Lecture Notes in

Mathematics
Edited by A. Dold and B. Eckmann

Subseries: Mathematisches Institut der Universit&t und

Max-Planck-lnstitut fer Mathematik, Bonn - vol. 3
Adviser. E Hirzebruch

1101

Vincent Cossart
Jean Giraud
Ulrich Orbanz

Resolution
of Surface Singularities
Three Lectures with an Appendix by H. Hironaka
Edited by U. Orbanz

Springer-Verlag
Berlin Heidelberg New York Tokyo 1984
Authors
Vincent Cossart
Universite Pierre et Marie Curie (Paris Vl), Mathematiques
4, Place Jussieu, 75005 Paris, France

Jean Giraud
Universite de Paris-Sud, Centre d'Orsay, Math6matique
B&t. 425, 91405 Orsay Cedex, France

Ulrich Orbanz
Max-Planck-lnstitut fiJr Mathematik
Gottfried-Claren-Str. 26, 5300 Bonn, Federal Republic of Germany

AMS Subject Classification (1980): 14 E15, 14J 17

ISBN 3-540-13904-4 Springer-Verlag Berlin Heidelberg New York Tokyo
ISBN 0-387-13904-4 Springer-Verlag New York Heidelberg Berlin Tokyo

This work ~s subject to copyright. All rights are reserved, whether the whole or part of the material
is concerned, specifically those of translation, reprinting, re-use of illustrations, broadcasting,
reproduction by photocopying machine or similar means, and storage in data banks. Under
§ 54 of the German Copyright Law where copies are made for other than private use, a fee is
payable to "Verwertungsgesellschaft Wort", Munich.
© by Springer-Verlag Berlin Heidelberg 1984
Printed in Germany
Printing and binding: Beltz Offsetdruck, Hemsbach/Bergstr.
2146/3140-543210
RESOLUTION OF S U R F A C E SINGULARITIES - THREE LECTURES

Introduction

The p r o b l e m of r e s o l u t i o n of the s i n g u l a r i t i e s of an a l g e b r a i c surface

has a long h i s t o r y reaching back to the last century. After its s o l u t i o n
(by R. W a l k e r in 1935), it was quite reasonable to ask for d e s i n g u l a r -
ization of a l g e b r a i c varieties of a r b i t r a r y dimensions. For the general
problem the surface case and the d i f f e r e n t methods of its s o l u t i o n w e r e
of special importance, not only as a tool for h i g h e r dimension, b u t also
as a testing ground for the g e n e r a l case. To quote D. Hilbert: "Viel-
leicht in den m e i s t e n Fillen, wo wir die A n t w o r t auf eine Frage vergeb-
lich suchen, liegt die U r s a c h e des M i B l i n g e n s darin, dab wir einfachere
noch u n v o l l k o m m e n erledigt haben. Es k o m m t dann alles darauf an, diese
leichteren Probleme aufzufinden und ihre L ~ s u n g e n mit m~glichst voll-
kommenen Hilfsmitteln und durch verallgemeinerungsf~hige Begriffe zu be-
werkstelligen."

Meanwhile the general problem of d e s i n g u l a r i z a t i o n in c h a r a c t e r i s t i c

zero has been s o l v e d by Hironaka, but there are some points a b o u t his
proof w h i c h are a c h a l l e n g e for further investigations. F i r s t of all,
since his proof is very complicated, it is natural to look for simplifi-
cations. One such s i m p l i f i c a t i o n for instance is the n o t i o n of i d e a l i s t i c
exponents, introduced by H i r o n a k a himself. Then, of course, the case of
positive characteristics remains open (in d i m e n s i o n > 3). I want to
stress another point which seems unsatisfactory: If H i r o n a k a ' s proof is
specialized for surfaces, there is still no s u b s t a n t i a l simplification of
his procedure. This is one reason to look for d i f f e r e n t methods to
desingularize surfaces, thereby hoping to find some more "natural" method
for d e s i n g u l a r i z a t i o n in the general case.

Another point of view is the c l a s s i f i c a t i o n of s i n g u l a r i t i e s by their

resolution (as in the case of plane curves). Of course, for this p u r p o s e
one needs a canonical procedure for resolution. In a vast g e n e r a l i z a t i o n
of the plane curve case, Abhyankar has developed a machinery that will
ultimately lead to a c a n o n i c a l procedure for d e s i n g u l a r i z a t i o n . But
there again the m e c h a n i s m is too c o m p l Z c a t e d to allow an easy d e s c r i p t i o n
in d i m e n s i o n 2.
 IV

Yet a n o t h e r a s p e c t of r e s o l u t i o n in d i m e n s i o n 7 1 is the p r o b l e m of
globalizing some local (or even punctual) algorithm. The m o s t compli-
cated p a r t of H i r o n a k a ' s paper is d e v o t e d to this p r o b l e m and also in
Abhyankar's proof of r e s o l u t i o n of surfaces in p o s i t i v e characteristics,
globalizing presents a serious difficulty.

To summarize, a good proof in the surface case should combine the

following features: It should be canonical, it should be easy in some
sense, and g l o b a l i z a t i o n s h o u l d be no problem.

The starting point for a new d e v e l o p m e n t in the r e s o l u t i o n of surfaces,

after H i r o n a k a ' s proof in 1964, was Zariski's paper [6] 1 in w h i c h the
author applied his new theory of e q u i s i n g u l a r i t y to give a new proof
for r e s o l u t i o n of surfaces. By using suitable projections as for equi-
singularity, he i n t r o d u c e d the n o t i o n of q u a s i - o r d i n a r y singularities.
He also e x t r a c t e d numerical invariants to m e a s u r e the i m p r o v e m e n t under
(suitably chosen) monoidal transformations. The result was the first
canonical proof for r e s o l u t i o n of surfaces that did not use n o r m a l i z a -
tion.

Abhyankar and H i r o n a k a followed these ideas along d i f f e r e n t lines.

Using a detailed description of the c h a r a c t e r i s t i c polyhedron, Hironaka
modified the d e f i n i t i o n of a q u a s i - o r d i n a r y singularity to give another
proof for r e s o l u t i o n of Surfaces, including characteristic p > O (see
appendix). Abhyankar made the d i s t i n c t i o n between "good" and "bad"
points, corresponding to their b e h a v i o u r along a s e q u e n c e of m o n o i d a l
transformations. This distinction is the result of a careful analysis
of Zariski's method of q u a s i - o r d i n a r y singularities, with the i n t e n t i o n
to suppress all p r o p e r t i e s of s i n g u l a r i t i e s which are not d i r e c t l y re-
lated to the process of resolution. A special feature of A b h y a n k a r ' s
method is that a s l i g h t (but very subtle) modification gives the stron-
ger r e s u l t of " e m b e d d e d resolution" (see first lecture for e x p l a n a t i o n ) .
This result in turn seems to have inspired Zariski to use his q u a s i -
ordinary singularities for a d i f f e r e n t proof of e m b e d d e d resolution in
1978 [7].

N O W we come to the c o n t e n t of the c o n t r i b u t i o n s of this volume, all of

which h a v e been p r e s e n t e d in a s e m i n a r on n u m e r i c a l characters of
singul a r i t i e s , held in Bonn 1979-1981. The first lecture is a c o m p l e t e l y

I see the r e f e r e n c e s of the f i r s t lecture

self-contained presentation of A b h y a n k a r ' s method for r e s o l u t i o n of
surfaces, b a s e d on a lecture of A b h y a n k a r ' s given in Paris in M a y 198Q.
Abhyankar's proof meets all r e q u i r e m e n t s stated above: It is canonical,
the local m e t h o d globalizes trivially , and the proof is easy and short
We note that m e a n w h i l e there are some e f f o r t s to use this p r o o f for the
classification of some surface singularities.

The second lecturebyGiraud introduces the various formulations of

resolution and r e l a t e d problems. Then some different methods for reso-
lution of surfaces are sketched, and finally an a t t e m p t is made to de-
scribe Zariski's ideas of 1967 clearly, without actually giving the
proof.

The third lecture by C o s s a r t contains a detailed comparison of

Hironaka's and A b h y a n k a r ' s generalizations of the n o t i o n of q u a s i -
ordinary singularities. Abhyankar's method is t r a n s l a t e d into the
langu a g e of c h a r a c t e r i s t i c polyhedra and q u a s i - o r d i n a r y points in the
sense of H i r o n a k a in order to d e s c r i b e more p r e c i s e l y the links and
the d i f f e r e n c e s between the m e t h o d s of A b h y a n k a r and Hironaka. Abhyan-
kar's m e t h o d is e x t e n d e d to p o s i t i v e characteristics, but only non-
embedded resolution is treated in this lecture.

Hironaka's lectures on the r e s o l u t i o n of e x c e l l e n t surfaces of any

characteristic, given at Bowdoin College and w r i t t e n down by B.M.
Bennett, have been distributed privately for m a n y years. Since these
notes are a b a s i c reference for the subject, they are i n c l u d e d here as
an a p p e n d i x in o r d e r to make them available for the m a t h e m a t i c a l com-
munity.

The three lectures differ considerably in style, due to the i n t e n t i o n s

and tastes of the authors. The first lecture uses the l a n g u a g e of local
algebra, and it a s s u m e s only a little background in this field. The
second lecture uses the l a n g u a g e of m o d e r n A l g e b r a i c Geometry. Since
this lecture does n o t c o n t a i n proofs, some familiarity with this lan-
guage is s u f f i c i e n t for the reader. Cossart's contribution is self
contained modulo the B o w d o i n College lectures by H i r o n a k a (see appendix)
and a t e c h n i c a l result in one of C o s s a r t ' s papers.

I gratefully acknowledge the s u p p o r t by the M a x - P l a n c k - I n s t i t u t f~r

Mathematik (Bonn) during the final p r e p a r a t i o n of this volume.

U. Orbanz
 CONTENTS

INTRODUCTION
III

Lecture I: Embedded resolution of a l g e b r a i c I

surfaces after A b h y a n k a r
(characteristic 0)
by U. O r b a n z

Lecture 2: Desingularization in low d i m e n s i o n 51

by J. G i r a u d

Lecture 3: Desingularization in d i m e n s i o n two 79

by V. C o s s a r t

Appendix: Desingularization of e x c e l l e n t surfaces, 99

Lectures by H. H i r o n a k a , Advanced Study
Seminar in A l g e b r a i c Geometry,
Bowdoin CQllege 1967,
Notes by B.M. Bennett
EMBEDDED RESOLUTION OF ALGEBRAIC SURFACES AFTER ABHYANKAR
(Characteristic O)

Ulrich Orbanz

This lecture is b a s e d on four s e m i n a r talks given by S.S. Abhyankar in

Paris between May 5 and May 8, 1980. I had the o p p o r t u n i t y to a t t e n d
these lectures and to discuss the topic with S.S. Abhyankar. We a g r e e d
that it w o u l d be w o r t h having a written exposition of his talks, not
only containing all the details that could not be given orally, but also
making the proof accessible to n o n - s p e c i a l i s t s , assUming only some
basic facts of c o m m u t a t i v e algebra.

The m a t e r i a l presented by A b h y a n k a r in Paris is the content of chapter

III, and I tried to keep this close to his instructive oral exposition.
My own contribution restricts itself to the a r r a n g e m e n t of the back-
ground material given in chapters I, II and IV.

According to Abhyankar, his results contained in this lecture have been

obtained by him in 1967, but the only p u b l i c a t i o n on this proof was
Lipma n ' s sketch in his A r c a t a Lectures [3]. M e a n w h i l e Abhyankar himself
has w r i t t e n two papers on the subject, for w h i c h my lecture may serve
as an introduction:

I) D e s i n g u l a r i z a t i o n of plane curves, in S i n g u l a r i t i e s , Amer. Math.

Soc., Proc. Symp. Pure Math. 40 (1983), 1-45.

This paper contains the results parallel to chapter II of the p r e s e n t

lecture, given in the t e r m i n o l o g y developed in [8].

2) Good points of a hypersurface, to appear in the A d v a n c e s in

Mathematics.

As the title indicates, this long p a p e r contains a generalization of

the n o t i o n of a "good point" to any hypersurface, and the techniques
for this notion are d e v e l o p e d for any dimension. At the end, the em-
bedded resolution of surfaces appears as an a p p l i c a t i o n of the more
general theory.
I am very much indepted to Prof. Abhyankar for his continual support
during the p r e p a r a t i o n of these notes.
Table of Contents

Introduction ...................................................... I

Table of contents ................................................. 3

Elementary properties of blowing up ........................... 4

§I Conventions and preliminaries .............................. 4
§2 Blowing up ................................................. 5
§3 A s s o c i a t e d graded rings .................................... 6
§4 Some special results ....................................... 8

II Resolution of plane curves by weighted initial forms .......... 11

§I W e i g h t e d initial form with respect to a f i x e d r e g u l a r
s y s t e m of p a r a m e t e r s ....................................... 11
§2 C r i t e r i o n for v(z;f) = v(f) ................................ 15
§3 Criterion for v(f) ....................................... 18
§4 Immediate quadratic transforms ............................. 19
§5 E m b e d d e d resolution of plane curves and principalization ... 21

III Resolution and embedded resolution of e m b e d d e d surfaces in

characteristic O .............................................. 24
§I General assumptions and statement of the results ........... 24
§2 Main example: Strict transform for f = zn+xay b ............. 25
§3 Good points: Definition and elementary properties .......... 27
§4 Strict transform: Power series case ........................ 29
§5 Bad points are finite in n u m b e r ............................ 33
§6 Review of the procedure to resolve F ....................... 34
§7 E m b e d d e d resolution - good triples ......................... 35
§8 E m b e d d e d resolution - power series case .................... 37
§9 Bad points for embedded resolution are finite in number .... 40
§10 Review of the procedure of embedded resolution ............ 41

IV Auxiliary results for surfaces ................................ 42

§I Equimultiple curves under quadratic transformations ........ 42
§2 Equimultiple curves under monoidal transformations ......... 43
§3 Equimultiple curves under completion ....................... 44
§4 Use of Tschirnhausen transformation ........................ 47

References ........................................................ 49
I Elementary properties of b l o w i n g up

§I C o n v e n t i o n s and preliminaries

All rings considered here are supposed to be c o m m u t a t i v e and to c o n t a i n

a unit element. By a local ring we mean a noetherian ring having a unique
maximal ideal. If R is a l o c a l ring, the m a x i m a l ideal of R w i l l be
denoted b y M(R) (and the d e f i n i t i o n of maximal ideal includes M(R) ~ R,
so t h a t R ~ 0). The completion of a l o c a l ring R will be denoted b y R ~.

For any local r i n g R, the function ord(R) :R + ~ U {~} is d e f i n e d by

ord(R) (f) = s u p {n 6 ~Z I f 6 M(R)n} , f 6 R.

So by Krull's intersection theorem

ord(R) (f) = ~ ~ f = O.

For the theory of regular local rings we refer to [4] o r [5]. W e w i l l

make use of the facts that regular local rings are U n i q u e Factorization
Domains, and that any localization of a r e g u l a r ring is r e g u l a r again.

Furthermore we need some facts about excellent rings t h a t c a n be found

in [4]. If R is a n e x c e l l e n t ring, t h e n so is a n y localization of R a n d
any finitely generated R-algebra (this includes homomorphic images). If
R is an e x c e l l e n t domain, then the integral closure R of R in its q u o -
tient field is a f i n i t e l y generated R-module. We w i l l a l s o use that the
formal fibres of a n e x c e l l e n t local r i n g R are r e g u l a r . This means that
given any prime ideal P of R a n d a n y p r i m e ideal Q of R ~ s u c h that
Q n R = P, the local ring (R~)Q/P(R*)Q is r e g u l a r . In p a r t i c u l a r if R
is r e d u c e d , then so is R ~.

Next we define the n o t i o n of normal crossing for a n y s e t of i d e a l s in

a regular local ring.

Definition. Let R be a regular local ring and let I* be a s e t of i d e a l s

of R. I* is s a i d t o h a v e normal crossings if R h a s a regular system of
parameters x = {Xl, .... , x r} w i t h the following property: If P is a n y
prime ideal of R a s s o c i a t e d to a n i d e a l I of I*, then P is g e n e r a t e d by
a subset of x.
Sometimes also elements of R w i l l be said to h a v e normal crossing, by
which we mean that the principal ideals generated by these elements
have normal crossings. For example, a nonzero element f of R has normal
crossings if a n d only if there is a regular system of p a r a m e t e r s
Xl, .... , x r of R s u c h that f-R is generated by a monomial in
x I, .... , x r. There is an o b v i o u s way to translate the definition of
normal crossing into geometry, i.e. to a r e g u l a r point P on a noetherian
scheme X and a set of c l o s e d subschemes passing through P.

§2 B l o w i n g up

We will describe blowing up o n l y locally. The verification of the global

nature of this process is left to the reader.

Given a local ring R and an ideal I of R, the b l o w i n g u p of R w i t h center

I is a family of h o m o m o r p h i s m s R ÷ R I of local rings obtained in the
following manner. Given any x 6 I, x ~ O, w e consider the ring Rx ob-
tained by inverting the powers of x. The subring of R x g e n e r a t e d by the
image of R (under the canonical homomorphism) and the elements a/x,
a 6 I, w i l l be denoted by R[I/x]. (Note that R[I/x] is g e n e r a t e d over
R by a l / x , .... , an/x, where a I, .... , a n is any system of generators of
I. T h e r e f o r e R[I/x] is n o e t h e r i a n . ) The homomorphism R ~R I belongs
to the blowing u p of R w i t h center I if and only if there is some
x 6 I, x ~ O, some prime ideal P in R [ I / x ] and an R-isomorphism
R[I/X]p ÷ RI making the diagram

R[I/x]

R1

commutative, where R ÷ R[I/X]p is the canonical homomorphism.

Let R I = R[I/X]p for some nonzero x in I and some prime ideal P of

R[I/x]. If a 6 I, then a/1 = x • (a/x) in R [ I / x ] , therefore I • R[I/X] =
x • R[I/x] and of course I - R I = x R I. W e note that x is a n o n - z e r o d i v i s o r
in R I s i n c e it is a non-zerodivisor in R [ I / x ] . If R is a domain (which
will be the case in the applications later on), RI w i l l be considered to
be a subring of the quotient field of R, so the h o m o m o r p h i s m R + R I is
simply the inclusion.

Let R I = R[I/x]p as above and let Q be the inverse image of P in R. If

P does not contain the image of x in R [ I / x ] , or equivalently if Q does
not contain I, then R I is R - i s o m o r p h i c to RQ. To see this, observe that
Rx maps canonically to RQ, and the image of R[I/x] under this map is
the same as the canonical image of R. So the more interesting case is
the one in w h i c h I R I = x R I is contained in M(RI) . T h e s e rings can also
be described by using associated graded rings (see §3) .

If R I is o b t a i n e d by blowing up R w i t h center I and if J is an ideal

of R contained in I, one can define an ideal J1 in RI, called the strict
transform of J in RI, with the following property: J1 contains JR I , and
the induced homomorphism R/J ÷ RI/J I isobtained by blowing up R / J with
center I/J. J1 is uniquely determined by these properties. In the appli-
cations we will only need the case where J is a nonzero principal ideal,
R is a r e g u l a r local ring and I an ideal such that R/I is regular again.
In this case, for any nonzero f 6 I let n be the largest integer such
that f 6 In and put fl = f/xn- Then fiR1 will be called the strict trans-
form of f R in R I. O n e checks easily that fl RI does not depend on the
choice of the element x ~ I for which I - R I = x R I. S o m e t i m e s we will also
call fl the strict transform of f, although f determines fl only up to
a unit. Finally one can check that fl RI has indeed the universal
property for the strict transform stated above. (To s e e this, note that
f.g 6 Im implies g E im-n for any g E R, since R and R/I were assumed
to be regular.)

If I is g e n e r a t e d by elements Xl, .... , x s a n d if R I is o b t a i n e d by

blowing up R with center I, then it is e a s y to see that R I is R - i s o -
morphic to a l o c a l i z a t i o n of R [ I / x j l for some j, I ! J ~ s.

§3 A s s o c i a t e d graded rings

For any ring R and any ideal I in R, the associated graded ring of R
with respect to I is defined to b e

gri(R ) = ~ In/In+1.
nkO
If R is local, the associated graded ring with respect to the maximal
ideal will simply be denoted by gr(R) . If f 6 R, the initial form ini(f)
of f with respect to I is the canonical image of f in I n / I n+1, where
n is the largest integer such that f 6 In (inIf = 0 if n = ~) . If R is
local, then the initial form of f with respect to M(R) will be denoted
by in(f). We will also have occasion to look at the initial ideal of
an ideal J of R w h e r e R is local. This is the ideal in gr(R) generated
by the initial forms of e l e m e n t s of J (with respect to M ( R ) ) , and it
will be denoted by gr(J,R) .

To d e s c r i b e the connection between blowing up a n d associated graded rings

we introduce some notation. Let
S = n~> S n be a graded ring and J a
O
homogeneous ideal. The homogeneous part of degree n of J will be denoted
by Jn- A homogeneous prime ideal Q of S will be called relevant if
QI ~ S I . By S(Q) we denote the subring of SQ c o n s i s t i n g of quotients
a/b, where a,b are homogeneous elements of the same degree in S a n d
b £ Q. If Q is relevant, then as for ordinary localization, the prime
ideals of S(Q) correspond to the homogeneous prime ideals of S contained
in Q in a o n e - t o - o n e manner.

Now let R be a local ring, I an ideal of R, x a nonzero element of I

and P a prime ideal of R[I/x]. Let R I = R[I/x]p and assume that
x - R I c M(RI) " We claim that there is a u n i q u e l y determined relevant
homogeneous prime ideal Q of gri(R) such that RI/XR I and gri(R) (Q) are
isomorphic as R - A l g e b r a s . To see this, take Q to be the ideal generated
by the initial forms ini(Y), where y 6 In~I n+1 for some n and y/x n 6 p,
or m o r e precisely Qn may be defined to c o n s i s t of all ini(y) where
y 6 In and y/x n E p. Then Q is a prime ideal, and since we may assume
x £ I~I 2, ini(x) ~ Q and Q is relevant. Now it is clear how to d e f i n e
a homomorphism from gri(R) (Q) to R I / x R I . If a / b 6 gri(R) (Q) , w h e r e a
and b are of degree n, let a = ini(y) a n d b = ini(z) for some y,z 6 In •
Then b £ Q implies z/x n £ P, and therefore the residue class of y/z =
( y / x n ) / ( z / x n) mod xR I will be defined to be the image of a/b. It is
left to the reader to v e r i f y that this defines indeed an isomorphism.

For the application of b l o w i n g up to r e s o l u t i o n of surfaces in c h a p t e r

III it is e n o u g h to look at those rings R I = R[I/X]p for w h i c h P is a
maximal ideal of R [ I / x ] . Writing R I / X R I ~ gri(R) (Q) as above, this means
that Q is a maximal element in the set of relevant homogeneous primes
of gri(R) . U n d e r suitable assumptions one can show that this implies
that Qo = M ( R ) / I . This is e a s y to see for the application we have in
mind, namely the case in w h i c h R is a r e g u l a r local ring of dimension 3
and R/I is a o n e d i m e n s i o n a l regular local ring. In this case gri(R)
is (isomorphic to) a polynomial ring in two v a r i a b l e s over R/I, a n d if
Q is a m a x i m a l homogeneous relevant prime in t h i s ring, t h e n ht(Q) = 2
and consequently Qo = M/I.

Now if Q o = M/I, then Q corresponds to a u n i q u e relevant homogeneous

prime Q o f gr1(R) ~ R R / M ( R ) , and
RI/M(R)RI = (grI(R) ~RR/N(R)) (Q)
as R - a l g e b r a s . Note that under the a s s u m p t i o n s made above, gri(R) ~ R R / M ( R )
is a p o l y n o m i a l ring in t w o v a r i a b l e s over R/M(R) and therefore Q is a
principal ideal.

It s h o u l d b e clear from these considerations that the o b j e c t of s t u d y

for the r e s o l u t i o n of s u r f a c e s (and a l s o of c u r v e s ) will be local
homomorphisms R ÷ RI, where d i m R I = d i m R.

Definition. A local homomorphism R ÷ R I of local rings will be c a l l e d

a quadratic transform, if this h o m o m o r p h i s m is o b t a i n e d by blowing up
R with center M(R) a n d d i m R I = d i m R.

Definition. A local homomorphism R + R I of local rings will be called

monoidal transform with center P if t h i s h o m o m o r p h i s m is o b t a i n e d by
blowing up R w i t h center P such that P ~ M and R/P is r e g u l a r , and
d i m R I = d i m R.

To conclude this section we remark that any quadratic (resp. monoidal)

transform R ÷ R I of R i n d u c e s uniquely a quadratic (resp. monoidal)
transform of the c o m p l e t i o n R ~ of R. A s n o t e d above, such a transform
is u n i q u e l y determined b y a (relevant) homogeneous prime ideal Q of
gr(R) (resp. grp(R) ~ R R / M ( R ) ) , and these graded rings do not change when
passing to c o m p l e t i o n .

§4 S o m e special results

Now let R b e a r e g u l a r local ring and P a prime ideal of R s u c h that

R/P is r e g u l a r , and l e t us l o o k a little closer a t the r i n g s obtained
by blowing up R w i t h center P. N o w R h a s a regular system of p a r a m e t e r s
Xl, .... , X r , X r + 1 , .... , x d s u c h t h a t P is g e n e r a t e d b y x I, .... , x r.
Let t = x I and l e t R I = R [ P / t ] N for s o m e m a x i m a l i d e a l N of R [ P / t ] .
Then R I / t R I is i s o m o r p h i c to grp(R) (Q) for some homogeneous prime ideal
Q of grp(R) such that i n p ( x I) £ Q. L e t X i = inp(xi), i = I, .... ,r, so
that grp(P) can be identified with the p o l y n o m i n a l ring
(R/P) [Xl, .... ,Xr]. Then gr(R) (Q) is a l o c a l i z a t i o n of the p o l y n o m i n a l
ring (R/P) [X2/Xl, .... ,Xr/X I] a n d h e n c e regular. Consequently R I is
regular and x I is p a r t of a r e g u l a r system of p a r a m e t e r s of R I.

We will describe a regular system of p a r a m e t e r s explicitely in the case

that d i m R = d i m R I = 3. A s s u m e first that P = (Xl,X2) . In this case
RI/M(R) R I = R 1 / ( X l , X 3 ) R I is i s o m o r p h i c to a l o c a l i z a t i o n of the p o l y -
nomial ring R/M(R) [X2/X I] at a p r i n c i p a l ideal, generated by ~ ( X 2 / X I)
say. Let p(T) 6 R[T] be o b t a i n e d by lifting the coefficients of ~ to
R. T h e n R1/(xl,p(x2/xl),X3)R I is a f i e l d and h e n c e x l , P ( X 2 / X l ) , X 3 is a
regular system of p a r a m e t e r s . Note that p is just the m i n i m a l polynomial
of the r e s i d u e of X 2 / X I m o d M ( R I) o v e r R/M(R) (up to a unit).

N e x t a s s u m e that P=(Xl,X2,X3)=M(R), a n d let RI/tRIZR/M(R)[ X 2 / X I , X 3 / X I] M'

where M is a m a x i m a l ideal of R/M(R) [ X 2 / X I , X 3 / X l ] . It
is n o w clear that M is g e n e r a t e d by two p o l y n o m i a l s P2,P3, the first of
which may be c h o s e n to be the m i n i m a l polynomial of the residue Of
x 2 / x I m o d M ( R I) o v e r R/M(R) . If Pi are obtained by lifting the coeffi-
cients of Pi to R (i = 2,3), then clearly xl,P2(X2/Xl),P3(X3/X I) w i l l
be a r e g u l a r system of p a r a m e t e r s of R I.

We w i l l need the following result related to the s t r i c t transform of an

element in R. A s s u m e that R is a r e g u l a r local ring and that R I is a
monoidal or q u a d r a t i c transform of R w i t h center P (so in p a r t i c u l a r
R/P is r e g u l a r ) . We assume t h a t ht(P) = 2. L e t f be a n o n z e r o element of
R and fl a s t r i c t transform of f in R I. We claim that if ord(R) (f) =
ord(R1) (f I) > O, then R I is u n i q u e l y determined and the h o m o m o r p h i s m
R ÷ R I induces an i s o m o r p h i s m of r e s i d u e fields. To p r o v e this, we fix
the following notation. Let n = ord(R) (f) and P = (x,y)R. Then
grp(R) ~ R R / M ( R ) can be identified with R/M(R) [X,Y], and R I / M ( R ) ' R I is
isomorphic to R/M(R) [X,Y](Q), Q a relevant homogeneous prime ideal. Let
g be the image of inp(f) in grp(R) ~ R R / M ( R ) . Then g is h o m o g e n e o u s of
degree ~ n and ord(R1) (f I) = n i m p l i e s that g 6 Q(n). But under our
assumption Q must be p r i n c i p a l . It f o l l o w s that Q is g e n e r a t e d by a
linear f o r m h and g = hn (up to a unit). This shows uniqueness of Q a n d
hence of RI, and since h is of d e g r e e I, R I / M ( R I) = R/M(R) . N o t e that
g must be of d e g r e e n, i.e. f 6 pn, and therefore g m a y be identified
with in(f) 6 gr(R) via the canonical homomorphism grp(R) ~ R R / M ( R ) ÷
 10

gr(R) . W e conclude with the following remark. Assume that PR I = xR I .

Since RI/M(R)R I = R/M(R), we have (y/x)-a 6 M ( R I) for a suitable a 6 R.
By the considerations made earlier it is clear that x, ( y / x ) - a will be
part of some regular system of parameters of R I.
 11

II R e s o l u t i o n of p l a n e curves by weighted initial forms

Throughout this chapter, R will denote a two-dimensional regular local

ring. We put M = M(R), k = R/M, and for any a 6 R, p(a) denotes the
residue class of a in k.

§1 Weighted initial form with respect to a f i x e d regular system of

parameters

Proposition I. L e t x , z be a r e g u l a r system of p a r a m e t e r s and let v,w,e

be p o s i t i v e real numbers. Assume that there are e l e m e n t s a(i,j) in R
(0 < i,j < n) such that

a(i,j)xiz j = ~ a(i,j)xiz j
iv+jw=e iv+jw>e

Then a(i,j) £ M for all (i,j) such that iv+jw=e.

Proof. Let

F(X,Z) = a(i,j)xlz 3 (1 .I)

iv+jw=e

and

a(i,j)xizJ. (I .2)
G(X,Z) = iv+3w> e

If a E M, say a = bx+cz, then axizj = bxi+IzJ + c x i z j+1, so to p r o v e

the a s s e r t i o n we may assume in a d d i t i o n that

for all (i,j) such that iv+jw=e we have that either

(I .3)
a(i,j) = O or a(i,j) is a u n i t in R. f

P u t s(j) = e / v - j (w/v) . T h e n (1.1) can be rewrftten as

d
F(X,Z) = ~. c(t) x S ( t ) z t, (i .4)
t=O

where again for a l l t

c(t) = O or c(t) is a u n i t in R. (1.5)

 12

We have to s h o w t h a t F(X,Z) = O, so a s s u m e the c o n t r a r y and c h o o s e d

in (I .4) such that c(d) ~ O. N o w let

Fo(X,Z) = F(X,Z) - c(d)xS(d)z d = d~ c(t) x S ( t ) z t (1 .6)

t=O

and

G(X,Z) = GI(X,Z) + G2(X,Z ) (1 .7)

where

GI(X,Z) = ~. a(i,j)xiz j
iv+jw>e
j~d
(1 .8)
G2(X,Z ) = ,~ a ( i , j ) x i z j.
iv+jw>e
j>d

T h e n we c a n w r i t e G2(X,Z) = zd+IH(x,z) and we get

c(d) -x s(d) zd _ zd+IH(x,z) = G I (x,z) - F o(x,z) . (i .9)

We have

s(t) > s(d) if t < d and

(I .10)
j < d and i v + j w > e ~ i > s(d) .

F r o m this w e see that

Fo(X,Z) 6 xS(d)+IR and G1(x,z) 6 xS(d)+IR.

Therefore

z d ( c ( d ) x s(d) - zH(x,z)) 6 xS(d)+IR

by (1.9), and we c o n c l u d e that

c ( d ) x s(d) - zH(x,z) 6 xS(d)+IR.

This clearly implies c(d) 6 M in c o n t r a d i c t i o n to (1.5) and c(d) ~ O.

 13

Remark. This proof shows t h a t the same result is v a l i d if x , z is a

regular sequence in a n y local r i n g R a n d M is r e p l a c e d b y x R + yR.

For the r e s t of t h i s section we fix a regular system of p a r a m e t e r s

x , z of R a n d a n o n z e r o element f of R, a n d w e p u t n = ord(R) (f) . W e
define

V(z,x;f) = {v 6 ]R I v > I and f = ~ a(i,j)zlx 3

iv+j~nv
for s o m e a(i,j) 6 ~ }.

Given v 6 V(z,x;f) a n d a(i,j) 6 R such that

f = ~-- a(i,j)z±x 3
iv+j~nv

we define w = min [j/(n-i) 1 i < n a n d a(i,j) # 0}. T h e n w ~ v,

w 6 (I/n!)~ and w £ V(z,x;f). It f o l l o w s that

sup V(z,x;f) 6 (1/n!)~ U {~}.

Let y 6 R be such that z R + y R = M a n d l e t us c o m p a r e V(z,x;f) and

V(z,y;f). First of all x = a z + b y for s o m e u n i t b of R. If v 6 V ( z , x ; f
and

f = ~ a ( i , j ) z l x 3,
iv+j~nv

by substituting x = az+by we get an equation of the form

f = ~ b ( r , s ) z r y s,
r,s

and we have

{ (r,s) I b(r,s) ~ O} _c { (i+j-k,k) I iv+j>nv and j>_k}.

Therefore if b(r,s) # O, it f o l l o w s that

rv + s = iv + jv - k v + k : iv + j + (v-l) (j-k) > nv

since v > I and j > k. T h e r e f o r e v 6 V(z,y;f) and by symmetry we obtain

V(z,x;f) = V(z,y;f)
 14

and in particular

sup V(z,x;f) = sup V(z,y;f).

Definition. a) v(z;f) = sup V(z,x;f) .

b) v(f) = sup {v(z;f) 1 z 6 M~M2}. So v(z;f) and v(f) are elements
of (I/n!)~ U {~}.
Clearly we have v ( z ; z n) = ~. If r is a nonnegative integer and
ir + j > nr, then

• . ] znR if i ~ n
zlx 3 6
Mr if i < n

It follows that if v ( z ; f ) = ~ then f 6 N (znR + Mr) = znR, and

therefore r

v(z;f) = ~ ~ fR = znR.

We proceed to d e f i n e the weighted initial form of f with respect to

z,x,which will be an element of the polynomial ring k[Z,X] (remember
k = R/M) . F o r this purpose k[Z,X] will not be identified with gr(R),
since the definition of the weighted initial form is not intrinsic
(but see remark below).

Definition. If v 6 V(z,x;f) and

f = ~ a ( i , j) z i x j
iv+j_>nv

we define

L(z,x;v) (f) = ~, p(a(i,j))Zlx 3

iv+j=nv

Remark. This is w e l l defined by Proposition I. If z and v are chosen such

that v = v(z;f) = v(f), then L(z,x;v) (f) can be defined intrinsically as
an element of gr(R) (see [8 ]).

Let v = v(z;f) and let w 6 V(z,x;f) be such that w < v. Then we have an
equation

f = ~ a(i, j) z i x j
iv+ j>nv
 15

and for i # n w e h a v e iw + j > nw. Therefore

L ( z , x ; w ) (f) = p ( a ( n , O ) ) -Z n.

Conversely assume that w 6 V(z,x;f) is s u c h that

L ( z , x ; w ) (f) = p ( a ) Z n, a E R.

Then

f = azn + .~ a(i,j) zix j,

iw+j>nw

and we define

u = inf {j/(n-i) I a(i,j) # 0 and i < n}.

We get that

u > w and u E V(z,x;f) .

Therefore, for a n y w 6 V ( z , x ; f ) we have

w < v(z;f) ~ L ( z , x ; w ) (f) = p(a) Z n for s o m e a E R. (1.1)

§2 C r i t e r i o n for v(z;f) = v(f)

Proposition 2. L e t z*,x* be a r e g u l a r system of p a r a m e t e r s of R, let

f be a nonzero element of R, n = ord(R) (f), and let

F(Z,X) = L(z~,x~;v(z~;f)) (f) 6 k[Z,X].

Then v(z~;f) < v(f) if a n d o n l y if t h e r e is a p o l y n o m i a l

G(Z,X) 6 k[Z,X] and an element a 6 R such that

F(Z,X) = p(a) G(Z,X) n.

The p r o o f of this p r o p o s i t i o n is t h e c o n t e n t of t h i s section and will

be obtained by an explicit description of the w e i g h t e d initial form
for a s e c o n d system of p a r a m e t e r s z,x of R.
 16

Let V 6 V(z,x;f) . Then f can be written as

f = az n + ~ a ( i , j ) z l x 3. (2.1)
iv+j>nv
i<[

Let w be a positive integer and z* = z - b x w, so that z*,x generate M.

Then

f = a(z*+bxW) n + ~-/ih_k i-k j+kw

a(i,j)z~ |kid y x 2.2)
iv+j>nv k=0 \ /
i<~

Note that (i-k)w + (j+kw) = iw+j. If n o w w < v and i < n, then

iv + j > n v ~ j > (n-i)v ~ j > (n-i)w ~ iw + j > nw.

Therefore, if w < v then

L(z*,x;w) (f) = L(z*,x;w) ( a ( z * + b x W ) n) = p(a) (Z + p ( b ) x w ) n. 2.3)

From (1.1) we conclude that v(z~;f) = w < v if p(a) ~ 0 ~ p(b) .

Therefore

If f is as in (2.1) with a 6 M and z* = z - bx w with ] 2.4)

b £ M, then v(z*;f) > v ~ w _> v.

Let us rewrite (2.2) as

~---{i h, k i - k j + k w
f = d-~ a(i,j) d__~.~l m y x 2.5)
i v + j_>nv k = O \'~j

where a(n,O) = a. N o w

[iv+j = dv and k = 0 if w > v

(i-k) v + J + k w dv 2

[iv+j = dv if W = v.

Therefore

p(a(i,j)) zix j if w > v

L(z*,x;v) (f) =~iv~=nv (2.6)
p(b)kzk-ix j+kv if w=v.
p ( a ( i , j) ~ = o < k )
~iv+j=nv

Define
 17

if w > v
G(Z,X) = L(z*,x;v)(z) = [Z
Z + p(b)X v if w = v

and define H(Z,X) : L(z,x;v)(f). Then (2.6) can be written as

L(z~,x;v) (f) = H(G(Z,X) ,X) . 2~7)

Furthermore in the case v = v(z;f), by (1.1) and (2.7) we have

W > V(Z;f) ~ V(Z~;f) = V(z;f) . 2.8)

We now turn to the proof of Proposition 2 and assume that we are glven
regular systems of parameters z,x and z ~ , x * of R such that

v(z*;f) < v(z;f). (2.9)

Assume first that v(z~;f) = I. W r i t i n g

f = ~ a(i, j) z ' i x ~j
i+j=n

we see that in gr(R)

in f = ~ p(a(i,j))in(z~)iin(x~) j = p(a)-in(z) n
i+j=n

for some a 6 R. So assume v(z~;f) > I. T h e n

in f = p(a)-in(z) n = p(a')in(z~) n, a,a' 6 R

and therefore z~,x generate M. Let w = v(z~;z). Then

z = cz~ + bx w

and c,b are units in R. We may assume c = I. L e t v = v(z~;f) 6 V(z,x;f)

and H(Z,X) = L(z,x;v) (f) . B y (1 .I) and (2.9) , H ( Z , X ) = p(a) "Z n for some
a 6 R, and so from (2.3) and (2.7) we see that

L(z~',x;v) (f) = p(a)-G(Z,X) n, v = v(z~;f) (2.10)

where G(Z,X) = L(z~,x;v)(z).

Assume conversely that (2.10) holds. Then necessarily we have

 18

G(Z,X) = p(c)Z + p ( d ) X v.

So if we define z = cz* + d x v, from (2.7) we conclude that

H(Z,X) = L(z,x;v)(f) = p(a) Z d

and therefore v(z;f) > v by (1.1).

§3 Criterion for v(f) =

Let f 6 R, ord(R) (f) = n < ~ as before and let R* be the completion of

R. If z,x is a regular system of parameters of R and v 6 V(z,x;f),
then for the weighted initial form L(z,x;v) (f) it does not make any
difference whether f is considered as element of R or as element of
R ~. Consequently, by (1.1) and Proposition 2, the numbers v(z;f) and
v(f) will be the same for R and R*.

Assume that v(f) = ~, and let zo = z, vo = V(Zo;f) . If vo < ~ = v(f)

then by the proof of Proposition 2 we can find e I, cI £ R so that

V I = V(Zl;f) > V o where zI = elz O + ci x v O .

Proceeding inductively, we either obtain some zi_ I 6 R such that fR =

z~ i R, or we can find
i--

v. 6 IN, e. 6 R, c. £ R and z. 6 R
1 1 1 1

such that

zI = eizi_ I + ci xvi-1, vi = v ( z .1; f ) > vi-1 = v(zi_1;f)

Actually we may assume e. = I for all i, so that

l

V.
zi+ I - zi 6 M i and lim v i = ~.
i+oo

Let

z* = lim z. £ R ~.

Then
 19

z* - z i 6 M vi for all i

and therefore

f 6 (z'R*) n + (MR*) v"l for all i.

Therefore

f R* = (z'R*) n

So we have shown
Proposition 3. v(f) = ~ if and only if there is a regular parameter z*
of R* such that fR* = (z'R*) n

Assume in addition that R is excellent and let P = f~. Then P'R* =

/~w. Therefore, if v(f) = ~, then R*/PR* = (R/P)* is regular and P is
generated by some regular parameter z of R. So we get
Proposition 3'. If R is excellent, then v(f) = ~ if and only if there is
a regular parameter z of R such that fR = znR.

§4 Immediate quadratic transforms

Let f be a nonzero element of R such that v(f) < ~. We choose a regular

system of parameters z,x of R such that v(f) = v(z;f) . We let n =
ord(R) (f) >O as before. Let R + R I be a quadratic transform of R and let
fl be a strict transform of f in RI, and assume ord(R1) (f I) = n. By
the observation made in I,§4, R I is uniquely determined and in(f) is
equal (up to a unit) to the n th power of a homogeneous element of degree
one in gr(R) . T h e r e f o r e v(f) > I, and by the choice of z we have in(f)
= in(z) n u p to a u n i t . Hence zI = z/x, x is a regular system of parame-
ters of RI, and RI/M(RI) = k. Let v I : V ( Z l ; f I) and write

f] : ~ b(i,j) z~x j with b(i,j) 6 R~

iv1+jhnv I
i<n

Then

f = fl x n b(i,j)zix j+n-1,
ivl+j>nv I
i<n
 2O

and

i(v1+1) + j + n - i = iv I +j+n _> n ( v 1 + 1 ) .

Since R I / M ( R I) = R/M, we may assume that

b(i,j) 6 R if iv I + j : nv I .

It follows that

v(f) > vI + I > 2

and furthermore if we put

L(z,x;v1+1) (f) = F(Z,X)

L(Zl,X;Vl) (f) = FI(X,Z)

then

F I (Z,X) = X n F(Z/X,X) .

From (1.1) we conclude that vI + I = v(f), and therefore by Proposition

2 it f o l l o w s that v I = v(fl) .

Now instead o f o r d ( R I) (fl) = n, assume

o r d ( R I) (fl) > O and v(f) > 2.

Then in(f) = p[a)-in(z) n for some a 6 R, so R I is the only quadratic

transform of R for which o r d ( R 1 ) (fl) > O, and z I = z/x, x is a r e g u l a r
system of parameters o f R I. W r i t i n g

f = ~ a(i, j) z l x 3
2i+j>2n
i<~

where v = v(f), we get

fl = ~ a ( i ' J ) z% x j + i - n
2i+j>2n
i<n

we see that fl 6 M(RI)n~ We summarize our results in

 21

Proposition 4. L e t f 6 R be such that o r d ( R ) (f) = n < ~ and v(f) < ~.

Choose z,x 6 R such that M = zR + x R a n d v(f) = v(z;f). Let R ÷ R I be
a quadratic transform of R and fl a strict transform of f i n RI, and
assume fl 6 M(RI). Then o r d ( R 1 ) (fl) = n if a n d only if v(f) ~ 2, a n d
in this case we have z I = z/x 6 M(RI) and

v ( f I) = V ( Z l ; f I) : v(f) - I.

§5 E m b e d d e d resolution of plane curves and principalization

For this section we fix a sequence

R = R O ÷ RI ÷ R2 ÷ ... + R.I ÷ "'"

of two-dimensional regular local rings and elements fi 6 R i such that

R i is a q u a d r a t i c transform of Ri_1\
for all i > I.
fi i s a s t r i c t transform of fi-1 S

We put f = fo a n d assume, of c o u r s e , that 0 < o r d ( R ) (f) < ~.

By Proposition 3' and Proposition 4 we know:

A
If R is e x c e l l e n t , then there is some io and for |
each i ~ io there are z i 6 M(Ri)~M(Ri )2 a n d a (5.1)
nonnegative integer t i such that fiRi = (ziRi)ti.

If z,x is a r e g u l a r system of parameters of R such that M(R) R I = x R I,

then either z/x is a unit in R I o r x,z/x is a r e g u l a r system of parameters
in R I . U s i n g this, one shows by induction that

(f/fi) "Ri has normal crossings for all i ~ O. (5.2)

Using the notation of (5.1), choose i ~ io and extend z i to a r e g u l a r

system of parameters zi = z , x o f R i. F u r t h e r m o r e let u,v be a regular
system of parameters of R i such that

(f/fi)Ri = uavbRi (5.3)

for some integers a,b. Assume that M ( R i ) R i + I = u R i + I. B y (5.1) and

(5.3) we have
 22

f R. = ztuavbR, where t = t..

1 1 1

If M ( R i ) R i + I = z R i + 1 , then fRi+ I = ua+b+t(v/u)bRi+1 has normal

crossings. If M ( R i ) R i + I = x R i + 1 , then

f Ri+ I = (z/x) t x a + b + t (v/u) b R i + 1 , (5.4)

and this has normal crossings again, if z / x or v / u is a u n i t in R i + I.

So let us look at the case that z/x 6 M ( R i + I) and v/u 6 M(Ri+I) • Then
z/x, X is a r e g u l a r system of parameters of Ri+ I . We define

Wo(a) = o r d ( R i / z R i) (a + zR i) for a 6 Ri,

w I (b) = ord(Ri+i/(z/x)Ri+ I) (b + (z/x) R i + I) for b 6 Ri+ I .

Then

Wo(U) > 0 and Wo(V) = w I (v).

Therefore

w 1(u-(v/u)) : w 1(v) < Wo(U-V).

It follows that after finitely many steps we obtain an equation like

(5.4) in which either z/x or v/u is a u n i t . This proves
Proposition 5. A s s u m e that R is e x c e l l e n t . Given f # 0 in R a n d any
sequence

R = R O ÷ RI ÷ ... ÷ R.1 ÷

of successive quadratic transforms, there is a n integer i I such that

fR i h a s normal crossings for all i ~ iI.

Given nonzero elements gl,..o,gm in R a n d applying Proposition 5 to

f = g1'''gm we obtain the following
Corollary. Assume that R is e x c e l l e n t and

R = R O ÷ RI ÷ ... ÷ R.l ÷

is a s e q u e n c e Of successive quadratic transforms. GiveD any nonzero

elements g1''-',gm of R, there is a n integer i I such that the set
 23

{ g i R i , - - - , g m R i}

has normal crossings for all i ~ iI.

Proposition 6. A s s u m e t h a t R is e x c e l l e n t and that

R = R O ÷ R I ÷ ... ÷ R 1 ÷

is a s e q u e n c e of s u c c e s s i v e quadratic transforms. For any ideal I of R

there is an i n t e g e r i 2 such that IR i is p r i n c i p a l for i ~ i2.

For the p r o o f we make induction o n the n u m b e r of g e n e r a t o r s of I, so

we may assume I = f R + gR, f J 0 # g. B y the c o r o l l a r y to P r o p o s i t i o n 5
we can choose i and a regular system of p a r a m e t e r s z,x of R i s u c h that

fR i = x a z b R i , g = x C z d R i .

We put t i = (a-c) (b-d) and we observe that

I R i is p r i n c i p a l if a n d o n l y if t i ~ O.

Assume t i < 0 a n d M(Ri) "Ri+ I = x R i + I . If z/x is a u n i t in R i + I, t h e n

IRi+ I = xa+bR xC+dR

i+I + i+I'

which is p r i n c i p a l . If z/x 6 M(Ri+I), we have

fRi+ I = x a+b,[z/x)
, .b Ri+1, g R i + I = x c + d (z/x) dRi+1

and therefore

2
ti+ I = (a+b-c-d) (b-d) = (b-d) + t. > t.,
1 1

and this finishes the p r o o f .

 24

III R e s o l u t i o n and E m b e d d e d Resolution of E m b e d d e d Surfaces in

Characteristic 0

§I G e n e r a l assumptions and statements of the results

By a s u r f a c e we will m e a n a noetherian scheme F, of pure d i m e n s i o n 2,

such that
the local ring at any p o i n t of F is a factor of an
excellent regular local ring c o n t a i n i n g a field of ~ (1.1)
characteristic O.
Also in w h a t follows, any regular local ring will be a s s u m e d to be
excellent and to c o n t a i n a field of c h a r a c t e r i s t i c O. A surface F will
be c a l l e d locally embedded, if for any closed p o i n t P 6 F, the local ring
of F at P is the h o m o m o r p h i c image of a 3 - d i m e n s i o n a l regular local
ring. F w i l l be c a l l e d embedded, if it is (isomorphic to) a closed sub-
scheme of a regular (excellent) scheme of pure dimension 3. The p u r p o s e
of this chapter is to give the main computations for the proof of the
following two theorems:

T h e o r e m A. If F is a reduced, locally embedded surface, there is a

surface F~ and a m o r p h i s m ~:F~ ÷ F such that
a) F~ is regular.
b) ~ i s the c o m p o s i t i o n of q u a d r a t i c and m o n o i d a l transformations.

T h e o r e m B. If F is a reduced, embedded surface, closed subscheme of

the 3-dimensional, regular excellent scheme Z, there is a 3-dimensional,
regular (excellent) scheme Z~ and a m o r p h i s m ~:Z~ ÷ Z such that
a) ~ is the c o m p o s i t i o n of q u a d r a t i c and m o n o i d a l transformations.
b) The (iterated) strict transform F* of F is regular.
c) ~-I(F) has normal crossings.

(For the general algebraic definition of normal crossings see I,§I) .

In fact it turns out that the q u a d r a t i c and m o n o i d a l transformations
used for ~ are of a certain restricted type; e.g. for T h e o r e m A, the
center of such a t r a n s f o r m a t i o n is c o n t a i n e d in the singular locus of
the surface in question. Therefore F* contains an open set U such that
maps U i s o m o r p h i c a l l y to the set of regular points of F.

The proofs of both theorems consist m a i n l y of local computations~ the

global argument being reduced to a certain finiteness statement (see
 25

§5 a n d §9). In t h i s chapter we give the local computation in the c o m -

plete case. The results needed to r e d u c e the g e n e r a l case to the c o m p l e t e
one, together with other technical details, are the content of the n e x t
chapter. Reviews of the m e t h o d of p r o o f for T h e o r e m s A and B will be
given later on (see §6 a n d §10), but we will anticipate two p o i n t s of
these reviews in o r d e r to g i v e a small background for the following
algebraic setup.

To prove t h e o r e m A, i t is o b v i o u s l y sufficient to o b t a i n a morphism

~:F • + F with b) and such t h a t F • has the following property: If e(F)
denotes the highest multiplicity which F has in its (closed) points,
t h e n e ( F ~) < e(F) . F o r the p r o o f of t h e o r e m B l e t us look at any
morphism ~:Z~+ Z which is the c o m p o s i t i o n of quadratic and (suitable)
monoidal transformations. Then ~-I(F) = F ~ U F, w h e r e F ~ is the
(iterated) strict transform a n d F the e x c e p t i o n a l divisor. One trick of
the p r o o f is to d e c o m p o s e F as F = G U H, w h e r e G is the "old" exceptional
divisor obtained in t h o s e steps in w h i c h the m a x i m a l multiplicity of the
surface (= i t e r a t e d strict transform of F) w a s b i g g e r than e(F~), and
H is t h e "new" exceptional divisor of those last steps in w h i c h the
maximal multiplicity d i d n o t drop. Since our surface F in the t h e o r e m
is s u p p o s e d to b e (locallyl embedded, locally we have to c o n s i d e r a 3-
dimensional regular local ring R (subject to the a s s u m p t i o n s made
earlier) and an element f ~ R which defines F locally. We denote by n
the n u m b e r ord(R) (f), i.e. the m u l t i p l i c i t y of F a t the p o i n t represented
b y R, a n d w e assume n > I and f~ = fR.

The notation R, f, n w i l l be kept fixed for this chapter, and all the
assumptions made before will be in f o r c e .

§2 M a i n example: Strict transform for f = zn + xay b

The reason to l o o k at this example is t h a t the g e n e r a l proof consists

of a r e d u c t i o n to this example. Therefore this special case shows in
the m o s t purified way the idea, and the simplicity, of the g e n e r a l proof.
In the f o l l o w i n g description we use some facts, which will be proved
later on, although they are s t a n d a r d for those people who are familiar
with the s u b j e c t .

We will view f = z n + x a y b as an e l e m e n t of the p o w e r series ring

R = K[[x,y,z]], K a f i e l d of c h a r a c t e r i s t i c O. The assumption n =
 26

ord(R) (f) implies a+b > n. Let us assume n > I. Then the n-fold curves
are among { (z,x), (z,y)}, and (z,x) (resp. (z,y)) is n-fold if and only
if a > n (resp. b > n), see §4 for proof.

Let a ~ n and ~:R ÷ R 1 be a monoidal transform with center (z,x), such

that a strict transform fl of f satisfies ord(R1) (f 1) = n and
dim R I = 3. Then R I = K[[x,Y,Zl]], zI = z/x and

n a-n b
fl = zl + x y .

Now let ~:R~R I be a quadratic transformation such that dim R I = 3 and

ord(R1) (f I) = n, where again fl is a strict transform of f. Then

fxR I or
(x,y,z)R I = ~YR1 "

If (x,y,z)R I = x R I, then

RI = S1[[zl]]

where

zI = z/x
SI = the completion of a quadratic transform of K[[x,y]]

and

fl = z nI + x a + b ~ n . ty/x)
, .b

(y/x may be a unit in $I) . L e t us write

a = c-n + a , O i
< a < n
b = d-n + b , O < b < n
r (a,b) = a +
n

After performing t = c + d monoidal transformations with an n-fold

curve as center we will ohtain a strict transform

n ab
ft = zt + x y .
 27

If n o w rn(a,b) = a + b < n, ft w i l l define a point on a surface of

multiplicity < n. T h i s is the typical example of a g o o d p o i n t . If
rn(a,b) ~ n, l e t us a p p l y a quadratic transformation to f l i k e above.
Then (up to units)

fl = znI + xal Yl b I ' Yl = y/x

where

a I = a+b-n, b I = b, o r b I : O.

If b I = O, r n ( a l , b I) < n, a n d if b I = b w e get

r n ( a l , b I) = a + b - n + b < rn(a,b) .

So b y r e p e a t e d quadratic transformations we will reach the s t a g e that

rn(a,b) < n.

L e t us c a l l a monoidal transformation permissible for f, if the c e n t e r

is an n - f o l d curve on the s u r f a c e defined by f. T h e n w e h a v e obtained
the following result: Along any sequence of q u a d r a t i c transformations,
there is a n i t e r a t e d strict transform of f for w h i c h reduction of
multiplicity below n can be achieved by permissible monoidal transfor-
mations. The number of s u c h transformations may be zero. We note that
the c o n d i t i o n rn(a,b) < n is s t a b l e under permissible monoidal transfor-
mations, but it is n o t stable under quadratic transformations. As an
e x a m p l e , take f = z 3 + x 4 y 4, w h i c h a f t e r q u a d r a t i c transformation gives
3 5 4
z I + x Yl (z I = z/x, Yl = y/x) . T h i s m e a n s (in the terminology to be
defined in §3) that the p r o p e r t y of a p o i n t to b e g o o d is s t a b l e under
permissible monoidal transformations, but unstable under quadratic
transformations.

§3 G o o d points: Definition and elementary properties

We start with R, f, n as b e f o r e . Let us r e c a l l that for a prime ideal P

(in a n y ring), p(s) denotes the s t h - s y m b o l i c power of P. If P is
generated by a regular sequence, then it is w e l l known t h a t p(S) = pS.

Definition. E(f,R) = {P c R I P prime ideal of h e i g h t 2 such that

f 6 p(n)}.
 28

So l o c a l l y E(f,R) is the set of n - f o l d curves on the surface defined

by f = O.

Definition. f is p r e - g o o d in R if
a) E(f,R) # ~ and
b) E(f,R) has normal crossings.

Let f be p r e - g o o d in R a n d let ~ : R ÷ R I be a m o n o i d a l transformation

with center Po 6 E(f,R) (and d i m R I = 3), such that for a s t r i c t
transform fl of f we h a v e ord(R1) (f I) = n. T h e n [I], (3.1o) contains a
detailed description of E ( f I , R 1) (the p r o o f will be r e p r o d u c e d in the
next chapter). If Q 6 E ( f I , R I ) , then either Q = R I N PRp for some
P £ E(f,R), P # Po' or

Q = M(SI) O RI,

where

S I = a quadratic transform of RPo

M(S I) = the m a x i m a l ideal of S
I

and

ord(S1) (f I) = n.

Therefore E(fI,RI) has normal crossings again, but it m a y be empty.

Definition. A monoidal transformation with center P c R is c a l l e d

permissible for f if R/P is r e g u l a r and P 6 E(f,R).

N o w we define f to be g o o d in R, if the p r o p e r t y of b e i n g pre-good is

stable under permissible monoidal transformations. To be p r e c i s e ~

Definition. f is g o o d in R if the following conditions are satisfied:

a) f is p r e - g o o d in R.
b) L e t
R = R ° ~I RI ÷ ... +~t R t
b e any sequence of p e r m i s s i b l e monoidal transformations, let
fo = f and fi = s t r i c t transform of fi-1 (i=1,...,t), and a s s u m e
that ord(Ri)(fi) = n for i=I, .... t-1. Then
o r d ( R t) (ft) < n or ft is p r e - g o o d in R t.
 29

(~i is s u p p o s e d to be p e r m i s s i b l e for fi-1' of course.)

So a p r e - g o o d point which is n o t g o o d has the p r o p e r t y (by d e f i n i t i o n )

that b y p e r m i s s i b l e monoidal transformations it is t r a n s f o r m e d into an
n-fold point which does not lie o n any n - f o l d curve. (This w i l l be
called an i s o l a t e d n-fold point.)

By localization, the p e r m i s s i b l e monoidal transformations correspond

to q u a d r a t i c transforms in an n - f o l d point of a p l a n e curve. Therefore
good points are c u r v e - l i k e , in the sense that for r e d u c t i o n of m u l t i -
plicity they b e h a v e like singularities of p l a n e curves.

§4 S t r i c t transform: Power series case

We consider again the c a s e R = K[[x,y,z]]. Then, by a suitable choice

of x, y, z we m a y assume that

n
f = z n + l=
~ ai zn-i, ~i6 K[[x,y]] = S

and

ord(S) (ei) _> i, 2 _< i _< n.

In S t h e r e is an e l e m e n t 8, u n i q u e up to a unit, with the following

properties :

~il~ i for i=2 ..... n. (4.1)

If y 6 S and ~ i l ~ i for i=2 .... ,n, t h e n ~iB- (4.2)

We will use the n o t a t i o n

8 -- g ' c ' d ' { ~ i I/i I 2 < i < n}. (4.3)

Now let P 6 E(f,R), and let D be the p a r t i a l derivative with respect

to z. T h e n DP(s) c p(S-1) for all s > I (as for a n y d e r i v a t i v e of R),
and therefore

D(n-1)f = n!z 6 p(1) = p.

 30

It follows that z 6 P, and therefore P : zR + tR f o r some prime element

t of S. This implies p(n) = pn and therefore

tilei , 2 ~ i ~ n,

or

tf~

by (4.2). Therefore

E(f,R) = {(z,t) I t 6 S prime and tI~}. (4.4)

From this we see that

f pre-good in R ~ 8 is a n o n - u n i t having normal crossings, (4.5)

and we note that for f pre-good in R, E(f,R) can have at most two
elements.
Before giving a criterion for f to b e good in R, let us prepare f a
little. Let ~:R ÷ R 1 be a quadratic transformation and fl a strict
transform such that o r d ( R 1 ) (fl) = n (and d i m R I = 3). Then

R 1 = S1[[Zl]]

where

S I = completion of a quadratic transform of S

z I = z/x or z I = z/y.

(See n e x t chapter for t h e details),. This means that

n
= ~iZl
i=2

where

• ~ i
di = ~i/xl or ~i = ei/y ' 2 < i < n. (4.6)

Therefore after t quadratic transformations we get a ring R t and an

iterated strict transform ft o f f for w h i c h the following holds:
If o r d ( R t ) ~ f t) = n, then
 31

R t = St[[zt] ]
S t = power series ring in 2 v a r i a b l e s

and

n
ft = zt + ~-~ ~ zt-i
i=2

Applying resolution of p l a n e c u r v e s to the (completed) sequence of

quadratic transformations

S ÷ SI ÷ ... ÷ S t

we m a y a s s u m e that

{~2,...,~n} has n o r m a l crossings in S t (4.7)

and
the ideal in S t g e n e r a t e d by { ~ ! / i I 2 ! i ! n} (4.8)
is p r i n c i p a l .

N o w by (4.6) there is some y 6 S t such that

i
~i = Y ei' 2 < i _< n,

and t h e r e f o r e the c o r r e s p o n d i n g properties to (4.7) and (4.8) also h o l d

for e 2 , . . . , e n .
N o w let us a s s u m e f r o m the b e g i n n i n g that

{~2,...,~n} has n o r m a l crossings in S (4.9)

and
the ideal in S g e n e r a t e d by {~!/n I 2 ~ i ~ n} is (4.10)
principal.

Then (up to u n i t s and for a s u i t a b l e choice of x,y)

~i = xaiybi, 2 _< i _< n (4.11)

and by (4.10) there is some u 6 {2, .... n} such t h a t

~U : g ' c ' d ' { ~ I / i 1 2 ~ i ~ n} (4.12)

 32

and by the p r e v i o u s considerations, (4.11) and (4.12) are stable under

quadratic transformations as l o n g as the m u l t i p l i c i t y of t h e strict
transform of f is n.
But these conditions are also stable under permissible monoidal
transformations. To see this, let

~:R ÷ R I

be a permissible monoidal transformation with center P for f s u c h that

o r d ( R 1 ) (f I) = n, w h e r e fl is a s t r i c t transform of f. T h e n

RI = S[[Zl]]
= ~z/x if P = (z,x)
Zl ~z/y if P = (z,y)

and

n
fl = Zl + E d. z n-i
i=2 l l

where

~i/xl if P = (x,z)
2 < i < n.
~i = [ ~ a / y I
if P = (z,y)

(For the d e s c r i p t i o n o f R I, see the n e x t chapter again.) So w e see that

under the a s s u m p t i o n (4.11) and (4.12) the b e h a v i o u r of f u n d e r
quadratic and permissible monoidal transformations is e x a c t l y the same
as t h a t of

= zU + ~ u = zU + x a u y b u '

and also f is g o o d in R if a n d o n l y if f is g o o d in R. W e summarize the

result as follows: Let

R = RO ÷ R I + R 2 ÷ ... ÷ R z ÷ "'"

be an infinite sequence of 3 - d i m e n s i o n a l regular local rings such that

R i is the c o m p l e t i o n of a q u a d r a t i c transformation of Ri_ I a n d let fi
be a s t r i c t transform of fi-1 (i ~ I). A s s u m e t h a t ord(Ri) (f i) = n
for a l l i ~ I. T h e n ft is g o o d in R t for some t. F o r the p r o o f w e m a y
assume f to s a t i s f y (4.11) and (4.12), which reduces the a s s e r t i o n to
the c a s e f = f. B u t this case was treated in §2.
 33

We emphasize that the e x p l i c i t description of the b e h a v i o u r of f u n d e r

quadratic and permissible monoidal transformations given above is o n l y
valid under the a s s u m p t i o n that the m u l t i p l i c i t y remains n. N o s u c h
description is p o s s i b l e in a s t e p in w h i c h the m u l t i p l i c i t y drops.

§5 B a d p o i n t s are finite in n u m b e r

L e t us fix s o m e n o t a t i o n . F will denote a surface (subject to the

assumptions made in §I). By a point of F w e m e a n a closed point, and
in f a c t w e w i l l identify F with its s e t of c l o s e d points. Let
m(F) = max {multp(F) I p 6 F} a n d M(F) = {P 6 F I multp(F) = m(F)}.
T h e n M(F) is a c l o s e d subset of F, and if F is n o t regular, then
d i m M(F) < I. W e a s s u m e from now on t h a t F is n o t r e g u l a r (i.e.
m(F) > 1) .

Definition. A point P 6 F is c a l l e d g o o d resp. pre-good if e i t h e r

multp(F) < m(F) o r P £ M(F) and the local r i n g o f F a t P is o f the form
R/fR with a regular local r i n g R of d i m e n s i o n 3 and f good resp. pre-
good in R. (This c o n d i t i o n does not depend on the c h o i c e of R and f.)

Granting that normal crossing of M(F) at s o m e point P is s t a b l e under

permissible monoidal transformations, we have the following geometric
description of b a d (= n o n - g o o d ) points. A point P 6 M(F) is b a d if it
is e i t h e r an isolated point of M(F), or a s i n g u l a r i t y of a o n e - d i m e n s i o n a l
component, or a point of intersection of o n e - d i m e n s i o n a l components of
M(F) which do not have normal crossings a t P, o r finally a point that is
transformed into one o f these three types by permissible monoidal
transformations.

To prove the finiteness of b a d p o i n t s , we define a subset B(F) of M(F)

such that:

a) B(F) is f i n i t e .
b) A l l points of F-B(F) are pre-good.
c) W e o b t a i n a surface of F * such that either m(F*) < m(F)
or M ( F ~) = B ( F *) is f i n i t e by repeated application of (5.1)
the f o l l o w i n g procedure: First replace F b y F' = F-B(F),
and if F' N M(F) has a one-dimensional component C, let
F I ÷ F' b e a m o n o i d a l transformation of F' w i t h center
C. C o n t i n u e with F I.
 34

We define

B(F) = {P 6 M(F) I P is an i s o l a t e d point of M(F) or a s i n g u l a r

point of M(F) }.

(We n o t e that the points of i n t e r s e c t i o n of d i f f e r e n t one-dimensional

components of M(F) are singular points of M(F) .) N o w a) a n d b) are
clear. As noted earlier in §3, the e f f e c t of a permissible monoidal
transformation on a regular point o f M(F) is the s a m e as t h a t o f a
quadratic transformation on a p l a n e curve. The m e a n s that there is a
well defined number v such that after exactly v permissible transforma-
tions the c o r r e s p o n d i n g point will either have a multiplicity less than
m(F) o r it w i l l be an isolated m(F)-fold point. Applying this remark
to all o n e - d i m e n s i o n a l components of M(F) N F' w e obtain c) above.
Now let

F ~ = F t ÷ Ft_ I ÷ ... ÷ F I ÷ F O = F

be a sequence of s u r f a c e s obtained by the p r o c e d u r e of (5.1 , c), and

let Pi:Fi + F be the corresponding morphism. Then we have

If P 6 F ~ - B ( F ~) , t h e n m u l t p ( F * ) < re(F) . (5.2)

For any P £ F~-B(F*), pt(P) is a g o o d p o i n t of F. (5.3)

From this it is c l e a r that the b a d points of F are contained in the

finite set

t
U Pi (B (F i) ) ,
i=O

where Po = idF"

§6 R e v i e w of the p r o c e d u r e to r e s o l v e F

Starting with a surface F (as d e f i n e d in §I) , we apply repeated quadratic

transformations to F c e n t e r e d a t the finitely many bad points of F. B y
this process we obtain a surface F* such t h a t m(F*) < m(F) or e v e r y point
of F* is a g o o d p o i n t . If m(F*) = m(F), we apply monoidal transforma-
tions to F * centered at (l-dimensional) components of M(F*) . W e n o t e
that these centers are globally defined. Now the m e r e definition of a
 35

good point says that by repeating these monoidal transformations, we

will eventually obtain a surface F *~ s u c h that m(F ~) < m(F), and this
proves Theorem A.

§7 Embedded resolution - good triples

As indicated a t the e n d of §I, the local object of s t u d y for the p r o o f

of Theorem B consists of t r i p l e s T = (f,g,h) of e l e m e n t s of a r e g u l a r
local r i n g R, for which we make the f o l l o w i n g assumptions:

f / ~ = fR, f J 0
gh has normal crossings
(7.1)
f a n d gh h a v e no common prime factor
n = ord(R)(f) ~ I

We will fix t h i s triple T = (f,g,h) with (7.1) for the r e s t of t h i s

chapter.

T will be c a l l e d resolved if n = I a n d f-g-h has normal crossings. If

n = I, T is r e s o l v e d if a n d o n l y if e i t h e r g'h is a u n i t or t h e following
condition is s a t i s f i e d : If P c R is a p r i m e ideal of h e i g h t 2 such t h a t
f 6 P and g'h 6 P, then f-g'h has normal crossings in Rp a n d R / P is
regular.

Definition. If n > I, w e put E(f,g,h;R) = E(f,R), and if n = I w e put

E(f,g,h;R) = {P c R I P prime, ht(P) = 2, f 6 P, g h £ P a n d

either R/P not regular or fgh d o e s not have
normal crossings in Rp}.

This definition is m a d e u p in s u c h ~ w a y t h a t T is r e s o l v e d if a n d o n l y
if n = I and E(f,g,h;R) = @. The geometric meaning of E ( f , g , h ; R ) in the
case n = I is the following. The set of p o i n t s of the surface defined by
fghR in w h i c h this surface does not have normal crossings is a p r o p e r
closed subset, and E(f,g,h;R) is the set of o n e - d i m e n s i o n a l components
of this subset.

If P is a n y p r i m e ideal o f R, w e d e f i n e the incidence number i(g,P) to

be the n u m b e r of (essentially) different prime factors of g c o n t a i n e d in
P, and we put
 36

i~ = m a x {i(g,P) I P 6 E(f,g,h;R)}
and
E(f,g,h;R)~ = {P 6 E ( f , g , h ; R ) 1 i(g,P) = i*}.

This means that E(f,g,h;R)~ consists of t h o s e curves of E ( f , g , h ; R )

which have maximal incidence with g.

Definition. (f,g,h) is c a l l e d pre-good in R if

a) E ( f , g , h ; R ) # ~.
b) E(f,g,h;R) has normal crossings
c) Every P 6 E(f,g,h;R)* has normal crossings with gh (i.e. {ghR,P}
has normal crossings).

T will be called good if the p r o p e r t y of b e i n g pre-good is s t a b l e along

any sequence of p e r m i s s i b l e monoid&l transformations. To m a k e this
precise, we h a v e to d e f i n e permissibility and the transform of T.

Definition. A monoidal transformation ~:R ÷ R I w i t h center P will be

called permissible for (f,g,h) if P ~ E ( f , g , h ; R ) • and P has normal
crossings with gh.

Definition. Let ~:R ÷ R I be a q u a d r a t i c transformation or a p e r m i s s i b l e

monoidal transformation for (f,g,h) . T h e n the transform TI = (fl,gl,hl)
of T w i l l be d e f i n e d as follows:

fl = s t r i c t transform of f

Let g1' be a s t r i c t transform of g and h I' = f g h / f l g 1 ' . Then

gl = Ig I' if °rd(R1) (fl) = n

=i'hi' if o r d ( R I) (fl) < n

and

if ord(R1) (f I) = n
(7.2)
if ord(R1) (f I) < n

Note that in e a c h case we h a v e that f-g-h = fl-gl-hl. The normal

crossing condition in the d e f i n i t i o n of p e r m i s s i b i l i t y ensures that
glhl will have normal crossings in RI, so T I is in f a c t a tripel in the
sense of (7.1). N o w we c a n d e f i n e good tripels.
 37

Definition. (f,g,h) is c a l l e d good in R, if t h e following properties

hold:
a) (f,g,h) is p r e - g o o d in R.
b) Let
R = R O ~I RI ~ 2 . .. ~ t R t

be any sequence of monoidal transformations such that ~i is p e r -

missible for Ti_ I = (fi_1,gi_1,hi_1), where T o = T and T i is a
transform of Ti_ I . Assume that o r d ( R i ) (f i) = n for I ~ i ~ t-1.
Then o r d ( R t ) (f t) < n or T t is p r e - g o o d in R t.

Let us note that if ~ : R ÷ R I is a p e r m i s s i b l e monoidal transformation

and TI = (fl,gl,hl) a transform of T, and if (f,g,h) was pre-good in
R, then T I need not be pre-good in R, even if E ( f l , g l , h l ; R 1 ) # ~. The
reason is t h a t the third condition in the definition of pre-good may be
violated. As an example take

f = z 3 + x7y 7
g = z + y ( x 2 + y 3)

h = I

Then (z,y) is a p e r m i s s i b l e center, and for the corresponding transform

we have

3 7 4
fl = z + x y
gl z + x 2 + y3
=
4
h 1 = y

Now E(fl,gl,hl;R)~ = { ( z , x ) , (z,y)} (the m a x i m a l incidence being O), but

neither (z,x) nor (z,y) has normal crossings with glhl = zy 4 • y 4 ( x 2 + y 3 ) .

I t is typical in t h i s example that E(f,g,h;R) consists of two elements.

If it c o n s i s t s of one element only, then E(fl,gT,hl;R I) c o n s i s t s of at
most one element (namely the exceptional divisor), therefore in t h i s
case pre-good is stable as long as E ( f l , g l , h l ; R 1 ) # @. We will use this
remark i n §9 to p r o v e that there are only finitely many bad points.

§8 E m b e d d e d resolution - power series case

Let us look at the case R = K[[x,y,z]] = S[{ z]] , S = K [ [ x , y ] ] , a g a i n . By

a suitable choice o f x, y, z we may assume again that
 38

n
f = zn + E ~.z n-i @i 6 S, 2 < i < n (8.1)
i= 2 1 ' -- --

(note that f = z if n = I) and in addition

g = pl...pr, pj irreducible

and

pj = z + Bj, Bj 6 S, I <_ j _< r. (8.2)

Now by the definition of a transform, the process had started somewhere

with h = I. So b y the description of the transformations of R given in
§4 (under the assumption that o r d ( R ) (f) is unchanged), we may assume
that

h 6 S. (8.3)

Let

U {@~[/i, B~! I 2 ! i ! n, I ! J ! r}

and assume

U has normal crossings (8.4)

and
For any u,v 6 U, uS + vS is a principal ideal. (8.5)

We will show that under these assumptions (f,g,h) is good in R if (and

only if) f is good in R. As in §3 w e will first consider the case

f = z n + x a y b, a+b > n (n > I) I) (8.6)

where x,y is a regular system of parameters of S such that h and the

@~'s are monomials in x a n d y.
3

Now since g has normal crossings, (8.2) and (8.5) imply that there are
at most two prime factors pj of g, and at most one of ~I' @2 is in
M(S) 2 (M(S) = the maximal ideal of S) . T h e r e f o r e , up to symmetry in x
and y and up to units, only the following six cases are possible:

I) If n = I, (f,g,h) is good in R if and only if g = I.

 39

Case I • r < I and g = z + xX , ~ > O.

Case 2. r = I and g = z + x l y p, ~ > I, ~ > I.

Case 3. r = I and g = z.
Case 4. r = 2 and Pl = z + x
P2 = z + 8x, 8 a unit # 1.
Case 5. r = 2 and Pl = z + x
P2 = z + x l y ~, ~ _> I a n d l+U _> 2.
Case 6. r = 2 and Pl = z + x

P2 = z.

Note that cases 3 and 6 cannot occur if n = I. A s s u m e now that f is good

in R. Let us examine case I for X > I. B y (8.5) we see that
x ( n = 1 ) ! a y ( n - 1 ) !bs + xn!ls is principal, so a > n and either b = O or
a ~ In, and E(f,g,h;R)~ = { ( z , x ) }. L e t ~:R ÷ R I be the unique permissible
monoidal transformation and let (fl,gl,hl) be a transform of (f,g,h).
If ord(R1) (f I) < n there is nothing to prove, so assume ord(R1) (f I) = n.
Then

fl = (z/x)n + xa-nyb and gl : z/x + x ~-I

which belongs to case I again, and which satisfies the assumption (8.5).
Similar considerations show that in e a c h case
a) (f,g,h) is pre-good in R,
b) a permissible monoidal transformation for which ord(R1) (f I) = n
does not change the assumptions (8.4) and (8.5) and it leads to
one of the cases I-6 again.
(It m a y be instructive for the reader to check all these cases.) So
eventually repeated permissible monoidal transformations will lead to
a triple ( f ~ , g ~ , h ~) in R ~ for which

]OrdR~ f~ < n if n > I

\f~g~h • has normal crossings if n = I.

We can summarize the result of these considerations in the following

way:
Let

R = R° + RI + ... ÷ Rt ÷ -..
be a sequence of quadratic transformations and let Ti = ( f i , g i , h i) be a
transform of T i _ I, w h e r e T O = T. Assume that ord(Ri) (f i) = n for all
i > I. Under the assumptions (8.4), (8.5) and (8.6) there is a t such

that T t is good in R t.
 40

Finally we remark that this result is true in g e n e r a l , since assumptions

(8.4) and (8.5) will be satisfied in R i for large i, a n d the g e n e r a l
case (8.1) c a n be r e d u c e d to (8.6) as in §4.

§9 B a d p o i n t s for e m b e d d e d resolution are finite in n u m b e r

Let Z be a regular, 3-dimensional, excellent scheme and let (F,G,H) be

a triple of s u r f a c e s on Z such that

F and G U H have no common components,

G U H has normal crossings, (9 .I)
F is r e d u c e d .

Let P be a point o n F, let R be the local r i n g of Z a t P a n d let F, G, H

be defined in R b y f, g, h r e s p e c t i v e l y .
Definition. P will be called a good (pre-good) point for (F,G,H) if
multp(F) < m(F) o r the tripel (f,g,h) is g o o d (pre-good) in R.

Now let C be a curve o n F.

Definition. C will be called permissible for (F,G,H) if
a) C is r e g u l a r ;
b) C c M(F)
c) among all curves on F s a t i s f y i n g a) a n d b), C is c o n t a i n e d in the
maximal possible number of c o m p o n e n t s of G;
d) C h a s normal crossings with G U H.

A monoidal transformation ~:Z I ÷ Z w i t h center C will be called per-

missible for (F,G,H) if C is p e r m i s s i b l e for (F,G,H) . N o t e that a per-
missible curve is p e r m i s s i b l e at each of its p o i n t s in the s e n s e of the
local definition o f §7.

Let ~:Z I + Z b e a q u a d r a t i c transformation centered at a point P 6 M(F)

or a permissible monoidal transformation for (F,G,H) . L e t F I (resp. G I)
be a strict transform of F (resp. G) under • and let H I = ~-I(H U C),
where C is the c e n t e r of ~. T h e n the transform of (F,G,H) is d e f i n e d to
be

(FI,GI,HI) if m ( F I) = m(F)
(FI,G I U HI,~) if m ( F I) < re(F) .
 41

As noted ~ n the l o c a l part, the transform of (F,G,H) will again be a

triple satisfying (9.1). Note that in b o t h cases ~-I(F U G U H) =
F I U G I U H I. T o p r o v e the f i n i t e n e s s of bad points for (F,G,H) we
proceed as in §5.
We define

] M(F) if re(F) > I

M(F,G,H)
1 {P 6 F I F U G U H does not have normal crossings
a t P} if m(F) = I

and we define B(F,G,H) to c o n s i s t of the i s o l a t e d points, the singular-

ities of M(F,G,H), and those regular points of M(F,G,H), at which G U H
does not have normal crossings w i t h M ( F , G , H ) . (Note t h a t M ( F , G , H ) is a
proper closed subset of F also in the case m(F) = I). If w e can show
that B(F,G,H) has properties analogous to (5.1), then the a r g u m e n t used
in §5 w i l l show that there are only finitely many points which are bad
(= n o t good) for (F,G,H) . a) a n d b) are clear, and so is c) in c a s e
m(F) > I. T o o b t a i n c) in case m(F) = I we use again the fact that per-
missible monoidal transformations correspond to q u a d r a t i c Zransformations
of p l a n e curves. Then the translation of c) into local algebra simply
means that along a quadratic sequence of 2-dimensional regular local
(excellent) rings any finite s e t of e l e m e n t s will have normal crossings
eventually.

§10 Review of the p r o c e d u r e of e m b e d d e d resolutions

TO describe an intermediate s t e p of e m b e d d e d resolution, it is c o n v e n i e n t

to s t a r t w i t h a pair (F,G) of s u r f a c e s rather than F only. Using the
notation of §9 w e assume F,G c Z, F a n d G h a v e no c o m m o n components and
G has normal crossings. Therefore (F,G,~) is a t r i p l e in the s e n s e of
(9.1). According to §9 there are o n l y finitely many bad points for
(F,G,@), and by §8 they can be removed by finitely many quadratic trans-
formations. If m(F) > I, a f i n i t e number of p e r m i s s i b l e monoidal trans-
formations will transform (F,G,H) into a tripl~ (F*,G*,@) with m(F*)
< m(F) . If m(F) = I, the same procedure will lead to a t r i p l e (F*,G*,H*)
such that F ~ U G* U H* has normal crossings, and this p r o v e s theorem B.
 42

IV ~uxiliary results for surfaces

In t h i s chapter, R denotes a three-dimensional, excellent, regular local

ring with maximal M and residue field K. W e assume that K has character-
istic O. (This assumption w i l l only be used in §3 and §4.) Furthermore w e fix
a nonzero element f in R s u c h that o r d ( R ) (f) = n > O.

§I E q u i m u l t i p l e curves under quadratic transformations

Let R ÷ R I be a quadratic transform of R, let fl a s t r i c t transform of

f in R I a n d assume o r d ( R 1 ) (fl) = n°

Let Po 6 E ( f I , R I) a n d assume first that Po N R = M. By the correspondence

between RI/MR I and gr(R) described in I, §3, Po corresponds to a h o m o -
geneous prime ideal Q of gr(R) , n e c e s s a r i l y of height one, such that
in(f) E Q(n). As observed earlier, these conditions imply that Q is
generated by in(z) for some z 6 M ~ M 2, a n d in(f) : in(z) n u p to a u n i t .
It follows that Po = (x,z/x), where x 6 R is chosen so t h a t MR I = xR I .
Furthermore Po is u n i q u e and R I / P o is regular.

If PI 6 E ( f I , R I) a n d PI N R ~ M, then

(RI)PI = Rp where P : PI n R.

Since f1-Rp = f-Rp, P 6 E(f,R) . So we obtain

,{PRp N R I 1P E E(f,R) and Rp D RI] or

E(fI,R I)
[{PRp R R I 1P 6 E(f,R) and Rp m R I} U {Po }.

Proposition I. L e t R ÷ RI, b e a quadratic transform and let fl b e a

strict transform of f in R I. A s s u m e that o r d ( R 1 ) (f I) = n a n d that
E(f,R) has normal crossings. Then E ( f I , R I) h a s normal crossings.

Proof. Let x,y,z be a regular system of p a r a m e t e r s of R w h o s e subsets

generate the elements of E(f,R) , and assume M R I = x R I . If PI 6 E(fI,RI) ,
then either PI n R = M o r PI N R = (y,z) , in w h i c h case we must have
V z
PI = (~,~) . T h e r e f o r e if E ( f I , R I) h a s only one element, then there is
nothing to p r o v e .
 43

Assume therefore that E(fI,RI) = {Po,PI}, where P o N R = M, PI =

(y/x)R1+(z/x)R I. S i n c e x, y/x, z/x is a r e g u l a r system of parameters of
R I and R I / P o is r e g u l a r , we have Po = xR1 + (a(y/x) + b(z/x))R1, where
a, b 6 R I a n d either a or b is a u n i t i n R I . If e . g . a = I, then we
write PI = ((y/x) / b(z/x))R I + ( z / x ) R I to s e e that E ( f I , R I) h a s normal
crossings.

§2 E q u i m u l t i p l e curves under monoidal transformations

Let R ÷ R I be a monoidal transform with center P 6 E(f,R) (remember

that monoidal includes the assumption that R/P is regular). Let fl b e
a strict transform of f in R I a n d assume o r d ( R 1 ) (f I) = n. B y the con-
siderations m a d e in I , § 4 , we can find a regular system of parameters x,
y, z of R w i t h the following properties:

P = x R + yR, PR 1 = xR I
x, y/x, z is a r e g u l a r system of parameters of R l j (2.1)
f - a y n 6 pn+1 + zpn for some unit a in R.

Assume that PI is a p r i m e ideal in R I s u c h that fl 6 PI and PI n R m P.

If PI ~ R = M, we must have PI = xR1 + zR1" From (2.1) we see that
fl-a(y/x)n 6 PI, which leads to the contradiction y/x 6 PI" Therefore
PI N R = P.

Applying this remark to a n element PI C E(fI,RI) such that PI N R D P

we see that (RI)P I dominates Rp, and actually (R1)p1 is the unique
quadratic transform of Rp such thatord((R1)P1) (f I) = n. O n the other
hand, if PI 6 E(fI,RI) such that Pl ~ R ~ P, then we conclude as in t h e
preceding section that PI n R 6 E(f,R) .

Proposition 2. L e t R ÷ R I be a monoidal transform of R with center

P 6 E(f,R), fl a s t r i c t transform of f in R I. A s s u m e that o r d ( R 1 ) (f I) =
n and that E(f,R) has normal crossings. Then E ( f I , R I) h a s normal
crossings.

Proof. Choose a regular system of p a r a m e t e r s x, y, z whose subsets

generate the elements of E ( f , R ) , and assume P = xR + yR and PR I = xR I .
If E ( f I , R I) = {PI} and x 6 PI' there is n o t h i n g to p r o v e . If
P2 6 E(fI,RI) such that x £ P2' then P2 N R = yR + zR a n d consequently
P2 = (Y/x)RI + zR1' so w e are done if E ( f I , R I) c o n s i s t s of o n e element
 44

only. Therefore we assume now t h a t E ( f I , R I) = { P I , P 2 } a n d x 6 PI,

P2 = (Y/X) RI + zR1" Since x, y/x, z is a r e g u l a r system of p a r a m e t e r s
o f RI, and since R I / P I is r e g u l a r , there are elements b,c 6 R I such
t h a t PI is g e n e r a t e d b y x a n d b(y/x) + cz, and either b o r c is a u n i t
in R I . As in P r o p o s i t i o n I we see that {PI,P2} has normal crossings.

§3 E q u i m u l t i p l e curves under completion

Let f 6 R be a nonzero element such t h a t ord(R) (f) = n > O a n d a s s u m e

that fR is n o t a n n th p o w e r of a n y p r i n c i p a l ideal of R.

Lemma I. If P~ 6 E ( f , R ~) t h e n P* N R 6 E(f,R).
Proof. Let P = P~ D R. W r i t e

nI ns
fR = Pl "'" "Ps R,

where Pl,--',Ps are the d i s t i n c t prime factors of f in R. T h e n f 6 P

implies Pi 6 P for some i, s a y Pl 6 P. L e t us show first t h a t ht(P) = 2.
If n o t then P = pTR and we conclude that

nI
f(R~)P~ = Pl (R~)P *"

Since R is e x c e l l e n t , R~p~/PiR*p, is r e g u l a r , necessarily of dimension

one, so o r d ( R ~ p ~ ) (pl) = I a n d we o b t a i n

n I = n, o r d ( R ) ( p l ) = I

n
so f R = pl R. Since this was excluded by assumption, we must have
ht(P) = 2. In p a r t i c u l a r P ~ is m i n i m a l among the p r i m e s o f R* c o n t a i n i n g
P R ~. U s i n g excellence again we see that R~p~/PR~p~ is r e g u l a r of d i m e n s i o n
zero and therefore PR~p~ = P~R~p~. Since Rp ÷ R ~ p ~ is f a i t h f u l l y flat,
we have

M(R~p~) m N Rp = pmR~p~ N Rp = p m R for all m

P

so t h a t f 6 pnRp; i.e. P 6 E(f,R) .

Using the n o t a t i o n of the p r o o f above, we note that if R/P is r e g u l a r ,

then R~/p~ is r e g u l a r a n d P ~ the u n i q u e element of E(f,R*) which contains
P. F r o m this remark and Lemma I we obtain immediately
 45

Lemma 2. If E(f,R) has normal crossings, t h e n E ( f , R ~) h a s normal

crossings. Therefore f pre-good in R i m p l i e s f pre-good in R*.

The converse of L e m m a 2 need not be true. As an example, take

R = K[x,y,Z](x,y,z ) and

f = z2 + [x 2 _ y2 + y 3 1 2

Then R ~ = K[[x,y,z]], and since x2-y 2 = (x+y) (x-y), x2-y2+y 3 can be

factored in R ~ b y H e n s e l ' s Lemma:

2 2 3
x - y + y = g1-(x-y)-g2- (x+y), g l , g 2 £ R*.

Therefore E ( f , R ~) = { (z,x-y), (z,x+y)} b u t E(f,R) = { ( z , x 2 - y 2 + y 3 ) }, so

E ( f , R ~) h a s normal crossings, b u t E(f,R) does not.

L e t P 6 E(f,R) and l e t P~ b e a m i n i m a l prime ideal of R ~ containing P R ~.

Then certainly P~ 6 E(f,R~). If n o w R ~ / P ~ is r e g u l a r but R/P is not, then
there must be another prime ideal Q~ of R ~ s u c h t h a t Q~ N R = P. (Note
t h a t P R • is an i n t e r s e c t i o n of p r i m e ideals since R is e x c e l l e n t . ) This
shows that the e x a m p l e given above is t y p i c a l in the following sense:
If E ( f , R ~) h a s normal crossings a n d E(f,R) does not, then there are
P~,Q~ 6 E ( f , R ~) s u c h t h a t P~ # Q* a n d P~ N R = Q~ n R. In p a r t i c u l a r
R/P~NR is n o t r e g u l a r .

Therefore ~ f E(f,R) has normal crossings, then the elements o f E(f,R)

a n d E ( f , R ~) c o r r e s p o n d uniquely to e a c h o t h e r by extension an4 con-
traction. Taking into account that any monoidal transformation of R
induces uniquely a monoidal transformation of R ~, w e h a v e

Lemma 3. If E(f,R) has normal crossings, then f is g o o d in R if a n d o n l y

if f is g o o d in R ~.

In II, §4, w e h a v e shown the following

Proposition 3 ~. L e t

R * = ~o~ ÷ R~+ "" . ÷ R *i + "."

be a sequence of t h r e e - d i m e n s i o n a l regular local rings such that each

R ~ is the c o m p l e t i o n of a quadratic transform of Ri_ I . L e t f = fo 6 R~
such that f # 0 a n d o r d ( R ~ ) (f) = n > O. Define fi 6 R *i i n d u c t i v e l y to b e
 46

a strict transform of fi-1, and assume t h a t o r d ( R ~ ) (f i) = n for all i.

Then for s o m e j, fj is g o o d in Rj.

We can use this now to p r o v e

Proposition 3. L e t

R = RO ÷ R I ÷ ... ÷ R i ÷ ...

be a sequence of e x c e l l e n t three-dimensional regular local rings such

t h a t e a c h R i is a q u a d r a t i c transform of R i _ I. L e t f = fo ~ R s u c h that
f ~ O a n d ord(R) (f) = n > O. Define fi 6 R i i n d u c t i v e l y to b e a s t r i c t
transform of fi-1, and assume t h a t ord(Ri) (f i) = n for all i. T h e n for
some j, fj is g o o d in Rj.

Proof. By Proposition 3 ~ we may assume f good in R ~. T h e n b y L e m m a 3,

if f is n o t good in R, E(f,R) does not have normal crossings. Now
E ( f I , R ~) h a s normal crossings by Proposition I, and it c o n t a i n s at most
one e l e m e n t PI such t h a t P~ ~ M ( R ~) and at most one element P2 such that
P2 m M(R~) • It f o l l o w s t h a t E ( f I , R I) h a s normal crossings, and by
Proposition I, E ( f i , R i) h a s n o r m a l c r o s s i n g s for all i ~ I. O f c o u r s e
+
fl m a y n o t be g o o d in R I (as s h o w n b y the e x a m p l e in I, §2). B u t b y
Proposition 3 ~ r fj is g o o d in Rt] for some j _> I. It f o l l o w s from Lemma 3
that fj is g o o d in Rj.

Let (f,g,h) be a triple in R in the sense o f II, (7.1). It is c l e a r that

corresponding to L e m m a 3 we have

Lemma 4. If E(f,R) has normal crossings, then (f,g,h) is g o o d in R if

and only if (f,g,h) is g o o d in R ~.

Therefore by the same argument as in P r o p o s i t i o n 3, the r e s u l t of II,

§8, yields

Proposition 4. L e t

R = Ro + R I ÷ R 2 + . .. + R1 ÷

be a s e q u e n c e of e x c e l l e n t three-dimensional regular local rings such

that each R i is a q u a d r a t i c transform of Ri_ I. L e t (f,g,h) = (fo,go,ho)
be a triple in R a n d d e f i n e (fi,gi,hi) inductively to be the triple of
 47

R i which is a t r a n s f o r m of (fi_],gi_1,hi_1) . Assume t h a t o r d ( R i ) (f i) = n

for all i ~ O. Then for s o m e j, (fj,gj,hj) is g o o d in Rj.

§4 U s e of the T s c h i r n h a u s e n transformation

In this section we t a k e R to b e the p o w e r series ring K[[x,y,z]] over a

f i e l d K of c h a r a c t e r i s t i c zero. Let f 6 R, f # O, a n d n = ord(R) (f) . B y
the W e i e r s t r a B Preparation Theorem and related results we may choose
x,y,z in s u c h a way that the ideal generated b y f is the same as the
ideal generated by zn + ~ I aizi-n' where a i 6 K[[x,y]]. The Tschirnhausen
transformation consists in r e p l a c i n g z by z* = z-al/n, so f b e c o m e s a
polynomial in z~ in w h i c h the c o e f f i c i e n t of (z*) n-1 is zero. So f r o m
now on we assume

n
f = zn + ~ a.z n - i 6 s = K[[x,y]], n > O.
l ' ai
i=2

Let R ÷ R I be a quadratic transform and fl a s t r i c t transform of f in

RI, and assume ord(R1) (fl) > O. W e claim that

= fxR I or
M(R) RI ~YR1.

Assume the contrary. Then x/z and y/z are in M ( R 1) a n d therefore

n
fl : f/z n = I + ~ ai/zl
i=2

is a u n i t in R]. (Note t h a t ord(R) (f) = n i m p l i e s ord(S) (a i) ~ i and

therefore ai/zi 6 M(RI). Note also that this argument is v a l i d w i t h o u t
assuming a I = O.)

Assume in a d d i t i o n t h a t ord(R1) (fl) = n, and let M(R) R I = x R I. L e t Q b e

the h o m o g e n e o u s prime ideal of gr(R) such t h a t R I / X R I = gr(R) (Q) . T h e n
in(f) 6 Q(n) . N o w

n
in(f) = in(z) n + ~ a i in(z) n-i
i:2

where ai is the c l a s s of a i m o d M(R) i+I. Using derivatives as in II, §4,

we see that in(z) 6 Q, w h i c h means that z/x 6 M(RI) and R I has as r e g u l a r
system of p a r a m e t e r s x, z/x, y* : F(y/x), where F(T) 6 S (see I, §4). It
follows that y* defines a quadratic transform S I of S a n d
 48

~ = s~[[z/x]]

We turn now to a m o n o i d a l transform R + R I with center P 6 E(f,R). As

usual, fl d e n o t e s a strict transform of f in R I. B y the o b s e r v a t i o n made
in II, §4, w e h a v e

P = zR + tR, t 6 S and tilai for all i.

Since ord(S) (t) = I, w e m a y assume t = x. A s in the c a s e of a quadratic

transform one sees t h a t o r d ( R 1 ) (f I) > 0 i m p l i e s P R I = x R I . If w e assume
o r d ( R 1 ) (f I) = n, then we have

RI/XRI+YR I = (grp(R) ® RK) (Q)

for s o m e h o m o g e n e o u s prime ideal Q, and the r e s i d u e class of inp(f)

belongs to Q(n) = Qn. A s above we conclude that z/x 6 M(RI), and
x, y, z/x is a r e g u l a r system of p a r a m e t e r s of R I. It f o l l o w s that

R I = S[ [z/x] ].
 49

References

[I] A b h y a n k a r S~. (1966) R e s o l u t i o n of s i n g u l a r i t i e s of e m b e d d e d algebraic

surfaces. A c a d e m i c Press, New York L o n d o n
[2] H i r o n a k a H. (1964) R e s o l u t i o n of s i n g u l a r i t i e s of an algebraic
variety over a field of c h a r a c t e r i s t i c zero I, II. Annals of
Mathematics 79:109-326
[3] L i p m a n J. (1975) I n t r o d u c t i o n to r e s o l u t i o n of singularities, in
Algebraic Geometry Arcata 1974. Amer. Math. Soc. Proc. Symp. Pure Math.
29:187-230
[4] M a t s u m u r a H. (1970) C o m m u t a t i v e algebra. Benjamin, New York
[5] Serre J-~ (1965) A l g ~ b r e L o c a l e - M u l t i p l i c i t ~ s . S p r i n g e r Lecture
Notes in M a t h e m a t i c s 11, Berlin H e i d e l b e r g New York
[6] Zariski 0. (1967) E x c e p t i o n a l s i n g u l a r i t i e s of an a l g e b r o i d surface
and their reduction, Accad. Naz. Lincei Rend. CI. Sci. Fis.Mat~Natur.
Serie VIII, 43:135-146
[7] Zariski O.(1978) A new proof of the total e m b e d d e d r e s o l u t i o n theorem
for algebraic surfaces (based on the theory of q u a s i - o r d i n a r y
singularities). Amer. J. Math. 1OO:411-442
[8] A b h y a n k a r &S. (1982) W e i g h t e d E x p a n s i o n s for C a n o n i c a l D e s i n g u l a r -
ization. S p r i n g e r Lecture Notes in M a t h e m a t i c s voi. 910
 DESINGULARIZATION IN LOW DIMENSION

Jean GIRAUD

§ I. Smart and coarse theorems.

There are two problems, namely desingularlzation and simplification

of a boundary and for each of them two levels of precision in the

corresponding theorems wich I will call coarse and smart.

Coarse desin~n~larization theorem. Let X be a scheme (or a complex analytic

space) . Assume that X is reduced. There exists a proper morphism

m
t -I
~: X " X , such that the induced morphism w (Xreg) rXreg is

an isomorphism, where Xreg is the set of points where the local ring
!
0X, x is regular, and X is regular.

This is a theorem in the complex analytic case or for X excellent

of characteristic zero or for X excellent and dim X ~ 2 .

Coarse Simplification of boundary. Let Z be a regular scheme (or smooth

complex analytic space), let Y be a closed subset of Z • There exiats

!
a proper and birational morphism w: Z , Z such that
!
(I) Z is regular

(2) -l(y) is a normal crossing divisor,

(for short we say that -1(y) is a d.n.c.) and each irreducible of

-1(y) is regular.

Of course if Y is regular we achieve that by blowing up Y •

There are various smart versions of these theorems and it is

not my intention to discuss them. I will only give one of them•

 52

Smart desin~ularization theorem. Let X be an excellent scheme of

characteristic zero.

There exists a sequence

x - Xo: -x I ------ x 2 ...... xN_ I •

v,
such that

(i) Yi+1 is regular, closed in X i , and contained in the

singular locus of Xi , 0 ~ i ~ N - 1 •

(ii) Xi+ I is the blowing up of Xi with center Yi+1

(iii) XN is regular

(iV) if Ei is the inverse image of Yi in ~ , I ~ i < N,

then %) Ei is a divisor with normal crossings.

1~i~N

Here the important condition is (i) , because if you start

with some embedding of X as a closed subscheme of a regular scheme Z ,

then by letting Zo = Z , and Zi+1= blowing up of Z i with center Yi+1

you get that each Xi is closed in a regular Z i . In other words if

X i is embedded in a regular Z , then the desingularization XN is

embedded in a regular ZN .

R~mark 1.If X is embedded as a hypersurface in a regular Z , then

the coarse simplification of boundary will produce 7: Z • Z such

that ~-I(x) is a d.n,c . Nence, if X is reduced and irreducible we

will get a birationnal desingularization Xt, • X , by taking for X'

the component of ~-I(X) whose image is a divisor in Z . This will

only be a coarse desingularization of X , unless we have a smart

simplification of boundary, by wich I mean some condition implying

condition (i) of the smart desingularization theorem.

Remark 2. On the other hand if X is a curve in ~ 3 = Z , then

simplification of boundary will not give a desingularization of X

since the strict transform of X is goin~ to be empty because one will

 53

have to blow up X if one wants ~-1(X) to be a Cartier divisor: this

is the universal property of blowing up.

Hence desingularlzation and simplification of boundary are

not at all the same problem. They are closely related in the hypersurface

case and in that case some people call it "embedded resolution". In the

general case desingularization and simplification are proved simultaneously

each one being useful for the proof of the other one but not in such a

naive ws~v as is suggested by the hypersurface case.

To end this paragraph I would like to point out that to prove

one of the three statements in dimension N then you need as an induction

hypothesis a very smart statement in dimension N-I . This is not exactly

true in low dimension, for instance use will see a nice proof of coarse

desingularization of X , if dim(X) = 2 due to J.Lipman. And also

simplification of a boundary Y in Z with dim(Z)=2 , is automatically

smart since one only has to blow up points. In some sense the case of

surfaces is somewhat misleading and specially the surface-hypersurface

case ( See § 4 and 5 ).

§ 2. What can be achieved by blowing up closed points.

If dim Z = I and Z is regular , then any closed subscheme is

a divisor with normal crossing: nothing to prove.

If dim × t I , and X excellent and reduc,,d, then X • is a

sln~
discrete set of closed points; by blowing them up one gets XI •X

which is different from X because for ~ 6 Xsing~mx, ~ is not

principal, hence the blowing up of ~ gives X' ~ X . Repeating this

operation one must reach the normalization of X since this normalization

is finite over X . Observe that for curves there is a unique coarse

desingularization, namely the normalization,which is also, as we just

saw, a smart desingularization.

 54

If dim Z = 2 and Z is regular, if J is a coherent sheaf of

ideals of Z , then the set of ~ e Z such that J Oz, ~ is not a

d.n.c, is a discrete set of closed points; call it F . Blowing up

of F in Z gives ZI which is regular and we get a new exceptionnal

set FI for J1 " JOz I , and so on. One can prove that by repeating

this process one gets Fn- ~ . In other words simplification of bou:ndarM

is automatic in dimension 2, and is achieved by repeated blowing up of

closed points.

Some isolated singularities of surfaces can be solved by

(repeated) blowing up of closed points. The most obvious example is

the cone over a projective smooth curve C since blowing up the vertex

of the cone gives the line bundle on C given by the inclusion C C N

Less obvious is the case of a normal surface X which admits a finite

and flat projection over a regular surface, such that the discriminant

has normal crossings, (car.o) . This is the basis of JUNG'S method.

Amother example is given by normal surfaces with rational singularities:

this is the basis of LIPMAN'S proof of coarse desingularization theorem

for excellent schemes of dimension two.

But it is easy to give an example of a normal surface such

that the blowing up of the closed point gives a surface with non isolated

singularity: z 3- x5+y 5 . The blowing up is covered by two pieces. The

first one has coordinates z'-z/x,x,y'=y/x and equation z'3=x,2(1+y,5)

and the second one is obtained by exchanging x and y .

§ 3. Coarse desin~ularization of surfaces.

Since we just saw that blowing up of points will not desingularize

surfaces, the next idea is to perform sucessively blowing up of closed

points (the singular locus) and normalization.Due to ZARISKI'S and

ABHYANKAR'S work we know that this procedure works but the simplest proof

has been given recently by LIPMAN. Let Bs state his result.

 55

Theorem Let X be an excellent scheme of dimension two. Assume that

is normal and quasi compact. Consider a sequence

X = Xo ~ X I' X 2 ..... X n" X n +.....

1
U U u U U

FI F2 F3 F Fn+2
n+1

Such that

(i) Fi+ I is a. discrete subset of Sing(X i) , i ~ o

(ii) X is the normalization of the blowing up of X. with

i+I 1
center Fi+ I , i ~ o
Then

(A) there exists an integer N such that for n ~N the inclusion

HI(XN,OXN ) C HI(Xn,Ox ) is bijective

n

(B) if Fi+1 = Sing (X i) for any i , then there exists an

such that FN+lis empty,hence ~ is regular.

The largest part of LIPMAN'S paper is devoted to the proof of

(A). This is due to the fact that he wants to give a self contained account

of local duality. If you accept it you know that there is an injective

map fn*(~Xn) ~Xo , where fn: Xn OXo is the projection and w

is the dualizing sheaf. Grauert-Riemenscheider vanishing theorem says

that • Ifn,(~ Xn) = 0 provided that X admits a desingularization,

but since he does not know yet the existence of a desingularization,

LIPMAN has to prove this vanishing theorem. When this is done

local d u a l i t y t e l l s you t h g t t h e l e n g t h o f HI(Xn,O X )/HI(Xo,O x ) is

n o
1;he length of ~JX /f* (~X) " Now one needs some kind of discriminant
o n
or trace map to show that there exist a d e ~x(X) (Say X is affine)

such that d~ x C fn* (~X ) for any n .

o n

The next step is to see that one only has to study points of X which
n
 56

are on the strict transform of the curve d = o and for these points

Ox x is regular for n big enough: this is a classical easy remark.

n t

This is essentially the argument for (A),see LIPMAN'S paper

for details!

Now to prove (B) it is enough to show that for a sequence of

normalized blowing up as in (B), such that H~(Xo,Ox ) = H~(Xn,Ox ) for

o n
all n ~O,X n is regular for n big enough.First there is an older lemma

of LIPMANsaying that in that case the blowing up remains normal: in

other words you don't have to normalize. Second the singularity has very

special features (all extracted from the cohomological hypothesis) wich

make the proof rather easy, at least for somebody who has some familiarity

with the manipulation of infinitely near points.

I apologize for all the imprecislons contained in this short

description of a rather long and very elegant paper .

The oldest idea for proving the coarse desingularization theorem

is due to JUNG and is 100 years old. It is still as fresh and beautiful

as on the first day: it only works in tl~rac~eristic 0 .

Say that X is a projective surface and look for a finite

morphisme w:X P S - ~2 given by some ~eneric linear projection.

Look for the discr~mlnant A of that projection: if X is not a

hypersurfaee one takes for A a suitable Fitting ~deal Of the relative

differentials ~Ix/S . This Fitting ideal is equal to the usual discriminant

in the hypersurface case; see for instance TEISSIER , Ai~cata 1975.

AS we saw in § 2, a suitable sequence of blowing up of closed points

in S will give p:S' ,S such that A'= p-I(A) is a divisor with

normal crossings. Let X' be the pull back of X by p and let X' be

its normalization. The claim is that X' can be desingularized by repeated

blowing up of closed points (one does not have to normalize again).

Before proving this we observe that if it is true we get

X'm.----~ ' ~ X , which is birationzl , with X" regular,

 57

but we may have modified X above some points ~ e X such that

0X, ~ is regular but A ~s not normal crossings at ~(~) E S . In

other words we cannot say that X" is a desingularization of X ,

because X" ,X is not an isomorphism above X (see the first

reg
statement page I). This is not difficult to arrange. In fact take an

open neighbourhood U of ~ in X such that U n w-1(Sing A) =

Such a U exists since Sing (A) is a finite set and ~: X • S is

finite. Then patch U and X"-q1(~)- along U - ~ ) ,"" where q:X" • X

is the projection. We have contracted q-1(~ )to ~ . If we do it for all

the ~ E -I (Sing A) such that 0X, ~ is regular, we get a coarse

deslngularization of X .

Now we go to the heart of JUNG's idea.

Definition Let X be a normal surface. We say that X can be desin-

gularized by blowing up of points if there exist a finite sequence

x - x o. ..........
U U U U U
FI F2 F3 FN FN+ I

such that (i) X. is normal for o < i < N

1

(ii) Fi+ I is the singular locus of X i for 0 4 i ~ N

(iJJ) Xi+ I is obtained by blowing up of Xi with center

Fi+ I , I < i < N .

(iv) FN+ I is empty.

Proposition: Let S be a regular algebraic surface over C . Let X ~S

be a finite covering. Assume that X is normal. Let A be a normal

crossing divisor of S such that if we let S = S-d and if ~ = X----~S

o o o o
is the restriction of w to S then w is etale . Then X can be
o o
desingularized by blowing up of points.

With this formulation it is obvious that we can replace X

and S by the corresponding analytic spaces: in fact blowing up of closed

 58

points commutes to that operation. Hence we can assume that we are in

the complex analytic case and that S = D X D and S = D' X D' , where
o
D is the unit disk of the complex line C and D' = D- { o } .

Let X be a connected component of X . It is a finite etale covering

0~ 0
of So , h e n c e Xo, a is given by a finite quotient group Q of

~l(So) = ~ X ZZ . L e t 0 = Ker(Z~ X Z~ ~Q) . If G = a Z~ X b ~ with

integers a and b , then the covering Xo, a is isomorphic to

D' X D' • DIX D' , (x,y) h-@ (xa,y b) . Furthermore such a covering can

be extended to a ramified covering of S by a smooth space namely

D X D P D X D = S , (x,y) t-@ (xa,yb). Of course G need not be such

a diagonal subgroup. Nevertheless there exists a diagonal subgroup

GI of ZZ X ~ such that GIC G c Z~ X ?Z and such that G/G I

is cyclic. This is an exercise about integral lattices whose solution

gives the following. Let (e I ,e2) be the canonical basis

of 2ZXZZ . There exist integers a,b,c, such that G =ZZae1+ZZ (bei+de 2) .

Let a=G.C.D.(a,b) and a'Re/a Let G1-?/ae1+?Za'd e 2 then GIC G ,

and G/G I is generated by ~2 = be,+de 2 , w~th a'~2£ G I . The ramified

covering corresponding to GI is D X D P D X D , (x,y) ~--, (xa,y a'd) ,

the Geiols group QI I (2Z X ZZ )/G I acts via two primitive roots of

unity ~ , ~ with ~a=wa'd=1 . The generator ~2 of the subgroup

G/G I acts by (x,y) ~-9 (~bx, W dy) which can be written as

(x,y) ~ (tPx,ty) with t = W d , with suitable p and t a'= I ,

(a',p) = I . Hence the quotient of D X D by G/G I is the well known

cyclic singularity labelled (a' ,1,p) . It is a toro~dal embedding

(in French it is called an eventa11). Now I claim that this ~ventail

is nothing but the closure in X of the component X we started with.

o,~
In fact one has a diagram
 59

D' X D' " DXD~

~u

• X

S • S
/4
0

where ¢(x,y)=(xa,y
a ' ~d - ) " ~ , and where u° exists since GI C G . The

morphism u can be extended to D X D (dotted arrow) because D X D

O

is smooth and X is finite over S . Since u gives an isomorphism

.~ O
D'XD'/(G/G I) ' X °' ~, the morphism u factors through the quotient

Xa = D X D/(G/GI) and we get

D×D

X X
V

Since Xs and X are finite over S the image of v is the closure

X of X in X . Since X is normal and X is a connected

0,~ OjG 0,~

component of X one has that X is a connected component of X .

0 0)~
Now Xo, ~ and X are both normal and v is an isomorphism when

restricted to S ; hence v is an isomorphism because v is finite.

0

Eventually we get that X is a finite union of toro~dal

embeddings. By explicit computation one knows that a toroldal embedding

can be desingularized by repeated blowing up of closed points.

This finishes the proof of the proposition.

In LAUFER's book,or HIHZEBRUCH's thesis one can find an explicit

description of the minimal resolution of X in terms of continued

fractions. This is a special feature of toro~dal embeddings of dimension

2 and a very useful one too. But if one follows this line of argument

it is not so easy to deduce from it that algebraic surfaces can be

 60

desingularized in the sense that the desingularization is also an

a l~Eai_ce variety: in other words X is only analytically isomorphic to

a toroidal embedding. It is to avoid this kind of difficulty that I

choosed to talk in terms of desingularization by blowing up of points,

a property which obviously goes back from analytic to algebraic geometry.

An old dream of many specialists is to extend JUNG's method to

dimension 3. We first notice that we would need simplification of the

boundary A inside S with dim S = 3 . The development of desingu -

larization theory has shown that this is more difficult than coarse

desingularizatlon of surfaces: it is more or less equivalent to smart

desingularization of a surface in the special case where this surface

is a hypersurface. Anyway simplification is known but we still face a

serious problem. If X---~S is a ramified covering with dim S=3

and such that the discriminant A has normal crossings, we can argue

as above and see that X is locally analytically a quotient singularity .

The prototype of them are toro~dal embeddings. One can desingularize

toro~dal embeddings of any dimension but, unfortunately,there are many

ways of doing that and there is no known rule to choose one of them

in such a way that the constructions made locally will ~atch to~ether

and give a global desingularization of X .

§ 4. De luxe JUNG's method.

ZARISKI has shown that JUNG's idea can be used to prove smart

desingularization for a surface-hypersurface. We just saw that for a

normal surface X , the existence of a finite projection ~:X , S

with normal crossing discriminant allows us to give an explicit description

of X as a union of toro~dal embeddings.ZARISKI'S method is based on

the fact that for a surface which is a hypersurface the existence of such
 61

a projection implies that the problem of lowering the multiplicity is

a purely combinatorial one. From that he is able to extract a proof of

the smart deslngularization theorem in that case: he says that you first

have to blow up points, he says when you stop blowing up polnts am~ start

blowing up ~- fold curves and of course he proves that you eventually

get rid of all u- fold points.Another version of an analogous idea

has been given some years ago by ABHYANKAR, but I will not talk about

it since it is going to be published in the proceedings of the

Reinhardtsbrunn Conference 1978.*

Let us give the main features of ZARISKI's proof. The argument

being limited to ¢~ar.O, we avoid unessential difficulties by declaring

that, up to the end of this lecture, scheme means of finite type over

the complex numbers.

Definition Let X be a scheme.

We say that X is a hypersurface locally at x if rank (M~,x/M2x,

x ) _- ~

= I + dim OX, x . One has a surjeetive morphism of graded rings

(,) k [MI_M -- , k -k(x) ,

r~O

we denote by ord (X) the smallest integer n such that (I) is

X

no'c injectlve in degree n .

We know that if X is a hypersurface at a closed point x then there

exists an etale nelghbourhood U of X and a closed immersion U---~Z

where Z is smooth over k and U is a Cartier divisor in Z

From this we deduce that X is a hypersurface at all points x' belonging

to the image of U and that Ordx(X ) is semicontinuous. Hence, there

exists a maximum ~ for Ordx(X) .

E d i t o r ' s Note: A b h y a n k a r d i d not s u b m i t a p a p e r to these p r o c e e d i n g s . The reader m a y

refer i n s t e a d to the article by U . O r b a n z in this volume, pp. 1-50.
 82

A v-fold scheme is a closed subscheme Y of X which is

smooth and such that ord (X) = v for shy x ~ Y ; we will also say
X

that Y is v- permissible. One can prove that by blowing up a v-fold

Y in X we get an X' which is still a hypersurface and that

v (X') < v(X) , see Prop. 1 underneath. If we can prove that there exists

a sequence of v-permissible blowing up

X = X o, XI~---.X2.....Xn_ I , Xn such that ~ (Xn) < v(X) we

obviously get smart desingularizatlon by repeating this process a finite

number of times (except for condition ( i v ) ) .

Now the first point is that if one has a transversal projection

~: X S and a v -permissible center Y in X then after blowing

up Y we still get a transversal projection. More precisely we have the

following

Proposition I. Let 7: X •S a flat and quasl-finite morphism.

Assume that

(i) S is regular and dim S = 2 and X is everywhere a

hypersurface. Let ~ = v(X) - max { o r d x ( X ) , x E X }.

(ii) For any x E X such that ordx(X) = v ,

one has length 0X,x (0X,x/ MS,s0X,x) = v , where s = ~(x) .

Then if x E X and v = Ordx(X) then locally for the etale

topology there exists an open neighbourhood U of x in X such that

(A) If Y is a closed ~-permissible subseheme of U the

projection w induces an isomorphism Wo: Y , ~ (Y) C S ; hence ~(Y)

is smooth.

(B) By blowing up of U with center Y and of ~(U) with center

~(Y) one gets a commutative diagram (*) and if v(U) = v(U') then

(i) and (ii) hold for ~' .

 6S

U' U'

~)
w • St
g

We leave to the reader the reduction to the following case:

S = Spec(A) and X = Spec(~/f~T]), f [~ Tg+ aiTg-1+..+au, a.Ex A .

aI
We perform the usual trick T~--*T + - w and we can assume that a I- 0 .

Let S (X) = IxE X , ordx(X) I ~ 1

Then I claim that T = 0 at all points of S (X) and this will obviously

prove (A) . Let us prove the claim. If x is a u-fold point, by which

I mean that x G Su(X ) , then f E M9 where Z = Spec (A [T] ) ,

- Z,x "

hence (ul). T = a u-1v-------~fE _Mz,x

aT hence T = 0 at x ow we are going

to prove (B) . Let Y be a u-permlssible center for X . By definition

Y is a smooth subscheme of Z of codimension 2 (resp.3) hence locally

around any point x E Z one can choose equations of Y in Z of the

form (T,x I) (resp. T,Xl,X2) . Such that the equations of ~(Y) in S

are (Xl) (resp.(xl,x2)) . In other words, if P is the ideal of Y

in Z and Q the ideal of ~(Y)

S one has P - Q A [ ~ + T ~ , in
ak pU-k
and Q - F ~A . ~ince f E pV , then ~ f E I ~ k ~ v-1 which
aT k ' ,
implies a.e Qi . We are now going to describ~ the diagram (*) of
i
prop I .

First case dim Y = I . We know that X' is covered by two afflne pieces.

In the first one we have POx,= TOx, and in the second one POx,= x10 X, ,

Since ale x~A , the unitary equation TV+ a2TU-2+ .. + a u can be written

t _U--2 %) a' a: 6 A ,
T~+ xla 2 T +-+Xl U ' 1

and from this follows that in fact X' is equal to the open set where

POx," XIOx, hence X' is affine and we have

!
El I Spec A [ TII / ( T I ~ 8,2 T1-2~-al ) , TII T/x I •
 64

As a consequense X' is finite over X .On the other hand S' - S since

w (Y) is a Cartier divisor a hence blowing up of ~ (Y) amounts to nothing.

Property (B) is obvious in that case.

Second Case dim Y=O, Q = (Xl,X 2) , P = (T,x 1,x 2) and of course

Y " l'lis a closed point.The blown up X' is covered by three affine pieces

but as above one sees that it is covered already by only two of them,namely

X'(i) " I ~ E X' /xiOx,,~'POx,,~ I, i " ~,2 •

The blown-up S' is also covered by two affine pieces

Zet
S'(i)
1 n e s' / Xi0s,,n= Q 0S,,n
S'(i) = Spec (A(i)) . Since a
n
E Qn
1
one has a
n
~ x.nA(i)
1
, hence

there exist a'2,a'3,...,a' 9 in A(i) such that a 2 = x~a' 2 ,

a 3 = x 3i a~ ,,., a = x~l a'~ and eventually

s'(i) = Spec A(1)

' -V-2 ,
for i " 1,2 , X'(i) = Spec A(i) [ ~ I C T ~ +
a 2 T1...+a 9 j
T I = T/x I
Hence property (B) is prove~ and we even have an explicit description

of the way the equation of X behaves by permissible blowing up.

Furthermore, observe that the discrimlnant 8 of f(T) is a weighted

homogeneous polynomial in the variables a. with total weight 9(~-I ) .

l
Hence if Q is the ideal of ~(Y) in S , since we know that
Qi Q~(V-1 )
aiE , we know that 8 E . After blowing up we know that

all ) ai/t x where t is the e%uation of the exceptiona~ divisor

in some affine piece of the blown up scheme S' . Hence we know that

the new discriminant is

(*) ~ = ~I t v(~-1)
As a consequence, if ~ had normal crossing in S then 6' also has normal

crosslng in S' . Proofs.if we blow up one point this is clear;

 65

if we blow up a 9-Fold curve we do not change S .

We start explaining how to use Prop. I .

One can cover X by affine pieces X each of which is equipped with

a projection w :X , S with the properties (i) and (ii) . For

each of them we have a discriminant A , hence a finite set F in

each Xa such that x ~ F means that ordx(X ~) ~ v and w(x) is a

point of S where A is not normal crossings. If we blow up Fa

in X we get a new famil~y X' S' and the X' still cover

X' . As we said in §2 , if we keep blowing up these points then the

Aa will eventually become normal crossing divisors. And this property

will be stable by any further v-permissible blowing up . But now we

will have to perform blowing up with one-dimensional support.

Here comes the fundamental lemma of Zariski ~ tells us

what happens.

Proposition 2. Let S be smooth of dimension 2 and

let X = Spec A[T]/(TV+ a2TV-2+ ..+av) , where S = Spec(A) .

Let Ul,U 2 in A such that UlU2=O is a normal crossing divisor of S .

Call this divisor A and let So = S-A. Assume that Xo=XXsS 0 is etale

above SO and that s is a point of S with u1=u2=O .

Let rI be the curve with equation T = u1= 0

Let F2 be the curve with equation T = u2= 0

Then Sv(X) C r I u r 2 ,

Furthermore there exist two numbers AI and ~2 such that

if we replace S by a convenient neighbourhood of we have

(i) AI~ I *-~r- I C Sv(X )

(ii) x2 ~ 1 - - r 2 c sv(x )
(iii) x1+ x ? I--{I} =l-lnr 2 ¢ Sv(X) .
 66

We already saw that T = 0 at any point of Sv(X) because

at= 0 . Furthermore ~(Sv(X)) is contained in the discriminant locus A

hence Sv(X ) C r I u F2 .

Replace S by a small enough complex analytic neighbourhood

of s ; there exists a complex analytic branched covering SN ~ S ,

given by u1= Xl N , u 2- x2N with smooth SN and such that the covering

X0/S 0 is trivialized by the base change SN D g for instance

let N be vl . In other words ~e have v morphisms r :SN, ,X ,

each one of them being given by complex analytic functions

N N
T = ra(Xi,X2) , u1=x I , u2=x 2

and we have H(T-ra~1,x2)) = f(T,Ul,U 2) .

Let R = G.C.D (r - ~ , s @ ~) .

Since the discriminant ~ = ~ ( r -r.)

• a b ml m2
is ~ z unlt.ulu 2 , one has that R = unit. x I x2 . We let

(I) ~I = m11N ' X2 " m2/N "

I claim that there exists a complex analytic function

g(ul,u 2) such that for each root r (Xl,X 2) one has

m I m2
(2) r~1,x2) = g(ul,u2) + x I x 2 h (Xl,X2) .

Let r a " Z r a,i,j x~ x~ be a root and define

g(ul,u2) =~ - i "
Nli,Nlj,(i < mlor j < m2) ra'i'jXl x~ .

mI m2
Since x I x2 divides r - rE we get that g~ does not depend

on ~ . Call it g(ul,u 2) .

To prove (2) we introduce the Galois group G of the covering

SN/S : this group is generated by

(Xl,X2)l >( ~Xl,X 2) and (Xl,X2) I >(Xl,~X 9)

 67

where ~ is a primitive N-root of unity. Since G acts on the roots,

if we look at a monomial ra,i, J x~ x~ with N~i and r ,i,j~3 /

then we must have i • mI and ~ • m 2 , since ra(~Xl,X 2) is a root ;

idem if N~j and this proves (2) .

We now prove the ~ part of (i) . If A~ > I, then m I) N

ffi ml m 2
and since f(T) ~~<v
l< (T-g-x I x 2 h a)

we have that f(T) belongs to the ideal (T-g,xIN) v

t

hence the curve [-I with equation T-g - u I ffi 0 is v-permissible.

As we saw that Sv(X ) C [-IU F 2 we have [-~ ~ [-I hence ~I is w-permissible

and uI must divide g .

Similarly for the ~ part of (iii), we have ~I + ~2> I hence

m I + m2 • N hence f E (T-g, (x 1,x2)N) v hence f 6 (T-g,u 1,u2 )v .

This exactly means that the point ~' defined by T-g - u I ffiu 2 = 0

is v-fold. As we already saw that T = 0 at any v-fold point we

conclude that ~' = ~ where ~ is defined by T = u I - u 2 -0 , in

other words I~I = FIO ['2 ' hence " of (iii) is true.

The proof of the converse is very tricky and since we don't

find a better one we refer to [Zariski 0 . ,Fxceptional singularities

of an algebroid surface and their reduction. Acad. Naz. dei lincei,

Serie VIII, Vol XLIII, fasc.3-4-(1967) p.135-146] which is reproduced

in volume I of [~ARISKI's collected papers, M.I.T. Press (1972)] .

The same paper also describes the behavior of 11 and ~2

under blowing up with v-fold center which is as follows

(A) if ~i• I and if we blow up FI then we don't change

S and don't change r-1 and [-2 but (11'A2) is replaced by (AI-I ,12 )

(B)if AI + ~2 • I and if we blow up I~I -- [-I or-2 then

the d.n.c, will acquire a new component namely the exceptional divisor
 68
of the blowing up of

(11,1e,~2)
s = #(~)

attached to
in

PI' r'e' F 2
S . Call it

will be
r

~I' Xl +
e
. The three numbers

12-1,X 2)_ . Picture:

~2
We will not say how this local study is used by ZARISKI to

get the desingularisation of surfaces for two reasons. The first one

is that we will mimic this part of his argument in the next § . The second

one is that one can find ZARISKI's proof in [ZARISKI 0., A new proof

of the total embedded resolution theorem for algebraic surfaces (based

on the theory of quasi-ordinary singularities), Amer.J. of Math. (1978) ] .

Since this paper is fairly cleverly organized it is not to be explained

in a few pages,andthe interested reader should look at it. He will see

that these two papers deserve the qualification of "de luxe".

§ 5. HIRONAKA's method applied to surfaces.

As ZARISKI does in the paper we have just quoted, we will

reduce the problem to a d.n.c, case, but here this condition does not

refer to any kind of discrimlnant but rather to the coefficients of the

equation of the singularity. Hence we will not use Prop. 2 of § 4. Here

surface means reduced scheme purely of dimension 2 and of finite type

over ~ . We will only treat the hypersurface case and refer to remark 2

at the end of the lecture for the general case.

We will repeatedly use in the proofs the local model we already

studied in § ~ , nsJnely:
 69

S is a smooth surface,

Z = Spec 0s[T]

x - Spec ( O s [ ~ ] / , ~ ( T ) )

With f(T) = Tv+ a2T~-2+..+ao , ale 0s(S) .

As above if P is the ideal in Z of a v-fold subscheme Y of X

one must have T E p and a.E1 Qi , where Q = P NO B.

Proof:~kf E pV-k . Converaely if Q is an ideal of 0S such that

~Tk
(2) Spec (0s/Q) is smooth and a.E
I Qi ' i =2,3,...,~ |

then P~Q 0Z+ T.O Z is the ideal in Z of a y-fold subscheme Y of X .

Hence the whole process of v-permissible blowing up can be

expressed in termaof S and the a. . After blowing up, the a.'s are
i I

only locally defined modulo the choice of a generator of the ideal of the

exception~] divisor. To avoid this difficulty we introduce ideals

(3) A i = a i~I/i0 S , i = 2,3,.., v , and we know that if Q is

the ideal of a smooth closed subscheme Y' of S then the s u b s c h e m e ~

of Z whose ideal is T 0z + Q 0Z will be a p-fold aubscheme of

X if and only if

, i = 2,3,.., v .

Furthermore if S' is the blowing up of S with center y,

(which is equal to S if Y' is a curve) then

(5) Ai is replaced by Q-VlAi0s, = A'.I "

(A) Normal crossing condition

We say that we have normal crossin~ at s e S if all the

ideals A2,A 3 .... ,Ao , ~ - Ai , ~ Ai are d.n.c.

2~ i~ v 2 ~ i~ v

in a neighbourhood of s . In that case there exists an integer i E [ 2,~]

such that for any j ~ [2,v|jAiD Aj (in other words Ai divides Aj)
 70

in a nelghbourhood U of s . Hence an ideal Q is v-fold in U if

and only if Spec(0u/Q 0 U) is smooth and A.Cz QVl (in U) . Furthermore

the corresponding y' nU has to be either a branch of the d.n.c or a

point belonging to the d.n.c, and after blowing it up we will have

again normal crossing. Furthermore for any point s' E S' lying over

s E 8 the same index i will be such that A'.D

x A'.j for any

j E [2,p] (See(5)) •

If s is a point of the d.n.c, we will have local coordinates

)'1 ~'2
(Xl,X2) and integers ( ~I' ~2 ) such that AiOs,s=(Xl x 2 ) Os, s

If the center of blo~rlng up Y' is a curve, its equation will

be for instance x1= o (locally at S) we will have S' = S and

A' i 0S,s =
(X~l 1 -v I , x2
X2 ) 0S, s

If the center of blowing up Y' is a branch point let

w:S! - S be the blowing up and let U' = #-I(u) . In U' we will

only have two branch points lying on the exceptionnal divisor with the

following rule ;

(i) let s I be the branch point lying above s and on the

strict transform of the curve x1= o . We have local coordinates

t = x2 and x' I = xl/t with

' ~" ~'-+ ;'2- Y l
A' Os, = (x 1 r,t I ) Os'
i ,s I ,s I

(ii) let s 2 be the branch point lying on the strict transfoEn

of the curve x2= o . We have local coordinates t = x I, x' 2- x 2 /t

with
~I + ~2-~I x'~2
A' i Os, =' ( t 2 ) Os "
,s 2 ' , ,S2

We will use later on the following

 71

Observation. ( X I < v| and Z2 < v| ) *-, (there is no v-fold curve

through s) . In that case let ~ = ~I + ~2 " Then (~> v I) *-~ (s is v-fold).

If we blow up s when Z! < v l ' X2 < v! , ~>o: , then the exceptionnal

divisor is not v-fold since the corresponding A is ~I + A2 - v! .

Furthermore, for the two branch points s I and s 2 lying above s we have

the numerical caracters

(Xl, #-v! , ~u + X l - V l ) at s1
(Xl,A2~) -
(X2, # - v i , # + X2-ol) at s2

Since 11,l 2 are < v~ one has ~ < 2 v! hence the hypotheses

(X I < v! and ~2 < v[) is still true at s I and s 2 . Furthermore one

has #(S I) < ,(s) and ~(s 2) < ~(s) .

Remark. If one knew that ~normal crossing condition ~ does not depend of

the local model we choosed at ~ E X we could blow up repeatedly the

)
discrete, finite, closed set where this condition is not satisf~ed~ As we

said earlier since dim S = 2 we would reach normal crossing after a

finite number of steps. Then it would be a purely combinatorial game

to get rid of all ~-fold points or curves. This is the line of

reasoning of Hironaka when he does his gardening. For the sake of

simplicity, we are going to argue differently but we will describe

an algorithm with the following properties

(i) if X' • X is an etale morphism if P(X) is the

J
result of the algorithm applied to X then X' x ~ ( X ) is P(X' ) .

(ii) one can describe the algorithm without having to know

anything about the way we prove that it works.

 72

(B) Blowing up of v-fold curves

Lemma I. Let X be a surface-hypersurface and let C be an irreducible

v-fold curve with v = v(x) . Let p:X' ,X be the blowing up of

X with center C . Then p is a finite morphism. Furthermore, if

(*) dim (sv(x')~-1(c)) ~ o

then p-1(C)re d is smooth and the projection ~ induces an i s o -

morphism p-1(C)re d ~ ,C .

Let ~ be a closed point of C • There existJ a complex

analytic neighbourhood U of ~ in X provided with a fi2ite and

flat projection

(I) ~: U ~S

such that S is smooth, and U is defined in Z = B X~1 by an equation

(2) f(T) ,, ~+ ,~2~-2+..+%

the ai being holomorphic function~on S . Furthermore one knows that

T - 0 on C hence w induces an isomorphism C , w(C) . Let t = o

be the equation of ~(C) (choose a smaller U if necessary) .

According to proposition I of § 4 we have a diagram

P U'
U /

where U' is the blowing up of U with center U n C. As we saw in the

proof of the proposition I of § 4, we have that p is a finite

morphism. Let TI=T/t be the variable in U' ; we know that TI= 0

on Sv(U' ) hence w'(Sv(U') ) = Sv(U') . It follows that the projection

~' induces an embedding of S~'(U') • p-1(C) into w'(p-1(C))=w(C) •

Hence if

(*) dim (Sv(U') n p-1(C) ) ~ 0

 7S

one has that w'(Sv(U')O p-1(C))= w(C) .

[ We should have chosen U such that C n U is connected and warned

that C has been replaced by C n U ].

The lemma follows from this. One could avoid the use of complex analytic

geometry by considering the completion of 0 or an etale neighbourhood

X,~
of ~ ; it is a matter of taste.

(C) Blowing up of a closed point.

Lemma 2 Let ~ be a u-fold point of X with ~- (x) . L,t

p:X' ,X be the blowing up of ~ .

Assume that

(*) dim (%(X') ~ p-1(~) ) • 0

then p-1(~)re d is a projective line contained in S~(X') .

Again we choose a local analytic model ~:X. , S and blow

u~ in X and s - ~(~) in S . We get a commutative diagram as in §~ .

P X'

~E x'IT / 1
S'
s ~

We know that w' maps S (X') isomorphically on its image. Hence

~'(S (X') o p-1(F.) ) is mapped isomorphically on a closed subscheme of

-I -I
q (s). Since q (s) is a projective line this image is either a finite

set of closed points or q-1(s) itself Since we have (*) we are in

the second case. Hence we get the conclusion. Picture: -~ )

@

We are ready to describe the algorithm.

 74

0~eration I. Datum: X where X is a reduced scheme of finite type

over ~ . X is purely of dimension 2 and for any x E X one has

rank (mx, n / m_X,

n2 ) = I + dim Ox,n. For short we say that X is a

surface-hypersurface~ .

Let v - maxIordx(X) , x E X } . IF v = I GO TO END .

Let S~(X) - I x E X , ordx(X) = v} with the reduced structure

In(X) " { x ~ X I x is isolated in S~(X) }

Zv(X) = S ~ ( X ) ~ I (X) (-1-dimenslonal part of S~(X) )

F(X) -- Sing ( Zv(x) )

IF Fv(X) @ @ apply Operation 2 to (X,Fv(X),P) •

IF F~(X) = #. IF E v ( X ) ~ @ apply operation 3 to (X,E ~(X) , ~) .

IF Fv(X) = @ , IF Z v ( X ) = # apply operation 2 to (X,Iv(X) , v) .

Operation 2 . Data: (X,F,v) where X is a surface -hypersurface,

F is a reduced closed subscheme of dimension zero of X , any point

of F is m-fold and v = v(X) . Blow up F in X and get p:X'.----~X.

Look at the union of those irreducible components of dimension I of

p-1(F)re d wich are v-fold, call this union C' . IF C' - # go to

Operation I for X' . If C' @ # go to operation 3 for (X',C',P) .

O~eration B . Data: (X,C,P) where X is a surface-hypersurface, C is

a m-fold curve(not necessarily connected) and v = v(X) .

Blow up C and get p:X' , X . Let C' be the union of the irreducible

components of dimension I of p-1(C)red wich are v-fold . IF

IF C' - # go to operation I for X' . IF C' ~ @ go to operation B

for (X',C',~)
 75

We first have to check that when we go from one operation

to another one,the new data satisfy the hypothesis of the second operation.

Since permissible blowing up preserve the surface-hypersurface condition

ors
the only difficulty may occur when we go from 2 to 3. In both cases one

still has v = v(X) since C' @ ¢ (otherwise we would go to I ).By

definition each irreducible component of C' is smooth since it is

v-fold. On the other hand, according to (B) and (C) there is at most one

irreducible component of C' above each irreducible component of the

center of blowing up Y (which is F for operation 2 or C for operation 3).

Since Y is smooth, two distinct irreducible components of Y do not

meet hence two distinct irreducible components of C' do not meet hence

C' is smooth and obviously v-fold.

Property I. Let ~:X' , X , be obtained from (X,Y,v) by performing

operation 2 (resp.3), with Y = F (resp. Y = C) . It is clear that Ev(X')

contains the strict transform lo(X)st of Zv(X) . Let E be the

union of those irreducible components of Ev(X') which are not contained

in Zu(X)st . We have that w(E) C Sv(X ) . In the case of operation 2 we

must have that w(E) C F hence according to (B), E is the union of a

finite set of mutually disjoint projective lines. Hence E = C' . In the

case of operation 3. for each irreducible component E of E one has

dim ~(E ) = I since ~:X' . •X is a finite morphism. Since

E ~ Zv (X)st one must have that E n C is dense in E

(since w(E ) C Zu(X ) and ~(Ea,)st = ¢) . Hence ~(E ) is a component

of C • Hence Ea is v-fold as we saw in (B), hence uEJ C' and

obviously U E = C' . As a conclusion we get

(D) r.v (x') - r.v (x)st U c'

with Ev(X)st ~ C' is finite and C' is a union of mutually disjoint

v-fold irreducible curves.

 76

Property 2. One can only perform operation 3 a finite number of times

before having to go back to operation I.

Remember that repeating operation 3 produces a succession of

finite and birationnal morphisms X = X o, XI~----X2...X n. By the

finiteness of the normalization of X one only has to show that

X i ~ Xi+1 for i = o,I,..,n-I. In other words, with the notation of

operation 3 , the morphism w:X' ~X cannot be an isomorphism. Since

v can only change during operation I, one has v~ 2 hence X is not

smooth at any point ~ E C ; since C is smooth it follows that C is

not a Cartier divisor of X hence ~ is not an isomorphism

(check that C @ ¢) . Alternate proof: look at what happens in the basis

S of any local model.

Property 3. When performing operation I for some X , when we first go

back to operation I we find some X' and we have that

Z~(X') = Zv(X)st is the strict transform of Zv(X).

By transitivity of the strict transform this follows from

(D). In fact when per~r~ing 2 or 3 , if C' ~ ¢ we go to operation 3 and

we can only perform operation 3 a finite number of times, say n times,

before having to go back to operation I . This means that we have a

sequence of modifications

X. X I" X2... Xn_ I , Xn

and ~-fold curves C I in XI,C 2 in X2,... , C n in X n such that for

i = 1,2,-,n-I , Xi+ I is obtained by blowing up Xi with center Ci .

Furthermore C = ¢ expresses the fact that we have to go back to

n

operation I . Formula (D) tells us that Z(X I) - Z(X)st U C I hence

Z(X 2) ~ Z(X)st uC1,st uC 2 but the strict transform of the center of

blowing up CI is empty hence Z(X 2) - Z(X)st UC? and by induction

Z(X n) = Z(X)st U C n U Z(X)st .

 77

Property 4. After a finite number of passages through operation I

one h a s Zv (X) = ¢ which means Sv (X) = Iv(X) . Furthermore once this

is achieved this remains true as long as v(X) remains constant.

Remark that if F (X) ~ ¢ then,with notation of property 3

Zv(X') = Zv(X)st and at any singular point of Zv(X) we have performed

some non-trivial blo~-ing up. Hence after a finite number of passages

Zv(X)st is smooth hence ~-fold. Then operation I tells us to go to 3

which means that the new strict transform will be empty and we get the

conclusion by (D).

To prove that the algorithm stops we can assume that

Iv (X) = Sv (X) and we only have to show that v eventually drops.

Since I (X) is finite we only have to look at some ~ E I p (X) and

we can replace X by some local model w:X~S, w(~) - s ,

as in (A) . We call

(I) X -- X ° XI~,------X2...Xn_ 1 9 X .

the result of the successive passages through operation I, and we have to

show that for n big enough one has ~ ~ p(Sv(Xn)) .

Let ~nE Sv (Xn)= Iv (Xn)

and call ~n-I~'~I ' ~o = ~ it's successive projections. Since operation (5)

does not change S (blowing up of a smooth curve) , we get a succession of

blowing up with finite and reduced centers G. in S called

i

S = So • SI ( S 2... Sn_ I ( Sn

Let Sn,Sn_ I,..., s I, So= s be the successive projections of the ~i "

Since we alws~s pass through operation 2 between Xi and Xi+ I , we know

that s.6 G. . As a consequence of simplification of boundary in S

l 1
(here dim S = 2) we know that there exists N such that for n > N , we

will have the d,n,c, condition at any point of G

n
With the notations of (A) at the point Sn we know that At< vl and

A2 < v| because we know that S v(x n) = Iv(X n) is of dimension 0 .

 78

Hence ~ (Sn) = AI+ X2< 2 v! . But we also know that ~ (Sn+ I) < ~ (sn)

hence for n > N + v[ we will c e r t i f y have Sv(Xn) = # since ~(Sn)< v:

This proves that the algorithm stops.

Remark I. We have proven the smart desingularization theorem except for the

condition (iv) which relates to normal crossing of the exception al divisor.

It is an easy exercise to modify operation I in such a way that the algorithm

also achievesthat.

Remark 2. If we only assume that X is reduced and purely of dimension 2

we let ~ (X) be the HILBERT-SAMUEL serlegof X at x ,

X

~X(x) ~n~ig(~,x/~l~) Tn. We modify operation ( 1 ) a s follows:

let ~ be the maximum of ~ (X).If v=(1-T) -2 GO TO END.

x /

let Sv(X) = { x E XiVx(X)= v} with the reduced structure.

We don't change the sequel.

In the text of operation 2 and 3 replace ~surfaee hypersurface~ by X is

reduced and purely of dimension 2.

I claim that this algorithm also stops. Of course the proof is

more complicated due to the fact that the local models must be obtained

by the use of a normalized standard basis for the ideal of a (local)

embedding X C Z . Furthermore one has to show that v cannot decrease

indefinitely. This two difficlties are not easier to overcome for

dim (X) = 2 than for the general X . But when this is done, the proof of

(A),(B),(C) and of the finiteness of the algorithm (for dim X = 2) are not

seriously affected by the lack of the hypersurface hypothesis.

Remark 3. The algorithm can be changed in a few different ways and still

enjoy the property (i) we stated at the beginning of this § . This is a

good exercice for the interested reader.

 DESINGULARIZATION IN D I M E N S I O N 2

VINCENT COSSART
Universit~ Pierre et M a r i e Curie (Paris VI)
Math~matiques
4~ P l a c e Jussieu
F - 75005 PARIS

INTRODUCTION. The a i m of t h i s paper is to l i n k four different

methods of d e s i n g u l a r i z a t i o n of e x c e l l e n t surfaces: The methods of
Jung, Zariski, Abhyankar and Hironaka. The basic reference is G i r a u d ' s
lecture in this volume ( [8] ). In this paper, Giraud shows how close
the m e t h o d s of Zariski and Jung are. So w e a r e g o i n g to s h o w t h a t the
proofs by Zariski, Abhyankar and Hironaka are almost the same f r o m the
point of v i e w of the c h a r a c t e r i s t i c polyhedron of s i n g u l a r i t y . Indeed
these three authors want to r e a c h this case: At e v e r y c l o s e d point of
the w o r s t Samuel stratum of X (= a r e d u c e d hypersurface of an e x -
cellent, regular scheme Z of d i m e n s i o n 3 ) there exists a regular
system of p a r a m e t e r s (y,ul,u 2) such that the p o l y h e d r o n A(J,U~,U 2)
(see [9] , 1.12)) has only one vertex, J being the ideal of X in
OZ, x . In t h a t case, Hironaka calls the singularity "quasi-ordinary"
and Abhyankar calls it "curve-like".

The first section defines properly the notion "A(j,ul,u 2) has

only one vertex" and shows how nice it is.

In t h e second section we t r a n s l a t e Zariski's method into terms

of the c h a r a c t e r i s t i c polyhedron.

In t h e third section we g i v e a new proof of A b h y a n k a r ' s t h e o r e m

with the help of the characteristic polyhedron.

The fourth section is just a reminder of H i r o n a k a ' s proof.

Acknowledgement

I would like to t h a n k U. Orbanz for his thorough and critical reading

of the d r a f t s of t h e s e notes. He a l s o carried out considerable editing
on t h e m a n d u l t i m a t e l y wrote t h e m u p in a m u c h b e t t e r final form than
my original version.
 80

I Quasi-ordinary points

To simplify the proofs of this paper, we will only look at the

surface-hypersurface case. Indeed the proofs are long enough
and the specialists will be convinced t h a t m o s t of t h e c o m p u t a t i o n s
may be generalized to the case of a surface X embedded in a regular
excellent scheme Z of any dimension.

NOTATION AND DEFINITION. Let us denote by X a reduced hypersur-

face of an excellent, regular scheme Z of dimension 3. We denote by
Sam(X) the locus of highest multiplicity of X . Let x 6 Sam(X)
be a closed point and let I(X) = (f) at x . We say that x is a
quasi-ordinary point if it satisfies the two conditions:

(I) dim Dx(X ) = 2 , where Dx(X) is the directrix of the tan-

gent cone CX, x of X at x . ( [7] , 1-31; this condition has the
following meaning: If v = ord (f) , then in (f) = c.Y ~ for some
x x
Y and some unit s.)

(2) There exists a regular system of parameters ( Y , U l , U 2) of

Z at x such that: (inx(Y)) = I(Dx(X)) c grx(Oz) , the polyhedron
A(f;u1~2;y) is m i n i m a l and has only one vertex (see [9] , (3.1 )o

REMARK. In the case where OZ, x and its residual field k have
the same characteristic, the ~hoice of Ul,U 2 defines a projection
A

X ~ Spec(k[Ul,U2~ ) , where X = Spec Ox, x is the completion of X

at x . Therefore, we recognize Jung's idea: We project X onto
a regular surface.

PROPOSITION I. Assume that x is a quasi-ordinary point of X .

Let (Y,Ul,U2) be paramete~ satisfying (2) r let (11,12 ) be the
vertex of &(f;u],u2) and let ~ = ord (f) . Then we have:
x
a) If DI = V ( y , u I) is permissible for X (see [12] , p. 107)
if w e blow up D I , then there is at most one point x' near to x .
Let Z' be the transform of Z . Then (y' = Y/Ul,Ul,U2) is a regular
system of parameters of Z' at x' . Furthermore, if x' is v e r y near
to x *), t h e n it is a q u a s i - o r d i n a r y point and £ ( f / u lv; u l , u 2 ) has only
one vertex with coordinates (I I - 1,12)

b) If neither V(y,ul) nor V(y,u2) is permissible for X and

if we blow up x , then there are at m o s t two points near to x. Let

Recall the notion of near and very near points to X : A point x I in some blowing
up x is called near to x if at x S the multiplicity is the same as at x .
x t is called very near to x , if in addition the dimension of the directrix is
unchanged ([6], introduction).
 81

!
Z' be the transform of Z . Then (y' = Y / U l , U l , U 2 = u 2 / u I) or
(y" = Y/U2,U ~ = Ul/U2,U2) is a r e g u l a r system of p a r a m e t e r s ~Z' at
the near points of x . If o n e of them is v e r y near to x , say w i t h
parameters (y,,ul,u~) , then it is q u a s i - o r d i n a r y and (f/ul,ul,u 2
has only one vertex with coordinates (I I + 12 - 1 , 1 2 ) .

Proof. a) If D1 is p e r m i s s i b l e and if we b l o w it up, then it is

very well known that there is at m o s t one near point x' to x and
that (y~ = Y/Ul,Ul,U 2) is a regular system of p a r a m e t e r s o f Z~ at
x' ( [IZ] , p.10?).We also know that A(f/ul;Ul,U2;Y') has only one
vertex v with coordinates (14 - 1,12 ) . Let v denote the vertex
of A ( f ; u l , u 2) and let

inv(f) = y~ + ~ ~ ( i ) Y ~ - I u ~ I U i~2
2 , ~(i) E k
i=I

where Y = inx(Y) , Uj = i n x ( U j) and k is the residual field at x

(For the definition of inv(f) see [9] ,(3.7)) . Then

~ ,~-i ,(I I -I) ,if^

inv~(f/u ) : ykm + [ ~(i) Y UI U2 z
i=I
th
where Y ' = in x , ( y ' ) , Uj ' = inx ,(u~)
J t h . As inv(f) is n o t a
power, also inx,(f/ul) is n o t a ~ p o w e r , so v' is n o t a s o l v a b l e
vertex of A(f/ul;Ul,U2;Y') (see [9] , (3.8) and (3.9)).

b) If n e i t h e r V ( y , u I) nor V ( y , u 2) is p e r m i s s i b l e , then
11 < I , 12 < I and 11 + 12 > I If we blow up x , it is v e r y well
known that the near points are in the two open sets 01 (resp. 0 2)
where uI (resp. u 2) is the equation of the exceptional divisor. Let
us w r i t e F = in6(f) (see [5], (6) or [6] , R e m a r k s on p. 27, for de-
finition) . Then, with obvious notation,

F = Y~ + ~ ~(i) Y U 1 IU 2 2 , e (i) 6 k
i=I

Suppose that x' is a n e a r point to x in 0 1 . Then at x t the

strict transform X' of X is d e f i n e d by

,~-i i(11+12-I) i12 A ) +

f/u~ : y,~ + ~ y uI [u~ (i uiB(i)]
i:I

where u~ = u2/u I and where A(i) 6 0Z,,x, has residue ~(i) in

the residue field at x' " If n o w u'2 is invertible at x' , then
ordx, (f/u~) < ~ since 0 ~ 11 + 12 - I < I , in c o n t r a d i c t i o n to the
assumption that x' is n e a r to x . Therefore (y, ,Ul,Uh) is a r e g u l a r
 82

system of parameters at xr and A(f/u ;Ul,U2;Y') has only one ver-

tex with coordinates (11 + 12 - 1'12) . Finally this polyhedron is
minimal because not both coordinates of the vertex can be integers
( [9] , (3.9.2))

PROPOSITION 2. Assume that x is a q u a s i - o r d i n a r y point of X

and let ( Y , U l , U 2) be a regular system of parameters satisfying (2)
Then the following algorithm stops: We blow up any one of the smallest
permissible ideals (Y,Ul) , (y,u 2) or ( Y , U l , U 2) and we do the same
in t h e successive transforms for any very near point of x and its
regular system of p a r a m e t e r s constructed in Proposition I.

Proof. Let (11,~2) be the coordinates of t h e vertex of ~(f;

U l , U 2) . Then by [5] , (4-2) or [6] , Remarks o n p. 27, we have
6(X,x) = 11 + 12 . For any very near point x' of x , Proposition I
Shows that 6 ( X ' , x r) < ~ ( X , x ) . Therefore the result follows from the
fact that 6(X,x) 6 (1/9!) • ~ ( [9] , (3.6)).

PROPOSITION 3. Assume that x is a quasi-ordinary point of X

and let ( Y , U l , U 2) be a regular system of p a r a m e t e r s satisfying (2)
If D is a n y equimultiple curve through x , then D = V ( y , u I) or
D = V ( y , u 2) . In particular D is r e g u l a r at x.

Proof. Let X' be the transform after blowing up x and let

D' denote the strict ~ansform of D . If D' = 0 , then it is e a s y
to see that D = V(y,ul) or D = V ( y , u 2) . If D ~ ~ 0 , then there
is a unique point x' on D' near to x , and by Proposition 2 we
may assume that x' is n o t very near to x . Since~ D' is a n e q u i -
multiple curve through x' , we conclude that dim Dx,(X') = I ( [12] ,
p.117). Furthermore, if (y',u~,uh). is t h e regular system of p a r a -
meters of Z ' at x' given by Proposition I, w e get that ~(f';u~,
J
u~;y') has only one vertex, with coordinates (11,1~) say, where

f' is the equation of X' in Z' . Now dim D , (X') = I implies

x •
I~ = 0 or l~ = 0 ( [12] , p.117)so we may assume I~ = 0 , 12i = I

Now after blowing up V(y',u~) there is n o m o r e point near to x' ,

which implies D' = V(y',u~) Projecting onto X we obtain D =V(y,u2).
Similarly, if 11' = I and 12' = 0 then D = V(y,ul)

PROPOSITION 4. Let ~(X) denote the highest multiplicity of

points on X , and assume that X satisfies:

(*) Every closed point x 6 Sam(X) with dLm Dx(X) = 2 is q u a s i -

ordinary.
 83

Consider any sequence of surfaces X(n) such that X(0) = X and

X(n) is a blowing up of X(n-1) centered at any closed subset of
Sam(X(n-1) . Then for n large enough we have

~(X(n)) < v(X)

Proof. Suppose we have made n blowing ups

X = X(0) + X(1) + X(2) ÷ ... ÷ X(n)

and ~(X(n)) = ~(X) . If x is a n isolated point of Sam(X) and

x(n) £ X(n) is a n y closed point near to x , then for the lexic~ographie
ordering we have

(dim D (X) , 6(X,x)) > (dim D (X(n) , 6 (X(n),x(n))

x x(n)

If x lies on an equimultiple curve D and if again x(n) £ X(n)

is any closed point near to x , then

(dim Dx(X) , 6(X,x)) h (dim Dx(n) (X(n)) , ~(X(n),x(n))) ,

and this inequality will be strict if D has been blown up d u r i n g the

algorithm. This allows to assign an invariant to x , which will drop
strictly (in t h e l e x i c o g r a p h i c order) in e v e r y step of our algorithm.
Namely, we take as components of our invariant the following numbers
(in t h e given order):

- number of components of dimension I of Sam(X);

- ~ 6(X,~) , where ~ are the generic points of one-dimensional

components of Sam(X) such that dim D~(X) = I ;

- sup{6(X, )Ix an isolated point of Sam(X) such that
dim D x(x) = I} ;
- number of isolated points of Sam(X) such that dim Dx(X) = I ;
- number of isolated points of Sam(X)

REMARK. As we can see, the algorithm is n o t canonical, since

there is obviously a choice at e a c h step.

II Jun~'s method locally (in c h a r a c t e r i s t i c 0 )

In this paragraph we assume that our given scheme Z is a k-scheme

for some algebraically closed field k of characteristic zero. In the
 84

first part of [18] , Zariski describes an algorithm by which one can

reach the following situation:

At every closed point x of Sam(X) there

is a transversal projection p :
(3)
Speck ~ul,u2~ ~S~ec 0x,x with a normal crossing

discriminant A such that I(Ared) c (ul,u2)

(where denotes the completion of 0X, x , of course.)

Xrx

PROPOSITION 4. If x is a closed point of Sam(X) satisfying

(3) , and if dim Dx(X) = 2 , then x is a q u a s i - o r d i n a r y point and
( Y , U l , U 2) satisfies (2) , where ( U l , U 2) is g i v e n by (3) and y
is c h o s e n such that A(f;ul,u2;Y) is m i n i m a l .

Proof. Let v be the multiplicity of X at x , let N = ~!

and consider the ramified covering S ~ S , where S = Spec k[Ul,U2] ,
S N = Speck [Xl,X 2 ] and x~-- = u I , x 7N = - u 2 . After applying Tschirn-
h^ a u s e n transformation to t h e equation f of X , f as an element of
0y, x ~ S k [ Xl,X 2 ~ can be w r i t t e n as

mI m2
f = U(y-r (Xl,X2)) , r = xI x2 h (Xl,X2) and at least

one h (0,0) ~ 0

Then I claim that A(f;ul,u2) has only one vertex of c o o r d i n a t e s

(XI,X 2) , where Xi = mi/N . First we note that

(4) A(f ;x1'x2;Y ) : (ml,m2) + ~ 2+

Now

~-i ' k
f = y + ~y a i , j , k u~ u 2 6 k I l y , u 1 ,u2]]

~-i jN x k N 6 k ~y,x 1,x2~

= y + ~y ai,j, k x I

Nj Nk
So, whenever ai,j, k ~ 0 , then ( ~- , -~- ) 6 A(f;xl,x2;Y)

2
and therefore (j/i,k/i) 6 ( 1 1 , 1 2 ) + 19+ , w h i c h implies

(5) A(f;u I • u2;Y ) c (X I r X2) + ]R 2+ •

 85

Since we h a v e applied Tschirnhausen transformation to f , we k n o w

that (f;ul,u2;Y) is m i n i m a l . So to p r o v e eguality in (5) , it
remains to s h o w that (11,12) 6 (f;ul,u2;Y) . Suppose not. Then we
know from [6] , Remarks on p. 27, that ~ ( f ; u l , u 2) > 11 + 12 So
whenever ai,j, k ~ 0 , then j + k > i(l I + A 2) and consequently

jN + kN > i(m I + m 2) ~ This would imply 6(f;xl,x2;Y) > i(m I + m 2)

in c o n t r a d i c t i o n to (4)

So we h a v e seen that condition (3) implies the condition (*)

of P r o p o s i t i o n 3, and t o g e t h e r with Propositions I and 2 this gives
an a l g o r i t h m for the d e s i n g u l a r i z a t i o n of surfaces over an a l g e b r a i -
ca~y closed field of c h a r a c t e r i s t i c zero. Moreover we note that once
condition (3) is s a t i s f i e d , this a ~ o r i t h m is the same as Zariski's
in [18]

EXAMPLE. Let

f y3 + 4 4 66 77 2
= 3yu~ u 2 + 2UlU 2 + U l U 2 ( U 1 + U 2 ) E { [Y,Ul,U2~

Then

2
A ( f ; u l , u 2) = (2,2) + ~+

The discriminant defined by the p r o j e c t i o n { [Ul,U2] c ~ { ~,Ul,U2~

is

4 4,3 6 6 7 7 2 )2
4(JUlU2J - 27(4UlU 2 + UlU2(U1+U2 )

13 13 2
= -27u I u 2 (u1+u2) (4 + U l U 2 ( U 1 + U ~ ) ) ,

and this discriminant does not h a v e normal crossings. This shows that
the converse of P r o p o s i t i o n 4 is false in g e n e r a l .

III Abhyankar's proof

In this paragraph, X is a r e d u c e d surface embedded in an ex-

cellent regular sc~me Z of d i m e n s i o n 3.

DEFINITION. We call bad points of X the closed points of Sam(X)

satisfying one of the following conditions:
 86

a) The point is isolated in Sam(X)

b) The point is a singular point of a one-dimensional component
of Sam(X)
c) The point is in the intersection of o n e - d i m e n s i o n a l components
of Sam(X)
d) If X = X(0) + X(1) + ... + X(n) is any sequence obtained
by successively blowing up any regular equimultiple curve
through the given points resp. near points to the given point,
then ~(X(i)) = ~(X) for all i .

Of course, a closed point of Sam(X) which is n o t bad is c a l l e d good.

PROPOSITION 5. There is o n l y a finite number of b a d points.

Proof. We have only to p r o v e that the points satisfying d)

are finite in n u m b e r . Take any one-dimensional component Y of Sam(X)
with genric point y . We apply the algorithm of blowing up first Y
and then closure of points near to y in t h e successive tran~orms of
X . From the theory of d e s i n g u l a r i z a t i o n of curves we know that this
algorithm is finite. Let n be the last step and X(n) the strict
transform of X . Since X(n) does not contain any point near to y ,
we see that above Y = {y} there is at m o s t a finite number of points
x' £ X(n) such that ~(x') = ~(X) The projection of these finitely
many points may be b a d points, but all other poin~of Y-Reg(Sam(X))
are good.

PROPOSITION 6. (Abhyankar) The following algorithm is finite:

We blow up every bad point of X and do the same for all the successive
transforms X(n) of X = X(0) as long as ~(X(n)) = v(X)

REMARK 6.1. Proposition 6 reduces the problem of resolution of

the singularities of X to the case where all closed points of Sam(X)
are good. In that particular case the following algorithm will be
finite: We b l o w up any curve of Sam(X) and do the same for the
successive transforms X(i) of X = X(0) as long as v(X(i)) = ~(X)
Furthermore, if X(n) is t h e last transform of X , then
~(X(n)) < v(X) . The proof of this Remark is c l e a r from the definition
of g o o d points.

We will give a proof of Proposition 6 based on o u r notion of a

quasi-ordinary singularity, thereby linking it to Zariski's proof.
The result will follow from two facts to be proven below:

a) By successively blowing up bad points with 2-dimensional

 87

directrix, a l l of t h e m w i l l eventually become quasi-ordinary, pro-

vided that they remain bad (Lemma 6.2).

b) By s u c c e s s i v e l y blowing up q u a s i - o r d i n a r y bad points, all of

them will eventually become good, provided that the multiplicity re-
mains unchanged (Lemma 6.3). *)

LEMMA 6.2. If w e apply the a l g o r i t h m of P r o p o s i t i o n 6 then, for

all large n , if x(n) £ X(n) is a n y b a d point such that
dimDx(n)(X(n)) = 2 , x(n) is q u a s i - o r d i n a r y .

For the proof we restrict ourselves to t h e m o s t difficult case,

namely that of p r i m e characteristic p such that ~(X) m 0(p)

REMARK 6.2.1. From [12] , p. I09 w e k n o w that for large n the

one dimensional components of Sam(X(n)) will be r e g u l a r and will
have normal crossings. Therefore, we make the additional assumption:

(**) The components of Sam(X) are regular and have normal crossings.

DEFINITION 6.2.2. Let x £ X be a bad point such that

d i m Dx(X) = 2 and assume that X satisfies (**) . Let

X(n) ~(n-1) X(n - I) , ... 7(I) , X(1) ~(0) , X(0) = X

be a s e q u e n c e obtained by blowing up a point x(i) £ X(i) near to

x at each step (x(0) = x) Let E. denote the strict transform
1,n
in X(n) of the e x c e p t i o n a l divisor of ~(i) • Let Y(0) denote the
one-dimensional components of Sam(X) (there are at m o s t 2) , and
for n > I let Y(n) denote the strict transform in X(n) of the
one-dimensional part of Sam(x) passing through x(n) (Note t h a t

~te from the editor: Here is a g a p for b a d p o i n t s with one dimen-

sional directrix, which may be filled in the following way: By P r o -
position 4 we know that the m u l t i p l i c i t y of such a point x will
strictly decrease by a suitable sequence of b l o w i n g up points and re-
gular equimultiple curves. Now we use a result of Z a r i s k i (Reduction
of t h e singularities of a l g e b r a i c three-dimensional varieties, Ann.
Math. 45 (1944) , 472 - 542) which is c a l l e d "dominant character of a
normal sequence" and which says that as l o n g as t h e m u l t i p l i c i t y does
not change, the o r d e r of b l o w i n g up a point and a curve may be i n t e r -
changed. Therefore, it w e f i r s t perform all b l o w i n g - u p s centered at
points, the point has become good since now multiplicity will be d e -
creased by blowing up curves only.
 88

either Y(n) : ~ or Y(n) is a r e g u l a r , irreducible curve for

n > I .) T h e n we define

~(x(n)) = 6 (X(n),x(n)) Y(n)-

~ 6(X(n),Y(n)) - i[[6(X(i)'x(i)) - 1]Ordx(n)Ei,n.

The picture below shows the characteristic polyhedron of x(n) in the

case that n > I and Y(n) % @ . In this case we may choose a regular
system of p a r a m e t e r s ( Y , U l , U 2) at x(n) (in Z(n) , the transform of
Z) , such that

I (u I) = I ( E n _ 1 , n) , (y,u 2) : I(Y(n)) and

A (f; Ul,U2; y) is m i n i m a l ,

f being the equation of X(n) ([12] , p. 28; see also the proof of
Lemma 6.2.3. below).

U,

~(f;ul,u 2 )

= 8(x(n))

12 (x (n)) ......... .
%

uh

11(x(n)) 6 (X(n) ,x(n) ) uI

Here 11(x(n)) : 8(X(n-1),x(n-1)) , 12(x(n)) : 6(X(n),Y(n)) and
: ~(x(n)) : 6(X(n),x(n)) - 11(x(n)) - 12(x(n)).

The proof of L e m m a 6.2. will consist in a d e s c r i p t i o n of the number

~(x(k)) under blowing up x(k) o To simplify notations, we restrict
the computations to the case that there is at m o s t one Ei, k passing
through x(k) , and we also assume that k > I . It w i l l become clear
that the other cases may be treated in e x a c t l y the same way. So we m a y
 89

assume that we can choose a regular system of p a r a m e t e r s ( Y , U l , U 2)

at x(k) with the following properties (Z(k) denotes the k-th
transform of Z ):

a) in x(k) (y) = I ( D x ( k ) (X(k))) c grx(k)0Z(k),x(k)

b) A(f;ul,u2;Y) is m i n i m a l , when f is t h e equation

of X (k) in Z (k)

c) If ordx(k) (Ei, k) % 0 for some i , then

(6)
I(Ei, k) = (u I) and
~(X(i),x(i)) - I : i n f { x l l (xl,x 2) 6 A(f;ul,u2;Y)} =
1 1 (x(i))

) If Y(k) % 0 , then I(Y(k)) = (y,u 2) and

6(X(k),Y(k)) = i n f { x 2 [ (Xl,X 2) E £ ( f ; u l , u 2) } =
12 (x (k))

(For k = 0 , the existence of such Y,Ul,U 2 is shown in [ 12] .)

We put 6 : 6(X(k),x(k)) , 11 = 11(x(i)) , 12 = 12(x(k)) and we note
that then

(7) ~(x(k)) = ~ - 11 - 12

( 11 = 0 if t h e r e is no Ei, k .) Z(k+1) is c o v e r e d by three affine

pieces. On the piece with exceptional divisor y = 0 there is no
point near to x(k) , so we assume that uI = 0 is the equation of
the exceptional divisor (again, the case u2 = 0 is similar)

Let K be the residue field at x(k) Then at x(k+1) we h a v e a re-

gular system of p a r a m e t e r (y',u1,~(1,u~)) , where y' = y/u I ,
u~ = u2/u I and where ~ ( U I , U 2) is the lifting of an irreducible homo-
geneous polynomial % ( U I , U 2) 6 K [ U I , U 2] (see [6] , Construction of
the parameters and Lemma 4).

Putting ~ = ~(X) , we m a y write

f = y9 + [yV-Ifi(ul,u2)
i:I
Then
F = in~(f) = Y~ + ~-IU 1 i ) u 2 8 (i) Q i ( U I , U 2 ) ,
i=I

where Y = in~(y) , Uj = in~(uj) , and Qi : 0 or Qi is h o m o g e n e o u s

of degree i~ - ~(i) - B ( i ) . We assume that Qi is d i v i s i b l e neither

by UI nor by U 2 : Lifting Qi to polynomials Qi with coeffi-
cients in 0 Z ( k ) , x ( k ) , we h a v e
 90

f fsu : + u hil
i=I

where h i 6 0Z(k),x(k) [ Y / U l , U 2 / U I] (see [6] , Lemma 4 ). N o w we

define

B(f',u1,~(1,u½),y' ) : inf{lordx(k+1)Q{1,u~)u~B(i)IQ i % O}

LEMMA 6.2.3. With the notations introduced above, we have

I
a) ~(x(k+1)) < e(x(k))
--
+- p .

b) If 6(X(k) , x(k)) 6 ~ and w(x(k+1)) < I , and if

Y(k+1) = ~ and no El,k+ I passes through x(k+1) , then x(k+1
is a good point.

c) If x(k+1) is a b a d point and not rational over x(k) ,

then

w(x(k+1)) < w(x(k))

d) The inequality of c) is strict except for the follow-

ing cases:

{ ~(x(k)) = 0 or
(x(k)) = I , B' : I and deg~ = p = 2 .

where B' = B(f',u1,~(1,u~),z') for a suitable choice of z,

Proof. If ( y ' , u l , u h) is a system of parameters, then it h a s

again the properties stated in (6) r and therefore,

w(x(k+1) < lord Q i (1'u'~)u'S(i)EL for every i such

-- i x(k+1)

that Qi % 0
and consequently

~(x(k+1) ! ~ - inf{~i(--i) IQi % O } - ~! G-i I - 12 : ~(x(k)).

So to p r o v e a) we may now assume that u 2' is invertible at x(k+1)

We write ~ = p sq with (p,q) = I , and we consider the following

conditions :

(A) 6 : 6(X(k),x(k)) ~ ~ ,

(B) Qi % 0 for some i , I < i ,< v _ - I

 91

(c) Qps = 0 ,
! I
(D) B = B(f',u1,~(1,u2),y ) ~ ~ •

The numbers (6-I,B) are the coordinates of the v e r t e x

v = v(f',u1,~(1,u~),y') of smallest abscissa of &(f';u1,~(1,u~);y')
I claim that if (A) , (B) , (C) or (D) is true, then v is not
solvable ( [9] , (3.8)). Indeed, if (A) or (D) holds, then the co-
ordinates of v are not integer (compare [9] , (3.9.2)), and if (B)
or (C) is s a t i s f i e d then the initial form inv(f' ~(1,u2),y,
)Ul, ,
is not a pS-th power (compare [9] ,(3.8) and (3.9)). Therefore, if
one of the conditions (A) - (D) is s a t i s f i e d , then

(8) w(x(k+1)) < B < inf{ ordx(k+1) (1,u 2) IQ i * O}

<inf {deg Qi/i-deg ¢IQ i * O}

< (6 ~(i)i B(i))/deg~i -< ~ ( x ( k ) ) / d e g ~ .

If m o r e o v e r deg~ > 2 (~ x(k+1) is not rational over x(k)) , we de-

duce

(9) ~ (x(k+1)) < ~ (k)/2

(pS) ~ (pS) S
Now assume for a m o m e n t that UI U Q p S ( U I , U 2) = P ( U I , U 2 )p
for some P 6 K [UI,U 2] . If we c h a n g e y to z = P ( U l , U 2) (P a
lifting of P ) , then the new parameters (Z,Ul,U 2) at x(k) still
satisfy (6) , and p u t t i n g Z = in6z we see t h a t the coefficient of
S
Z v-p in in~f' is zero, i.e. (C) is s a t i s f i e d for the new para-
meters. T h e r e f o r e , if none of the conditions (A) - (D) holds, we may
assume (after a suitable choice of our coordinates not affecting (6))
that

66~

Qi = 0 for all i such that pS~i , ISiSv

(E)
(pS)uB(pS) is not a pS-th power in
Qps(UI'U2)UI 2
K[UI,U 2 ]

If v = v ( f ' ; u 1 , ~ ( 1 , u 2') ;y') is still not solvable, then the inequali-

ties (8) and (9) hold. If v is s o l v a b l e , then we c h a n g e y' to
 92

z' = y' + a ~(1,ui)Su16-1 = y, + b u ~ - 1 , a 6 0Z(k+1),x(k+1) ,

and we put 8' = 8(f',u I,~(I,u 2)

T
,z') If v is the vertex of smallest
abscissa of A(f';u1,~(1,u½);z' ) , then

w
v 6 A(f';u1,~(1,u2)) (see [9] , 3.10))

and v' has c o o r d i n a t e s (6-1,8') (see [6] . We denote by b the

image of b in 0Z(k+1),x(k+1)/(ul,y ' ) ~ K[U~] (~(I,U2), (for this iso-

morphism se [6], C o n s t r u c t i o n of the parameters). Then we get

(pS) + ~ p S ) / p S
~(x(k+1)) < 8' < O r d x ( k + 1 ) (Qps(1,U 2)'U~

< ordx(k+1) ($(Q s ( I , U i ) U i B ( P S ) ) + I ) / p s ,

P

where 6 DerKP[u½P](K[U~]) . Now let r be maximal such that

u~(pS)u~ (ps)-WpS (UI,U2) = G pr for some G .

Then r < s If r = 0 then there exists 9 6 DerKP[U~(K[UI]) (either

= ~/~U~ or ~ is the e x t e n s i o n of some 8'6~er D(K) ) such that

K-

Q p s ( 1 ' U i ) U i 8(ps) =
U2~ (pS) "H otherwise.

Here H 6 K[U½] c K [ U 2 ] (~(I,U~)) and degu½(H) ~ d e g u ½ ( Q p s ) . Then

ord¢(1,U~ ) H < degu~ Qps(1,U½)< degQps(UI,U2)

If r > 0 then a computation similar to the one given above yields

s-r

ordx(k+1) ( H + bP ) <_ ordx(k+1) (H) + I

In any case we will have

(10) ~(x(k+1) < deg % s / p s • deg¢ + I/p s-r

< ~(x(k))/deg~ + I/p ,

 93

and this proves a) .

b). We have

(11) ~(X(k+1),x(k+1) = 6 - I + ~(x(k+1)) = integer + ~(x(k+1))

Let X" be the blowing up of X(k+1) at V(y',u~) . Then X" con-

tains at most one point x" which is near to x(k+1) , and the
characteristic polyhedron of x" (if it is near) will be
A(f';u~, ~(1,u~);y') translated by (- 1,0) ( [12] , p.127) . In par-
ticular ~(X",x")= 6(X(k+1),x(k+1~ - I . Now x" satisfies again the
assumptions of (6) and the condition (11) (where the pair (x(k),x(k+1~
is replaced by (x(k+1),x")) . So the assertion of b) follows by
descending induction on 6 .

c) In the cases (A) - (D) , the assertion follows from (8)

If (E) is satisfied and w(x(k+1~ < I , then x(k+1) is good by b)
The remaining case is e(x(k))/deg% + I/p h sup(~(x(k)),1) , in w h i c h
the result is clear.

If w(x(k)) = w(x(k+1)) = 0 , then these points are quasi-ordinary.

The case where ~(x(k+1)) = 8' = I will be studied in the next Lemma.

LEMMA 6.2.4. Assume that x(k+1) is bad and very near to x(k),
and that 6(X(k),x(k) 6 ~ . If, for some suitable choice of parameters
B(x(k+1)) = I , then ~(x(k+1)) = I . Let x" be any point of X(k+2)
near to x(k+1) . Then we have :

a) If x" 6 Ek+1,k+ 2 , then x" is quasi-ordinary or

dim Dx,,(X(k+2)) = I

b) If x" { Ek+1,k+ 2 , then either x" is good, or x" is

rational over x(k+1) and B(x") = w(x") = I

c) For i large enough, x(k+i) is quasi-ordinary.

Proof. By Lemma 6.2.3 , b) , it is clear that ~(x(k+1) = I

Now given x" , let Y(I,~) be a parameter at x" . If x" is ratio-
nal over x(k+1) , then by [6] , we have

2"~
B(f'/u I ,u1,Y(1,~)) < B (f' / u ~ , u 1 , ~ ( 1 , u ~ ) )

Therefore x" is good by Lemma 6.2.3 , b) . Assume now that

x" 6 Ek+l,k+ 2 is very near to x(k+1) . Then

 94

~(x") < 6- I + B(x(k+1)) I - (6(X(k+1),x(k+1)) - I)

= 6- I + B(x(k+1)) - 6(X(k+1),x(k+1)) = 0 ,

and therefore a) follows. In the situation of b) we know that

2~ ~ ,~(I u½)) ,
8 (f'/u I ,ui,~(I,~)) ! B(f'/ul,u I ,

and therefore either x" is good or it satisfies the same hypothesis

as x(k+1) . This proves b) and also c) , since the sequence of
~'s is not constant ( [12] )

LEMMA 6.2.5. With the hypotheses and notation of Lemma 6.2 ,

assume the sequence

x(k) ÷I x(k+1) ~ ... ~ x(k+n) , n > I

satisfies

w(x(k)) > 0 ,

l ~(x(k+1)
x(k+i) is
> ,~((x(k))
very near
for
to x(k)
0 < i <_ n
and bad
,
for 0 < i < n

Then this sequence is finite.

Proof. If x(k+i) ~ Ek+i_1,k+ i for all i , I < i < n , then

Hironaka's proof in [12] together with [6] implies that the sequence
w(x(k+j)) is constant by (8). From [12], we know that

w(x(k+i)) < 6(x(k+i-1)) - w(x(k+~-2) - 12(x(k+{-1)) ,

and by (8) we have

B(x(k+i-1)) < w(x(k+i-2)) + 12(x(k+i-1)) ,

and therefore w(x(k+i)) = 0 $ w(x(k)) . So we may now assume that (E)

is satisfied at x(k+j) for some j < i , and we choose j minimal for
this property. Then B(x(k+j+1)) ~ w(x(k)) + I/p by the proof of (10).
If now w(x(k)) s I/p then B(x(k+j÷1)) ~ 2 / p ~I , and the assertion
follows from Lemma 6.2.4 in the case B(x(k+j+1)) = I , while
B(x(k+j+1)) < I implies that x(k+j+1) is a good point. Assume finally
that w(x(k))> I/p . Then the result follows from the inequality
 95

I
~(x(k+1)) $ B(x(k+i-1)) - w(x(k)) $--
P
([12] , together with Lemma 6.2.3).

In t h e following Lemma, we denote by {z} the fractionary part

of a real number z .

Lemma 6.3. With the notations of Proposition I, a s s u m e that

x (k) 6 X (k) is a q u a s i - o r d i n a r y bad point (with dimx(k)X(k) = 2), and
let (11,12 ) be t h e vertex of the polyhedron A ( f ; u l , u 2) . Then

a) {,\1 } + {I 2 } >_ ]

b) After blowing up x(k) there are at m o s t two bad points very

near to x(k) , and these points are quasi-ordinary. The coordinates
of the vertices of the respective polyhedra are (l~,l~) (11+12-1,12)
and (11,l~) = (11,11+12-I)

Proof: a) Assume the contrary, i.e. {I I } + {I 2} < I . T h e n one of

the l's must be I since 6(X(k) ,x(k)) > I . T h e r e f o r e one of
V(y,ul) and V(y,u2) is p e r m i s s i b l e , and this is t h e only permissible
curve through x(k) by Proposition 3. If we blow up t h i s curve, Pro-
position I tells us that there is a t m o s t one point very near to x(k).
This point is a g a i n quasi-ordinary and { 1 1 } , { I 2} w i l l not change,
whereas 6 drops. S o we get the result by descending induction on 6

b) Assume first that {k I} + {I 2 } > I . T h e n

6(X(k),x(k)) = 11 + 12 { IN (case (A) in L e m m a 6.2.3) , and there-
fore ~ is zero at e v e r y point very near to x(k) . If u2/u I is in-
vertible at such a point, then the point is g o o d by a) . The remaining
points have local coordinates (y/u 1,u 1 , u 2 / u I) or (y/u 2 , u I / u 2 , u 2) ,
and for these the assertion is c l e a r . By a) it r e m a i n s to c o n s i d e r the
case {I I } + {I 2 } = I , w h i c h implies 1 1 + 12 = 6(X(k~x(k))6 I~ . Let x'
be a point very near to x(k) If u2/u I is invertible at x' , then
I I
u(x') < w(x(k)) + - by (10) But this means that x' is g o o d ,
-- p P
since it satisfies the hypotheses of part b) of Lemma 6.2.3. If x'
has local parameters (Y/Ul,Ul,U2/Ul) or ( Y / U 2 , U l / U 2 , U 2) , then it
is a g a i n quasi-ordinary, and so b y Lemma 6.2.3., b) it m u s t even be
a good point.

CONCLUSION. Using the notation of Lemma 6.3 w e have

{X~} + {k~} < {i I } + {k 2} and {k~} + {I~} < {I I } + {I 2 }

 96

Therefore, by Lemma 6.2 and 6.3, the algorithm of 6.1 will lead to a
situation where dim X < 1 for every bad point x of X . In t h i s
x --
case, the proof of Proposition 6 follows from [12] , p. 114.

IV Hironaka's method

We will use the assumptions and notations of III. Hironaka answers

the question by looking at the deformations of the characteristic poly-
hedra. There are three steps in h i s proof (given in [6] and [12] ):

Step 1. We can reduce the problem to the case where all points
in Sam(X) are isolated ( [12] , p. I09).

Step 2. The difficult case is t h e one in w h i c h the dimension of

the directrix of X at x is 2 .

Step 3. Let (~,B) be the vertex of smallest abscissa of

A(f;ul,u2) and choose Ul,U 2 so that B is m i n i m a l . Then B will

drop along the sequence

X = X(0) + X(1) + ... ÷ X(n) .....

where X(n) is o b t a i n e d from X(n-1) by b l o w i n g up Sam(X(n-1))

 97

REFERENCES.

[1] S. ABHYANKAR, Local u n i f o r m i z a t i o n on a l g e b r a i c surfaces

over g r o u n d fields of c h a r a c t e r i s t i c p # 0 . Ann. of Maths.,
63 (1956), p. 491-526.

[2] S. ABHYANKAR, On the r a m i f i c a t i o n of a l g e b r a i c functions, Ann.

of Maths., 77 (1955), p. 575-592.

[3] S. ABHYANKAR, R e d u c t i o n to m u l t i p l i c i t y less than p in a

p - c y c l i c e x t e n s i o n of a two d i m e n s i o n a l regular ring, Math.
Ann. 154 (1964), 28-55

[4] S. ABHYANKAR, N o n s p l i t t i n g of v a l u a t i o n s in e x t e n s i o n s of two

d i m e n s i o n a l regular local domains, Math. Ann. 170 (1967),
87-144.

[5] V. COSSART, Une n o u v e l l e d6finition de l ' i n v a r i a n t ~ ~ Un-

published.

[6] V. COSSART, D e s i n g u l a r i z a t i o n of e m b e d d e d excellent surfaces.

T o h o k u Math. Journ. 33 (1981), 25-33.

[7] J. GIRAUD, Etude locale des s i n g u l a r i t @ s r Orsay, Pub. no. 26

(1972)

[8] J. GIRAUD, Desingularization in low d i m e n s i o n T Lecture 2 in this

volume.

[9] H. HIRONAKA, C h a r a c t e r i s t i c p o l y h e d r a of s i n g u l a r i t i e s r J. Math.

Kyoto University, V. 7 (1967), 251-293.

[ i cI] H. HIRONAKA, B i m e r o m o r p h ~ smoothing of a complex analytic space r

U n i v e r s i t y of W a r w i c k (1967).

[11] H. HIRONAKA, I n t r o d u c t i o n to the theory of i n f i n i t e l y near

singular points~ M e m o r i a s de M a t e m a t i c a del I n s t i t u t o "Jorge
Juan" 28.

[12] H. HIRONAKA, D e s i n g u l a r i z a t i o n of e x c e l l e n t surfaces. A d v a n c e d

Science S e m i n a r in A l g e b r a i c G e o m e t r y (summer 1967). Notes
by B.Bennett, B o w d o i n College, r e p r i n t e d in the a p p e n d i x of
this volume.

[13] H. HIRONAKA, J.M. AROCA, J.U. VICENTE, The theory of the m a x i -

mal contact. M e m o r i a s de M a t e m a t i c a del I n s t i t u t o "Jorge Juan"
29.

[14] H. HIRONAKA, D e s i n g u l a r i z a t i o n Theorems. Memorias de M a t e m a t i c a

des I n s t i t u t o "Jorge Juan" 30.

[15] H.W.E. JUNG, D a r s t e l l u n g der F u n k t i o n e n eines a l g e b r a i s c h e n

K6rpers zweier u n a b h ~ n g i g e n V e r i n d e r l i c h e n in der U m g e b u n g einer
Stelle. J. Reine Angew. Math. 133 (1908), 289-314.

[16] J. LIPMAN, D e s i n g u l a r i z a t i o n of t w o - d i m e n s i o n a l schemes.

Ann. of Math. 107 (!978).
 98

[17] T. SANCHEZ, Teoria de s i n g u l a r i d a d e s de superficies a l g e b r o i d e s

sumergidas. M o n o g r a f i a s y m e m o r i a s de Matematica. IX, Pub del
I n s t i t u t o "Jorge Juan" de M a t e m a t i c a s . M a d r i d (1976).

[18] O. ZARISKI, E x c e p t i o n a l s i n g u l a r i t i e s of an a l g e b r o i d surface

and their reduction. Atti. Acad. Naz. dei Lincei, serie VIII,
vol. XLIII (1967), 135-146.
 APPENDIX

Desingularization of Excellent Surfaces

Heisuke Hironaka
Advanced Science Seminar in Algebraic Geometry
Bowdoin College, Summer 1967

Notes by Bruce Bennett

I . Fundamental Concepts

1. Excellence
Def. A scheme X is excellent if
(I) X is noetberian

(2) ~/:X' > X of finite type, Sin~ (X I) is

closed

(Sing (X) = ~x c X I ~ , x is not a regular local

ring) })
(3) ~ X" -fJ > X: g-> X both of finite type,

V x' ~ X' , if R = e'X, x, , let E"= Spec Rxx.X" .

(deduced from Spec R --> X s via base extension
by f)

and let b be the projection h:~" --> X" .

Thus: X"xx:Spec R = ~" h > X"

Spec R > X

Then h has tbe property'. Sing(~")= b'~(Sing(X")).

Remarks: Every scheme of finite type over an excellent

scheme is excellent, and in particular

Every closed subscheme of an excellent scheme is
excellent.
 100

Spec (any complete local ring) is excellent (a

theorem of Nagata and Grotbendieck, E.G.A.IV).
Any finite type scheme over ~ (= integers) is

excellent (Nagata).

2. TanRent Spaces and Cones

X ascheme, x ~ X . dim (X) = Krull dim ~X • x' d ~ = M a xx~X

(d~x(X) )

L6t ~ = ~ X,z, ~ = @~X~x.x(x~ = ~ / ~ , and define:

Tx(X) = Zariski tangent space to x at z = Spec ( s x ( x ) ( ~ / y ~ 2 ) ) °

(sx(x) (-) denotes symmetric algebra over ~(xi) •

Cx(X) = Tangential cone of X at x = Spec (Gr x(X))

Note: Cx(x) c__~ Tx(x ) via the canonical surjection

s~(x)( ~ / ~ ~I - > Gr~ (~) - > o.

~ x(X) = Strict. Tangent Space of X and X = maxi~ctm linear

subspace T of Tx(X) (passing through origin) such tbat

equivalently:
(i) Cx(X)= Cx(X)+ T (+ denotes addition of points in

(ii) Cx(X) <--~ T, and Cx(X) Z T x S • where

S C~.> Cx(X ) is some closed subspace.

Remark: The existence of ~x(X) is proved by showing that

if tu.~o linear subspaces T and T of Tx(X) bare property (i)
2

or (ii) 3 then so does T + T .

I 2

Observe x is a regular point of X < > Tx(X) = Cx(X) = (X).

Now suppose x ~ X r ~ Z. R = ~,x' ~ = ~•x,

M (resp. ~ I = m=~ iaoal o~ R (resp ~ . k = ~/~ = ~/~

J = ideal of X in Z at x, so that we have
 101

0 > J ---> R ----> ~ ,, > 0 .

We obtain:

0 ~> Gr x(X,z) ~> Gr x(Z) > Gr x(X) ....> 0

II II II
o > ~(J,R) > ~ = ~ o ~ / ~ ~+~ %~o~/~ ~ > o
where InM(J,R) may be described as
(i) the ideal of Or M(R) (= Grx(Z)) generated by

all InM(f) , f 8 J, where InM(f) = f(mod M ~u(f)) ,

where VM(f) = ~ i g h e s t power of M containing f if f ~ 0
~o if f = O

or (ii) the homogeneous ideal of GrM(R) whose v th piece is:

J R Mv + Mv+~/Mv+~
In this situation we may describe ~x(X) as follows:
minimum k submodule T of GrIM(R) such that(if kiT] denotes

Sk(~) ~--> GrNCR)) ,

(arx(x,z) n kiT]) Gr (Z) = ~rx(X,Z)
(i.e. I~(J,R) may be generated by Forms in kiT]).

Then ~z(X) -~ Spec (Grx(Z)/T Grx(Z)) in tbe sense that:

~x(X) C > Cx(X ) ~ > Tx(X ) C > Tx(Z ) .

(Note that Grx(Z) is a polynomial ring since R is regular.

kiT] is a polynomial subring and dim ~x(X) = dim Tx(Z)-dimkT)

Example: If Z = a%~, X C L > Z a curve, then ~ x ( X ) = a

point (as in the case of a node), or a line (as in the case

of a cusp). More precisely, if X is locally defined in

~2k(k = ~) by f, where x = (0,0), write f as a power series

f = f~ + f + (v = ord f)
V
wbeme fv =i=Hl (~iy'~i x) Gi" ~i s k. Then ~x(X) is a
 102

line <---> ali the ratios ~il ~ i a r e the same. (In fact, the

initial form of ~ at the orig~n is fu).

Terminology. x g ~,v a scheme, x is a re aular Point (or a

non-singular point or a simple point) of X if Ov is a

regular local ring. Otherwise, x is called a singular point

of X.

3- Normal Flatness

X an excellent scheme D ~ X closed, reduced,

irmeducible subscheme. Define

N(D,X) = T b e Normal . cone of x alon~ D = a family of cones

psmametrized by D = Spec (GrD(X)).

V
where QrD(X) = ideal of D i x),

viewed as a sheaf of g~aded 0D- algebras.

Definition: X <. "~ D ~ z. X i s..normall 7 flat alon S D.,at. x

if D is non-singular and Gr~(X) is locally i~ree at x V v.

X is normally flat along D if it is so at every point of D.

Note: It follows from general facts about flatness and the

definition of excellence that {x e DIX is normally flat along

D at x} is open in D.

x g D, can define Nx(D,X) = N(D,X)~DSpec ~(x)

= the fibre above x in the fibre system N(D,X).

Theorem. x s D ~---~ X, D nonsingular. Then the

following are equivalent:

i) X is normally flat along D at x.

2) cx(x) = ~ (O) x~x(D,X) (i.e., C (X)/~x(O) = N (D,X))

a~d Tx(D ) C---> ~ x ( X )

 103

Note that we always have canonical maps

Tx(D) > Cx(X) ......> Nx(D,X)

corresponding to:
sur}ective
< . . . . . < - -

3) If we suppose X is embedded in a non-singular

excellent scheme Z, then !) and 2) are equivalent to:

hypersurfaces H ,...,H m in Z (locally at x) such that

I

a) Hi ~ -X

b) ~ C (H i) = Cx(X)

(~ ~ H. = X at z, strict implication).
I

c) ~ i,H i has the same multiplicity at x as it has

at the generic point of D, (where both are re-

garded as points of Z).

c) means: if f i d e f i n e s Hi, (R,M) = ~ , x P ~ > R defines D

at x (P a regular prime in R), then vM(f i) = Vp(f i) ~/i.

~. Monoidal T~ansformations

X a scheme, D c > X closed subscheme, I ~-->

the ideal sheaf of D on X. Let X I = Pro~ ~) , where

v~o IV is viewed as a sheaf of graded ~ - a l g e b r a s . There is a

natttwal projection h-~X i .._> X induced by ~ -~-~> I ° ~----> ~ I v .

~X!,h) is called the mon oidal transform (or "blowing u~") of

X with center D.
, , ,

Remarks:

I) Let spec A ~" > X be a~ open affine, (go'''''gn)

a set of generators of l]Spec A. Then above Spec A, X t may
be described as
n

i~o Spec (A[go/g i .... , gn/gi])

 104

where A[go/gi, ...,gn/g i] is viewed as an A-subalgebra of Agl,

and h is induced by the natural maps

A ----> A[go/g i .... , gn/g i] -

2) i~XI is an invertible sheaf, generated on an affine

of type Spec (A[go/g i .... ,gn/g i]) (over Spec A~...,-> X) by gi"
Moreover (X:,h) is universal attracting for X-schemes Y
Y If > X s t I ~y is an invertible sheaf.

3) b'~(D) = X' x xD = Pro] (~=o iV/IV+l) = P(N(D,X)).

where we use the notation:

cone = Spec (graded algebra)

P(cone) = Proj (same graded algebra).
Similarly, i2 x ~ D, h-1(x) = P(Nx(D,X)).
b
It is easily shown that h induces an isamorphism X ~ - h - ~ > X - D .

~) i~ D = {x), a point, h-l(~) = P(Cx(X))=P(Nx(x~X)).

Therefore if x is non-singular, h a(x) = P(Tx(X)} = (x)

where m = dimx(X).

%) Suppose D g" > X t" > Z closed subschemes, and

consider the monoidal transform g:Z: ~ > Z of Z with
center D. The strict .transform of X by ~ is by definition

the smallest closed subscheme X: . > Z I which induces g-~(X-D)

on Z:-g-~(D). (i.e. such that Xt-g-I(D) ~ ~ X - D). Then
if h = g]X:, h : X t ~ > X is the monoidal transform of X with
center D.

5. Statement of the Problem.

, , , , , ,

Resolution of singularities of excellent schemes may be

stated in the form of the following
 105

Conjecture: X an excellent scheme. Then ~ a finite succes-

sion of monoidal transfomnations

f fo
n-l> X ,, > . > X > X = X
Xn n-l "" l o
such that O T h e center of fi is D. ~ , > X.) where
i i
X i is normally flat along D..
i
X n is empty (which is a slick way o£ saying that

(Xn_1)re d is non-singular).

Tbe Conjecture is kno~rn when l) X is char. 0 (i.e. when

char x(x) = 0 B/ x c X) and dim X is arbitrary (See Hironakals

Annals paper of Marcb 1964), and 2) dim X < 2 and char X

is arbitrary. The latter fact is the subject of the present

lectures, and we give a proof in the special case when X

is assumed to be embedded in a non-singular excellent scheme Z

of dimension 3, in other words, when X is locally a hyper-

surface in Z. We suppose in addition that Z is of finite

type over k = ~.

These additional hypotheses permit a considerable

lighteniu~ of the terminology (in particular the )'measure"

o~ the badness of a singularity becomes very simple) and tbe

necessity ot introducing technicalities to solve problems of

a purely algebraic nature is avoided. The essential character

of the proof remains unchanged, however, and in fact should

be more clearly evident.

No reference for the general proof is available at the

present time, but it probably will be published shortly.

We will need to assume the resqlution of singularities

of excellent curves (i.e. excellent schemes of dimension l)

in the form o£ the Con~ectur.e.

 106

II. Pro o~ of Resolution for Surfaces (in a Special Case,) `

I. Basic structure of the Proof.

The situation is: Z a 3-dimensional non-singular

algebraic scheme over k = ~, X ~ > Z a reduced surface.

Wewant to resolve the singularities of X in the sense of the

Con ~e cture.

Let x c X. R = ~,x ' M = maximal ideal of R, (f)R the

ideal of X in Z at x. Define Vx(X) = the multiplicity of X

at x = v M (f) = order of f at x.

Note that x is non-singular <~---~ Vx(X) = I <~---> f is a

regular parameter in R. Let

v = v(x) = ~ {vxCX)}
xaX
Remark. In general, if ~ is any integer, define

Sing ~(X) = {x e X] Vx(X) >_ ~)

Sing ,(X) = Sing v(X) = {xaXlVx(X) = v}

="maximal singular locus"
It is obvious that Sing~(X) _° Sing~+~(X), and we have the

following ~eneral fact:

X an excellent scheme, ~ an integer, then Sing.(X) is a

proper, closed subscheme of X. (Here Vx(X) , the multi-

plicity of OX, x, cannot in general be interpreted as the

order of a single element. The above fact is just the

assertion of the "upper semicontinuity" of multiplicity of

local rings of points on an excellent scheme).

In our case we can sketch a particularly simple proof

of the closedness of Sing~(X). Namely if char k = O,

vx(X) = ~ <--->7 a (pure) differential operator D on Z,

D = ~_ (~/~xi)a~ of order ~ , a i = ~ , and (Dr) x is a unit,

 107

bencB also a unit in some neighborhood U of x, so

Vy(X) ~ ~ V Y s U. (If char. k = p, use a "Hasse differ-

entiation" instead of an ordinary one) I.

Now consider a non-sin~ular closed subscbeme D ~-> Z

contained in Sing v(X). Thus D is a point or a non-singular

curve lying in X. Apply the monoidal transformations with

center D:
XI ~ .... > Z I

D ¢---> X C_ _~- > Z

In general, a monoidal tranformation is said to be permissible

if tbe scheme in question is normally flat along the center

of the transformation. In our case (since X is a hypersurface

in Z, and all the points of D have the same multiplicity as

points of X - namely v) condition (3) of the theorem cn

Normal Flatness, p. (4), implies that h is permissible.

Therefore
xCD) r > C T Z))
In fact, Tx(D) is a linear factor of Cx(X) (by condition (2)

of the above cited theorem, which asserts: X normally flat

along D < ~ > Cx(X) --'> Nx(X,D) x Tx(D)) , and D~x(X) is the

largest linear factor.

The following fact, essentially due to Zariski, and

which we will call Idea A, is of crucial importance in the

resolution. It states:

i. Differentials need not play a role in this type of

result. See for example the much more general result in
Hironaka:s Annals paper: T~an. l, Chap. III {3}, p.218, which
uses entirely different methods.
 108

Let X be. an excellent scheme~ D(---> X closed subscbeme.~ '

h:X T > X the monoidal transform ~ith center D~permissible.

Let x s D. xtc X:such that h(xl) = x. Then the sin~ularit ~

of x: is no worse than the singularity of x, and if the

singularity of x~is "as ba d as the sin~ul.arit~ of x"~ then

x' is contained in P ( ~ x ( X ) / T x ( D ) ) .

Remarks. l) We will not make precise at tbls time the

general meaning of "as bad as" and "no worse than". However,

in our situation, they mean Wx,(X') = Vx(X) and Vx,(~) ~ Vx(X)

respectively. Thus one consequence of Idea A is v(X I) ~ v(X).

2) P(~x(X)/Tx(D)) is viewed as a closed subscheme

of P(Nx(D,X)) ( = h-1(x) by Remark (3), P.5). In fact,

P(~x(X)/Tx(D))makes sense since Tx(D) ~ > ~x(X) as we

have Just seen. Moreover
by no~.~nal
Cx(X) = C z ~x(X) Nx(X,D) x Tx(D)
flatness
> C~(~'~x(X)/Tx(D)) = Nx(X,D) ,
so can view P(~x(X)/Tx(D)) ~----~ Nx(X,D).

3) A proof of Idea A in our special case will be given

in an appendix.
We can now list some i~mediate consequences of Idea A

in our situation h:X I • > X, center D c-> Sing v(X), h

permissible. (Terminolo6v: An x s Sing v(X) is called a

v-fold point. A curve contained in Sing v(X) is mlled a

v-fold curve. Let x e D, a v-fold point, e = dim ~ x ( X ) ,

d = dim D. Note that 0 ~ d ~_ i~ and 0 m e ~ 2 (since the

smallest vector space T such that in~ I (f) s kET] must have
at least dimension l, and dim ~y~x(X) = 3 - dim T in our

case. See page (3)). Note also that since Tx(D)~-->~x(X),

d~e.
,,,, J
 109

Case i. d = i. Then e = I or 2

a) e = I. Then dim ~ x ( X ) / T x ( D ) = 0, so

P(~x(X)/Tx(D)) is empty, so there is no v-fold

point in X I lying above x (by Idea A).

b) e -2. One can have at most one v-fold point x I

lying above x, and if there is, it is the point P ( ~ x ( X ) / T x ( D ) ) .

Case 2. d = 0. Then e = 0, l, or 2.

a) e = 0. N_~o v-fold points above x.

b) e = 1 at most one v-fold point above x, namely

PC SZx(x)ITxCD)).
c) e = 2 Then either there exists a finite number

of v-fold points above x (possibly none), or if there exists

a v-Zold curve, it must be precisely P ( ~ x ( X ) I T x ( D ) ) , which

is a line (in h'1(x)), isomorphic to l~(x).

~otice that if •
~ a ~-fold curve E I on X v
, , |
m a p p i n g onto
, ,

D by h ; then E t is unique and nop-sin~ular. Namely, if D is

a point, E t must be a specificSP ~ by Case 2c). And if D

is a curve then E t must be isomorphic to D by ~4 i.e., a

uniquely possible point in the fibre of every point of D,

(by case Ib).

W.e now give an outline of the Resolution procedure:

X ~ I- > Z 3 non-singular, v = v(X).

Step I. If Sing v(X) contains a singular irreducible curve,

apply quadratic transformations to each of its singular

points. W e end up with: Sing v(X) does not contain any

singular irreducible curve.

Here we use the resolution of singularities of curves

by monoidal transformations (necessarily quadratic). We

 110

of course view the transform of the curve as contained in the

transform of X. After a finite number of steps we resolve

the singularities of the curve. Need to note that we can

create no new irreducible singular ~Ufold curves in this

process. In fact, any v-fold curve lying above a blown up

point must be a certain I~, as we have seen; and if this I>:

is not v-fold, any v-fold point x ~ lying above x is either

isolated in Sing v(XW), or lies on the strict transform D I of

of some curve D in Sing v(X) (and it is well known that x t

is not a singularity of D I if x is not a singularity of D).

Note: A quadratic transformation is always permissible.

Step I I. If ~ a non-singular irreducible cur~e r in

Sing v(X), apply monoidal transformation with center r. By

induction applied to Spec (local ring at the generic point

P of r in X) we come to: Sing v(X) is a finite number of

points.

The point is that Spec (@X,p) has dimension one, so can

resolve it by permissible monoidal transformations, which we

may view as base extensions of transformations beginning with

X. Thus:

permissible ~ ~ permissible
(Spec ~X,p)'---~ X'
blowup
closed point
[ ~ blowupF
permissible since
Fc'~Sing~(X)

Spec &X~P > X

In this context, to say that we can resolve the singularities

of Spec ~ , p means that eventually we obtain a curve

above F which has non-singular generic points (in particular,

 111

points of multiplicity I), so by the upper semicontinuity of

multiplicity, only a finite number of points of ~ can be

v-fold.

Step III. After elirCunating v-fold curves as in Step I!,

apply quadratic transformations to each point in Sing v(X).

By doing this we m a y create a new v-fold curve. If so,

apply Step II, and if not apply Step III again. ~ .

We thus obtain a sequence of permissible monoidal transform-

ations:
h(~- ~) h=h(
.... > X(~1 ) XCm -i) > .... > X(~) " " o)> X
Then: Theorem. For some m , v(X(~ )) < v(X) = v.

The proof of this theorem is of course the central problem.

To prove it, we m a y start with a situation in which there

are only isolated ~-fold points (i.e. start with Step I!I),

in other words assume tbe v-fold locus is discrete. The

fact that v-fold curves m a y be subsequently created is not

crucial. The proof is done by contradiction.

We will start with: x ~ X, isolated in Sing v(X),

and distinguish two cases: e = 1 and e = 2. (e = d i m ~ x ( X ) ) .

Note that the case e = 0 is trivial, for then, by Idea A,

there can be no v-fold points above x. The case e = 2 is

more difficult, this is the "cu3p-type situation" in which

2-dimensional phenomena seem to be especially exhibited.

2. Tbe Case e = I.

Suppose we have an infinite sequence of permissible monoidal

tranformations:
 112

--> ...... > Z G > Z (3) non-singular

=~ > . . . . . > X
j" J
~----~> X (~)
1

x 8 X, an isolated v-fold point (1) with e = dim ~ x ( X ) = i.

We show by contradiction that

(~ At some finite stare there are no v-fold points lying

above x.

Let R = ~ , x ' N =~Z,x' R/(f)R = ~ , x " In

R = k[[y,u,t]] we may assume f = fv + ~ where v m (~) > v and
fv is a form of degree exactly v in y and u. Denote

= InN(f) = InM(fv). (In other words, y and u are a basis

for the vector space T defining the strict tangent space
~/x(X) (See Page (3)) and t extends this basis to a system

of parameters of R).

If there is no v-fold point lying above x we are done.

Assume, then, that x I c X is such a point. Then Idea A

asserts t h a t it is the unique point of the fibre at which

[/t = u/t = .O. In fact, since Tz(D) = 0 (D = {x}),

Idea A = > x s P(~x(X)) C > b-1(x). And in the blowing

up of Z with center x (in which the situation is embedded)

g-l(x) = Proj (k[y,u,t]) and P ( ~ x ( X ) ) = Proj(k[y,u,t]/(y,u ))

= the origin in Spec (k[Y/t,u/t]). Moreover if R'= ~ , x ! ,

then (Y/t' u/t,t) is a regular system of parameters in~R s .

(Let N: = max. (RS)).

Fact: X is defined in Z at x: by (f:)R:, f: = f/t v (2)

i 1

(1). ~!e are assuming that the sequence of monoidal trans-

formations corresponds to the procedure of steps I, II, III,
and that we sine starting at a Step III. In particular, we
may assume h is the quadratic trans£oz~nation with center x.
(2). This is a general fact about monoidal tranformations
o~ regular local rings whose center is a regular prime ideal.
See [1], chapter III, ~2, P 216.
 113

Thus ~t = £v/t v + ~/tV ' where fv/t v is a form in Y/t and

u/t , of degree v, and ~/t v is a power series in the new
parameters Y/t' u/t' and t, divisible by t! l)

Remark: Suppose we could conclude that:

'either (a) ~,(T'~ (= Ordy/t,u/t ,t(f~.) V

I or

.or
(b)~,(~') < v

(c)dim ~ x I ( X I) = 0
Then if (b) or (c), we are done, because (c) ~-> no v-fold
(7 1

points can lle above x I at any subsequent stage of the

sequence (Idea A) , and (b) --> ~M'(fl) = VM,(~I) < v,

i.e. Vxt(Xt) < v. And if (a), then we are in a situation

that is exactly the same as the original (i.e. Vx:(Xl) = v,

e = l, and In M I(~) is a form in yl and u ! (new parameters

are yt = Y/t' ul = u/t' tt = t.) Suppose then, that (a)

recurs infinitely many times, i.e., there is an infinite
sequence of points x(m)¢ X , each lying over the preceding

one, and such that the equation defining X(m ) in Z(m ) at

x (m) is:

f(m) = fv(Y/tm, u/tm ) + ~(m)

but f(m) = f/tVm , and we may assume ~(m) involves no de-
nominators in t larger than t (~-m). (See Footnote (I)).
Hence, multiplying by t vm, obtain:

f c (y,u)V + (tl~R.
And since this is true ~ m, f e (y,u)VR, so original x lies
on a ~-Zold curve defined by y = u = O. Contradiction.

a i ° •
(I) The point is that any re.Tin of type t y uJ/+v !a+i~j>~.a~_i)
may be ~ i t t e n in the form t a (y/~)i(u/+)J wher~ a,=a-~+i~j .
Moreover every-such term efZectiv~ly involves a power of t no
greater than v-1 in the denominator.
 114

At this point, we notice that:

I) There is a strong intuitive appeal for the truth

of ( . ) .
2) (*) is not true.

The point is that if neither possibilities a) or b) hold,

then something in ~:must contribute to the initial form of f l

but everything in ~I involves t, so a p~ori one needs all
three parameters to express the initial form (since both

yl and u: are involved in fv/t v). Hence one would like to

conclude ~x:(3~) = 0. The difficulty lies in the possibility

that there exists a new parametrization such that the initial

form does not involve all three new parameters. In the proof
(which follows), the concept of "T-preparation" is introduced
to handle this possibility.
Now wlth assumptions and notations .as above, we prove the
assertion (g) (p. 15).

We have a regular system of parameters (y,u,t) of R=~Z, x

such that if y, u, t are their initial forms, then

~x(z) = Spec Grx(z) = Spec k[~,~,~](k = ~(x)),
and ~,=!_x) = spe~ ( k [ ~ , ~ ' , ~ / ( ~ , ~ ) ) .
Note R = k[[y,u,t]]. R c > R. (f)R = ideal of X in Z at x.
Write oo

f = ~ j fij(t)yiu j , where fij(t) = ~. Fijata,Fij a ~ k.

i, j a=o
Note that the initial form @ of f is ~ =~4~Tj F ijoY:~U
j-'-
Now define:

i+J<v
a
(Think of the v------T~j) as a '~eighted system of exponents 'I).
 115

Let T = Yy,u,t (f) be the smallest element of A (I) • and let

IAI = the ~thullN of A = {q e Qlq > Y}.

. . . . _~ . . ~ -::- ~: ~

y (and hence IAI)is independent of the choice of the base

f of (f)R, but it does depend on the choice of parameters.
Remark: T,A are defined whenever we have expressions
• o

(E): f = ij~fij(t) yluJ ' fij(t) = ~ F i j a a t a

The important (and trivially verified) fact about T is that
given an 7 such f,
Ty°u°t,(f) < 1 <---> VM(f) < V
. , , , . .

Now let S T = {(i, j ) / ~ m ~ j / = y for some a such that Fija~ 0},

and define: [f]y,u,t = ~ + ~. 'Tterqns contributing to y", i.e.

= ~ + ~, ~ ~(v-i-j)~i~j
(i,j) Fij g v-i-J)
~S
T
where = L X(fl -

Definition: [f]y,u,t is solvable if ~ ~,~ c k such that

(Note that [ f ] solvable => y is an integer).

If [f]y,u,t is solvable, we make a 1,-preparatio.n,i.e. , we
replace y and u by new parameters

Yz Y - ~tY" ~i = u-1]tY
An easy computation then shows that:

T yz, ui, t (£) > Yy, u,t(£)

(1) The assumption that x is not on a v-fold curve y=u=O=>A is
non-emp~ty, and T>O. In fact, there must be some term
aF~''3~t~oYtbeT~i2~ tVmh~)F<~''3a .)~ 0 and z+~v" . And in such a term
 116

If [f]y~,u~,t is solvable, do the same. Suppose this is

possible infinitely many times. We get:

(Yl, U~) , (Y2,U2),...,CYm, Um),.--

Since each Yi is an integer, the sequences Yi and u i converge

in the M-adic t~pology.A Let y{~ = lim Yi' u{~ = lim ui(in R).
Tben f ~ (y{~,u{}) R. (in fact, y = Yy¢~,u{~,t(f) = ~ ,

so Fij a = 0 M i + j < V). Thus x lies on a formal v-fold

curve. But this yields a contradiction via the following

General Fact: Excellent ~ W -fold locus is comoatible with

completion (dimension remains tb~ same).

We may therefore assu~e F is T-prepared (i.e., not

solvable).
Now recall our situation:
Z g>Z
1"
h: X --> X h = quadratic
l
transformation
with center x.

both isolated v-fold points

~x(X) = Spec(k[~,G,E]/(~T,~). x: = the point P ( ~ x ( X ) ) .

In the fibre h'1(x), x t is tbe point ~/~ = ~/~ = O, so

(y',ut,t) = (y/t,u/t t) is a regular system of parameters

in R: = ~ , x z . The equation of X in Z at x ~ is
i I

f: = f/t ~ = ~.fij(t)/tv_(i+j)(y/t)i(u/t )j

The ' significant ' facts (verifiable by easy computations) are]

~. T~t u: t(f:) = yy,u,t(f) - 1

 117

Ell]v: ul t = [ f S y , u ~ t / ~ v

(Where Grx:(Z ~) = k[~, ~, ~I, Srx(Z) = k[~,G,~],

and we identify y' = y/~, u' = u/~).
In particular, IllS y, , u 1_solvable
,~ . . .
=> [ _f] y ~ u ~ t solvable.
......

~..If. ~Vt,u,t(f') > I, then l~¢,(f') = ~'

. . . . .

= ~(y',u') (= ~/~v , ~ = znN(f))"

In p,artic,ula~ the situation re m,ains, t.be s~ne in ,tb,i,s'.case

(i.e., dim ~ x : ( X ) = I, Vx.(X ) = v).

i I

By ~ and ~,after a finite number of quadratic transformations,

y <. I. So may assume in fact that y' = yyZ u~ t(f z) <_ 1 .
I~ yt < l, Vx1(X ) < v contradiction.
l
Suppose then that yI = 1. Then the initial form of f l is
equal to [f:Syl,ul,t • (i.e., the order of every term
contributing to yt is precisely v). Then there are two
possibilities (keeping in mind that t, and hence the initial
forms of all three parameters, appear non-trivially in

InM,(f:) = [ f : ] y , , u , t):
(i) ~ x ~ ( X ) = 0 (which is a contrsdiction because then,
l

by Idea A, no more v-fold points appear).

(ii) S parameters r,s in R: such that [f'Syl u. t=~(~,~),
where ~ is a form of degree v. (r = InM:(r), ~ = InM:(S))
But (ii) yields a contradiction as follows:
we may assume ~ = (a~: + b~' + ct), ~ = (a:~' + b:~' + c:~).
Have ~(~,u) + t-(something) = [f]y,u,t [f']y~,u~,t
--V
= ~(r,s) .t = ~(Y~t,~t) = ~((a~+b~+ct 2), ( a ~ + b ' ~ + c l t ~) )
Therefore ~(~,~) = ~((a~+b~), (a~+b'~)).
Notice then that ab:-ba' ~ O, for otherwise ~(~,~) is a
form of degree ~ in one parameter ay + b~ (=d(a:~+b~G), some
d ~ k) and contradicting e = 1. Now use
 118

Lemma: Let ~ = ~(X,Y),~ = ~(Z,W) be forms of degree v in

two v a r i a b l e s over k, and suppose that

~(y,u) = ~ay+bu),(aTy+blu)) c k[y,u], where a,a',b,blck and

ab: - ba: ~ 0. Then ~ c,c I e k 3 ~, ~ s k such that,

(in k [y,u,t]) ~(y+~t, u+~t) = ~((ay+bu+ct),(a'y+blu+clt)).

Proof: ~(y+~t,u+~t) = ~(a(y+~t) +b (u+~t), a' (y+~t) +b ' (u+~t))

= (ay+bu+ (a~+b~) t, a Iy+b lu+( a'~+b I~) t)

So we w a n t to solve:
a~ + b ~ = c
for ~ and
aI~+ b'~= c'
And this can be done if ab'-ba I ~ 0.
QED

Thus (letting t = t 2 in lemma snd r e c a l l i n g that

yt = l < > y= = 2) (il) =) ~'] solvable ~) ~]~,u,t is

solvable, which is a contradiction.

This completes the p r o o f of (I) for the case e = I.

2. e = 2 (cusp situation)

As before, we suppose that we have an infinite sequence

{b(m)} of p e r m i s s i b l e monoidal transformations:

... gCm)
~ > z( ) ~ > ... . . . . ~ > Z ° > Z
m i

h(m) > X(m) .> - > X h > X

... > z (m) ...> ....... ~ > xz > x all v-fold

points

where now x is an isolated v-fold point such that dim Z~x(X)=2.

R = @z,X, k = x(x), f defines X in Z at x. M = max (R)

Choose a s y s t e m of p a r a m e t e r s (y,u,t) of R so that

~x(X) = S p e c ( k [ ~ , ~ , ~ ] / ( ~ ) ) (~ = Ir~i(Y)) etc.

 119

Corresponding to the data (f;y,u,t), write

(F): f = .~ gi(t,u) ~ i
l

gi(u,t) = ~ GicdtCu d (Gic d ~ k)

c,d
(Note: I~(f) = ~ = Gv, o,o ~v)

Define
(a,b)
A = A(f;y,u,t) = U s Q2 [ Gi,a(v.i)b(v_i) F O}
i=o

<v_~i, d
=i~o v=i I ai, c,d F o }
and let
[AI = smallest convex set in l~ such that A c IA[

and (a,b) ~ [A[ ---> (a+r,b+s) c IA] ~ r,s _> 0

Let r ¢ EL We will use the notation:

S(r) = line through (r,O) with slope -1.

V(r) = vertical line through (r,O).
Remark: (1) A = A(f; y,u,t) is defined whenever we have
expressions (F) (i.e., without reference to any special

kind of x ~ X C___> Z). Then it is easy to check that

(i) The vertices of IAI are points of A, and they

all lie on the lattice

~II n) ~ ~Z/n) Q--> ~ (n = v')

(ii) vM(f) < v <~-> IA] contains a point on SCr)

w.i.tbr< 1 < : > there is a. vortex (atb) ' with

a+b<l.
(iii) Let ~ = a (f) = the distance of [A[ from
Y~ut.t
the b-axis = the smallest "a" appearing in any

(a,b) e IAI. Then m < 1 <~---> y = t = 0 is not

a v-fold curve through the point defined by f.

 120

(iv) Some vertex of I AI lies below the line

b = 1 (=> y = u = 0 is not a v-fold curve.

With m = ~y,u,t(f) as in Remark (iii) let ~ = ~T~u,t(f)

be the smallest such that (a,O) s IAI. ((~,~) is then

necessarily a vertex). Let y = Ty, u,t(£) be the smallest

number such that S(T) 0 IAI ~ ~, and let 6 = 6y, u,t(f) be

such that (~r-6,5) is the lowest point on S(¥) n IAI.

(m, ~) and (T-O~5) are the "m.ost important" vertices.

\ ......

.........

Solvabilit~ (a~b) - preparations and W ell-Pre2aredness

Let (a,b) be any vertex of IAI, and define

Jf~a,b = ~ + ~..terms contributing to (a,b)"
L Jy, U, t
-'r. v-~ )~b(v-i)
= o<~i<v Gi, a(v-i) ,b(v-i) ylta(
a,b
We say Lf~y,u,t is solvable ( or equivalently that the

vertex (a,b) is unprepared)if a,b are integers and

-y,u, t
For some ~ e k.
If ~f~a,b is solvable, w e c a n make a n
J y,u,t
"(a,b)-preparation" which is by definition a change of para-

meter
Y .. > y~ = y - ~ tan b

If we cannot make an (a,b)-preparation for any vertex

 121

(~,b) of 141 (i.e. if all the ~f~a,b are unsolvable)

' L Jy, u, t
then say we s.we "Well Prepared" ("W.P.")
Remark: (2) If y - - > Yl is an (a,b)-preparation, then
(~ Iz~(1:;yz, u,t) l c I-4(f;y,u,t) l - {(a,b)}.

(The vertex (a,b) is eliminated).

If (a:,b r) ~ (a,b) is another vertex, y - - > Yl

leaves it intact. (i.e., (aZ,b I) is a vertex of

IA(f;y ,u,t) I and ~fl az'bl =~f~a'~ in the
' [ jyl,u, t [ jy, u, t

sense that the former can be obtained by sub-

stituting the symbol ~i for the symbol y in the
latter).

Note: by the above, and the fact that 141 bas only a finite
number of vertices, and by an argument analogous to that of
Page (19),we deduce:
By a finite number of vertex preparations we may always
obtain a W.P. situation.

Translations in u and Very-Well Pre2aredness

In addition to the vertex preparations described above,

we will also need to consider translations of the type:
u > u = u-~ t n, ~ s k, n a positive integer.
i

Remark:(3) One checks easily that under u > u , a term

in f corresponding to a point (a~b) s IAI can affect only
those points strictl ~ below it .qp the line of slope -1/n
througb%t. In particular
 122

~y,u,t (~) = ~ , u ,t (~)

l

~y,u,t(f) = ~y, uft(f)

and J£~a, ~ is left intact.

Now suppose that (f;y,u,t) is l!.P. (can always assume to

be starting with this situation). Then choose ~ e k so
that afte~ u > u= u-~t (and subsequent W-preparation
z
y ~ > y ) we obtain the largest possible 6 .
l

Observe that by the previous Remark the highest point on

S(y) must remain, so that, if y ,u l are the new parameters s

Yyz, uft (f) = Yy, u,t ( £)"

Having obtained the-largest possible 5 in the above sense,
we distinguish two cases:

Case I (~-5~5) ~ (~,~) . Then we say that (f;y,u,t) is

"VerT-Well-Prepar ed" ("V.W. P. ")

Case Ii (y-g~5) = (a~)" Then consider the line U

through (~,~) with the largest slope sucb that none of IAI
lies below it. Let ~ = I slope of UG~I. (Note that ~ < 1),
and let (%1 ~) be the lowest point on Ua~ N IAI.
 123

If i / ~ is not an integer, say ~;e are V.W.P.

If I / ~ is an integer, choose ~ ~ k such that after

u > u = u -~t I/£ and subsequent well-preparation.

The numbe~ x becomes as large as possible (see Remark(3),

Page 26)). If we get to x = ~, get new U ~ and repeat.

We can undergo this procedure only finitely many times, and

wben we obtain tbe largest possible ~, say we are V.W.P.

Effects of Monoidal Transformations; Idea B: The Sequence

of Monoidal Transformations "Sharpens" IA I•

Given the date (f;y,u,t) ~e have the following operations:

({) Vertex Preparations at (a,b) (y-->y =y-~taub).

We bare remarked that this leaves other vertices

a~b :
(a',b t) intact, along ~ith their data {f}y,u,%.

Translations in u (u-->u --u-~t I/5- )

I

We have remarked that under this operation a

point (a,b) contributes only to points strictly

below it on the line of slope - 6 through it.

(if) ~uadratic and monoida ! transformations.

Remark (&) If we assume (f;y,u,t) is (a,~) - prepared,

then ~ is not affected by any operation in (i) above. In

particular ~ is not affected by passing from a W. P. to a

V. W. P. situation. We propose to prove Idea B in the form:

By the sequence {b~m ) } of monoidal transformations ~ is non-

....... ,,,H , , , •

increasing and eventually " _decreases. (This implies in essence

that we eventually arrive at a + ~ < l, so (by Remark 1)(if),

Page (2/~)) the multiplicity drops.)

Consider a permissible monoidal transformation

 124

Z: ~ > Z

b
X~ .. > X
x,x I are v-fold points
X ! ' > X
dim ~x(X) = 2.

(Think of this as occuring at some arbitrarily selected

stage in our sequence {b(m ) }; in particular x need not be

isolated in the v-fold locus).

Let R,M = ~ , x R',},i: = @'Z:,z:

(y,u,t) and (y:,u~,t I) are a system of parameters for

R and R I respectively.

(f;y,u,t) and (f~;yl,uX,t:) are data of X in Z at x and

X t in Z ~ at x I respectively.

We assume that ~z(X) = Spec(k[~,~,~]/(~)) and that

(f;y,u,t) is V.I,I.P. (As usual, ~ denotes InM(y), etc.)

B y Idea A there are four relevant ways in which x t can

be related to x:

T-I h is a quadratic .transformation with c e n t , e r x and

(y',~',t,) = (y/t,ui-~,t) , ~ ~ k.

T-2 h is a quadratic with center x and ( y l ~' t') : (y/~,u, t/~

T-3 h is a monoidal, t.r,ansf,ormation with cent. y = t=. 0 and

(y',u:,t:) = (y/t,u,t)

T-~ b is monoidal with center = u = 0 and (yZ~ut,t')

= (Y/u' u,t)

Since y defines ~z(X), and since we are assuming that we

have reached a stage in the resolving sequence where the

only irreducible v-fold curves are non-singular (see Page (12)

~e m a y always choose u and t so that one of the above occurs.

 125

We_now investigate the, ,.,effect,.of..each of the above on .#.

We will denote:

(~,~) = (~y,u,t(f),~y,u,t(f)) (a~:)=(ayT, u~t,(l~),~yl,u:tZ(fT))

(A(f;y,u,t)) = A(fS;y~,ul,t:), a transformation in the
plane.

T-3 Here f1=f/tv , and a is a translation to the left by 1

in fact:
yitCud >(y/t)iudtC-(V -i) = (y,)i(ul)d(t,)c-(v-i)

so the point (a,b) = (v-i •, ) moves to the point

(t-(v-i) a
v-i ' ~ ) = (a-l,b).
In particular o(~,~) =(a:,~:)(since ~ preserves the
shape and the orientation of I~I). .'. (~!,~r) = (~-l,~)-

does not change.

Note that T%3 occurs only when ~ > 1. Conversely, at any
point x whose m is > l, the only kind of monoidal transform-
ation in the sequence {h(m )} whicb affects it is T-3 (because

of the construction of the sequence {h(m )}; see Page (li~)).

Hence there can be only a finite number of successive T-31s.

T-~. See similarly that ~ is a translation down by l;

O(a,~3) = (~,p:) SO ~! = ~-!: ~ decreases.

Note that T-k occurs only when none of IAI lies below the
line b = l; conversely the only h(m ):s affecting zls whose
IAI have this property are T-k; there can be only a finite
number of successive T-_~k:s.
T-2. One checks that e(a,b) = (a,b+a-1). Therefore points
move vertically and thellnes S(r ) are transformed into

horizontal lines b = r-1. (In general lines of negative

slope are transformed into lines with increased slope). We
again have c(~,~) = (~I ~:). so (~:,~:) = (~,~+~-I). But
 126

< 1 (otber~:ise T-3)so ~ < ~: [ decreases.

Remark (5). In all the above caees, if (a,b) ¢ A, and
o(a,b) = (a~,b'), then

{~"y'u',t'
~a~b' = f {f }y,
%b u, t/~v T-3

Tbus (f;y,u,t) is (a, b) -prepared =>(f~;y~,u~,t ~) is (a~,b~) -

prepared.

In particular W.P. is preserved. Note that from a W.P.

situation we can pass to a V.W.P. situation without affecting
(~, ~).
T-I (Main Case).
a) C ~ 0 (Here something good happens because we may
assume to be V.W.P.). Apply tbe translation

u >u~ = u- ~t and well-prepare (via some y-->y ).

Note that this does not affect (m,~) but it might change

(7-5,~) (See Remark (3) Page (26). Denote

0~ = ~y~u,t (f) Y~ 7y,u~,t (f) •

Suppose we were originally (i.e., ~before y - - > y~,u --> u~) in
the situation of
Case I (Pa~e (~) (a~) / (T-5,5). In particular
have 8 < ~. And since 8 is defined for a V.W.P. situation,
6 <8<F.
I --

Suppose we were originally ~n the situation of

Case II (~,~) = (T-5,5). Then one can check that (T~O) is

a vertex of A(f;y~u ,t) and it is prep ared(1)(and bence is

!l) A sketch of t_be proof 'of "tbiQ assertion is: (y/~,u/.-~, t)

i -~ =P ~'_n~ -~ ) L . !n.par~'cular
there are terlus lik~ y~t c~e c ~ i n ~ from tbo~e ylt~uC ~hich
conwriouted to (a,~). Such y t are precisely those con-
tributing to (7,0) in A(f;y,u~,t). Then {£}I.0 ~ 0 and
{f}~,O . unsolvable.~ollow essentially b e c a ~ l i ~ is obtained
~,~l,~ from {f}~,~u,t by substituting ~t for u.
 127

not affected by the subsequent y --> yl). Then we must have

( ¥ - 6 ! , 6 I) = (y,O), so 01 = 0 < ~ (O<~because if ~ = O,

since m < 1 must have a + ~ < l, so multiplicity drops -
contradiction).
The point is that we bare reduced T-1 ~ / 0 to the case
T-1 ~ = 0 r Case I ( a ~ ) ~ (7-O,O)(which will be treated next),
except that we may no longer be V.W.P. This is inessential
however - we are still W.P. (the reduction to ~ = 0 is

~/ . . . . > U ).

T-1 b) ~=0
Here o(a,b) = (a+b-l,b) ; points move horizontally,

lines S(r ) become V(r_l ) snd in general lines cfnegative

slope decrease their slope. (See picture). Again we have,

that if o(a,b) = aT,bS), a:b:

then {f}y:,u~,t z={f)a,b
.y,u,t/~zv ,
so W.P. is preserved, and hence so is V.W.P., since (TI-G,~O)
must be some (xt,~). (See Page (~']).

o'-> ..

"\

Case. I (ar~) ~ (~-0~5) (for this case need only assume W.P.).

One checks by the above that o(y-O,O) = (m, ~) hence

~' = y < ~: ~ decreases.

''' .Sl , ] ,

(fl;yZ,ul,tt) is again V.W.P., and if ~ g 0 one checks

 128

easily that I / ~ t = I/E -I. (where E I is the absolute

value of the slope of the line through (a1,~ t) = (a,~) of

largest slope having ncne of IATI below it). In particular

: > ~ , and eventually El >. l; then we will be in

T-l, ~ = O~ Case I.

If the sequence of monoidal transformations involves only

the situations treated above, we will have an infinite number

of decreases of ~. But since (~,~) moves on a lattice, we

must eventually reach ~ = O, after which tbere can be no

further decreases - contradiction.

Thus we need only be concerned with the occurence

of the one remaining situation:

T-l, ~ = O, ( ~ ) =(y-5,~) and £ - 9 .

<¢

Here we must have ~ < i (otherwise T-~ since in this case

> 1 => all vertices bare b _> 1). Then (~ , ~I)=(~+~.1, ~)

and in fact c translates the whole figure to the left by

1-~. Thus the situation (ft;y:,ut,t s) is of the same type,

so the sequence of monoidal transformations corresponds to

a sequence of translations of iAl to the left by 1 - ~.

Therefore (since m < 1 otherwise T-3) we eventually come to

+ ~ <_ l, i.e., we get into the shaded area:

Ii/~-
 129

If we actually get ~ + ~ < I, then the multiplicity drops -

contradictien. If we get a boundary point (the only one to

w o r r y about is really (0,1)), then by an argument analogous

to that for the case e = 1 (Page (21)), we obtain

dim ~ x ( X ) < 2, so we are reduced to the case e = I.

Q.E.D.

Remarks about the .general theorem

Starting with a two dimensional excellent X (without

any hypothesis about X C > Z) the structure of the induc-

tion essentially allows us to reduce to the analysis of the

behavior of an isolated singular point under successive

monoidal transformations. Thus we may consider just a two

dimensional excellent local ring with only isolated

singularity (i.e., g localized at any non maximal prime is

regular), and, by the general theory of monoidal transform-

ations, we m a y even assume g is complete. Then we have

an embedding X = Spec(~) -- > Spec(R) where (R,M) is a

complete regular local ring. Thus ~ = R/j, where now

J = (f ,...,fn )-

i) Instead of vx(X ) we now use w~(X) or

equivalently v;:'(X,Z), which is the sequence:

(vMC~)'''',vMC~r),®, ~, .... )
(assuming the generators f. of J are suitably chosen.
I

In particular, they must be a "standard base").

2) I&(J;y,u,t) I m a y be defined in this more

general setting. (The y,u,t now stand for tuples of para-

meters).
 130

3) Idea A works (even with imperfect residue

field extension). In general the statement "The singularity

of x I in X: is as bad as the singularity of x in X must be

taken to mean:

v~~I (Xl ) >. v~(X) lexicographically

APPENDIX

We give a sketch of a proof of Idea A (in the special case

we have treated):

(R,M) a regular, local k-algebra. R/M = k, k = ~,

p ( > R a regular prime ideal, f c R. Suppose that
V = vM(f) = Vp(f). (This is normal flatness). Let (Rt,M I)

be a monoidal transform of ~ with center P such that

PR I = tRt(1)and R'/M t = R/M = k. Suppose that vM,(f/ty) -~ v.

Then if (yl,...,yr) are part of a system of parameters of R

such tbat

= kZzz(y ~) + .... + kZnN(y r)

is the smallest subspace of Gr~i for which I~i(f) c k[T],

we have:
yl/t,...,yr/t are regular parameters in R I .

Proof. By Normal Flatness we m a y assume the Yi are in P.

(In fact, by condition (2) of the Theorem on Normal Flatness,

page (4), we have "Tx(D)<--> ~x(X)" so the subspace T of

Gr~(R) defining ~'x(X) is contained in the subspace defining

Tx(D). But in our case D = V(P) C__> Spec R, so the subspace

(1) By definition, a m0noidal" transform of R with' 'center P

is a local R-algebra R' which is a localization of a ring o'f
type A = [p,/.,...,p,/.] with respect to a prime ideal N
containing MAY (where~the ~. are a set of generators of
P~ t ¢ P, and A is regarded ~s an R-subalgebra of R t) . Then
one can show that R t is necessarily the local ring ~f some
point lying over the closed point in the monoidal transform-
ation of Spec(R) with center V(P). (This assertion means
we may assume t = Pi for some i).
 131

defining Tx(D) is spanned by the initial forms of the para-

meters which generate P). So we have:

P = (Yx'''"Yr'Xx '''''xs)R' part of a regular system of

parameters of R. (And we may assume that t is an x or a y).

Now by the assumption v = vl~(f) = Vp(f), we may write:

oI o
f= =~ • "'Yr r +p where a ~ k
~,...,%1 a j ~ and pe ~ v + ~
I~-I =V

so f/tV = ~O aq(Y~/t) oi ...(Yr/t) or + p/tv

Claim that ~.VMT(f/tv) -~ V implies that all the Yi/t are in M I.

- - : , , ,, , ,

(Then, by the general theory of monoidal transformations,

it follows that the Yi/t are in fact regular parameters of R I,

which completes the proof). %Je know that for some

= (~,..-,~r) C k r, Z i = Yi/t - ~i are regular parameters

of R:.- We wish to sbow all the ~i = O. If not, let

Y/t = (~/t '''''Yr/ t ), Z = (Z ~, ... ,zr) and let F(Y) be the

form
O "i "'" , (of degree v over k).

Since VM:(f/tv ) -~ v, F(y/t) = F(Z + ~) ~ M'. But this can

only happen if F(Z+~) = F(Z) ( i.e., if after expanding
F(Z+~) all terms involving Z with total order < v are 0).

But then, by an easy argument about forms, after a suitable

change of coordinates we can express F as a form in fewer

than r variables, which leads to a contradiction regarding

the dimension of T.
Q.E.D.
 132

Bibliograpby

[1] Heisuke Hironaka, "Resolution of Singularities of an

Algebraic Variety over a field of

characteristic 0". Annals of

Mathematics, Vol. 79, No.2, March 1965.

[2] "On the Characters v ~:" and t ~:"of

Singularities". Columbia University

Mimeographed Notes.