TWO LANGUAGES OF SCHUBERT CALCULUS: GRASSMANNIANS

AND FLAG MANIFOLDS

FLORIAN BLOCK AND NICOLAS BRAY
ABSTRACT. We focus on the question why ag manifolds generalize the Grass-
mannian in terms of Schubert calculus. On our way we discuss Schubert cells
and Schubert varieties in these two languages. The latter ones form a basis for
the cohomology rings, respectively. This leads to structure constants. We intro-
duce Schubert polynomials and Schur polynomials and relate these to cohomol-
ogy. Finally, we present the translation between Grassmannianand ag manifold
Schubert calculus.
1. INTRODUCTION
In this paper, we will give a brief overview of the theory of Schubert calculus in
two contexts: On the Grassmannian Gr
k,n
(i.e. the space of k-planes in C
n
) and on
the ag manifold F l
n
. In fact, both of these are examples of partial ag manifolds,
the space F l
(d
1
,d
2
,...,d
m
)
n
consisting of partial ags F
1
F
2
F
m
in C
n
where F
i
is
a subvectorspace of C
n
having dimension d
i
. Schubert calculus can be formulated
in this more general setting in a fashion which encompasses both the Grassmannian
and ag theories. However, for most purposes this generality is not necessary and
to some extent it obscures the theory in the Grassmannian case. We will begin by
dening the basic geometric objects. We take the denition of the Grassmannian
for granted, but will briey review the denition of the ag manifold.
2. GRASSMANNIANS AND FLAG MANIFOLDS
A (complete) ag F

in C
n
is a strictly ascending chain 0 = F
0
F
1

F
n
= C
n
of subspaces of C
n
, where F
i
is a linear subspace of dimension i. Given
a ag F

, if another ag G

is choosen randomly (we make this precise later) then

we expect the intersections of the subspaces in F

and those in G

to be as small as
possible, i.e. dim(F
i
G
j
) should be max(i + j n, 0). In such a case, F

and G

are said to be in transverse position or simply transverse. In particular, for any ag

F

, if we dene the ag F

by setting F

i
to be the space perpendicular to F
ni
,
then F

and F

will be transverse.
The set of all ags can be thought of as the topological space of matrices in
GL
n
(C) modulo the relation where two matrices A and B are equivalent if, for all
i between 1 and n, the rst i columns of A span the same subpace as the rst i
columns of B. We dene the subspace X

(F

) to be the set of all ags transverse

to F

. These sets are open in the previous topology (intuitively, if we have a ag

Date: May 11, 2005.
1
2 FLORIAN BLOCK AND NICOLAS BRAY
G

transverse to F

and perturb it slightly then its intersection dimensions with F

,
these are F
i
G
j
, will not increase) and in fact give the set of all ags the structure
of a differentiable manifold which we will call F l
n
.
There is a natural transisitive action of GL
n
(C) on F l
n
given by A F

= 0
A F
1
A F
2
A X which is smooth in the above topology.
Lemma 1. The sets X

(F

) form an atlas on F l
n
of dimension

n
2

.
Proof. The sets X

(F

) certainly cover F l
n
since F

(F

). To prove the iso-

morphism statement, rst let E
i
be the subspace of C
n
spanned by the rst i vec-
tors of the standard basis of C
n
and consider the standard ag E

= 0 = E
0

E
1
E
2
E
n
= C
n
. We can write down an isomorphism between C
(
n
2
)
andX

(E

) explicitly by
(, , . . ., )

0 0 1
.
.
. 0 1
0 1
.
.
.
1

.
For any ags F

and G

, we can choose elements A

F
, A
G
GL
n
(C) such that
A
F
F

=A
G
G

=E

. The desired isomorphisms A

F
: C
(
n
2
)
X
o

(F

) and A
G
:
C
(
n
2
)
X
o

(G

) are C

on the preimage of the overlap of two open sets.

Flag manifolds and Grassmannians are quite different objects but note that for
any F l
n
and any 1 k n, there is the forgetful map
k
: F l
n
Gr
k,n
which maps
a ag F

to F
k
. This is clearly a smooth, surjective map and will later be quite
important for relating the Schubert calculi on these different spaces.
The Pl ucker embeddings. While the manifold structures of Gr
k,n
and F l
n
are
important, in order to do Schubert calculus, we actually must consider them as
algebraic varieties by embedding them in a higher dimensional projective space.
This is done by the Pl ucker embeddings. In the Grassmannian case, note that if
V is a k-dimensional subspace of C
n
, then its image
k
V in
k
C
n
will be a line.
In particular, if V is spanned by v
1
, . . ., v
k
, then its image will consist of the line
spanned by v
1
v
2
. . . v
k
. This suggests identifying points in Gr
k,n
with points in
P(
k
C
n
).
More specically, let M be a k n matrix whose rows span V. Since the rank of
M is k, there is at least one kk minor of M that is non-zero. For any matrix M and
any strictly ascending sequence I = (i
1
, i
2
, . . . , i
k
) with 1 i
1
<i
k
n, let M
I
be the
minor det(M
p,i
q
)
1p,qk
. If M
/
is another matrix whose rows span V then there is
an element A GL
n
(C) such that M = AM
/
which will imply that M
I
= det(A)M
/
I
for all I. Together these facts suggest the following denition::
Denition 2. For the Grassmannian Gr
k,n
, consider the projective space P(
k
C
n
)
and denote e
i
1
e
i
k
by X
i
1
,...,i
k
where 1 i
1
< < i
k
n and the e
i
are the
vectors in the standard basis of C
n
. For every V Gr
k,n
, choose a matrix M(V)
TWO LANGUAGES OF SCHUBERT CALCULUS: GRASSMANNIANS AND FLAG MANIFOLDS 3
1 1 3
4 2
3
4
FIGURE 1. Left: Young diagram associated to the partition =
(3, 2, 1, 1). Right: A possible numbering of this Young diagram.
whose rows span V and dene a map from Gr
k,n
to P(
k
C
n
) by
V

I=(i
1
,...,i
k
)
M(V)
I
X
i
1
,...,i
k
where the sum is over strictly ascending sequences. By the above, the image V
under this map is actually independent of the choice of M(V) and so we have a
well-dened map from Gr
k,n
to P(
k
C
n
) which is called the Pl ucker embedding
(that this map is, in fact, an embedding is slightly harder to show, see e.g. ??).
Now consider F l
n
as the subset of Gr
1,n
Gr
2,n
Gr
n1,n
consisting of
all (F
1
, F
2
, . . . , F
n1
) such that the F
i
form a ag. This can then be embedded into

n1
k=1
P(
k
C
n
) by the Pl ucker embeddings for the Grassmannian and the condition
of forming a ag cuts out a subvariety known as the incidence variety.
3. SCHUBERT CELLS AND SCHUBERT VARIETIES
3.1. Grassmannians. As a tool to deal with certain subsets of the Grassmannian
we need the notion of Young diagrams and later on also of numberings of Young
diagrams. For a more complete introduction to these objects see [3].
Denition 3. Each weakly decreasing sequence = (
1

2

k
) of non-
negative integers can be associated to a Young diagramconsisting of k left-adjusted
rows of boxes, row i having
i
boxes. A numbering of the Young diagram is an
assignment of the integers from 1 to k to the boxes such that each row is weakly
increasing and each column is strictly increasing.
For example (with a little bit of abuse of notation) if = (3, 2, 1, 1), the Young
diagram is shown on the left of Figure 1 while a numbering of is shown on the
right.
Denition 4. Let F

be a ag in C
n
and a Young diagram with k rows each
containing at most n k blocks. Then the (Grassmannian) Schubert cell
o

(F

)
is the set of V Gr
k,n
such that dim(V F
j
) = i when j is between n +i
i
and
n+i
i+1
. Equivalently,

(F

) =V Gr
k,n
: dim(V F
n+i
i
) = i, 1 i k.
The Grassmannian Schubert variety

(F

) is the closure of
o

(F

).
4 FLORIAN BLOCK AND NICOLAS BRAY
Proposition 5. The Grassmannian Schubert variety

(F

) is exactly V Gr
k,n
:
dim(V F
n+i
i
) i, 1 i k.
Proof. To see this, suppose V is an element of

(F

) which is not contained

in
o

(F

). This means that some of its intersections with members of F

have
higher dimension than allowed. One can then imagine perturbingV to V
/
which has
the appropriate dimensions. Clearly this is possible to achieve while only slightly
changing V and so one can nd a sequence of elements of
o

(F

) converging to
V.
Note that if and are two Young diagrams such that (i.e.,
i

i
) then
the Schubert conditions for a point to lie in

(F

) are a subset of those for

(F

)
and so we will have

(F

(F

). In fact, a stronger condition holds:

Lemma 6. The Schubert variety

(F

) is the disjoint union of the Schubert cells

(F

) for all .
Proof. That
o

(F

(F

) follows easily from Proposition 5. Conversely, if

V satises dim(V F
n+i
i
) i then one can imagine increasing
i
until the di-
mension in fact equals i. Doing this for every i would yield a such that
V
o

(F

). Finally, the fact that the union is disjoint follows immediately from
the denition.
3.2. Flag Manifolds. While most pairs of ags are transverse and therefore inter-
sect in a trivial fashion, in general two ags may have a more complicated inter-
section structure. In general one would want to consider the dimension table of F

and G

, D = D(F

, G

) where D
i j
= dim(F
i
G
j
). This table has a very restricted
structure, however, and this allows us to encode it in a very useful way. The key
features of the matrix are the positions where the dimension increases and by con-
sidering the what this set of positions looks like, it can be seen without too much
difculty that we can in fact encode this information in a permutation wS
n
. More
specically, for any two ags F

and G

there exists a unique permutation w S

n
such that
dim(F
i
G
j
) = rank(w[i, j]) for 1 i, j n.
Here, w[i, j] denotes the i j principal submatrix of the n n permutation matrix
w which is dened as follows: The entry (i, j) of w is one if w(i) = n +1 j and
zero otherwise (note that this is a ipped version of the standard denition of the
matrix of a permutation). Flags F

and G

are then said to be in relative position

w. In particular, F

and G

being in transverse position corresponds to them having

relative position id = (1, 2, . . ., n). As an example, (1, 4, 3, 2) has matrix

0 0 0 1
1 0 0 0
0 1 0 0
0 0 1 0

and ags F

= 0 span(e
1
) span(e
1
, e
2
) span(e
1
, e
2
, e
3
) C
4
and G

=
0 span(e
2
) span(e
2
, e
3
) span(e
2
, e
3
, e
4
) C
4
are in relative position(1, 4, 3, 2).
TWO LANGUAGES OF SCHUBERT CALCULUS: GRASSMANNIANS AND FLAG MANIFOLDS 5
Denition 7. The Schubert cell X
o
w
(F

) in F l
n
is the set of all ags G

which are
in relative position w with respect to F

. Equivalently,
X
o
w
(F

) =G

F l
n
: dim(F
i
G
j
) = rank(w[i, j]) for 1 i, j n.
The Schubert variety X
w
(F

) is the closure of X
o
w
(F

) in the ag manifold F l
n
.
In the Grassmannian case, we saw that the partial ordering on Young diagrams
corresponded to the inclusion ordering on Schubert cells. For ag manifolds, a
similar condition holds except now the Schubert cells are indexed by permuta-
tions and the correct partial ordering of the indices is the Bruhat order. For a
permutation w, the length l(w) of w is dened to be the number of inversions in
w, i.e. the number of i < j such that w(i) > w( j). Thus the identity permutation
has length zero and (n, n 1, . . ., 1) has length

n
2

. This gives a preorder on S

n
which can be strengthened to a partial order by dening v w if there is a se-
quence of permutations beginning at v and ending at w such that each is the result
of multiplying the last by a length-increasing transposition. Note, for example, that
(1, 2, . . ., n) w (n, n1, . . ., 1) for all w S
n
.
Lemma 8. If v w in the Bruhat order, then X
o
v
(F

) X
o
w
(F

) (and so X
v
(F

)
X
w
(F

)) for all F

. Morever, we in fact have that X

w
(F

) =
S
vw
X
o
v
(F

) where the
union is disjoint.
Proof. For essentially the same reasons as in the Grassmannian case, taking the
closure of X
o
w
(F

) corresponds to simply changing the equality in the denition

to dim(F
i
G
j
) >= rank(w[i, j]). It turns out that v w in the Bruhat order is
equivalent to rank(v[i, j]) rank(w[i, j]) which can be veried for considering only
the case where v is minimally greater (i.e. is the result of multiplying w with a
length increasing permutation as above) than w (see ?? where this is in fact taken
as the denition of the Bruhat order, though one must adjust for his dening the
Schubert cells in a order-reversing way than we have here).
As the name suggests, these Schubert varieties are in fact subvarieties of their
ambient space. The dimension of X
w
(F

) is given by

n
2

l(w). Note that with

0
= (n k)
k
= (n k, . . . , n k and w
0
= (n, n1, . . ., 1), we have that [
0
[ =
k(n k) and l(w
0
) =

n
2

and so

0
(F

) and X
w
0
(F

) both have dimension zero.

In particular, if the k-plane H is in the Schubert variety

0
(F

) then we have that

dim(HF
k
) k which implies that H = F
k
. Thus the Schubert variety
(nk)
k(F

)
consists of a single point. Similarly in the ag manifold, we have that if G

is in
X
w
o
(F

) then in particular dim(G

i
F
i
) =i and so G

=F

. Hence, also [X
w
o
(F

)[ =
1.
4. THE CLASS OF A SUBVARIETY
We nowattempt to give a brief description of howone may assign a cohomology
class to a subvariety X of a nonsingular algebraic variety V. If X is also nonsingular
then it is in fact an orientable manifold and thus has a fundamental class [X] which
then can be considered as an element of H

(V) by sending [X] to [V] i

([X]).
The difculty arises when the singular locus of X, i.e. the set of points where X is
6 FLORIAN BLOCK AND NICOLAS BRAY
singular, is non-empty. However even in this case, the nonsingular locus of X will
have less than full complex dimension and then one can essentially ignore it on the
level of homology. Intuitively, one can envision this as giving X the structure of
a simplicial complex in a way such that the singular locus of X forms a subcom-
plex of lower complex dimension and then one can dene the fundamental class
of X using only the top dimensional simplices in the complex. A more rigorous
denition uses a type of Borel-Moore homology.
This now allows us to assign a cohomology class (called a Schubert cycle) for
every Schubert variety. As it stands, this cycle will depend on the varietys index
(either a Young diagram or a permutation) and the reference ag. However, it turns
out that the class of a Schubert variety is independent of the reference ag. We will
prove this for the ag manifold and the same proof will work (mutatus mutandis)
for the Grassmannian as well.
Lemma 9. For any w and any two ags F

and G

, [X
w
(F

)] = [X
w
(G

)].
Proof. For any A GL
n
(C), dene the map i
A
: X
w
(F

) F l
n
to be G

A G

and so in particular i
id
is simply the inclusion of X
w
(F

) into F l
n
. If B is such
that F

= B G

then we will have by denition that i

B
([X
w
(F

)]) = [X
w
(G

)] since
i
B
(X
w
(F

)) = X
w
(G

). However since GL
n
(C) is connected, there exists a path
from id GL
n
(C) to B which will then induce a homotopy between i
id
and i
B
.
Since homotopic maps induce identical maps on homology, we will then have that
[X
w
(G

)] = i
B
([X
w
(F

)]) = i
id
([X
w
(F

)]) = [X
w
(F

)].
Thus when referring to a Schubert cycle, we can (and will) omit mention of the
reference ag used to dene it.
5. THE BASIS THEOREM AND STRUCTURE CONSTANTS
The key fact of Schubert calculus is that the Schubert cycles actually additively
generate the cohomology ring. To prove this we will need a technical lemma con-
cerning the Borel-Moore homology of algebraic varieties (for a proof, see Appen-
dix B of [3]).
Lemma 10. Given a ltration 0 = V
m
V
m1
V
0
= V of a variety V by
algebraic subsets V
i
such that V
i
V
i+1
is a disjoint union of algebraic varieties
U
i, j
all of which are isomorphic to some afne space over C, the cohomology ring
H

(V) is additively generated by the classes of the closures of the U

i, j
.
Theorem 11 (Schubert Basis Theorem). If X is either the Grassmannian or the
ag manifold, then the Schubert cycles in X additively generate its cohomology
ring.
Proof. We will prove the statement in the Grassmannian case, at which point it
follows for the ag manifold with a virtually identical argument. Letting F

be any
ag, we dene a ltration of Gr
k,n
by letting V
i
be the union of

(F

) over all
with [[ i. We clearly have that V
i+1
V
i
and, since
0
k (F

) = Gr
k,n
, V
0
= Gr
k,n
and so this is in fact a ltration. To see that this ltration satisies the condition
of the above lemma, we note that since

(F

) =
S

(F

) so that we also
TWO LANGUAGES OF SCHUBERT CALCULUS: GRASSMANNIANS AND FLAG MANIFOLDS 7
have V
i
=
S
[[i

(F

) and thus V
i
V
i+1
=
S
[[=i

(F

). Since every Schubert

cell is isomorphic to an afne space, the above lemma implies that H

(Gr
k,n
) is
generated by the classes of the closures of the Schubert cells
o

(F

), i.e. by the
Schubert cycles.
Note that if we have an additive basis X
i
for the cohomology ring of a space
then for any X
i
and X
j
we will have that X
i
X
j
=
k
c
k
i j
X
k
for some constants c
k
i j
.
These constants will be called the structure constants of the cohomology ring.
6. SCHUBERT POLYNOMIALS
To get a more explicit way to do Schubert calculus in the the ag manifold,
we want to understand the cohomology ring of this space. It will turn out that
the Schubert varieties in F l
n
are in one to one correspondence with the Schubert
polynomials which we will dene in the section. Using Schur polynomials (which
will be dened later) we can state the embedding H

(Gr
k,n
) H

(F l
n
) explicitly,
making the statement Flag manifolds are a generalization of the Grassmannian
precise.
For i 1, . . ., n1 let s
i
S
n
denote the transpositioninterchanging i and i +1.
The s
i
s generate the symmetric group S
n
. Further, S
n
acts on R
n
=Z[x
1
, . . . , x
n
] by
permuting the variables. For a polynomial f R
n
the polynomial f s
i
( f ) is anti-
symmetric in x
i
and x
i+1
meaning that if the variables x
i
and x
i+1
are interchanged
the sign of f s
i
( f ) switches. This implies that x
i
x
i+1
divides f s
i
( f ). So the
following is well dened.
Denition 12. Dene the Z-linear differential operator
i
on the polynomial ring
R
n
by

i
( f ) =
f s
i
( f )
x
i
x
i+1
for 1 i n1.
This operator has various properties. For example,
i
( f ) = 0 if f is symmetric
in x
i
and x
i+1
. Furthermore,
i
( f ) is symmetric in x
i
and x
i+1
. Hence
i
(
i
( f )) =0
for every f R
n
.
Denition 13. Given a permutation w=s
a
1
s
a
2
s
a
p
where p =l(w), then
a
1

a
2

a
p
is independent of the representation choosen (see Sections 10.3 and 10.4 in [3].
This allows us to dene the Schubert polynomial S
w
for every permutation w S
n
by
S
w
=
w
1
w
0
(x
n1
1
x
n2
2
x
n1
).
Notice that the degree of S
w
is l(w) since each application of
i
reduces the degree
by 1.
Computation of Schubert polynomials is not as bad as the denition might sug-
gest. Before we compute an example we collect rules which might help for the
computation.
Lemma 14. a) For any i,
i
(S
w
) = S
ws
i
if w(i) > w(i +1), and
i
(S
w
) = 0 oth-
erwise.
8 FLORIAN BLOCK AND NICOLAS BRAY
b) We have S
w
0
= x
n1
1
x
n2
2
x
2
n2
x
n1
.
c) For each i we have S
s
i
= x
1
+x
2
+ +x
i
.
Proof. Statement a) is Proposition 10.5a [3], b) follows from the denition. Using
a), we conclude
i
(S
s
i
) = 1 and
j
(S
s
i
) = 0 if i ,= j. This shows c).
The lemma gives rise to an algorithm to compute Schubert polynomials. We
illustrate it by example.
Example 15. Given the permutation w = (35142), we want to compute S
35142
.
We start out with S
54321
= x
4
1
x
3
2
x
2
3
x
4
. The idea is to iteratively interchange adjacent
numbers if they do not agree with the order in w. For example, we could start by
interchanging position 4 and 5. (Interchanging 2 and 3 would also work.) We get
by Lemma 14 a)
S
54312
=
4
(S
54321
) = x
4
1
x
3
2
x
2
3
.
Next, we interchange position 2 and 3 (now the only possibility) and end up with
S
53412
=
2
(S
54312
) = x
4
1
x
2
2
x
2
3
.
Now interchange position 3 and 4 (or 1 and 2):
S
53142
=
3
(S
53412
) = x
4
1
x
2
2
x
3
+x
4
1
x
2
2
x
4
.
Interchange position 1 and 2:
S
35142
=
1
(S
53142
) = x
3
1
x
2
2
x
3
+x
3
1
x
2
2
x
4
+x
2
1
x
3
2
x
3
+x
2
1
x
3
2
x
4
.
Since l(35142) = 6 we see that S
35142
indeed has the right degree.
We note that S
w
: w S
n
is a Z-basis for x
i
1
1
x
i
n1
n1
: i
j
n j which is
transversal to S, the subring of symmetric functions in R
n
. Hence the Schubert
polynomials are explicit polynomial representatives of an integral basis of the quo-
tient H
n
=R
n
/S. It will turn out that H
n

=H

(F l
n
) as rings where the isomorphism
is given by S
w
[X
w
].
7. SCHUR POLYNOMIALS
Like we used Schubert polynomials to write down Schubert varieties in the ag
manifold we now dene Schur polynomials to establish an analog in the Grass-
mannian. In particular, we are interested what the inclusion H

(Gr
k,n
) H

(F l
n
)
looks like on the level of polynomials.
Given a partition = (
1

2

k
) of lenght k. We will always assume

1
n k. With this partition we associate a Young diagram with at most k rows
and by abuse of notation we also call it (compare with Denition 3). We dene a
symmetric polynomial s

(x
1
, . . . , x
k
) as follows. For a numbering T of the Young
diagram we dene
x
T
=
k

i=1
(x
i
)
number of times i occurs in T
.
The Schur polynomial s

(x
1
, . . . , x
k
) is then given by
s

(x
1
, . . . , x
k
) =

x
T
TWO LANGUAGES OF SCHUBERT CALCULUS: GRASSMANNIANS AND FLAG MANIFOLDS 9
where the sumis over all possible numberings of the Young diagram with integers
from 1 to k. It is a fact that these polynomials are symmetric and form a basis of
the ring of symmetric polynomials Z[x
1
, . . . , x
k
]
S
k
. Also, I
n,k
=s

:
1
nk is
an ideal. Let A
n,k
be the quotient ring Z[x
1
, . . . , x
k
]
S
k
/I
n,k
.
Although the denitions of Schur polynomials and Schubert polynomials are
quite different, it turns out that the former are special instances of the latter. Given
a partition = (
1

2

k
) dene a permutation w S
n
by
w = w() = (
k
+1,
k1
+2, . . .,
1
+k, [remaining elements in increasing order]).
Since we assumed
1
n k this is well dened. For example, if = (3, 2, 1, 1)
then w() = (2, 3, 5, 7, 1, 4, 6). Note that in general the permutation w() has ex-
actly one descent at k, i.e. w(i) < w(i +1) if and only if i = k. On the other
hand, every permutation w with exactly one descent at k denes a partition by
(k+1i) =w(i) i and these two transformations are inverses of each other. We
will see later that S
w(
) = s

for every permutation of length k with

1
n k.
So we call a Schubert variety X
w
Grassmannian if w = w() for some permutation
with
1
nk.
8. TRANSLATION BETWEEN GRASSMANNIAN AND FLAG MANIFOLD
To understand the injection H

(Gr
k,n
) H

(F l
n
) on the level of polynomials
we need to establish a connection between these rings and the quotient ring A
n,k
of
Schur polynomials and the quotient ring H
n
of Schubert polynomials, respectively.
We begin with the latter.
On F l
n
we have a ltration of the trivial bundle E
F l
n
given by
0 =U
0
U
1
U
n
= E
F l
n
.
Here, the ber of U
i
over the point E

is simply given by E
i
. By taking quotients
we end up with n line bundles L
i
= U
i
/U
i1
with base space F l
n
. Our desired
isomorphism is given in terms of Chern classes. For the denition and details
about Chern classes see Chapter 14 in [5].
Proposition 16. The rst Chern classes c
1
(L
i
) H
2
(F l
n
) of these line bundles
generate the cohomology ring of F l
n
. More general, the map H
n
H

(F l
n
) given
by x
i
c
1
(L
i
) is an isomorphismof rings. Furthermore, this map induces a one to
one correspondence between Schubert polynomials and Schubert varieties, namely
S
w
[X
w
].
Proof. Let e
i
denote the ith symmetric polynomial in C[x
1
, . . ., x
n
], so e
i
is the
sum over all monomials of degree i. By Proposition 10.3 in [3] the Chern classes
c
1
(L
i
) generate H

(F l
n
), subject to the relations e
i
(c
1
(L
1
), . . . , c
1
(L
n
)). Since the
ith symmetric polynomials e
i
generate the ideal S of symmetric polynomials in
C[x
1
, . . . , x
n
] the map is an isomorphism.
In [1] it is shown that [X
w
(F

)] =
w
1
w
0
[F

] where we dene
u
[X
v
] = [X
vu
1]
for arbitrary permutations u, v S
n
. By the denition of the Schubert polynomials it
sufces to check

(x
n1
1
x
n2
2
x
n1
) = [F

]. This can be veried by an explicit

calculation.
10 FLORIAN BLOCK AND NICOLAS BRAY
Quite similiar to the isomorphism between Schubert polynomials and Schubert
varieties in the ag manifold we can write down the corresponding isomorphism
for the Grassmannian. Let d
1
, . . . , d
k
denote the Chern classes of the tautolog-
ical k-plane bundle of the Grassmannian Gr
k,n
. Then, by [6], the map A
n,k

H

(Gr
k,n
), x
i
d
i
is an isomorphism of rings. We can state the even stronger re-
sult which can be found in [3].
Proposition 17. The isomorphism between the quotient ring A
n,k
of Schur poly-
nomials and the cohomology ring H

(Gr
k,n
) induces a one to one correspondence
between Schur polynomials and Schubert varieties in the Grassmannian given by
s

].
Now we are in a good position to answer the question what we mean by the
statement
Flag manifolds are a generalization of Grassmannian manifolds.
Informally the answer is that every question in Schubert calculus stated in terms
of Grassmannians can be translated into a question concerning ag manifolds. We
start out on the level of polynomials.
Proposition 18. Let be a permutation of length k with
1
nk. Then
S
w()
= s

.
We prove this very nontrivial result using the followingtwo facts for those proofs
we refer to Lemma 10.12 in [3] and [4], respectively. The proof of the rst is
elementary. The proof of the second needs some amout of work. Originally, the
expression in Lemma 20 was the denition of the Schur polynomials.
Lemma 19. For w
0
= (k k 1. . . 2 1) S
k
, it holds that

w
0
=
1

wS
n
sgn(w) w,
where =
i<j
(x
i
x
j
) is the Vandermonde determinant.
Lemma 20. (Jacobi-Trudi formula)
For a partition , s

(x
1
, . . ., x
k
) =
det[x

i
+ki
j
]
1i, jk
det[x
ki
j
]
1i, jk
. Note, that the denominator is
equal to the the above.
Proof. (of Proposition 18) Let u S
n
with u(i +1) < u(i) for i < k and u(i +
1) > u(i) if i k. Let
a,b

i
=
i
if a > b and the identity operator otherwise.
A decomposition of u with minimal length is
u =u
0
[
n,u(1)
s
1
] [
n1,u(2)
s
2
] [
nk+1,u(k)
s
(
k)] s
k+1
s
k+2
s
n1

[
n1,u(1)
s
1
] [
n2,u(2)
s
2
] [
nk,u(k)
s
(
k)] s
k+1
s
k+2
s
n2
. . .
[
1,u(1)
s
1
].
TWO LANGUAGES OF SCHUBERT CALCULUS: GRASSMANNIANS AND FLAG MANIFOLDS 11
A simple example should clear this weird representation up. Applying rule a) in
Lemma 14 leads to S
u
= x
u(1)1
1
x
(u(2)1)
2
x
u(k)1
k
.
Let u = (k k 1. . . 2 1) S
n
) and let w
/
= w u where w = w(). By the rst
part of the proof we then have
S
w
/ = x
w(k)1
1
x
(w(k2)1)
2
x
w(1)1
k
= x

1
+r1
1
x

2
+r2
2
x

k
k
.
Again, by part a) of Lemma 14, S
w
=
u
(S
w
/ ). By the Leibniz formula for de-
terminants,
wS
k
sgn(w) wS
w
/ = det[x

i
+ki
j
]
1i, jk
. Using Lemma 19 and 20 we
conclude
S
w
=
det[x

i
+ki
j
]
1i, jk

= s

(x
1
, . . . , x
k
).

Finally, we can summerize the relation between Schubert calculus in the Grass-
mannian and in the ag manifold. The former is just a special case of the latter.
The translation is done by the canonical map

on cohomology.
Theorem 21. Let

: H

(Gr
k,n
) H

(F l
n
) be the map on cohomology induced
by the map : F l
n
Gr
k,n
, F

F
k
. Then

is given by

([

]) = [X
w()
].
Equivalently,

(s

) =S
w()
. On the level of polynomials,

is the identity map.

Proof. The equality

(s

) =S
w()
is precisely Proposition 10.9 in [3]. The equiv-
alence follows from Propositions 16 and 17. The last statement follows from
Proposition 18.
In order to appy Proposition 10.9 in [3] we should note that our Schubert vari-
eties X
w
are called dual Schubert varieties in [3] and are denoted by
w
.
Caution 22. Although

: H

(Gr
k,n
) H

(F l
n
) acts as identity on polynomials,
sending Schur polynomials, which are symmetric, into a ring which is a quotient
by the ideal of symmetric polynomials, this map is still injective. Note that the
Schur polynomial s

is a polynomial in k variables and symmetric in those. But

considered as a polynomial in Z[x
1
, . . . , x
n
] it is very far from being symmetric and
hence is not sent to 0 by

.
REFERENCES
[1] I. N. Bernstein, I. M. Gelfand, and S. I. Gelfand, Schubert cells and cohomology of the spaces
G/P, Russian Mathematical Surveys, 28 (1973): 1-26.
[2] Sarah Billey and Ravil Vakil, Intersections of Schubert Varieties and other Permutation Array
Schemes, arXiv: math.AG/0502468.
[3] William Fulton, Young Tableux, London Mathematical Society Student Texts 35, Cambridge,
1997.
[4] I. G. McDonald, Symmetric Functions and Hall Polynomials, Clarendon Press, Oxford, 1979.
[5] John W. Milnor and James D. Stasheff, Characteristic Classes, Annals of Mathematics Studies,
no. 76, Princeton, 1974.
[6] Frank Sottile, Pieris formula for ag manifolds and Schubert polynomials, Annales de
lInstitut Fourier, 46 (1996): 89-110.
12 FLORIAN BLOCK AND NICOLAS BRAY
[7] L. Manivel, Symmetric Functions, Schubert Polynomials and Degeneracy Loci, AMS, 2001.
E-mail address: [florian, bray]@math.berkeley.edu