ERASMUS MUNDUS MASTER ALGANT

UNIVERSIT
`
A DEGLI STUDI DI PADOVA
FACOLT
`
A DI SCIENZE MM. FF. NN.
CORSO DI LAUREA IN MATEMATICA
ELABORATO FINALE
INTEGRAL TRANSFORMS OF
CONSTRUCTIBLE FUNCTIONS
RELATORE: PROF. A. DAGNOLO
DIPARTIMENTO DI MATEMATICA PURA E APPLICATA
LAUREANDO: FARHAD BABAEE
ANNO ACCADEMICO 2009/2010
2
Contents
Introduction 5
1 Constructible sheaves and constructible functions 7
1.1 Constructible sheaves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.1.1 Subanalytic stratication . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.1.2 Denition of constructible sheaves . . . . . . . . . . . . . . . . . . . . . . 8
1.2 Constructible functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.2.1 Euler-Poincare index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.2.2 Operations on constructible functions . . . . . . . . . . . . . . . . . . . . 13
2 Integral transforms 17
2.1 Integral transforms of sheaves . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2 Integral transforms of constructible functions . . . . . . . . . . . . . . . . . . . . 19
3 Radon transform on Grassmannians 25
3.1 Basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.2 Inversion formulas for 1

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.2.1 Inverses of 1

when p +q = n . . . . . . . . . . . . . . . . . . . . . . . . 29
3.2.2 The general case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
A Duality formulas and Euler-Poincare index with compact support 39
B Micro-support of a sheaf 43
Introduction
The main objects of study in this thesis are constructible functions. A constructible function on
a real analytic manifold X is an integer valued function which is locally constant on every con-
nected component of a subanalytic stratication of X. On the other hand, an R-constructible
sheaf on X is an object of the derived category of sheaves of k-vector spaces (for k a eld)
having locally constant cohomologies on every component of a subanalytic stratication. It can
be shown that the Grothendieck group of the category of R-constructible sheaves is generated
by linear combinations of classes of constant sheaves on a locally closed subanalytic sets. Hence
one nds an isomorphism of this group with the category of constructible functions, given by:
k
m
Z
m.1
Z
, k
Z
[1] (1).1
Z
.
Here Z X is a compact contractible subanalytic set, m > 0 an integer and 1
Z
the
characteristic function of Z. This isomorphism allows one to translate the operations and
theorems from sheaf theory to the framework of constructible functions.
Among various operations on constructible sheaves, we will highlight the Radon transform. It
is dened in the following way:
Assume is a constructible function on X. Consider the double morphism of real analytic
manifolds:
S
f








g

?
?
?
?
?
?
?
X
Y,
the Radon transform of with the incidence relation S is dened by:
1
S
() =

g
f

.
Where the operation

g
plays the same role on consructible functions, as the proper direct
image, Rg
!
, does on derived category of sheaves.
Given a Radon transform, the question of existence of an inverse naturally arises. P. Schapira
in [8] considered this problem in general. He proved the existence of an inverse under certain
hypotheses. These hypotheses however, are not fullled if one looks at the Radon transforms
on Grassmannians.
In [7], Y. Matsui, under some assumptions, obtained an inverse for the Radon transform on
Grassmannians with the inclusion incidence relation. In order to do it, he considered transforms
6 Introduction
associated to dierent incidence relations and combined them using Cramers rule to obtain an
inverse.
In [6], C. Marastoni considered the transversality incidence relation for dual complex Grass-
mannians, in the framework of constructible sheaves. Using microlocal techniques he proved
that the associated Radon transfom is invertible.
In this thesis we consider the transversality incidence relation for general Grassmannians,
in the framework of constructible functions. We have two main results. On one hand, we
show that for dual Grassmannians the transversality incidence relation is self adjoint. On
the other hand, under certain assumptions, we prove that the associated Radon transform
has an inverse. Here, as in Matsuis, the idea is to consider transforms associated to dierent
incidence relations. However, in our proof we avoid the use of Schubert calculus. In addition,
our techniques could be adapted to obtain a simplied proof of Matsuis main result.
The rst chapter of this thesis is a brief introduction to the category of R-constructible sheaves,
in which we give the grounds for dening constructible functions, as well as the operations on
them.
In the second chapter, we will introduce a type of integral functor over sheaves. Then, we will
consider the analogue of this integral transform in the framework of constructible functions.
Which is in particular the Radon transform.
The third chapter is concentrated on Radon transforms on Grassmannians. In this chapter we
will prove our results.
The last parts of this thesis are appendix A and appendix B. Appendix A is a collection of du-
ality formulas and theorems related to Euler-Poincare index with compact support. Theorems
and formulas which are employed in the earlier chapters. Appendix B, contains a denition of
micro-support which is used in the second chapter. Few basic theorems on micro-supports are
also mentioned.
Acknowledgment: I wish to thank ALGANT consortium for providing me with the ALGANT
grant. I would like to thank Prof. Marco Garuti, for providing such a friendly environment
for the ALGANT students in university of Padova: Prof. Bas Edixhoven for the things that
I learned from him, and for the light he shed on certain theorems which I have employed in
this thesis. Finally, I acknowledge my gratitude to Prof. Andrea DAgnolo for his guidance
throughout this thesis and for his sheer patience in answering my questions.
Chapter 1
Constructible sheaves and
constructible functions
This chapter contains basic denitions and theorems for constructible functions. We will state
denitions of subanalytic stratication, constructible sheaves and nally constructible functions.
Next, we will explain how Euler-Poincare index provides an isomorphism between Grothendieck
group of category of R-constructible sheaves and constructible functions. This isomorphism
plays a key role in dening the operations on constructible functions motivated by the operations
on constructible sheaves. To justify denitions of operations on constructible functions we will
recall series of theorems from theory of sheaves. The main references for this chapter are [5]
and [3].
1.1 Constructible sheaves
In this section we rst introduce subanalytic stratications. Next, we will dene constructible
sheaves which have locally constant cohomology groups and are stable under the Grothendieck
operations.
Notation 1.1.1. Throughout the thesis for an algebraically closed eld k and a topological
manifold X, we denote by D
b
(k
X
), the derived category of bounded complexes of sheaves of k-
vector spaces. Moreover, k
X
is the locally constant sheaf on X with stalk k. If F Ob(D
b
(k
X
))
to lighten the notation we will write F D
b
(k
X
).
1.1.1 Subanalytic stratication
Subanalytic sets, are more general objects than semi-analytic sets (those which are locally
dened by inequalities of analytic functions). Family of subanalytic sets is closed under closure,
complement, inverse images and proper direct images. More precisely:
Denition 1.1.2. Let Y be a real analytic manifold and let Z be a subset of Y . One says
Z is subanalytic at y Y if there exist an open neighborhood U of y, compact manifolds
8 Constructible sheaves and constructible functions
X
i
j
(i = 1, 2, 1 j N) and morphisms f
i
j
: X
i
j
Y such that:
Z U = U
N

j=1
(f
1
j
(X
1
j
)`f
2
j
(X
2
j
)).
If Z is subanalytic at each y Y , one says Z is subanalytic in Y .
Subanalytic sets carry the following properties (for a proof, see [5], p.327).
Proposition 1.1.3. (i) Assume Z is subanalytic in Y . Then Z and Int(Z) are subanalytic
in Y . Moreover the connected components of Z are locally nite and subanalytic.
(ii) Assume Z
1
and Z
2
are subanalytic in Y . Then Z
1
Z
2
, Z
1
`Z
2
and Z
1
Z
2
are subanalytic.
(iii) Let f : X Y be a morphism of manifolds. If Z Y is subanalytic in Y then f
1
(Z)
is subanalytic in X. If W X is subanalytic in X and f is proper on W, then f(W) is
subanalytic in Y .
(iv) Let Z be a closed subanalytic subset of Y . Then there exists a manifold X and a proper
morphism f : X Y such that f(X) = Z.
Denition 1.1.4. For a closed subanalytic subset Y of X, a partition Y =

A
X

is called a
subanalytic stratication of Y , if it is locally nite, the X

s are subanalytic submanifolds and

for all pairs (, ) A A such that X

= one has X

. Each X

is called a
stratum.
1.1.2 Denition of constructible sheaves
Denition 1.1.5. Let F be an object of D
b
(k
X
), the derived category of bounded complexes
k-vector spaces. Assume that there exists a locally nite covering X =

iI
X
i
by subanalytic
subsets such that for all j Z, all i I, the sheaves H
j
(F)[
X
i
are locally constant of nite
rank. Then one says that F is R-constructible. We denote by D
b
Rc
(k
X
) the full triangulated
subcategory of D
b
(k
X
) consisting of R-constructible objects.
Example 1.1.6. Assume Z is a locally closed subanalytic subset of an analytic manifold X,
and denote by j : Z X the embedding. Note if X =

is a subanalytic stratication,
then so is X = (

Z) . (

(X`Z)). Let F D
b
Rc
(k
X
). For F
Z
= Rj
!
j
1
F one
has:

F
Z
[
Z
= F[
Z
;
F
Z
[
X\Z
= 0.
As a result F
Z
locally constant on both types of strata X

Z and X

(X`Z). Therefore,
F
Z
D
b
Rc
(k
X
).
Unlike the category locally constant sheaves, the category of constructible sheaves is preserved
under many natural sheaf theoretic operations.
Theorem 1.1.7. (i) Let f : X Y be an analytic morphism of real analytic spaces. Then
the following holds:
1.1 Constructible sheaves 9
(a) If G D
b
Rc
(k
Y
) then f
1
G D
b
Rc
(k
X
) and f
!
G D
b
Rc
(k
X
).
(b) If F D
b
Rc
(k
X
) and restriction of f to supp(F) is proper, then Rf

(F) and Rf
!
(F)
are R-constructible.
(ii) If F, G D
b
Rc
(k
X
) then F
L
G and RHom(F, G) D
b
Rc
(k
X
).
(iii) Let F D
b
(k
X
). Then F is R-constructible if and only if the dual
1
D
X
F is R-constuctible.
In particular, the dualizing complex
X
= D
X
k
X
is R-constructible.
(iv) If F D
b
Rc
(k
X
) then we have the following:
(a) The natural morphism F D
X
(D
X
(F)) is an isomorphism;
(b) For any x X one has (D
X
F)
x
RHom(R
x
(X, F), k).
Example 1.1.8. Let X = C and S = C0 and F a sheaf on S with supp(F) = x
n
: n N,
where x
n
=
1
n+1
and F
xn
= C
n
. We see that F is an R-constructible sheaf on S. Yet, the sheaf
i
!
F is not. Indeed, if i
!
F was an R-constructible sheaf we would have a subanalytic stratication
of X and the origin would be in a stratum dierent from each of x
n
. Such a stratication
though, could not be locally nite.
Recall that for a continuous map f : X Y and sheaf F D
b
(A
X
) and a sheaf G D
b
(A
Y
)
we have (i) f
!
(DG) D(f
1
G) and (ii) Rf

(DF) D(Rf
!
F). Therefore, by the last theorem
we get:
Corollary 1.1.9. Let f : X Y be an analytic morphism real analytic spaces. Then,
(i) if F D
b
Rc
(A
X
), then Rf
!
(DF) D(Rf

(F)),
(ii) if G D
b
Rc
(A
Y
), then f
1
(DG) D(f
!
(G)).
Another lemma will be useful later.
Lemma 1.1.10. Let F D
b
wRc
(A
X
) and let : X R
n
be a real analytic function.
Assume [
supp(F)
is proper. Then we have the natural isomorphism:
(i) R(
1
(B

); F) R(
1
(B

); F) R(
1
(0); F), for 0 < <1,
(ii) R

1
(0)
(X; F) R

1
(B

)
(X; F) R
c
(
1
(B

); F), for 0 < <1.

1
For a denition of dual see A.6.
10 Constructible sheaves and constructible functions
1.2 Constructible functions
Denition 1.2.1. A function : X Z is constructible if:
(i) for all m Z,
1
(m) is subanalytic,
(ii) the family
1
(m)
mZ
is locally nite.
We denote by CF(X) the set of constructible functions on X. Which is a ring under usual
operations of addition and multiplication. The presheaf U CF(U) (U open in X) is a sheaf.
We denote it by CF
X
.
In the above denition
1
(m)
mZ
gives a subanalytic, locally nite covering of X. Therefore,
by Hardt triangulation theorem([5], Proposition 8.2.5), there exists a locally nite, subanalytic
stratication

I
K

, such that all K

s are compact and contractible. Hence we simplify:

Lemma 1.2.2. A function : X Z is constructible if and only if there exists a locally nite
family of compact subanalytic contractible subsets K
i

iI
, such that:
=

iI
c
i
1
K
i
,
where c
i
Z and 1
A
is characteristic function of the subset A, i.e. 1
A
(x) =

1 if x A;
0 if x / A.
1.2.1 Euler-Poincare index
Theory of constructible functions is based on an observation on the local Euler-Poincare Index
of a constructible sheaf. Recall that for F an object of D
b
f
(k
X
), the derived category of nitely
generated k-vector spaces, the local Euler-Poincar`e index of F is dened to be:
(F) =

j
(1)
j
dimH
j
(F).
Denition 1.2.3. Let X be an analytic manifold and F D
b
Rc
(k
X
). For any x X one sets:
(F)(x) = (F
x
),

c
(F)(x) = (R
x
(X; F)) = (D(F))(x).
Moreover, if R(X; F) (resp. R
c
(X; F)) belongs to D
b
f
(k
X
), one sets:
(X; F) = (R(X; F))
(resp.
c
(X; F) = (R
c
(X; F))).
The integer (X; F) is called the Euler-Poincare index of F and the function (F)(x) is called
the local Euler-Poincare index of F.
1.2 Constructible functions 11
Remark 1.2.4. The equality
c
(F)(x) = (D(F))(x) in the preceding denition is implied by
Theorem 1.1.7
Example 1.2.5. (i) If F D
b
Rc
(k
X
), then by denition there exists a subanalytic strati-
cation

such that F[
X
are locally constant, and so for a connected component of
each X

, say Z, H
j
(F)[
Z
is a constant integer. Therefore (F) is a constructible function.
(ii) It is easy to see that (R, k
5
[0,1]
+k
[2,3]
[+1]) = 51
[0,1]
1
[2,3]
.
(iii) Assume Y is a locally closed subset of X, then:
(R
c
(X; k
Y
)) = (R
c
(X; i
!
k
Y
)) = (R
c
(Y ; i
!
i
1
k
X
)) =
c
(Y ).
Recall that satises:
(F G) = (F) +(G)
(F G) = (F).(G)
for F, G D
b
Rc
(k
X
). And for a distinguished triangle F

F F

+1
, we have:
(F) = (F

) +(F

).
As these properties suggest, it is suitable to look at Grothendieck group of D
b
Rc
(k
X
).
Denition 1.2.6. Let C be an abelian (resp. triangulated) category. One denotes by K(C)
the abelian group obtained as the quotient of the free abelian group generated by the objects of
C by the relation X = X

+ X

if there is an exact sequence 0 X

X X

0 (resp.
a distinguished triangle X

X X

+1
in C). One calls K(C) the Grothendieck Group of
C.
Remark 1.2.7. Let C be an abelian category and, D
b
(C) the derived category of bounded
complexes of C. Consider the distinguished triangles, (see [5], p.47):

n
(X) X
n+1
(X)
+1
,

n1
(X)
n
(X) H
n
(X)[n]
+1
,
H
n
(X)[n]
n
(X)
n+1
(X)
+1
.
By induction one can easily see that for an X D
b
(C), class of X, [X] K(D
b
(C)), can be
represented by:

j
[H
j
(X)[j]].
Therefore, i : C D
b
(C), X X, induces a group isomorphism K(C) K(D
b
(C)), which
its inverse is given by

j
(1)
j
[H
j
(X)].
We will denote the Grothendieck group of D
b
Rc
(k
X
) by K
Rc
(k
X
).
12 Constructible sheaves and constructible functions
Theorem 1.2.8. The map induces a group isomorphism from K
Rc
(k
X
) to CF(X).
Proof. We explain the proof of this theorem from [5]. The machinery of applying induction on
a subanalytic stratication is often useful.
I. Surjectivity. This is done exactly as in the Example 1.2.5. For a CF
X
, choose a
subanalytic stratication X =

A
X

such that =

A
m

1
X
, m Z. Let

= sgn(m

)
for m

= 0 and dene:
F =

k
|m|
X

where A

= A; m

= 0. Clearly F D
b
Rc
(k
X
), and (F) = . So the surjectivity follows.
II. Injectivity. To prove the injectivity, rst observe that every element u in K
Rc
(k
X
), is
represented by a nite sum u =

j
a
j
[F
j
], with a
j
Z. For any F D
b
Rc
(k
X
), we will set
F
n
= F ... F (n times), for n N and F
n
= F
n
[+1] for n Z and n < 0. Therefore,
we may rewrite u = [

j
F
a
j
j
]. Hence any u K
Rc
(k
X
) is represented by a single object
F =
j
F
a
j
j
D
b
Rc
.
Let X =

be a subanalytic stratication such that H

j
(F)[
Z
is constant for all j, all
. Let X
n
denote the union of the n-codimensional strata. Using the distinguished triangle
F
X
0
F F
X\X
0
+1
, gives:
[F] = [F
X
0
] + [F
X\X
0
]
and by induction we get F =

n
[F
Xn
]. Remark 1.2.7 implies:
[F] =

j,n
(1)
j
[H
j
(F)
Xn
] =

j,n
(1)
j
[k
dim(H
j
(F)
Xn
)
Xn
]. (1.1)
Hence (F) = 0 implies that for any :
dim

j even
H
j
(F)
Z
= dim

j odd
H
j
(F)
Z
.
Since k is a eld, this implies:

j even
H
j
(F)
Z
=

j odd
H
j
(F)
Z
and

j even
H
j
(F)
X
=

j odd
H
j
(F)
X
.
This shows [F] =

j,n
(1)
j
[H
j
(F)
Xn
] = 0.
Remark 1.2.9. The Equation 1.1 implies that [k
X
]s linearly generate K
Rc
(k
X
).
1.2 Constructible functions 13
1.2.2 Operations on constructible functions
Having the isomorphism of the Theorem 1.2.8 at hand, we can dene many operations on
constructible functions analogous to those of constructible sheaves.
I. External Product
CF
X
CF
Y
CF
XY
For , CF
X
, we can nd F, G D
b
Rc
(k
X
), such that = (F) and = (G). We have
= (F G). On the other hand by K unneth formula A.14, (F G) = (F)(G), which
suggests us to dene:
( )(x, y) = (x).(y).
II. Inverse Image
Let f : X Y be morphism of manifolds. Dening:
f

: f
1
CF
Y
CF
X
f

(x) = (f(x)),
we readily see that the diagram:
K
Rc
(A
Y
)

f
1

K
Rc
(k
X
)

f
1
CF
Y
f

CF
X
is commutative.
III. Integral
For
c
(X; CF
X
) we can choose F D
b
Rc
(k
X
) such that supp(F) = supp() (following the
proof of the Theorem 1.2.8 we see this choice is possible). One sets:

X
= (X; F).
Additivity of the functor (X, .) with respect to distinguished triangles and 1.2.8 show that the
number is only depends on . By 1.2.2 we may rene the expression =

1
X
, m Z,
to the case when X

is a family of locally nite, compact, contractible sets. For which:

X
=

(X; k
X
) =

.
Note that (

X
)(x) = (X; F
x
), for x X.
IV. Direct Image
Let f : X Y be a morphism of manifolds, and CF
X
. Assume f is proper on supp().
One denes the direct image of by f:
14 Constructible sheaves and constructible functions
(

f
)(y) =

X
.1
f
1
(y)
.
Which gives a morphism of sheaves:

f
: f
!
CF
X
CF
Y
.
Note that (Rf
!
G)(y) = (Y ; Gk
f
1
(y)
). Since the bers formula, A.3, gives:
(Rf
!
G)
y
R
c
(f
1
(y); G[
f
1
(y)
) R
c
(f
1
(y); i
1
G)
= R(X; i
!
i
1
G) R(Y ; Gk
f
1
(y)
).
Where i : f
1
(y) X is the inclusion. We will therefore have (f
!
G) =

f
.
V. Dual
There are two types of dual operators which we will introduce.
(a) For a CF
X
and F D
b
Rc
(k
X
), such that (F) = one denes D
X
: CF
X
CF
X
by
D
X
() = (D
X
F). This denition is well-dened, since D
X
is a triangulated functor. On
the other hand, by Denition 1.2.3:
(D
X
)(x) =
c
(F)(x) =
c
(R
x
(X; F)).
Now from Lemma 1.1.10, we can nd a chart (U, ) in X, containing x, such that (U)
B(0, ). Therefore:
R
x
(X; F) R
c
(
1
(B(0, )); F) R(X; F k

1
(B(0,))
).
Writing
1
(B(0, )) as B(x, ) gives an equivalent denition for the Dual:
(D
X
)(x) =

X
1
B(x,)

for 0 < <1.

(b) Second type of dual
2
, is a counterpart for D

X
.
If X is an oriented manifold, then by proposition A.7 one has the isomorphism:

X
k
X
[+n].
Therefore, for a F D
b
(k
X
) :
D

X
F = RHom(F, k
X
) = RHom(F,
X
[n]) = RHom(F[+n],
X
) = D
X
(F[+n]).
Consequently for CF
X
, let F D
b
Rc
(k
X
) such that = (F). Dene:
D

X
() = (D

X
F).
Note that D

X
() = (D
X
(F[+n])) = (1)
n
D
X
().
2
Refer to Denition A.6.
1.2 Constructible functions 15
We end this chapter by few relations which are followed easily from the Theorem 1.2.8 and
similar formulas for constuctible sheaves.
Proposition 1.2.10. (i) Let CF
X
. Then
D
X
D
X
= .
(ii) Let f : X Y be a morphism of manifolds and let (X; f
!
CF
X
). Then

f
D
X
= D
Y

f
.
(iii) Consider a Cartesian square of morphism of real analytic manifolds:
X

X
f

Y
(Recall that this means X

(y

, x) Y

X : g(y

) = f(x).)
Then for (X; f
!
CF
X
) one has:
g

f
=

(h

).
16 Constructible sheaves and constructible functions
Chapter 2
Integral transforms
In the rst chapter we dened constructible functions and constructible sheaves and explained
how the isomorphism lead to dening operations on constructible functions analogous to those
of constructible sheaves. In this chapter we introduce certain integral transforms in each setting.
We will introduce a so called Radon transform of constructible functions. We will prove general
properties of Radon transforms.
2.1 Integral transforms of sheaves
We recall the denition of integral transforms for sheaves from [6].
Let X, Y and Z be real analytic manifolds and k be a eld. Consider the diagram:
X Y Z
q
12
o
o
o
o
o
o
o
o
o
o
o
q
13

q
23

N
N
N
N
N
N
N
N
N
N
N
X Y
q
1

q
2

O
O
O
O
O
O
O
O
O
O
O
O
O
X Z
q

1
o
o
o
o
o
o
o
o
o
o
o
o
q

O
O
O
O
O
O
O
O
O
O
O
O
Y Z
q

1
o
o
o
o
o
o
o
o
o
o
o
o
q

X Y Z.
where all the maps are projections.
Denition 2.1.1. Let X, Y and Z be manifolds, and let K D
b
(k
XY
), K

D
b
(k
Y Z
). We
set composition K K

to be:
K K

= Rq
13!
(q
1
12
K q
1
23
K

) D
b
(k
XZ
).
If in the above denition we put X = pt, Y = X and Z = Y then by identication of ptX
with X and pt Y with Y we get the denition of the integral transforms for sheaves.
Denition 2.1.2. For any K D
b
(k
XY
), the integral transform of sheaves from X to Y with
kernel K is given by:
. K : D
b
(k
X
) D
b
(k
Y
), F K = Rq
2!
(q
1
1
F K).
18 Integral transforms
Using the following lemma one can compose integrals.
Lemma 2.1.3. Using the notations in Denitions 2.1.2 and 2.1.1 we have:
(F K) K

F (K K

).
Proof. Since the square:
X Y Z
q
12
o
o
o
o
o
o
o
o
o
o
o
q
23

N
N
N
N
N
N
N
N
N
N
N
X Y
q
2

O
O
O
O
O
O
O
O
O
O
O
O
O
Y Z
q

1
o
o
o
o
o
o
o
o
o
o
o
o
Y
.
is Cartesian (see Corollary A.4), therefore:
(F K) K

= Rq

2!

q
1
2
(Rq
2!
(K q
1
1
F))

Rq

2!

Rq
23!
q
1
12
(K q
1
1
F))

and
Rq

2!

Rq
23!
(q
1
12
K q
1
12
q
1
1
F))

= Rq

2!

Rq
23!
(q
1
12
K (q
1
q
12
)
1
F))

= Rq

2!

Rq
23!
(q
1
12
K (q

1
q
13
)
1
F))

.
By projection formula (Proposition A.1.15):
Rq

2!
Rq
23!

q
1
23
K

(q
1
12
K q
1
13
q

1
1
F))

R(q

2
q
23
)
!

q
1
23
K

(q
1
12
K q
1
13
q

1
1
F))

Rq

2!
Rq
13!

q
1
23
K

(q
1
12
K q
1
13
q

1
1
F))

.
Again by projection formula:
Rq

2!

Rq
13!
(q
1
23
K

q
1
12
K) q

1
1
F)

= F (K K

).
Corollary 2.1.4. In the situation of the preceding lemma assume Z = X, K K

X
[l]
and K

K k

Y
[l

] for some shifts l and l

(
X
and
Y
denote the diagonal of X X and
Y Y , respectively). Then l = l

and . K and . K

are equivalences of categories. One calls

. K

the inverse of the functor . K.

2.2 Integral transforms of constructible functions 19
Proof. We only need to observe that for an F D
b
(A
X
),
F k

X
= Rq
2!
((q
1
F)

) = F (2.1)
and use Lemma 2.1.3. Note that the isomorphisms K[l] K K

K K[l

], give l = l

.
Marastoni in [6] puts forward inverses for integral operators of sheaves under certain geometric
conditions
1
.
Theorem 2.1.5 ([6], Lemma 2.2). Assume X and Y are real analytic compact orientable
manifolds of the same dimension n. an open subanalytic subset of X Y . Let
t
be the
image of under the map .
t
: X Y Y X, (x, y) (y, x). Denote by j (resp. j
t
) the
embedding of into X Y (resp. of
t
into Y X). For any x X we set:

x
= y Y : (x, y) Y,
and similarly for y Y . Moreover consider the kernels:
K

= C

= j
!
j
1
C
XY
D
b
Rc
(C
XY
),
K

= D

XY
C

= Rj

j
1
C
XY
D
b
Rc
(C
XY
).
Under the hypotheses
(i) X is simply connected,
(ii) R(
x
; C

) =

0 for x = x

C for x = x

,
(iii) SS(C

) (T

X
X T

Y ) T

XY
(X Y ),
and similar conditions by interchanging X and Y , one has:
K K

t C

X
[n] and K

t K C

Y
[n].
In particular, the categories D
b
Rc
(C
X
) and D
b
Rc
(C
Y
) are equivalent.
2.2 Integral transforms of constructible functions
Consider the diagram of morphisms of real analytic manifolds:
X Y
q
1
.w
w
w
w
w
w
w
w
w
q
2

G
G
G
G
G
G
G
G
G
X
Y,
1
For the denition of micro-support see the Appendix B
20 Integral transforms
where q
1
and q
2
are projections. Assume k CF
XY
. Similar to 2.1.2, an integral transform
from CF
X
to CF
Y
with kernel k(x, y) is dened by the map:

q
2
k(x, y) q

1
(x).
Note that by denition of proper image, p
2
needs to be proper on supp(k(x, y)), for the integral
to make sense. Since k CF
XY
, by the Lemma 1.2.2, one writes:
k(x, y) =

1
S
(x, y),
such that S

is a locally nite family of subanalytic, compact and contractible subsets of

XY. As a consequence, an integral transform of constructible functions can be reduced to the
integral transform with kernels 1
S
, where S is locally closed subanalytic subset of XY. Namely,
the Radon transform: such that S

is a locally nite family of subanalytic, compact and

contractible subsets of XY. As a consequence, an integral transform of constructible functions
can be reduced to the integral transform with kernels 1
S
, where S is locally closed subanalytic
subset of X Y. Namely, the Radon transform:
Denition 2.2.1. Let S be a locally closed subanalytic subset of X Y . We denote by p
1
and
p
2
the rst and second projections dened on X Y , and by f and g the restrictions of p
1
and
p
2
to S.
We assume
p
2
is proper on

S, the closure of S in X Y. (2.2)
Then, for a CF
X
, we set:
1
S
() =

g
f

q
2
1
S
(q

1
).
We call 1
s
() the Radon transform of .
Remark 2.2.2. In the situation of the preceding theorem, we have 1
S
(1
{x}
)(y) = 1
S
(x, y).
Therefore, 1
S
= 0 implies S = . Since Radon transform is additive, we learn that the inverse
of a Radon transform, if it exists, it is unique.
By the same fashion of Lemma 2.1.3 one can prove a similar formula for of Radon transforms.
The following lemma is a part of Theorem 3.1 in [8].
Let S

Y X be a locally closed subanalytic subset, and again denote by p

2
and p
1
the rst
and second projections dened on Y X, and by f

and g

the restrictions of p
1
and p
2
to S

and by r the projection S

Y
S

X X.
We assume:
p
1
is proper on

S

, the closure of S

in Y X. (2.3)
Lemma 2.2.3. Under assumptions 2.2 and 2.3, for any CF
X
, we have:
1
S
1
S
() =

q
2

r
1
S
Y
S

1
. (2.4)
2.2 Integral transforms of constructible functions 21
First proof. Denote by h and h

the projections from S

Y
S

to S and S

respectively.
The square:
S
Y
S

S
g

Y
is Cartesian. Therefore one writes:
1
S
1
S
() =

(g

g
(f

)) =

((f h)

).
Considering the diagram:
S
Y
S

h
.x
x
x
x
x
x
x
x
x
r

G
G
G
G
G
G
G
G
G
S
f








g

F
F
F
F
F
F
F
F
F
F
X X
q
1
m
m
m
m
m
m
m
m
m
m
m
m
m
m
m
q
2

R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
S

.w
w
w
w
w
w
w
w
w
w
f

A
A
A
A
A
A
A
X Y
X,
gives:
1
S
1
S
() =

q
2

r
(r

1
) =

q
2

r
r

1
XX

1
.
Note that we have used the fact that for any CF
XX

r
r

1
XX

r
r

r
1
S
Y
S

.
For the last step (one can prove this by checking on characteristic functions of compact subsets
of X X, or by noting that it is followed by projection formula A.13).
Second proof. Let k be a eld and in the Lemma 2.1.3, put Z = X, K = k
S
D
b
Rc
(k
XY
),
K

= k
S
D
b
Rc
(k
Y X
) and note that:
(q
1
12
k
S
q
1
23
k
S
) = 1
S
Y
S
;
Moreover, the map r plays the same role as q
13
in 2.1.2.
To go further in the preceding Lemma, Schapira in [8] assumes:
There exists = Z such that:

r
1
S
Y
S
= 1

X
+1
XX\
X
.
Where
X
is diagonal of X X. Dierently put,
There exists = Z such that: (r
1
(x, x

)) =

if x = x

;
if x = x

.
(2.5)
We will refer to condition (2.5) as -condition.
The main theorem in [8] is the following:
22 Integral transforms
Theorem 2.2.4 ([8], Theorem 3.1). Assume (2.2), (2.3) and (2.5). Then for any CF
X
,
we have:
1
S
1
S
() = ( ) +

1
X
. (2.6)
Proof. Lemma 2.2.3 and (2.5) together with the equalities

q
2
1

X
q

1
= (2.7)
and
q
2
1
XX
q

1
=

X
,
oer the statement.
The upcoming lemma and proposition are useful to simplify (2.6) and the ideas in them will
be employed later on.
Denition 2.2.5. For a CF
Y
, we set
R
0
() =

p
1
1
Y X
(p

2
) =

p
1
(p

2
) =

1
X
.
Lemma 2.2.6 (Proposition 3.1 in [7]). Assume (2.2), (2.3) and (2.5), then for any CF
X
we have:
R
0
1
S
() =

X
()1
X
.
Proof. Having proved the equality for characteristic functions of compact subanalytic subsets
K X, by linearity of above operations we deduce the assertion.
Since the square
XY
p
2

p
1

Y
a
Y

X
a
X

Y
is Cartesian, we have:
R
0
1
S
(1
K
) =

p
1
p

p
2
(1
S
.p

1
1
K
) = a

a
Y

p
2
(1
S
.p

1
1
K
)
= a

a
X

p
1
(1
S
.p

1
1
K
).
For any CF
X
, we write:

a
X

(x) =

a
X

(pt) =

X
(x

)1
a
1
X
{pt}
(x

) =

1
X
(x).
Since we have set:
= ((x Y ) S) = (y Y : (x, y) S).
2.2 Integral transforms of constructible functions 23
We deduce:

p
1
1
S
.p

1
1
K

(x) =

XY
1
(({x}K)Y )S
(x

, y

) = 1
K
.
Which gives the desired result.
Denition 2.2.7. For a CF
Y
, we set:
R
1
() =

p
1
(1
S
t 1
Y X
)(p

2
) = 1
S
t () R
0
().
So the Proposition 3.2 in [7] follows:
Proposition 2.2.8. Assume (2.2), (2.3) and (2.5) and let CF
Y
. Then we have:
R
1
1
S
() = ( ).
In particular, if ( ) is not zero, we can reconstruct the original constructible function
from its Radon transform 1
S
() by dividing the last term by this constant ( ).
Proof.
R
1
1
S
() = 1
S
t 1
S
() R
0
1
S
() = ( ).
24 Integral transforms
Chapter 3
Radon transform on Grassmannians
In this chapter we concentrate on the Radon transforms of Grassmannians. In the rst section
we apply the theorems from previous chapter to draw basic conclusions on these Radon trans-
forms. In the second section we prove the main results of this thesis and we will try to extract
inversion formulas for the Radon transform with transversality incidence relation.
3.1 Basics
This section is mainly application of the theorems in the second chapter for the Radon
transform on Grassmannians.
We rst x few notations.
Denition 3.1.1. Let V be a vector space of dimention n over R or C (we will discuss the two
cases in parallel).
(i) G
n
(p) = set of p-planes in the vector space k
n
, the Grassmannian of p-planes in n-space
(ii) S
p,q
i
= (x, y) G
n
(p) G
n
(q) : dim(x y) = i, for xed integers 1 p q n and
i = 0, 1, .., p.
(iii) G
n
(p, q) = S
p,q
p
, the inclusion incidence relation.
(iv)
p,q
= S
p,q
0
, the transversality relation.
In the above, when p and q are xed we may drop the superscript p, q.
Remark 3.1.2. Natural action of the group G := SL(n) on each of G
n
(p) and G
n
(q), induces
an action on G
n
(p) G
n
(q) dened by g(x, y) := (gx, gy) for any g G. Orbits of this action
are clearly S
p,q
i
s. Each S
i
is a submanifold of G
n
(p) G
n
(q). For 0 i p, the closure of
S
i
,

S
i
, is given by

ji
S
j
. Thus, its complement

j=i
j=0
S
j
is open. In particular,

p
i=0
S
i
gives a
subanalytic stratication for G
n
(p) G
n
(q). Moreover, note that G
n
(p, q) (resp. ) is compact
(resp. open) submanifold of G
n
(p) G
n
(q).
Example 3.1.3. For any integers
0
,
1
, . . . ,
p
, the function

p
i=0

i
1
S
i
is a constructible
function.
26 Radon transform on Grassmannians
Example 3.1.4 (Inclusion incidence relation). Considering the diagram:
G
n
(p, q)
f
.t
t
t
t
t
t
t
t
t
g

K
K
K
K
K
K
K
K
K
K
G
n
(p) G
n
(q),
we verify -condition for S = G
n
(p, q) and S

= S
t
. As before, let r : S
Gn(q)
S

G
n
(p)
G
n
(p) be projection. For (x, x

) G
n
(p) G
n
(p) we have:
r
1
(x, x

) = y G
n
(q) : x y, x

y = y G
n
(q) : x +x

y
Suppose dim(xx

) = j for any 0 j p, then dim(x+x

) = 2p j. By taking quotient over

x x

we get:
r
1
(x, x

) G
n2p+j
(q 2p +j).
We recall that the formulas for Euler-Poincare Index of Grassmannians are given by
(G
n
(p)) =

0 if p(n p) is odd

E(
n
2
)
E(
p
2
)

if p(n p) is even
for the real Grassmannian, where E(x) is the integer part of x; And
(G
n
(p)) =

n
p

for the complex Grassmannian. Denote this numbers in both Real and Complex case by
n
(p).
Therefore, when p > 1 there is no hope for -condition to hold. On the other hand when p = 1,
j is either 0 or 1. Hence we have -condition:
(r
1
(x, x

)) =

n2
(q 2) if x = x

n1
(q 1) if x = x

.
Now by 2.6 we can write:
Proposition 3.1.5 ([8], Proposition 4.1). For any CF
Gn(1)
1
Gn(1,q)
1
Gn(q,1)
() = (
n1
(q 1)
n2
(q 2)) + [
n2
(q 2)

Gn(1)
]1
Gn(1).
(3.1)
Remark 3.1.6. The preceding example shows that for p > 1 -condition is not satised.
Although we could extract some information by considering dierent cases for intersections of
xx

, these information is not enough for nding an inverse. These cases are the main subject
of [7]. To extract more information, Matsui considers p + 1 cases for S

and uses that to nd

an inverse (See section 3.3.2 and Remark 3.2.17).
3.1 Basics 27
Example 3.1.7. Let S
n
be the n-dimensional sphere R
n
0/R
>0
. We denote by (S
n
)

the
set (R
n
)

0)/R
>0
, the dual of S
n
. One considers these spaces as oriented Grassmannians.
Let us take X = S
n
and Y = (S
n
)

. We set the incidence relation to be:

S = (x, y) X Y : x y.
Let S

= S
t
and r the projection S
Y
S

X X, as before. To verify -condition we

calculate (r
1
(x, x

)), for x, x

X. Assume x = x

, then:

(y (S
n
)

: x y, x

y) (S
(n1)
)

;
(y (S
n
)

: x y (S
(n2)
)

.
Recalling Cohomology Groups of spheres and using Poincare Duality (Corollary A.8)
(r
1
(x, x

)) =

2 if x = x

and n is odd ;
0 if x = x

and n is odd.
For n even, we get the numbers the other way around. Hence we can use (2.6) to get:
Proposition 3.1.8. Suppose n is odd. In the situation of Example 3.1.7, for each CF
S
we have:
1
S
1
S
() = .
This means that for a subanalytic subset K S
n
, by knowledge of 1
S
(1
K
) at each point of
(S
n
)

we can reconstruct K.
Example 3.1.9 ([8], p.8). Let V be an n-dimensional real ane space, P
n
its projective com-
pacticaion, i.e. P
n
= V . h

, where h

is the hyperplane at innity. Let P

n
be the dual
projective space. Therefore, P

n
`h

is the set of hyperplanes of V .

As in Example 3.1.4, consider X = G
n+1
(1) = P
n
and Y = G
n+1
(n) = P

n
. Let S = G
n+1
(1, n),
the inclusion incidence relation, and K a subanalytic subset of V . For a hyperplane in V one
has:
1
S
(1
K
)() =

q
2
1
S
.q

(1
K
)

() =

XY
1
S
.1
{(x,):xX,x}
.q

1
1
K
.
Hence:
1
S
(1
K
)() =

V
1
K
.1

= (K ).
By (2.6):
1
P

n
1
Pn
() =

if n is odd ,
+ [

Pn
]1
Pn
if n is even and n > 0 .
The above formula means, that to reconstruct a subanalytic subset K V in odd dimensions, it
suces to know the Euler-Poincare Index of every section of the set K with hyperplanes. Letting
n = 3, one might consider this result as an application of Radon transforms in tomography.
28 Radon transform on Grassmannians
3.2 Inversion formulas for 1

This section contains the main results of this thesis. We will attempt to nd inverses for
the Radon transform on Grassmannians with transversality incidence relation. In addition to
Denition 3.1.1 we x few others.
Denition 3.2.1. Assume 1 p q n are integers. We set: (S
p,q
i
)
x
= y G
n
(q) : (x, y)
S
p,q
i
, for a xed x G
n
(q).
Remark 3.2.2. Assume that l p n are integers and z G
n
(l) and x G
n
(p), are given.
Then the set
(S
p,q
0
)
x
(S
l,q
0
)
z
=
p,q
x

l,q
z
=

y G
n
(q) : y z = y x = 0

,
is non-empty if and only if p + q n. This follows from the fact that if l = p and p + q = n
the set is non-empty, as in preceding section. Having n > p + q or p l means more freedom
of choice and therefore the set being non-empty. Therefore, we have the suciency. Necessity
of the condition is obvious.
In the forthcoming sections we consider the cases when p+q = n and the general case, separately.
The following lemma will be frequently used.
Remark 3.2.3. Using Schubert cell decomposition of Grassmanian (see [4]) one can show that
the Grassmanians as well as objects of our study in the following sections are CW-complexes.
Moreover, as a result of Example 1.2.5, we will only deal with Euler-Poincare index with compact
support. Therefore, in these sections, by Euler characteristic of a manifod we mean Euler-
Poincare index with compact support of a constant sheaf with stalk k over that manifold.
Lemma 3.2.4. For an x G
n
(p), the Euler-Poincare index with compact support of
p,q
x
=
(S
p,q
0
)
x
G
n
(q) is given by:

(1)
qp
(G
np
(q)) in the real case;
(G
np
(q)) in the complex case.
More explicitly, recalling the formulas for Euler-Poincare indeces of Grassmannians (see Ex-
ample 3.1.4), we have:
(
p,q
x
) =

0 if q(n q p) is odd
(1)
qp

E(
np
2
)
E(
q
2
)

if q(n q p) is even
for the real Grassmannian;
(
p,q
x
) =

n p
q

for the complex Grassmannian.

3.2 Inversion formulas for 1

29
Proof. For simplicity we write
x
instead of
p,q
x
. To calculate
c
(
x
), it is convenient to
x an ordered basis for e
1
, e
2
, ..., e
n
for k
n
, such that x = e
1
, e
2
, ..., e
p
. Note that if y =
spanv
1
, ..., v
q
is an element of
x
, then no linear combination of v
i
s, 1 i q, is entirely
in spane
1
, ..., e
p
. Therefore, y in matrix form has a non-singular q q minor in q of the last
np columns. Let : spane
1
, e
2
, ..., e
n
spane
p+1
, ..., e
n
be the projection into the np
last coordinates. The restriction of to y G
n
(q) : y spane
1
, ..., e
p
= 0 induces a
map:

x
= y G
n
(q) : yspane
1
, ..., e
p
= 0 B = q-planes in spane
p+1
, ..., e
n
G
np
(q).
Which we will also denote it by . We claim that the map is surjective, furthermore, for each
b in B, the
1
(b) is isomorphic to k
qp
. To see this, note that a point b B is q-dimensional.
Hence, by a suitable change of coordinates in spane
p+1
, . . . , e
n
, it can be expressed by:
b =

1
q
0

q(np).
where 1
q
is q q identity matrix, and 0 is a q n p q zero matrix. Accordingly, the ber
of b,
1
(b), is given by:

1
(b) =

1
q
0

qn
,
where is a q p matrix of free variables in k. Inserting these facts into Corollary A.12, for
any b B we get:

c
(A) =
c
(B)
c
(
1
(b)) = (G
np
(q))
c
(k
qp
).
Replacing k with R and C and using examples A.10 and Example 3.1.4, give the formulas.
Corollary 3.2.5. If p +q = n, then

c
(
x
) =

(1)
pq
in the real case;
1 in the complex case.
3.2.1 Inverses of 1

when p +q = n
In this section we will try to nd an inverse for the Radon transform 1

when p + q = n.
This section is motivated by [6]. Marastoni in [6] proves that for the complex Grassmannians
X = G
n
(p) and Y = G
n
(q), if p + q = n, then X Y fullls the conditions of Theorem
2.1.5. As a consequence the inverse of the integral functor . C

, is given by . D

Y X
C

t .
Which indicates that the Radon transform with kernel D

Y X
(1

t ) ( CF
Y X
), is an inverse
for 1

. Explicit calculation of D

Y X
(1

t ) is not easy. However, we found that 1

t is an
inverse for 1

. As a result, by uniqueness of inverses for Radon transforms (Remark 2.2.2),

we learn that 1

t is the only inverse and we have 1

t = D

Y X
1

t .
We will show that and
t
, satisfy -condition, (2.5), in the complex case. To do this for
each (x, x

) G
n
(p) G
n
(p), we will calculate:

c
(r
1
(x, x

)) =
c
(
x

x
) =
c
(y G
q
(n) : y x = y x

= 0).
30 Radon transform on Grassmannians
Assume (x, x

) Z
j
(recall that this means dim(x x

) = j). When x = x

or equivalently
j = p we have:

c
(r
1
(x, x

)) =
c
(y G
q
(n) : y x = 0).
Which is already calculated in Corollary 3.2.5, therefore we only need to calculate the cases
when p j 1. First, we x (x, x

) such that dim(x x

) = j. Then, we nd bases for

x = spane
1
, ..., e
j
, e
j+1
, ..., e
p
and for x

= spane
1
, ...e
j
, e
p+1
, ..., e
2pj
. We extend x + x

=
spane
1
, ...e
j
, e
j+1
, ..., e
p
, e
p+1
, ..., e
2pj
to spane
1
, e
2
, ..., e
n
. Accordingly, in matrix form we
can represent:
x =

1
j
0 0 0
0 1
pj
0 0

pn
and x

1
j
0 0 0
0 0 1
pj
0

pn
,
where 1
m
is m by m identity matrix.
Moreover to write the elements of
x
in a matrix form, note that a vector v =

n
i=1

i
e
i
y
cannot be entirely in x = spane
1
, ..., e
p
, so at least one of the
i
s for i > p is non-zero. As we
are only interested in the span of such vectors, we can assume it is 1. Since we have dim(y) = q
we can choose q = np independent vectors in this way. By some elementary operations (again
since the span of such vectors only matters) on the matrix we can get another basis for y in
the matrix form ( [ 1
q
) or:
y =

[b
1
]
pjj
[b
3
]
pjpj
1
pj
0
pjqpj
[b
2
]
qpjj
[b
4
]
qpjpj
0
qpjpj
1
qpj

.
Knowing that:

x

x
= y
x
: x

y = 0,
for (x, x

) Z
j
. One has:
y =

b
1
b
3
1
pj
0
b
2
b
4
0 1
qpj

if and only if
det

1
j
0 0 0
0 0 1
pj
0
b
1
b
3
1
pj
0
b
2
b
4
0 1
qpj

= 0.
In addition
0 = det

1
j
0 0 0
0 0 1
pj
0
b
1
b
3
1
pj
0
b
2
b
4
0 1
qpj

= det

0 1
pj
0
b
3
1
pj
0
b
4
0 1
qpj

= det

0 1
pj
b
3
1
pj

= det b
3
.
3.2 Inversion formulas for 1

31
Therefore, in the real case, considering the projection:
:
x
R
(pj)
2
,

b
1
b
3
b
2
b
4

b
3
and application of Corollary A.12 and Example A.10, we get:

c
(
x

x
) = (1)
dim(b
1
)+dim(b
2
)+dim(b
4
)

c
(GL
pj
).
When p j = 1 (in real dimensions), we readily have:

c
(GL(p j)) = 2.
For p j > 1 we dene the map : GL(p j) SL(p j) by z
1
det(z)
z. Which is obviously
surjective with the ber k

above each point. Therefore

c
(GL(p j)) =
c
(k

)
c
(SL(p j)),
by Corollary A.12. Since SL(n) is a compact Lie group, by Lemma 3.2.9, we conclude:

c
(GL(p j)) = 0 for p j > 1.
Note that the natural real structure to C and considering it as R
2
, provides the complex case.
We recap:
Proposition 3.2.6. In the real case,

c
(
x

x
) =

0 if j < p 1
2(1)
pq
if j = p 1
(1)
pq
if j = p or x = x

.
In the complex case,

c
(
x

x
) =

0 if j < p or x = x

1 if j = p or x = x

.
The complex case in Proposition 3.2.6, suits ne in -condition (2.5). Using (2.6), we have
shown:
Theorem 3.2.7. The inverse of Radon transform 1

, when p +q = n, in the complex Grass-

mannian is 1

t . That is to say, for any CF

Gn(p)
we have:
1

t 1

() =
Remark 3.2.8. Using Theorem 3.2.11 below, one can nd the inverse also for the real case.
We now give a short proof for the fact about Lie groups which is used in the proof of Proposition
3.2.6.
Lemma 3.2.9. The Euler characteristic of every smooth compact Lie group of dimension
greater than zero is zero.
32 Radon transform on Grassmannians
Proof. Assume G is a compact Lie group and denote by l
x
: G G, y xy the left translation
by x G. Pick a non-zero vector X in the tangent space over the identity element e G.
Noting that for every x G, l
x
is a dieomorphism implies that 1(x) := T
e
(l
x
)X is a non-zero
vector eld on G. Hence by Poincare-Hopf theorem (G) = 0. Recall that the Poincare-Hopf
theorem (cf. [2]) states that the Euler characteristic of a compact, orientable and connected
manifold of dimension greater than zero, is sum of the indices of the singularities of a vector
eld on it. In particular, the Euler Characteristic a manifold is zero when a vector eld on it
has no singularities.
3.2.2 The general case
In the preceding section we found an inverse for 1

, on complex Grassmannians when

p +q = n. In this section we consider this problem in a general case for both real and complex
case. Unfortunately, the ideas of last section do not work here. To gain our results, we will
use a method introduced by Matsui in [7]. However, the computations in [7] include Schubert
calculus, and Matsui dealt with one Schubert variety. Our computations brought up the
question of intersection of two Schubert varieties; The problem of nding the intersection of
Schubert varieties is a long standing problem (see [4]). Formulas which have been extracted
to nd intersection of two Schubert varieties, solve the question for Schubert cells in the
general position. In our problem, on the other hand, non-general positions might occur and
contribute in the calculations of Euler-Poincare index. To perform our calculations, we looked
at brations
1
of the sets. One can easily apply these bration ideas in the next section to
recover main theorem of [7].
Let X = G
n
(p) and Y = G
n
(q). We assume p q, since by taking the dual we can deal with
the cases when p > q. We will use the notation in 3.1.1 and 3.2.1. Therefore
p,q
X Y and
S
q,p
i
Y X. Moreover, we set:
Z
j
= S
p,p
j
= (x, x

) X X : dim(x x

) = j.
For simplicity we write instead of
p,q
and S
i
instead of S
q,p
i
. By Remark 3.1.2, Y X =

p
i=0
S
i
is a subanalytic stratication. For each 0 i p we consider the diagram:

Y
S
i
h
.x
x
x
x
x
x
x
x
x
r

G
G
G
G
G
G
G
G
G

f








g

G
G
G
G
G
G
G
G
G
G
X X
q
1
m
m
m
m
m
m
m
m
m
m
m
m
m
m
m
q
2

R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
S
i
g

.v
v
v
v
v
v
v
v
v
v
f

A
A
A
A
A
A
A
A
X Y X.
In which all the maps are projections. Note that Z
p
= (x, x

) X X : x = x

, and as in
(2.7) for a CF
X
we have

q
2
1
Zp
q

1
= . The key idea here is to use Cramers rule and nd
a kernel such that

r
(kernel) = 1
Zp
.
1
Thanks to Prof. B. Edixhoven for this suggestion.
3.2 Inversion formulas for 1

33
One writes:
(

r
1

Y
S
i
)(x, x

) =
p

j=0
(

Y
S
i
1
r
1
(x,x

)r
1
(Z
j
)
)1
Z
j
. (3.2)
Accordingly, for a xed pair (x, x

) Z
j
,

Y
S
i
1
r
1
(x
1
,x
2
)r
1
(Z
j
)
=
c
(r
1
(x, x

)
Y
S
i
).
Note that:
r
1
(x, x

)
Y
S
i
= y G
n
(q) : x y = 0, dim(x

y) = i.
We dene the matrix T = (t
i,j
)
(p+1)(p+1)
by:
t
i,j
=
c
(r
1
(x, x

)
Y
S
i
) for any 0 i, j p. (3.3)
These numbers are independent of the choice of a pair (x, x

) Z
j
. Moreover, i + j > p
means dim(x

x) + dim(x

y) > dim(x

) which implies that dim(x

(x y)) 1 and
dim(x y) 1. From which we understand that t
i,j
= 0 for i + j > p, i.e. the matrix is
anti-triangular. In the section 2.3.2.1 we will provide recursive formulas to calculate all of the
entries of the matrix T in both real and complex cases.
Writing (3.2) in matrix form yields:
T

1
Z
0
1
Z
1
.
.
.
1
Zp

r
1

Y
S
0

r
1

Y
S
1
.
.
.

r
1

Y
Sp

. (3.4)
Since the matrix T is anti-triangular, its determinant is given by multiplication of entries of
anti-diagonal up to a sign. We will nd the elements of anti-diagonal explicitly in Lemma 3.2.15
and in Theorem 3.2.16 we will show when det(T) = 0.
By Cramers rule we can solve the equation (3.4) with respect to Z
p
and write:
det(T)1
Zp
=

t
0,0
t
0,1
. . . t
0,p1

r
1

Y
S
0
t
1,0
t
1,1
. . . t
1,p1

r
1

Y
S
1
t
2,0
t
2,1
. . . 0

r
1

Y
S
2
.
.
.
.
.
. .
.
. .
.
.
t
p1,0
t
p1,1
. . . 0

r
1

Y
S
p1
t
p,0
0 . . . 0

r
1

Y
Sp

.
This formula yields coecients for 1
S
i
s, say
i
s, such that
0
1
S
0
+ +
p
1
Sp
is an inverse
for 1

.
34 Radon transform on Grassmannians
Denition 3.2.10. (i) We set
K
p,q
= det

t
0,0
t
0,1
. . . t
0,p1
1
S
0
t
1,0
t
1,1
. . . t
1,p1
1
S
1
t
2,0
t
2,1
. . . 0 1
S
2
.
.
.
.
.
. .
.
. .
.
.
t
p1,0
t
p1,1
. . . 0 1
S
p1
t
p,0
0 . . . 0 1
Sp

CF
XY
.
(ii) We dene the map R
1
: CF
Y
CF
X
by

p
1
K
p,q
(p

2
).
The following proposition is an essential part of our main result.
Proposition 3.2.11. For a CF
X
,
R
1
1

() = det(T).
In particular, if det(T) = 0, then R

is invertible and its inverse is given by det(T)

1
R
1
.
Proof. By denition
R
1
1

() =

p
1

K
p,q

p
2
1

.p

q
2

det

t
0,0
t
0,1
. . . t
0,p1

r
1

Y
S
0
t
1,0
t
1,1
. . . t
1,p1

r
1

Y
S
1
t
2,0
t
2,1
. . . 0

r
1

Y
S
2
.
.
.
.
.
. .
.
. .
.
.
t
p1,0
t
p1,1
. . . 0

r
1

Y
S
p1
t
p,0
0 . . . 0

r
1

Y
Sp

q
2
det(T)1
Zp
q

= det(T).
For the second equality we have used Lemma 2.2.3. For the last equality recall that

q
2
1
Zp
q

1
=
.
3.2.2.1 Calculation of the entries of T
In the preceding section we showed how to nd the inverse for R

when det(T) = 0. Recall

that T = (t
i,j
) and t
i,j
=
c
(
p,q
x
(S
q,p
i
)
x
) for a pair (x, x

) S
p,p
j
. In this section we will give
a recursive method for nding entries of T. Next, we will use this method to nd the entries
of anti-diagonal. From which, we realize when T has a non-zero determinant.
3.2 Inversion formulas for 1

35
Consider the Grassmannians X = G
n
(p), Y = G
n
(q) and Z = G
n
(l), for integers 1 l p
q n. By denition:

S
p,q
0
X Y,
S
p,l
j
X Z,
S
q,l

Y Z.
Denition 3.2.12. (i) For a x G
n
(p), we set
g(n, q, p) = (
p,q
x
).
(ii) For given (x, z) S
p,l
j
we set:
f
p,j
(n, q, l, ) =
c
((S
p,q
0
)
x
(S
l,q

)
z
)
=
c
(

y G
n
(q) : y x = 0, dim(y z) =

).
Remark 3.2.13. The values of g(n, q, p) =
c
(
p,q
x
) =
c
((S
p,q
0
)
x
), is calculated in the Lemma
3.2.4.
Note that x y = 0 implies dim(x z) dim(z) dim(y z), which means that for the
quadruple (n, q, l, ) we must have j l . Moreover l = j, means z x and implies = 0.
Therefore:
f
p,j
(n, q, j, 0) = g(n, q, p). (3.5)
For an integer > 0, let:
B =

planes in z with zero intersection with (x z)

b G
l
() : b (x z) = 0

.
Therefore

c
(B) = g(l, , j). (3.6)
Lemma 3.2.14. The map:
: A =

y G
n
(q) : y x = 0, dim(y z) =

B
y y z,
is surjective. Moreover, for each b B one has:

c
(
1
(b)) = f
p,j
(n , q , l , 0).
Proof. Let b B be -plane. Then, in the quotient space k
n
/b let z

= z/b G
n
(l ) and
x

= x/(x b) G
n
(p). Then the set:
A

G
n
(q ) : y

= 0, y

= 0

,
is non-empty. This is followed from Remark 3.2.2 since p + q n and l p. Moreover, for
y

, we have y

b A and (y

b) = (y

b) (z

b) = b. Second part is proved by

taking quotient and noting that
1
(b) A

.
36 Radon transform on Grassmannians
By Lemma 3.2.14 and (3.6) and application of the Corollary A.12 we derive:
f
p,j
(n, q, l, ) = f
p,j
(n , q , l , 0).g(l, , j) for > 0. (3.7)
Using the equality of sets:

y G
n
(q) : y x = 0, dim(y z) = 0

y G
n
(q) : yx = 0

=1

y G
n
(q) : yx = 0, dim(yz) =

(l = dim(z))
and additivity of
c
, we deduce:
f
p,j
(n, q, l, 0) = g(n, q, p)
l

=1
f
p,j
(n, q, l, ).
Obviously we only need to consider 0 < < l j. So we can rene the formula to:
f
p,j
(n, q, l, 0) = g(n, q, p)
lj

=1
f
p,j
(n, q, l, ). (3.8)
In the above recursive formulas we have two steps. In the rst step (3.7) the third and fourth
coordinates of the quadruple (n, q, l, ) are reduced by . On the other hand, if is zero at
one stage then in the second step (3.8), 1 will replace . Hence, at each time, running
the two steps will make the quadruple closer some (, , j, 0), which its f is known by (3.5).
This implies that the procedure will end in nite number of steps and that it can calculate the
desired number.
We apply the recursive formulas (3.7) and (3.8) to nd the entries of anti-diagonal.
Lemma 3.2.15. Elements of the anti-diagonal of T are given by:
t
i,j
= t
i,pi
=

0 if (n i)(n q p) is odd
(1)
p
2
i
2

E(
npi
2
)
E(
qi
2
)

if (n i)(n q p) is even
in the real case. For the complex case:
t
i,j
= t
i,pi
=

n p i
q i

.
Proof. By our notations t
i,j
= f(n, q, p, i). By (3.7) for i = p j > 0 we have f(n, q, p, i) =
f(ni, q i, j, 0).g(n, p, j) by Lemma 3.2.4 and (3.5) both numbers on the right hand side are
known. If i = 0 then j = p or equivalently x = x

, then by denition f(n, q, p, i) = g(n, p, j).

Recalling the values g from Lemma 3.2.4 gives the assertion.
Now we can extract the key to the existence problem, and prove the main result of this thesis.
Theorem 3.2.16. If
3.2 Inversion formulas for 1

37
(i) p +q n in the complex Grassmannian.
(ii) p +q n and n p q is even in the real Grassmannian.
Then determinant of T in (3.4) is not zero. Moreover, R

is invertible and its inverse is given

by det(T)
1
R
1
.
Proof. By Lemma 3.2.15 and Proposition 3.2.11.
Remark 3.2.17. Matsui in [7] proves that 1
G(p,q)
, the Radon transform with inclusion inci-
dence relation, has an inverse in the following cases:
(i) p +q n in the complex Grassmannian,
(ii) p +q n and q p is even, in the real Grassmannian.
38 Radon transform on Grassmannians
Appendix A
Duality formulas and Euler-Poincare
index with compact support
Here we state some theorems which were used in the preceding chapters. The references for
this chapter are [1], [3], [5] and [9].
Let k be a eld and f : X Y be a continuous map of manifolds. Let F be a sheaf on X.
Recall that the proper direct image of F by f is dened by:
(f
!
F)(U) = s (f
1
(U); F) : f[
supp(s)
: supp(s) U is proper.
This functor is left exact and injective with respect to the family of c-soft sheaves (those sheaves,
for which (X; .) (K; .), for any compact K X, is surjective). Therefore f
!
has a right
derived functor Rf
!
: D
+
(k
X
) D
+
(k
Y
). Note that if we consider direct proper image by the
map a
X
: X pt, we get:
Ra
X!
(F) R
c
(X; F).
Proposition A.1. Let g : Y Z be another continuous map of locally compact spaces. Then,
there are canonical isomorphisms.
(g f)
!
g
!
f
!
and R(g f)
!
Rg
!
Rf
!
. (A.1)
In particular the diagram:
X
f

a
X

D
D
D
D
D
D
D
D
Y
a
Y
z
z
z
z
z
z
z
z
pt,
gives
R
c
(Y ; Rf
!
F) R
c
(X; F).
Remark A.2. The formula R(Y ; Rf

(F)) = R(X; F) might be considered as replacement

of Leray-Serre spectral sequence in the framework of derived categories. By the same token,
the formula R
c
(Y ; Rf
!
F) R
c
(X; F) might be seen as Leray-Serre sequence with compact
supports.
40 Duality formulas and Euler-Poincare index with compact support
Proposition A.3 (Fibers formula, [9] Proposition 1.11.4). For any F D
+
(k
X
), we have a
canonical isomorphism:
[Rf
!
(F)]
y
R
c
(f
1
(y); F[
f
1
(y)
).
Corollary A.4 (Cartesian square formula, [9] Proposition 1.11.6). Assume:
Y

Y
f

X
is a Cartesian square of locally compact spaces (i.e. Y

(x

, y) X

Y : g(x

) = f(x)).
Then, we have canonical isomorphisms:
f
!
g
1
h
1
f

!
and Rf
!
g
1
h
1
Rf

!
Theorem A.5 (Poincare-Verdier duality, [5] Theorem 3.1.5). Let f : X Y be a continuous
map of locally compact spaces that f
!
has nite cohomological dimension(i.e. there exists a
non-negative integer such that R
j
f
!
= 0 for j > r). Then there exist a functor of triangulated
categories f
!
: D
b
(k
Y
) D
b
(k
X
) and an isomorphism of bifunctors on D
b
(k
X
) D
b
(k
Y
) :
Hom
D
b
(k
Y
)
(Rf
!
(.), .) Hom
D
b
(k
X
)
(., f
!
(.)).
Namely, f
!
is a right adjoint to Rf
!
.
Denition A.6. (i) Let a
X
: X pt, and F D
b
(k
{pt}
). The dualizing complex on X,
denoted by
X
is:

X
= a
!
X
F D
b
(k
X
).
(ii) Assume X has a nite c-soft dimension and let F D
b
(k
X
). One sets:
D
X
F = RHom(F,
X
) , D

X
F = RHom(F, k
X
).
We call D
X
F, the dual of F. When there is no risk of confusion we might write D instead
of D
X
. Note that
X
= D
X
k
X
.
(iii) The orientation sheaf over X, Or
X
, with eld coecients is dened by:
U Hom(H
n
c
(U; k), k).
(iv) An orientation of the manifold M relative to k, the eld, is an isomorphism:
Or
X
k
X
.
If such an orientation exists, we will call the manifold orientable.
Proposition A.7 ([1] Proposition 3.5.1). Let X be an n-dimensional topological manifold with
boundary and Or
X
be the orientation sheaf on X. Then there is a canonical isomorphism:

X
Or
X
[+n],
in D
b
(k
X
).
41
Corollary A.8 (Classical Poincare duality, [1] p.70). Let k be a eld and X an n-dimensional
orientable manifold without boundary. We have the isomorphism:
H
ni
(X; k) Hom(H
i
c
(X; k), k)
Proof. I. By denition of orientable manifold and Proposition A.7, we have:
H
ni
(X; k
X
) H
ni
(X; Or
X
) = H
i
(X; Or
X
[+n]) H
i
(X;
X
).
II. For a manifold M, A D
b
(k
M
) and U M open, and each i Z, there exists a short exact
sequence (For more general case see Theorem 3.4.4 in [1]):
0 Ext(H
i+1
c
(U; A), k) H
i
(U; D
X
A) Hom(H
i
c
(U; A), k) 0.
III. Since a eld is an injective object of the category of k-vector spaces, Hom(., k) is an exact
functor. Therefore, for any A D
b
(k
M
), we have:
Ext(H
i+1
c
(M, A), k) = 0
IV. By denition and II, III it follows:
H
i
(X;
X
) H
i
(X; D
X
k
X
) Hom(H
i
c
(X; k
X
), k).
Composing IV and I yields the assertion.
By Classical Poincare duality it follows that:
Corollary A.9 ([3] Exercise 3.3.13). Let k be a eld and X be an orientable n-dimensional
topological manifold. Then for a locally constant k-vector space L on X we have:

c
(X, L) = (1)
n
(X, L).
In particular
c
(X, L) is invariant under homotopy equivalence up to a sign.
Example A.10. (i) Using Corollary A.9 we see that
c
(R
n
, R) = (1)
n
(R
n
; R). Since
(R
n
; R) is invariant under homotopy equivalence we have (R
n
, R) = (pt; R) = 1.
Since C
n
endows a real analytic structure of R
2n
, in particular we have
c
(C
n
, R) = 1.
(ii) One should be cautious that in the preceding corollary L is a locally constant sheaf. For
instance R
c
(R, R
[0,)
) = 0, and therefore
c
(R, R
[0,)
) = 0. Yet, R is homotopic to the
origin. Hence
c
in general is not invariant by homotopy, even up to a sign.
We state a proposition which is used in Chapter 3 for manifolds which are homotopy equivalent
to compact manifolds.
Proposition A.11 ([3] Corollary 2.5.5). Let F E B be a locally trivial bration such
that he base B and the ber F are homotopy equivalent to nite CW-complexes. Then the three
Euler characteristics (B), (F) and (E) are dened and:
(E) = (B)(F).
42 Duality formulas and Euler-Poincare index with compact support
Corollary A.12. In the situation of the preceding proposition, one has:

c
(E) =
c
(B)
c
(F).
where
c
(X) =
c
(X; k). We read
c
(X) the Euler Characteristic with compact support of X.
Proof. By Proposition A.11 and Corollary A.9. Note that the signs cancel out.
The following formula is used in the second chapter.
Proposition A.13 (Projection formula, [9] Proposition 1.13.4). Let f : X Y be a continuous
map between locally compact spaces. Then, for any F D
+
(k
X
) and any G D
+
(k
Y
), there
is a canonical isomorphism
G
L
Rf
!
F Rf
!
(f
1
G
L
F).
We end the appendix by mentioning the K unneth formula with compact support.
Corollary A.14 (K unneth formula with compact support,[3] Corollary 2.3.30). Let X
1
and X
2
be two topological spaces, p
i
: X
1
X
2
X
i
for i = 1, 2 be the two projections. Let F
i
D
b
(k
X
i
)
for i = 1, 2. Then:
R
c
(X
1
X
2
; p
1
1
F
1
L
p
1
2
F
2
) = R
c
(X
1
; F
1
)
L
R
c
(X
2
; F
2
).
Appendix B
Micro-support of a sheaf
In this section we introduce the notion of micro-support of a sheaf. Micro-support measures
how far a sheaf is from being locally constant. It is also a tool for realizing constructible sheaves.
For equivalent denitions see [5] p.221.
Denition B.1. Let X be a real analytic manifold, and denote by : T

X X. Let
F D
b
(k
X
) be a bounded complex. For a point p = (x
0
,
0
) T

X, the micro-support (or

characteristic variety) of complex F, denoted by SS(F) is the subset of cotangent bundle T

X
consisting of all the points p = (x
0
,
0
) such that the condition (C) is failed.
(C) There is an open neighborhood U of the point p such that for any point x
1
X and any
real smooth function f dened in a neighborhood of x
1
and satisfying f(x
1
) = 0 and df(x
1
) U
we have:
(R
{x:f(x)0}
F)
x
1
= 0.
Denition B.2. (i) Let (E, ) be a symplectic vector space and V a linear subspace of E.
We set:
V

= x E : (x, V ) = 0.
V is called isotropic (resp. Lagrangian, resp. involutive) if V V

(resp. V = V

, resp.
V V

).
(ii) Assume
1
is a conic subanalytic subset of T

X. One says that is isotropic (resp.

Lagrangian, resp. involutive), if at each regular point p , the tangent space of at
p, T
p
, has the corresponding property in T
p
T

X, with respect to the natural symplectic

form on T

X.
Following proposition yields some properties of micro-support.
Proposition B.3. (i) The micro-support SS(F) is a closed conic subset of T

X and SS(F)
T

X
X supp(F), where T

X
X is the zero section of the cotangent bundle of X;
(ii) SS(F) = SS(F[1]);
1
If is a subset, see [5] for the denition.
44 Micro-support of a sheaf
(iii) Let F
1
F
2
F
3
+1
be a distinguished triangle in D
b
(k
X
). Then SS(F
i
) SS(F
j
)
SS(F
k
) and (SS(F
i
)`SS(F
j
)) (SS(F
j
)`SS(F
i
)) SS(F
k
) for any permutation (i,j,k)
of (1,2,3);
(iv) SS(F)

j
SS(H
j
(F));
(v) Let M be a closed submanifold in X amd consider its conormal space in X given by:
T

M
X = (x, ) T

X; [
TxM
= 0.
Then SS(i
!
L) = T

M
X where i : M X is the inclusion and L is any non-zero locally constant
sheaf on M.
We end this appendix with a theorem which relates constructible sheaves to micro-support.
Theorem B.4. Let F D
b
(k
X
). Then the following conditions are equivalent:
(i) There exists a locally nite covering X =

iI
X
i
by subanalytic subsets such that for all
j Z, all i I, the sheaves H
j
(F)[
X
i
are locally constant.
(ii) SS(F) is contained in a closed conic subanalytic isotropic subset of T

X.
(iii) SS(F) is a closed conic subanalytic Lagrangian subset of T

X.
Bibliography
[1] Banagl, M., Topological Invariants of Stratied Spaces, Springer-Verlag, 2007.
[2] Bredon G., Topology and Geometry, Springer-Verlag, New York, 1993.
[3] Dimca A., Sheaves in topology, Springer-Verlag, 2003.
[4] Fulton W., Young Tableaux, Cambridge University Press, 1997.
[5] Kashiwara, M. and Schapira, P., Sheaves on manifolds, Springer-Verlag, Berlin, 1990.
[6] Marastoni C., Grassmann Duality for D-Modules, Annales Scientiques de lEcole Normale
Superieure, Volume 31, Number 4, 8 July 1998 , pp. 459-491.
[7] Matsui, Y., Radon Transforms of constructible functions on Grassmann manifolds, Publ.
Res. Inst. Math. Sci. 42 (2006), 551580.
[8] Schapira, P., Tomography of constructible functions, Applied Algebra, Algebraic Algo-
rithms and Error-Correcting Codes (G. Cohen, M. Giusti and T. Mora, eds.), Lecture
Notes in Comput. Sci. 948, Springer Berlin, 1995, pp. 427435.
[9] Schneiders, J. P., Introduction to Characteristic Classes and Index Theory, Textos de
Matematica, Volume 13, Faculdade de Ciencias da Universidade de Lisboa, 2000.
[10] Weibel C.A., An introduction to Homological Algebra, Cambridge University Press, 1995.