A Problem Course in Mathematical Logic: Stefan Bilaniuk

A Problem Course
in
Mathematical Logic
Volume II
Computability and Incompleteness
Stefan Bilaniuk
Author address:
Department of Mathematics
Trent University
Peterborough, Ontario
Canada K9J 7B8
E-mail address: [email protected]
1991 Mathematics Subject Classification. 03.
A Problem Course in Mathematical Logic

Volume II: Computability and Incompleteness
Version 1.3
Copyright 1994-1997
c by Stefan Bilaniuk.
Abstract. This is the Volume II of a text for a problem-oriented

undergraduate course in mathematical logic. It covers the basics
of computability, using Turing machines and recursive functions,
and Gödel’s Incompleteness Theorem, and could be used for a one-
semester course on these topics. Volume I, Propositional and First-
Order Logic, covers the basics of these topics through the Sound-
ness, Completeness, and Compactness Theorems.
Information on availability and the conditions under which this
book may be used and reproduced are given in the preface.
This book was typeset using LATEX, using the AMS-LATEX and
AMSFonts packages of the American Mathematical Society.
Contents
Preface v
Introduction 1
Computability 5
Chapter 10. Turing Machines 7
Chapter 11. Variations and Simulations 13
Chapter 12. Universal Turing Machines and the Halting Problem 17
Chapter 13. Computable and Non-Computable Functions 25
Chapter 14. Primitive Recursive Functions 29
Chapter 15. Recursive Functions 35
Incompleteness 41
Chapter 16. Preliminaries 43
Chapter 17. Coding First-Order Logic 45
Chapter 18. Defining Recursive Functions In Arithmetic 49
Chapter 19. The Incompleteness Theorem 53
Hints 57
Chapter 10. Hints 59
iii
iv CONTENTS

Bibliography 79
Index 81
Preface
This book is intended to be the basis for a problem-oriented full-year

course in mathematical logic for students with a modicum of mathe-
matical sophistication. Volume II covers the basics of computability,
using Turing machines and recursive functions, and incompleteness.
It could be used for a one-semester course on these topics. Volume
I covers the basics of propositional and first-order logic through the
Soundness, Completeness, and Compactness Theorems, plus some ma-
terial on applications of the Compactness Theorem; it could also be
used as for a one-semester course on these topics. However, part of
Volume II, particularly the chapters on incompleteness, assume some
familiarity with the basics of first-order logic.
In keeping with the modified Moore-method, this book supplies
definitions, problems, and statements of results, along with some ex-
planations, examples, and hints. The intent is for the students, indi-
vidually or in groups, to learn the material by solving the problems
and proving the results for themselves. Besides constructive criticism,
it will probably be necessary for the instructor to supply further hints
or direct the students to other sources from time to time. Just how
this text is used will, of course, depend on the instructor and students
in question. However, it is probably not appropriate for a conventional
lecture-based course nor for a really large class.
The material presented here is somewhat stripped-down. Various
concepts and topics that are often covered in introductory courses on
computability are given very short shrift or omitted entirely, among
them models of computation other than Turing machines and recursive
functions, formal languages, and computational complexity.1 Instruc-
tors might consider having students do projects on additional material
if they wish to cover it. It might also be expedient, in view of the
somewhat repetitive nature of devising Turing machines and recursive
functions for various purposes, to be selective about the problems the
students are required to do or to divide them up among the students.
1
Future versions of both volumes may include more – or less! – material. Feel
free to send suggestions, corrections, criticisms, and the like — I’ll feel free to ignore
them or use them.
v
vi PREFACE
Acknowledgements. Various people and institutions deserve the

credit for this text: All the people who developed the subject. My
teachers and colleagues, especially Gregory H. Moore, whose math-
ematical logic course convinced me that I wanted to do the stuff.
The students at Trent University who suffered, suffer, and will suf-
fer through assorted versions of this text. Trent University and the
taxpayers of Ontario, who paid my salary. Ohio University, where I
spent my sabbatical in 1995–96. All the people and organizations who
developed the software and hardware with which this book was pre-
pared. Anyone else I’ve missed.
Any blame properly accrues to the author.
Conditions. This book may be freely transmitted, stored, copied,
and used until 31 December, 1998, subject to the following restrictions:2
1. It may not be modified in any way without the express written
permission of the author.
2. It may not be sold at a profit, but only to recover the cost of
reproduction.
3. After 31 December, 1998, it may no longer be reproduced, stored,
or transmitted in any form without the express written permis-
sion of the author, except that printed copies existing as of 31
December, 1998, may be retained and used after this date.
The reason for the time-limit is that I hope to improve this book
and make a new version available.3
Availability. The URL of the home page for A Problem Course

In Mathematical Logic, with links to LATEX and PostScript source files
of the latest release of both volumes, is:
• http://www.trentu.ca/academic/math/sb/misc/pcml.html
A text-only information file and LATEX and PostScript source files of
the latest release of both volumes can be accessed by anonymous ftp
at:
• ftp://ftp.trentu.ca/pub/sbilaniuk/pcml/
Please note that in addition to LATEX you will need the AMS-LATEX
and AMSFonts packages to typeset and print either volume.
If you have any problems, feel free to contact the author at the
addresses given on the title page, preferably by e-mail, for assistance
or even to ask for a paper copy.
2
If you violate these restrictions, I will be flattered, but you will still be in the
wrong.
3
Who knows, it might even find a publisher . . .
PREFACE vii
Author’s Opinion. It’s not great, but the price is right!

viii PREFACE
Introduction
The Entscheidungsproblem. Recall that a logic satisfies the

Completeness Theorem if, whenever the truth of a set of sentences
Σ implies the truth of a sentence ϕ, there is a deduction of ϕ from
Σ. Propositional and first-order logics both satisfy the Completeness
Theorem, though second- and higher-order logics do not. In the case
of propositional logic, the Completeness Theorem leads to a rote pro-
cedure for determining whether Σ ` ϕ or not, so long as Σ is finite:
write out a complete truth table for all of Σ together with ϕ and check
whether every assignment that makes every sentence of Σ true also
makes ϕ true. It is natural to ask whether something of the sort can
be done for first-order logic. If so, it might be very useful: since most
of mathematics can be formalized in first-order logic, such a method
would have the obvious use of putting mathematicians out of business
. . . This question is the general Entscheidungsproblem 4 for first-order
logic:
Entscheidungsproblem. Given a set Σ of formulas of a first-
order language L and a formula ϕ of L, is there an effective method
for determining whether or not Σ ` ϕ?
Of course, the statement of the problem begs the question of what
“effective” is supposed to mean here. In this volume we’ll explore
two formalizations of the notion of “effective method”, namely Tur-
ing machines and recursive functions, and then use these to answer
the Entscheidungsproblem for first-order logic. The answer to the gen-
eral problem is negative, though decision procedures do exist for some
particular first-order languages and sets Σ.
Historically, the Entscheidungsproblem arose out of David Hilbert’s
scheme to secure the foundations of mathematics by axiomatizing math-
ematics in first-order logic and showing that the axioms do not give rise
to any contradictions. It did so in two related ways. First, given some
plausible set of axioms, it is necessary to show that they do not lead
to a contradiction, such as α ∧ (¬α). Second, it is desirable to know
4
Entscheidungsproblem ≡ decision problem.
1
2 INTRODUCTION
whether such a set of axioms is complete; i.e. given any sentence ϕ of

the language, that the axioms either prove or disprove ϕ.
In the course of trying to find a suitable formalization of the no-
tion of “effective method”, mathematicians developed several different
abstract models of computation in the 1930’s, including recursive func-
tions, λ-calculus, Turing machines, and grammars. Although these
models are very different from each other in spirit and formal defini-
tion, it turned out that they were all essentially equivalent in what they
could do. This suggested the (empirical!) principle:
Church’s Thesis. A function is effectively computable in princi-
ple in the real world if and only if it is computable by (any) one of the
abstract models mentioned above.
Of course, this is not a mathematical statement . . . We will study
Turing machines and recursive functions, and then use this knowledge
to formulate and answer a more precise version of the general Entschei-
dungsproblem for first-order logic.
The development of the theory of computation actually began be-
fore the development of electronic digital computers. In fact, the com-
puters and programming languages we use today owe much to the ab-
stract models of computation which preceded them. For two, the stan-
dard von Neumann architecture for digital computers was inspired by
Turing machines and the programming language LISP borrows much
of its structure from λ-calculus.
Approach. This book supplies definitions and statements of re-
sults, plus some explanations and a number of problems and examples,
but no proofs of the results. The hope is that you, gentle reader, will
learn the material presented here by solving the problems and proving
the results for yourself. Brief hints are supplied for almost all of the
problems and results, but if these do not suffice, you should consult
your peers, your instructor, or other texts.
Prerequisites. In principle, little is needed by way of prior math-
ematical knowledge to define and prove the basic facts about com-
putability. Some knowledge of the natural numbers and a little set the-
ory suffices. The material leading up to the Incompleteness Theorem
— the resolution of the general Entscheidungsproblem for first-order
logic — does require grounding in first-logic, such as that provided in
Volume I, as well as in computability.
What really is needed to get anywhere with all of the material
developed here is competence in handling abstraction and proofs, in-
cluding proofs by induction. The experience provided by a rigorous
INTRODUCTION 3
introductory course in algebra, analysis, or discrete mathematics ought

to be sufficient. Some problems and examples draw on concepts from
other parts of mathematics; students who are not already familiar with
these should consult texts in the appropriate subjects for the necessary
definitions.
Other Sources and Further Reading. [3], [5], [7], and [8] are
texts which go over at least some of the material, while [1] is a good
if terse reference for more advanced material. Entertaining accounts of
much of the material may be found in [6] and [9]; the latter discusses
the possibility that Church’s Thesis may not be true. Many of the
original sources for the material in this volume can be found in the
anthology [4].
4 INTRODUCTION
Computability
CHAPTER 10
Turing Machines
Of the various ways to formalize the notion an “effective method”,

the most commonly used are the simple abstract computers called Tur-
ing machines, which were introduced more or less simultaneously by
Alan Turing and Emil Post in 1936.1 Like most real-life digital com-
puters, Turing machines have two main parts, a processing unit and
a memory (which doubles as the input/output device), which we will
consider separately before seeing how they interact. The memory can
be thought of as a tape, without end in one direction, which is divided
up into cells like the frames of a movie. The Turing machine proper is
the processing unit. It has a scanner or “head” which can read from or
write to a single cell of the tape, and which can be moved to the left
or right one cell at a time.
Tapes. The first thing we have to do in describing a Turing ma-
chine is to specify what symbols it is able to read and write on its
tape.
Definition 10.1. An alphabet is a non-empty finite set Σ, the el-
ements of which are called symbols, such that 0 ∈
/ Σ.
The reason we don’t allow Σ to contain 0 is that we will use 0 to
mark all the otherwise blank cells on a tape.
Definition 10.2. Given an alphabet Σ, a tape (with entries from
Σ) is an infinite sequence
a = a0 a1 a2 a3 . . .
such that for each integer i the cell ai ∈ {0} ∪ Σ. The ith cell is said
to be blank if ai is 0, and marked if ai is a symbol from Σ.
A blank tape is one in which every cell is 0. It will be shown later on
that it is possible to restrict the alphabet to just one non-blank symbol
without essentially diminishing what a Turing machine can accomplish,
but it is usually convenient to have more symbols about when actually
devising a Turing machine for a particular task.
1
Both papers are reprinted in [4]. Post’s brief paper gives a particularly lucid
informal description.
7
8 10. TURING MACHINES
Example 10.1. A blank tape looks like:

000000000000000000000000 · · ·
The 0th cell is the leftmost one, cell 1 is the one immediately to the
right, cell 2 is the one immediately to the right of cell 1, and so on.
Letting our alphabet be Σ = { 1, x, y }, the following is a slightly
more exciting tape:
010xx1011x1y01yyx000000000000000 · · ·
In this case, cell 1 contains a 1, as do cells 5, 7, 8, 10, and 13; cells 3,
4, 9, and 16 each contain an x; cells 11, 14, and 15 each contain a y;
and all the rest contain a 0.
Problem 10.1. Write down tapes satisfying the following; you may
use any appropriate alphabets.
1. Entirely blank except for cells 3, 12, and 20.
2. Entirely marked except for cells 0, 2, and 3.
3. Entirely blank except for a block of five consecutive cells just to
the right of cell 0.
4. Entirely blank except that 1025 is written out in binary just to
the right of cell 2.
To keep track of which cell the Turing machine’s scanner is at, plus
some other information, we will usually attach additional information
to our description of the tape.
Definition 10.3. A tape position is a triple (i, s, a), where i and s
are natural numbers with s > 0, and a is a tape. Given a tape position
(i, s, a), we will refer to cell i as the scanned cell and to s as the state.
The number s mentioned above will be used to keep track of which
instruction the Turing machine is to execute next.
Conventions for tapes. Unless stated otherwise, we will assume

that all but finitely many cells of any given tape are blank, and that any
cells not explicitly described or displayed are blank. We will usually
depict as little of a tape as possible and omit the · · · s we used above.
Thus
010xx1011x1y01yyx
represents the tape given in the Example 10.1. In many cases we will
also use z n to abbreviate n consecutive copies of z, so the same tape
could be represented by
010x2 1012 x1y01y 2 x .
10. TURING MACHINES 9
Similarly, if σ is a finite sequence of elements of Σ ∪ {0}, we may write

σ n for the sequence consisting of n copies of σ stuck together end-to-
end. For example, (010)3 is 010010010.
In displaying tape positions we will usually underline the scanned
cell and write s to the right of the tape. For example, we would display
the tape position using the tape from Example 10.1 with cell 4 being
scanned and state 2 as follows:
010xx1012 x1y01y 2 x : 2
Problem 10.2. Using the tapes you gave in the corresponding part
of Problem 10.1, write down tape positions satisfying the following con-
ditions.
1. Cell 7 being scanned and state 4.
Turing machines. The “processing unit” of a Turing machine is
just a finite list of specifications describing what the machine will do
in various situations. (Remember, this is an abstract computer . . . )
The formal definition may not seem to amount to this at first glance.
Definition 10.4. A Turing machine (with alphabet Σ) is a func-
tion M such that for some natural number n,
dom(M) ⊆ {1, . . . , n} × ({0} ∪ Σ)
and
ran(M) ⊆ ({0} ∪ Σ) × {−1, 1} × {1, . . . , n} .
Note that M need not be defined for all possible pairs
(s, j) ∈ {1, . . . , n} × ({0} ∪ Σ) .
We will sometimes refer to a Turing machine simply as a machine
or TM . If n ≥ 1 is least such that M satisfies the definition above, we
shall say that M is an n-state Turing machine and that {1, . . . , n} is
the set of states of M.
Intuitively, we have a processing unit which has a finite list of basic
instructions, the states, which it can execute. Given a combination of
current state and the symbol marked in the cell of the tape currently
being scanned that it is equipped to handle, the processor specifies
• a symbol to be written in the currently scanned cell, overwriting
the symbol being read, then
• a move of the scanner one cell to the left or right, and then
• the next instruction to be executed.
That is, M(s, c) = (b, d, t) means that if our machine is in state s (i.e.
executing instruction number s), scanning the ith cell, and ai = c (i.e.
cell i contains c), then the machine M should
• set ai = b (i.e. write b instead of c in the scanned cell), then
• move the scanner to ai+d (i.e. move one cell left if d = −1 and
one cell right if d = 1), and then
• enter state t (i.e. go to instruction t).
If our processor isn’t equipped to handle input c for instruction s (i.e.
M(s, a) is undefined), then the computation in progress will simply
stop dead.
Example 10.2. We will usually present Turing machines in the
form of a table, with a row for each state and a column for each possible
entry in the scanned cell. Instead of −1 and 1, we will usually use L
and R when writing such tables in order to make them more readable.
Thus the table
M 0 1
1 1R2 0R1
2 0L2
defines a Turing machine M with alphabet {1} and two states such that
M(1, 0) = (1, 1, 2), M(1, 1) = (0, 1, 1), and M(2, 0) = (0, −1, 2), but
M(2, 1) is undefined. (So M has domain { (1, 0), (1, 1), (2, 0) } and
range { (1, 1, 2), (0, 1, 1), (0, −1, 2) }.) If the machine M were faced
with the tape position
01001111 : 1 ,
it would, being in state 1 and scanning a cell containing 0,
• write a 1 in the scanned cell,
• move the scanner one cell to the right, and
• go to state 2.
This would give the new tape position
01011111 : 2 .
Since M doesn’t know what to do on input 1 in state 2, the computation
could go no further.
Problem 10.3. In each case, give the table of a Turing machine
M meeting the given requirement.
1. M has alphabet {x, y, z} and has three states.
2. M has alphabet {1} and changes 0 to 1 and vice versa in any
cell it scans.
3. M is as simple as possible. How many possibilities are there
here?
10. TURING MACHINES 11
Computations. Informally, a computation is a sequence of actions

of a machine M on a tape according to the rules above, starting with
instruction 1 and the scanner at cell 0 on the given tape. A computation
ends (or halts) when and if the machine encounters a tape position
which it does not know what to do in or runs off the left end of the tape.
(If it never does either, the computation will never end — not quite
like real computers, Turing machines succeed only when they crash!)
The formal definition makes all this seem much more formidable.
Definition 10.5. Suppose M is a Turing machine. Then:
• If p = (i, s, a) is a tape position using the same alphabet as M
and M(s, ai ) = (b, d, t) is defined, then M(p) = (i + d, t, a0 ) is
the successor tape position, where a0i = b and a0j = aj for j 6= i.
• A partial computation with respect to M is a sequence p1 p2 . . . pk
of tape positions such that p`+1 = M(p` ) for each ` < k.
• A partial computation p1 p2 . . . pk with respect to M is a compu-
tation (with respect to M) with input tape a if p1 = (0, 1, a) and
M(pk ) is undefined. The output tape of the computation is the
tape of the final tape position pk .
Note that a computation must have only finitely many steps.
Example 10.3. Let’s see the machine M of Example 10.2 perform
a computation. Our input tape will be a = 1100, that is, the tape
which is entirely blank except that a0 = a1 = 1. The initial tape
position of the computation of M with input tape a is then
1100 : 1 .
The subsequent steps in the computation are:
0100 : 1
0000 : 1
0010 : 2
0010 : 2
We leave it to the reader to check that this is indeed a partial com-
putation with respect to M. Since M(2, 1) is undefined the process
terminates at this point and this partial computation is indeed a com-
putation.
Problem 10.4. Give the (partial) computation of the Turing ma-
chine M of Example 10.2 when the input tape is:
1. 00.
2. 110.
3. The tape with all cells marked with 1s and cell 5 being scanned.
Problem 10.5. For which possible input tapes does the partial com-
putation of the Turing machine M of Example 10.2 eventually termi-
nate?
Problem 10.6. Find a Turing machine that (eventually!) fills a
blank input tape with the pattern xyyz3xyyz3xyyz3 . . . .
Problem 10.7. Find a Turing machine with alphabet {$, @}that
never halts, no matter what is on the tape.
Example 10.4. The Turing machine P given below is intended to
produce output 01m on input 01n 01m whenever n, m > 0.
P 0 1
1 0R2
2 1R3
3 0R4 0R3
4 0R4 0R5
5 0L8 1L6
6 0L6 1R7
7 1R4
8 0L8 1L9
9 1L9
Trace P ’s computation on, say, input 013 014 to see how it works.
Problem 10.8. In each case, find a Turing machine (with the al-
phabet of your choice) that:
1. Halts with output 014 on input 0.
2. Halts with output 01n 0 on input 00n 1.
3. Halts with output 012n on input 01n .
4. Halts with output 0(10)n on input 01n .
5. Halts with output 01m 01n on input 01m 0k 1n , if n, m, k > 0.
6. Halts with output 01m 01n 01k on input 01n 01k 01m , if n, m, k > 0.
7. Halts with output 01m 01n 01k 01m 01n 01k on input 01m 01n 01k , if
n, m, k > 0.
8. On input 01m 01n , where m, n > 0, halts with output 01 if m 6= n
and output 011 if m = n.
It is quite possible to find such machines with just {1} as an alphabet.
Note. It doesn’t matter what the machine you define in each case
does on other inputs, so long as it does the right thing on the given
one(s).
CHAPTER 11
Variations and Simulations
The definition of a Turing machine given in Chapter 10 is arbitrary

in a number of ways, among them the use of an arbitrary finite al-
phabet, a single read-write scanner, and a single one-way infinite tape.
One could restrict the definition we gave by allowing
• the machine to move the scanner only to one of left or right in
each state,
• only {1} as an alphabet,
or both, among various possibilities. One could also define apparently
more powerful Turing machines by allowing the use of
• two-way infinite tapes,
• multiple tapes,
• two- and higher-dimensional tapes,
or various combinations of these, among many other possibilities. We
will construct a number of Turing machines that simulate others with
additional features; this will show that none of the modifications men-
tioned above really change what the machines can compute.
Example 11.1. Consider the following Turing machine:

M 0 1
1 1R2 0L1
2 0L2 1L1
Note that in state 1, this machine may move the scanner to either
the left or the right, depending on the contents of the cell being scanned.
We will construct a Turing machine, with alphabet {1}, that emulates
the action of M on any input, but which moves the scanner to only
one of left or right in each state. There is no problem with state 2 of
M, by the way, because in state 2 M always moves the scanner to the
left.
The basic idea is to add some states to M which replace part of the
description of state 1.
13
14 11. VARIATIONS AND SIMULATIONS
M0 0 1
1 1R2 0R3
2 0L2 1L1
3 0L4 1L4
4 0L1
This machine is just like M except that in state 1 with input 1,
instead of moving the scanner to the left and going to state 1, the
machine moves the scanner to the right and goes to the new state 3.
States 3 and 4 do nothing between them except move the scanner two
cells to the left without changing the tape, thus putting it where M
would have put it, and then entering state 1, as M would have.
Problem 11.1. Compare the computations of the machines M and
M 0 of Example 11.1 on the input tapes
1. 0
2. 011
and explain why is it not necessary to define M 0 for state 4 on input 1.
Problem 11.2. Given a Turing machine M with an arbitrary al-
phabet Σ, explain in detail how to construct a machine M 0 that simu-
lates what M does on any input, but which moves the scanner only to
one of left or right in each state.
Problem 11.3. Given a Turing machine M with an arbitrary al-
phabet Σ, explain in detail how to construct a machine N with alphabet
{1} that simulates M.
To define Turing machines with two-way infinite tapes we need
only change the definition of the tape: instead of having tapes be se-
quences a = ha0, a1, a2, . . . i indexed by N, we let them be sequences
b = h. . . , b−2, b−1 , b0, b1 , b2, . . . i indexed by Z. In defining computa-
tions for machines with two-way infinite tapes, we adopt the same
conventions that we did for machines with one-way infinite tapes, such
as the scanner starts off scanning cell 0 on the input tape. The only
real difference is that a machine with a two-way infinite tape cannot
halt by running off the left end of the tape . . .
Example 11.2. Consider the following two-way infinite tape Tur-
ing machine with alphabet {1}:
T 0 1
1 1L1 0R2
2 0R2 1L1
The biggest problem in trying to emulate T with a one-way infinite
tape Turing machine O is representing a two-way infinite tape on a
11. VARIATIONS AND SIMULATIONS 15
one-way infinite tape. To do this, we choose an alphabet for O with

malice aforethought:
{ S0 , S1 , 00 , 01 , 10 , 1
1
}
We can now represent the tape a,
. . . , a−2 , a−1 , a0, a1 , a2, . . . ,
for T by the tape a0 ,
a0
S
, a1
a−1
, a2
a−2
, ... ,
for O. In effect, this device allows us to split O’s tape into two tracks,
each of which accomodates half of the tape of T .
The key remaining idea is to split each state of T into a pair of states
for O: one for the lower track and one for the upper track. One must
take care to keep various details straight: when O changes a “cell” on
one track, it should not change the corresponding “cell” on the other
track; directions are reversed on the lower track; one has to “turn a
corner” moving past cell 0; and so on.
O 0 0
S
0
0
0
1
1
S
1
0
1
1
1 0 L1 S R3 0 L1 1 L1 S R2 0 R2 1 R2
1 1 1 1 0 0 0
2 00 R2 0S R2 00 R2 01 R2 S1 R3 10 L1 11 L1
3 01 R3 1S R3 01 R3 00 L4 S0 R2 11 R3 10 L4
4 00 L4 0S R2 00 L4 01 R3 S1 R3 10 L4 11 R3
States 1 and 3 are the upper- and lower-track versions, respectively,
of T ’s state 1; states 2 and 4 are the upper- and lower-track versions,
respectively, of T ’s state 2. We leave it to the reader to check that O
actually does simulate T . . .
Problem 11.4. Trace the (partial) computations of T , and their
counterparts for O, for each of the following input tapes for T , shown
with a bar over cell 0:
1. 0 (i.e. a blank tape)
2. 10
3. . . . 1111111 . . . (i.e. every cell marked with 1)
Problem 11.5. Explain in detail how, given a Turing machine N
with an arbitrary alphabet and a two-way infinite tape, one can con-
struct a Turing machine P with an one-way infinite tape that simulates
N.
Problem 11.6. Give a precise definition for Turing machines with
two tapes. Explain how, given any such machine, one could construct
a single-tape machine to simulate it.
16 11. VARIATIONS AND SIMULATIONS
Problem 11.7. Give a precise definition for Turing machines with

two-dimensional tapes. Explain how, given any such machine, one
could construct a single-tape machine to simulate it.
Taken together, these results mean that for the purposes of investi-
gating what can be computed in principle, we can use any of the above
variants on our definition of Turing machines without loss of generality.
CHAPTER 12
Universal Turing Machines and the Halting

Problem
In Chapter 11 we devised techniques for constructing, given a par-

ticular Turing machine (of some type), a Turing machine (of another
type) that would simulate it. We will go further and construct an uni-
versal Turing machine (sometimes referred to as an UTM ): a machine
U that, when given as input (a suitable description of) some Turing
machine M and an input tape a for M, simulates the computation of
M on input a. In effect, an universal Turing machine is a piece of
“hardware” that lets us treat Turing machines as “software”.
As a bonus, constructing such a machine will give us the tools we
will need to answer the following question:
The Halting Problem. Given a Turing machine M and an in-
put tape a, is there an effective method to determine whether or not
M eventually halts on input a?
An effective method to determine whether or not a given machine
will eventually halt on a given input — short of waiting forever! —
would be nice to have. For example, assuming Church’s Thesis is true,
such a method could let us identify computer programs which have
infinite loops before they tie computers up in knots.
An Universal Turing Machine. The first problem in trying to
build an universal Turing machine is finding a suitable way to describe
the machine which is to be simulated, as well as its input tape, on the
input tape of the universal Turing machine. We can simplify our task
somewhat by restricting our attention to simulating Turing machines
with one-way infinite tapes whose alphabet is just {1}. We lose no
real generality in doing so since, by the results in Chapter 11, such
machines can do just as much as any other type. Among the many
possible ways of describing such machines as input, the one given below
is fairly straightforward but woefully inefficient.1 Essentially, it consists
1
For an example of a different method, one could combine the methods devel-
oped in Chapter 15 of representing Turing machines and tapes by integers with the
represention of integers on a tape used in Chapter 13.
17
18 12. UNIVERSAL MACHINES & THE HALTING PROBLEM
of simply listing the table defining the Turing machine we wish to

specify.
Definition 12.1. Suppose M is a Turing machine with alphabet
{1} and m states. If 1 ≤ k ≤ m and i ∈ {0, 1} and M(k, i) = (j, d, `),
let
xM (k,i)y=11+k 0m−k 011+i 01−i 011+j 01−j 012+d 01−d 011+` 0m−` ;
if M(k, i) is not defined, just let j = d = ` = 0 above.
That is, xM(k, i)y is a string of 2m + 13 0s and 1s which represents
the (k, i)th entry of M’s table in five distinct blocks, separated by single
0s:
1. one of length m + 1 codes k, represented by k + 1 1s followed by
a padding of 0s;
2. one of length 2 codes i, represented by 11 if i = 1 and 10 if i = 0;
3. one of length 2 codes j, represented by 11 if j = 1 and 10 if
j = 0;
4. one of length 3 codes d, represented by 111 if d = 1, 100 if
d = −1, and 110 if d = 0 (i.e. if M(k, i) is undefined);
5. one of length m + 1 codes `, represented by ` + 1 1s followed by
a padding of 0s.
Definition 12.2. Suppose M is a Turing machine with alphabet
{1} and m states. The representation of M is
xM y=031m+1 03 xM (1,0)y02xM (1,1)y02xM (2,0)y02...02 xM (m,0)y02xM (m,1)y03 .
The representation of the machine M, xMy, then consists of an

initial string of three 0s, followed by m+1 1s giving the number of states
of M, three more 0s, the representations of the entries of the table of
M — including the empty ones! — listed in order and separated by
pairs of 0s, and a final string of three 0s.
Example 12.1. The representation of the Turing machine
E 0 1
1 0R2 ,
2 1L1 0R1
is
03 13 03 12 0010010013013 02 12 001201001200102 02 13 0100120102012 0 02 13 012 010013012 0 03 .
Problem 12.1. Pick a Turing machine different from the one in

Example 12.1 and give its representation.
Problem 12.2. Give a table for the Turing machine whose repre-
sentation is
0001100011010011011101100110110100111011000 .
12. UNIVERSAL MACHINES & THE HALTING PROBLEM 19
Problem 12.3. How many possible representations are there for a

given Turing machine?
Problem 12.4. Devise a more efficient way of representing Turing
machines than that given in Definition 12.2.
We now have at least one way of completely describing a given
Turing machine M to another Turing machine. Of course, we have not
yet considered how this description would actually be used, but we first
need to find a way to describe M’s input anyway. The most naive way
to do this is to simply have the input for M follow the description of M.
Unfortunately, this won’t do because we will need to keep track which
cell is being scanned and what the current state is while simulating
M. To solve this problem, we’ll go whole hog and describe not just the
tape a that M is reading but a complete tape position (i, s, a). (Recall
that i is an integer specifying the currently scanned cell, s is an integer
specifying which state M is in, and a is the tape.)
Definition 12.3. Suppose (i, s, a) is the tape position of an m-
state Turing machine with alphabet{1}. For each ` ∈ N, let
(
0 ` 6= i (i.e. cell ` is not being scanned)
c` =
1 ` = i (i.e. cell ` is the scanned cell).
Let n be the least positive integer such that ak = 0 for all k > n, if
such exists. If so, the representation of (i, s, a) is the finite sequence
x(i, s, a)y = 03 1s+1 0m−s 03 1c0 a0 1c1 a11c2 a2 . . . 1cn an 03 ;
otherwise, the representation of (i, s, a) is the infinite sequence
x(i, s, a)y = 03 1s+1 0m−s 03 1c0 a0 1c1 a11c2 a2 . . . .
That is, each cell of a is described by a triple of cells in the represen-
tation of a. The first cell of this triple is simply a marker, one indicates
whether or not the cell of a being described is the one being scanned
by M, and the last gives the content of the cell being described. The
representation of a tape position (i, s, a) then consists of three 0s, a
block of length m + 1 consisting of s + 1 1s padded out by m − s 0s,
another three 0s, the triples coding the cells of a, and, possibly, a final
three 0s to mark the end of the code. Since we assume tapes to have
all but finitely many cells blank unless stated otherwise, almost all of
the representations of tape positions we will encounter will be finite.
Example 12.2. Consider the tape position (2, 3, a) for a 5-state
Turing machine, where a = 01101. The representation of this tape
position is
000 111100 000 100 101 111 100 101 000 .
Note that the only use of the 1 at the beginning of the represen-
tation of each cell is to mark the fact that a cell is being represented:
moving a scanner along the representation of a, one can identify the be-
ginning and end (if any) of the representation because one encounters
a 0 instead of a 1 marking the representation of another cell.
Problem 12.5. What are the representations of the following tape
positions for a 5-state Turing machine?
1. (0, 4, 11111)
2. (3, 1, 101011)
Problem 12.6. What tape positions are represented by the follow-
ing?
1. 0001100000110100100101101100100101100000
2. 00011111000101111101101101101101 . . .
Problem 12.7. Devise a more efficient way of representing tape
positions than that given in Definition 12.3.
We can now define our representation for M together with a tape
position (i, s, a). Except for inserting some extra 0s, this just amounts
to xMy followed by x(i, s, a)y.
Definition 12.4. Suppose M is an m-state Turing machine with
alphabet {1} and (i, s, a) is a tape position for M (so 1 ≤ s ≤ m). Then
the representation of the machine M together with the tape position
(i, s, a) is the sequence
xMy03m+33 x(i, s, a)y .
The 3m+33 0s interpolated in the middle are intended to be used for
“scratch space” in the simulation of M by the universal Turing machine
we will construct in the following series of problems. Depending on
how these are solved, it may be possible to reduce — or necessary to
increase! — that 3m + 33.
Note. The statements of Problems 12.8–12.13 below assume the
use of the representation schemes given in Definitions 12.2–12.4. If you
would rather use the methods you devised in Problems 12.4 and 12.7,
by all means adapt the statements of Problems 12.8–12.13 accordingly.
You may find it convenient to give the machines you build in these
problems alphabets other than {1}, but one need not do so. (Why
doesn’t it really matter if one does?)
Problem 12.8. Suppose M is an m-state Turing machine with al-
phabet {1}, (i, s, a) is a tape position for M, and d = ±1. Find a
Turing machine H which on input

0xMy03m+30 12+d 01−d x(i, s, a)y
halts with output
0xMy03m+30 12+d 01−d x(i + d, s, a)y .
H should extend the representation of a if necessary.
Problem 12.9. Suppose M is an m-state Turing machine with al-
phabet {1}, (i, s, a) is a tape position for M. Find a Turing machine
R which on input
0xMy03m+33 x(i, s, a)y
halts with output
0xMy03m+31 11+c 01−c x(i, s, a)y ,
where c = ai .
Problem 12.10. Suppose M is an m-state Turing machine with
alphabet {1}, (i, s, a) is a tape position for M, and j ∈ {0, 1}. Find a
Turing machine W which on input
0xMy03m+31 11+j 01−j x(i, s, a)y
halts with output
0xMy03m+31 11+j 01−j x(i, s, a0)y ,
where a0 is identical to a except that a0i = j.
alphabet {1}, (i, s, a) is a tape position for M, and c ∈ {0, 1}. Find a
Turing machine E which on input
0xMy03m+31 11+c 01−c x(i, s, a)y
halts with output
0xMyxM(s, c)y0m+18 11+c 01−c x(i, s, a)y .
alphabet {1} and (i, s, a) is a tape position for M. Find a Turing
machine S which on input
halts with output
0xMy03m+33 xM(i, s, a)y .
Using these machines, we can finally assemble an universal Turing
machine.
Theorem 12.13. There is a Turing machine U such that, for any

m ≥ 1, m-state Turing machine M with alphabet {1}, and tape position
(i, s, a) for M, U acts on the input position
as follows:
• If M, starting from position (i, s, a), eventually halts in position
(j, t, b), then U eventually halts in the position
0xMy03m+33 x(j, t, b)y .
• If M, starting from the initial tape position (i, s, a), eventually
runs off the left end of its tape, U eventually runs off the left end
of its own tape.
• If M, starting from the initial tape position (i, s, a), never halts,
then U never halts.
The Halting Problem. Given that we are using Turing machines
to formalize the notion of an effective method, one of the difficulties
with solving the Halting Problem is representing a given Turing ma-
chine and its input tape as input for another machine. As this is exactly
what was done above, we can now formulate a precise version of the
Halting Problem and solve it.
The Halting Problem. Is there a Turing machine T which, for
any m ≥ 1, m-state Turing machine M with alphabet {1}, and tape a
for M, halts on input
0xMy03m+33 x(0, 1, a)y
with output 011 if M halts on input a, and with output 01 if M does
not halt on input a?
Note that this version of the Halting Problem is equivalent to our
original one only if Church’s Thesis is true.
Problem 12.14. Find a Turing machine C which, for any Turing
machine M with alphabet {1}, on input
0xMy
eventually halts with output
0x(0, 1, xMy)y .
Theorem 12.15. There is no Turing machine T which, for any
m ≥ 1, m-state Turing machine M with alphabet {1}, and tape a for
M, halts on input
0xMy03m+33 x(0, 1, a)y
with output 011 if M halts on input a, and with output 01 if M does

not halt on input a.
CHAPTER 13
Computable and Non-Computable Functions
So far, the only substantial facts we have about what Turing ma-
chines can do is that they can be used to simulate other Turing ma-
chines, but cannot solve the Halting Problem. Neither fact is trivial,
but neither is really interesting unless Turing machines can also be
used to handle more natural computational problems. Arithmetic is a
common source of such problems in the real world; indeed, any notion
of computation that can’t handle it is unlikely to be of great use.
Notation and conventions. To keep things as simple as pos-

sible, we will stick to computations involving the natural numbers,
i.e. the non-negative integers, the set of which is usually denoted by
N = { 0, 1, 2, . . . }.. The set of all k-tuples (n1 , . . . , nk ) of natural num-
bers is denoted by Nk . For all practical purposes, we may take N1 to
be N by identifying the 1-tuple (n) with the natural number n.
For k ≥ 1, f is a k-place function (from the natural numbers to the
natural numbers), often written as f : Nk → N, if it associates a value,
f(n1 , . . . , nk ), to each k-tuple (n1, n2 , . . . , nk ) ∈ Nk . Strictly speaking,
though we will frequently forget to be explicit about it, we will often
be working with k-place partial functions which might not be defined
for all the k-tuples in Nk . If f is a k-place partial function, the domain
of f is the set
dom(f) = { (n1 , . . . , nk ) ∈ Nk | f(n1 , . . . , nk ) is defined } .
Similarly, the image of f is the set
im(f) = { f(n1 , . . . , nk ) | (n1 , . . . , nk ) ∈ dom(f) } .
In subsequent chapters we will also work with relations on the nat-
ural numbers. Recall that a k-place relation on N is formally a subset
P of Nk ; P (n1 , . . . , nk ) is true if (n1 , . . . , nk ) ∈ P and false otherwise.
In particular, a 1-place relation is really just a subset of N.
Relations and functions are closely related. All one needs to know
about a k-place function f can be obtained from the (k + 1)-place
relation Pf given by
Pf (n1 , . . . , nk , nk+1 ) ⇐⇒ f(n1 , . . . , nk ) = nk+1 .
25
26 13. COMPUTABLE AND NON-COMPUTABLE FUNCTIONS
Similarly, all one needs to know about the k-place relation P can be
obtained from its characteristic function :
(
1 if P (n1 , . . . , nk ) is true;
χP (n1 , . . . , nk ) =
0 if P (n1 , . . . , nk ) is false.
The basic convention for representing natural numbers on the tape

of a Turing machine is a slight variation of unary notation : n is rep-
resented by 1n+1 . (Why would using 1n be a bad idea?) A k-tuple
(n1 , n2, . . . , nk ) ∈ N will be represented by 1n1 +1 01n2 +1 0 . . . 01nk +1 , i.e.
with the representations of the individual numbers separated by 0s.
This scheme is inefficient in its use of space — compared to binary
notation, for example — but it is simple and can be implemented on
Turing machines restricted to the alphabet {1}.
Computable functions. With suitable conventions for represent-

ing the input and output of a function on the natural numbers on the
tape of a Turing machine in hand, we can define what it means for a
function to be computable by a Turing machine.
Definition 13.1. A k-place function f is Turing computable, or
just computable, if there is a Turing machine M such that for any
k-tuple (n1 , . . . , nk ) ∈ dom(f) the computation of M with input tape
01n1 +1 01n2 +1 . . . 01nk +1 eventually halts with output tape 01f (n1 ,...,nk )+1 .
Such an M is said to compute f.
Note that for a Turing machine M to compute a function f, M
need only do the right thing on the right kind of input: what M does
in other situations does not matter. In particular, it does not matter
what M might do with k-tuple which is not in the domain of f.
Example 13.1. The identity function iN : N → N, i.e. iN(n) = n,
is computable. It is computed by M = ∅, the Turing machine with an
empty table that does absolutely nothing on any input.
Example 13.2. The projection function π12 : N2 → N given by
π12(n, m) = n is computed by the Turing machine:
P12 0 1
1 0R2
2 0R3 1R2
3 0L4 0R3
4 0L4 1L5
5 1L5
13. COMPUTABLE AND NON-COMPUTABLE FUNCTIONS 27
P12 acts as follows: it moves to the right past the first block of 1s
without disturbing it, erases the second block of 1s, and then returns
to the left of first block and halts.
The projection function π22 : N2 → N given by π12(n, m) = m is also
computable: the Turing machine P of Example 10.4 does the job.
Problem 13.1. Find Turing machines that compute the following
functions and explain how they work.
1. O(n) = 0.
2. S(n) = n + 1.
3. Sum(n, m) =(n + m.
n−1 n≥ 1
4. Pred(n) = .
0 n=0
(
n−m n≥m
5. Diff(n, m) = .
0 n<m
6. π23(p, q, r) = q.
We will see how to build complex functions computable by Turing
machines out of simpler ones in the next chapter.
A non-computable function. In the meantime, it is worth ask-
ing whether or not every function on the natural numbers is com-
putable. No such luck!
Problem 13.2. Show that there is some 1-place function f which
is not computable by comparing the number of such functions to the
number of Turing machines.
The argument hinted at above is unsatisfying in that it tells us there
is a non-computable function without actually producing an explicit
example. We can have some fun on the way to one.
Definition 13.2 (Busy Beaver Competition). A machine M is an
n-state entry in the busy beaver competition if:
• M has a two-way infinite tape and alphabet {1};
• M has n + 1 states, but state n + 1 is used only for halting (so
both M(n + 1, 0) and M(n + 1, 1) are undefined);
• M eventually halts when given a blank input tape.
M’s score in the competition is the number of 1’s on the output tape of
its computation from a blank input tape. The greatest possible score
of an n-state entry in the competition is denoted by Σ(n).
Note that there are only finitely many possible n-state entries in the
busy beaver competition because there are only finitely many (n + 1)-
state Turing machines with alphabet {1}. Since there is at least one
28 13. COMPUTABLE AND NON-COMPUTABLE FUNCTIONS
n-state entry in the busy beaver competition for every n ≥ 0 , it follows

that Σ(n) is well-defined for each n ∈ N.
Example 13.3. M = ∅ is the only 0-state entry in the busy beaver
competition, so Σ(0) = 0.
Example 13.4. The machine P given by
P 0 1
1 1R2 1L2
2 1L1 1L3
is a 2-state entry in the busy beaver competition with a score of 4, so
Σ(2) ≥ 4.
The function Σ grows extremely quickly. It is known that Σ(0) = 0,
Σ(1) = 1, Σ(2) = 4, Σ(3) = 6, and Σ(4) = 13. The value of Σ(5) is
still unknown, but must be quite large.1
Problem 13.3. Show that:
1. The 2-state entry given in Example 13.4 actually scores 4.
2. Σ(1) = 1.
3. Σ(3) ≥ 6.
4. Σ(n) < Σ(n + 1) for every n ∈ N.
Problem 13.4. Devise as high-scoring 4- and 5-state entries in the
busy beaver competition as you can.
The serious point of the busy beaver competition is that Σ is not a
Turing computable function.
Proposition 13.5. Σ is not computable by any Turing machine.
Anyone interested in learning more about the busy beaver com-
petition should start by reading the paper [10] in which it was first
introduced.
1
The best score known to the author as of this writing by a 5-state entry in
the busy beaver competition is 4098. One of the two machines achieving this score
does so in a computation that takes over 40 million steps! The other requires only
11 million or so . . .
CHAPTER 14
Primitive Recursive Functions
Starting with a small set of computable functions, and applying

computable ways of building functions from simpler ones, we will build
up a useful collection of computable functions. This will also go a long
way toward giving us a characterization of computable functions which
does not mention any particular computing devices.
The initial functions. The set of computable functions that will
be the fundamental building blocks for all that follows is infinite only
because of the presence of all the projection functions.
Definition 14.1. The following are the initial functions:
• O, the 1-place function such that O(n) = 0 for all n ∈ N;
• S, the 1-place function such that S(n) = n+ 1 for all n ∈ N; and,
• for each k ≥ 1 and 1 ≤ i ≤ k, πik , the k-place function such that
πik (n1 , . . . , nk ) = ni for all (n1 , . . . , nk ) ∈ Nk .
O is often referred to as the zero function, S is the successor function,
and the functions πik are called the projection functions.
Note that π11 is just the identity function on N. We have already ob-
served that O, S, π11, π12 , π22, and π23 are Turing computable in Chapter
13.
Problem 14.1. Show that all of the initial functions are Turing
computable.
Composition. The first of our methods for assembling computable
functions from simpler ones should be thoroughly familiar from many
parts of mathematics.
Definition 14.2. Suppose that m, k ≥ 1, g is an m-place function,
and h1, . . . , hm are k-place functions. Then the k-place function f is
said to be obtained from g, h1 , . . . , hm by composition, written as
f = g ◦ (h1 , . . . , hm ) ,
if for all (n1 , . . . , nk ) ∈ Nk ,
f(n1 , . . . , nk ) = g(h1 (n1 , . . . , nk ), . . . , hm (n1 , . . . , nk )).
29
30 14. PRIMITIVE RECURSIVE FUNCTIONS
Example 14.1. The constant function c11 , where c11 (n) = 1 for all
n, can be obtained by composition from the initial functions S and O.
For any n ∈ N,
c11 (n) = (S ◦ O)(n) = S(O(n)) = S(0) = 0 + 1 = 1 .
Problem 14.2. Suppose k ≥ 1 and a ∈ N. Use composition
to define the constant function cka , where cka (n1 , . . . , nk ) = a for all
(n1 , . . . , nk ) ∈ Nk , from the initial functions.
Proposition 14.3. Suppose that 1 ≤ k, 1 ≤ m, g is a Turing
computable m-place function, and h1, . . . , hm are Turing computable
k-place functions. Then g ◦ (h1 , . . . , hm ) is also Turing computable.
Unfortunately, one can’t do much else of interest using just the
initial functions and composition . . .
Proposition 14.4. Suppose f is a 1-place function obtained from
the initial functions by finitely many applications of composition. Then
there is a constant c ∈ N such that f(n) ≤ n + c for all n ∈ N.
Primitive recursion. Primitive recursion boils down to defining
a function inductively, using different functions to tell us what to do
at the base and inductive steps. Together with composition, it suffices
to build up just about all familiar arithmetic functions from the initial
functions.
Definition 14.3. Suppose that k ≥ 1, g is a k-place function, and
h is a k + 2-place function. Let f be the (k + 1)-place function such
that
1. f(n1 , . . . , nk , 0) = g(n1 , . . . , nk ) and
2. f(n1 , . . . , nk , m + 1) = h(n1 , . . . , nk , m, f(n1, . . . , nk , m)
for every (n1 , . . . , nk ) ∈ Nk and m ∈ N. Then f is said to be obtained
from g and h by primitive recursion.
That is, the initial values of f are given by g, and the rest are given
by h operating on the given input and the preceding value of f.
For a start, primitive recursion and composition let us define addi-
tion and multiplication from the initial functions.
Example 14.2. Sum(n, m) = n + m is obtained by primitive re-
cursion from the initial function π11 and the composition S ◦ π33 of initial
functions as follows:
• Sum(n, 0) = π11(n);
• Sum(n, m + 1) = (S ◦ π33 )(n, m, Sum(n, m)).
14. PRIMITIVE RECURSIVE FUNCTIONS 31
To see that this works, one can proceed by induction on m:

At the base step, m = 0, we have
Sum(n, 0) = π11(n) = n = n + 0 .
Assume that m ≥ 0 and Sum(n, m) = n + m. Then
Sum(n, m + 1) = (S ◦ π33 )(n, m, Sum(n, m))
= S(π33 (n, m, Sum(n, m)))
= S(Sum(n, m))
= Sum(n, m) + 1
= n + m + 1,
as desired.
As addition is to the successor function, so multiplication is to
addition.
Example 14.3. Mult(n, m) = nm is obtained by primitive recur-
sion from O and Sum ◦ (π33, π13):
• Mult(n, 0) = O(n);
• Mult(n, m + 1) = (Sum ◦ (π33, π13))(n, m, Mult(n, m)).
We leave it to the reader to check that this works.
Problem 14.5. Use composition and primitive recursion to obtain
each of the following functions from the initial functions or other func-
tions already obtained from the initial functions.
1. Exp(n, m) = nm
2. Pred(n) (defined in Problem 13.1)
3. Diff(n, m) (defined in Problem 13.1)
4. Fact(n) = n!
Proposition 14.6. Suppose k ≥ 1, g is a Turing computable k-
place function, and h is a Turing computable (k + 2)-place function. If
f is obtained from g and h by primitive recursion, then f is also Turing
computable.
Primitive recursive functions and relations. The collection of
functions which can be obtained from the initial functions by (possibly
repeatedly) using composition and primitive recursion is useful enough
to have a name.
Definition 14.4. A function f is primitive recursive if it can be
defined from the initial functions by finitely many applications of the
operations of composition and primitive recursion.
So we already know that all the initial functions, addition, and

multiplication, among others, are primitive recursive.
Problem 14.7. Show that each of the following functions is prim-
itive recursive.
1. For any k ≥ 0 and primitive resursive (k + 1)-place function g,
and Πm i=0 g(n1 , . . . , nk , i), the (k + 1)-place function f given by
f(n1 , . . . , nk , m) = Πm
i=0 g(n1 , . . . , nk , i)
= g(n1 , . . . , nk , 0) · . . . · g(n1 , . . . , nk , m)
for any (k + 1)-tuple (n1 , . . . , nk , m).(
0 n 6= a
2. For any constant a ∈ N, χ{a}(n) =
1 n = a.
(
f(n1 , . . . , nk ) (n1 , . . . , nk ) 6= (c1 , . . . , ck )
3. h(n1 , . . . , nk ) = , if
a (n1 , . . . , nk ) = (c1 , . . . , ck )
f is a primitive recursive k-place function and a, c1, . . . , ck ∈ N
are constants.
Theorem 14.8. Every primitive recursive function is Turing com-
putable.
Be warned, however, that there are computable functions which are
not primitive recursive.
We can extend the idea of “primitive recursive” to relations by using
their characteristic functions.
Definition 14.5. Suppose k ≥ 1. A k-place relation P ⊆ Nk is
primitive recursive if its characteristic function
(
1 (n1 , . . . , nk ) ∈ P
χP (n1 , . . . , nk ) =
0 (n1 , . . . , nk ) ∈
/P
is primitive recursive.
Example 14.4. P = {2} ⊂ N is primitive recursive since χ{2} is
recursive by Problem 14.7.
Problem 14.9. Show that the following relations and functions are
primitive recursive.
1. ¬P , i.e. Nk \ P , if P is a primitive recursive k-place relation.
2. P ∨ Q, i.e. P ∪ Q, if P and Q are primitive recursive k-place
relations.
3. P ∧ Q, i.e. P ∩ Q, if P and Q are primitive recursive k-place
relations.
14. PRIMITIVE RECURSIVE FUNCTIONS 33
4. Equal, where Equal(n, Pm m) ⇐⇒ n = m.

5. h(n1 , . . . , nk , m) = i=0 g(n1 , . . . , nk , i), for any k ≥ 0 and prim-
itive recursive (k + 1)-place function g.
6. Div, where Div(n, m) ⇐⇒ n | m.
7. IsPrime, where IsPrime(n) ⇐⇒ n is prime.
8. Prime(k) = pk , where p0 = 1 and pk is the kth prime if k ≥ 1.
9. Power(n, m) = k, where k ≥ 0 is maximal such that nk | m.
10. Length(n) = `, where ` is maximal such that p` | n.
11. Element(n, i) = n(i , if n = pn1 1 . . . pnk k (and ni = 0 if i > k).
ni+1 n
pni i pi+1 . . . pj j if 1 ≤ i ≤ j ≤ k
12. Subseq(n, i, j) = , whenever
0 otherwise
n = pn1 1 . . . pnk k .
13. Concat(n, m) = pn1 1 . . . pnk k pm ml n1 nk
k+1 . . . pk+` , if n = p1 . . . pk and
1
m1 m`
m = p1 . . . p` .
Parts of Problem 14.9 give us tools for representing finite sequences
of integers by single integers, as well as some tools for manipulating
these representations. This lets us reduce, in principle, all problems
involving primitive recursive functions and relations to problems in-
volving only 1-place primitive recursive functions and relations.
Theorem 14.10. A k-place g is primitive recursive if and only if
the 1-place function h given by h(n) = g(n1 , . . . , nk ) if n = pn1 1 . . . pnk k
is primitive recursive.
Note. It doesn’t matter what the function h may do on an n which
does not represent a sequence of length k.
Corollary 14.11. A k-place relation P is primitive recursive if
and only if the 1-place relation P 0 is primitive recursive, where
(n1, . . . , nk ) ∈ P ⇐⇒ pn1 1 . . . pnk k ∈ P 0 .
Computable non-primitive recursive functions. While primi-
tive recursion and composition do not suffice to build up all Turing com-
putable functions from the initial functions, they are powerful enough
that specific counterexamples are not all that easy to find.
Example 14.5 (Ackerman’s Function). Define the 2-place function
A from as follows:
• A(0, `) = S(`)
• A(S(k), 0) = A(k, 1)
• A(S(k), S(`)) = A(k, A(S(k), `))
Given A, define the 1-place function α by α(n) = A(n, n).
It isn’t too hard to show that A, and hence also α, are Turing
computable. However, though it takes considerable effort to prove it,
α grows faster with n than any primitive recursive function. (Try
working out the first few values of α . . . )
Problem 14.12. Show that the functions A and α defined in Ex-
ample 14.5 are Turing computable.
If you are very ambitious, you can try to prove the following theo-
rem.
Theorem 14.13. Suppose α is the function defined in Example
14.5 and f is any primitive recursive function. Then there is an n ∈ N
such that for all k > n, α(k) > f(k).
Corollary 14.14. The function α defined in Example 14.5 is not
. . . but if you aren’t, you can still try the following exercise.
Problem 14.15. Informally, define a computable function which
must be different from every primitive recursive function.
CHAPTER 15
Recursive Functions
We add one more computable method of building functions, un-

bounded minimalization, to our repertoire. The functions which can
be defined from the initial functions using unbounded minimalization,
as well as composition and primitive recursion, turn out to be precisely
the Turing computable functions.
Unbounded minimalization. Unbounded minimalization is the
counterpart for functions of “brute force” algorithms that try every
possibility until they succeed. (Which, of course, they might not . . . )
Definition 15.1. Suppose k ≥ 1 and g is a (k + 1)-place func-
tion. Then the unbounded minimalization of g is the k-place function
f defined by
f(n1 , . . . , nk ) = m where m is least so that g(n1 , . . . , nk , m) = 0.
This is often written as f(n1 , . . . , nk ) = µm[g(n1, . . . , nk , m) = 0].
Note. If there is no m such that g(n1 , . . . , nk , m) = 0, then the
unbounded minimalization of g is not defined on (n1 , . . . , nk ). This is
one reason we will occasionally need to deal with partial functions.
If the unbounded minimalization of a computable function is to be
computable, we have a problem even if we ask for some default out-
put (0, say) to ensure that it is defined for all k-tuples. The obvious
procedure which tests successive values of g to find the needed m will
run forever if there is no such m, and the incomputability of the Halt-
ing Problem suggests that other procedure’s won’t necessarily succeed
either. It follows that it is desirable to be careful, so far as possible,
which functions unbounded minimalization is applied to.
Definition 15.2. A (k + 1)-place function g is said to be regular
if for every (n1, . . . , nk ) ∈ Nk , there is at least one m ∈ N so that
g(n1 , . . . , nk , m) = 0.
That is, g is regular precisely if the obvious strategy of computing
g(n1 , . . . , nk , m) for m = 0, 1, . . . in succession until an m is found
with g(n1 , . . . , nk , m) = 0 always succeeds.
35
36 15. RECURSIVE FUNCTIONS
Proposition 15.1. If g is a Turing computable regular (k + 1)-

place function, then the unbounded minimalization of g is also Turing
computable.
While unbounded minimalization adds something essentially new to
our repertoire, it is worth noticing that bounded minimalization does
not.
Problem 15.2. Suppose g is a (k + 1)-place primitive recursive
regular function such that for some primitive recursive k-place function
h,
µm[g(n1, . . . , nk , m) = 0] ≤ h(n1, . . . , nk )
for all (n1 , . . . , nk ) ∈ N. Show that µm[g(n1, . . . , nk , m) = 0] is also
Recursive functions and relations. We can finally define an
equivalent notion of computability for functions on the natural numbers
which makes no mention of any computational device.
Definition 15.3. A k-place function f is recursive if it can be
defined from the initial functions by finitely many applications of com-
position, primitive recursion, and the unbounded minimalization of
regular functions.
Similarly, k-place partial function is recursive if it can be defined
from the initial functions by finitely many applications of composition,
primitive recursion, and the unbounded minimalization of (possibly
non-regular) functions.
In particular, every primitive recursive function is a recursive func-
tion.
Theorem 15.3. Every recursive function is Turing computable.
We shall show that every Turing computable function is recursive
later on. Similarly to primitive recursive relations we have the follow-
ing.
Definition 15.4. A k-place relation P is said to be recursive (Tur-
ing computable) if its characteristic function χP is recursive (Turing
computable).
Since every recursive function is Turing computable, and vice versa,
“recursive” is just a synonym of “Turing computable”, for functions and
relations alike.
Also, similarly to Theorem 14.10 and Corollary 14.11 we have the
following.
15. RECURSIVE FUNCTIONS 37
Theorem 15.4. A k-place g is recursive if and only if the 1-place

function h given by h(n) = g(n1 , . . . , nk ) if n = pn1 1 . . . pnk k is recursive.
As before, it doesn’t really matter what the function h does on an
n which does not represent a sequence of length k.
Corollary 15.5. A k-place relation P is recursive if and only if
the 1-place relation P 0 is recursive, where
(n1, . . . , nk ) ∈ P ⇐⇒ pn1 1 . . . pnk k ∈ P 0 .
Turing computable functions are recursive. By putting some
of the ideas in Chapters 12 and 14 together, we can use recursive func-
tions to simulate Turing machines. This will show that Turing com-
putable functions are recursive and, as a bonus, give us another way
of constructing an universal Turing machine. Since recursive functions
operate on integers, we first need to specify some way to code the tapes
of Turing machines by integers. We’ll try keep close to the representa-
tion schemes given in Definitions 12.2 and 12.3 in the process. As we
did in those definitions, we shall stick to Turing machines with alphabet
{1} for simplicity.
Definition 15.5. Suppose (i, s, a) is a tape position such that all
but finitely many cells of a are blank. Let n be any positive integer
such that ak = 0 for all k ∈ N with k > n. Then the code of (i, s, a) is
p(i, s, a)q = 2i 3s 5a0 7a1 11a2 . . . pan+3
n
.
Example 15.1. Consider the tape position (1, 2, 1001). Then
p(1, 2, 1001)q = 21 32 51 70 110 131 = 1170 .
Problem 15.6. Find the codes of the following tape positions.
1. (0, 1, a), where a is entirely blank.
2. (3, 4, a), where a is 1011100101.
Problem 15.7. What is the tape position whose code is 10314720?
We’ll also need to code sequences of tape positions when we deal
with computations.
Definition 15.6. Suppose t1t2 . . . tn is a sequence of tape posi-
tions. Then the code of this sequence is
pt1t2 . . . t2 q = 2pt1 q 3pt2 q . . . ppntn q .
Note. Both tape positions and sequences of tape positions also
have unique codes.
Problem 15.8. Pick some (short!) sequence of tape positions and
find its code.
Having defined how to represent tape positions as integers, we now

need to manipulate these representations using recursive functions.
The recursive functions and relations in Problem 14.9 provide the nec-
essary tools.
Problem 15.9. Show that each of the following is primitive recur-
sive.
1. The 4-place function Entry, where
Entry(j, w, t, n) =


 p(i + w − 1, t, a0)q if n = p(i, s, a)q, j ∈ {0, 1},

 w ∈ {0, 2}, and i + w − 1 ≥ 0; where

 a0k = ak for k 6= i and a0i = j;


0 otherwise.
2. For any Turing machine M with alphabet {1}, the 1-place func-
tion TMM , such that


pM(i, s, a)q if n = p(i, s, a)q
TMM (n) = and M(i, s, a) is defined;

0 otherwise.
3. For any Turing machine M with alphabet {1}, the 1-place rela-
tion CompM , where
CompM (n) ⇐⇒ n is the code of a computation of M.
The functions and relations above may be primitive recursive, but
the last step in showing that Turing computable functions are recursive
requires unbounded minimalization.
Theorem 15.10. Any k-place Turing computable function is recur-
sive.
One can push the techniques used above just a little farther to
get a recursive function that simulates any Turing machine. Since
any recursive function can be computed by some Turing machine, this
effectively gives us another universal Turing machine.
Problem 15.11. Devise a suitable definition for the code pMq of
a Turing machine M with alphabet {1}.
Problem 15.12. Show, using your definition of pMq from Problem
15.11, that the following are primitive recursive.
15. RECURSIVE FUNCTIONS 39
1. The 2-place function TM, where



 pM(i, s, a)q if m = pMq for some machine M,

 n = p(i, s, a)q,
TM(m, n) =

 and M(i, s, a) is defined;


0 otherwise.
2. The 2-place relation Comp, where
Comp(m, n) ⇐⇒ m = pMq
for some Turing machine M and n is the code of a computation
of M.
Problem 15.13. Show that the 2-place partial function Sim is re-
cursive, where, for any Turing machine M with alphabet {1} and input
tape a for M,
Sim(pMq, p(0, 1, a)q) = p(0, 1, b)q
if M halts with output b on input a.
Note that Sim(m, n) may be undefined on other inputs.
Recursively enumerable sets. The following notion is of partic-
ular interest in the advanced study of computability.
Definition 15.7. A subset (i.e. a 1-place relation) P of N is re-
cursively enumerable, often abbreviated as r.e., if there is a 1-place
recursive function f such that P = im(f) = { f(n) | n ∈ N }.
Since the image of any recursive 1-place function is recursively enu-
merable by definition, we do not lack for examples. For one, the set E
of even natural numbers is recursively enumerable, since it is the image
of f(n) = Mult(S(S(O(n))), n).
Proposition 15.14. If P is a 1-place recursive relation, then P is
recursively enumerable.
This proposition is not reversible, but it does come close.
Proposition 15.15. P ⊆ N is recursive if and only if both P and
N \ P are recursively enumerable.
Problem 15.16. Find an example of a recursively enumerable set
which is not recursive.
Problem 15.17. Is P ⊆ N primitive recursive if and only if both
P and N \ P are enumerable by primitive recursive functions?
Incompleteness
CHAPTER 16
Preliminaries
It was mentioned in the Introduction that one of the motivations for

the development of notions of computability was the following question.
Entscheidungsproblem. Given a reasonable set Σ of formulas
of a first-order language L and a formula ϕ of L, is there an effective
method for determining whether or not Σ ` ϕ?
Armed with knowledge of first-order logic on the one hand and of
computability on the other, we are in a position to formulate this ques-
tion precisely and then solve it. To cut to the chase, the answer is“no”
in general. Gödel’s Incompleteness Theorem asserts, roughly, that for
any computable set of axioms in a first-order language powerful enough
to prove certain facts about arithmetic, it is possible to formulate state-
ments in the language whose truth is not decided by the axioms. In
particular, it turns out that no consistent set of axioms can hope to
prove its own consistency.
We will tackle the Incompleteness Theorem in three stages. First,
we will code the formulas and proofs of a first-order language as num-
bers and show that the functions and relations involved are recursive.
This will, in particular, make it possible for us to define “computable
set of axioms” precisely. Second, we will show that all recursive func-
tions and relations can be defined by first-order formulas in the presence
of a fairly minimal set of axioms about elementary number theory. Fi-
nally, by putting recursive functions talking about first-order formulas
together with first-order formulas defining recursive functions, we will
manufacture a self-referential sentence which asserts its own unprov-
ability.
A language for first-order number theory. To keep things as
concrete as possible we will work with and in the following language
for first-order number theory, used as an example in Chapter 5.
Definition 16.1. LN is the first-order language with the following
symbols:
1. Parentheses: ( and )
2. Connectives: ¬ and →
43
44 16. PRELIMINARIES
3. Quantifier: ∀
4. Equality: =
5. Variable symbols: v0, v2, v3, . . .
6. Constant symbol: 0
7. 1-place function symbol: S
8. 2-place function symbols: +, ·, and E.
The non-logical symbols of LN , 0, S, +, ·, and E, are intended
to name, respectively, the number zero, and the successor, addition,
multiplication, and exponentiation functions on the natural numbers.
That is, the (standard!) structure this language is intended to discuss
is N = (N, 0, S, +, ·, E).
Note. Notation for and the definitions of terms, formulas, sen-
tences, structures, interpretations, logical axioms, deductions, and so
on, of first-order languages, plus various conventions involving these,
are given in Chapters 5-8 of Volume I. Look them up as (and if) you
need to.
Completeness. The notion of completeness mentioned in the In-
completeness Theorem is different from the one mentioned in the Com-
pleteness Theorem.1 “Completeness” in the latter sense is a property of
a logic: it asserts that whenever Γ |= σ (i.e. the truth of the sentence σ
follows from that of the set of sentences Γ), Γ ` σ (i.e. there is a deduc-
tion of σ from Γ). The sense of “completeness” in the Incompleteness
Theorem, defined below, is a property of a set of sentences.
Definition 16.2. A set of sentences Σ of a first-order language L
is said to be complete if for every sentence τ either Σ ` τ or Σ ` ¬τ .
That is, a set of sentences, or non-logical axioms, is complete if it
suffices to prove or disprove every sentence of the langage in in question.
Problem 16.1. Show that a consistent set Σ of sentences of a first-
order language L is complete if and only if the theory of Σ,
Th(Σ) = { τ | τ is a sentence of L and Σ ` τ } ,
is maximally consistent.
1
Which, to confuse the issue, was also first proved by Gödel.
CHAPTER 17
Coding First-Order Logic
We will encode the symbols, formulas, and deductions of LN as

natural numbers in such a way that the operations necessary to ma-
nipulate these codes are recursive. Although we will do so just for LN ,
any countable first-order language can be coded in a similar way.
Definition 17.1. To each symbol s of LN we assign an unique
positive integer psq, the Gödel code of s, as follows:
1. p(q = 1 and p)q = 2
2. p¬q = 3 and p→q = 4
3. p∀q = 5
4. p=q = 6.
5. pvk q = k + 12
6. p0q = 7
7. pSq = 8
8. p+q = 9, p·q = 10, and pEq = 11
Note that each positive integer is the Gödel code of one and only one
symbol of LN . We will also need to code sequences of the symbols of
LN , such as terms and formulas, as numbers, not to mention sequences
of sequences of symbols of LN , such as deductions.
Definition 17.2. Suppose s1s2 . . . sk is a sequence of symbols of
LN . Then the Gödel code of this sequence is
ps1 . . . sk q = pp1s1 q . . . ppksk q ,
where pn is the nth primes number.
Similarly, if σ1σ2 . . . σ` is a sequence of sequences of symbols of LN ,
then the Gödel code of this sequence is
pσ1 . . . σ` q = pp1σ1 q . . . ppkσ` q .
Example 17.1. p∀v1 = ·v1S0v1q = 25 313 56 7101113 138 177 1913 .
A particular integer n may simultaneously be the Gödel code of a
symbol, a sequence of symbols, and a sequence of sequences of symbols
of LN . We shall rely on context to avoid confusion, but, with some
45
46 17. CODING FIRST-ORDER LOGIC
more work, one could set things up so that no integer was the code of
more than one kind of thing.
We need to show that various relations and functions for recognizing
and manipulating Gödel codes are recursive.
Problem 17.1. Show that each of the following relations is prim-
itive recursive.
1. Term(n) ⇐⇒ n = ptq for some term t of LN .
2. Formula(n) ⇐⇒ n = pϕq for some formula ϕ of LN .
3. Sentence(n) ⇐⇒ n = pσq for some sentence σ of LN .
4. Logical(n) ⇐⇒ n = pγq for some logical axiom γ of LN .
Using these relations as building blocks, we will develop relations
and functions to handle deductions of LN . First, though, we need to
make “a computable set of formulas” precise.
Definition 17.3. A set ∆ of formulas of LN is said to be recursive
if the set of Gödel codes of formulas of ∆,
p∆q = { pδq | δ ∈ ∆ } ,
is recursive. Similarly, ∆ is said to be recursively enumerable if p∆q is
recursively enumerable.
Problem 17.2. Suppose ∆ is a recursive set of sentences of LN .
Show that each of the following relations is recursive.
1. Premiss∆ (n) ⇐⇒ n = pβq for some formula β of LN which
is either a logical axiom or in ∆.
2. Formulas(n) ⇐⇒ n = pϕ1 . . . ϕk q for some sequence ϕ1 . . . ϕk
of formulas of LN .
3. Inference(n, i, j) ⇐⇒ n = pϕ1 . . . ϕk q for some sequence
ϕ1 . . . ϕk of formulas of LN , 1 ≤ i, j ≤ k, and ϕk follows from ϕi
and ϕj by Modus Ponens.
4. Deduction∆ (n) ⇐⇒ n = pϕ1 . . . ϕk q for a deduction ϕ1 . . . ϕk
from ∆ in LN .
5. Conclusion∆ (n, m) ⇐⇒ n = pϕ1 . . . ϕk q for a deduction
ϕ1 . . . ϕk from ∆ in LN and m = pϕk q.
If p∆q is primitive recursive, which of these are primitive recursive?
It is at this point that the connection between computability and
completeness begins to appear.
Theorem 17.3. Suppose ∆ is a recursive set of sentences of LN .
Then pTh(∆)q is
1. recursively enumerable, and
2. recursive if and only if ∆ is complete.
17. CODING FIRST-ORDER LOGIC 47
Note. It follows that pTh(∆)q is an example of a recursively enu-

merable but not recursive set if ∆ is not complete.
48 17. CODING FIRST-ORDER LOGIC
CHAPTER 18
Defining Recursive Functions In Arithmetic
We will also need definitions and results complementing those ob-

tained in Chapter 17: a set of non-logical axioms in LN which prove
enough to let us define all the recursive functions by suitable formulas
of LN . The non-logical axioms in question essentially just ensure that
basic arithmetic works properly.
Definition 18.1. Let A be the following set of sentences of LN ,
written out in official form.
1. ∀v0 (¬ = Sv00)
2. ∀v0 ((¬ = v0 0) → (¬∀v1 (¬ = Sv1v0)))
3. ∀v0∀v1 (= Sv0Sv1 →= v0v1 )
4. ∀v0 = +v00v0
5. ∀v0∀v1 = +v0Sv1S + v0 v1
6. ∀v0 = ·v000
7. ∀v0∀v1 = ·v0Sv1 + ·v0 v1v0
8. ∀v0 = Ev00S0
9. ∀v0∀v1 = Ev0Sv1 · Ev0v1v0
Translated from the official forms, A consists of the following ax-
ioms about the natural numbers:
1. ∀x x + 1 6= 0
2. ∀x x 6= 0 → ∃y y + 1 = x
3. ∀x∀y x + 1 = y + 1 → x = y
4. ∀x x + 0 = x
5. ∀x∀y x + y + 1 = (x + y) + 1
6. ∀x x · 0 = 0
7. ∀x∀y x · (y + 1) = (x · y) + x
8. ∀x x0 = 1
9. ∀x∀y xy+1 = (xy ) · x
Each of the axioms in A is true in N = (N, 0, S, +, ·, E). However,
they are a long way from being able to prove all the sentences of first-
order arithmetic true in N. For example, though we won’t prove it,
it turns out that A is not enough to ensure that induction works:
that for every formula ϕ with at most the variable x free, if ϕx0 and
49
50 18. DEFINING RECURSIVE FUNCTIONS IN ARITHMETIC
∀y (ϕxy → ϕxSy ) hold, then so does ∀x ϕ. On the other hand, neither LN

nor A are quite as minimal as they might be. For example, one could
do without E and define it from · and +.
It will be convenient to use a couple of conventions in what follows.
First, we will often abbreviate the term of LN consisting of m Ss fol-
lowed by 0 by S m 0. For example, S 3 0 abbreviates SSS0. We will use
S m 0 to represent the natural number m in LN . (The interpretation of
S m 0 in N will, in fact, be the mth successor of 0, namely m.) Second, if
ϕ is a formula of LN with all of its free variables among v1 , . . . , vn , and
m0, m1, . . . , mn are natural numbers, we will write ϕ(S m1 0, . . . , S mk 0)
for the sentence ϕvS1m...v n
1 0,...,S mk 0 , i.e. ϕ with every free occurrence of vi
replaced by S mi 0. Note that since the term S mi 0 involves no variables,
it must be substitutable for vi in ϕ.
Definition 18.2. Suppose Σ is a set of sentences of LN . A k-place
function f is said to be representable in Th(Σ) = { τ | Σ ` τ } if there
is a formula ϕ of LN with at most v1, . . . , vk , and vk+1 as free variables
such that
f(n1 , . . . , nk ) = m ⇐⇒ ϕ(S n1 0, . . . , S nk 0, S m 0) ∈ Th(Σ)
⇐⇒ Σ ` ϕ(S n1 0, . . . , S nk 0, S m 0)
for all n1 , . . . , nk in N. Such a formula ϕ is said to represent f in
Th(Σ).
Similarly, a k-place relation P ⊆ Nk is said to be representable in
Th(Σ) if there is a formula ψ of LN with at most v1, . . . , vk as free
variables such that
P (n1 , . . . , nk ) ⇐⇒ ψ(S n1 0, . . . , S nk 0) ∈ Th(Σ)
⇐⇒ Σ ` ψ(S n1 0, . . . , S nk 0)
for all n1 , . . . , nk in N. Such a formula ψ is said to represent P in
Th(Σ).
We will use this definition mainly with Σ = A.
Example 18.1. The constant function c3 given by c3(n) = 3 is
representable in Th(A); v2 = SSS0 is a formula representing it. Note
that that this formula has no free variable for the input of the 1-place
function in question, but then the input is irrelevant . . .
Almost the same formula, v1 = SSS0, serves to represent the set
— i.e. 1-place relation — {3} in Th(A).
Example 18.2. The set of all even numbers is a 1-place relation is
representable in Th(A); ∃v0 v1 = S0 · v1 is a formula representing it.
18. DEFINING RECURSIVE FUNCTIONS IN ARITHMETIC 51
Example 18.3. The projection function π23 can be represented in

Th(A). v2 = v4 is one formula which represents π23; another is ∃v7 (v2 =
v7 ∧ v7 = v4 ).
Problem 18.1. Suppose Σ and Γ are sets of sentences of LN and
Σ ` Γ, i.e. Σ ` γ for every γ ∈ Γ. Then every function and relation
which is representable in Th(Γ) is representable in Th(Σ).
Problem 18.2. Suppose Σ is a set of sentences of LN and f is
a k-place function which is representable in Th(Σ). Then Σ must be
consistent.
It turns out that all recursive functions and relations are repre-
sentable in Th(A).
Problem 18.3. Show that the following functions are representable
in Th(A):
1. The zero function O(n) = 0.
2. The successor function S(n) = n + 1.
3. For every positive k and i ≤ k, the projection function πik .
Proposition 18.4. A k-place function f is representable in Th(A)
if and only if the k + 1-place relation Pf defined by
Pf (n1 , . . . , nk , nk+1 ) ⇐⇒ f(n1 , . . . , nk ) = nk+1
is representable in Th(A).
Also, a relation P ⊆ Nk is representable in Th(A) if and only if its
characteristic function χP is representable in Th(A).
Proposition 18.5. Suppose g1 , . . . , gm are k-place functions and
h is an m-place function, all of them representable in Th(A). Then
f = h ◦ (g1 , . . . , gm ) is a k-place function representable in Th(A).
Proposition 18.6. Suppose g is a k + 1-place regular function
which is representable in Th(A). Then the unbounded minimalization
of g is a k-place function representable in Th(A).
Between them, the above results supply most of the ingredients
needed to conclude that all recursive functions and relations on the
natural numbers are representable. The exception is showing that func-
tions defined by primitive recursion from representable functions are
also representable, which requires some additional effort. The basic
problem is that it is not obvious how a formula defining a function
can get at previous values of the function. To accomplish this, we will
borrow a trick from Chapter 14.
52 18. DEFINING RECURSIVE FUNCTIONS IN ARITHMETIC
Problem 18.7. Show that each of the following relations and func-
tions (first defined in Problem 14.9) is representable in Th(A).
1. Div(n, m) ⇐⇒ n | m
2. IsPrime(n) ⇐⇒ n is prime
3. Prime(k) = pk , where p0 = 1 and pk is the kth prime if k ≥ 1.
4. Power(n, m) = k, where k ≥ 0 is maximal such that nk | m.
5. Length(n) = `, where ` is maximal such that p` | n.
6. Element(n, i) = ni , where n = pn1 1 . . . pnk k (and ni = 0 if i > k).
Using the representable functions and relations given above, we can
represent the “history” function of any representable function . . .
Problem 18.8. Suppose f is a k-place function representable in
Th(A). Show that
f (n ,...,n ,0) f (n ,...,nk ,m)
F (n1, . . . , nk , m) = p1 1 k . . . pm+11
Ym
f (n ,...,n ,i)
= pi 1 k
i=0
is also representable in Th(A).

. . . and use it!
Proposition 18.9. Suppose g is a k-place function and h is a
k + 2-place function, both representable in Th(A). Then the k + 1-
place function f defined by primitive recursion from g and h is also
representable in Th(A).
Theorem 18.10. Recursive functions are representable in Th(A).
In particular, it follows that there are formulas of LN represent-
ing each of the functions from Chapter 17 for manipulating the codes
of formulas. This will permit us to construct formulas which encode
assertions about terms, formulas, and deductions; we will ultimately
prove the Incompleteness Theorem by showing there is a formula which
codes its own unprovability.
CHAPTER 19
The Incompleteness Theorem
By pulling the material in Chapters 16–18 together, we can finally

state and prove the Incompleteness Theorem.
Problem 19.1. Show that A is a recursive set of sentences of LN .
Problem 19.2. Show that the function


pϕ(S 0)q
k
if n = pϕq for a formula ϕ of LN
Sub(n, k) = with at most v1 free

0 otherwise
is recursive.
The key result needed to prove the Incompleteness Theorem is the
following lemma.
Lemma 19.3 (Fixed-Point Lemma). Suppose ϕ is a formula of LN
with only v1 as a free variable. Then there is a sentence σ of LN such
that
A ` σ ↔ ϕ(S pσq0) .
Note that σ must be different from the sentence ϕ(S pσq0): there is
no way to find a formula ϕ with one free variable and an integer k such
that pϕ(S k 0)q = k. (Think about how Gödel codes are defined . . . )
With the Fixed-Point Lemma in hand, Gödel’s Incompleteness The-
orem can be put away in fairly short order.
Theorem 19.4 (Gödel’s Incompleteness Theorem). Suppose Σ is
a consistent recursive set of sentences of LN such that Σ ` A. Then Σ
is not complete.
That is, any consistent set of sentences which proves at least as
much about the natural numbers as A does can’t be both complete
and recursive. The Incompleteness Theorem has many variations and
corollaries; [11] is a good place to learn about many of them.
Problem 19.5. Prove each of the following.
1. Let Γ be a complete set of sentences of LN such that Γ ∪ A is
consistent. Then Γ is not recursive.
53
54 19. THE INCOMPLETENESS THEOREM
2. Let ∆ be a recursive set of sentences such that ∆∪A is consistent.

Then ∆ is not complete.
3. The theory of N,
Th(N) = { σ | σ is a sentence of LN and N |= σ } ,
is not recursive.
There is nothing all that special about working in LN . The proof
of Gödel’s Incompleteness Theorem can be executed for any first order
language and recursive set of axioms which allow one to code and prove
enough facts about arithmetic. In particular, it can be done whenever
the language and axioms are powerful enough — as in Zermelo-Fraenkel
set theory, for example — to define the natural numbers and prove some
modest facts about them.
Gödel also proved a strengthened version of the Incompleteness
Theorem which asserts that a consistent recursive set of sentences Σ of
LN cannot prove its own consistency. To get at it, we need to express
the statement “Σ is consistent” in LN .
Problem 19.6. Suppose Σ is a recursive set of sentences of LN .
Find a sentence of LN , which we’ll denote by Con(Σ), such that Σ is
consistent if and only if A ` Con(Σ).
Theorem 19.7 (Gödel’s Second Incompleteness Theorem). Let Σ
be a consistent recursive set of sentences of LN such that Σ ` A. Then
Σ 0 Con(Σ).
As with the (First) Incompleteness Theorem, the Second Incom-
pleteness Theorem holds for any recursive set of sentences in a first-
order language which allow one to code and prove enough facts about
arithmetic. The perverse consequence of the Second Incompleteness
Theorem is that only an inconsistent set of axioms can prove its own
consistency . . .
The implications. Gödel’s Incompleteness Theorems have pro-

found implications.
Since almost all of mathematics can be formalized in first-order
logic, the First Incompleteness Theorem implies that there is no effec-
tive procedure that will find and prove all theorems. This might be
considered as job security for research mathematicians . . .
The Second Incompleteness Theorem, on the other hand, implies
that we can never be completely sure that any reasonable set of axioms
is actually consistent unless we take a more powerful set of axioms on
faith. It follows that one can never be completely sure — faith aside —
19. THE INCOMPLETENESS THEOREM 55
that the theorems proved in mathematics are really true. This might
be considered as job security for philosophers of mathematics . . .
Truth and definability. A close relative of the Incompleteness
Theorem is the assertion that truth in N = (N, S, +, ·, E, 0) is not
definable in N. To make sense of this, of course, we first need to define
what “definable in N means.
Definition 19.1. A k-place relation is definable in N if there is a
formula ϕ of LN with at most v1, . . . , vk as free variables such that
P (n1 , . . . , nk ) ⇐⇒ N |= ϕ[s(v1|n1 ) . . . (vk |nk )]
for every assignment s of N. Such a formula ϕ is said to define P in
N.
A definition of “function definable in N” could be made in a similar
way, of course. Definability is a close relative of representability:
Proposition 19.8. Suppose P is a k-place relation which is rep-
resentable in Th(A). Then P is definable in N.
Problem 19.9. Is the converse to Proposition 19.6 true?
A counterpart for definability of the Entscheidungsproblem is the
question of whether the truth in N is a definable relation in N, i.e.
whether the set of Gödel codes of sentences of LN true in N,
pTh(N)q = { pσq | σ is a sentence of LN and N |= σ } ,
is definable in N.
Theorem 19.10 (Tarski’s Undefinability Theorem). pTh(N)q is
not definable in N.
56 19. THE INCOMPLETENESS THEOREM
Hints
CHAPTER 10
Hints
10.1. This should be easy . . .

10.2. Ditto.
10.3. 1. Any machine with the given alphabet and a table with
three non-empty rows will do.
2. Every entry in the table in the 0 column must write a 1 in the
scanned cell; similarly, every entry in the 1 column must write a
0 in the scanned cell.
3. What’s the simplest possible table for a given alphabet?
10.4. Unwind the definitions step by step in each case. Not all of
these are computations . . .
10.5. Examine your solutions to the previous problem and, if nec-
essary, take the computations a little farther.
10.6. Have the machine run on forever to the right, writing down
the desired pattern as it goes no matter what may be on the tape
already.
10.7. Consider your solution to Problem 10.6 for one possible ap-
proach. It should be easy to find simpler solutions, though.
10.8. 1. Use four states to write the 1s, one for each.
2. The input has a convenient marker.
3. Run back and forth to move one marker n cells from the block
of 1’s while moving another through the block, and then fill in.
4. Modify the previous machine by having it delete every other 1
after writing out 12n .
5. Run back and forth to move the right block of 1s cell by cell to
the desired position.
6. Run back and forth to move the left block of 1s cell by cell
past the other two, and then apply a minor modification of the
previous machine.
7. Run back and forth between the blocks, moving a marker through
each. After the race between the markers to the ends of their
59
60 10. HINTS
respective blocks has been decided, erase everything and write

down the desired output.
CHAPTER 11
Hints
11.1. This ought to be easy.

11.2. Generalize the technique of Example 11.1, adding two new
states to help with each old state that may cause a move in different
directions. Be careful not to make a machine that would run off the
end of the tape when the original wouldn’t.
11.3. Note that the simulation must operate with coded versions
of Ms tape, unless Σ = {1}. The key idea is to use the tape of the
simulator in blocks of some fixed size, with the patterns of 0s and 1s
in each block corresponding to elements of Σ.
11.4. This should be straightforward, if somewhat tedious. You do
need to be careful in coming up with the appropriate input tapes for
O.
11.5. Generalize the technique of Example 11.2, splitting up the
tape of the simulator into upper and lower tracks and splitting each
state of N into two states in P . You will need to be quite careful in
describing just how the latter is to be done.
11.6. If you’re in doubt, go with one read/write scanner for each
tape, and have each entry in the table of a two-tape machine take
both scanners into account. Simulating such a machine is really just a
variation on the techniques used in Example 11.2.
11.7. Such a machine should be able to move its scanner to cells
up and down from the current one, as well to the side. (Diagonally too,
if you want to!) Simulating such a machine on a single tape machine is
a challenge. You might find it easier to first describe how to simulate
it on a suitable multiple-tape machine.
61
62 11. HINTS
CHAPTER 12
Hints
12.1. Pick as simple a Turing machine as you can think of . . .

12.2. Unwind the representation using Definition 12.2.
12.3. Trick question!
12.4. One could omit the representations of any empty entries in
the table of a Turing machine, for one thing.
12.5. Just use Definition 12.3 in each case.
12.6. Unwind each representation using Definition 12.3.
12.7. For one thing, is the leading 1 in the representation of each
cell really necessary?
12.8. H needs to make exactly two changes to the representation
of the tape position. Note that i = 0 is a special case.
12.9. R needs to check the representation of a particular cell.
12.10. W needs to make just one change to the representation of
the tape position.
12.11. E must first find and then copy the representation of par-
ticular entry in the table of M. Some of the machines in Problem 10.8
may come in handy here.
12.12. Put the machines developed in Problems 12.8–12.11 to-
gether.
12.13. Use the machine S of Problem 12.12 to do most of the work.
12.14. Essentially, all C does is unwind Definition 12.3.
12.15. Assume, by way of contradiction, that there was a Turing
machine T which solved the Halting Problem. Modify T to get a
machine Y which, for any Turing machine M, halts on input 0xMy if
and only if M does not halt on input 0xMy. Feed the input 0xYy to
Y ...
63
64 12. HINTS
CHAPTER 13
Hints
13.1. 1. Delete most of the input.

2. Add a little to the input.
3. Add a little to the input, and delete a little more elsewhere.
4. Delete a little from the input most of the time.
5. Run back and forth between the two blocks in the input, deleting
until one side disappears. Clean up appropriately!
6. Delete two of blocks and move the remaining one.
13.2. There are just as many functions N → N as there are real
numbers, of which there are many more than there are natural numbers.
13.3. 1. Trace the computation through step-by-step.
2. Consider the scores of each of the 1-state entries in the busy
beaver competition.
3. Find a 3-state entry in the busy beaver competition which scores
six.
4. Show how to turn an n-state entry in the busy beaver competi-
tion into an (n + 1)-state entry that scores just a little better.
13.4. You could start by looking at modifications of the 3-state
entry you devised in Problem 13.3.
13.5. Suppose Σ was computable by a Turing machine M. Modify
M to get an n-state entry in the busy beaver competition for some n
which achieves a score greater than Σ(n).
65
66 13. HINTS
CHAPTER 14
Hints
14.1. You only need to take care of the projection functions, and
these can be computed by Turing machines very similar to one another.
14.2. Generalize Example 14.1.
14.3. Use machines computing g, h1, . . . , hm as sub-machines of
the machine computing the composition. You might also find sub-
machines that copy the original input and various stages of the output
useful. It is important that each sub-machine get all the data it needs
and does not damage the data needed by other sun-machines.
14.4. Proceed by induction on the number of applications of com-
position used to define f from the initial functions.
14.5. 1. Exponentiation is to multiplication as multiplication is
to addition.
2. This is straightforward except for taking care of Pred(0) =
Pred(1) = 0.
3. Diff is to Pred as S is to Sum.
4. This is straightforward if you let 0! = 1.
14.6. Machines used to compute g and h are the principal parts
of the machine computing f, along with parts to copy, move, and/or
delete data on the tape between stages in the recursive process.
14.7. 1. f is to g as Fact is to the identity function.
2. Use Diff and a suitable constant function as the basic building
blocks.
3. This is a slight generalization of the preceding part.
14.8. Proceed by induction on the number of applications of prim-
itive recursion and composition.
14.9. 1. Use a composition including Diff, χP , and a suitable
constant function.
2. A suitable composition will do the job; it’s a little harder than
it looks.
67
68 14. HINTS
3. A suitable composition will do the job; it’s much more straight-

forward than the previous part.
4. Note tht n = m exactly when n − m = 0 = m − n.
5. Compare this with Problem 14.5.
6. First devise a characteristic function for the relation
Divides(n, k, m) ⇐⇒ nk = m ,
and then sum up.
7. Use Div and sum up.
8. Use IsPrime and some ingenuity.
9. Use Exp and Div and some more ingenuity.
10. A suitable combination of Prime with other things will do.
11. A suitable combination of Prime and Power will do.
12. Throw the kitchen sink at this one . . .
13. Ditto.
14.10. For the hard direction, do an induction on how g was built
up from the initial functions.
14.11. A straightforward application of Theorem 14.10.
14.12. This is not unlike, though a little more complicated than,
showing that primitive recursion preserves computability.
14.13. It’s not easy! Look it up . . .
14.14. This is a very easy consequence of Theorem 14.13.
14.15. Listing the definitions of all possible primitive recursive
functions is a computable task. Borrow a trick from Cantor’s proof
that the real numbers are uncountable. (A formal argument to this ef-
fect could be made using techniques similar to those used to show that
all Turing computable functions are recursive in the next chapter.)
CHAPTER 15
Hints
15.1. The strategy is obvious . . . Make sure that at each stage you
preserve a copy of the original input for use at later stages.
15.2. The primitive recursive function you define only needs to
check values of g(n1 , . . . , nk , m) for m such that 0 ≤ m ≤ h(n1 , . . . , nk ),
but it still needs to pick the least m such that g(n1 , . . . , nk , m) = 0.
15.3. This is very similar to Theorem 14.8.
15.4. This is virtually identical to Theorem 14.10.
15.5. This is virtually identical to Corollary 14.11.
15.6. In both cases, emulate Example 15.1.
15.7. Unwind Definition 15.5; you will need to do some factoring.
15.8. Find the codes of each of the positions in the sequence you
chose and then apply Definition 15.6.
15.9. 1. It will probably be convenient to first devise a function
which recognizes whether the input is of the correct form or not.
You may find it convenient to first show that following relation
is primitive recursive:
• TapePos, where TapePos(n) ⇐⇒ n is the code of a
tape position.
If the input is of the correct form, make the necessary changes
to n using the tools in Problem 14.9.
2. Piece TMM together by cases using the function Entry in each
case. You may wish to look back to the construction of an uni-
versal Turing machine in Chapter 12 for some idea of what needs
to be done.
3. You may find it convenient to first show that following relation
is primitive recursive:
• TapePosSeq, where TapePosSeq(n) ⇐⇒ n is the code
of a sequence of tape positions.
Use the function TMM to check that a sequence of tape positions
is a computation.
69
70 15. HINTS
15.10. The last part of Problem 15.9 and some unbounded mini-
malization are the key ingredients. You may also find Theorem 15.4
useful if you show that the following functions are recursive:
• Codek (n1 , . . . , nk ) = p(0, 1, 01n1 0 . . . 01nk )q for any fixed k ≥ 1.
• Decode(t) = n if t = p(i, k, 01n+1 )q (and anything you like
otherwise).
15.11. Take some creative inspiration from Definitions 15.5 and
15.6. Fpr example, if (s, i) ∈ dom(M) and M(s, i) = (j, d, t), you
could let the code of M(s, i) be
pM(s, i)q = 2s 3i 5j 7d+1 11t .
15.12. Much of what you need for both parts is just what was
needed for Problem 15.9. The additional ingredients mainly have to do
with using m = pMq properly.
15.13. Essentially, this is to Problem 15.12 as proving Theorem
15.10 is to Problem 15.9.
15.14. Use χP to help define a function f such that im(f) = P .
15.15. One direction is an easy application of Proposition 15.14.
For the other, given an n ∈ N, run the functions enumerating P and
N \ P concurrently until one or the other outputs n.
15.16. Consider the set of natural numbers coding (according to
some scheme you must devise) Turing machines together with input
tapes on which they halt.
15.17. See how far you can adapt your argument for Proposition
15.15.
CHAPTER 16
Hints
16.1. Compare Definition 16.2 with the definition of maximal con-

sistency.
71
72 16. HINTS
CHAPTER 17
Hints
17.1. In each case, use Definitions 17.1 and 17.2, together with the
appropriate definitions from first-order logic and the tools developed
in Problem 14.9.
17.2. In each case, use Definitions 17.1 and 17.2, together with the
appropriate definitions from first-order logic and the tools developed
in Problems 14.9 and 17.1. (They’re all primitive recursive if p∆q is,
by the way.)
17.3. 1. Use unbounded minimalization and the relations in
Problem 17.2 to define a function which, given n, returns the
nth integer which codes an element of Th(∆).
2. If ∆ is complete, then for any sentence σ, either dσe or d¬σ
must eventually turn up in an enumeration of pTh(∆)q. The
other direction is really just a matter of unwinding the definitions
involved.
73
74 17. HINTS
CHAPTER 18
Hints
18.1. Every deduction from Γ can be replaced by a dedudction of

Σ with the same conclusion.
18.2. If Σ were insconsistent it would prove entirely too much . . .
18.3. 1. Adapt Example 18.1.
2. Use the 1-place function symbol S of LN .
3. There is much less to this part than meets the eye . . .
18.4. In each case, you need to use the given representing formula
to define the one you need.
18.5. String together the formulas representing g1 , . . . , gm , and h
with ∧s and put some existential quantifiers in front.
18.6. First show that that < is representable in Th(A) and then
exploit this fact.
18.7. 1. n | m if and only if there is some k such that n · k = m.
2. n is prime if and only if there is no ` such that ` | n and 1 < ` < n.
3. pk is the first prime with exactly k − 1 primes less than it.
4. Note that k must be minimal such that nk+1 - m.
5. You’ll need a couple of the previous parts.
6. Ditto.
18.8. Problem 18.7 has most of the necessary ingredients needed
here.
18.9. Problems 18.7 and 18.8 have most of the necessary ingredi-
ents between them.
18.10. Proceed by induction on the numbers of applications of com-
position, primitive recursion, and unbounded minimalization in the re-
cursive definition f, using the previous results in Chapter 18 at the
basis and induction steps.
75
76 18. HINTS
CHAPTER 19
Hints
19.1. A is a finite set of sentences.

19.2. First show that recognizing that a formula has at most v1 as
a free variable is recursive. The rest boils down to checking that sub-
stituting a term for a free variable is also recursive, which has already
had to be done in the solutions to Problem 17.1.
19.3. Let ψ be the formula (with at most v1, v2 , and v3 free) which
represents the function f of Problem 19.2 in Th(A). Then the formula
∀v3 (ψ v2 v1 → ϕvv13 ) has only one variable free, namely v1 , and is very
close to being the sentence σ needed. To obtain σ you need to substitute
S k O for a suitable k for v1.
19.4. Try to prove this by contradiction. Observe first that if Σ is
recursive, then pTh(Σ)q is representable in Th(A).
19.5. 1. If Γ were recursive, you could get a contradiction to
the Incompleteness Theorem.
2. If ∆ were complete, it couldn’t also be recursive.
3. Note that A ⊂ Th(N).
19.6. Modify the formula representing the function ConclusionΣ
(defined in Problem 17.2) to get Con(Σ).
19.7. Try to do a proof by contradiction in three stages. First,
find a formula ϕ (with just v1 free) that represents “n is the code of
a sentence which cannot be proven from Σ” and use the Fixed-Point
Lemma to find a sentence τ such that Σ ` τ ↔ ϕ(S pτ q). Second, show
that if Σ is consistent, then Σ 0 τ . Third — the hard part — show
that Σ ` Con(Σ) → ϕ(S pτ q). This leads directly to a contradiction.
19.8. Note that N |= A.
19.9. If the converse was true, A would run afoul of the (First)
Incompleteness Theorem.
19.10. Suppose, by way of contradiction, that pTh(N)q was defin-
able in N. Now follow the proof of the (First) Incompleteness Theorem
as closely as you can.
77
78 19. HINTS
Bibliography
[1] Jon Barwise (ed.), Handbook of Mathematical Logic, North Holland, Amster-
dam, 1977, ISBN 0-7204-2285-X.
[2] C.C. Chang and H.J. Keisler, Model Theory, third ed., North Holland, Ams-
terdam, 1990.
[3] Martin Davis, Computability and Unsolvability, McGraw-Hill, New York, 1958;
Dover, New York, 1982, ISBN 0-486-61471-9.
[4] Martin Davis (ed.), The Undecidable; Basic Papers On Undecidable Propo-
sitions, Unsolvable Problems And Computable Functions, Raven Press, New
York, 1965.
[5] Herbert B. Enderton, A Mathematical Introduction to Logic, Academic Press,
New York, 1972.
[6] Douglas R. Hofstadter, Gödel, Escher, Bach, Random House, New York, 1979,
ISBN 0-394-74502-7.
[7] Jerome Malitz, Introduction to Mathematical Logic, Springer-Verlag, New
York, 1979, ISBN 0-387-90346-1.
[8] Yu.I. Manin, A Course in Mathematical Logic, Graduate Texts in Mathemat-
ics 53, Springer-Verlag, New York, 1977, ISBN 0-387-90243-0.
[9] Roger Penrose, The Emperor’s New Mind, Oxford University Press, Oxford,
1989.
[10] T. Rado, On non-computable functions, Bell System Tech. J. 41 (1962), 877–
884.
[11] Raymond M. Smullyan, Gödel’s Incompleteness Theorems, Oxford University
Press, Oxford, 1992, ISBN 0-19-504672-2.
79
80 BIBLIOGRAPHY
Index
A, 49 IsPrime, 33, 52
Con(Σ), 54 Length, 33, 52
p∆q, 46 Logical, 46
ϕ(S m1 0, . . . , S mk 0), 50 Mult, 31
f : Nk → N, 25 O, 27, 29, 51
iN , 26 Power, 33, 52
LN , 43 Pred, 27, 31
N, 25 Premiss∆ , 46
Nk \ P , 32 Prime, 33, 52
Nk , 25 Sentence, 46
N, 44 Sim, 39
P ∩ Q, 32 Subseq, 33
P ∪ Q, 32 Sub, 53
P ∧ Q, 32 Sum, 27, 30
P ∨ Q, 32 S, 27, 29, 51
S m 0, 50 TapePosSeq, 69
¬P , 32 TapePos, 69
πik , 29, 51 Term, 46
Th(N), 54 TM, 39
Th(Σ), 44 TMM , 38
A, 33 Ackerman’s Function, 33
α, 33 alphabet, 7, 13
Codek , 70 blank
Comp, 39 cell, 7
CompM , 38 tape, 7
Conclusion∆ , 46 bounded minimalization, 36
Decode, 70 busy beaver competition, 27
Deduction∆ , 46 n-state entry, 27
Diff, 27, 31 score, 27
Div, 33, 52
Element, 33, 52 cell
Entry, 38 blank, 7
Equal, 33 marked, 7
Exp, 31 scanned, 8
Fact, 31 characteristic function, 26
Formulas, 46 Church’s Thesis, 2
Formula, 46 code, 20
Inference, 46 sequence of tape positions, 37
81
82 INDEX
tape position, 37 (First) Incompleteness Theorem, 53

Turing machine, 38 Second Incompleteness Theorem, 54
complete, 2
set of sentences, 44 halt, 11
completeness, 44 Halting Problem, 17, 22
Completeness Theorem, 1
identity function, 26
composition, 29
image of a function, 25
computable
Incompleteness Theorem, 43
function, 26
Gödel’s First, 53
set of formulas, 46
Gödel’s Second, 54
computation, 11
initial function, 29
partial, 11
input tape, 11
constant function, 30
k-place
decision problem, 1
function, 25
definable
relation, 25
function, 55
relation, 55 language
domain of a function, 25 first-order number theory, 43
Entscheidungsproblem, 1, 43 machine, 9
first-order language Turing, 7, 9
for number theory, 43 marked cell, 7
Fixed-Point Lemma, 53 minimalization
function bounded, 36
bounded minimalization of, 36 unbounded, 35
composition of, 29
natural numbers, 25
computable, 26
n-state
constant, 30
entry in busy beaver competition,
definable in N, 55
27
domain of, 25
Turing machine, 9
identity, 26
number theory
initial, 29
first-order language for, 43
k-place, 25
partial, 25 output tape, 11
primitive recursion of, 30
primitive recursive, 31 partial
projection, 29 computation, 11
recursive, 36 function, 25
regular, 35 pig, yellow, 19
successor, 29 position
Turing computable, 26 tape, 8
unbounded minimalization of, 35 primitive recursion, 30
zero, 29 primitive recursive
function, 31
Gödel code relation, 32
sequences, 45 projection function, 29
symbols of LN , 45
Gödel Incompleteness Theorem, 43 r.e., 39
INDEX 83
recursion Turing computable

primitive, 30 function, 26
recursive relation, 36
function, 36 Turing machine, 7, 9
relation, 36 code of, 38
set of formulas, 46 n-state, 9
recursively enumerable, 39 representation of, 18
set of formulas, 46 table, 10
regular function, 35 universal, 17, 22, 38
relation two-way infinite tape, 13, 14
characteristic function, 26
definable in N, 55 unary notation, 26
k-place, 25 unbounded minimalization, 35
primitive recursive, 32 Undefinability Theorem
recursive, 36 Tarski’s, 55
Turing computable, 36 universal Turing machine, 17, 22, 38
representable UTM, 17
function, 50 zero function, 29
relation, 50
representation
of a tape position, 19
of a Turing machine, 18
scanned cell, 8
scanner, 13
score
busy beaver competition, 27
state, 8, 9
successor
function, 29
tape position, 11
table
Turing machine, 10
tape, 7, 13
blank, 7
input, 11
output, 11
two-way infinite, 13, 14
tape position, 8
code of, 37
code of a sequence of, 37
representation of, 19
successor, 11
Tarski’s Undefinability Theorem, 55
theory
of N, 54
of a set of sentences, 44
TM, 9

A Problem Course in Mathematical Logic: Stefan Bilaniuk

Uploaded by

Copyright:

Available Formats

A Problem Course in Mathematical Logic: Stefan Bilaniuk

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

A Problem Course in Mathematical Logic: Stefan Bilaniuk

Uploaded by

Copyright:

Available Formats

A Problem Course

A Problem Course in Mathematical Logic

Abstract. This is the Volume II of a text for a problem-oriented

Chapter 16. Hints 71

This book is intended to be the basis for a problem-oriented full-year

Acknowledgements. Various people and institutions deserve the

Availability. The URL of the home page for A Problem Course

Author’s Opinion. It’s not great, but the price is right!

The Entscheidungsproblem. Recall that a logic satisfies the

whether such a set of axioms is complete; i.e. given any sentence ϕ of

introductory course in algebra, analysis, or discrete mathematics ought

Of the various ways to formalize the notion an “effective method”,

Example 10.1. A blank tape looks like:

Conventions for tapes. Unless stated otherwise, we will assume

Similarly, if σ is a finite sequence of elements of Σ ∪ {0}, we may write

Computations. Informally, a computation is a sequence of actions

Variations and Simulations

The definition of a Turing machine given in Chapter 10 is arbitrary

Example 11.1. Consider the following Turing machine:

one-way infinite tape. To do this, we choose an alphabet for O with

Problem 11.7. Give a precise definition for Turing machines with

Universal Turing Machines and the Halting

In Chapter 11 we devised techniques for constructing, given a par-

of simply listing the table defining the Turing machine we wish to

The representation of the machine M, xMy, then consists of an

Problem 12.1. Pick a Turing machine different from the one in

Problem 12.3. How many possible representations are there for a

Turing machine H which on input

Theorem 12.13. There is a Turing machine U such that, for any

with output 011 if M halts on input a, and with output 01 if M does

Computable and Non-Computable Functions

Notation and conventions. To keep things as simple as pos-

The basic convention for representing natural numbers on the tape

Computable functions. With suitable conventions for represent-

n-state entry in the busy beaver competition for every n ≥ 0 , it follows

Primitive Recursive Functions

Starting with a small set of computable functions, and applying

To see that this works, one can proceed by induction on m:

So we already know that all the initial functions, addition, and

4. Equal, where Equal(n, Pm m) ⇐⇒ n = m.

We add one more computable method of building functions, un-

Proposition 15.1. If g is a Turing computable regular (k + 1)-

Theorem 15.4. A k-place g is recursive if and only if the 1-place

Having defined how to represent tape positions as integers, we now

1. The 2-place function TM, where

It was mentioned in the Introduction that one of the motivations for

Coding First-Order Logic

We will encode the symbols, formulas, and deductions of LN as

Note. It follows that pTh(∆)q is an example of a recursively enu-

Defining Recursive Functions In Arithmetic

We will also need definitions and results complementing those ob-

∀y (ϕxy → ϕxSy ) hold, then so does ∀x ϕ. On the other hand, neither LN

Example 18.3. The projection function π23 can be represented in

is also representable in Th(A).

The Incompleteness Theorem

By pulling the material in Chapters 16–18 together, we can finally

2. Let ∆ be a recursive set of sentences such that ∆∪A is consistent.

The implications. Gödel’s Incompleteness Theorems have pro-

10.1. This should be easy . . .