3528031379

i
ii
iii
iv
v
Voor Jeannette, Nils en Noah
Für Marlis
vii
Preface
Algebraic geometry is, loosely speaking, concerned with the study of zero sets of
polynomials (over an algebraically closed field). As one often reads in prefaces of intro-
ductory books on algebraic geometry, it is not so easy to develop the basics of algebraic
geometry without a proper knowledge of commutative algebra. On the other hand, the
commutative algebra one needs is quite difficult to understand without the geometric
motivation from which it has often developed.
Local analytic geometry is concerned with germs of zero sets of analytic functions,
that is, the study of such sets in the neighborhood of a point. It is not too big a surprise
that the basic theory of local analytic geometry is, in many respects, similar to the basic
theory of algebraic geometry. It would, therefore, appear to be a sensible idea to develop
the two theories simultaneously. This, in fact, is not what we will do in this book, as the
“commutative algebra” one needs in local analytic geometry is somewhat more difficult:
one has to cope with convergence questions. The most prominent and important example
is the substitution of division with remainder. Its substitution in local analytic geometry
is called the Weierstraß Division Theorem.
The above remarks motivated us to organize the first four chapters of this book
as follows. In Chapter 1 we discuss the algebra we need. Here, we assume the reader
attended courses on linear algebra and abstract algebra, including some Galois theory.
Probably the reader could just start with Chapter 2, referring to Chapter 1 only when
needed. Chapter 2 deals with the basics of affine algebraic geometry, up to say, Hilbert’s
Nullstellensatz and decomposition into irreducible components. In Chapter 3 we tackle
the corresponding basics for local analytic geometry. Here it is assumed that the reader
has knowledge of the theory of holomorphic functions in one variable. Although Chapter 4
is written in the language of local analytic geometry, most statements and proofs make
sense for affine algebraic geometry, too.
As an application of the general theory, in Chapter 5 we study the “simplest” germs
of local analytic spaces: plane curve singularities. Topics here are Puiseux expansion,
semigroups of curves, and resolutions of plane curve singularities.
Many of the topics in the rest of the book cannot usually be found in books on local
analytic geometry. The principle of conservation of number in Chapter 6 is such an exam-
ple. This is about interpretation of invariants in “families”. The simplest example of this
is the intersection number of two plane curve singularities. This intersection number can
be interpreted as the number of intersection points appearing after slightly “perturbing”
the singularities. To put the proofs in their proper context, we discuss and prove the
basic coherence theorems of Oka and Cartan, and the finite mapping theorem of Grauert
and Remmert. Next in line is the theory of standard bases in a power series ring, due to
Hironaka and Grauert. The corresponding notion in a polynomial ring is called Gröbner
basis, and has received much attention lately because of its applications in computer
algebra. We refer to the computer algebra system Singular [Singular 2000].1 One of
the main ideas is that standard bases allow a well-defined representative of an element
f modulo an ideal I, called the normal form of f . In particular, this normal form is zero
1 Our advise to the reader is to compute many examples of the theory in this book by means of a
computer algebra system, which can calculate in localizations of polynomials rings.
viii Preface
if and only if f is an element of I.

Chapter 8 is devoted to approximation theorems. Most famous, and the easiest to
prove, is the Artin Approximation Theorem. It states that if one has a formal solution
(that is, without regarding the convergence) of a set of analytic equations, then there
exists an analytic, that is, convergent solution to these equations. Further, a formal solu-
tion can be arbitrarily approximated by an analytic one. For applications it is sometimes
necessary that some solutions do not depend on some of the variables. This so-called
nested approximation theorem does not hold in general, however. Nevertheless, it does
hold under stronger assumptions. This is called Grauert’s Approximation Theorem.
In Chapter 9 we give the classification of simple hypersurface singularities in all
its details. One of the important tools here is the finite determinacy theorem. The fact
that an isolated hypersurface singularity is finitely determined is quite easy to prove: one
can prove it by an application of Newton’s Lemma, which is a “souped-up” version of
the implicit function theorem. For a sharper statement we use the Artin Approximation
Theorem. Finally, in Chapter 10, we give, as application of Grauert’s Approximation
Theorem, a proof of the existence of a semi-universal deformation of an isolated singu-
larity.
Some proofs in this book are new, or at least we have not seen them in this form in
the literature. We mention:
• The proof of the convergence of the Puiseux expansion by using Newton’s Lemma.
• The proof of the inversion theorem for Puiseux pairs.
• Direct construction of factors of a power series in two variables, without using the
Puiseux expansion.
• Certain parts of the proof of Grauert’s Approximation Theorem, in particular the
proof of Cartan’s Lemma by using standard bases.
• The fact that the resolution graph determines the semigroup of the curve.
• Parts of the proof of the Mather-Yau Theorem.
Assuming the reader has the knowledge mentioned above, it was our intention to
give either full proofs of all the statements in the book, or to put them into exercises
with sufficient hints. Except for the descriptive third section of Chapter 8, we hope to
have succeeded in this goal.
Acknowledgment: Various people read parts of the manuscript and gave us useful com-
ments. We mention Holger Cröni, Anne Frühbis Krüger, Tobias Hirsch, Rainer Kaenders
and Dorin Popescu. We thank Thomas Keilen, who also made some of the pictures for
us. We thank Pauline Bitsch for typing a large part of the manuscript. In particular we
thank Olaf Bachmann for helping us with LATEX, and explaining us the use of the cvs
system.
Last but not least, we thank our wives Jeannette and Marlis for being so patient
with us during the three years we were writing this book.
ix
Introductory Guide
2 3
6 5 7
8.1 8.2
8.3
9 10
x
Contents
Preface vii
Introductory Guide ix
Contents x
1 Algebra 1
1.1 Noetherian Rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3 Local Rings and Localization . . . . . . . . . . . . . . . . . . . . . . . 17
1.4 Primary Decomposition . . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.5 Finite and Integral Extensions . . . . . . . . . . . . . . . . . . . . . . 34
2 Affine Algebraic Geometry 46

2.1 Affine Hypersurfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
2.2 Affine Varieties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
2.3 Maps between Algebraic Sets . . . . . . . . . . . . . . . . . . . . . . . 66
3 Basics of Analytic Geometry 73

3.1 Holomorphic Functions of Several Complex Variables . . . . . . . . . 75
3.2 Weierstraß Division and Preparation Theorem . . . . . . . . . . . . . 86
3.3 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
3.4 Germs of Analytic Spaces . . . . . . . . . . . . . . . . . . . . . . . . . 107
4 Further Development of Analytic Geometry 124

4.1 Dimension Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
4.2 Hilbert-Samuel Function and Multiplicity . . . . . . . . . . . . . . . . 135
4.3 Regular Local Rings and the Jacobian Criterion . . . . . . . . . . . . 148
4.4 Normalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
5 Plane Curve Singularities 169

5.1 Puiseux Expansion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
5.2 Invariants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
5.3 Resolutions of Irreducible Plane Curve Singularities . . . . . . . . . . 193
5.4 Reducible Plane Curve Singularities . . . . . . . . . . . . . . . . . . . 208
6 The Principle of Conservation of Number 219

6.1 Sheaves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
6.2 Fundamental Properties of Coherent Sheaves . . . . . . . . . . . . . . 230
6.3 The Four Basic Coherence Theorems . . . . . . . . . . . . . . . . . . . 238
6.4 The Principle of Conservation of Number . . . . . . . . . . . . . . . . 243
6.5 Cohen-Macaulay Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . 252
xi
7 Standard Bases 269

7.1 The Division Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 270
7.2 Characterizations and Properties of Standard Bases . . . . . . . . . . 275
7.3 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282
8 Approximation Theorems 286

8.1 Artin’s Approximation Theorem . . . . . . . . . . . . . . . . . . . . . 286
8.2 Grauert’s Approximation Theorem . . . . . . . . . . . . . . . . . . . . 290
8.3 Some other Approximation Theorems . . . . . . . . . . . . . . . . . . 299
9 Classification of Simple Hypersurface Singularities 302

9.1 Finite Determinacy of Hypersurface Singularities . . . . . . . . . . . . 303
9.2 The A-D-E–singularities are simple. . . . . . . . . . . . . . . . . . . . 313
9.3 Orbits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322
10 Deformations of Singularities 329

10.1 Deformations of Functions . . . . . . . . . . . . . . . . . . . . . . . . . 330
10.2 Deformations of Singularities . . . . . . . . . . . . . . . . . . . . . . . 333
10.3 Existence of a Semi-Universal Deformation . . . . . . . . . . . . . . . 346
Bibliography 361
1
1 Algebra
In the first chapter of the book we collect some facts from algebra which we will use in
this book. We assume that the reader is familiar with the concepts of ring, ideal in a
ring, principal ideal domain, prime ideal, unique factorization domain, etc. Throughout
this book, a ring will mean a commutative ring with 1. All homomorphisms of rings
always take 1 to 1. The first topic we discuss in this chapter is Noetherian rings, that
is, rings in which every ideal is finitely generated. The main thing to prove here is the
Hilbert Basis Theorem. This says that the polynomial ring K[x1 , . . . ,xn ], with K a field,
is Noetherian. Then we will discuss the basic notions of the theory of modules over a
ring, a notion which is analogous to vector spaces over a field. In the third section we
will discuss local rings, that is, rings which have a unique maximal ideal. They play
a very important role in this book. For us, the most important example will appear
in the third chapter, the power series ring in n variables C {x1 , . . . ,xn }. Localization
of rings will be defined. Geometric interpretations of localization will be deferred to
Chapter two. In section four we discuss primary decomposition of ideals in Noetherian
rings. Geometrically, this corresponds to the decomposition of an algebraic or analytic
set into irreducible components: these notions will also be discussed in chapters two and
three. Algebraically, it can be seen as a generalization of unique factorization in Z, or in
K[x1 , . . . ,xn ]. In the final section of this chapter we treat ring extensions. The notions of
finite and algebraic extensions of fields here correspond to finite and integral extensions
of rings. A reduced ring is called normal (or integrally closed), if every element of the
total quotient ring Q(R) which is integral over R is already in R. The most important
examples are K[x1 , . . . ,xn ], and, as we will see in Chapter three, the power series rings
C {x1 , . . . ,xn }. We will prove the Cayley-Hamilton Theorem, and the very important
theorem of the finiteness of integral closure, which is due to Dedekind and Kronecker. It
says that, under some weak hypothesis, the normalization of a ring R is finitely generated
as an R–module. This is a nontrivial statement, and it is only because of this proof that
we need some Galois theory as prerequisite.
1.1 Noetherian Rings
Definition 1.1.1. A ring R is called Noetherian if every ideal in R is finitely generated.

This means that for every idealP I ⊂ R there exist finitely many elements f1 , . . . ,fs
s
in I such that I = (f1 , . . . ,fs ) = { ν=1 aν fν : aν ∈ R}. In particular, principal ideal
domains (domains in which every ideal can be generated by one element, for example
fields) are Noetherian. But the Noetherian condition is much less restrictive, and almost
all rings we consider in this book are Noetherian. Before giving some more examples we
give the following characterizations of Noetherian rings.
Lemma 1.1.2. Let R be a ring. Then the following conditions are equivalent:
(1) R is Noetherian.
2 1 Algebra
(2) Any chain of ideals in R
I1 ⊂ I2 ⊂ · · · ⊂ Ik ⊂ · · ·
becomes stationary, that is, there exists an n > 0 such that Ik = In for all k ≥ n.
(3) Every nonempty set of ideals in R has a maximal element with respect to inclusion.
Proof.
(1) =⇒ (2) It is easily checked that the union ∪∞ k=1 Ik is an ideal again, which by as-
sumption is finitely generated, say by a1 , . . . ,as . For n big enough, one has ai ∈ In for
i = 1, . . . ,s. Then Ik = In for all k ≥ n.
(2) =⇒ (1) Suppose that there exists an ideal I in R which is not finitely generated.
Then we can inductively define elements as follows:
• Take a1 ∈ I.
• Given a1 , . . . ,ak−1 take an element ak ∈ I \ (a1 , . . . ,ak−1 ).
But then the chain (a1 ) ⊂ (a1 ,a2 ) ⊂ · · · is not stationary, which is a contradiction.
(2) =⇒ (3) If (3) were false, then there would exist a set of ideals such that for any ideal
Ik in this set we can find an ideal Ik+1 with Ik $ Ik+1 . This would lead to a chain of
ideals which is not stationary, a contradiction.
(3) =⇒ (2) Apply (3) to the set of ideals Ik for k ≥ 1. If In is a maximal element, then
Ik = In for all k ≥ n.
Theorem 1.1.3 (Hilbert’s Basis Theorem). Let R be a Noetherian ring. Then the poly-
nomial ring R[x] is also Noetherian.
Proof. By contradiction. Let I ⊂ R[x] be an ideal in R[x] which is not finitely generated.
We then inductively define a sequence of polynomials f1 ,f2 , . . . in I by:
• Let f1 ∈ I be a polynomial of minimal degree.
• Given f1 , . . . ,fk−1 , let fk be a polynomial of minimal degree in I \ (f1 , . . . ,fk−1 ).
Let ak be the leading coefficient of fk . Then we claim that for all k ≥ 2:
(a1 , . . . ,ak−1 ) $ (a1 , . . . ,ak )
which would give us a chain of ideals in R which does not stabilize, in contradiction to the
assumption that R is Noetherian. Suppose on the contrary that ak ∈ (a1 , . . . ,ak−1 ) for
some k. We then can write ak = b1 a1 + . . . + bk−1 ak−1 for some bi ∈ R. The polynomial
fk − b1 xdeg(fk )−deg(f1 ) f1 − · · · − bk−1 xdeg(fk )−deg(fk−1 ) fk−1
is in I \ (f1 , . . . ,fk−1 ), but has degree smaller than fk , in contradiction to the choice of
fk .
The following corollary gives us the first class of examples of Noetherian rings using
the fact that fields are Noetherian.
1.1 Noetherian Rings 3
Corollary 1.1.4. Let R be a Noetherian ring, and I ⊂ R[x1 , . . . ,xn ] be an ideal. Then
the quotient ring R[x1 , . . . ,xn ]/I is Noetherian.
Proof. Using induction and Theorem 1.1.3 we obtain that R[x1 , . . . ,xn ] is Noetherian.
The rest is given as Exercise 1.1.20.
Definition 1.1.5.
(1) Let R be a ring. An R–algebra S is a ring S together with a ring homomorphism

ϕ : R −→ S. Via this ring homomorphism we can define multiplication with scalars
r · s := ϕ(r)s. Alternatively, one can define an R–algebra S to be a ring S which is
at the same time an R–module, with the condition that r · (st) = (rs) · t for r,s ∈ R
and t ∈ S.
(2) Let R be a ring and S = R[x1 , . . . ,xn ]/I be the quotient of the polynomial ring
by an ideal I. Then the R–algebra S is called a ring of finite type over R or,
alternatively, an affine algebra over R.
From Corollary 1.1.4 we obtain that affine algebras over a Noetherian ring (or in
another language, rings of finite type over a Noetherian ring) are Noetherian. Our most
important case is the case of affine algebras over a field (which is Noetherian).
Now we will investigate another special class of Noetherian rings.
Definition 1.1.6. Let K be a field and R be a K–algebra. R is called an Artinian
K–algebra if R, considered as K–vector space, has dimK (R) < ∞.
Lemma 1.1.7. Let R be an Artinian K–algebra. Then R is Noetherian.
Proof. Let I ⊂ R be an ideal, then I is a finite-dimensional K–vector space because R
is a finite-dimensional K–vector space. Any K–basis of I, in particular, generates I as
ideal and, therefore, I is finitely generated.
Lemma 1.1.8. Let R be an Artinian K–algebra. Then any chain of ideals in R
I1 ⊃ I2 ⊃ . . . ⊃ Ik ⊃ . . .
becomes stationary, that is, there exists an n > 0 such that Ik = In for all k ≥ n.1
Proof. A chain of ideals is especially a chain of K–vector spaces. Such a chain becomes
stationary because R is a finite-dimensional K–vector space.
Examples 1.1.9. Let K be a field.
(1) K[x]/(xn ), n ≥ 1, is an Artinian K–algebra.

(2) Let A be an Artinian K–algebra and I ⊂ A be an ideal. Then A/I is an Artinian
K–algebra.
At the end of this section, we introduce the notion of graded rings and prove some
other useful tools.
1 This is the usual way to define an Artinian ring. We need later on only Artinian K–algebras in the
sense of 1.1.6 which makes proofs a little simpler. Note that C [x]/(x2 ) is a Q–algebra which is an
Artinian ring, but not an Artinian Q–algebra in the sense of our definition.
4 1 Algebra
Definition 1.1.10. Let G be an abelian group. A G–graded ring is a ring R together

with a direct sum decomposition
M
(1.1) R= Rg ,
g∈G
such that
(1.2) Rg · Rh ⊂ Rg+h for all g,h ∈ G.
The elements of Rg are called homogeneous of degree g. A Z–graded ring is simply called
a graded ring.
Examples 1.1.11.
(1) Let S be a ring (for example a field), and R := S[x1 , . . . ,xn ] be the polynomial ring
in n variables with coefficients in S. Then R is a graded ring, by defining
( )
X
ν1 νn
Rd := cν1 ,...,νn x1 . . . xn ; cν1 ,...,νn ∈ S , and Rd = (0) for d < 0.
ν1 +···+νn =d
An element of Rd is called a homogeneous polynomial of degree d. Consider f ∈ R.

Then we can write
f = f0 + f1 + . . . + fd , fi ∈ Ri , fd 6= 0
in a unique way. We call d =: deg(f ) the degree of the polynomial f .
(2) More generally, consider the polynomial ring R := S[x1 , . . . ,xn ] as above and let
λ = (λ1 , . . . ,λn ) ∈ Nn . We call f ∈ S[x1 , . . . ,xn ] quasi-homogeneous of degree d
with respect to the weight λ if f can be written as
X
f= cν1 ,...,νn xν11 . . . xνnn .
ν1 λ1 +...+νn λn =d
Again, R = ⊕d≥0 Rd , where Rd consists of all quasi-homogeneous polynomials of

degree d.
Let f ∈ R, f = f0 + · · · + fd , fi ∈ Ri , fd 6= 0. If we wish to distinguish between
the usual degree and the quasi-homogeneous degree, we call d =: w-deg(f ) (the
weighted degree).
Theorem 1.1.12 (Chinese Remainder Theorem). Let R be a ring and I1 , . . . ,Im be
ideals in R. Assume that
m
(1) ∩ Iν = (0),
ν=1
(2) Iν + Iµ = R for ν 6= µ.
Then the canonical map
m
M
R −→ R/Iν ,
ν=1
x 7→ (x + I1 , . . . ,x + Im )
is an isomorphism of rings.
1.1 Noetherian Rings 5
Proof. The first condition implies that the map is injective. To prove the surjectivity it is
enough to see that for all j there exists an x ∈ R such that x ∈ Iν for all ν 6= j and x ≡ 1
mod Ij . Without loss of generality, it suffices to prove this for j = 1. By assumption
we have Iν + I1 = R for all ν 6= 1. In particular we can find xν ∈ Iν and yν ∈ I1 for
ν = 2, . . . ,m such that xν + yν = 1. We define
m
Y m
Y
x := xν = (1 − yν ).
ν=2 ν=2
As xν ∈ Iν for ν = 2, . . . ,m, it follows that x ∈ Iν for ν = 2, . . . ,m. As yν ∈ I1 , it follows

that x ≡ 1 mod I1 .
Finally, we prove a useful Lemma which will be needed in Chapters 4 and 6.
Lemma 1.1.13 (Prime Avoidance). Let R be a Noetherian ring and p1 , . . . ,pn be ideals
in R. Suppose that p1 , . . . ,pn−1 are prime ideals. Let I ⊂ ∪ni=1 pi be an ideal. Then there
exists k such that I ⊂ pk .
To put it differently, let I be an ideal such that I 6⊂ pi for all i. Then I 6⊂ ∪ni=1 pi . This
means that we can find f ∈ I with f ∈ / pi for all i, whence the name prime avoidance.
Proof. The proof is done by induction on n: the case n = 1 is obvious. So it suffices to
show that I ⊂ ∪j6=i pj for some i. This we will show by contradiction. So suppose that
I 6⊂ ∪j6=i pj for all i. Therefore, we can choose x1 , . . . ,xn ∈ I such that xi 6∈ ∪j6=i pj .
Because of the fact that xi ∈ I ⊂ ∪nj=1 pj we get xi ∈ pi .
Now we consider x1 + x2 . . . xn ∈ I. In particular, there exists a k such that
x1 + x2 . . . xn ∈ pk .
If k = 1 then x1 ∈ p1 and, therefore, x2 . . . xn ∈ p1 . This implies xν ∈ p1 for some
ν > 1 which is not possible by the choice of xν . If k > 1 then x2 . . . xn ∈ pk because
xk ∈ pk and, therefore, x1 ∈ pk which is again not possible. Altogether we obtain a
contradiction to the assumption that I 6⊂ ∪j6=i pj for all i. This shows the Lemma.
Exercises
1.1.14. Let I, J ⊂ R be ideals. The product I ·J ⊂ R is the ideal in R generated by the elements
x · y, where x runs through I and y runs through J. Show that I · J ⊂ I ∩ J.
1.1.15.
(1) Let I, J ⊂ R be ideals. Show that I : J := {x ∈ R : xJ ⊂ I} is an ideal in R, and that
I ⊂ I : J.
(2) Let a ∈ R. Prove that (I ∩ J) : a = (I : a) ∩ (J : a).
(3) Let I : f ∞ := {x ∈ R : xf s ∈ I for some s}. If R is Noetherian, show that I : f ∞ = I : f k
for some k ∈ N.
(4) Let I ⊂ R be an ideal, and x,y ∈ R. Prove that (I : x) : y = I : xy.
√
1.1.16. Let I ⊂ R be an ideal. The radical I of I is defined by:
√
I = {f ∈ R : f s ∈ I for some s ∈ N}.
√
An ideal is called radical if and only if I = I.
√
(1) Prove that I is an ideal.
√ p√ √
(2) Prove that I + J = I + J.
6 1 Algebra
√ √ √
(3) Find ideals I,J for which I + J % I + J .
√ √ √
(4) Prove that I ∩ J = I ∩ J. Generalize to finitely many ideals.
1.1.17. Let R be a ring. An element x ∈ R is called nilpotent if there exists an n ∈ N with
xn = 0. Show that the set of nilpotent elements is an ideal N .
The ring Rred := R/N is called the reduction of R. A ring is called reduced if it has no
nonzero nilpotent elements. Show that Rred is reduced.
P
1.1.18. Let R be a G–graded ring. An ideal I ⊂ R is called homogeneous if I = g∈G (I ∩ Rg ).
Prove the following.
(1) I is homogeneous if and only if it can be generated by homogeneous elements.
(2) Sums, products and intersections of homogeneous ideals are homogeneous.
1.1.19. An abelian group G is called ordered, if there is a total ordering < on G satisfying:
g < h =⇒ a + g < a + h for all a ∈ G.
For example, the group Z with its natural ordering is an ordered group. Let G be an ordered
group and R be a G–graded ring.
(1) Prove that a homogeneous ideal I is prime if and only if for all homogeneous elements
a,b ∈ R with ab ∈ I it follows that a ∈ I or b ∈ I.
(2) Prove that the radical of a homogeneous ideal is homogeneous.
1.1.20. Let R be a Noetherian ring and I ⊂ R be an ideal. Prove that R/I is Noetherian.
1.1.21. Consider two ideals I ⊃ J in a Noetherian ring R. Suppose that for all f ∈ I there
exists a k (depending on f) with f k ∈ J. Prove that some power of I is contained in J. In
√ k
particular I ⊂ I for some k.
1.2 Modules
The notion of a module over a ring is the analog of the notion of vector space over a
field, and is obtained by simply copying the definition. Here it is.
Definition 1.2.1. Let R be a ring. An R–module M is a set M with two operations:
(1) There is an addition + : M −→ M :
x,y ∈ M 7→ x + y ∈ M
(2) We have multiplication with scalars ·R × M −→ M :

x ∈ M, a ∈ R 7→ a · x = ax ∈ M.
These should satisfy the following conditions.

• The set M is an abelian group with respect to the addition +.
• a(x + y) = ax + ay
(ab)x = a(bx)
(a + b)x = ax + bx
1·x =x
This holds for all x,y ∈ M and a,b ∈ R.
1.2 Modules 7
As in the theory of rings, groups, and vector spaces, we want to consider modules
up to isomorphism. The definition should not surprise the reader:
Definition 1.2.2. Let M and N be R–modules, ϕ : M → N a map.

• ϕ is called a homomorphism of R–modules (or simply homomorphism) if
– ϕ(x + y) = ϕ(x) + ϕ(y)
– ϕ(ax) = aϕ(x)
for all x,y ∈ M and all a ∈ R.
• ϕ is called an isomorphism if and only if ϕ is a bijective homomorphism.

• Two modules M and N are called isomorphic, notation M ∼
= N , if and only if there
exists an isomorphism ϕ : M → N .
From the definition it is not immediate that isomorphy is an equivalence relation. Of
course, the fact that M is isomorphic to itself is trivial, as is the fact that the isomorphism
property is transitive. To prove that it is symmetric too, one has to show that the inverse
ϕ−1 : N → M of a bijective homomorphism ϕ : M → N is a homomorphism again. This
will be left to the reader as an exercise.
This property distinguishes algebra from for example topology. A bijective continu-
ous map between topological space need not be a homeomorphism. To give a very trivial
example, the identity map from a set with more than one element to itself, the first
equipped with the discrete topology, and the second equipped with the trivial topology,
is a bijective continuous map which is not a homeomorphism.
We now give some examples of modules.
Examples 1.2.3.
(1) A K–module, for a field K, is just a K–vector space.

(2) A Z–module is an abelian group.
(3) An ideal I of R is an R–module. In particular, R itself is an R–module.
(4) Consider the polynomial ring K[X], and let M be a K[X]–module. Then in partic-
ular M is a K–module, that is, a K–vector space. We also get a map (multiplication
with X):
X : M −→ M
which one checks to be K–linear. In fact, the data M is a K[X]–module, and M
is a K–vector space together with a linear map, are equivalent. Classifying K[X]–
modules up to isomorphism is the same as classifying endomorphisms of K–vector
spaces up to conjugacy, see Exercise 1.2.30.
We collect some more definitions.

Definition 1.2.4. Let R be a ring, M and N be R–modules.
8 1 Algebra
(1) Let ∅ 6= A ⊂ M . Then A is called a submodule of M if for all x,y ∈ A and a ∈ R we

have x + y and ax ∈ A. Then A is in a natural way an R–module, where addition
and scalar multiplication are induced by those in M . One shows without difficulty
that the image ϕ(M ) of a homomorphism ϕ : M −→ N is in a natural way a
submodule of N .
Let N ⊂ M be a submodule. Then the quotient module M/N is the set of equiv-
alence classes x := x + N where x runs through M . It is in a natural way an
R–module via
• (x + N ) + (y + N ) := (x + y) + N ,
• a(x + N ) := ax + N .
Here x,y ∈ M and a ∈ R. One checks that addition and scalar multiplication are
well-defined.
(2) Let ϕ : M −→ N be a homomorphism of R–modules. We define
• the kernel of ϕ, notation Ker(ϕ), as
Ker(ϕ) := {x ∈ M : ϕ(x) = 0}.
One checks, exactly as in linear algebra, that Ker(ϕ) = {0} if and only if ϕ is
injective. Moreover, Ker(ϕ) is a submodule of M ,
• the cokernel of ϕ, notation Coker(ϕ), to be the quotient module N/ϕ(M ).
(3) The direct sum M ⊕ N of M and N is defined to be the direct sum of abelian
groups M ⊕ N = {(m,n),m ∈ M,n ∈ N }, with obvious addition, and with scalar
multiplication defined by a(m,n) := (am,an). It is sometimes also called the direct
product M × N . Similarly, one defines the direct sum ⊕i∈I Mi of R–modules. Here
I is any index set. As a set:
M
Mi = {(mi )i∈I with only finitely many mi 6= 0}
i∈I
and we have the obvious addition and scalar multiplication. One also has the direct
product Y
Mi = {(mi )i∈I }.
i∈I
with obvious addition and scalar multiplication. Here it is allowed that infinitely
many mi are nonzero.
(4) An R–module is called a free module if it is isomorphic to a direct sum ⊕i∈I R of
copies of R. Often we will have the case that I is a finite set of cardinality t. This
free module is called the free module of rank t, denoted by Rt . The fact that the
rank is well-defined is not trivial and will be shown in the exercises.
(5) Let M be an R–module, m1 , . . . ,mt be elements of M . The submodule (m1 , . . . ,mt )
generated by m1 , . . . ,mt is by definition the submodule of M given by
{a1 m1 + . . . + at mt : a1 , . . . ,at ∈ R}.
Similarly one defines the submodule generated by an arbitrary number of elements
of M .
1.2 Modules 9
(6) An R–module M is called finitely generated if there exist finitely many elements
m1 , . . . ,mt of M such that M = (m1 , . . . ,mt ). If this is the case, the elements
m1 , . . . ,mt are called generators of M . Finite generation of a module is equivalent
to having a surjective homomorphism
ϕ : Rt −→ M,
(a1 , . . . ,at ) 7→ a1 m1 + . . . + at mt .
(7) A set {m1 , . . . ,mt } is called a basis of M if every m ∈ M can be written as an R–

linear combination of the elements m1 , . . . ,mt in a unique way, that is, for all m ∈ M
there exist uniquely defined a1 , . . . ,at ∈ R such that m = a1 m1 + . . . + at mt . One
can show (see the exercises) that a module has a basis if and only if it is isomorphic
to a finitely generated free module. Similarly an infinite set is called a basis of M
if every element of M can be written as an R–linear combination of finitely many
elements of the set in a unique way.
(8) The torsion module of M is defined by:
Tors(M ) := {m ∈ M : there exists a nonzerodivisor a ∈ R with a · m = 0}.
So in case R is an integral domain, we have to find a nonzero element a ∈ R with

am = 0. A module M is called torsion free if Tors(M ) = 0. M is called a torsion
module if Tors(M ) = M .
(9) The set HomR (M,N ) of all homomorphisms from M to N is an R–module via the
operations
(ϕ + ψ)(m) := ϕ(m) + ψ(m).

(aϕ)(m) := aϕ(m),
for all m ∈ M and a ∈ R.

(10) Let m ∈ M . The annihilator of m is defined by Ann(m) := {r ∈ R : r · m = 0}. The
annihilator of M by definition is Ann(M ) := {r ∈ R : r · m = 0 for all m ∈ M }.
The annihilator is obviously an ideal.
Despite the obvious similarity with vector spaces, the general theory of modules over
a ring is much more difficult and much richer. The inability of to divide is just too much
an obstacle. For example, we know from the theory of vector spaces, that any finitely
generated vector space over a field K is isomorphic to K t for some t. This is certainly
not true for rings, as the following trivial example shows:
Example 1.2.5.
(1) The Z–module (that is, abelian group) Z/(2) is not free. Indeed a free group either
has 1 or an infinite number of elements.
(2) More generally, any module M whose torsion submodule Tors(M ) is nontrivial, is
not a free module. Indeed, suppose the converse, that is, let {mi }i∈I be a basis of
M . Let m ∈ Tors(M ), m 6= 0 and assume
10 1 Algebra
m = a1 m 1 + . . . + at m t .
As m 6= 0 we have ai 6= 0 for some i. Because m ∈ Tors(M ) there exists a nonze-
rodivisor a ∈ R with am = 0. We obtain
a · m = aa1 m1 + . . . + aat mt = 0.
By definition of basis aai = 0, which is a contradiction to the fact that a is a
nonzerodivisor.
One could ask the question whether at least finitely generated torsion free modules
are free. But this is also much too much to ask for, as the following simple lemma shows:
Lemma 1.2.6. Let I ⊂ R be an ideal, with R an integral domain. Then I is a free
module if and only if I is principal.
Proof. The case where I = (0) is trivial. Otherwise, suppose that I is principal, that is,
I = (f ) for some f ∈ R. Then the map: R −→ I given by a 7→ af gives an isomorphism of
R–modules. It is injective because R is supposed to be an integral domain. To prove the
converse, let I be a free R-module. If the rank is one, let ϕ : R −→ I be an isomorphism.
It is then immediate that I is the ideal generated by ϕ(1). We are finished if we show
that an ideal cannot be isomorphic to a free module of rank at least two. For this, it
suffices to show that there does not exist an injective map α : Rn −→ I, as soon as
n ≥ 2. Suppose the converse. Without loss of generality we may assume n = 2. Let
f = α(1,0), and g = α(0,1). Since α is injective, both f and g are nonzero. But then
α(g, − f ) = gf − f g = 0, a contradiction.
So for example the ideal (x,y) in the polynomial ring K[x,y] is not a free K[x,y]–
module. From this lemma one might hope that at least for principal ideal domains (that is,
domains where every ideal can be generated by one element), a finitely generated torsion
free module is free. This is indeed the case, in fact one has the following classification
theorem:
Theorem 1.2.7. Let R be a principal ideal domain, and M be a finitely generated R–
module. Then
M∼ = Rs ⊕ Tors(M ).
The torsion module has the following form:
Tors(M ) ∼
= R/(a1 ) ⊕ · · · ⊕ R/(at ) with (a1 ) ⊃ (a2 ) ⊃ . . . ⊃ (at ) 6= (0).
For a proof, we refer to Lang, [Lang 1965], Chapter XV, §2. At one point in the
book we will need the statement that a finitely generated torsion free module over the
power series ring in one variable is free. This is a special case of the theorem, for which
an easier proof is available. We refer to 1.3.9.
Of course, the finitely generated free modules Rp with p ∈ N are the simplest R–
modules. Just as in linear algebra, maps between such modules can be given by matrices.
Consider Rp with basis {e1 , . . . ,ep }, and Rq with basis {f1 , . . . ,fq }. Let ϕ : Rp −→ Rq
be an R–module homomorphism. The matrix of ϕ (with Pq respect to the chosen bases) is
the matrix (aij ) with entries in R given by ϕ(ei ) = j=1 aji fj . All the results in linear
algebra which do not use division in their proofs carry over to the general case of rings. In
particular, the determinant of a map between free modules of the same rank is defined.
On the other hand, because in a general ring we cannot divide, Cramer’s rule does not
carry over. But at least we can save the following statement.
1.2 Modules 11
Theorem 1.2.8 (Cramer’s Rule). Let a ring R, an n × n matrix A = (aij ) with co-
efficients in R and vectors (x1 , . . . ,xn ) and (b1 , . . . ,bn ) with entries in R be given. For
all j let Aj be the matrix obtained from A by replacing its j–th column2 by (b1 , . . . ,bn )t .
Consider the system of equations
n
X
aij xj = bi i = 1, . . . ,n.
j=1
Then
det(A)xj = det(Aj ) for j = 1, . . . ,n..
For a proof, look in any textbook on linear algebra, and forget to do the division at
the end of the proof, or do the proof yourself, Exercise 1.2.28.
Definition 1.2.9. Let A be an n × n matrix with entries in R. The adjoint matrix
Aad = (aad
ij )
is defined by
aad
ij = (−1)
i+j
det(Aij )
where Aij is the matrix obtained from A by deleting the i–th column and j–th row.
For a matrix A with adjoint matrix Aad we have the following theorem, which follows
from Cramer’s rule:
Theorem 1.2.10.
Aad A = AAad = det(A) · Idn ,
where Idn denotes the n × n unit matrix.
So, up to a factor (the determinant), the adjoint matrix describes “the inverse ma-
trix”.
Definition 1.2.11. A (finite or infinite) sequence of R–modules and homomorphisms
αk−1 k α
· · · −→ Mk−1 −−−→ Mk −−→ Mk+1 −→ · · ·
is called a complex if Ker(αk ) ⊃ Im(αk−1 ). It is called exact at Mk if Ker(αk ) =

Im(αk−1 ). It is called exact if it is exact at all Mk . In particular we have:
α
• 0 −→ M −→ N is exact ⇐⇒ α is injective,
α
• M −→ N −→ 0 is exact ⇐⇒ α is surjective,
α
• 0 −→ M −→ N −→ 0 is exact ⇐⇒ α is an isomorphism.
An exact sequence of type
α β
0 −→ M1 −→ M2 −→ M3 −→ 0
is called a short exact sequence. This is equivalent to saying that α is injective, β is

surjective, and β induces an isomorphism M2 / Im(α) ∼= M3 .
2 Given a matrix A we denote by At the transposed matrix.
12 1 Algebra
Lemma 1.2.12 (Snake Lemma). Consider a commutative diagram3

f1 f2
0 −→ M1 −→ M2 −→ M3 −→ 0
h1 ↓ h2 ↓ h3 ↓
g1 g2
0 −→ N1 −→ N2 −→ N3 −→ 0
of R–modules, with exact rows. Then the sequence
1 f′ 2 f′ 1 δ 2 g g
0 → Ker(h1 ) −→ Ker(h2 ) −→ Ker(h3 ) −→ Coker(h1 ) −→ Coker(h2 ) −→ Coker(h3 ) → 0
is exact.
In this lemma f1′ and f2′ are restrictions of f1 and f2 , and g 1 and g2 are induced
by g1 and g2 . The map δ is defined as follows. Let x ∈ Ker(h3 ), that is, x ∈ M3 with
h3 (x) = 0. As the upper row is exact, it follows that f2 is surjective, hence we can
take y ∈ M2 with f2 (y) = x. Because of the commutativity of the diagram we have
g2 h2 (y) = h3 f2 (y) = h3 (x) = 0. Therefore, h2 (y) ∈ Ker(g2 ). Since the lower row is exact,
we have Ker(g2 ) = Im(g1 ). Hence we can find w ∈ N1 with g1 (w) = h2 (y). Now we put
δ(x) to be the class of w in Coker(h1 ). We have to prove that δ is well-defined and that
this sequence is exact. The proof of the Snake Lemma is left as Exercise 1.2.27.
Definition 1.2.13. Let R be a ring, and M be an R–module. Then M is called Noethe-
rian if every submodule N ⊂ M is finitely generated.
This just means that we can find finitely many elements m1 , . . . ,mt ∈ N such that
N = R · m1 + . . . + R · mt = (m1 , . . . ,mt ).
Note that a ring R is Noetherian if and only if R, considered as an R–module, is Noethe-

rian. The following lemma is left as Exercise 1.2.29.
Lemma 1.2.14.
• Submodules and quotient modules of Noetherian modules are Noetherian.

• Suppose that N ⊂ M are R–modules. Then
M is Noetherian ⇐⇒ N and M/N are Noetherian.
We could rephrase this by saying that, in a short exact sequence of modules, if any
two are Noetherian, then the third is Noetherian.
Noetherian modules over a Noetherian ring are easy to construct:
Lemma 1.2.15. Let R be a Noetherian ring, and M be a finitely generated R–module.
Then M is a Noetherian R–module.
f
3
M1 −
→ M2
A diagram h1 ↓ is called commutative if h2 f = gh1 , that is it does not matter which route one
h2 ↓
g
N1 −
→ N2
takes along the diagram.
1.2 Modules 13
Proof. As M is a quotient of a free module, we only have to prove the lemma for M = Rt ,
by the previous lemma. This will be done by induction on t. The case t = 1 is the
assumption. Apply the previous lemma to the exact sequence
0 1
1 0 0 1 0 ... 0 1
B0C
B.C @ .. .. .. .. A
@.A
. .. . .
0 0 0 0 ... 1
0 → R −−−−→ Rt −−−−−−−−−→ Rt−1 → 0.
Definition 1.2.16. (1) The tensor product M ⊗R N is an R–module with the following
universal property:
There exists a canonical bilinear map λ : M × N −→ M ⊗R N . If ϕ : M × N −→ P
is any bilinear map, then there exists a unique homomorphism h : M ⊗R N −→ P such
that h ◦ λ = ϕ.
λ /
M × NM M ⊗R N
MMM
MMM h
ϕ MMM
M&
P.
One can give a direct construction of the tensor product as follows. M ⊗R N is the
quotient of the free R–module generated by {m ⊗ n : m ∈ M, n ∈ N } divided out by the
submodule of the tensor product relations generated by
{(am + bn) ⊗ (cp + dq) − ac · m ⊗ p − ad · m ⊗ q − bc · n ⊗ p − bd · n ⊗ q | a,b,c,d ∈ R,
m,n ∈ M, p,q ∈ N }.
The bilinear form λ : M × N −→ M ⊗R N is defined by λ(m,n) = m ⊗ n and has the
property mentioned above (Exercise 1.2.45).
(2) Let S be an R–algebra, then M ⊗R S is an S–module with the multiplication
s · m ⊗ s′ := m ⊗ (ss′ ) (Exercise 1.2.44). For example, if V is a real vector space, then
V ⊗R C is a C –vector space, called the complexification of V .
(3) Let S,T be R–algebras, then S ⊗R T is an R–algebra with the multiplication
(s ⊗ t) · (s′ ⊗ t′ ) := (ss′ ) ⊗ (tt′ ). The simplest example here is the case that R = k is a
field, S = k[X] and T = k[Y ]. Then S ⊗R T = k[X,Y ].
Let i1 : S −→ S ⊗R T , i2 : T −→ S ⊗R T be defined by i1 (s) = s ⊗ 1, i2 (t) = 1 ⊗ t,
then
i1
S ⊗O R T o SO
i2
T o R
is commutative, and universal (cf. Exercise 1.2.45), that is: let A be an R–algebra and
j1 : S −→ A, j2 : T −→ A be homomorphisms such that
j1
AO o SO
j2
T o R
14 1 Algebra
is commutative, then there exists a unique homomorphism h : S ⊗R T −→ A such that

h ◦ iν = jν , ν = 1,2.
Definition 1.2.17. Let R be a G–graded ring. An R–module M is called G–graded if
there is a decomposition
M = ⊕g∈G Mg
such that
Rg Mh ⊂ Mg+h
for all g,h ∈ G. In particular a G–graded ring is a G–graded module over itself. A Z–
graded module is called a graded module. A homomorphism ϕ : M −→ N is called
homogeneous if ϕ(Mg ) ⊂ Ng for all g ∈ G. More generally, one says that ϕ : M −→ N is
homogeneous of degree h if ϕ(Mg ) ⊂ Ng+h for all g ∈ G. For a G–graded module M and
g ∈ G we define the G–graded module M (g). It has M as the underlying module, but has
different grading: M (g)h := Mg+h . So a homogeneous homomorphism ϕ : M −→ N of
degree g induces a homogeneous homomorphism ϕ : M −→ N (g) (or ϕ : M (−g) −→ N ).
Definition 1.2.18. Let R be a ring and ϕ : Rn −→ Rm be an R–module homomorphism.
Assume that ϕ is defined by the matrix A for some choice of bases in Rn and Rm . Then
for all 1 ≤ t ≤ min{n,m} the ideal generated by the t–minors of A does not depend on
A and is called the t–th Fitting ideal It (ϕ) of ϕ.
For convenience we define It (ϕ) = R if t ≤ 0.
Remark 1.2.19.
(1) The proof that It (ϕ) is independent on the choice of the bases is left as Exercise
1.2.47.
ϕ ψ
(2) Let Rn −→ Rm −→ M −→ 0 and Rs − → Rt −→ M −→ 0 be exact. Then Im−ν (ϕ) =
It−ν (ψ) for all ν. The proof uses some facts on so-called local rings, to be treated
in the next section. The proof is left as Exercise 1.3.28.
(3) If S is an R–algebra, then Iν (ϕ ⊗ idS ) = Iν (ϕ)S. This is left as Exercise 1.2.47.
Exercises
1.2.20. Prove that the ring of integers Z and the polynomial ring in one variable K[x] over a
field K are principal ideal domains.
(Hint: Use division with remainder.)
1.2.21.
(1) Let ϕ : Rp → Rs be an isomorphism between free R–modules. Show that p = s.
(Hint: Suppose not, for example p > s. Let ψ : Rs → Rp be the inverse of ϕ. Extend ϕ
resp. ψ to maps
ϕ′ : Rp → Rs ⊕ Rp−s
resp.
ψ ′ : Rs ⊕ Rp−s → Rp
in the obvious way. Derive a contradiction by looking at determinants.)
(2) Conclude that the rank of a free module is well-defined.
1.2.22. Let M be an R–module, and I = Ann(M ) = {x ∈ R : x · M = 0}. Then M is in a

natural way an R/I–module. Prove this.
1.2 Modules 15
1.2.23. Let R be a subring of S. Show that one can, in a natural way, view S as an R–module.
1.2.24. Prove that Tors(M ) is a submodule of M .
1.2.25. Consider an exact sequence

αk−1 k α
· · · −→ Mk−1 −→ Mk −→ Mk+1 −→ · · · .
Put Nk = Im(αk ) = Ker(αk+1 ). Show that for all k one has short exact sequences:
0 −→ Nk−1 −→ Mk −→ Nk −→ 0.
(So any exact sequence can be split into short exact sequences.)
1.2.26. Let K be a field, and consider an exact sequence
0 −→ M1 −→ M2 −→ · · · −→ Ms −→ 0
P
of finite-dimensional K–vector spaces. Prove that sk=1 (−1)k dimK (Mk ) = 0.
1.2.27. Prove the Snake Lemma 1.2.12.
1.2.28. Prove Cramer’s rule 1.2.8.
1.2.29. Prove Lemma 1.2.14.
1.2.30.
(1) Let M be a K[X]–module, which is finite-dimensional as a K–vector space. Show that M
is a torsion module.
(2) Prove the theorem on Jordan normal form of matrices over an algebraically closed field,
using the classification of finitely generated modules over a principal ideal domain, see
Theorem 1.2.7.
1.2.31. Let K be a field, S be a K–algebra, and M be an S–module. Suppose that Ann(M ) = (0)
and that dimK (M ) < ∞. Prove that dimK (S) < ∞.
(Hint: Show that one can embed S into HomS (M,M ).)
1.2.32. Let N be an R–module. Prove that the map φ : HomR (R,N ) −→ N , φ(ϕ) = ϕ(1), is
an isomorphism.
1.2.33. Let M,M ′ ,N,N ′ be R–modules and α : M ′ −→ M , β : N −→ N ′ be homomorphisms.

Prove that ϕ 7→ β ◦ ϕ ◦ α defines a homomorphism HomR (M,N ) −→ HomR (M ′ ,N ′ ).
n
1.2.34. Let M,N be R–modules and M = ⊕ Mi . Prove that
i=1
n
Hom(M,N ) ∼
= ⊕ Hom(Mi ,N )
i=1
n
Hom(N,M ) ∼
= ⊕ Hom(N,Mi ).
i=1
1.2.35. Consider an exact sequence
· · · −→ Mk+1 −→ Mk −→ Mk−1 −→ · · ·
of R–modules. Let x ∈ R be a nonzero divisor of Mk for all k. Prove that the induced sequence
· · · Mk+1 /xMk+1 −→ Mk /xMk −→ Mk−1 /xMk−1 −→ · · ·
is also exact.
(Hint: Split the exact sequence in many short exact sequences and use the Snake Lemma.)
16 1 Algebra
1.2.36. Let 0 −→ M −→ N −→ L be a complex of R–modules. Prove that the complex is exact

if and only if for all R–modules A
0 −→ HomR (A,M ) −→ HomR (A,N ) −→ HomR (A,L)
is exact.
1.2.37. Let M −→ N −→ L −→ 0 be a complex of R–modules. Prove that the complex is exact

if and only if for all R–modules A
0 −→ HomR (L,A) −→ HomR (N,A) −→ HomR (M,A)
is exact.
1.2.38. Let M,M ′ ,N,N ′ be R–modules and α : M ′ −→ M , β : N ′ −→ N be homomorphisms.

Prove that m′ ⊗ n′ 7→ α(m′ ) ⊗ β(n′ ) defines a homomorphism M ′ ⊗R N ′ −→ M ⊗R N .
1.2.39. Prove that M ⊗R N ∼

= N ⊗R M .
1.2.40. Prove that the canonical map φ : HomR (M ⊗R N,P ) −→ HomR (M, HomR (N,P )) de-
fined by φ(λ)(m)(n) := λ(m ⊗ n) is an isomorphism.
n n
1.2.41. Let M,N be R–modules and M = ⊕ Mi . Prove that M ⊗R N ∼
= ⊕ (Mi ⊗R N ).
i=1 i=1
1.2.42. Prove that M ⊗R R ∼

= M.
1.2.43. Let M −→ N −→ L −→ 0 be an exact sequence of R–modules and A be an R–module.

Prove that
M ⊗R A −→ N ⊗R A −→ L ⊗R A −→ 0
is exact.
(Hint: Use Exercises 1.2.37 and 1.2.40.)
1.2.44. Let M be an R–module and S be an R–algebra. Prove that with the multiplication
s · m ⊗ s′ = m ⊗ ss′ , M ⊗R S is an S–module.
1.2.45. Prove the universal properties mentioned in Definition 1.2.16.

P
1.2.46. Let V and W be vector spaces over K. For ϕ ∈ V ∗ and
P vi ⊗ wi ∈ V ⊗K W , consider
ϕ(vi )wi ∈ W . Show that we get a well defined map V ∗ −→ W . Show that V ⊗K W ∼ =
Hom(V ∗ ,W ).
1.2.47. Prove (1) and (3) of Remark 1.2.19.

(Hint: It is enough to consider the case of two matrices A and B, B obtained from A by one
column operation, and to prove that It (A) = It (B).)
ϕ ψ
1.2.48. Let M,N be R–modules and 0 −→ M −→ N −→ Rs −→ 0 be an exact sequence. Prove
that N ∼
= M ⊕ Rs , more precisely
ϕ ψ
0 −→ M −→ N −→ Rs −→ 0
|| ↑ ||
0 −→ M −→ M ⊕ Rs −→ Rs −→ 0
i π
commutes, i,π the canonical injection resp. projection. In this case one says that M is a direct
summand of N .
1.3 Local Rings and Localization 17
1.3 Local Rings and Localization
Definition 1.3.1. A ring R is called local if it has a unique maximal ideal m. One often
says that (R,m) is a local ring, to indicate that m is its unique maximal ideal.
Later on in this section we will see how to construct local rings from general rings by
means of “localization”. In this book, the local rings we are dealing with most frequently
are power series rings and their quotients, see Section 3.1. Here we define those rings for
the case of one variable.
P∞ 1.3.2. The formal power series ring C [[x]] in one variable consists of ele-
Definition
ments k=0 ak xk , with ak ∈ C . We have the obvious addition and multiplication.
P∞ The
convergent power series ring C {x} in one variable consists of elements k=0 ak xk , with
ak ∈ C , that converge in some open neighborhood of 0 in C .
Note that in the formal power series ring as well as in the convergent power series
ring, the elements which do not lie in (x), that is, those with nonzero constant term,
are units. Therefore, the fact that the power series rings are local rings follows from the
following lemma.
Lemma 1.3.3. Let R be a ring, and m ⊂ R be an ideal. Then R is local, with maximal
ideal m, if and only if R \ m is the set of units of R.
Proof. Suppose R is local with maximal ideal m. Take an element a ∈ R. If (a) 6= R,
then (a) is contained in a maximal ideal, which by definition of a local ring must be m.
Therefore, if a ∈
/ m, then (a) = R, that is, a is a unit. Moreover, if a ∈ m, then a is not
a unit, as otherwise (a) = R.
Suppose, on the other hand, that R \ m is the set of units. Take an ideal I 6= R. Then
the ideal I does not contain units, and therefore must be contained in m. This shows
that m is the unique maximal ideal of R.
Theorem 1.3.4 (Nakayama’s Lemma). Let (R,m) be a local ring and M be a finitely
generated R–module with m · M = M . Then M = 0.
Proof. By contradiction. Let m1 , . . . ,mt be a minimal number of generators for M . As
M = m · M there exist a1 , . . . ,at ∈ m such that mt = a1 m1 + . . . + at mt . Therefore
(1 − at )mt = a1 m1 + . . . + at−1 mt−1 .
Now 1 − at ∈ / m, and as R is a local ring it follows that 1 − at is a unit. Therefore

mt ∈ (m1 , . . . ,mt−1 ), in contradiction to the minimality of t.
For more a general version of Nakayama’s Lemma, we refer to Exercise 1.3.17. A
corollary of this important lemma is the following corollary.
Corollary 1.3.5 (Krull’s Intersection Theorem). Let (R,m) be a Noetherian local ring,
and M be a finitely generated R–module. Then ∩k∈N mk M = (0).
Proof. We will prove this theorem for the case M = R, the general case is similar, see
Exercise 1.3.24. Put I = ∩k∈N mk . Then I is an ideal (check this), finitely generated
because R is Noetherian. We will show that m · I = I, so that I = (0) by Nakayama’s
Lemma. For this purpose consider the following set of ideals
18 1 Algebra
A := {J ideal in R : J ∩ I = mI}.
Our ring R is Noetherian, so that this set has a maximal element which we call J.
Obviously mI ⊂ J. Note that it suffices to prove that mk ⊂ J for certain k. Indeed, as
obviously I ⊂ mk , it follows I ⊂ mk ⊂ J, hence I ⊂ J ∩ I = mI, and we can apply
Nakayama. As m is finitely generated, it suffices to find for each f ∈ m an α with f α ∈ J.
This we will do by showing J + (f α ) = J by using the maximality property of J. Thus it
remains to show that (J +(f α ))∩I = mI. The inclusion ⊃ is obvious. As R is Noetherian,
the chain J : f ⊂ J : f 2 ⊂ · · · stabilizes. Hence there exists an α with J : f α = J : f α+1 ,
and this is the α we need. Now let x ∈ (J + (f α )) ∩ I, say x = y + af α ∈ I for y ∈ J. It
follows that f α+1 a = f x − f y ∈ mI + J = J, as mI ⊂ J. Hence a ∈ J : f α = J : f α+1 .
Thus f α a ∈ J, so that x ∈ J follows. As x ∈ I it follows x ∈ J ∩ I = mI. This holds
for all x ∈ (J + (f α )) ∩ I so that (J + (f α )) ∩ I ⊂ mI follows. This is what we had to
show.
Another corollary is the following lemma on the number of generators of a module
over a local ring. The proof is left as Exercise 1.3.18.
Corollary 1.3.6. Let M be a finitely generated module over a local ring (R,m).
(1) Let f1 , . . . ,fs ∈ M such that the classes of the fi generate M/mM as R/m–vector
space. Then f1 , . . . ,fs generate M .
(2) The minimal number of generators of M is equal to dimR/m (M/mM ).
Definition 1.3.7. Let R be a ring, and M be an R–module. A presentation of M is a
short exact sequence
α β
(1.3) G −→ F −→ M −→ 0
of R–modules, in which F and G are free R–modules. In case R is a local ring, and M
is finitely generated, a presentation as in 1.3 is called minimal if the rank of F is equal
to the minimal number of generators of M , that is, equal to dimR/m (M/mM ), see 1.3.6.
This is equivalent to the statement Im(α) ⊂ m · F .
ϕ
Proposition 1.3.8. Let R be a local ring and Rn −→ Rm −→ M −→ 0 be a presentation
of M . The following conditions are equivalent:
(1) M is a free module of rank m − r,
(2) Im(ϕ) is a free module of rank r and a direct summand of Rm ,
(3) Ir (ϕ) = R and Ir+1 (ϕ) = (0).
Proof.
(3) ⇒ (2) The fact that Ir (ϕ) = R and R is local implies that for a matrix A representing
ϕ there is an r–minor which is not in the maximal ideal of R, hence a unit. By renumbering
we may assume that this minor is obtained by taking the determinant of the first r rows
ad
and columns. We write A = ( B∗ ∗∗ ) with B an r × r–matrix. Then det1 B B0 Idn−r 0 gives
n
an automorphism of R and we multiply A from the right with this matrix. So we may
assume that A is of type ( Id∗r ∗∗ ). As in linear algebra, we can now do row and column
operations, corresponding to a change of basis in target and source. We may then assume
that Ais of type Id0r 0∗ . As by assumption Ir+1 (ϕ) = 0, it then follows that A is of type
Idr 0 . This immediately implies (2).
0 0
(2) ⇒ (1) is trivial, using Exercise 1.3.26.

(1) ⇒ (3) By Exercise 1.2.48 and Exercise 1.3.26 we obtain that Rm ∼ = M ⊕ Im(ϕ),
and Im(ϕ) is free of rank r. Similarly, Rn ∼
= Ker(ϕ) ⊕ Im(ϕ) and Ker(ϕ) is free of rank
n − r. So we may assume that ϕ is given by the matrix Id0r 00 , see Remark 1.2.19. This
implies (3).
The following theorem is a special case of the classification of finitely generated
modules over principal ideal domains.
Theorem 1.3.9. Let M be a finitely generated torsion free C {x}–module. Then M is
free.
As a finitely generated vector space is free, the theorem immediately follows from
the following lemma.
Lemma 1.3.10. Let (R,m) be a local ring, M a finitely generated R–module, and x ∈ m
a nonzero divisor of M . Then
M is a free R–module ⇐⇒ M/xM is a free R/(x)–module.
Proof. The implication =⇒ is trivial. Suppose, therefore, that M/xM is a free R/(x)-
module. Take a map α : Rn = F → M , sending each basis element to a generator of M .
We might take n to be the minimal number of generators of M . It follows that the map
F/xF → M/xM is also surjective. By 1.3.6 M/mM ∼
n
= (R/m) . As we supposed that
M/xM is a free R/(x)–module the map α induces an isomorphism (R/(x))n ∼ = M/xM .
Let K be the kernel of the map α. We obtain a commutative diagram:
0 / K / F / M / 0
·x ·x ·x

0 / K / F / M / 0
where ·x denotes multiplication by x. By assumption, x is not a zero divisor of M , so
that the kernel of ·x : M → M is zero. It follows from the Snake Lemma that we have
an exact sequence:
0 −→ K/xK −→ F/xF −→ M/xM −→ 0.
From the remark above, we have that F/xF → M/xM is an isomorphism. It follows that
K/xK = 0. From Nakayama, in the version of Exercise 1.3.17 it follows that K = 0 or,
equivalently, that M ∼
= F is free.
There is a procedure to construct local rings from a given ring, called localization.
This is a special case of forming rings (and modules) of fractions, which is itself a gener-
alization of defining the quotient field of an integral domain. In order to define rings of
fractions, we have to introduce the concept of a multiplicatively closed set.
Definition 1.3.11. Let R be a ring. A set S ⊂ R is called multiplicatively closed if and
only if it satisfies the following conditions.
• 1 ∈ S,
• If f,g ∈ S, then f · g ∈ S.
20 1 Algebra
Definition 1.3.12. Let R be a ring and S be a multiplicatively closed set. We will define
the ring of fractions RS , also denoted by S −1 R. As a set it is the following:
RS = { sr : r ∈ R, s ∈ S}
and satisfies the following properties:

•
r1 r2
= ∈ RS
s1 s2
if and only if there exists an s ∈ S such that s(r1 s2 − r2 s1 ) = 0.
• Multiplication:
r1 r2 r1 r2
· := .
s1 s2 s1 s2
• Addition:
r1 r2 r1 s2 + r2 s1
+ := .
s1 s2 s1 s2
Similarly, for any R–module M , and any multiplicatively closed set S ⊂ R, one can
define the RS –module MS , also denoted S −1 M . In particular this can be done for an
ideal I ⊂ R. Its definition is straightforward, and is left to the reader.
Examples 1.3.13.
(1) Let R be an integral domain, and S = R \ {0}. Then RS is the quotient field Q(R)
of R. It is indeed a field, as Q(R) = { ab : a,b ∈ R, b 6= 0}. So if ab 6= 0, then a 6= 0,
and ab ∈ Q(R).
(2) More generally, for a ring R, one can take as multiplicatively closed set S the set of
all nonzero divisors. The resulting ring RS is called the total quotient ring, or total
ring of fractions of R, notation Q(R). We will prove in Proposition 1.4.27 that for
a reduced Noetherian ring Q(R) is a direct sum of fields.
(3) Consider the multiplicatively closed set S = {1,f,f 2 ,f 3 , . . .} for f ∈ R. In this case
one usually writes Rf instead of RS .
(4) Let p be a prime ideal of R. According to the definition of a prime ideal the set
R \ p is a multiplicatively closed set. The resulting ring of fractions is denoted by
Rp (instead of the “correct” RR\p ), and is called the localization of R in p. So
Rp = { ab : a ∈ R, b ∈
/ p}.
At first it might be surprising that the following lemma holds generally for prime
ideals, and not only for maximal ideals. Nevertheless, the proof is easy!
Lemma 1.3.14. Let R be a ring and p ⊂ R be a prime ideal. Then Rp is a local ring,
with maximal ideal pRp = { ab : a ∈ p, b ∈
/ p}.
Proof. By 1.3.3, it suffices to show that the elements of pRp are the nonunits of Rp . An
element in Rp \ pRp is of type ab , where a,b ∈
/ p. But then ab ∈ Rp , so that ab is a unit. Of
a
course, every element b , with a ∈ p is not a unit in Rp .
Theorem 1.3.15. Let R be a ring, S a multiplicatively closed set. There is a one-one

correspondence:
prime ideals p ⊂ R with p ∩ S = ∅ ←→ prime ideals pS ⊂ RS .
The correspondence is given by the following rules. Given a prime ideal pS ⊂ RS , the
ideal p := pS ∩ R is a prime ideal in R. Given a prime ideal p in R whose intersection
with S is empty, its localization pS is a prime ideal in RS .
Proof. Suppose p is a prime ideal in R, with p ∩ S = ∅. Then the inclusion pS $ RS is
strict. This is because otherwise 1 = as , with a ∈ p, s ∈ S. Hence there would exist a
t ∈ S such that t(s − a) = 0, that is ta = ts. But ts ∈ S, and ta ∈ p, which by assumption
cannot be the case. Now suppose as bt ∈ pS . It follows that ab ∈ p, and therefore either
a ∈ p or b ∈ p. Therefore either as ∈ pS or bt ∈ pS . The converse is left to the reader.
Exercises
1.3.16. Prove that IRf ∩ R = I : f ∞ (see 1.1.15).
1.3.17 (Nakayama’s Lemma). Let R be a ring, I ⊂ R be an ideal which is contained in all the
maximal ideals of R, and M be a finitely generated R–module.
(1) Suppose that M = I · M . Prove that M = 0.
(2) Let N be a submodule of M such that M = N + I · M . Prove that M = N .
1.3.18. Prove Corollary 1.3.6.

(Hint: Let N = (f1 , . . . ,fs ) and apply Nakayama’s Lemma.)
1.3.19. Let (R,m) be a Noetherian local ring, and J an ideal in R.

(1) Suppose mJ = (0).
(a) Prove that J is a finite-dimensional R/m–vector space.
(b) Prove that there exists an n ∈ N with J ∩ mn = (0).
(2) Show that that there exists an n with J ∩ mn ⊂ mJ.
(3) Prove that for such an n we have J/mJ ∼ = J + mn /mJ + mn .
1.3.20. Let K be a field, and R be a local K–algebra, with maximal ideal m.
(1) Suppose that dimK R = n < ∞. Show that mn = 0.
(Hint: Show that the sequence of ideals m ⊃ m2 ⊃ · · · stabilizes and use Nakayama’s
Lemma.)
(2) Let I ⊂ R be an ideal such that dimK (R/I) = n < ∞. Show that mn ⊂ I.
1.3.21. Let M be an R–module. Prove that the following conditions are equivalent.
(1) M = 0.
(2) Mp = 0 for all prime ideals p ⊂ R.
(3) Mm = 0 for all maximal ideals m ⊂ R.
(Hint: For (3) =⇒ (1) show that for all m ∈ M the ideal Ann(m) is not contained in any
maximal ideal of R.)
1.3.22. Let R be a ring.

(1) Let S ⊂ R a multiplicatively closed subset of R and M −→ N −→ P be an exact sequence
of R–modules. Prove that the induced sequence MS −→ NS −→ PS is exact. One phrases
this by saying that localization is an exact functor.
22 1 Algebra
(2) Let · · · −→ Mi−1 −→ Mi −→ Mi+1 −→ · · · be a complex. Prove that the complex is

exact if and only if for all prime ideals p ⊂ R the induced complex · · · −→ Mi−1p −→
Mip −→ Mi+1p −→ · · · is exact.
(3) Show that RS ⊗R M = ∼ MS .
(4) Show that (M ⊗R N )S ∼
= MS ⊗RS NS .
1.3.23. Let R be a ring, and p ⊂ R be a prime ideal. We define the n–th symbolic power p(n)
of p by
p(n) := pn Rp ∩ R.
We obviously have pn ⊂ p(n) .
(1) Let R = k[x,y,z]/(z 2 − xy), and p = (x,z). Show that p is prime, and that p(2) = (z 2 ,x).
In particular p2 6= p(2) .
/ p and f · g ∈ p(n) . Show that g ∈ p(n) .
(2) Let f,g ∈ R be given such that f ∈
1.3.24. Prove Krull’s Intersection Theorem 1.3.5 for modules.
1.3.25. Let R be a ring. The Jacobson radical J(R) of R by definition is the intersection of all
maximal ideals of R. Show that x ∈ J(R) implies that 1 + x is a unit.
1.3.26. Let R be a local ring and Rn = Rs ⊕ N . Prove that N is free of rank n − s.

(Hint: Use 1.3.6.) This statement does not hold for general (nonlocal) rings.
1.3.27. Let R be a ring and p1 , . . . ,pm be prime ideals. Let (0) 6= M be a finitely generated
R–module such that Mp j 6= (0) for all j. Prove that there exists an x ∈ M such that x 6∈ pj Mp j
for all j.
(Hint: Use the idea of the proof of prime avoidance, see 1.1.13: Use induction on m and assume,
therefore, y 6∈ pj Mp j for j = 1, . . . ,m − 1 and a suitable y. Assume the worst case, that is,
y ∈ pm Mp m . You may also assume that ∩m−1 m−1
j=1 pj 6⊂ pm and choose r ∈ ∩j=1 pj \ pm . Let
′ ′
M = (y,y2 , . . . ,yn ) and M = (y,ry2 , . . . ,ryn ) then Mp m = Mp m and y + ryi 6∈ pj Mp j for i ≥ 2
and j = 1, . . . ,m − 1. One of these elements is the element we are looking for because, otherwise,
y,y2 , . . . ,yn ∈ pm Mp m , that is, Mp m = (0).)
1.3.28. Prove (2) of Remark 1.2.19.

(Hint: Localize by prime ideals to reduce the assertion to the case of local rings. Let m be the
ϕ1 ϕ2
maximal ideal of R and Rn1 −→ Rm1 −→ M −→ 0, Rn2 −→ Rm2 −→ M −→ 0 be two
presentations of M . Show that we may assume that Ker(ϕ1 ) ⊂ mRn1 and Im(ϕ1 ) ⊂ mRm1 .
Suppose first that Ker(ϕi ) ⊂ mRni and Im(ϕi ) ⊂ mRmi . Prove that there exist isomorphisms
λ,µ such that
ϕ1
Rn1 / Rm1 /M
λ µ
ϕ2
Rn2 / Rm2 /M
commutes. Use Remark 1.2.19 (1) for this` case. Now
´ show that for a suitable choice of basis in
Rn2 and Rm2 the map ϕ2 has a matrix ϕ02 Id 0 0
0
with ϕ2 : Rn1 −→ Rm1 and Id the identity
matrix.)
1.4 Primary Decomposition
It is well-known that in Z every number is a product of prime numbers. More precisely,

let a ∈ Z and a 6= ±1,0, then
1.4 Primary Decomposition 23
Y
a = sign(a) pν11 · · · pνrr
with pairwise different prime numbers pi . The pi and the νi are uniquely determined (up
to permutation). It is easily shown that equivalently
(a) = (p1 )ν1 ∩ · · · ∩ (pr )νr
holds. This property can be generalized in the following manner:

Definition 1.4.1. Let R be an integral domain.
(1) An element 0 6= p ∈ R with (p) 6= R is called prime if p | gh implies that p | g or

p | h. An element f in R is called irreducible if f = gh implies that g or h is a unit.
(2) R is called a unique factorization domain, or a factorial ring, if every element

different from zero and not being a unit is a product of finitely many primes.
(3) Let R be a unique factorization domain and a1 , . . . ,an ∈ R elements different from
zero. An element d ∈ R is called a common divisor of a1 , . . . ,an if d | ai for i =
1, . . . ,n. An element d ∈ R is called a greatest common divisor , d = gcd(a1 , . . . ,an ),
if d is a common divisor of a1 , . . . ,an and every common divisor of a1 , . . . ,an divides
d.
Remarks 1.4.2.
(1) It follows directly from the definition that a prime element √ is irreducible. The
converse is not true in general. For example, in the
√ ring Z[√ −5] the element 2 is
irreducible, but not prime as 2 | 6, and
√ 6 = (1 + √ −5)(1 − −5). However, it can
be shown that 2 does not divide 1 + −5 and 1 − −5. To give a different example,
consider the ring C [x,y,z]/(z 2 − xy). Then we have z | xy, but z does not divide x
or y.
The converse does hold however in a unique factorization domain.
(2) One could, alternatively, define a unique factorization domain as a domain in which
each nonzero element which is not a unit has (up to units and renumbering) a unique
decomposition into irreducibles. This will be shown in the exercises.
(3) The existence of a greatest common divisor will be proved in Exercise 1.4.37.
Example 1.4.3. Any principal ideal domain, especially Z and the polynomial ring K[x]
over a field K in one variable, is a unique factorization domain. This will be proved in
Exercise 1.4.35. Moreover, it follows from the Lemma of Gauß, to be proved next, that
the polynomial ring K[x1 , . . . ,xn ] over any field K is a unique factorization domain.
Lemma 1.4.4 (Lemma of Gauß). Let R be a unique factorization domain. Then R[x]
is a unique factorization domain.
One of the main steps in the proof of this Theorem is the following lemma.
Lemma 1.4.5. Let R be a unique factorization domain, K = Q(R) its quotient field,
and f ∈ R[x] irreducible. Then f is also irreducible considered as element of K[x].
24 1 Algebra
Proof. Suppose the converse. Let f = g · h, g,h ∈ K[x] and deg(h) < deg(f ), deg(g) <
deg(f ). We choose c ∈ R such that cg =: g, ch =: h are in R[x] and obtain c2 f = gh.
Now c2 is not a unit in R. This is because f is by assumption irreducible in R[x]. So
we can write c2 = p1 · · · ps , pi ∈ R prime. Exercise 1.4.39 implies p1 |g or p1 |h. So we
can find g ′ and h′ in R[x] with p2 · · · ps f = g ′ h′ . Repeating this argument, shows that
f = g0 · h0 , deg(g0 ) = deg(g), deg(h0 ) = deg(h). This is a contradiction to the fact that
f is irreducible in R[x].
Proof of the Lemma of Gauß 1.4.4. Let K = Q(R) be the field of fractions ofP R. Let f ∈
m k
R[x]. We have to prove that f = f1 · · · fn , fP
i ∈ R[x] prime. We can write f = k=0 ak x .
m ak k
Define d = gcd(a0 , . . . ,am ). Then f = d · k=0 d x . We know, because R is a unique
factorization domain, that d is a unit or a product of finitely many primes. Because of
Exercise 1.4.39 they are also prime in R[x]. Therefore, we may assume that d = 1.
Let f = f1 · · · fn , fi ∈ R[x] be irreducible.
P Such a decomposition into finitely many
polynomials exists because deg(f ) = deg(fi ). Therefore, we may assume that f is
irreducible and we have to prove that f is prime. From Lemma 1.4.5 it follows that f is
irreducible in K[x]. From Exercise 1.4.33 it follows that f is prime in K[x].
Now we are ready to prove the theorem. Let f | g · h, g,h ∈ R[x]. Then f | g or f | h
in K[x]. Without loss of generality, suppose that f | h. Then it follows that f · b = h
for a suitable b ∈ K[x].Pm Then we choose c ∈ R such that c · b =: b ∈ R[x] and obtain
f b = c · h. Now f = k=0 ak xk and gcd(a0 , . . . ,am ) = 1 implies, using 1.4.39 that c | b,
that is, b ∈ R[x].
Corollary 1.4.6. Let R be a field or Z. Then R[x1 , . . . ,xn ] is a unique factorization
domain.
In the rest of the section, we discuss a generalization of this result for an arbitrary
Noetherian ring. From now on we fix a Noetherian ring R. First we need to find a
substitute of the notion of prime power for a general ring.
Definition 1.4.7. An ideal q $ R is called a primary ideal if ab ∈ q, a 6∈ q implies bn ∈ q
for some n ∈ N.
An ideal p is prime, if R/p is a domain, that is, has no zero divisors. Similarly, an
ideal I is primary, if every zerodivisor of R/I is nilpotent. An ideal I is radical , exactly
if R/I does not have any nilpotent elements, except for the zero element. This means
exactly that R/I is a reduced ring.
Lemma 1.4.8.
(1) Let m be a maximal ideal of R. Then m is a prime ideal.

√
(2) Let q be an ideal in R such that m := q is a maximal ideal. Then q is a primary
ideal.
√
(3) Let q ⊂ R be a primary ideal. Then q is prime.
(4) Let q ⊂ R be a primary ideal and d ∈
/ q be an element of R. Then q : d is a primary
ideal.
Proof.
(1) Let ab ∈ m and a 6∈ m. Then (a) + m = R because of the maximality of m. Let
1 = r · a + m for some r ∈ R, m ∈ m, then b = r · a · b + mb ∈ m because a · b and m · b ∈ m.
This implies that m is a prime ideal.
(2) Now let a · b ∈ q with a 6∈ q. If bs 6∈ q for all s ∈ N, that is, b 6∈ m, then again
(b) + m = R. Thus 1 = rb + m for some r ∈ R and m ∈ m. Hence a = rba + m · a.
Iterating this we obtain a = r(1 + m + m2 + · · · + mk )ba + mk+1 a for all k > 0. As
√
q = m, we have mk+1 ∈ q for some k (R is Noetherian). This is in contradiction to
a∈/ q. This implies b ∈ m and thus bs ∈ q for some s.
√ √
(3) Let ab ∈ q and a 6∈ q. Then (ab)m ∈ q for some m but am 6∈ q. This implies that
m s √
for some s, (b ) ∈ q and, therefore, b ∈ q.
(4) Exercise 1.4.42.
Remarks 1.4.9.
(1) It is not true that any power of a prime ideal is primary. For example, the ideal
p = (x,z) is prime in R := K[x,y,z]/(xy − z 2 ), because R/p ∼ = K[y] does not have
zero divisors. However, p2 is not primary. To see this, note that z 2 = xy ∈ p2 , x 6∈ p2
and no power of y is in p2 .
(2) It is always true from (f ) is a prime ideal it follows that f is an irreducible element.
The converse does not always hold, but it holds if R is a unique factorization
domain. Furthermore (f m ) is primary in this case. This applies for example for
rings like K[x1 , . . . ,xn ] or Z[x1 , . . . ,xn ] by the Lemma of Gauß.
One of the main goals in this section is to show that any ideal in a Noetherian ring is
a finite intersection of primary ideals. Before proving this, we need the following lemma.
Lemma 1.4.10. Let I ⊂ R be an ideal, and f ∈ R such that I : f = I : f 2 . Then
(I : f ) ∩ (I,f ) = I.
Proof. The inclusion I ⊂ (I : f ) ∩ (I,f ) is obvious. For the other inclusion, let g ∈ (I :
f ) ∩ (I,f ). We have to prove g ∈ I. Let g = x + y · f, x ∈ I, y ∈ R. As g ∈ I : f it follows
that f g = xf + y · f 2 ∈ I. Hence y · f 2 ∈ I, so that y ∈ I : f 2 = I : f . It follows that
yf ∈ I, and therefore g ∈ I.
Theorem 1.4.11 (Primary Decomposition). Let R be a Noetherian ring and I $ R
be an ideal. Then there exist finitely many primary ideals q1 , . . . ,qr in R such that I =
q1 ∩ . . . ∩ qr .
Proof. Suppose the converse. Then the set A of ideals for which the theorem is not true,
that is,
A := {I ideal in R : I is not an intersection of finitely many primary ideals},
is nonempty. The ring R is Noetherian by assumption, and therefore by 1.1.2 the set A
has a maximal element with respect to inclusion. Let I be such a maximal element. In
particular, I is not primary! Therefore, there exist a,b ∈ R such that a · b ∈ I, a 6∈ I
and bn 6∈ I for all n. We have a chain of ideals I : b ⊂ I : b2 ⊂ · · · . As our ring R is
Noetherian, we get by 1.1.2 that this chain is stationary. Hence, there exists an n with
the property I : bn = I : bn+1 = I : bn+2 = · · · = I : b2n . We use Lemma 1.4.10 to obtain
I = (I : bn ) ∩ (I,bn ).
26 1 Algebra
Now, since bn ∈ / I, we have I $ (I,bn ). Moreover as a ∈ / I and abn ∈ I we have

I $ (I : bn ). As I is maximal in A it follows that both (I : bn ),(I,bn ) ∈/ A . Therefore,
both ideals are intersections of finitely many primary ideals. But as I = (I : bn )∩(I,bn ), it
would follow that I is the intersection of finitely many primary ideals too, in contradiction
to our assumption.
Examples 1.4.12.
(1) (xy,xz) = (x) ∩ (y,z) ⊂ K[x,y,z];
(2) (xy − s,xz − s,yz − s) = (x,y,s) ∩ (x,z,s) ∩ (y,z,s) ∩ (x − y,y − z,x2 − s) ⊂ K[x,y,z,s].
We now turn our attention to the problem of uniqueness of the primary decomposi-
tion.
Definition 1.4.13.
√
• Let q be a primary ideal. The prime ideal p = q is called the associated prime of
q. In this case one says that q is a p–primary ideal.
• A primary decomposition I = q1 ∩ . . . ∩ qr is called irredundant4 if the following
two conditions are satisfied:
(1) None of the qi can be removed from the primary decomposition, that is for all
i, the inclusion I ⊂ ∩j6=i qj is strict.
√
(2) The associated primes pi := qi are pairwise different.
Lemma 1.4.14.
(1) Let q and q′ ⊂ R be p–primary ideals. Then q ∩ q′ is a p–primary ideal.
(2) Let R be Noetherian and I $ R be an ideal in R. Then there exists an irredundant
primary decomposition of I.
Proof.
√
(1) From 1.1.16, parts 2 and 4, we have q ∩ q′ = p. We have to prove that q ∩ q′ is
primary. So let ab ∈ q ∩ q′ , and a 6∈ q ∩ q′ . We may assume
√ without restriction of
√
generality that a 6∈ q. Then it follows that b ∈ q = p = q ∩ q′ , which shows the
lemma for q ∩ q′ .
(2) This follows from the first part of the lemma, as we can combine different p–primary
ideals into one p–primary ideal, for every prime ideal p.
It is important to note that an irredundant primary decomposition might not be
unique. The simplest example is:
Example 1.4.15.
(x2 ,xy) = (x) ∩ (x,y)2 = (x) ∩ (x2 ,y).
But at least the associated prime ideals are the same, that is, (x,y) and (x), in both
cases. This turns out to be the case in general.
4 In the literature this is often also called minimal, normal or reduced. Some authors, for example
Eisenbud, [Eisenbud 1995], reserve the notion irredundant primary decomposition for a primary de-
composition in which none of the terms can be dropped.
Theorem 1.4.16 (First Uniqueness Theorem of Primary Decomposition). Let

R be a Noetherian ring and I = q1 ∩ . . . ∩ qr be an irredundant primary decomposi-
√ √
tion of I. Then the prime ideals p1 = q1 , . . . ,pr = qr are uniquely determined by I,
up to permutation.
For the proof, we need the following lemma.
Lemma 1.4.17. Let R be a Noetherian ring and q be a primary ideal with associated
prime p and I = J ∩q for some ideal J 6⊂ q. Then there exists a d ∈ R such that I : d = p.
Proof.
Step 1. We first reduce to the case that J = R, that is, I = q. We choose d ∈ J and
d 6∈ q. Then it follows that
I : d = (J ∩ q) : d = (J : d) ∩ (q : d) = q : d.
As q : d is also primary by Exercise 1.4.47, and because of the property (I : x) : y = I : xy

(Exercise 1.1.15) we reduced the problem to the case J = R.
Step 2. The crucial thing to note is that for any g ∈ p \ q it follows that q $ q : g.
Indeed, g ∈ p \ q implies g m ∈ q for some m, and g 6∈ q implies m ≥ 2. If we choose m
minimal such that g m ∈ q, then g m−1 ∈ q : g, but g m−1 6∈ q. Note, moreover, that by
1.4.47 q : g is p–primary. In particular q : g ⊂ p.
Step 3. The proof of the lemma is by contradiction. So suppose that the set
A := {q ⊂ R p–primary : p 6= q : d for all d ∈ R}
is nonempty. As R is Noetherian, we can take a maximal element q in A. In particular,

p 6= q, so that we can take g ∈ p \ q with q $ q : g. Now q : g is p–primary, and by
maximality of q there exists a d with (q : g) : d = p. Hence q : (gd) = p, which is a
contradiction.
The prime ideals one can get by taking ideal quotients are called the associated
primes of I. More precisely:
Definition 1.4.18. Let I $ R be an ideal. A prime ideal p of R is called an associated
prime of I if it can be written as
p=I :d
for some d ∈ R. The set of associated primes of I is denoted by Ass(I).
Remarks 1.4.19.
(1) For a primary ideal q, we have now two definitions of an associated prime of q. We
will prove in Theorem 1.4.20 that the two definitions coincide.
(2) Alternatively, but very similarly, one can define the set of associated primes by
Ass(I) := {p : p prime, p = Ann(b), b ∈ R/I.}
The First Uniqueness Theorem 1.4.16 follows directly from the following theorem.
28 1 Algebra
Theorem 1.4.20. Let I = q1 ∩ · · · ∩ qr be an irredundant primary decomposition of I,

√
and pi = qi . Then
Ass(I) = {p1 , . . . ,pr },
that is, the set of associated primes is equal to the set of prime ideals which are associated
to the primary ideals occurring in an irredundant primary decomposition of I.
Proof. We apply 1.4.17 with J = ∩j6=i qj , and q = qi . We conclude that there exist di
such that I : di = pi for i = 1, . . . ,r. This gives the inclusion Ass(I) ⊃ {p1 , . . . ,pr }. To
show the other inclusion, let p = I : d ∈ Ass(I). Then
p = I : d = (q1 : d) ∩ · · · ∩ (qr : d).
Let {i1 , . . . ,it } be the set of indices with d ∈ / qij . This set is not empty because d ∈
/I=
∩qi . Then
p = (qi1 : d) ∩ · · · ∩ (qit : d).
p √
By Exercise 1.4.47, qik : d is pik –primary. In particular qik : d = qik = pik . By
Exercise 1.1.16 part 4 it follows that:
√ p p √ √
p = p = qi1 : d ∩ · · · ∩ qit : d = qi1 ∩ · · · ∩ qit = pi1 ∩ · · · ∩ pit
As p is a prime ideal it follows that p ⊃ pik for some k. As the other inclusion is clear it
follows that p = pik .
There is a second uniqueness theorem which we now want to formulate.
Definition 1.4.21.
(1) Let p ∈ Ass(I). Then we define
M (I,p) = {q : q ∈ Ass(I), q ⊂ p}.
(2) Suppose that p ∈ Ass(I) with M (I,p) = {p}. Then p is called a minimal associated
prime of I. This means that there does not exist an associated prime q of I with
q $ p.
(3) If p,q ∈ Ass(I), p $ q, then q is called an embedded prime ideal of I. This exactly
means that M (I,q) 6= {q}.
Example 1.4.22. Let I = (x2 ,xy) ⊂ K[x,y]. Then I : x = (x,y), I : y = (x). Ass(I) =
{(x),(x,y)}, M I,(x,y) = Ass(I). The ideal (x,y) is an embedded prime ideal.
Theorem 1.4.23 (Second Uniqueness Theorem of Primary Decomposition).
Let R be a Noetherian ring and I ⊂ R be an ideal, I = q1 ∩ · · · ∩ qr an irredundant
primary decomposition of I. Let p be an associated prime ideal of I, and put
√ √
M (I,p) = { qi1 , . . . , qis }.
Then qi1 ∩ · · · ∩ qis is uniquely determined (that is, independent of the irredundant de-
composition). In fact:
qi1 ∩ · · · ∩ qis = Ip ∩ R.
In particular, the primary ideals belonging to minimal associated primes are uniquely
determined. For such a minimal prime ideal pi one, therefore, has qi = Ipi ∩ R.
Proof. There is the following obvious generalization of 1.3.15:
primary ideals q ⊂ R with q ∩ S = ∅ ←→ primary ideals qS ⊂ RS .
The correspondence is that to a primary ideal q we assign qS . The ideal qS is primary

if q ∩ S = ∅, it is equal to RS if q ∩ S 6= ∅. On the other hand, for a primary ideal qS
in RS , we get a primary ideal q := qS ∩ R in R. For a proof, see Exercise 1.4.49. Using
Exercise 1.4.50 we get Ip = qi1 p ∩ . . . ∩ qis p and Ip ∩ R = qi1 ∩ · · · ∩ qis . This is what we
had to show.

Theorem 1.4.24. Let R be a Noetherian ring and Ass (0) = {p1 , . . . ,pk }. Then
p
(1) (0) = p1 ∩ · · · ∩ pk . In particular this means that the intersection of the minimal
prime ideals is the set of nilpotent elements.
(2) Let p ⊂ R be a prime ideal. Then pi ⊂ p for some i.
k
(3) ∪ pi \ {0} is the set of zerodivisors of R.
i=1
(4) If R is reduced, then the set of zerodivisors of R is equal to the union of the minimal
prime ideals of R.5
p
Proof. Let (0) = q1 ∩ · · · ∩ qk be an irredundant primary decomposition. Then (0) =
pp1 ∩· · ·∩pk follows, which is (1). For (2), as p is a prime ideal, it follows that p1 ∩· · ·∩pk =
(0) ⊂ p, so (2) follows.
To prove (3), let x be a zero divisor, that is, there is a y ∈ R, y 6= 0 and xy = 0. Let
√
(0) = q1 ∩ · · · ∩ qk be an irredundant primary decomposition and qi = pi . Since y 6= 0
there exists an i such that y 6∈ qi . This implies x ∈ pi . To prove the other direction, let
x ∈ pi for some i. If x is nilpotent, we are done. Otherwise xm ∈ qi for some m and
xm 6= 0. In particular, k ≥ 2.
But qi ·(q1 ∩· · ·∩qi−1 ∩qi+1 ∩· · · ∩qk ) ⊂ q1 ∩· · ·∩qk = (0) implies that there exists a
y 6= 0 such that xm y = 0. Let n ≥ 1 be minimal such that xn y = 0. Then x · (xn−1 y) = 0
and, consequently, x is a zero divisor. The final statement follows from the third.
As an application, we first want to study the structure of Artinian K–algebras (cf.
1.1.6).
Lemma 1.4.25. Let R be an Artinian K–algebra and p ⊂ R be a prime ideal. Then p
is a maximal ideal.6
Proof. R/p is an Artinian K–algebra, see 1.1.9, and an integral domain. We have to prove
that R/p is a field. Let x ∈ R/p, x 6= 0, then Lemma 1.1.8 implies that (xn ) = (xn+1 )
for some n. Hence, xn = xn+1 y for some y ∈ R/p. Since R/p is an integral domain, this
implies 1 = xy. So x is invertible, and R/p is a field.
Theorem 1.4.26. Let R be an Artinian K–algebra. Then R is uniquely (up to isomor-
phism) a finite direct sum of local Artinian K–algebras.
5 Note that minimal prime ideals of R are the minimal associated prime ideals of (0).
6 This is a very special property, and says that an Artinian K–algebra has Krull dimension zero. For a
discussion on these matters, see Section 4.1.
30 1 Algebra
Proof.
Step 1. We first prove the existence. Let (0) = q1 ∩ · · · ∩ qn be an irredundant primary
√
decomposition of the zero ideal. Then all the associated primes mi := qi are maximal
(1.4.25). By the Second Uniqueness Theorem 1.4.23 the qi are uniquely determined. We
claim that furthermore qi + qj = R if i 6= j. Indeed, otherwise m ⊃ qi + qj for some
√ √
maximal ideal m. But then m ⊃ qi = mi and m ⊃ qj = mj . This implies m = mi = mj
and, therefore, i = j (the decomposition was irredundant).
Therefore, the conditions to apply the Chinese Remainder Theorem 1.1.12 are ful-
filled and the canonical map
Mn
R −→ R/qi
i=1
is an isomorphism.
Because of 1.1.9 the R/qi are Artinian K–algebras. They are local because qi 6= qj
for i 6= j, again because the primary decomposition was irredundant.
Step 2. We now prove the uniqueness of the decomposition. So consider R ∼ = ⊕m
i=1 Ri with
Ri local. In step 1 we proved that an irredundant primary decomposition of the zero ideal
in an Artinian K–algebra is uniquely determined (up to permutation). We will show that
if qi := Ker(R −→ Ri ), then q1 ∩ . . . ∩ qm is an irredundant primary decomposition of
the zero ideal in R. This suffices to show the uniqueness, as Ri ∼
= R/qi .
(1) Consider the zero ideal in Ri . As Ri has only one maximal ideal, there is only one
primary ideal in an irredundant primary decomposition. This ideal has to be (0),
hence (0) is a primary ideal in Ri . As Ri = R/qi , this shows that the ideals qi are
primary.
(2) ∩m ∼ m
i=1 qi = (0) because R = ⊕i=1 Ri . Thus q1 ∩ · · · ∩ qm = (0) is a primary decompo-
sition.
(3) To prove that the primary decomposition is irredundant, it suffices to show that
√ √
qi 6= qj for i 6= j. For this it suffices to show that qi + qj = R for i 6= j.
Now, because R ∼ = ⊕mi=1 Ri we find x ∈ R corresponding to (0, . . . ,0,1,0, . . . ,0). So
x − 1 ∈ qi and x ∈ qj . This implies 1 = x + (1 − x) ∈ qi + qj .
Next we study the structure of quotient rings, as promised in the Examples 1.3.13.

Proposition 1.4.27. Let R = Q(R) be a reduced Noetherian ring and Ass (0) =
{p1 , . . . ,ps }. Then R ∼
= ⊕si=1 R/pi .
Proof. Note that p1 , . . . ,ps are minimal, because R is reduced. By prime avoidance,
Lemma 1.1.13, pi + pj 6⊂ ∪sℓ=1 pℓ if i 6= j.
By Theorem 1.4.24 there is a nonzerodivisor in pi + pj . As R = Q(R), all nonzerodi-
visors are units, so that pi +pj = R. As R is, moreover, reduced, we have (0) = p1 ∩· · ·∩ps
by 1.4.24. The conditions of the Chinese Remainder Theorem (1.1.12) are satisfied and
we get R ∼= ⊕si=1 R/pi .
Finally, we wish to study the rank of a module.
Definition 1.4.28. Let R be a Noetherian ring and M be a finitely generated R–module.
One says that M has rank r if M ⊗R Q(R) is a free Q(R)–module of rank r. We write
rank(M ) = r.
Lemma 1.4.29. Let R be a Noetherian ring and M be a finitely generated R–module.

M has rank r if and only if Mp is a free Rp –module of rank r for all p ∈ Ass (0) .
Proof. Since Rp = Q(R)pQ(R) , see Exercise 1.4.51, we may assume that R = Q(R). One
implication is clear. Assume that Mp is free of rank r for all prime ideals p. We use
induction on r. The case r = 0 is trivial. Suppose
r > 0. Now we choose x ∈ M such that
x 6∈ pMp for all prime ideals p in Ass (0) . Such an x exists because of Exercise 1.3.27.
Now x is an element of a minimal system of generators of Mp for all p, by Nakayama,
that is, an element of a basis of the free module Mp for all p. This implies that (M/xR)p
is free of rank r − 1 for all p. Using induction, we obtain that M/xR is free of rank r − 1
and, therefore, by Exercise 1.2.48, M ∼ = xR ⊕ M/xR is free of rank r because xR ∼ =R
∼
(xRp = Rp for all p).
Remark 1.4.30. If R is an integral domain, then every finitely generated R–module
M has a rank, namely rank(M ) = dimQ(R) M ⊗R Q(R). In general this is not true: let
R = C [x,y]/(xy) and M = (x). Then M(x) = (0) and M(y) = R(y) .
i π
Proposition 1.4.31. Let R be a Noetherian ring and 0 −→ M −→ N −→ P −→ 0 be
an exact sequence of R–modules. If two of the modules M,N,P have a rank then the third
one has a rank and rank(M )+ rank(P ) = rank(N ).
Proof. Because of 1.4.29, we may assume that R is local, and that m ∈ Ass((0)). Because
of Exercise 1.2.48 we may assume that N is always free. If N and P are free, we can use
1.3.26 to deduce that M is free. Now let N and M be free.
Suppose M ⊂ mN , then M = (0). Indeed, as m ∈ Ass((0)), there exists a d ∈ R
with m = (0) : d. It follows that dM = dmN = (0). As M is free, this implies M = (0).
If M 6⊂ mN we use induction on the rank of M . If the rank of M is 1 and m is a free
generator then i(m) 6∈ mN implies that i(m) can be extended to a set of free generators
of N . This implies that N ∼= M ⊕ P and P is free because of 1.3.26. If the rank of M is
bigger than 1 we can, for the same reason, write M = M ′ ⊕ R, N = N ′ ⊕ R such that
i(M ′ ) ⊂ N ′ and the induced sequence 0 −→ M ′ −→ N ′ −→ P −→ 0 is still exact. Then
P is free by induction.
Proposition 1.4.32. Let R be a Noetherian ring and F, G be free R–modules of finite
rank. Let ϕ : F −→ G be a homomorphism. Then the following conditions are equivalent:
(1) Im(ϕ) has rank r,
(2) Ir (ϕ) contains a nonzerodivisor and Ir+1 (ϕ) = 0.
Proof. (1) ⇒ (2).Assume that all elements of Ir (ϕ) are zerodivisors. Then Ir (ϕ) ⊂ p for
some p ∈ Ass (0) by prime avoidance. But Im(ϕ) ⊗R Q(R) is free of rank r, especially
Im(ϕ) ⊗R Rp is free of rank r. This implies that Ir (ϕ)Rp = Rp by Proposition1.3.8, and
is a contradiction to Ir (ϕ) ⊂ p.
Now assume Ir+1 (ϕ) 6= 0. As Im(ϕ) has rank r, Ir+1 (ϕ) does not contain a nonzero-
divisor. (Otherwise this would imply Ir+1 (ϕ)Q(R) = Q(R) and by Exercise 1.3.8 give that
Im(ϕ) has rank at least r + 1.) By prime avoidance and 1.4.24 there exists a p ∈ Ass (0)
with Ir+1 (ϕ)Rp ⊂ pRp . But we just proved that Ir (ϕ)Rp = Rp and, therefore, for a suit-
of bases in F ⊗R Rp and G ⊗R Rp the matrix corresponding to ϕ : Fp −→ Gp
able choice
is Id0r H0 with H 6= 0 because Ir+1 (ϕ) 6= 0. Now we can choose x ∈ p \ {0} such that
xp = (0), then xH = 0. This implies that Im(ϕ)⊗R Rp has torsion elements, and therefore
is not free which is a contradiction.
32 1 Algebra
For (2) ⇒ (1), the existence of a nonzerodivisor, implies Ir (ϕ)Q(R) = Q(R). Together
with Proposition 1.3.8 we get that Im(ϕ) has rank r.
Exercises
1.4.33. Let R be a principal ideal domain. Prove that f ∈ R is prime if and only if it is
irreducible.
(Hint: Irreducible elements in a principal ideal domain generate maximal ideals. Maximal ideals
are prime ideals.)
1.4.34. Prove that y 2 − x2 (x + 1) ∈ C [x,y] is irreducible.
1.4.35. Use Exercise 1.4.33 to prove that principal ideal domains are unique factorization do-
mains.
6 0
( Hint: Use the property that a principal ideal domain is Noetherian to prove that every f =
not being a unit is a finite product of irreducible elements.)
1.4.36. Let R be a unique factorization domain. Prove that R has the unique factorization
Q
n Q
m
property: let p1 , . . . ,pn , q1 , . . . ,qm be primes in R and pi = qi .
i=1 i=1
Prove:
(1) n = m.
(2) There exist a permutation π ∈ Sn and units u1 , . . . ,un ∈ R such that pi = ui qπ(i) .
1.4.37. Let R be a unique factorization domain and a1 , . . . ,an ∈ R be elements different from
zero. Prove that a1 , . . . ,an have a greatest common divisor.
1.4.38. Recall the Euclidean algorithm and prove that it computes in Z and K[x] the greatest
common divisor of two elements.
1.4.39. Let R be a unique factorization domain and p ∈ R prime. Let f,g ∈ R[x] be two
polynomials and assume p|f · g. Prove that p|f or p|g, that is, p is also prime in R[x].
1.4.40. Let a,b ∈ Z. Prove that (a) · (b) = (a) ∩ (b) iff gcd(a,b) = 1.
1.4.41. Let R be a unique factorization domain and f ∈ R be an irreducible element. Prove

that (f ) is a prime ideal and (f )m is primary for m ≥ 1.
1.4.42. Prove (4) of Lemma 1.4.8.
1.4.43. Let I ⊂ R be an ideal and q ⊃ I be a prime (respectively primary) ideal. Prove that
q = {f : f ∈ q} ⊂ R/I is prime (respectively primary).
1.4.44. Prove that (xi1 , . . . ,xik ) ⊂ K[x1 , . . . ,xn ] is a prime ideal.
1.4.45. Let q be a prime (respectively primary) ideal in a Noetherian ring R. Let I · J ⊂ q and
I 6⊂ q for two ideals I,J ⊂ R. Prove that J (respectively J s for some s) is contained in q.
1.4.46. Let I = (x2 yz 3 ,xyw,zw2 ) ⊂ K[x,y,z,w]. Compute the associated primes of I.
1.4.47. Let q be a primary ideal and p its associated prime ideal. Let x ∈ R, then
8
>
<R if x ∈ q
q : x = q if x 6∈ p
>
:primary with associated prime p if x 6∈ q.
1.4.48. Prove that

p p p
(1) (I,f · g) = (I,f ) ∩ (I,g).
√ p p
(2) I= IRf ∩ R ∩ (I,f ).
1.4.49. Formulate and prove a generalization of 1.3.15 to the case of primary ideals.
1.4.50. Let R be a Noetherian ring, p ⊂ R be a prime ideal.
(1) Let I,J be ideals in R. Prove that (I ∩ J)Rp = IRp ∩ JRp .
(2) Let q be a primary ideal. Show that
(a) qRp = Rp ⇐⇒ q 6⊂ p.
(b) Suppose that q ⊂ p. Then qRp is primary, and qRp ∩ R = q.
(3) Let q,q′ ⊂ p be two primary ideals with different associated primes. Prove that the asso-
ciated primes of qRp and q′ Rp are different.
(4) Let I = q1 ∩ . . . ∩ qs ∩ qs+1 ∩ . . . ∩ qt be an irredundant primary decomposition of I.
Suppose q1 , . . . ,qs ⊂ p, and qs+1 , . . . ,qt 6⊂ p Show that
(q1 ∩ . . . ∩ qs−1 )Rp 6⊂ qs Rp .
Deduce that
IRp = q1 Rp ∩ . . . ∩ qs Rp
is an irredundant primary decomposition of IRp .
(5) With the assumptions of (4), let q ∈ Ass(I) and q ⊂ p be a prime ideal. Suppose that IRp
is prime and prove IRp = qRp .
1.4.51. Let R be a Noetherian ring and p ∈ Ass((0)). Prove that Rp = ∼ Q(R)pQ(R) .
1.4.52. Let R be a√Noetherian ring and M be a finitely generated R–module, N ⊂ M a

submodule. Define M N = {x ∈ R : xq M ⊂ N for some q}, N : M = {x ∈ R : xM ⊂ N }
and Ann(M/N ) = {x ∈ R : xM/N = 0}. An element x ∈ R is called a zerodivisor in M/N
if there exists a nonzero m ∈ M/N such that xm = 0. N ⊂ M is primary in M if N 6= M
and every zerodivisor in M/N is nilpotent, that is, x being a zerodivisor in M/N implies that
xρ ∈ Ann(M/N ) for some ρ.
Prove the following statements.
√ p
(1) M N = Ann(M/N ).
√
(2) N being primary in M implies that N : M is a primary ideal (call N : M the associated
prime ideal to N ).
(3) N has an irredundant primary decomposition and the associated prime ideals are uniquely
determined.
(4) If p is an associated prime to N , then p = N : (m) for some m ∈ M .
(5) Let p1 , . . . ,ps be the set of associated primes of N , then the zerodivisors of M/N are
s
∪ pi \ {0}.
i=1
1.4.53.
(a) In 1.4.24 we proved that the set of nilpotent elements in a Noetherian ring is the inter-
section of all (minimal) prime ideals in the ring. Prove that this holds in fact for all rings
(Noetherian or not) by an application of Zorn’s Lemma.
√
(b) Let R be a ring and I ⊂ R be an ideal. Prove that I = ∩ p⊃I p.
p prime
1.4.54. Let (R,m) be a Noetherian local ring and M be an R–module. Suppose that for all
prime ideals p 6= m the localization Mp is the zero module. Show that there exists a k such the
mk · M = 0.
(Hint: Look at a primary decomposition of Ann(M ).)
1.4.55. Let (R,m) be a Noetherian local ring. Suppose that all x ∈ m \ m2 are zerodivisors.
Show that all x ∈ m are zerodivisors.
(Hint: Use 1.4.24.)
34 1 Algebra
1.5 Finite and Integral Extensions
Finite and integral extensions of rings are the ring counterparts of finite and algebraic
extensions of fields, which you (hopefully) know from a course in Galois theory. First we
give the relevant definitions.
Definition 1.5.1. Let R ⊂ S be rings.
• R ⊂ S is called a finite extension if S, considered as an R–module, is finitely
generated.
• An element α ∈ S is called integral over R if and only if α satisfies an equation:
αn + a1 αn−1 + . . . + an−1 α + an = 0
with all ai ∈ R. Such an equation is called an integral equation for α. The extension
R ⊂ S is called integral if every element x ∈ S is integral over R.
• On the other hand, R is called integrally closed in S, if every element in S which is
integral over R already belongs to R.
• R is called integrally closed if R is integrally closed in Q(R), the total quotient ring
of R.
• R is called normal if R is reduced and integrally closed.
We obtain the first lemma by copying the proofs of the corresponding statements in
field theory. As usual we denote by R[α] ⊂ S the smallest ring in S containing R and α.
Lemma 1.5.2. Let R ⊂ S and α ∈ S.
(1) Let R ⊂ S ⊂ T be ring extensions, with R ⊂ S and S ⊂ T finite. Then R ⊂ T is
finite.
(2) α is integral over R ⇐⇒ R[α] is a finitely generated R–module.
In particular, every finite extension R ⊂ S is an integral extension.
(3) Let R ⊂ S ⊂ T be ring extensions, with R ⊂ S and S ⊂ T integral. Then R ⊂ T is
an integral extension.
(4) Let S be a finitely generated R–algebra. Then S is finitely generated as an R–module
if and only if every element of S is integral over R.
Proof.
(1) If u1 , . . . ,us are generators of S as R–module and v1 , . . . ,vt are generators of T as
S–module, then one proves directly that the products ui vj generate T as an R−module.
(2) Let α be integral, with integral equation p(α) = 0, for some monic7 polynomial
p ∈ R[X] of degree n, and g(α) ∈ R[α] for some polynomial g ∈ R[X]. Then we can
perform division with remainder to write g = qp + r, where deg(r) < n (cf. Theorem
2.1.5). Plugging in α, we see that g(α) = r(α). This shows that the elements 1,α, . . . ,αn−1
generate R[α] as R−module. On the other hand, if we have finitely many generators
q1 , . . . ,qt of R[α] as R−module, and if we put n − 1 to be maximum of the degrees with
respect to α occurring in the qi , it follows that the element αn is an R−linear combination
of the qi . This gives an integral equation for α.
7 This means that the leading coefficient of p is 1.
1.5 Finite and Integral Extensions 35
(3) We give here a proof for the case that R is Noetherian. For the general case, see
Exercise 1.5.28. Let α ∈ T . By hypothesis, it satisfies an integral equation:
αn + a1 αn−1 + . . . + an−1 α + an = 0
with all ai ∈ S. So in fact α is integral over R[a1 , . . . ,an ]. Therefore R[a1 , . . . ,an ][α] is a
finitely generated R[a1 , . . . ,an ]–module, by part (2). As all elements ai are integral over
R by hypothesis, it follows from the second part and induction that R[a1 , . . . ,an ] is a
finitely generated R–module. Therefore, R[α] is a submodule of R[a1 , . . . ,an ,α] which is
also a finitely generated R–module by the first part. As R is Noetherian, it follows that
R[α] is finitely generated. Now use (2) again, to conclude that α is integral.
(4) This follows from part (2).
Definition 1.5.3. The integral closure of R in S is the set of all elements in S which
are integral over R. In the case that S is the total quotient ring of R, and R is reduced,
we simply call it the normalization of R. We denote the normalization by R. e
To use the standard proof from field theory that sum, difference and products of
integral elements are integral again we need the assumption that R is Noetherian. The
statement holds without this condition however, see Exercise 1.5.29.
Lemma 1.5.4. Let R ⊂ S be rings. Then the integral closure of R in S is a ring. The
integral closure of R in S is integrally closed in S. In particular, the normalization of a
ring is normal.
Proof. We give the proof here for the case that R is Noetherian. We have to show that if
α and β are integral, then α − β and αβ are integral. But those elements are in R[α,β],
which is a finitely generated R–module by 1.5.2. As we supposed R to be Noetherian, it
follows that R[αβ] and R[α − β] are finitely generated R−modules. By the second part
of 1.5.2 it follows that α − β and αβ are integral. The other statement follows from the
third part of 1.5.2.
To give examples of normal rings, we prove the following theorem:
Theorem 1.5.5. Let R be a unique factorization domain. Then R is normal.
r r
Proof. Suppose s ∈ Q(R) is in lowest terms, and is integral over R. So s satisfies an
equation:

r n

r n−1
s + a1 s + . . . + an = 0
with ai ∈ R. We multiply the equation with sn and conclude:
rn = −s(a1 rn−1 + . . . + an sn−1 )
Hence s divides rn , and thus s divides r. As we assumed that they have no common
factor, it follows that s is a unit in R, and so rs ∈ R.
In particular, polynomial rings over fields are normal! The converse of the theorem
does not hold, see Exercise 1.5.33.
Examples 1.5.6.
36 1 Algebra
(1) Consider the ring R := C [x,y]/(y 2 −x3 ). This ring is not normal. Indeed the element
2 3
t := xy satisfies t2 = xy 2 = xx2 = x. Therefore, t is integral, as it satisfies the integral
equation X 2 −x = 0. The normalization R e is in fact C [x,y,t]/(y 2 −x3 ,x−t2 ,y−t3 ) ∼=
C [t]. To show that t ∈ / R, consider the map ϕ : C [x,y] −→ C [t], which sends x to
t2 and y to t3 . Obviously y 2 − x3 is in the kernel of ϕ, and in fact the kernel is
generated by y 2 − x3 . Indeed, let g ∈ Ker(ϕ). Then write g = q(y 2 − x3 ) + r, where
r = r0 + r1 y, ri ∈ C [x]. Then g(t2 ,t3 ) = r0 (t2 ) + r1 (t2 ) · t3 = 0. The first term only
has terms in even degree in t, whereas the second has only odd terms in t. From
this it follows that r1 = r2 = 0. Hence C [x,y]/(y 2 − x3 ) = C [t2 ,t3 ], and t = xy is
not an element of R.
(2) Consider the ring R := C [x,y]/(x · y). This ring is not normal. Namely, x + y is not
x
a zerodivisor in R because x + y 6∈ (x) ∪ (y) (Theorem 1.4.24). Consider u = x+y .
2 2 3 2 2 2
Then u = u because x (x + y) = x = x(x + y ) = x(x + y) in R. But it can
easily be checked that u ∈/ R.
(3) Consider the ring R := C [x]/(x2 ). Then R = Q(R) and therefore R is integrally
closed. However, R is not normal, as R is not reduced.
In the second example we used quite essentially that R is not an integral domain.
In fact, the following theorem holds.
Theorem 1.5.7. Let R be a Noetherian normal local ring. Then R is an integral domain.
Proof. R is normal and, therefore, reduced. Let (0) = p1 ∩· · ·∩pr be an irredundant prime
decomposition. We have to prove that r = 1. Assuming r ≥ 2, we can find g ∈ p2 ∩· · ·∩pr ,
g 6∈ p1 (the decomposition is irredundant) and f ∈ p1 , f 6∈ p2 ∪ · · · ∪ pr (Prime avoidance
1.1.13). By the choice of f and g we have that f + g 6∈ pi for i = 1, . . . ,r. This implies
that f + g is not a zerodivisor (Theorem 1.4.24). On the other hand, f · g = 0 because
f
f · g ∈ p1 (p2 ∩ · · · ∩ pr ) ⊂ p1 ∩ · · · ∩ pr = (0). Consider u := f +g , then u2 = u because
2 3 2 2 2
f (f + g) = f = f (f + g ) = f (f + g) . This implies u ∈ R because R is normal. Note
f
that u 6= 1, as otherwise f +g = 1, and therefore g = 0. Now u(1 − u) = 0, hence u is a
zerodivisor in R. The element u is therefore contained in the maximal ideal of R. Hence
1 − u is a unit because R is local. This implies u = 0 and, therefore, f = 0 which is a
contradiction to the choice of f .
The assumption on R to be local in the theorem can be weakened to the assumption
that R has no idempotents different from 1.8 The last part of the proof just showed that
a local ring does not contain idempotents different from 1.
For more subtle arguments concerning integral elements given later, we need the
following important theorem:
Theorem 1.5.8 (Cayley-Hamilton). Let R ⊂ S, α ∈ S. Let M be an R[α]-module,
which is finitely generated as an R−module. Suppose moreover that M is a faithful9
R[α]–module. Then α is integral over R.
8 An element x in a ring is called idempotent if x2 = x.
9 An R–module M is called faithful if given a ∈ R such that a · m = 0 for all m ∈ M it follows that
a = 0.
Proof. Let u1 , . . . ,ut be generators of M as an R−module. Then for each i the element
αui is defined and is an P element of M . As the ui ′ s generate M there exist elements
t
ϕij ∈ R, such that αui = j=1 ϕij uj . In matrix notation:
 
u1
 ..  = 0, Id the unit matrix.
αId − (ϕij )  . 
ut
We multiply on the left with the adjoint (see 1.2.9) of αId − (ϕij ) to conclude that
det(αId − ϕij ) uk = 0 for k = 1, . . . ,t. This is Cramer’s
rule, see 1.2.8. As we assumed
M to be faithful, it follows that det αId − (ϕij ) = 0. Expand the determinant to find
an integral equation for α.
Remarks 1.5.9.
(1) Note that the ring S is “dummy” in some sense. We just need that our element
lives in some ring. To see why this theorem is called the Cayley-Hamilton Theorem,
we refer to Exercise 1.5.30.
(2) If moreover α sends M into I · M for some ideal I, then α satisfies an integral
equation of type
αn + a1 αn−1 + · · · + an = 0, aj ∈ I.
This motivates the following definition.
Definition 1.5.10. Let R ⊂ S be a ring extension, and I ⊂ R be an ideal. An element
x is called integral over I if it satisfies an equation of type10
xn + a1 xn−1 + . . . + an = 0, aj ∈ I.
Lemma 1.5.11. Let R ⊂ S be rings, I ⊂ R be an ideal, and x ∈ S. The following

conditions are equivalent.
(1) x is integral over I.

√
(2) The ring S ′ := R[x] is a finitely generated R–module, and x ∈ I · S ′.
(3) R[x] is contained
√ in a subring S ′ of S, where S ′ is a finitely generated R–module,
and x ∈ I · S .′
Proof. Of course, (2) =⇒ (3) is trivial. The equivalence (1) ⇐⇒ (2) can be proved as
in 1.5.2. The implication (3) =⇒ (1) uses the Cayley-Hamilton Theorem. Indeed, take
an m ∈ N such that xm ∈ I · S ′ . Therefore, multiplication with xm sends S ′ to I · S ′ .
It follows from the proof of the Cayley-Hamilton Theorem that xm satisfies an integral
equation with coefficients in I (S ′ is a faithful R[xm ]–module because 1 ∈ S ′ ).
1.5.12. Let R ⊂ S be rings, and I ⊂ R. Let R be the integral closure of R
Corollary √
in S. Then IR is the set of all elements of S which are integral over I.
10 There is a related notion of integrally dependence on I by using the condition aj ∈ I j . This notion
will not be used in this book.
38 1 Algebra
Proof. If x p∈ S and is√integral over I, then by√the second part of the previous lemma
we get x ∈ IR[x] ⊂ IR. Conversely, if x ∈ IR then xm ∈ IR for certain m. Hence
xm ∈ I · R[x1 , . . . ,xn ] for certain xi ∈ R, so that from (3) =⇒ (1) of the previous lemma
it follows that x is integral over I.
Next we will give a criterion for normality.
Theorem
√ 1.5.13 (Criterion for Normality). Let R be a Noetherian reduced ring. Let
I = I ⊂ R be a radical ideal with the following properties:
(1) I contains a nonzerodivisor of R.

(2) Let p be a prime ideal of R such that Rp is not normal. Then p ⊃ I.
Then R is normal if and only if the canonical inclusion R ⊂ HomR (I,I) is an equality.
To prove the theorem we need the following lemma.
e its normalization. Let I ⊂ R be
Lemma 1.5.14. Let R be a Noetherian reduced ring, R
an ideal containing a nonzerodivisor of R. Then
e
(1) R ⊂ HomR (I,I) ⊂ R.
√
e ∩ HomR (I,R).
(2) If, moreover, I = I, that is, I is radical, then HomR (I,I) = R
Proof.
(1) The canonical map R −→ HomR (I,I) is defined by the multiplication: to y ∈ R we
associate the map ϕy : I −→ I defined by ϕy (x) = yx. The injectivity of the canonical
map follows from the assumption that I contains a nonzerodivisor.
Now let a ∈ I be a nonzerodivisor and ϕ ∈ HomR (I,I). Then for every b ∈ I
we have ϕ(b) = b · ϕ(a)
a because aϕ(b) = ϕ(ab) = bϕ(a). Especially, ϕ(a)a ∈ Q(R) is
independent of the choice of the nonzerodivisor. The map ϕ 7→ ϕ(a)
a gives the inclusion
HomR (I,I) ⊂ Q(R). In particular, we obtain an identification
HomR (I,I) = {x ∈ Q(R) : xI ⊂ I}.
We claim that this module is already contained in the normalization Re of R. Let x ∈ Q(R)
and xI ⊂ I then I can be considered as an R[x]–module.
As an ideal containing a nonzerodivisor it is a faithful R[x]–module and we can apply
the Cayley-Hamilton Theorem 1.5.8 to obtain that x is integral over R. This implies that
e
HomR (I,I) ⊂ R.
(2) We just proved that HomR (I,I) ⊂ R. e Moreover, the inclusion HomR (I,I) ⊂ HomR (I,R)
e
is obvious. So we get HomR (I,I) ⊂ R ∩ HomR (I,R). As above, note that HomR (I,R) =
{x ∈ Q(R) : xI ⊂ R}.
To prove the converse inclusion, let z ∈ R e ∩ HomR (I,R). We have to prove that
n n−1
zI ⊂ I. Let z + an−1 z + · · · + a0 = 0 for suitable a0 , . . . ,an−1 ∈ R. Let f ∈ I, then
by multiplying the above equation with f n we obtain
(zf )n + an−1 f (zf )n−1 + · · · + a0 f n = 0.

√
This implies (zf )n ∈ I because zf ∈ R and f ∈ I. But I = I and, therefore, zf ∈ I.
This implies zI ⊂ I and proves the second part of the lemma.
Proof of Theorem 1.5.13. Let R be normal, that is, R = R.e Then Lemma 1.5.14 (1)
implies that R = HomR (I,I).
To prove the converse, assume that R = HomR (I,I). Suppose that the theorem is
e Let h ∈ R
wrong, that is, R 6= R. e \ R. Consider the ideal
J := {u ∈ R : hu ∈ R}
√ √
in R. We claim that I ⊂ J. As J = ∩p⊃J p is the intersection of all prime ideals
containing J, by Exercise 1.4.53, it suffices to show that p ⊃ I for all p ⊃ J. So let p be a
prime ideal with p ⊃ J. Now h ∈ / Rp . Indeed, otherwise ah ∈ R for some a ∈ / p. But then
also a ∈
/ J, in contradiction to the definition of J. So h gives an element of Q(Rp ) which
is integral over Rp (use the same integral equation) and, therefore, Rp is not normal.
This implies p ⊃ I by the second assumption, and proves the claim.
As our ring is Noetherian, we can now choose d ∈ N such that I d ⊂ J. This implies
d
hI ⊂ hJ ⊂ R. As h ∈ / R, we have d > 0. Let d be minimal with hI d ⊂ R. Thus
there exists an x ∈ I d−1
such that hx 6∈ R. But hx ∈ R e and hxI ⊂ hI d ⊂ R implies
e
hx ∈ R ∩ HomR (I,R) = HomR (I,I) = R. This is a contradiction to the choice of x.
Therefore, our original assumption h ∈ / R is wrong. This proves the theorem.
Remark 1.5.15. We will see later that for rings of finite type over a field (cf. Chapter
2) or analytic algebras (cf. Chapter 3) the ideal I describing the singular locus (which
can be computed using the Jacobian Criterion, cf. Chapter 4) has the properties required
in Theorem 1.5.13.
Example 1.5.16 (Continuation of Example 1.5.6).
(1) Let R = C [t2 ,t3 ], I = (t2 ,t3 ). Then
HomR (I,I) = {x ∈ Q(C [t2 ,t3 ]) = Q(C [t]) : xI ⊂ I} = C [t].
x
(2) Let R = C [x,y]/(x · y), I = (x,y) and u = x+y then R[u] = HomR (I,I) because
R[u] ⊂ Hom(I,I) and R[u] = R[u]/(u) × R[u]/(1 − u) is the product of the two
normal rings R[u]/(u) ∼
= C [y] and R[u]/(1 − u) = C [x].
We now come to the very important theorem of the finiteness of the normalization.
This theorem, due to Kronecker and Noether, is of utmost importance for the develop-
ment of the theory in this book, and will be used at several points. For example, it is
used in the proof of the Nullstellensatz in local analytic geometry, see 3.4.4, and in the
proof of the Local Parametrization Theorem, see 3.4.14, which is also very important
in our development of the theory of local analytic geometry. It is also used in Theorem
4.4.8, in order to prove that the “normalization of a space” always exists. Before stat-
ing the theorem of the finiteness of normalization, we recall some facts concerning the
discriminant of a polynomial in one variable.
Definition 1.5.17. Let k be a field (not necessarily algebraically closed), and f =
xr + a1 xr−1 + . . . + ar be a polynomial with coefficients in k. Consider a field extension
k ⊂ K in which f splits:
f = (x − α1 ) · · · (x − αr ).
Then the discriminant ∆ = ∆x (f ) with respect to the variable x is defined by:
Y
∆= (αi − αj ).
i6=j
40 1 Algebra
The discriminant therefore is zero exactly if f has a multiple root. It follows from
the Theorem on Symmetric Functions, and is a standard fact in field theory, that the
discriminant in fact is a polynomial in the coefficients a1 , . . . ,ar of f .
Lemma 1.5.18. Let R be an normal integral domain, K be its quotients field, and L ⊃ K
be a field extension. Let x ∈ L be integral over R. In particular, x is algebraic over K.
Then the minimal polynomial f (X) ∈ K[X] of x has coefficients in R.
Proof. Consider a complete set of conjugates xi of x in an algebraic closure of L. This
means that we can write Y
f (X) = (X − xi ).
(Each xi is counted pe times, where pe is the degree of inseparability. In particular xi
is counted once, if the characteristic is zero.) By a standard result in field theory, there
exists a K–isomorphism σi : K(x) ∼ = K(xi ), with σi (x) = xi for all i. It follows that also
xi is integral over R. Indeed, apply the isomorphism σi to an integral equation of x
xn + a1 xn−1 + . . . , + an = 0
to obtain the integral equation

xni + a1 xin−1 + . . . , + an = 0.
The coefficients of f (X) are the elementary symmetric functions of the xi . As sums and
products of integral elements are integral, it follows that the coefficients of f are integral
over R. As they are moreover in K, and R is normal by assumption, it follows that the
coefficients are in R.
Theorem 1.5.19 (Finiteness of the Normalization). We consider:
• a finite ring extension R ⊂ S of integral domains with R a normal ring;
• K, the quotient field of R and L the quotient field of S;
• α ∈ S a primitive element of the field extension K ⊂ L, that is L = K(α). We
suppose that α is separable; 11
• P (X) ∈ K[X] the minimal polynomial of α;
• 0 6= ∆ the discriminant of P (X).
Then we have the following statements.
(1) For all β ∈ S, the minimal polynomial of β has coefficients in R. In particular
P (X) ∈ R[X] and ∆ ∈ R.
(2) Let a polynomial q ∈ R[X] with q(α) = 0 be given. Then q ∈ (P ).
(3) Every element f in S is in the R–module generated by
1 αr−1
,..., .
∆ ∆
To put it in another way, for any f ∈ S there exists a polynomial q ∈ R[X] with
deg(q) < deg(P ) such that ∆f = q(α).
11 Recall that separable means that a minimal polynomial for α has no multiple roots.
(4) Let Se be the normalization of S. Then Se ⊂ 1

∆ R[α]. In particular Se is a finitely
generated S–module.
(5) S∆ = R∆ [α] is normal.
Proof. For the proof we need some Galois theory. Let r be the degree of P .
(1) The first statement is a consequence of Lemma 1.5.18.
(2) Note that a minimal polynomial P is monic. We perform division with remainder.
Thus q = HP + B, with degree of B smaller than r. If B is nonzero, then we divide B
by its leading coefficient. We get a monic polynomial B ′ in K[X] with B ′ (α) = 0. This
is a contradiction to the fact that P is the minimal polynomial of α. Thus B = 0, and
q ∈ (P ).
(3) We consider the splitting field M of P . The element α has r (the degree of P )
conjugates α1 , . . . ,αr in M , that is, the α = α1 ,α2 , . . . ,αr are the (pairwise different)
zeros of P in M . As P (αi ) = 0, it follows that the αi are integral over R. Now consider
an element f ∈ S ⊂ L. It is an easy fact from field theory that we can write:
(1.4) f = q0 · 1 + q1 · α + . . . + qr−1 · αr−1
with qi ∈ K for all12 i. We have to prove that ∆ · qi ∈ R for all i. Now define
f1 := q0 · 1 + q1 · α1 + . . . + qr−1 · αr−1
1
..
.
fr := q0 · 1 + q1 · αr + . . . + qr−1 · αr−1
r .
We need that the fi are also integral over R. For this we need some Galois theory. There
exist automorphisms σi of M over K mapping α to αi . If we let σi operate on equation
(1.4) it follows that fi = σi (f ). If we now let σi operate on an integral equation for f ,
it follows that the fi are also integral over R. We can view the previous equations as a
system of linear equations:
     
q0 f1 1 α1 . . . αr−11
     ..  ,
A  ...  =  ...  ; A =  ... ..
. . 
qr−1 fr 1 αr . . . αr−1
r
to which we apply our knowledge of linear algebra. First of all, the determinant D :=
det(A), as a sum and product of integral elements itself is integral. But we know more: the
matrix A isQ the well-known Vandermonde matrix and its determinant can be computed
to be D = i>j (αi − αj ). Furthermore, by definition of the discriminant ∆ = ±D2 . It
follows from Cramer’s rule 1.2.8 that
D · qν = Pν (fi ,αj )
for all ν for some polynomial Pν with integer coefficients. It follows that Dqν is integral
over R. As D also was integral we conclude that ∆qν = DDqν is integral over R. Now
∆ ∈ R ⊂ K, and the qν are in K, so that ∆ · qν is an element of K, the quotient ring of
R. We can now use our assumption that R is normal to conclude that ∆ · qν ∈ R for all
ν. This is what we had to prove.
12 Therefore, it is trivially true that for all f there exists a ∆f with ∆f · f ∈ R[α]. The point of the
theorem is, that we can take a universal ∆ which works for all f .
42 1 Algebra
(4) Let β ∈ S.e Then S[β] is a finitely generated R–module with quotient field L and,
therefore, by (3) of the theorem (applied to R ⊂ S[β]) included in R∆ [α]. This implies
that Se ⊂ R∆ [α] and is, therefore, finitely generated.
(5) Note that S∆ = R∆ [α] and, because of (4), Se∆ = R∆ [α].
Theorem 1.5.20 (Splitting of Normalization). Let R be a reduced Noetherian ring and
(0) = p1 ∩ · · · ∩ ps be an irredundant primary decomposition. Then there is a canonical
isomorphism between the normalization R ]i :
e and the direct sum of the normalizations R/p
s
M
e∼
R = ]i .
R/p
i=1
Proof. We have the following statements that are easy to prove, see Exercise 1.5.27.
(1) pi Q(R) is a prime ideal in Q(R).
(2) ∩si=1 pi Q(R) = (0).
(3) Q(R)/pi Q(R) = Q(R/pi ).
Now we use Proposition 1.4.27 and obtain
s
M s
M
Q(R) ∼
= ∼
Q(R)/pi Q(R) = Q(R/pi ).
i=1 i=1
The second isomorphism is the third statement from above. Let ui ∈ Q(R) correspond to
(0, . . . ,0,1,0, . . . ,0), with 1 at the i–th place. It follows immediately from the isomorphism
that u2i = ui . So ui ∈ R. e Note, moreover, that ui ∈ ∩j6=i pj Q(R), ui − 1 ∈ pi Q(R) for
i = 1, . . . ,s. Thus we get ui ∈ ∩j6=i (pj Q(R) ∩ R) e and ui − 1 ∈ pi Q(R) ∩ R.
e The conditions
of the Chinese Remainder Theorem are satisfied. We get an isomorphism
s
M
e∼
R = e i Q(R) ∩ R).
R/(p e
i=1
The theorem now follows because the canonical map

e i Q(R) ∩ R)
R/pi −→ R/(p e
]i ∼
induces an isomorphism R/p e i Q(R) ∩ R),
= R/(p e see Exercise 1.5.27.
Remark 1.5.21.
(1) Note that we reproved that a Noetherian normal local ring is an integral domain.
(2) Let K be a field, and S = K[x1 , . . . ,xn ]/p with p a prime ideal. We will prove in
2.2.9 (Noether normalization) that there exists R ⊂ S satisfying the assumptions
of the Theorem 1.5.19. This implies that the normalization Se is a finitely generated
S–module. In general this is not necessarily true. A similar statement holds for
so-called analytic algebras, which we will consider in the third chapter.
Theorem 1.5.22 (Lying-Over Theorem). Let R be a ring, R ⊂ S be an integral exten-

sion, and p ⊂ R be a prime ideal. Then there exists a prime ideal q in S with q ∩ R = p.
For the proof, we need the following two lemmas.
Lemma 1.5.23. Let R ⊂ S be an integral extension of rings, and p ⊂ R be a prime
ideal. Then pS ∩ R = p. In particular pS 6= S, so that there is an inclusion R/p ֒→ S/pS.
Proof. Let x ∈ pS ∩ R. Let S ′ ⊂ S be a subring such that RP⊂ S ′ is a finite extension
and x ∈ pS ′ . We can find such an S ′ as follows. Write x = ni=1 xi si with si ∈ S and
xi ∈ p. Then S ′ := R[s1 , . . . ,sn ]. It follows from 1.5.11 that x is integral over p. Let
xn + a1 xn−1 + . . . + an = 0,
with ai ∈ p. Now from x ∈ R it then follows xn ∈ p. The ideal p is prime, so that x ∈ p.

Thus pS ∩ R = p.
The following lemma is a generalization of the fact that every ideal is contained in
a maximal ideal, corresponding to the case S = {1} in the lemma.
Lemma 1.5.24. Let R be a ring, and S be a multiplicatively closed subset of R. Let I
be an ideal of R, with I ∩ S = ∅. Then there exists a prime ideal p of R with I ⊂ p and
p ∩ S = ∅.
Proof. As I ∩ S = ∅, we get IRS 6= RS . Let m ⊂ RS be a maximal ideal with IRS ⊂ m.
Then I ⊂ m ∩ R =: p.
Proof of the Lying-Over Theorem 1.5.22. By 1.5.23 we get pS ∩ N = ∅, for the multi-
plicatively closed subset N = R \ p. By 1.5.24 there exists a prime ideal P in S with
pS ⊂ P and P ∩ N = ∅. Therefore, P ∩ R ⊃ p follows from the first, and P ∩ R ⊂ p
follows from the second condition. This shows that P ∩ R = p.
Corollary 1.5.25 (Going-Up Theorem). Let R be a ring, R ⊂ S be an integral extension,
and q ⊂ p ⊂ R be prime ideals. Let Q be a prime ideal in S lying over q, that is Q∩R = q.
Then there exists a prime ideal P in S with P ∩ R = p and Q ⊂ P.
Proof. Apply the Lying-Over Theorem to the integral extension R/q ⊂ S/Q.
Theorem 1.5.26 (Going-Down Theorem). Let R ⊂ S be an integral extension of integral
domains, and suppose that R is normal. Let p ⊂ q be two prime ideals in R. Let Q be a
prime ideal in S with Q ∩ R = q. Then there exists a prime ideal P in S with P ⊂ Q
and P ∩ R = p.
Proof. We have multiplicatively closed sets N ′ := R \ p and N ′′ := S \ Q. Therefore,
N := N ′ · N ′′ is a multiplicatively closed subset of S.
Claim: We have pS ∩ N = ∅.
Otherwise, we have an element x ∈ pS ∩ N . By 1.5.11 x is integral over p. We will
show that the minimal polynomial P (X) = X n + a1 X n−1 + . . . + an of x over Q(R)
has coefficients in p. By Lemma 1.5.18 the ai are elements of R. Let x1 , . . . ,xn be the
zeros of P (X). Then x1 , . . . ,xn are also integral over p, as P (X) divides any polynomial
which gives an integral equation for x over p. By Corollary 1.5.12 it follows that the
elementary symmetric functions a1 , . . . ,an of the x1 , . . . ,xn are integral over p. As p is
prime, it follows that a1 , . . . ,an ∈ p.
44 1 Algebra
Moreover, as x ∈ N , we can write x = u · v, with u ∈ N ′ , and v ∈ N ′′ . Consider the

element v = ux . It is in S, and satisfies the equation
a1 n−1 an
vn + v + . . . + n = 0,
u u
which is a minimal equation for v. By Lemma 1.5.18, the uaii are elements of R. So we
can write
ai
= bi ,
ui
or, what is the same ai = bi ui . Now ai ∈ p, and, therefore, as ui ∈ /√p, it follows
√ that
bi ∈ p. This shows that the element v is integral over p. Therefore, v ∈ pS ⊂ qS ⊂ Q.
This is a contradiction to v ∈ N ′′ = S \ Q, and therefore proves the claim.
By 1.5.24 there exists a prime ideal P ⊂ S with P∩N = ∅ and pS ⊂ P. In particular
P ∩ N ′′ = ∅, so that P ⊂ Q. As P ∩ N ′ = ∅, it follows that P ∩ R ⊂ p, and therefore
p = P ∩ R.
The condition that R is normal in the Going-Down Theorem is necessary. See Ex-
ample 2.3.16.
Exercises
1.5.27. Show the unproved statements in the proof of Theorem 1.5.20.
]i ∼
(Hint: To prove that R/p e i Q(R) ∩ R)
= R/(p e consider the canonical inclusions
M M M
R⊂ R/pi ⊂ e i Q(R) ∩ R)
R/(p e ⊂ Q(R/pi ).)
1.5.28.
(1) Let R ⊂ S be a ring extension, and α ∈ S. Here R is not necessarily Noetherian. Use the
Cayley-Hamilton Theorem to prove that the following conditions are equivalent.
(a) α is integral over R.
(b) R[α] is a finitely generated R–module.
(c) R[α] is contained in a ring T ⊂ S, such that T is a finitely generated R–module.
(d) There exists a faithful R[α]–module M which is finitely generated as an R–module.
(2) Prove that 1.5.2 (3) holds without the hypothesis that R is Noetherian.
1.5.29. Use the Cayley-Hamilton Theorem to prove that the Noetherian hypothesis in 1.5.4 is
superfluous.
1.5.30.
(1) Let K be a field, V be a finite-dimensional vector space, and ϕ : V → V be a linear
map. Interprete 1.5.8 as the Theorem of Cayley-Hamilton you know from linear algebra.
Note that we could just as well take any ring R instead of the coefficient field K, and any
finitely generated module M instead of V .
(2) Let R be a ring, M be a finitely generated R−module, and ϕ : M → M be a surjective
R−module homomorphism. Prove that ϕ is an isomorphism.
(Hint: Consider M as R[X]–module, via X · m := ϕ(m), and extend by linearity. Then
(X) · M = M. Consider M as R[X,Y ]–module, via Y · m = Id(m) = m. Then apply
Remark 1.5.9 (2) to deduce that Y integral over X.)
fS = R
1.5.31. Let R be a reduced ring, and S ⊂ R be a multiplicatively closed set. Show that R eS .
Informally speaking, one says that taking integral closure commutes with localization.
1.5.32. Let R be a reduced ring. Prove that the following conditions are equivalent.
(1) R is normal.
(2) Rp is normal for all prime ideals p in R.
(3) Rm is normal for all maximal ideals m in R.
1.5.33. Prove that the ring R = k[x,y,z]/(z 2 − xy) is normal.

(Hint: Localize in the prime ideals (x,z) and in (y,z), and show that those localizations are
normal.) Show however, that R is not a unique factorization domain. This shows that the
converse of 1.5.5 in general does not hold.
1.5.34. Prove Nakayama’s Lemma using 1.5.8.
1.5.35. Let f = xsn + a1 xs−1 n + a2 xs−2

n + . . . + as with ai ∈ K[x1 , . . . ,xn−1 ]. Consider the
discriminant ∆ ∈ K[x1 , . . . ,xn−1 ] of f with respect to xn . Suppose that f , viewed as an element
of K[x1 , . . . ,xn ], is square-free.13 Prove that ∆ 6= 0.
1.5.36. To illustrate the Going-Up Theorem, let R = K[x,y] ⊂ S = K[x,y,z]/(z 2 − xy).

(1) Show that S is an integral extension of R.
(2) Find a prime ideal q in S with q ∩ R = (x). The ideal (x) is of course prime in R.
(3) Find a prime ideal q in S with q ∩ R = (y 2 − x3 ) ⊂ K[x,y].
13 A polynomial is by definition called square-free, if it does not have multiple factors.

46
2 Affine Algebraic Geometry
The purpose of this chapter is to introduce the reader to the basic definitions, ideas and
notations of affine algebraic geometry. Loosely speaking, an affine algebraic set is the zero
set of finitely many polynomials in an affine space K n , where K is an algebraically closed
field.1 Some of the results of this chapter are true and proved over arbitrary algebraically
closed fields. This is the reason why we do not consider only the case K = C as we will
do from Chapter 3 onwards. Admittedly, our main interest in this book is the behavior
of such zero sets in the neighborhood of a point. We will discuss and study the algebraic
case first, however, simply because the proofs of the statements in this case usually are
easier. This is because one does not have to bother one-selves with convergence questions.
Having grasped the main geometric and algebraic ideas in the affine case, the reader is
hopefully ready to tackle the local case in Chapter 3. A second reason for doing the affine
case, too, is that we need to do some algebraic geometry, in order to classify hypersurface
singularities, see Chapter 9.
Again for reasons of simplicity, we first study the zero set V (f ) of a single polyno-
mial f . These zero sets are called affine hypersurfaces. The main result in this section
is Study’s Lemma. It says that if f and g are polynomials, with f reduced and V (f ) a
subset of V (g), then f divides g. We give two proofs, whose ideas permeate large parts
of the first part of the book. Namely, we consider a “general” linear projection of the
hypersurface V (f ) on a hyperplane. This is called Noether normalization. One shows
that the projection is surjective (this is quite trivial), and the proof is completed by
using some elementary algebra, in particular, division with remainder. A second proof of
Study’s Lemma is given. Here it is not only proved that the projection is surjective, but
also the number of points in the fiber are studied.
Study’s Lemma allows us to interpret certain algebraic operations and statements
geometrically. We mention two of them.
The first is the decomposition of a hypersurface in irreducible components. A hy-
persurface is called reducible, if it can be written in a nontrivial way as the union of two
(or more) hypersurfaces. Now, if f and g are polynomials, the zero set of f · g, obviously,
is the union of the zero sets of f and g. Study’s Lemma allows us to conclude that for
the zero set of an irreducible polynomial f such a decomposition is not possible.
The second is the interpretation of the K–algebra K[x1 , . . . ,xn ]/(f ). On a hyper-
surface V = V (f ), where f is a polynomial, it is natural to consider functions which
are restrictions of polynomials on the ambient space K n . This set of functions is in a
natural way a K–algebra. If f is reduced, this algebra can be identified with the quotient
K[x1 , . . . ,xn ]/(f ) := K[V ]. In the proof, essential use of Study’s Lemma is made.
In the second section, we will study affine algebraic sets in general. That is, we study
zero sets V (I) := {a ∈ K n : f (a) = 0 for all f ∈ I} of ideals I in the polynomial ring
K[x1 , . . . ,xn ]. Hilbert’s Basis Theorem allows us to consider zero sets of only finitely
many polynomials. The main discussion is on a generalization of Study’s Lemma, called
Hilbert’s Nullstellensatz. A weaker form is given by the Weak Nullstellensatz. It says that
1 Just keep in mind the case K = C .
2.1 Affine Hypersurfaces 47
if I is a proper ideal, then V (I) 6= ∅. This statement is in fact equivalent to the structure
of maximal ideals in the polynomial ring: every maximal ideal in K[x1 , . . . ,xn ] is of type
(x1 − a1 , . . . ,xn − an ), for some (a1 , . . . ,an ) ∈ K n . So we have a one-one correspondence
between maximal ideals and points. The proof of the Weak Nullstellensatz is a conse-
quence of the Projection Theorem, which is a statement on how one can describe, under
certain circumstances, the image of an affine space under projection. To prove Hilbert’s
Nullstellensatz, we again use the Projection Theorem and Noether normalization. This
means that we study a general projection π : X = V (I) −→ K r , where r is determined
by X. Morally speaking, r is equal to the dimension of X, a property of X which is not
defined at this point of our study. The proof of the fact that the map π for a Noether
normalization then is surjective, is done by means of the Projection Theorem. Hilbert’s
Nullstellensatz itself is an easy consequence of the surjectivity of π, and is an immediate
generalization of the second part of the proof of Study’s Lemma. A shorter proof of the
Nullstellensatz out of the weak Nullstellensatz is by using the so-called Rabinowitch trick.
This is done in the exercises. We presented the other proof however, as a substitute for
the Rabinowitch trick in the local case does not, to our knowledge, exist. We want to
have similar proofs for both the affine and the analytic case.
In Section 2.3, we consider regular maps between affine algebraic sets. These maps
are restrictions of polynomial maps. In particular, we define isomorphisms between affine
algebraic sets. If ϕ : V −→ W is a regular map between affine algebraic varieties, we get,
by composition, a K–algebra homomorphism ϕ∗ : K[W ] −→ K[V ]. (Arrows reverse!)
Again geometric statements about maps between V and W are interpreted as algebraic
statements on the map ϕ∗ : K[W ] −→ K[V ]. For example, it is shown that ϕ is injective,
if and only if ϕ∗ is surjective.
2.1 Affine Hypersurfaces
Definition 2.1.1. Let K be a field, f ∈ K[x1 , . . . ,xn ] be a polynomial. The zero set
V (f ) ⊂ K n of f is defined by:
V (f ) := {(a1 , . . . ,an ) ∈ K n : f (a1 , . . . ,an ) = 0}.
A set V ⊂ K n is called an affine hypersurface if there exists a nonzero f ∈ K[x1 , . . . ,xn ],
such that V = V (f ).
We give some examples. For the examples, we use K = R for two reasons: real
pictures are easier to draw; R is not algebraically closed and almost everything which
fails over a field which is not algebraically closed fails already over R.
Examples 2.1.2.
(1) f = x2 + y 2 − 1 ∈ R[x,y]. The set V (f ) is a circle in the plane.
48 2 Affine Algebraic Geometry
(2) f = x2 − yz ∈ R[x,y,z]. The set V (f ) is a cone in three-dimensional space.
As is well-known, see 1.4.4, the polynomial ring K[x1 , . . . ,xn ] is factorial. It follows
easily from the definition, see Exercise 2.1.18, that f |g (or what is the same (f ) ⊃ (g))
implies that there is an inclusion of zero sets V (f ) ⊂ V (g). The converse does not hold,
as the following simple examples shows.
Examples 2.1.3.
(1) Take f = x2 + 1 ∈ R[x], g = x ∈ R[x]. Then ∅ = V (f ) ⊂ V (g) = {0}. But as the
degree of f is bigger than the degree of g, f cannot divide g.
(2) Let f be a nonconstant polynomial. Then V (f 2 ) = V (f ), so in particular V (f 2 ) ⊂
V (f ). But obviously f 2 does not divide f .
The phenomena occurring in the first example is typical for fields which are not
algebraically closed. In fact, one has the following theorem.
Theorem 2.1.4 (Study’s Lemma). Let K be an algebraically closed field,
f,g ∈ K[x1 , . . . ,xn ]. Suppose that f is irreducible, and that V (f ) ⊂ V (g). Then f |g.
There is a proof of Study’s Lemma using so-called resultants, see for example,
[Brieskorn-Knörrer 1986], page 192. We present two other proofs of Study’s Lemma here.
Both proofs are based on the following two facts.
• Division with remainder for polynomials.

• Noether normalization for hypersurfaces.
We will first discuss the well-known Division with Remainder Theorem.

Theorem 2.1.5 (Division with Remainder). Let R be a ring and f ∈ R[x] be a polyno-
mial. Suppose that the leading coefficient of f is a unit in R. Then for all g ∈ R[x] there
exist q,r ∈ R[x], with deg(r) < deg(f ) such that:
g = qf + r.
Moreover, q and r are uniquely determined by g and f .

Proof. Without loss of generality, we may assume that f is monic, that is, the leading
coefficient of f is equal to 1.
Step 1: Existence: The proof is by induction on deg(g). If deg(g) < deg(f ) then one
can take q = 0 and r = g. Otherwise, write
f = xd + bd−1 xd−1 + . . . + b0 ; g = am xm + am−1 xm−1 + . . . + a0 ,
with m ≥ d. Consider
g = g − am xm−d f = (am−1 − am bd−1 )xm−1 + . . .

e
The degree of ge is smaller than the degree of g. By induction e g = qef + r with

deg(r) < deg(f ). With q = qe + am xm−d the equality g = qf + r holds.
Step 2: Uniqueness. Suppose g = q1 f + r1 = q2 f + r2 with deg(r1 ), deg(r2 ) < deg(f ).
It suffices to show that q1 = q2 , as r1 = r2 follows immediately from it. Subtracting we
get:
(q1 − q2 )f = (r2 − r1 )
As the degree of r1 and r2 is smaller than the degree of f , all the terms on the left hand
side of degree at least d have to vanish. Because f is monic, and looking at the term
of highest degree, it follows that q1 and q2 have the same degree and the same leading
coefficient. Subtraction of this highest degree term from both q1 and q2 gives us q1′ and
q2′ with (q1′ − q2′ )f = (r2 − r1 ). By (descending) induction we have that q1′ = q2′ . Hence
q1 = q2 .
Lemma 2.1.6 (Noether Normalization for Hypersurfaces). Let K be a field with an
infinite number of elements, and let f ∈ K[x1 , . . . ,xn ] be a polynomial of degree s. Then
after a general linear change of coordinates, one has:
(2.1) f = c(xsn + a1 xns−1 + . . . + as )
with c ∈ K a nonzero constant, and the ai are elements of K[x1 , . . . ,xn−1 ].

Proof. We write
f = fs + fs−1 + . . . + f0
where each fi is homogeneous of degree i. The polynomial fs is nonzero, so by Exercise
2.1.19 for general (b1 , . . . ,bn ) ∈ K n the value c := fs (b1 , . . . ,bn ) 6= 0. After renumbering
the coordinates, we may furthermore assume that bn 6= 0. Take new coordinates y1 , . . . ,yn
related to x1 , . . . ,xn by the formula’s:
xi = bi yn + yi for 1 ≤ i ≤ n − 1, xn = bn yn
Then g(y1 , . . . ,yn ) := f (x1 (y1 , . . . ,yn ), . . . ,xn (y1 , . . . ,yn )) is such that g(0, . . . ,0,yn ) =
fs (b1 , . . . ,bn )yns + fs−1 (b1 , . . . ,bn )yns−1 + . . ., with fs (b1 , . . . ,bn ) = c 6= 0. Hence g =
c(yns + a1 yns−1 + . . . + as ), for certain ai ∈ K[y1 , . . . ,yn−1 ]. This shows the lemma.
The Noether normalization for hypersurfaces can be reformulated in a more algebraic
manner. The algebraic formulation is the one which is going to be generalized later.
Corollary 2.1.7 (Noether Normalization for Hypersurfaces). For a general coordinate
system,
K[x1 , . . . ,xn ]/(f )
is a finitely generated K[x1 , . . . ,xn−1 ]–module.2
2 In this case it is even a free module.
Proof. After a general change of coordinates we may assume that f is of type (2.1). We
claim that K[x1 , . . . ,xn ]/(f ) is as K[x1 , . . . ,xn−1 ]–module generated by {1,xn , . . . ,xns−1 }.
Consider a g ∈ K[x1 , . . . ,xn ]. We do division with remainder
g = qf + r
where r is a polynomial in xn of degree less than s with coefficients in K[x1 , . . . ,xn−1 ].

Hence, modulo f , the element g is indeed in the module generated by {1,xn , . . . ,xns−1 }.
Examples 2.1.8.
(1) Consider f = xy. Then obviously C [x] ⊂ C [x,y]/(xy), but C [x,y]/(xy) is not
a finitely generated C [x]–module. Indeed, if it was finitely generated, say by the
elements g1 , . . . ,gs ∈ C [x,y], then let d be the maximum of the degree in y in the
polynomials gi . Then y d+1 is not in the module generated by the g1 , . . . ,gs . The
module C [x,y]/(xy) is as C [x]–module generated by {1,y,y 2 , . . .}.
(2) Consider f = xy − 1. Then C [x] ⊂ C [x,y]/(xy − 1). But again C [x,y]/(xy − 1) =

C [x,x−1 ] is not a finitely generated C [x]–module. It is generated by {1,y,y 2 , . . .} =
{1,x−1 ,x−2 , . . .}.
f
These two examples are quite typical. In both cases, we considered the inclusion of
rings C [x] ⊂ C [x]/(f ). Consider the projection π : X = V (f ) −→ C which sends
(a,b) to a. In the first example, this map is surjective, but the fiber π −1 (0) consists of
infinitely many points, in fact the whole y–axis. In the second example, the map π is not
surjective: π −1 (0) = ∅. But, if f is as in formula (2.1), it turns out that the projection
π : V (f ) −→ C n−1 on the first n − 1 coordinates is surjective, and has only finitely many
points in each fiber. This fact, which is very easy to prove, is crucial for our first proof
of Study’s Lemma.
Lemma 2.1.9. Let K be an algebraically closed field, and f = xsn + a1 xns−1 + . . . + as ,
with ai ∈ K[x1 , . . . ,xn−1 ]. Consider the projection
π : V (f ) −→ K n−1 ; (p1 , . . . ,pn ) 7→ (p1 , . . . ,pn−1 ).
Then π is surjective.
Proof. This is quite obvious. For, let p = (p1 , . . . ,pn−1 ) ∈ K n−1 . Because K is alge-
braically closed, the equation
X s + a1 (p)X s−1 + . . . + as (p) = 0,
has a solution pn . Then (p1 , . . . ,pn ) ∈ V (f ).

First proof of Study’s Lemma 2.1.4. Let g ∈ K[x1 , . . . ,xn ] with V (f ) ⊂ V (g). We have
to show that f | g. By the Noether Normalization Lemma for Hypersurfaces, we may
suppose that f is of type
f = xsn + a1 xns−1 + . . . + as .
By 2.1.7 K[x1 , . . . ,xn ]/(f ) is a finitely generated K[x1 , . . . ,xn−1 ]–module. Consider the
class g of g in K[x1 , . . . ,xn ]/(f ). Hence, by 1.5.2, g is integral over K[x1 , . . . ,xn−1 ], that
is, it satisfies an equation of type
gk + b1 g k−1 + . . . + bk = 0,
with bi ∈ K[x1 , . . . ,xn−1 ]. This equation is to be considered in the ring K[x1 , . . . ,xn ]/(f ).
Therefore
g k + b1 g k−1 + . . . + bk ∈ (f ).
We take k minimal with this property. Take a point p ∈ V (f ), so that f (p) = 0. From
V (f ) ⊂ V (g) it follows that g(p) = 0. Therefore bk (p) = 0. But bk only depends on
the first n − 1 coordinates. As π : V (f ) −→ K n−1 is surjective by the previous lemma,
it follows that bk (p1 , . . . ,pn−1 ) = 0 for all (p1 , . . . ,pn−1 ) ∈ K n−1 . Hence bk is the zero
polynomial. Therefore
g · (g k−1 + b1 g k−2 + . . . + bk−1 ) ∈ (f ).
As f is irreducible, either f | g or f | g k−1 +b1 g k−2 +. . .+bk−1 . But the second possibility
is excluded by our choice of k to be minimal.
The second proof of Study’s Lemma we present looks somewhat more complicated. It
gives, however, also some more information on the fibers of the map π : V (f ) −→ K n−1 .
Moreover, the condition irreducible in this proof can be somewhat weakened: we only
need f to be square-free.
Lemma 2.1.10. Let K be an algebraically closed field. Consider a square-free polynomial

f ∈ K[x1 . . . ,xn ] of the form
f = xsn + a1 xns−1 + a2 xns−2 + . . . + as ; ai ∈ K[x1 , . . . ,xn−1 ].
Let ∆ ∈ K[x1 , . . . ,xn−1 ] be the discriminant of f (with respect to xn ), see 1.5.17. Con-
sider the projection
π : V (f ) −→ K n−1 ,
(p1 , . . . ,pn ) 7→ (p1 , . . . ,pn−1 ) =: q.
Then π −1 (q) has s, the degree of f , elements exactly when ∆(q) 6= 0.
Proof. Consider q = (p1 , . . . ,pn−1 ) ∈ K n−1 . Look at the polynomial
(2.2) xsn + a1 (q)xns−1 + a2 (q)xns−2 + . . . + as (q) ∈ K[xn ].
Because K is algebraically closed, this polynomial has s (distinct) roots exactly when its
discriminant, which is ∆(q), is nonzero. If b1 , . . . ,bt , t ≤ s are the roots of the polynomial
(2.2), then the points in the fiber π −1 (q) are (p1 , . . . ,pn−1 ,b1 ), . . . ,(p1 , . . . ,pn−1 ,bt ).
Example 2.1.11. Consider the curve C given by the equation y 2 − x2 (x + 1) = 0. The
discriminant is given by 4x2 (x + 1), so is zero for x = 0 and x = −1. Consider the
projection π from C to the x–axis.
y
We see indeed from the picture (over C !) that π −1 (p) consists of two points for
p 6= 0, − 1, whereas for p = 0 or −1 it consists of one point. For other examples see
Exercise 2.1.21.
Second proof of Study’s Lemma. After, if necessary, performing a linear change of coor-
dinates, we may assume, by Noether normalization for hypersurfaces, that
f = xsn + a1 xns−1 + a2 xns−2 + . . . + as
with ai ∈ K[x1 , . . . ,xn−1 ]. As f is square-free, by Exercise 2.1.20 the discriminant ∆ of

f is nonzero. We consider the projection π : V (f ) −→ K n−1 as in Lemma 2.1.10. Let
D = {q ∈ K n−1 : ∆(q) = 0}. We have shown that the inverse image of q 6∈ D under π
has exactly s points. Now do division with remainder for polynomials in the variable xn :
g = qf + r
r = r0 + r1 xn + . . . + rs−1 xns−1 , ri ∈ K[x1 , . . . ,xn−1 ].
Let (p1 , . . . ,pn−1 ,pn ) ∈ V (f ). As by assumption V (f ) ⊂ V (g), it follows that
g(p1 , . . . ,pn−1 ,pn ) = 0. As g = qf + r, it follows that r(p1 , . . . ,pn−1 ,pn ) = 0. Hence, if

q = (p1 . . . ,pn−1 ) 6∈ D the polynomial
r0 (q) + r1 (q)xn + . . . + rs−1 (q)xns−1
has s roots. But the polynomial r(p1 , . . . ,pn−1 ) has degree s − 1, has s roots, so must be
the zero polynomial. Therefore, its coefficients are zero. Hence
r0 (p1 , . . . ,pn−1 ) = · · · = rs−1 (p1 , . . . ,pn−1 ) = 0.
As we can take q = (p1 , . . . ,pn−1 ) general, it follows from Exercise 2.1.19 that the poly-
nomials, r0 , . . . ,rs−1 ∈ K[x1 , . . . ,xn−1 ] are zero. Hence r = 0. Therefore, g = qf , that is,
f | g.
Definition 2.1.12. An affine hypersurface C ⊂ K n is called irreducible if from C =
C1 ∪ C2 , with C1 and C2 affine hypersurfaces, it follows that either C1 = C or C2 = C.
A hypersurface is called reducible if it is not irreducible.
An irreducible component of an affine hypersurface C is an irreducible affine hyper-
surface contained in C.
A consequence of Study’s Lemma is that one can characterize irreducible hypersur-

faces by its defining equation.
Theorem 2.1.13. Let K be an algebraically closed field. Let C = V (f ) be an affine
hypersurface. Then the following conditions are equivalent.
(1) C is irreducible.
(2) There exists an irreducible g ∈ K[x1 , . . . ,xn ], and a k ∈ N with f = g k .
It follows that that for all affine hypersurfaces C we have a decomposition
C = C1 ∪ · · · ∪ Cs ,
where Ci are the irreducible components of C, and Ci 6= Cj for i 6= j.

Proof. Suppose first that C is irreducible. Suppose f = f1n1 · · · fsns . Then V (f ) =
V (f1 · · · fs ). So we may assume that f is square-free. We have to prove that f is ir-
reducible under this assumption. Suppose the converse. Then we can write f = f1 · f2 ,
where f1 and f2 do not have common factors and are not constant. In particular f1 ∤ f2
and f2 ∤ f1 . From Study’s Lemma it follows that V (f1 ) 6⊂ V (f2 ) and V (f2 ) 6⊂ V (f1 ).
Therefore C 6= V (f1 ) and C 6= V (f2 ). But as f = f1 ·f2 it follows that C = V (f1 )∪V (f2 ).
This is a contradiction to the assumption that C is irreducible.
To prove the converse, suppose C = C1 ∪ C2 , C = V (f ), C1 = V (f1 ) and C2 =
V (f2 ), and both C1 and C2 are not equal to C. Let h be irreducible and h | f1 . Then
V (h) ⊂ V (f1 ) $ V (f ) (Exercise 2.1.18). Study’s Lemma implies h | f . and thus h | g k .
As the polynomial ring is a unique factorization domain, and g is irreducible, it follows
that h = g · c for some c ∈ K. This implies V (h) = V (g) = V (f ) which is in contradiction
to the assumption V (h) $ V (f ).
Examples 2.1.14.
(1) f (x,y) = x · y. Then V (f ) has two components, given by x = 0 and y = 0.
x=0
y=0
(2) As f (x,y) = y 2 − x2 (x + 1), see 2.1.11, is an irreducible polynomial, see Exercise

1.4.34, the affine hypersurface V (f ) is irreducible.
We now consider polynomial functions on affine hypersurfaces.
Definition 2.1.15. Let C = V (f ) ⊂ K n be an affine hypersurface. A function g : C −→
K is called regular, if g is the restriction of a polynomial map g : K n −→ K. With the
obvious addition and multiplication the regular functions on C form a ring . It is in fact
a K–algebra. The ring of regular functions of C we denote by K[C].
A polynomial map g : K n −→ K is just given by a polynomial g ∈ K[x1 , . . . ,xn ].
Let such a g be given. The restrictions g and g + αf for any α ∈ K[x1 , . . . ,xn ] define the
same function on C = V (f ). Using Study’s Lemma, we have the converse, as soon as we
assume that the hypersurface C is defined by a square-free f .
Theorem 2.1.16. Let K be an algebraically closed field. Let f ∈ K[x1 , . . . xn ] be square-
free. Then the ring of regular functions on C = V (f ) is equal to K[x1 , . . . ,xn ]/(f ).
Proof. Let g1 ,g2 ∈ K[x1 , . . . xn ] be polynomial functions such that the functions g1|C and
g2|C are equal. We have to show that the classes of g1 and g2 in K[x1 , . . . ,xn ]/(f ) are
equal. By assumption g1 − g2 vanish on C, which means that V (f ) ⊂ V (g1 − g2 ). From
Study’s Lemma it follows that f |g1 − g2 , or, what is the same g1 − g2 ∈ (f ). This shows
the theorem.
The assumption that f is square-free is essential. For example, let f = x2 ∈ K[x,y],
and C = V (f ). Then x and 0 define the same function on V (f ), but x 6= 0 ∈ K[x,y]/(x2 ).
Exercises In these exercises, K is an algebraically closed field.

2.1.17. Deduce Study’s Lemma for square-free f from Study’s Lemma for irreducible f .
2.1.18. Let f,g ∈ K[x1 , . . . ,xn ] with f | g. Show that V (f ) ⊂ V (g).

2.1.19. Let K be a field with an infinite number of elements. Let f ∈ K[x1 , . . . ,xn ] be a
nonconstant polynomial. Prove, by induction on n, that for “general” points (a1 , . . . ,an ) ∈ K n
we have f (a1 , . . . ,an ) 6= 0. Specify what we mean by general.
2.2 Affine Varieties 55
2.1.20. Let f = xsn + a1 xs−1

n + a2 xs−2
n + . . . + as , with ai ∈ K[x1 , . . . ,xn−1 ]. Suppose that f is
square-free. Prove that the discriminant of f is nonzero.
2.1.21.
(1) Consider the curve defined by y 2 − xy + x2 = 0, and the projection π on the x–axis.
Compute for which points on the x–axis the inverse image of π consists of less than two
points.
(2) Consider the cone x2 − yz, see example 2.1.2., and the projection π on the y,z–plane.
Compute for which points on the y,z–plane the inverse image of π consists of less than
two points. What about projection on the x,y–plane?
2.1.22. Prove, using only the statement of Study’s Lemma, that V (f ) 6= ∅ for any nonunit
f ∈ K[x1 , . . . ,xn ].
2.2 Affine Varieties
In this rest of this chapter we assume, unless said otherwise explicitly that K is alge-
braically closed. In this section, we will, more generally, consider zero sets of any set of
elements in K[x1 , . . . ,xn ].
Definition 2.2.1. Let F ⊂ K[x1 , . . . ,xn ] be a subset. We put
V (F ) := {p ∈ K n : f (p) = 0 for all f ∈ F }.
and call it the zero set of F . A subset V of K n is called an (affine) algebraic set, or an
affine variety if there exists a subset F ⊂ K[x1 , . . . ,xn ] with V = V (F ). Conversely, for
any set A ⊂ K n we define
I (A) = {f ∈ K[x1 , . . . ,xn ] : f (p) = 0 for all p ∈ A},
and call it the ideal of (functions vanishing on) A.

Elementary properties of these notions are gathered in the following lemma.
Lemma 2.2.2. Let V,W be algebraic subsets of K n , A be a subset of K n , I,J,Iα be ideals
of K[x1 , . . . ,xn ], F be a subset of K[x1 , . . . ,xn ], and B be an index set.
(1) I (A) is an ideal. Even more, I (A) is a radical ideal.

(2) V ⊂ W =⇒ I (V ) ⊃ I (W ).
(3) I ⊂ J =⇒ V (I) ⊃ V (J).
(4) Let I be the ideal generated by F ⊂ K[x1 , . . . ,xn ]. Then V (F ) = V (I).
(5) ∅ and K n are algebraic sets.
(6) V (I · J) = V (I ∩ J) = V (I) ∪ V (J). In particular, (induction) finite unions of
algebraic sets are algebraic.
P
(7) V ( α∈B Iα ) = ∩α∈B V (Iα ). In particular, arbitrary intersections of algebraic sets
are algebraic sets.
(8) I (V ∪ W ) = I (V ) ∩ I (W ).
√
(9) I ⊂ I (V (I)).
Proof.
(1) The fact that I (A) is an ideal is easy. Moreover, if f m ∈ I (A), then f m (p) = 0 for
all p ∈ A. Hence f (p) = 0 for all p ∈ A, that is f ∈ I (A).
(2) and (3) are trivial.
(4) As F ⊂ I, Pnit follows that V (F ) ⊃ V (I). Suppose on the other hand that a ∈ V (F ),
and let f = i=1 ai fi , fi ∈ F be an element of I. As a ∈ V (F ), we have that fi (a) = 0
for all i. Hence f (a) = 0. As f is arbitrary it follows that a ∈ V (I).
(5) ∅ = V ({1}), hence is algebraic. Furthermore, K n = V ({0}), hence is algebraic too.
(6) We prove three inclusions.
(a) V (I) ∪ V (J) ⊂ V (I ∩ J): let a in V (I) ∪ V (J). Therefore, either f (a) = 0 for all
f ∈ I or f (a) = 0 for all f ∈ J. In particular, if f ∈ I ∩ J, it follows that f (a) = 0.
(b) V (I ∩ J) ⊂ V (I · J): this follows from the inclusion of ideals I · J ⊂ I ∩ J.
(c) V (I · J) ⊂ V (I) ∪ V (J): let a ∈ V (I · J). In particular f (a) · g(a) = 0 for all f ∈ I
and g ∈ J. Suppose a ∈ / V (J). Then we must show a ∈ V (I). As a ∈ / V (J) there
exists a g ∈ J with g(a) 6= 0. But as f (a)g(a) = 0 for all f ∈ I, it follows that
f (a) = 0 for all f ∈ I. This gives a ∈ V (I).
P
(7)PAs the ideal generated by ∪α∈B Iα is exactly α∈B Iα , it follows that V (∪α∈B Iα ) =
V ( α∈B Iα ). The equality V (∪α∈B Iα ) = ∩α∈B V (Iα ) is easy, and therefore left to the
reader.
(8) If f vanishes on V ∪ W , it must vanish on both. The inclusion I (V ∪ W ) ⊂ I (V ) ∩
I (W ) follows. Suppose on the other hand that f ∈ I (V ) ∩ I (W ), and let a ∈ V ∪ W .
Then either a ∈ V or a ∈ W . In any case, for f ∈ I (V ) ∩ I (W ) it follows that f (a) = 0.
(9) The inclusion I ⊂ I (V (I)) is a√tautology. As I (V (I)) is a radical ideal by the first
part of the lemma, it follows that I ⊂ I (V (I)).
Examples 2.2.3.
(1) Consider I = (xy,xz,yz). It is not too difficult to show, see Exercise 2.2.30, that
I = (x,y) ∩ (x,z) ∩ (y,z). Hence X := V (I) consist of the union of the coordinate
axes.
On the other hand, X is also the zero set of the ideal I ′ := (x,y) · (x,z) · (y,z) =
(x2 y,x2 z,xyz,xz 2 ,xy 2 ,yz 2 ,y 2 z). This ideal is smaller than I and, therefore, is not
radical. This can be seen immediately. For example, xy ∈ / I ′ but (xy)2 = y · x2 y is
′
in I . This is quite typical. Products of radical ideals in general are not radical.
(2) Consider I1 = (z,y 2 − x3 ), and I2 = (x,y), and X1 = V (I1 ),X2 = V (I2 ). Then
the union X = X1 ∪ X2 is the zero set of the ideal I = I1 · I2 = xz,yz,x(y 2 −
x3 ),y(y 2 − x3 ) . Again this ideal is not radical. Indeed, y 2 − x3 ∈
/ I, but (y 2 − x3 )2 =
2 3 2 2 3 2 3
√
y · y(y − x ) − x · x(y − x ) ∈ I, so that y − x ∈ I.
Definition 2.2.4. We define a topology on K n by taking the closed subsets to be the

algebraic subsets. That this is indeed a topology follows from Lemma 2.2.2 (5), (6) and
(7). This topology is called the Zariski topology of K n . More generally, for any algebraic
set V ⊂ K n , we define the Zariski topology on V to be the topology on V induced by
the Zariski topology on K n . So a closed subset in V is of type V (I) ∩ V for some ideal
I ⊂ K[x1 , . . . ,xn ].
The Zariski topology of C n is much coarser than the usual (Euclidean) topology of
n
C . For example, for n = 1, a subset of C is closed in the Zariski topology when this
subset is finite or equal to C . The Zariski topology does not behave well under products,
see Exercise 2.2.34.
Lemma 2.2.5. Let A ⊂ K n be a subset. Then A = V (I (A)).
Proof. The inclusion A ⊂ V (I (A)) is a tautology. Moreover, V (I (A)) is the smallest
closed subset containing A. Indeed, let B be a closed subset with A ⊂ B. As B is
closed, there exists an ideal J with B = V (J). Then J ⊂ I (V (J)), hence B = V (J) ⊃
V (I (V (J))) = V (I (B)) ⊃ V (I (A)). The last inclusion is by combining (2) and (3) of
the Lemma 2.2.2. As A is the smallest closed subset containing A, the lemma follows.
Theorem 2.2.6 (Hilbert’s Nullstellensatz). Let I ⊂ K[x1 , . . . ,xn ] be an ideal. Then
√
I = I (V (I)).
We will give several proofs of the Nullstellensatz in this book. For still other proofs,
we refer to the book of Eisenbud [Eisenbud 1995]. Hilbert’s Nullstellensatz is a gener-
alization of Study’s Lemma, see Exercise 2.2.31. Why is it called the Nullstellensatz
(German for Theorem of Zeros)? Well, let I $ K[x1 , . .. ,xn ]. Then it follows from the
Nullstellensatz that V (I) 6= ∅: indeed, otherwise I V (I) = I (∅) = K[x1 , . . . ,xn ], hence
√
I = (1). Therefore for some n we have 1 = 1n ∈ I and, therefore, I = K[x1 , . . . ,xn ], a
contradiction. Hence from the Nullstellensatz it follows:
Theorem 2.2.7 (Weak Nullstellensatz).
Let I $ K[x1 , . . . ,xn ] be an ideal. Then V (I) 6= ∅.
In fact, one is able to show via the so-called Rabinowitch trick, see Exercise 2.2.32,
that conversely, the weak Nullstellensatz implies the Nullstellensatz. In this reduction,
both the number of variables and the number of equations grow by one! In particular
we get an alternative proof of Study’s Lemma, by following the lines in this proof of
the Nullstellensatz, in which one would have to prove the weak Nullstellensatz for two
functions. We will prove the Weak Nullstellensatz by using the Projection Theorem,
which we treat now.
Theorem 2.2.8 (Projection Theorem). Consider a nonzero ideal I ⊂ K[x1 , . . . ,xn ].
Suppose that I contains an element f0 which is regular in xn , that is
f0 = xdn + a1 xd−1
n + . . . + a0
for certain elements ai ∈ K[x1 , . . . ,xn−1 ]. Let p : K n −→ K n−1 be the map which
sends (x1 , . . . ,xn ) to (x1 , . . . ,xn−1 ). Define moreover I ′ := I ∩ K[x1 , . . . ,xn−1 ], and put
X ′ = V (I ′ ). Then
p(X) = X ′ .
Proof. The inclusion p(X) ⊂ X ′ is the easy inclusion. Indeed if a ∈ p(X), then there
exists a b ∈ K with (a,b) ∈ X. Let f ∈ I ′ . Then f ∈ I and thus f (a,b) = 0. But f does
not depend on xn so that f (a) = 0.
For the inclusion X ′ ⊂ p(X) we will use the Nullstellensatz for the case n = 1.
This case is easy, as K[x] is a principal ideal domain, and every f has a root, since by
assumption K is algebraically closed. Assume that a ∈ / p(X). We will show that a ∈/ X ′.
′
To prove this, it suffices to find an g ∈ I with g(a) 6= 0. This g is constructed in two
steps.
Step 1. First we will show that for all f ∈ K[x1 , . . . ,xn ] there exists a representation
(2.3) f = g0 + g1 xn + . . . + gn−1 xd−1

n + hf
with hf ∈ I, gi ∈ K[x1 , . . . ,xn−1 ] and gi (a) = 0.

To show this we look at the map
ϕ : K[x1 , . . . ,xn ] −→ K[xn ], f 7→ f (a,xn ).
Now
∃b : b ∈ V (ϕ(I)) ⇐⇒ ∃b : f (a,b) = 0 for all f ∈ I ⇐⇒ a ∈ p(X),
By assumption a ∈ / p(X), so that V (ϕ(I)) = ∅. From the Nullstellensatz in one variable

we get ϕ(I) = K[xn ]. In particular for all f we can find an h′f ∈ I with ϕ(h′f ) = ϕ(f ).
With gf := f − h′f we get a decomposition
f = gf + h′f ,
such that h′f ∈ I and gf (a,xn ) = 0. We divide gf through f0 . We get
d−1
X
gf = qf0 + gj xjn .
j=0
Pd−1
If we plug in (a,xn ) we get 0 = q(a,xn )f0 (a,xn ) + j=0 gj (a)xjn . By assumption f0 (a,xn )
is a polynomial in xn of degree d. By the uniqueness statement in division with remainder
it follows that gj (a) = 0 for all j. Now we put hf := h′f +qf0 and we get the representation
(2.3).
Step 2. We apply (2.3) to the functions 1,xn , . . . ,xd−1
n . We get
1 = g0,0 +...+ g0,d−1 xd−1

n + h0
.. .. ..
. . .
xd−1
n = gd−1,0 + . . . + gd−1,d−1xd−1
n + hd−1
with gij (a) = 0 for all i,j and hj ∈ I for all j. In matrix notation
   
1 h0
   
(Id − gij )  ...  =  ...  .
xd−1
n hd−1
Let B be the adjoint matrix of (Id − gij ). We multiply with B from the left, and it
follows from Cramer’s rule that
   
1 h0
 ..   .. 
(2.4) det (Id − gij )  . =B . .
xd−1
n hd−1
We define g := det (Id − gij ) ∈ K[x1 , . . . ,xn−1 ]. From the first row of (2.4) we see that
g ∈ I, so g ∈ I ′ . As moreover gij (a) = 0, it follows that g(a) = 1 6= 0. This was our
goal.
Proof of the weak Nullstellensatz. The proof is by induction on n. The case I = (0) is
obvious. Otherwise there exists an f0 ∈ I, for which we may, by the Noether normaliza-
tion for hypersurfaces 2.1.6, assume to fulfill the conditions of the Projection Theorem.
Furthermore I ′ 6= (1), as otherwise I = (1). By induction V (I ′ ) 6= ∅, and thus by the
Projection Theorem p(X), and in particular X = V (I) is nonempty.
For the following theorem we do not need that K is algebraically closed. The state-
ment of the following so-called Noether normalization is a direct generalization of the
Noether normalization for hypersurfaces, see 2.1.7.
Theorem 2.2.9 (Noether Normalization). Let K be a field with infinitely many ele-
ments, A = K[x1 , . . . ,xn ]/I a finitely generated K–algebra. Then after a general linear
coordinate change, there exists a number r ≤ n, and an inclusion
K[x1 , . . . ,xr ] ⊂ A,
such that A is a finitely generated K[x1 , . . . ,xr ]–module. If moreover I 6= (0), then r < n.
If V = V (I), we also say that the projection on the first r coordinates
π : V −→ K r
is a Noether normalization of V .
Proof. If I = (0), we can take x1 , . . . ,xr to be x1 , . . . ,xn . Otherwise the proof goes
by induction on n, the case n = 0 being trivial. Let g ∈ I with g 6= 0. By Noether
normalization for hypersurfaces see 2.1.7, after a general (linear) change of coordinates,
we may assume that K[x1 , . . . ,xn ]/(g) is a finitely generated K[x1 , . . . ,xn−1 ]–module.
Because g ∈ I, then certainly K[x1 , . . . ,xn ]/I is a finitely generated K[x1 , . . . ,xn−1 ]/(I ∩
K[x1 , . . . ,xn−1 ])–module. We apply the induction hypotheses to I ∩K[x1 , . . . ,xn−1 ]. After
a general linear coordinate change in x1 , . . . ,xn−1 we have that

K[x1 , . . . ,xr ] ⊂ K[x1 , . . . ,xn−1 ]/ I ∩ K[x1 , . . . ,xn−1 ]
is finitely generated. Composing with the inclusion

K[x1 , . . . ,xn−1 ]/ I ∩ K[x1 , . . . ,xn−1 ] ⊂ K[x1 , . . . ,xn ]/I,
allows us to deduce that K[x1 , . . . ,xr ] ⊂ K[x1 , . . . ,xn ]/I is finitely generated. This proves
the theorem.
Corollary 2.2.10. Let π : V −→ K r be a Noether normalization. Then π is surjective.
Proof. This follows by induction from the Projection Theorem 2.2.8.
Corollary 2.2.11. Hilbert’s Nullstellensatz holds for prime ideals p ⊂ K[x1 , . . . ,xn ].
Proof. As a prime ideal is radical, the Nullstellensatz says that p = I (V (p)) for a prime
√
ideal p. The inclusion ( p = p ⊂ I (V (p)) is easy, see Lemma 2.2.2 (9). For the other
inclusion, consider a Noether normalization
K[x1 , . . . ,xr ] ֒→ K[x1 , . . . ,xn ]/p.
Now let f ∈ I (V (p)) be given. By looking at its class in K[x1 , . . . ,xn ]/p, we see that
there exists an integral equation
f s + a1 f s−1 + . . . + as = 0 ∈ K[x1 , . . . ,xn ]/p, ai ∈ K[x1 , . . . ,xr ].
We take s minimal with this property. From p ⊂ I (V (p)), it follows as ∈ I (V (p)). This
implies that as (p) = 0 for all p ∈ V (p). If p = (p1 , . . . ,pn ) then as (p) = as (p1 , . . . ,pr ).
Using Corollary 2.2.10 we know that V (p) −→ K r is surjective and, consequently, for
all (p1 , . . . ,pr ) ∈ K r there exists a p = (p1 , . . . ,pr ,pr+1 , . . . ,pn ) ∈ V (p) which implies
as (p1 , . . . ,pr ) = 0 for all (p1 , . . . ,pr ) ∈ K r . This implies as ∈ I (K r ) = 0. Hence
f (f s−1 + a1 f s−2 + . . . + as−1 ) ∈ p.
Because p is a prime ideal, and s was to be taken minimal, it follows that f ∈ p. This is
what we had to show.
To complete our proof of the Nullstellensatz it remains to show the following lemma.
Lemma 2.2.12. Suppose the Nullstellensatz holds for all prime ideals p ⊂ K[x1 , . . . ,xn ].
Then the Nullstellensatz holds for all ideals I ⊂ K[x1 , . . . ,xn ].
√
Proof. Let I = q1 ∩ . . . ∩ qr be a primary decomposition of I. Let pi = qi for all i. In
particular V (pi ) = V (qi ). Then V (I) = V (p1 ∩ . . . ∩ pr ) = V (p1 ) ∪ . . . ∪ V (pr ) by Lemma
2.2.2 (6). Hence by (8) of the same lemma
I (V (I)) = I (V (p1 )) ∩ . . . ∩ I (V (pr )).

√ p r √
Thus I= ∩i=1 qi = ∩ri=1 qi = ∩ri=1 pi = ∩ri=1 I (V (pi )) = I (V (I)).
Hilbert’s Nullstellensatz can be used to give a geometric interpretation of primary
decomposition for radical ideals in a polynomial ring over an algebraically closed field.
For this, we need the following definition.
Definition 2.2.13. Let X be a topological space. Then X is called irreducible if from
X = X1 ∪ X2 with X1 and X2 closed subsets of X it follows that either X1 = X or X2 =
X. The space X is called reducible if it is not irreducible. An irreducible decomposition
of a topological space is a decomposition
X = X1 ∪ . . . ∪ Xr ,
where the Xi are irreducible, and for all i 6= j, Xi is not contained in Xj .

We will prove that irreducible algebraic sets with the Zariski topology correspond
to zero sets of prime ideals. By an application of the Primary Decomposition Theorem
it will follow that an affine algebraic set X always has an irreducible decomposition.
Example 2.2.14. Let X = V (zx,zy) ⊂ K 2 , with the Zariski topology.
Then X = X1 ∪ X2 , where X1 = V (x,y) and X2 = V (z). Moreover, none of the Xi ,

for i = 1,2 is contained in the other.
Theorem 2.2.15.
(1) Let X ⊂ K n be an irreducible algebraic set. Then I (X) is a prime ideal.

(2) Conversely let p be a prime ideal in K[x1 , . . . ,xn ]. Then V (p) is irreducible.
Proof.
(1) Suppose f · g ∈ I (X). Then V (f ) ∪ V (g) = V (f · g) ⊃ V I (X) = X. The last
equality is by 2.2.5. Hence

X = X ∩ V (f ) ∪ X ∩ V (g) ,
so that by assumed irreducibility of X either X = X ∩ V (f ), or X = X ∩ V (g). So either

X ⊂ V (f ), of X ⊂ V (g), which means that either f ∈ I (X) or g ∈ I (X). This is what
we had to show.
(2) Suppose
V (p) = X1 ∪ X2 with X1 and X2 closed. Then by the Nullstellensatz p =
I V (p) = I (X1 ) ∩ I (X2 ). As p is prime, either p = I (X1 ) or p = I (X2 ). Therefore,
either V (p) = X1 or V (p) = X2 .
Theorem 2.2.16. Let I be a radical ideal and let I = p1 ∩ . . . ∩ pr be an irredundant
primary decomposition of I. Let Xi = V (pi ). Then X = X1 ∪ . . . ∪ Xr is an irreducible
decomposition of X. An irreducible decomposition of X is, up to permutation of the Xi ,
uniquely determined.
Proof. For the first statement, we have X = V (I) = V (p1 ) ∪ . . . ∪ V (pr ). The V (pi )
are irreducible by Theorem 2.2.15. Suppose V (pi ) ⊂ V (pj ). Then by the Nullstellensatz
pi = I (V (pi )) ⊃ I (V (pj )) = pj . As the primary decomposition of I is irredundant,
it follows that i = j. This shows that the Xi are pairwise not contained in each other.
Hence, X = X1 ∪ . . . ∪ Xr is a decomposition of X into irreducible components.
To show that an irreducible decomposition is uniquely determined, consider an ir-
reducible decomposition X = X1 ∪ . . . ∪ Xr . Let pi = I (Xi ). Then the pi are prime
ideals and I (X) = p1 ∩ . . . ∩ pr . The irredundancy of the primary decomposition follows
from pj 6⊂ pi for i 6= j, which follows from Xi 6⊂ Xj . By the Uniqueness Theorems of
a Primary Decomposition, a primary decomposition of a radical ideal is uniquely deter-
mined. Now Xi = V (I (Xi )) and it follows that the irreducible decomposition is uniquely
determined.
Lemma 2.2.17. Let X ⊂ K n be an irreducible algebraic set, U ⊂ X an open nonempty
subset. Then U = X. In particular, let h ∈ K[x1 , . . . ,xn ] be a polynomial with h 6∈ I (X).
Then X \ V (h) = X.
Proof. As X is closed the inclusion U ⊂ X is trivial. Suppose the inclusion is proper.
Then U is closed, and we can write X = U ∪X \ U. This gives a nontrivial decomposition
of X into closed subspaces, a contradiction. Hence the inclusion is an equality.
Despite the Nullstellensatz, it is often important to consider ideals which are not
radical.
Example 2.2.18. Let I = (y 2 ,xy) ∈ K[x,y]. Then, see 1.4.15, I = (y) ∩ (y 2 ,x) is a
primary decomposition. Hence V (I) = V (y) ∪ V (x2 ,y). Hence V (I) = {(x,y) : y =
0} ∪ {(0,0)}. One tries to visualize this by regarding the point (0,0) to be “embedded”
in the line {(x,y) : y = 0} as the following picture suggests.
This is the reason why, for an ideal I in a Noetherian ring R, the associated primes
of I which are not minimal are called embedded primes.
The next thing we do is to give a geometric interpretation of the the ideal quotient.
Theorem 2.2.19. Consider an ideal I ⊂ K[x1 , . . . ,xn ], and let d be an element in
K[x1 , . . . ,xn ]. Let I = q1 ∩ . . . ∩ qr be an irredundant primary decomposition of I. Then
[
V (I : d) = V (qi ).
i:d6∈qi
Proof. From (the easy) Exercise 1.1.15 it follows that I : d = (q1 : d)∩. . .∩(qr : d) so that
V (I : d) = V (q1 : d) ∪ . . . ∪ V (qr : d). Now if d ∈ qi , then qi : d is the whole polynomial
ring. In√particular V (qi : d) is empty. Suppose, on the other hand, that d ∈ / qi . Then
√
qi = qi : d by 1.4.47, so it follows that V (qi : d) = V (qi ). From this the theorem
follows.
For radical ideals I the theorem means that V (I : d) picks out the irreducible
components of V (I) on which the function d does not vanish identically. In mathematical
terms, we can write V (I : d) = V (I) \ V (d). Here the bar means taking the closure in
the Zariski topology. This will be slightly generalized in Exercise 2.2.38.
Examples 2.2.20.
(1) Let I = (xy,xz,yz) = (x,y)∩(x,z)∩(y,z). Then I is a radical ideal. Let d = x. Then

d vanishes on the y and z–axis so that V (I : d) is the x–axis. Indeed I : d = (y,z).
(2) Consider the ideal I = (x2 ,xy). Let d = x. Then I : d = (x,y). So V (I : d) does
consist of the origin. So the statement V (I : d) = V (I) \ V (d) does not generalize
to nonradical ideals.
The final topic in this section is the discussion of localizations. First a definition.
Definition 2.2.21. Let V ⊂ K n be an algebraic set
(1) A function f : V −→ K is called regular, if f is the restriction of a polynomial

map f : K n −→ K. The regular functions on V form a K–algebra, which we
denote by K[V ]. It is easy to see, see Exercise 2.2.27, that K[V ] is isomorphic to
K[x1 , . . . ,xn ]/I (V ). The ring K[V ] is called the coordinate ring, or ring of regular
functions of V .
(2) Let V be an irreducible algebraic set. A rational function on V is an element of
K(V ) := Q(K[V ]), the quotient field of K[V ].
Note that a rational function on V in general is not a genuine function on V : if
f = hg , then, in general, the value of f at a point p with h(p) = 0 is not defined.
Lemma 2.2.22.
(1) Let V be an irreducible algebraic set, and p a point of V . Let mp be the maximal
ideal of K[V ] of regular functions of V vanishing at the point p. Then K[V ]mp are
those rational functions f on V which have a well-defined value at p.
(2) More generally, let W ⊂ V be an irreducible subset of V , and p ⊂ K[V ] induced

by I (W ). Elements of K[V ]p are those rational functions on V which have a well
defined value on a nonempty Zariski open subset of W .
(3) Let f 6= 0 be an element of K[V ]. Then elements of K[V ]f have a well defined
value on V \ V (f ).
Proof. Elements f in K[V ]mp are of type f = hg with h ∈ / mp . In particular h(p) 6= 0.

g(p)
So f (p) = h(p) is a well defined element of K. The other two statements are proved in a
similar way.
Now about localizations for not necessarily irreducible V . First an example.
Example 2.2.23.
Consider the coordinate axes V (x · y), so K[V ] = K[x,y]/(xy). Let p = (x). Then all
elements which are not divisible by x are units in K[V ]p . In particular y is a unit. Hence
x = 0 in K[V ]p , so that K[V ]p consists of the rational functions on the y–axis.
This was an example for the following theorem, whose proof is left as an exercise.
Theorem 2.2.24. Let X = X1 ∪ . . . ∪ Xr be an irreducible decomposition of an algebraic
set X, and Y be an irreducible subset of X, defined by the prime ideal p. Then K[X]p ∼ =
⊕Y ⊂Xi K[Xi ]p . In particular, if pi is the ideal of Xi then K[X]pi ∼
= K(Xi ), the ring of
rational functions on Xi .
Remark 2.2.25. Let V be an algebraic set, with coordinate ring K[V ]. Then for any
set F ⊂ K[V ] we can define the zero set V (F ) which is a subset of V , and for any subset
A ⊂ V we can define the ideal of functions vanishing on A. All the properties of 2.2.2 hold
in this more general case, and the proofs are literally the same. Furthermore, the Null-
stellensatz holds in this more general case, and follows from the ordinary Nullstellensatz.
The proofs are left as Exercise 2.2.27.
Exercises
2.2.26.
(1) Let (a1 , . . . ,an ) ∈ K n be a point. Show that (x1 − a1 , . . . ,xn − an ) is a maximal ideal of
K[x1 , . . . ,xn ].
(Hint: Consider the map K[x1 , . . . ,xn ] −→ K : f 7→ f (a1 , . . . ,an )).
(2) Let m ⊂ K[x1 , . . . ,xn ] be a maximal ideal. Prove that the following two statements are
equivalent.
(a) V (m) 6= ∅.
(b) There exists (a1 , . . . ,an ) ∈ K n such that m = (x1 − a1 , . . . ,xn − an ).
(3) Let V ⊂ K n be an algebraic set. Prove that there is a one-one correspondence,

points of V ←→ maximal ideals of K[V ],
where K[V ] = K[x1 , . . . ,xn ]/I (V ). The point (a1 , . . . ,an ) ∈ V corresponds to the maxi-
mal ideal generated by the classes of x1 − a1 , . . . ,xn − an .
2.2.27.
(1) Show that K[V ] is isomorphic to K[x1 , . . . ,xn ]/I (V ).
(2) For a subset F ⊂ K[V ], define V (F ) := {a ∈ V : f (a) = 0 for all f ∈ F }. V (F ) is called
the zero set of F . Show that the properties of Lemma 2.2.2 also holds in this more general
context. Show that the Nullstellensatz holds in the ring K[V ].
(3) Show that V is irreducible if and only if K[V ] is an integral domain.
(4) Let a,b ∈ V . Show that there exists a regular function f ∈ K[V ] with f (a) 6= f (b).
(Hint: Reduce to the case V = K n . )
2.2.28. We proved hat V (I + J) = V (I) ∩ V (J). It is not true in general that I (V ∩ W ) =
I (V ) + I (W ). Give a counterexample.
2.2.29.
(1) Let I ⊂ K[x1 , . . . ,xn ] be an ideal, such that the vector space K[x1 , . . . ,xn ]/I has finite
dimension s. Show that the set V (I) consists of at most s points.
(Hint: Use the fact that an Artinian algebra is the direct sum of local Artinian algebras,
see 1.4.26.)
(2) Suppose moreover that I is radical. Show that V (I) consists of exactly s points.
(3) Show that for a Noether normalization π : X → K k , with s the degree of the corresponding
field extension K(x1 , . . . ,xk ) ⊂ K(X) = Q(K[X]), the number of points in a fiber is at
most s.
2.2.30.
(1) Prove that (x,y)∩(x,z)∩(y,z) is an irredundant primary decomposition of I = (xy,xz,yz).
(2) Prove that (z,y 2 − x3 ) ∩ (x,y) = (yz,xz,y 2 − x3 ).
2.2.31. Prove that Study’s Lemma is a consequence of Hilbert’s Nullstellensatz.
2.2.32. Show that the Nullstellensatz follows from the weak Nullstellensatz by using the so-
called Rabinowitch
` ´ trick which goes as follows.
Take f ∈ I V (I) , and consider the polynomial ring K[x1 , . . . ,xn ,y]. Take the ideal J generated
by I and yf − 1. Show that V (J) = ∅. Apply the weak Nullstellensatz to conclude that 1 ∈ J.
Then use the fact that modulo J, the equality y = f1 holds.
2.2.33. Consider affine varieties,
X = V (f1 , . . . ,fs ) ⊂ K n , Y = V (g1 , . . . ,gt ) ⊂ K m ,
where f1 , . . . ,fs ∈ K[x1 , . . . ,xn ], g1 , . . . ,gt ∈ K[y1 , . . . ,ym ]. Show that the Cartesian product
X × Y is also an affine variety, as zero set of the ideal
(f1 , . . . ,fs ,g1 , . . . ,gt ),
where the fi and gi are interpreted as elements of K[x1 , . . . ,xn ,y1 , . . . ,ym ].
2.2.34. Suppose m,n > 0. Prove that the Zariski topology of K n × K m is not equal to the
product of the Zariski topologies of K n and K m .
2.2.35. Let V be an algebraic set. Let mk = (x1 , . . . ,xk ) ⊂ K[x1 , . . . ,xk ], and let m be a maximal
ideal of K[V ]. A Noether normalization of K[V ]m , by definition is an inclusion
K[x1 , . . . ,xk ]m k ⊂ K[V ]m
for some k, such that K[V ]m is a finitely generated K[x1 , . . . ,xk ]m k –module. Show that after a
general coordinate change, a Noether normalization of K[V ]m exists.
2.2.36.
(1) Let V be an irreducible topological space, f : V −→ W be continuous. Show that f (V ) is
irreducible.
(2) Let V ⊂ W be topological spaces. Suppose that V is irreducible. Show that the closure V
of V in W is also irreducible.
2.2.37. In this exercise we look at the Going-Up Theorem in the special case that p is a maximal
ideal.
(1) Let R ⊂ S be an integral extension of integral domains. Show that R is a field if and only
if S is a field.
(2) Let R ⊂ S be an integral extension of rings. Let m be a maximal ideal in R, and q ⊂ S
be a prime ideal with q ∩ R = m. Prove that q is a maximal ideal.
Give geometric interpretations of these statements.
2.2.38. Let I,J be ideals in the polynomial ring K[x1 , . . . ,xn ]. Suppose I is radical. Prove that
V (I : J) = V (I) \ V (J).
2.2.39. Let Y ⊂ V be algebraic sets. Show that I (Y ) = ∩p∈Y mp .
2.3 Maps between Algebraic Sets
In this section we look at the morphisms we allow between algebraic sets. As an algebraic
set is defined by polynomial equations, it is a natural idea to allow between them only
polynomial mappings. The simple case of regular functions, that is, regular maps V −→ K
has already been considered in 2.2.21.
Definition 2.3.1. Let V ⊂ K n and W ⊂ K m be affine varieties. A map ϕ : V −→ W
is called a regular map between the affine varieties V and W if ϕ is the restriction of a
polynomial map ϕ : K n −→ K m . Regular maps V −→ K are called regular functions.
This means the following: suppose (x1 , . . . ,xn ) are coordinates on K n , and (y1 , . . . ,ym )
are coordinates on K m . Then there exist functions ϕ1 , . . . ,ϕm ∈ K[x1 , . . . ,xn ], such
that the point with coordinates (x1 , . . . ,xn ) ∈ V is sent to the point with coordinates
(y1 , . . . ,ym ) = ϕ1 (x1 , . . . ,xn ), . . . ,ϕm (x1 , . . . ,xn ) ∈ W .
Definition 2.3.2. An isomorphism ϕ : V −→ W is a regular map which has a two-sided
inverse. Two affine varieties V,W are called isomorphic if there exists an isomorphism
ϕ : V −→ W between them. Of course, being isomorphic is an equivalence relation.
It is easy to check that the composition of two polynomial maps is a polynomial map.
This in particular gives the following. Let ϕ : V −→ W be a regular map. Let f ∈ K[W ],
ϕ f
that is, f : W −→ K be a regular function. Then the composition f ◦ ϕ : V −→ W −→ K
is a regular function too. Hence we get a map
ϕ∗ : K[W ] −→ K[V ]; f 7→ ϕ ◦ f
which one checks to be a K–algebra homomorphism. Note that if moreover ψ : W −→ X,

we have the property (ψ ◦ ϕ)∗ = ϕ∗ ◦ ψ ∗ . So we assign to each regular map a K–algebra
homomorphism. There is a converse to this.
Theorem 2.3.3. Suppose a K–algebra homomorphism α : K[W ] −→ K[V ] is given.

Then there is a unique regular map ϕ : V −→ W with ϕ∗ = α.
Proof. We write
K[V ] = K[x1 , . . . ,xn ]/I (V ), K[W ] = K[y1 , . . . ,ym ]/I (W ).

2.3 Maps between Algebraic Sets 67
α
The composition K[y1 , . . . ,ym ] −→ K[W ] −→ K[V ] we also denote by α. Consider the
“coordinate” functions y1 , . . . ,ym ∈ K[y1 , . . . ,ym ]/I (W ). Then α(yi ) = ϕi (x1 , . . . ,xn ) ∈
K[V ] for some ϕi . Take lifts ϕi ∈ K[x1 , . . . ,xn ] of ϕi . Then ϕ := (ϕ1 , . . . ,ϕm ) is a
polynomial map ϕ : K n −→ K m . As both α and ϕ∗ are K–algebra maps, and they
agree on yi , it follows that they agree on all polynomials in the yi . Hence α = ϕ∗ :
K[y1 , . . . ,ym ] −→ K[V ]. So, uniqueness is clear. As different lifts differ by elements in
I (V ) the restriction of ϕ to V is well-determined. We need to show
ϕ(V ) ⊂ W.
Take p ∈ V , and consider ϕ(p). As W = V (I (W )) by 2.2.5 it suffices to show f (ϕ(p)) = 0

for all f ∈ I (W ). Now f (ϕ(p)) = f ◦ϕ(p) = ϕ∗ (f )(p) = α(f )(p). As α : K[W ] −→ K[V ],
it maps all functions in I (W ) to zero. Hence α(f ) is the zero function, in particular
α(f )(p) = 0.
The following follows immediately.
Corollary 2.3.4. Algebraic sets V and W are isomorphic if and only if K[V ] and K[W ]
are isomorphic as K–algebras.
As K–algebra homomorphisms between affine rings correspond to regular maps of

the corresponding varieties, one would like to translate certain properties of K–algebra
homomorphisms into properties of the corresponding regular map. The following theorem
gives a first example of this principle.
Theorem 2.3.5. Let algebraic sets V,W , algebraic subsets i : X ֒→ V , j : Y ֒→ W , and
a regular map ϕ : V −→ W be given. Then

(1) ϕ(X) ⊂ Y ⇐⇒ ϕ∗ I (Y ) ⊂ I (X),

(2) the fiber ϕ−1 (Y ) is the zero set of the set ϕ∗ I (Y ) ,
(3) a regular map is continuous in the Zariski topology.

Proof. (1) As I (X) ⊃ I (V ), we get a canonical surjection i∗ : K[V ] −→ K[X] which is
just restriction of regular maps. Similarly, we have a canonical surjection j ∗ : K[W ] −→
K[Y ]. Now if ϕ∗ (I (Y )) ⊂ I (X), we get an induced map ϕ∗ : K[Y ] −→ K[X], with
i∗ ◦ϕ∗ = ϕ∗ ◦j ∗ By the previous theorem we get a map ϕ : X −→ Y , such that j◦ϕ = ϕ◦i.
This exactly means that ϕ(X) ⊂ Y . If, on the other hand ϕ sends X to Y , that is, we
have a map ϕ : X −→ Y , we get an induced map ϕ∗ : K[Y ] −→ K[X], so a commutative
diagram
ϕ∗
K[W ] // K[V ]
i∗ j∗
ϕ∗

K[Y ] // K[X].

It follows that ϕ∗ I (Y ) ⊂ I (X).
(2) We apply the first part to X = ϕ−1 (Y ). We get

ϕ∗ I (Y ) ⊂ I ϕ−1 (Y ) ,
∗

so that, by taking zero
sets, V ϕ I (Y ) ⊃ ϕ−1 (Y ). Toprove the other
inclusion,
let p ∈ V ϕ I (Y ) . Then mp = I (p) ⊃ I V ϕ∗ I (Y ) ⊃ ϕ∗ I (Y ) . Therefore
∗
ϕ(p) ∈ Y follows from the first part applied to X = {p}. Hence p ∈ ϕ−1 (Y ) as was to be
proved.
(3) From the second part it follows that inverse images of Zariski closed sets are Zariski
closed. Thus, ϕ is continuous.
Example 2.3.6. Let V = V (I), where I = (y 2 − x) ⊂ K[x,y]. Consider the projection
ϕ : V −→ W = K, the projection on the x–axis. Then the map ϕ∗ is just the inclusion
map ϕ∗ : K[x] ⊂ K[x,y]/(y 2 −x). We take the subset Y of W = K defined by x(x−1) = 0,
so consisting of the points 0 and 1. Hence the fiber is defined by the ideal (x(x−1),y 2 −x))
in K 2 so consists of three points, namely (0,0),(1,1) and (1, − 1).
0 1
Lemma 2.3.7. Let a regular map ϕ : V −→ W be given, and let ϕ∗ : K[W ] −→ K[V ]
be the induced map on coordinate rings. Then
ϕ is injective ⇐⇒ ϕ∗ is surjective.
Proof. First suppose that ϕ∗ is surjective. Suppose that a,b ∈ V with ϕ(a) = ϕ(b). Then
for all g ∈ K[W ], it follows that ϕ∗ (g)(a) = g ◦ ϕ(a) = g ◦ ϕ(b) = ϕ∗ (g)(b). Suppose
a 6= b. Choose an f ∈ K[V ] with f (a) 6= f (b). This we can do by Exercise 2.2.27. Because
ϕ∗ is surjective, there is a g ∈ k[W ] with f = ϕ∗ (g). This is a contradiction.
For the converse, suppose ϕ is injective. Hence V can be viewed as a subset of W :
V ⊂ W ⊂ K n . Hence I (V ) ⊃ I (W ), so that K[W ] = K[x1 , . . . ,xn ]/I (W ) −→ k[V ] =
K[x1 , . . . ,xn ]/I (V ) is surjective.
The corresponding statement with injective and surjective interchanged, is not true,
by the following example which we have seen before.
Example 2.3.8. Consider V = {(x,y) ∈ K 2 : xy − 1 = 0}, let W be the x–axis, and
ϕ : V −→ W be the projection. Then ϕ∗ is the inclusion K[W ] = K[x] ⊂ K[V ] =
K[x,y]/(xy − 1) = K[x,x−1 ]. However, ϕ is not surjective: it misses the point 0.
Note that in this example, K[V ] is not a finitely generated K[W ]–module. If we
insist however on this, the converse holds. The following therefore is a generalization of
the fact that for a Noether normalization π : V −→ K r , the map π is surjective, and
that the fibers consist of finitely many points. The proof is left as Exercise 2.3.17.
Theorem 2.3.9. Let ϕ : V −→ W be a regular map, with ϕ∗ : K[W ] −→ K[V ] injective,

and such that K[V ] is a finitely generated K[W ]-module. Then ϕ is surjective, and the
fibers consist of finitely many points.
Note that although in Example 2.3.8, the map ϕ is not surjective, it is at least
“almost” surjective. To make this more precise, we need to understand localizations of
affine rings.
Lemma 2.3.10. Let V ⊂ K n be an algebraic set, with coordinate K[V ]. Take 0 6= f ∈
K[V ]. Then V \ V (f ) ⊂ V is homeomorphic in the Zariski topology to an algebraic set
X, whose coordinate ring K[X] is the localization K[V ]f .
Proof. Let K[V ] = K[x1 , . . . ,xn ]/I. Take a new coordinate y, and consider the set
X := V (I,y · f − 1) ⊂ K n+1 . The coordinate ring of X is K[x1 , . . . ,xn ,y]/(I,y · f − 1)
which is as K–algebra, by sending y to f1 , isomorphic to K[V ]f . The projection π :
X −→ K n on the first n coordinates is easily seen to be a one to one map of X onto its
image V r V (f ). The map π : X −→ V r V (f ) is regular, hence continuous. To show
that π is a homeomorphism, we have to show that a closed subset in X is sent to a closed
subset in V r V (f ). It suffices to show this for irreducible sets Y in X, corresponding
to a prime ideal pf in K[V ]f . By 1.3.15 pf ∩ K[V ] = p is a prime ideal in K[V ], and
conversely a prime ideal in K[V ] with f ∈ / p extends to a prime ideal in K[V ]f . We claim
that the image of V (pf ) is equal to V (p) r V (f ), which would prove the lemma. Indeed,
let a ∈ V (p) r V (f ). Then for all g ∈ p, and all k ∈ N we have fgk (a) = 0, as g(a) = 0,
and f (a) 6= 0. As pf consists of such elements it follows that a ∈ V (pf ). On the other
hand, if a ∈ V (pf ), then fgk (a) = 0 for all g ∈ p and k ∈ N. Hence g(a) = 0, and therefore
a ∈ V (p). This proves the lemma.
Definition 2.3.11. Consider irreducible affine varieties, and ϕ : V −→ W a regular
map. Then ϕ is called dominant if ϕ(V ) contains a nonempty Zariski open subset U . So
∅ 6= U ⊂ ϕ(V ).
Theorem 2.3.12. Let V,W be irreducible affine varieties, and ϕ : V −→ W be a regular
map. Then the following conditions are equivalent:
(1) ϕ∗ : K[W ] −→ K[V ] is injective.

(2) ϕ is dominant.
(3) ϕ(V ) = W .
Proof.
The implication (2) =⇒ (3) is easy, see Exercise 2.2.36.
(3) =⇒ (1). Take f,g ∈ K[W ]. Then we have the following two equivalences.
f = g ⇐⇒ f|ϕ(V ) = g|ϕ(V ) ⇐⇒ f ◦ ϕ = g ◦ ϕ.
The second equivalence is clear. For the first, the only not obvious implication is ⇐=. As
regular functions are continuous, it follows that (f − g)−1 (0) is closed. By assumption,
this set contains ϕ(V ), and therefore contains ϕ(V ) = W . Hence f = g on the whole of
W.
The implication (1) =⇒ (2) is the “difficult” case. We have to prove that W r V (f ) ⊂
ϕ(V ) for some 0 6= f ∈ K[W ]. The idea is to find a nonzero f ∈ K[W ] ⊂ K[V ] and an
r such that there exists an inclusion K[W ]f [x1 , . . . ,xr ] ⊂ K[V ]f , such that K[V ]f is a
finitely generated K[W ]f [x1 , . . . ,xr ]-module. Indeed, this inclusion gives

• a finite map, hence surjective map ϕ e : V \ V (f ) −→ W r V (f ) × K r ,

• a projection map π : W rV (f ) ×K r −→ W rV (f ), which is obviously surjective,
e Hence, the image of ϕ would contain W r V (f ), which is a nonempty

such that ϕ = π ◦ ϕ.
Zariski open subset of W .
Now we will find the f and r. For simplicity we write R = K[W ], and S = K[V ].
By assumption we have an inclusion R ⊂ S. Consider the following diagram:
R ֒→ S
↓ ↓
E = Q(R) ֒→ SRr{0} ⊂ Q(S).
Obviously, SRr{0} is a finitely generated E–algebra, and E is a field. In this situa-

tion we can apply the Noether Normalization Theorem, see Theorem 2.2.9. There exist
x1 , . . . ,xr ∈ SRr{0} such that
E[x1 , . . . ,xr ] ⊂ SRr{0} ,
and SRr{0} is finitely generated as E[x1 , . . . ,xr ]–module. By clearing denominators,

we can achieve that the xi are elements of S, and SRr{0} is still a finitely generated
E[x1 , . . . ,xr ]–module. Now consider
R[x1 , . . . ,xr ] ⊂ S.
This need not be a finitely generated extension. But we can do the following. Take gen-
erators α1 , . . . ,αs of SRr{0} as E[x1 , . . . ,xr ]–algebra. Again we may assume that in fact
αi in S. The elements αi satisfy an integral equation with coefficients in E[x1 , . . . ,xr ] =
Q(R)[x1 , . . . ,xr ]. Let f be the product of all the denominators that appear as coefficients
in all such equations. Therefore, over Rf [x1 , . . . ,xr ] every αi is integral. We therefore
forced that the extension
Rf [x1 , . . . ,xr ] ⊂ Sf
is integral. This was our goal.
Theorem 2.3.13. Consider an ideal I ⊂ K[x1 , . . . ,xn ,y1 , . . . ,ym ], and let V = V (I) ⊂
K n+m . Let p : K n+m −→ K n be the projection on the first n factors. Then
p(V ) = V (I ∩ K[x1 , . . . ,xn ]).
The closure is taken here in the Zariski topology.

Proof. To show the inclusion p(V ) ⊂ V (I ∩ K[x1 , . . . ,xn ]), it suffices to show p(V ) ⊂
V (I ∩ K[x1 , . . . ,xn ]), as the right hand side is closed. So let a ∈ p(V ). Hence there exists
a b ∈ K m with (a,b) ∈ V . Let f ∈ I ∩ K[x1 , . . . ,xn ] then f (a,b) = 0. But f does not
depend on the variables y1 , . . . ,ym at all, and therefore f (a) = 0. This holds for all
f ∈ I ∩ K[x1 , . . . ,xn ] so that the inclusion ⊂ follows.
For the converse inclusion consider
K[x1 , . . . ,xn ]/(I ∩ K[x1 , . . . ,xn ]) ⊂ K[x1 , . . . ,xn ,y1 , . . . ,ym ]/I,
so that p(V ) is indeed given by the zero set of the ideal I ∩ K[x1 , . . . ,xn ] by 2.3.12.
Remark 2.3.14. Let ϕ = (ϕ1 , . . . ,ϕm ) : V −→ W ⊂ K m be a regular map, V = V (I).
Consider the graph of ϕ:
Graph(ϕ) = {(x,ϕ(x)) : x ∈ V } ⊂ V × W ⊂ V × K m .
Then the zero set of
(y1 − ϕ1 , . . . ,ym − ϕm ) ⊂ K[V ] ⊗ K[W ]
is the graph of ϕ. The graph of ϕ is canonically isomorphic to V :

in the ideal (y1 − ϕ1 , . . . ,ym − ϕm ) one eliminates all y–variables, so that K[V ] ⊗
K[W ] /(y1 − ϕ1 , . . . ,ym − ϕm ) is isomorphic to K[V ]. Note that with this identifica-
tion the map of K–algebras

K[W ] −→ K[V ] ⊗ K[W ] /(y1 − ϕ1 , . . . ,ym − ϕm )
gives the map Graph(ϕ) −→ W , which maps (x,ϕ(x)) to ϕ(x). We come to the conclusion
that every map ϕ : V −→ W can be viewed as the restriction of a projection mapping.
This is a point of view which is sometimes useful. For example its local version, see 3.4.43
will be used in the proof of the Finite Mapping Theorem, see 6.3.5.
Example 2.3.15. Consider the regular map ϕ : K −→ K 2 given by ϕ(t) = (t2 ,t3 ). Let
the coordinates of K 2 be x,y. Then the graph of ϕ lies in K 3 and is the zero set of the
ideal I = (x − t2 ,y − t3 ) So the closure of the image of f is the zero set of the ideal
I ′ = (x − t2 ,y − t3 ) ∩ K[x,y] by 2.3.14 and 2.3.13. One shows that I ′ = (y 2 − x3 ). The
proof of this is contained in Example 1.5.6.
Finally, we give a counterexample to the Going-Down Theorem for nonnormal rings.
Example 2.3.16. Let K be an algebraically closed field. Consider the map
ϕ : K2 −→ K 3
(t,s) 7−→ (x,y,z) = (t2 − 1,t3 − t,s).
p p′
X′
X
Then one shows that the image of ϕ is exactly given by y 2 − x2 (x + 1) = 0. Moreover,

via
ϕ∗ , the module S := K[t,s] is a finitely generated R := K[x,y,z]/ y 2 −x2 (x+1) –module.
Indeed, from the definition of the map it follows that t2 − 1 − x = 0, hence t is integral.
Notice, moreover, that K[x,y,z]/ y 2 − x2 (x + 1) is not a normal ring; the element t = xy
is an integral element.
We now show that the Going-Down Theorem 1.5.26 does not hold for the ring
extension R ⊂ S. Consider the prime ideal P := (s − t) in S (in the picture X = V (P) ⊂
K 2 ), and the prime ideal p := P ∩ R in R (in the picture X ′ = V (p) ⊂ K 3 ). One easily
calculates that p = (x − z 2 + 1,y − xz). Now we claim that P is the unique prime ideal in
S which lies over p. This should be clear geometrically, but let us give a precise algebraic
argument. Indeed, such a prime ideal P contains at least t2 − 1 − s2 + 1 = t2 − s2 ,
corresponding to x − z 2 + 1, the first generator of P. Hence, either t − s ∈ P, or t + s ∈ P.
If t − s ∈ P, we are done. Therefore, suppose t + s ∈ P. As y − xz ∈ p, it follows that
t3 − t − s(t2 − 1) = 2t3 − 2t + (1 − t2 )(s + t) ∈ P. Hence P is either (t − 1,s + 1) or (t,s)
or (t + 1,s − 1). This cannot be the case because these are all maximal and P is not, see
Exercise 2.2.37.
Now consider the maximal ideal q = (x,y,z − 1) of R, corresponding to the point
p′ = (0,0,1) in the picture. There are two points in the preimage of p′ = (0,0,1) namely
(1,1) and p = (−1,1). In particular the maximal ideal Q : (t + 1,s − 1) corresponding to
p is such that Q ∩ R = q. However, P 6⊂ Q, as otherwise p ∈ X. So we conclude that the
statement of the Going-Down Theorem does not hold in this case.
Exercises
2.3.17.
Prove Theorem 2.3.9.
(Hint: Either use Theorem 2.3.5 to describe the fibers of y, and Exercise 2.2.37, or use the Graph
Construction 2.3.14 and the Projection Theorem 2.2.8.)
2.3.18.
(1) Consider the map f : K 2 −→ K 4 given by f (s,t) = (s,st,t2 ,t3 ) = (x,y,z,u). Show that the
ideal of the image is equal to I ′ = (y 2 − zx2 ,yz − ux,z 2 x − uy,u2 − z 3 ).
(Hint: Show that any g is modulo I ′ equivalent to ay + bu + c with a,c ∈ K[x,z] and
b ∈ K[x].)
(2) Consider the map f : K −→ K 3 , f (t) = (t3 ,t4 ,t5 ) = (x,y,z). Show that the image of f is
given by the zero set of the ideal (y 2 − xz,x3 − yz,z 2 − x2 y).
(3) Use a computer algebra system, for example [Singular 2000] to check the results in (1)
and (2).
73
3 Basics of Analytic Geometry
In this chapter we start studying local analytic geometry, that is, the zero sets of analytic
functions in a (small) neighborhood of a point. To see why we want to do so, we look
at the affine hypersurface in C 2 defined by f (x,y) = y 2 − x2 (x + 1) = 0. In a small
neighborhood U of (0,0) we see two “parts”, or components of f (x,y) = 0.
So we would like to say that the “singularity” consists of two components. But how small
should we take U ? The answer to this question is: as small as we need to. To make a
precise notion, we define the notion of a germ of a space. A germ of a topological space
X at a point p by definition is an equivalence class of open neighborhoods of p. But f ,
considered as element of C [x,y] is irreducible (Exercise 1.4.34), but reducible if considered
as an element of C {x,y}. Here C {x,y} is the ring of power series which converge in some
neighborhood of 0.1 Indeed, we have the following factorization in C {x,y}
√ √
f (x,y) = y − x x + 1 y + x x + 1 .
√
So we see that in a neighborhood on which √ x + 1 converges, the zero set √ f = 0 decom-
poses into two parts, namely {(x,y) : y = x x + 1} and {(x,y) : y = −x x + 1}.
As we assume that the reader heard a course on holomorphic functions in one vari-
able, we will, in Section 3.1, treat only the basic facts on holomorphic functions in several
variables. Many of the statements and the proofs are similar or direct generalizations of
the one variable case.
For polynomials f ∈ C [x1 , . . . ,xn ] one of the main invariants is the degree. It can
be interpreted (in case f is reduced) as the number of intersection points of the set {p ∈
C n : f (p) = 0} with a general line. By applying the Noether Normalization Theorem, one
may assume that this line is given by x1 = a1 , . . . ,xn−1 = an−1 . In case of convergent
power series f , its replacement is the multiplicity or order of the power series, and is
defined as the lowest order term of f . Assuming that f converges on the (small enough)
neighborhood U of 0, the multiplicity can be interpreted as the maximal number of
intersection points of f = 0 with a general line L in U , in case f is reduced. This
1 The neighborhood might depend on the power series.
74 3 Basics of Analytic Geometry
statement follows from the so-called continuity of roots. So for example the element
f (x,y) = y 2 − x2 (x + 1) has degree three considered as element of C [x,y], and has
multiplicity two considered as element of C {x,y}. There is a substitute for the Noether
normalization for the case f ∈ C {x1 , . . . ,xn }, so that we can always assume that the line
L from above points in the xn direction. In terms of the power series f of multiplicity b,
this is equivalent to saying that the term xbn occurs with nonzero coefficient in the power
series expansion of f . In this case one says that f is regular of order b in xn .
The replacement for division with remainder, which played an important role in
the proof of Study’s Lemma, or more generally the Nullstellensatz, is the Weierstraß
Division Theorem. The Weierstraß Division Theorem says the following: Let a convergent
power series f ∈ C {x1 , . . . ,xn } be regular of order b in xn , and a further power series
g ∈ C {x1 , . . . ,xn } be given. Then there exist uniquely defined q ∈ C {x1 , . . . ,xn } and
r ∈ C {x1 , . . . ,xn−1 }[xn ], where r has, considered as polynomial in xn , degree less than
b, such that
g = q · f + r.
The similarity with ordinary division with remainder is obvious. It follows from the
Weierstraß Division Theorem, that the C {x1 , . . . ,xn−1 }–module C {x1 , . . . ,xn }/(f ) is
finitely generated.2 As the ring C {x1 , . . . ,xn−1 } is local, it even follows by an application
of Nakayama’s Lemma, that the Weierstraß Division Theorem follows from the fact that
C {x1 , . . . ,xn }/(f ) is a finitely generated C {x1 , . . . ,xn−1 }–module. In Section 3.2 we will
prove a weak generalization of this seemingly weaker statement. If one looks closely at
the proof, we will “only” show the existence of q and r where r has degree less than or
equal to b. It turns out that in order to show the convergence of q and r, it is much easier
to prove this last statement.
In Section 3.3, we give applications of the Weierstraß Theorems. To mention the
most important ones, we show the Implicit and Inverse Function Theorems and New-
ton’s Lemma (a “cooked-up” version of the Implicit Function Theorem). Furthermore, we
show that the (formal) power series ring is Noetherian and factorial, prove the Noether
Normalization Theorem, and Hensel’s Lemma.
With this knowledge at hand we can start the study of germs of analytic spaces in
Section 3.4. The main theorem, and the most difficult in this section is the Nullstellensatz.
In the case of germs of analytic spaces, this theorem cannot be interpreted as a statement
on maximal ideals. We will prove the analogous statement to Corollary 2.2.10. That is,
let a germ of an analytic space (X,x) be given for which the ideal of (X,x) is prime, and
let π : (X,x) −→ (C k ,y) be a Noether normalization. Then π is surjective. As in the
affine case, the Nullstellensatz follows from this surjectivity. An essential ingredient in
the proof of the surjectivity of π is the Theorem of Finiteness of Normalization 1.5.19.
It will be used to describe the properties of a general projection of an irreducible germ
(X,x) onto a hypersurface singularity. The precise properties of this projection of X
onto a hypersurface are gathered in the Local Parametrization Theorem. This is a very
important theorem, as it, in many cases, allows the reduction of general statements
on germs of analytic spaces to the case of hypersurfaces, which are sometimes easier
to handle. The Local Parametrization Theorem will be used several times in the next
chapter.
2 This module is even free.
3.1 Holomorphic Functions of Several Complex Variables 75
3.1 Holomorphic Functions of Several Complex Variables
Before going on with geometry, we need some basic facts on holomorphic functions of
several complex variables. The facts we mention here are either proved, or the proof can
be easily generalized from the case of a function in one variable. In this case, no proof is
given.
Definition 3.1.1. Let n ∈ N, and νi , for i = 1, . . . ,n be nonnegative integers, x =
(x1 , . . . ,xn ) ∈ C n . We introduce the multi index notation
n
Y n
X
ν := (ν1 , . . . ,νn ); xν := xνi i ; |ν| := νi .
i=1 i=1
The number |ν| is called the degree of ν. Let r = (r1 , . . . ,rn ) ∈ Rn+ , ri > 0 for i = 1 . . . ,n.
The set
P r := {x ∈ C n : |xi | ≤ ri for i = 1, . . . ,n}
is called the closed polydisc (or polycylinder) with center {0} and polyradius r. Similarly
one defines open polydiscs Pr , and polydiscs with center p ∈ C n .
Definition 3.1.2. Let U ⊂ C n be a domain3 , f : U −→ C a function, p ∈ U . Then f is
called complex differentiable in p, if there exists an open neighborhood V ⊂ U of p and
functions ∆1 , . . . ,∆n : V −→ C which are continuous in p such that for all x ∈ V
n
X
f (x) = f (p) + (xi − pi )∆i (x).
i=1
f is called holomorphic in U if f is complex differentiable in p for all p ∈ U .

In this case it follows without much difficulty, by reducing it to the one-dimensional
∂f
case, that the values ∆i (p) are uniquely determined. Those values we denote by ∂x i
(p).
Furthermore, sums, products, and quotients (whenever defined) of holomorphic functions
are holomorphic. See Exercise 3.1.18.
It is known that one-variable functions are holomorphic if and only if they are locally
represented by their Taylor series. To have the analogous statement for the multi-variable
case, we need some definitions.
Definition 3.1.3.
(1) Let aν for ν ∈ Nn be complex numbers. The expression4

∞
X
aν (x − p0 )ν
ν=0
is called a formal power series in the variables x1 , . . . ,xn around p0 . With the
obvious addition and multiplication the set of formal power series is a ring, and
even an C –algebra. The ring of formal power series around p0 = 0 is denoted by
C [[x1 , . . . ,xn ]] or sometimes C [[x]] for short, and is called the formal power series
ring.
3 This means that U is an open and P connected subset.
4 This is an alternative notation for ν∈Nn aν (p − p0 )ν .
P∞
(2) Let p ∈ C n , and c ∈ C . One says that the formal power series ν=0 aν (x − p0 )ν
converges in p with limit c if for all ε > 0 there exists a finite set Iε ⊂ Nn such that
for all finite sets I with Iε ⊂ I ⊂ Nn

X

aν (p − p0 )ν − c < ε.

ν∈I
P ν
Note that P ν∈I aν (p − p0 ) makes sense because the summation is over a finite set.
∞
In case that ν=0 aν (x − p0 )ν converges in p the number c is uniquely determined,
and is called the value of the formal power series at p. We, therefore, simply write
P ∞ ν
ν=0 aν (p − p0 ) = c.
P∞ ν
P∞ formal powerν series ν=0 aν (x − p0 ) is called absolutely convergent in p if
The
ν=0 | aν (x − p0 ) | is convergent. An absolutely convergent power series is in fact
convergent.
Let U ⊂ C n , and f : U −→ C be a function. One says that the formal power series
(3) P
∞
aν (x − p0 )ν converges on U with limit f if for all p in U the formal power
ν=0 P
∞
series ν=0 aν (x − p0 )ν converges in p with limit f (p).
n
P∞U ⊂ C , andν f : U −→ C be a function. One says that the formal power
(4) Let again
series ν=0 aν (x − p0 ) converges uniformly to f on U if for all ε > 0 there exists
a finite set Iε ⊂ Nn such that for all finite sets I with Iε ⊂ I ⊂ Nn and for all p ∈ U

X
ν
aν (p − p0 ) − f (p) < ε.

ν∈I
The difference to ordinary convergence is that in case of ordinary convergence, the

set Iε might depend on p, whereas in case of uniform convergence, the set Iε is not
allowed to depend on p.
(5) Let U ⊂ C n be a domain, f : U −→ C be a function. f is called analytic if for
all p0 ∈ U , there exists a neighborhood V = V (p0 ) of p0 in U , and a power series
P ∞ ν
ν=0 aν (x − p0 ) which converges on V to f .
Let p ∈ C n be fixed. It will turn out that the formal power series that converge on
a neighborhood of p (the neighborhood might depend on the function) form a ring,
called the convergent power series ring On,p at p. The ring of convergent power
series at 0 is denoted by C {x} or C {x1 , . . . ,xn } or also often by On . instead of the
“correct” On,0 .
(6) Let 0 6= f ∈ C [[x1 , . . . ,xn ]] be a formal power series. We can write
f = fm + fm+1 + . . .
where fk are homogeneous polynomials of degree k in x1 , . . . ,xn , and fm 6= 0. Then

we define mult(f ) := ord(f ) := m and call it the order or multiplicity of f . One
obviously has that ord(f · g) = ord(f ) + ord(g).
(7) More generally, let λ = (λ1 , . . . ,λn ) ∈ Nn+ and let 0 6= f ∈ C [[x1 , . . . ,xn ]] be a
formal power series. We can write
f = fm + fm+1 + · · ·
where fk are quasi-homogeneous polynomials (with respect to λ) of weighted degree

k and fm 6= 0. Then we define w-ord(f ) := m and call it the weighted order of f .
The first thing we want to show is that analytic functions are holomorphic. In order
to show this, we need the following lemma.
P∞
Lemma 3.1.4. Consider a formal power series ν=0 aν xν , which converges at p ∈ C n .
Let p = (p1 , . . . ,pn ) and suppose that pi 6=P0∞for all i. Let the polyradius r be defined by
ri = |pi |. Then the formal power series ν=0 aν xν converges absolutely and uniformly
on all compact subsets of the open poly-disc Pr .
P
Proof. From the convergence of the series aν pν it follows that the set {aν pν } is
bounded, that is, there exists an M ∈ R with |aν pν | = |aν |rν < M . Consider a real
number q with 0 < q < 1. By definition of Pqr for all x ∈ Pqr it follows that |xν | < q |ν| rν .
Hence for all x ∈ Pqr
|aν xν | < |aν |q |ν| rν < M · q |ν| .
But n
X 1
|ν|
M q =M ,
ν
1−q
see Exercise 3.1.19. The set Nn is countable, hence there exists
n a bijection α : N −→
n α(k)
P∞ 1
N . Put bk (x) := aα(k) x . Hence, k=0 |bk (x)| ≤ M 1−q , so that this series is
P∞
increasing and bounded. It follows that k=0 bk (x) converges absolutely and uniformly
on Pqr so that
∞
X
∀ε > 0 there exists nε ∈ N with |bk (x)| < ε for all x ∈ Pqr .
k=nε
Set Iε := {α(i) : i = 0, . . . ,nε } and choose I with I ⊃ Iε . Then

∞ ∞

X X X X
ν
ν
|aν x | − |aν x | = |bk (x)| − |bk (x)|

ν=0 ν∈I k=0 k∈α−1 (I)
X ∞
X
= |bk (x)| ≤ |bk (x)| < ε.
/ −1 (I)
k∈α k=nε +1
This shows that the series converges absolutely and uniformly in Pqr . We still have to
show that the series converges on all compact subsets K in Pr . But Pqr for 0 < q < 1
is an open cover of K. As K is compact we have a finite subcover Pq1 r , . . . Pqs r . Let
q = max{q1 , . . . ,qs }. Hence K ⊂ Pqr , and thus the series converges uniformly on K.
Remarks 3.1.5.
(1) Let f be analytic on a neighborhood of 0. As f is analytic, it follows from the proof

of 3.1.4 that there exists a δ > 0 such that
X
kf kδ := |fα |δ |α| < ∞.
α
The fact that f is analytic is equivalent to this property. We could rephrase P this
as follows. For all δ > 0 we put a seminorm5 on C [[x1 , . . . ,xn ]]. Let f = fα xα ∈
C [[x1 , . . . ,xn ]] be a formal power series. Then we put for δ > 0
X
kf kδ := |fα |δ |α| ∈ R+ ∪ {∞}.
α
It follows from the proof of Lemma 3.1.4 that f is convergent exactly when there
exists a δ > 0 with kf kδ < ∞.
(2) Note that if kf kδ < ∞, and 0 < t < 1, then kf ktδ < ∞. If, moreover, f ∈
(x1 , . . . ,xn ) an easy estimate shows that kf ktδ ≤ t · kf kδ .
P∞ k
(3) Let f = k=0 fk xn be a convergent power series, with fk ∈ C [[x1 , . . . ,xn−1 ]].
From the absolute convergence of f in some domain in C n it follows that fi are
absolute convergent in some domain of C n−1 , see Exercise 3.1.23. It follows that
fk ∈ C {x1 , . . . ,xn−1 }.
Theorem 3.1.6. Let U ⊂ C n be open, f : U −→ C be analytic. Then f is holomorphic.
Proof.
P Without loss of generality, it suffices to show that f is holomorphic in 0. Let
ν
ν a ν x be a power series that converges on an open neighborhood V of 0 to f . We
write
X X
aν xν = a0...0 +x1 aν x1ν1 −1 xν22 · · · xνnn
ν ν1 6=0
X
+x2 aν x2ν2 −1 · · · xνnn
ν,ν2 6=0
ν1 =0
..
.
X
+xn aν xνnn −1
ν,νn 6=0
ν1 =···=νn−1 =0
= a0...0 +x1 ∆1 + . . . + xn ∆n .
P
Take a point p = (p1 , . . . ,pn ) for which ν aν pν is absolutely convergent. Such a point
exists by Lemma 3.1.4. We P may assume that all the pi are nonzero. It follows from
the absolute convergence of ν aν pν that each of the sub series pi ∆i (p) are absolute
convergent. As pi 6= 0, the series ∆i (p) are convergent. Applying Lemma 3.1.4, it follows
that the ∆i for i = 1, . . . ,n converge absolutely and uniformly on small neighborhoods of
0. From a standard theorem in analysis it then follows that the ∆i are continuous.
Theorem 3.1.7 (Osgood’s Lemma). Let U be an open set of C n , and f : U −→ C be
a continuous function. The following conditions are equivalent.
(1) f is analytic.
(2) f is holomorphic.
5 A seminorm is a map C [[x1 , . . . ,xn ]] → R ∪ {∞} satisfying all conditions of a norm (with the obvious
interpretation a + ∞ = ∞).
(3) f is holomorphic in each variable.
Proof. We already showed that (1) implies (2) in Theorem 3.1.6. As (2) implies (3)
is trivial, we only have to show (3) implies (1). Take a point p ∈ U . Without loss of
generality, we may assume that p = 0. Let P r be a closed poly-disc contained in U for
r = (r1 , . . . ,rn ). Take a point (ζ1 , . . . ,ζn−1 ) with |ζi | ≤ ri . The function
fn (xn ) := f (ζ1 , . . . ,ζn−1 ,xn )
by assumption is holomorphic in xn . By the Cauchy integral formula in one variable we

therefore have for all xn with |xn | < rn
Z
1 f (ζ1 , . . . ,ζn−1 ,ζn )
f (ζ1 , . . . ,ζn−1 ,xn ) = dζn .
2πi ζn − xn
|ζn |=rn
We can do the same trick with the penultimate variable,

Z
1 f (ζ1 , . . . ,ζn−1 ,xn )
f (ζ1 , . . . ,ζn−2 ,xn−1 ,xn ) = dζn−1
 2πi ζn−1 − x n−1
2 Z Z |ζn−1 |=rn−1
1  f (ζ1 , . . . ,ζn−1 ,ζn )  dζn−1
=  dζn 
2πi ζn − xn ζn−1 − xn−1
|ζn−1 |=rn−1 |ζn |=rn
Continuing like this we obtain
f (x1 , . . . ,xn ) =
     
n Z Z Z
1 1  1   f (ζ)   
 · · ·  dζn  dζn−1  · · ·  dζ1 .
2πi ζ1 − x1 ζ2 − x2 ζn − xn
|ζ1 |=r1 |ζ2 |=r2 |ζn |=rn
For fixed (x1 , . . . ,xn ) the integrand is continuous on T := {(ζ1 , . . . ,ζn ) : |ζi | = ri for
i = 1, . . . ,n}, which is a compact set. By Fubini
n Z
1 f (ζ)
f (x1 , . . . ,xn ) = dζ1 · · · dζn .
2πi (ζ1 − x1 ) · · · (ζn − xn )
T
It follows from Exercise 3.1.19 that for fixed (x1 , . . . ,xn ) the series
∞
X xν1 · · · xνn
1 1 n
=
(ζ1 − x1 ) · · · (ζn − xn ) ν=0 ζ1ν1 +1 · · · ζnνn +1
converges absolutely and uniformly on T . As f is continuous, it is bounded on T . There-

fore, for fixed (x1 , . . . ,xn ) the function
∞ ν νn
f (ζ1 , . . . ,ζn ) f (ζ1 , . . . ,ζn ) X x1 1 xn
= · ···
(ζ1 − x1 ) · · · (ζn − xn ) ζ1 . . . ζn ν=0
ζ1 ζn
converges absolutely and uniformly on T . It is a standard theorem from analysis, that

one may in such a case interchange summation and integration. We get
n Z
1 f (ζ)
f (x1 , . . . ,xn ) = dζ1 · · · dζn
2πi (ζ1 − x1 ) · · · (ζn − xn )
T
∞
X n Z
1 f (ζ)
= xν11 · · · xνnn · dζ1 · · · dζn
ν=0
2πi ζ1ν1 +1· · · ζnνn +1
T
∞
X
= aν xν , with
ν=0
n Z
1 f (ζ)
aν = dζ1 · · · dζn .
2πi ζ1ν1 +1 · · · ζnνn +1
T
Remarks 3.1.8.
(1) It follows that the convergent power series is a ring. Indeed, if f,g ∈ C {x1 , . . . ,xn },
then choose an open neighborhood U of 0 on which f converges, and an open
neighborhood V of 0 on which g converges. Then f and g are holomorphic on
U ∩ V , so that f ± g and f · g are holomorphic on V , so define an analytic function.
This fact could also have been proved directly, by using only Lemma 3.1.4. Note
moreover that C {x1 , . . . ,xn } is a local ring with maximal ideal (x1 , . . . ,xn ). Indeed,
for convergent power series f not in (x1 , . . . ,xn ) one has f (0) 6= 0. It follows that
1
f is also holomorphic in a neighborhood of 0, and therefore, by Osgood’s Lemma,
again is a convergent power series. So the elements of the complement of (x1 , . . . ,xn )
are units, and it follows from 1.3.3 that the power series ring is local.
(2) Let U ⊂ C n be open. A map F = (f1 , . . . ,fm ) : U −→ C n is called holomorphic
if each of the fi is holomorphic. If F (U ) ⊂ W ⊂ C m , and G : W −→ C p is
a holomorphic map, then the chain rule, see Exercise 3.1.22 gives that G ◦ F is
holomorphic. This in particular gives the following. Let g1 , . . . ,gn ∈ C {y1 , . . . ,ym },
with gi (0) = 0. Let f ∈ C {x1 , . . . ,xn }. Then f (g1 , . . . ,gn ) ∈ C {y1 , . . . ,ym }.
Theorem 3.1.9 (Identity Theorem). Let U ⊂ C n be a domain. Let f : U −→ C be a
holomorphic function, such that the restriction of f to a nonempty open subset V ⊂ U
is the zero map. Then f is identically zero on U .
Remark 3.1.10. In the case n = 1 we have a stronger version of the Identity Theorem,
as follows.
Let U ⊂ C be a domain and f : U −→ C be a holomorphic function. Let {xk } be
a convergent sequence, xk ∈ U and limk→∞ xk ∈ U . If f (xk ) = 0 for all k, then f is
identically zero on U .
Indeed, suppose lim xk = 0, and f not identically zero in a neighborhood of 0. Then
f = xb · u, where u(0) 6= 0. Therefore, for p with |p| small, it follows that u(p) 6= 0, and
therefore, f (p) 6= 0. The zeros of f therefore lie isolated. This is a contradiction, and
therefore, f vanishes in a neighborhood of 0.
One can reduce the proof of the Identity Theorem to the Identity Theorem in the
one variable case. The idea is the same as in the proof of the maximum principle, which
comes next. The proof of the Identity Theorem 3.1.9 is therefore left as Exercise 3.1.24.
Theorem 3.1.11 (Maximum Principle). Consider a domain subset U ⊂ C n . Let f :

U −→ C be a nonconstant holomorphic function. Then |f | does not have a maximum on
U.
In other words, if f : U −→ C takes its maximum absolute value at a point p in U ,
then f is constant.
Proof. In case of a holomorphic function in one variable, this is well-known. We will
prove the general statement by using the one-variable case and the Identity Theorem.
Let p ∈ U be a point on which |f | attains its maximum. Consider a complex line L
through p, and consider a small open ball B around p which lies inside U .
L
U
B
Then the restriction of f to B ∩ L is a holomorphic function of one variable. B ∩ L

is a disc in L. By the maximum principle for the one variable case, the restriction of f
to B ∩ L is constant. By varying the lines through p, we conclude that f is constant on
B. By the Identity Theorem, f is constant on U .
Definition 3.1.12.
(1) A set or space X ⊂ C n is called locally analytic, if for any point p ∈ X, there exists
an open subset V of p in C n , and finitely many holomorphic functions f1 , . . . ,fs
defined on V such that
X ∩ V = {x ∈ V : f1 (x) = . . . = fs (x) = 0}.
(2) Let U be an open subset of C n . A subset X ⊂ U is called an analytic subset of U ,

if X is locally analytic, and closed in U .
(3) Let X be an analytic subset of U , U ⊂ C n open. A function f : X −→ C is called
holomorphic if for all x ∈ X there exists an open neighborhood V of x in C n such
that f|V ∩X is the restriction of a holomorphic function on V .
Remarks 3.1.13.
(1) In this definition, we have to be more careful than in the affine case. Polynomials
are always globally defined. But holomorphic functions are usually only defined
in small open neighborhoods, so this is the reason why we have to choose open
neighborhoods V .
(2) The set X := {(x,y) : y = 0, Im(x) ≥ 0} in C 2
is not locally analytic. Indeed, let V be a connected open neighborhood of {0}, and
f be a function on V with
V ∩ X ⊂ {(x,y) : f (x,y) = 0.}
Then we can restrict f to the set V ∩ {y = 0}. By the Identity Theorem in one
variable, f must vanish identically on {y = 0} ∩ V . As this holds for all f , it follows
that X is not locally analytic.
(3) In the definition of X to be an analytic subset of U , we need the condition X is
closed in U to avoid phenomena as shown in the following picture.
Morally speaking, one wants an analytic subset of U , except for discrete points, to
“reach” the boundary. Even more severely, without the closedness condition, if X
is the shaded part of
X
U
then X would be analytic.

(4) Consider a sequence p1 ,p2 , . . . of nonzero points in C that converge to the origin.
Then the union X of the points is not an analytic set. Indeed, any holomorphic
function vanishing on U ∩X, for any open neighborhood U of 0 is the zero function.
This follows from the Identity Theorem in one variable.
It is well-known (see Remark 3.1.10) that the zeros of a holomorphic function in one
variable lie isolated. The following is a generalization of this fact.
Theorem 3.1.14. Let X ⊂ U be an analytic subset, with U connected. Suppose X 6= U .
Then the closure6 of U \ X in U is U .
Proof. The main idea of the proof is already contained in the proof of the fact that the
sets in 3.1.13, (2) and (3) are not analytic. The claim is that X has no inner points of
◦
U , that is, we claim that the set X of inner points of X is empty. Suppose the converse,
◦ ◦
that is, X contains a nonempty open subset X of U . We claim that X is closed. From
6 In the topological sense.
◦
this it would follow from the connectedness of U that X is equal to U , in contradiction
◦ ◦
to the assumption. So let us prove X is closed. Take a point p in the boundary of X
in U . As X is closed in U , the point p is in X. Hence, there exists an open, connected
neighborhood V of p and holomorphic functions f1 , . . . ,fs defined on V such that
X ∩ V = {x ∈ V : f1 (x) = . . . = fs (x) = 0}.
◦ ◦
As p is in the boundary of X, and p is in the open set V , it follows that V ∩ X 6= ∅.
V
p U
o
X
Therefore, for all i = 1, . . . ,s the function fi is identically zero on the (nonempty) open
◦
subset V ∩ X, hence fi is, by the Identity Theorem, identically zero on V . Hence V ⊂ X,
◦ ◦
and p ∈ X. So we proved that X is closed.
Theorem 3.1.15 (First Riemann Extension Theorem). Let U be an open connected
subset of C n , and X ⊂ U an analytic subset. Consider
f : U \ X −→ C
holomorphic. Suppose f is locally bounded, that is, for all p ∈ X there exists an open
neighborhood V of p such that the restriction of f to V \ (V ∩ X) is bounded. Then there
exists a holomorphic extension of f to U . More precisely, there exists a holomorphic
function fe : U −→ C such that fe|U\X = f .
Proof. Let p ∈ X, and Bp an open ball around p. It suffices to extend f to a holomorphic
function fep on Bp . Namely, if Bp ∩ Bq 6= ∅, and if fep ,feq are extensions of f to Bp ,
respectively Bq , they must agree on Bp ∩ Bq , as on W := Bp ∩ Bq ∩ (U \ X) they coincide
with f , and W is not empty because the closure of U \ X in U is U , see Theorem 3.1.14.
So we can use the Identity Theorem 3.1.9 to deduce that fp and fq coincide on Bp ∩ Bq .
Consider a line L through p. Then L ∩ X is an analytic subset of X. As X contains
no inner points, there exists a line not contained in X. As L ∩ X is an analytic subset
of an open subset of L ∼= C , hence locally given by the zero-set of one analytic function,
the points of L ∩ X lie isolated. By an affine change of coordinates, we may assume that
the line L is the xn –axis, given by x1 = . . . = xn−1 = 0, and p is equal to 0. Therefore,
there exists an rn > 0 such that
X ∩ {x = (x1 , . . . ,xn ) : x1 = . . . = xn−1 = 0, |xn | ≤ rn }
consists of just one point, namely 0. As the set {x = (x1 , . . . ,xn ) : x1 = . . . = xn−1 =
0, |xn | = rn } is compact7 , and is contained in the open set U \ X, it follows without much
difficulty that there exists an open neighborhood W of 0 in C n−1 such that W × {|xn | =
rn } ⊂ U \ X. So
7 It is a circle.
(3.1) X ∩ (W × {|xn | = rn }) = ∅.
Lζ L
U
X
ζ
p |xn | ≤ rn
W
We consider a point ζ = (ζ1 , . . . ,ζn−1 ) ∈ W fixed. As X is given by the vanishing of
holomorphic functions, and the restriction of such functions to the line Lζ given by x1 =
ζ1 , . . . ,xn−1 = ζn−1 do not all vanish identically because of (3.1), it follows that the set
X∩{(ζ1 , . . . ,ζn−1 )}×{|xn | ≤ rn } has only isolated points. The function f (ζ1 , . . . ,ζn−1 ,xn )
is a holomorphic function in xn , not defined in the finitely many points of X, and it is
locally bounded. Therefore, by the Riemann Extension Theorem in one variable, it has
a holomorphic extension to {(ζ1 , . . . ,ζn−1 )} × {|xn | ≤ rn } and is, in fact, given by the
integral
Z
f (ζ)
(3.2) f (ζ1 , . . . ,ζn−1 ,xn ) = dζn .
ζn − xn
|ζn |=rn
This holds for all (ζ1 , . . . ,ζn−1 ) ∈ W . Now we define fe for all ζ1 , . . . ,ζn−1 ,xn ∈ W ×{|xn | ≤
rn } by the right hand side of equation (3.2). It follows that fe is holomorphic with respect
to each variable. Therefore, by Osgood’s Lemma 3.1.7, fe is holomorphic.
Theorem 3.1.16 (Second Riemann Extension Theorem). Let U be an open subset of
C n , with n ≥ 2. Let p ∈ U , and
f : U \ {p} −→ C
be a holomorphic function. Then f has a holomorphic extension to the whole of U .

Proof. Consider a small open ball B around p which is contained in U . We will show
that f is bounded on B \ {p}, so that the First Riemann Extension Theorem gives that
f can be extended to a holomorphic function on B. Consider q ∈ B \ {p}. As n ≥ 2, we
can find a complex line L through q which misses p.
L U
B
q p
The restriction f to B ∩ L is holomorphic in one variable. By the maximum principle
|f (q)| ≤ max |f|∂B∩L | ≤ max |f|∂B |.
As this holds for all q ∈ B \ {p}, it follows that f is bounded, as wanted.

1
Remark 3.1.17. For n = 1 the theorem is not true: consider the function x on U =
C \ {0}.
Exercises
3.1.18.
(1) Prove that the values ∆i (p) in the definition of a holomorphic function are well-determined,
by reducing it to a corresponding statement in the one variable case.
P ν
(2) Let f = ν aν x be a convergent power series. Describe all partial derivatives of f in
terms of the aν .
3.1.19. Let q = (q1 , . . . ,qn ). Prove that
X n
Y 1
qν = .
ν∈Nn k=1
1 − qk
3.1.20. Consider the formal power series ring C [[x1 , . . . ,xn ]] and let m = (x1 , . . . ,xn ).
(1) Show that m is a maximal ideal.
(2) Let f,g ∈ C [[x1 , . . . ,xn ]]. Show that f = g if and only if f ≡ g mod mk for all k ∈ N.
(3) Show that in particular ∩∞ k
k=0 m = 0.
P
(4) Let g ∈ m, and ak ∈ C . Show that ∞ k
k=0 ak g is a well-defined element in C [[x1 , . . . ,xn ]].
(5) More generally, let elements g1 , . . . ,gs in m be given. Show that there is a well-defined
homomorphism of C –algebras
C [[y1 , . . . ,ys ]] −→ C [[x1 , . . . ,xn ]]

yi 7→ gi .
(6) Let ϕ : C [[y1 , . . . ,ys ]] −→ C[[x1 , . . . ,xn ]] be a C –algebra homomorphism. Prove that for
all f we have ϕ(f ) = f (ϕ(y1 ), . . . ,ϕ(ys )).
(7) Show that C [[x1 , . . . ,xn ]] is a local ring.
1
(Hint: Let g ∈ m. What is 1−g ?)
3.1.21. Prove the following statements.

(1) Let f be holomorphic in p. Then f is continuous at p.
(2) Sums, products and quotients (when defined) of holomorphic functions are holomorphic.
3.1.22.
(1) Let f ∈ C {x1 , . . . ,xn }, and g1 , . . . ,gn ∈ C {y1 , . . . ,ym } with gi (0) = 0 for i = 1, . . . ,n.
Show that f (g1 , . . . ,gm ) ∈ C {y1 , . . . ,ym } by using the seminorm kf kδ , see 3.1.5.
(2) Use Exercise 3.1.20 to prove that for a C –algebra homomorphism ϕ : C {y1 , . . . ,ys } −→
C{x1 , . . . ,xn } for all f ϕ(f ) = f (ϕ(y1 ), . . . ,ϕ(ys )) holds.
P
3.1.23. Let f = fk xkn , fk ∈ C [[x1 , . . . ,xn−1 ]], and f ∈ C {x1 , . . . ,xn }. Prove that fk ∈
C {x1 , . . . ,xn−1 }.
(Hint: Use 3.1.4 to show this for k = 0. Now use induction on k.)
3.1.24. Prove the Identity Theorem 3.1.9.
3.1.25. Let U be an open connected subset of C n , and O(U ) the ring of holomorphic functions
on U . Prove that O(U ) is an integral domain.
3.1.26. Let f be a holomorphic function on an open subset in C n , with n ≥ 2. Then f has no

zeros which lie isolated.
(Hint: Look at f1 .)
3.1.27. Consider an ideal I ⊂ C [x1 , . . . ,xn ], and f ∈ C [x1 , . . . ,xn ]. Suppose that
f ∈ I · C {x1 , . . . ,xn }. Show that f ∈ I.
3.1.28. Let I = (y 2 − xz,x3 − yz,z 2 − x2 y). Show that I ∩ C {x} = (0).

(Hint: Let α(y 2 − xz) + β(x3 − yz) + γ(z 2 − x2 y) ∈ C {x}. Show that modulo (x) we can
write α = Az,β = Ay + Cz,γ = Cy. Take α′ = Az + Cx2 ,β ′ = Ay + Cz,γ ′ = Cx + Cy,
and subtract these from α,β and γ. Iterate this and show that even in the formal power series
α(y 2 − xz) + β(x3 − yz) + γ(z 2 − x2 y) = 0.)
3.2 Weierstraß Division and Preparation Theorem
We now come to the famous Weierstraß Division Theorem, and its immediate corollary,
the Weierstraß Preparation Theorem. Before stating it, we need some definitions.
Definition 3.2.1.
(1) An element f ∈ C {x1 , . . . ,xn } is called regular of order b in xn if the power series
in the variable xn , defined by f (0, . . . ,0,xn ) has a zero of order b.
(2) An element
xbn + a1 xnb−1 + . . . + ab−1 xn + ab , ai ∈ C {x1 , . . . ,xn }
is called a Weierstraß polynomial if ai (0) = 0 for all i = 1, . . . ,b. Equivalently, it is

a monic polynomial of degree b in xn which is, as convergent power series, regular
of order b in xn .
For general coordinate systems (x1 , . . . ,xn ) an f ∈ C {x1 , . . . ,xn } is regular in xn .
In fact, we have the following result, which is analogous to the Noether Normalization
Theorem for Hypersurfaces, see 2.1.7. As the proof is similar it is left as Exercise 3.2.15.
Lemma 3.2.2. Let f ∈ C {x1 , . . . ,xn } be of order b. Then after a general linear change
of coordinates, f is regular of order b in xn .
3.2 Weierstraß Division and Preparation Theorem 87
We can now state the Weierstraß Theorems.

Theorem 3.2.3 (Weierstraß Division Theorem). Let f,g ∈ C {x1 , . . . ,xn }, and suppose
that f is regular of order b in xn . Then there exist uniquely determined q ∈ C {x1 , . . . ,xn },
and r ∈ C {x1 , . . . ,xn−1 }[xn ], with r of degree smaller than b in xn such that
g = qf + r.
Theorem 3.2.4 (Weierstraß Preparation Theorem). Let f ∈ C {x1 , . . . ,xn } be regular of

order b in xn . Then there exists a unit u ∈ C {x1 , . . . ,xn }, and a Weierstraß polynomial
p of degree b, such that
f = up.
These u and p are uniquely determined by f .
Proof. (Assuming the statement of the Weierstraß Division Theorem.) Apply the Weier-
straß Division Theorem to g = xbn to obtain a q ∈ C {x1 , . . . ,xn } and a polynomial r in
xn of degree smaller than b such that.
xbn = qf + r.
Now q is a unit, as otherwise in the expression qf + r there does not appear the term xbn ,
which would be a contradiction to the equality xbn = qf + r. So now we can put u = q −1
and p = xbn − r and get the equality f = up. It remains to show that p is a Weierstraß
polynomial. But if p were not a Weierstraß polynomial, one of the nonleading coefficients
of r would be a unit, and f would be regular of order smaller than b in xn . The fact that
u and p are uniquely determined follows from the corresponding fact in the Weierstraß
Division Theorem, see Exercise 3.2.16.
Remark 3.2.5. The Weierstraß Division and Preparation Theorem also hold in the
formal power series ring C [[x1 , . . . ,xn ]]. A proof for this case is much easier, as one does
not have to cope with convergence questions. A proof will be given in Exercise 3.2.18.
Corollary 3.2.6. Suppose that f ∈ C {x1 , . . . ,xn } is regular of order b. Then the
C {x1 , . . . ,xn−1 }–module C {x1 , . . . ,xn }/(f ) is finitely generated and free of rank b. A
basis is given by 1,xn , . . . ,xb−1
n .
The proof of this corollary is similar to the proof of 2.1.7 and therefore left as Exercise
3.2.17. It is in fact not so difficult to see that the Corollary is equivalent to the Weierstraß
Division Theorem. We can even do something better by invoking Nakayama’s Lemma.
Lemma 3.2.7. Let f ∈ C {x1 , . . . ,xn } be regular of order b. If C {x1 , . . . ,xn }/(f ) is a
finitely generated C {x1 , . . . ,xn−1 }–module, then the Weierstraß Division Theorem holds
for f .
Proof. Put M := C {x1 , . . . ,xn }/(f ). The maximal ideal ofC {x1 , . . . ,xn−1 } is the ideal
(x1 , . . . ,xn−1 ). Now M/(x1 , . . . ,xn−1 )M = C {xn }/ f (0,xn ) , hence is a finite-dimensional
C –vector space of dimension b. Indeed, a basis of this vector space is given by the classes
of 1,xn , . . . ,xb−1
n . By Corollary 1.3.6 (a corollary of Nakayama’s Lemma), it follows that
1,xn , . . . ,xb−1
n generate M as C {x1 , . . . ,xn−1 }–module. This exactly means that for all
g ∈ C {x1 , . . . ,xn } there exist q and r such that
g = qf + r
where r is a polynomial in xn of degree less that b with coefficients in C {x1 , . . . ,xn−1 }.

To finish the proof, we need to show that q and r are uniquely determined. So assume
that g = qf + r = q ′ f + r′ , with the degree of both r and r′ smaller than b. Then
(q − q ′ )f + r − r′ = 0. It therefore suffices to show that from qf + r = 0 with degree of
r less than b, it follows that q = r = 0. For this it obviously suffices to show that q = 0.
We now write
X X b−1
X
f= fi xin ; q= qi xin ; r= ri xin ,
i=0
with fi , qi and ri in C {x1 , . . . ,xn−1 }. As f is regular of order b in xn , it follows that

ord(fb ) = 0. If q 6= 0, we can find a minimal k such that ord(qk ) is minimal. This means
that for j < k we have ord(qj ) > ord(qk ), and for j > k we have ord(qj ) ≥ ord(qk ). As f
is regular of order b it follows that ord(fi ) > 0 for i = 0, . . . ,b − 1. We now look at the
coefficient of xb+k
n of the power series q · f . As the term xb+k
n does not appear in r, and
qf + r = 0, this coefficient is equal to
(3.3) fb+k q0 + · · · + fb+1 qk−1 + fb qk + fb−1 qk+1 + · · · + f0 qb+k = 0.
We know the following

• ord(fb qk ) = ord(qk ),
• ord(fb+j qk−j ) ≥ ord(qk−j ) > ord(qk ) for j > 0,
• ord(fb−j qk+j ) = ord(fb−j ) + ord(qk+j ) > ord(qk ) for j > 0.
The only terms of degree ord(qk ) in formula (3.3) come from fb qk , so the expression
(3.3) cannot be equal to zero. This is a contradiction. Hence q = 0 and the uniqueness is
proved.
So all we need to show now, in order to prove the Weierstraß Division Theorem,
is that C {x1 , . . . ,xn }/(f ) is a finitely generated C {x1 , . . . ,xn−1 }–module. As it is not
more difficult, we will in fact prove a more general version. This more general version
will play an important role in many arguments in this book. First we need a definition
and a simple lemma.
Definition 3.2.8.
(1) An analytic algebra (also called analytic C –algebra) is a C –algebra of type

C {x1 , . . . ,xn }/I, where I is an ideal in C {x1 , . . . ,xn }.
(2) A formal C –algebra is a C –algebra of type C [[x1 , . . . ,xn ]]/I, where I is an ideal in
C [[x1 , . . . ,xn ]].
Lemma 3.2.9. Let (R,mR ), (S,mS ) be analytic (or formal) C –algebras, ϕ : R → S a
morphism of C –algebras. Then ϕ(mR ) ⊂ mS .
Proof. Suppose the converse. Then there exists an f ∈ mR with ϕ(f ) ∈ / mS . Thus
ϕ(f )(0) = c 6= 0. Hence ϕ(f ) − c ∈ mS is not a unit. As ϕ is a ring homomorphism,
it sends 1 to 1. As a ring homomorphism, it therefore sends units in R to units in S.
However, as ϕ is a homomorphism of C –vector spaces, ϕ(f ) − c = ϕ(f − c). Now f − c is
a unit in R, and, therefore, ϕ(f ) − c is a unit in S. This is a contradiction, proving the
lemma.
Theorem 3.2.10 (General Weierstraß Division Theorem). Let R, S be analytic C –

algebras, ϕ : R → S a morphism of C –algebras. One can, via ϕ, view S as an R–module.
Then the following conditions are equivalent:
(1) S is a finitely generated R–module.
(2) dimC S/mR S < ∞.
Analogous statements hold for formal C –algebra’s.
Before giving the proof of Theorem 3.2.10, we look at two examples.
Examples 3.2.11.
(1) We show that the Weierstraß Division Theorem follows from the General Weierstraß
Division Theorem by using Lemma 3.2.7. So for f ∈ C {x1 , . . . ,xn } which is regu-
lar of order b in xn we have to show that C {x1 , . . . ,xn }/(f ) is a finitely generated
C {x1 , . . . ,xn−1 }–module. We take R = C {x1 , . . . ,xn−1 }, and S = C {x1 , . . . ,xn }/(f ).
The map ϕ sends xi to xi for i = 1, . . . ,n−1. Then S/mR S = C {x1 , . . . ,xn }/(x1 , . . . ,xn−1 ,f ) ∼
=
C {xn }/(f (0, . . . ,0,xn ) ∼
= C b . So the statement follows from the General Weierstraß
Division Theorem.
(2) We consider R = C {x}, S = C {t}, and the map ϕ : R −→ S be given by ϕ(x) = t2 .
Then S/mR S = C {t}/(t2 ) is a two-dimensional vector space.
P∞ We claim that C {t}
k
is as C {x}–module generated by 1 and t. Now let f = k=0 fk t ∈ C {t} be
arbitrary, with fk ∈ C . Now define two power series
∞
X ∞
X
a0 = f2k xk ; a1 = f2k+1 xk .
k=0 k=0
P∞ P∞
Then ϕ(a0 ) = k=0 f2k t2k and ϕ(a1 ) = k=0 f2k+1 t2k , so that
f = ϕ(a0 ) · 1 + ϕ(a1 ) · t.
We still have to show that a0 and a1 are convergent power series. In this case this
follows easily from the convergence of f , by using 3.1.4.
Proof of the General Weierstraß Division Theorem 3.2.10.
Step 1. We first show that (1) implies (2). This is quite obvious. Any set of generators of S
as an R–module will, by taking classes, be a set of generators for S/mR S as R/mR = C –
module. Hence, S/mR S is a finite-dimensional vector space.
Step 2. We first prove the implication (2) =⇒ (1) for the case that R = C {x1 , . . . ,xn }, S =
C {y1 , . . . ,ym }. We put gi = ϕ(xi ) for i = 1, . . . ,n. The assumption then is that
C {y1 , . . . ,ym }/(g1 , . . . ,gn )
is a finite-dimensional vector space. So S/(g1 , . . . ,gn ) is an Artinian C –algebra. Writing

m = mS for short, it follows from Exercise 1.3.20 that there exists a k ∈ N such that mk ⊂
(g1 , . . . ,gn ). Multiplying with m, and putting r = k+1, it follows that mr ⊂ m(g1 , . . . ,gn ).
In particular, for any monomial y α of degree r there exist elements pα,i ∈ m such that
n
X
(3.4) yα = pα,i gi .
i=1
The claim is that the monomials {y α : |α| < r} generate S as an R-module. This would
prove the finite generatedness of S as R–module in this case.
Step 2a. We first prove this “formally”. Let f ∈ S. We can write
X
(3.5) f = a(f ) + e
hα (f )y α .
α:|α|=r
such that e
hα (f )y α and e
hβ (f )y β have no common monomial if α 6= β. Moreover a(f ) is a
polynomial in the y ′ s of degree smaller than r. Plugging in (3.4) we get
n
X
(3.6) f = a(f ) + hi (f )gi ,
i=1
P
where hi (f ) = pα,i e
hα (f ).We now iterate (3.6). For a multi-index β = (β1 , . . . ,βn )
α:|α|=r
we define inductively
(1) aβ (f ) = a(f ) for β = (0, . . . ,0),

(2) hβ (f ) = f for β = (0, . . . ,0),
(3) hβ (f ) = 0 if some entry in β is negative,
(4) aβ (f ) = a(hβ (f )),
Pn
(5) hβ (f ) = i=1 hi hβ−ei (f ) . Here e1 = (1,0, . . . ,0) etc.
Note that the aβ (f ) are polynomials in the y1 , . . . ,ym of degree smaller than r. We claim
that for all k ∈ N
X X
(3.7) f= aβ (f )g β + hβ (f )g β .
|β|<k |β|=k
For k = 1 this is justP formula (3.6). We now apply (3.6) to the coefficients hβ (f ). We get
hβ (f ) = a(hβ (f )) + ni=1 hi (hβ (f ))gi . Plugging this in (3.7) and using the definition of
aβ (f ) and hβ (f ) gives the induction step.
As (g1 , . . . ,gn ) ∈ m, it follows that for all k ∈ N
X
f≡ aβ (f )g β mod mk .
β
So in the formal power series ring we have the equality

X
f= aβ (f )g β .
β
The aβ (f ) are polynomials in the y ′ s of degree smaller than r. So we write

X
(3.8) aβ (f ) = aβ,α (f )y α , aβ,α (f ) ∈ C .
α:|α|<r
Now for all α with |α| < r look at the formal power series
X
(3.9) Aα (f ) := aβ,α (f )xβ .
β
So if we look at the map ϕ : C [[x1 , . . . ,xn ]] −→ C [[y1 , . . . ,ym ]] which sends xi to gi it

follows that X
f= ϕ(Aα (f )) · y α ,
α:|α|<r
showing the theorem in the formal case.

Step 2b. WePhave to show the convergence
P of Aα (f ). We use the seminorm of 3.1.5. That
is, for f = α fα xα we have kf kδ = α |fα |δ |α| . Now choose a δ > 0 so small that
(1) kf kδ is finite.
(2) Let C be the number of monomials of degree r. Then for the pα,i of formula (3.4)
1
kpα,i kδ ≤ C.
We can find such a δ, because the elements pα,i are in the maximal ideal, see 3.1.5 (2).
We will show for all α with |α| < r
r
(3.10) Aα (f ) < ∞ for ε = δ .
ε 2n2
This will show the convergence of Aα (f ). We have the following estimates for the terms
in formula (3.5) and (3.6).
(1) ka(f )kδ ≤ kf kδ ,
(2) ke
hα (f )kδ ≤ δ −r kf kδ ,
(3) khi (f )kδ ≤ δ −r kf kδ .
The first two estimates follow from the definition of the seminorm, and the definition of
a(f ) and e
hα (f ). The final estimate follows because we have chosen δ so that kpα,i kδ ≤ C1 ,
C being the number of monomials of degree r. Inductively we get the estimates
(3.11) khβ (f )kδ ≤ n|β|−1 · δ −|β|r · kf kδ ,

(3.12) kaβ (f )kδ ≤ n|β|−1 · δ −|β|r · kf kδ .
P
Recall that aβ (f ) = α:|α|<r aβ,α (f )y α , for some aβ,α (f ) ∈ C .
δ −|α|
Obviously |aβ,α (f )| ≤ kaβ (f )kδ δ −|α| . From (3.12) it follows with Cα = n kf kδ that
|β|
|aβ,α (f )| ≤ Cα nδ −r .
δr
With ε = 2n2 we get
X X X n |β| X 1
aβ,α (f )xβ ε = |aβ,α (f )|ε|β| ≤ Cα ε|β| = Cα .
δ r (2n)|β|
β β β β
P∞ 1
As the number of β with |β| = k is at most nk this can be estimated as Cα k=0 2k < ∞.
Step 3. The implication (2) =⇒ (1) will be proved for the case that S = C {y1 , . . . ,ym }/J,
where J = (f1 , . . . ,fp ) is a finitely generated ideal. This would be always the case if we
knew that C {y1 , . . . ,ym } is Noetherian. Now, as the Weierstraß Division Theorem is
deduced from the General Weierstraß Division Theorem by using J = (f ), see Example
3.2.11 (1), and only the Weierstraß Division Theorem is used in the proof of the fact that
C {x1 , . . . ,xn } is Noetherian, see Lemma 3.2.7, this restriction is not essential.
Step 4. We now prove the general case, for J finitely generated.
Write R = C {x1 , . . . ,xn }/I,S = C {y1 , . . . ,ym }/J, where J = (f1 , . . . ,fp ). Define
new rings:
R′ := C {x1 , . . . ,xn ,xn+1 , . . . ,xn+p }; S ′ := C {y1 , . . . ,ym }.
Take representatives g1 , . . . ,gn ∈ S ′ = C {y1 , . . . ,ym } of ϕ(x1 ), . . . ,ϕ(xn ). The map ϕ′ is

then defined on generators by:
• ϕ′ (xi ) = gi for i = 1, . . . ,n,
• ϕ′ (xn+j ) = fj for j = 1, . . . ,p.
It is immediately clear that
S ′ /mR′ S ′ = C {y1 , . . . ,ym }/(g1 , . . . ,gn ,f1 , . . . ,fp ) = S/mR S,
which by assumption is finite-dimensional. Therefore, by Step 2 we may conclude that

S ′ is a finitely generated R′ –module. Hence, there exist v1 , . . . ,vs ∈ S ′ = C {y1 , . . . ,ym },
such that any h ∈ S ′ can be written as:
h = α1 v1 + . . . + αs vs αi ∈ ϕ′ (R′ ).
Reading this equation modulo (f1 , . . . ,fp ), we see that S is an R–module generated by
v1 , . . . ,vs , so, in particular, finitely generated.
Now we know (under the assumptions of the theorem) that S is a finitely generated
R–module, we can use Nakayama’s Lemma to get the following somewhat stronger form
of the theorem.
Corollary 3.2.12. Keep the notations of the previous Theorem 3.2.10. Suppose we
have elements v1 , . . . ,vs ∈ S whose classes generate S/mR S as a C –vector space. Then
v1 , . . . ,vs generate S as an R–module.
Corollary 3.2.13. Again keep the notation of Theorem 3.2.10.
(1) Suppose mR S = mS . Then the map ϕ is surjective.

(2) Suppose S is an analytic algebra such that mS has s generators. Then S is a quotient
of the power series ring C {x1 , . . . ,xs }.
Proof. The first statement follows from 3.2.12: indeed, it follows that 1 generates S as
an R–module, and thus in particular, ϕ is surjective. For the second statement, suppose
that f1 , . . . ,fs generate mS . Consider the map
ϕ : C {x1 , . . . ,xs } −→ S,
which sends xi to fi . It follows from the first part that ϕ is surjective.

There is a “direct” proof of the second part of this corollary, using only the statement
of the Weierstraß Division Theorem. See Exercise 3.2.21.
Corollary 3.2.14. Let R,S be analytic C –algebras, ϕ : R −→ S a morphism of C –
algebras. Let M be a finitely generated S–module such that dimC M/mR M < ∞. Then
M is a finitely generated R–module.
Proof. As M is an S/ AnnS (M )–module, we may assume that AnnS (M ) = (0). Exercise
1.2.31 applied to M/mR M implies that dimC S/mR S < ∞. By the previous theorem S
is a finitely generated R–module and, therefore, M is a finitely generated R–module.
Exercises
3.2.15. Prove Lemma 3.2.2.
3.2.16. Prove that u and p in the Weierstraß Preparation Theorem are uniquely determined.
3.2.17. Prove that the Weierstraß Division Theorem is equivalent to the statement in Corollary
3.2.6.
3.2.18. Prove the Weierstraß Division Theorem for the formal power series ring.
(Hint: Consider a formal power series f ∈ C [[x1 , . . . ,xn ]] which is regular of order b in xn . Write
(3.13) f = f0 + xbn uf .
As f is regular of order b in xn it follows that

• f0 is a polynomial of degree ≤ b − 1 in xn with coefficients in (x1 , . . . ,xn−1 ).
• uf is a unit.
f f0 f0
From (3.13) it follows that xbn = uf
− uf
with uf
∈ (x1 , . . . ,xn−1 ). Construct inductively
ri ,qi ,gi with the following properties.
(1) g = qi · f + ri + gi ,
(2) gi ,qi+1 − qi ,ri+1 − ri ∈ (x1 , . . . ,xn−1 )i ,
(3) ri is a polynomial of degree ≤ b − 1 in xn .
Use as initialization q0 = r0 = 0, and g0 = g, and divide gi by xbn . )
3.2.19. Let f be a Weierstraß polynomial in xn of degree b. Suppose that ord(f ) = b. Let g be
a (formal) power series of order t. Apply the Weierstraß Division Theorem, so write g = q · f + r,
for r a polynomial of degree smaller than b in xn . Prove that the order of r is at least t.
(Hint: Follow the steps in the proof of the formal Weierstraß Division Theorem, see Exercise
3.2.18.)
3.2.20.
(1) Let f ∈ C {x1 , . . . ,xn } be regular of order b in xn , assume that b = ord(f ) and let
g ∈ (x1 , . . . ,xn )c C {x1 , . . . ,xn }. Let g = qf + r as in 3.2.3. Prove that q ∈ (x1 , . . . ,xn )c−b .
(2) Generalize the result to the weighted order (cf. 3.1.3): let w1 , . . . ,wn ∈ N be weights for
x1 , . . . ,xn . Let f,g ∈ C {x1 , . . . ,xn } be regular of order b in xn , assume that w-ord(f ) =
bwn and w-ord(g) = c, let g = qf + r as in 3.2.3. Prove that w-ord(q) = c − bwn if q 6= 0.
3.2.21.
(1) Suppose I ⊂ C {x1 , . . . ,xn } contains an element which is regular of order one in xn . Use
the Weierstraß Division Theorem to show that C {x1 , . . . ,xn }/I is isomorphic to a quotient
of C {x1 , . . . ,xn−1 }. So we can “eliminate” the variable xn .
(2) Give a different proof of the second part of Corollary 3.2.13.
3.3 Applications
The first thing to note is that the Weierstraß Preparation Theorem implies the Implicit
Function Theorem!
Theorem 3.3.1 (Implicit Function Theorem). Let f ∈ C {x1 , . . . ,xn ,y} with f (0) = 0
and ∂f
∂y (0) 6= 0. Then there exists a unique ϕ ∈ (x1 , . . . ,xn )C {x1 , . . . ,xn } with
f (x1 , . . . ,xn ,y) = 0 ⇐⇒ y = ϕ(x1 , . . . ,xn ).
Proof. Note that the condition f (0) = 0 and ∂f

∂y (0) 6= 0 exactly means that f is regular
of order one in y. By the Weierstraß Preparation Theorem
f (x1 , . . . ,xn ,y) = u · (y − ϕ(x1 , . . . ,xn ))
for some unit u ∈ C {x1 , . . . ,xn ,y} and some ϕ ∈ C {x1 , . . . ,xn }. The Implicit Function
Theorem follows from this equation.
Example 3.3.2. Consider f (x,y) = (y + 1)2 − x − 1 = y 2 + 2y −√x. Then f satisfies
the conditions of the Implicit Function Theorem. Hence y = −1 + x + 1 exists as an
analytic function.
From the uniqueness statements in the Weierstraß Preparation Theorem the follow-
ing follows immediately.
Corollary 3.3.3. Let f ∈ C {x1 , . . . ,xn } be regular of order b in xn , and suppose that
f = u · p, where u,p ∈ C [[x1 , . . . ,xn ]], u a unit and p a Weierstraß polynomial. Then u
and p are in C {x1 , . . . ,xn }, that is, they are convergent.
A similar statement holds for the Weierstraß Division Theorem.
Example 3.3.4. Consider f (x,y) = 2y + y 2 − x. Then f is regular of order one in y.
Hence, by the Weierstraß Preparation Theorem f (x,y) = u(y − ϕ(x)) for some unit u. We
can generate ϕ and u as indicated by a proof of the formal Weierstraß Division Theorem
(Exercise 3.2.18).
ϕ(x) = 12 (2y + y 2 − x) + 12 x − 21 y 2
= ( 12 − 41 y)(2y + y 2 − x) + 12 x − 14 xy + 14 y 3
= ( 21 − 41 y − 18 x + 81 y 2 )(2y + y 2 − x) + 12 x − 81 x2 − 81 y 4 + 41 xy 2
= ....
So we see the beginning of the power series
ϕ(x) = 12 x − 18 x2 + . . .
√
beginning of the power series of ϕ(x) = −1 +
which is the √ x + 1. Similarly, we obtain
u = y + 1 + x + 1.
The Implicit Function Theorem holds just as well for the formal case. From the
uniqueness statement in the Implicit Function Theorem the following fact follows.
3.3 Applications 95
Corollary 3.3.5. Let f ∈ C {x1 , . . . ,xn ,y} be regular of order 1 in y. Suppose ϕ(x) ∈
C [[x1 , . . . ,xn ]] is such that f (x1 , . . . ,xn ,ϕ(x)) = 0. Then ϕ(x) ∈ C {x1 , . . . ,xn }.
We might phrase this by saying that a formal solution to f (x1 , . . . ,xn ,y) = 0 (solving
for y), already is analytic.
By induction we get the Implicit Mapping Theorem from the Implicit Function
Theorem.
Theorem 3.3.6 (Implicit Mapping Theorem).
∂fi
Let f1 , . . . ,fm ∈ C {x1 , . . . ,xn , y1 , . . . ,ym } such that fi (0) = 0 and det ∂y j
(0) 6=
0. Then there exist uniquely defined ϕ1 , . . . ,ϕm ∈ (x1 , . . . ,xn )C {x1 , . . . ,xn } such that
fi (x1 , . . . ,xn ,ϕ1 , . . . ,ϕm } = 0 for i = 1, . . . ,m.

∂fi
Proof. First suppose that the m×m matrix ∂y j
(0) is the identity matrix. We will prove
the statement by induction on m. The case m = 1 is the Implicit Function Theorem.
For the induction step, first apply the Implicit Function Theorem to the element fm . We
conclude that there exists a unique holomorphic function ψ = ψ(x1 , . . . ,xn ,y1 , . . . ,ym−1 )
with fm (x1 , . . . ,xn ,y1 , . . . ,ym−1 ,ψ) = 0. It is left as an exercise to show that the functions
fi (x1 , . . . ,xn ,y1 , . . . ,ym−1 ,ψ), i = 1, . . . m − 1,
satisfy the induction hypothesis.

−1
∂fi
For the general case, let A = ∂yj (0) . Then

∂fi
A· (0)
∂yj
is the identity matrix. Replacing f1 , . . . ,fm by linear combinations of those given by A,

we come back to the first case.
The proof of the Inverse Function Theorem, which follows from the Implicit Mapping
Theorem is standard, and left as Exercise 3.3.29.
Corollary 3.3.7 (Inverse Function Theorem).
Let f1 , . . . ,fn ∈ C {x1 , . . . ,xn } such that
∂fi
f1 (0) = · · · = fn (0) = 0. Then det
∂xj (0) 6
= 0 if and only if the C –algebra homomor-
phism
C {x1 , . . . ,xn } −→ C {x1 , . . . ,xn }

xi 7−→ fi
is an isomorphism. This again holds if and only if there exist open neighborhoods U,W
of 0 such that
F = (f1 , . . . ,fn )
defines a map F : U −→ W and this map has a holomorphic inverse.
The Inverse Function Theorem is the basis to introduce coordinate functions and
complex submanifolds of C n .
Definition 3.3.8.
(1) Let f1 , . . . ,fn be holomorphic on an open subset U . Let p ∈ U , and suppose f1 (p) =
··· =
fn (p)= 0. The set {f1 , . . . ,fn } is called a set of coordinate functions at p if
∂f
det ∂xi (p) 6= 0.
(2) A subset X ⊂ C n is called a complex submanifold of C n if for every x ∈ X there

exists an open subset U in C n and coordinate functions w1 , . . . ,wn of x such that
X ∩ U = {y ∈ U : w1 (y) = · · · = wm (y) = 0} for some m ≤ n.
Remark 3.3.9. A complex submanifold X ⊂ C n locally looks like an open subset of
C k for some k.
Example 3.3.10. Let f1 , . . . ,fk be holomorphic functions on the open set U ⊂ C n .
Then X = a1 , . . . ,an , f1 (a1 , . . . ,an ), . . . ,fk (a1 , . . . ,an ) : (a1 , . . . ,an ) ∈ U is a complex
submanifold of C n+k . To see this, let x1 , . . . ,xn ,y1 , . . . yk be coordinates on C n+k . Then
X can be given by
y1 − f1 (x1 , . . . ,xn ) = · · · = yk − fk (x1 , . . . ,xn ) = 0,
and one checks that these are part of a set of coordinate functions at every point of X.
We will see later in Chapter 4 that for an analytic subset of C n almost all points have
a neighborhood which is a zero set of coordinate functions. More precisely, an analytic
subset contains as an open set a complex submanifold. This gives rise to the following
definition.
Definition 3.3.11. Let U ⊂ C n be an open set and X ⊂ U an analytic subset. A point
x ∈ X is called regular, or X is called smooth at x, if there exists an open subset V in
C n , x ∈ V , such that X ∩ V is a complex submanifold of C n . If x ∈ X is not regular,
then x is called singular, or a singularity. The set of singular points of X is denoted by
Sing(X).
Examples 3.3.12.
(1) Every point in a complex submanifold of C n is a regular point.

(2) Let X = {(x,y) ∈ C 2 : x2 − y 3 = 0}. Then X is an analytic subset of C 2 ,
0 = (0,0) ∈ X is a singular point, see Exercise 3.3.37.
In chapter 4 we will obtain a characterization of the singular points in terms of the

Jacobian matrix (cf. Theorems 4.3.6 and 4.3.15).
We now turn to Newton’s Lemma. Newton’s Lemma, as you learned it in a calculus
course, is a method for approximating zeros of differentiable functions in one variable.
Consider a differentiable function y = f (x) in one real variable. The question is to find a
zero of f (x) = 0. The idea is to take an approximate solution ak , and to obtain a better
approximation ak+1 by intersecting the tangent line of the graph of f at (ak ,f (ak )) with
the x–axis.
3.3 Applications 97
ak+1
y =0
ak
A simple calculation shows
f (ak )
(3.14) ak+1 = ak − .
f ′ (ak )
Starting off with an initial approximation a0 we get a sequence a1 ,a2 , . . ., which, under
appropriate assumptions, converges to a number a with f (a) = 0.
We apply similar ideas to the equation f (x,y) = y 2 − 1 − x. We want to find √ a power
series, y = ϕ(x) with ϕ(x)2 = x + 1, that is, we want to find a power series for x + 1.
We start of with the initial approximation ϕ0 (x) = 1, and iterate with the obvious analog
of (3.14):

f x,ϕk (x)
(3.15) ϕk+1 (x) = ϕk (x) − ∂f .
∂y x,ϕk (x)
One calculates that

2
x
x x 4 x x2
ϕ1 (x) = 1 + , ϕ2 (x) = 1 + − =1+ − + ...,
2 2 2(1 + x/2) 2 8
√
so one sees the beginning of the power series of 1 + x.
We now generalize.
Corollary 3.3.13 (Newton’s Lemma: simple case).
Let f ∈ C {x1 , . . . ,xn ,y}, y(x) ∈ C {x1 , . . . ,xn } and c ≥ 1 with
2
∂f
f (x,y(x)) ≡ 0 mod x,y(x) (x)c .
∂y
Then there exists a y ∈ C {x1 , . . . ,xn } with

(1) f (x,y(x)) = 0.
c
(2) y ≡ y mod ∂f ∂y x,y(x) (x) .
∂f

Proof. Write ∂y x,y(x) = α ∈ C {x1 , . . . ,xn }. We choose a new variable t and substitute
y = y(x) + t · α
in f (x,y). We therefore want to find a t ∈ C {x1 , . . . ,xn } with


g(x,t) := f x,y(x) + t · α = 0.
The existence will be guaranteed by applying the Implicit Function Theorem. Indeed
Taylor’s formula gives
∂f
g(x,t) := f x,y(x) + tα = f x,y(x) + tα x,y(x) + α2 R = f x,y(x) + tα2 + α2 R
∂y
with R ∈ (t)2 . By assumption f (x,y(x)) = α2 · r with r ∈ (x)c . Therefore g(x,t) =

α2 (t + R + r). Hence it suffices to solve t + R + r = 0, with R ∈ (t)2 and r ∈ (x)c . Hence
this equation satisfies the assumption of the Implicit Function Theorem. Even more, by
Exercise 3.3.33 there exists a t ∈ (x)c with g(x,t) = 0.
A more general version of Newton’s Lemma, whose proof is only notationally more
difficult, will be proved in Exercise 3.3.34. This generalization will, for example, be used
in the proof of Artin’s Approximation Theorem, see Chapter 8.
Corollary 3.3.14. C {x1 , . . . ,xn } and C [[x1 , . . . ,xn ]] are Noetherian rings.
Proof. As the proofs are similar, we only give the proof for the analytic case. The proof is
by induction on n. For the case n = 0 we have the field C , which is obviously Noetherian.
Now take an ideal I ⊂ C {x1 , . . . ,xn }. We have to prove that I is finitely generated.
If I = (0), this is obviously the case. If I 6= (0), take 0 6= f ∈ I. After a general
linear change of coordinates we may assume that f is regular in xn , say of order b,
see 3.2.2. By 3.2.6, which is a corollary of the Weierstraß Division Theorem, it follows
that C {x1 , . . . ,xn }/(f ) is a finitely generated (even free) C {x1 , . . . ,xn−1 }–module. Now
C {x1 , . . . ,xn−1 } is Noetherian by induction, so that C {x1 , . . . ,xn }/(f ) is a Noetherian
C {x1 , . . . ,xn−1 }–module by 1.2.15. In particular I modulo (f ) is finitely generated as
C {x1 , . . . ,xn−1 }–module, and therefore certainly finitely generated as C {x1 , . . . ,xn }–
module, say by the classes of f1 , . . . ,ft . It follows that f,f1 , . . . ,ft generate I.
Remarks 3.3.15.
(1) Analogous ideas also give a new proof for the Hilbert Basis Theorem for the poly-
nomial ring C [x1 , . . . ,xn ].
(2) An alternative proof for the induction in the proof of Corollary 3.3.14 uses the
Hilbert Basis Theorem. It runs as follows. Let I 6= (0) be an ideal, and 0 6=
f ∈ I. Again, we may assume that C {x1 , . . . ,xn }/(f ) is a finitely generated
C {x1 , . . . ,xn−1 }–module. We may also assume that f is a Weierstraß polynomial
in xn . Hence C {x1 , . . . ,xn }/(f ) = C {x1 , . . . ,xn−1 }[xn ]/(f ). It follows from the
Weierstraß Division Theorem that
(I ∩ C {x1 , . . . ,xn−1 }[xn ]) C {x1 , . . . ,xn } = I.
By induction and the Hilbert Basis Theorem, the ring C {x1 , . . . ,xn−1 }[xn ] is Noethe-
rian. Hence I is finitely generated as C {x1 , . . . ,xn−1 }[xn ]–ideal, and in particular
finitely generated as C {x1 , . . . ,xn−1 ,xn }–ideal.
The next thing we want to show is that the power series ring is a unique factorization
domain. For the proof we need the following fact, which is interesting in its own right.
3.3 Applications 99
Lemma 3.3.16. Let h ∈ C {x1 , . . . ,xn−1 }[xn ] be a Weierstraß polynomial. Then

h is reducible in C {x1 , . . . ,xn−1 }[xn ] ⇐⇒ h is reducible in C {x1 , . . . ,xn }.
In particular, every irreducible factor of h in C {x1 , . . . ,xn } is xn –regular and, therefore,
every decomposition of h in C {x1 , . . . ,xn } gives a decomposition of h into Weierstraß
polynomials by multiplying with suitable units.
Similar statements hold for the formal power series ring.
Proof. We again only consider the analytic case, the formal case being similar.
⇐=: Let h = g1 · g2 be a factorization in C {x1 , . . . ,xn } with g1 and g2 nonunits. As h
is regular in xn it follows that g1 and g2 are regular in xn . As they are both nonunits,
they are regular of order at least one. By the Weierstraß Preparation Theorem we can
write g1 = u1 h1 , and g2 = u2 h2 , where u1 and u2 are units and h1 and h2 are Weierstraß
polynomials in xn . Hence h = u1 u2 h1 h2 . Now h1 h2 is a Weierstraß polynomial. It follows
from the uniqueness statement in the Weierstraß Preparation Theorem that u1 u2 = 1.
Therefore, h = h1 h2 , and h is reducible in C {x1 , . . . ,xn−1 }[xn ].
=⇒: Suppose h = g1 · g2 , with g1 and g2 nonunits in C {x1 , . . . ,xn−1 }[xn ]. We have to
show that g1 and g2 are also nonunits in C {x1 , . . . ,xn }. But suppose, for example, that
g1 is a unit. As g1 is not a unit in C {x1 , . . . ,xn−1 }[xn ] it follows that degxn (g1 ) ≥ 1.
So degxn (g2 ) < degxn (h). But h = g1 · g2 , so g2 is regular in xn . This implies that h is
xn –regular of order at most degxn (g2 ) which is a contradiction to the fact that h is a
Weierstraß polynomial.
Corollary 3.3.17. C {x1 , . . . ,xn } and C [[x1 , . . . ,xn ]] are factorial rings.
Proof. Again we only treat the convergent case. The proof is by induction on n. The case
n = 0 is trivial. Suppose therefore that C {x1 , . . . ,xn−1 } is factorial. Then by the Lemma
of Gauß (1.4.4) C {x1 , . . . ,xn−1 }[xn ] is factorial. Let f ∈ C {x1 , . . . ,xn }. After a general
linear change of coordinates we may assume that f is regular in xn . By the Weierstraß
Preparation Theorem f = uh, where u is a unit and h is a Weierstraß polynomial. As
by the previous lemma, the polynomial factors of h in both C {x1 , . . . ,xn−1 }[xn ] and
C {x1 , . . . ,xn } are equal, it follows that there is a unique factorization of h.
For the case of two variables we obtain the following result which will be very useful
in Chapter 5.
Corollary 3.3.18. Let f ∈ C {x,y} and y ∈ (x)C [[x]] such that f (x,y) = 0. Then
y ∈ C {x}.8
Proof. Suppose the converse, that is, assume that y 6∈ C {x}. First of all, we may
assume that f = y n + an−1 (x)y n−1 + · · · + a0 (x) is an irreducible Weierstraß poly-
nomial by (3.3.16). As y ∈ / C {x} it follows that n > 1. From Lemma 1.4.5 it fol-
lows that f is irreducible in Q(C {x})[y]. The ring Q(C {x})[y] is a principal ideal do-
main, hence (f, ∂f
∂y ) = (1). By clearing denominators, we can find p,q ∈ C {x}[y] and
0 6= h ∈ C {x} with qf + p ∂f ∂f
∂y = h. Plugging in y = y(x) it follows that ∂y (x,y) 6= 0.

Let ord ∂f ∂y (x,y) = m and choose y1 ∈ C [x] such that ord(y − y1 ) ≥ 2m + 1. Then
2
f (x,y1 ) ≡ 0 mod ∂f ∂y (x,y1 ) (x) = (x
2m+1
).
8 This corollary is also true for n variables x1 , . . . ,xn and can be proved, for instance, using Artin’s
Approximation Theorem from Chapter 8. In fact, it follows from Exercise 8.1.7.
Using Newton’s Lemma

3.3.13 we obtain an y2 ∈ C {x} such that f (x,y2 ) = 0 and
y − y2 ∈ (x) ∂f
∂y (x,y1 ) . Now f (x,y) = g(x,y)(y − y2 ), and g(x,y) = 0. Hence g is not a
unit, an we obtain a contradiction to the fact that f is irreducible.
We now come to the Noether Normalization Theorem. Its proof is analogous to the
proof in the affine case, see 2.2.9, but because it is such an important result, we spell it
out.
Corollary 3.3.19 (Noether Normalization). Let I ⊂ C {x1 , . . . ,xn } be an ideal. Then
after a general linear coordinate change9 we have that there exists a k such that
C {x1 , . . . ,xk } ⊂ C {x1 , . . . ,xn }/I
and such that C {x1 , . . . ,xn }/I is a finitely generated C {x1 , . . . ,xk }–module. If we have
such a situation we call
C {x1 , . . . ,xk } ⊂ C {x1 , . . . ,xn }/I
a Noether normalization. If moreover I 6= (0), then k < n.

Proof. If I = (0), we can take k = n. Otherwise, the theorem is proved by induction on n.
The case n = 0 is trivial. Take 0 6= g ∈ I. By 3.2.2 we have, after a general linear change
of coordinates, that g is regular of order b in xn . By Corollary 3.2.6, we know that
C {x1 , . . . ,xn }/g is a finitely generated C {x1 , . . . ,xn−1 }–module. Because g ∈ I, then
certainly C {x1 , . . . ,xn }/I is a finitely generated C {x1 , . . . ,xn−1 }/(I ∩ C {x1 , . . . ,xn−1 })–
module. We apply the induction hypothesis to the ideal I ∩ C {x1 , . . . ,xn−1 }. After a
general linear coordinate change in x1 , . . . ,xn−1 we have that
C {x1 , . . . ,xk } ⊂ C {x1 , . . . ,xn−1 }/(I ∩ C {x1 , . . . ,xn−1 })
is finitely generated. We compose this with the inclusion
C {x1 , . . . ,xn−1 }/(I ∩ C {x1 , . . . ,xn−1 }) ⊂ C {x1 , . . . ,xn }/I.
As a composition of finite rings extension is finite, see 1.5.2, we deduce that the ring
extension C {x1 , . . . ,xk } ⊂ C {x1 , . . . ,xn }/I is finitely generated. This proves the corol-
lary.
Due to the fact that for general coordinate changes we get a Noether normalization,
we can get the following situation, which will play an important role in the proof of the
Nullstellensatz.
Corollary 3.3.20. Suppose that p ⊂ C {x1 , . . . ,xn } is a prime ideal. Then we can get a
Noether normalization
C {x1 , . . . ,xk } ⊂ C {x1 , . . . ,xn }/p,
with the property that the class of xk+1 in Q(C {x1 , . . . ,xn }/p) is a primitive element of
the field extension
Q(C {x1 , . . . ,xk }) ⊂ Q(C {x1 , . . . ,xn }/p).
We call a Noether normalization with the above property a primitive Noether normaliza-
tion.
9 More precisely, there exists a Zariski open subset U of Gln (C ) such that the statement holds for all
coordinate transformations in U .
3.3 Applications 101
Proof. We know from the proof of the Theorem of the Primitive Element, see for example
[Van der Waerden 1971], that the class of ck+1 xk+1 + · · · + cn xn in Q(C {x1 , . . . ,xn }/p)
is a primitive element for the field extension Q(C {x1 , . . . ,xn }/p) ⊃ Q(C {x1 , . . . ,xk }) for
general ck+1 , . . . ,cn ∈ C . Doing the coordinate change x′k+1 = ck+1 xk+1 + · · · + cn xn the
corollary follows.
Corollary 3.3.21 (Hensel’s Lemma). Let F ∈ C {x1 , . . . ,xn }[T ] and suppose that F ≡
(T − c1 )s1 · · · (T − ce )se mod (x1 , . . . ,xn ), where the ci ∈ C are pairwise different.
Then F = F1 · · · Fe , Fi ∈ C {x1 , . . . ,xn }[T ], Fi ≡ (T − ci )si mod (x1 , . . . ,xn ), and
the Fi have degree si in T .
Proof. We use induction on e. The case e = 1 is trivial. We may assume that c1 = 0
(otherwise we perform the transformation T 7→ T + c1 ). Then F (0,T ) = T s1 · H(T ) and
H(0) 6= 0. Now we use the Weierstraß Preparation Theorem (3.2.4) and obtain F = E·F1 ,
where F1 is a Weierstraß polynomial of degree s1 with respect to T , and E is a unit in
C {x1 , . . . ,xn ,T }.
On the other hand, E also has to be a polynomial in T : divide F by F1 with remainder
in C {x1 , . . . ,xn }[T ], F = E1 F1 + R1 with E1 ,R1 ∈ C {x1 , . . . ,xn }[T ] and degT R1 < s1 .
By the uniqueness statements in the Weierstraß Division Theorem, we obtain R1 = 0
and E = E1 . Now F1 (0,T ) = T s1 and, consequently, E(0,T ) = (T − c2 )s2 · · · (T − ce )se .
The corollary is proved by induction hypothesis.
Zero set of F (0,T ) Zero set of F
Corollary 3.3.21 is equivalent to the following seemingly more general lemma:

Remark 3.3.22 (General Hensel’s Lemma). Let F,G,H ∈ C {x1 , . . . ,xn }[T ] such that
(1) F ≡ G · H mod (x1 , . . . ,xn );
(2) G is a monic polynomial;
(3) (G,H,x1 , . . . ,xn ) = C {x1 , . . . ,xn }[T ].
Then there exist G1 ,H1 ∈ C {x1 , . . . ,xn }[T ] such that F = G1 · H1 , Moreover G1 ≡ G
mod (x1 , . . . ,xn ), and H1 ≡ H mod (x1 , . . . ,xn ).
The proof of the equivalence is left as part of Exercise 3.3.36.
Corollary 3.3.23. Let f be a polynomial,
f = xkn + a1 xnk−1 + . . . + ak , ai ∈ C {x1 , . . . ,xn−1 }.
Consider an open set U ⊂ C n−1 on which the ai converge, and a point p ∈ U such that
the polynomial
xkn + a1 (p)xnk−1 + . . . + ak (p)

has k different roots in C . Then there exists an open neighborhood V ⊂ U of p, and k
different holomorphic functions α1 , . . . ,αk defined on V such that for all (x1 , . . . ,xn ) ∈
V ×C
k
Y
f (x1 , . . . ,xn ) = xn − αj (x1 , . . . ,xn−1 ) .
j=1
Proof. This is in fact Hensel’s Lemma for the case that all si are equal to one.
The following corollary we need in the proof of the Finite Mapping Theorem 6.3.5,
Corollary 3.3.24. Consider a monic polynomial F ∈ C {x1 , . . . ,xn }[T ].PSuppose that
F (0,T ) = (T − c1 )s1 · · · (T − ce )se for pairwise different ci ∈ C . Let s = ei=1 si be the
degree of F . Let pi := (0, . . . ,0,ci ) ∈ C n+1 . Then for all g1 , . . . ,ge with gi ∈ On+1,pi there
exist uniquely determined qi ∈ On+1,pi and a polynomial r ∈ C {x1 , . . . ,xn }[T ] of degree
less than s such that
gi = qi F + r
holds in On+1,pi = C {x1 , . . . ,xn ,T − ci } for i = 1, . . . ,e.
Proof. We first show existence. By Hensel’s Lemma, we may write F = F1 · · · Fe , with
ci )si , that is, the Fi are Weierstraß polynomials with respect to T − ci .
Fi (0,T ) = (T −Q
We write Ei = j6=i Fj , so that F = Ei · Fi . Note that Ei is a unit in On+1,pi . We apply
the Weierstraß Division Theorem in On+1,pi for i = 1, . . . ,e:
Ei−1 gi = qi′ Fi + ri
where ri ∈ C {x1 , . . . ,xn }[T ] has degree smaller

P than si . So gi = qi′ F + ri Ei . We set
′
r = r1 E1 + . . . + re Ee . Thus gi = qi F + r − j6=i rj Ej . In the ring On+1,pi the elements
Fj for j 6= i are units, so we get
X
gi = qi′ F + r − rj Fj−1 F.
j6=i
P
Hence we may take qi = qi′ − j6=i rj Fj−1 .
To show uniqueness, it suffices to show that from qi F − r = 0 for i = 1, . . . ,e with
degree of r in T smaller than s, it follows that r = 0. Thus we may write r = F1 · · · Fe qi
with qi ∈ On+1,pi . We define elements
r r r
P1 := , P2 := , . . . , Pe :=
F1 F1 F2 F1 F2 · · · Fe
Note that as r = F1 · · · Fe qi the Pi are elements of On+1,pi . Note that
r = P1 F1 ; Pj−1 = Pj Fj for j = 2, . . . ,e.
As Fi and r are polynomials in T − ci it follows from the uniqueness in the Weierstraß

Division Theorem in On+1,p1 and the fact we can do polynomial division that P1 is a
polynomial in T . Inductively, we get that P2 , . . . ,Pe are polynomials in T . So we get
r = Pe F1 F2 · · · Fe = Pe F . As the degree of F is equal to s, and r has smaller degree, it
follows that r = Pe = 0.
Corollary 3.3.25. Let R = On /I be an analytic algebra, and S be an R–algebra, which

is finitely generated as R–module. Suppose that S is a local ring. Then S is also an
analytic algebra.
Proof. First of all, take a Noether normalization C {x1 , . . . ,xk } = Ok ֒→ R. Therefore,
we may assume that R = Ok from the beginning. Let mR be the maximal ideal of R, and
mS be the maximal ideal of S. We have to prove that S ∼ = C {y1 , . . . ,ym }/J for some m
and J. Now S/mR S is a finitely generated R/mR = C –module, hence a finitely generated
vector space. As mR S ⊂ mS , it follows that S/mS is a finite extension of the field C . As
C is algebraically closed it follows that S/mS = C . Therefore for all u ∈ S, there exists
a c ∈ C such that u − c ∈ mS . It follows that we can take generators 1 = u0 ,u1 , . . . ,ut
of S as R–module with ui ∈ mS for i = 1, . . . ,t. We look at the surjective C –algebra
homomorphism
Ψ : Ok [y1 , . . . ,yt ] −→ S
yi 7→ ui
and conclude that S ∼ = Ok [y1 , . . . ,yt ]/J for J = Ker(Ψ). But we want to show that
S is a quotient of a power series ring. Therefore, we look at the canonical inclusion
Ok [y1 , . . . ,yt ] ⊂ Ok+t , and put Je := J · Ok+t . We obtain a canonical map:
e
θ : Ok [y1 , . . . ,yt ]/J → Ok+t /J.
Therefore, it suffices to show that θ is an isomorphism. So we have to prove two things.

Step 1: θ is surjective. As S is an integral extension of Ok , there exists for all i a monic
polynomial pi ∈ Ok [X] such that pi (ui ) = 0. This shows, by definition of the ideal J,
that pi (yi ) ∈ J. By the Weierstraß Division Theorem, any element in Ok+t is modulo Je
equivalent to a polynomial in y1 , . . . ,yt . This shows that the map θ is surjective.
Step 2: θ is injective. Take generators p1 , . . . ,ps for the ideal J. Take any element p ∈
e We have to show that p ∈ J. For this it suffices to show by
Ker(θ) = Ok [y1 , . . . ,yt ] ∩ J.
definition of J that Ψ(p) = 0. Now p is a polynomial in y1 , . . . ,yt , and we can write:
p = g1 p1 + . . . + gs ps , gi ∈ Ok+t .
The gi are power series in general, so it does not make sense to talk about ψ(gi ). We
therefore calculate modulo high powers of the yi , that is, for every natural number k we
write gi = gik + hik , where hik ∈ (y1 , . . . ,yt )k+1 and the gik are polynomials. Because
Ψ(yi ) = ui ∈ mS it follows that modulo mk+1 S :
Ψ(p) = Ψ(gik p1 + . . . + gsk ps ).
But Ψ(pi ) = 0, because the pi were supposed to be generators of J. It follows Ψ(p) ∈ mk+1
S
for all k. It follows from Krull’s Intersection Theorem 1.3.5 that Ψ(p) = 0, or, what is
the same, p ∈ J.
At the end of the section we look at generalization of this result, namely at the case
S is not necessarily local.
Corollary 3.3.26. Let R be an analytic algebra, and S be an R–algebra, which is finitely
generated as R–module. Then
r
M
S∼
= Si
i=1
and the Si are analytic algebras. So S is a finite direct sum of analytic algebras.
The proof uses the case that S is a local ring, see 3.3.25 and the following lemma.
Lemma 3.3.27. Let R = On , and S ⊃ R be an R–algebra, finitely generated as R–
module. Let m be the maximal ideal of R. Let y ∈ S be an idempotent modulo mS, that
is, y 2 − y ∈ mS. Then y can be lifted to an idempotent in S, that is, there exists a z ∈ S
with z 2 = z and z ≡ y mod mS.
√
Proof. As S is a finitely generated R–module, and y 2 − y ∈ mS ⊂ mS it follows from
Lemma 1.5.11 that y 2 − y is integral over m. From this integral equation it follows that
m
X
(y 2 − y)r = ck y k , for some ck ∈ m.
k=0
Define
m
X
F (T ) := (T 2 − T )r − ck T k ∈ R[T ].
k=0
r r
So F (T ) ≡ T (T − 1) mod m. We can now apply Hensel’s Lemma 3.3.21 and get
F = G · H, G(T ) ≡ T r mod m and H(T ) ≡ (T − 1)r mod m. In particular, from y 2 ≡ y
mod mS, we obtain G(y) ≡ y r ≡ y mod mS and H(y) ≡ (y − 1)r ≡ (−1)r (1 − y)
mod mS. Hence
1 = y + (1 − y) ≡ G(y) + (−1)r H(y) mod mS.
Let M be a maximal ideal of S. Then M ∩ R = m is the maximal ideal of R. This can be

proved directly, or follows from the Going-Up Theorem. It follows that mS is contained
in all maximal ideals. Hence all elements of 1 + mS are units in S,and the inverse is also
congruent to 1 modulo mS, see 1.3.25. Thus G(y) + (−1)r H(y) 1 + α = 1 for some
α ∈ mS. We get
1 = c · G(y) + dH(y)
for suitable c,d ∈ S, and c ≡ 1 mod mS. We now define z := c · G(y). Then z ≡ y
mod mS, and
z = c · G(y) + dH(y) z = z 2 + dH(y)cG(y) = z 2
because 0 = F (y) = G(y)H(y). This is what we had to show.
Proof of Corollary 3.3.26. Let m be the maximal ideal of R. The proof is by induction on
dimC (S/mS). The case that S is local has already been considered. So assume that S is
not a local ring. As S is a finitely generated R–module we get that S/mS is an Artinian
C –algebra (1.1.6). Because of Theorem 1.4.26 we have10
r
M
S/mS ∼
= Ci , Ci nonzero local Artinian C –algebras.
i=1
10 Recall that in the direct sum of algebras the multiplication is component wise.
The number r is bigger than one, as S/mS is not a local ring. In particular S/mS contains
an idempotent different from 1. (For example (1,0, . . . ,0) in ⊕ri=1 Ci ).
This idempotent can be lifted by Lemma 3.3.27 to an idempotent y ∈ S, y 6= 1. We
apply the Chinese Remainder Theorem 1.1.12 and get
S∼
= S/(y) × S/(1 − y).
To apply the Chinese Remainder Theorem, we need that (y) + (1 − y) = R (this is

obvious), and (y) ∩ (1 − y) = (0). To show this, let a ∈ (y) ∩ (1 − y). Then we can write
a = αy = β(1−y). Multiply with y and get αy 2 = 0. As y 2 = y it follows a = αy = 0. Note
that both y and (1−y) are not in mS. Indeed, they were chosen such that modulo mS the
element y is equivalent to (1,0, 1 − y is equivalent to (0,1, . . . ,1). So it follows
. . . ,0), and
that dimC (S/mS) > dimC S/ mS + (y) . Similarly dimC (S/mS) > dimC S/ mS +

(1 − y) . Thus we can apply the induction hypothesis, and the corollary follows.
Exercises
√
3.3.28. Compute the power series of y = −1 + x + 1 using the ideas of Example 3.3.4.
3.3.29. Prove the Inverse Function Theorem 3.3.7.
3.3.30 (General Noether Normalization). Let I ⊂ C {x1 , . . . ,xn } be an ideal. Then, after a
general linear coordinate change, we find that there exists a k such that
C {x1 , . . . ,xk } ⊂ C {x1 , . . . ,xn }/I
and for all j > k there are monic polynomials fj ∈ I ∩ C {x1 , . . . ,xj−1 }[xj ] (with respect to xj )
with degxj (fj ) = ord(fj ) (considered as a power series in C {x1 , . . . ,xj }).
If, moreover, I is a prime ideal then Q(C {x1 , . . . ,xn }/I) = Q(C {x1 , . . . ,xk })[xk+1 ]/(fk+1 ).
(Hint: Assume in the proof of 3.3.19 that the order of g ∈ I is b and apply 3.2.2 and the
Weierstraß Preparation Theorem 3.2.4. In case of I being a prime ideal, combine this idea with
the proof of 3.3.20.)
3.3.31. Consider a general Noether normalization Ok ⊂ OX,x = On /p in the sense of 3.3.30. Let
f be a power series of order t. Show that modulo p the element f is equivalent to a polynomial
in xk+1 , . . . ,xn which has order at least t.
(Hint: Reduce to the case that p = (Pk+1 , . . . ,Pn ), where the Pj are Weierstraß polynomials
in xj with coefficients in C {x1 , . . . ,xk }. For the case p = (Pn ), use Exercise 3.2.19, and apply
induction to the coefficients.)
3.3.32. Consider a prime ideal p ∈ C {x1 , . . . ,xn }, and a Noether normalization C {x1 , . . . ,xk } ⊂
C {x1 , . . . ,xn }/p. Let Pj (X) be an integral equation of xj for j = k + 1, . . . ,n of minimal degree.
Prove that the Pj are Weierstraß polynomials.
(Hint: Use the Weierstraß Preparation Theorem.)
3.3.33. Let f ∈ C {x1 , . . . ,xn ,y} with f (0) = 0 and ∂f ∂y

(0) 6= 0. Suppose moreover that f ∈
(y) + (x1 , . . . ,xn )c . Then there exists a ϕ ∈ (x1 , . . . ,xn )c with
f (x1 , . . . ,xn ,ϕ) = 0.
Generalize this to the case of m variables y1 , . . . ,ym . (Hint: Follow the steps in a proof of the
formal Implicit Function Theorem.)
3.3.34 (Newton’s Lemma). Let f1 , . . . ,fm ∈ C {x1 , . . . ,xn , y1 , . . . ,yN }, m “≤ N” and J =

∂fi
J(f1 , . . . ,fm ) the ideal generated by the m–minors of the Jacobian matrix ∂y . Suppose
` ´ 2
` ´ c N
j
fi x,y(x) ≡ 0 mod J y(x) (x1 , . . . ,xn ) for i = 1, . . . ,m, y(x) ∈ C {x1 , . . . ,xn } and c ≥ 1.
Then there exists a y ∈ C {x1 , . . . ,xn }N such that
` ´
fi x,y(x) = 0,` i =´ 1, . . . ,m
y ≡ y mod J y(x) (x1 , . . . ,xn )c .
Prove this lemma along the following lines.

` ´ “ ”
∂fi
(1) Let M1 , . . . ,Ms , with s = N
m
be the m–minors of ∂y j
(y(x)) . Take new variables tij
f1
!
Ps
.. . Show that
for 1 ≤ i ≤ N and 1 ≤ j ≤ s. Substitute yi = y i + tij Mj . Let f =
j=1 .
fm
it suffices to solve for tij an equation of type
0 P 1
t1j Mj
X X „ «B j C
∂fi B .. C
(3.16) 0 = f (y) = Mi Mj rij + Mi Mj hij + (y) B C.
∂yj @ P . A
tNj Mj
j
c
where the rij is a column vector of size m with entries in (x1 , . . . ,xn ) and hij is a column
vector of size m with entries in (t11 , . . . ,tNs )2 .
(2) Consider an m × N matrix A with entries in a ring R. Let M be an m–minor of A. Show
that there exists a N × m matrix B such that AB = M Idm , where Idm is the m × m
identity matrix.
(Hint: If m = N , one can take for B the adjoint of A, see 1.2.9.)
„ «
∂fi
(3) Use this to replace Mi Mj in (3.16) by Mj (y) Bi for certain matrices Bi .
∂yj
(4) Show that for all j with 1 ≤ j ≤ s it suffices to solve for tij the equation
0 1
s
t1j
X B C
Bi (rij + hij ) + @ ... A = 0.
i=1
tNj
Apply the Implicit Mapping Theorem to solve this system of equations.

3.3.35. Newton’s Lemma cannot be strengthened in the following sense, that is, the following
statement is not true in general.
Let f ∈ C {x1 , . . . ,xn ,y}, and y(x) ∈ C {x1 , . . . ,xn ,y} with
„ «
` ´ ∂f ` ´
f x,y(x) ≡ 0 mod x,y(x) (x)c .
∂y
Then there exists a y ∈ C {x1 , . . . ,xn ,y} with

` ´
(1) f x,y(x) = 0.
(2) y ≡ y mod (x)c .
Indeed, show that f (x,y) = y 2 − x3 is a counterexample.
3.3.36. A local C –algebra R with maximal ideal m is called Henselian if the statement of
Hensel’s Lemma (see 3.3.21), where C {x1 , . . . ,xn } is replaced by R, and (x1 , . . . ,xn ) is replaced
by m, holds. Prove the following statements.
(1) Let R be an analytic algebra. Then R is Henselian.
3.4 Germs of Analytic Spaces 107
(2) Show that K[x](x) is not Henselian, by looking at the polynomial F = T 2 − x − 1.

(3) Let p be a prime ideal. Prove that more generally K[x1 , . . . ,xn ]p is not Henselian.
(4) Prove Remark 3.3.22.
3.3.37. Prove that the point 0 ∈ X in Example 3.3.12 (2) is singular.
3.4 Germs of Analytic Spaces
In this section we study germs of analytic spaces. We already know what an analytic
space is, see 3.1.12, but we still have to define the notion of a germ.
Definition 3.4.1.
(1) Let X be a topological space and p ∈ X. Subsets A and B of X are called equivalent
at p if there exists a neighborhood U of p such that A∩U = B∩U . It is easy to check
that this is indeed an equivalence relation, see Exercise 3.4.33. The equivalence class
of A at p we call the germ of A at p. A is called a representative of the germ. We
write (A,p) for the germ of A at p.
(2) Let (X,x) and (Y,x) be germs. We define (X,x) ⊂ (Y,x) if there are representatives
X of (X,x) and Y of (Y,x) such that X ⊂ Y . It is not so difficult to see that
(X,x) = (Y,x) if and only if (X,x) ⊂ (Y,x) and (Y,x) ⊂ (X,x).
(3) Let (X,x) ⊂ (Y,x) and (Z,x) ⊂ (Y,x) be germs. Then we define (X,x) ∩ (Z,x) to
be the germ of X ∩ Z at x for any representative X of (X,x), and Z of (Z,x) at x.
We define (X,x) ∪ (Z,x) to be the germ of X ∪ Z at x for any representative X of
(X,x), and Z of (Z,x) at x.
We now define germs of analytic spaces.
Definition 3.4.2.
(1) A germ of an analytic space (X,x) is a germ at x of a locally analytic subset of C n .

(2) Let f ∈ On,x . Then we define a germ of an analytic hypersurface V (f ),x as
follows. Consider an open neighborhood U of x on which f converges. Consider the
analytic subset V (f ) := {p ∈ U : f (p) = 0} of U . Then V (f ),x is the germ of
V (f ) at zero, and is called the zero set of f .
(3) Let I = (f1 , . . . ,fs ) ⊂ On,x be an ideal. Then we define the germ of the analytic
space V (I),x by
s
\
V (I),x = V (fi ),x .
i=1
This definition is independent of the choice of generators fi ofI, as is easily checked.

Any germ of an analytic space is a germ of type V (I), 0 for some ideal I ⊂
C {x1 , . . . ,xn }.
(4) Let (X, 0) be a germ of an analytic space. Then define

I (X,x) = {f ∈ On,x : (X,x) ⊂ V (f ), 0 }.
The inclusion here is an inclusion of germs. So f ∈ I (X,x) if there exists a repre-

sentative X of (X, 0) and an open neighborhood U of x such that X is an analytic
subset of U , f converges on U , and its restriction to X is the zero function.

Remark 3.4.3. We had to be a little careful with the definition of V (I),x . The defi-
nition of V (I),x as ∩f ∈I V (f ),x would not make sense. In fact, in general it does not
make sense to talk about infinite intersections of germs. Namely, if we have germs (Xi ,x)
for i ∈ I, and if representatives are defined on open sets Ui , then it is very well possible
that ∩i∈I Ui = {0}, if the set I has infinitely many elements. If I is a finite set however
then ∩i∈I Ui is still an open neighborhood of x. Note that in the definition of V (I),x
we already used the basic fact that I is finitely generated, which follows from the fact
that C {x1 , . . . ,xn } is Noetherian.
We have properties similar to the statements in 2.2.2, except that, keeping in mind
the remark above, we are only allowed to take finite unions and intersections! The proofs
are all similar, and left to the reader. We also have:
√
Theorem 3.4.4 (Nullstellensatz). I V (I),x = I.
It obviously suffices to prove the Nullstellensatz for x = 0. For germs of analytic
spaces the Nullstellensatz cannot be interpreted as a statement on maximal ideals. This
is because the power series ring is a local ring, that is, it just has one maximal ideal!
Moreover, it seems that also the idea of the proof of the Projection Theorem 2.2.8 does
not work in the local case. The point is (in the notation of the Projection Theorem)
that we first have to define representatives, and in particular generators h1 , . . . ,ht for
I ′ = I ∩ C {x1 , . . . ,xn−1 }, which converge on some open neighborhood U of 0. The
proof of the Projection Theorem provides us with a holomorphic function g on U , which
vanishes on X and such that g(a) 6= 0. So it follows that the germ of g is in I ′ , but what
is not clear is that g is in the ideal generated by the h1 , . . . ,ht all over U .
It is therefore no surprise that proofs of the Nullstellensatz for the power series ring
tend to be somewhat more difficult.
As in the proof of the Nullstellensatz in the algebraic case, it suffices to prove the
Nullstellensatz for prime ideals p. As the proof is completely similar, see 2.2.12, we leave
the proof to the reader, and only formulate the result.
Lemma 3.4.5. Suppose the Nullstellensatz holds for prime ideals. Then the Nullstellen-
satz holds for all ideals.
So from now on we will look at prime ideals. As in the proof of the Nullstellensatz
in the affine case, we will look at a Noether normalization,
C {x1 , . . . ,xk } ⊂ C {x1 , . . . ,xn }/p.

Put (X, 0) = V (p), 0 . Then we would like to show that the linear projection (X, 0) −→
(C k , 0) is surjective. This should be analogous to 2.2.10, where we showed that for π :
X −→ K r a Noether normalization, the map π is surjective. This was the essential step
in the proof of the Nullstellensatz. Note that we did, until now, not even define what a
map between germs of spaces is!
Definition 3.4.6. Let (X,p) and (Y,q) be two germs of topological spaces. Agerm of
a continuous map f : (X,p) −→ (Y,q) is defined as an equivalence class of maps f :
U −→ W , with f (p) = q, and where U and W are representatives of (X,p) and (Y,p)
respectively. Two such maps f1 : U1 −→ W and f2 : U1 −→ W are called equivalent if
they agree on an open neighborhood of p contained in U1 ∩ U2 .
The problem with germs of continuous maps is, that it makes in general no sense to
talk about the image. By this we mean that in general it might happen that for every
subgerm (Z,q) ⊂ (Y,q), there does not exist an open neighborhood U of p in X such that
(Z,q) is represented by f (U ). For an example, see Exercise 3.4.34. In particular, it does
not make sense to talk about surjectivity. For special kind of maps, so-called finite maps,
this problem does not occur. Let us define this notion.
Definition 3.4.7. Let f : X → Y be a continuous map between topological spaces.
(1) f is called closed if the image f (A) ⊂ Y is closed for all closed subspaces A ⊂ X.
(2) f is called quasi-finite if for all p ∈ Y the fiber f −1 (p) consists of a finite number
of points.
(3) f is called finite if it is both closed and quasi-finite.
The simplest example of a mapping which is quasi-finite but not finite probably is
the inclusion mapping C \ {0} → C . (Here C has the standard Euclidean topology.)
Lemma 3.4.8. Let X and Y be topological spaces, and f : X −→ Y be closed. Let
p ∈ X and let q = f (p). Assume that f −1 f (p) = {p}. Let A,B ⊂ X with p ∈ A, p ∈ B,
defining the same germ at p. Then (f (A),q) = (f (B),q), that is, f (A) and f (B) define
the same germ in q.
Proof. By the definition of a germ, we have to find an open subset V of q in Y with
V ∩ f (A) = V ∩ f (B).
Step 1. We first show that there exists a neighborhood basis of p in X such that for
each element U of this basis we have the equality U = f −1 (f (U )). So let W be an open
neighborhood of p. Then X \ W is closed, so that f (X \ W ) is closed by assumption.
Then T := Y \ f (X \ W ) is open. Define U = f −1 (T ). Then U is open and we have the
following properties:

(1) As U = f −1 (T ), it follows f (U ) = f ◦ f −1 (T ) ⊂ T . Hence f −1 f (U ) ⊂ f −1 (T ) =
U . As the converse inclusion is trivial, it follows that f −1 f (U ) = U .
(2) U ⊂ W , as, otherwise, there exists an x ∈ U , x 6∈ W . By definition of T , it
immediately follows from x 6∈ W that f (x) ∈ f (X \ W ), so that f (x) 6∈ T . But we
have f (U ) ⊂ T , so this is a contradiction.
(3) We have p ∈ U . To show this, note that as p ∈ W and f −1 (f (p)) = {p} it follows
f (p) 6∈ f (X \ W ) and then f (p) ∈ T = Y \ f (X \ W ). Thus p ∈ f −1 (T ) = U .
Step 2. Let W be an open neighborhood of p such that A ∩ W = B ∩ W . Take U ⊂ W

and T ⊂ Y as in Step 1. In particular U ∩ A = U ∩ B. We claim that
f (U ∩ A) = T ∩ f (A).
As f (U ) ⊂ T , the inclusion ⊂ is a tautology. To show the converse inclusion, take a point
a ∈ T ∩ f (A). Then a = f (b) for some b ∈ A. Moreover, a ∈ T , so that b ∈ f −1 (T ) = U .
This shows that b ∈ U ∩ A, showing the converse inclusion.
As similarly f (U ∩ B) = T ∩ f (B), and U ∩ A = U ∩ B, it follows that T ∩ f (A) =
T ∩ f (B). This is what we had to show.
The lemma is the basis for the following definition:
Definition 3.4.9. Let (X,p) and (Y,q) be two germs of topological spaces. Let f :
(X,p) −→ (Y,q) be a germ of a continuous map.
(1) f is called finite, if there exists a representative f : X −→ Y which is finite.

(2) Assume that f is finite. Then the image of f is given by Im(f ) := f (X),q , where
f : X −→ Y is a representative such that f −1 f (p) = {p} and f is finite.
Note that because of Lemma 3.4.8 Im(f ) is well-defined.
(3) Assume that f is finite. Then f is called surjective if Im(f ) = (Y,q). Note that f is
surjective, if and only if f : X −→ Y is surjective for suitable representatives such
that f is finite and f −1 f (p) = {p}.
The first example we give of a finite map is the following.
Lemma 3.4.10. Let P = y r + a1 y r−1 + . . . + ar be a polynomial, with coefficients
ai ∈ Ok . Consider an open neighborhood U ⊂ C k of 0 such that the power series a1 , . . . ,ar
converge. Define:
X := {P = 0} ∩ (U × C ).
Then the canonical projection π : X −→ U is finite.
Proof. As the map obviously is quasi-finite, we only have to show that π is closed.
Therefore, let A ⊂ X be a closed subset, and take a point p ∈ π(A). We have to show
that p ∈ π(A). As p ∈ π(A) we have a sequence
(pi ,yi ) i = 1,2, . . .
with pi ∈ π(A), yi ∈ C , (pi ,yi ) ∈ A, and limi→∞ pi = p. Because (pi ,yi ) ∈ A ⊂ X, we
have
yir + a1 (pi )yir−1 + . . . + ar (pi ) = 0 for i ≥ 1.
As the aj are holomorphic, in particular continuous, we have that limi→∞ aj (pi ) = aj (p).
Therefore, the sequence aj (pi ), i = 1,2, . . . is bounded. We claim that the sequence yi
is bounded,
too. Indeed, |yi | < 1 or |yi | ≥ 1. But in the second case, we obviously have
aj (pi )
yj−1 ≤ |aj (pi )|, so that:
i

ar (pi )
|yi | = r−1 + . . . + a1 (pi ) ≤ |ar (pi )| + . . . + |a1 (pi )| ,
yi
hence the sequence yi is bounded. There exists therefore a convergent subsequence
yi1 ,yi2 , . . ., converging to say y. As A is closed, the point (p,y) = limj→∞ (pj ,yij ) be-
longs to A. Therefore, p ∈ π(A), which is what we had to prove.
Typical picture of the zero set of a monic polynomial.
Lemma 3.4.11 (Continuity of Roots). Consider a Weierstraß polynomial:
P (x,y) = y r + a1 y r−1 + . . . + ar ; ai ∈ m ⊂ C {x1 , . . . ,xk }.
(1) Then for every arbitrary small neighborhood V of 0 in C there exists a small neigh-
borhood U of 0 in C k such that the power series a1 , . . . ,ar converge on U and such
that for every p in U the r zeros (counted with multiplicity) of
y r + a1 (p)y r−1 + . . . + ar (p) = 0
lie in V .
(2) Let π : C k+1 −→ C k be the projection on the first k coordinates. Then π :
V (P ), 0 −→ (C k , 0) is surjective.
(3) Suppose that P is irreducible. Let ∆ be the discriminant of P and let U ⊂ C k be an

open set such that a1 , . . . ,ar converge on U . Let X = {(x,y) ∈ U × C : P (x,y) = 0},
D = {x ∈ U : ∆(x) = 0} and let π : X −→ U be the projection on the first k
coordinates. Then X \ π −1 (D) = X.
U
Typical picture of the zero set of of a Weierstraß polynomial.
Proof.
(1) The proof runs as the proof of the previous lemma. Suppose the converse. Then there
exists a small neighborhood V of 0 in C such that for all neighborhoods U of 0 in C k on
which the power series a1 , . . . ,ar converge, there exists a point p ∈ U and a root of the
polynomial P (p,y) = 0, which is not in V . Take a sequence Ui of open neighborhoods of
0 whose intersection is {0}, and points pi ∈ Ui such that there is a root yi of P (pi ,yi ) = 0
which does not lie in V . So there exist a C > 0, infinitely many yi with |yi | ≥ C and
infinitely many pi ∈ C k such that {pi } converges to 0 and P (pi ,yi ) = 0. But then it
follows as in the proof of 3.4.10 that
1 1
|yi | ≤ C r−1 |ar (pi )| + C r−2 |ar−1 (pi )| + . . . + |a1 (pi )| .
By assumption limi→∞ aj (pi ) = 0 for all j. So we can make yi as small as we want, in

contradiction to |yi | ≥ C.
(2) Let U ⊂ C k be an open neighborhood of 0 such that the power series a1 , . . . ,ar
converge. Define:
X := {(x,y) ∈ U × C : P (x,y) = 0}.
The canonical map π : X −→ U is finite (Lemma 3.4.10) and, obviously, surjective.
Moreover, because P is a Weierstraß polynomial, we have π −1 (0) = 0. This implies that
V (P ), 0) −→ (C k , 0) is surjective, see 3.4.9(3).
(3) We have to prove that for any point (a,b) ∈ π −1 (D) there exists a sequence (ai ,bi ) ∈
X \ π −1 (D) with limi→∞ (ai ,bi ) = (a,b). By Theorem 3.1.14 there exists a sequence ai in
U \ D with limi→∞ ai = a. We consider P at the point (a,b). Here we can write P = U ·Q,
U ∈ C {x,y} a unit, Q a Weierstraß polynomial with respect to x = x − a and y = y − b.
Now we use (1) and obtain for all i that there exists an open neighborhood Vi of a such
that for ai ∈ Vi all roots of Q(ai ,y) = 0 satisfy |y| < 1i . Therefore, we may choose a
sequence bi such that (ai ,bi ) ∈ X converging to (a,b) and ai 6∈ D.
Remarks 3.4.12.
(1) Alternatively, Lemma 3.4.11 could have been proved by using Rouché’s Theorem
R ′
of function theory. Even simpler, one could use the integral ff , which counts
the number of zeroes of a holomorphic function in one variable and a continuity
argument. The lemma in fact is the starting point for an alternative proof of the
Weierstraß Division and Preparation Theorem, see the book of Gunning and Rossi,
[Gunning-Rossi 1965].
(2) We will use the Nullstellensatz to generalize 3.4.11 (3). In fact we will prove that,
if X is a small representative of V (p),0 for a prime ideal p ⊂ C {x1 , . . . ,xn }, and
h ∈ C {x1 , . . . ,xn }, h 6∈ I (X, 0), then X \ {h = 0} = X.
Lemma 3.4.13. Let I ⊂ C {x1 , . . . ,xn } be an ideal.

(1) Suppose that there exists an f ∈ I which is regular in xn . Let
p : C n −→ C n−1 ; (x1 , . . . ,xn ) 7→ (x1 , . . . ,xn−1 ).

Then there exists a representative X of V (I), 0 and an open neighborhood U of
0 ∈ C n−1 such that the map p : X −→ U is a finite map.
(2) Suppose more generally, that C {x1 , . . . xn }/I is a finitely

generated C {x1 , . . . ,xk }–
module. Then there exists a representative X of V (I), 0 and an open neighborhood
U of C k such that the projection p : X −→ U is a finite map.
Proof. By the Weierstraß Preparation Theorem, we may assume that f = up, where p is
a Weierstraß polynomial, and u is a unit. So we may assume that f is a Weierstraß poly-
n−1
nomial. By 3.4.10 there exists an open neighborhood U of 0 in C and a representative
W of V (f ), 0 such that p : W −→ U is finite. Consider a representative X ⊂ W of
V (I), 0 . As X is a closed subset of W (being the zero set of finitely many holomorphic
functions), it follows that p : X −→ U is finite too.
For the second statement, it follows that each of the classes of xj for j = k + 1, . . . ,n
satisfy an integral equation with coefficients in C {x1 , . . . ,xk }. Hence there exist Weier-
straß polynomials fj (xj ) for j ≥ k + 1, with coefficients in C {x1 , . . . ,xk } which are
contained in I. By Exercise 3.4.35 there exists an open neighborhood U ⊂ C k such
that the projection map p : X −→ U is finite for a suitable representative X of
V (fk+1 , . . . ,fn ), 0 . So we can argue as above to deduce the lemma.
We will generalize the second part of Lemma 3.4.11 in the Local Parametrization
Theorem 3.4.14. In particular we will prove

Let p ⊂ C {x1 , . . . ,xn } be a prime ideal and π : V (p), 0 −→ (C k , 0) be a
primitive Noether normalization. Then π is finite and surjective.
This is the basis to prove the Nullstellensatz.

Proof of the Nullstellensatz 3.4.4. (Using the Local Parametrization Theorem 3.4.14.) It
suffices to show the Nullstellensatz
for prime ideals p. We choose a primitive Noether
normalization π : V (p), 0 −→ (C k , 0) and use the fact that π is surjective. Then the
proof is completely analogous to the proof of the Nullstellensatz in the affine case, see
2.2.11.

Recall that we want to prove that for some Noether normalization V (p), 0 =
(X, 0) −→ (C k , 0) is surjective. This will be proved in two steps, namely we will first
show that (X, 0) projects surjectively onto a hypersurface V (P ), 0 in (C k+1 , 0), and
then we project further onto (C k , 0). The first projection is “generically one to one”.
This projection is an important tool in several proofs in this book, and is called the
Local Parametrization Theorem.

Theorem 3.4.14 (Local Parametrization Theorem). Let (X, 0) = V (p), 0 for p ⊂
C {x1 , . . . ,xn } a prime ideal and C {x1 , . . . ,xk } ⊂ C {x1 , . . . ,xn }/p be a primitive Noether
normalization (cf. 3.3.20). Let π : C n −→ C k+1 be the projection on the first k + 1
coordinates, p : C k+1 −→ C k be the projection on the first k coordinates. Then we have
the following statements.
(1) p ∩ C {x1 , . . . ,xk+1 } = (P ), for P ∈ C {x1 , . . . ,xk }[xk+1 ] a Weierstraß polynomial.

(2) Let ∆ be the discriminant of P with respect to xk+1 . Then
(a) there exist qk+2 , . . . ,qn ∈ C {x1 , . . . ,xk }[xk+1 ] such that Qj := ∆xj − qj ∈ p
for j = k + 2, . . . ,n;
(b) there exist f1 , . . . ,fs ∈ C {x1 , . . . ,xk }[xk+1 , . . . ,xn ] such that
i. (f1 , . . . ,fs ) = p in the ring C {x1 , . . . ,xn }.

ii. fj ∈ C {x1 , . . . ,xk }[xk+j ] are Weierstraß polynomials for j = 1, . . . ,n − k
and P,Qk+2 , . . . ,Qn ∈ {f1 , . . . ,fs };
iii. (P,Qk+2 , . . . ,Qn )C {x1 , . . . ,xk }∆ [xk+1 , . . . ,xn ] =
(f1 , . . . ,fs )C {x1 , . . . ,xk }∆ [xk+1 , . . . ,xn ].
(3) There is an open neighborhood U of 0 in C k such that P,∆,Qk+2 , . . . ,Qn ,f1 , . . . ,fs
converge on U × C n−k and for
X = {x ∈ U × C n−k : f1 (x) = · · · = fs (x) = 0}

Y = {y ∈ U × C n−k : P (y) = Qk+2 (y) = · · · = Qn (y) = 0}
X ′ = {x′ ∈ U × C : P (x′ ) = 0}
D = {x ∈ U : ∆(x) = 0}
the following statements hold:

(a) Y \ (p ◦ π)−1 (D) = X \ (p ◦ π)−1 (D);11
(b) π : X \ (p ◦ π)−1 (D) −→ X ′ \ p−1 (D) is biholomorphic;
(c) X \ (p ◦ π)−1 (D) is a complex submanifold of U × C n−k ;
(d) the map p ◦ π : X −→ U is surjective and finite and (p ◦ π)−1 (0) = {0}.
(4) The map p ◦ π : (X, 0) −→ (C k , 0) is finite and surjective.

(5) The map π : (X, 0) −→ V (P ), 0 is finite and surjective.
Before giving the proof of the Local Parametrization Theorem we give an example.
In this example all the power series are in fact polynomials. The neighborhood U can be
taken to be the whole of the complex vector space.
Example 3.4.15. Consider the map C −→ C 3 given by t 7→ (t3 ,t4 ,t5 ) = (x,y,z). Its
image can be calculated, see 2.3.18, to be the zero set of the ideal I = (y 2 − xz,x3 −
yz,z 2 − x2 y). A Noether normalization map is given by the projection on the x–axis.
Indeed modulo I one has y 3 = xyz = x4 , and z 3 = x2 yz = x5 . This gives the integral
equation z 3 − x5 = 0, and y 3 − x4 = 0.12 To show that this is a Noether normalization
one has to show that I ∩ C {x} is zero. For this, see Exercise 3.1.28. We in fact have
a primitive Noether normalization. Here y plays the role of xk+1 . This follows because
2
in the quotient field z = yx . The polynomial P is equal to y 3 − x4 . The discriminant
is ∆ = 27x8 . So the Local Parametrization Theorem says that x8 z is a polynomial in
y with coefficients in C {x}. Indeed, even xz = y 2 . So in this case the polynomial q is
equal to 27x7 y 2 . The space X ′ is given by y 3 − x4 = 0. For each a 6= 0, and each y = b
2
with b3 − a4 , there is a unique point on X, with z–coordinate z = ba . Thus the map π
in the Local Parametrization Theorem is indeed surjective. In fact, it is easily seen that
for x 6= 0, the analytic set is a complex submanifold.
n` qk+2 (p) ´ o
11 The equation implies that X \ (p ◦ π)−1 (D) = p, ∆(p)
, . . . , q∆(p)
n (p)
: p ∈ U × C , P (p) = 0 . This
(together with 3.3.23) is the reason to call 3.4.14 Local Parametrization Theorem.
12 A different way to see this is by applying the general Weierstraß Division Theorem 3.2.10. Consider
the C {x}–module C {x,y,z}/I. Divide out the maximal ideal (x) of C {x}. We get the C –vector space
C {y,z}/(y 2 ,yz,z 2 ), which is finite-dimensional. A basis is given by the classes of 1,y,z.
Proof of the Local Parametrization Theorem 3.4.14.
Step 1. We first show (1) and (2a). We apply the Finiteness of Normalization Theo-
rem 1.5.19. We obtain that p ∩ C {x1 , . . . ,xk }[xk+1 ] = (P ), where P is the minimal
polynomial of the element xk+1 mod p in the field extension Q(C {x1 , . . . ,xn }/p) ⊃
Q(C {x1 , . . . ,xk }). By Exercise 3.3.32, P is a Weierstraß polynomial. Using the Weier-
straß Division Theorem 3.2.3, we obtain p ∩ C {x1 , . . . ,xk+1 } = (P ), as, otherwise, we
would find a nonzero element in p which is a polynomial in xk+1 of degree smaller than
the degree of P . This is in contradiction to the fact that P is the minimal polynomial
of the field extension. This proves (1). To prove (2a) we apply again 1.5.19. So for all
f ∈ On there exists qf ∈ C {x1 , . . . ,xk }[xk+1 ] with ∆ · f − qf ∈ p. In particular, this holds
for xk+2 , . . . ,xn . Therefore, there exist polynomials qk+2 , . . . ,qn ∈ C {x1 , . . . ,xk }[xk+1 ]
such that Qj := ∆xj − qj ∈ p for j = k + 2, . . . ,n. This shows (2a).
Step 2. We now prove (2b). Consider the minimal polynomials Pj ∈ C {x1 , . . . ,xk }[xj ] of
xj for j = k + 2, . . . ,n. Thus Pk+1 := P .
Step 2a. We first show that p has a system of generators in C {x1 , . . . ,xk }[xk+1 , . . . ,xn ].
Indeed, it follows that the extension C {x1 , . . . ,xk } ⊂ C {x1 , . . . ,xn }/(Pk+1 , . . . ,Pn ) is
finite. We put P i := Pi (0, . . . ,0,xk+1 , . . . ,xn ). It then easily follows that the vector space
C {xk+1 , . . . ,xn }/(P k+1 , . . . ,P n ) is generated by monomials in xk+1 , . . . ,xn . It follows
from Corollary 3.2.12 that every element in C {x1 , . . . ,xn } is modulo (Pk+1 , . . . ,Pn ) equiv-
alent to an element in C {x1 , . . . ,xk }[xk+1 , . . . ,xn ]. This then certainly holds modulo p.
Hence we can choose generators f1 , . . . ,fs of p such that
• f1 = Pk+1 = P, . . . ,fn−k = Pn ;
• fn−k+1 , . . . ,fs ∈ C {x1 , . . . ,xk }[xk+1 , . . . ,xn ];
• Qk+2 , . . . ,Qn ∈ {f1 , . . . ,fs }.
Step 2b. Put
J∆ := (P,Qk+2 , . . . ,Qn )C {x1 , . . . ,xk }∆ [xk+1 , . . . ,xn ]

I∆ := (f1 , . . . ,fs )C {x1 , . . . ,xk }∆ [xk+1 , . . . ,xn ].
To prove iii of 2b, We have to show J∆ = I∆ . The inclusion J∆ ⊂ I∆ is obvious and,

therefore, we have a surjection
(3.17) C {x1 , . . . ,xk }∆ [xk+1 , . . . ,xn ]/J∆ ։ C {x1 , . . . ,xk }∆ [xk+1 , . . . ,xn ]/I∆ .
q (x ,...,x )
The equation Qj = 0 says that xj = j 1 ∆ k+1 , so we can eliminate the variables
xk+2 , . . . ,xn in C {x1 , . . . ,xk }∆ [xk+1 , . . . ,xn ]/J∆ . After having done that, formula (3.17)
gives a surjection
C {x1 , . . . ,xk }∆ [xk+1 ]/(P ) ։ C {x1 , . . . ,xk }∆ [xk+1 , . . . ,xn ]/I∆ .
We claim that is in fact a bijection, that is, I∆ ∩ C {x1 , . . . ,xk }∆ [xk+1 ] = (P ). As I

generates p, it follows from the first part of the Theorem that I ∩ C {x1 , . . . ,xk }[xk+1 ] =
(P ). As (P ) is a prime ideal, and ∆ ∈/ (P ), the claim follows.
Step 3. The proof of (3a) now is easy. Let f1 , . . . ,fs ∈ C {x1 , . . . ,xk }[xk+1 , . . . ,xn ] be the
generators of p we constructed in Step 2. By (2b) we have for some b = b(i):
n
X
(3.18) ∆b fi = αj (fi )Qj + α(fi )P
j=k+2
for certain αj (fi ),α(fi ) ∈ C {x1 , . . . ,xk }[xk+1 , . . . ,xn ]. Choose U ⊂ C k an open neigh-
borhood of 0 such that P,∆,Qk+2 , . . . ,Qn , α(fi ),αj (fi ) converge on U × C n−k for all
i,j. Let p ∈ Y with ∆(p) 6= 0 be given. Then from (3.18) and P (p) = Qj (p) = 0 for
j = k + 2, . . . ,n, it follows that fi (p) = 0 for i = 1, . . . ,s. This exactly means that
Y \ (p ◦ π)−1 (D) = X \ (p ◦ π)−1 (D).
Step 4. We now prove (3b) and (3c). Consider a point a ∈ X ′ \ p−1 (D). From 3.3.23 X ′
is in a small neighborhood U ′ of a a complex submanifold of U ′ × W , given by the graph
of a function xk+1 = α(x1 , . . . ,xn ). By definition of Qj = ∆xj − qj . Therefore, there is a
unique point of X \ (p ◦ π)−1 (D) mapping to a. If the coordinates of a are (a1 , . . . ,ak+1 )
the unique point in π −1 (p) has coordinates

qk+2 (a1 , . . . ,ak+1 ) qn (a1 , . . . ,ak+1 )
a1 , . . . ,ak ,ak+1 , ,..., .
∆(a1 , . . . ,ak ) ∆(a1 , . . . ,ak )
So we have an inverse to the map π : X \ (p ◦ π)−1 (D) −→ X ′ \ p−1 (D), which we see to
be biholomorphic. Indeed, locally in U ′ × W × C n−k−1 , the set Y \ (p ◦ π)−1 (D) is given
by the set of equations
qk+2 (x1 , . . . ,xk+1 ) qn (x1 , . . . ,xk+1 )
xk+1 − α(x1 , . . . ,xk ) = xk+2 − = · · · = xn − = 0,
∆(x1 , . . . ,xk ) ∆(x1 , . . . ,xk )
and is therefore a complex submanifold.
Step 5. We prove (3d), (4) and (5). By Lemma 3.4.13 the map p ◦ π : X −→ U is finite.
The map p : X ′ \ p−1 (D) −→ U \ D is, obviously, surjective and because of (3b) the map
p ◦ π : X \ (p ◦ π)−1 (D) −→ U \ D is surjective. As p ◦ π is closed, the closure of U \ D is
in p ◦ π(X). Using Theorem 3.1.14 we obtain that U \ D = U and, therefore, π(X) = U .
Now (4) and (5) are immediate consequences (cf. 3.4.9 and 3.4.13).

Remarks 3.4.16. Let p be a prime ideal, and (X, 0) = V (p), 0 .
(1) Note that in the proof of the Local Parametrization Theorem, we showed that the
number of points in the general fiber of a Noether normalization π : X −→ U is
equal to the degree of P , that is, the degree of the field extension
Q(Ok ) ⊂ Q(OX, 0 ).
In fact, this holds generally for “finite maps between irreducible germs of analytic
spaces”, as will be shown in the exercises.
(2) We showed that locally, at a point a ∈ X \ (p ◦ π)−1 (D), the ideal I (X,a) ⊂
On,a is generated by Qk+2 , . . . ,Qn ,P . Moreover, On,a /I (X,a) is isomorphic to
Ok . It follows that if (f1 , . . . ,fs ) = I (X, 0), the ideal of (X, 0), then there exists a
neighborhood U of 0 in X such that the fi converge on U and for all a ∈ U ∩(X \D)
the ideal I (X,a) is generated by f1 , . . . ,fs .13
13 In fact, this holds for all a ∈ U . This is the Coherence Theorem of Oka-Cartan to be proved in Chapter
6.
(3) In the notation of the Local Parametrization Theorem, write OX ′ ,0 = Ok+1 /(P ).
The cokernel K := OX,0 /OX ′ ,0 is an Ok –module. It was shown in the proof of the
Local Parametrization Theorem, that there exists an a ∈ N such that ∆a · K = (0).
In particular, we can view K as an Ok /(∆a )–module.
Definition 3.4.17. Let (X,x) be a germ of an analytic space. Then (X,x) is called
irreducible if from (X,x) = (X1 ,x) ∪ (X2 ,x), with (X1 ,x) and (X2 ,x) germs of analytic
spaces it follows that either (X,x) = (X1 ,x) or (X,x) = (X2 ,x).
Corollary 3.4.18. Let (X,x) be a germ of an analytic space.
(1) (X,x) is irreducible if and only if I (X,x) is a prime ideal.
(2) Let (X,x) ⊂ (C n ,x) be irreducible, and h ∈ On , h 6∈ I (X,x). Then for small
enough representatives X \ {h = 0} = X.
(3) There is a, up to permutation, unique decomposition (X,x) = (X1 ,x) ∪ · · · ∪ (Xr ,x),
with (Xi ,x) irreducible and (Xj ,x) 6⊂ (Xi ,x) for i 6= j. This is called the irreducible
decomposition of (X,x). The (Xi ,x) are called irreducible components of (X,x).
Proof. The proof runs as in the algebraic case, see 2.2.15, 2.2.16 and 2.2.17. But note
that in the proof the Nullstellensatz is used!
Definition 3.4.19. Let (X,x) ⊂ (C n ,x) be a germ of an analytic space. A germ of
an analytic function f : (X,x) −→ (C ,y) is a germ of a map f : (X,x) −→ (C ,y)
such that some representative is the restriction to X of an analytic function on an open
neighborhood of x in C n . As is easily shown, the germs of analytic functions on (X,x)
form a C –algebra, which we denote by OX,x . This is called the ring of analytic functions
on (X,x). Sometimes OC n , 0 , which is equal to C {x1 , . . . ,xn } is denoted by On . A germ
of an analytic map or simply map, ϕ = (f1 , . . . ,fm ) : (X,x) −→ (Y,y ⊂ (C m ,y) is
similarly defined. The map ϕ : (X,x) −→ (Y,y) is called an isomorphism if ϕ has a two-
sided inverse ψ : (Y,y) −→ (X,x) which is an analytic map. (X,x) and (Y,y) are called
isomorphic if there exists an isomorphism ϕ : (X,x) −→ (Y,y).
The following lemma is easy, and left to the reader.
Lemma 3.4.20.
(1) Let (X,x) ⊂ (C n ,x) be a germ of an analytic space, and I (X,x) be the ideal of X.
Then OX,x = On,x /I (X,x).
(2) Let (X,x) ⊂ (C n ,x) be a germ of a submanifold. Then OX,x ∼
= C {x1 , . . . ,xk } for
some k.
(3) Let ϕ : (X,x) −→ (Y,y) be a germ of an analytic mapping. Then by composition ϕ
induces a map of C –algebras
ϕ∗ : OY,y −→ OX,x : f 7→ f ◦ ϕ.
Note that (ϕ ◦ ψ)∗ = ψ ∗ ◦ ϕ∗ for two germs of analytic mappings ϕ and ψ.

The converse of the third part of the lemma also holds:
Theorem 3.4.21. Let (X,x) ⊂ (C n ,x) and (Y,y) ⊂ (C m ,y) be germs of analytic spaces.
Let α : OY,y −→ OX,x be a C –algebra homomorphism. Then there exists a unique germ
of an analytic mapping ϕ : (X,x) −→ (Y,y) with ϕ∗ = α.
Proof. . Without loss of generality, we may assume that x = 0 and y = 0. In 3.2.9 it was
proved that α(mY,0 ) ⊂ mX,0 . As α is a ring homomorphism, we get
α(mkY,0 ) ⊂ mkX,0
for all k > 0. Take generators w1 , . . . ,wm of the maximal ideal of Om . We have a surjective
map Om −→ OY,0 = Om /I (Y, 0), sending wi to w i . Let
f1 = α(w 1 ), . . . ,fm = α(w m ).
So the fi are elements of OX,0 . Define
ϕ = (f1 , . . . ,fm ) : (X, 0) −→ (C m , 0).
Hence we get a map ϕ∗ : Om −→ OX,0 . Moreover consider the composition α
e : Om −→
α
OY,0 −→ OX,0 . We want to prove the following:
(1) ϕ∗ = α
e.

(2) ϕ (X, 0) ⊂ (Y, 0).
(1) By construction ϕ∗ (wi ) = α e(wi ). As they are both C –algebra homomorphisms, it
follows that ϕ∗ (g) = α
e(g) for all polynomials g ∈ C [w1 , . . . ,wm ]. But, we want to show
that ϕ∗ (g) = α
e(g) for all g ∈ Om . Given such a g, we can, for all k ∈ N write
g = gk + gk′ ,
where gk is a polynomial of degree smaller than k and gk′ ∈ mk . (Here m is the maximal
ideal of Om .) Now
ϕ∗ (gk ) = α
e(gk ).
As gk is a polynomial, and both ϕ∗ and α
e map mk into mkX,0 , it follows that
ϕ∗ (g) − α
e(g) = ϕ∗ (gk′ ) − α
e(gk′ ) ∈ mkX,o .
As this holds for all k it follows that ϕ∗ (g) − αe(g) ∈ ∩k mkX,0 . But as OX,0 is a Noetherian
k
local ring, it follows that ∩k mX,0 = 0, by Krull’s Intersection Theorem 1.3.5. Therefore,
ϕ∗ (g) = α
e(g).

(2) Because (Y, 0) = V I (Y, 0) , 0 , it suffices to show that for all g ∈ I (Y, 0), the
map
g ◦ ϕ : (X, 0) −→ (C , 0),
e is induced by α : OY,0 −→ OX,0 , the ideal I (Y, 0) is in the kernel
is the zero map. As α
of ϕ∗ : On −→ OX,0 . So the claim follows.
The uniqueness of ϕ is clear.
Corollary 3.4.22. Two germs of analytic spaces (X,x) and (Y,y) are isomorphic if
and only if OX,x and OY,y are isomorphic. So the study of germs of analytic spaces is
equivalent to the study of reduced analytic algebras. In particular, a germ of a complex
submanifold is isomorphic to (C k , 0) for some k.
Proof. The Nullstellensatz says that there is a 1 − 1 correspondence between reduced

analytic algebras and germs of analytic spaces. The previous theorem implies that the
isomorphism classes of reduced analytic algebras correspond to isomorphism classes of
germs of analytic spaces.
Remark 3.4.23. More generally, we will later on consider germs of complex spaces
(X,x). By this, we mean nothing else than analytic algebras OX,x which are, in general,
nonreduced. They pop up naturally, when one considers intersections, as we will see later.
Associated to such a germ of a complex space (X,x) is its reduction (Xred ,x), with local
ring OXred ,x the reduction of OX,x , obtained by dividing out the nilpotent elements.
We now characterize finite maps algebraically.
Theorem 3.4.24. Let f : (X,x) −→ (Y,y) be a map between germs of analytic spaces.
Then the following conditions are equivalent:
(1) f is finite.
(2) OX,x a finitely generated OY,y –module.
(3) OX,x /mY,y OX,x is a finite-dimensional C –vector space.
(4) f −1 (y) = {x}.
Proof.
(1) =⇒ (4) is easy. As f −1 (y) consist of a finite number of points, we can make X so
small that it just contains x.
(4) =⇒ (3). As f −1 (y) = {x}, the zero set of mY,y OX,x is {x}, and it follows from the
Nullstellensatz, see Exercise 3.4.32, that OX,x /mY,y OX,x is a finite-dimensional vector
space.
(3) ⇐⇒ (2). This is the general Weierstraß Division Theorem, see 3.2.10.
(2) =⇒ (1). As (Y,y) ⊂ (C n , 0) for some n, that is, we have a surjection OC n , 0 ։ OY,y ,
we may reduce to the case (Y,y) = (C n , 0). This case was treated in 3.4.13(2).
Example 3.4.25. A hypersurface singularity by definition is a germ of an analytic space
V (f ), 0 defined by a single square free f ∈ m ⊂ C {x1 , . . . ,xn }. The isomorphism class
of special such f have special names.
• Ak : f (x1 , . . . ,xn ) = xk+1
1 + x22 + . . . + x2n , k ≥ 1,
• Dk : f (x1 , . . . ,xn ) = x1 x22 + x1k−1 + x23 + . . . + x2n , k ≥ 4,
• E6 : f (x1 , . . . ,xn ) = x31 + x42 + x23 + . . . + x2n ,

• E7 : f (x1 , . . . ,xn ) = x31 + x1 x32 + x23 + . . . + x2n ,
• E8 : f (x1 , . . . ,xn ) = x31 + x52 + x23 + . . . + x2n .
Also for k = 0 the A0 is a germ of an analytic space, for which the local ring is isomorphic
to C {x2 , . . . ,xn }. So the A0 “singularity” defines a germ of a smooth space. The A-D-E–
singularities are the “simplest” singularities, and will be studied further in chapter nine.
We draw pictures of the Ak and Dk for n = 2 and small k.
A1 A2 A3
D4 D5 D6
Definition 3.4.26. Let f ∈ C {x1 , . . . ,xn }.
∂f ∂f
(1) We define the Jacobian ideal by J(f ) := ( ∂x 1
, . . . , ∂x n
).
(2) The Milnor algebra of f by definition is the C –algebra C {x1 , . . . ,xn }/J(f ). The
Milnor number of f , µ(f ), by definition is the C –vector space dimension of the
Milnor algebra:
µ(f ) = dimC C {x1 , . . . ,xn }/J(f ) .

(3) The Tjurina algebra of f is by definition the C –algebra C {x1 , . . . ,xn }/ f,J(f ) .
The Tjurina number of f , τ (f ) by definition is the C –vector space dimension of
the Tjurina algebra.
The Milnor number and Tjurina number might be infinite.

Theorem 3.4.27. Let f and g be in C {x1 , . . . ,xn }. Suppose that V (f ), 0 and V (g), 0
are isomorphic. Then the Tjurina algebras of f and of g are isomorphic.14 In fact, if ϕ
is an automorphism of C {x1 , . . . ,xn } such that ϕ (f ) = (g), then ϕ (f ) + J(f ) =
(g + J(g)). In particular the Tjurina numbers τ (f ) and τ (g) are equal.
Proof. Let ϕ = (ϕ1 , . . . ,ϕn ) : C {x1 , . . . ,xn } −→ C {x1 , . .. ,xn } be an automorphism

such that (ϕ(f )) = (g). It is not difficult to see that g,J(g) = ug,J(ug) for a unit u.
Therefore, we may assume that ϕ(f ) = g. The chain rule says that for all i
∂g X ∂f ∂ϕ
j
X ∂f ∂ϕj
= ◦ϕ · = ϕ · .
∂xi j
∂xj ∂xi j
∂xj ∂xi
So we see that J(g) ⊂ ϕ(J(f )). As ϕ−1 is also an isomorphism, it follows by symmetry
J(f ) ⊂ ϕ−1 (J(g)). Compose thiswith ϕ. We get ϕ(J(f )) ⊂ J(g). As moreover ϕ (f ) =
(g), it follows that ϕ (f ) + J(f ) ⊂ (g + J(g)). The converse inclusion follows because
of symmetry.
14 It is a famous theorem of Mather and Yau that the converse holds in case the Tjurina number is finite,
see 9.1.8.
Example 3.4.28. The Tjurina numbers of the Ak ,Dk and Ek singularities are equal to k.
The singularities Ak and Al are isomorphic if and only if k = l. Now let (X, 0) = V (f ), 0
for f 6= 0 be a germ of a regular point. Then one calculates directly that τ (f ) = 0.
In particular, all hypersurface singularities V (f ), 0 with 0 < τ (f ) < ∞ are genuine
singularities. So now we proved that the Ak for k ≥ 1 are indeed singularities. Similarly
Dk and Dl are isomorphic if and only if k = l, and these are indeed singularities.
It is also true that the Milnor number is an invariant of the hypersurface singular-
ity (X, 0). A proof however, is surprisingly much more difficult, and will be proved as
Corollary 6.4.10.
A1 –singularities can conveniently be described by the Tjurina, or Milnor number.

Theorem 3.4.29. Let (X, 0) = V (f ), 0 be a hypersurface singularity. Then µ(f ) =
τ (f ) = 1 if and only if (X, 0) is isomorphic to an A1 –singularity.
Proof. Suppose f ∈ m \ m2 . Then µ(f ) = τ (f ) = 0, as one easily checks. Hence we
may assume that f ∈ m2 , so that τ (f ) ≥ 1. A direct calculation shows that for an
A1 –singularity µ(f ) = τ (f ) = 1.
Conversely, as τ (f ) ≤ µ(f ) it follows that either µ(f ) = 0 or 1. But µ(f ) = 0 cannot
occur,Pas f ∈/ m \ m2 . So we may assume that µ(f ) = 1. We may write (nonuniquely)
f = i,j xi xj Hij , and we may assume that Hij is symmetric. We put hij = Hij (0).
Then
∂f X
=2 xj hij modulo m2 .
∂xi j
As µ(f ) = 1, it follows that J(f ) = m = (x1 , . . . ,xn ). So there exist αki with
X ∂f
xk = αki .
i
∂xi
Putting aij = αij (0) it follows that

X
xk = 2 aki hij xj .
i,j
P 2
So it follows that i aki hij = 12 δkj so that hij = ∂x∂i ∂x
f
j
(0) is an invertible matrix.
The theorem therefore follows from the famous Morse Lemma, to be proved next.
Lemma
2 3.4.30
(Morse Lemma). Let f ∈ C {x1 , . . . ,xn }, f ∈ m2 and suppose that
∂ f
∂xi ∂xj (0) has maximal rank n. Then there exists an automorphism ϕ of C {x1 , . . . ,xn }
such that ϕ(f ) = x21 + . . . + x2n .
Proof. We write
n
X
f= xi xj Hij (x1 , . . . ,xn ).
i,j=1
2
We may suppose that Hij = Hji for all i,j. The rank of ∂x∂i ∂x f
j
(0) is equal to the rank
of Hij (0). It follows from linear algebra that this rank is independent of the coordinates.
By induction on s we suppose we found coordinates (y1 , . . . ,yn ) such that
X
f (y1 , . . . ,yn ) = y12 + . . . + ys−1
2
+ yi yj Hij (y1 , . . . ,yn ).
i,j≥s
The case s = 0 is trivial. From the assumption that we have maximal rank, it follows
that at least one of the Hij (0) is not equal to zero. After a linear coordinate change
in the ys , . . . ,yn we may suppose that Hss (0) 6= 0. √ One can, therefore, by the Implicit
Function Theorem, consider the square root g := Hss which is in fact a unit. We take
new coordinates
1 X His (w1 , . . . ,wn )
yi = wi for i 6= s; ys (w1 , . . . ,wn ) = ws − 12 wi .
g(w1 , . . . ,wn ) i>s
Hss (w1 , . . . ,wn )
It follows from the Inverse Function Theorem that this is an invertible map. Now one
calculates that f (w1 , . . . ,wn ) is of the desired form.
Exercises
3.4.31. Suppose X ⊂ C n , and Y ⊂ C m are affine varieties. Let ϕ : X −→ Y be a morphism
and suppose that via ϕ∗ the coordinate ring K[X] is a finitely generated K[Y ]–module. Let
x ∈ X, and y = f (x) ∈ Y . Show that for the germs of the analytic spaces (X,x) and (Y,y) the
ring OX,x is a finitely generated OY,y module.
(Hint: Use the general Weierstraß Division Theorem 3.2.10.)
` ´
3.4.32. Let I ⊂ C {x1 , . . . ,xn } be an ideal. Use the Nullstellensatz to prove that V (I), 0 = {0}
if and only if dimC C {x1 , . . . ,xn }/I < ∞. Prove that this is the case if and only if I is an m–
primary ideal. Here m = (x1 , . . . ,xn ).
3.4.33. Prove that the relation in definition 3.4.1 is an equivalence relation.
3.4.34. Let f : (X,p) −→ (Y,q) be a germ of a continuous map. In general, it does not make
sense to talk about the image of the map f . Indeed, look at the map
f : R2 −→ R2 , (x,y) 7→ (z,w) = (x,xy).
Show that there exist arbitrary small representatives X of (X,p) and Y of (Y,q) such that for
all U open in R2
U ∩ f (X) 6= U ∩ f (Y ).
So it is impossible to define the image as the class of the image of a small enough representative.
Note that in this example, f is not finite.
3.4.35. Formulate and prove Lemma 3.4.10 and Lemma 3.4.11 for the case of finitely many
Weierstraß polynomials. (For every Weierstraß polynomial one takes a new variable).
3.4.36. Let X,Y be topological spaces, f : X −→ Y be a closed map. Then for each open
neighborhood U in X of a fiber f −1 (y) there exists an open neighborhood V of y in Y such that
f −1 (V ) ⊂ U . Prove this.
3.4.37. Let X be an irreducible affine variety. Formulate and prove a Local Parametrization
Theorem for X.
3.4.38. Let (X,x) be a germ of ` an analytic
´ space, OX,x be its local ring. Define for I ⊂ OX,x
the germ of the analytic space V (I),x ⊂ (X,x). For a subgerm (A,x) ⊂ (X,x), define I (A,x).
Prove the standard properties of these operations, including the Nullstellensatz.
` ´
3.4.39. Let (X, 0) = V (f ), 0 be a hypersurface singularity. One says that (X, 0) has an isolated
singularity, if there exists a representative of X such that X \ {0} is a complex submanifold.
Using the Nullstellensatz, show that (X, 0) has an isolated singularity if and only if τ (f ) < ∞.
3.4.40. Let ϕ : (X,x) −→ (Y,y) be finite map between germs of analytic spaces. The map ϕ is
called generically s to 1 if there exists a hypersurface (D,y) in (Y,y) such that for all small enough
representatives Y of (Y,y) and D of (D,y), the set π −1 (p) has cardinality s for all p ∈ X \ D.
(1) Show that if ϕ is generically s to 1 for some s ≥ 1, then ϕ is surjective.
(2) Show that ϕ∗ induces an isomorphism of total quotient rings ϕ∗ : Q(OY,y ) ∼= Q(OX,x ) if
and only ϕ is generically 1 − 1.
(Hint: First reduce to the case (X,x) and (Y,y) irreducible. Furthermore, if (Y,y) −→
(C k , 0) is a primitive Noether normalization of (Y,y), then the induced map (X,x) −→
(C k ,0) is a primitive Noether normalization of (X,x). Apply the Local Parametrization
Theorem.)
(3) Suppose that OY,y is a normal ring. Show that ϕ is generically s to 1, where s is equal to
the degree of the field extension Q(OX,x ) ⊃ Q(OY,y ).
(Hint: First recall from 1.5.7 that OY,y is an integral domain. Then check that in the proof
of the Local Parametrization Theorem we can replace C {x1 , . . . ,xk } by OY,y .)
(4) Show that this result can be generalized to general (Y,y) as soon as we know that a
“normalization” exists, see Section 4.4. This remark also holds for the next exercise.
This final item also holds for the next exercise
3.4.41. Let ϕ : (X,x) −→ (Y,y) be a finite analytic map.

(1) Suppose that ϕ∗ : OY,y −→ OX,x is injective. Show that ϕ is surjective, in the following
two cases.
(a) (X,x) and (Y,y) are irreducible, and ϕ∗ induces an isomorphism on quotient fields.
(b) OY,y is a normal ring.
(2) Suppose ϕ is surjective. Show that ϕ∗ is injective.
3.4.42.
(1) Let (X,x) and (Y,y) be isomorphic germs of analytic spaces. Show that for all k ∈ N the
C –algebras OX,x /mkX,x and OY,y /mkY,y are isomorphic. In particular they are isomorphic
as C –vector spaces.
(2) Reprove that Ak is isomorphic to Al if and only if k = l. In fact, show that all of the
A-D-E–singularities give different isomorphism classes.
3.4.43. Let (X, 0) ⊂ (C n , 0) and (Y, 0) ⊂ (C m , 0) be germs of analytic spaces, and let ϕ :
(X, 0) −→ (Y, 0) be a germ of an analytic map. Define the graph Graph(ϕ) ⊂ (C n+m , 0), which
is isomorphic to (X, 0). Show that we can view ϕ as the projection on the second factor.
124
4 Further Development of Analytic

Geometry
In this chapter we will continue with the basics of the theory of germs of analytic spaces
(and of affine varieties). Probably the most important invariant of (irreducible) germs of
analytic spaces is the dimension. There are various ways to introduce it. We decided to
give, in fact, four definitions of dimension. We introduce this many of them, so that we
have a flexible development of the theory.
Three of these definitions are given in Section 4.1, namely the Weierstraß dimension,
the Krull dimension and the Chevalley dimension. They all have a geometric motivation.
The Weierstraß dimension of a germ of an analytic space (X,x) is given by considering
a Noether normalization (X,x) −→ (C k , 0). It turns out that k is independent of the
Noether normalization of (X,x). This map is surjective and has finite fibers, so it should
be geometrically clear to the reader that (X,x) has the same dimension as (C k , 0), which
should be k. The Krull dimension is defined as the maximal length k of a chain of
irreducible germs (X0 ,x) $ (X1 ,x) $ · · · $ (Xk ,x) ⊂ (X,x). The Chevalley dimension
is defined by the minimal number k of hypersurfaces (H1 ,x), . . . ,(Hk ,x) one has to cut
(X,x) with in order to get a point. The main part of the first section is devoted to the
fact that these three definitions coincide. As suggested by the definition of Chevalley
dimension, we will have to take intersections. Now it might happen that, even if one
starts with a reduced space (X,x), the intersection with every hypersurface is nonreduced.
Algebraically, this means it might happen that for a reduced local ring (R,m) and every
f ∈ m the ring R/(f ) might be nonreduced. This is why we have to consider germs of
complex spaces, that is, nonreduced spaces. Algebraically, this just means that we study
analytic algebras.
Our discussion on dimension applies just as well to algebraic sets. One has to keep
in mind however, that dimension is a local property. For example, take X = V (zx,zy)
and the point p on the line as indicated in the following figure.
Then one would like to say that the dimension of X at the point p is one. If, however, p
is as in one of the following two pictures,
125
p
p
then the dimension of X at p is should be two. Therefore, when discussing dimension

of algebraic sets, we will consider localizations K[V ]m . A Noether normalization for this
case is discussed in Exercise 2.2.35.
In Section 4.2 we study a fourth characterization of dimension, by means of the

Hilbert-Samuel polynomial. This definition is less geometrically motivated, but it has the
advantage that in many interesting examples it can be calculated by means of a computer.
The idea of the Hilbert-Samuel polynomial is to “count” the number of functions defined
on the germ of the analytic space
(X,x). More precisely, we will study the behavior of the
function d 7→ dimC OX,x /md , the so-called Hilbert-Samuel function. The main result is
that for d ≫ 0, the Hilbert-Samuel function behaves like a polynomial, whose degree is
equal to the dimension of (X,x). As it is not more difficult, we will study more generally
the Hilbert-Samuel function of an arbitrary local ring, or even finitely generated modules
over a local ring. In order to prove that the degree of the Hilbert-Samuel polynomial is
smaller than the (Chevalley) dimension, we have to study even more generally Hilbert-
Samuel functions with respect to an m–primary ideal q, that is, we have to study the
function k 7→ dimC OX,x /qd .
It turns out that also the leading coefficient of the Hilbert-Samuel polynomial has
a nice geometrical interpretation. More precisely, if k is the dimension of (X,x), then k!
times the leading coefficient of the Hilbert-Samuel polynomial is called the multiplicity of
(X,x). If our space (X,x) has a representative X in a (small) open subset U in C n , then
the multiplicity is the number of intersections points of X with a general linear space
of dimension n − k. We have already studied this situation for hypersurfaces, see 3.4.11.
There we showed that the intersection of a general line near zero with the zero set of a
reduced power series of order d has d intersection points. It is, in fact, not difficult to
show that the multiplicity in this case is equal to d. The general case will be reduced to
the hypersurface case by using the Local Parametrization Theorem. Like the dimension,
the multiplicity can, in many cases, be calculated by computer.
In Section 4.3 we will consider tangent spaces, regular and singular points. We al-
ready defined in Chapter 3 what a smooth point of an analytic space is, see Definition
3.3.11. The first result is that the set of singular points is contained in a proper ana-
lytic subset. Essential use is made of the Local Parametrization Theorem. We have to
postpone the proof that the set of singular points in fact is an analytic subset to Chap-
ter 6. However, as a preliminary version, we will prove the Jacobian criterion. This says
that the set of singular points is contained in the analytic subset given by the maximal
minors of the Jacobian matrix. The fact that it is equal, uses the Coherence Theorem
of Oka-Cartan, which is a deep theorem, to be proved in Chapter 6. We will introduce
126 4 Further Development of Analytic Geometry
the Zariski tangent space and prove that singular points can be characterized by those
points for which the embedding dimension, which is equal to the vector space dimension
of the Zariski tangent space, is not equal to the dimension of the space. Finally we will
prove the general Jacobian criterion, which gives a statement on when a general point of
an analytic subset is smooth.
In the final section of this chapter we study the normalization of a germ of an analytic
space. In case (X,x) is irreducible, this is a normal analytic space (X,ee x) together with a
finite degree one map n : (X,e e x) −→ (X,x). For reducible (X,x), we need the notion of a
multi-germ. First of all, we will show that a normalization (X,ee x) of a space (X,x) always
exists. The proof uses the Finiteness of Normalization Theorem 1.5.19, Corollary 3.3.25,
and Corollary 3.3.26. These are used to show that the normalization of OX,x is a finite
direct sum of analytic algebras. We continue to study properties of normal analytic spaces.
It will be shown that the singular points of a normal analytic space are contained in a
subset of codimension at least two. In particular, normal curve singularities are smooth.
As the normalization always exists, this means that for irreducible curve singularities
(X,x) there exists a germ of a finite, generically 1 − 1 map n : (C , 0) −→ (X,x). Hence,
irreducible curve singularities can be parameterized! This property of curve singularities
will be studied further in the next chapter. The final topic in this chapter is a function
theoretic interpretation of normal analytic spaces, which can be phrased by saying that
an analytic space (X,x) is normal, if and only if the “Riemann Extension Theorem”
holds for (X,x). This means that if f is a holomorphic function on X \ Y , where Y
contains the singular locus of X, and such that f is locally bounded, then there exists an
extension of f to a holomorphic function on the whole of X. Again in the proof, Noether
normalization is used to reduce the statement to the case of C k , a case we already know,
see 3.1.15.
4.1 Dimension Theory
We start with a generalization of the notion of the germ of an analytic space. We need
this, because in our development of dimension theory nonreduced analytic algebras occur
naturally.
Definition 4.1.1. A germ of a complex space (X,x) consists of
(1) a germ of an analytic space (Xred ,x), called the reduction (X,x) and,
(2) an analytic algebra OX,x such that (OX,x )red = OXred ,x .

Example 4.1.2. Let I = (x2 ) ⊂ C {x}. Then V (I), 0 = (Xred , 0) is the point 0,
and OXred ,x = C . Now I defines a germ of a complex space (X, 0) using (Xred , 0) and
OX,x = C {x}/I.
Next we give three definitions of dimension.
Definition 4.1.3. Let (X,x) be a germ of a complex space and (OX,x ,m) be its local
ring.
• The Weierstraß dimension of (X,x) is the least number k, such that there exists a
Noether normalization Ok ⊂ OX,x of (X,x). (It will turn out that the number k in
a Noether normalization Ok ⊂ OX,x is independent of the Noether normalization.)
4.1 Dimension Theory 127
• The Chevalley dimension of (X,x) is the least number of generators for an m–

primary ideal of OX,x . If k is the Chevalley dimension of (X,x), and f1 , . . . ,fk
generate an m–primary ideal of OX,x , then f1 , . . . ,fk is called a system of parameters
for (X,x). Note that the definition of Chevalley dimension works for all (Noetherian)
local rings.
• The Krull dimension of (X,x) is the maximal length k of chains of prime ideals
p0 $ p1 $ . . . $ pk
in OX,x . Note that this definition also works for all (Noetherian) rings.
Although we will prove in this section that for germs of analytic spaces the three
definitions of dimension coincide, we call the Krull dimension simply the dimension
of (X,x), notation dim(X,x).
To motivate the definition of Weierstraß dimension, consider a Noether normalization

of (X,x). We have seen that we obtain a map (X,x) → (C k ,o) with finite fibers. It would
be natural to call k the dimension of (X,x). However, one has to show that the number k
is independent of the Noether normalization. This needs proof, and the proof will follow
from the fact that the Weierstraß dimension is equal to the Krull dimension. In order
to be careful, we defined the Weierstraß dimension to be the minimal k such that there
exists a Noether normalization (X,x) −→ (C k , 0).
If k is the Chevalley dimension of (X,x), then there exists a system of parameters
f1 , . . . ,fk and an s ∈ Z such that ms ⊂ (f1 , . . . ,fk ), as (f1 , . . . ,fk ) is m–primary. In fact,
the conditions are equivalent in this case, as obviously every element of OX,x /(f1 , . . . ,fk )
which is not a unit is nilpotent. It follows from the Nullstellensatz that the intersection
of (X,x) with the k hypersurfaces {f1 = 0} = . . . = {fk = 0} consists of just one point,
which should have dimension zero. The idea is that if (X,x) has dimension k, then a
general hypersurface section should have dimension k − 1, so that one has to intersect
(X,x) with at least k hypersurfaces to cut down to a single point.
By taking zero sets, it follows that the Krull dimension of (X,x) is the supremum of
the length n of a chain of irreducible subvarieties of (X,x):
{0} $ (X1 ,x) $ . . . $ (Xn ,x) j (X,x).
For the case OX,x = On , consider the following chain of prime ideals:
(0) $ (x1 ) $ (x1 ,x2 ) $ . . . $ (x1 , . . . ,xn )
from which it follows that the Krull dimension of (C n , 0) is at least n. It needs proof,
and is in fact nontrivial that the dimension of (C n , 0) is indeed equal to n. Of course, it
follows from the fact that the Krull dimension is equal to the Weierstraß dimension, or
alternatively, it is equal to the Chevalley dimension. We now start off proving that the
three definitions coincide.
Lemma 4.1.4. Let R ⊂ S be Noetherian local rings, mR ⊂ mS . Suppose that S is
a finitely generated R–module. Then the Krull dimension of S is equal to the Krull
dimension of R. In particular, if ϕ : (X,x) −→ (Y,y) is a finite surjective map of germs
of complex spaces, then dim(X,x) = dim(Y,y).
Proof.
Step 1. First let p0 $ · · · $ pd be a chain of prime ideals in R. By the Going-Up Theorem
1.5.25 there exists a chain of prime ideals P0 $ · · · $ Pd with Pi ∩ R = pi . This shows
that the Krull dimension of S is at least the Krull dimension of R.
Step 2. Now let P0 $ · · · $ Pd be a chain of prime ideals in S. Let pi := Pi ∩ R. Then
of course, the pi are prime ideals in R. It remains to show that for all i = 0, . . . ,d − 1
we have a proper inclusion pi $ pi+1 . By considering the quotient rings R/pi and S/Pi ,
which are integral domains, it obviously suffices to prove the following statement.
Let R ⊂ S be an integral extension, and (0) 6= J ⊂ S be an ideal containing a
nonzerodivisor. Then J ∩ R 6= (0).
To prove this, let x ∈ J be a nonzerodivisor. It satisfies an integral equation, which
we take of minimal degree
xn + . . . + an−1 x + an = 0, ai ∈ R.
Since x is a nonzerodivisor, it follows that an 6= 0, as otherwise we would get the integral

equation xn−1 + . . . + an−1 = 0. Thus an ∈ J ∩ R is a nonzero element.
Corollary 4.1.5. Let (X,x) be a germ of a complex space, and Ok ⊂ OX,x be a Noether
normalization. Then the Krull dimension of (X,x) is equal to k. In particular, the Weier-
straß dimension is equal to the Krull dimension.
Proof. From the previous lemma, it follows that the Krull dimension of (X,x) is equal to
the Krull dimension of (C k , 0). It remains to show that the Krull dimension of (C k , 0) is
equal to k. This will be shown by induction on k. If k = 0, this is obvious. We already
showed that the Krull dimension of (C k , 0) is at least k. Consider a chain of prime ideals
p0 $ p1 $ · · · $ pd in Ok of maximal length. We have to show d ≤ k. Consider the
ring Ok /p1 , which has Krull dimension d − 1. Since p1 6= (0), there exists a Noether
normalization Os ⊂ Ok /p1 with s < k. By induction, the Krull dimension of Os is equal
to s which, by the previous lemma, is equal to the Krull dimension of Ok /p1 . Therefore,
d − 1 = s < k, so that d ≤ k.
Lemma 4.1.6. Let (R,m) be a Noetherian local ring.
(1) The Krull dimension of R is equal to the Krull dimension of Rred . A similar state-
ment holds for the Chevalley dimension.
(2) The Krull dimension of R is zero, if and only if the Chevalley dimension of R is
zero.
Proof.
Step 1. A chain of prime ideals in Rred lifts to a chain of prime ideals in R. On the other
hand, the nilpotent elements of a ring are contained in all prime ideals of R. Therefore,
a chain of prime ideals in R gives also rise to a chain of prime ideals in Rred . Thus we
showed that the Krull dimensions of R and Rred are equal.
Step 2. Next we show that the Chevalley dimensions of R and Rred are equal. Let
(f1 , . . . ,fd ) be an m–primary ideal of R. Then, obviously, the classes of f1 , . . . ,fd also
generate an m–primary ideal in Rred .
On the other hand, suppose that we have elements f1 , . . . ,fd in R that generate
an m–primary ideal q = (f1 , . . . ,fd ) in Rred . This means that there exists a n ∈ N
with mn + N ⊂ q + N , where N is the set of nilpotent elements of R. However, as R is
Noetherian, there exists an s ∈ N with (q+N )s ⊂ q. It follows that mns ⊂ (mn +N )s ⊂ q.
Thus q is an m–primary ideal.
Step 3. First suppose that the Chevalley dimension is zero. We may now suppose that
the ring is reduced. We will show that (0) is a maximal ideal. Because the Chevalley
dimension is zero, it follows that there exists an m–primary ideal with 0 generators. Thus
the ideal (0) is m–primary. From the definition of primary, it follows that every zero
divisor is nilpotent, and hence zero, since R is reduced. Hence there are no zerodivisors
different from 0, which means that (0) is a prime ideal. As, moreover, (0) is m–primary,
it is equal to m. Hence the Krull dimension is zero.
Step 4. Suppose the Krull dimension is zero. Then m is the only prime ideal of R, as
otherwise we can construct a chain of length at least one. As the nilpotent elements
of R is the intersection of all minimal prime ideals, it follows that all elements of m
are nilpotent. Thus (0) is an m–primary ideal, which exactly says that the Chevalley
dimension of R is zero.
We now prove that the Chevalley dimension is smaller than or equal to the Krull
dimension. The proof works for a general Noetherian local ring.
Lemma 4.1.7. Let (R,m) be a Noetherian local ring. Then the Chevalley dimension is
smaller than or equal to the Krull dimension.
The main step in the proof is given by the following proposition.
Proposition 4.1.8. Let R be a Noetherian ring of Krull dimension d. Suppose that f
is an active element of R, that is, f is not contained in any of the minimal prime ideals
of R 1 . Then the Krull dimension of R/(f ) is at most d − 1.
Proof. Consider the ring R/(f ). Any chain of prime ideals p1 $ · · · $ ps in R/(f ) gives a
chain of prime ideals in R, with (f ) ⊂ p1 $ · · · $ ps . The ideal p1 is not minimal, because
otherwise f would be contained in a minimal prime ideal, contrary to the assumption.
Therefore, we can find a minimal prime ideal p0 contained in p1 . Hence s ≤ d, and it
follows that the Krull dimension of R/(f ) is at most d − 1.
Proof of Lemma 4.1.7. Let d be the Krull dimension of R. The proof goes by induction
on d. The case d = 0 was proved in Lemma 4.1.6.
Therefore, we may suppose d > 0. Then m is not a minimal prime ideal. Take an
element f ∈ m, which is not contained in ∪pi , where the pi are the minimal prime ideals of
R. Such an element exists because of Prime Avoidance, see 1.1.13. Applying the previous
lemma, we get that the Krull dimension of R/(f ) is at most d − 1. By induction we
have elements f1 , . . . ,fs , with s ≤ d − 1 generating an m–primary ideal in R/(f ). Then
f,f1 , . . . ,fs generate an m–primary ideal in R. Thus the Chevalley dimension is smaller
than or equal to s + 1 ≤ d.
Theorem 4.1.9. Let (X,x) be a germ of a complex space. Then the Weierstraß dimension
of (X,x) is equal to the Chevalley dimension of (X,x), and thus all three definitions of
dimension coincide for germs of complex spaces.
1 This exactly means that f is not a zerodivisor of Rred , see 1.4.24.
Proof. By Corollary 4.1.5 and Lemma 4.1.7 it remains to show that the Weierstraß
dimension is less than or equal to the Chevalley dimension of (X,x). This will be proved
in two steps.
Step 1. Let ϕ : Ok −→ OX,x be a C –algebra homomorphism, not necessarily injective.
Suppose that OX,x is, via ϕ, a finitely generated Ok –module. Then we claim that the
Weierstraß dimension of (X,x) is less than or equal to k.
Indeed, consider the injective map Ok / Ker(ϕ) ⊂ OX,x . This is an integral extension.
Now Ok / Ker(ϕ) is an analytic algebra. So there exists an s ≤ k and a Noether normal-
ization Os ⊂ Ok / Ker(ϕ). It follows that Os ⊂ OX,x is a finitely generated extension,
hence a Noether normalization. The claim follows.
Step 2. Now let f1 , . . . fs be generators of an m–primary ideal of OX,x . We consider the
map
ϕ : C {y1 , . . . ,ys } −→ OX,x

yi 7→ fi .
By assumption OX,x /(f1 , . . . ,fs ) is a finitely generated C –vector space, so that, by the
general Weierstraß Division Theorem 3.2.10, OX,x is a finitely generated C {y1 , . . . ,ys }–
module. By step 1 it follows that the Weierstraß dimension is less than or equal to s.
Corollary 4.1.10 (Active Lemma). Let f ∈ m ⊂ OX,x be an active element, (Y,x) :=
V (f ),x ⊂ (X,x). Then dim(Y,x) = dim(X,x) − 1.
Proof. We have shown in 4.1.8, dim(Y,x) ≤ dim(X,x) − 1. It cannot be smaller, as
otherwise we would be able to find s generators of a primary ideal with s < dim(X,x).
Example 4.1.11. The condition that f is an active element
is essential. For example,
the germ consisting coordinate
axes (X, 0) = V (xy), 0 has dimension one. Take f = x.
Then (Y,x) = V (x), 0 also has dimension one.
The following corollary is typical for the germ (C n , 0) and certainly does not hold
for general germs of analytic spaces.
Corollary 4.1.12 (Characterization of Hypersurfaces). Let (X, 0) ⊂ (C n , 0) be irre-
ducible, and suppose dim(X, 0) = n − 1. Then there exists an f with I (X, 0) = (f ), that
is (X, 0) is a hypersurface.
More generally, if every irreducible component of (X, 0) has dimension n − 1, it
follows that (X, 0) is a hypersurface.
Proof. It follows that I (X, 0) is a prime ideal, so that we can take an irreducible f ∈
I (X, 0). We claim that (f ) = I (X, 0). Indeed, otherwise there would be a g ∈ I (X, 0) \
(f ). In particular g would be a nonzerodivisor modulo (f ), and by the Active Lemma
dim(X, 0) ≤ dim V (f,g), 0 = n − 2, a contradiction.
To prove the second statement, let (X, 0) = (X1 , 0) ∪ · · · ∪ (Xr , 0) be an irre-
ducible decomposition.
Then we just proved that (Xi , 0) = V (fi ), 0 for some fi . Thus
V (f1 · · · fr ), 0 = V (f1 ), 0 ∪ · · · ∪ V (fr ), 0 = (X, 0).
Theorem 4.1.13. Suppose (X,x) is a germ of a complex space. Let (X,x) = (X1 ,x) ∪
· · · ∪ (Xr ,x) be an irreducible decomposition of (X,x). Then
dim(X,x) = max{dim(Xi ,x), i = 1, . . . ,r}.

Proof. We may assume that OX,x is reduced. We use the Krull dimension. The inequal-
ity ≥ is obvious, since a chain of irreducible germs in (Xi ,x) gives one in (X,x). Let
I (Xi ,x) = pi . Then pi is a minimal prime ideal of OX,x . Let (Y,x) be an irreducible
subgerm of (X,x). Then there exists an i with (Y,x) ⊂ (Xi ,x). This is because I (Y,x)
is a prime ideal, and therefore there exists a minimal prime ideal pi ⊂ I (Y,y). Taking
zero sets gives (Y,x) ⊂ (Xi ,x). So a maximal chain of irreducible subgerms is contained
in some irreducible component of (X,x). This shows the theorem.
Now consider an irreducible
germ of an analytic space (X,x), and let 0 6= f ∈ m ⊂
OX,x . Put (Y,x) := V (f ),x ⊂ (X,x). It may very well happen that (Y,x) is reducible.
Note that we proved (Active Lemma) that the dimension of (Y,x) is equal to dim(X,x)−1,
but we did not prove that each irreducible component of (Y,x) has dimension dim(X,x)−
1. We only proved that there is at least one component of this dimension. However,
the statement that for an active element f each irreducible component of V (f ) has
codimension one is in fact true, and holds for general Noetherian rings. This is Krull’s
Principal Ideal Theorem, to be proved in the next section, see 4.2.16. In this section, we
give a proof for germs of analytic spaces which uses the Going-Down Theorem 1.5.26.
First we need a definition.
Definition 4.1.14. Let R be a ring.
(1) Let p be prime ideal. We define the height ht(p) of p, to be the maximal m such
that there exists a chain of prime ideals
p0 $ p1 $ · · · $ pm = p.
(2) For a general ideal I in R we define the height of I by ht(I) := min{ht(p) : p ∈

Ass(I)}.
Remark 4.1.15. Note that from the interpretation of prime ideals in a localization, see
1.3.15, it follows that the height of a prime ideal is equal to the Krull dimension of Rp .
Furthermore, we have the inequality
(4.1) ht(I) + dim(R/I) ≤ dim(R),
as follows directly from the definition. Consider a germ of a complex space (X,x), and
let I ⊂ OX,x be an ideal. Equality in (4.1) would allow us to interpret the height of I as
the codimension of the zero set of I in (X, 0). The codimension of a germ of a complex
(Y,x) in (X,x) is simply defined as dim(X,x) − dim(Y,x). Of course, it suffices to show
this for irreducible germs . This will be shown now.
There exist examples of Noetherian rings, for which we have strict inequality in
(4.1), see Exercise 4.1.30.
Theorem 4.1.16. Let (X,x) ⊂ (C n , 0) be an irreducible germ of a complex space. All
maximal chains of prime ideals in OX,x have the same length, namely dim(X,x). In
particular, formula (4.1) holds for R = OX,x .
Proof. Let Ok ⊂ OX,x be a Noether Normalization, and consider a chain of prime ideals
P0 $ P1 $ · · · $ Ps
in OX,x which cannot be extended to a larger chain. We will show that s = k = dim(X,x).
Obviously s ≤ k.
Step 1. Let p be a prime ideal of height one in Ok . We will first show that the dimension
of Ok /p is k − 1. Now p 6= (0), and there exists an element 0 6= f ∈ p. As p is prime,
there exists an irreducible factor of f which lies in p. So we may assume that we have
an irreducible f which lies in p. As Ok is a unique factorization domain, (f ) is a prime
ideal. Since p has height one, it follows that p = (f ). By the Active Lemma 4.1.10, the
dimension of Ok /p is k − 1.
Step 2. Let pi = Pi ∩ Ok . Then we have seen in the proof of Lemma 4.1.4 that
(4.2) p0 $ p1 $ · · · $ ps
is a chain of prime ideals in Ok . We claim that this chain cannot be extended to a larger
chain of prime ideals in Ok . Suppose the converse, and suppose that there exists a prime
ideal q with say pi $ q $ pi+1 . (Here we take the convention ps+1 = Ok .) Take a Noether
normalization
Ot ⊂ Ok /pi .
We thus have Ot ⊂ Ok /pi ⊂ OX,x /Pi . As Ok /pi ⊂ OX,x /Pi is also a finite extension,
we have a Noether normalization Ot ⊂ OX,x /Pi . The prime ideals q and pi+1 induce
prime ideals in Ok /pi which we denote by the same name. Moreover, (0) $ q $ pi+1 .
Intersecting this chain with Ot , we get a chain of prime ideals in Ot
(0) $ q′ $ p′i+1 .
Moreover, by construction, Ot ∩ Pi+1 = p′i+1 . Furthermore, Ot is a normal ring (it is

even factorial), so that we can apply the Going-Down Theorem 1.5.26: there exists a
prime ideal Q in OX,x /Pi with (0) $ Q $ Pi+1 . We lift this prime ideal to a prime
ideal in OX,x , which we also call Q. So Pi $ Q $ Pi+1 which is in contradiction to our
assumption that the chain Po $ P1 $ · · · $ Ps is maximal. Hence the chain (4.2) is
maximal after all.
Step 3. We now show s = k = dim(X,x) by induction on k. The case k = 0 is obvious.
If k > 0, then the maximal ideal of OX,x is nonzero. So in a maximal chain Ps is the
maximal ideal and s ≥ 1. By step 1, Ok /p1 has dimension k − 1. Now p1 $ · · · $ ps
induces a maximal chain in Ok /p1 by step 2, and by induction, s − 1 = k − 1.
Corollary 4.1.17 (Krull’s Principal Ideal Theorem for OX,x ). Let (X,x) be an irre-
ducible germ
of an analytic space, and f ∈ OX,x . Then all irreducible components of
V (f ),x ⊂ (X,x) have dimension dim(X,x) − 1.
Definition 4.1.18.
(1) Let (X, 0) ⊂ (C n , 0) be a germ of an analytic space, and let I (X, 0) be the ideal
of (X, 0), and k be the minimal number of generators of I (X, 0). Then (X, 0) is
called a complete intersection if the dimension of (X, 0) is n − k.
(2) A germ of an analytic
space (X, 0) is called a set-theoretic complete intersection if
(X, 0) = V (I), 0 , and I can be generated by n − dim(X, 0) elements.
(3) Let R be a local ring. A sequence of nonunits {f1 , . . . ,fk } is called a regular sequence
if for all i = 1, . . . ,k (the class of) fi is a nonzero divisor of R/(f1 , . . . ,fi−1 ).
Example 4.1.19. We consider (X, 0), the coordinate axes in C 3 , given by the zero set of
the ideal I = (xy,xz,yz). This is a radical ideal as it is the intersection of the prime ideals
(x,y), (x,z) and (y,z). So I = I (X, 0), and by Nakayama’s Lemma it follows that the
minimal number of generators of I is equal to three. It is easy to see (using for example
the Chevalley dimension), that the dimension of (X, 0) is one. Therefore, (X, 0) is not a
complete intersection. It is however
a set-theoretic complete intersection. Indeed as one
easily checks, (X, 0) = V (J), 0 , where J = (xy,zy + zx).
Corollary 4.1.20. Let (X, 0) be a germ of a complex space, and suppose I (X, 0) is
minimally generated by f1 , . . . ,fk ∈ On . Then (X, 0) is a complete intersection, if and
only if {f1 , . . . ,fk } form a regular sequence in On . In this case, dim(X, 0) = n − k.
The proof is left as Exercise 4.1.25.
Definition 4.1.21. Let I ⊂ R be an ideal.
(1) I is called unmixed if all associated primes of I have the same height.
(2) I (or R/I) is called equidimensional, if for all minimal associated primes p we have
dim(Rp /IRp ) = dim(R/I).
In case I ⊂ On is unmixed, I does not have embedded components and is equidimen-
sional. By abuse of language, we also say that (X, 0) is unmixed, if the ideal I defining
(X, 0) is unmixed. Note that it very well might happen that unmixed ideals are not rad-
ical. As a corollary of dimension theory, we can give a nice characterization of unmixed
ideals in On in terms of a Noether normalization.
Theorem 4.1.22. Let I ⊂ On be an ideal and OX,0 = On /I. Suppose we have a Noether
normalization Ok ⊂ OX,0 of (X, 0). Then:
I is unmixed ⇐⇒ Every f ∈ Ok is a nonzero divisor of OX,0 .
Proof. Consider a primary decomposition of I:
I = q1 ∩ . . . ∩ qs .
√
Let pi := qi for i = 1, . . . ,s. By the Weierstraß characterization of dimension the ideal
I is unmixed, if and only if for all i = 1, . . . ,s the map
Ok −→ On /pi ,
defined by the composition Ok ⊂ On /I −→ On /qi −→ On /pi , is injective. Therefore, I is

unmixed if and only if for all f ∈ Ok we have f ∈
/ pi for all i, that is, f is a nonzerodivisor
according to the characterization of nonzerodivisors, see Theorem 1.4.24.

Example 4.1.23. Consider I = (x2 + xy,xz). Then (X, 0) = V (I), 0 consists of two
components. The first one is given by x = 0, and the second by z = 0, x = −y. So (X, 0)
is not equidimensional, and therefore not unmixed. A Noether normalization is given by
C {y,z} ⊂ C {x,y,z}/(x2 + xy,xz). So geometrically we are projecting on the y, z plane.
It is indeed an inclusion. Any function which vanishes on the union of the plane and the
line must also vanish on the plane x = 0. The extension is integral. Indeed, x satisfies
the integral equation x2 + xy = 0. So we see from the Noether normalization that (X, 0)
is not unmixed, since the element z is a zerodivisor.
Theorem 4.1.24 (Second Riemann Extension Theorem). Let U ⊂ C n be open, and

consider an analytic set X in U of codimension at least 2, that is, for all points p ∈ X,
we have dim(X,p) ≤ n − 2. Let
f : U \ X −→ C
be a holomorphic function. Then there exists a unique holomorphic extension of f to U .
Proof. The basic idea of the proof is the same as in the proof of the case that X consists
of a single point, see Theorem 3.1.16. We will show that f is locally bounded on U \ X,
so that we can apply the First Riemann Extension Theorem 3.1.15. Consider a Noether
normalization at p ∈ X:
(X,p) −→ (C k , 0).
The map is induced by the natural projection π : U −→ C k . By assumption k ≤ n − 2.
Step 1. We claim that for all small open balls V ⊂ C n−k there exists a small closed
neighborhood B of 0 ∈ C k , such that B × V ⊂ U and π −1 (B) ∩ X ⊂ B × V . In particular
(π −1 (B) ∩ X) ∩ (B × ∂V ) = ∅.
To show this, note that, by the Noether Normalization Theorem, we may assume that
OX,p ∼ = C {x1 , . . . ,xn }/J ⊃ C {x1 , . . . ,xk } is finite. We consider Weierstraß polynomials
Pk+1 , . . . ,Pn ∈ J, with Pi ∈ C {x1 , . . . ,xk }[xi ]. We may assume that the Pi converge on
U , and that X ⊂ {q ∈ U : Pk+1 (q) = · · · = Pn (q) = 0}. We now apply 3.4.11 to each of
the Pi and obtain an open ball B0 ⊂ C k , 0 ∈ B0 , such that π −1 (B0 ) ∩ X ⊂ B0 × V . Let
B ⊂ B0 be a closed ball with 0 ∈ B. Then π −1 (B) ∩ X ⊂ B × V .
Step 2. Now we choose a point q ∈ (B × V ) \ X. Then {π(q)} × V is also an open ball.
As the restriction of π to X is finite, the intersection of {π(q)} × V with X consists of a
finite number of points. We therefore can find a line Lq through q in {π(q)} × V which
misses ({π(q)} × V ) ∩ X, because of the fact that V has at least dimension 2.
Lq
q
V
X
B
The line Lq intersects the boundary {π(q)}×∂V . The restriction of f to Lq ∩({π(q)}×V )
is a function of one complex variable, so attains maximum absolute value on the boundary
{π(q)} × ∂V . Hence
4.2 Hilbert-Samuel Function and Multiplicity 135
|f (q)| ≤ max | f|{π(q)}×∂V |≤ max | f|B×∂V | .
As f is continuous, and B × ∂V is compact, it follows that f is locally bounded. So we

can apply the First Riemann Extension Theorem 3.1.15.
Exercises
4.1.26. Let (X,x) and (Y,y) be germs of analytic spaces. Show that
` ´
dim X × Y,(x,y) = dim(X,x) + dim(Y,y).
4.1.27. Let X ⊂ C n be an affine algebraic set, m = (x1 , . . . ,xn ). Let mk = (x1 , . . . ,xk ) ⊂
C [x1 , . . . ,xk ] and suppose
C [x1 , . . . ,xk ]m k ⊂ C [X]m
is a Noether normalization. Consider the germ of the analytic space (X, 0). Show that
Ok ⊂ OX,0
is a Noether normalization, cf. 2.2.35. Deduce that the dimension of X at 0, considered as an

algebraic set, is equal to the dimension of (X, 0),
` ´
4.1.28. Let (X, 0) = V (f1 , . . . ,fr ), 0 ,fi ∈ On be a germ of an analytic space. Consider the
map:
ϕ : Onr −→ On ,
given by the matrix (f1 , . . . ,fr ). Show that for all i < j the element rij = fj ei − fi ej (here
e1 , . . . ,er is the canonical basis of Onr ) is in the kernel of ϕ. Show that (X,0) has codimension r
in (C n , 0) if and only if the kernel of ϕ is generated by the rij .
4.1.29. Let p be a prime ideal in C {x1 , . . . ,xn }, with ht(p) = k. Show that there exists a regular
sequence f1 , . . . ,fk in p. Geometrically, this means that an irreducible space of codimension k is
contained as component in a complete intersection of codimension k.
4.1.30. Consider the ring R = C [[x]][y]. Show that the dimension of R is two. Consider the
principal ideal I := (xy − 1). Show that ht(I) = 1, and that I is a maximal ideal. Hence in (4.1)
we have a strict inequality.
4.1.31. Prove that for an irreducible affine space X, dim(X,p) is independent of the chosen
point p ∈ X.
4.1.32. Let X ⊂ C n be an irreducible affine variety and let C (X) be the quotient field of the
coordinate ring C [X]. Prove that the dimension of X is equal to the transcendence degree of the
field extension C ⊂ C (X). (In some books, this is used as the definition of dimension.)
4.2 Hilbert-Samuel Function and Multiplicity
Definition 4.2.1. Let (R,m) be a Noetherian local ring. The function
HSR : N −→ N

d 7→ dimR/m R/md
is called the Hilbert-Samuel function of R. In case R = OX,x is the local ring of a germ
of a complex space (X,x) we also write HSX,x and call it the Hilbert-Samuel function of
(X,x).
Example 4.2.2. We take (X,x) = (C n , 0). Then R/md is equal to the vector space of
all polynomials
of degree less than d in n variables. This is a vector space of dimension
n+d−1
n . This can be proved by inductionP for example, but the easiest way to see this is as
follows. To a monomial xα 1
1
· · · xαn
n with i < d we assign a monomial of degree d − 1
i αP
α1 αn d−1− i αi
in n + 1 variables, namely x1 · · · xn · x0 . So it suffices to count all monomials
of degree d − 1 in n + 1 variables. A monomial xα αn
0 · · · xn we represent by circles and
0
crosses. We first put α0 circles, then a cross, then α1 circles, then a cross, etc.
x x
α0 αn
On the other hand, one can assign a polynomial to such a diagram.
So we have to
choose n crosses out of n + d − 1 possibilities, and we get n+d−1 monomials. It follows
n
n
n+d−1

that the Hilbert-Samuel function HSC ,0 of (C , 0) is given by d 7→
n
n .
The aim of this section is to prove that for d ≫ 0, the Hilbert-Samuel function
behaves like a polynomial, and interpret its degree as the dimension of the space. Finally,
in case R is the local ring of a singularity we will interpret the leading coefficient of the
polynomial as its “multiplicity”.
Theorem 4.2.3. Let (R,m) be a Noetherian local ring.
(1) There exists a polynomial HSPR ∈ Q[t], such that HSPR (d) = HSR (d) for d suffi-
ciently large. We call HSPR the Hilbert-Samuel polynomial of R.
In case R is the local ring of (X,x), we also write HSPX,x for the Hilbert-Samuel
polynomial of R, and call it the Hilbert-Samuel polynomial of (X,x).
(2) deg(HSPR ) = dim(R).
The proof of 4.2.3 will be given on page 141. For the proof, we need to introduce a
much more general notion of Hilbert-Samuel function and polynomial.
Definition 4.2.4. Let (R,m) be a Noetherian local ring, q be an m–primary ideal and
M be a finitely generated R–module. A q–filtration {Mi } = {Mi : i ∈ N} is a sequence
of submodules M = M0 ⊃ M1 ⊃ . . . such that qMd ⊂ Md+1 for all d. It is called a stable
q–filtration (or q–stable filtration) if moreover there exists an n0 such that qMd = Md+1
for all d ≥ n0 .
Let {Mi } be a q–filtration. Then HS{Mi } : N −→ N with
HS{Mi } (d) := dimR/m M/Md
is called the Hilbert-Samuel function of the filtration {Mi }. (We will show in Theorem
4.2.6 that indeed dimR/m M/Md < ∞, so that the definition makes sense.)
If we take the filtration {Mi } with Mi = mi · M , we simply write HSM for HS{Mi } ,
and this is called the Hilbert-Samuel function of M . The Hilbert-Samuel function of
Definition 4.2.1 is the case M = R, q = m, and Mi = mi .
Remarks 4.2.5.
(1) We need the general concept of the Hilbert-Samuel function with respect to a filtra-
tion to obtain additivity. Namely, let N ⊂ M be a submodule, then the sequence
0 −→ N/N ∩ mn M −→ M/mn M −→ (M/N )/mn (M/N ) −→ 0
is exact and, therefore,
HSM = HSM/N + HS{N ∩mnM} .
But usually HSN 6= HS{N ∩mn M} .

(2) We need the Hilbert-Samuel function with respect to an m–primary ideal q in order
to be able to use the Noether Normalization Theorem. Namely, let Ok ⊂ OX,x
be a Noether normalization, and m be the maximal ideal of OX,x . Then q :=
(x1 , . . . ,xk )OX,x is m–primary. {qi } is obviously a (x1 , . . . ,xk )–stable filtration of
OX,x . Thus HSP{qi } is the Hilbert-Samuel polynomial of the Ok –module OX,x with
respect to (x1 , . . . ,xk ), and at the same time it is the Hilbert-Samuel polynomial
of OX,x with respect to the m–primary ideal q.
Moreover, we will prove that {mi } is a stable (x1 , . . . ,xk )–filtration of OX,x if we
have a general Noether normalization in the sense of 3.3.30. Therefore, the Hilbert-
Samuel polynomial HSPX,x can also be calculated as the Hilbert-Samuel polynomial
HSP{mi } of the stable (x1 , . . . ,xk )–filtration {mi } of the Ok –module OX,x .
The general notion of Hilbert-Samuel function with respect to an m– primary ideal
q is also used in order to prove second part of 4.2.3, using the Chevalley dimension.
In order to prove the first part of 4.2.3, we need to prove a more general version,
as written down in the next theorem.
Theorem 4.2.6. Let (R,m) be a local ring q be a primary ideal, and M be a finitely
generated R–module. Let r be the minimal number of generators of q, and let {Mi } be a
stable q-filtration of M . Then
(1) HS{Mi } (d) < ∞ for all d ≥ 0,

(2) there exists a polynomial HSP{Mi } of degree ≤ r, such that for all d sufficiently
large HS{Mi } (d) = HSP{Mi } (d),
(3) the degree and the leading coefficient of HSP{Mi } only depends on M and q, and
not on the chosen stable q–filtration.
The proof needs some more preparations, and will be given on page 140. First note
Pd−1
that, obviously HS{Mi } (d) = k=0 dimR/m Mk /Mk+1 . This leads to the following def-
inition.
Definition 4.2.7. Let (R,m) be a local ring, q an m–primary ideal and {Mi } be a
stable q–filtration. The associated graded ring Grq = Grq (R) by definition is Grq (R) :=
⊕d≥0 qd /qd+1. The associated graded module Gr{Mi } = Gr{Mi } (M ) by definition is
Gr{Mi } := ⊕d≥0 Md /Md+1 . Note that Grq is a graded ring, and that Gr{Mi } is a graded
Grq –module. This follows directly from the definition of a q–filtration.
The following lemma motivates the condition in the definition of a stable q–filtration.
Lemma 4.2.8. Let M be a finitely generated module over the Noetherian local ring
(R,m), q be an m–primary ideal and {Mi } be a stable q–filtration of M . Then Grq is a
Noetherian ring, and Gr{Mi } is a finitely generated Grq –module.
Proof.
Step 1. We first prove that Grq is Noetherian. By assumption R is Noetherian
and, there-
fore, q is finitely generated, say by a1 , . . . ,at . The algebra map R/q [x1 , . . . ,xt ] −→ Grq
which sends xi to the class of ai is surjective. The ring (R/q)[x1 , . . . ,xt ] is Noetherian,
and from the surjectivity it follows that Grq is Noetherian.
Step 2. Now we show that Gr{Mi } is a finitely generated Grq –module. Because {Mi }
is a stable q–filtration, there exists an n0 ∈ N such that Mn0 +ℓ = qℓ Mn0 for all ℓ ≥
0. This says that Gr{Mi } is generated by ⊕k≤n0 Mk /Mk+1 . Furthermore, because R
is Noetherian, the module Mk /Mk+1 is a finitely generated R–module. Itfollows that
Mk /Mk+1 is a finitely generated R/q–module. Therefore, ⊕k≤n0 Mk /Mk+1 is a finitely
generated R/q–module. So indeed Gr{Mi } is a finitely generated Grq –module.
This motivates us to study the graded situation.
Definition 4.2.9. Let K be a field, A = ⊕i≥0 Ai be a graded K–algebra. Suppose
that dimK A0 < ∞ and that A1 = (x1 , . . . ,xr )A0 generates A as an A0 –algebra. This
means that A is the ring A0 [x1 , . . . ,xr ].2 Let M = ⊕i≥0 Mi be a finitely generated graded
A–module. We define the Hilbert function HM : N −→ N of M by
HM (n) = dimK Mn .
The formal power series with integer coefficients

X
HPM (t) = HM (k)tk ∈ Z[[t]]
k
is called the Hilbert-Poincaré series of M .

Proposition 4.2.10. With the notations of 4.2.9, there exists a polynomial QM ∈ Z[t]
such that
QM (t)
HPM (t) = ,
(1 − t)r
that is, the Hilbert-Poincaré series is a rational function in t.
Proof. The proof is by induction on r.
Step 1. We first consider the case r = 0. In this case A1 = 0, so that A = A0 . Therefore,
M is a finitely generated module over A = A0 . Hence, from dimK (A0 ) < ∞, it follows
that dimK M < ∞. Thus Mk = 0 for large k. Hence, by definition of the Hilbert-Poincaré
series it follows that HPM (t) ∈ Z[t].
Step 2. The induction step. Let r > 0. Consider the following exact sequence of graded
modules:
ϕ
0 −→ Ker(ϕ) −→ M (−1) −→ M −→ Coker(ϕ) −→ 0.
Here M (−1) is isomorphic to M , but has different grading: M (−1)k = Mk−1 , see 1.2.17.
The map ϕ is multiplication with xr . If we have an exact sequence
2 More precisely, via the map yi 7→ xi the algebra A is a quotient of the polynomial ring A0 [y1 , . . . ,yr ].
0 −→ M1 −→ M2 −→ M3 −→ M4 −→ 0
of finite-dimensional K–vector spaces, then dimK (M1 ) + dimK (M3 ) = dimK (M2 ) +
dimK (M4 ), see Exercise 1.2.26. It follows from the definition of the Hilbert-Poincaré
series that
HPKer(ϕ) (t) + HPM (t) = HPM(−1) (t) + HPCoker(ϕ) (t).
From the definition of ϕ, it follows that Ker(ϕ) and Coker(ϕ) are finitely generated
A/xr –modules. By induction, we therefore have
QKer(ϕ) (t) QCoker(ϕ) (t)
HPKer(ϕ) (t) = , and HPCoker(ϕ) (t) = .
(1 − t)r−1 (1 − t)r−1
From the definitions of M (−1) and HP it follows immediately that HPM(−1) (t) =
t HPM (t), and we obtain
(1 − t) HPM (t) = HPM (t) − HPM(−1) (t)

= HPCoker(ϕ) (t) − HPKer(ϕ) (t)
QCoker(ϕ) (t) − QKer(ϕ) (t)
= .
(1 − t)r−1
This proves the proposition.
Corollary 4.2.11. With the conditions as in Definition 4.2.9 we have the following
statements.
(1) There exists a polynomial PM ∈ Q[t] of degree ≤ r − 1, the so-called Hilbert poly-
nomial of M , such that PM (d) = HM (d) for d sufficiently large.
(2) Let N = ⊕i≥0 Ni be a finitely generated graded A–module and ϕ : M −→ N be a
surjective homomorphism of graded A–modules. Then deg PM ≥ deg PN .
(3) deg PM ≤ deg PA .
(4) If Ann(m) = 0 for some homogeneous m ∈ M , then deg PM = deg PA .
Proof. (1) By definition HM (d) is the coefficient of td in the Hilbert-Poincaré series
QM (t)
HPM (t) = (1−t) r . We cancel powers of 1 − t in this fraction and get
G(t)
HPM (t) = , G(1) 6= 0,
(1 − t)s
Pk
with G(t) = i=0 gi ti ∈ Z[t] and s ≤ r. We expand (1 − t)−s in a power series. We leave
it as an exercise to show that the power series expansion of (1 − t)−s is
X ∞
−s s−1+i i
(1 − t) = t.
i=0
s−1
By multiplying G with (1 − t)−s and taking the coefficient of td we obtain for d ≫ 0
Xk
s−1+d−i
HM (d) = gi .
i=0
s−1
This is a polynomial in d of degree s− 1, which proves the first statement of the corollary.
(2) The surjectivity implies that HM (d) = dimK Md ≥ dimK Nd = HN (d). Hence
PM (d) ≥ PN (d) ≥ 0 for d sufficiently large. This is only possible if deg PM ≥ deg PN .
(3) Let M be generated by m1 , . . . ,ms . Let di be the degree of mi . Then we have a
surjective map of graded modules ⊕m . By the second part, deg P⊕A(di ) ≥
i=1 A(di ) ։ MP
deg PM . But PA(di ) (d) = PA (d+di ) and P⊕A(di ) = si=1 PA(di ) . Therefore deg P⊕A(di ) =
deg PA .
(4) If Ann(m) = 0, for some homogeneous m ∈ Mk , then there exists an injection of
graded A–modules
A(k) −→ M : 1→
7 m ∈ Mk .
This implies HA (d + k) = HA(k) (d) ≤ HM (d). This proves the corollary.
Example 4.2.12. Let A = C [x 1 , . . . ,xn ], Ai = {f ∈ A : f homogeneous of degree i}.
Then HA (d) = PA (d) = n+d−1
n−1 . Hence, deg(PA ) = n − 1 and
X∞
i+n−1 i 1
HPA (t) = t = .
i=0
n−1 (1 − t)n
We are now able to prove Theorem 4.2.6.

Proof of Theorem 4.2.6. (1) We have seen that fors all k the R/q module Mk /Mk+1 is
finitely generated.
So we have a surjection ։ Mk /Mk+1 . As q is m–primary,
R/q
dimR/m R/q < ∞. Hence, dimR/m Mk /Mk+1 < ∞. It follows that HS{Mi } (d) =
Pd−1
k=0 dimR/m Mk /Mk+1 < ∞.
(2) Define Ai = qi /qi+1 for all i ≥ 0. Then Grq = ⊕i≥0 Ai , and satisfies the con-
dition of Definition 4.2.9, and Gr{Mi } is a finitely generated Grq –module by 4.2.8.
Let r be the minimal number of generators of q. In the proof of 4.2.8, we saw that
then the minimal number of generators of A1 is at most r. It follows that the the
Hilbert polynomial HPGr{Mi } has degree at most r − 1 by Corollary 4.2.11. Note that
HPGr{Mi } (d) = HS{Mi } (d + 1) − HS{Mi } (d) for large d. It is a nice exercise, see 4.2.25,
that this implies (2).
(3) Let {Mfi } be another stable q–filtration of M . It is not difficult to show, see Exercise
4.2.31, that there exists an n0 such that Mn+n0 ⊂ M fn and M fn+n0 ⊂ Mn for all n. This
implies HS{Mi } (d) ≤ HS{M fi } (d) ≤ HS{Mi } (d + n0 ). As the degree
fi } (d + n0 ) and HS{M
of d 7→ HSP{Mi } (d + n0 ) in d is equal to the degree of HSP{Mi } , and similar for {M fi },

this shows that the degrees of HSP{Mi } and HSP{M fi } are equal. On the other hand the
inequalities give
HSP{Mi } (d) HSP{M

fi } (d)
≤ 1, and ≤ 1,
HSP{M
fi } (d + n0 ) HSP{Mi } (d + n0 )
for d large. This implies that the leading coefficients of HSP{Mi } and HSP{M
fi } are equal
and proves the theorem.
Before giving the proof of Theorem 4.2.3 we need two more results.
Lemma 4.2.13 (Artin-Rees Lemma). Let (R,m) be a Noetherian local ring, q be an

m–primary ideal, and {Mi } be a q–stable filtration of a finitely generated R–module M .
Let N ⊂ M be a submodule. Then the filtration {Ni } defined by Ni := Mi ∩ N is a stable
q–filtration of N .
Proof.
Step 1. Consider the ring R∗ := ⊕k≥0 qk . This is a graded ring in the obvious way. If q
is generated by a1 , . . . ,ar , then we get a surjective algebra map R[x1 , . . . ,xr ] −→ R∗ by
sending xi to ai . Hence R∗ is Noetherian.
Step 2. Now consider a filtration {Mi } of M , with qMk ⊂ Mk+1 , but which is not
necessarily q–stable. Then we can form the graded R∗ –module M ∗ := ⊕k≥0 Mk . We
claim:
{Mi } is a stable q–filtration ⇐⇒ M ∗ is a finitely generated R∗ –module.
To see this, consider for each n the submodule Mn′ of M ∗ generated by ⊕nk=0 Mk . It is
equal to
Mn′ = M0 ⊕ · · · ⊕ Mn ⊕ qMn ⊕ q2 Mn ⊕ · · · .
The Mn′ form an ascending chain of submodules of M ∗ . As R∗ is Noetherian by Step 1,
this chain becomes stable, if and only if M ∗ is a finitely generated R∗ –module. This is
the case, if and only if Mn′ 0 = M ∗ for some n0 . This again is the case if and only if
qr Mn0 = Mn0 +r for some n0 and all r ≥ 0, which is equivalent to the stability of the
filtration.

Step 3. We now can prove the Artin-Rees Lemma. We have qNk = q N ∩ Mk ⊂ qN ∩
qMk ⊂ N ∩ Mk+1 = Nk+1 , for all k ≥ 0 showing that {Nk } is a q–filtration. Therefore
N ∗ := ⊕k≥0 Nk is an R∗ –module, and in fact a submodule of M ∗ . As {Mi } is q–stable,
the module M ∗ is finitely generated, and therefore Noetherian by the previous step.
Therefore, N ∗ is finitely generated, and the filtration {Ni } is q–stable, again by step
2.
Proposition 4.2.14. Let (R,m) be a local ring, q be an m–primary ideal, and {Mi }
be the stable q–filtration defined by Mi = qi M . Let x ∈ R be a nonzerodivisor of M ,
M ′ := M/(x · M ), and {Mi′ } be the filtration on M ′ defined by Mi′ = qi M ′ . Then

deg HSP{Mi′ } ≤ deg HSP{Mi } − 1.
Proof. Consider N = x · M ⊂ M . Because x is not a zerodivisor, N is as R–module

isomorphic to M . Consider the filtration {Ni } given by Ni = N ∩ qi M . By the Artin
Rees Lemma 4.2.13, {Ni } is a q–stable filtration. So by 4.2.6 the degree and the leading
coefficient of HSP{Ni } and of HSP{Mi } are the same. We have for all k the short exact
sequence
0 −→ N/Nk −→ M/Mk −→ M ′ /Mk′ −→ 0.
Therefore, for d ≫ 0 we have that HSP{Mi′ } (d) = HSP{Mi } (d) − HSP{Ni } (d), so we see
that HSP{Mi′ } has smaller degree than HSP{Mi } .
Proof of Theorem 4.2.3. We apply Theorem 4.2.6 to the case M = R, and get the first
part. Now we show that the degree of the Hilbert-Samuel polynomial is equal to the
dimension of R. This is done in several steps.
Step 1. We first prove that the degree of HSPR is smaller than or equal to the Chevalley
dimension of R. Let q be any m–primary ideal, and consider the filtration qi R. We claim
that deg(HSP{qi } ) = deg(HSPR ) which was by definition equal to deg(HSP{mi R} ). To
see this, first note that for some s ∈ N we have m ⊃ q ⊃ ms . Thus for all d ≥ 0, we have
md ⊃ qd ⊃ msd . Thus for large d we get
HSPR (d) ≤ HSP{qi } (d) ≤ HSPR (sd).
Now HSPR (sd) is also a polynomial in d, and has the same degree as HSPR (d). Thus
the degrees of the polynomials are equal, proving the claim. The second part of Theorem
4.2.6, says that this degree is at most the minimal number of generators of q, which is
the Chevalley dimension of R.
Step 2. We now prove that the Krull dimension of R is less than or equal to deg(HSPR ).
As we already showed that the Chevalley dimension is less than or equal to the Krull
dimension, see 4.1.7, this suffices to prove the theorem. The inequality is proved by
induction on deg(HSPR ).
Step 2a. Suppose that deg(HSPR ) = 0. It follows that dim(R/md ) is independent of
d for d ≫ 0. Hence md = md+1 for d ≫ 0. It follows from Nakayama’s Lemma that
md = 0.Thus the zero ideal is m–primary. So the Chevalley dimension, and hence the
Krull dimension is zero.
Step 2b. We now do the induction step. Let s = dim(R), and p0 $ · · · $ ps be a maximal
chain of prime ideals in R. As the case s = 0 is trivial, we may suppose s > 0. In
particular p1 6= 0. Take an element x ∈ p1 \ p0 , and consider the ring S := R/p0 .3 We
claim that deg(HSPS/(x) ) ≤ deg(HSPR ) − 1. Indeed, x is a nonzerodivisor so that by
4.2.14 deg(HSPS/(x) ) ≤ deg(HSPS ) − 1. Moreover, we have a surjective map R/md ։
S/md . Therefore, deg(HSPS ) ≤ deg(HSPR ). Thus deg(HSPS/(x) ) ≤ deg(HSPR ) − 1. We
apply the induction hypothesis. Any chain in S/(x) has length at most deg(HSPR ) − 1.
But the image of the chain p1 $ · · · $ ps in S/(x) gives a chain of length s − 1. So
dim(R) − 1 = s − 1 ≤ deg(HSPR ) − 1. This proves the theorem.
Remark 4.2.15. Note that in this proof we showed that:
(1) for any Noetherian local ring the Krull dimension is equal to the Chevalley dimension;
(2) the degree of the Hilbert-Samuel polynomial of an R–module M does not depend on
the choice of the filtration on M , and is smaller than or equal to the dimension of R.
(Simply replace in the proof R by M .)
Theorem 4.2.16 (Krull’s Principal Ideal Theorem). Let R be a Noetherian ring, f ∈ R,
and p ⊃ (f ) be a minimal associated prime ideal of (f ). Then ht(p) ≤ 1. If moreover f
is an active element of R, for example a nonzerodivisor, then ht(p) = 1. In particular,
dim(R/(f ) = dim(R) − 1.
Proof. Consider an irredundant primary decomposition
(f ) = q1 ∩ · · · ∩ qr .
3 Note that if R is an integral domain, then S = R, because p0 = (0) is a prime ideal.
√ √
So qi = p for some i. Since p is minimal it follows that for all j 6= i we have qj 6⊂ p.
Consider the local ring Rp , with maximal ideal pRp . So we get the primary decomposition
(f )Rp = qi Rp .
Therefore (f )Rp is a primary ideal in Rp , and is even pRp –primary. It follows from the
definition that the (Chevalley) dimension of Rp is at most one. Thus ht(p) ≤ 1, by using
the Krull dimension and the first part of Remark 4.2.15.
Suppose moreover that f is an active element. If ht(p) = 0, then p is a minimal
prime ideal. As f ∈ p, this is in contradiction to the assumption that f is an active
element.
For a different proof of Krull’s Principal Ideal Theorem, which does not use the
Hilbert-Samuel function, see Exercise 4.2.32.
Definition P4.2.17. Let (R,m) be a Noetherian local ring of dimension e. We write
HSPR (t) = ek=0 ak tk . Then m(R) := e! · ae is called the multiplicity of R .
If R = OX,x is the local ring of a singularity (X,x), then we also write m(X,x) for
m(OX,x ) and call this the multiplicity of (X,x), or the multiplicity of X in x.
Remark 4.2.18. The multiplicity m(X,x) is positive, because m(X,x) e! is the coefficient
of the monomial of highest degree in HSPX,x and HSPX,x (d) > 0 for d sufficiently large.
We leave it is as Exercise 4.2.25, to show that m(X,x) is indeed an integer.
Examples 4.2.19.
(n+d−1)···d
(1) Let (X,x) = (C n , 0). Then HSPX,x (d) = n+d−1
n = n! = 1 n
n! d + . . . , which
is a polynomial in d of degree n. Therefore, m(X,x) = 1.
(2) Consider the A2 –singularity with local ring R = C {x,y}/(y 2 − x3 ). Consider an
element f ∈ C {x,y}. Then modulo (y 2 − x3 ) f has a representative Pf which is a
polynomial in y of degree at most one. It is even true that ord(Pf ) ≥ ord(f ), see
Exercise 3.2.19. So we get that R/md = C {x,y}/(y 2 − x3 ,xd−1 y,xd ), and it follows
that HSPR (d) = 2d − 1. So the cusp singularity has dimension one, and multiplicity
2.
We now generalize Example 4.2.19(2).
Lemma 4.2.20. Let f ∈ C {x1 , . . . ,xn }, ord(f ) = m and OX,x = On /(f ). Then
Xm
n+d−j−1
HSPX,x (d) = .
j=1
n−1
In particular the dimension of (X,x) is n−1, and the multiplicity is m. Note that we have
a Noether normalization On−1 ⊂ On /(f ) = OX,x , so that In particular, if f irreducible,
the multiplicity of (X,x) is equal to the degree of the field extension Q(OX,x ) ⊃ Q(On−1 ).
Proof. Without loss of generality we may suppose that f is a Weierstraß polynomial
in xn of degree equal to m = ord(f ). As in the example, any element g ∈ OX,x has a
representative Pg which is a polynomial in xn of degree at most m − 1. Furthermore,
ord(g) ≤ ord(Pg ). Now HSR (d) = dimC C {x1 , . . . ,xn }/((f ) + md ) is equal to
dimC C {x1 , . . . ,xn }/(f, {xα : α1 + . . . + αn = d, αn < m}).
For d > m a basis of this vector space is B = ∪m

j=1 Bj with Bj = {xα : αn = j − 1, α1 +
n+d−j−1

. . . + αn−1 ≤ d − j}. By Example 4.2.2 Bj has n−1 elements.
We now turn our attention to a geometric interpretation of the multiplicity. First
we need a lemma.
Lemma 4.2.21. Let (X,x) be an irreducible germ of an analytic space defined by a
prime ideal p ⊂ C {x1 , . . . ,xn }. Let Ok ⊂ OX,x := C {x1 , . . . ,xn }/p be a general Noether
normalization, that is, there exist monic polynomials Pj ∈ C {x1 , . . . ,xj−1 }[xj ] ∩ p for
j = k + 1, . . . ,n, such that
• degxj (Pj ) = ord(Pj ) =: nj ,

• Q(OX,x ) = Q(Ok )[xk+1 ]/(Pk+1 ).
Such a general Noether normalization exists according to Exercise 3.3.30. Let m be the
maximal ideal of OX,x , and mk be the maximal ideal of Ok . Consider the filtration {mi }
of OX,x Then {mi } is a stable mk –filtration.
Proof. Obviously, {mi } is an mk –filtration. Let f ∈ C {x1 , . . . ,xn } be a power series of
order t. It follows from Exercise 3.3.31 that modulo p f is equivalent to a polynomial
X νk+1
Pf = fνk+1 ,...,νn xk+1 · · · xνnn , fνk+1 ,...,νn ∈ Ok .
νj <nj
P
Moreover, Pf has order at least t. So we see that, if t ≥ j≥k+1 nj , we can write every
element of mt+1 as an element of mk · mt . This is what we needed to show.
Example 4.2.22. C {x} ⊂ C {x,y}/(y 2 − x) = OX,x = ∼ C {y} is not a general Noether
normalization (in the sense of 3.3.30) because degy (y 2 − x) = 2 > ord(y 2 − x) = 1. Now,
mk = (x), m = (x,y) and m2i = (y 2i ) = (xi ) implies mk · m2i = (xi+1 ) = m2i+2 6= m2i+1
for all i. This shows that {mi } is not a stable mk –filtration of OX,x .
Proposition 4.2.23. Let (X,x) be an irreducible germ of an analytic space, and Ok ⊂
OX,x be a general Noether normalization. Then
m(X,x) = [Q(OX,x ) : Q(Ok )],
that is, the multiplicity is equal to the degree of the field extension Q(OX,x ) ⊃ Q(Ok ).
Proof. We apply the Local Parametrization Theorem 3.4.14, in order to reduce the
statement to the case of a hypersurface singularity. This case was already considered
in Lemma 4.2.20. Let ∆ be the discriminant of P := Pk+1 . Let OX,x := On /p, OX ′ ,x′ :=
Ok [xk+1 ]/(P ), and K = OX,x /OX ′ ,x′ . It was shown in the first part of the Local
Parametrization Theorem that p ∩ Ok [xk+1 ] = (P ) so that we have an exact sequence of
Ok –modules.
(4.3) 0 −→ Ok [xk+1 ]/(P ) −→ On /p −→ K −→ 0.
Let mk be the maximal ideal of Ok . We consider the following filtrations on these three
Ok –modules.
• {mik OX,x } on OX,x ,

• {mik OX,x ∩ OX ′ ,x′ } on OX ′ ,x′ ,
• {mik K} on K.
These are all stable mk –filtrations, where the second one follows from the Artin-Rees
Lemma 4.2.13. It follows from exact sequence (4.3) that
HSP{mik OX,x } = HSP{mik OX,x ∩OX ′ ,x } + HSP{mik K} .
Let m be the maximal ideal of OX,x , and m′ be the maximal ideal of OX ′ ,x′ . Then the
i
filtrations {mi } and {m′ } are mk –stable by Lemma 4.2.21.
It was noted in 3.4.16, that as a consequence of the Local Parametrization Theorem,
the module K is in fact an Ok /∆a –module for some a ∈ N. The ring Ok /∆a has dimension
k − 1. It follows from the second part of Theorem 4.2.6, and the second part of Remark
4.2.15 that the Hilbert-Samuel polynomial of K has degree at most k − 1. By definition
m(X,x) is determined by the leading term of HSP{mi } , and m(X ′ ,x′ ) is determined by
the leading term of HSP{m′i } , which both have degree k, and those are independent of
the stable mk –filtration by Theorem 4.2.6. From these two facts it follows that m(X,x) =
m(X ′ ,x′ ). So we reduced the problem to the case of an irreducible hypersurface, as
promised.
We finally come to the geometric interpretation of multiplicity.

Theorem 4.2.24. Let (X,x) ⊂ (C n , 0) be an irreducible a germ of an analytic space,
and let dim(X,x) = k. There exists a Zariski open subset U ⊂ Mn−k,n (C )4 with the
following property: given M ∈ U , there exists
e of (X,x) in the open ball Bε with center 0 and
• an ε > 0 and a representative X
radius ε,
• an open and dense subset V of Bε ,

such that for p ∈ V the plane defined by
x1
!
M .. = pt
.
xn
e in exactly m(X,x) points.

intersects X
To put it more down to earth, if k = dim(X,x), a general n − k–dimensional plane
e of (X,x) in exactly m(X,x) points.
near the origin cuts a representative X
Proof. Let Ue ⊂ Gln (C ) be the Zariski open subset such that for Mf∈U e the correspond-
ing linear coordinate change defines, via the projection π : C n −→ C k onto the first k
coordinates, a general Noether normalization (X,x) −→ (C k , 0).
Let M be the part of Mf consisting of the first n−k rows. This part of M
f corresponds
f
to the direction of the projection π. Let U = {M | M ∈ U }.e
4 Mn−k,n (C ) is the vector space of (n − k) × n–matrices with entries in C .

Choosing M ∈ U restricts the situation to the case that a general primitive Noether
normalization OX,x ⊃ Ok is given and ! we may assume OX,x = C {x1 , . . . ,xn }/p ⊃
1 0 0 ... 0
C {x1 , . . . ,xk } and M = .. .. .. .
. . .
0 1 0 ... 0
x1
!
M .. =0
. e
xn X
p
Ck
Let V = {p | π(p) 6∈ D}, where D is the discriminant, given by V (∆), with ∆ the
discriminant of the minimal polynomial of the field extension Q(OX,x ) ⊃ Q(Ok ). We
! now x1
apply 3.4.16 (1) to obtain that the (n−k)–dimensional plane E defined by M .. = p,
.
xn
( x1 ! )
..
that is E = . : x1 = p1 , . . . ,xk = pk = π −1 π(p) , intersects a representative of
xn
(X,x) in [Q(OX,x ) : Q(Ok )] points. Now [Q(OX,x ) : Q(Ok )] is equal to m(X,x) by
Proposition 4.2.23.
Exercises
` ´
4.2.25. Define for all k ∈ Z the polynomial kt := k! 1
t(t − 1) · · · (t − k + 1) ∈ Q[t]. For f ∈ Q[t]
define ∆f by ∆f (t) := f (t + 1) − f (t) ∈ Q[t].
` ´ ` t ´
(1) Prove that ∆ kt = k−1 .
` ´ ` t ´
(2) For any f ∈ Q[t] we can write f (t) = c0 kt + c1 k−1 + . . . + ck for suitable ci ∈ Q. Show
this.
(3) Suppose moreover that f (n) ∈ Z for n ≫ 0, and k = deg(f ). Prove that ci ∈ Z for
i = 0, . . . ,k.
(Hint: Use induction on k, and consider ∆f .)
(4) Let f : N −→ N be any function. Suppose that d 7→ ∆f (d) = f (d+1)−f (d) is a polynomial
Q(d) for d ≫ 0 of degree k − 1. Prove that d 7→ f (d) is, for d ≫ 0, a polynomial of degree
k.
` t ´ ` t ´ ` ´ ` t ´
(Hint: Write ∆f (t) = c0 k−1 + c1 k−2 + . . . + ck−1 , and consider P = c0 kt + c1 k−1 +
`t´
. . . + ck−1 1 . Prove that ∆(f − P )(d) is a constant for d ≫ 0.)
4.2.26. Compute the Hilbert-Samuel function and polynomial of R = C {x,y,z}/(y 2 − xz,z 3 −

x5 ,z 2 − x2 y). Check the result by using the computer algebra system Singular.
4.2.27. Let I = (xz − y 2 , xw − yz, yw − z 2 ) ⊂ C {x,y,z,w}. Let (X,x) be the germ at 0 defined
by I. Compute the Hilbert-Samuel polynomial HSPX,x .
4.2.28. Let I ⊂ C [x1 , . . . ,xr ] = C [x] be an ideal, define C [x]≤s = {f | deg(f ) ≤ s} and I≤s =
I ∩ C [x]≤s . Define the affine Hilbert function by AHI (n) = dimC C [x]≤n /I≤n .
Pd `´
Prove that for sufficiently large n AHI (n) is a polynomial API (n) = bi si , bi ∈ Z, bd > 0.
i=0
(Hint: Let I h ⊂ C [t,x1 , . . . ,xr ] be the homogenization of I. Prove that (with the canonical
grading of C [t,x1 , . . . ,xr ]) HC [t,x1 ,...,xr ]/I h (n) = AHI (n).)
4.2.29. With the notations of Exercise 4.2.28 prove that deg API = dim C [x1 , . . . ,xr ]/I.
(Hint: Let I h be as in 4.2.28. Prove that
dim(C [t,x1 , . . . ,xr ]/I h ) = dim(C [x1 , . . . ,xr ]/I) + 1,
by using Noether normalization.)
4.2.30. Prove the following version of the Artin-Rees Lemma:

Let (R,m) be a local ring, q be an m–primary ideal, and M be a finitely generated module.
If N ⊂ M a submodule then there exists an n0 such that
` ´
(qn M ) ∩ N = qn−n0 (qn0 M ) ∩ N .
4.2.31. Prove: if {Mn }, {M fn } are stable q–filtrations of M then there exists an n0 such that
fn+n0 ⊂ Mn and M
M fn+n0 ⊂ Mn for all n ∈ N.
fn = qn M for all n.)
(Hint: It suffices to prove this for the special case M
4.2.32. Give a different proof of Krull’s Principal Ideal Theorem 4.2.16 along the following
steps.
(1) Reduce the second statement to the first, by using the characterization of zerodivisors.
(2) By looking at the localization Rp , reduce to the case that R is local, with maximal ideal
m = p. From now one we suppose this.
(3) From now suppose q 6= m is a prime ideal. Show that it suffices to show that height(q) =
dim Rq = 0.
(4) Show that it suffices to show that qRq is nilpotent.
(5) Show that in R/(f ) every descending chain of ideals becomes stationary using the fact
that mρ ⊂ (f ) for some ρ.
(6) Consider the symbolic powers q(n) := qn Rq ∩ R, see Exercise 1.3.23. Show that the chain
of ideals
(f ) + q(1) ⊃ (f ) + q(2) ⊃ (f ) + q(3) ⊃ . . .
becomes stationary, that is, there exists an n ∈ N with (f ) + q(n) = (f ) + q(n+1) .
(7) Prove that for this n we have q(n) = (f )q(n) + q(n+1) . (Use the second part of Exercise
1.3.23.)
(8) Apply Nakayama to see that q(n) = q(n+1) .
(9) Use the definition of symbolic power and Nakayama to see that qn Rq = 0.
(10) Deduce the theorem.
S
4.2.33. Let (X,x) be a germ of an analytic space and (X,x) = n i=1 (Xi ,x) be
Pan irreducible
decomposition. We set I = {i : dim(Xi ,x) = dim(X,x)}. Prove that m(X,x) = i∈I m(Xi ,x).
(Hint: Study the situation (X,x) = (X1 ,x) ∪ (X2 ,x), with (Xi ,x) reduced and defined by the
ideals Ii ⊂ On . Compare the Hilbert-Samuel functions by looking at the exact sequence
0 → On /I1 ∩ I2 → On /I1 ⊕ On /I2 → On /I1 + I2 → 0.
One may assume that dim(On /I1 + I2 ) < max{dim(On /I1 ), dim(On /I2 )}.)
4.3 Regular Local Rings and the Jacobian Criterion
We now study regular and singular points of analytic spaces. According to Definition
3.3.11 a point x of an analytic space X is called regular if a neighborhood of this point
is a complex submanifold of C n . In particular, the local ring OX,x of a smooth point
x of X is isomorphic to C {x1 , . . . ,xk } for some k, which is an integral domain. Since
for a reducible germ (X,x) we have that OX,x is not an integral domain the following
proposition follows.
Proposition 4.3.1. Let X be an analytic space, x ∈ X, and (X,x) be a reducible germ.
Then x ∈ Sing(X). In particular, if X =X1 ∪ · · · ∪ Xr with Xi 6⊂ Xj for i 6= j, then
S S
Sing(X) = i Sing(Xi ) ∪i6=j Xi ∩ Xj .
It might not be a big surprise for the reader that “almost all” points of an analytic
space are smooth. The proof however is a little bit tricky. For an irreducible space this is
proved by means of the Local Parametrization Theorem 3.4.14 (cf. Remark 3.4.16 (2)),
which essentially allows us to reduce the statement to the hypersurface case.
Proposition 4.3.2. Let U be an open subset of C n , and X be an analytic subset of U .
Then the set of singular points Sing(X) is contained in a proper analytic subset of X.
Proof. This is a local statement, that is, for all x ∈ X we have to prove it for a neighbor-
hood of x ∈ X. If (X,x) is an irreducible germ, it was proved in the Local Parametrization
Theorem that there exists a hypersurface (D,x) in (X,x), and representatives X and D
such that X \ D is a complex submanifold of an open set in C n .
Now if (X,x) is reducible, we can take an irreducible decomposition (X1 ,x) ∪ · · · ∪
(Xr ,x) of (X,x). For suitable representatives, we have X = X1 ∪ . . . ∪ Xr with Xi 6⊂ Xj
for i 6= j. We therefore have that Xi ∩ Xj for i 6= j is a proper analytic subset of X.
We may assume that Sing(Xi ) ⊂ Vi , Vi a proper analytic subset of Xi . On the other
hand, we have Sing(X) = ∪i Sing(Xi ) ∪ ∪i6=j (Xi ∩ Xj ) by the previous proposition.
Now ∪i6=j (Xi ∩ Xj ) is a proper analytic subset ofX. Therefore, if x is not contained in
the proper analytic subset ∪i Vi ∪ ∪i6=j (Xi ∩ Xj ) then x is a smooth point of X.
Our next goal would be to show that the set of singular points in fact forms an
analytic subset of X. However, as we will explain later, we are not able to prove this
in this section, since we will need the Oka-Cartan Theorem, see 6.3.2, concerning the
coherence of the ideal sheaf. So the fact that the singular locus is an analytic subset is
surprisingly deep. What we will show is, that we characterize the regular points by means
of the Jacobian Criterion.
Definition 4.3.3. Let (R,m) be a Noetherian local ring. The embedding dimension
edim(R) of R is defined by
edim(R) := dimR/m (m/m2 ).
If (X,x) is a germ of an analytic space, then edim(X,x) := edim(OX,x ). The ring R is

called a regular local ring, if the embedding dimension is equal to the dimension, that is,
if
edim(R) = dim(R).
4.3 Regular Local Rings and the Jacobian Criterion 149
By Nakayama’s Lemma, the embedding dimension is the minimal number of gener-

ators of the maximal ideal. As the dimension of R is equal to the Chevalley dimension
of R, see Remark 4.2.15, it follows that the number of generators of the maximal ideal
of a Noetherian local ring cannot be less than dim(R).
Example 4.3.4. Consider the power series ring C {x1 , . . . ,xn }. This is a ring of dimen-
sion n. The maximal ideal (x1 , . . . ,xn ) is generated by n = dim(C {x1 , . . . ,xn }) genera-
tors, so the power series ring is a regular local ring. Similarly, the formal power series
ring C [[x1 , . . . ,xn ]], and the localization C [x1 , . . . ,xn ]m , m = (x1 , . . . ,xn ) are regular
local rings.
The following lemma motivates the name embedding dimension.
Lemma 4.3.5. Let (X,x) be a germ of an analytic space, and let n ∈ N be the min-
imal number such that there exists an injective analytic map (X,x) ֒→ (C n , 0). Then
edim(X,x) = n. In particular, (X,x) is smooth at x if and only if OX,x is a regular local
ring.
Proof. Suppose ϕ : (X,x) ֒→ (C n , 0). So we have a surjective C –algebra map ϕ∗ :
OC n , 0 ։ OX,x . In particular ϕ induces a surjective map of vector spaces mC n ,0 /m2C n ,0 ։
mX,x /m2X,x . This shows that the minimal number of generators of mX,x is at most n, so
that edim(X,x) ≤ n. On the other hand, let s = edim(X,x). Then s is the number of
generators of mX,x , and it was proved in Corollary 3.2.13 that OX,x is a quotient of
C {x1 , . . . ,xs }. By dualizing we get an injective analytic map ϕ : (X,x) ֒→ (C s , 0), so
that s ≥ n. This proves the lemma.
We now come to the Jacobian criterion.
Theorem 4.3.6. Let (X, 0) ⊂ (C n , 0) be a germ of an analytic space, and let the ideal
of (X, 0) be generated by f1 , . . . ,fs ∈ C {x1 , . . . ,xn }. We denote by rank0 (f1 , . . . ,fs ) the
rank of the Jacobian matrix

∂fi ∂fi
(0) := (0) .
∂xj ∂xj 1≤i≤s,1≤j≤n
Then
edim(X, 0) + rank0 (f1 , . . . ,fs ) = n.
Proof. Put e := edim(X, 0) and r = rank0 (f1 , . . . ,fs ).
Step 1. Suppose we have functions f1 , . . . ,fs that generate the maximal ideal m of On .
Then we claim that the Jacobian matrix

∂fi
(0)
∂xj 1≤i≤s,1≤j≤n
has rank n. It follows from Nakayama’s Lemma that s ≥ n, and in fact already n of the
fi generate the Pmaximal ideal m. Without loss of generality, f1 , . . . ,fn are generators.
Now write fi = j Aij xj modulo m2 , for Aij ∈ C . As the fi generate m, we conversely
P
have xi = j Bij fj mod m2 , for some Bij ∈ C . Therefore (Aij ) is an invertible matrix

∂fi
with inverse (Bij ). Now ( ∂x j
(0) = (Aij ), so the Jacobian matrix has maximal rank n.
Step 2. We now prove r ≥ n − e. Let g1 , . . . ,ge ∈ OC n ,0 project to generators of the

maximal ideal mX,0 . Now
C ∼
= OX,0 /mX,0 = OC n ,0 /(f1 , . . . ,fs ,g1 , . . . ,ge ).
So f1 , . . . ,fs ,g1 , . . . ,ge generate the maximal ideal of OC n , 0 . Hence the Jacobian matrix
of (f1 , . . . ,fs ,g1 , . . . ,ge ) has rank n by step 1. Deleting a column reduces the rank by at
most one. So deleting the columns corresponding to g1 , . . . ,ge shows that the rank r of
∂fi
the Jacobian matrix ( ∂xj (0) is at least n − e.
Step 3. It remains to show that e ≤ n − r. This will be proved as an application of the

Implicit Mapping Theorem. After renumbering the fi and xj we may suppose that
∂f
i
det (0) 6= 0.
∂xj n−r+1≤i,j≤n
We can therefore apply the Implicit Mapping Theorem 3.3.6 and deduce that there exist
functions ϕn−r+1 (x1 , . . . ,xn−r ), . . . ,ϕn (x1 , . . . ,xn−r ) such that

fi x1 , . . . ,xn−r ,ϕn−r+1 (x1 , . . . ,xn−r ), . . . ,ϕn (x1 , . . . ,xn−r ) = 0,
for i = n − r + 1, . . . ,n. This shows that (X, 0) is a subgerm of the graph of the map
ϕ : C n−r −→ C r , that is, a subgerm of

V xn−r+1 − ϕn−r+1 , . . . ,xn − ϕn , 0 .
This is a smooth germ of (embedding) dimension n − r. Thus e ≤ n − r, as was to be

proved.
Remark 4.3.7. Let X be an analytic subset of an open subset U of C n , say locally
defined by holomorphic functions f1 , . . . ,fs on U . Suppose that for all x ∈ X, the germ
(X,x) has dimension n − c. It follows directly from the Jacobian Criterion and Lemma
4.3.5 that the singular locus of X is contained in the zero set of the c–minors of the
Jacobian matrix, which is an analytic set. However, a priori these two sets need not be
equal. Indeed, to complete the proof, one needs that for all x ∈ X, the classes of f1 , . . . ,fs
generate the ideal of (X,x), so it has to be a radical ideal. Now even if for some x ∈ X,
the ideal generated by f1 , . . . ,fs is radical in OC n ,x , it is not clear that for all y in a small
neighborhood of x, the ideal generated by the classes of f1 , . . . ,fs is radical in OC n ,y .
That this is in fact the case, is surprisingly difficult to prove. It is equivalent to the
Coherence Theorem of Oka-Cartan, see 6.3.2. From this Coherence Theorem it therefore
follows that the singular locus of an analytic set itself is analytic.
Definition 4.3.8. Let (X,p) be a germ of an analytic subset in C n , and let f1 , . . . ,fs
generate the ideal of (X,p). The (Zariski) tangent space TX,p of X at the point p =
(p1 , . . . ,pn ) is the affine space in C n defined by
n Xn
∂fi o
TX,p = p + q = (q1 , . . . ,qn ) ∈ C n : (p) · qj = 0, i = 1, . . . ,s .
j=1
∂xj
Lemma 4.3.9. TX,p is independent of the choice of f1 , . . . ,fs .

Proof. Suppose that g1 , . . . ,gm also generate I (X,p). Then we have an equality
s
X
gi = ξik · fk
k=1
for some ξik ∈ C {x1 , . . . ,xn }. We differentiate. By the product rule, and because fi (p) = 0
for all i, it follows
X s
∂gi ∂fk
(p) = ξik (p) (p).
∂xj ∂xj
k=1
P
n
∂fi
Now let (q1 , . . . ,qn ) ∈ C n be such that ∂xj (p)qj = 0 for i = 1, . . . ,s. Then
j=1
n
X s
X X n
∂gi ∂fk
(p)qj = ξik (p) (p)qj = 0.
j=1
∂xj j=1
∂xj
k=1
This shows one inclusion. The other inclusion follows by symmetry.

∂fi
Note that the rank of ∂x j
(p) is n − edim(X,p) by 4.3.6, so that the tangent space
has dimension edim(X,x). In particular, by 4.3.3, 4.3.5 and 3.4.20 we have the following.
Theorem 4.3.10. Let (X,p) be a germ of an analytic space. Then the following condi-
tions are equivalent.
(1) p is a regular point of X.
(2) OX,p is a regular local ring.
(3) OX,p ∼
= C {x1 , . . . ,xk } for some k.
(4) dim(OX,p ) = dimC (TX,p ).
Examples 4.3.11.
(1) Let X = {(x,y) ∈ C 2 : x2 − y = 0}. Then X is an analytic subset of C 2 , 0 =

(0,0) ∈ X and TX, 0 = {(q,0) : q ∈ C } ∼
= C.
TX,0
0
(2) Let X = {(x,y) ∈ C 2 : x3 − y 2 = 0}. Then again X is an analytic subset of

C 2 and 0 = (0,0) ∈ X, but TX,0 = C 2 . On the other hand, (1,1) ∈ X and
TX,(1,1) = {(q, 23 q − 21 ), q ∈ C }.
TX,(1,1)
Theorem 4.3.12. Let (X,p) ⊂ (C n ,p) and (Y,q) ⊂ (C m ,q) be germs of analytic subsets
and f = (f1 , . . . ,fm ) : (X,p) −→ (Y,q) be a germ of a holomorphic map. Then there
exists a canonical linear map
Tf,p : TX,p −→TY,q

∂fi
p + a 7−→q + (p) · at .
∂xj
If g : (Y,q) −→ (Z,r) is a further germ of a holomorphic map, then
Tg◦f,p = Tg,q ◦ Tf,p .
As TId,p = IdTX,p , it follows that if f is an isomorphism, then Tf,p is an isomorphism.

Proof. Without loss of generality, p = 0 and q = 0. Let I (X, 0) = (g1 , . . . ,gs ), and
I (Y, 0) = (h1 , . . . ,ht ). Then
Xn
∂gi
TX,0 = {a ∈ C n : (0)aj = 0, i = 1, . . . ,s}
j=1
∂xj
m
X ∂hi
TY,0 = {b ∈ C m : (0)bj = 0, i = 1, . . . ,t}.
j=1
∂yj
Now h1 ◦ f, . . . ,ht ◦ f vanish on X, so they are in the ideal generated by g1 , . . . ,gs . Hence
there exist sik ∈ OC n ,0 such that
s
X
hi ◦ f = sik gk .
k=1
We differentiate and get, using the chain rule,

m
X ∂hi ∂fk ∂
0 0 = (hi ◦ f )(0) =
∂yk ∂xj ∂xj
k=1
s
X s
X X s
∂sik ∂gk ∂gk
(0)gk (0)+ sik (0) (0) = sik (0) (0).
∂xj ∂xj ∂xj
k=1 k=1 k=1
Pn ∂gk
Now consider a point a = (a1 , . . . ,an ) ∈ TX,0 . So j=1 (0)aj = 0. It follows that
∂xj
m
X Xn
∂hi ∂fk
(0) (0)aj = 0.
∂yk j=1 ∂xj
k=1

∂fk
But this says exactly that (0) · at ∈ TY,0 .
∂xj
The second statement follows from the chain rule, and is left as an exercise.
Examples 4.3.13.
(1) Let π : C n −→ C k be the projection given by π(x1 , . . . ,xn ) = (x1 , . . . ,xk ), and
p = (p1 , . . . ,pn ) ∈ C n . Then
Tπ,p : TC n ,p = p + C n −→ TC k ,π(p) = π(p) + C k
is defined by Tπ,p (p + a) = π(p) + π(a) for a ∈ C n .

(2) Let I = {f1 , . . . ,fm } ⊂ C {x1 , . . . ,xn } be an ideal and i : V (I), 0 −→ (C n , 0) be
Pn ∂fi
the canonical inclusion. Then TV (I),0 = {a = (a1 , . . . ,an ) ∈ C n : j=1 (0)aj =
∂xj
0, i = 1, . . . ,m}, and Ti,0 : TV (I),0 −→ C n = TC n ,0 is the canonical inclusion.
(3) Let f : C −→ C 2 be defined by f (t) = (t2 ,t3 ),
Then Tf,0 : TC ,0 = C −→ TC 2 ,(0,0) = C 2 is the zero map. On the other hand

Tf,2 : TC ,2 = 2 + C −→ TC 2 ,(4,8) = (4,8) + C 2 is defined by
Tf,2 (2 + a) = (4,8) + (4a,12a).
The following theorem we only need in Chapter nine.

Theorem 4.3.14. Let X,Y ⊂ C n be irreducible affine algebraic sets, and ϕ : X −→ Y
be a dominant map, see 2.3.11. Then there exists a point p ∈ X such that with q = f (p)
Tϕ,p : TX,p −→ TY,q
is surjective.
Proof. In the proof of 2.3.12, it was shown that there exists an f ∈ C [Y ] ⊂ C [X], and
an r ∈ N such that ϕ factorizes on X \ V (f ) as follows
e
ϕ π
X \ V (f ) −→ Y \ V (f ) × C r −→ Y \ V (f ).
Moreover C [X]f is a finitely generated C [Y ]f [x1 , . . . ,xr ]–module. Here π is the projection
on the second factor, and thus for all q ′ ∈ Y \ V (f ) and a ∈ C r , the tangent map Tπ,(q′ ,a)
is surjective. So it suffices to show that there exists a point p ∈ X \ V (f ) with Tϕ,p e is
surjective, as an application of 4.3.12. As both X \ V (f ) and Y \ V (f ) × C r are affine,
see 2.3.10, we reduced the problem to the case of a finite map ϕ : X −→ Y of affine
algebraic sets, such that C [X] is a finitely generated C [Y ]–module.
So take a smooth point q ′ ∈ Y . By Theorem 2.3.9 the map ϕ is surjective, so there
exists a point p′ in ϕ−1 (q ′ ). It follows from 3.4.24 that the germ of the analytic map
ϕ : (X,p′ ) −→ (Y,q ′ ) is finite. Thus this is a Noether normalization, as by assumption q ′
is a smooth point of Y . It obviously suffices to prove the statement for an irreducible germ
(X,p′ ). It was proved in the Local Parametrization Theorem, see 3.4.14 that for “almost
all” points p in an open neighborhood of p′ the germ of the map ϕ : (X,p) −→ (Y,q) is
an isomorphism. In particular the tangent map Tϕ,p is an isomorphism. This concludes
the proof of the theorem.
For later applications, for example in the interpretation of normal rings, and in the
proof of Artin’s Approximation Theorem, we need a more general version of the Jacobian
Criterion.
Theorem 4.3.15 (Jacobian Criterion for Smoothness). Let I = (f1 , . . . ,fm ) ⊂
C {x1 , . . . ,xn }, p be a prime ideal associated to I, and q ⊃ p be a prime ideal containing
p. Then5

∂fi
C {x1 , . . . ,xn }q /Iq is a regular local ring ⇐⇒ rank ∂x j
mod q = ht(p).

∂fi
Furthermore, if ht(p) = k and rank ∂x j
mod q 1≤i≤k
= k then f1 , . . . ,fk generate
the ideal Iq .
We first discuss how to interpret this theorem.

∂fi
Examples 4.3.16. (1) For example, take q = m the maximal ideal. Then ∂x mod m
j
∂fi
is just the Jacobian matrix ∂x j
(0) . As all elements in C {x1 , . . . ,xn } \ m are already
invertible in C {x1 , . . . ,xn } we have the equality C {x1 , . . . ,xn }m /Im = C {x1 , . . . ,xn }/I.
It follows that C {x1 , . . . ,xn }/I is a regular local ring if and only the Jacobian matrix
has maximal rank. This we have already seen in 4.3.6.
(2) Suppose that C {x1 , . . . ,xn }q /Iq is a regular local ring. Put k = ht(p). Then k is the
codimension of V (p). Suppose p is a minimal
associated prime of I. Consider
the k–minors
∂fi ∂fi
∆1 , . . . ,∆s of the Jacobian matrix ∂x j
. Then rank ∂x j
mod q = ht(p) exactly
means that there exists a minor ∆i which is not in q. It follows that for all a in an open
subset of V (q) the variety
V (I) is smooth at the point a. Namely V (∆1 , . . . ,∆s ), 0 ∩
V (q), 0 $ V (q), 0 is a proper analytic subset. Now take representatives of all spaces
involved. It follows that for all a ∈ V (q)\V (∆1 , . . . ,∆s ) there is also a ∆i with ∆i (a) 6= 0.
“` ´”
5 ∂fi
Here rank ∂x mod q is the rank of the matrix over the quotient field Q(C {x1 , . . . ,xn }/q).
j
(3) Note that in the special case I = q = p, that C {x1 , . . . ,xn }p /pp is a field. As a field
obviously is a regular local ring, it follows that almost all points of V (p) are nonsingular.
This we have already seen in the Local Parametrization Theorem, see 3.4.14. In the proof
of the Jacobian criterion, this special case will be treated first and, in fact, in the proof
the Local Parametrization Theorem is used.
(4) We consider the explicit example of the Whitney umbrella defined by x2 − y 2 z = 0.
So p = I = (x2 − y 2 z). We take q = (x,y). Then q is the ideal of the singular locus. So
we expect that C {x,y,z}q /(x2 − y 2 z)q is not a regular local ring. Indeed, the minimal
number of generators of ideal (x,y)q is two, as one checks by using Nakayama’s Lemma.
However the dimension of C {x,y,z}q /(x2 − y 2 z)q is one.
On the other hand, looking at the y–axis, a “general” point of the y–axis is smooth.
We take the ideal q = (x,z), in which we localize. This is the same as localizing in the
multiplicative set S = {1,y,y 2 , . . .}. So y becomes a unit, and we can eliminate z. Hence
C {x,y,z}q /(x2 − y 2 z) ∼
= C {x,y}y , which one directly checks to be a regular local ring of
dimension one.
Before we can give the proof of the general Jacobian criterion, we need some basic
facts on regular local rings.
Theorem 4.3.17. Let (R,m) be a regular local ring. Then R is an integral domain.
Proof. The proof is by induction on the Krull-dimension of R.
Step 1. If dim(R) = 0, then edim(R) = 0, hence the maximal ideal has zero generators.
Thus m = 0, hence R = R/m is a field, which obviously has no zerodivisors.
Step 2. Now suppose dim(R) > 0. Let p1 , . . . ,ps be the minimal prime ideals of R. If we
prove that pi = (0), for some i, then i = s = 1 (because of minimality), and therefore
R = R/p1 = R/(0) is an integral domain.
As dim(R) > 0, it follows by Nakayama that m 6= m2 . By prime avoidance 1.1.13, we
can find an x ∈ m\m2 with x ∈ / ∪i pi . The last condition means that x is an active element
of R. By Krull’s
Principal Ideal Theorem 4.2.16, dim(R/(x)) = dim(R) − 1. Moreover,
m/ (x) + m2 is a vector space of dimension dim(R) − 1. Hence R/(x) is a regular local
ring, and by induction hypothesis it is an integral domain. Therefore, the ideal (x) is a
prime ideal. It cannot be a minimal prime ideal, since we have chosen x ∈ / ∪i pi . Therefore,
there must be a minimal prime ideal, say pi , which is contained in (x). Let y ∈ pi be
arbitrary. It follows that y = ax for some a ∈ R. As pi is prime, it follows a ∈ pi . Hence
we proved the equality xpi = pi . From Nakayama’s Lemma it follows that pi = 0, which
is what we had to prove.
Corollary 4.3.18. Let (R,m) be an r–dimensional regular local ring and x1 , . . . ,xi ∈ m.
The following conditions are equivalent.
(1) There exist xi+1 , . . . ,xr ∈ R such that m = (x1 , . . . ,xr ).

(2) The images of x1 , . . . ,xi in m/m2 are linearly independent over R/m.
(3) R/(x1 , . . . ,xi ) is an (r − i)–dimensional regular local ring.
Proof.
Step 1. We first show (1) =⇒ (2). The images of x1 , . . . ,xr generate m/m2 as R/m–vector
space. Since dimR/m (m/m2 ) = r they are linearly independent.
Step 2. (2) =⇒ (3). We choose xi+1 , . . . ,xr ∈ m such that the images of x1 , . . . ,xr form
a basis of m/m2 . Nakayama’s Lemma implies that m = (x1 , . . . ,xr ).
Now x1 is not a zero divisor of R by Theorem 4.3.17 . Thus dim R/(x1 ) = dim(R)−
1 by Krull’s Principal Ideal Theorem. The images of x2 , . . . ,xr generate the maximal ideal
in R/(x1 ). This implies that R/(x1 ) is regular. By induction we get that R/(x1 , . . . ,xi )
is regular.
Step 3. We finally show (3) =⇒ (1). Let m/(x1 , . . . ,xi ) be generated by the images of
xi+1 , . . . ,xr ∈ m. Then m = (x1 , . . . ,xi , xi+1 , . . . ,xr ).
Theorem 4.3.19. Let p be a prime ideal in C {x1 , . . . ,xn }. Then C {x1 , . . . ,xn }p is a
regular local ring.
Proof. The ring C {x1 , . . . ,xn }p has dimension ht(p). Consider a primitive Noether nor-
malization
C {x1 , . . . ,xk } −→ C {x1 , . . . ,xn }/p.
Then ht(p), which is the codimension of V (p), is equal to n − k. Hence we have to find
n − k generators of the maximal ideal pC {x1 , . . . ,xn }p . Let P ∈ C {x1 , . . . ,xk }][xk+1 ]
be the minimal polynomial of xk+1 ∈ C {x1 , . . . ,xn }/p, and ∆ ∈ C {x1 , . . . ,xk } be the
discriminant. In particular ∆ ∈ / p. Moreover, let Qj = xj ∆ − qj (xk+1 ) ∈ p for j =
k + 2, . . . ,n. Here the qj are polynomials with coefficients in C {x1 , . . . ,xk } which exist
according to the Finiteness of Normalization Theorem 1.5.19. Then Qn , . . . ,Qk+2 ,P ∈ p,
and we proved already in 3.4.14 that they generate the ideal pC {x1 , . . . ,xn }∆ . As ∆ ∈ / p,
they in particular generate the ideal pC {x1 , . . . ,xn }p .
Remark 4.3.20. It is quite generally true that localizations of regular local rings in
prime ideals are regular local rings. A proof of this more general result uses homological
methods, see for example [Eisenbud 1995], Corollary 19.14. The proof which we gave
works for the polynomial case too, and is a simple consequence of the Local Parametriza-
tion Theorem.
Proof of Theorem 4.3.15.
Step 1. We first consider the case (f1 , . . . ,fm ) = I = p = q.
The ring C {x1 , . . . ,xn }p /pp is a field and hence a regular local ring. Therefore, in
∂fi

this case we have to prove rank ( ∂x j
mod p)i≤m,j≤n = ht(p). Take a primitive Noether
normalization
C {x1 , . . . ,xk } −→ C {x1 , . . . ,xn }/p.
So we have to prove that the Jacobian matrix has rank n − k. We use the Local
Parametrization Theorem, and keep the notations of the proof of the previous Theo-
rem 4.3.19. We noted in the previous proof that the ideal pC {x1 , . . . ,xn }p is generated
by (P,Qk+2 , . . . ,Qn ). By Exercise 4.3.24 the rank of the Jacobian matrix is independent
of the generators of pC {x1 , . . . ,xn }p . As this ideal is generated by n − k elements, it
follows that the rank of the Jacobian matrix modulo p is at most n − k. Now the final
n − k columns of the Jacobian matrix for the generators (P,Qk+2 , . . . ,Qn ) look like
 
∂P
 ∂P ∂P  0 ... 0
 ∂xk+1
...  
 ∂xk+1 ∂xn   
   ∂qk+2 
 ∂Q ∂Qk+2   0 ... 0
 k+2  − ∂xk+1 ∆ 
 ...   
J :=  ∂xk+1 ∂xn  =  .. .. .
 . ..   . . 
 .   
 . .   .. .. 
 ∂Q
n ∂Qn    . . 

...  ∂qn 
∂xk+1 ∂xn − ∆
∂xk+1
∂P
So det(J) = ∆n−k−1 · ∂xk+1 . Because P is the minimal polynomial of the field extension
C {x1 , . . . ,xn }p /pp ⊃ Q(C {x1 , . . . ,xk }) we have ∂x∂P
k+1
6∈ p. So det(J) 6∈ p. This implies
that
∂fi
rank ( mod p)i≤m, j≤n = rank(J mod p) = n − k = ht(p).
∂xj
This shows the theorem in the special case I = p = q.
Step 2. Now suppose that C {x1 , . . . ,xn }q /Iq is a regular local ring. Then we claim that
(4.4) Iq = pq .
Indeed, p ⊃ I. By assumption, C {x1 , . . . ,xn }q /Iq is a regular local ring, so in particular

an integral domain by 4.3.17. So in the localization, C {x1 , . . . ,xn }q the ideal Iq is a prime
ideal. Formula (4.4) now follows from 1.4.50 (5), which says how a primary decomposition
behaves under localization.
So if we suppose that C {x1 , . . . ,xn }q /Iq is a regular local ring, we may suppose that
I = p is a prime ideal. Put a := ht(p), and b := ht(q). The dimension of C {x1 , . . . ,xn }q /Iq
is equal to b − a. It is not difficult to see, see 4.3.23, that the ideal pq in the ring
C {x1 , . . . ,xn }q can be generated by elements u1 , . . . ,ua ∈ q \ q2 . By 4.3.18 this sequence
can be extended to a regular sequence u1 , . . . ,ub ∈ q \ q2 which generates the ideal qq in
C {x1 , . . . ,xn }q . We apply step 1 and get
∂u
i
rank mod q i≤b, j≤n
= ht(q) = b,
∂xj
which is the maximal possible rank. Deleting the columns corresponding to ua+1 , . . . ,ub
we obtain ∂u
i
rank mod q i≤a, j≤n = a = ht(p),
∂xj
which is again the maximal possible rank. We deduce that the Jacobian matrix for the
f ’s also has maximal rank by applying again Exercise 4.3.24.

∂fi
Step 3. We now prove the converse. Therefore, assume rank ( ∂x j
mod q)i≤m, j≤n =
ht(p) = a. By renumbering, we may assume
∂f
i
rank ( mod q)i≤a, j≤a = a.
∂xj
We first show that f1 , . . . ,fa is part of a generating set for qq . Assume

a
X
ci f i ∈ q 2
i=1
for some c1 , . . . ,ca ∈ C . Then

∂ X
a Xa
∂fi
ci f i = ci ∈ q.
∂xj i=1 i=1
∂xj
∂f
i
As rank ( mod q)i≤a, j≤a = a it follows that ci = 0 for i = 1, . . . ,a. By Theo-
∂xj
rem 4.3.19 and Corollary 4.3.18, C {x1 , . . . ,xn }q /(f1 , . . . ,fa )q is a regular local ring. In
particular (f1 , . . . ,fa ) is a prime ideal in C {x1 , . . . ,xn }q . As f1 , . . . fa ∈ I ⊂ p we have
inclusions
(f1 , . . . ,fa )q ⊂ Iq ⊂ pq .
But as the outer ideals have the same height, and are both prime in C {x1 , . . . ,xn }q all
the ideals must be equal. In particular (f1 , . . . ,fa )q = Iq . Therefore, C {x1 , . . . ,xn }q /Iq
is a regular local ring. This proves the theorem.
Exercises
4.3.21. Prove that Tg◦f,p = Tg,q ◦ Tf,p , see 4.3.12.
4.3.22. Let X ⊂ C n , and Y ⊂ C m be algebraic set. Suppose f : X −→ Y is defined by
(f1 , . . . ,fm ). Consider the tangent map Tf,p : TX,p −→ TY,f (p) . Let ε be a new variable with
ε2 = 0. Suppose p + a ∈ TX,p , and Tf,p (p + a) = f (p) + b. Prove that f (p + εa) = f (p) + εb.
Make this to a precise statement!
4.3.23. Let (R,m) be a regular local ring, p ⊂ R a nonzero prime ideal.
(1) Suppose p is contained in m2 . Show that R/p is not a regular local ring.
(Hint: Use Krull’s Principal Ideal Theorem to prove that dim(R/p) < dim(R), and look
at the generators of the maximal ideal in R/p.)
(2) Suppose R/p is a regular local ring. Show that p can be generated by a regular sequence
u1 , . . . us , with ui ∈ m \ m2 .
4.3.24. Let q ⊂ C {x1 , . . . ,xn } be a prime ideal, and I = (f1 , . . . ,fm ) be a further ideal. Let
g1 , . . . ,gs ∈ I such that Iq = (g1 , . . . ,gs )q . Prove that
` ∂f ´ ` ∂g ´
rank ( ∂x i
j
mod q) = rank ( ∂x i
j
mod q) .
4.3.25. Let X ⊂ C n be an affine set. Prove that Sing(X) is an affine algebraic subset of X.
(Hint: Use the Primary Decomposition Theorem.)
4.3.26. Let X be an analytic space. Prove the Identity Theorem, cf. 3.1.9, for X.
(Hint: Use the fact that the singular points of X are contained in a proper analytic subset.)
4.3.27. Prove the following statement. Let (X, 0) ⊂ (C n , 0) be a germ of an analytic subset.
Let m ⊂ OX,0 be the maximal ideal. There is a canonical isomorphism of affine spaces TX,0 ∼ =
(m/m2 )∗ .
(Hint: Choose a representative Y ⊂ U of (X, 0), U ⊂ C n open, and f1 , . . . ,fs holomorphic
functions on U such that
“ Y =” {y ∈ U : f1 (y) = · · · = fs (y) = 0} and I (X, 0) = (f1 , . . . ,fs ).
∂fi
Then TX,0 = {a ∈ C n : ∂xj
(0) · at = 0}.
On the other hand, m is the maximal ideal in C {x1 , . . . ,xn }/(f1 , . . . ,fs ) generated by
x1 , . . . ,xn , with xi = xi mod (f1 , . . . ,fs ). Define φ : TX,0 −→ (m/m2 )∗ by φ(a)(xi ) = ai , for
a = (a1 , . . . ,an ). Pn Pn
Pn Prove that φ is P well-defined, P
n
that is
s
i=1 ci xi = 0 implies
2
i=1 ci ai = 0. Use that
i=1 c i xi = 0 implies i=1 ci x i + i=1 ξ i fi ∈ (x 1 , . . . ,x n ) for suitable ξi ∈ C {x1 , . . . ,xn }.
P
Prove that φ is an isomorphism using the fact that n ∂fi
j=1 ∂xj (0)xj = 0 in m/m2 .)
4.4 Normalization 159
4.4 Normalization of Germs of Analytic Spaces
The first aim of this section is to show that every germ of an analytic space has a
normalization. We start with the definition of a normal germ.
Definition 4.4.1. A germ of an analytic space (X,x) is called normal if the local ring
OX,x is normal.
The first property of normal germs is that they are irreducible.
Proposition 4.4.2. Let (X,x) a normal germ of an analytic space. Then (X,x) is irre-
ducible.
Proof. This is an immediate consequence of Theorem 1.5.7.
So we see that normality is a nontrivial condition. The first goal in this section is
to prove that for all germs there exists a normalization. But in order to reach this goal,
we have to allow more general spaces then germs of analytic spaces: we have to consider
multi-germs.
Definition 4.4.3. A multi-germ of analytic spaces (X,x) is a finite disjoint union (X,x) =
(X1 ,x1 ) ∪ . . . ∪ (Xr ,xr ) of germs of analytic spaces. The ring OX,x by definition is equal
to ⊕ri=1 OXi ,xi .
Let (Y,y) = (Y1 ,y1 )∪. . .∪(Ys ,ys ) be a further multi-germ. A map ϕ : (X,x) −→ (Y,y)
is given by a system of maps of ϕi : (Xi ,xi ) −→ (Yα(i) ,yα(i) ) for i = 1, . . . ,r, and
some α(i) ∈ {1, . . . ,s}. Such a map ϕ induces, and is induced by a C –algebra map
ϕ∗ : OY,y −→ OX,x . In the obvious way, one defines such a map to be finite, generic s to
1, etc.
A multi-germ (X,x) is called normal if OX,x is a normal ring.
Remarks 4.4.4.
(1) It is not so difficult to see that OX,x is a semi-local ring, see Exercise 4.4.16. This,
by definition, means that OX,x has only finitely many maximal ideals.
(2) It is an exercise, see 4.4.16, to show that a multi-germ (X,x) = (X1 ,x1 )∪. . .∪(Xr ,xr )
is normal if and only if (Xi ,xi ) is normal for i = 1, . . . ,r.
Now we can define what we mean by a normalization of a germ of an analytic space.
Definition 4.4.5. Let (X,x) be a germ of an analytic space. A normalization of (X,x)
e x) together together with a finite, generically 1 − 1 map
is a normal multi-germ (X,e
e x) −→ (X,x).
n : (X,e
Remark 4.4.6. It was proved in Exercise 3.4.40 that a generically 1 − 1 map ϕ between
irreducible germs of analytic spaces induces an isomorphism of quotient fields. This was a
simple consequence of the Local Parametrization Theorem. This holds more generally for
multi-germs, see Exercise 4.4.16. As a normalization should be generically 1−1, it follows
that the ring OX,e
e x is the normalization of the ring OX,x . So a normalization, assuming it
exists, is uniquely determined. Note that the normalization (X,ee x) has the same dimension
as (X,x). Furthermore, note that if (X,x) is irreducible, then the normalization (X,e e x)
must be a germ of an analytic space. This is because the total quotient ring of OX,x is
equal to the total quotient ring of OX,e
e x , which in case (X,x) is irreducible, is a field.
Examples 4.4.7.
(1) A smooth germ (C n , 0) is normal. This is because C {x1 , . . . ,xn } is a unique factor-
ization domain by Corollary 3.3.17, and unique factorization domains are normal,
see Theorem 1.5.5,6
(2) Consider the A1 –singularity, defined by the equation x2 − y 2 = 0. So OX,x =
C {x,y}/(x2 − y 2 ). The normalization (X,e e x) of (X,x) consists of two copies of
smooth spaces (X1 ,x1 ) with parameter t1 and (X2 ,x2 ) with parameter t2 . The
normalization map n is given by two maps n1 : (X1 ,x1 ) −→ (X,x) which sends t1
to (t1 ,t1 ), and n2 : (X2 ,x2 ) −→ (X,x) which sends t2 to (t2 , − t2 ).
(X1 ,x1 )
(X2 ,x2 )
(X,x)
(3) Let (X,x) be the A2 –singularity, with local ring OX,x = C {x,y}/(y 2 −x3 ). Then the
normalization of OX is equal to C {t}, where t = xy . Compare example 1.5.6. The
e x) is a smooth one-dimensional space with parameter t, and the
normalization (X,e
normalization map is given by t 7→ (x(t),y(t)) = (t2 ,t3 ). Note that in this example
the normalization map is “bijective”, but not an isomorphism.
(4) Let (X,x) be the A1 –singularity in three space, defined by the equation z 2 − xy.
Then X is already normal, cf. Exercise 1.5.33. The normalization of (X,x) is (X,x)
itself, and the normalization map is the identity.
(5) Let (X,x) be the “Whitney umbrella” defined by x2 − zy 2 = 0.
6 More generally, regular local rings are normal because they are unique factorization domains. But we
do not prove this.
2 2
The element s := xy is integral. Indeed, s2 = xy2 = zy
y 2 = z. We consider the ring
OX,x [s] generated by OX,x and s in the quotient ring Q(OX,x ) and claim that it is
equal to the normalization. Indeed, by applying the Weierstraß Division Theorem,
it is not so difficult to see that
OX,x [s] = C {x,y,z,s}/(x2 − zy 2 ,ys − x,s2 − z).
We can eliminate x and z and see that OX,x [s] is isomorphic to C {y,s}, which is
normal. So the normalization is smooth, with parameters y,s and the normalization
map is given by (y,s) 7→ (ys,y,s2 ).
Note that in this example the normalization map is indeed generically 1–1: for all
points (x,y,z) on the Whitney umbrella not on the z–axis there is only one point
in the preimage, but for all points on the z–axis except the origin there are two
points in the preimage of the normalization map.
In the third example we saw that the normalization of a cusp singularity is smooth.
This turns out to be the case for all curve singularities. So normality is a very strong
condition for curve singularities! It follows that for an irreducible curve singularity (X,x)
there always exists a finite generically 1–1 map from (C , 0) to (X,x). Such a map can
even be given in a particularly simple form, which is then called Puiseux expansion. This
will be studied in chapter five.
More generally the singular locus of a normal space must have at least codimen-
sion two, compare the fifth example. For hypersurface singularities, this condition is also
sufficient as suggested by the fourth example, but we are not able to prove this at the
moment, but only in the section on Cohen-Macaulay spaces, see in particular Corollary
6.5.9. So in higher dimensions the normality condition is not as strong as in the case of
curve singularities.
As noted before, if a normalization (X,ee x) of (X,x) exists, the ring O e must be

X,ex
equal to the normalization of the ring OX,x . The normalization of a ring always exists,
and it is our aim to show that this ring is a finite direct sum of analytic C –algebras.
Theorem 4.4.8 (Existence of the Normalization). Let (X,x) be a germ of an analytic
e x) −→ (X,x) of (X,x).
space. Then there exists a normalization n : (X,e
e x) is also an irreducible germ.
If moreover (X,x) is irreducible, then (X,e
Proof. Consider an irreducible decomposition (X,x) = (X1 ,x) ∪ . . . ∪ (Xr ,x) of (X,x). By
Splitting of Normalization, see Theorem 1.5.20, we have that O eX,x = OeX1 ,x ⊕. . .⊕ O
eXr ,x .
Thus we reduce to the case that (X,x) is irreducible. By the Finiteness of Normalization
Theorem 1.5.19 we deduce that O eX,x is a finitely generated OX,x –module. It was proved
in Corollary 3.3.26 that OeX,x is isomorphic to a direct sum ⊕s Si of nonzero analytic
i=1
algebras. As the total quotient ring of O eX,x is a field, we have that OeX,x is an integral
e
domain. Therefore s = 1, and it follows that OX,x is an analytic algebra. Thus it is equal
to OX,e e x).
e x for some germ of an irreducible analytic space (X,e
This normalization map is finite by 3.4.24, and it is generically 1-1 because of Exer-
cise 3.4.40.
To prove that curve singularities are not normal, it suffices to prove the following
purely algebraic theorem.
Theorem 4.4.9. Let R be a Noetherian local ring of dimension one. Then

R is regular ⇐⇒ R is normal.
Proof. Step 1. We first prove “=⇒”. As principal ideal domains are unique factorization
domains, see Example 1.4.3, and those are normal by 1.5.5, it suffices to show that R is
a principal ideal domain. The definition of regularity gives that dimC m/m2 = 1, so m is
a principal ideal. To show that an arbitrary ideal I ⊂ R is principal, first write m = (t)
for some t. Take an element x ∈ I, x 6= 0. We claim that there exists a unit u ∈ R and a
natural number k(x) such that x = utk(x) . This is obvious if x is a unit. Otherwise put
x = x1 t for some x1 ∈ R, and do the same with x1 . Either x1 is a unit, and we are done,
or x1 = x2 t, etc. This process must stop because, otherwise, x = xk tk ∈ mk for all k. But
∩mk = 0 by Krull’s Intersection Theorem, which is in contradiction to the choice of x.
This shows the claim.
Now take k to be the minimum of all k(x) where x runs over the nonzero elements
of I. It follows then I = (tk ), so, in particular, principal. Thus R is a principal ideal
domain, as was to be proved.
Step 2. “⇐=”. As a normal local ring, R is an integral domain by Theorem 1.5.7. Assume
that R is not regular. We consider the R–module m∗ = (R : m) = {x ∈ Q(R) : xm ⊂ R}.
Obviously, R ⊂ m∗ . We claim that this inclusion is strict. p
To prove the claim, take an arbitrary pa ∈ m, a 6= 0. Then (a) = m. Indeed,
by the Primary Decomposition Theorem, (a) is an intersection of prime ideals. But,
by assumption on dim(R) = 1, and because R is an integral domain, there is just one
nontrivial prime ideal, namely m. Because R is Noetherian, it follows that there exists an
s ∈ N with ms ⊂ (a). Let s be minimal with this property. If s = 1, then m = (a), hence
R would be a regular local ring. Take a nonzero element b ∈ ms−1 \ (a). Then ab ∈ / R.
b s b ∗
However, a m ⊂ R, because bm ⊂ m ⊂ (a). Thus a ∈ m \ R. This proves the claim.
From the definition of m∗ it follows that m ⊂ mm∗ ⊂ R. So either mm∗ = m
or mm∗ = R. We claim that the first equality cannot occur. Indeed, it would follow
that ab m ⊂ m. Therefore, m would be an R[ ab ]–module, finitely generated as an R–
module. By the Cayley-Hamilton Theorem 1.5.8, ab would be an integral element, which,
by assumption, would have to be in R. But ab was constructed in such a way that it does
not belong to R.
Therefore mm∗ = R. Now take any element t ∈ m \ m2 , which exists according to
Nakayama’s Lemma. Then tm∗ ⊂ R. It is impossible to have tm∗ ⊂ m, for, otherwise,
tR = tm∗ m ⊂ m2 , in contradiction to the choice of t. Therefore, tm∗ = R, and (t) = tR =
tm∗ m = m. This shows that m is principal.
Corollary 4.4.10. Let R = C {x1 , . . . ,xn }/I be a one-dimensional integral domain.
e is isomorphic to C {t}.
Then its normalization R
Proof. Because of Theorem 4.4.8 Re is a one-dimensional analytic algebra, which is nor-

e
mal. By Theorem 4.4.9, R is regular of dimension one, hence isomorphic to C {t}.
With a little bit more effort, one can give the following beautiful characterization of
a normal ring, due to Serre.
Theorem 4.4.11 (Serre’s R1 and S2 Criterion). Let R be a reduced ring. Then R is
normal, if and only if the following two conditions are satisfied:
(R1) for each prime ideal p of height one, Rp is a regular local ring;
(S2) let f ∈ R be a nonzerodivisor, then the ideal (f ) in R has no embedded primes.
Proof.
Step 1. Suppose first that R is normal. Then Rp , for a prime ideal p, is normal too, by
Exercise 1.5.32. Moreover, Rp has dimension one, so is regular by 4.4.9.
To check the second condition, suppose, on the contrary, that we have an embedded
prime p for some (f ). In particular, p is not a minimal prime ideal with p ⊃ (f ). By
Lemma 1.4.17 we can find an element b ∈ R such that p = (f ) : (b). Let m = pRp .
Considering m∗ := (Rp : m) = {x ∈ Q(R) : x · m ⊂ Rp }, one sees that fb ∈ m∗ . It follows,
as in the proof of 4.4.9, that if mm∗ = Rp , then m is principal. But then it follows that
the dimension of Rp is less than or equal to one. This is in contradiction to the fact that
p is not minimal over (f ), as the dimension of Rp is equal to the height of p.
So we may assume that mm∗ = m. We can now apply the Cayley-Hamilton Theorem
1.5.8: all elements of m∗ are integral over Rp . As we assumed R, and hence Rp , to be
normal, we deduce that m∗ = Rp . In particular, as fb ∈ m∗ . we have fb ∈ Rp . Therefore,
we can write fb = ac for some a ∈ / p. Now
p = (f ) : (b) = (f a) : (ba) = (f a) : (cf ) = (a) : (c).
In particular, a ∈ p, which is a contradiction. So our assumption that we had an embedded

prime p for (f ) is wrong. This proves one direction.
b
Step 2. Suppose on the other hand, that (R1) and (S2) hold, and suppose f is integral
over R. Consider an irredundant primary decomposition of (f ):
(f ) = q1 ∩ . . . ∩ qt .
√
As (f ) does not have embedded primes by (S2), all the pi := qi are minimal associated
primes. By Krull’s Principal Ideal Theorem, see 4.2.16, all the pi are prime ideals of
height one. By (R1) it follows that the Rpi are regular. Moreover they have dimension
one, so the Rpi are normal by Theorem 4.4.9. It follows that fb ∈ Rpi for all i. To put
it in another way: b ∈ f Rpi for all i. By the Second Uniqueness Theorem of Primary
Decomposition 1.4.23 we have qi = R ∩ f Rpi for all i. Therefore, b ∈ qi for all i. Hence,
b ∈ ∩qi = (f ), and thus fb ∈ R, which had to be proved.
We will now consider a function-theoretic interpretation of normal spaces. For this,
we will show that a space is normal if and only if “Riemann’s Extension Theorem” holds
for X. First we need a definition.
Definition 4.4.12. Let X be an analytic space, and Sing(X) be the singular locus of
X. A function f : X \ Sing(X) −→ C is called weakly holomorphic on X if
(1) f is holomorphic on X \ Sing(X);
(2) f is locally bounded.
Let (X,x) be a germ of an analytic space. A weakly holomorphic function germ f on
(X,x) is an equivalence class of weakly holomorphic functions f : X \ Sing(X) −→ C .
Here X is a representative of (X,x). Two such functions are called equivalent if they
agree on the intersection of the corresponding representatives.
Remarks 4.4.13.
(1) The weakly holomorphic functions on X form a ring. In fact, as Sing(X) is con-
tained in a proper analytic subset of X, it follows by continuity that the ring of
holomorphic functions on X embeds in the ring of weakly holomorphic functions on
X (an analytic function on X is already determined by its values on X \ Sing(X)).
′
(2) The germs of weakly holomorphic functions form a ring which we denote by OX,x .
′
As above one sees that OX,x ⊂ OX,x .
Examples 4.4.14. (1) Consider the coordinate axes, defined by xy = 0. Look at the
function f which is identically 1 on the the x–axis, except at the origin, and identically
0 on the y–axis except at the origin. Then f cannot be extended to the origin (not even
x
as a continuous function), but is bounded. Note that we can write f (x,y) = x+y .
(2) Consider the analytic subset X = {y − x = 0} ⊂ C . The function f = xy is
2 3 2
holomorphic, whenever
√ x 6= 0. So f is a holomorphic function on X \ {(0,0)}. We have
the estimate yx ≤ | x|, hence f defines a germ of a weakly holomorphic function on
(X,x). In this case, f extends to a continuous function on X, by f (0,0) = 0. However f
is not holomorphic.
Theorem 4.4.15.
(1) Let (X,x) be a germ of an analytic space. Then there is a canonical isomorphism
′
OX,x ∼
=O eX,x . To put it in another way, f is a weakly holomorphic function germ
if and only if f is in the integral closure of OX,x .
(2) A germ of an analytic space (X,x) is normal if and only if every weakly holomorphic
function germ on (X,x) can be extended to a holomorphic function on (X,x). To
put it in another way, a space (X,x) is normal, if and only if the First Riemann
Extension Theorem holds for (X,x).
Proof. The second statement follows immediately from the first one. The proof of the
first statement is quite long, and will be divided into several steps. Suppose the germ
(X,x) is in (C n , 0).
eX,x ⊂ O′ . Take f ∈ O
Step 1. First we prove O eX,x . Then f satisfies an equation
X,x
(4.5) f s + a1 f s−1 + . . . + as = 0, ai ∈ OX,x .
Consider a representative X of (X,x). Since in particular f ∈ Q(OX,x ) we can write

f = hg , for some g ∈ OX,x and h ∈ OX,x a nonzerodivisor. As h is a nonzerodivisor it
does not vanish identically on any of the components of (X,x).
By definition, X \ Sing(X) is a locally complex submanifold of an open subset in
C n , so that Riemann’s Extension Theorem holds for X \ Sing(X) (cf. Theorem 3.1.15).
Therefore, it suffices to prove that f is bounded on X \ Sing(X). This will be proved as
an application of the Continuity of Roots. We take X so small that the ai of Equation
(4.5) are holomorphic on X, and that X is an analytic subset of U . So we may lift the
ai to holomorphic functions on U . Now look at the equation
(4.6) F := T s + a1 T s−1 + . . . + as = 0.
We plug in 0 in the a1 , . . . ,as and factorize

F (0,T ) = (T − c1 )s1 · · · (T − ce )se .
By Hensel’s Lemma 3.3.21 F factorizes as F = F1 · · · Fe , with Fi (0,T ) = (T − ci )si , and

Fi a polynomial in T of degree si . In particular Fi (x1 , . . . ,xn ,T + ci ) is a Weierstraß
polynomial in T . Take a small open neighborhood Vi of ci in C , for i = 1, . . . ,e. By the
Continuity of Roots, there exists an open neighborhood Ui of 0, such for all p ∈ Ui , all
we take V = V1 ∪ . . . ∪ Ve ,
si zeros (counted with multiplicity) of Fi (p,T ) lie in Vi . So if P
and U = U1 ∩ . . . ∩ Ue it follows that for all p ∈ U the s = si zeros of F (p,T ) lie in
V . Now formula (4.5) says that f (p) satisfies the equation F (p,T ). Therefore, if p ∈ U ,
and f (p) is defined then f (p) ∈ V , in particular it is bounded. So we can extend f to a
holomorphic function on X \ Sing(X). By continuity, this extension is also bounded.
Step 2. We now turn our attention to the inclusion OX,x ′
⊂ O eX,x . We claim that it
suffices to show this for irreducible (X,x). To see this let (X,x) = (X1 ,x) ∪ . . . ∪ (Xr ,x)
be the irreducible decomposition of (X,x). Let f be a weakly holomorphic function on a
representative X = X1 ∪ . . . ∪ Xr of (X,x). As f is a genuine function outside the singular
of X, it follows that f is a holomorphic function on each Xi \ Sing(Xi ) ∪ ∪j6=i (Xi ∩
locus
Xj ) . By the Riemann Extension Theorem in C n , the function f can be extended to
a weakly holomorphic function on Xi . So, once we have shown the inclusion for an
irreducible (X,x), f defines an element of O eXi ,x for each each i. But by the Splitting of
Normalization, see 1.5.20, we get O eX,x = ⊕r O e e
i=1 Xi,x . This implies that f ∈ OX,x .
Step 3. We now reduce to the case that (X,x) is an irreducible hypersurface. By the Local
Parametrization Theorem we have a surjective map π from (X,x) onto a hypersurface
(X ′ ,x) which has the following properties:
• OX ′ ,x ⊂ OX,x .
• Q(OX,x ) = Q(OX ′ ,x ).
• There exists a representative π : X −→ X ′ and a proper analytic subset D of X ′
such that π : X \ π −1 (D) −→ X ′ \ D is biholomorphic.
• Sing(X ′ ) ⊂ D.
Let f be a weakly holomorphic function on X. Then f gives a bounded holomorphic
function on X ′ \ D. This is a smooth space, so by Riemann’s Extension Theorem 3.1.15
f can extend to a bounded holomorphic function on X ′ \ Sing(X ′ ). We obtain a weakly
holomorphic function on X ′ . So if the theorem is proved for a hypersurface we get f ∈
Q(OX ′ ,x′ ) and an integral equation for f with coefficients in OX ′ ,x . As OX ′ ,x ⊂ OX,x
this shows that f is in OeX,x , since f ∈ Q(OX ′ ,x ) = Q(OX,x ).
Step 4. Now let (X,x) be a germ of an analytic hypersurface, defined by a Weierstraß

polynomial
P = xsn + a1 xns−1 + . . . + as = 0.
′
So ai is in the maximal ideal of On−1 . We will first show that OX,x ⊂ Q(OX,x ). Let ∆

be the discriminant of P , (D,x) = V (∆),x and f be a weakly holomorphic function
germ on (X,x). We will show that on some representatives X and D we can write f as
g
∆, for a holomorphic function g on X.7 So we have to show that ∆ · f has a holomorphic
extension to X. Consider the projection
π : X −→ U ⊂ C n−1 .
We choose U so small, so that there exists a constant C > 0 such that the absolute value
of the xn –th coordinate of each point is smaller than C. This can be done by Continuity
of Roots, see 3.4.11. Furthermore, we may assume that |f (a)| < C for all a ∈ X \Sing(X),
because by assumption f is locally bounded. Take a point p ∈ U \ D. Locally around p,
there exists an open neighborhood V of p, and s holomorphic functions
α1 , . . . ,αs : V −→ C ,
such that π −1 (V ) is the disjoint union of s complex submanifolds, given by the equations
xn = αi (x1 , . . . ,xn−1 ), see 3.3.23. U was chosen to be so small that the αi are bounded.
Let q ∈ V , and consider the Lagrange interpolation polynomial in xn
s
X Y xn − αj (q)
f (q,αi (q)) .
i=1
αi (q) − αj (q)
j6=i
For q ∈ V , this is indeed a well-defined polynomial in xn . By construction,

the Lagrange
interpolation polynomial has the same values as f on the s points q,α1 (q) , . . . , q,αs (q) .
We now vary q, and we thus may assume that on π −1 (V ) f is equal to the above polyno-
mial in xn whose coefficients are holomorphic functions on V . Now on V the discriminant
is equal to Y
∆= (αi − αj ).
j6=i
Thus we see that ∆·f has as coefficients holomorphic functions on V . As one easily checks,
these functions are bounded by a number A(C), which only depends on C. Doing this for
all p ∈ U \ D, we see that ∆ · f is a polynomial in xn whose coefficients are holomorphic
functions on U \ D which are bounded. By the Riemann Extension Theorem for C n−1 ,
see 3.1.15, it follows that we can extend these to holomorphic functions on U . Hence ∆ · f
can be extended to a holomorphic function on all of X.
Step 5. We keep the notations of Step 4, and we finally show that any weakly holomorphic
function is integral over OX,x . For p ∈ U \ D and V an open neighborhood of p, look at
the holomorphic functions α1 , . . . ,αs : V −→ C , as above. The function q 7→ f (q,αi (q))
defines a holomorphic function on V . Moreover on π −1 (V ) the holomorphic function

f − f (x1 , . . . ,xn−1 ,α1 ) · · · f − f (x1 , . . . ,xn−1 ,αs )
is the zero function. Let bi be the i–th symmetric polynomial in the f (x1 , . . . ,xn−1 ,αj ).
Then the bi define holomorphic functions on V , and we have an equation
(4.7) f s + b1 f s−1 + . . . + bs = 0,
7 In view of the fact that we try to prove that f is in the integral closure of OX,x , and the fact that any
element in the integral closure has as universal denominator ∆, see 1.5.19, this is the obvious thing
to do.
as a holomorphic function on π −1 (V ). Moreover, the bi are bounded on V , because the

f (x1 , . . . ,xn−1 ,αi ) are. Doing this for all p, we get bounded functions b1 , . . . ,bs on U \ D.
By the Riemann Extension Theorem 3.1.15, they extend to a holomorphic function on
g
all of U . Now write f = ∆ . This we can do by Step 4. As ∆ does not vanish on U \ D,
we can multiply equation (4.7) with ∆s and get
(4.8) g s + b1 ∆g s−1 + . . . + bs ∆s = 0,
as holomorphic functions on X \ π −1 (D). By continuity, it follows that (4.8) holds on

X. We can now read (4.8) as an equation in OX,x . Upon dividing (4.8) by ∆s we get an
g
integral equation for f = ∆ .
Exercises
4.4.16. Let (X,x) = (X1 ,x1 ) ∪ · · · ∪ (Xr ,xr ) be a multi-germ of analytic spaces.
(1) Show that OX,x has exactly r maximal ideals, so in particular OX,x is a semi-local ring.
Describe those maximal ideals.
(2) Show that (X,x) is normal if and only if (Xi ,xi ) is normal for i = 1, . . . ,r.
(3) Let ϕ : (X,x) −→ (Y,y) be a map between multi-germs of analytic spaces. Show that
ϕ is generically 1 − 1 if and only if ϕ induces an isomorphism of total quotient rings
ϕ∗ : Q(OY,y ) −→ Q(OX,x ).
(4) Let ϕ : (X,x) −→ (Y,y) be a finite map between irreducible germs of analytic spaces.
Suppose that ϕ∗ : OY,y −→ OX,x is injective and let s be the degree of the field extension
Q(OX,x) ⊃ Q(OY,y ). Show that ϕ is generically s to 1 and that ϕ is surjective.
(Hint: Use 3.4.41 and 3.4.40)
4.4.17. Let ϕ : (X,x) −→ (Y,y) be a finite map of germs of analytic spaces. Prove that the
image of ϕ is a germ of an analytic space.
(Hint: Let I = Ker(ϕ∗ ). Show that the image of ϕ consists of all points a ∈ Y with f (a) = 0 for
all f ∈ I.)
4.4.18. The normalization of a local domain is not necessarily local. The local ring R, obtained
from C [x,y]/(y 2 − x3 − x2 ) by localizing in the maximal ideal (x,y) is an integral domain. Show
that the normalization of R is not local.
e x) −→ (X,x) be a normalization.
4.4.19. Let (X,x) be a germ of an analytic space, and n : (X,e
Let (Y,y) be a normal multi-germ, and ϕ : (Y,y) −→ (X,x) be a map of multi-germs. Show that
e x) making the following
e : (Y,y) −→ (X,e
there exists a uniquely determined map of multi-germs ϕ
diagram commutative.
e
ϕ
(Y,y) e x)
/ (X,e
FF
FF
FF n
ϕ FF
F#
(X,x)
(Hint: Prove a corresponding statement for rings.)
4.4.20. Let (X,x) be a germ of an analytic space, and Sing(X,x) be the singular locus of (X,x).
Let I ⊂ OX,x be the ideal of germs of analytic functions vanishing on the singular locus. Show
that (X,x) is normal if and only if HomOX,x (I,I) = OX,x .
(Hint: Use the fact that regular analytic rings
√ are normal. Follow the proof of 1.5.13, but now
use the Nullstellensatz 3.4.4 to prove I ⊂ J (notations of 1.5.13).)
“ ”
∂f ∂f
4.4.21. Let f ∈ C {x1 , . . . ,xn }. Prove that f k ∈ ∂x 1
, . . . , ∂x n
for some k ≥ 0.
` ∂f ∂f
´ ∂f ∂f
(Hint: Let (X, 0) = V (f, ∂x 1
, . . . , ∂xn
), 0 and (Y, 0) = (V (
∂x1
, . . . , ∂x n
), 0). Show that (X, 0) =
(Y, 0) in the following way. Suppose (X, 0) $ (Y, 0). Find a map n : (C , 0) −→ (Y, 0) with
n((C , 0)) 6⊂ (X, 0) by using the normalization of a curve (C, 0) ⊂ (Y, 0) which is not contained
d
in (X, 0). Compute dt (f ◦ n).)
e 0) −→ (C, 0) be given such

4.4.22. Let (C, 0) be an irreducible curve singularity. Let n : (C,e
e
that (C,e0) is smooth. Suppose that dimC (OC, ee /OC,0 ) < ∞. Show that n is the normalization
0
mapping.
Generalize this result to reducible germs.
(Hint: Show that the degree of the corresponding field extension is one.)
169
5 Plane Curve Singularities
This chapter is devoted to the study of plane curve singularities. First of all, from the
existence of the normalization, and the fact that one-dimensional normal singularities
are smooth, which are results proved in the previous chapter, it follows that one can
parameterize curve singularities. Thus, if R is the local ring of an irreducible plane
curve singularity, we have an injection R ⊂ C {t}. From the Finiteness of Normalization
Theorem 1.5.19 it follows without much difficulty that the codimension of R in C {t}
(as C –vector space) is finite. This codimension is called the δ–invariant. Here we prove
the converse of this statement, that is, given a subring R of C {t} of finite codimension,
generated by two elements, say x(t) and y(t), then R the local ring of an irreducible plane
curve singularity, say given by f (x,y) = 0. In order to obtain information on the order of
f , we introduce the intersection multiplicity of two (irreducible) plane curve singularities
(C, 0) and (D, 0). This number can be calculated as follows. Consider the normalization
C {t} of OC,0 , and let (D, 0) be given by g = 0. Via the inclusion OC,0 ⊂ C {t}, we
can consider at g as an element of C {t}. Then the intersection multiplicity is given
by the vanishing order of g. From this it easily follows that if R = C {x(t),y(t)}, and
f ∈ C {x,y} is irreducible with f (x(t),y(t)) = 0, then the order of f in y is equal to
ordt (x(t)), and similar for x. As an application we prove that if we have a convergent
power series f ∈ C {x,y}, which is irreducible when considered as a formal power series,
then it is already irreducible as a convergent power series. This in fact holds for power
series in any number of variables, and is an easy consequence of the Artin Approximation
Theorem, which will be proved in Chapter 8. Next we discuss Newton’s method. Here
we start with an f ∈ C {x,y}, and try to find x(t), y(t) with f (x(t),y(t)) = 0. We give
two proofs of the convergence. The first one is a consequence of Corollary 3.3.18, and is,
to our knowledge, new. The second one is standard. Finally, we discuss the irreducibility
of f in terms of its Newton polygon. It is shown that for an irreducible f the Newton
polygon has only one face. Our proof is algorithmic, that is, if the Newton polygon has
more than one face, we will give an algorithm to find a factorization of f .
In Section 5.2 we will study several invariants of an irreducible plane curve singularity
R. As above, we may assume that R ⊂ C {t}, which is the normalization. We already
mentioned the δ–invariant. Moreover, we introduce the semigroup Γ(R) ⊂ N, which
consists of the vanishing orders of the elements in R. Furthermore, the conductor c(R)
is introduced. It is the smallest natural number α with α + N ⊂ Γ(R). Our first aim is
to show the Theorem of Gorenstein, which says that c(R) = 2 · δ(R). The proof uses
duality theory for fractional ideals. Fractional ideals are R–modules in the quotient field
Q(R), which have a universal denominator. Given a fractional ideal I, we can look at
its dual I ∗ := HomR (I,R). For example, the normalization R e and the conductor ideal
e
AnnR (R/R) are fractional ideals, which are dual to each other. We have a canonical map
I −→ I ∗∗ , which one easily sees to be injective. One may ask oneself whether this map
is an isomorphism for all fractional ideals. In general, this turns out not to be the case.
Those curve singularities for which it is true are called Gorenstein, and we will prove
that irreducible plane curve singularities are indeed Gorenstein. We then discuss how
to compute the minimal generators of the semigroup Γ(R) from the parametrization. In
170 5 Plane Curve Singularities
general, this would be much too P difficult, so we first suppose that the parametrization
is of type x(t) = tn , y(t) = i
i≥m ai t with m ≥ n. After a coordinate change in t,
and interchanging x and y we may suppose that our parametrization is of this type.
Characteristic exponents k0 , . . . ,kg are introduced, in terms of n, and the i ∈ N with
ai 6= 0. A priori these are only invariants of the parametrization, but we will show
that they determine, and are determined by, the minimal generators β0 , . . . ,βg of the
semigroup Γ(R). We then discuss the Inversion Theorem, which is of importance in the
next section. This gives a method for computing the characteristic exponents in case
n > m.
In Section 5.3 we consider resolutions of irreducible plane curve singularities. First
of all, the blowing-up π : Bl0 C 2 −→ C 2 of C 2 in a point, say 0, is introduced. The space
Bl0 C 2 is a complex manifold of dimension one, in which the point 0 has been replaced by
all directions through 0. The set π −1 (0) is a projective line, called the exceptional curve
or exceptional divisor. If one now has a plane curve singularity, which has representative
C in some small open neighborhood U of 0, we can look at the strict transform C (1) :=
π −1 (C \ {0}), and iterate this process. That is, we blow up in the singular point of C (1) , if
present. The Theorem of Resolutions of Plane Curve Singularities says that this process
stops after finitely many steps. The main ingredient of the proof is the existence of a
parametrization of the irreducible plane curve singularity, as studied in the first section.
After some more blowing-ups, one can arrange that the strict transform intersects just
one exceptional curve, and intersects this curve transversally. This is then called the
standard resolution. Having such a resolution process, one can introduce two invariants.
First of all, one has the multiplicity sequence. If the standard resolution is obtained
after k blowing-ups, one has a sequence of numbers (m0 , . . . ,mk−1 ), where m0 is the
multiplicity of C, and mi is the multiplicity of the strict transform C (i) after i blowing-
ups. Furthermore, the resolution graph is introduced. It is a weighted graph with one
vertex for each exceptional divisor. The weight of a vertex is i if the curve is created
under the i–th blowing-up. Furthermore, for the strict transform we add a star ∗. Two
vertices are connected by an edge, if the corresponding divisors intersect. Furthermore,
the star is connected to the vertex corresponding to the unique exceptional divisor which
it intersects. The main result of the section is that the multiplicity sequence, the resolution
graph, the characteristic exponents, and the semigroup determine each other. In the
proof, essential use of the Inversion Theorem is made.
In Section 5.4 we study reducible plane curve singularities. Again we can study
a standard resolution. Its existence easily follows from the irreducible case. As a new
invariant the contact number between two irreducible components, also called branches
of the plane curve singularity, is introduced. The contact number of two irreducible
plane curve singularities is the number of blowing-ups one needs to separate the strict
transforms. It turns out that this number is finite. As in the irreducible case, we introduce
a resolution graph, but of course we need a star ∗ for each branch. We then show that
the resolution graph determines the resolution graphs of the irreducible components and
the contact numbers between the branches, and conversely. Finally, as an application of
the standard resolution we give two results. First of all, given irreducible plane curve
singularities, we show how to compute the intersection number from the multiplicity
sequences and the contact numbers. Secondly, for irreducible plane curve singularities,
we will prove a theorem of Max Noether, which gives a formula for the δ–invariant of an
irreducible plane curve singularity in terms of its multiplicity sequence. This result will
be generalized to the reducible case in the exercises.
5.1 Puiseux Expansion 171
5.1 Puiseux Expansion
From the results of the previous chapter it is easy to see that one can parameterize
irreducible (plane) curve singularities:
Theorem 5.1.1 (Puiseux Expansion). Let 0 6= f ∈ C {x,y} be irreducible. Then there
exist x(t), y(t) ∈ C {t} such that

(1) f x(t),y(t) = 0,

(2) dimC C {t}/C {x(t),y(t)} < ∞.
Proof. Let R be the normalization of C {x,y}/(f ). Because of Corollary 4.4.10 R is iso-
morphic to C {t}. Therefore, we have an injective map
C {x,y}/(f ) −→ C {t},
and C {t} is a finitely generated C {x,y}/(f )–module via this map. Let x(t) respectively
y(t) be the images of x mod (f ) respectively y mod (f ) in C {t}. Then obviously (1)
holds.
To prove (2) we may assume that f is a Weierstraß polynomial in y. Let ∆ be
its discriminant. As f is irreducible, it does not have multiple factors. In particular
∆ ∈ / (f ). It follows from the Theorem of Finiteness of Normalization, see 1.5.19, that
∆C {t} ⊂ C {x(t),y(t)}. We can view ∆ as an element of C {t}. Let k be the vanishing
order of ∆, that is, ∆ = tk u for some unit u. Then it follows that tk C {t} ⊂ C {x(t),y(t)},
so that {1, . . . ,tk−1 } generate C {t}/C {x(t),y(t)} as C –vector space.
Example 5.1.2. For the A2 –singularity y 2 −x3 = 0 we can take x(t) = t2 , and y(t) = t3 .
We have the following converse to 5.1.1.
Theorem 5.1.3. Let two power series x(t), y(t) ∈ tC {t} be given. Define R =
C {x(t),y(t)} and suppose that dimC C {t}/R < ∞. Then there exists an irreducible
f ∈ C {x,y} with R ∼
= C {x,y}/(f ). Moreover, R ⊂ C {t} is the normalization of R.
Proof. Consider the map ϕ : C {x,y} −→ C{t} defined by ϕ(x) = x(t), ϕ(y) = y(t). Then
C {x,y}/ Ker(ϕ) ∼ = R. From dimC C {t}/R < ∞ it follows that R ⊂ C {t} is a finite ring
extension, and as C {t} has dimension one, it follows that R has dimension one, see 4.1.4.
Because R ⊂ C {t}, it is an integral domain. Thus R is the local ring of an irreducible
plane curve singularity. Therefore Ker(ϕ) is a principal ideal by the Characterization of
Hypersurfaces 4.1.12. So Ker(ϕ) = (f ) for some f ∈ C {x,y}. As R is an integral domain,
we get that Ker(ϕ) is a prime ideal, and it follows that f is irreducible.
It remains to prove that C {t} is the normalization of R. As dimC C {t}/R < ∞,
it follows that I := AnnR (C {t}/R) 6= 0 (Exercise 5.1.30). Now I is an ideal in R =
C {x(t),y(t)}, but from the definition it follows immediately that I is also an ideal in
C {t}. As C {t} is a principal ideal domain it follows that I = tc ·C {t} for some c > 0. This
implies that tc C {t} ⊂ C {x(t),y(t)}. Let ψ ∈ tC {t}, then for all k ≥ c we have ψ k ∈ R.
k+1
In particular ψ = ψψk is in the quotient field of C {x(t),y(t)}. As certainly the constants
are in the quotient field of C {x(t),y(t)} too, it follows that the quotient fields of C {t}
and C {x(t),y(t)} are equal. As C {t} is a finitely generated C {x(t),y(t)}–module, every
element in C {t} is integral over C {x(t),y(t)}. Therefore C {t} is in the normalization
of C {x(t),y(t)}. As C {t} is normal itself, it follows that C {t} is the normalization of
C {x(t),y(t)}.
In this theorem we showed that we can find such an f , but we still have no in-
formation on its order. In order to answer this question, we introduce the intersection
multiplicity.
Definition 5.1.4. Let (X, 0) and (Y, 0) be plane curve singularities, not necessarily
reduced, not having a component in common. Let (X, 0) be defined by f = 0, and let
(Y, 0) be defined by g = 0. The intersection multiplicity or intersection number of (X, 0)
and (Y, 0) at 0 is defined by

(X, 0) · (Y, 0) := dimC C {x,y}/(f,g) .
In case (X, 0) and (Y, 0) are reduced, this intersection multiplicity has a geometric
interpretation as the number of intersection points in a small neighborhood of 0 after
a small perturbation of f and g. We are not able to prove this at the moment, but it
will be done in Chapter 6. The following important lemma tells us how to compute the
intersection number with the help of the normalization in case one of the curves, say
(X, 0), is irreducible.
Lemma 5.1.5. Suppose that (X, 0) is an irreducible plane curve singularity. Let OX,0 ⊂
∼
e 0 = C {t} be its normalization. Let (Y, 0) be a plane curve singularity, defined by
OX,e
g = 0. We can consider the class of g in OX,0 and, therefore, we can view g as an
element of C {t}. Then
(X, 0) · (Y, 0) = ordt (g).
Proof. Consider the diagram
·g
OX,e
e 0 / OX,e
e 0
O O
? ·g ?
OX,0 / OX,0 .
·g ·g
e 0 −→ OX,e
Then ordt (g) = dimC Coker(OX,e e 0 ) , and (X, 0)·(Y, 0) = dimC Coker(OX,0 −→

OX,0 ) . The problem is now reduced to a problem in linear algebra, and its solution is
left as Exercise 5.1.31.
Corollary 5.1.6. Under the conditions of Theorem 5.1.3, suppose that ord(x(t)) = n,
and that ord(y(t)) = m. Let f ∈ C {x,y} be irreducible with C {x(t),y(t)} ∼
= C {x,y}/(f ).
Then
(1) f is regular in y of order n, and regular in x of order m.
(2) The multiplicity ord(f ) of f is equal to min{n,m}.
Proof. Let (X, 0) is the singularity defined by f = 0, and (L, 0) the germ of the line
defined by x = 0. The first statement follows because, by Lemma 5.1.5,

ordy (f ) = dimC C {x,y}/(f,x) = (X, 0) · (L, 0) = ordt (x(t)).
Similarly for x.
To prove the second statement we may without loss of generality assume that n ≤ m,
and that f is regular in y of order ord(f ). This is always possible after a linear coordinate
change. It follows from the first part that
ord(f ) = ordy (f ) = (X, 0) · (L, 0) = ordt (x(t)) = n = min{n,m},
as claimed.
Theorem 5.1.7. Let ε be a primitive n–th root of unity. Consider the ring extension
C {x,y} ⊂ C {t,y}, where x = tn .
(1) Let f = y n + an−1 (x)y n−1 + · · · + a0 (x) be an irreducible Weierstraß polynomial.
Then there exists a y(t) ∈ tC {t} such that
n
Y
f= y − y(εi t) .
i=1
Qn
One also writes this as f = i=1 y − y(εi x1/n ) .

(2) Let y(t) ∈ tC {t}
such that dimC C {t}/C {tn ,y(t)} < ∞ be given. Then f =
Qn i
i=1 y − y(ε t) is an irreducible Weierstraß polynomial in C {x,y}.
Similar statements hold in the formal case.

Proof. (1) We first prove the first part. By Theorem 5.1.1 there exist x(t) and y(t) in
C {t} such that f x(t),y(t) = 0. By Corollary 5.1.5 it follows that ord(x(t)) = n, and we
may after a coordinate change in t, assume that x(t) = tn . Consider f (tn ,y) ∈ C {t,y}. As
f (tn ,y(t)) = 0, it follows from the Weierstraß Division Theorem that y − y(t) is a factor
of f (tn ,y). For all i with 1 ≤ i ≤ n it follows that f ((εi t)n ,y(εi t)) = f (tn ,y(εi t)) = 0. So
y − y(εi t) is also a factor of f (tn ,y). We claim that all these factors are different. That
is, we claim
y(εi t) = y(εj t) ⇐⇒ n | (i − j).
P
The direction “⇐=” is trivial. For the converse, write y(t) = i≥m ai ti . It is an elemen-
tary fact that for h ∈ C {tn ,y(t)} the order ordt (h) is in the semigroup generated by n
and all i with ai 6= 0. As C {t}/C {tn ,y(t)} is a finite-dimensional vector space we can find
k1 . . . ,ks ∈ N with gcd(n,k1 , . . . ,ks ) = 1, and akℓ 6= 0 for ℓ = 1, . . . ,s. Now y(εi t) = y(εj t)
implies that εkℓ i = εkℓ j for ℓ = 1, . . . ,s. Thus n | kℓ (i − j) for ℓ = 1, . . . ,s. It follows that
n | (i − j). (This is elementary number theory.) So the first part follows, because C {t,y}
is a unique factorization domain.
(2) Consider the kernel of the map ϕ : C {x,y} −→ C {t} which sends x to x(t) and y
to y(t). Now in Theorem 5.1.3 it was shown that Ker(ϕ) is principal, and in fact can be
generated by an irreducible Weierstraß polynomial in y of degree n by Corollary 5.1.6.
This polynomial we call f . From the first part it follows that f has in C {t,y} the desired
factorization.
The following corollary is most easily proved as an application of Artin’s Approxi-
mation Theorem. The proof we give now is more elementary, however.
Corollary 5.1.8. Consider an f ∈ C {x,y}. Then we have the following statements.
(1) Suppose f is irreducible in C {x,y}. Then f is also irreducible when considered as
element of C [[x,y]].
(2) Suppose f = f1 ·f2 with f1 , f2 ∈ C [[x,y]], and suppose f1 is a Weierstraß polynomial
in y. Then f1 and f2 are convergent.
(3) Suppose y = y(x) is a formal solution of f (x,y) = 0. Then y(x) is convergent.1
1 This we already proved as 3.3.18.
Proof. (1) We may assume that f is an irreducible

Q Weierstraß polynomial. As f is irre-
ducible we can, by Corollary
5.1.7, write f = ni=1 (y − y(εi t)), where x = tn . Moreover
dimC C {t}/C {tn ,y(t)} is finite. So for all k ≫ 0 the element tk can be written as a
n n
power series in t and y(t). Thus it follows that dimC C [[t]]/C [[t ,y(t)]] is finite. Ap-
plying the second part of Theorem 5.1.7 in the formal case, shows that f is irreducible
in C [[x,y]].
(2) Take a factorization f = u · g1 · · · gn of f , where the gi are Weierstraß polynomials
and u is a unit. Then u and g1 . . . ,gn are uniquely determined. By the first part, the gi
are also irreducible in C [[x,y]]. Taking similar factorizations of f1 and f2 . It follows that
f1 is a product of the g ′ s, hence convergent. But then f2 = ff1 is also convergent.
(3) If y = y(x) is a solution of f (x,y) = 0, we have that y − y(x) is a factor of f . Thus
(3) is a special case of the second part.
Now we wish to explain an explicit construction of the parametrization obtained in
Theorem 5.1.1. As a tool we need to introduce the Newton polygon.
P
Definition 5.1.9. Let f = aij xi y j ∈ C {x,y}. We call
supp(f ) = {(i,j) : aij 6= 0}
the support of f . Sometimes we will identify the pair (i,j) with the monomial xi y j . Let
S
(1) Γ+ (f ) = (i,j) + R2+ .
(i,j)∈ supp(f )
(2) Γ(f ) the boundary of the convex hull of Γ+ (f ) in R2 .

Then
(1) Γ(f ) is called the Newton polygon of f .

(2) The compact segments of Γ(f ) we call faces. The slope of the line through a face
is called the slope of the face.
(3) Let
P∆ be ia jface of Γ(f ). Then the restriction of f to ∆ by definition is f∆ =
aij x y .
(i,j)∈∆
(4) The face ∆ induces a grading on C {x,y} in the following way: if ∆ has slope
w1
−µ = − w 2
, in lowest terms, and w1 , w2 > 0. Then we set w-deg(x) = w1 , and
w-deg(y) = w2 . Note that f∆ is quasi-homogeneous with respect to these weights.
Example 5.1.10. Let f = x5 + x3 y 3 + y 3 + xy. Then Supp(f ) = {(5,0),(3,3),(0,3),(1,1)}.
Two faces
Γ+ (f ) Γ(f )
Let ∆ be the face defined by (1,1) and (5,0). Then f∆ = x5 + xy. The weights corre-
sponding to the grading defined by ∆ are w-deg(x) = 1 and w-deg(y) = 4.
Obviously, different f can have the same Newton polygon. For example, x5 + x3 y 3 +
y 3 + xy and x5 + y 3 + xy have the same Newton polygon. The element (3,3) does not
enlarge the Newton polygon.
Example 5.1.11. f = x3 y 3 + x2 y 4 .
One face
Γ(f )
Definition 5.1.12. f is called convenient if there exist positive integers n,m such that
(n,0),(0,m) ∈ supp(f ), that is, the noncompact parts of Γ(f ) are on the coordinate axis.
Remark 5.1.13. f in Example 5.1.10 is convenient, f in Example 5.1.11 is not conve-
nient. For any f ∈ C {x,y}, we can always write f = xa · y b · f1 , where f1 is convenient.
Therefore, from the point of view of the zeros of f the convenient case is the important
one.
We now come to Newton’s method to find a parametrization of a branch of a plane
curve singularity.
Theorem 5.1.14. Let f ∈ C {x,y}, and suppose that f is convenient. We have the fol-
lowing algorithm for finding a parametrization of an irreducible component (also called
branch) of f = 0.
w1
(1) Let ∆ be the face of Γ(f ) with smallest slope − w2
, which is in lowest terms, and with
w1 , w2 > 0. Then f∆ is quasi-homogeneous of say, degree d0 . Consider a zero of f∆ (1,y)
which we call a0 . Take new variables x1 and y1 and substitute
x = xw
1 ,
1
y = xw
1 (a0 + y1 ).
2
Then
1
f (1) (x1 ,y1 ) := f (xw w2
1 ,x1 (a0 + y1 )) ∈ C {x1 ,y1 }.
1
xd10
(2) Inductively find a parametrization x1 = ta , y1 = ϕ(t) of a branch of f (1) (x1 ,y1 ) in
case f (1) 6= 0. Then
x = taw1 , y = a0 taw2 + taw2 ϕ(t),
gives a parametrization of a branch of f = 0.
Before giving the proof we give an example.
Example 5.1.15. Let f (x,y) = y 4 − 2x3 y 2 + x6 − x7 . The Newton polygon consists of

one face.
Γ(f )
We have w1 = 2, w2 = 3. Then d0 = 12 and f∆ (x,y) = (y 2 − x3 )2 . Thus f∆ (1,y) =

(y 2 − 1)2 . We have the solution a0 = 1. We set
x = x21 , y = x31 (1 + y1 ).
We plug this into f (x,y) to get x12

1 f
(1)
(x1 ,y1 ). We obtain as result
f (1) (x1 ,y1 ) = y14 + 4y13 + 4y12 − x21 .
Of course, we can go on as in the first step, but we can also use some common sense.
Indeed,
f (1) (x1 ,y1 ) = (2y1 + y12 )2 − x21 ,
so that x1 = ±(2y1 + y12 ). We, therefore, see two solutions for y1 :
√ √
y1 = −1 + 1 − x1 ; y1 = −1 + 1 + x1 .
Hence, we obtain the solutions (setting x1 = t)

√
(1) x = t2 , y = t3 1 − t,
√
(2) x = t2 , y = t3 1 + t.
Hence, √ we can deduce easily, that f is reducible, as we can find g1 ∈ C {x,y} with
g1 (t2 ,t3 1 −√t) = 0, and g1 is a Weierstraß polynomial in y. Similarly we can find a g2
with g2 (t2 ,t3 1 + t) = 0. Note that the plane curve singularities V (g1 ), 0 and V (g2 ), 0
√ √
are such that their intersection is zero. Indeed (t2 ,t3 1 − t) = (t2 ,t3 1 − t) for t small
if and only if t = 0, as a simple calculation shows. Thus (g1 ) 6= (g2 ), and we get the
factorization f = g1 · g2 , since degy (gi ) = 2 for i = 1,2.
Proof of Theorem 5.1.14. As f is convenient, it is in particular regular in y, of order say
b.
Step 1. First of all, we show that the definition of f (1) makes sense. We can write
X
f= fd ,
d≥d0
where fd is quasi-homogeneous of degree d with weights w1 and w2 . From the definition

of quasi-homogeneity it follows that
X
f xw w2
1 ,x1 (a0 + y1 ) =
1
xd1 fd (1,a0 + y1 ),
d≥d0
(1)
so that indeed the definition of f makes sense. Obviously, if we can find a parametriza-
tion of a branch of f (1) = 0, we get a parametrization of a branch of f = 0.
Step 2. We now look at f (1) . Its definition was
X
(5.1) f (1) (x1 ,y1 ) = xd−d
1
0
fd (1,a0 + y1 ).
d≥d0
We claim that f (1) is again regular in y1 . In view of formula (5.1) we have to show that
fd0 (1,a0 + y1 ) is regular in y1 . We can factorize fd0 (1,y): (Recall that f is regular of order
b.)
fd0 (1,y) = c · (y − a0 ) · · · (y − ab−1 ),
for some constants c,a0 , . . . ,ab−1 ∈ C . From this we see that
fd0 (1,a0 + y1 ) = c · y1 (y1 + a0 − a1 ) · · · (y1 + a0 − ab−1 ),
so fd0 (1,a0 + y1 ) is regular in y1 . Thus f (1) is also regular in y1 , of order at most b. There
are two cases to consider.
(1) f (1) is also regular of order b. Then it follows that a0 = a1 = · · · = ab−1 , so that
fd0 (1,y) = c · (y − a0 )b . As fd0 is quasi-homogeneous and convenient it follows that
fd0 = c · (y − a0 xm )b . So w1 = 1, and m · b = d0 . Furthermore w2 = m, and
x1 = x, y = xm m
1 (a0 + y1 ) = x (a0 + y1 ).
(2) f (1) is regular of order b1 < b.

So if we iterate this procedure, that is, we define
xi+1 := (xi )1 , yi+1 := (yi )1 , f (i+1) := (f (i) )(1) ,
then the second case above can occur only finitely many times in the iteration. For the
rest of the proof, we may therefore assume that we are always in the first case, that is, we
may assume that x = x1 = x2 = · · · . So in this case one tries to find a parametrization
of type x = t, y = ϕ(t), that is a smooth branch y − y(x) = 0 of f = 0.
Step 3. So assuming we are always in the first case, we want to show that the algorithm
gives a solution y = y(x) of f (x,y) = 0. By Corollary 3.3.18, or alternatively, the third
part of Corollary 5.1.8 it suffices to show that y = y(x) is a formal solution of f (x,y) = 0.
So if ai is a solution of f (i) (0,yi ) = 0, the iteration gives
y = xm0 (a0 + y1 ), y1 = xm1 (a1 + y2 ), etc,
for certain m0 ,m1 , . . .. Plugging in we get
y = xm0 (a0 + xm1 (a1 + y2 )) = a0 xm0 + a1 xm0 +m1 + y2 xm0 +m1 = · · · .
We claim that the formal power series
∞
X
y(x) = ak xm0 +...+mk
k=0
is a formal solution of f (x,y) = 0. Indeed, f (x,y) = xd0 f (1) (x,y1 ) = · · · = xkd0 f (k) (x,yk ).
Plugging in y = y(x) we get f (x,y(x)) = xkd0 f (k) (x,yk (x)) for some power series yk (x).
Thus f (x,y(x)) is divisible by all powers of x, and hence is zero.
The next thing we study in this section is the irreducibility of a power series f , in
terms of the Newton polygon. Let us look at the simplest case. Suppose h = f · g, f
and g convenient, and suppose that the Newton polygon of both f and g has just one
face. Let the slope of the face of Γ(f ) be given by the endpoints (p1 ,0) and (0,q1 ), and
similarly the slope of the face of Γ(g) by the endpoints (p2 ,0) and (0,q2 ). We assume
that the slope of the face of Γ(f ) is less than or equal to the slope of the face of Γ(g).
Then h = f · g is convenient. The Newton polygon of h has two faces, resp. one face if
the slopes are equal. The first face has endpoints (0,q1 + q2 ), and (p1 ,q2 ), the second has
endpoints (p1 ,q2 ) and (p1 + p2 ,0).
q1 q1 + q2
Γ(f )
q2 Γ(g) q2
p1 p2 p1 p1 + p2
This is easy to prove. One has to show that each monomial xα y β of h lies right to both
faces. The condition that a monomial xα y β lies right to a line whose intersections with
the axes are (p,0) and (0,q) say, is simply αq + βp ≥ pq. Let xα1 y β1 be a monomial of f
and xα2 y β2 be a monomial of g. Then, by assumption on Γ(f ) and Γ(g) we have
q1 q2
(1) p1 ≥ p2 ;
(2) β1 ≥ − pq11 α1 + q1 ;
(3) β2 ≥ − pq22 α2 + q2 .
This implies that β1 +β2 ≥ − pq11 (α1 +α2 )+q1 +q2 and β1 +β2 ≥ − pq22 (α1 +α2 )+ pq22 (p1 +p2 ).
The first inequality follows by adding (2) and (3) and using (1). The second inequality
follows by adding pp21 q2 to (3) and using the inequality α1 q2 + β1 p2 ≥ p1 q2 . Therefore,
xα1 +α2 y β1 +β2 lies right to both faces. Moreover if xα1 +α2 y β1 +β2 lies on both faces, it
follows that β1 = 0, from which α1 = p1 , β2 = q2 and α2 = 0 easily follow. So the
monomial xp1 y q2 indeed occurs in the product f · g.
More generally, we need the following lemma. The proof is just more technical, and
therefore left to the reader.
Lemma 5.1.16. Let f1 ,f2 ∈ C {x,y} be convenient. Assume Γ(f2 ) has one face ∆1
defined by (p2 ,0) and (0,q1 − q2 ). Denote the faces of f1 by ∆2 , . . . ,∆r and let ∆i be
defined by (pi ,q i ) and (pi+1 ,q i+1 ) with p2 = 0, q 2 = q2 and q r+1 = 0. Assume furthermore
that the slope of ∆1 is smaller than the slopes of the ∆i , i ≥ 2.
Then the Newton polygon Γ(f ) of f = f1 · f2 has faces ∆1 , . . . ,∆r , which are given
by:
• ∆1 defined by (0,q1 ) and (p2 ,q2 );
• ∆i defined by (pi + p2 ,q i ) and (pi+1 + p2 ,q i+1 ) for i = 2, . . . ,r.
q1 − q2 q1
Γ(f1 · f2 )
∆1
∆1
Γ(f2 )
q2
p2
q2 ∆2
∆2
Γ(f1 )
p2
We now come to a converse of this lemma. It follows without much difficulty from
the explicit construction of the Puiseux expansion, that the converse must hold. We want
an explicit construction of the factorization, however.
Theorem 5.1.17. Let f ∈ C {x,y} be convenient, ∆1 , . . . ,∆r be the faces of Γ(f ) and
di the slope of ∆i . Then f = f1 · · · fr , where fi is convenient such that Γ(fi ) has only
one face of slope di , i = 1, . . . ,r.
In the proof of this theorem, we use the following lemma.
Lemma 5.1.18. Consider the polynomial ring C [x,y], with grading defined by w-deg(x) =
w1 > 0 and w-deg(y) = w2 > 0. Let g = xb + . . . be quasi-homogeneous of weighted degree
bw1 . Let h ∈ C [x,y] be quasi-homogeneous with weighted degree at least bw1 + aw2 for
some a ∈ N. Then h ∈ (y a ,g).
Proof. We apply Division with Remainder, see Theorem 2.1.5, and get
h = q · g + r,
where r is a polynomial in x of degree at most b − 1. From the algorithm of division with

remainder, it follows that also r is quasi-homogeneous, in fact w-deg(r) = w-deg(h) if
r 6= 0. Take a monomial xi y j in r. Then i < b. Hence
bw1 + aw2 ≤ w-deg(r) = w-deg(xi y j ) = iw1 + jw2 < bw1 + jw2 .
It follows that j > a, and every monomial of r is divisible by y a . Thus indeed h ∈

(y a ,g).
P
Proof of Theorem 5.1.17. Let f = aij xi y j , assume r > 1 and let ∆1 be the face with
minimal slope. Because f is convenient, ∆1 is the line segment from (0,q1 ) to (p2 ,q2 ):
(0,q1 )
∆1
(p2 ,q2 )
We define
1
g := · f∆1 , h := f − y q2 g.
y q2
We obviously have f = y q2 g + h.
As in Definition 5.1.9 we introduce the weighted degree on the monomials of C {x,y}
corresponding to ∆1 , that is, we put w-deg(x) = w1 = (q1 − q2 )/c, and w-deg(y) = w2 =
p2 /c, where c = gcd(q1 − q2 ,p2 ). With these weights, g is weighted homogeneous of degree
(q1 − q2 )p2 /c. We consider the elements of C {x,y} to be ordered by the corresponding
weighted order function, w-ord(f ) = min{w-deg(xi y j ) : (i,j) ∈ supp(f )}.
(n) (n)
We will now inductively construct f1 , f2 and h(n) such that
(n) (n)
• f = f1 · f2 + h(n) ,
(n)
• f1 = y q2 + terms of higher w-order not divisible by y q2 ,
(n)
• f2 = g + terms of higher w-order,
• p2 q1 /c < w-ord(h) < w-ord(h(2) ) < · · · < w-ord(h(n) ).
(1) (1)
The initialization of course is f1 = y q2 , f2 = g and h(1) = h. Note that the weighted
order of h is indeed bigger than p2 q1 /c, as w-ord(h) > w-ord(y q2 g) = p2 q1 /c. This is
because all monomials of h lie right to the line through the face ∆1 . Now let k be the
quasi-homogeneous part of h(n) of w–degree equal to w-ord(h(n) ). Because of w-deg(k) >
w-ord(y q2 g) we can find by Lemma 5.1.18 weighted homogeneous ξ,η such that k =
ξy q2 + ηg. Moreover one can arrange that no monomial of η is divisible by y q2 . Now set
(n+1) (n)
• f1 = f1 + η,
(n+1) (n)
• f2 = f2 + ξ,
(n+1) (n+1)
• h(n+1) = f − f1 · f2 .
(n) (n)
It is not difficult to check, using w-ord(f1 − y q2 ) > p2 q2 /c and w-ord(f2 − g) >
p2 (q1 − q2 )/c, that
w-ord(h(n+1) ) > w-ord(h(n) ).
Continuing this way, we get formal power series f1 and f2 with f = f1 ·f2 . In particular, by
construction, f1 is a Weierstraß polynomial with respect to y. Therefore, the convergence
of f1 and f2 follows from Corollary 5.1.8.
The Newton polygon Γ(f2 ) = Γ(g) looks like,
q1 − q2
p2
Using Lemma 5.1.16, the lemma follows by induction on r.

Example 5.1.19. We consider f = y 7 + x3 y 5 + x2 y 2 + x5 .
∆1
∆2
(1) (1)
We have w1 = 5, w2 = 2, g = f2 = y12 · f∆1 = y 5 + x2 . So with f1 = y 2 and
(1)
h = x5 + x3 y 5 we get
(1) (1)
f = f1 f2 + h(1) .
Now h(1) is weighted homogeneous of degree 25. The monomials of h(1) all are divisible
by x3 . In fact h(1) = (x2 + y 5 ) · x3 = g · x3 , so that
f = (y 2 + x3 )(y 5 + x2 ).
We have the following generalization.

Theorem 5.1.20. Let f ∈ C {x,y}P be convenient,
P and suppose Γ(f ) has just one face
∆. Suppose that there exist g = ν≥d gν , h = ν≥d hν and R such that
(1) f = g · h + R;
(2) the gν and hν are homogeneous of degree ν with respect to the grading introduced
by ∆;
(3) gν = hν for ν = d, . . . ,p − 1, and gp 6= hp ;
(4) (gd ,hp ) is (x,y)–primary;

(5) w-ord(R) > d + p.
Then f is reducible.
The proof runs as in the previous theorem, and is therefore left to the reader as an
exercise. One needs the following lemma, the proof of which will be given in Chapter
seven, see Lemma 7.3.
(0) (0) (0) (0)
Lemma 5.1.21. Let h1 and h2 be homogeneous and assume that (h1 ,h2 ) is (x,y)–
(0) (1) (0) (1) (1) (0)
primary. Let h1 = h1 + h1 , h2 = h2 + h2 , with w-ord(hi ) > w-ord(hi ). Suppose
i j
that iw1 + jw2 > w-ord(h1 · h2 ) for some i,j ∈ N. Then x y ∈ (h1 ,h2 ).
Examples 5.1.22. (1) In the previous example we were “lucky” to get factors of f
which are polynomials. In general of course, one cannot expect this. We now look at
the standard example f = y 2 − x2 (x + 1). We get f = (y − x)(y + x) + h(1) . Then
k = h(1) = −x3 ∈ (y − x,y + x). In fact −x3 = 21 x2 (y − x) − 12 x2 (y + x). Hence,
f = (y − x − 12 x2 )(y + x + 12 x2 ) + 14 x4 .
Now 41 x4 = − 81 x3 (y − x) + 81 x3 (y + x). Hence
f = (y − x − 12 x2 + 18 x3 ) · (y + x + 21 x2 − 81 x3 ) − 18 x5 + 1 6
64 x ,
etc. √ will not stop, because we get as the factorization f (x,y) = (y −

√ This process
x x + 1)(y + x x + 1).
(2) We consider f = y 4 − 2x3 y 2 + x6 − x7 . We have already seen in Example 5.1.15 that
f is reducible. To apply the theorem, we note that
f = (y 2 − x3 − x2 y + 12 x4 )(y 2 − x3 + x2 y + 12 x4 ) − 14 x8 .
so that the conditions of Theorem 5.1.20 are satisfied, and we get a second proof that f
is reducible.
Exercises
5.1.23. Use the algorithm for the Puiseux expansion
√ to find a solution of y 2 + 2y − x = 0, that
is, find the power series expansion of y = −1 + 1 + x.
5.1.24. Find a factorization of f (x,y) = y 4 − 2x3 y 2 + x6 − x7 in C {x,y}, see Example 5.1.22.
5.1.25. Find the Puiseux expansions of the following (irreducible) power series.
(1) f = y 4 − 2x3 y 2 − 4x5 y + x6 − x7 .
(2) f = x33 − 3x22 y 2 + 6x21 y − x20 + 3x11 y 4 + 2x10 y 3 − y 6 .
5.1.26. Prove Lemma 5.1.16.
5.1.27. Let C ((x)) be the quotient field of C {x} and C ((x)) its algebraic closure. Prove that
C ((x)) = ∪ C ((x1/n )).
n∈N
5.2 Invariants 183
5.1.28. This is an exercise for the reader who knows about resultants. Let (X, 0) and (Y, 0)
be plane curve singularities, which are given by Weierstraß polynomials f and g in y. Thus in
particular f, g ∈ C {x}[y]. Consider the resultant Rf,g ∈ C {x}. Prove that
(X, 0) · (Y, 0) = ordx (Rf,g )
(Hint: Reduce to the case that f and g are irreducible. Let x = tn , y = ϕ(t) be a Puiseux
expansion of (X, 0), and factorize f in C {t}[y]. Compute the resultant in C {t}.)
5.1.29. This exercise uses the previous exercises. Let (X, 0) be a plane curve singularity given
by a Weierstraß polynomial f in y. Let ∆ ∈ C {x} be the discriminant of f . Prove
ordx (∆) = µ(f ) + (X, 0) · (L, 0),
where (L, 0) is the line given by y = 0, and µ(f ) is the Milnor number.
∂f
(Hint: Consider (Y, 0) given by ∂x = 0. Compute the intersection number (X, 0) · (Y, 0) in two
ways. One way is by using the previous exercise, the other way is by using Puiseux expansion for
the branches of (Y, 0): if (ϕ1 (t),ϕ2 (t)) is a parametrization of a branch then look at the vanishing
d
order of t 7→ dt (f (ϕ1 (t),ϕ2 (t))), which differs by one from the intersection multiplicity.)
5.1.30. Let (R,m) be an analytic C –algebra and M a finitely generated R–module. Assume
that dimC M < ∞. Prove that AnnR (M ) 6= {0}.
(Hint: Choose x ∈ m, x 6= 0 and consider M ⊃ xM ⊃ . . ..)
5.1.31. Consider a vector space V over k and a subspace W ⊂ V . Let T : V −→ V be a

linear map, such that T (W ) ⊂ W and Ker(T ) ⊂ W . Suppose that both V /W and V /T (V ) are
finite-dimensional. Prove that dimk (V /T (V )) = dimk (W/T (W )).
5.1.32. Prove Theorem 5.1.20.
5.2 Invariants
In this chapter we wish to study invariants of irreducible plane curve singularities. An

invariant is here a map from the set of isomorphy classes of singularities to Zk for some k;
for example, the dimension of some vector space associated to the singularity, as we shall
soon see. Invariants play an important role in studying singularities. Let R = OX,x be
the local ring of an irreducible plane curve singularity and m the maximal ideal. By the
result of the previous section, we may assume that R ⊂ C {t} =: R, e the normalization of
R.
Definition 5.2.1.
e
(1) We call δ(R) := dimC (R/R) the δ–invariant of R.
(2) We call Γ(R) := {ordt (a) | a ∈ R, a 6= 0} the semi–group of values of R.
(3) We define the conductor by c(R) := min{α ∈ Γ(R) : α + N ⊂ Γ(R)}.

(4) Γ(R) is called symmetric, if there exists a k ∈ N with α ∈ Γ(R) ⇐⇒ k − α ∈
/ Γ(R).
Lemma 5.2.2.

(1) δ(R) = # N \ Γ(R) .
e
(2) AnnR (R/R) e
= tc(R) R.
Proof.
e
(1) Suppose that a1 , . . . ,aδ is a basis of R/R. Suppose that ord(ai ) = ord(aj ) = k,
c
so that ai = ci tk + . . . ,aj = cj tk + . . .. Then we may replace aj by aj − cji ai ,
which has higher vanishing order. Using this operation several times we can get
a basis a1 . . . . ,aδ with ord(ai ) 6= ord(aj ) for i 6= j. Similarly, we can assure that
ord(ai ) 6∈ Γ(R), then the ord(a1 ), . . . , ord(aδ ) are different elements of N \ Γ(R).
On the other hand, if we have different elements c1 < . . . < cγ of N \ Γ(R),
then b1 tc1 + . . . + bγ tcγ with bi ∈ C has vanishing order min{ci : bi 6= 0}. So
b1 tc1 + . . . + bγ tcγ ∈ R only when b1 = . . . = bγ = 0, so that δ(R) ≥ γ.
e
(2) In the proof of 5.1.3 we showed already that AnnR (R/R) is an ideal in R and Re
e ce
and, therefore AnnR (R/R) = t R for some c. We have to prove that c − 1 6∈ Γ(R).
But c − 1 ∈ Γ(R) and tc Re ⊂ R implies that tc−1 R
e ⊂ R and, consequently, tc−1 ∈
e
AnnR (R/R), e
which contradicts AnnR (R/R) e
= tc R.
This shows the Lemma.

Example 5.2.3. The δ invariant of A2 defined by y 2 − x3 is equal to one, the conductor
is equal to two. For the E8 , defined by y 3 − x5 = 0, the δ invariant is four, and the
conductor is eight.
Theorem 5.2.4 (Gorenstein).
c(R) = 2δ(R).
For the proof of this Theorem, we need a definition.
Definition 5.2.5.
(1) Let R be as above, K the quotient field of R. A set I ⊂ K is called a fractional

ideal if
(a) I is via multiplication in K an R–module, that is r · a ∈ I for all r ∈ R and
a ∈ I and, of course, a + a′ ∈ I for all a,a′ ∈ I.
(b) There exists a “universal denominator”, that is, there exists a d ∈ R,d 6= 0,
with d · I ⊂ R.
(2) Let I ⊂ K be a fractional ideal, then I ∗ := HomR (I,R).
(3) The ring R, or the curve singularity is called Gorenstein if for all fractional ideals
I the canonical map
I −→ I ∗∗ = HomR (HomR (I,R),R) a 7→ â,â(ϕ) := ϕ(a)
is an isomorphism.
Remarks 5.2.6.
e of R is a fractional ideal,
(1) Any ideal in R is a fractional ideal. The normalization R
which, in general, is not an ideal.
5.2 Invariants 185
(2) For fractional ideals I,J different from (0) one has the equality
HomR (I,J) = (J : I)K := {x ∈ K : x · I ⊂ J}.
In the remaining part of this Chapter we shall use (J : I) for (J : I)K to simplify
the notation.
The map (J : I) −→ HomR (I,J) sends an element a ∈ (J : I) to the multiplication
with this element in K. On the other hand, a homomorphism α ∈ HomR (I,J) can
be extended to an R–linear map α from K to K by putting α( rs ) := 1s α(r) for an
element r ∈ I and s ∈ R. Then one shows that the element α(1) ∈ (J : I), giving
the inverse map. In particular, we have for (0) 6= J ⊂ I always I ∗ ⊂ J ∗ .
e ∗ = (R : R)
(3) (R) e R = {c ∈ R : cR
e ⊂ R} = AnnR (R/R)
e e
= tc(R) R.
(4) For finite dimensional vector spaces V over a field k, one knows that the canonical
map V −→ V ∗∗ is an isomorphism. The corresponding statement for rings is false,
in general, even if one takes only fractional ideals. An example is given in the
Exercises. The canonical map is injective, but need not be surjective.
(5) As a fractional ideal is as R–module isomorphic to an ideal in R, it follows that

one has to prove the duality only for ideals.
Lemma 5.2.7. Let R be as above. Then R is Gorenstein if and only if dimC (m∗ /R) = 1.
Proof. First suppose that R is Gorenstein. Suppose the statement is wrong. Then we
can find a C –vectorspace I with R $ I $ m∗ . We claim that I is a fractional ideal. The
fact that I has a universal denominator follows immediately from the fact that m∗ has
a universal denominator. So we only have to show that I is an R–module. Obviously
a + a′ ∈ I if a,a′ ∈ I. Let f ∈ R, and a ∈ I. Then we can write f = c + r, with c ∈ C
and r ∈ m. Then r · a ∈ R, as a ∈ m∗ . Furthermore ca ∈ I as I is a vector space. Hence
f a ∈ I.
It follows from the assumption that m $ I ∗ $ R. This is a contradiction, as
dimC (R/m) = 1.
The converse is more difficult, and is proved in two steps.
Step 1. Let (0) 6= J ⊂ I be nonzero fractional ideals. Then we claim that dimC (I/J) ≥
dimC (J ∗ /I ∗ )2 . If J ⊂ L ⊂ I then dimC (I/J) = dimC (I/L)+dimC (L/J) and dimC (J ∗ /I ∗ ) =
dimC (J ∗ /L∗ ) + dimC (L∗ /I ∗ ). We can always find a vector space L with J $ L $ I, and
mL ⊂ J. Indeed, take any vector space L such that J has codimension one in L. By
Nakayama’s Lemma, it follows that mL ⊂ J. It follows as above that L is a fractional
ideal.
Therefore, we can use induction and it suffices to prove the claim for the case
dimC (I/J) = 1. We will construct an injective map
ϕ : J ∗ /I ∗ −→ m∗ /R,
2 Because of R being one–dimensional we have dimC (I/J) < ∞.
which, using the assumption, proves the claim. Take an element a ∈ I with (J,a) =
I. Then m(I/J) $ (I/J) by Nakayama’s Lemma. This implies m(I/J) = 0 because
dimC (I/J) = 1. In particular a · m ⊂ J. We now construct the map ϕ. Let b ∈ J ∗ . Then
ba · m = b · am ⊂ b · J ⊂ R. Thus ba ∈ m∗ . Thus the map ϕ sends b to ba. This gives a
well-defined map, as b ∈ I ∗ sens a ∈ I to an element in R. We now prove ϕ is injective.
So suppose ba ∈ R, for some b ∈ J ∗ . But then it sends all elements of I to R. Thus b ∈ I ∗ .
Step 2. Let I be a nonzero fractional ideal. We may assume that I is an ideal in R. Take
an element x ∈ I. We claim that we have the following equalities:

(∗) dimC (I/(x)) = dimC R/ (x) : I ,
(∗∗) dimC (R/I) = dimC (I ∗ /R).
Indeed, applying Step 1 gives

dimC (R/(x)) = dimC (R/I) + dimC I/(x)

≥ dimC (I ∗ /R) + dimC (x)∗ /I ∗

= dimC (x) : I/(x) + dimC (x) : (x) / (x) : I

= dimC (x) : I /(x) + dimC R/ (x) : I

= dimC R/(x) .

Here we used xI ∗ = x(R : I) = (x) : I and x(x)∗ = x R : (x) = (x) : (x) .
So all inequalities
have to be equalities
and the claim follows
because dimC I/(x) =
∗ ∗ ∗ ∗
dimC (x) /I = dim C x(x) /xI = dim C R/(x) : R . We apply (∗) to the ideal
(x) : I and obtain

dimC (R/I) = dimC (I ∗ /R) = dimC (x) : I /(x) = dimC R/ (x) : ((x) : I) .

Now (x) : I = xI ∗ and (x) : ((x) : I) = (x) : (xI ∗ ) = x(xI ∗ )∗ . As furthermore
(xI ∗ )∗ = x1 I ∗∗ , it follows that
dimC (R/I) = dimC (R/I ∗∗ ).
As I ⊂ I ∗∗ is trivial, the equality I = I ∗∗ follows. This is what we had to prove.

Corollary 5.2.8. Let R be the local ring of a plane curve singularity. Then R is Goren-
stein.
Remark 5.2.9. It follows from the proof that for any two fractional ideals I and J, we
have dimC (I/J) = dimC (J ∗ /I ∗ ) if both are finite. In particular, they are isomorphic as
C –vector spaces.
Proof. We will show that dimC (m∗ /R) = 1. Let 0 =
6 ϕ ∈ m ⊂ R. We consider the
so-called socle
Soc R/(ϕ) := {q ∈ R/(ϕ); m · q = 0 ∈ R/(ϕ)}.
We claim
5.2 Invariants 187

Soc R/(ϕ) ∼= m∗ /R.

Indeed, given g ∈ Soc R/(ϕ) we can assign to this the element ϕg . This is an element of
m∗ , because for all f ∈ m it follows that g · f ∈ (ϕ) and, hence, ϕg · f ∈ R. On the other

hand, suppose we have an element ab ∈ m∗ /R. Then we claim that ab ϕ ∈ Soc R/(ϕ) .
Let f ∈ m be arbitrary. Then ab ϕ · f = ab · f · ϕ ∈ (ϕ). As the maps are obviously inverse
to each other, the claim follows.
Now assume that R is defined by a Weierstraß polynomial f in y of degree m. It
suffices to show that the socle of C {x,y}/(f,x) has vector-space dimension 1. But as
C {x,y}/(f,x) ≃ C {y}/(y m ), we can easily calculate the socle. It is equal to C · y m−1 ,
hence has dimension 1.
Theorem 5.2.10. Let R be a Gorenstein curve singularity. Then c(R) = 2δ(R)
Proof. By duality R/I ≃ I ∗ /R as C –vector spaces for every nonzero ideal I. We apply
e ∗ . Then I ∗ = R,
this to the conductor ideal, which by definition is I = (R) e and we get
e e
R/I ≃ R/R. As the conductor is equal to dimC (R/I), (Lemma 5.2.2 (2)), and δ is equal
e
to dimC (R/R), the Theorem follows.
Definition 5.2.11. Let
β0 = min{a ∈ Γ(R); a 6= 0},

β1 = min{a ∈ Γ(R); gcd (a,β0 ) < β0 },
βν = min{a ∈ Γ(R); gcd (a,β0 , . . . ,βν−1 ) < gcd (β0 , . . . ,βν−1 )}, ν ≥ 2.
By this construction we obviously obtain only finitely many βν , say β0 , . . . ,βg′ . Let
G(R) := {β0 , . . . ,βg′ }.
Lemma 5.2.12.
(1) The set G(R) is the minimal set of generators of Γ(R).

(2) Let ϕ : R −→ R′ be an isomorphism, then G(R) = G(R′ ).
We leave the proof of the Lemma as an Exercise.

Examples 5.2.13.
(1) Let the cusp singularity be given by the parametrization x = t2 ,y = t3 . So R =

C {t2 ,t3 }, and the semi-group Γ(R) is generated by β0 = 2 and β1 = 3.
(2) We consider the parametrization x = t4 ,y = t6 + t7 . Then obviously β0 = 4 and
β1 = 6 are in Γ(R). But also β2 = 13 ∈ Γ(R), as it is the vanishing order of
y 2 − x3 = 2t13 + t14 . We shall soon see that Γ(R) is generated by 4,6,13.
The question is how to calculate generators of the semi-group Γ(R) from the parametriza-
tion. To answer this question, we introduce other invariants, the characteristic exponents.
P
Definition 5.2.14. Let R = C {tn ,y(t)}, y(t) = ai ti and am 6= 0. We define
i≥m
k0 = min{n,m},


min{i|ai 6= 0, gcd (i,k0 ) < k0 } if n ≤ m,
k1 = min{i|ai 6= 0, gcd (i,k0 ) < k0 } + n − m if n > m and m | n, ,


n if n > m and m ∤ n,
(
min{i|ai 6= 0, gcd (i,k0 , . . . ,kν−1 ) < gcd (k0 , . . . ,kν−1 )} if n ≤ m,
kν =
min{i|ai 6= 0, gcd (i,k0 , . . . ,kν−1 ) < gcd (k0 , . . . ,kν−1 )} + n − m if n > m,
ν ≥ 2.
By this construction again, we obviously obtain only finitely many kν , say k0 , . . . ,kg .
Note that in all cases k0 < k1 < . . . < kg and gcd(k0 , . . . ,kg ) = 1 because dimC (C {t}/R) <
∞. The set E(R) := {k0 ,k1 , . . . ,kg } is called the set of characteristic exponents of R.
Examples 5.2.15.
(1) A curve with just one characteristic exponent k0 = 1 is smooth.

(2) Consider the curve with parametrization x = t4 ,y = t6 + t7 . The characteristic
exponents are k0 = 4,k1 = 6,k2 = 7. We have seen in Example 5.2.13 that β0 =
4,β1 = 6 and β2 = 13.
A priori, the k0 , . . . ,kg depend on the parametrization, but in fact they do not! More
precisely, we have the following Theorem, implying this.
Theorem 5.2.16. The minimal set of generators of Γ(R), G(R) = {β0 , . . . ,βg′ }, and
the set of characteristic exponents, E(R) = {k0 , . . . ,kg }, satisfy the relations:
(1) g = g′,
i−1
X
gcd(β0 , . . . ,βj−1 )
(2) ki = βi − − 1 βj ,
j=1
gcd(β0 , . . . ,βj )
i−1
X
gcd(k0 , . . . ,kj−1 ) gcd(k0 , . . . ,kj )
(3) βi = ki + − kj .
j=1
gcd(k0 , . . . ,ki−1 ) gcd(k0 , . . . ,ki−1 )
It is left as Exercise 5.2.24 to show that the formulas in (2) and (3) are equivalent.
Corollary 5.2.17. Let ϕ : R −→ R′ be an isomorphism, then E(R) = E(R′ ).
Definition 5.2.18. Let E(R) = {k0 , . . . ,kg } be the characteristic exponents. Define
P (R) = {(m1 ,n1 ), . . . ,(mg ,ng )}, the set of Puiseux pairs as follows: let kk01 = m
n1
1
such
kν nν
that gcd(m1 ,n1 ) = 1. k0 = m1 ·...·mν such that gcd(mν ,nν ) = 1, ν = 2, . . . ,g.
To digest this definition a little bit, let us write the Puiseux expansion as
X
y= ak xk .
k∈Q
5.2 Invariants 189
If the curve is not smooth, there is a smallest k 1 with ak1 6= 0 and k 1 not an integer. We
n1
write k1 = m 1
in lowest terms, and then (m1 ,n1 ) is the first Puiseux pair of the curve. To
find the second Puiseux pair, we look for the smallest k 2 > k 1 which cannot be written
in the form ma1 . We write k 2 = m1n·m
2
2
with gcd(m2 ,n2 ) = 1 and call (m2 ,n2 ) the second
Puiseux pair. To find the third, we look for the smallest k 3 > k 2 which cannot be written
n3
in the form m1a·m2 . Write k 3 = m1 ·m 2 ·m3
with gcd(m3 ,n3 ) = 1. Then (m3 ,n3 ) the third
Puiseux pair, etc.
Remark 5.2.19. k0 = m1 · . . . · mg and ki = k0 k i , i = 1, . . . ,g. Therefore, P (R)

determines also E(R).
To prove Theorem 5.2.16 we need the following Lemma:
Q
n
Lemma 5.2.20. Let R = C {tn ,y(t)} and f = y − y(εi x1/n ) , ε a primitive n–th root
i=1
∂f n
g
P
of unity, then ord ∂y t ,y(t) = gcd(k0 , . . . ,kj−1 )−gcd(k0 , . . . ,kj ) kj +max{n−m,0}.
j=1
Q
n
Proof. From f = y − y(εi x1/n ) it follows that
i=1
Xn Y
∂f
= y − y(εj x1/n ) ,
∂y i=1 j6=i
and, hence,
∂f n n−1
Y
t ,y(t) = y(t) − y(εi t) .
∂y i=1
It thus suffices to calculate the vanishing order for the different y(t) − y(εi t), their sum
being the vanishing order of ∂f n
∂y (t ,y(t)). We shall do the calculation for n ≤ m and leave
the other cases to the reader.
In our situation k0 = n, and we may expand y(t) − y(εi t) as a power series:
X
y(t) − y(εi t) = aν (1 − εiν )tν .
ν≥m
n o
k0
Setting Nj = i; 1 ≤ i ≤ n − 1, gcd(k0 ,...,kj−1 )
| i, gcd(kk0 ,...,k
0
j)
∤i
and noting that
Sg
gcd(k0 , . . . ,kg ) = 1, we consider the following partition {1, . . . ,n − 1} = j=1 Nj . We

claim that ord y(t) − y(εi t) = kj for any i ∈ Nj .
If εiν 6= 1, then k0 = n ∤ iν, and hence gcd(k0 , . . . ,kj−1 ) ∤ ν, by assumption on i.
Therefore, if in addition aν 6= 0 then ν ≥ kj , by definition of kj , and thus the order of
y(t) − y(εi t) is at least kj . Conversely, we know that akj 6= 0, and thus it suffices to show
gcd(k ,...,k )
that εikj 6= 1, i. e. k0 = nmidikj . Writing q = gcd(k0 0 ,...,kj−1
j)
we get

kj k0 kj
(∗) gcd q, = gcd ,..., = 1.
gcd(k0 , . . . ,kj ) gcd(k0 , . . . ,kj ) gcd(k0 , . . . ,kj )
qk0 k0
By assumption on i we know that gcd(k0 ,...,kj−1 ) = gcd(k0 ,...,kj ) ∤ i. And hence,
i gcd(k0 ,...,kj−1 )
q ∤ k0 , which is an integer, again by assumption on i. From equation ∗ we
i gcd(k0 ,...,kj−1 )kj
deduce that q ∤ k0 gcd(k0 ,...,kj ) , and finally, k0 ∤ ikj . This proves the claim.
n o
Writing Mj = i; 1 ≤ i ≤ n − 1, gcd(kk0 0,...,kj ) |i we easily see that #Mj =
gcd(k0 , . . . ,kj )−1 and Nj = Mj−1 \Mj . Hence, #Nj = gcd(k0 , . . . ,kj−1 )−gcd(k0 , . . . ,kj ),
which finishes the proof.
P
Proof of Theorem 5.2.16. Let R = C {tn ,y(t)},y(t) = ai ti ,am 6= 0. We define β0 , . . . ,βg
i≥m
by the formulas in 5.2.16(3). So we have to show that β0 , . . . ,βg generate Γ(R). This will
be done by induction on g. The case g = 0 is easy, as we then have a smooth curve.
g=1
In this case it follows from the definition of the ki that gcd(k0 ,k1 ) = 1. We have to
prove that β0 = k0 and β1 = k1 generate Γ(R).
We first claim that we can arrange that R = C {tk0 + . . . , tk1 + . . .}, gcd(k0 ,k1 ) = 1
and k0 < k1 . For this we consider the different cases.
n = k0 and we may subtract suitable powers of tk0 from y(t) and

(1) If n ≤ m, thenP
obtain y(t) = ai ti with ak1 6= 0 and R = C {tk0 ,tk1 + . . .}.
i≥k1
(2) If n > m, m ∤ n then m = k0 and n = k1 and R = C {tk0 + . . . , tk1 }.

(3) If n > m and m | n then m = k0 and there exists a polynomial q ∈ C [t] such that

tn − q y(t) = tn−m+k1 + terms of higher order, k 1 = min{i|ai 6= 0, gcd(i,k0 ) = 1}.
We leave it as Exercise 5.2.25 to find q. Then, by definition, we have n−m+k1 = k1
and R = C {tk1 + . . . , tk0 + . . .}.
This proves the claim. By Corollary 5.1.6 there is a Weierstraß polynomial f ∈

C {x}[y], degy (f ) = k0 such that f (tk0 + . . . , tk1 + . . .) = 0. For an arbitrary h ∈ C {x,y},
we can, by the Weierstraß division Theorem, write h = a · f + r for a suitable polynomial
r ∈ C {x}[y], whose degree in y is less than k0 . Hence, to calculate the vanishing order
of h(tk0 + . . . ,tk1 + . . . ), we may suppose that h is a polynomial in y of degree less than
k0 = n. So write
0 −1
kX
h= ri (x)y i .
i=0
k0 +... k1 +... i
The vanishing order of ri (t )(t ) obviously is in k0 N + k1 N. To see that the
order of h(th0 + . . . , + th1 + . . .) is in k0 N + k1 N it is enough to show that all orders of
ri (tk0 +. . . )(tk1 +. . . )i are different, so that no canceling can occur in the sum. Therefore,
suppose
ord ri (tk0 + . . .)(tk1 + . . .)i = ord rj (tk0 + . . .)(tk1 + . . .)j
Then it follows that ik1 ≡ jk1 mod k0 . Since gcd(k0 ,k1 ) = 1, it follows k0 | i − j. As
both i and j are smaller than k0 , this can happen only if i = j.
Induction Step. Now assume g > 1.

5.2 Invariants 191
P i
Let x = tn ,y = ai t be thePparametrization of the curve and f an irreducible
Weierstraß polynomial with f (tn , ai ti ) = 0. For simplicity, we use for short Kj for
gcd(k0 , . . . ,kj ). We shall show by induction on d, that for all h ∈ C {x}[y] with degy (h) <
n
Kd we have
X d
ord h(tn ,y(t) ∈ βν N.
ν=0
This is sufficient because any h ∈ C {x,y} can be written as h = p · f + r, r ∈ C {x}[y]

with degy (r) < n as before.
d=1 We define a new curve by the parametrization
X
x = tn/K1 , y = ai ti/K1 .
m≤i<k2
Here k 2 = k2 − max{n − m,0}. This defines a curve with two characteristic exponents
k0 k1 n
K1 , K1 . Now let h ∈ C {x}[y], degy (h) < K1 . Applying the case g = 1 it follows that
β0 β1
ord h x(t), y(t) ∈ N+ N.
K1 K1
Hence
ord h tn , y(tK1 ) ∈ β0 N + β1 N,
as we replaced t by tK1 . By Taylor’s Theorem

h tn , y(t) = h tn , y(tK1 ) + R.
Therefore, the vanishing order of the first term is in β0 N + β1 N. We claim that R

has higher vanishing order. Because h is a polynomial in y of degree less than Kn1 , say
P i
h= qi y . As in case g = 1 we can show that the vanishing orders of qi (tn )tK1 i are
pairwise different. Hence, there exists an i such
that the vanishing order of qi (tn )tK1 i is
n K1
equal to the vanishing order of h t , y(t ) . Now in R terms of type
∂h n
t , y(tK1 ) ak2 tk2 + . . .
∂y
∂h P
occur, and ∂y = iqi y i−1 . The difference in the vanishing order between
qi (tn )tK1 (i−1) (tk2 + . . .)
and
qi (tn )tK1 i
is k 2 − K1 which is a positive number. Similar for the higher order derivatives of h.
Therefore, the vanishing order of h(tn ,y(t)) is in β0 N + β1 N.
d > 1 We shall first construct a function with vanishing order βd . To define this
function, we define a new curve R by the following parametrization:
X
x = tn/Kd−1 ; y = ai ti/Kd−1 .
m≤i<kd
Here k d = kd −max{n−m,0}. We consider a Weierstraß polynomial f ∈ C {x}[y] of degree

n k0 kd−1
Kd−1 such that f (x,y) = 0. The curve R has characteristic exponents Kd−1 , . . . , Kd−1 .
It follows from 5.2.20
X Kj−1
d−1
∂f n/Kd−1 Kj
ord t ,y = 2 − 2 kj + max{n − m,0}.
∂y j=1
Kd−1 Kd−1
Now consider
∂f n
f tn ,y(t) = f tn ,y(tKd−1 ) + t ,y(tKd−1 ) · (akd tkd + . . .) + R.
∂y
The first term is zero, the second term has vanishing order
X
d−1
Kj−1 Kj
X
d−1
Kj−1 Kj

kd + − kj + max{n − m,0} = kd + − kj = βd .
j=1
Kd−1 Kd−1 j=1
Kd−1 Kd−1
The fact that R has higher vanishing order is proved as in the case d = 1. Hence, we
have constructed an f with vanishing order βd .
Let h ∈ C {x}[y], degy (h) < Knd . By successive application of the Weierstraß division
Ps i n
Theorem we can write h = i=0 hi (x,y)f with hi ∈ C {x}[y], degy (hi ) < Kd−1 and
Kd−1
s< .
Kd
d−1
P
By induction we have ord hi tn ,y(t) ∈ βj N. To conclude the Theorem, again
j=0
we have to show that no canceling occurs. Therefore, suppose
i j
ord hi tn ,y(t) f tn ,y(t) = ord hj tn ,y(t) f tn ,y(t) .
Pd−1
Then (i − j)βd = ord hj tn ,y(t) − ord hi tn ,y(t) so that (i − j)βd ∈ j=0 βj N.
By Exercise 5.2.24, gcd(β0 , . . . ,βd−1 ) = Kd−1 | (i − j)βd . From the definition of βd it
Kd−1
then follows that Kd−1 | (i − j)kd , and, hence, | i − j.
Kd
Kd−1
But as i − j ≤ s < Kd , this implies i = j. This proves that {β0 , . . . ,βg } defined
by (3) in the Theorem generate Γ(R). It remains to prove that {β0 , . . . ,βg } is minimal,
which is left as Exercise 5.2.26.
Remark 5.2.21. Usually the k0 , . . . ,kg are introduced as invariants of the parametriza-
tion and not of R as follows:
k0 = n,
kν = min{i|ai 6= 0, gcd (i,k0 , . . . ,kν−1 ) < gcd (k0 , . . . ,kν−1 )}, for 1 ≤ ν ≤ g.
P
If, then, x = bν τ ν , y = τ m , is the parametrization obtained by τ m = y(τ ), and
ν≥n
k0′ = m, kν′ = min{i|bi 6= 0, gcd (i,k0′ , . . . ,kν−1
′
) < gcd (k0′ , . . . ,kν−1
′
)}, 1 ≤ ν ≤ g ′ , then
the so–called inversion Theorem gives a relation between {k0 , . . . ,kg′ ′ } and {k0 , . . . ,kg }
′
as follows:
5.3 Resolutions of Irreducible Plane Curve Singularities 193
If n < m < k1 then g ′ = g + 1, k1′ = n and kν+1 ′

= kν + n − m. If n = m or k1′ = m
then g ′ = g and kν′ = kν + n − m.
Using our way of defining k0 , . . . ,kg the proof of Theorem 5.2.16 includes also a proof
of the inversion Theorem.
Exercises
5.2.22. Let R be the local ring of a curve singularity, H its semi-group.
(1) Let α ∈ H. Prove that c(R) − 1 − α ∈
/ H.
(2) Prove that 2δ(R) ≤ c(R).
(3) Prove that 2δ(R) = c(R) if and only if H is symmetric, that is,
α ∈ H ⇐⇒ c(R) − 1 − α ∈
/ H.
5.2.23. Let ϕ : R −→ R′ be an isomorphism, then Γ(R) = Γ(R′ ). /Hint: use the fact that ϕ
extends to an isomorphism ϕ : C {t} −→ C {t} of the normalizations of R and R′ , ϕ(t) = t· unit.)
5.2.24.
(1) Show that (2) ⇐⇒ (3) in 5.2.16.
(2) Show that for d = 1, . . . ,g we have gcd(k0 , . . . ,kd ) = gcd(β0 , . . . ,βd ).
P
5.2.25. Let x = tn ,y = i≥m ai ti , with am = 1. Suppose n > m and m | n. Then there exists a
` ´
polynomial q ∈ C [t] such that tn −q y(t) = tn−m+k1 + terms of higher order, k1 = min{i|ai 6= 0,
gcd(i,k0 ) = 1}. P i
Hint: let n = r · m and assume am = 1, then tn − y(t)r = ci t , and if ci 6= 0 for
i>n
`r´ m(r−1)
i < n − m + k1 = (r − 1)m + k1 , the exponent of 1 t ak1 tk1 in y(t)r ), then m | i. If
ci 6= 0 for i < n − m + k1 , subtract again a suitable power of y(t) and continue up to the order
n − m + k1 .
5.2.26. Prove that G(R) = {β0 , . . . ,βg } for β0 , . . . ,βg defined by (3) of Theorem 5.2.16.
Hint: we know that β0 , . . . ,βg generate Γ(R). Use (3) to prove that gcd(β0 , . . . ,βi ) <
gcd(β0 , . . . ,βi−1 ) for i = 1, . . . ,g, and gcd(β0 , . . . ,βg ) = 1.
5.3 Resolutions of Irreducible Plane Curve Singularities
In this picture we blow up a point to a sphere to remove a singularity.

This section is devoted to the study of embedded resolutions of plane curve singu-
larities, which will be done by successive “blowing-ups” of the curve singularity. In order
to define the notion of blowing–up we need some other definitions first.
Definition 5.3.1. We define P1 = P1C to be the set of lines in C 2 which pass through
the origin.
Similarly one can define PnC for n ≥ 2. Alternatively, but very similarly, one could
define P1 as
P1 = (C 2 \ {0,0})/ ∼
Here (x,y) ∼ (x′ ,y ′ ) if and only if there exists a λ ∈ C \ {0} with (x′ ,y ′ ) = λ(x,y).
The equivalence class of (x,y) 6= (0,0) is denoted by (x : y). An equivalence class is called
a point of P1 . The (x : y) are called homogeneous coordinates of P1 .
As any line through (0,0) has a nonempty intersection with the unit sphere in C 2 =
R4 we also could have defined P1 as
{(x,y) ∈ C 2 :| x |2 + | y |2 = 1}/ ∼= §3 / ∼ ,
where (x,y) ∼ (x′ ,y ′ ) if ∃ λ ∈ C with |λ| = 1 so that (x,y) = λ(x′ ,y ′ ). We give P1 the
quotient topology. As a quotient of a compact topological space, P1 is compact. Moreover,
it is left to the reader to show that P1 is Hausdorff. In fact, P1 is homeomorphic to §2 .
This can be seen as follows. Consider a line with homogeneous coordinates (x : y) and
suppose that y 6= 0. Then there is a representative with y = 1, and x can be an arbitrary
element of C . So this subset of P1 is homeomorphic to C . There is only one point of
P1 with y = 0, as all (x,0) are equivalent to (1,0). Thus, it can be seen that P1C is
homeomorphic to the one point compactification of C = R2 , hence to §2 . Another way
to see this, is by considering an open covering
P1 = U0 ∪ U1 ,
where U0 = {(x : y) ∈ P1 : x 6= 0}, and U1 = {(x : y) ∈ P1 : y 6= 0}. Now U0 is

homeomorphic to C via the map f0 : U0 −→ C : (x : y) 7→ xy , and similarly U1 is
homeomorphic to C via the map f1 : U1 −→ C : (x : y) 7→ xy . Note that
f0 ◦ f1−1 : C \ {0} −→ C \ {0}
is the map which sends z to z1 , so is holomorphic. Therefore, P1 carries the structure

of a complex manifold. A complex manifold in general is a topological Hausdorff space
X which has an open covering X = ∪i∈I Ui . These Ui are all homeomorphic to an open
subset Vi in C n via fi , that is, fi : Ui −→ Vi is a homeomorphism. Moreover, for all
i,j ∈ I, the map
fi ◦ fj−1 : fj (Ui ∩ Uj ) −→ fi (Ui ∩ Uj )
should be holomorphic. This makes sense, as these sets are subsets of C n . We shall study
complex manifolds in Chapter 6. To study the blowing up here we need only to know the
P1 .
We now can define the blow-up of C 2 in the origin.
Definition 5.3.2. Consider X = {(x,y, u : v) | xv = yu} ⊂ C 2 × P1C and π : X −→
C ,π (x,y, u : v) = (x,y). π : X −→ C is called the blowing–up of (0,0) in C 2 .
2 2
This map has the following properties:

(1) π −1 (0,0) = {(0,0)} × P1C ,
(2) π|X\({(0,0)}×P1C ) : X \ ({(0,0)} × P1C ) −→ C 2 \ {(0,0)} is an isomorphism.

Remark 5.3.3. Let P1C = U1 ∪ U2 , U1 = {(1 : v) ∈ P1C }, U2 = {(u : 1) ∈ P1C }. Then
X ∩ (C 2 × U1 ) = {(x,xv,1 : v)} ≃ C 2 with coordinates (x,v),

2 2
X ∩ (C × U2 ) = {(yu,y,u : 1)} ≃ C with coordinates (y,u).
We shall now study the behavior of curves under blowing–up.

Let, as before, π : X −→ C 2 be the blowing–up of (0,0) ∈ C 2 . Let C ⊂ C 2 be a
curve through (0,0).

Definition 5.3.4. E = π −1 (0,0) = {(0,0)} × P1C is called the exceptional line or

exceptional divisor, C1 = π −1 C \ {(0,0)} , the closure of the preimage of the curve
without the origin is called the strict transform of C.
First case: C is a line.
If the equation of C is y = ax then on C 2 × U1 = {(x,xv,1 : v) | x,v ∈ C } we have
C1 ∩ (C 2 × U1 ) = {(x,ax, 1 : a) | x ∈ C },
E ∩ (C 2 × U1 ) = {(0,0, 1 : v) | v ∈ C }.
v
C1
a
y y
C
x x
On C 2 × U2 = {(yu,y,u : 1) | y,u ∈ C } we have

(n y o
2 ,y,1 : a | y ∈ C if a 6= 0,
C1 ∩ (C × U2 ) = a
∅ if a = 0,
2
E ∩ (C × U2 ) = {(0,0,u : 1) | u ∈ C }.
Altogether we see that C1 is a line intersecting E in (0,0,1 : a).

If C is defined by x = 0, then we obtain similarly
C1 ∩ (C 2 × U1 ) = ∅
C1 ∩ (C 2 × U2 ) = {(0,y,0 : 1) | y ∈ C },
and again C1 is a line intersecting E in (0,0,0 : 1).

Corollary 5.3.5. Let C be the union of the two lines defined by x = 0, respectively
y = 0, then C1 is the union of two lines which have no intersection point.
(1 : a)
π
E
(1 : b)
Second case: C is an P
irreducible curve. Because of Theorem 5.1.17 we may assume
that f = (y p + αxq )c + aij xi y j defines C with q ≥ p, α 6= 0 and gcd(p,q) = 1.
ip+jq>pqc
Let us assume first that q > p.
y 2 − x3 = 0
C
f (x,xv)
C1 ∩ (C 2 × U1 ) = {(x,xv,1 : v) | f1 (x,v) = 0} with f1 (x,v) = = (v p +
P xord(f )
αxq−p )c + aij xi+j−pc v j .
ip+jq>pqc
E ∩ (C 2 × U1 ) = {(0,0,1 : v) | v ∈ C },
C1 ∩ E(∩C 2 × U1 ) = {(0,0,1 : 0)}.
C (1) x
x − v2 = 0
E v
f (uy,y)
And C1 ∩ (C 2 × U2 ) = {(yu,y,u : 1) | f2 (u,y) = 0} with f2 (u,y) = ord(f ) =
P y
(1 + αy q−p uq )c + ip+jq>pqc aij ui y i+j−pc .
E ∩ (C 2 × U2 ) = {(0,0,u : 1) | u ∈ C },
C1 ∩ E ∩ (C 2 × U2 ) = ∅.
y
f2 = 1 − yu3
E u
The case p = q, that is p = q = 1 can be reduced by the transformation y 7→ y − αx

to the first case.
Altogether we obtain that the strict transform C1 intersects E at one point. Outside
this intersection point, C is smooth.
If [ pq ] > 1, then the multiplicity m(C1 ) = ord(f1 ) = pc = ord(f ) = m(C). If [ pq ] = 1
then the multiplicity drops: m(C1 ) = ord(f1 ) = (q − p)c.
Now we can iterate this procedure.
∼
We identify X ∩ (C 2 × U1 ) − → C 2 : (x,xv,1 : v) 7→ (x,v) and blow up (0,0) in C 2
again.
We shall denote the ambient space after the i–th blowing–up by Xi (we always
take the affine chart U1 or U2 , in which the curve meets the origin), the corresponding
exceptional line by Ei and the strict transform of Ci−1 by Ci :
π π
Xk k
−→ Xk−1 −→ . . . −→ X1 1
−→ C2
∪ ∪ ∪ ∪
Ck Ck−1 C1 C
mk = m(Ck ) mk−1 = m(Ck−1 ) ... m1 = m(C1 ) m0 = m(C).
If we denote by mi = m(Ci ) the multiplicity of Ci at (0,0), then mi ≤ mi−1 , as we have

already seen, and after a while it has to drop strictly. This implies that there is a k such
that Ck is smooth, that is, we have resolved the singularity after k times blowing–up. If
k
∪ Ev intersects Ck not transversally, we blow up again. Let us denote by π = π1 ◦ · · ·◦ πk
v=1
k
the composition of the blowing–ups and E = π −1 (0,0) = ∪ Ei the exceptional divisor.
i=1
π π
Definition 5.3.6. Xk −→ k
Xk−1 −→ . . . −→ X1 −→ 1
C 2 is called standard resolution of
the singularity of the irreducible curve C at (0,0) if either Ci−1 ⊂ Xi still has a singular
point and πi is the blowing–up in Xi−1 of this point or Ci−1 is smooth but the intersection
with Ei−1 in Xi−1 is not transversal and πi is the blowing–up of the intersection point
k
of Ei−1 and Ci−1 in Xi−1 and if, furthermore, Ck is smooth and intersects E = ∪ Ei
i=1
transversally.
(m0 , . . . ,mk−1 ) is called the multiplicity sequence of the singularity of C.
We have proved that every irreducible plane curve singularity has a standard reso-
lution. Furthermore, we leave it as an Exercise to prove that Ei ∩ Ej ∩ Ek = ∅ if i,j,k are
pairwise different.
Let us look for an example.
Example 5.3.7. Let C ⊂ C 2 be defined by f = y 4 − 2x3 y 2 − 4x5 y + x6 − x7 =
(y 2 − x3 )2 − 4x5 y − x7 .
m(C) = 4
The first blowing–up defined by y = xv gives C1 defined by f1 = (v 2 − x)2 − 4x2 v −

3
x = 0. E1 is defined by x = 0.
m(C1 ) = 2
E1
The second blowing–up defined by x = vu gives C2 defined by f2 = (v − u)2 − 4vu2 −

3
vu = 0. E2 is defined by v = 0.
E1
E2
m(C2 )) = 2
Now we change coordinates v v + u and obtain fe2 = f2 (v + u,u) = v 2 − 4u3 −

2 4 3
4u v − u − u v.
The third blowing–up v = xu gives C3 defined by f3 = x2 − 4u − 4ux − u2 − u2 x.
E3 is defined by u = 0.
E3
E2
m(C3 ) = 1
E1
The fourth blowing–up u = xy gives C4 defined by f4 = x − 4y − 4xy − xy 2 − x2 y 2 .

E4 is defined by x = 0.
E4
E3
E2
m(C4 ) = 1
E1
We change again coordinates x x + 4y and obtain fe4 (x,y) = f4 (x + 4y,y) =

2 2 3 2 2
x − 4xy − 16y − xy − 4y − (x + 4y) y .
The fifth blowing up x = yu gives C5 defined by f5 = u − 4uy − 4y − uy 2 − 4y 2 −
(u + 4)2 y 3 . E5 is defined by y = 0.
E5
E4
E3
m(C5) = 1
E2 E1
Now C5 is smooth and intersects the exceptional locus E = ∪Ei transversally.

The multiplicity sequence is (4,2,2,1,1).
Definition 5.3.8. Let π : Xk −→ C 2 be a standard resolution of the singular point
k
0 ∈ C ⊂ C 2 . Let π −1 (0) = E = ∪ Ei be the exceptional divisor and Ck the strict
i=1
transform of C. To this resolution we can associate a weighted graph, the the resolution
graph. The vertices of the graph are the exceptional lines Ei and the curve Ck . We connect
two vertices Ei and Ej if Ei ∩ Ej 6= ∅ and, similarly, we connect a vertex Ei to Ck if
Ei ∩ Ck 6= ∅, that is, we connect Ck to Ek . The weight of Ei in the graph is i. We shall
draw “•” for an Ei and “∗” for Ck .
For our example 5.3.7 we obtain:
∗
5 4
1 3 2
Now we are ready to prove the following Theorem:

Theorem 5.3.9. For a given irreducible curve singularity the following data determine
each other:
(1) the Puiseux pairs P (R);

(2) the characteristic exponents E(R);
(3) the minimal set of generators of the semi-group G(R);
(4) the multiplicity sequence (m0 , . . . ,mk−1 );
(5) the resolution graph.
We proved this already for (1), (2), and (3). To prove that (2), (4) and (5) determine
each other we need to P
study the resolution more carefully.
Let x = tk0 , y = ai ti define a singularity with characteristic exponents k0 , . . . ,kg .
i≥k1
Let
X1 = k1 ,
r1,1 = k0 .
We perform the Euclidean algorithm and obtain:
X1 = µ1,1 r1,1 + r1,2 ,

r1,1 = µ1,2 r1,2 + r1,3 ,
..
.
r1,w(1)−1 = µ1,w(1) r1,w(1) ,
For i = 2, . . . ,g, let Xi = ki − ki−1 and define inductively ri,1 = ri−1,w(i−1) .

We perform the Euclidian algorithm and obtain:
Xi = µi,1 ri,1 + ri,2 ,

ri,1 = µi,2 ri,2 + ri,3 ,
..
.
ri,w(i)−1 = µi,w(i) ri,w(i) ,
with
0 ≤ ri,j+1 < ri,j .

Notice that by the definition of the ki , always w(i) ≥ 2.
Using this data we construct a weighted graph as follows: The building blocks are
the graphs Sij :
• • • ... • • •
π(i,j) + 1 π(i,j) + 2 π(i,j) + 3 π(i,j) + µij − 1 π(i,j) + µij
with
π(1,1) = 0,

π(i,1) = π i − 1,w(i − 1) + µi−1,w(i−1) , for i = 2, . . . ,g,
π(i,j) = π(i,j − 1) + µi,j−1 , for 2 ≤ j ≤ w(i) and i = 1, . . . ,g.
The point with the smallest weight (respectively largest weight) is called the start
point (respectively end point) of the graph Sij .
Now we connect the end point of Sik with the start point of Sik+2 if k ≤ w(i) − 2,
w(i)−1 w(i)
the end point of Si with the end point of Si for all i. Then we connect the end
w(i) 1
point of Si with the start point of Si+1 if µi+1,1 6= 0, otherwise with the start point
3 2
of Si+1 if w(i + 1) ≥ 3 or with the end point of Si+1 if w(i + 1) = 2. Because w(i) ≥ 2,
w(i)
the end point of Si is, therefore, connected with three other points and usually called
w(g)
contact point of the graph. The end point of Sg we connect with the star. The result
is a graph of the following type:
∗
Contact points
...
...
..
.
... ...
... ...
We have g − 1 contact points and draw the graph in such a way that each contact
point creates a new level.
On the other hand it is obvious how to obtain the characteristic exponents from
the resolution graph: g − 1 is the number of contact points, points which are connected
to three other points. One can identify the Sij as the maximal connected subsets of the
graph such that the weights are strictly increasing (respectively decreasing). Starting
w(g)
with multiplicity 1 which occurs µg,w(g) –times in Sg we obtain rg,w(g)−1 = µg,w(g) · 1.
This way, going from bottom to top in the Euclidean algorithm we can reconstruct the
ki .
Let us now look at an example.

Example 5.3.10.
∗
10 11 12 13 14 16 15
9 8
1 2 6 7 5 4 3
From the diagram we can extract the number g of levels and the number ω(i) of
blocks in each level i, as well the blocks Sij themselves and, hence, the µi,j = #Sij . The
remaining data in the following table have to be calculated:
g=3
ω(1) = 3 S11 = {1,2} S12 = {3,4,5} S13 = {6,7}
µ1,1 = 2 µ1,2 = 3 µ1,3 = 2
r1,1 = 28 r1,2 = 8 r1,3 = 4 χ1 = 64
ω(2) = 2 S21 = ∅ S22 = {8,9}
µ2,1 = 0 µ2,2 = 2
r2,1 = 4 r2,2 = 2 χ2 = 2
ω(3) = 2 S31 = {10,11,12,13,14} S32 = {15,16}
µ3,1 = 5 µ3,2 = 2
r3,1 = 2 r3,2 = 1 χ3 = 11
In order to do the calculations we proceed as follows: the initialization step in level three
is done by setting
r3,2 = rg,ω(g) = 1.
We then go on with the following formulas:
r3,1 = µ3,2 · r3,2 = 2 · 1 = 2,

k3 − k2 = χ3 = µ3,1 · r3,1 + r3,2 = 5 · 2 + 1 = 11.
This finishes level three, and we may continue with level two:
r2,2 = r3,1 = 2,
r2,1 = µ2,2 · r2,2 = 2 · 2 = 4,
k2 − k1 = χ2 = µ2,1 · r2,1 + r2,2 = 0 · 2 + 2 = 2.
Finally, in level one we obtain:
r1,3 = r2,1 = 4,
r1,2 = µ1,3 · r1,3 = 2 · 4 = 8,
k0 = r1,1 = µ1,2 · r1,2 + r1,3 = 3 · 8 + 4 = 28,
k1 = χ1 = µ1,1 · r1,1 + r1,2 = 2 · 28 + 8 = 64.
We, therefore, find as characteristic exponents:
k0 = 28, k1 = 64, k2 = 66, k3 = 77.
Now Theorem 5.3.9 is a consequence of the following Theorem:

Theorem 5.3.11.
(1) The graph described above is the resolution graph of the singularity.
(2) The multiplicity sequence contains exactly µi,j times the multiplicity ri,j , i =
1, . . . ,q,j = 1, . . . ,w(i).
In our Example 5.3.7 we have (4,2,2,1,1) as the multiplicity sequence and

∗
S22
S11 S12
as the resolution graph. C was defined by f = y 4 − 2x3 y 2 − 4x5 y + x6 − x7 .

As we have already seen in Exercise 5.1.25 and Example 5.2.15, C has the parametriza-
tion x = t4 , y = t6 + t7 with the characteristic exponents k0 = 4, k1 = 6, k2 = 7. We
want now to calculate the ri,j and the µi,j using their definition. The initial values in the
first level are given by
χ1 = k1 = 6 and r1,1 = k0 = 4.
Proceeding along the lines of the algorithm, i. e. doing division with remainder, we find:
χ1 = 6 = 1 · 4 + 2 = µ1,1 · r1,1 + r1,2 ⇒ µ1,1 = 1, r1,2 = 2,

r1,1 = 4 = 2 · 2 = µ1,2 · r1,2 ⇒ µ1,2 = 2, ω(1) = 2.
The second level thus starts with
χ2 = k2 − k1 = 1 and r2,1 = r1,ω(1) = r1,2 = 2.
Doing division with remainder we then get:
χ2 = 1 = 0 · 2 + 1 = µ2,1 · r2,1 + r2,2 ⇒ µ2,1 = 0, r2,2 = 1,

r2,1 = 2 = 2 · 1 = µ2,2 · r2,2 ⇒ µ2,2 = 2, ω(2) = 2.
And this finishes the algorithm. One easily sees that starting from the diagram we would
get the same values for the µi,j and for the ri,j and, moreover, the reader immediately
verifies the connection between the (ri,j ,µi,j ) and the multiplicity sequence.
Proof of the Theorem. We prove the Theorem using induction on the length k of the
multiplicity sequence.
If the length k is 1, that is, the curve is smooth, defined by x = t, y = t and
k0 = k1 = 1. Consequently X1 = 1, r1,1 = 1 µ1,1 = 1. Thus, from now on we may
assume k > 1. P
Let x = tk0 , y = ai ti , ak1 = 1 and k0 , . . . ,kg be the characteristic exponents and
i≥k1
(m0 , . . . ,mk−1 ) be the multiplicity sequence.
k1 k0
gcd(k0 ,k1 )
The singularity has an equation f of the following type: f (x,y) = x gcd(k0 ,k1 ) −y gcd(k0 ,k1 ) +
P
bij xi y j .
k1 k0
ik1 +jk0 > gcd(k
0 ,k1 )
P The blowing up given by y = vx leads to the parametrization x = tk0 , y =

i−k0
ai t .
i≥k1
There are three possible cases:
(1) If k0 ≤ k1 − k0 , then k0 < k1 − k0 and we obtain as characteristic exponents

k0 ,k1 − k0 , . . . ,kg − k0 .
(2) If k0 > k1 − k0 and k1 − k0 ∤ k0 , then we have as characteristic exponents
k1 − k0 ,k0 ,k2 + k0 − k1 , . . . ,kg + k0 − k1 .
(3) If k0 > k1 − k0 and k1 − k0 | k0 , then k1 − k0 = gcd(k0 ,k1 ) and we obtain as
characteristic exponents k1 − k0 , k2 + k0 − k1 , . . . ,kg + k0 − k1 .
Now recall the following definitions:

X1 = k1 ,
r1,1 = k0 ,
X1 = µ1,1 r1,1 + r1,2 ,
r1,1 = µ1,2 r1,2 + r1,3 ,
..
.
r1,w(1)−1 = µ1,w(1) r1,w(1) ,
X2 = k2 − k1 ,
r2,1 = r1,w(1) ,
X2 = µ2,1 r2,1 + r2,2 ,
..
.
If we denote by X i ,r i,j ,µi,j ,k i the corresponding data after blowing up.
In case (1) we have:
X1 = X1 − k0 = (µ1,1 − 1)r1,1 + r1,2 ,
r1,1 = µ1,2 r1,2 + r1,3 ,
..
.
This implies:
µ1,1 = µ1,1 − 1,
µi,j = µi,j if (i,j) 6= (1,1),
r i,j = ri,j ,
X i = Xi i > 1.
k1 = X1 = k0 + r1,2 ,
that is, we start with
X 1 = r1,1 = µ1,2 r1,2 + r1,3 ,
..
.
This implies:
X 1 = r1,1 ,
µ1,j = µ1,j+1 ,
r 1,i = r1,j+1 ,
X i = Xi ,i > 1,
r i,j = ri,j ,i > 1,
µi,j = µi,j ,i > 1.
X1 = k1 ,
µ1,1 = 1, r1,1 = r1,w(1)−1 = k0 , r1,2 = r1,w(1) = k1 − k0 ,
r1,w(1)−1 = k0 = µ1,w(1) (k1 − k0 ),
X2 = k2 − k1 ,
r2,1 = k1 − k0 ,
X2 = µ2,1 r2,1 + r2,2 ,
..
.
We start with:
X 1 = X2 + k0 = X2 + µ1,w(1) (k1 − k0 ),
r 1,1 = k1 − k0 .
This implies:
X 1 = µ1,1 r 1,1 + r1,2 ,
µ1,1 = µ2,1 + µ1,w(1) ,
and we have:
X i = Xi+1 , i ≥ 2,
r i,j = ri+1,j , (i,j) 6= (1,1),
µi,j = µi+1,j , (i,j) 6= (1,1).
Now we use the induction hypothesis. In the first case, after blowing up the mul-
tiplicity, r1,1 = k0 occurs µ1,1 − 1 times and the multiplicities ri,j occur µi,j times if
(i,j) 6= (1,1). This proves the Theorem in this case.
In the second case, the multiplicity changes after blowing up, that is, m0 occurs
once. Then
X1 = k1 = 1 · k0 + (k1 − k0 ),k0 = X 1 , and k1 − k0 = r 1,1
determine the rest, as before.
In the third case, the multiplicity also changes after blowing up, that is, m0 occurs
once. m1 = r1,2 = r1,1 occurs (µ1,w(1) + µ2,1 ) times. This is covered by X 1 = X2 +
µ1,w(1) (k1 − k0 ). The rest remains unchanged.
This proves the second part of the Theorem.
Now, using induction hypothesis, assume that the singularity after one blowing up
has the resolution graph as described before and denote the corresponding building blocks
j
by S i .
∗
2
S11 S1
We have only to study the behavior at the first building block:
S11 S12
P
If we assume, as before, f (x,y) = (xq − y p )c + bij xi y j to be the equation
ip+jq>pq
for our singularity with gcd(p,q) = 1 and q > 1, then there are two possibilities for the
intersection of the exceptional divisor E1 with the curve after one blowing up:
(1) q−p > p gives rise to a transversal intersection: E1 E1 : x = 0,
(2) q−p < p means that the intersection has maximal contact:
E1 : x = 0.
The second case occurs if and only if the multiplicity changes.
E1 or E1
In the first case the next blowing–up gives:
E2 E2 .
j 1
This implies S i = Sij (with the weights shifted by one) for (i,j) 6= (1,1) and if S is
1 2 3 ℓ−2 ℓ−1
then S11
1 2 3 ℓ−1 ℓ
In the second case we have two possibilities, the number of characteristic exponents
remains constant (if q −p > 1) or drops (if q −p = 1) after the blowing–up (corresponding
to case (2) and (3) in the proof of the second part of the Theorem).
The next blowing–up gives:
E1 : x = 0
E2 : y = 0
The sequence of blowing up continues like
E1 E1
E3 Eℓ
E2
E2
Eℓ−1
up to the next multiplicity change in case q − p > 1, or after p − 1 steps in case

q − p = 1. In the latter case we have after p blowing–ups the equation fp = (x − y)c + . . .,
so that we have to change coordinates.
The next blowing–up in the case q − p > 1 gives:
Eℓ+1
Eℓ
Eℓ−1 E2
E1
that is, in the graph the point corresponding to E1 is connected with the end of
1 j−1
S 1 = S12 (weights shifted by 1) and S11 = {•}. Moreover, S1j = S 1 (weights shifted by
j
1) for j > 1 and Sij = S i (weights shifted by 1) for i > 1.
E2 E3 E4 Eℓ+1 E1
1
S1
In the case q − p = 1 we obtain:
Eℓ+1
Eℓ
E1 Eℓ−1 E2
that is, in the graph the point corresponding to E1 is connected with the ℓ-th point
1
inS1. E1 and the points corresponding to E2 , . . . ,Eℓ+1 give the first building block of
the resolution graph:
E2 E3 Eℓ Eℓ+1 E1
We obtain S11 : the point defined by E1 .
1
S12 the first ℓ–points of S 1 (weights shifted by 1), connected with S11 at the ℓ–th
point.
1
S21 the rest of S 1 (weights shifted by 1), which is not in S12 .
j
S2j = S 1 (weights shifted by 1) for j > 1.
j j
Si = Si−1 for i ≥ 3.
This proves the Theorem.
5.4 Reducible Plane Curve Singularities
As in the irreducible case, we can define the standard resolution of a reducible plane
curve singularity, as follows:
5.4 Reducible Plane Curve Singularities 209
Definition 5.4.1. Let C = ∪ri=1 Ci ⊂ U be a small representative of a reducible plane

πi π2 π1
curve singularity, with branches C1 , . . . ,Cr , r ≥ 2. Assume that Xi −→ . . . −→ X1 −→ U
is already defined. Denote by E (i) = (π1 ◦ · · · ◦ πi )−1 (0) the exceptional divisor and by
C (i) =
(π1 ◦ · · · ◦ πi )−1 (C \ {0}) the strict transform of C.
πi+1
Let Xi+1 −−−→ Xi be the blowing-up of Xi in all points of C (i) ∩ E (i) which are
still singular on C (i) or nontransversal intersection of C (i) with E (i) , that is, points with
intersection multiplicity of C (i) and E (i) greater than one, or where two exceptional
divisors and C (i) meet.
πk π1
Then Xk −→ · · · −→ X1 −→ U is called a standard resolution of (C, 0) if all branches
of C are smooth, do not intersect each-other, do intersect just one component of E (k) ,
(k)
and do intersect this component transversally.

Theorem 5.4.2. Every germ of a plane curve singularity has a standard resolution.
πi+1
Proof. We have to prove that the process of constructing Xi+1 ←−−− Xi as described
above stops after a while, that is, there exists an n such that C (n) has no singularities
on C (n) ∩ E (n) and intersects E (n) transversally. From the irreducible case, we know
that we can resolve the singularities of all branches such that after a while we have
at most nontransversal intersections of C (n) and E (n) . Now it might still happen that
two (smooth) branches of C (n) intersect each other. But we studied the intersections of
smooth curves under blowing-up, and we know that the intersection multiplicity drops
with each blowing-up, see ??. This proves the theorem.
Definition 5.4.3. Consider the standard resolution of a reducible plane curve singularity
(0) (j)
(C, 0) = ∪(Ci , 0). Write Ci := Ci . Furthermore, for some i and k assume that Ci ∩
(j) (j+1) (j+1)
Ck 6= ∅, but Ci ∩ Ck = ∅. Then γ(Ci ,Ck ) := j + 1 is called the contact number
of the branches Ci and Ck .
Definition 5.4.4. With the notations of Definition 5.4.1 we consider the following
weighted graph, the resolution graph of C:
(1) To each component of E (k) a point • is associated.

(2) To each component of C (k) a point ∗ is associated.
(3) Two points are connected by an edge if the corresponding components intersect.
(4) The points of type (1) are weighted. Let E be a component of E (k) . We give to the
corresponding point the weight i, if E is created in the i–th level of the blowing
ups, that is, i is minimal such that πi+1 ◦ · · · ◦ πn (E) is not a point.
Theorem 5.4.5. Let (C, 0) = ∪ri=1 (Ci , 0) be a plane curve singularity. The following
data determine each other:
(1) The resolution graph of C.

(2) The resolution graph of the branches (Ci , 0), together with the contact numbers
γ(Ci ,Cj ) for i 6= j.
Proof.
Step 1. First suppose we know the resolution graph of C. Take a branch (Ci , 0), and delete
all stars from the resolution graph which belong to the branches (Cj , 0) for j 6= i. Call a
point of the remaining graph contractible, if it satisfies one of the following conditions:
(1) It is not connected to the star, and connected only to points of lower weight.
(2) It is connected to a star, and to just one other point of lower weight.
This exactly means that the corresponding curve was created in a blowing-up, which
would not have been necessary for getting the standard resolution of the branch (Ci , 0).
From the resolution process, it follows that such a point is connected to at most two
points in the graph. If it is connected to just one other point, then delete the point of
highest weight, and the edge connected to it. If it is connected to two points A and B
say, then delete the point of highest weight, the connecting edges, and connect A and B
by an edge. This corresponds to “blowing-down” the curve corresponding to this highest
weight. Now iterate this process, until it is no longer possible. We then arrive at the
resolution graph of (Ci , 0). For example, if we have the following resolution graph
* *
* 5 4
2 3 1 3 2
we can get the resolution graph of the curve corresponding to the star on the left by the
following process
delete
* *
* 5 4 * 4
2 3 1 3 2 2 3 1 3 2
* * *
2 3 1 3 2 2 3 1 2 2 3 1
To get the contact number γ(Ci ,Cj ), we remove all the stars corresponding to the
branches Ck with k 6= i,j, and then successively remove points with property (1) and
(2). If this is no longer possible, then we are left with the resolution graph of the union
of (Ci , 0) and (Cj , 0). Consider the following numbers.
(1) The maximal natural number k such that the numbers 1, . . . ,k −1,k occur as weight
just once in the resolution graph of the union of (Ci , 0) and (Cj , 0)}.
(2) The weights of the vertices connected to the stars in the resolution graph of the
union of (Ci , 0) and (Cj , 0).
The contact number of (Ci , 0) and (Cj , 0) is the minimum of these numbers. So for
example in the resolution graph above, the contact number between the curves belonging
to the right two stars is equal to min{5,5,4} = 4, between the curve belonging to the star
on the left and the other two is one.
Step 2. Conversely, suppose we know the resolution graphs of the (Ci , 0) and the contact
numbers. We give an algorithm for finding the resolution graph of (C, 0). The proof
goes by induction on r. The case r = 1 is trivial. So suppose that the resolution graph
Γ of ∪r−1i=1 (Ci , 0) has been constructed out off the resolution graphs of the (Ci , 0) and
the contact numbers. For i = 1, . . . ,r − 1, define ℓ(i) = γ(Cr ,Ci ), and let l(r) be the
maximum of γ(Ci ,Cr ) for i 6= r. Let w(i) be the weight of the vertex in Γ belonging
to the exceptional curve which intersects the strict transform of Ci , for i = 1, . . . r − 1.
Let w(r) be the vertex of highest weight in the resolution graph Γr of (Cr , 0). It is
clear that the point in the resolution graph connected to the star, which corresponds to
(Ci , 0) has weight max{ℓ(i),w(i)}. So if ℓ(i) > w(i), we need, starting from the standard
resolution, ℓ(i) − w(i) extra blowing-ups. Therefore, if ℓ(i) > w(i), we modify the graph
Γ as suggested by the following picture.
∗
ℓ(i)
..
.
∗
... ... ... ...

w(i) w(i)
Γ Γ′
If ℓ(i) ≤ w(i) we leave Γ unchanged at the star corresponding to (Ci 0). We get a
modified graph Γ′ . Similarly, we get a modified graph Γ′r . Let k = max{ℓ(j) : j < r} be
the maximum of the contact numbers of (Cr , 0) with (Cj , 0) for j < r. Without loss of
generality, we may suppose that this maximum is achieved for j = 1. Let π1 ◦ · · · ◦ πk :
Xk −→ U be the map which occurs in the standard resolution process of ∪ri=1 (Ci , 0) after
(k) (k)
k steps. So by definition of contact number, the strict transforms C1 and Cr intersect
(k)
a common component E1 of E (k) , but do not intersect each other. In fact, by choice of
(k) (k)
k, Cr does not intersect any of the Cj . Let p be the point of Γ′ which corresponds to
(k)
E1 . If k > 1, it follows from the resolution process that there is a unique component
(k)
E ′ of E (k) which intersects E1 and was created under πk−1 . The point of Γ′ which
′
corresponds to E we call q. It has weight k − 1. There are two possible cases.
(k)
(1) k > 1, and the curve Cr intersects E ′ . Note that the points p and q in Γ′ are
connected by an edge. The following is a possible picture of a part of the resolution
graph after k blowing-ups.
(k)
C1
E′
(k)
E1
(k)
Cr
(k)
(2) k = 1, or the curve Cr does not intersect E ′ .
(k)
C1
E′
(k)
E1
(k)
Cr
In the second case, consider the subgraph Γ′′r of Γ′r which one gets by deleting all points
(k−1)
of weight less than k. So this is the resolution graph of Cr with all weights shifted
(k) (k)
by k − 1. As we already noted that none of the Cj intersects Cr , it follows that the
resolution graph of (C, 0) is obtained by identifying the unique point of Γ′′r of weight k
with p, which also has weight k.
In the first case, consider the subgraph Γ′′r of Γ′r which one gets by deleting all points
of weight less than k − 1. The graph Γ′′r looks like
k k−1
and we get the resolution graph (C, 0) by doing surgery: In Γ′ one replaces the
subgraph
p q
k k−1
by Γ′′r . So we see that we can construct the resolution graph (C, 0) from the resolution
graphs of (Ci 0) and the contact numbers.
Example 5.4.6. We consider the reducible plane curve singularity given by f = 0, where
f = f1 f2 f3 with
f1 = (y 2 − x3 )2 − 4x5 y − x7 ,
f2 = y 2 − x3 ,
f3 = y 3 − x2
The second and third branch are cusp singularities, and we know how the resolution for
those looks like, see the fourth part of Example ??. The resolution of the first branch has
been studied in Example 5.3.7. So we have to put the following three resolution graphs
into one graph.
*
5 4 * *
1 3 2 2 3 1 2 3 1
For this we have to determine the contact numbers.

(1) (1)
The interesting parts of the strict transforms C1 and C2 of the blowing-up of the
branches (C1 , 0) and (C2 , 0) are in the first chart:
(1)
C1 : defined by (y 2 − x)2 − 4x2 y − x3 = 0,
(1)
C2 : defined by y 2 − x = 0,
E1 : defined by x = 0.
In the second chart we obtain:
(1)
C3 : defined by y − x2 ,
E1 : defined by y = 0.
So we have the following picture
(1)
C2
(1) C3
(1)
C1
E1
and deduce that the contact number of (C1 , 0) and (C3 , 0) is equal to one. Similarly, the
contact number of (C2 , 0) and (C3 , 0) is equal to one. We still have to determine the
contact number of (C1 , 0) and (C2 , 0). So we look at the resolution of the union of (C1 , 0)
and (C2 , 0). One can calculate that the sequence of blowing-ups continues as suggested
in the following picture.
(3)
C2 (3)
C1
(2)
C2
(2)
C1 E3
E2
E1
E1
E2
(4) (4)
We see that the on the fourth blowing-up the branches C1 and C2 are separated.
Thus the contact number is four. It follows that the resolution graph of the curve (C, 0)
is equal to
* *
* 5 4
2 3 1 3 2
by using the algorithm described in the proof. Of course, we could also have gotten this
resolution graph by simply doing the whole resolution process.
Finally, summarizing all results, we obtain the following theorem:
Theorem 5.4.7. Let (C, 0) be a germ of a plane curve singularity. The following data
determine each other:
(1) the multiplicity sequences of the branches of C and the contact numbers between
the branches;
(2) the resolution graphs of the branches of C and the contact numbers between the
branches;
(3) the resolution graph of C.
Next we want to characterize the contact numbers of the branches by their intersec-
tion multiplicity.
Theorem 5.4.8. Let (C, 0) respectively (D, 0) be irreducible plane curve singularities
with multiplicity sequences (m0 , . . . ,mg ) respectively (n0 , . . . ,nh ). For k ≥ g + 1 define
mk = 1, and similarly, for k ≥ h + 1 define nk = 1. Then
γ(C,D)
X
(C, 0) · (D, 0) = mi · n i .
i=0
Proof. Of course we use Lemma 5.1.5 and induction on γ(C,D). After a general linear
coordinate change we may assume that (D, 0) is given by a Weierstraß polynomial f =
y n + a1 y n−1 + . . . + an of degree n = n0 in y, and that (C, 0) has as parametrization
x(t) = tm , y(t) = tq + . . . , where m = m0 .
By Lemma 5.1.5 we have (C, 0)·(D, 0) = ordt (f ). We look at what happens after blowing-
up. Then
x(t) = tm , v(t) = tq−m + . . . .
is a parametrization of the strict transform C (1) . An equation for the strict transform
D(1) of (D, 0) is given by f (x,v) = 0, where f (x,v) is defined by
f (x,v) = x−n f (x,xv).

Therefore,

ordt f x(t),y(t) = ordt f x(t),x(t)v(t)

= ordt x(t)n · ordt (f (x(t),v(t)))
=m0 · n0 + C (1) · D(1)
by induction.
Example 5.4.9. Consider the curve (C, 0) given by (y 2 − x3 )2 − 4x5 y − x7 = 0, and
(D, 0) given by g = y 2 − x3 = 0. The multiplicity sequence of (C, 0) is (4,2,2,1,1), and the
multiplicity sequence of (D, 0) is (2,1,1). We have seen that the contact number is four, see
Example 5.4.6. It follows that the intersection multiplicity is equal to 4·2+2·1+2·1+1·1 =
13. The curve (C, 0) is given by the parametrization x = t4 , y = t6 + t7 . So g is in C {t}
equal to (t6 + t7 )2 − (t4 )3 = 2t13 + t14 , so its order is thirteen. Therefore, conversely, one
can now deduce that the contact number is four.
As in the irreducible case, we define the δ–invariant.
e 0) its normalization. Then
Definition 5.4.10. Let (C, 0) be a curve singularity and (C,e
we define the δ–invariant by δ(C, 0) = dimC (OC,e
e 0 /OC,0 ).
The following easy lemma is useful if one wants to calculate the δ–invariant for
reducible singularities.
Lemma 5.4.11 (Lemma of Hironaka). Let (C, 0), and (D, 0) in (C n , 0) be reduced curve
singularities which do not have a component in common. Then
δ((C, 0) ∪ (D, 0)) = δ(C, 0) + δ(D, 0) + (C, 0) · (D, 0).
Proof. Set I = I (C, 0) ⊂ On and J = I (D, 0) ⊂ On . Consider the following exact

sequence
0 −→ On /(I ∩ J) −→ On /I ⊕ On /J −→ On /(I + J) −→ 0.
We conclude from this exact sequence that the vector space dimension of
OC,0 ⊕ OD,0 /OC∪D,0 and the intersection multiplicity (C, 0) · (D, 0) are equal. By split-
ting of the normalization we know that OC,e
e 0 ⊕ OD,e
e 0 = OC∪D,
^0 . Therefore δ(C ∪ D, 0) is

equal to dimC OC,e
e 0 ⊕ OD,e
e 0 /OC∪D,0 , which is equal to

e 0 ⊕ OD,e
dimC (OC,e e 0 )/(OC,0 ⊕ OD,0 ) + dimC (OC,0 ⊕ OD,0 )/OC∪D,0
= dimC (OC,e e 0 /OD, 0 ) + (C, 0) · (D, 0)

e 0 /OC,0 ) + dimC (OD,e
= δ(C, 0) + δ(D, 0) + (C, 0) · (D, 0).
We wish to finish this chapter with a proof of Max Noether’s Theorem. This theorem
tells us how to compute the δ–invariant from the multiplicity sequence of an irreducible
plane curve singularity. In the proof of Noether’s Theorem, we need the following lemma.

Lemma 5.4.12. Let (C, 0) = V (f ), 0 be an irreducible plane curve singularity. Let the
multiplicity be m = mult(C, 0), and c = c(C, 0) be the conductor. Suppose g ∈ OC,0 ⊂
e 0 = C {t} with ordt (g) ≥ c. Then the multiplicity of g is at least m − 1.
OC,
Proof. Suppose the converse. Then after a general linear coordinate change, we may
assume that g is regular in y of order k < m − 1 in y. We may also assume that f is
regular of order m in y, so that (C, 0) has parametrization x = tm , y = ts + . . . We now
look at h := xy · g ∈ OC,
e 0 . This is indeed in OC,
y
e 0 , since x = t
s−m
· unit ∈ C {t} = OC,
e 0.
Obviously ordt (h) ≥ c(C, 0), hence h ∈ OC,0 . Thus in OC,0 we have the equality yg = xh.
Therefore,
qf = yg − xh
for some q ∈ C {x,y}. Now on the right-hand side a term y k+1 occurs, as it occurs in yg
but not in xh. But it cannot occur on the left-hand side, as the multiplicity of f is m,
and k + 1 < m. This is a contradiction.
Theorem 5.4.13 (M. Noether). Let (C, 0) be an irreducible plane curve singularity,
(m0 , . . . ,mr ) be the multiplicity sequence of (C, 0). Then
r
X mi (mi − 1)
δ(C, 0) = .
i=0
2
Pr
Proof. As c(C, 0) = 2δ(C, 0) by 5.2.4, it suffices to show c(C, 0) = i=0 mi (mi − 1). The
formula, obviously, holds for the case of a smooth curve, since c(C, 0) = 0 and m0 = 1.
Let C (1) be the strict transform of (C, 0) under blowing-up U in the origin for some
sufficiently small neighborhood U . We claim that if the formula holds for C (1) , then it
holds for (C, 0). By induction on r, this suffices to prove the theorem. Suppose (C, 0) is
given by the parametrization
x = tm , y = ts + . . . , s > m.
Step 1. We first show α ∈ Γ(C (1) ) =⇒ α + m(m − 1) ∈ Γ(C, 0).

The strict transform C (1) has parametrization (x = x,y = xv):
x = tm , v = ts−m + . . . .
Let g = g(x,v) ∈ OC (1) . As we are looking at the vanishing order of g in OC, e 0 = C {t},
and C (1) is the zero set of a function which is regular of order ≤ m, we may, by applying
the Weierstraß Division Theorem, assume that degv (g) ≤ m − 1. We write
g = am−1 v m−1 + am−2 v m−2 + . . . + a0 , ai ∈ C {x}.
Multiplying g with xm−1 we get a function that we call g(x,y):
g(x,y) := xm−1 g(x,v) = am−1 y m−1 + xam−2 y m−2 + . . . + a0 xm−1 .
As (C, 0) and C (1) have the same normalization, it follows that g(x,y) ∈ OC,0 and has
vanishing order ordt (g) + m(m − 1). Hence it follows that
α ∈ Γ(C (1) ) =⇒ α + m(m − 1) ∈ Γ(C, 0),
as claimed.
Step 2. To finish the proof we show that
6 ∃ g ∈ OC,o with ordt (g) = c(C (1) ) + m(m − 1) − 1.
Suppose the converse, that is, let g ∈ OC,0 with ordt (g) = c(C (1) ) + m(m − 1) − 1. It is
left as Exercise 5.4.15 to show that we may take such g which is moreover reduced. By
the first step we know that c(C, 0) ≤ c(C (1) ) + m(m − 1) − 1. By the previous lemma, we
know that the multiplicity of g, say q, is at leastm − 1. Let g(x,v) = x−q g(x,xv). Then
5.4.8 gives that (C, 0) · V (g), 0 = C (1) · V (g), 0 + qm. Hence

C (1) · V (g), 0 = c(C (1) ) + m(m − 1) − 1 − qm,

because (C, 0) · V (g), 0 = ordt (g). Thus, we have found a function g ∈ OC (1) which
has vanishing order ordt (g) = c(C (1) ) + m(m − 1) − qm − 1. Now q ≥ m − 1, so that
xq−m+1 · g ∈ OC (1) . This function has vanishing order c(C (1) ) − 1. This is a contradiction
to the definition of the conductor c(C (1) ).
Example 5.4.14. Consider the plane curve singularity given by (y 2 − x3 )2 − 4x5 y − x7 =
0. The multiplicity sequence is (4,2,2,1,1), so the delta–invariant is equal to 6 + 1 + 1 = 8.
This can also be calculated immediately, with the aid of the characteristic exponents,
which are 4,6 and 7. Thus one can calculate the minimal set of generators of the semigroup
to be 4,6 and 13. The fact that δ = 8 is then easily seen by counting.
Exercises
5.4.15. Prove that one may take g in the proof of 5.4.13 to be reduced.
(Hint: Add elements of high order to some factors which are multiple.)
5.4.16. Let (C, 0) = ∪(Ci , 0) be a curve singularity. Consider the standard resolution of (C, 0),
for some small U :
Xk −→ . . . −→ X1 −→ U.
Let C (j) be the strict transform of C in Xj . An infinitely near point of C is, by definition, a
singular point of C (j) for some j, C (0) = C. Prove that
X mp (mp − 1)
δ(C, 0) =
p infinitely
2
near point
where mp is the multiplicity of C (j) at p.

(Hint: Use the Lemma of Hironaka 5.4.11, Lemma 5.4.8 and Lemma 5.4.13.)
5.4.17. Let (C, 0) be the plane curve singularity defined by f = (x3 − y 4 )(x3 − y 7 )(y 11 − x5 ) = 0.
Compute the multiplicity sequences of the branches of (C, 0), the contact numbers, and the res-
olution graph of (C, 0). Also construct the resolution graph of (C, 0) out off the resolution graphs
of the branches and their contact numbers.
5.4.18. Compute the δ–invariant of

(1) the D4 –singularity: y(y 2 − x2 ) = 0,
(2) the E7 –singularity: y(y 2 − x3 ) = 0, and
(3) m different lines through the origin.
5.4.19. Let (C, 0) be a reduced plane curve singularity given by a Weierstraß polynomial f
in y of degree b and multiplicity m. Let C (1) be the strict transform under blowing-up, E the
exceptional divisor. Prove:
P (1)
(1) p∈C (1) ∩E (E,p) · (C ,p) = b.
P
(2) µ(X, 0) − 1 = p∈C (1) ∩E µ(C (1) ,p) + m(m − 1).
(Hint: Use Exercise 5.1.29, and study the behavior of the terms in that formula under
blowing-up; use part 1 of this exercise and use Lemma 5.4.8.)
(3) Deduce Milnor’s formula
µ(C, 0) = 2 · δ(C, 0) − r(C, 0) + 1,
where r(C, 0) is the number of branches of (C, 0).

219
6 The Principle of Conservation of

Number
The aim of this chapter is to study the behavior of certain invariants of singularities in
a family. The idea is that the invariants should be constant in the following sense: if in
the family a singular point splits into several singular points by varying the parameter,
then the sum of the invariants of those singular points should be equal to the invariant
of the point we started with.
0 s
The picture describes a typical situation. In the fiber above the point 0, we have an
intersection of a line with a cusp and the intersection multiplicity at that point is three.
In a small neighborhood we have in the fiber three different intersection points and a
transversal intersection at each point of multiplicity 1. Altogether we see a conservation
of the intersection number. We will see that other invariants, such as the Milnor number
and the δ–invariant have the same property.
The basis to prove such this property is the theory of sheaves. Therefore, we start
the chapter with an overview of the basic facts of sheaf theory. In short, a presheaf S
of rings (or modules, or . . .) on a topological space gives for all open subsets U a ring
S (U ) (or module, or . . .), with some natural conditions. An element of S (U ) is called
a section over U . One can talk about the germ of a section at any point of U . The
set of germs at x is called the stalk at x, notation Sx . A sheaf is a presheaf, which
is, loosely speaking, determined by its stalks. The application we have in mind is that
these stalks are modules whose vector space dimension gives an invariant of a germ of
an analytic space. Thus, we want to compare the various stalks for a given sheaf. Now
general sheaves are like “sand” but coherent sheaves are not. This is reflected in the
statement that a map of sheaves which induces an isomorphism at the stalk at x, is an
isomorphism in a small neighborhood of x. Loosely speaking, the stalk determines the
sheaf in an open neighborhood. The problem with the theory of coherent sheaves is that
it is difficult to give nontrivial examples. The only examples one gets from the definition
are the zero sheaf, and finitely generated sheaves over a point. However, it is true that
most sheaves we consider are coherent. For example OX , that is, the sheaf of holomorphic
220 6 The Principle of Conservation of Number
functions, the ideal sheaf of an analytic set and the normalization sheaf of an analytic
space are coherent. Moreover, for finite maps coherence is preserved (Finite Mapping
Theorem). These theorems are all proved in Section 6.3. In Section 6.2 we develop the
technical tools. Here we prove the Meta-Theorem, which says that coherence of sheaves
is preserved under all “reasonable” algebraic operations. In Section 6.2 we discuss these
coherent sheaves, but give only two examples.
On this basis, in Section 6.4, the principle of conversion of number is studied. In
particular we study the behavior of the invariants mentioned above. Finally, in Section
6.5, we study Cohen-Macaulay spaces. This notion is closely related to the principle of
conservation of number. Hypersurfaces, more generally, complete intersections are Cohen-
Macaulay.
The space in the picture is not Cohen-Macaulay. There is an algebraic method of de-
ciding whether spaces are Cohen-Macaulay. This method is by using resolutions and the
Auslander-Buchsbaum formula. From this it follows that a space is Cohen-Macaulay if
and only if the resolution has the right length.
For lack of a better site, we will finally discuss the Hilbert-Burch Theorem. It says
that the ideal of Cohen-Macaulay codimension two spaces can be described by the t
minors of a certain t × (t + 1) matrix. It is used in Chapter 10 to give examples of
singularities for which there exists a semi-universal deformation over a smooth space.
6.1 Sheaves
Definition 6.1.1. Let X be a topological space. A presheaf S of rings on X consists of

the following data
(1) for each open set U of X a ring S (U ); elements of S (U ) we call sections of S
over U , elements of S (X) are called global sections of S ;
(2) for each inclusion V ⊂ U of open sets in X a ring homomorphism (called the
restriction map):
ρUV : S (U ) −→ S (V ).
These data have to satisfy the following conditions:

1. S (∅) = 0,
2. ρUU : S (U ) −→ S (U ) is the identity map,
3. if W ⊂ V ⊂ U are three open subsets of X then ρUW = ρV W ◦ ρUV .
6.1 Sheaves 221
For s ∈ S (U ) one usually writes s|V instead of ρUV (s). For all open subsets V we
get in a natural way a presheaf S|V which we call the restriction of S to V .
Similarly one defines presheaves of abelian groups, of vector spaces, C –algebras, etc.
Examples 6.1.2.
(1) Let X be the topological space consisting of just one point. Then a presheaf of
rings on X can be identified with a ring.
(2) Let X be a topological space. For all open U ⊂ X put
C (U ) := {f : U −→ C , f is continuous}.
For V ⊂ U , let ρUV be the obvious restriction of functions from U to V . Then one
shows without difficulty that this defines a presheaf of rings, or even of C –algebras
on X.
(3) Consider an open set X ⊂ C n . Put for U ⊂ X open:
O(U ) := {f : U −→ C , f is holomorphic},
and again for an open subset V of U , the map ρUV is given by restriction. This is
a presheaf of C –algebras.
(4) Consider an open set X ⊂ C n . For U ⊂ X open we put
O′ (U ) := {f : U −→ C , f is holomorphic},
but we define different restriction maps. For V $ U we put ρUV = 0, and ρUU = id.
Then O′ is a presheaf.
(5) Let A be a ring, X a topological space. The constant presheaf C is given by C (U ) =
A if U 6= ∅, and C (∅) = 0. The maps ρUV is the identity map if V 6= ∅. Moreover,
ρU∅ is the zero map for all open U .
Definition 6.1.3 (Sheaf Conditions). Let X be a topological space, S a presheaf on
X. Then S is called a sheaf if the following two sheaf axioms are satisfied.
(1) Let U be an open set in X, and U = ∪i∈I Vi be an open covering of U . Let s ∈ S (U )
be a section of S over U , such that s|Vi = 0 for all i. Then s = 0.
(2) Let U be an open set in X, and U = ∪i∈I Vi be an open covering of U . Let si ∈ S (Vi )
be given such that si|Vi ∩Vj = sj|Vi ∩Vj for all i,j ∈ I. Then there exists an s ∈ S (U )
such that s|Vi = si .
A presheaf which satisfies the sheaf axioms, so is in fact a sheaf, is also called a canonical
presheaf.
Remark 6.1.4. It follows from the first condition that such an s, existing according to
the second condition, is uniquely determined.
Example 6.1.5. In 6.1.2, the first three examples, in fact, are sheaves. The fourth
example, the presheaf O′ is not a sheaf, as it does not satisfy the first condition. Moreover,
the constant presheaf A is, in general, not a sheaf. In fact, suppose U,V are two open
subsets of X with U ∩ V = ∅. Let a,b ∈ A, with a 6= b. Then the restrictions of a and b
to U ∩ V = ∅ are equal, but there does not exist a section of U ∪ V , restricting to a and
b, respectively.
Definition 6.1.6. Let S be a presheaf (of rings) on a topological space X and p ∈ X

be a point. Consider the collection of rings:
{S (U ) : U ∋ p an open neighborhood of p}.
We define an equivalence relation on the disjoint union of this collection:

s ∈ S (U ) ∼p t ∈ S (V ) ⇐⇒ there exists an open neighborhood W ⊂ U ∩ V of p such
that s|W = t|W .
The equivalence class of s in p we call the germ of s in p, notation sp . The set of equivalence
classes we call the stalk of S at p, and will be denoted by Sp . It is not difficult to show
that, in a natural way, Sp is a ring again, see Exercise 6.1.28. The support of S is the
set of all points p ∈ X such that Sp 6= 0. The support of S is denoted by supp(S ).
Definition 6.1.7. Let two presheaves S and T (of rings) on a topological space, with
restriction mappings ρUV and ρ′UV respectively, be given. A morphism of presheaves
f : S −→ T is a collection of ring homomorphisms
fU : S (U ) −→ T (U ),
for all open subsets U of X, such that for all V ⊂ U ⊂ X open, the diagram
fU
S (U ) / T (U )
ρU V ρ′U V
fV

S (V ) / T (V )
commutes. A morphism of sheaves is given by a morphism of the underlying presheaves.

A subsheaf of a sheaf F is a sheaf G such that for all open subsets U of X G (U ) is
a subring of F (U ), and the restriction homomorphism are those induced by F , that is,
the inclusion mapping is a morphism of presheaves.
Examples 6.1.8.
(1) Consider the presheaves C and O on C n defined in Example 6.1.2 and f : O −→ C

defined by the canonical injection O(U ) ⊂ C (U ). Then f is a morphism of sheaves.
(2) The map defined by the canonical injection O′ (U ) ⊂ C (U ) is not a morphism of
presheaves.
Lemma 6.1.9. Let f : S −→ T be a morphism of presheaves on X. Then there is, for

all p ∈ X, a naturally defined ring homomorphism
fp : Sp −→ Tp .
This map is defined as follows. Take an element ap ∈ Sp , and a representative a ∈ S (U )

of a in some open neighborhood U of a. Then define fp (ap ) as the germ in p of f (a).
The fact that this map is well defined is left as Exercise 6.1.29.
The second part of the following theorem says, loosely speaking, that a sheaf is
determined by its stalks.
6.1 Sheaves 223
Theorem 6.1.10. Let f : S −→ T be a morphism of sheaves on a topological space X.

Let, for all p ∈ X, fp : Sp −→ Tp be the induced map on stalks. Then
(1) fU is injective for all U open in X if and only if fp is injective for all p;
(2) f is an isomorphism if and only if fp is an isomorphism for all p.
Proof. In both cases, one direction is quite obvious. If fU is injective (resp. an isomor-
phism) for all U , then fp is injective (resp. an isomorphism) for all p.
(1) We prove that fU is injective. Let s ∈ S (U ), with fU (s) = 0. In particular, for all
p ∈ U , the germs (fU (s))p = 0. Thus fp (sp ) = 0. By assumption sp = 0 for all p ∈ U .
This means that for all p there exists a neighborhood Up of p with s|Up = 0. By the first
sheaf axiom, it follows that s = 0. Thus fU is injective.
(2) We need to show that for all U open in X, the map fU : S (U ) −→ T (U ) is an
isomorphism. Namely, if this is the case, we can define the inverse map f −1 by (f −1 )U :=
(fU )−1 . It is a direct check that this gives indeed an inverse map. The first part gives that
fU is injective. So it remains to prove the surjectivity of fU . Take a section t ∈ T (U ). Let
p ∈ U , and consider tp . As fp is surjective by assumption, there exists an element sp with
fp (sp ) = tp . Take a representative sep ∈ S (Up ) of sp for some small open neighborhood
Up of p. We may, by making Up smaller if necessary, assume that fUp (e sp ) = t|Up . So we
have
(1) an open covering {Up }p∈U of U ,
(2) sections sep ∈ S (Up ),
sq )|Up ∩Uq for all p,q ∈ U .
sp )|Up ∩Uq = (e
(3) such that (e
The third condition follows as (e sq )|Up ∩Uq are both sections of S (Up ∩ Uq )
sp )|Up ∩Uq and (e
which are mapped under f|Up ∩Uq to t|Up ∩Uq . They are equal, because of the injectivity
proved in (1)!
It follows from the second sheaf property, that there exists a section s ∈ S (U ) with
s|Up = sep . By construction fU (s)|Up = t|Up for all p, so that we may conclude fU (s) = t
by using the first sheaf axiom.
This results motivates the following definition.
Definition 6.1.11. A sequence of morphisms of sheaves on X
f g
R −→ S −→ T
is called exact, if for all p ∈ X, the induced map on stalks

fp gp
Rp −→ Sp −→ Tp
is exact. In particular, a morphism S −→ T is called injective, if for all p ∈ X, the

induced map on stalks fp : Sp −→ Tp is injective. Similarly surjectivity is defined.
Remark 6.1.12. It is important to note that from the surjectivity of fp for all p it does
not follow that fU is surjective for all U in general. To give an example, consider the
following map of sheaves on X = C \ {0}:
∗
f : OX −→ OX , fU : g 7→ e2πig .
∗
Let U be an open subset of X, and OX (U ) = {h ∈ O(U ): h(p) 6= 0 for all p ∈ U }. Then
∗
fp is surjective for all p. Indeed, take hp ∈ OX,p , and a small simply connected open
neighborhood U of p on which h is a representative of hp . By continuity, we may assume
that U is so small that h is nonzero throughout U . From function theory, we know that
log(h) exists, and e2πi log(h) = h. Thus log(h)p maps to hp , showing that the map of
sheaves is surjective. However fp is not injective. With log(h) also log(h) + k for all k ∈ Z
maps to h. Let z be a coordinate on C . Then z ∈ O∗ (X) but z is not in the image of f ,
as log(z) is not a well defined holomorphic function on X \ {0}.
This example also shows that given a map f : G −→ H of sheaves the presheaf
defined by U 7→ H(U )/fU G (U ) is not necessarily a sheaf. The first sheaf axiom is
not satisfied: locally defined logarithms do not necessarily glue to a globally defined
logarithm. Thus we have problems in defining the cokernel of a map as sheaf. These
kinds of problems occur for several other constructions with sheaves as, for example, the
tensor product. This is the reason for the following construction, called sheafifying.
Theorem 6.1.13. Let X be a topological space and F be a presheaf on X. There exists
a sheaf F + and a morphism j : F −→ F + of presheaves with the following properties:
(1) jp : Fp −→ Fp+ is an isomorphism for all p.

(2) (F + ,j) is universal, that is, for any morphism ϕ : F −→ G of F to a sheaf G
there exists a unique morphism ψ : F + −→ G such that ϕ = ψ ◦ j.
Definition 6.1.14. The (uniquely) determined sheaf F + is called the sheaf associated
to the presheaf F . One says that F + is obtained from F by sheafifying.
Proof of Theorem 6.1.13. Let U be an open set and define F + (U ) to be the set of func-
tions f : U −→ ∪ Fx having the following two properties:
x∈U
(1) for each x ∈ U , f (x) ∈ Fx ,

(2) for each x ∈ U there is a neighborhood W ⊂ U of x and an element g ∈ F (W )
such that the germ gy of g in y ∈ W is equal to f (y) for all y ∈ W .
Now it is not difficult to verify, see Exercise 6.1.36, that F + together with the canonical
restriction maps is a sheaf. It is also not difficult to check that the canonical map j :
F −→ F + defined by jU (g)(y) = gy has the universal property mentioned, see Exercise
6.1.36.
Note that not all nonzero section of F (U ) give nonzero sections of F + (U ). Namely,
it might happen that all stalks are zero, but F (X) 6= 0. On the other hand, in the
∗
example f : OX −→ OX , fU (g) = e2πig the function z is in Im(f )(U ) as sheaf, but not
as presheaf. So F + (U ) might have more sections. This is because by definition sections
of F + (U ) are not sections of F (U ), but functions which are locally equal to sections of
F.
We are now prepared to give a lot of examples for sheaves.
Definition 6.1.15. Let X be a topological space, F , G be sheaves on X and f : F −→ G
a morphism.
6.1 Sheaves 225
(1) The kernel of f , Ker(f ) is the sheaf defined by Ker(f )(U ) = Ker(fU ).
(2) The cokernel of f , Coker(f
) is the sheaf associated to the presheaf defined by
U 7→ G (U )/fU F (U ) . A cokernel of a map of sheaves is also called a quotient
sheaf.
(3) The image of f , Im(f ) is the sheaf associated to the presheaf defined by U 7→
fU F (U ) .
(Note that Im(f ) is a subsheaf of G , Exercise 6.1.37.)
(4) Let A be a ring, the constant sheaf CA is the sheaf associated to the constant
presheaf defined by U 7→ A if U 6= ∅. In this case if U is the disjoint union of two
connected subsets, we have C (U ) = A ⊕ A. Sections of CA over U are now allowed
to differ on the two connected subsets.
The proof that the presheaf defined in (1) is already a sheaf will be given in Exercise
6.1.37.
We now relate (pre)sheaves which are defined on different topological spaces.
Definition 6.1.16. Let a continuous map f : X −→ Y between topological spaces be
given.
(1) Let S be a sheaf on X. We define a presheaf f∗ S on Y by putting for an open set

U ⊂Y
f∗ S (U ) := S f −1 (U ) ,
with the obvious restriction maps. (It can be proven without difficulty, see Exercise
6.1.30, that f∗ S is in fact a sheaf.)
(2) Let F be a sheaf on Y . f −1 F is the sheaf on X associated to the presheaf defined
by U 7→ lim F (V ).1
V ⊃f (U)
(3) If f is an injection and F a sheaf on Y , we define the restriction F|X of F to X

by f −1 F .
As a simple example, if f is the map which sends X to a point, then f∗ S can be

identified with a sheaf on a point, hence a ring. In this case we can identify f∗ S with
the global sections S (X) of X. On the other hand, if f : p −→ X is the injection of the
point p in the space X, then f −1 F is the sheaf on the point p, consisting of the stalk
Fp . More generally, if f : X −→ Y is the injection of a closed subspace, then f −1 G is a
sheaf whose stalks at p ∈ X are equal to Gp .
It is a direct and easy consequence of the definitions, that if one has a morphism
α : S −→ T of sheaves on X, and a continuous map f : X −→ Y , then one has
an induced map α∗ : f∗ S −→ f∗ T . One cannot hope in general, that for a surjective
morphism of sheaves α : S −→ T , the induced map α∗ : f∗ S −→ f∗ T is surjective
(look at the constant map f , that is, look at global sections). But, on the other hand,
we have the following theorem.
1 By definition lim F (V ) := ∪ F (V )/ ∼ where s ∼ s′ for s ∈ F (V ), s′ ∈ F (W ) if and only if
V ⊃f (U ) V ⊃f (U )
s|V ∩W = s′|V ∩W . Note that this is a generalization of the notion of germ, which corresponds to the
case that f is the injection of a point.
Theorem 6.1.17. Let X and Y be Hausdorff topological spaces and f : X −→ Y be a

finite continuous map. Consider an exact sequence of sheaves on X:
0 −→ S1 −→ S2 −→ S3 −→ 0.
Then the induced sequence of direct image sheaves
0 −→ f∗ S1 −→ f∗ S2 −→ f∗ S3 −→ 0
is also exact.
One phrases this by saying that for finite f , the functor f∗ is exact. This theorem
will play an important role in the proof of the Finite Mapping Theorem, see 6.3.5. The
theorem follows immediately from the fact that exactness means exactness of the stalks
and the following theorem.
Theorem 6.1.18. Let X,Y be Hausdorff spaces, and f : X −→ Y be a finite continuous
map. Let S be a sheaf on X. Let y ∈ Y , and x1 , . . . ,xt be the points in the fiber f −1 (y).
Then we have the equality of stalks:
t
M
(f∗ S )y = Sxi .
i=1
Proof. As X is Hausdorff, we can choose pairwise disjoint open neighborhoods U1′ , . . . ,Ut′
of x1 , . . . ,xt in X. As f is closed, there exists an open neighborhood V of y in Y , such
that f −1 (V ) ⊂ ∪ti=1 Ui′ , see Exercise 3.4.36.
f
x1 U1′ xt Ut′ y V
We define Ui := f −1 (V ) ∩ Ui′ . Then

(1) The U1 , . . . ,Ut are pairwise disjoint.
(2) f −1 (V ) = ∪ti=1 Ui .
Therefore,
f∗ S (V ) := S (f −1 (V )) = S (∪ti=1 Ui ) = ⊕ti=1 S (Ui ).
The last equality is due to the fact that the Ui are pairwise disjoint, and because of the
sheaf properties. Now use the definition of stalk to deduce the theorem.
There is a method to produce a lot of examples of sheaves, if we know them locally.
Proposition 6.1.19 (Gluing Sheaves). Let X be a topological space, X = ∪ Ui , Ui open
i∈I
∼
subsets in X. For every i let Fi be a sheaf on Ui and ϕij : Fi|Ui ∩Uj −→ Fj|Ui ∩Uj be
isomorphisms such that
(1) ϕii is the identity,

(2) ϕik = ϕjk ◦ ϕij on Ui ∩ Uj ∩ Uk .
6.1 Sheaves 227
∼
Then there exists a unique sheaf F on X and isomorphisms λi : F|Ui −→ Fi such that
for all i,j, ϕij ◦ λi = λj on Ui ∩ Uj .
In fact, for an open subset V of X we define
( )
Y
F (V ) := {si }i∈I ∈ Fi (Ui ∩ V ) : ϕij (si|Ui ∩Uj ∩V ) = sj|Ui ∩Uj ∩V .
i∈I
The proof that this defines a sheaf, and the uniqueness is left as Exercise 6.1.40.
Definition 6.1.20. A ringed space (X,OX ) is a topological space X together with a
sheaf of rings OX . Usually, if it is clear from the context what the sheaf of rings is, we
simply write X for the pair (X,OX ).
A sheaf F on X is called an OX –module if F (U ) is an OX (U )–module for all U ⊂ X
open, such that the restriction mapping F (U ) −→ F (V ) is OX (U )–linear for V ⊂ U .
(Note that F (V ) for V ⊂ U is also an OX (U )–module.)
An OX –module sheaf F which is a subsheaf of OX is called an ideal sheaf.
Examples 6.1.21 (Ringed Spaces).
(1) Let U be an open set of C n and OU be the sheaf of holomorphic functions on U .
(2) Let U be an open set in C n , f1 , . . . ,ft be holomorphic functions defined on U .
Consider X := V (f1 , . . . ,ft ). Let V ⊂ U be open. Consider the presheaf defined by
OX (V ∩ X) := O(V )/(f1 , . . . ,ft )O(V ) on X. This is easily checked to be a sheaf.
The ringed space constructed in this way is called a complex model space.
Consider X = {x ∈ U : f1 (x) = · · · = ft (x) = 0}. Consider the presheaf I (X) on
U defined by I (X)(W ) = {f ∈ OC n (W ) : f (x) = 0 for all x ∈ W ∩ X}. This is
easily checked to be a sheaf and is called the ideal sheaf of X.
(3) Let U ⊂ Rn be an open subset, and OU the sheaf of differentiable functions on U .
Similarly let CU be the sheaf of continuous functions on U .
(4) An affine variety X (with the Zariski topology) can also be seen as a ringed space
by taking as canonical presheaf the presheaf which assigns to each U ⊂ X the ring
g
OX (U ) = {f ∈ K(X) : ∀p ∈ U ∃g,h ∈ K[X] with f = and h(p) 6= 0}.
h
An element f in OX (U ) therefore can be viewed as a rational function f : U −→ K
which have a well-defined value for all p ∈ U .
(5) Spectrum of a ring: Let R be a ring (commutative with 1 as always). Then the
spectrum Spec(R) is defined as follows:
• As a set Spec(R) is the set of prime ideals p ⊂ R.
• One defines a topology on Spec(R) by defining the closed sets to be the sets
of type V (I) := {p ∈ Spec(R) : I ⊂ p}. This topology is called the Zariski
topology on Spec(R).

• We define a sheaf O of rings on Spec(R). Let U = D(f ) := Spec(R) \ V (f )
be an open set of Spec(R). Then O(D(f )) := Rf . We have to show that this
is independent of the chosen f . For this we refer to Exercise 6.1.33. As the
D(f ) form a basis for the topology by Exercise 6.1.34 this induces a presheaf
on Spec R. This presheaf is in fact a sheaf, see Exercise 6.1.34.
Definition 6.1.22. Let X be a ringed space and F , G be OX –module sheaves on X.
(1) HomX (F ,G ) is the sheaf defined by HomX (F ,G )(U ) = HomOX|U (F|U ,G|U ).
(2) F ⊗OX G is the sheaf associated to the presheaf defined by U 7→ F (U )⊗OX (U) G (U ).
(3) F ⊕ G is the sheaf defined by (F ⊕ G )(U ) = F (U ) ⊕ G (U ).
(4) Let F ,G ⊂ H, H an OX –module sheaf. Then F ∩ G is the sheaf defined by
(F ∩ G )(U ) = F (U ) ∩ G (U ). F + G is the sheaf associated to the presheaf defined
by U 7→ F (U ) + G (U ). F : G is the sheaf associated to the presheaf U 7→ F (U ) :
G (U ).
annihilator sheaf of F , is the sheaf associated to the presheaf U 7→

(5) Ann(F ), the
Ann F (U ) = {f ∈ OX (U ) : f · F (U ) = 0}.
The proof that the presheaves defined in (1), (3) and (4) are already sheaves is left
as Exercise 6.1.38.
Note that in all the examples considered above the stalk at each point is a local ring.
We put this phenomenon in a definition.
Definition 6.1.23. A locally ringed space (X,OX ) is a ringed space, such that for all
points p ∈ X the stalk OX,p is a local ring. A morphism between locally ringed spaces is
a pair: (f,f ∗ ) : (X,OX ) → (Y,OY ), where f : X → Y is a continuous map, and f ∗ is a
morphism of sheaves of OY −modules:
f ∗ : OY → f∗ OX
such that the induced map on stalks:
fp∗ : OY,f (p) → OX,p
sends the maximal ideal of OY,f (p) into the maximal ideal of OX,p . Usually, if there is no
danger of confusion, we just write f for the pair (f,f ∗ ).
Examples 6.1.24.
(1) The examples of Examples 6.1.21.

(2) The projective space PnC , the set of lines through the origin in C n+1 . As a set we
have PnC := {c ∈ C n+1 , c 6= 0}/ ∼ and c ∼ c′ if there exists a λ ∈ C such that
λc = c′ . The equivalence class of c = (c0 , . . . ,cn ) we write as (c0 : c1 : · · · : cn ).
We consider PnC with the quotient topology of the ordinary topology of C n+1 .
Let Ui = {c = (c0 : · · · : cn ) ∈ PnC , ci 6= 0}. Then Ui is an open subset
n
of PnC and ∪ Ui = PnC . We have homeomorphisms ϕi : Ui −→ C n defined by
i=0

ϕi (c0 : · · · : cn ) = cc0i , cc1i , . . . , ci−1 ci+1 cn
ci , ci , . . . , ci .
The map ϕj ◦ ϕ−1

i : ϕi (Ui ∩ Uj ) −→ ϕj (Ui ∩ Uj ) maps (z1 , . . . ,zn ) to
z1 zj−1 zj+1
zj , . . . , zj , zj , . . . , z1j , . . . , zznj and is obviously holomorphic.
OPn is the sheaf obtained by gluing the sheaves OC n on Ui .
6.1 Sheaves 229
The second example can be generalized as follows:

Proposition 6.1.25. Let X be a Hausdorff space and X = ∪i∈I Ui for open sets Ui .
Let ϕi : Ui −→ Wi be homeomorphisms to domains Wi in C n such that for all i,j and
Ui ∩ Uj 6= ∅
ϕj ◦ ϕ−1
i : Wi ∩ ϕi (Uj ) −→ Wj ∩ ϕj (Ui )
is holomorphic. Then X is a locally ringed space.
Proof. the structure sheaf OX is defined to be the gluing of the sheaves OUi = ϕ∗i OWi
(Proposition 6.1.19).
Definition 6.1.26. Let X be a Hausdorff space.
(1) Let (X,OX ) be a locally ringed space. (X,OX ) is called a complex manifold if every
x ∈ X has a neighborhood U such that (U,OX|U ) is isomorphic to (V,OV ), V ⊂ C n
a domain.
(2) (X,OX ) is called an analytic space if every x ∈ X has a neighborhood U such that
(U,OX|U ) is isomorphic to (V,OV ), V an analytic subset of an open set W in some

C n , and OV = OW /I (V ) |V .
(3) (X,OX ) is called a complex space if if every x ∈ X has a neighborhood U such

that (U,OX|U ) is isomorphic to (V,OV ), where (V,OV ) is a complex model space.
This means
that V is an analytic subset of an open set W in √ some C n and OV =
OW /J |V , J ⊂ OW an ideal sheaf such that for all x ∈ V , Jx = I (V )x . One
says that (Y,OY ) is a closed complex subspace of (X,OX ), if there exists a map of
locally ringed spaces (i,i∗ ) : (Y,OY ) −→ (X,OX ) such that i is injective, and i∗ is
surjective.
For the case (X,OX ) = (U,OU ) for U ⊂ C n open, we obtain that a closed complex
subspace is a complex model space.
Exercises
6.1.27. Consider the following presheaf on R. Let U ⊂ X be open and
B(U ) := {f : U −→ R : f bounded},
and ρU V the ordinary restrictions. Show that B is not a sheaf.
6.1.28. Let S be a sheaf of rings on a topological space X. Show that for each point x ∈ X
the stalk Sx is in a natural way a ring again.
6.1.29. Prove Lemma 6.1.9.
6.1.30. Prove that for f : X −→ Y continuous, S a sheaf on X, the direct image f∗ S is also
a sheaf.
6.1.31. Let R be a Noetherian ring. Prove that the topology on Spec(R) is indeed a topology
by proving the following statements:
` ´ ` ´
(1) V (0) = Spec(R); V (1) = ∅,
(2) V (I · J) = V (I) ∪ V (J),
P
(3) V ( j Ij ) = ∩j V (Ij ).
6.1.32. For a subset A of Spec(R) let I (A) = ∩ p. Prove the analog of Hilbert’s Nullstellen-
p∈A
√
satz: I = I (V (I)).
(Hint: Use primary decomposition of an ideal).
6.1.33. Let R be a ring and f,g ∈ R such that D(f ) = D(g) (equivalently V (f ) = V (g)). Prove
that Rf = Rg .
6.1.34. Let R be a ring. Show that the open sets D(f ) form a basis for the topology of Spec(R).
Show that the presheaf O on Spec(R), defined in 6.1.21, is in fact a sheaf.
6.1.35. Let f : X −→ Y be a continuous map, A ⊂ Y a closed subset, S a sheaf on X. Prove

that ` ´
(f∗ S )|A = f|f −1 (A) ∗ S|f −1 (A) .
6.1.36. Prove that F + defined in the proof of Theorem 6.1.13 is a sheaf and the canonical map
j : F −→ F + has the universal property.
6.1.37. Let f : F −→ G be a morphism of sheaves.

(1) Prove that there is a canonical injection i : Im(f ) −→ G .
(2) Prove that the presheaf defined by U 7→ Ker(fU ) is already a sheaf.
6.1.38. Let X be a ringed space and F ,G be OX –module sheaves on X.

(1) Suppose that both F and G are subsheaves of H . Prove that the presheaf defined by
U 7→ F (U ) ∩ G (U ) is a sheaf.
(2) Prove that the presheaf defined by U 7→ F (U ) ⊕ G (U ) is a sheaf.
(3) Prove that the presheaf defined by U 7→ HomOX|U (F|U ,G|U ) is a sheaf.
6.1.39. Let f : X −→ Y be a continuous map between the topological spaces X,Y , F be a

sheaf on X and G be a sheaf on Y .
(1) Prove that Gf (x) = (f −1 G )x .
(2) Show that there are canonical maps λ : f −1 f∗ F −→ F and ρ : G −→ f∗ f −1 G .
(3) Prove that Hom(f −1 G ,F ) = Hom(G ,f∗ F ).
6.1.40. Prove Proposition 6.1.19.
6.1.41. One can view X = C 2 \ {0} as a locally ringed space (X,OX ) just as one can see an
affine variety as a locally ringed space.
(1) Show that OX (X) = C [x,y].
(2) Show that X, as locally ringed space, is not isomorphic to an affine variety.
6.2 Fundamental Properties of Coherent Sheaves
In this section, unless otherwise stated explicitly, X will be a complex space, with struc-
ture sheaf OX , mX ⊂ OX the maximal ideal. S ,T , etc., will be sheaves of OX modules.
If (X,OX ) is fixed we shall often use O and m instead of OX and mX .
What is a coherent sheaf? Well, before defining it, let us mention some fundamental
properties of coherent sheaves, give some examples, and deduce some properties which
show that coherence is indeed a very useful property a sheaf can have. The properties we
shall mention will also justify the name “coherence”.
6.2 Fundamental Properties of Coherent Sheaves 231
Theorem 6.2.1 (Meta-Theorem for coherent sheaves). Let S ,T be coherent sheaves

of OX –modules then every reasonable operation with S ,T (as taking Hom,⊗, finitely
generated subsheaves . . .) produces again a coherent sheaf.
Theorem 6.2.2. Let S ,T be coherent sheaves of OX - modules. Let α : S −→ T be a
map of OX -modules. Suppose that at some point x ∈ X the maps of stalks
αx : Sx −→ Tx
is an isomorphism. Then there exists an open neighborhood U of x, such that for all
points p ∈ U the maps of stalks
αp : Sp −→ Tp
are isomorphisms, that is, the sheaves S and T are isomorphic, when restricted to U .
The Meta Theorem will be proved in Lemma 6.2.11, Corollary 6.2.12, Lemma 6.2.13
and Lemma 6.2.14. Theorem 6.2.2 is an immediate consequence of 6.2.12 (2) and 6.2.9.
So, to show that two coherent sheaves S and T are isomorphic on some open set
containing a point x, one has to find a map of sheaves α : S −→ T , and then show that
αx is an isomorphism. The last condition usually is pure algebra! As a sheaf is determined
by its stalks, the above property gives an easy method to give examples of sheaves which
are not coherent.
Example 6.2.3. Consider the ringed space consisting of the line C and the sheaf OC
of holomorphic functions.
For S we take the zero-sheaf (this sheaf will be coherent). To define T , consider a
sequence of points p1 ,p2 , . . . , converging to 0. Let T be the subsheaf of OC consisting of
all holomorphic functions vanishing at each point pi . We let α be the zero map. By the
identity Theorem for holomorphic functions in one variable, every holomorphic function
defined on some open neighborhood of U of 0 and vanishing at all points pi which lie
in U actually is zero. This shows that the stalk T0 is zero, hence α0 is an isomorphism.
However, T is not the zero sheaf (there exist nonzero sections over any open set not
containing zero). Therefore T is not a coherent sheaf. In fact, for all points p 6= pi and
p 6= 0, we have Tp = Op . For the pi we have Tpi = mpi , the germs of holomorphic
functions vanishing at the pi . Notice that T is an ideal sheaf but not finitely generated.
To prove this we show that for every domain U ⊂ C containing 0, every OC –module
homomorphism ϕ : OC |U −→ T|U has to be the zero map. Namely, because U contains
0 it contains almost all of the pi , then T (U ) = {0} because of the Identity Theorem
(Remark 3.1.10).
Now let W ⊂ U be an open subset and consider the diagram
ϕU
OC (U ) / T (U ) = {0}
ρU W ρ′U W
ϕW

OC (W ) / T (W ).
As ϕU (1) = 0 we have ϕW (1) = ϕW ρUW (1) = ρ′UW ϕU (1) = 0. This implies that
ϕW (g) = g · ϕW (1) = 0 for all g ∈ OC (W ). Hence ϕ = 0.
So for coherent sheaves such an ugly thing does not happen. The problem with the
theory of coherent sheaves (in complex analysis, not in algebra, see Exercise 6.2.19.)
is that one has to work hard to produce examples. It follows relatively easy from the
definition that sheaves behave well under most algebraic operations on sheaves (the Meta–
Theorem for coherent sheaves): kernels, cokernels of maps between coherent sheaves are
coherent. Direct sums of coherent sheaves are coherent. Finitely generated subsheaves
of coherent sheaves are coherent. Taking Hom, tensor product etc. preserves coherence.
These things will be discussed in this section. It is much harder to prove the four basic
coherence Theorems:
(1) The structure sheaf On is coherent. More generally, the structure sheaf of any
complex space is coherent.
(2) Coherence is preserved under finite mappings of complex spaces.
(3) The ideal sheaf of any analytic space is coherent.
(4) The normalization sheaf of any analytic space is coherent.
These Theorems will be proved in Section 6.3. Assuming these, we give some applications.
Corollary 6.2.4. Let J ⊂ OX be a coherent ideal sheaf then V (J) := {x ∈ X : Jx ⊂
mX,x } is an analytic space.
Proof. Let x ∈ X and assume that Jx ⊂ mX,x . Since OX,x is Noetherian, we have Jx =
(f1 , . . . ,fm ). Let U ⊂ X be an open set such that f1 , . . . ,fm ∈ OX (U ). Let J ⊂ OX|U
be the ideal sheaf generated by f1 , . . . ,fm . This is a coherent sheaf (Theorem 6.2.1) and
we have a canonical map α : J −→ J such that αx : J x −→ Jx is the identity. This
implies that J |W = JW for an open set W ⊂ U , x ∈ W (Theorem 6.2.2). Therefore,
V (J) ∩ W = {x ∈ W : f1 (x) = · · · = fm (x) = 0}.
Corollary 6.2.5. Let M be a coherent OX –module. Then the support of M :
supp(M ) := {x ∈ X : Mx 6= 0} is an analytic space.
Proof. Along with M , the annihilator is coherent (cf. Theorem 6.2.1 or Lemma 6.2.14),
and the annihilator is an ideal sheaf. Now the support
of M is given by the zero set of
the annihilator of M : Supp(M ) = V Ann(M ) .
Indeed, let x ∈
/ Supp(M ). Then Mx = 0, so that Ann(Mx ) = Ann(M )x = OX,x .
Therefore, x does not lie in the zero set of Ann(M ). Suppose on the other hand that
x∈ / V Ann(M ) . Then there exists an f ∈ Ann(Mx ) with f (x) 6= 0 and fx m = 0 for
all m ∈ Mx . As f (x) 6= 0 it follows that the germ fx is a unit, so that m = 0 for all
m ∈ Mx . Hence Mx = 0, and thus x ∈ / Supp(M ). Hence the support of M is an analytic
subset.
Corollary 6.2.6. Let F be a coherent sheaf and supp(F ) = {y : Fy 6= 0} = {x} then
(1) F (X) = Fx ;
(2) dimC Fx < ∞.
Proof. The first property holds without using the coherence. Consider the map F (X) −→
Fx . If for some s ∈ F (X) sx = 0 then there exists a neighborhood U of x such that
s|U = 0 because sy = 0 for all y 6= x. Let W := X \ {x} then X = U ∪ W and s|W = 0
implies s = 0. To prove the surjectivity let f ∈ Fx and choose a neighborhood U of x
such that f ∈ F (U ). On W := X \ {x} we have f|U∩W = 0. So the section 0 on W and
f on U glue to a section on X.
To prove (2) let I = Ann(F ) and assume that, in a neighborhood U of x, I is
generated by f1 , . . . ,fr . In OX,x we consider the primary decomposition of Ix = q1 ∩· · ·∩qs
and claim that s = 1 and q1 is mX,x –primary. If this is true, then mcX,x annihilates
Fx for a suitable c and, therefore, Fx is a finitely generated OX,x /mcX,x –module. But
dimC OX,x /mcX,x < ∞ implies dimC Fx < ∞.
√
To prove the claim assume Ix ⊂ q primary but q 6= mX,x . In this case V (I) ⊃
V (q) % {x}. This is a contradiction to the fact that Iy = OX,y for all y 6= x.
Definition 6.2.7.
(1) The sheaf S is said to be of finite type if for every point x ∈ X there is an open
neighborhood U of X and a surjective morphism of sheaves:
q
OX|U −→ S|U −→ 0.
(2) The sheaf S is said to be of relation finite type if for any open set U and any
q
morphism of sheaves: α : OX|U −→ S|U the kernel Ker(α) is a sheaf of finite type.
(3) A sheaf S of OX modules is called coherent, if S is of finite type, and of relation

finite type.
Lemma 6.2.8.
(1) A quotient sheaf of a sheaf of finite type, is a sheaf of finite type.

(2) A subsheaf of a sheaf of relation finite type, is a sheaf of relation finite type.
(3) A finitely generated subsheaf of a coherent sheaf is coherent.
Proof. Exercise 6.2.18.

Lemma 6.2.9. Let s1 , . . . ,sq be holomorphic functions on an open set U containing x.
Let S be a sheaf of finite type, and suppose we have a surjection of stalks:
(s1x ,...,sqx )
Oxq −−−−−−−→ Sx −→ 0.
Then on some open neighborhood V ⊂ U of x, we have a surjection:
q (s1 ,...,sq )
OX|V −−−−−−→ S|V −→ 0.
In particular, if a sheaf of finite type has zero stalk in x, then the sheaf is identically zero
in some open neighborhood V of x.
Proof. Because S is of finite type there exist an open neighborhood W of x, and sections
t1 , . . . ,ts generating S|V . But, because the
Psix generate the stalk at x, we can find germs
of holomorphic functions such that tix = j fijx sjx . Take an open neighborhood V ⊂ W
of x such that all these functions converge on V . Because the t′ s generate S , it follows
that the si also generate S on V .
The following Lemma is the basic Lemma, from which almost everything in this
section follows.
Proposition 6.2.10. Consider an exact sequence of sheaves on X:
ϕ1 ϕ2 ϕ3 ϕ4
S1 −→ S2 −→ S3 −→ S4 −→ S5 .
If Si is coherent for i = 1,2,4,5, then S3 is coherent.2

Proof. As S1 is of finite type, we may assume S1 = Oq for some q. Consider a map
β : Op −→ S3 . We have to prove that Ker(β) is finitely generated. As S4 is of relation
finite type, the map ϕ3 β : Op −→ S4 has a finitely generated kernel. This means that we
ψ ϕ3 β
have an exact sequence Os −→ Op −→ S 4. Obviously Ker(β) ⊂ Ker(ϕ3 β) = Im(psi).
By construction ϕ3 (β ◦ ψ) = 0, that is Im(β ◦ ψ) ⊂ Ker(ϕ3 ) which by exactness is equal
to Im(ϕ2 ). Therefore, we can find for every generator ei of Os an element ω(ei ) ∈ S2
with ω(ei ) = β ◦ ψ(ei ). Thus we get a map ω making a commutative diagram
Os D
DD ψ
DD
DD
D!
ω Op C
CC
CC
β CC
C
ϕ1 ϕ2 ϕ3 !
Oq / S2 / S3 / S4
Consider the map ϕ1 ⊕ ωOq ⊕ Os −→ S2 . As S2 is of relation finite type, the kernel

K is finitely generated. Let π : Oq ⊕ Os be the projection on the second summand. As
images of finitely generated sheaves are finitely generated, it follows that ψ ◦ π(K) is
finitely generated. So we are done if we show that Ker(β) = ψ ◦ π(K).
(1) We first show ψ ◦ π(K) ⊂ Ker(β). Hence suppose (a1 ,a2 ) ∈ K, that is ϕ1 (a1 ) =
−ω(a2 ). Then
0 = ϕ2 ϕ1 (a1 ) = −ϕ2 ω(a2 ) = −βψ(a2 ).
The first equality follows from exactness, the third because the diagram is commu-
tative. Hence ψ(a2 ) ∈ Ker(β). As a2 = π(a1 ,a2 ) the claim follows.
(2) We now show the converse inclusion. Let a ∈ Ker(β). Then certainly a ∈ Ker(ϕ3 β) =
Im(ψ). Hence there exists a b ∈ Os with ψ(b) = a. Then
0 = β(a) = βψ(b) = ϕ2 ω(b).
Thus ω(b) ∈ Ker(ϕ2 ) = Im(ϕ1 ). Hence there exists a c ∈ Os with ϕ1 (c) = ω(b).
But then (−c,b) ∈ K , and thus a = ψπ(−c,b) ∈ ψπ(K).
2 The proof shows that we only need that S2 , S3 are coherent, that S1 is of finite type, and that S5
is of relation finite typ.
Lemma 6.2.11. Consider a short exact sequence of sheaves of OX -modules:

φ ψ
0 −→ S1 −→ S2 −→ S3 −→ 0.
Then all sheaves S1 ,S2 ,S3 are coherent if two of them are coherent.
Corollary 6.2.12.
(1) The direct sum of finitely many coherent sheaves is coherent.
(2) Let α : S −→ T be a morphism of coherent sheaves. Then the sheaves Im(α), Ker(α)
and Coker(α) are coherent.
(3) Let S and T be coherent subsheaves of a coherent sheaf U . Then the sum S + T
and the intersection S ∩ T are coherent sheaves.
(4) Suppose that the structure sheaf OX is coherent. Then S is coherent, if and only
if for every x ∈ X there exists a small neighborhood U of x and an exact sequence:
q p
OX|U −→ OX|U −→ S|U −→ 0.
(5) If S and T are coherent, then S ⊗ T is coherent.
Proof. Exercise 6.2.17

It is somewhat more tricky to show that the sheaf Hom of coherent sheaves is
coherent:
Lemma 6.2.13. Let S and T be coherent sheaves.
(1) The natural map HomO (S ,T )x −→ HomOx (Sx ,Tx ) is bijective.

(2) HomO (S ,T ) is coherent.
Proof.
(1) The injectivity is easy: let ϕ : S|U −→ T|U be a map and assume that Ker(ϕx ) =
Sx then Theorem 6.2.2 implies that Ker(ϕ)|V = S|V for an open neighborhood of
x. This implies that ϕ is the zero map in HomO (S ,T )x . To show that the map is
surjective, we have to lift any map φx between germs to some open neighborhood
p
of x. This is easy for the case that S = OX , because we then just have to take
a lift in T of the images of the basis elements. Because S is finitely generated,
p
we can find a map ψ : O|U −→ S|U . By the remark made above, we can also find
p
a lift of the map ψx φx to χ : O|U −→ T|U , after possibly shrinking U . We have
Ker ψx ⊂ Ker χx . Because S is of relation finite type, Ker ψ is finitely generated,
and hence Ker ψU ⊂ Ker χU after possibly shrinking U . Therefore, χ induces the
p
desired map χ : S|U = O|U / Ker ψU −→ T|U . Note that in this proof we only made
use of the coherence of S .
(2) It follows directly from the definition of coherence, that if S is coherent we (locally)
have an exact sequence Oq −→ Op −→ S −→ 0. We want to show that
0 −→ HomO (S ,T ) −→ HomO (Op ,T ) −→ HomO (Oq ,T ) (∗)
is exact. This suffices, because HomO (Op ,T ) is isomorphic to T p , which is coher-

ent by assumption. As a kernel of a morphism between coherent sheaves is coherent
again, it then follows that HomO (S ,T ) is coherent. To prove the exactness of (∗),
look at the following commutative diagram:
0 // HomO (S ,T )x // HomO (Op ,T )x // HomO (Oq ,T )x

0 // HomOx (Sx ,Tx ) // HomOx (Oxp ,Tx ) // HomOx (Oxq ,Tx ).
The sequence at the bottom is exact, because that is just a statement about rings
and modules. The left vertical arrow is an isomorphism by the first part. Because,
as already remarked, HomO (Op ,T ) is isomorphic to T p , the fact that the other
vertical arrows are isomorphisms is trivial. Therefore, the upper row is exact. As
this is true for all x ∈ X, the claim follows.
The following now follows easily:

Lemma 6.2.14. Suppose that the structure sheaf is coherent. Let S be coherent, and
suppose that P and Q are coherent submodules of S . Then the sheaves Ann(S ) and
the transporter sheaf (Q : P) are coherent.
Proof. The annihilator sheaf is the kernel of the morphism
O −→ H omO (S ,S )
by sending an element to multiplication with this element. Therefore, by the previous

result, the Annihilator sheaf is coherent. The other result is Exercise 6.2.21.
The final basic property of coherent sheaves is the following extension principle:
Theorem 6.2.15 (Extension Principle). Let (Y,OY ) be a closed subspace of the complex
space (X,OX ), i : Y −→ X the inclusion. Suppose that the sheaf OX is coherent. Let S
be a sheaf of OY –modules. Then S is coherent, if and only if i∗ S is coherent.
In particular,it follows that OY is coherent.
Proof.
Step 1. We first prove that S is of finite type if and only if i∗ S is of finite type. Let
i∗ OY = OX /I , where I is a finitely generated (and hence coherent) OX –ideal sheaf.
Suppose that S is a finitely generated sheaf of OY –modules. Then we have a surjective
map OVp −→ S|V for some p ∈ N, and an open set V in Y . Then V = Y ∩ U , for some
open set U in X. As OU ։ i∗ OV = OU /I|U is surjective, it follows that the induced
p
map OU −→ i∗ S is surjective.
p
On the other hand, if OU −→ i∗ S is surjective, the fact that i∗ S is an i∗ OV =
OU /I|U –module implies that the map induces a surjective map OVp −→ S .
Step 2. Suppose that i∗ S is OX –coherent. By Step 1, it remains to show that S is of

relation finite type. Therefore, consider an open set V in Y , and a map of OV –modules
ϕ : OVp −→ S|V .
Now V = U ∩Y for some open set U in X. Let π : OU ։ i∗ OV be the canonical surjection.

p
The composed map ϕ ◦ π : OU −→ (i∗ S )|U of OU –modules has a finitely generated

kernel, because we assumed that i∗ S is OX –coherent. But as i∗ Ker(ϕ) = π Ker(ϕ ◦ π)
it follows that Ker(ϕ) is finitely generated.
So, in particular, as i∗ OY as a cokernel of a coherent sheaf by a finitely generated
ideal sheaf is coherent, it follows that OY itself is coherent.
Step 3. Now suppose that S is OY –coherent. It follows that we can write S locally as
a cokernel. Thus for all points p ∈ Y there exists an open neighborhood V of p in Y and
an exact sequence
OVq −→ OVp −→ S −→ 0.
Again we write V = U ∩Y for an open set U in X. Then i∗ OV is OU –coherent, and there-
fore i∗ OVq and i∗ OVp are OU –coherent. It follows that i∗ S , as cokernel of two coherent
OU –modules, itself is OU –coherent.
Lemma 6.2.16. Suppose the structure sheaf OX is coherent and let S be a coherent
OX –module. Then the set A = {x ∈ X : Sx is not a free OX,x -module} is a proper
analytic subset of X.
Proof. Let x ∈ X and U be an open neighborhood such that S|U has a representation
q ϕ P
OX|U −→ OX|U −→ S|U −→ 0.
This is possible because of 6.2.12 (4).

For y ∈ U , Sy is a free OX,y –module of rank p − r if and only if for the Fitting ideals
holds Ir (ϕy ) = OX,y and Ir+1 (ϕy ) = (0), (cf. Proposition 1.3.8). We may assume
that
on U , Ir+1 (ϕ) = (0) and Ir (ϕy ) 6= (0) for some y ∈ U , then A ∩ U = V Ir (ϕ) $ U .
Exercises
6.2.18. Prove Lemma 6.2.8.
6.2.19. Let X be an affine variety, and M be a finitely generated K[X]–module. Show that the
` ´
f, defined by M
associated sheaf M f D(f ) = Mf , is a coherent OX –module.
6.2.20. Assume that OX is coherent. Prove that the support of a coherent sheaf S is an analytic
set, by using the exact sequence
q M p
O|U −→ O|U −→ S|U −→ 0
and looking at minors of M .
6.2.21. Prove the second assertion of 6.2.14.

6.3 The Four Basic Coherence Theorems
In this section we prove the four basic Coherence Theorems in complex analysis.
Theorem 6.3.1 (Oka). The sheaf OC n is coherent. More generally, for any complex
space X, the sheaf OX is coherent.
Proof.
Step 1. A sheaf S is coherent, if for every point there exists an open neighborhood of this
point such that S|U is coherent. Thus to show that OX is coherent, it suffices to show this
for a complex model space (D,OD ). Such a complex model space injects, by definition,
into an open subset of C n . By the extension principle 6.2.15 it suffices therefore to prove
the coherence of OC n . This will be done by induction on n. The case n = 0 is trivial, see
the second example of ??.
Step 2. For a short notation we write On instead of OC n and, since there is no possible
confusion, OU instead of OC n |U for an open subset U ⊂ C n . So assume that On−1
is coherent. The statement is local, so we look at a neighborhood of a point p ∈ C n .
Without loss of generality we may assume p = 0. As the sheaf On is obviously of finite
type, we only have to show that it is of relation finite type. Therefore, we have to show
that for any map of sheaves:
F =(f1 ,...,fs )
Ons −−−−−−−−→ On
the kernel Ker(F ) is ofP finite type. Finding elements in Ker(F ) means finding relations,
s
that is, r1 , . . . ,rs with i=1 ri fi = 0. After a linear change of coordinates (if necessary),
we may assume that f1 , . . . ,fs are regular in xn . As it is easy to relate the relations
between f1 , . . . ,fs and u1 f1 , . . . ,us fs for units ui , we may, by applying the Weierstraß
Preparation Theorem, assume that the fi are Weierstraß polynomials in xn . Let di be
the degree of fi in xn , and put d = max{di }. We may assume, without loss of generality,
that the degree of fs is d.
For any k we denote by On−1 [xn ]≤k the free On−1 –module of rank k +1 of polynomi-
als in xn of degree less than or equal to k. By induction, On−1 is coherent and, therefore,
the free On−1 –module On−1 [xn ]≤k is also coherent for all k by 6.2.12. By definition of
coherence, the kernel of the following map of On−1 –modules:
(On−1 [xn ]≤d )s −→ On−1 [xn ]≤2d

(a1 , . . . ,as ) 7−→ a1 f1 + . . . + as fs .
is finitely generated. This exactly means that Ker(F ) ∩ (On−1 [xn ]≤d )s is a finitely gen-
erated On−1 –module.
Step 3. It remains to show that in some small open neighborhood of 0, the sheaf of
s
relations Ker(F ) is as an On –module generated by Ker(F ) ∩ (On−1 [xn ]≤d ) . This means
s
the following: Take generators h1 , . . . ,hl of Ker(F ) ∩ On−1 [xn ]≤d . Take a small open
neighborhood U of 0 on which the h1 , . . . ,hl and the f1 , . . . ,fs converge. Then the germs
h1,p , . . . hl,p of the hi in p generate the kernel of the map Fp for all p ∈ U . To prove this,
first note that elements of the following type (so-called Koszul relations):
(−fs ,0, . . . ,0,f1 ), . . . , (0, . . . ,0, − fs ,fs−1 )

6.3 The Four Basic Coherence Theorems 239
are in Ker(F ) ∩ On−1 [xn ]s≤d . Indeed, they are in Ker(F ), and also of degree less than or
equal to d in xn . We take an arbitrary point p = (p1 , . . . ,pn ) ∈ U . At the point p the
function fs,p is regular of order d′ ≤ d in xn − pn . Consider any relation between the fi
at the point p:
a1 f1,p + . . . + as fs,p = 0.
For i = 1, . . . ,s − 1 we perform Weierstraß division with respect to fs at p:
ai = qi fs,p + ci , deg(ci ) < d′ ≤ d.

Ps−1
Moreover, if we define cs := as + i=1 fi,p qi we see that
(a1 , . . . ,as ) = (c1 , . . . ,cs ) − (−fs,p ,0, . . . ,0,f1,p )q1 − . . . − (0, . . . ,0, − fs,p ,fs−1,p )qs−1 .
From this it follows that c1 f1,p + . . . + cs fs,p = 0 and, therefore, we expressed our relation
as an On –linear combination of elements of Ker(F ) ∩ On−1 [xn ]s≤d , as soon as we have
proved that cs is a polynomial of P degree less than or equal to d. This is easy, because we
s−1
have just proved that cs fs,p = − i=1 ci fi,p . The right-hand side is a polynomial in xn
′
of degree less than d + d . As we have assumed that the degree of fs is d, it follows that
cs has degree at most d′ ≤ d. This concludes the proof of Oka’s Theorem.
Theorem 6.3.2 (Oka-Cartan). Let X ⊂ U ⊂ C n be an analytic set. Then the ideal
sheaf I := I (X) of X is coherent.
Remark 6.3.3. Let the ideal I = (f1 , . . . ,fs ) ⊂ C {x1 , . . . ,xn } be a radical ideal , and
let (X, 0) = V (I), 0 . Take small enough representatives. Then it is a tautology that for
all a ∈ X
V (f1,a , . . . fs,a ) = (X,a)
but it is not clear that the ideal of (X,a) is generated by the fi,a . To put it in another
way, it is not clear a priori that the ideal (f1,a , . . . fs,a ) is a radical ideal in OX,a . That
this is indeed the case is the content of the Coherence Theorem of Cartan and Oka.
Step 1. Consider a point x ∈ X. Without loss of generality, we may assume that x =
0. Consider the ideal I0 ⊂ On,0 = C {x1 , . . . ,xn } of the germ (X, 0). As this ring is
Noetherian, this ideal is finitely generated. We take finitely many generators:
I0 = (f1 , . . . ,fs )
and choose a small open neighborhood U of 0 such that all fi converge on U . As a

finitely generated subsheaf of the coherent sheaf On the sheaf generated by (f1 , . . . ,fs )
is coherent by Lemma 6.2.8. We have a map of sheaves (f1 , . . . ,fs ) −→ I . This is an
isomorphism, if it induces an isomorphism on all stalks. So we only have to show
(6.1) Ip = (f1,p , . . . ,fs,p )
for all points p in a small open neighborhood V of 0.

Step 2. We reduce to the case that the germ (X, 0) is irreducible. Indeed, suppose (X, 0) =
(X1 , 0) ∪ · · · ∪ (Xr , 0). Take representatives of (Xi , 0), so that X = X1 ∪ . . . ∪ Xr in a
small open set U . For the ideal sheaves on U we have
I (X) = I (X1 ) ∩ . . . ∩ I (Xr ).

Therefore, if we show that the I (Xj ) are coherent it follows that I (X) is coherent by
by 6.2.12.
Step 3. Thus we may suppose that (X, 0) is irreducible. By Remark 3.4.16, as a conse-
quence of the Local Parametrization Theorem, there exists a representative X of (X, 0)
and a hypersurface D = V (∆) ⊂ X such that for p ∈ X \ D, Ip = (f1 , . . . ,fs )p . Hence
(6.1) holds for all points p outside a proper analytic subset. We now consider the following
ideal sheaf on U :
J := (f1 , . . . ,fs ) : (∆) ⊂ On .
This is a coherent sheaf. Indeed, since (f1 , . . . ,fs ) is a finitely generated subsheaf of the
coherent sheaf On , it is coherent by 6.2.8. Hence J is coherent by 6.2.14. Therefore, we
can take finitely many generators g1 , . . . ,gt for J . Looking at the stalk at zero, these
elements have the property that ∆0 gi,0 ∈ (f1 , . . . ,fs ) = I0 . But this ideal was supposed
to be prime, hence gi,0 ∈ (f1 , . . . ,fs ) ⊂ On,0 . By taking U smaller, we may assume
that gi,p ∈ (f1,p , . . . ,fs,p ) for all p ∈ U . Hence we reduced to the case that we have the
following equality of ideal sheaves:
(6.2) (f1 , . . . ,fs ) : (∆) = (f1 , . . . ,fs ).
Step 4. Take a point p in U and an element g ∈ Ip , the stalk of the ideal sheaf of X at
p. Take a representative g vanishing on V ∩ X, for a (very small) neighborhood V ⊂ U
of p. We have to show g ∈ (f1,p , . . . ,fs,p ).
p
V
On V , the sheaf
S := (f1 , . . . ,fs ) : (g)
is coherent by 6.3.1, 6.2.8 and 6.2.14. Hence, S has finitely many generators, say h1 , . . . ,hu ,
for holomorphic functions hi defined on V . Since (6.1) was true for all points p ∈ X \ D,
it follows that gp ∈ (f1,p , . . . ,fs,p ) for all p ∈ V \ D, so that we deduce Sp = On,p for all
p ∈ V \ D. Thus, the zero set V (h1 , . . . ,hu ) ⊂ D ∩ V . As ∆ also vanishes on D, it follows
from the Nullstellensatz that there exists an r ≥ 0 such that ∆r ∈ (h1,p , . . . ,hu,p ) for all
p ∈ D ∩ V . From the defining properties of the hi it follows that ∆r g ∈ (f1,p , . . . ,fs,p ).
Let r ≥ 0 be minimal with this property. To prove the theorem, we have to show that
r = 0. Suppose the converse, that is, r > 0. But then we have ∆r−1 g ∈ (f1,p , . . . ,fs,p ) :
(∆) = (f1,p , . . . ,fs,p ), by (6.2), giving a contradiction.
Corollary 6.3.4. Let X ⊂ C n be an analytic space. The set of singular points Sing(X)
of X is an analytic subspace of X.
6.3 The Four Basic Coherence Theorems 241
Proof. Let x ∈ X. By applying Proposition 4.3.1, we may assume that (X,x) is irre-
ducible. Now we may apply the Jacobian Criterion, 4.3.6, and the coherence of ideal
sheaves, 6.3.2.
Theorem 6.3.5 (Finite Mapping Theorem). Consider a finite mapping π : X −→ Y of
complex spaces, and let S be a coherent OX –sheaf. Then π∗ S is a coherent OY –sheaf.
Proof. Using the Extension principle (6.2.15) it is enough to prove the theorem for ana-
lytic spaces, that is, both X and Y are reduced. Let y ∈ Y be a point. Since coherence
is a local property it is enough to prove that (π∗ S )|U is coherent for a sufficiently small
neighborhood U of y. We may assume that y = 0. The theorem is proved in several steps.

Step 1. We first consider the case that X is a small representative of (X, 0) = V (f ), 0 ,
where f ∈ C {x1 , . . . xn } is a Weierstraß polynomial of degree b in xn , Y an open subset
of C n−1 , π : X −→ Y the projection on the first n − 1 coordinates, and S = OX .
Let V ⊂ C be a small open neighborhood of 0. We choose an open neighborhood
U ⊂ C n−1 of 0 such that for all p ∈ U the b zeros (counted with multiplicity) of f (p,xn )
lie in V (continuity of roots, see 3.4.11). We may assume that
X = {(p,x) : p ∈ U, f (p,x) = 0} ⊂ U × V
and π : X −→ U is induced by the canonical projection. For an open subset W ⊂ U we

have a canonical map:
b
OU (W ) −→ OU (W )[xn ] ⊂ OU×V (W × V ) −→ OX (W × V ) = π∗ OX (W )
b−1
X
(s0 , . . . sb−1 ) 7→ sj xjn .
j=0
We will prove that for all p ∈ U the induced map of the stalks
b
(6.3) OU,p −→ (π∗ OX )p
b
is an isomorphism. This will prove this step. Namely, as On−1|U is coherent by Oka’s
Theorem 6.3.1 , it follows that π∗ OX|(U×V ) is a coherent On−1|U –module.
Now let p ∈ U , x1 , . . . ,xt be the zeros of f (p,x
Ptn ) and b1 , . . . ,bt their multiplicities.
Put pi := (p,xi ) for i = 1, . . . ,t. In particular, i=1 bi = b. Using 6.1.18, we obtain
(π∗ OX )p = ⊕ti=1 OX,pi . To prove that (6.3) is surjective, consider an element (g1 , . . . ,gt ) ∈
Pb−1
(π∗ OX )p = ⊕ti=1 OX,pi . By Corollary 3.3.24 there exists an r = j=0 ri xjn , such that
r = gi in OX,pi for i = 1, . . . ,t. This shows the surjectivity. The injectivity of (6.3)
follows from the uniqueness of the decomposition of 3.3.24.
Step 2. We consider now the situation as in Step 1 but now S is an arbitrary coherent
sheaf. Because S is coherent, we have locally an exact sequence:
q p
OX −→ OX −→ S → 0.
We use Theorem 6.1.17 and obtain an exact sequence:

q p
π∗ OX −→ π∗ OX −→ π∗ S → 0.
q p
We proved in Step 1 that π∗ OX is coherent. Hence π∗ OX = (π∗ OX )q and π∗ OX are
coherent. Therefore, π∗ S as a cokernel is coherent.
Step 3. We now consider the case that i : X ⊂ V (f ) is a closed subspace, f regular in xn ,
Y an open subset of C n−1 , π ′ : V (f ) −→ Y the projection on the first n − 1 coordinates
and π = π ′ ◦ i.
Then i∗ S is coherent by the extension principle, and π∗ S = (π ′ ◦ i)∗ S = π∗′ (i∗ S )
is coherent by the previous step.
Step 4. We now consider the case that X is a small representative of (V (I), 0) ⊂ (C n , 0),
where C {x1 , . . . ,xn }/I is a finitely generated C {x1 , . . . ,xk }–module, Y is a small open
neighborhood of 0 in C k , and π : X −→ C k the obvious projection on the first k
coordinates. This case follows from Step 3 and induction on n − k.
Step 5. We now consider the general case. We can (locally) view any finite mapping as a
projection:
i / n
X C
π π′
j
Y / C k.
This is the graph construction, see 2.3.14 and Exercise 3.4.43. Here π ′ is the projection
on the first k coordinates. By the extension principle π∗ S is coherent if and only if
j∗ (π∗ S ) = π∗′ (i∗ S ) is coherent. But the last sheaf is coherent by Step 4.
For the proof of the final coherence theorem in this chapter, the following theorem
is of crucial importance.
Theorem 6.3.6 (Oka). Let X be an analytic space. Then the set of nonnormal points
of X forms an analytic subspace of X. In particular, the set of normal points is open in
X.
Proof. Consider a point 0 ∈ X, and let I ⊂ R := OX,0 be the ideal of the singular locus
of X at the point 0. Then Exercise 4.4.20, which is the local version of Theorem 1.5.13,
gives:
=
R ֒→ HomR (I,I) ⇐⇒ R is normal.
We now sheafify. We have proved in 6.3.2 that the ideal sheaf I of the singular locus (it
is an analytic space) is coherent. It follows from 6.2.13 that the sheaf HomOX (I ,I ) is
coherent, and that
HomOX (I ,I )p = HomOX,p (Ip ,Ip )
for all p ∈ X. It follows that the set of nonnormal points is given by the support of
the sheaf HomOX (I ,I )/OX . As this is the quotient of two coherent sheaves, it is a
coherent sheaf and, therefore, its support is an analytic set, see 6.2.5. This proves the
theorem.
Theorem 6.3.7 (Oka). Let (X,OX ) be an analytic space. The normalization sheaf O eX
eX,x at x by definition is the normalization of OX,x . Then the
is a sheaf whose stalk O
normalization sheaf OeX is OX –coherent.
6.4 The Principle of Conservation of Number 243
Proof. For simplicity of notation, we will assume that X is irreducible. We showed in

4.4.8 that there is a germ of a normal space (X,e e x) and a finite map n : (X,e
e x) −→ (X,x).
In particular, there exist functions f1 , . . . ,fs in Q(OX,x ) which generate OX,e e
e x = OX,x as
Ps
OX,x –module, that is, OX,e e x = i=1 fi OX,x .
e
Let X be a representative of (X,e e x) and X be a representative of (X,x) such that X e
is a normal analytic space, f1 , . . . ,fs are defined on X, e and the induced map n : Xe −→ X
is finite. This is possible because the set of normal points is open (6.3.6). Then n∗ OXe is
coherent by the Finite Mapping Theorem 6.3.5. P As the fi have a universal denominator
s ∼
∆, multiplication with ∆ gives Ps an isomorphism i=1 fi OX = S , where S is a finitely
generated ideal sheaf. Thus i=1 fi OX is coherent. To prove the theorem, it therefore
Ps eX .
suffices to show that n∗ OXe = Pi=1 fi OX and is equal to the normalization sheaf O
s
Using ?? the equality n∗ OXe = i=1 fi OX P holds as both sheaves are coherent, and we
s
can lift the isomorphism on stalks OX,e e x = i=1 fi OX,x to an isomorphism on a small
open neighborhood.
To show that it is equal to the normalization sheaf, note that the fi satisfy an
integral equation
(i) (i)
(6.4) fiki + a1 fiki −1 + . . . + ak = 0.
(i)
Consider an even smaller open neighborhood U of x, on which the aj converge. Then
P
the equation (6.4) Pholds on U . In particular, for all a ∈ U , we have si=1 fi OX,a ⊂ OeX,a .
s −1
This shows that i=1 fi OX|U is contained in the normalization sheaf. Now let n (a) =
{p1 , . . . ,pt }. By construction of the germ of the analytic space X e and Theorem 6.1.18 we
have an equality
s
X
fi OX,a = n∗ (OXe )a = OX,p
e 1 ⊕ . . . ⊕ OX,p
e t.
i=1
All of the rings OX,p

e
e Therefore, the direct
for i = 1, . . . ,t are normal by the choice of X.
Ps i eX,a .
sum, which is i=1 fi OX,a , is normal and, therefore, equal to the normalization O
This proves the theorem.
Exercises
6.3.8. Prove that any compact complex subspace of a domain in C n consists of a finite number
of points. (Hint: Use the Finite Mapping Theorem, and induction on n.)
6.4 The Principle of Conservation of Number
To illustrate the principle of conservation of number, let us formulate three theorems,

whose proofs will be given later in this Section.
Theorem 6.4.1. Let (C, 0) and (D, 0) be two plane curve singularities, given by f (x,y) =
0 and g(x,y) = 0, respectively. Suppose that (C, 0) and (D, 0) do not have components in
common. Consider deformations of f and g, that is, F (x,y,s) and G(x,y,s) in C {x,y,s}
with F (x,y,0) = f and G(x,y,0) = g.
We can choose for all sufficiently small open neighborhoods U of 0 in C 2 , an open
neighborhood V of 0 in C 2 such that
(1) both F and G converge on U × V ,

(2) if C resp. D are representatives of (C, 0) respectively (D, 0), then C and D intersect
only in 0 and C \ {0} and D \ {0} are smooth.
For s fixed, denote by Cs , respectively Ds the curve {F (x,y,s) = 0} ⊂ U , respectively
{G(x,y,s) = 0} ⊂ U . Then, for all U sufficiently small there exists a V such that for all
s∈V X
(C, 0) · (D, 0) = (Cs ,p) · (Ds ,p).
p∈Cs
Here (Cs ,p) · (Ds ,p) denotes the intersection number of Cs and Ds in the point p, see
5.1.4.
This theorem gives a geometric interpretation of the intersection number. Suppose
that Cs and Ds intersect transversally, that is, the intersection number (Cs · Ds )p = 1
for all p ∈ Cs ∩ Ds . The theorem then says, that the intersection number of C and D is
equal to the number of intersection points appearing after a small perturbation of C and
D.
Example 6.4.2. Let C be given by f (x,y) = y 2 − x3 = 0, and D be given by g(x,y) =
y = 0. Then
(C, 0) · (D, 0) = dimC C {x,y}/(y 2 − x3 ,y) = dimC C {x}/(x3 ) = 3.
We consider the small perturbations:
F (x,y,s) = y 2 − x2 (x − s), G(x,y,s) = y − s.
We see indeed three intersection points appearing.

Theorem 6.4.3. Let (C, 0) = ∪ri=1 (Ci , 0) be a germ of a plane curve singularity, and let
e 0) = ∪r (C
fi = 0 be the equation of Ci . Consider the normalization (C, ei , 0) −→ (C, 0)
i=1
of (C, 0). Let W ⊂ C be an open neighborhood of 0 and
(x(ti ),y(ti )) : W −→ C 2 for i = 1, . . . ,r

be a parametrization of the branch Ci , and consider a one-parameter deformation of the

parametrization with base space S ⊂ C
(x(ti ,s),y(ti ,s)) : W × S −→ C 2 , x(ti ,0) = x(ti ), y(ti ,0) = y(ti ) for i = 1, . . . ,r.
Then we can take W and S so small that the image of (x(ti ,s),y(ti ,s)) is Q
an analytic space
r
in W × S given by Fi (x,y,s) = 0 with Fi (x,y,0)Q= fi . Then F (x,y,s) = i=1 Fi (x,y,s) is
r
the induced one-parameter deformation of f = i=1 fi .
Moreover for all sufficiently small neighborhoods U of 0 in W , there exists an open
neighborhood V of 0 in S such that:
(1) F converges on U × V .
(2) The only singular point of f on U is 0.
Denote for fixed s by Cs the curve {F (x,y,s) = 0} ∩ U , then for all s ∈ V we have
X
δ(C) = δ(Cs ,p),
p∈Cs
where δ(Cs ,p) denotes the δ–invariant of Cs in the point p.

Example 6.4.4. We consider the previous example. We consider the curve singularity
(C ∪ D, 0) given by the equation y(y 2 − x3 ) = 0. The deformation in fact is given by the
following deformation of the parametrization:
x(t1 ,s) = t21 + s, y(t1 ,s) = t31 + st1 ; x(t2 ) = t2 , y(t2 ) = s,
(check this). As the δ–invariant of an A1 –singularity is one, we see from the picture
that δ(C ∪ D, 0) = 4.
Theorem 6.4.5. Let f = f (x1 , . . . ,xn ) : (C n , 0) −→ (C , 0) be a germ of a holomorphic
function with isolated singularity, µ(f ) its Milnor number. Consider a deformation of f :
F (x1 , . . . ,xn ,s) : (C n × S, 0) −→ (C × S, 0), F (x1 , . . . ,xn ,0) = f (x1 , . . . ,xn ).
Then for all sufficiently small open neighborhoods U of 0 there exists an open neighborhood
V of 0 such that for all s ∈ V :
X
µ(f ) = µ(fs ,p)
p∈U
where µ(fs ,p) is the Milnor number of the germ of the holomorphic function:
F (x1 , . . . ,xn ,s) = fs : (C n ,p) −→ (C ,fs (p))

Example 6.4.6. We consider the E7 –singularity given by f (x,y) = y(y 2 − x3 ), and the
deformation F (x,y,s) = (y − s)(y 2 − x2 (x − s)). The zero sets of f and fs are given
in the previous figures. In the picture of the zero set we see four A1 –singularities. But
since a function on a compact set must have a maximum and a minimum, we have that
in each compact region in the real plane bounded by the zero set of fs there must be
a singularity, too. One can in fact calculate that fs has seven A1 –singularities, so that
µ(f ) = 7.
How to prove that “an invariant is constant under deformation”, or, to put it in
another way, when do we have conservation of number? Usually, a numerical invariant
is defined by the vector space dimension of a certain module M over the local ring of a
singularity OX,0 . In the above theorems, we have respectively as local ring and as module:
(1) The local ring is OX,0 = C {x,y}/(f,g), the module M is OX,0 .

(2) The local ring is OC,0 , the module M is OC,
e 0 /OC,0 .
∂f ∂f
(3) The local ring is On , and the module M is On /J(f ), where J(f ) = ( ∂x 1
, . . . , ∂x n
)
is the Jacobian ideal.
When deforming, the local ring OX,0 will be a quotient of the local ring OXS ,0 , and
similarly M is a factor module of MS . Both the local ring and the module are in fact
also C {s}–modules. In the above examples:
(1) OXS ,0 = C {x,y,s}/(F,G), MS = OXS ,0 .
eCS,0 /OCS ,0 .
(2) The local ring is OCS ,0 = C {x,y,s}/F , the module MS is O
(3) The local ring is C {x1 , . . . ,xn ,s}, and the module is MS = C {x1 , . . . ,xn ,s}/J(F ),
∂F ∂F
where J(F ) = ( ∂x 1
, . . . , ∂x n
) is the “relative” Jacobian ideal.
In order to be able to compare the invariants at different points in the parameter space
S, we need to sheafify.
Let the singularity (X, 0) be embedded in (C n , 0), and let U be a sufficiently small
open neighborhood of 0. Then we consider a representative X of (X, 0) in U . Furthermore,
we choose a small open neighborhood V of 0 in the parameter space S = C . Let π :
U × V −→ V be the canonical projection and XS ⊂ U × V be a complex space such that
XS ∩ π −1 (0) = X. For s ∈ V we put Xs := XS ∩ π −1 (s).
U
X0 Xs
V
0 s
Theorem 6.4.7 (Principle of Conservation of Number). Let M be a sheaf on XS with

the following properties:
(1) M is a coherent OXS –module.3
(2) M0 , the stalk of M at 0, is a free C {s}–module of finite rank.
Then for all sufficiently small neighborhoods U ′ ⊂ U of 0 in C n there exists an open

neighborhood V ′ ⊂ V of 0 in C such that for all s ∈ V ′ :
X
dimC (M|X0 )0 = dimC (M|Xs )p ,
p∈Xs′
where Xs′ = Xs ∩ (U ′ × {s}).

Proof. As support of a coherent analytic sheaf, Z := Supp(M ) is an analytic space, see
6.2.5. We claim that the natural projection map f : (Z, 0) −→ (S, 0) is finite. This we
prove in three steps.
Step 1. We give Z the structure of a complex space by means of the ideal sheaf Ann(M ).
This means that OZ = OXS /Ann(M ). By Theorem 6.2.15 M is a coherent OZ –module.
From the second assumption it follows that M0 /mS,0M0 is a finite-dimensional vector
space.
Step 2. By construction AnnOZ,0 (M 0 ) = (0). We claim
p
I := AnnOZ,0 /mS,0 OZ,0 (M0 /mS,0 M0 ) ⊂ (0),
that is, I is a nilpotent ideal. To show this, let M0 as C {s}–module be generated by

f ∈ OZ,0 be a representative of an element in I, so that f M0 ⊂ mS,0 M0 ,
m1 , . . . ,mr . Let P
r
that is, f mi = j=1 ξij mj , i = 1, . . . ,r for suitable ξij ∈ mS,0 . By Cramer’s rule 1.2.8,

we get that det (ξij ) − f Idr mk = 0 for k = 1, . . . ,r. As AnnOZ,0 (M0 ) = (0), it follows
that det (ξij ) − f Idr = 0. This says that f r ∈ mS,0 OZ,0 .
3 The invariants we are interested in are the dimensions of the stalks of the restriction of M to the
fibers.
Step 3. Put R = OZ,0 /mS,0 OZ,0 and M = M0 /mS,0 M0 . Thus M is a finite-dimensional

C –vector space, and an R/I algebra. As R/I–algebra, the annihilator of M is zero. It
follows from Exercise 1.2.31 that R/I is a finite-dimensional
p vector space, hence R/I
has dimension zero. On the other hand, because I ⊂ (0) the dimensions dim(R) and
dim(R/I) are equal, hence dim(R) is zero. This implies that dimC (R) < ∞ because
R is a factor ring of a convergent power series ring. By Theorem 3.4.24 it follows that
f : (Z, 0) −→ (S, 0) is a finite map, as claimed.
We now finish the proof. We take a representative f : Z −→ S such that f −1 (0) =
{0}, and such that f is finite. From the Finite Mapping Theorem 6.3.5 it follows that
f∗ M is coherent, whose stalk at 0 by assumption is C {s}t for a suitable t. From the
coherence of f∗ M it follows that there exists an open neighborhood V ′ of 0 in S such
that f∗ M|V ′ ∼= OCt |V ′ (consider the map OCt |V ′ −→ f∗ M|V ′ induced by (f∗ M ) 0 ∼ = C {s}t
′
on a neighborhood V ⊂ S and use Lemma ??).
Take a point s0 ∈ V ′ , and let q1 , . . . ,qℓ ∈ Z be the points in the fiber of f . Because
f is finite, one has by Theorem 6.1.18:
C {s}t ∼
= (f∗ M )s0 = ⊕ℓi=1 Mqi .
Restricting this to the point s0 ∈ V ′ and applying Exercise 6.1.35 implies that
Ct ∼
= ⊕ℓi=1 (M|Xs0 )qi .
This proves the theorem.

Proofs of Theorems 6.4.1, 6.4.3 and 6.4.5.
Proof of 6.4.1. We have to show that we can apply the previous theorem to the sheaf
M := OU×V /(F,G). By the results of the previous section we know that M is coherent
and, therefore, the first assumption is satisfied. So we have to check the second condition.
Therefore, consider the stalk of M at 0 which is the ring C {x,y,s}/(F,G). This is a
C {s}–module. By assumption (C, 0) and (D, 0) have no component in common, so that
by the Nullstellensatz 3.4.4 the vector space C {x,y}/(f,g) is finite-dimensional. By the
general Weierstraß Division Theorem 3.2.10 it follows that C {x,y,s}/(F,G) is a finitely
generated C {s}–module. To prove that it is free, we have to show, by Theorem 1.3.9, that
it is torsion free, hence it remains to show that s is a nonzerodivisor. As (C, 0) ∩ (D, 0)
consists of one point by assumption, it follows from 4.1.20 that f, g is a regular sequence
in C {x,y}, that is, g is a nonzerodivisor of C {x,y}/(f ). From Exercise 4.1.28 it follows
that if
a · f + b · g = 0,
then a = Lg and b = −Lf for some element L in C {x,y}. Now suppose that s·Q ∈ (F,G).
We have to show that Q ∈ (F,G). We can write
s·Q = A·F +B·G
for some A,B ∈ C {x,y,s}. We put a = A(x,y,0) and b = B(x,y,0). By plugging in it

follows that af + bg = 0, so that a = Lg, b = −Lf . We now add 0 = (−LG)F + (LF )G
to the expression for sQ and get
s · Q = (A − LG)F + (B + LF )G.
The power series A − LG and B + LG are divisible by s so that

A − LG B + LF
Q= F+ G.
s s
Hence Q ∈ (F,G) as we wanted to show.
Proof of 6.4.3. As above, we have to show that the conditions of Theorem 6.4.7 are
satisfied. For notational convenience we will only consider the case that the curve C is
irreducible. So we have a map
W × S −→ C 2 × S (t,s) 7→ (x(t,s),y(t,s),s).
Let CS = {F = 0} be the image. F is a generator of the kernel of the map C {x,y,s} −→
C {t,s}, x 7→ x(t,s), y 7→ y(t,s), s 7→ s. In particular F is irreducible. Let m = (x,y,s)
be
the maximal ideal in C {x,y,s}/(F ). Then C {t,s}/mC {t,s} = C {t}/ x(t,0),y(t,0) C {t}
is a finite-dimensional C –vector space. This implies, using the General Weierstraß Di-
vision Theorem 3.2.10, that C {t,s} is a finitely generated C {x,y,s}/(F )–module. Using
Theorem 3.4.24 we may assume (after perhaps shrinking W and S) that the map n :
W × S −→ CS is finite. Using Theorem 6.3.5 we obtain that n∗ OW ×S is a coherent OCS –
module and, therefore, M := n∗ OW ×S /OCS is a coherent OCS –module, too. The first
condition of Theorem 6.4.7 is therefore fulfilled. The stalk at 0 is M 0 = OW ×S,0 /OCS ,0 =
C {t,s}/C {x(t,s),y(t,s),s}. Because M 0 /(s)M 0 = C {t}/C {x(t,0),y(t,0)} is a finite-
dimensional C –vector space, we can apply Corollary 3.2.14, and conclude that M0 is
a finitely generated C {s}–module.
We now show that s is a nonzerodivisor of OW ×S,0 /OCS,0 . To show this, we claim
that F (x,y,0) is a defining equation for the curve C, that is, up to multiplication with a
unit it is equal to f . Now we have the following isomorphisms:

C {x,y}/ F (x,y,0),x ∼ = C {x,y,s}/(F,x,s) ∼ = C {x(t,s),y(t,s),s}/ x(t,s),s

∼
= C {x(t),y(t)}/ x(t) ∼ = C {x,y}/(f,x).
This implies that both f and F (x,y,0) are y–regular of the same order. So the fact that
f is up to a unit equal to F (x,y,0) now follows from 3.3.16. Now suppose s · Q ∈ OCS,0
for some Q ∈ OC ×S,0 . We have to show that Q ∈ OCS,0 . We can write
s · Q = G(x(t,s),y(t,s),s)
for some G ∈ C {x,y,s}. Plugging in s = 0 yields that G(x(t),y(t),0) = 0, so that G(x,y,0)
is divisible by f , say G(x,y,0) = Lf . We may now, without loss of generality, assume
that F (x,y,0) is equal to f . Now add 0 = −LF to the expression for s · Q. Then
s · Q = (G − LF )(x(t,s),y(t,s),s).
In the ring C {x,y,s} we can divide G − LF by s, so that Q = G−LF s (x(t,s),y(t,s),s).
Hence Q is already an element of OCS,0 , so that s is a nonzerodivisor, as was to be shown.
It remains to show that (M|Cs )p = O eC ,e /OCs ,p for s ∈ S and p ∈ Cs . Looking at the
s p
integral equations for x(t,s) and y(t,s) for fixed s it follows that the map n : W × {s} −→
Cs is finite. As we just proved that in particular OW ×{s} /OCs is a finite-dimensional
vector space it follows from Exercise 4.4.22 that n : W × {s} −→ Cs is the normalization
for all points in Cs . This is what we needed to show.
Proof of 6.4.5. This is analogous to the proof of 6.4.1, and left as Exercise 6.4.11.
We may ask ourselves whether, for example, in the case of the Milnor number, a
deformation with only A1 –singularities always exists. This question is answered in the
following theorem.
Theorem 6.4.8. Consider a function f ∈ C {x1 , . . . ,xn } with an isolated singularity.
Let F : (C n × C n , 0) −→ (C , 0) be defined by
F (x1 , . . . ,xn ,t1 , . . . ,tn ) := f + t1 x1 + . . . + tn xn .
Then there exists a hypersurface (D, 0) ⊂ (C n , 0), such that for all small enough repre-
sentatives D and T of (D, 0) and (C n , 0), there exists a small neighborhood U of 0 in C n
such that for all t ∈ T \ D
Ft (x1 , . . . ,xn ,t1 , . . . ,tn ) : U −→ C
has only A1 –singularities. The number of these A1 –singularities is equal to the Milnor
number µ(f ) of f .
Proof. Let U be a neighborhood of 0 in C n such that f converges on U and let t ∈ C n .
A point p ∈ U is a singular point of Ft exactly if all partial derivatives with respect to
the xi vanish. In short, if p is in the zero set of ∂F ∂Ft
∂x1 , . . . , ∂xn . We therefore consider the
t
n n
space (Σ, 0) in (C × C , 0) defined by the ideal

∂F ∂F ∂f ∂f
,..., = + t1 , . . . , + tn .
∂x1 ∂xn ∂x1 ∂xn
Let Σ be a representative of (Σ, 0), We want a condition for a point (p,t) in Σ to have an
A1 –singularity. For this, we look at the projection to the second factor
π : Σ −→ C n .
Suppose that given (p,t) ∈ Σ, there exists a neighborhood of (p,t) which is mapped
biholomorphically under π onto an open subset T of C n . Then we claim that Ft has an
A1 –singularity at p. Indeed, by assumption we have that the inclusion of stalks
OT,t ⊂ OΣ,(p,t)
is a isomorphism. Dividing out the local parameters at t, we have

∼ ∂Ft ∂Ft
C = On,p / (p), . . . , (p) .
∂x1 ∂xn
Hence Ft has Milnor number one at p. This implies that it is an A1 –singularity at p.

We therefore have to investigate the map π : Σ −→ C n . We first look at the space
(Σ, 0). In the local ring OΣ,0 of this space, we can eliminate the functions t1 , . . . ,tn . Hence
OΣ,0 ∼= C {x1 , . . . ,xn }, so we see that (Σ, 0) is smooth. The local ring of the fiber of π
above 0 ∈ C n is given by
∂f ∂f ∂f ∂f
OC n ×C n ,0 / + t1 , . . . , + tn ,t1 , . . . ,tn = OC n ,0 / ,..., ,
∂x1 ∂xn ∂x1 ∂xn
which is a finite-dimensional vector space by assumption. Therefore, the map π : (Σ, 0) −→

(C n , 0) is finite by 3.2.10 and 3.4.24. As dim(Σ, 0) = dim(C n , 0) and π is closed, it fol-
lows that π is surjective. Therefore, π : (Σ, 0) −→ (C n , 0) is a Noether normalization.
We proved in 3.4.14 that there exists a hypersurface D ⊂ T , such that for all t ∈ T \ D,
there exists a small neighborhood V of t, such that π −1 (V ) is the union of s open subsets
in Σ, each of which is mapped biholomorphically on V . Therefore, by the discussion in
the beginning of the proof, for all t ∈ T \ D, the function Ft has only (finitely) many
A1 –singularities. The fact that this number is equal to µ(f ) is proved by means of con-
servation of number, see 6.4.5.
Example 6.4.9. Let f = x3 . Consider F = x3 + tx. Then Σ is given by
∂F
= 3x2 + t = 0.
∂x
The map π : Σ −→ T is given by the following picture.
So we see that for t 6= 0 we get two A1 –singularities.

Corollary 6.4.10. Consider f ∈ On with µ(f ) < ∞, and let u ∈ On be a unit. Then
µ(uf ) = µ(f ). In particular, for a germ of an analytic hypersurface (X, 0) with an isolated
singularity the Milnor number µ(X, 0) is well defined.
Proof. By symmetry, it suffices to show µ(uf ) ≥ µ(f ). Take a small open neighborhood
V of 0 in C n . Consider a family Ft ∈ C {x1 , . . . ,xn ,t} with F (x1 , . . . xn ,0) = f . Let U be
a small open neighborhood of 0, and suppose that the function Ft for t fixed has exactly
µ(f ) A1 –singularities in V . Then uFt is a family of functions with uF (x1 , . . . xn ,0) = uf .
If Ft has an A1 –singularity at p, then uFt also has an A1 –singularity at p. From the
principle of conservation of number, it follows that µ(uf ) ≥ µ(f ). This is what we
wanted to show.
Exercises
6.4.11. Prove 6.4.5.
6.4.12. For plane curve singularities (X, 0) and (Y, 0), prove Hironaka’s formula, see 5.4.11,using
the principle of conservation of number, in particular Theorems 6.4.1 and 6.4.3.
6.4.13. In the Theorem of the Principle of Conservation of Number, replace the second con-
dition by the condition that M0 is a finitely generated C {s}–module, or, what is the same by
Nakayama, that (M|X0 )0 is a finitely generated C –vector space. Prove that
` ´ X ` ´
dimC (M|X0 )0 ≥ dimC (M|Xs )p .
p∈Xs′
6.4.14. An ordinary m–tuple point is by definition a plane curve singularity (X, 0) = ∪m i=1 (Xi , 0)
such that all (Xi , 0) are smooth, and the intersection number (X i , 0) · (Xj , 0) = 1 for all i 6= j.
` ´
Prove that the δ–invariant of an ordinary m–tuple point is m 2
using
(1) Hironaka’s formula 5.4.11 or
(2) the principle of conservation of number.
Find a C –basis of OX,
e 0 /OX,0 .
` ´
6.4.15. Consider an irreducible plane curve singularity (X, 0) = V (f ), 0 in (C 2 , 0) of multi-
plicity m. Without loss of generality we may suppose that (X, 0) is given by the parametrization
x(t) = tm , y(t) = tm ϕ(t).
Consider a deformation of the parametrization as follows
x(t,s) = tm + s, y(t,s) = (tm + s)ϕ(t).

` ´
Let F ∈ C {x,y.x} be irreducible with F x(t,s),y(t,s) = 0 and F (x,y,0) = f .
(1) Show that one can take f as follows:
f (x,y) = y m + a1 (x)y m−1 + . . . + am (x) = 0,
where the order of ai is at least i.

(2) Show that F looks like
a1 (x − s) m−1 am (x − s) m
F = ym + y x +... + x = 0.
x−s (x − s)m
(3) Show that for small s 6= 0 the zero set of fs (x,y) := F (x,y,s) has two singular points:
1. an ordinary m–tuple point.
2. the strict transform of the blowing-up of the origin in C 2 .
For which values of t do these occur?
(4) Use the principle of conservation of number, in particular 6.4.3, to reprove Noether’s
formula 5.4.13.
(5) Generalize these results to reducible curves.
6.5 Cohen-Macaulay Spaces
The Cohen-Macaulay property (CM for short) of a space is a local property, related to
the principle of conservation of number. In this section we deal with the first properties of
Cohen-Macaulay spaces, hopefully showing that it is a good (and, therefore, quite strong)
property a space can have. Nevertheless, we will prove that complete intersections (in
particular hypersurface singularities) are Cohen-Macaulay. First we introduce the notion
of depth, needed in order to define Cohen-Macaulay. We introduce these matters directly
for modules over a local ring, in order to get a more flexible development of the theory.
6.5 Cohen-Macaulay Spaces 253
Definition 6.5.1. Let (R,m) be a local ring and M be an R–module.

(1) A sequence f1 , . . . ,fr of elements in m is called a regular sequence of M if f1 is not
a zerodivisor of M , and fi is not a zerodivisor of M/(f1 , . . . ,fi−1 )M for i = 2, . . . ,r.
(2) Let I ⊂ R be an ideal with IM 6= M . Then the I–depth of M , depth(I,M ) is the
maximal length of a regular sequence of M in I. If IM = M we define depth(I,M ) =
∞.
(3) The depth of M , depth(M ) is the maximal length of a regular sequence of M , that
is, depth(M ) = depth(m,M ). If we want to emphasize the ring R, we will write
depthR (M ).
(4) The module M is called Cohen-Macaulay, if the depth of M is equal to the di-
mension of M . The dimension of the module M is defined to be the dimension of
R/ Ann(M ).
(5) R is called Cohen-Macaulay, if R is a Cohen-Macaulay R–module.
(6) A germ (X,x) is called Cohen-Macaulay, if OX,x is Cohen-Macaulay.
As usual, the depth of a space at a point is defined as the depth of the corre-
sponding local ring. Note that because of the Active Lemma 4.1.10 we always have the
inequality depth(M ) ≤ dim(M ). If IM 6= M then depth(I,M ) ≤ depth(M ). Notice that
depth(I,M ) > depth(M ) implies IM = M and, especially for M = R, it implies I = R.
Furthermore, it is immediate from the definition that the depth of a module is zero ex-
actly if every element of m is a zerodivisor for M . From 1.4.52 it follows that m is an
associated prime ideal of M , so that m = M : f for some f ∈ m. Hence the following
lemma follows.
Lemma 6.5.2. Suppose that M is an R–module, and that depth(M ) = 0. Then m is an
associated prime of M , that is, there exists an f ∈ M \ {0} such that m · f = 0.
In case that M = R is the local ring of a germ of a complex space, the depth is
zero exactly if R is a zero-dimensional ring, or the special point is an embedded point.
More generally, the depth of a germ of a complex space at a point is the number of
hypersurface sections one has to take before one gets an embedded point. Note that if
a nonzero-dimensional germ of a complex space is reduced, it must have positive depth.
The Cohen-Macaulay property, therefore, states that an embedded point will not occur
when taking dim(X) successive hypersurface sections. It will turn out that the length
of a maximal regular sequence is independent of the maximal regular sequence. Let us,
before giving examples, prove this statement, for which we need the following lemma.
Lemma 6.5.3. Let R be a Noetherian local ring, M a finitely generated R–module. Let
{f,g} be a regular sequence of M . Then {g,f } is also a regular sequence.
Proof. The proof will be given in two steps.
Step 1. We first show that g is a nonzerodivisor. Suppose ga0 = 0, a0 ∈ M . In particular,
ga0 ∈ f · M . The assumption that {f,g} is a regular sequence, allows us to deduce that
a0 ∈ f · M . Write a0 = f a1 , a1 ∈ M . Then ga0 = f ga1 = 0, and f is not a zero divisor, so
ga1 = 0, etc. Therefore, for all k there exists an ak ∈ M such that a0 = ak f k . Therefore,
a0 ∈ ∩mk M which is zero according to Krull’s Intersection Theorem (Exercise 1.3.24).
Step 2. We have to show that f is a nonzerodivisor of M/g · M . Suppose f a ∈ g · M , that

is, f a = gb for some a,b ∈ M . Since by assumption, g is a nonzerodivisor of M/f · M , it
follows that b ∈ f · M , that is, b = f c for some c ∈ M . Therefore, f (a − gc) = 0. Because
f is not a zero divisor, a − cg = 0, that is, a ∈ g · M . This is what we had to show.
Lemma 6.5.4. Let R be a Noetherian local ring, I ⊂ R an ideal and M an R–module.
Any two maximal regular sequences in I for M have the same number of elements.
Proof. Let f1 , . . . ,fn and g1 , . . . ,gp be regular sequences in I, which cannot be extended.
We may assume that n ≥ p. We have to show that p = n. The lemma will be proved
by induction on n. The case n = 1 is trivial. We now claim that there exists an f ∈ I
such that the sequences f1 , . . . ,fn−1 and g1 , . . . ,gp−1 can be extended to regular se-
quences f1 , . . . ,fn−1 ,f and g1 , . . . ,gp−1 ,f . Let S be the union of the associated primes
of (f1 , . . . ,fn−1 )M and of (g1 , . . . ,gp−1 )M , then, by the characterization of nonzerodivi-
sors, see 1.4.52, the ideal (fn ,gp ) 6⊂ S, because (fn ,gp ) ⊂ S would imply (fn ,gp ) ⊂ p, p
one of the associated primes (cf. 1.1.13, Prime Avoidance). But by assumption fn or gp
are nonzerodivisors in M/(f1 , . . . ,fn−1 )M , resp. M/(g1 , . . . ,gp−1 )M . This implies that
afn + bgp =: f 6∈ S for suitable a,b ∈ R and has, therefore, the required property. Apply-
ing the case n = 1 to the ring R/(f1 , . . . ,fn−1 ) gives that the sequence f1 , . . . ,fn−1 ,f is
maximal. Similarly g1 , . . . ,gp−1 ,f is a maximal regular sequence.
By repeatedly applying the previous lemma, we find that f,f1 , . . . ,fn−1 and f,g1 , . . . ,gp−1
are maximal regular sequences. Applying the induction hypothesis to the ring R/(f )
proves the lemma.
Corollary 6.5.5. Let R be a Noetherian local ring, M be an R–module, I ⊂ R an ideal
and f ∈ I a nonzerodivisor of M . Then:
(1) depth(I,M/f M ) = depth(I,M ) − 1,
(2) M is Cohen-Macaulay ⇐⇒ M/f M is Cohen-Macaulay.
Proof. The first part follows from Lemma 6.5.4. For the second part, the implication
“⇐=” is trivial. For the other implication, we use the first part. It remains to show
dim(R/(Ann(M ),f ) = dim(R/ Ann(M ))−1. This holds, as from the assumption it follows
easily that f is a nonzerodivisor of Ann(M ), so that we can apply Krull’s Principal Ideal
Theorem 4.2.16.
Examples 6.5.6.
(1) Smooth spaces are Cohen-Macaulay.
(2) By applying the second part of Corollary 6.5.5 we obtain that complete intersection
singularities (in particular hypersurface singularities) are Cohen-Macaulay.
(3) Reduced curve singularities are Cohen-Macaulay. This is also obvious, as the depth
is positive and, therefore, only can be one, because the depth cannot be bigger than the
dimension.
(4) The singularity in three–space given by the ideal (xz,yz)
is reduced, but not Cohen-Macaulay. An example of a nonzerodivisor is z − x, which cuts

down the space to the following space with embedded point.
The local ring of the intersection is C {x,y,z}/(xz,yz,x−z) ∼

= C {x,y}/(x2 ,xy), hence
we see the embedded point appear. The depth of the germ the analytic space V (xz,yz), 0
is one. This space is therefore not Cohen-Macaulay.
(5) Consider C 4 with coordinates x,y,z,w, and consider the union X of the two planes
x = y = 0 and z = w = 0, intersecting in one point. The ideal of this union is (x,y) ∩
(z,w) = (xz,xw,yz,yw). It is geometrically clear that if we intersect with a generic three-
dimensional linear space, we have the union of two lines. But we obtain an embedded
point. If we intersect with y − w = 0, the local ring becomes:
C {x,y,z,w}/(xz,xw,yz,yw,y − w) ∼
= C {x,y,z}/(xz,xy,yz,y 2).
A primary decomposition of (xz,xy,yz,y 2 ) is given by:
(xz,xy,yz,y 2 ) = (y,x) ∩ (y,z) ∩ (y 2 ,x,z).
So we see an embedded point appearing on the A1 –singularity. The depth is one,

and, therefore, (X, 0) is not Cohen-Macaulay.
(6) To give an example of an irreducible space which is not Cohen-Macaulay, we consider
the map:
(C 2 , 0) −→ (C 4 , 0); (s,t) 7→ (x,y,z,u) = (s,st,t2 ,t3 ).
By Exercise 2.3.18 the ideal of the image is:
(y 2 − zx2 ,yz − ux,z 2 x − uy,u2 − z 3 ).
A general hyperplane section is given by x − z = 0. This section gives a space given by

the ideal
(y 2 − x3 ,x(y − u),x3 − uy,u2 − x3 ).
From x(y − u) = 0 it follows that either x = 0 or y = u. From the first possibility, it
follows that y = u = 0. The second possibility gives y = u, y 2 = x3 , a cusp singularity. In
total we get a cusp singularity with an embedded point. In fact, a primary decomposition
of the ideal is:
(y − u,y 2 − x3 ) ∩ (x,y 2 ,uy,u2 ).
Indeed, we need embedding dimension four to find an example of a reduced irreducible
space which is not Cohen-Macaulay, due to the Principal Ideal Theorem and the second
example.
The examples four and five are quite typical. In fact, a Cohen-Macaulay space is
equidimensional, even unmixed. For the definition of unmixed, see 4.1.22. In the fifth
example, the intersection of the two components has dimension zero. A necessary con-
dition for the union to be Cohen-Macaulay is the condition that the intersection has
codimension one, as we will see in 6.5.19. At least from these examples we see that the
Cohen-Macaulay property is indeed a nontrivial one.
Proposition 6.5.7. Let (R,mR ) and (S,mS ) be local rings, and R ⊂ S be a finite
ring extension of local rings. Let M be an S–module. Suppose depthR (M ) = 0. Then
depthS (M ) = 0.
Proof. As depthR (M ) = 0, we have that m is an associated prime of M . Thus there
exists an x ∈ M with x · m = 0. Take an element g ∈ mS . We have to show that g is a
zerodivisor of M . By 1.5.11 g is integral over mR , so we have an integral equation:
g n + a1 g n−1 + . . . + an = 0, ai ∈ mR .
As depthR (M ) = 0, we have ai x = 0 for i = 1, . . . ,n. Hence g n · x = 0. Thus g is a

zerodivisor of M .
This proposition will be generalized in Exercise 6.5.29.
Corollary 6.5.8.
(1) Let OX,0 be a singularity, Ok ֒→ OX,0 be a Noether normalization. Then (X, 0) is
Cohen-Macaulay if and only if OX,0 is a free Ok –module.
(2) A Cohen-Macaulay space is unmixed.
Proof. (1) Recall from 1.3.9 that if M is a finitely generated R–module, x ∈ m a nonze-
rodivisor of M , then M is a free R–module if and only if M/xM is a free R/(x)-module.
By 6.5.7 we can find an x ∈ Ok which is a nonzerodivisor of OX,0 . The first statement
follows by induction, since Corollary 6.5.5 gives that OX.0 /(x) is also Cohen-Macaulay.
(2) We know by the first part, that OX,0 is a free Ok –module, so, in particular, every
element of Ok is a nonzerodivisor of OX,0 . Therefore, (X, 0) is unmixed, by Theorem
4.1.22.
The second part of this corollary, applied to the case of a complete intersection is
usually called Macaulay’s Unmixedness Theorem.
Corollary 6.5.9. Let (X, 0) be an irreducible complete intersection singularity, and sup-
pose the singular locus of (X, 0) has codimension at least two. Then (X, 0) is normal.
Proof. By assumption, OX,0 satisfies (R1). We only need to show that it satisfies (S2)
by 4.4.11. The condition (S2) means that every hypersurface in (X, 0) has no embed-
ded primes. Because of Corollary 6.5.5 this hypersurface is also Cohen-Macaulay, and
therefore unmixed. Thus (S2) holds for Cohen-Macaulay singularities.
Theorem 6.5.10. Let (X, 0) be Cohen-Macaulay germ of an analytic space and π :
(X, 0) −→ (C k , 0) be a Noether normalization. Then there exist representatives X of
(X, 0) and U of (C k , 0) and a hypersurface D ⊂ U such that for all t ∈ U \ D the fiber
π −1 (t) consists of dimC OX,0 /mC k ,0 OX,0 many points.
Proof. As a Cohen-Macaulay space is unmixed, in particular equidimensional, we can

reduce to the case that (X, 0) is irreducible. As in the proof of 6.4.7 we choose X and U
such that π∗ OX ∼ = OU s
. This is possible because π∗ OX is coherent and (π∗ OX )0 = OX,0 is
a free OC k ,0 –module of finite rank s (Corollary 6.5.8). Let p ∈ U and π −1 (p) = {q1 , . . . ,qr }
then
s
OU,p ∼
= (π∗ OX )p = ⊕ri=1 OX,qi
and, therefore,
r
X
s= dimC OX,qi /mC k ,p OX,qi .
i=1
As in the proof of Theorem 6.4.8, we use the Local Parametrization Theorem 3.4.14
to find a hypersurface D ⊂ U with the following property: every point p ∈ U \ D has
a neighborhood V such that π −1 (V ) is a union of open subsets of X each of which
is mapped biholomorphically to V . This implies OX,qi ∼
= OU,p for all i and, therefore,
r = s.
Remark 6.5.11. The previous theorem does not hold if (X, 0) is not Cohen-Macaulay,
see the following examples. We can apply the theorem to the theory of intersection num-
ber. We consider f,g such that dimC C {x,y}/(f,g) < ∞, and deformations FS ,GS of f
and g, with OS,0 = Ok . Let (X, 0) = V (FS ,GS ), 0 . Then (X, 0) −→ (C k , 0) is a Noether
normalization. (X, 0) is a complete intersection, hence Cohen-Macaulay. The number of
points in Xs = V (Fs ,Gs ) for general s, therefore, is equal to dimC C {x,y}/(f,g).
Examples 6.5.12.
(1) Let I = (xy,x(x − z)). Then (X, 0) = V (I), 0 is the union of a plane and a line. The
projection (X, 0) −→ (C 2 , 0) which sends (a,b,c) to (b,c) is a Noether normalization. On
the level of rings we have a map
C {y,z} ⊂ C {x,y,z}/(xy,x2 − xz).
The number of points in a general fiber obviously is one. But

dimC C {x,y,z}/(I + (y,z)) = dimC C {x}/(x2 ) = 2.
Therefore, (X, 0) is not Cohen-Macaulay.

(2) We consider example 6.5.6 (6). So let I = (y 2 − zx2 ,yz − ux,z 2 x − uy,u2 − z 3 ). We
consider the projection π : (X, 0) = V (I), 0 −→ (C 2 , 0), which sends (a,b,c,d) to (a,c).
Recall that (X, 0) was given by the image of the map (C 2 , 0) −→ (C 4 , 0) which sends
(s,t) to (s,st,t2 ,t3 ). The composition is the map sending (s,t) to (s,t2 ). So we see that
(X, 0) −→ (C 2 , 0) is surjective, and in fact in a general fiber of π we have two elements.
Since π is surjective we have an inclusion of rings
C {x,z} ֒→ C {x,y,z,u}/I.
Now C {x,y,z,u}/(I + (x,z)) = C {y,u}/(y 2,uy,u2 ) which is a three-dimensional vector

space. By the General Division Theorem, it follows that C {x,y,z,u}/I is a finitely gen-
erated C {x,z}–module. Hence π : (X, 0) −→ (C 2 , 0) is a Noether normalization, has
two points in a general fiber, but the vector space C {x,y,z,u}/(I + (x,z)) has dimension
three. Thus we see that (X, 0) is not Cohen-Macaulay.
There is also a completely algebraic method by which one can decide whether a
ring (or module) is Cohen-Macaulay. It is based on resolutions of modules. This method
is often used in the literature for proving that for a certain invariant the principle of
conservation of number holds.
Definition 6.5.13. Let (R,m) be a Noetherian local ring and M be a finitely generated
R–module. A free resolution of M is an exact sequence (infinite of finite)
αk+1 1 0 α α
(6.5) · · · −→ Fk+1 −−−→ Fk −→ · · · −→ F0 −→ M −→ 0,
with finitely generated free R–modules Fi for i ≥ 0. One says that M has finite projective
dimension if there exists an exact sequence
n α 1 0 α α
(6.6) 0 −→ Fn −→ Fn−1 −→ · · · −→ F0 −→ M −→ 0,
with finitely generated free R–modules Fi . The minimal n with Fn 6= 0 is called the
projective dimension of M , denoted by pd(M ) = pdR (M ). If M does not have finite
projective dimension, we write pdR (M ) = ∞. The resolution (6.5) is called minimal, if
for all k ≥ 1, the map αk maps Fk into mFk−1 .
Theorem 6.5.14. (Minimal) free resolutions of finitely generated modules M over a
Noetherian local ring (R,m) exist. If a module M has a finite minimal resolution
n α 1 0 α α
0 −→ Fn −→ Fn−1 −→ · · · −→ F0 −→ M −→ 0,
with Fn 6= (0), then n = pdR (M ). To put it in other words, two minimal resolutions have
the same length. Moreover, the rank of Fk is independent of the minimal resolution, and
is called the k–th Betti number of M .
Proof. Consider a minimal set of generators b1 , . . . ,bs0 of M . We get a surjective map
α
F0 := Rs0 −→
0
M
with α0 (a1 , . . . ,as0 ) = a1 b1 + . . . + as0 bs0 . The map α0 is surjective, because b1 , . . . ,bs0
generate M . By Nakayama, α0 induces an isomorphism
α0 : F0 /mF0 ∼
= M/mM.
Let K1 be the kernel of α0 . Then K1 ⊂ mF0 , as α0 is an isomorphism. As K1 is a

submodule of the Noetherian module F0 , it is finitely generated. Hence, there exists a
surjective map
F1 ∼
= Rs1 −→ K1 −→ 0,
where s1 is the minimal number of generators of K1 . We denote the composition F1 −→
K1 −→ F0 by α1 . As K1 ⊂ mF0 , it follows that α1 (F1 ) ⊂ mF0 . We get an exact sequence
α1 0 α
F1 −→ F0 −→ M −→ 0.
Now put K2 = Ker(α1 ), take minimal number of generators of K2 etc. Continuing this
way, we get a minimal free resolution of M .
To show the invariance of the Betti-numbers, consider two minimal resolutions of
M:
αk+1 1 0 α α
· · · −→ Fk+1 −−−→ Fk −→ · · · −→ F0 −→ M −→ 0,
′ α′k+1 α′ α′
· · · −→ Fk+1 −−−→ Fk′ −→ · · · −→
1
F0′ −→
0
M −→ 0.
As both α1 and α′1 both have entries in m, we get isomorphisms F0 /mF0 ∼ = M/mM ∼ =
F0′ /mF0′ . Therefore rank(F0 ) = rank(F0′ ).
′ ′
Pf1 , . . .′ ,fs0 be a basis of F0 . As the α0 (fj ) generate
Let e1 , . . . ,es0 be a basis of F0 , and
M , there exist βij ∈ R with α0 (ei ) = βij α0 (fj ). Consider the map h1 = (βij ) : F0 −→
F0′ . So we get a commutative diagram
α0
F0 /M
h1
α′0
F0′ / M.
Modulo m the map h1 : F0 /mF0 −→ F0′ /mF0′ is an isomorphism, that is, det(βij )
mod m 6= 0. Then det(βij ) is a unit in R which implies that h1 is an isomorphism.
Hence h1 induces an isomorphism
h1 : Ker(α0 ) = K1 −→ K1′ = Ker(α′0 ).
As α2 and α′2 have entries in m, and we have surjections F1 −→ K1 and F1′ −→ K1′ , it
follows that rank(F1 ) = rank(F1′ ), etc.
Examples 6.5.15.
(1) An R–module M has projective dimension zero if and only if M is a free R–module.
(2) Let R = C {x,y}, I = (x,y), and M = R/I. We will show that
(−y
x ) (x y) res
0 −→ R −−−→ R2 −−−→ R −−→ R/I −→ 0
is a free resolution of R/I as R–module. Here res is just taking residue classes, and is
obviously surjective. The kernel of res is generated by x and y. We have to compute the
(x y)
kernel of R2 −−−→ R. So let (α,β) be in the kernel. Then x · α + y · β = 0. As (x,y) is a
regular sequence in C {x,y} it follows that α = y · m, for a certain m. Similarly β = x · ℓ
for a certain ℓ. Then xy · m + xy · ℓ = 0, hence ℓ = −m. Therefore, (α,β) = ℓ(−y,x). So
(−y
x ) (x y)
R −−−→ R2 −−−→ R
(−y
x )
is exact. Moreover, it is easy to see that R −−−→ R2 is injective.
(3) We consider the ring R = C {x,y}/(xy), and M = C {y} ⊕ C {x} = R/(x) ⊕ R/(y).
Then M = R,e the normalization of R. M can be described as the cokernel of the map
0
α1 =(x
0 y)
R2 −−−−−−→ R2 −→ M −→ 0.
The kernel of α1 is easily seen to be generated by (y,0) and (0,x). Hence we get an exact
sequence
0
α2 =(0y x0) α1 =(x
0 y)
R2 −−−−−−→ R2 −−−−−−→ R2 −→ M −→ 0.
Going on like this, we see that M has a minimal free resolution which is 2-periodic. In
particular we see that M does not have finite projective dimension.
So one sees that a finitely generated module over a local ring does not, in general,
have finite projective dimension. But it is true in the case that R is a regular local ring.
The analogous statement for graded modules over a polynomial ring is a famous theorem
of Hilbert.
Theorem 6.5.16 (Hilbert’s Syzygy Theorem). Let (R,m) be a regular local ring of di-
mension n, and M be a finitely generated R–module. Then pdR (M ) ≤ n. In particular,
the projective dimension of M is finite.
For the proof we need the following lemma.
Lemma 6.5.17. Let (R,m) be a Noetherian local ring, let M be a finitely generated R–
module, and x ∈ m be a nonzerodivisor of R and M . Consider a minimal free resolution
of M
0 −→ Fn −→Fn−1 −→ · · · −→F0 −→M −→ 0.
Then the induced sequence
(6.7) 0 −→ Fn /xFn −→Fn−1 /xFn−1 −→ · · · −→F0 /xF0 −→M/xM −→ 0
is a minimal free resolution of M/xM as R/(x)–module.

Moreover, pdR (M ) = pdR/(x) (M/xM ).
Proof. By 1.2.35 the sequence (6.7) is indeed exact. Moreover, from Fn /xFn = 0 it would
follow that Fn = xFn , hence Fn = 0 by Nakayama. Hence pdR (M ) = pdR/(x) (M/xM ).
Proof of Hilbert’s Syzygy Theorem 6.5.16. The proof is by induction on n. If n = 0, then

R is a field and M is a finitely generated vector space over R. Hence M is free, and the
projective dimension pdR (M ) = 0. Otherwise, consider a minimal free resolution
· · · −→ Fk −→ Fk−1 −→ · · · −→ F1 −→ F0 −→ M −→ 0
of M . We split this exact sequence into two exact sequences
· · · −→ Fk −→ Fk−1 −→ · · · −→ F1 −→ K −→ 0,
0 −→ K −→ F0 −→ M −→ 0,
with K = Ker(F0 → M ). Take any element x ∈ m \ m2 . As R is an integral domain, see
4.3.17, x is a nonzerodivisor of F0 . As K is a submodule of F0 , x is a nonzerodivisor of
K. Therefore x is a nonzerodivisor of Fi for all i ≥ 1 and of K, so that we get an exact
sequence
· · · −→ Fk /xFk −→ Fk−1 /xFk−1 −→ · · · −→ F1 /xF1 −→ K/xK −→ 0
The ring R is Noetherian, so that K/xK is a finitely generated R/(x)–module. As R/(x) is

a regular local ring of dimension n−1, it follows by induction that pdR/(x) (K/xK) ≤ n−1.
Hence Fk /xFk = 0 for k ≥ n + 1, because the resolution is minimal. By Nakayama, it
follows that Fk = 0 for k ≥ n + 1. This proves the theorem.
Lemma 6.5.18 (Depth Lemma). Let R be a Noetherian local ring, and consider a short
exact sequence of R–modules
0 −→ M1 −→ M2 −→ M3 −→ 0.
Then
depth(M2 ) ≥ min depth(M1 ), depth(M3 ) .
In case this inequality is strict we have, furthermore,
depth(M1 ) = depth(M3 ) + 1.
Proof. Suppose that all three modules have positive depth. From Prime Avoidance, see
Exercise 6.5.30, we can find an f ∈ R which is a nonzerodivisor of M1 ,M2 and M3 . By
the Snake Lemma 1.2.12, it follows that
0 −→ M1 /f M1 −→ M2 /f M2 −→ M3 /f M3 −→ 0
is also exact. Since the depth drops by one if one divides out a nonzerodivisor (Corollary
6.5.5), we can immediately reduce to the case that the depth of one of the three modules
M1 ,M2 or M3 is zero. So we distinguish three cases.
Case 1. Suppose depth(M1 ) = 0. We will prove that depth(M2 ) = 0. But this is obvious,
as any nonzerodivisor of M2 can be viewed as a nonzerodivisor of M1 , as M1 may be
viewed as a submodule of M2 .
Case 2. Suppose depth(M2 ) = 0. Then we have to show that either depth(M1 ) = 0 or
depth(M3 ) = 0. Suppose not, and let (by prime avoidance) f ∈ R be a nonzerodivisor of
M1 and M3 . Therefore, the multiplication by f on M1 and M3 has a trivial kernel. From
the Snake Lemma it follows that multiplication with f on M2 has trivial kernel, so f is
a nonzerodivisor of M2 . But as depth(M2 ) = 0 this cannot be the case, contradiction.
Case 3. Suppose depth(M3 ) = 0. We have to show that if, moreover, depth(M2 ) > 0,
then depth(M1 ) = 1. Take a nonzerodivisor f of M2 . As M1 is a submodule of M2 it is
also a nonzerodivisor of M1 . Hence, depth(M1 ) ≥ 1. By the Snake Lemma we obtain an
inclusion
(6.8) 0 −→ M ′ −→ M1 /f M1 ,
·f
where M ′ := Ker(M3 −→ M3 ). As depth(M3 ) = 0, f is a zerodivisor. Thus M ′ 6= 0. One
even has that there exists an element g ∈ M3 \ {0} with mg = 0. Hence, in particular, g ∈
M ′ . Any nonzerodivisor of M1 /f M1 would give a nonzerodivisor of M ′ by the inclusion
(6.8). But M ′ cannot have one by definition of g ∈ M3 . Therefore, depth(M1 /f M1) = 0,
and depth(M1 ) = 1.

Corollary 6.5.19. Let I,J ⊂ On be radical ideals. Suppose that (X, 0) = V (I), 0 and
(Y, 0) = V (J), 0 both have dimension k. Suppose (X, 0) intersects (Y, 0) properly, that
is, dim (X ∩ Y ), 0 ≤ k − 1. Suppose (X, 0) and (Y, 0) are Cohen-Macaulay. Then
(X ∪ Y, 0) is Cohen-Macaulay ⇐⇒ (X ∩ Y, 0) is Cohen-Macaulay of dimension k − 1.
Proof. Look at the exact sequence:
0 −→ On /I ∩ J −→ On /I ⊕ On /J −→ On /(I + J) −→ 0.
From the assumption dim(X ∩ Y, 0) ≤ k − 1 it follows that depth(On /(I + J)) ≤
k − 1. Because (X, 0) and (Y, 0) are Cohen-Macaulay of dimension k, we know that
depth(On /I ⊕ On /J) = k. The Depth Lemma gives that
depth(On /(I ∩ J)) = depth(On /(I + J)) + 1,
which can be interpreted as the statement of the theorem.
Theorem 6.5.20 (Auslander-Buchsbaum Formula). Let (R,m) be a Noetherian local

ring and M be a finitely generated R–module of finite projective dimension. Then:
depth(M ) + pdR (M ) = depth(R).
Proof. The proof is by induction on depth(R) and depth(R) − depth(M ).

Step 1. Suppose first that depth(R) = 0. Then we have to show that M is free, as
pdR (M ) = depth(M ) = 0 follows from it. As depth(R) = 0, there exists a nonzero x ∈ R
with x · m = 0. Now let
nα αn−1 1 α
0 −→ Fn −−→ Fn−1 −−−→ · · · −→ F0 −→ M −→ 0
be a minimal free resolution of M as an R–module. Suppose n > 0. We can view Fn

as a submodule of Fn−1 via αn . Because of the minimality, Fn ⊂ mFn−1 . Then xFn =
xmFn−1 = 0, hence Fn = 0 because Fn is free, a contradiction. It follows that pdR (M ) =
0, hence M is free.
Step 2. Suppose that depth(M ) = depth(R). Then we can find a maximal regular se-
quence f1 , . . . ,fs for both M and R. As M is free if and only M/xM is free for x an
nonzerodivisor, it follows that M is free. Hence pdR (M ) = 0.
Step 3. For the induction step we consider two cases.
Case 1. Suppose depth(M ) > 0 and depth(R) > 0. This case is done by induction on
depth(R). Take an element f ∈ m which is a nonzerodivisor of M and of R. Take a
presentation:
0 −→ K −→ F −→ M −→ 0.
From the Snake Lemma:
0 −→ K/f · K −→ F/f · F −→ M/f · M −→ 0
is exact. Because the projective dimension stays the same if one divides out a nonzero-
divisor, see 6.5.17, and the depth drops by one if one divides out a nonzerodivisor, the
theorem follows by induction on depth(R).
Case 2. Finally, suppose that depth(M ) = 0 and depth(R) > 0. The case pdR (M ) = 0 is
trivial, as M is then a free R–module. Otherwise, consider a minimal a presentation
0 −→ K −→ F −→ M −→ 0
with F free and K 6= 0. Then pdR (K) = pdR (M )−1. Since depth(M ) = 0, the inequality
in the Depth Lemma is strict, so that depth(K) = depth(M ) + 1 = 1. As depth(R) −
depth(K) = depth ∗R) − depth(M ) − 1, the Auslander Buchsbaum Formula holds for
K by induction, so depth(K) + pdR (K) = depth(R). Hence depth(M ) + pdR (M ) =
depth(R).

Corollary 6.5.21. Let I ⊂ C {x1 , . . . ,xn } = R be an ideal, (X, 0) = V (I), 0 . Consider
a minimal free resolution of R/I as R–module
0 −→ Fk −→Fk−1 −→ · · · −→F0 = R−→R/I −→ 0.

Then depth (X, 0) = n − k. In particular, (X, 0) is Cohen-Macaulay if and only if
k = n − dim (X, 0) .
This gives a possibility of checking on a computer whether a singularity, defined by

polynomials, is Cohen-Macaulay.
Example 6.5.22. Take the union of two planes in C 4 intersecting in a point again. The
local ring is R := C {x,y,z,w}/(xz,xw,yz,yw). A free resolution of R as O := C {x,y,z,w}–
module is given by
0 0 01
y1 w y 0
@ −w A @ −z 0 y 0A
z 0 −x 0 w
−x 0 0 −x −z (xz xw yz yw)
0 −→ O −−−−−→ O −−−−−−−−−−−−→ O 4 −−−−−−−−−−→ O −→ R −→ 0,
4
which shows that R as a O–module has projective dimension three. It follows from the
Auslander-Buchsbaum Formula that, as O–module, R has depth one. It follows that R
(as R–module) has depth one. We see again that R is not Cohen–Macaulay.
The following corollary of the Auslander-Buchsbaum formula will be used in the
proof of the Hilbert-Burch Theorem 6.5.26.
Corollary 6.5.23. Let (R,m) be a local ring and M be an R–module. Then

pdR (M ) ≥ depth Ann(M ),R .
Proof. We may assume that pdR (M ) < ∞. The proof is by induction on depth(M ). If
depth(M ) = 0, then, by the Auslander Buchsbaum Formula, pdR (M ) = depth(R) ≥
depth Ann(M ),R .
We now assume that depth(M ) > 0. In particular, from the Auslander-Buchsbaum
Formula, it follows that d := depth(R) > 0. Let 0 −→ Fn −→ · · · −→ F0 −→ M −→ 0
be a minimal free resolution of M . Choose a regular sequence x1 , . . . ,xd in R such that
xr+1 , . . . ,xd is a maximal regular sequence in Ann(M ). Since depth(M ) > 0, there exists
a nonzerodivisor of M . Thus in particular x1 is a nonzerodivisor of R/ Ann(M ). Hence
we have that r ≥ 1 and, by renumbering, we may suppose x1 is a nonzerodivisor of M .
We obtain, using 1.2.35, the exact sequence
0 −→ Fn /x1 Fn −→ · · · −→ F0 /x1 F0 −→ M/x1 M −→ 0.
By induction on depth(M ) we obtain

pdR/(x1 ) (M/x1 M ) ≥ depth Ann(M/x1 M ),R/(x1 ) .

Now pdR (M ) = pdR/(x1 ) (M/x1 M ) because of 6.5.17 and depth Ann(M/x1 M ),R/(x1 ) ≥

depth Ann(M ),R because of the special choice of x1 , . . . ,xd .
Finally, in this chapter we will prove the Hilbert-Burch Theorem. The proof needs
some preparations. First of all we need a characterization for a complex of type
ϕn ϕn−1 ϕ1
0 −→ Fn −→ Fn−1 −→ · · · −→ F0
being exact. We start with a lemma.

ϕn ϕ1
Lemma 6.5.24. Let R be a Noetherian local ring and 0 −→ Pn Fn −→ · · · −→ F0 be
a complex4 of finitely generated free R–modules. Let rν = j=ν (−1)j−ν
rank(Fj ) for
ν = 0, . . . ,n. Then the induced sequence
4 Note that there is no “−→ 0” at the right hand side of the complex.
(6.9) 0 −→ Fn ⊗R Q(R) −→ · · · −→ F0 ⊗R Q(R)

is exact if and only if depth Irν (ϕν ),R ≥ 1 for all ν. Moreover, if this is the case, then
Im(ϕν ) has rank rν for all ν.
Proof.
Step 1. First suppose that (6.9) is exact. To prove that depth Irν (ϕν ),R ≥ 1 means to
prove that Irν (ϕν ) contains a nonzerodivisor, which means that Irν (ϕν ) ⊗ Q(R) = Q(R).
ϕn ϕ1
Therefore, we may suppose that R = Q(R), and that 0 −→ Fn −→ · · · −→ F0 is exact,
and we have to prove that Irν (ϕν ) = R. For j = n, . . . ,1 we have exact sequences
0 −→ Im(ϕj ) −→ Fj−1 −→ Im(ϕj−1 ).
We proceed by descending induction. First of all Fn has rank rn , and thus, as ϕn is injec-
tive, Im(ϕn ) has rank rn . From 1.4.32 it follows that Irn (ϕn ) = R, and Irn +1 (ϕn ) = (0).
From 1.3.8 it follows that Im(ϕn−1 ) has rank rn−1 . Again 1.4.32 implies that Irn−1 (ϕn−1 ) =
R and Irn−1 +1 (ϕn−1 ) = (0). The induction step is proved in a completely similar manner.
Step 2. To prove the other direction, assume that depth(Irν (ϕν ),R) ≥ 1 for all ν. This
condition gives that Irν (ϕν ) contains a nonzerodivisor. As a nonzerodivisor is a unit
in Q(R), we get Irν (ϕν ) ⊗ Q(R) = Q(R). Therefore, we may assume that R = Q(R)
and thus that Irν (ϕν ) = R. Under these conditions we have to show exactness, that is,
Im(ϕν+1 ) = Ker(ϕν ) for all ν, and that Im(ϕν ) is free of rank rν . By applying 1.3.8 the
freeness of Im(ϕν ) follows if we moreover show that Irν +1 (ϕν ) = (0).
We will proceed by descending induction on ν. So, assume that we have already
proved that Irj +1 (ϕj ) = (0) for j ≥ ν + 1, and that the complex
0 −→ Fn −→ · · · −→ Fν+1 −→ Fν
is exact. Now consider
Fν+1 −→ Fν −→ Coker(ϕν+1 ) −→ 0.
By assumption we have Irν+1 (ϕν+1 ) = R, and by 1.3.8 it follows that Im(ϕν+1 ) is free,
necessarily of rank rν+1 . Moreover Proposition 1.3.8 says that
Fν = Im(ϕν+1 ) ⊕ Fν′
for a free module Fν′ . It easily follows from the definition of the rj that the rank of Fν′
is equal to rν . Since Im(ϕν+1 ) ⊂ Ker(ϕν ) it follows from the definition of Fitting ideals
that
(R =)Irν (ϕν ) = Irν (ϕν|Fν′ ).
Because of size reasons (rank(Fν′ ) = rν ), Irν +1 (ϕν|Fν′ ) = 0 and therefore Irν +1 (ϕν ) =
(0). Again from Proposition 1.3.8 it follows that Im(ϕν ) has rank rν , and that ϕν|Fν′ is
injective. This exactly says that Im(ϕν+1 ) = Ker(ϕν ), which is what we had to prove.
Theorem 6.5.25 (Buchsbaum-Eisenbud). Let R be a Noetherian local ring, and let
ϕn ϕ1
(6.10) 0 −→ Fn −→ · · · −→ F0
Pn
be a complex of free R–modules of finite rank. Let rν = j=ν (−1)j−ν rank(Fj ) for ν =

0, . . . ,n. Then the complex (6.10) is exact if and only if depth Irν (ϕν ),R ≥ ν for ν =
1, . . . ,n.
Proof.
Step 1. Assume the complex is exact. We have to prove that depth Irν (ϕν ),R ≥ ν.
We use induction on d = depth(R).
The case d = 0 is not difficult. The exactness of the
complex gives that pdR Im(ϕν ) < ∞ for ν = 1, . . . ,n. This implies that pdR Im(ϕν ) =
0 (Auslander-Buchsbaum 6.5.20), that is, Im(ϕν ) is free and of rank rν . By 1.4.32 Irν (ϕν )
contains a nonzerodivisor. As depth(R) = 0, the only nonzerodivisors
are the units.
Therefore Irν (ϕν ) = R and, consequently, depth Irν (ϕν ),R = ∞.
Now assume that d > 0. By Exercise 1.3.22 the sequence 0 −→ Fn ⊗Q(R) −→ · · · −→
F0 ⊗ Q(R) is exact. By the previous lemma depth Irν (ϕν ),R ≥ 1 for all ν. Therefore we
can find a nonzerodivisor x ∈ ∩nν=1Irν (ϕν ). If x is a unit then Irν (ϕν ) = R for all ν and
thus by definition depth Irν (ϕν ),R = ∞. Otherwise, x ∈ m is a nonzerodivisor, so that
by 1.2.35 we get that
0 −→ Fn /xFn −→ · · · −→ F1 /xF1

is exact. The induction hypothesis implies that depth Irν (ϕν ⊗ idR/(x) ),R/(x) ≥ ν − 1
for ν = 2, . . . ,n. This implies depth Irν (ϕν ),R ≥ ν because of 6.5.5 and the definition
of the Fitting ideals, see 1.2.18.
Step 2. For the converse we consider three cases.
Case 1. This is the case that d = depth(R) = 0. Then the only nonzerodivisors in R are
units. Thus R = Q(R) and the statement follows from the previous lemma.
Case 2. The case that d = depth(R) ≥ 2. This is proved by induction on the dimension
of R and on n. The case n = 0 is trivial, as obviously the sequence 0 −→ F0 is exact.
The case dim(R) ≤ 1 follows because then depth(R) ≤ 1, hence we are in the first case
or in the third case.
Now assume that dim(R) > 0 and n > 0. By induction hypothesis we may suppose
that
0 −→ Fn −→ · · · −→ F1
is exact. Let M = Im(ϕ2 ) and K = Ker(ϕ1 ). It remains to prove that K/M = 0. Let
p 6= m be a prime ideal. By Exercise 1.3.22 the localization of the sequence
0 −→ Fnp −→ · · · −→ F1p −→ F0p
is exact. As dim(Rp ) ≤ dim(R) − 1, it follows from the induction hypothesis with respect
to the dimension of the ring that (K/M )p = 0 for all prime ideals p ⊂ R different from
the maximal ideal m. By Exercise 1.4.54 this implies that a certain power of m annihilates
K/M . Thus K/M is a module of dimension zero. Hence depth(K/M ) = 0.
We consider the exact sequences
(6.11) 0 −→ Im(ϕν ) −→ Fν−1 −→ Im(ϕν−1 ) −→ 0
for ν = 2, . . . , min{d,n}. Recall that d = depth(R). Suppose that n ≤ d. Then Fn =

Im(ϕn ) is free, and so has depth d, see Exercise 6.5.27, and thus certainly depth ≥ n.
Suppose on the other hand that j > d. By assumption depth(Irj (ϕj ),R) ≥ j. As R has
only depth d, it follows that Irj (ϕj ) = R. Following the first part of the proof of Lemma
6.5.24 we obtain that Im(ϕj ) is free of rank rj for j ≥ d, In particular, Im(ϕd ) has depth
at least d. Applying the Depth Lemma 6.5.18 successively to the exact sequences (6.11)
gives that depth(Im(ϕj )) ≥ j for j ≥ 2. As M = Im(ϕ2 ) we get depth(M ) ≥ 2. Using
that F1 has positive depth, and applying the Depth Lemma to the exact sequence
0 −→ M −→ F1 −→ F1 /M −→ 0,
gives that depth(F1 /M ) ≥ 1. Note that also depth(F1 /K) ≥ 1. This holds because we
have an embedding F1 /K ⊂ F0 . Now assume that K/M 6= 0. We apply the Depth Lemma
to the exact sequence
0 −→ K/M −→ F1 /M −→ F1 /K −→ 0.
As both F1 /M and F1 /K have positive depth, it follows that depth(K/M ) ≥ 1. This is

a contradiction, as we have already proved that depth(K/M ) = 0. Thus K/M = 0, and
the sequence is exact.
Case 3. This is the case d = depth(R) = 1. By assumption depth(Ir2 (ϕ2 ),R) ≥ 2. It
follows that Ir2 (ϕ2 ) = R. Following the proof of case 2, it follows that Im(ϕ2 ) is free, and
F1 ∼
= Im(ϕ2 ) ⊕ F1′
for a suitable free module F1′ (1.3.8). It remains to prove that the restriction of ϕ1 to F1′
is injective. This is trivial if F1′ = 0. By 6.5.24 we have that rank(F1′ ) = r1 . As in the
second case we have that Ker(ϕ1 |F1′ ) is annihilated by a power of the maximal ideal m
and thus, if Ker(ϕ1 |F1′ ) 6= 0, it follows that depth(Ker(ϕ1 |F1′ )) = 0, and we have an exact
sequence
0 −→ Ker(ϕ1 |F1′ ) −→ F1′ −→ Im(ϕ1 ) −→ 0
As F1′ has depth one, it follows from the Depth Lemma that depth(Ker(ϕ1 |F1′ )) > 0. This
is a contradiction. Therefore, ϕ1 |F1′ is injective and 0 −→ Fn −→ · · · −→ F0 is exact.
Theorem 6.5.26 (Hilbert-Burch). Let R be a Noetherian local ring and I ⊂ R be an

ideal.
2 α
1 α
(1) Assume that 0 −→ F2 −→ F1 −→ R −→ R/I −→ 0 is a minimal free resolution.
Then
(1.1) rank(F1 ) = rank(F2 ) + 1,
(1.2) depth(Irank(F2 ) (α2 ),R) = 2,
(1.3) I = aIrank(F2 ) (α2 ) for a suitable nonzerodivisor a ∈ R. More precisely the
i–th entry of the matrix for α1 is (−1)i a times the minor obtained from α2 by
leaving out the i–th row.

(2) Let α2 be a (n − 1) × n–matrix and suppose that depth In−1 (α2 ),R ≥ 2. Let a ∈ R
be a nonzerodivisor and define α1 as in the description of (1.3). Let I := aIn−1 (α2 ).
α2 α1
Then 0 −→ Rn−1 −→ Rn −→ R −→ R/I −→ 0 is a free resolution of R/I.
Proof. The statement (2) is an immediate consequence of 6.5.25, as r2 = n−1, and r1 = 1.

We now prove (1.3). Let α1 be the map defined by the 1 × n–matrix having as i–th entry
(−1)i times the minor obtained from α2 by deleting the i–th row. Then Laplace expansion
gives α1 ◦ α2 = 0. On the other hand, by definition,
I1 (α1 ) = Im(α1 ) = Irank(F2 ) (α2 ). By
6.5.25 we obtain that depth Irank(F2 ) (α2 ),R ≥ 2. Therefore, we can apply again 6.5.25
to the dual complex
α∗ α∗
1
0 −→ R −→ F1∗ −→
2
F2∗ ,
using that I1 (α∗1 ) = I1 (α1 ) = Irank(F2 ) (α2 ), to obtain that this complex is exact. Now
consider the commutative diagram
α∗ α∗
R
1
/ F1∗ 2
/ F2∗
σ

α∗ α∗
0 /R 1
/ F1∗ 2
/ F2∗
The map σ with α∗1 ◦σ = α∗1 exists because of the exactness of the second row. Because σ ∈
HomR (R,R) there exists an a ∈ R such that σ(x) = ax for all x ∈ R. It remains to prove
that a is a nonzerodivisor. So suppose x · a = 0 for some x ∈ R. This implies xI1 (α1 ) = 0.
Therefore, I1 (α1 ) consists of zerodivisors. This is not possible, as the sequence 0 −→
α2 α1
F2 −→ F1 −→ F0 is exact and thus depth I1 (α1 ),R ≥ 1 by Theorem 6.5.25.
To prove (1.1) note that we have now proved that I contains a nonzerodivisor. Hence
I ⊗ Q(R) = Q(R) and R/I ⊗ Q(R) = 0. We tensorize the exact sequence with Q(R) and
obtain by Exercise 1.3.22 an exact sequence
0 −→ F2 ⊗R Q(R) −→ F1 ⊗R Q(R) −→ Q(R) −→ 0.
This implies rank(F1 ) = rank(F2 ) + 1. It remains to prove (1.2). Put I ′ = Irank(F2 ) (α2 ).
By dividing by a we get an exact sequence
α1
α
2
0 −→ F2 −→ a
F1 −→ R −→ R/I ′ −→ 0.
Obviously I ′ = Ann(R/I ′ ), so that from 6.5.23 we obtain 2 ≥ pdR (R/I ′ ) ≥ depth(I ′ ,R).
On the other hand, by 6.5.25, we have depth(I ′ ,R) ≥ 2. This implies depth(I ′ ,R) = 2.
Exercises
6.5.27. Suppose M is a free R–module. Then depth(M ) = depth(R).
6.5.28. Let R ⊂ S be a finite ring extension of local rings, and M be an S–module.

(1) Let I ⊂ R be be an ideal. Show that (I · R) · M = (I · S) · M .
(2) Let f1 , . . . ,fs be a regular sequence of M as R–module. Show that f1 , . . . ,fs is a regular
sequence of M as S–module.
6.5.29. Let R ⊂ S be a finite ring extension of local rings, and M be an S–module. Show that
depthR (M ) = depthS (M ).
6.5.30. Let M,N be R–modules of positive depth. Show that there exists an f ∈ R which is a
nonzerodivisor for both M and N . Generalize this statement to finitely many modules.
(Hint: Use Exercise 1.4.52 and Prime Avoidance 1.1.13.)
6.5.31. The assumption on R to be local in Lemma 6.5.3 is an essential assumption! Show that
in the ring C [x,y,z]/(xz − z) the sequence x,xy − y is a regular sequence, but xy − y,x is not a
regular sequence.
6.5.32. Consider the map f (s,t) = (s,st,t2 ,t3 ) = (x,y,z,u) of Exercise 2.3.18. The image is
given by the ideal I = (yz − ux,y 2 − zx2 ,z 2 x − uy,z 3 − u2 ), as shown in that exercise. Let
R = C {x,y,z,u}. Show that
0 1
0 u1 y u xz z 2
B −z 0 −u 0 C
@ −y A @ A
−z −x z −y u
x 4 0 −x 0 −y (yz−ux,y 2 −zx2 ,z 2 x−uy,z 3 −u2 )
−−−−−−−−−−−−−−→ R4 −−−−−−−−−−−−−−−−−−−−−−−→ R −→ R/I −→ 0
0 −→ R −−−−−→ R −
is a minimal free resolution

` of´ I. Conclude by the Auslander-Buchsbaum formula that the germ
of the complex space V (I), 0 is not Cohen-Macaulay.
6.5.33. Let R = C {x,y,z},I

` =´ (xy,xz,yz). Compute a free resolution of R/I as R–module.
Conclude that (X, 0) = V (I), 0 is Cohen-Macaulay.
6.5.34. Let R = C `{x,y,z},I´ = (xy,xz). Compute a free resolution of R/I as R–module. Con-
clude that (X, 0) = V (I), 0 is not Cohen-Macaulay.
6.5.35. Let R = C {x,y}/(xy), and M = R e be the normalization of R. Show that M is not a

free R–module. Show by using the Auslander Buchsbaum formula, that M does not have finite
projective dimension.
269
7 Standard Bases
Let an ideal I ⊂ C {x} be given. In this chapter we will define and study standard bases
of the ideal I. The main idea here is that one wants to have a “good” representative for
any coset of I. In fact, the Weierstraß Division Theorem provides us with the simplest
example of a standard basis. Consider f ∈ C {x} which is regular of order b in xn . For any
g ∈ C {x} we have a unique representation g = qf + r with r a polynomial of degree less
than b. From the uniqueness statement in the Weierstraß Division Theorem, it follows
without difficulty that the remainder r only depends on the coset g + (f ) of (f ). Of
course, the remainder r depends on the choice of xn .
In order to generalize, we would like to have division with remainder through an
ideal. The “choice” of xn is now replaced by the choice of a so-called monomial ordering
on the monomials of C {x}. Having such a monomial ordering, one can talk about the
leading term L(f ) of f , and the ideal L(I) = (L(f ) : f ∈ I \ {0}) of leading terms of
I. A standard basis of I is a set of elements f1 , . . . ,fm of I such that L(f1 ), . . . ,L(fm )
generate L(I). It is quite trivially true that a standard basis exists. Here one uses only
the fact that C {x} is Noetherian. It also turns out that a standard basis of I generates
the ideal I. Standard bases of ideals in C {x} generated by polynomials can be computed
effectively. The algorithms are explained in the exercises. For example, the computer
algebra system SINGULAR can be used for computation (cf. [Singular 2000]).
The first thing one would have is a kind of division with remainder through the fi ,
that is, one would like for any f ∈ C {x} a representation
m
X
(7.1) f= qi fi + r
i=1
where no term of r is divisible by the leading term of any fi . This replaces the condition
that r is polynomial of degree less than b in the Weierstraß Division Theorem. In fact,
we will give alternative proofs of the Weierstraß Division and Preparation Theorems. In
order to control the convergence of the qi and r, we will define
P seminorms1 on C {x}
n α
or even C [[x]] depending on ρ ∈ R+ defined by kf kρ := α |fα |ρ . This is also the
norm which is used in the books of Grauert and Remmert, see [Grauert-Remmert 1971],
and [Grauert-Remmert 1984] to give proofs of the Weierstraß Division and Preparation
Theorem. It turns out that if f1 , . . . ,fm is a standard basis, then the r is uniquely
determined, and is called the normal form NF(f ) = NF(f |I) of f . It depends, of course,
on the monomial ordering. In particular, it follows that f ∈ I if and only if NF(f ) = 0.
This will be used in an essential way in the proof of Grauert’s Approximation Theorem
in Chapter 8. We will give some applications in Section 7.3.
More generally, there is a theory of standard bases of modules. As it is not more
difficult, we will treat the general case from the beginning.
1 Here semi means that the norm might take infinite value.
270 7 Standard Bases
7.1 The Division Theorem
Definition 7.1.1. Let M = {xα |α ∈ Zn+ } be the set of monomials in C [x]. A monomial
ordering is a total ordering on M (respectively Zn+ ) compatible with the semi–group
structure, that is, xα > xβ implies xα+γ > xβ+γ .
We assume furthermore that the orderings considered here are well–orderings, that
is, every set of monomials has a minimal element with respect to the ordering. This is
equivalent to the property that 1 is the smallest monomial (cf. Exercise 7.1.9).
Examples 7.1.2.
(1) The lexicographical ordering >

lex
α β
x > x if and only if α 6= β and there is an i ≥ 1 such that αj = βj for i > j and
lex
αi > βi .
(2) The degree lexicographical ordering >
deglex
Let w = (w1 , . . . ,wn ) ∈ Rn and wi > 0, i = 1, . . . ,n. For α ∈ Zn+ define |α|w =
P n
i=1 αi wi (we shall also simply use |α| = |α|w if w is fixed).
xα > xβ if and only if |α| > |β| or |α| = |β| and xα > xβ
deglex lex
Definition 7.1.3. Now let > be a monomial ordering. Let f ∈ C {x} be a power series.
Then we may write X
f= ai xαi
i≥1
such that ai 6= 0, ai ∈ C and xαi < xαi+1 for all i.

We define
(1) L(f ) = xα1 the leading monomial of f ;
(2) C(f ) = a1 the coefficient of the leading monomial;
(3) tail(f ) = f − C(f ) · L(f ).
Let I ⊂ C {x} be an ideal, then L(I) = h{L(f )|f ∈ I, f 6= 0}i, the leading ideal, is
the ideal generated by all leading monomials of non–zero elements of I.
Example:. Let f = 2x51 + x62 + x42 + x1 and I = (f ). With respect to the orderings intro-
duced before (with w = (1,1)), we obtain for L(f ), C(f ), tail(f ) and L(I) the following
(the monomials of tail(f ) always ordered with respect to the monomial ordering):
c(f ) L(f ) L(I) tail(f )
lex 1 x42 hx42 i x62 + x1 + 2x51

deglex 1 x1 hx1 i x42 + 2x51 + x62
We wish to define orderings also for the monomials of the free module C {x}N . For a
compatible notation we choose, as a basis, the vectors (0, . . . ,1,0 . . . 0) and identify them,
i
as before, with xi−1
n+1 , xn+1 a new variable.
7.1 The Division Theorem 271
PN −1
Definition 7.1.4. Let < be a monomial ordering on C {x}. Let C {x}N = i=0 C {x}xin+1
and M = {xα |α ∈ Zn+1 N
+ ,0 ≤ αn+1 ≤ N −1} be the set of monomials of C {x} . A module
ordering <m is a total ordering of M compatible with the ordering < on C {x}, that is
(1) f,f ′ ∈ M and f <m f ′ implies xα f <m xα f ′ .
(2) f ∈ M and xα < xβ implies xα f <m xβ f .
Exercise 7.1.11 gives several examples for module orderings.
From now on in this chapter we use the following ordering:
Let w = (w1 , . . . ,wn+1 ) ∈ Rn+1
+ be fixed and < be the degree lexicographical ordering.
For xα and xβ ∈ M we define xα < xβ if and only if |α| < |β| or if |α| = |β| then α < β
with respect to the lexicographical ordering.
As in Definition 7.1.3 we define for f ∈ C {x}N the leading monomial L(f ), the
leading coefficient C(f ), the tail, tail(f ) and the leading module L(I) for a submodule
I ⊂ C {x}N .
Let m,m′ ∈ M be two monomials, m = xα , m′ = xβ . We define m | m′ if αn+1 =
βn+1 and xα | xβ .
The aim of this Chapter is to obtain information about ideals by studying their lead-
ing ideals (respectively modules) which are generated by monomials and are, therefore,
simpler.
Definition 7.1.5. Let I ⊂ C {x}N be a submodule, S = {f1 , . . . ,fm } ⊂ I is a standard
basis of I if L(I) = L(f1 ), . . . ,L(fm ) .
A standard basis is reduced if C(fi ) = 1 for all i and L(fi ) does not divide any
monomial occurring in fj for i 6= j and occurring in tail(fi ).
Example:. Let I = (x3 + y 2 ,y), w = (1,2), then {x3 + y 2 ,y,y 5 } is a standard basis of I
in C {x} but not a reduced standard basis of I, {x3 ,y} is a reduced standard basis.
Remark:. Standard bases were introduced by Hironaka (cf. [Hironaka 1964]), Grauert
(cf. [Grauert 1972]) for the power series case and Buchberger (cf. [Buchberger 1965]) for
polynomial rings. In case of polynomial rings, they are called Gröbner bases.
Proposition 7.1.6. Let I ⊂ C {x}N be a submodule, then I has a standard basis.
Proof. L(I) is, by definition, the submodule in C {x}N generated by the leading monomi-
als of the elements of I. C {x} is Noetherian and, therefore L(I) is generated by finitely
many such leading monomials, say xα1 , . . . ,xαm . Now choose f1 , . . . ,fm ∈ I such that
L(fi ) = xαi then {f1 , . . . ,fm } is a standard basis of I.
Later we shall see that a standard basis of I generates the module I.
We shall also prove that reduced standard bases exist and are uniquely determined.
Therefore, we need the notion of a normal form of an element f ∈ C {x}N with respect
to a set S = {f1 , . . . ,fm } of elements of C {x}N , that is, a representation in C {x}N
m
X
f= qi fi + r
i=1
such that no monomial of r is divisible by L(fi ) for all i.

The following result about the normal form is also a generalization of the Weierstraß
division Theorem and, therefore, also often called division Theorem. It is the essential
Theorem of this section with many consequences. We will need a quite general version.
The case s = 0 in the Theorem is the most important case, however.
Theorem 7.1.7 (Grauert’s Divison Theorem).

Let s ≥ 0 and f1 , . . . ,fm ∈ C {xs+1 , . . . ,xn }N , then for every f ∈ C {x}N there exists
a representation
X m
f= qν fν + r
ν=1
with the following properties.
(1) No monomial of r is divisible by L(fr ),ν = 1, . . . ,m;
(2) L(qν fν ) ≥ L(f ),ν = 1, . . . ,m2

(3) Let f ∈ Bρn for some ρ ∈ Rn+1 n+1
+ . For every ε > 0 there exist ρ ∈ R+ such that
• ρ ≤ ρ and ρi = ρi for i = 1, . . . ,s,

• for 0 < λ ≤ 1 let λ ◦ ρ = (ρ1 , . . . ,ρs ,λws+1 ρs+1 , . . . ,λwn+1 ρn+1 ), then
1
krkλ◦ρ ≤ kf kλ◦ρ
1−ε
1
kqν kλ◦ρ kL(fν )kλ◦ρ ≤ kf kλ◦ρ .
1−ε
Proof.
Special case: fi = xαi , i ∈ {1, . . . ,m}, let ∆ = hα1 , . . . ,αm i ⊂ Zn+1
+ be the Zn+ –semi–
module generated by α1 , . . . ,αm . We define inductively a partition of ∆ by
∆1 = hα1 i
∆i = hαi i ∩ (∆ r hα1 , . . . ,αi−1 i).
P
Let f = fα xα , we define
X
r := fα xα
α∈∆
/
1 X
qi = fα xα
xαi
α∈∆i
P
then f = qi fi + r. Let kf kρ < ∞, we choose ρ = ρ, then krkρ ≤ kf kρ and kqi kρ kfi kρ ≤
kf kρ . We shall use the construction of this step in the next one. Therefore, we define for
S = {xα1 , . . . ,xαm } being fixed maps
qi : C {x}N −→ C {x}N
r : C {x}N −→ C {x}N
1 P P P
by qi (f ) = x αi α∈∆i fα xα resp. r(f ) = α∈∆
/ fα xα and obtain f = qi (f )xαi + r(f ).
General case: Let L(fi ) = xαi and assume C(fi ) = 1. Let S = {xα1 , . . . ,xαm }. We define
a sequence νi ∈ C {x}N by
2 This condition is a kind of minimality condition. Note that in a sum one can always add two terms
which cancel. The condition written here, avoids this phenomenon.
7.1 The Division Theorem 273
ν0 = f
m
X
νj+1 = νj − qν (νj )fν − r(νj ).
ν=1
P
∞
The idea of the proof is now to define ν = νj and prove that
j=0
P
• f= qj (ν)fj + r(ν).
• ν ∈ C {x}N .
Lemma 7.1.8. Let g1 , . . . ,gm ∈ C {xs+1 , . . . ,xn }N for some s ≥ 0. Then, for every ε > 0
and ρ ∈ Rn+1 , there exist ρ ∈ Rn+1
+ such that ρ ≤ ρ and ρi = ρi for i = 1, . . . ,s and for
all λ, 0 < λ < 1,
k tail(gi )kλ◦ρ < εkL(gi )kλ◦ρ , i = 1, . . . ,r.
We shall prove the Lemma later.
ε
Now let 1 > ε > 0 be given. We apply the Lemma to f1 , . . . ,fm and m and choose
ε αi
ρ such that ktail (fP)k
i λ◦ρ < m (λ ◦ ρ) and ρ ≤ ρ and ρ i = ρ i for i =
Pm1, . . . ,s.
m
Now, kνj+1 kλ◦ρ ≤ i=1 k tail(fi )kλ◦ρ kqi (νj )kλ◦ρ because νj+1 = − i=0 qi (νj ) tail(fi ).
1
By definition of the qi we have kqi (νj )kλ◦ρ ≤ (λ◦ρ)αi kνj kλ◦ρ and, therefore, kνj+1 kλ◦ρ ≤
εkνj kλ◦ρ .
Pi
Let si := j=0 νj then
i
X i
X
ksi kλ◦ρ = k νj kλ◦ρ ≤ kνj kλ◦ρ
j=0 j=0
i
X
≤ εj kν0 kλ◦ρ
j=0
1
≤ kf kλ◦ρ .
1−ε
This implies
1
kνkλ◦ρ ≤ kf kλ◦ρ .
1−ε
Especially,
kxαi kλ◦ρ · kqi (ν)kλ◦ρ ≤ kνkλ◦ρ
and
kr(ν)kλ◦ρ ≤ kνkλ◦ρ
implies
1
kL(fi )kλ◦ρ kqi (ν)kλ◦ρ ≤ kf kλ◦ρ
1−ε
1
kr(ν)kλ◦ρ ≤ kf kλ◦ρ .
1−ε
It remains to prove that

m
X
f= qi (ν)fi + r(ν) and L qi (ν)fi ≥ L(f ).
i=1
Now
∞
X
f= (νj − νj+1 )
j=0
X∞ m
X
= νj − νj + qℓ (νj )fℓ + r(νj )
j=0 ℓ=1
Xm X∞ ∞
X
= qℓ (νj )fℓ + r(νj ).
ℓ=1 j=0 j=0
P∞ P∞
But j=0 qℓ (νj ) = qℓ (ν) and j=0 r(νj ) = r(ν) and, therefore, the first part is proved.
P∞
Now L qi (ν)fi = L j=0 qi (νj )fi ≥ L qi (ν0 )fi = L qi (f )fi because L(νj+1 ) >
L(νj ). Therefore,
L qi (ν)fi ≥ L qi (f )fi ≥ L(f ).
This proves the Theorem.
Proof of the Lemma. Because the g1 , . . . ,gm do not depend on x1 , . . . ,xs we have a free
choice for ρ1 , . . . ,ρs . So we may assume that s = 0.
P (i)
Let ε > 0 be given and assume that C(gi ) = 1 for all i. Let gi = α gα xα and
(i)
L(gi ) = xα . We have to prove that there exist ρ ∈ Rn+1 + such that for all 0 < λ < 1
(i)
sup{(2|α| + 2)n+2 | gα(i) | (λ ◦ ρ)α } < ε(2|α(i) | + 2)n+2 (λ ◦ ρ)α .
Now (λ ◦ ρ)α = λ|α| · ρα and, therefore, it is enough to find a ρ ∈ Rn+1

+ such that
(i)
|gα(i) | ρα < ερα for |α| = |α(i) | and i = 1, . . . ,m.
(i) (i)
Now |α| = |α(i) | and xα > xα implies xα > xα with respect to the lexicographical
(i) (i)
ordering, that is, for α = (α1 , . . . ,αn+1 ) and α(i) = (α1 , . . . ,αn+1 ) exist k such that
(i) (i) (i)
(∗) α1 = α1 , . . . ,αk−1 = αk−1 and αk > αk .
(i)
We choose k ∈ {1, . . . ,n + 1} maximal such that for at least one i and one α with gα 6= 0
(*) holds.
We choose any ρk+1 , . . . ,ρn+1 ∈ R+ smaller than 1 and choose ρk such that
(i) (i) (i)
(i) α α α α α
|gα | ρα n+1 n+1 k+1 n+1 (i)
k ρk+1 · . . . · ρn+1 < ερk ρk+1 · . . . · ρn+1 for all i and α such that |α| = |α |,
k k
(i)
gα 6= 0 and (*) holds, that is,
α −αk
(i)
ε α
(i)
−αk+1 α
(i)
−αn+1
(∗∗) ρk k < (i)
k+1
· ρk+1 n+1
· . . . · ρn+1 .
|gα |
7.2 Characterizations and Properties of Standard Bases 275
Assume ρℓ , . . . ,ρn+1 are already defined. We choose k ∈ {1, . . . ,ℓ − 1} maximal such

(i)
that for at least one i and one α with gα 6= 0 (∗) holds. We choose any ρk+1 , . . . ,ρℓ−1 ∈
R+ smaller than 1 and choose ρk such that (∗∗) holds. If such k does not exist we choose
any ρ1 , . . . ,ρℓ−1 ∈ R+ smaller than 1. Let ρ = (ρ1 , . . . ,ρn+1 ) then, by construction, we
(i) (i)
have |gα | ρα < ερα for |α| = |α(i) |.
This proves the Lemma.
Exercises
7.1.9. Prove that a monomial ordering is a well–ordering if and only if 1 is the smallest mono-
mial.
7.1.10. Give an algorithm which constructs in C {x} the functions q and r up to a given order.
7.1.11 (Examples for module orderings). Let < be a monomial ordering on C {x} and C {x}N =
PN−1 k α n+1
i=0 C {x}xn+1 be the identification as in Definition 7.1.4 and M = {x |α ∈ Z+ , 0 ≤ αn+1 ≤
N − 1}.
Let w = (w1 , . . . ,wn+1 ) ∈ Rn+1 , wn+1 6= 0. For xα and xβ ∈ M we define xα <m xβ if and
only if
P Pn+1
• |α|w = n+1
i=1 αi wi < i=1 βi wi = |β|w or
• |α|w = |β|w and xα αn β1 βn
1 · . . . · xn < x1 · . . . · xn .
1
Prove that <m is a module ordering. Notice that the following module orderings are often used:
(1) degree orderings w1 > 0, . . . ,wn+1 > 0;
(2) elimination orderings w1 = · · · = wn = 0.
7.1.12. For xα and xβ ∈ M we define xα <m xβ if and only if
• x1α1 · . . . · xα β1
n < x1 · . . . · xn or
n βn
• if (α1 , . . . ,αn ) = (β1 , . . . ,βn ) then αn+1 < βn+1 .

Prove that <m is a module ordering.
7.1.13. Let <m be an elimination ordering (7.1.11 (2)) with w1 · · · = wn = 0, wn+1 = 1. Let
I ⊂ C {x}N be aPsubmodule and S a standard basis of I. P
Prove that S ∩ ti=0 C {x}xin+1 is a standard basis of I ∩ ti=0 C {x}xin+1 for all t ≤ N − 1.
7.2 Characterizations and Properties of Standard Bases
Definition 7.2.1. Let S = {f1 , . . . ,fm } ⊂ C {x}N be Pan ordered set. For f ∈ C {x}
N
we can, by Grauert’s Division Theorem ??, write f = qi fi + r, such that no monomial

of r is divisible by any of the L(fi ). The r constructed in the proof of Grauert’s Division
Theorem is uniquely determined with respect to the ordered set S, see 7.1.7. This r
is called the normal form of f with respect to the (ordered) set S. We will write r =
NF(f |S), and if S is fixed, just NF(f ).
Remark 7.2.2. In general, the normal form depends on the ordering of S. For example,
f = x2 y + xy 2 + y 2 , f1 = y 2 − 1, f2 = xy, S1 = {f1 ,f2 } and S2 = {f2 ,f1 }. Then for the
degree lexicographic ordering with w = (1,1) we get
v = f +x v = f +1
q1 (v) = x+1 q1 (v) = x+y
q2 (v) = x q2 (v) = 1
r(v) = NF(f |S1 ) = x+1 r(v) = NF(f |S2 ) = 1.
We will show that if S is a standard basis however, then the normal form does not
depend on the ordering. First the definition.
Definition 7.2.3. Let I ⊂ C {x}N be a submodule. Then S = {f1 , . . . ,fm } ⊂ I is called
a standard basis of I if L(I) = L(f1 ), . . . ,L(fm ) . A standard basis is called reduced if
C(fi ) = 1 for all i and L(fi ) does not divide any monomial occurring in fj for i 6= j and
occurring in Tail(fi ).
Example 7.2.4. Let I = (x3 + y 2 ,y), w = (1,2). Then {x3 + y 2 ,y,y 5 } is a standard
basis of I in C {x} but not a reduced standard basis of I. The standard basis {x3 ,y} is
reduced.
Remark 7.2.5. Standard bases were introduced by Hironaka (cf. [Hironaka 1964]),
Grauert (cf. [Grauert 1972]) for the power series case and, for polynomial rings, by
Buchberger (cf. [Buchberger 1965]). In case of polynomial rings, they are also called
Gröbner bases. The idea of Gröbner bases is already contained in the work of Gordan,
see [Gordan 1899].
We now give some basic results.
Proposition 7.2.6. Let I ⊂ C {x}N be a submodule. Then I has a standard basis.
Proof. The module L(I) is, by definition, the submodule in C {x}N generated by the
leading monomials of the elements of I. The ring C {x} is Noetherian and, therefore
L(I) is generated by finitely many such leading monomials, say xα1 , . . . ,xαm . Choose
f1 , . . . ,fm ∈ I such that L(fi ) = xαi . Then {f1 , . . . ,fm } is a standard basis of I.
Corollary 7.2.7. Let S = {f1 , . . . ,fm } be a standard basis of I. Then I = (f1 , . . . ,fm ).
P
Proof. For f ∈ I we have f = qi fi + NF(f |S). This implies
that NF(f |S) ∈ I. If
NF(f |S) 6= 0, then L NF(f |S) ∈ L(I) = L(f1 ), . . . ,L(fm ) , which is a contradiction to
the definition of NF(f |S).
Corollary 7.2.8. Let S = {f1 , . . . ,fm } and T = {g1 , . . . ,gs } be standard bases of I.
Then for all f ∈ C {x}N we have NF(f |S) = NF(f |T ).
P P
Proof. f = qν fν + NF(f |S) = qν gν + NF(f |T
) implies NF(f |S) − NF(f |T ) ∈ I. If
NF(f |S) 6= NF(f |T ), then L NF(f |S) − NF(f |T ) ∈ L(I) is a monomial which appears
in NF(f |S) or in NF(f |T ) which is impossible by definition of NF.
This result makes the following definition possible.
Definition 7.2.9. Let I ⊂ C {x}N be a submodule and S a standard basis of I. Then for
any f ∈ C {x}N we define NF(f |I) = NF(f |S) to be the normal form of f with respect
to I.
Corollary 7.2.10. Let I be a submodule of C {x}N , and S = {f1 , . . . ,fm } be a standard
basis of I. Let f ∈ C {x}N . Then
(1) f ∈ I if and only if NF(f |S) = 0,
(2) f ∈ I if and only if NF(f |I) = 0,
(3) f − N F (f |I) ∈ I.
P
Proof. We know that f = qν fν +NF(f |S) and I = f1 , . . . ,fm . Therefore, NF(f |S) =
0 implies f∈ I. On the other hand, f ∈ I implies NF(f |S) ∈ I. If NF(f |S) 6= 0 then
L NF(f |S) ∈ L(I) which is not possible by the definition of NF.
Corollary 7.2.11. Let I ⊂ C {x}N be a submodule. Then I has a reduced standard
basis S = {f1 , . . . ,fm }, with the property L(f1 ) < · · · < L(fm ). Such a standard basis is
uniquely determined.
Proof.
Step 1. We first prove the existence. Let T = {g1 , . . . ,gm } be a standard basis of I,
and NF the normal form with respect to T . After reordering, and leaving out some of
the gi , we may assume L(g1 ) < · · · < L(gm ). Suppose that L(gi )|L(gj ) for some i 6= j.
Then {g1 , . . . ,gj−1 , gj+1 , . . . ,gm } is also a standard basis for I. Thus we may assume that
L(gi ) ∤ L(gj ) for i 6= j. Furthermore, dividing gi by C(gi ) we may assume that C(gi ) = 1
for all i. We define

fi := L(gi ) + NF Tail(gi )|I , i = 1, . . . ,m.
We claim Pthat f1 , . . . ,fm is a reduced

standard basis of I. Because gi = L(gi )+ Tail(gi ) =
L(gi ) + qiν gν + NF Tail(gi )|T , the fi are in I. Because L(gi ) = L(fi ) we obtain that
S := {f1 , . . . ,fm } is a standard basis of I. Furthermore, L(fi ) ∤ L(fj ) for i 6= j, and
by definition of normal form, no term of Tail(fi ) is divisible by any of the L(g1 ) =
L(f1 ), . . . ,L(gm ) = L(fm ). Thus f1 , . . . ,fm is a reduced standard basis.
Step 2. We now show uniqueness. Let S = {f1 , . . . ,fm } and T = {g1 , . . . ,gs } be two
reduced standard bases and L(f1 ) < · · · < L(fm ), L(g1 ) < · · · < L(gs ).
We first prove that s = m and L(fi ) = L(gi ). Suppose the converse. Now L(I) =
i ) ∤ L(fj ) for i 6= j (S is a reduced standard basis). Similarly
L(f1 ), . . . ,L(fm ) and L(f
L(I) = L(g1 ), . . . ,L(gs ) , L(gi ) ∤ L(gj ) for i 6= j. Assume that for some i ≥ 1 L(gj ) =
L(fj ) for j < i and L(gi ) 6= L(fi ). By symmetry we may assume L(gi ) < L(fi ).As they
are both standard basis, it follows that L(g1 ), . . . ,L(gs ) = L(f1 ), . . . ,L(fm ) . Hence
L(fν ) | L(gi ) for some ν. There are two cases.
(1) If ν ≥ i, then L(fν ) ≥ L(fi ). Furthermore, because of L(gi ) < L(fi ) we get L(gi ) <
L(fν ). Thus it is not possible that L(fν ) | L(gi ), contradiction.
(2) If ν < i, we have, by choice of i, L(fν ) = L(gν ). But as we have a reduced standard
basis L(gν ) ∤ L(gi ) for ν 6= i. This gives a contradiction because we had L(fν ) |
L(gi ).
So we proved that s = m and L(gi ) = L(fi ) for all i. Now consider fi − gi =
Tail(fi ) − Tail(gi ) ∈ I. If Tail(fi ) 6= Tail(gi ) then L Tail(fi ) − Tail(gi ) is a monomial in
L(I) and a monomial appearing in Tail(fi ) or Tail(gi ), which is a contradiction to the
property of both standard bases being reduced.
We will now give characterization of standard bases which is the theoretical back-
ground for Buchberger’s algorithm to compute standard bases.
Definition 7.2.12. Let f, g ∈ C {x}N , with L(f ) = xβ and L(g) = xγ .
α
(1) We define the least common multiple lcm L(f ),L(g) to be the monomial x with
α = max(β1 ,γ1 ), . . . , max(βn ,γn ),βn+1 if βn+1 = γn+1 and to be 0 if βn+1 6= γn+1 .
(2) Let m = lcm(L(f ),L(g). Then we define the syzygy polynomial of f and g by
m C(f ) m
spoly(f,g) = ·f − · · g.
L(f ) C(g) L(g)
(Note that in case m 6= 0 we have L(spoly(f,g)) > m.)
Examples 7.2.13.
(1) Let f = 2xy 2 + x4 , g = 5x2 y + y 4 ∈ C {x,y} and w = (1,1), then
spoly(f,g) = x5 − 25 y 5 .
(2) Consider the free C {x1 ,x2 }–module C {x1 ,x2 }2 = C {x1 ,x2 } ⊕ C {x1 ,x2 }x3 , and
w = (1,1,1) Let f = (x21 + 1,x2 ) = 1 + x21 + x2 x3 and g = (x2 ,x31 ) = x2 + x31 x3 . Then
L(f ) = 1, and L(g) = x2 , so that
spoly(f,g) = x2 f − g = (x21 x2 ,x22 − x31 ) = x21 x2 + x22 x3 − x31 x3 .

Pm
Definition 7.2.14. Let f1 , . . . ,fm ∈ C {x}N . Write C [x]m = i=1 C [x] · ei , ei =
(0, . . . ,0,1,0, . . . ,0).
(1) P
A syzygy on the leading terms C(f1 )L(f1 ), . . . ,C(fm )L(fm ) is an element s =
m m
i=1 si ei = (s1 , . . . ,sm ) ∈ C [x] such that
m
X
s(f1 , . . . ,fm ) := si C(fi )L(fi ) = 0.
i=1
(2) Let syz(f1 , . . . ,fm ) be the submodule of C [x]m of all syzygies of the leading terms
of f1 , . . . ,fm .
(3) Write xα(i,j) = lcm(L(fi ),L(fj )). Then
xα(i,j) C(fi ) xα(i,j)

s(i,j) := ei − ej ∈ syz(f1 , . . . ,fm ).
L(fi ) C(fj ) L(fj )
(Note that s(i,j) · (f1 , . . . ,fm ) = spoly(fi ,fj ), where the dot denotes the dot prod-
uct.)
Lemma 7.2.15. Consider f1 , . . . ,fm ∈ C {x}N , and keep the notations of the previous
definition.
(1) The C [x]–module syz(f1 , . . . ,fm ) admits a Zn+1 –grading, that is, we can write
syz(f1 , . . . ,fm ) = ⊕ν∈Nn+1 Sν with
m
X
Sν := ci xα(i) ei ∈ syz(f1 , . . . ,fm ) : ci ∈ C , xα(i) L(fi ) = xν .
i=1
(2) s(i,j) ∈ Sα(i,j) .

(3) The set {s(i,j) } generates syz(f1 , . . . ,fm ).

Proof.
Pm P
(1) Let s = i=1 si ei ∈ syz(f1 , . . . ,fm ) and write si = α∈Nn ciα xα ∈ C [x]. For ν ∈
Nn+1 put (
ciβ xβ if xβ L(fi ) = xν
siν =
0 if L(fi ) ∤ xv .
Pm P
Define sν := i=1 siν ei . Then obviously sν ∈ Sν and s = ν sν .
(2) This is clear.
(3) To prove the third statement, we may assume that C(fi ) = 1 for all i. Let L(fi ) = xαi .
Because ofP (1) it is enough to prove the statement for homogeneous s ∈ Sγ . We thus
m
write s = i=1 ci xβi , ci ∈ C and βi + αi = γ for all i with ci 6= 0. Let {i1 , . . . ,ik } be
the indices such that ciν 6= 0. We will prove by induction on k that s is in the module
generated by the s(i,j) . This is obvious for k = 0, as then s = 0. If k 6= 0, then k ≥ 2.
Note that as αij + βij = γ, it follows in particular that xγ ≥ lcm(L(fi1 ),L(fi2 )) =: xα .
Thus we can consider
xγ
s′ := s − c1 α s(i1 ,i2 ) .
x
We have s′ ∈ syz(f1 , . . . ,fm ), and s′ ∈ ⊕kν=2 C [x]eiν . By induction, this proves the lemma.
Theorem 7.2.16 (Buchberger’s Criterion). Let I = (f1 , . . . ,fm ) ⊂ C {x}N be a submod-

ule. The following conditions are equivalent:
(1) S = {f1 , . . . ,fm } is a standard basis for I.

(2) NF spoly(fi ,fj )|S = 0 for all i,j.
(3) Let G be a subset of {s(i,j) } generating syz(f1 , . . . ,fm ). Then NF spoly(fi ,fj )|S) =
0 for all i,j with s(i,j) ∈ G.
Proof. The implication (1) =⇒ (2) follows from Corollary 7.2.10, and (2) =⇒ (3) is
obvious.
It remains to show (3) =⇒ (1). Without loss of generality, we may assume that C(fi ) = 1
for all i. Let L(fi ) = xαi for i = 1, . . . ,m. Consider f ∈ I. We have
Pmto prove that L(f )
is divisible by L(fi ) for some i. As f ∈ I we can write f = i=1 gi fi . To such a
representation we associate the monomial xδ = min{L(gi fi )}. Assume that the repre-
sentation is chosen with maximal xδ between all possible representations. We will prove
that L(f ) = xδ . This would imply L(f ) = L(gi fi ) = L(gi ) · L(fi ) for some i, and thus
the theorem. Suppose the converse, that is L(f ) > xδ . We define
I1 := {i : L(gi fi ) = xδ }, I2 := {i : L(gi fi ) 6= xδ }.
Then we can write

X X X
f= C(gi )L(gi )fi + Tail(gi )fi + gi fi .
i∈I1 i∈I1 i∈I2
P
The assumption L(f ) > xδ implies i∈I1 C(gi )L(gi )L(fi ) = 0. Thus we have a ho-
mogeneous syzygy on the leading terms of the fi , in fact in Sδ ⊂ syz(f1 , . . . ,fm ).
By assumption on G and Lemma 7.2.15 this syzygy on the leading terms is of type
P αij (i,j)
i,j∈I1 cij x s ∈ Sδ for suitable cij ∈ C , and αij ∈ Nn . Taking the dot product
s(i,j) ∈G
with (f1 , . . . ,fm ) we obtain
X X
C(gi )L(gi )fi = cij xαij spoly(fi ,fj ).
i∈I1 i,j∈I1
s(i,j) ∈G
Now consider a nonzero term of the right hand side. Then there exist i,j ∈ I1 , with
s(i,j) ∈ G and cij spoly(fi ,fj ) 6= 0. As the syzygy on the leading terms is in Sδ we have
xαij lcm(L(fi ),L(fj )) = xδ . It was noted in the second part of Definition 7.2.12 that
L(xαij spoly(fi ,fj )) > xδ .
Now the third assumption implies NF spoly(fi ,fj )|S = 0 for s(i,j) ∈ G and, there-
Pm ij
fore, spoly(fi ,fj ) = ν=1 qν fν and L(qνij fν ) ≥ L spoly(fi ,fj ) . Thus in particular we
get xαij L(qνij fν ) > xδ . We get
X X m
X
C(gi )L(gi )fi = cij xαij qνij fν
i∈I1 i,j∈I1 ν=1
s(i,j) ∈G
Xm X
= cij xαij qνij fν .
ν=1 i,j∈I1
s(i,j) ∈G
Now we define g ν by
(
X gν , for ν ∈ I2
gν = cij xαij qνij +
i,j∈I1
Tail(gν ), for ν ∈ I1 .
s(i,j) ∈G
Pm
Then f = ν=1 g ν fν and L(gν fν ) > xδ for all ν. This is a contradiction to the choice of
δ and proves the theorem.
Example 7.2.17. Let f1 = x4 + x3 y 2 , f2 = y 4 + x2 y 3 , w = (1,1). Then {f1 ,f2 } is a
standard basis:
spoly(f1 ,f2 ) = x3 y 6 − x6 y 3 = −x2 y 3 f1 + x3 y 2 f2 .
The following lemma gives the possibility to find proper subsets of {s(i,j) } which
generate syz(f1 , . . . ,fm ). This is very useful for the application of 7.2.16.
Lemma 7.2.18 (Chain Criterion). Let f1 , . . . ,fm ∈ C {x}N and assume that a sub-
set G of {s(i,j) } generates
syz(f1 , . . . ,fm ). Suppose s(i,j) , s(i,k) and s(k,j) are in G and
L fi )| lcm(L(fj ),L(fk ) . Then G r {s(j,k) } generates syz(f1 , . . . ,fm ).
Proof. We may assume that C(fi ) = 1 for all i, then one computes that

lcm L(fj ),L(fk ) (i,j) lcm L(fj ),L(fk ) (i,k)
s(k,j) = s − s .
lcm L(fi ),L(fj ) lcm L(fi ),L(fk )
Thus the lemma follows from 7.2.16.
Exercises
7.2.19 (Product Criterion). Let f1 ,f2 ∈ C {x}, and assume that lcm(L(f1 ),L(f2 )) = L(f1 ) ·
L(f2 ). Prove that NF(spoly(f1 ,f2 )|{f1 ,f2 }) = 0.
7.2.20. Let <m be an elimination ordering (see 7.1.11 (2)) with w1 = · · · = wn = 0, wn+1 = 1.
Let I ⊂ C {x}N P
be a submodule and S a standard basis ofP I.
Prove that S ∩ ti=0 C {x}xin+1 is a standard basis of I ∩ ti=0 C {x}xin+1 for all t ≤ N − 1.
7.2.21 (Gröbner bases). In this exercise we will develop the theory of standard bases for poly-
nomial rings rewriting some passages of the Sections 7.1 and 7.2.
• Replace C {x} by C [x].
• Replace in Definition 7.1.3 xαi < xαi+1 by xαi > xαi+1 .
• Rewrite the examples.
• Replace (2) by L(qν fν ) ≤ L(f ), for ν = 1, . . . ,m in Theorem ??, and cancel Remark 7.1.7.
• Replace the proof of Theorem ?? by the following NF–algorithm (normal form algorithm):
r = NF(f |{f1 , . . . ,fm })
r=f
S = {f1 , . . . ,fm }
while r 6= 0 and L(fi ) | L(r) for some i
choose i minimal with L(fi ) | L(r)
C(r) L(r)
r = r − C(f i ) L(fi )
fi
6 0
if r =
r = C(r)L(r) + NF(Tail(r)|S)
return(r).
Prove that this algorithm
P terminates. Use the fact that < is a well-ordering and Definition
7.1.3. Prove that f = qν fν + r, r = NF(f |{f1 , . . . ,fm }) for suitable qν ∈ C [x] and (1)
and (2) of Theorem ?? are true.
• Check the proofs of 7.2.7, 7.2.8, 7.2.11 and 7.2.10, 7.2.15, and 7.2.16 (here we have to
change the definition of xδ = max{L(gi fi )} to min{L(gi fi )} and replace several times >
by <), and 7.2.18.
7.2.22 (Buchberger’s algorithm).
S = standard({f1 , . . . ,fm })
S = {f1 , . . . ,fm }
P = {(fi ,fj )}i<j
while P 6= ∅
choose (f,g) ∈ P
P = P r {(f,g)}
h = NF(spoly(f,g)|S)
if h 6= 0
P = P ∪ {(f,h)|h ∈ S}
S = S ∪ {h}
return(S).
• Prove that the algorithm terminates.
• Use 7.2.16 to prove the S is a standard basis of (f1 , . . . ,fm ).
7.2.23 (Standard bases for formal power series). Standard bases for formal power series can be
defined as in 7.2.3.
• Prove that the theory developed in this section holds in the formal case.
• Prove that {f1 , . . . ,fm } ⊂ C {x}N is a standard basis if and only if {f1 , . . . ,fm } ⊂ C [[x]]N
is a standard basis.
• Let {f1 , . . . ,fm } ⊂ C {x}N be a standard basis and f ∈ C {x}N .

Prove that NF(f |{f1 , . . . ,fm }) is the limit of the following sequence:
(1) v0 = f .
(2) Suppose vi is defined. Then vi+1 := vi − hfπ(i) . Here h = 0 and π(i) = 1 if no
monomial in vi is divisible by L(fj ) for all j. Otherwise, let cxα be the smallest term
in vi such that L(fj ) | xα for some j. Let π(i) be minimal with fπ(i) | xα . Then we
α
define h := c L(fx ) .
π(i)
7.2.24 (Computation of standard bases in the power series ring). Let < be a well ordering
on the set of monomials of C [x]. We choose a new variable t and extend this ordering in the
following way to C [x,t]:
x α ta > x β tb
if deg(xα ) + a > deg(xβ ) + b or in case of equality xα < xβ . This extension has the following
property: P
Let f ∈ C [x], f = i≥1 ai xαi , ai 6= 0 and xαi < xαi+1 , that is, the monomials in f are ordered
` ´
in the sense of a power series ring (cf. Definition 7.1.3). Let F = f h = tdeg(f ) f xt1 , . . . , xtn be
P αi αi
the homogenization of f , then F = i≥1 ai xαi tdeg(f )−deg(x ) and we have xαi tdeg(f )−deg(x ) >
αi+1
xαi+1 tdeg(f )−deg(x )
, that is, the monomials of F are ordered in the sense of a polynomial
ring (cf. Exercise 7.2.21).
Prove the following: Let f1 , . . . ,fm ∈ C [x] and I = (f1 , . . . ,fm )C {x}, respectively I =
(f1 , . . . ,fm )C [[x]]. Let < be a well-ordering on C [x] extended as before to C [x,t]. Let J =
(f1h , . . . ,fm
h
)C [x,t] and {G1 , . . . ,Gr } be a standard basis of J. Then {G1 (t = 1), . . . ,Gr (t = 1)}
is standard basis of I.3
7.3 Applications
Proposition 7.3.1. Let I ⊂ C {x} be a 0–dimensional ideal, then dimC C {x}/I =

dimC C {x}/L(I).
Proof. {xα |xα 6∈ L(I)} induces a C –basis of C {x}/L(I). We shall prove thatP this is also
a C –basis of C {x}/I. Let SP= {f1 , . . . ,fm } be a standard basis of I. Assume cα xα ∈ I
for some cα ∈ C , then NF( Pcα xα |S) = 0 but this is impossible, because S is a standard
basis and all monomials in cα xα are not divisible by L(fi ) for all i. This implies that
the {x |x 6∈ L(I)} induces a linearly independent set in C {x}/I. t Let xβ be a monomial
α α
in L(I) then NF(xβ |S) is a linear combination of the {xα |xα 6∈ L(I)} and, therefore, this
set is also a basis.
Now we want to prove that for w = (1, . . . ,1) the Hilbert–Samuel function of an ideal
I ⊂ C {x} coincides with the corresponding function of the leading ideals (cf. Chapter
4.3).
Corollary 7.3.2. Let I ⊂ C {x} be an ideal, then HSC {x}/I = HSC {x}/L(I) .
3 In practice this way to compute a standard basis is not very efficient. In the computer algebra sys-
tem Singular for this purpose Mora’s algorithm, a modified version of Buchberger’s algorithm, see
[Mora 1982] is implemented.

Proof. HSC {x}/I (c) = dimC C {x}/I + (x)c
= dimC C {x}/L I + (x)c
by 7.3.1
= dimC C {x}/L(I) + (x)c
= HSC {x}/L(I) (c).
Finally, we shall prove Lemma 5.1.21 from Chapter 5.
Let f ,g ∈ C [x,y] quasi-homogeneous deg(x) = q, deg(y) = p and (x,y)–primary,
let f = f + terms of higher degree, g = g+ terms of higher degree. Then xi y j ∈ (f,g) if
iq + jp > w-ord(f · g).
Proof. We choose w = (p,q) and compute a standard basis for (f,g). Because of the
fact that f ,g are quasi-homogeneous and (f ,g) is (x,y)–primary, we may assume that
L(f ) = xc0 and define g0 = f and g1 = NF(g|{g0 }), L(g1 ) = xc1 y b1 with c0 > c1 and
b1 > 0. Assume we have already constructed {g0 , . . . ,gi−1 } with the following properties:
(1) gj ∈ (f,g).
(2) L(gj ) = xcj y bj , cj > cj+1 ,bj+1 > bj .

(3) gj = NF spoly(gj−2 ,gj−1 )|{g0 , . . . ,gj−1 } .

We define gi := NF spoly(gi−2 ,gi−1 )|{g0 , . . . ,gi−1 } .
Let L(gi ) = xci y bi then bi > bi−1 because L spoly(gi−2 ,gi−1 ) < xci−2 y bi−1 . Simi-
larly, ci < ci−1 .
Because the ci are strictly decreasing and (f ,g) is (x,y)–primary there is an m such
that L(gm ) = y bm and gm+1 = 0. Namely, gm+1 = NF spoly(gm−1 ,g m ) | {g 0 , . . . ,g m } . If
cm−1 bm
gm+1 6= 0 then L(gm+1 ) > x y and, therefore, deg L(gm+1 ) > deg spoly(gm−1 ,gm ) =
pbm + qcm−1 . We shall prove now that in this situation L(gm+1 ) ∈ L(g0 ), . . . ,L(gm ) ,
which is a contradiction for gm+1 being in normal form. At the same time we prove
that {g0 , . . . ,gm } is a standard basis of (f,g). We use Theorem 7.2.16 and Lemma
7.2.18. The property cj > cj+1 and bj < bj+1 implies that L(gj−1 ) = xcj−1 y bj−1 |
lcm L(gj ),L(gk ) = x y for k ≤ j − 2. This implies that already G = {s(i,i−1) }i=1,...,m
ck bj
generates syz(g0 , . . . ,gm ). (Lemma 7.2.18.) But by definition gi =NF spoly(gi−2 ,gi−1 )
| {g0 , . . . ,gi−1 } and, therefore, NF spoly(gi−2 ,gi−1 )|{g0 , . . . ,gm } = 0. (4) of Theorem
7.2.16 implies that {g0 , . . . ,gm } is a standard basis of (f,g).
Because of (f ,g) being (x,y)–primary, L(gm ) = y bm . Now deg(g0 ) = c0 p (the
weighted degree) and
deg(gi ) = deg(gi−1 ) + (ci−2 − ci−1 )p

= deg(g1 ) + (c0 − ci−1 )p.
This implies that
deg(gm ) = deg(g1 ) + (c0 − cm−1 )p

= deg(g0 · g1 ) − cm−1 p
= deg(f · g) − cm−1 p
bm xcm−1 y bm
·xi y j
b1
x
cm−1 c0
The monomial xcm−1 y bm has, therefore, degree deg(f · g). It is the biggest monomial
which is not in (f,g). This implies that xi y j ∈ (f,g) if iq + jp > deg(f · g) and proves the
Lemma.
As a totally different application, we can study the ideal of the product of germs of
analytic spaces.
Definition 7.3.3. Let (X, 0) ⊂ (C n , 0) and (Y, 0) ⊂ (C m , 0) be germs of analytic spaces.
Let X and Y be representatives We define the product (X × Y, 0) to be the germ of
(X × Y ) in 0.
One checks without difficulty that this definition is independent on the choice of
the representatives. Let I ⊂ C {x} be the ideal of (X, 0), and J ⊂ C {y} be the ideal of
(Y, 0). Then it is a direct check that the product (X × Y, 0) is given as the zero set of
(I + J) · C {x,y}. That this are all functions in the ideal of (X × Y, 0) is not at all trivial:
Theorem 7.3.4. With the above notation one has
I (X × Y, 0) = (I + J) · C {x,y}.
Proof. The inclusion ⊃ is easy. For the converse, let f ∈ I (X × Y, 0). We have to show
that f ∈ (I + J) · C {x,y}. Take a standard basis of J. By Grauert’s division Theorem
7.1.7, we may assume that no monomial of f is in L(J). We write
X
f= fα y α , fα ∈ C {x}.
α
Therefore, for all α with fα 6= 0 we have that y α is not in the leading ideal of J. We
claim that all fα are in I, which suffices to prove the Theorem (Exercise 7.3.7). We choose
representatives X of (X, 0) and Y of (Y, 0) such that f is defined on X × Y and f (a,b) = 0
for all (a,b) ∈ X × Y . Suppose the converse. Then there exists an a ∈ X with fα (a) 6= 0
for some α. This is beause I = I (X, 0). For all b ∈ Y we have
X
f (a,b) = fα (a)bα = 0,
α
that is, X
f (a,y) = fα (a)y α ∈ I (Y, 0) = J.
α
This is a contradiction, as the leading term of f (a,y) does not lie in L(J).
This result motivates the following definition.
Definition 7.3.5. Let A = C {x}/I and B = C {y}/J be analytic algebras. Then we
b by
define the analytic tensor product A⊗B
b := C {x,y}/(I + J).
A⊗B
Thus in case that A and B are reduced, and are the local rings of (X, 0) and (Y, 0)
respectively, the analytic tensor product is the local ring of the product (X × Y, 0).
Remark 7.3.6. We have obvious maps inclusions i : A −→ A⊗B, b b
and j : B −→ A⊗B,
which corresponds geometrically to the projection on the factors, if A and B are reduced.
On the other hand, we also have a natural surjective map A⊗B b −→ A. This corresponds
geometrically with the inclusion of one of the factors in the product.
The analytic tensor product satisfies the following universal property. Let C be an
analytic algebra, and γ : A −→ C and δ : B −→ C be analytic maps. Then there exist a
uniquely determined map θ : A⊗B b −→ C such that γ = θ ◦ i and δ = θ ◦ j.
In the case that A and B are reduced, this is easy to prove and follows from the fact
that analytic maps (Z, 0) −→ (X, 0) and (Z, 0) −→ (Y, 0) induce obviously an analytic
map (Z, 0) −→ (X × Y, 0).
For the proof of the universal property and more facts on the analytic tensor product
we refer to the book Analytische Stellenalgebren, see [Grauert-Remmert 1971].
Exercises:
P
7.3.7. Let I ⊂ C {x1 , . . . ,xn } be an ideal and f ∈ C {x1 , . . . ,xn , y1 , . . . ,ym }, f = fα y α such
that fα ∈ I for all α. Prove that f ∈ IC {x1 , . . . ,xn ,y1 , . . . ,ym }.
(Hint: Use the fact that the normal form with respect to IC {x1 , . . . ,xn ,y1 , . . . ,ym } and the
degree ordering with w = (1, . . . ,1) is additive and maps (x1 , . . . ,ym )c to (x1 , . . . ,ym )c .)
7.3.8. Let I ⊂ C {x0 , . . . ,x4 } be the ideal generated by the 2–minors of the matrix ( xx01 xx12 xx23 xx34 ).
Prove that I is a prime ideal.
(Hint: Consider the map ϕ : C {x0 , . . . ,x4 } −→ C {x0 ,x1 }x0 defined by ϕ(x2 ) = x10 x21 , ϕ(x3 ) =
1 3
x and ϕ(x4 ) = x13 x41 . Prove that Ker(ϕ) = I : (x0 )∞ . Then prove that I = I : (x0 ). To
x2 1
0 0
see this, prove that (x1 ,x2 ,x3 )2 = L(I) the leading ideal of I with respect to the degree reverse
lexicographical ordering. The six 2–minors of the matrix above are already a standard basis of
I with respect to this ordering.)
286
8 Approximation Theorems
A canonical construction in local analytic geometry, which is possible without knowing

the convergence of the solution, should have a convergent solution. This is always our
wish and the underlying philosophy of the Approximation Theorems. We saw already in
Chapters 3, 5, 6 and 7 that a lot of canonical formally constructed power series simply
(and luckily) converge. Unfortunately, this is not always the case, as we will see in this
chapter. This is the reason why we need different Approximation Theorems, adapted to
the problem. We start with the famous Approximation Theorem of M. Artin: if a system
of analytic equations has a formal solution (which is sometimes really not convergent)
then we can find a convergent solution to this system. Even more, we can approximate
the formal one by convergent ones up to a given order.
Unfortunately, special properties of the formal solution (as not depending on some
variables) do not survive during the approximation process in general. Therefore, for
special situations, as needed later in deformation theory, we need special Approxima-
tion Theorems as Grauert’s Approximation Theorem proved in Section 8.2. In Section
8.3 we will give an overview about general Approximation Theorems and prove some
applications.
8.1 Artin’s Approximation Theorem
Here it comes:
Theorem 8.1.1 (Artin’s Approximation Theorem). Consider two sets of variables x =
(x1 , . . . ,xn ) and y = (y1 , . . . ,yN ). Let f1 , . . . ,fm ∈ C {x,y} and y ∈ (x)C [[x]]N such that
fi (y) = 0 for i = 1, . . . ,m. Let c be a positive integer. Then there exists a y ∈ C {x}N
such that:
• fi (y) = 0 for i = 1, . . . ,m.
• y ≡ y modulo (x)c .
The proof is by induction on n. The case n = 0 is trivial. The proof of the induction
step is quite long, and will be divided in several lemmas. The first two lemmas will
prepare us to prove the Artin Approximation Theorem for a special case. The general
case will then be reduced to this special case.
Lemma 8.1.2. Suppose that Artin’s Approximation Theorem holds for n − 1 variables
x1 , . . . ,xn−1 . Consider the following data:
• c, a positive integer.
• f1 , . . . ,fm ∈ C {x,y}.
• A formal Weierstraß polynomial P = xsn + a1 xns−1 + . . . + as ,
ai ∈ (x1 , . . . ,xn−1 )C [[x1 , . . . ,xn−1 ]].
8.1 Artin’s Approximation Theorem 287
• Elements: r1 , . . . ,r N ∈ C [[x1 , . . . ,xn−1 ]][xn ] of degree smaller than s in xn .
Suppose that
fi (r(x)) ≡ 0 modulo P for i = 1, . . . ,m.
Then there exists an analytic Weierstraß polynomial P of degree s in xn , and elements
r1 , . . . ,rN ∈ C {x1 , . . . ,xn−1 }[xn ] of degree smaller than s with respect to xn such that
fi (r(x)) ≡ 0 modulo P
and P ≡ P and ri ≡ ri modulo (x)c .
Proof. We may write:

s−1
X
ri = (r ij + cij )xjn r ij ∈ (x1 , . . . ,xn−1 ), cij ∈ C .
j=0
We consider new variables:
A1 , . . . ,As ; Rij , 1 ≤ i ≤ N, 0 ≤ j ≤ s − 1
and define
Pe := xsn + A1 xns−1 + . . . + As
s−1
X
ei :=
R (Rij + cij )xjn 1 ≤ i ≤ N.
j=0
We do m Weierstraß division with remainder for each fi and obtain

s−1
X
e1 , . . . ,R
fi (R eN ) = Qi · Pe + Fij xjn i = 1, . . . ,m,
j=0
for some Fij ∈ C {x1 , . . . ,xn−1 ,A1 , . . . ,As ,{Rij }}. Plugging in a1 , . . . ,as for A1 , . . . ,As
and, similarly, r ij for Rij for 1 ≤ i ≤ N, 0 ≤ j ≤ s − 1, we see that we have formal
solutions to the equations Fij = 0. We may apply the induction hypothesis to find the
P
convergent solutions ai and the rij . Then clearly P = Pe (a) and ri = (rij + cij )xjn
satisfy the statement of the lemma.
x1 , . . . ,xn−1 . Let g,f1 , . . . ,fm ∈ C {x,y}, c ∈ N and formal power series y ∈ (x)C [[x]]N
be given such that g(y) 6= 0 and
fi (y) ≡ 0 modulo g(y) i = 1, . . . ,m.
Then there exists a y ∈ C {x}N with y ≡ y modulo (x)c such that g(y) 6= 0 and
fi (y) ≡ 0 modulo g(y) i = 1, . . . ,m.

288 8 Approximation Theorems
Proof. Consider g(y) ∈ C [[x]]. After a linear coordinate change in the x1 , . . . ,xn , we may
assume that g(y) is regular of order s in xn . We may enlarge c so that c > s. We apply
the Weierstraß Preparation Theorem:
g(y) = U · P ,
where P is a Weierstraß polynomial in xn of degree s, and U is a unit. In particular

it follows that fi (y) ≡ 0 modulo P . For i = 1, . . . ,m we do Weierstraß divisions with
remainder:
y i = Qi P + r i ,
where r i ∈ C [[x1 , . . . ,xn−1 ]][xn ] are polynomials of degree smaller than s in xn . We will
approximate the Qi ,P , and the r i . From yi ≡ ri modulo P it follows that
g(r) ≡ g(y) ≡ 0 modulo P .
The first equivalence follows from the Taylor expansion, the second equivalence holds by
construction of P . Similarly
fi (r) ≡ fi (y) ≡ 0 modulo P .
By the previous lemma, applied to the functions g,f1 , . . . ,fm , we can find convergent
ri ∈ C {x}, and a convergent Weierstraß polynomial P with ri ≡ r i and P ≡ P modulo
(x)c such that
g(r) ≡ 0 and fi (r) ≡ 0 modulo P for i = 1, . . . ,m.
To define the yi , take any Qi ∈ C {x} such that Qi ≡ Qi modulo (x)c and put:
yi := Qi P + ri .
By construction yi ≡ y i modulo (x)c . Using Taylor’s formula again, as above, we see that
g(y) ≡ 0 and fi (y) ≡ 0 modulo P for i = 1, . . . ,m.
We are finished if we show that
(8.1) g(y) = U · P
for a unit U ∈ C {x}. It then follows immediately that fi (y) ≡ 0 modulo g(y) for
i = 1, . . . ,m, and g(y) 6= 0 because P 6= 0. To show (8.1), observe that by the Weierstraß
Division Theorem and the fact that g(y) ≡ 0 modulo P there exists a U ∈ C {x} with
g(y) = U · P . We only have to show that U is a unit. By assumption, g(y) is regular of
order s in xn . Because y ≡ y modulo xc and c > s it follows that g(y) is regular of order
s in xn . As P is a Weierstraß polynomial of degree s in xn , it follows that U is a unit.
This finishes the proof of the lemma.
x1 , . . . ,xn−1 . Then Artin’s Approximation Theorem for n variables x1 , . . . ,xn holds under
the following assumptions:
(1) m ≤ N .
8.1 Artin’s Approximation Theorem 289

∂fi
(2) There exists an m–minor ∆ of the Jacobian matrix ∂yj such that ∆(y) 6= 0.
Proof. This we will prove as an application of Newton’s Lemma, which itself follows
from the Implicit Mapping Theorem. Take any nonzero function h ∈ (x)c . Then because
fi (y) = 0, we certainly have:
fi (y) ≡ 0 modulo ∆2 (y)h.
We can apply the previous lemma to find a y ′ ∈ C {x}N , y ′ ≡ y modulo (x)c , such that
fi (y ′ ) ≡ 0 modulo ∆2 (y ′ )h.
Putting J to be the ideal generated by the m–minors of the Jacobian matrix we certainly
have
fi (y ′ ) ≡ 0 modulo J 2 (y ′ )(x)c ,
so that Newton’s Lemma 3.3.34 applies: there exists a y ∈ C {x}N , y ≡ y ′ modulo
J(y ′ )(x)c , such that fi (y) = 0 for all i = 1, . . . ,m. (Note that we needed m ≤ N in order
to apply Newton’s Lemma.)
Proof of Artin’s Approximation Theorem 8.1.1. Define the ideal p to be the kernel of the
map:
α : C {x,y} −→ C [[x]]
defined by x 7→ x, and y 7→ y(x). Note that as we have an injective map C {x,y}/p ֒→
C [[x]], it follows that p is a prime ideal. As f1 , . . . ,fm ∈ p it suffices to find a y(x) ∈
C {x}N such that g(x,y(x)) = 0 for all g ∈ p. So we may have assumed p = (f1 , . . . ,fm )
from the beginning.
Let ht(p) = k. By the Jacobian Criterion 4.3.15 applied to the case I = p = q, there
exist k variables z1 , . . . ,zk among the xi and the yi , such that (after renumbering the fi )

∂fi
∆′ := det ∈/ p.
∂zj 1≤i,j≤k
∂fi (x,y(x))
In particular ∆′ (x,y(x)) 6= 0. As 0 = fi (x,y(x)) ∈ C [[x]], it follows that ∂xj =0
for all j. From the chain rule it follows that
∂fi X ∂fi ∂y (x)

(x,y(x)) = − (x,y(x)) · k .
∂xj ∂yk ∂xj
k
Hence, ∆′ (x,y(x)) is in the ideal generated by the k–minors of the matrix

∂fi
(x,y(x)) .
∂yj 1≤i≤m
1≤j≤N
In particular, it follows that one of those minors is nonzero. We may, therefore, apply
the previous lemma to find a convergent y ∈ C {x}N with y ≡ y modulo (x)c and
f1 (x,y(x)) = . . . = fk (x,y(x)) = 0.
Now C {x,y}p /p is a regular local ring. (It is even a field.) In the Jacobian Criterion
4.3.15 it was shown that f1 , . . . ,fk generate pC {x,y}p , that is,
(f1 , . . . ,fk )C {x,y}p = pC {x,y}p .
Because C {x,y} is Noetherian, we can find h ∈

/ p such that
(8.2) ((f1 , . . . ,fk ) : h) = p.
As h ∈ / p = Ker(α) it follows that h(x,y(x)) 6= 0. By enlarging c we may assume that

h(x,y(x)) ∈ / (x)c . Hence y ≡ y modulo (x)c implies that h(x,y(x)) ∈ / (x)c . In particular
h(x,y(x)) 6= 0.
For any g ∈ p it follows from formula (8.2) that hg ∈ (f1 , . . . ,fk ). As h(x,y(x)) 6= 0
it follows that g(x,y(x)) = 0. This is what we had to show.
Exercises
8.1.5. Deduce from the Artin Approximation Theorem the following somewhat more general
version.
N
` Let ´f1 , . . . ,fm ∈ C {x,y}, and J ⊂ C {x} be an ideal. Let y(x) ∈ C [[x]] such that
fi x,y(x) ∈ J · C [[x]] for i = 1, . . . ,m. Then there exists a y(x) ∈ C {x}N such that fi (x,y(x)) ∈
J for all i = 1, . . . ,m.
8.1.6. Let the conditions be as in the previous exercise. Let I ⊂ C {x} be an ideal. Suppose we
have a formal solution y(x) whose components lie in I · C [[x]]. Then we can find a convergent
solution in I. Prove this.
8.1.7. Let p ⊂ C {x} be a prime ideal. Show that p · C [[x]] is a prime ideal.
(Hint: Consider f,g ∈ C [[x]] with f · g ∈ p · C [[x]]. Use Exercise 8.1.5 to find for all c ∈ N
fc ,gc ∈ C {x}, fc ≡ f modulo (x)c and gc ≡ g modulo (x)c , and fc · gc ∈ p. Deduce that either
f = limc→∞ fc ∈ p · C [[x]] or g = limc→∞ gc ∈ p · C [[x]].)
8.1.8. Let p ⊂ C {x} be a prime ideal. Suppose that C {x}/p is normal. Use Exercise 8.1.5 to
see that C [[x]]/p · C [[x]] is normal.
8.2 Grauert’s Approximation Theorem
In this section we prove Grauert’s Approximation Theorem. This approximation theorem

looks rather complicated, but is very useful. For example, we will use this theorem to
prove the existence of a ‘semi-universal’ deformation of an isolated singularity in Chapter
10.
Definition 8.2.1. We consider four sets of variables, x = (x1 , . . . ,xn ), s = (s1 , . . . ,sl ),
Φ = (Φ1 , . . . ,Φp ) and Ψ = (Ψ1 , . . . ,Ψq ). Let I ⊂ C {s} be an ideal. We can extend I to
an ideal I ⊂ C {x,s}. Consider an element F = (F1 , . . . ,Fk ) ∈ C {x,s,Φ,Ψ}k .
(1) An analytic solution of the equation F ≡ 0 modulo I is a pair (φ,ψ) ∈ C {s}p ×

C {x,s}q , that is power series φ1 , . . . ,φp ∈ C {s}, and ψ1 , . . . ,ψq ∈ C {x,s} such that
F (x,s,φ(s),ψ(x,s)) ≡ 0 modulo I · C {x,s}k ,
that is, the φ and the ψ have the property that
F1 (x,s,φ(s),ψ(x,s)), . . . ,Fk (x,s,φ(s),ψ(x,s)) ∈ I.

8.2 Grauert’s Approximation Theorem 291
(2) A solution of order e of the equation F ≡ 0 modulo I is a pair

(φ,ψ) ∈ C [s]p × C {x}[s]q such that
F (x,s,φ(s),ψ(x,s)) ≡ 0 modulo (I + me+1 ) · C {x,s}k ,
where m = (s) = (s1 , . . . ,sl ).
We can now formulate the theorem.

Theorem 8.2.2 (Grauert’s Approximation Theorem). Let the notation be as in the
previous definition. Let e0 ∈ N. Suppose that the system of equations
F ≡ 0 modulo I
has a solution (φ(e0 ) ,ψ (e0 ) ) of order e0 . Suppose, moreover, that for all e ≥ e0 every
solution (φ(e) ,ψ (e) ) of order e with φ(e) ≡ φ(e0 ) modulo me0 +1 , and ψ (e) ≡ ψ (e0 ) modulo
me0 +1 , extends to a solution (φ(e) + δ (e) ,ψ (e) + γ (e) ) of order e + 1, with δ (e) ∈ C [s]p and
γ (e) ∈ C {x}[s]q homogeneous of degree e + 1 in s. Then the system of equations F ≡ 0
modulo I has an analytic solution (φ,ψ), with φ ≡ φ(e0 ) modulo me0 +1 , and ψ ≡ ψ (e0 )
modulo me0 +1 .
Before giving the proof, we need some preparations. As in Chapter 7, we choose as
a basis for the free module C {x,s,Φ,Ψ}k the vectors ei := (0, . . . ,0,1,0, . . . ,0) (the 1 is on
the i–th spot) and identify ei with xi−1 n+1 by adding a new variable xn+1 and enlarging x
to x = (x1 , . . . ,xn+1 ).
We choose on C {x,s,Φ,Ψ}k the degree lexicographical ordering as in Chapter 7 with
weights 1 for all variables. For an element F ∈ C {x,s,Φ,Ψ}k there exists a polyradius
n+1+l+p+q
λ ∈ R+ such that F ∈ Bλk . We will write, in order to distinguish between the
p+q
variables x,s and φ,ψ, the polyradius as λ = (ρ,τ,σ), ρ ∈ Rn+1 l
+ , τ ∈ R+ , σ ∈ R+ .
Proposition 8.2.3. Let b1 , . . . bt ∈ C {x}k and ρ ∈ Rn+1

+ such that b1 , . . . bt ∈ Bρk be
given. Then for all ρ ≤ ρ there exists a K(ρ) ∈ R+ such that the following holds.
P
For all C ∈ C {x}[s]k , C = |β|=e Cβ s
β
with Cβ ∈ C {x} and for all y1 , . . . ,y t ∈
C [s] such that
(1) yj is a homogeneous polynomial of degree e in s for j = 1, . . . ,t,
P
t
(2) bj y j = C,
j=1
there exist yj ∈ C [s], j = 1, . . . ,t such that

(1) yj is a homogeneous polynomial of degree e in s for j = 1, . . . ,t,
P
t
(2) bj yj = C,
j=1
(3) kyj kτ ≤ K(ρ)kCk(ρ,τ ) for all j and all ρ ≤ ρ, τ ∈ Rl+ .

The important thing to note here, is that K(ρ) does not depend on C, e and τ . However,
it depends (of course) on the bi and on ρ.
Proof. This is an application of Cramer’s rule.

Step 1. We write
X X X
bi = bi,α xα , C = Cα,β xα sβ , y j = y j,β sβ .
α∈Nn+1
+ α∈Nn+1
+
|β|=e
|β|=e
The condition then reads

t
X
bj,α y j,β = Cα,β for all α ∈ Nn+1
+ and |β| = e.
j=1
This we can view as system of linear equations. More precisely, consider the linear sub-
space in C t spanned by the vectors
b(α) := (b1,α , . . . ,bt,α ), α ∈ Nn+1

+ .
Let r be the dimension of this subspace and choose a basis
{b(αi ) = (b1,αi , . . . ,bt,αi ), i = 1, . . . ,r}.

Pr
So for any α ∈ Nn+1 , we have an expression b(α) = i=1 ki,α b(αi ) for some ki,α ∈ C . By
linear algebra, there is an r–minor M of the matrix

bi,αj 1≤i≤t
1≤j≤r
which is nonzero. After renumbering the bi we may assume that M = det(bi,αj )1≤i,j≤r .
We define
P (i,β)
(1) for i = 1, . . . ,r, yi := |β|=e yi,β sβ with yi,β = MM ,
(2) yr+1 := · · · := yt := 0.
Here M (i,β) is obtained from M by replacing Pt the i–th column of M by the column vector
(Cα1 ,β , . . . ,Cαr ,β )t . We want to show j=1 bj,α yj,β = Cα,β for all α. By Cramer’s rule,
we know this is true for α = α1 , . . . ,αr . Now
X X X X X
Cα,β = bj,α y j,β = ki,α bj,αi y j,β = ki,α Cαi ,β = ki,α bj,αi yj,β = bj,α yj,β ,
j i,j i i,j j
as was to be shown.
Step 2. It remains to define the K = K(ρ) and to show the estimates. We take
|Mij |
K = max .
i,j ραj |M |
Here Mij is the determinant obtained from M by deleting the i–th row and the j–th
column. As yr+1 = · · · = yt = 0, we only have to show the estimates for the y1 , . . . ,yr .
By definition
X M (i,β) β X |M (i,β) |
kyi kτ = s τ = τβ.
M |M |
|β|=e |β|=e
Pr
Moreover, by linear algebra we have M (i,β) = j=1 Cαj ,β Mij and, therefore,
r
|M (i,β) | X |Mij |
≤ |Cαj ,β | .
|M | j=1
|M |
This implies
r
1 X X
kyi kτ ≤ |Cαj ,β | |Mij |τ β .
|M | j=1
|β|=e
On the other hand by the definition of the norm

r
X X
kCk(ρ,τ ) ≥ |Cαj ,β |τ β ραj .
|β|=e j=1
Thus with our choice of K the inequality
kyi kτ ≤ K · kCk(ρ,τ )
holds.
We need the following generalization of this proposition.
Lemma 8.2.4 (Cartan’s Lemma). Let a1 , . . . ,ar , b1 , . . . ,bt ∈ C {x}k and ρ ∈ Rn+1 + ,
τ ∈ Rl+ . Suppose that a1 , . . . ,ar , b1 , . . . ,bt ∈ Bρk . There exists a ρ ∈ Rn+1
+ , ρ ≤ ρ, and
K = K(ρ) ∈ R+ with the following property.
P
For all τ ∈ Rl+ with τ ≤ τ and for all C ∈ B(ρ,τ k
), C = |α|=e C
(α) α
s with C (α) ∈ C {x}
and for all y1 , . . . ,y t ∈ C [s] and z 1 , . . . z r ∈ C {x}[s] such that
(1) yj is a homogeneous polynomial of degree e in s for j = 1, . . . ,t, and z i is a homo-

geneous polynomial of degree e in s for i = 1, . . . ,r with coefficients in C {x},
P
r P
t
(2) ai z i + bj yj = C,
i=1 j=1
there exist zi ∈ C {x}[s], i = 1, . . . ,r and yj ∈ C [s], j = 1, . . . ,t, such that

(1) yj is a homogeneous polynomial of degree e in s for j = 1, . . . ,t, and zi is a homo-
geneous polynomial of degree e in s for i = 1, . . . ,r with coefficients in C {x},
P
r P
t
(2) ai z i + bj yj = C,
i=1 j=1
(3) kzi k(ρ,τ ) ≤ KkCk(ρ,τ ) and kyj kτ ≤ KkCk(ρ,τ ) for all i and j.
The important thing to note here, is that K does not depend on C, e, and τ . However,
it depends (of course) on the ai , the bi and ρ.
Proof. Step 1. The case r = 0 is dealt with in the previous proposition. We consider the
normal form NF with respect to the submodule (a1 , . . . ,ar ). Note that as the y j do not
Pr Pt
depend on x, we have NF(yj ) = y j . We take the normal form of i=1 ai z i + j=1 bj yj =
C and get
t
X
(8.3) NF(bj )y j = NF(C).
j=1
We apply the previous proposition. There exists a K0 (ρ) depending on the NF(bj ) and
P
ρ, and y1 , . . . ,yt ∈ C [s] homogeneous of degree e such that tj=1 NF(bj )yj = NF(C) and
(8.4) kyj kτ ≤ K0 (ρ)k NF(C)k(ρ,τ ) for all ρ ≤ ρ.
To define the zi , note that from 7.2.10 it follows that we can write
r
X r
X
(8.5) bj = αij ai + NF(bj ), C= βi ai + NF(C).
i=1 i=1
The precise αij and βi we take will be determined in the second step. Plugging (8.5) in
Pr Pt
i=1 ai z i + j=1 bj y j − C = 0 and using (8.3) we get
r
X t
X
zi + αij y j − βi ai = 0.
i=1 j=1
Now we define
t
X
zi := − αij yj + βi .
j=1
Obviously,
r
X t
X
zi + αij yj − βi ai = 0.
i=1 j=1
Pt
Working backwards, that is, using j=1 NF(bj )yj = NF(C) we get
r
X t
X
ai z i + bj yj = C.
i=1 j=1
Step 2. We still have to tell which αij and βi we take that fulfill (8.5), and show the
estimates. To do this, we take a standard basis f1 , . . . ,fm of the submodule (a1 , . . . ,ar ).
We can write
Xr
fν = hνi ai
i=1
for some hνi ∈ C {x}. Choose ρ ≤ ρ so small that a1 , . . . ,ar ,b1 , . . . ,bt ,f1 , . . . ,fm ,
h11 , . . . ,hmr ∈ Bρk . We apply Grauert’s Division Theorem, see Remark 7.1.7 with ε = 21 .
Therefore, by making ρ smaller we get equations
X
bj = bjν fν + NF(bj )
ν
X
C= Cν fν + NF(C),
ν
with the following estimates.

kb k
(1) kbjν kρ ≤ 2 kL(fjν )k
ρ
ρ
;
kCk
(2) kCν k(ρ,τ ) ≤ 2 kL(f(ρ,τ
ν )kρ
)
;
(3) k NF(C)k(ρ,τ ) ≤ 2kCk(ρ,τ ).
In particular from formula (8.4) we get for K0 := K0 (ρ)
(8.6) kyj kτ ≤ K0 k NF(C)k(ρ,τ ) ≤ 2K0 kCk(ρ,τ ) ,
so that we have an estimate for the yi . We define

X X
αij := bjν hνi , βi := hνi Cν .
ν ν
n o
khνi kρ
With these αij and βi the equations (8.5) hold. Let L = 2m · maxν,i kL(fν )kρ . Then
X
kαij kρ ≤ khνi kρ kbjν kρ ≤ Lkbj kρ ,
ν
kβi k(ρ,τ ) ≤ LkCk(ρ,τ ) .
From the definition of the zi and formula (8.6) we get

t
X
kzi k(ρ,τ ) ≤ kαij kρ kyj kτ + kβi k(ρ,τ )
j=1
t
X
≤ Lkbj kρ kyj kτ + LkCk(ρ,τ )
j=1
≤ L max{kbj kρ kyj kτ } + LkCk(ρ,τ )

j
≤ L max{2K0 kbj kρ + 1}kCk(ρ,τ ).

j
Let K ′ = L maxj {2K0 kbj kρ + 1} and K = max{2K0,K ′ }. With this K the desired
estimates hold.
We need one more result before we can give the proof of Grauert’s Approximation
Theorem.
Lemma 8.2.5. Let F (x,s,Θ) ∈ C {x,s,Θ}k be a vector of power series in three sets of
variables x = (x1 , . . . ,xn ), s = (s1 , . . . ,sl ) and Θ = (Θ1 , . . . ,Θt ). Let I ⊂ C {s} be an
k n+1+l+t
ideal. Assume that F ∈ B(ρ,τ ,σ) for a suitable polyradius (ρ,τ ,σ) ∈ R . Then for
l
all ε > 0, there exists a K = K(ε) > 0 and a τ ∈ R+ , τ ≤ τ , with the following property.
(e) (e)
Let e ∈ N and θ(e) (x,s) = θ1 (x,s), . . . ,θt (x,s) such that
(1) the θi (x,s) are polynomials of degree ≤ e in s with coefficients in C {x},

(2) the θi (x,s) are in normal form with respect to I,
(e)
(3) kθν (x,s)k(ρ,λτ ) ≤ K for ν = 1, . . . ,t and some λ, 0 < λ ≤ 1.
Let Fe ∈ C {x,s}k be the degree e + 1 part in s of the normal form of F (x,s,θ(e) (x,s))
with respect to I, then:
kFe k(ρ,λτ ) ≤ Kε.
Proof. Let 0 < α < 1 be given. Then there exists a polyradius τ ∈ Rl+ such that for all
f ∈ B(ρ,τ )
1
(8.7) k NF(f |I)k(ρ,τ ) ≤ kf k(ρ,τ ).
1−α
Given α and (ρ,τ ), this inequality still holds if we replace τ by λτ for any λ with 0 < λ ≤ 1,
see Remark 7.1.7. We develop F :
X
F (x,s,Θ) = Fν (x,s)Θν .
ν∈Nt
As this series is convergent, there exists an A > 0 such that kFν (x,s)k(ρ,τ ) ≤ A|ν| . (cf.
Exercise 8.2.6). We choose K so small that AK < 1. It is an exercise to show that we
can choose K = K(ε) so small that even
X 1 ε(1 − α)
(AK)|ν| = t
− 1 − tAK < K ,
(1 − AK) 3
|ν|≥2
see Exercise 8.2.7. We write:
Fν (x,s) = Fν (x,0) + Fν′ (x,s) with Fν′ ∈ (s) for |ν| ≤ 1.
We choose δ so small that δ < ε(1−α)K3 , and δ < ε(1−α)

3t . By making τ smaller we may
′
assume that kFν (x,s)k(ρ,τ ) ≤ δ for all ν with |ν| ≤ 1. Now
F (x,s,θ(e) (x,s)) = F0 (x,0) + F0′ (x,s)

X X X
+ Fν (x,0)θ(e) (x,s)ν + Fν′ (x,s)θ(e) (x,s)ν + Fν (x,s)θ(e) (x,s)ν .
|ν|=1 |ν|=1 |ν|≥2
By assumption the θ(e) (x,s) are in normal form with respect to P I. Moreover the Fν (x,0)
for all ν with |ν| ≤ 1 do not depend on s. It follows that F0 (x,0)+ |ν|=1 Fν (x,0)θ(e) (x,s)ν
P
is in normal form with respect to I. Furthermore F0 (x,0) + |ν|=1 Fν (x,0)θ(e) (x,s)ν does
not have terms of degree e + 1 in the variables s. It follows that Fe , which by definition
is the degree e + 1 part in s of the normal form of F (x,s,θ(e) (x,s)), is equal to the degree
e + 1 part of the normal form of
X X
F0′ (x,s) + Fν′ (x,s)θ(e) (x,s)ν + Fν (x,s)θ(e) (x,s)ν .
|ν|=1 |ν|≥2
Thus, because of (8.7)

X X
(1 − α)kFe k(ρ,λτ ) ≤ kF0′ (x,s) + Fν′ (x,s)θ(e) (x,s)ν + Fν (x,s)θ(e) (x,s)ν k(ρ,λτ )
|ν|=1 |ν|≥2
X
|ν|
≤ δ + tδK + (AK) .
|ν|≥2
By choice of K and δ each term on the right hand side is smaller than or equal to
(1 − α) εK
3 . Hence kFe k(ρ,λτ ) < εK, as was to be shown.

∂F
Step 1. We first reduce to the case that φ(e0 ) and ψ (e0 ) are in m := (s) and ∂Φ ,

∂F k
i s=0
∂Ψi s=0 ∈ C {x} , that is, they do not depend on the Φ and Ψ. Indeed, we can write
X X
φ(e0 ) = c + sν (cν + φ(e
ν
0 )′
) ψ (e0 ) = d + sν (dν + ψν(e0 )′ ).
(e )′ (e )′ (e0 )′ (e )′
Here c,d,cν and dν are in C {x}. Moreover φν 0 = {φi,ν0 }i=1,...,p , ψν = {ψi,ν0 }i=1,...,q
(e )′ (e )′
with φi,ν0 ∈ m, ψi,ν0 ∈ m · C {x,s}. Define new variables Φ′ = {Φ′iν },Ψ′ = {Ψ′iν }, and
define X X
F ′ (x,s,Φ′ ,Ψ′ ) := F (x,s,c + sν (cν + Φ′ν ),d + sν (dν + Ψ′ν )).
′ ′ ′ ′
Now F ′ has a solution φ(e0 ) ,ψ (e0 ) of order e0 . Replacing F,φ(e0 ) , ψ (e0 ) by F ′ ,φ(e0 ) ,ψ (e0 )
proves the claim.
Step 2. We can decide whether F (φ,ψ) ∈ I by looking at the normal form NF(F (φ,ψ)|I).
In fact, in 7.2.10 it is proved that:
F (φ,ψ) ∈ I ⇐⇒ NF(F (φ,ψ)|I) = 0.
The normal form NF(F (φ,ψ)|I) is a power series, all of whose terms are not in L(I), the
ideal of leading terms of I. Moreover, NF(F (φ,ψ)|I) is congruent to F (φ,ψ) modulo I. If
(φ,ψ) is a solution of F ≡ 0 modulo I, then it follows immediately that (NF(φ|I), NF(ψ|I))
is also a solution
P of F ≡ 0 modulo I. Similar remarks
P hold for solutions of order e, as
obviously aα sα being in normal form implies |α|≤e aα sα is in normal form. So we
will and may assume that all our solutions are in normal form.
n+1+l+p+q k
Step 3. We now choose the polyradius. Let (ρ,τ ,σ) ∈ R+ such that F ∈ B(ρ,τ ,σ) .
We consider the elements
∂F ∂F
s=0
∈ C {x}k and ∈ C {x}k , for i = 1, . . . ,p and j = 1, . . . ,q.
∂Φ i ∂Ψ j s=0
n+1
By Cartan’s

∂F
Lemma 8.2.4 we can choose a ρ ∈ R+ , ρ ≤ ρ and M
∂F k
= M (ρ) ≥ 1 depending
on ∂Ψ s=0 and ∂Φ s=0 with the following property. Let G ∈ B(ρ,τ ) , and δ 1 , . . . ,δ p ∈ C [s]
and γ 1 , . . . ,γ q ∈ C {x}[s] such that
p
X Xq
∂F ∂F
G+ δi s=0
+ γi = 0,
i=0
∂Φi i=0
∂Ψi s=0
and G, δ 1 , . . . ,δ p , γ 1 , . . . ,γ q are homogeneous with respect to s of degree e + 1 for e ≥ e0 .

Then there exist δ1 , . . . ,δp ∈ C [s] and γ1 , . . . ,γq ∈ C {x}[s] satisfying this equation being
homogeneous as before, where, moreover,
kδi kτ , kγi k(ρ,τ ) ≤ M kGk(ρ,τ ) for all τ ≤ τ .

1
We take ε = M . Let K = K(ε) and τ ≤ τ as in the previous lemma, applied to F (x,s,Φ,Ψ)
and let (φ ,ψ ), a solution of order e0 . Because φ(e0 ) ∈ m and ψ (e0 ) ∈ mC {x,s} we
(e0 ) (e0 )
can choose, by scaling, the polyradius τ so small that, moreover, kφ(e0 ) kτ < K and
kψ (e0 ) k(ρ,τ ) < K.
Step 4. We will for all e ≥ e0 +1 construct solutions (φ(e) ,ψ (e) ) of order e with (φ(e) ,ψ (e) ) ≡
(φ(e−1) ,ψ (e−1) ) modulo me and with the estimates:
kφe kτ , kψe k(ρ,τ ) < K.
Here φe and ψe are the degree e parts in s of φ(e) and ψ (e) . Let (φ(e) ,ψ (e) ) be a solution
of order e, such that (φ(e) ,ψ (e) ) are polynomials in s of degree ≤ e, in normal form with
respect to I and kφe kτ ≤ K, kψe k(ρ,τ ) ≤ K. Let Fe be the degree e + 1 part in s of
K
the normal form of F (x,s,φ(e) ,ψ (e) ). By the previous lemma, kFe k(ρ,τ ) < εK = M . By
(e)
assumption, there exist δ and γ (e) homogeneous of degree e + 1 in s and in normal
(e)
form with respect to I such that (φ(e) + δ ,ψ (e) + γ (e) ) is a solution of order e + 1. By
Taylor expansion and taking the degree e + 1 part of the normal form we obtain
Xp Xq
∂F (e) ∂F (e)
Fe + |s=0 δ i + |s=0 γ i = 0.
i=1
∂Φ i i=1
∂Ψ i
(e) (e)
By choice of M there exist such δi and γi , having moreover the estimates:
(e) (e) K
kδi kτ , kγi k(ρ,τ ) < M kFe k(ρ,τ ) < M = K.
M
Now φ(e+1) := φ(e) + δ (e) and ψ (e+1) := ψ (e) + γ (e) . Then φe+1 = δ (e) and ψe+1 = γ (e) ,
and the desired estimate holds.
P P∞
Step 5. We now look at the formal solution φ = ∞ e=0 φe , ψ = e=0 ψe . Then, because
φe and ψe are homogenous of degree e with respect to s, and kφe kτ < K, kψe k(ρ,τ ) < K
we obtain
∞
X ∞
X

1 e

1 e
kφkτ /2 < 2 · K = 2K, kψk(ρ,τ /2) < 2 · K = 2K,
e=0 e=0
showing the convergence of φ and ψ. This concludes the proof of Grauert’s Approximation
Theorem.
Exercises
P
8.2.6. Let f (x,Θ) = ν∈Nt fν Θν . Let (ρ,τ ) be a polyradius with kf k(ρ,τ ) < ∞. Show that there
exists an A with kfν kρ ≤ A|ν| .
1
8.2.7. Let c > 0, prove that for all x with |x| sufficiently small (1−x) t < (c + t)x + 1.
(Hint: Check the monotony behavior of the real function corresponding to the above equality.)
8.3 Some other Approximation Theorems 299
8.3 Some other Approximation Theorems
Now we would like to discuss, without proof, a generalization of Artin’s Approximation

Theorem 8.1.1:
Theorem 8.3.1. Let f1 , . . . ,fm ∈ C {x,y}. There is a function Θ : N → N with the
following property.
Let fi (y) ≡ 0 mod (x)Θ(c) , i = 1, . . . ,m, for y ∈ (x)C {x}N , then there exist
y ∈ C {x}N , y≡y mod (x)c and
fi (y) = 0, i = 1, . . . ,m.
The point here is that one does not even have to find a formal solution, but only
a solution up to high enough order. The weakness of this theorem however is, that
it seems not to be possible in general to compute the function Θ. The theorem is a
consequence of Artin’s Approximation Theorem and the following theorem, proved in
[Pfister-Popescu 1975], [Kurke et al. 1978].
Theorem 8.3.2. Let K be a field and f1 , . . . ,fm ∈ K[[x,y]], then there exists a function
Θ : N → N with the following property:
Let fi (y) ≡ 0 mod (x)Θ(c) , i = 1, . . . ,m, for y ∈ (x)K[[x]]N . Then there exists
y ∈ K[[x]]N such that y ≡ y mod (x)c and fi (y) = 0, i = 1, . . . ,m.
Artin’s Approximation Theorem 8.1.1 has been generalized in many directions. The
most general result is due to Dorin Popescu (cf. [Popescu 1985], [Popescu 1986, 1990])
and has been reproved by Marc Spivakovsky, see [Spivakovsky 1999] and Richard Swan
[Swan 1998].
Theorem 8.3.3 (Popescu). Let A be an excellent Noetherian, and with respect to an
bN
ideal a, Henselian ring. Let f1 , . . . ,fm ∈ A[Y ]. Suppose fi (y) = 0, i = 1, . . . ,m, y ∈ Aa
(the so-called a–adic completion). Let c > 0 be an integer. Then there exists a y ∈ AN
such that y ≡ y mod ac and fi (y) = 0, i = 1, . . . ,m.
We do not want to explain the notion of excellence. This can be found in Matsumura’s
book (cf. [Matsumura 1979]). Almost all rings occurring in this book, or more generally
in algebraic geometry, are excellent, as, for instance, the rings K[[x]],C {x},K[x] their
quotients and their polynomial extensions.
The ring A is Henselian with respect to the ideal a if Hensel’s Lemma holds with
respect to a (cf. Corollary 3.3.21, replace there (x) by a, cf. also [Kurke et al. 1975]).
Another important example of an excellent Henselian local ring is the ring Khxi of
algebraic power series over a field K, that is Khxi = {f ∈ K[[x]], F (f ) = 0 for a suitable
F ∈ K[x,T ], F 6= 0}.
We would like to finish this chapter with some important remarks concerning gen-
eralizations of the previous theorem.
Remark 8.3.4. A generalization of Theorem 8.3.3 similar to Theorem 8.3.1 is not pos-
sible (cf. a counterexample of Spivakovsky [Spivakovsky 1994]).
Remark 8.3.5. Assume that some of the y i in Theorem 8.1.1 do not depend on some of
the xi . In general, it is not possible to find convergent solutions y = (y1 , . . . ,yN ) with the
same property. Gabrielov, see [Gabrielov 1971] gave a counterexample of the following
type:
f = aY1 + bY2 + cY3 + d ∈ C {x1 ,x2 ,x3 , Y1 ,Y2 ,Y3 }, y = (y 1 ,y2 ,y3 ) ∈ C [[x1 ,x2 ,x3 ]]
such that f (y) = 0, y2 ,y3 ∈ C [[x1 ,x2 ]] and there are no y = (y1 ,y2 ,y3 ) ∈ C {x1 ,x2 ,x3 }
with f (y) = 0 and y2 ,y3 ∈ C {x1 ,x2 }.
Remark 8.3.6. In the following special situation the problem of Remark 8.3.5 can be
solved.
Let K be a field and f1 , . . . ,fm ∈ Khx,Y i, assume fi (y) = 0 for y ∈ (x)K[[x]].
Assume yi ∈ K[[x1 , . . . ,xvi ]] 1 ≤ v1 ≤ · · · ≤ vN ≤ n. Let c > 0 be an integer, then there
exists a y ∈ KhxiN such that
y ≡ y mod (x)c
yi ∈ Khx1 , . . . ,xvi i
fi (y) = 0, i = 1, . . . ,m.
As we will not explain here, this theorem is a consequence of Popescu’s general
theorem (cf. [Popescu 1985], [Popescu 1986, 1990]). We obtain the following corollary:
Corollary 8.3.7 (Grauert’s Theorem for Algebraic Power Series). We take the
notations of and assumptions of 8.2.1 and 8.2.2. Assume furthermore that I ⊂ C hsi
and F ∈ C hx,s,φ,ψik . Then the system of equations F ≡ 0 modulo I has an algebraic
solution (φ,ψ) with φ ≡ φ(e0 ) modulo me0 +1 , and ψ ≡ ψ (e0 ) modulo me0 +1 .
Artin’s Approximation Theorem and its generalizations have many applications. As
an application we give Artin’s answer to a question of Grauert.
Proposition 8.3.8. Let A,B,C be analytic C –algebras and w : C −→ B a homomor-
phism.
(1) Assume that for a surjective homomorphism ϕ : A −→ B the following diagram is
commutative:
B /B b
G OO OO
ϕ ϕ̂

w
A b
7/ A

ooooo
ooo
ooooo u
o
C
Then, for a given positive integer c, there exists a u : C −→ A such that ϕ ◦ u = w
and u ≡ u mod mcAb.
(2) Assume the following diagram is commutative:
Ab _?
O ?
?? u
??
??
AO ?B
~~~
v
~~w
~~
C
Then for a given positive integer c there exists a u : B −→ A such that v = u◦w and
b −→ A
u ≡ u mod mcAb. If û : B b is an isomorphism, then u is also an isomorphism.
8.3 Some other Approximation Theorems 301
Proof. Let A = C {x}/(f1 , . . . ,fa ), B = C {y}/(g1 , . . . ,gb ), C = C {z}/(h1, . . . ,hc ), x =

(x1 , . . . ,xn ), y = (y1 , . . . ,ym ), z = (z1 , . . . ,zr ).

In case of (1), we may assume y = x and ϕ xi mod (f1 , . . . ,fa ) = xi mod (g1 , . . . ,gb ).
Let zi (x) ∈ C {x} be representatives of w(zi ) and y i (x) ∈ C [[x]] be representatives
of u(zi ) (note that A b = C [[x]]/(f1 , . . . ,fa )). Then we have, because of the commutativity
of the diagram zi (x) ≡ yi (x) mod (g1 , . . . ,gb ), that is,
b
X
zi (x) − yi (x) = αij gj , i = 1, . . . ,r
j=1

for suitable αij ∈ C [[x]]. Furthermore, we have hj y(x) ≡ 0 mod (f1 , . . . ,fa ) because

of the definition of y and the fact that C = C {z}/(h1, . . . ,hc ). This implies hi y(x) =
P
a
β ij fj for suitable β ij ∈ C [[x]].
j=1
Now consider the following systems of equations
b
X
zi (x) − yi − Aij gj = 0
j=1
Xa
hi (y) − Bij fj = 0
j=1
in the variables y,{Aij },{Bij }. It has the formal solution (y,{αij },{β ij }). Using Artin’s

Approximation Theorem, we obtain for a given c an analytic solution y,{αij },{βij }
such that (y,{αij },{βij }) ≡ (y,{αij },{β ij }) mod (x1 , . . . ,xn )c .
We define u : C −→ A by u(zi ) = yi (x) mod (f1 , . . . ,fa ). Because of hi (y) ≡ 0
mod (f1 , . . . ,fa ) it is well-defined and because of zi (x) ≡ yi (x) mod (g1 , . . . ,gb ) we have
ϕ ◦ u = w.
To prove (2), let zi′ (x) ∈ C {x} (respectively zi′′ (y) ∈ C {y}) be representatives of v(zi )
(respectively w(zi )). Furthermore, let yi (x) ∈ C [[x]] be representatives of u(yi ). Similar
to case (1), we obtain
a
X

zi′′ y(x) − zi′ (x) = αij fj
j=1
Xa

gi y(x) = β ij fj
j=1
for suitable αij ,β ij ∈ C [[x]]. Similar to the first case, the approximation of this formal
solution defines the morphism u : B −→ A with the required properties.
We leave it as an exercise to prove that u is an isomorphism if û : B b −→ A b was
already an isomorphism.
302
9 Classification of Simple Hypersurface

Singularities
This chapter is devoted to the study of hypersurface singularities. There are various
equivalence relations one can consider on the set of all functions f ∈ C {x1 , . . . ,xn }. The
two most important ones are right equivalence and contact equivalence. Two functions
f and g are called right equivalent if there exists an automorphism ϕ of C {x1 , . . . ,xn }
such that ϕ(f ) = g. They are called contact equivalent if there exists an automorphism
ϕ and a unit u in C {x1 , . . . ,xn } such that ϕ(f ) = u · g. Until now we considered contact
equivalence in this book (without using this name), but in this chapter we will mainly
consider right equivalence, because in this case the theorems are somewhat easier to
formulate and to prove. (One does not have to worry about the unit.) The aim of this
chapter is to give a beginning of the classification of functions up to right equivalence.
The first thing is to consider Finite Determinacy Theorems. Let m be the maximal ideal
of C {x1 , . . . ,xn }. A function f is called (right) k–determined if for all elements g with
f − g ∈ mk+1 the functions f and g are right equivalent. In particular, a k–determined
f is right equivalent to a polynomial of degree at most k. A function is called finitely
determined, if there exists a k ∈ N such that f is k–determined. It is maybe a little
surprising that all functions which have an isolated singularity are finitely determined.
The proof, however, is not difficult at all: it is a direct application of Newton’s Lemma,
see 3.3.34. This result is not strong enough for our purposes, however. We want to find
a small k such that f is k–determined. So we prove that, if mk+1 ⊂ m2 J(f ), then f
is k–determined. (Here J(f ) is the Jacobian ideal, that is, the ideal generated by the
partial derivatives of f .) This is what is usually called the Finite Determinacy Theorem,
and is more difficult to prove. The idea of the proof is to connect f and g via the path
f + a(g − f ). So for a = 0 we get f , and for a = 1 we get g. For all a, we want to prove
that there exists a small neighborhood of a in C such that for all b ∈ U the functions
f + b(g − f ) and f + a(g − f ) are right equivalent. In order to reach this goal, we need a
criterion of local triviality. So, consider an element F ∈ C {x1 , . . . ,xn ,t} which we view
as a family of functions with parameter t. Then
we prove that this family is trivial if
∂F ∂F ∂F
and only if ∂t ∈ (x1 , . . . ,xn ) ∂x1 , . . . , ∂xn . In order to prove this, we construct an
automorphism in the formal power series ring. The existence of a convergent one then
follows from Artin’s Approximation Theorem. We will show that this condition holds in
the family f + ag, so that compactness of [0,1] then completes the proof of the Finite
Determinacy Theorem.
As an application of the local triviality we will prove the Mather-Yau Theorem.
It says that f and g are contact equivalent, if and only if their Tjurina algebras are
isomorphic. We will also prove a corresponding statement for right equivalence. The first
reduction in the proof of the Mather-Yau Theorem is to the case that (f ) + J(f ) =
(g) + J(g). Again we connect the functions by a path f + a(g − f ). But now one has to be
more careful, because it is certainly not true that for all a ∈ C the functions f + a(g − f )
and f are contact equivalent. For example, the functions f and −f are contact equivalent,
but for a = 12 we get that f + 21 (−f − f ) = 0, which is certainly not contact equivalent to
f . But at least we will show that for all a ∈ C , except for a discrete subset, the functions
9.1 Finite Determinacy of Hypersurface Singularities 303
f and f + a(g − f ) are contact equivalent. As the complement of a discrete subset is still
path connected, this is sufficient for the proof of the Mather-Yau Theorem.
In Section 9.2 we start with the classification of functions up to right equivalence. We
first define what a simple function is. This is done by putting a topology on C {x1 , . . . ,xn }.
A function f is then called simple, if there are only finitely many equivalence classes in a
sufficiently small neighborhood of f . The goal of this chapter is to prove that the simple
singularities are given by the A-D-E–singularities. We first prove the Splitting Lemma.
This allows us to study functions which lie in m3 . Then, via a sequence of arguments, it
is shown that the A-D-E–singularities are simple, and that for nonsimple singularities
f we may assume that either n ≥ 3 and f ∈ m3 , or n = 2 and f ∈ m4 , or n = 2 and
f ∈ (x,y 2 )3 or f is nonisolated.
In Section 9.3, we then show that functions with one of these properties are not
simple. The idea is to use the Finite Determinacy Theorem, to reduce the study of right
equivalence classes to the study of orbits under a group action in C N for some N ≫ 0.
In fact, we study the group action on the k–jets, which are the Taylor series of f up to
order k + 1. These orbits are orbits under the action of an algebraic group. This means
that all the actions are given by polynomial functions. We then apply some algebraic
geometry to prove that orbits of algebraic group actions are locally closed in the Zariski
topology. (That is, the intersection of an open and closed subset.) As all orbits contain
a smooth point, we can use the group action to show that all points of the orbit are
smooth. The next thing to do is to compute the tangent space. We will see that it is
equal to mJ(f ) + mk+1 /mk+1 . In particular, we get the dimension of the orbit of f . We
deduce that if g ∈ / mJ(f ), then in the family f + t · g there are only finitely many t such
that f + tg is right equivalent to f . The knowledge of the dimension of the orbits is our
main tool for showing that certain functions are not simple.
9.1 Finite Determinacy of Hypersurface Singularities
Definition 9.1.1. Let f, g ∈ m ⊂ C {x1 , . . . ,xn }, where m is, as usual, the maximal
ideal.
(1) One says that f is right equivalent to g, f ∼ g, if there exists an automorphism ϕ
R
of C {x1 , . . . ,xn } such that ϕ(f ) = g.
(2) One says that f is contact equivalent to g, f ∼ g, if there exists an automorphism
C
ϕ of C {x1 , . . . ,xn } such that (ϕ(f )) = (g), that is, there exists a unit u such
that ϕ(f ) = ug. Thus f and g are contact equivalent exactly if the C –algebras
C {x1 , . . . ,xn }/(f ) and C {x1 , . . . ,xn }/(g) are isomorphic. This is exactly the same
as saying that the germs of the analytic hypersurfaces defined by f and g are
isomorphic.
In this chapter, we will mainly work with right equivalence. Similar results hold for
contact equivalence, but the formulation and proofs will be left to the reader.
Definition 9.1.2. Let f ∈ m ⊂ C {x1 , . . . ,xn }. Then f is called k–determined if all
g ∈ C {x1 , . . . ,xn } with f − g ∈ mk+1 are right equivalent to f . If f is k–determined for
some k ∈ N, then f is called finitely determined. In particular, a k–determined f is right
equivalent to a polynomial.
304 9 Simple Hypersurface Singularities
The fact that functions which have an isolated singularity are finitely determined
follows quite easily from Newton’s Lemma 3.3.34, as we will show now.
Theorem 9.1.3. Let f ∈ m2 ⊂ C {x1 , . . . ,xn }. Suppose mk+1 ⊂ mJ(f )2 . Then f is
k–determined. In particular, functions with isolated singularities are finitely determined.
Proof. Let g ∈ m such that f − g ∈ mk+1 . Consider the following system of equations,
which we want to solve for y1 , . . . ,yn ∈ C {x1 , . . . ,xn }:
F (x,y) := f (y) − g(x) = 0.

∂F
∂f

We start with the initial approximation y i (x) = xi . Obviously ∂yi x,y(x) = ∂yi y(x) =
∂f
As by assumption F (x,y) = f (x) − g(x) ∈ mk+1 ⊂ mJ(f ) , we can apply Newton’s
∂xi .
2
Lemma 3.3.34. There exist y1 , . . . ,yn ∈ C {x1 , . . . ,xn } with

• F (x,y) = 0,
• yi (x) ≡ xi mod mJ(f ). Note that mJ(f ) ⊂ m2 .
This says that the y give a coordinate transformation transforming f into g.
So the fact that germs of holomorphic functions with an isolated singularity are right
equivalent to a polynomial is a relatively easy consequence of Newton’s Lemma. For our
purposes we need, however, a stronger version of the Finite Determinacy Theorem. The
statement of the above theorem can be strengthened to the following statement.
Theorem 9.1.4 (Finite Determinacy Theorem). Let f ∈ m2 ⊂ C {x1 , . . . ,xn }. Suppose
that mk+1 ⊂ m2 J(f ). Then f is k–determined. In particular, if k ≥ µ(f ) + 1, then f is
k–determined.
Note that from mk ⊂ mJ(f ) it follows that mk+1 ⊂ m2 J(f ). The last condition is
somewhat weaker. See Exercise 9.1.13.
The basic idea of the proof of the Finite Determinacy Theorem is to connect f and
g via the path Ft = f + t · (g − f ), for 0 ≤ t ≤ 1. We need a criterion whether for given t,s
the element Ft is right equivalent to Fs . The following theorem gives us such a criterion
“locally”. By this we mean that we give a criterion that Ft is right equivalent to F0 , for
small t.
Theorem 9.1.5. Let F ∈ C {x1 , . . . ,xn ,t} and c ≥ 0 be an integer. The following condi-
tions are equivalent:

(1) ∂F
∂t ∈ (x1 , . . . ,xn )c ∂F
∂x1 , . . . , ∂F
∂xn .
(2) There exists a ϕ = (ϕ1 , . . . ,ϕn ) ∈ C {x1 , . . . ,xn ,t}n such that
(a) ϕi (x1 , . . . ,xn ,0) = xi ,
(b) ϕi − xi ∈ (x1 , . . . ,xn )c ,
(c) F (ϕ1 , . . . ,ϕn ,0) = F (x1 , . . . ,xn ,t).

∂F ∂F ∂F
Corollary 9.1.6. Suppose that ∂t ∈ (x1 , . . . ,xn ) ∂x1 , . . . , ∂xn for some
F ∈ C {x1 , . . . ,xn ,t}. Then there exists a small neighborhood U of 0 in C such that
for all fixed a ∈ U the function Fa := F (x1 , . . . ,xn ,a) is right equivalent to F0 .
Proof. Consider the elements ϕ1 , . . . ,ϕn ∈ C {x1 , . . . ,xn ,t} which exist according to 9.1.5.
There exists a smallopen neighborhood
U of 0 in C , so that for all a ∈ U , the ϕi (x,a) are
∂ϕi (x,a)
convergent and det ∂xj (0) 6= 0. Thus for fixed a ∈ U , (ϕ1 , . . . ,ϕn ) is an automor-
phism of C {x1 , . . . ,xn }. Moreover, formula (c) in Theorem 9.1.5 says that it transforms
F (x1 , . . . ,xn ,0) into F (x1 , . . . ,xn ,a).
Step 1. We first prove the easy implication (2) =⇒ (1). As ϕi (x1 , . . . ,xn ,0) = xi it
follows easily from the Inverse Function Theorem that (ϕ1 , . . . ,ϕn ,t) is an automorphism
of C {x1 , . . . ,xn ,t}. Let ψ = (ψ1 , . . . ,ψn ,t) be its inverse. Then ψi − xi ∈ (x1 , . . . ,xn )c , as
one easily checks.
By assumption, we have F (ϕ,0) = F (x,t). Compose this with ψ. Then F (x,0) =
F (ψ,t). Hence the derivative of F (ψ1 , . . . ,ψn ,t) with respect to t is zero. Thus
n
X ∂F ∂ψi ∂F
(9.1) (ψ1 , . . . ,ψn ,t) + (ψ1 , . . . ,ψn ,t) = 0.
i=1
∂xi ∂t ∂t
Compose (9.1) with the automorphism (ϕ1 , . . . ,ϕn ,t). Then
Xn
∂F ∂ψi ∂F
(9.2) · (ϕ1 , . . . ,ϕn ,t) + = 0.
i=1
∂xi ∂t ∂t
We already noted that ψi −xi ∈ (x1 , . . . ,xn )c . But then also ∂ψ ∂t (ϕ1 , . . . ,ϕn ,t) ∈ (x1 , . . . ,xn )
i c
follows. So formula (9.2) gives the implication (2) =⇒ (1).

Step 2. We now assume that ∂F ∂t ∈ (x 1 , . . . ,xn )c ∂F
∂x1 , . . . , ∂F
∂xn . By Artin’s Approxima-
tion Theorem, or better its (slightly) more general version given in Exercise 8.1.6, it
follows that it suffices to find a formal ϕ = (ϕ1 , . . . ,ϕn ) ∈ C [[x1 , . . . ,xn ,t]]n such that
(1) ϕi (x1 , . . . ,xn ,0) = xi ,
(2) ϕi − xi ∈ (x1 , . . . ,xn )c ,
(3) F (ϕ1 , . . . ,ϕn ,0) = F (x1 , . . . ,xn ,t).
We now construct those formal ϕ1 , . . . ,ϕn . By assumption, we can write

X ∂F n
∂F
= ξi , ξi ∈ (x1 , . . . ,xn )c .
∂t i=1
∂xi
∂
Pn ∂
Consider the differential operator δ := − ∂t + i=1 ξi ∂xi . Let s be a new variable and
consider the map
Φ : C [[x1 , . . . ,xn ,t,s]] −→ C [[x1 , . . . ,xn ,t,s]]

X∞
1 k k
a 7−→ exp(sδ)(a) := k! δ (a)s for a ∈ C [[x1 , . . . ,xn ,t]]
k=0
s 7−→ s.
Put Φi := Φ(xi ) for i = 1, . . . ,n, and Φn+1 = Φ(t). Note that Φn+1 = Φ(t) = t − s. Now
define ϕi := Φi (x1 , . . . ,xn ,t,t) for i = 1, . . . ,n. We have to show that these ϕi satisfy our
conditions.
(1) The first condition is easy. Indeed, ϕi (x1 , . . . ,xn ,0) by definition is equal to δ 0 (xi ) =
xi .
(2) The second condition is also not too difficult. Indeed from the first part we have
X
1 k k
Φ(xi ) = xi + k! δ (xi )s .
k≥1
From the definition of δ, and the fact that the ξi ∈ (x1 , . . . ,xn )c it follows immediately
that Φ(xi ) − xi ∈ (x1 , . . . ,xn )c . The second condition now follows.
(3) The third condition is the most difficult one. The crucial thing to show is
∂F (Φ1 , . . . ,Φn+1 )
(9.3) = 0,
∂s
that is, F (Φ1 , . . . ,Φn ,Φn+1 ) does not depend on s. Indeed, assuming (9.3) it follows that
F (ϕ1 , . . . ,ϕn ,0) = F (Φ1 , . . . ,Φn ,Φn+1 )s=t =
F (Φ1 , . . . ,Φn ,Φn+1 )s=0 = F (x1 , . . . ,xn ,t).
The first equality holds by definition of the ϕi , the second equality because of (9.3). The
third equality is because Φi |s=0 = xi for i ≤ n, and Φn+1 |s=0 = t.
It therefore remains to show (9.3). We have the following formulae:
∂Φi
(9.4) = ξi (Φ1 , . . . ,Φn ,Φn+1 ),
∂s
∂Φn+1
(9.5) = −1,
∂s
(9.6) δ(F ) = 0.
The second and the third formula follow immediately from the fact that Φn+1 = t − s,
and the definition of δ. To prove formula (9.4), consider an arbitrary a ∈ C [[x1 , . . . ,xn ,t]],
and differentiate:
∞ ∞
∂Φ(a) X 1 X 1
= δ k (a)sk−1 = δ k−1 (δ(a))sk−1 = Φ(δ(a)).
∂s (k − 1)! (k − 1)!
k=1 k=1
In particular for i = 1, . . . ,n we have

∂Φi
= Φ δ(xi ) = Φ(ξi ) = ξi (Φ1 , . . . ,Φn ,Φn+1 ).
∂s
The last equality follows as Φi = Φ(xi ) for 1 ≤ i ≤ n, and Φ(t) = Φn+1 , by definition.
By using the chain rule and the formulae (9.4), (9.5) and (9.6) we get
n
∂F (Φ1 , . . . ,Φn+1 ) X ∂F ∂Φi ∂F
= (Φ1 , . . . ,Φn+1 ) − (Φ1 , . . . ,Φn+1 )
∂s i=1
∂xi ∂s ∂t
Xn
∂F ∂F
= (Φ1 , . . . ,Φn+1 )ξi (Φ1 . . . ,Φn+1 ) − (Φ1 , . . . ,Φn+1 )
i=1
∂xi ∂t
= δ(F )(Φ1 , . . . ,Φn+1 )
= 0.

Step 1. Let g ∈ mk+1 be given. Let 0 ≤ a ≤ 1. Consider
Fa (x1 , . . . ,xn ,t) := f + (a + t) · g.
For all a we will show that there exists an open neighborhood U of a such that for all
a+t∈ U
(9.7) f + (a + t) · g ∼ f + a · g.
R
Because the unit interval [0,1] is compact we can cover the unit interval with finitely
many U . As right equivalence is an equivalence relation, it then follows that f = F0 is
right equivalent to F1 = f + g.
Step 2. To prove (9.7) we will apply Corollary 9.1.6, and it therefore suffices to prove

∂Fa 2 ∂Fa ∂Fa
(9.8) ∈m ,..., .
∂t ∂x1 ∂xn
Here m = (x1 , . . . ,xn ). By applying Artin’s Approximation Theorem, it suffices to prove

this in the formal power series ring C [[x1 , . . . ,xn ,t]].
Step 3. We will first show the following intermediate result.
Let f ∈ C {x1 , . . . ,xn } and assume mk+1 ⊂ m2 J(f ). Then mk+1 ⊂ m2 J(f + g) for
all g ∈ mk+1 .
In order to prove this, put F := f + g. Let h ∈ mk+1 be given. We have to find

ξ i ∈ m2 such that
n
X ∂F
(9.9) h= ξi .
i=1
∂xi
2
By applying Artin’s Approximation Theorem, it suffices to find ξP i ∈ m C [[x1 , . . . ,xn ]]
k+1 n ∂f
satisfying (9.9). Given h ∈ m , we can by assumption write h = i=1 ξi ∂x i
, ξi ∈ m2 .
But then
Xn Xn
∂F ∂g
(9.10) h= ξi − ξi .
i=1
∂xi i=1
∂xi
∂g
As g ∈ mk+1 and ξi ∈ m2 it follows that ξi ∂x i
in mk+2 . By assumption mk+1 ⊂ m2 J(f ),
so that we can write
n
X n
X (1) ∂f
∂g (1)
(9.11) ξi = ξi , ξi ∈ m3 .
i=1
∂xi i=1
∂xi
(This is the reason we need mk+1 ⊂ m2 J(f ), as the condition mk+1 ⊂ mJ(f ) does not
suffice.) Plugging (9.11) in (9.10) we get
n
X n
X (1) ∂g n
X
(1) ∂F (1) ∂g
h= (ξi + ξi ) − ξ with ξi ∈ mk+3 .
i=1
∂xi i=1 i ∂xi i=1
∂xi
Repeating the procedure, we get ξ i ∈ m2 C [[x1 , . . . ,xn ]] satisfying formula (9.9).

Step 4. The rest of the proof runs similar to the third step. Applying the third step to
ag we obtain that mk+1 ⊂ m2 J(f + ag). Hence
Xn
∂(f + ag)
g= ξi , for some ξi ∈ m2 .
i=1
∂xi
This implies
X ∂Fa n X ∂g n
∂Fa
(9.12) =g= ξi −t· ξi .
∂t i=1
∂xi i=1
∂xi
∂g
As g ∈ mk+1 , it follows that ξi ∂x i
∈ mk+2 . Thus we can write
n
X n
X (1) ∂(f + ag)
∂g (1)
(9.13) ξi = ξi , ξi ∈ m3 .
i=1
∂xi i=1
∂xi
Plugging in (9.13) in (9.12) we get
X n
∂Fa (1) ∂Fa
g= = (ξi − tξi ) mod t2 mk+3 .
∂t i=1
∂xi
Iterating this procedure we obtain

X ∂Fa n
∂Fa
= ξi , ξ i ∈ m2 C [[x1 , . . . ,xn ,t]].
∂t i=1
∂xi
Step 5. For the final statement we apply Exercise 1.3.20. It follows that mk−1 ⊂ J(f ),
from which mk+1 ⊂ m2 J(f ) follows. So we can apply the first part of the theorem.
Theorem 9.1.5 can be generalized as follows:
Theorem 9.1.7 (Characterization of Local Analytic Triviality).
Let f ∈ C {x1 , . . . ,xn ,y1 , . . . ,ym } and c ≥ 0 be an integer. The following conditions are
equivalent:

∂f c ∂f ∂f
(1) ∂y i
∈ (x1 , . . . ,xn ) ∂x1 , . . . , ∂xn + (f ) for i = 1, . . . ,m.
(2) There exist ϕ1 , . . . ,ϕn ,u ∈ C {x1 , . . . ,xn ,y1 , . . . ,ym } such that
• u(x1 , . . . ,xn ,0, . . . ,0) = 1

• ϕi (x1 , . . . ,xn ,0, . . . ,0) = xi
• ϕi − xi ∈ (x1 , . . . ,xn )c
• f (x1 , . . . ,xn ,y1 , . . . ,ym ) = u · f (ϕ1 , . . . ,ϕn ,0, . . . ,0).

∂f ∂f ∂f
If moreover ∂y i
∈ (x1 , . . . ,xn )c ∂x 1
, . . . , ∂x n
for all i, then we can choose u = 1.
The proof is similar to the proof of 9.1.5 and will be left as Exercise 9.1.15. We now
come to the Theorem of Mather and Yau.
Theorem 9.1.8 (Mather-Yau). Let f and g ∈ m = (x1 , . . . ,xn ) ⊂ C {x1 , . . . ,xn }. Then
the following conditions are equivalent:
(1) C {x1 , . . . ,xn }/(f ) ∼
= C {x1 , . . . ,xn }/(g), that is, f and g are contact equivalent.

(2) C {x1 , . . . ,xn }/ (f ) + mJ(f ) ∼ = C {x1 , . . . ,xn }/ (g) + mJ(g) as C –algebras.
If f and g define isolated singularities these conditions are equivalent to

(3) C {x1 , . . . ,xn }/ (f ) + J(f ) =∼ C {x1 , . . . ,xn }/ (g) + J(g) as C –algebras.
To put it in another way, the function f and g are contact equivalent if and only if the
Tjurina algebras of f and g are isomorphic.
Remark 9.1.9. It is not true that functions are right equivalent if and only if their Milnor
algebras are isomorphic. Indeed, consider the family of functions ft = x4 + y 5 + t · x2 y 3 .
Then one can show, see Exercise 9.1.16:
(1) ft ∼ f1 ⇐⇒ t = a2 b2 , a4 = 1, b5 = 1.
R
5
(2) Consider the automorphism ϕα of C
{x,y} defined by ϕα (x) = α x, and ϕα (y) =
4 2 −1
α y, with α = t . Then ϕα J(ft ) = J(f1 ).
To find the correct theorem for the case of right equivalence, we consider the Milnor
algebra C {x1 , . . . ,xn }/J(f ) as C {t}–algebra with multiplication t · h := f h for h ∈
C {x1 , . . . ,xn }/J(f ).
Theorem 9.1.10 (Mather-Yau for Right Equivalence).
Let f and g ∈ m = (x1 , . . . ,xn ) ⊂ C {x1 , . . . ,xn }. Then the following conditions are
equivalent:
(1) f ∼ g.
R
(2) C {x1 , . . . ,xn }/mJ(f ) ∼

= C {x1 , . . . ,xn }/mJ(g) as C {t}–algebras.
If f and g define isolated singularities these conditions are equivalent to
∼ C {x1 , . . . ,xn }/J(g) as C {t}–algebras.
(3) C {x1 , . . . ,xn }/J(f ) =
The proof of this theorem is similar to the proof of 9.1.8 and left as Exercise 9.1.20.
Proof of the Mather-Yau Theorem for Contact Equivalence 9.1.8.
The equivalence (1) ⇐⇒ (2) is left as Exercise 9.1.17. The implication (1) =⇒ (3) is
easy, and has been proved in 3.4.27. The implication (3) =⇒ (1) is the most difficult
case, and will be proved in several steps.
Step 1. We first reduce to the case (f ) + J(f ) = (g) + J(g). Let ϕ : C {x1 , . . . ,xn }/ (f ) +
J(f ) ∼ = C {x1 , . . . ,xn }/ (g) + J(g) be a C –algebra isomorphism. We may assume that
), ord(g) ≥ 2. The isomorphism
ord(f ϕ induces an isomorphism between (f ) + J(f ) +
m2 /m2 and (g) + J(g) + m2 /m2 . These are finite-dimensional C –vector spaces. Let k
be its dimension, and let z1 , . . . ,zn ∈ C {x1 , . . . ,xn } induce a basis of m/m2 such that
z1 , . . . ,zk ∈ (f ) + J(f ). Let w1 , . . . ,wn ∈ C {x1 , . . . ,xn } induce a basis of m/m2 such that
w1 , . . . ,wk ∈ (g) + J(g) and ϕ([zk+i ]) = [wk+i ] for i ≥ 1. Define an automorphism ψ on
C {x1 , . . . ,xn } by ψ(zi ) = wi for all i. This is an automorphism by the Inverse Function
Theorem 3.3.7. Then the diagram
ψ
C {x1 , . . . ,xn } / C {x1 , . . . ,xn }
ϕ

C {x1 , . . . ,xn }/ (f ) + J(f ) / C {x1 , . . . ,xn }/ (g) + J(g)

commutes, that is, ψ (f ) + J(f ) = (g) + J(g). Now let h
:= ψ(f ). Then (h) + J(h) =
(g) + J(g), because ψ (f ) + J(f ) = (ψ(f )) + J(ψ(f )) , see 3.4.27. This is what we
wanted to show.
Step 2. We may now assume that (f ) + J(f ) = (g) + J(g). Let h := f − g and F :=
g + th ∈ C {x1 , . . . ,xn ,t}. So we are connecting f and g by a (complex) path: for t = 0 we
have g, and for t = 1 we have f . Unlike in the proof of the Finite Determinacy Theorem,
it is not true however that for all a, the functions g and g + ah are contact equivalent.
In this step we will show that at least for t small the functions g and g + th are contact
equivalent. From (f ) + J(f ) = (g) + J(g), it follows that (h) + J(h) ⊂ (g) + J(g). Hence
Xn
∂F ∂g
= h = ug + ξi
∂t i=1
∂xi
n n
!
X ∂F X ∂h
= uF + ξi − t uh − ξi
i=1
∂xi i=1
∂xi
n n
!
X ∂F X (1) ∂g
= uF + ξi + t u(1) g + ξi
i=1
∂xi i=1
∂xi
Iterating this process we obtain

X ∂F n
∂F
= uF + ξi
∂t i=1
∂xi
for suitable u,ξ i ∈ C [[x1 , . . . ,xn ,t]]. Using Artin’s Approximation Theorem 8.1.1 we ob-
tain
∂F ∂F ∂F
∈ F, ,..., .
∂t ∂x1 ∂xn
Now we can apply Theorem 9.1.7 to obtain

ϕ F (x1 , . . . ,xn ,0) = wF (x1 , . . . ,xn ,t)
for a suitable map ϕ = (ϕ1 , . . . ,ϕn ,t) : C {x1 , . . . ,xn ,t} −→ C {x1 , . . . ,xn ,t} and w a unit
in C {x1 , . . . ,xn ,t}. Now we want to prove that for small t fixed the map (ϕ1 , . . . ,ϕn ,t)
is an automorphism of C {x1 , . . . ,xn }. This is the place where we need the assumption
that f and g define isolated singularities.1
At least the map (ϕ1 , . . . ,ϕn ,t) is an automorphism of C {x1 , . . . ,xn ,t}, so has an
inverse (ψ1 , . . . ,ψn ,t), so that
g = F (x1 , . . . ,xn ,0) = w′ F (ψ1 , . . . ,ψn ,t)

1 Note that this is not necessary for the proof of (1) ⇐⇒ (2). Here the fact that for small t fixed
the map (ϕ1 , .“. . ,ϕn ,t) is an”automorphism follows directly from Theorem 9.1.7 because we have
∂F ∂F ∂F
∂t
∈ (F ) + m ∂x1
, . . . , ∂x .
n

for some unit w′ . Thus V (F ), 0 is as germ of an analytic
space isomorphic to V (g), 0 .
This is the product of an isolated singularity V (g), 0 in (C n , 0) with the germ of a

smooth one-dimensional space (C , 0). Thesingular locus of V (g), 0 in C n+1 , 0 is a
line, so that the singular locus of V (F ), 0 is a smooth curve. Now this curve must be
given by x1 = · · · = xn = 0, as F = g + t · (f − g), and we assumed that the order of
∂F ∂F ∂F
both f and g is at least two, and therefore F, ∂x1 , . . . , ∂xn , ∂t ⊂ (x1 , . . . ,xn ). Define
ai (t) := ψi (0, . . . ,0,t). Then ψi − ai ∈ (x1 , . . . ,xn ), thus for fixed sufficiently small t
(ψ1 − a1 , . . . ,ψn − an ) gives an automorphism of C {x1 , . . . ,xn }. We want to show that
ai = 0. For some small representative, the singular point (0, . . . ,0,t) is mapped via the
automorphism (ψ1 , . . . ,ψn ,t) to a singular point of V (g). As the only singular points of
V (g) are given by x1 = · · · = xn = 0, it follows that ψi (0, . . . ,0,t) = 0. As moreover
ψi (x1 , . . . ,xn ,0) = xi , it follows that ψi − xi ∈ t · (x1 , . . . ,xn ). Thus for fixed small t,
the (ψ1 , . . . ,ψn ), and hence (ϕ1 , . . . ,ϕn ) gives an automorphism of C {x1 , . . . ,xn } by the
Inverse Function Theorem.
(Note that in this step we only used the inclusion (h) + J(h) ⊂ (g) + J(g).)
Step 3. We want to redo the argument in step 2 for the function F := g + ah + th for
almost all a, that is, for all a except for a discrete subset of C . In order for the argument
to work, we need that (h) + J(h) ⊂ (g + ah) + J(g + ah). This would follow from the
statement
(9.14) (g + ah) + J(g + ah) = (g) + J(g)
Indeed, if this were true, then (h) + J(h) ⊂ (g) + J(g) = (g + ah) + J(g + ah) and we can
redo the argument of Step 2. Now (9.14) does not hold for all a ∈ C , but we will show
that it holds for almost all a ∈ C . To prove (9.14) for almost all a ∈ C , we use the theory
of coherent sheaves. Let U ⊂ C n be an open subset such that f and g converge on U and
∂G ∂G
0 ∈ U . Define G := g + th, and J(G) = ∂x 1
, . . . , ∂xn . We have (G) + J(G) ⊂ (g) + J(g).
The ideal sheaves (G) + J(G) and (g) + J(g) in OU×C are coherent sheaves. Therefore,
the quotient M := (g) + J(g) /((G) + J(G)) is a coherent sheaf. Supports of coherent
analytic sheaves are analytic sets by 6.2.5, so that Z := supp(M ) is an analytic subset of
U × C . Therefore, the intersection Ze := Z ∩ ({0} × C ) is an analytic subset of {0} × C .
Now Ze is the set of all points (0,a) such that M(0,a) 6= (0), and therefore includes the
set of all points for which the inclusion of stalks (G) + J(G) ⊂ (g) + J(g) is strict. As
an analytic subset of {0} × C , it is either {0} × C or a discrete set (Identity Theorem
3.1.10). But obviously (0,0) 6∈ Z, e therefore, Z e is discrete. This proves (9.14) for all a
except for a discrete subset of C . This discrete subset we call D.
Step 4. Because of (f ) + J(f ) = (g) + J(g), we see that 1 ∈
/ D. Choose a path γ : [0,1] −→
C \ D from 0 to 1. For all a ∈ γ ([0,1]), we have that there exists an open neighborhood
U of a such that for all b ∈ U , the function f + bh is contact equivalent to f + ah by
Step 3. As γ([0,1]) is compact, we can cover γ([0,1]) by a finite number of U ′ s. It follows
that f and f + g − f = g are contact equivalent.
Remark 9.1.11. The Morse Lemma follows from the Mather-Yau Theorem. As for
g = x21 + · · · + x2n we have J(g) = (x1 , . . . ,xn ) we have that for all f with J(f ) = m are
right equivalent to x21 + · · · + x2n . The C {t} module structure can only be trivial, that
is, multiplication with t is the zero map. Of course, this is a very difficult proof of the
Morse Lemma.
Exercises
9.1.12. Suppose that f and g are right equivalent. Let k ∈ N. Suppose that f is k–determined.
Prove that g is k–determined.
9.1.13. Show by counterexample that the statement mk+2 ⊂ m3 J(f ) does not imply that f is
k–determined in general.
9.1.14. Let R be a commutative ring and u ∈ R[[t]]. Prove that for given a ∈ R the differential
equation
dy
=u·y
dt
P y ∈ R[[t]]
has a unique solution Pwith y(0) = a.
(Hint: Write u = uk tk , y = yk tk and compare both sides of the equation.)
k k

(Hint: As an application of Artin’s Approximation Theorem, it suffices to find a formal automor-
phism. Now use the same idea as in Theorem 9.1.5 and induction on the number of parameters.
(k) (k)
Let ϕ(0) = (x1 , . . . ,xn ,y1 , . . . ,ym ) and u(0) = 1. Assume ϕ(k) = (ϕ1 , . . . ,ϕn ,y1 , . . . ,ym ) and
(k)
u are defined such that
“ ”
∂f ∂f ∂f
• u(k) = 1 in case of ∂y i
∈ (x 1 , . . . ,xn )
c
∂x1
, . . . , ∂xn
• u(k) (x1 , . . . ,xn ,y1 , . . . ,ym−k ,0, . . . ,0) = 1

(k)
• ϕi (x1 , . . . ,xn ,y1 , . . . ,ym−k ,0, . . . ,0) = xi
(k)
• ϕi − xi ∈ (x1 , . . . ,xn )c .
` (k) (k) ´
• f ϕ1 , . . . ,ϕn ,y1 , . . . ,ym−k ,0, . . . ,0 = u(k) f (x1 , . . . ,xn ,y1 , . . . ,ym ).
Choose ξ1 , . . . ,ξn ∈ (x1 , . . . ,xn )c C {x1 , . . . ,xn ,y1 , . . . ,ym } and v ∈ C {x1 , . . . ,xn ,y1 , . . . ,ym } such
Pn
that ∂y∂f = v · f + ξj ∂x∂f
j
(the case v = 0 included). Let ϕ be the automorphism defined
m−k
i=1
by the integral curve of the corresponding vector field and define ϕi := ϕ(xi ) i = 1, . . . ,n. Note
that ϕ(yi ) = yi − δi,m−k · t. Then ∂ϕ
∂t
i
= ξi (ϕ1 , . . . ,ϕn ,y1 , . . . ,ym−k−1 ,ym−k − t,0, . . . ,0). Use this
property to prove that
∂f (ϕ1 , . . . ,ϕn ,y1 , . . . ,ym−k−1 ,ym−k − t,0, . . . ,0)
=
∂t
v(ϕ1 , . . . ,ϕn ,y1 , . . . ,ym−k−1 ,ym−k − t,0, . . . ,0) · f (ϕ1 , . . . ,ϕn ,y1 , . . . ,ym−k−1 ,ym−k − t,0, . . . ,0).
Use Exercise 9.1.14 to choose s ∈ C [[x1 , . . . ,xn ,y1 , . . . ,ym−k ,t]] such that
∂s
= v(ϕ1 , . . . ,ϕn ,y1 , . . . ,ym−k−1 ,ym−k − t,0, . . . ,0) · s,
∂t
and s(t = 0) = 1. Use Exercise 9.1.14 again to prove that
f (ϕ1 , . . . ,ϕn ,y1 , . . . ,ym−k−1 ,ym−k − t,0, . . . ,0) = s · f (x1 , . . . ,xn ,y1 , . . . ,ym−k ,0, . . . ,0).
Let ϕi := ϕi |t=−ym−k and s = s |t=−ym−k then
f (ϕ1 , . . . ,ϕn ,y1 , . . . ,ym−k−1 ,0, . . . ,0) = s · f (x1 , . . . ,xn ,y1 , . . . ,ym−k ,0, . . . ,0).
(k+1)
Use the ϕi , s to define ϕi ,u(k+1) .
9.1.16. Prove the two statements in Remark 9.1.9.
9.1.17. Prove the equivalence (1) ⇔ (2) of the Theorem 9.1.8.

9.2 The A-D-E–singularities are simple. 313
9.1.18. Let g and f in m ⊂ C {x1 , . . . ,xn } be given such that J(g) ⊂ m2 J(f ). Prove that f and
g are right equivalent.
(Hint: Follow the steps in the proof of the Finite Determinacy Theorem 9.1.4.)
9.1.19. Let f ∈ C {x,y} of type

f = x3 + xy 3 + ψ,
with ψ ∈ (y 6 ,xy 4 ,x2 y 3 ,x3 y 2 ,x4 ). Prove that f is right equivalent to x3 + xy 3 .

(Hint: Use as in the proof of 9.1.8 the reduction to the case that J(f ) = J(g), resp. mJ(f ) =
mJ(g). The C {t}–algebra isomorphism implies that h := f − g ∈ J(g), resp. h := f − g ∈ mJ(g).
Now just continue as in the proof of 9.1.8.)
9.2 The A-D-E–singularities are simple.
We consider the automorphism group G of On = C {x1 , . . . ,xn }. Let f ∈ C {x1 , . . . ,xn }.

The right equivalence class of f is exactly the orbit of f under the following group action.2
G × On −→ On ; (G,f ) 7→ G(f ).
We denote the orbit of f by G · f := {G(f ) : G ∈ G }3 . The problem with the study

of such orbits is, that On as C –vector space is infinite-dimensional! In case of isolated
singularities, this is not a severe problem, as we can use the Finite Determinacy Theorem,
which allows us to reduce the study of this problem to the finite-dimensional case.
Definition 9.2.1. Let coordinates x1 , . . . ,xn on (C n , 0) be given, and let m be the
maximal ideal of On . Let k ≥ 1 be a natural number.
k–jet space by Jk := On /mk+1 . Each element in Jk has a represen-
(1) We define the P
tative of type |ν|≤k aν xν . Note that Jk is a finite-dimensional C –vector space.
P P
(2) Let f = ν aν xν ∈ On be given. The k − jet of f is defined by j k f = |ν|≤k aν xν .
(3) We can view G as a subset of ⊕ni=1 m. Indeed if G ∈ G then put gi := G(xi ). Then
G is uniquely determined by the element (g1 , . . . ,gn ), see Exercise 3.1.22. We now
define
Gk := {(g1 mod mk+1 , . . . ,gn mod mk+1 ) : (g1 , . . . ,gn ) ∈ G }.
(4) We get group actions

Gk × Jk −→ Jk .
induced by those of G . Here G · f := j k (G(f )). One immediately checks that this
is a group action and is well defined, see Exercise 9.2.17.
2 Group action means that (G1 · G2 )(f ) = G1 (G2 (f )).
3 In case of contact equivalence, we have to use the semi-direct product G ⋊On ∗ of G with O ∗ . Here O ∗
n n
is the group of units in On = C {x1 , . . . ,xn }. As a set the semi-direct product is just the Cartesian
product, but the multiplication is given by (ϕ1 ,w1 ) · (ϕ2 ,w2 ) := (ϕ1 ϕ2 , w1 · ϕ1 (w2 )). One needs this
strange looking product so that the action of G ⋊ On ∗ on C {x , . . . ,x } sending ((ϕ,u),f ) to u · ϕ(f )
1 n
is a group action.
Example 9.2.2. Let n = 1 and k = 2. Then J2 = {a0 + a1 x + a2 x2 : ai ∈ C }, and

G2 = {αx+ βx2 : α 6= 0}. So an element G = αx+ βx2 acts on a0 + a1 x+ a2 x2 by sending
it to a0 +a1 (αx+βx2 )+a2 (αx+βx2 )2 which in Jk is equal to a0 +αa1 x+(a1 β +a2 α2 )x2 .
Lemma 9.2.3. Let coordinates x1 , . . . ,xn on (C n , 0) be given. Consider f ∈ On , and
k ∈ N such that f is k–determined. Consider the group operation of Gk on Jk , and let
g ∈ On be given. Then
j k g ∈ Gk (j k f ) ⇐⇒ g is right equivalent to f.
Proof. The implication ⇐= is obvious. Suppose on the other hand that j k g ∈ Gk (j k f ).

Then there exists a G ∈ G such that f ≡ G(g) mod mk+1 . Because f is k–determined,
and f and G(g) have the same k–jet, we have that f and G(g) are right equivalent. By
transitivity, f and g are right equivalent.
Definition 9.2.4. On all jet spaces Jk , which are as vector spaces isomorphic to some
C N , we take the usual (Euclidean) C –topology. Consider the projection maps
πk : On = C {x1 , . . . ,xn } −→ Jk = C {x1 , . . . ,xn }/mk+1 .
We define a topology on C {x1 , . . . ,xn } as the coarsest topology such that all πk are
continuous.
Thus a basis for the topology for On is given by πk−1 (V ) for V ⊂ Jk . Informally
speaking, a power series is near zero when “some of its coefficients are small”, and the
coefficients of monomials which are not small are in a high power of the maximal ideal m.
On G we similarly define a topology. It is the topology induced by the product topology
of mn . The action of G on On is continuous. We omit the boring proof, as we do not
need the result.
Definition 9.2.5. We consider the action of G on On . Let f ∈ m ⊂ On . Then f is called
simple, if there exists an open neighborhood U of f , such that the number of orbits which
intersect U is finite.
Remark 9.2.6. There is a different way to phrase the definition of simple. Namely,
consider families
F (x,t) ∈ C {x1 , . . . ,xn ,t}
such that F (x,t) ∈ (x1 , . . . ,xn ), and F (x,0) = f . Families with this property are also
called deformations. Then f is called simple, if there exist finitely many right equivalence
classes G1 , . . . ,Gs such that for all families F (x,t) as above and t small and fixed the
germ Ft ∈ C {x1 , . . . ,xn } is right equivalent to one of the germs in G1 , . . . ,Gs . In case
that f is finitely determined, the proof that these two definitions are equivalent is left as
Exercise 9.2.16.
We will see later on, that simple germs have an isolated singularity. In fact, we
have the following Classification Theorem, due to Arnold, whose proof will occupy the
remainder of this chapter.
Theorem 9.2.7 (Classification of Simple Singularities). Suppose that f ∈ On is simple.
Then f is right equivalent to one of the singularities in the following list.
• Ak : f (x1 , . . . ,xn ) = xk+1

1 + x22 + . . . + x2n , k ≥ 0,4
• Dk : f (x1 , . . . ,xn ) = x1 x22 + x1k−1 + x23 + . . . + x2n , k ≥ 4,
• E6 : f (x1 , . . . ,xn ) = x31 + x42 + x23 + . . . + x2n ,
• E7 : f (x1 , . . . ,xn ) = x31 + x1 x32 + x23 + . . . + x2n ,
• E8 : f (x1 , . . . ,xn ) = x31 + x52 + x23 + . . . + x2n .
We give real pictures in dimension two of A2 , A3 , D5 , D6 , E6 and E7 .
In this section we will show that the A-D-E–singularities of Arnold’s list are simple.
In the next section we will show that there are no other simple singularities.
We first define the corank of a function. It is the coarsest invariant of right equiva-
lence classes.
Definition 9.2.8. Let f ∈ m2 ⊂ C {x1 , . . . ,xn }. The Hesse matrix of f is defined by
2
∂ f
H(f ) = (0) .
∂xi ∂xj 1≤i,j≤n
The corank of f is defined as Corank(f ) := n − rank(H(f )).
The following easy lemma is left as Exercise 9.2.20.

Lemma 9.2.9. Let f be right equivalent to g. Then Corank(f ) = Corank(g). Moreover,
let Ft be a holomorphic family of functions with Ft ∈ m2 for all t. Then for all t small
Corank(Ft ) ≤ Corank(F0 ).
We now prove the Splitting Lemma, which is a generalization of the Morse Lemma.
4 Note that A0 is the germ of a smooth map.
Lemma 9.2.10 (Splitting Lemma).
(1) Let f ∈ C {x1 , . . . ,xn }, with f ∈ m2 and Corank(f ) equal to s. Suppose that f is
finitely determined. Then f is right equivalent to an element of type
g(x1 , . . . ,xs ) + x2s+1 + . . . + x2n .
where g ∈ C {x1 , . . . ,xs } and g ∈ m3 .

(2) Let g1 and g2 ∈ m3 ⊂ C {x1 , . . . ,xs } both have isolated singularities. Then
g1 (x1 , . . . ,xs ) + x2s+1 + . . . + x2n is right equivalent to g2 (x1 , . . . ,xs ) + x2s+1 + . . . + x2n
if and only if g1 is right equivalent to g2 .
(3) Let f ∈ C {x1 , . . . ,xs } have an isolated singularity. Then f (x1 , . . . ,xs ) + x2s+1 +
. . . + x2n is simple if and only if f is simple.
Proof.
(1) By following the proof of the Morse Lemma 3.4.30, we may (after renaming coordi-
nates) assume that f is equivalent to a germ of type
X
(9.15) x2s+1 + . . . + x2n + xi xj Hij .
1≤i,j≤s
As the Corank is s, it follows that none of the Hij for 1 ≤ i,j ≤ s is a unit, hence Hij ∈ m.
We may rewrite (9.15) as
n
X
x2s+1 + . . . + x2n + g(x1 , . . . ,xs ) + xi Gi (x1 , . . . ,xn ) =
i=s+1
n
X
g+ (xi + 12 Gi (x1 , . . . ,xn ))2 − 41 Gi (x1 , . . . ,xn )2 .
i=s+1
As Hij ∈ m, it follows that Gi ∈ m2 for all i, and g(x1 , . . . ,xs ) ∈ m3 . We can therefore
do the following coordinate change:
x′i = xi for i = 1, . . . ,s,

x′i = xi + 1
2 Gi (x1 , . . . ,xn ) for i = s + 1, . . . ,n.
After this coordinate change we have that f is right equivalent to a germ of type
n
X
(9.16) x2s+1 + . . . + x2n + g(x1 , . . . ,xs ) + xi Gi (x1 , . . . ,xn ).
i=s+1
with Gi ∈ m3 . (Both the g and the Gi may have changed in this process). So we im-
proved the situation. Now we have Gi ∈ m3 , which is a stronger statement than Gi ∈ m2 .
We can continue like this, and we obtain that f is right equivalent to a germ of type
(9.16) with Gi ∈ mk . As f is k–determined it follows from Exercise 9.1.12 that (9.16) is
also k–determined. Thus (9.16) is right equivalent to its k–jet, and the first part of the
theorem follows.
(2) If g1 and g2 are right equivalent, then obviously g1 + x2s+1 + . . . + x2n and g2 +
x2s+1 + . . . + x2n are right equivalent. The converse follows without difficulty from the
Mather-Yau Theorem for right equivalence, see 9.1.10, but let us give a direct argument.
Suppose therefore that f1 = g1 (x1 , . . . ,xs ) + x2s+1 + . . . + x2n is right equivalent to f2 =
g2 (x1 , . . . ,xs ) + x2s+1 + . . . + x2n . Hence there exist
ϕ1 , . . . ,ϕn ∈ C {x1 , . . . ,xn }, with
∂li
ϕi = li + hi , hi ∈ m2 and li linear forms with det ∂xj 6= 0 such that
n
X n
X
(9.17) g1 (ϕ1 , . . . ,ϕs ) + ϕ2i = g2 (x1 , . . . ,xs ) + x2i .
i=s+1 i=s+1
Pn 2
Pn Pn
Note that i=s+1 li + i=s+1 hi (2li + hi ) = i=s+1 ϕ2i . By assumption, g1 ,g2 ∈ m3 , so
that
n
X n
X
(9.18) li2 = x2i .
i=s+1 i=s+1
Now suppose we can find ψs+1 , . . . ,ψn ∈ C {x1 , . . . ,xs } such that
(9.19) (2li + hi )(x1 , . . . ,xs ,ψs+1 , . . . ,ψn ) = 0 for i = s + 1, . . . ,n.
Then we can consider the endomorphism of C {x1 , . . . ,xs } given by
ϕ′1 := ϕ1 (x1 , . . . ,xs ,ψs+1 , . . . ,ψn ), . . . , ϕ′s := ϕs (x1 , . . . ,xs ,ψs+1 , . . . ,ψn ).
This is in fact an automorphism. By the Inverse Function Theorem it suffices to show that
(ϕ′1 , . . . ,ϕ′s ) = (x1 , . . . ,xs ). Now (ϕ1 , . . . ,ϕs ,ls+1 + 21 hs+1 , . . . ,ln + 12 hn ) is an automor-
phism of C {x1 , . . . ,xn }, again by the Inverse Function Theorem. Hence (ϕ1 , . . . ,ϕs ,ls+1 +
1 1
2 hs+1 , . . . ,ln + 2 hn ) is the the maximal ideal of On . It follows that for all k with 1 ≤ k ≤ s
we can write
X s Xn
1
xk = ξν ϕν + ξν lν + hν , k = 1, . . . ,s.
ν=1 ν=s+1
2
By plugging in ψi instead of xi for i ≥ s + 1 we get by using (9.19)

s
X
xk = ξν (x1 , . . . ,xs ,ψs+1 , . . . ,ψn )ϕ′ν .
ν=1
Therefore (ϕ′1 , . . . ,ϕ′s ) = (x1 , . . . ,xs ). Now
g1 (ϕ′1 , . . . ,ϕ′s ) = g2 (x1 , . . . ,xs )
follows from formulae (9.17), (9.18) and (9.19) so that we have an automorphism of
C {x1 , . . . ,xs } transforming g1 into g2 .
In order to construct the ψi , we therefore have to solve the system of equations
Fi = 0 with
Fi := 2li + hi , for i = s + 1, . . . ,n.
Now
∂Fi ∂li
det (0) = 2 det .
∂xj i,j≥s+1 ∂xj i,j≥s+1
Therefore, if we show that the right hand side is nonzero, we can apply
the Implicit
∂li
Mapping Theorem to solve this system of equations. The fact that det ∂x j
is
i,j≥s+1
nonzero, is linear algebra, and treated in Exercise 9.2.21.
(3) First suppose that f is simple. We have to prove that f P + x2s+1 + . . .+ x2n is simple too.
n
Consider Ft ∈ (x1 , . . . ,xn )C {x1 , . . . ,xn ,t} with F0 = f + i=s+1 x2i . Let k ≥ µ(f ) + 1.
For t small we have µ(Ft ) ≤ µ(f ). Then for all t small, Ft is k–determined by the Finite
Determinacy Theorem 9.1.4. We redo the argument of the first step simultaneously for
the whole family Ft . For t = 0, we get a coordinate change (in fact the identity), so that
n
for all small t we have a coordinate Pnchange2of (C , 0). As Ft is k–determined for Pall small
n
t we get a family Gt (x1 , . . . ,xs )+ i=s+1 xi , with G0 = f and Gt (x1 , . . . ,xs )+ i=s+1 x2i
is right equivalent to Ft for t small. As f is simple, there exist only finitely many right
equivalence classes in which Gt (x1 , . . . ,xs ) can lie. Hence there exist only finitely many
equivalence classes in which Ft can lie.
The converse statement is left to the reader.
Corollary 9.2.11. Let f ∈ m2 with µ(f ) < ∞, and Corank(f ) = 1. Then there exists a
k ≥ 2 such that f is right equivalent to Ak , that is, xk+1
1 + x22 + . . . + x2n . Moreover, the
Ak singularities are simple.
Proof. By the Splitting Lemma, we only have to consider a function in one variable
f (x) ∈ C {x}. If f (x) = 0 then µ(f ) = ∞. As this is not the case, there exists a unique
k with f (x)√= xk+1 u, where u is a unit in C {x}. By the Implicit Function Theorem, the
function k+1 u exists, is holomorphic and a unit in C {x}. We do the coordinate change
√ k+1
x′ = k+1 ux. In this coordinate f (x′ ) = x′ . This proves that f is right equivalent
to Ak . If Ft is a deformation of f , then µ(Ft ) ≤ µ(f ) for all t small. It follows that
an Ak singularity only deforms into an Al singularity with l ≤ k. In particular, the Ak
singularities are simple.
Now we turn our attention to functions of corank two. By the Splitting Lemma, we
only have to consider f (x,y) ∈ (x,y)3 ⊂ C {x,y}.
Proposition 9.2.12. Let f ∈ C{x,y} and f ∈ m3 . After a linear coordinate change we
may assume that j 3 f is of one of the following types.
1. j 3 f = xy(x + y), the zero set of j 3 f consists of three different lines.
2. j 3 f = x2 y, two lines coincide.
3. j 3 f = x3 , the three lines coincide.
4. j 3 f = 0.
Suppose Ft is a holomorphic family with F0 = f and Ft ∈ m2 for all small t. Then for
all t 6= 0 and small we have for the types (1), (2), resp (3):
(1) Either Corank(Ft ) = 0 or 1 or j 3 Ft = 0 consists of three different lines.

(2) Either Corank(Ft ) = 0 or 1 or j 3 Ft = 0 consists of three different lines, or two
lines coincide.
(3) Either Corank(Ft ) = 0 or 1 or j 3 Ft = 0 consists of three different lines, or two
lines coincide, or the three lines coincide.
Proof. Write
j 3 f = ax3 + bx2 y + cxy 2 + dy 3 .
Case 1. a = d = 0. Then j 3 f = xy(bx + cy).
(i) If b = c = 0 we get 4.
(ii) Suppose b = 0 or c = 0, but not both. Without loss of generality, let c = 0. Then
j 3 f = xy(bx). As b 6= 0, we can do the coordinate change y ′ = by. Then j 3 f = x2 y ′
which is 2.
(iii) Both b and c are nonzero. After the√coordinate change
√ x′ = bx, y ′ = cy we may
′
assume that b = c. By putting x = b · x, y = b · y we get j 3 f = x′ y ′ (x′ + y ′ )
3 ′ 3
which is 1.
Case 2. Suppose a 6= 0 or d 6= 0. Without
√ loss of generality, we may assume that a 6= 0
and, after the coordinate change x′ = 3 a · x, even a = 1. We factorize j 3 f :
j 3 f = (x − αy)(x − βy)(x − γy).
We do the coordinate change x′ = x − αy, so that we may assume α = 0. Hence
j 3 f = x · (x − βy)(x − γy).
(i) If β = γ = 0 we get 3.
(ii) Suppose that β 6= 0 or γ 6= 0. Without loss of generality β 6= 0. Take y ′ = x − βy.
Then j 3 f = x′ y ′ (x′ − δy ′ ) for some δ, and we are back in Case 1.
This proves the first part of the proposition. The second part of the proposition follows
from the fact that the corank can only drop, see 9.2.9, and the fact that the zeros of the
polynomial j 3 (Ft (x,1)) depend continuously on t.
Proposition 9.2.13. Let f ∈ C {x,y}, µ(f ) < ∞.
(1) Suppose j 3 f = xy(x + y). Then f is right equivalent to D4 .
(2) Suppose j 3 f = x2 y. Then there exists a k ≥ 5 such that f is right equivalent to Dk .
(3) The germs Dk for k ≥ 4 are simple.
Proof. (1) Suppose that j 3 f = xy(x + y). It is an exercise to show that from the Finite
Determinacy Theorem it follows that xy(x + y) is 3–determined. It follows without dif-
ficulty that f is right equivalent to D4 .
(2) Suppose j 3 f = x2 y. As we suppose µ(f ) < ∞, the element f is k–determined for

some k. Suppose that we have an s ≥ 4, such that
j s f = x2 y + ay s + bxy s−1 + x2 ϕ,
for a suitable ϕ ∈ ms−2 , and a,b ∈ C . This we certainly have for s = 4. We do the
coordinate transformation x′ = x + 21 by s−2 , y ′ = y + ϕ. Then a simple calculation shows
2 s
that j s f = x′ y ′ + ay ′ . Renaming x′ to x and y ′ to y we may therefore suppose that
s 2 s
j f = x y + ay . There are two cases to consider.
√ 1
Case 1. We have a 6= 0. Do the coordinate change y ′ = s a · y, x′ = 2s√ a
x. Then
s 2 s 2 s
j f = x y + y . As x y + y is s–determined (exercise), it follows that f is a Ds+1 sin-
gularity.
Case 2. We have a = 0. In this case we write down j s+1 f . Iterate the above coordinate
changes until we come in Case 1, or until s ≥ k. We will show that s ≥ k leads to a
contradiction. As f is k–determined it follows that f is right equivalent to j s f = x2 y+ay s .
If a = 0, then f does not have an isolated singularity, contrary to our assumption.
(3) Consider D4 , and a holomorphic family Ft with F0 = f = xy(x + y). We have that
µ(Ft ) < ∞. If for fixed t small, the corank of Ft is zero, then Ft has an A1 –singularity by
the Morse Lemma. If the corank is one, then Ft is right equivalent to Ak for some k by
9.2.11. As µ(Ft ) ≤ µ(D4 ), it follows that k ≤ 4.5 Hence D4 can only deform in finitely
many germs. If Corank(Ft ) = 2, it follows that j 3 Ft = 0 consists of three different lines
by 9.2.12. Hence there exist coordinates such that j 3 Ft = xy(x + y) for fixed t. Hence,
by what we just proved, j 3 Ft is right equivalent to D4 . This shows that D4 can deform
only in finitely many germs.
The proof that the Dk for k ≥ 5 are simple, is similar.
So in view of Proposition 9.2.12, the case to consider now are functions f ∈ C {x,y}
whose 3–jet is x3 .
Proposition 9.2.14. Let f ∈ C {x,y} with µ(f ) < ∞. Suppose j 3 f = x3 . Then either
(1) f is right equivalent to E6 , E7 or E8 or
(2) f is right equivalent to a function which lies in the ideal (x,y 2 )3 = (x3 ,x2 y 2 ,xy 4 ,y 6 ).
Moreover, E6 , E7 and E8 are simple.
Proof. Suppose j 3 f = x3 . We write down the four jet.
j 4 f = x3 + ay 4 + bxy 3 + x2 ϕ,
with a,b ∈ C and ϕ ∈ m2 . We do the coordinate change x′ = x + 13 ϕ. Hence we may

assume
j 4 f = x3 + ay 4 + bxy 3 .
√
Case 1. Suppose a 6= 0. Then by doing the coordinate change y ′ = 4 ay we may assume
a = 1. After doing the coordinate change y ′ = y + 14 bx we get j 4 f = x3 + y 4 + x2 ψ,
with ψ ∈ m2 . By the coordinate change x′ = x + 13 ψ, we can get rid of ψ to obtain
j 4 f = x3 + y 4 . As x3 + y 4 is 4–determined it follows that f is right equivalent to E6 .
Redoing this proof for a holomorphic family Ft with F0 = x3 + y 4 shows that E6 can
only deform into singularities of type Ak , Dk or E6 . As µ(Ft ) ≤ µ(f ) it follows that E6
is simple.
Case 2. Suppose a =√ 0, and b 6= 0. We may assume then that b = 1, by the coor-
dinate change y ′ = 3 by. From the Finite Determinacy Theorem it follows that f is
5–determined, but it does not follow from this theorem that f is 4–determined. Indeed,
one calculates that (xy 4 ,x2 y 3 ,x3 y 2 ,x4 ) ⊂ m2 J(f ), but y 5 ∈
/ m2 J(f ). But in fact f is
4–determined. To prove this, we write down the 5–jet of f
5 It follows from Exercise 9.3.22 that even k ≤ 3.
j 5 f = x3 + xy 3 + cy 5 + xϕ,
with c ∈ C and ϕ ∈ m4 . We do the coordinate change x′ = x + cy 2 , and get

3 2
j 5 f = x′ + x′ y 3 − 3cx′ y 2 + x′ ϕ′ ,
with ϕ′ ∈ m4 . Now we do the coordinate change y ′ = y − cx′ . Then

3 3 3
j 5 f = x′ + x′ y ′ − 3cx′ y ′ + ψ,
√
with ψ ∈ (x′ y ′ 4 ,x′ 2 y ′ 3 ,x′ 3 y ′ 2 ,x′ 4 ) ⊂ m2 J(f ). Replacing x′ by x′ 3 1 − 3cy ′ , and renaming
the coordinates we get
j 5 f = x3 + xy 3 + ψ,
with ψ ∈ (xy 4 ,x2 y 3 ,x3 y 2 ,x4 ) ⊂ m2 J(x3 + xy 3 ). By Exercise 9.1.19 it follows that f is
right equivalent to x3 + xy 3 , that is, we have an E7 –singularity. Similarly, one shows that
E7 is simple, and this proof is left to the reader.
Case 3. The final case to consider is a = b = 0, that is, j 4 f = x3 . So we consider the
5–jet:
j 5 f = x3 + cy 5 + dxy 4 + x2 ϕ,
with c,d ∈ C and ϕ ∈ m3 . Again we can get rid of the x2 ϕ term. The two remaining
cases are:
√
Case 3a. Suppose c 6= 0. We can after the coordinate change y ′ = 5 cy get that c = 1. Now
′ 1 5 3 ′5 2 3
put y = y+ 5 dx. Then j f = x +y +x ψ, with ψ ∈ m . By doing the coordinate change
3 5
x′ = x + 31 ψ, we can attain ψ = 0. Hence j 5 f = x′ + y ′ . By the Finite Determinacy
Theorem, it follows that f is right equivalent to an E8 –singularity. Similarly, it is proved
that E8 is simple.
Case 3b. Consider the case c = 0. Then f ∈ (x3 ,xy 4 ) + (x,y)6 ⊂ (x,y 2 )3 , as was to be
proved.
We now succeeded in our goal of proving that the A-D-E-singularities are simple.
In the next section, we will show that these are the only ones. Up to now, we already
observed the following
Remark 9.2.15. Suppose f ∈ On is not simple. Then either
(1) f has a nonisolated singularity, or
(2) Corank(f ) ≥ 3, or
(3) Corank(f ) = 2, and we may suppose f ∈ C {x,y} with f ∈ m4 , or
(4) Corank(f ) = 2, and we may suppose f ∈ C {x,y} with f ∈ (x,y 2 )3 .
Exercises
9.2.16. For finitely determined germs prove that the two definitions of simple, see Remark 9.2.6,
are equivalent.
9.2.17. Show that the group action Gk on Jk is well-defined and indeed a group action.
9.2.18. Prove the Morse Lemma by using only elementary linear algebra, and applying the
Finite Determinacy Theorem.
9.2.19.
(1) Let f be a germ of corank s. Prove that µ(f ) ≥ 2s .
(2) Let f ∈ m4 ⊂ C {x,y}. Show that µ(f ) ≥ 9.
(3) Let f ∈ (y,x2 )3 ⊂ C {x,y}. Prove that µ(f ) ≥ 10.
Conclude that germs with µ(f ) < 8 are simple.
9.2.20. Prove Lemma 9.2.9.
9.2.21. In this exercise we complete proof of the second part of the Splitting Lemma 9.2.10.
Consider matrices B of size (n P− s) × s and C of size (n − s) × (n − s). Let A =

(aij )s+1≤i≤n,1≤j≤n := (B,C). Let li = nj=s+1 aij xj . Suppose that
n
X n
X
(9.20) li2 = x2i .
i=s+1 i=s+1
To complete the proof of the Splitting Lemma it suffices to show det(C) = 1. Prove this.
(Hint: Use (9.20) to show that C is orthogonal.)
9.2.22. Let f ∈ C {x,y} with j 3 f = xy(x + y). Prove, without using the Finite Determinacy
Theorem, that f is right equivalent to D4 .
9.3 Orbits
In order to show that functions with one of the properties mentioned in Remark 9.2.15
are not simple, we need to study orbits of group actions more carefully. First a definition.
Definition 9.3.1. An affine algebraic group G is an affine variety G, together with
regular maps
• µ : G × G −→ G,
• ι : G −→ G,
which satisfy the groups axioms.
Examples 9.3.2.
(1) Consider Ga (k), k a field. As an affine variety, G ∼

= k. Furthermore µ(x,y) := x + y
and ι(x) := −x. This is called the additive group .
(2) Gm (k), the multiplicative group. As an affine variety it is equal to k \ {0}. Further-
more, µ(x,y) := x · y, and ι(x) := x−1 .
(3) Consider the group of GLn (k) of invertible n×n matrices. This is an algebraic group.
2
Indeed it is an affine algebraic set, because it is equal to k n \ {det(aij ) = 0}, so as
a complement of an affine hypersurface it is itself affine.
(4) The group Gk is also an algebraic group because, as in the previous example, it is
the complement of an affine hypersurface.
9.3 Orbits 323
Definition 9.3.3. An affine algebraic group acts on an affine variety X, if there exists
a regular map
ϕ : G × X −→ X : (g,x) 7→ g · x
such that (g1 g2 ) · x = g1 · (g2 · x).
Example 9.3.4. The action of Gk on Jk .
The main result we need concerning group actions is the following theorem.
Theorem 9.3.5. Let G be an affine algebraic group, acting on X. Then
(1) Every orbit G · x is locally closed in the Zariski topology.6
(2) Every orbit is smooth.
(3) The boundary of each orbit is a union of orbits of smaller dimension.
The proof will occupy the first part of this section. In fact we need to do some more
algebraic geometry, in particular, we need to study so-called constructible sets.
Definition 9.3.6. Let X be a topological space, Z ⊂ X a subset. Z is called constructible
if it is a finite union of locally closed subsets: Z = ∪ni=1 Ui ∩ Fi , with Ui open and Fi
closed.
Example 9.3.7. Consider C 2 with the Zariski topology, and let Z = U1 ∪ F2 , where U1
is defined by U1 := {(x,y) : x 6= 0} and F2 = {(x,y) : y = 0}.
The following set-theoretic lemma is left as Exercise 9.3.20.

Lemma 9.3.8. Let Z and W be constructible subsets of X. Then Z ∪ W , Z ∩ W and
Z \ W are constructible. In particular Z c (the complement) is constructible.
Lemma 9.3.9. Let X be an affine variety, and U ⊂ X be a (Zariski) open subset. Then
there exists a finite cover U = ∪ni=1 Ui with Ui affine.
Example 9.3.10. We have seen in Exercise 6.1.41 that C 2 \ {0}, as a locally ringed
space, is not isomorphic to an affine space. Let f ∈ C [x,y]. Then C 2 \ {f = 0} is affine
with coordinate ring C[x,y,z]/(z · f − 1). More generally, the complement of f = 0 in an
irreducible variety X is affine. Note that we can write

C 2 \ {0} = C 2 \ {y = 0} ∪ C 2 \ {x = 0} .
6 Recall that this means that it is the intersection of an open and a closed subset.
Proof of Lemma 9.3.9. Consider Y = X \ U . Then Y is Zariski closed. Consider the ideal
I (Y ) ⊂ K[X]. Write I (Y ) = (f1 , . . . ,fs ) ⊂ K[X] and define Ui := X \ V (fi ). Then Ui
is open and affine. We claim that U = ∪ni=1 Ui . As Ui ⊂ U , the inclusion ∪ni=1 Ui ⊂ U is
obvious.
For the other inclusion, suppose we have an element a ∈ U \ ∪ni=1 Ui . Then for all i,
we have a ∈/ Ui . Therefore fi (a) = 0 for all i. Hence a ∈ V (f1 , . . . ,fs ) = V (I (Y )) = Y .
Hence a ∈/ U , a contradiction.
Theorem 9.3.11 (Chevalley). Let X,Y be affine, and ϕ : X −→ Y be a morphism.
Then ϕ maps constructible sets in X to constructible sets in Y .
Proof. It suffices to prove that a locally closed subset of X is mapped to a constructible
set of Y . Let U ∩ F , with U open and F closed, be a locally closed subset. By Lemma
9.3.9, we can write U as a finite union of open affines. Therefore, it suffices to prove the
theorem for the case X = U , that is we have to show that the image of a closed subset
F of X is constructible. As F now is a finite union of irreducible varieties, it suffices to
prove the theorem for the case X = F is irreducible. So we have to show that the image
of X is constructible. As ϕ(X) is closed, it suffices to show that ϕ(X) is a finite union
of locally closed subsets of ϕ(X). Thus we may suppose Y = ϕ(X), and Y is irreducible.
The claim now follows by induction on dim(Y ). The case dim(Y ) = 0 is trivial. So assume
that the theorem has been proved for all Y ′ with Y ′ $ Y . As ϕ(X) = Y , the image ϕ(X)
contains a nonempty open subset U , see Theorem 2.3.12. Consider Y \ U = Z1 ∪ · · · ∪ Zs .
The set ϕ−1 (Zi ) is a closed subset of X, and has finitely many irreducible components.
Let Wij be such a component. By restriction we get a map ϕ : Wij −→ Zi . By induction,
the image is constructible. Therefore ϕ(X) is the union of U and a constructible set, and
therefore constructible itself.
We can now prove the important Theorem 9.3.5.
Proof of Theorem 9.3.5. (1) Let x ∈ X. Then G · x is constructible. Indeed, A := G · x
is the image of the map
G −→ X; g 7→ g · x.
Thus the orbit A = G · x is constructible by the previous Theorem 9.3.11. Write A =
∪si=1 Ui ∩ Fi . We may suppose that U1 ∩ F1 cannot be removed from this
union. Then
F1 \ (F2 ∪· · · ∪Fs ) is an open subset of F1 ∪· · · ∪Fs . Hence U := U1 ∩(F1 \ F2 ∪· · · ∪Fs )
is an open subset of F1 ∪ . . . ∪ Fs . As A is a closed subset of F1 ∪ . . . ∪ Fs , it follows that
U ⊂ A ⊂ A is open in A. Furthermore, by continuity we have that for all g ∈ G the set
g −1 (U ) := {g −1 · x : x ∈ U } is open in A. Applying this to g −1 ∈ G we see that g(U )
is open in A. As A is an orbit, it follows that A = ∪g∈G g(U ). Thus A is open in A, or
what is the same, A is locally closed.
(2) Let g ∈ G be fixed. The morphism g : X −→ X which sends x to g · x is an
isomorphism. Now let x ∈ X. Then G · x contains a smooth point. But if g · x = y, it
follows that y is also a smooth point. Therefore, all points of the orbit are smooth.
(3) Let A = G·x. Consider A, the closure of A. As multiplication with g ∈ G is continuous,
it follows that g −1 (A) is closed. As g −1 (A) = A, it follows by definition of closure that
g −1 (A) ⊃ A. Similarly g(A) ⊃ A. Now
A = g ◦ g −1 (A) ⊃ g(A) ⊃ A.
9.3 Orbits 325
Hence we have equality everywhere. Therefore, A is G–stable, and ∂A = A \ A is a union

of orbits. As A is open in A all the boundary components have smaller dimension. (Use
the Krull-dimension.)
Corollary 9.3.12. All orbits of the group action of Gk on Jk are complex submanifolds
of Jk .
The next step is the computation of the (co-)dimension of the orbits. It turns out
that this is closely related to the Milnor number.
Theorem 9.3.13. Let the irreducible algebraic group G act on the vector space C N . Let
x ∈ C N . Consider the orbit map
Ox : G −→ G · x; g 7→ g · x.
The differential of the orbit map
TOx ,id : TG,id −→ TG·x,x
is surjective. Thus the tangent space to the orbit of x is equal to the image of the differ-
ential map at the identity.
Proof. We proved in 4.3.14 that for a dominant map ϕ : X −→ Y there exists a point
p ∈ X such that Tϕ,p : TX,p −→ TY,ϕ(p) is surjective. We apply this to X = G, and
Y = G · x. So there exists a g ∈ G such that the differential of the orbit map at g is
surjective. As we already proved that the orbit is smooth, we only have to show that the
rank of the differential of the orbit map is independent of the point g ∈ G. For fixed h ∈ G,
consider the automorphism mh : G −→ G which sends g to hg. Let nh : G · x −→ G · x be
the automorphism which sends y to h · y. Consider the following commutative diagram
of holomorphic maps
Ox
G −−→ G·x
↓ mh ↓ nh
O x
G −−→ G·x
If we take derivatives we get a commutative diagram of linear maps
TOx ,id
TG,id −−−−→ TG·x,x
↓ Tmh ,id ↓ Tnh ,x
TO ,h
TG,h −−−x−→ TG·x,h·x,
in which the vertical arrows are isomorphisms. It follows that the differential of the orbit
map, the horizontal arrows, have constant rank.
Theorem 9.3.14. The tangent space to the orbit of f ∈ Jk under the action of Gk is
k+1
equal to mJ(fm)+m
k+1 .
Proof. As Gk can be viewed as an open subset of ⊕ni=1 m/mk+1 , it follows that the tangent
space of Gk is equal to the tangent space of ⊕ni=1 m/mk+1 . The differential of the orbit
map
Gk −→ Jk , h 7→ f ◦ h = h(f ),
∂f ∂f
at the identity is given by (h1 , . . . ,hn ) ∈ ⊕ni=1 m/mk+1 7→ h1 ∂x + . . . + hn ∂x , because
∂f ∂f
1 n
f (x1 + εh1 , . . . ,xn εhn ) − f = ε h1 ∂x1 + . . . + hn ∂xn , as a simple calculation shows. (Use
Exercise 4.3.22.)
Remark 9.3.15. Let f ∈ m2 ⊂ On have an isolated singularity, and suppose k satisfies

mk+1 ⊂ mJ(f ). Then the codimension of the orbit of f under Gk in Jk is equal to n+µ(f ).
Proof. Indeed, applying the previous theorem shows that this codimension is equal to
the vector space dimension of On /mJ(f ). Now we have the exact sequence
(9.21) 0 −→ J(f )/mJ(f ) −→ On /mJ(f ) −→ On /J(f ) −→ 0.
The vector space dimension of J(f )/mJ(f ) is by Nakayama’s Lemma the minimal number
of generators of J(f ). As f has an isolated singularity, V (J(f )) = {0}, or better, J(f ) is
an m–primary ideal. As the (Chevalley) dimension of (C n , 0) is n, it follows that number
of generators of J(f ) is at least n. But then the number of generators of J(f ) is exactly
∂f ∂f
n, because ∂x 1
, . . . , ∂x n
are n generators of J(f ). The statement therefore follows from
the exact sequence (9.21).
Theorem 9.3.16. Let f ∈ m2 ⊂ C {x1 , . . . ,xn }, with finite Milnor number. Let g ∈ /
m · J(f ). Then there exist only finitely many t ∈ C such that f + t · g is right equivalent
to f .
Proof. Choose k such that mk+1 ⊂ mJ(f ). Hence g does not lie in the tangent space to
the orbit through f in Jk under the action of Gk . This implies that the line f +tg in the jet
space Jk does not lie entirely in the orbit of f . Therefore, the line f + tg in Jk intersects
the orbit in only finitely many points, as the intersection of the orbit with the line is a
constructible subspace of the line. This is exactly the statement of the theorem.
Proposition 9.3.17. Let f ∈ C {x1 , . . . ,xn } be simple. Then f has an isolated singular-
ity.
Proof. Suppose not. Then the Milnor number µ(f ) is not finite. As mJ(f ) ⊂ J(f ), it
follows that in particular dimC C {x}/mJ(f ) = ∞, that is, for all k > 0, there exists a
gk ∈ mk \mJ(f )+mk+1 , as otherwise mk ⊂ mJ(f ) by Nakayama. Hence for all sufficiently
small t 6= 0 the germ fk := f + tgk is not right equivalent to f . This is because otherwise,
gk had to be in the tangent space of the orbit of f under Gk , but by assumption it is not.
Thus even fk and f are not in the same orbit under Gk . Moreover for k < s also fk is not
right equivalent to fs . Indeed otherwise the classes of fs and fk in the jet space Jk would
be Gk equivalent. But they are not, as in Jk the class of fs is equal to the class of f .
Hence there are infinitely many nonequivalent function germs in any small neighborhood
of f , and therefore f is not simple.
Examples 9.3.18. (1) Let a complex number a with a2 6= 4 be given. Then
fa (x,y) = x4 + y 4 + ax2 y 2
has an isolated singularity, hence a finite Milnor number. One calculates that m5 ⊂
m2 J(fa ). Hence fa is 4–determined. Now J(fa ) has two generators, and mJ(fa ) has 4
generators. Therefore, not all of the degree four monomials in x,y can be in mJ(fa ).
Indeed, mJ(fa ) looks like,
9.3 Orbits 327
Here dots denote nonzero monomials in O/mJ(fa ), that are however equal up to a nonzero
constant modulo mJ(fa ). We see that x2 y 2 ∈ / mJ(fa ). As this holds for all a ∈ C with
a2 6= 4, it follows that all germs fa are not simple. Indeed, suppose the converse. Then
there exists a (small) t0 such that fa+t = x4 + y 4 + (a + t)x2 y 2 are mutually right equiv-
alent for infinitely many t with 0 < t < t0 . This is impossible as for t small x2 y 2 is
not in mJ(fa+t ), so there can only be finitely many t’s in the family fa+t that are right
equivalent.
(2) Consider the family fa := x3 + y 6 + axy 4 , such that a is a complex number with
4a3 + 27 6= 0. The zero set of fa consists of three (different) parabolas. One calcu-
lates that fa then has an isolated singularity. Moreover, one shows that the monomial
xy 4 ∈
/ mJ(fa ), see Exercise 9.3.21 so that all germs fa are not simple as in the first
example.
(3) Consider the family fa := x3 + y 3 + z 3 + axyz with a3 + 27 6= 0. Then fa has an

isolated singularity, and xyz ∈
/ mJ(fa ), see Exercise 9.3.21. Thus again fa is not simple.
We now can reach our goal of proving that there are no other simple singularities
except for the A-D-E–singularities.
Theorem 9.3.19. Germs f ∈ On with one of the following properties are not simple.
(1) Nonisolated singularities.
(2) n = 2 and f ∈ m4 .
(3) n = 2 and f ∈ (x,y 2 )3 .
(4) f with Corank(f ) ≥ 3.
In particular, Arnold’s Classification Theorem of Simple Singularities 9.2.7 holds.
Proof. It was shown in the previous section, see 9.2.15 that either f is right equivalent
to an A-D-E–singularity, or right equivalent to a germ with one of the above listed
properties. So if we show that singularities with one of the above properties are not
simple, Arnold’s Classification Theorem follows.
(1) The fact that nonisolated singularities are not simple was proved in 9.3.17.
(2) We calculate modulo m5 . If f is simple, then j 4 f ∈ J4 has the property that in a
neighborhood U of j 4 f there are only finitely many orbits under the action of G4 . We
look at the tangent space at the orbit. As f ∈ m4 , the tangent space mJ(j 4 f ) + m5 /m5
is a subset of m4 /m5 . This tangent space has at most dimension 4, as mJ(j 4 f ) has four
generators which all lie in m4 . It follows that the orbit of j 4 f has dimension at most 4.
This holds for all f ∈ m4 . Note that m4 /m5 is G4 –stable, and has dimension 5. As a finite
union of subvarieties of dimension four never can fill up an open subset of a vector space
of dimension 5, it follows that all neighborhoods of j 4 f intersect infinitely many orbits.
In particular f is not simple.
(4) The argument is similar, but now we calculate modulo m4 . We know that mJ(f )/m4
has dimension at most n2 , but m3 /m4 has dimension n+2
3 , which is bigger.
(3) In this case the argument is somewhat more subtle. We calculate modulo m7 , so we
consider the action of G6 on the jet space J6 . We take the weighted order by defining
the weight of x to be equal to two, and the weight of y to be equal to one. So all
terms of elements in (x,y 2 )3 have weighted order at least six. Let f ∈ (x,y 2 )3 . The
four generators x ∂f ∂f ∂f ∂f
∂x ,y ∂x ,x ∂y ,y ∂y of mJ(f ) have weighted order at least six, five, seven
and six respectively. So we have at most three linearly independent elements in mJ(f ) of
weighted order six, namely those given by x ∂f 2 ∂f ∂f
∂x , y ∂x , and y ∂y . There are however four of
them in (x,y 2 )3 . We draw those weighted order six elements for the example f = x3 + y 6 .
The box with the cross means that that monomial is not in mJ(f ).
y
A counting argument shows that (x,y 2 )3 + m7 /m7 is a vector space of dimension sixteen.
It follows that the orbit G6 · f has dimension at most sixteen, as its tangent space has at
most dimension 16. However, G6 · f does not contain the whole of (y,x2 )3 modulo m7 , as
otherwise the tangent space of G6 ·f would contain (y,x2 )3 modulo m7 . So the intersection
of G6 · f with (y,x2 )3 modulo m7 has dimension at most fifteen. It now follows as in the
other cases that f is not simple.
Exercises
9.3.20. Prove Lemma 9.3.8.
9.3.21.
(1) Consider the family fa := x3 +y 6 +axy 4 for a ∈ C . Show that fa has an isolated singularity
exactly when 4a3 + 27 6= 0. Show that xy 4 ∈ / mJ(fa ).
(2) Do the same for the family fa = x3 + y 3 + z 3 + axyz.
9.3.22. Let Ft be a holomorphic family. Suppose that for all sufficiently small nonzero t,s
the germs Ft and Fs are right equivalent. Suppose that F0 is not right equivalent to Ft for t
sufficiently small and nonzero. Prove that µ(Ft ) < µ(F0 ).
(Hint: Use the third part of Theorem 9.3.5 and Remark 9.3.15.)
329
10 Deformations of Singularities
In the final chapter of this book, we study deformations of germs of complex spaces.
The ultimate goal is to prove the existence of a semi-universal deformation of (X, 0), in
case it has an isolated singularity. As the proof of this theorem is quite involved, we will
first treat some special cases. So, in Section 10.1, we will consider isolated hypersurface
singularities (X, 0) given by f = 0. A deformation (XS , 0) −→ (S, 0) over some germ of a
complex space (S, 0) is simply given by FS = 0 for some element FS ∈ C {x}⊗O b S,0 , with
FS (x,0) = f . There is a natural notion of isomorphism of deformations. Moreover, there
ϕ
is the notion of induced family, also called pull-back. If (T, 0) −→ (S, 0) is a holomorphic
map, we get a map ϕ : C {x}⊗O b S,0 −→ C {x}⊗O b T,0 , which we denote by the same name.
Then ϕ(FS ) is an element of C {x}⊗O b T,0 , which gives a deformation (XT , 0) −→ (T, 0) of
(X, 0). A very important special case is that (T, 0) ֒→ (S, 0), in which case one just restricts
the family to the subspace (T, 0). A versal deformation (XS , 0) −→ (S, 0) is a deformation
from which every other deformation (XT , 0) −→ (T, 0) can be induced. If, moreover, the
map on tangent spaces (mT,0 /m2T,0 )∗ −→ (mS,0 /m2S,0)∗ is uniquely determined, it is called
semi-universal. We will show in Remark ?? , that it is too much to ask for uniqueness of
the map (T, 0) −→ (S, 0). The main tool in the proof of the existence of a semi-universal
deformation of an isolated hypersurface singularity is Grauert’s Approximation Theorem.
In Section 10.2 we will consider deformations of germs of complex spaces, defined by
say f1 = · · · = fk = 0. In general, defining a deformation by simply deforming f1 , . . . ,fk
to F1S , . . . ,FkS as in the hypersurface case, does not lead to a good notion. This is because
the dimension might change, or components might disappear. In case (S, 0) is smooth
and one-dimensional, so that OS,0 = C {s}, we will show in Example 10.2.1 that it is a
good idea to take the condition that s is a nonzerodivisor of C {x}⊗O b S,0 /(F1S , . . . ,FkS ).
We will show that this property is equivalent to the fact that one can lift all the relations
(r1 , . . . ,rk ), that is f1 r1 + . . . + fk rk = 0, to a relation between the F1S , . . . ,FkS . This
is called flatness, and can be defined for general OS,0 , for example, C [ε]/(ε2 ), the local
ring of the “double point” (T, 0). As ε is always a zerodivisor, the flatness definition by
the lifting property of the relations makes more sense. We then classify all deformations
of (X, 0) over the double point (T, 0). These deformations are classified by the so-called
1
TX, 0
, which is a finite-dimensional vector space in case (X, 0) has an isolated singularity.
If one has a deformation over a germ of a complex space (S, 0), one can ∗ define the
Kodaira-Spencer map. To every element of the tangent space mS,0 /m2S,0 , one assigns
1
a deformation over (T, 0), hence an element of TX, 0
. We then prove the existence of a
versal deformation, again by applying Grauert’s Approximation Theorem, in case there
exists a deformation over a smooth space (S, 0) such that the Kodaira-Spencer map is
surjective. This case applies, for example, to isolated complete intersection singularities,
see Exercise 10.2.33, and to isolated Cohen-Macaulay codimension two singularities. An
essential ingredient is the Hilbert-Burch Theorem.
In the final section, we will prove the existence of a semi-universal deformation in
1
a more general setting, namely under the assumption that TX, 0
is a finite-dimensional
vector space. The basic idea is to do this order by order. Let (X, 0) be defined by f ∈
330 10 Deformations of Singularities
C {x}k . On first order, one does the following. Take a basis g1 , . . . ,gτ of TX,
1
0
, and take
a new ring C [[s1 , . . . ,sτ ]] with maximal ideal m = (s1 , . . . ,sτ ). We then have a flat
deformation
(10.1) f + s1 g1 + . . . ,sτ gτ
over the space with local ring C [[s1 , . . . ,sτ ]]/m2 . In general, it is not possible to lift this
family to a flat deformation over C [[s1 , . . . ,sτ ]]/m3 . We will show however, that there is
a minimal I with m3 ⊂ I ⊂ m2 such that there exists a deformation over C [[s1 , . . . ,sτ ]]/I
lifting (10.1). In general, suppose we have a deformation of (X, 0) over a space with
local ring O/J. We will show that there exists a minimal I with mJ ⊂ I ⊂ J, such
that there exists an extension of the deformation over the space with local ring O/I.
The existence of this minimal I will be shown by obstruction theory. We will construct
2 2
an element in TX, 0
⊗ I for all I with mJ ⊂ I ⊂ J, TX, 0
a vector space, the so called
obstruction space. It is zero if and only if the family can be lifted over the space with
local ring O/I. Then we will successively extend the family in a maximal way order by
order. This leads to a formal deformation. The fact that it is formally semi-universal
is not very difficult. In fact, by applying the ideas of Section 10.2, one can show that
it is semi-universal, by applying Grauert’s Approximation Theorem, in case the formal
deformation is in fact convergent. The final subtle point is to prove that there exists
a convergent deformation which is formally semi-universal. Also in this proof Grauert’s
Approximation Theorem is used. We moreover prove the following criterion for (formal)
semi-universality. Suppose a formal deformation of (X, 0) over a formal algebra C [[s]]/J
2
is given. Consider the obstruction element in TX, ⊗ J/mJ, which leads to the obstruction
2
∗
0
map TX,0 −→ J/mJ. Then the formal deformation is semi-universal if and only if the
Kodaira-Spencer map is an isomorphism, and the obstruction map is surjective.
10.1 Deformations of Functions
We make precise in the following definition the notion of deformations of functions.

Definition 10.1.1. Let f ∈ C {x1 , . . . ,xn }, S a germ of a complex space. A deformation
b S such that FS |s=0 = f .1
FS of f over S is a an element FS ∈ C {x1 , . . . ,xn }⊗O
Example 10.1.2. Let f,g ∈ C {x1 , . . . ,xn }. Then FS = f + s · g ∈ C {x1 , . . . ,xn ,s} =
b {s} is a deformation of f .
C {x1 , . . . ,xn }⊗C
Definition 10.1.3. Let FS be a deformation of f over S, and T −→ S be an analytic
map between germs of complex spaces. Then we have an induced map OS −→ OT on
the ring level, and hence a map:
b S −→ C {x1 , . . . ,xn }⊗O

C {x1 , . . . ,xn }⊗O b T.
The image of FS under this map we denote by FT . Then FT is a deformation of f over

T , which we call the pullback of FS over T −→ S, or the deformation of f induced by
T −→ S.
1 Recall that FS |s=0 is the image of the natural map C {x1 , . . . ,xn }⊗O b S /mS ∼
b S −→ C {x1 , . . . ,xn }⊗O =
C {x1 , . . . ,xn }.
10.1 Deformations of Functions 331
Examples 10.1.4. Restrictions, etc...

Definition 10.1.5. Let FS and GS be deformations of f over S. A morphism between
FS and GS is given by an analytic isomorphism
b S −→ C {x1 , . . . ,xn }⊗O

Φ : C {x1 , . . . ,xn }⊗O b S
such that it transforms fibers into fibers, that is the diagram

Φ
b S
C {x1 , . . . ,xn }⊗O b S
−→ C {x1 , . . . ,xn }⊗O
↓ ↓
OS = OS
is commutative, such that
• Φ(FS ) = GS ,
• Φ|s=0 = Id.
Definition 10.1.6. A deformation FS of F over S is called versal if for any deformation
GT of f over T there exists a map T −→ S such that GT is isomorphic to FT , the
deformation induced by T −→ S.
∂f ∂f
Theorem 10.1.7. Let f ∈ C {x1 , . . . ,xn } with µ = µ(f ) = dimC C {x1 , . . . ,xn }/( ∂x 1
, . . . ∂x 1
)<
∂f ∂f
∞. Let g1 , . . . ,gµ ∈ C {x1 , . . . ,xn } represent a basis of C {x1 , . . . ,xn }/( ∂x1 , . . . ∂x1 ). Then
µ
X
FS (x1 , . . . ,xn ) := f + si · gi ∈ C {x1 , . . . ,xn ,s1 , . . . ,sµ }
i=1
is a versal deformation of f .
Before giving the proof of this Theorem, we give some examples.
Example 10.1.8. We take the function f = x3 ∈ C {x}. A basis for the Milnor algebra is
given by {1,x}. Therefore, by the Theorem, a versal deformation is given by F (x,s1 ,s2 ) =
x3 + s1 x + s2 . Look at the deformation G(x,t) = x3 + tx2 of f . Its zero set looks like
Versality says that we can find a map T −→ S, that is, dually a map of rings
C {s1 ,s2 } −→ C {t},
or to put it in another way, two power series s1 (t),s2 (t) such that
F (x,s1 (t),s2 (t)) ≃ G(x,t).
This means that we have to find an automorphism Φ : C × T −→ C × T with

• Φ = (Φ(x,t),t),
• Φ(0,t) = Id,
• F (x,s1 (t),s2 (t)) = G(Φ(x,t),t).
We rewrite:
1 1 1
x3 + tx2 = (x + t)3 − t2 x − t3 .
3 3 27
With Φ(x,t) = x − 31 t we get
1 2
G(Φ(x,t),t) = x3 − t2 x + t3 .
3 27
We can therefore take
1 2 3
s1 (t) = − t2 ; s2 (t) = t .
3 27
therefore the map T −→ S is given by “the picture”
t
s2
s1
Example 10.1.9. We take G = f + ε · h ∈ C {x1 , . . . ,xn ,ε}/(ε2 ). Hence we can take h ∈

∂f ∂f
C {x1 , . . . ,xn }. Then G is a deformation of f over the double point T. As C {x1 , . . . ,xn }/( ∂x 1
, . . . , ∂x n
)
is a finite dimensional vectorspace we can write
µ
X n
X ∂f
h= bi gi + hi
i=1 i=1
∂xi
with bi ∈ C and hi ∈ C {x1 , . . . ,xn } and gi as in the statement of the Theorem. We

define the automorphism of C n × T by
Φ(xi ,ε) = xi − hi · ε.
By using Taylor Pnexpansion we get G(φ(x,ε),ε) = P f (x1 − h1 ε, . . . ,xn − hn ε) + εh =

∂f µ
f (x1 , . . . ,xn ) − i=1 εhi ∂x i
+ εh = f (x 1 , . . . ,xn ) + i=1 bi gi .
So G(Φ(x,ε),ε) = F (x,εb1 , . . . ,εbn ), which shows that G is isomorphic to a deforma-
tion induced by F and the map:
C {s1 . . . ,sµ } −→ C {ε}/(ε2 )
si 7→ εbi .
10.2 Deformations of Singularities 333
Proof of Theorem 10.1.7 The proof uses Grauert’s approximation Theorem and
the ideas of the last example. Let GT be a deformation of f over T , where OT =
C {t1 , . . . ,tq }/J for some ideal J. We have to find
• A map S −→ T , that is µ power series s(t) = s1 (t1 , . . . ,tq ), . . . ,sµ (t1 , . . . ,tq ) with
s ≡ 0 modulo t, and
• an automorphism Φ(x,t) with Φ(x,0) = Id,
such that
(∗) G(Φ(x,t),t) = F (x,s(t)).
This we can view as a system of equations, which we have to solve for Φ and s. Here Φ
may depend on x and t, but the µ functions si are only allowed to depend on t1 , . . . ,tq .
We show that this equation has a solution by applying Grauert’s approximation Theorem
8.2.2. To apply this Theorem, we have to show that every solution (Φ,s) of (∗) of order
i in t can be extended to a solution of (∗) of order i + 1. Having this solution (Φ,s) of
order i, we look at the homogeneous part of degree i + 1 of G(Φ(x,t),t) − F (x,s(t)) and
write it as X
tν h ν
|ν|=i+1
with hν ∈ C {x1 , . . . ,xn }. Just as in the second example we can write for all ν;
µ
X n
X ∂f
hν = bjν gj + hνj
j=1 j=1
∂xj
with bjν ∈ C and hνj ∈ C {x1 , . . . ,xn }. We put:

Φ′j =P
Φj − tν hνj j = 1, . . . ,n
s′j = sj + |ν|=i+1 tν bjν j = 1, . . . ,µ.
As G ≡ f modulo t, it follows from Taylor expansion that (Φ′ ,s′ ) = (Φ′1 , . . . ,Φ′n ,s′1 , . . . ,s′µ )
is a solution of (∗) of order i + 1. This proves the Theorem.
10.2 Deformations of Singularities
We now consider deformations of germs of complex spaces. So consider a singularity

(X, 0) defined by the equations
f1 (x) = . . . = fk (x) = 0,
for fi ∈ C {x1 , . . . ,xn } = C {x}. Naively, one probably would like to say that a deforma-
tion of (X, 0) over a space (S, 0) is given by the zero set of functions,
F1 (x,s) = . . . = Fk (x,s) = 0
with Fi ∈ C {x}⊗Ob S,0 and Fi (x,0) = fi (x) for i = 1, . . . k. However, this kind of defor-
mation is too general, because a “general fiber” of (XS , 0) −→ (S, 0) does not sufficiently
reflect the properties of (X, 0). To see this, look at the following example.
Example 10.2.1. Let (XS , 0) −→ (S, 0) be an analytic map. Here (XS , 0) is defined by
F1 = xy − s = 0
F2 = xz − s = 0
F3 = yz − s = 0
Although this is a space defined in 4-space, we would like to imagine the situation by the
following picture:
s=0 s 6= 0 s
The problem here is that s is zerodivisor of C {x,y,z,s}/(F1 ,F2 ,F3 ). The total space
(XS , 0) has four components, of which three are in the fiber s = 0. In fact, a primary
decomposition of (F1 ,F2 ,F3 ) is (cf. 1.4.12):
(F1 ,F2 ,F3 ) = (x,y,s) ∩ (x,z,s) ∩ (y,z,s) ∩ (x − y,y − z,x2 − s).
We therefore have a decomposition (XS , 0) = (X0 , 0) ∪ (X1 , 0), where (X0 , 0) is the union
of three coordinate axes, and (X1 , 0) is the parabola. Take a function f , vanishing on
(X1 , 0), but not on (X0 , 0). Then obviously s · f = 0 as elements of OXS ,0 , but both s
and f are nonzero in OXS ,0 . This expresses the fact that s is a zerodivisor. In fact it can
be proved by using the Active Lemma 4.1.10, see Exercise 10.2.28, that
Lemma 10.2.2. s is a zerodivisor ⇐⇒ there exists a component of (XS , 0) in the fiber
above zero.
The above example hopefully makes clear to the reader that for a ”nice” one parame-
ter family (with parameter s), one should impose the condition that s is a nonzerodivisor.
This we call flatness:
Definition 10.2.3. A C {s}-module M is called flat if and only if s is a nonzerodivisor
of M .
For example a finitely generated C {s}-module M is flat if and only if M is free, see
1.3.9.
It turns out that it is not so easy at all to construct (nontrivial) one parameter flat
deformations of singularities. Therefore, it is a natural idea to construct deformations
by “power-series expansion”. That is, construct deformations over (T, 0), then try to lift
over the germ of the complex space with local ring C [s]/(s3 ), etc. But then one has
the problem of defining flatness of a C [s]/(s2 )-module, as s is always a zerodivisor of
C [s]/(s2 ). So we need to give a different definition of flatness of a C {s}-module, which
gives a good generalization to rings with nilpotent elements. The following lemma is of
importance for solving this problem.
Lemma 10.2.4. Let (S, 0) = (C , 0), and consider a deformation of (X, 0) as above. Then
the deformation is flat (that is, s is a nonzerodivisor) if and only if for every relation
between the fi :
f1 r1 + . . . + fk rk = 0,
we can find a lift Ri (x,s), that is, Ri (x,0) = ri (x) and
F1 R1 + . . . + Fk Rk = 0.
P
Proof. Suppose that s is a nonzerodivisor, and take a relation fi ri = 0. Take any lift
Ri′ (x,s) of the ri , and look at F1 R1′ + . . . + Fk Rk′ . This might not be zero, but we know
it is if we plug in s = 0. This shows that this expression is divisible by s:
F1 R1′ + . . . + Fk Rk′ = sΦ.
This says that sΦ = 0 as element of OXS ,0 . As Ps is a nonzerodivisor it follows that Φ = 0

as element of OXS ,0 . Hence we can write Φ = αi Fi for some αi . Now put Ri = Ri′ −sαi .
It follows that F1 R1 + . . . + Fk Rk = 0, so we found a lift of the ri .
On the other hand, suppose that we can lift every relation. We need to show that s
is a nonzerodivisor. So suppose that sΦ = 0 ∈ OXS ,0 , that is
sΦ = F1 R1 + . . . + Fk Rk .
P
Putting Ps = 0 we get a relation fi ri = 0, which by assumption can be lifted to a
relation Fi Ri′ = 0. Then X
sΦ = Fi (Ri − Ri′ ).
As both Ri and Ri′ are lifts of the ri , it follows that Ri − Ri′ is divisible by s. Hence
R −R′
the power-series i s i exists. Because in the power-series ring s is a nonzerodivisor it
follows that
X Ri − R′
i
Φ= Fi ,
s
expressing the fact that Φ = 0 as element of OXS ,0 . This is what we had to show.
This result motivates the following definition.
Definition 10.2.5. Let (X, 0) ⊂ (C n , 0) be a germ of a complex space. Let I (X, 0) =
(f1 , . . . ,fk ). Let (S, 0) be a further germ of a complex space. Let (XS , 0) ⊂ (C n × S, 0)
be defined by F1 , . . . ,Fk , Fi ∈ C {x}⊗O b S,0 . Then (XS , 0) −→ (S, 0) is called a (flat)2
deformation of (X, 0) over (S, 0) if
(1) Fi (x,0) = fi (x) for i = 1, . . . k,
(2) for every relation r = (r1 , . . . ,rk ) ∈ C {x}k between the fi , that is f · r :=
Pk
b k
i=1 fi ri = 0, there exists a lift R = (R1 , . . . ,Rk ) ∈ (C {x}⊗OS,0 ) , that is
Ri (x,0) = ri (x), which is a relation between the Fi . This means that F · R :=
Pk 3
i=1 Fi Ri = 0.
2 Later on we will often write deformation instead of flat deformation.
3 b S,0 /(F1 , . . . ,Fk ) is a flat OS,0 –module in the general
This is, in fact, equivalent to saying that C {x}⊗O
set-up of flatness, see [Matsumura 1979], Theorem 49.
A priori the definition depends on the choice of the generators of (XS , 0) (and of
(X, 0)), but it is not so difficult to show that it in fact does not. See Exercise 10.2.29.
Proposition 10.2.6. Let (X, 0) be a germ of a complex space, defined by the ideal I =
(f1 , . . . ,fk ) in C {x}. Let g1 , . . . ,gk be further elements in C {x} =: O. Consider (T, 0),
the double point. Then f1 +εg1 , . . . ,fk +εgk defines a (flat) deformation (XT , 0) −→ (T, 0)
if and only if
f1 7→ g1 , . . . ,fk 7→ gk
gives a well-defined element of HomO (I,OX,0 ) =: NX,0 .4 The module NX,0 is called the
normal module of (X, 0).
Proof. Suppose first that f + εg := (f1 + εg1 , . . . ,fk + εgk ) defines a (flat) deformation
(XT , 0) −→ (T, 0). We have to show that P the map fi −→ gP i induces a well defined map
in HomO (I,OX,0 ). Thus an element h = rP
i fi is sent to ri gi . In order
P to show that
this
P is well-defined, we have to show that if ri f i = 0, it follows that ri gi ∈ I. But
ri fi = 0 means that r = (r1 , . . . ,rk ) is a relation, which by assumption can be lifted
to a relation r + εs between f + εg, hence
0 = (f + εg) · (r + εs) = f · r + ε(f · s + r · g) = ε(f · s + r · g)
because ε2 = 0. Hence r · g ∈ (f1 , . . . ,fk ) as was to be shown. The converse is proved by

reversing the arguments.
Lemma 10.2.7. Let (X, 0) be a germ of a complex space. Then (flat) deformations
(XT , 0) −→ (T, 0) are in one-one correspondence with the elements of the normal module
NX,0 .
Proof. We apply the previous proposition.

Step 1. Suppose that g1 , . . . ,gk and g1′ , . . . ,gk′ give the same element in NX,0 . We have to
show that the ideals (f1 + εg1, . . . ,fk + εgk ) and (f1 + εg1′ , . . . ,fk + εgk′ ) are equal. For this
′
it suffices, by symmetry, to show one inclusion.
′
P Now, by assumption, gi − gi ∈ 2I, that is,
there exist aij ∈ C {x} such that gi − gi = j aij fj . It then follows that, as ε = 0
X X
fi + εgi = fi + εgi′ + ε aij fj = (fi + εgi′ ) + ε aij (fj + εgj′ ),
j j
This gives the inclusion
(f1 + εg1 , . . . ,fk + εgk ) ⊂ (f1 + εg1′ , . . . ,fk + εgk′ ).
Step 2. The fact that equality of the ideals (f1 +εg1 , . . . ,fk +εgk ) and (f1 +εg1′ , . . . ,fk +εgk′ )
implies that g1 , . . . ,gk and g1′ , . . . ,gk′ give the same element in NX,0 = HomO (I,OX ) is
proved similarly.
Example 10.2.8. We consider the germ of the complex space (X, 0) defined by the
2–minors of the matrix5
4 Note that NX, 0 is not an invariant of the isomorphism class of (X, 0), as it might depend on the
embedding.
5 Readers knowing some algebraic geometry will recognize that this singularity is the cone over the
rational normal curve of degree four.

x0 x1 x2 x3
.
x1 x2 x3 x4
We will denote by fij the minor obtained by taking the i–th and j–th column. So for
example f13 = x0 x3 −x1 x2 . It has been shown in Exercise 7.3.8 that the ideal I defined by
the minors is a prime ideal. Thus (X, 0) is irreducible. We determine the normal module
NX,0 .
Step 1. We have to determine the relations between the fij . We get eight relations in the
following way: take three columns, double one of the rows, and compute the determinant.
As two rows are the same, the determinant is zero, so that we indeed get a relation. We
get relations rijk if we double the first row, and sijk if we double the second row. As an
example, we compute r123 . So we take the first three columns, and put the first row on
top. We get the following matrix
 
x0 x1 x2
 x0 x1 x2  ,
x1 x2 x3
and hence the relation x0 f23 −x1 f13 +x2 f12 = 0. We claim that all relations are generated
by the relations constructed in this way. We will only indicate how to prove this, leaving
the reader to fill in the details. First of all, we have that C {x0 ,x4 } ⊂ OX,0 is a Noether
normalization. Because OX,0 is a free C {x0 ,x4 }–module, see Exercise 7.3.8, it follows
that x0 ,x4 is a regular sequence in OX,0 . Thus all relations between the elements
−x21 =f12 (0,x1 ,x2 ,x3 ,0),

−x1 x2 =f13 (0,x1 ,x2 ,x3 ,0),
−x1 x3 =f14 (0,x1 ,x2 ,x3 ,0),
−x22 =f23 (0,x1 ,x2 ,x3 ,0)
−x2 x3 =f24 (0,x1 ,x2 ,x3 ,0),
−x23 =f34 (0,x1 ,x2 ,x3 ,0)
can be lifted to relations between the fij . The relations between these monomials are
easily determined, and there are indeed eight of them.
Step 2. We can easily generate 8 elements n1 , . . . ,n8 of the normal module NX,0 . These
we get by putting an extra ε at one of the eight spots of the matrix. The fact that the
relations can be lifted, follows because we can lift the relations rijk and sijk by again
taking three columns and double a row! The n1 , . . . ,n4 correspond to the four entries in
the top row, and the n5 , . . . ,n8 correspond to the four entries in the bottom row. To give
an example we determine n1 . We get the following matrix

x0 + ε x1 x2 x3
.
x1 x2 x3 x4
We take the first minor, and get x0 x2 + εx2 − x21 . Thus f12 → x2 etc. It turns out that
the normal module is generated by nine elements, namely by n1 , . . . ,n8 and one further
element n9 . We put these in the following table. The rows give the elements of NX,0 by
giving the value on fij in the columns.
f12 f23 f34 f13 f24 f14

n1 x2 0 0 x3 0 x4
n2 −x1 x3 0 0 x4 0
n3 0 −x2 x4 −x1 0 0
n4 0 0 −x3 0 −x2 −x1
n5 −x1 0 0 −x2 0 −x3
n6 x0 −x2 0 0 −x3 0
n7 0 x1 −x3 x0 0 0
n8 0 0 x2 0 x1 x0
n9 x0 −x2 x4 0 0 x2
The fact that n9 is also an element of the normal module is a direct check. We want to
show that n1 , . . . ,n9 generate the normal module. To show this, we first note that an
element in NX,0 is determined by the values on f12 ,f23 and f34 , that is, by the first three
columns. To see this, look for example at the relation r123 . It says that x0 f23 − x1 f13 +
x2 f12 = 0. So if ϕ ∈ NX,0 , it follows that
x1 ϕ(f13 ) = x0 ϕ(f23 ) + x2 ϕ(f12 ),
so we can compute x1 ϕ(f13 ). As x1 is a nonzerodivisor of OX,0 we can compute ϕ(f13 ).
Similarly one sees that one can compute ϕ(f24 ) and ϕ(f14 ).
By doing column operations, we see that we have normal module elements ϕ of the
following type:
(1) ϕ(f12 ) = xi , ϕ(f23 ) = 0, ϕ(f34 ) = 0, for i = 0,1,2;
(2) ϕ(f12 ) = 0, ϕ(f23 ) = xi , ϕ(f34 ) = 0, for i = 1,2,3;
(3) ϕ(f12 ) = 0, ϕ(f23 ) = 0, ϕ(f34 ) = xi , for i = 2,3,4.
Take an element ϕ of NX,0 . We claim that ϕ(f23 ) ∈ (x1 ,x2 ,x3 ). So calculate modulo
(x1 ,x2 ,x3 ). The ideal I modulo (x1 ,x2 ,x3 ) is generated by one element, namely x0 x4 . We
look at
0 = ϕ(r123 ) = x0 ϕ(f23 ) − x1 ϕ(f13 ) + x2 ϕ(f12 ).
Modulo (x1 ,x2 ,x3 ) we get x0 ϕ(f23 ) = 0. Hence a representative (also called ϕ(f23 )) of
ϕ(f23 ) in C {x} modulo (x1 ,x2 ,x3 ) is divisible by x4 . Similarly, using the relation s234 ,
we get that ϕ(f23 ) is divisible by x0 . Thus ϕ(f23 ) is divisible by x0 x4 , which is modulo
I equal to x1 x3 .
It is somewhat more difficult, but similar, to show that ϕ(f12 ) = 0 mod (x0 ,x1 ,x2 ),
and ϕ(f34 ) = 0 mod (x2 ,x3 ,x4 ). Altogether this shows that NX, 0 has the nine generators
written down above.
Definition 10.2.9. Let two deformations (XS , 0) −→ (S, 0) and (XS′ , 0) −→ (S, 0) of
(X, 0) over (S, 0) be given. A morphism of (XS , 0) −→ (S, 0) to (XS′ , 0) −→ (S, 0) is given
by a map Φ : (XS , 0) −→ (XS′ , 0) such that:
(1) Φ restricts to the identity map on (X, 0).
(2) Φ preserves fibers, that is, the diagram
Φ
(XS , 0) −→ (XS′ , 0)
↓ ↓
Id
(S, 0) −→ (S, 0)
is commutative.
Theorem 10.2.10. Let Φ : (XS , 0) −→ (XS′ , 0) be a morphism of deformations of (X, 0)
over (S, 0). Then Φ is an isomorphism.
Proof. This is a consequence of the Inverse Function Theorem. A morphism of deforma-
tions of (X, 0) induces the identity on (X, 0), and preserve fibers. Therefore, Φ∗ is induced
by a morphism which, by abuse of notation, we also call Φ∗
b S,0
C {x1 , . . . ,xn }⊗O b S,0 ,
−→ C {x1 , . . . ,xn }⊗O
xi 7→ xi + hi , i = 1, . . . n,
s ∈ OS,0 7→ s.
b S,0 , mS,0 the maximal ideal of OS,0 . To prove that Φ∗ is an

where hi ∈ mS,0 · C {x}⊗O
isomorphism, we may suppose that OS,0 = C {s}. Therefore, the determinant of the
Jacobian matrix of Φ∗ is equal to one. Hence the inverse of Φ∗ , and hence of Φ exists,
by the Inverse Function Theorem.
Remark 10.2.11. Let (X, 0) ⊂ (C n , 0) be a singularity, given by f = (f1 , . . . ,fk ).
Consider deformations (XS , 0) −→ (S, 0) and (XS′ , 0) −→ (S, 0) over (S, 0). Suppose
(XS , 0) is defined by FS = (F1 , . . . ,Fk ), and (XS′ , 0) is defined by FS′ = (F1′ , . . . ,Fk′ ), both
lifting (f1 , . . . ,fk ). To give an isomorphism Φ : (XS , 0) −→ (XS′ , 0) means that one has
to find:
b S,0 with Φi (x,0) = xi .
• Elements Φ1 , . . . ,Φn ∈ C {x1 , . . . ,xn }⊗O
b S,0 such that Λ|s=0 = Id.
• A k × k–matrix Λ with entries in C {x1 , . . . ,xn }⊗O
satisfying
FS′ = FS (Φ1 , . . . ,Φn ) ◦ Λ ∈ (C {x}⊗O
b S,0 )k .
(Here ◦ denotes matrix multiplication.) The elements Φi give the relative isomorphism,
inducing the identity on (X, 0), whereas the matrix Λ changes generators for the ideal of
(XS , 0).
Now we want to classify deformations of (X, 0) over the double point (T, 0). Suppose
that the ideal of (X, 0) is equal to I ⊂ On . Consider Θn , the free On –module of all
∂
derivations of On . Elements θ of Θn look like θ = h1 ∂x 1
+ . . . + hn ∂x∂ n for hi ∈ On . Note
that we have a map
α : Θn −→NX,0 = HomOn (I,OX,0 )

θ 7→ f 7→ θ(f ) .
So we can define:
1
Definition 10.2.12. TX, 0
:= Coker(α).6
Theorem 10.2.13. Let (X, 0) ⊂ (C n , 0) be a germ of a complex space. Isomorphism
classes of deformations of (X, 0) over (T, 0) are in one-one correspondence with elements
1
of TX, 0
.
6 1
It is, in fact true that TX, is an invariant of the isomorphism class of (X, 0), see Exercise 10.2.26.
0
Proof. Suppose (X, 0) is defined by f1 , . . . ,fk . Consider two deformations (XT , 0) −→

(T, 0) and (XT′ , 0) −→ (T, 0) given by f1 + εg1 , . . . ,fk + εgk and f1 + εg1′ , . . . ,fk + εgk′
respectively. A morphism of deformations between (XT , 0) and (XT′ , 0) is induced by a map
C {x} ⊗ C [ε]/(ε2 ) −→ C {x} ⊗ C [ε]/(ε2 ) with (x1 , . . . ,xn ,ε) 7→ (x1 + εh1 , . . . ,xn + εhn ,ε)
for some h1 , . . . hn ∈ C {x}. Then (XT′ , 0) is given by the zero set of
 
Xn
∂f i
fi (x1 + εh1 , . . . ,xn + εhn ) + εgi = fi + ε  hj + gi  for i = 1, . . . k.
j=1
∂xj
This holds because ε2 = 0. Hence we see that the map I −→ OX,0 given by fi 7→ gi − gi′
is in the image of Θn . Similarly, one shows that elements of Θn induce isomorphisms.
This is what we needed to show.
Examples 10.2.14.
1
(1) Let (X, 0) be a germ of an isolated hypersurface, defined by f ∈ C {x}. Then TX, 0
=
C {x}/(f,J(f )). This follows from Example ??.
(2) We take the singularity of Example 10.2.8. Recall that the normal module elements
are determined by their values on f12 , f23 and f34 . A direct calculation shows that
∂ ∂ ∂ ∂ ∂ 1
∂x0 = n1 , ∂x1 = n2 + n5 , ∂x2 = n3 + n6 , ∂x3 = n7 + n4 , ∂x4 = n8 . Thus TX,0 is generated
1
by four element n2 , n3 , n4 , and n9 . We claim that (x0 ,x1 ,x2 ,x3 ,x4 )TX, 0
= 0. Thus we
have to do twenty checks!! The reader hopefully forgives us for just doing one of those.
1
We will show that the class of x0 n4 is the zero element in TX, 0
.We look at the values on
f12 , f13 and f34 . They are 0, 0 and −x0 x3 which modulo I is equal to 0, 0 and −x2 x1 .
∂ 1
Thus it is equal to −x1 n8 = −x1 ∂x 4
, hence the trivial element in TX, 0
. Altogether we see
1
that dimC (TX,0 ) = 4, with a basis given by the classes of the normal module elements
n2 ,n3 ,n4 and n9 .
Theorem 10.2.15. Let (X, 0) be a germ of a complex space with an isolated singularity.
1
Then TX, 0
is a finite-dimensional C –vector space.
Proof. Let X ⊂ U be a small representative of (X, 0) such that 0 is the only singular
point of X. Consider the sheaf TX1 := HomOU (I ,OX )/ΘU . Here ΘU is the sheaf of
derivations on U , I is the sheaf of ideals defining X. The map ΘU −→ HomOU (I ,OX )
1
is defined as in the definition of TX, 0
. By the basic coherence theorems TX1 is a coherent
1
sheaf on X, whose stalks in p ∈ X is equal to TX,p . As X has only one singularity, these
are zero for p 6= 0, by Exercise 10.2.27. Hence TX1 is a coherent sheaf concentrated in the
point 0, and therefore the set of global sections Γ(TX1 ,U ) = TX, 1
0
is a finite-dimensional
C –vector space by Theorem 6.2.6.
The converse of this theorem does not hold, see Exercise 10.2.30. We now come to
the definition of induced family and of versal deformations.
Definition 10.2.16. Let (X, 0) be a germ of a complex space.
(1) Suppose (XS , 0) −→ (S, 0) is a deformation of (X, 0), defined by F1 , . . . ,Fk ∈
b S,0 . Let p : (T, 0) −→ (S, 0) be an analytic map. We define the pull-back,
C {x}⊗O
or the induced deformation
(X ×S T, 0) −→ (T, 0)
as follows. The map p induces a map p∗ : OS,0 −→ OT,0 , which in turns induces a
map p∗ : C {x}⊗Ob S,0 −→ C {x}⊗O b T,0 . Then (XT , 0) is given by p∗ (F1 ), . . . ,p∗ (Fk ).
It is proved in Exercise 10.2.32 that the map (X ×S T, 0) −→ (T, 0) is indeed flat.
(2) A deformation (XS , 0) −→ (S, 0) is called versal if for every deformation (XT , 0) −→
(T, 0) there exists a map p : (T, 0) −→ (S, 0) such that the deformations (XT , 0) −→
(T, 0) and (X ×S T, 0) −→ (T, 0) are isomorphic. The versal deformation is called
semi-universal if moreover the map on tangent spaces is uniquely determined.
Definition 10.2.17. Let (XS , 0) −→ (S, 0) be a deformation of (X, 0). Suppose (X, 0)
k
is defined by f ∈ C {x}k , and that the deformation is given by F ∈ C {x} b
P⊗OS,0 .
Suppose mS,0 has a minimal system of generators s1 , . . . ,sp . Write F = f + si gi . We
1
denote by [gi ] the class of gi mod mS,0 in TX, 0
. Then the map
∗
TS,0 = mS,0 /m2S,0 1
−→ TX, 0
X X
ai s∗i 7→ ai [gi ]
i i
is called the Kodaira-Spencer map. It is easily shown that the definition of the Kodaira-
Spencer map is independent of the choices made.
The main goal in this chapter is to prove that for (X, 0) with an isolated singularity,
a semi-universal deformation exists. In general, the proof of this theorem needs many
additional preparations, but we proved it already in the first section for hypersurface sin-
gularities. In the remainder of this chapter, we will prove a special case of this theorem,
which includes the case of hypersurface singularities and complete intersection singular-
ities (see Exercise 10.2.33). Moreover we treat the case of Cohen-Macaulay singularities
of codimension two by applying the Hilbert-Burch Theorem 6.5.26.
Theorem 10.2.18. Let (X, 0) be a germ of a complex space, defined by f ∈ C {x}k .
1
Suppose dimC (TX, 0
) < ∞. Let (S, 0) = (C τ , 0). Suppose that
(1) there exists a (flat) deformation (XS , 0) −→ (S, 0) defined by7
FS (x,s) := f + s1 g1 + . . . + sτ gτ ∈ (C {x} ⊗ C {s1 , . . . ,sτ })k ,
(2) the Kodaira-Spencer map of (XS , 0) −→ (S, 0) is surjective.

Then (XS , 0) −→ (S, 0) is a versal deformation of (X, 0). If moreover the Kodaira-Spencer
map is an isomorphism, then (XS , 0) −→ (S, 0) is a semi-universal deformation.
For the proof of this theorem, we need the following preparations, which we will also
need in the next section.
Definition 10.2.19. For a local Artinian C –algebra A we denote by (A, 0) the germ of
the complex space consisting of one point 0, but with local ring A.
Definition 10.2.20. Let A and B be local C –algebras, m the maximal ideal of B.
Suppose ϕ : B −→ A is a surjective morphism of C –algebras. Then
(1) ϕ is called a small surjection if m · Ker(ϕ) = 0,
7 Note that not necessarily gi ∈ C {x}.
(2) ϕ is called a simple surjection if Ker(ϕ) is a one-dimensional C –vector space.

Example 10.2.21. The surjection C [ε]/(ε3 ) −→ C [ε]/(ε2 ) is a simple surjection. The
surjection C [ε,η]/(ε,η)2 −→ C [ε,η]/(ε,η) is small, but not simple. The surjection ϕ :
C [ε,η]/(ε2 ,η 2 ) −→ C [ε,η]/(ε,η) is not small, as εη is not zero in C [ε,η]/(ε2 ,η 2 ).
Lemma 10.2.22. Let (X, 0) be a germ of a complex space, with f ∈ C {x}k generators
of the ideal of (X, 0). Let ϕ : B −→ A be a small surjection of local Artinian C –algebras.

Consider a deformation (XB , 0) −→ (B, 0), defined by FB ∈ C {x}⊗B b k . Let (XA , 0) −→
(A, 0) be the restriction of (XB , 0) −→ (B, 0) to (A, 0), defined by the image FA of FB in

C {x}⊗Ab k . Let rA be a relation between the FA , that is, FA · rA = 0. Then there exists
a lift rB of rA with FB · rB = 0.
Proof. The proof is by induction on the vector space dimension dimC (A). The case
dimC (A) = 1 is easy. In this case A = C , and the deformation (XA , 0) is equal to (X, 0).
By assumption on flatness, every relation r = rA can be lifted to a relation rB between
the FB .
Now assume dimC (A) > 1. We can always find 0 6= α ∈ A with mA · α = 0. Consider
A′ := A/(α). By induction we can find a lift rB ′ ′
of rA with rB ′
·FB = 0. Moreover, take any
′ ′ ′
lift reB of rA . As both ϕ(e
rB ) and ϕ(rB ) project to rA , it follows that ϕ(e
rB )−ϕ(rB ) = h·α
k ′
for some h ∈ C {x} . Since both ϕ(e rB ) and ϕ(rB ) are relations between the FA we get
h · FA · α = 0. Moreover FA · α = f · α, because mA · α = 0. Thus h is a relation
between the f , and can by flatness be lifted to a relation hB between the FB . Now define
′
rB := rB + α′ hB for any α′ with ϕ(α′ ) = α. Then rB is a relation, and lifts rA . This
proves the lemma.
Lemma 10.2.23. Consider:
(1) A singularity (X, 0) defined by f ∈ C {x}k .
(2) ϕ : B −→ A a small surjection of local Artinian C –algebras. Let α1 , . . . ,αp be a
C –basis of Ker(ϕ),
(3) A deformation (XA , 0) −→ (A, 0), defined by FA and two lifts (XB , 0) −→ (B, 0),
and (XB′ , 0) −→ (B, 0) of (XA , 0) −→ (A, 0), defined by FB and FB′ respectively, both
being lifts of FA .
Then there exist elements n1 , . . . ,np ∈ NX,0 with FB − FB′ = α1 n1 + . . . + αp np .
Conversely, let FB be given. Then FB + α1 n1 + . . . + αp np also defines a deformation

of (X, 0) over (B, 0) for all n1 , . . . ,np ∈ NX,0 .
Proof. We can write:
X X
FB = FA + αi gi ; FB′ = FA + αi gi′ .
Let r be a relation, that is, r · f = 0. We will see that r · (gi − gi′ ) ∈ (f1 , . . . ,fk ), that is,
gi − gi′ defines an element of NX,0 . Consider lifts RB and RB ′
of r, which are relations
′
between the FB and the FB respectively. By Lemma 10.2.22, we may assume that the
′
restrictions of RB and RB to C {x}k ⊗ A are equal. Therefore we can write:
X X
′
RB = RA + ci · αi ; RB = RA + c′i · αi .
for complex numbers ci and c′i . As B −→ A is a small extension, it follows that mαi = 0
for all i. Hence
X X
0 = RB · FB = RA · FA + (r · gi )αi + (ci · f )αi .
Similarly X X
′
0 = RB · FB′ = RA · FA + (r · gi′ )αi + (c′i · f )αi .
By subtracting and looking at the coefficients of αi we see that for all i = 1, . . . ,p:
r · (gi − gi′ ) ∈ (f1 , . . . ,fk ).
Therefore, ni := gi − gi′ is an element of NX,0 . The converse statement is much simpler,

and left as Exercise 10.2.34.
Proof of Theorem 10.2.18. Let OT,0 = C {t}/J, and (XT , 0) −→ (T, 0) be a deforma-
tion of (X, 0), given by equations FT (x,t) = 0. Versality means that we have to find
a map (T, 0) −→ (S, 0) such that the deformation (X ×S T, 0) −→ (T, 0) is isomorphic
to (XT , 0) −→ (T, 0). Dually the map (T, 0) −→ (S, 0) is given by an analytic map
OS,0 −→ OT,0 . Thus we have to find τ power series s1 (t), . . . ,sτ (t) ∈ C {t} to define this
map. The space (X ×S T, 0) is given by the equations FS (x,s(t)) = 0. To say that this
deformation is isomorphic to FT (x,t) = 0 means that we can find an automorphism Φ of
C {x,t} with Φ(t) = t, Φ(x)|t=0 = x, and a matrix Λ ∈ C {x,t} with Λ|t=0 the identity
matrix such that
(10.2) FS (x,s(t)) ≡ (FT ◦ Φ)(x,t) ◦ Λ mod J · C {x,t}k ,
with OT, 0 = C {t}/J. Let Φ(xi ) = φi and φ = (φ1 , . . . ,φn ).

We want to apply Grauert’s Approximation Theorem 8.2.2 in order to find a solution
to the equation (10.2). We have to solve for φ, Λ and s, but the s is only allowed to depend
on t, and not on x. The equation (10.2) has a solution of order zero, as it then reads
f = f . Thus we have to show that every solution of (10.2) of order e with respect to t
extends to a solution of (10.2) of order e + 1. Therefore, suppose (φ,Λ,s) is a solution of
(10.2) of order e. Thus
(10.3) FS (x,s(t)) ≡ FT (φ,t) ◦ Λ mod (J + (t)e+1 ) · C {x,t}k .
Here (t) is the maximal ideal of C {t}. We will change the order e + 1 terms in t of s,φ
and Λ in such a way that (10.3) holds modulo (J + (t)e+2 ).
Consider the small surjection
σ : B := C {t}/(J + (t)e+2 ) −→ C {t}/(J + (t)e+1 ) =: A.
Take a basis α1 , . . . ,αp of Ker(σ). We can factorize σ into p simple surjections σi , for
i = 1, . . . ,p. Thus, using induction, we may suppose that σ : B −→ A is a small surjection,
the kernel of σ is of dimension one, and Ker(σ) is generated by a monomial α of degree
e + 1.
Step 1. Change of s(t): We apply 10.2.23 to σ : B −→ A, FB = FS (x,s(t)) and FB′ =
FT (φ,t) ◦ Λ. The s(t), φ(x,t) and Λ(x,t) are given by induction. It follows that
(10.4) FS (x,s(t)) − FT (φ,t) ◦ Λ = n · α

for an element n ∈ C {x}k such that f 7→ n is an element of the normal module NX,0 .
By assumption the Kodaira-Spencer map is surjective, that is, the elements ai : f −→ gi
1 1
mod (s) for i = 1, . . . ,τ project to generators of TX, 0
. Thus as elements of TX, 0
, we can
write
n = b 1 a1 + . . . + b τ aτ
for suitable bi ∈ C . Now we can redefine the s(t):
s1 (t) := s1 (t) − b1 α, . . . , sτ (t) := sτ (t) − bτ α.
Step 2. Change of φ(x,t). After the change of s(t) we have just done, we have reached
1
the stage that the n = (n1 , . . . ,nk ) ∈ NX,0 in (10.4) defines a trivial element in TX, 0
.
−1
Then, obviously, nΛ|t=0 is a trivial element too. Therefore, there exist c1 , . . . ,cn ∈ C {x}
such that θ(f ) = nΛ−1 ∂ ∂
|t=0 , with θ := c1 ∂x1 + . . . + cn ∂xn . We redefine the automorphism
(φ1 , . . . ,φn ,t) in the following way.
φ1 := φ1 − c1 α, . . . , φn := φn − cn α.
It follows from Taylor’s expansion that after this redefinition, the n defines the trivial
element of NX,0 , that is, ni is the zero element in OX,0 for i = 1, . . . ,k.
Step 3. Change of Λ. After these redefinitions the elements n in (10.4) have entries ni
P in OX,0 = C {x}/(f1 , . . . ,fk ). Thus we can find elements
which all give the zero element
cji ∈ C {x} such that ni = j cji fj . Redefine the matrix Λ by putting Λ := Λ − (cji )α.
After this redefinition, we reached the stage that n = 0 ∈ C {x}k . This is what we had
to show.
To finish this section, we will prove the existence of a versal deformation for (X, 0) ⊂
(C n , 0) that have an isolated singularity, are Cohen-Macaulay, and have dimension n −
2. The Auslander-Buchsbaum formula 6.5.20 implies that the projective dimension of
OX,0 is equal to two. The main ingredient is now the Hilbert-Burch Theorem 6.5.26
and Theorem 10.2.18. Indeed, by the Hilbert-Burch Theorem, one has the following free
R := On – resolution of OX,0 = R/I, where I = I (X, 0):
2 α 1 α
0 −→ F2 −→ F1 −→ R −→ R/I −→ 0.
Here α2 is an (t − 1) × t matrix for some t. Generators f1 , . . . ,ft for the ideal I are
obtained by taking the (t−1)–minors of the matrix α2 . Equivalently, one has the following
resolution of the ideal I:
α2 α1
0 −→ F2 −→ F1 −→ I −→ 0.
Proposition 10.2.24. With the above notation, we get deformations of (X, 0) over any
germ of a complex space (S, 0) in the following way. Let E be a matrix with entries in
b S,0 . Consider the matrix α2 + E, and consider It−1 (α2 + E). This is an ideal in
On ⊗m
b S,0 , and defines a space (XS , 0) such that (XS , 0) −→ (S, 0) is a flat deformation
On ⊗O
of (X, 0). Moreover, all deformations of (X, 0) are obtained this way.
Proof. Step 1. Consider α1 + E ′ , the 1 × t matrix where the i–th entry equals (−1)i times
the minor obtained from α2 + E by leaving out the i–th row. Then the sequence
α +E α +E ′
b S,0 )t−1 −−2−−→ (On ⊗O
(On ⊗O b S,0 )t −−1−−→ It−1 (α2 + E) −→ 0
is certainly a complex, that is, (α1 + E ′ ) ◦ (α2 + E) = 0. This implies that we can lift
all the relations between the minors of α2 , which are generators of the ideal I. So we get
indeed a flat deformation.
Step 2. Conversely, suppose we have a deformation (XS , 0) of (X, 0) over (S, 0). The
flatness says that we can lift the relations. Hence we get a complex
α′ α′
(10.5) b S,0 )t−1 −→
(On ⊗O 2
b S,0 )t −→
(On ⊗O 1
b S, 0 −→ OXS ,0 −→ 0.
On ⊗O
where α′1 = α1 modulo mS,0 and α′2 = α2 modulo mS,0 . We leave it as Exercise 10.2.37
to prove that the sequence (10.5) is exact, that is, the kernel of α′1 is generated by the
lifts of the set of t − 1 generators of the kernel of α. We claim that the map α′2 is in
fact injective. To show this suppose α′2 (a1 , . . . ,at−1 ) = 0. Calculating modulo mS,0 , we
see that (a1 , . . . at−1 ) is zero modulo mS,0 . This is because
P α2 is injective. Take a basis
b1 , . . . ,bp of mS,0 /m2S,0 , and write (a1 , . . . ,at−1 ) = (ai1 , . . . ,ait−1 )bi modulo m2S,0 for
aij ∈ On . From α′2 (a1 , . . . ,at−1 ) = 0 it follows by calculating modulo m2S,0 that aij = 0.
Thus (a1 , . . . at−1 ) is zero modulo m2S,0 . Going on like this, we see that (a1 , . . . at−1 ) is
zero modulo mkS,0 for all k, hence 0 by Krull’s Intersection Theorem. So we get an exact
sequence
α′ α′
b S,0 )t−1 −→
0 −→ (On ⊗O 2
b S,0 )t −→
(On ⊗O 1
b S, 0 −→ OXS ,0 −→ 0.
On ⊗O
Let I = I (XS , 0) ⊂ On ⊗O b S,0 . It follows from the Hilbert-Burch Theorem that I =

aIt−1 (α′2 ) for some nonzerodivisor a. Calculating modulo mS,0 we see that a ≡ 1 mod mS,0 .
Thus a is a unit in On ⊗Ob S,0 . Then we change the minors of α′2 to a−1 times the minors,
to get the desired generators of the ideal I.
Theorem 10.2.25. Let (X, 0) be a Cohen-Macaulay codimension two isolated singular-
ity, whose ideal is given by the (t − 1)–minors of a matrix A with entries in On . Let
1
τ = dimC (TX, 0
). Consider matrices αi such that the deformations given by the (t − 1)–
minors of a matrix A + εαi form a basis of the vector space of isomorphism classes
of deformations
Pτ over (T, 0). Then the ideal given by the (t − 1)–minors of the matrix
A + i=1 si αi , which has entries in On ⊗Cb {s1 , . . . ,sτ } is the ideal of a space (XS , 0)
which gives a semi-universal deformation of (X, 0) over (S, 0) ∼ = (C τ , 0).
Proof. The theorem follows from Proposition 10.2.24 and Theorem 10.2.18.
Exercises
10.2.26.
(1) Let φ : C {x} −→ C {x} =: O be an automorphism, and I ⊂ O be an ideal. Prove that
λ 7→ ϕλϕ−1 induces an isomorphism HomO (I,O/I) −→ HomO (ϕ(I),O/ϕ(I)).
(2) Let (X, 0) ∼ 1
= (Y, 0) be germs of complex spaces. Prove that TX, ∼ 1
0 = TY,0 .
1
10.2.27. Let (X, 0) be a germ of a smooth space. Show that TX, 0 = 0.
10.2.28. Prove Lemma 10.2.2.
10.2.29. Prove that the definition of flat deformation is independent of the choice of generators
of (XS , 0).
10.2.30.
(1) Consider the germ of the complex space (X, 0) defined by the ideal I = (xz,xw,yz,yw) ⊂
C {x,y,z,w}. Thus V (I) is the union of two planes in C 4 intersecting in a point. Show
1
that TX, 0 = 0. Singularities with this property are called rigid.
1
(2) Give an example of a nonisolated singularity (X, 0) with finite-dimensional TX, 0.
` ´
10.2.31. Consider the curve singularity V (xy,xz,yz), 0 . Compute a semi-universal deforma-
tion of this curve singularity.
10.2.32. Prove that for a (flat) deformation (XS , 0) −→ (S, 0), and an analytic map p : (T, 0) −→
(S, 0) the map (X ×S T, 0) −→ (T, 0) is flat.
10.2.33. Let (X, 0) be a complete intersection with isolated singularity. Describe a versal de-
formation of (X, 0).
10.2.34. Finish the proof of Lemma 10.2.23.
10.2.35. Consider t(t − 1) variables xij for 1 ≤ i ≤ t − 1 and 1 ≤ j ≤ t. Consider the generic
determinantal variety (X, 0) ⊂ (C t(t−1) , 0) defined by the ((t − 1)–minors of the matrix (xij ).
Prove that (X, 0) is rigid.
10.2.36. Use a computer algebra system (for example SINGULAR) to compute Example 10.2.8
(Hint: Try in SINGULAR the following:
ring r =0,(x(0..4)),(c,dp);
matrix m[2][4] = x(0),x(1),x(2),x(3),x(1),x(2),x(3),x(4);
ideal i = minor(m,2);
module s = syz(i);
qring q = std(i);
module s = imap(r,s);
syz(transpose(s));)
10.2.37. Prove that the sequence (10.5) is exact.
(Hint: Use the same method as in the proof of the injectivity of α′2 .)
10.3 Existence of a Semi-Universal Deformation
In this section we will finally give the proof that for an isolated singularity a versal
deformation exists. We need many preparations.
Definition 10.3.1. Let (X, 0) be a germ of a complex space, A a local Artinian C –
algebra with maximal ideal mA . Suppose me+1
A = 0. Then a deformation (XA , 0) −→ (A, 0)
is called a deformation of order e.
Definition 10.3.2. Let (X, 0) be a germ of a complex space with local ring OX,0 =
C {x}/(f1 , . . . ,fk ). Let A = C [[s]]/J be a local formal C –algebra in the variables s1 , . . . ,sp ,
with maximal ideal mA . By abuse of notation, we denote by J also the ideal J · C {x}[[s]].
(1) A formal deformation (XA , 0) of (X, 0) over A is given by a quotient of C {x}[[s]]/J
by an ideal (F1 , . . . Fk ) such that
(a) Fi (x,0) = fi for i = 1, . . . ,k.
(b) For every relation r ∈ C {x}k , that is r · f = 0, there exists R ∈ C {x}[[s]]k
such that R · F = 0 as elements of C {x}[[s]]/J and R(x,0) = r.
10.3 Existence of a Semi-Universal Deformation 347
(2) For all e ∈ N, consider the local Artinian C –algebra Ae defined by Ae = C [[s]]/(J +
me+1
A ). Note that a formal deformation induces a deformation (XAe , 0) −→ (Ae , 0)
of order e for all e.
(3) A formal deformation (XA , 0) is called versal if for all local Artinian C –algebras
B, and all deformations (XB , 0) −→ (B, 0) there exists a homomorphism Ae −→ B
for some e ≫ 0 and an isomorphism of deformations (XAe ×Ae B, 0) ∼ = (XB , 0).
It is called semi-universal, if the induced homomorphism on the tangent spaces
(mA /m2A )∗ −→ (mB /m2B )∗ is uniquely determined.
Remark 10.3.3. Let (XA , 0) and (XB , 0) be two formal semi-universal deformations
of a germ of a complex space (X, 0). Then the formal analytic algebras A and B are
isomorphic.
To see this, note that, because (XA , 0) is semi-universal, we have a unique map α :
A1 −→ B1 , inducing (XB1 , 0) −→ (B1 , 0). Similarly, we get a map β : B1 −→ A1 . By semi-
universality, the composition β ◦ α is the identity on A1 , and α ◦ β is the identity on B1 .
Hence A1 and B1 are isomorphic. Now let e ∈ N be given and consider (XBe , 0) −→ (Be , 0).
By semi-universality, for some q there exists a map Aq −→ Be , so that (XBe , 0) −→ (Be , 0)
is induced by (XAq , 0) −→ (Aq , 0). As me+1
Be = 0 by definition, we can take q = e. Thus
we get a map φ : Ae −→ Be . Conversely, we get a map ψ : Be −→ Ae . The composition
ψ ◦ φ : Ae −→ Ae is such that the pull-back of the deformation (XAe , 0) −→ (Ae , 0)
is isomorphic to (XAe , 0) −→ (Ae , 0). By semi-universality, this map is uniquely defined
up to first order, and is the identity modulo the square of the maximal ideal. By the
Inverse Function Theorem8 , the map ψ ◦ φ is an isomorphism, and similarly φ ◦ ψ is an
isomorphism. Note that in particular dimC (Ae ) = dimC (Be ) for all e ∈ N.
Remark 10.3.4. Suppose that (XA , 0) is a formal semi-universal deformation of (X, 0).
Then the Kodaira-Spencer map (mA /m2A )∗ −→ TX, 1
0
is an isomorphism.
1
Namely, to construct the inverse mapping, consider an element of TX, 0
, that is,
2
a deformation over the space defined by the local ring C [[ε]]/ε . Because the formal
deformation is semi-universal, we get a unique map A1 −→ C [[ε]]/ε2 . By looking at the
coefficients of ε of the values of all elements in mA /m2A we get an element of (mA /mA )∗ .
One checks that this gives the inverse of the Kodaira-Spencer map.
Thus, as already said, we first want to show the existence of a formal semi-universal
deformation. The idea is to do this order by order, that is, define deformations (XAe , 0) →
(Ae , 0) of order e for all e ∈ N. For e = 1, we define a deformation over (A1 , 0), with
A1 = C [[s1 , . . . ,sτ ]]/m2 , and τ is the dimension of TX,
1
0
. Given Ae , it might be quite
obvious that we would like to have a deformation of order e + 1 over an as big as possible
space Ae+1 . In order to see that such a space exists, we will develop obstruction theory.
2
To formulate this, we need to define a further OX,0 –module, called TX, 0
.
Definition 10.3.5. Let a singularity (X, 0), with local ring OX,0 = On /I be given.
For simplicity, we write On = O. Take generators f1 , . . . ,fk of I. We consider the map
F := Ok −→ O, which sends the ei = (0, . . . ,0,1,0, . . . ,0) to fi . We let R be the kernel
of this map. Thus, we have an exact sequence
0 −→ R −→ F −→ O −→ OX,0 −→ 0.
8 This theorem holds in the formal case too, just ignore the convergence.
Let R0 ⊂ R be the submodule generated by the so-called Koszul relations fi ej −fj ei ∈ R.

Note that fi ej − fj ei is indeed in R, as obviously fi · fj − fj · fi = 0. Now we can define
2
TX, 0
:= HomO (R/R0 ,OX,0 )/ HomO (F ,OX,0 ).
2
Lemma 10.3.6. TX, 0
is a well-defined OX,0 –module.
Proof. The fact that it is an OX,0 –module is easy, as we are looking at homomorphisms
to OX,0 . Thus if we multiply a homomorphism with something in the ideal of (X, 0), it
will certainly become the zero homomorphism. As R ⊂ F we get dually a map
HomO (F ,OX,0 ) −→ HomO (R,OX,0 )
Now R0 is generated by elements of type fi ej − fj ei . If ϕ ∈ Hom(F ,OX,0 ), then this

element is sent to fi ϕ(ej ) − fj ϕ(ei ) = 0 ∈ OX,0 . This shows that we get an induced map
HomO (F ,OX,0 ) −→ HomO (R/R0 ,OX,0 ).

It remains to prove that the definition is independent of the choice of the generators
f1 , . . . ,fk of the ideal of OX,0 . This is left as Exercise 10.3.25.
Theorem 10.3.7. Let (X, 0) be a germ of a complex space with an isolated singularity.
2
Then TX, 0
is a finite-dimensional C –vector space.
The proof is left as Exercise 10.3.26. The main theorem on obstruction theory is the
following.
Theorem 10.3.8. Let (X, 0) be a germ of a complex space. Consider a local Artinian
(resp. formal, resp. analytic) C –algebra A, and a small surjection B −→ A, with kernel
J. This means that we have an exact sequence
0 −→ J −→ B −→ A −→ 0.
and m · J = 0, m the maximal ideal of B. Therefore, J is a vector space over C = B/m.

Suppose we have a deformation (XA , 0) −→ (A, 0) of (X, 0). For a short notation we write
ξ for this deformation. Then we have the following statements.
(1) There exists a well-defined obstruction element
2
ob(ξ)B→A ∈ TX, 0
⊗ J.
This obstruction element ob(ξ)B→A is zero, if and only if there exists a flat defor-
mation of (XB , 0) −→ (B, 0) extending the given deformation ξ,
(2) Suppose we have a commutative diagram
0 // J // B // A // 0

0 // J ′ // B ′ // A′ // 0
such that the rows are small extensions. Let ξ ′ be the deformation over A′ induced
2
by ξ. Let β : TX, 0
2
⊗ J −→ TX, 0
⊗ J ′ be the induced map. Then

ob(ξ ′ )B ′ →A′ = β ob(ξ)B→A .
Proof. We give the proof for local Artinian C –algebras, the remaining cases we leave
as Exercise 10.3.22. We first construct the obstruction element ob(ξ)B→A . Suppose the
k
deformation ξ, that is (XA , 0) −→ (A, 0), is given by the elements FA ∈ C {x} ⊗ A . Let
r ∈ R be a relation. By flatness we get a relation rA between the FA . Now take arbitrary
lifts FB ∈ C {x} ⊗ B of FA and a lift rB of rA . Then we consider
rB · FB ∈ OX,0 ⊗ B.
As rB is a lift of rA , and FB is a lift of FA it follows that the image of rB ·FB in C {x} ⊗ A

is zero. Thus we have in fact
rB · FB ∈ OX,0 ⊗ J.
We claim that the map r 7→ rB · FB defines an element in HomO (R/R0 ,OX,0 ⊗ J) =
HomO (R/R0 ,OX,0 ) ⊗ J,9 which induces the obstruction element ob(ξ)B→A ∈ TX,
2
0
⊗ J.
Step 1. We first show that the element rB · FB depends on r and on FB , but not on the
′
particular lift rB . In particular it is independent of the lift rA . So, take another lift rA
′ ′ ′
of r, with rA · FA = 0, and a lift rB of rA . First we show that we may assume that the
′ ′
restrictions rA of rB and rA of rB to A are equal. The idea of the proof is the same as
in the proof of Lemma 10.2.22, so it is done by induction on dimC (A). This part of the
proof does not use the fact that B −→ A is a small extension. The case dimC (A) = 1 is
our assumption. To do the induction step, take an element 0 6= α ∈ A with mA · α = 0.
′
We may suppose by induction that the image of rB and rB in A/(α) are equal. So we
′ k ′
can write rA − rA = h · α for some h ∈ C {x} . By assumption rA · FA = rA · FA = 0.
Hence h · FA α = 0. Now mA α = 0, and FA ≡ f modulo mA , so that h · f = 0. Thus h is
a relation, which can be lifted to a relation hA between the FA . We lift hA to hB , and
′ ′′ ′
change rB to rB := rB − hB α′ , for a suitable α′ which maps to α. Then rB ′′
is also a lift
′ ′′ ′′
of r. Moreover rB · FB is equal to rB · FB , and the restrictions of rB and rB to A are
equal.
′
Thus we now reached the stage that rB − rB has entries in C {x} ⊗ J, J being the
kernel of the map B −→ A. As mB J = 0 by assumption, and FB ≡ f modulo mB it
′ ′
follows that (rB − rB ) · FB = (rB − rB ) · f , which, as the fi are zero in OX,0 , is zero in
OX,0 ⊗ J. This shows that rB · FB is independent of the lift rB of r.
Therefore the map r 7→ rB · FB gives a well-defined element in HomO (R,OX,0 ) ⊗ J.
This element might depend on the choice of the lift FB .
Step 2. The restriction of the homomorphism to the submodule R0 generated by the
Koszul relations is zero. Indeed, we can take as lift of r = fi ej − fj ei simply the element
rB := FBi ej − FBj ei . Then rB · FB = 0.
2
Step 3. We now show that the class of rB ·FB gives a well-defined element in TX, 0
⊗J. Take
′ ′ k ′
another lift FB of FA . Then FB −FB ∈ C{x} ⊗J. We look at rB ·(FB −FB ). As mB J = 0,
and rB ≡ r modulo mB , it follows that this is equal to r · (FB − FB′ ). It follows that the
map r 7→ rB · (FB − FB′ ) is equal to the map r 7→ r · (FB − FB′ ) ∈ HomO (F ,OX,0 ) ⊗ J.
This is what we had to show.
Step 4. Suppose that an extension of ξ to a deformation (XB , 0) −→ (B, 0) exists. Then
we can find FB and for every r ∈ R an element rB such that rB · FB = 0 ∈ C {x} ⊗ J.
Thus, certainly, the obstruction element is zero.
9 This equality holds because the tensor product is taken over C and, therefore, all modules are free.
Step 5. On the other hand, suppose that ob(ξ)B→A = 0. Take an arbitrary lift FB′ =
′ ′
(FB1 , . . . ,FBk ) of FA . Because ob(ξ)B→A = 0, the element in HomO (R/R0 ,OX,0 ) ⊗ J
given by
r 7→ rB · FB′ , rB any lift of r,
is, in fact, in HomO (F ,OX,0 ) ⊗ J. Hence, there exist h1 , . . . ,hk ∈ C {x} ⊗ J such that
the map is given by
k
X
r = (r1 , . . . ,rk ) 7→ ri hi .
i=1
Now we define
′ ′
FB = (FB1 , . . . ,FBk ) := (FB1 , . . . ,FBk ) − (h1 , . . . ,hk ).
By construction, the element in HomO (R,OX,0 ) ⊗ J = HomO (R,OX,0 ⊗ J), given by
r 7→ rB · FB , rB any lift of r,
is the zero map. We claim that the FB defines a deformation which extends FA . To show
′ ′
this, for every relation r ∈ R a lift rB such that rB · FB = 0 has to be found. Now by
construction rB · FB = 0 as element of OX,0 ⊗ J. Therefore, there exist a1 , . . . ,ak in J
with rB · FB = a1 f1 + . . . + ak fk . Put a = (a1 , . . . ,ak ). As mB J = 0, and FB ≡ f modulo
′ ′
mB , it follows that a · FB = a · f . Now define rB := rB − a. Then rB is a lift of r with
′
rB · FB = 0. As we can do this for all r, it follows that FB gives a (flat) deformation of
(X, 0).
Step 6. The second statement of the theorem is immediate from the construction of the
obstruction element.
Corollary 10.3.9. Let (X, 0) be a germ of a complex space, and A = O/J a local Ar-
tinian (resp. formal, resp. analytic) C –algebra, and A′ := O/mJ. Let ξ be a deformation
(XA , 0) −→ (A, 0), and consider the small extension
0 −→ J/mJ −→ O/mJ −→ O/J −→ 0,

2
and let ob(ξ)A′ →A ∈ TX, 0
⊗ (J/mJ) be the obstruction element. By Exercise 1.2.46 this
obstruction element gives a map which we, by abuse of notation, give the same name:
2
∗
ob(ξ)A′ →A : TX, 0
−→ J/mJ.
Let I mod mJ be the image of the map ob(ξ)A′ →A . Then I is the unique minimal ideal
(with respect to inclusion) with the properties:
(1) mJ ⊂ I ⊂ J
(2) Let B = O/I. Then there exists a deformation (XB , 0) −→ (B, 0) of (X, 0) inducing
the deformation ξ.
The small surjection B −→ A we call a maximal extension with respect to the deformation
(XA , 0) −→ (A, 0).
Proof. We can write
ob(ξ)A′ −→A = α1 ⊗ h1 + . . . + αp ⊗ hp ,
2
for some linear independent elements α1 , . . . ,αp of TX, 0
and h1 , . . . hp of J. Let I =
(h1 , . . . ,hp ) + mJ. To show that a deformation (XB , 0) −→ (B, 0) exists, we consider the
local Artinian C –algebra O/I. Then ob(ξ)O/I→A is equal to α1 ⊗ h1 + . . . + αp ⊗ hp ,
2
but this time considered as element of TX, 0
⊗ (J/I), thus ob(ξ)O/I→A is zero. (See the
second part of Theorem 10.3.8.) Hence we forced the obstruction element to be zero, and
therefore there exists an extension of ξ over the Artinian C –algebra O/I. To show that I
is minimal and unique with respect to the two properties, let K be another ideal satisfying
mJ ⊂ K ⊂ J, such that there exists an extension of ξ over O/K. The obstruction element
is α1 ⊗ h1 + . . . + αp ⊗ hp , and therefore is zero considered as element of J/K. Hence
I = (h1 , . . . ,hp ) + mJ ⊂ K, which is what we had to prove.
Lemma 10.3.10. Let A = C [[s]]/J be a formal algebra, (XA , 0) be a formal deformation
of (X, 0) over A, and (XAe , 0) be the induced deformation of order e for all e. Consider
the small surjection
(10.6) C [[s]]/J + me+1 −→ C [[s]]/J + me .
Assume that
(1) C [[s]]/J + me −→ C [[s]]/J + me−1 is a maximal extension for the deformation
(XAe−1 , 0) −→ (Ae−1 , 0),

J+me+1 J+me
(2) dimC mJ+m e+1 = dim C mJ+me .
Then (10.6) is a maximal extension for the deformation (XAe , 0).

Proof. If C [[s]]/J + me+1 is not maximal, then there exists an ideal I $ J + me+1 with
mJ + me+1 ⊂ I ⊂ J + me , and such that the deformation extends to a deformation over
C [[s]]/I. Then the inclusion I + me ⊂ J + me is in fact an equality because of the first
e
condition. Therefore, the minimal number of generators of I modulo m is equal
to ethe
e
I+m J+m
minimal number of generators of J modulo me , that is, dimC mI+me = dimC mJ+m e .
As I $ J + me+1 , it follows that there is a minimal set of generators of J modulo me+1 ,
which has an element of degree e. In formula’s

J+me I+me I
dimC mJ+m e = dimC mI+m e ≤ dimC mI+m e+1 =

I J+me+1
dimC mJ+m e+1 < dimC mJ+m e+1 .
But this is in contradiction to the second condition.

It is time for an example.
Example 10.3.11. We consider the germ of the complex space (X, 0) defined by the
2–minors of the matrix
x0 x1 x2 x3
.
x1 x2 x3 x4
1
The modules NX,0 and TX, 0
have been described in 10.2.8 and 10.2.14. Thus we have the
following deformation ξ over C [[s1 ,s2 ,s3 ,t]]/m2 =: A, with m = (s1 ,s2 ,s3 ,t):
F12 = f12 − s1 x1 + tx0
F23 = f23 + s1 x3 − s2 x2 − tx2
F34 = f34 + s2 x4 − s3 x3 + tx4
F13 = f13 − s2 x1
F24 = f24 + s1 x4 − s3 x2
F14 = f14 − s3 x1 + tx2
The fij are the 2–minors of the above matrix as in Example 10.2.8. The Fij give a (flat)
deformation over C [[s1 ,s2 ,s3 ,t]]/(s1 ,s2 ,s3 ,t)2 , so we must be able to lift the relations.
Now we give the lift Rijk and Sijk of the relations rijk and sijk .
R123 = x0 F23 − (x1 + s1 )F13 + (x2 + s2 )F12
R124 = x0 F24 − (x1 + s1 )F14 + (x3 + s3 )F12 − tF13
R134 = x0 F34 − (x2 + s2 )F14 + (x3 + s3 )F13 − tF14 − tF23
R234 = (x1 + s1 )F34 − (x2 + s2 + t)F24 + (x3 + s3 )F23
S123 = x1 F23 − (x2 + t)F13 + x3 F12
S234 = x2 F34 − x3 F24 + x4 F23
S124 = x1 F24 − (x2 + t)F14 + x4 F12 − tF23
S134 = x1 F34 − x3 F14 + x4 F13 − tF24
These Rijk and Sijk are all zero modulo (s1 ,s2 ,s3 ,t)2 expressing the fact that we have a
(flat) deformation. We obtain the obstruction element by looking at the quadratic terms
in s and t, which one gets by simply expanding the terms:
r123 →
7 R123 = s2 tx0
r124 →7 R124 = s3 tx0 − s1 tx2 + s2 tx1
r134 →
7 R134 = s3 tx1 − s1 tx3
r234 →7 R234 = 0
s123 7→ S123 = s2 tx1
s124 7→ S124 = s3 tx1 − s1 tx3 + s2 tx2
s134 7→ S134 = −s1 tx4 + s3 tx2
s234 7→ S234 = 0
This map of R −→ OX,0 ⊗ J gives the element ob(ξ)A′ →A in TX, 2
0
⊗ J, where A′ =
3 2 3
C [[s1 ,s2 ,s3 ,t]]/m and J = m /m . Now, according ∗ to Corollary 10.3.9, we need a basis
2 2
of TX, 0
to describe the map ob(ξ) ′
A →A : T X,0 −→ m2 /m3 , and compute I, the image
of this map. We will see that we already computed part of a basis, which is enough for
our purpose. We have in particular that the elements α1 , α2 and α3 given in the first
three columns of the following table, (obtained by looking at the coefficients of s1 t, s2 t
and s3 t) give three elements in HomO (R/R0 ,OX,0 ). The last six columns correspond to
the images of the relations with respect to the canonical basis of HomO (F ,OX,0 ). We will
see that α1 , α2 and α3 are linearly independent modulo HomO (F ,OX,0 ), and therefore
2
induce linear independent elements of TX, 0
.
r123 7→ 0 x0 0 x2 x0 0 −x1 0 0
r124 7 → −x2 x1 x0 x3 0 0 0 x0 −x1
r134 7→ −x3 0 x1 0 0 x0 x3 0 −x2
r234 7 → 0 0 0 0 x3 x1 0 −x2 0
(10.7)
s123 7→ 0 x1 0 x3 x1 0 −x2 0 0
s124 7 → −x3 x2 x1 x4 0 0 0 x1 −x2
s134 7 → −x4 0 x2 0 0 x1 x4 0 −x3
s234 7 → 0 0 0 0 x4 x2 0 −x3 0
2
To see whether these elements are linearly independent in TX, 0
we have to look at the
2
“trivial” elements in TX,0 , that is, elements of Hom(F ,OX,0 ) which are given by the final
six columns of the above table (10.7). These six columns are the duals of f12 , f23 , f34 ,
f13 , f24 , and f14 respectively. Thus, for example, the first row and the final six columns
says that r123 is the relation x2 f12 +x0 f23 +0f34 −x1 f13 +0f24 +0f14 = 0. An elementary
but boring check now shows that α1 , α2 and α3 represent linearly independent elements
2 2
in TX, 0
. They can therefore be considered as part of a basis of TX, 0
and ob(ξ)A′ →A =
α1 ⊗ s1 t + α2 ⊗ s2 t + α3 ⊗ s3 t. Hence the minimal ideal I of the corollary is given by
the ideal (s1 t,s2 t,s3 t) + (s1 ,s2 ,s3 ,t)3 . Note, however, that the family considered above, in
fact, defines a deformation over C {s1 ,s2 ,s3 ,t}/(s1 t,s2 t,s3 t), as all the Rijk · F and Sijk · F
vanish modulo (s1 t,s2 t,s3 t) and not only modulo (s1 t,s2 t,s3 t) + (s1 ,s2 ,s3 ,t)3 .
Theorem 10.3.12. Let A = C [[s]]/J be a formal local C –algebra, and (X, 0) be a
germ of a complex space. Let (XA , 0) be a formal deformation of (X, 0). For all k put
Ak := A/mk+1 and Jk = J + mk+1 . Suppose that
(1) the Kodaira-Spencer map of (XA , 0) is surjective (resp. an isomorphism),
(2) for all e ≥ 2, the small extension Ae+1 −→ Ae is a maximal extension with respect
to the deformation (XAe , 0) −→ (Ae , 0).
Then (XA , 0) is a formal versal (resp. semi-universal) deformation.
Example 10.3.13. The family written down in the previous example is formally versal.
It is a deformation over A = C [[s1 ,s2 ,s3 ,t]]/(s1 t,s2 t,s3 t). Details are left as an exercise.
The basic idea of the proof is as the proof of Theorem 10.2.18, but there is a subtle
difficulty popping up, which is taken care of by Proposition 10.3.15. In its proof we need
the following lemma.
Lemma 10.3.14. Suppose that ϕ : B −→ A is a simple surjection,and let ψ : C −→ B
be a map of local Artinian C –algebras such that
(1) ϕ ◦ ψ : C −→ A is surjective,
(2) ϕ does not have a section.
Then ψ is surjective.
Proof. Suppose the converse. Let J = Ker(ψ). Then we have a injective map ψ : C/J −→
B. Moreover ϕ ◦ ψ : C/J −→ A is surjective. As ψ : C/J −→ B is not surjective, and
dimC (B) = dimC (A) + 1, it follows that dimC (C/J) = dimC (A). In Exercise 10.3.23
you will prove that a surjective algebra map of Artinian C –algebras of the same (vector
space) dimension is an isomorphism. Thus ϕ ◦ ψ is an isomorphism. Denote the inverse
by β : A −→ C/J. Then the map ψ ◦ β is a section of ϕ. This is a contradiction.
Proposition 10.3.15. Let local Artinian C –algebras A = C [[s1 , . . . ,sp ]]/I, and A′ =
C [[s1 , . . . ,sp ]]/J be given, for some ideals I, J with mJ ⊂ I ⊂ J. Let two further Artinian
C –algebras B and B ′ be given. Let ψ : A −→ A′ be the canonical surjection. Let ξ :
(XA′ , 0) −→ (A′ , 0) be a deformation of (X, 0). Suppose that
(1) the small extension A −→ A′ is a maximal extension with respect to the deformation
ξ : (XA′ , 0) −→ (A′ , 0);
(2) we have a simple surjection π : B −→ B ′ ;
(3) we have a map φ′ : A′ −→ B ′ and a deformation (XB , 0) −→ (B, 0) inducing
Cartesian diagrams
(XB′ , 0) −→ (XA′ , 0) (XB′ , 0) −→ (XB , 0)
↓ ↓ ↓ ↓
φ′
(B′ , 0) −→ (A′ , 0) (B′ , 0) −→ (B, 0)
Then there exists a map α : A −→ B making the following diagram commutative:

α
A −→ B
(10.8) ↓ψ ↓π
φ′
A′ −→ B′.
Proof. We consider the so-called fibered product
A′ ×B ′ B := {(a,b) ∈ A′ × B : φ′ (a) = π(b)}.
Defining addition and multiplication component wise A′ ×B ′ B is a local C –algebra. We
have a natural projection pr1 : A′ ×B ′ B −→ A′ which sends (a,b) to a. Note that pr1 is
also a simple surjection. Indeed, pr1 (a,b) = 0 implies that a = 0, and thus, by definition
π(b) = 0. Thus b is in the kernel of π, which by assumption is one-dimensional. It is
immediate from the definition that making the diagram (10.8) commutative is the same
as making the diagram
A′:: ×B ′ B
φ vv
v
vvv pr1
v
vv ψ
// A′
A
commutative.
Case 1. Suppose that pr1 has a section, that is, there exists a map s : A′ −→ A′ ×B ′ B
with pr1 ◦ s the identity. This is the easy case, as we then can define φ := s ◦ ψ.
Case 2. Suppose that pr1 does not have a section.
Step 1. We will show that A′ ×B ′ B is a quotient of C [[s1 , . . . .sp ]].10 We first construct
a map ω making the diagram
ω // A′ ×B ′ B
C [[s1 , . . . .sp ]]
pr1
ψ
A // A′
10 Note that both A and A′ are quotients of the same C [[s1 , . . . .sp ]].
commutative. Indeed, by assumption A′ is a quotient of C [[s1 , . . . .sp ]]. Consider the

composition
φ′
β : C [|s1 , . . . ,sp ]] −→ A′ −→ B ′ .
As π : B −→ B ′ is surjective, we can find bi ∈ B with π(bi ) = β(si ) for all i = 1, . . . ,p.
Now define
ω(si ) = (si ,bi ) ∈ A′ ×B ′ B.
As C [[s1 , . . . ,sp ]] is a formal free algebra, we can extend ω to an algebra map. Applying
Lemma 10.3.14 to the case C = C [[s1 , . . . ,sp ]]/ Ker(ω), ψ = ω,B = A′ ×B ′ B, A = A′
and ϕ = pr1 , we obtain A′ ×B ′ B ∼ = C [[s1 , . . . ,sp ]]/ Ker(ω).
Step 2. Write O = C [[s1 , . . . ,sp ]]. Put K := Ker(ω). Thus we get
A′ ×B ′ B ∼
= O/K ։ O/J = A′ .
Recall that A = O/I. We will show that ω induces a map φ : A −→ A′ ×B ′ B. To show

this we have to show that I ⊂ K. Now O/K ։ O/J is a simple surjection, thus in
particular small by Exercise 10.3.24. Hence mJ ⊂ K ⊂ J. As by assumption A −→ A′ is
a maximal extension with respect to (XA′ , 0) −→ (A′ , 0), we have I ⊂ K, if there exists
a deformation over O/K inducing the deformation ξ. This we show in Step 3.
Step 3. We now write D = A′ ×B ′ B for short. To complete the proof, it therefore suffices
to show that there exists a deformation
(XD , 0) −→ (D, 0)

inducing ξ. Suppose ξ is given by functions FA′ ∈ C {x}⊗Ab ′ k . Moreover, by assumption

we have deformations (XB , 0) given by say FB ∈ C {x}⊗B b k , and (XB′ , 0) given by

say FB ′ ∈ C {x}⊗Bb ′ k . As the deformation over (A′ , 0) and (B, 0) induces the same
deformation over B ′ , it can be easily seen that we may assume that
φ′ (FA′ ) = FB ′ = π(FB ).

Thus the pair (FA′ ,FB ) defines an element in C {x}⊗D b k . To show that it defines a (flat)
deformation, we have to lift relations. So let r ∈ C {x}k with r · f = 0 be an arbitrary
relation. By flatness we can find RA′ ∈ C {x}k ⊗ A′ with RA′ · FA′ = 0. Let RB ′ be the
image of RA′ under the natural map C {x}k ⊗ A′ −→ C {x}k ⊗ B ′ . By Lemma 10.2.22 we
can find an RB ∈ C {x}k ⊗ B lifting RB ′ with RB · FB = 0. The pair (RA′ ,RB ) therefore
is an element of C {x}k ⊗Db = C {x}k ⊗(A b ′ ×B ′ B), and (RA′ ,RB ) · (FA′ ,FB ) = 0. Hence
the relation r lifts. This finishes the proof of the proposition.
Proof of Theorem 10.3.12. Consider a local Artinian C –algebra B, and suppose that
we have a deformation (XB , 0) −→ (B, 0) of (X, 0). Let Ae := A/me+1 = C [[s]]/(J +
me+1 ). We have an induced deformation (XAe , 0) −→ (Ae , 0). Consider the following
small surjection
(10.9) 0 −→ J + me+1 /(mJ + me+2 ) −→ C [[s]]/(mJ + me+2 ) −→ Ae −→ 0
By assumption, for all e the ideal J mod me+2 is minimal among those ideals for which
there exists an extension of (XAe , 0) −→ (Ae , 0) over some small extension of Ae .
To show the versality, we have to show that there exists an e ∈ N and a mor-
phism of C –algebras Ae −→ B such that the deformation (XB , 0) −→ (B, 0) is iso-
morphic to (XAe ×Ae B, 0) −→ (B, 0). This we will prove by induction on the vector
space dimension dimC (B). We suppose that XAe is defined by elements F (e) (x,s) =
(e) (e)
F1 (x,s), . . . ,Fk (x,s) ∈ C {x}k ⊗ Ae = C {x}k ⊗ C [[s]]/Je . (Recall that Je = J +
me+1 .)
Step 1. Consider the induced deformation over B/m2B . As the Kodaira Spencer map is
surjective, we can find a map A1 −→ B inducing the deformation over B/m2B . This map
is uniquely determined in case the Kodaira-Spencer map is an isomorphism.
Step 2. Now take a nonzero element α ∈ B, such that mB α = 0. Put B ′ = B/(α), and
denote by π : B −→ B ′ the canonical projection. Then dimC (B ′ ) = dimC (B) − 1, and
therefore, by induction, we have for suitable e ∈ N a morphism of C –algebras
σ′
Ae −→ B ′
that induces a deformation which is isomorphic to the given deformation (XB′ , 0) −→

(B′ , 0), induced by (XB , 0) −→ (B, 0) via the projection π. By Proposition 10.3.15, there
exists a map σ : Ae+1 −→ B, making the diagram
σ
Ae+1 −→ B
↓ ↓π
σ′
Ae −→ B′
commutative.
Step 3. Now that we have our map σ we get two different deformations over (B, 0): The
given deformation (XB , 0) −→ (B, 0), and the deformation which is induced by σ. By
assumption the restriction of the two deformations to (B′ , 0) are isomorphic. Hence there
exist
(1) a k × k matrix Λ′ with entries in C {x} ⊗ B ′ , lifting the identity matrix and
(2) a B ′ – automorphism Φ′ defined by φ′1 = Φ′ (x1 ), . . . ,φ′n = Φ′ (xn ) ∈ C {x} ⊗ B ′
lifting the identity,
such that
F (e) (x,σ ′ (s1 ), . . . ,σ ′ (sp )) = FB′ (φ′1 , . . . ,φ′n ,t) ◦ Λ′ .
We now take any lift Λ of Λ′ , and φ1 , . . . ,φn of φ′1 , . . . ,φ′n . We will change σ, Λ, and
φ1 , . . . ,φn in such a way that:
F (e+1) (x,σ(s1 ), . . . ,σ(sp )) = FB (φ1 , . . . ,φn ,t) ◦ Λ.
This change of σ, φi , and Λ can be done exactly as in the proof of 10.2.18.

Theorem 10.3.16 (Schlessinger). Let (X, 0) be a germ of a complex space. Suppose
1
that TX, 0
is a finite-dimensional vector space, of dimension p, and a1 , . . . ap are elements
1
of NX,0 projecting to a basis of TX, 0
. The following construction gives a formal semi-
universal deformation of (X, 0).
(1) Consider new variables s1 , . . . ,sp , and consider A := C [[s1 , . . . ,sp ]], with maximal
ideal m = (s1 , . . . ,sp ). We set A1 := A/m2 and define a first order deformation:
(XA1 , 0) −→ (A1 , 0),

Pp
by taking F (1) (x,s) = f + i=1 si ai ∈ C {x}k ⊗ A1 .
(2) Now suppose the deformation (XAe , 0) −→ (Ae , 0), for Ae = O/Je has already been
constructed. We apply 10.3.9 to find an ideal Je+1 , minimal with respect to the
properties mentioned in that corollary. Then define Ae+1 := O/Je+1 . So by 10.3.9,
we can find (XAe+1 , 0) −→ (Ae+1 , 0) defined by F (e) (x,s). Continuing like this we
get formal power series F ∈ C {x}[[s]]/J, which defines a (flat) deformation over
J = ∩e Je .
Moreover, all formal semi-universal deformations of (X, 0) are obtained by this construc-
tion.
Lemma 10.3.17. Consider the ideals J and Je constructed in the statement of Theorem
10.3.16. Then Je = J + me+1 .
Proof. As Je+k ⊂ Je for all k by construction, and J = ∩e Je , it suffices to show Je+1 +
me+1 = Je and iterate this equation. This is proved by induction on e. The case e = 1 is
trivial, as J1 = m2 . So assume we proved Je +me = Je−1 . As me+1 ⊂ mJe−1 ⊂ Je ⊂ Je−1 ,
and Je+1 ⊂ Je by construction, the inclusion Je+1 +me+1 ⊂ Je follows. To show the other
inclusion, observe that there is a deformation of order e over the space with local ring
O/Je+1 + me+1 (lifting the deformation (XAe−1 , 0) −→ (Ae−1 , 0)) by simply restricting.
Hence, by minimality of Je we only have to show
mJe−1 ⊂ Je+1 + me+1 ⊂ Je−1 .
The inclusion Je+1 + me+1 ⊂ Je−1 is easy, as Je+1 ⊂ Je ⊂ Je−1 , and me+1 ⊂ me ⊂ Je−1
by construction. For the other inclusion, we use mJe−1 = mJe +me+1 ⊂ Je+1 +me+1 . The
equality follows from the induction hypothesis, the inclusion holds because mJe ⊂ Je+1
by definition.
Proof of Theorem 10.3.16. The fact that the constructed deformation is semi-universal
is an immediate consequence of Theorem 10.3.12, Lemma 10.3.17 and Corollary 10.3.9.
It remains to show that all formal versal deformations come from a construction as
in Theorem 10.3.16. This we do by contradiction. Thus let (XA , 0) be a formal semi-
universal deformation. Now certainly, because of the versality, the Kodaira-Spencer map
is surjective. Furthermore dimC (mA /m2A ) is equal to the dimension of TX,
1
0
, because the
deformation is semi-universal. This implies that the Kodaira-Spencer map is an isomor-
phism. Now suppose that for some e the small extension Ae+1 −→ Ae is not maximal
with respect to the deformation (XAe , 0) −→ (Ae , 0). We can follow the construction of
Theorem 10.3.16 to find a formal semi-universal deformation over A′ with A′e = Ae , and
A′e+1 ։ Ae+1 . But we have formal semi-universal deformations over both A and A′ it
follows that dimC (A′e+1 ) = dimC (Ae+1 ), see Remark 10.3.3. Thus Ae+1 was maximal
after all, proving the claim.
Remark 10.3.18. Note that in fact we proved something stronger. Indeed, under the
assumptions of Theorem 10.3.16, suppose we have a deformation (XB , 0) −→ (B, 0) of
order q. We have an induced deformation of order q − 1 over the space defined by the
Artinian ring Bq−1 := B + mq /mq . Suppose we have a map ϕ : Ae −→ Bq−1 inducing
this deformation. Then there exists a lift φe : Af −→ B for some f ≥ e, that is, we have
a commutative diagram
e
φ
Af // B
φ
Ae // Bq−1 ,
such that φe induces the given deformation (XB , 0) −→ (B, 0). This property of the con-
structed versal deformation is of great use in the proof of the existence of a convergent
semi-universal deformation.
Theorem 10.3.19. Let (XS , 0) and (S, 0) be germs of complex spaces, and suppose
(XS , 0) −→ (S, 0) is formally semi-universal. Then (XS , 0) is a semi-universal defor-
mation.
Proof. Let a deformation (XT , 0) −→ (T, 0) be given. Suppose that this deformation is

b T,0 k . We have to find a map (T, 0) −→ (S, 0) such
given by functions FT (x,t) ∈ C {x}⊗O
that the deformation (XT , 0) −→ (T, 0) is isomorphic to (XS ×S T, 0) −→ (T, 0). Write
OS,0 = C {s}/J, and OT,0 = C {t}/J ′ . To give a map (T, 0) −→ (S, 0) means that dually
we have to find a map φ : C {s} −→ C {t} with φ(J) ⊂ J ′ . The induced deformation
(XS ×S T, 0) −→ (T, 0) is given by FS (x,φ(s)) ∈ C {x,t}k . This has to be isomorphic
to the given deformation defined by FT . Therefore, we have to find an automorphism
Φ : C {x,t}n −→ C {x,t}n , and a matrix Λ ∈ C {x,t}k such that
(10.10) FS (x,φ(s)) = (FT ◦ Φ)(x,t) ◦ Λ mod J ′ · C {x,t}.
This is a system of system of equations we want to solve for φ, Φ and Λ. We want

to apply Grauert’s Approximation Theorem, and we therefore have to show that every
solution of (10.10) of order e can be lifted to a solution of order e + 1. By assumption,
(XS , 0) is formally semi-universal, hence is given by a sequence of maximal extensions by
Theorem 10.3.16. Note that Remark 10.3.18 says that every solution of order e can be
lifted to a solution of order e + 1. Thus we may indeed apply Grauert’s Approximation
Theorem.
Corollary 10.3.20. Let (X, 0) be a germ of a complex space, and let (XA , 0) be a (formal)
deformation over A = C [[s]]/J. Then (XA , 0) is (formally) semi-universal if and only if
(1) The Kodaira-Spencer map is an isomorphism,
2
∗
(2) The obstruction map TX, 0
−→ J/mJ is surjective.
A similar statement holds for convergent deformations.
Proof. From Theorem 10.3.19, it suffices to consider the formal case.
Step 1. First suppose that the Kodaira-Spencer map is an isomorphism, and that the
obstruction map is surjective. Consider the small extensions
J/mJ // C [[s]]/mJ // C [[s]]/J

(J + me+1 )/(mJ + me+2 ) // C [[s]]/mJ + me+2 // C [[s]]/J + me+1 .
We obtain, using the second part of Theorem 10.3.8, for the corresponding obstruction
2 ∗
map TX, 0
−→ J + me+1 /mJ + me+2 a commutative diagram
2
∗
TX, // J/mJ
0
QQQ
QQQ
QQQ
QQQ
Q((
J + me+1 /mJ + me+2 .
2
∗
As by the second assumption, the obstruction map TX, 0
−→ J/mJ is surjective, the
image of the obstruction map of the second extension is thus (J + me+2 )/(mJ + me+2 ).
By Corollary 10.3.9 we see that for all e the extension C [[s]]/J + me+2 = Ae+1 −→ Ae =
C [[s]]/J + me+1 is maximal with respect to the deformation (XAe , 0) −→ (Ae , 0). Thus
we have a formal versal deformation by Theorem 10.3.12.
Step 2. Now suppose (XA , 0) is formally semi-universal. Consider the formal deforma-
tion constructed in Theorem 10.3.16. By construction, for all e, the obstruction element
associated to the extension
0 −→ (J + me+1 )/(mJ + me+2 ) −→ C [[s]]/mJ + me+2 −→ C [[s]]/J + me+1 −→ 0,
maps surjectively onto Je+1 /mJe+1 = J + me+2 /mJ + me+2 . Now for some e ≫ 0 the
minimal number of generators of Je+1 mod me+2 is equal to the minimal number of
generators of J by Exercise 1.3.19. Thus the obstruction map is surjective. The Kodaira-
Spencer map is an isomorphism by definition of the first order deformation.
Now we come to the proof of the main theorem.
Theorem 10.3.21 (Grauert). Let (X, 0) be a germ of a complex space. Suppose that
1
dimC (TX, 0
) < ∞. Then there exists an analytic semi-universal deformation of (X, 0).
Proof. The previous theorem says that it remains to give a convergent deformation which
is formally semi-universal. Let (X, 0) be defined by f = (f1 , . . . ,fk ) ∈ C {x}k , and let
r1 , . . . ,rt ∈ C {x}k generate the module of relations between the f1 , . . . ,fk . Take a formal
semi-universal deformation of (X, 0) over A = C [[s]]/J, and let m = (s). Let e be a
natural number, such that the minimal number of generators of J is equal to the minimal
number of generators of J modulo me+1 . As C [[s]] is Noetherian local ring, such an e
exists according to Exercise 1.3.19. Let g1 , . . . ,gp be the generators of J +me+1 /me+1 . We
may assume that the gi are polynomials in s. In order to apply Grauert’s Approximation
Theorem, we will have to interpret the flatness condition in terms of equations. Now
flatness means that we can find lifts F,R1 , . . . ,Rt ∈ C {x,s} of f,r1 , . . . ,rt and vectors
M1 , . . . ,Mt ∈ C {x,s}p with
(10.11) Ri · F − g · Mi = 0 for i = 1, . . . t.
We want to apply Grauert’s Approximation Theorem 8.2.2 to this set of equations. Our
formal semi-universal deformation of (X, 0) over A = C [[s]]/J gives solutions modulo
(q) (q)
mq+1 of (10.11), F (q) , Ri , g (q) , and Mi for all q ≥ e, lifting those for q = e. Now
′ ′
′ (q) ′ (q)
suppose F (q) , Ri , g (q) and Mi is a solution modulo mq+1 of (10.11) lifting F (e) ,
(e) (e) (e)
Ri , g , and Mi . In order to apply Grauert’s Approximation Theorem, we have to
′ (q)′ ′ (q)′
prove that the F (q) , Ri , g (q) and Mi can be lifted to solutions modulo mq+2 . The
′ ′
(q) (q)
ideal Jq′ = (g1 , . . . ,gp ) has p generators, just as many as Je . From Lemma 10.3.10 it
follows that the surjection C [[s]]/Jq′ −→ C [[s]]/Je is induced by a sequence of maximal
small extensions. Therefore Jq′ can by 10.3.16 be extended to an ideal J ′ , so that there
exists a formal semi-universal deformation over C [[s]]/J ′ . Hence from Remark 10.3.3 it
follows that the formal algebras C [[s]]/J ′ and C [[s]]/J are isomorphic, and thus that
C [[s]]/Jq′ and C [[s]]/Jq are isomorphic for all q. Thus for dimension reasons the minimal
ideal I with
′
(1) mJq+1 ⊂ I ⊂ Jq′ and,
(2) there exists a lift of the deformation over (A′q , 0) to one over (A′q+1 , 0) with A′q+1 =
C [[s]]/I,
′
is the ideal Jq+1 , as this is also the case for Jq+1 . Thus there does indeed exists a lift
′ (q+1)′ ′ (q+1)′ ′ (q)′ ′ (q)′
F (q+1) , Ri , g (q+1) and Mi of F (q) , Ri , g (q) and Mi . Thus Grauert’s
Approximation Theorem 8.2.2 applies, and the theorem is proved.
Exercises
10.3.22. Prove Theorem 10.3.8 for the formal and the analytic case.
(Hint: Use Exercise 1.3.19 to show that there exists an n ∈ N with OX,0 ⊗J ∼
= OX,0 ⊗(J +mn /mn ),
and use the theorem already proved for the Artinian case.)
10.3.23. Let ϕ : B −→ A be a surjective map of Artinian C –algebras. Suppose that dimC (A) =
dimC (B). Show that ϕ is an isomorphism.
10.3.24. Prove that a simple surjection of Artinian C –algebras is small.

2
10.3.25. Show that the definition of of TX, 0 is independent of the choice of generators of the
ideal of (X, 0).

1
(Hint: Use similar arguments as in the corresponding proof for TX, 0 .)
10.3.27. Prove Theorem 10.3.21 for the algebraic case: Let (X, 0) be an isolated singularity,
1
defined by algebraic power series. Suppose dimC (TX, 0 ) < ∞. Then there exists a semi-universal
deformation of (X, 0) which is defined by algebraic power series.
(Hint: Use Corollary 8.3.7.)
361
Bibliography
[Artin 1968] Artin, M.: On the solution of analytic equations. Inv. math.
5, 277–291.
[Brieskorn-Knörrer 1986] Brieskorn, E., Knörrer, H.: Plane Algebraic Curves.
Birkhäuser, Basel, Boston, Stuttgart.
[Buchberger 1965] Buchberger, B.: Ein Algorithmus zum Auffinden der Ba-
siselemente des Restklassenringes nach einem nulldimen-
sionalen Polynomideal. Dissertation Innsbruck.
[Eisenbud 1995] Eisenbud, D.: Commutative Algebra with a View Toward

Algebraic Geometry. GTM 150. Springer-Verlag, Berlin
etc..
[Gabrielov 1971] Gabrielov, A.M.: The formal relation between analytic func-
tions. Funktional Anal i Priložen 5, no. 4, 64–65.
[Gordan 1899] Gordan, P.: Neuer Beweis des Hilbert’schen Satzes über ho-
mogene Funktionen. Nachrichten König. Ges. der Wiss. zu
Göttingen, 240–242.
[Grauert 1972] Grauert, H.: Über die Deformation isolierter Singularitäten
analytischer Mengen. Inv. Math 15, 171–198.
[Grauert-Remmert 1971] Grauert, H., Remmert, R.: Analytische Stellenalgebren.
Grundl. 176. Springer Verlag, Berlin etc.
[Grauert-Remmert 1984] Grauert, H., Remmert, R.: Coherent Analytic Sheaves.
Grundl. 265. Springer Verlag, Berlin etc.
[Gunning-Rossi 1965] Gunning, R., Rossi, H.: Analytic Functions of Several Com-
plex Variables. Prentice-Hall, Englewood Cliffs, NJ.
[Hironaka 1964] Hironaka, H.: Resolution of Singularities of an Algebraic
Variety over a Field of Characteristic Zero. Ann. of Math.
79, 109–326.
[Kurke et al. 1975] Kurke, H., Pfister, G., Roczen, M.: Henselsche Ringe und
algebraische Geometrie. VEB Deutscher Verlag der Wis-
senschaften, Berlin.
[Kurke et al. 1978] Kurke, H., Pfister, G., Popescu, D.; Roczen, M.; Mostowski,
T.: Die Approximationseigenschaft lokaler Ringe. Springer
Lecture Notes 634.
[Lang 1965] Lang, S.: Algebra. Addison-Wesley, Reading, Mas-
sachusetts.
362
[Matsumura 1979] Matsumura, H.: Commutative Algebra. W.A. Benjamin,

New York.
[Mora 1982] Mora, T.: An Algorithm to Compute the Equations of Tan-

gent Cones. Proc. EUROCAL 82, Lecture Notes in Comp.
Sci. 144, 158–165.
[Popescu 1985] Popescu, D.: General Néron Desingularization. Nagoya
Math. J. 100, 97–126.
[Popescu 1986, 1990] Popescu, D.: General Néron Desingularization and Approx-
imation. Nagoya Math. J. 104, 85–115; 118, 45–53.
[Pfister-Popescu 1975] Pfister, G., Popescu, D.: Die strenge Approximationseigen-
schaft lokaler Ringe. Inv. Math. 30 (82), 145–147.
[Schlessinger 1968] Schlessinger, M.; Functors of Artin rings. Trans. Am. Math.
Soc. 130, 208–222.
[Singular 2000] Greuel, G.-M., Pfister, G., Schönemann,H.: The Comput-
eralgebra System singular,
http://www.singular.mathematik.uni-kl.de.
[Spivakovsky 1994] Spivakovsky, M.: Non–existence of the Artin Function for
Henselian Pairs. Math. Ann. 299, 727–729.
[Swan 1998] Swan, R.: Neron–Popescu Desingularization, in Algebra
and Geometry, Ed. Ming-chang Kang, International Press,
Cambridge, 135–192.
[Spivakovsky 1999] Spivakovsky, M.: A new proof of D. Popescu’s Theorem on
smoothing of ring homomorphisms. J. of AMS 294, 381–
444.
[Van der Waerden 1971] Van der Waerden, B.L.: Moderne Algebra, eighth ed.
Springer Verlag, Berlin etc.

3528031379

Uploaded by

Copyright:

Available Formats

3528031379

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

3528031379

Uploaded by

Copyright:

Available Formats

i

Voor Jeannette, Nils en Noah

if and only if f is an element of I.

2 Affine Algebraic Geometry 46

3 Basics of Analytic Geometry 73

4 Further Development of Analytic Geometry 124

5 Plane Curve Singularities 169

6 The Principle of Conservation of Number 219

7 Standard Bases 269

8 Approximation Theorems 286

9 Classification of Simple Hypersurface Singularities 302

10 Deformations of Singularities 329

1.1 Noetherian Rings

Definition 1.1.1. A ring R is called Noetherian if every ideal in R is finitely generated.

(2) Any chain of ideals in R

(a1 , . . . ,ak−1 ) $ (a1 , . . . ,ak )

fk − b1 xdeg(fk )−deg(f1 ) f1 − · · · − bk−1 xdeg(fk )−deg(fk−1 ) fk−1

(1) Let R be a ring. An R–algebra S is a ring S together with a ring homomorphism

(1) K[x]/(xn ), n ≥ 1, is an Artinian K–algebra.

Definition 1.1.10. Let G be an abelian group. A G–graded ring is a ring R together

An element of Rd is called a homogeneous polynomial of degree d. Consider f ∈ R.

Again, R = ⊕d≥0 Rd , where Rd consists of all quasi-homogeneous polynomials of

As xν ∈ Iν for ν = 2, . . . ,m, it follows that x ∈ Iν for ν = 2, . . . ,m. As yν ∈ I1 , it follows

(2) We have multiplication with scalars ·R × M −→ M :

These should satisfy the following conditions.

Definition 1.2.2. Let M and N be R–modules, ϕ : M → N a map.

• ϕ is called an isomorphism if and only if ϕ is a bijective homomorphism.

(1) A K–module, for a field K, is just a K–vector space.

We collect some more definitions.

(1) Let ∅ 6= A ⊂ M . Then A is called a submodule of M if for all x,y ∈ A and a ∈ R we

(7) A set {m1 , . . . ,mt } is called a basis of M if every m ∈ M can be written as an R–

Tors(M ) := {m ∈ M : there exists a nonzerodivisor a ∈ R with a · m = 0}.

So in case R is an integral domain, we have to find a nonzero element a ∈ R with

(ϕ + ψ)(m) := ϕ(m) + ψ(m).

for all m ∈ M and a ∈ R.

is called a complex if Ker(αk ) ⊃ Im(αk−1 ). It is called exact at Mk if Ker(αk ) =

is called a short exact sequence. This is equivalent to saying that α is injective, β is

Lemma 1.2.12 (Snake Lemma). Consider a commutative diagram3

of R–modules, with exact rows. Then the sequence

Note that a ring R is Noetherian if and only if R, considered as an R–module, is Noethe-

• Submodules and quotient modules of Noetherian modules are Noetherian.

M is Noetherian ⇐⇒ N and M/N are Noetherian.

is commutative, then there exists a unique homomorphism h : S ⊗R T −→ A such that

1.2.22. Let M be an R–module, and I = Ann(M ) = {x ∈ R : x · M = 0}. Then M is in a

1.2.24. Prove that Tors(M ) is a submodule of M .

1.2.25. Consider an exact sequence

1.2.26. Let K be a field, and consider an exact sequence

1.2.27. Prove the Snake Lemma 1.2.12.

1.2.28. Prove Cramer’s rule 1.2.8.

1.2.29. Prove Lemma 1.2.14.

1.2.33. Let M,M ′ ,N,N ′ be R–modules and α : M ′ −→ M , β : N −→ N ′ be homomorphisms.

1.2.35. Consider an exact sequence

· · · Mk+1 /xMk+1 −→ Mk /xMk −→ Mk−1 /xMk−1 −→ · · ·

1.2.36. Let 0 −→ M −→ N −→ L be a complex of R–modules. Prove that the complex is exact

0 −→ HomR (A,M ) −→ HomR (A,N ) −→ HomR (A,L)

1.2.37. Let M −→ N −→ L −→ 0 be a complex of R–modules. Prove that the complex is exact

0 −→ HomR (L,A) −→ HomR (N,A) −→ HomR (M,A)

1.2.38. Let M,M ′ ,N,N ′ be R–modules and α : M ′ −→ M , β : N ′ −→ N be homomorphisms.

1.2.39. Prove that M ⊗R N ∼