Full text of "Dual graphs of exceptional divisors"

See other formats

DUAL GRAPHS OF EXCEPTIONAL DIVISORS

CN ' JANOS KOLLAR

o

■ Let X be a complex algebraic or analytic variety. Its local topology near a point

C^ I X € X is completely described by the link L{x g X), which is obtained as the

intersection of X with a sphere of radius < e -C 1 centered at x. A regular

neighborhood of a; G X is homcomorphic to the cone over L{x E X); cf. |GM88[

CN ; p.41].

One can study the local topology of X by choosing a resolution of singularities
TT : Y ^ X such that E^ ■= 7r~^(x) CZ Y is a simple normal crossing divisor
r^ I and then relating the topology of E^ to the topology of the link L{x G X). This

^ • approach, initiated in |Mum61| . has been especially successful for surfaces.

. ' The topology of a simple normal crossing divisor E can in turn be understood

'■^, . in 2 steps. First, the Ei are smooth projective varieties, and their topology is

C^ ' much studied. A second layer of complexity comes from how the components Ei

are glued together. This gluing process can be naturally encoded by a finite cell

complex 'D{E), called the dual graph or dual complex of E] see Definition [5l Given

X € X and a resolution tt : F — > X, the dual complex V{Ex) depends on the

►^ ' resolution chosen, but its homotopy type does not; we denote it by 'DTZ{x G X)

O' Csee |Thu07l[Ste08llABWlT])

Tij" ' Using this approach }KK11| proved that for every finitely presented group F there

>0 . is a complex algebraic singularity {x G X) (of dimension 3) such that tti{^L{x G

X)^ = F. The proof starts with first constructing a simple normal crossing variety
E such that ttiIE) ^ tti (V{E)) = F and then realizing E as the exceptional divisor

(N

m

I~i I on a resolution of a singularity {x G X), yielding a chain of isomorphisms

Tri{L{x G X)) = 7^l(I?7^(0 G X)) = tti{V{E)) = F.

The aim of this note is to go further and prove that not just tti(T>TZ{0 G X))
k> ' but 'DTZ{0 G X) can be arbitrary.

^ ' Theorem 1. Let T be a connected, finite cell complex. Then there is a normal

- - - singularity (0 G Xt) whose dual complex T)TZ{0 G Xt) is homotopy equivalent to

T.

It is interesting to connect properties of VTl{Q G X) with algebraic or geometric
properties of the singularity (0 G X). A quasi projective variety X has rational
singularities if for one (equivalently every) resolution of singularities p : Y —^ X
and for every algebraic (or holomorphic) vector bundle i^ on A, the natural maps
W{X,F) — > H^{Y,p*F) are isomorphisms. That is, for purposes of computing
cohomology of vector bundles, X behaves like a smooth variety. See |KM98[ Sec. 5.1]
for details.

It is known that if X has rational singularities then 'DTZ{0 G X) is Q-acyclic,
that is, W{Vn{0 G A),Q) = for i > 0, see for example |KK11[ Lem.39].
The fundamental groups of the VTZ{Q G X) for rational singularities were deter-
mined in [KKllj : these are exactly those finitely presented groups G for which

2 JANOS KOLLAR

H^{G,Q) — H'^{G,Q) = (sometimes called Q-superperfect groups). Our next
result determines the possible homotopy types of 'DTZ{0 G X) for rational singular-
ities.

Theorem 2. Let T be a connected, finite, Q-acyclic cell complex. Then there is a
rational singularity (0 G X) whose dual complex VTZ{0 G X) is homotopy equivalent
toT.

While these are much stronger results than the fundamental group versions, most
of the work needed to prove these theorems was already done in |Kolll| IKKll] .

The main technical result of jKKll) proves that for every compact simplicial
complex T there is a projective simple normal crossing variety Z such that T^iZ)
is homotopy equivalent to T, while the main technical result of |Kolll) shows that
for every projective simple normal crossing variety Z there is a normal singularity
(0 G X) and a partial resolution tt : X' ^ X such that Z ^ Tr^^{0) C X' .

If dim Z < 4 then X' has only very simple singularities which are easy to resolve.
This was sufficient to control tti (T>TZ{0 G X)) . However, as dim Z increases, X'
has more and more complicated singularities given locally by equations of the form

(yii ■■■ yim
'■ '■

Ural ' ' ' Uimra

where n,m are arbitrary and {xi,yij,t) are local coordinates.

If dim Z = 2, then the only singularity that appears is the ordinary 3-fold double
point (xia;2 — tyn). The first somewhat complicated singularity

(xiX2a;3 = i • {yiiy22 - Vv2y2i)) c C*

appears when dimZ = 6.

In this paper we start with the varieties constructed in jKoUHlKKll] and resolve
these singularities. Surprisingly, the resolution process described in pTUlG]) leaves
the dual complex unchanged and we get the following.

Theorem 3. Let Z he a projective simple normal crossing variety of dimension n.
Then there is a normal singularity (0 G X) of dimension (n + 1) and a resolution
TT : Y -^ X such that E :— 7r~^(0) GY is a simple normal crossing divisor and its
dual complex 'L){E) is naturally identified with D{Z).

4 (Open problems).

(jUl) It might be possible to describe all complexes that occur on resolutions of
(n + l)-dimensional varieties. It is clear that dimV{E) < n but I do not know any
other restrictions.

Starting with an n-dimensional complex T, the constructions of jKollll IKKllj
give a (2n -I- l)-dimensional singularity (0 G X) such that 'DTZ{0 G X) is homotopy
equivalent to T. This increase of the dimension may not be necessary.

((312) The singularities constructed in Theorems [T}{5] are not isolated. It would
be interesting to construct isolated examples.

((4l3) As shown by |KK11) . links of isolated singularities are much more compli-
cated topologically than smooth projective varieties. As a starting point of further
investigations, it would be useful to understand the precise relationship between
'DTZ{0 G X) and the topology of the link of an isolated singularity.

DUAL GRAPHS OF EXCEPTIONAL DIVISORS 3

([114) As we noted, given a singularity (0 G X) and a resolution tt : Y ^ X
such that E := 7r^^(0) is a simple normal crossing divisor, the homotopy type
'DTZ{0 G X) of the dual complex ^{E) does not depend on the choice of tt : y ^- X.

Note that if p : X' — >■ X is a proper birational morphism such that E' :— 7r^^(0)
is a simple normal crossing divisor, then the dual complex 'D(E') is defined even if
X' is singular.

It is possible that 'D{E') is in fact homotopy equivalent to 'DTZ{0 G X) as long
as X' has rational singularities. (The latter condition is actually quite weak, for
instance it holds if none of the strata of E are contained in SingX.)

(H5) Assume that {X, A) is dlt |KM98[ 2.37]. Since dlt implies rational, VTZix G
X) a Q-acyclic for every point x & X. Furthermore, tti (VTZ{x G X)) = 1 by [Kol93|
and [TakOSj . Thus 'DTZ{x G X) is contractible iff it is Z-acyclic. It would be very
interesting to decide whether 'DTZ{x G X) is contractible or not. For quotient
singularities this is proved in [KSllj .

Related results are treated in |Kol07a] and |IIX09) .

Definition 5 (Dual graphs). Let AT be a variety with irreducible components
{Xi : i G /}. We say that A is a simple normal crossing variety (abbreviated as snc)
if the Xi are smooth and every point p E X has an open (Euclidean) neighborhood
p £ Up C X and an embedding Up ^^ Cp+^ such that Up C (zi • • • z„+i = 0).
A stratum of X is any irreducible component of an intersection Di^jXi for some
J Cl.

The combinatorics of X is encoded by a cell complex X'(A) whose vertices are
labeled by the irreducible components of X and for every stratum Z C Clu^jXi
we attach a (| J| — l)-dimensional cell. Note that for any j € J there is a unique
irreducible component of r\i^j\[jyXi that contains Z; this specifies the attaching
map. 2?(A) is called the dual complex or dual graph of X. (Although 'D{X) is not
a simplicial complex in general, it is an unordered A-complex in the terminology
of |Hat02l p.534].)

Definition 6 (Dual graphs associated to a singularity). Let A be a normal variety
and x G A a point. Choose a resolution of singularities tt : Y ^ X such that
Ex :— 7r^^(x) CZ Y is a simple normal crossing divisor. Thus it has a dual complex
V{Ex).

The dual graph of a normal surface singularity has a long history. Higher dimen-
sional versions appear in |Kul77[ iPerTTI IGorSOl IFM83) but systematic investigations
were started only recently; sec [ThuOT', 'StjOgl |Pay09[ [PiyTTI .

It is proved in ThuO?^ Ste08, ABWlT] that the homotopy type of V{E,j;) is
independent of the resolution Y —>■ X. We denote it by 'DTZ{x G X).

The proof of Theorem [T] starts with the following, which is a combination of
Theorem 29 and Lemma 39 of [KK11| .

Theorem 7. Let T be a finite cell complex. Then there is a projective simple
normal crossing variety Zt such that

(1) T){Zt) is homotopy equivalent to T ,

(2) tii{Zt) =7ri(r) and

(3) H'{Zt,Ozt) = H\T,C) for every i>0. D

8 (Summary of the construction of [Kolll] ). Let Z he a projective local complete
intersection variety of dimension n and choose any embedding Z C P into a smooth

z

c

Y

i

i-K

e

X

4 JANOS KOLLAR

projective variety of dimension N . (We can take P = F^ for TV ^ 1.) Let L be a
sufficiently ample line bundle on P. Let Z C Yi C P he the complete intersection
of {N — n ~ 1) general sections oi L{—Z). Set

Y := B^_z)Yi := Proj^^ E^^o^nC"^^)-

(Note that this is not the blow-up of Z but the blow-up of its inverse in the class
group.)

It is proved in jKoUl] that the birational transform of Z in F is a Cartier divisor
isomorphic to Z and there is a contraction morphism

(01)

such that Y\Z = X\{0}. If F is smooth then T>n{0 G X) = V{Z) and we are
done with Theorem[TJ However, the construction of jKoUl] yields a smooth variety
Y only if dim Z — \ ov Z is smooth. (It is easy to see that not every simple normal
crossing variety Z can be realized as a hypersurface on a smooth variety, so this
limitation is not unexpected.)

Thus we need to understand the singularities of Y and resolve them.

In order to do this, we need a very detailed description of the singularities of Y .
This is a local question, so we may assume that Z C C^ is a complete intersection
defined by /i = ■ • ■ = fn-n — 0. Let Z C li C C^ be a general complete
intersection defined by equations

hi,ifi -\ h hi^N-nlN-n = for i = 1, . . . , iV - n - 1.

Let H = {hij) be the {N — n — 1) x {N — n) matrix of the system and Hi the
submatrix obtained by removing the ith column. By |Kolllj or |Koll2[ Sec. 3. 2], an
open neighborhood oi Z gY \s defined by the equations

{f, = {-lY -t-deiH, :i = l,...,Ar-n) CC^ xQ. (02)

Assume now that Z has simple normal crossing singularities. Up-to permuting the
fi and passing to a smaller open set, we may assume that d/2, . . . , dfM-n are linearly
independent everywhere along Z . Then the singularities of Y all come from the
equation

/i = -t-detiJi. (03)

Our aim is to write down local normal forms for Y along Z .

On C^ there is a stratification C^ = Ro D Ri D • • • where Ri is the set of
points where rankiJi < {N — ?i — 1) — z. Since the hij are general, codimv^ Ri — i^
and we may assume that every stratum of Z is transversal to each Ri \ Ri+i pop .

Let S G Z he any stratum and p G S a, point such that p € R„i \ Rm+i- We
can choose local coordinates {xi, . . . ,0;^} and {y^s '■ ^ 1^ t,s < m\ such that, in a
neighborhood of p,

fi — xi ■ ■ ■ Xd and det Hi = det (ajj-s : 1 ^ ?", s < m) .

Note that m? < dim S = n — d, thus we can add n ^ d — m? further coordinates xjij
to get a complete local coordinate system on S.
Then the n coordinates {x^, yij} determine a map

cr : C^ X Ct -^ C" X Ct

DUAL GRAPHS OF EXCEPTIONAL DIVISORS 5

such that (j(F) is defined by the equation

xi ■ ■ ■ Xd = t ■ det (jjrs : 1 < r, s < m) .

Since (i/2, . . . , dfN-n are Hnearly independent along Z, we see that cr|y is etale along
Z CY.

We can summarize these considerations as follows.

Proposition 9. Let Z be a normal crossing variety of dimension n. Then there is
a normal singularity (0 £ X) of dimension n + 1 and a proper, birational morphism
TT : Y ^ X such that red7r^^(0) = Z and for every point p £ 7r^^(0) we can choose
local etale or analytic coordinates called {xi : i G Ip} and {yrs '■ ^ ^ r, s < rup} (plus
possibly other unnamed coordinates) such that one can write the local equations of
Z cY as

iIl^eI,^^ = t = 0) C [U.ei,^. = t ■ det{yrs : 1 < r, s < m^) j C C"+2. D

10 (Detcrminantal varieties). We have used the following basic properties of de-
terminantal varieties. These are quite easy to prove directly; see |Har95[ 12.2 and
14.16] for a more general case.

Let T^ be a smooth, affine variety, and C C Oy a finite dimensional sub vector
space without common zeros. Let H = ihij) be an n x n matrix whose entries
are general elements in L. For a point p G ^ set mp = corank7J(p). Then there
are local analytic coordinates {yrs '■ ^ < r,s < nip} (plus possibly other unnamed
coordinates) such that, in a neighborhood of p,

det H — dei[yrs '■ ^ <r,s < nip) .

In particular, multp(detiJ) = corank_ff(p), for every m the set of points Rm C V
where corankiJ(p) > to is a subvariety of pure codimension m^ and Singi?™ —

Rm+l-

11 (Inductive set-up for resolution). The object we try to resolve is a triple

iY,E,F) := (r,E.e/i?.,E,ej«.^.) (Ell)

where y is a variety over C, Ei,Fj are codimension 1 subvarieties and Oj € N. (The
construction () produces a triple {Y,E := Z,F := 0) . The role of the Fj is to keep
track of the exceptional divisors as we resolve the singularities oiY.)

We assume that E is a simple normal crossing variety and for every point p G
E there is a (Euclidean) open neighborhood p G Yp C Y, an embedding Cp :
Yp ^^ (j^dimy+i^ subsets Ip C I and Jp C J, a natural number nip G N and local
coordinates in cdimy+i ca^^g^j

{xi :i e Ip}, {yrs : 1 < -r, s < rUp}, {zj : j G Jp} and t

(plus possibly other unnamed coordinates) such that one can write the local equa-
tion of ap{Yp) C C^ini^+i as

Il^eI,^^ = t ■ det(y,, : 1 < r, s < to^) • Uje.i,^T ■ Cll2)

Furthermore, ap(Ei) — {t — Xi = Q) f^ ap{Yp) for i G Ip and ap{Fj) — (zj —
0)r\ap(Yp) for j G Jp. (We do not impose any compatibility condition between the
local equations on overlapping charts.)

We say that {Y, E, F) is resolved at p if F is smooth at p.

The key technical result of the paper is the following.

6 JANOS KOLLAR

Proposition 12. Let (Y,E,F) be a triple as above. Then there is a resolution of
singularities n : (Y',E',F') -^ (Y,E,F) such that

(1) Y' is smooth and E' is a simple normal crossing divisor,

(2) E' = n-\E),

(3) every stratum of E' is mapped birationally to a stratum of E and

(4) TT induces an identification !?(£'') = 'D(E).

Proof. The resolution will be a composite of explicit blow-ups of smooth subva-
rieties (except at the last step) . We use the local equations to describe the blow-up
centers locally. Thus we need to know which local subvarieties can be defined glob-
ally. For example, choosing a divisor Fj-^ specifies the local divisor {zj-^ — 0) at every
point p G Fj-^ . Similarly, choosing two divisors Ei-^^ , Ei^ gives the local subvarieties
(t = Xi-^ = Xij = 0) at every point p £ Ei-^ n Ei^ . (Here it is quite important that
the divisors Ei are themselves smooth. The algorithm does not seem to work if the
Ei have self-intersections.) Note that by contrast (a;,^ = Xi^ = 0) C Y defines a
local divisor which has no global meaning. Similarly, the vanishing of any of the
coordinate functions i/rs has no global meaning.

To a point p e Sing E we associate the local invariant

Deg(p) := (deg^(p),degy(p),degjp)) ^ (|/p|,TOp, EjeJp«j)-

It is clear that deg3.(p) and deg_^(p) do not depend on the local coordinates cho-
sen. We see in (fT4l) that degy{p) is also well defined if p G Singi?. The degrees
deg^{p),degy{p),deg^{p) are constructible and upper semi continuous functions on
Sing^;.

Note that Y is smooth at p iff either Deg(p) = (1, *, *) or Deg(p) — {*, 0, 0). If
deg^(p) = 1 then we can rewrite the equation ((TTJ2) as

x' ^t- rijZj' where x' -.^ xi + t ■ [l - det(yrs)) • Dj-^^j' '

so if yis smooth then {Y, E + F^ has only simple normal crossings along E. Thus
the resolution constructed in Theorem  is a log resolution.

The usual method of Hironaka would start by blowing up the highest multiplicity
points. This introduces new and rather complicated exceptional divisors and I have
not been able to understand explicitly how the dual complex changes.

In our case, it turns out to be much better to look at a locus where degy{p) is
maximal but instead of maximizing deg^ {p) or deg^ (p) we maximize the dimension.
Thus we blow up subvarieties along which Y is not equimultiple. Usually this leads
to a morass, but our equations separate the variables into distinct groups which
makes these blow-ups easy to compute.

One can think of this as mixing the main step of the Hironaka method with the
order reduction for monomial ideals (see, for instance, jKol07b| Step 3 of 3.111]).

After some preliminary remarks about blow-ups of simple normal crossing vari-
eties the proof of (jT^ is carried out in a series of steps pTHTC| .

We start with the locus where degj, {p) is maximal and by a sequence of blow-ups
we eventually achieve that degy{p) < 1 for every singular point p. This, however,
increases deg^. Then in 3 similar steps we lower the maximum of deg^ until we
achieve that deg^(p) < 1 for every singular point p. Finally we take care of the
singular points where degy{p) + deg^(p) > 1. D

13 (Blowing up simple normal crossing varieties). Let Z he a simple normal crossing
variety and W C Z & subvariety. We say that W has simple normal crossing with Z

DUAL GRAPHS OF EXCEPTIONAL DIVISORS 7

if for each point p £ Z there is an open neighborhood Zp, an embedding Zp 'H- C"+^
and subsets /p, Jp C {0, . . . ,n\ such that

Zp = (^^G/,^* = 0) and Wr\Zp^{xj={):j e Jp) .

This imphes that for every stratum Zj C Z the intersection W D Zj is smooth
(even scheme theoretically).

If W has simple normal crossing with Z then the blow-up By/Z is again a simple
normal crossing variety. If W is one of the strata of Z, then V{BwZ) is obtained
from T^iZ) by removing the cell corresponding to W and every other cell whose
closure contains it. Otherwise 'D{B\yZ) — T>{Z). (In the terminology of |Koll2[
Sec. 2. 4], BwZ -^ Z is a thrifty modification.)

As an example, let Z = {xiX2X-i = 0) C C^. There are 7 strata and 2?(Z) is the
2-simplex whose vertices correspond to the planes {xi = 0).

Let us blow up a point W — {p} C Z to get BpZ C Bp€? . Note that the
exceptional divisor E C BpC^ is not a part of BpZ and BpZ still has 3 irreducible
components.

If p is the origin, then the triple intersection is removed and ^{BpZ) is the
boundary of the 2-simplex.

If p is not the origin, then BpZ still has 7 strata naturally corresponding to the
strata of Z and 'D{BpZ) is the 2-simplex.

We will be interested in situations where F is a hypersurface in C""''^ and Z dY
is a Cartier divisor that is a simple normal crossing variety. Let W <ZY hea, smooth,
irreducible subvariety, not contained in Z such that

(1) the scheme theoretic intersection WCiZ has simple normal crossing with Z

(2) mult^nw Z = multvF Y. (Note that this holds if W^ C Sing Y and multznw^ Z
2.)

Choose local coordinates (xq, . . . , a:„,i) such that W = (xq = ■ ■ ■ Xi = 0) and
Z — {t — 0) CY. Let /(xo, . . . , Xn, t) = be the local equation of Y.

Blow up W to get TT : BwY — s> Y. Up to permuting the indices 0, . . . , i, the
blow-up B\yY is covered by coordinate charts described by the coordinate change

\X{) , Xi , . . . , X^, Xi-\-\ , . . . , Xm f^j — V'^O' ^1^0' ■ • ■ 1 ■^2*^0' -^^-t-l 1 ■ • ■ 1 •^715 ^j ■

If multvi-' Y = d then the local equation of BwY in the above chart becomes
By assumption (2), [x'^Y is also the largest power that divides

I \^0i ^l^Qn ■ ■ ■ ^^i^0^ X^+l^ • ■ • 7 2;„, Uj ,

hence 7r^^(Z) = BwnzZ.

Observe finally that the conditions (1-2) can not be fulfilled in any interesting
way if Y is smooth. Since we want Z n VF to be scheme theoretically smooth, if Y
is smooth then condition (1) implies that Z n VF is disjoint from SingZ.

(As an example, let Y = £? and Z = [xyz = 0). Take W := {x = y = z).
Note that W is transversal to every irreducible component of Z but W C] Z \s
a non-reduced point. The preimage of Z in BwY does not have simple normal
crossings.)

There are, however, plenty of examples where Y is singular along Z nW and
these are exactly the singular points that we want to resolve.

JANOS KOLLAR

14 (Resolving the determinantal part). Let m be the largest size of a determinant
occurring at a non- resolved point. Assume that m > 2 and let p G y be a non-
resolved point with nip = m.

Away from E U F the local equation of Y is

DiG/p^i = det(2/rs : 1 < r, s < m).

Thus, the singular set of Yp\{EU F) is

|J(i^i/)(rank(j/rs) < m - 2) n (a;^ = Xf = O)

where the union runs through all 2-element subsets {i, i'} C Ip. Thus the irreducible
components of Sing Y \ {E U F) are in natural one-to-one correspondence with the
irreducible components of Sing E and the value of to = deg„ (p) is determined by
the multiplicity of any of these irreducible components at p.
Pick 11,^2 G / and we work locally with a subvariety

Wp{ii,i2) ■■= {r&nk{yrs) < m - 2) n {xi, = Xi^ = O).

Note that W'{ii,i2) is singular if m > 2 and the subset of its highest multiplicity
points is given by rank(yrs) = 0. Therefore the locally defined subvarieties

Wp{ii,i2) := (y-rs =0:l<r, s<m)n [xi^ = Xi^ = O).

glue together to a well defined global smooth subvariety W := W{ii, 12).

E is defined by {t = 0) thus ECiW has the same local equations as Wp{ii,i2)- In
particular, E ClW has simple normal crossings with E and _E n H^ is not a stratum
of E; its codimension in the stratum {xi-^ — Xi^ — 0) is m^.

Furthermore, E has multiplicity 2 along E n W, hence (fT3l2) also holds and so

V{BEnw) ='D{E).

We blow wp W G Y . We will check that the new triple is again of the form
pT|) . The local degree Deg(p) is unchanged over Y \ W . The key assertion is that,
over W , the maximum value of Deg(p) (with respect to the lexicographic ordering)
decreases. By repeating this procedure for every irreducible components of Sing E,
we decrease the maximum value of Deg(j)). We can repeat this until we reach
deg (p) < 1 for every non- resolved point p ^Y .

(Note that this procedure requires an actual ordering of the irreducible compo-
nents of Sing E', which is a very non-canonical choice. If a finite groups acts on Y ^
the resolution usually can not be chosen equivariant.)

Now to the local computation of the blow-up. Fix a point p E W and set
Ip ■— Ip \ {«!, 12}- We write the local equation of Y as

z-

jeJp"'3

Xi-^Xi2 ■ L ~ t ■ det{yrs) ■ R where L-=Y[i£i*^i ^^'^ ^-—11

P

There are two types of local charts on the blow-up.
(1) There are two charts of the first type. Up to interchanging the subscripts
1,2, these are given by the coordinate change

{xi^,Xi^,yrs : 1 < r,s < to) = (x-^ , a;-^^-^ , y^^a;-^ : 1 < r,s < to).

After setting z^ '■— ^'i the new local equation is

<.i = i.det(y;j.(z™'-2.i?).

The exceptional divisor is added to the F-divisors with coefficient m^ — 2
and the new degree is [deg^{p) — l,degy(p),deg2(p) + to^ — 2).

DUAL GRAPHS OF EXCEPTIONAL DIVISORS 9

(2) There are m^ charts of the second type. Up to re-indexing the m^ pairs
(r, s) these are given by the coordinate change

except when r = s = m where we set ymm — Vmra- It is convenient to set
y'mm — 1 ^-iid '^w '■= y'mm- Then the new local equation is

:>^,y,^ -L^t- det{yl, : 1 < r, s < m) • (zf-^ ■ R) .

Note that the {m, m) entry of (y^s) is 1- By row and column operations we
see that

det(y;, : 1 < r, s, < to) = det(y;, - y^y^,. : 1 < r, s, < to - l).

By setting j/"^ :— y'^^ — yrrnVms '^6 have new local equations

x'^y.^L = t ■ det{y';^ : 1 < r, s, < to - 1) • {zf-^ ■ R)

and the new degree is (deg^. (p) , deg^^ (p) — 1, deg^(p) + m"^ ~2).

Outcome. After these blow ups we have a triple {Y, E, F) such that at non-
resolved points the local equations are

U.ei,^^=t-yU,eJ,^-' or U^eI,^^=t■U,eJ,^T■ (El^)

(Note that we can not just declare that y is also a z-variable. The Zj are local
equations of the divisors Fj while (y = 0) has no global meaning.)

15 (Resolving the monomial part). Following ((Til 3). the local equations are

Il^eI,x^ =t-y''- rijej^^r where c e {0, 1}.

We lower the degree of the z-monomial in 3 steps.

Step 1. Assume that there is a non- resolved point with Ojj > 2.
The singular set of Fj-^ is then

U(i,i')(^ji = Xi = Xii = 0)

where the union runs through all 2-element subsets {i,i'} C /. Pick an irreducible
component of it, call it W{ii,i2,ji) '■= {zj^^ = Xi^ = Xi^ — O).

Set /* :— Ip \ {ii, 12}, Jp '■= Jp \ {ji} and write the local equations as

Xi^Xi^ ■ L = tz°j' ■ R where L := Wifzj.Xi and R := y'' ■ Hj-gj-^i'-

There are 3 local charts on the blow-up:

(1) (xij , Xij , Zj ) = {x[_^,x^^x^_^, Zjx'^_^) and, after setting z^ '■— x'^_^ the new local

equation is

xl^-L^t-z-^-^zp -R.

The new degree is (deg^(p) - l,degy(p),deg^(p) + aj - 2)

(2) Same as above with the subscripts 1, 2 interchanged.

(3) (xij , Xi^ , Zj) — (x^j^z'-, x^^z', z'A with new local equation

2^11 Xi2 ■ L — t ■ Zj ^ ■ R.

The new degree is (deg^(p),degy(p), deg^(p) - 2).

10 JANOS KOLLAR

Step 2. Assume that there is a non- resolved point with Oj^ ~ aj.-^ = 1.
The singular set of Fj^ n Fj^ is then

where the union runs through all 2-elenient subsets {i,i'} C /. Pick an irreducible
component of it, call it W{ii,i2,ji,J2) '■= {zj^ = Zj^ = Xi^ = Xi^ = O).

Set /* :— Ip \ {«i, 22}, Jp '■= Jp \ {ji, J2} and we write the local equations as

Xi-^Xi^ ■ L — tzj-^Zj2 ■ R where L :— YiiPi'^i ^^^ ^ ■= U'^ ' Yi

3<^'K

z-

There are two types of local charts on the blow-up.

(1) In the chart {xi^,Xi^,Zj^,Zj^) ~ {x^^,x'^^x^^,z'j^x^^,Zj^x^^) the new local

equation is

<-^ = ^-44-^-

and the new degree is (deg^(p) — 1, degj^(p), deg^(p)). A similar chart is
obtained by interchanging the subscripts ii, 12-
(2) In the chart {x.i^,Xi^,Zj^,Zj^) = {x'^^z'^^,x'^^Zj^, Zj^, Zj^ZjJ. the new local
equation is

The new degree is (deg^(p),degj,(p), deg^(p) — l).

A similar chart is obtained by interchanging the subscripts j'l, j2-

By repeated application of these two steps we are reduced to the case where
deg^{p) < 1 at all non-resolved points.

Step 3. Assume that there is a non- resolved point with degy{p) — deg^{p) — 1.
The singular set of Y is

U(^,^')(2^ = ^ = ^' =^«' =0)-

Pick an irreducible component of it, call it W{ii,i2) :— (y = z = Xi^ = Xi^ ~ O).
The blow up computation is the same as in Step 2.

As before we see that at each step the conditions ([T51 1-2) hold, hence X'(£') is
unchanged.

Outcome. After these blow-ups we have a triple (Y, E, F) such that at non-
resolved points the local equations are

U.ieipX,=t-y, U^eI^^^ = t ■ zi or ll^eI^^^ = '^- CS^)

As before, the y and z variables have different meaning, but we can rename zi as
y. Thus we have only one non-resolved local form left: Yi^i ~ ^V-

16 (Resolving the multiplicity 2 part). Here we have a local equation x^^ • • ■ Xi^ = ty
where d > 2. We would like to blow up {xi^ = y = 0), but, as we noted, this
subvariety is not globally defined. However, a rare occurrence helps us out. Usually
the blow-up of a smooth subvariety determines its center uniquely. However, this
is not the case for codimension 1 centers. Thus we could get a globally well defined
blow-up even from centers that are not globally well defined.

Note that the inverse of (xi^ — y — 0) in the local Picard group of Y is Ei-^ =
{xi^ = t = 0), which is globally defined. Thus

ProjyS^>oCy(m£;,J

DUAL GRAPHS OF EXCEPTIONAL DIVISORS 11

is well defined, and locally it is isomorphic to the blow-up B(^x. ^y^Q-jY. (A priori,
we would need to take the normalization of B^,^.. =y^Q)Y , but it is actually normal.)
Thus we have 2 local charts.

(1) {xi-^ , y) = {x'^_^ , y'x'^_^ ) and the new local equation is {xi^ ■ ■ ■ Xi^ — ty'^ . The
new local degree is (d — 1, 1, 0).

(2) [xi^ , y) — {x[_^ y' , y') and the new local equation is (x^^ -Xi^ ■ ■ ■ Xi^ = t) . The
new local degree is (d, 0,0).

Outcome. After all these blow-ups we have a triple (^, X^iez-^iiX^iG/'^i^i)
where J2i<£i^i i^ ^ simple normal crossing divisor and Y is smooth along X^ie/^j-

This completes the proof of Proposition [T^l □

17 (Proof of Theorem). Assume that T is Q-acyclic. Then, by (O there is a simple
normal crossing variety Zt such that H^(^Zt,Ozt) = for i > 0. Then [KoUli
Prop. 9] shows that, for L sufficiently ample, the singularity (0 S Xt) constructed
in © and ® is rational. By ^ we conclude that Vn{0 G Xt) = V{Zt) is
homotopy equivalent to T.

Acknowledgments. I thank M. Kapovich, P. Ozsvath and S. Payne for comments
and corrections. Partial financial support was provided by the NSF under grant
number DMS-07-58275.

References

[ABWll] D. Arapura, P. Bakhtary, and J. Wlodarczyk, Weights on cohomology, invariants of

singularities, and dual complexes, ArXiv e-prints (2011).
[FM83] Robert Friedman and David R. Morrison (eds.), The birational geometry of degenerations,

Progr. Math., vol. 29, Birkhiiuser Boston, Mass., 1983. MR 690262 {84g: 14032)
[GM88] Mark Goresky and Robert MacPherson, Stratified Morse theory, Ergebnisse der Math-

cmatik und ihrer Grenzgebiete (3), vol. 14, Springer- Verlag, Berlin, 1988. MR 932724

(90d:57039)
[Gor80] Gerald Leonard Gordon, On a simplicial complex associated to the monodromy. Trans.

Amer. Math. Soc. 261 (1980), no. 1, 93-101. MR 576865 (81j:32017)
[Har95] Joe Harris, Algebraic geometry. Graduate Texts in Mathematics, vol. 133, Springer-

Verlag, New York, 1995, A first course. Corrected reprint of the 1992 original.

MR MR1416564 (97e:14001)
[Hat02] Allen Hatcher, Algebraic topology, Cambridge University Press, Cambridge, 2002.

MR 1867354 (2002k:55001)
[HX09] Amit Hogadi and Chenyang Xu, Degenerations of rationally connected varieties. Trans.

Amer. Math. Soc. 361 (2009), no. 7, 3931-3949. MR 2491906 (20101:14091)
[KKll] Michael Kapovich and Janos KoUar, Fundamental groups of links of isolated singularities,

ArXiv e-prints (2011).
[KM98] Janos KoUar and Shigefumi Mori, Birational geometry of algebraic varieties, Cambridge

Tracts in Mathematics, vol. 134, Cambridge University Press, Cambridge, 1998, With the

collaboration of C. H. Clemens and A. Corti, Translated from the 1998 Japanese original.

MR 1658959 (2000b: 14018)
[Kol93] Janos KoUar, Shafarevich maps and plurigenera of algebraic varieties. Invent. Math. 113

(1993), no. 1, 177-215. MR 1223229 (94m:14018)
[Kol07a] , A conjecture of Ax and degenerations of Fano varieties, Israel J. Math. 162

(2007), 235-251. MR 2365862 (2008j:14017)
[Kol07b] , Lectures on resolution of singularities, Annals of Mathematics Studies, vol. 166,

Princeton University Press, Princeton, NJ, 2007. MR 2289519

[KoUl] , New examples of terminal and log canonical singularities. [arXiv: 1107.2864 2011.

[Koll2] , Singularities of the minimal model program, Cambridge University Press, Cam-
bridge, 2012, With the collaboration of S. Kovacs (to appear).

12 JANOS KOLLAR

[KSll] M. Kerz and S. Saito, Cohomological Hasse principle and McKay principle for weight
homology, ArXiv e-prints (2011).

[Kul77] Vik. S. Kulikov, Degenerations of K3 surfaces and Enriques surfaces, Izv. Akad. Nauk
SSSR Ser. Mat. 41 (1977), no. 5, 1008-1042, 1199. MR 0506296 (58 #22087b)

[Muni61] David Mumford, The topology of normal singularities of an algebraic surface and
a criterion for simplicity, Inst. Hautes Etudes Sci. Publ. Math. (1961), no. 9, 5—22.
MR 0153682 (27 #3643)

[Pay09] Sam Payne, Lecture at M5-RJ,[ http://www.msri.org/web/msri/online-videos/-/video/show Video/36741
2009.

[Payll] Sam Payne, Boundary complexes and weight filtrations, ArXiv e-prints (2011).

[Pcr77] Ulf Persson, On degenerations of algebraic surfaces, Mem. Amer. Math. Soc. 11 (1977),
no. 189, xv+144. MR 0466149 (57 #6030)

[Ste08] D. A. Stepanov, A note on resolution of rational and hypersurface singularities, Proc.
Amer. Math. Soc. 136 (2008), no. 8, 2647-2654. MR 2399025 (2009g:32060)

[Tak03] Shigeharu Takayama, Local simple connectedness of resolutions of log-terminal singular-
ities, Internat. J. Math. 14 (2003), no. 8, 825-836. MR 2013147 (2004m:14023)

[Thu07] Amaury ThuiUier, Geometric toroidale et geometric analytique non archimedienne. Ap-
plication au type d'homotopie de certains schemas formels, Manuscripta Math. 123
(2007), no. 4, 381-451. MR 2320738 (2008g:14038)

Princeton University, Princeton NJ 08544-1000
kollarOmath . princeton . edu 