User’s Guide

to

PARI / GP

(version 2.13.3)

The PARI Group

Institut de Mathématiques de Bordeaux, UMR 5251 du CNRS.

Université de Bordeaux, 351 Cours de la Libération
F-33405 TALENCE Cedex, FRANCE
e-mail: [email protected]

Home Page:
http://pari.math.u-bordeaux.fr/
Copyright c 2000–2020 The PARI Group

Permission is granted to make and distribute verbatim copies of this manual provided the copyright
notice and this permission notice are preserved on all copies.

Permission is granted to copy and distribute modified versions, or translations, of this manual
under the conditions for verbatim copying, provided also that the entire resulting derived work is
distributed under the terms of a permission notice identical to this one.

PARI/GP is Copyright c 2000–2020 The PARI Group

PARI/GP is free software; you can redistribute it and/or modify it under the terms of the GNU
General Public License as published by the Free Software Foundation. It is distributed in the hope
that it will be useful, but WITHOUT ANY WARRANTY WHATSOEVER.
 Table of Contents
Chapter 1: Overview of the PARI system . . . . . . . . . . . . . . . . . . . . . . . . . . .
5
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5
1.2 Multiprecision kernels / Portability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6
1.3 The PARI types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
7
1.4 The PARI philosophy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9
1.5 Operations and functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Chapter 2: The gp Calculator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2 The general gp input line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.3 The PARI types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.4 GP operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.5 Variables and symbolic expressions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.6 Variables and Scope . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.7 User defined functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.8 Member functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
2.9 Strings and Keywords . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
2.10 Errors and error recovery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
2.11 Interfacing GP with other languages . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
2.12 Defaults . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
2.13 Simple metacommands . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
2.14 The preferences file . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
2.15 Using readline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
2.16 GNU Emacs and PariEmacs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
Chapter 3: Functions and Operations Available in PARI and GP . . . . . . . . . . 67
3.1 Programming in GP: control statements . . . . . . . . . . . . . . . . . . . . . . . . . . 69
3.2 Programming in GP: other specific functions . . . . . . . . . . . . . . . . . . . . . . . 85
3.3 Parallel programming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
3.4 GP defaults . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
3.5 Standard monadic or dyadic operators . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
3.6 Conversions and similar elementary functions or commands . . . . . . . . . . . . . . . 139
3.7 Combinatorics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
3.8 Arithmetic functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
3.9 Polynomials and power series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
3.10 Vectors, matrices, linear algebra and sets . . . . . . . . . . . . . . . . . . . . . . . . . 255
3.11 Transcendental functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 293
3.12 Sums, products, integrals and similar functions . . . . . . . . . . . . . . . . . . . . . . 313
3.13 General number fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 347
3.14 Associative and central simple algebras . . . . . . . . . . . . . . . . . . . . . . . . . . 441
3.15 Elliptic curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 469
3.16 L-functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 515
3.17 Modular forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 535
3.18 Modular symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 571
3.19 Plotting functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 591
Appendix A: Installation Guide for the UNIX Versions . . . . . . . . . . . . . . . . 601
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 612
4
 Chapter 1:
Overview of the PARI system

1.1 Introduction.
PARI/GP is a specialized computer algebra system, primarily aimed at number theorists, but has
been put to good use in many other different fields, from topology or numerical analysis to physics.
Although quite an amount of symbolic manipulation is possible, PARI does badly compared
to systems like Axiom, Magma, Maple, Mathematica, Maxima, or Reduce on such tasks, e.g. mul-
tivariate polynomials, formal integration, etc. On the other hand, the three main advantages of
the system are its speed, the possibility of using directly data types which are familiar to mathe-
maticians, and its extensive algebraic number theory module (from the above-mentioned systems,
only Magma provides similar features).
Non-mathematical strong points include the possibility to program either in high-level scripting
languages or with the PARI library, a mature system (development started in the mid eighties) that
was used to conduct and disseminate original mathematical research, while building a large user
community, linked by helpful mailing lists and a tradition of great user support from the developers.
And, of course, PARI/GP is Free Software, covered by the GNU General Public License, either
version 2 of the License or (at your option) any later version.
PARI is used in three different ways:
1) as a library libpari, which can be called from an upper-level language application, for
instance written in ANSI C or C++;
2) as a sophisticated programmable calculator, named gp, whose language GP contains most
of the control instructions of a standard language like C;
3) the compiler gp2c translates GP code to C, and loads it into the gp interpreter. A
typical script compiled by gp2c runs 3 to 10 times faster. The generated C code can be edited and
optimized by hand. It may also be used as a tutorial to libpari programming.
The present Chapter 1 gives an overview of the PARI/GP system; gp2c is distributed separately
and comes with its own manual. Chapter 2 describes the GP programming language and the gp
calculator. Chapter 3 describes all routines available in the calculator. Programming in library
mode is explained in Chapters 4 and 5 in a separate booklet: User’s Guide to the PARI library
(libpari.dvi.
A tutorial for gp is provided in the standard distribution: A tutorial for PARI/GP (tuto-
rial.dvi) and you should read this first. You can then start over and read the more boring stuff
which lies ahead. You can have a quick idea of what is available by looking at the gp reference card
(refcard.dvi or refcard.ps). In case of need, you can refer to the complete function description
in Chapter 3.

5
How to get the latest version. Everything can be found on PARI’s home page:

http://pari.math.u-bordeaux.fr/.

From that point you may access all sources, some binaries, version information, the complete mailing
list archives, frequently asked questions and various tips. All threaded and fully searchable.

How to report bugs. Bugs are submitted online to our Bug Tracking System, available from
PARI’s home page, or directly from the URL

http://pari.math.u-bordeaux.fr/Bugs/.

Further instructions can be found on that page.

1.2 Multiprecision kernels / Portability.

The PARI multiprecision kernel comes in three non exclusive flavors. See Appendix A for how
to set up these on your system; various compilers are supported, but the GNU gcc compiler is the
definite favorite.
A first version is written entirely in ANSI C, with a C++-compatible syntax, and should be
portable without trouble to any 32 or 64-bit computer having no drastic memory constraints. We
do not know any example of a computer where a port was attempted and failed.
In a second version, time-critical parts of the kernel are written in inlined assembler. At present
this includes
• the whole ix86 family (Intel, AMD, Cyrix) starting at the 386, up to the Xbox gaming
console, including the Opteron 64 bit processor.
• three versions for the Sparc architecture: version 7, version 8 with SuperSparc processors,
and version 8 with MicroSparc I or II processors. UltraSparcs use the MicroSparc II version;
• the DEC Alpha 64-bit processor;
• the Intel Itanium 64-bit processor;
• the PowerPC equipping old macintoshs (G3, G4, etc.);
• the HPPA processors (both 32 and 64 bit);
A third version uses the GNU MP library to implement most of its multiprecision kernel. It
improves significantly on the native one for large operands, say 100 decimal digits of accuracy or
more. You should enable it if GMP is present on your system. Parts of the first version are still in
use within the GMP kernel, but are scheduled to disappear.
A historical version of the PARI/GP kernel, written in 1985, was specific to 680x0 based
computers, and was entirely written in MC68020 assembly language. It ran on SUN-3/xx, Sony
News, NeXT cubes and on 680x0 based Macs. It is no longer part of the PARI distribution; to run
PARI with a 68k assembler micro-kernel, use the GMP kernel!

6
1.3 The PARI types.
The GP language is not typed in the traditional sense; in particular, variables have no type.
In library mode, the type of all PARI objects is GEN, a generic type. On the other hand, it is
dynamically typed: each object has a specific internal type, depending on the mathematical object
it represents.
The crucial word is recursiveness: most of the PARI types are recursive. For example, the basic
internal type t_COMPLEX exists. However, the components (i.e. the real and imaginary part) of such
a “complex number” can be of any type. The only sensible ones are integers (we are then in Z[i]),
rational numbers (Q[i]), real numbers (R[i] = C), or even elements of Z/nZ (in (Z/nZ)[t]/(t2 +1)),
or p-adic numbers when p ≡ 3 mod 4 (Qp [i]). This feature must not be used too rashly in library
mode: for example you are in principle allowed to create objects which are “complex numbers of
complex numbers”. (This is not possible under gp.) But do not expect PARI to make sensible use
of such objects: you will mainly get nonsense.
On the other hand, it is allowed to have components of different, but compatible, types, which
can be freely mixed in basic ring operations + or ×. For example, taking again complex numbers,
the real part could be an integer, and the imaginary part a rational number. On the other hand,
if the real part is a real number, the imaginary part cannot be an integer modulo n !
Let us now describe the types. As explained above, they are built recursively from basic
types which are as follows. We use the letter T to designate any type; the symbolic names t_xxx
correspond to the internal representations of the types.
type t_INT Z Integers (with arbitrary precision)
type t_REAL R Real numbers (with arbitrary precision)
type t_INTMOD Z/nZ Intmods (integers modulo n)
type t_FRAC Q Rational numbers (in irreducible form)
type t_FFELT Fq Finite field element
type t_COMPLEX T [i] Complex numbers
type t_PADIC Qp p-adic numbers
type t_QUAD Q[w] Quadratic Numbers (where [Z[w] : Z] = 2)
type t_POLMOD T [X]/(P ) Polmods (polynomials modulo P ∈ T [X])
type t_POL T [X] Polynomials
type t_SER T ((X)) Power series (finite Laurent series)
type t_RFRAC T (X) Rational functions (in irreducible form)
type t_VEC Tn Row (i.e. horizontal) vectors
type t_COL Tn Column (i.e. vertical) vectors
type t_MAT Mm,n (T ) Matrices
type t_LIST Tn Lists
type t_STR Character strings
type t_CLOSURE Functions
type t_ERROR Error messages
type t_INFINITY −∞ and +∞
and where the types T in recursive types can be different in each component. The first nine basic
types, from t_INT to t_POLMOD, are called scalar types because they essentially occur as coefficients
of other more complicated objects. Type t_POLMOD is used to define algebraic extensions of a base
ring, and as such is a scalar type.
In addition, there exist types t_QFR and t_QFI for integral binary quadratic forms, and the in-
ternal type t_VECSMALL. The latter holds vectors of small integers, whose absolute value is bounded

7
by 231 (resp. 263 ) on 32-bit, resp. 64-bit, machines. They are used internally to represent permu-
tations, polynomials or matrices over a small finite field, etc.

Every PARI object (called GEN in the sequel) belongs to one of these basic types. Let us have
a closer look.

1.3.1 Integers and reals. They are of arbitrary and varying length (each number carrying in its
internal representation its own length or precision) with the following mild restrictions (given for
32-bit machines, the restrictions for 64-bit machines being so weak as to be considered nonexistent):
integers must be in absolute value less than 2536870815 (i.e. roughly 161614219 decimal digits). The
precision of real numbers is also at most 161614219 significant decimal digits, and the binary
exponent must be in absolute value less than 229 , resp. 261 , on 32-bit, resp. 64-bit machines.

Integers and real numbers are nonrecursive types.

1.3.2 Intmods, rational numbers, p-adic numbers, polmods, and rational functions.
These are recursive, but in a restricted way.

For intmods or polmods, there are two components: the modulus, which must be of type
integer (resp. polynomial), and the representative number (resp. polynomial).

For rational numbers or rational functions, there are also only two components: the numerator
and the denominator, which must both be of type integer (resp. polynomial).

Finally, p-adic numbers have three components: the prime p, the “modulus” pk , and an ap-
proximation to the p-adic number. Here Zp is considered as the projective limit lim Z/pk Z via
←−
its finite quotients, and Qp as its field of fractions. Like real numbers, the codewords contain an
exponent, giving the p-adic valuation of the number, and also the information on the precision of
the number, which is redundant with pk , but is included for the sake of efficiency.

1.3.3 Finite field elements. The exact internal format depends of the finite field size, but it
includes the field characteristic p, an irreducible polynomial T ∈ Fp [X] defining the finite field
Fp [X]/(T ) and the element expressed as a polynomial in (the class of) X.

1.3.4 Complex numbers and quadratic numbers. Quadratic numbers √ are numbers of the
form a + bw, where
√ w is such that [Z[w] : Z] = 2, and more precisely w = d/2 when d ≡ 0 mod 4,
and w = (1 + d)/2 when d ≡ 1 mod 4, where d is the discriminant
√ of a quadratic order. Complex
numbers correspond to the important special case w = −1.

Complex numbers are partially recursive: the two components a and b can be of type t_INT,
t_REAL, t_INTMOD, t_FRAC, or t_PADIC, and can be mixed, subject to the limitations mentioned
above. For example, a + bi with a and b p-adic is in Qp [i], but this is equal to Qp when p ≡ 1 mod 4,
hence we must exclude these p when one explicitly uses a complex p-adic type. Quadratic numbers
are more restricted: their components may be as above, except that t_REAL is not allowed.

8
1.3.5 Polynomials, power series, vectors, matrices. They are completely recursive, over a
commutative base ring: their components can be of any type, and types can be mixed (however
beware when doing operations). Note in particular that a polynomial in two variables is simply a
polynomial with polynomial coefficients. Polynomials or matrices over noncommutative rings are
not supported.
In the present version 2.13.3 of PARI, it is not possible to handle conveniently power series of
power series, i.e. power series in several variables. However power series of polynomials (which are
power series in several variables of a special type) are OK. This is a difficult design problem: the
mathematical problem itself contains some amount of imprecision, and it is not easy to design an
intuitive generic interface for such beasts.

1.3.6 Strings. These contain objects just as they would be printed by the gp calculator.

1.3.7 Zero. What is zero? This is a crucial question in all computer systems. The answer we
give in PARI is the following. For exact types, all zeros are equivalent and are exact, and thus
are usually represented as an integer zero. The problem becomes nontrivial for imprecise types:
there are infinitely many distinct zeros of each of these types! For p-adics and power series the
answer is as follows: every such object, including 0, has an exponent e. This p-adic or X-adic zero
is understood to be equal to O(pe ) or O(X e ) respectively.
Real numbers also have exponents and a real zero is in fact O(2e ) where e is now usually a
negative binary exponent. This of course is printed as usual for a floating point number (0.00 · · · or
0.Exx depending on the output format) and not with a O symbol as with p-adics or power series.
With respect to the natural ordering on the reals we make the following convention: whatever its
exponent a real zero is smaller than any positive number, and any two real zeroes are equal.

1.4 The PARI philosophy.

The basic principles which govern PARI is that operations and functions should, firstly, give
as exact a result as possible, and secondly, be permitted if they make any kind of sense.
In this respect, we make an important distinction between exact and inexact objects: by
definition, types t_REAL, t_PADIC or t_SER are imprecise. A PARI object having one of these
imprecise types anywhere in its tree is inexact, and exact otherwise. No loss of accuracy (rounding
error) is involved when dealing with exact objects. Specifically, an exact operation between exact
objects will yield an exact object. For example, dividing 1 by 3 does not give 0.333 · · ·, but the
rational number (1/3). To get the result as a floating point real number, evaluate 1./3 or 0.+1/3.
Conversely, the result of operations between imprecise objects, although inexact by nature,
will be as precise as possible. Consider for example the addition of two real numbers x and y. The
accuracy of the result is a priori unpredictable; it depends on the precisions of x and y, on their
sizes, and also on the size of x + y. From this data, PARI works out the right precision for the
result. Even if it is working in calculator mode gp, where there is a notion of default precision, its
value is only used to convert exact types to inexact ones.
In particular, if an operation involves objects of different accuracies, some digits will be dis-
regarded by PARI. It is a common source of errors to forget, for instance, that a real number is
given as r + 2e ε where r is a rational approximation, e a binary exponent and ε is a nondescript
real number less than 1 in absolute value. Hence, any number less than 2e may be treated as an
exact zero:

9
 ? 0.E-28 + 1.E-100
%1 = 0.E-28
? 0.E100 + 1
%2 = 0.E100
As an exercise, if a = 2^(-100), why do a + 0. and a * 1. differ?
The second principle is that PARI operations are in general quite permissive. For instance
taking the exponential of a vector should not make sense. However, it frequently happens that one
wants to apply a given function to all elements in a vector. This is easily done using a loop, or
using the apply built-in function, but in fact PARI assumes that this is exactly what you want to
do when you apply a scalar function to a vector. Taking the exponential of a vector will do just
that, so no work is necessary. Most transcendental functions work in the same way*.
In the same spirit, when objects of different types are combined they are first automatically
mapped to a suitable ring, where the computation becomes meaningful:
? 1/3 + Mod(1,5)
%1 = Mod(3, 5)
? I + O(5^9)
%2 = 2 + 5 + 2*5^2 + 5^3 + 3*5^4 + 4*5^5 + 2*5^6 + 3*5^7 + O(5^9)
? Mod(1,15) + Mod(1,10)
%3 = Mod(2, 5)
The first example is straightforward: since 3 is invertible mod 5, (1/3) is easily mapped to
Z/5Z. In the second example, I stands for the customary square root of −1; we obtain a 5-adic
number, 5-adically close to a square root of −1. The final example is more problematic, but there
are natural maps from Z/15Z and Z/10Z to Z/5Z, and the computation takes place there.

1.5 Operations and functions.

The available operations and functions in PARI are described in detail in Chapter 3. Here is
a brief summary:

1.5.1 Standard arithmetic operations.

Of course, the four standard operators +, -, *, / exist. We emphasize once more that division is, as
far as possible, an exact operation: 4 divided by 3 gives (4/3). In addition to this, operations on
integers or polynomials, like \ (Euclidean division), % (Euclidean remainder) exist; for integers, \/
computes the quotient such that the remainder has smallest possible absolute value. There is also
the exponentiation operator ^, when the exponent is of type integer; otherwise, it is considered as a
transcendental function. Finally, the logical operators ! (not prefix operator), && (and operator),
|| (or operator) exist, giving as results 1 (true) or 0 (false).

1.5.2 Conversions and similar functions.

Many conversion functions are available to convert between different types. For example floor,
ceiling, rounding, truncation, etc. . . . Other simple functions are included like real and imaginary
part, conjugation, norm, absolute value, changing precision or creating an intmod or a polmod.

* An ambiguity arises with square matrices. PARI always considers that you want to do com-
ponentwise function evaluation in this context, hence to get for example the standard exponential
of a square matrix you would need to implement a different function.

10
1.5.3 Transcendental functions.
They usually operate on any complex number, power series, and some also on p-adics. The list is
ever-expanding and of course contains all the elementary functions (exp/log, trigonometric func-
tions), plus many others (modular functions, Bessel functions, polylogarithms. . . ). Recall that by
extension, PARI usually allows a transcendental function to operate componentwise on vectors or
matrices.

1.5.4 Arithmetic functions.

Apart from a few like the factorial function or the Fibonacci numbers, these are functions which
explicitly use the prime factor decomposition of integers. The standard functions are included. A
number of factoring methods are used by a rather sophisticated factoring engine (to name a few,
Shanks’s SQUFOF, Pollard’s rho, Lenstra’s ECM, the MPQS quadratic sieve). These routines
output strong pseudoprimes, which may be certified by the APRCL test.
There is also a large package to work with algebraic number fields. All the usual operations on
elements, ideals, prime ideals, etc. are available. More sophisticated functions are also implemented,
like solving Thue equations, finding integral bases and discriminants of number fields, computing
class groups and fundamental units, computing in relative number field extensions, Galois and class
field theory, and also many functions dealing with elliptic curves over Q or over local fields.

1.5.5 Other functions.

Quite a number of other functions dealing with polynomials (e.g. finding complex or p-adic roots,
factoring, etc), power series (e.g. substitution, reversion), linear algebra (e.g. determinant, charac-
teristic polynomial, linear systems), and different kinds of recursions are also included. In addi-
tion, standard numerical analysis routines like univariate integration (using the double exponential
method), real root finding (when the root is bracketed), polynomial interpolation, infinite series
evaluation, and plotting are included.

And now, you should really have a look at the tutorial before proceeding.

11
12
 Chapter 2:
The gp Calculator

2.1 Introduction.

Originally, gp was designed as a debugging device for the PARI system library. Over the
years, it has become a powerful user-friendly stand-alone calculator. The mathematical functions
available in PARI and gp are described in the next chapter. In the present one, we describe the
specific use of the gp programmable calculator.

EMACS: If you have GNU Emacs and use the PariEmacs package, you can work in a special Emacs shell,
described in Section 2.16. Specific features of this Emacs shell are indicated by an EMACS sign in
the left margin.

We briefly mention at this point GNU TeXmacs (http://www.texmacs.org/), a free wysiwyg

editing platform that allows to embed an entire gp session in a document, and provides a nice
alternative to PariEmacs.

2.1.1 Startup.

To start the calculator, the general command line syntax is:

gp [-D key=val ] [files]

where items within brackets are optional. The [files] argument is a list of files written in the GP
scripting language, which will be loaded on startup. There can be any number of arguments of the
form -D key=val , setting some internal parameters of gp, or defaults: each sets the default key to
the value val . See Section 2.12 below for a list and explanation of all defaults. These defaults can
be changed by adding parameters to the input line as above, or interactively during a gp session,
or in a preferences file also known as gprc.

If a preferences file (to be discussed in Section 2.14) is found, gp then reads it and executes the
commands it contains. This provides an easy way to customize gp. The files argument is processed
right after the gprc.

A copyright banner then appears which includes the version number, and a lot of useful tech-
nical information. After the copyright, the computer writes the top-level help information, some
initial defaults, and then waits after printing its prompt, which is ’? ’ by default . Whether ex-
tended on-line help and line editing are available or not is indicated in this gp banner, between the
version number and the copyright message. Consider investigating the matter with the person who
installed gp if they are not. Do this as well if there is no mention of the GMP kernel.

13
2.1.2 Getting help.
To get help, type a ? and hit return. A menu appears, describing the main categories of
available functions and how to get more detailed help. If you now type ?n with n = 1, 2, . . ., you
get the list of commands corresponding to category n and simultaneously to Section 3.n of this
manual. If you type ?functionname where functionname is the name of a PARI function, you will
get a short explanation of this function.
If extended help (see Section 2.13.1) is available on your system, you can double or triple the ?
sign to get much more: respectively the complete description of the function (e.g. ??sqrt), or a list
of gp functions relevant to your query (e.g. ???"elliptic curve" or ???"quadratic field").
If gp was properly installed (see Appendix A), a line editor is available to correct the command
line, get automatic completions, and so on. See Section 2.15 or ??readline for a short summary
of the line editor’s commands.
If you type ?\ you will get a short description of the metacommands (keyboard shortcuts).
Finally, typing ?. will return the list of available (pre-defined) member functions. These
are functions attached to specific kind of objects, used to retrieve easily some information from
complicated structures (you can define your own but they won’t be shown here). We will soon
describe these commands in more detail.
More generally, commands starting with the symbols \ or ?, are not computing commands, but
are metacommands which allow you to exchange information with gp. The available metacommands
can be divided into default setting commands (explained below) and simple commands (or keyboard
shortcuts, to be dealt with in Section 2.13).

2.1.3 Input.
Just type in an instruction, e.g. 1 + 1, or Pi. No action is undertaken until you hit the
<Return> key. Then computation starts, and a result is eventually printed. To suppress printing
of the result, end the expression with a ; sign. Note that many systems use ; to indicate end of
input. Not so in gp: a final semicolon means the result should not be printed. (Which is certainly
useful if it occupies several screens.)

2.1.4 Interrupt, Quit.

Typing quit at the prompt ends the session and exits gp. At any point you can type Ctrl-C
(that is press simultaneously the Control and C keys): the current computation is interrupted and
control given back to you at the gp prompt, together with a message like

*** at top-level: gcd(a,b)

*** ^--------
*** gcd: user interrupt after 236 ms.
telling you how much time elapsed since the last command was typed in and in which GP function
the computation was aborted. It does not mean that that much time was spent in the function,
only that the evaluator was busy processing that specific function when you stopped it.

14
2.2 The general gp input line.
The gp calculator uses a purely interpreted language GP. The structure of this language is
reminiscent of LISP with a functional notation, f(x,y) rather than (f x y): all programming
constructs, such as if, while, etc. . . are functions*, and the main loop does not really execute,
but rather evaluates (sequences of) expressions. Of course, it is by no means a true LISP, and has
been strongly influenced by C and Perl since then.

2.2.1 Introduction. User interaction with a gp session proceeds as follows. First, one types a
sequence of characters at the gp prompt; see Section 2.15 for a description of the line editor. When
you hit the <Return> key, gp gets your input, evaluates it, then prints the result and assigns it to
an “history” array.
More precisely, the input is case-sensitive and, outside of character strings, blanks are com-
pletely ignored. Inputs are either metacommands or sequences of expressions. Metacommands are
shortcuts designed to alter gp’s internal state, such as the working precision or general verbosity
level; we shall describe them in Section 2.13, and ignore them for the time being.
The evaluation of a sequence of instructions proceeds in two phases: your input is first digested
(byte-compiled) to a bytecode suitable for fast evaluation, in particular loop bodies are compiled
only once but a priori evaluated many times; then the bytecode is evaluated.
An expression is formed by combining constants, variables, operator symbols, functions and
control statements. It is evaluated using the conventions about operator priorities and left to right
associativity. An expression always has a value, which can be any PARI object:
? 1 + 1
%1 = 2 \\ an ordinary integer
? x
%2 = x \\ a polynomial of degree 1 in the unknown x
? print("Hello")
Hello \\ void return value, ’Hello’ printed as side effect
? f(x) = x^2
%4 = (x)->x^2 \\ a user function
In the third example, Hello is printed as a side effect, but is not the return value. The print
command is a procedure, which conceptually returns nothing. But in fact procedures return a
special void object, meant to be ignored (but which evaluates to 0 in a numeric context, and
stored as 0 in the history or results). The final example assigns to the variable f the function
x 7→ x2 , the alternative form f = x->x^2 achieving the same effect; the return value of a function
definition is, unsurprisingly, a function object (of type t_CLOSURE).
Several expressions are combined on a single line by separating them with semicolons (’;’).
Such an expression sequence will be called a seq. A seq also has a value, which is the value of the
last expression in the sequence. Under gp, the value of the seq, and only this last value, becomes
an history entry. The values of the other expressions in the seq are discarded after the execution
of the seq is complete, except of course if they were assigned into variables. In addition, the value
of the seq is printed if the line does not end with a semicolon ;.

* Not exactly, since not all their arguments need be evaluated. For instance it would be stupid
to evaluate both branches of an if statement: since only one will apply, only this one is evaluated.

15
2.2.2 The gp history of results.
This is not to be confused with the history of your commands, maintained by readline. The
gp history contains the results they produced, in sequence.
The successive elements of the history array are called %1, %2, . . . As a shortcut, the latest
computed expression can also be called %, the previous one %‘, the one before that %‘‘ and so on.
When you suppress the printing of the result with a semicolon, it is still stored in the history,
but its history number will not appear either. It is a better idea to assign it to a variable for later
use than to mentally recompute what its number is. Of course, on the next line, you may just use
%.
The time used to compute that history entry is also stored as part of the entry and can be
recovered using the %# operator: %#1, %#2, %#‘; %# by itself returns the time needed to compute
the last result (the one returned by %). The output is a vector with two components [cpu, real]
where cpu is the CPU time and real is the wall clock time.

Remark. The history “array” is in fact better thought of as a queue: its size is limited to 5000
entries by default, after which gp starts forgetting the initial entries. So %1 becomes unavailable as
gp prints %5001. You can modify the history size using histsize.

2.2.3 Special editing characters. A GP program can of course have more than one line. Since
your commands are executed as soon as you have finished typing them, there must be a way to tell
gp to wait for the next line or lines of input before doing anything. There are three ways of doing
this.
The first one is to use the backslash character \ at the end of the line that you are typing,
just before hitting <Return>. This tells gp that what you will write on the next line is the physical
continuation of what you have just written. In other words, it makes gp forget your newline
character. You can type a \ anywhere. It is interpreted as above only if (apart from ignored
whitespace characters) it is immediately followed by a newline. For example, you can type
? 3 + \
4
instead of typing 3 + 4.
The second one is a variation on the first, and is mostly useful when defining a user function
(see Section 2.7): since an equal sign can never end a valid expression, gp disregards a newline
immediately following an =.
? a =
123
%1 = 123
The third one is in general much more useful, and uses braces { and }. An opening brace {
signals that you are typing a multi-line command, and newlines are ignored until you type a closing
brace }. There are two important, but easily obeyed, restrictions: first, braces do not nest; second,
inside an open brace-close brace pair, all input lines are concatenated, suppressing any newlines.
Thus, all newlines should occur after a semicolon (;), a comma (,) or an operator (for clarity’s
sake, never split an identifier over two lines in this way). For instance, the following program
{
a = b

16
 b = c
}
would silently produce garbage, since this is interpreted as a=bb=c which assigns the value of c to
both bb and a. It should have been written

{
a = b;
b = c;
}

2.3 The PARI types.

We see here how to input values of the different data types known to PARI. Recall that blanks are
ignored in any expression which is not a string (see below).

A note on efficiency. The following types are provided for convenience, not for speed: t_INTMOD,
t_FRAC, t_PADIC, t_QUAD, t_POLMOD, t_RFRAC. Indeed, they always perform a reduction of some
kind after each basic operation, even though it is usually more efficient to perform a single
P reduction
at the end of some complex computation. For instance, in a convolution product i+j=n xi yj in
Z/N Z — common when multiplying polynomials! —, it is quite wasteful to perform n reductions
modulo N . In short, basic individual operations on these types are fast, but recursive objects
with such components could be handled more efficiently: programming with libpari will save large
constant factors here, compared to GP.

2.3.1 Integers (t_INT). After an (optional) leading + or -, type in the decimal digits of your
integer. No decimal point!

? 1234567
%1 = 1234567
? -3
%2 = -3
? 1. \\ oops, not an integer
%3 = 1.000000000000000000000000000
Integers can be input in hexadecimal notation by prefixing them with 0x; hexadecimal digits
(a, . . . , f ) can be input either in lowercase or in uppercase:

? 0xF
%4 = 15
? 0x1abcd
%5 = 109517
Integers can also be input in binary by prefixing them with 0b:

? 0b010101
%6 = 21

17
2.3.2 Real numbers (t_REAL).
Real numbers are represented (approximately) in a floating point system, internally in base 2,
but converted to base 10 for input / output purposes. A t_REAL object has a given bit accuracy
(or bit precision) ` ≥ 0; it comprises
• a sign s: +1, −1 or 0;
• a mantissa m: a multiprecision integer, 0 ≤ m < 2` ;
• an exponent e: a small integer in [−2B , 2B [, where B = 31 on a 32-bit machine and 63
otherwise.
This data may represent any real number x such that

|x − sm2e | < 2e−` .

We consider that a t_REAL with sign s = 0 has accuracy ` = 0, so that its mantissa is useless, but
it still has an exponent e and acts like a machine epsilon for all accuracies < e.
After an (optional) leading + or -, type a number with a decimal point. Leading zeroes may
be omitted, up to the decimal point, but trailing zeroes are important: your t_REAL is assigned
an internal precision, which is the supremum of the input precision, one more than the number of
decimal digits input, and the default realprecision. For example, if the default precision is 38
digits, typing 2. yields a precision of 38 digits, but 2.0. . . 0 with 45 zeros gives a number with
internal decimal precision at least 45, although less may be printed.
You can also use scientific notation with the letter E or e. As usual, en is interpreted as ×10n
for all integers n. Since the result is converted to a t_REAL, you may often omit the decimal point
in this case: 6.02 E 23 or 1e-5 are fine, but e10 is not.
By definition, 0.E n returns a real 0 of exponent n, whereas 0. returns a real 0 “of default
precision” (of exponent −realprecision), see Section 1.3.7, behaving like the machine epsilon for
the current default accuracy: any float of smaller absolute value is indistinguishable from 0.

Note on output formats. A zero real number is printed in e format as 0.Exx where xx is the
(usually negative) decimal exponent of the number (cf. Section 1.3.7). This allows the user to check
the accuracy of that particular zero.
When the integer part of a real number x is not known exactly because the exponent of x is
greater than the internal precision, the real number is printed in e format.

Technical note. The internal precision is actually expressed in bits and can be viewed and
manipulated globally in interactive use via realprecision (decimal digits, as explained above;
shortcut \p) or realbitprecision (bits; shortcut \ps), the latter allowing finer granularity. See
Section 3.11 for details. In programs we advise to leave this global variable alone and adapt precision
locally for a given sequence of computations using localbitprec.
Note that most decimal floating point numbers cannot be converted exactly in binary, the
(binary) number actually stored is a rounded version of the (decimal) number input. Analogously,
a decimal output is rounded from the internal binary representation.

18
2.3.3 Intmods (t_INTMOD). To create the image of the integer a in Z/bZ (for some nonzero integer
b), type Mod(a,b); not a%b. Internally, all operations are done on integer representatives belonging
to [0, b − 1].
Note that this type is available for convenience, not for speed: each elementary operation
involves a reduction modulo b.
If x is a t_INTMOD Mod(a,b), the following member function is defined:
x.mod: return the modulus b.

2.3.4 Rational numbers (t_FRAC). All fractions are automatically reduced to lowest terms, so it
is impossible to work with reducible fractions. To enter n/m just type it as written. As explained
in Section 3.5.8, floating point division is not performed, only reduction to lowest terms.
Note that rational computation are almost never the fastest method to proceed: in the PARI
implementation, each elementary operation involves computing a gcd. It is generally a little more
efficient to cancel denominators and work with integers only:
? P = Pol( vector(10^3,i, 1/i) ); \\ big polynomial with small rational coeffs
? P^2
time = 1,392 ms.
? c = content(P); c^2 * (P/c)^2; \\ same computation in integers
time = 1,116 ms.
And much more efficient (but harder to setup) to use homomorphic imaging schemes and modular
computations. As the simple example below indicates, if you only need modular information, it
is very worthwhile to work with t_INTMODs directly, rather than deal with t_FRACs all the way
through:
? p = nextprime(10^7);
? sum(i=1, 10^5, 1/i) % p
time = 13,288 ms.
%1 = 2759492
? sum(i=1, 10^5, Mod(1/i, p))
time = 60 ms.
%2 = Mod(2759492, 10000019)

2.3.5 Finite field elements (t_FFELT). Let T ∈ Fp [X] be a monic irreducible polynomial defining
your finite field over Fp , for instance obtained using ffinit. Then the ffgen function creates a
generator of the finite field as an Fp -algebra, namely the class of X in Fp [X]/(T ), from which you
can build all other elements. For instance, to create the field F28 , we write
? T = ffinit(2, 8);
? y = ffgen(T, ’y);
? y^0 \\ the unit element in the field
%3 = 1
? y^8
%4 = y^6 + y^5 + y^4 + y^3 + y + 1
The second (optional) parameter to ffgen is only used to display the result; it is customary to
use the name of the variable we assign the generator to. If g is a t_FFELT, the following member
functions are defined:

19
 g.pol: the polynomial (with reduced integer coefficients) expressing g in term of the field
generator.
g.p: the characteristic of the finite field.
g.f: the dimension of the definition field over its prime field; the cardinality of the definition
field is thus pf .
g.mod: the minimal polynomial (with reduced integer coefficients) of the field generator.

2.3.6 Complex numbers

√ (t_COMPLEX). To enter x + iy, type x + I*y. (That’s I, not i!) The
letter I stands for −1. The “real” and “imaginary” parts x and y can be of type t_INT, t_REAL,
t_INTMOD, t_FRAC, or t_PADIC.

2.3.7 p-adic numbers (t_PADIC):. Typing O(p^k), where p is a prime and k is an integer,
yields a p-adic 0 of accuracy k, representing any p-adic number whose valuation is ≥ k. To input a
general nonzero p-adic number, write a suitably precise rational or integer approximation and add
O(p^k) to it. For example, you can type in the 7-adic number
2*7^(-1) + 3 + 4*7 + 2*7^2 + O(7^3)
exactly as shown, or equivalently as 905/7 + O(7^3).
Note that it is not checked whether p is indeed prime but results are undefined if this is not
the case: you can try to work on 10-adics if you want, but disasters will happen as soon as you do
something nontrivial. For instance:
? t = 2 * (1/10 + O(10^5));
? lift(t)
%2 = 2/10 \\ not reduced (invalid t_FRAC)
? factor(x^2-t)
*** at top-level: factor(x^2-%1)
*** ^--------------
*** factor: impossible inverse in Fl_inv: Mod(2, 10000).
Note that O(25) is not the same as O(5^2); you want the latter!
If a is a t_PADIC, the following member functions are defined:
a.mod: returns the modulus pk .
a.p: returns p.
Note that this type is available for convenience, not for speed: internally, t_PADICs are stored
as p-adic units modulo some pk . Each elementary operation involves updating pk (multiplying or
dividing by powers of p) and a reduction mod pk . In particular, additions are slow.
? n = 1+O(2^20); for (i=1,10^6, n++)
time = 841 ms.
? n = Mod(1,2^20); for (i=1,10^6, n++)
time = 441 ms.
? n = 1; for (i=1,10^6, n++)
time = 328 ms.
The penalty attached to maintaining pk decreases steeply as p increases (and updates become
rare). But t_INTMODs remain at least 25% more efficient. (On the other hand, they do not allow
denominators!)

20
2.3.8 Quadratic numbers (t_QUAD). This type is used to work in the quadratic order of discrim-
inant d, where d is a nonsquare integer congruent to 0 or 1 (modulo 4). The command
w = quadgen(d,’w)
assigns to w the “canonical” generator
√ √ for the integer basis of the order of discriminant d, i.e. w =
d/2 if d ≡ 0 mod 4, and w = (1 + d)/2 if d ≡ 1 mod 4 and set its name to w. The name ’w is used
for printing and we advise to store it in a variable of the same name. Beware, two t_QUADs with
different discriminants can be printed in the same way and not be equal; however, gp will refuse to
add or multiply them for example, so use different names for different discriminants.
Since the order is Z + wZ, any other element can √ be input as z = x+y*w for some integers x
and y. In fact, you may work in its fraction field Q( d) and use t_FRAC values for x and y.
The member function z.disc retrieves the discriminant d; x and y are obtained via real(z)
and imag(z) respectively.

2.3.9 Polmods (t_POLMOD). Exactly as for intmods, to enter x mod y (where x and y are poly-
nomials), type Mod(x,y), not x%y. Note that when y is an irreducible polynomial in one variable,
polmods whose modulus is y are simply algebraic numbers in the finite extension defined by the
polynomial y. This allows us to work easily in number fields, finite extensions of the p-adic field
Qp , or finite fields.
Note that this type is available for convenience, not for speed: each elementary operation
involves a reduction modulo y. If p is a t_POLMOD, the following member functions are defined:
p.pol: return a representative of the polynomial class of minimal degree.
p.mod: return the modulus.
Important remark. Mathematically, the variables occurring in a polmod are not free variables.
But internally, a congruence class in R[t]/(y) is represented by its representative of lowest degree,
which is a t_POL in R[t], and computations occur with polynomials in the variable t. PARI will not
recognize that Mod(y, y^2 + 1) is “the same” as Mod(x, x^2 + 1), since x and y are different
variables.
To avoid inconsistencies, polmods must use the same variable in internal operations (i.e. be-
tween polmods) and variables of lower priority for external operations, typically between a poly-
nomial and a polmod. See Section 2.5.3 for a definition of “priority” and a discussion of (PARI’s
idea of) multivariate polynomial arithmetic. For instance:
? Mod(x, x^2+ 1) + Mod(x, x^2 + 1)
%1 = Mod(2*x, x^2 + 1) \\ 2i (or −2i), with i2 = −1
? x + Mod(y, y^2 + 1)
%2 = x + Mod(y, y^2 + 1) \\ in Q(i)[x]
? y + Mod(x, x^2 + 1)
%3 = Mod(x + y, x^2 + 1) \\ in Q(y)[i]
The first two are straightforward, but the last one may not be what you want: y is treated here as
a numerical parameter, not as a polynomial variable.
If the main variables are the same, it is allowed to mix t_POL and t_POLMODs. The result is
the expected t_POLMOD. For instance
? x + Mod(x, x^2 + 1)
%1 = Mod(2*x, x^2 + 1)

21
2.3.10 Polynomials (t_POL). Type the polynomial in a natural way, not forgetting to put a “∗”
between a coefficient and a formal variable;
? 1 + 2*x + 3*x^2
%1 = 3*x^2 + 2*x + 1
This assumes that x is still a ”free variable”.
? x = 1; 1 + 2*x + 3*x^2
%2 = 6
generates an integer, not a polynomial! It is good practice to never assign values to polynomial
variables to avoid the above problem, but a foolproof construction is available using ’x instead of x:
’x is a constant evaluating to the free variable with name x, independently of the current value
of x.
? x = 1; 1 + 2*’x + 3*’x^2
%3 = 1 + 2*x + 3*x^2
? x = ’x; 1 + 2*x + 3*x^2
%4 = 1 + 2*x + 3*x^2
You may also use the functions Pol or Polrev:
? Pol([1,2,3]) \\ Pol creates a polynomial in x by default
%1 = x^2 + 2*x + 3
? Polrev([1,2,3])
%2 = 3*x^2 + 2*x + 1
? Pol([1,2,3], ’y) \\ we use ’y, safer than y
%3 = y^2 + 2*y + 3
The latter two are much more efficient constructors than an explicit summation (the latter is
quadratic in the degree, the former linear):
? for (i=1, 10^4, Polrev( vector(100, i,i) ) )
time = 124ms
? for (i=1, 10^4, sum(i = 1, 100, (i+1) * ’x^i) )
time = 3,985ms
Polynomials are always printed as univariate polynomials over a commutative base ring, with
monomials sorted by decreasing degree:
? (x+y+1)^2
%1 = x^2 + (2*y + 2)*x + (y^2 + 2*y + 1)
(Univariate polynomial in x whose coefficients are polynomials in y.) See Section 2.5 for valid vari-
able names, and a discussion of multivariate polynomial rings. Polynomials over noncommutative
rings are not supported.

2.3.11 Power series (t_SER). Typing O(X^k), where k is an integer, yields an X-adic 0 of accu-
racy k, representing any power series in X whose valuation is ≥ k. Of course, X can be replaced by
any other variable name! To input a general nonzero power series, type in a polynomial or rational
function (in X, say), and add O(X^k) to it. The discussion in the t_POL section about variables
remains valid; a constructor Ser replaces Pol and Polrev. Power series over noncommutative rings
are not supported.

22
Caveat. Power series with inexact coefficients sometimes have a nonintuitive behavior: if k signif-
icant terms are requested, an inexact zero is counted as significant, even if it is the coefficient of
lowest degree. This means that useful higher order terms may be disregarded.
If a series with a zero leading coefficient must be inverted, then as a desperation measure that
coefficient is discarded, and a warning is issued:
? C = 0. + y + O(y^2);
? 1/C
*** _/_: Warning: normalizing a series with 0 leading term.
%2 = y^-1 + O(1)
The last output could be construed as a bug since it is a priori impossible to deduce such a result
from the input (0. represents any sufficiently small real number). But it was thought more useful
to try and go on with an approximate computation than to raise an early exception.
If the series precision is insufficient, errors may occur (mostly division by 0), which could have
been avoided by a better global understanding of the computation:
? A = 1/(y + 0.); B = 1. + O(y);
? B * denominator(A)
%2 = 0.E-28 + O(y)
? A/B
*** _/_: Warning: normalizing a series with 0 leading term.
%3 = 1.000000000000000000000000000*y^-1 + O(1)
? A*B
*** _*_: Warning: normalizing a series with 0 leading term.
%4 = 1.000000000000000000000000000*y^-1 + O(1)

2.3.12 Rational functions (t_RFRAC). As for fractions, all rational functions are automatically
reduced to lowest terms. All that was said about fractions in Section 2.3.4 remains valid here.

2.3.13 Binary quadratic forms of positive or negative discriminant (t_QFR and t_QFI).
These are input using the function Qfb. For example Qfb(1,2,3) creates the binary form q =
x2 + 2xy + 3y 2 . It is imaginary (of internal type t_QFI) since its discriminant 22 − 4 × 3 = −8 is
negative. Although imaginary forms could be positive or negative definite, only positive definite
forms are implemented.
The discriminant can be retrieved via poldisc. The individual components are obtained via
either of
[a,b,c] = Vec(q);
a = component(q,1);
b = component(q,2);
c = component(q,3);
In the case of forms with positive discriminant (t_QFR), you may add an optional fourth
component (related to the regulator, more precisely to Shanks and Lenstra’s distance), which
must be a real number. See also the function qfbprimeform which creates a prime form of given
discriminant.

23
2.3.14 Row and column vectors (t_VEC and t_COL). To enter a row vector, type the compo-
nents separated by commas “,”, and enclosed between brackets “[ ” and “ ]”, e.g. [1,2,3]. To
enter a column vector, type the vector horizontally, and add a tilde “˜” to transpose. [ ] yields the
empty (row) vector. The function Vec can be used to transform any object into a vector (see Chap-
ter 3). The construction [i..j], where i ≤ j are two integers returns the vector [i, i + 1, . . . , j − 1, j]
? [1,2,3]
%1 = [1, 2, 3]
? [-2..3]
%2 = [-2, -1, 0, 1, 2, 3]
Let the variable v contain a (row or column) vector:
• v[m] refers to its m-th entry; you can assign any value to v[m], i.e. write something like
v[m] = expr .
• v[i..j], where i ≤ j, returns the vector slice containing elements v[i], . . . , v[j]; you can not
assign a result to v[i..j].
• v[^i] returns the vector whose i-th entry has been removed; you can not assign a result to
v[^i].
In the last two constructions v[i..j] and v[^i], i and j are allowed to be negative integers, in
which case, we start counting from the end of the vector: e.g., −1 is the index of the last element.
? v = [1,2,3,4];
? v[2..4]
%2 = [2, 3, 4]
? v[^3]
%3 = [1, 2, 4]
? v[^-1]
%3 = [1, 2, 3]
? v[-3..-1]
%4 = [2, 3, 4]

Remark. vector is the standard constructor for row vectors whose i-th entry is given by a simple
function of i; vectorv is similar for column vectors:
? vector(10, i, i^2+1)
%1 = [2, 5, 10, 17, 26, 37, 50, 65, 82, 101]

The functions Vec and Col convert objects to row and column vectors respectively (as well as
Vecrev and Colrev, which revert the indexing):
? T = poltchebi(5) \\ 5-th Chebyshev polynomial
%1 = 16*x^5 - 20*x^3 + 5*x
? Vec(T)
%2 = [16, 0, -20, 0, 5, 0] \\ coefficients of T
? Vecrev(T)
%3 = [0, 5, 0, -20, 0, 16] \\ ... in reverse order

24
Remark. For v a t_VEC, t_COL, t_VECSMALL, t_LIST or t_MAT, the alternative set-notations
[g(x) | x <- v, f(x)]
[x | x <- v, f(x)]
[g(x) | x <- v]

are available as shortcuts for

apply(g, select(f, Vec(v)))
select(f, Vec(v))
apply(g, Vec(v))

respectively, and may serve as t_VEC constructors:

? [ p | p <- primes(10), isprime(p+2) ]
%2 = [3, 5, 11, 17, 29]

returns the primes p (among the first 10 primes) such that (p, p + 2) is a twin pair;
? [ p^2 | p <- primes(10), p % 4 == 1 ]
%1 = [25, 169, 289, 841]

returns the squares of the primes congruent to 1 modulo 4, where p runs among the first 10 primes.

2.3.15 Matrices (t_MAT). To enter a matrix, type the components row by row, the components
being separated by commas “,”, the rows by semicolons “;”, and everything enclosed in brackets
“[ ” and “ ]”, e.g. [x,y; z,t; u,v]. [;] yields an empty (0 × 0) matrix. The function Mat
transforms any object into a matrix, and matrix creates matrices whose (i, j)-th entry is described
by a function f (i, j):
? Mat(1)
%1 =
[1]
? matrix(2,2, i,j, 2*i+j)
%2 =
[3 4]
[5 6]
Matrix multiplication assumes that the base ring containing the matrix entries is commutative.
Let the variable M contain a matrix, and let i, j, k, l denote four integers:
• M[i,j] refers to its (i, j)-th entry; you can assign any result to M[i,j].
• M[i,] refers to its i-th row; you can assign a t_VEC of the right dimension to M[i,].
• M[,j] refers to its j-th column; you can assign a t_COL of the right dimension to M[,j].
But M[i] is meaningless and triggers an error. The “range” i..j and “caret” ^c notations are
available as for vectors; you can not assign to any of these:
• M[i..j, k..l], i ≤ j, k ≤ l, returns the submatrix built from the rows i to j and columns
k to l of M .

25
 • M[i..j,] returns the submatrix built from the rows i to j of M .

• M[,i..j] returns the submatrix built from the columns i to j of M .

• M[i..j, ^k], i ≤ j, returns the submatrix built from the rows i to j and column k removed.

• M[^k,] returns the submatrix with row k removed.

• M[,^k] returns the submatrix with column k removed.

Finally,

• M[i..j, k] returns the t_COL built from the k-th column (entries i to j).

• M[^i, k] returns the t_COL built from the k-th column (entry i removed).

• M[k, i..j] returns the t_VEC built from the k-th row (entries i to j).

• M[k, ^i] returns the t_VEC built from the k-th row (entry i removed).

? M = [1,2,3;4,5,6;7,8,9];
? M[1..2, 2..3]
%2 =
[2 3]
[5 6]
? M[1..2,]
%3 =
[1 2 3]
[4 5 6]
? M[,2..3]
%4 =
[2 3]
[5 6]
[8 9]

All this is recursive, so if M is a matrix of matrices of . . . , an expression such as M[1,1][,3][4]

= 1 is perfectly valid (and actually identical to M[1,1][4,3] = 1), assuming that all matrices along
the way have compatible dimensions.

26
Technical note (design flaw). Matrices are internally represented as a vector of columns. All
matrices with 0 columns are thus represented by the same object (internally, an empty vector), and
there is no way to distinguish between them. Thus it is not possible to create or represent matrices
with zero columns and an actual nonzero number of rows. The empty matrix [;] is handled as
though it had an arbitrary number of rows, exactly as many as needed for the current computation
to make sense:

? [1,2,3; 4,5,6] * [;]

%1 = [;]

The empty matrix on the first line is understood as a 3 × 0 matrix, and the result as a 2 × 0 matrix.
On the other hand, it is possible to create matrices with a given positive number of columns, each
of which has zero rows, e.g. using Mat as above or using the matrix function.

Note that although the internal representation is essentially the same, a row vector of column
vectors is not a matrix; for example, multiplication will not work in the same way. It is easy to go
from one representation to the other using Vec / Mat, though:

? [1,2,3;4,5,6]
%1 =
[1 2 3]
[4 5 6]
? Vec(%)
%2 = [[1, 4]~, [2, 5]~, [3, 6]~]
? Mat(%)
%3 =
[1 2 3]
[4 5 6]

2.3.16 Lists (t_LIST). Lists can be input directly, as in List([1,2,3,4]); but in most cases, one
creates an empty list, then appends elements using listput:

? L = List(); listput(~L,1); listput(~L,2);

? L
%2 = List([1, 2])

Note the ~L: this means that the function is called with a reference to L and changes L in place.
Elements can be accessed directly as with the vector types described above.

2.3.17 Strings (t_STR). To enter a string, enclose it between double quotes ", as in: "this is a
string". The function Str can be used to transform any object into a string.

27
2.3.18 Small vectors (t_VECSMALL). This type codes in an efficient way vectors containing only
small integers, such as permutations. Most gp functions will refuse to operate on these objects,
notable exceptions being vecsort and conversion functions such as VEC, but you can retrieve
entries and assign to them as for ordinary vectors. You can also convert back and forth between
t_VECSMALL and t_VEC objects using Vec and Vecsmall.
? v = Vecsmall([2, 4, 6])
%1 = Vecsmall([1, 2, 3])
? v[1]
%2 = 2
? v[1] = 3; v
%3 = Vecsmall([3, 2, 3])
? v[2..3]
%4 = Vecsmall([2, 3])
? v[^2]
%5 = Vecsmall([3, 3])
? Vec(v)
%6 = [3, 2, 3]
Allowed entries for a t_VECMALL are signed integer x such that |x| < 231 on a 32-bit architecture,
resp. |x| < 263 on a 64-bit architecture Assigning a larger integer to a t_VECSMALL entry triggers
an exception:
? v[1] = 2^63
*** at top-level: v[1]=2^63
*** ^----------
*** incorrect type in t_VECSMALL assignment (t_INT).

2.3.19 Functions (t_CLOSURE). We will explain this at length in Section 2.7. For the time being,
suffice it to say that functions can be assigned to variables, as any other object, and the following
equivalent basic forms are available to create new ones
f = (x,y) -> x^2 + y^2
f(x,y) = x^2 + y^2

2.3.20 Error contexts (t_ERROR). An object of this type is created whenever an error occurs: it
contains some information about the error and the error context. Usually, an appropriate error is
printed immediately, the computation is aborted, and GP enters the “break loop”:
? 1/0; 1 + 1
*** at top-level: 1/0;1+1
*** ^------
*** _/_: division by a noninvertible object
*** Break loop: type ’break’ to go back to the GP prompt

Here the computation is aborted as soon as we try to evaluate 1/0, and 1 + 1 is never executed.
Exceptions can be trapped using iferr, however: we can evaluate some expression and either
recover an ordinary result (no error occurred), or an exception (an error did occur).
? i = Mod(6,12); iferr(1/i, E, print(E)); 1 + 1
error("impossible inverse modulo: Mod(6, 12).")
%1 = 2

28
One can ignore the exception, print it as above, or extract non trivial information from the error
context:
? i = Mod(6,12); iferr(1/i, E, print(component(E,1)));
Mod(6, 12)
We can also rethrow the exception: error(E).

2.3.21 Infinity (t_INFINITY).

There are only two objects of this type +oo and -oo, representing ±∞. This type only contain
only two elements oo and -oo, They are used in functions sur as intnum or polrootsreal, to
encode infinite real intervals. These objects can only be negated and compared to real numbers
(t_INT, t_REAL, t_FRAC), but not included in any computation, i.e. 1+oo is an error, not kbdoo
again.

2.4 GP operators.
Loosely speaking, an operator is a function, usually attached to basic arithmetic operations, whose
name contains only nonalphanumeric characters. For instance + or -, but also = or +=, or even [ ]
(the selection operator). As all functions, operators take arguments, and return a value; assignment
operators also have side effects: besides returning a value, they change the value of some variable.
Each operator has a fixed and unchangeable priority, which means that, in a given expression,
the operations with the highest priority is performed first. Unless mentioned otherwise, opera-
tors at the same priority level are left-associative (performed from left to right), unless they are
assignments, in which case they are right-associative. Anything enclosed between parenthesis is
considered a complete subexpression, and is resolved recursively, independently of the surrounding
context. For instance,
a + b + c --> (a + b) + c \\ left-associative
a = b = c --> a = (b = c) \\ right-associative
Assuming that op 1 , op 2 , op 3 are binary operators with increasing priorities (think of +, *, ^),

x op 1 y op 2 z op 2 x op 3 y

is equivalent to
x op 1 ((y op 2 z) op 2 (x op 3 y)).

GP contains many different operators, either unary (having only one argument) or binary, plus
a few special selection operators. Unary operators are defined as either prefix or postfix , meaning
that they respectively precede (op x) and follow (x op) their single argument. Some symbols are
syntactically correct in both positions, like !, but then represent different operators: the ! symbol
represents the negation and factorial operators when in prefix and postfix position respectively.
Binary operators all use the (infix) syntax x op y.
Most operators are standard (+, %, =), some are borrowed from the C language (++, <<),
and a few are specific to GP (\, #). Beware that some GP operators differ slightly from their C
counterparts. For instance, GP’s postfix ++ returns the new value, like the prefix ++ of C, and the
binary shifts <<, >> have a priority which is different from (higher than) that of their C counterparts.
When in doubt, just surround everything by parentheses; besides, your code will be more legible.

29
Here is the list of available operators, ordered by decreasing priority, binary and left-associative
unless mentioned otherwise. An expression is an lvalue if something can be assigned to it. (The
name comes from left-value, to the left of a = operator; e.g. x, or v[1] are lvalues, but x + 1 is
not.)
• Priority 14
: as in x:small, is used to indicate to the GP2C compiler that the variable on the left-hand
side always contains objects of the type specified on the right hand-side (here, a small integer) in
order to produce more efficient or more readable C code. This is ignored by GP.
• Priority 13
( ) is the function call operator. If f is a closure and args is a comma-separated list of
arguments (possibly empty), f (args) evaluates f on those arguments.
• Priority 12
++ and -- (unary, postfix): if x is an lvalue, x++ assigns the value x + 1 to x, then returns
the new value of x. This corresponds to the C statement ++x: there is no prefix ++ operator in GP.
x-- does the same with x − 1. These operators are not associative, i.e. x++++ is invalid, since x++
is not an lvalue.
• Priority 11
.member (unary, postfix): x.member extracts member from structure x (see Section 2.8).
[ ] is the selection operator. x[i] returns the i-th component of vector x; x[i,j], x[,j]
and x[i,] respectively return the entry of coordinates (i, j), the j-th column, and the i-th row of
matrix x. If the assignment operator (=) immediately follows a sequence of selections, it assigns its
right hand side to the selected component. E.g x[1][1] = 0 is valid; but beware that (x[1])[1]
= 0 is not (because the parentheses force the complete evaluation of x[1], and the result is not
modifiable).
• Priority 10
’ (unary, postfix): derivative with respect to the main variable. If f is a function (t_CLOSURE),
f 0 is allowed and defines a new function, which will perform numerical derivation when evaluated
at a scalar x; this is defined as (f (x + ε) − f (x − ε))/2ε for a suitably small epsilon depending on
current precision.
? (x^2 + y*x + y^2)’ \\ derive with respect to main variable x
%1 = 2*x + y
? SIN = cos’
%2 = cos’
? SIN(Pi/6) \\ numerical derivation
%3 = -0.5000000000000000000000000000
? cos’(Pi/6) \\ works directly: no need for intermediate SIN
%4 = -0.5000000000000000000000000000

~ (unary, postfix): vector/matrix transpose.

! (unary, postfix): factorial. x! = x(x − 1) · · · 1.
! (unary, prefix): logical not. !x returns 1 if x is equal to 0 (specifically, if gequal0(x)==1),
and 0 otherwise.
• Priority 9
# (unary, prefix): cardinality; #x returns length(x).

30
• Priority 8
^: powering. This operator is right associative: 2 ^3^4 is understood as 2 ^(3^4).
• Priority 7
+, - (unary, prefix): - toggles the sign of its argument, + has no effect whatsoever.
• Priority 6
*: multiplication.
/: exact division (3/2 yields 3/2, not 1.5).
\, %: Euclidean quotient and remainder, i.e. if x = qy + r, then x\y = q, x%y = r. If x and y
are scalars, then q is an integer and r satisfies 0 ≤ r < |y|; if x and y are polynomials, then q and
r are polynomials such that deg r < deg y and the leading terms of r and x have the same sign.
\/: rounded Euclidean quotient for integers (rounded towards +∞ when the exact quotient
would be a half-integer).
<<, >>: left and right binary shift. By definition, x<<n = x ∗ 2n if n > 0, and truncate(x2−n )
otherwise. Right shift is defined by x>>n = x<<(-n).
• Priority 5
+, -: addition/subtraction.
• Priority 4
<, >, <=, >=: the usual comparison operators, returning 1 for true and 0 for false. For
instance, x<=1 returns 1 if x ≤ 1 and 0 otherwise.
<>, !=: test for (exact) inequality.
==: test for (exact) equality. t_QFR having the same coefficients but a different distance
component are tested as equal.
===: test whether two objects are identical component-wise. This is stricter than ==: for
instance, the integer 0, a 0 polynomial or a vector with 0 entries, are all tested equal by ==, but
they are not identical.
• Priority 3
&&: logical and .
||: logical (inclusive) or . Any sequence of logical or and and operations is evaluated from left
to right, and aborted as soon as the final truth value is known. Thus, for instance,
x == 0 || test(1/x)
will never produce an error since test(1/x) is not even evaluated when the first test is true (hence
the final truth value is true). Similarly
type(p) == "t_INT" && isprime(p)
does not evaluate isprime(p) if p is not an integer.
• Priority 2
= (assignment, lvalue = expr ). The result of x = y is the value of the expression y, which
is also assigned to the variable x. This assignment operator is right-associative. This is not the
equality test operator; a statement like x = 1 is always true (i.e. nonzero), and sets x to 1; the
equality test would be x == 1. The right hand side of the assignment operator is evaluated before
the left hand side.

31
 It is crucial that the left hand-side be an lvalue there, it avoids ambiguities in expressions like
1 + x = 1. The latter evaluates as 1 + (x = 1), not as (1 + x) = 1, even though the priority
of = is lower than the priority of +: 1 + x is not an lvalue.
If the expression cannot be parsed in a way where the left hand side is an lvalue, raise an error.

? x + 1 = 1
*** syntax error, unexpected ’=’, expecting $end or ’;’: x+1=1
*** ^--
Assignment to all variables is a deep copy: after x = y, modifying a component of y will not change
x. To globals it is a full copy to the heap. Space used by local objects in local variables is released
when they go out of scope or when the value changes in local scope. Assigning a value to a vector
or matrix entry allocates room for that entry only (on the heap).
op=, where op is any binary operator among +, -, *, %, /, \, \/, <<, or >> (composed assignment
lvalue op= expr ). The expression x op= y assigns (x op y) to x, and returns the new value of x.
The result is not an lvalue; thus

(x += 2) = 3
is invalid. These assignment operators are right-associative:

? x = ’x; x += x *= 2
%1 = 3*x
• Priority 1
-> (function definition): (vars)->expr returns a function object, of type t_CLOSURE.

Remark. Use the op= operators as often as possible since they make complex assignments more
legible. Compare

v[i+j-1] = v[i+j-1] + 1 --> v[i+j-1]++

M[i,i+j] = M[i,i+j] * 2 --> M[i,i+j] *= 2

Remark about efficiency. the operators ++ and -- are usually a little more efficient than their
expended counterpart:

? N = 10^7;
? i = 0; for(k = 1, N, i=i+1)
time = 949 ms.
? i = 0; for(k = 1, N, i++)
time = 933 ms.
On the other hand, this is not the case for the op= operators which may even be a little less efficient:

? i = 0; for(k = 1, N, i=i+10)
time = 949 ms.
? i = 0; for(k = 1, N, i+=10)
time = 1,064 ms.

32
2.5 Variables and symbolic expressions.
In this section we use variable in the standard mathematical sense, symbols representing
algebraically independent elements used to build rings of polynomials and power series, and explain
the all-important concept of variable priority. In the next Section 2.6, we shall no longer consider
only free variables, but adopt the viewpoint of computer programming and assign values to these
symbols: (bound) variables are names attached to values in a given scope.

2.5.1 Variable names. A valid name starts with a letter, followed by any number of keyword
characters: or alphanumeric characters ([A-Za-z0-9]). As a rule, the built-in function names are
reserved
√ and cannot be used; see the list with \c, including the constants Pi, Euler, Catalan,
I = −1 and oo = ∞. Beware in particular of gamma, omega, theta, sum or O, none of which are
free to use. (We shall see in Section 2.6 how this rule can be circumvented. It is possible to name
a lexical variable gamma.)

GP names are case

√ sensitive. For instance, the symbol i is perfectly safe to use, and will not
be mistaken for I = −1; analogously, o is not synonymous to O.

In GP you can use up to 16383 variable names (up to 65535 on 64-bit machines). If you ever
need thousands of variables and this becomes a serious limitation, you should probably be using
vectors instead: e.g. instead of variables X1, X2, X3, . . . , you might equally well store their values
in X[1], X[2], X[3], . . .

2.5.2 Variables and polynomials. The quote operator ’t registers a new free variable with the
interpreter, which will be written as t, and evaluates to a monomial of degree 1 in the said variable.

Caveat. For reasons of backward compatibility, there is no such thing as an “unbound” (unini-
tialized) variable in GP. If you use a valid variable name in an expression, t say, for the first time
before assigning a value into it, it is interpreted as ’t rather than raising an exception. One should
not rely on this feature in serious programs, which would otherwise break if some unexpected as-
signment (e.g. t = 1) occurs: use ’t directly or t = ’t first, then t. A statement like t = ’t in
effect restores t as a free variable.

? t = ’t; t^2 + 1
%1 = t^2 + 1
? t = 2; t^2 + 1
%2 = 5
? %1
%3 = t^2 + 1
? eval(%1)
%4 = 5

In the above, we initialize t to a monomial, then bind it to 2. Assigning a value to a polynomial

variable does not affect previous expressions involving it; to take into account the new variable’s
value, one must force a new evaluation, using the function eval (see Section 3.9.6).

33
Caveat2. The use of an explicit quote operator avoids the following kind of problems:

? t = ’t; p = t^2 + 1; subst(p, t, 2)

%1 = 5
? t = 2;
? subst(p, t, 3) \\ t is no longer free: it evaluates to 2
*** at top-level: subst(p,t,3)
*** ^----
*** variable name expected.
? subst(p, ’t, 3) \\ OK
%3 = 10

2.5.3 Variable priorities, multivariate objects. A multivariate polynomial in PARI is just a

polynomial (in one variable), whose coefficients are themselves polynomials, arbitrary but for the
fact that they do not involve the main variable. (PARI currently has no sparse representation for
polynomials, listing only nonzero monomials.) All computations are then done formally on the
coefficients as if the polynomial was univariate.
This is not symmetrical. So if I enter ’x + ’y in a clean session, what happens? This is
understood as
x1 + (y 1 + 0 ∗ y 0 ) ∗ x0 ∈ (Z[y])[x]

but how do we know that x is “more important” than y ? Why not y 1 + x ∗ y 0 , which is the same
mathematical entity after all?
The answer is that variables are ordered implicitly by the interpreter: when a new identifier
(e.g x, or y as above) is input, the corresponding variable is registered as having a strictly lower
priority than any variable in use at this point*. To see the ordering used by gp at any given time,
type variable().
Given such an ordering, multivariate polynomials are stored so that the variable with the
highest priority is the main variable. And so on, recursively, until all variables are exhausted. A
different storage pattern (which could only be obtained via libpari programming and low-level
constructors) would produce an invalid object, and eventually a disaster.
In any case, if you are working with expressions involving several variables and want to have
them ordered in a specific manner in the internal representation just described, the simplest is just
to write down the variables one after the other under gp before starting any real computations.
You may also define variables from your gprc to have a consistent ordering of common variable
names in all your gp sessions, e.g read in a file variables.gp containing

’x; ’y; ’z; ’t; ’a;

There is no way to change the priority of existing variables, but you may always create new ones
with well-defined priorities using varhigher or varlower.

* This is not strictly true: the variables x and y are predefined, and satisfy x > y. Variables of
higher priority than x can be created using varhigher.

34
Important note. PARI allows Euclidean division of multivariate polynomials, but assumes that
the computation takes place in the fraction field of the coefficient ring (if it is not an integral
domain, the result will a priori not make sense). This can become tricky. For instance assume x
has highest priority, then y:
? x % y
%1 = 0
? y % x
%2 = y \\ these two take place in Q(y)[x]
? x * Mod(1,y)
%3 = Mod(1, y)*x \\ in (Q(y)/yQ(y))[x] ∼ Q[x]
? Mod(x,y)
%4 = 0
In the last example, the division by y takes place in Q(y)[x], hence the Mod object is a coset
in (Q(y)[x])/(yQ(y)[x]), which is the null ring since y is invertible! So be very wary of variable
ordering when your computations involve implicit divisions and many variables. This also affects
functions like numerator/denominator or content:
? denominator(x / y)
%1 = 1
? denominator(y / x)
%2 = x
? content(x / y)
%3 = 1/y
? content(y / x)
%4 = y
? content(2 / x)
%5 = 2
Can you see why? Hint: x/y = (1/y) ∗ x is in Q(y)[x] and denominator is taken with respect to
Q(y)(x); y/x = (y ∗ x0 )/x is in Q(y)(x) so y is invertible in the coefficient ring. On the other hand,
2/x involves a single variable and the coefficient ring is simply Z.
These problems arise because the variable ordering defines an implicit variable with respect
to which division takes place. This is the price to pay to allow % and / operators on polynomials
instead of requiring a more cumbersome divrem(x, y, var ) (which also exists). Unfortunately,
in some functions like content and denominator, there is no way to set explicitly a main variable
like in divrem and remove the dependence on implicit orderings. This will hopefully be corrected
in future versions.

2.5.4 Multivariate power series. Just like multivariate polynomials, power series are funda-
mentally single-variable objects. It is awkward to handle many variables at once, since PARI’s
implementation cannot handle multivariate error terms like O(xi y j ). (It can handle the polyno-
mial O(y j ) × xi which is a very different thing, see below.)
The basic assumption in our model is that if variable x has higher priority than y, then y does
not depend on x: setting y to a function of x after some computations with bivariate power series
does not make sense a priori. This is because implicit constants in expressions like O(xi ) depend
on y (whereas in O(y j ) they can not depend on x). For instance
? O(x) * y

35
 %1 = O(x)
? O(y) * x
%2 = O(y)*x
Here is a more involved example:
? A = 1/x^2 + 1 + O(x); B = 1/x + 1 + O(x^3);
? subst(z*A, z, B)
%2 = x^-3 + x^-2 + x^-1 + 1 + O(x)
? B * A
%3 = x^-3 + x^-2 + x^-1 + O(1)
? z * A
%4 = z*x^-2 + z + O(x)
The discrepancy between %2 and %3 is surprising. Why does %2 contain a spurious constant term,
which cannot be deduced from the input? Well, we ignored the rule that forbids to substitute
an expression involving high-priority variables to a low-priority variable. The result %4 is correct
according to our rules since the implicit constant in O(x) may depend on z. It is obviously wrong
if z is allowed to have negative valuation in x. Of course, the correct error term should be O(xz),
but this is not possible in PARI.

2.6 Variables and Scope.

This section is rather technical, and strives to explain potentially confusing concepts. Skip to
the last subsection for practical advice, if the next discussion does not make sense to you. After
learning about user functions, study the example in Section 2.7.3 then come back.

Definitions.
A scope is an enclosing context where names and values are attached. A user’s function body,
the body of a loop, an individual command line, all define scopes; the whole program defines the
global scope. The argument of eval is evaluated in the enclosing scope.
Variables are bound to values within a given scope. This is traditionally implemented in two
different ways:
• lexical (or static) scoping: the binding makes sense within a given block of program text.
The value is private to the block and may not be accessed from outside. Where to find the value
is determined at compile time.
• dynamic scoping: introducing a local variable, say x, pushes a new value on a stack attached
to the name x (possibly empty at this point), which is popped out when the control flow leaves the
scope. Evaluating x in any context, possibly outside of the given block, always yields the top value
on this dynamic stack.
GP implements both lexical and dynamic scoping, using the keywords* my (lexical) and local
(dynamic):
x = 0;
f() = x
g() = my(x = 1); f()

* The names are borrowed from the Perl scripting language.

36
 h() = local(x = 1); f()
The function g returns 0 since the global x binding is unaffected by the introduction of a private
variable of the same name in g. On the other hand, h returns 1; when it calls f(), the binding stack
for the x identifier contains two items: the global binding to 0, and the binding to 1 introduced in
h, which is still present on the stack since the control flow has not left h yet.
The rule mentionned in the previous section about built-in function names being reserved does
not apply to lexically scoped variables. Those may temporarily shadow an existing function name:
my(gamma = 0) ; ...
Without the my, this would be invalid since gamma is the Γ function.

2.6.1 Scoping rules.

Named parameters in a function definition, as well as all loop indices**, have lexical scope
within the function body and the loop body respectively.
p = 0;
forprime (p = 2, 11, print(p)); p \\ prints 0 at the end
x = 0;
f(x) = x++;
f(1) \\ returns 2, and leave global x unaffected (= 0)
If you exit the loop prematurely, e.g. using the break statement, you must save the loop index in
another variable since its value prior the loop will be restored upon exit. For instance
for(i = 1, n,
if (ok(i), break);
);
if (i > n, return(failure));
is incorrect, since the value of i tested by the (i > n) is quite unrelated to the loop index. One ugly
workaround is
for(i = 1, n,
if (ok(i), isave = i; break);
);
if (isave > n, return(failure));
But it is usually more natural to wrap the loop in a user function and use return instead of break:
try() =
{
for(i = 1, n,
if (ok(i), return (i));
);
0 \\ failure
}
A list of variables can be lexically or dynamically scoped (to the block between the declaration
and the end of the innermost enclosing scope) using a my or local declaration:

** More generally, in all iterative constructs which use a variable name (for, prod, sum, vector,
matrix, plot, etc.) the given variable is lexically scoped to the construct’s body.

37
 for (i = 1, 10,
my(x, y, z, i2 = i^2); \\ temps needed within the loop body
...
)

Note how the declaration can include (optional) initial values, i2 = i^2 in the above. Variables
for which no explicit default value is given in the declaration are initialized to 0. It would be more
natural to initialize them to free variables, but this would break backward compatibility. To obtain
this behavior, you may explicitly use the quoting operator:

my(x = ’x, y = ’y, z = ’z);

A more complicated example:

for (i = 1, 3,
print("main loop");
my(x = i); \\ local to the outermost loop
for (j = 1, 3,
my (y = x^2); \\ local to the innermost loop
print (y + y^2);
x++;
)
)

When we leave the loops, the values of x, y, i, j are the same as before they were started.

Note that eval is evaluated in the given scope, and can access values of lexical variables:

? x = 1;
? my(x = 0); eval("x")
%2 = 0 \\ we see the local x scoped to this command line, not the global one

Variables dynamically scoped using local should more appropriately be called temporary val-
ues since they are in fact local to the function declaring them and any subroutine called from
within. In practice, you almost certainly want true private variables, hence should use almost
exclusively my.

We strongly recommended to explicitly scope (lexically) all variables to the smallest possible
block. Should you forget this, in expressions involving such “rogue” variables, the value used will
be the one which happens to be on top of the value stack at the time of the call; which depends on
the whole calling context in a nontrivial way. This is in general not what you want.

38
2.7 User defined functions.
User-defined functions are ordinary GP objects, bound to variables just like any other object.
Those variables are subject to scoping rules as any other: while you can define all your functions
in global scope, it is usually possible and cleaner to lexically scope your private helper functions to
the block of text where they will be needed.
Whenever gp meets a construction of the form expr(argument list) and the expression expr
evaluates to a function (an object of type t_CLOSURE), the function is called with the proper
arguments. For instance, constructions like funcs[i](x) are perfectly valid, assuming funcs is an
array of functions.
As regards argument passing conventions, GP functions support both
• call by value: the function operates on a copy of a variable, changes made to the argument
in the function do not affect the original variable;
• and call by reference: the function receives a reference to the variable and original data is
affected.

2.7.1 Defining a function.

A user function is defined as follows:
(list of formal variables) -> seq.
The list of formal variables is a comma-separated list of distinct variable names and allowed to be
empty. It there is a single formal variable, the parentheses are optional. This list corresponds to
the list of parameters you will supply to your function when calling it. By default, GP functions
use call by value to pass arguments; a variable name may be prefixed by a tilde ~ to use instead a
call by reference.
In most cases you want to assign a function to a variable immediately, as in
R = (x,y) -> sqrt( x^2+y^2 );
sq = x -> x^2; \\ or equivalently (x) -> x^2
but it is quite possible to define short-lived anonymous functions. The trailing semicolon is not
part of the definition, but as usual prevents gp from printing the result of the evaluation, i.e. the
function object. The construction
f(list of formal variables) = seq
is available as an alias for
f = (list of formal variables) -> seq
Using that syntax, it is not possible to define anonymous functions (obviously), and the above two
examples become:
R(x,y) = sqrt( x^2+y^2 );
sq(x) = x^2;
The semicolon serves the same purpose as above: preventing the printing of the resulting function
object; compare
? sq(x) = x^2; \\ no output
? sq(x) = x^2 \\ print the result: a function object

39
 %2 = (x)->x^2
Of course, the sequence seq can be arbitrarily complicated, in which case it will look better written
on consecutive lines, with properly scoped variables:
{
f(x0, x1, . . . ) =
my(t0, t1, . . . ); \\ variables lexically scoped to the function body
...
}
Note that the following variant would also work:
f(x0, x1, . . . ) =
{
my(t0, t1, . . . ); \\ variables lexically scoped to the function body
...
}
(the first newline is disregarded due to the preceding = sign, and the others because of the enclosing
braces). The my statements can actually occur anywhere within the function body, scoping the
variables to more restricted blocks than the whole function body.
Formal parameters are lexically scoped to the function body. It is not allowed to use the same
variable name for different parameters of your function:
? f(x,x) = 1
*** variable declared twice: f(x,x)=1
*** ^----
By default, arguments are passed by value, not as variables: modifying a function’s argument
in the function body is allowed, but does not modify its value in the calling scope. In fact, a copy
of the actual parameter is assigned to the formal parameter when the function is called. (This is
not litterally true: a form of copy-on-write is implemented so an object is not duplicated unless
modified in the function.) If an argument is prefixed by a tilde ~ in the function declaration and
the call, it is passed by reference. (If either the declaration or the call is missing a tilde, we revert
to a call by value.)
? x = [1];
? f(v) = v[1]++;
? F(~v) = v[1]++;
? f(x)
%4 = 2
? x \\ unchanged
%5 = [1]
? F(~x)
%6 = 2
? x \\ incremented
%7 = [2]
? F(x) \\ forgot the ~: call by value
%8 = 3
? x \\ => contents of x did not change
%9 = [2]

40
 ? f(~x) \\ adding a ~ in call, missing in declaration
%10 = 3
? x \\ => call by value
%11 = [2]

Caveat. In GP, a call by reference means that the function accesses the value and may change
the original variable content, but only if it is a container type (a vector, list or matrice), as shown
above with the vector x. It will not alter its value in other cases !

? v = 1 \\ not a container
? F(~v) = v++
? F(~v)
%3 = 2
? v \\ components of v could be altered, not v itself
%4 = 1

Functions taking an unlimited number of arguments.

A function taking an unlimited number of arguments is called variadic. To create such a

function, use the syntax

(list of formal variables, var [..]) -> seq

The parameter var is replaced by a vector containing all the remaining arguments. The name may
not be prefixed by a tilde to (absurdly) indicate a call by reference.

? f(c[..]) = sum(i=1,#c,c[i]);
? f(1,2,3)
%1 = 6
? sep(s,v[..]) = for(i=1,#v-1,print1(v[i],s)); if (#v, print(v[#v]));
? sep(":", 1, 2, 3)
1:2:3

Finishing touch. You can add a specific help message for your function using addhelp, but the
online help system already handles it. By default ?name will print the definition of the function
name: the list of arguments, as well as their default values, the text of seq as you input it. Just as
\c prints the list of all built-in commands, \u outputs the list of all user-defined functions.

Backward compatibility (lexical scope). Lexically scoped variables were introduced in ver-
sion 2.4.2. Before that, the formal parameters were dynamically scoped. If your script depends on
this behavior, you may use the following trick: replace the initial f(x) = by

f(x_orig) = local(x = x_orig)

41
Backward compatibility (disjoint namespaces). Before version 2.4.2, variables and functions
lived in disjoint namespaces and it was not possible to have a variable and a function share the
same name. Hence the need for a kill function allowing to reuse symbols. This is no longer the
case.
There is now no distinction between variable and function names: we have PARI objects
(functions of type t_CLOSURE, or more mundane mathematical entities, like t_INT, etc.) and
variables bound to them. There is nothing wrong with the following sequence of assignments:
? f = 1 \\ assigns the integer 1 to f
%1 = 1;
? f() = 1 \\ a function with a constant value
%2 = ()->1
? f = x^2 \\ f now holds a polynomial
%3 = x^2
? f(x) = x^2 \\ . . . and now a polynomial function
%4 = (x)->x^2
? g(fun) = fun(Pi);\\ a function taking a function as argument
? g(cos)
%6 = -1.000000000000000000000000000

Previously used names can be recycled as above: you are just redefining the variable. The previous
definition is lost of course.

Important technical note. Built-in functions are a special case since they are read-only (you
cannot overwrite their default meaning), and they use features not available to user functions,
in particular pointer arguments. In the present version 2.13.3, it is possible to assign a built-in
function to a variable, or to use a built-in function name to create an anonymous function, but
some special argument combinations may not be available:
? issquare(9, &e)
%1 = 1
? e
%2 = 3
? g = issquare;
? g(9)
%4 = 1
? g(9, &e) \\ pointers are not implemented for user functions
*** unexpected &: g(9,&e)
*** ^---

2.7.2 Function call, Default arguments.

You may now call your function, as in f(1,2), supplying values for the formal variables.
The number of parameters actually supplied may be less than the number of formal variables in
the function definition. An uninitialized formal variable is given an implicit default value of (the
integer) 0, i.e. after the definition
f(x, y) = ...
you may call f(1, 2), supplying values for the two formal parameters, or for example
f(2) equivalent to f(2,0),

42
 f() f(0,0),
f(,3) f(0,3). (“Empty argument” trick)

This implicit default value of 0, is actually deprecated and setting

default(strictargs, 1)

allows to disable it (see Section 3.4.42).

The recommended practice is to explicitly set a default value: in the function definition, you
can append =expr to a formal parameter, to give that variable a default value. The expression gets
evaluated the moment the function is called, and may involve the preceding function parameters:
a default value for xi may involve xj for j < i. For instance, after

f(x = 1, y = 2, z = y+1) = ....

typing in f(3,4) would give you f(3,4,5). In the rare case when you want to set some far
away argument, and leave the defaults in between as they stand, use the “empty argument” trick:
f(6,,1) would yield f(6,2,1). Of course, f() by itself yields f(1,2,3) as was to be expected.

In short, the argument list is filled with user supplied values, in order. A comma or closing
parenthesis, where a value should have been, signals we must use a default value. When no input
arguments are left, the defaults are used instead to fill in remaining formal parameters. A final
example:

f(x, y=2, z=3) = print(x, ":", y, ":", z);

defines a function which prints its arguments (at most three of them), separated by colons.

? f(6,7)
6:7:3
? f(,5)
0:5:3
? f()
0:2:3

If strictargs is set (recommended), x is now a mandatory argument, and the above becomes:

? default(strictargs,1)
? f(6,7)
6:7:3
? f(,5)
*** at top-level: f(,5)
*** ^-----
*** in function f: x,y=2,z=3
*** ^---------
*** missing mandatory argument ’x’ in user function.

43
Example. We conclude with an amusing example, intended to illustrate both user-defined func-
tions and the power of the sumalt function. Although the Riemann zeta-function is included (as
zeta) among the standard functions, let us assume that we want to check other implementations.
Since we are highly interested in the critical strip, we use the classical formula
X
(21−s − 1)ζ(s) = (−1)n n−s , <s > 0.
n≥1

The implementation is obvious:

ZETA(s) = sumalt(n=1, (-1)^n*n^(-s)) / (2^(1-s) - 1)
Note that n is automatically lexically scoped to the sumalt “loop”, so that it is unnecessary to add
a my(n) declaration to the function body. Surprisingly, this gives very good accuracy in a larger
region than expected:
? check = z -> ZETA(z) / zeta(z);
? check(2)
%1 = 1.000000000000000000000000000
? check(200)
%2 = 1.000000000000000000000000000
? check(0)
%3 = 0.9999999999999999999999999994
? check(-5)
%4 = 1.00000000000000007549266557
? check(-11)
%5 = 0.9999752641047824902660847745
? check(1/2+14.134*I) \\ very close to a nontrivial zero
%6 = 1.000000000000000000003747432 + 7.62329066 E-21*I
? check(-1+10*I)
%7 = 1.000000000000000000000002511 + 2.989950968 E-24*I
Now wait a minute; not only are we summing a series which is certainly no longer alternating (it
has complex coefficients), but we are also way outside of the region of convergence, and still get
decent results! No programming mistake this time: sumalt is a “magic” function*, providing very
good convergence acceleration; in effect, we are computing the analytic continuation of our original
function. To convince ourselves that sumalt is a nontrivial implementation, let us try a simpler
example:
? sum(n=1, 10^7, (-1)^n/n, 0.) / (-log(2)) \\ approximates the well-known formula
time = 7,417 ms.
%1 = 0.9999999278652515622893405457
? sumalt(n=1, (-1)^n/n) / (-log(2)) \\ accurate and fast
time = 0 ms.
%2 = 1.000000000000000000000000000
No, we are not using a powerful simplification tool here, only numerical computations. Remember,
PARI is not a computer algebra system!

* sumalt is heuristic, but its use can be rigorously justified for a given function, in particular our
ζ(s) formula. Indeed, Peter Borwein (An efficient algorithm for the Riemann zeta function, CMS
Conf. Proc. 27 (2000), pp. 29–34) proved that the formula √ used in sumalt with n terms computes
(1 − 2 )ζ(s) with a relative error of the order of (3 + 8) |Γ(s)|−1 .
1−s −n

44
2.7.3 Beware scopes. Be extra careful with the scopes of variables. What is wrong with the
following definition?
FirstPrimeDiv(x) =
{ my(p);
forprime(p=2, x, if (x%p == 0, break));
p
}
? FirstPrimeDiv(10)
%1 = 0

Hint. The function body is equivalent to

{ my(newp = 0);
forprime(p=2, x, if (x%p == 0, break));
newp
}

Detailed explanation. The index p in the forprime loop is lexically scoped to the loop and is
not visible to the outside world. Hence, it will not survive the break statement. More precisely,
at this point the loop index is restored to its preceding value. The initial my(p), although well-
meant, adds to the confusion: it indeed scopes p to the function body, with initial value 0, but the
forprime loop introduces another variable, unfortunately also called p, scoped to the loop body,
which shadows the one we wanted. So we always return 0, since the value of the p scoped to the
function body never changes and is initially 0.
To sum up, the routine returns the p declared local to it, not the one which was local to
forprime and ran through consecutive prime numbers. Here is a corrected version:
? FirstPrimeDiv(x) = forprime(p=2, x, if (x%p == 0, return(p)))

2.7.4 Recursive functions. Recursive functions can easily be written as long as one pays proper
attention to variable scope. Here is an example, used to retrieve the coefficient array of a multivari-
ate polynomial (a nontrivial task due to PARI’s unsophisticated representation for those objects):

coeffs(P, nbvar) =
{
if (type(P) != "t_POL",
for (i=1, nbvar, P = [P]);
return (P)
);
vector(poldegree(P)+1, i, coeffs(polcoef(P, i-1), nbvar-1))
}
If P is a polynomial in k variables, show that after the assignment v = coeffs(P,k), the coefficient
of xn1 1 . . . xknk in P is given by v[n1 +1][. . . ][nk +1].
When the operating system allows querying the maximum size of the process stack, we auto-
matically limit the recursion depth:
? dive(n) = dive(n+1)

45
 ? dive(0);
*** [...] at: dive(n+1)
*** ^---------
*** in function dive: dive(n+1)
*** ^---------
\\ (last 2 lines repeated 19 times)
*** deep recursion.
All Unix variants support this mechanism and the recursion limit may be different from one machine
to the next; other systems may crash on deep recursion. There is no way to increase the limit from
within gp. On a Unix system, you may increase it before launching gp with ulimit or limit,
depending on your shell, and raise the process available stack space (increase stacksize).

2.7.5 Function which take functions as parameters. This is done as follows:

? calc(f, x) = f(x)
? calc(sin, Pi)
%2 = -5.04870979 E-29
? g(x) = x^2;
? calc(g, 3)
%4 = 9
If we do not need g elsewhere, we should use an anonymous function here, calc(x->x^2, 3). Here
is a variation:
? funs = [cos, sin, tan, x->x^3+1]; \\ an array of functions
? call(i, x) = funs[i](x)
evaluates the appropriate function on argument x, provided 1 ≤ i ≤ 4. Finally, a more useful
example:
APPLY(f, v) = vector(#v, i, f(v[i]))
applies the function f to every element in the vector v. (The built-in function apply is more
powerful since it also applies to lists and matrices.)

2.7.6 Defining functions within a function. Defining a single function is easy:

init(x) = (add = y -> x+y);
Basically, we are defining a global variable add whose value is the function y->x+y. The parentheses
were added for clarity and are not mandatory.
? init(5);
? add(2)
%2 = 7
A more refined approach is to avoid global variables and return the function:
init(x) = y -> x+y
add = init(5)
Then add(2) still returns 7, as expected! Of course, if add is in global scope, there is no gain, but
we can lexically scope it to the place where it is useful:
my ( add = init(5) );

46
 How about multiple functions then? We can use the last idea and return a vector of functions,
but if we insist on global variables? The first idea

init(x) = add(y) = x+y; mul(y) = x*y;

does not work since in the construction f() = seq, the function body contains everything until
the end of the expression. Hence executing init defines the wrong function add (itself defining a
function mul). The way out is to use parentheses for grouping, so that enclosed subexpressions will
be evaluated independently:

? init(x) = ( add(y) = x+y ); ( mul(y) = x*y );

? init(5);
? add(2)
%3 = 7
? mul(3)
%4 = 15

This defines two global functions which have access to the lexical variables private to init! The
following would work in exactly the same way:

? init5() = my(x = 5); ( add(y) = x+y ); ( mul(y) = x*y );

2.7.7 Closures as Objects. Contrary to what you might think after the preceding examples, GP’s
closures may not be used to simulate true “objects”, with private and public parts and methods
to access and manipulate them. In fact, closures indeed incorporate an existing context (they may
access lexical variables that existed at the time of their definition), but then may not change it.
More precisely, they access a copy, which they are welcome to change, but a further function call
still accesses the original context, as it existed at the time the function was defined:

init() =
{ my(count = 0);
(inc()=count++);
(dec()=count--);
}
? init();
? inc()
%1 = 1
? inc()
%2 = 1
? dec()
%3 = -1
? dec()
%4 = -1

47
2.8 Member functions.
Member functions use the ‘dot’ notation to retrieve information from complicated structures.
The built-in structures are bid, ell, galois, ff, nf, bnf, bnr and prid, which will be described at length
in Chapter 3. The syntax structure.member is taken to mean: retrieve member from structure,
e.g. E.j returns the j-invariant of the elliptic curve E, or outputs an error message if E is not a
proper ell structure. To define your own member functions, use the syntax
var .member = seq,
where the formal variable var is scoped to the function body seq. This is of course reminiscent of
a user function with a single formal variable var . For instance, the current implementation of the
ell type is a vector, the j-invariant being the thirteenth component. It could be implemented as
x.j =
{
if (type(x) != "t_VEC" || #x < 14, error("not an elliptic curve: " x));
x[13]
}
As for user functions, you can redefine your member functions simply by typing new definitions.
On the other hand, as a safety measure, you cannot redefine the built-in member functions, so
attempting to redefine x.j as above would in fact produce an error; you would have to call it
e.g. x.myj in order for gp to accept it.
Member functions use call by reference to pass arguments, your function may modify in place
the contents of a variable (of container type).
Rationale. In most cases, member functions are simple accessors of the form
x.a = x[1];
x.b = x[2];
x.c = x[3];
where x is a vector containing relevant data. There are at least three alternative approaches to the
above member functions: 1) hardcode x[1], etc. in the program text, 2) define constant global
variables AINDEX = 1, BINDEX = 2 and hardcode x[AINDEX], 3) user functions a(x) = x[1] and
so on.
Even if 2) improves on 1), these solutions are neither elegant nor flexible, and they scale badly.
3) is a genuine possibility, but the main advantage of member functions is that their namespace is
independent from the variables (and functions) namespace, hence we can use very short identifiers
without risk. The j-invariant is a good example: it would clearly not be a good idea to define j(E)
= E[13], because clashes with loop indices are likely.
Beware that there is no guarantee that a built-in member function is a simple accessor and it
could involve a computation. Thus you should not use them on a constant object in tight loops:
store them in a variable before the loop.
Note. Typing \um will output all user-defined member functions.
Member function names. A valid name starts with a letter followed by any number of keyword
characters: or alphanumeric characters ([A-Za-z0-9]). The built-in member function names are
reserved and cannot be used (see the list with ?.). Finally, names starting with e or E followed
by a digit are forbidden, due to a clash with the floating point exponent notation: we understand
1.e2 as 100.000 . . ., not as extracting member e2 of object 1.

48
2.9 Strings and Keywords.

2.9.1 Strings. GP variables can hold values of type character string (internal type t_STR). This
section describes how they are actually used, as well as some convenient tricks (automatic concate-
nation and expansion, keywords) valid in string context.
As explained above, the general way to input a string is to enclose characters between quotes ".
This is the only input construct where whitespace characters are significant: the string will contain
the exact number of spaces you typed in. Besides, you can “escape” characters by putting a \ just
before them; the translation is as follows
\e: <Escape>
\n: <Newline>
\t: <Tab>
For any other character x, \x is expanded to x. In particular, the only way to put a " into a
string is to escape it. Thus, for instance, "\"a\"" would produce the string whose content is “a”.
This is definitely not the same thing as typing "a", whose content is merely the one-letter string a.
You can concatenate two strings using the concat function. If either argument is a string, the
other is automatically converted to a string if necessary (it will be evaluated first).
? concat("ex", 1+1)
%1 = "ex2"
? a = 2; b = "ex"; concat(b, a)
%2 = "ex2"
? concat(a, b)
%3 = "2ex"
Some functions expect strings for some of their arguments: print would be an obvious example,
Str is a less obvious but useful one (see the end of this section for a complete list). While typing
in such an argument, you will be said to be in string context. The rest of this section is devoted to
special syntactical tricks which can be used with such arguments (and only here; you will get an
error message if you try these outside of string context):
• Writing two strings alongside one another will just concatenate them, producing a longer
string. Thus it is equivalent to type in "a " "b" or "a b". A little tricky point in the first
expression: the first whitespace is enclosed between quotes, and so is part of a string; while the
second (before the "b") is completely optional and gp actually suppresses it, as it would with any
number of whitespace characters at this point (i.e. outside of any string).
• If you insert any expression when a string is expected, it gets “expanded”: it is evaluated
as a standard GP expression, and the final result (as would have been printed if you had typed
it by itself) is then converted to a string, as if you had typed it directly. For instance "a" 1+1
"b" is equivalent to "a2b": three strings get created, the middle one being the expansion of 1+1,
and these are then concatenated according to the rule described above. Another tricky point here:
assume you did not assign a value to aaa in a GP expression before. Then typing aaa by itself in
a string context will actually produce the correct output (i.e. the string whose content is aaa), but
in a fortuitous way. This aaa gets expanded to the monomial of degree one in the variable aaa,
which is of course printed as aaa, and thus will expand to the three letters you were expecting.

49
Warning. Expression involving strings are not handled in a special way; even in string context,
the largest possible expression is evaluated, hence print("a"[1]) is incorrect since "a" is not an
object whose first component can be extracted. On the other hand print("a", [1]) is correct
(two distinct argument, each converted to a string), and so is print("a" 1) (since "a"1 is not
a valid expression, only "a" gets expanded, then 1, and the result is concatenated as explained
above).

2.9.2 Keywords. Since there are cases where expansion is not desirable, we now distinguish
between “Keywords” and “Strings”. String is what has been described so far. Keywords are
special relatives of Strings which are automatically assumed to be quoted, whether you actually
type in the quotes or not. Thus expansion is never performed on them. They get concatenated,
though. The analyzer supplies automatically the quotes you have “forgotten” and treats Keywords
just as normal strings otherwise. For instance, if you type "a"b+b in Keyword context, you will get
the string whose contents are ab+b. In String context, on the other hand, you would get a2*b.
All GP functions have prototypes (described in Chapter 3 below) which specify the types of
arguments they expect: either generic PARI objects (GEN), or strings, or keywords, or unevaluated
expression sequences. In the keyword case, only a very small set of words will actually be meaningful
(the default function is a prominent example).

Reference. The arguments of the following functions are processed in string context:
Str
addhelp (second argument)
default (second argument)
error
extern
plotstring (second argument)
plotterm (first argument)
read and readvec
system
all the printxxx functions
all the writexxx functions
The arguments of the following functions are processed as keywords:
alias
default (first argument)
install (all arguments but the last)
trap (first argument)
whatnow

2.9.3 Useful example. The function Str converts its arguments into strings and concatenate
them. Coupled with eval, it is very powerful. The following example creates generic matrices:
? genmat(u,v,s="x") = matrix(u,v,i,j, eval( Str(s,i,j) ))
? genmat(2,3) + genmat(2,3,"m")
%1 =
[x11 + m11 x12 + m12 x13 + m13]
[x21 + m21 x22 + m22 x23 + m23]

50
2.10 Errors and error recovery.

2.10.1 Errors. Your input program is first compiled to a more efficient bytecode; then the latter
is evaluated, calling appropriate functions from the PARI library. Accordingly, there are two kind
of errors: syntax errors produced by the compiler, and runtime errors produced by the PARI
library either by the evaluator itself, or in a mathematical function. Both kinds are fatal to your
computation: gp will report the error and perform some cleanup (restore variables modified while
evaluating the erroneous command, close open files, reclaim unused memory, etc.).
At this point, the default is to return to the usual prompt, but if the recover option (Sec-
tion 3.4.37) is off then gp exits immediately. This can be useful for batch-mode operation to make
untrapped errors fatal.
When reporting a syntax error , gp gives meaningful context by copying (part of) the expression
it was trying to compile, indicating where the error occurred with a caret ^-, as in
? factor()
*** too few arguments: factor()
*** ^-
? 1+
*** syntax error, unexpected $end: 1+
*** ^-
possibly enlarged to a full arrow given enough trailing context
? if (isprime(1+, do_something())
*** syntax error, unexpected ’,’: if(isprime(1+,do_something()))
*** ^----------------
These error messages may be mysterious, because gp cannot guess what you were trying to do, and
the error may occur once gp has been sidetracked. The first error is straightforward: factor has
one mandatory argument, which is missing.
The other two are simple typos involving an ill-formed addition 1 + missing its second
operand. The error messages differ because the parsing context is slightly different: in the first case
we reach the end of input ($end) while still expecting a token, and in the second one, we received
an unexpected token (the comma).
Here is a more complicated one:
? factor(x
*** syntax error, unexpected $end, expecting )-> or ’,’ or ’)’: factor(x
*** ^-
The error is a missing parenthesis, but from gp’s point of view, you might as well have intended to
give further arguments to factor (this is possible and useful, see the description of the function).
In fact gp expected either a closing parenthesis, or a second argument separated from the first by
a comma. And this is essentially what the error message says: we reached the end of the input
($end) while expecting a ’)’ or a ’,’.
Actually, a third possibility is mentioned in the error message )->, which could never be valid
in the above context, but a subexpression like (x)->sin(x), defining an inline closure would be
valid, and the parser is not clever enough to rule that out, so we get the same message as in
? (x

51
 *** syntax error, unexpected $end, expecting )-> or ’,’ or ’)’: (x
*** ^-

where all three proposed continuations would be valid.

Runtime errors from the evaluator are nicer because they answer a correctly worded query,
otherwise the bytecode compiler would have protested first; here is a slightly pathological case:

? if (siN(x) < eps, do_something())

*** at top-level: if(siN(x)<eps,do_someth
*** ^--------------------
*** not a function in function call

(no arrow!) The code is syntactically correct and compiled correctly, even though the siN function,
a typo for sin, was not defined at this point. When trying to evaluate the bytecode, however, it
turned out that siN is still undefined so we cannot evaluate the function call siN(x).

Library runtime errors are even nicer because they have more mathematical content, which is
easier to grasp than a parser’s logic:

? 1/Mod(2,4)
*** at top-level: 1/Mod(2,4)
*** ^---------
*** _/_: impossible inverse in Fp_inv: Mod(2, 4).

telling us that a runtime error occurred while evaluating the binary / operator (the surrounding
the operator are placeholders), more precisely the Fp inv library function was fed the argument
Mod(2,4) and could not invert it. More context is provided if the error occurs deep in the call
chain:

? f(x) = 1/x;
? g(N) = for(i = -N, N, f(i + O(5)));
? g(10)
*** at top-level: g(10)
*** ^-----
*** in function g: for(i=-N,N,f(i))
*** ^-----
*** in function f: 1/x
*** ^--
*** _/_: impossible inverse in ginv: O(5).

In this example, the debugger reports (at least) 3 enclosed frames: last (innermost) is the body of
user function f , the body of g, and the top-level (global scope). In fact, the for loop in g’s body
defines an extra frame, since there exist variables scoped to the loop body.

52
2.10.2 Error recovery.
It is annoying to wait for a program to finish and find out the hard way that there was a
mistake in it (like the division by 0 above), sending you back to the prompt. First you may lose
some valuable intermediate data. Also, correcting the error may not be obvious; you might have to
change your program, adding a number of extra statements and tests to narrow down the problem.
A different situation, still related to error recovery, is when you actually foresee that some
error may occur, are unable to prevent it, but quite capable of recovering from it, given the chance.
Examples include lazy factorization, where you knowingly use a pseudo prime N as if it were prime;
you may then encounter an “impossible” situation, but this would usually exhibit a factor of N ,
enabling you to refine the factorization and go on. Or you might run an expensive computation
at low precision to guess the size of the output, hence the right precision to use. You can then
encounter errors like “precision loss in truncation”, e.g when trying to convert 1E1000, known to
28 digits of accuracy, to an integer; or “division by 0”, e.g inverting 0E1000 when all accuracy has
been lost, and no significant digit remains. It would be enough to restart part of the computation
at a slightly higher precision.
We now describe error trapping, a useful mechanism which alleviates much of the pain in the
first situation (the break loop debugger), and provides satisfactory ways out of the second one (the
iferr exception handler).

2.10.3 Break loop.

A break loop is a special debugging mode that you enter whenever a user interrupt (Control-C)
or runtime error occurs, freezing the gp state, and preventing cleanup until you get out of the loop.
By runtime error, we mean an error from the evaluator, the library or a user error (from error),
not syntax errors. When a break loop starts, a prompt is issued (break>). You can type in a gp
command, which is evaluated when you hit the <Return> key, and the result is printed as during
the main gp loop, except that no history of results is kept. Then the break loop prompt reappears
and you can type further commands as long as you do not exit the loop. If you are using readline,
the history of commands is kept, and line editing is available as usual. If you type in a command
that results in an error, you are sent back to the break loop prompt: errors do not terminate the
loop.
To get out of a break loop, you can use next, break, return, or type C-d (EOF), any of which
will let gp perform its usual cleanup, and send you back to the gp prompt. Note that C-d is slightly
dangerous, since typing it twice will not only send you back to the gp prompt, but to your shell
prompt! (Since C-d at the gp prompt exits the gp session.)
If the break loop was started by a user interrupt Control-C, and not by an error, inputting an
empty line, i.e hitting the <Return> key at the break> prompt, resumes the temporarily interrupted
computation. A single empty line has no effect in case of a fatal error, to avoid getting get out of
the loop prematurely, thereby losing valuable debugging data. Any of next, break, return, or C-d
will abort the computation and send you back to the gp prompt as above.
Break loops are useful as a debugging tool. You may inspect the values of gp variables to
understand why an error occurred, or change gp’s state in the middle of a computation (increase
debugging level, start storing results in a log file, set variables to different values. . . ): hit C-c, type
in your modifications, then let the computation go on as explained above. A break loop looks like
this:
? v = 0; 1/v

53
 *** at top-level: v=0;1/v
*** ^--
*** _/_: impossible inverse in gdiv: 0.
*** Break loop (type ’break’ to go back to the GP prompt)
break>
So the standard error message is printed first. The break> at the bottom is a prompt, and hitting
v then <Return>, we see:
break> v
0
explaining the problem. We could have typed any gp command, not only the name of a variable,
of course. Lexically-scoped variables are accessible to the evaluator during the break loop:
? for(v = -2, 2, print(1/v))
-1/2
-1
*** at top-level: for(v=-2,2,print(1/v))
*** ^----
*** _/_: impossible inverse in gdiv: 0.
*** Break loop (type ’break’ to go back to the GP prompt)
break> v
0
Even though loop indices are automatically lexically scoped and no longer exist when the break
loop is run, enough debugging information is retained in the bytecode to reconstruct the evaluation
context. Of course, when the error occurs in a nested chain of user function calls, lexically scoped
variables are available only in the corresponding frame:
? f(x) = 1/x;
? g(x) = for(i = 1, 10, f(x+i));
? for(j = -5,5, g(j))
*** at top-level: for(j=-5,5,g(j))
*** ^-----
*** in function g: for(i=1,10,f(x+i))
*** ^-------
*** in function f: 1/x
*** ^--
*** _/_: impossible inverse in gdiv: 0.
*** Break loop: type ’break’ to go back to GP prompt
break> [i,j,x] \\ the x in f ’s body.
[i, j, 0]
break> dbg_up \\ go up one frame
*** at top-level: for(j=-5,5,g(j))
*** ^-----
*** in function g: for(i=1,10,f(x+i))
*** ^-------
break> [i,j,x] \\ the x in g’s body, i in the for loop.
[5, j, -5]
The following GP commands are available during a break loop to help debugging:

54
 dbg_up(n): go up n frames, as seen above.
dbg_down(n): go down n frames, cancelling previous dbg up’s.
dbg_x(t): examine t, as \x but more flexible.
dbg_err(): returns the current error context t_ERROR. The error components often provide
useful additional information:
? O(2) + O(3)
*** at top-level: O(2)+O(3)
*** ^-----
*** _+_: inconsistent addition t_PADIC + t_PADIC.
*** Break loop: type ’break’ to go back to GP prompt
break> E = dbg_err()
error("inconsistent addition t_PADIC + t_PADIC.")
break> Vec(E)
["e_OP", "+", O(2), O(3)]

Note. The debugger is enabled by default, and fires up as soon as a runtime error occurs. If you
do not like this behavior, you may disable it by setting the default breakloop to 0 in for gprc. A
runtime error will send you back to the prompt. Note that the break loop is automatically disabled
when running gp in non interactive mode, i.e. when the program’s standard input is not attached
to a terminal.

Technical Note. When you enter a break loop due to a PARI stack overflow, the PARI stack is
reset so that you can run commands. Otherwise the stack would immediately overflow again! Still,
as explained above, you do not lose the value of any gp variable in the process.

2.10.4 Protecting code. The expression

iferr(statements, ERR, recovery)
evaluates and returns the value of statements, unless an error occurs during the evaluation in which
case the value of recovery is returned. As in an if/else clause, with the difference that statements
has been partially evaluated, with possible side effects. We shall give a lot more details about
the ERR argument shortly; it is the name of a variable, lexically scoped to the recovery expression
sequence, whose value is set by the exception handler to help the recovery code decide what to do
about the error.
For instance one can define a fault tolerant inversion function as follows:
? inv(x) = iferr(1/x, ERR, "oo") \\ ERR is unused...
? for (i=-1,1, print(inv(i)))
-1
oo
1
Protected codes can be nested without adverse effect. Let’s now see how ERR can be used; as
written, inv is too tolerant:
? inv("blah")
%2 = "oo"

55
 Let’s improve it by checking that we caught a “division by 0” exception, and not an unrelated
one like the type error 1 / "blah".
? inv2(x) = {
iferr(1/x,
ERR, if (errname(ERR) != "e_INV", error(ERR)); "oo")
}
? inv2(0)
%3 = "oo" \\ as before
? inv2("blah")
*** at top-level: inv2("blah")
*** ^------------
*** in function inv2: ...f(errname(ERR)!="e_INV",error(ERR));"oo")
*** ^-----------------
*** error: forbidden division t_INT / t_STR.

In the inv2("blah") example, the error type was not expected, so we rethrow the exception:
error(ERR) triggers the original error that we mistakenly trapped. Since the recovery code should
always check whether the error is the one expected, this construction is very common and can be
simplified to
? inv3(x) = iferr(1/x,
ERR, "oo",
errname(ERR) == "e_INV")
More generally
iferr(statements, ERR, recovery, predicate)
only catches the exception if predicate (allowed to check various things about ERR, not only its
name) is nonzero.
Rather than trapping everything, then rethrowing whatever we do not like, we advise to only
trap errors of a specific kind, as above. Of course, sometimes, one just want to trap everything
because we do not know what to expect. The following function check whether install works
correctly in your gp:
broken_install() =
{ \\ can we install?
iferr(install(addii,GG),
ERR, return ("OS"));
\\ can we use the installed function?
iferr(if (addii(1,1) != 2, return("BROKEN")),
ERR, return("USE"));
return (0);
}
The function returns OS if the operating system does not support install, USE if using an installed
function triggers an error, BROKEN if the installed function did not behave as expected, and 0 if
everything works.
The ERR formal parameter contains more useful data than just the error name, which we
recovered using errname(ERR). In fact, a t_ERROR object usually has extra components, which can

56
be accessed as component(ERR,1), component(ERR,2), and so on. Or globally by casting the error
to a t_VEC: Vec(ERR) returns the vector of all components at once. See Section 3.1.24 for the list
of all exception types, and the corresponding contents of ERR.

2.11 Interfacing GP with other languages.

The PARI library was meant to be interfaced with C programs. This specific use is dealt with
extensively in the User’s guide to the PARI library. Of course, gp itself provides a convenient
interpreter to execute rather intricate scripts (see Section 3.1).
Scripts, when properly written, tend to be shorter and clearer than C programs, and are
certainly easier to write, maintain or debug. You don’t need to deal with memory management,
garbage collection, pointers, declarations, and so on. Because of their intrinsic simplicity, they
are more robust as well. They are unfortunately somewhat slower. Thus their use will remain
complementary: it is suggested that you test and debug your algorithms using scripts, before
actually coding them in C if speed is paramount. The GP2C compiler often eases this part.
The install command (see Section 3.2.40) efficiently imports foreign functions for use under
gp, which can of course be written using other libraries than PARI. Thus you may code only critical
parts of your program in C, and still maintain most of the program as a GP script.
We are aware of three PARI-related Free Software packages to embed PARI in other lan-
guages. We neither endorse nor support any of them, but you may want to give them a try if you
are familiar with the languages they are based on. The first is the Python-based SAGE system
(http://sagemath.org/). The second is the Math::Pari Perl module (see any CPAN mirror),
written by Ilya Zakharevich. Finally, Michael Stoll and Sam Steingold have integrated PARI into
CLISP (http://clisp.cons.org/), a Common Lisp implementation.
These provide interfaces to gp functions for use in python, perl, or Lisp programs, respectively.

2.12 Defaults.
There are many internal variables in gp, defining how the system will behave in certain situations,
unless a specific override has been given. Most of them are a matter of basic customization (colors,
prompt) and will be set once and for all in your preferences file (see Section 2.14), but some of
them are useful interactively (set timer on, increase precision, etc.).
The function used to manipulate these values is called default, which is described in Sec-
tion 3.2.12. The basic syntax is
default(def , value),
which sets the default def to value. In interactive use, most of these can be abbreviated using gp
metacommands (mostly, starting with \), which we shall describe in the next section.
Available defaults are described in the reference guide, Section 3.4, the most important one
being parisizemax. Just be aware that typing default by itself will list all of them, as well as
their current values (see \d).

57
Note. The suffixes k, M or G can be appended to a value which is a numeric argument, with the
effect of multiplying it by 103 , 106 and 109 respectively. Case is not taken into account there, so
for instance 30k and 30K both stand for 30000. This is mostly useful to modify or set the defaults
parisize and parisizemax which typically involve a lot of trailing zeroes.

(somewhat technical) Note. As we saw in Section 2.9, the second argument to default is
subject to string context expansion, which means you can use run-time values. In other words,
something like

a = 3;
default(logfile, "file" a ".log")

logs the output in file3.log.

Some special defaults, corresponding to file names and prompts, expand further the resulting
value at the time they are set. Two kinds of expansions may be performed:

• time expansion: the string is sent through the library function strftime. This means that
%char combinations have a special meaning, usually related to the time and date. For instance, %H
= hour (24-hour clock) and %M = minute [00,59] (on a Unix system, you can try man strftime at
your shell prompt to get a complete list). This is applied to prompt and logfile. For instance,

default(prompt,"(%H:%M) ? ")

will prepend the time of day, in the form (hh:mm) to gp’s usual prompt.

• environment expansion: When the string contains a sequence of the form $SOMEVAR,
e.g. $HOME, the environment is searched and if SOMEVAR is defined, the sequence is replaced by
the corresponding value. Also the ~ symbol has the same meaning as in many shells — ~ by itself
stands for your home directory, and ~user is expanded to user’s home directory. This is applied
to all file names.

2.13 Simple metacommands.

Simple metacommands are meant as shortcuts and should not be used in GP scripts (see Sec-
tion 3.1). Beware that these, as all of gp input, are case sensitive. For example, \Q is not identical
to \q. Two kinds of arguments are allowed: numbers (denoted n below) and names (denoted
filename below); braces are used to denote optional arguments, , e.g. {n} means that a numeric
argument is expected but can be omitted. Names can be optionally surrounded by double quotes
and in this case can contain whitespace, e.g. "a b" and are treated as ordinary character strings,
see Section 2.9 for details.

Whitespace (or spaces) between the metacommand and its arguments and within unquoted
arguments is optional. (This can cause problems with \w, when you insist on having a file name
whose first character is a digit, and with \r or \w, if the file name itself contains a space. In such
cases, just quote filenames or use the underlying read or write function).

58
2.13.1 ?{command }. The gp on-line help interface. If you type ?n where n is a number from 1 to
11, you will get the list of functions in Section 3.n of the manual (the list of sections being obtained
by simply typing ?).
These names are in general not informative enough. More details can be obtained by typing
?function, which gives a short explanation of the function’s calling convention and effects. A help
string is also attached to a symbolic operator, where arguments are replaced by a placeholder
character :
? ?sin
sin(x): sine of x.
? ?_*_
x*y: product of x and y.
? ?!_
!a: boolean operator "not".
? ?_!
n!: factorial of n.
? ? _^_
x^y: compute x to the power y.

Of course, to have complete information, read Chapter 3 of this manual. The source code is at
your disposal as well, though a trifle less readable.
If the line before the copyright message indicates that extended help is available (this means
perl is present on your system and the PARI distribution was correctly installed), you can add
more ? signs for extended functionality:
?? keyword yields the function description as it stands in this manual, usually in Chapter 2
or 3. If you’re not satisfied with the default chapter chosen, you can impose a given chapter by
ending the keyword with @ followed by the chapter number, e.g. ?? Hello@2 will look in Chapter 2
for section heading Hello (which doesn’t exist, by the way).
All operators (e.g. +, &&, etc.) are accepted by this extended help, as well as a few other
keywords describing key gp concepts, e.g. readline (the line editor), integer, nf (“number field”
as used in most algebraic number theory computations), ell (elliptic curves), etc.
In case of conflicts between function and default names (e.g log, simplify), the function has
higher priority. To get the default help, use
?? default(log)
?? default(simplify)
??? pattern produces a list of sections in Chapter 3 of the manual related to your query. As
before, if pattern ends by @ followed by a chapter number, that chapter is searched instead; you
also have the option to append a simple @ (without a chapter number) to browse through the whole
manual.
If your query contains dangerous characters (e.g ? or blanks) it is advisable to enclose it within
double quotes, as for GP strings (e.g ??? "elliptic curve").
Note that extended help is more powerful than the short help, since it knows about operators
as well: you can type ?? * or ?? &&, whereas a single ? would just yield a not too helpful

59
 &&: unknown identifier.
message. Also, you can ask for extended help on section number n in Chapter 3, just by typing
?? n (where ?n would yield merely a list of functions). Finally, a few key concepts in gp are
documented in this way: metacommands (e.g ?? "??"), defaults (e.g ?? default(log)) not to be
mistaken with ?? log (the natural logarithm) and type names (e.g t_INT or integer), as well as
various miscellaneous keywords such as edit (short summary of line editor commands), operator,
member, "user defined", nf, ell, . . .
Last but not least: ?? without argument will open a dvi previewer (xdvi by default, $GPXDVI
if it is defined in your environment) containing the full user’s manual. ??tutorial and ??refcard
do the same with the tutorial and reference card respectively.
Technical note. This functionality is provided by an external perl script that you are free to
use outside any gp session (and modify to your liking, if you are perl-knowledgeable). It is called
gphelp, lies in the doc subdirectory of your distribution (just make sure you run Configure first,
see Appendix A) and is really two programs in one. The one which is used from within gp is
gphelp which runs TEX on a selected part of this manual, then opens a previewer. gphelp -detex
is a text mode equivalent, which looks often nicer especially on a colour-capable terminal (see
misc/gprc.dft for examples). The default help selects which help program will be used from
within gp. You are welcome to improve this help script, or write new ones (and we would like to
know about it so that we may include them in future distributions). By the way, outside of gp you
can give more than one keyword as argument to gphelp.

2.13.2 /*...*/. A comment. Everything between the stars is ignored by gp. These comments
can span any number of lines.

2.13.3 \\. A one-line comment. The rest of the line is ignored by gp.

2.13.4 \a {n}. Prints the object number n (%n) in raw format. If the number n is omitted, print
the latest computed object (%).

2.13.5 \c. Prints the list of all available hardcoded functions under gp, not including opera-
tors written as special symbols (see Section 2.4). More information can be obtained using the ?
metacommand (see above). For user-defined functions / member functions, see \u and \um.

2.13.6 \d. Prints the defaults as described in the previous section (shortcut for default(), see
Section 3.2.12).

2.13.7 \e {n}. Switches the echo mode on (1) or off (0). If n is explicitly given, set echo to n.

2.13.8 \g {n}. Sets the debugging level debug to the nonnegative integer n.

2.13.9 \gf {n}. Sets the file usage debugging level debugfiles to the nonnegative integer n.

2.13.10 \gm {n}. Sets the memory debugging level debugmem to the nonnegative integer n.

2.13.11 \h {m-n}. Outputs some debugging info about the hashtable. If the argument is a number
n, outputs the contents of cell n. Ranges can be given in the form m-n (from cell m to cell n, $
= last cell). If a function name is given instead of a number or range, outputs info on the internal
structure of the hash cell this function occupies (a struct entree in C). If the range is reduced to
a dash (’-’), outputs statistics about hash cell usage.

60
2.13.12 \l {logfile}. Switches log mode on and off. If a logfile argument is given, change the
default logfile name to logfile and switch log mode on.

2.13.13 \m. As \a, but using prettymatrix format.

2.13.14 \o {n}. Sets output mode to n (0: raw, 1: prettymatrix, 3: external prettyprint).

2.13.15 \p {n}. Sets realprecision to n decimal digits. Prints its current value if n is omitted.

2.13.16 \pb {n}. Sets realbitprecision to n bits. Prints its current value if n is omitted.

2.13.17 \ps {n}. Sets seriesprecision to n significant terms. Prints its current value if n is
omitted.

2.13.18 \q. Quits the gp session and returns to the system. Shortcut for quit() (see Sec-
tion 3.2.61).

2.13.19 \r {filename}. Reads into gp all the commands contained in the named file as if they
had been typed from the keyboard, one line after the other. Can be used in combination with the
\w command (see below). Related but not equivalent to the function read (see Section 3.2.62); in
particular, if the file contains more than one line of input, there will be one history entry for each of
them, whereas read would only record the last one. If filename is omitted, re-read the previously
used input file (fails if no file has ever been successfully read in the current session). If a gp binary
file (see Section 3.2.86) is read using this command, it is silently loaded, without cluttering the
history.
Assuming gp figures how to decompress files on your machine, this command accepts com-
pressed files in compressed (.Z) or gzipped (.gz or .z) format. They will be uncompressed on
the fly as gp reads them, without changing the files themselves.

2.13.20 \s. Prints the state of the PARI stack and heap. This is used primarily as a debugging
device for PARI.

2.13.21 \t. Prints the internal longword format of all the PARI types. The detailed bit or byte
format of the initial codeword(s) is explained in Chapter 4, but its knowledge is not necessary for
a gp user.

2.13.22 \u. Prints the definitions of all user-defined functions.

2.13.23 \um. Prints the definitions of all user-defined member functions.

2.13.24 \v. Prints the version number and implementation architecture (680x0, Sparc, Alpha,
other) of the gp executable you are using.

2.13.25 \w {n} {filename}. Writes the object number n ( %n ) into the named file, in raw format.
If the number n is omitted, writes the latest computed object ( % ). If filename is omitted, appends
to logfile (the GP function write is a trifle more powerful, as you can have arbitrary file names).

61
2.13.26 \x {n}. Prints the complete tree with addresses and contents (in hexadecimal) of the
internal representation of the object number n ( %n ). If the number n is omitted, uses the latest
computed object in gp. As for \s, this is used primarily as a debugging device for PARI, and the
format should be self-explanatory. The underlying GP function dbg_x is more versatile, since it
can be applied to other objects than history entries.

2.13.27 \y {n}. Switches simplify on (1) or off (0). If n is explicitly given, set simplify to n.

2.13.28 #. Switches the timer on or off.

2.13.29 ##. Prints the time taken by the latest computation. Useful when you forgot to turn on
the timer.

2.14 The preferences file.

This file, called gprc in the sequel, is used to modify or extend gp default behavior, in all gp
sessions: e.g customize default values or load common user functions and aliases. gp opens the
gprc file and processes the commands in there, before doing anything else, e.g. creating the PARI
stack. If the file does not exist or cannot be read, gp will proceed to the initialization phase at
once, eventually emitting a prompt. If any explicit command line switches are given, they override
the values read from the preferences file.

2.14.1 Syntax. The syntax in the gprc file (and valid in this file only) is simple-minded, but
should be sufficient for most purposes. The file is read line by line; as usual, white space is ignored
unless surrounded by quotes and the standard multiline constructions using braces, \, or = are
available (multiline comments between /* . . . */ are also recognized).

2.14.1.1 Preprocessor:. Two types of lines are first dealt with by a preprocessor:
• comments are removed. This applies to all text surrounded by /* . . . */ as well as to
everything following \\ on a given line.
• lines starting with #if boolean are treated as comments if boolean evaluates to false, and
read normally otherwise. The condition can be negated using either #if not (or #if !). If the
rest of the current line is empty, the test applies to the next line (same behavior as = under gp).
The following tests can be performed:
EMACS: true if gp is running in an Emacs or TeXmacs shell (see Section 2.16).
READL: true if gp is compiled with readline support (see Section 2.15).
VERSION op number : where op is in the set {>, <, <=, >=}, and number is a PARI version
number of the form Major .Minor .patch, where the last two components can be omitted (i.e. 1 is
understood as version 1.0.0). This is true if gp’s version number satisfies the required inequality.
BITS IN LONG == number : number is 32 (resp. 64). This is true if gp was built for a 32-bit
(resp. 64-bit) architecture.

62
2.14.1.2 Commands:. After preprocessing, the remaining lines are executed as sequence of ex-
pressions (as usual, separated by ; if necessary). Only two kinds of expressions are recognized:
• default = value, where default is one of the available defaults (see Section 2.12), which will
be set to value on actual startup. Don’t forget the quotes around strings (e.g. for prompt or help).
• read "some GP file" where some GP file is a regular GP script this time, which will be
read just before gp prompts you for commands, but after initializing the defaults. In particular,
file input is delayed until the gprc has been fully loaded. This is the right place to input files
containing alias commands, or your favorite macros.
For instance you could set your prompt in the following portable way:
\\ self modifying prompt looking like (18:03) gp >
prompt = "(%H:%M) \e[1mgp\e[m > "
\\ readline wants nonprinting characters to be braced between ^A/^B pairs
#if READL prompt = "(%H:%M) ^A\e[1m^Bgp^A\e[m^B > "
\\ escape sequences not supported under emacs
#if EMACS prompt = "(%H:%M) gp > "
Note that any of the last two lines could be broken in the following way
#if EMACS
prompt = "(%H:%M) gp > "
since the preprocessor directive applies to the next line if the current one is empty.
A sample gprc file called misc/gprc.dft is provided in the standard distribution. It is a good
idea to have a look at it and customize it to your needs. Since this file does not use multiline
constructs, here is one (note the terminating ; to separate the expressions):
#if VERSION > 2.2.3
{
read "my_scripts"; \\ syntax errors in older versions
new_galois_format = 1; \\ default introduced in 2.2.4
}
#if ! EMACS
{
colors = "9, 5, no, no, 4, 1, 2";
help = "gphelp -detex -ch 4 -cb 0 -cu 2";
}

2.14.2 The gprc location. When gp is started, it looks for a customization file, or gprc in the
following places (in this order, only the first one found will be loaded):
• gp checks whether the environment variable GPRC is set. On Unix, this can be done with something
like:
GPRC=/my/dir/anyname; export GPRC in sh syntax (for instance in your .profile),
setenv GPRC /my/dir/anyname in csh syntax (in your .login or .cshrc file).
env GPRC=/my/dir/anyname gp on the command line launching gp.
If so, the file named by $GPRC is the gprc.
• If GPRC is not set, and if the environment variable HOME is defined, gp then tries

63
 $HOME/.gprc on a Unix system
$HOME\gprc.txt on a DOS, OS/2, or Windows system.
• If no gprc was found among the user files mentioned above we look for /etc/gprc for a system-
wide gprc file (you will need root privileges to set up such a file yourself).
• Finally, we look in pari’s datadir for a file named
.gprc on a Unix system
gprc.txt on a DOS, OS/2, or Windows system. If you are using our Windows installer, this
is where the default preferences file is written.
Note that on Unix systems, the gprc’s default name starts with a ’.’ and thus is hidden to regular
ls commands; you need to type ls -a to list it.

2.15 Using readline.

This very useful library provides line editing and contextual completion to gp. You are en-
couraged to read the readline user manual, but we describe basic usage here.

A (too) short introduction to readline. In the following, C- stands for “the Control key
combined with another” and the same for M- with the Meta key; generally C- combinations act
on characters, while the M- ones operate on words. The Meta key might be called Alt on some
keyboards, will display a black diamond on most others, and can safely be replaced by Esc in any
case.
Typing any ordinary key inserts text where the cursor stands, the arrow keys enabling you
to move in the line. There are many more movement commands, which will be familiar to the
Emacs user, for instance C-a/C-e will take you to the start/end of the line, M-b/M-f move the
cursor backward/forward by a word, etc. Just press the <Return> key at any point to send your
command to gp.
All the commands you type at the gp prompt are stored in a history, a multiline command
being saved as a single concatenated line. The Up and Down arrows (or C-p/C-n) will move you
through the history, M-</M-> sending you to the start/end of the history. C-r/C-s will start an
incremental backward/forward search. You can kill text (C-k kills till the end of line, M-d to the
end of current word) which you can then yank back using the C-y key (M-y will rotate the kill-ring).
C- will undo your last changes incrementally (M-r undoes all changes made to the current line).
C-t and M-t will transpose the character (word) preceding the cursor and the one under the cursor.
Keeping the M- key down while you enter an integer (a minus sign meaning reverse behavior)
gives an argument to your next readline command (for instance M-- C-k will kill text back to the
start of line). If you prefer Vi–style editing, M-C-j will toggle you to Vi mode.
Of course you can change all these default bindings. For that you need to create a file named
.inputrc in your home directory. For instance (notice the embedding conditional in case you would
want specific bindings for gp):
$if Pari-GP
set show-all-if-ambiguous
"\C-h": backward-delete-char
"\e\C-h": backward-kill-word

64
 "\C-xd": dump-functions
(: "\C-v()\C-b" # can be annoying when copy-pasting!
[: "\C-v[]\C-b"
$endif

C-x C-r will re-read this init file, incorporating any changes made to it during the current session.

Note. By default, ( and [ are bound to the function pari-matched-insert which, if “electric
parentheses” are enabled (default: off) will automatically insert the matching closure (respectively
) and ]). This behavior can be toggled on and off by giving the numeric argument −2 to ( (M--2(),
which is useful if you want, e.g to copy-paste some text into the calculator. If you do not want a
toggle, you can use M--0 / M--1 to specifically switch it on or off).

Note. In some versions of readline (2.1 for instance), the Alt or Meta key can give funny re-
sults (output 8-bit accented characters for instance). If you do not want to fall back to the Esc
combination, put the following two lines in your .inputrc:

set convert-meta on
set output-meta off

Command completion and online help. Hitting <TAB> will complete words for you. This
mechanism is context-dependent: gp will strive to only give you meaningful completions in a given
context (it will fail sometimes, but only under rare and restricted conditions).

For instance, shortly after a ~, we expect a user name, then a path to some file. Directly after
default( has been typed, we would expect one of the default keywords. After a ’.’, we expect a
member keyword. And generally of course, we expect any GP symbol which may be found in the
hashing lists: functions (both yours and GP’s), and variables.

If, at any time, only one completion is meaningful, gp will provide it together with

• an ending comma if we are completing a default,

• a pair of parentheses if we are completing a function name. In that case hitting <TAB> again
will provide the argument list as given by the online help. (Recall that you can always undo the
effect of the preceding keys by hitting C- ; this applies here.)

Otherwise, hitting <TAB> once more will give you the list of possible completions. Just ex-
periment with this mechanism as often as possible, you will probably find it very convenient. For
instance, you can obtain default(seriesprecision,10), just by hitting def<TAB>se<TAB>10,
which saves 18 keystrokes (out of 27).

Hitting M-h will give you the usual short online help concerning the word directly beneath the
cursor, M-H will yield the extended help corresponding to the help default program (usually opens
a dvi previewer, or runs a primitive tex-to-ASCII program). None of these disturb the line you
were editing.

65
2.16 GNU Emacs and PariEmacs.
If you install the PariEmacs package (see Appendix A), you may use gp as a subprocess in
Emacs. You then need to include in your .emacs file the following lines:
(autoload ’gp-mode "pari" nil t)
(autoload ’gp-script-mode "pari" nil t)
(autoload ’gp "pari" nil t)
(autoload ’gpman "pari" nil t)
(setq auto-mode-alist
(cons ’("\\.gp$" . gp-script-mode) auto-mode-alist))
which autoloads functions from the PariEmacs package and ensures that file with the .gp suffix
are edited in gp-script mode.
Once this is done, under GNU Emacs if you type M-x gp (where as usual M is the Meta key), a
special shell will be started launching gp with the default stack size and prime limit. You can then
work as usual under gp, but with all the facilities of an advanced text editor. See the PariEmacs
documentation for customizations, menus, etc.

66
 Chapter 3:
Functions and Operations Available in PARI and GP

The functions and operators available in PARI and in the GP/PARI calculator are numerous and
ever-expanding. Here is a description of the ones available in version 2.13.3. It should be noted that
many of these functions accept quite different types as arguments, but others are more restricted.
The list of acceptable types will be given for each function or class of functions. Except when stated
otherwise, it is understood that a function or operation which should make natural sense is legal.
On the other hand, many routines list explicit preconditions for some of their argument, e.g.
p is a prime number, or q is a positive definite quadratic form. For reasons of efficiency, all trust
the user input and only perform minimal sanity checks. When a precondition is not satisfied, any
of the following may occur: a regular exception is raised, the PARI stack overflows, a SIGSEGV or
SIGBUS signal is generated, or we enter an infinite loop. The function can also quietly return a
mathematically meaningless result: junk in, junk out.
In this chapter, we will describe the functions according to a rough classification. The general
entry looks something like:
foo(x, {flag = 0}): short description.
The library syntax is GEN foo(GEN x, long fl = 0).
This means that the GP function foo has one mandatory argument x, and an optional one, flag,
whose default value is 0. (The {} should not be typed, it is just a convenient notation we will use
throughout to denote optional arguments.) That is, you can type foo(x,2), or foo(x), which is
then understood to mean foo(x,0). As well, a comma or closing parenthesis, where an optional
argument should have been, signals to GP it should use the default. Thus, the syntax foo(x,) is
also accepted as a synonym for our last expression. When a function has more than one optional
argument, the argument list is filled with user supplied values, in order. When none are left, the
defaults are used instead. Thus, assuming that foo’s prototype had been

foo({x = 1}, {y = 2}, {z = 3}),

typing in foo(6,4) would give you foo(6,4,3). In the rare case when you want to set some far
away argument, and leave the defaults in between as they stand, you can use the “empty arg”
trick alluded to above: foo(6,,1) would yield foo(6,2,1). By the way, foo() by itself yields
foo(1,2,3) as was to be expected.
In this rather special case of a function having no mandatory argument, you can even omit
the (): a standalone foo would be enough (though we do not recommend it for your scripts, for
the sake of clarity). In defining GP syntax, we strove to put optional arguments at the end of the
argument list (of course, since they would not make sense otherwise), and in order of decreasing
usefulness so that, most of the time, you will be able to ignore them.
Finally, an optional argument (between braces) followed by a star, like {x }∗, means that any
number of such arguments (possibly none) can be given. This is in particular used by the various
print routines.

67
Flags. A flag is an argument which, rather than conveying actual information to the routine,
instructs it to change its default behavior, e.g. return more or less information. All such flags are
optional, and will be called flag in the function descriptions to follow. There are two different kind
of flags

• generic: all valid values for the flag are individually described (“If flag is equal to 1, then. . . ”).

• binary: use customary binary notation as a compact way to represent many toggles with
just one integer. Let (p0 , . . . , pn ) be a list of switches (i.e. of properties which take either the value
0 or 1), the number 23 + 25 = 40 means that p3 and p5 are set (that is, set to 1), and none of the
others are (that is, they are set to 0). This is announced as “The binary digits of flag mean 1: p0 ,
2: p1 , 4: p2 ”, and so on, using the available consecutive powers of 2.

Mnemonics for binary flags. Numeric flags as mentioned above are obscure, error-prone, and
quite rigid: should the authors want to adopt a new flag numbering scheme, it would break backward
compatibility. The only advantage of explicit numeric values is that they are fast to type, so their
use is only advised when using the calculator gp.

As an alternative, one can replace a binary flag by a character string containing symbolic
identifiers (mnemonics). In the function description, mnemonics corresponding to the various
toggles are given after each of them. They can be negated by prepending no to the mnemonic, or by
removing such a prefix. These toggles are grouped together using any punctuation character (such
as ’,’ or ’;’). For instance (taken from description of ploth(X = a, b, expr , {flag = 0}, {n = 0}))
Binary digits of flags mean: 1 = Parametric, 2 = Recursive, . . .

so that, instead of 1, one could use the mnemonic "Parametric; no Recursive", or simply "Para-
metric" since Recursive is unset by default (default value of flag is 0, i.e. everything unset).
People used to the bit-or notation in languages like C may also use the form "Parametric |
no Recursive".

Pointers. If a parameter in the function prototype is prefixed with a & sign, as in

foo(x, &e)

it means that, besides the normal return value, the function may assign a value to e as a side effect.
When passing the argument, the & sign has to be typed in explicitly. As of version 2.13.3, this
pointer argument is optional for all documented functions, hence the & will always appear between
brackets as in Z issquare(x, {&e}).

About library programming. The library function foo, as defined at the beginning of this
section, is seen to have two mandatory arguments, x and flag: no function seen in the present
chapter has been implemented so as to accept a variable number of arguments, so all arguments
are mandatory when programming with the library (usually, variants are provided corresponding
to the various flag values). We include an = default value token in the prototype to signal how
a missing argument should be encoded. Most of the time, it will be a NULL pointer, or -1 for a
variable number. Refer to the User’s Guide to the PARI library for general background and details.

68
3.1 Programming in GP: control statements.
A number of control statements are available in GP. They are simpler and have a syntax
slightly different from their C counterparts, but are quite powerful enough to write any kind of
program. Some of them are specific to GP, since they are made for number theorists. As usual,
X will denote any simple variable name, and seq will always denote a sequence of expressions,
including the empty sequence.

Caveat. In constructs like

for (X = a,b, seq)

the variable X is lexically scoped to the loop, leading to possibly unexpected behavior:

n = 5;
for (n = 1, 10,
if (something_nice(), break);
);
\\ at this point n is 5 !

If the sequence seq modifies the loop index, then the loop is modified accordingly:

? for (n = 1, 10, n += 2; print(n))

3
6
9
12

3.1.1 break({n = 1}). Interrupts execution of current seq, and immediately exits from the n
innermost enclosing loops, within the current function call (or the top level loop); the integer n
must be positive. If n is greater than the number of enclosing loops, all enclosing loops are exited.

3.1.2 breakpoint(). Interrupt the program and enter the breakloop. The program continues when
the breakloop is exited.

? f(N,x)=my(z=x^2+1);breakpoint();gcd(N,z^2+1-z);
? f(221,3)
*** at top-level: f(221,3)
*** ^--------
*** in function f: my(z=x^2+1);breakpoint();gcd(N,z
*** ^--------------------
*** Break loop: type <Return> to continue; ’break’ to go back to GP
break> z
10
break>
%2 = 13

69
3.1.3 dbg down({n = 1}). (In the break loop) go down n frames. This allows to cancel a previous
call to dbg up.
? x = 0;
? g(x) = x-3;
? f(x) = 1 / g(x+1);
? for (x = 1, 5, f(x+1))
*** at top-level: for(x=1,5,f(x+1))
*** ^-------
*** in function f: 1/g(x+1)
*** ^-------
*** _/_: impossible inverse in gdiv: 0.
*** Break loop: type ’break’ to go back to GP prompt
break> dbg_up(3) \\ go up 3 frames
*** at top-level: for(x=1,5,f(x+1))
*** ^-----------------
break> x
0
break> dbg_down()
*** at top-level: for(x=1,5,f(x+1))
*** ^-------
break> x
1
break> dbg_down()
*** at top-level: for(x=1,5,f(x+1))
*** ^-------
break> x
1
break> dbg_down()
*** at top-level: for(x=1,5,f(x+1))
*** ^-------
*** in function f: 1/g(x+1)
*** ^-------
break> x
2
The above example shows that the notion of GP frame is finer than the usual stack of function
calls (as given for instance by the GDB backtrace command): GP frames are attached to variable
scopes and there are frames attached to control flow instructions such as a for loop above.

3.1.4 dbg err(). In the break loop, return the error data of the current error, if any. See iferr
for details about error data. Compare:
? iferr(1/(Mod(2,12019)^(6!)-1),E,Vec(E))
%1 = ["e_INV", "Fp_inv", Mod(119, 12019)]
? 1/(Mod(2,12019)^(6!)-1)
*** at top-level: 1/(Mod(2,12019)^(6!)-
*** ^--------------------
*** _/_: impossible inverse in Fp_inv: Mod(119, 12019).
*** Break loop: type ’break’ to go back to GP prompt

70
 break> Vec(dbg_err())
["e_INV", "Fp_inv", Mod(119, 12019)]

3.1.5 dbg up({n = 1}). (In the break loop) go up n frames, which allows to inspect data of the
parent function. To cancel a dbg_up call, use dbg_down.
? x = 0;
? g(x) = x-3;
? f(x) = 1 / g(x+1);
? for (x = 1, 5, f(x+1))
*** at top-level: for(x=1,5,f(x+1))
*** ^-------
*** in function f: 1/g(x+1)
*** ^-------
*** _/_: impossible inverse in gdiv: 0.
*** Break loop: type ’break’ to go back to GP prompt
break> x
2
break> dbg_up()
*** at top-level: for(x=1,5,f(x+1))
*** ^-------
break> x
1
break> dbg_up()
*** at top-level: for(x=1,5,f(x+1))
*** ^-------
break> x
1
break> dbg_up()
*** at top-level: for(x=1,5,f(x+1))
*** ^-----------------
break> x
0
break> dbg_down() \\ back up once
*** at top-level: for(x=1,5,f(x+1))
*** ^-------
break> x
1
The above example shows that the notion of GP frame is finer than the usual stack of function
calls (as given for instance by the GDB backtrace command): GP frames are attached to variable
scopes and there are frames attached to control flow instructions such as a for loop above.

3.1.6 dbg x(A, {n}). Print the inner structure of A, complete if n is omitted, up to level n otherwise.
This is useful for debugging. This is similar to \x but does not require A to be an history entry. In
particular, it can be used in the break loop.

3.1.7 for(X = a, b, seq). Evaluates seq, where the formal variable X goes from a to b. Nothing is
done if a > b. a and b must be in R. If b is set to +oo, the loop will not stop; it is expected that
the caller will break out of the loop itself at some point, using break or return.

71
3.1.8 forcomposite(n = a, {b}, seq). Evaluates seq, where the formal variable n ranges over the
composite numbers between the nonnegative real numbers a to b, including a and b if they are
composite. Nothing is done if a > b.

? forcomposite(n = 0, 10, print(n))

4
6
8
9
10

Omitting b means we will run through all composites ≥ a, starting an infinite loop; it is expected
that the user will break out of the loop himself at some point, using break or return.

Note that the value of n cannot be modified within seq:

? forcomposite(n = 2, 10, n = [])

*** at top-level: forcomposite(n=2,10,n=[])
*** ^---
*** index read-only: was changed to [].

3.1.9 fordiv(n, X, seq). Evaluates seq, where the formal variable X ranges through the divisors of
n (see divisors, which is used as a subroutine). It is assumed that factor can handle n, without
negative exponents. Instead of n, it is possible to input a factorization matrix, i.e. the output of
factor(n).

This routine uses divisors as a subroutine, then loops over the divisors. In particular, if n is
an integer, divisors are sorted by increasing size.

To avoid storing all divisors, possibly using a lot of memory, the following (slower) routine
loops over the divisors using essentially constant space:

FORDIV(N)=
{ my(F = factor(N), P = F[,1], E = F[,2]);
forvec(v = vector(#E, i, [0,E[i]]), X = factorback(P, v));
}
? for(i=1, 10^6, FORDIV(i))
time = 11,180 ms.
? for(i=1, 10^6, fordiv(i, d, ))
time = 2,667 ms.

Of course, the divisors are no longer sorted by inreasing size.

72
3.1.10 fordivfactored(n, X, seq). Evaluates seq, where the formal variable X ranges through
[d, factor(d)], where d is a divisors of n (see divisors, which is used as a subroutine). Note that
such a pair is accepted as argument to all multiplicative functions.

It is assumed that factor can handle n, without negative exponents. Instead of n, it is possible
to input a factorization matrix, i.e. the output of factor(n). This routine uses divisors(, 1) as
a subroutine, then loops over the divisors. In particular, if n is an integer, divisors are sorted by
increasing size.

This function is particularly useful when n is hard to factor and one must evaluate multiplica-
tive function on its divisors: we avoid refactoring each divisor in turn. It also provides a small
speedup when n is easy to factor; compare

? A = 10^8; B = A + 10^5;
? for (n = A, B, fordiv(n, d, eulerphi(d)));
time = 2,091 ms.
? for (n = A, B, fordivfactored(n, d, eulerphi(d)));
time = 1,298 ms. \\ avoid refactoring the divisors
? forfactored (n = A, B, fordivfactored(n, d, eulerphi(d)));
time = 1,270 ms. \\ also avoid factoring the consecutive n’s !

3.1.11 foreach(V, X, seq). Evaluates seq, where the formal variable X ranges through the com-
ponents of V (t_VEC, t_COL, t_LIST or t_MAT). A matrix argument is interpreted as a vector
containing column vectors, as in Vec(V ).

3.1.12 forell(E, a, b, seq, {flag = 0}). Evaluates seq, where the formal variable E = [name, M, G]
ranges through all elliptic curves of conductors from a to b. In this notation name is the curve
name in Cremona’s elliptic curve database, M is the minimal model, G is a Z-basis of the free part
of the Mordell-Weil group E(Q). If flag is nonzero, select only the first curve in each isogeny class.

? forell(E, 1, 500, my([name,M,G] = E); \

if (#G > 1, print(name)))
389a1
433a1
446d1
? c = 0; forell(E, 1, 500, c++); c \\ number of curves
%2 = 2214
? c = 0; forell(E, 1, 500, c++, 1); c \\ number of isogeny classes
%3 = 971

The elldata database must be installed and contain data for the specified conductors.

The library syntax is forell(void *data, long (*f)(void*,GEN), long a, long b, long
flag).

73
3.1.13 forfactored(N = a, b, seq). Evaluates seq, where the formal variable N is [n, factor(n)]
and n goes from a to b; a and b must be integers. Nothing is done if a > b.

This function is√only implemented for |a|, |b| < 264 (232 on a 32-bit machine). It uses a sieve
and runs in time O( b + b − a). It should be at least √ 3 times faster than regular factorization as
long as the interval length b − a is much larger than b and get√relatively faster as the bounds
increase. The function slows down dramatically if primelimit < b.

? B = 10^9;
? for (N = B, B+10^6, factor(N))
time = 4,538 ms.
? forfactored (N = B, B+10^6, [n,fan] = N)
time = 1,031 ms.
? B = 10^11;
? for (N = B, B+10^6, factor(N))
time = 15,575 ms.
? forfactored (N = B, B+10^6, [n,fan] = N)
time = 2,375 ms.
? B = 10^14;
? for (N = B, B+10^6, factor(N))
time = 1min, 4,948 ms.
? forfactored (N = B, B+10^6, [n,fan] = N)
time = 58,601 ms.
√
The last timing is with the default primelimit (500000) which is much less than √ B + 106 ; it
goes down to 26,750ms if primelimit gets bigger than that bound. In any case B + 106 is much
larger than the interval length 106 so forfactored gets relatively slower for that reason as well.

Note that all PARI multiplicative functions accept the [n,fan] argument natively:

? s = 0; forfactored(N = 1, 10^7, s += moebius(N)*eulerphi(N)); s

time = 6,001 ms.
%1 = 6393738650
? s = 0; for(N = 1, 10^7, s += moebius(N)*eulerphi(N)); s
time = 28,398 ms. \\ slower, we must factor N. Twice.
%2 = 6393738650

The following loops over the fundamental dicriminants less than X:

? X = 10^8;
? forfactored(d=1,X, if (isfundamental(d),));
time = 34,030 ms.
? for(d=1,X, if (isfundamental(d),))
time = 1min, 24,225 ms.

74
3.1.14 forpart(X = k, seq, {a = k}, {n = k}). Evaluate seq over the partitions X = [x1 , . . . xn ] of
the integer k, i.e. increasing sequences x1 ≤ x2 . . . ≤ xn of sum x1 + . . . + xn = k. By convention, 0
admits only the empty partition and negative numbers have no partitions. A partition is given by a
t_VECSMALL, where parts are sorted in nondecreasing order. The partitions are listed by increasing
size and in lexicographic order when sizes are equal:

? forpart(X=4, print(X))
Vecsmall([4])
Vecsmall([1, 3])
Vecsmall([2, 2])
Vecsmall([1, 1, 2])
Vecsmall([1, 1, 1, 1])

Optional parameters n and a are as follows:

• n = nmax (resp. n = [nmin, nmax ]) restricts partitions to length less than nmax (resp.
length between nmin and nmax), where the length is the number of nonzero entries.

• a = amax (resp. a = [amin, amax ]) restricts the parts to integers less than amax (resp.
between amin and amax ).

By default, parts are positive and we remove zero entries unless amin ≤ 0, in which case we
fix the size #X = nmax :

\\ at most 3 nonzero parts, all <= 4

? forpart(v=5,print(Vec(v)), 4, 3)
[1, 4]
[2, 3]
[1, 1, 3]
[1, 2, 2]
\\ between 2 and 4 parts less than 5, fill with zeros
? forpart(v=5,print(Vec(v)),[0,5],[2,4])
[0, 0, 1, 4]
[0, 0, 2, 3]
[0, 1, 1, 3]
[0, 1, 2, 2]
[1, 1, 1, 2]
\\ no partitions of 1 with 2 to 4 nonzero parts
? forpart(v=1,print(v),[0,5],[2,4])
?

The behavior is unspecified if X is modified inside the loop.

The library syntax is forpart(void *data, long (*call)(void*,GEN), long k, GEN a,

GEN n).

75
3.1.15 forperm(a, p, seq). Evaluates seq, where the formal variable p goes through some per-
mutations given by a t_VECSMALL. If a is a positive integer then P goes through the permu-
tations of {1, 2, ..., a} in lexicographic order and if a is a small vector then p goes through the
(multi)permutations lexicographically larger than or equal to a.

? forperm(3, p, print(p))
Vecsmall([1, 2, 3])
Vecsmall([1, 3, 2])
Vecsmall([2, 1, 3])
Vecsmall([2, 3, 1])
Vecsmall([3, 1, 2])
Vecsmall([3, 2, 1])

When a is itself a t_VECSMALL or a t_VEC then p iterates through multipermutations

? forperm([2,1,1,3], p, print(p))
Vecsmall([2, 1, 1, 3])
Vecsmall([2, 1, 3, 1])
Vecsmall([2, 3, 1, 1])
Vecsmall([3, 1, 1, 2])
Vecsmall([3, 1, 2, 1])
Vecsmall([3, 2, 1, 1])

3.1.16 forprime(p = a, {b}, seq). Evaluates seq, where the formal variable p ranges over the prime
numbers between the real numbers a to b, including a and b if they are prime. More precisely, the
value of p is incremented to nextprime(p + 1), the smallest prime strictly larger than p, at the
end of each iteration. Nothing is done if a > b.

? forprime(p = 4, 10, print(p))

5
7

Setting b to +oo means we will run through all primes ≥ a, starting an infinite loop; it is expected
that the caller will break out of the loop itself at some point, using break or return.

Note that the value of p cannot be modified within seq:

? forprime(p = 2, 10, p = [])

*** at top-level: forprime(p=2,10,p=[])
*** ^---
*** prime index read-only: was changed to [].

76
3.1.17 forprimestep(p = a, b, q, seq). Evaluates seq, where the formal variable p ranges over the
prime numbers p in an arithmetic progression in [a, b]: q is either an integer (p ≡ a (mod q)) or
an intmod Mod(c,N) and we restrict to that congruence class. Nothing is done if a > b.
? forprimestep(p = 4, 30, 5, print(p))
19
29
? forprimestep(p = 4, 30, Mod(1,5), print(p))
11
Setting b to +oo means we will run through all primes ≥ a, starting an infinite loop; it is expected
that the caller will break out of the loop itself at some point, using break or return.
The current implementation restricts the modulus of the arithmetic progression to an unsigned
long (64 or 32 bits).
? forprimestep(p=2,oo,2^64,print(p))
*** at top-level: forprimestep(p=2,oo,2^64,print(p))
*** ^----------------------------------
*** forprimestep: overflow in t_INT-->ulong assignment.
Note that the value of p cannot be modified within seq:
? forprimestep(p = 2, 10, 3, p = [])
*** at top-level: forprimestep(p=2,10,3,p=[])
*** ^---
*** prime index read-only: was changed to [].

3.1.18 forsquarefree(N = a, b, seq). Evaluates seq, where the formal variable N is [n, factor(n)]
and n goes through squarefree integers from a to b; a and b must be integers. Nothing is done if
a > b.
? forsquarefree(N=-3,9,print(N))
[-3, [-1, 1; 3, 1]]
[-2, [-1, 1; 2, 1]]
[-1, Mat([-1, 1])]
[1, matrix(0,2)]
[2, Mat([2, 1])]
[3, Mat([3, 1])]
[5, Mat([5, 1])]
[6, [2, 1; 3, 1]]
[7, Mat([7, 1])]
This function is√only implemented for |a|, |b| < 264 (232 on a 32-bit machine). It uses a sieve
and runs in time O( b+b−a). It should be at least√ 5 times faster than regular factorization as long
as the interval length b − a is much larger than b and get√ relatively faster as the bounds increase.
The function slows down dramatically if primelimit < b. It is comparable to forfactored, but
about ζ(2) = π 2 /6 times faster due to the relative density of squarefree integers.
? B = 10^9;
? for (N = B, B+10^6, factor(N))
time = 4,392 ms.
? forfactored (N = B, B+10^6, [n,fan] = N)

77
 time = 915 ms.
? forsquarefree (N = B, B+10^6, [n,fan] = N)
time = 532 ms.
? B = 10^11;
? for (N = B, B+10^6, factor(N))
time = 13,053 ms.
? forfactored (N = B, B+10^6, [n,fan] = N)
time = 1,976 ms.
? forsquarefree (N = B, B+10^6, [n,fan] = N)
time = 1,245 ms.
? B = 10^14;
? for (N = B, B+10^6, factor(N))
time = 50,612 ms.
? forsquarefree (N = B, B+10^6, [n,fan] = N)
time = 46,309 ms.
√
The last timing is with the default primelimit (500000) which is much less than √ B + 106 ; it
goes down to 20,396ms if primelimit gets bigger than that bound. In any case B + 106 is much
larger than the interval length 106 so forsquarefree gets relatively slower for that reason as well.

Note that all PARI multiplicative functions accept the [n,fan] argument natively:

? s = 0; forsquarefree(N = 1, 10^7, s += moebius(N)*eulerphi(N)); s

time = 3,788 ms.
%1 = 6393738650
? s = 0; for(N = 1, 10^7, s += moebius(N)*eulerphi(N)); s
time = 28,630 ms. \\ slower, we must factor N. Twice.
%2 = 6393738650

The following loops over the fundamental dicriminants less than X:

? X = 10^8;
? for(d=1,X, if (isfundamental(d),))
time = 1min, 29,066 ms.
? forfactored(d=1,X, if (isfundamental(d),));
time = 42,387 ms.
? forsquarefree(d=1,X, D = quaddisc(d); if (D <= X, ));
time = 32,479 ms.

Note that in the last loop, the fundamental discriminants D are not evaluated in order (since
quaddisc(d) for squarefree d is either d or 4d). This is the price we pay for a faster evaluation,
and the set of numbers we run through is the same.

We can run through negative fundamental discriminants in the same way

? forsquarefree(d=-X,-1, D = quaddisc(d); if (D >= -X, ));

78
3.1.19 forstep(X = a, b, s, seq). Evaluates seq, where the formal variable X goes from a to b in
increments of s. Nothing is done if s > 0 and a > b or if s < 0 and a < b. s must be in R∗
or an intmod Mod(c,N) (restrict to the corresponding arithmetic progression) or a vector of steps
[s1 , . . . , sn ] (the successive steps in R∗ are used in the order they appear in s).
? forstep(x=5, 10, 2, print(x))
5
7
9
? forstep(x=5, 10, Mod(1,3), print(x))
7
10
? forstep(x=5, 10, [1,2], print(x))
5
6
8
9
Setting b to +oo will start an infinite loop; it is expected that the caller will break out of the loop
itself at some point, using break or return.

3.1.20 forsubgroup(H = G, {bound }, seq). Evaluates seq for each subgroup H of the abelian
group G (given in SNF form or as a vector of elementary divisors).
If bound is present, and is a positive integer, restrict the output to subgroups of index less than
bound . If bound is a vector containing a single positive integer B, then only subgroups of index
exactly equal to B are computed
The subgroups are not ordered in any obvious way, unless G is a p-group in which case
Birkhoff’s algorithm produces them by decreasing index. A subgroup is given as a matrix whose
columns give its generators on the implicit generators of G. For example, the following prints all
subgroups of index less than 2 in G = Z/2Zg1 × Z/2Zg2 :
? G = [2,2]; forsubgroup(H=G, 2, print(H))
[1; 1]
[1; 2]
[2; 1]
[1, 0; 1, 1]
The last one, for instance is generated by (g1 , g1 + g2 ). This routine is intended to treat huge
groups, when subgrouplist is not an option due to the sheer size of the output.
For maximal speed the subgroups have been left as produced by the algorithm. To print them
in canonical form (as left divisors of G in HNF form), one can for instance use
? G = matdiagonal([2,2]); forsubgroup(H=G, 2, print(mathnf(concat(G,H))))
[2, 1; 0, 1]
[1, 0; 0, 2]
[2, 0; 0, 1]
[1, 0; 0, 1]
Note that in this last representation, the index [G : H] is given by the determinant. See galois-
subcyclo and galoisfixedfield for applications to Galois theory.

79
 The library syntax is forsubgroup(void *data, long (*call)(void*,GEN), GEN G, GEN
bound).

3.1.21 forsubset(nk , s, seq). If nk is a nonnegative integer n, evaluates seq, where the formal
variable s goes through all subsets of {1, 2, . . . , n}; if nk is a pair [n, k] of integers, s goes through
subsets of size k of {1, 2, . . . , n}. In both cases s goes through subsets in lexicographic order among
subsets of the same size and smaller subsets come first.
? forsubset([5,3], s, print(s))
Vecsmall([1, 2, 3])
Vecsmall([1, 2, 4])
Vecsmall([1, 2, 5])
Vecsmall([1, 3, 4])
Vecsmall([1, 3, 5])
Vecsmall([1, 4, 5])
Vecsmall([2, 3, 4])
Vecsmall([2, 3, 5])
Vecsmall([2, 4, 5])
Vecsmall([3, 4, 5])
? forsubset(3, s, print(s))
Vecsmall([])
Vecsmall([1])
Vecsmall([2])
Vecsmall([3])
Vecsmall([1, 2])
Vecsmall([1, 3])
Vecsmall([2, 3])
Vecsmall([1, 2, 3])
The running time is proportional to the number of subsets enumerated, respectively 2n and
binomial(n, k):
? c = 0; forsubset([40,35],s,c++); c
time = 128 ms.
%4 = 658008
? binomial(40,35)
%5 = 658008

3.1.22 forvec(X = v, seq, {flag = 0}). Let v be an n-component vector (where n is arbitrary)
of two-component vectors [ai , bi ] for 1 ≤ i ≤ n, where all entries ai , bi are real numbers. This
routine lets X vary over the n-dimensional box given by v with unit steps: X is an n-dimensional
vector whose i-th entry X[i] runs through ai , ai + 1, ai + 2, . . . stopping with the first value greater
than bi (note that neither ai nor bi − ai are required to be integers). The values of X are ordered
lexicographically, like embedded for loops, and the expression seq is evaluated with the successive
values of X. The type of X is the same as the type of v: t_VEC or t_COL.
If flag = 1, generate only nondecreasing vectors X, and if flag = 2, generate only strictly
increasing vectors X.
? forvec (X=[[0,1],[-1,1]], print(X));
[0, -1]

80
 [0, 0]
[0, 1]
[1, -1]
[1, 0]
[1, 1]
? forvec (X=[[0,1],[-1,1]], print(X), 1);
[0, 0]
[0, 1]
[1, 1]
? forvec (X=[[0,1],[-1,1]], print(X), 2)
[0, 1]

3.1.23 if(a, {seq1 }, {seq2 }). Evaluates the expression sequence seq1 if a is nonzero, otherwise the
expression seq2 . Of course, seq1 or seq2 may be empty:
if (a,seq) evaluates seq if a is not equal to zero (you don’t have to write the second comma),
and does nothing otherwise,
if (a,,seq) evaluates seq if a is equal to zero, and does nothing otherwise. You could get the
same result using the ! (not) operator: if (!a,seq).
The value of an if statement is the value of the branch that gets evaluated: for instance
x = if(n % 4 == 1, y, z);
sets x to y if n is 1 modulo 4, and to z otherwise.
Successive ’else’ blocks can be abbreviated in a single compound if as follows:
if (test1, seq1,
test2, seq2,
...
testn, seqn,
seqdefault);
is equivalent to
if (test1, seq1
, if (test2, seq2
, ...
if (testn, seqn, seqdefault)...));
For instance, this allows to write traditional switch / case constructions:
if (x == 0, do0(),
x == 1, do1(),
x == 2, do2(),
dodefault());

Remark. The boolean operators && and || are evaluated according to operator precedence as
explained in Section 2.4, but, contrary to other operators, the evaluation of the arguments is
stopped as soon as the final truth value has been determined. For instance
if (x != 0 && f(1/x), ...)
is a perfectly safe statement.

81
Remark. Functions such as break and next operate on loops, such as forxxx, while, until.
The if statement is not a loop. (Obviously!)

3.1.24 iferr(seq1 , E, seq2 , {pred }). Evaluates the expression sequence seq1 . If an error occurs,
set the formal parameter E set to the error data. If pred is not present or evaluates to true,
catch the error and evaluate seq2 . Both pred and seq2 can reference E . The error type is given
by errname(E), and other data can be accessed using the component function. The code seq2
should check whether the error is the one expected. In the negative the error can be rethrown
using error(E) (and possibly caught by an higher iferr instance). The following uses iferr to
implement Lenstra’s ECM factoring method

? ecm(N, B = 1000!, nb = 100)=

{
for(a = 1, nb,
iferr(ellmul(ellinit([a,1]*Mod(1,N)), [0,1]*Mod(1,N), B),
E, return(gcd(lift(component(E,2)),N)),
errname(E)=="e_INV" && type(component(E,2)) == "t_INTMOD"))
}
? ecm(2^101-1)
%2 = 7432339208719

The return value of iferr itself is the value of seq2 if an error occurs, and the value of seq1
otherwise. We now describe the list of valid error types, and the attached error data E ; in each
case, we list in order the components of E , accessed via component(E,1), component(E,2), etc.

Internal errors, “system” errors.

• "e ARCH". A requested feature s is not available on this architecture or operating system. E
has one component (t_STR): the missing feature name s.

• "e BUG". A bug in the PARI library, in function s. E has one component (t_STR): the
function name s.

• "e FILE". Error while trying to open a file. E has two components, 1 (t_STR): the file type
(input, output, etc.), 2 (t_STR): the file name.

• "e IMPL". A requested feature s is not implemented. E has one component, 1 (t_STR): the
feature name s.

• "e PACKAGE". Missing optional package s. E has one component, 1 (t_STR): the package
name s.

82
Syntax errors, type errors.
• "e DIM". The dimensions of arguments x and y submitted to function s does not match up.
E.g., multiplying matrices of inconsistent dimension, adding vectors of different lengths,. . . E has
three component, 1 (t_STR): the function name s, 2: the argument x, 3: the argument y.
• "e FLAG". A flag argument is out of bounds in function s. E has one component, 1 (t_STR):
the function name s.
• "e NOTFUNC". Generated by the PARI evaluator; tried to use a GEN x which is not a
t_CLOSURE in a function call syntax (as in f = 1; f(2);). E has one component, 1: the offending
GEN x.
• "e OP". Impossible operation between two objects than cannot be typecast to a sensible
common domain for deeper reasons than a type mismatch, usually for arithmetic reasons. As in
O(2) + O(3): it is valid to add two t_PADICs, provided the underlying prime is the same; so the
addition is not forbidden a priori for type reasons, it only becomes so when inspecting the objects
and trying to perform the operation. E has three components, 1 (t_STR): the operator name op,
2: first argument, 3: second argument.
• "e TYPE". An argument x of function s had an unexpected type. (As in factor("blah").)
E has two components, 1 (t_STR): the function name s, 2: the offending argument x.
• "e TYPE2". Forbidden operation between two objects than cannot be typecast to a sensible
common domain, because their types do not match up. (As in Mod(1,2) + Pi.) E has three
components, 1 (t_STR): the operator name op, 2: first argument, 3: second argument.
• "e PRIORITY". Object o in function s contains variables whose priority is incompatible
with the expected operation. E.g. Pol([x,1], ’y): this raises an error because it’s not possible to
create a polynomial whose coefficients involve variables with higher priority than the main variable.
E has four components: 1 (t_STR): the function name s, 2: the offending argument o, 3 (t_STR):
an operator op describing the priority error, 4 (t_POL): the variable v describing the priority error.
The argument satisfies variable(x) opvariable(v).
• "e VAR". The variables of arguments x and y submitted to function s does not match up.
E.g., considering the algebraic number Mod(t,t^2+1) in nfinit(x^2+1). E has three component,
1 (t_STR): the function name s, 2 (t_POL): the argument x, 3 (t_POL): the argument y.

Overflows.
• "e COMPONENT". Trying to access an inexistent component in a vector/matrix/list in a
function: the index is less than 1 or greater than the allowed length. E has four components, 1
(t_STR): the function name 2 (t_STR): an operator op (< or >), 2 (t_GEN): a numerical limit l
bounding the allowed range, 3 (GEN): the index x. It satisfies x op l.
• "e DOMAIN". An argument is not in the function’s domain. E has five components, 1 (t_STR):
the function name, 2 (t_STR): the mathematical name of the out-of-domain argument 3 (t_STR):
an operator op describing the domain error, 4 (t_GEN): the numerical limit l describing the domain
error, 5 (GEN): the out-of-domain argument x. The argument satisfies x op l, which prevents it
from belonging to the function’s domain.
• "e MAXPRIME". A function using the precomputed list of prime numbers ran out of primes.
E has one component, 1 (t_INT): the requested prime bound, which overflowed primelimit or 0
(bound is unknown).

83
 • "e MEM". A call to pari_malloc or pari_realloc failed. E has no component.
• "e OVERFLOW". An object in function s becomes too large to be represented within PARI’s
hardcoded limits. (As in 2^2^2^10 or exp(1e100), which overflow in lg and expo.) E has one
component, 1 (t_STR): the function name s.
• "e PREC". Function s fails because input accuracy is too low. (As in floor(1e100) at
default accuracy.) E has one component, 1 (t_STR): the function name s.
• "e STACK". The PARI stack overflows. E has no component.

Errors triggered intentionally.

• "e ALARM". A timeout, generated by the alarm function. E has one component (t_STR): the
error message to print.
• "e USER". A user error, as triggered by error(g1 , . . . , gn ). E has one component, 1 (t_VEC):
the vector of n arguments given to error.

Mathematical errors.
• "e CONSTPOL". An argument of function s is a constant polynomial, which does not make
sense. (As in galoisinit(Pol(1)).) E has one component, 1 (t_STR): the function name s.
• "e COPRIME". Function s expected coprime arguments, and did receive x, y, which were not.
E has three component, 1 (t_STR): the function name s, 2: the argument x, 3: the argument y.
• "e INV". Tried to invert a noninvertible object x in function s. E has two components, 1
(t_STR): the function name s, 2: the noninvertible x. If x = Mod(a, b) is a t_INTMOD and a is not 0
mod b, this allows to factor the modulus, as gcd(a, b) is a nontrivial divisor of b.
• "e IRREDPOL". Function s expected an irreducible polynomial, and did receive T , which was
not. (As in nfinit(x^2-1).) E has two component, 1 (t_STR): the function name s, 2 (t_POL):
the polynomial x.
• "e MISC". Generic uncategorized error. E has one component (t_STR): the error message to
print.
• "e MODULUS". moduli x and y submitted to function s are inconsistent. As in
nfalgtobasis(nfinit(t^3-2), Mod(t,t^2+1)

E has three component, 1 (t_STR): the function s, 2: the argument x, 3: the argument x.
• "e PRIME". Function s expected a prime number, and did receive p, which was not. (As in
idealprimedec(nf, 4).) E has two component, 1 (t_STR): the function name s, 2: the argument
p.
• "e ROOTS0". An argument of function s is a zero polynomial, and we need to consider its
roots. (As in polroots(0).) E has one component, 1 (t_STR): the function name s.
• "e SQRTN". Trying to compute an n-th root of x, which does not exist, in function s. (As in
sqrt(Mod(-1,3)).) E has two components, 1 (t_STR): the function name s, 2: the argument x.

84
3.1.25 next({n = 1}). Interrupts execution of current seq, resume the next iteration of the
innermost enclosing loop, within the current function call (or top level loop). If n is specified,
resume at the n-th enclosing loop. If n is bigger than the number of enclosing loops, all enclosing
loops are exited.

3.1.26 return({x = 0}). Returns from current subroutine, with result x. If x is omitted, return
the (void) value (return no result, like print).

3.1.27 until(a, seq). Evaluates seq until a is not equal to 0 (i.e. until a is true). If a is initially
not equal to 0, seq is evaluated once (more generally, the condition on a is tested after execution
of the seq, not before as in while).

3.1.28 while(a, seq). While a is nonzero, evaluates the expression sequence seq. The test is made
before evaluating the seq, hence in particular if a is initially equal to zero the seq will not be
evaluated at all.

3.2 Programming in GP: other specific functions.

In addition to the general PARI functions, it is necessary to have some functions which will
be of use specifically for gp, though a few of these can be accessed under library mode. Before we
start describing these, we recall the difference between strings and keywords (see Section 2.9): the
latter don’t get expanded at all, and you can type them without any enclosing quotes. The former
are dynamic objects, where everything outside quotes gets immediately expanded.

3.2.1 Strchr(x). Deprecated alias for strchr.

The library syntax is GEN pari_strchr(GEN x).

3.2.2 Strexpand({x}∗). Deprecated alias for strexpand

The library syntax is GEN strexpand(const char *x*).

3.2.3 Strprintf(fmt, {x}∗). Deprecated alias for strprintf.

The library syntax is GEN Strprintf(const char *fmt, const char *x*).

3.2.4 Strtex({x}∗). Deprecated alias for strtex.

The library syntax is GEN strtex(const char *x*).

85
3.2.5 addhelp(sym, str ). Changes the help message for the symbol sym. The string str is expanded
on the spot and stored as the online help for sym. It is recommended to document global variables
and user functions in this way, although gp will not protest if you don’t.
You can attach a help text to an alias, but it will never be shown: aliases are expanded by
the ? help operator and we get the help of the symbol the alias points to. Nothing prevents you
from modifying the help of built-in PARI functions. But if you do, we would like to hear why you
needed it!
Without addhelp, the standard help for user functions consists of its name and definition.
gp> f(x) = x^2;
gp> ?f
f =
(x)->x^2

Once addhelp is applied to f , the function code is no longer included. It can still be consulted by
typing the function name:
gp> addhelp(f, "Square")
gp> ?f
Square
gp> f
%2 = (x)->x^2
The library syntax is void addhelp(const char *sym, const char *str).

3.2.6 alarm({s = 0}, {code}). If code is omitted, trigger an e ALARM exception after s seconds
(wall-clock time), cancelling any previously set alarm; stop a pending alarm if s = 0 or is omitted.
Otherwise, if s is positive, the function evaluates code, aborting after s seconds. The return
value is the value of code if it ran to completion before the alarm timeout, and a t_ERROR object
otherwise.
? p = nextprime(10^25); q = nextprime(10^26); N = p*q;
? E = alarm(1, factor(N));
? type(E)
%3 = "t_ERROR"
? print(E)
%4 = error("alarm interrupt after 964 ms.")
? alarm(10, factor(N)); \\ enough time
%5 =
[ 10000000000000000000000013 1]
[100000000000000000000000067 1]
Here is a more involved example: the function timefact(N,sec) below tries to factor N and gives
up after sec seconds, returning a partial factorization.
\\ Time-bounded partial factorization
default(factor_add_primes,1);
timefact(N,sec)=
{
F = alarm(sec, factor(N));

86
 if (type(F) == "t_ERROR", factor(N, 2^24), F);
}
We either return the factorization directly, or replace the t_ERROR result by a simple bounded
factorization factor(N, 2^24). Note the factor_add_primes trick: any prime larger than 224
discovered while attempting the initial factorization is stored and remembered. When the alarm
rings, the subsequent bounded factorization finds it right away.

Caveat. It is not possible to set a new alarm within another alarm code: the new timer erases the
parent one.
The library syntax is GEN gp_alarm(long s, GEN code = NULL).

3.2.7 alias(newsym, sym). Defines the symbol newsym as an alias for the symbol sym:
? alias("det", "matdet");
? det([1,2;3,4])
%1 = -2
You are not restricted to ordinary functions, as in the above example: to alias (from/to) member
functions, prefix them with ‘ .’; to alias operators, use their internal name, obtained by writing
in lieu of the operators argument: for instance, ! and ! are the internal names of the factorial
and the logical negation, respectively.
? alias("mod", "_.mod");
? alias("add", "_+_");
? alias("_.sin", "sin");
? mod(Mod(x,x^4+1))
%2 = x^4 + 1
? add(4,6)
%3 = 10
? Pi.sin
%4 = 0.E-37
Alias expansion is performed directly by the internal GP compiler. Note that since alias is
performed at compilation-time, it does not require any run-time processing, however it only affects
GP code compiled after the alias command is evaluated. A slower but more flexible alternative is
to use variables. Compare
? fun = sin;
? g(a,b) = intnum(t=a,b,fun(t));
? g(0, Pi)
%3 = 2.0000000000000000000000000000000000000
? fun = cos;
? g(0, Pi)
%5 = 1.8830410776607851098 E-39
with
? alias(fun, sin);
? g(a,b) = intnum(t=a,b,fun(t));
? g(0,Pi)
%2 = 2.0000000000000000000000000000000000000
? alias(fun, cos); \\ Oops. Does not affect *previous* definition!

87
 ? g(0,Pi)
%3 = 2.0000000000000000000000000000000000000
? g(a,b) = intnum(t=a,b,fun(t)); \\ Redefine, taking new alias into account
? g(0,Pi)
%5 = 1.8830410776607851098 E-39
A sample alias file misc/gpalias is provided with the standard distribution.
The library syntax is void alias0(const char *newsym, const char *sym).

3.2.8 allocatemem({s = 0}). This special operation changes the stack size after initialization.
The argument s must be a nonnegative integer. If s > 0, a new stack of at least s bytes is allocated.
We may allocate more than s bytes if s is way too small, or for alignment reasons: the current
formula is max(16 ∗ ds/16e , 500032) bytes.
If s = 0, the size of the new stack is twice the size of the old one.
This command is much more useful if parisizemax is nonzero, and we describe this case first.
With parisizemax enabled, there are three sizes of interest:
• a virtual stack size, parisizemax, which is an absolute upper limit for the stack size; this is
set by default(parisizemax, ...).
• the desired typical stack size, parisize, that will grow as needed, up to parisizemax; this
is set by default(parisize, ...).
• the current stack size, which is less that parisizemax, typically equal to parisize but
possibly larger and increasing dynamically as needed; allocatemem allows to change that one
explicitly.
The allocatemem command forces stack usage to increase temporarily (up to parisizemax of
course); for instance if you notice using \gm2 that we seem to collect garbage a lot, e.g.
? \gm2
debugmem = 2
? default(parisize,"32M")
*** Warning: new stack size = 32000000 (30.518 Mbytes).
? bnfinit(’x^2+10^30-1)
*** bnfinit: collecting garbage in hnffinal, i = 1.
*** bnfinit: collecting garbage in hnffinal, i = 2.
*** bnfinit: collecting garbage in hnffinal, i = 3.
and so on for hundred of lines. Then, provided the breakloop default is set, you can interrupt the
computation, type allocatemem(100*10^6) at the break loop prompt, then let the computation
go on by typing <Enter>. Back at the gp prompt, the desired stack size of parisize is restored.
Note that changing either parisize or parisizemax at the break loop prompt would interrupt the
computation, contrary to the above.
In most cases, parisize will increase automatically (up to parisizemax) and there is no need
to perform the above maneuvers. But that the garbage collector is sufficiently efficient that a given
computation can still run without increasing the stack size, albeit very slowly due to the frequent
garbage collections.

88
Deprecated: when parisizemax is unset. This is currently still the default behavior in order not
to break backward compatibility. The rest of this section documents the behavior of allocatemem
in that (deprecated) situation: it becomes a synonym for default(parisize,...). In that case,
there is no notion of a virtual stack, and the stack size is always equal to parisize. If more memory
is needed, the PARI stack overflows, aborting the computation.
Thus, increasing parisize via allocatemem or default(parisize,...) before a big compu-
tation is important. Unfortunately, either must be typed at the gp prompt in interactive usage, or
left by itself at the start of batch files. They cannot be used meaningfully in loop-like constructs,
or as part of a larger expression sequence, e.g
allocatemem(); x = 1; \\ This will not set x!
In fact, all loops are immediately exited, user functions terminated, and the rest of the sequence
following allocatemem() is silently discarded, as well as all pending sequences of instructions. We
just go on reading the next instruction sequence from the file we are in (or from the user). In
particular, we have the following possibly unexpected behavior: in
read("file.gp"); x = 1
were file.gp contains an allocatemem statement, the x = 1 is never executed, since all pending
instructions in the current sequence are discarded.
The reason for these unfortunate side-effects is that, with parisizemax disabled, increasing the
stack size physically moves the stack, so temporary objects created during the current expression
evaluation are not correct anymore. (In particular byte-compiled expressions, which are allocated
on the stack.) To avoid accessing obsolete pointers to the old stack, this routine ends by a longjmp.
The library syntax is void gp_allocatemem(GEN s = NULL).

3.2.9 apply(f, A). Apply the t_CLOSURE f to the entries of A.

• If A is a scalar, return f(A).
ai xi (+O(xN )), apply f on all coefficients and return
P
P • If Ai is a polynomial
N
or power series
f (ai )x (+O(x )).
• If A is a vector or list [a1 , . . . , an ], return the vector or list [f (a1 ), . . . , f (an )]. If A is a matrix,
return the matrix whose entries are the f (A[i, j]).
? apply(x->x^2, [1,2,3,4])
%1 = [1, 4, 9, 16]
? apply(x->x^2, [1,2;3,4])
%2 =
[1 4]
[9 16]
? apply(x->x^2, 4*x^2 + 3*x+ 2)
%3 = 16*x^2 + 9*x + 4
? apply(sign, 2 - 3* x + 4*x^2 + O(x^3))
%4 = 1 - x + x^2 + O(x^3)
Note that many functions already act componentwise on vectors or matrices, but they almost never
act on lists; in this case, apply is a good solution:
? L = List([Mod(1,3), Mod(2,4)]);

89
 ? lift(L)
*** at top-level: lift(L)
*** ^-------
*** lift: incorrect type in lift.
? apply(lift, L);
%2 = List([1, 2])

Remark. For v a t_VEC, t_COL, t_VECSMALL, t_LIST or t_MAT, the alternative set-notations
[g(x) | x <- v, f(x)]
[x | x <- v, f(x)]
[g(x) | x <- v]
are available as shortcuts for
apply(g, select(f, Vec(v)))
select(f, Vec(v))
apply(g, Vec(v))
respectively:
? L = List([Mod(1,3), Mod(2,4)]);
? [ lift(x) | x<-L ]
%2 = [1, 2]
The library syntax is genapply(void *E, GEN (*fun)(void*,GEN), GEN a).

3.2.10 arity(C). Return the arity of the closure C, i.e., the number of its arguments.
? f1(x,y=0)=x+y;
? arity(f1)
%1 = 2
? f2(t,s[..])=print(t,":",s);
? arity(f2)
%2 = 2
Note that a variadic argument, such as s in f2 above, is counted as a single argument.
The library syntax is GEN arity0(GEN C).

3.2.11 call(f, A). A = [a1 , . . . , an ] being a vector and f being a function, returns the evaluation
of f (a1 , . . . , an ). f can also be the name of a built-in GP function. If #A = 1, call(f, A)
= apply(f, A)[1]. If f is variadic, the variadic arguments must grouped in a vector in the last
component of A.
This function is useful
• when writing a variadic function, to call another one:
fprintf(file,format,args[..]) = write(file,call(strprintf,[format,args]))
• when dealing with function arguments with unspecified arity
The function below implements a global memoization interface:
memo=Map();
memoize(f,A[..])=

90
 {
my(res);
if(!mapisdefined(memo, [f,A], &res),
res = call(f,A);
mapput(memo,[f,A],res));
res;
}
for example:
? memoize(factor,2^128+1)
%3 = [59649589127497217,1;5704689200685129054721,1]
? ##
*** last result computed in 76 ms.
? memoize(factor,2^128+1)
%4 = [59649589127497217,1;5704689200685129054721,1]
? ##
*** last result computed in 0 ms.
? memoize(ffinit,3,3)
%5 = Mod(1,3)*x^3+Mod(1,3)*x^2+Mod(1,3)*x+Mod(2,3)
? fibo(n)=if(n==0,0,n==1,1,memoize(fibo,n-2)+memoize(fibo,n-1));
? fibo(100)
%7 = 354224848179261915075
• to call operators through their internal names without using alias
matnbelts(M) = call("_*_",matsize(M))
The library syntax is GEN call0(GEN f, GEN A).

3.2.12 default({key}, {val }). Returns the default corresponding to keyword key. If val is present,
sets the default to val first (which is subject to string expansion first). Typing default() (or \d)
yields the complete default list as well as their current values. See Section 2.12 for an introduction
to GP defaults, Section 3.4 for a list of available defaults, and Section 2.13 for some shortcut alter-
natives. Note that the shortcuts are meant for interactive use and usually display more information
than default.
The library syntax is GEN default0(const char *key = NULL, const char *val = NULL)
.

3.2.13 errname(E). Returns the type of the error message E as a string.

? iferr(1 / 0, E, print(errname(E)))
e_INV
? ?? e_INV
[...]
* "e_INV". Tried to invert a noninvertible object x in function s.
[...]
The library syntax is GEN errname(GEN E).

91
3.2.14 error({str }∗). Outputs its argument list (each of them interpreted as a string), then
interrupts the running gp program, returning to the input prompt. For instance

error("n = ", n, " is not squarefree!")

3.2.15 export(x{= ...}, ..., z{= ...}). Export the variables x, . . . , z to the parallel world. Such
variables are visible inside parallel sections in place of global variables, but cannot be modified
inside a parallel section. export(a) set the variable a in the parallel world to current value of a.
export(a=z) set the variable a in the parallel world to z, without affecting the current value of a.

? fun(x)=x^2+1;
? parvector(10,i,fun(i))
*** mt: please use export(fun).
? export(fun)
? parvector(10,i,fun(i))
%4 = [2,5,10,17,26,37,50,65,82,101]

3.2.16 exportall(). Declare all current dynamic variables as exported variables. Such variables
are visible inside parallel sections in place of global variables.

? fun(x)=x^2+1;
? parvector(10,i,fun(i))
*** mt: please use export(fun).
? exportall()
? parvector(10,i,fun(i))
%4 = [2,5,10,17,26,37,50,65,82,101]

The library syntax is void exportall().

3.2.17 extern(str ). The string str is the name of an external command (i.e. one you would type
from your UNIX shell prompt). This command is immediately run and its output fed into gp, just
as if read from a file.

The library syntax is GEN gpextern(const char *str).

3.2.18 externstr(str ). The string str is the name of an external command (i.e. one you would
type from your UNIX shell prompt). This command is immediately run and its output is returned
as a vector of GP strings, one component per output line.

The library syntax is GEN externstr(const char *str).

92
3.2.19 fileclose(n). Close the file descriptor n, created via fileopen or fileextern. Finitely
many files can be opened at a given time, closing them recycles file descriptors and avoids running
out of them:

? n = 0; while(n++, fileopen("/tmp/test", "w"))

*** at top-level: n=0;while(n++,fileopen("/tmp/test","w"))
*** ^--------------------------
*** fileopen: error opening requested file: ‘/tmp/test’.
*** Break loop: type ’break’ to go back to GP prompt
break> n
65533

This is a limitation of the operating system and does not depend on PARI: if you open too many
files in gp without closing them, the operating system will also prevent unrelated applications from
opening files. Independently, your operating system (e.g. Windows) may prevent other applications
from accessing or deleting your file while it is opened by gp. Quitting gp implicitly calls this function
on all opened file descriptors.

On files opened for writing, this function also forces a write of all buffered data to the file
system and completes all pending write operations. This function is implicitly called for all open
file descriptors when exiting gp but it is cleaner and safer to call it explicitly, for instance in case
of a gp crash or general system failure, which could cause data loss.

? n = fileopen("./here");
? while(l = fileread(n), print(l));
? fileclose(n);
? n = fileopen("./there", "w");
? for (i = 1, 100, filewrite(n, i^2+1))
? fileclose(n)

Until a fileclose, there is no guarantee that the file on disk contains all the expected data
from previous filewrites. (And even then the operating system may delay the actual write to
hardware.)

Closing a file twice raises an exception:

? n = fileopen("/tmp/test");
? fileclose(n)
? fileclose(n)
*** at top-level: fileclose(n)
*** ^------------
*** fileclose: invalid file descriptor 0

The library syntax is void gp_fileclose(long n).

93
3.2.20 fileextern(str ). The string str is the name of an external command, i.e. one you would
type from your UNIX shell prompt. This command is immediately run and the function returns a
file descriptor attached to the command output as if it were read from a file.
? n = fileextern("ls -l");
? while(l = filereadstr(n), print(l))
? fileclose(n)
If the secure default is set, this function will raise en exception.
The library syntax is long gp_fileextern(const char *str).

3.2.21 fileflush({n}). Flushes the file descriptor n, created via fileopen or fileextern. On files
opened for writing, this function forces a write of all buffered data to the file system and completes
all pending write operations. This function is implicitly called by fileclose but you may want to
call it explicitly at synchronization points, for instance after writing a large result to file and before
printing diagnostics on screen. (In order to be sure that the file contains the expected content on
inspection.)
If n is omitted, flush all descriptors to output streams.
? n = fileopen("./here", "w");
? for (i = 1, 10^5, \
filewrite(n, i^2+1); \
if (i % 10000 == 0, fileflush(n)))
Until a fileflush or fileclose, there is no guarantee that the file contains all the expected
data from previous filewrites.
The library syntax is void gp_fileflush0(GEN n = NULL). But the direct and more specific
variant void gp_fileflush(long n) is also available.

3.2.22 fileopen(path, mode). Open the file pointed to by ’path’ and return a file descriptor which
can be used with other file functions.
The mode can be
• "r" (default): open for reading; allow fileread and filereadstr.
• "w": open for writing, discarding existing content; allow filewrite, filewrite1.
• "a": open for writing, appending to existing content; same operations allowed as "w".
Eventually, the file should be closed and the descriptor recycled using fileclose.
? n = fileopen("./here"); \\ "r" by default
? while (l = filereadstr(n), print(l)) \\ print successive lines
? fileclose(n) \\ done
In read mode, raise an exception if the file does not exist or the user does not have read permission.
In write mode, raise an exception if the file cannot be written to. Trying to read or write to a file
that was not opend with the right mode raises an exception.
? n = fileopen("./read", "r");
? filewrite(n, "test") \\ not open for writing
*** at top-level: filewrite(n,"test")
*** ^-------------------

94
 *** filewrite: invalid file descriptor 0
The library syntax is long gp_fileopen(const char *path, const char *mode).

3.2.23 fileread(n). Read a logical line from the file attached to the descriptor n, opened for
reading with fileopen. Return 0 at end of file.
A logical line is a full command as it is prepared by gp’s preprocessor (skipping blanks and
comments or assembling multiline commands between braces) before being fed to the interpreter.
The function filereadstr would read a raw line exactly as input, up to the next carriage return
\n.
Compare raw lines

? n = fileopen("examples/bench.gp");
? while(l = filereadstr(n), print(l));
{
u=v=p=q=1;
for (k=1, 2000,
[u,v] = [v,u+v];
p *= v; q = lcm(q,v);
if (k%50 == 0,
print(k, " ", log(p)/log(q))
)
)
}
and logical lines

? n = fileopen("examples/bench.gp");
? while(l = fileread(n), print(l));
u=v=p=q=1;for(k=1,2000,[u,v]=[v,u+v];p*=v;q=lcm(q,v);[...]
The library syntax is GEN gp_fileread(long n).

3.2.24 filereadstr(n). Read a raw line from the file attached to the descriptor n, opened for
reading with fileopen, discarding the terminating newline. In other words the line is read exactly
as input, up to the next carriage return \n. By comparison, fileread would read a logical line, as
assembled by gp’s preprocessor (skipping blanks and comments for instance).
The library syntax is GEN gp_filereadstr(long n).

3.2.25 filewrite(n, s). Write the string s to the file attached to descriptor n, ending with a newline.
The file must have been opened with fileopen in "w" or "a" mode. There is no guarantee that s
is completely written to disk until fileclose(n) is executed, which is automatic when quitting gp.
If the newline is not desired, use filewrite1.

95
Variant. The high-level function write is expensive when many consecutive writes are expected
because it cannot use buffering. The low-level interface fileopen / filewrite / fileclose is
more efficient:
? f = "/tmp/bigfile";
? for (i = 1, 10^5, write(f, i^2+1))
time = 240 ms.
? v = vector(10^5, i, i^2+1);
time = 10 ms. \\ computing the values is fast
? write("/tmp/bigfile2",v)
time = 12 ms. \\ writing them in one operation is fast
? n = fileopen("/tmp/bigfile", "w");
? for (i = 1, 10^5, filewrite(n, i^2+1))
time = 24 ms. \\ low-level write is ten times faster
? fileclose(n);
In the final example, the file needs not be in a consistent state until the ending fileclose is
evaluated, e.g. some lines might be half-written or not present at all even though the corresponding
filewrite was executed already. Both a single high-level write and a succession of low-level
filewrites achieve the same efficiency, but the latter is often more natural. In fact, concatenating
naively the entries to be written is quadratic in the number of entries, hence much more expensive
than the original write operations:
? v = []; for (i = 1, 10^5, v = concat(v,i))
time = 1min, 41,456 ms.
The library syntax is void gp_filewrite(long n, const char *s).

3.2.26 filewrite1(n, s). Write the string s to the file attached to descriptor n. The file must have
been opened with fileopen in "w" or "a" mode.
If you want to append a newline at the end of s, you can use Str(s,"\n") or filewrite.
The library syntax is void gp_filewrite1(long n, const char *s).

3.2.27 fold(f, A). Apply the t_CLOSURE f of arity 2 to the entries of A, in order to return
f(. . . f(f(A[1],A[2]),A[3]). . . ,A[#A]).
? fold((x,y)->x*y, [1,2,3,4])
%1 = 24
? fold((x,y)->[x,y], [1,2,3,4])
%2 = [[[1, 2], 3], 4]
? fold((x,f)->f(x), [2,sqr,sqr,sqr])
%3 = 256
? fold((x,y)->(x+y)/(1-x*y),[1..5])
%4 = -9/19
? bestappr(tan(sum(i=1,5,atan(i))))
%5 = -9/19
The library syntax is GEN fold0(GEN f, GEN A). Also available is GEN genfold(void *E,
GEN (*fun)(void*, GEN, GEN), GEN A).

96
3.2.28 getabstime(). Returns the CPU time (in milliseconds) elapsed since gp startup. This
provides a reentrant version of gettime:
my (t = getabstime());
...
print("Time: ", strtime(getabstime() - t));
For a version giving wall-clock time, see getwalltime.
The library syntax is long getabstime().

3.2.29 getenv(s). Return the value of the environment variable s if it is defined, otherwise return
0.
The library syntax is GEN gp_getenv(const char *s).

3.2.30 getheap(). Returns a two-component row vector giving the number of objects on the heap
and the amount of memory they occupy in long words. Useful mainly for debugging purposes.
The library syntax is GEN getheap().

3.2.31 getlocalbitprec(). Returns the current dynamic bit precision.

3.2.32 getlocalprec(). Returns the current dynamic precision, in decimal digits.

3.2.33 getrand(). Returns the current value of the seed used by the pseudo-random number gener-
ator random. Useful mainly for debugging purposes, to reproduce a specific chain of computations.
The returned value is technical (reproduces an internal state array), and can only be used as an
argument to setrand.
The library syntax is GEN getrand().

3.2.34 getstack(). Returns the current value of top − avma, i.e. the number of bytes used up to
now on the stack. Useful mainly for debugging purposes.
The library syntax is long getstack().

3.2.35 gettime(). Returns the CPU time (in milliseconds) used since either the last call to
gettime, or to the beginning of the containing GP instruction (if inside gp), whichever came last.
For a reentrant version, see getabstime.
For a version giving wall-clock time, see getwalltime.
The library syntax is long gettime().

3.2.36 getwalltime(). Returns the time (in milliseconds) elapsed since 00:00:00 UTC Thursday
1, January 1970 (the Unix epoch).
my (t = getwalltime());
...
print("Time: ", strtime(getwalltime() - t));
The library syntax is GEN getwalltime().

97
3.2.37 global(listof variables). Obsolete. Scheduled for deletion.

3.2.38 inline(x, ..., z). Declare x, . . . , z as inline variables. Such variables behave like lexically
scoped variable (see my()) but with unlimited scope. It is however possible to exit the scope by
using uninline(). When used in a GP script, it is recommended to call uninline() before the
script’s end to avoid inline variables leaking outside the script. DEPRECATED, use export.

3.2.39 input(). Reads a string, interpreted as a GP expression, from the input file, usually
standard input (i.e. the keyboard). If a sequence of expressions is given, the result is the result
of the last expression of the sequence. When using this instruction, it is useful to prompt for the
string by using the print1 function. Note that in the present version 2.19 of pari.el, when using
gp under GNU Emacs (see Section 2.16) one must prompt for the string, with a string which ends
with the same prompt as any of the previous ones (a "? " will do for instance).

The library syntax is GEN gp_input().

3.2.40 install(name, code, {gpname}, {lib}). Loads from dynamic library lib the function name.
Assigns to it the name gpname in this gp session, with prototype code (see below). If gpname is
omitted, uses name. If lib is omitted, all symbols known to gp are available: this includes the whole
of libpari.so and possibly others (such as libc.so).

Most importantly, install gives you access to all nonstatic functions defined in the PARI
library. For instance, the function

GEN addii(GEN x, GEN y)

adds two PARI integers, and is not directly accessible under gp (it is eventually called by the +
operator of course):

? install("addii", "GG")
? addii(1, 2)
%1 = 3

It also allows to add external functions to the gp interpreter. For instance, it makes the function
system obsolete:

? install(system, vs, sys,/*omitted*/)

? sys("ls gp*")
gp.c gp.h gp_rl.c

This works because system is part of libc.so, which is linked to gp. It is also possible to compile
a shared library yourself and provide it to gp in this way: use gp2c, or do it manually (see the
modules build variable in pari.cfg for hints).

Re-installing a function will print a warning and update the prototype code if needed. However,
it will not reload a symbol from the library, even if the latter has been recompiled.

98
Prototype. We only give a simplified description here, covering most functions, but there are
many more possibilities. The full documentation is available in libpari.dvi, see
??prototype
• First character i, l, u, v : return type int / long / ulong / void. (Default: GEN)
• One letter for each mandatory argument, in the same order as they appear in the argument
list: G (GEN), & (GEN*), L (long), U (ulong), s (char *), n (variable).
• p to supply realprecision (usually long prec in the argument list), b to supply realbit-
precision (usually long bitprec), P to supply seriesprecision (usually long precdl).
We also have special constructs for optional arguments and default values:
• DG (optional GEN, NULL if omitted),
• D& (optional GEN*, NULL if omitted),
• Dn (optional variable, −1 if omitted),
For instance the prototype corresponding to
long issquareall(GEN x, GEN *n = NULL)
is lGD&.

Caution. This function may not work on all systems, especially when gp has been compiled
statically. In that case, the first use of an installed function will provoke a Segmentation Fault (this
should never happen with a dynamically linked executable). If you intend to use this function,
please check first on some harmless example such as the one above that it works properly on your
machine.
The library syntax is void gpinstall(const char *name, const char *code, const char
*gpname, const char *lib).

3.2.41 kill(sym). Restores the symbol sym to its “undefined” status, and deletes any help messages
attached to sym using addhelp. Variable names remain known to the interpreter and keep their
former priority: you cannot make a variable “less important” by killing it!
? z = y = 1; y
%1 = 1
? kill(y)
? y \\ restored to ‘‘undefined’’ status
%2 = y
? variable()
%3 = [x, y, z] \\ but the variable name y is still known, with y > z !
For the same reason, killing a user function (which is an ordinary variable holding a t_CLOSURE)
does not remove its name from the list of variable names.
If the symbol is attached to a variable — user functions being an important special case —,
one may use the quote operator a = ’a to reset variables to their starting values. However, this
will not delete a help message attached to a, and is also slightly slower than kill(a).
? x = 1; addhelp(x, "foo"); x
%1 = 1

99
 ? x = ’x; x \\ same as ’kill’, except we don’t delete help.
%2 = x
? ?x
foo
On the other hand, kill is the only way to remove aliases and installed functions.
? alias(fun, sin);
? kill(fun);
? install(addii, GG);
? kill(addii);
The library syntax is void kill0(const char *sym).

3.2.42 listcreate({n}). This function is obsolete, use List.

Creates an empty list. This routine used to have a mandatory argument, which is now ignored
(for backward compatibility).

3.2.43 listinsert( L, x, n). Inserts the object x at position n in L (which must be of type t_LIST).
This has complexity O(#L−n+1): all the remaining elements of list (from position n+1 onwards)
are shifted to the right. If n is greater than the list length, appends x.
? L = List([1,2,3]);
? listput(~L, 4); L \\ listput inserts at end
%4 = List([1, 2, 3, 4])
? listinsert(~L, 5, 1); L \\insert at position 1
%5 = List([5, 1, 2, 3, 4])
? listinsert(~L, 6, 1000); L \\ trying to insert beyond position #L
%6 = List([5, 1, 2, 3, 4, 6]) \\ ... inserts at the end
Note the ~L: this means that the function is called with a reference to L and changes L in place.
The library syntax is GEN listinsert(GEN ~L, GEN x, long n).

3.2.44 listkill( L). Obsolete, retained for backward compatibility. Just use L = List() instead
of listkill(L). In most cases, you won’t even need that, e.g. local variables are automatically
cleared when a user function returns.
The library syntax is void listkill(GEN ~L).

3.2.45 listpop( list, {n}). Removes the n-th element of the list list (which must be of type t_LIST).
If n is omitted, or greater than the list current length, removes the last element. If the list is already
empty, do nothing. This runs in time O(#L − n + 1).
? L = List([1,2,3,4]);
? listpop(~L); L \\ remove last entry
%2 = List([1, 2, 3])
? listpop(~L, 1); L \\ remove first entry
%3 = List([2, 3])
Note the ~L: this means that the function is called with a reference to L and changes L in place.
The library syntax is void listpop0(GEN ~list, long n).

100
3.2.46 listput( list, x, {n}). Sets the n-th element of the list list (which must be of type t_LIST)
equal to x. If n is omitted, or greater than the list length, appends x. The function returns the
inserted element.
? L = List();
? listput(~L, 1)
%2 = 1
? listput(~L, 2)
%3 = 2
? L
%4 = List([1, 2])
Note the ~L: this means that the function is called with a reference to L and changes L in place.
You may put an element into an occupied cell (not changing the list length), but it is easier
to use the standard list[n] = x construct.
? listput(~L, 3, 1) \\ insert at position 1
%5 = 3
? L
%6 = List([3, 2])
? L[2] = 4 \\ simpler
%7 = List([3, 4])
? L[10] = 1 \\ can’t insert beyond the end of the list
*** at top-level: L[10]=1
*** ^------
*** nonexistent component: index > 2
? listput(L, 1, 10) \\ but listput can
%8 = 1
? L
%9 = List([3, 2, 1])
This function runs in time O(#L) in the worst case (when the list must be reallocated), but in
time O(1) on average: any number of successive listputs run in time O(#L), where #L denotes
the list final length.
The library syntax is GEN listput0(GEN ~list, GEN x, long n).

3.2.47 listsort( L, {flag = 0}). Sorts the t_LIST list in place, with respect to the (somewhat
arbitrary) universal comparison function cmp. In particular, the ordering is the same as for sets
and setsearch can be used on a sorted list. No value is returned. If flag is nonzero, suppresses all
repeated coefficients.
? L = List([1,2,4,1,3,-1]); listsort(~L); L
%1 = List([-1, 1, 1, 2, 3, 4])
? setsearch(L, 4)
%2 = 6
? setsearch(L, -2)
%3 = 0
? listsort(~L, 1); L \\ remove duplicates
%4 = List([-1, 1, 2, 3, 4])

101
Note the ~L: this means that the function is called with a reference to L and changes L in place:
this is faster than the vecsort command since the list is sorted in place and we avoid unnecessary
copies.

? v = vector(100,i,random); L = List(v);
? for(i=1,10^4, vecsort(v))
time = 162 ms.
? for(i=1,10^4, vecsort(L))
time = 162 ms.
? for(i=1,10^4, listsort(~L))
time = 63 ms.

The library syntax is void listsort(GEN ~L, long flag).

3.2.48 localbitprec(p). Set the real precision to p bits in the dynamic scope. All computations
are performed as if realbitprecision was p: transcendental constants (e.g. Pi) and conversions
from exact to floating point inexact data use p bits, as well as iterative routines implicitly using a
floating point accuracy as a termination criterion (e.g. solve or intnum). But realbitprecision
itself is unaffected and is “unmasked” when we exit the dynamic (not lexical) scope. In effect, this
is similar to

my(bit = default(realbitprecision));
default(realbitprecision,p);
...
default(realbitprecision, bit);

but is both less cumbersome, cleaner (no need to manipulate a global variable, which in fact never
changes and is only temporarily masked) and more robust: if the above computation is interrupted
or an exception occurs, realbitprecision will not be restored as intended.

Such localbitprec statements can be nested, the innermost one taking precedence as ex-
pected. Beware that localbitprec follows the semantic of local, not my: a subroutine called
from localbitprec scope uses the local accuracy:

? f()=bitprecision(1.0);
? f()
%2 = 128
? localbitprec(1000); f()
%3 = 1024

Note that the bit precision of data (1.0 in the above example) increases by steps of 64 (32 on a
32-bit machine) so we get 1024 instead of the expected 1000; localbitprec bounds the relative
error exactly as specified in functions that support that granularity (e.g. lfun), and rounded to the
next multiple of 64 (resp. 32) everywhere else.

102
Warning. Changing realbitprecision or realprecision in programs is deprecated in favor of
localbitprec and localprec. Think about the realprecision and realbitprecision defaults
as interactive commands for the gp interpreter, best left out of GP programs. Indeed, the above
rules imply that mixing both constructs yields surprising results:

? \p38
? localprec(19); default(realprecision,1000); Pi
%1 = 3.141592653589793239
? \p
realprecision = 1001 significant digits (1000 digits displayed)

Indeed, realprecision itself is ignored within localprec scope, so Pi is computed to a low

accuracy. And when we leave the localprec scope, realprecision only regains precedence, it is
not “restored” to the original value.

3.2.49 localprec(p). Set the real precision to p in the dynamic scope and return p. All computa-
tions are performed as if realprecision was p: transcendental constants (e.g. Pi) and conversions
from exact to floating point inexact data use p decimal digits, as well as iterative routines implicitly
using a floating point accuracy as a termination criterion (e.g. solve or intnum). But realpre-
cision itself is unaffected and is “unmasked” when we exit the dynamic (not lexical) scope. In
effect, this is similar to

my(prec = default(realprecision));
default(realprecision,p);
...
default(realprecision, prec);

but is both less cumbersome, cleaner (no need to manipulate a global variable, which in fact never
changes and is only temporarily masked) and more robust: if the above computation is interrupted
or an exception occurs, realprecision will not be restored as intended.

Such localprec statements can be nested, the innermost one taking precedence as expected.
Beware that localprec follows the semantic of local, not my: a subroutine called from localprec
scope uses the local accuracy:

? f()=precision(1.);
? f()
%2 = 38
? localprec(19); f()
%3 = 19

103
Warning. Changing realprecision itself in programs is now deprecated in favor of localprec.
Think about the realprecision default as an interactive command for the gp interpreter, best left
out of GP programs. Indeed, the above rules imply that mixing both constructs yields surprising
results:

? \p38
? localprec(19); default(realprecision,100); Pi
%1 = 3.141592653589793239
? \p
realprecision = 115 significant digits (100 digits displayed)

Indeed, realprecision itself is ignored within localprec scope, so Pi is computed to a low

accuracy. And when we leave localprec scope, realprecision only regains precedence, it is not
“restored” to the original value.

3.2.50 mapdelete( M, x). Removes x from the domain of the map M .

? M = Map(["a",1; "b",3; "c",7]);

? mapdelete(M,"b");
? Mat(M)
["a" 1]
["c" 7]

The library syntax is void mapdelete(GEN ~M, GEN x).

3.2.51 mapget(M, x). Returns the image of x by the map M .

? M=Map(["a",23;"b",43]);
? mapget(M,"a")
%2 = 23
? mapget(M,"b")
%3 = 43

Raises an exception when the key x is not present in M .

? mapget(M,"c")
*** at top-level: mapget(M,"c")
*** ^-------------
*** mapget: nonexistent component in mapget: index not in map

The library syntax is GEN mapget(GEN M, GEN x).

104
3.2.52 mapisdefined(M, x, {&z}). Returns true (1) if x has an image by the map M , false (0)
otherwise. If z is present, set z to the image of x, if it exists.

? M1 = Map([1, 10; 2, 20]);

? mapisdefined(M1,3)
%1 = 0
? mapisdefined(M1, 1, &z)
%2 = 1
? z
%3 = 10

? M2 = Map(); N = 19;
? for (a=0, N-1, mapput(M2, a^3%N, a));
? {for (a=0, N-1,
if (mapisdefined(M2, a, &b),
printf("%d is the cube of %d mod %d\n",a,b,N)));}
0 is the cube of 0 mod 19
1 is the cube of 11 mod 19
7 is the cube of 9 mod 19
8 is the cube of 14 mod 19
11 is the cube of 17 mod 19
12 is the cube of 15 mod 19
18 is the cube of 18 mod 19

The library syntax is GEN mapisdefined(GEN M, GEN x, GEN *z = NULL).

3.2.53 mapput( M, x, y). Associates x to y in the map M . The value y can be retrieved with
mapget.

? M = Map();
? mapput(~M, "foo", 23);
? mapput(~M, 7718, "bill");
? mapget(M, "foo")
%4 = 23
? mapget(M, 7718)
%5 = "bill"
? Vec(M) \\ keys
%6 = [7718, "foo"]
? Mat(M)
%7 =
[ 7718 "bill"]
["foo" 23]

The library syntax is void mapput(GEN ~M, GEN x, GEN y).

105
3.2.54 print({str }∗). Outputs its arguments in raw format ending with a newline. The arguments
are converted to strings following the rules in Section 2.9.
? m = matid(2);
? print(m) \\ raw format
[1, 0; 0, 1]
? printp(m) \\ prettymatrix format
[1 0]
[0 1]

3.2.55 print1({str }∗). Outputs its arguments in raw format, without ending with a newline. Note
that you can still embed newlines within your strings, using the \n notation ! The arguments are
converted to strings following the rules in Section 2.9.

3.2.56 printf(fmt, {x}∗). This function is based on the C library command of the same name. It
prints its arguments according to the format fmt, which specifies how subsequent arguments are
converted for output. The format is a character string composed of zero or more directives:
• ordinary characters (not %), printed unchanged,
• conversions specifications (% followed by some characters) which fetch one argument from
the list and prints it according to the specification.
More precisely, a conversion specification consists in a %, one or more optional flags (among #,
0, -, +, ‘ ’), an optional decimal digit string specifying a minimal field width, an optional precision
in the form of a period (‘.’) followed by a decimal digit string, and the conversion specifier (among
d,i, o, u, x,X, p, e,E, f, g,G, s).

The flag characters. The character % is followed by zero or more of the following flags:
• #: the value is converted to an “alternate form”. For o conversion (octal), a 0 is prefixed
to the string. For x and X conversions (hexa), respectively 0x and 0X are prepended. For other
conversions, the flag is ignored.
• 0: the value should be zero padded. For d, i, o, u, x, X e, E, f, F, g, and G conversions, the
value is padded on the left with zeros rather than blanks. (If the 0 and - flags both appear, the 0
flag is ignored.)
• -: the value is left adjusted on the field boundary. (The default is right justification.) The
value is padded on the right with blanks, rather than on the left with blanks or zeros. A - overrides
a 0 if both are given.
• ‘ ’ (a space): a blank is left before a positive number produced by a signed conversion.
• +: a sign (+ or -) is placed before a number produced by a signed conversion. A + overrides
a space if both are used.

The field width. An optional decimal digit string (whose first digit is nonzero) specifying a
minimum field width. If the value has fewer characters than the field width, it is padded with
spaces on the left (or right, if the left-adjustment flag has been given). In no case does a small field
width cause truncation of a field; if the value is wider than the field width, the field is expanded
to contain the conversion result. Instead of a decimal digit string, one may write * to specify that
the field width is given in the next argument.

106
The precision. An optional precision in the form of a period (‘.’) followed by a decimal digit
string. This gives the number of digits to appear after the radix character for e, E, f, and F
conversions, the maximum number of significant digits for g and G conversions, and the maximum
number of characters to be printed from an s conversion. Instead of a decimal digit string, one
may write * to specify that the field width is given in the next argument.

The length modifier. This is ignored under gp, but necessary for libpari programming. De-
scription given here for completeness:
• l: argument is a long integer.
• P: argument is a GEN.

The conversion specifier. A character that specifies the type of conversion to be applied.
• d, i: a signed integer.
• o, u, x, X: an unsigned integer, converted to unsigned octal (o), decimal (u) or hexadecimal
(x or X) notation. The letters abcdef are used for x conversions; the letters ABCDEF are used for X
conversions.
• e, E: the (real) argument is converted in the style [ -]d.ddd e[ -]dd, where there is one
digit before the decimal point, and the number of digits after it is equal to the precision; if the
precision is missing, use the current realprecision for the total number of printed digits. If the
precision is explicitly 0, no decimal-point character appears. An E conversion uses the letter E
rather than e to introduce the exponent.
• f, F: the (real) argument is converted in the style [ -]ddd.ddd, where the number of digits
after the decimal point is equal to the precision; if the precision is missing, use the current real-
precision for the total number of printed digits. If the precision is explicitly 0, no decimal-point
character appears. If a decimal point appears, at least one digit appears before it.
• g, G: the (real) argument is converted in style e or f (or E or F for G conversions) [ -]ddd.ddd,
where the total number of digits printed is equal to the precision; if the precision is missing, use
the current realprecision. If the precision is explicitly 0, it is treated as 1. Style e is used when
the decimal exponent is < −4, to print 0., or when the integer part cannot be decided given the
known significant digits, and the f format otherwise.
• c: the integer argument is converted to an unsigned char, and the resulting character is
written.
• s: convert to a character string. If a precision is given, no more than the specified number
of characters are written.
• p: print the address of the argument in hexadecimal (as if by %#x).
• %: a % is written. No argument is converted. The complete conversion specification is %%.
Examples:
? printf("floor: %d, field width 3: %3d, with sign: %+3d\n", Pi, 1, 2);
floor: 3, field width 3: 1, with sign: +2
? printf("%.5g %.5g %.5g\n",123,123/456,123456789);
123.00 0.26974 1.2346 e8
? printf("%-2.5s:%2.5s:%2.5s\n", "P", "PARI", "PARIGP");

107
 P :PARI:PARIG
\\ min field width and precision given by arguments
? x = 23; y=-1/x; printf("x=%+06.2f y=%+0*.*f\n", x, 6, 2, y);
x=+23.00 y=-00.04
\\ minimum fields width 5, pad left with zeroes
? for (i = 2, 5, printf("%05d\n", 10^i))
00100
01000
10000
100000 \\ don’t truncate fields whose length is larger than the minimum width
? printf("%.2f |%06.2f|", Pi,Pi)
3.14 | 3.14|
All numerical conversions apply recursively to the entries of vectors and matrices:
? printf("%4d", [1,2,3]);
[ 1, 2, 3]
? printf("%5.2f", mathilbert(3));
[ 1.00 0.50 0.33]
[ 0.50 0.33 0.25]
[ 0.33 0.25 0.20]

Technical note. Our implementation of printf deviates from the C89 and C99 standards in a
few places:
• whenever a precision is missing, the current realprecision is used to determine the number
of printed digits (C89: use 6 decimals after the radix character).
• in conversion style e, we do not impose that the exponent has at least two digits; we never
write a + sign in the exponent; 0 is printed in a special way, always as 0.Eexp.
• in conversion style f, we switch to style e if the exponent is greater or equal to the precision.
• in conversion g and G, we do not remove trailing zeros from the fractional part of the result;
nor a trailing decimal point; 0 is printed in a special way, always as 0.Eexp.

3.2.57 printp({str }∗). Outputs its arguments in prettymatrix format, ending with a newline. The
arguments are converted to strings following the rules in Section 2.9.
? m = matid(2);
? print(m) \\ raw format
[1, 0; 0, 1]
? printp(m) \\ prettymatrix format
[1 0]
[0 1]

3.2.58 printsep(sep, {str }∗). Outputs its arguments in raw format, ending with a newline. The
arguments are converted to strings following the rules in Section 2.9. Successive entries are separated
by sep:
? printsep(":", 1,2,3,4)
1:2:3:4

108
3.2.59 printsep1(sep, {str }∗). Outputs its arguments in raw format, without ending with a
newline. The arguments are converted to strings following the rules in Section 2.9. Successive
entries are separated by sep:

? printsep1(":", 1,2,3,4);print("|")
1:2:3:4|

3.2.60 printtex({str }∗). Outputs its arguments in TEX format. This output can then be used in
a TEX manuscript, see strtex for details. The arguments are converted to strings following the
rules in Section 2.9. The printing is done on the standard output. If you want to print it to a file
you should use writetex (see there).

Another possibility is to enable the log default (see Section 2.12). You could for instance do:

default(logfile, "new.tex");
default(log, 1);
printtex(result);

3.2.61 quit({status = 0}). Exits gp and return to the system with exit status status, a small
integer. A nonzero exit status normally indicates abnormal termination. (Note: the system actually
sees only status mod 256, see your man pages for exit(3) or wait(2)).

3.2.62 read({filename}). Reads in the file filename (subject to string expansion). If filename
is omitted, re-reads the last file that was fed into gp. The return value is the result of the last
expression evaluated.

If a GP binary file is read using this command (see Section 3.2.86), the file is loaded and
the last object in the file is returned.

In case the file you read in contains an allocatemem statement (to be generally avoided), you
should leave read instructions by themselves, and not part of larger instruction sequences.

Variants. readvec allows to read a whole file at once; fileopen followed by either fileread
(evaluated lines) or filereadstr (lines as nonevaluated strings) allows to read a file one line at a
time.

The library syntax is GEN gp_read_file(const char *filename).

3.2.63 readstr({filename}). Reads in the file filename and return a vector of GP strings, each
component containing one line from the file. If filename is omitted, re-reads the last file that was
fed into gp.

The library syntax is GEN readstr(const char *filename).

109
3.2.64 readvec({filename}). Reads in the file filename (subject to string expansion). If filename
is omitted, re-reads the last file that was fed into gp. The return value is a vector whose components
are the evaluation of all sequences of instructions contained in the file. For instance, if file contains
1
2
3
then we will get:
? \r a
%1 = 1
%2 = 2
%3 = 3
? read(a)
%4 = 3
? readvec(a)
%5 = [1, 2, 3]
In general a sequence is just a single line, but as usual braces and \ may be used to enter
multiline sequences.
The library syntax is GEN gp_readvec_file(const char *filename). The underlying li-
brary function GEN gp_readvec_stream(FILE *f) is usually more flexible.

3.2.65 select(f, A, {flag = 0}). We first describe the default behavior, when flag is 0 or omitted.
Given a vector or list A and a t_CLOSURE f, select returns the elements x of A such that f (x) is
nonzero. In other words, f is seen as a selection function returning a boolean value.
? select(x->isprime(x), vector(50,i,i^2+1))
%1 = [2, 5, 17, 37, 101, 197, 257, 401, 577, 677, 1297, 1601]
? select(x->(x<100), %)
%2 = [2, 5, 17, 37]
returns the primes of the form i2 + 1 for some i ≤ 50, then the elements less than 100 in the
preceding result. The select function also applies to a matrix A, seen as a vector of columns, i.e.
it selects columns instead of entries, and returns the matrix whose columns are the selected ones.

Remark. For v a t_VEC, t_COL, t_VECSMALL, t_LIST or t_MAT, the alternative set-notations
[g(x) | x <- v, f(x)]
[x | x <- v, f(x)]
[g(x) | x <- v]
are available as shortcuts for
apply(g, select(f, Vec(v)))
select(f, Vec(v))
apply(g, Vec(v))
respectively:
? [ x | x <- vector(50,i,i^2+1), isprime(x) ]
%1 = [2, 5, 17, 37, 101, 197, 257, 401, 577, 677, 1297, 1601]

110
If flag = 1, this function returns instead the indices of the selected elements, and not the elements
themselves (indirect selection):
? V = vector(50,i,i^2+1);
? select(x->isprime(x), V, 1)
%2 = Vecsmall([1, 2, 4, 6, 10, 14, 16, 20, 24, 26, 36, 40])
? vecextract(V, %)
%3 = [2, 5, 17, 37, 101, 197, 257, 401, 577, 677, 1297, 1601]
The following function lists the elements in (Z/N Z)∗ :
? invertibles(N) = select(x->gcd(x,N) == 1, [1..N])
Finally
? select(x->x, M)
selects the nonzero entries in M. If the latter is a t_MAT, we extract the matrix of nonzero columns.
Note that removing entries instead of selecting them just involves replacing the selection function
f with its negation:
? select(x->!isprime(x), vector(50,i,i^2+1))
The library syntax is genselect(void *E, long (*fun)(void*,GEN), GEN a). Also avail-
able is GEN genindexselect(void *E, long (*fun)(void*, GEN), GEN a), corresponding to
flag = 1.

3.2.66 self(). Return the calling function or closure as a t_CLOSURE object. This is useful for
defining anonymous recursive functions.
? (n -> if(n==0,1,n*self()(n-1)))(5)
%1 = 120 \\ 5!
? (n -> if(n<=1, n, self()(n-1)+self()(n-2)))(20)
%2 = 6765 \\ Fibonacci(20)
The library syntax is GEN pari_self().

3.2.67 setrand(n). Reseeds the random number generator using the seed n. No value is returned.
The seed is a small positive integer 0 < n < 264 used to generate deterministically a suitable state
array. All gp session start by an implicit setrand(1), so resetting the seed to this value allows to
replay all computations since the session start. Alternatively, running a randomized computation
starting by setrand(n) twice with the same n will generate the exact same output.
In the other direction, including a call to setrand(getwalltime()) from your gprc will cause
GP to produce different streams of random numbers in each session. (Unix users may want to use
/dev/urandom instead of getwalltime.)
For debugging purposes, one can also record a particular random state using getrand (the
value is encoded as a huge integer) and feed it to setrand:
? state = getrand(); \\ record seed
...
? setrand(state); \\ we can now replay the exact same computations
The library syntax is void setrand(GEN n).

111
3.2.68 strchr(x). Converts integer or vector of integers x to a string, translating each integer (in
the range [1, 255]) into a character using ASCII encoding.

? strchr(97)
%1 = "a"
? Vecsmall("hello world")
%2 = Vecsmall([104, 101, 108, 108, 111, 32, 119, 111, 114, 108, 100])
? strchr(%)
%3 = "hello world"

The library syntax is GEN pari_strchr(GEN x).

3.2.69 strexpand({x}∗). Converts its argument list into a single character string (type t_STR,
the empty string if x is omitted). Then perform environment expansion, see Section 2.12. This
feature can be used to read environment variable values.

? strexpand("$HOME/doc")
%1 = "/home/pari/doc"
? module = "aprcl"; n = 10;
? strexpand("$HOME/doc/", module, n, ".tex")
%3 = "/home/pari/doc/aprcl10.tex"

The individual arguments are read in string context, see Section 2.9.

3.2.70 strjoin(v, {p = ””}). Joins the strings in vector v, separating them with delimiter p. The
reverse operation is strsplit.

? v = ["abc", "def", "ghi"]

? strjoin(v, "/")
%2 = "abc/def/ghi"
? strjoin(v)
%3 = "abcdefghi"

The library syntax is GEN strjoin(GEN v, GEN p = NULL).

3.2.71 strprintf(fmt, {x}∗). Returns a string built from the remaining arguments according to the
format fmt. The format consists of ordinary characters (not %), printed unchanged, and conversions
specifications. See printf.

? dir = "/home/pari"; file = "aprcl"; n = 10;

? strprintf("%s/%s%ld.tex", dir, file, n)
%2 = "/home/pari/aprcl10.tex"

112
3.2.72 strsplit(s, {p = ””}). Splits the string s into a vector of strings, with p acting as a delimiter.
If p is empty or omitted, split the string into characters.
? strsplit("abc::def::ghi", "::")
%1 = ["abc", "def", "ghi"]
? strsplit("abc", "")
%2 = ["a", "b", "c"]
? strsplit("aba", "a")
If s starts (resp. ends) with the pattern p, then the first (resp. last) entry in the vector is the empty
string:
? strsplit("aba", "a")
%3 = ["", "b", ""]
The library syntax is GEN strsplit(GEN s, GEN p = NULL).

3.2.73 strtex({x}∗). Translates its arguments to TeX format, and concatenates the results into a
single character string (type t_STR, the empty string if x is omitted).
The individual arguments are read in string context, see Section 2.9.
? v = [1, 2, 3]
%1 [1, 2, 3]
? strtex(v)
%2 = "\\pmatrix{ 1&2&3\\cr}\n"

TEX-nical notes. The TeX output engine was originally written for plain TeX and designed
for maximal portability. Unfortunately later LaTeX packages have obsoleted valid TEX primitives,
leading us to replace TeX’s \over by LaTeX’s \frac in PARI’s TeX output. We have decided
not to update further our TeX markup and let the users of various LaTeX engines customize their
preambles. The following documents the precise changes you may need to include in your style files
to incorporate PARI TeX output verbatim:
• if you enabled bit 4 in TeXstyle default, you must define \PARIbreak; see ??TeXstyle;
• if you use plain TeX only: you must define \frac as follows
\def\frac#1#2{{#1\over#2}}
• if you use LaTeX and amsmath, \pmatrix is obsoleted in favor of the pmatrix environment;
see examples/parigp.sty for how to re-enable the deprecated construct.

3.2.74 strtime(t). Return a string describing the time t in milliseconds in the format used by the
GP timer.
? print(strtime(12345678))
3h, 25min, 45,678 ms
? {
my(t=getabstime());
F=factor(2^256+1);t=getabstime()-t;
print("factor(2^256+1) took ",strtime(t));
}
factor(2^256+1) took 1,320 ms
The library syntax is GEN strtime(long t).

113
3.2.75 system(str ). str is a string representing a system command. This command is executed,
its output written to the standard output (this won’t get into your logfile), and control returns to
the PARI system. This simply calls the C system command.
The library syntax is void gpsystem(const char *str).

3.2.76 trap({e}, {rec}, seq). This function is obsolete, use iferr, which has a nicer and much
more powerful interface. For compatibility’s sake we now describe the obsolete function trap.
This function tries to evaluate seq, trapping runtime error e, that is effectively preventing it
from aborting computations in the usual way; the recovery sequence rec is executed if the error
occurs and the evaluation of rec becomes the result of the command. If e is omitted, all exceptions
are trapped. See Section 2.10.2 for an introduction to error recovery under gp.
? \\ trap division by 0
? inv(x) = trap (e_INV, INFINITY, 1/x)
? inv(2)
%1 = 1/2
? inv(0)
%2 = INFINITY
Note that seq is effectively evaluated up to the point that produced the error, and the recovery
sequence is evaluated starting from that same context, it does not ”undo” whatever happened in
the other branch (restore the evaluation context):
? x = 1; trap (, /* recover: */ x, /* try: */ x = 0; 1/x)
%1 = 0

Note. The interface is currently not adequate for trapping individual exceptions. In the current
version 2.13.3, the following keywords are recognized, but the name list will be expanded and
changed in the future (all library mode errors can be trapped: it’s a matter of defining the keywords
to gp):
e ALARM: alarm time-out
e ARCH: not available on this architecture or operating system
e STACK: the PARI stack overflows
e INV: impossible inverse
e IMPL: not yet implemented
e OVERFLOW: all forms of arithmetic overflow, including length or exponent overflow (when a
larger value is supplied than the implementation can handle).
e SYNTAX: syntax error
e MISC: miscellaneous error
e TYPE: wrong type
e USER: user error (from the error function)
The library syntax is GEN trap0(const char *e = NULL, GEN rec = NULL, GEN seq =
NULL).

114
3.2.77 type(x). This is useful only under gp. Returns the internal type name of the PARI object
x as a string. Check out existing type names with the metacommand \t. For example type(1)
will return ”t_INT”.
The library syntax is GEN type0(GEN x). The macro typ is usually simpler to use since it
returns a long that can easily be matched with the symbols t_*. The name type was avoided since
it is a reserved identifier for some compilers.

3.2.78 unexport(x, ..., z). Remove x, . . . , z from the list of variables exported to the parallel world.
See export.

3.2.79 unexportall(). Empty the list of variables exported to the parallel world.
The library syntax is void unexportall().

3.2.80 uninline(). Exit the scope of all current inline variables. DEPRECATED, use export /
unexport.

3.2.81 version(). Returns the current version number as a t_VEC with three integer compo-
nents (major version number, minor version number and patchlevel); if your sources were obtained
through our version control system, this will be followed by further more precise arguments, in-
cluding e.g. a git commit hash.
This function is present in all versions of PARI following releases 2.3.4 (stable) and 2.4.3
(testing).
Unless you are working with multiple development versions, you probably only care about the
3 first numeric components. In any case, the lex function offers a clever way to check against
a particular version number, since it will compare each successive vector entry, numerically or as
strings, and will not mind if the vectors it compares have different lengths:
if (lex(version(), [2,3,5]) >= 0,
\\ code to be executed if we are running 2.3.5 or more recent.
,
\\ compatibility code
);
On a number of different machines, version() could return either of
%1 = [2, 3, 4] \\ released version, stable branch
%1 = [2, 4, 3] \\ released version, testing branch
%1 = [2, 6, 1, 15174, ""505ab9b"] \\ development
In particular, if you are only working with released versions, the first line of the gp introductory
message can be emulated by
[M,m,p] = version();
printf("GP/PARI CALCULATOR Version %s.%s.%s", M,m,p);

If you are working with many development versions of PARI/GP, the 4th and/or 5th components
can be profitably included in the name of your logfiles, for instance.

115
Technical note. For development versions obtained via git, the 4th and 5th components are
liable to change eventually, but we document their current meaning for completeness. The 4th
component counts the number of reachable commits in the branch (analogous to svn’s revision
number), and the 5th is the git commit hash. In particular, lex comparison still orders correctly
development versions with respect to each others or to released versions (provided we stay within
a given branch, e.g. master)!

The library syntax is GEN pari_version().

3.2.82 warning({str }∗). Outputs the message “user warning” and the argument list (each of them
interpreted as a string). If colors are enabled, this warning will be in a different color, making it
easy to distinguish.

warning(n, " is very large, this might take a while.")

3.2.83 whatnow(key). If keyword key is the name of a function that was present in GP version
1.39.15, outputs the new function name and syntax, if it changed at all. Functions that where
introduced since then, then modified are also recognized.

? whatnow("mu")
New syntax: mu(n) ===> moebius(n)
moebius(x): Moebius function of x.
? whatnow("sin")
This function did not change

When a function was removed and the underlying functionality is not available under a com-
patible interface, no equivalent is mentioned:

? whatnow("buchfu")
This function no longer exists

(The closest equivalent would be to set K = bnfinit(T) then access K.fu.)

3.2.84 write(filename, {str }∗). Writes (appends) to filename the remaining arguments, and ap-
pends a newline (same output as print).

Variant. The high-level function write is expensive when many consecutive writes are expected
because it cannot use buffering. The low-level interface fileopen / filewrite / fileclose is
more efficient. It also allows to truncate existing files and replace their contents.

3.2.85 write1(filename, {str }∗). Writes (appends) to filename the remaining arguments without
a trailing newline (same output as print1).

116
3.2.86 writebin(filename, {x}). Writes (appends) to filename the object x in binary format. This
format is not human readable, but contains the exact internal structure of x, and is much faster to
save/load than a string expression, as would be produced by write. The binary file format includes
a magic number, so that such a file can be recognized and correctly input by the regular read or
\r function. If saved objects refer to polynomial variables that are not defined in the new session,
they will be displayed as tn for some integer n (the attached variable number). Installed functions
and history objects can not be saved via this function.

If x is omitted, saves all user variables from the session, together with their names. Reading
such a “named object” back in a gp session will set the corresponding user variable to the saved
value. E.g after

x = 1; writebin("log")

reading log into a clean session will set x to 1. The relative variables priorities (see Section 2.5.3)
of new variables set in this way remain the same (preset variables retain their former priority, but
are set to the new value). In particular, reading such a session log into a clean session will restore
all variables exactly as they were in the original one.

Just as a regular input file, a binary file can be compressed using gzip, provided the file name
has the standard .gz extension.

In the present implementation, the binary files are architecture dependent and compatibility
with future versions of gp is not guaranteed. Hence binary files should not be used for long term
storage (also, they are larger and harder to compress than text files).

The library syntax is void gpwritebin(const char *filename, GEN x = NULL).

3.2.87 writetex(filename, {str }∗). As write, in TEX format. See strtex for details: this function
is essentially equivalent to calling strtex on remaining arguments and writing them to file.

3.3 Parallel programming.

These function are only available if PARI was configured using Configure --mt=. . . . Two
multithread interfaces are supported:

• POSIX threads

• Message passing interface (MPI)

As a rule, POSIX threads are well-suited for single systems, while MPI is used by most clusters.
However the parallel GP interface does not depend on the chosen multithread interface: a properly
written GP program will work identically with both.

117
3.3.1 parapply(f, x). Parallel evaluation of f on the elements of x. The function f must not
access global variables or variables declared with local(), and must be free of side effects.

parapply(factor,[2^256 + 1, 2^193 - 1])

factors 2256 + 1 and 2193 − 1 in parallel.

{
my(E = ellinit([1,3]), V = vector(12,i,randomprime(2^200)));
parapply(p->ellcard(E,p), V)
}

computes the order of E(Fp ) for 12 random primes of 200 bits.

The library syntax is GEN parapply(GEN f, GEN x).

3.3.2 pareval(x). Parallel evaluation of the elements of x, where x is a vector of closures. The
closures must be of arity 0, must not access global variables or variables declared with local and
must be free of side effects.

Here is an artificial example explaining the MOV attack on the elliptic discrete log problem
(by reducing it to a standard discrete log over a finite field):

{
my(q = 2^30 + 3, m = 40 * q; p = 1 + m^2); \\ p, q are primes
my(E = ellinit([0,0,0,1,0] * Mod(1,p)));
my([P, Q] = ellgenerators(E));
\\ E(F_p) ~ Z/m P + Z/m Q and the order of the
\\ Weil pairing <P,Q> in (Z/p)^* is m
my(F = [m,factor(m)], e = random(m), R, wR, wQ);
R = ellpow(E, Q, e);
wR = ellweilpairing(E,P,R,m);
wQ = ellweilpairing(E,P,Q,m); \\ wR = wQ^e
pareval([()->znlog(wR,wQ,F), ()->elllog(E,R,Q), ()->e])
}

Note the use of my to pass ”arguments” to the functions we need to evaluate while satisfying the
listed requirements: closures of arity 0 and no global variables (another possibility would be to
use export). As a result, the final three statements satisfy all the listed requirements and are run
in parallel. (Which is silly for this computation but illustrates the use of pareval.) The function
parfor is more powerful but harder to use.

The library syntax is GEN pareval(GEN x).

118
3.3.3 parfor(i = a, {b}, expr1 , {r}, {expr2 }). Evaluates in parallel the expression expr1 in the
formal argument i running from a to b. If b is set to +oo, the loop runs indefinitely. If r and
expr2 are present, the expression expr2 in the formal variables r and i is evaluated with r running
through all the different results obtained for expr1 and i takes the corresponding argument.
The computations of expr1 are started in increasing order of i; otherwise said, the computation
for i = c is started after those for i = 1, . . . , c − 1 have been started, but before the computation for
i = c + 1 is started. Notice that the order of completion, that is, the order in which the different r
become available, may be different; expr2 is evaluated sequentially on each r as it appears.
The following example computes the sum of the squares of the integers from 1 to 10 by
computing the squares in parallel and is equivalent to parsum (i=1, 10, i^2):
? s=0;
? parfor (i=1, 10, i^2, r, s=s+r)
? s
%3 = 385
More precisely, apart from a potentially different order of evaluation due to the parallelism,
the line containing parfor is equivalent to
? my (r); for (i=1, 10, r=i^2; s=s+r)
The sequentiality of the evaluation of expr2 ensures that the variable s is not modified con-
currently by two different additions, although the order in which the terms are added is nondeter-
ministic.
It is allowed for expr2 to exit the loop using break/next/return. If that happens for i = c,
then the evaluation of expr1 and expr2 is continued for all values i < c, and the return value is
the one obtained for the smallest i causing an interruption in expr2 (it may be undefined if this
is a break/next). In that case, using side-effects in expr2 may lead to undefined behavior, as the
exact number of values of i for which it is executed is nondeterministic. The following example
computes nextprime(1000) in parallel:
? parfor (i=1000, , isprime (i), r, if (r, return (i)))
%1 = 1009

3.3.4 parforeach(V, x, expr1 , {r}, {expr2 }). Evaluates in parallel the expression expr1 in the
formal argument x, where x runs through all components of V . If r and expr2 are present,
evaluate sequentially the expression expr2, in which the formal variables x and r are replaced
by the successive arguments and corresponding values. The sequential evaluation ordering is not
specified:
? parforeach([50..100], x,isprime(x), r, if(r,print(x)))
53
67
71
79
83
89
97
73
59
61

119
3.3.5 parforprime(p = a, {b}, expr1 , {r}, {expr2 }). Behaves exactly as parfor, but loops only
over prime values p. Precisely, the functions evaluates in parallel the expression expr1 in the formal
argument p running through the primes from a to b. If b is set to +oo, the loop runs indefinitely.
If r and expr2 are present, the expression expr2 in the formal variables r and p is evaluated
with r running through all the different results obtained for expr1 and p takes the corresponding
argument.
It is allowed fo expr2 to exit the loop using break/next/return; see the remarks in the
documentation of parfor for details.

3.3.6 parforprimestep(p = a, {b}, q, expr1 , {r}, {expr2 }). Behaves exactly as parfor, but loops
only over prime values p in an arithmetic progression Precisely, the functions evaluates in parallel
the expression expr1 in the formal argument p running through the primes from a to b in an
arithmetic progression of the form a + kq. (p ≡ a (mod q)) or an intmod Mod(c,N). If b is set
to +oo, the loop runs indefinitely. If r and expr2 are present, the expression expr2 in the formal
variables r and p is evaluated with r running through all the different results obtained for expr1
and p takes the corresponding argument.
It is allowed fo expr2 to exit the loop using break/next/return; see the remarks in the
documentation of parfor for details.

3.3.7 parforvec(X = v, expr1 , {j}, {expr2 }, {flag}). Evaluates the sequence expr2 (dependent on
X and j) for X as generated by forvec, in random order, computed in parallel. Substitute for j
the value of expr1 (dependent on X).
It is allowed fo expr2 to exit the loop using break/next/return, however in that case, expr2
will still be evaluated for all remaining value of p less than the current one, unless a subsequent
break/next/return happens.

3.3.8 parselect(f, A, {flag = 0}). Selects elements of A according to the selection function f , done
in parallel. If flagis 1, return the indices of those elements (indirect selection) The function f must
not access global variables or variables declared with local(), and must be free of side effects.
The library syntax is GEN parselect(GEN f, GEN A, long flag).

3.3.9 parsum(i = a, b, expr ). Sum of expression expr , the formal parameter going from a to b,
evaluated in parallel in random order. The expression expr must not access global variables or
variables declared with local(), and must be free of side effects.
? parsum(i=1,1000,ispseudoprime(2^prime(i)-1))
cpu time = 1min, 26,776 ms, real time = 5,854 ms.
%1 = 20
returns the number of prime numbers among the first 1000 Mersenne numbers.

3.3.10 parvector(N, i, expr ). As vector(N,i,expr) but the evaluations of expr are done in
parallel. The expression expr must not access global variables or variables declared with local(),
and must be free of side effects.
parvector(10,i,quadclassunit(2^(100+i)+1).no)
computes the class numbers in parallel.

120
 3.4 GP defaults.
This section documents the GP defaults, that can be set either by the GP function default
or in your GPRC. Be sure to check out parisize and parisizemax !

3.4.1 TeXstyle. The bits of this default allow gp to use less rigid TeX formatting commands in
the logfile. This default is only taken into account when log = 3. The bits of TeXstyle have the
following meaning
2: insert \right / \left pairs where appropriate.
4: insert discretionary breaks in polynomials, to enhance the probability of a good line break.
You must then define \PARIbreak as follows:
\def\PARIbreak{\hskip 0pt plus \hsize\relax\discretionary{}{}{}}
The default value is 0.

3.4.2 breakloop. If true, enables the “break loop” debugging mode, see Section 2.10.3.
The default value is 1 if we are running an interactive gp session, and 0 otherwise.

3.4.3 colors. This default is only usable if gp is running within certain color-capable terminals.
For instance rxvt, color xterm and modern versions of xterm under X Windows, or standard
Linux/DOS text consoles. It causes gp to use a small palette of colors for its output. With xterms,
the colormap used corresponds to the resources Xterm*colorn where n ranges from 0 to 15 (see
the file misc/color.dft for an example). Accepted values for this default are strings "a1 ,. . . ,ak "
where k ≤ 7 and each ai is either
• the keyword no (use the default color, usually black on transparent background)
• an integer between 0 and 15 corresponding to the aforementioned colormap
• a triple [c0 , c1 , c2 ] where c0 stands for foreground color, c1 for background color, and c2 for
attributes (0 is default, 1 is bold, 4 is underline).
The output objects thus affected are respectively error messages, history numbers, prompt,
input line, output, help messages, timer (that’s seven of them). If k < 7, the remaining ai are
assumed to be no. For instance
default(colors, "9, 5, no, no, 4")
typesets error messages in color 9, history numbers in color 5, output in color 4, and does not affect
the rest.
A set of default colors for dark (reverse video or PC console) and light backgrounds respectively
is activated when colors is set to darkbg, resp. lightbg (or any proper prefix: d is recognized as
an abbreviation for darkbg). A bold variant of darkbg, called boldfg, is provided if you find the
former too pale.
EMACS: In the present version, this default is incompatible with PariEmacs. Changing it will just fail silently
(the alternative would be to display escape sequences as is, since Emacs will refuse to interpret
them). You must customize color highlighting from the PariEmacs side, see its documentation.
The default value is "" (no colors).

121
3.4.4 compatible. Obsolete. This default is now a no-op.

3.4.5 datadir. The name of directory containing the optional data files. For now, this includes
the elldata, galdata, galpol, seadata packages.
The default value is /usr/local/share/pari, or the override specified via Configure --
datadir=.
Windows-specific note. On Windows operating systems, the special value @ stands for “the
directory where the gp binary is installed”. This is the default value.

3.4.6 debug. Debugging level. If it is nonzero, some extra messages may be printed, according to
what is going on (see \g).
The default value is 0 (no debugging messages).

3.4.7 debugfiles. File usage debugging level. If it is nonzero, gp will print information on file
descriptors in use, from PARI’s point of view (see \gf).
The default value is 0 (no debugging messages).

3.4.8 debugmem. Memory debugging level (see \gm). If this is nonzero, gp will print increasingly
precise notifications about memory use:
• debugmem > 0, notify when parisize changes (within the boundaries set by parisizemax);
• debugmem > 1, indicate any important garbage collection and the function it is taking place
in;
• debugmem > 2, indicate the creation/destruction of “blocks” (or clones); expect lots of
messages.
Important Note: if you are running a version compiled for debugging (see Appendix A) and
debugmem > 1, gp will further regularly print information on memory usage, notifying whenever
stack usage goes up or down by 1 MByte. This functionality is disabled on non-debugging builds
as it noticeably slows down the performance.
The default value is 1.

3.4.9 echo. This default can be 0 (off), 1 (on) or 2 (on, raw). When echo mode is on, each
command is reprinted before being executed. This can be useful when reading a file with the \r
or read commands. For example, it is turned on at the beginning of the test files used to check
whether gp has been built correctly (see \e). When echo is set to 1 the input is cleaned up,
removing white space and comments and uniting multi-line input. When set to 2 (raw), the input
is written as-is, without any pre-processing.
The default value is 0 (no echo).

3.4.10 factor add primes. This toggle is either 1 (on) or 0 (off). If on, the integer factorization
machinery calls addprimes on prime factors that were difficult to find (larger than 224 ), so they
are automatically tried first in other factorizations. If a routine is performing (or has performed)
a factorization and is interrupted by an error or via Control-C, this lets you recover the prime
factors already found. The downside is that a huge addprimes table unrelated to the current
computations will slow down arithmetic functions relying on integer factorization; one should then
empty the table using removeprimes.
The default value is 0.

122
3.4.11 factor proven. This toggle is either 1 (on) or 0 (off). By default, the factors output by
the integer factorization machinery are only pseudo-primes, not proven primes. If this toggle is set,
a primality proof is done for each factor and all results depending on integer factorization are fully
proven. This flag does not affect partial factorization when it is explicitly requested. It also does
not affect the private table managed by addprimes: its entries are included as is in factorizations,
without being tested for primality.
The default value is 0.

3.4.12 format. Of the form x.n, where x (conversion style) is a letter in {e, f, g}, and n (precision)
is an integer; this affects the way real numbers are printed:
• If the conversion style is e, real numbers are printed in scientific format, always with an
explicit exponent, e.g. 3.3 E-5.
• In style f, real numbers are generally printed in fixed floating point format without exponent,
e.g. 0.000033. A large real number, whose integer part is not well defined (not enough significant
digits), is printed in style e. For instance 10.^100 known to ten significant digits is always printed
in style e.
• In style g, nonzero real numbers are printed in f format, except when their decimal exponent
is < −4, in which case they are printed in e format. Real zeroes (of arbitrary exponent) are printed
in e format.
The precision n is the number of significant digits printed for real numbers, except if n < 0
where all the significant digits will be printed (initial default 28, or 38 for 64-bit machines). For
more powerful formatting possibilities, see printf and strprintf.
The default value is "g.28" and "g.38" on 32-bit and 64-bit machines, respectively.

3.4.13 graphcolormap. A vector of colors, to be used by hi-res graphing routines. Its length
is arbitrary, but it must contain at least 3 entries: the first 3 colors are used for background,
frame/ticks and axes respectively. All colors in the colormap may be freely used in plotcolor
calls.
A color is either given as in the default by character strings or by an RGB code. For valid color
names, see the standard rgb.txt file in X11 distributions, where we restrict to lowercase letters and
remove all whitespace from color names. An RGB code is a vector with 3 integer entries between
0 and 255 or a # followed by 6 hexadecimal digits. For instance [250, 235, 215], "#faebd7" and
"antiquewhite" all represent the same color.
The default value is ["white", "black", "blue", "violetred", "red", "green", "grey",
"gainsboro"].

3.4.14 graphcolors. Entries in the graphcolormap that will be used to plot multi-curves. The
successive curves are drawn in colors
graphcolormap[graphcolors[1]], graphcolormap[graphcolors[2]], . . .
cycling when the graphcolors list is exhausted.
The default value is [4,5].

3.4.15 help. Name of the external help program to use from within gp when extended help is
invoked, usually through a ?? or ??? request (see Section 2.13.1), or M-H under readline (see
Section 2.15).

123
Windows-specific note. On Windows operating systems, if the first character of help is @, it is
replaced by “the directory where the gp binary is installed”.
The default value is the path to the gphelp script we install.

3.4.16 histfile. Name of a file where gp will keep a history of all input commands (results are
omitted). If this file exists when the value of histfile changes, it is read in and becomes part
of the session history. Thus, setting this default in your gprc saves your readline history between
sessions. Setting this default to the empty string "" changes it to <undefined>. Note that, by
default, the number of history entries saved is not limited: set history-size in readline’s .inputrc
to limit the file size.
The default value is <undefined> (no history file).

3.4.17 histsize. gp keeps a history of the last histsize results computed so far, which you can
recover using the % notation (see Section 2.13.4). When this number is exceeded, the oldest values
are erased. Tampering with this default is the only way to get rid of the ones you do not need
anymore.
The default value is 5000.

3.4.18 lines. If set to a positive value, gp prints at most that many lines from each result,
terminating the last line shown with [+++] if further material has been suppressed. The various
print commands (see Section 3.2) are unaffected, so you can always type print(%) or \a to view
the full result. If the actual screen width cannot be determined, a “line” is assumed to be 80
characters long.
The default value is 0.

3.4.19 linewrap. If set to a positive value, gp wraps every single line after printing that many
characters.
The default value is 0 (unset).

3.4.20 log. This can be either 0 (off) or 1, 2, 3 (on, see below for the various modes). When
logging mode is turned on, gp opens a log file, whose exact name is determined by the logfile
default. Subsequently, all the commands and results will be written to that file (see \l). In case a
file with this precise name already existed, it will not be erased: your data will be appended at the
end.
The specific positive values of log have the following meaning
1: plain logfile
2: emit color codes to the logfile (if colors is set).
3: write LaTeX output to the logfile (can be further customized using TeXstyle).
The default value is 0.

124
Note. Logging starts as soon as log is set to a nonzero value. In particular, when log is set in gprc,
warnings and errors triggered from the rest of the file will be written in the logfile. For instance, on
clean startup, the logfile will start by Done. (from the Reading GPRC:. . . Done. diagnostic printed
when starting gp), then the gp header and prompt.

3.4.21 logfile. Name of the log file to be used when the log toggle is on. Environment and time
expansion are performed.
The default value is "pari.log".

3.4.22 nbthreads. This default is specific to the parallel version of PARI and gp (built via
Configure --mt=prthread or mpi) and is ignored otherwise. In parallel mode, it governs the
number of threads to use for parallel computing. The exact meaning and default value depend on
the mt engine used:
• single: not used (always a single thread).
• pthread: number of threads (unlimited, default: number of cores)
• mpi: number of MPI processes to use (limited to the number allocated by mpirun, default:
use all allocated processes).
See also threadsize and threadsizemax.

3.4.23 new galois format. This toggle is either 1 (on) or 0 (off). If on, the polgalois command
will use a different, more consistent, naming scheme for Galois groups. This default is provided to
ensure that scripts can control this behavior and do not break unexpectedly.
The default value is 0. This value will change to 1 (set) in the next major version.

3.4.24 output. There are three possible values: 0 (= raw ), 1 (= prettymatrix ), or 3 (= external
prettyprint). This means that, independently of the default format for reals which we explained
above, you can print results in three ways:
• raw format, i.e. a format which is equivalent to what you input, including explicit multiplica-
tion signs, and everything typed on a line instead of two dimensional boxes. This can have several
advantages, for instance it allows you to pick the result with a mouse or an editor, and to paste it
somewhere else.
• prettymatrix format: this is identical to raw format, except that matrices are printed as
boxes instead of horizontally. This is prettier, but takes more space and cannot be used for input.
Column vectors are still printed horizontally.
• external prettyprint: pipes all gp output in TeX format to an external prettyprinter, according
to the value of prettyprinter. The default script (tex2mail) converts its input to readable two-
dimensional text.
Independently of the setting of this default, an object can be printed in any of the three formats
at any time using the commands \a and \m and \B respectively.
The default value is 1 (prettymatrix ).

125
3.4.25 parisize. gp, and in fact any program using the PARI library, needs a stack in which to do
its computations; parisize is the stack size, in bytes. It is recommended to increase this default
using a gprc, to the value you believe PARI should be happy with, given your typical computation.
We strongly recommend to also set parisizemax to a much larger value in your gprc, about what
you believe your machine can stand: PARI will then try to fit its computations within about
parisize bytes, but will increase the stack size if needed (up to parisizemax). Once the memory
intensive computation is over, PARI will restore the stack size to the originally requested parisize.
The default value is 4M, resp. 8M on a 32-bit, resp. 64-bit machine.

3.4.26 parisizemax. gp, and in fact any program using the PARI library, needs a stack in which
to do its computations. If nonzero, parisizemax is the maximum size the stack can grow to, in
bytes. If zero, the stack will not automatically grow, and will be limited to the value of parisize.
When parisizemax is set, PARI tries to fit its computations within about parisize bytes, but
will increase the stack size if needed, roughly doubling it each time (up to parisizemax of course!)
and printing a message such as Warning: increasing stack size to some value. Once the
memory intensive computation is over, PARI will restore the stack size to the originally requested
parisize without printing further messages.
We strongly recommend to set parisizemax permanently to a large nonzero value in your gprc,
about what you believe your machine can stand. It is possible to increase or decrease parisizemax
inside a running gp session, just use default as usual.
The default value is 0, for backward compatibility reasons.

3.4.27 path. This is a list of directories, separated by colons ’:’ (semicolons ’;’ in the DOS world,
since colons are preempted for drive names). When asked to read a file whose name is not given
by an absolute path (does not start with /, ./ or ../), gp will look for it in these directories, in
the order they were written in path. Here, as usual, . means the current directory, and .. its
immediate parent. Environment expansion is performed.
The default value is ".:~:~/gp" on UNIX systems, ".;C:\;C:\GP" on DOS, OS/2 and Win-
dows, and "." otherwise.

3.4.28 plothsizes. If the graphic driver allows it, the array contains the size of the terminal, the
size of the font, the size of the ticks.

3.4.29 prettyprinter. The name of an external prettyprinter to use when output is 3 (alternate
prettyprinter). Note that the default tex2mail looks much nicer than the built-in “beautified
format” (output = 2).
The default value is "tex2mail -TeX -noindent -ragged -by par".

3.4.30 primelimit. gp precomputes a list of all primes less than primelimit at initialization time,
and can build fast sieves on demand to quickly iterate over primes up to the square of primelimit.
These are used by many arithmetic functions, usually for trial division purposes. The maximal
value is 232 − 2049 (resp 264 − 2049) on a 32-bit (resp. 64-bit) machine, but values beyond 108 ,
allowing to iterate over primes up to 1016 , do not seem useful.
Since almost all arithmetic functions eventually require some table of prime numbers, PARI
guarantees that the first 6547 primes, up to and including 65557, are precomputed, even if prime-
limit is 1.
This default is only used on startup: changing it will not recompute a new table.

126
 Deprecated feature. primelimit was used in some situations by algebraic number theory func-
tions using the nf_PARTIALFACT flag (nfbasis, nfdisc, nfinit, . . . ): this assumes that all primes
p > primelimit have a certain property (the equation order is p-maximal). This is never done by
default, and must be explicitly set by the user of such functions. Nevertheless, these functions now
provide a more flexible interface, and their use of the global default primelimit is deprecated.

Deprecated feature. factor(N, 0) was used to partially factor integers by removing all prime
factors ≤ primelimit. Don’t use this, supply an explicit bound: factor(N, bound), which avoids
relying on an unpredictable global variable.

The default value is 500k.

3.4.31 prompt. A string that will be printed as prompt. Note that most usual escape sequences
are available there: \e for Esc, \n for Newline, . . . , \\ for \. Time expansion is performed.

This string is sent through the library function strftime (on a Unix system, you can try man
strftime at your shell prompt). This means that % constructs have a special meaning, usually
related to the time and date. For instance, %H = hour (24-hour clock) and %M = minute [00,59] (use
%% to get a real %).

If you use readline, escape sequences in your prompt will result in display bugs. If you have
a relatively recent readline (see the comment at the end of Section 3.4.3), you can brace them
with special sequences (\[ and \]), and you will be safe. If these just result in extra spaces in
your prompt, then you’ll have to get a more recent readline. See the file misc/gprc.dft for an
example.

EMACS: Caution: PariEmacs needs to know about the prompt pattern to separate your input from previous
gp results, without ambiguity. It is not a trivial problem to adapt automatically this regular
expression to an arbitrary prompt (which can be self-modifying!). See PariEmacs’s documentation.

The default value is "? ".

3.4.32 prompt cont. A string that will be printed to prompt for continuation lines (e.g. in
between braces, or after a line-terminating backslash). Everything that applies to prompt applies
to prompt cont as well.

The default value is "".

3.4.33 psfile. This default is obsolete, use one of plotexport, plothexport or plothrawexport
functions and write the result to file.

3.4.34 readline. Switches readline line-editing facilities on and off. This may be useful if you
are running gp in a Sun cmdtool, which interacts badly with readline. Of course, until readline is
switched on again, advanced editing features like automatic completion and editing history are not
available.

The default value is 1.

127
3.4.35 realbitprecision. The number of significant bits used to convert exact inputs given to
transcendental functions (see Section 3.11), or to create absolute floating point constants (input
as 1.0 or Pi for instance). Unless you tamper with the format default, this is also the number of
significant bits used to print a t_REAL number; format will override this latter behavior, and allow
you to have a large internal precision while outputting few digits for instance.
Note that most PARI’s functions currently handle precision on a word basis (by increments
of 32 or 64 bits), hence bit precision may be a little larger than the number of bits you expected.
For instance to get 10 bits of precision, you need one word of precision which, on a 64-bit machine,
correspond to 64 bits. To make things even more confusing, this internal bit accuracy is converted
to decimal digits when printing floating point numbers: now 64 bits correspond to 19 printed
decimal digits (19 < log10 (264 ) < 20).
The value returned when typing default(realbitprecision) is the internal number of sig-
nificant bits, not the number of printed decimal digits:
? default(realbitprecision, 10)
? \pb
realbitprecision = 64 significant bits
? default(realbitprecision)
%1 = 64
? \p
realprecision = 3 significant digits
? default(realprecision)
%2 = 19
Note that realprecision and \p allow to view and manipulate the internal precision in decimal
digits.
The default value is 128, resp. 96, on a 64-bit, resp .32-bit, machine.

3.4.36 realprecision. The number of significant digits used to convert exact inputs given to
transcendental functions (see Section 3.11), or to create absolute floating point constants (input
as 1.0 or Pi for instance). Unless you tamper with the format default, this is also the number
of significant digits used to print a t_REAL number; format will override this latter behavior, and
allow you to have a large internal precision while outputting few digits for instance.
Note that PARI’s internal precision works on a word basis (by increments of 32 or 64 bits),
hence may be a little larger than the number of decimal digits you expected. For instance to get
2 decimal digits you need one word of precision which, on a 64-bit machine, actually gives you 19
digits (19 < log10 (264 ) < 20). The value returned when typing default(realprecision) is the
internal number of significant digits, not the number of printed digits:
? default(realprecision, 2)
realprecision = 19 significant digits (2 digits displayed)
? default(realprecision)
%1 = 19
The default value is 38, resp. 28, on a 64-bit, resp. 32-bit, machine.

3.4.37 recover. This toggle is either 1 (on) or 0 (off). If you change this to 0, any error becomes
fatal and causes the gp interpreter to exit immediately. Can be useful in batch job scripts.
The default value is 1.

128
3.4.38 secure. This toggle is either 1 (on) or 0 (off). If on, the system and extern command are
disabled. These two commands are potentially dangerous when you execute foreign scripts since
they let gp execute arbitrary UNIX commands. gp will ask for confirmation before letting you (or
a script) unset this toggle.
The default value is 0.

3.4.39 seriesprecision. Number of significant terms when converting a polynomial or rational

function to a power series (see \ps).
The default value is 16.

3.4.40 simplify. This toggle is either 1 (on) or 0 (off). When the PARI library computes some-
thing, the type of the result is not always the simplest possible. The only type conversions which the
PARI library does automatically are rational numbers to integers (when they are of type t_FRAC
and equal to integers), and similarly rational functions to polynomials (when they are of type
t_RFRAC and equal to polynomials). This feature is useful in many cases, and saves time, but can
be annoying at times. Hence you can disable this and, whenever you feel like it, use the func-
tion simplify (see Chapter 3) which allows you to simplify objects to the simplest possible types
recursively (see \y).
The default value is 1.

3.4.41 sopath. This is a list of directories, separated by colons ’:’ (semicolons ’;’ in the DOS
world, since colons are preempted for drive names). When asked to install an external symbol
from a shared library whose name is not given by an absolute path (does not start with /, ./ or
../), gp will look for it in these directories, in the order they were written in sopath. Here, as
usual, . means the current directory, and .. its immediate parent. Environment expansion is
performed.
The default value is "", corresponding to an empty list of directories: install will use the
library name as input (and look in the current directory if the name is not an absolute path).

3.4.42 strictargs. This toggle is either 1 (on) or 0 (off). If on, all arguments to new user functions
are mandatory unless the function supplies an explicit default value. Otherwise arguments have
the default value 0.
In this example,
fun(a,b=2)=a+b
a is mandatory, while b is optional. If strictargs is on:
? fun()
*** at top-level: fun()
*** ^-----
*** in function fun: a,b=2
*** ^-----
*** missing mandatory argument ’a’ in user function.
This applies to functions defined while strictargs is on. Changing strictargs does not
affect the behavior of previously defined functions.
The default value is 0.

129
3.4.43 strictmatch. Obsolete. This toggle is now a no-op.

3.4.44 threadsize. This default is specific to the parallel version of PARI and gp (built via Con-
figure --mt=prthread or mpi) and is ignored otherwise. In parallel mode, each thread allocates its
own private stack for its computations, see parisize. This value determines the size in bytes of the
stacks of each thread, so the total memory allocated will be parisize + nbthreads × threadsize.
If set to 0, the value used is the same as parisize. It is not easy to estimate reliably a sufficient
value for this parameter because PARI itself will parallelize computations and we recommend to
not set this value explicitly unless it solves a specific problem for you. For instance if you see
frequent messages of the form

*** Warning: not enough memory, new thread stack 10000002048

(Meaning that threadsize had to be temporarily increased.) On the other hand we strongly
recommend to set parisizemax and threadsizemax to a nonzero value.
The default value is 0.

3.4.45 threadsizemax. This default is specific to the parallel version of PARI and gp (built
via Configure --mt=pthread or mpi) and is ignored otherwise. In parallel mode, each threads
allocates its own private stack for its computations, see parisize and parisizemax. The values
of threadsize and threadsizemax determine the usual and maximal size in bytes of the stacks of
each thread, so the total memory allocated will be between parisize + nbthreads × threadsize.
and parisizemax + nbthreads × threadsizemax.
If set to 0, the value used is the same as threadsize. We strongy recommend to set both
parisizemax and threadsizemax to a nonzero value.
The default value is 0.

3.4.46 timer. This toggle is either 1 (on) or 0 (off). Every instruction sequence in the gp calculator
(anything ended by a newline in your input) is timed, to some accuracy depending on the hardware
and operating system. When timer is on, each such timing is printed immediately before the
output as follows:

? factor(2^2^7+1)
time = 108 ms. \\ this line omitted if ’timer’ is 0
%1 =
[ 59649589127497217 1]
[5704689200685129054721 1]
(See also # and ##.)
The time measured is the user CPU time, not including the time for printing the results. If
the time is negligible (< 1 ms.), nothing is printed: in particular, no timing should be printed when
defining a user function or an alias, or installing a symbol from the library.
The default value is 0 (off).

130
3.5 Standard monadic or dyadic operators.
3.5.1 Boolean operators.
Any nonzero value is interpreted as true and any zero as false (this includes empty vectors
or matrices). The standard boolean operators || (inclusive or), && (and) and ! in prefix notation
(not) are available. Their value is 1 (true) or 0 (false):
? a && b \\ 1 iff a and b are nonzero
? a || b \\ 1 iff a or b is nonzero
? !a \\ 1 iff a is zero

3.5.2 Comparison. The standard real comparison operators <=, <, >=, >, are available in GP.
The result is 1 if the comparison is true, 0 if it is false. These operators allow to compare integers
(t_INT), rational (t_FRAC) or real (t_REAL) numbers, real quadratic numbers (t_QUAD of positive
discriminant) and infinity (oo, t_INFINITY).
By extension, two character strings (t_STR) are compared using the standard lexicographic
order. Comparing a string to an object of a different type raises an exception. See also the cmp
universal comparison function.

3.5.3 Equality. Two operators allow to test for equality: == (equality up to type coercion) and
=== (identity). The result is 1 if equality is decided, else 0.
The operator === is strict: objects of different type or length are never identical, polynomials
in different variables are never identical, even if constant. On the contrary, == is very liberal:
a == b decides whether there is a natural map sending a to the domain of b or sending b to the
domain of a, such that the comparison makes sense and equality holds. For instance
? 4 == Mod(1,3) \\ equal
%1 = 1
? 4 === Mod(1,3) \\ but not identical
%2 = 0
? ’x == ’y \\ not equal (nonconstant and different variables)
%3 = 0
? Pol(0,’x) == Pol(0,’y) \\ equal (constant: ignore variable)
%4 = 1
? Pol(0,’x) == Pol(0,’y) \\ not identical
%5 = 0
? 0 == Pol(0) \\ equal
%6 = 1
? [0] == 0 \\ equal
%7 = 1
? [0, 0] == 0 \\ equal
%8 = 1
? [0] == [0,0] \\ not equal
%9 = 1
In particular == is not transitive in general; it is transitive when used to compare objects known to
have the same type. The operator === is transitive. The == operator allows two equivalent negated
forms: != or <>; there is no negated form for ===.
Do not mistake = for ==: it is the assignment statement.

131
3.5.4 +/-. The expressions +x and -x refer to monadic operators: the first does nothing, the
second negates x.
The library syntax is GEN gneg(GEN x) for -x.

3.5.5 +. The expression x + y is the sum of x and y. Addition between a scalar type x and a t_COL
or t_MAT y returns respectively [y[1] + x, y[2], . . .] and y + xId. Other additions between a scalar
type and a vector or a matrix, or between vector/matrices of incompatible sizes are forbidden.
The library syntax is GEN gadd(GEN x, GEN y).

3.5.6 -. The expression x - y is the difference of x and y. Subtraction between a scalar type x
and a t_COL or t_MAT y returns respectively [y[1] − x, y[2], . . .] and y − xId. Other subtractions
between a scalar type and a vector or a matrix, or between vector/matrices of incompatible sizes
are forbidden.
The library syntax is GEN gsub(GEN x, GEN y) for x - y.

3.5.7 *. The expression x * y is the product of x and y. Among the prominent impossibilities are
multiplication between vector/matrices of incompatible sizes, between a t_INTMOD or t_PADIC Re-
stricted to scalars, * is commutative; because of vector and matrix operations, it is not commutative
in general.
Multiplication between two t_VECs or two t_COLs is not allowed; to take the scalar product of
two vectors of the same length, transpose one of the vectors (using the operator ~ or the function
mattranspose, see Section 3.10) and multiply a line vector by a column vector:
? a = [1,2,3];
? a * a
*** at top-level: a*a
*** ^--
*** _*_: forbidden multiplication t_VEC * t_VEC.
? a * a~
%2 = 14
If x, y are binary quadratic forms, compose them; see also qfbnucomp and qfbnupow. If x, y
are t_VECSMALL of the same length, understand them as permutations and compose them.
The library syntax is GEN gmul(GEN x, GEN y) for x * y. Also available is GEN gsqr(GEN x)
for x * x.

3.5.8 /. The expression x / y is the quotient of x and y. In addition to the impossibilities for
multiplication, note that if the divisor is a matrix, it must be an invertible square matrix, and in
that case the result is x∗y −1 . Furthermore note that the result is as exact as possible: in particular,
division of two integers always gives a rational number (which may be an integer if the quotient
is exact) and not the Euclidean quotient (see x \ y for that), and similarly the quotient of two
polynomials is a rational function in general. To obtain the approximate real value of the quotient
of two integers, add 0. to the result; to obtain the approximate p-adic value of the quotient of two
integers, add O(p^k) to the result; finally, to obtain the Taylor series expansion of the quotient of
two polynomials, add O(X^k) to the result or use the taylor function (see Section 3.9.57).
The library syntax is GEN gdiv(GEN x, GEN y) for x / y.

132
3.5.9 \. The expression x \ y is the Euclidean quotient of x and y. If y is a real scalar, this is
defined as floor(x/y) if y > 0, and ceil(x/y) if y < 0 and the division is not exact. Hence the
remainder x - (x\y)*y is in [0, |y|[.

Note that when y is an integer and x a polynomial, y is first promoted to a polynomial of

degree 0. When x is a vector or matrix, the operator is applied componentwise.

The library syntax is GEN gdivent(GEN x, GEN y) for x \ y.

3.5.10 \/. The expression x \/ y evaluates to the rounded Euclidean quotient of x and y. This is
the same as x \ y except for scalar division: the quotient is such that the corresponding remainder
is smallest in absolute value and in case of a tie the quotient closest to +∞ is chosen (hence the
remainder would belong to [−|y|/2, |y|/2[).

When x is a vector or matrix, the operator is applied componentwise.

The library syntax is GEN gdivround(GEN x, GEN y) for x \/ y.

3.5.11 %. The expression x % y evaluates to the modular Euclidean remainder of x and y, which we
now define. When x or y is a nonintegral real number, x%y is defined as x - (x\y)*y. Otherwise,
if y is an integer, this is the smallest nonnegative integer congruent to x modulo y. (This actually
coincides with the previous definition if and only if x is an integer.) If y is a polynomial, this is the
polynomial of smallest degree congruent to x modulo y. For instance:

? (1/2) % 3
%1 = 2
? 0.5 % 3
%2 = 0.5000000000000000000000000000
? (1/2) % 3.0
%3 = 1/2

Note that when y is an integer and x a polynomial, y is first promoted to a polynomial of

degree 0. When x is a vector or matrix, the operator is applied componentwise.

The library syntax is GEN gmod(GEN x, GEN y) for x % y.

3.5.12 op=. When op is a binary arithmetic operator among +, -, %, /, \ or \/, the construct
xop = y is a shortcut for x = xopy.

? v[1] += 10 \\ increment v[1] by 10

? a /= 2 \\ divide a by 2

3.5.13 ++. x++ is a shortcut for x = x + 1.

3.5.14 --. x-- is a shortcut for x = x - 1.

133
3.5.15 ^. The expression x^n is powering.
• If the exponent n is an integer, then exact operations are performed using binary (left-shift)
powering techniques. By definition, x0 is (an empty product interpreted as) an exact 1 in the
underlying prime ring:
? 0.0 ^ 0
%1 = 1
? (1 + O(2^3)) ^ 0
%2 = 1
? (1 + O(x)) ^ 0
%3 = 1
? Mod(2,4)^0
%4 = Mod(1,4)
? Mod(x,x^2)^0
%5 = Mod(1, x^2)
If x is a p-adic number, its precision will increase if vp (n) > 0 and n 6= 0. Powering a binary
quadratic form (types t_QFI and t_QFR) returns a representative of the class, which is reduced if
the input was. (In particular, x ^1 returns x itself, whether it is reduced or not.)
PARI rewrites the multiplication x ∗ x of two identical objects as x2 . Here, identical means
the operands are reference the same chunk of memory; no equality test is performed. This is no
longer true when more than two arguments are involved.
? a = 1 + O(2); b = a;
? a * a \\ = a^2, precision increases
%2 = 1 + O(2^3)
? a * b \\ not rewritten as a^2
%3 = 1 + O(2)
? a*a*a \\ not rewritten as a^3
%4 = 1 + O(2)
• If the exponent is a rational number p/q the behaviour depends on x. If x is a complex
number, return exp(n log x) (principal branch), in an exact form if possible:
? 4^(1/2) \\ 4 being a square, this is exact
%1 = 2
? 2^(1/2) \\ now inexact
%2 = 1.4142135623730950488016887242096980786
? (-1/4)^(1/2) \\ exact again
%3 = 1/2*I
? (-1)^(1/3)
%4 = 0.500...+ 0.866...*I
Note that even though −1 is an exact cube root of −1, it is not exp(log(−1)/3); the latter is
returned.
Otherwise return a solution y of y q = xp if it exists; beware that this is defined up to q-th
roots of 1 in the base field. Intmods modulo composite numbers are not supported.
? Mod(7,19)^(1/2)
%1 = Mod(11, 19) \\ is any square root
? sqrt(Mod(7,19))

134
 %2 = Mod(8, 19) \\ is the smallest square root
? Mod(1,4)^(1/2)
*** at top-level: Mod(1,4)^(1/2)
*** ^------
*** _^_: not a prime number in gpow: 4.
• If the exponent is a negative integer or rational number, an inverse must be computed. For
noninvertible t_INTMOD x, this will fail and (for n an integer) implicitly exhibit a factor of the
modulus:
? Mod(4,6)^(-1)
*** at top-level: Mod(4,6)^(-1)
*** ^-----
*** _^_: impossible inverse modulo: Mod(2, 6).
Here, a factor 2 is obtained directly. In general, take the gcd of the representative and the modulus.
This is most useful when performing complicated operations modulo an integer N whose factoriza-
tion is unknown. Either the computation succeeds and all is well, or a factor d is discovered and
the computation may be restarted modulo d or N/d.
For noninvertible t_POLMOD x, the behavior is the same:
? Mod(x^2, x^3-x)^(-1)
*** at top-level: Mod(x^2,x^3-x)^(-1)
*** ^-----
*** _^_: impossible inverse in RgXQ_inv: Mod(x^2, x^3 - x).
Note that the underlying algorihm (subresultant) assumes that the base ring is a domain:
? a = Mod(3*y^3+1, 4); b = y^6+y^5+y^4+y^3+y^2+y+1; c = Mod(a,b);
? c^(-1)
*** at top-level: Mod(a,b)^(-1)
*** ^-----
*** _^_: impossible inverse modulo: Mod(2, 4).
In fact c is invertible, but Z/4Z is not a domain and the algorithm fails. It is possible for the
algorithm to succeed in such situations and any returned result will be correct, but chances are
that an error will occur first. In this specific case, one should work with 2-adics. In general, one
can also try the following approach
? inversemod(a, b) =
{ my(m, v = variable(b));
m = polsylvestermatrix(polrecip(a), polrecip(b));
m = matinverseimage(m, matid(#m)[,1]);
Polrev(m[1..poldegree(b)], v);
}
? inversemod(a,b)
%2 = Mod(2,4)*y^5 + Mod(3,4)*y^3 + Mod(1,4)*y^2 + Mod(3,4)*y + Mod(2,4)
This is not guaranteed to work either since matinverseimage must also invert pivots. See Sec-
tion 3.10.
For a t_MAT x, the matrix is expected to be square and invertible, except in the special case
x^(-1) which returns a left inverse if one exists (rectangular x with full column rank).

135
 ? x = Mat([1;2])
%1 =
[1]
[2]
? x^(-1)
%2 =
[1 0]
• Finally, if the exponent n is not an rational number, powering is treated as the transcendental
function exp(n log x), although it will be more precise than the latter when n and x are exact:
? s = 1/2 + 10^14 * I
? localprec(200); z = 2^s \\ for reference
? exponent(2^s - z)
%3 = -127 \\ perfect
? exponent(exp(s * log(2)) - z)
%4 = -84 \\ not so good
The second computation is less precise because log(2) is first computed to 38 decimal digits, then
multiplied by s, which has a huge imaginary part amplifying the error.
In this case, x 7→ xn is treated as a transcendental function and and in particular acts com-
ponentwise on vector or matrices, even square matrices ! (See Section 3.11.) If x is 0 and n is an
inexact 0, this will raise an exception:
? 4 ^ 1.0
%1 = 4.0000000000000000000000000000000000000
? 0^ 0.0
*** at top-level: 0^0.0
*** ^----
*** _^_: domain error in gpow(0,n): n <= 0
The library syntax is GEN gpow(GEN x, GEN n, long prec) for x^n.

3.5.16 cmp(x, y). Gives the result of a comparison between arbitrary objects x and y (as −1, 0
or 1). The underlying order relation is transitive, the function returns 0 if and only if x === y. It
has no mathematical meaning but satisfies the following properties when comparing entries of the
same type:
• two t_INTs compare as usual (i.e. cmp(x, y) < 0 if and only if x < y);
• two t_VECSMALLs of the same length compare lexicographically;
• two t_STRs compare lexicographically.
In case all components are equal up to the smallest length of the operands, the more complex
is considered to be larger. More precisely, the longest is the largest; when lengths are equal, we
have matrix > vector > scalar. For example:
? cmp(1, 2)
%1 = -1
? cmp(2, 1)
%2 = 1
? cmp(1, 1.0) \\ note that 1 == 1.0, but (1===1.0) is false.

136
 %3 = -1
? cmp(x + Pi, [])
%4 = -1
This function is mostly useful to handle sorted lists or vectors of arbitrary objects. For instance, if
v is a vector, the construction vecsort(v, cmp) is equivalent to Set(v).
The library syntax is GEN cmp_universal(GEN x, GEN y).

3.5.17 divrem(x, y, {v}). Creates a column vector with two components, the first being the
Euclidean quotient (x \ y), the second the Euclidean remainder (x - (x\y)*y), of the division of
x by y. This avoids the need to do two divisions if one needs both the quotient and the remainder.
If v is present, and x, y are multivariate polynomials, divide with respect to the variable v.
Beware that divrem(x,y)[2] is in general not the same as x % y; no GP operator corresponds
to it:
? divrem(1/2, 3)[2]
%1 = 1/2
? (1/2) % 3
%2 = 2
? divrem(Mod(2,9), 3)[2]
*** at top-level: divrem(Mod(2,9),3)[2
*** ^--------------------
*** forbidden division t_INTMOD \ t_INT.
? Mod(2,9) % 6
%3 = Mod(2,3)
The library syntax is GEN divrem(GEN x, GEN y, long v = -1) where v is a variable number.
Also available is GEN gdiventres(GEN x, GEN y) when v is not needed.

3.5.18 lex(x, y). Gives the result of a lexicographic comparison between x and y (as −1, 0 or 1).
This is to be interpreted in quite a wide sense: it is admissible to compare objects of different types
(scalars, vectors, matrices), provided the scalars can be compared, as well as vectors/matrices of
different lengths; finally, when comparing two scalars, a complex number a + I ∗ b is interpreted as
a vector [a, b] and a real number a as [a, 0]. The comparison is recursive.
In case all components are equal up to the smallest length of the operands, the more complex
is considered to be larger. More precisely, the longest is the largest; when lengths are equal, we
have matrix > vector > scalar. For example:
? lex([1,3], [1,2,5])
%1 = 1
? lex([1,3], [1,3,-1])
%2 = -1
? lex([1], [[1]])
%3 = -1
? lex([1], [1]~)
%4 = 0
? lex(2 - I, 1)
%5 = 1
? lex(2 - I, 2)

137
 %6 = 2

The library syntax is GEN lexcmp(GEN x, GEN y).

3.5.19 max(x, y). Creates the maximum of x and y when they can be compared.

The library syntax is GEN gmax(GEN x, GEN y).

3.5.20 min(x, y). Creates the minimum of x and y when they can be compared.

The library syntax is GEN gmin(GEN x, GEN y).

3.5.21 shift(x, n). Shifts x componentwise left by n bits if n ≥ 0 and right by |n| bits if n < 0.
May be abbreviated as x << n or x >> (−n). A left shift by n corresponds to multiplication by 2n .
A right shift of an integer x by |n| corresponds to a Euclidean division of x by 2|n| with a remainder
of the same sign as x, hence is not the same (in general) as x\2n .

The library syntax is GEN gshift(GEN x, long n).

3.5.22 shiftmul(x, n). Multiplies x by 2n . The difference with shift is that when n < 0, ordinary
division takes place, hence for example if x is an integer the result may be a fraction, while for
shifts Euclidean division takes place when n < 0 hence if x is an integer the result is still an integer.

The library syntax is GEN gmul2n(GEN x, long n).

3.5.23 sign(x). sign (0, 1 or −1) of x, which must be of type integer, real or fraction; t_QUAD with
positive discriminants and t_INFINITY are also supported.

The library syntax is GEN gsigne(GEN x).

3.5.24 vecmax(x, {&v}). If x is a vector or a matrix, returns the largest entry of x, otherwise
returns a copy of x. Error if x is empty.

If v is given, set it to the index of a largest entry (indirect maximum), when x is a vector. If
x is a matrix, set v to coordinates [i, j] such that x[i, j] is a largest entry. This flag is ignored if x
is not a vector or matrix.

? vecmax([10, 20, -30, 40])

%1 = 40
? vecmax([10, 20, -30, 40], &v); v
%2 = 4
? vecmax([10, 20; -30, 40], &v); v
%3 = [2, 2]

The library syntax is GEN vecmax0(GEN x, GEN *v = NULL). When v is not needed, the
function GEN vecmax(GEN x) is also available.

138
3.5.25 vecmin(x, {&v}). If x is a vector or a matrix, returns the smallest entry of x, otherwise
returns a copy of x. Error if x is empty.
If v is given, set it to the index of a smallest entry (indirect minimum), when x is a vector. If
x is a matrix, set v to coordinates [i, j] such that x[i, j] is a smallest entry. This is ignored if x is
not a vector or matrix.
? vecmin([10, 20, -30, 40])
%1 = -30
? vecmin([10, 20, -30, 40], &v); v
%2 = 3
? vecmin([10, 20; -30, 40], &v); v
%3 = [2, 1]
The library syntax is GEN vecmin0(GEN x, GEN *v = NULL). When v is not needed, the
function GEN vecmin(GEN x) is also available.

3.6 Conversions and similar elementary functions or commands.

Many of the conversion functions are rounding or truncating operations. In this case, if the argu-
ment is a rational function, the result is the Euclidean quotient of the numerator by the denomi-
nator, and if the argument is a vector or a matrix, the operation is done componentwise. This will
not be restated for every function.

3.6.1 Col(x, {n}). Transforms the object x into a column vector. The dimension of the resulting
vector can be optionally specified via the extra parameter n.
If n is omitted or 0, the dimension depends on the type of x; the vector has a single component,
except when x is
• a vector or a quadratic form (in which case the resulting vector is simply the initial object
considered as a row vector),
• a polynomial or a power series. In the case of a polynomial, the coefficients of the vector start
with the leading coefficient of the polynomial, while for power series only the significant coefficients
are taken into account, but this time by increasing order of degree. In this last case, Vec is the
reciprocal function of Pol and Ser respectively,
• a matrix (the column of row vector comprising the matrix is returned),
• a character string (a vector of individual characters is returned).
In the last two cases (matrix and character string), n is meaningless and must be omitted or
an error is raised. Otherwise, if n is given, 0 entries are appended at the end of the vector if n > 0,
and prepended at the beginning if n < 0. The dimension of the resulting vector is |n|.
See ??Vec for examples.
The library syntax is GEN gtocol0(GEN x, long n). GEN gtocol(GEN x) is also available.

3.6.2 Colrev(x, {n}). As Col(x, −n), then reverse the result. In particular, Colrev is the reciprocal
function of Polrev: the coefficients of the vector start with the constant coefficient of the polynomial
and the others follow by increasing degree.
The library syntax is GEN gtocolrev0(GEN x, long n). GEN gtocolrev(GEN x) is also
available.

139
3.6.3 List({x = [ ]}). Transforms a (row or column) vector x into a list, whose components are
the entries of x. Similarly for a list, but rather useless in this case. For other types, creates a list
with the single element x.
The library syntax is GEN gtolist(GEN x = NULL). The variant GEN mklist(void) creates
an empty list.

3.6.4 Map({x}). A “Map” is an associative array, or dictionary: a data type composed of a

collection of (key, value) pairs, such that each key appears just once in the collection. This function
converts the matrix [a1 , b1 ; a2 , b2 ; . . . ; an , bn ] to the map ai 7→ bi .
? M = Map(factor(13!));
? mapget(M,3)
%2 = 5
If the argument x is omitted, creates an empty map, which may be filled later via mapput.
The library syntax is GEN gtomap(GEN x = NULL).

3.6.5 Mat({x = [ ]}). Transforms the object x into a matrix. If x is already a matrix, a copy of
x is created. If x is a row (resp. column) vector, this creates a 1-row (resp. 1-column) matrix,
unless all elements are column (resp. row) vectors of the same length, in which case the vectors are
concatenated sideways and the attached big matrix is returned. If x is a binary quadratic form,
creates the attached 2 × 2 matrix. Otherwise, this creates a 1 × 1 matrix containing x.
? Mat(x + 1)
%1 =
[x + 1]
? Vec( matid(3) )
%2 = [[1, 0, 0]~, [0, 1, 0]~, [0, 0, 1]~]
? Mat(%)
%3 =
[1 0 0]
[0 1 0]
[0 0 1]
? Col( [1,2; 3,4] )
%4 = [[1, 2], [3, 4]]~
? Mat(%)
%5 =
[1 2]
[3 4]
? Mat(Qfb(1,2,3))
%6 =
[1 1]
[1 3]
The library syntax is GEN gtomat(GEN x = NULL).

140
3.6.6 Mod(a, b). In its basic form, create an intmod or a polmod (a mod b); b must be an integer
or a polynomial. We then obtain a t_INTMOD and a t_POLMOD respectively:

? t = Mod(2,17); t^8
%1 = Mod(1, 17)
? t = Mod(x,x^2+1); t^2
%2 = Mod(-1, x^2+1)

If a%b makes sense and yields a result of the appropriate type (t_INT or scalar/t_POL), the operation
succeeds as well:

? Mod(1/2, 5)
%3 = Mod(3, 5)
? Mod(7 + O(3^6), 3)
%4 = Mod(1, 3)
? Mod(Mod(1,12), 9)
%5 = Mod(1, 3)
? Mod(1/x, x^2+1)
%6 = Mod(-x, x^2+1)
? Mod(exp(x), x^4)
%7 = Mod(1/6*x^3 + 1/2*x^2 + x + 1, x^4)

If a is a complex object, “base change” it to Z/bZ or K[x]/(b), which is equivalent to, but
faster than, multiplying it by Mod(1,b):

? Mod([1,2;3,4], 2)
%8 =
[Mod(1, 2) Mod(0, 2)]
[Mod(1, 2) Mod(0, 2)]
? Mod(3*x+5, 2)
%9 = Mod(1, 2)*x + Mod(1, 2)
? Mod(x^2 + y*x + y^3, y^2+1)
%10 = Mod(1, y^2 + 1)*x^2 + Mod(y, y^2 + 1)*x + Mod(-y, y^2 + 1)

This function is not the same as x % y, the result of which has no knowledge of the intended
modulus y. Compare

? x = 4 % 5; x + 1
%11 = 5
? x = Mod(4,5); x + 1
%12 = Mod(0,5)

Note that such “modular” objects can be lifted via lift or centerlift. The modulus of a
t_INTMOD or t_POLMOD z can be recovered via z.mod.

The library syntax is GEN gmodulo(GEN a, GEN b).

141
3.6.7 Pol(t, {v =0 x}). Transforms the object t into a polynomial with main variable v. If t is
a scalar, this gives a constant polynomial. If t is a power series with nonnegative valuation or
a rational function, the effect is similar to truncate, i.e. we chop off the O(X k ) or compute the
Euclidean quotient of the numerator by the denominator, then change the main variable of the
result to v.
The main use of this function is when t is a vector: it creates the polynomial whose coefficients
are given by t, with t[1] being the leading coefficient (which can be zero). It is much faster to
evaluate Pol on a vector of coefficients in this way, than the corresponding formal expression
an X n + . . . + a0 , which is evaluated naively exactly as written (linear versus quadratic time in n).
Polrev can be used if one wants x[1] to be the constant coefficient:
? Pol([1,2,3])
%1 = x^2 + 2*x + 3
? Polrev([1,2,3])
%2 = 3*x^2 + 2*x + 1
The reciprocal function of Pol (resp. Polrev) is Vec (resp. Vecrev).
? Vec(Pol([1,2,3]))
%1 = [1, 2, 3]
? Vecrev( Polrev([1,2,3]) )
%2 = [1, 2, 3]

Warning. This is not a substitution function. It will not transform an object containing variables
of higher priority than v.
? Pol(x + y, y)
*** at top-level: Pol(x+y,y)
*** ^----------
*** Pol: variable must have higher priority in gtopoly.
The library syntax is GEN gtopoly(GEN t, long v = -1) where v is a variable number.

3.6.8 Polrev(t, {v =0 x}). Transform the object t into a polynomial with main variable v. If t is a
scalar, this gives a constant polynomial. If t is a power series, the effect is identical to truncate,
i.e. it chops off the O(X k ).
The main use of this function is when t is a vector: it creates the polynomial whose coefficients
are given by t, with t[1] being the constant term. Pol can be used if one wants t[1] to be the leading
coefficient:
? Polrev([1,2,3])
%1 = 3*x^2 + 2*x + 1
? Pol([1,2,3])
%2 = x^2 + 2*x + 3
The reciprocal function of Pol (resp. Polrev) is Vec (resp. Vecrev).
The library syntax is GEN gtopolyrev(GEN t, long v = -1) where v is a variable number.

142
3.6.9 Qfb(a, b, c, {D = 0.}). Creates the binary quadratic form ax2 + bxy + cy 2 . If b2 − 4ac > 0,
initialize Shanks’ distance function to D. Negative definite forms are not implemented, use their
positive definite counterpart instead.
The library syntax is GEN Qfb0(GEN a, GEN b, GEN c, GEN D = NULL, long prec)
. Also available are GEN qfi(GEN a, GEN b, GEN c) (assumes b2 − 4ac < 0) and GEN qfr(GEN a,
GEN b, GEN c, GEN D) (assumes b2 − 4ac > 0).

3.6.10 Ser(s, {v =0 x}, {d = seriesprecision}). Transforms the object s into a power series with
main variable v (x by default) and precision (number of significant terms) equal to d ≥ 0 (d =
seriesprecision by default). If s is a scalar, this gives a constant power series in v with precision
d. If s is a polynomial, the polynomial is truncated to d terms if needed
? \ps
seriesprecision = 16 significant terms
? Ser(1) \\ 16 terms by default
%1 = 1 + O(x^16)
? Ser(1, ’y, 5)
%2 = 1 + O(y^5)
? Ser(x^2,, 5)
%3 = x^2 + O(x^7)
? T = polcyclo(100)
%4 = x^40 - x^30 + x^20 - x^10 + 1
? Ser(T, ’x, 11)
%5 = 1 - x^10 + O(x^11)
The function is more or less equivalent with multiplication by 1 + O(v d ) in theses cases, only faster.
For the remaining types, vectors and power series, we first explain what occurs if d is omitted.
In this case, the function uses exactly the amount of information given in the input:
• If s is already a power series in v, we return it verbatim;
• If s is a vector, the coefficients of the vector are understood to be the coefficients of the
power series starting from the constant term (as in Polrev(x)); in other words we convert t_VEC /
t_COL to the power series whose significant terms are exactly given by the vector entries.
On the other hand, if d is explicitly given, we abide by its value and return a series, truncated
or extended with zeros as needed, with d significant terms.
? v = [1,2,3];
? Ser(v, t) \\ 3 terms: seriesprecision is ignored!
%7 = 1 + 2*t + 3*t^2 + O(t^3)
? Ser(v, t, 7) \\ 7 terms as explicitly requested
%8 = 1 + 2*t + 3*t^2 + O(t^7)
? s = 1+x+O(x^2);
? Ser(s)
%10 = 1 + x + O(x^2) \\ 2 terms: seriesprecision is ignored
? Ser(s, x, 7) \\ extend to 7 terms
%11 = 1 + x + O(x^7)
? Ser(s, x, 1) \\ truncate to 1 term
%12 = 1 + O(x)

143
The warning given for Pol also applies here: this is not a substitution function.
The library syntax is GEN Ser0(GEN s, long v = -1, GEN d = NULL, long precdl) where
v is a variable number.

3.6.11 Set({x = [ ]}). Converts x into a set, i.e. into a row vector, with strictly increasing entries
with respect to the (somewhat arbitrary) universal comparison function cmp. Standard container
types t_VEC, t_COL, t_LIST and t_VECSMALL are converted to the set with corresponding elements.
All others are converted to a set with one element.
? Set([1,2,4,2,1,3])
%1 = [1, 2, 3, 4]
? Set(x)
%2 = [x]
? Set(Vecsmall([1,3,2,1,3]))
%3 = [1, 2, 3]
The library syntax is GEN gtoset(GEN x = NULL).

3.6.12 Str({x}∗). Converts its argument list into a single character string (type t_STR, the empty
string if x is omitted). To recover an ordinary GEN from a string, apply eval to it. The arguments
of Str are evaluated in string context, see Section 2.9.
? x2 = 0; i = 2; Str(x, i)
%1 = "x2"
? eval(%)
%2 = 0
This function is mostly useless in library mode. Use the pair strtoGEN/GENtostr to convert
between GEN and char*. The latter returns a malloced string, which should be freed after usage.

3.6.13 Vec(x, {n}). Transforms the object x into a row vector. The dimension of the resulting
vector can be optionally specified via the extra parameter n. If n is omitted or 0, the dimension
depends on the type of x; the vector has a single component, except when x is
• a vector or a quadratic form: returns the initial object considered as a row vector,
• a polynomial or a power series: returns a vector consisting of the coefficients. In the case of a
polynomial, the coefficients of the vector start with the leading coefficient of the polynomial, while
for power series only the significant coefficients are taken into account, but this time by increasing
order of degree. In particular the valuation is ignored (which makes the function useful for series
of negative valuation):
? Vec(3*x^2 + x)
%1 = [3, 1, 0]
? Vec(x^2 + 3*x^3 + O(x^5))
%2 = [1, 3, 0]
? Vec(x^-2 + 3*x^-1 + O(x))
%3 = [1, 3, 0]
Vec is the reciprocal function of Pol for a polynomial and of Ser for power series of valuation 0.
• a matrix: returns the vector of columns comprising the matrix,
? m = [1,2,3;4,5,6]

144
 %4 =
[1 2 3]
[4 5 6]
? Vec(m)
%5 = [[1, 4]~, [2, 5]~, [3, 6]~]

• a character string: returns the vector of individual characters,

? Vec("PARI")
%6 = ["P", "A", "R", "I"]

• a map: returns the vector of the domain of the map,

• an error context (t_ERROR): returns the error components, see iferr.

In the last four cases (matrix, character string, map, error), n is meaningless and must be
omitted or an error is raised. Otherwise, if n is given, 0 entries are appended at the end of the
vector if n > 0, and prepended at the beginning if n < 0. The dimension of the resulting vector is
|n|. This allows to write a conversion function for series that takes positive valuations into account:

? serVec(s) = Vec(s, -serprec(s,variable(s)));

? Vec(x^2 + 3*x^3 + O(x^5))
%2 = [0, 0, 1, 3, 0]

(That function is not intended for series of negative valuation.)

The library syntax is GEN gtovec0(GEN x, long n). GEN gtovec(GEN x) is also available.

3.6.14 Vecrev(x, {n}). As Vec(x, −n), then reverse the result. In particular, Vecrev is the
reciprocal function of Polrev: the coefficients of the vector start with the constant coefficient of
the polynomial and the others follow by increasing degree.

The library syntax is GEN gtovecrev0(GEN x, long n). GEN gtovecrev(GEN x) is also
available.

3.6.15 Vecsmall(x, {n}). Transforms the object x into a row vector of type t_VECSMALL. The
dimension of the resulting vector can be optionally specified via the extra parameter n.

This acts as Vec(x, n), but only on a limited set of objects: the result must be representable
as a vector of small integers. If x is a character string, a vector of individual characters in ASCII
encoding is returned (strchr yields back the character string).

The library syntax is GEN gtovecsmall0(GEN x, long n). GEN gtovecsmall(GEN x) is also
available.

145
3.6.16 binary(x). Outputs the vector of the binary digits of |x|. Here x can be an integer, a
real number (in which case the result has two components, one for the integer part, one for the
fractional part) or a vector/matrix.
? binary(10)
%1 = [1, 0, 1, 0]
? binary(3.14)
%2 = [[1, 1], [0, 0, 1, 0, 0, 0, [...]]
? binary([1,2])
%3 = [[1], [1, 0]]
For integer x ≥ 1, the number of bits is logint(x, 2) + 1. By convention, 0 has no digits:
? binary(0)
%4 = []
The library syntax is GEN binaire(GEN x).

3.6.17 bitand(x, y). Bitwise and of two integers x and y, that is the integer
X
(xi and yi )2i
i

Negative numbers behave 2-adically, i.e. the result is the 2-adic limit of bitand(xn , yn ), where
xn and yn are nonnegative integers tending to x and y respectively. (The result is an ordinary
integer, possibly negative.)
? bitand(5, 3)
%1 = 1
? bitand(-5, 3)
%2 = 3
? bitand(-5, -3)
%3 = -7
The library syntax is GEN gbitand(GEN x, GEN y). Also available is GEN ibitand(GEN x,
GEN y), which returns the bitwise and of |x| and |y|, two integers.

3.6.18 bitneg(x, {n = −1}). bitwise negation of an integer x, truncated to n bits, n ≥ 0, that is

the integer
n−1
X
not(xi )2i .
i=0

The special case n = −1 means no truncation: an infinite sequence of leading 1 is then represented
as a negative number.
See Section 3.6.17 for the behavior for negative arguments.
The library syntax is GEN gbitneg(GEN x, long n).

146
3.6.19 bitnegimply(x, y). Bitwise negated imply of two integers x and y (or not (x ⇒ y)), that
is the integer X
(xi andnot(yi ))2i

See Section 3.6.17 for the behavior for negative arguments.

The library syntax is GEN gbitnegimply(GEN x, GEN y). Also available is GEN ibitnegim-
ply(GEN x, GEN y), which returns the bitwise negated imply of |x| and |y|, two integers.

3.6.20 bitor(x, y). bitwise (inclusive) or of two integers x and y, that is the integer
X
(xi or yi )2i

See Section 3.6.17 for the behavior for negative arguments.

The library syntax is GEN gbitor(GEN x, GEN y). Also available is GEN ibitor(GEN x, GEN
y), which returns the bitwise or of |x| and |y|, two integers.

3.6.21 bitprecision(x, {n}). The function behaves differently according to whether n is present
or not. If n is missing, the function returns the (floating point) precision in bits of the PARI object
x.
If x is an exact object, the function returns +oo.
? bitprecision(exp(1e-100))
%1 = 512 \\ 512 bits
? bitprecision( [ exp(1e-100), 0.5 ] )
%2 = 128 \\ minimal accuracy among components
? bitprecision(2 + x)
%3 = +oo \\ exact object
Use getlocalbitprec() to retrieve the working bit precision (as modified by possible localbit-
prec statements).
If n is present and positive, the function creates a new object equal to x with the new bit-
precision roughly n. In fact, the smallest multiple of 64 (resp. 32 on a 32-bit machine) larger than
or equal to n.
For x a vector or a matrix, the operation is done componentwise; for series and polynomials,
the operation is done coefficientwise. For real x, n is the number of desired significant bits. If n
is smaller than the precision of x, x is truncated, otherwise x is extended with zeros. For exact or
non-floating-point types, no change.
? bitprecision(Pi, 10) \\ actually 64 bits ~ 19 decimal digits
%1 = 3.141592653589793239
? bitprecision(1, 10)
%2 = 1
? bitprecision(1 + O(x), 10)
%3 = 1 + O(x)
? bitprecision(2 + O(3^5), 10)
%4 = 2 + O(3^5)

The library syntax is GEN bitprecision00(GEN x, GEN n = NULL).

147
3.6.22 bittest(x, n). Outputs the nth bit of x starting from the right (i.e. the coefficient of 2n in
the binary expansion of x). The result is 0 or 1. For x ≥ 1, the highest 1-bit is at n = logint(x)
(and bigger n gives 0).
? bittest(7, 0)
%1 = 1 \\ the bit 0 is 1
? bittest(7, 2)
%2 = 1 \\ the bit 2 is 1
? bittest(7, 3)
%3 = 0 \\ the bit 3 is 0
See Section 3.6.17 for the behavior at negative arguments.
The library syntax is GEN gbittest(GEN x, long n). For a t_INT x, the variant long
bittest(GEN x, long n) is generally easier to use, and if furthermore n ≥ 0 the low-level function
ulong int_bit(GEN x, long n) returns bittest(abs(x),n).

3.6.23 bitxor(x, y). Bitwise (exclusive) or of two integers x and y, that is the integer
X
(xi xor yi )2i

See Section 3.6.17 for the behavior for negative arguments.

The library syntax is GEN gbitxor(GEN x, GEN y). Also available is GEN ibitxor(GEN x,
GEN y), which returns the bitwise xor of |x| and |y|, two integers.

3.6.24 ceil(x). Ceiling of x. When x is in R, the result is the smallest integer greater than or equal
to x. Applied to a rational function, ceil(x) returns the Euclidean quotient of the numerator by
the denominator.
The library syntax is GEN gceil(GEN x).

3.6.25 centerlift(x, {v}). Same as lift, except that t_INTMOD and t_PADIC components are lifted
using centered residues:
• for a t_INTMOD x ∈ Z/nZ, the lift y is such that −n/2 < y ≤ n/2.
• a t_PADIC x is lifted in the same way as above (modulo ppadicprec(x) ) if its valuation v is
nonnegative; if not, returns the fraction pv centerlift(xp−v ); in particular, rational reconstruction
is not attempted. Use bestappr for this.
For backward compatibility, centerlift(x,’v) is allowed as an alias for lift(x,’v).
The library syntax is centerlift(GEN x).

3.6.26 characteristic(x). Returns the characteristic of the base ring over which x is defined (as
defined by t_INTMOD and t_FFELT components). The function raises an exception if incompatible
primes arise from t_FFELT and t_PADIC components.
? characteristic(Mod(1,24)*x + Mod(1,18)*y)
%1 = 6
The library syntax is GEN characteristic(GEN x).

148
3.6.27 component(x, n). Extracts the nth -component of x. This is to be understood as follows:
every PARI type has one or two initial code words. The components are counted, starting at 1,
after these code words. In particular if x is a vector, this is indeed the nth -component of x, if x
is a matrix, the nth column, if x is a polynomial, the nth coefficient (i.e. of degree n − 1), and for
power series, the nth significant coefficient.
For polynomials and power series, one should rather use polcoeff, and for vectors and matri-
ces, the [ ] operator. Namely, if x is a vector, then x[n] represents the nth component of x. If x is
a matrix, x[m,n] represents the coefficient of row m and column n of the matrix, x[m,] represents
the mth row of x, and x[,n] represents the nth column of x.
Using of this function requires detailed knowledge of the structure of the different PARI types,
and thus it should almost never be used directly. Some useful exceptions:
? x = 3 + O(3^5);
? component(x, 2)
%2 = 81 \\ p^(p-adic accuracy)
? component(x, 1)
%3 = 3 \\ p
? q = Qfb(1,2,3);
? component(q, 1)
%5 = 1
The library syntax is GEN compo(GEN x, long n).
3.6.28 conj(x). Conjugate of x. The meaning of this is clear, except that for real quadratic
numbers, it means conjugation in the real quadratic field. This function has no effect on integers,
reals, intmods, fractions or p-adics. The only forbidden type is polmod (see conjvec for this).
The library syntax is GEN gconj(GEN x).
3.6.29 conjvec(z). Conjugate vector representation of z. If z is a polmod, equal to Mod(a, T ), this
gives a vector of length degree(T ) containing:
• the complex embeddings of z if T has rational coefficients, i.e. the a(r[i]) where r =
polroots(T );
• the conjugates of z if T has some intmod coefficients;
2 n−1
if z is a finite field element, the result is the vector of conjugates [z, z p , z p , . . . , z p ] where n =
degree(T ).
If z is an integer or a rational number, the result is z. If z is a (row or column) vector, the result
is a matrix whose columns are the conjugate vectors of the individual elements of z.
The library syntax is GEN conjvec(GEN z, long prec).
3.6.30 denominator(f, {D}). Denominator of f . The meaning of this is clear when f is a rational
number or function. If f is an integer or a polynomial, it is treated as a rational number or function,
respectively, and the result is equal to 1. For polynomials, you probably want to use
denominator( content(f) )
instead. As for modular objects, t_INTMOD and t_PADIC have denominator 1, and the denominator
of a t_POLMOD is the denominator of its lift.
If f is a recursive structure, for instance a vector or matrix, the lcm of the denominators of its
components (a common denominator) is computed. This also applies for t_COMPLEXs and t_QUADs.

149
Warning. Multivariate objects are created according to variable priorities, with possibly surprising
side effects (x/y is a polynomial, but y/x is a rational function). See Section 2.5.3.
The optional argument D allows to control over which ring we compute the denominator and
get a more predictable behaviour:
• 1: we only consider the underlying Q-structure and the denominator is a (positive) rational
integer
• a simple variable, say ’x: all entries as rational functions in K(x) and the denominator is a
polynomial in x.
? f = x + 1/y + 1/2;
? denominator(f) \\ a t_POL in x
%2 = 1
? denominator(f, 1) \\ Q-denominator
%3 = 2
? denominator(f, x) \\ as a t_POL in x, seen above
%4 = 1
? denominator(f, y) \\ as a rational function in y
%5 = 2*y
The library syntax is GEN denominator(GEN f, GEN D = NULL). Also available are GEN
denom(GEN x) which implements the not very useful default behaviour (D is NULL) and GEN
Q_denom(GEN x) (D = 1).

3.6.31 digits(x, {b = 10}). Outputs the vector of the digits of |x| in base b, where x and b are
integers (b = 10 by default). For x ≥ 1, the number of digits is logint(x, b) + 1. See fromdigits
for the reverse operation.
? digits(1230)
%1 = [1, 2, 3, 0]
? digits(10, 2) \\ base 2
%2 = [1, 0, 1, 0]
By convention, 0 has no digits:
? digits(0)
%3 = []
The library syntax is GEN digits(GEN x, GEN b = NULL).

3.6.32 exponent(x). When x is a t_REAL, the result is the binary exponent e of x. For a nonzero
x, this is the unique integer e such that 2e ≤ |x| < 2e+1 . For a real 0, this returns the PARI
exponent e attached to x (which may represent any floating-point number less than 2e in absolute
value).
? exponent(Pi)
%1 = 1
? exponent(4.0)
%2 = 2
? exponent(0.0)
%3 = -128
? default(realbitprecision)

150
 %4 = 128

This definition extends naturally to nonzero integers, and the exponent of an exact 0 is −oo by
convention.

For convenience, we define the exponent of a t_FRAC a/b as the difference of exponent(a) and
exponent(b); note that, if e0 denotes the exponent of a/b * 1.0, then the exponent e we return is
either e0 or e0 + 1, thus 2e+1 is an upper bound for |a/b|.

? [ exponent(9), exponent(10), exponent(9/10), exponent(9/10*1.) ]

%5 = [3, 3, 0, -1]

For a PARI object of type t_COMPLEX, t_POL, t_SER, t_VEC, t_COL, t_MAT this returns the
largest exponent found among the components of x. Hence 2e+1 is a quick upper bound for the
sup norm of real matrices or polynomials; and 2e+(3/2) for complex ones.

? exponent(3*x^2 + 15*x - 100)

%5 = 6
? exponent(0)
%6 = -oo

The library syntax is GEN gpexponent(GEN x).

3.6.33 floor(x). Floor of x. When x is in R, the result is the largest integer smaller than or equal
to x. Applied to a rational function, floor(x) returns the Euclidean quotient of the numerator by
the denominator.

The library syntax is GEN gfloor(GEN x).

3.6.34 frac(x). Fractional part of x. Identical to x − floor(x). If x is real, the result is in [0, 1[.

The library syntax is GEN gfrac(GEN x).

3.6.35 fromdigits(x, {b = 10}). Gives the integer formed by the elements of x seen as the digits
of a number in base b (b = 10 by default). This is the reverse of digits:

? digits(1234,5)
%1 = [1,4,4,1,4]
? fromdigits([1,4,4,1,4],5)
%2 = 1234

By convention, 0 has no digits:

? fromdigits([])
%3 = 0

The library syntax is GEN fromdigits(GEN x, GEN b = NULL).

151
3.6.36 imag(x). Imaginary part of x. When x is a quadratic number, this is the coefficient of ω
in the “canonical” integral basis (1, ω).

? imag(3 + I)
%1 = 1
? x = 3 + quadgen(-23);
? imag(x) \\ as a quadratic number
%3 = 1
? imag(x * 1.) \\ as a complex number
%4 = 2.3979157616563597707987190320813469600

The library syntax is GEN gimag(GEN x).

3.6.37 length(x). Length of x; #x is a shortcut for length(x). This is mostly useful for

• vectors: dimension (0 for empty vectors),

• lists: number of entries (0 for empty lists),

• maps: number of entries (0 for empty maps),

• matrices: number of columns,

• character strings: number of actual characters (without trailing \0, should you expect it
from C char*).

? #"a string"
%1 = 8
? #[3,2,1]
%2 = 3
? #[]
%3 = 0
? #matrix(2,5)
%4 = 5
? L = List([1,2,3,4]); #L
%5 = 4
? M = Map([a,b; c,d; e,f]); #M
%6 = 3

The routine is in fact defined for arbitrary GP types, but is awkward and useless in other
cases: it returns the number of non-code words in x, e.g. the effective length minus 2 for integers
since the t_INT type has two code words.

The library syntax is long glength(GEN x).

152
3.6.38 lift(x, {v}). If v is omitted, lifts intmods from Z/nZ in Z, p-adics from Qp to Q (as
truncate), and polmods to polynomials. Otherwise, lifts only polmods whose modulus has main
variable v. t_FFELT are not lifted, nor are List elements: you may convert the latter to vectors
first, or use apply(lift,L). More generally, components for which such lifts are meaningless (e.g.
character strings) are copied verbatim.

? lift(Mod(5,3))
%1 = 2
? lift(3 + O(3^9))
%2 = 3
? lift(Mod(x,x^2+1))
%3 = x
? lift(Mod(x,x^2+1))
%4 = x

Lifts are performed recursively on an object components, but only by one level : once a
t_POLMOD is lifted, the components of the result are not lifted further.

? lift(x * Mod(1,3) + Mod(2,3))

%4 = x + 2
? lift(x * Mod(y,y^2+1) + Mod(2,3))
%5 = y*x + Mod(2, 3) \\ do you understand this one?
? lift(x * Mod(y,y^2+1) + Mod(2,3), ’x)
%6 = Mod(y, y^2 + 1)*x + Mod(Mod(2, 3), y^2 + 1)
? lift(%, y)
%7 = y*x + Mod(2, 3)

To recursively lift all components not only by one level, but as long as possible, use liftall. To
lift only t_INTMODs and t_PADICs components, use liftint. To lift only t_POLMODs components,
use liftpol. Finally, centerlift allows to lift t_INTMODs and t_PADICs using centered residues
(lift of smallest absolute value).

The library syntax is GEN lift0(GEN x, long v = -1) where v is a variable number. Also
available is GEN lift(GEN x) corresponding to lift0(x,-1).

3.6.39 liftall(x). Recursively lift all components of x from Z/nZ to Z, from Qp to Q (as truncate),
and polmods to polynomials. t_FFELT are not lifted, nor are List elements: you may convert the
latter to vectors first, or use apply(liftall,L). More generally, components for which such lifts
are meaningless (e.g. character strings) are copied verbatim.

? liftall(x * (1 + O(3)) + Mod(2,3))

%1 = x + 2
? liftall(x * Mod(y,y^2+1) + Mod(2,3)*Mod(z,z^2))
%2 = y*x + 2*z

The library syntax is GEN liftall(GEN x).

153
3.6.40 liftint(x). Recursively lift all components of x from Z/nZ to Z and from Qp to Q (as
truncate). t_FFELT are not lifted, nor are List elements: you may convert the latter to vectors
first, or use apply(liftint,L). More generally, components for which such lifts are meaningless
(e.g. character strings) are copied verbatim.
? liftint(x * (1 + O(3)) + Mod(2,3))
%1 = x + 2
? liftint(x * Mod(y,y^2+1) + Mod(2,3)*Mod(z,z^2))
%2 = Mod(y, y^2 + 1)*x + Mod(Mod(2*z, z^2), y^2 + 1)
The library syntax is GEN liftint(GEN x).

3.6.41 liftpol(x). Recursively lift all components of x which are polmods to polynomials. t_FFELT
are not lifted, nor are List elements: you may convert the latter to vectors first, or use ap-
ply(liftpol,L). More generally, components for which such lifts are meaningless (e.g. character
strings) are copied verbatim.
? liftpol(x * (1 + O(3)) + Mod(2,3))
%1 = (1 + O(3))*x + Mod(2, 3)
? liftpol(x * Mod(y,y^2+1) + Mod(2,3)*Mod(z,z^2))
%2 = y*x + Mod(2, 3)*z
The library syntax is GEN liftpol(GEN x).

3.6.42 norm(x). Algebraic norm of x, i.e. the product of x with its conjugate (no square roots
are taken), or conjugates for polmods. For vectors and matrices, the norm is taken componentwise
and hence is not the L2 -norm (see norml2). Note that the norm of an element of R is its square,
so as to be compatible with the complex norm.
The library syntax is GEN gnorm(GEN x).

3.6.43 numerator(f, {D}). Numerator of f . This is defined as f * denominator(f,D), see

denominator for details. The optional argument D allows to control over which ring we compute
the denominator:
• 1: we only consider the underlying Q-structure and the denominator is a (positive) rational
integer
• a simple variable, say ’x: all entries as rational functions in K(x) and the denominator is a
polynomial in x.
? f = x + 1/y + 1/2;
? numerator(f) \\ a t_POL in x
%2 = x + ((y + 2)/(2*y))
? numerator(f, 1) \\ Q-denominator is 2
%3 = x + ((y + 2)/y)
? numerator(f, y) \\ as a rational function in y
%5 = 2*y*x + (y + 2)
The library syntax is GEN numerator(GEN f, GEN D = NULL). Also available are GEN nu-
mer(GEN x) which implements the not very useful default behaviour (D is NULL) and GEN
Q_remove_denom(GEN x, GEN *ptd) (D = 1) and also returns the denominator (coding 1 as
NULL).

154
3.6.44 oo. Returns an object meaning +∞, for use in functions such as intnum. It can be negated
(-oo represents −∞), and compared to real numbers (t_INT, t_FRAC, t_REAL), with the expected
meaning: +∞ is greater than any real number and −∞ is smaller.
The library syntax is GEN mkoo().

3.6.45 padicprec(x, p). Returns the absolute p-adic precision of the object x; this is the minimum
precision of the components of x. The result is +oo if x is an exact object (as a p-adic):
? padicprec((1 + O(2^5)) * x + (2 + O(2^4)), 2)
%1 = 4
? padicprec(x + 2, 2)
%2 = +oo
? padicprec(2 + x + O(x^2), 2)
%3 = +oo
The function raises an exception if it encounters an object incompatible with p-adic computations:
? padicprec(O(3), 2)
*** at top-level: padicprec(O(3),2)
*** ^-----------------
*** padicprec: inconsistent moduli in padicprec: 3 != 2
? padicprec(1.0, 2)
*** at top-level: padicprec(1.0,2)
*** ^----------------
*** padicprec: incorrect type in padicprec (t_REAL).
The library syntax is GEN gppadicprec(GEN x, GEN p). Also available is the function long
padicprec(GEN x, GEN p), which returns LONG_MAX if x = 0 and the p-adic precision as a long
integer.

3.6.46 precision(x, {n}). The function behaves differently according to whether n is present or
not. If n is missing, the function returns the floating point precision in decimal digits of the PARI
object x. If x has no floating point component, the function returns +oo.
? precision(exp(1e-100))
%1 = 154 \\ 154 significant decimal digits
? precision(2 + x)
%2 = +oo \\ exact object
? precision(0.5 + O(x))
%3 = 38 \\ floating point accuracy, NOT series precision
? precision( [ exp(1e-100), 0.5 ] )
%4 = 38 \\ minimal accuracy among components
Using getlocalprec() allows to retrieve the working precision (as modified by possible localprec
statements).
If n is present, the function creates a new object equal to x with a new floating point precision
n: n is the number of desired significant decimal digits. If n is smaller than the precision of a
t_REAL component of x, it is truncated, otherwise it is extended with zeros. For non-floating-point
types, no change.

155
 The library syntax is GEN precision00(GEN x, GEN n = NULL). Also available are GEN
gprec(GEN x, long n) and long precision(GEN x). In both, the accuracy is expressed in
words (32-bit or 64-bit depending on the architecture).

3.6.47 random({N = 231 }). Returns a random element in various natural sets depending on the
argument N .
• t_INT: returns an integer uniformly distributed between 0 and N −1. Omitting the argument
is equivalent to random(2^31).
• t_REAL: returns a real number in [0, 1[ with the same accuracy as N (whose mantissa has
the same number of significant words).
• t_INTMOD: returns a random intmod for the same modulus.
• t_FFELT: returns a random element in the same finite field.
• t_VEC of length 2, N = [a, b]: returns an integer uniformly distributed between a and b.
• t_VEC generated by ellinit over a finite field k (coefficients are t_INTMODs modulo a prime
or t_FFELTs): returns a “random” k-rational affine point on the curve. More precisely if the curve
has a single point (at infinity!) we return it; otherwise we return an affine point by drawing an
abscissa uniformly at random until ellordinate succeeds. Note that this is definitely not a uniform
distribution over E(k), but it should be good enough for applications.
• t_POL return a random polynomial of degree at most the degree of N . The coefficients are
drawn by applying random to the leading coefficient of N .

? random(10)
%1 = 9
? random(Mod(0,7))
%2 = Mod(1, 7)
? a = ffgen(ffinit(3,7), ’a); random(a)
%3 = a^6 + 2*a^5 + a^4 + a^3 + a^2 + 2*a
? E = ellinit([3,7]*Mod(1,109)); random(E)
%4 = [Mod(103, 109), Mod(10, 109)]
? E = ellinit([1,7]*a^0); random(E)
%5 = [a^6 + a^5 + 2*a^4 + 2*a^2, 2*a^6 + 2*a^4 + 2*a^3 + a^2 + 2*a]
? random(Mod(1,7)*x^4)
%6 = Mod(5, 7)*x^4 + Mod(6, 7)*x^3 + Mod(2, 7)*x^2 + Mod(2, 7)*x + Mod(5, 7)

These variants all depend on a single internal generator, and are independent from your oper-
ating system’s random number generators. A random seed may be obtained via getrand, and reset
using setrand: from a given seed, and given sequence of randoms, the exact same values will be
generated. The same seed is used at each startup, reseed the generator yourself if this is a problem.
Note that internal functions also call the random number generator; adding such a function call in
the middle of your code will change the numbers produced.

156
Technical note. Up to version 2.4 included, the internal generator produced pseudo-random
numbers by means of linear congruences, which were not well distributed in arithmetic pro-
gressions. We now use Brent’s XORGEN algorithm, based on Feedback Shift Registers, see
http://wwwmaths.anu.edu.au/~brent/random.html. The generator has period 24096 − 1, passes
the Crush battery of statistical tests of L’Ecuyer and Simard, but is not suitable for cryptographic
purposes: one can reconstruct the state vector from a small sample of consecutive values, thus
predicting the entire sequence.

The library syntax is GEN genrand(GEN N = NULL).

Also available: GEN ellrandom(GEN E) and GEN ffrandom(GEN a).

3.6.48 real(x). Real part of x. When x is a quadratic number, this is the coefficient of 1 in the
“canonical” integral basis (1, ω).

? real(3 + I)
%1 = 3
? x = 3 + quadgen(-23);
? real(x) \\ as a quadratic number
%3 = 3
? real(x * 1.) \\ as a complex number
%4 = 3.5000000000000000000000000000000000000

The library syntax is GEN greal(GEN x).

3.6.49 round(x, {&e}). If x is in R, rounds x to the nearest integer (rounding to +∞ in case

of ties), then and sets e to the number of error bits, that is the binary exponent of the difference
between the original and the rounded value (the “fractional part”). If the exponent of x is too large
compared to its precision (i.e. e > 0), the result is undefined and an error occurs if e was not given.

Important remark. Contrary to the other truncation functions, this function operates on every
coefficient at every level of a PARI object. For example

2.4 ∗ X 2 − 1.7
 
truncate = 2.4 ∗ X,
X

whereas
2.4 ∗ X 2 − 1.7 2 ∗ X2 − 2
 
round = .
X X

An important use of round is to get exact results after an approximate computation, when theory
tells you that the coefficients must be integers.

The library syntax is GEN round0(GEN x, GEN *e = NULL). Also available are GEN grnd-
toi(GEN x, long *e) and GEN ground(GEN x).

157
3.6.50 serchop(s, {n = 0}). Remove all terms of degree strictly less than n in series s. When the
series contains no terms of degree < n, return O(xn ).
? s = 1/x + x + 2*x^2 + O(x^3);
? serchop(s)
%2 = x + 2*x^3 + O(x^3)
? serchop(s, 2)
%3 = 2*x^2 + O(x^3)
? serchop(s, 100)
%4 = O(x^100)
The library syntax is GEN serchop(GEN s, long n).

3.6.51 serprec(x, v). Returns the absolute precision of x with respect to power series in the
variable v; this is the minimum precision of the components of x. The result is +oo if x is an exact
object (as a series in v):
? serprec(x + O(y^2), y)
%1 = 2
? serprec(x + 2, x)
%2 = +oo
? serprec(2 + x + O(x^2), y)
%3 = +oo
The library syntax is GEN gpserprec(GEN x, long v) where v is a variable number. Also
available is long serprec(GEN x, GEN p), which returns LONG_MAX if x = 0, otherwise the series
precision as a long integer.

3.6.52 simplify(x). This function simplifies x as much as it can. Specifically, a complex or

quadratic number whose imaginary part is the integer 0 (i.e. not Mod(0,2) or 0.E-28) is converted
to its real part, and a polynomial of degree 0 is converted to its constant term. Simplifications
occur recursively.
This function is especially useful before using arithmetic functions, which expect integer argu-
ments:
? x = 2 + y - y
%1 = 2
? isprime(x)
*** at top-level: isprime(x)
*** ^----------
*** isprime: not an integer argument in an arithmetic function
? type(x)
%2 = "t_POL"
? type(simplify(x))
%3 = "t_INT"
Note that GP results are simplified as above before they are stored in the history. (Unless you
disable automatic simplification with \y, that is.) In particular
? type(%1)
%4 = "t_INT"
The library syntax is GEN simplify(GEN x).

158
3.6.53 sizebyte(x). Outputs the total number of bytes occupied by the tree representing the PARI
object x.
The library syntax is long gsizebyte(GEN x). Also available is long gsizeword(GEN x)
returning a number of words.

3.6.54 sizedigit(x). This function is DEPRECATED, essentially meaningless, and provided for
backwards compatibility only. Don’t use it!
outputs a quick upper bound for the number of decimal digits of (the components of) x, off
by at most 1. More precisely, for a positive integer x, it computes (approximately) the ceiling of

floor(1 + log2 x) log10 2,

To count the number of decimal digits of a positive integer x, use #digits(x). To estimate
(recursively) the size of x, use normlp(x).
The library syntax is long sizedigit(GEN x).

3.6.55 truncate(x, {&e}). Truncates x and sets e to the number of error bits. When x is in R,
this means that the part after the decimal point is chopped away, e is the binary exponent of the
difference between the original and the truncated value (the “fractional part”). If the exponent of
x is too large compared to its precision (i.e. e > 0), the result is undefined and an error occurs if
e was not given. The function applies componentwise on vector / matrices; e is then the maximal
number of error bits. If x is a rational function, the result is the “integer part” (Euclidean quotient
of numerator by denominator) and e is not set.
Note a very special use of truncate: when applied to a power series, it transforms it into a
polynomial or a rational function with denominator a power of X, by chopping away the O(X k ).
Similarly, when applied to a p-adic number, it transforms it into an integer or a rational number
by chopping away the O(pk ).
The library syntax is GEN trunc0(GEN x, GEN *e = NULL). The following functions are also
available: GEN gtrunc(GEN x) and GEN gcvtoi(GEN x, long *e).

3.6.56 valuation(x, p). Computes the highest exponent of p dividing x. If p is of type integer, x
must be an integer, an intmod whose modulus is divisible by p, a fraction, a q-adic number with
q = p, or a polynomial or power series in which case the valuation is the minimum of the valuation
of the coefficients.
If p is of type polynomial, x must be of type polynomial or rational function, and also a power
series if x is a monomial. Finally, the valuation of a vector, complex or quadratic number is the
minimum of the component valuations.
If x = 0, the result is +oo if x is an exact object. If x is a p-adic numbers or power series, the
result is the exponent of the zero. Any other type combinations gives an error.
The library syntax is GEN gpvaluation(GEN x, GEN p). Also available is long gvalua-
tion(GEN x, GEN p), which returns LONG_MAX if x = 0 and the valuation as a long integer.

159
3.6.57 varhigher(name, {v}). Return a variable name whose priority is higher than the priority
of v (of all existing variables if v is omitted). This is a counterpart to varlower.
? Pol([x,x], t)
*** at top-level: Pol([x,x],t)
*** ^------------
*** Pol: incorrect priority in gtopoly: variable x <= t
? t = varhigher("t", x);
? Pol([x,x], t)
%3 = x*t + x
This routine is useful since new GP variables directly created by the interpreter always have lower
priority than existing GP variables. When some basic objects already exist in a variable that is
incompatible with some function requirement, you can now create a new variable with a suitable
priority instead of changing variables in existing objects:
? K = nfinit(x^2+1);
? rnfequation(K,y^2-2)
*** at top-level: rnfequation(K,y^2-2)
*** ^--------------------
*** rnfequation: incorrect priority in rnfequation: variable y >= x
? y = varhigher("y", x);
? rnfequation(K, y^2-2)
%3 = y^4 - 2*y^2 + 9

Caution 1. The name is an arbitrary character string, only used for display purposes and need
not be related to the GP variable holding the result, nor to be a valid variable name. In particular
the name can not be used to retrieve the variable, it is not even present in the parser’s hash tables.
? x = varhigher("#");
? x^2
%2 = #^2

Caution 2. There are a limited number of variables and if no existing variable with the given
display name has the requested priority, the call to varhigher uses up one such slot. Do not create
new variables in this way unless it’s absolutely necessary, reuse existing names instead and choose
sensible priority requirements: if you only need a variable with higher priority than x, state so
rather than creating a new variable with highest priority.
\\ quickly use up all variables
? n = 0; while(1,varhigher("tmp"); n++)
*** at top-level: n=0;while(1,varhigher("tmp");n++)
*** ^-------------------
*** varhigher: no more variables available.
*** Break loop: type ’break’ to go back to GP prompt
break> n
65510
\\ infinite loop: here we reuse the same ’tmp’
? n = 0; while(1,varhigher("tmp", x); n++)
The library syntax is GEN varhigher(const char *name, long v = -1) where v is a variable
number.

160
3.6.58 variable({x}). Gives the main variable of the object x (the variable with the highest
priority used in x), and p if x is a p-adic number. Return 0 if x has no variable attached to it.
? variable(x^2 + y)
%1 = x
? variable(1 + O(5^2))
%2 = 5
? variable([x,y,z,t])
%3 = x
? variable(1)
%4 = 0
The construction
if (!variable(x),...)
can be used to test whether a variable is attached to x.
If x is omitted, returns the list of user variables known to the interpreter, by order of decreasing
priority. (Highest priority is initially x, which come first until varhigher is used.) If varhigher or
varlower are used, it is quite possible to end up with different variables (with different priorities)
printed in the same way: they will then appear multiple times in the output:
? varhigher("y");
? varlower("y");
? variable()
%4 = [y, x, y]
Using v = variable() then v[1], v[2], etc. allows to recover and use existing variables.
The library syntax is GEN gpolvar(GEN x = NULL). However, in library mode, this function
should not be used for x non-NULL, since gvar is more appropriate. Instead, for x a p-adic (type
t_PADIC), p is gel(x, 2); otherwise, use long gvar(GEN x) which returns the variable number of x
if it exists, NO VARIABLE otherwise, which satisfies the property varncmp(NO VARIABLE, v) > 0 for
all valid variable number v, i.e. it has lower priority than any variable.

3.6.59 variables({x}). Returns the list of all variables occurring in object x (all user variables
known to the interpreter if x is omitted), sorted by decreasing priority.
? variables([x^2 + y*z + O(t), a+x])
%1 = [x, y, z, t, a]
The construction
if (!variables(x),...)
can be used to test whether a variable is attached to x.
If varhigher or varlower are used, it is quite possible to end up with different variables (with
different priorities) printed in the same way: they will then appear multiple times in the output:
? y1 = varhigher("y");
? y2 = varlower("y");
? variables(y*y1*y2)
%4 = [y, y, y]
The library syntax is GEN variables_vec(GEN x = NULL).

161
 Also available is GEN variables_vecsmall(GEN x) which returns the (sorted) variable num-
bers instead of the attached monomials of degree 1.

3.6.60 varlower(name, {v}). Return a variable name whose priority is lower than the priority of
v (of all existing variables if v is omitted). This is a counterpart to varhigher.
New GP variables directly created by the interpreter always have lower priority than existing
GP variables, but it is not easy to check whether an identifier is currently unused, so that the
corresponding variable has the expected priority when it’s created! Thus, depending on the session
history, the same command may fail or succeed:

? t; z; \\ now t > z
? rnfequation(t^2+1,z^2-t)
*** at top-level: rnfequation(t^2+1,z^
*** ^--------------------
*** rnfequation: incorrect priority in rnfequation: variable t >= t
Restart and retry:

? z; t; \\ now z > t
? rnfequation(t^2+1,z^2-t)
%2 = z^4 + 1
It is quite annoying for package authors, when trying to define a base ring, to notice that the
package may fail for some users depending on their session history. The safe way to do this is as
follows:

? z; t; \\ In new session: now z > t

...
? t = varlower("t", ’z);
? rnfequation(t^2+1,z^2-2)
%2 = z^4 - 2*z^2 + 9
? variable()
%3 = [x, y, z, t]

? t; z; \\ In new session: now t > z

...
? t = varlower("t", ’z); \\ create a new variable, still printed "t"
? rnfequation(t^2+1,z^2-2)
%2 = z^4 - 2*z^2 + 9
? variable()
%3 = [x, y, t, z, t]
Now both constructions succeed. Note that in the first case, varlower is essentially a no-op,
the existing variable t has correct priority. While in the second case, two different variables are
displayed as t, one with higher priority than z (created in the first line) and another one with lower
priority (created by varlower).

162
Caution 1. The name is an arbitrary character string, only used for display purposes and need
not be related to the GP variable holding the result, nor to be a valid variable name. In particular
the name can not be used to retrieve the variable, it is not even present in the parser’s hash tables.

? x = varlower("#");
? x^2
%2 = #^2

Caution 2. There are a limited number of variables and if no existing variable with the given
display name has the requested priority, the call to varlower uses up one such slot. Do not create
new variables in this way unless it’s absolutely necessary, reuse existing names instead and choose
sensible priority requirements: if you only need a variable with higher priority than x, state so
rather than creating a new variable with highest priority.

\\ quickly use up all variables

? n = 0; while(1,varlower("x"); n++)
*** at top-level: n=0;while(1,varlower("x");n++)
*** ^-------------------
*** varlower: no more variables available.
*** Break loop: type ’break’ to go back to GP prompt
break> n
65510
\\ infinite loop: here we reuse the same ’tmp’
? n = 0; while(1,varlower("tmp", x); n++)

The library syntax is GEN varlower(const char *name, long v = -1) where v is a variable
number.

3.7 Combinatorics.

Permutations are represented in gp as t_VECSMALLs and can be input directly as Vecs-

mall([1,3,2,4]) or obtained from the iterator forperm:

? forperm(3, p, print(p)) \\ iterate through S_3

Vecsmall([1, 2, 3])
Vecsmall([1, 3, 2])
Vecsmall([2, 1, 3])
Vecsmall([2, 3, 1])
Vecsmall([3, 1, 2])
Vecsmall([3, 2, 1])

Permutations can be multiplied via *, raised to some power using ^, inverted using ^(-1), con-
jugated as p * q * p^(-1). Their order and signature is available via permorder and permsign.

163
3.7.1 bernfrac(n). Bernoulli number Bn , where B0 = 1, B1 = −1/2, B2 = 1/6,. . . , expressed as
a rational number. The argument n should be a nonnegative integer. The function bervec creates
a cache of successive Bernoulli numbers which greatly speeds up later calls to bernfrac:
? bernfrac(20000);
time = 107 ms.
? bernvec(10000); \\ cache B_0, B_2, ..., B_20000
time = 35,957 ms.
? bernfrac(20000); \\ now instantaneous
?
The library syntax is GEN bernfrac(long n).

3.7.2 bernpol(n, {v =0 x}). Bernoulli polynomial Bn in variable v.

? bernpol(1)
%1 = x - 1/2
? bernpol(3)
%2 = x^3 - 3/2*x^2 + 1/2*x
The library syntax is GEN bernpol(long n, long v = -1) where v is a variable number.

3.7.3 bernreal(n). Bernoulli number Bn , as bernfrac, but Bn is returned as a real number (with
the current precision). The argument n should be a nonnegative integer. The function slows down
as the precision increases:
? \p1000
? bernreal(200000);
time = 5 ms.
? \p10000
? bernreal(200000);
time = 18 ms.
? \p100000
? bernreal(200000);
time = 84 ms.
The library syntax is GEN bernreal(long n, long prec).

3.7.4 bernvec(n). Returns a vector containing, as rational numbers, the Bernoulli numbers B0 ,
B2 ,. . . , B2n :
? bernvec(5) \\ B_0, B_2..., B_10
%1 = [1, 1/6, -1/30, 1/42, -1/30, 5/66]
? bernfrac(10)
%2 = 5/66
This routine uses a lot of memory but is much faster than repeated calls to bernfrac:
? forstep(n = 2, 10000, 2, bernfrac(n))
time = 18,245 ms.
? bernvec(5000);
time = 1,338 ms.

164
The computed Bernoulli numbers are stored in an incremental cache which makes later calls to
bernfrac and bernreal instantaneous in the cache range: re-running the same previous bernfracs
after the bernvec call gives:
? forstep(n = 2, 10000, 2, bernfrac(n))
time = 1 ms.
The time and space complexity of this function are Õ(n2 ); in the feasible range n ≤ 105 (requires
about two hours), the practical time complexity is closer to Õ(nlog2 6 ).
The library syntax is GEN bernvec(long n).
 
x
3.7.5 binomial(x, {k}). binomial coefficient . Here k must be an integer, but x can be any
k
PARI object.
? binomial(4,2)
%1 = 6
? n = 4; vector(n+1, k, binomial(n,k-1))
%2 = [1, 4, 6, 4, 1]
The argument k may be omitted if x = n is a nonnegative integer; in this case, return the vector
with n + 1 components whose k + 1-th entry is binomial(n, k)
? binomial(4)
%3 = [1, 4, 6, 4, 1]
The library syntax is GEN binomial0(GEN x, GEN k = NULL).

3.7.6 eulerfrac(n). Euler number En , where E0 = 1, E1 = 0, E2 = −1, . . . , are integers such that

1 X En
= tn .
cosh t n!
n≥0

The argument n should be a nonnegative integer.

? vector(10,i,eulerfrac(i))
%1 = [0, -1, 0, 5, 0, -61, 0, 1385, 0, -50521]
? eulerfrac(20000);
? sizedigit(%))
%3 = 73416
The library syntax is GEN eulerfrac(long n).

3.7.7 eulerianpol(n, {v =0 x}). Eulerian polynomial An in variable v.

? eulerianpol(2)
%1 = x + 1
? eulerianpol(5, ’t)
%2 = t^4 + 26*t^3 + 66*t^2 + 26*t + 1

The library syntax is GEN eulerianpol(long n, long v = -1) where v is a variable number.

165
3.7.8 eulerpol(n, {v =0 x}). Euler polynomial En in variable v.
? eulerpol(1)
%1 = x - 1/2
? eulerpol(3)
%2 = x^3 - 3/2*x^2 + 1/4
The library syntax is GEN eulerpol(long n, long v = -1) where v is a variable number.

3.7.9 eulervec(n). Returns a vector containing, as rational numbers, the nonzero Euler numbers
E0 , E2 ,. . . , E2n :
? eulervec(5) \\ E_0, E_2..., E_10
%1 = [1, -1, 5, -61, 1385, -50521]
? eulerfrac(10)
%2 = -50521
This routine uses more memory but is a little faster than repeated calls to eulerfrac:
? forstep(n = 2, 8000, 2, eulerfrac(n))
time = 46,851 ms.
? eulervec(4000);
time = 30,588 ms.
The library syntax is GEN eulervec(long n).

3.7.10 fibonacci(x). xth Fibonacci number.

The library syntax is GEN fibo(long x).

3.7.11 hammingweight(x). If x is a t_INT, return the binary Hamming weight of |x|. Otherwise
x must be of type t_POL, t_VEC, t_COL, t_VECSMALL, or t_MAT and the function returns the number
of nonzero coefficients of x.
? hammingweight(15)
%1 = 4
? hammingweight(x^100 + 2*x + 1)
%2 = 3
? hammingweight([Mod(1,2), 2, Mod(0,3)])
%3 = 2
? hammingweight(matid(100))
%4 = 100
The library syntax is long hammingweight(GEN x).

3.7.12 numbpart(n). Gives the number of unrestricted partitions of n, usually called p(n) in the
literature; in other words the number of nonnegative integer solutions to a + 2b + 3c + · · · = n.
n must be of type integer and n < 1015 (with trivial values p(n) = 0 for n < 0 and p(0) = 1).
The algorithm uses the Hardy-Ramanujan-Rademacher formula. To explicitly enumerate them, see
partitions.
The library syntax is GEN numbpart(GEN n).

166
3.7.13 numtoperm(n, k). Generates the k-th permutation (as a row vector of length n) of the
numbers 1 to n. The number k is taken modulo n! , i.e. inverse function of permtonum. The
numbering used is the standard lexicographic ordering, starting at 0.

The library syntax is GEN numtoperm(long n, GEN k).

3.7.14 partitions(k, {a = k}, {n = k}). Returns the vector of partitions of the integer k as a sum
of positive integers (parts); for k < 0, it returns the empty set [], and for k = 0 the trivial partition
(no parts). A partition is given by a t_VECSMALL, where parts are sorted in nondecreasing order:

? partitions(3)
%1 = [Vecsmall([3]), Vecsmall([1, 2]), Vecsmall([1, 1, 1])]

correspond to 3, 1 + 2 and 1 + 1 + 1. The number of (unrestricted) partitions of k is given by

numbpart:

? #partitions(50)
%1 = 204226
? numbpart(50)
%2 = 204226

Optional parameters n and a are as follows:

• n = nmax (resp. n = [nmin, nmax ]) restricts partitions to length less than nmax (resp.
length between nmin and nmax), where the length is the number of nonzero entries.

• a = amax (resp. a = [amin, amax ]) restricts the parts to integers less than amax (resp.
between amin and amax ).

? partitions(4, 2) \\ parts bounded by 2

%1 = [Vecsmall([2, 2]), Vecsmall([1, 1, 2]), Vecsmall([1, 1, 1, 1])]
? partitions(4,, 2) \\ at most 2 parts
%2 = [Vecsmall([4]), Vecsmall([1, 3]), Vecsmall([2, 2])]
? partitions(4,[0,3], 2) \\ at most 2 parts
%3 = [Vecsmall([4]), Vecsmall([1, 3]), Vecsmall([2, 2])]

By default, parts are positive and we remove zero entries unless amin ≤ 0, in which case nmin is
ignored and we fix #X = nmax :

? partitions(4, [0,3]) \\ parts between 0 and 3

%1 = [Vecsmall([0, 0, 1, 3]), Vecsmall([0, 0, 2, 2]),\
Vecsmall([0, 1, 1, 2]), Vecsmall([1, 1, 1, 1])]
? partitions(1, [0,3], [2,4]) \\ no partition with 2 to 4 nonzero parts
%2 = []

The library syntax is GEN partitions(long k, GEN a = NULL, GEN n = NULL).

167
3.7.15 permcycles(x). Given a permutation x on n elements, return the orbits of {1, . . . , n} under
the action of x as cycles.
? permcycles(Vecsmall([1,2,3]))
%1 = [Vecsmall([1]),Vecsmall([2]),Vecsmall([3])]
? permcycles(Vecsmall([2,3,1]))
%2 = [Vecsmall([1,2,3])]
? permcycles(Vecsmall([2,1,3]))
%3 = [Vecsmall([1,2]),Vecsmall([3])]
The library syntax is GEN permcycles(GEN x).

3.7.16 permorder(x). Given a permutation x on n elements, return its order.

? p = Vecsmall([3,1,4,2,5]);
? p^2
%2 = Vecsmall([4,3,2,1,5])
? p^4
%3 = Vecsmall([1,2,3,4,5])
? permorder(p)
%4 = 4
The library syntax is GEN permorder(GEN x).

3.7.17 permsign(x). Given a permutation x on n elements, return its signature.

? p = Vecsmall([3,1,4,2,5]);
? permsign(p)
%2 = -1
? permsign(p^2)
%3 = 1
The library syntax is long permsign(GEN x).

3.7.18 permtonum(x). Given a permutation x on n elements, gives the number k such that
x = numtoperm(n, k), i.e. inverse function of numtoperm. The numbering used is the standard
lexicographic ordering, starting at 0.
The library syntax is GEN permtonum(GEN x).

3.7.19 stirling(n, k, {flag = 1}). Stirling number of the first kind s(n, k) (flag = 1, default) or of
the second kind S(n, k) (flag=2), where n, k are nonnegative integers. The former is (−1)n−k times
the number of permutations of n symbols with exactly k cycles; the latter is the number of ways
of partitioning a set of n elements into k nonempty subsets. Note that if all s(n, k) are needed, it
is much faster to compute
X
s(n, k)xk = x(x − 1) . . . (x − n + 1).
k

Similarly, if a large number of S(n, k) are needed for the same k, one should use

X xk
S(n, k)xn = .
n
(1 − x) . . . (1 − kx)

168
(Should be implemented using a divide and conquer product.) Here are simple variants for n fixed:

/* list of s(n,k), k = 1..n */

vecstirling(n) = Vec( factorback(vector(n-1,i,1-i*’x)) )
/* list of S(n,k), k = 1..n */
vecstirling2(n) =
{ my(Q = x^(n-1), t);
vector(n, i, t = divrem(Q, x-i); Q=t[1]; simplify(t[2]));
}
/* Bell numbers, B_n = B[n+1] = sum(k = 0, n, S(n,k)), n = 0..N */
vecbell(N)=
{ my (B = vector(N+1));
B[1] = B[2] = 1;
for (n = 2, N,
my (C = binomial(n-1));
B[n+1] = sum(k = 1, n, C[k]*B[k]);
); B;
}
The library syntax is GEN stirling(long n, long k, long flag). Also available are GEN
stirling1(ulong n, ulong k) (flag = 1) and GEN stirling2(ulong n, ulong k) (flag = 2).

3.8 Arithmetic functions.

These functions are by definition functions whose natural domain of definition is either Z (or
Z>0 ). The way these functions are used is completely different from transcendental functions in
that there are no automatic type conversions: in general only integers are accepted as arguments.
An integer argument N can be given in the following alternate formats:
• t_MAT: its factorization fa = factor(N ),
• t_VEC: a pair [N , fa] giving both the integer and its factorization.
This allows to compute different arithmetic functions at a given N while factoring the latter
only once.

? N = 10!; faN = factor(N);

? eulerphi(N)
%2 = 829440
? eulerphi(faN)
%3 = 829440
? eulerphi(S = [N, faN])
%4 = 829440
? sigma(S)
%5 = 15334088

169
3.8.1 Arithmetic functions and the factoring engine. All arithmetic functions in the narrow
sense of the word — Euler’s totient function, the Moebius function, the sums over divisors or
powers of divisors etc.— call, after trial division by small primes, the same versatile factoring
machinery described under factorint. It includes Shanks SQUFOF, Pollard Rho, ECM and
MPQS stages, and has an early exit option for the functions moebius and (the integer function
underlying) issquarefree. This machinery relies on a fairly strong probabilistic primality test, see
ispseudoprime, but you may also set
default(factor_proven, 1)
to ensure that all tentative factorizations are fully proven. This should not slow down PARI too
much, unless prime numbers with hundreds of decimal digits occur frequently in your application.

3.8.2 Orders in finite groups and Discrete Logarithm functions.

The following functions compute the order of an element in a finite group: ellorder (the
rational points on an elliptic curve defined over a finite field), fforder (the multiplicative group of
a finite field), znorder (the invertible elements in Z/nZ). The following functions compute discrete
logarithms in the same groups (whenever this is meaningful) elllog, fflog, znlog.
All such functions allow an optional argument specifying an integer N , representing the order
of the group. (The order functions also allows any nonzero multiple of the order, with a minor loss
of efficiency.) That optional argument follows the same format as given above:
• t_INT: the integer N ,
• t_MAT: the factorization fa = factor(N ),
• t_VEC: this is the preferred format and provides both the integer N and its factorization in
a two-component vector [N , fa].
When the group is fixed and many orders or discrete logarithms will be computed, it is much
more efficient to initialize this data once and for all and pass it to the relevant functions, as in
? p = nextprime(10^30);
? v = [p-1, factor(p-1)]; \\ data for discrete log & order computations
? znorder(Mod(2,p), v)
%3 = 500000000000000000000000000028
? g = znprimroot(p);
? znlog(2, g, v)
%5 = 543038070904014908801878611374

3.8.3 Dirichlet characters.

∗
Q Z) can be written G = ⊕i≤n (Z/di Z)gi , with dn | . . . | d2 |
The finite abelian group G = (Z/N
d1 (SNF condition), all di > 0, and i di = φ(N ).
The SNF condition makes the di unique, but the generators gi , of respective order di , are
definitely
Q ni not unique. The ⊕ notation means that all elements of G can be written uniquely as
g
i i where ni ∈ Z/di Z. The gi are the so-called SNF generators of G.
• a character on the abelian group ⊕(Z/dj Z)gj is given by a row vector χ = [a1 , . . . , an ] of
integers 0 ≤ ai < di such that
Q χ(g j) = P
e(aj /dj ) for all j, with the standard notation e(x) :=
nj
exp(2iπx). In other words, χ( gj ) = e( aj nj /dj ).

170
 This will be generalized to more general abelian groups in later sections (Hecke characters),
but in the present case of (Z/N Z)∗ , there is a useful alternate convention : namely, it is Q not
necessary to impose the SNF condition and we can use Chinese remainders instead. If N = pep
is the factorization of N into primes, the so-called Conrey generators of G are the generators of
the (Z/pep Z)∗ lifted to (Z/N Z)∗ by requesting that they be congruent to 1 modulo N/pep (for p
odd we take the smallest positive primitive root mod p2 , and for p = 2 we take −1 Qif e2 > 1 and
additionally 5 if e2 > 2). We can again write G = ⊕i≤n (Z/Di Z)Gi , where again i Di = φ(N ).
These generators don’t satisfy the SNF condition in general since their orders are now (p − 1)pep −1
2 and 5 has order 2e2 −2 (e2 > 2). Nevertheless, any
for p odd; for p = 2, the generator −1 has orderQ
m ∈ (Z/N Z)∗ can be uniquely decomposed as Gm i
i
for some mi modulo Di and we can define a
character by χ(Gj ) = e(mj /Dj ) for all j.
• The column vector of the mj , 0 ≤ mj < Dj is called the Conrey logarithm of m (discrete
logarithm in terms of the Conrey generators). Note that discrete logarithms in PARI/GP are
always expressed as t_COLs.
• The attached character is called the Conrey character attached to m.
To sum up a Dirichlet character can be defined by a t_INT (the Conrey label m), a t_COL
(the Conrey logarithm of m, in terms of the Conrey generators) or a t_VEC (in terms of the SNF
generators). The t_COL format, i.e. Conrey logarithms, is the preferred (fastest) representation.
Concretely, this works as follows:
G = znstar(N, 1) initializes (Z/N Z)∗ , which must be given as first arguments to all functions
handling Dirichlet characters.
znconreychar transforms t_INT and t_COL to a SNF character.
znconreylog transforms t_INT and t_VEC to a Conrey logarithm.
znconreyexp transforms t_VEC and t_COL to a Conrey label.
Also available are charconj, chardiv, charmul, charker, chareval, charorder, zncharin-
duce, znconreyconductor (also computes the primitive character attached to the input character).
The prefix char indicates that the function applies to all characters, the prefix znchar that it is
specific to Dirichlet characters (on (Z/N Z)∗ ) and the prefix znconrey that it is specific to Conrey
representation.

3.8.4 addprimes({x = [ ]}). Adds the integers contained in the vector x (or the single integer x) to
a special table of “user-defined primes”, and returns that table. Whenever factor is subsequently
called, it will trial divide by the elements in this table. If x is empty or omitted, just returns the
current list of extra primes.
? addprimes(37975227936943673922808872755445627854565536638199)
? factor(15226050279225333605356183781326374297180681149613806\
88657908494580122963258952897654000350692006139)
%2 =
[37975227936943673922808872755445627854565536638199 1]
[40094690950920881030683735292761468389214899724061 1]
? ##
*** last result computed in 0 ms.
The entries in x must be primes: there is no internal check, even if the factor_proven default
is set. To remove primes from the list use removeprimes.

171
 The library syntax is GEN addprimes(GEN x = NULL).

3.8.5 bestappr(x, {B}). Using variants of the extended Euclidean algorithm, returns a rational
approximation a/b to x, whose denominator is limited by B, if present. If B is omitted, returns the
best approximation affordable given the input accuracy; if you are looking for true rational numbers,
presumably approximated to sufficient accuracy, you should first try that option. Otherwise, B
must be a positive real scalar (impose 0 < b ≤ B).
• If x is a t_REAL or a t_FRAC, this function uses continued fractions.
? bestappr(Pi, 100)
%1 = 22/7
? bestappr(0.1428571428571428571428571429)
%2 = 1/7
? bestappr([Pi, sqrt(2) + ’x], 10^3)
%3 = [355/113, x + 1393/985]
By definition, a/b is the best rational approximation to x if |bx − a| < |vx − u| for all integers
(u, v) with 0 < v ≤ B. (Which implies that n/d is a convergent of the continued fraction of x.)
• If x is a t_INTMOD modulo N or a t_PADIC of precision N = pk , this function performs
rational modular reconstruction modulo N . The routine then returns the unique rational number
a/b in coprime integers |a| < N/2B andp b ≤ B which is congruent to x modulo N . Omitting B
amounts to choosing it of the order of N/2. If rational reconstruction is not possible (no suitable
a/b exists), returns [].
? bestappr(Mod(18526731858, 11^10))
%1 = 1/7
? bestappr(Mod(18526731858, 11^20))
%2 = []
? bestappr(3 + 5 + 3*5^2 + 5^3 + 3*5^4 + 5^5 + 3*5^6 + O(5^7))
%2 = -1/3
In most concrete uses, B is a prime power and we performed Hensel lifting to obtain x.
The function applies recursively to components of complex objects (polynomials, vectors, . . . ).
If rational reconstruction fails for even a single entry, returns [].
The library syntax is GEN bestappr(GEN x, GEN B = NULL).

3.8.6 bestapprPade(x, {B}). Using variants of the extended Euclidean algorithm (Padé approx-
imants), returns a rational function approximation a/b to x, whose denominator is limited by B, if
present. If B is omitted, return the best approximation affordable given the input accuracy; if you
are looking for true rational functions, presumably approximated to sufficient accuracy, you should
first try that option. Otherwise, B must be a nonnegative real (impose 0 ≤ degree(b) ≤ B).
• If x is a t_POLMOD modulo N this function performs rational modular reconstruction mod-
ulo N . The routine then returns the unique rational function a/b in coprime polynomials, with
degree(b) ≤ B and degree(a) minimal, which is congruent to x modulo N . Omitting B amounts to
choosing it equal to the floor of degree(N )/2. If rational reconstruction is not possible (no suitable
a/b exists), returns [].
? T = Mod(x^3 + x^2 + x + 3, x^4 - 2);
? bestapprPade(T)

172
 %2 = (2*x - 1)/(x - 1)
? U = Mod(1 + x + x^2 + x^3 + x^5, x^9);
? bestapprPade(U) \\ internally chooses B = 4
%3 = []
? bestapprPade(U, 5) \\ with B = 5, a solution exists
%4 = (2*x^4 + x^3 - x - 1)/(-x^5 + x^3 + x^2 - 1)

• If x is a t_SER, we implicitly convert the input to a t_POLMOD modulo N = tk where k is the

series absolute precision.

? T = 1 + t + t^2 + t^3 + t^4 + t^5 + t^6 + O(t^7); \\ mod t^7

? bestapprPade(T)
%1 = 1/(-t + 1)

• If x is a t_RFRAC, we implicitly convert the input to a t_POLMOD modulo N = tk where

k = 2B + 1. If B was omitted, we return x:

? T = (4*t^2 + 2*t + 3)/(t+1)^10;

? bestapprPade(T,1)
%2 = [] \\ impossible
? bestapprPade(T,2)
%3 = 27/(337*t^2 + 84*t + 9)
? bestapprPade(T,3)
%4 = (4253*t - 3345)/(-39007*t^3 - 28519*t^2 - 8989*t - 1115)

The function applies recursively to components of complex objects (polynomials, vectors, . . . ). If

rational reconstruction fails for even a single entry, return [].

The library syntax is GEN bestapprPade(GEN x, long B).

3.8.7 bezout(x, y). Deprecated alias for gcdext

The library syntax is GEN gcdext0(GEN x, GEN y).

3.8.8 bigomega(x). Number of prime divisors of the integer |x| counted with multiplicity:

? factor(392)
%1 =
[2 3]
[7 2]
? bigomega(392)
%2 = 5; \\ = 3+2
? omega(392)
%3 = 2; \\ without multiplicity

The library syntax is long bigomega(GEN x).

173
3.8.9 charconj(cyc,Pchi ). Let cyc represent a finite abelian group by its elementary divisors,
i.e. (dj ) represents j≤k Z/dj Z with dk | . . . | d1 ; any object which has a .cyc method is also
allowed, e.g. the output of znstar Q nor bnrinit. APcharacter on this group is given by a row vector
χ = [a1 , . . . , an ] such that χ( gj j ) = exp(2πi aj nj /dj ), where gj denotes the generator (of
order dj ) of the j-th cyclic component.
This function returns the conjugate character.
? cyc = [15,5]; chi = [1,1];
? charconj(cyc, chi)
%2 = [14, 4]
? bnf = bnfinit(x^2+23);
? bnf.cyc
%4 = [3]
? charconj(bnf, [1])
%5 = [2]
For Dirichlet characters (when cyc is znstar(q,1)), characters in Conrey representation are avail-
able, see Section 3.8.3 or ??character:
? G = znstar(8, 1); \\ (Z/8Z)^*
? charorder(G, 3) \\ Conrey label
%2 = 2
? chi = znconreylog(G, 3);
? charorder(G, chi) \\ Conrey logarithm
%4 = 2
The library syntax is GEN charconj0(GEN cyc, GEN chi). Also available is GEN char-
conj(GEN cyc, GEN chi), when cyc is known to be a vector of elementary divisors and chi a
compatible character (no checks).

3.8.10 chardiv(cyc,Pa, b). Let cyc represent a finite abelian group by its elementary divisors,
i.e. (dj ) represents j≤k Z/dj Z with dk | . . . | d1 ; any object which has a .cyc method is also
allowed, e.g. the output of znstar Q nj or bnrinit.PA character on this group is given by a row vector
a = [a1 , . . . , an ] such that χ( gj ) = exp(2πi aj nj /dj ), where gj denotes the generator (of order
dj ) of the j-th cyclic component.
Given two characters a and b, return the character a/b = ab.
? cyc = [15,5]; a = [1,1]; b = [2,4];
? chardiv(cyc, a,b)
%2 = [14, 2]
? bnf = bnfinit(x^2+23);
? bnf.cyc
%4 = [3]
? chardiv(bnf, [1], [2])
%5 = [2]
For Dirichlet characters on (Z/N Z)∗ , additional representations are available (Conrey labels, Con-
rey logarithm), see Section 3.8.3 or ??character. If the two characters are in the same format, the
result is given in the same format, otherwise a Conrey logarithm is used.
? G = znstar(100, 1);

174
 ? G.cyc
%2 = [20, 2]
? a = [10, 1]; \\ usual representation for characters
? b = 7; \\ Conrey label;
? c = znconreylog(G, 11); \\ Conrey log
? chardiv(G, b,b)
%6 = 1 \\ Conrey label
? chardiv(G, a,b)
%7 = [0, 5]~ \\ Conrey log
? chardiv(G, a,c)
%7 = [0, 14]~ \\ Conrey log
The library syntax is GEN chardiv0(GEN cyc, GEN a, GEN b). Also available is GEN
chardiv(GEN cyc, GEN a, GEN b), when cyc is known to be a vector of elementary divisors
and a, b are compatible characters (no checks).

3.8.11 chareval(G, chi , x, {z}). Let G be an abelian group structure affording a discrete logarithm
method, e.g G = znstar(N, 1) for (Z/N Z)∗ or a bnr structure, let x be an element of G and let
chi be a character of G (see the note below for details). This function returns the value of chi at
x.
Note on characters. Let K be some field. If G is an abelian group, let χ : G → K ∗ be a character
of finite order and let o be a multiple of the character order such that χ(n) = ζ c(n) for some fixed
ζ ∈ K ∗ of multiplicative order o and a unique morphism c : G → (Z/oZ, +). Our usual convention
is to write
G = (Z/o1 Z)g1 ⊕ · · · ⊕ (Z/od Z)gd
for some generators (gi ) of Q
respective order di , where the group has exponent o := lcmi oi . Since
ζ o = 1, the vector (ci ) in (Z/oi Z) defines a character χ on G via χ(gi ) = ζ ci (o/oi ) for all i.
Classical Dirichlet characters have values in K = C and we can take ζ = exp(2iπ/o).
Note on Dirichlet characters. In the special case where bid is attached to G = (Z/qZ)∗ (as
per G = znstar(q,1)), the Dirichlet character chi can be written in one of the usual 3 formats:
a t_VEC in terms of bid.gen as above, a t_COL in terms of the Conrey generators, or a t_INT
(Conrey label); see Section 3.8.3 or ??character.
The character value is encoded as follows, depending on the optional argument z:
• If z is omitted: return the rational number c(x)/o for x coprime to q, where we normalize
0 ≤ c(x) < o. If x can not be mapped to the group (e.g. x is not coprime to the conductor of a
Dirichlet or Hecke character) we return the sentinel value −1.
• If z is an integer o, then we assume that o is a multiple of the character order and we return
the integer c(x) when x belongs to the group, and the sentinel value −1 otherwise.
• z can be of the form [zeta, o], where zeta is an o-th root of 1 and o is a multiple of the character
order. We return ζ c(x) if x belongs to the group, and the sentinel value 0 otherwise. (Note that this
coincides with the usual extension of Dirichlet characters to Z, or of Hecke characters to general
ideals.)
• Finally, z can be of the form [vzeta, o], where vzeta is a vector of powers ζ 0 , . . . , ζ o−1 of some
o-th root of 1 and o is a multiple of the character order. As above, we return ζ c(x) after a table
lookup. Or the sentinel value 0.
The library syntax is GEN chareval(GEN G, GEN chi, GEN x, GEN z = NULL).

175
3.8.12 chargalois(cyc, {ORD}). Let cyc represent a finite abelian group by its elementary divisors
(any object which has a .cyc method is also allowed, i.e. the output of znstar or bnrinit). Return
a list of representatives for the Galois orbits of complex characters of G. If ORD is present, select
characters depending on their orders:
• if ORD is a t_INT, restrict to orders less than this bound;
• if ORD is a t_VEC or t_VECSMALL, restrict to orders in the list.
? G = znstar(96);
? #chargalois(G) \\ 16 orbits of characters mod 96
%2 = 16
? #chargalois(G,4) \\ order less than 4
%3 = 12
? chargalois(G,[1,4]) \\ order 1 or 4; 5 orbits
%4 = [[0, 0, 0], [2, 0, 0], [2, 1, 0], [2, 0, 1], [2, 1, 1]]
Given a character χ, of order n (charorder(G,chi)), the elements in its orbit are the φ(n) char-
acters χi , (i, n) = 1.
The library syntax is GEN chargalois(GEN cyc, GEN ORD = NULL).

3.8.13 charker(cyc,Pchi ). Let cyc represent a finite abelian group by its elementary divisors,
i.e. (dj ) represents j≤k Z/dj Z with dk | . . . | d1 ; any object which has a .cyc method is also
allowed, e.g. the output of znstar Q nor bnrinit. APcharacter on this group is given by a row vector
χ = [a1 , . . . , an ] such that χ( gj j ) = exp(2πi aj nj /dj ), where gj denotes the generator (of
order dj ) of the j-th cyclic component.
This function returns the kernel of χ, as a matrix K in HNF which is a left-divisor of matdiag-
onal(d). Its columns express in terms of the gj the generators of the subgroup. The determinant
of K is the kernel index.
? cyc = [15,5]; chi = [1,1];
? charker(cyc, chi)
%2 =
[15 12]
[ 0 1]
? bnf = bnfinit(x^2+23);
? bnf.cyc
%4 = [3]
? charker(bnf, [1])
%5 =
[3]
Note that for Dirichlet characters (when cyc is znstar(q, 1)), characters in Conrey representation
are available, see Section 3.8.3 or ??character.
? G = znstar(8, 1); \\ (Z/8Z)^*
? charker(G, 1) \\ Conrey label for trivial character
%2 =
[1 0]
[0 1]

176
 The library syntax is GEN charker0(GEN cyc, GEN chi). Also available is GEN charker(GEN
cyc, GEN chi), when cyc is known to be a vector of elementary divisors and chi a compatible
character (no checks).

3.8.14 charmul(cyc, Pa, b). Let cyc represent a finite abelian group by its elementary divisors,
i.e. (dj ) represents j≤k Z/dj Z with dk | . . . | d1 ; any object which has a .cyc method is also
allowed, e.g. the output of znstar Q n or bnrinit.PA character on this group is given by a row vector
a = [a1 , . . . , an ] such that χ( gj j ) = exp(2πi aj nj /dj ), where gj denotes the generator (of order
dj ) of the j-th cyclic component.
Given two characters a and b, return the product character ab.
? cyc = [15,5]; a = [1,1]; b = [2,4];
? charmul(cyc, a,b)
%2 = [3, 0]
? bnf = bnfinit(x^2+23);
? bnf.cyc
%4 = [3]
? charmul(bnf, [1], [2])
%5 = [0]
For Dirichlet characters on (Z/N Z)∗ , additional representations are available (Conrey labels, Con-
rey logarithm), see Section 3.8.3 or ??character. If the two characters are in the same format,
their product is given in the same format, otherwise a Conrey logarithm is used.
? G = znstar(100, 1);
? G.cyc
%2 = [20, 2]
? a = [10, 1]; \\ usual representation for characters
? b = 7; \\ Conrey label;
? c = znconreylog(G, 11); \\ Conrey log
? charmul(G, b,b)
%6 = 49 \\ Conrey label
? charmul(G, a,b)
%7 = [0, 15]~ \\ Conrey log
? charmul(G, a,c)
%7 = [0, 6]~ \\ Conrey log
The library syntax is GEN charmul0(GEN cyc, GEN a, GEN b). Also available is GEN char-
mul(GEN cyc, GEN a, GEN b), when cyc is known to be a vector of elementary divisors and a, b
are compatible characters (no checks).

3.8.15 charorder(cyc, P chi ). Let cyc represent a finite abelian group by its elementary divisors,
i.e. (dj ) represents j≤k Z/dj Z with dk | . . . | d1 ; any object which has a .cyc method is also
allowed, e.g. the output of znstar Q nor bnrinit. AP character on this group is given by a row vector
χ = [a1 , . . . , an ] such that χ( gj j ) = exp(2πi aj nj /dj ), where gj denotes the generator (of
order dj ) of the j-th cyclic component.
This function returns the order of the character chi.
? cyc = [15,5]; chi = [1,1];
? charorder(cyc, chi)

177
 %2 = 15
? bnf = bnfinit(x^2+23);
? bnf.cyc
%4 = [3]
? charorder(bnf, [1])
%5 = 3
For Dirichlet characters (when cyc is znstar(q, 1)), characters in Conrey representation are
available, see Section 3.8.3 or ??character:
? G = znstar(100, 1); \\ (Z/100Z)^*
? charorder(G, 7) \\ Conrey label
%2 = 4
The library syntax is GEN charorder0(GEN cyc, GEN chi). Also available is GEN
charorder(GEN cyc, GEN chi), when cyc is known to be a vector of elementary divisors and
chi a compatible character (no checks).

3.8.16 charpow(cyc, Pa, n). Let cyc represent a finite abelian group by its elementary divisors,
i.e. (dj ) represents j≤k Z/dj Z with dk | . . . | d1 ; any object which has a .cyc method is also
allowed, e.g. the output of znstar Q n or bnrinit.PA character on this group is given by a row vector
a = [a1 , . . . , an ] such that χ( gj j ) = exp(2πi aj nj /dj ), where gj denotes the generator (of order
dj ) of the j-th cyclic component.
Given n ∈ Z and a character a, return the character an .
? cyc = [15,5]; a = [1,1];
? charpow(cyc, a, 3)
%2 = [3, 3]
? charpow(cyc, a, 5)
%2 = [5, 0]
? bnf = bnfinit(x^2+23);
? bnf.cyc
%4 = [3]
? charpow(bnf, [1], 3)
%5 = [0]
For Dirichlet characters on (Z/N Z)∗ , additional representations are available (Conrey labels, Con-
rey logarithm), see Section 3.8.3 or ??character and the output uses the same format as the
input.
? G = znstar(100, 1);
? G.cyc
%2 = [20, 2]
? a = [10, 1]; \\ standard representation for characters
? b = 7; \\ Conrey label;
? c = znconreylog(G, 11); \\ Conrey log
? charpow(G, a,3)
%6 = [10, 1] \\ standard representation
? charpow(G, b,3)
%7 = 43 \\ Conrey label
? charpow(G, c,3)

178
 %8 = [1, 8]~ \\ Conrey log
The library syntax is GEN charpow0(GEN cyc, GEN a, GEN n). Also available is GEN char-
pow(GEN cyc, GEN a, GEN n), when cyc is known to be a vector of elementary divisors (no
check).

3.8.17 chinese(x, {y}). If x and y are both intmods or both polmods, creates (with the same
type) a z in the same residue class as x and in the same residue class as y, if it is possible.
? chinese(Mod(1,2), Mod(2,3))
%1 = Mod(5, 6)
? chinese(Mod(x,x^2-1), Mod(x+1,x^2+1))
%2 = Mod(-1/2*x^2 + x + 1/2, x^4 - 1)
This function also allows vector and matrix arguments, in which case the operation is recursively
applied to each component of the vector or matrix.
? chinese([Mod(1,2),Mod(1,3)], [Mod(1,5),Mod(2,7)])
%3 = [Mod(1, 10), Mod(16, 21)]
For polynomial arguments in the same variable, the function is applied to each coefficient; if the
polynomials have different degrees, the high degree terms are copied verbatim in the result, as if
the missing high degree terms in the polynomial of lowest degree had been Mod(0,1). Since the
latter behavior is usually not the desired one, we propose to convert the polynomials to vectors of
the same length first:
? P = x+1; Q = x^2+2*x+1;
? chinese(P*Mod(1,2), Q*Mod(1,3))
%4 = Mod(1, 3)*x^2 + Mod(5, 6)*x + Mod(3, 6)
? chinese(Vec(P,3)*Mod(1,2), Vec(Q,3)*Mod(1,3))
%5 = [Mod(1, 6), Mod(5, 6), Mod(4, 6)]
? Pol(%)
%6 = Mod(1, 6)*x^2 + Mod(5, 6)*x + Mod(4, 6)
If y is omitted, and x is a vector, chinese is applied recursively to the components of x,
yielding a residue belonging to the same class as all components of x.
Finally chinese(x, x) = x regardless of the type of x; this allows vector arguments to contain
other data, so long as they are identical in both vectors.
The library syntax is GEN chinese(GEN x, GEN y = NULL). GEN chinese1(GEN x) is also
available.

3.8.18 content(x, {D}). Computes the gcd of all the coefficients of x, when this gcd makes
sense. This is the natural definition if x is a polynomial (and by extension a power series) or a
vector/matrix. This is in general a weaker notion than the ideal generated by the coefficients:
? content(2*x+y)
%1 = 1 \\ = gcd(2,y) over Q[y]
If x is a scalar, this simply returns the absolute value of x if x is rational (t_INT or t_FRAC),
and either 1 (inexact input) or x (exact input) otherwise; the result should be identical to gcd(x,
0).

179
 The content of a rational function is the ratio of the contents of the numerator and the de-
nominator. In recursive structures, if a matrix or vector coefficient x appears, the gcd is taken not
with x, but with its content:
? content([ [2], 4*matid(3) ])
%1 = 2
The content of a t_VECSMALL is computed assuming the entries are signed integers.
The optional argument D allows to control over which ring we compute and get a more pre-
dictable behaviour:
• 1: we only consider the underlying Q-structure and the denominator is a (positive) rational
number
• a simple variable, say ’x: all entries are considered as rational functions in K(x) for some
field K and the content is an element of K.
? f = x + 1/y + 1/2;
? content(f) \\ as a t_POL in x
%2 = 1/(2*y)
? content(f, 1) \\ Q-content
%3 = 1/2
? content(f, y) \\ as a rational function in y
%4 = 1/2
? g = x^2*y + y^2*x;
? content(g, x)
%6 = y
? content(g, y)
%7 = x
The library syntax is GEN content0(GEN x, GEN D = NULL).

3.8.19 contfrac(x, {b}, {nmax }). Returns the row vector whose components are the partial quo-
tients of the continued fraction expansion of x. In other words, a result [a0 , . . . , an ] means that
x ≈ a0 + 1/(a1 + . . . + 1/an ). The output is normalized so that an 6= 1 (unless we also have n = 0).
The number of partial quotients n + 1 is limited by nmax. If nmax is omitted, the expansion
stops at the last significant partial quotient.
? \p19
realprecision = 19 significant digits
? contfrac(Pi)
%1 = [3, 7, 15, 1, 292, 1, 1, 1, 2, 1, 3, 1, 14, 2, 1, 1, 2, 2]
? contfrac(Pi,, 3) \\ n = 2
%2 = [3, 7, 15]
x can also be a rational function or a power series.
If a vector b is supplied, the numerators are equal to the coefficients of b, instead of all equal to
1 as above; more precisely, x ≈ (1/b0 )(a0 +b1 /(a1 +. . .+bn /an )); for a numerical continued fraction
(x real), the ai are integers, as large as possible; if x is a rational function, they are polynomials
with deg ai = deg bi + 1. The length of the result is then equal to the length of b, unless the next
partial quotient cannot be reliably computed, in which case the expansion stops. This happens

180
when a partial remainder is equal to zero (or too small compared to the available significant digits
for x a t_REAL).
A direct implementation of the numerical continued fraction contfrac(x,b) described above
would be
\\ "greedy" generalized continued fraction
cf(x, b) =
{ my( a= vector(#b), t );
x *= b[1];
for (i = 1, #b,
a[i] = floor(x);
t = x - a[i]; if (!t || i == #b, break);
x = b[i+1] / t;
); a;
}
There is some degree of freedom when choosing the ai ; the program above can easily be modified to
derive variants of the standard algorithm. In the same vein, although no builtin function implements
the related Engel expansion (a special kind of Egyptian fraction decomposition: x = 1/a1 +
1/(a1 a2 ) + . . . ), it can be obtained as follows:
\\ n terms of the Engel expansion of x
engel(x, n = 10) =
{ my( u = x, a = vector(n) );
for (k = 1, n,
a[k] = ceil(1/u);
u = u*a[k] - 1;
if (!u, break);
); a
}

Obsolete hack. (don’t use this): if b is an integer, nmax is ignored and the command is understood
as contfrac(x, , b).
The library syntax is GEN contfrac0(GEN x, GEN b = NULL, long nmax). Also available
are GEN gboundcf(GEN x, long nmax), GEN gcf(GEN x) and GEN gcf2(GEN b, GEN x).

3.8.20 contfracpnqn(x, {n = −1}). When x is a vector or a one-row matrix, x is considered as

the list of partial quotients [a0 , a1 , . . . , an ] of a rational number, and the result is the 2 by 2 matrix
[pn , pn−1 ; qn , qn−1 ] in the standard notation of continued fractions, so pn /qn = a0 + 1/(a1 + . . . +
1/an ). If x is a matrix with two rows [b0 , b1 , . . . , bn ] and [a0 , a1 , . . . , an ], this is then considered as
a generalized continued fraction and we have similarly pn /qn = (1/b0 )(a0 + b1 /(a1 + . . . + bn /an )).
Note that in this case one usually has b0 = 1.
If n ≥ 0 is present, returns all convergents from p0 /q0 up to pn /qn . (All convergents if x is too
small to compute the n + 1 requested convergents.)
? a = contfrac(Pi,10)
%1 = [3, 7, 15, 1, 292, 1, 1, 1, 3]
? allpnqn(x) = contfracpnqn(x,#x) \\ all convergents
? allpnqn(a)

181
 %3 =
[3 22 333 355 103993 104348 208341 312689 1146408]
[1 7 106 113 33102 33215 66317 99532 364913]
? contfracpnqn(a) \\ last two convergents
%4 =
[1146408 312689]
[ 364913 99532]
? contfracpnqn(a,3) \\ first three convergents
%5 =
[3 22 333 355]
[1 7 106 113]

The library syntax is GEN contfracpnqn(GEN x, long n). also available is GEN pnqn(GEN x)
for n = −1.

3.8.21 core(n, {flag = 0}). If n is an integer written as n = df 2 with d squarefree, returns d. If

flag is nonzero, returns the two-element row vector [d, f ]. By convention, we write 0 = 0 × 12 , so
core(0, 1) returns [0, 1].

The library syntax is GEN core0(GEN n, long flag). Also available are GEN core(GEN n)
(flag = 0) and GEN core2(GEN n) (flag = 1)

3.8.22 coredisc(n, {flag = 0}). A fundamental discriminant is an integer of the form t ≡ 1 mod 4
or 4t ≡ 8, 12 mod 16, with t squarefree (i.e. 1 or the discriminant of a quadratic number field).
Given a nonzero integer n, this routine returns the (unique) fundamental discriminant d such that
n = df 2 , f a positive rational number. If flag is nonzero, returns the two-element row vector [d, f ].
If n is congruent to 0 or 1 modulo 4, f is an integer, and a half-integer otherwise.

By convention, coredisc(0, 1)) returns [0, 1].

Note that quaddisc(n) returns the same value as coredisc(n), and also works with rational
inputs n ∈ Q∗ .

The library syntax is GEN coredisc0(GEN n, long flag). Also available are GEN core-
disc(GEN n) (flag = 0) and GEN coredisc2(GEN n) (flag = 1)

3.8.23 dirdiv(x, y). x and y being vectors of perhaps different lengths but with y[1] 6= 0 considered
as Dirichlet series, computes the quotient of x by y, again as a vector.

The library syntax is GEN dirdiv(GEN x, GEN y).

182
3.8.24 direuler(p = a, b, expr , {c}). Computes the Dirichlet series attached to the Euler product
of expression expr as p ranges through the primes from a to b. expr must be a polynomial or
rational function in another variable than p (say X) and expr (X) is understood as the local factor
expr (p−s ).
The series is output as a vector of coefficients. If c is omitted, output the first b coefficients of
the series; otherwise, output the first c coefficients. The following command computes the sigma
function, attached to ζ(s)ζ(s − 1):
? direuler(p=2, 10, 1/((1-X)*(1-p*X)))
%1 = [1, 3, 4, 7, 6, 12, 8, 15, 13, 18]
? direuler(p=2, 10, 1/((1-X)*(1-p*X)), 5) \\ fewer terms
%2 = [1, 3, 4, 7, 6]
Setting c < b is useless (the same effect would be achieved by setting b = c). If c > b, the computed
coefficients are “missing” Euler factors:
? direuler(p=2, 10, 1/((1-X)*(1-p*X)), 15) \\ more terms, no longer = sigma !
%3 = [1, 3, 4, 7, 6, 12, 8, 15, 13, 18, 0, 28, 0, 24, 24]
The library syntax is direuler(void *E, GEN (*eval)(void*,GEN), GEN a, GEN b)

3.8.25Pdirmul(x, y). P
x and y being vectors of perhaps different lengths representing the Dirichlet
series n xn n−s and n yn n−s , computes the product of x by y, again as a vector.
? dirmul(vector(10,n,1), vector(10,n,moebius(n)))
%1 = [1, 0, 0, 0, 0, 0, 0, 0, 0, 0]
The product length is the minimum of #x∗v(y) and #y∗v(x), where v(x) is the index of the first
nonzero coefficient.
? dirmul([0,1], [0,1]);
%2 = [0, 0, 0, 1]
The library syntax is GEN dirmul(GEN x, GEN y).

x
3.8.26 dirpowerssum(n, x). For positive integer n and complex number x, return the sum 1√ +
x x
2 + . . . + n . This is the same as vecsum(dirpowers(n,x)), but faster and using only O( n)
memory instead of O(n).
? dirpowers(5, 2)
%1 = [1, 4, 9, 16, 25]
? vecsum(%)
%2 = 55
? dirpowerssum(5, 2)
%3 = 55
? \p200
? dirpowerssum(10^7, 1/2 + I * sqrt(3));
time = 29,884 ms.
? vecsum(dirpowers(10^7, 1/2 + I * sqrt(3)))
time = 41,894 ms.

The penultimate command works with default stack size, the last one requires a stacksize of at
least 5GB.

183
 When n ≤ 0, the function returns 0.

The library syntax is GEN dirpowerssum0(GEN n, GEN x, long prec). Also available is GEN
dirpowerssum(ulong n, GEN x, long prec).

3.8.27 divisors(x, {flag = 0}). Creates a row vector whose components are the divisors of x. The
factorization of x (as output by factor) can be used instead. If flag = 1, return pairs [d, factor(d)].

By definition, these divisors are the products of the irreducible factors of n, as produced by
factor(n), raised to appropriate powers (no negative exponent may occur in the factorization). If
n is an integer, they are the positive divisors, in increasing order.

? divisors(12)
%1 = [1, 2, 3, 4, 6, 12]
? divisors(12, 1) \\ include their factorization
%2 = [[1, matrix(0,2)], [2, Mat([2, 1])], [3, Mat([3, 1])],
[4, Mat([2, 2])], [6, [2, 1; 3, 1]], [12, [2, 2; 3, 1]]]
? divisors(x^4 + 2*x^3 + x^2) \\ also works for polynomials
%3 = [1, x, x^2, x + 1, x^2 + x, x^3 + x^2, x^2 + 2*x + 1,
x^3 + 2*x^2 + x, x^4 + 2*x^3 + x^2]

This function requires a lot of memory if x has many divisors. The following idiom runs
through all divisors using very little memory, in no particular order this time:

F = factor(x); P = F[,1]; E = F[,2];

forvec(e = vectorv(#E,i,[0,E[i]]), d = factorback(P,e); ...)

If the factorization of d is also desired, then [P, e] almost provides it but not quite: e may
contain 0 exponents, which are not allowed in factorizations. These must be sieved out as in:

tofact(P,E) =
my(v = select(x->x, E, 1)); Mat([vecextract(P,v), vecextract(E,v)]);
? tofact([2,3,5,7]~, [4,0,2,0]~)
%4 =
[2 4]
[5 2]

We can then run the above loop with tofact(P,e) instead of, or together with, factorback.

The library syntax is GEN divisors0(GEN x, long flag). The functions GEN divisors(GEN
N) (flag = 0) and GEN divisors_factored(GEN N) (flag = 1) are also available.

184
3.8.28 divisorslenstra(N, r, s). Given three integers N > s > r ≥ 0 such that (r, s) = 1 and
s3 > N , find all divisors d of N such that d ≡ r (mod s). There are at most 11 such divisors
(Lenstra).
? N = 245784; r = 19; s = 65 ;
? divisorslenstra(N, r, s)
%2 = [19, 84, 539, 1254, 3724, 245784]
? [ d | d <- divisors(N), d % s == r]
%3 = [19, 84, 539, 1254, 3724, 245784]
When the preconditions are not met, the result is undefined:
? N = 4484075232; r = 7; s = 1303; s^3 > N
%4 = 0
? divisorslenstra(N, r, s)
? [ d | d <- divisors(N), d % s == r ]
%6 = [7, 2613, 9128, 19552, 264516, 3407352, 344928864]
(Divisors were missing but s3 < N .)
The library syntax is GEN divisorslenstra(GEN N, GEN r, GEN s).

3.8.29 eulerphi(x). Euler’s φ (totient) function of the integer |x|, in other words |(Z/xZ)∗ |.
? eulerphi(40)
%1 = 16
According to this definition we let φ(0) := 2, since Z∗ = {−1, 1}; this is consistent with znstar(0):
we have znstar(n).no = eulerphi(n) for all n ∈ Z.
The library syntax is GEN eulerphi(GEN x).

3.8.30 factor(x, {D}). Factor x over domain D; if D is omitted, it is determined from x. For
instance, if x is an integer, it is factored in Z, if it is a polynomial with rational coefficients, it is
factored in Q[x], etc., see below for details. The result is a two-column matrix: the first contains
the irreducibles dividing x (rational or Gaussian primes, irreducible polynomials), and the second
the exponents. By convention, 0 is factored as 01 .

x ∈ Q. See factorint for the algorithms used. The factorization includes the unit −1 when x < 0
and all other factors are positive; a denominator is factored with negative exponents. The factors
are sorted in increasing order.
? factor(-7/106)
%1 =
[-1 1]
[ 2 -1]
[ 7 1]
[53 -1]
By convention, 1 is factored as matrix(0,2) (the empty factorization, printed as [;]).
Large rational “primes” > 264 in the factorization are in fact pseudoprimes (see ispseudo-
prime), a priori not rigorously proven primes. Use isprime to prove primality of these factors, as
in

185
 ? fa = factor(2^2^7 + 1)
%2 =
[59649589127497217 1]
[5704689200685129054721 1]
? isprime( fa[,1] )
%3 = [1, 1]~ \\ both entries are proven primes
Another possibility is to globally set the default factor_proven, which will perform a rigorous
primality proof for each pseudoprime factor but will slow down PARI.
A t_INT argument D can be added, meaning that we only trial divide by all primes p < D
and the addprimes entries, then skip all expensive factorization methods. The limit D must be
nonnegative. In this case, one entry in the factorization may be a composite number: all factors
less than D2 and primes from the addprimes table are actual primes. But (at most) one entry may
not verify this criterion, and it may be prime or composite: it is only known to be coprime to all
other entries and not a pure power..
? factor(2^2^7 +1, 10^5)
%4 =
[340282366920938463463374607431768211457 1]

Deprecated feature. Setting D = 0 is the same as setting it to primelimit + 1.

This routine uses trial division and perfect power tests, and should not be used for huge
values of D (at most 109 , say): factorint(, 1 + 8) will in general be faster. The latter does
not guarantee that all small prime factors are found, but it also finds larger factors and in a more
efficient way.
? F = (2^2^7 + 1) * 1009 * (10^5+3); factor(F, 10^5) \\ fast, incomplete
time = 0 ms.
%5 =
[1009 1]
[34029257539194609161727850866999116450334371 1]
? factor(F, 10^9) \\ slow
time = 3,260 ms.
%6 =
[1009 1]
[100003 1]
[340282366920938463463374607431768211457 1]
? factorint(F, 1+8) \\ much faster and all small primes were found
time = 8 ms.
%7 =
[1009 1]
[100003 1]
[340282366920938463463374607431768211457 1]
? factor(F) \\ complete factorization
time = 60 ms.

186
 %8 =
[1009 1]
[100003 1]
[59649589127497217 1]
[5704689200685129054721 1]

Setting D = I will factor in the Gaussian integers Z[i]:

x ∈ Q(i). The factorization is performed with Gaussian primes in Z[i] and includes Gaussian units
in {±1, ±i}; factors are sorted by increasing norm. Except for a possible leading unit, the Gaussian
factors are normalized: rational factors are positive and irrational factors have positive imaginary
part (a canonical represneta.
Unless factor_proven is set, large factors are actually pseudoprimes, not proven primes; a
rational factor is prime if less than 264 and an irrational one if its norm is less than 264 .
? factor(5*I)
%9 =
[ 2 + I 1]
[1 + 2*I 1]
One can force the factorization of a rational number by setting the domain D = I:
? factor(-5, I)
%10 =
[ I 1]
[ 2 + I 1]
[1 + 2*I 1]
? factorback(%)
%11 = -5

Univariate polynomials and rational functions. PARI can factor univariate polynomials in
K[t]. The following base fields K are currently supported: Q, R, C, Qp , finite fields and number
fields. See factormod and factorff for the algorithms used over finite fields and nffactor for
the algorithms over number fields. The irreducible factors are sorted by increasing degree and
normalized: they are monic except when K = Q where they are primitive in Z[t].
The content is not included in the factorization, in particular factorback will in general
recover the original x only up to multiplication by an element of K ∗ : when K 6= Q, this scalar
is pollead(x) (since irreducible factors are monic); and when K = Q you can either ask for the
Q-content explicitly of use factorback:
? P = t^2 + 5*t/2 + 1; F = factor(P)
%12 =
[t + 2 1]
[2*t + 1 1]
? content(P, 1) \\ Q-content
%13 = 1/2
? pollead(factorback(F)) / pollead(P)

187
 %14 = 2
You can specify K using the optional “domain” argument D as follows
• K = Q : D a rational number (t_INT or t_FRAC),
• K = Z/pZ with p prime : D a t_INTMOD modulo p; factoring modulo a composite number
is not supported.
• K = Fq : D a t_FFELT encoding the finite field; you can also use a t_POLMOD of t_INTMOD
modulo a prime p but this is usualy less convenient;
• K = Q[X]/(T ) a number field : D a t_POLMOD modulo T ,
• K = Q(i) (alternate syntax for special case): D = I,
• K = Q(w) a quadratic number field (alternate syntax for special case): D a t_QUAD,
• K = R : D a real number (t_REAL); truncate the factorization at accuracy precision(D).
If x is inexact and precision(x) is less than precision(D), then the precision of x is used instead.
• K = C : D a complex number with a t_REAL component, e.g. I * 1.; truncate the
factorization as for K = R,
• K = Qp : D a t_PADIC; truncate the factorization at p-adic accuracy padicprec(D), possibly
less if x is inexact with insufficient p-adic accuracy;
? T = x^2+1;
? factor(T, 1); \\ over Q
? factor(T, Mod(1,3)) \\ over F_3
? factor(T, ffgen(ffinit(3,2,’t))^0) \\ over F_{3^2}
? factor(T, Mod(Mod(1,3), t^2+t+2)) \\ over F_{3^2}, again
? factor(T, O(3^6)) \\ over Q_3, precision 6
? factor(T, 1.) \\ over R, current precision
? factor(T, I*1.) \\ over C
? factor(T, Mod(1, y^3-2)) \\ over Q(2^{1/3})
In most cases, it is possible and simpler to call a specialized variant rather than use the above
scheme:
? factormod(T, 3) \\ over F_3
? factormod(T, [t^2+t+2, 3]) \\ over F_{3^2}
? factormod(T, ffgen(3^2, ’t)) \\ over F_{3^2}
? factorpadic(T, 3,6) \\ over Q_3, precision 6
? nffactor(y^3-2, T) \\ over Q(2^{1/3})
? polroots(T) \\ over C
? polrootsreal(T) \\ over R (real polynomial)
It is also possible to let the routine use the smallest field containing all coefficients, taking into
account quotient structures induced by t_INTMODs and t_POLMODs (e.g. if a coefficient in Z/nZ is
known, all rational numbers encountered are first mapped to Z/nZ; different moduli will produce
an error):
? T = x^2+1;
? factor(T); \\ over Q
? factor(T*Mod(1,3)) \\ over F_3

188
 ? factor(T*ffgen(ffinit(3,2,’t))^0) \\ over F_{3^2}
? factor(T*Mod(Mod(1,3), t^2+t+2)) \\ over F_{3^2}, again
? factor(T*(1 + O(3^6)) \\ over Q_3, precision 6
? factor(T*1.) \\ over R, current precision
? factor(T*(1.+0.*I)) \\ over C
? factor(T*Mod(1, y^3-2)) \\ over Q(2^{1/3})

Multiplying by a suitable field element equal to 1 ∈ K in this way is error-prone and is not
recommanded. Factoring existing polynomials with obvious fields of coefficients is fine, the domain
argument D should be used instead ad hoc conversions.

Note on inexact polynomials. Polynomials with inexact coefficients (e.g. floating point or
p-adic numbers) are first rounded to an exact representation, then factored to (potentially) infinite
accuracy and we return a truncated approximation of that virtual factorization. To avoid pitfalls,
we advise to only factor exact polynomials:

? factor(x^2-1+O(2^2)) \\ rounded to x^2 + 3, irreducible in Q_2

%1 =
[(1 + O(2^2))*x^2 + O(2^2)*x + (1 + 2 + O(2^2)) 1]
? factor(x^2-1+O(2^3)) \\ rounded to x^2 + 7, reducible !
%2 =
[ (1 + O(2^3))*x + (1 + 2 + O(2^3)) 1]
[(1 + O(2^3))*x + (1 + 2^2 + O(2^3)) 1]
? factor(x^2-1, O(2^2)) \\ no ambiguity now
%3 =
[ (1 + O(2^2))*x + (1 + O(2^2)) 1]
[(1 + O(2^2))*x + (1 + 2 + O(2^2)) 1]

Note about inseparable polynomials. Polynomials with inexact coefficients are considered to
be squarefree: indeed, there exist a squarefree polynomial arbitrarily close to the input, and they
cannot be distinguished at the input accuracy. This means that irreducible factors are repeated
according to their apparent multiplicity. On the contrary, using a specialized function such as
factorpadic with an exact rational input yields the correct multiplicity when the (now exact)
input is not separable. Compare:

? factor(z^2 + O(5^2)))
%1 =
[(1 + O(5^2))*z + O(5^2) 1]
[(1 + O(5^2))*z + O(5^2) 1]
? factor(z^2, O(5^2))
%2 =
[1 + O(5^2))*z + O(5^2) 2]

189
Multivariate polynomials and rational functions. PARI recursively factors multivariate poly-
nomials in K[t1 , . . . , td ] for the same fields K as above and the argument D is used in the same way
to specify K. The irreducible factors are sorted by their main variable (least priority first) then by
increasing degree.
? factor(x^2 + y^2, Mod(1,5))
%1 =
[ x + Mod(2, 5)*y 1]
[Mod(1, 5)*x + Mod(3, 5)*y 1]
? factor(x^2 + y^2, O(5^2))
%2 =
[ (1 + O(5^2))*x + (O(5^2)*y^2 + (2 + 5 + O(5^2))*y + O(5^2)) 1]
[(1 + O(5^2))*x + (O(5^2)*y^2 + (3 + 3*5 + O(5^2))*y + O(5^2)) 1]
? lift(%)
%3 =
[ x + 7*y 1]
[x + 18*y 1]
Note that the implementation does not really support inexact real fields (R or C) and usually
misses factors even if the input is exact:
? factor(x^2 + y^2, I) \\ over Q(i)
%4 =
[x - I*y 1]
[x + I*y 1]
? factor(x^2 + y^2, I*1.) \\ over C
%5 =
[x^2 + y^2 1]
The library syntax is GEN factor0(GEN x, GEN D = NULL).
GEN factor(GEN x) GEN boundfact(GEN x, ulong lim).

3.8.31 factorback(f, {e}). Gives back the factored object corresponding to a factorization. The
integer 1 corresponds to the empty factorization.
If e is present, e and f must be vectors of the same length (e being integral), and the corre-
sponding factorization is the product of the f [i]e[i] .
If not, and f is vector, it is understood as in the preceding case with e a vector of 1s: we return
the product of the f [i]. Finally, f can be a regular factorization, as produced with any factor
command. A few examples:
? factor(12)
%1 =
[2 2]
[3 1]
? factorback(%)
%2 = 12
? factorback([2,3], [2,1]) \\ 2^3 * 3^1

190
 %3 = 12
? factorback([5,2,3])
%4 = 30

The library syntax is GEN factorback2(GEN f, GEN e = NULL). Also available is GEN fac-
torback(GEN f) (case e = NULL).

3.8.32 factorcantor(x, p). This function is obsolete, use factormod.

The library syntax is GEN factmod(GEN x, GEN p).

3.8.33 factorff(x, {p}, {a}). Obsolete, kept for backward compatibility: use factormod.

The library syntax is GEN factorff(GEN x, GEN p = NULL, GEN a = NULL).

3.8.34 factorial(x). Factorial of x. The expression x! gives a result which is an integer, while
factorial(x) gives a real number.

The library syntax is GEN mpfactr(long x, long prec). GEN mpfact(long x) returns x! as
a t_INT.

3.8.35 factorint(x, {flag = 0}). Factors the integer n into a product of pseudoprimes (see ispseu-
doprime), using a combination of the Shanks SQUFOF and Pollard Rho method (with modifications
due to Brent), Lenstra’s ECM (with modifications by Montgomery), and MPQS (the latter adapted
from the LiDIA code with the kind permission of the LiDIA maintainers), as well as a search for
pure powers. The output is a two-column matrix as for factor: the first column contains the
“prime” divisors of n, the second one contains the (positive) exponents.

By convention 0 is factored as 01 , and 1 as the empty factorization; also the divisors are by
default not proven primes if they are larger than 264 , they only failed the BPSW compositeness
test (see ispseudoprime). Use isprime on the result if you want to guarantee primality or set the
factor_proven default to 1. Entries of the private prime tables (see addprimes) are also included
as is.

This gives direct access to the integer factoring engine called by most arithmetical functions.
flag is optional; its binary digits mean 1: avoid MPQS, 2: skip first stage ECM (we may still
fall back to it later), 4: avoid Rho and SQUFOF, 8: don’t run final ECM (as a result, a huge
composite may be declared to be prime). Note that a (strong) probabilistic primality test is used;
thus composites might not be detected, although no example is known.

You are invited to play with the flag settings and watch the internals at work by using gp’s
debug default parameter (level 3 shows just the outline, 4 turns on time keeping, 5 and above show
an increasing amount of internal details).

The library syntax is GEN factorint(GEN x, long flag).

191
3.8.36 factormod(f, {D}, {flag = 0}). Factors the polynomial f over the finite field defined by
the domain D as follows:
• D = p a prime: factor over Fp ;
• D = [T, p] for a prime p and T (y) an irreducible polynomial over Fp : factor over Fp [y]/(T )
(as usual the main variable of T must have lower priority than the main variable of f );
• D a t_FFELT: factor over the attached field;
• D omitted: factor over the field of definition of f , which must be a finite field.
The coefficients of f must be operation-compatible with the corresponding finite field. The
result is a two-column matrix, the first column being the irreducible polynomials dividing f , and the
second the exponents. By convention, the 0 polynomial factors as 01 ; a nonzero constant polynomial
has empty factorization, a 0 × 2 matrix. The irreducible factors are ordered by increasing degree
and the result is canonical: it will not change across multiple calls or sessions.
? factormod(x^2 + 1, 3) \\ over F_3
%1 =
[Mod(1, 3)*x^2 + Mod(1, 3) 1]
? liftall( factormod(x^2 + 1, [t^2+1, 3]) ) \\ over F_9
%2 =
[ x + t 1]
[x + 2*t 1]
\\ same, now letting GP choose a model
? T = ffinit(3,2,’t)
%3 = Mod(1, 3)*t^2 + Mod(1, 3)*t + Mod(2, 3)
? liftall( factormod(x^2 + 1, [T, 3]) )
%4 = \\ t is a root of T !
[ x + (t + 2) 1]
[x + (2*t + 1) 1]
? t = ffgen(t^2+Mod(1,3)); factormod(x^2 + t^0) \\ same using t_FFELT
%5 =
[ x + t 1]
[x + 2*t 1]
? factormod(x^2+Mod(1,3))
%6 =
[Mod(1, 3)*x^2 + Mod(1, 3) 1]
? liftall( factormod(x^2 + Mod(Mod(1,3), y^2+1)) )
%7 =
[ x + y 1]
[x + 2*y 1]
If flag is nonzero, outputs only the degrees of the irreducible polynomials (for example to
compute an L-function). By convention, a constant polynomial (including the 0 polynomial) has
empty factorization. The degrees appear in increasing order but need not correspond to the ordering
with flag = 0 when multiplicities are present.
? f = x^3 + 2*x^2 + x + 2;
? factormod(f, 5) \\ (x+2)^2 * (x+3)

192
 %1 =
[Mod(1, 5)*x + Mod(2, 5) 2]
[Mod(1, 5)*x + Mod(3, 5) 1]
? factormod(f, 5, 1) \\ (deg 1) * (deg 1)^2
%2 =
[1 1]
[1 2]
The library syntax is GEN factormod0(GEN f, GEN D = NULL, long flag).

3.8.37 factormodDDF(f, {D}). Distinct-degree factorization of the squarefree polynomial f over

the finite field defined by the domain D as follows:
• D = p a prime: factor over Fp ;
• D = [T, p] for a prime p and T an irreducible polynomial over Fp : factor over Fp [x]/(T );
• D a t_FFELT: factor over the attached field;
• D omitted: factor over the field of definition of f , which must be a finite field.
This is somewhat faster than full factorization. The coefficients of f must be operation-
compatible with the corresponding finite field. The result is a two-column matrix:
• the first column contains monic (squarefree) pairwise coprime polynomials dividing f , all of
whose irreducible factors have degree d;
• the second column contains the degrees of the irreducible factors.
The factors are ordered by increasing degree and the result is canonical: it will not change
across multiple calls or sessions.
? f = (x^2 + 1) * (x^2-1);
? factormodSQF(f,3) \\ squarefree over F_3
%2 =
[Mod(1, 3)*x^4 + Mod(2, 3) 1]
? factormodDDF(f, 3)
%3 =
[Mod(1, 3)*x^2 + Mod(2, 3) 1] \\ two degree 1 factors
[Mod(1, 3)*x^2 + Mod(1, 3) 2] \\ irred of degree 2
? for(i=1,10^5,factormodDDF(f,3))
time = 424 ms.
? for(i=1,10^5,factormod(f,3)) \\ full factorization is slower
time = 464 ms.
? liftall( factormodDDF(x^2 + 1, [3, t^2+1]) ) \\ over F_9
%6 =
[x^2 + 1 1] \\ product of two degree 1 factors
? t = ffgen(t^2+Mod(1,3)); factormodDDF(x^2 + t^0) \\ same using t_FFELT
%7 =
[x^2 + 1 1]
? factormodDDF(x^2-Mod(1,3))

193
 %8 =
[Mod(1, 3)*x^2 + Mod(2, 3) 1]

The library syntax is GEN factormodDDF(GEN f, GEN D = NULL).

3.8.38 factormodSQF(f, {D}). Squarefree factorization of the polynomial f over the finite field
defined by the domain D as follows:
• D = p a prime: factor over Fp ;
• D = [T, p] for a prime p and T an irreducible polynomial over Fp : factor over Fp [x]/(T );
• D a t_FFELT: factor over the attached field;
• D omitted: factor over the field of definition of f , which must be a finite field.
This is somewhat faster than full factorization. The coefficients of f must be operation-
compatible with the corresponding finite field. The result is a two-column matrix:
• the first column contains monic squarefree pairwise coprime polynomials dividing f ;
• the second column contains the power to which the polynomial in column 1 divides f ;
The factors are ordered by increasing degree and the result is canonical: it will not change
across multiple calls or sessions.

? f = (x^2 + 1)^3 * (x^2-1)^2;

? factormodSQF(f, 3) \\ over F_3
%1 =
[Mod(1, 3)*x^2 + Mod(2, 3) 2]
[Mod(1, 3)*x^2 + Mod(1, 3) 3]
? for(i=1,10^5,factormodSQF(f,3))
time = 192 ms.
? for(i=1,10^5,factormod(f,3)) \\ full factorization is slower
time = 409 ms.
? liftall( factormodSQF((x^2 + 1)^3, [3, t^2+1]) ) \\ over F_9
%4 =
[x^2 + 1 3]
? t = ffgen(t^2+Mod(1,3)); factormodSQF((x^2 + t^0)^3) \\ same using t_FFELT
%5 =
[x^2 + 1 3]
? factormodSQF(x^8 + x^7 + x^6 + x^2 + x + Mod(1,2))
%6 =
[ Mod(1, 2)*x + Mod(1, 2) 2]
[Mod(1, 2)*x^2 + Mod(1, 2)*x + Mod(1, 2) 3]
The library syntax is GEN factormodSQF(GEN f, GEN D = NULL).

194
3.8.39 ffcompomap(f, g). Let k, l, m be three finite fields and f a (partial) map from l to m and
g a (partial) map from k to l, return the (partial) map f ◦ g from k to m.

a = ffgen([3,5],’a); b = ffgen([3,10],’b); c = ffgen([3,20],’c);

m = ffembed(a, b); n = ffembed(b, c);
rm = ffinvmap(m); rn = ffinvmap(n);
nm = ffcompomap(n,m);
ffmap(n,ffmap(m,a)) == ffmap(nm, a)
%5 = 1
ffcompomap(rm, rn) == ffinvmap(nm)
%6 = 1

The library syntax is GEN ffcompomap(GEN f, GEN g).

3.8.40 ffembed(a, b). Given two finite fields elements a and b, return a map embedding the
definition field of a to the definition field of b. Assume that the latter contains the former.

? a = ffgen([3,5],’a);
? b = ffgen([3,10],’b);
? m = ffembed(a, b);
? A = ffmap(m, a);
? minpoly(A) == minpoly(a)
%5 = 1

The library syntax is GEN ffembed(GEN a, GEN b).

3.8.41 ffextend(a, P, {v}). Extend the field K of definition of a by a root of the polynomial
P ∈ K[X] assumed to be irreducible over K. Return [r, m] where r is a root of P in the extension
field L and m is a map from K to L, see ffmap. If v is given, the variable name is used to display
the generator of L, else the name of the variable of P is used. A generator of L can be recovered
using b = f f gen(r). The image of P in L[X] can be recovered using P L = f f map(m, P ).

? a = ffgen([3,5],’a);
? P = x^2-a; polisirreducible(P)
%2 = 1
? [r,m] = ffextend(a, P, ’b);
? r
%3 = b^9+2*b^8+b^7+2*b^6+b^4+1
? subst(ffmap(m, P), x, r)
%4 = 0
? ffgen(r)
%5 = b

The library syntax is GEN ffextend(GEN a, GEN P, long v = -1) where v is a variable
number.

195
3.8.42 fffrobenius(m, {n = 1}). Return the n-th power of the Frobenius map over the field of
definition of m.
? a = ffgen([3,5],’a);
? f = fffrobenius(a);
? ffmap(f,a) == a^3
%3 = 1
? g = fffrobenius(a, 5);
? ffmap(g,a) == a
%5 = 1
? h = fffrobenius(a, 2);
? h == ffcompomap(f,f)
%7 = 1
The library syntax is GEN fffrobenius(GEN m, long n).

3.8.43 ffgen(k, {v =0 x}). Return a generator for the finite field k as a t_FFELT. The field k can
be given by
• its order q
• the pair [p, f ] where q = pf
• a monic irreducible polynomial with t_INTMOD coefficients modulo a prime.
• a t_FFELT belonging to k.
If v is given, the variable name is used to display g, else the variable of the polynomial or the
t_FFELT is used, else x is used.
When only the order is specified, the function uses the polynomial generated by ffinit and
is deterministic: two calls to the function with the same parameters will always give the same
generator.
For efficiency, the characteristic is not checked to be prime; similarly if a polynomial is given,
we do not check whether it is irreducible.
To obtain a multiplicative generator, call ffprimroot on the result.
? g = ffgen(16, ’t);
? g.mod \\ recover the underlying polynomial.
%2 = t^4+t^3+t^2+t+1
? g.pol \\ lift g as a t_POL
%3 = t
? g.p \\ recover the characteristic
%4 = 2
? fforder(g) \\ g is not a multiplicative generator
%5 = 5
? a = ffprimroot(g) \\ recover a multiplicative generator
%6 = t^3+t^2+t
? fforder(a)
%7 = 15
The library syntax is GEN ffgen(GEN k, long v = -1) where v is a variable number.
To create a generator for a prime finite field, the function GEN p_to_GEN(GEN p, long v)
returns ffgen(p,v)^0.

196
3.8.44 ffinit(p, n, {v =0 x}). Computes a monic polynomial of degree n which is irreducible over
Fp , where p is assumed to be prime. This function uses a fast variant of Adleman and Lenstra’s
algorithm.
It is useful in conjunction with ffgen; for instance if P = ffinit(3,2), you can represent
elements in F32 in term of g = ffgen(P,’t). This can be abbreviated as g = ffgen(3^2, ’t),
where the defining polynomial P can be later recovered as g.mod.
The library syntax is GEN ffinit(GEN p, long n, long v = -1) where v is a variable
number.

3.8.45 ffinvmap(m). m being a map from K to L two finite fields, return the partial map p from
L to K such that for all k ∈ K, p(m(k)) = k.
? a = ffgen([3,5],’a);
? b = ffgen([3,10],’b);
? m = ffembed(a, b);
? p = ffinvmap(m);
? u = random(a);
? v = ffmap(m, u);
? ffmap(p, v^2+v+2) == u^2+u+2
%7 = 1
? ffmap(p, b)
%8 = []
The library syntax is GEN ffinvmap(GEN m).

3.8.46 fflog(x, g, {o}). Discrete logarithm of the finite field element x in base g, i.e. an e in Z such
that g e = o. If present, o represents the multiplicative order of g, see Section 3.8.2; the preferred
format for this parameter is [ord, factor(ord)], where ord is the order of g. It may be set as a
side effect of calling ffprimroot. The result is undefined if e does not exist. This function uses
• a combination of generic discrete log algorithms (see znlog)
• a cubic sieve index calculus algorithm for large fields of degree at least 5.
• Coppersmith’s algorithm for fields of characteristic at most 5.
? t = ffgen(ffinit(7,5));
? o = fforder(t)
%2 = 5602 \\ not a primitive root.
? fflog(t^10,t)
%3 = 10
? fflog(t^10,t, o)
%4 = 10
? g = ffprimroot(t, &o);
? o \\ order is 16806, bundled with its factorization matrix
%6 = [16806, [2, 1; 3, 1; 2801, 1]]
? fforder(g, o)
%7 = 16806
? fflog(g^10000, g, o)
%8 = 10000
The library syntax is GEN fflog(GEN x, GEN g, GEN o = NULL).

197
3.8.47 ffmap(m, x). Given a (partial) map m between two finite fields, return the image of x by
m. The function is applied recursively to the component of vectors, matrices and polynomials. If
m is a partial map that is not defined at x, return [].

? a = ffgen([3,5],’a);
? b = ffgen([3,10],’b);
? m = ffembed(a, b);
? P = x^2+a*x+1;
? Q = ffmap(m,P);
? ffmap(m,poldisc(P)) == poldisc(Q)
%6 = 1

The library syntax is GEN ffmap(GEN m, GEN x).

3.8.48 ffmaprel(m, x). Given a (partial) map m between two finite fields, express x as an algebraic
element over the codomain of m in a way which is compatible with m. The function is applied
recursively to the component of vectors, matrices and polynomials.

? a = ffgen([3,5],’a);
? b = ffgen([3,10],’b);
? m = ffembed(a, b);
? mi= ffinvmap(m);
? R = ffmaprel(mi,b)
%5 = Mod(b,b^2+(a+1)*b+(a^2+2*a+2))

In particular, this function can be used to compute the relative minimal polynomial, norm and
trace:

? minpoly(R)
%6 = x^2+(a+1)*x+(a^2+2*a+2)
? trace(R)
%7 = 2*a+2
? norm(R)
%8 = a^2+2*a+2

The library syntax is GEN ffmaprel(GEN m, GEN x).

3.8.49 ffnbirred(q, n, {fl = 0}). Computes the number of monic irreducible polynomials over Fq
of degree exactly n, (flag = 0 or omitted) or at most n (flag = 1).

The library syntax is GEN ffnbirred0(GEN q, long n, long fl). Also available are GEN
ffnbirred(GEN q, long n) (for flag = 0) and GEN ffsumnbirred(GEN q, long n) (for flag = 1).

198
3.8.50 fforder(x, {o}). Multiplicative order of the finite field element x. If o is present, it represents
a multiple of the order of the element, see Section 3.8.2; the preferred format for this parameter
is [N, factor(N)], where N is the cardinality of the multiplicative group of the underlying finite
field.

? t = ffgen(ffinit(nextprime(10^8), 5));
? g = ffprimroot(t, &o); \\ o will be useful!
? fforder(g^1000000, o)
time = 0 ms.
%5 = 5000001750000245000017150000600250008403
? fforder(g^1000000)
time = 16 ms. \\ noticeably slower, same result of course
%6 = 5000001750000245000017150000600250008403

The library syntax is GEN fforder(GEN x, GEN o = NULL).

3.8.51 ffprimroot(x, {&o}). Return a primitive root of the multiplicative group of the definition
field of the finite field element x (not necessarily the same as the field generated by x). If present, o is
set to a vector [ord, fa], where ord is the order of the group and fa its factorization factor(ord).
This last parameter is useful in fflog and fforder, see Section 3.8.2.

? t = ffgen(ffinit(nextprime(10^7), 5));
? g = ffprimroot(t, &o);
? o[1]
%3 = 100000950003610006859006516052476098
? o[2]
%4 =
[2 1]
[7 2]
[31 1]
[41 1]
[67 1]
[1523 1]
[10498781 1]
[15992881 1]
[46858913131 1]
? fflog(g^1000000, g, o)
time = 1,312 ms.
%5 = 1000000

The library syntax is GEN ffprimroot(GEN x, GEN *o = NULL).

199
3.8.52 gcd(x, {y}). Creates the greatest common divisor of x and y. If you also need the u and v
such that x ∗ u + y ∗ v = gcd(x, y), use the gcdext function. x and y can have rather quite general
types, for instance both rational numbers. If y is omitted and x is a vector, returns the gcd of all
components of x, i.e. this is equivalent to content(x).
When x and y are both given and one of them is a vector/matrix type, the GCD is again taken
recursively on each component, but in a different way. If y is a vector, resp. matrix, then the result
has the same type as y, and components equal to gcd(x, y[i]), resp. gcd(x, y[,i]). Else if x is
a vector/matrix the result has the same type as x and an analogous definition. Note that for these
types, gcd is not commutative.
The algorithm used is a naive Euclid except for the following inputs:
• integers: use modified right-shift binary (“plus-minus” variant).
• univariate polynomials with coefficients in the same number field (in particular rational):
use modular gcd algorithm.
• general polynomials: use the subresultant algorithm if coefficient explosion is likely (non
modular coefficients).
If u and v are polynomials in the same variable with inexact coefficients, their gcd is defined
to be scalar, so that
? a = x + 0.0; gcd(a,a)
%1 = 1
? b = y*x + O(y); gcd(b,b)
%2 = y
? c = 4*x + O(2^3); gcd(c,c)
%3 = 4
A good quantitative check to decide whether such a gcd “should be” nontrivial, is to use polre-
sultant: a value close to 0 means that a small deformation of the inputs has nontrivial gcd. You
may also use gcdext, which does try to compute an approximate gcd d and provides u, v to check
whether ux + vy is close to d.
The library syntax is GEN ggcd0(GEN x, GEN y = NULL). Also available are GEN ggcd(GEN
x, GEN y), if y is not NULL, and GEN content(GEN x), if y = NULL.

3.8.53 gcdext(x, y). Returns [u, v, d] such that d is the gcd of x, y, x ∗ u + y ∗ v = gcd(x, y), and
u and v minimal in a natural sense. The arguments must be integers or polynomials.
? [u, v, d] = gcdext(32,102)
%1 = [16, -5, 2]
? d
%2 = 2
? gcdext(x^2-x, x^2+x-2)
%3 = [-1/2, 1/2, x - 1]
If x, y are polynomials in the same variable and inexact coefficients, then compute u, v, d such
that x ∗ u + y ∗ v = d, where d approximately divides both and x and y; in particular, we do not
obtain gcd(x,y) which is defined to be a scalar in this case:
? a = x + 0.0; gcd(a,a)
%1 = 1

200
 ? gcdext(a,a)
%2 = [0, 1, x + 0.E-28]
? gcdext(x-Pi, 6*x^2-zeta(2))
%3 = [-6*x - 18.8495559, 1, 57.5726923]
For inexact inputs, the output is thus not well defined mathematically, but you obtain explicit
polynomials to check whether the approximation is close enough for your needs.
The library syntax is GEN gcdext0(GEN x, GEN y).

3.8.54 halfgcd(x, y). Let inputs x and y be both integers, or both polynomials in the same
variable. Return a vector [M, [a,b]~], where M is an invertible 2 × 2 matrix such that M*[x,y]~=
[a,b]~, where b is small. More precisely,
• polynomial case: det M has degree 0 and we have

deg a ≥ dmax(deg x, deg y))/2e > deg b.

• integer case: det M = ±1 and we have

lp m
a≥ max(|x|, |y|) > b.

Assuming x and y are nonnegative, then M −1 has nonnegative coefficients, and det M is equal to
the sign of both main diagonal terms M [1, 1] and M [2, 2].
The library syntax is GEN ghalfgcd(GEN x, GEN y).

3.8.55 hilbert(x, y, {p}). Hilbert symbol of x and y modulo the prime p, p = 0 meaning the place
at infinity (the result is undefined if p 6= 0 is not prime).
It is possible to omit p, in which case we take p = 0 if both x and y are rational, or one of them
is a real number. And take p = q if one of x, y is a t_INTMOD modulo q or a q-adic. (Incompatible
types will raise an error.)
The library syntax is long hilbert(GEN x, GEN y, GEN p = NULL).

3.8.56 isfundamental(D). True (1) if D is equal to 1 or to the discriminant of a quadratic field,

false (0) otherwise. D can be input in factored form as for arithmetic functions:
? isfundamental(factor(-8))
%1 = 1
\\ count fundamental discriminants up to 10^8
? c = 0; forfactored(d = 1, 10^8, if (isfundamental(d), c++)); c
time = 40,840 ms.
%2 = 30396325
? c = 0; for(d = 1, 10^8, if (isfundamental(d), c++)); c
time = 1min, 33,593 ms. \\ slower !
%3 = 30396325
The library syntax is long isfundamental(GEN D).

201
3.8.57 ispolygonal(x, s, {&N }). True (1) if the integer x is an s-gonal number, false (0) if not.
The parameter s > 2 must be a t_INT. If N is given, set it to n if x is the n-th s-gonal number.
? ispolygonal(36, 3, &N)
%1 = 1
? N
The library syntax is long ispolygonal(GEN x, GEN s, GEN *N = NULL).

3.8.58 ispower(x, {k}, {&n}). If k is given, returns true (1) if x is a k-th power, false (0) if not.
What it means to be a k-th power depends on the type of x; see issquare for details.
If k is omitted, only integers and fractions are allowed for x and the function returns the
maximal k ≥ 2 such that x = nk is a perfect power, or 0 if no such k exist; in particular ispower(-
1), ispower(0), and ispower(1) all return 0.
If a third argument &n is given and x is indeed a k-th power, sets n to a k-th root of x.
For a t_FFELT x, instead of omitting k (which is not allowed for this type), it may be natural to
set
k = (x.p ^ x.f - 1) / fforder(x)
The library syntax is long ispower(GEN x, GEN k = NULL, GEN *n = NULL). Also available
is long gisanypower(GEN x, GEN *pty) (k omitted).

3.8.59 ispowerful(x). True (1) if x is a powerful integer, false (0) if not; an integer is powerful if
and only if its valuation at all primes dividing x is greater than 1.
? ispowerful(50)
%1 = 0
? ispowerful(100)
%2 = 1
? ispowerful(5^3*(10^1000+1)^2)
%3 = 1
The library syntax is long ispowerful(GEN x).

3.8.60 isprime(x, {flag = 0}). True (1) if x is a prime number, false (0) otherwise. A prime number
is a positive integer having exactly two distinct divisors among the natural numbers, namely 1 and
itself.
This routine proves or disproves rigorously that a number is prime, which can be very slow
when x is indeed a large prime integer. For instance a 1000 digits prime should require 15 to 30
minutes with default algorithms. Use ispseudoprime to quickly check for compositeness. Use
primecert in order to obtain a primality proof instead of a yes/no answer; see also factor.
The function accepts vector/matrices arguments, and is then applied componentwise.
If flag = 0, use a combination of
• Baillie-Pomerance-Selfridge-Wagstaff compositeness test (see ispseudoprime),
• Selfridge “p − 1” test if x − 1 is smooth enough,
• Adleman-Pomerance-Rumely-Cohen-Lenstra (APRCL) for general medium-sized x (less than
1500 bits),

202
 • Atkin-Morain’s Elliptic Curve Primality Prover (ECPP) for general large x.

If flag = 1, use Selfridge-Pocklington-Lehmer “p − 1” test; this requires partially factoring

various auxilliary integers and is likely to be very slow.

If flag = 2, use APRCL only.

If flag = 3, use ECPP only.

The library syntax is GEN gisprime(GEN x, long flag).

3.8.61 isprimepower(x, {&n}). If x = pk is a prime power (p prime, k > 0), return k, else return
0. If a second argument &n is given and x is indeed the k-th power of a prime p, sets n to p.

The library syntax is long isprimepower(GEN x, GEN *n = NULL).

3.8.62 ispseudoprime(x, {flag}). True (1) if x is a strong pseudo prime (see below), false (0)
otherwise. If this function returns false, x is not prime; if, on the other hand it returns true, it
is only highly likely that x is a prime number. Use isprime (which is of course much slower) to
prove that x is indeed prime. The function accepts vector/matrices arguments, and is then applied
componentwise.

If flag = 0, checks whether x has no small prime divisors (up to 101 included) and is a Baillie-
Pomerance-Selfridge-Wagstaff pseudo prime. Such a pseudo prime passes a Rabin-Miller test for
base 2, followed by a Lucas test for the sequence (P, 1), where P ≥ 3 is the smallest odd integer
such that P 2 − 4 is not a square mod x. (Technically, we are using an “almost extra strong Lucas
test” that checks whether Vn is ±2, without computing Un .)

There are no known composite numbers passing the above test, although it is expected that
infinitely many such numbers exist. In particular, all composites ≤ 264 are correctly detected
(checked using http://www.cecm.sfu.ca/Pseudoprimes/index-2-to-64.html).

If flag > 0, checks whether x is a strong Miller-Rabin pseudo prime for flag randomly chosen
bases (with end-matching to catch square roots of −1).

The library syntax is GEN gispseudoprime(GEN x, long flag).

3.8.63 ispseudoprimepower(x, {&n}). If x = pk is a pseudo-prime power (p pseudo-prime as

per ispseudoprime, k > 0), return k, else return 0. If a second argument &n is given and x is
indeed the k-th power of a prime p, sets n to p.

More precisely, k is always the largest integer such that x = nk for some integer n and, when
n ≤ 264 the function returns k > 0 if and only if n is indeed prime. When n > 264 is larger
than the threshold, the function may return 1 even though n is composite: it only passed an
ispseudoprime(n) test.

The library syntax is long ispseudoprimepower(GEN x, GEN *n = NULL).

203
3.8.64 issquare(x, {&n}). True (1) if x is a square, false (0) if not. What “being a square” means
depends on the type of x: all t_COMPLEX are squares, as well as all nonnegative t_REAL; for exact
types such as t_INT, t_FRAC and t_INTMOD, squares are numbers of the form s2 with s in Z, Q and
Z/N Z respectively.
? issquare(3) \\ as an integer
%1 = 0
? issquare(3.) \\ as a real number
%2 = 1
? issquare(Mod(7, 8)) \\ in Z/8Z
%3 = 0
? issquare( 5 + O(13^4) ) \\ in Q_13
%4 = 0
If n is given, a square root of x is put into n.
? issquare(4, &n)
%1 = 1
? n
%2 = 2
For polynomials, either we detect that the characteristic is 2 (and check directly odd and
even-power monomials) or we assume that 2 is invertible and check whether squaring the truncated
power series for the square root yields the original input.
For t_POLMOD x, we only support t_POLMODs of t_INTMODs encoding finite fields, assuming
without checking that the intmod modulus p is prime and that the polmod modulus is irreducible
modulo p.
? issquare(Mod(Mod(2,3), x^2+1), &n)
%1 = 1
? n
%2 = Mod(Mod(2, 3)*x, Mod(1, 3)*x^2 + Mod(1, 3))
The library syntax is long issquareall(GEN x, GEN *n = NULL). Also available is long is-
square(GEN x). Deprecated GP-specific functions GEN gissquare(GEN x) and GEN gissquare-
all(GEN x, GEN *pt) return gen 0 and gen 1 instead of a boolean value.

3.8.65 issquarefree(x). True (1) if x is squarefree, false (0) if not. Here x can be an integer or a
polynomial with coefficients in an integral domain.
? issquarefree(12)
%1 = 0
? issquarefree(6)
%2 = 1
? issquarefree(x^3+x^2)
%3 = 0
? issquarefree(Mod(1,4)*(x^2+x+1)) \\ Z/4Z is not a domain !
*** at top-level: issquarefree(Mod(1,4)*(x^2+x+1))
*** ^--------------------------------
*** issquarefree: impossible inverse in Fp_inv: Mod(2, 4).
A polynomial is declared squarefree if gcd(x, x0 ) is 1. In particular a nonzero polynomial with
inexact coefficients is considered to be squarefree. Note that this may be inconsistent with factor,

204
which first rounds the input to some exact approximation before factoring in the apropriate domain;
this is correct when the input is not close to an inseparable polynomial (the resultant of x and x0
is not close to 0).

An integer can be input in factored form as in arithmetic functions.

? issquarefree(factor(6))
%1 = 1
\\ count squarefree integers up to 10^8
? c = 0; for(d = 1, 10^8, if (issquarefree(d), c++)); c
time = 3min, 2,590 ms.
%2 = 60792694
? c = 0; forfactored(d = 1, 10^8, if (issquarefree(d), c++)); c
time = 45,348 ms. \\ faster !
%3 = 60792694

The library syntax is long issquarefree(GEN x).

3.8.66 istotient(x, {&N }). True (1) if x = φ(n) for some integer n, false (0) if not.

? istotient(14)
%1 = 0
? istotient(100)
%2 = 0

If N is given, set N = n as well.

? istotient(4, &n)
%1 = 1
? n
%2 = 10

The library syntax is long istotient(GEN x, GEN *N = NULL).

3.8.67 kronecker(x, y). Kronecker symbol (x|y), where x and y must be of type integer. By
definition, this is the extension of Legendre symbol to Z × Z by total multiplicativity in both
arguments with the following special rules for y = 0, −1 or 2:

• (x|0) = 1 if |x| = 1 and 0 otherwise.

• (x| − 1) = 1 if x ≥ 0 and −1 otherwise.

• (x|2) = 0 if x is even and 1 if x = 1, −1 mod 8 and −1 if x = 3, −3 mod 8.

The library syntax is long kronecker(GEN x, GEN y).

205
3.8.68 lcm(x, {y}). Least common multiple of x and y, i.e. such that lcm(x, y) ∗ gcd(x, y) = x ∗ y,
up to units. If y is omitted and x is a vector, returns the lcm of all components of x. For integer
arguments, return the nonnegative lcm.
When x and y are both given and one of them is a vector/matrix type, the LCM is again taken
recursively on each component, but in a different way. If y is a vector, resp. matrix, then the result
has the same type as y, and components equal to lcm(x, y[i]), resp. lcm(x, y[,i]). Else if x is
a vector/matrix the result has the same type as x and an analogous definition. Note that for these
types, lcm is not commutative.
Note that lcm(v) is quite different from
l = v[1]; for (i = 1, #v, l = lcm(l, v[i]))
Indeed, lcm(v) is a scalar, but l may not be (if one of the v[i] is a vector/matrix). The computa-
tion uses a divide-conquer tree and should be much more efficient, especially when using the GMP
multiprecision kernel (and more subquadratic algorithms become available):
? v = vector(10^5, i, random);
? lcm(v);
time = 546 ms.
? l = v[1]; for (i = 1, #v, l = lcm(l, v[i]))
time = 4,561 ms.
The library syntax is GEN glcm0(GEN x, GEN y = NULL).

3.8.69 logint(x, b, {&z}). Return the largest integer e so that be ≤ x, where the parameters b > 1
and x > 0 are both integers. If the parameter z is present, set it to be .
? logint(1000, 2)
%1 = 9
? 2^9
%2 = 512
? logint(1000, 2, &z)
%3 = 9
? z
%4 = 512
The number of digits used to write b in base x is 1 + logint(x,b):
? #digits(1000!, 10)
%5 = 2568
? logint(1000!, 10)
%6 = 2567
This function may conveniently replace
floor( log(x) / log(b) )
which may not give the correct answer since PARI does not guarantee exact rounding.
The library syntax is long logint0(GEN x, GEN b, GEN *z = NULL).

3.8.70 moebius(x). Moebius µ-function of |x|; x must be a nonzero integer.

The library syntax is long moebius(GEN x).

206
3.8.71 nextprime(x). Finds the smallest pseudoprime (see ispseudoprime) greater than or equal
to x. x can be of any real type. Note that if x is a pseudoprime, this function returns x and not
the smallest pseudoprime strictly larger than x. To rigorously prove that the result is prime, use
isprime.

The library syntax is GEN nextprime(GEN x).

3.8.72 numdiv(x). Number of divisors of |x|. x must be of type integer.

The library syntax is GEN numdiv(GEN x).

3.8.73 omega(x). Number of distinct prime divisors of |x|. x must be of type integer.

? factor(392)
%1 =
[2 3]
[7 2]
? omega(392)
%2 = 2; \\ without multiplicity
? bigomega(392)
%3 = 5; \\ = 3+2, with multiplicity

The library syntax is long omega(GEN x).

3.8.74 precprime(x). Finds the largest pseudoprime (see ispseudoprime) less than or equal to
x. x can be of any real type. Returns 0 if x ≤ 1. Note that if x is a prime, this function returns x
and not the largest prime strictly smaller than x. To rigorously prove that the result is prime, use
isprime.

The library syntax is GEN precprime(GEN x).

3.8.75 prime(n). The nth prime number

? prime(10^9)
%1 = 22801763489
11
Uses checkpointing and a naive O(n) algorithm. Will need about
p30 minutes for n up to 10 ; make
√
sure to start gp with primelimit at least pn , e.g. the value n log(n log n) is guaranteed to be
sufficient.

The library syntax is GEN prime(long n).

207
3.8.76 primecert(N, {flag = 0}). If N is a prime, return a PARI Primality Certificate for the
prime N , as described below. Otherwise, return 0. A Primality Certificate c can be checked using
primecertisvalid(c).
If flag = 0 (default), return an ECPP certificate (Atkin-Morain)
A PARI ECPP Primality Certificate for the prime N is either a prime integer N < 264 or
a vector C of length ` whose ith component C[i] is a vector [Ni , ti , si , ai , Pi ] of length 5 where
N1 = N . It is said to be valid if for each i = 1, . . . , `, all of the following conditions are satisfied
• Ni is a positive integer
• ti is an integer such that t2i < 4Ni
• si is a positive integer which divides mi where mi = Ni + 1 − ti
mi
• If we set qi = si , then
1/4
• qi > (Ni + 1)2
• qi = Ni+1 if 1 ≤ i < l
• q` ≤ 264 is prime
• ai is an integer
• P[i] is a vector of length 2 representing the affine point Pi = (xi , yi ) on the elliptic curve
E : y 2 = x3 + ai x + bi modulo Ni where bi = yi2 − x3i − ai xi satisfying the following:
• mi Pi = ∞
• si Pi 6= ∞
Using the following theorem, the data in the vector C allows to recursively certify the primality
of N (and all the qi ) under the single assumption that q` be prime.

Theorem. If N is an integer and there exist positive integers m, q and a point P on the elliptic
curve E : y 2 = x3 + ax + b defined modulo N such that q > (N 1/4 + 1)2 , q is a prime divisor of m,
mP = ∞ and m q P 6= ∞, then N is prime.

? primecert(10^35 + 69)
%1 = [[100000000000000000000000000000000069, 5468679110354
52074, 2963504668391148, 0, [60737979324046450274283740674
208692, 24368673584839493121227731392450025]], [3374383076
4501150277, -11610830419, 734208843, 0, [26740412374402652
72 4, 6367191119818901665]], [45959444779, 299597, 2331, 0
, [18022351516, 9326882 51]]]
? primecert(nextprime(2^64))
%2 = [[18446744073709551629, -8423788454, 160388, 1, [1059
8342506117936052, 2225259013356795550]]]
? primecert(6)
%3 = 0
? primecert(41)
%4 = 41

If flag = 1 (very slow), return an N − 1 certificate (Pocklington Lehmer)

208
 A PARI N − 1 Primality Certificate for the prime N is either a prime integer N < 264 or a
pair [N, C], where C is a vector with ` elements which are either a single integer pi < 264 or a triple
[pi , ai , Ci ] with pi > 264 satisfying the following properties:
• pi is a prime divisor of N − 1;
−1 (N −1)/pi
• ai is an integer such that aN
i ≡1 (mod N ) and ai − 1 is coprime with N ;
• Ci is an N − 1 Primality Certificate for pi
vp (N −1)
• The product F of the pi i is strictly larger than N 1/3 . Provided that all pi are indeed
primes, this implies that any divisor of N is congruent to 1 modulo F .
• The Billhart, Lehmer, Selfridge criterion is satisfied: when we write N = 1 + c1 F + c2 F 2 in
base F the polynomial 1 + c1 X + c2 X 2 is irreducible over Z, i.e. c21 − 4c2 is not a square. This
implies that N is prime.
This algorithm requires factoring partially p−1 for various prime integers p with an unfactored
parted ≤ p2/3 and this may be exceedingly slow compared to the default.
The algorithm fails if one of the pseudo-prime factors is not prime, which is exceedingly unlikely
and well worth a bug report. Note that if you monitor the algorithm at a high enough debug level,
you may see warnings about untested integers being declared primes. This is normal: we ask for
partial factorizations (sufficient to prove primality if the unfactored part is not too large), and
factor warns us that the cofactor hasn’t been tested. It may or may not be tested later, and may
or may not be prime. This does not affect the validity of the whole Primality Certificate.
The library syntax is GEN primecert(GEN N, long flag).

3.8.77 primecertexport(cert, {format = 0}). Returns a string suitable for print/write to display
a primality certificate from primecert, the format of which depends on the value of format:
• 0 (default): Human-readable format. See ??primecert for the meaning of the successive
N, t, s, a, m, q, E, P . The integer D is the negative fundamental discriminant coredisc(t2 − 4N ).
• 1: Primo format 4.
• 2: MAGMA format.
Currently, only ECPP Primality Certificates are supported.
? cert = primecert(10^35+69);
? s = primecertexport(cert); \\ Human-readable
? print(s)
[1]
N = 100000000000000000000000000000000069
t = 546867911035452074
s = 2963504668391148
a = 0
D = -3
m = 99999999999999999453132088964547996
q = 33743830764501150277
E = [0, 1]
P = [21567861682493263464353543707814204,
49167839501923147849639425291163552]

209
 [2]
N = 33743830764501150277
t = -11610830419
s = 734208843
a = 0
D = -3
m = 33743830776111980697
q = 45959444779
E = [0, 25895956964997806805]
P = [29257172487394218479, 3678591960085668324]
\\ Primo format
? s = primecertexport(cert,1); write("cert.out", s);
\\ Magma format, write to file
? s = primecertexport(cert,2); write("cert.m", s);
? cert = primecert(10^35+69, 1); \\ N-1 certificate
? primecertexport(cert)
*** at top-level: primecertexport(cert)
*** ^---------------------
*** primecertexport: sorry, N-1 certificate is not yet implemented.

The library syntax is GEN primecertexport(GEN cert, long format).

3.8.78 primecertisvalid(cert). Verifies if cert is a valid PARI ECPP Primality certificate, as

described in ??primecert.

? cert = primecert(10^35 + 69)

%1 = [[100000000000000000000000000000000069, 5468679110354
52074, 2963504668391148, 0, [60737979324046450274283740674
208692, 24368673584839493121227731392450025]], [3374383076
4501150277, -11610830419, 734208843, 0, [26740412374402652
72 4, 6367191119818901665]], [45959444779, 299597, 2331, 0
, [18022351516, 9326882 51]]]
? primecertisvalid(cert)
%2 = 1
? cert[1][1]++; \\ random perturbation
? primecertisvalid(cert)
%4 = 0 \\ no longer valid
? primecertisvalid(primecert(6))
%5 = 0

The library syntax is long primecertisvalid(GEN cert).

210
3.8.79 primepi(x). The prime counting function. Returns the number of primes p, p ≤ x.

? primepi(10)
%1 = 4;
? primes(5)
%2 = [2, 3, 5, 7, 11]
? primepi(10^11)
%3 = 4118054813

Uses
√ checkpointing and a naive O(x) algorithm; make sure to start gp with primelimit at least
x.

The library syntax is GEN primepi(GEN x).

3.8.80 primes(n). Creates a row vector whose components are the first n prime numbers. (Returns
the empty vector for n ≤ 0.) A t_VEC n = [a, b] is also allowed, in which case the primes in [a, b]
are returned

? primes(10) \\ the first 10 primes

%1 = [2, 3, 5, 7, 11, 13, 17, 19, 23, 29]
? primes([0,29]) \\ the primes up to 29
%2 = [2, 3, 5, 7, 11, 13, 17, 19, 23, 29]
? primes([15,30])
%3 = [17, 19, 23, 29]

The library syntax is GEN primes0(GEN n).

3.8.81 qfbclassno(D, {flag = 0}). Ordinary class number of the quadratic order of discriminant
D, for “small” values of D.

• if D > 0 or flag = 1, use a O(|D|1/2 ) algorithm (compute L(1, χD ) with the approximate
functional equation). This is slower than quadclassunit as soon as |D| ≈ 102 or so and is not
meant to be used for large D.

• if D < 0 and flag = 0 (or omitted), use a O(|D|1/4 ) algorithm (Shanks’s baby-step/giant-step
method). It should be faster than quadclassunit for small values of D, say |D| < 1018 .

Important warning. In the latter case, this function only implements part of Shanks’s method
(which allows to speed it up considerably). It gives unconditionnally correct results for |D| < 2·1010 ,
but may give incorrect results for larger values if the class group has many cyclic factors. We thus
recommend to double-check results using the function quadclassunit, which is about 2 to 3 times
slower in the range |D| ∈ [1010 , 1018 ], assuming GRH. We currently have no counter-examples but
they should exist: we would appreciate a bug report if you find one.

211
Warning. Contrary to what its name implies, this routine does not compute the number of classes
of binary primitive forms of discriminant D, which is equal to the narrow class number. The two
notions are the same when D < 0 or the fundamental unit ε has negative norm; when D > 0 and
N ε > 0, the number of classes of forms is twice the ordinary class number. This is a problem
which we cannot fix for backward compatibility reasons. Use the following routine if you are only
interested in the number of classes of forms:
QFBclassno(D) =
qfbclassno(D) * if (D < 0 || norm(quadunit(D)) < 0, 1, 2)
Here are a few examples:
? qfbclassno(400000028) \\ D > 0: slow
time = 3,140 ms.
%1 = 1
? quadclassunit(400000028).no
time = 20 ms. \\ much faster, assume GRH
%2 = 1
? qfbclassno(-400000028) \\ D < 0: fast enough
time = 0 ms.
%3 = 7253
? quadclassunit(-400000028).no
time = 0 ms.
%4 = 7253
See also qfbhclassno.
The library syntax is GEN qfbclassno0(GEN D, long flag).

3.8.82 qfbcompraw(x, y). composition of the binary quadratic forms x and y, without reduction
of the result. This is useful e.g. to compute a generating element of an ideal. The result is undefined
if x and y do not have the same discriminant.
The library syntax is GEN qfbcompraw(GEN x, GEN y).

3.8.83 qfbhclassno(x). Hurwitz class number of x, when x is nonnegative and congruent to 0 or

3 modulo 4, and 0 for other values. For x > 5 · 105 , we assume the GRH, and use quadclassunit
with default parameters.
? qfbhclassno(1) \\ not 0 or 3 mod 4
%1 = 0
? qfbhclassno(3)
%2 = 1/3
? qfbhclassno(4)
%3 = 1/2
? qfbhclassno(23)
%4 = 3
The library syntax is GEN hclassno(GEN x).

212
3.8.84 qfbnucomp(x, y, L). composition of the primitive positive definite binary quadratic forms
x and y (type t_QFI) using the NUCOMP and NUDUPL algorithms of Shanks, à la Atkin.
L is any positive constant, but for optimal speed, one should take L = |D/4|1/4 , i.e. sqrt-
nint(abs(D)>>2,4), where D is the common discriminant of x and y. When x and y do not have
the same discriminant, the result is undefined.
The current implementation is slower than the generic routine for small D, and becomes faster
when D has about 45 bits.
The library syntax is GEN nucomp(GEN x, GEN y, GEN L). Also available is GEN nudupl(GEN
x, GEN L) when x = y.

3.8.85 qfbnupow(x, n, {L}). n-th power of the primitive positive definite binary quadratic
form x using Shanks’s NUCOMP and NUDUPL algorithms; if set, L should be equal to sqrt-
nint(abs(D)>>2,4), where D < 0 is the discriminant of x.
The current implementation is slower than the generic routine for small discriminant D, and
becomes faster for D ≈ 245 .
The library syntax is GEN nupow(GEN x, GEN n, GEN L = NULL).

3.8.86 qfbpowraw(x, n). n-th power of the binary quadratic form x, computed without doing
any reduction (i.e. using qfbcompraw). Here n must be nonnegative and n < 231 .
The library syntax is GEN qfbpowraw(GEN x, long n).

3.8.87 qfbprimeform(x, p). Prime binary quadratic form of discriminant x whose first coefficient
is p, where |p| is a prime number. By abuse of notation, p = ±1 is also valid and returns the unit
form. Returns an error if x is not a quadratic residue mod p, or if x < 0 and p < 0. (Negative
definite t_QFI are not implemented.) In the case where x > 0, the “distance” component of the
form is set equal to zero according to the current precision.
The library syntax is GEN primeform(GEN x, GEN p, long prec).

3.8.88 qfbred(x, {flag = 0}, {d}, {isd }, {sd }). Reduces the binary quadratic form x (updating
Shanks’s distance function if x is indefinite). The binary digits of flag are toggles meaning
1: perform a single reduction step
2: don’t update Shanks’s distance
j√ k √
The arguments d, isd , sd , if present, supply the values of the discriminant, d , and d
respectively (no checking is done of these facts). If d < 0 these values are useless, and all references
to Shanks’s distance are irrelevant.
The library syntax is GEN qfbred0(GEN x, long flag, GEN d = NULL, GEN isd = NULL,
GEN sd = NULL). Also available are
GEN redimag(GEN x) (for definite x),
and for indefinite forms:
GEN redreal(GEN x)
GEN rhoreal(GEN x) (= qfbred(x,1)),
GEN redrealnod(GEN x, GEN isd) (= qfbred(x,2,,isd)),
GEN rhorealnod(GEN x, GEN isd) (= qfbred(x,3,,isd)).

213
3.8.89 qfbredsl2(x, {data}). Reduction of the (real or imaginary) binary quadratic form x, return
[y, g] where y is reduced and g in SL(2, Z) is such that g · x = y; data, if present, must be equal
to [D, sqrtint(D)], where D > 0 is the discriminant of x. In case x is a t_QFR, the distance
component is unaffected.

The library syntax is GEN qfbredsl2(GEN x, GEN data = NULL).

3.8.90 qfbsolve(Q, n, {flag = 0}). Solve the equation Q(x, y) = n in coprime integers x and y
(primitive solutions), where Q is a binary quadratic form and n an integer, up to the action of the
special orthogonal group G = SO(Q, Z), which is isomorphic to the group of units of positive norm
of the quadratic order of discriminant D = discQ. If D > 0, G is infinite. If D < −4, G is of order
2, if D = −3, G is of order 6 and if D = −4, G is of order 4.

Binary digits of flag mean: 1: return all solutions if set, else a single solution; return [] if a
single solution is wanted (bit unset) but none exist. 2: also include imprimitive solutions.

When flag = 2 (return a single solution, possibly imprimitive), the algorithm returns a solution
with minimal content; in particular, a primitive solution exists if and only if one is returned.

The integer n can be given by its factorization matrix fa = factor(n) or by the pair [n, fa].

? qfbsolve(Qfb(1,0,2), 603) \\ a single primitive solution

%1 = [5, 17]
? qfbsolve(Qfb(1,0,2), 603, 1) \\ all primitive solutions
%2 = [[5, 17], [-19, -11], [19, -11], [5, -17]]
? qfbsolve(Qfb(1,0,2), 603, 2) \\ a single, possibly imprimitive solution
%3 = [5, 17] \\ actually primitive
? qfbsolve(Qfb(1,0,2), 603, 3) \\ all solutions
%4 = [[5, 17], [-19, -11], [19, -11], [5, -17], [-21, 9], [-21, -9]]
? N = 2^128+1; F = factor(N);
? qfbsolve(Qfb(1,0,1),[N,F],1)
%3 = [[-16382350221535464479,8479443857936402504],
[18446744073709551616,-1],[-18446744073709551616,-1],
[16382350221535464479,8479443857936402504]]

For fixed Q, assuming the factorisation of n is given, the algorithm runs in probabilistic
polynomial time in log p, where p is the largest prime divisor of n, through the computation of
square roots of D modulo 4p). The dependency on Q is more complicated: polynomial time in
log |D| if Q is imaginary, but exponential time if Q is real (through the computation of a full cycle
of reduced forms). In the latter case, note that bnfisprincipal provides a solution in heuristic
subexponential time assuming the GRH.

The library syntax is GEN qfbsolve(GEN Q, GEN n, long flag).

214
3.8.91 quadclassunit(D, {flag = 0}, {tech = [ ]}). Buchmann-McCurley’s sub-exponential algo-
rithm for computing the class group of a quadratic order of discriminant D.
This function should be used instead of qfbclassno or quadregulator when D < −1025 ,
D > 1010 , or when the structure is wanted. It is a special case of bnfinit, which is slower, but
more robust.
The result is a vector v whose components should be accessed using member functions:
• v.no: the class number
• v.cyc: a vector giving the structure of the class group as a product of cyclic groups;
• v.gen: a vector giving generators of those cyclic groups (as binary quadratic forms).
• v.reg: the regulator, computed to an accuracy which is the maximum of an internal accuracy
determined by the program and the current default (note that once the regulator is known to a
small accuracy it is trivial to compute it to very high accuracy, see the tutorial).
The flag is obsolete and should be left alone. In older versions, it supposedly computed the
narrow class group when D > 0, but this did not work at all; use the general function bnfnarrow.
Optional parameter tech is a row vector of the form [c1 , c2 ], where c1 ≤ c2 are nonnegative real
numbers which control the execution time and the stack size, see 3.13.7. The parameter is used as a
threshold to balance the relation finding phase against the final linear algebra. Increasing the default
c1 means that relations are easier to find, but more relations are needed and the linear algebra will
be harder. The default value for c1 is 0 and means that it is taken equal to c2 . The parameter c2 is
mostly obsolete and should not be changed, but we still document it for completeness: we compute
a tentative class group by generators and relations using a factorbase of prime ideals ≤ c1 (log |D|)2 ,
then prove that ideals of norm ≤ c2 (log |D|)2 do not generate a larger group. By default an optimal
c2 is chosen, so that the result is provably correct under the GRH — a famous result of Bach states
that c2 = 6 is fine, but it is possible to improve on this algorithmically. You may provide a smaller
c2 , it will be ignored (we use the provably correct one); you may provide a larger c2 than the default
value, which results in longer computing times for equally correct outputs (under GRH).
The library syntax is GEN quadclassunit0(GEN D, long flag, GEN tech = NULL, long
prec). If you really need to experiment with the tech parameter, it is usually more convenient
to use GEN Buchquad(GEN D, double c1, double c2, long prec). If only the class number
is needed, GEN quadclassno(GEN D) will be faster (still assuming the GRH), but will not provide
the group structure. For negative D, |D| < 1020 , qfbclassno should be faster but may return a
wrong result.
√
3.8.92 quaddisc(x). Discriminant of the étale algebra Q( x), where x ∈ Q∗ . This is the same as
coredisc(d) where d is the integer squarefree part of x, so x = df 2 with f ∈ Q∗ √and d ∈ Z. This
returns 0 for x = 0, 1 for x square and the discriminant of the quadratic field Q( x) otherwise.
? quaddisc(7)
%1 = 28
? quaddisc(-7)
%2 = -7
The library syntax is GEN quaddisc(GEN x).

215
 √
3.8.93 quadgen(D, {v =0 w}). Creates the quadratic number ω = (a + D)/2 where a = 0 if
D ≡ 0 mod 4, a = 1 if D ≡ 1 mod 4, so that (1, ω) is an integral basis for the quadratic order of
discriminant D. D must be an integer congruent to 0 or 1 modulo 4, which is not a square. If v is
given, the variable name is used to display g else ’w’ is used.
? w = quadgen(5, ’w); w^2 - w - 1
%1 = 0
? w = quadgen(0, ’w)
*** at top-level: w=quadgen(0)
*** ^----------
*** quadgen: domain error in quadpoly: issquare(disc) = 1
The library syntax is GEN quadgen0(GEN D, long v = -1) where v is a variable number.
When v does not matter, the function GEN quadgen(GEN D) is also available.

3.8.94 quadhilbert(D). Relative equation defining the Hilbert class field of the quadratic field of
discriminant D.
If D < 0, uses complex multiplication (Schertz’s variant).
If D > 0 Stark units are used and (in rare cases) a vector of extensions may be returned whose
compositum is the requested class field. See bnrstark for details.
The library syntax is GEN quadhilbert(GEN D, long prec).

3.8.95 quadpoly(D, {v =0 x}). Creates the “canonical” quadratic polynomial (in the variable v)
corresponding to the discriminant D, i.e. the minimal polynomial of quadgen(D). D must be an
integer congruent to 0 or 1 modulo 4, which is not a square.
? quadpoly(5,’y)
%1 = y^2 - y - 1
? quadpoly(0,’y)
*** at top-level: quadpoly(0,’y)
*** ^--------------
*** quadpoly: domain error in quadpoly: issquare(disc) = 1
The library syntax is GEN quadpoly0(GEN D, long v = -1) where v is a variable number.

3.8.96 quadray(D, f ). Relative equation for the ray class field of conductor f for the quadratic
field of discriminant D using analytic methods. A bnf for x2 − D is also accepted in place of D.
For D < 0, uses the σ function and Schertz’s method.
For D > 0, uses Stark’s conjecture, and a vector of relative equations may be returned. See
bnrstark for more details.
The library syntax is GEN quadray(GEN D, GEN f, long prec).

3.8.97 quadregulator(x). Regulator of the quadratic field of positive discriminant x. Returns an

error if x is not a discriminant (fundamental or not) or if x is a square. See also quadclassunit if
x is large.
The library syntax is GEN quadregulator(GEN x, long prec).

216
 √
3.8.98 quadunit(D, {v =0 w}). Fundamental unit u of the real quadratic field Q( D) where
D is the positive discriminant of the field. If D is not a fundamental discriminant, this probably
gives the fundamental unit of the corresponding order. D must be an integer congruent to 0 or 1
modulo 4, which is not a square; the result is a quadratic number (see Section 3.8.93). If v is given,
the variable name
√ is used√to display u else ’w’ is used. The algorithm computes the continued
fraction of (1 + √D)/2 or D/2 (see GTM 138, algorithm 5.7.2). Although the continued fraction
length is only O( D), the function still runs in time Õ(D), in part because the output size is not
polynomially bounded in terms of log D. See bnfinit and bnfunits for a better alternative for
large D, running in time subexponential in log D and returning the fundamental units in compact
form (as a short list of S-units of size O(log D)3 raised to possibly large exponents).
The library syntax is GEN quadunit0(GEN D, long v = -1) where v is a variable number.
When v does not matter, the function GEN quadunit(GEN D) is also available.

3.8.99 ramanujantau(n). Compute the value of Ramanujan’s tau function at an individual n,

assuming the truth of the GRH (to compute quickly class numbers of imaginary quadratic fields
using quadclassunit). Algorithm in Õ(n1/2 ) using O(log n) space. If all values up to N are
required, then X Y
τ (n)q n = q (1 − q n )24
n≥1

will produce them in time Õ(N ), against Õ(N 3/2 ) for individual calls to ramanujantau; of course
the space complexity then becomes Õ(N ).
? tauvec(N) = Vec(q*eta(q + O(q^N))^24);
? N = 10^4; v = tauvec(N);
time = 26 ms.
? ramanujantau(N)
%3 = -482606811957501440000
? w = vector(N, n, ramanujantau(n)); \\ much slower !
time = 13,190 ms.
? v == w
%4 = 1
The library syntax is GEN ramanujantau(GEN n).

3.8.100 randomprime({N = 231 }, {q}). Returns a strong pseudo prime (see ispseudoprime) in
[2, N − 1]. A t_VEC N = [a, b] is also allowed, with a ≤ b in which case a pseudo prime a ≤ p ≤ b is
returned; if no prime exists in the interval, the function will run into an infinite loop. If the upper
bound is less than 264 the pseudo prime returned is a proven prime.
? randomprime(100)
%1 = 71
? randomprime([3,100])
%2 = 61
? randomprime([1,1])
*** at top-level: randomprime([1,1])
*** ^------------------
*** randomprime: domain error in randomprime:
*** floor(b) - max(ceil(a),2) < 0
? randomprime([24,28]) \\ infinite loop

217
 If the optional parameter q is an integer, return a prime congruent to 1 mod q; if q is an intmod,
return a prime in the given congruence class. If the class contains no prime in the given interval,
the function will raise an exception if the class is not invertible, else run into an infinite loop
? randomprime(100, 4) \\ 1 mod 4
%1 = 71
? randomprime(100, 4)
%2 = 13
? randomprime([10,100], Mod(2,5))
%3 = 47
? randomprime(100, Mod(0,2)) \\ silly but works
%4 = 2
? randomprime([3,100], Mod(0,2)) \\ not invertible
*** at top-level: randomprime([3,100],Mod(0,2))
*** ^-----------------------------
*** randomprime: elements not coprime in randomprime:
0
2
? randomprime(100, 97) \\ infinite loop
The library syntax is GEN randomprime0(GEN N = NULL, GEN q = NULL). Also available is
GEN randomprime(GEN N = NULL).

3.8.101 removeprimes({x = [ ]}). Removes the primes listed in x from the prime number table.
In particular removeprimes(addprimes()) empties the extra prime table. x can also be a single
integer. List the current extra primes if x is omitted.
The library syntax is GEN removeprimes(GEN x = NULL).

3.8.102 sigma(x, {k = 1}). Sum of the k th powers of the positive divisors of |x|. x and k must be
of type integer.
The library syntax is GEN sumdivk(GEN x, long k). Also available is GEN sumdiv(GEN n)
, for k = 1.

3.8.103 sqrtint(x, {&r}). Returns the integer square root of x, i.e. the largest integer y such that
y 2 ≤ x, where x a nonnegative integer. If r is present, set it to the remainder r = x − y 2 , which
satisfies 0 ≤ r ≤ 2y
? x = 120938191237; sqrtint(x)
%1 = 347761
? sqrt(x)
%2 = 347761.68741970412747602130964414095216
? y = sqrtint(x, &r)
%3 = 347761
? x - y^2
%4 = 478116
The library syntax is GEN sqrtint0(GEN x, GEN *r = NULL). Also available is GEN
sqrtint(GEN a).

218
3.8.104 sqrtnint(x, n). Returns the integer n-th root of x, i.e. the largest integer y such that
y n ≤ x, where x is a nonnegative integer.
? N = 120938191237; sqrtnint(N, 5)
%1 = 164
? N^(1/5)
%2 = 164.63140849829660842958614676939677391
The special case n = 2 is sqrtint
The library syntax is GEN sqrtnint(GEN x, long n).

3.8.105 sumdedekind(h, k). Returns the Dedekind sum attached to the integers h and k, corre-
sponding to a fast implementation of
s(h,k) = sum(n = 1, k-1, (n/k)*(frac(h*n/k) - 1/2))

The library syntax is GEN sumdedekind(GEN h, GEN k).

3.8.106 sumdigits(n, {B = 10}). Sum of digits in the integer |n|, when written in base B > 1.
? sumdigits(123456789)
%1 = 45
? sumdigits(123456789, 2)
%1 = 16
Note that the sum of bits in n is also returned by hammingweight. This function is much faster
than vecsum(digits(n,B)) when B is 10 or a power of 2, and only slightly faster in other cases.
The library syntax is GEN sumdigits0(GEN n, GEN B = NULL). Also available is GEN sumdig-
its(GEN n), for B = 10.

3.8.107 znchar(D). Given a datum D describing a group (Z/N Z)∗ and a Dirichlet character χ,
return the pair [G, chi], where G is znstar(N, 1)) and chi is a GP character.
The following possibilities for D are supported
• a nonzero t_INT congruent to 0, 1 modulo 4, return the real character modulo D given by
the Kronecker symbol (D/.);
• a t_INTMOD Mod(m, N), return the Conrey character modulo N of index m (see znconrey-
log).
• a modular form space as per mfinit([N, k, χ]) or a modular form for such a space, return
the underlying Dirichlet character χ (which may be defined modulo a divisor of N but need not be
primitive).
In the remaining cases, G is initialized by znstar(N, 1).
• a pair [G, chi], where chi is a standard GP Dirichlet character c = (cj ) on G (generic
character t_VEC or Conrey characters t_COL or t_INT); given generators G = ⊕(Z/dj Z)gj , χ(gj ) =
e(cj /dj ).
• a pair [G, chin], where chin is a normalized representation [n, c̃] of the Dirichlet character
c; χ(gj ) = e(c̃j /n) where n is minimal (order of χ).

219
 ? [G,chi] = znchar(-3);
? G.cyc
%2 = [2]
? chareval(G, chi, 2)
%3 = 1/2
? kronecker(-3,2)
%4 = -1
? znchartokronecker(G,chi)
%5 = -3
? mf = mfinit([28, 5/2, Mod(2,7)]); [f] = mfbasis(mf);
? [G,chi] = znchar(mf); [G.mod, chi]
%7 = [7, [2]~]
? [G,chi] = znchar(f); chi
%8 = [28, [0, 2]~]
The library syntax is GEN znchar(GEN D).

3.8.108 zncharconductor(G, chi ). Let G be attached to (Z/qZ)∗ (as per G = znstar(q, 1))
and chi be a Dirichlet character on (Z/qZ)∗ (see Section 3.8.3 or ??character). Return the
conductor of chi:
? G = znstar(126000, 1);
? zncharconductor(G,11) \\ primitive
%2 = 126000
? zncharconductor(G,1) \\ trivial character, not primitive!
%3 = 1
? zncharconductor(G,1009) \\ character mod 5^3
%4 = 125
The library syntax is GEN zncharconductor(GEN G, GEN chi).

chi , Q). Let N = p pep and a Dirichlet character χ, we have a

Q
3.8.109 znchardecompose(G,
decomposition χ = p χp into character modulo N where the conductor of χp divides pep ; it equals
Q
pep for all p if and only if χ is primitive.
∗
Given
Q a znstar G describing a group (Z/N Z) , a Dirichlet character chi and an integer Q,
return p|(Q,N ) χp . For instance, if Q = p is a prime divisor of N , the function returns χp (as a
character modulo N ), given as a Conrey character (t_COL).
? G = znstar(40, 1);
? G.cyc
%2 = [4, 2, 2]
? chi = [2, 1, 1];
? chi2 = znchardecompose(G, chi, 2)
%4 = [1, 1, 0]~
? chi5 = znchardecompose(G, chi, 5)
%5 = [0, 0, 2]~
? znchardecompose(G, chi, 3)
%6 = [0, 0, 0]~
? c = charmul(G, chi2, chi5)
%7 = [1, 1, 2]~ \\ t_COL: in terms of Conrey generators !

220
 ? znconreychar(G,c)
%8 = [2, 1, 1] \\ t_VEC: in terms of SNF generators
The library syntax is GEN znchardecompose(GEN G, GEN chi, GEN Q).

3.8.110 znchargauss(G, chi , {a = 1}). Given a Dirichlet character χ on G = (Z/N Z)∗ (see
znchar), return the complex Gauss sum

N
X
g(χ, a) = χ(n)e(an/N )
n=1

? [G,chi] = znchar(-3); \\ quadratic Gauss sum: I*sqrt(3)

? znchargauss(G,chi)
%2 = 1.7320508075688772935274463415058723670*I
? [G,chi] = znchar(5);
? znchargauss(G,chi) \\ sqrt(5)
%2 = 2.2360679774997896964091736687312762354
? G = znstar(300,1); chi = [1,1,12]~;
? znchargauss(G,chi) / sqrt(300) - exp(2*I*Pi*11/25) \\ = 0
%4 = 2.350988701644575016 E-38 + 1.4693679385278593850 E-39*I
? lfuntheta([G,chi], 1) \\ = 0
%5 = -5.79[...] E-39 - 2.71[...] E-40*I
The library syntax is GEN znchargauss(GEN G, GEN chi, GEN a = NULL, long bitprec)
.

3.8.111 zncharinduce(G, chi , N ). Let G be attached to (Z/qZ)∗ (as per G = znstar(q,1)) and
let chi be a Dirichlet character on (Z/qZ)∗ , given by
• a t_VEC: a standard character on bid.gen,
• a t_INT or a t_COL: a Conrey index in (Z/qZ)∗ or its Conrey logarithm; see Section 3.8.3 or
??character.
Let N be a multiple of q, return the character modulo N extending chi. As usual for arithmetic
functions, the new modulus N can be given as a t_INT, via a factorization matrix or a pair [N,
factor(N)], or by znstar(N,1).
? G = znstar(4, 1);
? chi = znconreylog(G,1); \\ trivial character mod 4
? zncharinduce(G, chi, 80) \\ now mod 80
%3 = [0, 0, 0]~
? zncharinduce(G, 1, 80) \\ same using directly Conrey label
%4 = [0, 0, 0]~
? G2 = znstar(80, 1);
? zncharinduce(G, 1, G2) \\ same
%4 = [0, 0, 0]~
? chi = zncharinduce(G, 3, G2) \\ extend the nontrivial character mod 4
%5 = [1, 0, 0]~
? [G0,chi0] = znchartoprimitive(G2, chi);
? G0.mod

221
 %7 = 4
? chi0
%8 = [1]~

Here is a larger example:

? G = znstar(126000, 1);
? label = 1009;
? chi = znconreylog(G, label)
%3 = [0, 0, 0, 14, 0]~
? [G0,chi0] = znchartoprimitive(G, label); \\ works also with ’chi’
? G0.mod
%5 = 125
? chi0 \\ primitive character mod 5^3 attached to chi
%6 = [14]~
? G0 = znstar(N0, 1);
? zncharinduce(G0, chi0, G) \\ induce back
%8 = [0, 0, 0, 14, 0]~
? znconreyexp(G, %)
%9 = 1009

The library syntax is GEN zncharinduce(GEN G, GEN chi, GEN N).

3.8.112 zncharisodd(G, chi ). Let G be attached to (Z/N Z)∗ (as per G = znstar(N,1)) and let
chi be a Dirichlet character on (Z/N Z)∗ , given by

• a t_VEC: a standard character on G.gen,

• a t_INT or a t_COL: a Conrey index in (Z/qZ)∗ or its Conrey logarithm; see Section 3.8.3 or
??character.

Return 1 if and only if chi(−1) = −1 and 0 otherwise.

? G = znstar(8, 1);
? zncharisodd(G, 1) \\ trivial character
%2 = 0
? zncharisodd(G, 3)
%3 = 1
? chareval(G, 3, -1)
%4 = 1/2

The library syntax is long zncharisodd(GEN G, GEN chi).

222
3.8.113 znchartokronecker(G, chi , {flag = 0}). Let G be attached to (Z/N Z)∗ (as per G =
znstar(N,1)) and let chi be a Dirichlet character on (Z/N Z)∗ , given by

• a t_VEC: a standard character on bid.gen,

• a t_INT or a t_COL: a Conrey index in (Z/qZ)∗ or its Conrey logarithm; see Section 3.8.3 or
??character.

If flag = 0, return the discriminant D if chi is real equal to the Kronecker symbol (D/.) and
0 otherwise. The discriminant D is fundamental if and only if chi is primitive.

If flag = 1, return the fundamental discriminant attached to the corresponding primitive

character.

? G = znstar(8,1); CHARS = [1,3,5,7]; \\ Conrey labels

? apply(t->znchartokronecker(G,t), CHARS)
%2 = [4, -8, 8, -4]
? apply(t->znchartokronecker(G,t,1), CHARS)
%3 = [1, -8, 8, -4]

The library syntax is GEN znchartokronecker(GEN G, GEN chi, long flag).

3.8.114 znchartoprimitive(G, chi ). Let G be attached to (Z/qZ)∗ (as per G = znstar(q, 1))
and chi be a Dirichlet character on (Z/qZ)∗ , of conductor q0 | q.

? G = znstar(126000, 1);
? [G0,chi0] = znchartoprimitive(G,11)
? G0.mod
%3 = 126000
? chi0
%4 = 11
? [G0,chi0] = znchartoprimitive(G,1);\\ trivial character, not primitive!
? G0.mod
%6 = 1
? chi0
%7 = []~
? [G0,chi0] = znchartoprimitive(G,1009)
? G0.mod
%4 = 125
? chi0
%5 = [14]~

Note that znconreyconductor is more efficient since it can return χ0 and its conductor q0 without
needing to initialize G0 . The price to pay is a more cryptic format and the need to initalize G0
later, but that needs to be done only once for all characters with conductor q0 .

The library syntax is GEN znchartoprimitive(GEN G, GEN chi).

223
3.8.115 znconreychar(G, m). Given a znstar G attached to (Z/qZ)∗ (as per G = znstar(q,1)),
this function returns the Dirichlet character attached to m ∈ (Z/qZ)∗ via Conrey’s logarithm,
which establishes a “canonical” bijection between (Z/qZ)∗ and its dual.
Let q = p pep be the factorization of q into distinct primes. For all odd p with ep > 0, let gp
Q
be the element in (Z/qZ)∗ which is
• congruent to 1 mod q/pep ,
• congruent mod pep to the smallest positive integer that generates (Z/p2 Z)∗ .
For p = 2, we let g4 (if 2e2 ≥ 4) and g8 (if furthermore (2e2 ≥ 8) be the elements in (Z/qZ)∗
which are
• congruent to 1 mod q/2e2 ,
• g4 = −1 mod 2e2 ,
• g8 = 5 mod 2e2 .
Then the gp (and the extra g4 and g8 if 2e2Q≥ 2) are independent generators of (Z/qZ)∗ , i.e.
m
every m in (Z/qZ)∗ can be written uniquely as p gp p , where mp is defined modulo the order op
of gp and p ∈ Sq , the set of prime divisors of q together with 4 if 4 | q and 8 if 8 | q. Note that the
gp are in general not SNF generators as produced by znstar whenever ω(q) ≥ 2, although their
number is the same. They however allow to handle the finite abelian group (Z/qZ)∗ in a fast and
elegant way. (Which unfortunately does not generalize to ray class groups or Hecke characters.)
The Conrey logarithm of m is the vector (mp )p∈Sq , obtained via znconreylog. The Conrey
character χq (m, ·) attached to m mod q maps each gp , p ∈ Sq to e(mp /op ), where e(x) = exp(2iπx).
This function returns the Conrey character expressed in the standard PARI way in terms of the
SNF generators G.gen.
? G = znstar(8,1);
? G.cyc
%2 = [2, 2] \\ Z/2 x Z/2
? G.gen
%3 = [7, 3]
? znconreychar(G,1) \\ 1 is always the trivial character
%4 = [0, 0]
? znconreychar(G,2) \\ 2 is not coprime to 8 !!!
*** at top-level: znconreychar(G,2)
*** ^-----------------
*** znconreychar: elements not coprime in Zideallog:
2
8
*** Break loop: type ’break’ to go back to GP prompt
break>
? znconreychar(G,3)
%5 = [0, 1]
? znconreychar(G,5)
%6 = [1, 1]
? znconreychar(G,7)
%7 = [1, 0]

224
We indeed get all 4 characters of (Z/8Z)∗ .

For convenience, we allow to input the Conrey logarithm of m instead of m:

? G = znstar(55, 1);
? znconreychar(G,7)
%2 = [7, 0]
? znconreychar(G, znconreylog(G,7))
%3 = [7, 0]

The library syntax is GEN znconreychar(GEN G, GEN m).

3.8.116 znconreyconductor(G, chi , {&chi0 }). Let G be attached to (Z/qZ)∗ (as per G = zn-
star(q, 1)) and chi be a Dirichlet character on (Z/qZ)∗ , given by

• a t_VEC: a standard character on bid.gen,

• a t_INT or a t_COL: a Conrey index in (Z/qZ)∗ or its Conrey logarithm; see Section 3.8.3 or
??character.

Return the conductor of chi, as the t_INT bid.mod if chi is primitive, and as a pair [N, faN]
(with faN the factorization of N ) otherwise.

If chi0 is present, set it to the Conrey logarithm of the attached primitive character.

? G = znstar(126000, 1);
? znconreyconductor(G,11) \\ primitive
%2 = 126000
? znconreyconductor(G,1) \\ trivial character, not primitive!
%3 = [1, matrix(0,2)]
? N0 = znconreyconductor(G,1009, &chi0) \\ character mod 5^3
%4 = [125, Mat([5, 3])]
? chi0
%5 = [14]~
? G0 = znstar(N0, 1); \\ format [N,factor(N)] accepted
? znconreyexp(G0, chi0)
%7 = 9
? znconreyconductor(G0, chi0) \\ now primitive, as expected
%8 = 125

The group G0 is not computed as part of znconreyconductor because it needs to be computed

only once per conductor, not once per character.

The library syntax is GEN znconreyconductor(GEN G, GEN chi, GEN *chi0 = NULL).

225
3.8.117 znconreyexp(G, chi ). Given a znstar G attached to (Z/qZ)∗ (as per G = znstar(q,
1)), this function returns the Conrey exponential of the character chi : it returns the integer
m ∈ (Z/qZ)∗ such that znconreylog(G, m) is chi .
The character chi is given either as a
• t_VEC: in terms of the generators G.gen;
• t_COL: a Conrey logarithm.
? G = znstar(126000, 1)
? znconreylog(G,1)
%2 = [0, 0, 0, 0, 0]~
? znconreyexp(G,%)
%3 = 1
? G.cyc \\ SNF generators
%4 = [300, 12, 2, 2, 2]
? chi = [100, 1, 0, 1, 0]; \\ some random character on SNF generators
? znconreylog(G, chi) \\ in terms of Conrey generators
%6 = [0, 3, 3, 0, 2]~
? znconreyexp(G, %) \\ apply to a Conrey log
%7 = 18251
? znconreyexp(G, chi) \\ ... or a char on SNF generators
%8 = 18251
? znconreychar(G,%)
%9 = [100, 1, 0, 1, 0]
The library syntax is GEN znconreyexp(GEN G, GEN chi).

3.8.118 znconreylog(G, m). Given a znstar attached to (Z/qZ)∗ (as per G = znstar(q,1)), this
function returns the Conrey logarithm of m ∈ (Z/qZ)∗ .
Let q = p pep be the factorization of q into distinct primes, where we assume e2 = 0 or
Q
e2 ≥ 2. (If e2 = 1, we can ignore 2 from the factorization, as if we replaced q by q/2, since
(Z/qZ)∗ ∼ (Z/(q/2)Z)∗ .)
For all odd p with ep > 0, let gp be the element in (Z/qZ)∗ which is
• congruent to 1 mod q/pep ,
• congruent mod pep to the smallest positive integer that generates (Z/p2 Z)∗ .
For p = 2, we let g4 (if 2e2 ≥ 4) and g8 (if furthermore (2e2 ≥ 8) be the elements in (Z/qZ)∗
which are
• congruent to 1 mod q/2e2 ,
• g4 = −1 mod 2e2 ,
• g8 = 5 mod 2e2 .
Then the gp (and the extra g4 and g8 if 2eQ 2
≥ 2) are independent generators of Z/qZ∗ , i.e.
∗ m
every m in (Z/qZ) can be written uniquely as p gp p , where mp is defined modulo the order op
of gp and p ∈ Sq , the set of prime divisors of q together with 4 if 4 | q and 8 if 8 | q. Note that the
gp are in general not SNF generators as produced by znstar whenever ω(q) ≥ 2, although their

226
number is the same. They however allow to handle the finite abelian group (Z/qZ)∗ in a fast and
elegant way. (Which unfortunately does not generalize to ray class groups or Hecke characters.)

The Conrey logarithm of m is the vector (mp )p∈Sq . The inverse function znconreyexp recovers
the Conrey label m from a character.

? G = znstar(126000, 1);
? znconreylog(G,1)
%2 = [0, 0, 0, 0, 0]~
? znconreyexp(G, %)
%3 = 1
? znconreylog(G,2) \\ 2 is not coprime to modulus !!!
*** at top-level: znconreylog(G,2)
*** ^-----------------
*** znconreylog: elements not coprime in Zideallog:
2
126000
*** Break loop: type ’break’ to go back to GP prompt
break>
? znconreylog(G,11) \\ wrt. Conrey generators
%4 = [0, 3, 1, 76, 4]~
? log11 = ideallog(,11,G) \\ wrt. SNF generators
%5 = [178, 3, -75, 1, 0]~

For convenience, we allow to input the ordinary discrete log of m, ideallog(, m, bid), which
allows to convert discrete logs from bid.gen generators to Conrey generators.

? znconreylog(G, log11)
%7 = [0, 3, 1, 76, 4]~

We also allow a character (t_VEC) on bid.gen and return its representation on the Conrey gener-
ators.

? G.cyc
%8 = [300, 12, 2, 2, 2]
? chi = [10,1,0,1,1];
? znconreylog(G, chi)
%10 = [1, 3, 3, 10, 2]~
? n = znconreyexp(G, chi)
%11 = 84149
? znconreychar(G, n)
%12 = [10, 1, 0, 1, 1]

The library syntax is GEN znconreylog(GEN G, GEN m).

227
3.8.119 zncoppersmith(P, N, X, {B = N }). Coppersmith’s algorithm. N being an integer and
P ∈ Z[t], finds in polynomial time in log(N ) and d = deg(P ) all integers x with |x| ≤ X such that

gcd(N, P (x)) ≥ B.

This is a famous application of the LLL algorithm meant to help in the factorization of N . Notice
that P may be reduced modulo N Z[t] without affecting the situation. The parameter X must not
be too large: assume for now that the leading coefficient of P is coprime to N , then we must have

d log X log N < log2 B,

i.e., X < N 1/d when B = N . Let now P0 be the gcd of the leading coefficient of P and N . In
applications to factorization, we should have P0 = 1; otherwise, either P0 = N and we can reduce
the degree of P , or P0 is a non trivial factor of N . For completeness, we nevertheless document the
exact conditions that X must satisfy in this case: let p := logN P0 , b := logN B, x := logN X, then
• either p ≥ d/(2d − 1) is large and we must have xd < 2b − 1;
• or p < d/(2d − 1) and we must have both p < b < 1 − p + p/d and x(d + p(1 − 2d)) < (b − p)2 .
Note that this reduces to xd < b2 when p = 0, i.e., the condition described above.
Some x larger than X may be returned if you are very lucky. The routine runs in polynomial
time in log N and d but the smaller B, or the larger X, the slower. The strength of Coppersmith
method is the ability to find roots modulo a general composite N : if N is a prime or a prime power,
polrootsmod or polrootspadic will be much faster.
We shall now present two simple applications. The first one is finding nontrivial factors of N ,
given some partial information on the factors; in that case B must obviously be smaller than the
largest nontrivial divisor of N .
setrand(1); \\ to make the example reproducible
[a,b] = [10^30, 10^31]; D = 20;
p = randomprime([a,b]);
q = randomprime([a,b]); N = p*q;
\\ assume we know 0) p | N; 1) p in [a,b]; 2) the last D digits of p
p0 = p % 10^D;
? L = zncoppersmith(10^D*x + p0, N, b \ 10^D, a)
time = 1ms.
%6 = [738281386540]
? gcd(L[1] * 10^D + p0, N) == p
%7 = 1
and we recovered p, faster than by trying all possibilities x < 1011 .
The second application is an attack on RSA with low exponent, when the message x is short
and the padding P is known to the attacker. We use the same RSA modulus N as in the first
example:
setrand(1);
P = random(N); \\ known padding
e = 3; \\ small public encryption exponent
X = floor(N^0.3); \\ N^(1/e - epsilon)
x0 = random(X); \\ unknown short message

228
 C = lift( (Mod(x0,N) + P)^e ); \\ known ciphertext, with padding P
zncoppersmith((P + x)^3 - C, N, X)
\\ result in 244ms.
%14 = [2679982004001230401]
? %[1] == x0
%15 = 1
We guessed an integer of the order of 1018 , almost instantly.
The library syntax is GEN zncoppersmith(GEN P, GEN N, GEN X, GEN B = NULL).

3.8.120 znlog(x, g, {o}). This functions allows two distinct modes of operation depending on g:
• if g is the output of znstar (with initialization), we compute the discrete logarithm of x
with respect to the generators contained in the structure. See ideallog for details.
• else g is an explicit element in (Z/N Z)∗ , we compute the discrete logarithm of x in (Z/N Z)∗
in base g. The rest of this entry describes the latter possibility.
The result is [] when x is not a power of g, though the function may also enter an infinite loop
in this case.
If present, o represents the multiplicative order of g, see Section 3.8.2; the preferred format
for this parameter is [ord, factor(ord)], where ord is the order of g. This provides a definite
speedup when the discrete log problem is simple:
? p = nextprime(10^4); g = znprimroot(p); o = [p-1, factor(p-1)];
? for(i=1,10^4, znlog(i, g, o))
time = 163 ms.
? for(i=1,10^4, znlog(i, g))
time = 200 ms. \\ a little slower
The result is undefined if g is not invertible mod N or if the supplied order is incorrect.
This function uses
• a combination of generic discrete log algorithms (see below).
• in (Z/N Z)∗ when N is prime: a linear sieve index calculus method, suitable for N < 1050 ,
say, is used for large prime divisors of the order.
The generic discrete log algorithms are:
• Pohlig-Hellman algorithm, to reduce to groups of prime order q, where q|p − 1 and p is an
odd prime divisor of N ,
• Shanks baby-step/giant-step (q < 232 is small),
• Pollard rho method (q > 232 ).
√
The latter two algorithms require O( q) operations in the group on average, hence will not
be able to treat cases where q > 1030 , say. In addition, Pollard rho is not able to handle the case
where there are no solutions: it will enter an infinite loop.
? g = znprimroot(101)
%1 = Mod(2,101)
? znlog(5, g)

229
 %2 = 24
? g^24
%3 = Mod(5, 101)
? G = znprimroot(2 * 101^10)
%4 = Mod(110462212541120451003, 220924425082240902002)
? znlog(5, G)
%5 = 76210072736547066624
? G^% == 5
%6 = 1
? N = 2^4*3^2*5^3*7^4*11; g = Mod(13, N); znlog(g^110, g)
%7 = 110
? znlog(6, Mod(2,3)) \\ no solution
%8 = []

For convenience, g is also allowed to be a p-adic number:

? g = 3+O(5^10); znlog(2, g)
%1 = 1015243
? g^%
%2 = 2 + O(5^10)

The library syntax is GEN znlog0(GEN x, GEN g, GEN o = NULL). The function GEN zn-
log(GEN x, GEN g, GEN o) is also available

3.8.121 znorder(x, {o}). x must be an integer mod n, and the result is the order of x in the
multiplicative group (Z/nZ)∗ . Returns an error if x is not invertible. The parameter o, if present,
represents a nonzero multiple of the order of x, see Section 3.8.2; the preferred format for this
parameter is [ord, factor(ord)], where ord = eulerphi(n) is the cardinality of the group.

The library syntax is GEN znorder(GEN x, GEN o = NULL). Also available is GEN order(GEN
x).

3.8.122 znprimroot(n). Returns a primitive root (generator) of (Z/nZ)∗ , whenever this latter
group is cyclic (n = 4 or n = 2pk or n = pk , where p is an odd prime and k ≥ 0). If the group is
not cyclic, the result is undefined. If n is a prime power, then the smallest positive primitive root
is returned. This may not be true for n = 2pk , p odd.

Note that this function requires factoring p − 1 for p as above, in order to determine the exact
order of elements in (Z/nZ)∗ : this is likely to be costly if p is large.

The library syntax is GEN znprimroot(GEN n).

230
3.8.123 znstar(n, {flag = 0}). Gives the structure of the multiplicative group (Z/nZ)∗ . The
output G depends on the value of flag:
• flag = 0 (default), an abelian group structure [h, d, g], where h = φ(n) is the order (G.no), d
(G.cyc) is a k-component row-vector d of integers di such that di > 1, di | di−1 for i ≥ 2 and

k
Y
∗
(Z/nZ) ' (Z/di Z),
i=1

and g (G.gen) is a k-component row vector giving generators of the image of the cyclic groups
Z/di Z.
• flag = 1 the result is a bid structure; this allows computing discrete logarithms using znlog
(also in the noncyclic case!).
? G = znstar(40)
%1 = [16, [4, 2, 2], [Mod(17, 40), Mod(21, 40), Mod(11, 40)]]
? G.no \\ eulerphi(40)
%2 = 16
? G.cyc \\ cycle structure
%3 = [4, 2, 2]
? G.gen \\ generators for the cyclic components
%4 = [Mod(17, 40), Mod(21, 40), Mod(11, 40)]
? apply(znorder, G.gen)
%5 = [4, 2, 2]
For user convenience, we define znstar(0) as [2, [2], [-1]], corresponding to Z∗ , but flag = 1
is not implemented in this trivial case.
The library syntax is GEN znstar0(GEN n, long flag).

3.9 Polynomials and power series.

We group here all functions which are specific to polynomials or power series. Many other
functions which can be applied on these objects are described in the other sections. Also, some of
the functions described here can be applied to other types.

3.9.1 O(p^e). If p is an integer greater than 2, returns a p-adic 0 of precision e. In all other cases,
returns a power series zero with precision given by ev, where v is the X-adic valuation of p with
respect to its main variable.
The library syntax is GEN ggrando(). GEN zeropadic(GEN p, long e) for a p-adic and GEN
zeroser(long v, long e) for a power series zero in variable v.

3.9.2 bezoutres(A, B, {v}). Deprecated alias for polresultantext

The library syntax is GEN polresultantext0(GEN A, GEN B, long v = -1) where v is a
variable number.

231
3.9.3 deriv(x, {v}). Derivative of x with respect to the main variable if v is omitted, and with
respect to v otherwise. The derivative of a scalar type is zero, and the derivative of a vector or
matrix is done componentwise. One can use x0 as a shortcut if the derivative is with respect to the
main variable of x; and also use x00 , etc., for multiple derivatives altough derivn is often preferrable.
By definition, the main variable of a t_POLMOD is the main variable among the coefficients from
its two polynomial components (representative and modulus); in other words, assuming a polmod
represents an element of R[X]/(T (X)), the variable X is a mute variable and the derivative is taken
with respect to the main variable used in the base ring R.
? f = (x/y)^5;
? deriv(f)
%2 = 5/y^5*x^4
? f’
%3 = 5/y^5*x^4
? deriv(f, ’x) \\ same since ’x is the main variable
%4 = 5/y^5*x^4
? deriv(f, ’y)
%5 = -5/y^6*x^5
This function also operates on closures, in which case the variable must be omitted. It returns
a closure performing a numerical differentiation as per derivnum:
? f(x) = x^2;
? g = deriv(f)
? g(1)
%3 = 2.0000000000000000000000000000000000000
? f(x) = sin(exp(x));
? deriv(f)(0)
%5 = 0.54030230586813971740093660744297660373
? cos(1)
%6 = 0.54030230586813971740093660744297660373
The library syntax is GEN deriv(GEN x, long v = -1) where v is a variable number.

3.9.4 derivn(x, n, {v}). n-th derivative of x with respect to the main variable if v is omitted, and
with respect to v otherwise; the integer n must be nonnegative. The derivative of a scalar type is
zero, and the derivative of a vector or matrix is done componentwise. One can use x0 , x00 , etc., as
a shortcut if the derivative is with respect to the main variable of x.
By definition, the main variable of a t_POLMOD is the main variable among the coefficients from
its two polynomial components (representative and modulus); in other words, assuming a polmod
represents an element of R[X]/(T (X)), the variable X is a mute variable and the derivative is taken
with respect to the main variable used in the base ring R.
? f = (x/y)^5;
? derivn(f, 2)
%2 = 20/y^5*x^3
? f’’
%3 = 20/y^5*x^3
? derivn(f, 2, ’x) \\ same since ’x is the main variable
%4 = 20/y^5*x^3

232
 ? derivn(f, 2, ’y)
%5 = 30/y^7*x^5
This function also operates on closures, in which case the variable must be omitted. It returns
a closure performing a numerical differentiation as per derivnum:

? f(x) = x^10;
? g = derivn(f, 5)
? g(1)
%3 = 30240.000000000000000000000000000000000
? derivn(zeta, 2)(0)
%4 = -2.0063564559085848512101000267299604382
? zeta’’(0)
%5 = -2.0063564559085848512101000267299604382
The library syntax is GEN derivn(GEN x, long n, long v = -1) where v is a variable
number.

3.9.5 diffop(x, v, d, {n = 1}). Let v be a vector of variables, and d a vector of the same length,
return the image of x by the n-power (1 if n is not given) of the differential operator D that
assumes the value d[i] on the variable v[i]. The value of D on a scalar type is zero, and D
applies componentwise to a vector or matrix. When applied to a t_POLMOD, if no value is provided
for the variable of the modulus, such value is derived using the implicit function theorem.

Examples. This function can be used to differentiate formal expressions: if E = exp(X 2 ) then we
have E 0 = 2 ∗ X ∗ E. We derivate X ∗ exp(X 2 ) as follows:

? diffop(E*X,[X,E],[1,2*X*E])
%1 = (2*X^2 + 1)*E
Let Sin and Cos be two function such that Sin2 + Cos2 = 1 and Cos0 = −Sin. We can
differentiate Sin/Cos as follows, PARI inferring the value of Sin0 from the equation:

? diffop(Mod(’Sin/’Cos,’Sin^2+’Cos^2-1),[’Cos],[-’Sin])
%1 = Mod(1/Cos^2, Sin^2 + (Cos^2 - 1))
Compute the Bell polynomials (both complete and partial) via the Faa di Bruno formula:

Bell(k,n=-1)=
{ my(x, v, dv, var = i->eval(Str("X",i)));
v = vector(k, i, if (i==1, ’E, var(i-1)));
dv = vector(k, i, if (i==1, ’X*var(1)*’E, var(i)));
x = diffop(’E,v,dv,k) / ’E;
if (n < 0, subst(x,’X,1), polcoef(x,n,’X));
}
The library syntax is GEN diffop0(GEN x, GEN v, GEN d, long n).
For n = 1, the function GEN diffop(GEN x, GEN v, GEN d) is also available.

233
3.9.6 eval(x). Replaces in x the formal variables by the values that have been assigned to them
after the creation of x. This is mainly useful in GP, and not in library mode. Do not confuse this
with substitution (see subst).

If x is a character string, eval(x) executes x as a GP command, as if directly input from the

keyboard, and returns its output.

? x1 = "one"; x2 = "two";
? n = 1; eval(Str("x", n))
%2 = "one"
? f = "exp"; v = 1;
? eval(Str(f, "(", v, ")"))
%4 = 2.7182818284590452353602874713526624978

Note that the first construct could be implemented in a simpler way by using a vector x =
["one","two"]; x[n], and the second by using a closure f = exp; f(v). The final example
is more interesting:

? genmat(u,v) = matrix(u,v,i,j, eval( Str("x",i,j) ));

? genmat(2,3) \\ generic 2 x 3 matrix
%2 =
[x11 x12 x13]
[x21 x22 x23]

A syntax error in the evaluation expression raises an e SYNTAX exception, which can be trapped
as usual:

? 1a
*** syntax error, unexpected variable name, expecting $end or ’;’: 1a
*** ^-
? E(expr) =
{
iferr(eval(expr),
e, print("syntax error"),
errname(e) == "e_SYNTAX");
}
? E("1+1")
%1 = 2
? E("1a")
syntax error

The library syntax is geval(GEN x).

234
3.9.7 factorpadic(pol , p, r). p-adic factorization of the polynomial pol to precision r, the re-
sult being a two-column matrix as in factor. Note that this is not the same as a factorization
over Z/pr Z (polynomials over that ring do not form a unique factorization domain, anyway), but
approximations in Q/pr Z of the true factorization in Qp [X].
? factorpadic(x^2 + 9, 3,5)
%1 =
[(1 + O(3^5))*x^2 + O(3^5)*x + (3^2 + O(3^5)) 1]
? factorpadic(x^2 + 1, 5,3)
%2 =
[ (1 + O(5^3))*x + (2 + 5 + 2*5^2 + O(5^3)) 1]
[(1 + O(5^3))*x + (3 + 3*5 + 2*5^2 + O(5^3)) 1]
The factors are normalized so that their leading coefficient is a power of p. The method used is a
modified version of the round 4 algorithm of Zassenhaus.
If pol has inexact t_PADIC coefficients, this is not always well-defined; in this case, the poly-
nomial is first made integral by dividing out the p-adic content, then lifted to Z using truncate
coefficientwise. Hence we actually factor exactly a polynomial which is only p-adically close to the
input. To avoid pitfalls, we advise to only factor polynomials with exact rational coefficients.
The library syntax is factorpadic(GEN f,GEN p, long r) . The function factorpadic0 is
deprecated, provided for backward compatibility.

3.9.8 fft(w, P ). Let w = [1, z, . . . , z N −1 ] from some primitive N -roots of unity z where N is a
power of 2, and P be a polynomial < N , return the unnormalized discrete Fourier transform P of Pi ,
{P (w[i]), 1 ≤ i ≤ N }. Also allow P to be a vector [p0 , . . . , pn ] representing the polynomial pi X .
Composing fft and fftinv returns N times the original input coefficients.
? w = rootsof1(4); fft(w, x^3+x+1)
%1 = [3, 1, -1, 1]
? fftinv(w, %)
%2 = [4, 4, 0, 4]
? Polrev(%) / 4
%3 = x^3 + x + 1
? w = powers(znprimroot(5),3); fft(w, x^3+x+1)
%4 = [Mod(3,5),Mod(1,5),Mod(4,5),Mod(1,5)]
? fftinv(w, %)
%5 = [Mod(4,5),Mod(4,5),Mod(0,5),Mod(4,5)]
The library syntax is GEN FFT(GEN w, GEN P).

3.9.9 fftinv(w, P ). Let w = [1, z, . . . , z N −1 ] from some primitive N -roots of unity z where N is
a power of 2, and P be a polynomial < N , return the unnormalized discrete Fourier transform of
P, {P (1/w[i]),
P
i
1 ≤ i ≤ N }. Also allow P to be a vector [p0 , . . . , pn ] representing the polynomial
pi X . Composing fft and fftinv returns N times the original input coefficients.
? w = rootsof1(4); fft(w, x^3+x+1)
%1 = [3, 1, -1, 1]
? fftinv(w, %)
%2 = [4, 4, 0, 4]
? Polrev(%) / 4

235
 %3 = x^3 + x + 1
? N = 512; w = rootsof1(N); T = random(1000 * x^(N-1));
? U = fft(w, T);
time = 3 ms.
? V = vector(N, i, subst(T, ’x, w[i]));
time = 65 ms.
? exponent(V - U)
%7 = -97
? round(Polrev(fftinv(w,U) / N)) == T
%8 = 1
The library syntax is GEN FFTinv(GEN w, GEN P).

3.9.10 intformal(x, {v}). formal integration of x with respect to the variable v (wrt. the main
variable if v is omitted). Since PARI cannot represent logarithmic or arctangent terms, any such
term in the result will yield an error:
? intformal(x^2)
%1 = 1/3*x^3
? intformal(x^2, y)
%2 = y*x^2
? intformal(1/x)
*** at top-level: intformal(1/x)
*** ^--------------
*** intformal: domain error in intformal: residue(series, pole) != 0
The argument x can be of any type. When x is a rational function, we assume that the base
ring is an integral domain of characteristic zero.
By definition, the main variable of a t_POLMOD is the main variable among the coefficients
from its two polynomial components (representative and modulus); in other words, assuming a
polmod represents an element of R[X]/(T (X)), the variable X is a mute variable and the integral
is taken with respect to the main variable used in the base ring R. In particular, it is meaningless
to integrate with respect to the main variable of x.mod:
? intformal(Mod(1,x^2+1), ’x)
*** intformal: incorrect priority in intformal: variable x = x
The library syntax is GEN integ(GEN x, long v = -1) where v is a variable number.

3.9.11 padicappr(pol , a). Vector of p-adic roots of the polynomial pol congruent to the p-adic
number a modulo p, and with the same p-adic precision as a. The number a can be an ordinary p-
adic number (type t_PADIC, i.e. an element of Zp ) or can be an integral element of a finite unramified
extension Qp [X]/(T ) of Qp , given as a t_POLMOD Mod(A, T ) at least one of whose coefficients is a
t_PADIC and T irreducible modulo p. In this case, the result is the vector of roots belonging to the
same extension of Qp as a. The polynomial pol should have exact coefficients; if not, its coefficients
are first rounded to Q or Q[X]/(T ) and this is the polynomial whose roots we consider.
The library syntax is GEN padicappr(GEN pol, GEN a). Also available is GEN Zp_appr(GEN
f, GEN a) when a is a t_PADIC.

236
3.9.12 padicfields(p, N, {flag = 0}). Returns a vector of polynomials generating all the extensions
of degree N of the field Qp of p-adic rational numbers; N is allowed to be a 2-component vector
[n, d], in which case we return the extensions of degree n and discriminant pd .
The list is minimal in the sense that two different polynomials generate nonisomorphic exten-
sions; in particular, the number of polynomials is the number of classes of nonisomorphic extensions.
If P is a polynomial in this list, α is any root of P and K = Qp (α), then α is the sum of a uni-
formizer and a (lift of a) generator of the residue field of K; in particular, the powers of α generate
the ring of p-adic integers of K.
If flag = 1, replace each polynomial P by a vector [P, e, f, d, c] where e is the ramification
index, f the residual degree, d the valuation of the discriminant, and c the number of conjugate
fields. If flag = 2, only return the number of extensions in a fixed algebraic closure (Krasner’s
formula), which is much faster.
The library syntax is GEN padicfields0(GEN p, GEN N, long flag). Also available is GEN
padicfields(GEN p, long n, long d, long flag), which computes extensions of Qp of degree
n and discriminant pd .

3.9.13 polchebyshev(n, {flag = 1}, {a =0 x}). Returns the nth Chebyshev polynomial of the first
kind Tn (flag = 1) or the second kind Un (flag = 2), evaluated at a (’x by default). Both series of
polynomials satisfy the 3-term relation

Pn+1 = 2xPn − Pn−1 ,

and are determined by the initial conditions U0 = T0 = 1, T1 = x, U1 = 2x. In fact Tn0 = nUn−1
and, for all complex numbers z, we have Tn (cos z) = cos(nz) and Un−1 (cos z) = sin(nz)/ sin z. If
n ≥ 0, then these polynomials have degree n. For n < 0, Tn is equal to T−n and Un is equal to
−U−2−n . In particular, U−1 = 0.
The library syntax is GEN polchebyshev_eval(long n, long flag, GEN a = NULL). Also
available are GEN polchebyshev(long n, long flag, long v), GEN polchebyshev1(long n,
long v) and GEN polchebyshev2(long n, long v) for Tn and Un respectively.

3.9.14 polclass(D, {inv = 0}, {x =0 x}). Return a polynomial in Z[x] generating the Hilbert class
field for the imaginary quadratic discriminant D. If inv is 0 (the default), use the modular j-
function and return the classical Hilbert polynomial, otherwise use a class invariant. The following
invariants correspond to the different values of inv, where f denotes Weber’s function weber, and
wp,q the double eta quotient given by wp,q = η(x/p)η(x/q)
η(x)η(x/pq)

The invariants wp,q are not allowed unless they satisfy the following technical conditions en-
suring they do generate the Hilbert class field and not a strict subfield:
√
• if p 6= q, we need them both noninert, prime to the conductor of Z[ D]. Let P, Q be prime
±1 ±1
ideals above p and √ q; if both are unramified, we further require that P Q be all distinct in the
class group of Z[ D]; if both are ramified, we require that P Q 6= 1 in the class group.
• if p = q, we want it split and prime to the conductor and the prime ideal above it must have
order 6= 1, 2, 4 in the class group.
Invariants are allowed under the additional conditions on D listed below.
•0: j

237
 • 1 : f , D = 1 mod 8 and D = 1, 2 mod 3;
• 2 : f 2 , D = 1 mod 8 and D = 1, 2 mod 3;
• 3 : f 3 , D = 1 mod 8;
• 4 : f 4 , D = 1 mod 8 and D = 1, 2 mod 3;
• 5 : γ2 = j 1/3 , D = 1, 2 mod 3;
• 6 : w2,3 , D = 1 mod 8 and D = 1, 2 mod 3;
• 8 : f 8 , D = 1 mod 8 and D = 1, 2 mod 3;
• 9 : w3,3 , D = 1 mod 2 and D = 1, 2 mod 3;
• 10: w2,5 , D 6= 60 mod 80 and D = 1, 2 mod 3;
• 14: w2,7 , D = 1 mod 8;
• 15: w3,5 , D = 1, 2 mod 3;
• 21: w3,7 , D = 1 mod 2 and 21 does not divide D
2
• 23: w2,3 , D = 1, 2 mod 3;
2
• 24: w2,5 , D = 1, 2 mod 3;
• 26: w2,13 , D 6= 156 mod 208;
2
• 27: w2,7 , D 6= 28 mod 112;
2
• 28: w3,3 , D = 1, 2 mod 3;
• 35: w5,7 , D = 1, 2 mod 3;
• 39: w3,13 , D = 1 mod 2 and D = 1, 2 mod 3;
The algorithm for computing the polynomial does not use the floating point approach, which
would evaluate a precise modular function in a precise complex argument. Instead, it relies on a
faster Chinese remainder based approach modulo small primes, in which the class invariant is only
defined algebraically by the modular polynomial relating the modular function to j. So in fact,
any of the several roots of the modular polynomial may actually be the class invariant, and more
precise assertions cannot be made.
For instance, while polclass(D) returns the minimal polynomial of j(τ ) with τ (any) quadratic
integer for the discriminant D, the polynomial returned by polclass(D, 5) can be the minimal
polynomial of any of γ2 (τ ), ζ3 γ2 (τ ) or ζ32 γ2 (τ ), the three roots of the modular polynomial j = γ23 ,
in which j has been specialised to j(τ ).
(f 24 −16)3
The modular polynomial is given by j = f 24 for Weber’s function f .
For the double eta quotients of level N = pq, all functions are covered such that the modular
curve X0+ (N ), the function field of which is generated by the functions invariant under Γ0 (N ) and
the Fricke–Atkin–Lehner involution, is of genus 0 with function field generated by (a power of) the
double eta quotient w. This ensures that the full Hilbert class field (and not a proper subfield) is
generated by class invariants from these double eta quotients. Then the modular polynomial is of
degree 2 in j, and of degree ψ(N ) = (p + 1)(q + 1) in w.
? polclass(-163)

238
 %1 = x + 262537412640768000
? polclass(-51, , ’z)
%2 = z^2 + 5541101568*z + 6262062317568
? polclass(-151,1)
x^7 - x^6 + x^5 + 3*x^3 - x^2 + 3*x + 1
The library syntax is GEN polclass(GEN D, long inv, long x = -1) where x is a variable
number.

3.9.15 polcoef(x, n, {v}). Coefficient of degree n of the polynomial x, with respect to the main
variable if v is omitted, with respect to v otherwise. If n is greater than the degree, the result is
zero.
Naturally applies to scalars (polynomial of degree 0), as well as to rational functions whose
denominator is a monomial. It also applies to power series: if n is less than the valuation, the result
is zero. If it is greater than the largest significant degree, then an error message is issued.
The library syntax is GEN polcoef(GEN x, long n, long v = -1) where v is a variable
number.

3.9.16 polcoeff(x, n, {v}). Deprecated alias for polcoef.

The library syntax is GEN polcoef(GEN x, long n, long v = -1) where v is a variable
number.

3.9.17 polcyclo(n, {a =0 x}). n-th cyclotomic polynomial, evaluated at a (’x by default). The
integer n must be positive.
Algorithm used: reduce to the case where n is squarefree; to compute
Q the dcyclotomic polyno-
p µ(n/d)
mial, use Φnp (x) = Φn (x )/Φ(x); to compute it evaluated, use Φn (x) = d|n (x − 1) . In the
d
evaluated case, the algorithm assumes that a − 1 is either 0 or invertible, for all d | n. If this is
not the case (the base ring has zero divisors), use subst(polcyclo(n),x,a).
The library syntax is GEN polcyclo_eval(long n, GEN a = NULL). The variant GEN polcy-
clo(long n, long v) returns the n-th cyclotomic polynomial in variable v.

3.9.18 polcyclofactors(f ). Returns a vector of polynomials, whose product is the product of

distinct cyclotomic polynomials dividing f .
? f = x^10+5*x^8-x^7+8*x^6-4*x^5+8*x^4-3*x^3+7*x^2+3;
? v = polcyclofactors(f)
%2 = [x^2 + 1, x^2 + x + 1, x^4 - x^3 + x^2 - x + 1]
? apply(poliscycloprod, v)
%3 = [1, 1, 1]
? apply(poliscyclo, v)
%4 = [4, 3, 10]
In general, the polynomials are products of cyclotomic polynomials and not themselves irreducible:
? g = x^8+2*x^7+6*x^6+9*x^5+12*x^4+11*x^3+10*x^2+6*x+3;
? polcyclofactors(g)
%2 = [x^6 + 2*x^5 + 3*x^4 + 3*x^3 + 3*x^2 + 2*x + 1]
? factor(%[1])

239
 %3 =
[ x^2 + x + 1 1]
[x^4 + x^3 + x^2 + x + 1 1]
The library syntax is GEN polcyclofactors(GEN f).

3.9.19 poldegree(x, {v}). Degree of the polynomial x in the main variable if v is omitted, in the
variable v otherwise.
The degree of 0 is -oo. The degree of a nonzero scalar is 0. Finally, when x is a nonzero
polynomial or rational function, returns the ordinary degree of x. Raise an error otherwise.
The library syntax is GEN gppoldegree(GEN x, long v = -1) where v is a variable number.
Also available is long poldegree(GEN x, long v), which returns -LONG_MAX if x = 0 and the
degree as a long integer.

3.9.20 poldisc(pol , {v}). Discriminant of the polynomial pol in the main variable if v is omitted,
in v otherwise. Uses a modular algorithm over Z or Q, and the subresultant algorithm otherwise.
? T = x^4 + 2*x+1;
? poldisc(T)
%2 = -176
? poldisc(T^2)
%3 = 0
For convenience, the function also applies to types t_QUAD and t_QFI/t_QFR:
? z = 3*quadgen(8) + 4;
? poldisc(z)
%2 = 8
? q = Qfb(1,2,3);
? poldisc(q)
%4 = -8
The library syntax is GEN poldisc0(GEN pol, long v = -1) where v is a variable number.

3.9.21 poldiscfactors(T, {flag = 0}). Given a polynomial T with integer coefficients, return
[D, faD] where D is the discriminant of T and faD is a cheap partial factorization of |D|: entries
in its first column are coprime and not perfect powers but need not be primes. The factors are
obtained by a combination of trial division, testing for perfect powers, factorizations in coprimes,
and computing Euclidean remainder sequences for (T, T 0 ) modulo composite factors d of D (which
is likely to produce 0-divisors in Z/dZ). If flag is 1, finish the factorization using factorint.
? T = x^3 - 6021021*x^2 + 12072210077769*x - 8092423140177664432;
? [D,faD] = poldiscfactors(T); print(faD); D
[3, 3; 7, 2; 373, 2; 500009, 2; 24639061, 2]
%2 = -27937108625866859018515540967767467
? T = x^3 + 9*x^2 + 27*x - 125014250689643346789780229390526092263790263725;
? [D,faD] = poldiscfactors(T); print(faD)
[2, 6; 3, 3; 125007125141751093502187, 4]
? [D,faD] = poldiscfactors(T, 1); print(faD)
[2, 6; 3, 3; 500009, 12; 1000003, 4]
The library syntax is GEN poldiscfactors(GEN T, long flag).

240
3.9.22 poldiscreduced(f ). Reduced discriminant vector of the (integral, monic) polynomial f .
This is the vector of elementary divisors of Z[α]/f 0 (α)Z[α], where α is a root of the polynomial f .
The components of the result are all positive, and their product is equal to the absolute value of
the discriminant of f .
The library syntax is GEN reduceddiscsmith(GEN f).

3.9.23 polgraeffe(f ). Returns the Graeffe transform g of f , such that g(x2 ) = f (x)f (−x).
The library syntax is GEN polgraeffe(GEN f).

3.9.24 polhensellift(A, B, p, e). Given a prime p, an integral polynomial A whose leading co-
efficient is a p-unit, a vector B of integral polynomials that are monic and pairwise relatively
prime modulo p, and whose product is congruent to A/lc(A) modulo p, lift the elements of B to
polynomials whose product is congruent to A modulo pe .
More generally, if T is an integral polynomial irreducible mod p, and B is a factorization of
A over the finite field Fp [t]/(T ), you can lift it to Zp [t]/(T, pe ) by replacing the p argument with
[p, T ]:

? { T = t^3 - 2; p = 7; A = x^2 + t + 1;
B = [x + (3*t^2 + t + 1), x + (4*t^2 + 6*t + 6)];
r = polhensellift(A, B, [p, T], 6) }
%1 = [x + (20191*t^2 + 50604*t + 75783), x + (97458*t^2 + 67045*t + 41866)]
? liftall( r[1] * r[2] * Mod(Mod(1,p^6),T) )
%2 = x^2 + (t + 1)
The library syntax is GEN polhensellift(GEN A, GEN B, GEN p, long e).

3.9.25 polhermite(n, {a =0 x}, {flag = 0}). nth Hermite polynomial Hn evaluated at a (’x by
default), i.e.
2 d
n 2
Hn (x) = (−1)n ex n
e−x .
dx
If flag is nonzero and n > 0, return [Hn−1 (a), Hn (a)].

? polhermite(5)
%1 = 32*x^5 - 160*x^3 + 120*x
? polhermite(5, -2) \\ H_5(-2)
%2 = 16
? polhermite(5,,1)
%3 = [16*x^4 - 48*x^2 + 12, 32*x^5 - 160*x^3 + 120*x]
? polhermite(5,-2,1)
%4 = [76, 16]
The library syntax is GEN polhermite_eval0(long n, GEN a = NULL, long flag). The
variant GEN polhermite(long n, long v) returns the n-th Hermite polynomial in variable v. To
obtain Hn (a), use GEN polhermite_eval(long n, GEN a).

241
3.9.26 polinterpolate(X, {Y }, {t =0 x}, {&e}). Given the data vectors X and Y of the same
length n (X containing the x-coordinates, and Y the corresponding y-coordinates), this function
finds the interpolating polynomial P of minimal degree passing through these points and evaluates
it at t. If Y is omitted, the polynomial P interpolates the (i, X[i]).
? v = [1, 2, 4, 8, 11, 13];
? P = polinterpolate(v) \\ formal interpolation
%1 = 7/120*x^5 - 25/24*x^4 + 163/24*x^3 - 467/24*x^2 + 513/20*x - 11
? [ subst(P,’x,a) | a <- [1..6] ]
%2 = [1, 2, 4, 8, 11, 13]
? polinterpolate(v,, 10) \\ evaluate at 10
%3 = 508
? subst(P, x, 10)
%4 = 508
? P = polinterpolate([1,2,4], [9,8,7])
%5 = 1/6*x^2 - 3/2*x + 31/3
? [subst(P, ’x, a) | a <- [1,2,4]]
%6 = [9, 8, 7]
? P = polinterpolate([1,2,4], [9,8,7], 0)
%7 = 31/3
If the goal is to extrapolate a function at a unique point, it is more efficient to use the t argument
rather than interpolate formally then evaluate:
? x0 = 1.5;
? v = vector(20, i,random([-10,10]));
? for(i=1,10^3, subst(polinterpolate(v),’x, x0))
time = 352 ms.
? for(i=1,10^3, polinterpolate(v,,x0))
time = 111 ms.
? v = vector(40, i,random([-10,10]));
? for(i=1,10^3, subst(polinterpolate(v), ’x, x0))
time = 3,035 ms.
? for(i=1,10^3, polinterpolate(v,, x0))
time = 436 ms.
The threshold depends on the base field. Over small prime finite fields, interpolating formally first
is more efficient
? bench(p, N, T = 10^3) =
{ my (v = vector(N, i, random(Mod(0,p))));
my (x0 = Mod(3, p), t1, t2);
gettime();
for(i=1, T, subst(polinterpolate(v), ’x, x0));
t1 = gettime();
for(i=1, T, polinterpolate(v,, x0));
t2 = gettime(); [t1, t2];
}
? p = 101;
? bench(p, 4, 10^4) \\ both methods are equivalent
%3 = [39, 40]

242
 ? bench(p, 40) \\ with 40 points formal is much faster
%4 = [45, 355]
As the cardinality increases, formal interpolation requires more points to become interesting:
? p = nextprime(2^128);
? bench(p, 4) \\ formal is slower
%3 = [16, 9]
? bench(p, 10) \\ formal has become faster
%4 = [61, 70]
? bench(p, 100) \\ formal is much faster
%5 = [1682, 9081]
? p = nextprime(10^500);
? bench(p, 4) \\ formal is slower
%7 = [72, 354]
? bench(p, 20) \\ formal is still slower
%8 = [1287, 962]
? bench(p, 40) \\ formal has become faster
%9 = [3717, 4227]
? bench(p, 100) \\ faster but relatively less impressive
%10 = [16237, 32335]
If t is a complex numeric value and e is present, e will contain an error estimate on the
returned value. More precisely, let P be the interpolation polynomial on the given n points; there
exist a subset of n − 1 points and Q the attached interpolation interpolation polynomial such that
e = exponent(P (t) − Q(t)) (Neville’s algorithm).
? f(x) = 1 / (1 + 25*x^2);
? x0 = 975/1000;
? test(X) =
{ my (P, e);
P = polinterpolate(X, [f(x) | x <- X], x0, &e);
[ exponent(P - f(x0)), e ];
}
\\ equidistant nodes vs. Chebyshev nodes
? test( [-10..10] / 10 )
%4 = [6, 5]
? test( polrootsreal(polchebyshev(21)) )
%5 = [-15, -10]
? test( [-100..100] / 100 )
%7 = [93, 97] \\ P(x0) is way different from f(x0)
? test( polrootsreal(polchebyshev(201)) )
%8 = [-60, -55]
This is an example of Runge’s phenomenon: increasing the number of equidistant nodes makes
extrapolation much worse. Note that the error estimate is not a guaranteed upper bound (cf %4),
but is reasonably tight in practice.
The library syntax is GEN polint(GEN X, GEN Y = NULL, GEN t = NULL, GEN *e = NULL)
.

243
3.9.27 poliscyclo(f ). Returns 0 if f is not a cyclotomic polynomial, and n > 0 if f = Φn , the
n-th cyclotomic polynomial.
? poliscyclo(x^4-x^2+1)
%1 = 12
? polcyclo(12)
%2 = x^4 - x^2 + 1
? poliscyclo(x^4-x^2-1)
%3 = 0
The library syntax is long poliscyclo(GEN f).

3.9.28 poliscycloprod(f ). Returns 1 if f is a product of cyclotomic polynomial, and 0 otherwise.

? f = x^6+x^5-x^3+x+1;
? poliscycloprod(f)
%2 = 1
? factor(f)
%3 =
[ x^2 + x + 1 1]
[x^4 - x^2 + 1 1]
? [ poliscyclo(T) | T <- %[,1] ]
%4 = [3, 12]
? polcyclo(3) * polcyclo(12)
%5 = x^6 + x^5 - x^3 + x + 1
The library syntax is long poliscycloprod(GEN f).

3.9.29 polisirreducible(pol ). pol being a polynomial (univariate in the present version 2.13.3),
returns 1 if pol is nonconstant and irreducible, 0 otherwise. Irreducibility is checked over the
smallest base field over which pol seems to be defined.
The library syntax is long polisirreducible(GEN pol).

(a)
3.9.30 pollaguerre(n, {a = 0}, {b =0 x}, {flag = 0}). nth Laguerre polynomial Ln of degree n
and parameter a evaluated at b (’x by default), i.e.

x−a ex dn −x n+a

L(a)
n (x) = e x .
n! dxn
(a) (a)
If flag is 1, return [Ln−1 (b), Ln (b)].
The library syntax is GEN pollaguerre_eval0(long n, GEN a = NULL, GEN b = NULL, long
flag). To obtain the n-th Laguerre polynomial in variable v, use GEN pollaguerre(long n, GEN
(a)
a, GEN b, long v). To obtain Ln (b), use GEN pollaguerre_eval(long n, GEN a, GEN b)
.

3.9.31 pollead(x, {v}). Leading coefficient of the polynomial or power series x. This is computed
with respect to the main variable of x if v is omitted, with respect to the variable v otherwise.
The library syntax is GEN pollead(GEN x, long v = -1) where v is a variable number.

244
3.9.32 pollegendre(n, {a =0 x}, {flag = 0}). nth Legendre polynomial Pn evaluated at a (’x by
default), where
1 dn 2
Pn (x) = n (x − 1)n .
2 n! dxn
If flag is 1, return [Pn−1 (a), Pn (a)].
The library syntax is GEN pollegendre_eval0(long n, GEN a = NULL, long flag). To
obtain the n-th Legendre polynomial Pn in variable v, use GEN pollegendre(long n, long v)
. To obtain Pn (a), use GEN pollegendre_eval(long n, GEN a).

3.9.33 polmodular(L, {inv = 0}, {x =0 x}, {y =0 y}, {derivs = 0}). Return the modular polyno-
mial of prime level L in variables x and y for the modular function specified by inv. If inv is 0
(the default),
p use the modular j function, if inv is 1 use the Weber-f function, and if inv is 5 use
γ2 = [3]j. See polclass for the full list of invariants. If x is given as Mod(j, p) or an element
j of a finite field (as a t_FFELT), then return the modular polynomial of level L evaluated at j. If
j is from a finite field and derivs is nonzero, then return a triple where the last two elements are
the first and second derivatives of the modular polynomial evaluated at j.
? polmodular(3)
%1 = x^4 + (-y^3 + 2232*y^2 - 1069956*y + 36864000)*x^3 + ...
? polmodular(7, 1, , ’J)
%2 = x^8 - J^7*x^7 + 7*J^4*x^4 - 8*J*x + J^8
? polmodular(7, 5, 7*ffgen(19)^0, ’j)
%3 = j^8 + 4*j^7 + 4*j^6 + 8*j^5 + j^4 + 12*j^2 + 18*j + 18
? polmodular(7, 5, Mod(7,19), ’j)
%4 = Mod(1, 19)*j^8 + Mod(4, 19)*j^7 + Mod(4, 19)*j^6 + ...
? u = ffgen(5)^0; T = polmodular(3,0,,’j)*u;
? polmodular(3, 0, u,’j,1)
%6 = [j^4 + 3*j^2 + 4*j + 1, 3*j^2 + 2*j + 4, 3*j^3 + 4*j^2 + 4*j + 2]
? subst(T,x,u)
%7 = j^4 + 3*j^2 + 4*j + 1
? subst(T’,x,u)
%8 = 3*j^2 + 2*j + 4
? subst(T’’,x,u)
%9 = 3*j^3 + 4*j^2 + 4*j + 2
The library syntax is GEN polmodular(long L, long inv, GEN x = NULL, long y = -1,
long derivs) where y is a variable number.

3.9.34 polrecip(pol ). Reciprocal polynomial of pol with respect to its main variable, i.e. the
coefficients of the result are in reverse order; pol must be a polynomial.
? polrecip(x^2 + 2*x + 3)
%1 = 3*x^2 + 2*x + 1
? polrecip(2*x + y)
%2 = y*x + 2
The library syntax is GEN polrecip(GEN pol).

245
3.9.35 polresultant(x, y, {v}, {flag = 0}). Resultant of the two polynomials x and y with exact
entries, with respect to the main variables of x and y if v is omitted, with respect to the variable
v otherwise. The algorithm assumes the base ring is a domain. If you also need the u and v such
that x ∗ u + y ∗ v = Res(x, y), use the polresultantext function.

If flag = 0 (default), uses the algorithm best suited to the inputs, either the subresultant
algorithm (Lazard/Ducos variant, generic case), a modular algorithm (inputs in Q[X]) or Sylvester’s
matrix (inexact inputs).

If flag = 1, uses the determinant of Sylvester’s matrix instead; this should always be slower
than the default.

If x or y are multivariate with a huge polynomial content, it is advisable to remove it before

calling this function. Compare:

? a = polcyclo(7) * ((t+1)/(t+2))^100;
? b = polcyclo(11)* ((t+2)/(t+3))^100);
? polresultant(a,b);
time = 3,833 ms.
? ca = content(a); cb = content(b); \
polresultant(a/ca,b/cb)*ca^poldegree(b)*cb*poldegree(a); \\ instantaneous

The function only removes rational denominators and does not compute automatically the content
because it is generically small and potentially very expensive (e.g. in multivariate contexts). The
choice is yours, depending on your application.

The library syntax is GEN polresultant0(GEN x, GEN y, long v = -1, long flag) where
v is a variable number.

3.9.36 polresultantext(A, B, {v}). Finds polynomials U and V such that A∗U +B∗V = R, where
R is the resultant of U and V with respect to the main variables of A and B if v is omitted, and
with respect to v otherwise. Returns the row vector [U, V, R]. The algorithm used (subresultant)
assumes that the base ring is a domain.

? A = x*y; B = (x+y)^2;
? [U,V,R] = polresultantext(A, B)
%2 = [-y*x - 2*y^2, y^2, y^4]
? A*U + B*V
%3 = y^4
? [U,V,R] = polresultantext(A, B, y)
%4 = [-2*x^2 - y*x, x^2, x^4]
? A*U+B*V
%5 = x^4

The library syntax is GEN polresultantext0(GEN A, GEN B, long v = -1) where v is a

variable number. Also available is GEN polresultantext(GEN x, GEN y).

246
3.9.37 polroots(T ). Complex roots of the polynomial T , given as a column vector where each
root is repeated according to its multiplicity and given as floating point complex numbers at the
current realprecision:
? polroots(x^2)
%1 = [0.E-38 + 0.E-38*I, 0.E-38 + 0.E-38*I]~
? polroots(x^3+1)
%2 = [-1.00... + 0.E-38*I, 0.50... - 0.866...*I, 0.50... + 0.866...*I]~
The algorithm used is a modification of Schönhage’s root-finding algorithm, due to and orig-
inally implemented by Gourdon. It runs in polynomial time in deg(T ) and the precision. If
furthermore T has rational coefficients, roots are guaranteed to the required relative accuracy. If
the input polynomial T is exact, then the ordering of the roots does not depend on the precision:
they are ordered by increasing |=z|, then by increasing <z; in case of tie (conjugates), the root
with negative imaginary part comes first.
The library syntax is GEN roots(GEN T, long prec).

3.9.38 polrootsbound(T, {tau = 0.01}). Return a sharp upper bound B for the modulus of the
largest complex root of the polynomial T with complex coefficients with relative error τ . More
precisely, we have |z| ≤ B for all roots and there exist one root such that |z0 | ≥ B exp(−2τ ). Much
faster than either polroots or polrootsreal.
? T=poltchebi(500);
? vecmax(abs(polroots(T)))
time = 5,706 ms.
%2 = 0.99999506520185816611184481744870013191
? vecmax(abs(polrootsreal(T)))
time = 1,972 ms.
%3 = 0.99999506520185816611184481744870013191
? polrootsbound(T)
time = 217 ms.
%4 = 1.0098792554165905155
? polrootsbound(T, log(2)/2) \\ allow a factor 2, much faster
time = 51 ms.
%5 = 1.4065759938190154354
? polrootsbound(T, 1e-4)
time = 504 ms.
%6 = 1.0000920717983847741
? polrootsbound(T, 1e-6)
time = 810 ms.
%7 = 0.9999960628901692905
? polrootsbound(T, 1e-10)
time = 1,351 ms.
%8 = 0.9999950652993869760
The library syntax is GEN polrootsbound(GEN T, GEN tau = NULL).

3.9.39 polrootsff(x, {p}, {a}). Obsolete, kept for backward compatibility: use factormod.
The library syntax is GEN polrootsff(GEN x, GEN p = NULL, GEN a = NULL).

247
3.9.40 polrootsmod(f, {D}). Vector of roots of the polynomial f over the finite field defined by
the domain D as follows:
• D = p a prime: factor over Fp ;
• D = [T, p] for a prime p and T (y) an irreducible polynomial over Fp : factor over Fp [y]/(T )
(as usual the main variable of T must have lower priority than the main variable of f );
• D a t_FFELT: factor over the attached field;
• D omitted: factor over the field of definition of f , which must be a finite field.
Multiple roots are not repeated.
? polrootsmod(x^2-1,2)
%1 = [Mod(1, 2)]~
? polrootsmod(x^2+1,3)
%2 = []~
? polrootsmod(x^2+1, [y^2+1,3])
%3 = [Mod(Mod(1, 3)*y, Mod(1, 3)*y^2 + Mod(1, 3)),
Mod(Mod(2, 3)*y, Mod(1, 3)*y^2 + Mod(1, 3))]~
? polrootsmod(x^2 + Mod(1,3))
%4 = []~
? liftall( polrootsmod(x^2 + Mod(Mod(1,3),y^2+1)) )
%5 = [y, 2*y]~
? t = ffgen(y^2+Mod(1,3)); polrootsmod(x^2 + t^0)
%6 = [y, 2*y]~
The library syntax is GEN polrootsmod(GEN f, GEN D = NULL).

3.9.41 polrootspadic(f, p, r). Vector of p-adic roots of the polynomial pol , given to p-adic pre-
cision r; the integer p is assumed to be a prime. Multiple roots are not repeated. Note that this
is not the same as the roots in Z/pr Z, rather it gives approximations in Z/pr Z of the true roots
living in Qp :
? polrootspadic(x^3 - x^2 + 64, 2, 4)
%1 = [2^3 + O(2^4), 2^3 + O(2^4), 1 + O(2^4)]~
? polrootspadic(x^3 - x^2 + 64, 2, 5)
%2 = [2^3 + O(2^5), 2^3 + 2^4 + O(2^5), 1 + O(2^5)]~
As the second commands show, the first two roots are distinct in Qp , even though they are equal
modulo 24 .
More generally, if T is an integral polynomial irreducible mod p and f has coefficients in
Q[t]/(T ), the argument p may be replaced by the vector [T, p]; we then return the roots of f in the
unramified extension Qp [t]/(T ).
? polrootspadic(x^3 - x^2 + 64*y, [y^2+y+1,2], 5)
%3 = [Mod((2^3 + O(2^5))*y + (2^3 + O(2^5)), y^2 + y + 1),
Mod((2^3 + 2^4 + O(2^5))*y + (2^3 + 2^4 + O(2^5)), y^2 + y + 1),
Mod(1 + O(2^5), y^2 + y + 1)]~
If pol has inexact t_PADIC coefficients, this need not well-defined; in this case, the polynomial
is first made integral by dividing out the p-adic content, then lifted to Z using truncate coeffi-
cientwise. Hence the roots given are approximations of the roots of an exact polynomial which is

248
p-adically close to the input. To avoid pitfalls, we advise to only factor polynomials with exact
rational coefficients.
The library syntax is GEN polrootspadic(GEN f, GEN p, long r).

3.9.42 polrootsreal(T, {ab}). Real roots of the polynomial T with real coefficients, multiple roots
being included according to their multiplicity. If the polynomial does not have rational coefficients,
it is first rescaled and rounded. The roots are given to a relative accuracy of realprecision. If
argument ab is present, it must be a vector [a, b] with two components (of type t_INT, t_FRAC or
t_INFINITY) and we restrict to roots belonging to that closed interval.
? \p9
? polrootsreal(x^2-2)
%1 = [-1.41421356, 1.41421356]~
? polrootsreal(x^2-2, [1,+oo])
%2 = [1.41421356]~
? polrootsreal(x^2-2, [2,3])
%3 = []~
? polrootsreal((x-1)*(x-2), [2,3])
%4 = [2.00000000]~
The algorithm used is a modification of Uspensky’s method (relying on Descartes’s rule of
sign), following Rouillier and Zimmerman’s article “Efficient isolation of a polynomial real roots”
(http://hal.inria.fr/inria-00072518/). Barring bugs, it is guaranteed to converge and to give
the roots to the required accuracy.

Remark. If the polynomial T is of the form Q(xh ) for some h ≥ 2 and ab is omitted, the routine
will apply the algorithm to Q (restricting to nonnegative roots when h is even), then take h-th
roots. On the other hand, if you want to specify ab, you should apply the routine to Q yourself
and a suitable interval [a0 , b0 ] using approximate h-th roots adapted to your problem: the function
will not perform this change of variables if ab is present.
The library syntax is GEN realroots(GEN T, GEN ab = NULL, long prec).

3.9.43 polsturm(T, {ab}). Number of distinct real roots of the real polynomial T . If the argument
ab is present, it must be a vector [a, b] with two real components (of type t_INT, t_REAL, t_FRAC
or t_INFINITY) and we count roots belonging to that closed interval.
If possible, you should stick to exact inputs, that is avoid t_REALs in T and the bounds a, b: the
result is then guaranteed and we use a fast algorithm (Uspensky’s method, relying on Descartes’s
rule of sign, see polrootsreal). Otherwise, the polynomial is rescaled and rounded first and the
result may be wrong due to that initial error. If only a or b is inexact, on the other hand, the
interval is first thickened using rational endpoints and the result remains guaranteed unless there
exist a root very close to a nonrational endpoint (which may be missed or unduly included).
? T = (x-1)*(x-2)*(x-3);
? polsturm(T)
%2 = 3
? polsturm(T, [-oo,2])
%3 = 2
? polsturm(T, [1/2,+oo])
%4 = 3

249
 ? polsturm(T, [1, Pi]) \\ Pi inexact: not recommended !
%5 = 3
? polsturm(T*1., [0, 4]) \\ T*1. inexact: not recommended !
%6 = 3
? polsturm(T^2, [0, 4]) \\ not squarefree: roots are not repeated!
%7 = 3
The library syntax is long RgX_sturmpart(GEN T, GEN ab) or long sturm(GEN T) (for the
case ab = NULL). The function long sturmpart(GEN T, GEN a, GEN b) is obsolete and deprecated.

3.9.44 polsubcyclo(n, d, {v =0 x}). Gives polynomials (in variable v) defining the sub-Abelian
extensions of degree d of the cyclotomic field Q(ζn ), where d | φ(n).
If there is exactly one such extension the output is a polynomial, else it is a vector of polyno-
mials, possibly empty. To get a vector in all cases, use concat([], polsubcyclo(n,d)).
The function galoissubcyclo allows to specify exactly which sub-Abelian extension should
be computed.
The library syntax is GEN polsubcyclo(long n, long d, long v = -1) where v is a variable
number.

3.9.45 polsylvestermatrix(x, y). Forms the Sylvester matrix corresponding to the two polyno-
mials x and y, where the coefficients of the polynomials are put in the columns of the matrix (which
is the natural direction for solving equations afterwards). The use of this matrix can be essential
when dealing with polynomials with inexact entries, since polynomial Euclidean division doesn’t
make much sense in this case.
The library syntax is GEN sylvestermatrix(GEN x, GEN y).

3.9.46 polsym(x, n). Creates the column vector of the symmetric powers of the roots of the
polynomial x up to power n, using Newton’s formula.
The library syntax is GEN polsym(GEN x, long n).

3.9.47 poltchebi(n, {v =0 x}). Deprecated alias for polchebyshev

The library syntax is GEN polchebyshev1(long n, long v = -1) where v is a variable
number.

3.9.48 polteichmuller(T, p, r). Given T ∈ Fp [X] return the polynomial P ∈ Zp [X] whose roots
(resp. leading coefficient) are the Teichmuller lifts of the roots (resp. leading coefficient) of T , to
p-adic precision r. If T is monic, P is the reduction modulo pr of the unique monic polynomial
congruent to T modulo p such that P (X p ) = 0 (mod P (X), pr ).
? T = ffinit(3, 3, ’t)
%1 = Mod(1,3)*t^3 + Mod(1,3)*t^2 + Mod(1,3)*t + Mod(2,3)
? P = polteichmuller(T,3,5)
%2 = t^3 + 166*t^2 + 52*t + 242
? subst(P, t, t^3) % (P*Mod(1,3^5))
%3 = Mod(0, 243)
? [algdep(a+O(3^5),2) | a <- Vec(P)]
%4 = [x - 1, 5*x^2 + 1, x^2 + 4*x + 4, x + 1]

250
When T is monic and irreducible mod p, this provides a model Qp [X]/(P ) of the unramified
extension Qp [X]/(T ) where the Frobenius has the simple form X mod P 7→ X p mod P .
The library syntax is GEN polteichmuller(GEN T, ulong p, long r).

(m)
3.9.49 polzagier(n, m). Creates Zagier’s polynomial Pn used in the functions sumalt and
sumpos (with flag = 1), see “Convergence acceleration of alternating series”, Cohen et al., Experi-
ment. Math., vol. 9, 2000, pp. 3–12.
(m) (0)
If m < 0 or m ≥ n, Pn = 0. We have Pn := Pn is Tn (2x − 1), where Tn is the Legendre
(m)
polynomial of the second kind. For n > m > 0, Pn is the m-th difference with step 2 of the
m+1
sequence n Pn ; in this case, it satisfies

dm+1
2Pn(m) (sin2 t) = (sin(2t)m sin(2(n − m)t)).
dtm+1

The library syntax is GEN polzag(long n, long m).

3.9.50 serconvol(x,
Py). Convolution (or
P Hadamard product) of the two power series x and y; in
other words if x = ak ∗ X k and y = bk ∗ X k then serconvol(x, y) = ak ∗ bk ∗ X k .
P

The library syntax is GEN convol(GEN x, GEN y).

3.9.51
P serlaplace(x). x must be a power
P series with nonnegative exponents or a polynomial. If
k k
x = (ak /k!) ∗ X then the result is ak ∗ X .
The library syntax is GEN laplace(GEN x).

3.9.52 serreverse(s). Reverse power series of s, i.e. the series t such that t(s) = x; s must be a
power series whose valuation is exactly equal to one.
? \ps 8
? t = serreverse(tan(x))
%2 = x - 1/3*x^3 + 1/5*x^5 - 1/7*x^7 + O(x^8)
? tan(t)
%3 = x + O(x^8)
The library syntax is GEN serreverse(GEN s).

3.9.53 subst(x, y, z). Replace the simple variable y by the argument z in the “polynomial” ex-
pression x. If z is a vector, return the vector of the evaluated expressions subst(x, y, z[i]).
Every type is allowed for x, but if it is not a genuine polynomial (or power series, or rational
function), the substitution will be done as if the scalar components were polynomials of degree
zero. In particular, beware that:
? subst(1, x, [1,2; 3,4])
%1 =
[1 0]
[0 1]
? subst(1, x, Mat([0,1]))
*** at top-level: subst(1,x,Mat([0,1])

251
 *** ^--------------------
*** subst: forbidden substitution by a non square matrix.
If x is a power series, z must be either a polynomial, a power series, or a rational function. If x is
a vector, matrix or list, the substitution is applied to each individual entry.
Use the function substvec to replace several variables at once, or the function substpol to
replace a polynomial expression.
The library syntax is GEN gsubst(GEN x, long y, GEN z) where y is a variable number.

3.9.54 substpol(x, y, z). Replace the “variable” y by the argument z in the “polynomial” expres-
sion x. Every type is allowed for x, but the same behavior as subst above apply.
The difference with subst is that y is allowed to be any polynomial here. The substitution is
done moding out all components of x (recursively) by y − t, where t is a new free variable of lowest
priority. Then substituting t by z in the resulting expression. For instance
? substpol(x^4 + x^2 + 1, x^2, y)
%1 = y^2 + y + 1
? substpol(x^4 + x^2 + 1, x^3, y)
%2 = x^2 + y*x + 1
? substpol(x^4 + x^2 + 1, (x+1)^2, y)
%3 = (-4*y - 6)*x + (y^2 + 3*y - 3)
The library syntax is GEN gsubstpol(GEN x, GEN y, GEN z). Further, GEN gdeflate(GEN
T, long v, long d) attempts to write T (x) in the form t(xd ), where x =pol x(v), and returns
NULL if the substitution fails (for instance in the example %2 above).

3.9.55 substvec(x, v, w). v being a vector of monomials of degree 1 (variables), w a vector

of expressions of the same length, replace in the expression x all occurrences of vi by wi . The
substitutions are done simultaneously; more precisely, the vi are first replaced by new variables in
x, then these are replaced by the wi :
? substvec([x,y], [x,y], [y,x])
%1 = [y, x]
? substvec([x,y], [x,y], [y,x+y])
%2 = [y, x + y] \\ not [y, 2*y]
The library syntax is GEN gsubstvec(GEN x, GEN v, GEN w).

3.9.56 sumformal(f, {v}). formal sum of the polynomial expression f with respect to the main
variable if v is omitted, with respect to the variable v otherwise; it is assumed that the base ring
has characteristic zero. In other words, considering f as a polynomial function in the variable v,
returns F , a polynomial in v vanishing at 0, such that F (b) − F (a) = sumbv=a+1 f (v):
? sumformal(n) \\ 1 + ... + n
%1 = 1/2*n^2 + 1/2*n
? f(n) = n^3+n^2+1;
? F = sumformal(f(n)) \\ f(1) + ... + f(n)
%3 = 1/4*n^4 + 5/6*n^3 + 3/4*n^2 + 7/6*n
? sum(n = 1, 2000, f(n)) == subst(F, n, 2000)
%4 = 1
? sum(n = 1001, 2000, f(n)) == subst(F, n, 2000) - subst(F, n, 1000)

252
 %5 = 1
? sumformal(x^2 + x*y + y^2, y)
%6 = y*x^2 + (1/2*y^2 + 1/2*y)*x + (1/3*y^3 + 1/2*y^2 + 1/6*y)
? x^2 * y + x * sumformal(y) + sumformal(y^2) == %
%7 = 1
The library syntax is GEN sumformal(GEN f, long v = -1) where v is a variable number.

3.9.57 taylor(x, t, {d = seriesprecision}). Taylor expansion around 0 of x with respect to the

simple variable t. x can be of any reasonable type, for example a rational function. Contrary to
Ser, which takes the valuation into account, this function adds O(td ) to all components of x.
? taylor(x/(1+y), y, 5)
%1 = (y^4 - y^3 + y^2 - y + 1)*x + O(y^5)
? Ser(x/(1+y), y, 5)
*** at top-level: Ser(x/(1+y),y,5)
*** ^----------------
*** Ser: main variable must have higher priority in gtoser.
The library syntax is GEN tayl(GEN x, long t, long precdl) where t is a variable number.

3.9.58 thue(tnf , a, {sol }). Returns all solutions of the equation P (x, y) = a in integers x and y,
where tnf was created with thueinit(P ). If present, sol must contain the solutions of Norm(x) = a
modulo units of positive norm in the number field defined by P (as computed by bnfisintnorm).
If there are infinitely many solutions, an error is issued.
It is allowed to input directly the polynomial P instead of a tnf , in which case, the function
first performs thueinit(P,0). This is very wasteful if more than one value of a is required.
If tnf was computed without assuming GRH (flag 1 in thueinit), then the result is uncondi-
tional. Otherwise, it depends in principle of the truth of the GRH, but may still be unconditionally
correct in some favorable cases. The result is conditional on the GRH if a 6= ±1 and P has a single
irreducible rational factor, whose attached tentative class number h and regulator R (as computed
assuming the GRH) satisfy
• h > 1,
• R/0.2 > 1.5.
Here’s how to solve the Thue equation x13 − 5y 13 = −4:
? tnf = thueinit(x^13 - 5);
? thue(tnf, -4)
%1 = [[1, 1]]
In this case, one checks that bnfinit(x^13 -5).no is 1. Hence, the only solution is (x, y) = (1, 1)
and the result is unconditional. On the other hand:
? P = x^3-2*x^2+3*x-17; tnf = thueinit(P);
? thue(tnf, -15)
%2 = [[1, 1]] \\ a priori conditional on the GRH.
? K = bnfinit(P); K.no
%3 = 3
? K.reg

253
 %4 = 2.8682185139262873674706034475498755834
This time the result is conditional. All results computed using this particular tnf are likewise
conditional, except for a right-hand side of ±1. The above result is in fact correct, so we did not
just disprove the GRH:
? tnf = thueinit(x^3-2*x^2+3*x-17, 1 /*unconditional*/);
? thue(tnf, -15)
%4 = [[1, 1]]
Note that reducible or nonmonic polynomials are allowed:
? tnf = thueinit((2*x+1)^5 * (4*x^3-2*x^2+3*x-17), 1);
? thue(tnf, 128)
%2 = [[-1, 0], [1, 0]]
Reducible polynomials are in fact much easier to handle.

Note. When P is irreducible without a real root, the default strategy is to use brute force enu-
meration in time |a|1/ deg P and avoid computing a tough bnf attached to P , see thueinit. Besides
reusing a quantity you might need for other purposes, the default argument sol can also be used
to use a different strategy and prove that there are no solutions; of course you need to compute a
bnf on you own to obtain sol . If there are solutions this won’t help unless P is quadratic, since
the enumeration will be performed in any case.
The library syntax is GEN thue(GEN tnf, GEN a, GEN sol = NULL).

3.9.59 thueinit(P, {flag = 0}). Initializes the tnf corresponding to P , a nonconstant univariate
polynomial with integer coefficients. The result is meant to be used in conjunction with thue to
solve Thue equations P (X/Y )Y deg P = a, where a is an integer. Accordingly, P must either have
at least two distinct irreducible factors over Q, or have one irreducible factor T with degree > 2 or
two conjugate complex roots: under these (necessary and sufficient) conditions, the equation has
finitely many integer solutions.
? S = thueinit(t^2+1);
? thue(S, 5)
%2 = [[-2, -1], [-2, 1], [-1, -2], [-1, 2], [1, -2], [1, 2], [2, -1], [2, 1]]
? S = thueinit(t+1);
*** at top-level: thueinit(t+1)
*** ^-------------
*** thueinit: domain error in thueinit: P = t + 1
The hardest case is when deg P > 2 and P is irreducible with at least one real root. The routine
then uses Bilu-Hanrot’s algorithm.
If flag is nonzero, certify results unconditionally. Otherwise, assume GRH, this being much
faster of course. In the latter case, the result may still be unconditionally correct, see thue. For
instance in most cases where P is reducible (not a pure power of an irreducible), or conditional
computed class groups are trivial or the right hand side is ±1, then results are unconditional.

254
Note. The general philosophy is to disprove the existence of large solutions then to enumer-
ate bounded solutions naively. The implementation will overflow when there exist huge solutions
and the equation has degree > 2 (the quadratic imaginary case is special, since we can stick to
bnfisintnorm, there are no fundamental units):

? thue(t^3+2, 10^30)
*** at top-level: L=thue(t^3+2,10^30)
*** ^-----------------
*** thue: overflow in thue (SmallSols): y <= 80665203789619036028928.
? thue(x^2+2, 10^30) \\ quadratic case much easier
%1 = [[-1000000000000000, 0], [1000000000000000, 0]]

Note. It is sometimes possible to circumvent the above, and in any case obtain an important speed-
up, if you can write P = Q(xd ) for some d > 1 and Q still satisfying the thueinit hypotheses.
You can then solve the equation attached to Q then eliminate all solutions (x, y) such that either
x or y is not a d-th power.

? thue(x^4+1, 10^40); \\ stopped after 10 hours

? filter(L,d) =
my(x,y); [[x,y] | v<-L, ispower(v[1],d,&x)&&ispower(v[2],d,&y)];
? L = thue(x^2+1, 10^40);
? filter(L, 2)
%4 = [[0, 10000000000], [10000000000, 0]]
The last 2 commands use less than 20ms.

Note. When P is irreducible without a real root, the equation can be solved unconditionnally
in time |a|1/ deg P . When this latter quantity is huge and the equation has no solutions, this fact
may still be ascertained via arithmetic conditions but this now implies solving norm equations,
computing a bnf and possibly assuming the GRH. When there is no real root, the code does not
compute a bnf (with certification if flag = 1) if it expects this to be an “easy” computation (because
the result would only be used for huge values of a). See thue for a way to compute an expensive
bnf on your own and still get a result where this default cheap strategy fails.
The library syntax is GEN thueinit(GEN P, long flag, long prec).

3.10 Vectors, matrices, linear algebra and sets.

Note that most linear algebra functions operating on subspaces defined by generating sets
(such as mathnf, qflll, etc.) take matrices as arguments. As usual, the generating vectors are
taken to be the columns of the given matrix.
Since PARI does not have a strong typing system, scalars live in unspecified commutative base
rings. It is very difficult to write robust linear algebra routines in such a general setting. We thus
assume that the base ring is a domain and work over its field of fractions. If the base ring is not a
domain, one gets an error as soon as a nonzero pivot turns out to be noninvertible. Some functions,
e.g. mathnf or mathnfmod, specifically assume that the base ring is Z.

255
3.10.1 algdep(z, k, {flag = 0}). z being real/complex, or p-adic, finds a polynomial (in the vari-
able ’x) of degree at most k, with integer coefficients, having z as approximate root. Note that the
polynomial which is obtained is not necessarily the “correct” one. In fact it is not even guaranteed
to be irreducible. One can check the closeness either by a polynomial evaluation (use subst), or
by computing the roots of the polynomial given by algdep (use polroots or polrootspadic).

Internally, lindep([1, z, . . . , z k ], flag) is used. A nonzero value of flag may improve on the
default behavior if the input number is known to a huge accuracy, and you suspect the last bits
are incorrect: if flag > 0 the computation is done with an accuracy of flag decimal digits; to get
meaningful results, the parameter flag should be smaller than the number of correct decimal digits
in the input. But default values are usually sufficient, so try without flag first:

? \p200
? z = 2^(1/6)+3^(1/5);
? algdep(z, 30); \\ right in 280ms
? algdep(z, 30, 100); \\ wrong in 169ms
? algdep(z, 30, 170); \\ right in 288ms
? algdep(z, 30, 200); \\ wrong in 320ms
? \p250
? z = 2^(1/6)+3^(1/5); \\ recompute to new, higher, accuracy !
? algdep(z, 30); \\ right in 329ms
? algdep(z, 30, 200); \\ right in 324ms
? \p500
? algdep(2^(1/6)+3^(1/5), 30); \\ right in 677ms
? \p1000
? algdep(2^(1/6)+3^(1/5), 30); \\ right in 1.5s

The changes in realprecision only affect the quality of the initial approximation to 21/6 + 31/5 ,
algdep itself uses exact operations. The size of its operands depend on the accuracy of the input
of course: more accurate input means slower operations.

Proceeding by increments of 5 digits of accuracy, algdep with default flag produces its first
correct result at 195 digits, and from then on a steady stream of correct results:

\\ assume T contains the correct result, for comparison

forstep(d=100, 250, 5, localprec(d);\
print(d, " ", algdep(2^(1/6)+3^(1/5),30) == T))

The above example is the test case studied in a 2000 paper by Borwein and Lisonek: Appli-
cations of integer relation algorithms, Discrete Math., 217, p. 65–82. The version of PARI tested
there was 1.39, which succeeded reliably from precision 265 on, in about 200 as much time as the
current version.

The library syntax is GEN algdep0(GEN z, long k, long flag). Also available is GEN
algdep(GEN z, long k) (flag = 0).

256
3.10.2 bestapprnf(V, T, {rootT }). T being an integral polynomial and V being a scalar, vector,
or matrix with complex coefficients, return a reasonable approximation of V with polmods modulo
T . T can also be any number field structure, in which case the minimal polynomial attached to the
structure (T .pol) is used. The rootT argument, if present, must be an element of polroots(T ) (or
T .pol), i.e. a complex root of T fixing an embedding of Q[x]/(T ) into C.
? bestapprnf(sqrt(5), polcyclo(5))
%1 = Mod(-2*x^3 - 2*x^2 - 1, x^4 + x^3 + x^2 + x + 1)
? bestapprnf(sqrt(5), polcyclo(5), exp(4*I*Pi/5))
%2 = Mod(2*x^3 + 2*x^2 + 1, x^4 + x^3 + x^2 + x + 1)
When the output has huge rational coefficients, try to increase the working realbitprecision:
if the answer does not stabilize, consider that the reconstruction failed. Beware that if T is not
Galois over Q, some embeddings may not allow to reconstruct V :
? T = x^3-2; vT = polroots(T); z = 3*2^(1/3)+1;
? bestapprnf(z, T, vT[1])
%2 = Mod(3*x + 1, x^3 - 2)
? bestapprnf(z, T, vT[2])
%3 = 4213714286230872/186454048314072 \\ close to 3*2^(1/3) + 1
The library syntax is GEN bestapprnf(GEN V, GEN T, GEN rootT = NULL, long prec)
.

3.10.3 charpoly(A, {v =0 x}, {flag = 5}). characteristic polynomial of A with respect to the
variable v, i.e. determinant of v ∗ I − A if A is a square matrix.
? charpoly([1,2;3,4]);
%1 = x^2 - 5*x - 2
? charpoly([1,2;3,4],, ’t)
%2 = t^2 - 5*t - 2
If A is not a square matrix, the function returns the characteristic polynomial of the map “multi-
plication by A” if A is a scalar:
? charpoly(Mod(x+2, x^3-2))
%1 = x^3 - 6*x^2 + 12*x - 10
? charpoly(I)
%2 = x^2 + 1
? charpoly(quadgen(5))
%3 = x^2 - x - 1
? charpoly(ffgen(ffinit(2,4)))
%4 = Mod(1, 2)*x^4 + Mod(1, 2)*x^3 + Mod(1, 2)*x^2 + Mod(1, 2)*x + Mod(1, 2)
The value of flag is only significant for matrices, and we advise to stick to the default value.
Let n be the dimension of A.
If flag = 0, same method (Le Verrier’s) as for computing the adjoint matrix, i.e. using the
traces of the powers of A. Assumes that n! is invertible; uses O(n4 ) scalar operations.
If flag = 1, uses Lagrange interpolation which is usually the slowest method. Assumes that n!
is invertible; uses O(n4 ) scalar operations.
If flag = 2, uses the Hessenberg form. Assumes that the base ring is a field. Uses O(n3 ) scalar
operations, but suffers from coefficient explosion unless the base field is finite or R.

257
 If flag = 3, uses Berkowitz’s division free algorithm, valid over any ring (commutative, with
unit). Uses O(n4 ) scalar operations.
If flag = 4, x must be integral. Uses a modular algorithm: Hessenberg form for various small
primes, then Chinese remainders.
If flag = 5 (default), uses the “best” method given x. This means we use Berkowitz unless the
base ring is Z (use flag = 4) or a field where coefficient explosion does not occur, e.g. a finite field
or the reals (use flag = 2).
The library syntax is GEN charpoly0(GEN A, long v = -1, long flag) where v is a variable
number. Also available are GEN charpoly(GEN x, long v) (flag = 5), GEN caract(GEN A, long
v) (flag = 1), GEN carhess(GEN A, long v) (flag = 2), GEN carberkowitz(GEN A, long v)
(flag = 3) and GEN caradj(GEN A, long v, GEN *pt). In this last case, if pt is not NULL, *pt
receives the address of the adjoint matrix of A (see matadjoint), so both can be obtained at once.

3.10.4 concat(x, {y}). Concatenation of x and y. If x or y is not a vector or matrix, it is considered

as a one-dimensional vector. All types are allowed for x and y, but the sizes must be compatible.
Note that matrices are concatenated horizontally, i.e. the number of rows stays the same. Using
transpositions, one can concatenate them vertically, but it is often simpler to use matconcat.
? x = matid(2); y = 2*matid(2);
? concat(x,y)
%2 =
[1 0 2 0]
[0 1 0 2]
? concat(x~,y~)~
%3 =
[1 0]
[0 1]
[2 0]
[0 2]
? matconcat([x;y])
%4 =
[1 0]
[0 1]
[2 0]
[0 2]
To concatenate vectors sideways (i.e. to obtain a two-row or two-column matrix), use Mat instead,
or matconcat:
? x = [1,2];
? y = [3,4];
? concat(x,y)
%3 = [1, 2, 3, 4]
? Mat([x,y]~)
%4 =
[1 2]

258
 [3 4]
? matconcat([x;y])
%5 =
[1 2]
[3 4]

Concatenating a row vector to a matrix having the same number of columns will add the row
to the matrix (top row if the vector is x, i.e. comes first, and bottom row otherwise).

The empty matrix [;] is considered to have a number of rows compatible with any operation,
in particular concatenation. (Note that this is not the case for empty vectors [ ] or [ ]~.)

If y is omitted, x has to be a row vector or a list, in which case its elements are concatenated,
from left to right, using the above rules.

? concat([1,2], [3,4])
%1 = [1, 2, 3, 4]
? a = [[1,2]~, [3,4]~]; concat(a)
%2 =
[1 3]
[2 4]
? concat([1,2; 3,4], [5,6]~)
%3 =
[1 2 5]
[3 4 6]
? concat([%, [7,8]~, [1,2,3,4]])
%5 =
[1 2 5 7]
[3 4 6 8]
[1 2 3 4]

The library syntax is GEN gconcat(GEN x, GEN y = NULL). GEN gconcat1(GEN x) is a short-
cut for gconcat(x,NULL).

3.10.5 dirpowers(n, x). For nonnegative n and complex number x, return the vector with n
components [1x , 2x , . . . , nx ].

? dirpowers(5, 2)
%1 = [1, 4, 9, 16, 25]
? dirpowers(5, 1/2)
%2 = [1, 1.414..., 1.732..., 2.000..., 2.236...]

When n ≤ 0, the function returns the empty vector [].

The library syntax is GEN dirpowers(long n, GEN x, long prec).

259
3.10.6 forqfvec(v, q, b, expr ). q being a square and symmetric integral matrix representing a
positive definite quadratic form, evaluate expr for all pairs of nonzero vectors (−v, v) such that
q(v) ≤ b. The formal variable v runs through representatives of all such pairs in turn.
? forqfvec(v, [3,2;2,3], 3, print(v))
[0, 1]~
[1, 0]~
[-1, 1]~
The library syntax is void forqfvec0(GEN v, GEN q = NULL, GEN b). The following func-
tions are also available: void forqfvec(void *E, long (*fun)(void *, GEN, GEN, double),
GEN q, GEN b): Evaluate fun(E,U,v,m) on all v such that q(U v) < b, where U is a t_MAT, v is a
t_VECSMALL and m = q(v) is a C double. The function fun must return 0, unless forqfvec should
stop, in which case, it should return 1.
void forqfvec1(void *E, long (*fun)(void *, GEN), GEN q, GEN b): Evaluate fun(E,v)
on all v such that q(v) < b, where v is a t_COL. The function fun must return 0, unless forqfvec
should stop, in which case, it should return 1.

3.10.7 lindep(v, {flag = 0}). finds a small nontrivial integral linear combination between com-
ponents of v. If none can be found return an empty vector.
If v is a vector with real/complex entries we use a floating point (variable precision) LLL
algorithm. If flag = 0 the accuracy is chosen internally using a crude heuristic. If flag > 0 the
computation is done with an accuracy of flag decimal digits. To get meaningful results in the latter
case, the parameter flag should be smaller than the number of correct decimal digits in the input.
? lindep([sqrt(2), sqrt(3), sqrt(2)+sqrt(3)])
%1 = [-1, -1, 1]~
If v is p-adic, flag is ignored and the algorithm LLL-reduces a suitable (dual) lattice.
? lindep([1, 2 + 3 + 3^2 + 3^3 + 3^4 + O(3^5)])
%2 = [1, -2]~
If v is a matrix (or a vector of column vectors, or a vector of row vectors), flag is ignored and
the function returns a non trivial kernel vector if one exists, else an empty vector.
? lindep([1,2,3;4,5,6;7,8,9])
%3 = [1, -2, 1]~
? lindep([[1,0], [2,0]])
%4 = [2, -1]~
? lindep([[1,0], [0,1]])
%5 = []~
If v contains polynomials or power series over some base field, finds a linear relation with
coefficients in the field.
? lindep([x*y, x^2 + y, x^2*y + x*y^2, 1])
%4 = [y, y, -1, -y^2]~
For better control, it is preferable to use t_POL rather than t_SER in the input, otherwise one gets
a linear combination which is t-adically small, but not necessarily 0. Indeed, power series are first
converted to the minimal absolute accuracy occurring among the entries of v (which can cause
some coefficients to be ignored), then truncated to polynomials:

260
 ? v = [t^2+O(t^4), 1+O(t^2)]; L=lindep(v)
%1 = [1, 0]~
? v*L
%2 = t^2+O(t^4) \\ small but not 0
The library syntax is GEN lindep0(GEN v, long flag).

3.10.8 matadjoint(M, {flag = 0}). adjoint matrix of M , i.e. a matrix N of cofactors of M ,

satisfying M ∗ N = det(M ) ∗ Id. M must be a (not necessarily invertible) square matrix of
dimension n. If flag is 0 or omitted, we try to use Leverrier-Faddeev’s algorithm, which assumes
that n! invertible. If it fails or flag = 1, compute T = charpoly(M ) independently first and return
(−1)n−1 (T (x) − T (0))/x evaluated at M .
? a = [1,2,3;3,4,5;6,7,8] * Mod(1,4);
? matadjoint(a)
%2 =
[Mod(1, 4) Mod(1, 4) Mod(2, 4)]
[Mod(2, 4) Mod(2, 4) Mod(0, 4)]
[Mod(1, 4) Mod(1, 4) Mod(2, 4)]
Both algorithms use O(n4 ) operations in the base ring. Over a field, they are usually slower than
computing the characteristic polynomial or the inverse of M directly.
The library syntax is GEN matadjoint0(GEN M, long flag). Also available are GEN adj(GEN
x) (flag=0) and GEN adjsafe(GEN x) (flag=1).

3.10.9 matcompanion(x). The left companion matrix to the nonzero polynomial x.

The library syntax is GEN matcompanion(GEN x).

3.10.10 matconcat(v). Returns a t_MAT built from the entries of v, which may be a t_VEC
(concatenate horizontally), a t_COL (concatenate vertically), or a t_MAT (concatenate vertically each
column, and concatenate vertically the resulting matrices). The entries of v are always considered
as matrices: they can themselves be t_VEC (seen as a row matrix), a t_COL seen as a column
matrix), a t_MAT, or a scalar (seen as an 1 × 1 matrix).
? A=[1,2;3,4]; B=[5,6]~; C=[7,8]; D=9;
? matconcat([A, B]) \\ horizontal
%1 =
[1 2 5]
[3 4 6]
? matconcat([A, C]~) \\ vertical
%2 =
[1 2]
[3 4]
[7 8]
? matconcat([A, B; C, D]) \\ block matrix
%3 =
[1 2 5]
[3 4 6]

261
 [7 8 9]

If the dimensions of the entries to concatenate do not match up, the above rules are extended as
follows:

• each entry vi,j of v has a natural length and height: 1 × 1 for a scalar, 1 × n for a t_VEC of
length n, n × 1 for a t_COL, m × n for an m × n t_MAT

• let Hi be the maximum over j of the lengths of the vi,j , let Lj be the maximum over i of the
heights of the vi,j . The dimensions of the (i, j)-th block in the concatenated matrix are Hi × Lj .

• a scalar s = vi,j is considered as s times an identity matrix of the block dimension min(Hi , Lj )

• blocks are extended by 0 columns on the right and 0 rows at the bottom, as needed.

? matconcat([1, [2,3]~, [4,5,6]~]) \\ horizontal

%4 =
[1 2 4]
[0 3 5]
[0 0 6]
? matconcat([1, [2,3], [4,5,6]]~) \\ vertical
%5 =
[1 0 0]
[2 3 0]
[4 5 6]
? matconcat([B, C; A, D]) \\ block matrix
%6 =
[5 0 7 8]
[6 0 0 0]
[1 2 9 0]
[3 4 0 9]
? U=[1,2;3,4]; V=[1,2,3;4,5,6;7,8,9];
? matconcat(matdiagonal([U, V])) \\ block diagonal
%7 =
[1 2 0 0 0]
[3 4 0 0 0]
[0 0 1 2 3]
[0 0 4 5 6]
[0 0 7 8 9]

The library syntax is GEN matconcat(GEN v).

262
3.10.11 matdet(x, {flag = 0}). Determinant of the square matrix x.
If flag = 0, uses an appropriate algorithm depending on the coefficients:
• integer entries: modular method due to Dixon, Pernet and Stein.
• real or p-adic entries: classical Gaussian elimination using maximal pivot.
• intmod entries: classical Gaussian elimination using first nonzero pivot.
• other cases: Gauss-Bareiss.
If flag = 1, uses classical Gaussian elimination with appropriate pivoting strategy (maximal
pivot for real or p-adic coefficients). This is usually worse than the default.
The library syntax is GEN det0(GEN x, long flag). Also available are GEN det(GEN x)
(flag = 0), GEN det2(GEN x) (flag = 1) and GEN ZM_det(GEN x) for integer entries.

3.10.12 matdetint(B). Let B be an m × n matrix with integer coefficients. The determinant D

of the lattice generated by the columns of B is the square root of det(B T B) if B has maximal rank
m, and 0 otherwise.
This function uses the Gauss-Bareiss algorithm to compute a positive multiple of D. When B
is square, the function actually returns D = | det B|.
This function is useful in conjunction with mathnfmod, which needs to know such a multiple.
If the rank is maximal but the matrix is nonsquare, you can obtain D exactly using

matdet( mathnfmod(B, matdetint(B)) )

Note that as soon as one of the dimensions gets large (m or n is larger than 20, say), it will often
be much faster to use mathnf(B, 1) or mathnf(B, 4) directly.
The library syntax is GEN detint(GEN B).

3.10.13 matdetmod(x, d). Given a matrix x with t_INT entries and d an arbitrary positive
integer, return the determinant of x modulo d.

? A = [4,2,3; 4,5,6; 7,8,9]

? matdetmod(A,27)
%2 = 9
Note that using the generic function matdet on a matrix with t_INTMOD entries uses Gaussian
reduction and will fail in general when the modulus is not prime.

? matdet(A * Mod(1,27))
*** at top-level: matdet(A*Mod(1,27))
*** ^------------------
*** matdet: impossible inverse in Fl_inv: Mod(3, 27).
The library syntax is GEN matdetmod(GEN x, GEN d).

263
3.10.14 matdiagonal(x). x being a vector, creates the diagonal matrix whose diagonal entries
are those of x.
? matdiagonal([1,2,3]);
%1 =
[1 0 0]
[0 2 0]
[0 0 3]
Block diagonal matrices are easily created using matconcat:
? U=[1,2;3,4]; V=[1,2,3;4,5,6;7,8,9];
? matconcat(matdiagonal([U, V]))
%1 =
[1 2 0 0 0]
[3 4 0 0 0]
[0 0 1 2 3]
[0 0 4 5 6]
[0 0 7 8 9]
The library syntax is GEN diagonal(GEN x).

3.10.15 mateigen(x, {flag = 0}). Returns the (complex) eigenvectors of x as columns of a matrix.
If flag = 1, return [L, H], where L contains the eigenvalues and H the corresponding eigenvectors;
multiple eigenvalues are repeated according to the eigenspace dimension (which may be less than
the eigenvalue multiplicity in the characteristic polynomial).
This function first computes the characteristic polynomial of x and approximates its complex
roots (λi ), then tries to compute the eigenspaces as kernels of the x − λi . This algorithm is ill-
conditioned and is likely to miss kernel vectors if some roots of the characteristic polynomial are
close, in particular if it has multiple roots.
? A = [13,2; 10,14]; mateigen(A)
%1 =
[-1/2 2/5]
[ 1 1]
? [L,H] = mateigen(A, 1);
? L
%3 = [9, 18]
? H
%4 =
[-1/2 2/5]
[ 1 1]
? A * H == H * matdiagonal(L)
%5 = 1
For symmetric matrices, use qfjacobi instead; for Hermitian matrices, compute
A = real(x);
B = imag(x);

264
 y = matconcat([A, -B; B, A]);
and apply qfjacobi to y.
The library syntax is GEN mateigen(GEN x, long flag, long prec). Also available is GEN
eigen(GEN x, long prec) (flag = 0)

3.10.16 matfrobenius(M, {flag}, {v =0 x}). Returns the Frobenius form of the square matrix M.
If flag = 1, returns only the elementary divisors as a vector of polynomials in the variable v. If
flag = 2, returns a two-components vector [F,B] where F is the Frobenius form and B is the basis
change so that M = B −1 F B.
The library syntax is GEN matfrobenius(GEN M, long flag, long v = -1) where v is a
variable number.

3.10.17 mathess(x). Returns a matrix similar to the square matrix x, which is in upper Hessenberg
form (zero entries below the first subdiagonal).
The library syntax is GEN hess(GEN x).

3.10.18 mathilbert(n). x being a long, creates the Hilbert matrixof order x, i.e. the matrix
whose coefficient (i,j) is 1/(i + j − 1).
The library syntax is GEN mathilbert(long n).

3.10.19 mathnf(M, {flag = 0}). Let R be a Euclidean ring, equal to Z or to K[X] for some
field K. If M is a (not necessarily square) matrix with entries in R, this routine finds the upper
triangular Hermite normal form of M . If the rank of M is equal to its number of rows, this is a
square matrix. In general, the columns of the result form a basis of the R-module spanned by the
columns of M .
The values of flag are:
• 0 (default): only return the Hermite normal form H
• 1 (complete output): return [H, U ], where H is the Hermite normal form of M , and U is a
transformation matrix such that M U = [0|H]. The matrix U belongs to GL(R). When M has a
large kernel, the entries of U are in general huge.
For these two values, we use a naive algorithm, which behaves well in small dimension only. Larger
values correspond to different algorithms, are restricted to integer matrices, and all output the
unimodular matrix U . From now on all matrices have integral entries.
• flag = 4, returns [H, U ] as in “complete output” above, using a variant of LLL reduction
along the way. The matrix U is provably small in the L2 sense, and often close to optimal; but the
reduction is in general slow, although provably polynomial-time.
If flag = 5, uses Batut’s algorithm and output [H, U, P ], such that H and U are as before
and P is a permutation of the rows such that P applied to M U gives H. This is in general faster
than flag = 4 but the matrix U is usually worse; it is heuristically smaller than with the default
algorithm.
When the matrix is dense and the dimension is large (bigger than 100, say), flag = 4 will be
fastest. When M has maximal rank, then
H = mathnfmod(M, matdetint(M))

265
will be even faster. You can then recover U as M −1 H.
? M = matrix(3,4,i,j,random([-5,5]))
%1 =
[ 0 2 3 0]
[-5 3 -5 -5]
[ 4 3 -5 4]
? [H,U] = mathnf(M, 1);
? U
%3 =
[-1 0 -1 0]
[ 0 5 3 2]
[ 0 3 1 1]
[ 1 0 0 0]
? H
%5 =
[19 9 7]
[ 0 9 1]
[ 0 0 1]
? M*U
%6 =
[0 19 9 7]
[0 0 9 1]
[0 0 0 1]
For convenience, M is allowed to be a t_VEC, which is then automatically converted to a t_MAT,
as per the Mat function. For instance to solve the generalized extended gcd problem, one may use
? v = [116085838, 181081878, 314252913,10346840];
? [H,U] = mathnf(v, 1);
? U
%2 =
[ 103 -603 15 -88]
[-146 13 -1208 352]
[ 58 220 678 -167]
[-362 -144 381 -101]
? v*U
%3 = [0, 0, 0, 1]
This also allows to input a matrix as a t_VEC of t_COLs of the same length (which Mat would
concatenate to the t_MAT having those columns):
? v = [[1,0,4]~, [3,3,4]~, [0,-4,-5]~]; mathnf(v)
%1 =
[47 32 12]
[ 0 1 0]

266
 [ 0 0 1]

The library syntax is GEN mathnf0(GEN M, long flag). Also available are GEN hnf(GEN M)
(flag = 0) and GEN hnfall(GEN M) (flag = 1). To reduce huge relation matrices (sparse with small
entries, say dimension 400 or more), you can use the pair hnfspec / hnfadd. Since this is quite
technical and the calling interface may change, they are not documented yet. Look at the code in
basemath/hnf snf.c.

3.10.20 mathnfmod(x, d). If x is a (not necessarily square) matrix of maximal rank with integer
entries, and d is a multiple of the (nonzero) determinant of the lattice spanned by the columns of
x, finds the upper triangular Hermite normal form of x.

If the rank of x is equal to its number of rows, the result is a square matrix. In general, the
columns of the result form a basis of the lattice spanned by the columns of x. Even when d is
known, this is in general slower than mathnf but uses much less memory.

The library syntax is GEN hnfmod(GEN x, GEN d).

3.10.21 mathnfmodid(x, d). Outputs the (upper triangular) Hermite normal form of x concate-
nated with the diagonal matrix with diagonal d. Assumes that x has integer entries. Variant: if d
is an integer instead of a vector, concatenate d times the identity matrix.

? m=[0,7;-1,0;-1,-1]
%1 =
[ 0 7]
[-1 0]
[-1 -1]
? mathnfmodid(m, [6,2,2])
%2 =
[2 1 1]
[0 1 0]
[0 0 1]
? mathnfmodid(m, 10)
%3 =
[10 7 3]
[ 0 1 0]
[ 0 0 1]

The library syntax is GEN hnfmodid(GEN x, GEN d).

267
3.10.22 mathouseholder(Q, v). applies a sequence Q of Householder transforms, as returned by
matqr(M, 1) to the vector or matrix v.
? m = [2,1; 3,2]; \\ some random matrix
? [Q,R] = matqr(m);
? Q
%3 =
[-0.554... -0.832...]
[-0.832... 0.554...]
? R
%4 =
[-3.605... -2.218...]
[0 0.277...]
? v = [1, 2]~; \\ some random vector
? Q * v
%6 = [-2.218..., 0.277...]~
? [q,r] = matqr(m, 1);
? exponent(r - R) \\ r is the same as R
%8 = -128
? q \\ but q has a different structure
%9 = [[0.0494..., [5.605..., 3]]]]
? mathouseholder(q, v) \\ applied to v
%10 = [-2.218..., 0.277...]~
The point of the Householder structure is that it efficiently represents the linear operator v 7→ Qv
in a more stable way than expanding the matrix Q:
? m = mathilbert(20); v = vectorv(20,i,i^2+1);
? [Q,R] = matqr(m);
? [q,r] = matqr(m, 1);
? \p100
? [q2,r2] = matqr(m, 1); \\ recompute at higher accuracy
? exponent(R - r)
%5 = -127
? exponent(R - r2)
%6 = -127
? exponent(mathouseholder(q,v) - mathouseholder(q2,v))
%7 = -119
? exponent(Q*v - mathouseholder(q2,v))
%8 = 9
We see that R is OK with or without a flag to matqr but that multiplying by Q is considerably
less precise than applying the sequence of Householder transforms encoded by q.
The library syntax is GEN mathouseholder(GEN Q, GEN v).

3.10.23 matid(n). Creates the n × n identity matrix.

The library syntax is GEN matid(long n).

268
3.10.24 matimage(x, {flag = 0}). Gives a basis for the image of the matrix x as columns of a
matrix. A priori the matrix can have entries of any type. If flag = 0, use standard Gauss pivot. If
flag = 1, use matsupplement (much slower: keep the default flag!).

The library syntax is GEN matimage0(GEN x, long flag). Also available is GEN image(GEN
x) (flag = 0).

3.10.25 matimagecompl(x). Gives the vector of the column indices which are not extracted
by the function matimage, as a permutation (t_VECSMALL). Hence the number of components of
matimagecompl(x) plus the number of columns of matimage(x) is equal to the number of columns
of the matrix x.

The library syntax is GEN imagecompl(GEN x).

3.10.26 matimagemod(x, d, &U ). Gives a Howell basis (unique representation for submodules
of (Z/dZ)n ) for the image of the matrix x modulo d as columns of a matrix H. The matrix x must
have t_INT entries, and d can be an arbitrary positive integer. If U is present, set it to a matrix
such that AU = H.

? A = [2,1;0,2];
? matimagemod(A,6,&U)
%2 =
[1 0]
[0 2]
? U
%3 =
[5 1]
[3 4]
? (A*U)%6
%4 =
[1 0]
[0 2]

Caveat. In general the number of columns of the Howell form is not the minimal number of
generators of the submodule. Example:

? matimagemod([1;2],4)
%5 =
[2 1]
[0 2]

269
Caveat 2. In general the matrix U is not invertible, even if A and H have the same size. Example:
? matimagemod([4,1;0,4],8,&U)
%6 =
[2 1]
[0 4]
? U
%7 =
[0 0]
[2 1]
The library syntax is GEN matimagemod(GEN x, GEN d, GEN *U = NULL).

3.10.27 matindexrank(M ). M being a matrix of rank r, returns a vector with two t_VECSMALL
components y and z of length r giving a list of rows and columns respectively (starting from 1) such
that the extracted matrix obtained from these two vectors using vecextract(M, y, z) is invertible.
The vectors y and z are sorted in increasing order.
The library syntax is GEN indexrank(GEN M).

3.10.28 matintersect(x, y). x and y being two matrices with the same number of rows each of
whose columns are independent, finds a basis of the vector space equal to the intersection of the
spaces spanned by the columns of x and y respectively. The faster function idealintersect can be
used to intersect fractional ideals (projective ZK modules of rank 1); the slower but more general
function nfhnf can be used to intersect general ZK -modules.
The library syntax is GEN intersect(GEN x, GEN y).

3.10.29 matinverseimage(x, y). Given a matrix x and a column vector or matrix y, returns a
preimage z of y by x if one exists (i.e such that xz = y), an empty vector or matrix otherwise. The
complete inverse image is z + Kerx, where a basis of the kernel of x may be obtained by matker.
? M = [1,2;2,4];
? matinverseimage(M, [1,2]~)
%2 = [1, 0]~
? matinverseimage(M, [3,4]~)
%3 = []~ \\ no solution
? matinverseimage(M, [1,3,6;2,6,12])
%4 =
[1 3 6]
[0 0 0]
? matinverseimage(M, [1,2;3,4])
%5 = [;] \\ no solution
? K = matker(M)
%6 =
[-2]
[1]
The library syntax is GEN inverseimage(GEN x, GEN y).

270
3.10.30 matinvmod(x, d). Computes a left inverse of the matrix x modulo d. The matrix x must
have t_INT entries, and d can be an arbitrary positive integer.

? A = [3,1,2;1,2,1;3,1,1];
? U = matinvmod(A,6)
%2 =
[1 1 3]
[2 3 5]
[1 0 5]
? (U*A)%6
%3 =
[1 0 0]
[0 1 0]
[0 0 1]
? matinvmod(A,5)
*** at top-level: matinvmod(A,5)
*** ^--------------
*** matinvmod: impossible inverse in gen_inv: 0.

The library syntax is GEN matinvmod(GEN x, GEN d).

3.10.31 matisdiagonal(x). Returns true (1) if x is a diagonal matrix, false (0) if not.

The library syntax is GEN isdiagonal(GEN x).

3.10.32 matker(x, {flag = 0}). Gives a basis for the kernel of the matrix x as columns of a matrix.
The matrix can have entries of any type, provided they are compatible with the generic arithmetic
operations (+, × and /).

If x is known to have integral entries, set flag = 1.

The library syntax is GEN matker0(GEN x, long flag). Also available are GEN ker(GEN x)
(flag = 0), GEN ZM_ker(GEN x) (flag = 1).

3.10.33 matkerint(x, {flag = 0}). Gives an LLL-reduced Z-basis for the lattice equal to the kernel
of the matrix x with rational entries. flag is deprecated, kept for backward compatibility.

The library syntax is GEN matkerint0(GEN x, long flag). Use directly GEN kerint(GEN x)
if x is known to have integer entries, and Q_primpart first otherwise.

271
3.10.34 matkermod(x, d, &im). Gives a Howell basis (unique representation for submodules
of (Z/dZ)n , cf. matimagemod) for the kernel of the matrix x modulo d as columns of a matrix. The
matrix x must have t_INT entries, and d can be an arbitrary positive integer. If im is present, set
it to a basis of the image of x (which is computed on the way).
? A = [1,2,3;5,1,4]
%1 =
[1 2 3]
[5 1 4]
? K = matkermod(A,6)
%2 =
[2 1]
[2 1]
[0 3]
? (A*K)%6
%3 =
[0 0]
[0 0]
The library syntax is GEN matkermod(GEN x, GEN d, GEN *im = NULL).

3.10.35 matmuldiagonal(x, d). Product of the matrix x by the diagonal matrix whose diagonal
entries are those of the vector d. Equivalent to, but much faster than x ∗ matdiagonal(d).
The library syntax is GEN matmuldiagonal(GEN x, GEN d).

3.10.36 matmultodiagonal(x, y). Product of the matrices x and y assuming that the result is a
diagonal matrix. Much faster than x ∗ y in that case. The result is undefined if x ∗ y is not diagonal.
The library syntax is GEN matmultodiagonal(GEN x, GEN y).

3.10.37 matpascal(n, {q}). Creates as a matrix the lower triangular Pascal triangle of order x + 1
(i.e. with binomial coefficients up to x). If q is given, compute the q-Pascal triangle (i.e. using
q-binomial coefficients).
The library syntax is GEN matqpascal(long n, GEN q = NULL). Also available is GEN mat-
pascal(GEN x).

3.10.38 matpermanent(x). Permanent of the square matrix x using Ryser’s formula in Gray
code order.
? n = 20; m = matrix(n,n,i,j, i!=j);
? matpermanent(m)
%2 = 895014631192902121
? n! * sum(i=0,n, (-1)^i/i!)
%3 = 895014631192902121
This function runs in time O(2n n) for a matrix of size n and is not implemented for n large.
The library syntax is GEN matpermanent(GEN x).

272
3.10.39 matqr(M, {flag = 0}). Returns [Q, R], the QR-decomposition of the square invertible
matrix M with real entries: Q is orthogonal and R upper triangular. If flag = 1, the orthogonal
matrix is returned as a sequence of Householder transforms: applying such a sequence is stabler
and faster than multiplication by the corresponding Q matrix. More precisely, if
[Q,R] = matqr(M);
[q,r] = matqr(M, 1);
then r = R and mathouseholder(q, M ) is (close to) R; furthermore
mathouseholder(q, matid(#M)) == Q~
the inverse of Q. This function raises an error if the precision is too low or x is singular.
The library syntax is GEN matqr(GEN M, long flag, long prec).

3.10.40 matrank(x). Rank of the matrix x.

The library syntax is long rank(GEN x).

3.10.41 matreduce(m). Let m be a factorization matrix, i.e., a 2-column matrix whose columns
contains arbitrary “generators” and integer “exponents” respectively. Returns the canonical form
of m: the first column is sorted with unique elements and the second one contains the merged
“exponents” (exponents of identical entries in the first column of m are added, rows attached to
0 exponents are deleted). The generators are sorted with respect to the universal cmp routine; in
particular, this function is the identity on true integer factorization matrices, but not on other
factorizations (in products of polynomials or maximal ideals, say). It is idempotent.
For convenience, this function also allows a vector m, which is handled as a factorization with
all exponents equal to 1, as in factorback.
? A=[x,2;y,4]; B=[x,-2; y,3; 3, 4]; C=matconcat([A,B]~)
%1 =
[x 2]
[y 4]
[x -2]
[y 3]
[3 4]
? matreduce(C)
%2 =
[3 4]
[y 7]
? matreduce([x,x,y,x,z,x,y]) \\ vector argument
%3 =
[x 4]
[y 2]
[z 1]
The library syntax is GEN matreduce(GEN m).

273
3.10.42 matrix(m, {n = m}, {X}, {Y }, {expr = 0}). Creation of the m × n matrix whose coef-
ficients are given by the expression expr . There are two formal parameters in expr , the first one
(X) corresponding to the rows, the second (Y ) to the columns, and X goes from 1 to m, Y goes
from 1 to n. If one of the last 3 parameters is omitted, fill the matrix with zeroes. If n is omitted,
return a square m × m matrix.

3.10.43 matrixqz(A, {p = 0}). A being an m × n matrix in Mm,n (Q), let ImQ A (resp. ImZ A)
the Q-vector space (resp. the Z-module) spanned by the columns of A. This function has varying
behavior depending on the sign of p:
If p ≥ 0, A is assumed to have maximal rank n ≤ m. The function returns a matrix B ∈
Mm,n (Z), with ImQ B = ImQ A, such that the GCD of all its n × n minors is coprime to p; in
particular, if p = 0 (default), this GCD is 1.
If p = −1, returns a basis of the lattice Zn ∩ ImZ A.
If p = −2, returns a basis of the lattice Zn ∩ ImQ A.

Caveat. (p = −1 or −2) For efficiency reason, we do not compute the HNF of the resulting basis.
? minors(x) = vector(#x[,1], i, matdet(x[^i,]));
? A = [3,1/7; 5,3/7; 7,5/7]; minors(A)
%1 = [4/7, 8/7, 4/7] \\ determinants of all 2x2 minors
? B = matrixqz(A)
%2 =
[3 1]
[5 2]
[7 3]
? minors(%)
%3 = [1, 2, 1] \\ B integral with coprime minors
? matrixqz(A,-1)
%4 =
[3 1]
[5 3]
[7 5]
? matrixqz(A,-2)
%5 =
[3 1]
[5 2]
[7 3]

The library syntax is GEN matrixqz0(GEN A, GEN p = NULL).

3.10.44 matsize(x). x being a vector or matrix, returns a row vector with two components, the
first being the number of rows (1 for a row vector), the second the number of columns (1 for a
column vector).
The library syntax is GEN matsize(GEN x).

274
3.10.45 matsnf(X, {flag = 0}). If X is a (singular or nonsingular) matrix outputs the vector
of elementary divisors of X, i.e. the diagonal of the Smith normal form of X, normalized so that
dn | dn−1 | . . . | d1 . X must have integer or polynomial entries; in the latter case, X must be a
square matrix.

The binary digits of flag mean:

1 (complete output): if set, outputs [U, V, D], where U and V are two unimodular matrices
such that U XV is the diagonal matrix D. Otherwise output only the diagonal of D. If X is not a
square matrix, then D will be a square diagonal matrix padded with zeros on the left or the top.

4 (cleanup): if set, cleans up the output. This means that elementary divisors equal to 1 will be
deleted, i.e. outputs a shortened vector D0 instead of D. If complete output was required, returns
[U 0 , V 0 , D0 ] so that U 0 XV 0 = D0 holds. If this flag is set, X is allowed to be of the form ‘vector of
elementary divisors’ or [U, V, D] as would normally be output with the cleanup flag unset.

The library syntax is GEN matsnf0(GEN X, long flag).

3.10.46 matsolve(M, B). Let M be a left-invertible matrix and B a column vector such that
there exists a solution X to the system of linear equations M X = B; return the (unique) solution
X. This has the same effect as, but is faster, than M −1 ∗ B. Uses Dixon p-adic lifting method if
M and B are integral and Gaussian elimination otherwise. When there is no solution, the function
returns an X such that M X − B is nonzero although it has at least #M zero entries:

? M = [1,2;3,4;5,6];
? B = [4,6,8]~; X = matsolve(M, B)
%2 = [-2, 3]~
? M*X == B
%3 = 1
? B = [1,2,4]~; X = matsolve(M, [1,2,4]~)
%4 = [0, 1/2]~
? M*X - B
%5 = [0, 0, -1]~

Raises an exception if M is not left-invertible, even if there is a solution:

? M = [1,1;1,1]; matsolve(M, [1,1]~)

*** at top-level: matsolve(M,[1,1]~)
*** ^------------------
*** matsolve: impossible inverse in gauss: [1, 1; 1, 1].

The function also works when B is a matrix and we return the unique matrix solution X provided
it exists.

The library syntax is GEN gauss(GEN M, GEN B). For integral input, the function GEN
ZM_gauss(GEN M, GEN B) is also available.

275
3.10.47 matsolvemod(M, D, B, {flag = 0}). M being any integral matrix, D a column vector
of nonnegative integer P
moduli, and B an integral column vector, gives an integer solution to the
system of congruences i mi,j xj ≡ bi (mod di ) if one exists, otherwise returns zero. Shorthand
notation: B (resp. D) can be given as a single integer, in which case all the bi (resp. di ) above are
taken to be equal to B (resp. D).
? M = [1,2;3,4];
? matsolvemod(M, [3,4]~, [1,2]~)
%2 = [10, 0]~
? matsolvemod(M, 3, 1) \\ M X = [1,1]~ over F_3
%3 = [2, 1]~
? matsolvemod(M, [3,0]~, [1,2]~) \\ x + 2y = 1 (mod 3), 3x + 4y = 2 (in Z)
%4 = [6, -4]~
If flag = 1, all solutions are returned in the form of a two-component row vector [x, u], where
x is an integer solution to the system of congruences and u is a matrix whose columns give a
basis of the homogeneous system (so that all solutions can be obtained by adding x to any linear
combination of columns of u). If no solution exists, returns zero.
The library syntax is GEN matsolvemod(GEN M, GEN D, GEN B, long flag). Also available
are GEN gaussmodulo(GEN M, GEN D, GEN B) (flag = 0) and GEN gaussmodulo2(GEN M, GEN D,
GEN B) (flag = 1).

3.10.48 matsupplement(x). Assuming that the columns of the matrix x are linearly independent
(if they are not, an error message is issued), finds a square invertible matrix whose first columns
are the columns of x, i.e. supplement the columns of x to a basis of the whole space.
? matsupplement([1;2])
%1 =
[1 0]
[2 1]
Raises an error if x has 0 columns, since (due to a long standing design bug), the dimension
of the ambient space (the number of rows) is unknown in this case:
? matsupplement(matrix(2,0))
*** at top-level: matsupplement(matrix
*** ^--------------------
*** matsupplement: sorry, suppl [empty matrix] is not yet implemented.
The library syntax is GEN suppl(GEN x).

3.10.49 mattranspose(x). Transpose of x (also x˜). This has an effect only on vectors and
matrices.
The library syntax is GEN gtrans(GEN x).

3.10.50 minpoly(A, {v =0 x}). minimal polynomial of A with respect to the variable v., i.e. the
monic polynomial P of minimal degree (in the variable v) such that P (A) = 0.
The library syntax is GEN minpoly(GEN A, long v = -1) where v is a variable number.

276
3.10.51 norml2(x). Square of the L2 -norm of x. More precisely, if x is a scalar, norml2(x)
is defined to be the square of the complex modulus of x (real t_QUADs are not supported). If
x is a polynomial, a (row or column) vector or a matrix, norml2(x) is defined recursively as
P
Pi norml2(x i ),
Pwhere2 (xi ) run through the components of x. In particular, this yields the usual
|xi |2 (resp. |xi,j | ) if x is a polynomial or vector (resp. matrix) with complex components.

? norml2( [ 1, 2, 3 ] ) \\ vector
%1 = 14
? norml2( [ 1, 2; 3, 4] ) \\ matrix
%2 = 30
? norml2( 2*I + x )
%3 = 5
? norml2( [ [1,2], [3,4], 5, 6 ] ) \\ recursively defined
%4 = 91

The library syntax is GEN gnorml2(GEN x).

3.10.52 normlp(x, {p = oo}). Lp -norm of x; sup norm if p is omitted or +oo. More precisely, if x
is a scalar, normlp(x, p) is defined to be abs(x). If x is a polynomial, a (row or column) vector or
a matrix:

• if p is omitted or +oo, then normlp(x) is defined recursively as maxi normlp(xi )), where (xi )
run through the components of x. In particular, this yields the usual sup norm if x is a polynomial
or vector with complex components.

p) is defined recursively as ( i normlpp (xi , p))1/p . In particular, this

P
• otherwise, normlp(x,
yields the usual ( |xi |p )1/p if x is a polynomial or vector with complex components.
P

? v = [1,-2,3]; normlp(v) \\ vector

%1 = 3
? normlp(v, +oo) \\ same, more explicit
%2 = 3
? M = [1,-2;-3,4]; normlp(M) \\ matrix
%3 = 4
? T = (1+I) + I*x^2; normlp(T)
%4 = 1.4142135623730950488016887242096980786
? normlp([[1,2], [3,4], 5, 6]) \\ recursively defined
%5 = 6
? normlp(v, 1)
%6 = 6
? normlp(M, 1)
%7 = 10
? normlp(T, 1)
%8 = 2.4142135623730950488016887242096980786

The library syntax is GEN gnormlp(GEN x, GEN p = NULL, long prec).

277
3.10.53 powers(x, n, {x0 }). For nonnegative n, return the vector with n + 1 components
[1, x, . . . , xn ] if x0 is omitted, and [x0 , x0 ∗ x, ..., x0 ∗ xn ] otherwise.

? powers(Mod(3,17), 4)
%1 = [Mod(1, 17), Mod(3, 17), Mod(9, 17), Mod(10, 17), Mod(13, 17)]
? powers(Mat([1,2;3,4]), 3)
%2 = [[1, 0; 0, 1], [1, 2; 3, 4], [7, 10; 15, 22], [37, 54; 81, 118]]
? powers(3, 5, 2)
%3 = [2, 6, 18, 54, 162, 486]

When n < 0, the function returns the empty vector [].

The library syntax is GEN gpowers0(GEN x, long n, GEN x0 = NULL). Also available is GEN
gpowers(GEN x, long n) when x0 is NULL.

3.10.54 qfauto(G, {fl }). G being a square and symmetric matrix with integer entries representing
a positive definite quadratic form, outputs the automorphism group of the associate lattice. Since
this requires computing the minimal vectors, the computations can become very lengthy as the
dimension grows. G can also be given by an qfisominit structure. See qfisominit for the
meaning of fl .

The output is a two-components vector [o, g] where o is the group order and g is the list of
generators (as a vector). For each generator H, the equality G = t HGH holds.

The interface of this function is experimental and will likely change in the future.

This function implements an algorithm of Plesken and Souvignier, following Souvignier’s im-
plementation.

The library syntax is GEN qfauto0(GEN G, GEN fl = NULL). The function GEN qfauto(GEN
G, GEN fl) is also available where G is a vector of zm matrices.

3.10.55 qfautoexport(qfa, {flag}). qfa being an automorphism group as output by qfauto,

export the underlying matrix group as a string suitable for (no flags or flag = 0) GAP or (flag = 1)
Magma. The following example computes the size of the matrix group using GAP:

? G = qfauto([2,1;1,2])
%1 = [12, [[-1, 0; 0, -1], [0, -1; 1, 1], [1, 1; 0, -1]]]
? s = qfautoexport(G)
%2 = "Group([[-1, 0], [0, -1]], [[0, -1], [1, 1]], [[1, 1], [0, -1]])"
? extern("echo \"Order("s");\" | gap -q")
%3 = 12

The library syntax is GEN qfautoexport(GEN qfa, long flag).

3.10.56 qfbil(x, y, {q}). This function is obsolete, use qfeval.

The library syntax is GEN qfbil(GEN x, GEN y, GEN q = NULL).

278
3.10.57 qfeval({q}, x, {y}). Evaluate the quadratic form q (given by a symmetric matrix) at the
vector x; if y is present, evaluate the polar form at (x, y); if q omitted, use the standard Euclidean
scalar product, corresponding to the identity matrix.
Roughly equivalent to x~* q * y, but a little faster and more convenient (does not distinguish
between column and row vectors):
? x = [1,2,3]~; y = [-1,3,1]~; q = [1,2,3;2,2,-1;3,-1,9];
? qfeval(q,x,y)
%2 = 23
? for(i=1,10^6, qfeval(q,x,y))
time = 661ms
? for(i=1,10^6, x~*q*y)
time = 697ms
The speedup is noticeable for the quadratic form, compared to x~* q * x, since we save almost
half the operations:
? for(i=1,10^6, qfeval(q,x))
time = 487ms
The special case q = Id is handled faster if we omit q altogether:
? qfeval(,x,y)
%6 = 8
? q = matid(#x);
? for(i=1,10^6, qfeval(q,x,y))
time = 529 ms.
? for(i=1,10^6, qfeval(,x,y))
time = 228 ms.
? for(i=1,10^6, x~*y)
time = 274 ms.
We also allow t_MATs of compatible dimensions for x, and return x~* q * x in this case as
well:
? M = [1,2,3;4,5,6;7,8,9]; qfeval(,M) \\ Gram matrix
%5 =
[66 78 90]
[78 93 108]
[90 108 126]
? q = [1,2,3;2,2,-1;3,-1,9];
? for(i=1,10^6, qfeval(q,M))
time = 2,008 ms.
? for(i=1,10^6, M~*q*M)
time = 2,368 ms.
? for(i=1,10^6, qfeval(,M))
time = 1,053 ms.
? for(i=1,10^6, M~*M)
time = 1,171 ms.

279
 If q is a t_QFI or t_QFR, it is implicitly converted to the attached symmetric t_MAT. This is
done more efficiently than by direct conversion, since we avoid introducing a denominator 2 and
rational arithmetic:
? q = Qfb(2,3,4); x = [2,3];
? qfeval(q, x)
%2 = 62
? Q = Mat(q)
%3 =
[ 2 3/2]
[3/2 4]
? qfeval(Q, x)
%4 = 62
? for (i=1, 10^6, qfeval(q,x))
time = 758 ms.
? for (i=1, 10^6, qfeval(Q,x))
time = 1,110 ms.
Finally, when x is a t_MAT with integral coefficients, we allow a t_QFI or t_QFR for q and
return the binary quadratic form q ◦ M . Again, the conversion to t_MAT is less efficient in this case:
? q = Qfb(2,3,4); Q = Mat(q); x = [1,2;3,4];
? qfeval(q, x)
%2 = Qfb(47, 134, 96)
? qfeval(Q,x)
%3 =
[47 67]
[67 96]
? for (i=1, 10^6, qfeval(q,x))
time = 701 ms.
? for (i=1, 10^6, qfeval(Q,x))
time = 1,639 ms.
The library syntax is GEN qfeval0(GEN q = NULL, GEN x, GEN y = NULL).

3.10.58 qfgaussred(q). decomposition into squares of the quadratic form represented by the
symmetric matrix q. The result is a matrix whose diagonal entries are the coefficients of the
squares, and the off-diagonal entries on each line represent the bilinear forms. More precisely, if
(aij ) denotes the output, one has
X X
q(x) = aii (xi + aij xj )2
i j6=i

? qfgaussred([0,1;1,0])
%1 =
[1/2 1]
[-1 -1/2]
This means that 2xy = (1/2)(x + y)2 − (1/2)(x − y)2 . Singular matrices are supported, in which
case some diagonal coefficients will vanish:

280
 ? qfgaussred([1,1;1,1])
%1 =
[1 1]
[1 0]

This means that x2 + 2xy + y 2 = (x + y)2 .

The library syntax is GEN qfgaussred(GEN q). GEN qfgaussred_positive(GEN q) assumes

that q is positive definite and is a little faster; returns NULL if a vector with negative norm occurs
(non positive matrix or too many rounding errors).

3.10.59 qfisom(G, H, {fl }, {grp}). G, H being square and symmetric matrices with integer entries
representing positive definite quadratic forms, return an invertible matrix S such that G = t SH ×
S. This defines a isomorphism between the corresponding lattices. Since this requires computing
the minimal vectors, the computations can become very lengthy as the dimension grows. See
qfisominit for the meaning of fl . If grp is given it must be the automorphism group of H. It will
be used to speed up the computation.

G can also be given by an qfisominit structure which is preferable if several forms H need
to be compared to G.

This function implements an algorithm of Plesken and Souvignier, following Souvignier’s im-
plementation.

The library syntax is GEN qfisom0(GEN G, GEN H, GEN fl = NULL, GEN grp = NULL)
. Also available is GEN qfisom(GEN G, GEN H, GEN fl, GEN grp) where G is a vector of zm, and
H is a zm, and grp is either NULL or a vector of zm.

3.10.60 qfisominit(G, {fl }, {m}). G being a square and symmetric matrix with integer entries
representing a positive definite quadratic form, return an isom structure allowing to compute
isomorphisms between G and other quadratic forms faster.

The interface of this function is experimental and will likely change in future release.

If present, the optional parameter fl must be a t_VEC with two components. It allows to
specify the invariants used, which can make the computation faster or slower. The components are

• fl[1] Depth of scalar product combination to use.

• fl[2] Maximum level of Bacher polynomials to use.

If present, m must be the set of vectors of norm up to the maximal of the diagonal entry of G,
either as a matrix or as given by qfminim. Otherwise this function computes the minimal vectors
so it become very lengthy as the dimension of G grows.

The library syntax is GEN qfisominit0(GEN G, GEN fl = NULL, GEN m = NULL). Also
available is GEN qfisominit(GEN F, GEN fl) where F is a vector of zm.

281
3.10.61 qfjacobi(A). Apply Jacobi’s eigenvalue algorithm to the real symmetric matrix A. This
returns [L, V ], where
• L is the vector of (real) eigenvalues of A, sorted in increasing order,
• V is the corresponding orthogonal matrix of eigenvectors of A.

? \p19
? A = [1,2;2,1]; mateigen(A)
%1 =
[-1 1]
[ 1 1]
? [L, H] = qfjacobi(A);
? L
%3 = [-1.000000000000000000, 3.000000000000000000]~
? H
%4 =
[ 0.7071067811865475245 0.7071067811865475244]
[-0.7071067811865475244 0.7071067811865475245]
? norml2( (A-L[1])*H[,1] ) \\ approximate eigenvector
%5 = 9.403954806578300064 E-38
? norml2(H*H~ - 1)
%6 = 2.350988701644575016 E-38 \\ close to orthogonal
The library syntax is GEN jacobi(GEN A, long prec).

3.10.62 qflll(x, {flag = 0}). LLL algorithm applied to the columns of the matrix x. The columns
of x may be linearly dependent. The result is by default a unimodular transformation matrix T such
that x · T is an LLL-reduced basis of the lattice generated by the column vectors of x. Note that if
x is not of maximal rank T will not be square. The LLL parameters are (0.51, 0.99), meaning that
the Gram-Schmidt coefficients for the final basis satisfy |µi,j | ≤ 0.51, and the Lovász’s constant is
0.99.
If flag = 0 (default), assume that x has either exact (integral or rational) or real floating point
entries. The matrix is rescaled, converted to integers and the behavior is then as in flag = 1.
If flag = 1, assume that x is integral. Computations involving Gram-Schmidt vectors are
approximate, with precision varying as needed (Lehmer’s trick, as generalized by Schnorr). Adapted
from Nguyen and Stehlé’s algorithm and Stehlé’s code (fplll-1.3).
If flag = 2, x should be an integer matrix whose columns are linearly independent. Returns
a partially reduced basis for x, using an unpublished algorithm by Peter Montgomery: a basis is
said to be partially reduced if |vi ± vj | ≥ |vi | for any two distinct basis vectors vi , vj . This is faster
than flag = 1, esp. when one row is huge compared to the other rows (knapsack-style), and should
quickly produce relatively short vectors. The resulting basis is not LLL-reduced in general. If LLL
reduction is eventually desired, avoid this partial reduction: applying LLL to the partially reduced
matrix is significantly slower than starting from a knapsack-type lattice.
If flag = 3, as flag = 1, but the reduction is performed in place: the routine returns x · T . This
is usually faster for knapsack-type lattices.

282
 If flag = 4, as flag = 1, returning a vector [K, T ] of matrices: the columns of K represent a
basis of the integer kernel of x (not LLL-reduced in general) and T is the transformation matrix
such that x · T is an LLL-reduced Z-basis of the image of the matrix x.
If flag = 5, case as case 4, but x may have polynomial coefficients.
If flag = 8, same as case 0, but x may have polynomial coefficients.
? \p500
realprecision = 500 significant digits
? a = 2*cos(2*Pi/97);
? C = 10^450;
? v = powers(a,48); b = round(matconcat([matid(48),C*v]~));
? p = b * qflll(b)[,1]; \\ tiny linear combination of powers of ’a’
time = 4,470 ms.
? exponent(v * p / C)
%5 = -1418
? p3 = qflll(b,3)[,1]; \\ compute in place, faster
time = 3,790 ms.
? p3 == p \\ same result
%7 = 1
? p2 = b * qflll(b,2)[,1]; \\ partial reduction: faster, not as good
time = 343 ms.
? exponent(v * p2 / C)
%9 = -1190
The library syntax is GEN qflll0(GEN x, long flag). Also available are GEN lll(GEN x)
(flag = 0), GEN lllint(GEN x) (flag = 1), and GEN lllkerim(GEN x) (flag = 4).

3.10.63 qflllgram(G, {flag = 0}). Same as qflll, except that the matrix G = x~ ∗ x is the Gram
matrix of some lattice vectors x, and not the coordinates of the vectors themselves. In particular,
G must now be a square symmetric real matrix, corresponding to a positive quadratic form (not
necessarily definite: x needs not have maximal rank). The result is a unimodular transformation
matrix T such that x · T is an LLL-reduced basis of the lattice generated by the column vectors of
x. See qflll for further details about the LLL implementation.
If flag = 0 (default), assume that G has either exact (integral or rational) or real floating point
entries. The matrix is rescaled, converted to integers and the behavior is then as in flag = 1.
If flag = 1, assume that G is integral. Computations involving Gram-Schmidt vectors are
approximate, with precision varying as needed (Lehmer’s trick, as generalized by Schnorr). Adapted
from Nguyen and Stehlé’s algorithm and Stehlé’s code (fplll-1.3).
flag = 4: G has integer entries, gives the kernel and reduced image of x.
flag = 5: same as 4, but G may have polynomial coefficients.
The library syntax is GEN qflllgram0(GEN G, long flag). Also available are GEN lll-
gram(GEN G) (flag = 0), GEN lllgramint(GEN G) (flag = 1), and GEN lllgramkerim(GEN G)
(flag = 4).

283
3.10.64 qfminim(x, {B}, {m}, {flag = 0}). x being a square and symmetric matrix of dimension
d representing a positive definite quadratic form, this function deals with the vectors of x whose
norm is less than or equal to B, enumerated using the Fincke-Pohst algorithm, storing at most m
pairs of vectors: only one vector is given for each pair ±v. There is no limit if m is omitted: beware
that this may be a huge vector! The vectors are returned in no particular order.
The function searches for the minimal nonzero vectors if B is omitted. The behavior is unde-
fined if x is not positive definite (a “precision too low” error is most likely, although more precise
error messages are possible). The precise behavior depends on flag.
• If flag = 0 (default), return [N, M, V ], where N is the number of vectors enumerated (an
even number, possibly larger than 2m), M ≤ B is the maximum norm found, and V is a matrix
whose columns are found vectors.
• If flag = 1, ignore m and return [M, v], where v is a nonzero vector of length M ≤ B. If no
nonzero vector has length ≤ B, return []. If no explicit B is provided, return a vector of smallish
norm, namely the vector of smallest length (usually the first one but not always) in an LLL-reduced
basis for x.
In these two cases, x must have integral small entries: more precisely, we definitely must have
d · kxk2∞ < 253 but even that may not be enough. The implementation uses low precision floating
point computations for maximal speed and gives incorrect results when x has large entries. That
condition is checked in the code and the routine raises an error if large rounding errors occur. A
more robust, but much slower, implementation is chosen if the following flag is used:
• If flag = 2, x can have non integral real entries, but this is also useful when x has large
integral entries. Return [N, M, V ] as in case flag = 0, where M is returned as a floating point
number. If x is inexact and B is omitted, the “minimal” vectors in V only have approximately the
same norm (up to the internal working accuracy). This version is very robust but still offers no
hard and fast guarantee about the result: it involves floating point operations performed at a high
floating point precision depending on your input, but done without rigorous tracking of roundoff
errors (as would be provided by interval arithmetic for instance). No example is known where the
input is exact but the function returns a wrong result.
? x = matid(2);
? qfminim(x) \\ 4 minimal vectors of norm 1: ±[0, 1], ±[1, 0]
%2 = [4, 1, [0, 1; 1, 0]]
? { x = \\ The Leech lattice
[4, 2, 0, 0, 0,-2, 0, 0, 0, 0, 0, 0, 1,-1, 0, 0, 0, 1, 0,-1, 0, 0, 0,-2;
2, 4,-2,-2, 0,-2, 0, 0, 0, 0, 0, 0, 0,-1, 0, 0, 0, 0, 0,-1, 0, 1,-1,-1;
0,-2, 4, 0,-2, 0, 0, 0, 0, 0, 0, 0,-1, 1, 0, 0, 1, 0, 0, 1,-1,-1, 0, 0;
0,-2, 0, 4, 0, 0, 0, 0, 0, 0, 0, 0, 1,-1, 0, 0, 0, 1,-1, 0, 1,-1, 1, 0;
0, 0,-2, 0, 4, 0, 0, 0, 1,-1, 0, 0, 1, 0, 0, 0,-2, 0, 0,-1, 1, 1, 0, 0;
-2, -2,0, 0, 0, 4,-2, 0,-1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0,-1, 1, 1;
0, 0, 0, 0, 0,-2, 4,-2, 0, 0, 0, 0, 0, 1, 0, 0, 0,-1, 0, 0, 0, 1,-1, 0;
0, 0, 0, 0, 0, 0,-2, 4, 0, 0, 0, 0,-1, 0, 0, 0, 0, 0,-1,-1,-1, 0, 1, 0;
0, 0, 0, 0, 1,-1, 0, 0, 4, 0,-2, 0, 1, 1, 0,-1, 0, 1, 0, 0, 0, 0, 0, 0;
0, 0, 0, 0,-1, 0, 0, 0, 0, 4, 0, 0, 1, 1,-1, 1, 0, 0, 0, 1, 0, 0, 1, 0;
0, 0, 0, 0, 0, 0, 0, 0,-2, 0, 4,-2, 0,-1, 0, 0, 0,-1, 0,-1, 0, 0, 0, 0;
0, 0, 0, 0, 0, 0, 0, 0, 0, 0,-2, 4,-1, 1, 0, 0,-1, 1, 0, 1, 1, 1,-1, 0;
1, 0,-1, 1, 1, 0, 0,-1, 1, 1, 0,-1, 4, 0, 0, 1, 0, 1, 1, 0, 1, 0, 1,-1;
-1,-1, 1,-1, 0, 0, 1, 0, 1, 1,-1, 1, 0, 4, 1, 1, 0, 0, 1, 1, 0, 1, 0, 1;

284
 0, 0, 0, 0, 0, 0, 0, 0, 0,-1, 0, 0, 0, 1, 4, 0, 0, 0, 1, 0, 0, 0, 0, 0;
0, 0, 0, 0, 0, 0, 0, 0,-1, 1, 0, 0, 1, 1, 0, 4, 0, 0, 0, 0, 1, 1, 0, 0;
0, 0, 1, 0,-2, 0, 0, 0, 0, 0, 0,-1, 0, 0, 0, 0, 4, 1, 1, 1, 0, 0, 1, 1;
1, 0, 0, 1, 0, 0,-1, 0, 1, 0,-1, 1, 1, 0, 0, 0, 1, 4, 0, 1, 1, 0, 1, 0;
0, 0, 0,-1, 0, 1, 0,-1, 0, 0, 0, 0, 1, 1, 1, 0, 1, 0, 4, 0, 1, 1, 0, 1;
-1, -1,1, 0,-1, 1, 0,-1, 0, 1,-1, 1, 0, 1, 0, 0, 1, 1, 0, 4, 0, 0, 1, 1;
0, 0,-1, 1, 1, 0, 0,-1, 0, 0, 0, 1, 1, 0, 0, 1, 0, 1, 1, 0, 4, 1, 0, 1;
0, 1,-1,-1, 1,-1, 1, 0, 0, 0, 0, 1, 0, 1, 0, 1, 0, 0, 1, 0, 1, 4, 0, 1;
0,-1, 0, 1, 0, 1,-1, 1, 0, 1, 0,-1, 1, 0, 0, 0, 1, 1, 0, 1, 0, 0, 4, 1;
-2,-1, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0,-1, 1, 0, 0, 1, 0, 1, 1, 1, 1, 1, 4]; }
? qfminim(x,,0) \\ 0: don’t store minimal vectors
time = 121 ms.
%4 = [196560, 4, [;]] \\ 196560 minimal vectors of norm 4
? qfminim(x) \\ store all minimal vectors !
time = 821 ms.
? qfminim(x,,0,2); \\ safe algorithm. Slower and unnecessary here.
time = 5,540 ms.
%6 = [196560, 4.000061035156250000, [;]]
? qfminim(x,,,2); \\ safe algorithm; store all minimal vectors
time = 6,602 ms.
In this example, storing 0 vectors limits memory use; storing all of them requires a parisize about
50MB. All minimal vectors are nevertheless enumerated in both cases of course, which means the
speedup is likely to be marginal.
The library syntax is GEN qfminim0(GEN x, GEN B = NULL, GEN m = NULL, long flag,
long prec). Also available are GEN minim(GEN x, GEN B = NULL, GEN m = NULL) (flag = 0),
GEN minim2(GEN x, GEN B = NULL, GEN m = NULL) (flag = 1). GEN minim_raw(GEN x, GEN B
= NULL, GEN m = NULL) (do not perform LLL reduction on x and return NULL on accuracy error).
GEN minim_zm(GEN x, GEN B = NULL, GEN m = NULL) (flag = 0, return vectors as t_VECSMALL
to save memory)

3.10.65 qfnorm(x, {q}). This function is obsolete, use qfeval.

The library syntax is GEN qfnorm(GEN x, GEN q = NULL).

3.10.66 qforbits(G, V ). Return the orbits of V under the action of the group of linear transfor-
mation generated by the set G. It is assumed that G contains minus identity, and only one vector
in {v, −v} should be given. If G does not stabilize V , the function return 0.
In the example below, we compute representatives and lengths of the orbits of the vectors of
norm ≤ 3 under the automorphisms of the lattice Z6 .
? Q=matid(6); G=qfauto(Q); V=qfminim(Q,3);
? apply(x->[x[1],#x],qforbits(G,V))
%2 = [[[0,0,0,0,0,1]~,6],[[0,0,0,0,1,-1]~,30],[[0,0,0,1,-1,-1]~,80]]
The library syntax is GEN qforbits(GEN G, GEN V).

285
3.10.67 qfparam(G, sol , {flag = 0}). Coefficients of binary quadratic forms that parametrize the
solutions of the ternary quadratic form G, using the particular solution sol . flag is optional and
can be 1, 2, or 3, in which case the flag-th form is reduced. The default is flag=0 (no reduction).
? G = [1,0,0;0,1,0;0,0,-34];
? M = qfparam(G, qfsolve(G))
%2 =
[ 3 -10 -3]
[-5 -6 5]
[ 1 0 1]
Indeed, the solutions can be parametrized as

(3x2 − 10xy − 3y 2 )2 + (−5x2 − 6xy + 5y 2 )2 − 34(x2 + y 2 )2 = 0.

? v = y^2 * M*[1,x/y,(x/y)^2]~
%3 = [3*x^2 - 10*y*x - 3*y^2, -5*x^2 - 6*y*x + 5*y^2, -x^2 - y^2]~
? v~*G*v
%4 = 0
The library syntax is GEN qfparam(GEN G, GEN sol, long flag).

3.10.68 qfperfection(G). G being a square and symmetric matrix with integer entries representing
a positive definite quadratic form, outputs the perfection rank of the form. That is, gives the rank
of the family of the s symmetric matrices vv t , where v runs through the minimal vectors.
The algorithm computes the minimal vectors and its runtime is exponential in the dimension
of x.
The library syntax is GEN qfperfection(GEN G).

3.10.69 qfrep(q, B, {flag = 0}). q being a square and symmetric matrix with integer entries
representing a positive definite quadratic form, count the vectors representing successive integers.
• If flag = 0, count all vectors. Outputs the vector whose i-th entry, 1 ≤ i ≤ B is half the
number of vectors v such that q(v) = i.
• If flag = 1, count vectors of even norm. Outputs the vector whose i-th entry, 1 ≤ i ≤ B is
half the number of vectors such that q(v) = 2i.
? q = [2, 1; 1, 3];
? qfrep(q, 5)
%2 = Vecsmall([0, 1, 2, 0, 0]) \\ 1 vector of norm 2, 2 of norm 3, etc.
? qfrep(q, 5, 1)
%3 = Vecsmall([1, 0, 0, 1, 0]) \\ 1 vector of norm 2, 0 of norm 4, etc.
This routine uses a naive algorithm based on qfminim, and will fail if any entry becomes larger
than 231 (or 263 ).
The library syntax is GEN qfrep0(GEN q, GEN B, long flag).

286
3.10.70 qfsign(x). Returns [p, m] the signature of the quadratic form represented by the symmetric
matrix x. Namely, p (resp. m) is the number of positive (resp. negative) eigenvalues of x. The
result is computed using Gaussian reduction.

The library syntax is GEN qfsign(GEN x).

3.10.71 qfsolve(G). Given a square symmetric matrix G of dimension n ≥ 1, solve over Q

the quadratic equation X t GX = 0. The matrix G must have rational coefficients. The solution
might be a single nonzero vector (vectorv) or a matrix (whose columns generate a totally isotropic
subspace).

If no solution exists, returns an integer, that can be a prime p such that there is no local
solution at p, or −1 if there is no real solution, or −2 if n = 2 and − det G is positive but not a
square (which implies there is a real solution, but no local solution at some p dividing det G).

? G = [1,0,0;0,1,0;0,0,-34];
? qfsolve(G)
%1 = [-3, -5, 1]~
? qfsolve([1,0; 0,2])
%2 = -1 \\ no real solution
? qfsolve([1,0,0;0,3,0; 0,0,-2])
%3 = 3 \\ no solution in Q_3
? qfsolve([1,0; 0,-2])
%4 = -2 \\ no solution, n = 2

The library syntax is GEN qfsolve(GEN G).

3.10.72 seralgdep(s, p, r). finds a linear relation between powers (1, s, . . . , sp ) of the series s,
with polynomial coefficients of degree ≤ r. In case no relation is found, return 0.

? s = 1 + 10*y - 46*y^2 + 460*y^3 - 5658*y^4 + 77740*y^5 + O(y^6);

? seralgdep(s, 2, 2)
%2 = -x^2 + (8*y^2 + 20*y + 1)
? subst(%, x, s)
%3 = O(y^6)
? seralgdep(s, 1, 3)
%4 = (-77*y^2 - 20*y - 1)*x + (310*y^3 + 231*y^2 + 30*y + 1)
? seralgdep(s, 1, 2)
%5 = 0

The series main variable must not be x, so as to be able to express the result as a polynomial in x.

The library syntax is GEN seralgdep(GEN s, long p, long r).

287
3.10.73 setbinop(f, X, {Y }). The set whose elements are the f(x,y), where x,y run through X,Y.
respectively. If Y is omitted, assume that X = Y and that f is symmetric: f (x, y) = f (y, x) for all
x, y in X.
? X = [1,2,3]; Y = [2,3,4];
? setbinop((x,y)->x+y, X,Y) \\ set X + Y
%2 = [3, 4, 5, 6, 7]
? setbinop((x,y)->x-y, X,Y) \\ set X - Y
%3 = [-3, -2, -1, 0, 1]
? setbinop((x,y)->x+y, X) \\ set 2X = X + X
%2 = [2, 3, 4, 5, 6]
The library syntax is GEN setbinop(GEN f, GEN X, GEN Y = NULL).

3.10.74 setintersect(x, y). Intersection of the two sets x and y (see setisset). If x or y is not a
set, the result is undefined.
The library syntax is GEN setintersect(GEN x, GEN y).

3.10.75 setisset(x). Returns true (1) if x is a set, false (0) if not. In PARI, a set is a row vector
whose entries are strictly increasing with respect to a (somewhat arbitrary) universal comparison
function. To convert any object into a set (this is most useful for vectors, of course), use the
function Set.
? a = [3, 1, 1, 2];
? setisset(a)
%2 = 0
? Set(a)
%3 = [1, 2, 3]
The library syntax is long setisset(GEN x).

3.10.76 setminus(x, y). Difference of the two sets x and y (see setisset), i.e. set of elements of
x which do not belong to y. If x or y is not a set, the result is undefined.
The library syntax is GEN setminus(GEN x, GEN y).

3.10.77 setsearch(S, x, {flag = 0}). Determines whether x belongs to the set S (see setisset).
We first describe the default behavior, when flag is zero or omitted. If x belongs to the set S,
returns the index j such that S[j] = x, otherwise returns 0.
? T = [7,2,3,5]; S = Set(T);
? setsearch(S, 2)
%2 = 1
? setsearch(S, 4) \\ not found
%3 = 0
? setsearch(T, 7) \\ search in a randomly sorted vector
%4 = 0 \\ WRONG !
If S is not a set, we also allow sorted lists with respect to the cmp sorting function, without repeated
entries, as per listsort(L, 1); otherwise the result is undefined.
? L = List([1,4,2,3,2]); setsearch(L, 4)

288
 %1 = 0 \\ WRONG !
? listsort(L, 1); L \\ sort L first
%2 = List([1, 2, 3, 4])
? setsearch(L, 4)
%3 = 4 \\ now correct
If flag is nonzero, this function returns the index j where x should be inserted, and 0 if it already
belongs to S. This is meant to be used for dynamically growing (sorted) lists, in conjunction with
listinsert.
? L = List([1,5,2,3,2]); listsort(L,1); L
%1 = List([1,2,3,5])
? j = setsearch(L, 4, 1) \\ 4 should have been inserted at index j
%2 = 4
? listinsert(L, 4, j); L
%3 = List([1, 2, 3, 4, 5])
The library syntax is long setsearch(GEN S, GEN x, long flag).

3.10.78 setunion(x, y). Union of the two sets x and y (see setisset). If x or y is not a set, the
result is undefined.
The library syntax is GEN setunion(GEN x, GEN y).

3.10.79 trace(x). This applies to quite general x. If x is not a matrix, it is equal to the sum of x
and its conjugate, except for polmods where it is the trace as an algebraic number.
For x a square matrix, it is the ordinary trace. If x is a nonsquare matrix (but not a vector),
an error occurs.
The library syntax is GEN gtrace(GEN x).

3.10.80 vecextract(x, y, {z}). Extraction of components of the vector or matrix x according to

y. In case x is a matrix, its components are the columns of x. The parameter y is a component
specifier, which is either an integer, a string describing a range, or a vector.
If y is an integer, it is considered as a mask: the binary bits of y are read from right to left,
but correspond to taking the components from left to right. For example, if y = 13 = (1101)2 then
the components 1,3 and 4 are extracted.
If y is a vector (t_VEC, t_COL or t_VECSMALL), which must have integer entries, these entries
correspond to the component numbers to be extracted, in the order specified.
If y is a string, it can be
• a single (nonzero) index giving a component number (a negative index means we start
counting from the end).
• a range of the form "a..b", where a and b are indexes as above. Any of a and b can be
omitted; in this case, we take as default values a = 1 and b = −1, i.e. the first and last components
respectively. We then extract all components in the interval [a, b], in reverse order if b < a.
In addition, if the first character in the string is ^, the complement of the given set of indices
is taken.

289
 If z is not omitted, x must be a matrix. y is then the row specifier, and z the column specifier,
where the component specifier is as explained above.
? v = [a, b, c, d, e];
? vecextract(v, 5) \\ mask
%1 = [a, c]
? vecextract(v, [4, 2, 1]) \\ component list
%2 = [d, b, a]
? vecextract(v, "2..4") \\ interval
%3 = [b, c, d]
? vecextract(v, "-1..-3") \\ interval + reverse order
%4 = [e, d, c]
? vecextract(v, "^2") \\ complement
%5 = [a, c, d, e]
? vecextract(matid(3), "2..", "..")
%6 =
[0 1 0]
[0 0 1]
The range notations v[i..j] and v[^i] (for t_VEC or t_COL) and M[i..j, k..l] and friends
(for t_MAT) implement a subset of the above, in a simpler and faster way, hence should be preferred
in most common situations. The following features are not implemented in the range notation:
• reverse order,
• omitting either a or b in a..b.
The library syntax is GEN extract0(GEN x, GEN y, GEN z = NULL).

3.10.81 vecprod(v). Return the product of the components of the vector v. Return 1 on an
empty vector.
? vecprod([1,2,3])
%1 = 6
? vecprod([])
%2 = 1
The library syntax is GEN vecprod(GEN v).

3.10.82 vecsearch(v, x, {cmpf }). Determines whether x belongs to the sorted vector or list v:
return the (positive) index where x was found, or 0 if it does not belong to v.
If the comparison function cmpf is omitted, we assume that v is sorted in increasing order,
according to the standard comparison function lex, thereby restricting the possible types for x and
the elements of v (integers, fractions, reals, and vectors of such). We also transparently allow a
t_VECSMALL x in this case, for the natural ordering of the integers.
If cmpf is present, it is understood as a comparison function and we assume that v is sorted
according to it, see vecsort for how to encode comparison functions.
? v = [1,3,4,5,7];
? vecsearch(v, 3)
%2 = 2
? vecsearch(v, 6)

290
 %3 = 0 \\ not in the list
? vecsearch([7,6,5], 5) \\ unsorted vector: result undefined
%4 = 0
Note that if we are sorting with respect to a key which is expensive to compute (e.g. a discriminant),
one should rather precompute all keys, sort that vector and search in the vector of keys, rather
than searching in the original vector with respect to a comparison function.
By abuse of notation, x is also allowed to be a matrix, seen as a vector of its columns; again by
abuse of notation, a t_VEC is considered as part of the matrix, if its transpose is one of the matrix
columns.
? v = vecsort([3,0,2; 1,0,2]) \\ sort matrix columns according to lex order
%1 =
[0 2 3]
[0 2 1]
? vecsearch(v, [3,1]~)
%2 = 3
? vecsearch(v, [3,1]) \\ can search for x or x~
%3 = 3
? vecsearch(v, [1,2])
%4 = 0 \\ not in the list

The library syntax is long vecsearch(GEN v, GEN x, GEN cmpf = NULL).

3.10.83 vecsort(x, {cmpf }, {flag = 0}). Sorts the vector x in ascending order, using a mergesort
method. x must be a list, vector or matrix (seen as a vector of its columns). Note that mergesort
is stable, hence the initial ordering of “equal” entries (with respect to the sorting criterion) is not
changed.
If cmpf is omitted, we use the standard comparison function lex, thereby restricting the
possible types for the elements of x (integers, fractions or reals and vectors of those). We also
transparently allow a t_VECSMALL x in this case, for the standard ordering on the integers.
If cmpf is present, it is understood as a comparison function and we sort according to it. The
following possibilities exist:
• an integer k: sort according to the value of the k-th subcomponents of the components of x.
• a vector: sort lexicographically according to the components listed in the vector. For example,
if cmpf = [2, 1, 3], sort with respect to the second component, and when these are equal, with respect
to the first, and when these are equal, with respect to the third.
• a comparison function: t_CLOSURE with two arguments x and y, and returning a real number
which is < 0, > 0 or = 0 if x < y, x > y or x = y respectively.
• a key: t_CLOSURE with one argument x and returning the value f (x) with respect to which
we sort.
? vecsort([3,0,2; 1,0,2]) \\ sort columns according to lex order
%1 =
[0 2 3]
[0 2 1]

291
 ? vecsort(v, (x,y)->y-x) \\ reverse sort
? vecsort(v, (x,y)->abs(x)-abs(y)) \\ sort by increasing absolute value
? vecsort(v, abs) \\ sort by increasing absolute value, using key
? cmpf(x,y) = my(dx = poldisc(x), dy = poldisc(y)); abs(dx) - abs(dy);
? v = [x^2+1, x^3-2, x^4+5*x+1] vecsort(v, cmpf) \\ comparison function
? vecsort(v, x->abs(poldisc(x))) \\ key

The abs and cmpf examples show how to use a named function instead of an anonymous function.
It is preferable to use a key whenever possible rather than include it in the comparison function as
above since the key is evaluated O(n) times instead of O(n log n), where n is the number of entries.

A direct approach is also possible and equivalent to using a sorting key:

? T = [abs(poldisc(x)) | x<-v];
? perm = vecsort(T,,1); \\ indirect sort
? vecextract(v, perm)

This also provides the vector T of all keys, which is interesting for instance in later vecsearch
calls: it is more efficient to sort T (T = vecextract(T, perm)) then search for a key in T rather
than to search in v using a comparison function or a key. Note also that mapisdefined is often
easier to use and faster than vecsearch.

The binary digits of flag mean:

• 1: indirect sorting of the vector x, i.e. if x is an n-component vector, returns a permutation

of [1, 2, . . . , n] which applied to the components of x sorts x in increasing order. For example,
vecextract(x, vecsort(x,,1)) is equivalent to vecsort(x).

• 4: use descending instead of ascending order.

• 8: remove “duplicate” entries with respect to the sorting function (keep the first occurring
entry). For example:

? vecsort([Pi,Mod(1,2),z], (x,y)->0, 8) \\ make everything compare equal

%1 = [3.141592653589793238462643383]
? vecsort([[2,3],[0,1],[0,3]], 2, 8)
%2 = [[0, 1], [2, 3]]

The library syntax is GEN vecsort0(GEN x, GEN cmpf = NULL, long flag).

3.10.84 vecsum(v). Return the sum of the components of the vector v. Return 0 on an empty
vector.

? vecsum([1,2,3])
%1 = 6
? vecsum([])
%2 = 0

The library syntax is GEN vecsum(GEN v).

292
3.10.85 vector(n, {X}, {expr = 0}). Creates a row vector (type t_VEC) with n components whose
components are the expression expr evaluated at the integer points between 1 and n. If the last
two arguments are omitted, fills the vector with zeroes.
? vector(3,i, 5*i)
%1 = [5, 10, 15]
? vector(3)
%2 = [0, 0, 0]
The variable X is lexically scoped to each evaluation of expr . Any change to X within expr
does not affect subsequent evaluations, it still runs 1 to n. A local change allows for example
different indexing:
vector(10, i, i=i-1; f(i)) \\ i = 0, ..., 9
vector(10, i, i=2*i; f(i)) \\ i = 2, 4, ..., 20
This per-element scope for X differs from for loop evaluations, as the following example shows:
n = 3
v = vector(n); vector(n, i, i++) ----> [2, 3, 4]
v = vector(n); for (i = 1, n, v[i] = i++) ----> [2, 0, 4]

3.10.86 vectorsmall(n, {X}, {expr = 0}). Creates a row vector of small integers (type
t_VECSMALL) with n components whose components are the expression expr evaluated at the integer
points between 1 and n.

3.10.87 vectorv(n, {X}, {expr = 0}). As vector, but returns a column vector (type t_COL).

3.11 Transcendental functions.

Since the values of transcendental functions cannot be exactly represented, these functions will
always return an inexact object: a real number, a complex number, a p-adic number or a power
series. All these objects have a certain finite precision.
As a general rule, which of course in some cases may have exceptions, transcendental functions
operate in the following way:
• If the argument is either a real number or an inexact complex number (like 1.0 + I or
Pi*I but not 2 - 3*I), then the computation is done with the precision of the argument. In the
example below, we see that changing the precision to 50 digits does not matter, because x only had
a precision of 19 digits.
? \p 15
realprecision = 19 significant digits (15 digits displayed)
? x = Pi/4
%1 = 0.785398163397448
? \p 50
realprecision = 57 significant digits (50 digits displayed)
? sin(x)
%2 = 0.7071067811865475244
Note that even if the argument is real, the result may be complex (e.g. acos(2.0) or acosh(0.0)).
See each individual function help for the definition of the branch cuts and choice of principal value.

293
 • If the argument is either an integer, a rational, an exact complex number or a quadratic
number, it is first converted to a real or complex number using the current precision, which can be
view and manipulated using the defaults realprecision (in decimal digits) or realbitprecision
(in bits). This precision can be changed indifferently
• in decimal digits: use \p or default(realprecision,...).
• in bits: use \pb or default(realbitprecision,...).
After this conversion, the computation proceeds as above for real or complex arguments.
In library mode, the realprecision does not matter; instead the precision is taken from the
prec parameter which every transcendental function has. As in gp, this prec is not used when the
argument to a function is already inexact. Note that the argument prec stands for the length in
words of a real number, including codewords. Hence we must have prec ≥ 3. (Some functions allow
a bitprec argument instead which allow finer granularity.)
Some accuracies attainable on 32-bit machines cannot be attained on 64-bit machines for parity
reasons. For example the default gp accuracy is 28 decimal digits on 32-bit machines, corresponding
to prec having the value 5, but this cannot be attained on 64-bit machines.
• If the argument is a polmod (representing an algebraic number), then the function is evaluated
for every possible complex embedding of that algebraic number. A column vector of results is
returned, with one component for each complex embedding. Therefore, the number of components
equals the degree of the t_POLMOD modulus.
• If the argument is an intmod or a p-adic, at present only a few functions like sqrt (square
root), sqr (square), log, exp, powering, teichmuller (Teichmüller character) and agm (arithmetic-
geometric mean) are implemented.
Note that in the case of a 2-adic number, sqr(x) may not be identical to x ∗ x: for example if
x = 1 + O(25 ) and y = 1 + O(25 ) then x ∗ y = 1 + O(25 ) while sqr(x) = 1 + O(26 ). Here, x ∗ x yields
the same result as sqr(x) since the two operands are known to be identical . The same statement
holds true for p-adics raised to the power n, where vp (n) > 0.

Remark. If we wanted to be strictly consistent with the PARI philosophy, we should have x ∗ y =
(4 mod 8) and sqr(x) = (4 mod 32) when both x and y are congruent to 2 modulo 4. However,
since intmod is an exact object, PARI assumes that the modulus must not change, and the result is
hence (0 mod 4) in both cases. On the other hand, p-adics are not exact objects, hence are treated
differently.
• If the argument is a polynomial, a power series or a rational function, it is, if necessary,
first converted to a power series using the current series precision, held in the default series-
precision. This precision (the number of significant terms) can be changed using \ps or de-
fault(seriesprecision,...). Then the Taylor series expansion of the function around X = 0
(where X is the main variable) is computed to a number of terms depending on the number of
terms of the argument and the function being computed.
Under gp this again is transparent to the user. When programming in library mode, however,
it is strongly advised to perform an explicit conversion to a power series first, as in
x = gtoser(x, gvar(x), seriesprec)
where the number of significant terms seriesprec can be specified explicitly. If you do not do
this, a global variable precdl is used instead, to convert polynomials and rational functions to a

294
power series with a reasonable number of terms; tampering with the value of this global variable is
deprecated and strongly discouraged.
• If the argument is a vector or a matrix, the result is the componentwise evaluation of the
function. In particular, transcendental functions on square matrices, which are not implemented
in the present version 2.13.3, will have a different name if they are implemented some day.
P (−1)n
3.11.1 Catalan. Catalan’s constant G = n>=0 (2n+1) 2 = 0.91596 · · ·. Note that Catalan is one

of the few reserved names which cannot be used for user variables.
The library syntax is GEN mpcatalan(long prec).

3.11.2 Euler. Euler’s constant γ = 0.57721 · · ·. Note that Euler is one of the few reserved names
which cannot be used for user variables.
The library syntax is GEN mpeuler(long prec).
√
3.11.3 I. The complex number −1.
The library syntax is GEN gen_I().

3.11.4 Pi. The constant π (3.14159 · · ·). Note that Pi is one of the few reserved names which
cannot be used for user variables.
The library syntax is GEN mppi(long prec).

3.11.5 abs(x). Absolute value of x (modulus if x is complex). Rational functions are not allowed.
Contrary to most transcendental functions, an exact argument is not converted to a real number
before applying abs and an exact result is returned if possible.
? abs(-1)
%1 = 1
? abs(3/7 + 4/7*I)
%2 = 5/7
? abs(1 + I)
%3 = 1.414213562373095048801688724
If x is a polynomial, returns −x if the leading coefficient is real and negative else returns x. For a
power series, the constant coefficient is considered instead.
The library syntax is GEN gabs(GEN x, long prec).
√
3.11.6 acos(x). Principal branch of cos−1 (x) = −i log(x + i 1 − x2 ). In particular, <(acos(x)) ∈
[0, π] and if x ∈ R and |x| > 1, then acos(x) is complex. The branch cut is in two pieces:
] − ∞, −1] , continuous with quadrant II, and [1, +∞[, continuous with quadrant IV. We have
acos(x) = π/2 − asin(x) for all x.
The library syntax is GEN gacos(GEN x, long prec).

3.11.7 acosh(x). Principal branch of cosh−1 (x) = 2 log( (x + 1)/2 + (x − 1)/2). In particular,
p p

<(acosh(x)) ≥ 0 and =(acosh(x)) ∈] − π, π]; if x ∈ R and x < 1, then acosh(x) is complex.

The library syntax is GEN gacosh(GEN x, long prec).

295
3.11.8 agm(x, y). Arithmetic-geometric mean of x and y. In the case of complex or negative
numbers, the optimal AGM is returned (the largest in absolute value over all choices of the signs of
the square roots). p-adic or power series arguments are also allowed. Note that a p-adic agm exists
only if x/y is congruent to 1 modulo p (modulo 16 for p = 2). x and y cannot both be vectors or
matrices.
The library syntax is GEN agm(GEN x, GEN y, long prec).

3.11.9 airy(z). Airy [Ai, Bi] functions of argument z.

? [A,B] = airy(1);
? A
%2 = 0.13529241631288141552414742351546630617
? B
%3 = 1.2074235949528712594363788170282869954

The library syntax is GEN airy(GEN z, long prec).

3.11.10 arg(x). Argument of the complex number x, such that −π < arg(x) ≤ π.
The library syntax is GEN garg(GEN x, long prec).
√
3.11.11 asin(x). Principal branch of sin−1 (x) = −i log(ix + 1 − x2 ). In particular, <(asin(x)) ∈
[−π/2, π/2] and if x ∈ R and |x| > 1 then asin(x) is complex. The branch cut is in two pieces:
] − ∞, −1], continuous with quadrant II, and [1, +∞[ continuous with quadrant IV. The function
satisfies iasin(x) = asinh(ix).
The library syntax is GEN gasin(GEN x, long prec).
√
3.11.12 asinh(x). Principal branch of sinh−1 (x) = log(x + 1 + x2 ). In particular =(asinh(x)) ∈
[−π/2, π/2]. The branch cut is in two pieces: ] − i∞, −i], continuous with quadrant III and
[+i, +i∞[, continuous with quadrant I.
The library syntax is GEN gasinh(GEN x, long prec).

3.11.13 atan(x). Principal branch of tan−1 (x) = log((1 + ix)/(1 − ix))/2i. In particular the real
part of atan(x) belongs to ] − π/2, π/2[. The branch cut is in two pieces: ] − i∞, −i[, continuous
with quadrant IV, and ]i, +i∞[ continuous with quadrant II. The function satisfies atan(x) =
−iatanh(ix) for all x 6= ±i.
The library syntax is GEN gatan(GEN x, long prec).

3.11.14 atanh(x). Principal branch of tanh−1 (x) = log((1 + x)/(1 − x))/2. In particular the
imaginary part of atanh(x) belongs to [−π/2, π/2]; if x ∈ R and |x| > 1 then atanh(x) is complex.
The library syntax is GEN gatanh(GEN x, long prec).

3.11.15 besselh1(nu, x). H 1 -Bessel function of index nu and argument x.

The library syntax is GEN hbessel1(GEN nu, GEN x, long prec).

3.11.16 besselh2(nu, x). H 2 -Bessel function of index nu and argument x.

The library syntax is GEN hbessel2(GEN nu, GEN x, long prec).

296
3.11.17 besseli(nu, x). I-Bessel function of index nu and argument x. If x converts to a power
series, the initial factor (x/2)ν /Γ(ν + 1) is omitted (since it cannot be represented in PARI when
ν is not integral).
The library syntax is GEN ibessel(GEN nu, GEN x, long prec).

3.11.18 besselj(nu, x). J-Bessel function of index nu and argument x. If x converts to a power
series, the initial factor (x/2)ν /Γ(ν + 1) is omitted (since it cannot be represented in PARI when
ν is not integral).
The library syntax is GEN jbessel(GEN nu, GEN x, long prec).

3.11.19 besseljh(n, x). J-Bessel function of half integral index. More precisely, besseljh(n, x)
computes Jn+1/2 (x) where n must be of type integer, and x is any element of C. In the present
version 2.13.3, this function is not very accurate when x is small.
The library syntax is GEN jbesselh(GEN n, GEN x, long prec).

3.11.20 besselk(nu, x). K-Bessel function of index nu and argument x.

The library syntax is GEN kbessel(GEN nu, GEN x, long prec).

3.11.21 besseln(nu, x). Deprecated alias for bessely.

The library syntax is GEN nbessel(GEN nu, GEN x, long prec).

3.11.22 bessely(nu, x). Y -Bessel function of index nu and argument x.

The library syntax is GEN nbessel(GEN nu, GEN x, long prec).

3.11.23 cos(x). Cosine of x. Note that, for real x, cosine and sine can be obtained simultaneously
as
cs(x) = my(z = exp(I*x)); [real(z), imag(z)];
and for general complex x as
cs2(x) = my(z = exp(I*x), u = 1/z); [(z+u)/2, (z-u)/2];
Note that the latter function suffers from catastrophic cancellation when z 2 ≈ ±1.
The library syntax is GEN gcos(GEN x, long prec).

3.11.24 cosh(x). Hyperbolic cosine of x.

The library syntax is GEN gcosh(GEN x, long prec).

3.11.25 cotan(x). Cotangent of x.

The library syntax is GEN gcotan(GEN x, long prec).

3.11.26 cotanh(x). Hyperbolic cotangent of x.

The library syntax is GEN gcotanh(GEN x, long prec).

297
3.11.27 dilog(x).P Principal branch of the dilogarithm of x, i.e. analytic continuation of the power
n 2
series log2 (x) = n≥1 x /n .
The library syntax is GEN dilog(GEN x, long prec).
R ∞ −t
3.11.28 eint1(x, {n}). Exponential integral x e t dt = incgam(0, x), where the latter expression
extends the function definition from real x > 0 to all complex x 6= 0.
If n is present, we must have x > 0; the function returns the n-dimensional vector
[eint1(x), . . . , eint1(nx)]. Contrary to other transcendental functions, and to the default case
(n omitted), the values are correct up to a bounded absolute, rather than relative, error 10−n ,
where n is precision(x) if x is a t_REAL and defaults to realprecision otherwise. (In the most
important application, to the computation of L-functions via approximate functional equations,
those values appear as weights in long sums and small individual relative errors are less useful
than controlling the absolute error.) This is faster than repeatedly calling eint1(i * x), but less
precise.
The library syntax is GEN veceint1(GEN x, GEN n = NULL, long prec). Also available is
GEN eint1(GEN x, long prec).

3.11.29 ellE(k). Complete elliptic integral of the second kind

Z π/2
E(k) = (1 − k 2 sin(t)2 )1/2 dt
0

for the complex parameter k using the agm.

The library syntax is GEN ellE(GEN k, long prec).

3.11.30 ellK(k). Complete elliptic integral of the first kind

Z π/2
K(k) = (1 − k 2 sin(t)2 )−1/2 dt
0

for the complex parameter k using the agm.

The library syntax is GEN ellK(GEN k, long prec).
√ R∞ 2
3.11.31 erfc(x). Complementary
√ error function, analytic continuation of (2/ π) x e−t dt =
sign(x)incgam(1/2, x2 )/ π for real x 6= 0. The latter expression extends the function definition
from real x to complex x with positive real part (or zero real part and positive imaginary part).
This is extended to the whole complex plane by the functional equation erfc(−x) = 2 − erfc(x).
? erfc(0)
%1 = 1.0000000000000000000000000000000000000
? erfc(1)
%2 = 0.15729920705028513065877936491739074071
? erfc(1+I)
%3 = -0.31615128169794764488027108024367036903
- 0.19045346923783468628410886196916244244*I
The library syntax is GEN gerfc(GEN x, long prec).

298
 Q∞
3.11.32 eta(z, {flag = 0}). Variants of Dedekind’s η function. If flag = 0, return n=1 (1 − q n ),
where q depends on x in the following way:
• q = e2iπx if x is a complex number (which must then have positive imaginary part); notice
that the factor q 1/24 is missing!
• q = x if x is a t_PADIC, or can be converted to a power series (which must then have positive
valuation).
IfQflag is nonzero, x is converted to a complex number and we return the true η function,
∞
q 1/24 n=1 (1 − q n ), where q = e2iπx .
The library syntax is GEN eta0(GEN z, long flag, long prec).
Also available is GEN trueeta(GEN x, long prec) (flag = 1).

3.11.33 exp(x). Exponential of x. p-adic arguments with positive valuation are accepted.
The library syntax is GEN gexp(GEN x, long prec). For a t_PADIC x, the function GEN
Qp_exp(GEN x) is also available.

3.11.34 expm1(x). Return exp(x) − 1, computed in a way that is also accurate when the real
part of x is near 0. A naive direct computation would suffer from catastrophic cancellation; PARI’s
direct computation of exp(x) alleviates this well known problem at the expense of computing exp(x)
to a higher accuracy when x is small. Using expm1 is recommended instead:
? default(realprecision, 10000); x = 1e-100;
? a = expm1(x);
time = 4 ms.
? b = exp(x)-1;
time = 4 ms.
? default(realprecision, 10040); x = 1e-100;
? c = expm1(x); \\ reference point
? abs(a-c)/c \\ relative error in expm1(x)
%7 = 1.4027986153764843997 E-10019
? abs(b-c)/c \\ relative error in exp(x)-1
%8 = 1.7907031188259675794 E-9919
As the example above shows, when x is near 0, expm1 is more accurate than exp(x)-1.
The library syntax is GEN gexpm1(GEN x, long prec).

3.11.35 gamma(s). For s a complex number, evaluates Euler’s gamma function

Z ∞
Γ(s) = ts−1 exp(−t) dt.
0

Error if s is a nonpositive integer, where Γ has a pole.

For s a t_PADIC, evaluates the Morita gamma function at s, Q0that is the unique continuous
p-adic function on the p-adic integers extending Γp (k) = (−1)k j<k j, where the prime means
that p does not divide j.
? gamma(1/4 + O(5^10))
%1= 1 + 4*5 + 3*5^4 + 5^6 + 5^7 + 4*5^9 + O(5^10)

299
 ? algdep(%,4)
%2 = x^4 + 4*x^2 + 5
The library syntax is GEN ggamma(GEN s, long prec). For a t_PADIC x, the function GEN
Qp_gamma(GEN x) is also available.

3.11.36 gammah(x). Gamma function evaluated at the argument x + 1/2.

The library syntax is GEN ggammah(GEN x, long prec).

3.11.37 gammamellininv(G, t, {m = 0}). Returns the value at t of the inverse Mellin transform
G initialized by gammamellininvinit. If the optional parameter m is present, return the m-th
derivative G(m) (t).
? G = gammamellininvinit([0]);
? gammamellininv(G, 2) - 2*exp(-Pi*2^2)
%2 = -4.484155085839414627 E-44
The shortcut
gammamellininv(A,t,m)
for
gammamellininv(gammamellininvinit(A,m), t)
is available.
The library syntax is GEN gammamellininv(GEN G, GEN t, long m, long bitprec).

3.11.38 gammamellininvasymp(A, n, {m = 0}). Return the first n terms of the asymptotic

expansion at infinity of the m-th derivative K (m) (t) of the inverse Mellin transform of the function

f (s) = ΓR (s + a1 ) . . . ΓR (s + ad ) ,

where A is the vector [a1 , . . . , ad ] and ΓR (s) = π −s/2 Γ(s/2) (Euler’s gamma). The result is a vector
[M [1]...M [n]] with M[1]=1, such that
q
2/d X
K (m) (t) = 2d+1 /dta+m(2/d−1) e−dπt M [n + 1](πt2/d )−n
n≥0

P
with a = (1 − d + 1≤j≤d aj )/d. We also allow A to be the output of gammamellininvinit.
The library syntax is GEN gammamellininvasymp(GEN A, long precdl, long n).

3.11.39 gammamellininvinit(A, {m = 0}). Initialize data for the computation by gam-

mamellininv of the m-th derivative of the inverse Mellin transform of the function

f (s) = ΓR (s + a1 ) . . . ΓR (s + ad )

where A is the vector [a1 , . . . , ad ] and ΓR (s) = π −s/2 Γ(s/2) (Euler’s gamma). This is the special
case of Meijer’s G functions used to compute L-values via the approximate functional equation. By
extension, A is allowed to be an Ldata or an Linit, understood as the inverse Mellin transform of
the L-function gamma factor.

300
Caveat. Contrary to the PARI convention, this function guarantees an absolute (rather than
relative) error bound.
For instance, the inverse Mellin transform of ΓR (s) is 2 exp(−πz 2 ):
? G = gammamellininvinit([0]);
? gammamellininv(G, 2) - 2*exp(-Pi*2^2)
%2 = -4.484155085839414627 E-44
The inverse Mellin transform of ΓR (s + 1) is 2z exp(−πz 2 ), and its second derivative is
4πz exp(−πz 2 )(2πz 2 − 3):
? G = gammamellininvinit([1], 2);
? a(z) = 4*Pi*z*exp(-Pi*z^2)*(2*Pi*z^2-3);
? b(z) = gammamellininv(G,z);
? t(z) = b(z) - a(z);
? t(3/2)
%3 = -1.4693679385278593850 E-39
The library syntax is GEN gammamellininvinit(GEN A, long m, long bitprec).

3.11.40 hypergeom({N }, {D}, z). General hypergeometric function, where N and D are the vector
of parameters in the numerator and denominator respectively, evaluated at the complex argument
z.
This function implements hypergeometric functions
Q
(ai )n z n
Q1≤i≤p
X
p Fq ((ai )1≤i≤p , (bj )1≤j≤q ; z) = ,
n≥0 1≤j≤q (bj )n n!

where (a)n = a(a + 1) · · · (a + n − 1) is the rising Pochammer symbol. For this to make sense, none
of the bj must be a negative or zero integer. The corresponding general GP command is
hypergeom([a1,a2,...,ap], [b1,b2,...,bq], z)
Whenever p = 1 or q = 1, a one-element vector can be replaced by the element it contains.
Whenever p = 0 or q = 0, an empty vector can be omitted. For instance hypergeom(,b,z) computes
0 F1 (; b; z).

We distinguish three kinds of such functions according to their radius of convergence R:

• q ≥ p: R = ∞.
• q = p − 1: R = 1. Nonetheless, by integral representations, p Fq can be analytically continued
outside the disc of convergence.
• q ≤ p − 2: R = 0. By integral representations, one can make sense of the function in a
suitable domain.
The list of implemented functions and their domain of validity in our implementation is as
follows:
F01: hypergeom(,a,z) (or [a]). This is essentially a Bessel function and computed as such.
R = ∞.
F10: hypergeom(a,,z) This is (1 − z)−a .

301
 F11: hypergeom(a,b,z) is the Kummer confluent hypergeometric function, computed by sum-
ming the series. R = ∞
F20: hypergeom([a,b],,z). R = 0, computed as
Z ∞
1
ta−1 (1 − zt)−b e−t dt .
Γ(a) 0

F21: hypergeom([a,b],c,z) (or [c]). R = 1, extended by

Z 1
Γ(c)
tb−1 (1 − t)c−b−1 (1 − zt)a dt .
Γ(b)Γ(c − b) 0

This is Gauss’s Hypergeometric function, and almost all of the implementation work is done for
this function.
F31: hypergeom([a,b,c],d,z) (or [d]). R = 0, computed as
Z ∞
1
ta−1 e−t 2 F1 (b, c; d; tz) dt .
Γ(a) 0

F32: hypergeom([a,b,c],[d,e],z). R = 1, extended by

Z 1
Γ(e)
tc−1 (1 − t)e−c−1 2 F1 (a, b; d; tz) dt .
Γ(c)Γ(e − c) 0

For other inputs: if R = ∞ or R = 1 and |z| < 1−ε is not too close to the circle of convergence,
we simply sum the series.

? hypergeom([3,2], 3.4, 0.7) \\ 2F1(3,2; 3.4; 0.7)

%1 = 7.9999999999999999999999999999999999999
? a=5/3; T1=hypergeom([1,1,1],[a,a],1) \\ 3F2(1,1,1; a,a; 1)
%2 = 3.1958592952314032651578713968927593818
? T2=hypergeom([2,1,1],[a+1,a+1],1)
%3 = 1.6752931349345765309211012564734179541
? T3=hypergeom([2*a-1,1,1],[a+1,a+1],1)
%4 = 1.9721037126267142061807688820853354440
? T1 + (a-1)^2/(a^2*(2*a-3)) * (T2-2*(a-1)*T3) \\
- gamma(a)^2/((2*a-3)*gamma(2*a-2))
%5 = -1.880790961315660013 E-37 \\ ~ 0

This identity is due to Bercu.

The library syntax is GEN hypergeom(GEN N = NULL, GEN D = NULL, GEN z, long prec)
.

302
3.11.41 hyperu(a, b, z). U -confluent hypergeometric function with complex parameters a, b, z.
Note that 2 F0 (a, b, z) = (−z)−a U (a, a + 1 − b, −1/z),
? hyperu(1, 3/2, I)
%1 = 0.23219... - 0.80952...*I
? -I * hypergeom([1, 1+1-3/2], [], -1/I)
%2 = 0.23219... - 0.80952...*I
The library syntax is GEN hyperu(GEN a, GEN b, GEN z, long prec).
R∞
3.11.42 incgam(s, x, {g}). Incomplete gamma function x e−t ts−1 dt, extended by analytic con-
tinuation to all complex x, s not both 0. The relative error is bounded in terms of the precision of
s (the accuracy of x is ignored when determining the output precision). When g is given, assume
that g = Γ(s). For small |x|, this will speed up the computation.
The library syntax is GEN incgam0(GEN s, GEN x, GEN g = NULL, long prec). Also
available is GEN incgam(GEN s, GEN x, long prec).

3.11.43 incgamc(s, x). Complementary incomplete gamma function. The arguments x and s are
complex numbers such that s is not a pole of Γ and |x|/(|s|+1)
Rx is not much larger than 1 (otherwise
the convergence is very slow). The result returned is 0 e−t ts−1 dt.
The library syntax is GEN incgamc(GEN s, GEN x, long prec).

3.11.44 lambertw(y). Lambert W function, solution of the implicit equation xex = y, for a
positive real number y. This is the restriction to the positive reals of the complex principal branch
W0 , which is not implemented outside of R∗+ . Other branches Wk for k 6= 0 are not implemented
either.
The library syntax is GEN glambertW(GEN y, long prec).

3.11.45 lngamma(x). Principal branch of the logarithm of the gamma function of x. This
function is analytic on the complex plane with nonpositive integers removed, and can have much
larger arguments than gamma itself.
For x a power series such that x(0) is not a pole of gamma, compute the Taylor expansion.
(PARI only knows about regular power series and can’t include logarithmic terms.)
? lngamma(1+x+O(x^2))
%1 = -0.57721566490153286060651209008240243104*x + O(x^2)
? lngamma(x+O(x^2))
*** at top-level: lngamma(x+O(x^2))
*** ^-----------------
*** lngamma: domain error in lngamma: valuation != 0
? lngamma(-1+x+O(x^2))
*** lngamma: Warning: normalizing a series with 0 leading term.
*** at top-level: lngamma(-1+x+O(x^2))
*** ^--------------------
*** lngamma: domain error in intformal: residue(series, pole) != 0
The library syntax is GEN glngamma(GEN x, long prec).

303
3.11.46 log(x). Principal branch of the natural logarithm of x ∈ C∗ , i.e. such that =(log(x)) ∈
] − π, π]. The branch cut lies along the negative real axis, continuous with quadrant 2, i.e. such
that limb→0+ log(a + bi) = log a for a ∈ R∗ . The result is complex (with imaginary part equal to
π) if x ∈ R and x < 0. In general, the algorithm uses the formula

π
log(x) ≈ − m log 2,
2agm(1, 4/s)

if s = x2m is large enough. (The result is exact to B bits provided s > 2B/2 .) At low accuracies,
the series expansion near 1 is used.

p-adic arguments are also accepted for x, with the convention that log(p) = 0. Hence in
particular exp(log(x))/x is not in general equal to 1 but to a (p − 1)-th root of unity (or ±1 if
p = 2) times a power of p.

The library syntax is GEN glog(GEN x, long prec). For a t_PADIC x, the function GEN
Qp_log(GEN x) is also available.

3.11.47 log1p(x). Return log(1 + x), computed in a way that is also accurate when the real part
of x is near 0. This is the reciprocal function of expm1(x) = exp(x) − 1.

? default(realprecision, 10000); x = Pi*1e-100;

? (expm1(log1p(x)) - x) / x
%2 = -7.668242895059371866 E-10019
? (log1p(expm1(x)) - x) / x
%3 = -7.668242895059371866 E-10019

When x is small, this function is both faster and more accurate than log(1 + x):

? \p38
? x = 1e-20;
? localprec(100); c = log1p(x); \\ reference point
? a = log1p(x); abs((a - c)/c)
%6 = 0.E-38
? b = log(1+x); abs((b - c)/c) \\ slightly less accurate
%7 = 1.5930919111324522770 E-38
? for (i=1,10^5,log1p(x))
time = 81 ms.
? for (i=1,10^5,log(1+x))
time = 100 ms. \\ slower, too

The library syntax is GEN glog1p(GEN x, long prec).

304
3.11.48 polylog(m, x, {flag = 0}). One of the different polylogarithms, depending on flag:
If flagP= 0 or is omitted: mth polylogarithm of x, i.e. analytic continuation of the power series
Lim (x) = n≥1 xn /nm (x < 1). Uses the functional equation linking the values at x and 1/x to
restrict to the case |x| ≤ 1, then the power series when |x|2 ≤ 1/2, and the power series expansion
in log(x) otherwise.
Using flag, computes a modified mth polylogarithm of x. We use Zagier’s notations; let <m
denote < or = depending on whether m is odd or even:
If flag = 1: compute D̃m (x), defined for |x| ≤ 1 by

m−1
!
X (− log |x|)k (− log |x|)m−1
<m Lim−k (x) + log |1 − x| .
k! m!
k=0

If flag = 2: compute Dm (x), defined for |x| ≤ 1 by

m−1
!
X (− log |x|)k 1 (− log |x|)m
<m Lim−k (x) − .
k! 2 m!
k=0

If flag = 3: compute Pm (x), defined for |x| ≤ 1 by

m−1
!
X 2k Bk 2m−1 Bm
<m (log |x|)k Lim−k (x) − (log |x|)m .
k! m!
k=0

These three functions satisfy the functional equation fm (1/x) = (−1)m−1 fm (x).
The library syntax is GEN polylog0(long m, GEN x, long flag, long prec). Also available
is GEN gpolylog(long m, GEN x, long prec) (flag= 0).

3.11.49 polylogmult(s, {z}, {t = 0}). For s a vector of positive integers and z a vector of complex
numbers of the same length, returns the multiple polylogarithm value (MPV)
X Y
ζ(s1 , . . . , sr ; z1 , . . . , zr ) = zini /nsi i .
n1 >...>nr >0 1≤i≤r

If z is omitted, assume z = [1, . . . , 1], i.e., Multiple Zeta Value. More generally, return Yamamoto’s
interpolation between ordinary multiple polylogarithms (t = 0) and star polylogarithms (t = 1,
using the condition n1 ≥ . . . ≥ nr > 0), evaluated at t.
We must have |z1 · · · zi | ≤ 1 for all i, and if s1 = 1 we must have z1 6= 1.
? 8*polylogmult([2,1],[-1,1]) - zeta(3)
%1 = 0.E-38

305
Warning. The algorithm used converges when the zi are ±1. It may not converge as some zi 6= 1
becomes too close to 1, even at roots of 1 of moderate order:
? polylogmult([2,1], (99+20*I)/101 * [1,1])
*** polylogmult: sorry, polylogmult in this range is not yet implemented.
? polylogmult([2,1], exp(I*Pi/20)* [1,1])
*** polylogmult: sorry, polylogmult in this range is not yet implemented.
More precisely, if yi := 1/(z1 · · · zi ) and

v := min |(1 − yi )yj | > 1/4

i<j;yi 6=1

−b 2
then the algorithm computes the value up
Pto a 2 absolute error in O(k N ) operations on floating
point numbers of O(N ) bits, where k = i si is the weight and N = b/ log2 (4v).
The library syntax is GEN polylogmult_interpolate(GEN s, GEN z = NULL, GEN t =
NULL, long prec). Also available is GEN polylogmult(GEN s, GEN z, long prec) (t is NULL).

3.11.50 psi(x). The ψ-function of x, i.e. the logarithmic derivative Γ0 (x)/Γ(x).

The library syntax is GEN gpsi(GEN x, long prec).

3.11.51 rootsof1(N ). Return the column vector v of all complex N -th roots of 1, where N is a
positive integer. In other words, v[k] = exp(2I(k − 1)π/N ) for k = 1, . . . , N . Rational components
(e.g., the roots ±1 and ±I) are given exactly, not as floating point numbers:
? rootsof1(4)
%1 = [1, I, -1, -I]~
? rootsof1(3)
%2 = [1, -1/2 + 0.866025...*I, -1/2 - 0.866025...*I]~
The library syntax is GEN grootsof1(long N, long prec).

3.11.52 sin(x). Sine of x. Note that, for real x, cosine and sine can be obtained simultaneously as
cs(x) = my(z = exp(I*x)); [real(z), imag(z)];
and for general complex x as
cs2(x) = my(z = exp(I*x), u = 1/z); [(z+u)/2, (z-u)/2];
Note that the latter function suffers from catastrophic cancellation when z 2 ≈ ±1.
The library syntax is GEN gsin(GEN x, long prec).

3.11.53 sinc(x). Cardinal sine of x, i.e. sin(x)/x if x 6= 0, 1 otherwise. Note that this function
also allows to compute
(1 − cos(x))/x2 = sinc(x/2)2 /2
accurately near x = 0.
The library syntax is GEN gsinc(GEN x, long prec).

3.11.54 sinh(x). Hyperbolic sine of x.

The library syntax is GEN gsinh(GEN x, long prec).

306
3.11.55 sqr(x). Square of x. This operation is not completely straightforward, i.e. identical to x∗x,
since it can usually be computed more efficiently (roughly one-half of the elementary multiplications
can be saved). Also, squaring a 2-adic number increases its precision. For example,
? (1 + O(2^4))^2
%1 = 1 + O(2^5)
? (1 + O(2^4)) * (1 + O(2^4))
%2 = 1 + O(2^4)
Note that this function is also called whenever one multiplies two objects which are known to be
identical , e.g. they are the value of the same variable, or we are computing a power.
? x = (1 + O(2^4)); x * x
%3 = 1 + O(2^5)
? (1 + O(2^4))^4
%4 = 1 + O(2^6)
(note the difference between %2 and %3 above).
The library syntax is GEN gsqr(GEN x).
√
3.11.56 sqrt(x). Principal branch of the square root of x, defined as x = exp(log x/2). In
particular, we have Arg(sqrt(x)) ∈ ] − π/2, π/2], and if x ∈ R and x < 0, then the result is complex
with positive imaginary part.
Intmod a prime p, t_PADIC and t_FFELT are allowed as arguments. In the first 2 cases
(t_INTMOD, t_PADIC), the square root (if it exists) which is returned is the one whose first p-adic
digit is in the interval [0, p/2]. For other arguments, the result is undefined.
The library syntax is GEN gsqrt(GEN x, long prec). For a t_PADIC x, the function GEN
Qp_sqrt(GEN x) is also available.

3.11.57 sqrtn(x, n, {&z}). Principal branch of the nth root of x, i.e. such that Arg(sqrtn(x)) ∈
] − π/n, π/n]. Intmod a prime and p-adics are allowed as arguments.
If z is present, it is set to a suitable root of unity allowing to recover all the other roots. If it
was not possible, z is set to zero. In the case this argument is present and no nth root exist, 0 is
returned instead of raising an error.
? sqrtn(Mod(2,7), 2)
%1 = Mod(3, 7)
? sqrtn(Mod(2,7), 2, &z); z
%2 = Mod(6, 7)
? sqrtn(Mod(2,7), 3)
*** at top-level: sqrtn(Mod(2,7),3)
*** ^-----------------
*** sqrtn: nth-root does not exist in gsqrtn.
? sqrtn(Mod(2,7), 3, &z)
%2 = 0
? z
%3 = 0
The following script computes all roots in all possible cases:
sqrtnall(x,n)=

307
 { my(V,r,z,r2);
r = sqrtn(x,n, &z);
if (!z, error("Impossible case in sqrtn"));
if (type(x) == "t_INTMOD" || type(x)=="t_PADIC",
r2 = r*z; n = 1;
while (r2!=r, r2*=z;n++));
V = vector(n); V[1] = r;
for(i=2, n, V[i] = V[i-1]*z);
V
}
addhelp(sqrtnall,"sqrtnall(x,n):compute the vector of nth-roots of x");

The library syntax is GEN gsqrtn(GEN x, GEN n, GEN *z = NULL, long prec). If x is a
t_PADIC, the function GEN Qp_sqrtn(GEN x, GEN n, GEN *z) is also available.

3.11.58 tan(x). Tangent of x.

The library syntax is GEN gtan(GEN x, long prec).

3.11.59 tanh(x). Hyperbolic tangent of x.

The library syntax is GEN gtanh(GEN x, long prec).

3.11.60 teichmuller(x, {tab}). Teichmüller character of the p-adic number x, i.e. the unique
(p − 1)-th root of unity congruent to x/pvp (x) modulo p. If x is of the form [p, n], for a prime p
and integer n, return the lifts to Z of the images of i + O(pn ) for i = 1, . . . , p − 1, i.e. all roots of 1
ordered by residue class modulo p. Such a vector can be fed back to teichmuller, as the optional
argument tab, to speed up later computations.
? z = teichmuller(2 + O(101^5))
%1 = 2 + 83*101 + 18*101^2 + 69*101^3 + 62*101^4 + O(101^5)
? z^100
%2 = 1 + O(101^5)
? T = teichmuller([101, 5]);
? teichmuller(2 + O(101^5), T)
%4 = 2 + 83*101 + 18*101^2 + 69*101^3 + 62*101^4 + O(101^5)
As a rule of thumb, if more than

p / 2(log2 (p) + hammingweight(p))

values of teichmuller are to be computed, then it is worthwile to initialize:

? p = 101; n = 100; T = teichmuller([p,n]); \\ instantaneous
? for(i=1,10^3, vector(p-1, i, teichmuller(i+O(p^n), T)))
time = 60 ms.
? for(i=1,10^3, vector(p-1, i, teichmuller(i+O(p^n))))
time = 1,293 ms.
? 1 + 2*(log(p)/log(2) + hammingweight(p))
%8 = 22.316[...]
Here the precomputation induces a speedup by a factor 1293/60 ≈ 21.5.

308
Caveat. If the accuracy of tab (the argument n above) is lower than the precision of x, the former
is used, i.e. the cached value is not refined to higher accuracy. It the accuracy of tab is larger,
then the precision of x is used:
? Tlow = teichmuller([101, 2]); \\ lower accuracy !
? teichmuller(2 + O(101^5), Tlow)
%10 = 2 + 83*101 + O(101^5) \\ no longer a root of 1
? Thigh = teichmuller([101, 10]); \\ higher accuracy
? teichmuller(2 + O(101^5), Thigh)
%12 = 2 + 83*101 + 18*101^2 + 69*101^3 + 62*101^4 + O(101^5)
The library syntax is GEN teichmuller(GEN x, GEN tab = NULL).
Also available are the functions GEN teich(GEN x) (tab is NULL) as well as GEN teichmul-
lerinit(long p, long n).

3.11.61 theta(q, z). Jacobi sine theta-function

X
θ1 (z, q) = 2q 1/4 (−1)n q n(n+1) sin((2n + 1)z).
n≥0

The library syntax is GEN theta(GEN q, GEN z, long prec).

3.11.62 thetanullk(q, k). k-th derivative at z = 0 of theta(q, z).

The library syntax is GEN thetanullk(GEN q, long k, long prec).
i
d θ
GEN vecthetanullk(GEN q, long k, long prec) returns the vector of all dz i (q, 0) for all odd

i = 1, 3, . . . , 2k − 1. GEN vecthetanullk_tau(GEN tau, long k, long prec) returns vecthetan-

ullk tau at q = exp(2iπtau).

3.11.63 weber(x, {flag = 0}). One of Weber’s three f functions. If flag = 0, returns

f (x) = exp(−iπ/24) · η((x + 1)/2) / η(x) such that j = (f 24 − 16)3 /f 24 ,

where j is the elliptic j-invariant (see the function ellj). If flag = 1, returns

f1 (x) = η(x/2) / η(x) such that j = (f124 + 16)3 /f124 .

Finally, if flag = 2, returns

√
f2 (x) = 2η(2x) / η(x) such that j = (f224 + 16)3 /f224 .
√
Note the identities f 8 = f18 + f28 and f f1 f2 = 2.
The library syntax is GEN weber0(GEN x, long flag, long prec). Also available are GEN
weberf(GEN x, long prec), GEN weberf1(GEN x, long prec) and GEN weberf2(GEN x, long
prec).

309
 −s
P
3.11.64 zeta(s). For s 6= 1 a complex number, Riemann’s zeta function ζ(s) = n≥1 n ,
computed using the Euler-Maclaurin summation formula, except when s is of type integer, in
which case it is computed using Bernoulli numbers for s ≤ 0 or s > 0 and even, and using modular
forms for s > 0 and odd. Power series are also allowed:
? zeta(2) - Pi^2/6
%1 = 0.E-38
? zeta(1+x+O(x^3))
%2 = 1.0000000000000000000000000000000000000*x^-1 + \
0.57721566490153286060651209008240243104 + O(x)
For s 6= 1 a p-adic number, Kubota-Leopoldt zeta function at s, that is the unique continuous
p-adic function on the p-adic integers that interpolates the values of (1 − p−k )ζ(k) at negative
integers k such that k ≡ 1 (mod p − 1) (resp. k is odd) if p is odd (resp. p = 2). Power series are
not allowed in this case.
? zeta(-3+O(5^10))
%1 = 4*5^-1 + 4 + 3*5 + 4*5^3 + 4*5^5 + 4*5^7 + O(5^9)))))
? (1-5^3) * zeta(-3)
%2 = -1.0333333333333333333333333333333333333
? bestappr(%)
%3 = -31/30
? zeta(-3+O(5^10)) - (-31/30)
%4 = O(5^9)
The library syntax is GEN gzeta(GEN s, long prec).

−s
P
3.11.65 zetahurwitz(s, x, {der = 0}). Hurwitz zeta function ζ(s, x) = n≥0 (n + x) and
analytically continued, with s 6= 1 and x not a negative or zero integer. Note that ζ(s, 1) = ζ(s). s
can also be a polynomial, rational function, or power series. If der is positive, compute the der’th
derivative with respect to s. Note that the derivative with respect to x is simply −sζ(s + 1, x).
? zetahurwitz(Pi,Pi)
%1 = 0.056155444497585099925180502385781494484
? zetahurwitz(2,1) - zeta(2)
%2 = -2.350988701644575016 E-38
? zetahurwitz(Pi,3) - (zeta(Pi)-1-1/2^Pi)
%3 = -2.2040519077917890774 E-39
? zetahurwitz(-7/2,1) - zeta(-7/2)
%4 = -2.295887403949780289 E-41
? zetahurwitz(-2.3,Pi+I*log(2))
%5 = -5.1928369229555125820137832704455696057\
- 6.1349660138824147237884128986232049582*I
? zetahurwitz(-1+x^2+O(x^3),1)
%6 = -0.083333333333333333333333333333333333333\
- 0.16542114370045092921391966024278064276*x^2 + O(x^3)
? zetahurwitz(1+x+O(x^4),2)
%7 = 1.0000000000000000000000000000000000000*x^-1\
- 0.42278433509846713939348790991759756896\
+ 0.072815845483676724860586375874901319138*x + O(x^2)
? zetahurwitz(2,1,2) \\ zeta’’(2)

310
 %8 = 1.9892802342989010234208586874215163815
The library syntax is GEN zetahurwitz(GEN s, GEN x, long der, long bitprec).

3.11.66 zetamult(s, {t = 0}). For s a vector of positive integers such that s[1] ≥ 2, returns the
multiple zeta value (MZV)
X
ζ(s1 , . . . , sk ) = n−s
1
1
. . . n−s
k
k

n1 >...>nk >0
P
of length k and weight i si . More generally, return Yamamoto’s t-MZV interpolation evaluated
at t: for t = 0, this is the ordinary MZV; for t = 1, we obtain the MZSV star value, with ≥ instead
of strict inequalities; and of course, for t = 0 x we obtain Yamamoto’s one-variable polynomial.
? zetamult([2,1]) - zeta(3) \\ Euler’s identity
%1 = 0.E-38
? zetamult([2,1], 1) \\ star value
%2 = 2.4041138063191885707994763230228999815
? zetamult([2,1], ’x)
%3 = 1.20205[...]*x + 1.20205[...]
If the bit precision is B, this function runs in time Õ(k(B + k)2 ) if t = 0, and Õ(kB 3 ) otherwise.
In addition to the above format (avec), the function also accepts a binary word format evec
(each si is replaced by si bits, all of them 0 but the last one) giving the MZV representation as an
iterated integral, and an index format (if e is the positive integer attached the evec vector of bits,
the index is the integer e + 2k−2 ). The function zetamultconvert allows to pass from P one format
to the other; the function zetamultall computes simultaneously all MZVs of weight i≤k si up
to n.
The library syntax is GEN zetamult_interpolate(GEN s, GEN t = NULL, long prec). Also
available is GEN zetamult(GEN s, long prec) for t = 0.

3.11.67 zetamultall(k, {flag = 0}). List of all multiple zeta values (MZVs) for weight s1 + . . . + sr
up to k. Binary digits of flag mean : 0 = star values if set; 1 = values up to to duality if set (see
zetamultdual, ignored if star values); 2 = values of weight k if set (else all values up to weight k);
3 = return the 2-component vector [Z, M], where M is the vector of the corresponding indices m,
i.e., such that zetamult(M[i]) = Z[i]. Note that it is necessary to use zetamultconvert to have
the corresponding avec (s1 , . . . , sr ).
With default flag flag = 0, the function returns a vector with 2k−1 − 1 components whose i-th
entry is the MZV of index i (see zetamult). If the bit precision is B, this function runs in time
O(2k kB 2 ) for an output of size O(2k B).
? Z = zetamultall(5); #Z \\ 2^4 - 1 MZVs of weight <= 5
%1 = 15
? Z[10]
%2 = 0.22881039760335375976874614894168879193
? zetamultconvert(10)
%3 = Vecsmall([3, 2]) \\ index 10 corresponds to ζ(3, 2)
? zetamult(%) \\ double check
%4 = 0.22881039760335375976874614894168879193
? zetamult(10) \\ we can use the index directly

311
 %5 = 0.22881039760335375976874614894168879193
If we use flag bits 1 and 2, we avoid unnecessary computations and copying, saving a potential
factor 4: half the values are in lower weight and computing up to duality save another rough factor
2. Unfortunately, the indexing now no longer corresponds to the new shorter vector of MZVs:
? Z = zetamultall(5, 2); #Z \\ up to duality
%6 = 9
? Z = zetamultall(5, 2); #Z \\ only weight 5
%7 = 8
? Z = zetamultall(5, 2 + 4); #Z \\ both
%8 = 4
So how to recover the value attached to index 10 ? Flag bit 3 returns the actual indices used:
? [Z, M] = zetamultall(5, 2 + 8); M \\ other indices were not included
%9 = Vecsmall([1, 2, 4, 5, 6, 8, 9, 10, 12])
? Z[8] \\ index m = 10 is now in M[8]
%10 = 0.22881039760335375976874614894168879193
? [Z, M] = zetamultall(5, 2 + 4 + 8); M
%11 = Vecsmall([8, 9, 10, 12])
? Z[3] \\ index m = 10 is now in M[3]
%12 = 0.22881039760335375976874614894168879193
The following construction automates the above programmatically, looking up the MZVs of index
10 (= ζ(3, 2)) in all cases, without inspecting the various index sets M visually:
? Z[vecsearch(M, 10)] \\ works in all the above settings
%13 = 0.22881039760335375976874614894168879193
The library syntax is GEN zetamultall(long k, long flag, long prec).

3.11.68 zetamultconvert(a, {fl = 1}). a being either an evec, avec, or index m, converts into
evec (fl=0), avec (fl=1), or index m (fl=2).
? zetamultconvert(10)
%1 = Vecsmall([3, 2])
? zetamultconvert(13)
%2 = Vecsmall([2, 2, 1])
? zetamultconvert(10, 0)
%3 = Vecsmall([0, 0, 1, 0, 1])
? zetamultconvert(13, 0)
%4 = Vecsmall([0, 1, 0, 1, 1])
The last two lines imply that [3, 2] and [2, 2, 1] are dual (reverse order of bits and swap 0 and 1 in
evec form). Hence they have the same zeta value:
? zetamult([3,2])
%5 = 0.22881039760335375976874614894168879193
? zetamult([2,2,1])
%6 = 0.22881039760335375976874614894168879193
The library syntax is GEN zetamultconvert(GEN a, long fl).

312
3.11.69 zetamultdual(s). s being either an evec, avec, or index m, return the dual sequence
in avec format. The dual of a sequence of length r and weight k has length k − r and weight k.
Duality is an involution and zeta values attached to dual sequences are the same:
? zetamultdual([4])
%1 = Vecsmall([2, 1, 1])
? zetamultdual(%)
%2 = Vecsmall([4])
? zetamult(%1) - zetamult(%2)
%3 = 0.E-38
In evec form, duality simply reverses the order of bits and swaps 0 and 1:
? zetamultconvert([4], 0)
%4 = Vecsmall([0, 0, 0, 1])
? zetamultconvert([2,1,1], 0)
%5 = Vecsmall([0, 1, 1, 1])
The library syntax is GEN zetamultdual(GEN s).

3.12 Sums, products, integrals and similar functions.

Although the gp calculator is programmable, it is useful to have a number of preprogrammed
loops, including sums, products, and a certain number of recursions. Also, a number of functions
from numerical analysis like numerical integration and summation of series will be described here.
One of the parameters in these loops must be the control variable, hence a simple variable name.
In the descriptions, the letter X will always denote any simple variable name, and represents the
formal parameter used in the function. The expression to be summed, integrated, etc. is any legal
PARI expression, including of course expressions using loops.

Library mode. Since it is easier to program directly the loops in library mode, these functions
are mainly useful for GP programming. On the other hand, numerical routines code a function (to
be integrated, summed, etc.) with two parameters named
GEN (*eval)(void*,GEN)
void *E; \\ context: eval(E, x) must evaluate your function at x.
see the Libpari manual for details.

Numerical integration. Starting with version 2.2.9 the “double exponential” univariate integra-
tion method is implemented in intnum and its variants. Romberg integration is still available under
the name intnumromb, but superseded. It is possible to compute numerically integrals to thousands
of decimal places in reasonable time, as long as the integrand is regular. It is also reasonable to
compute numerically integrals in several variables, although more than two becomes lengthy. The
integration domain may be noncompact, and the integrand may have reasonable singularities at
endpoints. To use intnum, you must split the integral into a sum of subintegrals where the function
has no singularities except at the endpoints. Polynomials in logarithms are not considered singular,
and neglecting these logs, singularities are assumed to be algebraic (asymptotic to C(x − a)−α for
some α > −1 when x is close to a), or to correspond to simple discontinuities of some (higher)
derivative of the function. For instance, the point 0 is a singularity of abs(x).
See also the discrete summation methods below, sharing the prefix sum.

313
3.12.1 asympnum(expr , {alpha = 1}). Asymptotic expansion of expr , corresponding to a se-
quence u(n), assuming it has the shape
X
u(n) ≈ ai n−iα
i≥0

with rational coefficients ai with reasonable height; the algorithm is heuristic and performs repeated
calls to limitnum, with alpha as in limitnum. As in limitnum, u(n) may be given either by a closure
n 7→ u(n) or as a closure N 7→ [u(1), . . . , u(N )], the latter being often more efficient.
? f(n) = n! / (n^n*exp(-n)*sqrt(n));
? asympnum(f)
%2 = [] \\ failure !
? localprec(57); l = limitnum(f)
%3 = 2.5066282746310005024157652848110452530
? asympnum(n->f(n)/l) \\ normalize
%4 = [1, 1/12, 1/288, -139/51840, -571/2488320, 163879/209018880,
5246819/75246796800]
and we indeed get a few terms of Stirling’s expansion. Note that it definitely helps to normalize
with a limit computed to higher accuracy (as a rule of thumb, multiply the bit accuracy by 1.612):
? l = limitnum(f)
? asympnum(n->f(n) / l) \\ failure again !!!
%6 = []
We treat again the example of the Motzkin numbers Mn given in limitnum:
\\ [M_k, M_{k*2}, ..., M_{k*N}] / (3^n / n^(3/2))
? vM(N, k = 1) =
{ my(q = k*N, V);
if (q == 1, return ([1/3]));
V = vector(q); V[1] = V[2] = 1;
for(n = 2, q - 1,
V[n+1] = ((2*n + 1)*V[n] + 3*(n - 1)*V[n-1]) / (n + 2));
f = (n -> 3^n / n^(3/2));
return (vector(N, n, V[n*k] / f(n*k)));
}
? localprec(100); l = limitnum(n->vM(n,10)); \\ 3/sqrt(12*Pi)
? \p38
? asympnum(n->vM(n,10)/l)
%2 = [1, -3/32, 101/10240, -1617/1638400, 505659/5242880000, ...]
If alpha is not a rational number, loss of accuracy is expected, so it should be precomputed
to double accuracy, say:
? \p38
? asympnum(n->log(1+1/n^Pi),Pi)
%1 = [0, 1, -1/2, 1/3, -1/4, 1/5]
? localprec(76); a = Pi;
? asympnum(n->log(1+1/n^Pi), a) \\ more terms
%3 = [0, 1, -1/2, 1/3, -1/4, 1/5, -1/6, 1/7, -1/8, 1/9, -1/10, 1/11, -1/12]
? asympnum(n->log(1+1/sqrt(n)),1/2) \\ many more terms

314
 %4 = 49
The expression is evaluated for n = 1, 2, . . . , N for an N = O(B) if the current bit accuracy is
B. If it is not defined for one of these values, translate or rescale accordingly:
? asympnum(n->log(1-1/n)) \\ can’t evaluate at n = 1 !
*** at top-level: asympnum(n->log(1-1/n))
*** ^-----------------------
*** in function asympnum: log(1-1/n)
*** ^----------
*** log: domain error in log: argument = 0
? asympnum(n->-log(1-1/(2*n)))
%5 = [0, 1/2, 1/8, 1/24, ...]
? asympnum(n->-log(1-1/(n+1)))
%6 = [0, 1, -1/2, 1/3, -1/4, ...]

The library syntax is asympnum(void *E, GEN (*u)(void *,GEN,long), GEN alpha,
long prec), where u(E, n, prec) must return either u(n) or [u(1), . . . , u(n)] in precision prec.
Also available is GEN asympnum0(GEN u, GEN alpha, long prec), where u is a closure as above
or a vector of sufficient length.

3.12.2 asympnumraw(expr , N, {alpha = 1}). Return the N + 1 first terms of asymptotic ex-
pansion of expr , corresponding to a sequence u(n), as floating point numbers. Assume that the
expansion has the shape X
u(n) ≈ ai n−iα
i≥0

and return approximation of [a0 , a1 , . . . , aN ]. The algorithm is heuristic and performs repeated calls
to limitnum, with alpha as in limitnum. As in limitnum, u(n) may be given either by a closure
n 7→ u(n) or as a closure N 7→ [u(1), . . . , u(N )], the latter being often more efficient. This function
is related to, but more flexible than, asympnum, which requires rational asymptotic expansions.
? f(n) = n! / (n^n*exp(-n)*sqrt(n));
? asympnum(f)
%2 = [] \\ failure !
? v = asympnumraw(f, 10);
? v[1] - sqrt(2*Pi)
%4 = 0.E-37
? bestappr(v / v[1], 2^60)
%5 = [1, 1/12, 1/288, -139/51840, -571/2488320, 163879/209018880,...]
and we indeed get a few terms of Stirling’s expansion (the first 9 terms are correct). If u(n) has an
asymptotic expansion in n−α with α not an integer, the default alpha = 1 is inaccurate:
? f(n) = (1+1/n^(7/2))^(n^(7/2));
? v1 = asympnumraw(f,10);
? v1[1] - exp(1)
%8 = 4.62... E-12
? v2 = asympnumraw(f,10,7/2);
? v2[1] - exp(1)
%7 0.E-37

315
As in asympnum, if alpha is not a rational number, loss of accuracy is expected, so it should be
precomputed to double accuracy, say.
The library syntax is asympnumraw(void *E, GEN (*u)(void *,GEN,long), long N,
GEN alpha, long prec), where u(E, n, prec) must return either u(n) or [u(1), . . . , u(n)] in pre-
cision prec. Also available is GEN asympnumraw0(GEN u, GEN alpha, long prec) where u is
either a closure as above or a vector of sufficient length.

3.12.3 contfraceval(CF , t, {lim = −1}). Given a continued fraction CF output by contfracinit,

evaluate the first lim terms of the continued fraction at t (all terms if lim is negative or omitted;
if positive, lim must be less than or equal to the length of CF.
The library syntax is GEN contfraceval(GEN CF, GEN t, long lim).
P
3.12.4 contfracinit(M, {lim = −1}). Given M representing the power series S = n≥0 M [n +
1]z n , transform it into a continued fraction in Euler form, using the quotient-difference algorithm;
restrict to n ≤ lim if latter is nonnegative. M can be a vector, a power series, a polynomial; if
the limiting parameter lim is present, a rational function is also allowed (and converted to a power
series of that accuracy).
The result is a 2-component vector [A, B] such that S = M [1]/(1 + A[1]z + B[1]z 2 /(1 + A[2]z +
B[2]z 2 /(1 + . . . 1/(1 + A[lim/2]z)))). Does not work if any coefficient of M vanishes, nor for series
for which certain partial denominators vanish.
The library syntax is GEN contfracinit(GEN M, long lim). Also available is GEN
quodif(GEN M, long n) which returns the standard continued fraction, as a vector C such that
S = c[1]/(1 + c[2]z/(1 + c[3]z/(1 + . . . ...c[lim]z))).

3.12.5 derivnum(X = a, expr , {ind = 1}). Numerical derivation of expr with respect to X at
X = a. The order of derivation is 1 by default.
? derivnum(x=0, sin(exp(x))) - cos(1)
%1 = 0.E-38
A clumsier approach, which would not work in library mode, is
? f(x) = sin(exp(x))
? f’(0) - cos(1)
%2 = 0.E-38
• When a is a numerical type (integer, rational number, real number or t_COMPLEX of such),
performs numerical derivation.
• When a is a (polynomial, rational function or) power series, compute derivnum(t=a,f) as
f 0 (a) = (f (a))0 /a0 :
? derivnum(x = 1 + t, sqrt(x))
%1 = 1/2 - 1/4*t + 3/16*t^2 - 5/32*t^3 + ... + O(t^16)
? derivnum(x = 1/(1 + t), sqrt(x))
%2 = 1/2 + 1/4*t - 1/16*t^2 + 1/32*t^3 + ... + O(t^16)
? derivnum(x = 1 + t + O(t^17), sqrt(x))
%3 = 1/2 - 1/4*t + 3/16*t^2 - 5/32*t^3 + ... + O(t^16)
If the parameter ind is present, it can be

316
 • a nonnegative integer m, in which case we return f (m) (x);

• or a vector of orders, in which case we return the vector of derivatives.

? derivnum(x = 0, exp(sin(x)), 16) \\ 16-th derivative

%1 = -52635599.000000000000000000000000000000
? round( derivnum(x = 0, exp(sin(x)), [0..13]) ) \\ 0-13-th derivatives
%2 = [1, 1, 1, 0, -3, -8, -3, 56, 217, 64, -2951, -12672, 5973, 309376]

The library syntax is derivfunk(void *E, GEN (*eval)(void*,GEN), GEN a, GEN ind,
long prec). Also available is GEN derivfun(void *E, GEN (*eval)(void *, GEN), GEN a,
long prec). If a is a numerical type (t_INT, t_FRAC, t_REAL or t_COMPLEX of such, we have GEN
derivnumk(void *E, GEN (*eval)(void *, GEN, long), GEN a, GEN ind, long prec)
and GEN derivnum(void *E, GEN (*eval)(void *, GEN, long prec), GEN a, long prec)

3.12.6 intcirc(X = a, R, expr , {tab}). Numerical integration of (2iπ)−1 expr with respect to X on
the circle |X − a| = R. In other words, when expr is a meromorphic function, sum of the residues
in the corresponding disk; tab is as in intnum, except that if computed with intnuminit it should
be with the endpoints [-1, 1].

? \p105
? intcirc(s=1, 0.5, zeta(s)) - 1
time = 496 ms.
%1 = 1.2883911040127271720 E-101 + 0.E-118*I

The library syntax is intcirc(void *E, GEN (*eval)(void*,GEN), GEN a,GEN R,GEN tab,
long prec).

3.12.7 intfuncinit(t = a, b, f, {m = 0}). Initialize tables for use with integral transforms (such as
Fourier, Laplace or Mellin transforms) in order to compute

Z b
f (t)k(t, z) dt
a

for some kernel k(t, z). The endpoints a and b are coded as in intnum, f is the function to which
the integral transform is to be applied and the nonnegative integer m is as in intnum: multiply
the number of sampling points roughly by 2m , hopefully increasing the accuracy. This function is
particularly useful when the function f is hard to compute, such as a gamma product.

Limitation. The endpoints a and b must be at infinity, with the same asymptotic behavior.
Oscillating types are not supported. This is easily overcome by integrating vectors of functions, see
example below.

317
Examples.
• numerical Fourier transform
Z +∞
F (z) = f (t)e−2iπzt dt.
−∞

First the easy case, assume that f decrease exponentially:

f(t) = exp(-t^2);
A = [-oo,1];
B = [+oo,1];
\p200
T = intfuncinit(t = A,B , f(t));
F(z) =
{ my(a = -2*I*Pi*z);
intnum(t = A,B, exp(a*t), T);
}
? F(1) - sqrt(Pi)*exp(-Pi^2)
%1 = -1.3... E-212
Now the harder case, f decrease slowly: we must specify the oscillating behavior. Thus, we cannot
precompute usefully since everything depends on the point we evaluate at:
f(t) = 1 / (1+ abs(t));
\p200
\\ Fourier cosine transform
FC(z) =
{ my(a = 2*Pi*z);
intnum(t = [-oo, a*I], [+oo, a*I], cos(a*t)*f(t));
}
FC(1)
• Fourier coefficients: we must integrate over a period, but intfuncinit does not support
finite endpoints. The solution is to integrate a vector of functions !
FourierSin(f, T, k) = \\ first k sine Fourier coeffs
{
my (w = 2*Pi/T);
my (v = vector(k+1));
intnum(t = -T/2, T/2,
my (z = exp(I*w*t));
v[1] = z;
for (j = 2, k, v[j] = v[j-1]*z);
f(t) * imag(v)) * 2/T;
}
FourierSin(t->sin(2*t), 2*Pi, 10)
The same technique can be used instead of intfuncinit to integrate f (t)k(t, z) whenever the list
of z-values is known beforehand.
Note that the above code includes an unrelated optimization: the sin(jwt) are computed as
imaginary parts of exp(ijwt) and the latter by successive multiplications.

318
 • numerical Mellin inversion
Z c+i∞ Z +∞
−1 −s −1
F (z) = (2iπ) f (s)z ds = (2π) f (c + it)e− log z(c+it) dt.
c−i∞ −∞

We take c = 2 in the program below:

f(s) = gamma(s)^3; \\ f(c+it) decrease as exp(-3Pi|t|/2)

c = 2; \\ arbitrary
A = [-oo,3*Pi/2];
B = [+oo,3*Pi/2];
T = intfuncinit(t=A,B, f(c + I*t));
F(z) =
{ my (a = -log(z));
intnum(t=A,B, exp(a*I*t), T)*exp(a*c) / (2*Pi);
}

The library syntax is intfuncinit(void *E, GEN (*eval)(void*,GEN), GEN a,GEN b,long
m, long prec).

3.12.8 intnum(X = a, b, expr , {tab}). Numerical integration of expr on ]a, b[ with respect to X,
using the double-exponential method, and thus O(D log D) evaluation of the integrand in precision
D. The integrand may have values belonging to a vector space over the real numbers; in particular,
it can be complex-valued or vector-valued. But it is assumed that the function is regular on ]a, b[.
If the endpoints a and b are finite and the function is regular there, the situation is simple:

? intnum(x = 0,1, x^2)

%1 = 0.3333333333333333333333333333
? intnum(x = 0,Pi/2, [cos(x), sin(x)])
%2 = [1.000000000000000000000000000, 1.000000000000000000000000000]

An endpoint equal to ±∞ is coded as +oo or -oo, as expected:

? intnum(x = 1,+oo, 1/x^2)

%3 = 1.000000000000000000000000000

In basic usage, it is assumed that the function does not decrease exponentially fast at infinity:

? intnum(x=0,+oo, exp(-x))
*** at top-level: intnum(x=0,+oo,exp(-
*** ^--------------------
*** exp: overflow in expo().

We shall see in a moment how to avoid that last problem, after describing the last optional argument
tab.

319
The tab argument. The routine uses weights wi , which are mostly independent of the P function
being integrated, evaluated at many sampling points xi and approximates the integral by wi f (xi ).
If tab is
• a nonnegative integer m, we multiply the number of sampling points by 2m , hopefully
increasing accuracy. Note that the running time increases roughly by a factor 2m . One may try
consecutive values of m until they give the same value up to an accepted error.
• a set of integration tables containing precomputed xi and wi as output by intnuminit. This
is useful if several integrations of the same type are performed (on the same kind of interval and
functions, for a given accuracy): we skip a precomputation of O(D log D) elementary functions in
accuracy D, whose running time has the same order of magnitude as the evaluation of the integrand.
This is in particular useful for multivariate integrals.

Specifying the behavior at endpoints. This is done as follows. An endpoint a is either given
as such (a scalar, real or complex, oo or -oo for ±∞), or as a two component vector [a, α], to
indicate the behavior of the integrand in a neighborhood of a.
If a is finite, the code [a, α] means the function has a singularity of the form (x − a)α , up to
logarithms. (If α ≥ 0, we only assume the function is regular, which is the default assumption.) If
a wrong singularity exponent is used, the result will lose decimals:
? c = -9/10;
? intnum(x=0, 1, x^c) \\ assume x−9/10 is regular at 0
%1 = 9.9999839078827082322596783301939063944
? intnum(x=[0,c], 1, x^c) \\ no, it’s not
%2 = 10.000000000000000000000000000000000000
? intnum(x=[0,c/2], 1, x^c) \\ using a wrong exponent is bad
%3 = 9.9999999997122749095442279375719919769
If a is ±∞, which is coded as +oo or -oo, the situation is more complicated, and [±oo, α]
means:
• α = 0 (or no α at all, i.e. simply ±oo) assumes that the integrand tends to zero moderately
quickly, at least as O(x−2 ) but not exponentially fast.
• α > 0 assumes that the function tends to zero exponentially fast approximately as exp(−α|x|).
This includes oscillating but quickly decreasing functions such as exp(−x) sin(x).
? intnum(x=0, +oo, exp(-2*x))
*** at top-level: intnum(x=0,+oo,exp(-
*** ^--------------------
*** exp: exponent (expo) overflow
? intnum(x=0, [+oo, 2], exp(-2*x)) \\ OK!
%1 = 0.50000000000000000000000000000000000000
? intnum(x=0, [+oo, 3], exp(-2*x)) \\ imprecise exponent, still OK !
%2 = 0.50000000000000000000000000000000000000
? intnum(x=0, [+oo, 10], exp(-2*x)) \\ wrong exponent ⇒ disaster
%3 = 0.49999999999952372962457451698256707393
As the last exemple shows, the exponential decrease rate must be indicated to avoid overflow, but
the method is robust enough for a rough guess to be acceptable.

320
 • α < −1 assumes that the function tends to 0 slowly, like xα . Here the algorithm is less robust
and it is essential to give a sharp α, unless α ≤ −2 in which case we use the default algorithm as
if α were missing (or equal to 0).
? intnum(x=1, +oo, x^(-3/2)) \\ default
%1 = 1.9999999999999999999999999999646391207
? intnum(x=1, [+oo,-3/2], x^(-3/2)) \\ precise decrease rate
%2 = 2.0000000000000000000000000000000000000
? intnum(x=1, [+oo,-11/10], x^(-3/2)) \\ worse than default
%3 = 2.0000000000000000000000000089298011973

The last two codes are reserved for oscillating functions. Let k > 0 real, and g(x) a nonoscil-
lating function tending slowly to 0 (e.g. like a negative power of x), then
• α = k ∗ I assumes that the function behaves like cos(kx)g(x).
• α = −k ∗ I assumes that the function behaves like sin(kx)g(x).
Here it is critical to give the exact value of k. If the oscillating part is not a pure sine or cosine, one
must expand it into a Fourier series, use the above codings, and sum the resulting contributions.
Otherwise you will get nonsense. Note that cos(kx), and similarly sin(kx), means that very function,
and not a translated version such as cos(kx + a).

Note. If f (x) = cos(kx)g(x) where g(x) tends to zero exponentially fast as exp(−αx), it is up to the
user to choose between [±oo, α] and [±oo, k ∗ I], but a good rule of thumb is that if the oscillations
are weaker than the exponential decrease, choose [±oo, α], otherwise choose [±oo, k ∗ I], although
the latter can reasonably be used in all cases, while the former cannot. To take a specific example,
in most inverse Mellin transforms, the integrand is a product of an exponentially decreasing and an
oscillating factor. If we choose the oscillating type of integral we perhaps obtain the best results,
at the expense of having to recompute our functions for a different value of the variable z giving
the transform, preventing us to use a function such as intfuncinit. On the other hand using the
exponential type of integral, we obtain less accurate results, but we skip expensive recomputations.
See intfuncinit for more explanations.

Power series limits. The limits a and b can be power series of nonnegative valuation, giving a
power series expansion for the integral – provided it exists.
? intnum(t=0,X + O(X^3), exp(t))
%4 = 1.000...*X - 0.5000...*X^2 + O(X^3)
? bestappr( intnum(t=0,X + O(X^17), exp(t)) )- exp(X) + 1
%5 = O(X^17)
R 1/X
The valuation of the limit cannot be negative since 0 (1 + t2 )−1 dt = π/2 − sign(X) + O(X 2 ).
Polynomials and rational functions are also allowed and converted to power series using current
seriesprecision:
? bestappr( intnum(t=1,1+X, 1/t) )
%6 = X - 1/2*X^2 + 1/3*X^3 - 1/4*X^4 + [...] + 1/15*X^15 + O(X^16)
The function does not work if the integral is singular with the constant coefficient of the series as
limit:
? intnum(t=X^2+O(X^4),1, 1/sqrt(t))
%8 = 2.000... - 6.236608109630992528 E28*X^2 + O(X^4)

321
however you can use

? intnum(t=[X^2+O(X^4),-1/2],1, 1/sqrt(t))
%10 = 2.000000000000000000000000000-2.000000000000000000000000000*X^2+O(X^4)
whis is translated internally to

? intnum(t=[0,-1/2],1, 1/sqrt(t))-intnum(t=[0,-1/2],X^2+O(X^4), 1/sqrt(t))

For this form the argument tab can be used only as an integer, not a table precomputed by
intnuminit.

We shall now see many examples to get a feeling for what the various parameters achieve. All
examples below assume precision is set to 115 decimal digits. We first type

? \p 115

Apparent singularities. In many cases,R apparent singularities can be ignored. For instance, if
∞
f (x) = 1/(exp(x) − 1) − exp(−x)/x, then 0 f (x) dx = γ, Euler’s constant Euler. But

? f(x) = 1/(exp(x)-1) - exp(-x)/x

? intnum(x = 0, [oo,1], f(x)) - Euler
%1 = 0.E-115
But close to 0 the function f is computed with an enormous loss of accuracy, and we are in fact
lucky that it get multiplied by weights which are sufficiently close to 0 to hide this:

? f(1e-200)
%2 = -3.885337784451458142 E84
A more robust solution is to define the function differently near special points, e.g. by a Taylor
expansion

? F = truncate( f(t + O(t^10)) ); \\ expansion around t = 0

? poldegree(F)
%4 = 7
? g(x) = if (x > 1e-18, f(x), subst(F,t,x)); \\ note that 7 · 18 > 105
? intnum(x = 0, [oo,1], g(x)) - Euler
%2 = 0.E-115
It is up to the user to determine constants such as the 10−18 and 10 used above.

True singularities. With true singularities the result is worse. For instance

? intnum(x = 0, 1, x^(-1/2)) - 2
%1 = -3.5... E-68 \\ only 68 correct decimals
? intnum(x = [0,-1/2], 1, x^(-1/2)) - 2
%2 = 0.E-114 \\ better

322
Oscillating functions.

? intnum(x = 0, oo, sin(x) / x) - Pi/2

%1 = 16.19.. \\ nonsense
? intnum(x = 0, [oo,1], sin(x)/x) - Pi/2
%2 = -0.006.. \\ bad
? intnum(x = 0, [oo,-I], sin(x)/x) - Pi/2
%3 = 0.E-115 \\ perfect
? intnum(x = 0, [oo,-I], sin(2*x)/x) - Pi/2 \\ oops, wrong k
%4 = 0.06...
? intnum(x = 0, [oo,-2*I], sin(2*x)/x) - Pi/2
%5 = 0.E-115 \\ perfect
? intnum(x = 0, [oo,-I], sin(x)^3/x) - Pi/4
%6 = -0.0008... \\ bad
? sin(x)^3 - (3*sin(x)-sin(3*x))/4
%7 = O(x^17)
We may use the above linearization and compute two oscillating integrals with endpoints [oo, -I]
and [oo, R-3*I] respectively, or notice the obvious change of variable, and reduce to the single
∞
integral 21 0 sin(x)/x dx. We finish with some more complicated examples:

? intnum(x = 0, [oo,-I], (1-cos(x))/x^2) - Pi/2

%1 = -0.0003... \\ bad
? intnum(x = 0, 1, (1-cos(x))/x^2) \
+ intnum(x = 1, oo, 1/x^2) - intnum(x = 1, [oo,I], cos(x)/x^2) - Pi/2
%2 = 0.E-115 \\ perfect
? intnum(x = 0, [oo, 1], sin(x)^3*exp(-x)) - 0.3
%3 = -7.34... E-55 \\ bad
? intnum(x = 0, [oo,-I], sin(x)^3*exp(-x)) - 0.3
%4 = 8.9... E-103 \\ better. Try higher m
? tab = intnuminit(0,[oo,-I], 1); \\ double number of sampling points
? intnum(x = 0, oo, sin(x)^3*exp(-x), tab) - 0.3
%6 = 0.E-115 \\ perfect

Warning. Like sumalt, intnum often assigns a reasonable value to diverging integrals. Use these
values at your own risk! For example:

? intnum(x = 0, [oo, -I], x^2*sin(x))

%1 = -2.0000000000...
Note the formula Z ∞
sin(x)/xs dx = cos(πs/2)Γ(1 − s) ,
0

a priori valid only for 0 < <(s) < 2, but the right hand side provides an analytic continuation
which may be evaluated at s = −2. . .

323
Multivariate integration. Using successive univariate integration with respect to different formal
parameters, it is immediate to do naive multivariate integration. But it is important to use a suitable
intnuminit to precompute data for the internal integrations at least!
For example, to compute the double integral on the unit disc x2 + y 2 ≤ 1 of the function
x + y 2 , we can write
2

? tab = intnuminit(-1,1);
? intnum(x=-1,1, intnum(y=-sqrt(1-x^2),sqrt(1-x^2), x^2+y^2, tab),tab) - Pi/2
%2 = -7.1... E-115 \\ OK

The first tab is essential, the second optional. Compare:

? tab = intnuminit(-1,1);
time = 4 ms.
? intnum(x=-1,1, intnum(y=-sqrt(1-x^2),sqrt(1-x^2), x^2+y^2));
time = 3,092 ms. \\ slow
? intnum(x=-1,1, intnum(y=-sqrt(1-x^2),sqrt(1-x^2), x^2+y^2, tab), tab);
time = 252 ms. \\ faster
? intnum(x=-1,1, intnum(y=-sqrt(1-x^2),sqrt(1-x^2), x^2+y^2, tab));
time = 261 ms. \\ the internal integral matters most
The library syntax is intnum(void *E, GEN (*eval)(void*,GEN), GEN a,GEN b,GEN tab,
long prec), where an omitted tab is coded as NULL.

3.12.9 intnumgauss(X = a, b, expr , {tab}). Numerical integration of expr on the compact in-
terval [a, b] with respect to X using Gauss-Legendre quadrature; tab is either omitted or precom-
puted with intnumgaussinit. As a convenience, it can be an integer n in which case we call
intnumgaussinit(n) and use n-point quadrature.
? test(n, b = 1) = T=intnumgaussinit(n);\
intnumgauss(x=-b,b, 1/(1+x^2),T) - 2*atan(b);
? test(0) \\ default
%1 = -9.490148553624725335 E-22
? test(40)
%2 = -6.186629001816965717 E-31
? test(50)
%3 = -1.1754943508222875080 E-38
? test(50, 2) \\ double interval length
%4 = -4.891779568527713636 E-21
? test(90, 2) \\ n must almost be doubled as well!
%5 = -9.403954806578300064 E-38
On the other hand, we recommend to split the integral and change variables rather than increasing
n too much:
? f(x) = 1/(1+x^2);
? b = 100;
? intnumgauss(x=0,1, f(x)) + intnumgauss(x=1,1/b, f(1/x)*(-1/x^2)) - atan(b)
%3 = -1.0579449157400587572 E-37
The library syntax is GEN intnumgauss0(GEN X, GEN b, GEN expr, GEN tab = NULL, long
prec).

324
3.12.10 intnumgaussinit({n}). Initialize tables for n-point Gauss-Legendre integration of a
smooth function f on a compact interval [a, b]. If n is omitted, make a default choice n ≈ B/4,
where B is realbitprecision, suitable for analytic functions on [−1, 1]. The error is bounded by
(b − a)2n+1 (n!)4 f (2n)
(ξ), a < ξ < b.
(2n + 1)!(2n)! (2n)!
If r denotes the distance of the nearest pole to the interval [a, b], then this is of the order of
((b − a)/(4r))2n . In particular, the integral must be subdivided if the interval length b − a becomes
close to 4r. The default choice n ≈ B/4 makes this quantity of order 2−B when b − a = r, as is the
case when integrating 1/(1 + t) on [0, 1] for instance. If the interval length increases, n should be
increased as well.
Specifically, the function returns a pair of vectors [x, w], where x contains the nonnegative
roots of the n-th Legendre polynomial Pn and w the corresponding Gaussian integration weights
Qn (xj )/Pn0 (xj ) = 2/((1 − x2j )Pn0 (xj ))2 such that
Z 1
f (t) dt ≈ wj f (xj ) .
−1

? T = intnumgaussinit();
? intnumgauss(t=-1,1,exp(t), T) - exp(1)+exp(-1)
%1 = -5.877471754111437540 E-39
? intnumgauss(t=-10,10,exp(t), T) - exp(10)+exp(-10)
%2 = -8.358367809712546836 E-35
? intnumgauss(t=-1,1,1/(1+t^2), T) - Pi/2 \\ b - a = 2r
%3 = -9.490148553624725335 E-22 \\ ... loses half the accuracy
? T = intnumgaussinit(50);
? intnumgauss(t=-1,1,1/(1+t^2), T) - Pi/2
%5 = -1.1754943508222875080 E-38
? intnumgauss(t=-5,5,1/(1+t^2), T) - 2*atan(5)
%6 = -1.2[...]E-8
On the other hand, we recommend to split the integral and change variables rather than
increasing n too much, see intnumgauss.
The library syntax is GEN intnumgaussinit(long n, long prec).

3.12.11 intnuminit(a, b, {m = 0}). Initialize tables for integration from a to b, where a and b
are coded as in intnum. Only the compactness, the possible existence of singularities, the speed
of decrease or the oscillations at infinity are taken into account, and not the values. For instance
intnuminit(-1,1) is equivalent to intnuminit(0,Pi), and intnuminit([0,-1/2],oo) is equiva-
lent to intnuminit([-1,-1/2], -oo); on the other hand, the order matters and intnuminit([0,-
1/2], [1,-1/3]) is not equivalent to intnuminit([0,-1/3], [1,-1/2]) !
If m is present, it must be nonnegative and we multiply the default number of sampling points
by 2m (increasing the running time by a similar factor).
The result is technical and liable to change in the future, but we document it here for com-
pleteness. Let x = φ(t), t ∈] − ∞, ∞[ be an internally chosen change of variable, achieving double
exponential decrease of the integrand at infinity. The integrator intnum will compute
X
h φ0 (nh)F (φ(nh))
|n|<N

325
for some integration step h and truncation parameter N . In basic use, let
[h, x0, w0, xp, wp, xm, wm] = intnuminit(a,b);
• h is the integration step
• x0 = φ(0) and w0 = φ0 (0),
• xp contains the φ(nh), 0 < n < N ,
• xm contains the φ(nh), 0 < −n < N , or is empty.
• wp contains the φ0 (nh), 0 < n < N ,
• wm contains the φ0 (nh), 0 < −n < N , or is empty.
The arrays xm and wm are left empty when φ is an odd function. In complicated situations,
intnuminit may return up to 3 such arrays, corresponding to a splitting of up to 3 integrals of
basic type.
If the functions to be integrated later are of the form F = f (t)k(t, z) for some kernel k (e.g.
Fourier, Laplace, Mellin, . . . ), it is useful to also precompute the values of f (φ(nh)), which is
accomplished by intfuncinit. The hard part is to determine the behavior of F at endpoints,
depending on z.
The library syntax is GEN intnuminit(GEN a, GEN b, long m, long prec).

3.12.12 intnumromb(X = a, b, expr , {flag = 0}). Numerical integration of expr (smooth in ]a, b[),
with respect to X. Suitable for low accuracy; if expr is very regular (e.g. analytic in a large region)
and high accuracy is desired, try intnum first.
Set flag = 0 (or omit it altogether) when a and b are not too large, the function is smooth,
and can be evaluated exactly everywhere on the interval [a, b].
If flag = 1, uses a general driver routine for doing numerical integration, making no particular
assumption (slow).
flag = 2 is tailored for being used when a or b are infinite using the change of variable t = 1/X.
One must have ab > 0, and in fact if for example b = +∞, then it is preferable to have a as large
as possible, at least a ≥ 1.
If flag = 3, the function is allowed to be undefined at a (but right continuous) or b (left
continuous), for example the function sin(x)/x between x = 0 and 1.
The user should not require too much accuracy: realprecision about 30 decimal digits
(realbitprecision about 100 bits) is OK, but not much more. In addition, analytical cleanup
of the integral must have been done: there must be no singularities in the interval or at the
boundaries. In practice this can be accomplished with a change of variable. Furthermore, for
improper integrals, where one or both of the limits of integration are plus or minus infinity, the
function must decrease sufficiently rapidly at infinity, which can often be accomplished through
integration by parts. Finally, the function to be integrated should not be very small (compared to
the current precision) on the entire interval. This can of course be accomplished by just multiplying
by an appropriate constant.
Note that infinity can be represented with essentially no loss of accuracy by an appropriate
huge number. However beware of real underflow when dealing with rapidly decreasing functions.

326
 R∞ 2
For example, in order to compute the 0 e−x dx to 28 decimal digits, then one can set infinity
equal to 10 for example, and certainly not to 1e1000.
The library syntax is GEN intnumromb_bitprec(void *E, GEN (*eval)(void*, GEN), GEN
a, GEN b, long flag, long bitprec), where eval(x, E) returns the value of the function at x.
You may store any additional information required by eval in E, or set it to NULL. The historical
variant intnumromb, where prec is expressed in words, not bits, is obsolete and should no longer
be used.

3.12.13 laurentseries(f, {M = seriesprecision}, {x =0 x}). Expand f as a Laurent series around

x = 0 to order M . This function computes f (x+O(xn )) until n is large enough: it must be possible
to evaluate f on a power series with 0 constant term.
? laurentseries(t->sin(t)/(1-cos(t)), 5)
%1 = 2*x^-1 - 1/6*x - 1/360*x^3 - 1/15120*x^5 + O(x^6)
? laurentseries(log)
*** at top-level: laurentseries(log)
*** ^------------------
*** in function laurentseries: log
*** ^---
*** log: domain error in log: series valuation != 0
Note that individual Laurent coefficients of order ≤ M can be retrieved from s =
laurentseries(f, M ) via polcoef(s,i) for any i ≤ M . The series s may occasionally be more
precise that the required O(xM +1 ).
With respect to successive calls to derivnum, laurentseries is both faster and more precise:
? laurentseries(t->log(3+t),1)
%1 = 1.0986122886681096913952452369225257047 + 1/3*x - 1/18*x^2 + O(x^3)
? derivnum(t=0,log(3+t),1)
%2 = 0.33333333333333333333333333333333333333
? derivnum(t=0,log(3+t),2)
%3 = -0.11111111111111111111111111111111111111
? f = x->sin(exp(x));
? polcoef(laurentseries(x->f(x+2), 1), 1)
%5 = 3.3129294231043339804683687620360224365
? exp(2) * cos(exp(2));
%6 = 3.3129294231043339804683687620360224365
? derivnum(x = 2, f(x))
%7 = 3.3129294231043339804683687620360224364 \\ 1 ulp off
? default(realprecision,115);
? for(i=1,10^4, laurentseries(x->f(x+2),1))
time = 279 ms.
? for(i=1,10^4, derivnum(x=2,f(x))) \\ ... and slower
time = 1,134 ms.
The library syntax is laurentseries(void *E, GEN (*f)(void*,GEN,long), long M, long
v, long prec).

327
3.12.14 limitnum(expr , {alpha = 1}). Lagrange-Zagier numerical extrapolation of expr , corre-
sponding to a sequence un , either given by a closure n->u(n). I.e., assuming that un tends to a
finite limit `, try to determine `.
The routine assume that un has an asymptotic expansion in n−α :
X
un = ` + ai n−iα
i≥1

for some ai . It is purely numerical and heuristic, thus may or may not work on your examples.
The expression will be evaluated for n = 1, 2, . . . , N for an N = O(B) at a bit accuracy bounded
by 1.612B.
? limitnum(n -> n*sin(1/n))
%1 = 1.0000000000000000000000000000000000000
? limitnum(n -> (1+1/n)^n) - exp(1)
%2 = 0.E-37
? limitnum(n -> 2^(4*n+1)*(n!)^4 / (2*n)! /(2*n+1)! ) - Pi
%3 = 0.E -37
It is not mandatory to specify α when the un have an asymptotic expansion in n−1 . However, if
the series in n−1 is lacunary, specifying α allows faster computation:
? \p1000
? limitnum(n->(1+1/n^2)^(n^2)) - exp(1)
time = 1min, 44,681 ms.
%4 = 0.E-1001
? limitnum(n->(1+1/n^2)^(n^2), 2) - exp(1)
time = 27,271 ms.
%5 = 0.E-1001 \\ still perfect, 4 times faster
When un has an asymptotic expansion in n−α with α not an integer, leaving α unspecified will
bring an inexact limit. Giving a satisfying optional argument improves precision; the program runs
faster when the optional argument gives non lacunary series.
? \p50
? limitnum(n->(1+1/n^(7/2))^(n^(7/2))) - exp(1)
time = 982 ms.
%6 = 4.13[...] E-12
? limitnum(n->(1+1/n^(7/2))^(n^(7/2)), 1/2) - exp(1)
time = 16,745 ms.
%7 = 0.E-57
? limitnum(n->(1+1/n^(7/2))^(n^(7/2)), 7/2) - exp(1)
time = 105 ms.
%8 = 0.E-57
Alternatively, un may be given by a closure N → 7 [u1 , . . . , uN ] which can often be programmed in
a more efficient way, for instance when un+1 is a simple function of the preceding terms:
? \p2000
? limitnum(n -> 2^(4*n+1)*(n!)^4 / (2*n)! /(2*n+1)! ) - Pi
time = 1,755 ms.

328
 %9 = 0.E-2003
? vu(N) = \\ exploit hypergeometric property
{ my(v = vector(N)); v[1] = 8./3;\
for (n=2, N, my(q = 4*n^2); v[n] = v[n-1]*q/(q-1));\
return(v);
}
? limitnum(vu) - Pi \\ much faster
time = 106 ms.
%11 = 0.E-2003

All sums and recursions can be handled in the same way. In the above it is essential that un be
defined as a closure because it must be evaluated at a higher precision than the one expected for the
limit. Make sure that the closure does not depend on a global variable which would be computed
at a priori fixed accuracy. For instance, precomputing v1 = 8.0/3 first and using v1 in vu above
would be wrong because the resulting vector of values will use the accuracy of v1 instead of the
ambient accuracy at which limitnum will call it.

Alternatively, and more clumsily, un may be given by a vector of values: it must be long and
precise enough for the extrapolation to make sense. Let B be the current realbitprecision, the
vector length must be at least 1.102B and the values computed with bit accuracy 1.612B.

? limitnum(vector(10,n,(1+1/n)^n))
*** ^--------------------
*** limitnum: nonexistent component in limitnum: index < 43
\\ at this accuracy, we must have at least 43 values
? limitnum(vector(43,n,(1+1/n)^n)) - exp(1)
%12 = 0.E-37
? v = vector(43);
? s = 0; for(i=1,#v, s += 1/i; v[i]= s - log(i));
? limitnum(v) - Euler
%15 = -1.57[...] E-16
? v = vector(43);
\\ ~ 128 bit * 1.612
? localbitprec(207);\
s = 0; for(i=1,#v, s += 1/i; v[i]= s - log(i));
? limitnum(v) - Euler
%18 = 0.E-38

Because of the above problems, the preferred format is thus a closure, given either a single
value or the vector of values [u1 , . . . , uN ]. The function distinguishes between the two formats by
evaluating the closure at N 6= 1 and 1 and checking whether it yields vectors of respective length
N and 1 or not.

329
Warning. The expression is evaluated for n = 1, 2, . . . , N for an N = O(B) if the current bit
accuracy is B. If it is not defined for one of these values, translate or rescale accordingly:
? limitnum(n->log(1-1/n)) \\ can’t evaluate at n = 1 !
*** at top-level: limitnum(n->log(1-1/n))
*** ^-----------------------
*** in function limitnum: log(1-1/n)
*** ^----------
*** log: domain error in log: argument = 0
? limitnum(n->-log(1-1/(2*n)))
%19 = -6.11[...] E-58
We conclude with a complicated example. Since the function is heuristic, it is advisable to
check whether it produces the same limit for un , u2n , . . . ukm for a suitable small multiplier k. The
following function implements the recursion for the Motzkin numbers Mn which count the number
of ways to draw non intersecting chords between n points on a circle:
X
Mn = Mn−1 + Mi Mn−2−i = ((n + 1)Mn−1 + (3n − 3)Mn−2 )/(n + 2).
i<n−1

n+1
It is known that Mn ∼ √3 .
12πn3

\\ [M_k, M_{k*2}, ..., M_{k*N}] / (3^n / n^(3/2))

vM(N, k = 1) =
{ my(q = k*N, V);
if (q == 1, return ([1/3]));
V = vector(q); V[1] = V[2] = 1;
for(n = 2, q - 1,
V[n+1] = ((2*n + 1)*V[n] + 3*(n - 1)*V[n-1]) / (n + 2));
f = (n -> 3^n / n^(3/2));
return (vector(N, n, V[n*k] / f(n*k)));
}
? limitnum(vM) - 3/sqrt(12*Pi) \\ complete junk
%1 = 35540390.753542730306762369615276452646
? limitnum(N->vM(N,5)) - 3/sqrt(12*Pi) \\ M_{5n}: better
%2 = 4.130710262178469860 E-25
? limitnum(N->vM(N,10)) - 3/sqrt(12*Pi) \\ M_{10n}: perfect
%3 = 0.E-38
? \p2000
? limitnum(N->vM(N,10)) - 3/sqrt(12*Pi) \\ also at high accuracy
time = 409 ms.
%4 = 1.1048895470044788191 E-2004
In difficult cases such as the above a multiplier of 5 to 10 is usually sufficient. The above example
is typical: a good multiplier usually remains sufficient when the requested precision increases!
The library syntax is limitnum(void *E, GEN (*u)(void *,GEN,long), GEN alpha, long
prec), where u(E, n, prec) must return u(n) in precision prec. Also available is GEN limit-
num0(GEN u, GEN alpha, long prec), where u must be a vector of sufficient length as above.

330
3.12.15 prod(X = a, b, expr , {x = 1}). Product of expression expr , initialized at x, the formal
parameter X going from a to b. As for sum, the main purpose of the initialization parameter x is
to force the type of the operations being performed. For example if it is set equal to the integer 1,
operations will start being done exactly. If it is set equal to the real 1., they will be done using real
numbers having the default precision. If it is set equal to the power series 1 + O(X k ) for a certain
k, they will be done using power series of precision at most k. These are the three most common
initializations.
As an extreme example, compare
? prod(i=1, 100, 1 - X^i); \\ this has degree 5050 !!
time = 128 ms.
? prod(i=1, 100, 1 - X^i, 1 + O(X^101))
time = 8 ms.
%2 = 1 - X - X^2 + X^5 + X^7 - X^12 - X^15 + X^22 + X^26 - X^35 - X^40 + \
X^51 + X^57 - X^70 - X^77 + X^92 + X^100 + O(X^101)
Of course, in this specific case, it is faster to use eta, which is computed using Euler’s formula.
? prod(i=1, 1000, 1 - X^i, 1 + O(X^1001));
time = 589 ms.
? \ps1000
seriesprecision = 1000 significant terms
? eta(X) - %
time = 8ms.
%4 = O(X^1001)
The library syntax is produit(GEN a, GEN b, char *expr, GEN x).

3.12.16 prodeuler(p = a, b, expr ). Product of expression expr , initialized at 1.0 (i.e. to a floating
point number equal to 1 to the current realprecision), the formal parameter p ranging over the
prime numbers between a and b.
? prodeuler(p = 2, 10^4, 1 - p^-2)
%1 = 0.60793306911405513018380499671124428015
? P = 1; forprime(p = 2, 10^4, P *= (1 - p^-2))
? exponent(numerator(P))
%3 = 22953
The function returns a floating point number because, as the second expression shows, such products
are usually intractably large rational numbers when computed symbolically. If the expression is a
rational funtction, prodeulerrat computes the product over all primes:
? prodeulerrat(1 - p^-2)
%4 = 0.60792710185402662866327677925836583343
? 6/Pi^2
%3 = 0.60792710185402662866327677925836583343
The library syntax is prodeuler(void *E, GEN (*eval)(void*,GEN), GEN a,GEN b, long
prec).

331
 F (ps ), where the product is taken over prime
Q
3.12.17 prodeulerrat(F, {s = 1}, {a = 2}). p≥a
numbers and F is a rational function.
? prodeulerrat(1+1/q^3,1)
%1 = 1.1815649490102569125693997341604542605
? zeta(3)/zeta(6)
%2 = 1.1815649490102569125693997341604542606
The library syntax is GEN prodeulerrat(GEN F, GEN s = NULL, long a, long prec)
.

3.12.18 prodinf(X = a, expr , {flag = 0}). infinite product of expression expr , the formal pa-
rameter X starting at a. The evaluation stops when the relative error of the expression minus 1
is less than the default precision. In particular, divergent products result in infinite loops. The
expressions must always evaluate to an element of C.
If flag = 1, do the product of the (1 + expr ) instead.
The library syntax is prodinf(void *E, GEN (*eval)(void*,GEN), GEN a, long prec)
(flag = 0), or prodinf1 with the same arguments (flag = 1).
Q
3.12.19 prodnumrat(F, a). n≥a F (n), where F − 1 is a rational function of degree less than or
equal to −2.
? prodnumrat(1+1/x^2,1)
%1 = 3.6760779103749777206956974920282606665
The library syntax is GEN prodnumrat(GEN F, long a, long prec).

3.12.20 solve(X = a, b, expr ). Find a real root of expression expr between a and b, under the con-
dition expr (X = a) ∗ expr (X = b) ≤ 0. (You will get an error message roots must be bracketed
in solve if this does not hold.) This routine uses Brent’s method and can fail miserably if expr is
not defined in the whole of [a, b] (try solve(x=1, 2, tan(x))).
The library syntax is zbrent(void *E,GEN (*eval)(void*,GEN),GEN a,GEN b,long prec).

3.12.21 solvestep(X = a, b, step, expr , {flag = 0}). Find zeros of a continuous function in the real
interval [a, b] by naive interval splitting. This function is heuristic and may or may not find the
intended zeros. Binary digits of flag mean
• 1: return as soon as one zero is found, otherwise return all zeros found;
• 2: refine the splitting until at least one zero is found (may loop indefinitely if there are no
zeros);
• 4: do a multiplicative search (we must have a > 0 and step > 1), otherwise an additive
search; step is the multiplicative or additive step.
• 8: refine the splitting until at least one zero is very close to an integer.
? solvestep(X=0,10,1,sin(X^2),1)
%1 = 1.7724538509055160272981674833411451828
? solvestep(X=1,12,2,besselj(4,X),4)
%2 = [7.588342434..., 11.064709488...]

The library syntax is solvestep(void *E, GEN (*eval)(void*,GEN), GEN a,GEN b, GEN
step,long flag,long prec).

332
3.12.22 sum(X = a, b, expr , {x = 0}). Sum of expression expr , initialized at x, the formal
parameter going from a to b. As for prod, the initialization parameter x may be given to force the
type of the operations being performed.
As an extreme example, compare
? sum(i=1, 10^4, 1/i); \\ rational number: denominator has 4345 digits.
time = 236 ms.
? sum(i=1, 5000, 1/i, 0.)
time = 8 ms.
%2 = 9.787606036044382264178477904

3.12.23 sumalt(X = a, expr , {flag = 0}). Numerical summation of the series expr , which should
be an alternating series (−1)k ak , the formal variable X starting at a. Use an algorithm of Cohen,
Villegas and Zagier (Experiment. Math. 9 (2000), no. 1, 3–12).
If flag = 0, assuming√ that the ak are the moments of a positive measure on [0, 1], the
relative error is O(3 √+ 8)−n after using ak for k ≤ n. If realprecision is p, we thus set
n = log(10)p/ log(3 + 8) ≈ 1.3p; besides the time needed to compute the ak , k ≤ n, the algorithm
overhead is negligible: time O(p2 ) and space O(p).
If flag = 1, use a variant with more complicated polynomials, see polzagier. If the ak are
the moments of w(x)dx where w (or only xw(x2 )) is a smooth function extending analytically
to the whole complex plane, convergence is in O(14.4−n ). If xw(x2 ) extends analytically to a
smaller region, we still have exponential convergence, with worse constants. Usually faster when the
computation of ak is expensive. If realprecision is p, we thus set n = log(10)p/ log(14.4) ≈ 0.86p;
besides the time needed to compute the ak , k ≤ n, the algorithm overhead is not negligible: time
O(p3 ) and space O(p2 ). Thus, even if the analytic conditions for rigorous use are met, this variant
is only worthwile if the ak are hard to compute, at least O(p2 ) individually on average: otherwise
we gain a small constant factor (1.5, say) in the number of needed ak at the expense of a large
overhead.
The conditions for rigorous use are hard to check but the routine is best used heuristically:
even divergent alternating series can sometimes be summed by this method, as well as series which
are not exactly alternating (see for example Section 2.7). It should be used to try and guess the
value of an infinite sum. (However, see the example at the end of Section 2.7.1.)
If the series already converges geometrically, suminf is often a better choice:
? \p38
? sumalt(i = 1, -(-1)^i / i) - log(2)
time = 0 ms.
%1 = 0.E-38
? suminf(i = 1, -(-1)^i / i) \\ Had to hit Ctrl-C
*** at top-level: suminf(i=1,-(-1)^i/i)
*** ^------
*** suminf: user interrupt after 10min, 20,100 ms.
? \p1000
? sumalt(i = 1, -(-1)^i / i) - log(2)
time = 90 ms.
%2 = 4.459597722 E-1002
? sumalt(i = 0, (-1)^i / i!) - exp(-1)

333
 time = 670 ms.
%3 = -4.03698781490633483156497361352190615794353338591897830587 E-944
? suminf(i = 0, (-1)^i / i!) - exp(-1)
time = 110 ms.
%4 = -8.39147638 E-1000 \\ faster and more accurate
The library syntax is sumalt(void *E, GEN (*eval)(void*,GEN),GEN a,long prec). Also
available is sumalt2 with the same arguments (flag = 1).

3.12.24 sumdiv(n, X, expr ). Sum of expression expr over the positive divisors of n. This function
is a trivial wrapper essentially equivalent to
D = divisors(n);
sum (i = 1, #D, my(X = D[i]); eval(expr))
If expr is a multiplicative function, use sumdivmult.

3.12.25 sumdivmult(n, d, expr ). Sum of multiplicative expression expr over the positive di-
visors d of n. Assume that expr evaluates to f (d) where f is multiplicative: f (1) = 1 and
f (ab) = f (a)f (b) for coprime a and b. The library syntax is sumdivmultexpr(void *E, GEN
(*eval)(void*,GEN), GEN d)

F (ps ), where the sum is taken over prime

P
3.12.26 sumeulerrat(F, {s = 1}, {a = 2}). p≥a
numbers and F is a rational function.
? sumeulerrat(1/p^2)
%1 = 0.45224742004106549850654336483224793417
? sumeulerrat(1/p, 2)
%2 = 0.45224742004106549850654336483224793417
The library syntax is GEN sumeulerrat(GEN F, GEN s = NULL, long a, long prec).

3.12.27 suminf(X = a, expr ). Naive summation of expression expr , the formal parameter X going
from a to infinity. The evaluation stops when the relative error of the expression is less than the
default bit precision for 3 consecutive evaluations. The expressions must evaluate to a complex
number.
If the expression tends slowly to 0, like n−a for some a > 1, make sure b = realbitprecision
is low: indeed, the algorithm will require O(2b/a ) function evaluations and we expect only about
b(1 − 1/a) correct bits in the answer. If the series is alternating, we can expect b correct bits
but the sumalt function should be used instead since its complexity is polynomial in b, instead of
exponential. More generally, sumpos should be used if the terms have a constant sign and sumnum
if the function is C ∞ .
? \pb25
realbitprecision = 25 significant bits (7 decimal digits displayed)
? exponent(suminf(i = 1, (-1)^i / i) + log(2))
time = 2min, 2,602 ms.
%1 = -29
? \pb45
realbitprecision = 45 significant bits (13 decimal digits displayed)
? exponent(suminf(i = 1, 1 / i^2) - zeta(2))
time = 2,186 ms.

334
 %2 = -23
\\ alternatives are much faster
? \pb 10000
realbitprecision = 10000 significant bits (3010 decimal digits displayed)
? exponent(sumalt(i = 1, (-1)^i / i) + log(2))
time = 25 ms.
%3 = -10043
? \pb 4000
realbitprecision = 4000 significant bits (1204 decimal digits displayed)))
? exponent(sumpos(i = 1, 1 / i^2) - zeta(2))
time = 22,593 ms.
%4 = -4030
? exponent(sumnum(i = 1, 1 / i^2) - zeta(2))
time = 7,032 ms.
%5 = -4031
\\ but suminf is perfect for geometrically converging series
? exponent(suminf(i = 1, 2^-i) - 1)
time = 25 ms.
%6 = -4003

The library syntax is suminf˙bitprec(void *E, GEN (*eval)(void*,GEN), GEN a, long

prec). The historical variant GEN suminf(\dots, long prec), where prec is expressed in words,
not bits, is obsolete and should no longer be used.

3.12.28 sumnum(n = a, f, {tab}). Numerical summation of f (n) at high accuracy using Euler-
MacLaurin, the variable n taking values from a to +∞, where f is assumed to have positive values
and is a C ∞ function; a must be an integer and tab, if given, is the output of sumnuminit. The
latter precomputes abscissas and weights, speeding up the computation; it also allows to specify
the behavior at infinity via sumnuminit([+oo, asymp]).

? \p500
? z3 = zeta(3);
? sumpos(n = 1, n^-3) - z3
time = 2,332 ms.
%2 = 2.438468843 E-501
? sumnum(n = 1, n^-3) - z3 \\ here slower than sumpos
time = 2,752 ms.
%3 = 0.E-500

335
Complexity. The function f will be evaluated at O(D log D) real arguments, where D ≈
realprecision · log(10). The routine is geared towards slowly decreasing functions: if f decreases
exponentially fast, then one of suminf or sumpos should be preferred. If f satisfies the stronger hy-
potheses required for Monien summation, i.e. if f (1/z) is holomorphic in a complex neighbourhood
of [0, 1], then sumnummonien will be faster since it only requires O(D/ log D) evaluations:
? sumnummonien(n = 1, 1/n^3) - z3
time = 1,985 ms.
%3 = 0.E-500
The tab argument precomputes technical data not depending on the expression being summed and
valid for a given accuracy, speeding up immensely later calls:
? tab = sumnuminit();
time = 2,709 ms.
? sumnum(n = 1, 1/n^3, tab) - z3 \\ now much faster than sumpos
time = 40 ms.
%5 = 0.E-500
? tabmon = sumnummonieninit(); \\ Monien summation allows precomputations too
time = 1,781 ms.
? sumnummonien(n = 1, 1/n^3, tabmon) - z3
time = 2 ms.
%7 = 0.E-500
The speedup due to precomputations becomes less impressive when the function f is expensive to
evaluate, though:
? sumnum(n = 1, lngamma(1+1/n)/n, tab);
time = 14,180 ms.
? sumnummonien(n = 1, lngamma(1+1/n)/n, tabmon); \\ fewer evaluations
time = 717 ms.

Behaviour at infinity. By default, sumnum assumes that expr decreases slowly at infinity, but
at least like O(n−2 ). If the function decreases like nα for some −2 < α < −1, then it must be
indicated via
tab = sumnuminit([+oo, alpha]); /* alpha < 0 slow decrease */
otherwise loss of accuracy is expected. If the functions decreases quickly, like exp(−αn) for some
α > 0, then it must be indicated via
tab = sumnuminit([+oo, alpha]); /* alpha > 0 exponential decrease */
otherwise exponent overflow will occur.
? sumnum(n=1,2^-n)
*** at top-level: sumnum(n=1,2^-n)
*** ^----
*** _^_: overflow in expo().
? tab = sumnuminit([+oo,log(2)]); sumnum(n=1,2^-n, tab)
%1 = 1.000[...]
As a shortcut, one can also input
sumnum(n = [a, asymp], f)

336
instead of

tab = sumnuminit(asymp);
sumnum(n = a, f, tab)

Further examples.

? \p200
? sumnum(n = 1, n^(-2)) - zeta(2) \\ accurate, fast
time = 200 ms.
%1 = -2.376364457868949779 E-212
? sumpos(n = 1, n^(-2)) - zeta(2) \\ even faster
time = 96 ms.
%2 = 0.E-211
? sumpos(n=1,n^(-4/3)) - zeta(4/3) \\ now much slower
time = 13,045 ms.
%3 = -9.980730723049589073 E-210
? sumnum(n=1,n^(-4/3)) - zeta(4/3) \\ fast but inaccurate
time = 365 ms.
%4 = -9.85[...]E-85
? sumnum(n=[1,-4/3],n^(-4/3)) - zeta(4/3) \\ with decrease rate, now accurate
time = 416 ms.
%5 = -4.134874156691972616 E-210
? tab = sumnuminit([+oo,-4/3]);
time = 196 ms.
? sumnum(n=1, n^(-4/3), tab) - zeta(4/3) \\ faster with precomputations
time = 216 ms.
%5 = -4.134874156691972616 E-210
? sumnum(n=1,-log(n)*n^(-4/3), tab) - zeta’(4/3)
time = 321 ms.
%7 = 7.224147951921607329 E-210

Note that in the case of slow decrease (α < 0), the exact decrease rate must be indicated,
while in the case of exponential decrease, a rough value will do. In fact, for exponentially decreasing
functions, sumnum is given for completeness and comparison purposes only: one of suminf or sumpos
should always be preferred.

? sumnum(n=[1, 1], 2^-n) \\ pretend we decrease as exp(-n)

time = 240 ms.
%8 = 1.000[...] \\ perfect
? sumpos(n=1, 2^-n)
%9 = 1.000[...] \\ perfect and instantaneous

337
Beware cancellation. The function f (n) is evaluated for huge values of n, so beware of cancel-
lation in the evaluation:

? f(n) = 2 - 1/n - 2*n*log(1+1/n); \\ result is O(1/n^2)

? z = -2 + log(2*Pi) - Euler;
? sumnummonien(n=1, f(n)) - z
time = 149 ms.
%12 = 0.E-212 \\ perfect
? sumnum(n=1, f(n)) - z
time = 116 ms.
%13 = -948.216[...] \\ junk

As sumnum(n=1, print(n)) shows, we evaluate f (n) for n > 1e233 and our implementation of
f suffers from massive cancellation since we are summing two terms of the order of O(1) for a
result in O(1/n2 ). You can either rewrite your sum so that individual terms are evaluated without
cancellation or locally replace f (n) by an accurate asymptotic expansion:

? F = truncate( f(1/x + O(x^30)) );

? sumnum(n=1, if(n > 1e7, subst(F,x,1/n), f(n))) - z
%15 = 1.1 E-212 \\ now perfect

The library syntax is sumnum((void *E, GEN (*eval)(void*, GEN), GEN a, GEN tab,
long prec)) where an omitted tab is coded as NULL.

3.12.29 sumnumap(n = a, f, {tab}). Numerical summation of f (n) at high accuracy using Abel-
Plana, the variable n taking values from a to +∞, where f is holomorphic in the right half-place
<(z) > a; a must be an integer and tab, if given, is the output of sumnumapinit. The latter
precomputes abscissas and weights, speeding up the computation; it also allows to specify the
behavior at infinity via sumnumapinit([+oo, asymp]).

? \p500
? z3 = zeta(3);
? sumpos(n = 1, n^-3) - z3
time = 2,332 ms.
%2 = 2.438468843 E-501
? sumnumap(n = 1, n^-3) - z3 \\ here slower than sumpos
time = 2,565 ms.
%3 = 0.E-500

338
Complexity. The function f will be evaluated at O(D log D) real arguments and O(D) complex
arguments, where D ≈ realprecision · log(10). The routine is geared towards slowly decreasing
functions: if f decreases exponentially fast, then one of suminf or sumpos should be preferred. The
default algorithm sumnum is usually a little slower than sumnumap but its initialization function
sumnuminit becomes much faster as realprecision increases.
If f satisfies the stronger hypotheses required for Monien summation, i.e. if f (1/z) is holomor-
phic in a complex neighbourhood of [0, 1], then sumnummonien will be faster since it only requires
O(D/ log D) evaluations:
? sumnummonien(n = 1, 1/n^3) - z3
time = 1,128 ms.
%3 = 0.E-500
The tab argument precomputes technical data not depending on the expression being summed and
valid for a given accuracy, speeding up immensely later calls:
? tab = sumnumapinit();
time = 2,567 ms.
? sumnumap(n = 1, 1/n^3, tab) - z3 \\ now much faster than sumpos
time = 39 ms.
%5 = 0.E-500
? tabmon = sumnummonieninit(); \\ Monien summation allows precomputations too
time = 1,125 ms.
? sumnummonien(n = 1, 1/n^3, tabmon) - z3
time = 2 ms.
%7 = 0.E-500
The speedup due to precomputations becomes less impressive when the function f is expensive to
evaluate, though:
? sumnumap(n = 1, lngamma(1+1/n)/n, tab);
time = 10,762 ms.
? sumnummonien(n = 1, lngamma(1+1/n)/n, tabmon); \\ fewer evaluations
time = 205 ms.

Behaviour at infinity. By default, sumnumap assumes that expr decreases slowly at infinity, but
at least like O(n−2 ). If the function decreases like nα for some −2 < α < −1, then it must be
indicated via
tab = sumnumapinit([+oo, alpha]); /* alpha < 0 slow decrease */
otherwise loss of accuracy is expected. If the functions decreases quickly, like exp(−αn) for some
α > 0, then it must be indicated via
tab = sumnumapinit([+oo, alpha]); /* alpha > 0 exponential decrease */
otherwise exponent overflow will occur.
? sumnumap(n=1,2^-n)
*** at top-level: sumnumap(n=1,2^-n)
*** ^----
*** _^_: overflow in expo().
? tab = sumnumapinit([+oo,log(2)]); sumnumap(n=1,2^-n, tab)

339
 %1 = 1.000[...]
As a shortcut, one can also input
sumnumap(n = [a, asymp], f)
instead of
tab = sumnumapinit(asymp);
sumnumap(n = a, f, tab)

Further examples.
? \p200
? sumnumap(n = 1, n^(-2)) - zeta(2) \\ accurate, fast
time = 169 ms.
%1 = -4.752728915737899559 E-212
? sumpos(n = 1, n^(-2)) - zeta(2) \\ even faster
time = 79 ms.
%2 = 0.E-211
? sumpos(n=1,n^(-4/3)) - zeta(4/3) \\ now much slower
time = 10,518 ms.
%3 = -9.980730723049589073 E-210
? sumnumap(n=1,n^(-4/3)) - zeta(4/3) \\ fast but inaccurate
time = 309 ms.
%4 = -2.57[...]E-78
? sumnumap(n=[1,-4/3],n^(-4/3)) - zeta(4/3) \\ decrease rate: now accurate
time = 329 ms.
%6 = -5.418110963941205497 E-210
? tab = sumnumapinit([+oo,-4/3]);
time = 160 ms.
? sumnumap(n=1, n^(-4/3), tab) - zeta(4/3) \\ faster with precomputations
time = 175 ms.
%5 = -5.418110963941205497 E-210
? sumnumap(n=1,-log(n)*n^(-4/3), tab) - zeta’(4/3)
time = 258 ms.
%7 = 9.125239518216767153 E-210
Note that in the case of slow decrease (α < 0), the exact decrease rate must be indicated, while
in the case of exponential decrease, a rough value will do. In fact, for exponentially decreasing
functions, sumnumap is given for completeness and comparison purposes only: one of suminf or
sumpos should always be preferred.
? sumnumap(n=[1, 1], 2^-n) \\ pretend we decrease as exp(-n)
time = 240 ms.
%8 = 1.000[...] \\ perfect
? sumpos(n=1, 2^-n)
%9 = 1.000[...] \\ perfect and instantaneous
The library syntax is sumnumap((void *E, GEN (*eval)(void*,GEN), GEN a, GEN tab,
long prec)) where an omitted tab is coded as NULL.

340
 P
3.12.30 sumnumapinit({asymp}). Initialize tables for Abel–Plana summation of a series f (n),
where f is holomorphic in a right half-plane. If given, asymp is of the form [+oo, α], as in intnum
and indicates the decrease rate at infinity of functions to be summed. A positive α > 0 encodes
an exponential decrease of type exp(−αn) and a negative −2 < α < −1 encodes a slow polynomial
decrease of type nα .
? \p200
? sumnumap(n=1, n^-2);
time = 163 ms.
? tab = sumnumapinit();
time = 160 ms.
? sumnumap(n=1, n^-2, tab); \\ faster
time = 7 ms.
? tab = sumnumapinit([+oo, log(2)]); \\ decrease like 2^-n
time = 164 ms.
? sumnumap(n=1, 2^-n, tab) - 1
time = 36 ms.
%5 = 3.0127431466707723218 E-282
? tab = sumnumapinit([+oo, -4/3]); \\ decrease like n^(-4/3)
time = 166 ms.
? sumnumap(n=1, n^(-4/3), tab);
time = 181 ms.
The library syntax is GEN sumnumapinit(GEN asymp = NULL, long prec).

3.12.31 sumnuminit({asymp}). Initialize tables for Euler–MacLaurin delta summation of a series

with positive terms. If given, asymp is of the form [+oo, α], as in intnum and indicates the decrease
rate at infinity of functions to be summed. A positive α > 0 encodes an exponential decrease of
type exp(−αn) and a negative −2 < α < −1 encodes a slow polynomial decrease of type nα .
? \p200
? sumnum(n=1, n^-2);
time = 200 ms.
? tab = sumnuminit();
time = 188 ms.
? sumnum(n=1, n^-2, tab); \\ faster
time = 8 ms.
? tab = sumnuminit([+oo, log(2)]); \\ decrease like 2^-n
time = 200 ms.
? sumnum(n=1, 2^-n, tab)
time = 44 ms.
? tab = sumnuminit([+oo, -4/3]); \\ decrease like n^(-4/3)
time = 200 ms.
? sumnum(n=1, n^(-4/3), tab);
time = 221 ms.
The library syntax is GEN sumnuminit(GEN asymp = NULL, long prec).

341
3.12.32 sumnumlagrange(n = a, f, {tab}). Numerical summation of f (n) from n = a to +∞
using Lagrange summation; a must be an integer, and the optional argument tab is the output of
sumnumlagrangeinit. By default, the program assumes that the N th remainder has an asymptotic
expansion in integral powers of 1/N . If not, initialize tab using sumnumlagrangeinit(al), where
the asymptotic expansion of the remainder is integral powers of 1/N al ; al can be equal to 1 (default),
1/2, 1/3, or 1/4, and also equal to 2, but in this latter case it is the N th remainder minus one half
of the last summand which has an asymptotic expansion in integral powers of 1/N 2 .
? \p1000
? z3 = zeta(3);
? sumpos(n = 1, n^-3) - z3
time = 4,440 ms.
%2 = -2.08[...] E-1001
? sumnumlagrange(n = 1, n^-3) - z3 \\ much faster than sumpos
time = 25 ms.
%3 = 0.E-1001
? tab = sumnumlagrangeinit();
time = 21 ms.
? sumnumlagrange(n = 1, n^-3, tab) - z3
time = 2 ms. /* even faster */
%5 = 0.E-1001
? \p115
? tab = sumnumlagrangeinit([1/3,1/3]);
time = 316 ms.
? sumnumlagrange(n = 1, n^-(7/3), tab) - zeta(7/3)
time = 24 ms.
%7 = 0.E-115
? sumnumlagrange(n = 1, n^(-2/3) - 3*(n^(1/3)-(n-1)^(1/3)), tab) - zeta(2/3)
time = 32 ms.
%8 = 1.0151767349262596893 E-115
Complexity. The function f is evaluated at O(D) integer arguments, where D ≈ realprecision·
log(10).
The library syntax is sumnumlagrange((void *E, GEN (*eval)(void*, GEN), GEN a,
GEN tab, long prec)) where an omitted tab is coded as NULL.

3.12.33 sumnumlagrangeinit({asymp}, {c1 }). InitializeP tables for Lagrange summation of a

series. By default, assume that the remainder R(n) = m≥n f (m) has an asymptotic expansion
X X
R(n) = f (n) ≈ ai /ni
m≥n i≥1

at infinity. The argument asymp allows to specify different expansions:

• a real number β means X
R(n) = n−β ai /ni
i≥1

• a t_CLOSURE g means X
R(n) = g(n) ai /ni
i≥1

342
(The preceding case corresponds to g(n) = n−β .)

• a pair [α, β] where β is as above and α ∈ {2, 1, 1/2, 1/3, 1/4}. We let R2 (n) = R(n) − f (n)/2
and Rα (n) = R(n) for α 6= 2. Then
X
Rα (n) = g(n) ai /niα
i≥1

Note that the initialization times increase considerable for the α is this list (1/4 being the slowest).

The constant c1 is technical and computed by the program, but can be set by the user: the
number of interpolation steps will be chosen close to c1 · B, where B is the bit accuracy.

? \p2000
? sumnumlagrange(n=1, n^-2);
time = 173 ms.
? tab = sumnumlagrangeinit();
time = 172 ms.
? sumnumlagrange(n=1, n^-2, tab);
time = 4 ms.
? \p115
? sumnumlagrange(n=1, n^(-4/3)) - zeta(4/3);
%1 = -0.1093[...] \\ junk: expansion in n^(1/3)
time = 84 ms.
? tab = sumnumlagrangeinit([1/3,0]); \\ alpha = 1/3
time = 336 ms.
? sumnumlagrange(n=1, n^(-4/3), tab) - zeta(4/3)
time = 84 ms.
%3 = 1.0151767349262596893 E-115 \\ now OK
? tab = sumnumlagrangeinit(1/3); \\ alpha = 1, beta = 1/3: much faster
time = 3ms
? sumnumlagrange(n=1, n^(-4/3), tab) - zeta(4/3) \\ ... but wrong
%5 = -0.273825[...] \\ junk !
? tab = sumnumlagrangeinit(-2/3); \\ alpha = 1, beta = -2/3
time = 3ms
? sumnumlagrange(n=1, n^(-4/3), tab) - zeta(4/3)
%6 = 2.030353469852519379 E-115 \\ now OK

in The final example with ζ(4/3), the remainder R1 (n) is of the form n−1/3 i≥0 ai /ni , i.e.
P

n2/3 i≥1 ai /ni . The explains the wrong result for β = 1/3 and the correction with β = −2/3.
P

The library syntax is GEN sumnumlagrangeinit(GEN asymp = NULL, GEN c1 = NULL, long
prec).

343
 P
3.12.34 sumnummonien(n = a, f, {tab}). Numerical summation n≥a f (n) at high accuracy,
the variable n taking values from the integer a to +∞ using Monien summation, which assumes
that f (1/z) has a complex analytic continuation in a (complex) neighbourhood of the segment
[0, 1].
The function f is evaluated at O(D/ log D) real arguments, where D ≈ realprecision·log(10).
By default, assume that f (n) = O(n−2 ) and has a nonzero asymptotic expansion
X
f (n) = ai n−i
i≥2

at infinity. To handle more complicated behaviors and allow time-saving precomputations (for a
given realprecision), see sumnummonieninit.
The library syntax is GEN sumnummonien0(GEN n, GEN f, GEN tab = NULL, long prec)
.

3.12.35 sumnummonieninit({asymp},
P {w}, {n0 = 1}). Initialize tables for Monien summation
of a series n≥n0 f (n) where f (1/z) has a complex analytic continuation in a (complex) neigh-
bourhood of the segment [0, 1].
By default, assume that f (n) = O(n−2 ) and has a nonzero asymptotic expansion
X
f (n) = ai /ni
i≥2

at infinity. Note that the sum starts at i = 2! The argument asymp allows to specify different
expansions:
• a real number β > 0 means X
f (n) = ai /ni+β
i≥1

(Now the summation starts at 1.)

• a vector [α, β] of reals, where we must have α > 0 and α + β > 1 to ensure convergence,
means that X
f (n) = ai /nαi+β
i≥1

Note that asymp = [1, β] is equivalent to asymp = β.

? \p57
? s = sumnum(n = 1, sin(1/sqrt(n)) / n); \\ reference point
? \p38
? sumnummonien(n = 1, sin(1/sqrt(n)) / n) - s
%2 = -0.001[...] \\ completely wrong
? t = sumnummonieninit(1/2); \\ f(n) = sum_i 1 / n^(i+1/2)
? sumnummonien(n = 1, sin(1/sqrt(n)) / n, t) - s
%3 = 0.E-37 \\ now correct
(As a matter of fact, in the above summation, the result given by sumnum at \p38 is slighly incorrect,
so we had to increase the accuracy to \p57.)

344
 The argument w is used to sum expressions of the form
X
f (n)w(n),
n≥n0

for varying f as above, and fixed weight function w, where we further assume that the auxiliary
sums X
gw (m) = w(n)/nαm+β
n≥n0

converge for all m ≥ 1. Note that for nonnegative integers k, and weight w(n) = (log n)k , the
function gw (m) = ζ (k) (αm + β) has a simple expression; for general weights, gw is computed using
sumnum. The following variants are available
• an integer k ≥ 0, to code w(n) = (log n)k ;
• a t_CLOSURE computing the values w(n), where we assume that w(n) = O(n ) for all  > 0;
• a vector [w, fast], where w is a closure as above and fast is a scalar; we assume that w(n) =
O(nfast+ ); note that w = [w, 0] is equivalent to w = w. Note that if w decreases exponentially,
suminf should be used instead.
The subsequent calls to sumnummonien must use the same value of n0 as was used here.
? \p300
? sumnummonien(n = 1, n^-2*log(n)) + zeta’(2)
time = 328 ms.
%1 = -1.323[...]E-6 \\ completely wrong, f does not satisfy hypotheses !
? tab = sumnummonieninit(, 1); \\ codes w(n) = log(n)
time = 3,993 ms.
? sumnummonien(n = 1, n^-2, tab) + zeta’(2)
time = 41 ms.
%3 = -5.562684646268003458 E-309 \\ now perfect
? tab = sumnummonieninit(, n->log(n)); \\ generic, slower
time = 9,808 ms.
? sumnummonien(n = 1, n^-2, tab) + zeta’(2)
time = 40 ms.
%5 = -5.562684646268003458 E-309 \\ identical result
The library syntax is GEN sumnummonieninit(GEN asymp = NULL, GEN w = NULL, GEN n0
= NULL, long prec).
P
3.12.36 sumnumrat(F, a). n≥a F (n), where F is a rational function of degree less than or equal
to −2 and where poles of F at integers ≥ a are omitted from the summation. The argument a
must be a t_INT or -oo.
? sumnumrat(1/(x^2+1)^2,0)
%1 = 1.3068369754229086939178621382829073480
? sumnumrat(1/x^2, -oo) \\ value at x=0 is discarded
%2 = 3.2898681336964528729448303332920503784
? 2*zeta(2)
%3 = 3.2898681336964528729448303332920503784

345
When deg F = −1, we define
∞
X X
F (n) := (F (n) + F (−1 − n)) :
−∞ n≥0

? sumnumrat(1/x, -oo)
%4 = 0.E-38
The library syntax is GEN sumnumrat(GEN F, GEN a, long prec).

3.12.37 sumpos(X = a, expr , {flag = 0}). Numerical summation of the series expr , which must
be a series of terms having the same sign, the formal variable X starting at a. The algorithm uses
Van Wijngaarden’s trick for converting such a series into an alternating one, then sumalt. For
regular functions, the function sumnum is in general much faster once the initializations have been
made using sumnuminit. Contrary to sumnum, sumpos allows functions defined only at integers:
? sumnum(n = 0, 1/n!)
*** at top-level: sumnum(n=1,1/n!)
*** ^---
*** incorrect type in gtos [integer expected] (t_FRAC).
? sumpos(n = 0, 1/n!) - exp(1)
%2 = -1.0862155548773347717 E-33
On the other hand, when the function accepts general real numbers, it is usually advantageous to
replace n by n * 1.0 in the sumpos call in particular when rational functions are involved:
? \p500
? sumpos(n = 0, n^7 / (n^9+n+1));
time = 6,108 ms.
? sumpos(n = 0, n *= 1.; n^7 / (n^9+n+1));
time = 2,788 ms.
? sumnumrat(n^7 / (n^9+n+1), 0);
time = 4 ms.
In the last example, sumnumrat is of course much faster but it only applies to rational functions.
The routine is heuristic and assumes that expr is more or less a decreasing function of X. In
particular, the result will be completely wrong if expr is 0 too often. We do not check either that
all terms have the same sign: as sumalt, this function should be used to try and guess the value of
an infinite sum.
If flag = 1, use sumalt(, 1) instead of sumalt(, 0), see Section 3.12.23. Requiring more stringent
analytic properties for rigorous use, but allowing to compute fewer series terms.
To reach accuracy 10−p , both algorithms require O(p2 ) space; furthermore, assuming the terms
decrease polynomially (in O(n−C )), both need to compute O(p2 ) terms. The sumpos(, 1) variant
has a smaller implied constant (roughly 1.5 times smaller). Since the sumalt(, 1) overhead is now
small compared to the time needed to compute series terms, this last variant should be about
1.5 faster. On the other hand, the achieved accuracy may be much worse: as for sumalt, since
conditions for rigorous use are hard to check, the routine is best used heuristically.
The library syntax is sumpos(void *E, GEN (*eval)(void*,GEN),GEN a,long prec). Also
available is sumpos2 with the same arguments (flag = 1).

346
3.13 General number fields.
In this section, we describe functions related to general number fields. Functions related to
quadratic number fields are found in Section 3.8 (Arithmetic functions).
3.13.1 Number field structures.
Let K = Q[X]/(T ) a number field, ZK its ring of integers, T ∈ Z[X] is monic. Three basic
number field structures can be attached to K in GP:
• nf denotes a number field, i.e. a data structure output by nfinit. This contains the basic
arithmetic data attached to the number field: signature, maximal order (given by a basis nf.zk),
discriminant, defining polynomial T , etc.
• bnf denotes a “Buchmann’s number field”, i.e. a data structure output by bnfinit. This
contains nf and the deeper invariants of the field: units U (K), class group Cl(K), as well as
technical data required to solve the two attached discrete logarithm problems.
• bnr denotes a “ray number field”, i.e. a data structure output by bnrinit, corresponding to
the ray class group structure of the field, for some modulus f . It contains a bnf , the modulus f ,
the ray class group Clf (K) and data attached to the discrete logarithm problem therein.
3.13.2 Algebraic numbers and ideals.
An algebraic number belonging to K = Q[X]/(T ) is given as
• a t_INT, t_FRAC or t_POL (implicitly modulo T ), or
• a t_POLMOD (modulo T ), or
• a t_COL v of dimension N = [K : Q], representing the element in terms of the computed
integral basis, as sum(i = 1, N, v[i] * nf.zk[i]). Note that a t_VEC will not be recognized.

An ideal is given in any of the following ways:

• an algebraic number in one of the above forms, defining a principal ideal.
• a prime ideal, i.e. a 5-component vector in the format output by idealprimedec or ideal-
factor.
• a t_MAT, square and in Hermite Normal Form (or at least upper triangular with nonnegative
coefficients), whose columns represent a Z-basis of the ideal.
One may use idealhnf to convert any ideal to the last (preferred) format.
• an extended ideal is a 2-component vector [I, t], where I is an ideal as above and t is
an algebraic number, representing the ideal (t)I. This is useful whenever idealred is involved,
implicitly working in the ideal class group, while keeping track of principal ideals. The following
multiplicative ideal operations update the principal part: idealmul, idealinv, idealpow and
idealred; e.g. using idealmul on [I, t], [J, u], we obtain [IJ, tu]. In all other functions, the extended
part is silently discarded, e.g. using idealadd with the above input produces I + J.
The “principal part” t in an extended ideal may be represented in any of the above forms,
and also as a factorization matrix (in terms of number field elements, not ideals!), possibly the
empty factorization matrix factor(1) representing 1; the empty matrix [;] is also accepted as a
synonym for 1. When t is such a factorization matrix, elements stay in factored form, or famat
for factorization matrix, which is a convenient way to avoid coefficient explosion. To recover the
conventional expanded form, try nffactorback; but many functions already accept famats as
input, for instance ideallog, so expanding huge elements should never be necessary.

347
3.13.3 Finite abelian groups.
A finite abelian group G in user-readable format is given by its Smith Normal Form as a pair
[h, d] or triple [h, d, g]. Here h is the cardinality of G, (di ) is the vector of elementary divisors,
Q and
(gi ) is a vector of generators. In short, G = ⊕i≤n (Z/di Z)gi , with dn | . . . | d2 | d1 and di = h.
This information can also be retrieved as G.no, G.cyc and G.gen.
• aQcharacter on the P
abelian group ⊕(Z/dj Z)gj is given by a row vector χ = [a1 , . . . , an ] such
n
that χ( gj j ) = exp(2πi aj nj /dj ).
• given such a structure, a subgroup H is input as a square matrix in HNF, whose columns
express generators of H on the given generators gi . Note that the determinant of that matrix is
equal to the index (G : H).

3.13.4 Relative extensions.

We now have a look at data structures attached to relative extensions of number fields L/K,
and to projective ZK -modules. When defining a relative extension L/K, the nf attached to the
base field K must be defined by a variable having a lower priority (see Section 2.5.3) than the
variable defining the extension. For example, you may use the variable name y to define the base
field K, and x to define the relative extension L/K.

Basic definitions.
• rnf denotes a relative number field, i.e. a data structure output by rnfinit, attached to the
extension L/K. The nf attached to be base field K is rnf.nf.
• A relative matrix is an m × n matrix whose entries are elements of K, in any form. Its m
columns Aj represent elements in K n .
• An ideal list is a row vector of fractional ideals of the number field nf .
• A pseudo-matrix is a 2-component row vector (A, I) where A is a relative m × n matrix and
I an ideal list of length n. If I = {a1 , . . . , an } and the columns of A are (A1 , . . . , An ), this data
defines the torsion-free (projective) ZK -module a1 A1 ⊕ an An .
• An integral pseudo-matrix is a 3-component row vector w(A, I, J) where A = (ai,j ) is an
m × n relative matrix and I = (b1 , . . . , bm ), J = (a1 , . . . , an ) are ideal lists, such that ai,j ∈ bi a−1
j
for all i, j. This data defines two abstract projective ZK -modules N = a1 ω1 ⊕ · · · ⊕ an ωn in K n ,
P = b1 η1 ⊕ · · · ⊕ bm ηm in K m , and a ZK -linear map f : N → P given by
X X 
f( αj ωj ) = ai,j αj ηi .
i

This data defines the ZK -module M = P/f (N ).

• Any projective ZK -module M of finite type in K m can be given by a pseudo matrix (A, I).
• An arbitrary ZK modules of finite type in K m , with nontrivial torsion, is given by an integral
pseudo-matrix (A, I, J)

348
Algebraic numbers in relative extension.
We are given a number field K = nfinit(T ), attached to K = Q[Y ]/(T ), T ∈ Q[Y ], and
a relative extension L = rnfinit(K, P ), attached to L = K[X]/(P ), P ∈ K[X]. In all contexts
(except rnfeltabstorel, see below), an algebraic number is given as
• a t_INT, t_FRAC or t_POL in Q[Y ] (implicitly modulo T ) or a t_POL in K[X] (implicitly
modulo P ),
• a t_POLMOD (modulo T or P ), or
• a t_COL v of dimension m = [K : Q], representing the element in terms of the integral basis
K.zk;
• if an absolute nf structure Labs was attached to L, via Labs = nfinit(L), then we can also
use a t_COL v of dimension [L : Q], representing the element in terms of the computed integral basis
Labs.zk. Be careful that in the degenerate case L = K, then the previous interpretation (with
respect to K.zk) takes precedence. This is no concern when K = Q or if P = X − Y (because
in that case the primitive polynomial Labs.pol defining L of Q is nf.pol and the computation of
nf.zk is deterministic); but in other cases, the integer bases attached to K and Labs may differ.

Special case: rnfabstorel. This function assumes that elements are given in absolute represen-
tation (with respect to Labs.zk or modulo Labs.pol and converts them to relative representation
modulo L.pol. In that function (only), a t_POL in X is implicitly understood modulo Labs.pol
and a t_COL of length [L : Q] refers to the integral basis Labs.zk in all cases, including L = K.

Pseudo-bases, determinant.
• The pair (A, I) is a pseudo-basis of the module it generates if the aj are nonzero, and the Aj
are K-linearly independent. We call n the size of the pseudo-basis. If A is a relative matrix, the
latter condition means it is square with nonzero determinant; we say that it is in Hermite Normal
Form (HNF) if it is upper triangular and all the elements of the diagonal are equal to 1.
• For instance, the relative integer basis rnf.zk is a pseudo-basis (A, I) of ZL , where A =
rnf.zk[1] is a vector of elements of L, which are K-linearly independent. Most rnf routines return
and handle ZK -modules contained in L (e.g. ZL -ideals) via a pseudo-basis (A0 , I 0 ), where A0 is a
relative matrix representing a vector of elements of L in terms of the fixed basis rnf.zk[1]
• The determinant of a pseudo-basis (A, I) is the ideal equal to the product of the determinant
of A by all the ideals of I. The determinant of a pseudo-matrix is the determinant of any pseudo-
basis of the module it generates.

3.13.5 Class field theory.

A modulus, in the sense of class field theory, is a divisor supported on the real and finite places
of K. In PARI terms, this means either an ordinary ideal I as above (no Archimedean component),
or a pair [I, a], where a is a vector with r1 {0, 1}-components, corresponding to the infinite part of
the divisor. More precisely, the i-th component of a corresponds to the real embedding attached
to the i-th real root of K.roots. (That ordering is not canonical, but well defined once a defining
polynomial for K is chosen.) For instance, [1, [1,1]] is a modulus for a real quadratic field,
allowing ramification at any of the two places at infinity, and nowhere else.
A bid or “big ideal” is a structure output by idealstar needed to compute in (ZK /I)∗ , where
I is a modulus in the above sense. It is a finite abelian group as described above, supplemented by
technical data needed to solve discrete log problems.

349
 Finally we explain how to input ray number fields (or bnr ), using class field theory. These
are defined by a triple A, B, C, where the defining set [A, B, C] can have any of the following
forms: [bnr ], [bnr , subgroup], [bnr , character ], [bnf , mod ], [bnf , mod , subgroup]. The last two forms
are kept for backward compatibility, but no longer serve any real purpose (see example below); no
newly written function will accept them.
• bnf is as output by bnfinit, where units are mandatory unless the modulus is trivial; bnr
is as output by bnrinit. This is the ground field K.
• mod is a modulus f, as described above.
• subgroup a subgroup of the ray class group modulo f of K. As described above, this is input
as a square matrix expressing generators of a subgroup of the ray class group bnr .clgp on the
given generators. We also allow a t_INT n for n · Clf .
• character is a character χ of the ray class group modulo f, representing the subgroup Kerχ.
The corresponding bnr is the subfield of the ray class field of K modulo f, fixed by the given
subgroup.
? K = bnfinit(y^2+1);
? bnr = bnrinit(K, 13)
? %.clgp
%3 = [36, [12, 3]]
? bnrdisc(bnr); \\ discriminant of the full ray class field
? bnrdisc(bnr, [3,1;0,1]); \\ discriminant of cyclic cubic extension of K
? bnrconductor(bnr, [3,1]); \\ conductor of chi: g1->zeta_12^3, g2->zeta_3
We could have written directly
? bnrdisc(K, 13);
? bnrdisc(K, 13, [3,1;0,1]);
avoiding one bnrinit, but this would actually be slower since the bnrinit is called internally
anyway. And now twice!

3.13.6 General use.

All the functions which are specific to relative extensions, number fields, Buchmann’s number
fields, Buchmann’s number rays, share the prefix rnf, nf, bnf, bnr respectively. They take as first
argument a number field of that precise type, respectively output by rnfinit, nfinit, bnfinit,
and bnrinit.
However, and even though it may not be specified in the descriptions of the functions below,
it is permissible, if the function expects a nf , to use a bnf instead, which contains much more
information. On the other hand, if the function requires a bnf, it will not launch bnfinit for you,
which is a costly operation. Instead, it will give you a specific error message. In short, the types

nf ≤ bnf ≤ bnr

are ordered, each function requires a minimal type to work properly, but you may always substitute
a larger type.
The data types corresponding to the structures described above are rather complicated. Thus,
as we already have seen it with elliptic curves, GP provides “member functions” to retrieve data

350
from these structures (once they have been initialized of course). The relevant types of number
fields are indicated between parentheses:
bid (bnr ) : bid ideal structure.
bnf (bnr , bnf ) : Buchmann’s number field.
clgp (bnr , bnf ) : classgroup. This one admits the following three subclasses:
cyc : cyclic decomposition (SNF).
gen : generators.
no : number of elements.
diff (bnr , bnf , nf ) : the different ideal.
codiff (bnr , bnf , nf ) : the codifferent (inverse of the different in the ideal group).
disc (bnr , bnf , nf ) : discriminant.
fu ( bnf ) : fundamental units.
index (bnr , bnf , nf ) : index of the power order in the ring of integers.
mod (bnr ) : modulus.
nf (bnr , bnf , nf ) : number field.
pol (bnr , bnf , nf ) : defining polynomial.
r1 (bnr , bnf , nf ) : the number of real embeddings.
r2 (bnr , bnf , nf ) : the number of pairs of complex embeddings.
reg ( bnf ) : regulator.
roots (bnr , bnf , nf ) : roots of the polynomial generating the field.
sign (bnr , bnf , nf ) : signature [r1, r2].
t2 (bnr , bnf , nf ) : the T2 matrix (see nfinit).
tu ( bnf ) : a generator for the torsion units.
zk (bnr , bnf , nf ) : integral basis, i.e. a Z-basis of the maximal order.
zkst (bnr ) : structure of (ZK /m)∗ .
The member functions .codiff, .t2 and .zk perform a computation and are relatively ex-
pensive in large degree: move them out of tight loops and store them in variables.
For instance, assume that bnf = bnfinit(pol ), for some polynomial. Then bnf .clgp retrieves
the class group, and bnf .clgp.no the class number. If we had set bnf = nfinit(pol ), both would
have output an error message. All these functions are completely recursive, thus for instance
bnr .bnf.nf.zk will yield the maximal order of bnr , which you could get directly with a simple
bnr .zk.

3.13.7 Class group, units, and the GRH.

Some of the functions starting with bnf are implementations of the sub-exponential algorithms
for finding class and unit groups under GRH, due to Hafner-McCurley, Buchmann and Cohen-
Diaz-Olivier. The general call to the functions concerning class groups of general number fields
(i.e. excluding quadclassunit) involves a polynomial P and a technical vector

tech = [c1 , c2 , nrpid ],

where the parameters are to be understood as follows:

P is the defining polynomial for the number field, which must be in Z[X], irreducible and
monic. In fact, if you supply a nonmonic polynomial at this point, gp issues a warning, then
transforms your polynomial so that it becomes monic. The nfinit routine will return a different
result in this case: instead of res, you get a vector [res,Mod(a,Q)], where Mod(a,Q) = Mod(X,P)
gives the change of variables. In all other routines, the variable change is simply lost.

351
 The tech interface is obsolete and you should not tamper with these parameters. Indeed, from
version 2.4.0 on,

• the results are always rigorous under GRH (before that version, they relied on a heuristic
strengthening, hence the need for overrides).

• the influence of these parameters on execution time and stack size is marginal. They can
be useful to fine-tune and experiment with the bnfinit code, but you will be better off modifying
all tuning parameters in the C code (there are many more than just those three). We nevertheless
describe it for completeness.

The numbers c1 ≤ c2 are nonnegative real numbers. By default they are chosen so that the
result is correct under GRH. For i = 1, 2, let Bi = ci (log |dK |)2 , and denote by S(B) the set of
maximal ideals of K whose norm is less than B. We want S(B1 ) to generate Cl(K) and hope that
S(B2 ) can be proven to generate Cl(K).

More precisely, S(B1 ) is a factorbase used to compute a tentative Cl(K) by generators and
relations. We then check explicitly, using essentially bnfisprincipal, that the elements of S(B2 )
belong to the span of S(B1 ). Under the assumption that S(B2 ) generates Cl(K), we are done.
User-supplied ci are only used to compute initial guesses for the bounds Bi , and the algorithm
increases them until one can prove under GRH that S(B2 ) generates Cl(K). A uniform result of
Bach says that c2 = 12 is always suitable, but this bound is very pessimistic and a direct algorithm
due to Belabas-Diaz-Friedman is used to check the condition, assuming GRH. The default values
are c1 = c2 = 0. When c1 is equal to 0 the algorithm takes it equal to c2 .

nrpid is the maximal number of small norm relations attached to each ideal in the factor base.
Set it to 0 to disable the search for small norm relations. Otherwise, reasonable values are between
4 and 20. The default is 4.

Warning. Make sure you understand the above! By default, most of the bnf routines depend
on the correctness of the GRH. In particular, any of the class number, class group structure, class
group generators, regulator and fundamental units may be wrong, independently of each other.
Any result computed from such a bnf may be wrong. The only guarantee is that the units given
generate a subgroup of finite index in the full unit group. You must use bnfcertify to certify the
computations unconditionally.

Remarks.

You do not need to supply the technical parameters (under the library you still need to send
at least an empty vector, coded as NULL). However, should you choose to set some of them, they
must be given in the requested order. For example, if you want to specify a given value of nrpid ,
you must give some values as well for c1 and c2 , and provide a vector [c1 , c2 , nrpid ].

Note also that you can use an nf instead of P , which avoids recomputing the integral basis
and analogous quantities.

352
3.13.8 bnfcertify(bnf , {flag = 0}). bnf being as output by bnfinit, checks whether the result is
correct, i.e. whether it is possible to remove the assumption of the Generalized Riemann Hypothesis.
It is correct if and only if the answer is 1. If it is incorrect, the program may output some error
message, or loop indefinitely. You can check its progress by increasing the debug level. The bnf
structure must contain the fundamental units:
? K = bnfinit(x^3+2^2^3+1); bnfcertify(K)
*** at top-level: K=bnfinit(x^3+2^2^3+1);bnfcertify(K)
*** ^-------------
*** bnfcertify: precision too low in makeunits [use bnfinit(,1)].
? K = bnfinit(x^3+2^2^3+1, 1); \\ include units
? bnfcertify(K)
%3 = 1
If flag is present, only certify that the class group is a quotient of the one computed in bnf
(much simpler in general); likewise, the computed units may form a subgroup of the full unit group.
In this variant, the units are no longer needed:
? K = bnfinit(x^3+2^2^3+1); bnfcertify(K, 1)
%4 = 1
The library syntax is long bnfcertify0(GEN bnf, long flag). Also available is GEN bn-
fcertify(GEN bnf) (flag = 0).

3.13.9 bnfdecodemodule(nf , m). If m is a module as output in the first component of an

extension given by bnrdisclist, outputs the true module.
? K = bnfinit(x^2+23); L = bnrdisclist(K, 10); s = L[2]
%1 = [[[Vecsmall([8]), Vecsmall([1])], [[0, 0, 0]]],
[[Vecsmall([9]), Vecsmall([1])], [[0, 0, 0]]]]
? bnfdecodemodule(K, s[1][1])
%2 =
[2 0]
[0 1]
? bnfdecodemodule(K,s[2][1])
%3 =
[2 1]
[0 1]
The library syntax is GEN decodemodule(GEN nf, GEN m).

3.13.10 bnfinit(P, {flag = 0}, {tech = [ ]}). Initializes a bnf structure. Used in programs such as
bnfisprincipal, bnfisunit or bnfnarrow. By default, the results are conditional on the GRH,
see 3.13.7. The result is a 10-component vector bnf .
This implements Buchmann’s sub-exponential algorithm for computing the class group, the
regulator and a system of fundamental units of the general algebraic number field K defined by the
irreducible polynomial P with integer coefficients. The meaning of flagis as follows:
• flag = 0 (default). This is the historical behavior, kept for compatibility reasons and speed.
It has severe drawbacks but is likely to be a little faster than the alternative, twice faster say, so
only use it if speed is paramount, you obtain a useful speed gain for the fields under consideration,

353
and you are only interested in the field invariants such as the classgroup structure or its regulator.
The computations involve exact algebraic numbers which are replaced by floating point embeddings
for the sake of speed. If the precision is insufficient, gp may not be able to compute fundamental
units, nor to solve some discrete logarithm problems. It may be possible to increase the precision
of the bnf structure using nfnewprec but this may fail, in particular when fundamental units are
large. In short, the resulting bnf structure is correct and contains useful information but later
function calls to bnfisprincpal or bnrclassfield may fail.
When flag = 1, we keep an exact algebraic version of all floating point data and this allows
to guarantee that functions using the structure will always succeed, as well as to compute the
fundamental units exactly. The units are computed in compact form, as a product of small S-
units, possibly with huge exponents. This flag also allows bnfisprincipal to compute generators
of principal ideals in factored form as well. Be warned that expanding such products explicitly can
take a very long time, but they can easily be mapped to floating point or `-adic embeddings of
bounded accuracy, or to K ∗ /(K ∗ )` , and this is enough for applications. In short, this flag should
be used by default, unless you have a very good reason for it, for instance building massive tables
of class numbers, and you do not care about units or the effect large units would have on your
computation.
tech is a technical vector (empty by default, see 3.13.7). Careful use of this parameter may
speed up your computations, but it is mostly obsolete and you should leave it alone.

The components of a bnf are technical. In fact: never access a component directly, always
use a proper member function. However, for the sake of completeness and internal documentation,
their description is as follows. We use the notations explained in the book by H. Cohen, A Course
in Computational Algebraic Number Theory, Graduate Texts in Maths 138, Springer-Verlag, 1993,
Section 6.5, and subsection 6.5.5 in particular.
bnf [1] contains the matrix W , i.e. the matrix in Hermite normal form giving relations for the
class group on prime ideal generators (pi )1≤i≤r .
bnf [2] contains the matrix B, i.e. the matrix containing the expressions of the prime ideal
factorbase in terms of the pi . It is an r × c matrix.
bnf [3] contains the complex logarithmic embeddings of the system of fundamental units which
has been found. It is an (r1 + r2 ) × (r1 + r2 − 1) matrix.
bnf [4] contains the matrix MC00 of Archimedean components of the relations of the matrix
(W |B).
bnf [5] contains the prime factor base, i.e. the list of prime ideals used in finding the relations.
bnf [6] contains a dummy 0.
bnf [7] or bnf .nf is equal to the number field data nf as would be given by nfinit.
bnf [8] is a vector containing the classgroup bnf .clgp as a finite abelian group, the regulator
bnf .reg, the number of roots of unity and a generator bnf .tu, the fundamental units in expanded
form bnf .fu. If the fundamental units were omitted in the bnf , bnf .fu returns the sentinel value
0. If flag = 1, this vector contain also algebraic data corresponding to the fundamental units and
to the discrete logarithm problem (see bnfisprincipal). In particular, if flag = 1 we may only
know the units in factored form: the first call to bnf .fu expands them, which may be very costly,
then caches the result.

354
 bnf [9] is a vector used in bnfisprincipal only and obtained as follows. Let D = U W V
obtained by applying the Smith normal form algorithm to the matrix W (= bnf [1]) and let Ur
be the reduction of U modulo D. The first elements of the factorbase are given (in terms of
bnf.gen) by the columns of Ur , with Archimedean component ga ; let also GDa be the Archimedean
components of the generators of the (principal) ideals defined by the bnf.gen[i]^bnf.cyc[i].
Then bnf [9] = [Ur , ga , GDa ], followed by technical exact components which allow to recompute ga
and GDa to higher accuracy.

bnf [10] is by default unused and set equal to 0. This field is used to store further information
about the field as it becomes available, which is rarely needed, hence would be too expensive
to compute during the initial bnfinit call. For instance, the generators of the principal ideals
bnf.gen[i]^bnf.cyc[i] (during a call to bnrisprincipal), or those corresponding to the relations
in W and B (when the bnf internal precision needs to be increased).

The library syntax is GEN bnfinit0(GEN P, long flag, GEN tech = NULL, long prec)
.

Also available is GEN Buchall(GEN P, long flag, long prec), corresponding to tech =
NULL, where flag is either 0 (default) or nf_FORCE (include all data in algebraic form). The function
GEN Buchall_param(GEN P, double c1, double c2, long nrpid, long flag, long prec)
gives direct access to the technical parameters.

3.13.11 bnfisintnorm(bnf , x). Computes a complete system of solutions (modulo units of positive
norm) of the absolute norm equation Norm(a) = x, where a is an integer in bnf . If bnf has not
been certified, the correctness of the result depends on the validity of GRH.

See also bnfisnorm.

The library syntax is GEN bnfisintnorm(GEN bnf, GEN x). The function GEN bnfisint-
normabs(GEN bnf, GEN a) returns a complete system of solutions modulo units of the abso-
lute norm equation |Norm(x)| = |a|. As fast as bnfisintnorm, but solves the two equations
Norm(x) = ±a simultaneously.

3.13.12 bnfisnorm(bnf , x, {flag = 1}). Tries to tell whether the rational number x is the norm of
some element y in bnf . Returns a vector [a, b] where x = N orm(a) ∗ b. Looks for a solution which
is an S-unit, with S a certain set of prime ideals containing (among others) all primes dividing
x. If bnf is known to be Galois, you may set flag = 0 (in this case, x is a norm iff b = 1). If
flag is nonzero the program adds to S the following prime ideals, depending on the sign of flag. If
flag > 0, the ideals of norm less than flag. And if flag < 0 the ideals dividing flag.

Assuming GRH, the answer is guaranteed (i.e. x is a norm iff b = 1), if S contains all primes
less than 12 log(disc(Bnf ))2 , where Bnf is the Galois closure of bnf .

See also bnfisintnorm.

The library syntax is GEN bnfisnorm(GEN bnf, GEN x, long flag).

355
3.13.13 bnfisprincipal(bnf , x, {flag = 1}). bnf being the number field data output by bnfinit,
and x being an ideal, this function tests whether the ideal is principal or not. The result is more
complete than a simple true/false answer and solves a general discrete logarithm problem. Assume
the class group is ⊕(Z/di Z)gi (where the generators gi and their orders di are respectively given
by bnf.gen and bnf.cyc). The routine returns a row vector [e, t], where e is a vector of exponents
0 ≤ ei < di , and t is a number field element such that
Y
x = (t) giei .
i

For given gi (i.e. for a given bnf), the ei are unique, and t is unique modulo units.
In particular, x is principal if and only if e is the zero vector. Note that the empty vector,
which is returned when the class number is 1, is considered to be a zero vector (of dimension 0).
? K = bnfinit(y^2+23);
? K.cyc
%2 = [3]
? K.gen
%3 = [[2, 0; 0, 1]] \\ a prime ideal above 2
? P = idealprimedec(K,3)[1]; \\ a prime ideal above 3
? v = bnfisprincipal(K, P)
%5 = [[2]~, [3/4, 1/4]~]
? idealmul(K, v[2], idealfactorback(K, K.gen, v[1]))
%6 =
[3 0]
[0 1]
? % == idealhnf(K, P)
%7 = 1
The binary digits of flagmean:
• 1: If set, outputs [e, t] as explained above, otherwise returns only e, which is much easier to
compute. The following idiom only tests whether an ideal is principal:
is_principal(bnf, x) = !bnfisprincipal(bnf,x,0);
• 2: It may not be possible to recover t, given the initial accuracy to which the bnf structure
was computed. In that case, a warning is printed and t is set equal to the empty vector []~. If this
bit is set, increase the precision and recompute needed quantities until t can be computed. Warning:
setting this may induce lengthy computations and you should consider using flag 4 instead.
• 4: Return t in factored form (compact representation), as a small product of S-units for
a small set of finite places S, possibly with huge exponents. This kind of result can be cheaply
mapped to K ∗ /(K ∗ )` or to C or Qp to bounded accuracy and this is usually enough for applications.
Explicitly expanding such a compact representation is possible using nffactorback but may be
very costly. The algorithm is guaranteed to succeed if the bnf was computed using bnfinit(,1).
If not, the algorithm may fail to compute a huge generator in this case (and replace it by []~).
This is orders of magnitude faster than flag 2 when the generators are indeed large.
The library syntax is GEN bnfisprincipal0(GEN bnf, GEN x, long flag). Instead of the
above hardcoded numerical flags, one should rather use an or-ed combination of the symbolic flags
nf_GEN (include generators, possibly a place holder if too difficult), nf_GENMAT (include generators
in compact form) and nf_FORCE (insist on finding the generators, a no-op if nf_GENMAT is included).

356
3.13.14 bnfissunit(bnf , sfu, x). This function is obsolete, use bnfisunit.
The library syntax is GEN bnfissunit(GEN bnf, GEN sfu, GEN x).

3.13.15 bnfisunit(bnf , x, {U }). bnf being the number field data output by bnfinit and x being
an algebraic number (type integer, rational or polmod), this outputs the decomposition of x on
the fundamental units and the roots of unity if x is a unit, the empty vector otherwise. More
precisely, if u1 ,. . . ,ur are the fundamental units, and ζ is the generator of the group of roots of
unity (bnf.tu), the output is a vector [x1 , . . . , xr , xr+1 ] such that x = ux1 1 · · · uxr r · ζ xr+1 . The xi
are integers but the last one (i = r + 1) is only defined modulo the order w of ζ and is guaranteed
to be in [0, w[.
Note that bnf need not contain the fundamental units explicitly: it may contain the placeholder
0 instead:
? setrand(1); bnf = bnfinit(x^2-x-100000);
? bnf.fu
%2 = 0
? u = [119836165644250789990462835950022871665178127611316131167, \
379554884019013781006303254896369154068336082609238336]~;
? bnfisunit(bnf, u)
%3 = [-1, 0]~
The given u is 1/u1 , where u1 is the fundamental unit implicitly stored in bnf . In this case, u1
was not computed and stored in algebraic form since the default accuracy was too low. Re-run the
bnfinit command at \g1 or higher to see such diagnostics.
This function allows x to be given in factored form, but it then assumes that x is an actual
unit. (Because it is general too costly to check whether this is the case.)
? { v = [2, 85; 5, -71; 13, -162; 17, -76; 23, -37; 29, -104; [224, 1]~, -66;
[-86, 1]~, 86; [-241, 1]~, -20; [44, 1]~, 30; [124, 1]~, 11; [125, -1]~, -11;
[-214, 1]~, 33; [-213, -1]~, -33; [189, 1]~, 74; [190, -1]~, 104;
[-168, 1]~, 2; [-167, -1]~, -8]; }
? bnfisunit(bnf,v)
%5 = [1, 0]~
Note that v is the fundamental unit of bnf given in compact (factored) form.
If the argument U is present, as output by bnfunits(bnf, S), then the function decomposes
x on the S-units generators given in U[1].
? bnf = bnfinit(x^4 - x^3 + 4*x^2 + 3*x + 9, 1);
? bnf.sign
%2 = [0, 2]
? S = idealprimedec(bnf,5); #S
%3 = 2
? US = bnfunits(bnf,S);
? g = US[1]; #g \\ #S = #g, four S-units generators, in factored form
%5 = 4
? g[1]
%6 = [[6, -3, -2, -2]~ 1]
? g[2]

357
 %7 =
[[-1, 1/2, -1/2, -1/2]~ 1]
[ [4, -2, -1, -1]~ 1]
? [nffactorback(bnf, x) | x <- g]
%8 = [[6, -3, -2, -2]~, [-5, 5, 0, 0]~, [-1, 1, -1, 0]~,
[1, -1, 0, 0]~]
? u = [10,-40,24,11]~;
? a = bnfisunit(bnf, u, US)
%9 = [2, 0, 1, 4]~
? nffactorback(bnf, g, a) \\ prod_i g[i]^a[i] still in factored form
%10 =
[[6, -3, -2, -2]~ 2]
[ [0, 0, -1, -1]~ 1]
[ [2, -1, -1, 0]~ -2]
[ [1, 1, 0, 0]~ 2]
[ [-1, 1, 1, 1]~ -1]
[ [1, -1, 0, 0]~ 4]
? nffactorback(bnf,%) \\ u = prod_i g[i]^a[i]
%11 = [10, -40, 24, 11]~
The library syntax is GEN bnfisunit0(GEN bnf, GEN x, GEN U = NULL). Also available is
GEN bnfisunit(GEN bnf, GEN x) for U = NULL.

3.13.16 bnflog(bnf , l). Let bnf be a bnf structure attached to the number field F and let l be a
prime number (hereafter denoted ` for typographical reasons). Return the logarithmic `-class group
Cl
f F of F . This is an abelian group, conjecturally finite (known to be finite if F/Q is abelian). The
function returns if and only if the group is indeed finite (otherwise it would run into an infinite
loop). Let S = {p1 , . . . , pk } be the set of `-adic places (maximal ideals containing `). The function
returns [D, G(`), G0 ], where

• D is the vector of elementary divisors for Cl

fF .

• G(`) is the vector of elementary divisors for the (conjecturally finite) abelian group
X
Cl(`) = {a =
f ai pi : degF a = 0},
i≤k

where the pi are the `-adic places of F ; this is a subgroup of f

Cl.
• G0 is the vector of elementary divisors for the `-Sylow Cl0 of the S-class group of F ; the
Cl maps to Cl0 with a simple co-kernel.
group f
The library syntax is GEN bnflog(GEN bnf, GEN l).

358
3.13.17 bnflogdegree(nf , A, l). Let nf be a nf structure attached to a number field F , and let
l be a prime number (hereafter denoted `). The `-adified group of idèles of F quotiented by the
group of logarithmic units is identified to the `-group of logarithmic divisors ⊕Z` [p], generated by
the maximal ideals of F .
The degree map degF is additive with values in Z` , defined by degF p = f˜p deg` p, where the
integer f˜p is as in bnflogef and deg` p is log` p for p 6= `, log` (1 + `) for p = ` 6= 2 and log` (1 + 22 )
for p = ` = 2.
Let A = pnp be an ideal and let à =
Q P
np [p] be the attached logarithmic divisor. Return
the exponential of the `-adic logarithmic degree degF A, which is a natural number.
The library syntax is GEN bnflogdegree(GEN nf, GEN A, GEN l).

3.13.18 bnflogef(nf , pr ). Let nf be a nf structure attached to a number field F and let pr be

a prid structure attached to a maximal ideal p/p. Return [ẽ(Fp /Qp ), f˜(Fp /Qp )] the logarithmic
ramification and residue degrees. Let Qcp /Qp be the cyclotomic Zp -extension, then ẽ = [Fp : Fp ∩Qcp ]
and f˜ = [Fp ∩ Qcp : Qp ]. Note that ẽf˜ = e(p/p)f (p/p), where e(p/p) and f (p/p) denote the usual
ramification and residue degrees.
? F = nfinit(y^6 - 3*y^5 + 5*y^3 - 3*y + 1);
? bnflogef(F, idealprimedec(F,2)[1])
%2 = [6, 1]
? bnflogef(F, idealprimedec(F,5)[1])
%3 = [1, 2]
The library syntax is GEN bnflogef(GEN nf, GEN pr).

3.13.19 bnfnarrow(bnf ). bnf being as output by bnfinit, computes the narrow class group
of bnf . The output is a 3-component row vector v analogous to the corresponding class group
component bnf .clgp: the first component is the narrow class number v.no, the second component
is a vector containing the SNF cyclic components v.cyc of the narrow class group, and the third
is a vector giving the generators of the corresponding v.gen cyclic groups. Note that this function
is a special case of bnrinit; the bnf need not contain fundamental units.
The library syntax is GEN bnfnarrow(GEN bnf).

3.13.20 bnfsignunit(bnf ). bnf being as output by bnfinit, this computes an r1 × (r1 + r2 − 1)

matrix having ±1 components, giving the signs of the real embeddings of the fundamental units.
The following functions compute generators for the totally positive units:
/* exponents of totally positive units generators on K.tu, K.fu */
tpuexpo(K)=
{ my(M, S = bnfsignunit(K), [m,n] = matsize(S));
\\ m = K.r1, n = r1+r2-1
S = matrix(m,n, i,j, if (S[i,j] < 0, 1,0));
S = concat(vectorv(m,i,1), S); \\ add sign(-1)
M = matkermod(S, 2);
if (M, mathnfmodid(M, 2), 2*matid(n+1))
}
/* totally positive fundamental units of bnf K */
tpu(K)=

359
 { my(ex = tpuexpo(K)[,^1]); \\ remove ex[,1], corresponds to 1 or -1
my(v = concat(K.tu[2], K.fu));
[ nffactorback(K, v, c) | c <- ex];
}
The library syntax is GEN signunits(GEN bnf).

3.13.21 bnfsunit(bnf , S). Computes the fundamental S-units of the number field bnf (output by
bnfinit), where S is a list of prime ideals (output by idealprimedec). The output is a vector v
with 6 components.
v[1] gives a minimal system of (integral) generators of the S-unit group modulo the unit group.
v[2] contains technical data needed by bnfissunit.
v[3] is an obsoleted component, now the empty vector.
v[4] is the S-regulator (this is the product of the regulator, the S-class number and the natural
logarithms of the norms of the ideals in S).
v[5] gives the S-class group structure, in the usual abelian group format: a vector whose three
components give in order the S-class number, the cyclic components and the generators.
v[6] is a copy of S.
The library syntax is GEN bnfsunit(GEN bnf, GEN S, long prec).

3.13.22 bnfunits(bnf , {S}). Return the fundamental units of the number field bnf output by
bnfinit; if S is present and is a list of prime ideals, compute fundamental S-units instead. The
first component of the result contains independent integral S-units generators: first nonunits, then
r1 +r2 −1 fundamental units, then the torsion unit. The result may be used as an optional argument
to bnfisunit. The units are given in compact form: no expensive computation is attempted if the
bnf does not already contain units.
? bnf = bnfinit(x^4 - x^3 + 4*x^2 + 3*x + 9, 1);
? bnf.sign \\ r1 + r2 - 1 = 1
%2 = [0, 2]
? U = bnfunits(bnf); u = U[1];
? #u \\ r1 + r2 = 2 units
%5 = 2;
? u[1] \\ fundamental unit as factorization matrix
%6 =
[[0, 0, -1, -1]~ 1]
[[2, -1, -1, 0]~ -2]
[ [1, 1, 0, 0]~ 2]
[ [-1, 1, 1, 1]~ -1]
? u[2] \\ torsion unit as factorization matrix
%7 =
[[1, -1, 0, 0]~ 1]
? [nffactorback(bnf, z) | z <- u] \\ same units in expanded form
%8 = [[-1, 1, -1, 0]~, [1, -1, 0, 0]~]

360
 Now an example involving S-units for a nontrivial S:
? S = idealprimedec(bnf,5); #S
%9 = 2
? US = bnfunits(bnf, S); uS = US[1];
? g = [nffactorback(bnf, z) | z <- uS] \\ now 4 units
%11 = [[6, -3, -2, -2]~, [-5, 5, 0, 0]~, [-1, 1, -1, 0]~, [1, -1, 0, 0]~]
? bnfisunit(bnf,[10,-40,24,11]~)
%12 = []~ \\ not a unit
? e = bnfisunit(bnf, [10,-40,24,11]~, US)
%13 = [2, 0, 1, 4]~ \\ ...but an S-unit
? nffactorback(bnf, g, e)
%14 = [10, -40, 24, 11]~
? nffactorback(bnf, uS, e) \\ in factored form
%15 =
[[6, -3, -2, -2]~ 2]
[ [0, 0, -1, -1]~ 1]
[ [2, -1, -1, 0]~ -2]
[ [1, 1, 0, 0]~ 2]
[ [-1, 1, 1, 1]~ -1]
[ [1, -1, 0, 0]~ 4]

Note that in more complicated cases, any nffactorback fully expanding an element in factored
form could be very expensive. On the other hand, the final example expands a factorization whose
components are themselves in factored form, hence the result is a factored form: this is a cheap
operation.
The library syntax is GEN bnfunits(GEN bnf, GEN S = NULL).

3.13.23 bnrL1(bnr , {H}, {flag = 0}). Let bnr be the number field data output by bnrinit and
H be a square matrix defining a congruence subgroup of the ray class group corresponding to bnr
(the trivial congruence subgroup if omitted). This function returns, for each character χ of the ray
class group which is trivial on H, the value at s = 1 (or s = 0) of the abelian L-function attached
to χ. For the value at s = 0, the function returns in fact for each χ a vector [rχ , cχ ] where

L(s, χ) = c · sr + O(sr+1 )

near 0.
The argument flag is optional, its binary digits mean 1: compute at s = 0 if unset or s = 1 if
set, 2: compute the primitive L-function attached to χ if unset or the L-function with Euler factors
at prime ideals dividing the modulus of bnr removed if set (that is LS (s, χ), where S is the set
of infinite places of the number field together with the finite prime ideals dividing the modulus of
bnr ), 3: return also the character if set.
K = bnfinit(x^2-229);
bnr = bnrinit(K,1);
bnrL1(bnr)

361
returns the order and the first√nonzero term of L(s, χ) at s = 0 where χ runs through the characters
of the class group of K = Q( 229). Then
bnr2 = bnrinit(K,2);
bnrL1(bnr2,,2)
returns the order and the first nonzero terms of LS (s, χ) at s = 0 where χ runs through the
characters of the class group of K and S is the set of infinite places of K together with the finite
prime 2. Note that the ray class group modulo 2 is in fact the class group, so bnrL1(bnr2,0)
returns the same answer as bnrL1(bnr,0).
This function will fail with the message
*** bnrL1: overflow in zeta_get_N0 [need too many primes].
if the approximate functional equation requires us to sum too many terms (if the discriminant of
K is too large).
The library syntax is GEN bnrL1(GEN bnr, GEN H = NULL, long flag, long prec).

3.13.24 bnrchar(bnr , g, {v}). Returns all characters χ on bnr.clgp such that χ(gi ) = e(vi ), where
e(x) = exp(2iπx). If v is omitted, returns all characters that are trivial on the gi . Else the vectors
g and v must have the same length, the gi must be ideals in any form, and each vi is a rational
number whose denominator must divide the order of gi in the ray class group. For convenience,
the vector of the gi can be replaced by a matrix whose columns give their discrete logarithm, as
given by bnrisprincipal; this allows to specify abstractly a subgroup of the ray class group.
? bnr = bnrinit(bnfinit(x), [160,[1]], 1); /* (Z/160Z)^* */
? bnr.cyc
%2 = [8, 4, 2]
? g = bnr.gen;
? bnrchar(bnr, g, [1/2,0,0])
%4 = [[4, 0, 0]] \\ a unique character
? bnrchar(bnr, [g[1],g[3]]) \\ all characters trivial on g[1] and g[3]
%5 = [[0, 1, 0], [0, 2, 0], [0, 3, 0], [0, 0, 0]]
? bnrchar(bnr, [1,0,0;0,1,0;0,0,2])
%6 = [[0, 0, 1], [0, 0, 0]] \\ characters trivial on given subgroup
The library syntax is GEN bnrchar(GEN bnr, GEN g, GEN v = NULL).

3.13.25 bnrclassfield(bnr , {subgp}, {flag = 0}). bnr being as output by bnrinit, returns a
relative equation for the class field corresponding to the congruence group defined by (bnr , subgp)
(the full ray class field if subgp is omitted). The subgroup can also be a t_INT n, meaning n · Clf .
The function also handles a vector of subgroup, e.g, from subgrouplist and returns the vector of
individual results in this case.
If flag = 0, returns a vector of polynomials such that the compositum of the corresponding
fields is the class field; if flag = 1 returns a single polynomial; if flag = 2 returns a single absolute
polynomial.
? bnf = bnfinit(y^3+14*y-1); bnf.cyc
%1 = [4, 2]
? pol = bnrclassfield(bnf,,1) \\ Hilbert class field
%2 = x^8 - 2*x^7 + ... + Mod(11*y^2 - 82*y + 116, y^3 + 14*y - 1)

362
 ? rnfdisc(bnf,pol)[1]
%3 = 1
? bnr = bnrinit(bnf,3*5*7); bnr.cyc
%4 = [24, 12, 12, 2]
? bnrclassfield(bnr,2) \\ maximal 2-elementary subextension
%5 = [x^2 + (-21*y - 105), x^2 + (-5*y - 25), x^2 + (-y - 5), x^2 + (-y - 1)]
\\ quadratic extensions of maximal conductor
? bnrclassfield(bnr, subgrouplist(bnr,[2]))
%6 = [[x^2 - 105], [x^2 + (-105*y^2 - 1260)], [x^2 + (-105*y - 525)],
[x^2 + (-105*y - 105)]]
? #bnrclassfield(bnr,subgrouplist(bnr,[2],1)) \\ all quadratic extensions
%7 = 15

When the subgroup contains nClf , where n is fixed, it is advised to directly compute the bnr
modulo n to avoid expensive discrete logarithms:

? bnf = bnfinit(y^2-5); p = 1594287814679644276013;

? bnr = bnrinit(bnf,p); \\ very slow
time = 24,146 ms.
? bnrclassfield(bnr, 2) \\ ... even though the result is trivial
%3 = [x^2 - 1594287814679644276013]
? bnr2 = bnrinit(bnf,p,,2); \\ now fast
time = 1 ms.
? bnrclassfield(bnr2, 2)
%5 = [x^2 - 1594287814679644276013]

This will save a lot of time when the modulus contains a maximal ideal whose residue field is large.

The library syntax is GEN bnrclassfield(GEN bnr, GEN subgp = NULL, long flag, long
prec).

3.13.26 bnrclassno(A, {B}, {C}). Let A, B, C define a class field L over a ground field K (of
type [bnr ], [bnr , subgroup], or [bnf , modulus], or [bnf , modulus,subgroup], Section 3.13.5);
this function returns the relative degree [L : K].

In particular if A is a bnf (with units), and B a modulus, this function returns the correspond-
ing ray class number modulo B. One can input the attached bid (with generators if the subgroup
C is non trivial) for B instead of the module itself, saving some time.

This function is faster than bnrinit and should be used if only the ray class number is desired.
See bnrclassnolist if you need ray class numbers for all moduli less than some bound.

The library syntax is GEN bnrclassno0(GEN A, GEN B = NULL, GEN C = NULL). Also avail-
able is GEN bnrclassno(GEN bnf, GEN f) to compute the ray class number modulo f .

363
3.13.27 bnrclassnolist(bnf , list). bnf being as output by bnfinit, and list being a list of moduli
(with units) as output by ideallist or ideallistarch, outputs the list of the class numbers of the
corresponding ray class groups. To compute a single class number, bnrclassno is more efficient.

? bnf = bnfinit(x^2 - 2);

? L = ideallist(bnf, 100, 2);
? H = bnrclassnolist(bnf, L);
? H[98]
%4 = [1, 3, 1]
? l = L[1][98]; ids = vector(#l, i, l[i].mod[1])
%5 = [[98, 88; 0, 1], [14, 0; 0, 7], [98, 10; 0, 1]]

The weird l[i].mod[1], is the first component of l[i].mod, i.e. the finite part of the con-
ductor. (This is cosmetic: since by construction the Archimedean part is trivial, I do not want to
see it). This tells us that the ray class groups modulo the ideals of norm 98 (printed as %5) have
respectively order 1, 3 and 1. Indeed, we may check directly:

? bnrclassno(bnf, ids[2])
%6 = 3

The library syntax is GEN bnrclassnolist(GEN bnf, GEN list).

3.13.28 bnrconductor(A, {B}, {C}, {flag = 0}). Conductor f of the subfield of a ray class field
as defined by [A, B, C] (of type [bnr ], [bnr , subgroup], [bnf , modulus] or [bnf , modulus,
subgroup], Section 3.13.5)

If flag = 0, returns f .

If flag = 1, returns [f, Clf , H], where Clf is the ray class group modulo f , as a finite abelian
group; finally H is the subgroup of Clf defining the extension.

If flag = 2, returns [f, bnr (f ), H], as above except Clf is replaced by a bnr structure, as output
by bnrinit(, f ), without generators unless the input contained a bnr with generators.

In place of a subgroup H, this function also accepts a character chi = (aj ), expressed as usual
in terms of the generators bnr.gen: χ(gj ) = exp(2iπaj /dj ), where gj has order dj = bnr.cyc[j].
In which case, the function returns respectively

If flag = 0, the conductor f of Kerχ.

If flag = 1, [f, Clf , χf ], where χf is χ expressed on the minimal ray class group, whose modulus
is the conductor.

If flag = 2, [f, bnr (f ), χf ].

364
Note. Using this function with flag 6= 0 is usually a bad idea and kept for compatibility and
convenience only: flag = 1 has always been useless, since it is no faster than flag = 2 and returns
less information; flag = 2 is mostly OK with two subtle drawbacks:
• it returns the full bnr attached to the full ray class group, whereas in applications we only
need Clf modulo N -th powers, where N is any multiple of the exponent of Clf /H. Computing
directly the conductor, then calling bnrinit with optional argument N avoids this problem.
• computing the bnr needs only be done once for each conductor, which is not possible using
this function.
For maximal efficiency, the recommended procedure is as follows. Starting from data (character
or congruence subgroups) attached to a modulus m, we can first compute the conductors using this
function with default flag = 0. Then for all data with a common conductor f | m, compute (once!)
the bnr attached to f using bnrinit (modulo N -th powers for a suitable N !) and finally map
original data to the new bnr using bnrmap.
The library syntax is GEN bnrconductor0(GEN A, GEN B = NULL, GEN C = NULL, long
flag).
Also available are GEN bnrconductor(GEN bnr, GEN H, long flag) and GEN bnrconduc-
tormod(GEN bnr, GEN H, long flag, GEN cycmod) which returns ray class groups modulo
cycmod-th powers.

3.13.29 bnrconductorofchar(bnr , chi ). This function is obsolete, use bnrconductor.

The library syntax is GEN bnrconductorofchar(GEN bnr, GEN chi).

3.13.30 bnrdisc(A, {B}, {C}, {flag = 0}). A, B, C defining a class field L over a ground field K (of
type [bnr ], [bnr , subgroup], [bnr , character ], [bnf , modulus] or [bnf , modulus, subgroup],
Section 3.13.5), outputs data [N, r1 , D] giving the discriminant and signature of L, depending on
the binary digits of flag:
• 1: if this bit is unset, output absolute data related to L/Q: N is the absolute degree [L : Q],
r1 the number of real places of L, and D the discriminant of L/Q. Otherwise, output relative data
for L/K: N is the relative degree [L : K], r1 is the number of real places of K unramified in L (so
that the number of real places of L is equal to r1 times N ), and D is the relative discriminant ideal
of L/K.
• 2: if this bit is set and if the modulus is not the conductor of L, only return 0.
The library syntax is GEN bnrdisc0(GEN A, GEN B = NULL, GEN C = NULL, long flag)
.

3.13.31 bnrdisclist(bnf , bound , {arch}). bnf being as output by bnfinit (with units), computes
a list of discriminants of Abelian extensions of the number field by increasing modulus norm up to
bound bound . The ramified Archimedean places are given by arch; all possible values are taken if
arch is omitted.
The alternative syntax bnrdisclist(bnf , list) is supported, where list is as output by ideal-
list or ideallistarch (with units), in which case arch is disregarded.
The output v is a vector, where v[k] is itself a vector w, whose length is the number of ideals
of norm k.

365
 • We consider first the case where arch was specified. Each component of w corresponds to
an ideal m of norm k, and gives invariants attached to the ray class field L of bnf of conductor
[m, arch]. Namely, each contains a vector [m, d, r, D] with the following meaning: m is the prime
ideal factorization of the modulus, d = [L : Q] is the absolute degree of L, r is the number of real
places of L, and D is the factorization of its absolute discriminant. We set d = r = D = 0 if m is
not the finite part of a conductor.
• If arch was omitted, all t = 2r1 possible values are taken and a component of w has the form
[m, [[d1 , r1 , D1 ], . . . , [dt , rt , Dt ]]], where m is the finite part of the conductor as above, and [di , ri , Di ]
are the invariants of the ray class field of conductor [m, vi ], where vi is the i-th Archimedean
component, ordered by inverse lexicographic order; so v1 = [0, . . . , 0], v2 = [1, 0 . . . , 0], etc. Again,
we set di = ri = Di = 0 if [m, vi ] is not a conductor.
Finally, each prime ideal pr = [p, α, e, f, β] in the prime factorization m is coded as the integer
p · n2 + (f − 1) · n + (j − 1), where n is the degree of the base field and j is such that
pr = idealprimedec(nf ,p)[j].
m can be decoded using bnfdecodemodule.
Note that to compute such data for a single field, either bnrclassno or bnrdisc are (much)
more efficient.
The library syntax is GEN bnrdisclist0(GEN bnf, GEN bound, GEN arch = NULL).

3.13.32 bnrgaloisapply(bnr , mat, H). Apply the automorphism given by its matrix mat to the
congruence subgroup H given as a HNF matrix. The matrix mat can be computed with bnrgalo-
ismatrix.
The library syntax is GEN bnrgaloisapply(GEN bnr, GEN mat, GEN H).

3.13.33 bnrgaloismatrix(bnr , aut). Return the matrix of the action of the automorphism aut
of the base field bnf.nf on the generators of the ray class field bnr.gen; aut can be given as a
polynomial, an algebraic number, or a vector of automorphisms or a Galois group as output by
galoisinit, in which case a vector of matrices is returned (in the later case, only for the generators
aut.gen).
The generators bnr.gen need not be explicitly computed in the input bnr , which saves time:
the result is well defined in this case also.
? K = bnfinit(a^4-3*a^2+253009); B = bnrinit(K,9); B.cyc
%1 = [8400, 12, 6, 3]
? G = nfgaloisconj(K)
%2 = [-a, a, -1/503*a^3 + 3/503*a, 1/503*a^3 - 3/503*a]~
? bnrgaloismatrix(B, G[2]) \\ G[2] = Id ...
%3 =
[1 0 0 0]
[0 1 0 0]
[0 0 1 0]
[0 0 0 1]
? bnrgaloismatrix(B, G[3]) \\ automorphism of order 2
%4 =

366
 [799 0 0 2800]
[ 0 7 0 4]
[ 4 0 5 2]
[ 0 0 0 2]
? M = %^2; for (i=1, #B.cyc, M[i,] %= B.cyc[i]); M
%5 = \\ acts on ray class group as automorphism of order 2
[1 0 0 0]
[0 1 0 0]
[0 0 1 0]
[0 0 0 1]
See bnrisgalois for further examples.
The library syntax is GEN bnrgaloismatrix(GEN bnr, GEN aut). When aut is a polynomial
or an algebraic number, GEN bnrautmatrix(GEN bnr, GEN aut) is available.

3.13.34 bnrinit(bnf , f, {flag = 0}, {cycmod }). bnf is as output by bnfinit (including fundamen-
tal units), f is a modulus, initializes data linked to the ray class group structure corresponding
to this module, a so-called bnr structure. One can input the attached bid with generators for f
instead of the module itself, saving some time. (As in idealstar, the finite part of the conductor
may be given by a factorization into prime ideals, as produced by idealfactor.)
If the positive integer cycmod is present, only compute the ray class group modulo cycmod,
which may save a lot of time when some maximal ideals in f have a huge residue field. In appli-
cations, we are given a congruence subgroup H and study the class field attached to Clf /H. If
that finite Abelian group has an exponent which divides cycmod, then we have changed nothing
theoretically, while trivializing expensive discrete logs in residue fields (since computations can be
made modulo cycmod-th powers). This is useful in bnrclassfield, for instance when computing
p-elementary extensions.
The following member functions are available on the result: .bnf is the underlying bnf , .mod
the modulus, .bid the bid structure attached to the modulus; finally, .clgp, .no, .cyc, .gen
refer to the ray class group (as a finite abelian group), its cardinality, its elementary divisors, its
generators (only computed if flag = 1).
The last group of functions are different from the members of the underlying bnf , which refer to
the class group; use bnr .bnf.xxx to access these, e.g. bnr .bnf.cyc to get the cyclic decomposition
of the class group.
They are also different from the members of the underlying bid , which refer to (ZK /f )∗ ; use
bnr .bid.xxx to access these, e.g. bnr .bid.no to get φ(f ).
If flag = 0 (default), the generators of the ray class group are not explicitly computed, which
saves time. Hence bnr .gen would produce an error. Note that implicit generators are still fixed
and stored in the bnr (and guaranteed to be the same for fixed bnf and bid inputs), in terms of
bnr.bnf.gen and bnr.bid.gen. The computation which is not performed is the expansion of such
products in the ray class group so as to fix eplicit ideal representatives.
If flag = 1, as the default, except that generators are computed.
The library syntax is GEN bnrinitmod(GEN bnf, GEN f, long flag, GEN cycmod = NULL)
. Instead of the above hardcoded numerical flags, one should rather use GEN Buchraymod(GEN

367
bnf, GEN module, long flag, GEN cycmod) where an omitted cycmod is coded as NULL and
flag is an or-ed combination of nf GEN (include generators) and nf INIT (if omitted, return just
the cardinality of the ray class group and its structure), possibly 0. Or simply GEN Buchray(GEN
bnf, GEN module, long flag) when cycmod is NULL.

3.13.35 bnrisconductor(A, {B}, {C}). Fast variant of bnrconductor(A, B, C); A, B, C represent

an extension of the base field, given by class field theory (see Section 3.13.5). Outputs 1 if this
modulus is the conductor, and 0 otherwise. This is slightly faster than bnrconductor when the
character or subgroup is not primitive.
The library syntax is long bnrisconductor0(GEN A, GEN B = NULL, GEN C = NULL).

3.13.36 bnrisgalois(bnr , gal , H). Check whether the class field attached to the subgroup H
is Galois over the subfield of bnr.nf fixed by the group gal , which can be given as output by
galoisinit, or as a matrix or a vector of matrices as output by bnrgaloismatrix, the second
option being preferable, since it saves the recomputation of the matrices. Note: The function
assumes that the ray class field attached to bnr is Galois, which is not checked.
In the following example, we lists the congruence subgroups of subextension of degree at most
3 of the ray class field of conductor 9 which are Galois over the rationals.
? K = bnfinit(a^4-3*a^2+253009); B = bnrinit(K,9); G = galoisinit(K);
? [H | H<-subgrouplist(B,3), bnrisgalois(B,G,H)];
time = 160 ms.
? M = bnrgaloismatrix(B,G);
? [H | H<-subgrouplist(B,3), bnrisgalois(B,M,H)]
time = 1 ms.
The second computation is much faster since bnrgaloismatrix(B,G) is computed only once.
The library syntax is long bnrisgalois(GEN bnr, GEN gal, GEN H).

3.13.37 bnrisprincipal(bnr , x, {flag = 1}). Let bnr be the ray class group data output by
bnrinit(, , 1) and let x be an ideal in any form, coprime to the modulus f = bnr.mod. Solves
the discrete logarithm problem in the ray class group, with respect to the generators bnr.gen, in
a way similar to bnfisprincipal. If x is not coprime to the modulus of bnr the result is unde-
fined. Note that bnr need not contain the ray class group generators, i.e. it may be created with
bnrinit(, , 0); in that case, although bnr.gen is undefined, we can still fix natural generators for
the ray class group (in terms of the generators in bnr.bnf.gen and bnr.bid.gen) and compute
with respect to them.
The binary digits of flag (default flag = 1) mean:
• 1: If set returns a 2-component vector [e, α] where e is the vector of components
Q eof x on the
∗
ray class group generators, α is an element congruent to 1 mod f such that x = α i gi . If unset,
i

returns only e.
• 4: If set, returns [e, α] where α is given in factored form (compact representation). This is
orders of magnitude faster.
? K = bnfinit(x^2 - 30); bnr = bnrinit(K, [4, [1,1]]);
? bnr.clgp \\ ray class group is isomorphic to Z/4 x Z/2 x Z/2
%2 = [16, [4, 2, 2]]
? P = idealprimedec(K, 3)[1]; \\ the ramified prime ideal above 3

368
 ? bnrisprincipal(bnr,P) \\ bnr.gen undefined !
%5 = [[3, 0, 0]~, 9]
? bnrisprincipal(bnr,P, 0) \\ omit principal part
%5 = [3, 0, 0]~
? bnr = bnrinit(bnr, bnr.bid, 1); \\ include explicit generators
? bnrisprincipal(bnr,P) \\ ... alpha is different !
%7 = [[3, 0, 0]~, 1/128625]
It may be surprising that the generator α is different although the underlying bnf and bid
are the same. This defines unique generators for the ray class group as ideal classes, whether we
use bnrinit(,0) or bnrinit(,1). But the actual ideal representatives (implicit if the flag is 0,
computed and stored in the bnr if the flag is 1) are in general different and this is what happens
here. Indeed, the implicit generators are naturally expressed expressed in terms of bnr.bnf.gen
and bnr.bid.gen and then expanded and simplified (in the same ideal class) so that we obtain
ideal representatives for bnr.gen which are as simple as possible. And indeed the quotient of the
two α found is 1 modulo the conductor (and positive at the infinite places it contains), and this is
the only guaranteed property.
Beware that, when bnr is generated using bnrinit(, cycmod), the results are given in Clf
modulo cycmod-th powers:
? bnr2 = bnrinit(K, bnr.mod,, 2); \\ modulo squares
? bnr2.clgp
%9 = [8, [2, 2, 2]] \\ bnr.clgp tensored by Z/2Z
? bnrisprincipal(bnr2,P, 0)
%10 = [1, 0, 0]~
The library syntax is GEN bnrisprincipal(GEN bnr, GEN x, long flag). Instead of hard-
coded numerical flags, one should rather use GEN isprincipalray(GEN bnr, GEN x) for flag = 0,
and if you want generators:
bnrisprincipal(bnr, x, nf_GEN)
Also available is GEN bnrisprincipalmod(GEN bnr, GEN x, GEN mod, long flag) that
returns the discrete logarithm of x modulo the t_INT mod; the value mod = NULL is treated as 0
(full discrete logarithm), and flag = 1 is not allowed if mod is set.

3.13.38 bnrmap(A, B). This function has two different uses:

• if A and B are bnr structures for the same bnf attached to moduli mA and mB with mB | mA ,
return the canonical surjection from A to B, i.e. from the ray class group moodulo mA to the ray
class group modulo mB . The map is coded by a triple [M, cyc A , cyc B ]: M gives the image of
the fixed ray class group generators of A in terms of the ones in B, cyc A and cyc B are the cyclic
structures A.cyc and B.cyc respectively. Note that this function does not need A or B to contain
explicit generators for the ray class groups: they may be created using bnrinit(,0).
If B is only known modulo N -th powers (from bnrinit(,N)), the result is correct provided N
is a multiple of the exponent of A.
• if A is a projection map as above and B is either a congruence subgroup H, or a ray class
character χ, or a discrete logarithm (from bnrisprincipal) modulo mA whose conductor divides
mB , return the image of the subgroup (resp. the character, the discrete logarighm) as defined
modulo mB . The main use of this variant is to compute the primitive subgroup or character

369
attached to a bnr modulo their conductor. This is more efficient than bnrconductor in two respects:
the bnr attached to the conductor need only be computed once and, most importantly, the ray class
group can be computed modulo N -th powers, where N is a multiple of the exponent of ClmA /H
(resp. of the order of χ). Whereas bnrconductor is specified to return a bnr attached to the
full ray class group, which may lead to untractable discrete logarithms in the full ray class group
instead of a tiny quotient.
The library syntax is GEN bnrmap(GEN A, GEN B).

3.13.39 bnrrootnumber(bnr , chi , {flag = 0}). If χ = chi is a character over bnr , not necessarily
primitive, let L(s, χ) = id χ(id)N (id)−s be the attached Artin L-function. Returns the so-called
P
Artin root number, i.e. the complex number W (χ) of modulus 1 such that

Λ(1 − s, χ) = W (χ)Λ(s, χ)

where Λ(s, χ) = A(χ)s/2 γχ (s)L(s, χ) is the enlarged L-function attached to L.

You can set flag = 1 if the character is known to be primitive. Example:
bnf = bnfinit(x^2 - x - 57);
bnr = bnrinit(bnf, [7,[1,1]]);
bnrrootnumber(bnr, [2,1])
√
returns the root number of the character χ of Cl7∞1 ∞2 (Q( 229)) defined by χ(g1a g2b ) = ζ12a ζ2b . Here
g1 , g2 are the generators of the ray-class group given by bnr.gen and ζ1 = e2iπ/N1 , ζ2 = e2iπ/N2
where N1 , N2 are the orders of g1 and g2 respectively (N1 = 6 and N2 = 3 as bnr.cyc readily tells
us).
The library syntax is GEN bnrrootnumber(GEN bnr, GEN chi, long flag, long prec)
.

3.13.40 bnrstark(bnr , {subgroup}). bnr being as output by bnrinit, finds a relative equation for
the class field corresponding to the modulus in bnr and the given congruence subgroup (as usual,
omit subgroup if you want the whole ray class group).
The main variable of bnr must not be x, and the ground field and the class field must be
totally real. When the base field is Q, the vastly simpler galoissubcyclo is used instead. Here is
an example:
bnf = bnfinit(y^2 - 3);
bnr = bnrinit(bnf, 5);
bnrstark(bnr)
√
returns the ray class field of Q( 3) modulo 5. Usually, one wants to apply to the result one of
rnfpolredbest(bnf, pol) \\ compute a reduced relative polynomial
rnfpolredbest(bnf, pol, 2) \\ compute a reduced absolute polynomial
The routine uses Stark units and needs to find a suitable auxiliary conductor, which may not
exist when the class field is not cyclic over the base. In this case bnrstark is allowed to return a
vector of polynomials defining independent relative extensions, whose compositum is the requested
class field. We decided that it was useful to keep the extra information thus made available, hence
the user has to take the compositum herself, see nfcompositum.

370
 Even if it exists, the auxiliary conductor may be so large that later computations become
unfeasible. (And of course, Stark’s conjecture may simply be wrong.) In case of difficulties, try
bnrclassfield:

? bnr = bnrinit(bnfinit(y^8-12*y^6+36*y^4-36*y^2+9,1), 2);

? bnrstark(bnr)
*** at top-level: bnrstark(bnr)
*** ^-------------
*** bnrstark: need 3919350809720744 coefficients in initzeta.
*** Computation impossible.
? bnrclassfield(bnr)
time = 20 ms.
%2 = [x^2 + (-2/3*y^6 + 7*y^4 - 14*y^2 + 3)]

The library syntax is GEN bnrstark(GEN bnr, GEN subgroup = NULL, long prec).

3.13.41 dirzetak(nf , b). Gives as a vector the first b coefficients of the Dedekind zeta function of
the number field nf considered as a Dirichlet series.

The library syntax is GEN dirzetak(GEN nf, GEN b).

3.13.42 factornf(x, t). This function is obsolete, use nffactor.

factorization of the univariate polynomial x over the number field defined by the (univariate)
polynomial t. x may have coefficients in Q or in the number field. The algorithm reduces to
factorization over Q (Trager’s trick). The direct approach of nffactor, which uses van Hoeij’s
method in a relative setting, is in general faster.

The main variable of t must be of lower priority than that of x (see Section 2.5.3). However
if nonrational number field elements occur (as polmods or polynomials) as coefficients of x, the
variable of these polmods must be the same as the main variable of t. For example

? factornf(x^2 + Mod(y, y^2+1), y^2+1);

? factornf(x^2 + y, y^2+1); \\ these two are OK
? factornf(x^2 + Mod(z,z^2+1), y^2+1)
*** at top-level: factornf(x^2+Mod(z,z
*** ^--------------------
*** factornf: inconsistent data in rnf function.
? factornf(x^2 + z, y^2+1)
*** at top-level: factornf(x^2+z,y^2+1
*** ^--------------------
*** factornf: incorrect variable in rnf function.

The library syntax is GEN polfnf(GEN x, GEN t).

371
3.13.43 galoischardet(gal , chi , {o = 1}). Let G be the group attached to the galoisinit struc-
ture gal , and let χ be the character of some representation ρ of the group G, where a polynomial
variable is to be interpreted as an o-th root of 1. For instance, if [T,o] = galoischartable(gal)
the characters χ are input as the columns of T.

Return the degree-1 character det ρ as the list of det ρ(g), where g runs through representatives
of the conjugacy classes in galoisconjclasses(gal), with the same ordering.

? P = x^5 - x^4 - 5*x^3 + 4*x^2 + 3*x - 1;

? polgalois(P)
%2 = [10, 1, 1, "D(5) = 5:2"]
? K = nfsplitting(P);
? gal = galoisinit(K); \\ dihedral of order 10
? [T,o] = galoischartable(gal);
? chi = T[,1]; \\ trivial character
? galoischardet(gal, chi, o)
%7 = [1, 1, 1, 1]~
? [galoischardet(gal, T[,i], o) | i <- [1..#T]] \\ all characters
%8 = [[1, 1, 1, 1]~, [1, 1, -1, 1]~, [1, 1, -1, 1]~, [1, 1, -1, 1]~]

The library syntax is GEN galoischardet(GEN gal, GEN chi, long o).

3.13.44 galoischarpoly(gal , chi , {o = 1}). Let G be the group attached to the galoisinit
structure gal , and let χ be the character of some representation ρ of the group G, where a polynomial
variable is to be interpreted as an o-th root of 1, e.g., if [T,o] = galoischartable(gal) and χ is
a column of T. Return the list of characteristic polynomials det(1 − ρ(g)T ), where g runs through
representatives of the conjugacy classes in galoisconjclasses(gal), with the same ordering.

? T = x^5 - x^4 - 5*x^3 + 4*x^2 + 3*x - 1;

? polgalois(T)
%2 = [10, 1, 1, "D(5) = 5:2"]
? K = nfsplitting(T);
? gal = galoisinit(K); \\ dihedral of order 10
? [T,o] = galoischartable(gal);
? o
%5 = 5
? galoischarpoly(gal, T[,1], o) \\ T[,1] is the trivial character
%6 = [-x + 1, -x + 1, -x + 1, -x + 1]~
? galoischarpoly(gal, T[,3], o)
%7 = [x^2 - 2*x + 1,
x^2 + (y^3 + y^2 + 1)*x + 1,
-x^2 + 1,
x^2 + (-y^3 - y^2)*x + 1]~

The library syntax is GEN galoischarpoly(GEN gal, GEN chi, long o).

372
3.13.45 galoischartable(gal ). Compute the character table of G, where G is the underlying group
of the galoisinit structure gal . The input gal is also allowed to be a t_VEC of permutations that
is closed under products. Let N be the number of conjugacy classes of G. Return a t_VEC [M, e]
where e ≥ 1 is an integer and M is a square t_MAT of size N giving the character table of G.
• Each column corresponds to an irreducible character; the characters are ordered by increasing
dimension and the first column is the trivial character (hence contains only 1’s).
• Each row corresponds to a conjugacy class; the conjugacy classes are ordered as specified
by galoisconjclasses(gal), in particular the first row corresponds to the identity and gives the
dimension χ(1) of the irreducible representation attached to the successive characters χ.
The value M [i, j] of the character j at the conjugacy class i is represented by a polynomial in
y whose variable should be interpreted as an e-th root of unity, i.e. as the lift of
Mod(y, polcyclo(e,’y))
(Note that M is the transpose of the usual orientation for character tables.)
The integer e divides the exponent of the group G and is chosen as small as posible; for instance
e = 1 when the characters are all defined over Q, as is the case for Sn . Examples:
? K = nfsplitting(x^4+x+1);
? gal = galoisinit(K);
? [M,e] = galoischartable(gal);
? M~ \\ take the transpose to get the usual orientation
%4 =
[1 1 1 1 1]
[1 -1 -1 1 1]
[2 0 0 -1 2]
[3 -1 1 0 -1]
[3 1 -1 0 -1]
? e
%5 = 1
? {G = [Vecsmall([1, 2, 3, 4, 5]), Vecsmall([1, 5, 4, 3, 2]),
Vecsmall([2, 1, 5, 4, 3]), Vecsmall([2, 3, 4, 5, 1]),
Vecsmall([3, 2, 1, 5, 4]), Vecsmall([3, 4, 5, 1, 2]),
Vecsmall([4, 3, 2, 1, 5]), Vecsmall([4, 5, 1, 2, 3]),
Vecsmall([5, 1, 2, 3, 4]), Vecsmall([5, 4, 3, 2, 1])];}
\\G = D10
? [M,e] = galoischartable(G);
? M~
%8 =
[1 1 1 1]
[1 -1 1 1]
[2 0 -y^3 - y^2 - 1 y^3 + y^2]
[2 0 y^3 + y^2 -y^3 - y^2 - 1]
? e
%9 = 5
The library syntax is GEN galoischartable(GEN gal).

373
3.13.46 galoisconjclasses(gal ). gal being output by galoisinit, return the list of conjugacy
classes of the underlying group. The ordering of the classes is consistent with galoischartable
and the trivial class comes first.
? G = galoisinit(x^6+108);
? galoisidentify(G)
%2 = [6, 1] \\ S_3
? S = galoisconjclasses(G)
%3 = [[Vecsmall([1,2,3,4,5,6])],
[Vecsmall([3,1,2,6,4,5]),Vecsmall([2,3,1,5,6,4])],
[Vecsmall([6,5,4,3,2,1]),Vecsmall([5,4,6,2,1,3]),
Vecsmall([4,6,5,1,3,2])]]
? [[permorder(c[1]),#c] | c <- S ]
%4 = [[1,1], [3,2], [2,3]]
This command also accepts subgroups returned by galoissubgroups:
? subs = galoissubgroups(G); H = subs[5];
? galoisidentify(H)
%2 = [2, 1] \\ Z/2
? S = galoisconjclasses(subgroups_of_G[5]);
? [[permorder(c[1]),#c] | c <- S ]
%4 = [[1,1], [2,1]]

The library syntax is GEN galoisconjclasses(GEN gal).

3.13.47 galoisexport(gal , {flag}). gal being be a Galois group as output by galoisinit, export
the underlying permutation group as a string suitable for (no flags or flag = 0) GAP or (flag = 1)
Magma. The following example compute the index of the underlying abstract group in the GAP
library:
? G = galoisinit(x^6+108);
? s = galoisexport(G)
%2 = "Group((1, 2, 3)(4, 5, 6), (1, 4)(2, 6)(3, 5))"
? extern("echo \"IdGroup("s");\" | gap -q")
%3 = [6, 1]
? galoisidentify(G)
%4 = [6, 1]
This command also accepts subgroups returned by galoissubgroups.
To import a GAP permutation into gp (for galoissubfields for instance), the following GAP
function may be useful:
PermToGP := function(p, n)
return Permuted([1..n],p);
end;
gap> p:= (1,26)(2,5)(3,17)(4,32)(6,9)(7,11)(8,24)(10,13)(12,15)(14,27)
(16,22)(18,28)(19,20)(21,29)(23,31)(25,30)
gap> PermToGP(p,32);
[ 26, 5, 17, 32, 2, 9, 11, 24, 6, 13, 7, 15, 10, 27, 12, 22, 3, 28, 20, 19,
29, 16, 31, 8, 30, 1, 14, 18, 21, 25, 23, 4 ]
The library syntax is GEN galoisexport(GEN gal, long flag).

374
3.13.48 galoisfixedfield(gal , perm, {flag}, {v = y}). gal being be a Galois group as output by
galoisinit and perm an element of gal .group, a vector of such elements or a subgroup of gal as
returned by galoissubgroups, computes the fixed field of gal by the automorphism defined by the
permutations perm of the roots gal .roots. P is guaranteed to be squarefree modulo gal .p.
If no flags or flag = 0, output format is the same as for nfsubfield, returning [P, x] such that
P is a polynomial defining the fixed field, and x is a root of P expressed as a polmod in gal .pol.
If flag = 1 return only the polynomial P .
If flag = 2 return [P, x, F ] where P and x are as above and F is the factorization of gal .pol
over the field defined by P , where variable v (y by default) stands for a root of P . The priority of
v must be less than the priority of the variable of gal .pol (see Section 2.5.3). In this case, P is also
expressed in the variable v for compatibility with F . Example:
? G = galoisinit(x^4+1);
? galoisfixedfield(G,G.group[2],2)
%2 = [y^2 - 2, Mod(- x^3 + x, x^4 + 1), [x^2 - y*x + 1, x^2 + y*x + 1]]
√ √
computes the factorization x4 + 1 = (x2 − 2x + 1)(x2 + 2x + 1)
The library syntax is GEN galoisfixedfield(GEN gal, GEN perm, long flag, long v =
-1) where v is a variable number.

3.13.49 galoisgetgroup(a, {b}). Query the galpol package for a group of order a with index b
in the GAP4 Small Group library, by Hans Ulrich Besche, Bettina Eick and Eamonn O’Brien.
The current version of galpol supports groups of order a ≤ 143. If b is omitted, return the
number of isomorphism classes of groups of order a.
The library syntax is GEN galoisgetgroup(long a, long b). Also available is GEN ga-
loisnbpol(long a) when b is omitted.

3.13.50 galoisgetname(a, b). Query the galpol package for a string describing the group of
order a with index b in the GAP4 Small Group library, by Hans Ulrich Besche, Bettina Eick and
Eamonn O’Brien. The strings were generated using the GAP4 function StructureDescription.
The command below outputs the names of all abstract groups of order 12:
? o = 12; N = galoisgetgroup(o); \\ # of abstract groups of order 12
? for(i=1, N, print(i, ". ", galoisgetname(o,i)))
1. C3 : C4
2. C12
3. A4
4. D12
5. C6 x C2
The current version of galpol supports groups of order a ≤ 143. For a ≥ 16, it is possible for
different groups to have the same name:
? o = 20; N = galoisgetgroup(o);
? for(i=1, N, print(i, ". ", galoisgetname(o,i)))
1. C5 : C4
2. C20
3. C5 : C4

375
 4. D20
5. C10 x C2
The library syntax is GEN galoisgetname(long a, long b).

3.13.51 galoisgetpol(a, {b}, {s}). Query the galpol package for a polynomial with Galois group
isomorphic to GAP4(a,b), totally real if s = 1 (default) and totally complex if s = 2. The current
version of galpol supports groups of order a ≤ 143. The output is a vector [pol, den] where
• pol is the polynomial of degree a
• den is the denominator of nfgaloisconj(pol). Pass it as an optional argument to ga-
loisinit or nfgaloisconj to speed them up:
? [pol,den] = galoisgetpol(64,4,1);
? G = galoisinit(pol);
time = 352ms
? galoisinit(pol, den); \\ passing ’den’ speeds up the computation
time = 264ms
? % == %‘
%4 = 1 \\ same answer
If b and s are omitted, return the number of isomorphism classes of groups of order a.
The library syntax is GEN galoisgetpol(long a, long b, long s). Also available is GEN
galoisnbpol(long a) when b and s are omitted.

3.13.52 galoisidentify(gal ). gal being be a Galois group as output by galoisinit, output the
isomorphism class of the underlying abstract group as a two-components vector [o, i], where o is
the group order, and i is the group index in the GAP4 Small Group library, by Hans Ulrich Besche,
Bettina Eick and Eamonn O’Brien.
This command also accepts subgroups returned by galoissubgroups.
The current implementation is limited to degree less or equal to 127. Some larger “easy” orders
are also supported.
The output is similar to the output of the function IdGroup in GAP4. Note that GAP4
IdGroup handles all groups of order less than 2000 except 1024, so you can use galoisexport and
GAP4 to identify large Galois groups.
The library syntax is GEN galoisidentify(GEN gal).

3.13.53 galoisinit(pol , {den}). Computes the Galois group and all necessary information for
computing the fixed fields of the Galois extension K/Q where K is the number field defined by
pol (monic irreducible polynomial in Z[X] or a number field as output by nfinit). The extension
K/Q must be Galois with Galois group “weakly” super-solvable, see below; returns 0 otherwise.
Hence this permits to quickly check whether a polynomial of order strictly less than 48 is Galois or
not.
The algorithm used is an improved version of the paper “An efficient algorithm for the com-
putation of Galois automorphisms”, Bill Allombert, Math. Comp, vol. 73, 245, 2001, pp. 359–375.
A group G is said to be “weakly” super-solvable if there exists a normal series
{1} = H0 / H1 / · · · / Hn−1 / Hn

376
 such that each Hi is normal in G and for i < n, each quotient group Hi+1 /Hi is cyclic, and
either Hn = G (then G is super-solvable) or G/Hn is isomorphic to either A4 , S4 or the group
(3 × 3) : 4 (GAP4(36,9)) then [o1 , . . . , og ] ends by [3, 3, 4].
In practice, almost all small groups are WKSS, the exceptions having order 48(2), 56(1), 60(1),
72(3), 75(1), 80(1), 96(10), 112(1), 120(3) and ≥ 144.
This function is a prerequisite for most of the galoisxxx routines. For instance:
P = x^6 + 108;
G = galoisinit(P);
L = galoissubgroups(G);
vector(#L, i, galoisisabelian(L[i],1))
vector(#L, i, galoisidentify(L[i]))
The output is an 8-component vector gal .
gal [1] contains the polynomial pol (gal .pol).
gal [2] is a three-components vector [p, e, q] where p is a prime number (gal .p) such that pol
totally split modulo p , e is an integer and q = pe (gal .mod) is the modulus of the roots in gal .roots.
gal [3] is a vector L containing the p-adic roots of pol as integers implicitly modulo gal .mod.
(gal .roots).
gal [4] is the inverse of the Vandermonde matrix of the p-adic roots of pol , multiplied by gal [5].
gal [5] is a multiple of the least common denominator of the automorphisms expressed as
polynomial in a root of pol .
gal [6] is the Galois group G expressed as a vector of permutations of L (gal .group).
gal [7] is a generating subset S = [s1 , . . . , sg ] of G expressed as a vector of permutations of L
(gal .gen).
gal [8] contains the relative orders [o1 , . . . , og ] of the generators of S (gal .orders).
Let Hn be as above, we have the following properties:
• if G/Hn ' A4 then [o1 , . . . , og ] ends by [2, 2, 3].
• if G/Hn ' S4 then [o1 , . . . , og ] ends by [2, 2, 3, 2].
• if G/Hn ' (3 × 3) : 4 (GAP4(36,9)) then [o1 , . . . , og ] ends by [3, 3, 4].
• for 1 ≤ i ≤ g the subgroup of G generated by [s1 , . . . , si ] is normal, with the exception of
i = g − 2 in the A4 case and of i = g − 3 in the S4 case.
• the relative order oi of si is its order in the quotient group G/hs1 , . . . , si−1 i, with the same
exceptions.
• for any x ∈ G there exists a unique family [e1 , . . . , eg ] such that (no exceptions):
– for 1 ≤ i ≤ g we have 0 ≤ ei < oi
– x = g1e1 g2e2 . . . gnen
If present den must be a suitable value for gal [5].
The library syntax is GEN galoisinit(GEN pol, GEN den = NULL).

377
3.13.54 galoisisabelian(gal , {flag = 0}). gal being as output by galoisinit, return 0 if gal is
not an abelian group, and the HNF matrix of gal over gal.gen if flag = 0, 1 if flag = 1, and the
SNF matrix of gal if flag = 2.
This command also accepts subgroups returned by galoissubgroups.
The library syntax is GEN galoisisabelian(GEN gal, long flag).

3.13.55 galoisisnormal(gal , subgrp). gal being as output by galoisinit, and subgrp a subgroup
of gal as output by galoissubgroups,return 1 if subgrp is a normal subgroup of gal , else return 0.
This command also accepts subgroups returned by galoissubgroups.
The library syntax is long galoisisnormal(GEN gal, GEN subgrp).

3.13.56 galoispermtopol(gal , perm). gal being a Galois group as output by galoisinit and
perm a element of gal .group, return the polynomial defining the Galois automorphism, as output
by nfgaloisconj, attached to the permutation perm of the roots gal .roots. perm can also be a
vector or matrix, in this case, galoispermtopol is applied to all components recursively.
Note that
G = galoisinit(pol);
galoispermtopol(G, G[6])~
is equivalent to nfgaloisconj(pol), if degree of pol is greater or equal to 2.
The library syntax is GEN galoispermtopol(GEN gal, GEN perm).

3.13.57 galoissubcyclo(N, H, {fl = 0}, {v}). Computes the subextension of Q(ζn ) fixed by the
subgroup H ⊂ (Z/nZ)∗ . By the Kronecker-Weber theorem, all abelian number fields can be
generated in this way (uniquely if n is taken to be minimal).
The pair (n, H) is deduced from the parameters (N, H) as follows
• N an integer: then n = N ; H is a generator, i.e. an integer or an integer modulo n; or a
vector of generators.
• N the output of znstar(n) or znstar(n, 1). H as in the first case above, or a matrix, taken
to be a HNF left divisor of the SNF for (Z/nZ)∗ (N .cyc), giving the generators of H in terms of
N .gen.
• N the output of bnrinit(bnfinit(y), m) where m is a module. H as in the first case,
or a matrix taken to be a HNF left divisor of the SNF for the ray class group modulo m (of type
N .cyc), giving the generators of H in terms of N .bid.gen (= N .gen if N includes generators).
In this last case, beware that H is understood relatively to N ; in particular, if the infinite
place does not divide the module, e.g if m is an integer, then it is not a subgroup of (Z/nZ)∗ , but
of its quotient by {±1}.
If f l = 0, compute a polynomial (in the variable v ) defining the subfield of Q(ζn ) fixed by the
subgroup H of (Z/nZ)∗ .
If f l = 1, compute only the conductor of the abelian extension, as a module.
If f l = 2, output [pol, N ], where pol is the polynomial as output when f l = 0 and N the
conductor as output when f l = 1.

378
 The following function can be used to compute all subfields of Q(ζn ) (of exact degree d, if d is
set):
subcyclo(n, d = -1)=
{ my(bnr,L,IndexBound);
IndexBound = if (d < 0, n, [d]);
bnr = bnrinit(bnfinit(y), [n,[1]]);
L = subgrouplist(bnr, IndexBound, 1);
vector(#L,i, galoissubcyclo(bnr,L[i]));
}
Setting L = subgrouplist(bnr, IndexBound) would produce subfields of exact conductor n∞.
The library syntax is GEN galoissubcyclo(GEN N, GEN H = NULL, long fl, long v = -1)
where v is a variable number.

3.13.58 galoissubfields(G, {flag = 0}, {v}). Outputs all the subfields of the Galois group G, as
a vector. This works by applying galoisfixedfield to all subgroups. The meaning of flag is the
same as for galoisfixedfield.
The library syntax is GEN galoissubfields(GEN G, long flag, long v = -1) where v is
a variable number.

3.13.59 galoissubgroups(G). Outputs all the subgroups of the Galois group gal. A subgroup is
a vector [gen, orders], with the same meaning as for gal .gen and gal .orders. Hence gen is a vector
of permutations generating the subgroup, and orders is the relatives orders of the generators. The
cardinality of a subgroup is the product of the relative orders. Such subgroup can be used instead
of a Galois group in the following command: galoisisabelian, galoissubgroups, galoisexport
and galoisidentify.
To get the subfield fixed by a subgroup sub of gal , use
galoisfixedfield(gal,sub[1])
The library syntax is GEN galoissubgroups(GEN G).

3.13.60 idealadd(nf , x, y). Sum of the two ideals x and y in the number field nf . The result is
given in HNF.
? K = nfinit(x^2 + 1);
? a = idealadd(K, 2, x + 1) \\ ideal generated by 2 and 1+I
%2 =
[2 1]
[0 1]
? pr = idealprimedec(K, 5)[1]; \\ a prime ideal above 5
? idealadd(K, a, pr) \\ coprime, as expected
%4 =
[1 0]
[0 1]
This function cannot be used to add arbitrary Z-modules, since it assumes that its arguments are
ideals:
? b = Mat([1,0]~);

379
 ? idealadd(K, b, b) \\ only square t_MATs represent ideals
*** idealadd: nonsquare t_MAT in idealtyp.
? c = [2, 0; 2, 0]; idealadd(K, c, c) \\ nonsense
%6 =
[2 0]
[0 2]
? d = [1, 0; 0, 2]; idealadd(K, d, d) \\ nonsense
%7 =
[1 0]
[0 1]

In the last two examples, we get wrong results since the matrices c and d do not correspond to
an ideal: the Z-span of their columns (as usual interpreted as coordinates with respect to the
integer basis K.zk) is not an OK -module. To add arbitrary Z-modules generated by the columns
of matrices A and B, use mathnf(concat(A,B)).
The library syntax is GEN idealadd(GEN nf, GEN x, GEN y).

3.13.61 idealaddtoone(nf , x, {y}). x and y being two co-prime integral ideals (given in any
form), this gives a two-component row vector [a, b] such that a ∈ x, b ∈ y and a + b = 1.
The alternative syntax idealaddtoone(nf , v), is supported, where v is a k-component vector
of ideals (given in any form)Pwhich sum to ZK . This outputs a k-component vector e such that
e[i] ∈ x[i] for 1 ≤ i ≤ k and 1≤i≤k e[i] = 1.
The library syntax is GEN idealaddtoone0(GEN nf, GEN x, GEN y = NULL).

3.13.62 idealappr(nf , x, {flag}). If x is a fractional ideal (given in any form), gives an element
α in nf such that for all prime ideals p such that the valuation of x at p is nonzero, we have
vp (α) = vp (x), and vp (α) ≥ 0 for all other p.
The argument x may also be given as a prime ideal factorization, as output by idealfactor,
but allowing zero exponents. This yields an element α such that for all prime ideals p occurring in
x, vp (α) = vp (x); for all other prime ideals, vp (α) ≥ 0.
flag is deprecated (ignored), kept for backward compatibility.
The library syntax is GEN idealappr0(GEN nf, GEN x, long flag). Use directly GEN
idealappr(GEN nf, GEN x) since flagis ignored.

3.13.63 idealchinese(nf , x, {y}). x being a prime ideal factorization (i.e. a 2-columns matrix
whose first column contains prime ideals and the second column contains integral exponents), y a
vector of elements in nf indexed by the ideals in x, computes an element b such that
vp (b − yp ) ≥ vp (x) for all prime ideals in x and vp (b) ≥ 0 for all other p.
? K = nfinit(t^2-2);
? x = idealfactor(K, 2^2*3)
%2 =
[[2, [0, 1]~, 2, 1, [0, 2; 1, 0]] 4]
[ [3, [3, 0]~, 1, 2, 1] 1]
? y = [t,1];

380
 ? idealchinese(K, x, y)
%4 = [4, -3]~

The argument x may also be of the form [x, s] where the first component is as above and s is
a vector of signs, with r1 components si in {−1, 0, 1}: if σi denotes the i-th real embedding of the
number field, the element b returned satisfies further sign(σi (b)) = si for all i such that si = ±1.
In other words, the sign is fixed to si at the i-th embedding whenever si is nonzero.

? idealchinese(K, [x, [1,1]], y)

%5 = [16, -3]~
? idealchinese(K, [x, [-1,-1]], y)
%6 = [-20, -3]~
? idealchinese(K, [x, [1,-1]], y)
%7 = [4, -3]~

If y is omitted, return a data structure which can be used in place of x in later calls and allows
to solve many chinese remainder problems for a given x more efficiently.

? C = idealchinese(K, [x, [1,1]]);

? idealchinese(K, C, y) \\ as above
%9 = [16, -3]~
? for(i=1,10^4, idealchinese(K,C,y)) \\ ... but faster !
time = 80 ms.
? for(i=1,10^4, idealchinese(K,[x,[1,1]],y))
time = 224 ms.

Finally, this structure is itself allowed in place of x, the new s overriding the one already
present in the structure. This allows to initialize for different sign conditions more efficiently when
the underlying ideal factorization remains the same.

? D = idealchinese(K, [C, [1,-1]]); \\ replaces [1,1]

? idealchinese(K, D, y)
%13 = [4, -3]~
? for(i=1,10^4,idealchinese(K,[C,[1,-1]]))
time = 40 ms. \\ faster than starting from scratch
? for(i=1,10^4,idealchinese(K,[x,[1,-1]]))
time = 128 ms.

The library syntax is GEN idealchinese(GEN nf, GEN x, GEN y = NULL). Also available is
GEN idealchineseinit(GEN nf, GEN x) when y = NULL.

3.13.64 idealcoprime(nf , x, y). Given two integral ideals x and y in the number field nf , returns
a β in the field, such that β · x is an integral ideal coprime to y.

The library syntax is GEN idealcoprime(GEN nf, GEN x, GEN y).

381
3.13.65 idealdiv(nf , x, y, {flag = 0}). Quotient x · y −1 of the two ideals x and y in the number
field nf . The result is given in HNF.
If flag is nonzero, the quotient x·y −1 is assumed to be an integral ideal. This can be much faster
when the norm of the quotient is small even though the norms of x and y are large. More precisely,
the algorithm cheaply removes all maximal ideals above rational primes such that vp (N x) = vp (N y).
The library syntax is GEN idealdiv0(GEN nf, GEN x, GEN y, long flag). Also available
are GEN idealdiv(GEN nf, GEN x, GEN y) (flag = 0) and GEN idealdivexact(GEN nf, GEN x,
GEN y) (flag = 1).

3.13.66 idealdown(nf , x). Let nf be a number field as output by nfinit, and x a fractional ideal.
This function returns the nonnegative rational generator of x ∩ Q. If x is an extended ideal, the
extended part is ignored.
? nf = nfinit(y^2+1);
? idealdown(nf, -1/2)
%2 = 1/2
? idealdown(nf, (y+1)/3)
%3 = 2/3
? idealdown(nf, [2, 11]~)
%4 = 125
? x = idealprimedec(nf, 2)[1]; idealdown(nf, x)
%5 = 2
? idealdown(nf, [130, 94; 0, 2])
%6 = 130
The library syntax is GEN idealdown(GEN nf, GEN x).

3.13.67 idealfactor(nf , x, {lim}). Factors into prime ideal powers the ideal x in the number field
nf . The output format is similar to the factor function, and the prime ideals are represented in the
form output by the idealprimedec function. If lim is set, return partial factorization, including
only prime ideals above rational primes < lim.
? nf = nfinit(x^3-2);
? idealfactor(nf, x) \\ a prime ideal above 2
%2 =
[[2, [0, 1, 0]~, 3, 1, ...] 1]
? A = idealhnf(nf, 6*x, 4+2*x+x^2)
%3 =
[6 0 4]
[0 6 2]
[0 0 1]
? idealfactor(nf, A)
%4 =
[[2, [0, 1, 0]~, 3, 1, ...] 2]
[[3, [1, 1, 0]~, 3, 1, ...] 2]
? idealfactor(nf, A, 3) \\ restrict to primes above p < 3
%5 =

382
 [[2, [0, 1, 0]~, 3, 1, ...] 2]

The library syntax is GEN gpidealfactor(GEN nf, GEN x, GEN lim = NULL). This function
should only be used by the gp interface. Use directly GEN idealfactor(GEN x) or GEN idealfac-
tor_limit(GEN x, ulong lim).

3.13.68 idealfactorback(nf , f, {e}, {flag = 0}). Gives back the ideal corresponding to a factor-
ization. The integer 1 corresponds to the empty factorization. If e is present, e and f must be
vectors of the same length (e being integral), and the corresponding factorization is the product of
the f [i]e[i] .

If not, and f is vector, it is understood as in the preceding case with e a vector of 1s: we return
the product of the f [i]. Finally, f can be a regular factorization, as produced by idealfactor.

? nf = nfinit(y^2+1); idealfactor(nf, 4 + 2*y)

%1 =
[[2, [1, 1]~, 2, 1, [1, 1]~] 2]
[[5, [2, 1]~, 1, 1, [-2, 1]~] 1]
? idealfactorback(nf, %)
%2 =
[10 4]
[0 2]
? f = %1[,1]; e = %1[,2]; idealfactorback(nf, f, e)
%3 =
[10 4]
[0 2]
? % == idealhnf(nf, 4 + 2*y)
%4 = 1

If flag is nonzero, perform ideal reductions (idealred) along the way. This is most useful
if the ideals involved are all extended ideals (for instance with trivial principal part), so that the
principal parts extracted by idealred are not lost. Here is an example:

? f = vector(#f, i, [f[i], [;]]); \\ transform to extended ideals

? idealfactorback(nf, f, e, 1)
%6 = [[1, 0; 0, 1], [2, 1; [2, 1]~, 1]]
? nffactorback(nf, %[2])
%7 = [4, 2]~

The extended ideal returned in %6 is the trivial ideal 1, extended with a principal generator
given in factored form. We use nffactorback to recover it in standard form.

The library syntax is GEN idealfactorback(GEN nf, GEN f, GEN e = NULL, long flag)
.

383
3.13.69 idealfrobenius(nf , gal , pr ). Let K be the number field defined by nf and assume K/Q
be a Galois extension with Galois group given gal=galoisinit(nf), and that pr is an unramified
prime ideal p in prid format. This function returns a permutation of gal.group which defines
the Frobenius element Frobp attached to p. If p is the unique prime number in p, then Frob(x) ≡
xp mod p for all x ∈ ZK .
? nf = nfinit(polcyclo(31));
? gal = galoisinit(nf);
? pr = idealprimedec(nf,101)[1];
? g = idealfrobenius(nf,gal,pr);
? galoispermtopol(gal,g)
%5 = x^8
This is correct since 101 ≡ 8 mod 31.
The library syntax is GEN idealfrobenius(GEN nf, GEN gal, GEN pr).

3.13.70 idealhnf(nf , u, {v}). Gives the Hermite normal form of the ideal uZK + vZK , where u
and v are elements of the number field K defined by nf .
? nf = nfinit(y^3 - 2);
? idealhnf(nf, 2, y+1)
%2 =
[1 0 0]
[0 1 0]
[0 0 1]
? idealhnf(nf, y/2, [0,0,1/3]~)
%3 =
[1/3 0 0]
[0 1/6 0]
[0 0 1/6]
If b is omitted, returns the HNF of the ideal defined by u: u may be an algebraic number
(defining a principal ideal), a maximal ideal (as given by idealprimedec or idealfactor), or a
matrix whose columns give generators for the ideal. This last format is a little complicated, but
useful to reduce general modules to the canonical form once in a while:
• if strictly less than N = [K : Q] generators are given, u is the ZK -module they generate,
• if N or more are given, it is assumed that they form a Z-basis of the ideal, in particular that
the matrix has maximal rank N . This acts as mathnf since the ZK -module structure is (taken for
granted hence) not taken into account in this case.
? idealhnf(nf, idealprimedec(nf,2)[1])
%4 =
[2 0 0]
[0 1 0]
[0 0 1]
? idealhnf(nf, [1,2;2,3;3,4])
%5 =
[1 0 0]

384
 [0 1 0]
[0 0 1]
2
√ u = Qfb(a,b,c), provided b − 4ac =
Finally, when K is quadratic with discriminant DK , we allow
DK . As usual, this represents the ideal aZ + (1/2)(−b + DK )Z.

? K = nfinit(x^2 - 60); K.disc

%1 = 60
? idealhnf(K, qfbprimeform(60,2))
%2 =
[2 1]
[0 1]
? idealhnf(K, Qfb(1,2,3))
*** at top-level: idealhnf(K,Qfb(1,2,3
*** ^--------------------
*** idealhnf: Qfb(1, 2, 3) has discriminant != 60 in idealhnf.

The library syntax is GEN idealhnf0(GEN nf, GEN u, GEN v = NULL). Also available is GEN
idealhnf(GEN nf, GEN a).

3.13.71 idealintersect(nf , A, B). Intersection of the two ideals A and B in the number field nf .
The result is given in HNF.

? nf = nfinit(x^2+1);
? idealintersect(nf, 2, x+1)
%2 =
[2 0]
[0 2]

This function does not apply to general Z-modules, e.g. orders, since its arguments are replaced
by the ideals they generate. The following script intersects Z-modules A and B given by matrices
of compatible dimensions with integer coefficients:

ZM_intersect(A,B) =
{ my(Ker = matkerint(concat(A,B)));
mathnf( A * Ker[1..#A,] )
}

The library syntax is GEN idealintersect(GEN nf, GEN A, GEN B).

3.13.72 idealinv(nf , x). Inverse of the ideal x in the number field nf , given in HNF. If x is an
extended ideal, its principal part is suitably updated: i.e. inverting [I, t], yields [I −1 , 1/t].

The library syntax is GEN idealinv(GEN nf, GEN x).

385
3.13.73 idealismaximal(nf , x). Given nf a number field as output by nfinit and an ideal x,
return 0 if x is not a maximal ideal. Otherwise return a prid structure nf attached to the ideal.
This function uses ispseudoprime and may return a wrong result in case the underlying rational
pseudoprime is not an actual prime number: apply isprime(pr.p) to guarantee correctness. If x
is an extended ideal, the extended part is ignored.

? K = nfinit(y^2 + 1);
? idealismaximal(K, 3) \\ 3 is inert
%2 = [3, [3, 0]~, 1, 2, 1]
? idealismaximal(K, 5) \\ 5 is not
%3 = 0
? pr = idealprimedec(K,5)[1] \\ already a prid
%4 = [5, [-2, 1]~, 1, 1, [2, -1; 1, 2]]
? idealismaximal(K, pr) \\ trivial check
%5 = [5, [-2, 1]~, 1, 1, [2, -1; 1, 2]]
? x = idealhnf(K, pr)
%6 =
[5 3]
[0 1]
? idealismaximal(K, x) \\ converts from matrix form to prid
%7 = [5, [-2, 1]~, 1, 1, [2, -1; 1, 2]]
This function is noticeably faster than idealfactor since it never involves an actually factorization,
in particular when x ∩ Z is not a prime number.
The library syntax is GEN idealismaximal(GEN nf, GEN x).

3.13.74 idealispower(nf , A, n, {&B}). Let nf be a number field and n > 0 be a positive integer.
Return 1 if the fractional ideal A = B n is an n-th power and 0 otherwise. If the argument B is
present, set it to the n-th root of A, in HNF.

? K = nfinit(x^3 - 2);
? A = [46875, 30966, 9573; 0, 3, 0; 0, 0, 3];
? idealispower(K, A, 3, &B)
%3 = 1
? B
%4 =
[75 22 41]
[ 0 1 0]
[ 0 0 1]
? A = [9375, 2841, 198; 0, 3, 0; 0, 0, 3];
? idealispower(K, A, 3)
%5 = 0

The library syntax is long idealispower(GEN nf, GEN A, long n, GEN *B = NULL).

386
3.13.75 ideallist(nf , bound , {flag = 4}). Computes the list of all ideals of norm less or equal to
bound in the number field nf . The result is a row vector with exactly bound components. Each
component is itself a row vector containing the information about ideals of a given norm, in no
specific order, depending on the value of flag:

The possible values of flag are:

0: give the bid attached to the ideals, without generators.

1: as 0, but include the generators in the bid .

2: in this case, nf must be a bnf with units. Each component is of the form [bid , U ], where
bid is as case 0 and U is a vector of discrete logarithms of the units. More precisely, it gives the
ideallogs with respect to bid of (ζ, u1 , . . . , ur ) where ζ is the torsion unit generator bnf.tu[2]
and (ui ) are the fundamental units in bnf.fu. This structure is technical, and only meant to be
used in conjunction with bnrclassnolist or bnrdisclist.

3: as 2, but include the generators in the bid .

4: give only the HNF of the ideal.

? nf = nfinit(x^2+1);
? L = ideallist(nf, 100);
? L[1]
%3 = [[1, 0; 0, 1]] \\ A single ideal of norm 1
? #L[65]
%4 = 4 \\ There are 4 ideals of norm 4 in Z[i]

If one wants more information, one could do instead:

? nf = nfinit(x^2+1);
? L = ideallist(nf, 100, 0);
? l = L[25]; vector(#l, i, l[i].clgp)
%3 = [[20, [20]], [16, [4, 4]], [20, [20]]]
? l[1].mod
%4 = [[25, 18; 0, 1], []]
? l[2].mod
%5 = [[5, 0; 0, 5], []]
? l[3].mod
%6 = [[25, 7; 0, 1], []]

where we ask for the structures of the (Z[i]/I)∗ for all three ideals of norm 25. In fact, for all
moduli with finite part of norm 25 and trivial Archimedean part, as the last 3 commands show.
See ideallistarch to treat general moduli.

The library syntax is GEN ideallist0(GEN nf, long bound, long flag).

387
3.13.76 ideallistarch(nf , list, arch). list is a vector of vectors of bid’s, as output by ideallist
with flag 0 to 3. Return a vector of vectors with the same number of components as the original
list. The leaves give information about moduli whose finite part is as in original list, in the same
order, and Archimedean part is now arch (it was originally trivial). The information contained is
of the same kind as was present in the input; see ideallist, in particular the meaning of flag.
? bnf = bnfinit(x^2-2);
? bnf.sign
%2 = [2, 0] \\ two places at infinity
? L = ideallist(bnf, 100, 0);
? l = L[98]; vector(#l, i, l[i].clgp)
%4 = [[42, [42]], [36, [6, 6]], [42, [42]]]
? La = ideallistarch(bnf, L, [1,1]); \\ add them to the modulus
? l = La[98]; vector(#l, i, l[i].clgp)
%6 = [[168, [42, 2, 2]], [144, [6, 6, 2, 2]], [168, [42, 2, 2]]]
Of course, the results above are obvious: adding t places at infinity will add t copies of Z/2Z
to (ZK /f )∗ . The following application is more typical:
? L = ideallist(bnf, 100, 2); \\ units are required now
? La = ideallistarch(bnf, L, [1,1]);
? H = bnrclassnolist(bnf, La);
? H[98];
%4 = [2, 12, 2]
The library syntax is GEN ideallistarch(GEN nf, GEN list, GEN arch).

3.13.77 ideallog({nf }, x, bid ). nf is a number field, bid is as output by idealstar(nf, D, . . . )

and x an element of nf which must have valuation equal to 0 at all prime ideals in the support of
D and need not be integral. This function computes the discrete logarithm of x on the generators
given in bid .gen. In other words, if gi are these generators, of orders di respectively, the result is
a column vector of integers (xi ) such that 0 ≤ xi < di and
Y
∗
x≡ gixi (mod D) .
i

Note that when the support of D contains places at infinity, this congruence implies also sign
conditions on the attached real embeddings. See znlog for the limitations of the underlying discrete
log algorithms.
When nf is omitted, take it to be the rational number field. In that case, x must be a t_INT
and bid must have been initialized by znstar(N,1).
The library syntax is GEN ideallog(GEN nf = NULL, GEN x, GEN bid). Also available are
GEN Zideallog(GEN bid, GEN x) when nf is NULL, and GEN ideallogmod(GEN nf, GEN x, GEN
bid, GEN mod) that returns the discrete logarithm of x modulo the t_INT mod; the value mod = NULL
is treated as 0 (full discrete logarithm), but nf = NULL is not implemented with nonzero mod.

3.13.78 idealmin(nf , ix , {vdir }). This function is useless and kept for backward compatibility only,
use idealred. Computes a pseudo-minimum of the ideal x in the direction vdir in the number
field nf .
The library syntax is GEN idealmin(GEN nf, GEN ix, GEN vdir = NULL).

388
3.13.79 idealmul(nf , x, y, {flag = 0}). Ideal multiplication of the ideals x and y in the number
field nf ; the result is the ideal product in HNF. If either x or y are extended ideals, their principal
part is suitably updated: i.e. multiplying [I, t], [J, u] yields [IJ, tu]; multiplying I and [J, u] yields
[IJ, u].

? nf = nfinit(x^2 + 1);
? idealmul(nf, 2, x+1)
%2 =
[4 2]
[0 2]
? idealmul(nf, [2, x], x+1) \\ extended ideal * ideal
%3 = [[4, 2; 0, 2], x]
? idealmul(nf, [2, x], [x+1, x]) \\ two extended ideals
%4 = [[4, 2; 0, 2], [-1, 0]~]

If flag is nonzero, reduce the result using idealred.

The library syntax is GEN idealmul0(GEN nf, GEN x, GEN y, long flag).

See also GEN idealmul(GEN nf, GEN x, GEN y) (flag = 0) and GEN idealmulred(GEN nf, GEN
x, GEN y) (flag 6= 0).

3.13.80 idealnorm(nf , x). Computes the norm of the ideal x in the number field nf .

The library syntax is GEN idealnorm(GEN nf, GEN x).

3.13.81 idealnumden(nf , x). Returns [A, B], where A, B are coprime integer ideals such that
x = A/B, in the number field nf .

? nf = nfinit(x^2+1);
? idealnumden(nf, (x+1)/2)
%2 = [[1, 0; 0, 1], [2, 1; 0, 1]]

The library syntax is GEN idealnumden(GEN nf, GEN x).

3.13.82 idealpow(nf , x, k, {flag = 0}). Computes the k-th power of the ideal x in the number
field nf ; k ∈ Z. If x is an extended ideal, its principal part is suitably updated: i.e. raising [I, t] to
the k-th power, yields [I k , tk ].

If flag is nonzero, reduce the result using idealred, throughout the (binary) powering process;
in particular, this is not the same as idealpow(nf , x, k) followed by reduction.

The library syntax is GEN idealpow0(GEN nf, GEN x, GEN k, long flag).

See also GEN idealpow(GEN nf, GEN x, GEN k) and GEN idealpows(GEN nf, GEN x, long k)
(flag = 0). Corresponding to flag = 1 is GEN idealpowred(GEN nf, GEN vp, GEN k).

389
3.13.83 idealprimedec(nf , p, {f = 0}). Computes the prime ideal decomposition of the (positive)
prime number p in the number field K represented by nf . If a nonprime p is given the result is
undefined. If f is present and nonzero, restrict the result to primes of residue degree ≤ f .
The result is a vector of prid structures, each representing one of the prime ideals above p in
the number field nf . The representation pr = [p, a, e, f, mb] of a prime ideal means the following:
a is an algebraic integer in the maximal order ZK and the prime ideal is equal to p = pZK + aZK ;
e is the ramification index; f is the residual index; finally, mb is the multiplication table attached
to the algebraic integer b is such that p−1 = ZK + b/pZK , which is used internally to compute
valuations. In other words if p is inert, then mb is the integer 1, and otherwise it is a square t_MAT
whose j-th column is b · nf.zk[j].
The algebraic number a is guaranteed to have a valuation equal to 1 at the prime ideal (this
is automatic if e > 1).
The components of pr should be accessed by member functions: pr.p, pr.e, pr.f, and pr.gen
(returns the vector [p, a]):
? K = nfinit(x^3-2);
? P = idealprimedec(K, 5);
? #P \\ 2 primes above 5 in Q(2^(1/3))
%3 = 2
? [p1,p2] = P;
? [p1.e, p1.f] \\ the first is unramified of degree 1
%5 = [1, 1]
? [p2.e, p2.f] \\ the second is unramified of degree 2
%6 = [1, 2]
? p1.gen
%7 = [5, [2, 1, 0]~]
? nfbasistoalg(K, %[2]) \\ a uniformizer for p1
%8 = Mod(x + 2, x^3 - 2)
? #idealprimedec(K, 5, 1) \\ restrict to f = 1
%9 = 1 \\ now only p1
The library syntax is GEN idealprimedec_limit_f(GEN nf, GEN p, long f).

3.13.84 idealprincipalunits(nf , pr , k). Given a prime ideal in idealprimedec format, returns

the multiplicative group (1 + pr )/(1 + pr k ) as an abelian group. This function is much faster than
idealstar when the norm of pr is large, since it avoids (useless) work in the multiplicative group
of the residue field.
? K = nfinit(y^2+1);
? P = idealprimedec(K,2)[1];
? G = idealprincipalunits(K, P, 20);
? G.cyc
%4 = [512, 256, 4] \\ Z/512 x Z/256 x Z/4
? G.gen
%5 = [[-1, -2]~, 1021, [0, -1]~] \\ minimal generators of given order
The library syntax is GEN idealprincipalunits(GEN nf, GEN pr, long k).

390
3.13.85 idealramgroups(nf , gal , pr ). Let K be the number field defined by nf and assume that
K/Q is Galois with Galois group G given by gal=galoisinit(nf). Let pr be the prime ideal P
in prid format. This function returns a vector g of subgroups of gal as follows:
• g[1] is the decomposition group of P,
• g[2] is G0 (P), the inertia group of P,
and for i ≥ 2,
• g[i] is Gi−2 (P), the i − 2-th ramification group of P.
The length of g is the number of nontrivial groups in the sequence, thus is 0 if e = 1 and f = 1,
and 1 if f > 1 and e = 1. The following function computes the cardinality of a subgroup of G, as
given by the components of g:
card(H) =my(o=H[2]); prod(i=1,#o,o[i]);
? nf=nfinit(x^6+3); gal=galoisinit(nf); pr=idealprimedec(nf,3)[1];
? g = idealramgroups(nf, gal, pr);
? apply(card,g)
%3 = [6, 6, 3, 3, 3] \\ cardinalities of the G_i
? nf=nfinit(x^6+108); gal=galoisinit(nf); pr=idealprimedec(nf,2)[1];
? iso=idealramgroups(nf,gal,pr)[2]
%5 = [[Vecsmall([2, 3, 1, 5, 6, 4])], Vecsmall([3])]
? nfdisc(galoisfixedfield(gal,iso,1))
%6 = -3
The field fixed by the inertia group of 2 is not ramified at 2.
The library syntax is GEN idealramgroups(GEN nf, GEN gal, GEN pr).

3.13.86 idealred(nf , I, {v = 0}). LLL reduction of the ideal I in the number field K attached to
nf , along the direction v. The v parameter is best left omitted, but if it is present, it must be an
nf.r1 + nf.r2-component vector of nonnegative integers. (What counts is the relative magnitude
of the entries: if all entries are equal, the effect is the same as if the vector had been omitted.)
This function finds an a ∈ K ∗ such that J = (a)I is “small” and integral (see the end for
technical details). The result is the Hermite normal form of the “reduced” ideal J.
? K = nfinit(y^2+1);
? P = idealprimedec(K,5)[1];
? idealred(K, P)
%3 =
[1 0]
[0 1]
More often than not, a principal ideal yields the unit ideal as above. This is a quick and dirty
way to check if ideals are principal, but it is not a necessary condition: a nontrivial result does not
prove that the ideal is nonprincipal. For guaranteed results, see bnfisprincipal, which requires
the computation of a full bnf structure.
If the input is an extended ideal [I, s], the output is [J, sa]; in this way, one keeps track of the
principal ideal part:

391
 ? idealred(K, [P, 1])
%5 = [[1, 0; 0, 1], [2, -1]~]
meaning that P is generated by [2, −1] . The number field element in the extended part is an
algebraic number in any form or a factorization matrix (in terms of number field elements, not
ideals!). In the latter case, elements stay in factored form, which is a convenient way to avoid
coefficient explosion; see also idealpow.

Technical note. The routine computes an LLL-reduced basis for the lattice I −1 equipped with
the quadratic form
rX
1 +r2
2
||x||v = 2vi εi |σi (x)|2 ,
i=1

where as usual the σi are the (real and) complex embeddings and εi = 1, resp. 2, for a real,
resp. complex place. The element a is simply the first vector in the LLL basis. The only reason
you may want to try to change some directions and set some vi 6= 0 is to randomize the elements
found for a fixed ideal, which is heuristically useful in index calculus algorithms like bnfinit and
bnfisprincipal.

Even more technical note. In fact, the above is a white lie. We do not use || · ||v exactly but a
rescaled rounded variant which gets us faster and simpler LLLs. There’s no harm since we are not
using any theoretical property of a after all, except that it belongs to I −1 and that aI is “expected
to be small”.
The library syntax is GEN idealred0(GEN nf, GEN I, GEN v = NULL).

3.13.87 idealredmodpower(nf , x, n, {B = primelimit}). Let nf be a number field, x an ideal in

nf and n > 0 be a positive integer. Return a number field element b such that xbn = v is small. If
x is integral, then v is also integral.
More precisely, idealnumden reduces the problem to x integral. Then, factoring out the prime
ideals dividing a rational prime p ≤ B, we rewrite x = IJ n where the ideals I and J are both
integral and I is B-smooth. Then we return a small element b in J −1 .
The bound B avoids a costly complete factorization of x; as soon as the n-core of x is B-smooth
(i.e., as soon as I is n-power free), then J is as large as possible and so is the expected reduction.
? T = x^6+108; nf = nfinit(T); a = Mod(x,T);
? setrand(1); u = (2*a^2+a+3)*random(2^1000*x^6)^6;
? sizebyte(u)
%3 = 4864
? b = idealredmodpower(nf,u,2);
? v2 = nfeltmul(nf,u, nfeltpow(nf,b,2))
%5 = [34, 47, 15, 35, 9, 3]~
? b = idealredmodpower(nf,u,6);
? v6 = nfeltmul(nf,u, nfeltpow(nf,b,6))
%7 = [3, 0, 2, 6, -7, 1]~
The last element v6, obtained by reducing modulo 6-th powers instead of squares, looks smaller
than v2 but its norm is actually a little larger:
? idealnorm(nf,v2)
%8 = 81309

392
 ? idealnorm(nf,v6)
%9 = 731781
The library syntax is GEN idealredmodpower(GEN nf, GEN x, ulong n, ulong B).

3.13.88 idealstar({nf }, N, {flag = 1}, {cycmod }). Outputs a bid structure, necessary for com-
puting in the finite abelian group G = (ZK /N )∗ . Here, nf is a number field and N is a modulus:
either an ideal in any form, or a row vector whose first component is an ideal and whose second
component is a row vector of r1 0 or 1. Ideals can also be given by a factorization into prime ideals,
as produced by idealfactor.
If the positive integer cycmod is present, only compute the group modulo cycmod-th powers,
which may save a lot of time when some maximal ideals in the modulus have a huge residue
field. Whereas you might only be interested in quadratic or cubic residuosity; see also bnrinit for
applications in class field theory.
This bid is used in ideallog to compute discrete logarithms. It also contains useful information
which can be conveniently retrieved as bid .mod (the modulus), bid .clgp (G as a finite abelian
group), bid .no (the cardinality of G), bid .cyc (elementary divisors) and bid .gen (generators).
If flag = 1 (default), the result is a bid structure without generators: they are well defined
but not explicitly computed, which saves time.
If flag = 2, as flag = 1, but including generators.
If flag = 0, only outputs (ZK /N )∗ as an abelian group, i.e as a 3-component vector [h, d, g]: h
is the order, d is the vector of SNF cyclic components and g the corresponding generators.
If nf is omitted, we take it to be the rational number fields, N must be an integer and we
return the structure of (Z/N Z)∗ . In other words idealstar(, N, flag) is short for
idealstar(nfinit(x), N, flag)
but faster. The alternative syntax znstar(N, flag) is also available for an analogous effect but,
due to an unfortunate historical oversight, the default value of flag is different in the two functions
(znstar does not initialize by default, you probably want znstar(N,1)).
The library syntax is GEN idealstarmod(GEN nf = NULL, GEN N, long flag, GEN cycmod =
NULL). Instead the above hardcoded numerical flags, one should rather use GEN Idealstarmod(GEN
nf, GEN ideal, long flag, GEN cycmod) or GEN Idealstar(GEN nf, GEN ideal, long flag)
(cycmod is NULL), where flag is an or-ed combination of nf_GEN (include generators) and nf_INIT
(return a full bid, not a group), possibly 0. This offers one more combination: gen, but no init.

3.13.89 idealtwoelt(nf , x, {a}). Computes a two-element representation of the ideal x in the

number field nf , combining a random search and an approximation theorem; x is an ideal in any
form (possibly an extended ideal, whose principal part is ignored)
• When called as idealtwoelt(nf,x), the result is a row vector [a, α] with two components
such that x = aZK + αZK and a is chosen to be the positive generator of x ∩ Z, unless x was given
as a principal ideal in which case we may choose a = 0. The algorithm uses a fast lazy factorization
of x ∩ Z and runs in randomized polynomial time.
? K = nfinit(t^5-23);
? x = idealhnf(K, t^2*(t+1), t^3*(t+1))
%2 = \\ some random ideal of norm 552*23

393
 [552 23 23 529 23]
[ 0 23 0 0 0]
[ 0 0 1 0 0]
[ 0 0 0 1 0]
[ 0 0 0 0 1]
? [a,alpha] = idealtwoelt(K, x)
%3 = [552, [23, 0, 1, 0, 0]~]
? nfbasistoalg(K, alpha)
%4 = Mod(t^2 + 23, t^5 - 23)
• When called as idealtwoelt(nf,x,a) with an explicit nonzero a supplied as third argument,
the function assumes that a ∈ x and returns α ∈ x such that x = aZK + αZK . Note that we must
factor a in this case, and the algorithm is generally slower than the default variant and gives larger
generators:

? alpha2 = idealtwoelt(K, x, 552)

%5 = [-161, -161, -183, -207, 0]~
? idealhnf(K, 552, alpha2) == x
%6 = 1
Note that, in both cases, the return value is not recognized as an ideal by GP functions; one must
use idealhnf as above to recover a valid ideal structure from the two-element representation.
The library syntax is GEN idealtwoelt0(GEN nf, GEN x, GEN a = NULL). Also available are
GEN idealtwoelt(GEN nf, GEN x) and GEN idealtwoelt2(GEN nf, GEN x, GEN a).

3.13.90 idealval(nf , x, pr ). Gives the valuation of the ideal x at the prime ideal pr in the number
field nf , where pr is in idealprimedec format. The valuation of the 0 ideal is +oo.
The library syntax is GEN gpidealval(GEN nf, GEN x, GEN pr). Also available is long
idealval(GEN nf, GEN x, GEN pr), which returns LONG_MAX if x = 0 and the valuation as a long
integer.

3.13.91 matalgtobasis(nf , x). This function is deprecated, use apply.

nf being a number field in nfinit format, and x a (row or column) vector or matrix, apply
nfalgtobasis to each entry of x.
The library syntax is GEN matalgtobasis(GEN nf, GEN x).

3.13.92 matbasistoalg(nf , x). This function is deprecated, use apply.

nf being a number field in nfinit format, and x a (row or column) vector or matrix, apply
nfbasistoalg to each entry of x.
The library syntax is GEN matbasistoalg(GEN nf, GEN x).

394
3.13.93 modreverse(z). Let z = Mod(A, T) be a polmod, and Q be its minimal polynomial, which
must satisfy deg(Q) = deg(T ). Returns a “reverse polmod” Mod(B, Q), which is a root of T .
This is quite useful when one changes the generating element in algebraic extensions:
? u = Mod(x, x^3 - x -1); v = u^5;
? w = modreverse(v)
%2 = Mod(x^2 - 4*x + 1, x^3 - 5*x^2 + 4*x - 1)
which means that x3 − 5x2 + 4x − 1 is another defining polynomial for the cubic field
Q(u) = Q[x]/(x3 − x − 1) = Q[x]/(x3 − 5x2 + 4x − 1) = Q(v),
and that u → v 2 − 4v + 1 gives an explicit isomorphism. From this, it is easy to convert elements
between the A(u) ∈ Q(u) and B(v) ∈ Q(v) representations:
? A = u^2 + 2*u + 3; subst(lift(A), ’x, w)
%3 = Mod(x^2 - 3*x + 3, x^3 - 5*x^2 + 4*x - 1)
? B = v^2 + v + 1; subst(lift(B), ’x, v)
%4 = Mod(26*x^2 + 31*x + 26, x^3 - x - 1)
If the minimal polynomial of z has lower degree than expected, the routine fails
? u = Mod(-x^3 + 9*x, x^4 - 10*x^2 + 1)
? modreverse(u)
*** modreverse: domain error in modreverse: deg(minpoly(z)) < 4
*** Break loop: type ’break’ to go back to GP prompt
break> Vec( dbg_err() ) \\ ask for more info
["e_DOMAIN", "modreverse", "deg(minpoly(z))", "<", 4,
Mod(-x^3 + 9*x, x^4 - 10*x^2 + 1)]
break> minpoly(u)
x^2 - 8
The library syntax is GEN modreverse(GEN z).
3.13.94 newtonpoly(x, p). Gives the vector of the slopes of the Newton polygon of the polynomial
x with respect to the prime number p. The n components of the vector are in decreasing order,
where n is equal to the degree of x. Vertical slopes occur iff the constant coefficient of x is zero
and are denoted by +oo.
The library syntax is GEN newtonpoly(GEN x, GEN p).
3.13.95 nfalgtobasis(nf , x). Given an algebraic number x in the number field nf , transforms it
to a column vector on the integral basis nf .zk.
? nf = nfinit(y^2 + 4);
? nf.zk
%2 = [1, 1/2*y]
? nfalgtobasis(nf, [1,1]~)
%3 = [1, 1]~
? nfalgtobasis(nf, y)
%4 = [0, 2]~
? nfalgtobasis(nf, Mod(y, y^2+4))
%5 = [0, 2]~
This is the inverse function of nfbasistoalg.
The library syntax is GEN algtobasis(GEN nf, GEN x).

395
3.13.96 nfbasis(T, {&dK }). Let T (X) be an irreducible polynomial with integral coefficients.
This function returns an integral basis of the number field defined by T , that is a Z-basis of its
maximal order. If present, dK is set to the discriminant of the returned order. The basis elements
are given as elements in K = Q[X]/(T ), in Hermite normal form with respect to the Q-basis
(1, X, . . . , X deg T −1 ) of K, lifted to Q[X]. In particular its first element is always 1 and its i-th
element is a polynomial of degree i − 1 whose leading coefficient is the inverse of an integer: the
product of those integers is the index of Z[X]/(T ) in the maximal order ZK :
? nfbasis(x^2 + 4) \\ Z[X]/(T) has index 2 in Z_K
%1 = [1, x/2]
? nfbasis(x^2 + 4, &D)
%2 = [1, x/2]
? D
%3 = -4
This function uses a modified version of the round 4 algorithm, due to David Ford, Sebastian
Pauli and Xavier Roblot.
Local basis, orders maximal at certain primes.
Obtaining the maximal order is hard: it requires factoring the discriminant D of T . Obtaining
an order which is maximal at a finite explicit set of primes is easy, but it may then be a strict
suborder of the maximal order. To specify that we are interested in a given set of places only,
we can replace the argument T by an argument [T, listP ], where listP encodes the primes we are
interested in: it must be a factorization matrix, a vector of integers or a single integer.
• Vector: we assume that it contains distinct prime numbers.
• Matrix: we assume that it is a two-column matrix of a (partial) factorization of D; namely
the first column contains distinct primes and the second one the valuation of D at each of these
primes.
• Integer B: this is replaced by the vector of primes up to B. Note that the function will
use at least O(B) time: a small value, about 105 , should be enough for most applications. Values
larger than 232 are not supported.
In all these cases, the primes may or may not divide the discriminant D of T . The function
then returns a Z-basis of an order whose index is not divisible by any of these prime numbers.
The result may actually be a global integral basis, in particular if all the prime divisors of the field
discriminant are included, but this is not guaranteed! Note that nfinit has built-in support for
such a check:
? K = nfinit([T, listP]);
? nfcertify(K) \\ we computed an actual maximal order
%2 = [];
The first line initializes a number field structure incorporating nfbasis([T, listP] in place of a
proven integral basis. The second line certifies that the resulting structure is correct. This allows
to create an nf structure attached to the number field K = Q[X]/(T ), when the discriminant of
T cannot be factored completely, whereas the prime divisors of discK are known. If present, the
argument dK is set to the discriminant of the returned order, and is equal to the field discriminant
if and only if the order is maximal.
Of course, if listP contains a single prime number p, the function returns a local integral basis
for Zp [X]/(T ):

396
 ? nfbasis(x^2+x-1001)
%1 = [1, 1/3*x - 1/3]
? nfbasis( [x^2+x-1001, [2]] )
%2 = [1, x]
The following function computes the index iT of Z[X]/(T ) in the order generated by the Z-basis
B:
nfbasisindex(T, B) = vecprod([denominator(pollead(Q)) | Q <- B]);
In particular, B is a basis of the maximal order if and only if poldisc(T )/i2T is equal to the field
discriminant. More generally, this formula gives the square of index of the order given by B in
ZK . For instance, assume that P is a vector of prime numbers containing (at least) all prime
divisors of the field discriminant, then the following construct allows to provably compute the field
discriminant and to check whether the returned basis is actually a basis of the maximal order
? B = nfbasis([T, P], &D);
? dK = sign(D) * vecprod([p^valuation(D,p) | p<-P]);
? dK * nfbasisindex(T, B)^2 == poldisc(T)
The variable dK contains the field discriminant and the last command returns 1 if and only if B is
a Z-basis of the maximal order. Of course, the nfinit / nfcertify approach is simpler, but it is
also more costly.

The Buchmann-Lenstra algorithm.

We now complicate the picture: it is in fact allowed to include composite numbers instead of
primes in listP (Vector or Matrix case), provided they are pairwise coprime. The result may still
be a correct integral basis if the field discriminant factors completely over the actual primes in the
list; again, this is not guaranteed. Adding a composite C such that C 2 divides D may help because
when we consider C as a prime and run the algorithm, two good things can happen: either we
succeed in proving that no prime dividing C can divide the index (without actually needing to find
those primes), or the computation exhibits a nontrivial zero divisor, thereby factoring C and we
go on with the refined factorization. (Note that including a C such that C 2 does not divide D is
useless.) If neither happen, then the computed basis need not generate the maximal order. Here is
an example:
? B = 10^5;
? listP = factor(poldisc(T), B); \\ primes <= B dividing D + cofactor
? basis = nfbasis([T, listP], &D)
If the computed discriminant D factors completely over the primes less than B (together with the
primes contained in the addprimes table), then everything is certified: D is the field discriminant
and basis generates the maximal order. This can be tested as follows:
F = factor(D, B); P = F[,1]; E = F[,2];
for (i = 1, #P,
if (P[i] > B && !isprime(P[i]), warning("nf may be incorrect")));
This is a sufficient but not a necessary condition, hence the warning, instead of an error.
The function nfcertify speeds up and automates the above process:
? B = 10^5;
? nf = nfinit([T, B]);

397
 ? nfcertify(nf)
%3 = [] \\ nf is unconditionally correct
? [basis, disc] = [nf.zk, nf.disc];
The library syntax is GEN nfbasis(GEN T, GEN *dK = NULL).

3.13.97 nfbasistoalg(nf , x). Given an algebraic number x in the number field nf , transforms it
into t_POLMOD form.
? nf = nfinit(y^2 + 4);
? nf.zk
%2 = [1, 1/2*y]
? nfbasistoalg(nf, [1,1]~)
%3 = Mod(1/2*y + 1, y^2 + 4)
? nfbasistoalg(nf, y)
%4 = Mod(y, y^2 + 4)
? nfbasistoalg(nf, Mod(y, y^2+4))
%5 = Mod(y, y^2 + 4)
This is the inverse function of nfalgtobasis.
The library syntax is GEN basistoalg(GEN nf, GEN x).

3.13.98 nfcertify(nf ). nf being as output by nfinit, checks whether the integer basis is known
unconditionally. This is in particular useful when the argument to nfinit was of the form
[T, listP], specifying a finite list of primes when p-maximality had to be proven, or a list of
coprime integers to which Buchmann-Lenstra algorithm was to be applied.
The function returns a vector of coprime composite integers. If this vector is empty, then
nf.zk and nf.disc are correct. Otherwise, the result is dubious. In order to obtain a certified
result, one must completely factor each of the given integers, then addprime each of their prime
factors, then check whether nfdisc(nf.pol) is equal to nf.disc.
The library syntax is GEN nfcertify(GEN nf).

3.13.99 nfcompositum(nf , P, Q, {flag = 0}). Let nf be a number field structure attached to the
field K and let P and Q be squarefree polynomials in K[X] in the same variable. Outputs the
simple factors of the étale K-algebra A = K[X, Y ]/(P (X), Q(Y )). The factors are given by a list
of polynomials R in K[X], attached to the number field K[X]/(R), and sorted by increasing degree
(with respect to lexicographic ordering for factors of equal degrees). Returns an error if one of the
polynomials is not squarefree.
Note that it is more efficient to reduce to the case where P and Q are irreducible first. The
routine will not perform this for you, since it may be expensive, and the inputs are irreducible in
most applications anyway. In this case, there will be a single factor R if and only if the number
fields defined by P and Q are linearly disjoint (their intersection is K).
The binary digits of flag mean
1: outputs a vector of 4-component vectors [R, a, b, k], where R ranges through the list of all
possible compositums as above, and a (resp. b) expresses the root of P (resp. Q) as an element of
K[X]/(R). Finally, k is a small integer such that b + ka = X modulo R.

398
 2: assume that P and Q define number fields that are linearly disjoint: both polynomials are
irreducible and the corresponding number fields have no common subfield besides K. This allows
to save a costly factorization over K. In this case return the single simple factor instead of a vector
with one element.

A compositum is often defined by a complicated polynomial, which it is advisable

√ to reduce
before further work. Here is an example involving the field K(ζ5 , 51/10 ), K = Q( 5):

? K = nfinit(y^2-5);
? L = nfcompositum(K, x^5 - y, polcyclo(5), 1); \\ list of [R, a, b, k]
? [R, a] = L[1]; \\ pick the single factor, extract R, a (ignore b, k)
? lift(R) \\ defines the compositum
%4 = x^10 + (-5/2*y + 5/2)*x^9 + (-5*y + 20)*x^8 + (-20*y + 30)*x^7 + \
(-45/2*y + 145/2)*x^6 + (-71/2*y + 121/2)*x^5 + (-20*y + 60)*x^4 + \
(-25*y + 5)*x^3 + 45*x^2 + (-5*y + 15)*x + (-2*y + 6)
? a^5 - y \\ a fifth root of y
%5 = 0
? [T, X] = rnfpolredbest(K, R, 1);
? lift(T) \\ simpler defining polynomial for K[x]/(R)
%7 = x^10 + (-11/2*y + 25/2)
? liftall(X) \\ root of R in K[x]/(T (x))
%8 = (3/4*y + 7/4)*x^7 + (-1/2*y - 1)*x^5 + 1/2*x^2 + (1/4*y - 1/4)
? a = subst(a.pol, ’x, X); \\ a in the new coordinates
? liftall(a)
%10 = (-3/4*y - 7/4)*x^7 - 1/2*x^2
? a^5 - y
%11 = 0

The main variables of P and Q must be the same and have higher priority than that of nf
(see varhigher and varlower).

The library syntax is GEN nfcompositum(GEN nf, GEN P, GEN Q, long flag).

3.13.100 nfdetint(nf , x). Given a pseudo-matrix x, computes a nonzero ideal contained in

(i.e. multiple of) the determinant of x. This is particularly useful in conjunction with nfhnfmod.

The library syntax is GEN nfdetint(GEN nf, GEN x).

3.13.101 nfdisc(T ). field discriminant of the number field defined by the integral, preferably
monic, irreducible polynomial T (X). Returns the discriminant of the number field Q[X]/(T ),
using the Round 4 algorithm.

399
Local discriminants, valuations at certain primes.
As in nfbasis, the argument T can be replaced by [T, listP ], where listP is as in nfbasis:
a vector of pairwise coprime integers (usually distinct primes), a factorization matrix, or a single
integer. In that case, the function returns the discriminant of an order whose basis is given by
nfbasis(T,listP), which need not be the maximal order, and whose valuation at a prime entry
in listP is the same as the valuation of the field discriminant.
In particular, if listP is [p] for a prime p, we can return the p-adic discriminant of the maximal
order of Zp [X]/(T ), as a power of p, as follows:
? padicdisc(T,p) = p^valuation(nfdisc([T,[p]]), p);
? nfdisc(x^2 + 6)
%2 = -24
? padicdisc(x^2 + 6, 2)
%3 = 8
? padicdisc(x^2 + 6, 3)
%4 = 3
The following function computes the discriminant of the maximal order under the assumption that
P is a vector of prime numbers containing (at least) all prime divisors of the field discriminant:
globaldisc(T, P) =
{ my (D = nfdisc([T, P]));
sign(D) * vecprod([p^valuation(D,p) | p <-P]);
}
? globaldisc(x^2 + 6, [2, 3, 5])
%1 = -24
The library syntax is nfdisc(GEN T). Also available is GEN nfbasis(GEN T, GEN *d), which
returns the order basis, and where *d receives the order discriminant.

3.13.102 nfdiscfactors(T ). Given a polynomial T with integer coefficients, return [D, faD] where
D is nfdisc(T ) and faD is the factorization of |D|. All the variants [T,listP] are allowed (see
??nfdisc), in which case faD is the factorization of the discriminant underlying order (which need
not be maximal at the primes not specified by listP) and the factorization may contain large
composites.
? T = x^3 - 6021021*x^2 + 12072210077769*x - 8092423140177664432;
? [D,faD] = nfdiscfactors(T); print(faD); D
[3, 3; 500009, 2]
%2 = -6750243002187]
? T = x^3 + 9*x^2 + 27*x - 125014250689643346789780229390526092263790263725;
? [D,faD] = nfdiscfactors(T); print(faD); D
[3, 3; 1000003, 2]
%4 = -27000162000243
? [D,faD] = nfdiscfactors([T, 10^3]); print(faD)
[3, 3; 125007125141751093502187, 2]
In the final example, we only get a partial factorization, which is only guaranteed correct at primes
≤ 103 .

400
 The function also accept number field structures, for instance as output by nfinit, and returns
the field discriminant and its factorization:

? T = x^3 + 9*x^2 + 27*x - 125014250689643346789780229390526092263790263725;

? nf = nfinit(T); [D,faD] = nfdiscfactors(T); print(faD); D
%2 = -27000162000243
? nf.disc
%3 = -27000162000243

The library syntax is GEN nfdiscfactors(GEN T).

3.13.103 nfeltadd(nf , x, y). Given two elements x and y in nf , computes their sum x + y in the
number field nf .

? nf = nfinit(1+x^2);
? nfeltadd(nf, 1, x) \\ 1 + I
%2 = [1, 1]~

The library syntax is GEN nfadd(GEN nf, GEN x, GEN y).

3.13.104 nfeltdiv(nf , x, y). Given two elements x and y in nf , computes their quotient x/y in
the number field nf .

The library syntax is GEN nfdiv(GEN nf, GEN x, GEN y).

3.13.105 nfeltdiveuc(nf , x, y). Given two elements x and y in nf , computes an algebraic integer
q in the number field nf such that the components of x − qy are reasonably small. In fact, this is
functionally identical to round(nfdiv(nf ,x,y)).

The library syntax is GEN nfdiveuc(GEN nf, GEN x, GEN y).

3.13.106 nfeltdivmodpr(nf , x, y, pr ). This function is obsolete, use nfmodpr.

Given two elements x and y in nf and pr a prime ideal in modpr format (see nfmodprinit),
computes their quotient x/y modulo the prime ideal pr .

The library syntax is GEN nfdivmodpr(GEN nf, GEN x, GEN y, GEN pr). This function
is normally useless in library mode. Project your inputs to the residue field using nf to Fq, then
work there.

3.13.107 nfeltdivrem(nf , x, y). Given two elements x and y in nf , gives a two-element row vector
[q, r] such that x = qy + r, q is an algebraic integer in nf , and the components of r are reasonably
small.

The library syntax is GEN nfdivrem(GEN nf, GEN x, GEN y).

401
3.13.108 nfeltembed(nf , x, {pl }). Given an element x in the number field nf , return the (real
or) complex embeddings of x specified by optional argument pl , at the current realprecision:
• pl omitted: return the vector of embeddings at all r1 + r2 places;
• pl an integer between 1 and r1 + r2 : return the i-th embedding of x, attached to the i-th
root of nf.pol, i.e. nf.roots[i];
• pl a vector or t_VECSMALL: return the vector of embeddings; the i-th entry gives the embed-
ding at the place attached to the pl [i]-th real root of nf.pol.
? nf = nfinit(’y^3 - 2);
? nf.sign
%2 = [1, 1]
? nfeltembed(nf, ’y)
%3 = [1.25992[...], -0.62996[...] + 1.09112[...]*I]]
? nfeltembed(nf, ’y, 1)
%4 = 1.25992[...]
? nfeltembed(nf, ’y, 3) \\ there are only 2 arch. places
*** at top-level: nfeltembed(nf,’y,3)
*** ^-----------------
*** nfeltembed: domain error in nfeltembed: index > 2
The library syntax is GEN nfeltembed(GEN nf, GEN x, GEN pl = NULL, long prec).

3.13.109 nfeltmod(nf , x, y). Given two elements x and y in nf , computes an element r of nf of

the form r = x − qy with q and algebraic integer, and such that r is small. This is functionally
identical to
x − nfmul(nf , round(nfdiv(nf , x, y)), y).

The library syntax is GEN nfmod(GEN nf, GEN x, GEN y).

3.13.110 nfeltmul(nf , x, y). Given two elements x and y in nf , computes their product x ∗ y in
the number field nf .
The library syntax is GEN nfmul(GEN nf, GEN x, GEN y).

3.13.111 nfeltmulmodpr(nf , x, y, pr ). This function is obsolete, use nfmodpr.

Given two elements x and y in nf and pr a prime ideal in modpr format (see nfmodprinit),
computes their product x ∗ y modulo the prime ideal pr .
The library syntax is GEN nfmulmodpr(GEN nf, GEN x, GEN y, GEN pr). This function
is normally useless in library mode. Project your inputs to the residue field using nf to Fq, then
work there.

3.13.112 nfeltnorm(nf , x). Returns the absolute norm of x.

The library syntax is GEN nfnorm(GEN nf, GEN x).

3.13.113 nfeltpow(nf , x, k). Given an element x in nf , and a positive or negative integer k,

computes xk in the number field nf .
The library syntax is GEN nfpow(GEN nf, GEN x, GEN k). GEN nfinv(GEN nf, GEN x)
correspond to k = −1, and GEN nfsqr(GEN nf, GEN x) to k = 2.

402
3.13.114 nfeltpowmodpr(nf , x, k, pr ). This function is obsolete, use nfmodpr.
Given an element x in nf , an integer k and a prime ideal pr in modpr format (see nfmodprinit),
computes xk modulo the prime ideal pr .
The library syntax is GEN nfpowmodpr(GEN nf, GEN x, GEN k, GEN pr). This function
is normally useless in library mode. Project your inputs to the residue field using nf to Fq, then
work there.

3.13.115 nfeltreduce(nf , a, id ). Given an ideal id in Hermite normal form and an element a of

the number field nf , finds an element r in nf such that a − r belongs to the ideal and r is small.
The library syntax is GEN nfreduce(GEN nf, GEN a, GEN id).

3.13.116 nfeltreducemodpr(nf , x, pr ). This function is obsolete, use nfmodpr.

Given an element x of the number field nf and a prime ideal pr in modpr format compute a
canonical representative for the class of x modulo pr .
The library syntax is GEN nfreducemodpr(GEN nf, GEN x, GEN pr). This function is normally
useless in library mode. Project your inputs to the residue field using nf to Fq, then work there.

3.13.117 nfeltsign(nf , x, {pl }). Given an element x in the number field nf , returns the signs of
the real embeddings of x specified by optional argument pl :
• pl omitted: return the vector of signs at all r1 real places;
• pl an integer between 1 and r1 : return the sign of the i-th embedding of x, attached to the
i-th real root of nf.pol, i.e. nf.roots[i];
• pl a vector or t_VECSMALL: return the vector of signs; the i-th entry gives the sign at the
real place attached to the pl [i]-th real root of nf.pol.
? nf = nfinit(polsubcyclo(11,5,’y)); \\ Q(cos(2 pi/11))
? nf.sign
%2 = [5, 0]
? x = Mod(’y, nf.pol);
? nfeltsign(nf, x)
%4 = [-1, -1, -1, 1, 1]
? nfeltsign(nf, x, 1)
%5 = -1
? nfeltsign(nf, x, [1..4])
%6 = [-1, -1, -1, 1]
? nfeltsign(nf, x, 6) \\ there are only 5 real embeddings
*** at top-level: nfeltsign(nf,x,6)
*** ^-----------------
*** nfeltsign: domain error in nfeltsign: index > 5
The library syntax is GEN nfeltsign(GEN nf, GEN x, GEN pl = NULL).

3.13.118 nfelttrace(nf , x). Returns the absolute trace of x.

The library syntax is GEN nftrace(GEN nf, GEN x).

403
3.13.119 nfeltval(nf , x, pr , {&y}). Given an element x in nf and a prime ideal pr in the format
output by idealprimedec, computes the valuation v at pr of the element x. The valuation of 0 is
+oo.
? nf = nfinit(x^2 + 1);
? P = idealprimedec(nf, 2)[1];
? nfeltval(nf, x+1, P)
%3 = 1
This particular valuation can also be obtained using idealval(nf ,x,pr ), since x is then converted
to a principal ideal.
If the y argument is present, sets y = xτ v , where τ is a fixed “anti-uniformizer” for pr : its
valuation at pr is −1; its valuation is 0 at other prime ideals dividing pr .p and nonnegative at all
other primes. In other words y is the part of x coprime to pr . If x is an algebraic integer, so is y.
? nfeltval(nf, x+1, P, &y); y
%4 = [0, 1]~
For instance if x = i xei i is known to be coprime to pr , where Q
Q
the xi are algebraic integers
and ei ∈ Z then, if vi = nfeltval(nf , xi , pr , &yi ), we still have x = i yiei , where the yi are still
algebraic integers but now all of them are coprime to pr . They can then be mapped to the residue
field of pr more efficiently than if the product had been expanded beforehand: we can reduce mod
pr after each ring operation.
The library syntax is GEN gpnfvalrem(GEN nf, GEN x, GEN pr, GEN *y = NULL). Also
available are long nfvalrem(GEN nf, GEN x, GEN pr, GEN *y = NULL), which returns LONG_MAX
if x = 0 and the valuation as a long integer, and long nfval(GEN nf, GEN x, GEN pr), which
only returns the valuation (y = NULL).

3.13.120 nffactor(nf , T ). Factorization of the univariate polynomial (or rational function) T over
the number field nf given by nfinit; T has coefficients in nf (i.e. either scalar, polmod, polynomial
or column vector). The factors are sorted by increasing degree.
The main variable of nf must be of lower priority than that of T , see Section 2.5.3. However
if the polynomial defining the number field occurs explicitly in the coefficients of T as modulus of
a t_POLMOD or as a t_POL coefficient, its main variable must be the same as the main variable of
T . For example,
? nf = nfinit(y^2 + 1);
? nffactor(nf, x^2 + y); \\ OK
? nffactor(nf, x^2 + Mod(y, y^2+1)); \\ OK
? nffactor(nf, x^2 + Mod(z, z^2+1)); \\ WRONG
It is possible to input a defining polynomial for nf instead, but this is in general less efficient
since parts of an nf structure will then be computed internally. This is useful in two situations: when
you do not need the nf elsewhere, or when you cannot initialize an nf due to integer factorization
difficulties when attempting to compute the field discriminant and maximal order. In all cases,
the function runs in polynomial time using Belabas’s variant of van Hoeij’s algorithm, which copes
with hundreds of modular factors.

404
Caveat. nfinit([T, listP]) allows to compute in polynomial time a conditional nf structure,
which sets nf.zk to an order which is not guaranteed to be maximal at all primes. Always either
use nfcertify first (which may not run in polynomial time) or make sure to input nf.pol instead
of the conditional nf : nffactor is able to recover in polynomial time in this case, instead of
potentially missing a factor.
The library syntax is GEN nffactor(GEN nf, GEN T).

3.13.121 nffactorback(nf , f, {e}). Gives back the nf element corresponding to a factorization.

The integer 1 corresponds to the empty factorization.
If e is present, e and f must be vectors of the same length (e being integral), and the corre-
sponding factorization is the product of the f [i]e[i] .
If not, and f is vector, it is understood as in the preceding case with e a vector of 1s: we return
the product of the f [i]. Finally, f can be a regular factorization matrix.

? nf = nfinit(y^2+1);
? nffactorback(nf, [3, y+1, [1,2]~], [1, 2, 3])
%2 = [12, -66]~
? 3 * (I+1)^2 * (1+2*I)^3
%3 = 12 - 66*I
The library syntax is GEN nffactorback(GEN nf, GEN f, GEN e = NULL).

3.13.122 nffactormod(nf , Q, pr ). This routine is obsolete, use nfmodpr and factormod.

Factors the univariate polynomial Q modulo the prime ideal pr in the number field nf . The
coefficients of Q belong to the number field (scalar, polmod, polynomial, even column vector) and
the main variable of nf must be of lower priority than that of Q (see Section 2.5.3). The prime ideal
pr is either in idealprimedec or (preferred) modprinit format. The coefficients of the polynomial
factors are lifted to elements of nf :

? K = nfinit(y^2+1);
? P = idealprimedec(K, 3)[1];
? nffactormod(K, x^2 + y*x + 18*y+1, P)
%3 =
[x + (2*y + 1) 1]
[x + (2*y + 2) 1]
? P = nfmodprinit(K, P); \\ convert to nfmodprinit format
? nffactormod(K, x^2 + y*x + 18*y+1)
%5 =
[x + (2*y + 1) 1]
[x + (2*y + 2) 1]
Same result, of course, here about 10% faster due to the precomputation.
The library syntax is GEN nffactormod(GEN nf, GEN Q, GEN pr).

405
3.13.123 nfgaloisapply(nf , aut, x). Let nf be a number field as output by nfinit, and let aut
be a Galois automorphism of nf expressed by its image on the field generator (such automorphisms
can be found using nfgaloisconj). The function computes the action of the automorphism aut on
the object x in the number field; x can be a number field element, or an ideal (possibly extended).
Because of possible confusion with elements and ideals, other vector or matrix arguments are
forbidden.
? nf = nfinit(x^2+1);
? L = nfgaloisconj(nf)
%2 = [-x, x]~
? aut = L[1]; /* the nontrivial automorphism */
? nfgaloisapply(nf, aut, x)
%4 = Mod(-x, x^2 + 1)
? P = idealprimedec(nf,5); /* prime ideals above 5 */
? nfgaloisapply(nf, aut, P[2]) == P[1]
%6 = 0 \\ !!!!
? idealval(nf, nfgaloisapply(nf, aut, P[2]), P[1])
%7 = 1
The surprising failure of the equality test (%7) is due to the fact that although the corresponding
prime ideals are equal, their representations are not. (A prime ideal is specified by a uniformizer,
and there is no guarantee that applying automorphisms yields the same elements as a direct ide-
alprimedec call.)
The automorphism can also be given as a column vector, representing the image of Mod(x,
nf.pol) as an algebraic number. This last representation is more efficient and should be preferred
if a given automorphism must be used in many such calls.
? nf = nfinit(x^3 - 37*x^2 + 74*x - 37);
? aut = nfgaloisconj(nf)[2]; \\ an automorphism in basistoalg form
%2 = -31/11*x^2 + 1109/11*x - 925/11
? AUT = nfalgtobasis(nf, aut); \\ same in algtobasis form
%3 = [16, -6, 5]~
? v = [1, 2, 3]~; nfgaloisapply(nf, aut, v) == nfgaloisapply(nf, AUT, v)
%4 = 1 \\ same result...
? for (i=1,10^5, nfgaloisapply(nf, aut, v))
time = 463 ms.
? for (i=1,10^5, nfgaloisapply(nf, AUT, v))
time = 343 ms. \\ but the latter is faster
The library syntax is GEN galoisapply(GEN nf, GEN aut, GEN x).

3.13.124 nfgaloisconj(nf , {flag = 0}, {d}). nf being a number field as output by nfinit, com-
putes the conjugates of a root r of the nonconstant polynomial x = nf [1] expressed as polynomials
in r. This also makes sense when the number field is not Galois since some conjugates may lie in
the field. nf can simply be a polynomial.
If no flags or flag = 0, use a combination of flag 4 and 1 and the result is always complete.
There is no point whatsoever in using the other flags.
If flag = 1, use nfroots: a little slow, but guaranteed to work in polynomial time.

406
 If flag = 4, use galoisinit: very fast, but only applies to (most) Galois fields. If the field
is Galois with weakly super-solvable Galois group (see galoisinit), return the complete list of
automorphisms, else only the identity element. If present, d is assumed to be a multiple of the least
common denominator of the conjugates expressed as polynomial in a root of pol .
This routine can only compute Q-automorphisms, but it may be used to get K-automorphism
for any base field K as follows:
rnfgaloisconj(nfK, R) = \\ K-automorphisms of L = K[X] / (R)
{
my(polabs, N,al,S, ala,k, vR);
R *= Mod(1, nfK.pol); \\ convert coeffs to polmod elts of K
vR = variable(R);
al = Mod(variable(nfK.pol),nfK.pol);
[polabs,ala,k] = rnfequation(nfK, R, 1);
Rt = if(k==0,R,subst(R,vR,vR-al*k));
N = nfgaloisconj(polabs) % Rt; \\ Q-automorphisms of L
S = select(s->subst(Rt, vR, Mod(s,Rt)) == 0, N);
if (k==0, S, apply(s->subst(s,vR,vR+k*al)-k*al,S));
}
K = nfinit(y^2 + 7);
rnfgaloisconj(K, x^4 - y*x^3 - 3*x^2 + y*x + 1) \\ K-automorphisms of L
The library syntax is GEN galoisconj0(GEN nf, long flag, GEN d = NULL, long prec)
. Use directly GEN galoisconj(GEN nf, GEN d), corresponding to flag = 0, the others only have
historical interest.

3.13.125 nfgrunwaldwang(nf , Lpr , Ld , pl , {v =0 x}). Given nf a number field in nf or bnf

format, a t_VEC Lpr of primes of nf and a t_VEC Ld of positive integers of the same length, a
t_VECSMALL pl of length r1 the number of real places of nf , computes a polynomial with coefficients
in nf defining a cyclic extension of nf of minimal degree satisfying certain local conditions:
• at the prime Lpr[i], the extension has local degree a multiple of Ld[i];
• at the i-th real place of nf , it is complex if pl[i] = −1 (no condition if pl[i] = 0).
The extension has degree the LCM of the local degrees. Currently, the degree is restricted to
be a prime power for the search, and to be prime for the construction because of the rnfkummer
restrictions.
When nf is Q, prime integers are accepted instead of prid structures. However, their primality
is not checked and the behavior is undefined if you provide a composite number.

Warning. If the number field nf does not contain the n-th roots of unity where n is the degree of
the extension to be computed, the function triggers the computation of the bnf of nf (ζn ), which
may be costly.
? nf = nfinit(y^2-5);
? pr = idealprimedec(nf,13)[1];
? pol = nfgrunwaldwang(nf, [pr], [2], [0,-1], ’x)
%3 = x^2 + Mod(3/2*y + 13/2, y^2 - 5)
The library syntax is GEN nfgrunwaldwang(GEN nf, GEN Lpr, GEN Ld, GEN pl, long v =
-1) where v is a variable number.

407
3.13.126 nfhilbert(nf , a, b, {pr }). If pr is omitted, compute the global quadratic Hilbert symbol
(a, b) in nf , that is 1 if x2 − ay 2 − bz 2 has a non trivial solution (x, y, z) in nf , and −1 otherwise.
Otherwise compute the local symbol modulo the prime ideal pr , as output by idealprimedec.
The library syntax is long nfhilbert0(GEN nf, GEN a, GEN b, GEN pr = NULL).
Also available is long nfhilbert(GEN bnf, GEN a, GEN b) (global quadratic Hilbert symbol).

3.13.127 nfhnf(nf , x, {flag = 0}). Given a pseudo-matrix (A, I), finds a pseudo-basis (B, J) in
Hermite normal form of the module it generates. If flag is nonzero, also return the transformation
matrix U such that AU = [0|B].
The library syntax is GEN nfhnf0(GEN nf, GEN x, long flag). Also available:
GEN nfhnf(GEN nf, GEN x) (flag = 0).
GEN rnfsimplifybasis(GEN bnf, GEN x) simplifies the pseudo-basis x = (A, I), returning a
pseudo-basis (B, J). The ideals in the list J are integral, primitive and either trivial (equal to the
full ring of integer) or nonprincipal.

3.13.128 nfhnfmod(nf , x, detx ). Given a pseudo-matrix (A, I) and an ideal detx which is con-
tained in (read integral multiple of) the determinant of (A, I), finds a pseudo-basis in Hermite
normal form of the module generated by (A, I). This avoids coefficient explosion. detx can be
computed using the function nfdetint.
The library syntax is GEN nfhnfmod(GEN nf, GEN x, GEN detx).

3.13.129 nfinit(pol , {flag = 0}). pol being a nonconstant irreducible polynomial in Q[X], prefer-
ably monic and integral, initializes a number field structure (nf) attached to the field K defined
by pol . As such, it’s a technical object passed as the first argument to most nfxxx functions, but
it contains some information which may be directly useful. Access to this information via member
functions is preferred since the specific data organization given below may change in the future.
Currently, nf is a row vector with 9 components:
nf [1] contains the polynomial pol (nf .pol).
nf [2] contains [r1, r2] (nf .sign, nf .r1, nf .r2), the number of real and complex places of K.
nf [3] contains the discriminant d(K) (nf .disc) of K.
nf [4] contains the index of nf [1] (nf .index), i.e. [ZK : Z[θ]], where θ is any root of nf [1].
nf [5] is a vector containing 7 matrices M , G, roundG, T , MD, TI , MDI and a vector vP
defined as follows:
• M is the (r1+r2)×n matrix whose columns represent the numerical values of the conjugates
of the elements of the integral basis.
• G is an n × n matrix such that T 2 = t GG, where T 2 is the quadratic form T2 (x) =
|σ(x)|2 , σ running over the embeddings of K into C.
P

• roundG is a rescaled copy of G, rounded to nearest integers.

• T is the n × n matrix whose coefficients are Tr(ωi ωj ) where the ωi are the elements of
the integral basis. Note also that det(T ) is equal to the discriminant of the field K. Also, when
understood as an ideal, the matrix T −1 generates the codifferent ideal.

408
 • The columns of M D (nf .diff) express a Z-basis of the different of K on the integral
basis.
• TI is equal to the primitive part of T −1 , which has integral coefficients.
• MDI is a two-element representation (for faster ideal product) of d(K) times the codifferent
ideal (nf .disc∗nf .codiff, which is an integral ideal). This is used in idealinv.
• vP is the list of prime divisors of the field discriminant, i.e, the ramified primes (nf .p);
nfdiscfactors(nf) is the preferred way to access that information.
nf [6] is the vector containing the r1+r2 roots (nf .roots) of nf [1] corresponding to the r1+r2
embeddings of the number field into C (the first r1 components are real, the next r2 have positive
imaginary part).
nf [7] is a Z-basis for dZK , where d = [ZK : Z(θ)], expressed on the powers of θ. The
multiplication by d ensures that all polynomials have integral coefficients and nf [7]/d (nf .zk) is
an integral basis for ZK . Its first element is guaranteed to be 1. This basis is LLL-reduced with
respect to T2 (strictly speaking, it is a permutation of such a basis, due to the condition that the
first element be 1).
nf [8] is the n × n integral matrix expressing the power basis in terms of the integral basis, and
finally
nf [9] is the n × n2 matrix giving the multiplication table of the integral basis.
If a non monic or non integral polynomial is input, nfinit will transform it, and return a
structure attached to the new (monic integral) polynomial together with the attached change of
variables, see flag = 3. It is allowed, though not very useful given the existence of nfnewprec, to
input a nf or a bnf instead of a polynomial. It is also allowed to input a rnf , in which case an nf
structure attached to the absolute defining polynomial polabs is returned (flagis then ignored).
? nf = nfinit(x^3 - 12); \\ initialize number field Q[X] / (X^3 - 12)
? nf.pol \\ defining polynomial
%2 = x^3 - 12
? nf.disc \\ field discriminant
%3 = -972
? nf.index \\ index of power basis order in maximal order
%4 = 2
? nf.zk \\ integer basis, lifted to Q[X]
%5 = [1, x, 1/2*x^2]
? nf.sign \\ signature
%6 = [1, 1]
? factor(abs(nf.disc )) \\ determines ramified primes
%7 =
[2 2]
[3 5]
? idealfactor(nf, 2)
%8 =
[[2, [0, 0, -1]~, 3, 1, [0, 1, 0]~] 3] \\ p32

409
Huge discriminants, helping nfdisc.
In case pol has a huge discriminant which is difficult to factor, it is hard to compute from
scratch the maximal order. The following special input formats are also accepted:
• [pol , B] where pol is a monic integral polynomial and B is the lift of an integer basis, as
would be computed by nfbasis: a vector of polynomials with first element 1 (implicitly modulo
pol ). This is useful if the maximal order is known in advance.
• [pol , B, P ] where pol and B are as above (a monic integral polynomial and the lift of an
integer basis), and P is the list of ramified primes in the extension.
• [pol , listP] where pol is a rational polynomial and listP specifies a list of primes as in
nfbasis. Instead of the maximal order, nfinit then computes an order which is maximal at these
particular primes as well as the primes contained in the private prime table, see addprimes. The
result has a good chance of being correct when the discriminant nf.disc factors completely over
this set of primes but this is not guaranteed. The function nfcertify automates this:
? pol = polcompositum(x^5 - 101, polcyclo(7))[1];
? nf = nfinit( [pol, 10^3] );
? nfcertify(nf)
%3 = []
A priori, nf.zk defines an order which is only known to be maximal at all primes ≤ 103 (no prime
≤ 103 divides nf.index). The certification step proves the correctness of the computation. Had it
failed, that particular nf structure could not have been trusted and may have caused routines using
it to fail randomly. One particular function that remains trustworthy in all cases is idealprimedec
when applied to a prime included in the above list of primes or, more generally, a prime not dividing
any entry in nfcertify output.

If flag = 2: pol is changed into another polynomial P defining the same number field, which is as
simple as can easily be found using the polredbest algorithm, and all the subsequent computations
are done using this new polynomial. In particular, the first component of the result is the modified
polynomial.
If flag = 3, apply polredbest as in case 2, but outputs [nf , Mod(a, P )], where nf is as before
and Mod(a, P ) = Mod(x, pol ) gives the change of variables. This is implicit when pol is not monic
or not integral: first a linear change of variables is performed, to get a monic integral polynomial,
then polredbest.
The library syntax is GEN nfinit0(GEN pol, long flag, long prec). Also available are
GEN nfinit(GEN x, long prec) (flag = 0), GEN nfinitred(GEN x, long prec) (flag = 2), GEN
nfinitred2(GEN x, long prec) (flag = 3). Instead of the above hardcoded numerical flags in
nfinit0, one should rather use
GEN nfinitall(GEN x, long flag, long prec), where flag is an or-ed combination of
• nf_RED: find a simpler defining polynomial,
• nf_ORIG: if nf_RED set, also return the change of variable,
• nf_ROUND2: Deprecated . Slow down the routine by using an obsolete normalization algorithm
(do not use this one!),
• nf_PARTIALFACT: Deprecated . Lazy factorization of the polynomial discriminant. Result is
conditional unless nfcertify can certify it.

410
3.13.130 nfisideal(nf , x). Returns 1 if x is an ideal in the number field nf , 0 otherwise.
The library syntax is long isideal(GEN nf, GEN x).

3.13.131 nfisincl(f, g, {flag = 0}). Let f and g define number fields, where f and g are irreducible
polynomials in Q[X] and nf structures as output by nfinit. Tests whether the number field f
is conjugate to a subfield of the field g. If they are not, the output is the integer 0. If they are,
the output is a vector of polynomials (flag = 0, default) or a single polynomial flag = 1, each
polynomial a representing an embedding i.e. being such that g | f ◦ a. If either f or g is not
irreducible, the result is undefined.
? T = x^6 + 3*x^4 - 6*x^3 + 3*x^2 + 18*x + 10;
? U = x^3 + 3*x^2 + 3*x - 2
? v = nfisincl(U, T);
%2 = [24/179*x^5-27/179*x^4+80/179*x^3-234/179*x^2+380/179*x+94/179]
? subst(U, x, Mod(v[1],T))
%3 = Mod(0, x^6 + 3*x^4 - 6*x^3 + 3*x^2 + 18*x + 10)
? #nfisincl(x^2+1, T) \\ two embeddings
%4 = 2
\\ same result with nf structures
? nfisincl(U, L = nfinit(T)) == v
%5 = 1
? nfisincl(K = nfinit(U), T) == v
%6 = 1
? nfisincl(K, L) == v
%7 = 1
\\ comparative bench: an nf is a little faster, esp. for the subfield
? B = 10^3;
? for (i=1, B, nfisincl(U,T))
time = 712 ms.
? for (i=1, B, nfisincl(K,T))
time = 485 ms.
? for (i=1, B, nfisincl(U,L))
time = 704 ms.
? for (i=1, B, nfisincl(K,L))
time = 465 ms.
Using an nf structure for the potential subfield is faster if the structure is already available. On
the other hand, the gain in nfisincl is usually not sufficient to make it worthwhile to initialize
only for that purpose.
? for (i=1, B, nfinit(U))
time = 308 ms.
The library syntax is GEN nfisincl0(GEN f, GEN g, long flag). Also available is GEN
nfisisom(GEN a, GEN b) (flag = 0).

411
3.13.132 nfisisom(f, g). As nfisincl, but tests for isomorphism. More efficient if f or g is a
number field structure.

? f = x^6 + 30*x^5 + 495*x^4 + 1870*x^3 + 16317*x^2 - 22560*x + 59648;

? g = x^6 + 42*x^5 + 999*x^4 + 8966*x^3 + 36117*x^2 + 21768*x + 159332;
? h = x^6 + 30*x^5 + 351*x^4 + 2240*x^3 + 10311*x^2 + 35466*x + 58321;
? #nfisisom(f,g) \\ two isomorphisms
%3 = 2
? nfisisom(f,h) \\ not isomorphic
%4 = 0
\\ comparative bench
? K = nfinit(f); L = nfinit(g); B = 10^3;
? for (i=1, B, nfisisom(f,g))
time = 6,124 ms.
? for (i=1, B, nfisisom(K,g))
time = 3,356 ms.
? for (i=1, B, nfisisom(f,L))
time = 3,204 ms.
? for (i=1, B, nfisisom(K,L))
time = 3,173 ms.

The function is usually very fast when the fields are nonisomorphic, whenever the fields can be
distinguished via a simple invariant such as degree, signature or discriminant. It may be slower
when the fields share all invariants, but still faster than computing actual isomorphisms:

\\ usually very fast when the answer is ’no’:

? for (i=1, B, nfisisom(f,h))
time = 32 ms.
\\ but not always
? u = x^6 + 12*x^5 + 6*x^4 - 377*x^3 - 714*x^2 + 5304*x + 15379
? v = x^6 + 12*x^5 + 60*x^4 + 166*x^3 + 708*x^2 + 6600*x + 23353
? nfisisom(u,v)
%13 = 0
? polsturm(u) == polsturm(v)
%14 = 1
? nfdisc(u) == nfdisc(v)
%15 = 1
? for(i=1,B, nfisisom(u,v))
time = 1,821 ms.
? K = nfinit(u); L = nfinit(v);
? for(i=1,B, nfisisom(K,v))
time = 232 ms.

The library syntax is GEN nfisisom(GEN f, GEN g).

412
3.13.133 nfislocalpower(nf , pr , a, n). Let nf be a nf structure attached to a number field K, let
a ∈ K and let pr be a prid structure attached to a maximal ideal v. Return 1 if a is an n-th power
in the completed local field Kv , and 0 otherwise.
? K = nfinit(y^2+1);
? P = idealprimedec(K,2)[1]; \\ the ramified prime above 2
? nfislocalpower(K,P,-1, 2) \\ -1 is a square
%3 = 1
? nfislocalpower(K,P,-1, 4) \\ ... but not a 4-th power
%4 = 0
? nfislocalpower(K,P,2, 2) \\ 2 is not a square
%5 = 0
? Q = idealprimedec(K,5)[1]; \\ a prime above 5
? nfislocalpower(K,Q, [0, 32]~, 30) \\ 32*I is locally a 30-th power
%7 = 1
The library syntax is long nfislocalpower(GEN nf, GEN pr, GEN a, GEN n).

3.13.134 nfkermodpr(nf , x, pr ). This function is obsolete, use nfmodpr.

Kernel of the matrix a in ZK /pr , where pr is in modpr format (see nfmodprinit).
The library syntax is GEN nfkermodpr(GEN nf, GEN x, GEN pr). This function is normally
useless in library mode. Project your inputs to the residue field using nfM to FqM, then work there.

3.13.135 nfmodpr(nf , x, pr ). Map x to a t_FFELT in the residue field modulo pr . The argument
pr is either a maximal ideal in idealprimedec format or, preferably, a modpr structure from
nfmodprinit. The function nfmodprlift allows to lift back to ZK .
Note that the function applies to number field elements and not to vector / matrices / poly-
nomials of such. Use apply to convert recursive structures.
? K = nfinit(y^3-250);
? P = idealprimedec(K, 5)[2];
? modP = nfmodprinit(K, P, ’t);
? K.zk
%4 = [1, 1/5*y, 1/25*y^2]
? apply(t->nfmodpr(K,t,modP), K.zk)
%5 = [1, t, 2*t + 1]
? %[1].mod
%6 = t^2 + 3*t + 4
? K.index
%7 = 125
For clarity, we represent elements in the residue field F5 [t]/(T ) as polynomials in the variable t.
Whenever the underlying rational prime does not divide K.index, it is actually the case that t is
the reduction of y in Q[y]/(K.pol) modulo an irreducible factor of K.pol over Fp . In the above
example, 5 divides the index and t is actually the reduction of y/5.
The library syntax is GEN nfmodpr(GEN nf, GEN x, GEN pr).

413
3.13.136 nfmodprinit(nf , pr , {v = variable(nf .pol )}). Transforms the prime ideal pr into modpr
format necessary for all operations modulo pr in the number field nf . The functions nfmodpr and
nfmodprlift allow to project to and lift from the residue field. The variable v is used to display
finite field elements (see ffgen).

? K = nfinit(y^3-250);
? P = idealprimedec(K, 5)[2];
? modP = nfmodprinit(K, P, ’t);
? K.zk
%4 = [1, 1/5*y, 1/25*y^2]
? apply(t->nfmodpr(K,t,modP), K.zk)
%5 = [1, t, 2*t + 1]
? %[1].mod
%6 = t^2 + 3*t + 4
? K.index
%7 = 125
For clarity, we represent elements in the residue field F5 [t]/(T ) as polynomials in the variable t.
Whenever the underlying rational prime does not divide K.index, it is actually the case that t is
the reduction of y in Q[y]/(K.pol) modulo an irreducible factor of K.pol over Fp . In the above
example, 5 divides the index and t is actually the reduction of y/5.
The library syntax is GEN nfmodprinit0(GEN nf, GEN pr, long v) = -1) where v) is a
variable number.

3.13.137 nfmodprlift(nf , x, pr ). Lift the t_FFELT x (from nfmodpr) in the residue field modulo
pr to the ring of integers. Vectors and matrices are also supported. For polynomials, use apply
and the present function.
The argument pr is either a maximal ideal in idealprimedec format or, preferably, a modpr
structure from nfmodprinit. There are no compatibility checks to try and decide whether x is
attached the same residue field as defined by pr : the result is undefined if not.
The function nfmodpr allows to reduce to the residue field.

? K = nfinit(y^3-250);
? P = idealprimedec(K, 5)[2];
? modP = nfmodprinit(K,P);
? K.zk
%4 = [1, 1/5*y, 1/25*y^2]
? apply(t->nfmodpr(K,t,modP), K.zk)
%5 = [1, y, 2*y + 1]
? nfmodprlift(K, %, modP)
%6 = [1, 1/5*y, 2/5*y + 1]
? nfeltval(K, %[3] - K.zk[3], P)
%7 = 1
The library syntax is GEN nfmodprlift(GEN nf, GEN x, GEN pr).

414
3.13.138 nfnewprec(nf ). Transforms the number field nf into the corresponding data using
current (usually larger) precision. This function works as expected if nf is in fact a bnf or a
bnr (update structure to current precision). If the original bnf structure was not computed by
bnfinit(,1), then this may be quite slow and even fail: many generators of principal ideals have
to be computed and the algorithm may fail because the accuracy is not sufficient to bootstrap the
required generators and fundamental units.

The library syntax is GEN nfnewprec(GEN nf, long prec). See also GEN bnfnewprec(GEN
bnf, long prec) and GEN bnrnewprec(GEN bnr, long prec).

3.13.139 nfpolsturm(nf , T, {pl }). Given a polynomial T with coefficients in the number field nf ,
returns the number of real roots of the s(T ) where s runs through the real embeddings of the field
specified by optional argument pl :

• pl omitted: all r1 real places;

• pl an integer between 1 and r1 : the embedding attached to the i-th real root of nf.pol, i.e.
nf.roots[i];

• pl a vector or t_VECSMALL: the embeddings attached to the pl [i]-th real roots of nf.pol.

? nf = nfinit(’y^2 - 2);
? nf.sign
%2 = [2, 0]
? nf.roots
%3 = [-1.414..., 1.414...]
? T = x^2 + ’y;
? nfpolsturm(nf, T, 1) \\ subst(T,y,sqrt(2)) has two real roots
%5 = 2
? nfpolsturm(nf, T, 2) \\ subst(T,y,-sqrt(2)) has no real root
%6 = 0
? nfpolsturm(nf, T) \\ all embeddings together
%7 = [2, 0]
? nfpolsturm(nf, T, [2,1]) \\ second then first embedding
%8 = [0, 2]
? nfpolsturm(nf, x^3) \\ number of distinct roots !
%9 = [1, 1]
? nfpolsturm(nf, x, 6) \\ there are only 2 real embeddings !
*** at top-level: nfpolsturm(nf,x,6)
*** ^-----------------
*** nfpolsturm: domain error in nfpolsturm: index > 2

The library syntax is GEN nfpolsturm(GEN nf, GEN T, GEN pl = NULL).

415
3.13.140 nfroots({nf }, x). Roots of the polynomial x in the number field nf given by nfinit
without multiplicity (in Q if nf is omitted). x has coefficients in the number field (scalar, polmod,
polynomial, column vector). The main variable of nf must be of lower priority than that of x
(see Section 2.5.3). However if the coefficients of the number field occur explicitly (as polmods)
as coefficients of x, the variable of these polmods must be the same as the main variable of t (see
nffactor).
It is possible to input a defining polynomial for nf instead, but this is in general less efficient
since parts of an nf structure will then be computed internally. This is useful in two situations: when
you do not need the nf elsewhere, or when you cannot initialize an nf due to integer factorization
difficulties when attempting to compute the field discriminant and maximal order.

Caveat. nfinit([T, listP]) allows to compute in polynomial time a conditional nf structure,

which sets nf.zk to an order which is not guaranteed to be maximal at all primes. Always either use
nfcertify first (which may not run in polynomial time) or make sure to input nf.pol instead of
the conditional nf : nfroots is able to recover in polynomial time in this case, instead of potentially
missing a factor.
The library syntax is GEN nfroots(GEN nf = NULL, GEN x). See also GEN nfrootsQ(GEN x)
, corresponding to nf = NULL.

3.13.141 nfrootsof1(nf ). Returns a two-component vector [w, z] where w is the number of roots
of unity in the number field nf , and z is a primitive w-th root of unity. It is possible to input a
defining polynomial for nf instead.
? K = nfinit(polcyclo(11));
? nfrootsof1(K)
%2 = [22, [0, 0, 0, 0, 0, -1, 0, 0, 0, 0]~]
? z = nfbasistoalg(K, %[2]) \\ in algebraic form
%3 = Mod(-x^5, x^10 + x^9 + x^8 + x^7 + x^6 + x^5 + x^4 + x^3 + x^2 + x + 1)
? [lift(z^11), lift(z^2)] \\ proves that the order of z is 22
%4 = [-1, -x^9 - x^8 - x^7 - x^6 - x^5 - x^4 - x^3 - x^2 - x - 1]
This function guesses the number w as the gcd of the #k(v)∗ for unramified v above odd primes,
then computes the roots in nf of the w-th cyclotomic polynomial. The algorithm is polynomial
time with respect to the field degree and the bitsize of the multiplication table in nf (both of them
polynomially bounded in terms of the size of the discriminant). Fields of degree up to 100 or so
should require less than one minute.
The library syntax is GEN nfrootsof1(GEN nf).

3.13.142 nfsnf(nf , x, {flag = 0}). Given a torsion ZK -module x attached to the square integral
invertible pseudo-matrix (A, I, J), returns an ideal list D = [d1 , . . . , dn ] which is the Smith normal
form of x. In other words, x is isomorphic to ZK /d1 ⊕ · · · ⊕ ZK /dn and di divides di−1 for i ≥ 2.
If flag is nonzero return [D, U, V ], where U AV is the identity.
See Section 3.13.4 for the definition of integral pseudo-matrix; briefly, it is input as a 3-
component row vector [A, I, J] where I = [b1 , . . . , bn ] and J = [a1 , . . . , an ] are two ideal lists, and A
is a square n × n matrix with columns (A1 , . . . , An ), seen as elements in K n (with canonical basis
(e1 , . . . , en )). This data defines the ZK module x given by

(b1 e1 ⊕ · · · ⊕ bn en )/(a1 A1 ⊕ · · · ⊕ an An ) ,

416
The integrality condition is ai,j ∈ bi a−1
j for all i, j. If it is not satisfied, then the di will not be
integral. Note that every finitely generated torsion module is isomorphic to a module of this form
and even with bi = ZK for all i.
The library syntax is GEN nfsnf0(GEN nf, GEN x, long flag). Also available:
GEN nfsnf(GEN nf, GEN x) (flag = 0).

3.13.143 nfsolvemodpr(nf , a, b, P ). This function is obsolete, use nfmodpr.

Let P be a prime ideal in modpr format (see nfmodprinit), let a be a matrix, invertible
over the residue field, and let b be a column vector or matrix. This function returns a solution of
a · x = b; the coefficients of x are lifted to nf elements.
? K = nfinit(y^2+1);
? P = idealprimedec(K, 3)[1];
? P = nfmodprinit(K, P);
? a = [y+1, y; y, 0]; b = [1, y]~
? nfsolvemodpr(K, a,b, P)
%5 = [1, 2]~
The library syntax is GEN nfsolvemodpr(GEN nf, GEN a, GEN b, GEN P). This function is
normally useless in library mode. Project your inputs to the residue field using nfM to FqM, then
work there.

3.13.144 nfsplitting(P, {d}). Defining polynomial over Q for the splitting field of P ∈ Q[x], that
is the smallest field over which P is totally split. If irreducible, the polynomial P can also be given
by a nf structure, which is more efficient. If d is given, it must be a multiple of the splitting field
degree. Note that if P is reducible the splitting field degree can be smaller than the degree of P .
? K = nfinit(x^3-2);
? nfsplitting(K)
%2 = x^6 + 108
? nfsplitting(x^8-2)
%3 = x^16 + 272*x^8 + 64
? S = nfsplitting(x^6-8) // reducible
%4 = x^4+2*x^2+4
? lift(nfroots(subst(S,x,a),x^6-8))
%5 = [-a,a,-1/2*a^3-a,-1/2*a^3,1/2*a^3,1/2*a^3+a]
Specifying the degree of the splitting field can make the computation faster.
? nfsplitting(x^17-123);
time = 3,607 ms.
? poldegree(%)
%2 = 272
? nfsplitting(x^17-123,272);
time = 150 ms.
? nfsplitting(x^17-123,273);
*** nfsplitting: Warning: ignoring incorrect degree bound 273
time = 3,611 ms.
The complexity of the algorithm is polynomial in the degree d of the splitting field and the bitsize
of T ; if d is large the result will likely be unusable, e.g. nfinit will not be an option:

417
 ? nfsplitting(x^6-x-1)
[... degree 720 polynomial deleted ...]
time = 11,020 ms.
The library syntax is GEN nfsplitting(GEN P, GEN d = NULL).

3.13.145 nfsubfields(pol , {d = 0}, {fl = 0}). Finds all subfields of degree d of the number field
defined by the (monic, integral) polynomial pol (all subfields if d is null or omitted). The result is a
vector of subfields, each being given by [g, h] (default) or simply g (flag=1), where g is an absolute
equation and h expresses one of the roots of g in terms of the root x of the polynomial defining nf .
This routine uses
• Allombert’s galoissubfields when nf is Galois (with weakly supersolvable Galois group).
• Klüners’s or van Hoeij–Klüners–Novocin algorithm in the general case. The latter runs in
polynomial time and is generally superior unless there exists a small unramified prime p such that
pol has few irreducible factors modulo p.
An input of the form [nf, fa] is also allowed, where fa is the factorisation of nf.pol over nf ,
expressed as a famat of polynomials with coefficients in the variable of nf, in which case the van
Hoeij–Klüners–Novocin algorithm is used.
? pol = x^4 - x^3 - x^2 + x + 1;
? nfsubfields(pol)
%2 = [[x, 0], [x^2 - x + 1, x^3 - x^2 + 1], [x^4 - x^3 - x^2 + x + 1, x]]
? nfsubfields(pol,,1)
%2 = [x, x^2 - x + 1, x^4 - x^3 - x^2 + x + 1]
? y=varhigher("y"); fa = nffactor(pol,subst(pol,x,y));
? #nfsubfields([pol,fa])
%5 = 3
The library syntax is GEN nfsubfields0(GEN pol, long d, long fl). Also available is GEN
nfsubfields(GEN nf, long d), corresponding to flag= 0.

3.13.146 nfsubfieldscm(nf , {fl = 0}). Compute the maximal CM subfield of nf . Return 0 if nf

does not have a CM subfield, otherwise return [g, h] (default) or g (flag=1) where g is an absolute
equation and h expresses a root of g in terms of the generator of nf . Moreover, the CM involution
is given by X mod g(X) 7→ −X mod g(X), i.e. X mod g(X) is a totally imaginary element.
An input of the form [nf, fa] is also allowed, where fa is the factorisation of nf.pol over nf ,
and nf is also allowed to be a monic defining polynomial for the number field.
? nf = nfinit(x^8 + 20*x^6 + 10*x^4 - 4*x^2 + 9);
? nfsubfieldscm(nf)
%2 = [x^4 + 4480*x^2 + 3612672, 3*x^5 + 58*x^3 + 5*x]
? pol = y^16-8*y^14+29*y^12-60*y^10+74*y^8-48*y^6+8*y^4+4*y^2+1;
? fa = nffactor(pol, subst(pol,y,x));
? nfsubfieldscm([pol,fa])
%5 = [y^8 + ... , ...]
The library syntax is GEN nfsubfieldscm(GEN nf, long fl).

418
3.13.147 nfsubfieldsmax(nf , {fl = 0}). Compute the list of maximal subfields of nf . The result
is a vector as in nfsubfields.
An input of the form [nf, fa] is also allowed, where fa is the factorisation of nf.pol over nf ,
and nf is also allowed to be a monic defining polynomial for the number field.
The library syntax is GEN nfsubfieldsmax(GEN nf, long fl).

3.13.148 polcompositum(P, Q, {flag = 0}). P and Q being squarefree polynomials in Z[X] in

the same variable, outputs the simple factors of the étale Q-algebra A = Q(X, Y )/(P (X), Q(Y )).
The factors are given by a list of polynomials R in Z[X], attached to the number field Q(X)/(R),
and sorted by increasing degree (with respect to lexicographic ordering for factors of equal degrees).
Returns an error if one of the polynomials is not squarefree.
Note that it is more efficient to reduce to the case where P and Q are irreducible first. The
routine will not perform this for you, since it may be expensive, and the inputs are irreducible in
most applications anyway. In this case, there will be a single factor R if and only if the number
fields defined by P and Q are linearly disjoint (their intersection is Q).
Assuming P is irreducible (of smaller degree than Q for efficiency), it is in general much faster
to proceed as follows
nf = nfinit(P); L = nffactor(nf, Q)[,1];
vector(#L, i, rnfequation(nf, L[i]))
to obtain the same result. If you are only interested in the degrees of the simple factors, the
rnfequation instruction can be replaced by a trivial poldegree(P) * poldegree(L[i]).
The binary digits of flag mean
1: outputs a vector of 4-component vectors [R, a, b, k], where R ranges through the list of all
possible compositums as above, and a (resp. b) expresses the root of P (resp. Q) as an element of
Q(X)/(R). Finally, k is a small integer such that b + ka = X modulo R.
2: assume that P and Q define number fields which are linearly disjoint: both polynomials are
irreducible and the corresponding number fields have no common subfield besides Q. This allows
to save a costly factorization over Q. In this case return the single simple factor instead of a vector
with one element.
A compositum is often defined by a complicated polynomial, which it is advisable to reduce
before further work. Here is an example involving the field Q(ζ5 , 51/5 ):
? L = polcompositum(x^5 - 5, polcyclo(5), 1); \\ list of [R, a, b, k]
? [R, a] = L[1]; \\ pick the single factor, extract R, a (ignore b, k)
? R \\ defines the compositum
%3 = x^20 + 5*x^19 + 15*x^18 + 35*x^17 + 70*x^16 + 141*x^15 + 260*x^14\
+ 355*x^13 + 95*x^12 - 1460*x^11 - 3279*x^10 - 3660*x^9 - 2005*x^8 \
+ 705*x^7 + 9210*x^6 + 13506*x^5 + 7145*x^4 - 2740*x^3 + 1040*x^2 \
- 320*x + 256
? a^5 - 5 \\ a fifth root of 5
%4 = 0
? [T, X] = polredbest(R, 1);
? T \\ simpler defining polynomial for Q[x]/(R)
%6 = x^20 + 25*x^10 + 5

419
 ? X \\ root of R in Q[y]/(T (y))
%7 = Mod(-1/11*x^15 - 1/11*x^14 + 1/22*x^10 - 47/22*x^5 - 29/11*x^4 + 7/22,\
x^20 + 25*x^10 + 5)
? a = subst(a.pol, ’x, X) \\ a in the new coordinates
%8 = Mod(1/11*x^14 + 29/11*x^4, x^20 + 25*x^10 + 5)
? a^5 - 5
%9 = 0
In the above example, x5 − 5 and the 5-th cyclotomic polynomial are irreducible over Q; they have
coprime degrees so define linearly disjoint extensions and we could have started by
? [R,a] = polcompositum(x^5 - 5, polcyclo(5), 3); \\ [R, a, b, k]
The library syntax is GEN polcompositum0(GEN P, GEN Q, long flag). Also available are
GEN compositum(GEN P, GEN Q) (flag = 0) and GEN compositum2(GEN P, GEN Q) (flag = 1).

3.13.149 polgalois(T ). Galois group of the nonconstant polynomial T ∈ Q[X]. In the present
version 2.13.3, T must be irreducible and the degree d of T must be less than or equal to 7. If the
galdata package has been installed, degrees 8, 9, 10 and 11 are also implemented. By definition,
if K = Q[x]/(T ), this computes the action of the Galois group of the Galois closure of K on the d
distinct roots of T , up to conjugacy (corresponding to different root orderings).
The output is a 4-component vector [n, s, k, name] with the following meaning: n is the cardi-
nality of the group, s is its signature (s = 1 if the group is a subgroup of the alternating group Ad ,
s = −1 otherwise) and name is a character string containing name of the transitive group according
to the GAP 4 transitive groups library by Alexander Hulpke.
k is more arbitrary and the choice made up to version 2.2.3 of PARI is rather unfortunate:
for d > 7, k is the numbering of the group among all transitive subgroups of Sd , as given in “The
transitive groups of degree up to eleven”, G. Butler and J. McKay, Communications in Algebra,
vol. 11, 1983, pp. 863–911 (group k is denoted Tk there). And for d ≤ 7, it was ad hoc, so as
to ensure that a given triple would denote a unique group. Specifically, for polynomials of degree
d ≤ 7, the groups are coded as follows, using standard notations

In degree 1: S1 = [1, 1, 1].

In degree 2: S2 = [2, −1, 1].

In degree 3: A3 = C3 = [3, 1, 1], S3 = [6, −1, 1].

In degree 4: C4 = [4, −1, 1], V4 = [4, 1, 1], D4 = [8, −1, 1], A4 = [12, 1, 1], S4 = [24, −1, 1].

In degree 5: C5 = [5, 1, 1], D5 = [10, 1, 1], M20 = [20, −1, 1], A5 = [60, 1, 1], S5 = [120, −1, 1].

In degree 6: C6 = [6, −1, 1], S3 = [6, −1, 2], D6 = [12, −1, 1], A4 = [12, 1, 1], G18 = [18, −1, 1],
S4− = [24, −1, 1], A4 × C2 = [24, −1, 2], S4+ = [24, 1, 1], G− +
36 = [36, −1, 1], G36 = [36, 1, 1], S4 ×
C2 = [48, −1, 1], A5 = P SL2 (5) = [60, 1, 1], G72 = [72, −1, 1], S5 = P GL2 (5) = [120, −1, 1],
A6 = [360, 1, 1], S6 = [720, −1, 1].

In degree 7: C7 = [7, 1, 1], D7 = [14, −1, 1], M21 = [21, 1, 1], M42 = [42, −1, 1], P SL2 (7) =
P SL3 (2) = [168, 1, 1], A7 = [2520, 1, 1], S7 = [5040, −1, 1].

This is deprecated and obsolete, but for reasons of backward compatibility, we cannot change
this behavior yet. So you can use the default new_galois_format to switch to a consistent naming

420
scheme, namely k is always the standard numbering of the group among all transitive subgroups
of Sn . If this default is in effect, the above groups will be coded as:

In degree 1: S1 = [1, 1, 1].

In degree 2: S2 = [2, −1, 1].

In degree 3: A3 = C3 = [3, 1, 1], S3 = [6, −1, 2].

In degree 4: C4 = [4, −1, 1], V4 = [4, 1, 2], D4 = [8, −1, 3], A4 = [12, 1, 4], S4 = [24, −1, 5].

In degree 5: C5 = [5, 1, 1], D5 = [10, 1, 2], M20 = [20, −1, 3], A5 = [60, 1, 4], S5 = [120, −1, 5].

In degree 6: C6 = [6, −1, 1], S3 = [6, −1, 2], D6 = [12, −1, 3], A4 = [12, 1, 4], G18 = [18, −1, 5],
A4 × C2 = [24, −1, 6], S4+ = [24, 1, 7], S4− = [24, −1, 8], G− +
36 = [36, −1, 9], G36 = [36, 1, 10], S4 ×
C2 = [48, −1, 11], A5 = P SL2 (5) = [60, 1, 12], G72 = [72, −1, 13], S5 = P GL2 (5) = [120, −1, 14],
A6 = [360, 1, 15], S6 = [720, −1, 16].

In degree 7: C7 = [7, 1, 1], D7 = [14, −1, 2], M21 = [21, 1, 3], M42 = [42, −1, 4], P SL2 (7) =
P SL3 (2) = [168, 1, 5], A7 = [2520, 1, 6], S7 = [5040, −1, 7].

Warning. The method used is that of resolvent polynomials and is sensitive to the current preci-
sion. The precision is updated internally but, in very rare cases, a wrong result may be returned if
the initial precision was not sufficient.
The library syntax is GEN polgalois(GEN T, long prec). To enable the new format in
library mode, set the global variable new_galois_format to 1.

3.13.150 polred(T, {flag = 0}). This function is deprecated , use polredbest instead. Finds
polynomials with reasonably small coefficients defining subfields of the number field defined by T .
One of the polynomials always defines Q (hence has degree 1), and another always defines the same
number field as T if T is irreducible.
All T accepted by nfinit are also allowed here; in particular, the format [T, listP] is
recommended, e.g. with listP = 105 or a vector containing all ramified primes. Otherwise, the
maximal order of Q[x]/(T ) must be computed.
The following binary digits of flag are significant:
1: Possibly use a suborder of the maximal order. The primes dividing the index of the order
chosen are larger than primelimit or divide integers stored in the addprimes table. This flag is
deprecated , the [T, listP] format is more flexible.
2: gives also elements. The result is a two-column matrix, the first column giving primitive
elements defining these subfields, the second giving the corresponding minimal polynomials.
? M = polred(x^4 + 8, 2)
%1 =
[ 1 x - 1]
[ 1/2*x^2 + 1 x^2 - 2*x + 3]
[-1/2*x^2 + 1 x^2 - 2*x + 3]
[ 1/2*x^2 x^2 + 2]
[ 1/4*x^3 x^4 + 2]

421
 ? minpoly(Mod(M[2,1], x^4+8))
%2 = x^2 + 2
The library syntax is polred(GEN T) (flag = 0). Also available is GEN polred2(GEN T) (flag =
2). The function polred0 is deprecated, provided for backward compatibility.

3.13.151 polredabs(T, {flag = 0}). Returns a canonical defining polynomial P for the number
field Q[X]/(T ) defined by T , such that the sum of the squares of the modulus of the roots (i.e. the
T2 -norm) is minimal. Different T defining isomorphic number fields will yield the same P . All
T accepted by nfinit are also allowed here, e.g. nonmonic polynomials, or pairs [T, listP]
specifying that a nonmaximal order may be used. For convenience, any number field structure (nf ,
bnf ,. . . ) can also be used instead of T .
? polredabs(x^2 + 16)
%1 = x^2 + 1
? K = bnfinit(x^2 + 16); polredabs(K)
%2 = x^2 + 1

Warning 1. Using a t_POL T requires computing and fully factoring the discriminant dK of the
maximal order which may be very hard. You can use the format [T, listP], where listP encodes
a list of known coprime divisors of disc(T ) (see ??nfbasis), to help the routine, thereby replacing
this part of the algorithm by a polynomial time computation But this may only compute a suborder
of the maximal order, when the divisors are not squarefree or do not include all primes dividing
dK . The routine attempts to certify the result independently of this order computation as per
nfcertify: we try to prove that the computed order is maximal. If the certification fails, the
routine then fully factors the integers returned by nfcertify. You can also use polredbest to
avoid this factorization step; in this case, the result is small but no longer canonical.

Warning 2. Apart from the factorization of the discriminant of T , this routine runs in polynomial
time for a fixed degree. But the complexity is exponential in the degree: this routine may be
exceedingly slow when the number field has many subfields, hence a lot of elements of small T2 -
norm. If you do not need a canonical polynomial, the function polredbest is in general much
faster (it runs in polynomial time), and tends to return polynomials with smaller discriminants.
The binary digits of flag mean
1: outputs a two-component row vector [P, a], where P is the default output and Mod(a, P)
is a root of the original T .
4: gives all polynomials of minimal T2 norm; of the two polynomials P (x) and ±P (−x), only
one is given.
16: (OBSOLETE) Possibly use a suborder of the maximal order, without attempting to certify
the result as in Warning 1. This makes polredabs behave like polredbest. Just use the latter.
? T = x^16 - 136*x^14 + 6476*x^12 - 141912*x^10 + 1513334*x^8 \
- 7453176*x^6 + 13950764*x^4 - 5596840*x^2 + 46225
? T1 = polredabs(T); T2 = polredbest(T);
? [ norml2(polroots(T1)), norml2(polroots(T2)) ]
%3 = [88.0000000, 120.000000]
? [ sizedigit(poldisc(T1)), sizedigit(poldisc(T2)) ]
%4 = [75, 67]
The precise definition of the output of polredabs is as follows.

422
 • Consider the finite list of characteristic polynomials of primitive elements of K that are
in ZK and minimal for the T2 norm; now remove from the list the polynomials whose discriminant
do not have minimal absolute value. Note that this condition is restricted to the original list of
polynomials with minimal T2 norm and does not imply that the defining polynomial for the field
with smallest discriminant belongs to the list !
• To a polynomial P (x) = xn + . . . + an ∈ R[x] we attach the sequence S(P ) given by
|a1 |, a1 , . . . , |an |, an . Order the polynomials P by the lexicographic order on the coefficient vectors
S(P ). Then the output of polredabs is the smallest polynomial in the above list for that order. In
other words, the monic polynomial which is lexicographically smallest with respect to the absolute
values of coefficients, favouring negative coefficients to break ties, i.e. choosing x3 − 2 rather than
x3 + 2.
The library syntax is GEN polredabs0(GEN T, long flag). Instead of the above hardcoded
numerical flags, one should use an or-ed combination of
• nf_PARTIALFACT (OBSOLETE): possibly use a suborder of the maximal order, without
attempting to certify the result.
• nf_ORIG: return [P, a], where Mod(a, P) is a root of T .
• nf_RAW: return [P, b], where Mod(b, T) is a root of P . The algebraic integer b is the raw
result produced by the small vectors enumeration in the maximal order; P was computed as the
characteristic polynomial of Mod(b, T). Mod(a, P) as in nf_ORIG is obtained with modreverse.
• nf_ADDZK: if r is the result produced with some of the above flags (of the form P or [P, c]),
return [r,zk], where zk is a Z-basis for the maximal order of Q[X]/(P ).
• nf_ALL: return a vector of results of the above form, for all polynomials of minimal T2 -norm.

3.13.152 polredbest(T, {flag = 0}). Finds a polynomial with reasonably small coefficients defin-
ing the same number field as T . All T accepted by nfinit are also allowed here (e.g. nonmonic
polynomials, nf, bnf, [T,Z K basis]). Contrary to polredabs, this routine runs in polynomial
time, but it offers no guarantee as to the minimality of its result.
This routine computes an LLL-reduced basis for an order in Q[X]/(T ), then examines small lin-
ear combinations of the basis vectors, computing their characteristic polynomials. It returns the sep-
arable polynomial P of smallest discriminant, the one with lexicographically smallest abs(Vec(P))
in case of ties. This is a good candidate for subsequent number field computations since it guaran-
tees that the denominators of algebraic integers, when expressed in the power basis, are reasonably
small. With no claim of minimality, though.
It can happen that iterating this functions yields better and better polynomials, until it sta-
bilizes:
? \p5
? P = X^12+8*X^8-50*X^6+16*X^4-3069*X^2+625;
? poldisc(P)*1.
%2 = 1.2622 E55
? P = polredbest(P);
? poldisc(P)*1.
%4 = 2.9012 E51
? P = polredbest(P);
? poldisc(P)*1.

423
 %6 = 8.8704 E44
In this example, the initial polynomial P is the one returned by polredabs, and the last one is
stable.
If flag = 1: outputs a two-component row vector [P, a], where P is the default output and
Mod(a, P) is a root of the original T .

? [P,a] = polredbest(x^4 + 8, 1)
%1 = [x^4 + 2, Mod(x^3, x^4 + 2)]
? charpoly(a)
%2 = x^4 + 8
In particular, the map Q[x]/(T ) → Q[x]/(P ), x 7→ Mod(a, P) defines an isomorphism of number
fields, which can be computed as

subst(lift(Q), ’x, a)
if Q is a t_POLMOD modulo T ; b = modreverse(a) returns a t_POLMOD giving the inverse of the
above map (which should be useless since Q[x]/(P ) is a priori a better representation for the number
field and its elements).
The library syntax is GEN polredbest(GEN T, long flag).

3.13.153 polredord(x). This function is obsolete, use polredbest.

The library syntax is GEN polredord(GEN x).

3.13.154 poltschirnhaus(x). Applies a random Tschirnhausen transformation to the polynomial

x, which is assumed to be nonconstant and separable, so as to obtain a new equation for the étale
algebra defined by x. This is for instance useful when computing resolvents, hence is used by the
polgalois function.
The library syntax is GEN tschirnhaus(GEN x).

3.13.155 rnfalgtobasis(rnf , x). Expresses x on the relative integral basis. Here, rnf is a relative
number field extension L/K as output by rnfinit, and x an element of L in absolute form, i.e.
expressed as a polynomial or polmod with polmod coefficients, not on the relative integral basis.
The library syntax is GEN rnfalgtobasis(GEN rnf, GEN x).

3.13.156 rnfbasis(bnf , M ). Let K the field represented by bnf , as output by bnfinit. M is a

projective ZK -module of rank n (M ⊗ K is an n-dimensional K-vector space), given by a pseudo-
basis of size n. The routine returns either a true ZK -basis of M (of size n) if it exists, or an
n + 1-element generating set of M if not.
It is allowed to use a monic irreducible polynomial P in K[X] instead of M , in which case, M
is defined as the ring of integers of K[X]/(P ), viewed as a ZK -module.

424
Huge discriminants, helping rnfdisc. The format [T, B] is also accepted instead of T and
computes an order which is maximal at all maximal ideals specified by B, see ??rnfinit: the
valuation of D is then correct at all such maximal ideals but may be incorrect at other primes.

The library syntax is GEN rnfbasis(GEN bnf, GEN M).

3.13.157 rnfbasistoalg(rnf , x). Computes the representation of x as a polmod with polmods

coefficients. Here, rnf is a relative number field extension L/K as output by rnfinit, and x an
element of L expressed on the relative integral basis.

The library syntax is GEN rnfbasistoalg(GEN rnf, GEN x).

3.13.158 rnfcharpoly(nf , T, a, {var =0 x}). Characteristic polynomial of a over nf , where a

belongs to the algebra defined by T over nf , i.e. nf [X]/(T ). Returns a polynomial in variable v (x
by default).

? nf = nfinit(y^2+1);
? rnfcharpoly(nf, x^2+y*x+1, x+y)
%2 = x^2 + Mod(-y, y^2 + 1)*x + 1

The library syntax is GEN rnfcharpoly(GEN nf, GEN T, GEN a, long var = -1) where
var is a variable number.

3.13.159 rnfconductor(bnf , T, {flag = 0}). Given a bnf structure attached to a number field K,
as produced by bnfinit, and T an irreducible polynomial in K[x] defining an Abelian extension
L = K[x]/(T ), computes the class field theory conductor of this Abelian extension. If T does not
define an Abelian extension over K, the result is undefined; it may be the integer 0 (in which case
the extension is definitely not Abelian) or a wrong result.

The result is a 3-component vector [f, bnr , H], where f is the conductor of the extension given
as a 2-component row vector [f0 , f∞ ], bnr is the attached bnr structure and H is a matrix in
HNF defining the subgroup of the ray class group on the ray class group generators bnr.gen; in
particular, it is a left divisor of the diagonal matrix attached to bnr.cyc and | det H| = N = deg T .

If flagis set, return [f, bnrmod , H], where bnrmod is now attached to Clf /ClN f , and H is as
before since it contains the N -th powers. This is useful when f contains a maximal ideal with
huge residue field, since the corresponding tough discrete logarithms are trivialized: in the quotient
group, all elements have small order dividing N . This allows to work in Clf /H but no longer in
Clf .

Huge discriminants, helping rnfdisc. The format [T, B] is also accepted instead of T and
computes the conductor of the extension provided it factors completely over the maximal ideals
specified by B, see ??rnfinit: the valuation of f0 is then correct at all such maximal ideals but
may be incorrect at other primes.

The library syntax is GEN rnfconductor0(GEN bnf, GEN T, long flag). Also available is
GEN rnfconductor(GEN bnf, GEN T) when flag = 0.

425
3.13.160 rnfdedekind(nf , pol , {pr }, {flag = 0}). Given a number field K coded by nf and a
monic polynomial P ∈ ZK [X], irreducible over K and thus defining a relative extension L of K,
applies Dedekind’s criterion to the order ZK [X]/(P ), at the prime ideal pr . It is possible to set pr
to a vector of prime ideals (test maximality at all primes in the vector), or to omit altogether, in
which case maximality at all primes is tested; in this situation flag is automatically set to 1.
The default historic behavior (flag is 0 or omitted and pr is a single prime ideal) is not so useful
since rnfpseudobasis gives more information and is generally not that much slower. It returns a
3-component vector [max , basis, v]:
• basis is a pseudo-basis of an enlarged order O produced by Dedekind’s criterion, containing
the original order ZK [X]/(P ) with index a power of pr . Possibly equal to the original order.
• max is a flag equal to 1 if the enlarged order O could be proven to be pr -maximal and to 0
otherwise; it may still be maximal in the latter case if pr is ramified in L,
• v is the valuation at pr of the order discriminant.
If flag is nonzero, on the other hand, we just return 1 if the order ZK [X]/(P ) is pr -maximal
(resp. maximal at all relevant primes, as described above), and 0 if not. This is much faster than
the default, since the enlarged order is not computed.
? nf = nfinit(y^2-3); P = x^3 - 2*y;
? pr3 = idealprimedec(nf,3)[1];
? rnfdedekind(nf, P, pr3)
%3 = [1, [[1, 0, 0; 0, 1, 0; 0, 0, 1], [1, 1, 1]], 8]
? rnfdedekind(nf, P, pr3, 1)
%4 = 1
In this example, pr3 is the ramified ideal above 3, and the order generated by the cube roots of y
is already pr3-maximal. The order-discriminant has valuation 8. On the other hand, the order is
not maximal at the prime above 2:
? pr2 = idealprimedec(nf,2)[1];
? rnfdedekind(nf, P, pr2, 1)
%6 = 0
? rnfdedekind(nf, P, pr2)
%7 = [0, [[2, 0, 0; 0, 1, 0; 0, 0, 1], [[1, 0; 0, 1], [1, 0; 0, 1],
[1, 1/2; 0, 1/2]]], 2]
The enlarged order is not proven to be pr2-maximal yet. In fact, it is; it is in fact the maximal
order:
? B = rnfpseudobasis(nf, P)
%8 = [[1, 0, 0; 0, 1, 0; 0, 0, 1], [1, 1, [1, 1/2; 0, 1/2]],
[162, 0; 0, 162], -1]
? idealval(nf,B[3], pr2)
%9 = 2
It is possible to use this routine with nonmonic P = i≤n pi X i ∈ ZK [X] if flag = 1; in this case,
P
we test maximality of Dedekind’s order generated by
1, pn α, pn α2 + pn−1 α, . . . , pn αn−1 + pn−1 αn−2 + · · · + p1 α.
The routine will fail if P vanishes on the projective line over the residue field ZK /pr (FIXME).
The library syntax is GEN rnfdedekind(GEN nf, GEN pol, GEN pr = NULL, long flag)
.

426
3.13.161 rnfdet(nf , M ). Given a pseudo-matrix M over the maximal order of nf , computes its
determinant.

The library syntax is GEN rnfdet(GEN nf, GEN M).

3.13.162 rnfdisc(nf , T ). Given an nf structure attached to a number field K, as output by nfinit,

and a monic irreducible polynomial T ∈ K[x] defining a relative extension L = K[x]/(T ), compute
the relative discriminant of L. This is a vector [D, d], where D is the relative ideal discriminant
and d is the relative discriminant considered as an element of K ∗ /K ∗ 2 . The main variable of nf
must be of lower priority than that of T , see Section 2.5.3.

Huge discriminants, helping rnfdisc. The format [T, B] is also accepted instead of T and
computes an order which is maximal at all maximal ideals specified by B, see ??rnfinit: the
valuation of D is then correct at all such maximal ideals but may be incorrect at other primes.

The library syntax is GEN rnfdiscf(GEN nf, GEN T).

3.13.163 rnfeltabstorel(rnf , x). Let rnf be a relative number field extension L/K as output by
rnfinit and let x be an element of L expressed as a polynomial modulo the absolute equation
rnf .pol, or in terms of the absolute Z-basis for ZL if rnf contains one (as in rnfinit(nf,pol,1),
or after a call to nfinit(rnf)). Computes x as an element of the relative extension L/K as a
polmod with polmod coefficients.

? K = nfinit(y^2+1); L = rnfinit(K, x^2-y);

? L.polabs
%2 = x^4 + 1
? rnfeltabstorel(L, Mod(x, L.polabs))
%3 = Mod(x, x^2 + Mod(-y, y^2 + 1))
? rnfeltabstorel(L, 1/3)
%4 = 1/3
? rnfeltabstorel(L, Mod(x, x^2-y))
%5 = Mod(x, x^2 + Mod(-y, y^2 + 1))
? rnfeltabstorel(L, [0,0,0,1]~) \\ Z_L not initialized yet
*** at top-level: rnfeltabstorel(L,[0,
*** ^--------------------
*** rnfeltabstorel: incorrect type in rnfeltabstorel, apply nfinit(rnf).
? nfinit(L); \\ initialize now
? rnfeltabstorel(L, [0,0,0,1]~)
%6 = Mod(Mod(y, y^2 + 1)*x, x^2 + Mod(-y, y^2 + 1))

The library syntax is GEN rnfeltabstorel(GEN rnf, GEN x).

427
3.13.164 rnfeltdown(rnf , x, {flag = 0}). rnf being a relative number field extension L/K as
output by rnfinit and x being an element of L expressed as a polynomial or polmod with polmod
coefficients (or as a t_COL on nfinit(rnf).zk), computes x as an element of K as a t_POLMOD if
flag = 0 and as a t_COL otherwise. If x is not in K, a domain error occurs.

? K = nfinit(y^2+1); L = rnfinit(K, x^2-y);

? L.pol
%2 = x^4 + 1
? rnfeltdown(L, Mod(x^2, L.pol))
%3 = Mod(y, y^2 + 1)
? rnfeltdown(L, Mod(x^2, L.pol), 1)
%4 = [0, 1]~
? rnfeltdown(L, Mod(y, x^2-y))
%5 = Mod(y, y^2 + 1)
? rnfeltdown(L, Mod(y,K.pol))
%6 = Mod(y, y^2 + 1)
? rnfeltdown(L, Mod(x, L.pol))
*** at top-level: rnfeltdown(L,Mod(x,x
*** ^--------------------
*** rnfeltdown: domain error in rnfeltdown: element not in the base field
? rnfeltdown(L, Mod(y, x^2-y), 1) \\ as a t_COL
%7 = [0, 1]~
? rnfeltdown(L, [0,1,0,0]~) \\ not allowed without absolute nf struct
*** rnfeltdown: incorrect type in rnfeltdown (t_COL).
? nfinit(L); \\ add absolute nf structure to L
? rnfeltdown(L, [0,1,0,0]~) \\ now OK
%8 = Mod(y, y^2 + 1)

If we had started with L = rnfinit(K, x^2-y, 1), then the final would have worked directly.

The library syntax is GEN rnfeltdown0(GEN rnf, GEN x, long flag). Also available is GEN
rnfeltdown(GEN rnf, GEN x) (flag = 0).

3.13.165 rnfeltnorm(rnf , x). rnf being a relative number field extension L/K as output by
rnfinit and x being an element of L, returns the relative norm NL/K (x) as an element of K.

? K = nfinit(y^2+1); L = rnfinit(K, x^2-y);

? rnfeltnorm(L, Mod(x, L.pol))
%2 = Mod(x, x^2 + Mod(-y, y^2 + 1))
? rnfeltnorm(L, 2)
%3 = 4
? rnfeltnorm(L, Mod(x, x^2-y))

The library syntax is GEN rnfeltnorm(GEN rnf, GEN x).

428
3.13.166 rnfeltreltoabs(rnf , x). rnf being a relative number field extension L/K as output by
rnfinit and x being an element of L expressed as a polynomial or polmod with polmod coefficients,
computes x as an element of the absolute extension L/Q as a polynomial modulo the absolute
equation rnf .pol.
? K = nfinit(y^2+1); L = rnfinit(K, x^2-y);
? L.pol
%2 = x^4 + 1
? rnfeltreltoabs(L, Mod(x, L.pol))
%3 = Mod(x, x^4 + 1)
? rnfeltreltoabs(L, Mod(y, x^2-y))
%4 = Mod(x^2, x^4 + 1)
? rnfeltreltoabs(L, Mod(y,K.pol))
%5 = Mod(x^2, x^4 + 1)
The library syntax is GEN rnfeltreltoabs(GEN rnf, GEN x).

3.13.167 rnfelttrace(rnf , x). rnf being a relative number field extension L/K as output by
rnfinit and x being an element of L, returns the relative trace T rL/K (x) as an element of K.
? K = nfinit(y^2+1); L = rnfinit(K, x^2-y);
? rnfelttrace(L, Mod(x, L.pol))
%2 = 0
? rnfelttrace(L, 2)
%3 = 4
? rnfelttrace(L, Mod(x, x^2-y))
The library syntax is GEN rnfelttrace(GEN rnf, GEN x).

3.13.168 rnfeltup(rnf , x, {flag = 0}). rnf being a relative number field extension L/K as output
by rnfinit and x being an element of K, computes x as an element of the absolute extension L/Q.
As a t_POLMOD modulo rnf .pol if flag = 0 and as a t_COL on the absolute field integer basis if
flag = 1.
? K = nfinit(y^2+1); L = rnfinit(K, x^2-y);
? L.pol
%2 = x^4 + 1
? rnfeltup(L, Mod(y, K.pol))
%3 = Mod(x^2, x^4 + 1)
? rnfeltup(L, y)
%4 = Mod(x^2, x^4 + 1)
? rnfeltup(L, [1,2]~) \\ in terms of K.zk
%5 = Mod(2*x^2 + 1, x^4 + 1)
? rnfeltup(L, y, 1) \\ in terms of nfinit(L).zk
%6 = [0, 1, 0, 0]~
? rnfeltup(L, [1,2]~, 1)
%7 = [1, 2, 0, 0]~
The library syntax is GEN rnfeltup0(GEN rnf, GEN x, long flag).

429
3.13.169 rnfequation(nf , pol , {flag = 0}). Given a number field nf as output by nfinit (or
simply a polynomial) and a polynomial pol with coefficients in nf defining a relative extension L
of nf , computes an absolute equation of L over Q.

The main variable of nf must be of lower priority than that of pol (see Section 2.5.3). Note
that for efficiency, this does not check whether the relative equation is irreducible over nf , but only
if it is squarefree. If it is reducible but squarefree, the result will be the absolute equation of the
étale algebra defined by pol . If pol is not squarefree, raise an e DOMAIN exception.

? rnfequation(y^2+1, x^2 - y)
%1 = x^4 + 1
? T = y^3-2; rnfequation(nfinit(T), (x^3-2)/(x-Mod(y,T)))
%2 = x^6 + 108 \\ Galois closure of Q(2^(1/3))

If flag is nonzero, outputs a 3-component row vector [z, a, k], where

• z is the absolute equation of L over Q, as in the default behavior,

• a expresses as a t_POLMOD modulo z a root α of the polynomial defining the base field nf ,

• k is a small integer such that θ = β + kα is a root of z, where β is a root of pol . It is

guaranteed that k = 0 whenever Q(β) = L.

? T = y^3-2; pol = x^2 +x*y + y^2;

? [z,a,k] = rnfequation(T, pol, 1);
? z
%3 = x^6 + 108
? subst(T, y, a)
%4 = 0
? alpha= Mod(y, T);
? beta = Mod(x*Mod(1,T), pol);
? subst(z, x, beta + k*alpha)
%7 = 0

The library syntax is GEN rnfequation0(GEN nf, GEN pol, long flag). Also available
are GEN rnfequation(GEN nf, GEN pol) (flag = 0) and GEN rnfequation2(GEN nf, GEN pol)
(flag = 1).

3.13.170 rnfhnfbasis(bnf , x). Given bnf as output by bnfinit, and either a polynomial x with
coefficients in bnf defining a relative extension L of bnf , or a pseudo-basis x of such an extension,
gives either a true bnf -basis of L in upper triangular Hermite normal form, if it exists, and returns
0 otherwise.

The library syntax is GEN rnfhnfbasis(GEN bnf, GEN x).

430
3.13.171 rnfidealabstorel(rnf , x). Let rnf be a relative number field extension L/K as output by
rnfinit and let x be an ideal of the absolute extension L/Q. Returns the relative pseudo-matrix
in HNF giving the ideal x considered as an ideal of the relative extension L/K, i.e. as a ZK -module.
Let Labs be an (absolute) nf structure attached to L, obtained via Labs = nfinit(rnf)).
Then rnf “knows” about Labs and x may be given in any format attached to Labs, e.g. a prime
ideal or an ideal in HNF wrt. Labs.zk:
? K = nfinit(y^2+1); rnf = rnfinit(K, x^2-y); Labs = nfinit(rnf);
? m = idealhnf(Labs, 17, x^3+2); \\ some ideal in HNF wrt. Labs.zk
? B = rnfidealabstorel(rnf, m)
%3 = [[1, 8; 0, 1], [[17, 4; 0, 1], 1]] \\ pseudo-basis for m as Z_K-module
? A = rnfidealreltoabs(rnf, B)
%4 = [17, x^2 + 4, x + 8, x^3 + 8*x^2] \\ Z-basis for m in Q[x]/(rnf.polabs)
? mathnf(matalgtobasis(Labs, A)) == m
%5 = 1
If on the other hand, we do not have a Labs at hand, because it would be too expensive to compute,
but we nevertheless have a Z-basis for x, then we can use the function with this basis as argument.
The entries of x may be given either modulo rnf.polabs (absolute form, possibly lifted) or modulo
rnf.pol (relative form as t_POLMODs):
? K = nfinit(y^2+1); rnf = rnfinit(K, x^2-y);
? rnfidealabstorel(rnf, [17, x^2 + 4, x + 8, x^3 + 8*x^2])
%2 = [[1, 8; 0, 1], [[17, 4; 0, 1], 1]]
? rnfidealabstorel(rnf, Mod([17, y + 4, x + 8, y*x + 8*y], x^2-y))
%3 = [[1, 8; 0, 1], [[17, 4; 0, 1], 1]]
The library syntax is GEN rnfidealabstorel(GEN rnf, GEN x).

3.13.172 rnfidealdown(rnf , x). Let rnf be a relative number field extension L/K as output by
rnfinit, and x an ideal of L, given either in relative form or by a Z-basis of elements of L (see
Section 3.13.171). This function returns the ideal of K below x, i.e. the intersection of x with K.
The library syntax is GEN rnfidealdown(GEN rnf, GEN x).

3.13.173 rnfidealfactor(rnf , x). Factor into prime ideal powers the ideal x in the attached
absolute number field L = nfinit(rnf ). The output format is similar to the factor function, and
the prime ideals are represented in the form output by the idealprimedec function for L.
? rnf = rnfinit(nfinit(y^2+1), x^2-y+1);
? rnfidealfactor(rnf, y+1) \\ P_2^2
%2 =
[[2, [0,0,1,0]~, 4, 1, [0,0,0,2;0,0,-2,0;-1,-1,0,0;1,-1,0,0]] 2]
? rnfidealfactor(rnf, x) \\ P_2
%3 =
[[2, [0,0,1,0]~, 4, 1, [0,0,0,2;0,0,-2,0;-1,-1,0,0;1,-1,0,0]] 1]
? L = nfinit(rnf);
? id = idealhnf(L, idealhnf(L, 25, (x+1)^2));
? idealfactor(L, id) == rnfidealfactor(rnf, id)
%6 = 1

431
Note that ideals of the base field K must be explicitly lifted to L via rnfidealup before they can
be factored.
The library syntax is GEN rnfidealfactor(GEN rnf, GEN x).

3.13.174 rnfidealhnf(rnf , x). rnf being a relative number field extension L/K as output by
rnfinit and x being a relative ideal (which can be, as in the absolute case, of many different
types, including of course elements), computes the HNF pseudo-matrix attached to x, viewed as a
ZK -module.
The library syntax is GEN rnfidealhnf(GEN rnf, GEN x).

3.13.175 rnfidealmul(rnf , x, y). rnf being a relative number field extension L/K as output by
rnfinit and x and y being ideals of the relative extension L/K given by pseudo-matrices, outputs
the ideal product, again as a relative ideal.
The library syntax is GEN rnfidealmul(GEN rnf, GEN x, GEN y).

3.13.176 rnfidealnormabs(rnf , x). Let rnf be a relative number field extension L/K as output
by rnfinit and let x be a relative ideal (which can be, as in the absolute case, of many different
types, including of course elements). This function computes the norm of the x considered as an
ideal of the absolute extension L/Q. This is identical to
idealnorm(rnf, rnfidealnormrel(rnf,x))
but faster.
The library syntax is GEN rnfidealnormabs(GEN rnf, GEN x).

3.13.177 rnfidealnormrel(rnf , x). Let rnf be a relative number field extension L/K as output
by rnfinit and let x be a relative ideal (which can be, as in the absolute case, of many different
types, including of course elements). This function computes the relative norm of x as an ideal of
K in HNF.
The library syntax is GEN rnfidealnormrel(GEN rnf, GEN x).

3.13.178 rnfidealprimedec(rnf , pr ). Let rnf be a relative number field extension L/K as output
by rnfinit, and pr a maximal ideal of K (prid ), this function completes the rnf with a nf structure
attached to L (see Section 3.13.182) and returns the prime ideal decomposition of pr in L/K.
? K = nfinit(y^2+1); rnf = rnfinit(K, x^3+y+1);
? P = idealprimedec(K, 2)[1];
? S = rnfidealprimedec(rnf, P);
? #S
%4 = 1
The argument pr is also allowed to be a prime number p, in which case the function returns
a pair of vectors [SK,SL], where SK contains the primes of K above p and SL[i] is the vector of
primes of L above SK[i].
? [SK,SL] = rnfidealprimedec(rnf, 5);
? [#SK, vector(#SL,i,#SL[i])]
%6 = [2, [2, 2]]
The library syntax is GEN rnfidealprimedec(GEN rnf, GEN pr).

432
3.13.179 rnfidealreltoabs(rnf , x, {flag = 0}). Let rnf be a relative number field extension L/K
as output by rnfinit and let x be a relative ideal, given as a ZK -module by a pseudo matrix [A, I].
This function returns the ideal x as an absolute ideal of L/Q. If flag = 0, the result is given by a
vector of t_POLMODs modulo rnf.pol forming a Z-basis; if flag = 1, it is given in HNF in terms of
the fixed Z-basis for ZL , see Section 3.13.182.
? K = nfinit(y^2+1); rnf = rnfinit(K, x^2-y);
? P = idealprimedec(K,2)[1];
? P = rnfidealup(rnf, P)
%3 = [2, x^2 + 1, 2*x, x^3 + x]
? Prel = rnfidealhnf(rnf, P)
%4 = [[1, 0; 0, 1], [[2, 1; 0, 1], [2, 1; 0, 1]]]
? rnfidealreltoabs(rnf,Prel)
%5 = [2, x^2 + 1, 2*x, x^3 + x]
? rnfidealreltoabs(rnf,Prel,1)
%6 =
[2 1 0 0]
[0 1 0 0]
[0 0 2 1]
[0 0 0 1]
The reason why we do not return by default (flag = 0) the customary HNF in terms of a fixed
Z-basis for ZL is precisely because a rnf does not contain such a basis by default. Completing the
structure so that it contains a nf structure for L is polynomial time but costly when the absolute
degree is large, thus it is not done by default. Note that setting flag = 1 will complete the rnf .
The library syntax is GEN rnfidealreltoabs0(GEN rnf, GEN x, long flag). Also available
is GEN rnfidealreltoabs(GEN rnf, GEN x) (flag = 0).

3.13.180 rnfidealtwoelt(rnf , x). rnf being a relative number field extension L/K as output by
rnfinit and x being an ideal of the relative extension L/K given by a pseudo-matrix, gives a
vector of two generators of x over ZL expressed as polmods with polmod coefficients.
The library syntax is GEN rnfidealtwoelement(GEN rnf, GEN x).

3.13.181 rnfidealup(rnf , x, {flag = 0}). Let rnf be a relative number field extension L/K as
output by rnfinit and let x be an ideal of K. This function returns the ideal xZL as an absolute
ideal of L/Q, in the form of a Z-basis. If flag = 0, the result is given by a vector of polynomials
(modulo rnf.pol); if flag = 1, it is given in HNF in terms of the fixed Z-basis for ZL , see
Section 3.13.182.
? K = nfinit(y^2+1); rnf = rnfinit(K, x^2-y);
? P = idealprimedec(K,2)[1];
? rnfidealup(rnf, P)
%3 = [2, x^2 + 1, 2*x, x^3 + x]
? rnfidealup(rnf, P,1)
%4 =
[2 1 0 0]
[0 1 0 0]
[0 0 2 1]

433
 [0 0 0 1]
The reason why we do not return by default (flag = 0) the customary HNF in terms of a fixed
Z-basis for ZL is precisely because a rnf does not contain such a basis by default. Completing the
structure so that it contains a nf structure for L is polynomial time but costly when the absolute
degree is large, thus it is not done by default. Note that setting flag = 1 will complete the rnf .
The library syntax is GEN rnfidealup0(GEN rnf, GEN x, long flag). Also available is GEN
rnfidealup(GEN rnf, GEN x) (flag = 0).

3.13.182 rnfinit(nf , T, {flag = 0}). Given an nf structure attached to a number field K, as

output by nfinit, and a monic irreducible polynomial T in ZK [x] defining a relative extension
L = K[x]/(T ), this computes data to work in L/K The main variable of T must be of higher
priority (see Section 2.5.3) than that of nf , and the coefficients of T must be in K.
The result is a row vector, whose components are technical. We let m = [K : Q] the degree of
the base field, n = [L : K] the relative degree, r1 and r2 the number of real and complex places of
K. Access to this information via member functions is preferred since the specific data organization
specified below will change in the future.
If flag = 1, add an nf structure attached to L to rnf . This is likely to be very expensive
if the absolute degree mn is large, but fixes an integer basis for ZL as a Z-module and allows to
input and output elements of L in absolute form: as t_COL for elements, as t_MAT in HNF for
ideals, as prid for prime ideals. Without such a call, elements of L are represented as t_POLMOD,
etc. Note that a subsequent nfinit(rnf ) will also explicitly add such a component, and so will the
following functions rnfidealmul, rnfidealtwoelt, rnfidealprimedec, rnfidealup (with flag 1)
and rnfidealreltoabs (with flag 1). The absolute nf structure attached to L can be recovered
using nfinit(rnf).
rnf [1](rnf.pol) contains the relative polynomial T .
rnf [2] contains the integer basis [A, d] of K, as (integral) elements of L/Q. More precisely, A
is a vector of polynomial with integer coefficients, d is a denominator, and the integer basis is given
by A/d.
rnf [3] (rnf.disc) is a two-component row vector [d(L/K), s] where d(L/K) is the relative
ideal discriminant of L/K and s is the discriminant of L/K viewed as an element of K ∗ /(K ∗ )2 , in
other words it is the output of rnfdisc.
rnf [4](rnf.index) is the ideal index f, i.e. such that d(T )ZK = f2 d(L/K).
rnf [5](rnf.p) is the list of rational primes dividing the norm of the relative discriminant ideal.
rnf [7] (rnf.zk) is the pseudo-basis (A, I) for the maximal order ZL as a ZK -module: A is
the relative integral pseudo basis expressed as polynomials (in the variable of T ) with polmod
coefficients in nf , and the second component I is the ideal list of the pseudobasis in HNF.
rnf [8] is the inverse matrix of the integral basis matrix, with coefficients polmods in nf .
rnf [9] is currently unused.
rnf [10] (rnf.nf) is nf .
rnf [11] is an extension of rnfequation(K, T, 1). Namely, a vector [P, a, k, K.pol, T ] de-
scribing the absolute extension L/Q: P is an absolute equation, more conveniently obtained as

434
rnf.polabs; a expresses the generator α = y mod K.pol of the number field K as an element of L,
i.e. a polynomial modulo the absolute equation P ;
k is a small integer such that, if β is an abstract root of T and α the generator of K given
above, then P (β + kα) = 0. It is guaranteed that k = 0 if Q(β) = L.

Caveat. Be careful if k 6= 0 when dealing simultaneously with absolute and relative quantities
since L = Q(β + kα) = K(α), and the generator chosen for the absolute extension is not the same
as for the relative one. If this happens, one can of course go on working, but we advise to change
the relative polynomial so that its root becomes β + kα. Typical GP instructions would be
[P,a,k] = rnfequation(K, T, 1);
if (k, T = subst(T, x, x - k*Mod(y, K.pol)));
L = rnfinit(K, T);
rnf [12] is by default unused and set equal to 0. This field is used to store further information
about the field as it becomes available (which is rarely needed, hence would be too expensive to
compute during the initial rnfinit call).

Huge discriminants, helping rnfdisc. When T has a discriminant which is difficult to factor, it
is hard to compute ZL . As in nfinit, the special input format [T, B] is also accepted, where T is a
polynomial as above and B specifies a list of maximal ideals. The following formats are recognized
for B:
• an integer: the list of all maximal ideals above a rational prime p < B.
• a vector of rational primes or prime ideals: the list of all maximal ideals dividing an element
in the list.
Instead of ZL , this produces an order which is maximal at all such maximal ideals primes.
The result may actually be a complete and correct rnf structure if the relative ideal discriminant
factors completely over this list of maximal ideals but this is not guaranteed. In general, the order
may not be maximal at primes p not in the list such that p2 divides the relative ideal discriminant.
The library syntax is GEN rnfinit0(GEN nf, GEN T, long flag). Also available is GEN
rnfinit(GEN nf, GEN T) (flag = 0).

3.13.183 rnfisabelian(nf , T ). T being a relative polynomial with coefficients in nf , return 1 if it

defines an abelian extension, and 0 otherwise.
? K = nfinit(y^2 + 23);
? rnfisabelian(K, x^3 - 3*x - y)
%2 = 1
The library syntax is long rnfisabelian(GEN nf, GEN T).

3.13.184 rnfisfree(bnf , x). Given bnf as output by bnfinit, and either a polynomial x with
coefficients in bnf defining a relative extension L of bnf , or a pseudo-basis x of such an extension,
returns true (1) if L/bnf is free, false (0) if not.
The library syntax is long rnfisfree(GEN bnf, GEN x).

435
3.13.185 rnfislocalcyclo(rnf ). Let rnf be a relative number field extension L/K as output by
rnfinit whose degree [L : K] is a power of a prime `. Return 1 if the `-extension is locally
cyclotomic (locally contained in the cyclotomic Z` -extension of Kv at all places v|`), and 0 if not.
? K = nfinit(y^2 + y + 1);
? L = rnfinit(K, x^3 - y); /* = K(zeta_9), globally cyclotomic */
? rnfislocalcyclo(L)
%3 = 1
\\ we expect 3-adic continuity by Krasner’s lemma
? vector(5, i, rnfislocalcyclo(rnfinit(K, x^3 - y + 3^i)))
%5 = [0, 1, 1, 1, 1]
The library syntax is long rnfislocalcyclo(GEN rnf).

3.13.186 rnfisnorm(T, a, {flag = 0}). Similar to bnfisnorm but in the relative case. T is as
output by rnfisnorminit applied to the extension L/K. This tries to decide whether the element
a in K is the norm of some x in the extension L/K.
The output is a vector [x, q], where a = Norm(x)∗q. The algorithm looks for a solution x which
is an S-integer, with S a list of places of K containing at least the ramified primes, the generators
of the class group of L, as well as those primes dividing a. If L/K is Galois, then this is enough but
you may want to add more primes to S to produce different elements, possibly smaller; otherwise,
flag is used to add more primes to S: all the places above the primes p ≤ flag (resp. p|flag) if
flag > 0 (resp. flag < 0).
The answer is guaranteed (i.e. a is a norm iff q = 1) if the field is Galois, or, under GRH, if S
contains all primes less than 12 log2 |disc(M )|, where M is the normal closure of L/K.
If rnfisnorminit has determined (or was told) that L/K is Galois, and flag 6= 0, a Warning
is issued (so that you can set flag = 1 to check whether L/K is known to be Galois, according to
T ). Example:
bnf = bnfinit(y^3 + y^2 - 2*y - 1);
p = x^2 + Mod(y^2 + 2*y + 1, bnf.pol);
T = rnfisnorminit(bnf, p);
rnfisnorm(T, 17)
checks whether 17 is a norm in the Galois extension Q(β)/Q(α), where α3 + α2 − 2α − 1 = 0 and
β 2 + α2 + 2α + 1 = 0 (it is).
The library syntax is GEN rnfisnorm(GEN T, GEN a, long flag).

3.13.187 rnfisnorminit(pol , polrel , {flag = 2}). Let K be defined by a root of pol , and L/K the
extension defined by the polynomial polrel . As usual, pol can in fact be an nf , or bnf , etc; if pol
has degree 1 (the base field is Q), polrel is also allowed to be an nf , etc. Computes technical data
needed by rnfisnorm to solve norm equations N x = a, for x in L, and a in K.
If flag = 0, do not care whether L/K is Galois or not.
If flag = 1, L/K is assumed to be Galois (unchecked), which speeds up rnfisnorm.
If flag = 2, let the routine determine whether L/K is Galois.
The library syntax is GEN rnfisnorminit(GEN pol, GEN polrel, long flag).

436
3.13.188 rnfkummer(bnr , {subgp}). This function is deprecated, use bnrclassfield.
The library syntax is GEN rnfkummer(GEN bnr, GEN subgp = NULL, long prec).

3.13.189 rnflllgram(nf , pol , order ). Given a polynomial pol with coefficients in nf defin-
ing a relative extension L and a suborder order of L (of maximal rank), as output by
rnfpseudobasis(nf , pol ) or similar, gives [[neworder ], U ], where neworder is a reduced order and
U is the unimodular transformation matrix.
The library syntax is GEN rnflllgram(GEN nf, GEN pol, GEN order, long prec).

3.13.190 rnfnormgroup(bnr , pol ). bnr being a big ray class field as output by bnrinit and pol a
relative polynomial defining an Abelian extension, computes the norm group (alias Artin or Takagi
group) corresponding to the Abelian extension of bnf =bnr.bnf defined by pol , where the module
corresponding to bnr is assumed to be a multiple of the conductor (i.e. pol defines a subextension of
bnr). The result is the HNF defining the norm group on the given generators of bnr.gen. Note that
neither the fact that pol defines an Abelian extension nor the fact that the module is a multiple of
the conductor is checked. The result is undefined if the assumption is not correct, but the function
will return the empty matrix [;] if it detects a problem; it may also not detect the problem and
return a wrong result.
The library syntax is GEN rnfnormgroup(GEN bnr, GEN pol).

3.13.191 rnfpolred(nf , pol ). This function is obsolete: use rnfpolredbest instead. Relative
version of polred. Given a monic polynomial pol with coefficients in nf , finds a list of relative
polynomials defining some subfields, hopefully simpler and containing the original field. In the
present version 2.13.3, this is slower and less efficient than rnfpolredbest.
Remark. This function is based on an incomplete reduction theory of lattices over number fields,
implemented by rnflllgram, which deserves to be improved.
The library syntax is GEN rnfpolred(GEN nf, GEN pol, long prec).

3.13.192 rnfpolredabs(nf , pol , {flag = 0}). Relative version of polredabs. Given an irreducible
monic polynomial pol with coefficients in the maximal order of nf , finds a canonical relative poly-
nomial defining the same field, hopefully with small coefficients. Note that the equation is only
canonical for a fixed nf , using a different defining polynomial in the nf structure will produce a
different relative equation.
The binary digits of flag correspond to 1: add information to convert elements to the new
representation, 2: absolute polynomial, instead of relative, 16: possibly use a suborder of the
maximal order. More precisely:
0: default, return P
1: returns [P, a] where P is the default output and a, a t_POLMOD modulo P , is a root of pol .
2: returns Pabs, an absolute, instead of a relative, polynomial. This polynomial is canonical
and does not depend on the nf structure. Same as but faster than
polredabs(rnfequation(nf, pol))
3: returns [Pabs, a, b], where Pabs is an absolute polynomial as above, a, b are t_POLMOD
modulo Pabs, roots of nf.pol and pol respectively.
16: (OBSOLETE) possibly use a suborder of the maximal order. This makes rnfpolredabs
behave as rnfpolredbest. Just use the latter.

437
Warning. The complexity of rnfpolredabs is exponential in the absolute degree. The function
rnfpolredbest runs in polynomial time, and tends to return polynomials with smaller discrimi-
nants. It also supports polynomials with arbitrary coefficients in nf , neither integral nor necessarily
monic.
The library syntax is GEN rnfpolredabs(GEN nf, GEN pol, long flag).

3.13.193 rnfpolredbest(nf , pol , {flag = 0}). Relative version of polredbest. Given a polynomial
pol with coefficients in nf , finds a simpler relative polynomial P defining the same field. As opposed
to rnfpolredabs this function does not return a smallest (canonical) polynomial with respect to
some measure, but it does run in polynomial time.
The binary digits of flag correspond to 1: add information to convert elements to the new
representation, 2: absolute polynomial, instead of relative. More precisely:
0: default, return P
1: returns [P, a] where P is the default output and a, a t_POLMOD modulo P , is a root of pol .
2: returns Pabs, an absolute, instead of a relative, polynomial. Same as but faster than

rnfequation(nf, rnfpolredbest(nf,pol))
3: returns [Pabs, a, b], where Pabs is an absolute polynomial as above, a, b are t_POLMOD
modulo Pabs, roots of nf.pol and pol respectively.

? K = nfinit(y^3-2); pol = x^2 +x*y + y^2;

? [P, a] = rnfpolredbest(K,pol,1);
? P
%3 = x^2 - x + Mod(y - 1, y^3 - 2)
? a
%4 = Mod(Mod(2*y^2+3*y+4,y^3-2)*x + Mod(-y^2-2*y-2,y^3-2),
x^2 - x + Mod(y-1,y^3-2))
? subst(K.pol,y,a)
%5 = 0
? [Pabs, a, b] = rnfpolredbest(K,pol,3);
? Pabs
%7 = x^6 - 3*x^5 + 5*x^3 - 3*x + 1
? a
%8 = Mod(-x^2+x+1, x^6-3*x^5+5*x^3-3*x+1)
? b
%9 = Mod(2*x^5-5*x^4-3*x^3+10*x^2+5*x-5, x^6-3*x^5+5*x^3-3*x+1)
? subst(K.pol,y,a)
%10 = 0
? substvec(pol,[x,y],[a,b])
%11 = 0
The library syntax is GEN rnfpolredbest(GEN nf, GEN pol, long flag).

438
3.13.194 rnfpseudobasis(nf , T ). Given an nf structure attached to a number field K, as output
by nfinit, and a monic irreducible polynomial T in ZK [x] defining a relative extension L =
K[x]/(T ), computes the relative discriminant of L and a pseudo-basis (A, J) for the maximal order
ZL viewed as a ZK -module. This is output as a vector [A, J, D, d], where D is the relative ideal
discriminant and d is the relative discriminant considered as an element of K ∗ /K ∗ 2 .

? K = nfinit(y^2+1);
? [A,J,D,d] = rnfpseudobasis(K, x^2+y);
? A
%3 =
[1 0]
[0 1]
? J
%4 = [1, 1]
? D
%5 = [0, -4]~
? d
%6 = [0, -1]~

Huge discriminants, helping rnfdisc. The format [T, B] is also accepted instead of T and
produce an order which is maximal at all prime ideals specified by B, see ??rnfinit.

? p = 585403248812100232206609398101;
? q = 711171340236468512951957953369;
? T = x^2 + 3*(p*q)^2;
? [A,J,D,d] = V = rnfpseudobasis(K, T); D
time = 22,178 ms.
%10 = 3
? [A,J,D,d] = W = rnfpseudobasis(K, [T,100]); D
time = 5 ms.
%11 = 3
? V == W
%12 = 1
? [A,J,D,d] = W = rnfpseudobasis(K, [T, [3]]); D
%13 = 3
? V == W
%14 = 1

In this example, the results are identical since D ∩ Z factors over primes less than 100 (and in fact,
over 3). Had it not been the case, the order would have been guaranteed maximal at primes p|p for
p ≤ 100 only (resp. p|3). And might have been nonmaximal at any other prime ideal p such that
p2 divided D.

The library syntax is GEN rnfpseudobasis(GEN nf, GEN T).

439
3.13.195 rnfsteinitz(nf , x). Given a number field nf as output by nfinit and either a polynomial
x with coefficients in nf defining a relative extension L of nf , or a pseudo-basis x of such an extension
as output for example by rnfpseudobasis, computes another pseudo-basis (A, I) (not in HNF in
general) such that all the ideals of I except perhaps the last one are equal to the ring of integers
of nf , and outputs the four-component row vector [A, I, D, d] as in rnfpseudobasis. The name of
this function comes from the fact that the ideal class of the last ideal of I, which is well defined, is
the Steinitz class of the ZK -module ZL (its image in SK0 (ZK )).

The library syntax is GEN rnfsteinitz(GEN nf, GEN x).

3.13.196 subgrouplist(cyc, {bound }, {flag = 0}). cyc being a vector of positive integers giving
the cyclic components for a finite Abelian group G (or any object which has a .cyc method),
outputs the list of subgroups of G. Subgroups are given as HNF left divisors of the SNF matrix
corresponding to G.

If flag = 0 (default) and cyc is a bnr structure output by bnrinit, gives only the subgroups
whose modulus is the conductor. Otherwise, all subgroups are given.

If bound is present, and is a positive integer, restrict the output to subgroups of index less than
bound . If bound is a vector containing a single positive integer B, then only subgroups of index
exactly equal to B are computed. For instance

? subgrouplist([6,2])
%1 = [[6, 0; 0, 2], [2, 0; 0, 2], [6, 3; 0, 1], [2, 1; 0, 1], [3, 0; 0, 2],
[1, 0; 0, 2], [6, 0; 0, 1], [2, 0; 0, 1], [3, 0; 0, 1], [1, 0; 0, 1]]
? subgrouplist([6,2],3) \\ index less than 3
%2 = [[2, 1; 0, 1], [1, 0; 0, 2], [2, 0; 0, 1], [3, 0; 0, 1], [1, 0; 0, 1]]
? subgrouplist([6,2],[3]) \\ index 3
%3 = [[3, 0; 0, 1]]
? bnr = bnrinit(bnfinit(x), [120,[1]], 1);
? L = subgrouplist(bnr, [8]);

In the last example, L corresponds to the 24 subfields of Q(ζ120 ), of degree 8 and conductor 120∞
(by setting flag, we see there are a total of 43 subgroups of degree 8).

? vector(#L, i, galoissubcyclo(bnr, L[i]))

will produce their equations. (For a general base field, you would have to rely on bnrstark, or
bnrclassfield.)

Warning. This function requires factoring the exponent of G. If you are only interested in
subgroups of index n (or dividing n), you may considerably speed up the function by computing
the subgroups of G/Gn , whose cyclic components are apply(x->gcd(n,x), C) (where C gives the
cyclic components of G). If you want the bnr variant, now is a good time to use bnrinit(,,, n)
as well, to directly compute the ray class group modulo n-th powers.

The library syntax is GEN subgrouplist0(GEN cyc, GEN bound = NULL, long flag).

440
3.14 Associative and central simple algebras.
This section collects functions related to associative algebras and central simple algebras (CSA)
over number fields.

3.14.1 Algebra definitions.

Let A be a finite-dimensional unital associative algebra over a field K. The algebra A is central
if its center is K and it is simple if it has no nontrivial two-sided ideals.

We provide functions to handle associative algebras of finite dimension over Q or Fp . We

represent them by the left multiplication table on a basis over the prime subfield; the function
algtableinit creates the object representing an associative algebra. We also provide functions to
handle central simple algebras over a number field K. We represent them either by the left multi-
plication table on a basis over the center K or by a cyclic algebra (see below); the function alginit
creates the object representing a central simple algebra.

The set of elements of an algebra A that annihilate every simple left A-module is a two-sided
ideal, called the Jacobson radical of A. If the Jacobson radical is trivial, the algebra is semisimple:
it is isomorphic to a direct product of simple algebras. The dimension of a CSA over its center K
is always a square d2 and the integer d is called the degree of the algebra over K. A CSA over a
field K is always isomorphic to Mk (D) for some integer k and some central division algebra D of
degree e: the integer e is the index of the algebra.

Let L/K be a cyclic extension of degree d, let σ be a generator of Gal(L/K) and let b ∈ K ∗ .
Ld−1
Then the cyclic algebra (L/K, σ, b) is the algebra i=0 xi L with xd = b and `x = xσ(`) for all ` ∈ L.
The algebra (L/K, σ, b) is a central simple K-algebra of degree d, and it is an L-vector space. Left
multiplication is L-linear and induces a K-algebra isomorphism (L/K, σ, b) ⊗K L → Md (L).

Let K be a nonarchimedean local field with uniformizer π, and let L/K be the unique unram-
ified extension of degree d. Then every central simple algebra A of degree d over K is isomorphic
to (L/K, Frob, π h ) for some integer h. The element h/d ∈ Q/Z is called the Hasse invariant of A.

3.14.2 Orders in algebras.

Let A be an algebra of finite dimension over Q. An order in A is a finitely generated

Z-submodule O such that QO = A, that is also a subring with unit. By default the data
computed by alginit contains a Z-basis of a maximal order O0 . We define natural orders in
central simple algebrasLdefined by a cyclic algebra or by a multiplication table over the center.
d−1
Let A = (L/K, σ, b) = i=0 xi L be a cyclic algebra over a number field K of degree n with ring of
integers Z . Let ZL be the ring of integers of L, and assume that b is integral. Then the submod-
LKd−1
ule O = i=0 xi ZL is an order in A, called the natural order . Let ω0 , . . . , ωnd−1 be a Z-basis of ZL .
The natural basis of O is b0 , . . . , bnd2 −1 where bi = xi/(nd) ω(i mod nd) . Now let A be a central simple
algebra of degree d over a number field K of degree n with ring of integers ZK . Let e0 , . . . , ed2 −1
be a basis of A over K and assume that the left multiplication table of A on (ei ) is integral. Then
Ld2 −1
the submodule O = i=0 ZK ei is an order in A, called the natural order . Let ω0 , . . . , ωn−1 be
a Z-basis of ZK . The natural basis of O is b0 , . . . , bnd2 −1 where bi = ω(i mod n) ei/n .

441
3.14.3 Lattices in algebras.

We also provide functions to handle full lattices in algebras over Q. A full lattice J ⊂ A is
represented by a 2-component t_VEC [I, t] representing J = tI, where

• I is an integral nonsingular upper-triangular matrix representing a sublattice of O0 expressed

on the integral basis, and

• t ∈ Q>0 is a t_INT or t_FRAC.

For the sake of efficiency you should use matrices I that are primitive and in Hermite Normal
Form; this makes the representation unique. No GP function uses this property, but all GP functions
return lattices in this form. The prefix for lattice functions is alglat.

3.14.4 GP conventions for algebras.

As with number fields, we represent elements of central simple algebras in two ways, called
the algebraic representation and the basis representation, and you can convert betweeen the two
with the functions algalgtobasis and algbasistoalg. In every central simple algebra object,
we store a Z-basis of an order O0 , and the basis representation is simply a t_COL with coefficients
in Q expressing the element in that basis. If no maximal order was computed by alginit, then O0
is the natural order. If a maximal order was computed, then O0 is a maximal order containing
the natural order. For a cyclic algebra A = (L/K, σ, b), the algebraic representation is a t_COL
Ld−1
with coefficients in L representing the element in the decomposition A = i=0 xi L. For a central
simple algebra defined by a multiplication table over its center K on a basis (ei ), the algebraic
representation is a t_COL with coefficients in K representing the element on the basis (ei ).
Ld−1
Warning. The coefficients in the decomposition A = i=0 xi L are not the same as those in the
Ld−1 i
decomposition A = i=0 Lx ! The i-th coefficients are related by conjugating by xi , which on L
amounts to acting by σ i .

Warning. For a central simple algebra over Q defined by a multiplication table, we cannot
distinguish between the basis and the algebraic representations from the size of the vectors. The
behavior is then to always interpret the column vector as a basis representation if the coefficients
are t_INT or t_FRAC, and as an algebraic representation if the coefficients are t_POL or t_POLMOD.

3.14.5 algadd(al , x, y). Given two elements x and y in al , computes their sum x + y in the
algebra al .

? A = alginit(nfinit(y),[-1,1]);
? algadd(A,[1,0]~,[1,2]~)
%2 = [2, 2]~

Also accepts matrices with coefficients in al .

The library syntax is GEN algadd(GEN al, GEN x, GEN y).

442
3.14.6 algalgtobasis(al , x). Given an element x in the central simple algebra al output by al-
ginit, transforms it to a column vector on the integral basis of al . This is the inverse function of
algbasistoalg.
? A = alginit(nfinit(y^2-5),[2,y]);
? algalgtobasis(A,[y,1]~)
%2 = [0, 2, 0, -1, 2, 0, 0, 0]~
? algbasistoalg(A,algalgtobasis(A,[y,1]~))
%3 = [Mod(Mod(y, y^2 - 5), x^2 - 2), 1]~
The library syntax is GEN algalgtobasis(GEN al, GEN x).

3.14.7 algaut(al ). Given a cyclic algebra al = (L/K, σ, b) output by alginit, returns the auto-
morphism σ.
? nf = nfinit(y);
? p = idealprimedec(nf,7)[1];
? p2 = idealprimedec(nf,11)[1];
? A = alginit(nf,[3,[[p,p2],[1/3,2/3]],[0]]);
? algaut(A)
%5 = -1/3*x^2 + 1/3*x + 26/3
The library syntax is GEN algaut(GEN al).

3.14.8 algb(al ). Given a cyclic algebra al = (L/K, σ, b) output by alginit, returns the element
b ∈ K.
nf = nfinit(y);
? p = idealprimedec(nf,7)[1];
? p2 = idealprimedec(nf,11)[1];
? A = alginit(nf,[3,[[p,p2],[1/3,2/3]],[0]]);
? algb(A)
%5 = Mod(-77, y)
The library syntax is GEN algb(GEN al).

3.14.9 algbasis(al ). Given a central simple algebra al output by alginit, returns a Z-basis of
the order O0 stored in al with respect to the natural order in al . It is a maximal order if one has
been computed.
A = alginit(nfinit(y), [-1,-1]);
? algbasis(A)
%2 =
[1 0 0 1/2]
[0 1 0 1/2]
[0 0 1 1/2]
[0 0 0 1/2]
The library syntax is GEN algbasis(GEN al).

443
3.14.10 algbasistoalg(al , x). Given an element x in the central simple algebra al output by
alginit, transforms it to its algebraic representation in al . This is the inverse function of algal-
gtobasis.
? A = alginit(nfinit(y^2-5),[2,y]);
? z = algbasistoalg(A,[0,1,0,0,2,-3,0,0]~);
? liftall(z)
%3 = [(-1/2*y - 2)*x + (-1/4*y + 5/4), -3/4*y + 7/4]~
? algalgtobasis(A,z)
%4 = [0, 1, 0, 0, 2, -3, 0, 0]~
The library syntax is GEN algbasistoalg(GEN al, GEN x).

3.14.11 algcenter(al ). If al is a table algebra output by algtableinit, returns a basis of the

center of the algebra al over its prime field (Q or Fp ). If al is a central simple algebra output by
alginit, returns the center of al , which is stored in al .
A simple example: the 2 × 2 upper triangular matrices over Q, generated by I2 , a = [0, 1; 0, 0]
and b = [0, 0; 0, 1], such that a2 = 0, ab = a, ba = 0, b2 = b: the diagonal matrices form the center.
? mt = [matid(3),[0,0,0;1,0,1;0,0,0],[0,0,0;0,0,0;1,0,1]];
? A = algtableinit(mt);
? algcenter(A) \\ = (I_2)
%3 =
[1]
[0]
[0]
An example in the central simple case:
? nf = nfinit(y^3-y+1);
? A = alginit(nf, [-1,-1]);
? algcenter(A).pol
%3 = y^3 - y + 1
The library syntax is GEN algcenter(GEN al).

3.14.12 algcentralproj(al , z, {maps = 0}). Given a table algebra al output by algtableinit

and a t_VEC z = [z1 , . . . , zn ] of orthogonal central idempotents, returns a t_VEC [al1 , . . . , aln ] of
algebras such that ali = zi al. If maps = 1, each ali is a t_VEC [quo, proj, lif t] where quo is the
quotient algebra, proj is a t_MAT representing the projection onto this quotient and lift is a t_MAT
representing a lift.
A simple example: F2 × F4 , generated by 1 = (1, 1), e = (1, 0) and x such that x2 + x + 1 = 0.
We have e2 = e, x2 = x + 1 and ex = 0.
? mt = [matid(3), [0,0,0; 1,1,0; 0,0,0], [0,0,1; 0,0,0; 1,0,1]];
? A = algtableinit(mt,2);
? e = [0,1,0]~;
? e2 = algsub(A,[1,0,0]~,e);
? [a,a2] = algcentralproj(A,[e,e2]);
? algdim(a)
%6 = 1

444
 ? algdim(a2)
%7 = 2
The library syntax is GEN alg_centralproj(GEN al, GEN z, long maps).

3.14.13 algchar(al ). Given an algebra al output by alginit or algtableinit, returns the

characteristic of al .
? mt = [matid(3), [0,0,0; 1,1,0; 0,0,0], [0,0,1; 0,0,0; 1,0,1]];
? A = algtableinit(mt,13);
? algchar(A)
%3 = 13
The library syntax is GEN algchar(GEN al).

3.14.14 algcharpoly(al , b, {v =0 x}, {abs = 0}). Given an element b in al , returns its characteristic
polynomial as a polynomial in the variable v. If al is a table algebra output by algtableinit or if
abs = 1, returns the absolute characteristic polynomial of b, which is an element of Fp [v] or Q[v];
if al is a central simple algebra output by alginit and abs = 0, returns the reduced characteristic
polynomial of b, which is an element of K[v] where K is the center of al .
? al = alginit(nfinit(y), [-1,-1]); \\ (-1,-1)_Q
? algcharpoly(al, [0,1]~)
%2 = x^2 + 1
? algcharpoly(al, [0,1]~,,1)
%3 = x^4 + 2*x^2 + 1
? nf = nfinit(y^2-5);
? al = alginit(nf,[-1,y]);
? a = [y,1+x]~*Mod(1,y^2-5)*Mod(1,x^2+1);
? P = lift(algcharpoly(al,a))
%7 = x^2 - 2*y*x + (-2*y + 5)
? algcharpoly(al,a,,1)
%8 = x^8 - 20*x^6 - 80*x^5 + 110*x^4 + 800*x^3 + 1500*x^2 - 400*x + 25
? lift(P*subst(P,y,-y)*Mod(1,y^2-5))^2
%9 = x^8 - 20*x^6 - 80*x^5 + 110*x^4 + 800*x^3 + 1500*x^2 - 400*x + 25
Also accepts a square matrix with coefficients in al .
The library syntax is GEN algcharpoly(GEN al, GEN b, long v = -1, long abs) where v
is a variable number.

3.14.15 algdegree(al ). Given a central simple algebra al output by alginit, returns the degree
of al .
? nf = nfinit(y^3-y+1);
? A = alginit(nf, [-1,-1]);
? algdegree(A)
%3 = 2
The library syntax is long algdegree(GEN al).

445
3.14.16 algdim(al , {abs = 0}). If al is a table algebra output by algtableinit or if abs = 1,
returns the dimension of al over its prime subfield (Q or Fp ). If al is a central simple algebra
output by alginit and abs = 0, returns the dimension of al over its center.

? nf = nfinit(y^3-y+1);
? A = alginit(nf, [-1,-1]);
? algdim(A)
%3 = 4
? algdim(A,1)
%4 = 12

The library syntax is long algdim(GEN al, long abs).

3.14.17 algdisc(al ). Given a central simple algebra al output by alginit, computes the discrim-
inant of the order O0 stored in al , that is the determinant of the trace form Tr : O0 × O0 → Z.

? nf = nfinit(y^2-5);
? A = alginit(nf, [-3,1-y]);
? [PR,h] = alghassef(A)
%3 = [[[2, [2, 0]~, 1, 2, 1], [3, [3, 0]~, 1, 2, 1]], Vecsmall([0, 1])]
? n = algdegree(A);
? D = algdim(A,1);
? h = vector(#h, i, n - gcd(n,h[i]));
? n^D * nf.disc^(n^2) * idealnorm(nf, idealfactorback(nf,PR,h))^n
%4 = 12960000
? algdisc(A)
%5 = 12960000

The library syntax is GEN algdisc(GEN al).

3.14.18 algdivl(al , x, y). Given two elements x and y in al , computes their left quotient x\y in the
algebra al : an element z such that xz = y (such an element is not unique when x is a zerodivisor).
If x is invertible, this is the same as x−1 y. Assumes that y is left divisible by x (i.e. that z exists).
Also accepts matrices with coefficients in al .

The library syntax is GEN algdivl(GEN al, GEN x, GEN y).

3.14.19 algdivr(al , x, y). Given two elements x and y in al , returns xy −1 . Also accepts matrices
with coefficients in al .

The library syntax is GEN algdivr(GEN al, GEN x, GEN y).

446
3.14.20 alggroup(gal , {p = 0}). Initializes the group algebra K[G] over K = Q (p omitted) or Fp
where G is the underlying group of the galoisinit structure gal . The input gal is also allowed to
be a t_VEC of permutations that is closed under products.
Example:

? K = nfsplitting(x^3-x+1);
? gal = galoisinit(K);
? al = alggroup(gal);
? algissemisimple(al)
%4 = 1
? G = [Vecsmall([1,2,3]), Vecsmall([1,3,2])];
? al2 = alggroup(G, 2);
? algissemisimple(al2)
%8 = 0
The library syntax is GEN alggroup(GEN gal, GEN p = NULL).

3.14.21 alggroupcenter(gal , {p = 0}, {&cc}). Initializes the center Z(K[G]) of the group alge-
bra K[G] over K = Q (p = 0 or omitted) or Fp where G is the underlying group of the galoisinit
structure gal . The input gal is also allowed to be a t_VEC of permutations that is closed under
products. Sets cc to a t_VEC [elts, conjclass, rep, flag] where elts is a sorted t_VEC containing the
list of elements of G, conjclass is a t_VECSMALL of the same length as elts containing the index of
the conjugacy class of the corresponding element (an integer between 1 and the number of conju-
gacy classes), and rep is a t_VECSMALL of length the number of conjugacy classes giving for each
conjugacy class the index in elts of a representative of this conjugacy class. Finally flag is 1 if and
only if the permutation representation of G is transitive, in which case the i-th element of elts is
characterized by g[1] = i; this is always the case when gal is a galoisinit structure. The basis
of Z(K[G]) as output consists of the indicator functions of the conjugacy classes in the ordering
given by cc. Example:

? K = nfsplitting(x^4+x+1);
? gal = galoisinit(K); \\ S4
? al = alggroupcenter(gal,,&cc);
? algiscommutative(al)
%4 = 1
? #cc[3] \\ number of conjugacy classes of S4
%5 = 5
? gal = [Vecsmall([1,2,3]),Vecsmall([1,3,2])]; \\ C2
? al = alggroupcenter(gal,,&cc);
? cc[3]
%8 = Vecsmall([1, 2])
? cc[4]
%9 = 0
The library syntax is GEN alggroupcenter(GEN gal, GEN p = NULL, GEN *cc = NULL)
.

447
3.14.22 alghasse(al , pl ). Given a central simple algebra al output by alginit and a prime ideal
or an integer between 1 and r1 + r2 , returns a t_FRAC h : the local Hasse invariant of al at the
place specified by pl .

? nf = nfinit(y^2-5);
? A = alginit(nf, [-1,y]);
? alghasse(A, 1)
%3 = 1/2
? alghasse(A, 2)
%4 = 0
? alghasse(A, idealprimedec(nf,2)[1])
%5 = 1/2
? alghasse(A, idealprimedec(nf,5)[1])
%6 = 0

The library syntax is GEN alghasse(GEN al, GEN pl).

3.14.23 alghassef(al ). Given a central simple algebra al output by alginit, returns a t_VEC
[PR, hf ] describing the local Hasse invariants at the finite places of the center: PR is a t_VEC of
primes and hf is a t_VECSMALL of integers modulo the degree d of al . The Hasse invariant of al at
the primes outside PR is 0, but PR can include primes at which the Hasse invariant is 0.

? nf = nfinit(y^2-5);
? A = alginit(nf, [-1,2*y-1]);
? [PR,hf] = alghassef(A);
? PR
%4 = [[19, [10, 2]~, 1, 1, [-8, 2; 2, -10]], [2, [2, 0]~, 1, 2, 1]]
? hf
%5 = Vecsmall([1, 0])

The library syntax is GEN alghassef(GEN al).

3.14.24 alghassei(al ). Given a central simple algebra al output by alginit, returns a t_VECSMALL
hi of r1 integers modulo the degree d of al , where r1 is the number of real places of the center: the
local Hasse invariants of al at infinite places.

? nf = nfinit(y^2-5);
? A = alginit(nf, [-1,y]);
? alghassei(A)
%3 = Vecsmall([1, 0])

The library syntax is GEN alghassei(GEN al).

448
3.14.25 algindex(al , {pl }). Returns the index of the central simple algebra A over K (as output
by alginit), that is the degree e of the unique central division algebra D over K such that A is
isomorphic to some matrix algebra Mk (D). If pl is set, it should be a prime ideal of K or an integer
between 1 and r1 + r2 , and in that case return the local index at the place pl instead.
? nf = nfinit(y^2-5);
? A = alginit(nf, [-1,y]);
? algindex(A, 1)
%3 = 2
? algindex(A, 2)
%4 = 1
? algindex(A, idealprimedec(nf,2)[1])
%5 = 2
? algindex(A, idealprimedec(nf,5)[1])
%6 = 1
? algindex(A)
%7 = 2
The library syntax is long algindex(GEN al, GEN pl = NULL).

3.14.26 alginit(B, C, {v}, {maxord = 1}). Initializes the central simple algebra defined by data
B, C and variable v, as follows.
• (multiplication table) B is the base number field K in nfinit form, C is a “multiplication
table” over K. As a K-vector space, the algebra is generated by a basis (e1 = 1, . . . , en ); the
table is given as a t_VEC of n matrices in Mn (K), giving the left multiplication by the basis
elements
Lnei , in the given basis. Assumes that e1 = 1, that the multiplication table is integral, and
that ( i=1 Kei , C) describes a central simple algebra over K.
{ mi = [0,-1,0, 0;
1, 0,0, 0;
0, 0,0,-1;
0, 0,1, 0];
mj = [0, 0,-1,0;
0, 0, 0,1;
1, 0, 0,0;
0,-1, 0,0];
mk = [0, 0, 0, 0;
0, 0,-1, 0;
0, 1, 0, 0;
1, 0, 0,-1];
A = alginit(nfinit(y), [matid(4), mi,mj,mk], 0); }
represents (in a complicated way) the quaternion algebra (−1, −1)Q . See below for a simpler
solution.
• (cyclic algebra) B is an rnf structure attached to a cyclic number field extension L/K of
degree d, C is a t_VEC [sigma,b] with 2 components: sigma is a t_POLMOD representing an automor-
phism generating Gal(L/K), b is an element in K ∗ . This represents the cyclic algebra (L/K, σ, b).
Currently the element b has to be integral.
? Q = nfinit(y); T = polcyclo(5, ’x); F = rnfinit(Q, T);

449
 ? A = alginit(F, [Mod(x^2,T), 3]);
defines the cyclic algebra (L/Q, σ, 3), where L = Q(ζ5 ) and σ : ζ 7→ ζ 2 generates Gal(L/Q).
• (quaternion algebra, special case of the above) B is an nf structure attached to a number
field K, C = [a, b] is a vector containing two elements of K ∗ with a not a square in K, returns
the quaternion algebra (a, b)K . The variable v (’x by default) must have higher priority than the
variable of K.pol and is used to represent elements in the splitting field L = K[x]/(x2 − a).
? Q = nfinit(y); A = alginit(Q, [-1,-1]); \\ (−1, −1)Q
• (algebra/K defined by local Hasse invariants) B is an nf structure attached to a number
field K, C = [d, [PR, hf ], hi ] is a triple containing an integer d > 1, a pair [PR, hf ] describing the
Hasse invariants at finite places, and hi the Hasse invariants at archimedean (real) places. A local
Hasse invariant belongs to (1/d)Z/Z ⊂ Q/Z, and is given either as a t_FRAC (lift to (1/d)Z), a
t_INT or t_INTMOD modulo d (lift to Z/dZ); a whole vector of local invariants can also be given as
a t_VECSMALL, whose entries are handled as t_INTs. PR is a list of prime ideals (prid structures),
and hf is a vector of the same length giving the local invariants at those maximal ideals. The
invariants at infinite real places are indexed by the real roots K.roots: if the Archimedean place
v is attached to the j-th root, the value of hv is given by hi [j], must be 0 or 1/2 (or d/2 modulo d),
and can be nonzero only if d is even.
By class field theory, provided the local invariants hv sum to 0, up to Brauer equivalence, there
is a unique central simple algebra over K with given local invariants and trivial invariant elsewhere.
In particular, up to isomorphism, there is a unique such algebra A of degree d.
We realize A as a cyclic algebra through class field theory. The variable v (’x by default) must
have higher priority than the variable of K.pol and is used to represent elements in the (cyclic)
splitting field extension L/K for A.
? nf = nfinit(y^2+1);
? PR = idealprimedec(nf,5); #PR
%2 = 2
? hi = [];
? hf = [PR, [1/3,-1/3]];
? A = alginit(nf, [3,hf,hi]);
? algsplittingfield(A).pol
%6 = x^3 - 21*x + 7
• (matrix algebra, toy example) B is an nf structure attached to a number field K, C = d is
a positive integer. Returns a cyclic algebra isomorphic to the matrix algebra Md (K).
In all cases, this function computes a maximal order for the algebra by default, which may
require a lot of time. Setting maxord = 0 prevents this computation.
The pari object representing such an algebra A is a t_VEC with the following data:
• A splitting field L of A of the same degree over K as A, in rnfinit format, accessed with
algsplittingfield.
• The Hasse invariants at the real places of K, accessed with alghassei.
• The Hasse invariants of A at the finite primes of K that ramify in the natural order of A,
accessed with alghassef.
• A basis of an order O0 expressed on the basis of the natural order, accessed with algbasis.

450
 • A basis of the natural order expressed on the basis of O0 , accessed with alginvbasis.

• The left multiplication table of O0 on the previous basis, accessed with algmultable.

• The characteristic of A (always 0), accessed with algchar.

• The absolute traces of the elements of the basis of O0 .

• If A was constructed as a cyclic algebra (L/K, σ, b) of degree d, a t_VEC [σ, σ 2 , . . . , σ d−1 ].

The function algaut returns σ.

• If A was constructed as a cyclic algebra (L/K, σ, b), the element b, accessed with algb.

• If A was constructed with its multiplication table mt over K, the t_VEC of t_MAT mt, accessed
with algrelmultable.

• If A was constructed with its multiplication table mt over K, a t_VEC with three components:
a t_COL representing an element of A generating the splitting field L as a maximal subfield of A,
a t_MAT representing an L-basis B of A expressed on the Z-basis of O0 , and a t_MAT representing
the Z-basis of O0 expressed on B. This data is accessed with algsplittingdata.

The library syntax is GEN alginit(GEN B, GEN C, long v = -1, long maxord) where v is
a variable number.

3.14.27 alginv(al , x). Given an element x in al , computes its inverse x−1 in the algebra al .
Assumes that x is invertible.

? A = alginit(nfinit(y), [-1,-1]);
? alginv(A,[1,1,0,0]~)
%2 = [1/2, 1/2, 0, 0]~

Also accepts matrices with coefficients in al .

The library syntax is GEN alginv(GEN al, GEN x).

3.14.28 alginvbasis(al ). Given an central simple algebra al output by alginit, returns a Z-basis
of the natural order in al with respect to the order O0 stored in al .

A = alginit(nfinit(y), [-1,-1]);
? alginvbasis(A)
%2 =
[1 0 0 -1]
[0 1 0 -1]
[0 0 1 -1]
[0 0 0 2]

The library syntax is GEN alginvbasis(GEN al).

451
3.14.29 algisassociative(mt, p = 0). Returns 1 if the multiplication table mt is suitable for
algtableinit(mt,p), 0 otherwise. More precisely, mt should be a t_VEC of n matrices in Mn (K),
giving the left multiplications by the basis elements e1 , . . . , en (structure constants). We check
whether the first basis element e1 is 1 and ei (ej ek ) = (ei ej )ek for all i, j, k.

? mt = [matid(3),[0,0,0;1,0,1;0,0,0],[0,0,0;0,0,0;1,0,1]];
? algisassociative(mt)
%2 = 1
May be used to check a posteriori an algebra: we also allow mt as output by algtableinit (p
is ignored in this case).
The library syntax is GEN algisassociative(GEN mt, GEN p).

3.14.30 algiscommutative(al ). al being a table algebra output by algtableinit or a central

simple algebra output by alginit, tests whether the algebra al is commutative.

? mt = [matid(3),[0,0,0;1,0,1;0,0,0],[0,0,0;0,0,0;1,0,1]];
? A = algtableinit(mt);
? algiscommutative(A)
%3 = 0
? mt = [matid(3), [0,0,0; 1,1,0; 0,0,0], [0,0,1; 0,0,0; 1,0,1]];
? A = algtableinit(mt,2);
? algiscommutative(A)
%6 = 1
The library syntax is GEN algiscommutative(GEN al).

3.14.31 algisdivision(al , {pl }). Given a central simple algebra al output by alginit, tests
whether al is a division algebra. If pl is set, it should be a prime ideal of K or an integer be-
tween 1 and r1 + r2 , and in that case tests whether al is locally a division algebra at the place pl
instead.

? nf = nfinit(y^2-5);
? A = alginit(nf, [-1,y]);
? algisdivision(A, 1)
%3 = 1
? algisdivision(A, 2)
%4 = 0
? algisdivision(A, idealprimedec(nf,2)[1])
%5 = 1
? algisdivision(A, idealprimedec(nf,5)[1])
%6 = 0
? algisdivision(A)
%7 = 1
The library syntax is GEN algisdivision(GEN al, GEN pl = NULL).

452
3.14.32 algisdivl(al , x, y, {&z}). Given two elements x and y in al , tests whether y is left divisible
by x, that is whether there exists z in al such that xz = y, and sets z to this element if it exists.
? A = alginit(nfinit(y), [-1,1]);
? algisdivl(A,[x+2,-x-2]~,[x,1]~)
%2 = 0
? algisdivl(A,[x+2,-x-2]~,[-x,x]~,&z)
%3 = 1
? z
%4 = [Mod(-2/5*x - 1/5, x^2 + 1), 0]~
Also accepts matrices with coefficients in al .
The library syntax is GEN algisdivl(GEN al, GEN x, GEN y, GEN *z = NULL).

3.14.33 algisinv(al , x, {&ix }). Given an element x in al , tests whether x is invertible, and sets ix
to the inverse of x.
? A = alginit(nfinit(y), [-1,1]);
? algisinv(A,[-1,1]~)
%2 = 0
? algisinv(A,[1,2]~,&ix)
%3 = 1
? ix
%4 = [Mod(Mod(-1/3, y), x^2 + 1), Mod(Mod(2/3, y), x^2 + 1)]~
Also accepts matrices with coefficients in al .
The library syntax is GEN algisinv(GEN al, GEN x, GEN *ix = NULL).

3.14.34 algisramified(al , {pl }). Given a central simple algebra al output by alginit, tests
whether al is ramified, i.e. not isomorphic to a matrix algebra over its center. If pl is set, it should
be a prime ideal of K or an integer between 1 and r1 + r2 , and in that case tests whether al is
locally ramified at the place pl instead.
? nf = nfinit(y^2-5);
? A = alginit(nf, [-1,y]);
? algisramified(A, 1)
%3 = 1
? algisramified(A, 2)
%4 = 0
? algisramified(A, idealprimedec(nf,2)[1])
%5 = 1
? algisramified(A, idealprimedec(nf,5)[1])
%6 = 0
? algisramified(A)
%7 = 1
The library syntax is GEN algisramified(GEN al, GEN pl = NULL).

453
3.14.35 algissemisimple(al ). al being a table algebra output by algtableinit or a central
simple algebra output by alginit, tests whether the algebra al is semisimple.

? mt = [matid(3),[0,0,0;1,0,1;0,0,0],[0,0,0;0,0,0;1,0,1]];
? A = algtableinit(mt);
? algissemisimple(A)
%3 = 0
? m_i=[0,-1,0,0;1,0,0,0;0,0,0,-1;0,0,1,0]; \\ quaternion algebra (-1,-1)
? m_j=[0,0,-1,0;0,0,0,1;1,0,0,0;0,-1,0,0];
? m_k=[0,0,0,-1;0,0,-1,0;0,1,0,0;1,0,0,0];
? mt = [matid(4), m_i, m_j, m_k];
? A = algtableinit(mt);
? algissemisimple(A)
%9 = 1

The library syntax is GEN algissemisimple(GEN al).

3.14.36 algissimple(al , {ss = 0}). al being a table algebra output by algtableinit or a central
simple algebra output by alginit, tests whether the algebra al is simple. If ss = 1, assumes that
the algebra al is semisimple without testing it.

? mt = [matid(3),[0,0,0;1,0,1;0,0,0],[0,0,0;0,0,0;1,0,1]];
? A = algtableinit(mt); \\ matrices [*,*; 0,*]
? algissimple(A)
%3 = 0
? algissimple(A,1) \\ incorrectly assume that A is semisimple
%4 = 1
? m_i=[0,-1,0,0;1,0,0,0;0,0,0,-1;0,0,1,0];
? m_j=[0,0,-1,0;0,0,0,1;1,0,0,0;0,-1,0,0];
? m_k=[0,0,0,-1;0,0,b,0;0,1,0,0;1,0,0,0];
? mt = [matid(4), m_i, m_j, m_k];
? A = algtableinit(mt); \\ quaternion algebra (-1,-1)
? algissimple(A)
%10 = 1
? mt = [matid(3), [0,0,0; 1,1,0; 0,0,0], [0,0,1; 0,0,0; 1,0,1]];
? A = algtableinit(mt,2); \\ direct product F_4 x F_2
? algissimple(A)
%13 = 0

The library syntax is GEN algissimple(GEN al, long ss).

454
3.14.37 algissplit(al , {pl }). Given a central simple algebra al output by alginit, tests whether al
is split, i.e. isomorphic to a matrix algebra over its center. If pl is set, it should be a prime ideal
of K or an integer between 1 and r1 + r2 , and in that case tests whether al is locally split at the
place pl instead.
? nf = nfinit(y^2-5);
? A = alginit(nf, [-1,y]);
? algissplit(A, 1)
%3 = 0
? algissplit(A, 2)
%4 = 1
? algissplit(A, idealprimedec(nf,2)[1])
%5 = 0
? algissplit(A, idealprimedec(nf,5)[1])
%6 = 1
? algissplit(A)
%7 = 0
The library syntax is GEN algissplit(GEN al, GEN pl = NULL).

3.14.38 alglatadd(al , lat1 , lat2 , {&ptinter }). Given an algebra al and two lattices lat1 and lat2
in al , computes the sum lat1 + lat2. If ptinter is present, set it to the intersection lat1 ∩ lat2.
? al = alginit(nfinit(y^2+7), [-1,-1]);
? lat1 = alglathnf(al,[1,1,0,0,0,0,0,0]~);
? lat2 = alglathnf(al,[1,0,1,0,0,0,0,0]~);
? latsum = alglatadd(al,lat1,lat2,&latinter);
? matdet(latsum[1])
%5 = 4
? matdet(latinter[1])
%6 = 64
The library syntax is GEN alglatadd(GEN al, GEN lat1, GEN lat2, GEN *ptinter = NULL)
.

3.14.39 alglatcontains(al , lat, x, {&ptc}). Given an algebra al , a lattice lat and x in al , tests
whether x ∈ lat. If ptc is present, sets it to the t_COL of coordinates of x in the basis of lat.
? al = alginit(nfinit(y^2+7), [-1,-1]);
? a1 = [1,-1,0,1,2,0,1,2]~;
? lat1 = alglathnf(al,a1);
? alglatcontains(al,lat1,a1,&c)
%4 = 1
? c
%5 = [-1, -2, -1, 1, 2, 0, 1, 1]~
The library syntax is GEN alglatcontains(GEN al, GEN lat, GEN x, GEN *ptc = NULL)
.

455
3.14.40 alglatelement(al , lat, c). Given an algebra al , a lattice lat and a t_COL c, returns the
element of al whose coordinates on the Z-basis of lat are given by c.

? al = alginit(nfinit(y^2+7), [-1,-1]);
? a1 = [1,-1,0,1,2,0,1,2]~;
? lat1 = alglathnf(al,a1);
? c = [1..8]~;
? elt = alglatelement(al,lat1,c);
? alglatcontains(al,lat1,elt,&c2)
%6 = 1
? c==c2
%7 = 1

The library syntax is GEN alglatelement(GEN al, GEN lat, GEN c).

3.14.41 alglathnf(al , m, {d = 0}). Given an algebra al and a matrix m with columns representing
elements of al , returns the lattice L generated by the columns of m. If provided, d must be a
rational number such that L contains d times the natural basis of al . The argument m is also
allowed to be a t_VEC of t_MAT, in which case m is replaced by the concatenation of the matrices,
or a t_COL, in which case m is replaced by its left multiplication table as an element of al .

? al = alginit(nfinit(y^2+7), [-1,-1]);
? a = [1,1,-1/2,1,1/3,-1,1,1]~;
? mt = algtomatrix(al,a,1);
? lat = alglathnf(al,mt);
? lat[2]
%5 = 1/6

The library syntax is GEN alglathnf(GEN al, GEN m, GEN d).

3.14.42 alglatindex(al , lat1 , lat2 ). Given an algebra al and two lattices lat1 and lat2 in al ,
computes the generalized index of lat1 relative to lat2 , i.e. |lat2/lat1 ∩ lat2|/|lat1/lat1 ∩ lat2|.

? al = alginit(nfinit(y^2+7), [-1,-1]);
? lat1 = alglathnf(al,[1,1,0,0,0,0,0,0]~);
? lat2 = alglathnf(al,[1,0,1,0,0,0,0,0]~);
? alglatindex(al,lat1,lat2)
%4 = 1
? lat1==lat2
%5 = 0

The library syntax is GEN alglatindex(GEN al, GEN lat1, GEN lat2).

456
3.14.43 alglatinter(al , lat1 , lat2 , {&ptsum}). Given an algebra al and two lattices lat1 and lat2
in al , computes the intersection lat1 ∩ lat2. If ptsum is present, sets it to the sum lat1 + lat2.
? al = alginit(nfinit(y^2+7), [-1,-1]);
? lat1 = alglathnf(al,[1,1,0,0,0,0,0,0]~);
? lat2 = alglathnf(al,[1,0,1,0,0,0,0,0]~);
? latinter = alglatinter(al,lat1,lat2,&latsum);
? matdet(latsum[1])
%5 = 4
? matdet(latinter[1])
%6 = 64
The library syntax is GEN alglatinter(GEN al, GEN lat1, GEN lat2, GEN *ptsum = NULL)
.

3.14.44 alglatlefttransporter(al , lat1 , lat2 ). Given an algebra al and two lattices lat1 and lat2
in al , computes the left transporter from lat1 to lat2 , i.e. the set of x ∈ al such that x · lat1 ⊂ lat2.
? al = alginit(nfinit(y^2+7), [-1,-1]);
? lat1 = alglathnf(al,[1,-1,0,1,2,0,5,2]~);
? lat2 = alglathnf(al,[0,1,-2,-1,0,0,3,1]~);
? tr = alglatlefttransporter(al,lat1,lat2);
? a = alglatelement(al,tr,[0,0,0,0,0,0,1,0]~);
? alglatsubset(al,alglatmul(al,a,lat1),lat2)
%6 = 1
? alglatsubset(al,alglatmul(al,lat1,a),lat2)
%7 = 0
The library syntax is GEN alglatlefttransporter(GEN al, GEN lat1, GEN lat2).

3.14.45 alglatmul(al , lat1 , lat2 ). Given an algebra al and two lattices lat1 and lat2 in al , com-
putes the lattice generated by the products of elements of lat1 and lat2 . One of lat1 and lat2
is also allowed to be an element of al ; in this case, computes the product of the element and the
lattice.
? al = alginit(nfinit(y^2+7), [-1,-1]);
? a1 = [1,-1,0,1,2,0,1,2]~;
? a2 = [0,1,2,-1,0,0,3,1]~;
? lat1 = alglathnf(al,a1);
? lat2 = alglathnf(al,a2);
? lat3 = alglatmul(al,lat1,lat2);
? matdet(lat3[1])
%7 = 29584
? lat3 == alglathnf(al, algmul(al,a1,a2))
%8 = 0
? lat3 == alglatmul(al, lat1, a2)
%9 = 0
? lat3 == alglatmul(al, a1, lat2)
%10 = 0
The library syntax is GEN alglatmul(GEN al, GEN lat1, GEN lat2).

457
3.14.46 alglatrighttransporter(al , lat1 , lat2 ). Given an algebra al and two lattices lat1 and lat2
in al , computes the right transporter from lat1 to lat2 , i.e. the set of x ∈ al such that lat1·x ⊂ lat2.
? al = alginit(nfinit(y^2+7), [-1,-1]);
? lat1 = alglathnf(al,matdiagonal([1,3,7,1,2,8,5,2]));
? lat2 = alglathnf(al,matdiagonal([5,3,8,1,9,8,7,1]));
? tr = alglatrighttransporter(al,lat1,lat2);
? a = alglatelement(al,tr,[0,0,0,0,0,0,0,1]~);
? alglatsubset(al,alglatmul(al,lat1,a),lat2)
%6 = 1
? alglatsubset(al,alglatmul(al,a,lat1),lat2)
%7 = 0
The library syntax is GEN alglatrighttransporter(GEN al, GEN lat1, GEN lat2).

3.14.47 alglatsubset(al , lat1 , lat2 , {&ptindex }). Given an algebra al and two lattices lat1
and lat2 in al , tests whether lat1 ⊂ lat2. If it is true and ptindex is present, sets it to the
index of lat1 in lat2 .
? al = alginit(nfinit(y^2+7), [-1,-1]);
? lat1 = alglathnf(al,[1,1,0,0,0,0,0,0]~);
? lat2 = alglathnf(al,[1,0,1,0,0,0,0,0]~);
? alglatsubset(al,lat1,lat2)
%4 = 0
? latsum = alglatadd(al,lat1,lat2);
? alglatsubset(al,lat1,latsum,&index)
%6 = 1
? index
%7 = 4
The library syntax is GEN alglatsubset(GEN al, GEN lat1, GEN lat2, GEN *ptindex =
NULL).

3.14.48 algmakeintegral(mt, {maps = 0}). mt being a multiplication table over Q in the same
format as the input of algtableinit, computes an integral multiplication table mt2 for an isomor-
phic algebra. When maps = 1, returns a t_VEC [mt2 , S , T ] where S and T are matrices respectively
representing the map from the algebra defined by mt to the one defined by mt2 and its inverse.
? mt = [matid(2),[0,-1/4;1,0]];
? algtableinit(mt);
*** at top-level: algtableinit(mt)
*** ^----------------
*** algtableinit: domain error in algtableinit: denominator(mt) != 1
? mt2 = algmakeintegral(mt);
? al = algtableinit(mt2);
? algisassociative(al)
%4 = 1
? [mt2, S, T] = algmakeintegral(mt,1);
? S
%6 =
[1 0]

458
 [0 1/4]
? T
%7 =
[1 0]
[0 4]
? vector(#mt, i, S * (mt * T[,i]) * T) == mt2
%8 = 1
The library syntax is GEN algmakeintegral(GEN mt, long maps).

3.14.49 algmul(al , x, y). Given two elements x and y in al , computes their product xy in the
algebra al .

? A = alginit(nfinit(y), [-1,-1]);
? algmul(A,[1,1,0,0]~,[0,0,2,1]~)
%2 = [2, 3, 5, -4]~
Also accepts matrices with coefficients in al .
The library syntax is GEN algmul(GEN al, GEN x, GEN y).

3.14.50 algmultable(al ). Returns a multiplication table of al over its prime subfield (Q or

Fp ), as a t_VEC of t_MAT: the left multiplication tables of basis elements. If al was output by
algtableinit, returns the multiplication table used to define al . If al was output by alginit,
returns the multiplication table of the order O0 stored in al .

? A = alginit(nfinit(y), [-1,-1]);
? M = algmultable(A);
? #M
%3 = 4
? M[1] \\ multiplication by e_1 = 1
%4 =
[1 0 0 0]
[0 1 0 0]
[0 0 1 0]
[0 0 0 1]
? M[2]
%5 =
[0 -1 1 0]
[1 0 1 1]
[0 0 1 1]
[0 0 -2 -1]
The library syntax is GEN algmultable(GEN al).

459
3.14.51 algneg(al , x). Given an element x in al , computes its opposite −x in the algebra al .

? A = alginit(nfinit(y), [-1,-1]);
? algneg(A,[1,1,0,0]~)
%2 = [-1, -1, 0, 0]~

Also accepts matrices with coefficients in al .

The library syntax is GEN algneg(GEN al, GEN x).

3.14.52 algnorm(al , x, {abs = 0}). Given an element x in al , computes its norm. If al is a table
algebra output by algtableinit or if abs = 1, returns the absolute norm of x , which is an element
of Fp of Q; if al is a central simple algebra output by alginit and abs = 0 (default), returns the
reduced norm of x , which is an element of the center of al .

? mt = [matid(3), [0,0,0; 1,1,0; 0,0,0], [0,0,1; 0,0,0; 1,0,1]];

? A = algtableinit(mt,19);
? algnorm(A,[0,-2,3]~)
%3 = 18
? nf = nfinit(y^2-5);
? B = alginit(nf,[-1,y]);
? b = [x,1]~;
? n = algnorm(B,b)
%7 = Mod(-y + 1, y^2 - 5)
? algnorm(B,b,1)
%8 = 16
? nfeltnorm(nf,n)^algdegree(B)
%9 = 16

Also accepts a square matrix with coefficients in al .

The library syntax is GEN algnorm(GEN al, GEN x, long abs).

3.14.53 algpoleval(al , T, b). Given an element b in al and a polynomial T in K[X], computes T (b)
in al . Also accepts as input a t_VEC [b, mb] where mb is the left multiplication table of b.

? nf = nfinit(y^2-5);
? al = alginit(nf,[y,-1]);
? b = [1..8]~;
? pol = algcharpoly(al,b,,1);
? algpoleval(al,pol,b)==0
%5 = 1
? mb = algtomatrix(al,b,1);
? algpoleval(al,pol,[b,mb])==0
%7 = 1

The library syntax is GEN algpoleval(GEN al, GEN T, GEN b).

460
3.14.54 algpow(al , x, n). Given an element x in al and an integer n, computes the power xn in
the algebra al .
? A = alginit(nfinit(y), [-1,-1]);
? algpow(A,[1,1,0,0]~,7)
%2 = [8, -8, 0, 0]~
Also accepts a square matrix with coefficients in al .
The library syntax is GEN algpow(GEN al, GEN x, GEN n).

3.14.55 algprimesubalg(al ). al being the output of algtableinit representing a semisimple al-

gebra of positive characteristic, returns a basis of the prime subalgebra of al . The prime subalgebra
of al is the subalgebra fixed by the Frobenius automorphism of the center of al . It is abstractly
isomorphic to a product of copies of Fp .
? mt = [matid(3), [0,0,0; 1,1,0; 0,0,0], [0,0,1; 0,0,0; 1,0,1]];
? A = algtableinit(mt,2);
? algprimesubalg(A)
%3 =
[1 0]
[0 1]
[0 0]
The library syntax is GEN algprimesubalg(GEN al).

3.14.56 algquotient(al , I, {maps = 0}). al being a table algebra output by algtableinit and I
being a basis of a two-sided ideal of al represented by a matrix, returns the quotient al /I . When
maps = 1, returns a t_VEC [al /I , proj , lift] where proj and lift are matrices respectively representing
the projection map and a section of it.
? mt = [matid(3), [0,0,0; 1,1,0; 0,0,0], [0,0,1; 0,0,0; 1,0,1]];
? A = algtableinit(mt,2);
? AQ = algquotient(A,[0;1;0]);
? algdim(AQ)
%4 = 2
The library syntax is GEN alg_quotient(GEN al, GEN I, long maps).

3.14.57 algradical(al ). al being a table algebra output by algtableinit, returns a basis of the
Jacobson radical of the algebra al over its prime field (Q or Fp ).
Here is an example with A = Q[x]/(x2 ), with the basis (1, x):
? mt = [matid(2),[0,0;1,0]];
? A = algtableinit(mt);
? algradical(A) \\ = (x)
%3 =
[0]
[1]
Another one with 2 × 2 upper triangular matrices over Q, with basis I2 , a = [0, 1; 0, 0] and
b = [0, 0; 0, 1], such that a2 = 0, ab = a, ba = 0, b2 = b:

461
 ? mt = [matid(3),[0,0,0;1,0,1;0,0,0],[0,0,0;0,0,0;1,0,1]];
? A = algtableinit(mt);
? algradical(A) \\ = (a)
%6 =
[0]
[1]
[0]
The library syntax is GEN algradical(GEN al).
3.14.58 algramifiedplaces(al ). Given a central simple algebra al output by alginit, returns a
t_VEC containing the list of places of the center of al that are ramified in al . Each place is described
as an integer between 1 and r1 or as a prime ideal.
? nf = nfinit(y^2-5);
? A = alginit(nf, [-1,y]);
? algramifiedplaces(A)
%3 = [1, [2, [2, 0]~, 1, 2, 1]]
The library syntax is GEN algramifiedplaces(GEN al).
3.14.59 algrandom(al , b). Given an algebra al and an integer b, returns a random element in al
with coefficients in [−b, b].
The library syntax is GEN algrandom(GEN al, GEN b).
3.14.60 algrelmultable(al ). Given a central simple algebra al output by alginit defined by a
multiplication table over its center (a number field), returns this multiplication table.
? nf = nfinit(y^3-5); a = y; b = y^2;
? {m_i = [0,a,0,0;
1,0,0,0;
0,0,0,a;
0,0,1,0];}
? {m_j = [0, 0,b, 0;
0, 0,0,-b;
1, 0,0, 0;
0,-1,0, 0];}
? {m_k = [0, 0,0,-a*b;
0, 0,b, 0;
0,-a,0, 0;
1, 0,0, 0];}
? mt = [matid(4), m_i, m_j, m_k];
? A = alginit(nf,mt,’x);
? M = algrelmultable(A);
? M[2] == m_i
%8 = 1
? M[3] == m_j
%9 = 1
? M[4] == m_k
%10 = 1
The library syntax is GEN algrelmultable(GEN al).

462
3.14.61 algsimpledec(al , {maps = 0}). al being the output of algtableinit, returns a t_VEC
[J, [al 1 , al 2 , . . . , al n ]] where J is a basis of the Jacobson radical of al and al /J is isomorphic to
the direct product of the simple algebras al i . When maps = 1, each al i is replaced with a t_VEC
[al i , proj i , lift i ] where proj i and lift i are matrices respectively representing the projection map al →
al i and a section of it. Modulo J, the images of the lift i form a direct sum in al /J, so that the
images of 1 ∈ al i under lift i are central primitive idempotents of al /J. The factors are sorted by
increasing dimension, then increasing dimension of the center. This ensures that the ordering of
the isomorphism classes of the factors is deterministic over finite fields, but not necessarily over Q.
The library syntax is GEN algsimpledec(GEN al, long maps).

3.14.62 algsplit(al , {v =0 x}). If al is a table algebra over Fp output by algtableinit that

represents a simple algebra, computes an isomorphism between al and a matrix algebra Md (Fpn )
where N = nd2 is the dimension of al . Returns a t_VEC [map, mapi], where:
• map is a t_VEC of N matrices of size d × d with t_FFELT coefficients using the variable v ,
representing the image of the basis of al under the isomorphism.
• mapi is an N × N matrix with t_INT coefficients, representing the image in al by the inverse
isomorphism of the basis (bi ) of Md (Fp [α]) (where α has degree n over Fp ) defined as follows:
let Ei,j be the matrix having all coefficients 0 except the (i, j)-th coefficient equal to 1, and define

bi3 +n(i2 +di1 )+1 = Ei1 +1,i2 +1 αi3 ,

where 0 ≤ i1 , i2 < d and 0 ≤ i3 < n.

Example:

? al0 = alginit(nfinit(y^2+7), [-1,-1]);

? al = algtableinit(algmultable(al0), 3); \\ isomorphic to M_2(F_9)
? [map,mapi] = algsplit(al, ’a);
? x = [1,2,1,0,0,0,0,0]~; fx = map*x
%4 =
[2*a 0]
[ 0 2]
? y = [0,0,0,0,1,0,0,1]~; fy = map*y
%5 =
[1 2*a]
[2 a + 2]
? map*algmul(al,x,y) == fx*fy
%6 = 1
? map*mapi[,6]
%7 =
[0 0]
[a 0]

463
Warning. If al is not simple, algsplit(al) can trigger an error, but can also run into an infinite
loop. Example:

? al = alginit(nfinit(y),[-1,-1]); \\ ramified at 2
? al2 = algtableinit(algmultable(al),2); \\ maximal order modulo 2
? algsplit(al2); \\ not semisimple, infinite loop

The library syntax is GEN algsplit(GEN al, long v = -1) where v is a variable number.

3.14.63 algsplittingdata(al ). Given a central simple algebra al output by alginit defined by a

multiplication table over its center K (a number field), returns data stored to compute a splitting
of al over an extension. This data is a t_VEC [t,Lbas,Lbasinv] with 3 components:

• an element t of al such that L = K(t) is a maximal subfield of al ;

• a matrix Lbas expressing a L-basis of al (given an L-vector space structure by multiplication

on the right) on the integral basis of al ;

• a matrix Lbasinv expressing the integral basis of al on the previous L-basis.

? nf = nfinit(y^3-5); a = y; b = y^2;
? {m_i = [0,a,0,0;
1,0,0,0;
0,0,0,a;
0,0,1,0];}
? {m_j = [0, 0,b, 0;
0, 0,0,-b;
1, 0,0, 0;
0,-1,0, 0];}
? {m_k = [0, 0,0,-a*b;
0, 0,b, 0;
0,-a,0, 0;
1, 0,0, 0];}
? mt = [matid(4), m_i, m_j, m_k];
? A = alginit(nf,mt,’x);
? [t,Lb,Lbi] = algsplittingdata(A);
? t
%8 = [0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0]~;
? matsize(Lb)
%9 = [12, 2]
? matsize(Lbi)
%10 = [2, 12]

The library syntax is GEN algsplittingdata(GEN al).

464
3.14.64 algsplittingfield(al ). Given a central simple algebra al output by alginit, returns an
rnf structure: the splitting field of al that is stored in al , as a relative extension of the center.

nf = nfinit(y^3-5);
a = y; b = y^2;
{m_i = [0,a,0,0;
1,0,0,0;
0,0,0,a;
0,0,1,0];}
{m_j = [0, 0,b, 0;
0, 0,0,-b;
1, 0,0, 0;
0,-1,0, 0];}
{m_k = [0, 0,0,-a*b;
0, 0,b, 0;
0,-a,0, 0;
1, 0,0, 0];}
mt = [matid(4), m_i, m_j, m_k];
A = alginit(nf,mt,’x);
algsplittingfield(A).pol
%8 = x^2 - y

The library syntax is GEN algsplittingfield(GEN al).

3.14.65 algsqr(al , x). Given an element x in al , computes its square x2 in the algebra al .

? A = alginit(nfinit(y), [-1,-1]);
? algsqr(A,[1,0,2,0]~)
%2 = [-3, 0, 4, 0]~

Also accepts a square matrix with coefficients in al .

The library syntax is GEN algsqr(GEN al, GEN x).

3.14.66 algsub(al , x, y). Given two elements x and y in al , computes their difference x − y in the
algebra al .

? A = alginit(nfinit(y), [-1,-1]);
? algsub(A,[1,1,0,0]~,[1,0,1,0]~)
%2 = [0, 1, -1, 0]~

Also accepts matrices with coefficients in al .

The library syntax is GEN algsub(GEN al, GEN x, GEN y).

465
3.14.67 algsubalg(al , B). al being a table algebra output by algtableinit and B being a basis
of a subalgebra of al represented by a matrix, computes an algebra al2 isomorphic to B .
Returns [al2 , B2 ] where B2 is a possibly different basis of the subalgebra al2 , with respect to
which the multiplication table of al2 is defined.
? mt = [matid(3), [0,0,0; 1,1,0; 0,0,0], [0,0,1; 0,0,0; 1,0,1]];
? A = algtableinit(mt,2);
? B = algsubalg(A,[1,0; 0,0; 0,1]);
? algdim(A)
%4 = 3
? algdim(B[1])
%5 = 2
? m = matcompanion(x^4+1);
? mt = [m^i | i <- [0..3]];
? al = algtableinit(mt);
? B = [1,0;0,0;0,1/2;0,0];
? al2 = algsubalg(al,B);
? algdim(al2[1])
? al2[2]
%13 =
[1 0]
[0 0]
[0 1]
[0 0]
The library syntax is GEN algsubalg(GEN al, GEN B).

3.14.68 algtableinit(mt, {p = 0}). Initializes the associative algebra over K = Q (p omitted)

or Fp defined by the multiplication table mt. As a K-vector space, the algebra is generated
by a basis (e1 = 1, e2 , . . . , en ); the table is given as a t_VEC of n matrices in Mn (K), giving
the left multiplication by the basis elements ei , in the given basis. Assumes that e1 = 1, that
Ke1 ⊕ . . . ⊕ Ken ] describes an associative algebra over K, and in the case K = Q that the
multiplication table is integral. If the algebra is already known to be central and simple, then the
case K = Fp is useless, and one should use alginit directly.
The point of this function is to input a finite dimensional K-algebra, so as to later compute
its radical, then to split the quotient algebra as a product of simple algebras over K.
The pari object representing such an algebra A is a t_VEC with the following data:
• The characteristic of A, accessed with algchar.
• The multiplication table of A, accessed with algmultable.
• The traces of the elements of the basis.
A simple example: the 2 × 2 upper triangular matrices over Q, generated by I2 , a = [0, 1; 0, 0]
and b = [0, 0; 0, 1], such that a2 = 0, ab = a, ba = 0, b2 = b:
? mt = [matid(3),[0,0,0;1,0,1;0,0,0],[0,0,0;0,0,0;1,0,1]];
? A = algtableinit(mt);
? algradical(A) \\ = (a)

466
 %6 =
[0]
[1]
[0]
? algcenter(A) \\ = (I_2)
%7 =
[1]
[0]
[0]
The library syntax is GEN algtableinit(GEN mt, GEN p = NULL).

3.14.69 algtensor(al1 , al2 , {maxord = 1}). Given two algebras al1 and al2 , computes their tensor
product. Computes a maximal order by default. Prevent this computation by setting maxord = 0.
Currently only implemented for cyclic algebras of coprime degree over the same center K, and
the tensor product is over K.
The library syntax is GEN algtensor(GEN al1, GEN al2, long maxord).

3.14.70 algtomatrix(al , x, {abs = 1}). Given an element x in al , returns the image of x under a
homomorphism to a matrix algebra. If al is a table algebra output by algtableinit or if abs = 1,
returns the left multiplication table on the integral basis; if al is a central simple algebra and abs = 0,
returns φ(x) where φ : A ⊗K L → Md (L) (where d is the degree of the algebra and L is an extension
of L with [L : K] = d) is an isomorphism stored in al . Also accepts a square matrix with coefficients
in al .
? A = alginit(nfinit(y), [-1,-1]);
? algtomatrix(A,[0,0,0,2]~)
%2 =
[Mod(x + 1, x^2 + 1) Mod(Mod(1, y)*x + Mod(-1, y), x^2 + 1)]
[Mod(x + 1, x^2 + 1) Mod(-x + 1, x^2 + 1)]
? algtomatrix(A,[0,1,0,0]~,1)
%2 =
[0 -1 1 0]
[1 0 1 1]
[0 0 1 1]
[0 0 -2 -1]
? algtomatrix(A,[0,x]~,1)
%3 =
[-1 0 0 -1]
[-1 0 1 0]
[-1 -1 0 -1]
[ 2 0 0 1]
Also accepts matrices with coefficients in al .
The library syntax is GEN algtomatrix(GEN al, GEN x, long abs).

467
3.14.71 algtrace(al , x, {abs = 0}). Given an element x in al , computes its trace. If al is a table
algebra output by algtableinit or if abs = 1, returns the absolute trace of x , which is an element
of Fp or Q; if al is the output of alginit and abs = 0 (default), returns the reduced trace of x ,
which is an element of the center of al .
? A = alginit(nfinit(y), [-1,-1]);
? algtrace(A,[5,0,0,1]~)
%2 = 11
? algtrace(A,[5,0,0,1]~,1)
%3 = 22
? nf = nfinit(y^2-5);
? A = alginit(nf,[-1,y]);
? a = [1+x+y,2*y]~*Mod(1,y^2-5)*Mod(1,x^2+1);
? t = algtrace(A,a)
%7 = Mod(2*y + 2, y^2 - 5)
? algtrace(A,a,1)
%8 = 8
? algdegree(A)*nfelttrace(nf,t)
%9 = 8
Also accepts a square matrix with coefficients in al .
The library syntax is GEN algtrace(GEN al, GEN x, long abs).

3.14.72 algtype(al ). Given an algebra al output by alginit or by algtableinit, returns an

integer indicating the type of algebra:
• 0: not a valid algebra.
• 1: table algebra output by algtableinit.
• 2: central simple algebra output by alginit and represented by a multiplication table over
its center.
• 3: central simple algebra output by alginit and represented by a cyclic algebra.
? algtype([])
%1 = 0
? mt = [matid(3), [0,0,0; 1,1,0; 0,0,0], [0,0,1; 0,0,0; 1,0,1]];
? A = algtableinit(mt,2);
? algtype(A)
%4 = 1
? nf = nfinit(y^3-5);
? a = y; b = y^2;
? {m_i = [0,a,0,0;
1,0,0,0;
0,0,0,a;
0,0,1,0];}
? {m_j = [0, 0,b, 0;
0, 0,0,-b;
1, 0,0, 0;
0,-1,0, 0];}
? {m_k = [0, 0,0,-a*b;

468
 0, 0,b, 0;
0,-a,0, 0;
1, 0,0, 0];}
? mt = [matid(4), m_i, m_j, m_k];
? A = alginit(nf,mt,’x);
? algtype(A)
%12 = 2
? A = alginit(nfinit(y), [-1,-1]);
? algtype(A)
%14 = 3
The library syntax is long algtype(GEN al).

3.15 Elliptic curves.

3.15.1 Elliptic curve structures. An elliptic curve is given by a Weierstrass model
y 2 + a1 xy + a3 y = x3 + a2 x2 + a4 x + a6 ,
whose discriminant is nonzero. Affine points on E are represented as two-component vectors [x,y];
the point at infinity, i.e. the identity element of the group law, is represented by the one-component
vector [0].
Given a vector of coefficients [a1 , a2 , a3 , a4 , a6 ], the function ellinit initializes and returns an
ell structure. An additional optional argument allows to specify the base field in case it cannot be
inferred from the curve coefficients. This structure contains data needed by elliptic curve related
functions, and is generally passed as a first argument. Expensive data are skipped on initialization:
they will be dynamically computed when (and if) needed, and then inserted in the structure. The
precise layout of the ell structure is left undefined and should never be used directly. The following
member functions are available, depending on the underlying domain.
All domains.
• a1, a2, a3, a4, a6: coefficients of the elliptic curve.
• b2, b4, b6, b8: b-invariants of the curve; in characteristic 6= 2, for Y = 2y + a1 x + a3, the
curve equation becomes
Y 2 = 4x3 + b2 x2 + 2b4 x + b6 =: g(x).

• c4, c6: c-invariants of the curve; in characteristic 6= 2, 3, for X = x + b2 /12 and Y =

2y + a1 x + a3, the curve equation becomes
Y 2 = 4X 3 − (c4 /12)X − (c6 /216).

• disc: discriminant of the curve. This is only required to be nonzero, not necessarily a unit.
• j: j-invariant of the curve.
These are used as follows:
? E = ellinit([0,0,0, a4,a6]);
? E.b4
%2 = 2*a4
? E.disc
%3 = -64*a4^3 - 432*a6^2

469
Curves over C.
This in particular includes curves defined over Q. All member functions in this section return
data, as it is currently stored in the structure, if present; and otherwise compute it to the default
accuracy, that was fixed at the time of ellinit (via a t_REAL D domain argument, or realprecision
by default). The function ellperiods allows to recompute (and cache) the following data to current
realprecision.
• area: volume of the complex lattice defining E.
• roots is a vector whose three components contain the complex roots of the right hand side
g(x) of the attached b-model Y 2 = g(x). If the roots are all real, they are ordered by decreasing
value. If only one is real, it is the first component.
• omega: [ω1 , ω2 ], periods forming a basis of the complex lattice defining E. The first
component ω1 is the (positive) real period, in other words the integral of the Néron differential
dx/(2y + a1 x + a3 ) over the connected component of the identity component of E(R). The second
ω1
component ω2 is a complex period, such that τ = ω 2
belongs to Poincaré’s half-plane (positive
imaginary part); not necessarily to the standard fundamental domain. It is normalized so that
=(ω2 ) < 0 and either <(ω2 ) = 0, when E.disc > 0 (E(R) has two connected components), or
<(ω2 ) = ω1 /2
• eta is a row vector containing the quasi-periods η1 and η2 such that ηi = 2ζ(ωi /2), where
ζ is the Weierstrass zeta function attached to the period lattice; see ellzeta. In particular, the
Legendre relation holds: η2 ω1 − η1 ω2 = 2πi.

Warning. As for the orientation of the basis of the period lattice, beware that many sources use
the inverse convention where ω2 /ω1 has positive imaginary part and our ω2 is the negative of theirs.
Our convention τ = ω1 /ω2 ensures that the action of PSL2 is the natural one:

[a, b; c, d] · τ = (aτ + b)/(cτ + d) = (aω1 + bω2 )/(cω1 + dω2 ),

instead of a twisted one. (Our τ is −1/τ in the above inverse convention.)

Curves over Qp .
We advise to input a model defined over Q for such curves. In any case, if you input an
approximate model with t_PADIC coefficients, it will be replaced by a lift to Q (an exact model
“close” to the one that was input) and all quantities will then be computed in terms of this lifted
model.
For the time being only curves with multiplicative reduction (split or nonsplit), i.e. vp (j) < 0,
are supported by nontrivial functions. In this case the curve is analytically isomorphic to Q̄∗p /q Z :=
Eq (Q̄p ), for some p-adic integer q (the Tate period). In particular, we have j(q) = j(E).
• p is the residual characteristic
• roots is a vector with a single component, equal to the p-adic root e1 of the right hand side
g(x) of the attached b-model Y 2 = g(x). The point (e1 , 0) corresponds to −1 ∈ Q̄∗p /q Z under the
Tate parametrization.
• tate returns [u2 , u, q, [a, b], Ei, L]
√ in the notation of Henniart-Mestre (CRAS t. 308, p. 391–
395, 1989): q is as above, u ∈ Qp ( −c6 ) is such that φ∗ dx/(2y + a1 x + a3) = udt/t, where
φ : Eq → E is an isomorphism (well defined up to sign) and dt/t is the canonical invariant
differential on the Tate curve; u2 ∈ Qp does not depend on φ. (Technicality: if u 6∈ Qp , it is

470
 p
stored as a quadratic t_POLMOD.) The parameters [a, b] satisfy 4u2 b · agm( a/b, 1)2 = 1 as in
Theorem 2 (loc. cit.). Ei describes the sequence of 2-isogenous curves (with kernel generated by
[0, 0]) Ei : y 2 = x(x + Ai )(x + Ai − Bi ) converging quadratically towards the singular curve E∞ .
Finally, L is Mazur-Tate-Teitelbaum’s L-invariant, equal to logp q/vp (q).

Curves over Fq .

• p is the characteristic of Fq .

• no is #E(Fq ).

• cyc gives the cycle structure of E(Fq ).

• gen returns the generators of E(Fq ).

• group returns [no, cyc, gen], i.e. E(Fq ) as an abelian group structure.

Curves over Q.

All functions should return a correct result, whether the model is minimal or not, but it is a
good idea to stick to minimal models whenever gcd(c4 , c6 ) is easy to factor (minor speed-up). The
construction

E = ellminimalmodel(E0, &v)

replaces the original model E0 by a minimal model E, and the variable change v allows to go
between the two models:

ellchangepoint(P0, v)
ellchangepointinv(P, v)

respectively map the point P0 on E0 to its image on E, and the point P on E to its pre-image on
E0 .

A few routines — namely ellgenerators, ellidentify, ellsearch, forell — require the

optional package elldata (John Cremona’s database) to be installed. In that case, the function
ellinit will allow alternative inputs, e.g. ellinit("11a1"). Functions using this package need to
load chunks of a large database in memory and require at least 2MB stack to avoid stack overflows.

• gen returns the generators of E(Q), if known (from John Cremona’s database)

Curves over number fields.

• nf return the nf structure attached to the number field over which E is defined.

• bnf return the bnf structure attached to the number field over which E is defined or raise
an error (if only an nf is available).

• omega, eta, area: vectors of complex periods, quasi-periods and lattice areas attached to
the complex embeddings of E, in the same order as E.nf.roots.

471
3.15.2 Reduction. Let E be a curve defined over Qp given by a p-integral model; if the curve has
good reduction at p, we may define its reduction Ẽ over the finite field Fp :
? E = ellinit([-3,1], O(5^10)); \\ E/Q5
? Et = ellinit(E, 5)
? ellcard(Et) \\ Ẽ/F5 has 7 points
%3 = 7
? ellinit(E, 7)
*** at top-level: ellinit(E,7)
*** ^------------
*** ellinit: inconsistent moduli in ellinit: 5 != 7
Likewise, if a curve is defined over a number field K and p is a maximal ideal with finite residue
field Fq , we define the reduction Ẽ/Fq provided E has good reduction at p. E/Q is an important
special case:
? E = ellinit([-3,1]);
? factor(E.disc)
%2 =
[2 4]
[3 4]
? Et = ellinit(E, 5);
? ellcard(Et) \\ Ẽ / F5 has 7 points
%4 = 7
? ellinit(E, 3) \\ bad reduction at 3
%5 = []
General number fields are similar:
? K = nfinit(x^2+1); E = ellinit([x,x+1], K);
? idealfactor(K, E.disc) \\ three primes of bad reduction
%2 =
[ [2, [1, 1]~, 2, 1, [1, -1; 1, 1]] 10]
[ [5, [-2, 1]~, 1, 1, [2, -1; 1, 2]] 2]
[[5, [2, 1]~, 1, 1, [-2, -1; 1, -2]] 2]
? P = idealprimedec(K, 3); \\ a prime of good reduction
? idealnorm(K, P)
%4 = 9
? Et = ellinit(E, P);
? ellcard(Et) \\ Ẽ / F9 has 4 points
%6 = 4
If the model is not locally minimal at p, the above will fail: elllocalred and ellchangecurve
allow to reduce to that case.
Some functions such as ellap, ellcard, ellgroup and ellissupersingular even implicitly
replace the given equation by a local minimal model and consider the group of nonsingular points
Ẽ ns so they make sense even when the curve has bad reduction.

472
3.15.3 ellL1(E, {r = 0}). Returns the value at s = 1 of the derivative of order r of the L-function
of the elliptic curve E.
? E = ellinit("11a1"); \\ order of vanishing is 0
? ellL1(E)
%2 = 0.2538418608559106843377589233
? E = ellinit("389a1"); \\ order of vanishing is 2
? ellL1(E)
%4 = -5.384067311837218089235032414 E-29
? ellL1(E, 1)
%5 = 0
? ellL1(E, 2)
%6 = 1.518633000576853540460385214
The main use of this function, after computing at low accuracy the order of vanishing using el-
lanalyticrank, is to compute the leading term at high accuracy to check (or use) the Birch and
Swinnerton-Dyer conjecture:
? \p18
realprecision = 18 significant digits
? E = ellinit("5077a1"); ellanalyticrank(E)
time = 8 ms.
%1 = [3, 10.3910994007158041]
? \p200
realprecision = 202 significant digits (200 digits displayed)
? ellL1(E, 3)
time = 104 ms.
%3 = 10.3910994007158041387518505103609170697263563756570092797[. . .]
The library syntax is GEN ellL1_bitprec(GEN E, long r, long bitprec).

3.15.4 elladd(E, z1 , z2 ). Sum of the points z1 and z2 on the elliptic curve corresponding to E.
The library syntax is GEN elladd(GEN E, GEN z1, GEN z2).

3.15.5 ellak(E, n). Computes the coefficient an of the L-function of the elliptic curve E/Q,
i.e. coefficients of a newform of weight 2 by the modularity theorem (Taniyama-Shimura-Weil
conjecture). E must be an ell structure over Q as output by ellinit. E must be given by an
integral model, not necessarily minimal, although a minimal model will make the function faster.
? E = ellinit([1,-1,0,4,3]);
? ellak(E, 10)
%2 = -3
? e = ellchangecurve(E, [1/5,0,0,0]); \\ made not minimal at 5
? ellak(e, 10) \\ wasteful but works
%3 = -3
? E = ellminimalmodel(e); \\ now minimal
? ellak(E, 5)
%5 = -3
If the model is not minimal at a number of bad primes, then the function will be slower on those
n divisible by the bad primes. The speed should be comparable for other n:

473
 ? for(i=1,10^6, ellak(E,5))
time = 699 ms.
? for(i=1,10^6, ellak(e,5)) \\ 5 is bad, markedly slower
time = 1,079 ms.
? for(i=1,10^5,ellak(E,5*i))
time = 1,477 ms.
? for(i=1,10^5,ellak(e,5*i)) \\ still slower but not so much on average
time = 1,569 ms.
The library syntax is GEN akell(GEN E, GEN n).

3.15.6 ellan(E, n). Computes the vector of the first n Fourier coefficients ak corresponding to the
elliptic curve E defined over a number field. If E is defined over Q, the curve may be given by an
arbitrary model, not necessarily minimal, although a minimal model will make the function faster.
Over a more general number field, the model must be locally minimal at all primes above 2 and 3.
The library syntax is GEN ellan(GEN E, long n). Also available is GEN ellanQ_zv(GEN e,
long n), which returns a t_VECSMALL instead of a t_VEC, saving on memory.

3.15.7 ellanalyticrank(E, {eps}). Returns the order of vanishing at s = 1 of the L-function of

the elliptic curve E and the value of the first nonzero derivative. To determine this order, it is
assumed that any value less than eps is zero. If eps is omitted, 2−b/2 is used, where b is the current
bit precision.

? E = ellinit("11a1"); \\ rank 0
? ellanalyticrank(E)
%2 = [0, 0.2538418608559106843377589233]
? E = ellinit("37a1"); \\ rank 1
? ellanalyticrank(E)
%4 = [1, 0.3059997738340523018204836835]
? E = ellinit("389a1"); \\ rank 2
? ellanalyticrank(E)
%6 = [2, 1.518633000576853540460385214]
? E = ellinit("5077a1"); \\ rank 3
? ellanalyticrank(E)
%8 = [3, 10.39109940071580413875185035]
The library syntax is GEN ellanalyticrank_bitprec(GEN E, GEN eps = NULL, long bit-
prec).

3.15.8 ellap(E, {p}). Let E be an ell structure as output by ellinit, attached to an elliptic
curve E/K. If the field K = Fq is finite, return the trace of Frobenius t, defined by the equation
#E(Fq ) = q + 1 − t.
For other fields of definition and p defining a finite residue field Fq , return the trace of Frobenius
for the reduction of E: the argument p is best left omitted if K = Q` (else we must have p = `)
and must be a prime number (K = Q) or prime ideal (K a general number field) with residue field
Fq otherwise. The equation need not be minimal or even integral at p; of course, a minimal model
will be more efficient.

474
 For a number field K, the trace of Frobenius is the ap coefficient in the Euler product defining
the curve L-series, whence the function name:
Y Y
L(E/K, s) = (1 − ap (N p)−s )−1 (1 − ap (N p)−s + (N p)1−2s )−1 .
bad p good p

When the characteristic of the finite field is large, the availability of the seadata package will
speed up the computation.
? E = ellinit([0,1]); \\ y^2 = x^3 + 0.x + 1, defined over Q
? ellap(E, 7) \\ 7 necessary here
%2 = -4 \\ #E(F_7) = 7+1-(-4) = 12
? ellcard(E, 7)
%3 = 12 \\ OK
? E = ellinit([0,1], 11); \\ defined over F_11
? ellap(E) \\ no need to repeat 11
%4 = 0
? ellap(E, 11) \\ ... but it also works
%5 = 0
? ellgroup(E, 13) \\ ouch, inconsistent input!
*** at top-level: ellap(E,13)
*** ^-----------
*** ellap: inconsistent moduli in Rg_to_Fp:
11
13
? a = ffgen(ffinit(11,3), ’a); \\ defines F_q := F_{11^3}
? E = ellinit([a+1,a]); \\ y^2 = x^3 + (a+1)x + a, defined over F_q
? ellap(E)
%8 = -3
If the curve is defined over a more general number field than Q, the maximal ideal p must be
explicitly given in idealprimedec format. There is no assumption of local minimality at p.
? K = nfinit(a^2+1); E = ellinit([1+a,0,1,0,0], K);
? fa = idealfactor(K, E.disc)
%2 =
[ [5, [-2, 1]~, 1, 1, [2, -1; 1, 2]] 1]
[[13, [5, 1]~, 1, 1, [-5, -1; 1, -5]] 2]
? ellap(E, fa[1,1])
%3 = -1 \\ nonsplit multiplicative reduction
? ellap(E, fa[2,1])
%4 = 1 \\ split multiplicative reduction
? P17 = idealprimedec(K,17)[1];
? ellap(E, P17)
%6 = 6 \\ good reduction
? E2 = ellchangecurve(E, [17,0,0,0]);
? ellap(E2, P17)
%8 = 6 \\ same, starting from a nonmiminal model
? P3 = idealprimedec(K,3)[1];

475
 ? ellap(E, P3) \\ OK: E is minimal at P3
%10 = -2
? E3 = ellchangecurve(E, [3,0,0,0]);
? ellap(E3, P3) \\ not integral at P3
*** at top-level: ellap(E3,P3)
*** ^------------
*** ellap: impossible inverse in Rg_to_ff: Mod(0, 3).

Algorithms used. If E/Fq has CM by a principal imaginary quadratic order we use a fast
explicit formula (involving essentially Kronecker symbols and Cornacchia’s algorithm), in O(log q)2
bit operations. Otherwise, we use Shanks-Mestre’s baby-step/giant-step method, which runs in time
Õ(q 1/4 ) using Õ(q 1/4 ) storage, hence becomes unreasonable when q has about 30 digits. Above
this range, the SEA algorithm becomes available, heuristically in Õ(log q)4 , and primes of the order
of 200 digits become feasible. In small characteristic we use Mestre’s (p=2), Kohel’s (p=3,5,7,13),
Satoh-Harley (all in Õ(p2 n2 )) or Kedlaya’s (in Õ(pn3 )) algorithms.

The library syntax is GEN ellap(GEN E, GEN p = NULL).

3.15.9 ellbil(E, z1 , z2 ). Deprecated alias for ellheight(E,P,Q).

The library syntax is GEN bilhell(GEN E, GEN z1, GEN z2, long prec).

3.15.10 ellbsd(E). The object E being an elliptic curve over a number field, returns a real number
(r)
c such that the BSD conjecture predicts that LE (1)/r! = cRS where r is the rank, R the regulator
and S the cardinal of the Tate-Shafarevich group.

? e = ellinit([0,-1,1,-10,-20]); \\ rank 0
? ellbsd(e)
%2 = 0.25384186085591068433775892335090946105
? lfun(e,1)
%3 = 0.25384186085591068433775892335090946104
? e = ellinit([0,0,1,-1,0]); \\ rank 1
? P = ellheegner(e);
? ellbsd(e)*ellheight(e,P)
%6 = 0.30599977383405230182048368332167647445
? lfun(e,1,1)
%7 = 0.30599977383405230182048368332167647445
? e = ellinit([1+a,0,1,0,0],nfinit(a^2+1)); \\ rank 0
? ellbsd(e)
%9 = 0.42521832235345764503001271536611593310
? lfun(e,1)
%10 = 0.42521832235345764503001271536611593309

The library syntax is GEN ellbsd(GEN E, long prec).

476
3.15.11 ellcard(E, {p}). Let E be an ell structure as output by ellinit, attached to an elliptic
curve E/K. If K = Fq is finite, return the order of the group E(Fq ).
? E = ellinit([-3,1], 5); ellcard(E)
%1 = 7
? t = ffgen(3^5,’t); E = ellinit([t,t^2+1]); ellcard(E)
%2 = 217
For other fields of definition and p defining a finite residue field Fq , return the order of the reduction
of E: the argument p is best left omitted if K = Q` (else we must have p = `) and must be a
prime number (K = Q) or prime ideal (K a general number field) with residue field Fq otherwise.
The equation need not be minimal or even integral at p; of course, a minimal model will be more
efficient. The function considers the group of nonsingular points of the reduction of a minimal
model of the curve at p, so also makes sense when the curve has bad reduction.
? E = ellinit([-3,1]);
? factor(E.disc)
%2 =
[2 4]
[3 4]
? ellcard(E, 5) \\ as above !
%3 = 7
? ellcard(E, 2) \\ additive reduction
%4 = 2
When the characteristic of the finite field is large, the availability of the seadata package will
speed the computation. See also ellap for the list of implemented algorithms.
The library syntax is GEN ellcard(GEN E, GEN p = NULL). Also available is GEN ell-
card(GEN E, GEN p) where p is not NULL.

3.15.12 ellchangecurve(E, v). Changes the data for the elliptic curve E by changing the coor-
dinates using the vector v=[u,r,s,t], i.e. if x0 and y 0 are the new coordinates, then x = u2 x0 + r,
y = u3 y 0 + su2 x0 + t. E must be an ell structure as output by ellinit. The special case v = 1 is
also used instead of [1, 0, 0, 0] to denote the trivial coordinate change.
The library syntax is GEN ellchangecurve(GEN E, GEN v).

3.15.13 ellchangepoint(x, v). Changes the coordinates of the point or vector of points x using the
vector v=[u,r,s,t], i.e. if x0 and y 0 are the new coordinates, then x = u2 x0 +r, y = u3 y 0 +su2 x0 +t
(see also ellchangecurve).
? E0 = ellinit([1,1]); P0 = [0,1]; v = [1,2,3,4];
? E = ellchangecurve(E0, v);
? P = ellchangepoint(P0,v)
%3 = [-2, 3]
? ellisoncurve(E, P)
%4 = 1
? ellchangepointinv(P,v)
%5 = [0, 1]
The library syntax is GEN ellchangepoint(GEN x, GEN v). The reciprocal function GEN
ellchangepointinv(GEN x, GEN ch) inverts the coordinate change.

477
3.15.14 ellchangepointinv(x, v). Changes the coordinates of the point or vector of points x using
the inverse of the isomorphism attached to v=[u,r,s,t], i.e. if x0 and y 0 are the old coordinates,
then x = u2 x0 + r, y = u3 y 0 + su2 x0 + t (inverse of ellchangepoint).

? E0 = ellinit([1,1]); P0 = [0,1]; v = [1,2,3,4];

? E = ellchangecurve(E0, v);
? P = ellchangepoint(P0,v)
%3 = [-2, 3]
? ellisoncurve(E, P)
%4 = 1
? ellchangepointinv(P,v)
%5 = [0, 1] \\ we get back P0

The library syntax is GEN ellchangepointinv(GEN x, GEN v).

3.15.15 ellconvertname(name). Converts an elliptic curve name, as found in the elldata

database, from a string to a triplet [conductor , isogeny class, index ]. It will also convert a triplet
back to a curve name. Examples:

? ellconvertname("123b1")
%1 = [123, 1, 1]
? ellconvertname(%)
%2 = "123b1"

The library syntax is GEN ellconvertname(GEN name).

3.15.16 elldivpol(E, n, {v =0 x}). n-division polynomial fn for the curve E in the variable v. In
standard notation, for any affine point P = (X, Y ) on the curve and any integer n ≥ 0, we have

[n]P = (φn (P )ψn (P ) : ωn (P ) : ψn (P )3 )

for some polynomials φn , ωn , ψn in Z[a1 , a2 , a3 , a4 , a6 ][X, Y ]. We have fn (X) = ψn (X) for n odd,
and fn (X) = ψn (X, Y )(2Y + a1 X + a3 ) for n even. We have

f0 = 0, f1 = 1, f2 = 4X 3 + b2 X 2 + 2b4 X + b6 , f3 = 3X 4 + b2 X 3 + 3b4 X 2 + 3b6 X + b8,

f4 = f2 (2X 6 + b2 X 5 + 5b4 X 4 + 10b6 X 3 + 10b8 X 2 + (b2 b8 − b4 b6 )X + (b8 b4 − b26 )), . . .

When n is odd, the roots of fn are the X-coordinates of the affine points in the n-torsion subgroup
E[n]; when n is even, the roots of fn are the X-coordinates of the affine points in E[n] \ E[2] when
n > 2, resp. in E[2] when n = 2. For n < 0, we define fn := −f−n .

The library syntax is GEN elldivpol(GEN E, long n, long v = -1) where v is a variable
number.

478
3.15.17 elleisnum(w, k, {flag = 0}). k being an even positive integer, computes the numerical
value of the Eisenstein series of weight k at the lattice w, as given by ellperiods, namely
 X 
(2iπ/ω2 )k 1 + 2/ζ(1 − k) nk−1 q n /(1 − q n ) ,
n≥1

where q = exp(2iπτ ) and τ := ω1 /ω2 belongs to the complex upper half-plane. It is also possible
to directly input w = [ω1 , ω2 ], or an elliptic curve E as given by ellinit.
? w = ellperiods([1,I]);
? elleisnum(w, 4)
%2 = 2268.8726415508062275167367584190557607
? elleisnum(w, 6)
%3 = -3.977978632282564763 E-33
? E = ellinit([1, 0]);
? elleisnum(E, 4)
%5 = -48.000000000000000000000000000000000000
When flag is nonzero and k = 4 or 6, returns the elliptic invariants g2 or g3 , such that
y 2 = 4x3 − g2 x − g3
is a Weierstrass equation for E.
? g2 = elleisnum(E, 4, 1)
%6 = -4.0000000000000000000000000000000000000
? g3 = elleisnum(E, 6, 1) \\ ~ 0
%7 = 0.E-114 - 3.909948178422242682 E-57*I
The library syntax is GEN elleisnum(GEN w, long k, long flag, long prec).

3.15.18 elleta(w). Returns the quasi-periods [η1 , η2 ] attached to the lattice basis w = [ω1 , ω2 ].
Alternatively, w can be an elliptic curve E as output by ellinit, in which case, the quasi periods
attached to the period lattice basis E.omega (namely, E.eta) are returned.
? elleta([1, I])
%1 = [3.141592653589793238462643383, 9.424777960769379715387930149*I]
The library syntax is GEN elleta(GEN w, long prec).

3.15.19 ellformaldifferential(E, {n = seriesprecision}, {t =0 x}). Let ω := dx/(2y + a1 x + a3 )

be the invariant differential form attached to the model E of some elliptic curve (ellinit form),
and η := x(t)ω. Return n terms (seriesprecision by default) of f (t), g(t) two power series in the
formal parameter t = −x/y such that ω = f (t)dt, η = g(t)dt:
f (t) = 1 + a1 t + (a21 + a2 )t2 + . . . , g(t) = t−2 + . . .

? E = ellinit([-1,1/4]); [f,g] = ellformaldifferential(E,7,’t);

? f
%2 = 1 - 2*t^4 + 3/4*t^6 + O(t^7)
? g
%3 = t^-2 - t^2 + 1/2*t^4 + O(t^5)
The library syntax is GEN ellformaldifferential(GEN E, long precdl, long n = -1)
where n is a variable number.

479
3.15.20 ellformalexp(E, {n = seriesprecision}, {z =0 x}). The elliptic formal exponential Exp
attached to E is the isomorphism from the formal additive law to the formal group of E. It is
normalized so as to be the inverse of the elliptic logarithm (see ellformallog): Exp ◦ L = Id.
Return n terms of this power series:

? E=ellinit([-1,1/4]); Exp = ellformalexp(E,10,’z)

%1 = z + 2/5*z^5 - 3/28*z^7 + 2/15*z^9 + O(z^11)
? L = ellformallog(E,10,’t);
? subst(Exp,z,L)
%3 = t + O(t^11)

The library syntax is GEN ellformalexp(GEN E, long precdl, long n = -1) where n is a
variable number.

3.15.21 ellformallog(E, {n = seriesprecision}, {v =0 x}). The formal elliptic logarithm is a series

L in tK[[t]] such that dL = ω = dx/(2y + a1 x + a3 ), the canonical invariant differential attached
to the model E. It gives an isomorphism from the formal group of E to the additive formal group.

? E = ellinit([-1,1/4]); L = ellformallog(E, 9, ’t)

%1 = t - 2/5*t^5 + 3/28*t^7 + 2/3*t^9 + O(t^10)
? [f,g] = ellformaldifferential(E,8,’t);
? L’ - f
%3 = O(t^8)

The library syntax is GEN ellformallog(GEN E, long precdl, long n = -1) where n is a
variable number.

3.15.22 ellformalpoint(E, {n = seriesprecision}, {v =0 x}). If E is an elliptic curve, return the

coordinates x(t), y(t) in the formal group of the elliptic curve E in the formal parameter t = −x/y
at ∞:
x = t−2 − a1 t−1 − a2 − a3 t + . . .

y = −t−3 − a1 t−2 − a2 t−1 − a3 + . . .

Return n terms (seriesprecision by default) of these two power series, whose coefficients are in
Z[a1 , a2 , a3 , a4 , a6 ].

? E = ellinit([0,0,1,-1,0]); [x,y] = ellformalpoint(E,8,’t);

? x
%2 = t^-2 - t + t^2 - t^4 + 2*t^5 + O(t^6)
? y
%3 = -t^-3 + 1 - t + t^3 - 2*t^4 + O(t^5)
? E = ellinit([0,1/2]); ellformalpoint(E,7)
%4 = [x^-2 - 1/2*x^4 + O(x^5), -x^-3 + 1/2*x^3 + O(x^4)]

The library syntax is GEN ellformalpoint(GEN E, long precdl, long n = -1) where n is
a variable number.

480
3.15.23 ellformalw(E, {n = seriesprecision}, {t =0 x}). Return the formal power series w at-
tached to the elliptic curve E, in the variable t:
w(t) = t3 (1 + a1 t + (a2 + a21 )t2 + · · · + O(tn )),
which is the formal expansion of −1/y in the formal parameter t := −x/y at ∞ (take n =
seriesprecision if n is omitted). The coefficients of w belong to Z[a1 , a2 , a3 , a4 , a6 ].
? E=ellinit([3,2,-4,-2,5]); ellformalw(E, 5, ’t)
%1 = t^3 + 3*t^4 + 11*t^5 + 35*t^6 + 101*t^7 + O(t^8)
The library syntax is GEN ellformalw(GEN E, long precdl, long n = -1) where n is a
variable number.

3.15.24 ellfromeqn(P ). Given a genus 1 plane curve, defined by the affine equation f (x, y) = 0,
return the coefficients [a1 , a2 , a3 , a4 , a6 ] of a Weierstrass equation for its Jacobian. This allows to
recover a Weierstrass model for an elliptic curve given by a general plane cubic or by a binary
quartic or biquadratic model. The function implements the f 7→ f ∗ formulae of Artin, Tate and
Villegas (Advances in Math. 198 (2005), pp. 366–382).
In the example below, the function is used to convert between twisted Edwards coordinates
and Weierstrass coordinates.
? e = ellfromeqn(a*x^2+y^2 - (1+d*x^2*y^2))
%1 = [0, -a - d, 0, -4*d*a, 4*d*a^2 + 4*d^2*a]
? E = ellinit(ellfromeqn(y^2-x^2 - 1 +(121665/121666*x^2*y^2)),2^255-19);
? isprime(ellcard(E) / 8)
%3 = 1
The elliptic curve attached to the sum of two cubes is given by
? ellfromeqn(x^3+y^3 - a)
%1 = [0, 0, -9*a, 0, -27*a^2]
Congruent number problem. Let n be an integer, if a2 + b2 = c2 and ab = 2n, then by
substituting b by 2n/a in the first equation, we get ((a2 + (2n/a)2 ) − c2 )a2 = 0. We set x = a,
y = ac.
? En = ellfromeqn((x^2 + (2*n/x)^2 - (y/x)^2)*x^2)
%1 = [0, 0, 0, -16*n^2, 0]
For example 23 is congruent since the curve has a point of infinite order, namely:
? ellheegner( ellinit(subst(En, n, 23)) )
%2 = [168100/289, 68053440/4913]
The library syntax is GEN ellfromeqn(GEN P).

3.15.25 ellfromj(j). Returns the coefficients [a1 , a2 , a3 , a4 , a6 ] of a fixed elliptic curve with j-
invariant j.
The library syntax is GEN ellfromj(GEN j).

3.15.26 ellgenerators(E). If E is an elliptic curve over the rationals, return a Z-basis of the free
part of the Mordell-Weil group attached to E. This relies on the elldata database being installed
and referencing the curve, and so is only available for curves over Z of small conductors. If E is
an elliptic curve over a finite field Fq as output by ellinit, return a minimal set of generators for
the group E(Fq ).

481
Caution. When the group is not cyclic, of shape Z/d1 Z × Z/d2 Z with d2 | d1 , the points [P, Q]
returned by ellgenerators need not have order d1 and d2 : it is true that P has order d1 , but we
only know that Q is a generator of E(Fq )/ < P > and that the Weil pairing w(P, Q) has order d2 ,
see ??ellgroup. If you need generators [P, R] with R of order d2 , find x such that R = Q − [x]P
has order d2 by solving the discrete logarithm problem [d2 ]Q = [x]([d2 ]P ) in a cyclic group of order
d1 /d2 . This will be very expensive if d1 /d2 has a large prime factor.

The library syntax is GEN ellgenerators(GEN E).

3.15.27 ellglobalred(E). Let E be an ell structure as output by ellinit attached to an elliptic

curve defined over a number field. This function calculates the arithmetic conductor and the global
Tamagawa number c. The result [N, v, c, F, L] is slightly different if E is defined over Q (domain
D = 1 in ellinit) or over a number field (domain D is a number field structure, including
nfinit(x) representing Q !):

• N is the arithmetic conductor of the curve,

• v is an obsolete field, left in place for backward compatibility. If E is defined over Q, v gives
the coordinate change for E to the standard minimal integral model (ellminimalmodel provides
it in a cheaper way); if E is defined over another number field, v gives a coordinate change to an
integral model (ellintegralmodel provides it in a cheaper way).

• c is the product of the local Tamagawa numbers cp , a quantity which enters in the Birch and
Swinnerton-Dyer conjecture,

• F is the factorization of N ,

• L is a vector, whose i-th entry contains the local data at the i-th prime ideal divisor of N ,
i.e. L[i] = elllocalred(E,F[i,1]). If E is defined over Q, the local coordinate change has been
deleted and replaced by a 0; if E is defined over another number field the local coordinate change
to a local minimal model is given relative to the integral model afforded by v (so either start from
an integral model so that v be trivial, or apply v first).

The library syntax is GEN ellglobalred(GEN E).

3.15.28 ellgroup(E, {p}, {flag}). Let E be an ell structure as output by ellinit, attached to
an elliptic curve E/K. We first describle the function when the field K = Fq is finite, it computes
the structure of the finite abelian group E(Fq ):

• if flag = 0, return the structure [] (trivial group) or [d1 ] (nontrivial cyclic group) or [d1 , d2 ]
(noncyclic group) of E(Fq ) ∼ Z/d1 Z × Z/d2 Z, with d2 | d1 .

• if flag = 1, return a triple [h, cyc, gen], where h is the curve cardinality, cyc gives the group
structure as a product of cyclic groups (as per flag = 0). More precisely, if d2 > 1, the output is
[d1 d2 , [d1 , d2 ], [P, Q]] where P is of order d1 and [P, Q] generates the curve.

482
Caution. It is not guaranteed that Q has order d2 , which in the worst case requires an expensive
discrete log computation. Only that ellweilpairing(E, P, Q, d1 ) has order d2 .
For other fields of definition and p defining a finite residue field Fq , return the structure of
the reduction of E: the argument p is best left omitted if K = Q` (else we must have p = `) and
must be a prime number (K = Q) or prime ideal (K a general number field) with residue field Fq
otherwise. The curve is allowed to have bad reduction at p and in this case we consider the (cyclic)
group of nonsingular points for the reduction of a minimal model at p.
If flag = 0, the equation not be minimal or even integral at p; of course, a minimal model will
be more efficient.
If flag = 1, the requested generators depend on the model, which must then be minimal at p,
otherwise an exception is thrown. Use ellintegralmodel and/or ellocalred first to reduce to
this case.
? E = ellinit([0,1]); \\ y^2 = x^3 + 0.x + 1, defined over Q
? ellgroup(E, 7)
%2 = [6, 2] \\ Z/6 x Z/2, noncyclic
? E = ellinit([0,1] * Mod(1,11)); \\ defined over F_11
? ellgroup(E) \\ no need to repeat 11
%4 = [12]
? ellgroup(E, 11) \\ ... but it also works
%5 = [12]
? ellgroup(E, 13) \\ ouch, inconsistent input!
*** at top-level: ellgroup(E,13)
*** ^--------------
*** ellgroup: inconsistent moduli in Rg_to_Fp:
11
13
? ellgroup(E, 7, 1)
%6 = [12, [6, 2], [[Mod(2, 7), Mod(4, 7)], [Mod(4, 7), Mod(4, 7)]]]
Let us now consider curves of bad reduction, in this case we return the structure of the (cyclic)
group of nonsingular points, satisfying #Ens (Fp ) = p − ap :
? E = ellinit([0,5]);
? ellgroup(E, 5, 1)
%2 = [5, [5], [[Mod(4, 5), Mod(2, 5)]]]
? ellap(E, 5)
%3 = 0 \\ additive reduction at 5
? E = ellinit([0,-1,0,35,0]);
? ellgroup(E, 5, 1)
%5 = [4, [4], [[Mod(2, 5), Mod(2, 5)]]]
? ellap(E, 5)
%6 = 1 \\ split multiplicative reduction at 5
? ellgroup(E, 7, 1)
%7 = [8, [8], [[Mod(3, 7), Mod(5, 7)]]]
? ellap(E, 7)
%8 = -1 \\ nonsplit multiplicative reduction at 7
The library syntax is GEN ellgroup0(GEN E, GEN p = NULL, long flag). Also available is
GEN ellgroup(GEN E, GEN p), corresponding to flag= 0.

483
3.15.29 ellheegner(E). Let E be an elliptic curve over the rationals, assumed to be of (analytic)
rank 1. This returns a nontorsion rational point on the curve, whose canonical height is equal to
the product of the elliptic regulator by the analytic Sha.
This uses the Heegner point method, described in Cohen GTM 239; the complexity is propor-
tional to the product of the square root of the conductor and the height of the point (thus, it is
preferable to apply it to strong Weil curves).
? E = ellinit([-157^2,0]);
? u = ellheegner(E); print(u[1], "\n", u[2])
69648970982596494254458225/166136231668185267540804
538962435089604615078004307258785218335/67716816556077455999228495435742408
? ellheegner(ellinit([0,1])) \\ E has rank 0 !
*** at top-level: ellheegner(E=ellinit
*** ^--------------------
*** ellheegner: The curve has even analytic rank.
The library syntax is GEN ellheegner(GEN E).

3.15.30 ellheight(E, {P }, {Q}). Let E be an elliptic curve defined over K = Q or a number field,
as output by ellinit; it needs not be given by a minimal model although the computation will be
faster if it is.
• Without arguments P, Q, returns the Faltings height of the curve E using Deligne
normalization. For a rational curve, the normalization is such that the function returns -
(1/2)*log(ellminimalmodel(E).area).
• If the argument P ∈ E(K) is present, returns the global Néron-Tate height h(P ) of the point,
using the normalization in Cremona’s Algorithms for modular elliptic curves.
• If the argument Q ∈ E(K) is also present, computes the value of the bilinear form (h(P +
Q) − h(P − Q))/4.
The library syntax is GEN ellheight0(GEN E, GEN P = NULL, GEN Q = NULL, long prec)
. Also available is GEN ellheight(GEN E, GEN P, long prec) (Q omitted).

3.15.31 ellheightmatrix(E, x). x being a vector of points, this function outputs the Gram matrix
of x with respect to the Néron-Tate height, in other words, the (i, j) component of the matrix is equal
to ellbil(E,x[i],x[j]). The rank of this matrix, at least in some approximate sense, gives the
rank of the set of points, and if x is a basis of the Mordell-Weil group of E, its determinant is equal
to the regulator of E. Note our height normalization follows Cremona’s Algorithms for modular
elliptic curves: this matrix should be divided by 2 to be in accordance with, e.g., Silverman’s
normalizations.
The library syntax is GEN ellheightmatrix(GEN E, GEN x, long prec).

3.15.32 ellidentify(E). Look up the elliptic curve E, defined by an arbitrary model over Q, in
the elldata database. Return [[N, M, G], C] where N is the curve name in Cremona’s elliptic
curve database, M is the minimal model, G is a Z-basis of the free part of the Mordell-Weil group
E(Q) and C is the change of coordinates from E to M , suitable for ellchangecurve.
The library syntax is GEN ellidentify(GEN E).

484
3.15.33 ellinit(x, {D = 1}). Initialize an ell structure, attached to the elliptic curve E. E is
either
• a 5-component vector [a1 , a2 , a3 , a4 , a6 ] defining the elliptic curve with Weierstrass equation

Y 2 + a1 XY + a3 Y = X 3 + a2 X 2 + a4 X + a6 ,

• a 2-component vector [a4 , a6 ] defining the elliptic curve with short Weierstrass equation

Y 2 = X 3 + a4 X + a6 ,

• a character string in Cremona’s notation, e.g. "11a1", in which case the curve is retrieved
from the elldata database if available.
The optional argument D describes the domain over which the curve is defined:
• the t_INT 1 (default): the field of rational numbers Q.
• a t_INT p, where p is a prime number: the prime finite field Fp .
• an t_INTMOD Mod(a, p), where p is a prime number: the prime finite field Fp .
• a t_FFELT, as returned by ffgen: the corresponding finite field Fq .
• a t_PADIC, O(pn ): the field Qp , where p-adic quantities will be computed to a relative
accuracy of n digits. We advise to input a model defined over Q for such curves. In any case, if
you input an approximate model with t_PADIC coefficients, it will be replaced by a lift to Q (an
exact model “close” to the one that was input) and all quantities will then be computed in terms
of this lifted model, at the given accuracy.
• a t_REAL x: the field C of complex numbers, where floating point quantities are by default
computed to a relative accuracy of precision(x). If no such argument is given, the value of
realprecision at the time ellinit is called will be used.
• a number field K, given by a nf or bnf structure; a bnf is required for ellminimalmodel.
• a prime ideal p, given by a prid structure; valid if x is a curve defined over a number field
K and the equation is integral and minimal at p.
This argument D is indicative: the curve coefficients are checked for compatibility, possibly
changing D; for instance if D = 1 and an t_INTMOD is found. If inconsistencies are detected, an
error is raised:
? ellinit([1 + O(5), 1], O(7));
*** at top-level: ellinit([1+O(5),1],O
*** ^--------------------
*** ellinit: inconsistent moduli in ellinit: 7 != 5
If the curve coefficients are too general to fit any of the above domain categories, only basic
operations, such as point addition, will be supported later.
If the curve (seen over the domain D) is singular, fail and return an empty vector [].
? E = ellinit([0,0,0,0,1]); \\ y^2 = x^3 + 1, over Q
? E = ellinit([0,1]); \\ the same curve, short form
? E = ellinit("36a1"); \\ sill the same curve, Cremona’s notations

485
 ? E = ellinit([0,1], 2) \\ over F2: singular curve
%4 = []
? E = ellinit([’a4,’a6] * Mod(1,5)); \\ over F_5[a4,a6], basic support !

The result of ellinit is an ell structure. It contains at least the following information in its
components:
a1 , a2 , a3 , a4 , a6 , b2 , b4 , b6 , b8 , c4 , c6 , ∆, j.

All are accessible via member functions. In particular, the discriminant is E.disc, and the j-
invariant is E.j.

? E = ellinit([a4, a6]);
? E.disc
%2 = -64*a4^3 - 432*a6^2
? E.j
%3 = -6912*a4^3/(-4*a4^3 - 27*a6^2)
Further components contain domain-specific data, which are in general dynamic: only com-
puted when needed, and then cached in the structure.

? E = ellinit([2,3], 10^60+7); \\ E over F_p, p large

? ellap(E)
time = 4,440 ms.
%2 = -1376268269510579884904540406082
? ellcard(E); \\ now instantaneous !
time = 0 ms.
? ellgenerators(E);
time = 5,965 ms.
? ellgenerators(E); \\ second time instantaneous
time = 0 ms.
See the description of member functions related to elliptic curves at the beginning of this
section.
The library syntax is GEN ellinit(GEN x, GEN D = NULL, long prec).

3.15.34 ellintegralmodel(E, {&v}). Let E be an ell structure over a number field K or Qp .

This function returns an integral model. If v is present, sets v = [u, 0, 0, 0] to the corresponding
change of variable: the return value is identical to that of ellchangecurve(E, v).

? e = ellinit([1/17,1/42]);
? e = ellintegralmodel(e,&v);
? e[1..5]
%3 = [0, 0, 0, 15287762448, 3154568630095008]
? v
%4 = [1/714, 0, 0, 0]
The library syntax is GEN ellintegralmodel(GEN E, GEN *v = NULL).

486
3.15.35 ellisdivisible(E, P, n, {&Q}). Given E/K a number field and P in E(K) return 1 if
P = [n]R for some R in E(K) and set Q to one such R; and return 0 otherwise. The integer n ≥ 0
may be given as ellxn(E,n), if many points need to be tested.
? K = nfinit(polcyclo(11,t));
? E = ellinit([0,-1,1,0,0], K);
? P = [0,0];
? ellorder(E,P)
%4 = 5
? ellisdivisible(E,P,5, &Q)
%5 = 1
? lift(Q)
%6 = [-t^7-t^6-t^5-t^4+1, -t^9-2*t^8-2*t^7-3*t^6-3*t^5-2*t^4-2*t^3-t^2-1]
? ellorder(E, Q)
%7 = 25
The algebraic complexity of the underlying algorithm is in O(n4 ), so it is advisable to first factor
n, then use a chain of checks attached to the prime divisors of n: the function will do it itself unless
n is given in ellxn form.
The library syntax is long ellisdivisible(GEN E, GEN P, GEN n, GEN *Q = NULL).

3.15.36 ellisogeny(E, G, {only image = 0}, {x =0 x}, {y =0 y}). Given an elliptic curve E, a finite
subgroup G of E is given either as a generating point P (for a cyclic G) or as a polynomial whose
roots vanish on the x-coordinates of the nonzero elements of G (general case and more efficient if
available). This function returns the [a1 , a2 , a3 , a4 , a6 ] invariants of the quotient elliptic curve E/G
and (if only image is zero (the default)) a vector of rational functions [f, g, h] such that the isogeny
E → E/G is given by (x, y) 7→ (f (x)/h(x)2 , g(x, y)/h(x)3 ).
? E = ellinit([0,1]);
? elltors(E)
%2 = [6, [6], [[2, 3]]]
? ellisogeny(E, [2,3], 1) \\ Weierstrass model for E/<P>
%3 = [0, 0, 0, -135, -594]
? ellisogeny(E,[-1,0])
%4 = [[0,0,0,-15,22], [x^3+2*x^2+4*x+3, y*x^3+3*y*x^2-2*y, x+1]]
The library syntax is GEN ellisogeny(GEN E, GEN G, long only_image, long x = -1,
long y = -1) where x, y are variable numbers.

3.15.37 ellisogenyapply(f, g). Given an isogeny of elliptic curves f : E 0 → E (being the result
of a call to ellisogeny), apply f to g:
• if g is a point P in the domain of f , return the image f (P );
• if g : E 00 → E 0 is a compatible isogeny, return the composite isogeny f ◦ g : E 00 → E.
? one = ffgen(101, ’t)^0;
? E = ellinit([6, 53, 85, 32, 34] * one);
? P = [84, 71] * one;
? ellorder(E, P)
%4 = 5
? [F, f] = ellisogeny(E, P); \\ f: E->F = E/<P>

487
 ? ellisogenyapply(f, P)
%6 = [0]
? F = ellinit(F);
? Q = [89, 44] * one;
? ellorder(F, Q)
%9 = 2
? [G, g] = ellisogeny(F, Q); \\ g: F->G = F/<Q>
? gof = ellisogenyapply(g, f); \\ gof: E -> G
The library syntax is GEN ellisogenyapply(GEN f, GEN g).

3.15.38 ellisomat(E, {p = 0}, {fl = 0}). Given an elliptic curve E defined over a number field
K, compute representatives of the isomorphism classes of elliptic curves defined over K and K-
isogenous to E. We assume that E does not have CM over K (otherwise that set would be infinite).
For any such curve Ei , let fi : E → Ei be a rational isogeny of minimal degree and let gi : Ei → E
be the dual isogeny; and let M be the matrix such that Mi,j is the minimal degree for an isogeny
Ei → Ej .
The function returns a vector [L, M ] where L is a list of triples [Ei , fi , gi ] (flag = 0), or simply
the list of Ei (flag = 1, which saves time). The curves Ei are given in [a4 , a6 ] form and the first
curve E1 is isomorphic to E by f1 .
If p is set, it must be a prime number; in this which case only isogenies of degree a power of p
are considered.
Over a number field, the possible isogeny degrees are determined by Billerey algorithm.
? E = ellinit("14a1");
? [L,M] = ellisomat(E);
? LE = apply(x->x[1], L) \\ list of curves
%3 = [[215/48,-5291/864],[-675/16,6831/32],[-8185/48,-742643/864],
[-1705/48,-57707/864],[-13635/16,306207/32],[-131065/48,-47449331/864]]
? L[2][2] \\ isogeny f_2
%4 = [x^3+3/4*x^2+19/2*x-311/12,
1/2*x^4+(y+1)*x^3+(y-4)*x^2+(-9*y+23)*x+(55*y+55/2),x+1/3]
? L[2][3] \\ dual isogeny g_2
%5 = [1/9*x^3-1/4*x^2-141/16*x+5613/64,
-1/18*x^4+(1/27*y-1/3)*x^3+(-1/12*y+87/16)*x^2+(49/16*y-48)*x
+(-3601/64*y+16947/512),x-3/4]
? apply(E->ellidentify(ellinit(E))[1][1], LE)
%6 = ["14a1","14a4","14a3","14a2","14a6","14a5"]
? M
%7 =
[1 3 3 2 6 6]
[3 1 9 6 2 18]
[3 9 1 6 18 2]
[2 6 6 1 3 3]
[6 2 18 3 1 9]
[6 18 2 3 9 1]
The library syntax is GEN ellisomat(GEN E, long p, long fl).

488
3.15.39 ellisoncurve(E, z). Gives 1 (i.e. true) if the point z is on the elliptic curve E, 0 otherwise.
If E or z have imprecise coefficients, an attempt is made to take this into account, i.e. an imprecise
equality is checked, not a precise one. It is allowed for z to be a vector of points in which case a
vector (of the same type) is returned.
The library syntax is GEN ellisoncurve(GEN E, GEN z). Also available is int oncurve(GEN
E, GEN z) which does not accept vectors of points.

3.15.40 ellisotree(E). Given an elliptic curve E defined over Q or a set of Q-isogenous curves as
given by ellisomat, return a pair [L, M ] where
• L lists the minimal models of the isomorphism classes of elliptic curves Q-isogenous to E (or
in the set of isogenous curves),
• M is the adjacency matrix of the prime degree isogenies tree: there is an edge from Ei to Ej
if there is an isogeny Ei → Ej of prime degree such that the Néron differential forms are preserved.
? E = ellinit("14a1");
? [L,M] = ellisotree(E);
? M
%3 =
[0 0 3 2 0 0]
[3 0 0 0 2 0]
[0 0 0 0 0 2]
[0 0 0 0 0 3]
[0 0 0 3 0 0]
[0 0 0 0 0 0]
? [L2,M2] = ellisotree(ellisomat(E,2,1));
%4 =
[0 2]
[0 0]
? [L3,M3] = ellisotree(ellisomat(E,3,1));
? M3
%6 =
[0 0 3]
[3 0 0]
[0 0 0]
Compare with the result of ellisomat.
? [L,M]=ellisomat(E,,1);
? M
%7 =
[1 3 3 2 6 6]
[3 1 9 6 2 18]
[3 9 1 6 18 2]
[2 6 6 1 3 3]
[6 2 18 3 1 9]

489
 [6 18 2 3 9 1]

The library syntax is GEN ellisotree(GEN E).

3.15.41 ellissupersingular(E, {p}). Return 1 if the elliptic curve E defined over a number field,
Qp or a finite field is supersingular at p, and 0 otherwise. If the curve is defined over a number
field, p must be explicitly given, and must be a prime number, resp. a maximal ideal, if the curve
is defined over Q, resp. a general number field: we return 1 if and only if E has supersingular good
reduction at p.

Alternatively, E can be given by its j-invariant in a finite field. In this case p must be omitted.

? setrand(1); \\ make the choice of g deterministic

? g = ffprimroot(ffgen(7^5))
%1 = 4*x^4 + 5*x^3 + 6*x^2 + 5*x + 6
? [g^n | n <- [1 .. 7^5 - 1], ellissupersingular(g^n)]
%2 = [6]
? K = nfinit(y^3-2); P = idealprimedec(K, 2)[1];
? E = ellinit([y,1], K);
? ellissupersingular(E, P)
%5 = 1
? Q = idealprimedec(K,5)[1];
? ellissupersingular(E, Q)
%6 = 0

The library syntax is GEN ellissupersingular(GEN E, GEN p = NULL). Also available is int
elljissupersingular(GEN j) where j is a j-invariant of a curve over a finite field.

3.15.42 ellj(x). Elliptic j-invariant. x must be a complex number with positive imaginary part,
or convertible into a power series or a p-adic number with positive valuation.

The library syntax is GEN jell(GEN x, long prec).

3.15.43 elllocalred(E, {p}). Calculates the Kodaira type of the local fiber of the elliptic curve
E at p. E must be an ell structure as output by ellinit, over Q` (p better left omitted, else
equal to `) over Q (p a rational prime) or a number field K (p a maximal ideal given by a prid
structure). The result is a 4-component vector [f, kod, v, c]. Here f is the exponent of p in the
arithmetic conductor of E, and kod is the Kodaira type which is coded as follows:

1 means good reduction (type I0 ), 2, 3 and 4 mean types II, III and IV respectively, 4 + ν with
ν > 0 means type Iν ; finally the opposite values −1, −2, etc. refer to the starred types I∗0 , II∗ , etc.
The third component v is itself a vector [u, r, s, t] giving the coordinate changes done during the
local reduction; u = 1 if and only if the given equation was already minimal at p. Finally, the last
component c is the local Tamagawa number cp .

The library syntax is GEN elllocalred(GEN E, GEN p = NULL).

490
3.15.44 elllog(E, P, G, {o}). Given two points P and G on the elliptic curve E/Fq , returns the
discrete logarithm of P in base G, i.e. the smallest nonnegative integer n such that P = [n]G.
See znlog for the limitations of the underlying discrete log algorithms. If present, o represents the
order of G, see Section 3.8.2; the preferred format for this parameter is [N, factor(N)], where N
is the order of G.
If no o is given, assume that G generates the curve. The function also assumes that P is a
multiple of G.
? a = ffgen(ffinit(2,8),’a);
? E = ellinit([a,1,0,0,1]); \\ over F_{2^8}
? x = a^3; y = ellordinate(E,x)[1];
? P = [x,y]; G = ellmul(E, P, 113);
? ord = [242, factor(242)]; \\ P generates a group of order 242. Initialize.
? ellorder(E, G, ord)
%4 = 242
? e = elllog(E, P, G, ord)
%5 = 15
? ellmul(E,G,e) == P
%6 = 1
The library syntax is GEN elllog(GEN E, GEN P, GEN G, GEN o = NULL).

3.15.45 elllseries(E, s, {A = 1}). This function is deprecated, use lfun(E,s) instead.

E being an elliptic curve, given by an arbitrary model over Q as output by ellinit, this
function computes the value of the L-series of E at the (complex) point s. This function uses an
O(N 1/2 ) algorithm, where N is the conductor.
The optional parameter A fixes a cutoff point for the integral and is best left omitted; the result
must be independent of A, up to realprecision, so this allows to check the function’s accuracy.
The library syntax is GEN elllseries(GEN E, GEN s, GEN A = NULL, long prec).

3.15.46 ellminimaldisc(E). E being an elliptic curve defined over a number field output by
ellinit, return the minimal discriminant ideal of E.
The library syntax is GEN ellminimaldisc(GEN E).

3.15.47 ellminimalmodel(E, {&v}). Let E be an ell structure over a number field K. This
function determines whether E admits a global minimal integral model. If so, it returns it and
sets v = [u, r, s, t] to the corresponding change of variable: the return value is identical to that of
ellchangecurve(E, v).
Q (v ∆−δ )/12
Else return the (nonprincipal) Weierstrass class of E, i.e. the class of p p p where
δ
∆ = E.disc is the model’s discriminant and pp is the local minimal discriminant. This function
requires either that E be defined over the rational field Q (with domain D = 1 in ellinit), in
which case a global minimal model always exists, or over a number field given by a bnf structure.
The Weierstrass class is given in bnfisprincipal format, i.e. in terms of the K.gen generators.
The resulting model has integral coefficients and is everywhere minimal, the coefficients a1 and
a3 are reduced modulo 2 (in terms of the fixed integral basis K.zk) and a2 is reduced modulo 3.
Over Q, we further require that a1 and a3 be 0 or 1, that a2 be 0 or ±1 and that u > 0 in the
change of variable: both the model and the change of variable v are then unique.

491
 ? e = ellinit([6,6,12,55,233]); \\ over Q
? E = ellminimalmodel(e, &v);
? E[1..5]
%3 = [0, 0, 0, 1, 1]
? v
%4 = [2, -5, -3, 9]

? K = bnfinit(a^2-65); \\ over a nonprincipal number field

? K.cyc
%2 = [2]
? u = Mod(8+a, K.pol);
? E = ellinit([1,40*u+1,0,25*u^2,0], K);
? ellminimalmodel(E) \\ no global minimal model exists over Z_K
%6 = [1]~
The library syntax is GEN ellminimalmodel(GEN E, GEN *v = NULL).

3.15.48 ellminimaltwist(E, {flag = 0}). Let E be an elliptic curve defined over Q, return a
discriminant D such that the twist of E by D is minimal among all possible quadratic twists, i.e.
if flag = 0, its minimal model has minimal discriminant, or if flag = 1, it has minimal conductor.
In the example below, we find a curve with j-invariant 3 and minimal conductor.

? E = ellminimalmodel(ellinit(ellfromj(3)));
? ellglobalred(E)[1]
%2 = 357075
? D = ellminimaltwist(E,1)
%3 = -15
? E2 = ellminimalmodel(ellinit(elltwist(E,D)));
? ellglobalred(E2)[1]
%5 = 14283
In the example below, flag = 0 and flag = 1 give different results.

? E = ellinit([1,0]);
? D0 = ellminimaltwist(E,0)
%7 = 1
? D1 = ellminimaltwist(E,1)
%8 = 8
? E0 = ellminimalmodel(ellinit(elltwist(E,D0)));
? [E0.disc, ellglobalred(E0)[1]]
%10 = [-64, 64]
? E1 = ellminimalmodel(ellinit(elltwist(E,D1)));
? [E1.disc, ellglobalred(E1)[1]]
%12 = [-4096, 32]
The library syntax is GEN ellminimaltwist0(GEN E, long flag). Also available are GEN
ellminimaltwist(E) for flag = 0, and GEN ellminimaltwistcond(E) for flag = 1.

492
3.15.49 ellmoddegree(e). e being an elliptic curve defined over Q output by ellinit, compute
the modular degree of e divided by the square of the Manin constant c. It is conjectured that c = 1
for the strong Weil curve in the isogeny class (optimal quotient of J0 (N )) and this can be proven
using ellweilcurve when the conductor N is moderate.
? E = ellinit("11a1"); \\ from Cremona table: strong Weil curve and c = 1
? [v,smith] = ellweilcurve(E); smith \\ proof of the above
%2 = [[1, 1], [5, 1], [1, 1/5]]
? ellmoddegree(E)
%3 = 1
? [ellidentify(e)[1][1] | e<-v]
%4 = ["11a1", "11a2", "11a3"]
? ellmoddegree(ellinit("11a2"))
%5 = 5
? ellmoddegree(ellinit("11a3"))
%6 = 1/5
The modular degree of 11a1 is 1 (because ellweilcurve or Cremona’s table prove that the Manin
constant is 1 for this curve); the output of ellweilcurve also proves that the Manin constants of
11a2 and 11a3 are 1 and 5 respectively, so the actual modular degree of both 11a2 and 11a3 is 5.
The library syntax is GEN ellmoddegree(GEN e).

3.15.50 ellmodulareqn(N, {x}, {y}). Given a prime N < 500, return a vector [P, t] where P (x, y)
is a modular equation of level N , i.e. a bivariate polynomial with integer coefficients; t indicates
the type of this equation: either canonical (t = 0) or Atkin (t = 1). This function requires the
seadata package and its only use is to give access to the package contents. See polmodular for a
more general and more flexible function.
Let j be the j-invariant function. The polynomial P satisfies the functional equation,

P (f, j) = P (f | WN , j | WN ) = 0

for some modular function f = fN (hand-picked for each fixed N to minimize its size, see below),
where WN (τ ) = −1/(N τ ) is the Atkin-Lehner involution. These two equations allow to compute
the values of the classical modular polynomial ΦN , such that ΦN (j(τ ), j(N τ )) = 0, while being
much smaller than the latter. More precisely, we have j(WN (τ )) = j(N τ ); the function f is
invariant under Γ0 (N ) and also satisfies
• for Atkin type: f | WN = f ;
• for canonical type: let s = 12/gcd(12, N − 1), then f | WN = N s /f . In this case, f has a

2s
simple definition: f (τ ) = N s η(N τ )/η(τ ) , where η is Dedekind’s eta function.
The following GP function returns values of the classical modular polynomial by eliminating
fN (τ ) in the above functional equation, for N ≤ 31 or N ∈ {41, 47, 59, 71}.
classicaleqn(N, X=’X, Y=’Y)=
{
my([P,t] = ellmodulareqn(N), Q, d);
if (poldegree(P,’y) > 2, error("level unavailable in classicaleqn"));
if (t == 0, \\ Canonical
my(s = 12/gcd(12,N-1));

493
 Q = ’x^(N+1) * substvec(P,[’x,’y],[N^s/’x,Y]);
d = N^(s*(2*N+1)) * (-1)^(N+1);
, \\ Atkin
Q = subst(P,’y,Y);
d = (X-Y)^(N+1));
polresultant(subst(P,’y,X), Q) / d;
}
The library syntax is GEN ellmodulareqn(long N, long x = -1, long y = -1) where x, y
are variable numbers.

3.15.51 ellmul(E, z, n). Computes [n]z, where z is a point on the elliptic curve E. The exponent
n is in Z, or may be a complex quadratic integer if the curve E has complex multiplication by n
(if not, an error message is issued).
? Ei = ellinit([1,0]); z = [0,0];
? ellmul(Ei, z, 10)
%2 = [0] \\ unsurprising: z has order 2
? ellmul(Ei, z, I)
%3 = [0, 0] \\ Ei has complex multiplication by Z[i]
? ellmul(Ei, z, quadgen(-4))
%4 = [0, 0] \\ an alternative syntax for the same query
? Ej = ellinit([0,1]); z = [-1,0];
? ellmul(Ej, z, I)
*** at top-level: ellmul(Ej,z,I)
*** ^--------------
*** ellmul: not a complex multiplication in ellmul.
? ellmul(Ej, z, 1+quadgen(-3))
%6 = [1 - w, 0]
The simple-minded algorithm for the CM case assumes that we are in characteristic 0, and
that the quadratic order to which n belongs has small discriminant.
The library syntax is GEN ellmul(GEN E, GEN z, GEN n).

3.15.52 ellneg(E, z). Opposite of the point z on elliptic curve E.

The library syntax is GEN ellneg(GEN E, GEN z).

3.15.53 ellnonsingularmultiple(E, P ). Given an elliptic curve E/Q (more precisely, a model

defined over Q of a curve) and a rational point P ∈ E(Q), returns the pair [R, n], where n is the
least positive integer such that R := [n]P has good reduction at every prime. More precisely, its
image in a minimal model is everywhere nonsingular.
? e = ellinit("57a1"); P = [2,-2];
? ellnonsingularmultiple(e, P)
%2 = [[1, -1], 2]
? e = ellinit("396b2"); P = [35, -198];
? [R,n] = ellnonsingularmultiple(e, P);
? n
%5 = 12
The library syntax is GEN ellnonsingularmultiple(GEN E, GEN P).

494
3.15.54 ellorder(E, z, {o}). Gives the order of the point z on the elliptic curve E, defined over a
finite field or a number field. Return (the impossible value) zero if the point has infinite order.

? E = ellinit([-157^2,0]); \\ the "157-is-congruent" curve

? P = [2,2]; ellorder(E, P)
%2 = 2
? P = ellheegner(E); ellorder(E, P) \\ infinite order
%3 = 0
? K = nfinit(polcyclo(11,t)); E=ellinit("11a3", K); T = elltors(E);
? ellorder(E, T.gen[1])
%5 = 25
? E = ellinit(ellfromj(ffgen(5^10)));
? ellcard(E)
%7 = 9762580
? P = random(E); ellorder(E, P)
%8 = 4881290
? p = 2^160+7; E = ellinit([1,2], p);
? N = ellcard(E)
%9 = 1461501637330902918203686560289225285992592471152
? o = [N, factor(N)];
? for(i=1,100, ellorder(E,random(E)))
time = 260 ms.
The parameter o, is now mostly useless, and kept for backward compatibility. If present,
it represents a nonzero multiple of the order of z, see Section 3.8.2; the preferred format for this
parameter is [ord, factor(ord)], where ord is the cardinality of the curve. It is no longer needed
since PARI is now able to compute it over large finite fields (was restricted to small prime fields
at the time this feature was introduced), and caches the result in E so that it is computed and
factored only once. Modifying the last example, we see that including this extra parameter provides
no improvement:

? o = [N, factor(N)];
? for(i=1,100, ellorder(E,random(E),o))
time = 260 ms.
The library syntax is GEN ellorder(GEN E, GEN z, GEN o = NULL). The obsolete form GEN
orderell(GEN e, GEN z) should no longer be used.

3.15.55 ellordinate(E, x). Gives a 0, 1 or 2-component vector containing the y-coordinates of the
points of the curve E having x as x-coordinate.
The library syntax is GEN ellordinate(GEN E, GEN x, long prec).

3.15.56 ellpadicL(E, p, n, {s = 0}, {r = 0}, {D = 1}). Returns the value (or r-th derivative) on a
character χs of Z∗p of the p-adic L-function of the elliptic curve E/Q, twisted by D, given modulo
pn .

495
Characters. The set of continuous characters of Gal(Q(µp∞ )/Q) is identified to Z∗p via the
∗
cyclotomic character χ with values in Qp . Denote by τ : Z∗p → Z∗p the Teichmüller character,
with values in the (p − 1)-th roots of 1 for p 6= 2, and {−1, 1} for p = 2; finally, let hχi = χτ −1 ,
with values in 1 + 2pZp . In GP, the continuous character of Gal(Q(µp∞ )/Q) given by hχis1 τ s2 is
represented by the pair of integers s = (s1 , s2 ), with s1 ∈ Zp and s2 mod p − 1 for p > 2, (resp.
mod 2 for p = 2); s may be also an integer, representing (s, s) or χs .

The p-adic L function.

R The p-adic L function Lp is defined on the set of continuous characters
of Gal(Q(µp∞ )/Q), as Z∗ χs dµ for a certain p-adic distribution µ on Z∗p . The derivative is given
p
by Z
L(r) s
p (E, χ ) = logrp (a)χs (a)dµ(a).
Z∗
p

More precisely:
1
• When E has good supersingular reduction, Lp takes its values in D := HdR (E/Q) ⊗Q Qp
and satisfies
(1 − p−1 F )−2 Lp (E, χ0 ) = (L(E, 1)/Ω) · ω
where F is the Frobenius, L(E, 1) is the value of the complex L function at 1, ω is the Néron
differential and Ω the attached period on E(R). Here, χ0 represents the trivial character.
(r)
The function returns the components of Lp (E, χs ) in the basis (ω, F ω).
• When E has ordinary good reduction, this method only defines the projection of Lp (E, χs )
on the α-eigenspace, where α is the unit eigenvalue for F . This is what the function returns. We
have
(1 − α−1 )−2 Lp,α (E, χ0 ) = L(E, 1)/Ω.

Two supersingular examples:

? cxL(e) = bestappr( ellL1(e) / e.omega[1] );
? e = ellinit("17a1"); p=3; \\ supersingular, a3 = 0
? L = ellpadicL(e,p,4);
? F = [0,-p;1,ellap(e,p)]; \\ Frobenius matrix in the basis (omega,F(omega))
? (1-p^(-1)*F)^-2 * L / cxL(e)
%5 = [1 + O(3^5), O(3^5)]~ \\ [1,0]~
? e = ellinit("116a1"); p=3; \\ supersingular, a3 != 0~
? L = ellpadicL(e,p,4);
? F = [0,-p; 1,ellap(e,p)];
? (1-p^(-1)*F)^-2*L~ / cxL(e)
%9 = [1 + O(3^4), O(3^5)]~
Good ordinary reduction:
? e = ellinit("17a1"); p=5; ap = ellap(e,p)
%1 = -2 \\ ordinary
? L = ellpadicL(e,p,4)
%2 = 4 + 3*5 + 4*5^2 + 2*5^3 + O(5^4)
? al = padicappr(x^2 - ap*x + p, ap + O(p^7))[1];
? (1-al^(-1))^(-2) * L / cxL(e)
%4 = 1 + O(5^4)

496
 Twist and Teichmüller:
? e = ellinit("17a1"); p=5; \\ ordinary
\\ 2nd derivative at tau^1, twist by -7
? ellpadicL(e, p, 4, [0,1], 2, -7)
%2 = 2*5^2 + 5^3 + O(5^4)
We give an example of non split multiplicative reduction (see ellpadicbsd for more examples).
? e=ellinit("15a1"); p=3; n=5;
? L = ellpadicL(e,p,n)
%2 = 2 + 3 + 3^2 + 3^3 + 3^4 + O(3^5)
? (1 - ellap(e,p))^(-1) * L / cxL(e)
%3 = 1 + O(3^5)
This function is a special case of mspadicL and it also appears as the first term of mspadic-
series:
? e = ellinit("17a1"); p=5;
? L = ellpadicL(e,p,4)
%2 = 4 + 3*5 + 4*5^2 + 2*5^3 + O(5^4)
? [M,phi] = msfromell(e, 1);
? Mp = mspadicinit(M, p, 4);
? mu = mspadicmoments(Mp, phi);
? mspadicL(mu)
%6 = 4 + 3*5 + 4*5^2 + 2*5^3 + 2*5^4 + 5^5 + O(5^6)
? mspadicseries(mu)
%7 = (4 + 3*5 + 4*5^2 + 2*5^3 + 2*5^4 + 5^5 + O(5^6))
+ (3 + 3*5 + 5^2 + 5^3 + O(5^4))*x
+ (2 + 3*5 + 5^2 + O(5^3))*x^2
+ (3 + 4*5 + 4*5^2 + O(5^3))*x^3
+ (3 + 2*5 + O(5^2))*x^4 + O(x^5)
These are more cumbersome than ellpadicL but allow to compute at different characters, or
successive derivatives, or to twist by a quadratic character essentially for the cost of a single call
to ellpadicL due to precomputations.
The library syntax is GEN ellpadicL(GEN E, GEN p, long n, GEN s = NULL, long r, GEN
D = NULL).

3.15.57 ellpadicbsd(E, p, n, {D = 1}). Given an elliptic curve E over Q, its quadratic twist ED
and a prime number p, this function is a p-adic analog of the complex functions ellanalyticrank
and ellbsd. It calls ellpadicL with initial accuracy pn and may increase it internally; it returns
a vector [r, Lp ] where
• Lp is a p-adic number (resp. a pair of p-adic numbers if E has good supersingular reduction)
defined modulo pN , conjecturally equal to Rp S, where Rp is the p-adic regulator as given by
ellpadicregulator (in the basis (ω, F ω)) and S is the cardinal of the Tate-Shafarevich group for
the quadratic twist ED .
• r is an upper bound for the analytic rank of the p-adic L-function attached to ED : we know
for sure that the i-th derivative of Lp (ED , .) at χ0 is O(pN ) for all i < r and that its r-th derivative
is nonzero; it is expected that the true analytic rank is equal to the rank of the Mordell-Weil group

497
ED (Q), plus 1 if the reduction of ED at p is split multiplicative; if r = 0, then both the analytic
rank and the Mordell-Weil rank are unconditionnally 0.
Recall that the p-adic BSD conjecture (Mazur, Tate, Teitelbaum, Bernardi, Perrin-Riou) pre-
dicts an explicit link between Rp S and

(1 − p−1 F )−2 · L(r) 0

p (ED , χ )/r!

where r is the analytic rank of the p-adic L-function attached to ED and F is the Frobenius on
1
HdR ; see ellpadicL for definitions.
? E = ellinit("11a1"); p = 7; n = 5; \\ good ordinary
? ellpadicbsd(E, 7, 5) \\ rank 0,
%2 = [0, 1 + O(7^5)]
? E = ellinit("91a1"); p = 7; n = 5; \\ non split multiplicative
? [r,Lp] = ellpadicbsd(E, p, n)
%5 = [1, 2*7 + 6*7^2 + 3*7^3 + 7^4 + O(7^5)]
? R = ellpadicregulator(E, p, n, E.gen)
%6 = 2*7 + 6*7^2 + 3*7^3 + 7^4 + 5*7^5 + O(7^6)
? sha = Lp/R
%7 = 1 + O(7^4)
? E = ellinit("91b1"); p = 7; n = 5; \\ split multiplicative
? [r,Lp] = ellpadicbsd(E, p, n)
%9 = [2, 2*7 + 7^2 + 5*7^3 + O(7^4)]
? ellpadicregulator(E, p, n, E.gen)
%10 = 2*7 + 7^2 + 5*7^3 + 6*7^4 + 2*7^5 + O(7^6)
? [rC, LC] = ellanalyticrank(E);
? [r, rC]
%12 = [2, 1] \\ r = rC+1 because of split multiplicative reduction
? E = ellinit("53a1"); p = 5; n = 5; \\ supersingular
? [r, Lp] = ellpadicbsd(E, p, n);
? r
%15 = 1
? Lp
%16 = [3*5 + 2*5^2 + 2*5^5 + O(5^6), \
5 + 3*5^2 + 4*5^3 + 2*5^4 + 5^5 + O(5^6)]
? R = ellpadicregulator(E, p, n, E.gen)
%17 = [3*5 + 2*5^2 + 2*5^5 + O(5^6), 5 + 3*5^2 + 4*5^3 + 2*5^4 + O(5^5)]
\\ expect Lp = R*#Sha, hence (conjecturally) #Sha = 1
? E = ellinit("84a1"); p = 11; n = 6; D = -443;
? [r,Lp] = ellpadicbsd(E, 11, 6, D) \\ Mordell-Weil rank 0, no regulator
%19 = [0, 3 + 2*11 + O(11^6)]
? lift(Lp) \\ expected cardinal for Sha is 5^2
%20 = 25
? ellpadicbsd(E, 3, 12, D) \\ at 3
%21 = [1, 1 + 2*3 + 2*3^2 + O(3^8)]
? ellpadicbsd(E, 7, 8, D) \\ and at 7
%22 = [0, 4 + 3*7 + O(7^8)]
The library syntax is GEN ellpadicbsd(GEN E, GEN p, long n, GEN D = NULL).

498
3.15.58 ellpadicfrobenius(E, p, n). If p > 2 is a prime and E is an elliptic curve on Q with
good reduction at p, return the matrix of the Frobenius endomorphism ϕ on the crystalline module
1
Dp (E) = Qp ⊗ HdR (E/Q) with respect to the basis of the given model (ω, η = xω), where ω =
dx/(2y + a1 x + a3 ) is the invariant differential. The characteristic polynomial of ϕ is x2 − ap x + p.
The matrix is computed to absolute p-adic precision pn .
? E = ellinit([1,-1,1,0,0]);
? F = ellpadicfrobenius(E,5,3);
? lift(F)
%3 =
[120 29]
[ 55 5]
? charpoly(F)
%4 = x^2 + O(5^3)*x + (5 + O(5^3))
? ellap(E, 5)
%5 = 0
The library syntax is GEN ellpadicfrobenius(GEN E, long p, long n).

3.15.59 ellpadicheight(E, p, n, P, {Q}). Cyclotomic p-adic height of the rational point P on the
elliptic curve E (defined over Q), given to n p-adic digits. If the argument Q is present, computes
the value of the bilinear form (h(P + Q) − h(P − Q))/4.
1
Let D := HdR (E) ⊗Q Qp be the Qp vector space spanned by ω (invariant differential dx/(2y +
a1 x + a3) related to the given model) and η = xω. Then the cyclotomic p-adic height hE associates
to P ∈ E(Q) an element f ω + gη in D. This routine returns the vector [f, g] to n p-adic digits. If
P ∈ E(Q) is in the kernel of reduction mod p and if its reduction at all finite places is non singular,
then g = −(logE P )2 , where logE is the logarithm for the formal group of E at p.
If furthermore the model is of the form Y 2 = X 3 + aX + b and P = (x, y), then

f = logp (denominator(x)) − 2 logp (σ(P ))

where σ(P ) is given by ellsigma(E, P ).

Recall (Advanced topics in the arithmetic of elliptic curves, Theorem 3.2) that the local height
function over the complex numbers is of the form

λ(z) = − log(|E.disc|)/6 + <(zη(z)) − 2 log(σ(z)).

(N.B. our normalization for local and global heights is twice that of Silverman’s).
? E = ellinit([1,-1,1,0,0]); P = [0,0];
? ellpadicheight(E,5,3, P)
%2 = [3*5 + 5^2 + 2*5^3 + O(5^4), 5^2 + 4*5^4 + O(5^5)]
? E = ellinit("11a1"); P = [5,5]; \\ torsion point
? ellpadicheight(E,19,6, P)
%4 = [0, 0]
? E = ellinit([0,0,1,-4,2]); P = [-2,1];
? ellpadicheight(E,3,3, P)
%6 = [2*3^2 + 2*3^3 + 3^4 + O(3^5), 2*3^2 + 3^4 + O(3^5)]
? ellpadicheight(E,3,5, P, elladd(E,P,P))

499
 %7 = [3^2 + 2*3^3 + O(3^7), 3^2 + 3^3 + 2*3^4 + 3^5 + O(3^7)]
• When E has good ordinary reduction at p or non split multiplicative reduction, the “canon-
ical” p-adic height is given by
s2 = ellpadics2(E,p,n);
ellpadicheight(E, p, n, P) * [1,-s2]~
Since s2 does not depend on P , it is preferable to compute it only once:
? E = ellinit("5077a1"); p = 5; n = 7; \\ rank 3
? s2 = ellpadics2(E,p,n);
? M = ellpadicheightmatrix(E,p, n, E.gen) * [1,-s2]~;
? matdet(M) \\ p-adic regulator on the points in E.gen
%4 = 5 + 5^2 + 4*5^3 + 2*5^4 + 2*5^5 + 2*5^6 + O(5^7)
• When E has split multiplicative reduction at p (Tate curve), the “canonical” p-adic height
is given by
Ep = ellinit(E[1..5], O(p^(n))); \\ E seen as a Tate curve over Qp
[u2,u,q] = Ep.tate;
ellpadicheight(E, p, n, P) * [1,-s2 + 1/log(q)/u2]]~
where s2 is as above. For example,
? E = ellinit("91b1"); P =[-1, 3]; p = 7; n = 5;
? Ep = ellinit(E[1..5], O(p^(n)));
? s2 = ellpadics2(E,p,n);
? [u2,u,q] = Ep.tate;
? H = ellpadicheight(E,p, n, P) * [1,-s2 + 1/log(q)/u2]~
%5 = 2*7 + 7^2 + 5*7^3 + 6*7^4 + 2*7^5 + O(7^6)
These normalizations are chosen so that p-adic BSD conjectures are easy to state, see ell-
padicbsd.
The library syntax is GEN ellpadicheight0(GEN E, GEN p, long n, GEN P, GEN Q = NULL)
.

3.15.60 ellpadicheightmatrix(E, p, n, Q). Q being a vector of points, this function returns the
“Gram matrix” [F, G] of the cyclotomic p-adic height hE with respect to the basis (ω, η) of D =
1
HdR (E) ⊗Q Qp given to n p-adic digits. In other words, if ellpadicheight(E, p, n, Q[i], Q[j]) =
[f, g], corresponding to f ω + gη in D, then F [i, j] = f and G[i, j] = g.
? E = ellinit([0,0,1,-7,6]); Q = [[-2,3],[-1,3]]; p = 5; n = 5;
? [F,G] = ellpadicheightmatrix(E,p,n,Q);
? lift(F) \\ p-adic entries, integral approximation for readability
%3 =
[2364 3100]
[3100 3119]
? G
%4 =
[25225 46975]
[46975 61850]

500
 ? [F,G] * [1,-ellpadics2(E,p,n)]~
%5 =
[4 + 2*5 + 4*5^2 + 3*5^3 + O(5^5) 4*5^2 + 4*5^3 + 5^4 + O(5^5)]
[ 4*5^2 + 4*5^3 + 5^4 + O(5^5) 4 + 3*5 + 4*5^2 + 4*5^3 + 5^4 + O(5^5)]

The library syntax is GEN ellpadicheightmatrix(GEN E, GEN p, long n, GEN Q).

3.15.61 ellpadiclambdamu(E, p, {D = 1}, {i = 0}). Let p be a prime number and let E/Q
be a rational elliptic curve with good or bad multiplicative reduction at p. Return the Iwasawa
invariants λ and µ for the p-adic L function Lp (E), twisted by (D/.) and the i-th power of the
Teichmüller character τ , see ellpadicL for details about Lp (E).

Let χ be the cyclotomic character and choose γ in Gal(Qp (µp∞ )/Qp ) such that χ(γ) = 1 + 2p.
Let L̂(i),D ∈ Qp [[X]] ⊗ Dcris such that

(< χ >s τ i )(L̂(i),D (γ − 1)) = Lp E, < χ >s τ i (D/.) .

• When E has good ordinary or bad multiplicative reduction at p. By Weierstrass’s preparation

theorem the series L̂(i),D can be written pµ (X λ + pG(X)) up to a p-adic unit, where G(X) ∈ Zp [X].
The function returns [λ, µ].

• When E has good supersingular reduction, we define a sequence of polynomials Pn in Qp [X]

of degree < pn (and bounded denominators), such that

n
L̂(i),D ≡ Pn ϕn+1 ωE − ξn Pn−1 ϕn+2 ωE mod (1 + X)p − 1 Qp [X] ⊗ Dcris ,

where ξn = polcyclo(pn , 1 + X). Let λn , µn be the invariants of Pn . We find that

• µn is nonnegative and decreasing for n of given parity hence µ2n tends to a limit µ+ and
µ2n+1 tends to a limit µ− (both conjecturally 0).

• there exists integers λ+ , λ− in Z (denoted with a ˜ in the reference below) such that

lim λ2n + 1/(p + 1) = λ+ and lim λ2n+1 + p/(p + 1) = λ− .

n→∞ n→∞

The function returns [[λ+ , λ− ], [µ+ , µ− ]].

Reference: B. Perrin-Riou, Arithmétique des courbes elliptiques à réduction supersingulière en p,

Experimental Mathematics, 12, 2003, pp. 155-186.

The library syntax is GEN ellpadiclambdamu(GEN E, long p, long D, long i).

501
3.15.62 ellpadiclog(E, p, n, P ). Given E defined over K = Q or Qp and P = [x, y] on E(K)
in the kernel of reduction mod p, let t(P ) = −x/y be the formal group parameter; this function
returns L(t), where L denotes the formal logarithm (mapping the formal group of E to the additive
formal group) attached to the canonical invariant differential: dL = dx/(2y + a1 x + a3 ).

? E = ellinit([0,0,1,-4,2]); P = [-2,1];
? ellpadiclog(E,2,10,P)
%2 = 2 + 2^3 + 2^8 + 2^9 + 2^10 + O(2^11)
? E = ellinit([17,42]);
? p=3; Ep = ellinit(E,p); \\ E mod p
? P=[114,1218]; ellorder(Ep,P) \\ the order of P on (E mod p) is 2
%5 = 2
? Q = ellmul(E,P,2) \\ we need a point of the form 2*P
%6 = [200257/7056, 90637343/592704]
? ellpadiclog(E,3,10,Q)
%7 = 3 + 2*3^2 + 3^3 + 3^4 + 3^5 + 3^6 + 2*3^8 + 3^9 + 2*3^10 + O(3^11)
The library syntax is GEN ellpadiclog(GEN E, GEN p, long n, GEN P).

3.15.63 ellpadicregulator(E, p, n, S). Let E/Q be an elliptic curve. Return the determinant of
the Gram matrix of the vector of points S = (S1 , · · · , Sr ) with respect to the “canonical” cyclotomic
p-adic height on E, given to n (p-adic) digits.
When E has ordinary reduction at p, this is the expected Gram deteterminant in Qp .
In the case of supersingular reduction of E at p, the definition requires care: the regulator R is
1
an element of D := HdR (E) ⊗Q Qp , which is a two-dimensional Qp -vector space spanned by ω and
η = xω (which are defined over Q) or equivalently but now over Qp by ω and F ω where F is the
Frobenius endomorphism on D as defined in ellpadicfrobenius. On D we define the cyclotomic
height hE = f ω + gη (see ellpadicheight) and a canonical alternating bilinear form [., .]D such
that [ω, η]D = 1.
For any ν ∈ D, we can define a height hν := [hE , ν]D from E(Q) to Qp and h·, ·iν the attached
bilinear form. In particular, if hE = f ω + gη, then hη = [hE , η]D = f and hω = [hE , ω]D = −g
hence hE = hη ω − hω η. Then, R is the unique element of D such that

[ω, ν]r−1
D [R, ν]D = det(hSi , Sj iν )

for all ν ∈ D not in Qp ω. The ellpadicregulator function returns R in the basis (ω, F ω), which
was chosen so that p-adic BSD conjectures are easy to state, see ellpadicbsd.
Note that by definition
[R, η]D = det(hSi , Sj iη )

and
[R, ω + η]D = det(hSi , Sj iω+η ).

The library syntax is GEN ellpadicregulator(GEN E, GEN p, long n, GEN S).

502
3.15.64 ellpadics2(E, p, n). If p > 2 is a prime and E/Q is an elliptic curve with ordinary good
reduction at p, returns the slope of the unit eigenvector of ellpadicfrobenius(E,p,n), i.e., the
1
action of Frobenius ϕ on the crystalline module Dp (E) = Qp ⊗ HdR (E/Q) in the basis of the given
model (ω, η = xω), where ω is the invariant differential dx/(2y + a1 x + a3 ). In other words, η + s2 ω
is an eigenvector for the unit eigenvalue of ϕ.

? e=ellinit([17,42]);
? ellpadics2(e,13,4)
%2 = 10 + 2*13 + 6*13^3 + O(13^4)

This slope is the unique c ∈ 3−1 Zp such that the odd solution σ(t) = t + O(t2 ) of

1 dσ
−d( ) = (x(t) + c)ω
σ ω

is in tZp [[t]].

It is equal to b2 /12 − E2 /12 where E2 is the value of the Katz p-adic Eisenstein series of weight
2 on (E, ω). This is used to construct a canonical p-adic height when E has good ordinary reduction
at p as follows

s2 = ellpadics2(E,p,n);
h(E,p,n, P, s2) = ellpadicheight(E, [p,[1,-s2]],n, P);

Since s2 does not depend on the point P , we compute it only once.

The library syntax is GEN ellpadics2(GEN E, GEN p, long n).

3.15.65 ellperiods(w, {flag = 0}). Let w describe a complex period lattice (w = [w1 , w2 ] or an
ellinit structure). Returns normalized periods [W1 , W2 ] generating the same lattice such that
τ := W1 /W2 has positive imaginary part and lies in the standard fundamental domain for SL2 (Z).

If flag = 1, the function returns [[W1 , W2 ], [η1 , η2 ]], where η1 and η2 are the quasi-periods
attached to [W1 , W2 ], satisfying η2 W1 − η1 W2 = 2iπ.

The output of this function is meant to be used as the first argument given to ellwp, ellzeta,
ellsigma or elleisnum. Quasi-periods are needed by ellzeta and ellsigma only.

? L = ellperiods([1,I],1);
? [w1,w2] = L[1]; [e1,e2] = L[2];
? e2*w1 - e1*w2
%3 = 6.2831853071795864769252867665590057684*I
? ellzeta(L, 1/2 + 2*I)
%4 = 1.5707963... - 6.283185307...*I
? ellzeta([1,I], 1/2 + 2*I) \\ same but less efficient
%4 = 1.5707963... - 6.283185307...*I

The library syntax is GEN ellperiods(GEN w, long flag, long prec).

503
3.15.66 ellpointtoz(E, P ). If E/C ' C/Λ is a complex elliptic curve (Λ = E.omega), computes a
complex number z, well-defined modulo the lattice Λ, corresponding to the point P ; i.e. such that
P = [℘Λ (z), ℘0Λ (z)] satisfies the equation

y 2 = 4x3 − g2 x − g3 ,

where g2 , g3 are the elliptic invariants.

If E is defined over R and P ∈ E(R), we have more precisely, 0 ≤ <(t) < w1 and 0 ≤ =(t) <
=(w2), where (w1, w2) are the real and complex periods of E.
? E = ellinit([0,1]); P = [2,3];
? z = ellpointtoz(E, P)
%2 = 3.5054552633136356529375476976257353387
? ellwp(E, z)
%3 = 2.0000000000000000000000000000000000000
? ellztopoint(E, z) - P
%4 = [2.548947057811923643 E-57, 7.646841173435770930 E-57]
? ellpointtoz(E, [0]) \\ the point at infinity
%5 = 0
If E is defined over a general number field, the function returns the values corresponding to
the various complex embeddings of the curve and of the point, in the same order as E.nf.roots:
? E=ellinit([-22032-15552*x,0], nfinit(x^2-2));
? P=[-72*x-108,0];
? ellisoncurve(E,P)
%3 = 1
? ellpointtoz(E,P)
%4 = [-0.52751724240790530394437835702346995884*I,
-0.090507650025885335533571758708283389896*I]
? E.nf.roots
%5 = [-1.4142135623730950488016887242096980786, \\ x-> -sqrt(2)
1.4142135623730950488016887242096980786] \\ x-> sqrt(2)
If E/Qp has multiplicative reduction, then E/Q̄p is analytically isomorphic to Q̄∗p /q Z (Tate
curve) for some p-adic integer q. The behavior is then as follows:
• If the reduction is split (E.tate[2] is a t_PADIC), we have an isomorphism φ : E(Qp ) ' Q∗p /q Z
and the function returns φ(P ) ∈ Qp .
• If the reduction is not split (E.tate[2] is a t_POLMOD), we only have an isomorphism φ :
E(K) ' K ∗ /q Z over the unramified quadratic extension K/Qp . In this case, the output φ(P ) ∈ K
is a t_POLMOD.
? E = ellinit([0,-1,1,0,0], O(11^5)); P = [0,0];
? [u2,u,q] = E.tate; type(u) \\ split multiplicative reduction
%2 = "t_PADIC"
? ellmul(E, P, 5) \\ P has order 5
%3 = [0]
? z = ellpointtoz(E, [0,0])
%4 = 3 + 11^2 + 2*11^3 + 3*11^4 + 6*11^5 + 10*11^6 + 8*11^7 + O(11^8)
? z^5

504
 %5 = 1 + O(11^9)
? E = ellinit(ellfromj(1/4), O(2^6)); x=1/2; y=ellordinate(E,x)[1];
? z = ellpointtoz(E,[x,y]); \\ t_POLMOD of t_POL with t_PADIC coeffs
? liftint(z) \\ lift all p-adics
%8 = Mod(8*u + 7, u^2 + 437)
The library syntax is GEN zell(GEN E, GEN P, long prec).

3.15.67 ellpow(E, z, n). Deprecated alias for ellmul.

The library syntax is GEN ellmul(GEN E, GEN z, GEN n).

3.15.68 ellratpoints(E, h, {flag = 0}). E being an integral model of elliptic curve , return a
vector containing the affine rational points on the curve of naive height less than h. If flag = 1,
stop as soon as a point is found; return either an empty vector or a vector containing a single point.
See hyperellratpoints for how h can be specified.
? E=ellinit([-25,1]);
? ellratpoints(E,10)
%2 = [[-5,1],[-5,-1],[-3,7],[-3,-7],[-1,5],[-1,-5],
[0,1],[0,-1],[5,1],[5,-1],[7,13],[7,-13]]
? ellratpoints(E,10,1)
%3 = [[-5,1]]
The library syntax is GEN ellratpoints(GEN E, GEN h, long flag).

3.15.69 ellrootno(E, {p}). E being an ell structure over Q as output by ellinit, this function
computes the local root number of its L-series at the place p (at the infinite place if p = 0). If
p is omitted, return the global root number and in this case the curve can also be defined over a
number field.
Note that the global root number is the sign of the functional equation and conjecturally is
the parity of the rank of the Mordell-Weil group. The equation for E needs not be minimal at p,
but if the model is already minimal the function will run faster.
The library syntax is long ellrootno(GEN E, GEN p = NULL).

3.15.70 ellsea(E, {tors = 0}). Let E be an ell structure as output by ellinit, defined over
a finite field Fq . This low-level function computes the order of the group E(Fq ) using the SEA
algorithm; compared to the high-level function ellcard, which includes SEA among its choice of
algorithms, the tors argument allows to speed up a search for curves having almost prime order
and whose quadratic twist may also have almost prime order. When tors is set to a nonzero value,
the function returns 0 as soon as it detects that the order has a small prime factor not dividing
tors; SEA considers modular polynomials of increasing prime degree ` and we return 0 as soon as
we hit an ` (coprime to tors) dividing #E(Fq ):
? ellsea(ellinit([1,1], 2^56+3477), 1)
%1 = 72057594135613381
? forprime(p=2^128,oo, q = ellcard(ellinit([1,1],p)); if(isprime(q),break))
time = 6,571 ms.
? forprime(p=2^128,oo, q = ellsea(ellinit([1,1],p),1);if(isprime(q),break))
time = 522 ms.

505
In particular, set tors to 1 if you want a curve with prime order, to 2 if you want to allow a cofactor
which is a power of two (e.g. for Edwards’s curves), etc. The early exit on bad curves yields a
massive speedup compared to running the cardinal algorithm to completion.
When tors is negative, similar checks are performed for the quadratic twist of the curve.
The following function returns a curve of prime order over Fp .
cryptocurve(p) =
{
while(1,
my(E, N, j = Mod(random(p), p));
E = ellinit(ellfromj(j));
N = ellsea(E, 1); if (!N, continue);
if (isprime(N), return(E));
\\ try the quadratic twist for free
if (isprime(2*p+2 - N), return(ellinit(elltwist(E))));
);
}
? p = randomprime([2^255, 2^256]);
? E = cryptocurve(p); \\ insist on prime order
%2 = 47,447ms
The same example without early abort (using ellcard(E) instead of ellsea(E, 1)) runs for about
5 minutes before finding a suitable curve.
The availability of the seadata package will speed up the computation, and is strongly rec-
ommended. The generic function ellcard should be preferred when you only want to compute the
cardinal of a given curve without caring about it having almost prime order:
• If the characteristic is too small (p ≤ 7) or the field cardinality is tiny (q ≤ 523) the generic
algorithm ellcard is used instead and the tors argument is ignored. (The reason for this is that
SEA is not implemented for p ≤ 7 and that if q ≤ 523 it is likely to run into an infinite loop.)
• If the field cardinality is smaller than about 250 , the generic algorithm will be faster.
• Contrary to ellcard, ellsea does not store the computed cardinality in E.
The library syntax is GEN ellsea(GEN E, long tors).

3.15.71 ellsearch(N ). This function finds all curves in the elldata database satisfying the
constraint defined by the argument N :
• if N is a character string, it selects a given curve, e.g. "11a1", or curves in the given isogeny
class, e.g. "11a", or curves with given conductor, e.g. "11";
• if N is a vector of integers, it encodes the same constraints as the character string above,
according to the ellconvertname correspondance, e.g. [11,0,1] for "11a1", [11,0] for "11a"
and [11] for "11";
• if N is an integer, curves with conductor N are selected.
If N codes a full curve name, for instance "11a1" or [11,0,1], the output format is
[N, [a1 , a2 , a3 , a4 , a6 ], G] where [a1 , a2 , a3 , a4 , a6 ] are the coefficients of the Weierstrass equation
of the curve and G is a Z-basis of the free part of the Mordell-Weil group attached to the curve.

506
 ? ellsearch("11a3")
%1 = ["11a3", [0, -1, 1, 0, 0], []]
? ellsearch([11,0,3])
%2 = ["11a3", [0, -1, 1, 0, 0], []]

If N is not a full curve name, then the output is a vector of all matching curves in the above
format:
? ellsearch("11a")
%1 = [["11a1", [0, -1, 1, -10, -20], []],
["11a2", [0, -1, 1, -7820, -263580], []],
["11a3", [0, -1, 1, 0, 0], []]]
? ellsearch("11b")
%2 = []
The library syntax is GEN ellsearch(GEN N). Also available is GEN ellsearchcurve(GEN N)
that only accepts complete curve names (as t_STR).

3.15.72 ellsigma(L, {z =0 x}, {flag = 0}). Computes the value at z of the Weierstrass σ function
attached to the lattice L as given by ellperiods(, 1): including quasi-periods is useful, otherwise
there are recomputed from scratch for each new z.
Y  z  ωz + z22
σ(z, L) = z 1− e 2ω .
∗
ω
ω∈L

It is also possible to directly input L = [ω1 , ω2 ], or an elliptic curve E as given by ellinit

(L = E.omega).
? w = ellperiods([1,I], 1);
? ellsigma(w, 1/2)
%2 = 0.47494937998792065033250463632798296855
? E = ellinit([1,0]);
? ellsigma(E) \\ at ’x, implicitly at default seriesprecision
%4 = x + 1/60*x^5 - 1/10080*x^9 - 23/259459200*x^13 + O(x^17)
If flag = 1, computes an arbitrary determination of log(σ(z)).
The library syntax is GEN ellsigma(GEN L, GEN z = NULL, long flag, long prec).

3.15.73 ellsub(E, z1 , z2 ). Difference of the points z1 and z2 on the elliptic curve corresponding
to E.
The library syntax is GEN ellsub(GEN E, GEN z1, GEN z2).

3.15.74 elltamagawa(E). The object E being an elliptic curve over a number field, returns the
global Tamagawa number of the curve (including the factor at infinite places).
? e = ellinit([1, -1, 1, -3002, 63929]); \\ curve "90c6" from elldata
? elltamagawa(e)
%2 = 288
? [elllocalred(e,p)[4] | p<-[2,3,5]]
%3 = [6, 4, 6]
? vecprod(%) \\ since e.disc > 0 the factor at infinity is 2
%4 = 144
The library syntax is GEN elltamagawa(GEN E).

507
3.15.75 elltaniyama(E, {n = seriesprecision}). Computes the modular parametrization of the el-
liptic curve E/Q, where E is an ell structure as output by ellinit. This returns a two-component
vector [u, v] of power series, given to n significant terms (seriesprecision by default), charac-
terized by the following two properties. First the point (u, v) satisfies the equation of the elliptic
curve. Second, let N be the conductor of E and Φ : X0 (N ) → E be a modular parametrization;
the pullback by Φ of the Néron differential du/(2v + a1 u + a3 ) is equal to 2iπf (z)dz, a holomor-
phic differential form. The variable used in the power series for u and v is x, which is implicitly
understood to be equal to exp(2iπz).
The algorithmPassumes that E is a strong Weil curve and that the Manin constant is equal to
1: in fact, f (x) = n>0 ellak(E, n)xn .
The library syntax is GEN elltaniyama(GEN E, long precdl).

3.15.76 elltatepairing(E, P, Q, m). Let E be an elliptic curve defined over a finite field k and
m ≥ 1 be an integer. This function computes the (nonreduced) Tate pairing of the points P and Q
on E, where P is an m-torsion point. More precisely, let fm,P denote a Miller function with divisor
m[P ] − m[OE ]; the algorithm returns fm,P (Q) ∈ k ∗ /(k ∗ )m .
The library syntax is GEN elltatepairing(GEN E, GEN P, GEN Q, GEN m).

3.15.77 elltors(E). If E is an elliptic curve defined over a number field or a finite field, outputs
the torsion subgroup of E as a 3-component vector [t,v1,v2], where t is the order of the torsion
group, v1 gives the structure of the torsion group as a product of cyclic groups (sorted by decreasing
order), and v2 gives generators for these cyclic groups. E must be an ell structure as output by
ellinit.
? E = ellinit([-1,0]);
? elltors(E)
%1 = [4, [2, 2], [[0, 0], [1, 0]]]
Here, the torsion subgroup is isomorphic to Z/2Z × Z/2Z, with generators [0, 0] and [1, 0].
The library syntax is GEN elltors(GEN E).

3.15.78 elltwist(E, {P }). Returns the coefficients [a1 , a2 , a3 , a4 , a6 ] of the twist of the elliptic
curve E by the quadratic extension of the coefficient ring defined by P (when P is a polynomial)
or quadpoly(P) when P is an integer. If E is defined over a finite field, then P can be omitted,
in which case a random model of the unique nontrivial twist is returned. If E is defined over a
number field, the model should be replaced by a minimal model (if one exists).
Example: Twist by discriminant −3:
? elltwist(ellinit([0,a2,0,a4,a6]),-3)
%1 = [0,-3*a2,0,9*a4,-27*a6]
Twist by the Artin-Schreier extension given by x2 + x + T in characteristic 2:
? lift(elltwist(ellinit([a1,a2,a3,a4,a6]*Mod(1,2)),x^2+x+T))
%1 = [a1,a2+a1^2*T,a3,a4,a6+a3^2*T]
Twist of an elliptic curve defined over a finite field:
? E=ellinit([1,7]*Mod(1,19));lift(elltwist(E))
%1 = [0,0,0,11,12]
The library syntax is GEN elltwist(GEN E, GEN P = NULL).

508
3.15.79 ellweilcurve(E, {&ms}). If E 0 is an elliptic curve over Q, let LE 0 be the sub-Z-module
of HomΓ0 (N ) (∆0 , Q) attached to E 0 (It is given by x[3] if [M, x] = msfromell(E 0 ).)
On the other hand, if N is the conductor of E and f is the modular form for Γ0 (N ) attached
to E, let Lf be the lattice of the f -component of HomΓ0 (N ) (∆0 , Q) given by the elements φ such
that φ({0, γ −1 0}) ∈ Z for all γ ∈ Γ0 (N ) (see mslattice).
Let E 0 run through the isomorphism classes of elliptic curves isogenous to E as given by
ellisomat (and in the same order). This function returns a pair [vE,vS] where vE contains
minimal models for the E 0 and vS contains the list of Smith invariants for the lattices LE 0 in Lf .
The function also accepts the output of ellisomat, i.e. the isogeny class. If the optional argument
ms is present, it contains the output of msfromell(vE, 0), i.e. the new modular symbol space M
of level N and a vector of triples [x+ , x− , L] attached to each curve E 0 .
In particular, the strong Weil curve amongst the curves isogenous to E is the one whose Smith
invariants are [c, c], where c is the Manin constant, conjecturally equal to 1.
? E = ellinit("11a3");
? [vE, vS] = ellweilcurve(E);
? [n] = [ i | i<-[1..#vS], vS[i]==[1,1] ] \\ lattice with invariant [1,1]
%3 = [2]
? ellidentify(vE[n]) \\ ... corresponds to strong Weil curve
%4 = [["11a1", [0, -1, 1, -10, -20], []], [1, 0, 0, 0]]
? [vE, vS] = ellweilcurve(E, &ms); \\ vE,vS are as above
? [M, vx] = ms; msdim(M) \\ ... but ms contains more information
%6 = 3
? #vx
%7 = 3
? vx[1]
%8 = [[1/25, -1/10, -1/10]~, [0, 1/2, -1/2]~, [1/25,0; -3/5,1; 2/5,-1]]
? forell(E, 11,11, print(msfromell(ellinit(E[1]), 1)[2]))
[1/5, -1/2, -1/2]~
[1, -5/2, -5/2]~
[1/25, -1/10, -1/10]~
The last example prints the modular symbols x+ in M + attached to the curves 11a1, 11a2 and
11a3.
The library syntax is GEN ellweilcurve(GEN E, GEN *ms = NULL).

3.15.80 ellweilpairing(E, P, Q, m). Let E be an elliptic curve defined over a finite field and m ≥ 1
be an integer. This function computes the Weil pairing of the two m-torsion points P and Q on E,
which is an alternating bilinear map. More precisely, let fm,R denote a Miller function with divisor
m[R] − m[OE ]; the algorithm returns the m-th root of unity

ε(P, Q)m · fm,P (Q)/fm,Q (P ),

where f (R) is the extended evaluation of f at the divisor [R] − [OE ] and ε(P, Q) ∈ {±1} is given
by Weil reciprocity: ε(P, Q) = 1 if and only if P, Q, OE are not pairwise distinct.
The library syntax is GEN ellweilpairing(GEN E, GEN P, GEN Q, GEN m).

509
3.15.81 ellwp(w, {z =0 x}, {flag = 0}). Computes the value at z of the Weierstrass ℘ function
attached to the lattice w as given by ellperiods. It is also possible to directly input w = [ω1 , ω2 ],
or an elliptic curve E as given by ellinit (w = E.omega).

? w = ellperiods([1,I]);
? ellwp(w, 1/2)
%2 = 6.8751858180203728274900957798105571978
? E = ellinit([1,1]);
? ellwp(E, 1/2)
%4 = 3.9413112427016474646048282462709151389

One can also compute the series expansion around z = 0:

? E = ellinit([1,0]);
? ellwp(E) \\ ’x implicitly at default seriesprecision
%5 = x^-2 - 1/5*x^2 + 1/75*x^6 - 2/4875*x^10 + O(x^14)
? ellwp(E, x + O(x^12)) \\ explicit precision
%6 = x^-2 - 1/5*x^2 + 1/75*x^6 + O(x^9)

Optional flag means 0 (default): compute only ℘(z), 1: compute [℘(z), ℘0 (z)].

For instance, the Dickson elliptic functions sm and sn can be implemented as follows

smcm(z) =
{ my(a, b, E = ellinit([0,-1/(4*27)])); \\ ell. invariants (g2,g3)=(0,1/27)
[a,b] = ellwp(E, z, 1);
[6*a / (1-3*b), (3*b+1)/(3*b-1)];
}
? [s,c] = smcm(0.5);
? s
%2 = 0.4898258757782682170733218609
? c
%3 = 0.9591820206453842491187464098
? s^3+c^3
%4 = 1.000000000000000000000000000
? smcm(’x + O(’x^11))
%5 = [x - 1/6*x^4 + 2/63*x^7 - 13/2268*x^10 + O(x^11),
1 - 1/3*x^3 + 1/18*x^6 - 23/2268*x^9 + O(x^10)]

The library syntax is GEN ellwp0(GEN w, GEN z = NULL, long flag, long prec). For
flag = 0, we also have GEN ellwp(GEN w, GEN z, long prec), and GEN ellwpseries(GEN E,
long v, long precdl) for the power series in variable v.

510
3.15.82 ellxn(E, n, {v =0 x}). For any affine point P = (t, u) on the curve E, we have

[n]P = (φn (P )ψn (P ) : ωn (P ) : ψn (P )3 )

for some φn , ωn , ψn in Z[a1 , a2 , a3 , a4 , a6 ][t, u] modulo the curve equation. This function returns
a pair [A, B] of polynomials in Z[a1 , a2 , a3 , a4 , a6 ][v] such that [A(t), B(t)] = [φn (P ), ψn (P )2 ] in
the function field of E, whose quotient give the abscissa of [n]P . If P is an n-torsion point, then
B(t) = 0.
? E = ellinit([17,42]); [t,u] = [114,1218];
? T = ellxn(E, 2, ’X)
%2 = [X^4 - 34*X^2 - 336*X + 289, 4*X^3 + 68*X + 168]
? [a,b] = subst(T,’X,t);
%3 = [168416137, 5934096]
? a / b == ellmul(E, [t,u], 2)[1]
%4 = 1
The library syntax is GEN ellxn(GEN E, long n, long v = -1) where v is a variable number.

3.15.83 ellzeta(w, {z =0 x}). Computes the value at z of the Weierstrass ζ function attached to
the lattice w as given by ellperiods(, 1): including quasi-periods is useful, otherwise there are
recomputed from scratch for each new z.

1 X 1
ζ(z, L) = + z2 2
.
z ∗
ω (z − ω)
ω∈L

It is also possible to directly input w = [ω1 , ω2 ], or an elliptic curve E as given by ellinit

(w = E.omega). The quasi-periods of ζ, such that

ζ(z + aω1 + bω2 ) = ζ(z) + aη1 + bη2

for integers a and b are obtained as ηi = 2ζ(ωi /2). Or using directly elleta.
? w = ellperiods([1,I],1);
? ellzeta(w, 1/2)
%2 = 1.5707963267948966192313216916397514421
? E = ellinit([1,0]);
? ellzeta(E, E.omega[1]/2)
%4 = 0.84721308479397908660649912348219163647
One can also compute the series expansion around z = 0 (the quasi-periods are useless in this case):
? E = ellinit([0,1]);
? ellzeta(E) \\ at ’x, implicitly at default seriesprecision
%4 = x^-1 + 1/35*x^5 - 1/7007*x^11 + O(x^15)
? ellzeta(E, x + O(x^20)) \\ explicit precision
%5 = x^-1 + 1/35*x^5 - 1/7007*x^11 + 1/1440257*x^17 + O(x^18)

The library syntax is GEN ellzeta(GEN w, GEN z = NULL, long prec).

511
3.15.84 ellztopoint(E, z). E being an ell as output by ellinit, computes the coordinates [x, y]
on the curve E corresponding to the complex or p-adic parameter z. Hence this is the inverse
function of ellpointtoz.
• If E is defined over a p-adic field and has multiplicative reduction, then z is understood as
an element on the Tate curve Q̄∗p /q Z .
? E = ellinit([0,-1,1,0,0], O(11^5));
? [u2,u,q] = E.tate; type(u)
%2 = "t_PADIC" \\ split multiplicative reduction
? z = ellpointtoz(E, [0,0])
%3 = 3 + 11^2 + 2*11^3 + 3*11^4 + 6*11^5 + 10*11^6 + 8*11^7 + O(11^8)
? ellztopoint(E,z)
%4 = [O(11^9), O(11^9)]
? E = ellinit(ellfromj(1/4), O(2^6)); x=1/2; y=ellordinate(E,x)[1];
? z = ellpointtoz(E,[x,y]); \\ nonsplit: t_POLMOD with t_PADIC coefficients
? P = ellztopoint(E, z);
? P[1] \\ y coordinate is analogous, more complicated
%8 = Mod(O(2^4)*x + (2^-1 + O(2^5)), x^2 + (1 + 2^2 + 2^4 + 2^5 + O(2^7)))
• If E is defined over the complex numbers (for instance over Q), z is understood as a complex
number in C/ΛE . If the short Weierstrass equation is y 2 = 4x3 − g2 x − g3 , then [x, y] represents
the Weierstrass ℘-function and its derivative. For a general Weierstrass equation we have

x = ℘(z) − b2 /12, y = ℘0 (z)/2 − (a1 x + a3 )/2.

If z is in the lattice defining E over C, the result is the point at infinity [0].
? E = ellinit([0,1]); P = [2,3];
? z = ellpointtoz(E, P)
%2 = 3.5054552633136356529375476976257353387
? ellwp(E, z)
%3 = 2.0000000000000000000000000000000000000
? ellztopoint(E, z) - P
%4 = [2.548947057811923643 E-57, 7.646841173435770930 E-57]
? ellztopoint(E, 0)
%5 = [0] \\ point at infinity
The library syntax is GEN pointell(GEN E, GEN z, long prec).

3.15.85 genus2red(PQ, {p}). Let P Q be a polynomial P , resp. a vector [P, Q] of polynomials,

with rational coefficients. Determines the reduction at p > 2 of the (proper, smooth) genus 2 curve
C/Q, defined by the hyperelliptic equation y 2 = P (x), resp. y 2 + Q(x) ∗ y = P (x). (The special
fiber Xp of the minimal regular model X of C over Z.)
If p is omitted, determines the reduction type for all (odd) prime divisors of the discriminant.
This function was rewritten from an implementation of Liu’s algorithm by Cohen and Liu (1994),
genus2reduction-0.3, see http://www.math.u-bordeaux.fr/~liu/G2R/.

512
CAVEAT. The function interface may change: for the time being, it returns [N, FaN , T, V ] where
N is either the local conductor at p or the global conductor, FaN is its factorization, y 2 = T defines
a minimal model over Z[1/2] and V describes the reduction type at the various considered p.
Unfortunately, the program is not complete for p = 2, and we may return the odd part of the
conductor only: this is the case if the factorization includes the (impossible) term 2−1 ; if the
factorization contains another power of 2, then this is the exact local conductor at 2 and N is the
global conductor.
? default(debuglevel, 1);
? genus2red(x^6 + 3*x^3 + 63, 3)
(potential) stable reduction: [1, []]
reduction at p: [III{9}] page 184, [3, 3], f = 10
%1 = [59049, Mat([3, 10]), x^6 + 3*x^3 + 63, [3, [1, []],
["[III{9}] page 184", [3, 3]]]]
? [N, FaN, T, V] = genus2red(x^3-x^2-1, x^2-x); \\ X_1(13), global reduction
p = 13
(potential) stable reduction: [5, [Mod(0, 13), Mod(0, 13)]]
reduction at p: [I{0}-II-0] page 159, [], f = 2
? N
%3 = 169
? FaN
%4 = Mat([13, 2]) \\ in particular, good reduction at 2 !
? T
%5 = x^6 + 58*x^5 + 1401*x^4 + 18038*x^3 + 130546*x^2 + 503516*x + 808561
? V
%6 = [[13, [5, [Mod(0, 13), Mod(0, 13)]], ["[I{0}-II-0] page 159", []]]]
We now first describe the format of the vector V = Vp in the case where p was specified (local re-
duction at p): it is a triple [p, stable, red ]. The component stable = [type, vecj ] contains information
about the stable reduction after a field extension; depending on types, the stable reduction is
• 1: smooth (i.e. the curve has potentially good reduction). The Jacobian J(C) has potentially
good reduction.
• 2: an elliptic curve E with an ordinary double point; vecj contains j mod p, the modular
invariant of E. The (potential) semi-abelian reduction of J(C) is the extension of an elliptic curve
(with modular invariant j mod p) by a torus.
• 3: a projective line with two ordinary double points. The Jacobian J(C) has potentially
multiplicative reduction.
• 4: the union of two projective lines crossing transversally at three points. The Jacobian
J(C) has potentially multiplicative reduction.
• 5: the union of two elliptic curves E1 and E2 intersecting transversally at one point; vecj
contains their modular invariants j1 and j2 , which may live in a quadratic extension of Fp and need
not be distinct. The Jacobian J(C) has potentially good reduction, isomorphic to the product of
the reductions of E1 and E2 .
• 6: the union of an elliptic curve E and a projective line which has an ordinary double point,
and these two components intersect transversally at one point; vecj contains j mod p, the modular
invariant of E. The (potential) semi-abelian reduction of J(C) is the extension of an elliptic curve
(with modular invariant j mod p) by a torus.

513
 • 7: as in type 6, but the two components are both singular. The Jacobian J(C) has potentially
multiplicative reduction.
The component red = [NUtype, neron] contains two data concerning the reduction at p without
any ramified field extension.
The NUtype is a t_STR describing the reduction at p of C, following Namikawa-Ueno, The
complete classification of fibers in pencils of curves of genus two, Manuscripta Math., vol. 9,
(1973), pages 143-186. The reduction symbol is followed by the corresponding page number or
page range in this article.
The second datum neron is the group of connected components (over an algebraic closure of
Fp ) of the Néron model of J(C), given as a finite abelian group (vector of elementary divisors).

If p = 2, the red component may be omitted altogether (and replaced by [], in the case where
the program could not compute it. When p was not specified, V is the vector of all Vp , for all
considered p.

Notes about Namikawa-Ueno types.

• A lower index is denoted between braces: for instance, [I{2}-II-5] means [I 2-II-5].
• If K and K 0 are Kodaira symbols for singular fibers of elliptic curves, then [K-K 0 -m] and
0
[K -K-m] are the same.
We define a total ordering on Kodaira symbol by fixing I < I∗ < II < II∗, . . .. If the reduction
type is the same, we order by the number of components, e.g. I2 < I4 , etc. Then we normalize our
output so that K ≤ K 0 .
• [K-K 0 -−1] is [K-K 0 -α] in the notation of Namikawa-Ueno.
• The figure [2I 0-m] in Namikawa-Ueno, page 159, must be denoted by [2I 0-(m+1)].
The library syntax is GEN genus2red(GEN PQ, GEN p = NULL).

3.15.86 hyperellcharpoly(X). X being a nonsingular hyperelliptic curve defined over a finite

field, return the characteristic polynomial of the Frobenius automorphism. X can be given either by
a squarefree polynomial P such that X : y 2 = P (x) or by a vector [P, Q] such that X : y 2 + Q(x) ×
y = P (x) and Q2 + 4P is squarefree.
The library syntax is GEN hyperellcharpoly(GEN X).

3.15.87 hyperellpadicfrobenius(Q, q, n). Let X be the curve defined by y 2 = Q(x), where Q is

a polynomial of degree d over Q and q ≥ d is a prime such that X has good reduction at q. Return
1
the matrix of the Frobenius endomorphism ϕ on the crystalline module Dp (X) = Qp ⊗ HdR (X/Q)
g−1
with respect to the basis of the given model (ω, xω, . . . , x ω), where ω = dx/(2y) is the invariant
differential, where g is the genus of X (either d = 2g + 1 or d = 2g + 2). The characteristic
polynomial of ϕ is the numerator of the zeta-function of the reduction of the curve X modulo q.
The matrix is computed to absolute q-adic precision q n .
Alternatively, q may be of the form [T, p] where p is a prime, T is a polynomial with integral
coefficients whose projection to Fp [t] is irreducible, X is defined over K = Q[t]/(T ) and has good
1
reduction to the finite field Fq = Fp [t]/(T ). The matrix of ϕ on Dq (X) = Qq ⊗ HdR (X/K) is
n
computed to absolute p-adic precision p .
? M=hyperellpadicfrobenius(x^5+’a*x+1,[’a^2+1,3],10);

514
 ? liftall(M)
[48107*a + 38874 9222*a + 54290 41941*a + 8931 39672*a + 28651]
[ 21458*a + 4763 3652*a + 22205 31111*a + 42559 39834*a + 40207]
[ 13329*a + 4140 45270*a + 25803 1377*a + 32931 55980*a + 21267]
[15086*a + 26714 33424*a + 4898 41830*a + 48013 5913*a + 24088]
? centerlift(simplify(liftpol(charpoly(M))))
%8 = x^4+4*x^2+81
? hyperellcharpoly((x^5+Mod(a,a^2+1)*x+1)*Mod(1,3))
%9 = x^4+4*x^2+81
The library syntax is GEN hyperellpadicfrobenius0(GEN Q, GEN q, long n). The functions
GEN hyperellpadicfrobenius(GEN H, ulong p, long n) and GEN nfhyperellpadicfrobe-
nius(GEN H, GEN T, ulong p, long n) are also available.

3.15.88 hyperellratpoints(X, h, {flag = 0}). X being a nonsingular hyperelliptic curve given by

an rational model, return a vector containing the affine rational points on the curve of naive height
less than h.a If flag = 1, stop as soon as a point is found; return either an empty vector or a vector
containing a single point.
X is given either by a squarefree polynomial P such that X : y 2 = P (x) or by a vector [P, Q]
such that X : y 2 + Q(x)y = P (x) and Q2 + 4P is squarefree.
The parameter h can be
• an integer H: find the points [n/d, y] whose abscissas x = n/d have naive height (=
max(|n|, d)) less than H;
• a vector [N, D] with D ≤ N : find the points [n/d, y] with |n| ≤ N , d ≤ D.
• a vector [N, [D1 , D2 ]] with D1 < D2 ≤ N find the points [n/d, y] with |n| ≤ N and D1 ≤
d ≤ D2 .
The library syntax is GEN hyperellratpoints(GEN X, GEN h, long flag).

3.16 L-functions.
This section describes routines related to L-functions. We first introduce the basic concept
and notations, then explain how to represent them in GP. Let ΓR (s) = π −s/2 Γ(s/2), where Γ is
Euler’s gamma
Q function. Given d ≥ 1 and a d-tuple A = [α1 , . . . , αd ] of complex numbers, we let
γA (s) = α∈A ΓR (s + α).
Given a sequence a = (an )n≥1 of complex numbers (such that a1 = 1), a positive conductor
N ∈ Z, and a gamma factor γA as above, we consider the Dirichlet series
X
L(a, s) = an n−s
n≥1

and the attached completed function

Λ(a, s) = N s/2 γA (s) · L(a, s).

Such a datum defines an L-function if it satisfies the three following assumptions:

515
 • [Convergence] The an = O (nk1 + ) have polynomial growth, equivalently L(s) converges
absolutely in some right half-plane <(s) > k1 + 1.
• [Analytic continuation] L(s) has a meromorphic continuation to the whole complex plane
with finitely many poles.
• [Functional equation] There exist an integer k, a complex number  (usually of modulus 1),
and an attached sequence a∗ defining both an L-function L(a∗ , s) satisfying the above two assump-
tions and a completed function Λ(a∗ , s) = N s/2 γA (s) · L(a∗ , s), such that

Λ(a, k − s) = Λ(a∗ , s)

for all regular points.

More often than not in number theory we have a∗ = a (which forces || = 1), but this needs
not be the case. If a is a real sequence and a = a∗ , we say that L is self-dual . We do not assume
that the an are multiplicative, nor equivalently that L(s) has an Euler product.
Remark. Of course, a determines the L-function, but the (redundant) datum a, a∗ , A, N, k, 
describes the situation in a form more suitable for fast computations; knowing the polar part r of
Λ(s) (a rational function such that Λ − r is holomorphic) is also useful. A subset of these, including
only finitely many an -values will still completely determine L (in suitable families), and we provide
routines to try and compute missing invariants from whatever information is available.
Important Caveat. The implementation assumes that the implied constants in the O are small.
In our generic framework, it is impossible to return proven results without more detailed information
about the L function. The intended use of the L-function package is not to prove theorems, but to
experiment and formulate conjectures, so all numerical results should be taken with a grain of salt.
One can always increase realbitprecision and recompute: the difference estimates the actual
absolute error in the original output.
Note. The requested precision has a major impact on runtimes. Because of this, most L-function
routines, in particular lfun itself, specify the requested precision in bits, not in decimal digits.
This is transparent for the user once realprecision or realbitprecision are set. We advise to
manipulate precision via realbitprecision as it allows finer granularity: realprecision increases
by increments of 64 bits, i.e. 19 decimal digits at a time.

3.16.1 Theta functions.

Given an L-function as above, we define an attached theta function via Mellin inversion: for
any positive real t > 0, we let
Z
1
θ(a, t) := t−s Λ(s) ds
2πi <(s)=c

where c is any positive real number c > k1 + 1 such that c + <(a) > 0 for all a ∈ A. In fact, we
have Z
X 1
θ(a, t) = 1/2
an K(nt/N ) where K(t) := t−s γA (s) ds.
2πi <(s)=c
n≥1

Note that this function is analytic and actually makes sense for complex t, such that <(t2/d ) > 0,
i.e. in a cone containing the positive real half-line. The functional equation for Λ translates into

θ(a, 1/t) − tk θ(a∗ , t) = PΛ (t),

516
where PΛ is an explicit polynomial in t and log t given by the Taylor development of the polar part
of Λ: there are no log’s if all poles are simple, and P = 0 if Λ is entire. The values θ(t) are generally
easier to compute than the L(s), and this functional equation provides a fast way to guess possible
values for missing invariants in the L-function definition.

3.16.2 Data structures describing L and theta functions.

We have 3 levels of description:
• an Lmath is an arbitrary description of the underlying mathematical situation (to which e.g.,
we associate the ap as traces of Frobenius elements); this is done via constructors to be described
in the subsections below.
• an Ldata is a computational description of situation, containing the complete datum
(a, a∗ , A, k, N, , r). Where a and a∗ describe the coefficients (given n, b we must be able to compute
[a1 , . . . , an ] with bit accuracy b), A describes the Euler factor, the (classical) weight is k, N is the
conductor, and r describes the polar part of L(s). This is obtained via the function lfuncreate.
N.B. For motivic L-functions, the motivic weight w is w = k − 1; but we also support nonmotivic
L-functions.

Technical note. When some components of an Ldata cannot be given exactly, usually r or , the
Ldata may be given as a closure. When evaluated at a given precision, the closure must return all
components as exact data or floating point numbers at the requested precision, see ??lfuncreate.
The reason for this technicality is that the accuracy to which we must compute is not bounded a
priori and unknown at this stage: it depends on the domain where we evaluate the L-function.
• an Linit contains an Ldata and everything needed for fast numerical computations. It
specifies the functions to be considered (either L(j) (s) or θ(j) (t) for derivatives of order j ≤ m, where
m is now fixed) and specifies a domain which limits the range of arguments (t or s, respectively to
certain cones and rectangular regions) and the output accuracy. This is obtained via the functions
lfuninit or lfunthetainit.
All the functions which are specific to L or theta functions share the prefix lfun. They take
as first argument either an Lmath, an Ldata, or an Linit. If a single value is to be computed, this
makes no difference, but when many values are needed (e.g. for plots or when searching for zeros),
one should first construct an Linit attached to the search range and use it in all subsequent calls.
If you attempt to use an Linit outside the range for which it was initialized, a warning is issued,
because the initialization is performed again, a major inefficiency:
? Z = lfuncreate(1); \\ Riemann zeta
? L = lfuninit(Z, [1/2, 0, 100]); \\ zeta(1/2+it), |t| < 100
? lfun(L, 1/2) \\ OK, within domain
%3 = -1.4603545088095868128894991525152980125
? lfun(L, 0) \\ not on critical line !
*** lfun: Warning: lfuninit: insufficient initialization.
%4 = -0.50000000000000000000000000000000000000
? lfun(L, 1/2, 1) \\ attempt first derivative !
*** lfun: Warning: lfuninit: insufficient initialization.
%5 = -3.9226461392091517274715314467145995137
For many L-functions, passing from Lmath to an Ldata is inexpensive: in that case one may
use lfuninit directly from the Lmath even when evaluations in different domains are needed. The
above example could equally have skipped the lfuncreate:

517
 ? L = lfuninit(1, [1/2, 0, 100]); \\ zeta(1/2+it), |t| < 100
In fact, when computing a single value, you can even skip lfuninit:
? L = lfun(1, 1/2, 1); \\ zeta’(1/2)
? L = lfun(1, 1+x+O(x^5)); \\ first 5 terms of Taylor development at 1
Both give the desired results with no warning.
Complexity. The implementation requires O(N (|t| + 1))1/2 coefficients an to evaluate L of con-
ductor N at s = σ + it.
We now describe the available high-level constructors, for built-in L functions.

3.16.3 Dirichlet L-functions.

Given a Dirichlet character χ : (Z/N Z)∗ → C, we let
X
L(χ, s) = χ(n)n−s .
n≥1

Only primitive characters are supported. Given a nonzero integer D, the t_INT D encodes the
function L((D0 /.), s), for the quadratic Kronecker symbol attached to the fundamental discriminant
D0 = coredisc(D). This includes Riemann ζ function via the special case D = 1.
More general characters can be represented in a variety of ways:
• via Conrey notation (see znconreychar): χN (m, ·) is given as the t_INTMOD Mod(m,N).
• via a znstar structure describing the abelian group (Z/N Z)∗ , where the character is given
in terms of the znstar generators:
? G = znstar(100, 1); \\ (Z/100Z)^*
? G.cyc \\ ~ Z/20 . g1 + Z/2 . g2 for some generators g1 and g2
%2 = [20, 2]
? G.gen
%3 = [77, 51]
? chi = [a, b] \\ maps g1 to e(a/20) and g2 to e(b/2); e(x) = exp(2ipi x)
More generally, let (Z/N Z)∗ = ⊕(Z/di Z)gi be given via a znstar structure G (G.cyc gives the
di Q
and G.gen the giP ). A character χ on G is given by a row vector v = [a1 , . . . , an ] such that
χ( gini ) = exp(2πi ai ni /di ). The pair [G, v] encodes the primitive character attached to χ.
• in fact, this construction [G, m] describing a character is more general: m is also allowed to
be a Conrey label as seen above, or a Conrey logarithm (see znconreylog), and the latter format
is actually the fastest one. Instead of a single character as above, one may use the construction
lfuncreate([G, vchi]) where vchi is a nonempty vector of characters of the same conductor
(more precisely, whose attached primitive characters have the same conductor) and same parity.
The function is then vector-valued, where the values are ordered as the characters in vchi. Conrey
labels cannot be used in this last format because of the need to distinguish a single character given
by a row vector of integers and a vector of characters given by their labels: use znconreylog(G,i)
first to convert a label to Conrey logarithm.
• it is also possible to view Dirichlet characters as Hecke characters over K = Q (see below),
for a modulus [N, [1]] but this is both more complicated and less efficient.
In all cases, a nonprimitive character is replaced by the attached primitive character.

518
3.16.4 Hecke L-functions.
The Dedekind zeta function of a number field K = Q[X]/(T ) is encoded either by the defining
polynomial T , or any absolute number fields structure (preferably at least a bnf ).
Given a finite order Hecke character χ : Clf (K) → C, we let
X
−s
L(χ, s) = χ(A) NK/Q A .
A⊂OK

Let Clf (K) = ⊕(Z/di Z)gi given by a bnr structure with generators: the di are given by
K.cyc and the gi by K.gen. Q A character χ Pon the ray class group is given by a row vector v =
[a1 , . . . , an ] such that χ( gini ) = exp(2πi ai ni /di ). The pair [bnr , v] encodes the primitive
character attached to χ.

? K = bnfinit(x^2-60);
? Cf = bnrinit(K, [7, [1,1]], 1); \\ f = 7 oo_1 oo_2
? Cf.cyc
%3 = [6, 2, 2]
? Cf.gen
%4 = [[2, 1; 0, 1], [22, 9; 0, 1], [-6, 7]~]
? lfuncreate([Cf, [1,0,0]]); \\ χ(g1 ) = ζ6 , χ(g2 ) = χ(g3 ) = 1
Dirichlet characters on (Z/N Z)∗ are a special case, where K = Q:

? Q = bnfinit(x);
? Cf = bnrinit(Q, [100, [1]]); \\ for odd characters on (Z/100Z)*
For even characters, replace by bnrinit(K, N). Note that the simpler direct construction in the
previous section will be more efficient. Instead of a single character as above, one may use the
construction lfuncreate([Cf, vchi]) where vchi is a nonempty vector of characters of the same
conductor (more precisely, whose attached primitive characters have the same conductor). The
function is then vector-valued, where the values are ordered as the characters in vchi.

3.16.5 Artin L functions.

Given a Galois number field N/Q with group G = galoisinit(N ), a representation ρ of G
over the cyclotomic field Q(ζn ) is specified by the matrices giving the images of G.gen by ρ. The
corresponding Artin L function is created using lfunartin.

P = quadhilbert(-47); \\ degree 5, Galois group D_5

N = nfinit(nfsplitting(P)); \\ Galois closure
G = galoisinit(N);
[s,t] = G.gen; \\ order 5 and 2
L = lfunartin(N,G, [[a,0;0,a^-1],[0,1;1,0]], 5); \\ irr. degree 2
In the above, the polynomial variable (here a) represents ζ5 := exp(2iπ/5) and the two matrices
give the images of s and t. Here, priority of a must be lower than the priority of x.

519
3.16.6 L-functions of algebraic varieties.
L-function of elliptic curves over number fields are supported.

? E = ellinit([1,1]);
? L = lfuncreate(E); \\ L-function of E/Q
? E2 = ellinit([1,a], nfinit(a^2-2));
? L2 = lfuncreate(E2); \\ L-function of E/Q(sqrt(2))
L-function of hyperelliptic genus-2 curve can be created with lfungenus2. To create the L
function of the curve y 2 + (x3 + x2 + 1)y = x2 + x:

? L = lfungenus2([x^2+x, x^3+x^2+1]);
Currently, the model needs to be minimal at 2, and if the conductor is even, its valuation at
2 might be incorrect (a warning is issued).

3.16.7 Eta quotients / Modular forms.

An eta quotient is created by applying lfunetaquo to a matrix with 2 columns [m, rm ] repre-
senting
Y
f (τ ) := η(mτ )rm .
m

It is currently
P assumed that f is a self-dual cuspidal form on Γ0 (N ) for some N . For instance, the
L-function τ (n)n−s attached to Ramanujan’s ∆ function is encoded as follows

? L = lfunetaquo(Mat([1,24]));
? lfunan(L, 100) \\ first 100 values of tau(n)
More general modular forms defined by modular symbols will be added later.

3.16.8 Low-level Ldata format.

When no direct constructor is available, you can still input an L function directly by supplying
[a, a∗ , A, k, N, , r] to lfuncreate (see ??lfuncreate for details).
It is strongly suggested to first check consistency of the created L-function:

? L = lfuncreate([a, as, A, k, N, eps, r]);

? lfuncheckfeq(L) \\ check functional equation

3.16.9 lfun(L, s, {D = 0}). Compute the L-function value L(s), or if D is set, the derivative of
order D at s. The parameter L is either an Lmath, an Ldata (created by lfuncreate, or an Linit
(created by lfuninit), preferrably the latter if many values are to be computed.
The argument s is also allowed to be a power series; for instance, if s = α + x + O(xn ), the
function returns the Taylor expansion of order n around α. The result is given with absolute error
less than 2−B , where B = realbitprecision.

520
Caveat. The requested precision has a major impact on runtimes. It is advised to manipulate
precision via realbitprecision as explained above instead of realprecision as the latter allows
less granularity: realprecision increases by increments of 64 bits, i.e. 19 decimal digits at a time.

? lfun(x^2+1, 2) \\ Lmath: Dedekind zeta for Q(i) at 2

%1 = 1.5067030099229850308865650481820713960
? L = lfuncreate(ellinit("5077a1")); \\ Ldata: Hasse-Weil zeta function
? lfun(L, 1+x+O(x^4)) \\ zero of order 3 at the central point
%3 = 0.E-58 - 5.[...] E-40*x + 9.[...] E-40*x^2 + 1.7318[...]*x^3 + O(x^4)
\\ Linit: zeta(1/2+it), |t| < 100, and derivative
? L = lfuninit(1, [100], 1);
? T = lfunzeros(L, [1,25]);
%5 = [14.134725[...], 21.022039[...]]
? z = 1/2 + I*T[1];
? abs( lfun(L, z) )
%7 = 8.7066865533412207420780392991125136196 E-39
? abs( lfun(L, z, 1) )
%8 = 0.79316043335650611601389756527435211412 \\ simple zero

The library syntax is GEN lfun0(GEN L, GEN s, long D, long bitprec).

3.16.10 lfunabelianrelinit(bnfL, bnfK , polrel , sdom, {der = 0}). Returns the Linit structure
attached to the Dedekind zeta function of the number field L (see lfuninit), given a subfield K
such that L/K is abelian. Here polrel defines L over K, as usual with the priority of the variable
of bnfK lower than that of polrel. sdom and der are as in lfuninit.

? D = -47; K = bnfinit(y^2-D);
? rel = quadhilbert(D); T = rnfequation(K.pol, rel); \\ degree 10
? L = lfunabelianrelinit(T,K,rel, [2,0,0]); \\ at 2
time = 84 ms.
? lfun(L, 2)
%4 = 1.0154213394402443929880666894468182650
? lfun(T, 2) \\ using parisize > 300MB
time = 652 ms.
%5 = 1.0154213394402443929880666894468182656

As the example shows, using the (abelian) relative structure is more efficient than a direct compu-
tation. The difference becomes drastic as the absolute degree increases while the subfield degree
remains constant.

The library syntax is GEN lfunabelianrelinit(GEN bnfL, GEN bnfK, GEN polrel, GEN
sdom, long der, long bitprec).

521
3.16.11 lfunan(L, n). Compute the first n terms of the Dirichlet series attached to the L-function
given by L (Lmath, Ldata or Linit).
? lfunan(1, 10) \\ Riemann zeta
%1 = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
? lfunan(5, 10) \\ Dirichlet L-function for kronecker(5,.)
%2 = [1, -1, -1, 1, 0, 1, -1, -1, 1, 0]

The library syntax is GEN lfunan(GEN L, long n, long prec).

3.16.12 lfunartin(nf , gal , rho, n). Returns the Ldata structure attached to the Artin L-function
provided by the representation ρ of the Galois group of the extension K/Q, defined over the
cyclotomic field Q(ζn ), where nf is the nfinit structure attached to K, gal is the galoisinit structure
attached to K/Q, and rho is given either
• by the values of its character on the conjugacy classes (see galoisconjclasses and galois-
chartable)
• or by the matrices that are the images of the generators gal .gen.
Cyclotomic numbers in rho are represented by polynomials, whose variable is understood as
the complex number exp(2iπ/n).
In the following example we build the Artin L-functions attached to the two irreducible degree
2 representations of the dihedral
√ group D10 defined over Q(ζ5 ), for the extension H/Q where H
is the Hilbert class field of Q( −47). We show numerically some identities involving Dedekind ζ
functions and Hecke L series.
? P = quadhilbert(-47)
%1 = x^5 + 2*x^4 + 2*x^3 + x^2 - 1
? N = nfinit(nfsplitting(P));
? G = galoisinit(N); \\ D_10
? [T,n] = galoischartable(G);
? T \\ columns give the irreducible characters
%5 =
[1 1 2 2]
[1 -1 0 0]
[1 1 -y^3 - y^2 - 1 y^3 + y^2]
[1 1 y^3 + y^2 -y^3 - y^2 - 1]
? n
%6 = 5
? L2 = lfunartin(N,G, T[,2], n);
? L3 = lfunartin(N,G, T[,3], n);
? L4 = lfunartin(N,G, T[,4], n);
? s = 1 + x + O(x^4);
? lfun(-47,s) - lfun(L2,s)
%11 ~ 0
? lfun(1,s)*lfun(-47,s)*lfun(L3,s)^2*lfun(L4,s)^2 - lfun(N,s)
%12 ~ 0
? lfun(1,s)*lfun(L3,s)*lfun(L4,s) - lfun(P,s)

522
 %13 ~ 0
? bnr = bnrinit(bnfinit(x^2+47),1,1);
? bnr.cyc
%15 = [5] \\ Z/5Z: 4 nontrivial ray class characters
? lfun([bnr,[1]], s) - lfun(L3, s)
%16 ~ 0
? lfun([bnr,[2]], s) - lfun(L4, s)
%17 ~ 0
? lfun([bnr,[3]], s) - lfun(L3, s)
%18 ~ 0
? lfun([bnr,[4]], s) - lfun(L4, s)
%19 ~ 0
The first identity identifies the nontrivial abelian character with (−47, ·); the second is the
factorization of the regular representation of D10 ; the third is the factorization of the natural
representation of D10 ⊂ S5 ; and the final four are the expressions of the degree 2 representations
as induced from degree 1 representations.
The library syntax is GEN lfunartin(GEN nf, GEN gal, GEN rho, long n, long bitprec)
.

3.16.13 lfuncheckfeq(L, {t}). Given the data attached to an L-function (Lmath, Ldata or Linit),
check whether the functional equation is satisfied. This is most useful for an Ldata constructed
“by hand”, via lfuncreate, to detect mistakes.
If the function has poles, the polar part must be specified. The routine returns a bit accuracy
b such that |w − ŵ| < 2b , where w is the root number contained in data, and

ŵ = θ(1/t)t−k /θ(t)

is a computed value derived from the assumed functional equation. If the parameter t is omitted,
we try random samples on the real line in the segment [1, 1.25]. Of course, a large negative value
of the order of realbitprecision is expected but if θ is very small all over the sampled segment,
you should first increase realbitprecision by − log2 |θ(t)| (which is positive if θ is small) to get
a meaningful result.
If t is given, it should be close to the unit disc for efficiency and such that θ(t) 6= 0. We
then check the functional equation at that t. Again, if θ(t) is very small, you should first increase
realbitprecision to get a useful result.
? \pb 128 \\ 128 bits of accuracy
? default(realbitprecision)
%1 = 128
? L = lfuncreate(1); \\ Riemann zeta
? lfuncheckfeq(L)
%3 = -124
i.e. the given data is consistent to within 4 bits for the particular check consisting of estimating
the root number from all other given quantities. Checking away from the unit disc will either fail
with a precision error, or give disappointing results (if θ(1/t) is large it will be computed with a
large absolute error)
? lfuncheckfeq(L, 2+I)

523
 %4 = -115
? lfuncheckfeq(L,10)
*** at top-level: lfuncheckfeq(L,10)
*** ^------------------
*** lfuncheckfeq: precision too low in lfuncheckfeq.
The library syntax is long lfuncheckfeq(GEN L, GEN t = NULL, long bitprec).

3.16.14 lfunconductor(L, {setN = 10000}, {flag = 0}). Compute the conductor of the given L-
function (if the structure contains a conductor, it is ignored). Two methods are available, depending
on what we know about the conductor, encoded in the setN parameter:
• setN is a scalar: we know nothing but expect that the conductor lies in the interval [1, setN].
If flag is 0 (default), give either the conductor found as an integer, or a vector (possibly empty)
of conductors found. If flag is 1, same but give the computed floating point approximations to the
conductors found, without rounding to integers. It flag is 2, give all the conductors found, even
those far from integers.

Caveat. This is a heuristic program and the result is not proven in any way:
? L = lfuncreate(857); \\ Dirichlet L function for kronecker(857,.)
? \p19
realprecision = 19 significant digits
? lfunconductor(L)
%2 = [17, 857]
? lfunconductor(L,,1) \\ don’t round
%3 = [16.99999999999999999, 857.0000000000000000]
? \p38
realprecision = 38 significant digits
? lfunconductor(L)
%4 = 857
Increasing setN or increasing realbitprecision slows down the program but gives better accuracy
for the result. This algorithm should only be used if the primes dividing the conductor are unknown,
which is uncommon.
• setN is a vector of possible conductors; for instance of the form D1 * divisors(D2), where
D1 is the known part of the conductor and D2 is a multiple of the contribution of the bad primes.
In that case, flag is ignored and the routine uses lfuncheckfeq. It returns [N, e] where N is
the best conductor in the list and e is the value of lfuncheckfeq for that N . When no suitable
conductor exist or there is a tie among best potential conductors, return the empty vector [].
? E = ellinit([0,0,0,4,0]); /* Elliptic curve y^2 = x^3+4x */
? E.disc \\ |disc E| = 2^12
%2 = -4096
\\ create Ldata by hand. Guess that root number is 1 and conductor N
? L(N) = lfuncreate([n->ellan(E,n), 0, [0,1], 2, N, 1]);
\\ lfunconductor ignores conductor = 1 in Ldata !
? lfunconductor(L(1), divisors(E.disc))
%5 = [32, -127]
? fordiv(E.disc, d, print(d,": ",lfuncheckfeq(L(d)))) \\ direct check

524
 1: 0
2: 0
4: -1
8: -2
16: -3
32: -127
64: -3
128: -2
256: -2
512: -1
1024: -1
2048: 0
4096: 0
The above code assumed that root number was 1; had we set it to −1, none of the lfuncheckfeq
values would have been acceptable:
? L2 = lfuncreate([n->ellan(E,n), 0, [0,1], 2, 0, -1]);
? lfunconductor(L2, divisors(E.disc))
%7 = []
The library syntax is GEN lfunconductor(GEN L, GEN setN = NULL, long flag, long
bitprec).

3.16.15 lfuncost(L, {sdom}, {der = 0}). Estimate the cost of running lfuninit(L,sdom,der)
at current bit precision. Returns [t, b], to indicate that t coefficients an will be computed, as well
as t values of gammamellininv, all at bit accuracy b. A subsequent call to lfun at s evaluates a
polynomial of degree t at exp(hs) for some real parameter h, at the same bit accuracy b. If L is
already an Linit, then sdom and der are ignored and are best left omitted; the bit accuracy is also
inferred from L: in short we get an estimate of the cost of using that particular Linit.
? \pb 128
? lfuncost(1, [100]) \\ for zeta(1/2+I*t), |t| < 100
%1 = [7, 242] \\ 7 coefficients, 242 bits
? lfuncost(1, [1/2, 100]) \\ for zeta(s) in the critical strip, |Im s| < 100
%2 = [7, 246] \\ now 246 bits
? lfuncost(1, [100], 10) \\ for zeta(1/2+I*t), |t| < 100
%3 = [8, 263] \\ 10th derivative increases the cost by a small amount
? lfuncost(1, [10^5])
%3 = [158, 113438] \\ larger imaginary part: huge accuracy increase
? L = lfuncreate(polcyclo(5)); \\ Dedekind zeta for Q(zeta_5)
? lfuncost(L, [100]) \\ at s = 1/2+I*t), |t| < 100
%5 = [11457, 582]
? lfuncost(L, [200]) \\ twice higher
%6 = [36294, 1035]
? lfuncost(L, [10^4]) \\ much higher: very costly !
%7 = [70256473, 45452]
? \pb 256
? lfuncost(L, [100]); \\ doubling bit accuracy
%8 = [17080, 710]

525
In fact, some L functions can be factorized algebraically by the lfuninit call, e.g. the Dedekind
zeta function of abelian fields, leading to much faster evaluations than the above upper bounds. In
that case, the function returns a vector of costs as above for each individual function in the product
actually evaluated:
? L = lfuncreate(polcyclo(5)); \\ Dedekind zeta for Q(zeta_5)
? lfuncost(L, [100]) \\ a priori cost
%2 = [11457, 582]
? L = lfuninit(L, [100]); \\ actually perform all initializations
? lfuncost(L)
%4 = [[16, 242], [16, 242], [7, 242]]
The Dedekind function of this abelian quartic field is the product of four Dirichlet L-functions
attached to the trivial character, a nontrivial real character and two complex conjugate characters.
The nontrivial characters happen to have the same conductor (hence same evaluation costs), and
correspond to two evaluations only since the two conjugate characters are evaluated simultaneously.
For a total of three L-functions evaluations, which explains the three components above. Note that
the actual cost is much lower than the a priori cost in this case.
The library syntax is GEN lfuncost0(GEN L, GEN sdom = NULL, long der, long bitprec)
. Also available is GEN lfuncost(GEN L, GEN dom, long der, long bitprec) when L is not an
Linit; the return value is a t_VECSMALL in this case.

3.16.16 lfuncreate(obj ). This low-level routine creates Ldata structures, needed by lfun functions,
describing an L-function and its functional equation. We advise using a high-level constructor when
one is available, see ??lfun, and this function accepts them:
? L = lfuncreate(1); \\ Riemann zeta
? L = lfuncreate(5); \\ Dirichlet L-function for quadratic character (5/.)
? L = lfuncreate(x^2+1); \\ Dedekind zeta for Q(i)
? L = lfuncreate(ellinit([0,1])); \\ L-function of E/Q: y^2=x^3+1
One can then use, e.g., lfun(L,s) to directly evaluate the respective L-functions at s, or
lfuninit(L, [c,w,h] to initialize computations in the rectangular box <(s − c) ≤ w, =(s) ≤ h.
We now describe the low-level interface, used to input nonbuiltin L-functions. The input is now
a 6 or 7 component vector V = [a, astar, V ga, k, N, eps, poles], whose components are as follows:
• V[1]=a encodes the Dirichlet series coefficients (an ). The preferred format is a closure of
arity 1: n->vector(n,i,a(i)) giving the vector of the first n coefficients. The closure is allowed
to return a vector of more than n coefficients (only the first n will be considered) or even less than
n, in which case loss of accuracy will occur and a warning that #an is less than expected is issued.
This allows to precompute and store a fixed large number of Dirichlet coefficients in a vector v and
use the closure n->v, which does not depend on n. As a shorthand for this latter case, you can
input the vector v itself instead of the closure.
? z = lfuncreate([n->vector(n,i,1), 1, [0], 1, 1, 1, 1]); \\ Riemann zeta
? lfun(z,2) - Pi^2/6
%2 = -5.877471754111437540 E-39
A second format is limited to L-functions affording an Euler product. It is a closure of arity
2 (p,d)->F(p) giving the local factor Lp (X) at p as a rational function, to be evaluated at p−s
as in direuler; d is set to logint(n, p) + 1, where n is the total number of Dirichlet coefficients

526
(a1 , . . . , an ) that will be computed. In other words, the smallest integer d such that pd > n. This
parameter d allows to compute only part of Lp when p is large and Lp expensive to compute:
any polynomial (or t_SER) congruent to Lp modulo X d is acceptable since only the coefficients
of X 0 , . . . , X d−1 are needed to expand the Dirichlet series. The closure can of course ignore this
parameter:
? z = lfuncreate([(p,d)->1/(1-x), 1, [0], 1, 1, 1, 1]); \\ Riemann zeta
? lfun(z,2) - Pi^2/6
%4 = -5.877471754111437540 E-39
One can describe separately the generic local factors coefficients and the bad local factors by setting
dir = [F, Lbad ], were Lbad = [[p1 , Lp1 ], . . . , [pk , Lpk ]], where F describes the generic local factors as
above, except that when p = pi for some i ≤ k, the coefficient ap is directly set to Lpi instead of
calling F .
N = 15;
E = ellinit([1, 1, 1, -10, -10]); \\ = "15a1"
F(p,d) = 1 / (1 - ellap(E,p)*’x + p*’x^2);
Lbad = [[3, 1/(1+’x)], [5, 1/(1-’x)]];
L = lfuncreate([[F,Lbad], 0, [0,1], 2, N, ellrootno(E)]);
Of course, in this case, lfuncreate(E) is preferable!
• V[2]=astar is the Dirichlet series coefficients of the dual function, encoded as a above. The
sentinel values 0 and 1 may be used for the special cases where a = a∗ and a = a∗ , respectively.
• V[3]=Vga is the vector of αj such that the gamma factor of the L-function is equal to
Y
γA (s) = ΓR (s + αj ),
1≤j≤d

where ΓR (s) = π −s/2 Γ(s/2). This same syntax is used in the gammamellininv functions. In
particular the length d of Vga is the degree of the L-function. In the present implementation, the
αj are assumed to be exact rational numbers. However when calling theta functions with complex
(as opposed to real) arguments, determination problems occur which may give wrong results when
the αj are not integral.
• V[4]=k is a positive integer k. The functional equation relates values at s and k − s. For
instance, for an Artin L-series such as a Dedekind zeta function we have k = 1, for an elliptic curve
k = 2, and for a modular form, k is its weight. For motivic L-functions, the motivic weight w is
w = k − 1.
By default we assume that an = O (nk1 + ), where k1 = w and even k1 = w/2 when the L
function has no pole (Ramanujan-Petersson). If this is not the case, you can replace the k argument
by a vector [k, k1 ], where k1 is the upper bound you can assume.
• V[5]=N is the conductor, an integer N ≥ 1, such that Λ(s) = N s/2 γA (s)L(s) with γA (s) as
above.
• V[6]=eps is the root number ε, i.e., the complex number (usually of modulus 1) such that
Λ(a, k − s) = εΛ(a∗ , s).
• The last optional component V[7]=poles encodes the poles of the L or Λ-functions, and is
omitted if they have no poles. A polar part is given by a list of 2-component vectors [β, Pβ (x)], where

527
β is a pole and the power series Pβ (x) describes the attached polar part, such that L(s) − Pβ (s − β)
is holomorphic in a neighbourhood of β. For instance Pβ = r/x + O(1) for a simple pole at β or
r1 /x2 + r2 /x + O(1) for a double pole. The type of the list describing the polar part allows to
distinguish between L and Λ: a t_VEC is attached to L, and a t_COL is attached to Λ. Unless
a = a∗ (coded by astar equal to 0 or 1), it is mandatory to specify the polar part of Λ rather
than those of L since the poles of L∗ cannot be infered from the latter ! Whereas the functional
equation allows to deduce the polar part of Λ∗ from the polar part of Λ.

Finally, if a = a∗ , we allow a shortcut to describe the frequent situation where L has at most
simple pole, at s = k, with residue r a complex scalar: you may then input poles = r. This value
r can be set to 0 if unknown and it will be computed.

When one component is not exact. Alternatively, obj can be a closure of arity 0 returning
the above vector to the current real precision. This is needed if some components are not available
exactly but only through floating point approximations. The closure allows algorithms to recompute
them to higher accuracy when needed. Compare

? Ld1() = [n->lfunan(Mod(2,7),n),1,[0],1,7,((-13-3*sqrt(-3))/14)^(1/6)];
? Ld2 = [n->lfunan(Mod(2,7),n),1,[0],1,7,((-13-3*sqrt(-3))/14)^(1/6)];
? L1 = lfuncreate(Ld1);
? L2 = lfuncreate(Ld2);
? lfun(L1,1/2+I*200) \\ OK
%5 = 0.55943925130316677665287870224047183265 -
0.42492662223174071305478563967365980756*I
? lfun(L2,1/2+I*200) \\ all accuracy lost
%6 = 0.E-38 + 0.E-38*I

The accuracy lost in Ld2 is due to the root number being given to an insufficient precision. To see
what happens try

? Ld3() = printf("prec needed: %ld bits",getlocalbitprec());Ld1()

? L3 = lfuncreate(Ld3);
prec needed: 64 bits
? z3 = lfun(L3,1/2+I*200)
prec needed: 384 bits
%16 = 0.55943925130316677665287870224047183265 -
0.42492662223174071305478563967365980756*I

The library syntax is GEN lfuncreate(GEN obj).

3.16.17 lfundiv(L1 , L2 ). Creates the Ldata structure (without initialization) corresponding to

the quotient of the Dirichlet series L1 and L2 given by L1 and L2. Assume that vz (L1 ) ≥ vz (L2 )
at all complex numbers z: the construction may not create new poles, nor increase the order of
existing ones.

The library syntax is GEN lfundiv(GEN L1, GEN L2, long bitprec).

528
3.16.18 lfundual(L). Creates the Ldata structure (without initialization) corresponding to the
dual L-function L̂ of L. If k and ε are respectively the weight and root number of L, then the
following formula holds outside poles, up to numerical errors:

Λ(L, s) = εΛ(L̂, k − s).

? L = lfunqf(matdiagonal([1,2,3,4]));
? eps = lfunrootres(L)[3]; k = L[4];
? M = lfundual(L); lfuncheckfeq(M)
%3 = -127
? s= 1+Pi*I;
? a = lfunlambda(L,s);
? b = eps * lfunlambda(M,k-s);
? exponent(a - b)
%7 = -130
The library syntax is GEN lfundual(GEN L, long bitprec).

3.16.19 lfunetaquo(MQn ). Returns the Ldata structure attached to the L function attached to the
modular form z 7→ i=1 η(Mi,1 z)Mi,2 It is currently
P assumed that f is a self-dual cuspidal form on
Γ0 (N ) for some N . For instance, the L-function τ (n)n−s attached to Ramanujan’s ∆ function
is encoded as follows
? L = lfunetaquo(Mat([1,24]));
? lfunan(L, 100) \\ first 100 values of tau(n)
For convenience, a t_VEC is also accepted instead of a factorization matrix with a single row:
? L = lfunetaquo([1,24]); \\ same as above
The library syntax is GEN lfunetaquo(GEN M).

3.16.20 lfungenus2(F ). Returns the Ldata structure attached to the L function attached to the
genus-2 curve defined by y 2 = F (x) or y 2 + Q(x)y = P (x) if F = [P, Q]. Currently, the model
needs to be minimal at 2, and if the conductor is even, its valuation at 2 might be incorrect (a
warning is issued).
The library syntax is GEN lfungenus2(GEN F).

3.16.21 lfunhardy(L, t). Variant of the Hardy Z-function given by L, used for plotting or locating
zeros of L(k/2 + it) on the critical line. The precise definition is as follows: let k/2 be the center
of the critical strip, d be the degree, Vga = (αj )j≤d given thePgamma factors, and ε be the root
number; we set s = k/2 + it = ρeiθ and 2E = d(k/2 − 1) + <( 1≤j≤d αj ). Assume first that Λ is
self-dual, then the computed function at t is equal to

Z(t) = ε−1/2 Λ(s) · ρ−E edtθ/2 ,

which is a real function of t vanishing exactly when L(k/2 + it) does on the critical line. The
normalizing factor |s|−E edtθ/2 compensates the exponential decrease of γA (s) as t → ∞ so that
Z(t) ≈ 1. For non-self-dual Λ, the definition is the same except we drop the ε−1/2 term (which is
not well defined since it depends on the chosen dual sequence a∗ (n)): Z(t) is still of the order of 1
and still vanishes where L(k/2 + it) does, but it needs no longer be real-valued.

529
 ? T = 100; \\ maximal height
? L = lfuninit(1, [T]); \\ initialize for zeta(1/2+it), |t|<T
? \p19 \\ no need for large accuracy
? ploth(t = 0, T, lfunhardy(L,t))

Using lfuninit is critical for this particular applications since thousands of values are computed.
Make sure to initialize up to the maximal t needed: otherwise expect to see many warnings for
unsufficient initialization and suffer major slowdowns.

The library syntax is GEN lfunhardy(GEN L, GEN t, long bitprec).

3.16.22 lfuninit(L, sdom, {der = 0}). Initalization function for all functions linked to the
computation of the L-function L(s) encoded by L, where s belongs to the rectangular domain
sdom = [center , w, h] centered on the real axis, |<(s) − center | ≤ w, |=(s)| ≤ h, where all three
components of sdom are real and w, h are nonnegative. der is the maximum order of derivation
that will be used. The subdomain [k/2, 0, h] on the critical line (up to height h) can be encoded as
[h] for brevity. The subdomain [k/2, w, h] centered on the critical line can be encoded as [w, h] for
brevity.

The argument L is an Lmath, an Ldata or an Linit. See ??Ldata and ??lfuncreate for how
to create it.

The height h of the domain is a crucial parameter: if you only need L(s) for real s, set h to 0.
The running time is roughly proportional to

(B/d + πh/4)d/2+3 N 1/2 ,

where B is the default bit accuracy, d is the degree of the L-function, and N is the conductor (the
exponent d/2 + 3 is reduced to d/2 + 2 when d = 1 and d = 2). There is also a dependency on w,
which is less crucial, but make sure to use the smallest rectangular domain that you need.

? L0 = lfuncreate(1); \\ Riemann zeta

? L = lfuninit(L0, [1/2, 0, 100]); \\ for zeta(1/2+it), |t| < 100
? lfun(L, 1/2 + I)
? L = lfuninit(L0, [100]); \\ same as above !

The library syntax is GEN lfuninit0(GEN L, GEN sdom, long der, long bitprec).

3.16.23 lfunlambda(L, s, {D = 0}). Compute the completed L-function Λ(s) = N s/2 γ(s)L(s),
or if D is set, the derivative of order D at s. The parameter L is either an Lmath, an Ldata (created
by lfuncreate, or an Linit (created by lfuninit), preferrably the latter if many values are to be
computed.

The result is given with absolute error less than 2−B |γ(s)N s/2 |, where B = realbitprecision.

The library syntax is GEN lfunlambda0(GEN L, GEN s, long D, long bitprec).

530
3.16.24 lfunmfspec(L). Let L be the L-function attached to a modular eigenform f of weight k,
as given by lfunmf. In even weight, returns [ve,vo,om,op], where ve (resp., vo) is the vector of
even (resp., odd) periods of f and om and op the corresponding real numbers ω − and ω + normalized
in a noncanonical way. In odd weight ominus is the same as op and we return [v,op] where v is
the vector of all periods.

? D = mfDelta(); mf = mfinit(D); L = lfunmf(mf, D);

? [ve, vo, om, op] = lfunmfspec(L)
%2 = [[1, 25/48, 5/12, 25/48, 1], [1620/691, 1, 9/14, 9/14, 1, 1620/691],\
0.0074154209298961305890064277459002287248,\
0.0050835121083932868604942901374387473226]
? DS = mfsymbol(mf, D); bestappr(om*op / mfpetersson(DS), 10^8)
%3 = 8192/225
? mf = mfinit([4, 9, -4], 0);
? F = mfeigenbasis(mf)[1]; L = lfunmf(mf, F);
? [v, om] = lfunmfspec(L)
%6 = [[1, 10/21, 5/18, 5/24, 5/24, 5/18, 10/21, 1],\
1.1302643192034974852387822584241400608]
? FS = mfsymbol(mf, F); bestappr(om^2 / mfpetersson(FS), 10^8)
%7 = 113246208/325

The library syntax is GEN lfunmfspec(GEN L, long bitprec).

3.16.25 lfunmul(L1 , L2 ). Creates the Ldata structure (without initialization) corresponding to

the product of the Dirichlet series given by L1 and L2.

The library syntax is GEN lfunmul(GEN L1, GEN L2, long bitprec).

3.16.26 lfunorderzero(L, {m = −1}). Computes the order of the possible zero of the L-function
at the center k/2 of the critical strip; return 0 if L(k/2) does not vanish.

If m is given and has a nonnegative value, assumes the order is at most m. Otherwise, the
algorithm chooses a sensible default:

• if the L argument is an Linit, assume that a multiple zero at s = k/2 has order less than
or equal to the maximal allowed derivation order.

• else assume the order is less than 4.

You may explicitly increase this value using optional argument m; this overrides the default
value above. (Possibly forcing a recomputation of the Linit.)

The library syntax is long lfunorderzero(GEN L, long m, long bitprec).

531
3.16.27 lfunqf(Q). Returns the Ldata structure attached to the Θ function of the lattice attached
to the primitive form proportional to the definite positive quadratic form Q.

? L = lfunqf(matid(2));
? lfunqf(L,2)
%2 = 6.0268120396919401235462601927282855839
? lfun(x^2+1,2)*4
%3 = 6.0268120396919401235462601927282855839

The following computes the Madelung constant:

? L1=lfunqf(matdiagonal([1,1,1]));
? L2=lfunqf(matdiagonal([4,1,1]));
? L3=lfunqf(matdiagonal([4,4,1]));
? F(s)=6*lfun(L2,s)-12*lfun(L3,s)-lfun(L1,s)*(1-8/4^s);
? F(1/2)
%5 = -1.7475645946331821906362120355443974035

The library syntax is GEN lfunqf(GEN Q, long prec).

3.16.28 lfunrootres(data). Given the Ldata attached to an L-function (or the output of lfun-
thetainit), compute the root number and the residues.

The output is a 3-component vector [[[a1 , r1 ], · · · , [an , rn ], [[b1 , R1 ], · · · , [bm , Rm ]] , w], where ri
is the polar part of L(s) at ai , Ri is is the polar part of Λ(s) at bi or [0, 0, r] if there is no pole, and
w is the root number. In the present implementation,

• either the polar part must be completely known (and is then arbitrary): the function deter-
mines the root number,

? L = lfunmul(1,1); \\ zeta^2
? [r,R,w] = lfunrootres(L);
? r \\ single pole at 1, double
%3 = [[1, 1.[...]*x^-2 + 1.1544[...]*x^-1 + O(x^0)]]
? w
%4 = 1
? R \\ double pole at 0 and 1
%5 = [[1,[...]], [0,[...]]]~

• or at most a single pole is allowed: the function computes both the root number and the
residue (0 if no pole).

The library syntax is GEN lfunrootres(GEN data, long bitprec).

532
3.16.29 lfunshift(L, d, {flag}). Creates the Ldata structure (without initialization) corresponding
to the shift of L by d, that is to the function Ld such that Ld (s) = L(s − d). If flag = 1, return the
product L × Ld instead.
? Z = lfuncreate(1); \\ zeta(s)
? L = lfunshift(Z,1); \\ zeta(s-1)
? normlp(Vec(lfunlambda(L,s)-lfunlambda(L,3-s)))
%3 = 0.E-38 \\ the expansions coincide to ’seriesprecision’
? lfun(L,1)
%4 = -0.50000000000000000000000000000000000000 \\ = zeta(0)
? M = lfunshift(Z,1,1); \\ zeta(s)*zeta(s-1)
? normlp(Vec(lfunlambda(M,s)-lfunlambda(M,2-s)))
%6 = 2.350988701644575016 E-38
? lfun(M,2) \\ simple pole at 2, residue zeta(2)
%7 = 1.6449340668482264364724151666460251892*x^-1+O(x^0)
The library syntax is GEN lfunshift(GEN L, GEN d, long flag, long bitprec).

3.16.30 lfunsympow(E, m). Returns the Ldata structure attached to the L function attached to
the m-th symmetric power of the elliptic curve E defined over the rationals.
The library syntax is GEN lfunsympow(GEN E, ulong m).

3.16.31 lfuntheta(data, t, {m = 0}). Compute the value of the m-th derivative at t of the theta
function attached to the L-function given by data. data can be either the standard L-function
data, or the output of lfunthetainit. The result is given with absolute error less than 2−B , where
B = realbitprecision.
P p
The theta function is defined by the formula Θ(t) = n≥1 a(n)K(nt/ (N )), where a(n) are
the coefficients of the Dirichlet series, N is the conductor, and K is the inverse Mellin transform of
the gamma product defined by the Vga component. Its Mellin transform is equal to Λ(s) − P (s),
where Λ(s) is the completed L-function and the rational function P P (s) its polar part. In particular,
if the L-function is the L-function of a modular form f (τ ) = n≥0 a(n)q n with q = exp(2πiτ ), we
√
have Θ(t) = 2(f (it/ N ) − a(0)). Note that a(0) = −L(f, 0) in this case.
The library syntax is GEN lfuntheta(GEN data, GEN t, long m, long bitprec).

3.16.32 lfunthetacost(L, {tdom}, {m = 0}). This function estimates the cost of running lfun-
thetainit(L,tdom,m) at current bit precision. Returns the number of coefficients an that would
be computed. This also estimates the cost of a subsequent evaluation lfuntheta, which must com-
pute that many values of gammamellininv at the current bit precision. If L is already an Linit,
then tdom and m are ignored and are best left omitted: we get an estimate of the cost of using
that particular Linit.
? \pb 1000
? L = lfuncreate(1); \\ Riemann zeta
? lfunthetacost(L); \\ cost for theta(t), t real >= 1
%1 = 15
? lfunthetacost(L, 1 + I); \\ cost for theta(1+I). Domain error !
*** at top-level: lfunthetacost(1,1+I)
*** ^--------------------
*** lfunthetacost: domain error in lfunthetaneed: arg t > 0.785

533
 ? lfunthetacost(L, 1 + I/2) \\ for theta(1+I/2).
%2 = 23
? lfunthetacost(L, 1 + I/2, 10) \\ for theta^((10))(1+I/2).
%3 = 24
? lfunthetacost(L, [2, 1/10]) \\ cost for theta(t), |t| >= 2, |arg(t)| < 1/10
%4 = 8
? L = lfuncreate( ellinit([1,1]) );
? lfunthetacost(L) \\ for t >= 1
%6 = 2471
The library syntax is long lfunthetacost0(GEN L, GEN tdom = NULL, long m, long
bitprec).

3.16.33 lfunthetainit(L, {tdom}, {m = 0}). Initalization function for evaluating the m-th deriva-
tive of theta functions with argument t in domain tdom. By default (tdom omitted), t is real, t ≥ 1.
Otherwise, tdom may be
• a positive real scalar ρ: t is real, t ≥ ρ.
• a nonreal complex number: compute at this particular t; this allows to compute θ(z) for any
complex z satisfying |z| ≥ |t| and | arg z| ≤ | arg t|; we must have |2 arg z/d| < π/2, where d is the
degree of the Γ factor.
• a pair [ρ, α]: assume that |t| ≥ ρ and | arg t| ≤ α; we must have |2α/d| < π/2, where d is the
degree of the Γ factor.
? \p500
? L = lfuncreate(1); \\ Riemann zeta
? t = 1+I/2;
? lfuntheta(L, t); \\ direct computation
time = 30 ms.
? T = lfunthetainit(L, 1+I/2);
time = 30 ms.
? lfuntheta(T, t); \\ instantaneous
The T structure would allow to quickly compute θ(z) for any z in the cone delimited by t as
explained above. On the other hand
? lfuntheta(T,I)
*** at top-level: lfuntheta(T,I)
*** ^--------------
*** lfuntheta: domain error in lfunthetaneed: arg t > 0.785398163397448
The initialization is equivalent to
? lfunthetainit(L, [abs(t), arg(t)])
The library syntax is GEN lfunthetainit(GEN L, GEN tdom = NULL, long m, long bit-
prec).

3.16.34 lfuntwist(L, chi ). Creates the Ldata structure (without initialization) corresponding to
the twist of L by the primitive character attached to the Dirichlet character chi. The conductor
of the character must be coprime to the conductor of the L-function L.
The library syntax is GEN lfuntwist(GEN L, GEN chi, long bitprec).

534
3.16.35 lfunzeros(L, lim, {divz = 8}). lim being either a positive upper limit or a nonempty real
interval, computes an ordered list of zeros of L(s) on the critical line up to the given upper limit
or in the given interval. Use a naive algorithm which may miss some zeros: it assumes that two
consecutive zeros at height T ≥ 1 differ at least by 2π/ω, where

ω := divz · d log(T /2π) + d + 2 log(N/(π/2)d ) .

To use a finer search mesh, set divz to some integral value larger than the default (= 8).
? lfunzeros(1, 30) \\ zeros of Rieman zeta up to height 30
%1 = [14.134[...], 21.022[...], 25.010[...]]
? #lfunzeros(1, [100,110]) \\ count zeros with 100 <= Im(s) <= 110
%2 = 4
The algorithm also assumes that all zeros are simple except possibly on the real axis at s = k/2 and
that there are no poles in the search interval. (The possible zero at s = k/2 is repeated according
to its multiplicity.)
If you pass an Linit to the function, the algorithm assumes that a multiple zero at s = k/2 has
order less than or equal to the maximal derivation order allowed by the Linit. You may increase
that value in the Linit but this is costly: only do it for zeros of low height or in lfunorderzero
instead.
The library syntax is GEN lfunzeros(GEN L, GEN lim, long divz, long bitprec).

3.17 Modular forms.

This section describes routines for working with modular forms and modular form spaces.

3.17.1 Modular form spaces.

These structures are initialized by the mfinit command; supported modular form spaces with
corresponding flags are the following:
• The full modular form space Mk (Γ0 (N ), χ), where k is an integer or a half-integer and χ a
Dirichlet character modulo N (flag 4, the default).
• The cuspidal space Sk (Γ0 (N ), χ) (flag 1).
• The Eisenstein space Ek (Γ0 (N ), χ) (flag 3), so that Mk = Ek ⊕ Sk .
• The new space Sknew (Γ0 (N ), χ) (flag 0).
• The old space Skold (Γ0 (N ), χ) (flag 2), so that Sk = Sknew ⊕ Skold .
These resulting mf structure contains a basis of modular forms, which is accessed by the
function mfbasis; the elements of this basis have Fourier coefficients in the cyclotomic field Q(χ).
These coefficients are given algebraically, as rational numbers or t_POLMODs. The member function
mf.mod recovers the modulus used to define Q(χ), which is a cyclotomic polynomial Φn (t). When
needed, the elements of Q(χ) are considered to be canonically embedded into C via Mod(t, Φn (t)) 7→
exp(2iπ/n).
The basis of eigenforms for the new space is obtained by the function mfeigenbasis: the
elements of this basis now have Fourier coefficients in a relative field extension of Q(χ). Note that
if the space is larger than the new space (i.e. is the cuspidal or full space) we nevertheless obtain
only the eigenbasis for the new space.

535
3.17.2 Generalized modular forms.
A modular form is represented in a special internal format giving the possibility to compute
an arbitrary number of terms of its Fourier coefficients at infinity [a(0), a(1), ..., a(n)] using the
function mfcoefs. These coefficients are given algebraically, as rational numbers or t_POLMODs.
The member function f.mod recovers the modulus used in the coefficients of f , which will be the
same as for k = Q(χ) (a cyclotomic polynomial), or define a number field extension K/k.
Modular forms are obtained either directly from other mathematical objects, e.g., elliptic
curves, or by a specific formula, e.g., Eisenstein series or Ramanujan’s Delta function, or by applying
standard operators to existing forms (Hecke operators, Rankin–Cohen brackets, . . . ). A function
mfparams is provided so that one can recover the level, weight, character and field of definition
corresponding to a given modular form.
A number of creation functions and operations are provided. It is however important to note
that strictly speaking some of these operations create objects which are not modular forms: typical
examples are derivation or integration of modular forms, the Eisenstein series E2 , eta quotients, or
quotients of modular forms. These objects are nonetheless very important in the theory, so are not
considered as errors; however the user must be aware that no attempt is made to check that the
objects that he handles are really modular. When the documentation of a function does not state
that it applies to generalized modular forms, then the output is undefined if the input is not a true
modular form.

3.17.3 getcache(). Returns information about various auto-growing caches. For each resource,
we report its name, its size, the number of cache misses (since the last extension), the largest cache
miss and the size of the cache in bytes.
The caches are initially empty, then set automatically to a small inexpensive default value,
then grow on demand up to some maximal value. Their size never decreases, they are only freed
on exit.
The current caches are
• Hurwitz class numbers H(D) for |D| ≤ N , computed in time O(N 3/2 ) using O(N ) space.
• Factorizations of small integers up to N , computed in time O(N 1+ε ) using O(N log N ) space.
• Divisors of small integers up to N , computed in time O(N 1+ε ) using O(N log N ) space.
• Coredisc’s of negative integers down to −N , computed in time O(N 1+ε ) using O(N ) space.
• Primitive dihedral forms of weight 1 and level up to N , computed in time O(N 2+ε ) and
space O(N 2 ).
? getcache() \\ on startup, all caches are empty
%1 =
[ "Factors" 0 0 0 0]
[ "Divisors" 0 0 0 0]
[ "H" 0 0 0 0]
["CorediscF" 0 0 0 0]
[ "Dihedral" 0 0 0 0]
? mfdim([500,1,0],0); \\ nontrivial computation
time = 540 ms.

536
 ? getcache()
%3 =
[ "Factors" 50000 0 0 4479272]
["Divisors" 50000 1 100000 5189808]
[ "H" 50000 0 0 400008]
["Dihedral" 1000 0 0 2278208]
The library syntax is GEN getcache().

3.17.4 lfunmf(mf , {F }). If F is a modular form in mf, output the L-functions corresponding to
its [Q(F ) : Q(χ)] complex embeddings, ready for use with the lfun package. If F is omitted,
output the L-functions attached to all eigenforms in the new space; the result is a vector whose
length is the number of Galois orbits of newforms. Each entry contains the vector of L-functions
corresponding to the d complex embeddings of an orbit of dimension d over Q(χ).
? mf = mfinit([35,2],0);mffields(mf)
%1 = [y, y^2 - y - 4]
? f = mfeigenbasis(mf)[2]; mfparams(f) \\ orbit of dimension two
%2 = [35, 2, 1, y^2 - y - 4, t - 1]
? [L1,L2] = lfunmf(mf, f); \\ Two L-functions
? lfun(L1,1)
%4 = 0.81018461849460161754947375433874745585
? lfun(L2,1)
%5 = 0.46007635204895314548435893464149369804
? [ lfun(L,1) | L <- concat(lfunmf(mf)) ]
%6 = [0.70291..., 0.81018..., 0.46007...]
The concat instruction concatenates the vectors corresponding to the various (here two) orbits, so
that we obtain the vector of all the L-functions attached to eigenforms.
The library syntax is GEN lfunmf(GEN mf, GEN F = NULL, long bitprec).

3.17.5 mfDelta(). Mf structure corresponding to the Ramanujan Delta function ∆.

? mfcoefs(mfDelta(),4)
%1 = [0, 1, -24, 252, -1472]
The library syntax is GEN mfDelta().

3.17.6 mfEH(k). k being in 1/2 + Z≥0 , return the mf structure corresponding to the Cohen-
Eisenstein series Hk of weight k on Γ0 (4).
? H = mfEH(13/2); mfcoefs(H,4)
%1 = [691/32760, -1/252, 0, 0, -2017/252]
The coefficients of H are given by the Cohen-Hurwitz function H(k − 1/2, N ) and can be
obtained for moderately large values of N (the algorithm uses Õ(N ) time):
? mfcoef(H,10^5+1)
time = 55 ms.
%2 = -12514802881532791504208348
? mfcoef(H,10^7+1)
time = 6,044 ms.

537
 %3 = -1251433416009877455212672599325104476
The library syntax is GEN mfEH(GEN k).

3.17.7 mfEk(k). K being an even nonnegative integer, return the mf structure corresponding to
the standard Eisenstein series Ek .
? mfcoefs(mfEk(8), 4)
%1 = [1, 480, 61920, 1050240, 7926240]
The library syntax is GEN mfEk(long k).

3.17.8 mfTheta({psi = 1}). The unary theta function corresponding to the primitive Dirichlet
character ψ. Its level is 4F (ψ)2 and its weight is 1 − ψ(−1)/2.
? Ser(mfcoefs(mfTheta(),30))
%1 = 1 + 2*x + 2*x^4 + 2*x^9 + 2*x^16 + 2*x^25 + O(x^31)
? f = mfTheta(8); Ser(mfcoefs(f,30))
%2 = 2*x - 2*x^9 - 2*x^25 + O(x^31)
? mfparams(f)
%3 = [256, 1/2, 8, y, t + 1]
? g = mfTheta(-8); Ser(mfcoefs(g,30))
%4 = 2*x + 6*x^9 - 10*x^25 + O(x^31)
? mfparams(g)
%5 = [256, 3/2, 8, y, t + 1]
? h = mfTheta(Mod(2,5)); mfparams(h)
%6 = [100, 3/2, Mod(7, 20), y, t^2 + 1]
The library syntax is GEN mfTheta(GEN psi = NULL).

3.17.9 mfatkin(mfatk , f ). Given a mfatk output by mfatk = mfatkininit(mf,Q) and a modular

form f belonging to the pace mf, returns the modular form g = C × f |WQ , where C = mfatk[3]
is a normalizing constant such that g has the same field of coefficients as f ; mfatk[3] gives the
constant C, and mfatk[1] gives the modular form space to which g belongs (or is set to 0 if it is
mf).
? mf = mfinit([35,2],0); [f] = mfbasis(mf);
? mfcoefs(f, 4)
%2 = [0, 3, -1, 0, 3]
? mfatk = mfatkininit(mf,7);
? g = mfatkin(mfatk, f); mfcoefs(g, 4)
%4 = [0, 1, -1, -2, 7]
? mfatk = mfatkininit(mf,35);
? g = mfatkin(mfatk, f); mfcoefs(g, 4)
%6 = [0, -3, 1, 0, -3]
The library syntax is GEN mfatkin(GEN mfatk, GEN f).

538
3.17.10 mfatkineigenvalues(mf , Q). Given a modular form space mf of integral weight k and a
primitive divisor Q of the level N of mf, outputs the Atkin–Lehner eigenvalues of wQ on the new
space, grouped by orbit. If the Nebentypus χ of mf is a (trivial or) quadratic character defined
modulo N/Q, the result is rounded and the eigenvalues are ±ik .
? mf = mfinit([35,2],0); mffields(mf)
%1 = [y, y^2 - y - 4] \\ two orbits, dimension 1 and 2
? mfatkineigenvalues(mf,5)
%2 = [[1], [-1, -1]]
? mf = mfinit([12,7,Mod(3,4)],0);
? mfatkineigenvalues(mf,3)
%4 = [[I, -I, -I, I, I, -I]] \\ one orbit
To obtain the eigenvalues on a larger space than the new space, e.g., the full space, you can
directly call [mfB,M,C]=mfatkininit and compute the eigenvalues as the roots of the characteristic
polynomial of M/C, by dividing the roots of charpoly(M) by C. Note that the characteristic poly-
nomial is computed exactly since M has coefficients in Q(χ), whereas C may be given by a complex
number. If the coefficients of the characteristic polynomial are polmods modulo T they must be
embedded to C first using subst(lift(), t, exp(2*I*Pi/n)), when T is poliscyclo(n); note
that T = mf.mod.
The library syntax is GEN mfatkineigenvalues(GEN mf, long Q, long prec).

3.17.11 mfatkininit(mf , Q). Given a modular form space with parameters N, k, χ and a primitive
divisor Q of the level N , initializes data necessary for working with the Atkin–Lehner operator WQ ,
for now only the function mfatkin. We write χ ∼ χQ χN/Q where the two characters are primitive
with (coprime) conductors dividing Q and N/Q respectively. For F ∈ Mk (Γ0 (N ), χ), the form
F |WQ still has level N and weight k but its Nebentypus may no longer be χ: it becomes χQ χN/Q )
if k is integral and χQ χN/Q )(4Q/·) if not.
The result is a technical 4-component vector [mfB, MC, C, mf], where
• mfB encodes the modular form space to which F |WQ belongs when F ∈ Mk (Γ0 (N ), χ): an
mfinit corresponding to a new Nebentypus or the integer 0 when the character does not change.
This does not depend on F .
• MC is the matrix of WQ on the bases of mf and mfB multiplied by a normalizing constant
C(k, χ, Q). This matrix has polmod coefficients in Q(χ).
• C is the complex constant C(k, χ, Q). For k integral, let A(k, χ, Q) = Qε /g(χQ ), where ε = 0
for k even and 1/2 for k odd and where g(χQ ) is the Gauss sum attached to χQ ). (A similar, more
complicated, definition holds in half-integral weight depending on the parity of k − 1/2.) Then if
M denotes the matrix of WQ on the bases of mf and mfB, A · M has coefficients in Q(χ). If A
is rational, we let C = 1 and C = A as a floating point complex number otherwise, and finally
MC := M · C.
? mf=mfinit([32,4],0); [mfB,MC,C]=mfatkininit(mf,32); MC
%1 =
[5/16 11/2 55/8]
[ 1/8 0 -5/4]
[1/32 -1/4 11/16]
? C

539
 %2 = 1
? mf=mfinit([32,4,8],0); [mfB,MC,C]=mfatkininit(mf,32); MC
%3 =
[ 1/8 -7/4]
[-1/16 -1/8]
? C
%4 = 0.35355339059327376220042218105242451964
? algdep(C,2) \\ C = 1/sqrt(8)
%5 = 8*x^2 - 1
The library syntax is GEN mfatkininit(GEN mf, long Q, long prec).

3.17.12 mfbasis(NK , {space = 4}). If N K = [N, k, CHI ] as in mfinit, gives a basis of the
corresponding subspace of Mk (Γ0 (N ), χ). N K can also be the output of mfinit, in which case
space can be omitted. To obtain the eigenforms, use mfeigenbasis.
If space is a full space Mk , the output is the union of first, a basis of the space of Eisenstein
series, and second, a basis of the cuspidal space.
? see(L) = apply(f->mfcoefs(f,3), L);
? mf = mfinit([35,2],0);
? see( mfbasis(mf) )
%2 = [[0, 3, -1, 0], [0, -1, 9, -8], [0, 0, -8, 10]]
? see( mfeigenbasis(mf) )
%3 = [[0, 1, 0, 1], [Mod(0, z^2 - z - 4), Mod(1, z^2 - z - 4), \
Mod(-z, z^2 - z - 4), Mod(z - 1, z^2 - z - 4)]]
? mf = mfinit([35,2]);
? see( mfbasis(mf) )
%5 = [[1/6, 1, 3, 4], [1/4, 1, 3, 4], [17/12, 1, 3, 4], \
[0, 3, -1, 0], [0, -1, 9, -8], [0, 0, -8, 10]]
? see( mfbasis([48,4],0) )
%6 = [[0, 3, 0, -3], [0, -3, 0, 27], [0, 2, 0, 30]]
The library syntax is GEN mfbasis(GEN NK, long space).

3.17.13 mfbd(F, d). F being a generalized modular form, return B(d)(F ), where B(d) is the
expanding operator τ 7→ dτ .
? D2=mfbd(mfDelta(),2); mfcoefs(D2, 6)
%1 = [0, 0, 1, 0, -24, 0, 252]
The library syntax is GEN mfbd(GEN F, long d).

3.17.14 mfbracket(F, G, {m = 0}). Compute the m-th Rankin–Cohen bracket of the generalized
modular forms F and G.
? E4 = mfEk(4); E6 = mfEk(6);
? D1 = mfbracket(E4,E4,2); mfcoefs(D1,5)/4800
%2 = [0, 1, -24, 252, -1472, 4830]
? D2 = mfbracket(E4,E6,1); mfcoefs(D2,10)/(-3456)
%3 = [0, 1, -24, 252, -1472, 4830]
The library syntax is GEN mfbracket(GEN F, GEN G, long m).

540
3.17.15 mfcoef(F, n). Compute the n-th Fourier coefficient a(n) of the generalized modular form
F . Note that this is the n + 1-st component of the vector mfcoefs(F,n) as well as the second
component of mfcoefs(F,1,n).

? mfcoef(mfDelta(),10)
%1 = -115920
The library syntax is GEN mfcoef(GEN F, long n).

3.17.16 mfcoefs(F, n, {d = 1}). Compute the vector of Fourier coefficients [a[0], a[d], ..., a[nd]] of
the generalized modular form F ; d must be positive and d = 1 by default.

? D = mfDelta();
? mfcoefs(D,10)
%2 = [0, 1, -24, 252, -1472, 4830, -6048, -16744, 84480, -113643, -115920]
? mfcoefs(D,5,2)
%3 = [0, -24, -1472, -6048, 84480, -115920]
? mfcoef(D,10)
%4 = -115920
This function also applies when F is a modular form space as output by mfinit; it then returns
the matrix whose columns give the Fourier expansions of the elements of mfbasis(F ):

? mf = mfinit([1,12]);
? mfcoefs(mf,5)
%2 =
[691/65520 0]
[ 1 1]
[ 2049 -24]
[ 177148 252]
[ 4196353 -1472]
[ 48828126 4830]
The library syntax is GEN mfcoefs(GEN F, long n, long d).

3.17.17 mfconductor(mf , F ). mf being output by mfinit for the cuspidal space and F a modular
form,Pgives the smallest level at which F is defined. In particular, if F is cuspidal and we write
F = j B(dj )fj for new forms fj of level Nj (see mftonew), then its conductor is the least common
multiple of the dj Nj .

? mf=mfinit([96,6],1); vF = mfbasis(mf); mfdim(mf)

%1 = 72
? vector(10,i, mfconductor(mf, vF[i]))
%2 = [3, 6, 12, 24, 48, 96, 4, 8, 12, 16]
The library syntax is long mfconductor(GEN mf, GEN F).

541
3.17.18 mfcosets(N ). Let N be F a positive integer. Return the list of right cosets of Γ0 (N )\Γ,
i.e., matrices γj ∈ Γ such that Γ = j Γ0 (N )γj . The γj are chosen in the form [a, b; c, d] with c | N .
? mfcosets(4)
%1 = [[0, -1; 1, 0], [1, 0; 1, 1], [0, -1; 1, 2], [0, -1; 1, 3],\
[1, 0; 2, 1], [1, 0; 4, 1]]
We also allow the argument N to be a modular form space, in which case it is replaced by the level
of the space:
? M = mfinit([4, 12, 1], 0); mfcosets(M)
%2 = [[0, -1; 1, 0], [1, 0; 1, 1], [0, -1; 1, 2], [0, -1; 1, 3],\
[1, 0; 2, 1], [1, 0; 4, 1]]

Warning. In the present implementation, the trivial coset is represented by [1, 0; N, 1] and is the
last in the list.
The library syntax is GEN mfcosets(GEN N).

3.17.19 mfcuspisregular(NK , cusp). In the space defined by NK = [N,k,CHI] or NK = mf, de-

termine if cusp in canonical format (oo or denominator dividing N ) is regular or not.
? mfcuspisregular([4,3,-4],1/2)
%1 = 0
The library syntax is long mfcuspisregular(GEN NK, GEN cusp).

3.17.20 mfcusps(N ). Let N be a positive integer. Return the list of cusps of Γ0 (N ) in the form
a/b with b | N .
? mfcusps(24)
%1 = [0, 1/2, 1/3, 1/4, 1/6, 1/8, 1/12, 1/24]
We also allow the argument N to be a modular form space, in which case it is replaced by the level
of the space:
? M = mfinit([4, 12, 1], 0); mfcusps(M)
%2 = [0, 1/2, 1/4]
The library syntax is GEN mfcusps(GEN N).

3.17.21 mfcuspval(mf , F, cusp). Valuation of modular form F in the space mf at cusp, which
can be either ∞ or any rational number. The result is either a rational number or ∞ if F is zero.
Let χ be the Nebentypus of the space mf; if Q(F ) 6= Q(χ), return the vector of valuations attached
to the [Q(F ) : Q(chi)] complex embeddings of F .
? T=mfTheta(); mf=mfinit([12,1/2]); mfcusps(12)
%1 = [0, 1/2, 1/3, 1/4, 1/6, 1/12]
? apply(x->mfcuspval(mf,T,x), %1)
%2 = [0, 1/4, 0, 0, 1/4, 0]
? mf=mfinit([12,6,12],1); F=mfbasis(mf)[5];
? apply(x->mfcuspval(mf,F,x),%1)
%4 = [1/12, 1/6, 1/2, 2/3, 1/2, 2]
? mf=mfinit([12,3,-4],1); F=mfbasis(mf)[1];
? apply(x->mfcuspval(mf,F,x),%1)

542
 %6 = [1/12, 1/6, 1/4, 2/3, 1/2, 1]
? mf = mfinit([625,2],0); [F] = mfeigenbasis(mf); mfparams(F)
%7 = [625, 2, 1, y^2 - y - 1, t - 1] \\ [Q(F):Q(chi)] = 2
? mfcuspval(mf, F, 1/25)
%8 = [1, 2] \\ one conjugate has valuation 1, and the other is 2
? mfcuspval(mf, F, 1/5)
%9 = [1/25, 1/25]
The library syntax is GEN mfcuspval(GEN mf, GEN F, GEN cusp, long bitprec).

3.17.22 mfcuspwidth(N, cusp). Width of cusp in Γ0 (N ).

? mfcusps(12)
%1 = [0, 1/2, 1/3, 1/4, 1/6, 1/12]
? [mfcuspwidth(12,c) | c <- mfcusps(12)]
%2 = [12, 3, 4, 3, 1, 1]
? mfcuspwidth(12, oo)
%3 = 1
We also allow the argument N to be a modular form space, in which case it is replaced by the level
of the space:

? M = mfinit([4, 12, 1], 0); mfcuspwidth(M, 1/2)

%4 = 1
The library syntax is long mfcuspwidth(GEN N, GEN cusp).

3.17.23 mfderiv(F, {m = 1}). m-th formal derivative of the power series corresponding to the
generalized modular form F , with respect to the differential operator qd/dq (default m = 1).

? D=mfDelta();
? mfcoefs(D, 4)
%2 = [0, 1, -24, 252, -1472]
? mfcoefs(mfderiv(D), 4)
%3 = [0, 1, -48, 756, -5888]
The library syntax is GEN mfderiv(GEN F, long m).

3.17.24 mfderivE2(F, {m = 1}). Compute the Serre derivative (q.d/dq)F − kE2 F/12 of the
generalized modular form F , which has weight k + 2; if F is a true modular form, then its Serre
derivative is also modular. If m > 1, compute the m-th iterate, of weight k + 2m.

? mfcoefs(mfderivE2(mfEk(4)),5)*(-3)
%1 = [1, -504, -16632, -122976, -532728]
? mfcoefs(mfEk(6),5)
%2 = [1, -504, -16632, -122976, -532728]
The library syntax is GEN mfderivE2(GEN F, long m).

543
3.17.25 mfdescribe(F, {&G}). Gives a human-readable description of F , which is either a mod-
ular form space or a generalized modular form. If the address of G is given, puts into G the vector
of parameters of the outermost operator defining F ; this vector is empty if F is a leaf (an atomic
object such as mfDelta(), not defined in terms of other forms) or a modular form space.
? E1 = mfeisenstein(4,-3,-4); mfdescribe(E1)
%1 = "F_4(-3, -4)"
? E2 = mfeisenstein(3,5,-7); mfdescribe(E2)
%2 = "F_3(5, -7)"
? E3 = mfderivE2(mfmul(E1,E2), 3); mfdescribe(E3,&G)
%3 = "DERE2^3(MUL(F_4(-3, -4), F_3(5, -7)))"
? mfdescribe(G[1][1])
%4 = "MUL(F_4(-3, -4), F_3(5, -7))"
? G[2]
%5 = 3
? for (i = 0, 4, mf = mfinit([37,4],i); print(mfdescribe(mf)));
S_4^new(G_0(37, 1))
S_4(G_0(37, 1))
S_4^old(G_0(37, 1))
E_4(G_0(37, 1))
M_4(G_0(37, 1))
The library syntax is GEN mfdescribe(GEN F, GEN *G = NULL).

3.17.26 mfdim(NK , {space = 4}). If N K = [N, k, CHI ] as in mfinit, gives the dimension of the
corresponding subspace of Mk (Γ0 (N ), χ). N K can also be the output of mfinit, in which case
space must be omitted.
The subspace is described by the small integer space: 0 for the newspace Sknew (Γ0 (N ), χ), 1
for the cuspidal space Sk , 2 for the oldspace Skold , 3 for the space of Eisenstein series Ek and 4 for
the full space Mk .

Wildcards. As in mfinit, CHI may be the wildcard 0 (all Galois orbits of characters); in this
case, the output is a vector of [order , conrey, dim, dimdih] corresponding to the nontrivial spaces,
where
• order is the order of the character,
• conrey its Conrey label from which the character may be recovered via znchar(conrey),
• dim the dimension of the corresponding space,
• dimdih the dimension of the subspace of dihedral forms corresponding to Hecke characters if
k = 1 (this is not implemented for the old space and set to −1 for the time being) and 0 otherwise.
The spaces are sorted by increasing order of the character; the characters are taken up to Galois
conjugation and the Conrey number is the minimal one among Galois conjugates. In weight 1, this
is only implemented when the space is 0 (newspace), 1 (cusp space), 2(old space) or 3(Eisenstein
series).

Wildcards for sets of characters. CHI may be a set of characters, and we return the set of
[dim, dimdih].

544
Wildcard for Mk (Γ1 (N )). Additionally, the wildcard CHI = −1 is available in which case we
output the total dimension of the corresponding subspace of Mk (Γ1 (N )). In weight 1, this is not
implemented when the space is 4 (fullspace).
? mfdim([23,2], 0) \\ new space
%1 = 2
? mfdim([96,6], 0)
%2 = 10
? mfdim([10^9,4], 3) \\ Eisenstein space
%1 = 40000
? mfdim([10^9+7,4], 3)
%2 = 2
? mfdim([68,1,-1],0)
%3 = 3
? mfdim([68,1,0],0)
%4 = [[2, Mod(67, 68), 1, 1], [4, Mod(47, 68), 1, 1]]
? mfdim([124,1,0],0)
%5 = [[6, Mod(67, 124), 2, 0]]
This last example shows that there exists a nondihedral form of weight 1 in level 124.
The library syntax is GEN mfdim(GEN NK, long space).

3.17.27 mfdiv(F, G). Given two generalized modular forms F and G, compute F/G assuming
that the quotient will not have poles at infinity. If this is the case, use mfshift before doing the
division.
? D = mfDelta(); \\ Delta
? H = mfpow(mfEk(4), 3);
? J = mfdiv(H, D)
*** at top-level: J=mfdiv(H,mfdeltac
*** ^--------------------
*** mfdiv: domain error in mfdiv: ord(G) > ord(F)
? J = mfdiv(H, mfshift(D,1));
? mfcoefs(J, 4)
%4 = [1, 744, 196884, 21493760, 864299970]
The library syntax is GEN mfdiv(GEN F, GEN G).

3.17.28 mfeigenbasis(mf ). Vector of the eigenforms for the space mf. The initial basis of forms
computed by mfinit before splitting is also available via mfbasis.
? mf = mfinit([26,2],0);
? see(L) = for(i=1,#L,print(mfcoefs(L[i],6)));
? see( mfeigenbasis(mf) )
[0, 1, -1, 1, 1, -3, -1]
[0, 1, 1, -3, 1, -1, -3]
? see( mfbasis(mf) )
[0, 2, 0, -2, 2, -4, -4]
[0, -2, -4, 10, -2, 0, 8]
The eigenforms are internally expressed as (algebraic) linear combinations of mfbasis(mf) and
it is very inefficient to compute many coefficients of those forms individually: you should rather

545
use mfcoefs(mf) to expand the basis once and for all, then multiply by mftobasis(mf,f) for the
forms you’re interested in:
? mf = mfinit([96,6],0); B = mfeigenbasis(mf); #B
%1 = 8;
? vector(#B, i, mfcoefs(B[i],1000)); \\ expanded individually: slow
time = 7,881 ms.
? M = mfcoefs(mf, 1000); \\ initialize once
time = 982 ms.
? vector(#B, i, M * mftobasis(mf,B[i])); \\ then expand: much faster
time = 623 ms.
When the eigenforms are defined over an extension field of Q(χ) for a nonrational character,
their coefficients are hard to read and you may want to lift them or to express them in an absolute
number field. In the construction below T defines Q(f ) over Q, a is the image of the generator
Mod(t, t2 + t + 1) of Q(χ) in Q(f ) and y − ka is the image of the root y of f.mod:
? mf = mfinit([31, 2, Mod(25,31)], 0); [f] = mfeigenbasis(mf);
? f.mod
%2 = Mod(1, t^2 + t + 1)*y^2 + Mod(2*t + 2, t^2 + t + 1)
? v = liftpol(mfcoefs(f,5))
%3 = [0, 1, (-t - 1)*y - 1, t*y + (t + 1), (2*t + 2)*y + 1, t]
? [T,a,k] = rnfequation(mf.mod, f.mod, 1)
%4 = [y^4 + 2*y^2 + 4, Mod(-1/2*y^2 - 1, y^4 + 2*y^2 + 4), 0]
? liftpol(substvec(v, [t,y], [a, y-k*a]))
%5 = [0, 1, 1/2*y^3 - 1, -1/2*y^3 - 1/2*y^2 - y, -y^3 + 1, -1/2*y^2 - 1]
Beware that the meaning of y has changed in the last line is different: it now represents of root
of T , no longer of f.mod (the notions coincide if k = 0 as here but it will not always be the case).
This can be avoided with an extra variable substitution, for instance
? [T,a,k] = rnfequation(mf.mod, subst(f.mod,’y,’x), 1)
%6 = [x^4 + 2*x^2 + 4, Mod(-1/2*x^2 - 1, x^4 + 2*x^2 + 4), 0]
? liftpol(substvec(v, [t,y], [a, x-k*a]))
%7 = [0, 1, 1/2*x^3 - 1, -1/2*x^3 - 1/2*x^2 - x, -x^3 + 1, -1/2*x^2 - 1]
The library syntax is GEN mfeigenbasis(GEN mf).

3.17.29 mfeigensearch(NK , {AP }). Search for a normalized rational eigen cuspform with
quadratic character given restrictions on a few initial coefficients. The meaning of the parame-
ters is as follows:
• NK governs the limits of the search: it is of the form [N, k]: search for given level N , weight
k and quadratic character; note that the character (D/.) is uniquely determined by (N, k). The
level N can be replaced by a vector of allowed levels.
• AP is the search criterion, which can be omitted: a list of pairs [. . . , [p, ap ], . . .], where p is a
prime number and ap is either a t_INT (the p-th Fourier coefficient must match ap exactly) or a
t_INTMOD Mod(a, b) (the p-th coefficient must be congruent to a modulo b).
The result is a vector of newforms f matching the search criteria, sorted by increasing level
then increasing |D|.
? #mfeigensearch([[1..80],2], [[2,2],[3,-1]])

546
 %1 = 1
? #mfeigensearch([[1..80],2], [[2,2],[5,2]])
%2 = 1
? v = mfeigensearch([[1..20],2], [[3,Mod(2,3)],[7,Mod(5,7)]]); #v
%3 = 1
? F=v[1]; [mfparams(F)[1], mfcoefs(F,15)]
%4 = [11, [0, 1, -2, -1, 2, 1, 2, -2, 0, -2, -2, 1, -2, 4, 4, -1]]
The library syntax is GEN mfeigensearch(GEN NK, GEN AP = NULL).

3.17.30 mfeisenstein(k, {CHI1 }, {CHI2 }). Create the Eisenstein series Ek (χ1 , χ2 ), where k ≥ 1,
χi are Dirichlet characters and an omitted character is considered as trivial. This form belongs to
Ek (Γ0 (N ), χ) with χ = χ1 χ2 and N is the product of the conductors of χ1 and χ2 .
? CHI = Mod(3,4);
? E = mfeisenstein(3, CHI);
? mfcoefs(E, 6)
%2 = [-1/4, 1, 1, -8, 1, 26, -8]
? CHI2 = Mod(4,5);
? mfcoefs(mfeisenstein(3,CHI,CHI2), 6)
%3 = [0, 1, -1, -10, 1, 25, 10]
? mfcoefs(mfeisenstein(4,CHI,CHI), 6)
%4 = [0, 1, 0, -28, 0, 126, 0]
? mfcoefs(mfeisenstein(4), 6)
%5 = [1/240, 1, 9, 28, 73, 126, 252]
Note that mfeisenstein(k) is 0 for k odd and −Bk /(2k) · Ek for k even, where
X
Ek (q) = 1 − (2k/Bk ) σk−1 (n)q n
n≥1

is the standard Eisenstein series. In other words it is normalized so that its linear coefficient is 1.

Important note. This function is currently implemented only when Q(χ) is the field of definition
of Ek (χ1 , χ2 ). If it is a strict subfield, an error is raised:
? mfeisenstein(6, Mod(7,9), Mod(4,9));
*** at top-level: mfeisenstein(6,Mod(7,9),Mod(4,9))
*** ^---------------------------------
*** mfeisenstein: sorry, mfeisenstein for these characters is not
*** yet implemented.
The reason for this is that each modular form is attached to a modular form space Mk (Γ0 (N ), χ).
This is a C-vector space but it allows a basis of forms defined over Q(χ) and is only implemented
as a Q(χ)-vector space: there is in general no mechanism to take linear combinations of forms in
the space with coefficients belonging to a larger field. (Due to their importance, eigenforms are the
single exception to this restriction; for an eigenform F , Q(F ) is built on top of Q(χ).) When the
property Q(χ) = Q(Ek (χ1 , χ2 ) does not hold, we cannot express E as a Q(χ)-linear combination
of the basis forms and many operations will fail. For this reason, the construction is currently
disabled.
The library syntax is GEN mfeisenstein(long k, GEN CHI1 = NULL, GEN CHI2 = NULL)
.

547
3.17.31 mfembed(f, {v}). Let f be a generalized modular form with parameters [N, k, χ, P ]
(see mfparams, we denote Q(χ) the subfield of C generated by the values of χ and Q(f ) the
field of definition of f . In this context Q(χ) has a single canonical complex embeding given by
s : Mod(t, polcyclo(n, t)) 7→ exp(2iπ/n) and the number field Q(f ) has [Q(f ) : Q(χ)] induced
embeddings attached to the complex roots of the polynomial s(P ). If Q(f ) is stricly larger than
Q(χ) we only allow an f which is an eigenform, produced by mfeigenbasis.
This function is meant to create embeddings of Q(f ) and/or apply them to the object v,
typically a vector of Fourier coefficients of f from mfcoefs.
• If v is omitted and f is a modular form as above, we return the embedding of Q(χ) if
Q(χ) = Q(f ) and a vector containing [Q(f ) : Q(χ)] embeddings of Q(f ) otherwise.
• If v is given, it must be a scalar in Q(f ), or a vector/matrix of such, we apply the embeddings
coefficientwise and return either a single result if Q(f ) = Q(χ) and a vector of [Q(f ) : Q(χ)] results
otherwise.
• Finally f can be replaced by a single embedding produced by mfembed(f ) (v was omitted)
and we apply that particular embedding to v.
? mf = mfinit([35,2,Mod(11,35)], 0);
? [f] = mfbasis(mf);
? f.mod \\ Q(χ) = Q(ζ3 )
%3 = t^2 + t + 1
? v = mfcoefs(f,5); lift(v) \\ coefficients in Q(χ)
%4 = [0, 2, -2*t - 2, 2*t, 2*t, -2*t - 2]
? mfembed(f, v) \\ single embedding
%5 = [0, 2, -1 - 1.7320...*I, -1 + 1.73205...*I, -1 + 1.7320...*I, ...]
? [F] = mfeigenbasis(mf);
? mffields(mf)
%7 = [y^2 + Mod(-2*t, t^2 + t + 1)] \\ [Q(f ) : Q(χ)] = 2
? V = liftpol( mfcoefs(F,5) );
%8 = [0, 1, y + (-t - 1), (t + 1)*y + t, (-2*t - 2)*y + t, -t - 1]
? vall = mfembed(F, V); #vall
%9 = 2 \\ 2 embeddings, both applied to V
? vall[1] \\ the first
%10 = [0, 1, -1.2071... - 2.0907...*I, 0.2071... - 0.3587...*I, ...]
? vall[2] \\ and the second one
%11 = [0, 1, 0.2071... + 0.3587...*I, -1.2071... + 2.0907...*I, ...]
? vE = mfembed(F); #vE \\ same 2 embeddings
%12 = 2
? mfembed(vE[1], V) \\ apply first embedding to V
%13 = [0, 1, -1.2071... - 2.0907...*I, 0.2071... - 0.3587...*I, ...]
For convenience, we also allow a modular form space from mfinit instead of f , corresponding
to the single embedding of Q(χ).
? [mfB,MC,C] = mfatkininit(mf,7); MC \\ coefs in Q(χ)
%13 =
[ Mod(2/7*t, t^2 + t + 1) Mod(-1/7*t - 2/7, t^2 + t + 1)]
[Mod(-1/7*t - 2/7, t^2 + t + 1) Mod(2/7*t, t^2 + t + 1)]

548
 ? C \\ normalizing constant
%14 = 0.33863... - 0.16787*I
? M = mfembed(mf, MC) / C \\ the true matrix for the action of w_7
[-0.6294... + 0.4186...*I -0.3625... - 0.5450...*I]
[-0.3625... - 0.5450...*I -0.6294... + 0.4186...*I]
? exponent(M*conj(M) - 1) \\ M * conj(M) is close to 1
%16 = -126
The library syntax is GEN mfembed0(GEN f, GEN v = NULL, long prec).

3.17.32 mfeval(mf , F, vtau). Computes the numerical value of the modular form F , belonging to
mf , at the complex number vtau or the vector vtau of complex numbers in the completed upper-
half plane. The result is given with absolute error less than 2−B , where B = realbitprecision.
If the field of definition Q(F ) is larger than Q(χ) then F may be embedded into C in d =
[Q(F ) : Q(χ)] ways, in which case a vector of the d results is returned.
? mf = mfinit([11,2],0); F = mfbasis(mf)[1]; mfparams(F)
%1 = [11, 2, 1, y, t-1] \\ Q(F) = Q(chi) = Q
? mfeval(mf,F,I/2)
%2 = 0.039405471130100890402470386372028382117
? mf = mfinit([35,2],0); F = mfeigenbasis(mf)[2]; mfparams(F)
%3 = [35, 2, 1, y^2 - y - 4, t - 1] \\ [Q(F) : Q(chi)] = 2
? mfeval(mf,F,I/2)
%4 = [0.045..., 0.0385...] \\ sigma_1(F) and sigma_2(F) at I/2
? mf = mfinit([12,4],1); F = mfbasis(mf)[1];
? mfeval(mf, F, 0.318+10^(-7)*I)
%6 = 3.379... E-21 + 6.531... E-21*I \\ instantaneous !
In order to maximize the imaginary part of the argument, the function computes (f |k γ)(γ −1 · τ )
for a suitable γ not necessarily in Γ0 (N ) (in which case f | γ is evaluated using mfslashexpansion).
? T = mfTheta(); mf = mfinit(T); mfeval(mf,T,[0,1/2,1,oo])
%1 = [1/2 - 1/2*I, 0, 1/2 - 1/2*I, 1]
The library syntax is GEN mfeval(GEN mf, GEN F, GEN vtau, long bitprec).

3.17.33 mffields(mf ). Given mf as output by mfinit with parameters (N, k, χ), returns the vector
of polynomials defining each Galois orbit of newforms over Q(χ).
? mf = mfinit([35,2],0); mffields(mf)
%1 = [y, y^2 - y - 4]
Here the character is trivial so Q(χ) = Q) and there are 3 newforms: one is rational (corresponding
to y), the other two are conjugate and defined over the quadratic field Q[y]/(y 2 − y − 4).
? [G,chi] = znchar(Mod(3,35));
? zncharconductor(G,chi)
%2 = 35
? charorder(G,chi)
%3 = 12
? mf = mfinit([35, 2, [G,chi]],0); mffields(mf)
%4 = [y, y]

549
 Here the character is primitive of order 12 and the two newforms are defined over Q(χ) =
Q(ζ12 ).
? mf = mfinit([35, 2, Mod(13,35)],0); mffields(mf)
%3 = [y^2 + Mod(5*t, t^2 + 1)]
This time the character has order 4 and there are two conjugate newforms over Q(χ) = Q(i).
The library syntax is GEN mffields(GEN mf).

3.17.34 mffromell(E). E being an elliptic curve defined over Q given by an integral model in
ellinit format, computes a 3-component vector [mf,F,v], where F is the newform corresponding
to E by modularity, mf is the newspace to which F belongs, and v gives the coefficients of F on
mfbasis(mf).
? E = ellinit("26a1");
? [mf,F,co] = mffromell(E);
? co
%2 = [3/4, 1/4]~
? mfcoefs(F, 5)
%3 = [0, 1, -1, 1, 1, -3]
? ellan(E, 5)
%4 = [1, -1, 1, 1, -3]
The library syntax is GEN mffromell(GEN E).

3.17.35 mffrometaquo(eta, {flag = 0}). Modular form corresponding to the eta quotient matrix
eta. If the valuation v at infinity is fractional, return 0. If the eta quotient is not holomorphic but
simply meromorphic, return 0 if flag=0; return the eta quotient (divided by q to the power −v if
v < 0, i.e., with valuation 0) if flag is set.
? mffrometaquo(Mat([1,1]),1)
%1 = 0
? mfcoefs(mffrometaquo(Mat([1,24])),6)
%2 = [0, 1, -24, 252, -1472, 4830, -6048]
? mfcoefs(mffrometaquo([1,1;23,1]),10)
%3 = [0, 1, -1, -1, 0, 0, 1, 0, 1, 0, 0]
? F = mffrometaquo([1,2;2,-1]); mfparams(F)
%4 = [16, 1/2, 1, y, t - 1]
? mfcoefs(F,10)
%5 = [1, -2, 0, 0, 2, 0, 0, 0, 0, -2, 0]
? mffrometaquo(Mat([1,-24]))
%6 = 0
? f = mffrometaquo(Mat([1,-24]),1); mfcoefs(f,6)
%7 = [1, 24, 324, 3200, 25650, 176256, 1073720]
For convenience, a t_VEC is also accepted instead of a factorization matrix with a single row:
? f = mffrometaquo([1,24]); \\ also valid
The library syntax is GEN mffrometaquo(GEN eta, long flag).

550
3.17.36 mffromlfun(L). Let L being an L-function in any of the lfun formats represent-
ing a self-dual modular form (for instance an eigenform). Return [NK,space,v] when mf =
mfinit(NK,space) is the modular form space containing the form and mftobasis(mf, v) will
represent it on the space basis. If L has rational coefficients, this will be enough to recognize the
modular form in mf :
? L = lfuncreate(x^2+1);
? lfunan(L,10)
%2 = [1, 1, 0, 1, 2, 0, 0, 1, 1, 2]
? [NK,space,v] = mffromlfun(L); NK
%4 = [4, 1, -4]
? mf=mfinit(NK,space); w = mftobasis(mf,v)
%5 = [1.0000000000000000000000000000000000000]~
? [f] = mfbasis(mf); mfcoefs(f,10) \\ includes a_0 !
%6 = [1/4, 1, 1, 0, 1, 2, 0, 0, 1, 1, 2]
If L has inexact complex coefficients, one can for instance compute an eigenbasis for mf and
check whether one of the attached L-function is reasonably close to L. In the example, we cheat
by producing the L function from an eigenform in a known space, but the function does not use
this information:
? mf = mfinit([32,6,Mod(5,32)],0);
? [poldegree(K) | K<-mffields(mf)]
%2 = [19] \\ one orbit, [Q(F) : Q(chi)] = 19
? L = lfunmf(mf)[1][1]; \\ one of the 19 L-functions attached to F
? lfunan(L,3)
%4 = [1, 5.654... - 0.1812...*I, -7.876... - 19.02...*I]
? [NK,space,v] = mffromlfun(L); NK
%5 = [32, 6, Mod(5, 32)]
? vL = concat(lfunmf(mf)); \\ L functions for all cuspidal eigenforms
? an = lfunan(L,10);
? for (i = 1, #vL, if (normlp(lfunan(vL[i],10) - an, oo) < 1e-10, print(i)));
1
The library syntax is GEN mffromlfun(GEN L, long prec).

3.17.37 mffromqf(Q, {P }). Q being an even integral positive definite quadratic form and P a
homogeneous spherical polynomial for Q, computes a 3-component vector [mf , F, v], where F is
the theta function corresponding to (Q, P ), mf is the corresponding space of modular forms (from
mfinit), and v gives the coefficients of F on mfbasis(mf).
? [mf,F,v] = mffromqf(2*matid(10)); v
%1 = [64/5, 4/5, 32/5]~
? mfcoefs(F, 5)
%2 = [1, 20, 180, 960, 3380, 8424]
? mfcoef(F, 10000) \\ number of ways of writing 10000 as sum of 10 squares
%3 = 128205250571893636
? mfcoefs(F, 10000); \\ fast !
time = 220ms
? [mf,F,v] = mffromqf([2,0;0,2],x^4-6*x^2*y^2+y^4);
? mfcoefs(F,10)

551
 %6 = [0, 4, -16, 0, 64, -56, 0, 0, -256, 324, 224]
? mfcoef(F,100000) \\ instantaneous
%7 = 41304367104
Odd dimensions are supported, corresponding to forms of half-integral weight:
? [mf,F,v] = mffromqf(2*matid(3));
? mfisequal(F, mfpow(mfTheta(),3))
%2 = 1
? mfcoefs(F, 32) \\ illustrate Legendre’s 3-square theorem
%3 = [ 1,
6, 12, 8, 6, 24, 24, 0, 12,
30, 24, 24, 8, 24, 48, 0, 6,
48, 36, 24,24, 48, 24, 0, 24,
30, 72, 32, 0, 72, 48, 0, 12]
The library syntax is GEN mffromqf(GEN Q, GEN P = NULL).

3.17.38 mfgaloisprojrep(mf , F ). mf being an mf output by mfinit in weight 1, return a poly-

nomial defining the field fixed by the kernel of the projective Artin representation attached to F
(by Deligne–Serre). Currently only implemented for projective image A4 and S4 .
\\ A4 example
? mf = mfinit([4*31,1,Mod(87,124)],0);
? F = mfeigenbasis(mf)[1];
? mfgaloistype(mf,F)
%3 = -12
? pol = mfgaloisprojrep(mf,F)
%4 = x^12 + 68*x^10 + 4808*x^8 + ... + 4096
? G = galoisinit(pol); galoisidentify(G)
%5 = [12,3] \\A4
? pol4 = polredbest(galoisfixedfield(G,G.gen[3], 1))
%6 = x^4 + 7*x^2 - 2*x + 14
? polgalois(pol4)
%7 = [12, 1, 1, "A4"]
? factor(nfdisc(pol4))
%8 =
[ 2 4]
[31 2]
\\ S4 example
? mf = mfinit([4*37,1,Mod(105,148)],0);
? F = mfeigenbasis(mf)[1];
? mfgaloistype(mf,F)
%11 = -24
? pol = mfgaloisprojrep(mf,F)
%12 = x^24 + 24*x^22 + 256*x^20 + ... + 255488256
? G = galoisinit(pol); galoisidentify(G)
%13 = [24, 12] \\S4
? pol4 = polredbest(galoisfixedfield(G,G.gen[3..4], 1))
%14 = x^4 - x^3 + 5*x^2 - 7*x + 12

552
 ? polgalois(pol4)
%15 = [24, -1, 1, "S4"]
? factor(nfdisc(pol4))
%16 =
[ 2 2]
[37 3]

The library syntax is GEN mfgaloisprojrep(GEN mf, GEN F, long prec).

3.17.39 mfgaloistype(NK , {F }). NK being either [N,1,CHI] or an mf output by mfinit in weight

1, gives the vector of types of Galois representations attached to each cuspidal eigenform, unless
the modular form F is specified, in which case only for F (note that it is not tested whether F
belongs to the correct modular form space, nor whether it is a cuspidal eigenform). Types A4 , S4 ,
A5 are represented by minus their cardinality −12, −24, or −60, and type Dn is represented by its
cardinality, the integer 2n:

? mfgaloistype([124,1, Mod(67,124)]) \\ A4
%1 = [-12]
? mfgaloistype([148,1, Mod(105,148)]) \\ S4
%2 = [-24]
? mfgaloistype([633,1, Mod(71,633)]) \\ D10, A5
%3 = [10, -60]
? mfgaloistype([239,1, -239]) \\ D6, D10, D30
%4 = [6, 10, 30]
? mfgaloistype([71,1, -71])
%5 = [14]
? mf = mfinit([239,1, -239],0); F = mfeigenbasis(mf)[2];
? mfgaloistype(mf, F)
%7 = 10

The function may also return 0 as a type when it failed to determine it; in this case the correct
type is either −12 or −60, and most likely −12.

The library syntax is GEN mfgaloistype(GEN NK, GEN F = NULL).

3.17.40 mfhecke(mf , F, n). F being a modular form in modular form space mf , returns T (n)F ,
where T (n) is the n-th Hecke operator.

553
Warning. If F is of level M < N , then T (n)F is in general not the same in Mk (Γ0 (M ), χ) and in
Mk (Γ0 (N ), χ). We take T (n) at the same level as the one used in mf.

? mf = mfinit([26,2],0); F = mfbasis(mf)[1]; mftobasis(mf,F)

%1 = [1, 0]~
? G2 = mfhecke(mf,F,2); mftobasis(mf,G2)
%2 = [0, 1]~
? G5 = mfhecke(mf,F,5); mftobasis(mf,G5)
%3 = [-2, 1]~
Modular forms of half-integral weight are supported, in which case n must be a perfect square, else
Tn will act as 0 (the operator Tp for p | N is not supported yet):

? F = mfpow(mfTheta(),3); mf = mfinit(F);
? mfisequal(mfhecke(mf,F,9), mflinear([F],[4]))
%2 = 1
(F is an eigenvector of all Tp2 , with eigenvalue p + 1 for odd p.)

Warning. When n is a large composite, resp. the square of a large composite in half-integral
weight, it is in general more efficient to use mfheckemat on the mftobasis coefficients:

? mfcoefs(mfhecke(mf,F,3^10), 10)
time = 917 ms.
%3 = [324, 1944, 3888, 2592, 1944, 7776, 7776, 0, 3888, 9720, 7776]
? M = mfheckemat(mf,3^10) \\ instantaneous
%4 =
[324]
? G = mflinear(mf, M*mftobasis(mf,F));
? mfcoefs(G, 10) \\ instantaneous
%6 = [324, 1944, 3888, 2592, 1944, 7776, 7776, 0, 3888, 9720, 7776]
The library syntax is GEN mfhecke(GEN mf, GEN F, long n).

3.17.41 mfheckemat(mf , vecn). If vecn is an integer, matrix of the Hecke operator T (n) on the
basis formed by mfbasis(mf). If it is a vector, vector of such matrices, usually faster than calling
each one individually.

? mf=mfinit([32,4],0); mfheckemat(mf,3)
%1 =
[0 44 0]
[1 0 -10]
[0 -2 0]
? mfheckemat(mf,[5,7])
%2 = [[0, 0, 220; 0, -10, 0; 1, 0, 12], [0, 88, 0; 2, 0, -20; 0, -4, 0]]
The library syntax is GEN mfheckemat(GEN mf, GEN vecn).

554
3.17.42 mfinit(NK , {space = 4}). Create the space of modular forms corresponding to the data
contained in NK and space. NK is a vector which can be either [N, k] (N level, k weight) correspond-
ing to a subspace of Mk (Γ0 (N )), or [N, k, CHI ] (CHI a character) corresponding to a subspace of
Mk (Γ0 (N ), χ). Alternatively, it can be a modular form F or modular form space, in which case we
use mfparams to define the space parameters.
The subspace is described by the small integer space: 0 for the newspace Sknew (Γ0 (N ), χ), 1
for the cuspidal space Sk , 2 for the oldspace Skold , 3 for the space of Eisenstein series Ek and 4 for
the full space Mk .
Wildcards. For given level and weight, it is advantageous to compute simultaneously spaces
attached to different Galois orbits of characters, especially in weight 1. The parameter CHI may
be set to 0 (wildcard), in which case we return a vector of all mfinit(s) of non trivial spaces in
Sk (Γ1 (N )), one for each Galois orbit (see znchargalois). One may also set CHI to a vector of
characters and we return a vector of all mfinits of subspaces of Mk (G0 (N ), χ) for χ in the list, in
the same order. In weight 1, only S1new , S1 and E1 support wildcards.
The output is a technical structure S, or a vector of structures if CHI was a wildcard, which
contains the following information: [N, k, χ] is given by mfparams(S), the space dimension is
mfdim(S) and a C-basis for the space is mfbasis(S). The structure is entirely algebraic and
does not depend on the current realbitprecision.
? S = mfinit([36,2], 0); \\ new space
? mfdim(S)
%2 = 1
? mfparams
%3 = [36, 2, 1, y] \\ trivial character
? f = mfbasis(S)[1]; mfcoefs(f,10)
%4 = [0, 1, 0, 0, 0, 0, 0, -4, 0, 0, 0]
? vS = mfinit([36,2,0],0); \\ with wildcard
? #vS
%6 = 4 \\ 4 non trivial spaces (mod Galois action)
? apply(mfdim,vS)
%7 = [1, 2, 1, 4]
? mfdim([36,2,0], 0)
%8 = [[1, Mod(1, 36), 1, 0], [2, Mod(35, 36), 2, 0], [3, Mod(13, 36), 1, 0],
[6, Mod(11, 36), 4, 0]]
The library syntax is GEN mfinit(GEN NK, long space).

3.17.43 mfisCM(F ). Tests whether the eigenform F is a CM form. The answer is 0 if it is not,
and if it is, either the unique negative discriminant of the CM field, or the pair of two negative
discriminants of CM fields, this latter case occurring only in weight 1 when the projective image is
D2 = C2 × C2 , i.e., coded 4 by mfgaloistype.
? F = mffromell(ellinit([0,1]))[2]; mfisCM(F)
%1 = -3
? mf = mfinit([39,1,-39],0); F=mfeigenbasis(mf)[1]; mfisCM(F)
%2 = Vecsmall([-3, -39])
? mfgaloistype(mf)
%3 = [4]
The library syntax is GEN mfisCM(GEN F).

555
3.17.44 mfisequal(F, G, {lim = 0}). Checks whether the modular forms F and G are equal. If
lim is nonzero, only check equality of the first lim + 1 Fourier coefficients and the function then
also applies to generalized modular forms.
? D = mfDelta(); F = mfderiv(D);
? G = mfmul(mfEk(2), D);
? mfisequal(F, G)
%2 = 1
The library syntax is long mfisequal(GEN F, GEN G, long lim).

3.17.45 mfisetaquo(f, {flag = 0}). If the generalized modular form f is a holomorphic eta
quotient, return the eta quotient matrix, else return 0. If flagis set, also accept meromorphic eta
quotients: check whether f = q −v(g) g(q) for some eta quotient g; if so, return the eta quotient
matrix attached to g, else return 0. See mffrometaquo.
? mfisetaquo(mfDelta())
%1 =
[1 24]
? f = mffrometaquo([1,1;23,1]);
? mfisetaquo(f)
%3 =
[ 1 1]
[23 1]
? f = mffrometaquo([1,-24], 1);
? mfisetaquo(f) \\ nonholomorphic
%5 = 0
? mfisetaquo(f,1)
%6 =
[1 -24]
The library syntax is GEN mfisetaquo(GEN f, long flag).

3.17.46 mfkohnenbasis(mf ). mf being a cuspidal space of half-integral weight k ≥ 3/2 with level
N and character χ, gives a basis B of the Kohnen +-space of mf as a matrix whose columns are
the coefficients of B on the basis of mf. The conductor of either χ or χ · (−4/.) must divide N/4.
? mf = mfinit([36,5/2],1); K = mfkohnenbasis(mf); K~
%1 =
[-1 0 0 2 0 0]
[ 0 0 0 0 1 0]
? (mfcoefs(mf,20) * K)~
%4 =
[0 -1 0 0 2 0 0 0 0 0 0 0 0 -6 0 0 8 0 0 0 0]
[0 0 0 0 0 1 0 0 -2 0 0 0 0 0 0 0 0 1 0 0 2]
? mf = mfinit([40,3/2,8],1); mfkohnenbasis(mf)
*** at top-level: mfkohnenbasis(mf)
*** ^-----------------
*** mfkohnenbasis: incorrect type in mfkohnenbasis [incorrect CHI] (t_VEC).

556
 In the final example both χ = (8/.) and χ · (−4/.) have conductor 8, which does not divide
N/4 = 10.
The library syntax is GEN mfkohnenbasis(GEN mf).

3.17.47 mfkohnenbijection(mf ). mf being a cuspidal space of half-integral weight, returns

[mf2,M,K,shi], where M is a matrix giving a Hecke-module isomorphism from the cuspidal space
mf2 giving S2k−1 (Γ0 (N ), χ2 ) to the Kohnen +-space Sk+ (Γ0 (4N ), χ), K represents a basis B of the
Kohnen +-space as a matrix whose columns are the coefficients of B on the basis of mf; shi is a
vector of pairs (ti , ni ) gives the linear combination of Shimura lifts giving M −1 : ti is a squarefree
positive integer and ni is a small nonzero integer.
? mf=mfinit([60,5/2],1); [mf2,M,K,shi]=mfkohnenbijection(mf); M
%2 =
[-3 0 5/2 7/2]
[ 1 -1/2 -7 -7]
[ 1 1/2 0 -3]
[ 0 0 5/2 5/2]
? shi
%2 = [[1, 1], [2, 1]]
This last command shows that the map giving the bijection is the sum of the Shimura lift with
t = 1 and the one with t = 2.
Since it gives a bijection of Hecke modules, this matrix can be used to transport modular
form data from the easily computed space of level N and weight 2k − 1 to the more difficult space
of level 4N and weight k: matrices of Hecke operators, new space, splitting into eigenspaces and
eigenforms. Examples:
? K^(-1)*mfheckemat(mf,121)*K /* matrix of T_11^2 on K. Slowish. */
time = 1,280 ms.
%1 =
[ 48 24 24 24]
[ 0 32 0 -20]
[-48 -72 -40 -72]
[ 0 0 0 52]
? M*mfheckemat(mf2,11)*M^(-1) /* instantaneous via T_11 on S_{2k-1} */
time = 0 ms.
%2 =
[ 48 24 24 24]
[ 0 32 0 -20]
[-48 -72 -40 -72]
[ 0 0 0 52]
? mf20=mfinit(mf2,0); [mftobasis(mf2,b) | b<-mfbasis(mf20)]
%3 = [[0, 0, 1, 0]~, [0, 0, 0, 1]~]
? F1=M*[0,0,1,0]~
%4 = [1/2, 1/2, -3/2, -1/2]~
? F2=M*[0,0,0,1]~

557
 %5 = [3/2, 1/2, -9/2, -1/2]
? K*F1
%6 = [1, 0, 0, 1, 1, 0, 0, 1, -3, 0, 0, -3, 0, 0]~
? K*F2
%7 = [3, 0, 0, 3, 1, 0, 0, 1, -9, 0, 0, -3, 0, 0]~
+
This gives a basis of the new space of S5/2 (Γ0 (60)) expressed on the initial basis of S5/2 (Γ0 (60)).
If we want the eigenforms, we write instead:

? BE=mfeigenbasis(mf20);[E1,E2]=apply(x->K*M*mftobasis(mf2,x),BE)
%1 = [[1, 0, 0, 1, 0, 0, 0, 0, -3, 0, 0, 0, 0, 0]~,\
[0, 0, 0, 0, 1, 0, 0, 1, 0, 0, 0, -3, 0, 0]~
? EI1 = mflinear(mf, E1); EI2=mflinear(mf, E2);

These are the two eigenfunctions in the space mf, the first (resp., second) will have Shimura image
a multiple of BE[1] (resp., BE[2]). The function mfkohneneigenbasis does this directly.

The library syntax is GEN mfkohnenbijection(GEN mf).

3.17.48 mfkohneneigenbasis(mf , bij ). mf being a cuspidal space of half-integral weight

k ≥ 3/2 and bij being the output of mfkohnenbijection(mf), outputs a 3-component vector
[mf0,BNEW,BEIGEN], where BNEW and BEIGEN are two matrices whose columns are the coefficients
of a basis of the Kohnen new space and of the eigenforms on the basis of mf respectively, and mf0
is the corresponding new space of integral weight 2k − 1.

? mf=mfinit([44,5/2],1);bij=mfkohnenbijection(mf);
? [mf0,BN,BE]=mfkohneneigenbasis(mf,bij);
? BN~
%2 =
[2 0 0 -2 2 0 -8]
[2 0 0 4 14 0 -32]
? BE~
%3 = [1 0 0 Mod(y-1, y^2-3) Mod(2*y+1, y^2-3) 0 Mod(-4*y-4, y^2-3)]
? lift(mfcoefs(mf,20)*BE[,1])
%4 = [0, 1, 0, 0, y - 1, 2*y + 1, 0, 0, 0, -4*y - 4, 0, 0,\
-5*y + 3, 0, 0, 0, -6, 0, 0, 0, 7*y + 9]~

The library syntax is GEN mfkohneneigenbasis(GEN mf, GEN bij).

3.17.49 mflinear(vF , v). vF being a vector of generalized modular forms and v a vector of co-
efficients of same length, compute the linear combination of the entries of vF with coefficients
v.

558
Note. Use this in particular to subtract two forms F and G (with vF = [F, G] and v = [1, −1]),
or to multiply an form by a scalar λ (with vF = [F ] and v = [λ]).
? D = mfDelta(); G = mflinear([D],[-3]);
? mfcoefs(G,4)
%2 = [0, -3, 72, -756, 4416]
For user convenience, we allow
• a modular form space mf as a vF argument, which is understood as mfbasis(mf);
• in this case, we also allow a modular form f as v, which is understood as mftobasis(mf , f ).
? T = mfpow(mfTheta(),7); F = mfShimura(T,-3); \\ Shimura lift for D=-3
? mfcoefs(F,8)
%2 = [-5/9, 280, 9240, 68320, 295960, 875280, 2254560, 4706240, 9471000]
? mf = mfinit(F); G = mflinear(mf,F);
? mfcoefs(G,8)
%4 = [-5/9, 280, 9240, 68320, 295960, 875280, 2254560, 4706240, 9471000]
This last construction allows to replace a general modular form by a simpler linear combination of
basis functions, which is often more efficient:
? T10=mfpow(mfTheta(),10); mfcoef(T10, 10^4) \\ direct evaluation
time = 399 ms.
%5 = 128205250571893636
? mf=mfinit(T10); F=mflinear(mf,T10); \\ instantaneous
? mfcoef(F, 10^4) \\ after linearization
time = 67 ms.
%7 = 128205250571893636
The library syntax is GEN mflinear(GEN vF, GEN v).

3.17.50 mfmanin(FS ). Given the modular symbol F S associated to an eigenform F by mfsym-

bol(mf,F), computes the even and odd special polynomials as well as the even and odd periods
ω + and ω − as a vector [[P + , P − ], [ω + , ω − , r]], where r = =(ω + ω − )/ < F, F >. If F has several
embeddings into C, give the vector of results corresponding to each embedding.
? D=mfDelta(); mf=mfinit(D); DS=mfsymbol(mf,D);
? [pols,oms]=mfmanin(DS); pols
%2 = [[4*x^9 - 25*x^7 + 42*x^5 - 25*x^3 + 4*x],\
[-36*x^10 + 691*x^8 - 2073*x^6 + 2073*x^4 - 691*x^2 + 36]]
? oms
%3 = [0.018538552324740326472516069364750571812,\
-0.00033105361053212432521308691198949874026*I, 4096/691]
? mf=mfinit([11,2],0); F=mfeigenbasis(mf)[1]; FS=mfsymbol(mf,F);
? [pols,oms]=mfmanin(FS);pols
%5 = [[0, 0, 0, 1, 1, 0, 0, -1, -1, 0, 0, 0],\
[2, 0, 10, 5, -5, -10, -10, -5, 5, 10, 0, -2]]
? oms[3]
%6 = 24/5
The library syntax is GEN mfmanin(GEN FS, long bitprec).

559
3.17.51 mfmul(F, G). Multiply the two generalized modular forms F and G.

? E4 = mfEk(4); G = mfmul(mfmul(E4,E4),E4);
? mfcoefs(G, 4)
%2 = [1, 720, 179280, 16954560, 396974160]
? mfcoefs(mfpow(E4,3), 4)
%3 = [1, 720, 179280, 16954560, 396974160]

The library syntax is GEN mfmul(GEN F, GEN G).

3.17.52 mfnumcusps(N ). Number of cusps of Γ0 (N )

? mfnumcusps(24)
%1 = 8
? mfcusps(24)
%1 = [0, 1/2, 1/3, 1/4, 1/6, 1/8, 1/12, 1/24]

The library syntax is GEN mfnumcusps(GEN N).

3.17.53 mfparams(F ). If F is a modular form space, returns [N,k,CHI,space,Φ], level, weight,

character χ, and space code; where Φ is the cyclotomic polynomial defining the field of values of
CHI. If F is a generalized modular form, returns [N,k,CHI,P,Φ], where P is the (polynomial giving
the) field of definition of F as a relative extension of the cyclotomic field Q(χ) = Q[t]/(Φ): in that
case the level N may be a multiple of the level of F and the polynomial P may define a larger field
than Q(F ). If you want the true level of F from this result, use mfconductor(mfinit(F),F). The
polynomial P defines an extension of Q(χ) = Q[t]/(Φ(t)); it has coefficients in that number field
(polmods in t).

In contrast with mfparams(F)[4] which always gives the polynomial P defining the relative
extension Q(F )/Q(χ), the member function F .mod returns the polynomial used to define Q(F )
over Q (either a cyclotomic polynomial or a polynomial with cyclotomic coefficients).

? E1 = mfeisenstein(4,-3,-4); E2 = mfeisenstein(3,5,-7); E3 = mfmul(E1,E2);

? apply(mfparams, [E1,E2,E3])
%2 = [[12, 4, 12, y, t-1], [35, 3, -35, y, t-1], [420, 7, -420, y, t-1]]
? mf = mfinit([36,2,Mod(13,36)],0); [f] = mfeigenbasis(mf); mfparams(mf)
%3 = [36, 2, Mod(13, 36), 0, t^2 + t + 1]
? mfparams(f)
%4 = [36, 2, Mod(13, 36), y, t^2 + t + 1]
? f.mod
%5 = t^2 + t + 1
? mf = mfinit([36,4,Mod(13,36)],0); [f] = mfeigenbasis(mf);
? lift(mfparams(f))
%7 = [36, 4, 13, y^3 + (2*t-2)*y^2 + (-4*t+6)*y + (10*t-1), t^2+t+1]

The library syntax is GEN mfparams(GEN F).

560
3.17.54 mfperiodpol(mf , f, {flag = 0}). Period polynomial of the cuspidal part of the form f , in
R i∞
other words 0 (X − τ )k−2 f (τ ) dτ . If flag is 0, ordinary period polynomial. If it is 1 or −1, even
or odd part of that polynomial. f can also be the modular symbol output by mfsymbol(mf,f).
? D = mfDelta(); mf = mfinit(D,0);
? PP = mfperiodpol(mf, D, -1); PP/=polcoef(PP, 1); bestappr(PP)
%1 = x^9 - 25/4*x^7 + 21/2*x^5 - 25/4*x^3 + x
? PM = mfperiodpol(mf, D, 1); PM/=polcoef(PM, 0); bestappr(PM)
%2 = -x^10 + 691/36*x^8 - 691/12*x^6 + 691/12*x^4 - 691/36*x^2 + 1
The library syntax is GEN mfperiodpol(GEN mf, GEN f, long flag, long bitprec).

3.17.55 mfperiodpolbasis(k, {flag = 0}). Basis of period polynomials for weight k. If flag=1 or
−1, basis of odd or even period polynomials.
? mfperiodpolbasis(12,1)
%1 = [x^8 - 3*x^6 + 3*x^4 - x^2, x^10 - 1]
? mfperiodpolbasis(12,-1)
%2 = [4*x^9 - 25*x^7 + 42*x^5 - 25*x^3 + 4*x]
The library syntax is GEN mfperiodpolbasis(long k, long flag).

3.17.56 mfpetersson(fs, {gs}). Petersson scalar product of the modular forms f and g belonging
to the same modular form space mf, given by the corresponding “modular symbols” fs and gs
output by mfsymbol (also in weight 1 and half-integral weight, where symbols do not exist). If
gs is omitted it is understood to be equal to fs. The scalar product is normalized by the factor
1/[Γ : Γ0 (N )]. Note that f and g can both be noncuspidal, in which case the program returns an
error if the product is divergent. If the fields of definition Q(f ) and Q(g) are equal to Q(χ) the
result is a scalar. If [Q(f ) : Q(χ)] = d > 1 and [Q(g) : Q(χ)] = e > 1 the result is a d × e matrix
corresponding to all the embeddings of f and g. In the intermediate cases d = 1 or e = 1 the result
is a row or column vector.
? D=mfDelta(); mf=mfinit(D); DS=mfsymbol(mf,D); mfpetersson(DS)
%1 = 1.0353620568043209223478168122251645932 E-6
? mf=mfinit([11,6],0);B=mfeigenbasis(mf);BS=vector(#B,i,mfsymbol(mf,B[i]));
? mfpetersson(BS[1])
%3 = 1.6190120685220988139111708455305245466 E-5
? mfpetersson(BS[1],BS[2])
%4 = [-3.826479006582967148 E-42 - 2.801547395385577002 E-41*I,\
1.6661127341163336125 E-41 + 1.1734725972345985061 E-41*I,\
0.E-42 - 6.352626992842664490 E-41*I]~
? mfpetersson(BS[2])
%5 =
[ 2.7576133733... E-5 2.0... E-42 6.3... E-43 ]
[ -4.1... E-42 6.77837030070... E-5 3.3...E-42 ]
[ -6.32...E-43 3.6... E-42 2.27268958069... E-5]
? mf=mfinit([23,2],0); F=mfeigenbasis(mf)[1]; FS=mfsymbol(mf,F);
? mfpetersson(FS)
%5 =
[0.0039488965740025031688548076498662860143 -3.56 ... E-40]

561
 [ -3.5... E-40 0.0056442542987647835101583821368582485396]
Noncuspidal example:
? E1=mfeisenstein(5,1,-3);E2=mfeisenstein(5,-3,1);
? mf=mfinit([12,5,-3]); cusps=mfcusps(12);
? apply(x->mfcuspval(mf,E1,x),cusps)
%3 = [0, 0, 1, 0, 1, 1]
? apply(x->mfcuspval(mf,E2,x),cusps)
%4 = [1/3, 1/3, 0, 1/3, 0, 0]
? E1S=mfsymbol(mf,E1);E2S=mfsymbol(mf,E2);
? mfpetersson(E1S,E2S)
%6 = -1.884821671646... E-5 - 1.9... E-43*I
Weight 1 and 1/2-integral weight example:
? mf=mfinit([23,1,-23],1);F=mfbasis(mf)[1];FS=mfsymbol(mf,F);
? mfpetersson(mf,FS)
%2 = 0.035149946790370230814006345508484787443
? mf=mfinit([4,9/2],1);F=mfbasis(mf)[1];FS=mfsymbol(mf,F);
? mfpetersson(FS)
%4 = 0.00015577084407139192774373662467908966030
The library syntax is GEN mfpetersson(GEN fs, GEN gs = NULL).

3.17.57 mfpow(F, n). Compute F n , where n is an integer and F is a generalized modular form:
? G = mfpow(mfEk(4), 3); \\ E4^3
? mfcoefs(G, 4)
%2 = [1, 720, 179280, 16954560, 396974160]
The library syntax is GEN mfpow(GEN F, long n).

3.17.58 mfsearch(NK , V, {space}). NK being of the form [N,k] with k possibly half-integral,
search for a modular form with rational coefficients, of weight k and level N , whose initial coefficients
a(0),... are equal to V ; space specifies the modular form spaces in which to search, in mfinit or
mfdim notation. The output is a list of matching forms with that given level and weight. Note that
the character is of the form (D/.), where D is a (positive or negative) fundamental discriminant
dividing N . The forms are sorted by increasing |D|.
The parameter N can be replaced by a vector of allowed levels, in which case the list of forms
is sorted by increasing level, then increasing |D|. If a form is found at level N , any multiple of N
with the same D is not considered. Some useful possibilities are
• [N1 ..N2 ]: all levels between N1 and N2 , endpoints included;
• F * [N1 ..N2 ]: same but levels divisible by F ;
• divisors(N0 ): all levels dividing N0 .
Note that this is different from mfeigensearch, which only searches for rational eigenforms.
? F = mfsearch([[1..40], 2], [0,1,2,3,4], 1); #F
%1 = 3
? [ mfparams(f)[1..3] | f <- F ]
%2 = [[38, 2, 1], [40, 2, 8], [40, 2, 40]]

562
 ? mfcoefs(F[1],10)
%3 = [0, 1, 2, 3, 4, -5, -8, 1, -7, -5, 7]
The library syntax is GEN mfsearch(GEN NK, GEN V, long space).

3.17.59 mfshift(F, s). Divide the generalized modular form F by q s , omitting the remainder if
there is one. One can have s < 0.
? D=mfDelta(); mfcoefs(mfshift(D,1), 4)
%1 = [1, -24, 252, -1472, 4830]
? mfcoefs(mfshift(D,2), 4)
%2 = [-24, 252, -1472, 4830, -6048]
? mfcoefs(mfshift(D,-1), 4)
%3 = [0, 0, 1, -24, 252]
The library syntax is GEN mfshift(GEN F, long s).

3.17.60 mfshimura(mf , F, {D = 1}). F being a modular form of half-integral weight k ≥ 3/2

and t a positive squarefree integer, returns the Shimura lift G of weight 2k − 1 corresponding to D.
This function returns [mf2 , G, v] where mf2 is a modular form space containing G and v expresses
G in terms of mfbasis(mf2 ); so that G is mflinear(mf2 , v).
? F = mfpow(mfTheta(), 7); mf = mfinit(F);
? [mf2, G, v] = mfshimura(mf, F, 3); mfcoefs(G,5)
%2 = [-5/9, 280, 9240, 68320, 295960, 875280]
? mfparams(G) \\ the level may be lower than expected
%3 = [1, 6, 1, y, t - 1]
? mfparams(mf2)
%4 = [2, 6, 1, 4, t - 1]
? v
%5 = [280, 0]~
? mfcoefs(mf2, 5)
%6 =
[-1/504 -1/504]
[ 1 0]
[ 33 1]
[ 244 0]
[ 1057 33]
[ 3126 0]
? mf = mfinit([60,5/2],1); F = mflinear(mf,mfkohnenbasis(mf)[,1]);
? mfparams(mfshimura(mf,F)[2])
%8 = [15, 4, 1, y, t - 1]
? mfparams(mfshimura(mf,F,6)[2])
%9 = [15, 4, 1, y, t - 1]
The library syntax is GEN mfshimura(GEN mf, GEN F, long D).

563
3.17.61 mfslashexpansion(mf , f, g, n, flrat, {&params}). Let mf be a modular form space in
level N , f a modular form belonging to mf and let g be in M2+ (Q). This function computes the
Fourier expansion of f |k g to n terms. We first describe the behaviour when flrat is 0: the result
is a vector v of floating point complex numbers such that
X
f |k g(τ ) = q α v[m + 1]q m/w ,
m≥0

where q = e(τ ), w is the width of the cusp g(i∞) (namely (N/(c2 , N ) if g is integral) and α is a
rational number. If params is given, it is set to the parameters [α, w, matid(2)].
If flrat is 1, the program tries to rationalize the expression, i.e., to express the coefficients as
rational numbers or polmods. We write g = λ · M · A where λ ∈ Q∗ , M ∈ SL2 (Z) and A = [a, b; 0, d]
is upper triangular, integral and primitive with a > 0, d > 0 and 0 ≤ b < d. Let α and w by the
parameters attached to the expansion of F := f |k M as above, i.e.
X
F (τ ) = q α v[m + 1]q m/w .
m≥0

The function returns the expansion v of F = f |k M and sets the parameters to [α, w, A]. Finally,
the desired expansion is (a/d)k/2 F (τ + b/d). The latter is identical to the returned expansion when
A is the identity, i.e. when g ∈ PSL2 (Z). If this is not the case, the expansion differs from v by the
multiplicative constant (a/d)k/2 e(αb/(dw)) and a twist by a root of unity q 1/w → e(b/(dw))q 1/w .
The complications introduced by this extra matrix A allow to recognize the coefficients in a much
smaller cyclotomic field, hence to obtain a simpler description overall. (Note that this rationaliza-
tion step may result in an error if the program cannot perform it.)
? mf = mfinit([32,4],0); f = mfbasis(mf)[1];
? mfcoefs(f, 10)
%2 = [0, 3, 0, 0, 0, 2, 0, 0, 0, 47, 0]
? mfatk = mfatkininit(mf,32); mfcoefs(mfatkin(mfatk,f),10) / mfatk[3]
%3 = [0, 1, 0, 16, 0, 22, 0, 32, 0, -27, 0]
? mfatk[3] \\ here normalizing constant C = 1, but need in general
%4 = 1
? mfslashexpansion(mf,f,[0,-1;1,0],10,1,&params) * 32^(4/2)
%5 = [0, 1, 0, 16, 0, 22, 0, 32, 0, -27, 0]
? params
%6 = [0, 32, [1, 0; 0, 1]]
? mf = mfinit([12,8],0); f = mfbasis(mf)[1];
? mfslashexpansion(mf,f,[1,0;2,1],7,0)
%7 = [0, 0, 0, 0.6666666... + 0.E-38*I, 0, -3.999999... + 6.92820...*I, 0,\
-11.99999999... - 20.78460969...*I]
? mfslashexpansion(mf,f,[1,0;2,1],7,1, &params)
%8 = [0, 0, 0, 2/3, 0, Mod(8*t, t^2+t+1), 0, Mod(-24*t-24, t^2+t+1)]
? params
%9 = [0, 3, [1, 0; 0, 1]]
If [Q(f ) : Q(χ)] > 1, the coefficients may be polynomials in y, where y is any root of the
polynomial giving the field of definition of f (f.mod or mfparams(f)[4]).
? mf=mfinit([23,2],0);f=mfeigenbasis(mf)[1];

564
 ? mfcoefs(f,5)
%1 = [Mod(0, y^2 - y - 1), Mod(1, y^2 - y - 1), Mod(-y, y^2 - y - 1),\
Mod(2*y - 1, y^2 - y - 1), Mod(y - 1, y^2 - y - 1), Mod(-2*y, y^2 - y - 1)]
? mfslashexpansion(mf,f,[1,0;0,1],5,1)
%2 = [0, 1, -y, 2*y - 1, y - 1, -2*y]
? mfslashexpansion(mf,f,[0,-1;1,0],5,1)
%3 = [0, -1/23, 1/23*y, -2/23*y + 1/23, -1/23*y + 1/23, 2/23*y]

Caveat. In half-integral weight, we define the “slash” operation as

2k
(f |k g)(τ ) := (cτ + d)−1/2 f (g · τ ),

with the principal determination of the square root. In particular, the standard cocycle condition
is no longer satisfied and we only have f |(gg 0 ) = ±(f |g)|g 0 .
The library syntax is GEN mfslashexpansion(GEN mf, GEN f, GEN g, long n, long flrat,
GEN *params = NULL, long prec).

3.17.62 mfspace(mf , {f }). Identify the modular space mf , resp. the modular form f in mf if
present, as the flag given to mfinit. Returns 0 (newspace), 1 (cuspidal space), 2 (old space), 3
(Eisenstein space) or 4 (full space).
? mf = mfinit([1,12],1); mfspace(mf)
%1 = 1
? mfspace(mf, mfDelta())
%2 = 0 \\ new space
This function returns −1 when the form f is modular but does not belong to the space.
? mf = mfinit([1,2]; mfspace(mf, mfEk(2))
%3 = -1
When f is not modular and is for instance only quasi-modular, the function returns nonsense:
? M6 = mfinit([1,6]);
? dE4 = mfderiv(mfEk(4)); \\ not modular !
? mfspace(M6,dE4) \\ asserts (wrongly) that E4’ belongs to new space
%3 = 0
The library syntax is long mfspace(GEN mf, GEN f = NULL).

3.17.63 mfsplit(mf , {dimlim = 0}, {flag = 0}). mf from mfinit with integral weight containing
the new space (either the new space itself or the cuspidal space or the full space), and preferably
the newspace itself for efficiency, split the space into Galois orbits of eigenforms of the newspace,
satisfying various restrictions.
The functions returns [vF, vK], where vF gives (Galois orbit of) eigenforms and vK is a list
of polynomials defining each Galois orbit. The eigenforms are given in mftobasis format, i.e. as a
matrix whose columns give the forms with respect to mfbasis(mf).
If dimlim is set, only the Galois orbits of dimension ≤ dimlim are computed (i.e. the rational
eigenforms if dimlim = 1 and the character is real). This can considerably speed up the function
when a Galois orbit is defined over a large field.

565
 flag speeds up computations when the dimension is large: if f lag = d > 0, when the dimension
of the eigenspace is > d, only the Galois polynomial is computed.

Note that the function mfeigenbasis returns all eigenforms in an easier to use format (as
modular forms which can be input as is in other functions); mfsplit is only useful when you can
restrict to orbits of small dimensions, e.g. rational eigenforms.

? mf=mfinit([11,2],0); f=mfeigenbasis(mf)[1]; mfcoefs(f,16)

%1 = [0, 1, -2, -1, ...]
? mf=mfinit([23,2],0); f=mfeigenbasis(mf)[1]; mfcoefs(f,16)
%2 = [Mod(0, z^2 - z - 1), Mod(1, z^2 - z - 1), Mod(-z, z^2 - z - 1), ...]
? mf=mfinit([179,2],0); apply(poldegree, mffields(mf))
%3 = [1, 3, 11]
? mf=mfinit([719,2],0);
? [vF,vK] = mfsplit(mf, 5); \\ fast when restricting to small orbits
time = 192 ms.
? #vF \\ a single orbit
%5 = 1
? poldegree(vK[1]) \\ of dimension 5
%6 = 5
? [vF,vK] = mfsplit(mf); \\ general case is slow
time = 2,104 ms.
? apply(poldegree,vK)
%8 = [5, 10, 45] \\ because degree 45 is large...

The library syntax is GEN mfsplit(GEN mf, long dimlim, long flag).

3.17.64 mfsturm(NK ). Gives the Sturm bound for modular forms on Γ0 (N ) and weight k, i.e.,
an upper bound for the order of the zero at infinity of a nonzero form. NK is either
Q
• a pair [N, k], in which case the bound is the floor of (kN/12) · p|N (1 + 1/p);

• or the output of mfinit in which case the exact upper bound is returned.

? NK = [96,6]; mfsturm(NK)
%1 = 97
? mf=mfinit(NK,1); mfsturm(mf)
%2 = 76
? mfdim(NK,0) \\ new space
%3 = 72

The library syntax is long mfsturm(GEN NK).

566
3.17.65 mfsymbol(mf , f ). Initialize data for working with all period polynomials of the mod-
ular form f : this is essential for efficiency for functions such as mfsymboleval, mfmanin, and
mfpetersson. An mfsymbol contains an mf structure and can always be used whenever an mf
would be needed.
? mf=mfinit([23,2],0);F=mfeigenbasis(mf)[1];
? FS=mfsymbol(mf,F);
? mfsymboleval(FS,[0,oo])
%3 = [8.762565143790690142 E-39 + 0.0877907874...*I,
-5.617375463602574564 E-39 + 0.0716801031...*I]
? mfpetersson(FS)
%4 =
[0.0039488965740025031688548076498662860143 1.2789721111175127425 E-40]
[1.2630501762985554269 E-40 0.0056442542987647835101583821368582485396]
By abuse of language, initialize data for working with mfpetersson in weight 1 and half-integral
weight (where no symbol exist); the mf argument may be an mfsymbol attached to a form on the
space, which avoids recomputing data independent of the form.
? mf=mfinit([12,9/2],1); F=mfbasis(mf);
? fs=mfsymbol(mf,F[1]);
time = 476 ms
? mfpetersson(fs)
%2 = 1.9722437519492014682047692073275406145 E-5
? f2s = mfsymbol(mf,F[2]);
time = 484 ms.
? mfpetersson(f2s)
%4 = 1.2142222531326333658647877864573002476 E-5
? gs = mfsymbol(fs,F[2]); \\ re-use existing symbol, a little faster
time = 430 ms.
? mfpetersson(gs) == %4 \\ same value
%6 = 1
For simplicity, we also allow mfsymbol(f) instead of mfsymbol(mfinit(f), f):
The library syntax is GEN mfsymbol(GEN mf, GEN f = NULL, long bitprec).

3.17.66 mfsymboleval(fs, path, {ga = id }). Evaluation of the modular symbol f s (corresponding
to the modular form f ) output by mfsymbol on the given path path, where path is either a
vector [s1 , s2 ] or an integral matrix [a, b; c, d] representing the path [a/c, b/d]. In both cases s1
or s2 (or a/c or b/d) can also be elements of the upper half-plane. To avoid possibly lengthy
mfsymbol computations, the program also accepts f s of the form [mf,F], but in that case s1 and
Rs2s2are limited
k−2
to oo and elements of the upper half-plane. The result is the polynomial equal to
s1
(X − τ ) F (τ ) dτ , the integral being computed along a geodesic joining s1 and s2 . If ga in
GL+ 2 (Q) is given, replace F by F |k γ. Note that if the integral diverges, the result will be a rational
function. If the field of definition Q(f ) is larger than Q(χ) then f can be embedded into C in
d = [Q(f ) : Q(χ)] ways, in which case a vector of the d results is returned.
? mf=mfinit([35,2],1);f=mfbasis(mf)[1];fs=mfsymbol(mf,f);
? mfsymboleval(fs,[0,oo])
%1 = 0.31404011074188471664161704390256378537*I

567
 ? mfsymboleval(fs,[1,3;2,5])
%2 = -0.1429696291... - 0.2619975641...*I
? mfsymboleval(fs,[I,2*I])
%3 = 0.00088969563028739893631700037491116258378*I
? E2=mfEk(2);E22=mflinear([E2,mfbd(E2,2)],[1,-2]);mf=mfinit(E22);
? E2S = mfsymbol(mf,E22);
? mfsymboleval(E2S,[0,1])
%6 = (-1.00000...*x^2 + 1.00000...*x - 0.50000...)/(x^2 - x)

The rational function which is given in case the integral diverges is easy to interpret. For
instance:

? E4=mfEk(4);mf=mfinit(E4);ES=mfsymbol(mf,E4);
? mfsymboleval(ES,[I,oo])
%2 = 1/3*x^3 - 0.928067...*I*x^2 - 0.833333...*x + 0.234978...*I
? mfsymboleval(ES,[0,I])
%3 = (-0.234978...*I*x^3 - 0.833333...*x^2 + 0.928067...*I*x + 0.333333...)/x

mfsymboleval(ES,[a,oo]) is the limit as T → ∞ of

Z iT
(X − τ )k−2 F (τ ) dτ + a(0)(X − iT )k−1 /(k − 1) ,
a

where a(0) is the 0th coefficient of F at infinity. Similarly, mfsymboleval(ES,[0,a]) is the limit
as T → ∞ of Z a
(X − τ )k−2 F (τ ) dτ + b(0)(1 + iT X)k−1 /(k − 1) ,
i/T

where b(0) is the 0th coefficient of F |k S at infinity.

The library syntax is GEN mfsymboleval(GEN fs, GEN path, GEN ga = NULL, long bit-
prec).

3.17.67 mftaylor(F, n, {flreal = 0}). F being a form in Mk (SL2 (Z)), computes the first n + 1
canonical Taylor expansion of F around τ = I. If flreal=0, computes only an algebraic equivalence
class. If flreal is set, compute pn such that for τ close enough to I we have
X
f (τ ) = (2I/(τ + I))k pn ((τ − I)/(τ + I))n .
n>=0

? D=mfDelta();
? mftaylor(D,8)
%2 = [1/1728, 0, -1/20736, 0, 1/165888, 0, 1/497664, 0, -11/3981312]

The library syntax is GEN mftaylor(GEN F, long n, long flreal, long prec).

568
3.17.68 mftobasis(mf , F, {flag = 0}). Coefficients of the form F on the basis given by mfba-
sis(mf). A q-expansion or vector of coefficients can also be given instead of F , but in this case an
error message may occur if the expansion is too short. An error message is also given if F does not
belong to the modular form space. If flag is set, instead of error messages the output is an affine
space of solutions if a q-expansion or vector of coefficients is given, or the empty column otherwise.

? mf = mfinit([26,2],0); mfdim(mf)
%1 = 2
? F = mflinear(mf,[a,b]); mftobasis(mf,F)
%2 = [a, b]~
A q-expansion or vector of coefficients can also be given instead of F .

? Th = 1 + 2*sum(n=1, 8, q^(n^2), O(q^80));

? mf = mfinit([4,5,Mod(3,4)]);
? mftobasis(mf, Th^10)
%3 = [64/5, 4/5, 32/5]~
If F does not belong to the corresponding space, the result is incorrect and simply matches
the coefficients of F up to some bound, and the function may either return an empty column or an
error message. If flag is set, there are no error messages, and the result is an empty column if F is
a modular form; if F is supplied via a series or vector of coefficients which does not contain enough
information to force a unique (potential) solution, the function returns [v, K] where v is a solution
and K is a matrix of maximal rank describing the affine space of potential solutions v + K · x.

? mf = mfinit([4,12],1);
? mftobasis(mf, q-24*q^2+O(q^3), 1)
%2 = [[43/64, -63/8, 800, 21/64]~, [1, 0; 24, 0; 2048, 768; -1, 0]]
? mftobasis(mf, [0,1,-24,252], 1)
%3 = [[1, 0, 1472, 0]~, [0; 0; 768; 0]]
? mftobasis(mf, [0,1,-24,252,-1472], 1)
%4 = [1, 0, 0, 0]~ \\ now uniquely determined
? mftobasis(mf, [0,1,-24,252,-1472,0], 1)
%5 = [1, 0, 0, 0]~ \\ wrong result: no such form exists
? mfcoefs(mflinear(mf,%), 5) \\ double check
%6 = [0, 1, -24, 252, -1472, 4830]
? mftobasis(mf, [0,1,-24,252,-1472,0])
*** at top-level: mftobasis(mf,[0,1,
*** ^--------------------
*** mftobasis: domain error in mftobasis: form does not belong to space
? mftobasis(mf, mfEk(10))
*** at top-level: mftobasis(mf,mfEk(
*** ^--------------------
*** mftobasis: domain error in mftobasis: form does not belong to space
? mftobasis(mf, mfEk(10), 1)
%7 = []~
The library syntax is GEN mftobasis(GEN mf, GEN F, long flag).

569
3.17.69 mftocoset(N, M, Lcosets). M being a matrix in SL2 (Z) and Lcosets being mfcosets(N),
a list of right cosets of Γ0 (N ), find the coset to which M belongs. The output is a pair [γ, i] such
that M = γLcosets[i], γ ∈ Γ0 (N ).
? N = 4; L = mfcosets(N);
? mftocoset(N, [1,1;2,3], L)
%2 = [[-1, 1; -4, 3], 5]
The library syntax is GEN mftocoset(long N, GEN M, GEN Lcosets).

3.17.70 mftonew(mf , F ). mf being being a full or cuspidal space with parameters [N, k, χ] and F
a cusp form in that space, returns a vector of 3-component vectors [M, d, G], where f (χ) | M | N ,
d | N/M , and G is a form in Sknew (Γ0 (M ), χ) such that F is equal to the sum of the B(d)(G) over
all these 3-component vectors.
? mf = mfinit([96,6],1); F = mfbasis(mf)[60]; s = mftonew(mf,F); #s
%1 = 1
? [M,d,G] = s[1]; [M,d]
%2 = [48, 2]
? mfcoefs(F,10)
%3 = [0, 0, -160, 0, 0, 0, 0, 0, 0, 0, -14400]
? mfcoefs(G,10)
%4 = [0, 0, -160, 0, 0, 0, 0, 0, 0, 0, -14400]
The library syntax is GEN mftonew(GEN mf, GEN F).

3.17.71 mftraceform(NK , {space = 0}). If N K = [N, k, CHI, .] as in mfinit with k integral,

gives the trace form in the corresponding subspace of Sk (Γ0 (N ), χ). The supported values for space
are 0: the newspace (default), 1: the full cuspidal space.
? F = mftraceform([23,2]); mfcoefs(F,16)
%1 = [0, 2, -1, 0, -1, -2, -5, 2, 0, 4, 6, -6, 5, 6, 4, -10, -3]
? F = mftraceform([23,1,-23]); mfcoefs(F,16)
%2 = [0, 1, -1, -1, 0, 0, 1, 0, 1, 0, 0, 0, 0, -1, 0, 0, -1]
The library syntax is GEN mftraceform(GEN NK, long space).

3.17.72 mftwist(F, D). F being a generalized modular form, returns the twist of F by the integer
D, i.e., the form G such that mfcoef(G,n)=(D/n)mfcoef(F,n), where (D/n) is the Kronecker
symbol.
? mf = mfinit([11,2],0); F = mfbasis(mf)[1]; mfcoefs(F, 5)
%1 = [0, 1, -2, -1, 2, 1]
? G = mftwist(F,-3); mfcoefs(G, 5)
%2 = [0, 1, 2, 0, 2, -1]
? mf2 = mfinit([99,2],0); mftobasis(mf2, G)
%3 = [1/3, 0, 1/3, 0]~
Note that twisting multiplies the level by D2 . In particular it is not an involution:
? H = mftwist(G,-3); mfcoefs(H, 5)
%4 = [0, 1, -2, 0, 2, 1]
? mfparams(G)
%5 = [99, 2, 1, y, t - 1]
The library syntax is GEN mftwist(GEN F, GEN D).

570
3.18 Modular symbols.
Let ∆0 := Div0 (P1 (Q)) be the abelian group of divisors of degree 0 on the rational projective
line. The standard GL(2, Q) action on P1 (Q) via homographies naturally extends to ∆0 . Given
• G a finite index subgroup of SL(2, Z),
• a field F and a finite dimensional representation V /F of GL(2, Q),
we consider the space of modular symbols M := HomG (∆0 , V ). This finite dimensional F -vector
space is a G-module, canonically isomorphic to Hc1 (X(G), V ), and allows to compute modular forms
for G.
Currently, we only support the groups Γ0 (N ) (N > 0 an integer) and the representations
Vk = Q[X, Y ]k−2 (k ≥ 2 an integer) over Q. We represent a space of modular symbols by an ms
structure, created by the function msinit. It encodes basic data attached to the space: chosen
Z[G]-generators (gi ) for ∆0 (and relations among those) and an F -basis of M . A modular symbol
s is thus given either in terms of this fixed basis, or as a collection of values s(gi ) satisfying certain
relations.
A subspace of M (e.g. the cuspidal or Eisenstein subspaces, the new or old modular symbols,
etc.) is given by a structure allowing quick projection and restriction of linear operators; its first
component is a matrix whose columns form an F -basis of the subspace.

3.18.1 msatkinlehner(M, Q, {H}). Let M be a full modular symbol space of level N , as given
by msinit, let Q | N , (Q, N/Q) = 1, and let H be a subspace stable under the Atkin-Lehner
involution wQ . Return the matrix of wQ acting on H (M if omitted).
? M = msinit(36,2); \\ M_2(Gamma_0(36))
? w = msatkinlehner(M,4); w^2 == 1
%2 = 1
? #w \\ involution acts on a 13-dimensional space
%3 = 13
? M = msinit(36,2, -1); \\ M_2(Gamma_0(36))^-
? w = msatkinlehner(M,4); w^2 == 1
%5 = 1
? #w
%6 = 4
The library syntax is GEN msatkinlehner(GEN M, long Q, GEN H = NULL).

3.18.2 mscosets(gen, inH ). gen being a system of generators for a group G and H being a
subgroup of finite index in G, return a list of right cosets of H\G and the right action of G
on H\G. The subgroup H is given by a criterion inH (closure) deciding whether an element of
G belongs to H. The group G is restricted to types handled by generic multiplication (*) and
inversion (g^(-1)), such as matrix groups or permutation groups.
Let gens = [g1 , . . . , gr ]. The function returns [C, M ] where C lists the h = [G : H] representa-
tives [γ1 , . . . , γh ] for the right cosets Hγ1 , . . . , Hγh ; γ1 is always the neutral element in G. For all
i ≤ h, j ≤ r, if M [i][j] = k then Hγi gj = Hγk .
? PSL2 = [[0,1;-1,0], [1,1;0,1]]; \\ S and T
\\ G = PSL2, H = Gamma0(2)
? [C, M] = mscosets(PSL2, g->g[2,1] % 2 == 0);

571
 ? C \\ three cosets
%3 = [[1, 0; 0, 1], [0, 1; -1, 0], [0, 1; -1, -1]]
? M
%4 = [Vecsmall([2, 1]), Vecsmall([1, 3]), Vecsmall([3, 2])]
Looking at M [1] we see that S belongs to the second coset and T to the first (trivial) coset.
The library syntax is GEN mscosets0(GEN gen, GEN inH). Also available is the function GEN
mscosets(GEN G, void *E, long (*inH)(void *, GEN))

3.18.3 mscuspidal(M, {flag = 0}). M being a full modular symbol space, as given by msinit,
return its cuspidal part S. If flag = 1, return [S, E] its decomposition into cuspidal and Eisenstein
parts.
A subspace is given by a structure allowing quick projection and restriction of linear operators;
its first component is a matrix with integer coefficients whose columns form a Q-basis of the
subspace.
? M = msinit(2,8, 1); \\ M_8(Gamma_0(2))^+
? [S,E] = mscuspidal(M, 1);
? E[1] \\ 2-dimensional
%3 =
[0 -10]
[0 -15]
[0 -3]
[1 0]
? S[1] \\ 1-dimensional
%4 =
[ 3]
[30]
[ 6]
[-8]
The library syntax is GEN mscuspidal(GEN M, long flag).

3.18.4 msdim(M ). M being a full modular symbol space or subspace, for instance as given by
msinit or mscuspidal, return its dimension as a Q-vector space.
? M = msinit(11,4); msdim(M)
%1 = 6
? M = msinit(11,4,1); msdim(M)
%2 = 4 \\ dimension of the ’+’ part
? [S,E] = mscuspidal(M,1);
? [msdim(S), msdim(E)]
%4 = [2, 2]
Note that mfdim([N,k]) is going to be much faster if you only need the dimension of the space
and not really to work with it. This function is only useful to quickly check the dimension of an
existing space.
The library syntax is long msdim(GEN M).

572
3.18.5 mseisenstein(M ). M being a full modular symbol space, as given by msinit, return its
Eisenstein subspace. A subspace is given by a structure allowing quick projection and restriction of
linear operators; its first component is a matrix with integer coefficients whose columns form a Q-
basis of the subspace. This is the same basis as given by the second component of mscuspidal(M, 1).
? M = msinit(2,8, 1); \\ M_8(Gamma_0(2))^+
? E = mseisenstein(M);
? E[1] \\ 2-dimensional
%3 =
[0 -10]
[0 -15]
[0 -3]
[1 0]
? E == mscuspidal(M,1)[2]
%4 = 1
The library syntax is GEN mseisenstein(GEN M).

3.18.6 mseval(M, s, {p}). Let ∆0 := Div0 (P1 (Q)). Let M be a full modular symbol space, as
given by msinit, let s be a modular symbol from M , i.e. an element of HomG (∆0 , V ), and let
p = [a, b] ∈ ∆0 be a path between two elements in P1 (Q), return s(p) ∈ V . The path extremities
a and b may be given as t_INT, t_FRAC or oo = (1 : 0); it is also possible to describe the path by
a 2 × 2 integral matrix whose columns give the two cusps. The symbol s is either
• a t_COL coding a modular symbol in terms of the fixed basis of HomG (∆0 , V ) chosen in M ;
if M was initialized with a nonzero sign (+ or −), then either the basis for the full symbol space
or the ±-part can be used (the dimension being used to distinguish the two).
• a t_MAT whose columns encode modular symbols as above. This is much faster than evalu-
ating individual symbols on the same path p independently.
• a t_VEC (vi ) of elements of V , where the vi = s(gi ) give the image of the generators gi of
∆0 , see mspathgens. We assume that s is a proper symbol, i.e. that the vi satisfy the mspathgens
relations.
If p is omitted, convert a single symbol s to the second form: a vector of the s(gi ). A t_MAT
is converted to a vector of such.
? M = msinit(2,8,1); \\ M_8(Gamma_0(2))^+
? g = mspathgens(M)[1]
%2 = [[+oo, 0], [0, 1]]
? N = msnew(M)[1]; #N \\ Q-basis of new subspace, dimension 1
%3 = 1
? s = N[,1] \\ t_COL representation
%4 = [-3, 6, -8]~
? S = mseval(M, s) \\ t_VEC representation
%5 = [64*x^6-272*x^4+136*x^2-8, 384*x^5+960*x^4+192*x^3-672*x^2-432*x-72]
? mseval(M,s, g[1])
%6 = 64*x^6 - 272*x^4 + 136*x^2 - 8
? mseval(M,S, g[1])
%7 = 64*x^6 - 272*x^4 + 136*x^2 - 8

573
Note that the symbol should have values in V = Q[x, y]k−2 , we return the de-homogenized values
corresponding to y = 1 instead.
The library syntax is GEN mseval(GEN M, GEN s, GEN p = NULL).

3.18.7 msfarey(F, inH , {&CM }). F being a Farey symbol attached to a group G contained in
PSL2 (Z) and H a subgroup of G, return a Farey symbol attached to H. The subgroup H is given
by a criterion inH (closure) deciding whether an element of G belongs to H. The symbol F can be
created using
• mspolygon: G = Γ0 (N ), which runs in time Õ(N );
• or msfarey itself, which runs in time O([G : H]2 ).
If present, the argument CM is set to mscosets(F[3]), giving the right cosets of H\G and the
action of G by right multiplication. Since msfarey’s algorithm is quadratic in the index [G : H], it
is advisable to construct subgroups by a chain of inclusions if possible.
\\ Gamma_0(N)
G0(N) = mspolygon(N);
\\ Gamma_1(N): direct construction, slow
G1(N) = msfarey(mspolygon(1), g -> my(a = g[1,1]%N, c = g[2,1]%N);\
c == 0 && (a == 1 || a == N-1));
\\ Gamma_1(N) via Gamma_0(N): much faster
G1(N) = msfarey(G0(N), g -> my(a=g[1,1]%N); a==1 || a==N-1);
\\ Gamma(N)
G(N) = msfarey(G1(N), g -> g[1,2]%N==0);
G_00(N) = msfarey(G0(N), x -> x[1,2]%N==0);
G1_0(N1,N2) = msfarey(G0(1), x -> x[2,1]%N1==0 && x[1,2]%N2==0);
\\ Gamma_0(91) has 4 elliptic points of order 3, Gamma_1(91) has none
D0 = mspolygon(G0(91), 2)[4];
D1 = mspolygon(G1(91), 2)[4];
write("F.tex","\\documentclass{article}\\usepackage{tikz}\\begin{document}",\
D0,"\n",D1,"\\end{document}");
The library syntax is GEN msfarey0(GEN F, GEN inH, GEN *CM = NULL). Also available is
GEN msfarey(GEN F, void *E, long (*inH)(void *, GEN), GEN *pCM).

3.18.8 msfromcusp(M, c). Returns the modular symbol attached to the cusp c, where M is a
modular symbol space of level N , attached to G = Γ0 (N ). The cusp c in P1 (Q)/G is given either
as oo (= (1 : 0)) or as a rational number a/b (= (a : b)). The attached symbol maps the path
[b] − [a] ∈ Div0 (P1 (Q)) to Ec (b) − Ec (a), where Ec (r) is 0 when r 6= c and X k−2 | γr otherwise,
where γr · r = (1 : 0). These symbols span the Eisenstein subspace of M .
? M = msinit(2,8); \\ M_8(Gamma_0(2))
? E = mseisenstein(M);
? E[1] \\ two-dimensional
%3 =
[0 -10]
[0 -15]
[0 -3]

574
 [1 0]
? s = msfromcusp(M,oo)
%4 = [0, 0, 0, 1]~
? mseval(M, s)
%5 = [1, 0]
? s = msfromcusp(M,1)
%6 = [-5/16, -15/32, -3/32, 0]~
? mseval(M,s)
%7 = [-x^6, -6*x^5 - 15*x^4 - 20*x^3 - 15*x^2 - 6*x - 1]
In case M was initialized with a nonzero sign, the symbol is given in terms of the fixed basis
of the whole symbol space, not the + or − part (to which it need not belong).
? M = msinit(2,8, 1); \\ M_8(Gamma_0(2))^+
? E = mseisenstein(M);
? E[1] \\ still two-dimensional, in a smaller space
%3 =
[ 0 -10]
[ 0 3]
[-1 0]
? s = msfromcusp(M,oo) \\ in terms of the basis for M_8(Gamma_0(2)) !
%4 = [0, 0, 0, 1]~
? mseval(M, s) \\ same symbol as before
%5 = [1, 0]
The library syntax is GEN msfromcusp(GEN M, GEN c).

3.18.9 msfromell(E, {sign = 0}). Let E/Q be an elliptic curve of conductor N . For ε = ±1, we
define the (cuspidal, new) modular symbol xε in Hc1 (X0 (N ), Q)ε attached to E. For all primes p
not dividing N we have Tp (xε ) = ap xε , where ap = p + 1 − #E(Fp ).
Let Ω+ = E.omega[1] be the real period of E (integration of the Néron differential dx/(2y +
a1 x + a3) on the connected component of E(R), i.e. the generator of H1 (E, Z)+ ) normalized by
Ω+ > 0. Let iΩ− the integral on a generator of H1 (E, Z)− with Ω− ∈ R>0 . If c∞ is the number of
connected components of E(R), Ω− is equal to (−2/c∞ ) × imag(E.omega[2]). The complex modular
symbol is defined by Z
F : δ → 2iπ f (z)dz
δ

The modular symbols x are normalized so that F = x+ Ω+ + x− iΩ− . In particular, we have

ε

x+ ([0] − [∞]) = L(E, 1)/Ω+ ,

which defines x± unless L(E, 1) = 0. Furthermore, for all fundamental discriminants D such that
ε · D > 0, we also have
X
(D|a)xε ([a/|D|] − [∞]) = L(E, (D|.), 1)/Ωε ,
0≤a<|D|

where (D|.) is the Kronecker symbol. The period Ω− is also 2/c∞ × the real period of the twist
E (−4) = elltwist(E, −4).

575
 This function returns the pair [M, x], where M is msinit(N, 2) and x is xsign as above when
sign = ±1, and x = [x+ , x− , LE ] when sign is 0, where LE is a matrix giving the canonical Z-lattice
attached to E in the sense of mslattice applied to Qx+ + Qx− . Explicitly, it is generated by
(x+ , x− ) when E(R) has two connected components and by (x+ − x− , 2x− ) otherwise.
The modular symbols x± are given as a t_COL (in terms of the fixed basis of HomG (∆0 , Q)
chosen in M ).
? E=ellinit([0,-1,1,-10,-20]); \\ X_0(11)
? [M,xp]= msfromell(E,1);
? xp
%3 = [1/5, -1/2, -1/2]~
? [M,x]= msfromell(E);
? x \\ x^+, x^- and L_E
%5 = [[1/5, -1/2, -1/2]~, [0, 1/2, -1/2]~, [1/5, 0; -1, 1; 0, -1]]
? p = 23; (mshecke(M,p) - ellap(E,p))*x[1]
%6 = [0, 0, 0]~ \\ true at all primes, including p = 11; same for x[2]
? (mshecke(M,p) - ellap(E,p))*x[3] == 0
%7 = 1
Instead of a single curve E, one may use instead a vector of isogenous curves. The function then
returns M and the vector of attached modular symbols.
The library syntax is GEN msfromell(GEN E, long sign).

3.18.10 msfromhecke(M, v, {H}). Given a msinit M and a vector v of pairs [p, P ] (where p is
prime and P is a polynomial with integer coefficients), return a basis of all modular symbols such
that P (Tp )(s) = 0. If H is present, it must be a Hecke-stable subspace and we restrict to s ∈ H.
When Tp has a rational eigenvalue and P (x) = x − ap has degree 1, we also accept the integer ap
instead of P .
? E = ellinit([0,-1,1,-10,-20]) \\11a1
? ellap(E,2)
%2 = -2
? ellap(E,3)
%3 = -1
? M = msinit(11,2);
? S = msfromhecke(M, [[2,-2],[3,-1]])
%5 =
[ 1 1]
[-5 0]
[ 0 -5]
? mshecke(M, 2, S)
%6 =
[-2 0]
[ 0 -2]
? M = msinit(23,4);
? S = msfromhecke(M, [[5, x^4-14*x^3-244*x^2+4832*x-19904]]);
? factor( charpoly(mshecke(M,5,S)) )
%9 =

576
 [x^4 - 14*x^3 - 244*x^2 + 4832*x - 19904 2]
The library syntax is GEN msfromhecke(GEN M, GEN v, GEN H = NULL).

3.18.11 msgetlevel(M ). M being a full modular symbol space, as given by msinit, return its
level N .
The library syntax is long msgetlevel(GEN M).

3.18.12 msgetsign(M ). M being a full modular symbol space, as given by msinit, return its
sign: ±1 or 0 (unset).
? M = msinit(11,4, 1);
? msgetsign(M)
%2 = 1
? M = msinit(11,4);
? msgetsign(M)
%4 = 0
The library syntax is long msgetsign(GEN M).

3.18.13 msgetweight(M ). M being a full modular symbol space, as given by msinit, return its
weight k.
? M = msinit(11,4);
? msgetweight(M)
%2 = 4
The library syntax is long msgetweight(GEN M).

3.18.14 mshecke(M, p, {H}). M being a full modular symbol space, as given by msinit, p being
a prime number, and H being a Hecke-stable subspace (M if omitted) return the matrix of Tp
acting on H (Up if p divides N ). Result is undefined if H is not stable by Tp (resp. Up ).
? M = msinit(11,2); \\ M_2(Gamma_0(11))
? T2 = mshecke(M,2)
%2 =
[3 0 0]
[1 -2 0]
[1 0 -2]
? M = msinit(11,2, 1); \\ M_2(Gamma_0(11))^+
? T2 = mshecke(M,2)
%4 =
[ 3 0]
[-1 -2]
? N = msnew(M)[1] \\ Q-basis of new cuspidal subspace
%5 =
[-2]
[-5]
? p = 1009; mshecke(M, p, N) \\ action of T_1009 on N
%6 =

577
 [-10]
? ellap(ellinit("11a1"), p)
%7 = -10

The library syntax is GEN mshecke(GEN M, long p, GEN H = NULL).

3.18.15 msinit(G, V, {sign = 0}). Given G a finite index subgroup of SL(2, Z) and a finite
dimensional representation V of GL(2, Q), creates a space of modular symbols, the G-module
HomG (Div0 (P1 (Q)), V ). This is canonically isomorphic to Hc1 (X(G), V ), and allows to compute
modular forms for G. If sign is present and nonzero, it must be ±1 and we consider the subspace
defined by Ker(σ − sign), where σ is induced by [-1,0;0,1]. Currently the only supported groups
are the Γ0 (N ), coded by the integer N > 0. The only supported representation is Vk = Q[X, Y ]k−2 ,
coded by the integer k ≥ 2.

? M = msinit(11,2); msdim(M) \\ Gamma0(11), weight 2

%1 = 3
? mshecke(M,2) \\ T_2 acting on M
%2 =
[3 1 1]
[0 -2 0]
[0 0 -2]
? msstar(M) \\ * involution
%3 =
[1 0 0]
[0 0 1]
[0 1 0]
? Mp = msinit(11,2, 1); msdim(Mp) \\ + part
%4 = 2
? mshecke(Mp,2) \\ T_2 action on M^+
%5 =
[3 2]
[0 -2]
? msstar(Mp)
%6 =
[1 0]
[0 1]

The library syntax is GEN msinit(GEN G, GEN V, long sign).

578
3.18.16 msissymbol(M, s). M being a full modular symbol space, as given by msinit, check
whether s is a modular symbol attached to M . If A is a matrix, check whether its columns
represent modular symbols and return a 0 − 1 vector.
? M = msinit(7,8, 1); \\ M_8(Gamma_0(7))^+
? A = msnew(M)[1];
? s = A[,1];
? msissymbol(M, s)
%4 = 1
? msissymbol(M, A)
%5 = [1, 1, 1]
? S = mseval(M,s);
? msissymbol(M, S)
%7 = 1
? [g,R] = mspathgens(M); g
%8 = [[+oo, 0], [0, 1/2], [1/2, 1]]
? #R \\ 3 relations among the generators g_i
%9 = 3
? T = S; T[3]++; \\ randomly perturb S(g_3)
? msissymbol(M, T)
%11 = 0 \\ no longer satisfies the relations
The library syntax is GEN msissymbol(GEN M, GEN s).

3.18.17 mslattice(M, {H}). Let ∆0 := Div0 (P1 (Q)) and Vk = Q[x, y]k−2 . Let M be a full
modular symbol space, as given by msinit and let H be a subspace, e.g. as given by mscuspidal.
This function returns a canonical Z structure on H defined as follows. Consider the map c : M =
HomΓ0 (N ) (∆0 , Vk ) → H 1 (Γ0 (N ), Vk ) given by φ 7→ class(γ → φ({0, γ −1 0})). Let Lk = Z[x, y]k−2
be the natural Z-structure of Vk . The result of mslattice is a Z-basis of the inverse image by c of
H 1 (Γ0 (N ), Lk ) in the space of modular symbols generated by H.
For user convenience, H can be defined by a matrix representing the Q-basis of H (in terms
of the canonical Q-basis of M fixed by msinit and used to represent modular symbols).
If omitted, H is the cuspidal part of M as given by mscuspidal. The Eisenstein part
HomΓ0 (N ) (Div(P1 (Q)), Vk ) is in the kernel of c, so the result has no meaning for the Eisenstein
part H.
? M=msinit(11,2);
? [S,E] = mscuspidal(M,1); S[1] \\ a primitive Q-basis of S
%2 =
[ 1 1]
[-5 0]
[ 0 -5]
? mslattice(M,S)
%3 =
[-1/5 -1/5]
[ 1 0]
[ 0 1]
? mslattice(M,E)
%4 =

579
 [1]
[0]
[0]
? M=msinit(5,4);
? S=mscuspidal(M); S[1]
%6 =
[ 7 20]
[ 3 3]
[-10 -23]
[-30 -30]
? mslattice(M,S)
%7 =
[-1/10 -11/130]
[ 0 -1/130]
[ 1/10 6/65]
[ 0 1/13]

The library syntax is GEN mslattice(GEN M, GEN H = NULL).

3.18.18 msnew(M ). M being a full modular symbol space, as given by msinit, return the
new part of its cuspidal subspace. A subspace is given by a structure allowing quick projection
and restriction of linear operators; its first component is a matrix with integer coefficients whose
columns form a Q-basis of the subspace.

? M = msinit(11,8, 1); \\ M_8(Gamma_0(11))^+

? N = msnew(M);
? #N[1] \\ 6-dimensional
%3 = 6

The library syntax is GEN msnew(GEN M).

3.18.19 msomseval(Mp, PHI , path). Return the vectors of moments of the p-adic distribution
attached to the path path by the overconvergent modular symbol PHI.

? M = msinit(3,6,1);
? Mp= mspadicinit(M,5,10);
? phi = [5,-3,-1]~;
? msissymbol(M,phi)
%4 = 1
? PHI = mstooms(Mp,phi);
? ME = msomseval(Mp,PHI,[oo, 0]);

The library syntax is GEN msomseval(GEN Mp, GEN PHI, GEN path).

580
3.18.20 mspadicL(mu, {s = 0}, {r = 0}). Returns the value (or r-th derivative) on a character
χs of Z∗p of the p-adic L-function attached to mu.
Let Φ be the p-adic distribution-valued overconvergent symbol attached to a modular symbol
φ for Γ0 (N ) (eigenvector for TN (p) for the eigenvalue ap ). Then Lp (Φ, χs ) = Lp (µ, s) is the p-adic
L function defined by Z
Lp (Φ, χs ) = χs (z)dµ(z)
Z∗
p

where µ is the distribution on Z∗p defined by the restriction of Φ([∞]−[0]) to Z∗p . The r-th derivative
is taken in direction hχi: Z
L(r) s
p (Φ, χ ) = χs (z)(log z)r dµ(z).
Z∗
p

In the argument list,

• mu is as returned by mspadicmoments (distributions attached to Φ by restriction to discs
a + pν Zp , (a, p) = 1).
• s = [s1 , s2 ] with s1 ∈ Z ⊂ Zp and s2 mod p − 1 or s2 mod 2 for p = 2, encoding the p-adic
character χs := hχis1 τ s2 ; here χ is the cyclotomic character from Gal(Qp (µp∞ )/Qp ) to Z∗p , and
τ is the Teichmüller character (for p > 2 and the character of order 2 on (Z/4Z)∗ if p = 2); for
convenience, the character [s, s] can also be represented by the integer s.
When ap is a p-adic unit, Lp takes its values in Qp . When ap is not a unit, it takes its values
in the two-dimensional Qp -vector space Dcris (M (φ)) where M (φ) is the “motive” attached to φ,
and we return the two p-adic components with respect to some fixed Qp -basis.
? M = msinit(3,6,1); phi=[5, -3, -1]~;
? msissymbol(M,phi)
%2 = 1
? Mp = mspadicinit(M, 5, 4);
? mu = mspadicmoments(Mp, phi); \\ no twist
\\ End of initializations
? mspadicL(mu,0) \\ L_p(chi^0)
%5 = 5 + 2*5^2 + 2*5^3 + 2*5^4 + ...
? mspadicL(mu,1) \\ L_p(chi), zero for parity reasons
%6 = [O(5^13)]~
? mspadicL(mu,2) \\ L_p(chi^2)
%7 = 3 + 4*5 + 4*5^2 + 3*5^5 + ...
? mspadicL(mu,[0,2]) \\ L_p(tau^2)
%8 = 3 + 5 + 2*5^2 + 2*5^3 + ...
? mspadicL(mu, [1,0]) \\ L_p(<chi>)
%9 = 3*5 + 2*5^2 + 5^3 + 2*5^7 + 5^8 + 5^10 + 2*5^11 + O(5^13)
? mspadicL(mu,0,1) \\ L_p’(chi^0)
%10 = 2*5 + 4*5^2 + 3*5^3 + ...
? mspadicL(mu, 2, 1) \\ L_p’(chi^2)
%11 = 4*5 + 3*5^2 + 5^3 + 5^4 + ...
Now several quadratic twists: mstooms is indicated.
? PHI = mstooms(Mp,phi);
? mu = mspadicmoments(Mp, PHI, 12); \\ twist by 12

581
 ? mspadicL(mu)
%14 = 5 + 5^2 + 5^3 + 2*5^4 + ...
? mu = mspadicmoments(Mp, PHI, 8); \\ twist by 8
? mspadicL(mu)
%16 = 2 + 3*5 + 3*5^2 + 2*5^4 + ...
? mu = mspadicmoments(Mp, PHI, -3); \\ twist by -3 < 0
? mspadicL(mu)
%18 = O(5^13) \\ always 0, phi is in the + part and D < 0
One can locate interesting symbols of level N and weight k with msnew and mssplit. Note
that instead of a symbol, one can input a 1-dimensional Hecke-subspace from mssplit: the function
will automatically use the underlying basis vector.
? M=msinit(5,4,1); \\ M_4(Gamma_0(5))^+
? L = mssplit(M, msnew(M)); \\ list of irreducible Hecke-subspaces
? phi = L[1]; \\ one Galois orbit of newforms
? #phi[1] \\... this one is rational
%4 = 1
? Mp = mspadicinit(M, 3, 4);
? mu = mspadicmoments(Mp, phi);
? mspadicL(mu)
%7 = 1 + 3 + 3^3 + 3^4 + 2*3^5 + 3^6 + O(3^9)
? M = msinit(11,8, 1); \\ M_8(Gamma_0(11))^+
? Mp = mspadicinit(M, 3, 4);
? L = mssplit(M, msnew(M));
? phi = L[1]; #phi[1] \\ ... this one is two-dimensional
%11 = 2
? mu = mspadicmoments(Mp, phi);
*** at top-level: mu=mspadicmoments(Mp,ph
*** ^--------------------
*** mspadicmoments: incorrect type in mstooms [dim_Q (eigenspace) > 1]
The library syntax is GEN mspadicL(GEN mu, GEN s = NULL, long r).

3.18.21 mspadicinit(M, p, n, {flag}). M being a full modular symbol space, as given by msinit,
and p a prime, initialize technical data needed to compute with overconvergent modular symbols,
modulo pn . If flag is unset, allow all symbols; else initialize only for a restricted range of symbols
depending on flag: if flag = 0 restrict to ordinary symbols, else restrict to symbols φ such that
Tp (φ) = ap φ, with vp (ap ) ≥ flag, which is faster as flag increases. (The fastest initialization is
obtained for flag = 0 where we only allow ordinary symbols.) For supersingular eigensymbols, such
that p | ap , we must further assume that p does not divide the level.
? E = ellinit("11a1");
? [M,phi] = msfromell(E,1);
? ellap(E,3)
%3 = -1
? Mp = mspadicinit(M, 3, 10, 0); \\ commit to ordinary symbols
? PHI = mstooms(Mp,phi);
If we restrict the range of allowed symbols with flag(for faster initialization), exceptions will
occur if vp (ap ) violates this bound:

582
 ? E = ellinit("15a1");
? [M,phi] = msfromell(E,1);
? ellap(E,7)
%3 = 0
? Mp = mspadicinit(M,7,5,0); \\ restrict to ordinary symbols
? PHI = mstooms(Mp,phi)
*** at top-level: PHI=mstooms(Mp,phi)
*** ^---------------
*** mstooms: incorrect type in mstooms [v_p(ap) > mspadicinit flag] (t_VEC).
? Mp = mspadicinit(M,7,5); \\ no restriction
? PHI = mstooms(Mp,phi);

This function uses O(N 2 (n + k)2 p) memory, where N is the level of M .

The library syntax is GEN mspadicinit(GEN M, long p, long n, long flag).

3.18.22 mspadicmoments(Mp, PHI , {D = 1}). Given Mp from mspadicinit, an overconvergent

eigensymbol PHI from mstooms and a fundamental discriminant D coprime to p, let PHID denote
the twisted symbol. This function computes the distribution µ = PHID ([0] − ∞]) | Z∗p restricted to
Z∗p . More precisely, it returns the moments of the p − 1 distributions PHID ([0] − [∞]) | (a + pZp ),
0 < a < p. We also allow PHI to be given as a classical symbol, which is then lifted to an
overconvergent symbol by mstooms; but this is wasteful if more than one twist is later needed.

The returned data µ (p-adic distributions attached to PHI) can then be used in mspadicL or
mspadicseries. This precomputation allows to quickly compute derivatives of different orders or
values at different characters.

? M = msinit(3,6, 1);
? phi = [5,-3,-1]~;
? msissymbol(M, phi)
%3 = 1
? p = 5; mshecke(M,p) * phi \\ eigenvector of T_5, a_5 = 6
%4 = [30, -18, -6]~
? Mp = mspadicinit(M, p, 10, 0); \\ restrict to ordinary symbols, mod p^10
? PHI = mstooms(Mp, phi);
? mu = mspadicmoments(Mp, PHI);
? mspadicL(mu)
%8 = 5 + 2*5^2 + 2*5^3 + ...
? mu = mspadicmoments(Mp, PHI, 12); \\ twist by 12
? mspadicL(mu)
%10 = 5 + 5^2 + 5^3 + 2*5^4 + ...

The library syntax is GEN mspadicmoments(GEN Mp, GEN PHI, long D).

583
3.18.23 mspadicseries(mu, {i = 0}). Let Φ be the p-adic distribution-valued overconvergent
symbol attached to a modular symbol φ for Γ0 (N ) (eigenvector for TN (p) for the eigenvalue ap ).
If µ is the distribution on Z∗p defined by the restriction of Φ([∞] − [0]) to Z∗p , let
Z
i
L̂p (µ, τ )(x) = τ i (t)(1 + x)logp (t)/ logp (u) dµ(t)
Z∗
p

Here, τ is the Teichmüller character and u is a specific multiplicative generator of 1+2pZp . (Namely
1 + p if p > 2 or 5 if p = 2.) To explain the formula, let G∞ := Gal(Q(µp∞ )/Q), let χ : G∞ → Z∗p
be the cyclotomic character (isomorphism) and γ the element of G∞ such that χ(γ) = u; then
χ(γ)logp (t)/ logp (u) = hti.
The p-padic precision of individual terms is maximal given the precision of the overconvergent
symbol µ.
? [M,phi] = msfromell(ellinit("17a1"),1);
? Mp = mspadicinit(M, 5,7);
? mu = mspadicmoments(Mp, phi,1); \\ overconvergent symbol
? mspadicseries(mu)
%4 = (4 + 3*5 + 4*5^2 + 2*5^3 + 2*5^4 + 5^5 + 4*5^6 + 3*5^7 + O(5^9)) \
+ (3 + 3*5 + 5^2 + 5^3 + 2*5^4 + 5^6 + O(5^7))*x \
+ (2 + 3*5 + 5^2 + 4*5^3 + 2*5^4 + O(5^5))*x^2 \
+ (3 + 4*5 + 4*5^2 + O(5^3))*x^3 \
+ (3 + O(5))*x^4 + O(x^5)
An example with nonzero Teichmüller:
? [M,phi] = msfromell(ellinit("11a1"),1);
? Mp = mspadicinit(M, 3,10);
? mu = mspadicmoments(Mp, phi,1);
? mspadicseries(mu, 2)
%4 = (2 + 3 + 3^2 + 2*3^3 + 2*3^5 + 3^6 + 3^7 + 3^10 + 3^11 + O(3^12)) \
+ (1 + 3 + 2*3^2 + 3^3 + 3^5 + 2*3^6 + 2*3^8 + O(3^9))*x \
+ (1 + 2*3 + 3^4 + 2*3^5 + O(3^6))*x^2 \
+ (3 + O(3^2))*x^3 + O(x^4)
Supersingular example (not checked)
? E = ellinit("17a1"); ellap(E,3)
%1 = 0
? [M,phi] = msfromell(E,1);
? Mp = mspadicinit(M, 3,7);
? mu = mspadicmoments(Mp, phi,1);
? mspadicseries(mu)
%5 = [(2*3^-1 + 1 + 3 + 3^2 + 3^3 + 3^4 + 3^5 + 3^6 + O(3^7)) \
+ (2 + 3^3 + O(3^5))*x \
+ (1 + 2*3 + O(3^2))*x^2 + O(x^3),\
(3^-1 + 1 + 3 + 3^2 + 3^3 + 3^4 + 3^5 + 3^6 + O(3^7)) \
+ (1 + 2*3 + 2*3^2 + 3^3 + 2*3^4 + O(3^5))*x \
+ (3^-2 + 3^-1 + O(3^2))*x^2 + O(3^-2)*x^3 + O(x^4)]
Example with a twist:

584
 ? E = ellinit("11a1");
? [M,phi] = msfromell(E,1);
? Mp = mspadicinit(M, 3,10);
? mu = mspadicmoments(Mp, phi,5); \\ twist by 5
? L = mspadicseries(mu)
%5 = (2*3^2 + 2*3^4 + 3^5 + 3^6 + 2*3^7 + 2*3^10 + O(3^12)) \
+ (2*3^2 + 2*3^6 + 3^7 + 3^8 + O(3^9))*x \
+ (3^3 + O(3^6))*x^2 + O(3^2)*x^3 + O(x^4)
? mspadicL(mu)
%6 = [2*3^2 + 2*3^4 + 3^5 + 3^6 + 2*3^7 + 2*3^10 + O(3^12)]~
? ellpadicL(E,3,10,,5)
%7 = 2 + 2*3^2 + 3^3 + 2*3^4 + 2*3^5 + 3^6 + 2*3^7 + O(3^10)
? mspadicseries(mu,1) \\ must be 0
%8 = O(3^12) + O(3^9)*x + O(3^6)*x^2 + O(3^2)*x^3 + O(x^4)
The library syntax is GEN mspadicseries(GEN mu, long i).

3.18.24 mspathgens(M ). Let ∆0 := Div0 (P1 (Q)). Let M being a full modular symbol space, as
given by msinit, return a set of Z[G]-generators for ∆0 . The output is [g, R], where g is a minimal
system of generators and R the vector of Z[G]-relations between the given generators.PA relation is
coded by a vector of pairs [ai , i] with ai ∈ Z[G] and i the index of a generator, so that i ai g[i] = 0.
An element [v] − [u] in ∆0 is coded by the “path” [u, v], where oo denotes the point at infinity
(1 : 0) on the projective line. An element of Z[G] is either an integer n (= n[id2 ]) or a “factorization
matrix”: the P first column contains distinct elements gi of G and the second integers ni and the
matrix codes ni [gi ]:
? M = msinit(11,8); \\ M_8(Gamma_0(11))
? [g,R] = mspathgens(M);
? g
%3 = [[+oo, 0], [0, 1/3], [1/3, 1/2]] \\ 3 paths
? #R \\ a single relation
%4 = 1
? r = R[1]; #r \\ ...involving all 3 generators
%5 = 3
? r[1]
%6 = [[1, 1; [1, 1; 0, 1], -1], 1]
? r[2]
%7 = [[1, 1; [7, -2; 11, -3], -1], 2]
? r[3]
%8 = [[1, 1; [8, -3; 11, -4], -1], 3]
P
The given relation is of the form i (1 − γi )gi = 0, with γi ∈ Γ0 (11). There will always be a single
relation involving all generators (corresponding to a round trip along all cusps), then relations
involving a single generator (corresponding to 2 and 3-torsion elements in the group:
? M = msinit(2,8); \\ M_8(Gamma_0(2))
? [g,R] = mspathgens(M);
? g
%3 = [[+oo, 0], [0, 1]]
Note that the output depends only on the group G, not on the representation V .

585
 The library syntax is GEN mspathgens(GEN M).

3.18.25 mspathlog(M, p). Let ∆0 := Div0 (P1 (Q)). Let M being a full modular symbol space,
as given by msinit, encoding fixed Z[G]-generators (gi ) of ∆0 (see mspathgens). A path p = [a, b]
between two elements in P1 (Q) corresponds to [b] − [a] ∈ ∆0 . The path extremities a and b may
be given as t_INT, t_FRAC or oo = (1 : 0). Finally, we also allow to input a path as a 2 × 2 integer
matrix, whose first and second column give a and b respectively, with the convention [x, y]˜ = (x : y)
in P1 (Q).
P
Returns (pi ) in Z[G] such that p = i pi gi .
? M = msinit(2,8); \\ M_8(Gamma_0(2))
? [g,R] = mspathgens(M);
? g
%3 = [[+oo, 0], [0, 1]]
? p = mspathlog(M, [1/2,2/3]);
? p[1]
%5 =
[[1, 0; 2, 1] 1]
? p[2]
%6 =
[[1, 0; 0, 1] 1]
[[3, -1; 4, -1] 1]
? mspathlog(M, [1,2;2,3]) == p \\ give path via a 2x2 matrix
%7 = 1
Note that the output depends only on the group G, not on the representation V .
The library syntax is GEN mspathlog(GEN M, GEN p).

3.18.26 mspetersson(M, {F }, {G = F }). M being a full modular symbol space for Γ = Γ0 (N ),

as given by msinit, calculate the intersection product {F, G} of modular symbols F and G on
M = HomΓ (∆0 , Vk ) extended to an hermitian bilinear form on M ⊗C whose radical is the Eisenstein
subspace of M .
Suppose that f1 and f2 are two parabolic forms. Let F1 and F2 be the attached modular
symbols Z
Fi (δ) = fi (z) · (zX + Y )k−2 dz
δ

and let F1R , F2R be the attached real modular symbols

Z
FiR (δ) = < fi (z) · (zX + Y )k−2 dz


δ

Then we have
{F1R , F2R } = −2(2i)k−2 · =(< f1 , f2 >Petersson )
and
{F1 , F¯2 } = (2i)k−2 < f1 , f2 >Petersson
In weight 2, the intersection product {F, G} has integer values on the Z-structure on M given by
1
mslattice and defines a Riemann form on Hpar (Γ, R).

586
 For user convenience, we allow F and G to be matrices and return the attached Gram matrix.
If F is omitted: treat it as the full modular space attached to M ; if G is omitted, take it equal to
F.
? M = msinit(37,2);
? C = mscuspidal(M)[1];
? mspetersson(M, C)
%3 =
[ 0 -17 -8 -17]
[17 0 -8 -25]
[ 8 8 0 -17]
[17 25 17 0]
? mspetersson(M, mslattice(M,C))
%4 =
[0 -1 0 -1]
[1 0 0 -1]
[0 0 0 -1]
[1 1 1 0]
? E = ellinit("33a1");
? [M,xpm] = msfromell(E); [xp,xm,L] = xpm;
? mspetersson(M, mslattice(M,L))
%7 =
[0 -3]
[3 0]
? ellmoddegree(E)
%8 = [3, -126]
The coefficient 3 in the matrix is the degree of the modular parametrization.
The library syntax is GEN mspetersson(GEN M, GEN F = NULL, GEN G = NULL).

3.18.27 mspolygon(M, {flag = 0}). M describes a subgroup G of finite index in the modular
group PSL2 (Z), as given by msinit or a positive integer N (encoding the group G = Γ0 (N )), or
by msfarey (arbitrary subgroup). Return an hyperbolic polygon (Farey symbol) attached to G.
More precisely:
• Its vertices are an ordered list in P1 (Q) and contain a representatives of all cusps.
• Its edges are hyperbolic arcs joining two consecutive vertices; each edge e is labelled by an
integer µ(e) ∈ {∞, 2, 3}.
• Given a path (a, b) between two elements of P1 (Q), let (a, b) = (b, a) be the opposite path.
There is an involution e → e∗ on the edges. We have µ(e) = ∞ if and only if e 6= e∗ ; when µ(e) 6= 3,
e is G-equivalent to e∗ , i.e. there exists γe ∈ G such that e = γe e∗ ; if µ(e) = 3 there exists γe ∈ G
of order 3 such that the hyperbolic triangle (e, γe e, γe2 e) is invariant by γe . In all cases, to each
edge we have attached γe ∈ G of order µ(e).
The polygon is given by a triple [E, A, g]
• The list E of its consecutive edges as matrices in M2 (Z).
• The permutation A attached to the involution: if e = E[i] is the i-th edge, then A[i] is the
index of e∗ in E.

587
 • The list g of pairing matrices γe . Remark that γe∗ = γe−1 if µ(e) 6= 3, i.e., g[i]−1 = g[A[i]]
whenever i 6= A[i] (µ(g[i]) = 1) or µ(g[i]) = 2 (g[i]2 = 1). Modulo these trivial relations, the pairing
matrices form a system of independant generators of G. Note that γe is elliptic if and only if e∗ = e.
The above data yields a fundamental domain for G acting on Poincaré’s half-plane: take the convex
hull of the polygon defined by
• The edges in E such that e 6= e∗ or e∗ = e, where the pairing matrix γe has order 2;
• The edges (r, t) and (t, s) where the edge e = (r, s) ∈ E is such that e = e∗ and γe has order
3 and the triangle (r, t, s) is the image of (0, exp(2iπ/3), ∞) by some element of P SL2 (Q) formed
around the edge.
Binary digits of flag mean:
1: return a normalized hyperbolic polygon if set, else a polygon with unimodular edges (ma-
trices of determinant 1). A polygon is normalized in the sense of compact orientable surfaces if
the distance d(a, a∗ ) between an edge a and its image by the involution a∗ is less than 2, with
equality if and only if a is linked with another edge b (a, b, a∗ et b∗ appear consecutively in E up
to cyclic permutation). In particular, the vertices of all edges such that that d(a, a∗ ) 6= 1 (distance
is 0 or 2) are all equivalent to 0 modulo G. The external vertices of aa∗ such that d(a, a∗ ) = 1 are
also equivalent to 0; the internal vertices a ∩ a∗ (a single point), together with 0, form a system of
representatives of the cusps of G\P1 (Q). This is useful to compute the homology group H1 (G, Z)
as it gives a symplectic basis for the intersection pairing. In this case, the number of parabolic
matrices (trace 2) in the system of generators G is 2(t − 1), where t is the number of non equivalent
cusps for G. This is currently only implemented for G = Γ0 (N ).
2: add graphical representations (in LaTeX form) for the hyperbolic polygon in Poincaré’s
half-space and the involution a → a∗ of the Farey symbol. The corresponding character strings can
be included in a LaTeX document provided the preamble contains \usepackage{tikz}.
? [V,A,g] = mspolygon(3);
? V
%2 = [[-1, 1; -1, 0], [1, 0; 0, 1], [0, 1; -1, 1]]
? A
%3 = Vecsmall([2, 1, 3])
? g
%4 = [[-1, -1; 0, -1], [1, -1; 0, 1], [1, -1; 3, -2]]
? [V,A,g, D1,D2] = mspolygon(11,2); \\ D1 and D2 contains pictures
? {write("F.tex",
"\\documentclass{article}\\usepackage{tikz}\\begin{document}"
D1, "\n", D2,
"\\end{document}");}
? [V1,A1] = mspolygon(6,1); \\ normalized
? V1
%8 = [[-1, 1; -1, 0], [1, 0; 0, 1], [0, 1; -1, 3],
[1, -2; 3, -5], [-2, 1; -5, 2], [1, -1; 2, -1]]
? A1
%9 = Vecsmall([2, 1, 4, 3, 6, 5])
? [V0,A0] = mspolygon(6); \\ not normalized V[3]^* = V[6], d(V[3],V[6]) = 3
? A0
%11 = Vecsmall([2, 1, 6, 5, 4, 3])

588
 ? [V,A] = mspolygon(14, 1);
? A
%13 = Vecsmall([2, 1, 4, 3, 6, 5, 9, 10, 7, 8])
One can see from this last example that the (normalized) polygon has the form

(a1 , a∗1 , a2 , a∗2 , a3 , a∗3 , a4 , a5 , a∗4 , a∗5 ),

that X0 (14) is of genus 1 (in general the genus is the number of blocks of the form aba∗ b∗ ), has
no elliptic points (A has no fixed point) and 4 cusps (number of blocks of the form aa∗ plus 1).
The vertices of edges a4 and a5 all project to 0 in X0 (14): the paths a4 and a5 project as loops in
X0 (14) and give a symplectic basis of the homology H1 (X0 (14), Z).
? [V,A] = mspolygon(15);
? apply(matdet, V) \\ all unimodular
%2 = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
? [V,A] = mspolygon(15,1);
? apply(matdet, V) \\ normalized polygon but no longer unimodular edges
%4 = [1, 1, 1, 1, 2, 2, 47, 11, 47, 11]
The library syntax is GEN mspolygon(GEN M, long flag).

3.18.28 msqexpansion(M, projH , {B = seriesprecision}). M being a full modular symbol space,

as given by msinit, and projH being a projector on a Hecke-simple subspace (as given by mssplit),
return the Fourier coefficients an , n ≤ B of the corresponding normalized newform. If B is omitted,
use seriesprecision.
This function uses a naive O(B 2 d3 ) algorithm, where d = O(kN ) is the dimension of
Mk (Γ0 (N )).
? M = msinit(11,2, 1); \\ M_2(Gamma_0(11))^+
? L = mssplit(M, msnew(M));
? msqexpansion(M,L[1], 20)
%3 = [1, -2, -1, 2, 1, 2, -2, 0, -2, -2, 1, -2, 4, 4, -1, -4, -2, 4, 0, 2]
? ellan(ellinit("11a1"), 20)
%4 = [1, -2, -1, 2, 1, 2, -2, 0, -2, -2, 1, -2, 4, 4, -1, -4, -2, 4, 0, 2]
The shortcut msqexpansion(M, s, B) is available for a symbol s, provided it is a Hecke eigenvector:
? E = ellinit("11a1");
? [M,S] = msfromell(E); [sp,sm] = S;
? msqexpansion(M,sp,10) \\ in the + eigenspace
%3 = [1, -2, -1, 2, 1, 2, -2, 0, -2, -2]
? msqexpansion(M,sm,10) \\ in the - eigenspace
%4 = [1, -2, -1, 2, 1, 2, -2, 0, -2, -2]
? ellan(E, 10)
%5 = [1, -2, -1, 2, 1, 2, -2, 0, -2, -2]
The library syntax is GEN msqexpansion(GEN M, GEN projH, long precdl).

589
3.18.29 mssplit(M, {H}, {dimlim}). Let M denote a full modular symbol space, as given by
msinit(N, k, 1) or msinit(N, k, −1) and let H be a Hecke-stable subspace of msnew(M ) (the full
new subspace if H is omitted). This function splits H into Hecke-simple subspaces. If dimlim
is present and positive, restrict to subspaces of dimension ≤ dimlim. A subspace is given by a
structure allowing quick projection and restriction of linear operators; its first component is a
matrix with integer coefficients whose columns form a Q-basis of the subspace.
? M = msinit(11,8, 1); \\ M_8(Gamma_0(11))^+
? L = mssplit(M); \\ split msnew(M)
? #L
%3 = 2
? f = msqexpansion(M,L[1],5); f[1].mod
%4 = x^2 + 8*x - 44
? lift(f)
%5 = [1, x, -6*x - 27, -8*x - 84, 20*x - 155]
? g = msqexpansion(M,L[2],5); g[1].mod
%6 = x^4 - 558*x^2 + 140*x + 51744
To a Hecke-simple subspace corresponds an orbit of (normalized) newforms, defined over a number
field. In the above example, we printed the polynomials defining the said fields, as well as the first
5 Fourier coefficients (at the infinite cusp) of one such form.
The library syntax is GEN mssplit(GEN M, GEN H = NULL, long dimlim).

3.18.30 msstar(M, {H}). M being a full modular symbol space, as given by msinit, return the
matrix of the * involution, induced by complex conjugation, acting on the (stable) subspace H (M
if omitted).
? M = msinit(11,2); \\ M_2(Gamma_0(11))
? w = msstar(M);
? w^2 == 1
%3 = 1
The library syntax is GEN msstar(GEN M, GEN H = NULL).

3.18.31 mstooms(Mp, phi ). Given Mp from mspadicinit, lift the (classical) eigen symbol phi
to a p-adic distribution-valued overconvergent symbol in the sense of Pollack and Stevens. More
precisely, let φ belong to the space W of modular symbols of level N , vp (N ) ≤ 1, and weight k which
is an eigenvector for the Hecke operator TN (p) for a nonzero eigenvalue ap and let N0 = lcm(N, p).
Under the action of TN0 (p), φ generates a subspace Wφ of dimension 1 (if p | N ) or 2 (if p does
not divide N ) in the space of modular symbols of level N0 .
Let Vp = [p, 0; 0, 1] and Cp = [ap , pk−1 ; −1, 0]. When p does not divide N and ap is divisible
by p, mstooms returns the lift Φ of (φ, φ|k Vp ) such that

TN0 (p)Φ = Cp Φ

When p does not divide N and ap is not divisible by p, mstooms returns the lift Φ of φ−α−1 φ|k Vp
which is an eigenvector of TN0 (p) for the unit eigenvalue where α2 − ap α + pk−1 = 0.
The resulting overconvergent eigensymbol can then be used in mspadicmoments, then mspadicL
or mspadicseries.

590
 ? M = msinit(3,6, 1); p = 5;
? Tp = mshecke(M, p); factor(charpoly(Tp))
%2 =
[x - 3126 2]
[ x - 6 1]
? phi = matker(Tp - 6)[,1] \\ generator of p-Eigenspace, a_p = 6
%3 = [5, -3, -1]~
? Mp = mspadicinit(M, p, 10, 0); \\ restrict to ordinary symbols, mod p^10
? PHI = mstooms(Mp, phi);
? mu = mspadicmoments(Mp, PHI);
? mspadicL(mu)
%7 = 5 + 2*5^2 + 2*5^3 + ...

A non ordinary symbol.

? M = msinit(4,6,1); p = 3;
? Tp = mshecke(M, p); factor(charpoly(Tp))
%2 =
[x - 244 3]
[ x + 12 1]
? phi = matker(Tp + 12)[,1] \\ a_p = -12 is divisible by p = 3
%3 = [-1/32, -1/4, -1/32, 1]~
? msissymbol(M,phi)
%4 = 1
? Mp = mspadicinit(M,3,5,0);
? PHI = mstooms(Mp,phi);
*** at top-level: PHI=mstooms(Mp,phi)
*** ^---------------
*** mstooms: incorrect type in mstooms [v_p(ap) > mspadicinit flag] (t_VEC).
? Mp = mspadicinit(M,3,5,1);
? PHI = mstooms(Mp,phi);

The library syntax is GEN mstooms(GEN Mp, GEN phi).

3.19 Plotting functions.

Although plotting is not even a side purpose of PARI, a number of plotting functions are
provided. There are three types of graphic functions.

3.19.1 High-level plotting functions. (all the functions starting with ploth) in which the user
has little to do but explain what type of plot he wants, and whose syntax is similar to the one used
in the preceding section.

591
3.19.2 Low-level plotting functions. (called rectplot functions, sharing the prefix plot), where
every drawing primitive (point, line, box, etc.) is specified by the user. These low-level functions
work as follows. You have at your disposal 16 virtual windows which are filled independently, and
can then be physically ORed on a single window at user-defined positions. These windows are
numbered from 0 to 15, and must be initialized before being used by the function plotinit, which
specifies the height and width of the virtual window (called a rectwindow in the sequel). At all
times, a virtual cursor (initialized at [0, 0]) is attached to the window, and its current value can be
obtained using the function plotcursor.
A number of primitive graphic objects (called rect objects) can then be drawn in these windows,
using a default color attached to that window (which can be changed using the plotcolor function)
and only the part of the object which is inside the window will be drawn, with the exception of
polygons and strings which are drawn entirely. The ones sharing the prefix plotr draw relatively
to the current position of the virtual cursor, the others use absolute coordinates. Those having the
prefix plotrecth put in the rectwindow a large batch of rect objects corresponding to the output
of the related ploth function.
Finally, the actual physical drawing is done using plotdraw. The rectwindows are preserved
so that further drawings using the same windows at different positions or different windows can be
done without extra work. To erase a window, use plotkill. It is not possible to partially erase a
window: erase it completely, initialize it again, then fill it with the graphic objects that you want
to keep.
In addition to initializing the window, you may use a scaled window to avoid unnecessary
conversions. For this, use plotscale. As long as this function is not called, the scaling is simply
the number of pixels, the origin being at the upper left and the y-coordinates going downwards.
Plotting functions are platform independent, but a number of graphical drivers are available
for screen output: X11-windows (including Openwindows and Motif), Windows’s Graphical Device
Interface, the Qt and FLTK graphical libraries and one may even write the graphical objects to
a PostScript or SVG file and use an external viewer to open it. The physical window opened by
plotdraw or any of the ploth* functions is completely separated from gp (technically, a fork is
done, and all memory unrelated to the graphics engine is immediately freed in the child process),
which means you can go on working in the current gp session, without having to kill the window
first. This window can be closed, enlarged or reduced using the standard window manager functions.
No zooming procedure is implemented though.

3.19.3 Functions for PostScript or SVG output. in the same way that printtex allows you to
have a TEX output corresponding to printed results, the functions plotexport, plothexport and
plothrawexport convert a plot to a character string in either PostScript or Scalable Vector
Graphics format. This string can then be written to a file in the customary way, using write.
These export routines are available even if no Graphic Library is.

3.19.4 parploth(X = a, b, expr , {flags = 0}, {n = 0}). Parallel version of ploth. High precision
plot of the function y = f (x) represented by the expression expr , x going from a to b. This opens
a specific window (which is killed whenever you click on it), and returns a four-component vector
giving the coordinates of the bounding box in the form [xmin, xmax , ymin, ymax ].

592
Important note. parploth may evaluate expr thousands of times; given the relatively low
resolution of plotting devices, few significant digits of the result will be meaningful. Hence you
should keep the current precision to a minimum (e.g. 9) before calling this function.
The parameter n specifies the number of reference point on the graph, where a value of 0
means we use the hardwired default values; the binary digits of flag have the same meaning as
in ploth: 1 = Parametric; 2 = Recursive; 4 = no Rescale; 8 = no X axis; 16 = no Y axis;
32 = no Frame; 64 = no Lines; 128 = Points too; 256 = Splines; 512 = no X ticks; 1024 =
no Y ticks; 2048 = Same ticks; 4096 = Complex.
For instance:
\\ circle
parploth(X=0,2*Pi,[sin(X),cos(X)], "Parametric")
\\ two entwined sinusoidal curves
parploth(X=0,2*Pi,[sin(X),cos(X)])
\\ circle cut by the line y = x
parploth(X=0,2*Pi,[X,X,sin(X),cos(X)], "Parametric")
\\ circle
parploth(X=0,2*Pi,exp(I*X), "Complex")
\\ circle cut by the line y = x
parploth(X=0,2*Pi,[(1+I)*X,exp(I*X)], "Complex")
The library syntax is parploth(GEN a,GEN b,GEN code, long flag, long n, long prec).

3.19.5 parplothexport(fmt, X = a, b, expr , {flags = 0}, {n = 0}). Parallel version of plothex-

port. Plot of expression expr , X goes from a to b in high resolution, returning the resulting picture
as a character string which can then be written to a file.
The format fmt is either "ps" (PostScript output) or "svg" (Scalable Vector Graphics). All
other parameters and flags are as in ploth.
? s = parplothexport("svg", x=1,10, x^2+3);
? write("graph.svg", s);
The above only works if graph.svg does not already exist, otherwise write will append to the
existing file and produce an invalid svg. Here is a version that truncates an existing file (beware!):
? n = fileopen("graph.svg", "w");
? filewrite(n, s);
? fileclose(n);
This is intentionally more complicated.
The library syntax is parplothexport(GEN fmt, GEN a, GEN b, GEN code, long flags,
long n, long prec),

3.19.6 plot(X = a, b, expr , {Ymin}, {Ymax }). Crude ASCII plot of the function represented by
expression expr from a to b, with Y ranging from Ymin to Ymax . If Ymin (resp. Ymax ) is not
given, the minimum (resp. the maximum) of the computed values of the expression is used instead.
The library syntax is pariplot(void *E, GEN (*eval)(void*, GEN), GEN a, GEN b, GEN
ymin, GEN ymax, long prec)

593
3.19.7 plotbox(w, x2 , y2 , {filled = 0}). Let (x1, y1) be the current position of the virtual cursor.
Draw in the rectwindow w the outline of the rectangle which is such that the points (x1, y1) and
(x2, y2) are opposite corners. Only the part of the rectangle which is in w is drawn. The virtual
cursor does not move. If filled = 1, fill the box.
The library syntax is void plotbox(long w, GEN x2, GEN y2, long filled).

3.19.8 plotclip(w). ‘clips’ the content of rectwindow w, i.e remove all parts of the drawing that
would not be visible on the screen. Together with plotcopy this function enables you to draw on
a scratchpad before committing the part you’re interested in to the final picture.
The library syntax is void plotclip(long w).

3.19.9 plotcolor(w, c). Set default color to c in rectwindow w. Return [R,G,B] value attached to
color. Possible values for c are
• a t_VEC or t_VECSMALL [R, G, B] giving the color RGB value (all 3 values are between 0 and
255), e.g. [250,235,215] or equivalently [0xfa, 0xeb, 0xd7] for antiquewhite;
• a t_STR giving a valid colour name (see the rgb.txt file in X11 distributions), e.g. "an-
tiquewhite" or an RGV value given by a # followed by 6 hexadecimal digits, e.g. "#faebd7" for
antiquewhite;
• a t_INT, an index in the graphcolormap default, factory setting are
1=black, 2=blue, 3=violetred, 4=red, 5=green, 6=grey, 7=gainsborough.
but this can be extended if needed.
? plotinit(0,100,100);
? plotcolor(0, "turquoise")
%2 = [64, 224, 208]
? plotbox(0, 50,50,1);
? plotmove(0, 50,50);
? plotcolor(0, 2) \\ blue
%4 = [0, 0, 255]
? plotbox(0, 50,50,1);
? plotdraw(0);
The library syntax is GEN plotcolor(long w, GEN c).

3.19.10 plotcopy(sourcew , destw , dx , dy, {flag = 0}). Copy the contents of rectwindow sourcew
to rectwindow destw with offset (dx,dy). If flag’s bit 1 is set, dx and dy express fractions of the size
of the current output device, otherwise dx and dy are in pixels. dx and dy are relative positions of
northwest corners if other bits of flag vanish, otherwise of: 2: southwest, 4: southeast, 6: northeast
corners
The library syntax is void plotcopy(long sourcew, long destw, GEN dx, GEN dy, long
flag).

3.19.11 plotcursor(w). Give as a 2-component vector the current (scaled) position of the virtual
cursor corresponding to the rectwindow w.
The library syntax is GEN plotcursor(long w).

594
3.19.12 plotdraw(w, {flag = 0}). Physically draw the rectwindow w. More generally, w can be of
the form [w1 , x1 , y1 , w2 , x2 , y2 , . . .] (number of components must be divisible by 3; the windows w1 ,
w2 , etc. are physically placed with their upper left corner at physical position (x1 , y1 ), (x2 , y2 ),. . .
respectively, and are then drawn together. Overlapping regions will thus be drawn twice, and the
windows are considered transparent. Then display the whole drawing in a window on your screen.
If flag 6= 0, x1 , y1 etc. express fractions of the size of the current output device
The library syntax is void plotdraw(GEN w, long flag).

3.19.13 plotexport(fmt, list, {flag = 0}). Draw list of rectwindows as in plotdraw(list,flag),

returning the resulting picture as a character string which can then be written to a file. The format
fmt is either "ps" (PostScript output) or "svg" (Scalable Vector Graphics).
? plotinit(0, 100, 100);
? plotbox(0, 50, 50);
? plotcolor(0, 2);
? plotbox(0, 30, 30);
? plotdraw(0); \\ watch result on screen
? s = plotexport("svg", 0);
? write("graph.svg", s); \\ dump result to file
The library syntax is GEN plotexport(GEN fmt, GEN list, long flag).

3.19.14 ploth(X = a, b, expr , {flag = 0}, {n = 0}). High precision plot of the function y = f (x)
represented by the expression expr , x going from a to b. This opens a specific window (which is
killed whenever you click on it), and returns a four-component vector giving the coordinates of the
bounding box in the form [xmin, xmax , ymin, ymax ].

Important note. ploth may evaluate expr thousands of times; given the relatively low resolution
of plotting devices, few significant digits of the result will be meaningful. Hence you should keep
the current precision to a minimum (e.g. 9) before calling this function.
n specifies the number of reference point on the graph, where a value of 0 means we use the
hardwired default values (1000 for general plot, 1500 for parametric plot, and 8 for recursive plot).
If no flag is given, expr is either a scalar expression f (X), in which case the plane curve
y = f (X) will be drawn, or a vector [f1 (X), . . . , fk (X)], and then all the curves y = fi (X) will be
drawn in the same window.
The binary digits of flag mean:
• 1 = Parametric: parametric plot. Here expr must be a vector with an even number of
components. Successive pairs are then understood as the parametric coordinates of a plane curve.
Each of these are then drawn.
For instance:
ploth(X=0,2*Pi,[sin(X),cos(X)], "Parametric")
ploth(X=0,2*Pi,[sin(X),cos(X)])
ploth(X=0,2*Pi,[X,X,sin(X),cos(X)], "Parametric")
draw successively a circle, two entwined sinusoidal curves and a circle cut by the line y = x.
• 2 = Recursive: recursive plot. If this is set, only one curve can be drawn at a time, i.e. expr
must be either a two-component vector (for a single parametric curve, and the parametric flag has

595
to be set), or a scalar function. The idea is to choose pairs of successive reference points, and if their
middle point is not too far away from the segment joining them, draw this as a local approximation
to the curve. Otherwise, add the middle point to the reference points. This is fast, and usually
more precise than usual plot. Compare the results of
ploth(X=-1,1, sin(1/X), "Recursive")
ploth(X=-1,1, sin(1/X))
for instance. But beware that if you are extremely unlucky, or choose too few reference points,
you may draw some nice polygon bearing little resemblance to the original curve. For instance you
should never plot recursively an odd function in a symmetric interval around 0. Try
ploth(x = -20, 20, sin(x), "Recursive")
to see why. Hence, it’s usually a good idea to try and plot the same curve with slightly different
parameters.
The other values toggle various display options:
• 4 = no Rescale: do not rescale plot according to the computed extrema. This is used in
conjunction with plotscale when graphing multiple functions on a rectwindow (as a plotrecth
call):
s = plothsizes();
plotinit(0, s[2]-1, s[2]-1);
plotscale(0, -1,1, -1,1);
plotrecth(0, t=0,2*Pi, [cos(t),sin(t)], "Parametric|no_Rescale")
plotdraw([0, -1,1]);
This way we get a proper circle instead of the distorted ellipse produced by
ploth(t=0,2*Pi, [cos(t),sin(t)], "Parametric")
• 8 = no X axis: do not print the x-axis.
• 16 = no Y axis: do not print the y-axis.
• 32 = no Frame: do not print frame.
• 64 = no Lines: only plot reference points, do not join them.
• 128 = Points too: plot both lines and points.
• 256 = Splines: use splines to interpolate the points.
• 512 = no X ticks: plot no x-ticks.
• 1024 = no Y ticks: plot no y-ticks.
• 2048 = Same ticks: plot all ticks with the same length.
• 4096 = Complex: is a parametric plot but where each member of expr is considered a complex
number encoding the two coordinates of a point. For instance:
ploth(X=0,2*Pi,exp(I*X), "Complex")
ploth(X=0,2*Pi,[(1+I)*X,exp(I*X)], "Complex")
will draw respectively a circle and a circle cut by the line y = x.
The library syntax is ploth(void *E, GEN (*eval)(void*, GEN), GEN a, GEN b, long
flag, long n, long prec),

596
3.19.15 plothexport(fmt, X = a, b, expr , {flags = 0}, {n = 0}). Plot of expression expr , X goes
from a to b in high resolution, returning the resulting picture as a character string which can then
be written to a file.
The format fmt is either "ps" (PostScript output) or "svg" (Scalable Vector Graphics). All
other parameters and flags are as in ploth.
? s = plothexport("svg", x=1,10, x^2+3);
? write("graph.svg", s);
The library syntax is plothexport(GEN fmt, void *E, GEN (*eval)(void*, GEN), GEN
a, GEN b, long flags, long n, long prec),

3.19.16 plothraw(X, Y, {flag = 0}). Given X and Y two vectors of equal length, plots (in high
precision) the points whose (x, y)-coordinates are given in X and Y . Automatic positioning and
scaling is done, but with the same scaling factor on x and y. If flag is 1, join points, other nonzero
flags toggle display options and should be combinations of bits 2k , k ≥ 3 as in ploth.
The library syntax is GEN plothraw(GEN X, GEN Y, long flag).

3.19.17 plothrawexport(fmt, X, Y, {flag = 0}). Given X and Y two vectors of equal length,
plots (in high precision) the points whose (x, y)-coordinates are given in X and Y , returning the
resulting picture as a character string which can then be written to a file. The format fmt is either
"ps" (PostScript output) or "svg" (Scalable Vector Graphics).
Automatic positioning and scaling is done, but with the same scaling factor on x and y. If flag
is 1, join points, other nonzero flags toggle display options and should be combinations of bits 2k ,
k ≥ 3 as in ploth.
The library syntax is GEN plothrawexport(GEN fmt, GEN X, GEN Y, long flag).

3.19.18 plothsizes({flag = 0}). Return data corresponding to the output window in the form of
a 8-component vector: window width and height, sizes for ticks in horizontal and vertical directions
(this is intended for the gnuplot interface and is currently not significant), width and height of
characters, width and height of display, if applicable. If display has no sense, e.g. for svg plots or
postscript plots, then width and height of display are set to 0.
If flag = 0, sizes of ticks and characters are in pixels, otherwise are fractions of the screen size
The library syntax is GEN plothsizes(long flag).

3.19.19 plotinit(w, {x}, {y}, {flag = 0}). Initialize the rectwindow w, destroying any rect objects
you may have already drawn in w. The virtual cursor is set to (0, 0). The rectwindow size is set
to width x and height y; omitting either x or y means we use the full size of the device in that
direction. If flag = 0, x and y represent pixel units. Otherwise, x and y are understood as fractions
of the size of the current output device (hence must be between 0 and 1) and internally converted
to pixels.
The plotting device imposes an upper bound for x and y, for instance the number of pixels
for screen output. These bounds are available through the plothsizes function. The following
sequence initializes in a portable way (i.e independent of the output device) a window of maximal
size, accessed through coordinates in the [0, 1000] × [0, 1000] range:
s = plothsizes();

597
 plotinit(0, s[1]-1, s[2]-1);
plotscale(0, 0,1000, 0,1000);
The library syntax is void plotinit(long w, GEN x = NULL, GEN y = NULL, long flag)
.

3.19.20 plotkill(w). Erase rectwindow w and free the corresponding memory. Note that if you
want to use the rectwindow w again, you have to use plotinit first to specify the new size. So
it’s better in this case to use plotinit directly as this throws away any previous work in the given
rectwindow.
The library syntax is void plotkill(long w).

3.19.21 plotlines(w, X, Y, {flag = 0}). Draw on the rectwindow w the polygon such that the
(x,y)-coordinates of the vertices are in the vectors of equal length X and Y . For simplicity, the
whole polygon is drawn, not only the part of the polygon which is inside the rectwindow. If flag is
nonzero, close the polygon. In any case, the virtual cursor does not move.
X and Y are allowed to be scalars (in this case, both have to). There, a single segment will be
drawn, between the virtual cursor current position and the point (X, Y ). And only the part thereof
which actually lies within the boundary of w. Then move the virtual cursor to (X, Y ), even if it
is outside the window. If you want to draw a line from (x1, y1) to (x2, y2) where (x1, y1) is not
necessarily the position of the virtual cursor, use plotmove(w,x1,y1) before using this function.
The library syntax is void plotlines(long w, GEN X, GEN Y, long flag).

3.19.22 plotlinetype(w, type). This function is obsolete and currently a no-op.

Change the type of lines subsequently plotted in rectwindow w. type −2 corresponds to frames,
−1 to axes, larger values may correspond to something else. w = −1 changes highlevel plotting.
The library syntax is void plotlinetype(long w, long type).

3.19.23 plotmove(w, x, y). Move the virtual cursor of the rectwindow w to position (x, y).
The library syntax is void plotmove(long w, GEN x, GEN y).

3.19.24 plotpoints(w, X, Y ). Draw on the rectwindow w the points whose (x, y)-coordinates are
in the vectors of equal length X and Y and which are inside w. The virtual cursor does not move.
This is basically the same function as plothraw, but either with no scaling factor or with a scale
chosen using the function plotscale.
As was the case with the plotlines function, X and Y are allowed to be (simultaneously)
scalar. In this case, draw the single point (X, Y ) on the rectwindow w (if it is actually inside w),
and in any case move the virtual cursor to position (x, y).
If you draw few points in the rectwindow, they will be hard to see; in this case, you can use
filled boxes instead. Compare:
? plotinit(0, 100,100); plotpoints(0, 50,50);
? plotdraw(0)
? plotinit(1, 100,100); plotmove(1,48,48); plotrbox(1, 4,4, 1);
? plotdraw(1)
The library syntax is void plotpoints(long w, GEN X, GEN Y).

598
3.19.25 plotpointsize(w, size). This function is obsolete. It is currently a no-op.

Changes the “size” of following points in rectwindow w. If w = −1, change it in all rectwindows.

The library syntax is void plotpointsize(long w, GEN size).

3.19.26 plotpointtype(w, type). This function is obsolete and currently a no-op.

change the type of points subsequently plotted in rectwindow w. type = −1 corresponds to a

dot, larger values may correspond to something else. w = −1 changes highlevel plotting.

The library syntax is void plotpointtype(long w, long type).

3.19.27 plotrbox(w, dx , dy, {filled }). Draw in the rectwindow w the outline of the rectangle which
is such that the points (x1, y1) and (x1 + dx, y1 + dy) are opposite corners, where (x1, y1) is the
current position of the cursor. Only the part of the rectangle which is in w is drawn. The virtual
cursor does not move. If filled = 1, fill the box.

The library syntax is void plotrbox(long w, GEN dx, GEN dy, long filled).

3.19.28 plotrecth(w, X = a, b, expr , {flag = 0}, {n = 0}). Writes to rectwindow w the curve
output of ploth(w, X = a, b, expr , flag, n). Returns a vector for the bounding box.

3.19.29 plotrecthraw(w, data, {flags = 0}). Plot graph(s) for data in rectwindow w; flag has the
same meaning here as in ploth, though recursive plot is no longer significant.

The argument data is a vector of vectors, each corresponding to a list a coordinates. If

parametric plot is set, there must be an even number of vectors, each successive pair corresponding
to a curve. Otherwise, the first one contains the x coordinates, and the other ones contain the
y-coordinates of curves to plot.

The library syntax is GEN plotrecthraw(long w, GEN data, long flags).

3.19.30 plotrline(w, dx , dy). Draw in the rectwindow w the part of the segment (x1, y1) − (x1 +
dx, y1 + dy) which is inside w, where (x1, y1) is the current position of the virtual cursor, and move
the virtual cursor to (x1 + dx, y1 + dy) (even if it is outside the window).

The library syntax is void plotrline(long w, GEN dx, GEN dy).

3.19.31 plotrmove(w, dx , dy). Move the virtual cursor of the rectwindow w to position (x1 +
dx, y1 + dy), where (x1, y1) is the initial position of the cursor (i.e. to position (dx, dy) relative to
the initial cursor).

The library syntax is void plotrmove(long w, GEN dx, GEN dy).

599
3.19.32 plotrpoint(w, dx , dy). Draw the point (x1 + dx, y1 + dy) on the rectwindow w (if it is
inside w), where (x1, y1) is the current position of the cursor, and in any case move the virtual
cursor to position (x1 + dx, y1 + dy).
If you draw few points in the rectwindow, they will be hard to see; in this case, you can use
filled boxes instead. Compare:
? plotinit(0, 100,100); plotrpoint(0, 50,50); plotrpoint(0, 10,10);
? plotdraw(0)
? thickpoint(w,x,y)= plotmove(w,x-2,y-2); plotrbox(w,4,4,1);
? plotinit(1, 100,100); thickpoint(1, 50,50); thickpoint(1, 60,60);
? plotdraw(1)
The library syntax is void plotrpoint(long w, GEN dx, GEN dy).

3.19.33 plotscale(w, x1 , x2 , y1 , y2 ). Scale the local coordinates of the rectwindow w so that x

goes from x1 to x2 and y goes from y1 to y2 (x2 < x1 and y2 < y1 being allowed). Initially, after
the initialization of the rectwindow w using the function plotinit, the default scaling is the graphic
pixel count, and in particular the y axis is oriented downwards since the origin is at the upper left.
The function plotscale allows to change all these defaults and should be used whenever functions
are graphed.
The library syntax is void plotscale(long w, GEN x1, GEN x2, GEN y1, GEN y2)
.

3.19.34 plotstring(w, x, {flags = 0}). Draw on the rectwindow w the String x (see Section 2.9),
at the current position of the cursor.
flag
is used for justification: bits 1 and 2 regulate horizontal alignment: left if 0, right if 2, center
if 1. Bits 4 and 8 regulate vertical alignment: bottom if 0, top if 8, v-center if 4. Can insert
additional small gap between point and string: horizontal if bit 16 is set, vertical if bit 32 is set
(see the tutorial for an example).
The library syntax is void plotstring(long w, const char *x, long flags).

3.19.35 psdraw(list, {flag = 0}). This function is obsolete, use plotexport and write the result to
file.
The library syntax is void psdraw(GEN list, long flag).

3.19.36 psploth(X = a, b, expr , {flags = 0}, {n = 0}). This function is obsolete, use plothexport
and write the result to file.
The library syntax is GEN psploth0(GEN X, GEN b, GEN expr, long flags, long prec)
.

3.19.37 psplothraw(listx , listy, {flag = 0}). This function is obsolete, use plothrawexport and
write the result to file.
The library syntax is GEN psplothraw(GEN listx, GEN listy, long flag).

600
 Appendix A:
Installation Guide for the UNIX Versions

3.1 Required tools.

Compiling PARI requires an ANSI C or a C++ compiler. If you do not have one, we suggest
that you obtain the gcc/g++ compiler. As for all GNU software mentioned afterwards, you can
find the most convenient site to fetch gcc at the address

http://www.gnu.org/order/ftp.html

(On Mac OS X, this is also provided in the Xcode tool suite; or the lightweight “Command-Line
Tools (CLT) for Xcode”.) You can compile PARI with a different compiler, but the PARI kernel
takes advantage of optimizations provided by gcc. Compiling with GCC results in at least 20%
speedup on most architectures.

Optional libraries and programs. The following programs and libraries are useful in conjunction
with gp, but not mandatory. In any case, get them before proceeding if you want the functionality
they provide. All of libraries and programs are free. The download page on our website

http://pari.math.u-bordeaux.fr/download.html

contains pointers on how to get them.

• GNU MP (GMP) library. GMP provides an alternative multiprecision kernel, which is faster
than PARI’s native one. When using GMP the binaries become incompatible, so the resulting
PARI library SONAME is libpari-gmp.

• GNU readline library. Readline provides line editing under gp, an automatic context-
dependent completion, and an editable history of commands.

• GNU emacs and the PariEmacs package. The gp calculator can be run in an Emacs buffer,
with all the obvious advantages if you are familiar with this editor. Note that readline is still
useful in this case since it provides a better automatic completion than is provided by Emacs’s
GP-mode.

• GNU gzip/gunzip/gzcat package enables gp to read compressed data.

• perl provides extended online help (full text from the manual) about functions and concepts.
The script handling this online help can be used under gp or independently.

601
3.2 Compiling the library and the gp calculator.

3.2.1 Basic configuration. Type

./Configure
in the toplevel directory. This attempts to configure PARI/GP without outside help. Note that if
you want to install the end product in a nonstandard place, you can use the --prefix option, as
in
./Configure --prefix=/an/exotic/directory
(the default prefix is /usr/local). For example, to build a package for a Linux distribution, you
may want to use
./Configure --prefix=/usr
The command
./Configure --builddir=DIR
configures the build for a particular system. It extracts some files and creates the build directory
DIR, where the object files and executables will be built. Thus you can build PARI/GP for several
different machines from the same source tree. When building for multiple machines the builds are
independent and can be performed simultaneously.
If the --builddir=DIR argument is omitted, the build directory is named Oosname-arch,
where the osname and arch components depends on your architecture and operating system. A de-
bug build (Configure -g) or a profiling build (Configure -pg) bring in an extra .dbg (resp. .prf)
suffix.
Decide whether you agree with what Configure printed on your screen, in particular the ar-
chitecture, compiler and optimization flags. Look for messages prepended by ###, which report
genuine problems. Look especially for the gmp, readline and X11 libraries, and the perl and gun-
zip (or zcat) binaries. If anything should have been found and was not, consider that Configure
failed and follow the instructions in section 3.
The Configure script creates a file config.log in the build directory, which contains debug-
ging information — in particular, all messages from compilers — that may help diagnose problems.
This file is erased and recreated from scratch each time Configure is run.

3.2.2 Advanced configuration. Configure accepts many other flags, and you may use any
number of them to build quite a complicated configuration command. See Configure --help for
a complete list. In particular, GMP is configured using the set of flags --with-gmp*, and Readline
is configured using the set of flags --with-readline*.
--with-gmp: enables a particular installation of GMP. If the GMP library was installed in
/an/exotic/directory, then you can use --with-gmp=/an/exotic/directory to find the headers
and libraries.
If you are using GMP, tune it first, then tune PARI. Make sure you tune PARI on the machine
that will actually run your computations. Do not use a heavily loaded machine for tunings.
--with-readline: enables a particular installation of Readline. If Readline was installed in
/an/exotic/directory, then you can use --with-readline=/an/exotic/directory to find the
headers and libraries.

602
 make -jN: You may speed up the compilation by using a parallel make. The command below
tells make to execute 4 recipes simultaneously.
env MAKE="make -j4" Configure
Next, we focus on some of the nonobvious options used by Configure:
--tune: fine tunes the library for the host used for compilation. Tuning the package ad-
justs thresholds by running a large number of comparative tests and creates a file tune.h in the
build directory, that will be used from now on, overriding the ones in src/kernel/none/ and
src/kernel/gmp/. It will take a while: about 30 minutes. Expect a small performance boost,
perhaps a 10% speed increase compared to default settings.
--graphic=lib: enables a particular graphic library. The default is X11 on most platforms,
but PARI can use Qt, fltk or win32 (GDI), or even dump a ps or svg file and open it using an
external viewer.
--time=function: chooses a timing function. The default usually works fine, however you can
use a different one that better fits your needs. PARI can use getrusage, clock gettime, times or
ftime as timing functions. (Not all timing functions are available on all platforms.) The three first
functions give timings in terms of CPU usage of the current task, approximating the complexity of
the algorithm. The last one, ftime, gives timings in terms of absolute (wall-clock) time. Moreover,
the clock gettime function is more precise, but much slower (at the time of this writing), than
getrusage or times.
--with-runtime-perl=perl : absolute path to the runtime perl binary to be used by the
gphelp and tex2mail scripts. Defaults to the path found by Configure on the build host (usually
/usr/bin/perl). For cross-compiling builds, when the target and build hosts have mismatched
configurations; suggested values are
/usr/bin/env perl: the first perl executable found in user’s PATH,
/usr/bin/perl: perl’s standard location.
The remaining options are specific to parallel programming. We provide an Introduction to
parallel GP programming in the file doc/parallel.dvi, and to multi-threaded libpari programs
in Appendix D. Beware that these options change the library ABI:
--mt=engine: specify the engine used for parallel computations. Supported value are
• single: (default) no parallellism.
• pthread: use POSIX threads. This is well-suited for multi-core systems. Setting this option
also set --enable-tls, see below. This option requires the pthread library. For benchmarking, it
is often useful to set --time=ftime so that GP report wall-clock instead of the sum of the time
spent by each thread.
• mpi: use the MPI interface to parallelism. This allows PARI to take advantage of clusters
using MPI. This option requires a MPI library. It is usually necessary to set the environment
variable CC to mpicc.
--enable-tls: build the thread-safe version of the library. Implied by --mt=pthread. This
tends to slow down the shared library libpari.so by about 25%, so you probably want to use the
static library libpari.a instead.

603
3.2.3 Compilation. To compile the gp binary and build the documentation, type

make all

from the build directory. Recall the build directory is (by default) of the form Oosname-arch or
passed to Configure using the option --build=DIR. To only compile the gp binary, type

make gp

in the toplevel directory (this will correctly determine the Oosname-arch default directory) or make
-C DIR gp (it is also possible to cd DIR first).

If your make program supports parallel make, you can speed up the process by going to the
build directory that Configure created and doing a parallel make here, for instance make -j4 with
GNU make. It should also work from the toplevel directory.

3.2.4 Basic tests.

To test the binary, type make bench. This runs a quick series of tests, for a few seconds on
modern machines.

In many cases, this will also build a different binary (named gp-sta or gp-dyn) linked in a
slightly different way and run the tests with both. (In exotic configurations, one may pass all the
tests while the other fails and we want to check for this.) To test only the default binary, use make
dobench which starts the bench immediately.

If a [BUG] message shows up, something went wrong. The testing utility directs you to files
containing the differences between the test output and the expected results. Have a look and decide
for yourself if something is amiss. If it looks like a bug in the Pari system, we would appreciate a
report, see the last section.

3.2.5 Cross-compiling.

When cross-compiling, you can set the environment variable RUNTEST to a program that is able
to run the target binaries, e.g. an emulator. It will be used for both the Configure tests and make
bench.

3.3 Troubleshooting and fine tuning.

In case the default Configure run fails miserably, try

./Configure -a

(interactive mode) and answer all the questions: there are about 30 of them, and default answers
are provided. If you accept all default answers, Configure will fail just the same, so be wary. In
any case, we would appreciate a bug report (see the last section).

604
3.3.1 Installation directories. The install location of programs and libraries is controlled by the
--prefix configure option. The default location is /usr/local.

The precise destinations of programs and libraries are as follows: the gp binary, the scripts
gphelp and tex2mail go to $prefix/bin. The pari library goes to $prefix/lib and include files
to $prefix/include/pari. Other system-dependent data go to $prefix/lib/pari.

Architecture independent files go to various subdirectories of $share prefix, which defaults

to $prefix/share, and can be specified via the --share-prefix argument. Man pages go into
$share prefix/man, and other system-independent data under $share prefix/pari: documenta-
tion, sample GP scripts and C code, extra packages like elldata or galdata.

You can also set directly --bindir (executables), --libdir (library), --includedir (include
files), --mandir (manual pages), --datadir (other architecture-independent data), and finally
--sysdatadir (other architecture-dependent data).

3.3.2 Environment variables. Configure lets the following environment variable override the
defaults if set:

CC: C compiler.

DLLD: Dynamic library linker.

LD: Static linker.

For instance, Configure may avoid /bin/cc on some architectures due to various problems which
may have been fixed in your version of the compiler. You can try

env CC=cc Configure

and compare the benches. Also, if you insist on using a C++ compiler and run into trouble with a
fussy g++, try to use g++ -fpermissive.

The contents of the following variables are appended to the values computed by Configure:

CFLAGS: Flags for CC.

CPPFLAGS: Flags for CC (preprocessor).

LDFLAGS: Flags for LD.

The contents of the following variables are prepended to the values computed by Configure:

C INCLUDE PATH is prepended to the list of directories searched for include files. Note that
adding -I flags to CFLAGS is not enough since Configure sometimes relies on finding the include
files and parsing them, and it does not parse CFLAGS at this time.

LIBRARY PATH is prepended to the list of directories searched for libraries.

You may disable inlining by adding -DDISABLE INLINE to CFLAGS, and prevent the use of the
volatile keyword with -DDISABLE VOLATILE.

605
3.3.3 Debugging/profiling.: If you also want to debug the PARI library,

Configure -g

creates a directory Oxxx.dbg containing a special Makefile ensuring that the gp and PARI library
built there is suitable for debugging. If you want to profile gp or the library, using gprof for
instance,

Configure -pg

will create an Oxxx.prf directory where a suitable version of PARI can be built.

The gp binary built above with make all or make gp is optimized. If you have run Configure
-g or -pg and want to build a special purpose binary, you can cd to the .dbg or .prf directory and
type make gp there. You can also invoke make gp.dbg or make gp.prf directly from the toplevel.

3.3.4 Multiprecision kernel. The kernel can be specified via the

--kernel=fully qualified kernel name

switch. The PARI kernel consists of two levels: Level 0 (operation on words) and Level 1 (operation
on multi-precision integers and reals), which can take the following values.

Level 0: auto (as detected), none (portable C) or one of the assembler micro-kernels

alpha
hppa hppa64
ia64
ix86 x86_64
m68k
ppc ppc64
sparcv7 sparcv8_micro sparcv8_super

Level 1: auto (as detected), none (native code only), or gmp

• A fully qualified kernel name is of the form Level0 -Level1 , the default value being auto-auto.

• A name not containing a dash ’-’ is an alias for a fully qualified kernel name. An alias stands for
name-none, but gmp stands for auto-gmp.

3.3.5 Problems related to readline. Configure does not try very hard to find the readline
library and include files. If they are not in a standard place, it will not find them. You can invoke
Configure with one of the following arguments:

--with-readline[=prefix to lib/libreadline.xx and include/readline.h]

--with-readline-lib=path to libreadline.xx

--with-readline-include=path to readline.h

606
Known problems.
• on Linux: Linux distributions have separate readline and readline-devel packages. You
need both of them installed to compile gp with readline support. If only readline is installed,
Configure will complain. Configure may also complain about a missing libncurses.so, in which
case, you have to install the ncurses-devel package (some distributions let you install readline-
devel without ncurses-devel, which is a bug in their package dependency handling).
• on OS X.4 or higher: these systems comes equipped with a fake readline, which is not
sufficient for our purpose. As a result, gp is built without readline support. Since readline is not
trivial to install in this environment, a step by step solution can be found in the PARI FAQ, see

http://pari.math.u-bordeaux.fr/.

3.3.6 Testing.
3.3.6.1 Known problems. if BUG shows up in make bench
• If when running gp-dyn, you get a message of the form
ld.so: warning: libpari.so.xxx has older revision than expected xxx
(possibly followed by more errors), you already have a dynamic PARI library installed and a broken
local configuration. Either remove the old library or unset the LD LIBRARY PATH environment
variable. Try to disable this variable in any case if anything very wrong occurs with the gp-dyn
binary, like an Illegal Instruction on startup. It does not affect gp-sta.
• Some implementations of the diff utility (on HPUX for instance) output No differences
encountered or some similar message instead of the expected empty input, thus producing a
spurious [BUG] message.
3.3.6.2 Some more testing. [Optional]
You can test gp in compatibility mode with make test-compat. If you want to test the graphic
routines, use make test-ploth. You will have to click on the mouse button after seeing each image.
There will be eight of them, probably shown twice (try to resize at least one of them as a further
test).
The make bench, make test-compat and make test-ploth runs all produce a Postscript file
pari.ps in Oxxx which you can send to a Postscript printer. The output should bear some similarity
to the screen images.
3.3.6.3 Heavy-duty testing. [Optional] There are a few extra tests which should be useful only
for developers.
make test-kernel checks whether the low-level kernel seems to work, and provides simple
diagnostics if it does not. Only useful if make bench fails horribly, e.g. things like 1+1 do not work.
make test-all runs all available test suites. Thorough, but slow. Some of the tests require
extra packages (elldata, galdata, etc.) to be available. If you want to test such an extra package
before make install (which would install it to its final location, where gp expects to find it), run
env GP_DATA_DIR=$PWD/data make test-all
from the PARI toplevel directory, otherwise the test will fail.

607
 make test-io tests writing to and reading from files. It requires a working system() command
(fails on Windows + MingW).
make test-time tests absolute and relative timers. This test has a tendency to fail when the
machine is heavily loaded or if the granularity of the chosen system timer is bigger than 2ms. Try
it a few times before reporting a problem.
make test-install tests the GP function install. This may not be available on your plat-
form, triggering an error message (“not yet available for this architecture”). The implementation
may be broken on your platform triggering an error or a crash when an install’ed function is used.

3.4 Installation.
When everything looks fine, type
make install
You may have to do this with superuser privileges, depending on the target directories. (Tip for
MacOS X beginners: use sudo make install.) In this case, it is advised to type make all first
to avoid running unnecessary commands as root.

Caveat. Install directories are created honouring your umask settings: if your umask is too re-
strictive, e.g. 077, the installed files will not be world-readable. (Beware that running sudo may
change your user umask.)
This installs in the directories chosen at Configure time the default gp executable (probably
gp-dyn) under the name gp, the default PARI library (probably libpari.so), the necessary include
files, the manual pages, the documentation and help scripts.
To save on disk space, you can manually gzip some of the documentation files if you wish:
usersch*.tex and all dvi files (assuming your xdvi knows how to deal with compressed files); the
online-help system can handle it.

3.4.1 Static binaries and libraries. By default, if a dynamic library libpari.so can be built,
the gp binary we install is gp-dyn, pointing to libpari.so. On the other hand, we can build a gp
binary into which the libpari is statically linked (the library code is copied into the binary); that
binary is not independent of the machine it was compiled on, and may still refer to other dynamic
libraries than libpari.
You may want to compile your own programs in the same way, using the static libpari.a
instead of libpari.so. By default this static library libpari.a is not created. If you want it as
well, use the target make install-lib-sta. You can install a statically linked gp with the target
make install-bin-sta. As a rule, programs linked statically (with libpari.a) may be slightly
faster (about 5% gain, possibly up to 20% when using pthreads), but use more disk space and
take more time to compile. They are also harder to upgrade: you will have to recompile them all
instead of just installing the new dynamic library. On the other hand, there is no risk of breaking
them by installing a new pari library.

608
3.4.2 Extra packages. The following optional packages endow PARI with some extra capabilities:
• elldata: This package contains the elliptic curves in John Cremona’s database. It is needed
by the functions ellidentify, ellsearch, forell and can be used by ellinit to initialize a curve
given by its standard code.
• galdata: The default polgalois function can only compute Galois groups of polynomials
of degree less or equal to 7. Install this package if you want to handle polynomials of degree bigger
than 7 (and less than 11).
• seadata: This package contains the database of modular polynomials extracted from the
ECHIDNA databases and computed by David R. Kohel. It is used to speed up the functions ellap,
ellcard and ellgroup for primes larger than 1020 .
• galpol: This package contains the GALPOL database of polynomials defining Galois exten-
sions of the rationals, accessed by galoisgetpol.

To install package pack , you need to fetch the separate archive: pack .tgz which you can
download from the pari server. Copy the archive in the PARI toplevel directory, then extract its
contents; these will go to data/pack /. Typing make install installs all such packages.

3.4.3 The GPRC file. Copy the file misc/gprc.dft (or gprc.dos if you are using GP.EXE) to
$HOME/.gprc. Modify it to your liking. For instance, if you are not using an ANSI terminal,
remove control characters from the prompt variable. You can also enable colors.
If desired, read $datadir/misc/gpalias from the gprc file, which provides some common
shortcuts to lengthy names; fix the path in gprc first. (Unless you tampered with this via Configure,
datadir is $prefix/share/pari.) If you have superuser privileges and want to provide system-
wide defaults, copy your customized .gprc file to /etc/gprc.
In older versions, gphelp was hidden in pari lib directory and was not meant to be used from
the shell prompt, but not anymore. If gp complains it cannot find gphelp, check whether your
.gprc (or the system-wide gprc) does contain explicit paths. If so, correct them according to the
current misc/gprc.dft.

3.5 Getting Started.

3.5.1 Printable Documentation. Building gp with make all also builds its documentation.
You can also type directly make doc. In any case, you need a working (plain) TEX installation.
After that, the doc directory contains various dvi files: libpari.dvi (manual for the PARI
library), users.dvi (manual for the gp calculator), tutorial.dvi (a tutorial), and refcard.dvi
(a reference card for GP). You can send these files to your favorite printer in the usual way, probably
via dvips. The reference card is also provided as a PostScript document, which may be easier to
print than its dvi equivalent (it is in Landscape orientation and assumes A4 paper size).
If pdftex is part of your TEX setup, you can produce these documents in PDF format, which may
be more convenient for online browsing (the manual is complete with hyperlinks); type
make docpdf
All these documents are available online from PARI home page (see the last section).

609
3.5.2 C programming. Once all libraries and include files are installed, you can link your C
programs to the PARI library. A sample makefile examples/Makefile is provided to illustrate the
use of the various libraries. Type make all in the examples directory to see how they perform on
the extgcd.c program, which is commented in the manual.
This should produce a statically linked binary extgcd-sta (standalone), a dynamically linked
binary extgcd-dyn (loads libpari at runtime) and a shared library libextgcd, which can be used
from gp to install your new extgcd command.
The standalone binary should be bulletproof, but the other two may fail for various reasons. If
when running extgcd-dyn, you get a message of the form “DLL not found”, then stick to statically
linked binaries or look at your system documentation to see how to indicate at linking time where
the required DLLs may be found! (E.g. on Windows, you will need to move libpari.dll somewhere
in your PATH.)

3.5.3 GP scripts. Several complete sample GP programs are also given in the examples directory,
for example Shanks’s SQUFOF factoring method, the Pollard rho factoring method, the Lucas-
Lehmer primality test for Mersenne numbers and a simple general class group and fundamental
unit algorithm. See the file examples/EXPLAIN for some explanations.

3.5.4 The PARI Community. PARI’s home page at the address

http://pari.math.u-bordeaux.fr/

maintains an archive of mailing lists dedicated to PARI, documentation (including Frequently

Asked Questions), a download area and our Bug Tracking System (BTS). Bug reports should be
submitted online to the BTS, which may be accessed from the navigation bar on the home page or
directly at
http://pari.math.u-bordeaux.fr/Bugs/
Further information can be found at that address but, to report a configuration problem, make
sure to include the relevant *.dif files in the Oxxx directory and the file pari.cfg.

There are a number of mailing lists devoted to PARI/GP, and most feedback should be directed
there. Instructions and archives can be consulted at

http://pari.math.u-bordeaux.fr/lists-index.html

The most important are:

• pari-announce (read-only): to announce major version changes. You cannot write to this
one, but you should probably subscribe.
• pari-dev: for everything related to the development of PARI, including suggestions, tech-
nical questions or patch submissions. Bug reports can be discussed here, but as a rule it is better
to submit them directly to the BTS.
• pari-users: for everything else.
You may send an email to the last two without being subscribed. To subscribe, send an message
respectively to
[email protected]
[email protected]

610
 [email protected]
with the word subscribe in the Subject:. You can also write to us at the address
[email protected]
but we cannot promise you will get an individual answer.

If you have used PARI in the preparation of a paper, please cite it in the following form
(BibTeX format):
@preamble{\usepackage{url}}
@manual{PARI2,
organization = "{The PARI~Group}",
title = "{PARI/GP version 2.13.3}",
year = 2020,
address = "Bordeaux",
note = "available from \url{http://pari.math.u-bordeaux.fr/}"
}

In any case, if you like this software, we would be indebted if you could send us an email message
giving us some information about yourself and what you use PARI for.

Good luck and enjoy!

611
 Index

SomeWord refers to PARI-GP concepts. algebraic number . . . . . . . . . . . 347, 348

SomeWord is a PARI-GP keyword. alggroup . . . . . . . . . . . . . . . 446, 447
SomeWord is a generic index entry. alggroupcenter . . . . . . . . . . . . . . 447
alghasse . . . . . . . . . . . . . . . 447, 448
+ alghassef . . . . . . . . . . . . . . . . . 448
+oo . . . . . . . . . . . . . . . . . . 155, 158 alghassei . . . . . . . . . . . . . . . . . 448
algindex . . . . . . . . . . . . . . . 448, 449
- alginit 442, 443, 444, 445, 446, 447, 448, 449,
451, 452, 453, 454, 459, 460,
-LONG_MAX . . . . . . . . . . . . . . . . . 240 462, 463, 464, 466, 467, 468
alginv . . . . . . . . . . . . . . . . . . . 451
A
alginvbasis . . . . . . . . . . . . . . . . 451
a1 . . . . . . . . . . . . . . . . . . . . . . 469 algisassociative . . . . . . . . . . 451, 452
a2 . . . . . . . . . . . . . . . . . . . . . . 469 algiscommutative . . . . . . . . . . . . . 452
a3 . . . . . . . . . . . . . . . . . . . . . . 469 algisdivision . . . . . . . . . . . . . . . 452
a4 . . . . . . . . . . . . . . . . . . . . . . 469 algisdivl . . . . . . . . . . . . . . 452, 453
a6 . . . . . . . . . . . . . . . . . . . . . . 469 algisinv . . . . . . . . . . . . . . . . . . 453
Abelian extension . . . . . . . . . . 425, 437 algisramified . . . . . . . . . . . . . . . 453
abs . . . . . . . . . . . . . . . . . . . . . 295 algissemisimple . . . . . . . . . . 453, 454
accuracy . . . . . . . . . . . . . . . . . . . . 9 algissimple . . . . . . . . . . . . . . . . 454
acos . . . . . . . . . . . . . . . . . . . . . 295 algissplit . . . . . . . . . . . . . . 454, 455
acosh . . . . . . . . . . . . . . . . . . . . 295 alglatadd . . . . . . . . . . . . . . . . . 455
addhelp . . . . . . . . . . . . . . . 50, 85, 86 alglatcontains . . . . . . . . . . . . . . 455
addprimes . . . 122, 171, 191, 397, 410, 421 alglatelement . . . . . . . . . . . . 455, 456
adj . . . . . . . . . . . . . . . . . . . . . 261 alglathnf . . . . . . . . . . . . . . . . . 456
adjoint matrix . . . . . . . . . . . . . . . 261 alglatindex . . . . . . . . . . . . . . . . 456
adjsafe . . . . . . . . . . . . . . . . . . . 261 alglatinter . . . . . . . . . . . . . . . . 456
agm . . . . . . . . . . . . . . . . . . 295, 296 alglatlefttransporter . . . . . . . . . 457
airy . . . . . . . . . . . . . . . . . . . . . 296 alglatmul . . . . . . . . . . . . . . . . . 457
akell . . . . . . . . . . . . . . . . . . . . 474 alglatrighttransporter . . . . . . 457, 458
alarm . . . . . . . . . . . . . . . . . . 84, 86 alglatsubset . . . . . . . . . . . . . . . 458
algadd . . . . . . . . . . . . . . . . . . . 442 algmakeintegral . . . . . . . . . . 458, 459
algalgtobasis . . . . . . . . . . . . 442, 443 algmul . . . . . . . . . . . . . . . . . . . 459
algaut . . . . . . . . . . . . . . . . . . . 443 algmultable . . . . . . . . . . . . . . . . 459
algb . . . . . . . . . . . . . . . . . . . . . 443 algneg . . . . . . . . . . . . . . . . . . . 459
algbasis . . . . . . . . . . . . . . . . . . 443 algnorm . . . . . . . . . . . . . . . . . . . 460
algbasistoalg . . . . . . . . . 442, 443, 444 algpoleval . . . . . . . . . . . . . . . . . 460
algcenter . . . . . . . . . . . . . . . . . 444 algpow . . . . . . . . . . . . . . . . . . . 460
algcentralproj . . . . . . . . . . . . . . 444 algprimesubalg . . . . . . . . . . . . . . 461
algchar . . . . . . . . . . . . . . . . . . . 445 algquotient . . . . . . . . . . . . . . . . 461
algcharpoly . . . . . . . . . . . . . . . . 445 algradical . . . . . . . . . . . . . . 461, 462
algdegree . . . . . . . . . . . . . . . . . 445 algramifiedplaces . . . . . . . . . . . . 462
algdep . . . . . . . . . . . . . . . . 255, 256 algrandom . . . . . . . . . . . . . . . . . 462
algdep0 . . . . . . . . . . . . . . . . . . . 256 algrelmultable . . . . . . . . . . . . . . 462
algdim . . . . . . . . . . . . . . . . 445, 446 algsimpledec . . . . . . . . . . . . 462, 463
algdisc . . . . . . . . . . . . . . . . . . . 446 algsplit . . . . . . . . . . . . . . . . . . 463
algdivl . . . . . . . . . . . . . . . . . . . 446 algsplittingdata . . . . . . . . . . 463, 464
algdivr . . . . . . . . . . . . . . . . . . . 446 algsplittingfield . . . . . . . . . 464, 465
algebraic dependence . . . . . . . . . 255, 287 algsqr . . . . . . . . . . . . . . . . . . . 465

612
algsub . . . . . . . . . . . . . . . . . . . 465 bernvec . . . . . . . . . . . . . . . . 164, 165
algsubalg . . . . . . . . . . . . . . 465, 466 bervec . . . . . . . . . . . . . . . . . . . 163
algtableinit . 444, 445, 451, 452, 453, 454, besselh1 . . . . . . . . . . . . . . . . . . 296
458, 459, 460, 461, 462, 463, besselh2 . . . . . . . . . . . . . . . . . . 296
465, 466, 467, 468 besseli . . . . . . . . . . . . . . . . . . . 296
algtensor . . . . . . . . . . . . . . . . . 467 besselj . . . . . . . . . . . . . . . . . . . 297
algtobasis . . . . . . . . . . . . . . . . . 395 besseljh . . . . . . . . . . . . . . . . . . 297
algtomatrix . . . . . . . . . . . . . . . . 467 besselk . . . . . . . . . . . . . . . . . . . 297
algtrace . . . . . . . . . . . . . . . 467, 468 besseln . . . . . . . . . . . . . . . . . . . 297
algtype . . . . . . . . . . . . . . . . 468, 469 bessely . . . . . . . . . . . . . . . . . . . 297
alg_centralproj . . . . . . . . . . . . . 445 bestappr . . . . . . . . . . . . . . . 148, 172
alg_quotient . . . . . . . . . . . . . . . 461 bestapprnf . . . . . . . . . . . . . . 256, 257
alias . . . . . . . . . . . . . . . . . . 50, 87 bestapprPade . . . . . . . . . . . . 172, 173
alias0 . . . . . . . . . . . . . . . . . . . 88 Bezout relation . . . . . . . . . . . . . . . 200
allocatemem . . . . . . . . . . . . . . 88, 109 bezout . . . . . . . . . . . . . . . . . . . 173
alternating series . . . . . . . . . . . . . . 333 bezoutres . . . . . . . . . . . . . . . . . 231
and . . . . . . . . . . . . . . . . . . . . . 130 bid . . . . . . . . . . . . . . . . . . . . 47, 349
and . . . . . . . . . . . . . . . . . . . . . 146 bid . . . . . . . . . . . . . . . . . . . . . 350
apply . . . . . . . . . . . . . . . . 10, 89, 90 bigomega . . . . . . . . . . . . . . . . . . 173
area . . . . . . . . . . . . . . . . . . 470, 471 bilhell . . . . . . . . . . . . . . . . . . . 476
arg . . . . . . . . . . . . . . . . . . . . . 296 binaire . . . . . . . . . . . . . . . . . . . 146
arity . . . . . . . . . . . . . . . . . . . . 90 binary file . . . . . . . . . . . . . . . . . . 117
arity0 . . . . . . . . . . . . . . . . . . . 90 binary file . . . . . . . . . . . . . . 61, 109
Artin L-function . . . . . . . . . . . . . . 370 binary flag . . . . . . . . . . . . . . . . . 68
Artin root number . . . . . . . . . . . . . 370 binary quadratic form . . . . . . . . . 23, 142
asin . . . . . . . . . . . . . . . . . . . . . 296 binary . . . . . . . . . . . . . . . . . . . 145
asinh . . . . . . . . . . . . . . . . . . . . 296 binomial coefficient . . . . . . . . . . . . . 165
asympnum . . . . . . . . . . . . . . . 313, 315 binomial . . . . . . . . . . . . . . . . . . 165
asympnum0 . . . . . . . . . . . . . . . . . 315 binomial0 . . . . . . . . . . . . . . . . . 165
asympnumraw . . . . . . . . . . . . . . . . 315 Birch and Swinnerton-Dyer conjecture . . 482
asympnumraw0 . . . . . . . . . . . . . . . 315 bitand . . . . . . . . . . . . . . . . . . . 146
atan . . . . . . . . . . . . . . . . . . . . . 296 bitneg . . . . . . . . . . . . . . . . . . . 146
atanh . . . . . . . . . . . . . . . . . . . . 296 bitnegimply . . . . . . . . . . . . . . . . 146
automatic simplification . . . . . . . . . . 129 bitor . . . . . . . . . . . . . . . . . . . . 147
available commands . . . . . . . . . . . . 60 bitprecision . . . . . . . . . . . . . . . 147
bitprecision00 . . . . . . . . . . . . . . 147
B bittest . . . . . . . . . . . . . . . . 147, 148
bitwise and . . . . . . . . . . . . . . . . . 146
b2 . . . . . . . . . . . . . . . . . . . . . . 469 bitwise exclusive or . . . . . . . . . . . . 148
b4 . . . . . . . . . . . . . . . . . . . . . . 469 bitwise inclusive or . . . . . . . . . . . . . 147
b6 . . . . . . . . . . . . . . . . . . . . . . 469 bitwise negation . . . . . . . . . . . . . . 146
b8 . . . . . . . . . . . . . . . . . . . . . . 469 bitxor . . . . . . . . . . . . . . . . . . . 148
backslash character . . . . . . . . . . . . 16 bnf . . . . . . . . . . . . . . . . . . . . 47, 347
basistoalg . . . . . . . . . . . . . . . . . 398 bnf . . . . . . . . . . . . . . . . . . 350, 471
bernfrac . . . . . . . . . . . . . . . 163, 164 bnfcertify . . . . . . . . . . . . . . 352, 353
Bernoulli numbers . . . . . . . 163, 164, 309 bnfcertify0 . . . . . . . . . . . . . . . . 353
Bernoulli polynomial . . . . . . . . . . . . 164 bnfdecodemodule . . . . . . . . . . 353, 366
bernpol . . . . . . . . . . . . . . . . . . . 164 bnfinit . . . . . . . . . . . 214, 347, 353, 392
bernreal . . . . . . . . . . . . . . . . . . 164 bnfinit0 . . . . . . . . . . . . . . . . . . 355

613
bnfisintnorm . . . . . . . . . . . . . . . 355 break loop . . . . . . . . . . . . . . . . . . 53
bnfisintnormabs . . . . . . . . . . . . . 355 break . . . . . . . . . . . . . . . . . . 53, 69
bnfisnorm . . . . . . . . . . . . . . . . . 355 breakloop . . . . . . . . . . . . . 55, 88, 121
bnfisprincipal . . . 214, 354, 355, 368, 392 breakpoint . . . . . . . . . . . . . . . . . 69
bnfisprincipal0 . . . . . . . . . . . . . 356 Buchall . . . . . . . . . . . . . . . . . . . 355
bnfissunit . . . . . . . . . . . . . . 356, 357 Buchall_param . . . . . . . . . . . . . . . 355
bnfisunit . . . . . . . . . . . . . . 357, 358 Buchmann . . . . . . . . . . . . . . 351, 353
bnfisunit0 . . . . . . . . . . . . . . . . . 358 Buchmann-McCurley . . . . . . . . . . . . 214
bnflog . . . . . . . . . . . . . . . . . . . 358 Buchquad . . . . . . . . . . . . . . . . . . 215
bnflogdegree . . . . . . . . . . . . 358, 359 Buchray . . . . . . . . . . . . . . . . . . . 367
bnflogef . . . . . . . . . . . . . . . 358, 359 Buchraymod . . . . . . . . . . . . . . . . . 367
bnfnarrow . . . . . . . . . . . . . . 215, 359
bnfnewprec . . . . . . . . . . . . . . . . . 415 C
bnfsignunit . . . . . . . . . . . . . . . . 359
bnfsunit . . . . . . . . . . . . . . . . . . 360 c4 . . . . . . . . . . . . . . . . . . . . . . 469
bnfunits . . . . . . . . . . . . . . . 360, 361 c6 . . . . . . . . . . . . . . . . . . . . . . 469
bnr . . . . . . . . . . . . . . . . . . . . 47, 347 call by reference . . . . . . . . . . . . . . 39
bnrautmatrix . . . . . . . . . . . . . . . 367 call by value . . . . . . . . . . . . . . . . 38
bnrchar . . . . . . . . . . . . . . . . . . . 362 call . . . . . . . . . . . . . . . . . . . . . 90
bnrclassfield . . . . . . . 362, 363, 370, 440 call0 . . . . . . . . . . . . . . . . . . . . 91
bnrclassno . . . . . . . . . . . . . . 363, 366 caract . . . . . . . . . . . . . . . . . . . 258
bnrclassno0 . . . . . . . . . . . . . . . . 363 caradj . . . . . . . . . . . . . . . . . . . 258
bnrclassnolist . . . . . . . . 363, 364, 387 carberkowitz . . . . . . . . . . . . . . . 258
bnrconductor . . . . . . . . . . . . . . . . 365 carhess . . . . . . . . . . . . . . . . . . . 258
bnrconductor . . . . . . . . . 364, 365, 369 Catalan . . . . . . . . . . . . . . . . . . . 295
bnrconductor0 . . . . . . . . . . . . . . . 365 ceil . . . . . . . . . . . . . . . . . . . . . 148
bnrconductormod . . . . . . . . . . . . . 365 centerlift . . . . . . . . . . . 141, 148, 153
bnrconductorofchar . . . . . . . . . . . 365 character string . . . . . . . . . . . . . . . 27
bnrdisc . . . . . . . . . . . . . . . . 365, 366 character . . . . . . . . . . 170, 348, 518, 519
bnrdisc0 . . . . . . . . . . . . . . . . . . 365 character . . . . . . . . . . . . . . . 361, 370
bnrdisclist . . . . . . . . . . . . . 365, 387 characteristic polynomial . . . . . . . . . 257
bnrdisclist0 . . . . . . . . . . . . . . . 366 characteristic . . . . . . . . . . . . . . 148
bnrgaloisapply . . . . . . . . . . . . . . 366 charconj . . . . . . . . . . . . . . . 173, 174
bnrgaloismatrix . . . . . . . . . . 366, 367 charconj0 . . . . . . . . . . . . . . . . . 174
bnrinit . . . . . . . . . . . . . 350, 364, 367 chardiv . . . . . . . . . . . . . . . . 174, 175
bnrinitmod . . . . . . . . . . . . . . . . . 367 chardiv0 . . . . . . . . . . . . . . . . . . 175
bnrisconductor . . . . . . . . . . . . . . 368 chareval . . . . . . . . . . . . . . . . . . 175
bnrisconductor0 . . . . . . . . . . . . . 368 chargalois . . . . . . . . . . . . . . 175, 176
bnrisgalois . . . . . . . . . . . . . . . . 368 charker . . . . . . . . . . . . . . . . . . . 176
bnrisprincipal . . . . . . . . 355, 368, 369 charker0 . . . . . . . . . . . . . . . . . . 176
bnrisprincipalmod . . . . . . . . . . . . 369 charmul . . . . . . . . . . . . . . . . . . . 177
bnrL1 . . . . . . . . . . . . . . . . . 361, 362 charmul0 . . . . . . . . . . . . . . . . . . 177
bnrmap . . . . . . . . . . . . . . . . 369, 370 charorder . . . . . . . . . . . . . . 177, 178
bnrnewprec . . . . . . . . . . . . . . . . . 415 charorder0 . . . . . . . . . . . . . . . . . 178
bnrrootnumber . . . . . . . . . . . . . . . 370 charpoly . . . . . . . . . . . . . . . 257, 258
bnrstark . . . . . . . . . . 216, 370, 371, 440 charpoly0 . . . . . . . . . . . . . . . . . 258
Boolean operators . . . . . . . . . . . . . 130 charpow . . . . . . . . . . . . . . . . . . . 178
boundfact . . . . . . . . . . . . . . . . . 190 charpow0 . . . . . . . . . . . . . . . . . . 178
brace characters . . . . . . . . . . . . . . 16 Chebyshev . . . . . . . . . . . . . . . . . 237

614
chinese . . . . . . . . . . . . . . . . . . . 179 cotanh . . . . . . . . . . . . . . . . . . . 297
chinese1 . . . . . . . . . . . . . . . . . . 179 CPU time . . . . . . . . . . . . . . . . . . 130
clgp . . . . . . . . . . . . . . . . . . . . . 350 cyc . . . . . . . . . . . . . . . . . . 351, 471
CLISP . . . . . . . . . . . . . . . . . . . . 57
cmdtool . . . . . . . . . . . . . . . . . . . 127 D
cmp . . . . . . . . . . 101, 131, 136, 144, 288
datadir . . . . . . . . . . . . . . . . . . . 121
cmp_universal . . . . . . . . . . . . . . . 137
dbg_down . . . . . . . . . . . . . . 54, 69, 71
code words . . . . . . . . . . . . . . . . . 148
codiff . . . . . . . . . . . . . . . . . . . 351 dbg_err . . . . . . . . . . . . . . . . . 55, 70
Col . . . . . . . . . . . . . . . . . . . 24, 139 dbg_up . . . . . . . . . . . . . . . . . 54, 71
colors . . . . . . . . . . . . . . . . . . . 121 dbg_x . . . . . . . . . . . . . . . . 55, 61, 71
Colrev . . . . . . . . . . . . . . . . . 24, 139 debug . . . . . . . . . . . . . . . 60, 122, 191
column vector . . . . . . . . . . . . . . 7, 23 debugfiles . . . . . . . . . . . . . . . 60, 122
comparison operators . . . . . . . . . . . 131 debugmem . . . . . . . . . . . . . . . . 60, 122
compatible . . . . . . . . . . . . . . . . . 121 decodemodule . . . . . . . . . . . . . . . 353
completion . . . . . . . . . . . . . . . . . 64 decomposition into squares . . . . . . . . 280
complex number . . . . . . . . . . . . 7, 8, 20 Dedekind sum . . . . . . . . . . . . . . . 219
compo . . . . . . . . . . . . . . . . . . . . 149 Dedekind . . . . . . . . . . . . 298, 371, 425
component . . . . . . . . . . . . . . . 82, 148 deep recursion . . . . . . . . . . . . . . . 45
composition . . . . . . . . . . . . . . . . . 212 def,factor_add_primes . . . . . . . . . 122
compositum . . . . . . . . . . . . . . 398, 419 def,factor_proven . . . . . . . . . . . . 122
compositum . . . . . . . . . . . . . . . . . 420 def,new_galois_format . . . . . . . . . 125
compositum2 . . . . . . . . . . . . . . . . 420 def,prompt_cont . . . . . . . . . . . . . 127
compress . . . . . . . . . . . . . . . . . . 61 default precision . . . . . . . . . . . . . . . 9
concat . . . . . . . . . . . . . . . . . 49, 258 default . . . . . . . . . . . . . . . 50, 91, 120
conj . . . . . . . . . . . . . . . . . . . . . 149 default0 . . . . . . . . . . . . . . . . . . 91
conjvec . . . . . . . . . . . . . . . . . . . 149 defaults . . . . . . . . . . . . . . . . . 57, 60
Conrey character . . . . . . . . . . . . . . 171 denom . . . . . . . . . . . . . . . . . . . . 150
Conrey generators . . . . . . . . . . . . . 170 denominator . . . . . . . . . . . 35, 149, 150
Conrey logarithm . . . . . . . . . . . . . . 171 deriv . . . . . . . . . . . . . . . . . 231, 232
content . . . . . . . . . . . . . . 35, 179, 200 derivfun . . . . . . . . . . . . . . . . . . 317
content0 . . . . . . . . . . . . . . . . . . 180 derivfunk . . . . . . . . . . . . . . . . . 317
contfrac . . . . . . . . . . . . . . . . . . 180 derivn . . . . . . . . . . . . . . . . 232, 233
contfrac0 . . . . . . . . . . . . . . . . . 181 derivnum . . . . . . . . . . . . 316, 317, 327
contfraceval . . . . . . . . . . . . . . . 316 derivnumk . . . . . . . . . . . . . . . . . 317
contfracinit . . . . . . . . . . . . . . . 316 det . . . . . . . . . . . . . . . . . . . . . 263
contfracpnqn . . . . . . . . . . . . 181, 182 det0 . . . . . . . . . . . . . . . . . . . . . 263
continued fraction . . . . . . . . . . . . . 180 det2 . . . . . . . . . . . . . . . . . . . . . 263
convol . . . . . . . . . . . . . . . . . . . 251 detint . . . . . . . . . . . . . . . . . . . 263
Coppersmith . . . . . . . . . . . . . . . . 227 diagonal . . . . . . . . . . . . . . . . . . 264
core . . . . . . . . . . . . . . . . . . . . . 182 diff . . . . . . . . . . . . . . . . . . . . . 351
core0 . . . . . . . . . . . . . . . . . . . . 182 difference . . . . . . . . . . . . . . . . . . 132
core2 . . . . . . . . . . . . . . . . . . . . 182 diffop . . . . . . . . . . . . . . . . . . . 233
coredisc . . . . . . . . . . . . . . . . . . 182 diffop0 . . . . . . . . . . . . . . . . . . . 233
coredisc0 . . . . . . . . . . . . . . . . . 182 digits . . . . . . . . . . . . . . . . . . . 150
coredisc2 . . . . . . . . . . . . . . . . . 182 dilog . . . . . . . . . . . . . . . . . . . . 297
cos . . . . . . . . . . . . . . . . . . . . . 297 dirdiv . . . . . . . . . . . . . . . . . . . 182
cosh . . . . . . . . . . . . . . . . . . . . . 297 direuler . . . . . . . . . . . . . . . 182, 183
cotan . . . . . . . . . . . . . . . . . . . . 297 Dirichlet series . . . . . . . . . 182, 183, 371

615
dirmul . . . . . . . . . . . . . . . . . . . 183 ellfromeqn . . . . . . . . . . . . . . . . . 481
dirpowers . . . . . . . . . . . . . . . . . 259 ellfromj . . . . . . . . . . . . . . . . . . 481
dirpowerssum . . . . . . . . . . . . . . . 183 ellgenerators . . . . . . . . . 471, 481, 482
dirpowerssum0 . . . . . . . . . . . . . . . 183 ellglobalred . . . . . . . . . . . . . . . 482
dirzetak . . . . . . . . . . . . . . . . . . 371 ellgroup . . . . . . . . . . . . . . . 482, 483
disc . . . . . . . . . . . . . . . . . . 351, 469 ellgroup0 . . . . . . . . . . . . . . . . . 483
divisors . . . . . . . . . . . . . . . . 72, 184 ellheegner . . . . . . . . . . . . . . 483, 484
divisors0 . . . . . . . . . . . . . . . . . 184 ellheight . . . . . . . . . . . . . . . . . 484
divisorslenstra . . . . . . . . . . 184, 185 ellheight0 . . . . . . . . . . . . . . . . . 484
divisors_factored . . . . . . . . . . . . 184 ellheightmatrix . . . . . . . . . . . . . 484
divrem . . . . . . . . . . . . . . . . . 35, 137 ellidentify . . . . . . . . . . . . . 471, 484
dvi . . . . . . . . . . . . . . . . . . . . . . 65 ellinit . . . . . 469, 471, 481, 484, 485, 486
dynamic scoping . . . . . . . . . . . . . . 36 ellintegralmodel . . . . . . . . . . 482, 486
ellisdivisible . . . . . . . . . . . 486, 487
E ellisogeny . . . . . . . . . . . . . . . . . 487
ellisogenyapply . . . . . . . . . . 487, 488
echo . . . . . . . . . . . . . . . . . . . 60, 122 ellisomat . . . . . . . . . . . . . . . . . 488
ECM . . . . . . . . . . . . . . . . . . 169, 191 ellisoncurve . . . . . . . . . . . . 488, 489
editing characters . . . . . . . . . . . . . 16 ellisotree . . . . . . . . . . . . . . 489, 490
Egyptian fraction . . . . . . . . . . . . . . 181 ellissupersingular . . . . . . . . . . . 490
eigen . . . . . . . . . . . . . . . . . . . . 265 ellj . . . . . . . . . . . . . . . . . . . . . 490
eint1 . . . . . . . . . . . . . . . . . 297, 298 elljissupersingular . . . . . . . . . . . 490
elementary divisors . . . . . . . . . . . . . 274 ellK . . . . . . . . . . . . . . . . . . . . . 298
ell . . . . . . . . . . . . . . . . . 47, 469, 485 ellL1 . . . . . . . . . . . . . . . . . . . . 472
ell . . . . . . . . . . . . . . . . . . . . . 484 ellL1_bitprec . . . . . . . . . . . . . . . 473
elladd . . . . . . . . . . . . . . . . . . . 473 elllocalred . . . . . . . . . . . . . . . . 490
ellak . . . . . . . . . . . . . . . . . . . . 473 elllog . . . . . . . . . . . . . . . . 490, 491
ellan . . . . . . . . . . . . . . . . . . . . 474 elllseries . . . . . . . . . . . . . . . . . 491
ellanalyticrank . . . . . . . 473, 474, 497 ellminimaldisc . . . . . . . . . . . . . . 491
ellanalyticrank_bitprec . . . . . . . . 474 ellminimalmodel . . . . . . . 482, 491, 492
ellanQ_zv . . . . . . . . . . . . . . . . . 474 ellminimaltwist . . . . . . . . . . . . . 492
ellap . . . . . . . . . . . . . . 474, 476, 477 ellminimaltwist0 . . . . . . . . . . . . . 492
ellbil . . . . . . . . . . . . . . . . . . . 476 ellminimaltwistcond . . . . . . . . . . . 492
ellbsd . . . . . . . . . . . . . . . . 476, 497 ellmoddegree . . . . . . . . . . . . 492, 493
ellcard . . . . . . . . . . . . . . . . 476, 477 ellmodulareqn . . . . . . . . . . . . 493, 494
ellchangecurve . . . . . . . . . . . . . . 477 ellmul . . . . . . . . . . . . . . . . 494, 505
ellchangepoint . . . . . . . . . . . . . . 477 ellneg . . . . . . . . . . . . . . . . . . . 494
ellchangepointinv . . . . . . . . . 477, 478 ellnonsingularmultiple . . . . . . . . . 494
ellconvertname . . . . . . . . . . . 478, 506 ellorder . . . . . . . . . . . . . . . 494, 495
elldata . . . 73, 471, 478, 481, 484, 485, 506 ellordinate . . . . . . . . . . . . . 156, 495
elldivpol . . . . . . . . . . . . . . . . . 478 ellpadicbsd . . . . . . . . . . 497, 498, 500
ellE . . . . . . . . . . . . . . . . . . . . . 298 ellpadicfrobenius . . . . . . . . . 498, 499
elleisnum . . . . . . . . . . . . . . 478, 479 ellpadicheight . . . . . . . . . . . 499, 502
elleta . . . . . . . . . . . . . . . . 479, 511 ellpadicheight0 . . . . . . . . . . . . . 500
ellformaldifferential . . . . . . . . . 479 ellpadicheightmatrix . . . . . . . 500, 501
ellformalexp . . . . . . . . . . . . 479, 480 ellpadicL . . . . . . . . . . . 495, 497, 498
ellformallog . . . . . . . . . . . . 479, 480 ellpadiclambdamu . . . . . . . . . . . . . 501
ellformalpoint . . . . . . . . . . . . . . 480 ellpadiclog . . . . . . . . . . . . . 501, 502
ellformalw . . . . . . . . . . . . . . 480, 481 ellpadicregulator . . . . . . . . . 497, 502

616
ellpadics2 . . . . . . . . . . . . . . 502, 503 eulerianpol . . . . . . . . . . . . . . . . 165
ellperiods . . . 469, 478, 503, 507, 509, 511 eulerphi . . . . . . . . . . . . . . . . . . 185
ellpointtoz . . . . . . . . . . . . . . . . 503 eulerpol . . . . . . . . . . . . . . . . . . 165
ellpow . . . . . . . . . . . . . . . . . . . 505 eulervec . . . . . . . . . . . . . . . . . . 166
ellrandom . . . . . . . . . . . . . . . . . 157 eval . . . . . . . . . . . . . . . 36, 38, 50, 233
ellratpoints . . . . . . . . . . . . . . . 505 exp . . . . . . . . . . . . . . . . . . . . . 299
ellrootno . . . . . . . . . . . . . . . . . 505 expm1 . . . . . . . . . . . . . . . . . . . . 299
ellsea . . . . . . . . . . . . . . . . 505, 506 exponent . . . . . . . . . . . . . . . . . . 150
ellsearch . . . . . . . . . . . 471, 506, 507 export . . . . . . . . . . . . . . . . . . . 91
ellsearchcurve . . . . . . . . . . . . . . 507 exportall . . . . . . . . . . . . . . . . . 92
ellsigma . . . . . . . . . . . . . . . . . . 507 expression sequence . . . . . . . . . . . . 15
ellsub . . . . . . . . . . . . . . . . . . . 507 expression . . . . . . . . . . . . . . . . . . 15
elltamagawa . . . . . . . . . . . . . . . . 507 extended gcd . . . . . . . . . . . . . . . . 200
elltaniyama . . . . . . . . . . . . . 507, 508 extern . . . . . . . . . . . . . . . 50, 92, 128
elltatepairing . . . . . . . . . . . . . . 508 external prettyprint . . . . . . . . . . . . . 125
elltors . . . . . . . . . . . . . . . . . . . 508 externstr . . . . . . . . . . . . . . . . . 92
elltwist . . . . . . . . . . . . . . . . . . 508 extract0 . . . . . . . . . . . . . . . . . . 290
ellweilcurve . . . . . . . . . . . . 508, 509
ellweilpairing . . . . . . . . . . . . . . 509 F
ellwp . . . . . . . . . . . . . . . . . 509, 510
ellwp0 . . . . . . . . . . . . . . . . . . . 510 factmod . . . . . . . . . . . . . . . . . . . 191
ellwpseries . . . . . . . . . . . . . . . . 510 factor . . . . . . . . . . . . . 184, 185, 190
ellxn . . . . . . . . . . . . . . . . . 510, 511 factor0 . . . . . . . . . . . . . . . . . . . 190
ellzeta . . . . . . . . . . . . . . . . 470, 511 factorback . . . . . . . . . . . . . . 190, 191
ellztopoint . . . . . . . . . . . . . . . . 511 factorback2 . . . . . . . . . . . . . . . . 191
Emacs . . . . . . . . . . . . . . . . . . . . 65 factorcantor . . . . . . . . . . . . . . . 191
Engel expansion . . . . . . . . . . . . . . 181 factorff . . . . . . . . . . . . . . . 187, 191
environment expansion . . . . . . . . . . . 112 factorial . . . . . . . . . . . . . . . . . 191
environment expansion . . . . . . . . . 58 factorint . . . . . . . . . . . . . . 185, 191
environment variable . . . . . . . . . . . . 112 factormod . . . . . . . . . . . . . . 187, 191
erfc . . . . . . . . . . . . . . . . . . . . . 298 factormod0 . . . . . . . . . . . . . . . . . 193
errname . . . . . . . . . . . . . . . . . . . 91 factormodDDF . . . . . . . . . . . . . . . 193
error recovery . . . . . . . . . . . . . . . . 52 factormodSQF . . . . . . . . . . . . . . . 194
error trapping . . . . . . . . . . . . . . . . 53 factornf . . . . . . . . . . . . . . . . . . 371
error . . . . . . . . . . . . . . 50, 53, 84, 91 factorpadic . . . . . . . . . . . . . 234, 235
error(E) . . . . . . . . . . . . . . . . . . 82 factor_add_primes . . . . . . . . . . . . 87
eta . . . . . . . . . . . . . 298, 331, 470, 471 factor_proven . . . . . . . 171, 186, 187, 191
eta0 . . . . . . . . . . . . . . . . . . . . . 299 famat . . . . . . . . . . . . . . . . . . . . 347
Euclid . . . . . . . . . . . . . . . . . . . . 200 ff . . . . . . . . . . . . . . . . . . . . . . . 47
Euclidean quotient . . . . . . . . . . 132, 133 ffcompomap . . . . . . . . . . . . . . 194, 195
Euclidean remainder . . . . . . . . . . . . 133 ffembed . . . . . . . . . . . . . . . . . . . 195
Euler numbers . . . . . . . . . . . . 165, 166 ffextend . . . . . . . . . . . . . . . . . . 195
Euler polynomial . . . . . . . . . . . . . . 165 fffrobenius . . . . . . . . . . . . . 195, 196
Euler product . . . . . . . . . . . . . 182, 331 ffgen . . . . . . . . . . . . 19, 196, 197, 485
Euler totient function . . . . . . . . 169, 185 ffinit . . . . . . . . . . . . . . 19, 196, 197
Euler . . . . . . . . . . . . . . . . . . . . 295 ffinvmap . . . . . . . . . . . . . . . . . . 197
Euler-Maclaurin . . . . . . . . . . . . . . 309 fflog . . . . . . . . . . . . . . . . . 197, 199
eulerfrac . . . . . . . . . . . . . . . . . 165 ffmap . . . . . . . . . . . . . . . . . 197, 198
Eulerian polynomial . . . . . . . . . . . . 165 ffmaprel . . . . . . . . . . . . . . . . . . 198

617
ffnbirred . . . . . . . . . . . . . . . . . 198 forsubgroup . . . . . . . . . . . . . . . . 79
ffnbirred0 . . . . . . . . . . . . . . . . . 198 forsubset . . . . . . . . . . . . . . . . . 80
fforder . . . . . . . . . . . . . . . . 198, 199 forvec . . . . . . . . . . . . . . . . . . . 80
ffprimroot . . . . . . . . . . . . . . 197, 199 frac . . . . . . . . . . . . . . . . . . . . . 151
ffrandom . . . . . . . . . . . . . . . . . . 157 free variable . . . . . . . . . . . . . . . . . 33
ffsumnbirred . . . . . . . . . . . . . . . 198 fromdigits . . . . . . . . . . . . . . . . . 151
fft . . . . . . . . . . . . . . . . . . . . . 235 fu . . . . . . . . . . . . . . . . . . . . . . 351
FFT . . . . . . . . . . . . . . . . . . . . . 235 fundamental units . . . . . . . 216, 351, 353
fftinv . . . . . . . . . . . . . . . . . . . 235
FFTinv . . . . . . . . . . . . . . . . . . . 236 G
fibo . . . . . . . . . . . . . . . . . . . . . 166
fibonacci . . . . . . . . . . . . . . . . . 166 gabs . . . . . . . . . . . . . . . . . . . . . 295
field discriminant . . . . . . . . . . . . . . 399 gacos . . . . . . . . . . . . . . . . . . . . 295
fileclose . . . . . . . . . . . . . . . . . 92 gacosh . . . . . . . . . . . . . . . . . . . 295
fileextern . . . . . . . . . . . . . . . . . 93 gadd . . . . . . . . . . . . . . . . . . . . . 132
fileflush . . . . . . . . . . . . . . . . . 94 galdata . . . . . . . . . . . . . . . . . . . 420
filename . . . . . . . . . . . . . . . . . . . 58 galois . . . . . . . . . . . . . . . . . . . . 47
fileopen . . . . . . . . . . . . . . . . . . 94 Galois . . . . 79, 355, 405, 406, 418, 420, 436
fileread . . . . . . . . . . . . . . . . . . 95 galoisapply . . . . . . . . . . . . . . . . 406
filereadstr . . . . . . . . . . . . . . . . 95 galoischardet . . . . . . . . . . . . 371, 372
filewrite . . . . . . . . . . . . . . . . . 95 galoischarpoly . . . . . . . . . . . . . . 372
filewrite1 . . . . . . . . . . . . . . . . . 96 galoischartable . . . . . . . . . . 372, 373
finite field element . . . . . . . . . . . 7, 8, 19 galoisconj . . . . . . . . . . . . . . . . . 407
finite field . . . . . . . . . . . . . . . . . . 21 galoisconj0 . . . . . . . . . . . . . . . . 407
fixed floating point format . . . . . . . . 123 galoisconjclasses . . . . . . . . . 373, 374
flag . . . . . . . . . . . . . . . . . . . . . . 67 galoisexport . . . . . . . . . . . . 374, 376
floor . . . . . . . . . . . . . . . . . . . . 151 galoisfixedfield . . . . . . . . 79, 374, 375
fold . . . . . . . . . . . . . . . . . . . . . 96 galoisgetgroup . . . . . . . . . . . . . . 375
fold0 . . . . . . . . . . . . . . . . . . . . 96 galoisgetname . . . . . . . . . . . . . . . 375
for . . . . . . . . . . . . . . . . . . . . . 71 galoisgetpol . . . . . . . . . . . . . . . 376
forcomposite . . . . . . . . . . . . . . . 71 galoisidentify . . . . . . . . . . . . . . 376
Ford . . . . . . . . . . . . . . . . . . . . . 396 galoisinit . . . . . . 368, 374, 376, 377, 406
fordiv . . . . . . . . . . . . . . . . . . . 72 galoisisabelian . . . . . . . . . . 377, 378
fordivfactored . . . . . . . . . . . . . . 72 galoisisnormal . . . . . . . . . . . . . . 378
foreach . . . . . . . . . . . . . . . . . . . 73 galoisnbpol . . . . . . . . . . . . . 375, 376
forell . . . . . . . . . . . . . . . . . 73, 471 galoispermtopol . . . . . . . . . . . . . 378
forfactored . . . . . . . . . . . . . . . . 73 galoissubcyclo . . . . 79, 250, 370, 378, 379
formal integration . . . . . . . . . . . . . 236 galoissubfields . . . . . . . 374, 379, 418
formal sum . . . . . . . . . . . . . . . . . 252 galoissubgroups . . . . . . . . . . . . . 379
format . . . . . . . . . . . . . 123, 127, 128 gamma . . . . . . . . . . . . . . . . . . . . 299
forpart . . . . . . . . . . . . . . . . . 74, 75 gamma-function . . . . . . . . . . . . . . 299
forperm . . . . . . . . . . . . . . . . . . . 75 gammah . . . . . . . . . . . . . . . . . . . 300
forprime . . . . . . . . . . . . . . . . . . 76 gammamellininv . . . . . . . . . . . 300, 525
forprimestep . . . . . . . . . . . . . . . 76 gammamellininvasymp . . . . . . . . . . . 300
forqfvec . . . . . . . . . . . . . . . 259, 260 gammamellininvinit . . . . . . . . 300, 301
forqfvec0 . . . . . . . . . . . . . . . . . 260 garg . . . . . . . . . . . . . . . . . . . . . 296
forqfvec1 . . . . . . . . . . . . . . . . . 260 gasin . . . . . . . . . . . . . . . . . . . . 296
forsquarefree . . . . . . . . . . . . . . . 77 gasinh . . . . . . . . . . . . . . . . . . . 296
forstep . . . . . . . . . . . . . . . . . . . 78 gatan . . . . . . . . . . . . . . . . . . . . 296

618
gatanh . . . . . . . . . . . . . . . . . . . 296 getrand . . . . . . . . . . . . . . . . . 97, 156
gauss . . . . . . . . . . . . . . . . . . . . 275 getstack . . . . . . . . . . . . . . . . . . 97
gaussmodulo . . . . . . . . . . . . . . . . 276 gettime . . . . . . . . . . . . . . . . . . . 97
gaussmodulo2 . . . . . . . . . . . . . . . 276 getwalltime . . . . . . . . . . . . . . . . 97
gbitand . . . . . . . . . . . . . . . . . . . 146 geval . . . . . . . . . . . . . . . . . . . . 234
gbitneg . . . . . . . . . . . . . . . . . . . 146 gexp . . . . . . . . . . . . . . . . . . . . . 299
gbitnegimply . . . . . . . . . . . . . . . 146 gexpm1 . . . . . . . . . . . . . . . . . . . 299
gbitor . . . . . . . . . . . . . . . . . . . 147 gfloor . . . . . . . . . . . . . . . . . . . 151
gbittest . . . . . . . . . . . . . . . . . . 148 gfrac . . . . . . . . . . . . . . . . . . . . 151
gbitxor . . . . . . . . . . . . . . . . . . . 148 ggamma . . . . . . . . . . . . . . . . . . . 299
gboundcf . . . . . . . . . . . . . . . . . . 181 ggammah . . . . . . . . . . . . . . . . . . . 300
gcd . . . . . . . . . . . . . . . . . . . . . 199 ggcd . . . . . . . . . . . . . . . . . . . . . 200
gcdext . . . . . . . . . . . . . . . . 199, 200 ggcd0 . . . . . . . . . . . . . . . . . . . . 200
gcdext0 . . . . . . . . . . . . . . . . 173, 201 ggrando . . . . . . . . . . . . . . . . . . . 231
gceil . . . . . . . . . . . . . . . . . . . . 148 ghalfgcd . . . . . . . . . . . . . . . . . . 201
gcf . . . . . . . . . . . . . . . . . . . . . 181 gimag . . . . . . . . . . . . . . . . . . . . 151
gcf2 . . . . . . . . . . . . . . . . . . . . . 181 gisanypower . . . . . . . . . . . . . . . . 202
gconcat . . . . . . . . . . . . . . . . . . . 259 gisprime . . . . . . . . . . . . . . . . . . 203
gconcat1 . . . . . . . . . . . . . . . . . . 259 gispseudoprime . . . . . . . . . . . . . . 203
gconj . . . . . . . . . . . . . . . . . . . . 149 gissquare . . . . . . . . . . . . . . . . . 204
gcos . . . . . . . . . . . . . . . . . . . . . 297 gissquareall . . . . . . . . . . . . . . . 204
gcosh . . . . . . . . . . . . . . . . . . . . 297 glambertW . . . . . . . . . . . . . . . . . 303
gcotan . . . . . . . . . . . . . . . . . . . 297 glcm0 . . . . . . . . . . . . . . . . . . . . 206
gcotanh . . . . . . . . . . . . . . . . . . . 297 glength . . . . . . . . . . . . . . . . . . . 152
gcvtoi . . . . . . . . . . . . . . . . . . . 159 glngamma . . . . . . . . . . . . . . . . . . 303
gdeflate . . . . . . . . . . . . . . . . . . 252 global . . . . . . . . . . . . . . . . . . . 97
gdiv . . . . . . . . . . . . . . . . . . . . . 132 glog . . . . . . . . . . . . . . . . . . . . . 304
gdivent . . . . . . . . . . . . . . . . . . . 133 glog1p . . . . . . . . . . . . . . . . . . . 304
gdiventres . . . . . . . . . . . . . . . . . 137 gmax . . . . . . . . . . . . . . . . . . . . . 138
gdivround . . . . . . . . . . . . . . . . . 133 gmin . . . . . . . . . . . . . . . . . . . . . 138
gen (member function) . . . . . . . . . . 351 gmod . . . . . . . . . . . . . . . . . . . . . 133
GEN . . . . . . . . . . . . . . . . . . . . . . 8 gmodulo . . . . . . . . . . . . . . . . . . . 141
gen . . . . . . . . . . . . . . . . . . . . . 471 gmul . . . . . . . . . . . . . . . . . . . . . 132
genapply . . . . . . . . . . . . . . . . . . 90 gmul2n . . . . . . . . . . . . . . . . . . . 138
generic matrix . . . . . . . . . . . . . . . 50 gneg . . . . . . . . . . . . . . . . . . . . . 132
genfold . . . . . . . . . . . . . . . . . . . 96 gnorm . . . . . . . . . . . . . . . . . . . . 154
genindexselect . . . . . . . . . . . . . . 111 gnorml2 . . . . . . . . . . . . . . . . . . . 277
genrand . . . . . . . . . . . . . . . . . . . 156 gnormlp . . . . . . . . . . . . . . . . . . . 277
genselect . . . . . . . . . . . . . . . . . 111 gp . . . . . . . . . . . . . . . . . . . . . . . 5
GENtostr . . . . . . . . . . . . . . . . . . 144 GP . . . . . . . . . . . . . . . . . . . . . . . 5
genus2red . . . . . . . . . . . . . . 512, 514 gp . . . . . . . . . . . . . . . . . . . . . . 13
gen_I . . . . . . . . . . . . . . . . . . . . 295 gp2c . . . . . . . . . . . . . . . . . . . . . . 5
gerfc . . . . . . . . . . . . . . . . . . . . 298 gpexponent . . . . . . . . . . . . . . . . . 151
getabstime . . . . . . . . . . . . . . . 96, 97 gpextern . . . . . . . . . . . . . . . . . . 92
getcache . . . . . . . . . . . . . . . 536, 537 gphelp . . . . . . . . . . . . . . . . . . . 60
getenv . . . . . . . . . . . . . . . . . . . 97 gpidealfactor . . . . . . . . . . . . . . . 382
getheap . . . . . . . . . . . . . . . . . . . 97 gpidealval . . . . . . . . . . . . . . . . . 394
getlocalbitprec . . . . . . . . . . . . . 97 gpinstall . . . . . . . . . . . . . . . . . 99
getlocalprec . . . . . . . . . . . . . . . 97 gpnfvalrem . . . . . . . . . . . . . . . . . 404

619
gpolvar . . . . . . . . . . . . . . . . . . . 161 gsqrt . . . . . . . . . . . . . . . . . . . . 307
gpolylog . . . . . . . . . . . . . . . . . . 305 gsqrtn . . . . . . . . . . . . . . . . . . . 308
gpow . . . . . . . . . . . . . . . . . . . . . 136 gsub . . . . . . . . . . . . . . . . . . . . . 132
gpowers . . . . . . . . . . . . . . . . . . . 278 gsubst . . . . . . . . . . . . . . . . . . . 252
gpowers0 . . . . . . . . . . . . . . . . . . 278 gsubstpol . . . . . . . . . . . . . . . . . 252
gppadicprec . . . . . . . . . . . . . . . . 155 gsubstvec . . . . . . . . . . . . . . . . . 252
gppoldegree . . . . . . . . . . . . . . . . 240 gtan . . . . . . . . . . . . . . . . . . . . . 308
gprc . . . . . . . . . . . . . . . . . 13, 34, 62 gtanh . . . . . . . . . . . . . . . . . . . . 308
GPRC . . . . . . . . . . . . . . . . . . . . . 63 gtocol . . . . . . . . . . . . . . . . . . . 139
gprc . . . . . . . . . . . . . . . . . . 125, 126 gtocol0 . . . . . . . . . . . . . . . . . . . 139
gprec . . . . . . . . . . . . . . . . . . . . 155 gtocolrev . . . . . . . . . . . . . . . . . 139
gpserprec . . . . . . . . . . . . . . . . . 158 gtocolrev0 . . . . . . . . . . . . . . . . . 139
gpsi . . . . . . . . . . . . . . . . . . . . . 306 gtolist . . . . . . . . . . . . . . . . . . . 140
gpsystem . . . . . . . . . . . . . . . . . . 114 gtomap . . . . . . . . . . . . . . . . . . . 140
gpvaluation . . . . . . . . . . . . . . . . 159 gtomat . . . . . . . . . . . . . . . . . . . 140
gpwritebin . . . . . . . . . . . . . . . . . 117 gtopoly . . . . . . . . . . . . . . . . . . . 142
gp_alarm . . . . . . . . . . . . . . . . . . 87 gtopolyrev . . . . . . . . . . . . . . . . . 142
gp_allocatemem . . . . . . . . . . . . . . 89 gtoset . . . . . . . . . . . . . . . . . . . 144
gp_fileclose . . . . . . . . . . . . . . . 93 gtovec . . . . . . . . . . . . . . . . . . . 145
gp_fileextern . . . . . . . . . . . . . . . 93 gtovec0 . . . . . . . . . . . . . . . . . . . 145
gp_fileflush . . . . . . . . . . . . . . . 94 gtovecrev . . . . . . . . . . . . . . . . . 145
gp_fileflush0 . . . . . . . . . . . . . . . 94 gtovecrev0 . . . . . . . . . . . . . . . . . 145
gp_fileopen . . . . . . . . . . . . . . . . 94 gtovecsmall . . . . . . . . . . . . . . . . 145
gp_fileread . . . . . . . . . . . . . . . . 95 gtovecsmall0 . . . . . . . . . . . . . . . 145
gp_filereadstr . . . . . . . . . . . . . . 95 gtrace . . . . . . . . . . . . . . . . . . . 289
gp_filewrite . . . . . . . . . . . . . . . 96 gtrans . . . . . . . . . . . . . . . . . . . 276
gp_filewrite1 . . . . . . . . . . . . . . . 96 gtrunc . . . . . . . . . . . . . . . . . . . 159
gp_getenv . . . . . . . . . . . . . . . . . 97 gvaluation . . . . . . . . . . . . . . . . . 159
gp_input . . . . . . . . . . . . . . . . . . 98 gvar . . . . . . . . . . . . . . . . . . . . . 161
gp_readvec_file . . . . . . . . . . . . . 110 gzeta . . . . . . . . . . . . . . . . . . . . 310
gp_readvec_stream . . . . . . . . . . . . 110 gzip . . . . . . . . . . . . . . . . . . . 61, 117
gp_read_file . . . . . . . . . . . . . . . 109
Graeffe . . . . . . . . . . . . . . . . . . . 241 H
graphcolormap . . . . . . . . . . . . 123, 594
graphcolors . . . . . . . . . . . . . . . . 123 Hadamard product . . . . . . . . . . . . . 251
greal . . . . . . . . . . . . . . . . . . . . 157 halfgcd . . . . . . . . . . . . . . . . . . . 201
GRH . . . . . . . . . . 254, 351, 352, 355, 436 hammingweight . . . . . . . . . . . . 166, 219
grndtoi . . . . . . . . . . . . . . . . . . . 157 hbessel1 . . . . . . . . . . . . . . . . . . 296
grootsof1 . . . . . . . . . . . . . . . . . 306 hbessel2 . . . . . . . . . . . . . . . . . . 296
ground . . . . . . . . . . . . . . . . . . . 157 hclassno . . . . . . . . . . . . . . . . . . 212
group . . . . . . . . . . . . . . . . . . . . 471 heap . . . . . . . . . . . . . . . . . . . . . 61
gshift . . . . . . . . . . . . . . . . . . . 138 help . . . . . . . . . . . . . . . . . . . . . 123
gsigne . . . . . . . . . . . . . . . . . . . 138 Hermite normal form . 265, 267, 349, 384, 408
gsin . . . . . . . . . . . . . . . . . . . . . 306 Hermite . . . . . . . . . . . . . . . . . . . 241
gsinc . . . . . . . . . . . . . . . . . . . . 306 hess . . . . . . . . . . . . . . . . . . . . . 265
gsinh . . . . . . . . . . . . . . . . . . . . 306 Hilbert class field . . . . . . . . . . . . . 216
gsizebyte . . . . . . . . . . . . . . . . . 159 Hilbert matrix . . . . . . . . . . . . . . . 265
gsizeword . . . . . . . . . . . . . . . . . 159 Hilbert symbol . . . . . . . . . . . . 201, 407
gsqr . . . . . . . . . . . . . . . . . . 132, 307 hilbert . . . . . . . . . . . . . . . . . . . 201

620
histfile . . . . . . . . . . . . . . . . . . 124 ideallist0 . . . . . . . . . . . . . . . . . 387
histsize . . . . . . . . . . . . . . . . 16, 124 ideallistarch . . . . . . . . . . . . 387, 388
hnf . . . . . . . . . . . . . . . . . . . . . 267 ideallog . . . . . . . . . . 229, 347, 388, 393
hnfall . . . . . . . . . . . . . . . . . . . 267 ideallogmod . . . . . . . . . . . . . . . . 388
hnfmod . . . . . . . . . . . . . . . . . . . 267 idealmin . . . . . . . . . . . . . . . . . . 388
hnfmodid . . . . . . . . . . . . . . . . . . 267 idealmul . . . . . . . . . . . . . . . 388, 389
Householder transform . . . . . . . . 267, 272 idealmul0 . . . . . . . . . . . . . . . . . 389
Hurwitz class number . . . . . . . . . . . 212 idealmulred . . . . . . . . . . . . . . . . 389
hyperellcharpoly . . . . . . . . . . . . . 514 idealnorm . . . . . . . . . . . . . . . . . 389
hyperellpadicfrobenius . . . . . . 514, 515 idealnumden . . . . . . . . . . . . . . . . 389
hyperellpadicfrobenius0 . . . . . . . . 515 idealpow . . . . . . . . . . . . . . . 389, 391
hyperellratpoints . . . . . . . . . . . . 515 idealpow0 . . . . . . . . . . . . . . . . . 389
hypergeom . . . . . . . . . . . . . . 301, 302 idealpowred . . . . . . . . . . . . . . . . 389
hyperu . . . . . . . . . . . . . . . . . . . 302 idealpows . . . . . . . . . . . . . . . . . 389
idealprimedec . . . . . . . . . 389, 390, 405
I idealprimedec_limit_f . . . . . . . . . 390
idealprincipalunits . . . . . . . . . . . 390
I . . . . . . . . . . . . . . . . . . . . . 20, 295 idealramgroups . . . . . . . . . . . 390, 391
ibessel . . . . . . . . . . . . . . . . . . . 296 idealred . . . . . . . . . . . . 347, 383, 391
ibitand . . . . . . . . . . . . . . . . . . . 146 idealred0 . . . . . . . . . . . . . . . . . 392
ibitnegimply . . . . . . . . . . . . . . . 146 idealredmodpower . . . . . . . . . . . . . 392
ibitor . . . . . . . . . . . . . . . . . . . 147 idealstar . . . . . . . . . . . 367, 390, 393
ibitxor . . . . . . . . . . . . . . . . . . . 148 Idealstar . . . . . . . . . . . . . . . . . 393
ideal (extended) . . . . . . 347, 385, 388, 389 idealstarmod . . . . . . . . . . . . . . . 393
ideal list . . . . . . . . . . . . . . . . . . . 348 Idealstarmod . . . . . . . . . . . . . . . 393
ideal . . . . . . . . . . . . . . . . . . . . . 347 idealtwoelt . . . . . . . . . . . . . 393, 394
idealadd . . . . . . . . . . . . . . . 379, 380 idealtwoelt0 . . . . . . . . . . . . . . . 394
idealaddtoone . . . . . . . . . . . . . . . 380 idealtwoelt2 . . . . . . . . . . . . . . . 394
idealaddtoone0 . . . . . . . . . . . . . . 380 idealval . . . . . . . . . . . . . . . . . . 394
idealappr . . . . . . . . . . . . . . . . . 380 if . . . . . . . . . . . . . . . . . . . . . . 81
idealappr0 . . . . . . . . . . . . . . . . . 380 iferr . . . . . . . . . 28, 53, 70, 82, 114, 145
idealchinese . . . . . . . . . . . . 380, 381 imag . . . . . . . . . . . . . . . . . . . . . 151
idealchineseinit . . . . . . . . . . . . . 381 image . . . . . . . . . . . . . . . . . . . . 269
idealcoprime . . . . . . . . . . . . . . . 381 imagecompl . . . . . . . . . . . . . . . . . 269
idealdiv . . . . . . . . . . . . . . . . . . 381 incgam . . . . . . . . . . . . . . . . 302, 303
idealdiv0 . . . . . . . . . . . . . . . . . 381 incgam0 . . . . . . . . . . . . . . . . . . . 303
idealdivexact . . . . . . . . . . . . . . . 381 incgamc . . . . . . . . . . . . . . . . . . . 303
idealdown . . . . . . . . . . . . . . . . . 382 inclusive or . . . . . . . . . . . . . . . . . 130
idealfactor . . . . . . . . . . 367, 382, 393 index . . . . . . . . . . . . . . . . . . . . 351
idealfactorback . . . . . . . . . . . . . 383 index . . . . . . . . . . . . . . . . . . . . 351
idealfactor_limit . . . . . . . . . . . . 382 indexrank . . . . . . . . . . . . . . . . . 270
idealfrobenius . . . . . . . . . . . 383, 384 infinite product . . . . . . . . . . . . . . . 332
idealhnf . . . . . . . . . . . . . . . 384, 385 infinity . . . . . . . . . . . . . . . . . . . . 326
idealhnf0 . . . . . . . . . . . . . . . . . 385 inline . . . . . . . . . . . . . . . . . . . 97
idealintersect . . . . . . . . . . . 270, 385 input . . . . . . . . . . . . . . . . . . . . 98
idealinv . . . . . . . . . . . . . . . 385, 409 install . . . . . . . . . . . . . 50, 56, 98, 129
idealismaximal . . . . . . . . . . . 385, 386 intcirc . . . . . . . . . . . . . . . . . . . 317
idealispower . . . . . . . . . . . . . . . 386 integ . . . . . . . . . . . . . . . . . . . . 236
ideallist . . . . . . . . . . . . . . 386, 387 integer . . . . . . . . . . . . . . . . . . 7, 8, 17

621
integral basis . . . . . . . . . . . . . . . . 395 kill . . . . . . . . . . . . . . . . . . . . . 99
integral pseudo-matrix . . . . . . . . . . . 348 kill0 . . . . . . . . . . . . . . . . . . . . 100
internal longword format . . . . . . . . . 61 Kodaira . . . . . . . . . . . . . . . . . . . 490
internal representation . . . . . . . . . . . 61 Kronecker symbol . . . . . . . . . . . . . 205
interpolating polynomial . . . . . . . . . . 241 kronecker . . . . . . . . . . . . . . . . . 205
intersect . . . . . . . . . . . . . . . . . 270
intformal . . . . . . . . . . . . . . . . . 236 L
intfuncinit . . . . . . . . . . 317, 319, 326
Laguerre polynomial . . . . . . . . . . . . 244
intmod . . . . . . . . . . . . . . . . . . . . . 7
lambertw . . . . . . . . . . . . . . . . . . 303
intmod . . . . . . . . . . . . . . . . . . . 8, 18
laplace . . . . . . . . . . . . . . . . . . . 251
intnum 102, 103, 313, 319, 324, 326, 340, 341
intnumgauss . . . . . . . . . . . . . 324, 325 laurentseries . . . . . . . . . . . . . . . 327
intnumgauss0 . . . . . . . . . . . . . . . 324 lcm . . . . . . . . . . . . . . . . . . . . . 205
intnumgaussinit . . . . . . . . . . 324, 325 Ldata . . . . . . . . . . . . . . . . . 517, 526
intnuminit . . . . . . . . . . . 319, 325, 326 Leech lattice . . . . . . . . . . . . . . . . 285
intnumromb . . . . . . . . . . . . . . . . . 326 Legendre polynomial . . . . . . . . . . . . 244
intnumromb_bitprec . . . . . . . . . . . 326 Legendre symbol . . . . . . . . . . . . . . 205
int_bit . . . . . . . . . . . . . . . . . . . 148 length . . . . . . . . . . . . . . . . . . . 152
inverse . . . . . . . . . . . . . . . . . . . . 135 Lenstra . . . . . . . . . . . . . . . . . . . 191
inverseimage . . . . . . . . . . . . . . . 270 lex . . . . . . . . . . . . . . . . . . . . . 137
isdiagonal . . . . . . . . . . . . . . . . . 271 lexcmp . . . . . . . . . . . . . . . . . . . 137
isfundamental . . . . . . . . . . . . . . . 201 lexical scoping . . . . . . . . . . . . . . . 36
isideal . . . . . . . . . . . . . . . . . . . 411 lfun . . . . . . . . . . . . . . . . . . . . . 520
ispolygonal . . . . . . . . . . . . . . . . 201 lfun0 . . . . . . . . . . . . . . . . . . . . 521
ispower . . . . . . . . . . . . . . . . . . . 202 lfunabelianrelinit . . . . . . . . . . . 521
ispowerful . . . . . . . . . . . . . . . . . 202 lfunan . . . . . . . . . . . . . . . . . . . 521
isprime . . . . . . . . . . . . . . . . 202, 203 lfunartin . . . . . . . . . . . 519, 522, 523
isprimepower . . . . . . . . . . . . . . . 203 lfuncheckfeq . . . . . . . . . . . . 523, 524
isprincipalray . . . . . . . . . . . . . . 369 lfunconductor . . . . . . . . . . . . 524, 525
ispseudoprime . 191, 202, 203, 206, 207, 217 lfuncost . . . . . . . . . . . . . . . 525, 526
ispseudoprimepower . . . . . . . . . . . 203 lfuncost0 . . . . . . . . . . . . . . . . . 526
issquare . . . . . . . . . . . . 202, 203, 204 lfuncreate . . . . . . . . . . . 517, 526, 528
issquareall . . . . . . . . . . . . . . . . 204 lfundiv . . . . . . . . . . . . . . . . . . . 528
issquarefree . . . . . . . . . 169, 204, 205 lfundual . . . . . . . . . . . . . . . 528, 529
istotient . . . . . . . . . . . . . . . . . 205 lfunetaquo . . . . . . . . . . . . . . 520, 529
lfungenus2 . . . . . . . . . . . . . . . . . 529
J lfunhardy . . . . . . . . . . . . . . 529, 530
lfuninit . . . . . . . . . . . . 517, 521, 530
j . . . . . . . . . . . . . . . . . . . . . . . 469 lfuninit0 . . . . . . . . . . . . . . . . . 530
jacobi . . . . . . . . . . . . . . . . . . . 282 lfunlambda . . . . . . . . . . . . . . . . . 530
jbessel . . . . . . . . . . . . . . . . . . . 297 lfunlambda0 . . . . . . . . . . . . . . . . 530
jbesselh . . . . . . . . . . . . . . . . . . 297 lfunmf . . . . . . . . . . . . . . . . . . . 537
jell . . . . . . . . . . . . . . . . . . . . . 490 lfunmfspec . . . . . . . . . . . . . . 530, 531
lfunmul . . . . . . . . . . . . . . . . . . . 531
K
lfunorderzero . . . . . . . . . . . . . . . 531
kbessel . . . . . . . . . . . . . . . . . . . 297 lfunqf . . . . . . . . . . . . . . . . 531, 532
ker . . . . . . . . . . . . . . . . . . . . . 271 lfunrootres . . . . . . . . . . . . . . . . 532
kerint . . . . . . . . . . . . . . . . . . . 271 lfunshift . . . . . . . . . . . . . . 532, 533
keyword . . . . . . . . . . . . . . . . . . . 49 lfunsympow . . . . . . . . . . . . . . . . . 533

622
lfuntheta . . . . . . . . . . . . . . . . . 533 logfile . . . . . . . . . . . . . . . . . . . 124
lfunthetacost . . . . . . . . . . . . . . . 533 logint . . . . . . . . . . . . . . . . . . . 206
lfunthetacost0 . . . . . . . . . . . . . . 534 logint0 . . . . . . . . . . . . . . . . . . . 206
lfunthetainit . . . . . . . . . . . . 517, 534 LONG_MAX . . . . . . . 155, 158, 159, 394, 404
lfuntwist . . . . . . . . . . . . . . . . . 534 lvalue . . . . . . . . . . . . . . . . . . 29, 32
lfunzeros . . . . . . . . . . . . . . 534, 535 lvalue . . . . . . . . . . . . . . . . . . . 30
libpari . . . . . . . . . . . . . . . . . . . . 5
LiDIA . . . . . . . . . . . . . . . . . . . . 191 M
lift . . . . . . . . . . . . . 141, 148, 152, 153
lift0 . . . . . . . . . . . . . . . . . . . . 153 Map . . . . . . . . . . . . . . . . . . . . . 140
liftall . . . . . . . . . . . . . . . . . . . 153 mapdelete . . . . . . . . . . . . . . . . . 104
liftint . . . . . . . . . . . . . . . . 153, 154 mapget . . . . . . . . . . . . . . . . 104, 105
liftpol . . . . . . . . . . . . . . . . 153, 154 mapisdefined . . . . . . . . . 104, 105, 292
limit . . . . . . . . . . . . . . . . . . . . 46 mapput . . . . . . . . . . . . . . . . 105, 140
limitnum . . . . . . . . . . . . . . . 327, 330 Mat . . . . . . . . . . . 25, 26, 140, 258, 266
limitnum0 . . . . . . . . . . . . . . . . . 330 matadjoint . . . . . . . . . . . . . . 258, 261
lindep . . . . . . . . . . . . . . . . 255, 260 matadjoint0 . . . . . . . . . . . . . . . . 261
lindep0 . . . . . . . . . . . . . . . . . . . 261 matalgtobasis . . . . . . . . . . . . . . . 394
line editor . . . . . . . . . . . . . . . . . . 64 matbasistoalg . . . . . . . . . . . . . . . 394
linear dependence . . . . . . . . . . . . . 260 matcompanion . . . . . . . . . . . . . . . 261
lines . . . . . . . . . . . . . . . . . . . . 124 matconcat . . . . . . . . . 258, 261, 262, 264
linewrap . . . . . . . . . . . . . . . . . . 124 matdet . . . . . . . . . . . . . . . . . . . 262
Linit . . . . . . . . . . . . . . . . . . . . 517 matdetint . . . . . . . . . . . . . . . . . 263
Lisp . . . . . . . . . . . . . . . . . . . . . 57 matdetmod . . . . . . . . . . . . . . . . . 263
list . . . . . . . . . . . . . . . . . . . . . 7, 27 matdiagonal . . . . . . . . . . . . . . . . 263
List . . . . . . . . . . . . . . . . . . . . . 139 mateigen . . . . . . . . . . . . . . . 264, 265
listcreate . . . . . . . . . . . . . . . . . 100 matfrobenius . . . . . . . . . . . . . . . 265
listinsert . . . . . . . . . . . . . . . . . 100 Math::Pari . . . . . . . . . . . . . . . . . 57
listkill . . . . . . . . . . . . . . . . . . 100 mathess . . . . . . . . . . . . . . . . . . . 265
listpop . . . . . . . . . . . . . . . . . . . 100 mathilbert . . . . . . . . . . . . . . . . . 265
listpop0 . . . . . . . . . . . . . . . . . . 100 mathnf . . . . . . . . . . . . . . . . 255, 265
listput . . . . . . . . . . . . . . . . . . . 100 mathnf0 . . . . . . . . . . . . . . . . . . . 267
listput0 . . . . . . . . . . . . . . . . . . 101 mathnfmod . . . . . . . . . . . . . . . . . 267
listsort . . . . . . . . . . . . 101, 102, 288 mathnfmodid . . . . . . . . . . . . . . . . 267
LLL . . . . . . . . . . 227, 265, 271, 282, 391 mathouseholder . . . . . . . . . . . 267, 268
lll . . . . . . . . . . . . . . . . . . . . . 283 matid . . . . . . . . . . . . . . . . . . . . 268
lllgram . . . . . . . . . . . . . . . . . . . 283 matimage . . . . . . . . . . . . . . . . . . 268
lllgramint . . . . . . . . . . . . . . . . . 283 matimage0 . . . . . . . . . . . . . . . . . 269
lllgramkerim . . . . . . . . . . . . . . . 283 matimagecompl . . . . . . . . . . . . . . . 269
lllint . . . . . . . . . . . . . . . . . . . 283 matimagemod . . . . . . . . . . . . . 269, 270
lllkerim . . . . . . . . . . . . . . . . . . 283 matindexrank . . . . . . . . . . . . . . . 270
Lmath . . . . . . . . . . . . . . . . . . . . 517 matintersect . . . . . . . . . . . . . . . 270
lngamma . . . . . . . . . . . . . . . . . . . 303 matinverseimage . . . . . . . . . . . . . 270
local . . . . . . . . . . . . . . . 36, 102, 103 matinvmod . . . . . . . . . . . . . . 270, 271
localbitprec . . . . . . . . . . . . . 18, 102 matisdiagonal . . . . . . . . . . . . . . . 271
localprec . . . . . . . . . . . . . . . . . 103 matker . . . . . . . . . . . . . . . . . . . 271
log . . . . . . . . . . . 59, 60, 109, 124, 303 matker0 . . . . . . . . . . . . . . . . . . . 271
log1p . . . . . . . . . . . . . . . . . . . . 304 matkerint . . . . . . . . . . . . . . . . . 271
logfile . . . . . . . . . . . . . . . . . . . . 109 matkerint0 . . . . . . . . . . . . . . . . . 271

623
matkermod . . . . . . . . . . . . . . 271, 272 mfeval . . . . . . . . . . . . . . . . . . . 549
matmuldiagonal . . . . . . . . . . . . . . 272 mffields . . . . . . . . . . . . . . . 549, 550
matmultodiagonal . . . . . . . . . . . . . 272 mffromell . . . . . . . . . . . . . . . . . 550
matpascal . . . . . . . . . . . . . . . . . 272 mffrometaquo . . . . . . . . . . . . . . . 550
matpermanent . . . . . . . . . . . . . . . 272 mffromlfun . . . . . . . . . . . . . . 550, 551
matqpascal . . . . . . . . . . . . . . . . . 272 mffromqf . . . . . . . . . . . . . . . 551, 552
matqr . . . . . . . . . . . . . . . . . 272, 273 mfgaloisprojrep . . . . . . . . . . 552, 553
matrank . . . . . . . . . . . . . . . . . . . 273 mfgaloistype . . . . . . . . . . . . . . . 553
matreduce . . . . . . . . . . . . . . . . . 273 mfhecke . . . . . . . . . . . . . . . . 553, 554
matrix . . . . . . . . . . . . . . . . 7, 8, 25, 50 mfheckemat . . . . . . . . . . . . . . . . . 554
matrix . . . . . . . . . . . . . . . . . 25, 273 mfinit . . . . . . . . . . . . . 554, 555, 566
matrixqz . . . . . . . . . . . . . . . . . . 274 mfisCM . . . . . . . . . . . . . . . . . . . 555
matrixqz0 . . . . . . . . . . . . . . . . . 274 mfisequal . . . . . . . . . . . . . . 555, 556
matsize . . . . . . . . . . . . . . . . . . . 274 mfisetaquo . . . . . . . . . . . . . . . . . 556
matsnf . . . . . . . . . . . . . . . . . . . 274 mfkohnenbasis . . . . . . . . . . . . . . . 556
matsnf0 . . . . . . . . . . . . . . . . . . . 275 mfkohnenbijection . . . . . . . . . 557, 558
matsolve . . . . . . . . . . . . . . . . . . 275 mfkohneneigenbasis . . . . . . . . . . . 558
matsolvemod . . . . . . . . . . . . . 275, 276 mflinear . . . . . . . . . . . . . . . 558, 559
matsupplement . . . . . . . . . . . . . . . 276 mfmanin . . . . . . . . . . . . . . . . . . . 559
mattranspose . . . . . . . . . . . . . . . 276 mfmul . . . . . . . . . . . . . . . . . 559, 560
max . . . . . . . . . . . . . . . . . . . . . 138 mfnumcusps . . . . . . . . . . . . . . . . . 560
member functions . . . . . . . . 47, 350, 469 mfparams . . . . . . . . . . . . . . . . . . 560
mfatkin . . . . . . . . . . . . . . . . . . . 538 mfperiodpol . . . . . . . . . . . . . 560, 561
mfatkineigenvalues . . . . . . . . 538, 539 mfperiodpolbasis . . . . . . . . . . . . . 561
mfatkininit . . . . . . . . . . . . . 539, 540 mfpetersson . . . . . . . . . . . . . 561, 562
mfbasis . . . . . . . . . . . . . . . . . . . 540 mfpow . . . . . . . . . . . . . . . . . . . . 562
mfbd . . . . . . . . . . . . . . . . . . . . . 540 mfsearch . . . . . . . . . . . . . . . 562, 563
mfbracket . . . . . . . . . . . . . . . . . 540 mfshift . . . . . . . . . . . . . . . . . . . 563
mfcoef . . . . . . . . . . . . . . . . 540, 541 mfshimura . . . . . . . . . . . . . . . . . 563
mfcoefs . . . . . . . . . . . . . . . . . . . 541 mfslashexpansion . . . . . . . . . . 563, 565
mfconductor . . . . . . . . . . . . . . . . 541 mfspace . . . . . . . . . . . . . . . . . . . 565
mfcosets . . . . . . . . . . . . . . . 541, 542 mfsplit . . . . . . . . . . . . . . . . 565, 566
mfcuspisregular . . . . . . . . . . . . . 542 mfsturm . . . . . . . . . . . . . . . . . . . 566
mfcusps . . . . . . . . . . . . . . . . . . . 542 mfsymbol . . . . . . . . . . . . . . . 566, 567
mfcuspval . . . . . . . . . . . . . . 542, 543 mfsymboleval . . . . . . . . . . . . 567, 568
mfcuspwidth . . . . . . . . . . . . . . . . 543 mftaylor . . . . . . . . . . . . . . . . . . 568
mfDelta . . . . . . . . . . . . . . . . . . . 537 mfTheta . . . . . . . . . . . . . . . . . . . 538
mfderiv . . . . . . . . . . . . . . . . . . . 543 mftobasis . . . . . . . . . . . . . . 568, 569
mfderivE2 . . . . . . . . . . . . . . . . . 543 mftocoset . . . . . . . . . . . . . . 569, 570
mfdescribe . . . . . . . . . . . . . . 543, 544 mftonew . . . . . . . . . . . . . . . . . . . 570
mfdim . . . . . . . . . . . . . . . . . 544, 545 mftraceform . . . . . . . . . . . . . . . . 570
mfdiv . . . . . . . . . . . . . . . . . . . . 545 mftwist . . . . . . . . . . . . . . . . . . . 570
mfEH . . . . . . . . . . . . . . . . . . . . . 537 min . . . . . . . . . . . . . . . . . . . . . 138
mfeigenbasis . . . . . . . . . . . . 545, 546 minim . . . . . . . . . . . . . . . . . . . . 285
mfeigensearch . . . . . . . . . . . . 546, 547 minim2 . . . . . . . . . . . . . . . . . . . 285
mfeisenstein . . . . . . . . . . . . . . . 547 minimal model . . . . . . . . . . . . . . . 491
mfEk . . . . . . . . . . . . . . . . . . . . . 538 minimal polynomial . . . . . . . . . . . . 276
mfembed . . . . . . . . . . . . . . . . . . . 547 minimal vector . . . . . . . . . . . . . . . 285
mfembed0 . . . . . . . . . . . . . . . . . . 549 minim_raw . . . . . . . . . . . . . . . . . 285

624
minim_zm . . . . . . . . . . . . . . . . . . 285 mssplit . . . . . . . . . . . . . . . . 589, 590
minpoly . . . . . . . . . . . . . . . . . . . 276 msstar . . . . . . . . . . . . . . . . . . . 590
mklist . . . . . . . . . . . . . . . . . . . 140 mstooms . . . . . . . . . . . . . 581, 590, 591
mkoo . . . . . . . . . . . . . . . . . . . . . 155 multivariate polynomial . . . . . . . . . . 45
Mod . . . . . . . . . . . . . . . . . . . . . 140 my . . . . . . . . . . . . . . . 36, 40, 102, 103
mod . . . . . . . . . . . . . . . . . . . . . 351
modpr . . . . . . . . . . . . . . . . . . . . 413 N
modprinit . . . . . . . . . . . . . . . . . 405
modreverse . . . . . . . . . . . 394, 395, 423 nbessel . . . . . . . . . . . . . . . . . . . 297
modulus . . . . . . . . . . . . . . . . . . . 349 nbthreads . . . . . . . . . . . . . . . . . 125
Moebius . . . . . . . . . . . . . . . . 169, 206 newtonpoly . . . . . . . . . . . . . . . . . 395
moebius . . . . . . . . . . . . . . . . 169, 206 new_galois_format . . . . . . . . . 420, 421
Mordell-Weil group . . . . . 481, 484, 505, 506 next . . . . . . . . . . . . . . . . . . . 53, 84
mpcatalan . . . . . . . . . . . . . . . . . 295 nextprime . . . . . . . . . . . . . . 206, 207
mpeuler . . . . . . . . . . . . . . . . . . . 295 nf . . . . . . . . . . . . . . . . . . . . 47, 347
mpfact . . . . . . . . . . . . . . . . . . . 191 nf . . . . . . . . . . . . . . . . . . . 351, 471
mpfactr . . . . . . . . . . . . . . . . . . . 191 nfadd . . . . . . . . . . . . . . . . . . . . 401
mppi . . . . . . . . . . . . . . . . . . . . . 295 nfalgtobasis . . . . . . . . . . . . 394, 395
MPQS . . . . . . . . . . . . . . . . . 169, 191 nfbasis . . . . . 126, 395, 398, 400, 409, 410
msatkinlehner . . . . . . . . . . . . . . . 571 nfbasistoalg . . . . . . . . . . . . 394, 398
mscosets . . . . . . . . . . . . . . . 571, 572 nfcertify . . . . . . . . . 397, 398, 410, 422
mscosets0 . . . . . . . . . . . . . . . . . 572 nfcompositum . . . . . . . . . . . . 398, 399
mscuspidal . . . . . . . . . . . . . . . . . 572 nfdetint . . . . . . . . . . . . . . . . . . 399
msdim . . . . . . . . . . . . . . . . . . . . 572 nfdisc . . . . . . . . . . . . . 126, 399, 400
mseisenstein . . . . . . . . . . . . 572, 573 nfdiscfactors . . . . . . . . . . . . 400, 401
mseval . . . . . . . . . . . . . . . . 573, 574 nfdiv . . . . . . . . . . . . . . . . . . . . 401
msfarey . . . . . . . . . . . . . . . . . . . 574 nfdiveuc . . . . . . . . . . . . . . . . . . 401
msfarey0 . . . . . . . . . . . . . . . . . . 574 nfdivmodpr . . . . . . . . . . . . . . . . . 401
msfromcusp . . . . . . . . . . . . . . 574, 575 nfdivrem . . . . . . . . . . . . . . . . . . 401
msfromell . . . . . . . . . . . . . . 575, 576 nfeltadd . . . . . . . . . . . . . . . . . . 401
msfromhecke . . . . . . . . . . . . . . . . 576 nfeltdiv . . . . . . . . . . . . . . . . . . 401
msgetlevel . . . . . . . . . . . . . . . . . 577 nfeltdiveuc . . . . . . . . . . . . . . . . 401
msgetsign . . . . . . . . . . . . . . . . . 577 nfeltdivmodpr . . . . . . . . . . . . . . . 401
msgetweight . . . . . . . . . . . . . . . . 577 nfeltdivrem . . . . . . . . . . . . . . . . 401
mshecke . . . . . . . . . . . . . . . . 577, 578 nfeltembed . . . . . . . . . . . . . . 401, 402
msinit . . . . . . . . . . . . . . . . 571, 578 nfeltmod . . . . . . . . . . . . . . . . . . 402
msissymbol . . . . . . . . . . . . . . 578, 579 nfeltmul . . . . . . . . . . . . . . . . . . 402
mslattice . . . . . . . . . . . 509, 579, 580 nfeltmulmodpr . . . . . . . . . . . . . . . 402
msnew . . . . . . . . . . . . . . . . . . . . 580 nfeltnorm . . . . . . . . . . . . . . . . . 402
msomseval . . . . . . . . . . . . . . . . . 580 nfeltpow . . . . . . . . . . . . . . . . . . 402
mspadicinit . . . . . . . . . . . . . 582, 583 nfeltpowmodpr . . . . . . . . . . . . . . . 402
mspadicL . . . . . . . 497, 580, 582, 583, 590 nfeltreduce . . . . . . . . . . . . . . . . 403
mspadicmoments . . . . . . . . 581, 583, 590 nfeltreducemodpr . . . . . . . . . . . . . 403
mspadicseries . . . . . . . 497, 583, 585, 590 nfeltsign . . . . . . . . . . . . . . . . . 403
mspathgens . . . . . . . . . . . 573, 585, 586 nfelttrace . . . . . . . . . . . . . . . . . 403
mspathlog . . . . . . . . . . . . . . . . . 586 nfeltval . . . . . . . . . . . . . . . . . . 403
mspetersson . . . . . . . . . . . . . 586, 587 nffactor . . . . . . . . . . 187, 371, 404, 405
mspolygon . . . . . . . . . . . . . . 587, 589 nffactorback . . . . . . . . . 347, 383, 405
msqexpansion . . . . . . . . . . . . . . . 589 nffactormod . . . . . . . . . . . . . . . . 405

625
nfgaloisapply . . . . . . . . . . . . . . . 405 nftrace . . . . . . . . . . . . . . . . . . . 403
nfgaloisconj . . . . . . . . . . . . 376, 406 nfval . . . . . . . . . . . . . . . . . . . . 404
nfgrunwaldwang . . . . . . . . . . . . . . 407 nfvalrem . . . . . . . . . . . . . . . . . . 404
nfhilbert . . . . . . . . . . . . . . 407, 408 nf_ADDZK . . . . . . . . . . . . . . . . . . 423
nfhilbert0 . . . . . . . . . . . . . . . . . 408 nf_ALL . . . . . . . . . . . . . . . . . . . 423
nfhnf . . . . . . . . . . . . . . . . . 270, 408 nf_FORCE . . . . . . . . . . . . . . . 355, 356
nfhnf0 . . . . . . . . . . . . . . . . . . . 408 nf_GEN . . . . . . . . . . . . . . . . 356, 393
nfhnfmod . . . . . . . . . . . . . . . . . . 408 nf_GENMAT . . . . . . . . . . . . . . . . . 356
nfhyperellpadicfrobenius . . . . . . . 515 nf_INIT . . . . . . . . . . . . . . . . . . . 393
nfinit . . . . . 126, 347, 376, 408, 410, 421, nf_ORIG . . . . . . . . . . . . . . . . 410, 423
422, 423, 449 nf_PARTIALFACT . . . . . . . . 126, 410, 423
nfinit0 . . . . . . . . . . . . . . . . . . . 410 nf_RAW . . . . . . . . . . . . . . . . . . . 423
nfinitall . . . . . . . . . . . . . . . . . 410 nf_RED . . . . . . . . . . . . . . . . . . . 410
nfinitred . . . . . . . . . . . . . . . . . 410 nf_ROUND2 . . . . . . . . . . . . . . . . . 410
nfinitred2 . . . . . . . . . . . . . . . . . 410 no . . . . . . . . . . . . . . . . . . . 351, 471
nfinv . . . . . . . . . . . . . . . . . . . . 402 norm . . . . . . . . . . . . . . . . . . . . . 154
nfisideal . . . . . . . . . . . . . . . . . 410 norml2 . . . . . . . . . . . . . . . . . . . 276
nfisincl . . . . . . . . . . . . . . . . . . 411 normlp . . . . . . . . . . . . . . . . . . . 277
nfisincl0 . . . . . . . . . . . . . . . . . 411 not . . . . . . . . . . . . . . . . . . . . . . 130
nfisisom . . . . . . . . . . . . . . . 411, 412 nucomp . . . . . . . . . . . . . . . . . . . 213
nfislocalpower . . . . . . . . . . . 412, 413 nudupl . . . . . . . . . . . . . . . . . . . 213
nfkermodpr . . . . . . . . . . . . . . . . . 413 number field . . . . . . . . . . . . . . . . 21
nfmod . . . . . . . . . . . . . . . . . . . . 402 numbpart . . . . . . . . . . . . . . . 166, 167
nfmodpr . . . . . . . . . . . . . . . . 413, 414 numdiv . . . . . . . . . . . . . . . . . . . 207
nfmodprinit . . . . . . . . 401, 402, 413, 414 numer . . . . . . . . . . . . . . . . . . . . 154
nfmodprinit0 . . . . . . . . . . . . . . . 414 numerator . . . . . . . . . . . . . . . 35, 154
nfmodprlift . . . . . . . . . . . . . 413, 414 numerical derivation . . . . . . . . . . . . 30
nfmul . . . . . . . . . . . . . . . . . . . . 402 numerical integration . . . . . . . . . . . 313
nfmulmodpr . . . . . . . . . . . . . . . . . 402 numtoperm . . . . . . . . . . . . . . 166, 168
nfnewprec . . . . . . . . . . . 409, 414, 415 nupow . . . . . . . . . . . . . . . . . . . . 213
nfnorm . . . . . . . . . . . . . . . . . . . 402
nfpolsturm . . . . . . . . . . . . . . . . . 415 O
nfpow . . . . . . . . . . . . . . . . . . . . 402
O . . . . . . . . . . . . . . . . . . . . . . . 231
nfpowmodpr . . . . . . . . . . . . . . . . . 403
omega . . . . . . . . . . . . . . . . . . . . 215
nfreduce . . . . . . . . . . . . . . . . . . 403
omega . . . . . . . . . . . . . . 207, 470, 471
nfreducemodpr . . . . . . . . . . . . . . . 403
oncurve . . . . . . . . . . . . . . . . . . . 489
nfroots . . . . . . . . . . . . . . . . 415, 416
oo . . . . . . . . . . . . . . . . . . . . . . 154
nfrootsof1 . . . . . . . . . . . . . . . . . 416
operator . . . . . . . . . . . . . . . . . . . 29
nfrootsQ . . . . . . . . . . . . . . . . . . 416
or . . . . . . . . . . . . . . . . . . . . . . 130
nfsnf . . . . . . . . . . . . . . . . . 416, 417
or . . . . . . . . . . . . . . . . . . . 147, 148
nfsnf0 . . . . . . . . . . . . . . . . . . . 417
order . . . . . . . . . . . . . . . . . . . . 230
nfsolvemodpr . . . . . . . . . . . . . . . 417
orderell . . . . . . . . . . . . . . . . . . 495
nfsplitting . . . . . . . . . . . . . 417, 418
output . . . . . . . . . . . . . . . . . 61, 125
nfsqr . . . . . . . . . . . . . . . . . . . . 402
nfsubfield . . . . . . . . . . . . . . . . . 375
P
nfsubfields . . . . . . . . . . . . . . . . 418
nfsubfields0 . . . . . . . . . . . . . . . 418 p . . . . . . . . . . . . . . . . . . . . 470, 471
nfsubfieldscm . . . . . . . . . . . . . . . 418 p-adic number . . . . . . . . . . . . . 7, 8, 20
nfsubfieldsmax . . . . . . . . . . . 418, 419 padicappr . . . . . . . . . . . . . . . . . 236

626
padicfields . . . . . . . . . . . . . 236, 237 plotinit . . . . . . . . . . . . . . . . . . 597
padicfields0 . . . . . . . . . . . . . . . 237 plotkill . . . . . . . . . . . . . . . . . . 598
padicprec . . . . . . . . . . . . . . . . . 155 plotlines . . . . . . . . . . . . . . . . . 598
parametric plot . . . . . . . . . . . . . . . 595 plotlinetype . . . . . . . . . . . . . . . 598
parapply . . . . . . . . . . . . . . . . . . 117 plotmove . . . . . . . . . . . . . . . . . . 598
pareval . . . . . . . . . . . . . . . . . . . 118 plotpoints . . . . . . . . . . . . . . . . . 598
parfor . . . . . . . . . . . . . . . . . . . 118 plotpointsize . . . . . . . . . . . . 598, 599
parforeach . . . . . . . . . . . . . . . . . 119 plotpointtype . . . . . . . . . . . . . . . 599
parforprime . . . . . . . . . . . . . . . . 119 plotrbox . . . . . . . . . . . . . . . . . . 599
parforprimestep . . . . . . . . . . . . . 120 plotrecth . . . . . . . . . . . . . . 596, 599
parforvec . . . . . . . . . . . . . . . . . 120 plotrecthraw . . . . . . . . . . . . . . . 599
PariEmacs . . . . . . . . . . . . . . . . . 601 plotrline . . . . . . . . . . . . . . . . . 599
pariplot . . . . . . . . . . . . . . . . . . 593 plotrmove . . . . . . . . . . . . . . . . . 599
parisize . . . . . . . . . . . . . 88, 120, 125 plotrpoint . . . . . . . . . . . . . . . . . 599
parisizemax . . . . . . 57, 88, 120, 125, 126 plotscale . . . . . . . . . . . . . . 596, 600
pari_malloc . . . . . . . . . . . . . . . . 83 plotstring . . . . . . . . . . . . . . . 50, 600
pari_realloc . . . . . . . . . . . . . . . 83 plotterm . . . . . . . . . . . . . . . . . . 50
pari_self . . . . . . . . . . . . . . . . . 111 pnqn . . . . . . . . . . . . . . . . . . . . . 182
pari_strchr . . . . . . . . . . . . . . 85, 112 pointell . . . . . . . . . . . . . . . . . . 512
pari_version . . . . . . . . . . . . . . . 116 pointer . . . . . . . . . . . . . . . . . . . . 68
parploth . . . . . . . . . . . . . . . 592, 593 Pol . . . . . . . . . . . . . . . . 22, 141, 142
parplothexport . . . . . . . . . . . . . . 593 pol . . . . . . . . . . . . . . . . . . . . . 351
parselect . . . . . . . . . . . . . . . . . 120 polchebyshev . . . . . . . . . . . . . . . 237
parsum . . . . . . . . . . . . . . . . . . . 120 polchebyshev1 . . . . . . . . . . . . 237, 250
partitions . . . . . . . . . . . . . . 166, 167 polchebyshev2 . . . . . . . . . . . . . . . 237
parvector . . . . . . . . . . . . . . . . . 120 polchebyshev_eval . . . . . . . . . . . . 237
Pascal triangle . . . . . . . . . . . . . . . 272 polclass . . . . . . . . . . . . . . . 237, 239
path . . . . . . . . . . . . . . . . . . . . . 126 polcoef . . . . . . . . . . . . . . . . . . . 239
Pauli . . . . . . . . . . . . . . . . . . . . . 396 polcoeff . . . . . . . . . . . . . . . 149, 239
Perl . . . . . . . . . . . . . . . . . . . . . 57 polcompositum . . . . . . . . . . . . . . . 419
Perl . . . . . . . . . . . . . . . . . . . . . 36 polcompositum0 . . . . . . . . . . . . . . 420
permcycles . . . . . . . . . . . . . . . . . 167 polcyclo . . . . . . . . . . . . . . . . . . 239
permorder . . . . . . . . . . . . . . . . . 168 polcyclofactors . . . . . . . . . . 239, 240
permsign . . . . . . . . . . . . . . . . . . 168 polcyclo_eval . . . . . . . . . . . . . . . 239
permtonum . . . . . . . . . . . . . . 166, 168 poldegree . . . . . . . . . . . . . . . . . 240
Pi . . . . . . . . . . . . . . . . . . . . . . 295 poldisc . . . . . . . . . . . . . . . . . . . 240
plot . . . . . . . . . . . . . . . . . . . . . 593 poldisc0 . . . . . . . . . . . . . . . . . . 240
plotbox . . . . . . . . . . . . . . . . 593, 594 poldiscfactors . . . . . . . . . . . . . . 240
plotclip . . . . . . . . . . . . . . . . . . 594 poldiscreduced . . . . . . . . . . . . . . 240
plotcolor . . . . . . . . . . . . . . 123, 594 polfnf . . . . . . . . . . . . . . . . . . . 371
plotcopy . . . . . . . . . . . . . . . . . . 594 polgalois . . . . . . . . . . . 125, 420, 421
plotcursor . . . . . . . . . . . . . . . . . 594 polgraeffe . . . . . . . . . . . . . . . . . 241
plotdraw . . . . . . . . . . . . . . . 594, 595 polhensellift . . . . . . . . . . . . . . . 241
plotexport . . . . . . . . . . . . . . . . . 595 polhermite . . . . . . . . . . . . . . . . . 241
ploth . . . . . . . . . . . . . . . 68, 595, 596 polhermite_eval . . . . . . . . . . . . . 241
plothexport . . . . . . . . . . . . . 596, 597 polhermite_eval0 . . . . . . . . . . . . . 241
plothraw . . . . . . . . . . . . . . . . . . 597 polint . . . . . . . . . . . . . . . . . . . 243
plothrawexport . . . . . . . . . . . . . . 597 polinterpolate . . . . . . . . . . . . . . 241
plothsizes . . . . . . . . . . . . . . 126, 597 poliscyclo . . . . . . . . . . . . . . 243, 244

627
poliscycloprod . . . . . . . . . . . . . . 244 precision . . . . . . . . . . . . . . . . . . . 293
polisirreducible . . . . . . . . . . . . . 244 precision . . . . . . . . . . . . . . . . . 155
pollaguerre . . . . . . . . . . . . . . . . 244 precision00 . . . . . . . . . . . . . . . . 155
pollaguerre_eval . . . . . . . . . . . . . 244 precprime . . . . . . . . . . . . . . . . . 207
pollaguerre_eval0 . . . . . . . . . . . . 244 preferences file . . . . . . . . . . . 13, 57, 62
Pollard Rho . . . . . . . . . . . . . . 169, 191 prettymatrix format . . . . . . . . . . . . 125
pollead . . . . . . . . . . . . . . . . . . . 244 prettyprinter . . . . . . . . . . . . 125, 126
pollegendre . . . . . . . . . . . . . 244, 245 prid . . . . . . . . . . . . . . . . . . . 47, 389
pollegendre_eval . . . . . . . . . . . . . 245 prime . . . . . . . . . . . . . . . . . . . . 207
pollegendre_eval0 . . . . . . . . . . . . 245 primecert . . . . . . . . . . . 202, 207, 209
polmod . . . . . . . . . . . . . . . . . . . . . 7 primecertexport . . . . . . . . . . 209, 210
polmod . . . . . . . . . . . . . . . . . . 8, 21 primecertisvalid . . . . . . . . . . 207, 210
polmodular . . . . . . . . . . . . . . 245, 493 primeform . . . . . . . . . . . . . . . . . 213
polrecip . . . . . . . . . . . . . . . . . . 245 primelimit . . . . . . . . . . . . . . 126, 421
polred . . . . . . . . . . . . . . . . . . . 421 primepi . . . . . . . . . . . . . . . . 210, 211
polred2 . . . . . . . . . . . . . . . . . . . 421 primes . . . . . . . . . . . . . . . . . . . 211
polredabs . . . . . . . . . . . . . . 422, 423 primes0 . . . . . . . . . . . . . . . . . . . 211
polredabs0 . . . . . . . . . . . . . . . . . 423 principal ideal . . . . . . . . . . . . 355, 391
polredbest . . . . . . 410, 421, 422, 423, 424 print . . . . . . . . . . . . . . . . 49, 50, 105
polredord . . . . . . . . . . . . . . . . . 424 print1 . . . . . . . . . . . . . . . . . . . 106
polresultant . . . . . . . . . . . . 200, 245 printf . . . . . . . . . . . . . 106, 108, 123
polresultant0 . . . . . . . . . . . . . . . 246 printp . . . . . . . . . . . . . . . . . . . 108
polresultantext . . . . . . . . . . 245, 246 printsep . . . . . . . . . . . . . . . . . . 108
polresultantext0 . . . . . . . . . . 231, 246 printsep1 . . . . . . . . . . . . . . . . . 108
Polrev . . . . . . . . . . . 22, 141, 142, 143 printtex . . . . . . . . . . . . . . . . . . 109
polroots . . . . . . . . . . . . . . . 246, 255 priority . . . . . . . . . . . . . . . . . . . 29
polrootsbound . . . . . . . . . . . . . . . 247 prod . . . . . . . . . . . . . . . . . . . . . 330
polrootsff . . . . . . . . . . . . . . . . . 247 prodeuler . . . . . . . . . . . . . . . . . 331
polrootsmod . . . . . . . . . . 228, 247, 248 prodeulerrat . . . . . . . . . . . . 331, 332
polrootspadic . . . . . . . 228, 248, 249, 255 prodinf . . . . . . . . . . . . . . . . . . . 332
polrootsreal . . . . . . . . . . . . . . . 249 prodinf1 . . . . . . . . . . . . . . . . . . 332
polsturm . . . . . . . . . . . . . . . . . . 249 prodnumrat . . . . . . . . . . . . . . . . . 332
polsubcyclo . . . . . . . . . . . . . . . . 250 product . . . . . . . . . . . . . . . . . . . 132
polsylvestermatrix . . . . . . . . . . . 250 produit . . . . . . . . . . . . . . . . . . . 331
polsym . . . . . . . . . . . . . . . . . . . 250 programming . . . . . . . . . . . . . . . . 68
poltchebi . . . . . . . . . . . . . . . . . 250 projective module . . . . . . . . . . . . . . 348
polteichmuller . . . . . . . . . . . . . . 250 prompt . . . . . . . . . . . . . . . . . . . 127
poltschirnhaus . . . . . . . . . . . . . . 424 psdraw . . . . . . . . . . . . . . . . . . . 600
polylog . . . . . . . . . . . . . . . . . . . 304 pseudo-basis . . . . . . . . . . . . . . . . . 349
polylog0 . . . . . . . . . . . . . . . . . . 305 pseudo-matrix . . . . . . . . . . . . . . . . 348
polylogmult . . . . . . . . . . . . . 305, 306 psfile . . . . . . . . . . . . . . . . . . . 127
polylogmult_interpolate . . . . . . . . 306 psi . . . . . . . . . . . . . . . . . . . . . 306
polynomial . . . . . . . . . . . . . . . 7, 8, 21 psploth . . . . . . . . . . . . . . . . . . . 600
polzag . . . . . . . . . . . . . . . . . . . 251 psploth0 . . . . . . . . . . . . . . . . . . 600
polzagier . . . . . . . . . . . . . . 251, 333 psplothraw . . . . . . . . . . . . . . . . . 600
PostScript . . . . . . . . . . . . . . . . . 592 Python . . . . . . . . . . . . . . . . . . . 57
power series . . . . . . . . . . . . . . . 7, 8, 22 p_to_GEN . . . . . . . . . . . . . . . . . . 196
powering . . . . . . . . . . . . . . . . . . 133
powers . . . . . . . . . . . . . . . . . . . 277 Q

628
qfauto . . . . . . . . . . . . . . . . . . . 278 quadclassno . . . . . . . . . . . . . . . . 215
qfauto0 . . . . . . . . . . . . . . . . . . . 278 quadclassunit . . . . . . . . . 211, 214, 217
qfautoexport . . . . . . . . . . . . . . . 278 quadclassunit0 . . . . . . . . . . . . . . 215
Qfb . . . . . . . . . . . . . . . . . . . . . 142 quaddisc . . . . . . . . . . . . . . . 182, 215
Qfb0 . . . . . . . . . . . . . . . . . . . . . 142 quadgen . . . . . . . . . . . . . . . . 215, 216
qfbclassno . . . . . . . . . . . 211, 214, 215 quadgen0 . . . . . . . . . . . . . . . . . . 216
qfbclassno0 . . . . . . . . . . . . . . . . 212 quadhilbert . . . . . . . . . . . . . . . . 216
qfbcompraw . . . . . . . . . . . . . . . . . 212 quadpoly . . . . . . . . . . . . . . . . . . 216
qfbhclassno . . . . . . . . . . . . . . . . 212 quadpoly0 . . . . . . . . . . . . . . . . . 216
qfbil . . . . . . . . . . . . . . . . . . . . 278 quadratic number . . . . . . . . . . . 7, 8, 20
qfbnucomp . . . . . . . . . . . . . . . . . 212 quadray . . . . . . . . . . . . . . . . . . . 216
qfbnupow . . . . . . . . . . . . . . . . . . 213 quadregulator . . . . . . . . . . . . 214, 216
qfbpowraw . . . . . . . . . . . . . . . . . 213 quadunit . . . . . . . . . . . . . . . 216, 217
qfbprimeform . . . . . . . . . . . . . . . 213 quadunit0 . . . . . . . . . . . . . . . . . 217
qfbred . . . . . . . . . . . . . . . . . . . 213 quit . . . . . . . . . . . . . . . . . . . 61, 109
qfbred0 . . . . . . . . . . . . . . . . . . . 213 quodif . . . . . . . . . . . . . . . . . . . 316
qfbredsl2 . . . . . . . . . . . . . . 213, 214 quote . . . . . . . . . . . . . . . . . . . . 99
qfbsolve . . . . . . . . . . . . . . . . . . 214 quotient . . . . . . . . . . . . . . . . . . . 132
qfeval . . . . . . . . . . . . . . . . . . . 278 Q_denom . . . . . . . . . . . . . . . . . . . 150
qfeval0 . . . . . . . . . . . . . . . . . . . 280 Q_primpart . . . . . . . . . . . . . . . . . 271
qfgaussred . . . . . . . . . . . . . . 280, 281 Q_remove_denom . . . . . . . . . . . . . . 154
qfgaussred_positive . . . . . . . . . . . 281
qfi . . . . . . . . . . . . . . . . . . . . . 142 R
qfisom . . . . . . . . . . . . . . . . . . . 281
qfisom0 . . . . . . . . . . . . . . . . . . . 281 r1 . . . . . . . . . . . . . . . . . . . . . 351
.
qfisominit . . . . . . . . . . . . . . . . . 281 r2 . . . . . . . . . . . . . .
. . . . . . . . 351
qfisominit0 . . . . . . . . . . . . . . . . 281 ramanujantau . . . . . . .
. . . . . . . . 217
qfjacobi . . . . . . . . . . . . . . . 264, 281 ramification group . . . . .
. . . . . . . . 391
qflll . . . . . . . . . . . . . . 255, 282, 283 random . . . . . . . . . . .
. . . . . . 97, 156
qflll0 . . . . . . . . . . . . . . . . . . . 283 randomprime . . . . . . . .
. . . . . 217, 218
qflllgram . . . . . . . . . . . . . . . . . 283 randomprime0 . . . . . . .
. . . . . . . . 218
qflllgram0 . . . . . . . . . . . . . . . . . 283 rank . . . . . . . . . . . . .
. . . . . . . . 273
qfminim . . . . . . . . . . . . . . . . 283, 286 rational function . . . . . .
. . . . . . . 7, 23
qfminim0 . . . . . . . . . . . . . . . . . . 285 rational number . . . . . .
. . . . . . 7, 8, 19
qfnorm . . . . . . . . . . . . . . . . . . . 285 raw format . . . . . . . . .
. . . . . . . . 125
qforbits . . . . . . . . . . . . . . . . . . 285 read . . . . . . . . . . . . .
. 50, 58, 109, 116
qfparam . . . . . . . . . . . . . . . . 285, 286 readline . . . . . . . . . .
. . . . . . . . 127
qfperfection . . . . . . . . . . . . . . . 286 readstr . . . . . . . . . . .
. . . . . . . . 109
qfr . . . . . . . . . . . . . . . . . . . . . 142 readvec . . . . . . . . . . .
. . . . . . 50, 109
qfrep . . . . . . . . . . . . . . . . . . . . 286 real number . . . . . . . . .
. . . . . . 7, 8, 17
qfrep0 . . . . . . . . . . . . . . . . . . . 286 real . . . . . . . . . . . . .
. . . . . . . . 157
qfsign . . . . . . . . . . . . . . . . . . . 286 realbitprecision . . . . .
61, 102, 127, 293,
qfsolve . . . . . . . . . . . . . . . . 286, 287 326, 516, 520
Qp_exp . . . . . . . . . . . . . . . . . . . 299 realprecision . . . . . 18, 61, 103, 128, 293,
Qp_gamma . . . . . . . . . . . . . . . . . . 299 326, 516, 520
Qp_log . . . . . . . . . . . . . . . . . . . 304 realroots . . . . . . . . . . . . . . . . . 249
Qp_sqrt . . . . . . . . . . . . . . . . . . . 307 recover . . . . . . . . . . . . . . . . . . . 128
Qp_sqrtn . . . . . . . . . . . . . . . . . . 308 recursion depth . . . . . . . . . . . . . . . 45
QR-decomposition . . . . . . . . . . . . . 272 recursion . . . . . . . . . . . . . . . . . . 45

629
recursive plot . . . . . . . . . . . . . . . . 595 rnfidealup . . . . . . . . . . . . . . 433, 434
redimag . . . . . . . . . . . . . . . . . . . 213 rnfidealup0 . . . . . . . . . . . . . . . . 434
redreal . . . . . . . . . . . . . . . . . . . 213 rnfinit . . . . . . . . . . . . . . . . 434, 435
redrealnod . . . . . . . . . . . . . . . . . 213 rnfinit0 . . . . . . . . . . . . . . . . . . 435
reduceddiscsmith . . . . . . . . . . . . . 241 rnfisabelian . . . . . . . . . . . . . . . 435
reduction . . . . . . . . . . . . . . . 212, 213 rnfisfree . . . . . . . . . . . . . . . . . 435
reference card . . . . . . . . . . . . . . . . 60 rnfislocalcyclo . . . . . . . . . . 435, 436
reg . . . . . . . . . . . . . . . . . . . . . 351 rnfisnorm . . . . . . . . . . . . . . . . . 436
removeprimes . . . . . . . . . . . . 122, 218 rnfisnorminit . . . . . . . . . . . . . . . 436
return . . . . . . . . . . . . . . . . . 53, 84 rnfkummer . . . . . . . . . . . . . . . . . 436
RgX_sturmpart . . . . . . . . . . . . . . . 250 rnflllgram . . . . . . . . . . . . . . 436, 437
rhoreal . . . . . . . . . . . . . . . . . . . 213 rnfnormgroup . . . . . . . . . . . . . . . 437
rhorealnod . . . . . . . . . . . . . . . . . 213 rnfpolred . . . . . . . . . . . . . . . . . 437
Riemann zeta-function . . . . . . . . . 43, 309 rnfpolredabs . . . . . . . . . . . . 437, 438
rnf . . . . . . . . . . . . . . . . . . . . . . 348 rnfpolredbest . . . . . . . . . . . . 437, 438
rnfalgtobasis . . . . . . . . . . . . . . . 424 rnfpseudobasis . . . . . . . . . . . 438, 439
rnfbasis . . . . . . . . . . . . . . . . . . 424 rnfsimplifybasis . . . . . . . . . . . . . 408
rnfbasistoalg . . . . . . . . . . . . . . . 424 rnfsteinitz . . . . . . . . . . . . . 439, 440
rnfcharpoly . . . . . . . . . . . . . . . . 425 Roblot . . . . . . . . . . . . . . . . . . . . 396
rnfconductor . . . . . . . . . . . . . . . 425 roots . . . . . . . . . . . . . . 247, 351, 470
rnfconductor0 . . . . . . . . . . . . . . . 425 rootsof1 . . . . . . . . . . . . . . . . . . 306
rnfdedekind . . . . . . . . . . . . . 425, 426 round 4 . . . . . . . . . . . . . . . . 235, 396
rnfdet . . . . . . . . . . . . . . . . 426, 427 round . . . . . . . . . . . . . . . . . . . . 157
rnfdisc . . . . . . . . . . . . . . . . . . . 427 round0 . . . . . . . . . . . . . . . . . . . 157
rnfdiscf . . . . . . . . . . . . . . . . . . 427 row vector . . . . . . . . . . . . . . . . 7, 23
rnfeltabstorel . . . . . . . . . . . . . . 427
rnfeltdown . . . . . . . . . . . . . . 427, 428 S
rnfeltdown0 . . . . . . . . . . . . . . . . 428
rnfeltnorm . . . . . . . . . . . . . . . . . 428 Scalable Vector Graphics . . . . . . . 592
rnfeltreltoabs . . . . . . . . . . . 428, 429 scalar product . . . . . . . . . . . . . . . 132
rnfelttrace . . . . . . . . . . . . . . . . 429 scalar type . . . . . . . . . . . . . . . . . . 7
rnfeltup . . . . . . . . . . . . . . . . . . 429 Schertz . . . . . . . . . . . . . . . . . . . 216
rnfeltup0 . . . . . . . . . . . . . . . . . 429 Schönage . . . . . . . . . . . . . . . . . . 246
rnfequation . . . . . . . . . . . . . 429, 430 scientific format . . . . . . . . . . . . . . 123
rnfequation0 . . . . . . . . . . . . . . . 430 SEA . . . . . . . . . . . . . . . . . . . . . 476
rnfequation2 . . . . . . . . . . . . . . . 430 secure . . . . . . . . . . . . . . . . . . . 128
rnfhnfbasis . . . . . . . . . . . . . . . . 430 select . . . . . . . . . . . . . . . . . . . 110
rnfidealabstorel . . . . . . . . . . 430, 431 self . . . . . . . . . . . . . . . . . . . . . 111
rnfidealdown . . . . . . . . . . . . . . . 431 Ser . . . . . . . . . . . . . . . . 22, 143, 253
rnfidealfactor . . . . . . . . . . . . . . 431 Ser0 . . . . . . . . . . . . . . . . . . . . . 143
rnfidealhnf . . . . . . . . . . . . . . . . 432 seralgdep . . . . . . . . . . . . . . . . . 287
rnfidealmul . . . . . . . . . . . . . . . . 432 serchop . . . . . . . . . . . . . . . . . . . 157
rnfidealnormabs . . . . . . . . . . . . . 432 serconvol . . . . . . . . . . . . . . . . . 251
rnfidealnormrel . . . . . . . . . . . . . 432 seriesprecision 61, 129, 294, 479, 480, 507
rnfidealprimedec . . . . . . . . . . . . . 432 serlaplace . . . . . . . . . . . . . . . . . 251
rnfidealreltoabs . . . . . . . . . . 432, 433 serprec . . . . . . . . . . . . . . . . . . . 158
rnfidealreltoabs0 . . . . . . . . . . . . 433 serreverse . . . . . . . . . . . . . . . . . 251
rnfidealtwoelement . . . . . . . . . . . 433 Set . . . . . . . . . . . . . . . . . . . . . 144
rnfidealtwoelt . . . . . . . . . . . . . . 433 setbinop . . . . . . . . . . . . . . . . . . 287

630
setintersect . . . . . . . . . . . . . . . 288 string context . . . . . . . . . . . . . . . . 49
setisset . . . . . . . . . . . . . . . . . . 288 string . . . . . . . . . . . . . . . . . 7, 27, 49
setminus . . . . . . . . . . . . . . . . . . 288 strjoin . . . . . . . . . . . . . . . . . . . 112
setrand . . . . . . . . . . . . . . 97, 111, 156 Strprintf . . . . . . . . . . . . . . . . . 85
setsearch . . . . . . . . . . . 101, 288, 289 strprintf . . . . . . . . . . . . . . 112, 123
setunion . . . . . . . . . . . . . . . . . . 289 strsplit . . . . . . . . . . . . . . . 112, 113
Shanks SQUFOF . . . . . . . . . . . 169, 191 Strtex . . . . . . . . . . . . . . . . . . . 85
Shanks . . . . . . . . . . . . 142, 211, 212, 213 strtex . . . . . . . . . . . . . . 85, 113, 117
shift . . . . . . . . . . . . . . . . . . . . 138 strtime . . . . . . . . . . . . . . . . . . . 113
shiftmul . . . . . . . . . . . . . . . . . . 138 strtoGEN . . . . . . . . . . . . . . . . . . 144
sigma . . . . . . . . . . . . . . . . . 182, 218 sturm . . . . . . . . . . . . . . . . . . . . 250
sign . . . . . . . . . . . . . . . . . . . . . 138 sturmpart . . . . . . . . . . . . . . . . . 250
sign . . . . . . . . . . . . . . . . . . 138, 351 subfield . . . . . . . . . . . . . . . . . . . 418
signunits . . . . . . . . . . . . . . . . . 360 subgroup . . . . . . . . . . . . . . . . . . . 348
simplify . . . . . . . . . . . . . 59, 129, 158 subgroup . . . . . . . . . . . . . . . . . . 79
sin . . . . . . . . . . . . . . . . . . . . . 306 subgrouplist . . . . . . . . . . 79, 362, 440
sinc . . . . . . . . . . . . . . . . . . . . . 306 subgrouplist0 . . . . . . . . . . . . . . . 440
sinh . . . . . . . . . . . . . . . . . . . . . 306 subresultant algorithm . . . . . 200, 240, 245
sizebyte . . . . . . . . . . . . . . . . . . 158 subst . . . . . . . . . . . . . . . . . 251, 255
sizedigit . . . . . . . . . . . . . . . . . 159 substpol . . . . . . . . . . . . . . . . . . 252
Smith normal form . . . . . 79, 274, 351, 354, substvec . . . . . . . . . . . . . . . . . . 252
359, 393, 416 sum . . . . . . . . . . . . . . . . . . . . . 132
SNF generators . . . . . . . . . . . . . . . 170 sum . . . . . . . . . . . . . . . . . . . . . 332
solve . . . . . . . . . . . . . . 102, 103, 332 sumalt . . . . . . . . . . . 323, 333, 334, 346
solvestep . . . . . . . . . . . . . . . . . 332 sumalt2 . . . . . . . . . . . . . . . . . . . 334
sopath . . . . . . . . . . . . . . . . . . . 129 sumdedekind . . . . . . . . . . . . . . . . 219
sqr . . . . . . . . . . . . . . . . . . . . . 306 sumdigits . . . . . . . . . . . . . . . . . 219
sqrt . . . . . . . . . . . . . . . . . . . . . 307 sumdigits0 . . . . . . . . . . . . . . . . . 219
sqrtint . . . . . . . . . . . . . . . . 218, 219 sumdiv . . . . . . . . . . . . . . . . 218, 334
sqrtint0 . . . . . . . . . . . . . . . . . . 218 sumdivk . . . . . . . . . . . . . . . . . . . 218
sqrtn . . . . . . . . . . . . . . . . . . . . 307 sumdivmult . . . . . . . . . . . . . . . . . 334
sqrtnint . . . . . . . . . . . . . . . 218, 219 sumdivmultexpr . . . . . . . . . . . . . . 334
stack . . . . . . . . . . . 61, 125, 126, 129, 130 sumeulerrat . . . . . . . . . . . . . . . . 334
stacksize . . . . . . . . . . . . . . . . . 46 sumformal . . . . . . . . . . . . . . 252, 253
Stark units . . . . . . . . . . . . . . 216, 370 suminf . . . . . . . . . . . . . 333, 334, 335
startup . . . . . . . . . . . . . . . . . . . 62 suminf_bitprec . . . . . . . . . . . . . . 335
Steinitz class . . . . . . . . . . . . . . . . 439 sumnum . . . . . . . . . . . . . . . . 335, 338
Stirling number . . . . . . . . . . . . . . . 168 sumnumap . . . . . . . . . . . . . . . 338, 340
stirling . . . . . . . . . . . . . . . 168, 169 sumnumapinit . . . . . . . . . . . . 340, 341
stirling1 . . . . . . . . . . . . . . . . . 169 sumnuminit . . . . . . . . . . . . . . . . . 341
stirling2 . . . . . . . . . . . . . . . . . 169 sumnumlagrange . . . . . . . . . . . 341, 342
Str . . . . . . . . . . . . . . . . . 49, 50, 144 sumnumlagrangeinit . . . . . . . . 342, 343
Strchr . . . . . . . . . . . . . . . . . . . 85 sumnummonien . . . . . . . . . 335, 338, 343
strchr . . . . . . . . . . . . . . . . 111, 145 sumnummonien0 . . . . . . . . . . . . . . . 344
Strexpand . . . . . . . . . . . . . . . . . 85 sumnummonieninit . . . . . . . . . . 344, 345
strexpand . . . . . . . . . . . . . . . 85, 112 sumnumrat . . . . . . . . . . . . . . 345, 346
strftime . . . . . . . . . . . . . . . . 58, 127 sumpos . . . . . . . . . . . . . . . . . . . 346
strictargs . . . . . . . . . . . . . . . 43, 129 sumpos2 . . . . . . . . . . . . . . . . . . . 346
strictmatch . . . . . . . . . . . . . . . . 129 suppl . . . . . . . . . . . . . . . . . . . . 276

631
sylvestermatrix . . . . . . . . . . . . . 250 t_LIST . . . . . . . . . . . . . . . . . . 7, 27
symmetric powers . . . . . . . . . . . . . 250 t_MAT . . . . . . . . . . . . . . . . . . . 7, 25
system . . . . . . . . . . . . 50, 98, 113, 128 t_PADIC . . . . . . . . . . . . . . . . . . 7, 20
t_POL . . . . . . . . . . . . . . . . . . . 7, 21
T t_POLMOD . . . . . . . . . . . . . . . . . 7, 21
t_QFI . . . . . . . . . . . . . . . . . . . 7, 23
t2 . . . . . . . . . . . . . . . . . . . . . . 351
t_QFR . . . . . . . . . . . . . . . . . . . 7, 23
Tamagawa number . . . . . . . . . . 482, 490
t_QUAD . . . . . . . . . . . . . . . . . . 7, 20
tan . . . . . . . . . . . . . . . . . . . . . 308
t_REAL . . . . . . . . . . . . . . . . . . 7, 17
tanh . . . . . . . . . . . . . . . . . . . . . 308
t_RFRAC . . . . . . . . . . . . . . . . . . 7, 23
Taniyama-Shimura-Weil conjecture . . . . 473
t_SER . . . . . . . . . . . . . . . . . . . 7, 22
tate . . . . . . . . . . . . . . . . . . . . . 470 t_STR . . . . . . . . . . . . . . . . . . . 7, 27
tayl . . . . . . . . . . . . . . . . . . . . . 253 t_VEC . . . . . . . . . . . . . . . . . . . 7, 23
Taylor series . . . . . . . . . . . . . . . . 132 t_VECSMALL . . . . . . . . . . . . . . . . 7, 27
taylor . . . . . . . . . . . . . . . . . . . 253
teich . . . . . . . . . . . . . . . . . . . . 309 U
teichmuller . . . . . . . . . . . . . 308, 309
teichmullerinit . . . . . . . . . . . . . 309 ulimit . . . . . . . . . . . . . . . . . . . 46
tex2mail . . . . . . . . . . . . . . . 125, 126 unexport . . . . . . . . . . . . . . . . . . 115
TeXstyle . . . . . . . . . . . . 113, 121, 124 unexportall . . . . . . . . . . . . . . . . 115
theta . . . . . . . . . . . . . . . . . . . . 309 uninline . . . . . . . . . . . . . . . . . . 115
thetanullk . . . . . . . . . . . . . . . . . 309 until . . . . . . . . . . . . . . . . . . . . 85
threadsize . . . . . . . . . . . . . . . . . 129 user defined functions . . . . . . . . . . . 38
threadsizemax . . . . . . . . . . . . . . . 130
thue . . . . . . . . . . . . . . . . . . 253, 254 V
thueinit . . . . . . . . . . . . 253, 254, 255 valuation . . . . . . . . . . . . . . . . . 159
time expansion . . . . . . . . . . . . . . 58 van Hoeij . . . . . . . . . . . . . . . 371, 404
timer . . . . . . . . . . . . . . . . . . . . 130 varhigher . . . . . . . 34, 159, 160, 161, 162
trace . . . . . . . . . . . . . . . . . . . . 289 variable (priority) . . . . . . . . . . . 21, 34
Trager . . . . . . . . . . . . . . . . . . . . 371 variable scope . . . . . . . . . . . . . . . . 36
trap . . . . . . . . . . . . . . . . . . . 50, 114 variable . . . . . . . . . . . . . . . . . 21, 32
trap0 . . . . . . . . . . . . . . . . . . . . 114 variable . . . . . . . . . . . . . . . . 34, 160
trueeta . . . . . . . . . . . . . . . . . . . 299 variables . . . . . . . . . . . . . . . . . 161
trunc0 . . . . . . . . . . . . . . . . . . . 159 variables_vec . . . . . . . . . . . . . . . 161
truncate . . . . . . . 152, 153, 159, 235, 248 variables_vecsmall . . . . . . . . . . . 161
tschirnhaus . . . . . . . . . . . . . . . . 424 varlower . . . . . . . . . . . . 159, 162, 163
tu . . . . . . . . . . . . . . . . . . . . . . 351 Vec . . . . . . . . . . . . . 23, 24, 26, 27, 144
tutorial . . . . . . . . . . . . . . . . . . . 60 veceint1 . . . . . . . . . . . . . . . . . . 298
type . . . . . . . . . . . . . . . . . . . . . 114 vecextract . . . . . . . . . . . . . . 270, 289
type0 . . . . . . . . . . . . . . . . . . . . 114 vecmax . . . . . . . . . . . . . . . . . . . 138
t_CLOSURE . . . . . . . . . . . . . . . . 7, 28 vecmax0 . . . . . . . . . . . . . . . . . . . 138
t_COL . . . . . . . . . . . . . . . . . . . 7, 23 vecmin . . . . . . . . . . . . . . . . 138, 139
t_COMPLEX . . . . . . . . . . . . . . . . 7, 20 vecmin0 . . . . . . . . . . . . . . . . . . . 139
t_ERROR . . . . . . . . . . . . . . . . . . 7, 28 vecprod . . . . . . . . . . . . . . . . . . . 290
t_FFELT . . . . . . . . . . . . . . . . . . 7, 19 Vecrev . . . . . . . . . . . . . . . . . 24, 145
t_FRAC . . . . . . . . . . . . . . . . . . 7, 19 vecsearch . . . . . . . . . . . 290, 291, 292
t_INFINITY . . . . . . . . . . . . . . . . 7, 29 vecsmall . . . . . . . . . . . . . . . . . . . . 7
t_INT . . . . . . . . . . . . . . . . . . . 7, 17 Vecsmall . . . . . . . . . . . . . . . . 27, 145
t_INTMOD . . . . . . . . . . . . . . . . . 7, 18 vecsort . . . . . . . . . . . . . . . . 290, 291

632
vecsort0 . . . . . . . . . . . . . . . . . . 292 zetamultdual . . . . . . . . . . . . 312, 313
vecsum . . . . . . . . . . . . . . . . . . . 292 zetamult_interpolate . . . . . . . . . . 311
vecthetanullk . . . . . . . . . . . . . . . 309 Zideallog . . . . . . . . . . . . . . . . . 388
vecthetanullk_tau . . . . . . . . . . . . 309 zk . . . . . . . . . . . . . . . . . . . . . . 351
vector . . . . . . . . . . . . . . . . . . . . . 8 zkst . . . . . . . . . . . . . . . . . . . . . 351
vector . . . . . . . . . . . . . . 24, 292, 293 ZM_det . . . . . . . . . . . . . . . . . . . 263
vectorsmall . . . . . . . . . . . . . . . . 293 ZM_gauss . . . . . . . . . . . . . . . . . . 275
vectorv . . . . . . . . . . . . . . . . . 24, 293 ZM_ker . . . . . . . . . . . . . . . . . . . 271
version number . . . . . . . . . . . . . . . 61 znchar . . . . . . . . . . . . . . . . 219, 220
version . . . . . . . . . . . . . . . . . . . 115 zncharconductor . . . . . . . . . . . . . 220
Vi . . . . . . . . . . . . . . . . . . . . . . 64 znchardecompose . . . . . . . . . . 220, 221
znchargauss . . . . . . . . . . . . . . . . 221
W zncharinduce . . . . . . . . . . . . 221, 222
warning . . . . . . . . . . . . . . . . . . . 116 zncharisodd . . . . . . . . . . . . . . . . 222
weber . . . . . . . . . . . . . . . . . . . . 309 znchartokronecker . . . . . . . . . 222, 223
weber0 . . . . . . . . . . . . . . . . . . . 309 znchartoprimitive . . . . . . . . . . . . 223
weberf . . . . . . . . . . . . . . . . . . . 309 znconreychar . . . . . . . . . 223, 225, 518
weberf1 . . . . . . . . . . . . . . . . . . . 309 znconreyconductor . . . . . . . . . . . . 225
weberf2 . . . . . . . . . . . . . . . . . . . 309 znconreyexp . . . . . . . . . . . . . 225, 226
Weierstrass ℘-function . . . . . . . . . . . 512 znconreylog . . . . . . . . 224, 226, 227, 518
Weierstrass equation . . . . . . . . . . . . 469 zncoppersmith . . . . . . . . . . . . 227, 229
Weil curve . . . . . . . . . . . . . . . . . 508 znlog . . . . . . . . . 197, 229, 230, 388, 490
whatnow . . . . . . . . . . . . . . . . . 50, 116 znlog0 . . . . . . . . . . . . . . . . . . . 230
while . . . . . . . . . . . . . . . . . . . . 85 znorder . . . . . . . . . . . . . . . . . . . 230
write . . . . . . . . . . . . . . 50, 58, 61, 116 znprimroot . . . . . . . . . . . . . . . . . 230
write1 . . . . . . . . . . . . . . . . . . . 116 znstar . . . . . . . . . . . . . . . . 229, 230
writebin . . . . . . . . . . . . . . . . . . 116 znstar0 . . . . . . . . . . . . . . . . . . . 231
writetex . . . . . . . . . . . . . . . . . . 117 Zp_appr . . . . . . . . . . . . . . . . . . . 236

X
x[,n] . . . . . . . . . . . . . . . . . . . . 149
x[m,n] . . . . . . . . . . . . . . . . . . . 149
x[m,] . . . . . . . . . . . . . . . . . . . . 149
x[n] . . . . . . . . . . . . . . . . . . . . . 149

Z
Zassenhaus . . . . . . . . . . . . . . . . . 235
zbrent . . . . . . . . . . . . . . . . . . . 332
zell . . . . . . . . . . . . . . . . . . . . . 505
zero . . . . . . . . . . . . . . . . . . . . . . 9
zeropadic . . . . . . . . . . . . . . . . . 231
zeroser . . . . . . . . . . . . . . . . . . . 231
zeta function . . . . . . . . . . . . . . . . 43
zeta . . . . . . . . . . . . . . . . . . . . . 309
zetahurwitz . . . . . . . . . . . . . . . . 310
zetamult . . . . . . . . . . . . . . . . . . 311
zetamultall . . . . . . . . . . . . . 311, 312
zetamultconvert . . . . . . . . . . . . . 312

633