IAS/Park City Mathematics Series

Volume 00, Pages 000–000

S 1079-5634(XX)0000-0

Perverse sheaves and the topology of algebraic varieties

Mark Andrea A. de Cataldo

Dedicato a Mikki

Contents
1 Lecture 1: The decomposition theorem 3
1.1 Deligne’s theorem in cohomology 3
1.2 The global invariant cycle theorem 4
1.3 Cohomological decomposition theorem 5
1.4 The local invariant cycle theorem 6
1.5 Deligne’s theorem 7
1.6 The decomposition theorem 8
1.7 Exercises for Lecture 1 10
2 Lecture 2: The category of perverse sheaves P(Y) 17
2.1 Three “Whys", and a brief history of perverse sheaves 17
2.2 The constructible derived category D(Y) 19
2.3 Definition of perverse sheaves 21
2.4 Artin vanishing and Lefschetz hyperplane theorems 22
2.5 The perverse t-structure 24
2.6 Intersection complexes 25
2.7 Exercises for Lecture 2 26
3 Lecture 3: Semi-small maps 28
3.1 Semi-small maps 29
3.2 The decomposition theorem for semi-small maps 31
3.3 Hilbert schemes of points on surfaces and Heisenberg algebras 31
3.4 The endomorphism algebra End(f∗ QX ) 33
3.5 Geometric realization of the representations of the Weyl group 34
3.6 Exercises for Lecture 3 35
4 Lecture 4: Symmetries: VD, RHL, IC splits off 37
4.1 Verdier duality and the decomposition theorem 37
4.2 Verdier duality and the decomposition theorem with large fibers 38
4.3 The relative hard Lefschetz theorem 39
Received by the editors July 2, 2015.
Partially supported by N.S.F. grant DMS-1301761 and by a grant from the Simons Foundation
(#296737 to Mark de Cataldo).

©0000 (copyright holder)

1
2 Perverse sheaves and the topology of algebraic varieties

4.4 Application of RHL: Stanley’s theorem 40

4.5 Intersection cohomology of the target as a direct summand 41
4.6 Pure Hodge structure on intersection cohomology groups 43
4.7 Exercises for Lecture 4 44
5 Lecture 5: the perverse filtration 46
5.1 The perverse spectral sequence and the perverse filtration 47
5.2 Geometric description of the perverse filtration 48
5.3 Hodge-theoretic consequences 50
5.4 Character variety and Higgs moduli: P = W 50
5.5 Let us conclude with a motivic question 53
5.6 Exercises for Lecture 5 54

Goal of the lectures. The goal of these lectures is to introduce the novice to the
use of perverse sheaves in complex algebraic geometry and to what is perhaps the
deepest known fact relating the homological invariants of the source and target of
a proper map of complex algebraic varieties, namely the decomposition theorem.
Notation. A variety is a complex algebraic variety, which we do not assume
to be irreducible, nor reduced. We work with cohomology with Q-coefficients
as Z-coefficients do not fit well in our story. As we rarely focus on a single co-
homological degree, for the most part we consider the total, graded cohomology
groups, which we denote by H∗ (X, Q).
Bibliographical references. The main reference is the survey [19] and the
extensive bibliography contained in it, most of which is not reproduced here.
This allowed me to try to minimize the continuous distractions related to the
peeling apart of the various versions of the results and of the attributions. The
reader may also consult the discussions in [18] that did not make it into the very
different final version [19].
Style of the lectures and of the lecture notes. I hope to deliver my lectures in
a rather informal style. I plan to introduce some main ideas, followed by what I
believe to be a striking application, often with an idea of proof. The lecture notes
are not intended to replace in any way the existing literature on the subject, they
are a mere amplification of what I can possibly touch upon during the five one-
hour lectures. As it is usual when meeting a new concept, the theorems and the
applications are very important, but I also believe that working with examples,
no matter how lowly they may seem, can be truly illuminating and useful in
building one’s own local and global picture. Because of the time factor, I cannot
possibly fit many of these examples in the flow of the lectures. This is why there
are plenty of exercises, which are not just about examples, but at times deal head-
on with actual important theorems. I could have laid-out several more exercises
(you can look at my lecture notes [22], or at my little book [9] for more exercises),
 Mark Andrea A. de Cataldo 3

but I tried to choose ones that would complement well the lectures; too much of
anything is not a good thing anyway.
What is missing from these lectures? A lot! Two related topics come to
mind: vanishing/nearby cycles and constructions of perverse sheaves; see the
survey [19] for a quick introduction to both. To compound this infamy, there is
no discussion of the equivariant picture [3].
An afterthought. The 2015 PCMI is now over. Even though I have been away
from Mikki, Caterina, Amelie (Amie!) and Dylan for three weeks, my PCMI
experience has been wonderful. If you love math, then you should consider par-
ticipating in future PCMIs. Now, let us get to Lecture 1.

1. Lecture 1: The decomposition theorem

Summary of Lecture 1. Deligne’s theorem on the degeneration of the Leray spectral
sequence for smooth projective maps; this is the 1968 prototype of the 1982 decomposition
theorem. Application, via the use of the theory of mixed Hodge structures, to the global in-
variant cycle theorem, a remarkable topological property enjoyed by families of projective
manifolds and compactifications of their total spaces. The main theorem of these lectures,
the decomposition theorem, stated in cohomology. Application to a proof of the local in-
variant cycle theorem, another remarkable topological property concerning degenerations
of families of projective manifolds. Deligne’s theorem, including semi-simplicity of the di-
rect image sheaves, in the derived category. The decomposition theorem: the direct image
complex splits in the derived category into a direct sum of shifted and twisted intersection
complexes supported on the target of a proper map.
1.1. Deligne’s theorem in cohomology
Let us start with a warm-up: the Künneth formula and a question.
Let Y, F be varieties. Then
M
(1.1.1) H∗ (Y × F, Q) = H∗−q (Y, Q) ⊗ Hq (F, Q).
q>0

Note that the restriction map H∗ (Y × F, Q) → H∗ (F, Q) is surjective. This surjec-

tivity fails in the context of (differentiable) fiber bundles: take the Hopf fibration
b : S3 → S2 (cf. Exercise 1.7.2), for example. It is a remarkable fact that, in the
context of algebraic geometry, one has indeed this surjectivity property, and more.
Let us start discussing this phenomenon by asking the following

Question 1.1.2. Let f : X → Y be a family of projective manifolds. What can we say

about the restriction maps H∗ (X, Q) → H∗ (f−1 (y), Q)? Let X be a projective mani-
fold completing X (i.e. X is open and Zariski-dense in X). What can we say about the
restriction maps H∗ (X, Q) → H∗ (f−1 (y), Q)?

Answer: The answers are given, respectively, by (1.2.1) and by the global in-
variant cycle Theorem 1.2.2. Both rely on Deligne’s Theorem, which we review
next.
4 Perverse sheaves and the topology of algebraic varieties

The decomposition theorem has an important precursor in Deligne’s theorem,

which can be viewed as the decomposition theorem in the absence of singularities
of the domain, of the target and of the map. We start by stating the cohomological
version of his theorem.

Theorem 1.1.3. (Blanchard-Deligne 1968 theorem in cohomology [24]) For any

smooth projective map1 f : X → Y of algebraic manifolds, there is an isomorphism
∼
M
(1.1.4) H∗ (X, Q) = H∗−q (Y, Rq f∗ QX ),
q>0

where Rq f∗ QX denotes the q-th direct image sheaf of the sheaf QX via the morphism f; see
§1.2. More precisely, the Leray spectral sequence (see §1.7) of the map f is E2 -degenerate.

Proof. Exercise 1.7.3 guides you through Deligne’s classical trick (the Deligne-
Lefschetz criterion) of using the hard Lefschetz theorem on the fibers to force the
triviality of the differentials of the Leray spectral sequence. 
Compare (1.1.1) and (1.1.4): both present cohomological shifts; both express
the cohomology of the l.h.s. via cohomology groups on Y; in the former case, we
have cohomology with constant coefficients; in the latter, and this is an important
distinction, we have cohomology with locally constant coefficients.
Deligne’s theorem is central in the study of the topology of algebraic varieties.
Let us discuss one striking application of this result: the global invariant cycle
theorem.
1.2. The global invariant cycle theorem
Let f : X → Y be a smooth and projective map of algebraic manifolds, let
j : X → X be an open immersion into a projective manifold and let y ∈ Y. What are
the images of H∗ (X, Q) and H∗ (X, Q) via the restriction maps into H∗ (f−1 (y), Q)?
The answer is the global invariant cycle Theorem 1.2.2 below.
The direct image sheaf Rq := Rq f∗ QX on Y is the sheaf associated with the
pre-sheaf
U 7→ Hq (f−1 (U), Q).
In view of Ehresmann’s lemma, the proper2 submersion f is a C∞ fiber bundle.
The sheaf Rq is then locally constant with stalk
Rq
y = H (f (y), Q).
q −1

The fundamental group π1 (Y, y) acts via linear transformations on Rq

y : pick a loop
γ(t) at y and use a trivialization of the bundle along the loop to move vectors in
Rq q q
y along Rγ(t) , back to Ry (monodromy action for the locally constant sheaf R ).
q

Global sections of R identify with the monodromy invariants (Ry ) ⊆ Rq

q q π1
y.
Note, further, that this subspace is defined topologically. The cohomology group
1 Smooth: submersion; projective: factors as X → Y × P → Y (closed embedding, projection).
2 Proper := the pre-image of compact is compact; it is the “relative” version of compactness.
 Mark Andrea A. de Cataldo 5

Rqy = H (f (y), Q) has it own Hodge (p, p )-decomposition (pure Hodge struc-
q −1 0

ture of weight q), an algebro-geometric structure.

How is (Rq q
y ) 1 ⊆ Ry placed with respect to the Hodge structure?
π

The E2 -degeneration Theorem 1.1.3 yields the following immediate, yet, re-
markable, consequence:
surj
(1.2.1) Hq (X, Q) −→ (Rq
y)
π1
⊆ Rq
y = H (f (y), Q),
q −1

i.e. the restriction map in cohomology, which automatically factors through the invariants,
maps surjectively onto them.
The theory of mixed Hodge structures now tells us that the monodromy invari-
ant subspace (Rq q
y ) 1 ⊆ Ry (a topological gadget) is in fact a Hodge substructure,
π

i.e. it inherits the Hodge (p, p 0 )-decomposition (the algebro-geometric gadget).

The same mixed theory implies the highly non-trivial fact (Exercise 1.7.14) that
the images of the restriction maps from H∗ (X, Q) and H∗ (X, Q) into H∗ (f−1 (y), Q)
coincide.
We have reached the following conclusion, proved by Deligne in 1972.

Theorem 1.2.2. (Global invariant cycle theorem [26]) Let f : X → Y be a smooth

and projective map of algebraic manifolds, let j : X → X be an open immersion into
a projective manifold and let y ∈ Y. Then the images of H∗ (X, Q) and H∗ (X, Q) into
H∗ (f−1 (y), Q) coincide with the subspace of monodromy invariants. In particular, this
latter is a Hodge substructure of the pure Hodge structure Hq (f−1 (y), Q).

This theorem provides a far-reaching answer to Question 1.1.2. Note that the
Hopf examples in Exercise 1.7.2 show that such a nice general answer is not pos-
sible outside of the realm of complex algebraic geometry: there are two obstacles,
i.e. the non E2 -degeneration, and the absence of the special kind of global con-
straints imposed by mixed Hodge structures.
1.3. Cohomological decomposition theorem
The decomposition theorem is a generalization of Deligne’s Theorem 1.1.3 for
smooth proper maps to the case of arbitrary proper maps of algebraic varieties:
compare (1.1.4) and (1.3.2). It was first proved by Beilinson-Bernstein-Deligne-
Gabber in their monograph [2, Théorème 6.2.5] on perverse sheaves.
A possible initial psychological drawback, when compared with Deligne’s the-
orem, is that even if one insists in dealing with maps of projective manifolds,
the statement is not about cohomology with locally constant coefficients, but re-
quires the Goresky-MacPherson intersection cohomology groups with twisted
coefficients on various subvarieties of the target of the map. However, this is
precisely why this theorem is so striking!
To get to the point, for now we simply say that we have the intersection coho-
mology groups IH∗ (S, Q) of an irreducible variety S; they agree with the ordinary
cohomology groups when S is nonsingular. The theory is very flexible as it al-
lows for twisted coefficients: given a locally constant sheaf L on a dense open
6 Perverse sheaves and the topology of algebraic varieties

subvariety So ⊆ Sreg ⊆ S, we get the intersection cohomology groups IH∗ (S, L) of

S. We may call such pairs (S, L), enriched varieties (see [39, p.222]; this explains
the notation EV below.

Theorem 1.3.1. (Cohomological decomposition theorem) Let f : X → Y be a proper

map of complex algebraic varieties. For every q > 0, there is a finite collection EVq of
pairs (S, L) with S ⊆ Y pairwise distinct closed subvarieties of Y, and an isomorphism
∼
M M
(1.3.2) IH∗ (X, Q) = IH∗−q (S, L).
q>0 (S,L)∈EVq

Note that the same S could appear for distinct q’s, hence the notation EVq .
Deligne’s Theorem 1.1.3 in cohomology is a special case. In particular, we can
deduce an appropriate version of the global invariant cycle theorem [2, 6.2.8]. Let
us instead focus on its local counterpart.
1.4. The local invariant cycle theorem
The decomposition theorem (1.3.2) has a local flavor over the target Y, in both
the Zariski and in the classical topology: if we replace Y by an open set U ⊆ Y, X
by f−1 (U), and S by S ∩ U, then (1.3.2) remains valid.
Let us focus on the classical topology. Let X be nonsingular; this is for the sake
of our discussion, for then IH∗ (X, Q) = H∗ (X, Q).
Let y ∈ Y be a point and let us pick a small Euclidean “ball” By ⊆ Y centered
at y, so that (1.3.2) reads:
M
H∗ (f−1 (y), Q) = H∗ (f−1 (By ), Q) = IH∗−q (S ∩ By , L),
q>0,(S,L)∈EVq

where the second equality stems directly from (1.3.2), and the first one can be
seen as follows: the constructibility of the direct image complex Rf∗ QX ensures
that the second term can be identified with the stalk (R∗ f∗ QX )y , and, in turn, the
proper base change theorem ensures that this latter is the first term; see Fact 2.2.1.
Let f be surjective. Let fo : Xo → Y o the restriction of the map f over the open
subvariety of Y of regular values for f. Let yo ∈ By be a regular value for f.
By looking at Deligne theorem for the map fo it seems reasonable to expect
that for every q one of the summands in (1.3.2) should be IH∗−q (Y, Lq ), where
Lq is the locally constant sheaf Rq fo∗ Q. This is indeed the case.
If follows that for every q > 0, IH0 (By , Lq |Y o ∩By ) is a direct summand of
Hq (f−1 (y), Q), let us even say that the latter surjects onto the former. Note that
we did not assume that y ∈ Y o .
The intersection cohomology group IH0 (Y, Lq ) is the space of monodromy in-
variants for the representation π1 (Y o ∩ By , yo ) → GL(Hq (f−1 (yo ), Q). Abbreviate
the fundamental group notation to π1,loc .
We have reached a very important conclusion:

Theorem 1.4.1. (Local invariant cycle theorem, [7] and [2, (6.2.9)] Let f : X → Y
be a proper surjective map of algebraic varieties with X nonsingular. Let y ∈ Y be any
 Mark Andrea A. de Cataldo 7

point, let By be a small Euclidean ball on Y at y, let yo ∈ By be a regular value of f.

Then H∗ (f−1 (y), Q) = H∗ (f−1 (By ), Q) surjects onto the local monodromy invariants
H∗ (f−1 (yo ), Q)π1,loc .

1.5. Deligne’s theorem

In fact, Deligne proved something stronger than his cohomological theorem
(1.1.4), he proved a decomposition theorem for the derived direct image complex
under a smooth proper map.
Pre-warm-up: cohomological shifts. Given a Z-graded object K = ⊕i∈Z Ki ,
like the total cohomology of a variety, or a complex (of sheaves, for example) on
it, or the total cohomology of such a complex, etc., and given an integer a ∈ Z,
we can shift by the amount a and get a new graded object (with Ki+a in degree i)
M
(1.5.1) K[a] := Ki+a .
i∈Z
If a > 0, then the effect of this operation is to “shift K back by a units.” Again,
if K has non zero entries contained in an interval [m, n], then K[a] has non zero
entries contained in [m − a, n − a]. We have the following basic relation, e.g. for
complexes of sheaves
Hi (K[a]) = Hi+a (K).
A sheaf F can be viewed as a complex placed in cohomological degree zero; we
can then take the F[a]’s. We can take a collection of Fq ’s and form ⊕q Fq [−q],
which is a complex with trivial differentials. Then
H∗ (Y, ⊕q Fq [−q]) = ⊕q H∗−q (Y, Fq ).
Warm-up: Künneth for the derived direct image. Let f : X := Y × F → Y be
the projection. Then there is a canonical isomorphism
(1.5.2) Rf∗ QX = ⊕q>0 Hq (F)[−q]
where Hq (F) is the constant sheaf on Y with stalk Hq (F, Q). The isomorphism
takes place in the derived category of the category of sheaves of rational vector
spaces on Y; this is where we find the direct image complex Rf∗ QX , whose coho-
mology is the cohomology of X: H∗ (Y, Rf∗ QX ) = H∗ (X, Q). Exercise 1.7.23 asks
you to prove (1.5.2).
Now to Deligne’s 1968 theorem.

Theorem 1.5.3. (Deligne’s 1968 theorem [24]; semi-simplicity in 1972 [26, §4.2] )
Let f : X → Y be a smooth proper map of algebraic varieties. “The derived direct image
complex has trivial differentials”, more precisely, there is an isomorphism in the derived
category
∼
M
(1.5.4) Rf∗ QX = Rq f∗ Q [−q].
q>0

Moreover, the locally constant direct image sheaves Rq f∗ QX are semi-simple.

8 Perverse sheaves and the topology of algebraic varieties

Equation (1.5.4), which is proved by means of an E2 -degeneration argument3

along the lines of the one in Exercise 1.7.3, is the “derived” version of (1.1.4),
which follows by taking cohomology on both sides of (1.5.4). In addition to [24],
you may want to consult the first two pages of [30]. The semi-simplicity result is
one of the many amazing applications of the theory of weights (Hodge-theoretic,
or Frobenius).
Terminology and facts about semi-simple locally constant sheaves. To give
a locally constant sheaf on Y is the same as giving a representation of the funda-
mental group of Y (Exercise 1.7.10). By borrowing from the language of represen-
tations, we have the notions of simple (no non-trivial locally constant subsheaf;
a.k.a. irreducible) and semi-simple (direct sum of simples; a.k.a. completely re-
ducible), indecomposable (no non-trivial direct sum decomposition) locally con-
stant sheaves.
Once one has semi-simplicity, one can decompose further. For a semi-simple
locally constant sheaf L, we have the canonical isotypic direct sum decomposition
(1.5.5) L = ⊕χ Lχ ,
where each summand is the span of all mutually isomorphic simple subobjects,
and the direct sum ranges over the set of isomorphism classes of irreducible
representations of the fundamental group. In particular, in (1.5.4), we have
Rq f∗ QX = Rq = ⊕χ Rq χ.
What is semi-simplicity good for? Here is the beginning of an answer: look
at Exercise 4.7.4, where it is put to good use to give Deligne’s proof in [29] of
the Hard Lefschetz theorem. In the context of the decomposition theorem, the
semi-simplicity of the perverse direct images is an essential ingredient in the
proof of the relative hard Lefschetz theorem; see [2] and [16, especially, §5.1 and
§6.4]. Keep in mind that the local systems arising naturally in algebraic geometry
are typically not semisimple (cf. G. Williamson’s Math Overflow post on “non
semi-simple monodromy in an algebraic family").
1.6. The decomposition theorem
As we have seen, Deligne’s theorem in cohomology has a counterpart in the
derived category. The cohomological decomposition Theorem 1.3.1 also has a
stronger counterpart in the derived category, i.e Theorem 1.6.2.
In these lectures, we adopt a version of the decomposition theorem that is
more general (and simpler to state!) than the one in [2, 6.2.5] (coefficients of
geometric origin) and of [48] (coefficients in polarizable variations of pure Hodge
structures). The version we adopt is due essentially to T. Mochizuki [41, 42] (with
important contributions of C. Sabbah [47]) and it involves semi-simple coefficients.
Mochizuki’s [41, 42] deals with projective maps of quasi-projective varieties and
3 People refer to it as the Deligne-Lefschetz criterion.
 Mark Andrea A. de Cataldo 9

with C-coefficients; one needs a little bit of tinkering to reach the same conclu-
sions for proper maps of complex varieties with Q-coefficients (to my knowledge,
this is not in the literature).
Warning: IC vs. IC. We are about to meet the main protagonists of our
lectures, the intersection complexes ICS (L) with twisted coefficients; in fact, the
actual protagonists are the shifted (see (1.5.1) for the notion of shift):
(1.6.1) ICS (L) := ICS (L)[dim S],
which are perverse sheaves on S and on any variety Y for which S ⊆ Y is
closed. Note that ICS (L) only has non-trivial cohomology sheaves in the inter-
val [0, dim S − 1], the analogous interval for ICS (L) is [− dim S, −1]. Both IC and
IC are called intersection complexes. Instead of discussing the pros and cons of
either notation, let us move on.
Brief on intersection complexes. The intersection cohomology groups of an
enriched variety (S, L) are in fact the cohomology groups of S with coefficients
in a very special complex of sheaves called the intersection complex of S with
coefficients in L and denoted by ICS (L): we have IH∗ (S, L) = H∗ (S, ICS (L)). If
S is nonsingular, and L is constant of rank one, then ICS = ICS (Q) = QS . The
decomposition theorem in cohomology (1.3.2) is the shadow in cohomology of
a decomposition of the direct image complex Rf∗ ICX in the derived category of
sheaves of rational vector spaces on Y. In fact, the decomposition theorem holds
in the greater generality of semi-simple coefficients.

Theorem 1.6.2. (Decomposition theorem) Let f : X → Y be a proper map of complex

algebraic varieties. Let ICX (M) be the intersection complex of X with semi-simple twisted
coefficients M. For every q > 0, there is a finite collection EVq of pairs (S, L) with S
pairwise distinct4 and L semi-simple, and an isomorphism
Rf∗ ICX (M) =∼L
(S,L)∈EVq ICS (L)[−q].
L
(1.6.3) q>0

In particular, by taking cohomology:

∼
M
(1.6.4) IH∗ (X, M) = IH∗−q (S, L).
q,EVq

We have the isotypic decompositions (1.5.5), which can be plugged into (1.6.4).

Remark 1.6.5. The fact that there may be summands associated with S 6= Y should
not come as a surprise. It is a natural fact arising from to the singularities (de-
viation from being smooth) of the map f. One does not need the decomposition
theorem to get convinced: the reader can work out the case of the blowing up of
the affine plane at the origin; see also Exercise 1.7.20. In general, it is difficult to
predict which S will appear in the decomposition theorem; see parts 5 and 7 of
Exercise 1.7.21.

4 The same S could appear for distinct q’s.

10 Perverse sheaves and the topology of algebraic varieties

1.7. Exercises for Lecture 1

Exercise 1.7.1. (Ehresmann’s lemma and local constancy of higher direct images
for proper submersions) Let f : X → Y be a map of varieties and recall that the
q-th direct image sheaf Rq := Rq f∗ QX is defined to be the sheafification of the
presheaf Y ⊇ U 7→ Hq (f−1 (U), Q). If f admits the structure of a C∞ fiber bundle,
then the sheaves Rq are locally constant, with stalks the cohomology of the fibers.
Give examples of maps f : X → Y, where the stalks (Rq f∗ QX )y of the direct
image of the constant sheaf differ from the cohomology groups H∗ (f−1 (y), Q)
of the corresponding fibers (hint: the maps cannot be proper). If f is a proper
smooth map of complex algebraic varieties, then it admits a structure of C∞ fiber
bundle (Ehresmann’s lemma). Deduce that nonsingular hypersurfaces of fixed
degree in complex projective space are all diffeomorphic to each other. Is the
same true in real projective space? Why?
Quick review of the Leray spectral sequence (see Grothendieck’s gem [34],
a.k.a. “Tohoku”). The Leray spectral sequence for a map f : X → Y (and for
the sheaf QX ) is a gadget denoted Epq2 = Hp (Y, Rq f∗ QX ) ⇒ Hp+q (X, Q). There
are the differentials dr : Epq
r → Erp+r,q−r+1 , d2r = 0, with r > 2 and for which
∗
Er+1 = H (Er , dr ). E2 -degeneration means that dr = 0 for every r > 2, so
that one has a cohomological decomposition H∗ (X, Q) = ∼ ⊕q>0 H∗−q (Y, Rq f∗ QX ).
Note that with Z coefficients, E2 -degeneration does not imply the existence of an
analogous splitting.
Exercise 1.7.2. (Maps of Hopf-type) Let a : C2 \ o → P1 = ∼ S2 be the usual map
(x, y) 7→ (x : y) from the affine plane with the origin o deleted onto the projec-
tive line. It induces maps, b : S3 → S2 and c : HS := (C2 \ o)/Z → P1 (where
1 ∈ Z acts as multiplication by two). These three maps are fiber bundles. Show
that there cannot be a cohomological decomposition as in (1.1.4). Deduce that
their Leray spectral sequences are not E2 -degenerate. Observe that the conclu-
sion of the global invariant cycle theorem concerning the surjectivity onto the
monodromy invariants fails in all three cases.
Exercise 1.7.3. (Proof of the cohomological decomposition (1.1.4) via hard Lef-
schetz) Let us recall the hard Lefschetz theorem: let X be a projective manifold of
dimension d, and let η ∈ H2 (X, Q) be the first Chern class of an ample line bun-
dle on X; then the iterated cup product maps ηd−q : Hq (X, Q) → H2d−q (X, Q)
are isomorphisms for every q > 0. Deduce the primitive Lefschetz decomposi-
tion: for every q 6 d, set Hq prim := Ker {η
d−q+1 : Hq → H2d−q+2 }; then, for

every 0 6 q 6 d, we have Hq = ⊕j>0 Hq−2j prim , and, for d 6 q 6 2d, we have

Hq = ηq−d ∪ (⊕j>0 Hq−2d−2j
prim ). Let f : X → Y be as in (1.1.4), i.e. smooth and
projective and let d := dim X − dim Y. Apply the hard Lefschetz theorem to the
fibers of the smooth map f and deduce the analogue of the primitive Lefschetz
decomposition for the direct image sheaves Rq := Rf∗ QX . Argue that in order to
deduce (1.1.4) it is enough to show the differentials dr of the spectral sequence
 Mark Andrea A. de Cataldo 11

vanish on Hp (Y, Rq
prim ) for every q 6 d. Use the following commutative diagram,
with some entries left blank on purpose for you to fill-in, to deduce that indeed
we have the desired vanishing:
d?
H? (Y, Rq / H? (Y, ?)
prim )

η? η?
 d?

H? (Y, ?) / H? (Y, ?).

(Hint: the right power of η kills a primitive class in degree q, but is injective
in degree q − 1.) Remark: the refined decomposition (1.5.4) is proved in a sim-
ilar way by replacing the spectral sequence above with the analogous one for
Hom(Rq [−q], Rf∗ QX ): first you prove it is E2 -degenerate; then you lift the iden-
tity Rq → Rq to a map in Hom(Rq [−q], Rf∗ QX ) inducing the identity on Rq ;
see [30].

Heuristics for E2 -degeneration and for semi-simplicity of the Rq f∗ QX via

weights. (What follows should be considered as a very informal fireside chat.) It
seems that Deligne guessed at E2 -degeneration by looking at the same situation
over the algebraic closure of a finite field by considerations (“the yoga of weights”
[25, 28]) of the size (weight) of the eigenvalues of the action of Frobenius on the
entries Epqr : they should have weight something analogous to exp (p + q) (we are
using the exponential function as an analogy only, one needs to say more, but we
shan’t) so the Frobenius-compatible differentials must be zero. There are similar
heuristic considerations for Deligne’s theorem to the effect that the Rq are semi-
simple: if 0 → M → Rq → N → 0 is a short exact sequence, then Frobenius acts
on Ext1 (N, M) with weight exp (1); take M ⊆ Rq to be the maximal semi-simple
subobject; then the corresponding extension is invariant under Frobenius and
has weight zero (1 = exp (0)); it follows that the extension splits and the biggest
semi-simple in Rq splits off: Rq = ∼ M ⊕ N; if the resulting quotient N were non-
trivial, then it would contain a non-trivial simple that then, by the splitting, would
enlarge the biggest semi-simple M in Rq ; contradiction. This kind of heuristic is
now firmly based in deep theorems by Deligne and others [2, 29] for varieties
over finite fields and their algebraic closure, and by M. Saito [48] in the context
of mixed Hodge modules over complex algebraic varieties.

Exercise 1.7.4. (Rank one locally constant sheaves) Take [0, 1] × Q and identify
the two ends, {0} × Q and {1} × Q, by multiplication by −1 ∈ Q∗ . Interpret this
as a rank one locally constant sheaf on S1 that is not constant. Do the same,
but multiply by 2. Do the same, but first replace Q with Q and multiply by a
root of unity. Show that the tensor product operation (L, M) → L ⊗ M induces
the structure of an abelian group on the set of isomorphisms classes of rank one
locally constant sheaves on S1 . Determine the torsion elements of this group
when you replace Q with Q. Show that if we replace S1 with any connected
12 Perverse sheaves and the topology of algebraic varieties

variety and Q with C, then we obtain the structure of a complex Lie group (a.k.a.
the character variety for rank one complex representations; one can define it for
arbitrary rank, but needs geometric invariant theory to do so).
Exercise 1.7.5. (Locally constant sheaves and representations of the fundamen-
tal group) A locally constant sheaf (a.k.a. a local system) L on Y gives rise to
a representation ρL : π1 (Y, y) → GL(Ly ): pick a loop γ(t) at y and use local
trivializations of L along the loop to move vectors in Ly along Lγ(t) , back to Ly .
Exercise 1.7.6. (Representations of the fundamental group and locally constant
sheaves) Given a representation ρ : π1 (Y, y) → GL(V) into a finite dimensional
vector space, consider a universal cover (Y,
ey e) → (Y, y), build the quotient space
(V × Y)/π1 (Y, y), take the natural map (projection) to Y and take the sheaf of
e
its local sections. Show that this is a locally constant sheaf whose associated
representation is ρ.
Exercise 1.7.7. (Zeroth cohomology of a local system) Let X be a connected space.
Let L be a local system on X, and write M for the associated π1 (X)-representation.
Show that
H0 (X; L) = Mπ1 (X) ,
where the right hand side is the fixed part of M under the π1 (X)-action.
Exercise 1.7.8. (Cohomology of local systems on a circle) Fix an orientation of
S1 and the generator T ∈ π1 (S1 ) that comes with it. Let L be a local system on S1
with associated monodromy representation M. Show that
H0 (S1 , L) = ker((T − id) : M → M), H1 (S1 , L) = coker((T − id) : M → M),
and that H>1 (S1 , L) = 0.
(Hint: one way to proceed is to use Cech cohomology. Alternatively, embed
S1 as the boundary of a disk and use relative cohomology (or dualize and use
compactly supported cohomology, where the orientation is easier to get a handle
on)).
Exercise 1.7.9. (Fiber bundles over a circle: the Wang sequence) This is an exten-
sion of the previous exercise. Let f : E → S1 be a locally trivial fibration with fibre
F and monodromy isomorphism T : F → F. Show that the Leray spectral sequence
gives rise to a short exact sequence
0 → H1 (S1 , Rq−1 f∗ Q) → Hq (E, Q) → H0 (S1 , Rq f∗ Q) → 0.
Use the previous exercise to put these together into a long exact sequence
. . . → Hq (E, Q) −→ Hq (F, Q) −→ Hq (F, Q) → Hq+1 (E) → . . .
where the middle map Hq (F; Q) → Hq (F; Q) is given by T ∗ − id. Make a connec-
tion with the theory of nearby cycles (which are not discussed in these notes).
Exercise 1.7.10. (The abelian category Loc(Y)) Show that the abelian category
Loc (Y) of locally constant sheaves of finite rank on Y is equivalent to the abelian
 Mark Andrea A. de Cataldo 13

category of finite dimensional π1 (Y, y)-representations. Show that both cate-

gories are noetherian (acc ok!), artinian (dcc ok!) and have a duality anti-self-
equivalence.

Exercise 1.7.11 below is in striking contrast with the category Loc, but also
with the one of perverse sheaves, which admits, by its very definition, the anti-
self-equivalence given by Verdier duality.

Exercise 1.7.11. (The abelian category Shc (Y) is not artinian.) Show that in
the presence of such an anti-self-equivalence, noetherian is equivalent to artinian.
Observe that the category Shc (Y) whose objects are the constructible sheaves (i.e.
`
there is a finite partition of Y = Yi into locally closed subvarieties to which the
sheaf restricts to a locally constant one (always assumed to be of finite rank!) is
abelian and noetherian, but it is not artinian. Deduce that Shc (Y) does not admit
an anti-self-equivalence. Give an explicit example of the failure of dcc in Shc (Y).
Prove that Shc (Y) is artinian if and only if dim Y = 0.

Exercise 1.7.12. (Cyclic coverings) Show that the direct image sheaf sheaf R0 f∗ Q
for the map S1 → S1 , t → tn is a semi-simple locally constant sheaf of rank n;
find its simple summands (one of them is the constant sheaf QS1 and the resulting
splitting is given by the trace map). Do the same for R0 f∗ Q.

Exercise 1.7.13. (Indecomposable non simple) The rank two locally constant
sheaf on S1 given by the non-trivial unipotent 2 × 2 Jordan block is indecom-
posable and is neither simple nor semi-simple. Make a connection between this
locally constant sheaf and the Picard-Lefschetz formula for the degeneration of a
curve of genus one to a nodal curve.

Amusing monodromy dichotomy. There is an important and amusing di-

chotomy concerning local systems in algebraic geometry (which we state infor-
mally): the global local systems arising in complex algebraic geometry via the
decomposition theorem are semi-simple (i.e. completely reducible; related to the
Zariski closure of the image of the fundamental group in the general linear group
being reductive); the restriction of these local systems to small punctured disks
with centers at infinity (degenerations), are quasi-unipotent, i.e. unipotent after
taking a finite cyclic covering if necessary. This local quasi-unipotency is in some
sense the opposite of the global complete reducibility.
Quick review of Deligne’s 1972 and 1974 theory of mixed Hodge structures
[25–27]; see also [31]. Deligne discovered the existence of a remarkable structure,
a mixed Hodge structure, on the singular cohomology of a complex algebraic va-
riety X (not necessarily smooth, complete or irreducible): there is an increasing fil-
tration Wk H∗ (X, Q) and a decreasing filtration Fp H∗ (X, C) (with conjugate filtra-
pq
k H (X, C) = ⊕p+q=k Hk ,
tion denoted by F) such that the graded quotients GrW ∗

where the splitting is induced by the (conjugate and opposite) filtrations F, F; i.e.
(W, F, F) induce pure Hodge structure of weight k on GrW k . This structure is
14 Perverse sheaves and the topology of algebraic varieties

canonical and functorial for maps of complex algebraic varieties. Important: one
has that a map of mixed Hodge structures f : A → B is automatically strict,
i.e. if f(a) ∈ Wk B, then there is a 0 ∈ Wk A with f(a 0 ) = f(a). Kernels, im-
ages and cokernels of pull-back maps in cohomology inherit such a structure.
If X is a projective manifold, we get the known Hodge (p, q)-decomposition:
Hi (X, C) = ⊕p+q=i Hpq (X). It is important to take note that for each fixed i
pq
we have Hi (X, C) = ⊕k GrW k H (X, C) = ⊕k ⊕p+q=k Hk (X) which may admit
i

several non zero k summands for k 6= i. In this case, we say that the mixed
Hodge structure is mixed. This happens for the projective nodal cubic: H1 = H0,00 ,
1 1,1
and for the punctured affine line H = H2 . Here are some “inequalities” for the
weight filtration: GrW d
k H = 0 for k ∈/ [0, 2d]; if X is complete, then GrW d
k>d H = 0;
if X is nonsingular, then GrW d d
k<d H = 0 and Wd H is the image of the restriction
map from any nonsingular completion (open immersion in proper nonsingular);
if X → Y is surjective and X is complete nonsingular, then the kernel of the pull-
back to Hd (X) is Wd−1 Hd (Y).
Exercise 1.7.14. (Amazing weights) Let Z → U → X be a closed immersion with
Z complete followed by an open dense immersion into a complete nonsingular
variety. Use some of the weight inequalities listed above, together with strictness,
to show that the images of H∗ (X, Q) and H∗ (U, Q) into H∗ (Z, Q) coincide. Build
a counterexample in complex geometry (Hopf!). Build a counterexample in real
algebraic geometry (circle, bi-punctured sphere, sphere).
The reader is invited to produce an explicit example of a projective normal sur-
face having mixed singular cohomology. Morally speaking, as soon as you leave
the world of projective manifolds and dive into the one of projective varieties,
“mixedness” is the norm.
For an explicit example of a proper map with no cohomological decomposition
analogous to (1.1.4), see Exercise 1.7.18. We can produce many by pure-thought
using Deligne’s theory of mixed Hodge structures. Here is how.
Exercise 1.7.15. (In general, there is no decomposition Rf∗ QX = ∼ ⊕Rq f∗ QX [−q])
Pick a normal projective variety Y whose singular cohomology is a non pure
mixed Hodge structure. Resolve the singularities f : X → Y. Use Zariski’s main
theorem to show that R0 f∗ QX = QY . Show that, in view of the the mixed-not-
pure assumption, the map of mixed Hodge structures f∗ is not injective. Deduce
that QY is not a direct summand of Rf∗ QX and that, in particular, there is no
decomposition Rf∗ QX = ∼ ⊕Rq f∗ QX [−q] in this case. (In some sense, the absence
of such a decomposition is the norm for proper maps of varieties.)
Exercise 1.7.16. (The affine cone Y over a projective manifold V) Let V d ⊆ P
be an embedded projective manifold of dimension d and let Y d+1 ⊆ A be its
affine cone with vertex o. Let j : U := Y \ {o} → Y be the open embedding. Show
that U is the C∗ -bundle over V of the dual to the hyperplane line bundle for the
given embedding V ⊆ P. Determine H∗ (U, Q). Answer: for every for 0 6 q 6
 Mark Andrea A. de Cataldo 15

d, Hq (U) = Hq prim (V) and H

1+d+q (U) = Hd−q (V). Show that R0 j Q
prim ∗ U = QY
and that, for q > 0, Rq j∗ QU is skyscraper at o with stalk Hq (U, Q). Compute
Hqc (Y, Q). Give a necessary and sufficient condition on the cohomology of V that
ensures that Y satisfies Poincaré duality Hq (Y, Q) = ∼ Hc2d+2−q (Y, Q). Observe
that if V is a curve this condition boils down to it having genus zero. Remark:
once you know about a bit about Verdier duality, this exercise tells you that the
complex QY [dim Y] is Verdier self dual if and only if V meets the condition you
have identified above; in particular, it does not if V is a curve of positive genus.

Fact 1.7.17. (ICY , Y a cone over a projective manifold V) Let things be as in

Exercise 1.7.16. By adopting the definition of the intersection complex as an
iterated push-forward followed by truncations, as originally given by Goresky-
MacPherson, the intersection complex of Y is defined to be ICY := τ6d Rj∗ QU ,
where we are truncating the image direct complex Rj∗ QU in the following way:
keep the same entries up to degree d − 1, replace the d-th entry by the kernel of
the differential exiting it and setting the remaining entries to be zero; the resulting
cohomology sheaves are the same as the ones for Rj∗ QU up to degree d included,
and they are zero afterwards. More precisely, the cohomology sheaves of this
complex are as follows: Hq = 0 for q ∈ / [0, d], H0 = QY , and for 1 6 q 6 d, Hi
q
is skyscaper at o with stalk Hprim (V, Q). Here is a justification for this definition:
while Q[dim Y] usually fails to be Verdier self-dual, one can verify directly that
the intersection complex ICY := ICY [dim Y] is Verdier self-dual. In general, if
we were to truncate at any other spot, then we would not get this self-duality
behavior (unless we truncate at minus 1 and get zero). Note that the knowledge
of the cohomology sheaves of a complex, e.g. ICY , is important information, but
it does not characterize the complex up to isomorphism.

Exercise 1.7.18. (Example of no decomposition Rf∗ QX = ∼ ⊕Rq f∗ QX [−q]) Let

things be as in Exercise 1.7.16 and assume that V is a curve of positive genus,
so that Y is a surface. Let f : X → Y be the resolution obtained by blowing up the
vertex o ∈ Y. Use the failure of the self-duality of QY [2] to deduce that QY is not
a direct summand of Rf∗ QX . Deduce that Rf∗ QX 6' ⊕Rq f∗ QX [−q]. (The reader
is invited to check [17, §3.1] out: it contains an explicit computation dealing with
this example showing that as you try split QY off Rf∗ QX , you meet an obstruc-
tion; instead, you end up splitting ICY off Rf∗ QX , provided you have defined ICY
as the truncated push-forward as above.)

Fact 1.7.19. (Intersection complexes on curves) Let Y o be a nonsingular curve

and L be a locally constant sheaf on it. Let j : Y o → Y be an open immersion
into another curve (e.g. a compactification). Then ICY (L) = j∗ L[1] (definition of
IC via push-forward/truncation). Note that if y ∈ Y is a nonsingular point, then
the stalk (j∗ L)y is given by the local monodromy invariants of L around a small
loop about y. The complex Rj∗ L[1] may fail to be Verdier self-dual, whereas its
truncation τ6−1 Rj∗ L[1] = j∗ L[1] = ICY (L) is Verdier self-dual. Note that we have
16 Perverse sheaves and the topology of algebraic varieties

a factorization Rj! L[1] → ICY (L) → Rj∗ L[1]. This is not an accident; see the end of
§2.6.

Exercise 1.7.20. (Blow-ups) Compute the direct image sheaves Rq f∗ Q for the
blowup of Cm ⊆ Cn (start with m = 0; observe that there is a product decom-
position of the situation that allows you to reduce to the case m = 0). Same
question for the composition of the blow up of C1 ⊆ C3 , followed by the blow-
ing up of a positive dimensional fiber of the first blow up. Observe that in all
cases, one gets an the decomposition Rf∗ Q =∼ ⊕Rq f∗ Q[−q]. Guess the shape of
the decomposition theorem in both cases.

Exercise 1.7.21. (Examples of the decomposition theorem) Guess the exact form
of the cohomological and “derived” decomposition theorem in the following
cases: 1) the normalization of a cubic curve with a node and of a cubic curve
with a cusp; 2) the blowing up of a smooth subvariety of an algebraic manifold; 3)
compositions of various iterations of blowing ups of nonsingular varieties along
smooth centers; 4) a projection F × Y → Y; 5) the blowing up of the vertex of
the affine cone over the nonsingular quadric in P3 ; 6) same but for the projective
cone; 7) blow up the same affine and projective cones but along a plane through
the vertex of the cone; 8) the blowing up of the vertex of the affine/projective
cone over an embedded projective manifold.

Exercise 1.7.22. (decomposition theorem for Lefschetz pencils) Guess the shape
of the decomposition theorem for a Lefschetz pencil f : X e → P1 on a nonsingular
projective surface X. Work out explicitly the invariant cycle theorems in this case.
Do the same for a nonsingular projective manifold. When do we get skyscraper
contributions?

Exercise 1.7.23. (Künneth for the derived image complex) One needs a little bit
of working experience with the derived category to carry out what follows below.
But try anyway. Let f : X := Y × F → Y. A class aq ∈ Hq (X, Q) is the same thing
as a map in the derived category aq : QX → QX [q]. First pushing forward via
Rf∗ , then observing that Rf∗ f∗ QY = Rf∗ QX , and, finally, pre-composing with the
adjunction map QY → Rf∗ QX , yields a map aq : QY → Rf∗ QX [q]. Take aq to be
of the form pr∗F αq . Obtain a map αq : Hq (F) → Rf∗ QX [q]. Next, shift this map to
get αq : Hq (F)[−q] → Rf∗ QX . Show that the map induces the “identity” on the
q-th direct image sheaf and zero on the other direct image sheaves. Deduce that
P
q αq : ⊕q H (F)[−q] → Rf∗ QX is an isomorphism in the derived category in-
q

ducing the “identity” on the cohomology sheaves. Observe that you did not make
any choice in what above, i.e. the resulting isomorphism is canonical, whereas in
Deligne’s theorem 1.5.3, one does not obtain a canonical isomorphism.

Exercise 1.7.24. (Deligne’s theorem as a special case of the decomposition the-

orem) Keeping in mind that if So = S, then ICS (L) = L, recover the Deligne
theorem from the decomposition theorem.
 Mark Andrea A. de Cataldo 17

2. Lecture 2: The category of perverse sheaves P(Y)

Summary of Lecture 2. The constructible derived category. Definition of perverse
sheaves. Artin vanishing and its relation to a proof of the Lefschetz hyperplane theo-
rem for perverse sheaves. The perverse t-structure (really, only the perverse cohomology
functors!). Beilinson’s and Nori’s equivalence theorems. Several equivalent definitions of
intersection complexes.
2.1. Three “Whys", and a brief history of perverse sheaves
Why intersection cohomology? Let us look at (1.3.2) for X and Y nonsingular:
∼ ⊕q,EV IH∗−q (S, L),
H∗ (X, Q) = q

i.e. the l.h.s. is ordinary cohomology, but the r.h.s. is not any kind of ordinary co-
homology on Y: we need intersection cohomology to state the decomposition the-
orem, even when X and Y are nonsingular. The intersection cohomology groups
of a projective variety enjoy a battery of wonderful properties (Poincaré-Hodge-
Lefschetz package). In some sense, intersection cohomology nicely replaces sin-
gular cohomology on singular varieties, but with a funny twist: singular coho-
mology is functorial, but has no Poincaré duality; intersection cohomology has
Poincaré duality, but is not functorial!
Why the constructible derived category? The cohomological Deligne theorem
(1.1.4) for smooth projective maps is a purely cohomological statement and it can
be proved via purely cohomological methods (hard Lefschetz + Leray spectral
sequence). The cohomological decomposition theorem (1.3.2) is also a cohomo-
logical statement. However, there is no known proof of this statement that does
not make use of the formalism of the middle perversity t-structure present in the
constructible derived category: one proves the derived version (1.6.3) and then
deduces the cohomological one (1.3.2) by taking cohomology. Actually, the defi-
nition of perverse sheaves does not make sense if we take the whole derived cat-
egory, we need to take complexes with cohomology sheaves supported at closed
subvarieties (not just classically closed subsets). We thus restrict to an agreeable,
yet flexible, class of complexes: the “constructible complexes”.
Why perverse sheaves? Intersection complexes, i.e. the objects appearing on
both sides of the decomposition theorem (1.6.3) are very special perverse sheaves.
In fact, in a precise way, they form the building blocks of the category of perverse
sheaves: every perverse sheaf is an iterated extension of a collection of intersec-
tion complexes. Perverse sheaves satisfy their own set of beautiful properties:
Artin vanishing theorem, Lefschetz hyperplane theorem, stability via duality, sta-
bility via vanishing and nearby cycle functors. As mentioned above, the known
proofs of the decomposition theorem use the machinery of perverse sheaves.
A brief history of perverse sheaves.
Intersection complexes were invented by Goresky-MacPherson as a tool to sys-
tematize, strengthen and widen the scope of their own intersection cohomology
18 Perverse sheaves and the topology of algebraic varieties

theory. For example, their original geometric proof of Poincaré duality can be
replaced by the self-duality property of the intersection complex. See also S.
Kleiman’s very entertaining survey [38].
The conditions leading to the definition of perverse sheaves appeared first in
connection with the Riemann-Hilbert correspondence established by Kashiwara
and by Mekbouth: their result is an equivalence of categories between the con-
structible derived category (which we have been procrastinating to define) and
the derived category of regular holonomic D-modules (which we shall not de-
fine); the standard t-structures, given by the standard truncations met in Exer-
cise 1.7.17, of these two categories do not correspond to each other under the
Riemann-Hilbert equivalence; the conditions leading to the “conditions of sup-
port” defining of perverse sheaves are the (non-trivial) translation in the con-
structible derived category of the conditions on the D-module side stating that a
complex of D-modules has trivial cohomology D-modules in positive degree. It
is a seemingly unrelated, yet remarkable and beautiful fact, that the conditions of
support so obtained are precisely what makes the Artin vanishing Theorem 2.4.1
work on an affine variety.
As mentioned above, Gelfand and MacPherson conjectured the decomposition
theorem for Rf∗ ICX . Meanwhile, Deligne had developed a theory of pure com-
plexes for varieties defined over finite fields and established the invariance of
purity under push-forward by proper maps. Gabber proved that the intersection
complex of a pure local system, in that context, is pure. The four authors of [2]
introduced and developed systematically the basis for the theory of t-structures,
especially with respect to the middle perversity. They then proved that the no-
tions of purity and perverse t-structure are compatible: a pure complex splits
over the algebraic closure of the finite field as prescribed by the r.h.s. of (1.6.3).
The decomposition theorem over the algebraic closure of a finite field follows
when considering the purity result for the proper direct image mentioned above.
The whole of Ch. 6 in [2], aptly named “De F à C”, is devoted to explaining
how these kind of results over the algebraic closure of a finite field yield results
over the field of complex numbers. This established the original proof of the
decomposition theorem over the complex numbers for semi-simple complexes of
geometric origin (see [2, 6.2.4, 6.2.5]), such as ICX .
M. Saito has developed in [48] the theory of mixed Hodge modules which
yields the desired decomposition theorem when M underlies a variation of polar-
izable pure Hodge structures.
M. A. de Cataldo and L. Migliorini have given a proof based on classical Hodge
theory of the decomposition theorem when M is constant [16].
Finally, the decomposition theorem stated in (1.6.3) is the most general state-
ment currently available over the complex numbers and is due to work of C.
Sabbah [47] and T. Mochizuki [41, 42] (where this is done in the essential case of
projective maps of quasi projective manifolds; it is possible to extend it to proper
 Mark Andrea A. de Cataldo 19

maps of algebraic varieties). The methods (tame harmonic bundles, D-modules)

are quite different from the ones discussed in these lectures.
2.2. The constructible derived category D(Y)
The decomposition theorem isomorphisms (1.6.3) take place in the “construct-
ible derived category” D(Y). It is probably a good time to try and give an idea
what this category is.
Constructible sheaf. A sheaf F on Y is constructible if there is a finite disjoint
`
union decomposition Y = a Sa into locally closed subvarieties such that the
restriction F|Sa are locally constant sheaves of finite rank. This is a good time to
look at Exercise 2.7.1.
Constructible complex. A complex C of sheaves of rational vector spaces on Y
is said to be constructible if it is bounded (all but finitely many of its cohomology
sheaves are zero) and its cohomology sheaves are constructible sheaves. See the
most-important Fact 2.2.1.
Constructible derived category. The definition of D(Y) is kind of a mouthful:
it is the full subcategory of the derived category D(Sh(Y, Q)) of the category of sheaves
of rational vector spaces whose objects are the constructible complexes.
It usually takes time to absorb these notions and to absorb the apparatus it
gives rise to. We take a different approach and we try to isolate some of the
aspects of the theory that are more relevant to the decomposition theorem. We
do not dwell on technical details.
Cohomology. Of course, the first functors to consider are cohomology and
cohomology with compact supports Hi (Y, −), Hic (Y, −) : D(Y) → D(point). They
can be seen as special cases of derived direct images.
Derived direct images Rf∗ , Rf! . We can define derived direct image maps
Rf∗ , Rf! : D(X) → D(Y), for every map f : X → Y. The first thing to know is that
H∗ (X, C) = H∗ (Y, Rf∗ C) and that H∗c (X, C) = H∗c (Y, Rf! C), so that we may view
them as generalizing cohomology.
Pull-backs. The pull-back functor f∗ is probably the most intuitive one. The
extraordinary pull-back functor f! is tricky and we will not dwell on it. It is the
right adjoint to Rf! ; for open immersions, f! = f∗ ; for closed immersions it is the
derived version of the sheaf of sections supported on the closed subvariety; for
smooth maps of relative dimension d, f! = f∗ [2d]. A down-to-earth reference for
f! and duality I like is [35] (good also, among other things, as an introduction to
Borel-Moore homology). I also like [32]. There is also the seemingly inescapable,
and nearly encyclopedic [37].

Fact 2.2.1. A good reference is [4]. Given C ∈ D(Y), and y ∈ Y there is a system
of “standard neighborhoods” Uy () (think of 0 <   1 as the radius of an
Euclidean ball; of course, our Uy () are singular, if Y is singular at y) such that
H∗ (Uy (), C) and H∗c (Uy (), C) are “constant” (make the meaning of constant
precise) for 0 <   1. The Uy () are cofinal in the system of neighborhoods of
20 Perverse sheaves and the topology of algebraic varieties

y. We have a canonical identification H∗ (C)y = H∗ (Uy (), C) for 0 <   1. This

is very important for many reasons. Let us give one. Let f : X → Y be a map and
let K ∈ D(X). Since Rf∗ K is constructible, we have (R∗ f∗ K)y = H∗ (f−1 (Uy (), K)
for 0 <   1. Caution: there is a natural map to H∗ (f−1 (y), K), but this map
is in general neither surjective, nor injective; if the map is proper, then it is an
isomorphism (proper base change).
∼ D(Y) whose
Verdier duality. This is an anti-self equivalence (−)∨ : D(Y)op =
∗ ∨
defining property is a natural perfect pairing H (Y, C ) × Hc (Y, C) −→ Q, or,
−∗

equivalently, of a canonical isomorphism

(2.2.2) ∼ H−∗ (Y, C)∨ .
H∗ (Y, C∨ ) = c

If Y is nonsingular irreducible, then Q∨ Y = QY [2 dim Y], and we get an identifica-

tion H ∗+2 dim Y −∗ ∨
(Y, Q) = Hc (Y, Q) , i.e. Poincaré duality.
Stability of constructibility. It is by no means obvious, nor easy, that if C is
constructible, then Rf∗ C is constructible. This can be deduced from the Thom
isotopy lemmas5 [33]. This is a manifestation of the important principle that con-
structibility is preserved under all the “usual” operations on the derived category
of sheaves on Y (see [4]). The list above is more complete once we include the de-
rived RHom, the tensor product of complexes (it is automatically derived when
using Q-coefficients), the nearby and vanishing cycle functors, etc.
Duality exchanges. Verdier duality exchanges Rf∗ with Rf! , and f∗ with f! , i.e.,
Rf∗ (C∨ ) = (Rf! C)∨ and f∗ (K∨ ) = (f! K)∨ . Here is a nice consequence: let f be
proper (so that Rf∗ = Rf! ), then if C is self-dual, then so is Rf∗ C.
The importance of being proper. Proper maps are important for many rea-
sons: Rf! = Rf∗ ; the duality exchanges simplify; the proper base change theorem
holds, a special case of which tells us that (Rq f∗ QX )y = Hq (f−1 (y), Q) (see Ex-
ercise 1.7.1); Rf∗ preserves pure complexes (Frobenius, mixed Hodge modules);
Grothendieck trace formula is about Rf! ; the decomposition theorem is about
proper maps.
Adjunctions. We have adjoint pairs (f∗ , Rf∗ ) and (Rf! , f! ), hence natural trans-
formations: Id → Rf∗ f∗ and Rf! f! → Id. By applying cohomology to the first
one, we get the pull-back map in cohomology, and by applying cohomology with
compact supports to the second, we obtain the push-forward in cohomology with
compact supports. The functors RHom and ⊗ also form an adjoint pair (you
should formalize this).
The attaching triangles and the long exact sequences we already know. By
combining adjunction maps, we get some familiar situations from algebraic topol-
ogy. Let j : U → Y ← Z : i be a complementary pair of open/closed embeddings.
5 The
` a map f : X → Y of complex algebraic varieties, there is a dis-
main two points are: 1) given
joint union decomposition Y = i Yi into locally closed subvarieties such that Xi := f−1 Yi → Yi
is a topological fiber bundle for the classical topology; 2) algebraic maps can be completed compat-
ibly with the previous assertion (you may want to sit down and come up with a reasonable precise
statement yourself).
 Mark Andrea A. de Cataldo 21

Given C ∈ D(Y), we have the distinguished triangle i∗ i! C → C → j∗ j∗ C → i! i! C[1]

and, by applying sheaf cohomology, we get the long exact sequence of relative
cohomology
. . . → Hq (Y, U, C) → Hq (Y, C) → Hq (U, C|U ) → Hq+1 (Y, U, C) → . . .
If g : Y → Z is a map, then we can push forward the attaching (distinguished)
triangle and obtain a distinguished triangle, which will also give rise to hosts of
long exact sequences when fed to cohomological functors. By dualizing, we get
j! j∗ C → C → i∗ i∗ C → j! j∗ C[1] and by taking cohomology with compact supports,
we get the long exact sequence of cohomology with compact supports
. . . → Hq q q q+1
c (U, C) → Hc (Y, C) → Hc (Z, C|Z ) → Hc (U, C) → . . .
This is nice, and used very often, because it gives the usual nice relation between
the compactly supported Betti numbers of the three varieties (U, Y, Z).
Exercise 2.7.11 asks you to use the first attaching triangle and its push-forward
when studying the resolution of singularities of a germ of an isolated surface sin-
gularity. This is an important example: it shows how the intersection complex of
the singular surface arises; it relates the non degeneracy of the intersection form
on the curves contracted by the resolution, to the decomposition theorem for the
resolution map. The careful study of this example allows to extract many general
ideas and patterns, specifically how to relate the non degeneracy of certain local
intersection forms to a proof of the decomposition theorem.
2.3. Definition of perverse sheaves
Before we define perverse sheaves. The category of perverse sheaves on an
algebraic variety is abelian, noetherian, anti-self-equivalent under Verdier duality,
artinian. The cohomology groups of perverse sheaves satisfy Poincaré duality,
Artin vanishing and the Lefschetz hyperplane theorem. They are stable under
the nearby and vanishing cycle functors. The simple perverse sheaves, i.e. the in-
tersection complexes with simple coefficients, satisfy the decomposition theorem
and the relative hard Lefschetz theorem, and their cohomology groups satisfy the
Hard Lefschetz theorem and, when the coefficients are “Hodge-theoretic”, the
Hodge-Riemann bilinear relations. Perverse sheaves play important roles in the
topology of algebraic varieties, arithmetic algebraic geometry, singularity theory,
combinatorics, representation theory, geometric Langlands program.
Even though not everyone agrees (e.g. I), one may say that perverse sheaves
are more natural than constructible sheaves; see Remark 2.4.3. Perverse sheaves
on singular varieties are close in spirit to locally constant sheaves on algebraic
manifolds.
Perverse sheaves have at least one drawback: they are not sheaves!6
6 The “sheaves” in perverse sheaves is because the objects and the arrows in P(Y) can be glued from
local data, exactly like sheaves; this is false for the derived category and for D(Y)!; do you know
why? As to the term “perverse”, see M. Goresky’s post on Math Overflow: What is the etymology of
the term perverse sheaf?
22 Perverse sheaves and the topology of algebraic varieties

The conditions of support and of co-support. We say that a constructible

complex C ∈ D(Y) satisfies the conditions of support if dim supp Hi (C) 6 −i, for
every i ∈ Z, and that it satisfies the conditions of co-support if its Verdier dual C∨
satisfies the conditions of support.

Definition 2.3.1. (Definition of the category P(Y) of perverse sheaves) We say

that P ∈ D(Y) is a perverse sheaf if P satisfies the conditions of support and of
co-support. The category P(Y) of perverse sheaves on Y is the full subcategory of
the constructible derived category D(Y) with objects the perverse sheaves.

Conditions of (co)support and vectors of dimensions. To fix ideas, suppose

that dim Y = 4. The following vector exemplifies the upper bounds for the dimen-
sions of the supports of the cohomology sheaves Hi for −4 6 i 6 0 (outside of
this interval, the cohomology sheaves of a perverse sheaf can be shown to be zero)
(4, 3, 2, 1, 0). For comparison, the analogous vector for an intersection complex of
the form ICY (L) is (4, 2, 1, 0, 0). For a table giving a good visual for the conditions
of support and co-support for perverse sheaves and for intersection complexes,
see [19, p.556].
It is important to keep in mind that the support conditions are conditions on
the stalks (direct limits of cohomology over neighborhoods) of the cohomology
sheaves, whereas the conditions of co-support are conditions on the co-stalk (in-
verse limits of cohomology with compact supports over neighborhoods) and as
such are maybe a bit less intuitive; see Exercise 1.7.16, where this issue is tackled
for the constant sheaf on the affine cone over an embedded projective manifold.
On the other hand, if for some reason we know that complex C ∈ D(Y) is Verdier
self-dual, then it is perverse if and only if it satisfies the conditions of support.
Note that the derived direct image via a proper map of a self-dual complex is
self-dual. This simple remark is very helpful in practice.
This is a good time to carry out Exercises 2.7.3 and 2.7.4.
2.4. Artin vanishing and Lefschetz hyperplane theorems
Conditions of support and Artin vanishing theorem. The conditions of sup-
port seem to have first appeared in the proof of the Artin vanishing theorem
for constructible sheaves in SGA 4.3.XIV, Théorème 3.1, p.159: let Y be affine and
F ∈ Shc (Y); then H∗ (Y, F) = 0 for ∗ > dim Y. Note that F[dim Y] satisfies the con-
ditions of support, and that Artin’s wonderful proof works for a constructible
complex satisfying the conditions of support. Note also that if Y is nonsingular
and F = QY , the result is affordable by means of Morse theory7 !
Since perverse sheaves satisfy the conditions of support and co-support, we
see that they automatically satisfy the “improved” version of the Artin vanishing
theorem.

7 The singular case is one of the reasons for the existence of the book [33].
 Mark Andrea A. de Cataldo 23

Theorem 2.4.1. (Artin vanishing theorem for perverse sheaves)8 Let P be a per-
verse sheaf on the affine variety Y. Then H∗ (Y, P) = 0, for ∗ not in [− dim Y, 0] and
H∗c (Y, P) = 0, for ∗ not in [0, dim Y].

“Proof.” See Exercise 2.7.5.

Exactly as in the Morse-theory proof of the weak Lefschetz theorem, the im-
proved Artin vanishing theorem implies

Theorem 2.4.2. (Lefschetz hyperplane theorem for perverse sheaves). Let Y be

a quasi projective variety, let P ∈ P(Y) and let Y1 ⊆ Y be a general hyperplane section.
Then H∗ (Y, P) → H∗ (Y1 , P|Y1 ) is an isomorphism for ∗ 6 −2, and is injective for ∗ = −1.
There is a similar statement for compactly supported cohomology (guess it!).

Proof. We give the proof when Y is projective. Let j : U := Y \ Y1 → Y ← Y1 : i

be the natural maps. We have the attaching triangle Rj! j∗ P → P → Ri∗ i∗ P → and
the long exact sequence of cohomology (= cohomology with compact supports
because Y is projective!)
∗
. . . → H−k
c (Y, Rj! j P) → H
−k
(Y, P) → H−k (Y1 , P|Y1 ) → H−k+1 (Y, Rj! j∗ P) → . . .
We need to show that H−k ∗ −k ∗
c (Y, Rj! j P) = Hc (U, j P) = 0 for −k < 0, but this is
Artin vanishing for perverse sheaves on the affine U. 

Remark 2.4.3. (Lefschetz hyperplane theorem: Perverse sheaves vs. sheaves)

As the proof given above shows, once we assume the projectivity of Y, any hy-
perplane section Y1 will do. This is similar to the classical proof of the Lefschetz
hyperplane theorem due to Andreotti-Frankel (following a suggestion by Thom)
and contained in Milnor’s Morse Theory (jewel) book [40], where, if Y is projec-
tive, then we only need to pick a hyperplane section that contains all the singu-
larities of Y, so that the desired vanishing stems from Lefschetz duality and from
Morse theory. This shows that even the constant sheaf is not well-behaved on
singular spaces! If we try and repeat the proof above for the constant sheaf on a
singular space, we stumble into the realization that we do not have the necessary
Artin vanishing for cohomology with compact supports for the constant sheaf on
the possibly singular U. This issue disappears if we use perverse sheaves!

The simple perverse sheaves are the intersection complexes. Since the cate-
gory P(Y) is artinian, every perverse sheaf P ∈ P(Y) admits an increasing finite
filtration with quotients simple perverse sheaves. The important fact is that the
simple perverse sheaves are the intersection complexes ICS (L) seen above with L
simple! (Exercise 2.7.10). There is a shift involved: ICS (L) is not perverse on the
nose, but if we set
(2.4.4) ICS (L) := ICS (L)[dim S],
8 Note that there is no sheaf analogue of Artin vanishing for compactly supported cohomology: the

Verdier dual of a constructible sheaf is not a sheaf!

24 Perverse sheaves and the topology of algebraic varieties

then the result is a perverse sheaf on Y. One views these complexes on the closed
subvarieties i : S → Y as complexes on Y supported on S via i∗ (we do not do this
to simplify the notation).
There is an important collection of results in intersection cohomology that gen-
eralize to the singular setting the beloved collection of classical results that hold
for complex algebraic manifolds. See Exercise 2.7.7.
2.5. The perverse t-structure
The constructible derived category D(Y) comes equipped with the standard
t-structure—i.e. the truncation functors are the standard ones—whose heart is
the abelian category Shc (Y) ⊆ D(Y) of constructible sheaves. A t-structure on a
triangulated category is an abstraction of the notion of standard truncation [2]. A
triangulated category may carry several inequivalent t-structures.
The middle perversity t-structure on D(Y). The category of perverse sheaves
P(Y) is also the heart of a t-structure on D(Y), the middle-perversity t-structure.
Instead of dwelling on the axioms, here is a short discussion.
The perverse sheaf cohomology functors. Every t-structure on a triangulated
category comes with its own cohomology functors; the standard one comes with
the cohomology sheaf functors. The perverse t-structure then comes with the
perverse cohomology sheaves pHi : D(Y) → P(Y) which are . . . cohomological,
i.e. turn distinguished triangles into long exact sequences
A → B → C → A[1] =⇒ . . . → pHi (A) → pHi (B) → pHi (C) → pHi+1 (A) → . . . ,
and, moreover, we have
(2.5.1) p
Hi (C[j]) = pHi+j (C).
Let us mention that C ∈ D(Y) satisfies the conditions of support if and only if
its perverse cohomology sheaves are zero in positive degrees; similarly, for the
conditions of co-support (swap positive with negative).
Kernels, cokernels. Once you have the cohomology functors, you can verify
that P(Y) is abelian: take an arrow a : P → Q in P(Y), form its cone C ∈ D(Y), and
then you need to verify that pH−1 (C) → P is the kernel and that Q → pH0 (C) is
the cokernel. What is the image?
Verdier duality exchange:
(2.5.2) p
Hi (C∨ ) = pH−i (C)∨
∼ C∨ , then ( pHi (C))∨ = pH−i (C) and if, in addition, f is proper, then
If C =
(2.5.3) ∼ Rf∗ C.
(Rf∗ C)∨ =
The perverse cohomology sheaves of a complex do not determine the complex.
However, Exercise 4.7.8 tells us that in the decomposition theorem (1.6.3) we may
write
∼
M
Rf∗ ICX (M) = H (Rf∗ ICX (M))[−c].
p c

c∈Z
 Mark Andrea A. de Cataldo 25

The perverse cohomology sheaf construction is a way to get perverse sheaves

out of any complex. So there are plenty of perverse sheaves. In fact, we have the
following two rather deep and very (!) surprising facts:
Theorem 2.5.4. (The constructible derived category as a derived category) Let Y
be a variety.
(1) ([46]9 ) D(Y, Z) with its standard t-structure is equivalent to Db (Shc (Y, Z))
with its standard t-structure.
(2) ([1]10 ) D(Y, Q)) with its perverse t-structure is equivalent to Db (P(Y, Q)) with
its standard t-structure.
Exercise 2.7.9 introduces another construction that leads to special perverse
sheaves, i.e. the intermediate extension functor j!∗ . This is crucial, in view of
the fact that the intersection cohomology complex ICS (L) can be defined as the
intermediate extension of L[dim S] from the open subvariety So ⊆ Sreg ⊆ S on
which L is defined to the whole of S, and thus to any variety that contains S as a
closed subvariety.
2.6. Intersection complexes
Recall the conditions of support and co-support for a complex P to be a com-
plex of perverse sheaves: dim supp Hi (C) 6 −i, and the same for C∨ .
The original definition of intersection complex ICS (L) of an enriched variety
(where we start with L a locally constant sheaf on some non-empty So ⊆ Sreg ⊆ S)
involves repeatedly pushing-forward and standard-truncating across the strata of
a suitable stratification of S, starting from So ; see [2, Proposition 2.1.11]. It is a
fact that shrinking So does not effect the end result (this is an excellent exercise).
The end result can be characterized as follows.
Conditions of (co)support for intersection complexes. The intersection com-
plex ICS (L) of an enriched variety (S, L) is the complex C, unique up to unique iso-
morphism, subject to the conditions of support and co-support: C|So = L[dim S],
dim supp Hi 6 −i − 1 for every i 6= − dim S, and the “same” for C∨ . Recall the
two vectors exemplifying the conditions of support in dimension four for per-
verse sheaves (4, 3, 2, 1, 0), and for intersection complexes (4, 2, 1, 0, 0). Using this
characterization, the reader should verify that direct images under finite maps
preserve intersection complexes and that the same is true for small maps (see §3).
Another characterization of intersection complexes. The intersection complex
ICS (L) is the unique perverse sheaf extending its own restriction to an open dense
subvariety U ⊆ S so that the extension is “minimal” in the following sense: it has
no non zero perverse subobject or quotient supported on the boundary S \ U11 .
9 Nori’s paper contains a lovely proof of Artin vanishing in characteristic zero.
10 Beilinsonalso proves his wonderful Lemma 3.3, a strengthening of the Artin vanishing in arbitrary
characteristic; Nori calls it the “Basic Lemma” in [46].
11 It may have, however, a non zero subquotient supported on the boundary; the formation of this

kind of minimal—a.k.a. intermediate extension—is not exact on the relevant abelian categories: it
preserves injective and surjective maps, but it does not preserve exact sequences; see [19, p.562].
26 Perverse sheaves and the topology of algebraic varieties

Intersection complexes as intermediate extensions. Let j : U → S be a lo-

cally closed embedding. Let P ∈ P(U). Take the natural map (forget the sup-
ports) Rj! P → Rj∗ P. Take the map induced at the level of 0-th perverse coho-
mology sheaves: a : pH0 (Rj! P) → pH0 (Rj∗ P). Define the intermediate exten-
sion of P on U to Y by setting j!∗ P := Im a ∈ P(S). Let (S, L) be an enriched
variety and let j : So → S; we apply the intermediate extension functor to
P := L[dim S] ∈ P(U := So ) and we end up with ICS (L)! The same conclusion
holds if we take any U and P := (ICS (L))|U ) (the intersection complex is the
intermediate extension of its restriction to any dense open subvariety).
Since P(Y) is Noetherian and closed under Verdier duality, it is artinian, so
that the Jordan-Holder theorem holds. Now it is a good time to carry out Exer-
cise 2.7.10, which shows how to produce an “explicit” Jordan-Holder decomposi-
tion for perverse sheaves. The method also makes it clear that the simple objects
in P(Y) are the ICS (L) with S ⊆ Y an irreducible closed subvariety and with L a
simple locally constant sheaf on some Zariski dense open subset of the regular
part of S.
2.7. Exercises for Lecture 2

Exercise 2.7.1. (Some very non constructible sheaves) Provide a definition of

analytic constructibility in the context of the analytic Zariski topology which is
parallel to the one given in §2.2 by dropping the condition on the finiteness of the
collection of locally closed subvarieties. Use the closed embedding i : C → A1
of the Cantor set into the complex affine line to show that the direct image sheaf
i∗ QC is not analytically constructible.

Exercise 2.7.2. (Skyscraper sheaves) Classify the sheaves of rational vector spaces
on a variety which are both constructible and injective.

Example 2.7.3. (First (non) examples) If Y is of pure dimension and F ∈ Shc (Y),
then F[dim Y] satisfies the conditions of support. In general, F[dim Y] is not per-
verse as its Verdier dual may fail to satisfy the condition of support. For example,
the Verdier dual of QY [dim Y] is the shifted dualizing complex ωY [− dim Y] and
the singularities of Y dictate whether or not it satisfies the conditions of support;
see [18, §4.3.5-7]. If Y is nonsingular of pure dimension, and L is locally constant,
then L[dim Y] is perverse, for its Verdier dual is L∨ [dim Y].

Exercise 2.7.4. (Some perverse sheaves) The derived direct image of a perverse
sheaf via a finite map is perverse. Give examples showing that the derived direct
image via a quasi-finite map of a perverse sheaf may fail to be perverse. Let
j : X := C∗ → C =: Y be the natural open embedding; show that the natural
map Rj! QX [1] → Rj∗ QX [1] in D(Y) is, in fact, in P(Y); determine kernel, cokernel
and image. Show that if we replace C∗ with Cn \ {0}, then the map above is not
one of perverse sheaves. Show that if instead of removing the origin, we remove
a finite configuration of hypersurfaces, then we get a map of perverse sheaves.
 Mark Andrea A. de Cataldo 27

Show more generally that an affine open immersion j is such that Rj! and Rj∗
preserve perverse sheaves (hint: push-forward and use freely the Stein (rather
than the affine) version of the Artin vanishing theorem to verify the conditions of
support on small balls centered at points on the hypersurfaces at the boundary).
Show that the direct image Rf∗ QX [2] with f : X → Y a resolution of singularities
of a surface is perverse. Let f : X2d → Y 2d be proper and birational, with X
nonsingular and irreducible, let y ∈ Y and let f be an isomorphism over Y \ y; give
an “if and only if" condition that ensures that Rf∗ Q[dim X] is perverse. Determine
the pairs m 6 n such that the blowing up of f : X → Y of Cm ⊆ Cn =: Y is such
that Rf∗ QX [n] is perverse.

Exercise 2.7.5. (Artin vanishing: from constructible to perverse sheaves; co-

homological dimension) Assume the Artin vanishing theorem for constructible
sheaves and deduce the one for the cohomology of perverse sheaves by use of
the Grothendieck spectral sequence Hp (Y, Hq (P)) =⇒ Hp+q (Y, P). (Hint: the
supports of the cohomology sheaves are closed affine subvarieties.) Dualize the
result to obtain the Artin vanishing theorem for the cohomology with compact
support of a perverse sheaf. Use a suitable affine covering of a quasi-projective
variety Y to show that the cohomology and cohomology with compact supports
of a perverse sheaf live in the interval [− dim Y, + dim Y]. What about a non quasi
projective Y?

Exercise 2.7.6. (Intersection complex via push-forward and truncation) The orig-
inal Goresky-MacPherson definition of intersection complex, suggested to them
by Deligne, involves repeated push-forward and truncation across the strata of a
Whitney-stratification. Let us take j : Cn \ Cm=0 =: U → Y := Cn . The formula
reads ICY := τ6−1 Rj∗ QU [n]. Verify that the result is QY [n]. Do this for other
values of m and verify that you get QY [n], again. Take a complete flag of linear
subspaces in Cn , apply the general formula given by iterated push-forward and
truncation, and verify that you get QY [n]. What is your conclusion?

Exercise 2.7.7. (Hodge-Lefschetz package for intersection cohomology) Guess

the precise statements of the following results concerning intersection cohomol-
ogy groups: Poincaré duality, Artin vanishing theorem, Lefschetz hyperplane
theorem, existence of pure and mixed Hodge structures, hard Lefschetz theorem,
primitive Lefschetz decomposition, Hodge-Riemann bilinear relations.

Exercise 2.7.8. (Injective or surjective?) Let j : A1 \ {0} → A1 be the natural open

embedding. Verify that: the natural map j! Q → j∗ Q in Shc (A1 ) is injective; the
natural map j! [1]Q → j∗ Q[1] is surjective in P(A1 ).

Exercise 2.7.9. (Perverse cohomology sheaves and the intermediate extension

functor) For j : Cn \ {0} → Cn the natural open embedding, compute pHi (Rj! Q)
and pHi (Rj∗ Q). Let j : U → X be an open embedding and let P ∈ P(U). Let
a : Rj! P → Rj∗ P be the natural map. Show that the assignment defined by
28 Perverse sheaves and the topology of algebraic varieties

P 7→ j!∗ P := Im{ pH0 (Rj! P) → pH0 (Rj∗ P)} is functorial. This is the intermediate
extension functor. Find an example showing that it is not exact (exact:= it sends
short exact sequences in P(U) into ones in P(X)). (Hint: punctured disk and rank
two unipotent and non-diagonal matrices). Compute j!∗ QU [dim U] when j is the
embedding of a Zariski-dense open subset of a nonsingular and irreducible vari-
ety. Same for the embedding of affine cones over projective manifolds minus their
vertex into the cone. Compute j!∗ L[1] where U = C∗ and L is a locally constant
sheaf on U.

Exercise 2.7.10. (Jordan-Holder for perverse sheaves) Recall our standing as-
sumptions: varieties are not assumed to be irreducible, nor pure-dimensional.
Let P ∈ P(Y). Find a non empty open nonsingular irreducible subvariety j : U ⊆ Y
such that Q := j∗ P = L[dim U] for a locally constant sheaf on U. Produce the
natural commutative diagram with a 0 epimorphic and a 00 monomorphic
6P

(
pH0 (Rj
! Q)
a / pH0 (Rj∗ Q)
6
a0 a 00

(
ICU (L) := Ima 0 .

Deduce formally, from the fact that the image of P contains Ima 0 , that P contains
a subobject P 0 together with a surjective map b : P 0 → ICU (L); you can even
choose P 0 to be maximal with this property, but it does not matter in what fol-
lows. Deduce that we have a filtration Ker b ⊆ P 0 ⊆ P with P 0 /Ker b = ICU (L).
Use noetherian induction to prove that we can refine this two-step filtration to
a filtration with successive quotients of the form ICS (L). Each local system L
appearing in this way admits a finite filtration with simple quotients. Refine fur-
ther to obtain a finite increasing filtration of P with successive quotients of the
form ICS (L) with L simple. (In the last step you need to use the fact that the
intermediate extension functor, while not exact, preserves injective maps.)

Exercise 2.7.11. (Attaching the vertex to a cone) Use attaching triangles and re-
sulting long exact sequences of cohomology to study C = Rf∗ QX , where f : X → Y
is the resolution of the cone (affine and projective) over a nonsingular embedded
projective curve obtained by blowing up the vertex. (See [17].)

3. Lecture 3: Semi-small maps

Summary of Lecture 3. Definition of semi-small map. Hard Lefschetz and Hodge-
Riemann bilinear relations for semi-small maps. Special form of the decomposition theo-
rem for semi-small maps. Hilbert schemes of points on smooth surfaces and the picture
 Mark Andrea A. de Cataldo 29

of Grojnowski-Nakajima. The endomorphism and correspondence algebras are isomorphic

and semi-simple. Hint of a relation to the Springer picture.
3.1. Semi-small maps Semi-small maps are a very special class of maps, e.g.
they are necessarily generically finite. On the other hand, the blowing up of
point in C3 is not semi-small. Resolutions of singularities are very rarely semi-
small (except in dimension 6 2). It is remarkable that semi-small maps appear
in important situations, e.g. holomorphic symplectic contractions, quiver vari-
eties, moduli of bundles on surfaces, Springer resolutions, convolution on affine
grassmannians, standard resolutions of theta divisors and Hilbert-Chow maps for
Hilbert schemes of points on surfaces. References include [14] and the beautiful
book [6].
We now discuss some of their features. To simplify the discussion, we work
with proper surjective maps f : X → Y with X nonsingular and irreducible.
We start with what is likely to be the quickest possible definition of semi-small
map.

Definition 3.1.1. (Definition of semi-small map) The map f is said to be semi-

small if dim X ×Y X = dim X.

Quick is good, but not always transparent. The standard definition involves
consideration of the dimension of the locally closed loci Sk ⊆ Y where the fibers
of the map have fixed dimension k. Then semi-smallness is the requirement that
dim Sk + 2k 6 dim X for every k > 0; see Exercise 3.6.1.
Small maps. We say that the map f is small if it is semi-small and X ×Y X
has a unique irreducible component of maximal dimension dim X (which one?).
For semi-small maps, this is equivalent to having dim Sk + 2k < dim X for every
k > 0.
The blowing ups of Cm ⊆ Cn , m 6 n − 2 have positive-dimensional fibers
isomorphic to Pn−m−1 and are semi-small if and only if m = n − 2. None of
these is small. The blowing up of the affine cone over the nonsingular quadric in
P3 along a plane thru the vertex is a small map, with fiber over the vertex of the
cone isomorphic to P1 . The blowing up of the vertex, which has fiber over the
vertex isomorphic to the nonsingular quadric, is not.
The Springer resolution of the nilpotent cone in a semi-simple Lie algebra is
semi-small and Grothendieck-Springer simultaneous resolution is small. We shall
meet both a bit later and show how they interact beautifully to give us a “decom-
position theorem argument” for the presence of an action of the Weyl group of
the Lie algebra on the cohomology of the fibers of the Springer resolution. The
Weyl group does not act on the fibers!
The following beautiful result of D. Kaledin is a source of many examples of
highly non-trivial semi-small maps.
30 Perverse sheaves and the topology of algebraic varieties

Theorem 3.1.2. (Holomorphic symplectic contractions are semi-small [36]) A

projective birational map from a holomorphic symplectic12 nonsingular variety is semi-
small.
It is amusing to realize that, for projective maps onto a projective target, semi-
smallness and the Hard Lefschetz phenomenon are essentially equivalent. In fact,
we have the following
Theorem 3.1.3. (Hard Lefschetz for semi-small maps [14]) Let f : X → Y be a
surjective projective map of projective varieties with X nonsingular and let η := f∗ L ∈
H2 (X, Q) be the first Chern class of the pull-back to X of an ample line bundle L on Y. The
iterated cup product maps ηr : Hdim X−r (X, Q) → Hdim X+r (X, Q) are isomorphisms for
every r > 0 if and only if the map f is semi-small. In the semi-small case, we have the
primitive Lefschetz decomposition and the Hodge-Riemann bilinear relations.
Hodge-index theorem for semi-small maps. There is an important phenom-
enon concerning projective maps that is worth mentioning, i.e. the signature of
certain local intersection forms [16]; for a discussion of these, see [17]. The sit-
uation is more transparent in the case of semi-small maps, where it is directly
related to the Hodge-Riemann bilinear relations associated with Theorem 3.1.3.
To have a clearer picture, let us limit ourselves to stating a simple, revealing
and important special case. Let f : X → Y be a surjective semi-small projective
map with X nonsingular of some even dimension 2d. Assume that f−1 (y) is d-
dimensional for some y ∈ Y. By intersecting in X, we obtain the refined symmetric
intersection pairing H2d (f−1 (y)) × H2d (f−1 (y)) → Q, where we are intersecting
the fundamental classes of the irreducible components of top-dimension d of this
special fiber inside of X. The following is a generalization of a result of a famous
result of Grauert’s for d = 1.
Theorem 3.1.4. (Refined intersection forms have a precise sign) The refined inter-
section pairing above is (−1)d -positive-definite.
By looking carefully at every proper map, similar refined intersection forms
reveal themselves. One can prove that the decomposition theorem is (essentially)
equivalent to the non-degeneracy of these refined intersection forms together
with Deligne’s semi-simplicity of monodromy theorem; see [16].
Exercise 3.6.2 relates perverse sheaves and semi-small maps: if X in nonsingu-
lar, then f : X → Y is semi-small if and only if Rf∗ QX [dim X] is perverse.
The decomposition theorem then tells us that
M
(3.1.5) f∗ QX [dim X] = pH0 (f∗ QX [dim X]) = ICS (L).
(S,L)∈EV0

Question 3.1.6. What “are” the summands appearing in the decomposition theo-
rem for semi-small maps?
dim X
12 Aneven-dimensional variety with a closed holomorphic 2-form ω that is non-degenerate: ω 2

is nowhere vanishing.
 Mark Andrea A. de Cataldo 31

3.2. The decomposition theorem for semi-small maps

A little bit about stratifications. Even if not logically necessary, it simplifies
matters to use the stratification theory of maps to clarify the picture a bit. There
`
is a finite disjoint union decomposition Y = a∈A Sa into locally closed non-
singular irreducible subvarieties Sa ⊆ Y such that f−1 (Sa ) → Sa is locally (for
the classical topology) topologically a product over Sa . It is clear that this de-
composition refines the one above given by the dimension of the fibers, so that
dim Sa + 2 dim f−1 (s) 6 dim X for every s ∈ Sa . Since the map is assumed to be
proper, it is also clear that all direct image sheaves Rq restrict to locally constant
sheaves on every Sa . We call the Sa the strata (of a stratification of the map f).

Definition 3.2.1. (Relevant stratum) We say that Sa is relevant if we have:

dim Sa + 2 dim f−1 (s) = dim X.
We denote by Arel ⊆ A the set of relevant strata.

Exercise 3.6.3 shows that, for each relevant stratum Sa , the direct image sheaf
Rdim X−dim Sa (cf. §1.2) restricted to Sa is locally constant, semi-simple, with finite
monodromy. We denote this restriction by La .

Theorem 3.2.2. (Decomposition theorem for semi-small maps) If f : X → Y is

proper, surjective and semi-small with X nonsingular, then there is a direct sum decompo-
sition M
f∗ QX [dim X] = ICSa (La ).
a∈Arel

Since the locally constant sheaf La is semi-simple, it admits the isotypical direct
sum decomposition (1.5.5), i.e. we have La = ⊕χ La,χ ⊗ Ma,χ where χ ranges over
a finite set of distinct isomorphism classes of simple locally constant sheaves on
Sa and Ma,χ is a vector space of rank the multiplicity ma,χ of the locally constant
sheaf Lχ in La . The decomposition theorem then reads
M
(3.2.3) f∗ QX [dim X] = ICSa (La,χ ⊗ Ma,χ ).
a,χ

3.3. Hilbert schemes of points on surfaces and Heisenberg algebras

An excellent reference is [43]. Let X be a nonsingular complex surface. For
every n > 0 we have the Hilbert scheme X[n] of n points on X. It is irreducible,
nonsingular and of dimension 2n. There is a proper birational surjective map
π : X[n] → X(n) onto the n-th symmetric product sending a length n zero di-
mensional subscheme of X to its support counting multiplicities. There is a nat-
ural stratification for the symmetric product of the map, which we now describe.
` (n)
First, we stratify the symmetric product variety X(n) = ν∈P(n) Xν , where
P(n) is the set of partitions ν = {νj } of the integer n13 obtained by taking the lo-
cally closed, irreducible, nonsingular subvarieties of dimension 2l(ν) consisting
13 The νj are a set of l(ν) positive integers adding up to n: l(ν) is called the length of ν.
32 Perverse sheaves and the topology of algebraic varieties
P
of points of the form i νi xi ∈ X(n) , with type {νi } given by ν. The remarkable
(n)
fact is that the fibers π−1 (xν ) of the points xν ∈ Xν are irreducible of dimen-
P
sion j (νj − 1) = n − l(ν). If we ignore the scheme structures, these fibers are
isomorphic to products of punctual Hilbert schemes. It follows that the map π
` (n) ` (n)
is semi-small. In fact, X[n] = P(n) π−1 (Xν ) → P(n) Xν is the aforemen-
tioned stratification of the semi-small map π. All of its strata on X(n) are relevant.
Since the fibers are all irreducible, the relevant locally constant sheaves are all
constant of rank one. In particular, the decomposition theorem for π takes the
form
Rπ∗ QX[n] [2n] = ⊕ν∈P(n) IC (n) .
Xν
(n)
A second remarkable fact is that the normalization of Xν can be identified with
Q
a product of symmetric products X(ν) := n i=1 X
(ai ) , where a is the number of
i
times that i appears in ν. This is a variety obtained by dividing a nonsingular
variety by the action of a finite group; in particular, its intersection complex is
the constant sheaf. By the IC normalization principle (Fact 4.5.3), we see that
IC (n) is the push-forward of the shifted constant sheaf from the normalization.
Xν
Taking care of shifts, and waiting for the the dust to settle, we obtain the Göttsche
formula: (recall that X(n) := {point})
M M M
(3.3.1) H(X) := H∗ (X[n] ) = H∗−cl(ν) (X(0) , Q),
n>0 n>0 ν∈P(n)

where the colength is given by cl(ν) := n − l(ν).

If we take X = C2 , then something remarkable emerges: look at (3.3.2) and
(3.3.3). The formula above, taken for every n > 0 gives
∞
X ∞
Y
[n] 1
(3.3.2) dim H∗ (C2 , Q) = .
1 − qj
n=0 j=1

Let R := Q[x1 , x2 , . . .] be the algebra of polynomials in the infinitely many inde-

terminates xi , declared to be of degree i. The infinite dimensional Heisenberg
algebra H is the Lie algebra whose underlying rational vector space has basis
{{di }i<0 , c0 , {mi }i>0 } and subject to the following relations: c0 is central, the di ’s
commute with each other, the mi ’s commute with each other, and the commuta-
tor [di , mj ] = δ−i,j c0 . Then R is an irreducible H-module generated by 1 where
di acts as formal derivation by xi and mj by multiplication by xj . The dimension
an of the space of homogeneous polynomials of degree n is given by:
∞
X ∞
Y 1
(3.3.3) an qn = .
1 − qj
n=0 j=1

Well, isn’t this a coincidence! The operators di , mi change the homogeneous de-
gree of x-monomials by ±i. This, together with the formalism of correspondences
in products, suggests that there should be geometrically meaningful cohomology
 Mark Andrea A. de Cataldo 33

[n] [n±i]
classes in H∗ (C2 × C2 , Q) that reflect, on the Hilbert scheme side, the
Heisenberg algebra action on the polynomial side.
That this is indeed the case is due to Grojnowski and to Nakajima: they
guessed what above, constructed algebraic cycles on the products of Hilbert
schemes above that would be good candidates and then verified the Heisenberg
Lie algebra relations. In fact, for every nonsingular surface X, there is an associ-
ated (Heisenberg-Clifford) algebra H(X) that acts geometrically and irreducibly
on H(X) (3.3.1).
3.4. The endomorphism algebra End(f∗ QX )
A reference here is [15].
Semi-simple algebras. A semi-simple algebra is an associative artinian (dcc)
algebra over a field with trivial Jacobson ideal (the ideal killing all simple left
modules). The Artin-Wedderburn theorem classifies the semi-simple algebras
over a field as the ones which are finite Cartesian products of matrix algebras
over finite dimensional division algebras over the field.
Warm-up. Show that Md×d (Q) is semi-simple. Show that the upper triangular
matrices do not form a semi-simple algebra. Hence if f := pr2 : P1 × P1 → P1 ,
then EndD(P1 ) (Rf∗ Q) is not a semi-simple algebra.

Theorem 3.4.1. (Semi-small maps and semi-simplicity of the endomorphism

algebra) In the setting of Theorem 3.2.2, the endomorphism Q-algebra
EndD(Y) f∗ QX [dim X]
is semi-simple.

Proof. We see more important properties of intersection complexes at play, i.e. the
Schur lemma phenomena for simple perverse sheaves.
By simplicity, Hom(ICSa (Lχ ), ICSb (Lψ )) = δχ,ψ δa,b End(ICSa (Lχ )) (i.e., there
are no non zero maps if they differ): in fact, look at kernel and cokernel and
use simplicity. This leaves us with considering terms of the form End(ICSa (Lχ ))
whose elements, for the same reason as above, are either zero, or are isomor-
phisms. These terms are thus division algebras Da,χ . It follows that
Y
EndD(Y) (Rf∗ QX [dim X]) = Mda,χ ×da,χ (Da,χ ),
a,χ

which is a semi-simple algebra by the Artin-Wedderburn theorem. 

The endomorphism algebra as a geometric convolution algebra. A refer-
ences here is also [6]. We can realize the algebra EndD(Y) (Rf∗ QX ) of endomor-
phisms in the derived category in geometric terms as the convolution algebra
HBM
2 dim X (X ×Y X), which is thus semi-simple. Let us discuss this a bit.
34 Perverse sheaves and the topology of algebraic varieties

Let X be a nonsingular projective variety. Then, we have an isomorphism of

algebras between the first and last term
End(H∗ (X)) =1 H∗ (X)∨ ⊗ H∗ (X)
(3.4.2)
∼ 2 H∗ (X) ⊗ H∗ (X) =
= ∼ 3 H∗ (X × X) =
∼ 4 H∗ (X × X)

where: =1 : linear algebra; =∼ 2 : Poincaré duality; =∼ 3 : Künneth; =

∼ 4 : Poincaré
duality; and where: the algebra structure on the last term is the one given by the
formalism of composition of correspondences in products (see Exercise 3.6.4).
The classes Γ ∈ Hγ (X × X) appearing in Exercise 3.6.4 are called correspon-
dences. This picture generalizes well, but not trivially, to proper maps f : X → Y
from nonsingular varieties as follows.
Theorem 3.4.3. (Correspondences and endomaps in the derived category) Let
f : X → Y be a proper map from a nonsingular variety. There is a natural isomorphism
∼ HBM (X ×Y X) of Q-algebras.
EndD(Y) (f∗ QX ) = 2 dim X

For our semi-small maps, Exercise 3.6.7 provides an evident geometric basis of
the vector space HBM2 dim X (X ×Y X).
2 dim X (X ×Y X) = EndD(Y) (Rf∗ QX ) given by algebraic
Since there is a basis of HBM
cycles, a formal linear algebra manipulation shows that if X is projective, then
decomposition H∗ (X, Q) = ⊕a∈Arel IH∗−codim(Sa ) (Sa , La ) is compatible with the
Hodge (p, q)-decomposition, i.e. it is given by pure Hodge substructures; see
Exercise 3.6.8. In fact, one even has a canonical decomposition of Chow motives
reflecting the decomposition theorem for semi-small maps; see [15]. Look at the
related (deeper) Question 5.5.2.
3.5. Geometric realization of the representations of the Weyl group
Excellent references here are [5, 6].
There is a well-developed theory of representations of finite groups G (charac-
ter theory) into finite dimensional complex vector spaces. In a nearly tautological
sense, this theory is equivalent to the representation theory of the group alge-
bra C[G].
If we take the Weyl group W of any of the usual suspects, e.g. SLn (C) with
Weyl group the symmetric group Sn , then we can ask whether we can realize the
irreducible representations of W by using the fact that W is a Weyl group.
Springer realized that this was indeed possible, and in geometric terms! In
what follows, we do not reproduce this amazing story, but we limit ourselves to
showing how the decomposition theorem14 allows us to introduce the action of
the Weyl group on the cohomology of the Springer fibers. The Weyl group does
not act algebraically on these fibers!
Take the Lie algebra sln (C) of traceless n × n matrices. Inside of it there is the
cone N with vertex the origin given by the nilpotent matrices. Take the flag variety
F, i.e. the space of complete flags f in Cn . Set N
e := {(n, f)|, n stabilizes f} ⊆ N × F.

14 There are other ways.

 Mark Andrea A. de Cataldo 35

Then N e → F can be shown to be the projection T ∗ F → F for the cotangent bundle

(and N e is thus a holomorphic symplectic manifold) and the projection π : N e →N
is a resolution of the singularities on the nilpotent cone N.
The map π is semi-small! We know this, for example, from Kaledin’s Theo-
rem 3.1.2. In fact, it was known much earlier, by the work of many. We can
partition N according to the Jordan canonical form. This gives rise to a stratifica-
tion of N and of π. Every stratum of this stratification turns out to be relevant for
the semi-small map π.
It is amusing to realize that the intersection form associated with the deepest
stratum (vertex) gives rise to ± the Euler number of F and that the one associated
with the codimension two stratum yields (−1)times the Cartan matrix for sln (C).
The fibers of the map π are called Springer fibers. Springer proved that all the
irreducible representations of the Weyl group occur as direct summands of the
action of the Weyl group on the homology of the Springer fibers. This beautiful
result tells us that indeed one can realize geometrically such representations.
Note that the Weyl group does not act algebraically on the Springer fibers.
In what follows we aim at explaining how the Weyl group acts on the perverse
sheaf Rπ∗ QN e . In turn, by taking stalks, this explains why the Weyl group acts on
the homology of the Springer fibers.
Instead of sticking with π : N e → N, we consider p : sl^ n (C) → sln (C) defined
by the same kind of incidence relation. The difference is that if we take the
Zariski open set U given by diagonalizable matrices with n distinct eigenvalues,
then p is a topological Galois cover with group the Weyl group (permutation of
eigenvalues). We have a Weyl group action! Unfortunately N ∩ U = ∅! On the
other hand, p is . . . small! So Rp∗ Q = IC(L), where L is the local system on U
associated with the Galois cover with group the Weyl group. Then W acts on L.
Hence it acts on IC(L) by functoriality of the intermediate extension construction.
Since the map p is proper, and it restricts to π over N, we see that the restriction
of Rp∗ Q = IC(L) to N is Rπ∗ Q which thus finds itself endowed, almost by the
trick of a magician, with the desired W action!
3.6. Exercises for Lecture 3

Exercise 3.6.1. (Semi-smallness and fibers) Show that, for any maps f : X → Y,
we always have dim X ×Y X > dim X. Use Chevalley’s result on the upper semi-
continuity of the dimensions of the fibers of maps of algebraic varieties to produce
`
a finite disjoint union decomposition Y = k>0 Sk into locally closed subvari-
−1
eties with dim f (y) = k for every y ∈ Sk . Show that f is semi-small iff we have
dim Sk + 2k 6 dim X15 for every k > 0. Observe that f semi-small implies that f is
generically finite, i.e. that S0 is open and dense. Observe that f−1 (S0 ) ×S0 f−1 (S0 )
15 Think
of it as a vary special upper bound on the dimension of the “stratum” where the fibers are
k-dimensional.
36 Perverse sheaves and the topology of algebraic varieties

has dimension dim X. Give examples of semi-small maps where X ×Y X has at

least two irreducible components of dimension dim X.
Exercise 3.6.2. (Semi-small maps and perverse sheaves) A proper map f : X → Y
with X nonsingular is semi-small if and only if f∗ QX [dim X] is perverse on Y.
Exercise 3.6.3. (Relevant locally constant sheaves) Let Sa ∈ Arel be relevant.
−1
Show that Rdim X−dim Sa is locally constant on Sa with stalks H2 dim f (s) (f−1 (s)).
The monodromy of this locally constant sheaf, denoted by La , factors through
the finite group of symmetries of the set of irreducible components of maximal
dimension 12 (dim X − dim Sa ) of a typical fiber f−1 (s). (Note that, a priori, the
monodromy could send the fundamental class of such a component to minus
itself, thus contradicting the claim just made; that this is not the case follows, for
example, from a theorem of Grothendieck’s in EGA IV, 15.6.4. See the nice general
discussion in B.C. Ngo’s preprint [44, §7.1.1,], a discussion which we could not
locate in the published version [45]). In particular, La is semi-simple. Note that
if we switch from Q-coefficients to Q-coefficients simple objects may split further.
Do they stay semi-simple?
Exercise 3.6.4. (Formalism of correspondences in products) Unwind the isomor-
phisms (3.4.2) to deduce that under them, a class Γ ∈ Hγ (X × X) defines a linear
map Γ∗ : H∗ (X) → H∗+γ−2 dim X (X) given by a 7→ PD(pr2 ∗ (pr∗1 a ∩ Γ )). Conversely,
show that any graded linear map H∗ (X) → H∗+γ−2 dim X (X) is given by a unique
such Γ ∈ Hγ (X × X).
Exercise 3.6.5. (Sheaf theoretic definition of Borel-Moore homology) Let X be
a variety. Recall (one) definition of Borel-Moore homology: embed X as a closed
subvariety of smooth variety Y, then set
HBM
i (X) := H2dim Y−i (Y, Y − X),
where the right-hand-side is relative cohomology. By interpreting relative coho-
mology sheaf theoretically, give a sheaf theoretic definition of Bore-Moore homol-
ogy.
(Hint: consider the distinguished triangle i∗ i! → id → Rj∗ j∗ →, if you apply this
to the constant sheaf and take cohomology (or equivalently push to a point), what
long exact sequence do you get?)
Exercise 3.6.6. (Borel-Moore homology and Ext/convolution algebras) Fix a pro-
per morphism of varieties π : X → Y, with X smooth. Form a cartesian square
p1
Z /X
p2 π
 
X
π /Y

For sheaves A, B on a variety Z, let Exti (A, B) = HomDb (Z) (A, B[i]). I.e., Ext∗
denotes (shifted) Hom in the derived category. Show the following:
 Mark Andrea A. de Cataldo 37

(1) Ext∗ (Rπ∗ Q, Rπ∗ Q) = Ext∗ (Q, π! Rπ∗ Q). (Hint: adjunction property).
(2) Ext∗ (Q, π! Rπ∗ Q) = Ext∗ (Q, Rp2∗ p1! Q). (Hint: proper base change).
(3) Ext∗ (Q, Rp2∗ p1! Q) = Ext∗ (Q, p1! Q) (Hint: push-forward and hom).
(4) Ext∗ (Q, p1! Q) = HBM
2dim X−∗ (Z). (Hint: use the sheaf-theoretic definition of
Borel-Moore homology; the dimensional shift suggests the use of some
kind of duality).

Exercise 3.6.7. (Geometric basis for HBM2 dim X (X ×Y X) when f is semi-small) Show
that if f is semi-small, then the rational vector space HBM
2 dim X (X ×Y X) has a basis
formed by the fundamental classes of the irreducible components of X ×Y X of
maximal dimension dim X. Describe these irreducible components in terms of
monodromy over the relevant strata.

Exercise 3.6.8. (Hodge-theoretic decomposition theorem for f semi-small) Show

that if f : X → Y is semi-small with X projective nonsingular, then the decompo-
sition H∗ (X, Q) = ⊕a∈Arel IH∗−codim(Sa ) (Sa , La ) is one by pure Hodge structures
(i.e. compatible with the Hodge (p, q)-decomposition of H∗ (X, C). (Hint: the pro-
jectors onto the direct sums are given by algebraic cycles.)

4. Lecture 4: Symmetries: VD, RHL, IC splits off

Summary of Lecture 4. Discussion of the two main symmetries in the decomposition
theorem for projective maps: Verdier duality and the relative hard Lefschetz theorem.
Hard Lefschetz in intersection cohomology and Stanley’s theorem for rational simplicial
polytopes. A proof that the intersection complex of the image is always a direct summand.
Pure Hodge structures on the intersection cohomology of a projective surface.

Remark 4.0.1. We are going to discuss two symmetries for projective maps: Ver-
dier duality and the relative Hard Lefschetz theorem. Both these statements have
to do with direct image perverse sheaves. In fact, if the target is projective, then
we can take the shadow of these two symmetries in cohomology and notice that
there are two additional symmetries: Verdier duality and Hard Lefschetz theorem
on the individual summands IH∗ (S, L). Exercise 4.7.7 asks you to make an explicit
list in a low-dimensional case.

4.1. Verdier duality and the decomposition theorem

Verdier duality and the decomposition theorem. Recall the statement (1.6.3)
of the decomposition theorem
∼
M
Rf∗ ICX (M) = ICS (L)[−q].
q>0,EVq

Switching to the perverse intersection complex. If (S, L) is an enriched variety,

then ICS (L) is not a perverse sheaf. Recall (1.6.1): the perverse object is ICS (L) :=
ICS (L)[dim S]. In order to emphasize better certain symmetric aspects of the
decomposition theorem, we switch to the perverse intersection complex. This
38 Perverse sheaves and the topology of algebraic varieties

entails a a minor headache when re-writing (1.6.3), which becomes (verify it as

an exercise)
∼
MM
(4.1.1) Rf∗ ICX (M) = ICS (L)[−b],
b∈Z EVb

where (S, L) ∈ EVb if and only if (S, L) ∈ EVb+dim X−dim S .

The perverse cohomology sheaves of the derived direct image. Note that,
now, every b-th direct summand above is a perverse sheaf, so that, in view of
Exercise 4.7.8, we have that:
(4.1.2) Hb (Rf∗ ICX (M)) = ⊕EVb ICS (L).
p

The case when M is self-dual (and semi-simple). Let M be self-dual, e.g.

a constant sheaf, a polarizable variation of pure Hodge structures, or even the
direct sum of any M with its dual; self-dual local systems appear frequently
in complex algebraic geometry. Then so is ICX (M) and, by the duality ex-
change property for proper maps, so is Rf∗ ICX (M). In view of the duality relation
(2.5.2) between perverse cohomology sheaves, we see that pHb (Rf∗ ICX (M))) = ∼
pH−b (Rf IC (M)))∨ . By combining with (4.1.2), we get
∗ X
M  M  M 
∼
(4.1.3) Rf∗ ICX (M) = ICS (L)[−b] ⊕ ICS (L) ⊕ ICS (L∨ )[b] .
b<0 EV0 b<0
EVb EVb

In other words, the direct image is palindromic, i.e. it reads the same, up to shifts
and dualities, from right to left and from left to right. Just like the cohomology
of a compact oriented manifold!
The defect of semi-smallness. When trying to determine the precise shape
of the decomposition theorem, one important invariant is the minimal interval
[−r, r] out of which the perverse cohomology sheaves are zero. In this direction,
note that if M is constant and X is nonsingular, then r = dim X ×Y X − dim X > 0.
This difference is called the defect of semi-smallness in [16]. In this situation, r = 0
if and only if the map is semi-small.
4.2. Verdier duality and the decomposition theorem with large fibers
Here is a nice consequence of Verdier duality, more precisely of (4.1.3). It is an
observation due to Goresky and MacPherson and it is used by B.C. Ngô in his
proof of the support theorem, a key technical and geometric result in his proof of
the fundamental lemma in the Langlands’ program. See [45, §7.3].
Theorem 4.2.1. Let f : X → Y be proper with X nonsingular and equidimensional fibers
of dimension d. Assume a subvariety S appears in the decomposition theorem (1.6.3) for
Rf∗ QX . Then codim(S) 6 d.
Proof. There is a maximum index b+ S ∈ Z for which a term ICS (L)[−bS ] appears.
+
+
By the palindromicity (4.1.3), we may assume that bS > 0. Recall that L is defined
on some open dense So ⊆ S. Let U ⊆ Y be open such that its trace on S is So .
Replace Y with U. Denote by i : So → Y the closed embedding. Then Rf∗ Q[dim X]
 Mark Andrea A. de Cataldo 39

admits i∗ L[dim S][−b+

S ] as a+ direct summand. Then i∗ L is a non-trivial direct
summand of Rdim X−dim S+bS f∗ Q. Since the fibers have dimension d and b+ S > 0,
+
we have dim X − dim S 6 dim X − dim S + bS 6 2d. Since dim X = dim Y + d, the
conclusion follows. 
4.3. The relative hard Lefschetz theorem
Poincaré duality vs. hard Lefschetz. Let X be a projective manifold and let
η ∈ H2 (X, Q) be the class of a hyperplane section. Then we have two separate
phenomena:
Hdim X−r (X, Q) = Hdim X+r (X, Q)∨ (Poincaré duality),
ηr : Hdim X−r (X, Q) =∼ Hdim X+r (X, Q) (hard Lefschetz).
R
The first statement is that the pairing X − ∧ − between cohomology in comple-
R
mentary degrees is non-degenerate. The second one is that the pairing X ηr − ∧−
on a cohomology group Hdim X−r is non-degenerate. They both imply the usual
symmetry of Betti numbers. The latter implies also their unimodality, i.e. that
bd > bd−2 > . . ., where d is either dim X, or dim X − 1 (compare with the Hopf
surface, which has the former, but not the latter).
Exercise 4.7.4 discusses two proofs of the classical hard Lefschetz theorem.
Both proofs generalize and, with some work, afford proofs of the relative hard
Lefschetz theorem. In particular, they yield proofs of the hard Lefschetz theorem
in intersection cohomology.
We have mentioned how, in the context of singular varieties, Poincaré duality
in cohomology is lost but found again in intersection cohomology. The same is
true for the hard Lefschetz theorem for the intersection cohomology of projective
varieties! This brings us back to the theme harping the importance of the derived
category and of perverse sheaves: the statement of the hard Lefschetz for inter-
section cohomology is cohomological, but there is no known proof that avoids
perverse sheaves.
Let f : X → Y be a map of varieties and let η ∈ H2 (X, Q) be a cohomology
class. It is a general fact that, for every constructible complex C ∈ D(X), the class
η induces, η : C → C[2], which induces η : Rf∗ C → Rf∗ C[2], which, by taking
perverse cohomology sheaves, induces η : pHb (C) → pHb+2 (C). It follows that,
for every b > 0, we obtain maps ηb : pH−b (C) → pHb (C) in P(Y).
Let f : X → Y be a projective map with and let ICX (M) ∈ P(X). Let η ∈ H2 (X, Q)
be the first Chern class of an f-ample line bundle, i.e. a line bundle on X which
is ample on every fiber of f. It is a general fact that η ∈ H2 (X, Q) induces,
η : ICX (M) → ICX (M)[2], which induces η : Rf∗ ICX (M) → Rf∗ ICX (M)[2], which,
by taking perverse cohomology sheaves, induces η : pHb → pHb+2 . It follows
that, for every b > 0, we obtain maps ηb : pH−b → pHb in P(Y).

Theorem 4.3.1. (Relative hard Lefschetz [2, 16, 41, 42, 47, 48]) Let f : X → Y be a
projective map and let η ∈ H2 (X, Q) be the first Chern class of an f-ample line bundle
40 Perverse sheaves and the topology of algebraic varieties

on X16 . Let ICX (M) be semi-simple, i.e. X irreducible and M semi-simple. For every
b > 0 the iterated cup product map ηb : pH−b (Rf∗ ICX (M)) → pHb (Rf∗ ICX (M)) is
an isomorphism.

Hard Lefschetz for the intersection cohomology of projective varieties. The

special case when Y is a point and M is constant yields the hard Lefschetz theorem
for the intersection cohomology groups of a projective variety.
The same Deligne-Lefschetz criterion employed in the proof of the derived
Deligne theorem, allows one to deduce formally a first approximation to the
decomposition theorem (this argument does not afford the semi-simplicity part
of the decomposition theorem)
∼
M
(4.3.2) Rf∗ ICX (M) = H (Rf∗ ICX (M))[−b].
p b

b
Exercise 4.7.3 gets you a bit more acquainted with the primitive Lefschetz de-
compositions. Exercise 4.7.4 draws a parallel between the classical inductive ap-
proach to the Hard Lefschetz theorem via the Lefschetz hyperplane section theo-
rem and the semi-simplicity of monodromy for the family of hyperplane sections;
see Deligne’s second paper on the Weil Conjectures [29, §4.1].
4.4. Application of RHL: Stanley’s theorem
An excellent reference is [49]. For more details, see [19].
A convex polytope is the convex hull of a finite set in real Euclidean space. It
is said to be simplicial if all its faces are simplices. Example: a triangle. Non-
example: a square. Example: two square-based pyramids joined at the bases.
Let P be a d-dimensional simplicial convex polytope with fi i-dimensional faces,
0 6 i 6 d − 1. The f-vector (f for faces) of P is the vector f(P) = (f0 , . . . , fd−1 ). The
h-vector of P is defined by setting h(P) = (h0 , . . . , hd ) with
Xi 
d−j

hi = (−1)i−j fj−1 (f−1 := 1)
d−i
j=0

The f and h-vectors determine each other.

Question 4.4.1. When is a vector (f0 , . . . , fd−1 ) an f(P)-vector for some P?

A reformulation of) P. McMullen’s 1971 conjecture. A vector f is an f(P)-

vector for some P if and only if 1) hi = hd−i and 2) there is a graded commutative
Q-algebra R = ⊕i>0 Ri , that is generated by R1 , has R0 = Q and, for 1 6 i 6 bd/2c,
has dim Ri = hi − hi−1 . In particular, h0 6 h1 6 . . . 6 hd .
Associated simplicial toric variety. Stanley himself writes: “we are led to
suspect the existence of a smooth d-dimensional projective variety X(P) for which
(the Betti numbers) b2i = hi ”, for which Heven (X, Q) is generated by H2 (X, Q)

is one such line bundle: since f is projective, we can factor f as X → Y × P → Y (closed

16 Here

embedding, followed by the projection); pull-back the hyperplane bundle from P to Y × P and restrict
to X.
 Mark Andrea A. de Cataldo 41

and we can take for R := Heven (X, Q)/(η) (quotient by the ideal generated by
hyperplane class).
Then 1) above would be Poincaré duality and 2) would be a direct consequence
of hard Lefschetz on the smooth projective X (unimodality).
Let me describe briefly what is going on.
The combinatorial data of the simplicial P gives rise to, perhaps after some
wiggling to achieve rational vertices, a simplicial toric variety X(P).
Saying that P is simplicial means that X(P), while possibly singular, has singu-
larities of the type “vector space modulo a finite group”.
Necessity of the conditions. It is a fact that H∗ (X(P), Q) = Heven (X(P), Q) and
that b2i = hi . The basic idea is that: faces give rise to torus orbits; torus orbits
assemble into cells with the shape of affine spaces modulo finite groups; then
X(P) is a disjoint union of such cells; since the are cells automatically of even
real dimension, the cohomology has graded bases labelled by these cells; the only
issue is to count these cells properly; this is indeed the explanation of the relation
f(P) ↔ h(P) (in the simplicial case).
Exercise 4.7.11 tells us that ICX(P) = QX(P) . It follows that H∗ (X(P), Q) satisfies
Poincaré duality. We thus get the necessity of 1) in McMullen’s conjecture.
The necessity of 2) would follow if we knew the hard Lefschetz theorem for
H∗ (X(P), Q). But we do know this: we know it for the intersection cohomology
groups by the relative Hard Lefschetz theorem applied to the case when we map
a variety to a point! and, since ICX(P) = QX(P) , the rational intersection cohomol-
ogy groups are the ordinary rational cohomology groups.
We thus have the following

Theorem 4.4.2. (Simplicial polytopes: if and only if for f being an f(P) vector)
The McMullen conditions are necessary (Stanley: discussion above) and sufficient (Billera
and Lee: construction).

4.5. Intersection cohomology of the target as a direct summand

A funny situation. Singular cohomology is functorial, but, in general, f∗ is not
injective, not even if f is proper17 . On the other hand, intersection cohomology is
not functorial, but for proper surjective maps, the decomposition theorem exhibits
the intersection cohomology of the target as a direct summand of the intersection
cohomology of the source.
The following theorem is one of the most striking and useful applications of
the decomposition theorem. It is usually used, stated and proved in the context
of proper birational maps. The proof in the presence of generic large fibers is not
more difficult. In fact, we give a proof as it is also a chance to meet and use some
very useful general principles of the theory we have been talking about in these
lectures.
17
One can prove that it is injective when f is a proper map of algebraic manifolds by using the trace
map.
42 Perverse sheaves and the topology of algebraic varieties

Theorem 4.5.1. (Intersection complex as a direct summand) Let f : X → Y be a

proper map of irreducible varieties with image Y 0 . Then IH∗ (Y 0 ) is a direct summand of
IH∗ (X). More precisely, ICY 0 is a direct summand of f∗ ICX .

Let us state three general and useful principles. Recall that for a given (S, L),
the locally constant sheaf L is only defined on a suitable open dense subvariety
So ⊆ Sreg of S and that one can shrink So , if necessary.

Fact 4.5.2. (IC Localization Principle) Let ICS (L) ∈ P(Y), so that S ⊆ Y is closed,
and let U ⊆ Y be open. Then ICS (L)|U = ICS∩U (L|So ∩U ).

Fact 4.5.3. (IC Normalization Principle) Let ν : Ŷ → Y be the normalization of a va-

riety. Then ν∗ ICŶ (L) = ICY (L) (here, ν is finite, so that Rν∗ = ν∗ , derived=underived).

These first two principles hold because intersection complexes with coefficients
are characterized by the strengthened conditions of support and by restricting to
the locally constant sheaf on some Zariski dense open subset of the regular part,
and both conditions are preserved under restriction to any open set and under a
finite birational map. See also Exercise 4.7.12.

Fact 4.5.4. (Decomposition Theorem Localization Principle) A summand ICS (L)

appears in the decomposition theorem on Y if and only if there is an open U ⊆ Y meeting
S such that the restriction ICS (L)|U appears in the decomposition theorem on U.

This last principle is very important. It fails, for example, for the map from
the Hopf surface to P1 in the following sense: there is no decomposition theorem
over P1 (else we would have E2 -degeneration of the Leray spectral sequence),
but the Hopf map is locally trivial over any open proper subset U ⊆ P1 , so that
the decomposition theorem holds there (Künneth). It is important because when
looking for summands in the decomposition theorem, it may be easier to detect
them over some Zariski open subset. For example, if X is nonsingular, given
f : X → Y, there is the open subset Yreg(f) ⊆ Y of regular values of f. Let Rq be
the locally constant sheaves given by the cohomology of the fibers of f over Yreg(f) .
Deligne’s theorem applies to the map over Yreg(f) . The reader can now observe
that the decomposition theorem localization principle allows us to deduce that
all the ICY (Rq )[−q] are direct summands of Rf∗ QX .
The principle follows from the validity of the decomposition theorem on Y and
on every U and from the fact that the summands of the decomposition theorem
over U are uniquely determined (this is left as Exercise 4.7.9).
In the proof of Theorem 4.5.1 we shall also make use of a simple fact concerning
topological coverings that we leave as Exercise 4.7.13
Proof. (of Theorem 4.5.1)
• WLOG, we may assume that Y 0 := f(X) = Y, i.e. that f is surjective.
∼ ⊕q>0 ⊕(S,L)∈EV ICS (L)[−q].
• We have f∗ ICX = q
 Mark Andrea A. de Cataldo 43

• By the two localization principles above, we can replace Y with any of its
Zariski-dense open subsets.
• We may thus assume that there are no enriched proper subvarieties in
the decomposition theorem:
∼ ⊕q>0 ICY (Lq )[−q].
f∗ ICX =
• By constructibility, and by further shrinking if necessary, may also assume
that ICY (Lq ) = Lq is locally constant and we get
∼
M M
Rf∗ ICX = Rq f∗ ICq [−q] = Lq [−q].
q>0 q>0

• We may also assume that ICY = QY .

• WLOG, we may assume that X is normal. In fact, take the normalization
ν : X 0 → X; by the IC Normalization Principle, we have Rν∗ ICX 0 = ICX ;
on the other hand, we have R(f ◦ ν)∗ = Rf∗ ◦ Rν∗ (ν is finite, so Rν∗ = R0 ν∗ ;
but we do not need this here).
• FACT: regardless of normality, there is always a natural map QX → ICX
in place. Since X is normal, this map induces QX = ∼ H0 (ICX ), i.e. we have
[1]
a distinguished triangle QX → ICX → τ>1 ICX → . We push it forward
and deduce
R0 f∗ QX = R0 f∗ ICX .
• We are thus reduced to showing that, up to further-shrinkage, QY is a
direct summand of L0 = R0 f∗ ICX = R0 f∗ QX .
• Stein-factorize f := h ◦ g, with h finite and with g having connected fibers.
• Because of connected fibers, we have R0 g∗ QX = QY . We are thus reduced
to showing that, after shrinking, QY is a direct summand of R0 h∗ QY 0 .
• Since h is finite, by shrinking the target if necessary, h becomes a covering
map.
• The claim now follows by a standard trace argument (Exercise 4.7.13). 
4.6. Pure Hodge structure on intersection cohomology groups
The theory of mixed Hodge modules of M. Saito [48] endows the intersection
cohomology groups of complex varieties with a mixed Hodge structure.
Let us use the intersection complex as a direct summand Theorem 4.5.1 to
endow the intersection cohomology of a complete surface with a pure Hodge
structure. This is merely to illustrate the method, which works for any algebraic
variety [11, 12, 20]. This special case is simple because the resolution is semi-small,
yet illuminating because its simplified set-up allows to focus on the main ideas
without distractions.
Let Y be a complete surface. We are interested in IH∗ (Y, Q), so that we may
assume that Y is normal, for the intersection cohomology groups do not change
under normalization. In particular, Y has isolated singularities. Let S be the finite
set of singular points. Pick a resolution of the singularities f : X → Y that leave
44 Perverse sheaves and the topology of algebraic varieties

Yreg untouched. By the decomposition theorem, Rf∗ QX [2] = ICY ⊕ ⊕b VSb [−b],
where VSb is a skyscraper sheaf at S. Since all fibers have dimension 6 1, we have
Rq>3 f∗ QX = 0. It follows that VSb>1 = 0. By the symmetries of Verdier duality,
we have VSb6−1 = 0 and we have Rf∗ QX [2] = ∼ ICY ⊕ V 0 . This is perverse and
S
self-dual. In this case, t associated pairing yields the usual intersection pairing
on H∗ (X, Q) = IH∗ (Y, Q) ⊕ VS . The two summands are orthogonal for this pair-
ing (there are no maps ICY → VS0 !). The l.h.s. is a pure Hodge structure. The
pairing is a map of pure Hodge structures. In order to conclude that IH∗ (Y, Q) is
a pure Hodge substructure (our goal!), it is enough to show that VS0 ⊆ H2 (X, Q)
is a pure Hodge substructure. By the support conditions for IC, H0 (ICY ) = 0. It
follows that VS0 = R2 f∗ QX = H2 (f−1 (S), Q), which is generated by the fundamen-
tal classes of the curve fibers, which are of (p, q)-type (1, 1), i.e. they form a pure
Hodge structure of weight two.
4.7. Exercises for Lecture 4

Exercise 4.7.1. (Failure of local Poincaré duality; QY not a direct summand) Let
Y be the affine cone over a nonsingular embedded projective curve of genus g > 1.
Use the defining property ωY to show that ωY 6= QY [4] so that the usual local
Poincaré duality fails. Take the usual resolutions f : X → Y. Use the fundamental
relation Rf∗ (C∨ ) = (Rf∗ C)∨ and deduce, by using the failure of Poincaré duality
in neighborhoods of the vertex, that QY is not a direct summand of Rf∗ QX .

Exercise 4.7.2. (Goresky-MacPherson’s estimate) Let f : X → Y be proper with

X nonsingular (or at least with ICX = QX ). Assume that S ⊆ Y appears in the
decomposition theorem for Rf∗ QX [dim X]. Show that
dim X − dim S 6 2 dim f−1 (s), ∀s ∈ S.
Observe that equality implies that S appears only in perversity zero (i.e. with shift
b = 0 only). Deduce from this that, for example, in the decomposition theorem
for the small resolution of the affine cone over P1 × P1 ⊆ P3 , the vertex does
not contribute direct summands. What happens if the map f is of pure relative
dimension 1? (Relate the answer to the number of irreducible components in the
fibers and consider what kind of very special property the intersection complexes
appearing in the decomposition theorem should enjoy). What happens if, in
addition to having f of relative dimension one, Y is also nonsingular, or at least
has ICY = QY ?

Exercise 4.7.3. (Primitive Lefschetz decomposition) Familiarize yourself with

the primitive Lefschetz decomposition (PLD) associated with the hard Lefschetz
theorem for compact Kähler manifolds; see [9], for example. Observe that the
same proof of the PLD holds if you start with a bounded graded object H∗ in an
abelian category with Hb = 0 for |b|  0 and endowed with a degree two operator
∼ Hb for every b > 0. Deduce the appropriate PLD for the
η satisfying ηb : H−b =
graded object H arising from the relative hard Lefschetz theorem 4.3.1. State an
p ∗
 Mark Andrea A. de Cataldo 45

appropriate version of the unimodality of the Betti numbers of Kähler manifolds

in the context of the graded object pH∗ ∈ P(Y).

Exercise 4.7.4. (Proofs of hard (vache (!), in French) Lefschetz ) Let i : Y → X

be a nonsingular hyperplane section of a nonsingular projective manifold. Use
Poincaré duality and the slogan “cup product in cohomology = transverse inter-
section in homology," to show that we have commutative diagrams for every r > 1
(for r = 1 we get a triangle!)
ηr
(4.7.5) Hdim X−r (X, Q) / Hdim X+r (X, Q)
O
restriction i∗ i! Gysin
 ηr−1
|Y
Hdim Y−(r−1) (Y, Q) / Hdim Y+(r−1) (X, Q)

Assume the Hard Lefschetz for Y (induction). Use the Lefschetz hyperplane theo-
rem and deduce the hard Lefschetz for X, but only for r > 2 (for r = 0 it is trivial).
For r = 1, we have the commutative triangle with i∗ injective and i! surjective
η
(4.7.6) Hdim X−1 (X, Q) / Hdim X+1 (X, Q)
6
i∗ i!

restriction ( Gysin
Hdim Y (Y, Q).
Hard Lefschetz boils down to the statement that Im i∗ ∩ Ker i! = {0}. Show that
hard Lefschetz is equivalent to the statement: (∗) the non-degenerate intersection
form on Hdim Y (Y, Q) stays non-degenerate when restricted to i∗ Hdim X−1 (X, Q).
At this point, we have two options. Option 1: use the Hodge-Riemann bilinear
relations for Y: a class i∗ a 6= 0 in the intersection would be primitive and the same
would be true for its (p, q)-components; argue that we may assume a to be of
type (p, q); the Hodge-Riemann relations would then yield a contradiction since
R R
0 6= Y (i∗ a)2 = X η ∧ a2 = 0,. Option 2: put Y in a pencil X e → P1 with smooth
total space (blow up X); let Σ be the set of critical values of f; use the global
invariant cycle theorem and Deligne semi-simplicity to show that the semisimple
π1 (P1 \ Σ)-module Hdim Y (Y, Q) has i∗ Hdim Y (X, Q) as its module of invariants.
Conclude by first proving and then by using the following lemma ([29, p.218]):
let V a completely reducible linear representation of a group π, endowed with a
π-invariant and non-degenerate bilinear form Φ; then the restriction of Φ to the
invariants V π is non-degenerate.

Exercise 4.7.7. (Four symmetries) Let f : X → Y be a projective map of projec-

tive varieties with X nonsingular and Y of dimension 4. Assume that the direct
image perverse cohomology sheaves pHb of Rf∗ QX [dim X] live in the interval
[−2, 2] and that the enriched varieties appearing in the decomposition theorem
are supported at a point S0 , at a curve S1 , and at a surface S2 on Y. Denote
the b-th direct summands of H∗ (X, Q) stemming from (4.1.3) by H∗b (X, Q). Each
46 Perverse sheaves and the topology of algebraic varieties

of these groups further decomposes into direct summands labelled by the Sk :

H∗b (X, Q) = ⊕k H∗b,Sk (X, Q). List the four kind of symmetries among the various
H∗b,Sk (X, Q): Verdier Duality and Relative Hard Lefschetz for b and −b; Verdier
duality and Hard Lefschetz in intersection cohomology for the for the same b.
(See [16, §2.4].)

Exercise 4.7.8. (Perverse cohomology sheaves and the decomposition theorem)

Show that if P ∈ P(Y), then P = pH0 (P). Show that if C =
L
Pb [−b] with
Pb ∈ P(Y), then H (C) = Pb . Let j : C \ o → C be the natural open em-
p b n n

bedding (where o ∈ Cn is the origin). Compute pHi (Rj! Q) and pHi (Rj∗ Q). Let
j : U → X be an open embedding and let P ∈ P(U). Deduce that in the decompo-
sition theorem for C := Rf∗ ICS (L), we have that pHb (C) = ⊕EVb ICS (L), whereas
in the one for K := Rf∗ ICS (L) we have a less simple expression (involving the
dimensions of the varieties S).

Exercise 4.7.9. (The summands in the decomposition theorem are uniquely de-
termined) Prove that the direct summands in the decomposition theorem are
uniquely determined. How non-unique is the isomorphism in the statement of
the decomposition theorem?

Exercise 4.7.10. (No extra supports) Let f : X → Y be a proper map with X a non-
singular surface and Y a curve. Use the proof of Theorem 4.2.1 to show that if all
fibers are irreducible, then the enriched varieties appearing in the decomposition
theorem for Rf∗ QX are supported on all of Y.

Exercise 4.7.11. (IC = Q for Cn /G) Let G be a finite group acting on a complex
vector space V and let f : X := V → Y := V/G be the resulting finite quotient
map. Show that the natural map QY → R0 f∗ QX splits. Deduce that QY [dim Y]
is Verdier self-dual. Deduce that QY [dim Y] is perverse. Show that moreover, it
satisfies the conditions of support that characterize the intersection complex ICY .

Exercise 4.7.12. (IC and finite maps) Prove that the direct image of an intersection
complex under a finite map is an intersection complex.

Exercise 4.7.13. (Coverings traces) Let f : X → Y be a proper submersion of fiber

dimension zero (finite topological covering). Prove (without using the fancy semi-
simplicity results seen above) that L := f∗ QX is a semi-simple locally constant
sheaf admitting QY as a direct summand (trace map). If the covering is normal
(a.k.a. Galois), use the language of representations of the fundamental group to
reach the same conclusion.

5. Lecture 5: the perverse filtration

Summary of Lecture 5. The classical “topologists”’ Leray spectral sequence for fiber
bundles. Grothendieck spectral sequence, Leray as a special case. Verdier’s spectral objects.
Geometric description of the perverse Leray filtration. Relation to the topologists’ point
 Mark Andrea A. de Cataldo 47

of view. Hodge-theoretic applications. The P=W theorem and conjecture in non abelian
Hodge theory. A sample perversity calculation. A motivic question on the projectors that
can be associated with the decomposition theorem.
5.1. The perverse spectral sequence and the perverse filtration
It is important to keep in mind that given f : X → Y, the Leray filtration L on
H (X, Q) is defined a priori, independently of the Leray spectral sequence. The
∗

latter is machinery that tells you something about the graded pieces GrL i H (X, Q)
d

of the Leray filtration, i.e. GrL i H (X, Q) = E∞

d d−i,i .

The Leray spectral sequence for a fiber bundle. As references, you can con-
sult Spanier’s or Hatcher’s algebraic topology textbooks. Let p : E → B be a
topological fiber bundle with fiber F. Assume you are given a cell complex struc-
ture B• on B: in short, we have the p-th skeleta Bp , Bp \ Bp−1 is a disjoint union
of p-cells, Hr (Bp , Bp−1 ) = H ∼ δrp Q# (we call this the
e r (Bouquet of # p-spheres) =
cellularity condition), etc. The cohomology of the complex Hp (Bp , Bp−1 ) with
differentials given by consideration of the coboundary operators in the long ex-
act sequence of the triples (Bp , Bp−1 , Bp−2 ), computes H∗ (B, Q). In fact, there
is a spectral sequence Ep,q 1 = Hp+q (Bs , Bs−1 ), but the cellularity condition re-
duces the spectral sequence to a complex. The same kind of spectral sequence
for the pre-images E• of B• reads E1p,q = Hp+q (Ep , Ep−1 ) and it does not re-
duce to a complex. The bundle structure and the cellularity condition tell us that
Ep,q
1 = Hp (Bp , Bp−1 ) ⊗ Hq (F). One then argues that Epq 2 = Hp (B, Rq ). This is
probably close in spirit to the original way of viewing the Leray spectral sequence
for a fiber bundle. The increasing Leray filtration on the cohomology of the total
space is given by the kernel of the restriction maps to pre images of the skeleta
KerH? (E, Q) → H? (E?? , Q). Let us not worry about indexing schemes.
Grothendieck’s Leray spectral sequence. You can consult Grothendieck’s “To-
hoku” paper [34]. Grothendieck gave a sheaf-theoretic approach to this picture:
start with a complex of sheaves C on Y; take a Cartan-Eilenberg resolution for C,
i.e. an injective resolution C → I that “is” also an injective resolution for the trun-
cated complexes τ6i C and for the cohomology sheaves Hi (C); the complex of
global sections Γ (Y, I) is filtered; the Grothendieck spectral sequence is the spec-
tral sequence for this filtered complex, and it abuts to the standard (Grothendieck)
filtration given by Im H∗ (Y, τ6i C) ⊆ H∗ (Y, C). Given f : X → Y and C ∈ D(X), the
(Grothendieck-)Leray spectral sequence is the Grothendieck spectral sequence for
Rf∗ C, and the (Grothendieck)-Leray filtration is the standard filtration for Rf∗ C.
Perverse and perverse Leray spectral sequences and filtrations. The main in-
put leading to the machinery above uses of injective resolutions together with the
system of standard truncation maps. If we replace the standard truncation with
the perverse truncation maps, we obtain the perverse and perverse Leray spectral
sequences and the perverse and perverse Leray filtrations. For C ∈ D(Y), the
perverse filtration is given by setting Pb H∗ (Y, C) := Im H∗ (Y, pτ6b C) ⊆ H∗ (Y, C).
48 Perverse sheaves and the topology of algebraic varieties

Similarly for the perverse Leray filtration

Pb H∗ (X, Rf∗ K) := Im H∗ (Y, pτ6b Rf∗ K) ⊆ H∗ (Y, Rf∗ K) = H∗ (X, K).
In fact, every t-structure on D(Y), and there are many different ones!, gives rise to
the same kind of picture outlined above. The Grothendieck and Leray filtrations
correspond to the standard t-structure on D(Y).
Verdier’s spectral objects. There is at least one other convenient way to view
these mechanisms, one that, once you are given a cohomological functor, avoids
injective resolutions, namely Verdier’s spectral objects [30]: the input is a cohomo-
logical functor on a t-category; the output is a spectral sequence abutting to the
filtration defined in cohomology by the t-truncations. Even if not logically neces-
sary in our context, this is a useful tool when there are not enough injectives (e.g.
the category of perverse sheaves does not have enough injectives), and it is (yet
another!) cool way to look at spectral sequences.
Why bother with the perverse Leray filtration? Because in the context of the
decomposition theorem
∼ ⊕b ⊕EV ICS (L)[−b] = ⊕b pHb (C)[−b],
C := Rf∗ ICX (M) = b

the perverse Leray spectral sequence is E2 -degenerate, the graded pieces of the
perverse Leray filtration are the cohomology groups H∗ (Y, pHb (C)) and the coho-
mological decomposition theorem gives splittings of the perverse Leray filtration.
In particular, every cohomology class in IH∗ (X, M) has b-components, which
split further into EVb -components. The perverse filtration is canonical, so are
its graded pieces, as well as their splitting into EVb -pieces; what is usually not
canonical, is the choice of a splitting. In particular, if you know that the perverse
filtration has some property, e.g. it carries a Hodge structure, then the graded
b-pieces inherit such a structure as well, and so will the individual EVb -pieces.
See Corollary 5.3.1 and Remark 5.3.2.
Skeleta in algebraic geometry? Let us go back to Leray for fiber bundles. In
that topological context, it is natural to work with a cell complex structure. We
can do that with smooth projective maps f : X → Y in complex algebraic geometry
(varieties can be triangulated), but the skeleta will not be algebraic subvarieties. It
is hard to predict the properties of the Leray filtration if it is described as a kernel
of a restriction map to some closed subspace that is not a subvariety. We may say
that in the context of topological fiber bundles, the Leray filtration is described
geometrically (using the geometry at hand) via the kernels of the pull-back maps
to pre-images of skeleta. For maps of complex algebraic varieties, we would like
a geometric description of the perverse and perverse Leray filtrations, but the
skeleta of a cell-decomposition do not seem to be immediately helpful.
5.2. Geometric description of the perverse filtration

Question 5.2.1. Can we describe the perverse filtration Pb H∗ (Y, C) geometrically?

 Mark Andrea A. de Cataldo 49

Let us approach this problem in the special case that is reminiscent of the
decomposition theorem, i.e. let us assume that C = ⊕b Pb [−b] ∈ D(Y) with
Pb ∈ P(Y). In this case, the perverse filtration comes to us already canonically
split:
 
(5.2.2) Pb H∗ (Y, C) = Im ⊕b 0 6b H∗−b (Y, Pb ) ⊆ H∗ (Y, C).
This is a good place to remark that the decomposition theorem asserts the ex-
istence of a direct sum decomposition, not that one can find a natural one. In
general, there is no such thing. A relatively ample line bundle provides you with
the possibility of choosing some distinguished splittings; see [30] and [12].
Exercise 5.6.1 proves half of the following fact: let P be a perverse sheaf on a
quasi projective variety Y and let Y• be a general flag of linear sections of Y for
some embedding in projective space; here Yk ⊆ Y has codimension k in Y; then
P|Yk [−k] ∈ P(Yk ).
Exercise 5.6.2 first asks you to apply repeatedly the Lefschetz hyperplane theo-
rem for perverse sheaves to the elements of the flag Y• to show that the restriction
maps H∗ (Y, P) → H∗ (Yk , P|Yk ) are injective for every ∗ 6 −k. Next, it asks you to
specialize the situation to the case when Y is affine, to use Artin vanishing and
deduce that, for Y affine, the restriction maps H∗ (Y, P) → H∗ (Yk , P|Yk ) are zero
for ∗ > −k.
We conclude that, when C = ⊕b Pb [−b] ∈ D(Y), with Pb ∈ P(Y) and Y is affine,
we have a geometric description of the perverse filtration
(5.2.3) Ker(H∗ (Y, C) → H∗ (Yk , C|Yk )) = P∗+k−1 H∗ (Y, C).
In other words, for cohomological purposes, we may consider the Y• as the skeleta
of a “cell” decomposition; the term cells now refer to the fact that the relative
cohomology groups H∗ (Yk , Yk+1 , P) are non zero in at most one cohomological
degree, which is reminiscent of the analogous fact for cell complexes (vanishing
for bouquet of spheres).
By renumbering 5.2.3, we get
(5.2.4) Pb H∗ (Y, C) = Ker(H∗ (Y, C) → H∗ (Yb−∗+1 , C|Yb−∗+1 )).
What if C ∈ D(Y) is not split and Y is not affine?
If Y is affine, then the exact same description, but with a different proof, re-
mains valid for every C ∈ D(Y).
If Y is quasi projective, then we can use the Jouanolou trick (Exercise 5.6.4), to
reduce to the affine situation; see [11]. The use of this trick is not necessary and
one can work directly on Y, but one has to use a general pair of flags coming from
a suitable embedding in projective space.
Let us state the end result for the perverse Leray filtration in the special, but
key case of a map to an affine variety. Note that the map need not to be proper
and that it applies to every complex, not just one whose direct image splits as
above. For a discussion of the case of the standard Leray filtration, see [10].
50 Perverse sheaves and the topology of algebraic varieties

Theorem 5.2.5. (Geometric description of the perverse Leray filtration [11, 21])
Let f : X → Y be a map of varieties with Y affine and let K ∈ D(X). Then there is a flag
Y• ⊆ Y, with pre-image flag X• ⊆ X, such that
Pb H∗ (X, K) = Ker (H∗ (X, K) → H∗ (Xb−∗+1 , K|Xb−∗+1 ) .

5.3. Hodge-theoretic consequences

Here is a corollary that exemplifies the utility of having a geometric description
of the perverse filtration as the kernels of restriction maps.

Corollary 5.3.1. (Perverse Leray and MHS) Let f : X → Y be a map of varieties. Then
the subspaces Pb ⊆ H∗ (X, Q) of the perverse Leray filtration on the cohomology of the
domain are mixed Hodge substructures. In particular, the graded pieces carry a natural
mixed Hodge structure.

Remark 5.3.2. (Streamlined Hodge-theoretic proof of the decomposition theo-

rem) Corollary 5.3.1 can be used as the basis for a streamlined Hodge-theoretic
proof of the decomposition and allied results for the push-forward of the inter-
section complex of a variety. This would shorten considerably the proofs in [16].

5.4. Character variety and Higgs moduli: P = W

There is a version of the story outlined in what follows for any complex re-
ductive group; one can even consider higher dimensional projective manifolds
instead of just curves. However, we stick with GL(2, C) and curves.
Fix a smooth projective curve X of genus g > 2 and a point x ∈ X.
Character variety MB . Let MB be the moduli space of irreducible representa-
tions of π1 (X \ x) → GL(2, C) subject to the condition that a small loop circuiting
x maps to −Id. This is a nonsingular affine variety of dimension 2a := 8g − 6.
Higgs moduli space and Hitchin map. Let MD be the moduli space of stable
rank two and degree one Higgs bundles (E, φ) on X, where φ is a one form with
coefficients in End(E). This is a nonsingular quasi projective variety of the same
dimension 2a. It is neither affine, nor quasi projective: it carries the projective
Hitchin map h : MD → A = ∼ Ca , (E, φ) 7→ (trace(φ), det(φ)) (sections of the
∗
cotangent line bundle T X and of its tensor square line bundle), and the general
fiber is an abelian variety of dimension a.
Non abelian Hodge theorem. Part of the non abelian Hodge theorem states
that these two varieties are naturally diffeomorphic. They are not biholomorphic.
There is also a third moduli space in the picture, related to flat connections; but
we stick to our limited set-up above.
Mixed Hodge structures on H∗ (MD ) and H∗ (MB ). The mixed Hodge theory
of both sides is relatively well-understood: the one for MD is pure (Exercise 5.6.6),
so that the weight filtration WD on H∗ (MD , Q) is the filtration by cohomological
degree: 0 = WD,d−1 Hd ⊆ WD,d Hd = Hd , or GrW D d
d H = Hd . The mixed
Hodge structure on H∗ (MB ) is more interesting: the odd graded pieces GrW B d
odd H
WB d
are zero, while the even ones Gr2b H are of pure type (b, b) (Hodge-Tate) and
 Mark Andrea A. de Cataldo 51

the non-purity, here, refers to the fact that there are degrees d for which we
have Hd = ⊕b GrW B d
2b H with more than one non-trivial summand; the non-trivial
graded pieces live in the interval [0, 4a].
WB and WD do not match. By what above, it is clear that the weight filtrations
WB (mixedness) and WD (purity) do not correspond under the diffeomorphism
MB = ∼ MD .
The curious hard Lefschetz (CHL) phenomenon on H∗ (MB , Q). There is a
distinguished cohomology class α ∈ H2 (M, Q) which is linked to a curious phe-
nomenon concerning the MB -side, i.e. there is a sort of hard Lefschetz statement
of the form:
∼
=
(5.4.1) αb ∪ − : GrW B ∗ WB
2a−2b H (MB ) −→ Gr2a+2b H
∗+2b
(MB ).
It is called the curious hard Lefschetz because it looks like a hard Lefschetz-
kind of statement. However, we are not in an actual classical hard-Lefscehtz-type
situation, for MB is affine and α is known to be a (2, 2) class!

Question 5.4.2. Via the non abelian Hodge theorem isomorphism

∼ H∗ (MD , Q),
H∗ (MB , Q) =
what corresponds to WB together with its curious hard Lefschetz, on the H∗ (MD , Q)-
side?

To answer this question, we first normalize the perverse Leray filtration P on

H∗ (MD , Q) for the Hitchin map so that its graded pieces are trivial outside the
interval [0, 2a]. We denote the result by PD . This means that instead of working
with Rf∗ QMD , we work with Rf∗ QMD [a] (Exercise 5.6.7).
Recall that the analogous interval for the weight filtration WB is [0, 4a] and
that the odd graded pieces are trivial. This makes the following answer to Ques-
tion 5.4.2 “numerically” plausible: intervals match by halving, in view of [0, 2a]
and [0, 4a] and GrW B
odd = 0. On the CHL/RHL-side, the answer is also made plau-
sible by the fact that the class α is ample on the fibers of the Hitchin map, so that
indeed it gives rise to a RHL.

Theorem 5.4.3. (P=W [13]) For G = GL/SL/PGL(2, C), via the non abelian Hodge
∼ MD , we have:
theorem MB =
WB,2b ←→ PD,b ∀b, CHL ←→ RHL.

Of course, even if numerically plausible, the fact that the subspaces of the
filtrations match and that CHL turns into RHL seems striking to some of us.
The proof of Theorem 5.4.3 makes an essential use of the geometric description
of the perverse filtration for the Hitchin map based on Theorem 5.2.5: generators
and relations for the cohomology ring H∗ (MB , Q) are known (Hausel-Thaddeus,
Hausel-Rodriguez Villegas); the generators are of pure type (p, p) (for various
values of p) hence live in WB,2p ; every cohomology class is a sum of monomials
in these generators; such monomials have type which is the sum of the types of
52 Perverse sheaves and the topology of algebraic varieties

the factors; their level in WB is the sum of the levels of the factors; the proof then
hinges on the verification that all monomials of level 2b in WB live in PD,b ; in
turn this follows by verifying that the generators have this property and, critically,
that PD is multiplicative (i.e. the level of a cup product is not more than the sum
of the levels of the factors).
In the GL2 /SL2 /PGL2 -case, it is not hard to verify that the generators have
the required property. The heart of the proof of the P=W Theorem 5.4.3 in [13]
consists of showing that the perverse Leray filtration for the Hitchin map is multi-
plicative with respect to the cup product. This is automatic for the Leray filtration
of any map, but fails in general for the perverse Leray filtration (Exercise 5.6.8).
In our case, the Leray filtration differs from the perverse Leray filtration.
Let us illustrate the use of Theorem 5.2.5 with a calculation whose result
tells us that a certain generator, let us call it β ∈ H4 (M, Q), of type (2, 2) in
WB,4 H4 (MB , Q), in fact lies in PD,2 H4 (MD , Q). See [13, §3.1].
By keeping in mind the normalization above of the perverse filtration, the
geometric description of the perverse filtration Theorem 5.2.5 requires us to verify
that the class β vanishes over the pre-image of a generic affine line in A = ∼ Ca
(end of Exercise 5.6.7).
The class β ∈ H4 (MD , Q) is known to be a multiple of the second Chern class
of the tangent bundle of MD . Since the generic fiber is an abelian variety, it is
clear that β vanishes over the pre-image of a generic point. More is true: every
linear function on A gives rise to a Hamiltonian vector field tangent to the fibers
of the Hitchin map; since the tangent bundle of MD,reg (pre-image of regular
values Areg of h) is an extension of the pull-back of the (trivial) tangent bundle
of Areg by the relative tangent bundle (also trivialized by the Hamiltonian vector
fields above), we see that in fact β is trivial on MD,reg .
Ngô’s striking support theorem [45] tells us that there is a Zariski dense open
set Aell ⊆ A with closed complement of codimension > g − 2 such that the de-
∼
composition theorem over Aell is of the form Rhell ∗ Q = ⊕q>0 ICAell (R )[−q] (in
q

[13], we reach this conclusion directly and complement it by showing that these
intersection complexes are in fact sheaves on Aell ). A generic line will avoid the
small closed complement (at least if g > 3; g = 2 can be dealt with separately),
where Rq are the locally constant direct image sheaves over the regular part.
Pick a generic line Λ and observe that the decomposition theorem for the
Hitchin map restricted over the line reads:
∼
M
∗ Q=
RhΛ ICΛ (Rq )[−q];
q>0

this is because the restriction of an (un-shifted) intersection complex IC to a gen-

eral linear section is an (un-shifted) intersection complex.
 Mark Andrea A. de Cataldo 53

Let j : Λreg → Λ be the open immersion of the set of regular values of hΛ .

Since Λ is a nonsingular curve, we know that
∼
M
RhΛ∗ Q= j∗ Rq [−q];
q>0

this is because intersection complexes on nonsingular curves are obtained via the
ordinary sheaf-theoretic push-forward (Fact 1.7.19).
Note that our perverse sheaves are now just sheaves (up to shift). It follows that
the perverse spectral sequence for hΛ is just the ordinary Leray spectral sequence
and the same holds for map hΛreg over the set of regular values of hΛ .
By the functoriality of the Leray spectral sequence and by Artin vanishing on
the affine curves Λ and Λreg (H>1 = 0!), we have a commutative diagram

0 / H1 (Λ, j∗ R3 ) / H4 (MΛ ) / H0 (A, j∗ R4 ) /0

(5.4.4) =
  
0 / H1 (Λreg , R3 ) / H4 (MΛ ) / H0 (R4 ) / 0,
reg

of short exact sequences (the edge sequences for the Leray spectral sequences for
the maps h) where the first vertical map is injective (edge sequence for the Leray
spectral sequence for the map j), and the third is an isomorphism (definition of
direct image sheaf).
A simple diagram chase, tells us that the vertical restriction map in the middle
of (5.4.4) is also injective.
On the other hand, the class β|MΛ 7→ β|MΛreg = 0 by what seen earlier (β re-
stricts to zero over Areg , hence over Λreg ).
By the injectivity statement above, we see that β vanishes over the generic line
and we deduce that β ∈ PD,2 H4 (MD , Q), as predicted by P = W.
Question 5.4.5. We can formulate P = W for every complex reductive group. Does it
hold, at least for GLn ?
There are indications that this should hold for GLn , n small.
5.5. Let us conclude with a motivic question
Let f : X → Y be a projective map of projective varieties with X nonsingular. By
the decomposition theorem, there is an isomorphism
(5.5.1) ∼ ⊕q,EV IH∗−q (S, L).
φ : H∗ (X, Q) = q

This implies that for each (S, L) in EVq we obtain a projector (map that squares
to itself) on H∗ (X, Q) with image φ(IH∗−q (S, L)). We view this projector as a
cohomology class πφ := H2 dim X (X × X, Q).
It is possible to endow each term on the r.h.s. of (5.5.1) with a natural pure
Hodge structure and then to choose an isomorphism φ (5.5.1) that is an isomor-
phism of pure Hodge structures. This implies that πφ is rational and that is has
(p, q)-type (dim X, dim X), i.e. it is a Hodge class.
54 Perverse sheaves and the topology of algebraic varieties

According to the Hodge conjecture, πφ should be algebraic (cohomology class

of an algebraic cycle in X × X).

Question 5.5.2. (Motivic decomposition theorem) Can we chose φ so that the re-
sulting projectors πφ are given by algebraic cycles?

The answer is positive for semi-small maps (Exercise 3.6.8). We have no idea
if/why this should be true. We can prove something much weaker: the projectors
are absolute Hodge (in the sense of Deligne), even motivated (in the sense of
André); see [23].
If the answer were true, then applying it to the blowing up of the projec-
tive cone over an embedded projective manifold would yield the Grothendieck
standard conjecture of Lefschetz type (the inverse to the Hard Lefschetz isomor-
phisms are induced by algebraic cycles in the product); see the introduction to [8].
5.6. Exercises for Lecture 5

Exercise 5.6.1. (Restricting perverse sheaves to general linear sections) Let Y

be quasi projective and P ∈ P(Y). Show that if i : Yk → Y is a codimension k
general linear section of Y relative to any fixed embedding in Y → PN , then
P|Yk [−k] ∈ P(Yk ). To do so first verify directly that P|Yk [−k] satisfies the condi-
tions of support (use the Bertini theorem to cut down the supports of cohomol-
ogy sheaves). It is not so trivial to verify the conditions of co-support, i.e. the
conditions of support for the dual of P|Yk [−k]. Here, we simply say that we can
choose Yk general, depending on P, so that i! = i∗ [−2k] and that the desired
conditions follow formally from this and from the duality exchange property:
(i∗ (P[−k]))∨ = i! P∨ [k] = i∗ P∨ [−k]. (See [16], for example).

Exercise 5.6.2. (Restriction to general linear sections and maps in cohomology)

Let Y be quasi projective, let P ∈ D(Y) and let Yk be the complete intersection of
k general linear sections of Y relative to any embedding of Y in some projective
space. Use the Lefschetz hyperplane theorem to show that the restriction maps
H∗ (Y, P) → H∗ (Yk , P|Yk ) are injective for every ∗ 6 −k. Assume in addition that Y
is affine. Use the Artin vanishing theorem for perverse sheaves and Exercise 5.6.1
to show that the same restriction maps are zero for ∗ > −k.

Exercise 5.6.3. (Geometric description of perverse Leray) Write out explicitly the
conclusion of Theorem 5.2.5 in the case when f is the blow-up at the vertex of the
affine cone over P1 × P1 and verify it.

Exercise 5.6.4. (Jouanolou’s trick) Let Y be quasi projective. There is a map

p : Y → Y with Y affine and that is a Zariski locally trivial bundle with fiber affine
spaces Am . There are various different ways to do this; following is a series
of hints to reach this goal. Choose a projective completion Y ⊆ Y so that the
embedding is affine (e.g. with boundary a divisor). Argue that we may assume
that Y is projective. Pick a suitably positive rank dim Y + 1 vector bundle E on
 Mark Andrea A. de Cataldo 55

Y, where suitably positive here means that the associated line bundle OP(E) (1)
is very ample and has a section s that is not identically zero on any projective
fiber. Show that taking P1 × P1 \ ∆ → P1 yields a special case of the construction
above. Show that Y := P(E) \ (s = 0) does the job. The usefulness of this trick in
our situation is that if we start with f : X → Y, with Y quasi projective, we can
base change to g : X → Y so that now the target is affine and the properties of p
(smooth map with “contractible” fibers) allow us to prove the assertions on f by
first proving them for g and then “descending” them to f.

Exercise 5.6.5. (Leray and perverse Leray) In this exercise use the following: if
j : So → S is an open embedding of nonsingular curves and L is a locally constant
sheaf on So , then ICS (L) = R0 j∗ L. Let p : P1 × P1 → P1 be a projection and let
b : X → P1 × P1 be the blowing up at a point. Let f := p ◦ b. Determine and com-
pare the perverse Leray and the Leray filtrations for f on H∗ (X, Q). (Renumber
the perverse Leray one so that 1 ∈ P0 \ P−1 ; this way they both “start” at the same
“time”.) Do the same thing, but for a Lefschetz pencil of plane curves. Note how
the graded spaces for the two filtrations differ in the “middle”.

Exercise 5.6.6. (H∗ (MD ) is pure). There is a natural C∗ -action on MD obtained

by multiplying the Higgs field φ by a scalar. Show that the Hitchin map is equi-
variant for this action. Use the C∗ -action and the properness of the Hitchin map
to show that the closed embedding of the fiber h−1 (0) → MD induces an isomor-
phism in cohomology. Use the weight inequalities listed at the end of our quick
review of mixed Hodge theory in §1.7, to deduce the Hj (MD ) is pure of weight j
for every j.

Exercise 5.6.7. (Normalizing the perverse filtration) Assume that

C = ⊕b Pb [−b] ∈ D(Y), with Pb ∈ P(Y).
Show that Pb H∗ (Y, C[m])) = Pb+m H∗+m (Y, C) (in fact, this is true in general,
i.e. without assuming that C splits). Deduce that if H∗ (Y, C) 6= 0, then there
is a unique m ∈ Z such that GrP ∗ P ∗
b<0 H (Y, C[m]) = 0 and Gr0 H (Y, C[m]) 6= 0.
The Hitchin map is of pure relative dimension a with a nonsingular domain; use
these two facts, and the decomposition theorem, to deduce that if C = Rf∗ QMD ,
then the m = a, i.e. the perverse Leray filtration for H∗ (A, Rf∗ QMD [a]) “starts” at
level zero, and that it “ends” at level 2a (trivial graded pieces after that). Reality
check: verify that a class lives in PD,2 H4 (MD , Q) if and only if it restricts to zero
over a general line Λ1 ⊆ Aa .

Exercise 5.6.8. (The perverse filtration is not multiplicative in general) First, let
X 0 = S × C (surface times curve, both projective and nonsingular), then let X be
the blowing up of a point in X 0 and let f : X → C be the natural map (blow-down
followed by projection. This is a flat map of relative dimension 2. Let P be the
perverse Leray filtration on H∗ (X, Q) = H∗−2 (C, Rf∗ QX [2]). Verify that it lives in
the interval [0, 4]. Verify that the class e of the exceptional divisor lies in P1 \ P0
56 References

and that e2 lives in P3 \ P2 . Deduce that the perverse filtration is not multiplicative
in general (multiplicative:= Pi ∪ Pj → Pi+j ).

Exercise 5.6.9. (When does Question 5.5.2 have an easy answer?) List some
classes of proper maps f : X → Y such that Question 5.5.2 has an affirmative
answer.

Acknowledgments. I would like to thank M. Goresky, L. Migliorini and R.

Virk for their very useful comments. Special thanks to the anonymous referee for
the excellent suggestions.

References
[1] A. A. Beı̆linson, On the derived category of perverse sheaves, K-theory, arithmetic and geometry
(Moscow, 1984), Lecture Notes in Math., vol. 1289, Springer, Berlin, 1987, pp. 27–41. MR923133
←25
[2] A. A. Beı̆linson, J. Bernstein, and P. Deligne, Faisceaux pervers, Analysis and topology on singular
spaces, I (Luminy, 1981), Astérisque, vol. 100, Soc. Math. France, Paris, 1982, pp. 5–171 (French).
MR751966 ←5, 6, 8, 11, 18, 24, 25, 39
[3] J. Bernstein and V. Lunts, Equivariant sheaves and functors, Lecture Notes in Mathematics, vol. 1578,
Springer-Verlag, Berlin, 1994. MR1299527 ←3
[4] A. Borel et al., Intersection cohomology, Modern Birkhäuser Classics, Birkhäuser Boston, Inc.,
Boston, MA, 2008. Notes on the seminar held at the University of Bern, Bern, 1983; Reprint
of the 1984 edition. MR2401086 ←19, 20
[5] W. Borho and R. MacPherson, Représentations des groupes de Weyl et homologie d’intersection pour
les variétés nilpotentes, C. R. Acad. Sci. Paris Sér. I Math. 292 (1981), no. 15, 707–710 (French, with
English summary). MR618892 ←34
[6] N. Chriss and V. Ginzburg, Representation theory and complex geometry, Birkhäuser Boston, Inc.,
Boston, MA, 1997. MR1433132 ←29, 33, 34
[7] C. H. Clemens, Degeneration of Kähler manifolds, Duke Math. J. 44 (1977), no. 2, 215–290. MR0444662
←6
[8] A. Corti and M. Hanamura, Motivic decomposition and intersection Chow groups. I, Duke Math. J.
103 (2000), no. 3, 459–522. MR1763656 ←54
[9] M. A. de Cataldo, The Hodge theory of projective manifolds, Imperial College Press, London, 2007.
MR2351108 ←2, 44
[10] M. A. A. de Cataldo, The standard filtration on cohomology with compact supports with an appendix
on the base change map and the Lefschetz hyperplane theorem, Interactions of classical and numerical
algebraic geometry, Contemp. Math., vol. 496, Amer. Math. Soc., Providence, RI, 2009, pp. 199–
220. MR2555955 ←49
[11] M. A. A. de Cataldo, The perverse filtration and the Lefschetz hyperplane theorem, II, J. Algebraic
Geom. 21 (2012), no. 2, 305–345. MR2877437 ←43, 49, 50
[12] M. A. A. de Cataldo, Hodge-theoretic splitting mechanisms for projective maps, J. Singul. 7 (2013),
134–156. With an appendix containing a letter from P. Deligne. MR3077721 ←43, 49
[13] M. A. A. de Cataldo, T. Hausel, and L. Migliorini, Topology of Hitchin systems and Hodge theory of
character varieties: the case A1 , Ann. of Math. (2) 175 (2012), no. 3, 1329–1407. MR2912707 ←51, 52
[14] M. A. A. de Cataldo and L. Migliorini, The hard Lefschetz theorem and the topology of semismall
maps, Ann. Sci. École Norm. Sup. (4) 35 (2002), no. 5, 759–772 (English, with English and French
summaries). MR1951443 ←29, 30
[15] M. A. A. de Cataldo and L. Migliorini, The Chow motive of semismall resolutions, Math. Res. Lett.
11 (2004), no. 2-3, 151–170. MR2067464 ←33, 34
[16] M. A. A. de Cataldo and L. Migliorini, The Hodge theory of algebraic maps, Ann. Sci. École Norm.
Sup. (4) 38 (2005), no. 5, 693–750 (English, with English and French summaries). MR2195257 ←8,
18, 30, 38, 39, 46, 50, 54
 References 57

[17] M. A. A. de Cataldo and L. Migliorini, Intersection forms, topology of maps and motivic decomposition
for resolutions of threefolds, Algebraic cycles and motives. Vol. 1, London Math. Soc. Lecture Note
Ser., vol. 343, Cambridge Univ. Press, Cambridge, 2007, pp. 102–137. MR2385301 ←15, 28, 30
[18] M. A. A. de Cataldo and L. Migliorini, The decomposition theorem and the topology of algebraic maps,
“First version”, available at arXiv:0712.0349v1. ←2, 26
[19] M. A. A. de Cataldo and L. Migliorini, The decomposition theorem, perverse sheaves and the topology
of algebraic maps, Bull. Amer. Math. Soc. (N.S.) 46 (2009), no. 4, 535–633. MR2525735 ←2, 3, 22, 25,
40
[20] M. A. A. de Cataldo and L. Migliorini, Hodge-theoretic aspects of the decomposition theorem, Algebraic
geometry—Seattle 2005. Part 2, Proc. Sympos. Pure Math., vol. 80, Amer. Math. Soc., Providence,
RI, 2009, pp. 489–504. MR2483945 ←43
[21] M. A. A. de Cataldo and L. Migliorini, The perverse filtration and the Lefschetz hyperplane theorem,
Ann. of Math. (2) 171 (2010), no. 3, 2089–2113. MR2680404 ←50
[22] M. A. de Cataldo and L. Migliorini, The Hodge theory of maps, Hodge theory, Math. Notes, vol. 49,
Princeton Univ. Press, Princeton, NJ, 2014, pp. 273–296. Lectures 4–5 by de Cataldo. MR3290128
←2
[23] M. A. de Cataldo and L. Migliorini, The projectors of the decomposition theorem are motivated, Math.
Res. Lett. 22 (2015), no. 4, 1061–1088. MR3391877 ←54
[24] P. Deligne, Théorème de Lefschetz et critères de dégénérescence de suites spectrales, Inst. Hautes Études
Sci. Publ. Math. 35 (1968), 259–278 (French). MR0244265 ←4, 7, 8
[25] P. Deligne, Théorie de Hodge. I, Actes du Congrès International des Mathématiciens (Nice, 1970),
Gauthier-Villars, Paris, 1971, pp. 425–430 (French). MR0441965 ←11, 13
[26] P. Deligne, Théorie de Hodge. II, Inst. Hautes Études Sci. Publ. Math. 40 (1971), 5–57 (French).
MR0498551 ←5, 7, 13
[27] P. Deligne, Théorie de Hodge. III, Inst. Hautes Études Sci. Publ. Math. 44 (1974), 5–77 (French).
MR0498552 ←13
[28] P. Deligne, Poids dans la cohomologie des variétés algébriques, Proceedings of the International Con-
gress of Mathematicians (Vancouver, B. C., 1974), Canad. Math. Congress, Montreal, Que., 1975,
pp. 79–85. MR0432648 ←11
[29] P. Deligne, La conjecture de Weil. II, Inst. Hautes Études Sci. Publ. Math. 52 (1980), 137–252 (French).
MR601520 ←8, 11, 40, 45
[30] P. Deligne, Décompositions dans la catégorie dérivée, Motives (Seattle, WA, 1991), Proc. Symp. Pure
Math., vol. 55, Amer. Math. Soc., Providence, RI, 1994, pp. 115–128 (French). MR1265526 ←8, 11,
48, 49
[31] A. H. Durfee, A naive guide to mixed Hodge theory, Singularities, Part 1 (Arcata, Calif., 1981), Proc.
Sympos. Pure Math., vol. 40, Amer. Math. Soc., Providence, RI, 1983, pp. 313–320. MR713069 ←13
[32] S. I. Gelfand and Y. I. Manin, Methods of homological algebra, Springer-Verlag, Berlin, 1996. Trans-
lated from the 1988 Russian original. MR1438306 ←19
[33] M. Goresky and R. MacPherson, Stratified Morse theory, Ergebnisse der Mathematik und ihrer
Grenzgebiete (3) [Results in Mathematics and Related Areas (3)], vol. 14, Springer-Verlag, Berlin,
1988. MR932724 ←20, 22
[34] A. Grothendieck, Sur quelques points d’algèbre homologique, Tôhoku Math. J. (2) 9 (1957), 119–221
(French). MR0102537 ←10, 47
[35] B. Iversen, Cohomology of sheaves, Universitext, Springer-Verlag, Berlin, 1986. MR842190 ←19
[36] D. Kaledin, Symplectic singularities from the Poisson point of view, J. Reine Angew. Math. 600 (2006),
135–156. MR2283801 ←30
[37] M. Kashiwara and P. Schapira, Sheaves on manifolds, Grundlehren der Mathematischen Wis-
senschaften [Fundamental Principles of Mathematical Sciences], vol. 292, Springer-Verlag, Berlin,
1994. With a chapter in French by Christian Houzel; Corrected reprint of the 1990 original.
MR1299726 ←19
[38] S. L. Kleiman, The development of intersection homology theory, Pure Appl. Math. Q. 3 (2007), no. 1,
Special Issue: In honor of Robert D. MacPherson., 225–282. MR2330160 ←18
[39] R. MacPherson, Global questions in the topology of singular spaces, Proceedings of the International
Congress of Mathematicians, Vol. 1, 2 (Warsaw, 1983), PWN, Warsaw, 1984, pp. 213–235. MR804683
←6
[40] J. Milnor, Morse theory, Based on lecture notes by M. Spivak and R. Wells. Annals of Mathematics
Studies, No. 51, Princeton University Press, Princeton, N.J., 1963. MR0163331 ←23
58 References

[41] T. Mochizuki, Asymptotic behaviour of tame harmonic bundles and an application to pure twistor D-
modules. I, Mem. Amer. Math. Soc. 185 (2007), no. 869, xii+324. MR2281877 ←8, 18, 39
[42] T. Mochizuki, Asymptotic behaviour of tame harmonic bundles and an application to pure twistor D-
modules. II, Mem. Amer. Math. Soc. 185 (2007), no. 870, xii+565. MR2283665 ←8, 18, 39
[43] H. Nakajima, Lectures on Hilbert schemes of points on surfaces, University Lecture Series, vol. 18,
American Mathematical Society, Providence, RI, 1999. MR1711344 ←31
[44] B. C. Ngô, Le lemme fondamental pour les algebres de Lie (2008), available at arXiv:0801.0446v1.
←36
[45] B. C. Ngô, Le lemme fondamental pour les algèbres de Lie, Publ. Math. Inst. Hautes Études Sci. 111
(2010), 1–169 (French). MR2653248 ←36, 38, 52
[46] M. V. Nori, Constructible sheaves, Algebra, arithmetic and geometry, Part I, II (Mumbai, 2000), Tata
Inst. Fund. Res. Stud. Math., vol. 16, Tata Inst. Fund. Res., Bombay, 2002, pp. 471–491. MR1940678
←25
[47] C. Sabbah, Polarizable twistor D-modules, Astérisque 300 (2005), vi+208 (English, with English and
French summaries). MR2156523 ←8, 18, 39
[48] M. Saito, Mixed Hodge modules, Publ. Res. Inst. Math. Sci. 26 (1990), no. 2, 221–333. MR1047415 ←8,
11, 18, 39, 43
[49] R. P. Stanley, Combinatorial applications of the hard Lefschetz theorem, Proceedings of the Interna-
tional Congress of Mathematicians, Vol. 1, 2 (Warsaw, 1983), PWN, Warsaw, 1984, pp. 447–453.
MR804700 ←40

Department of Mathematics, SUNY at Stony Brook, Stony Brook, NY 11794, USA

E-mail address: [email protected]