See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/2107493

Multiplicative rule of Schubert classes

Article in Inventiones mathematicae · June 2003

DOI: 10.1007/s00222-004-0394-z

CITATIONS READS
50 61

1 author:

Haibao Duan

79 PUBLICATIONS 475 CITATIONS

SEE PROFILE

All content following this page was uploaded by Haibao Duan on 28 September 2015.

The user has requested enhancement of the downloaded file.

 Multiplicative rule of Schubert classes
arXiv:math/0306227v2 [math.AG] 1 Sep 2003

Haibao Duan
Institute of Mathematics, Chinese Academy of Sciences
Beijing 100080, [email protected]

Abstract
Let G be a compact connected Lie group and H, the centralizer of
a one-parameter subgroup in G. Combining the ideas of Bott-Samelson
resolutions of Schubert varieties and the enumerative formula on a twisted
product of 2 spheres obtained in [Du2 ], we obtain an explicit formula for
multiplying Schubert classes in the flag manifold GH.
2000 Mathematical Subject Classification: 14N15 (14M10).
Key words and phrases: Schubert varieties, interesection multiplicities,
Cartan numbers

1 Introduction
Let G be a compact connected Lie group and H, the centralizer of a one-parameter
subgroup in G. The Weyl of G (resp. of H) is denoted by W (resp. W ′ ). The
set W/W ′ of left cosets of W ′ in W can be identified with the subset of W :
W = {w ∈ W | l(w1 ) ≥ l(w) for all w1 ∈ wW ′},
where l : W → Z is the length function relative to a fixed maximal torus T in G.
It is known from Bruhat-Chevalley that the flag manifold G/H = {gH | g ∈
G} admits a canonical decomposition into cells, indexed by elements of W ,
G/H = ∪ Xw (H), dim Xw = 2l(w),
w∈W
with each cell Xw (H) the closure of an algebraic affine space, known as a Schubert
variety in G/H [BGG]. Since only even dimensional cells are involved, the set
of fundamental classes [Xw (H)] ∈ H2l(w) (G/H), w ∈ W , form an additive basis
of the homology H∗ (G/H). The cocycle class Pw (H) ∈ H 2l(w) (G/H), w ∈ W ,
defined by the Kronecker pairing as hPw (H), [Xu (H)]i = δw,u , w, u ∈ W , is called
the Schubert class corresponding to w. Clearly one has
Basis Theorem. The set of Schubert classes {Pw (H) | w ∈ W } constitutes
an additive basis for the cohomology H ∗ (G/H).

1
 One immediate consequence is that the product of two arbitrary Schubert
classes can be expressed in terms of Schubert classes. Precisely, given u, v ∈ W ,
one has the expression P
Pu (H) · Pv (H) = aw w
u,v Pw (H), au,v ∈ Z
l(w)=l(u)+l(v),w∈W
in H ∗ (G/H). Since the Chow ring A∗ (G/H) is canonically isomorphic to the inte-
gral cohomology H ∗(G/H), the following Problem is of fundamental importance
in the intersection theory of G/H.
Problem. Find the number aw u,v for given w, u, v ∈ W , l(w) = l(u) + l(v).

If G is the unitary group U(n) of rank n and H = U(k) × U(n − k), the
flag manifold G/H is the Grassmannian Gn,k of k-planes through the origin in
Cn . In this case, a combinatorial description for aw u,v is given by the Littlewood-
Richardson rule, one cornerstone of the Schubert calculus for Gn,k [S]. It was first
stated by Littlewood and Richardson in 1934 [LR]. Complete proofs appeared
only in the 1970s (see “Note and references” in [M, p.148]).
Another special case is when H = T (a maximal torus in G) and if either
l(u) = 1 or l(v) = 1. The number aw u,v is seen as certain Cartan number of
G from the Chevalley formula. Chevalley announced the formula at the end of
his address at the 1958 ICM in Edinburgh [Ch1 ], while a proof was given in his
famous manuscript [Ch2 ]. Although [Ch2 ] remained unpublished until 1994, this
formula became part of the official literature after the publications of [BGG] by
Bernstein et al in 1973, and [De2 ] by Demazure in 1974, where both authors
verified it using different methods (cf. introduction to [Ch2 ] by Borel).
In recent years, inspired by theory of Schubert polynomials of Lascoux and
Schützenberger [LS], many achievements have been made in generalizing the clas-
sical Pieri formula, which handles the problem for the case where G is a matrix
group and where one of Pu and Pv is a special Schubert class. (See [FP, Section
9.10] for more recent progresses and relevant references).
While the problem in its natural generality remains unsolved1 , the further
problem of determining the multiplicative rule of Schubert classes in the quantum
cohomology of G/H has appeared on the agenda, where the analogue of the
coefficients awu,v are known as Gromov-Witten numbers (cf. [FP, p.134], [CF]).
It was announced in [Du2 ] that, combining the ideas of Bott-Samelson resolu-
tions of Schubert varieties and the enumerative formula on a twisted products of
2-spheres obtained in [Du2 ], it is possible to find a unified formula that expresses
awu,v in terms of certain Cartan numbers of G. This paper is devoted to complete
this project.
1
We quote from Fulton and Pragacz [FP]: there is no analogue of the Littlewood Richardson
rule for explicitly multiplying Schubert classes in a flag manifold; from Sottile [S]: the analog
of the Littlewood Richardson rule is not known for most other flag variety G/P.
For the cases of matrix groups, the theory of Schubert polynomials was developed to make
explicit computation with Schubert classes possible (cf. introduction to [BH]).

2
2 Main result

A few notations will be needed in presenting our result. Throughout this paper
G is a compact connected Lie group with a fixed maximal torus T . We set
n = dim T .

2.1. Geometry of Cartan subalgebra. Equip the Lie algebra L(G) of G

with an inner product (, ) so that the adjoint representation acts as isometries of
L(G). The Cartan subalgebra of G is the Euclidean subspace L(T ) of L(G).
The restriction of the exponential map exp : L(G) → G to L(T ) defines a
set D(G) of m = 12 (dim G − n) hyperplanes in L(T ), i.e. the set of singular
hyperplanes through the origin in L(T ). These planes divide L(T ) into finitely
many convex cones, called the Weyl chambers of G. The reflections σ of L(T )
in the these planes generate the Weyl group W of G. Let Φ ⊂ L(T ) be the
root system associated to W ([Hu]). Recall that if β, β ′ ∈ Φ, the Cartan number
β ◦ β ′ = 2(β, β ′)/(β ′, β ′ ) is an integer (only 0, ±1, ±2, ±3 can occur).
Fix, once and for all, a regular point α ∈ L(T )\ ∪ L, and let Φ+ (resp.
L∈D(G)
∆) be the set of positive roots (resp. simple roots) relative to α. The set D(G)
can now be indexed by Φ+ as {Lβ | β ∈ Φ+ }, where Lβ is the singular plane
corresponding to the root β. For a β ∈ Φ+ , write σβ ∈ W for the reflection of
L(T ) in Lβ . If β ∈ ∆ we call σβ a simple reflection.

It is known that the set of simple reflections {σβ | β ∈ ∆} generates W . That

is, any w ∈ W admits a factorization of the form
(2.1) w = σβ1 ◦ · · · ◦ σβk , βi ∈ ∆.
Definition 1. The length l(w) of an w ∈ W is the least number of factors
in all decompositions of w in the form (2.1). The decomposition (2.1) is said
reduced if k = l(w).
If (2.1) is a reduced decomposition, the k×k (strictly upper triangular) matrix
Aw = (ai,j ) with
0 if i ≥ j;
ai,j = {
−βi ◦ βj if i < j
is called the Cartan matrix of w associated to the decomposition (2.1).

Example 1. By resorting to the geometry of the Cartan subalgebra L(T )

there is geometric method to find a reduced decomposition (hence a Cartan ma-
trix) of an w ∈ W .
Picture W as the W -orbit {w(α) ∈ L(T ) | w ∈ W } of the regular point
α. Given an w ∈ W let Cw be a straight line segment in L(T ) from the Weyl
chamber containing α to w(α) that crosses the planes in D(G) one at a time.
Assume that they are met in the order Lα1 , · · · , Lαk , αi ∈ Φ+ . We have l(w) = k
and w = σαk ◦ · · · ◦ σα1 (cf. [Du2 ] or [Han]). Set

3
 β1 = α1 , β2 = σα1 (α2 ), · · · , βk = σα1 ◦ · · · ◦ σαk−1 (αk ).
Then βi ∈ ∆. Moreover, from
σβi = σα1 ◦ · · · ◦ σαi−1 ◦ σαi ◦ σαi−1 ◦ · · · ◦ σα1 ,
one verifies easily that w = σβ1 ◦ · · · ◦ σβk . This decomposition is reduced because
of βi ∈ ∆ and l(w) = k.
As an example of the method consider the case of G2 , the exceptional Lie
group of rank 2. With dim L(T ) = 2, the singular lines, denoted by Li , i ≤ 6,
are depicted in the figure. Taking a regular point α ∈ L(T ) as marked, the set
of simply roots is ∆ = {β1 , β2 }. Let W be the Weyl group of G2 .

For the elements w, w′ ∈ W specified by the vectors w(α), w′(α) ∈ L(T ) in

the figure, we get from the segments Cw and Cw′ the reduced decompositions
w = σβ2 ◦ σβ1 ◦ σβ2 ◦ σβ1 ◦ σβ2 ;
w′ = σβ1 ◦ σβ2 ◦ σβ1 ◦ σβ2 ◦ σβ1 .
From the Cartan matrix ofG2 ([Hu,
 p.59]) 
β1 ◦ β1 β1 ◦ β2 2 −1
=
β2 ◦ β1 β2 ◦ β2 −3 2
we read that the Cartan matrices of w and w ′ associated to the decompositions
are respectively
   
0 3 −2 3 −2 0 1 −2 1 −2
 0 0 1 −2 1   0 0 3 −2 3 
   
Aw =   0 0 0 3 −2  and Aw′ =  0 0 0
  1 −2 .

 0 0 0 0 1   0 0 0 0 3 
0 0 0 0 0 0 0 0 0 0

4
 2.2. The triangular operators. Let Z[x1 , · · · , xk ] = ⊕r≥0 Z[x1 , · · · , xk ](r)
be the ring of integral polynomials in x1 , · · · , xk , graded by | xi |= 1.
Definition 2. Given an k × k strictly upper triangular integer matrix
A = (ai,j ) the triangular operator associated to A is the homomorphism TA :
Z[x1 , · · · , xk ](k) → Z defined recursively by the following elimination laws.
1) if h ∈ Z[x1 , · · ·, xk−1](k) , then TA (h) = 0;
2) if k = 1 (consequently A = (0)), then TA (x1 ) = 1;
3) if h ∈ Z[x1 , · · ·, xk−1](k−r) with r ≥ 1, then
TA (hxrk ) = TA′ (h(a1,k x1 + · · · + ak−1,k xk−1 )r−1 ),
where A′ is the ((k − 1) × (k − 1) strictly upper triangular) matrix obtained from
A by deleting the k th column and the k th row.
By additivity, TA is defined for every f ∈ Z[x1 , · · · , xk ](k) using the unique
expansion f = Σhr xrk with hr ∈ Z[x1 , · · ·, xk−1 ](k−r) .

Example 2. Definition  2 gives  an effective algorithm to evaluate TA .

0 a
For k = 2 and A1 = , then TA1 : Z[x1 , x2 ](2) → Z is given by
0 0
TA1 (x21 ) = 0,
TA1 (x1 x2 ) = TA′1 (x1 ) = 1 and
TA1 (x22 ) = TA′1 (ax1 ) = a.
 
0 a b
For k = 3 and A2 =  0 0 c , then A′2 = A1 and TA2 : Z[x1 , x2 , x3 ](3) → Z
0 0 0
is given by
0, if r3 = 0 and
TA2 (xr11 xr22 xr33 ) = {
TA1 (xr11 xr22 (bx1 + cx2 )r3 −1 ), if r3 ≥ 1,
where r1 + r2 + r3 = 3, and where TA1 is calculated in the above.

It is straightforward from Definition 2 that

Corollary 1. We have TA (x1 · · · xk ) = 1 and
TA (xr11 · · · xrkk ) = 0
whenever r1 + · · · + ri > i for some 1 ≤ i < k.

The operator TA can also be given by explicit formula. For a sequence

(r1 , · · · , rk ) of k non-negative integers with Σri = k, let C(r1 , · · · , rk ) be the
set of all strictly upper triangular k × k matrices C = (ci,j ) with non-negative
integer entries satisfying
Σcs,i = ri − 1 + Σci,j , 1 ≤ i ≤ k.
s j
Definition 2, together with induction on k, yields ci,j
a
Corollary 2. TA (xr11 · · · xrkk ) = Σ Π(Σci,j )!Π ci,j
i,j !
.
(ci,j )∈C(r1 ,··· ,rk ) j i i,j

5
 2.3. The formula. It is well known that simply connected semi-simple Lie
groups are classified by their Cartan matrices [Hu, p.55]. So, conceivably, any
geometric invariant associated to G/H can be reduced in principle to Cartan
numbers of G (entries in the Cartan matrix of G). Explicit and direct relationship
may become more desirable if one wants to find an expression of the invariant in
its natural generality (i.e. uniformly for all G/H) rather than for special cases.
We present both a formula and an algorithm, which evaluate the number aw u,v in
term of Cartan numbers of G.

Assume that w = σβ1 ◦ · · · ◦ σβk , βi ∈ ∆, is a reduced decomposition of an

w ∈ W , and let Aw = (ai,j )k×k be the associated Cartan matrix. For a subset
L = [i1 , · · · , ir ] ⊆ [1, · · · , k] we put | L |= r and set
σL = σβi1 ◦ · · · ◦ σβir ∈ W ; xL = xi1 · · · xir ∈ Z[x1 , · · · , xk ].
Our solution to the problem is
Theorem. If u, v P ∈ W with l(w) P = l(u) + l(v), then
awu,v = TAw [( x L )( xK )],
|L|=l(u) |K|=l(v)
σL =u σK =v
where L, K ⊆ [1, · · · , k].

The proof of the Theorem will be developed in such a way as to suggest its
analogue for multiplying generalized Schubert classes in the focal manifolds of
isoparametric submanifolds. In particular, the Theorem is valid for the focal
manifolds of an isoparametric submanifold with equal multiplicities 2, in which
the classical flag manifolds G/H are special cases (cf. 7.2, 7.5 and [HPT]).
In [Bi] S. Billey obtained a recurrence [Bi, (5.5)] that can be used to derive an
expression for aw v s
u,v by using all of the quantities ξ (t) |α , ξ (t) |α and πt |α with
u ≤ s < t ≤ w , where the ξ s (t) and the πt for t, s ∈ W are certain polynomials
in the simple roots of G defined respectively by Kostant and Kumar in [KK] and
by Billey in [Bi], and where |α means evaluating the polynomials at the regular
point α. In view of [Bi, Theorem 4], our theorem expresses aw u,v only in terms of
the data required to describe ξ u (w) and ξ v (w).

Example 3. Let W be the Weyl group of G2 . Continuing from Example 1

we express the product Pu Pv in H ∗ (G2 /T ) in terms of Schubert classes, where
u, v ∈ W are specified by the vectors u(α), v(α) ∈ L(T ) in the figure. We
note that l(u) = 3, l(v) = 2 and w, w ′ ∈ W are the only elements with length
l(u) + l(v) = 5.
Referring to the reduced decomposition w = σβ2 ◦ σβ1 ◦ σβ2 ◦ σβ1 ◦ σβ2 obtained
in Example 1, the solutions in L ⊂ [1, 2, 3, 4, 5] to the equations σL = u, | L |= 3
are
L = (1, 2, 3), (1, 2, 5), (1, 4, 5), (3, 4, 5);
and the solutions in K ⊂ [1, 2, 3, 4, 5] to the equations σK = v, | K |= 2 are
K = (2, 3), (2, 5), (4, 5).

6
Using the Theorem we compute
aw
u,v = TAw [(x1 x2 x3 + x1 x2 x5 + x1 x4 x5 + x3 x4 x5 )(x2 x3 + x2 x5 + x4 x5 )]
= 2 + 2TAw (x1 x2 x4 x25 ) + TAw (x1 x24 x25 )
+TAw (x2 x23 x4 x5 ) + TAw (x2 x3 x4 x25 ) + TAw (x3 x24 x25 ),
where the second equality follows from the additivity of TAw and an application
of Corollary 1. With the matrix Aw being determined in Example 1, we find that
TAw (x1 x2 x4 x25 ) TAw (x1 x24 x25 ) TAw (x2 x23 x4 x5 ) TAw (x2 x3 x4 x25 ) TAw (x3 x24 x25 )
.
1 −2 1 −1 −1
Consequently, aw u,v = 1.
Similarly, from the reduced decomposition w′ = σβ1 ◦ σβ2 ◦ σβ1 ◦ σβ2 ◦ σβ1 we
find P P
xL = x2 x3 x4 , xK = x1 x2 + x1 x4 + x3 x4 .
|L|=l(u) |K|=l(v)
σL (α)=u(α) σK (α)=v(α)
From the Theorem we get
′
aw
u,v = TAw [(x2 x3 x4 )(x1 x2 + x1 x4 + x3 x4 )] = 0 (by Corollary 1).
Summarizing, Pu Pv = Pw .

Remark 1. It follows from intersection theory that the coefficients aw

u,v are
always non-negative. It is therefore attempted to have a method to compute
the number without cancellation involved (i.e. positively multiplying Schubert
classes). From the computation in Example 3 one finds that our method is not
positive. However, the geometric reason behind this phenomenon can be easily
clarified. This will be discussed in subsection 7.6.

2.4. The algorithm. In concrete situations one prefers to see the practical
value of awu,v rather than the closed formula, for this could reveal in a direct way
the intersection multiplicities of Xu with Xv in the variety Xw . For this purpose
the Theorem does indicate an effective algorithm to evaluate aw u,v , as the following
recipe shows (see also Examples 1-3).
(1) starting from the Cartan matrix of G, a program to enumerate all elements
in a coset W of the Weyl group W by their minimal reduced decompositions is
available in [DZZ];
(2) for an w ∈ W with a reduced decomposition, the corresponding Cartan
matrix Aw can be read directly from Cartan matrix of G (cf. Example 1);
(3) for an w ∈ W with a reduced decomposition w = σβ1 ◦ · · · ◦ σβk and an
u ∈ W with l(u) = r < k, the solutions in the subsequence [i1 , · · · , ir ] ⊆ [1, · · · , k]
to the equation σβi1 ◦ · · · ◦ σβir = u in W agree with the solutions to the vector
equation σβi1 ◦ · · · ◦ σβir (α) = u(α) in the linear space L(T ), where α ∈ L(T ) is
a fixed regular point;
(4) the evaluation the operator TAw on a polynomial can be easily programmed
(cf. Example 2).
Combining the ideas explained above, a program to compute the numbers
w
au,v has been compiled [DZ2]. It uses Cartan matrix as the only input and

7
computations in various flag manifolds G/H can be performed by the single
program.
As for the efficiency of the program, we refer the reader to the computa-
tional results tabulated in [DZ1 ]. They were produced by a similar program that
implements Steenrod operations on Schubert classes.

2.5. Arrangement of the paper. The rest sections of the paper are so
arranged. Section 3 develops preliminary results from algebraic topology. We
recall from [Du2 ] the cohomology of twisted product of 2-spheres (Lemma 3.3),
and the enumerative formula on these manifolds (Lemma 3.4). In particular, we
introduce divided differences for spherical represented involutions, and their basic
properties are established in Lemma 3.2.
We shall see in Section 4 that, by resorting to the geometry of the adjoint
representation, Bott-Samelson cycles in the space G/T appears as certain twisted
product of 2-spheres that are parameterized by ordered sequences of roots, and
the divided differences on the integral cohomology of G/T arise naturally from the
geometric fact that the involution on G/T corresponding to a root is spherical
representable. After determining the induced action of Bott-Samelson cycles
(corresponding to a sequence of simply roots) on Schubert classes in Lemma 5.1
(Section 5), the Theorem is established in Section 6.

In many literatures ranging from topology, algebraic and differential geometry

to representation theory, one finds analogues of the terminologies that we work
with, such as Bott-Samelson cycles (or schemes), divided differences and Schu-
bert varieties, but with seemingly different appearances. In particular, Schubert
varieties were originally introduced and extensively studied in the context of al-
gebraic geometry, but we will work with Lie groups in the real compact form
so that our method are ready to apply to general situations (cf. 7.5). In order
to merge our presentations into the existing literatures and, at the same time,
not to interrupt our exposition, Section 7, entitled Historical remarks, is devoted
to recall those historical events illustrating the readiness and necessity of the
conceptual development in our paper, and will be referred to from time to time.

Finally, a brief account for the method of the proof. In 1973 Hansen discovered
that the celebrated K-cycles on the flag manifold G/T constructed by Bott and
Samelson in 1958 (cf. [BS2 ] or 7.1) provided a degree 1 map gw from a twisted
products Γw of 2-spheres onto the Schubert variety Xw (cf. [Han] or 7.4). This
suggests that the intersection product in Xw can be translated as part of the
intersection product in Γw via the homomorphism induced by gw . However, the
latter is much easier to work with for the following reasons (cf. Lemma 3.3 and
Lemma 3.4 in Section 3).
a) the space Γw admits a natural cell decomposition with each cell, again, a
twisted product of 2-spheres;

8
 b) the cohomology of Γw is a polynomial ring Z[x1 , · · · , xk ] generated by xi ’s
in dimension 2, subject to relations occurring only in dimension 4;
c) the intersection product in Γw is handled by a triangular operator TA .
Therefore, the intersection multiplicity aw
u,v in question can be calculated by com-
putations in a space like Γw , rather than in the Schubert variety Xw itself.

3 Preliminaries in topology
In this paper all homologies (resp. cohomologies) will have integer coefficients
unless otherwise stated. If f : X → Y is a continuous map between two topolog-
ical spaces, f∗ (resp. f ∗ ) is the homology (resp. cohomology) map induced by f .
Write S r for the r-dimensional sphere. If M is an oriented closed manifold (resp.
a connected projective variety) [M] ∈ Hdim M (M) stands for the orientation class.
The Kronecker pairing, between cohomology and homology of a space X, will be
denoted by <, >: H ∗ (X) × H∗ (X) → Z.

3.1. Sphere bundle with a cross section. Let p : E → M be a smooth,

oriented r-sphere bundle over an oriented manifold M which has a cross section
s : M → E. Let the normal bundle ξ of the embedding s be oriented by p, and
let e ∈ H r (M) be the Euler class of ξ with respect to this orientation.
The integral cohomology H ∗ (E) can be described as follows. Denote by i :
S r → E for the fiber inclusion of p over a point z ∈ M, and write by J : E → E
for the involution given by the antipodal map in each fiber sphere. We have the
following result from [Du2 , Lemma 4].
Lemma 3.1. There exists a unique class x ∈ H r (E) such that
s∗ (x) = 0 ∈ H ∗ (M) and < i∗ (x), [S r ] >= 1.
Furthermore
(1) H ∗ (E), as a module over H ∗ (M), has the basis {1, x} subject to the
relation x2 + p∗ (e)x = 0;
(2) the induced cohomology map J ∗ acts identically on the subset Im p∗ ⊂
∗
H (E) and
J ∗ (x) = (−1)r−1 x − p∗ (e).
Remark 2. If r is odd, 2e = 0.

3.2. Divided difference of a spherical represented involution. A

self-map σ of a manifold M is called an involution if σ 2 = id : M → M. An r-
spherical representation of the involution (M; σ) is a system f : (E; J) → (M; σ)
in which
(1) E is the total space of an oriented r-sphere bundle p : E → M with a
cross section s;
(2) f is a continuous map E → M that satisfies the following two constrains

9
(3.1) f ◦ s = id : M → M; and
(3.2) f ◦ J = σ ◦ f : E → M,
where J : E → E is the involution on E given by the antipodal map in the fibers.

In view of the H ∗ (M)-module structure of H ∗ (E) specified by Lemma 3.1,

an r-spherical representation f of the involution (M, σ) gives rise to an additive
operator θf : H m (M) → H m−r (M) of degree −r that is characterized uniquely
as follows.
The induced homomorphism f ∗ : H ∗ (M) → H ∗ (E) satisfies
(3.3) f ∗ (z) = p∗ (z) + p∗ (θf (z))x
for all z ∈ H ∗ (M).

Useful properties of θf can be derived directly from its definition (3.3).

Lemma 3.2. Let e ∈ H r (M) be as that in Lemma 3.1. We have
(1) σ ∗ = Id − eθf : H ∗ (M) → H ∗ (M);
(2) θf (z1 z2 ) = θf (z1 )z2 + σ ∗ (z1 )θf (z2 ), z1 , z2 ∈ H ∗ (M).
(3) if r is even, then θf (e) = 2( ∈ H 0 (M) = Z); if r is odd, then θf (e) = 0;
(4) if r is even, 2θf ◦ θf = 0 : H ∗ (M) → H ∗ (M).
Proof. Since f ∗ is a ring map, we have
f ∗ (z1 z2 ) = [p∗ (z1 ) + p∗ (θf (z1 ))x][p∗ (z2 ) + p∗ (θf (z2 ))x]
= p∗ (z1 z2 ) + p∗ [θf (z1 )z2 + z1 θf (z2 )]x + p∗ [θf (z1 )θf (z2 )]x2
= p∗ (z1 z2 ) + p∗ [θf (z1 )z2 + z1 θf (z2 ) − eθf (z1 )θf (z2 )]x,
where the last equality is obtained from x2 = −p∗ (e)x (because of the relation
x2 + p∗ (e)x = 0). Comparing this with (3.3) yields
(3.4) θf (z1 z2 ) = θf (z1 )z2 + z1 θf (z2 ) − eθf (z1 )θf (z2 )
= θf (z1 )z2 + (z1 − eθf (z1 ))θf (z2 ).
Applying J ∗ to (3.3) gives
J ∗ f ∗ (z) = p∗ (z) + p∗ (θf (z))J ∗ (x)
by (2) of Lemma 3.1. From J ∗ f ∗ = f ∗ σ ∗ (by (3.2)) and J ∗ (x) = (−1)r−1 x − p∗ (e)
(by (2) of Lemma 3.1) we get
f ∗ (σ ∗ (z)) = p∗ (z − eθf (z)) + p∗ ((−1)r−1 θf (z))x.
Comparing this with (3.3) gives rise to
(3.5) σ ∗ (z) = z − eθf (z); and
(3.6) θf (σ ∗ (z)) = (−1)r−1 θf (z).
(1) is verified by (3.5). Combining (3.4) with (3.5) shows (2).
Substituting (3.5) in the left hand side of (3.6) and rewriting the resulting
equation by using (3.4) gives
(3.7) (1 + (−1)r )θf (z) = θf (e)θf (z) + (e − θf (e)e)θf (θf (z)).
Taking z = e in (3.7) and noting that θf (θf (e)) ∈ H −r (M) = 0 we get the
equation (1 + (−1)r )θf (e) = θf (e)2 in H 0 (M) = Z. This proves (3).
Now (3.7) becomes eθf (θf (z)) = 0 for all z ∈ H ∗ (M) by (3). Consequently,
taking z = θf (u) in (3.5) yields σ ∗ (θf (u)) = θf (u). Now (3.6) implies that if r is

10
even, 2θf (θf (u)) = 0 for all u ∈ H ∗ (M). This verifies (4), hence completes the
proof of Lemma 3.2.

We observe from (1) of Lemma 3.2 that, for any z ∈ H ∗ (M), the difference
z − σ ∗ (z) ∈ H ∗ (M) is always divisible by e with quotient θf (z).
Definition 3. The operator θf is called the divided difference of the spherical
representation f of the involution (M, σ) (Compare with the discussion in 7.5).

3.3. Twisted product of 2-spheres. The following definition singles out

a class spaces in which we will be particularly interested.
Definition 4. A smooth manifold M is called an oriented twisted product of
2-spheres of rank k, denoted by M = ∝ S 2 , if there is a tower of smooth maps
1≤i≤k
pk−1 pk−2 p2 p1
M = Mk → Mk−1 → · · · → M2 → M1
in which
1) M1 is diffeomorphic to S 2 with an orientation;
2) pi is the projection of an oriented smooth S 2 bundle over Mi ;
3) pi has a fixed cross section si , 1 ≤ i ≤ k − 1.

Let M = ∝ S 2 be a twisted product of 2-spheres of rank k. The cross

1≤i≤k
sections si yield a sequence of embeddings
s1 s2 sk−1
M1 → M2 → · · · → Mk = M.
In view of this we shall make no notational distinction between a subspace N ⊂
Mi and its image under si in Mj , j ≥ i. Take a base point x0 ∈ M1 and let each
Mi has x0 as its base point.
For a subset L = [i1 , · · · , ir ] ⊆ [1, · · · , k] a smooth submanifold
S(L) ⊂ Mik ⊂ Mik +1 ⊂ · · · ⊂ M
can be introduced inductively as follows.
1) S(1) = M1 ;
2) if i > 1, S(i) ⊂ Mi is the fiber sphere of pi−1 over the base point;
′
3) assume that S(L ) ⊂ Mir−1 , where L′ = [i1 , · · · , ir−1 ], has been defined
and consider the case L = [i1 , · · · , ir−1 , j]. Then S(L) ⊂ Mj is the total space of
′
the restricted bundle of pj−1 : Mj → Mj−1 to the subspace S(L ) ⊂ Mj−1 . The
′
natural bundle map over the inclusion S(L ) ⊂ Mir−1 → Mj−1 gives rise to the
desired embedding S(L) ⊆ Mj ⊆ M.

The integral homology (resp. cohomology ) of an M = ∝ S 2 can be de-

1≤i≤k
scribed as follows. Consider the normal bundle πi : Ei → Mi−1 of the embedding
si−1 : Mi−1 → Mi , 2 ≤ i ≤ k. It is a 2-plane bundle with a natural orientation
inherited from that on pi−1 . Let e′i ∈ H 2 (Mi−1 ) be the Euler class of πi . We set
e1 = 0 and for i ≥ 2
ei = (pi−1 ◦ pi ◦ · · · ◦ pk−1 )∗ e′i ∈ H 2 (M).

11
 Lemma 3.3. For an M = ∝ S 2 let [S(L)] ∈ H2|L| (M) be the fundamental
1≤i≤k
class of the cycle S(L) ⊂ M. Then
(1) the set {[S(L)] | L ⊆ [1, · · · , k]} of homology classes is an additive basis
for the graded Z-module H∗ (M).
Further, let {xL ∈ H 2|L| (M) | L ⊆ [1, · · · , k]} be the basis of H ∗ (M) Kro-
necker dual to the basis {[S(L)] | L ⊆ [1, · · · , k]} in homology. Then
(2) xL = xi1 · · · xir if L = [i1 , · · · , ir ];
(3) H ∗ (M) = Z[x1 , · · ·, xk ]/ < x2i + ei xi ; 1 ≤ i ≤ k >, where ei is a polynomial
in x1 , · · ·, xi−1 .
Proof. By setting Tr (M) = ∪ S(L) ⊆ M, 1 ≤ r ≤ k, we get a filtration of
|L|=r
subspaces
(3.8) T1 (M) ⊂ T2 (M) ⊂ · · · ⊂ Tk (M) = M.
Moreover, the subspace Tr (M)\Tr−1 (M) consisting of 2r-dimensional open cells
in one-to-one correspondence with the subset L ⊆ [1, · · · , k] with | L |= r. This
implies that (3.8) dominates M by a cell complex with only even dimensional
cells. This proves (1).
Assertions (2) and (3) follow easily from (1) of Lemma 3.1 together with
induction on k.

For degree reasons the polynomials ei are all homogeneous of degree 1 in

x1 , · · ·, xi−1 , i.e.
ei = a1,i x1 + · · · + ai−1,i xi−1 , ai,j ∈ Z, 1 ≤ i ≤ k.
Definition 5. With ai,j = 0 for i ≥ j being understood, the strictly upper
triangular matrix A = (−ai,j )k×k is called the structure matrix of M = ∝ S 2 .
1≤i≤k
Remark 3. It was shown in [Du2 , Proposition 1] that any strictly upper
triangular matrix A of rank k can be realized as the structure matrix of an
M = ∝ S 2.
1≤i≤k

3.4. Integration along a twisted product of 2-spheres. Recall from

Lemma 3.3 that the integral cohomology of an M = ∝ S 2 can be specified as
1≤i≤k
a quotient of a free polynomial ring
H ∗ (M) = Z[x1 , · · ·, xk ]/ < x2i + ei xi ; 1 ≤ i ≤ k >.
Write Z[x1 , · · ·, xk ](r) for the subset of all homogeneous polynomials of degree r
in Z[x1 , · · ·, xk ] and let pM : Z[x1 , · · ·, xk ](r) R→ H 2r (M) be the obvious quotient
(k)
map. Consider R the additive correspondence M : Z[x1 , · · ·, xk ] → Z defined by
M
h =< pM (h), [M] >,
where [M] ∈ H2k (M) = Z is the orientation class. R
As being indicated by the notation, the operator M can be interpreted as
“integration along M” in De Rham theory. In view of discussions in [Du1 ],
the problem of effective computation R of the intersection product in M asks an
effective algorithm to evaluate M . The idea of structure matrix of M, together

12
with the operator TA introduced in 2.2, is useful in presenting such an algorithm.
The following result was shown in [Du2 , Proposition 2].
Lemma 3.4. R If M has the structure matrix A = (aij )k×k , then
(k)
M
= TA : Z[x1 , · · ·, xk ] → Z.

4 Geometry from adjoint representation

Since the point α ∈ L(T ) is regular, the adjoint representation Ad : G → L(G)
yields a smooth embedding
ϕ : G/T → L(G) by ϕ(gT ) = Adg (α).
In this way G/T becomes a submanifold of the Euclidean space L(G).

4.1. Preliminaries. Let Φ+ be the set of positive roots relative to α (cf.

2.1). Assume that the Cartan decomposition of the Lie algebra L(G) relative to
T ⊂ G is
L(G) = L(T ) ⊕β∈Φ+ Fβ ,
where Fβ is the root space, viewed as a real 2-plane, belonging to the root β ∈ Φ+
([Hu, p.35]). Let [, ] be the Lie bracket on L(G). We make a simultaneous choice
of an orientation on each Fβ by the following rule: take a v ∈ Fβ \{0} and let
v′ ∈ Fβ be such that [v, v ′] = β. Then Fβ is oriented by the ordered basis {v, v ′}.
Clearly this orientation on Fβ is independent of a specific choice of v.

We quote from [HPT, p.426-427] for relevant information concerning the ge-
ometries of the submanifold G/T ⊂ L(G).
(4.1) The subspaces ⊕β∈Φ+ Fβ and L(T ) of L(G) are tangent and normal to
G/T at α respectively;
(4.2) The tangent bundle to G/T has a canonical orthogonal decomposition
into the sum of m integrable 2-plane bundles ⊕β∈Φ+ Eβ with Eβ (α) = Fβ .
In general, Eβ (y) = Adg (Fβ ) if y = Adg (α) ∈ G/T .
(4.3) The leaf of the integrable subbundle Eβ through a point y ∈ G/T ,
denoted by S(y; β), is a 2-sphere.
In view of (4.2) and (4.3) we adopt the following convention.
Convention. Each bundle Eβ , β ∈ Φ+ , is oriented such that the identifi-
cation Eβ (α) = Fβ is orientation preserving. Let the 2-sphere S(y; β) have the
orientation inherited from that on its tangent plane Eβ (y) at y ∈ S(y; β).

By considering G/T as a submanifold in the Euclidean space L(G) via the

embedding ϕ, the geometric features singled out in (4.2) and (4.3) will dominate
all of our geometric constructions and computations in this section.

4.2. Realization of roots as homology and cohomology classes. We

start by relating the set Φ of roots with certain 2-dimensional homology (resp.
cohomology) classes of G/T .

13
 For a β ∈ Φ+ write by eβ ∈ H 2 (G/T ) for the Euler class of the oriented
bundle Eβ , and let ϕβ : S(y; β) → G/T be inclusion of the leaf sphere (cf. (4.3)).
Lemma 4.1. The Kronecker paring H 2 (G/T ) × H2 (G/T ) → Z can be ex-
pressed in term of Cartan numbers as
< eγ , ϕβ∗ [S(y; β)] >= β ◦ γ, β, γ ∈ Φ,
where [S(y; β)] ∈ H 2 (S(y; β))( = Z) is the orientation class.
Remark 4. The class ϕβ∗ [S(y; β)] ∈ H2 (G/T ) is independent of the choice of
the point y ∈ G/T . Let ε : [0, 1] → G/T be a path that joins y to some y ′. Then
the continuous one-parameter family of 2-spheres S(ε(t); β) ⊂ G/T , t ∈ [0, 1] is
an isotopy from S(y; β) to S(y ′; β).

The Weyl group W of G, acting as isometries of the Cartan subalgebra L(T ),

has the effect to permute the set Φ of roots [Hu]. On the other hand, via the
embedding ϕ, the canonical action of W on G/T [BS2 ] is given by
(4.4) w(x) = Adg (w(α)), x = Adg (α) ∈ G/T , w ∈ W .
Let w ∗ : H ∗ (G/T ) → H ∗ (G/T ) be the induced action on cohomology.
Lemma 4.2. For an w ∈ W one has
(1) w(S(y; β)) = S(w(y); w(β));
(2) w ∗ (Eβ ) = Ew−1 (β) ; and
(3) w ∗ (eβ ) = ew−1 (β) .
Proof. In term of the W -action on the set Φ of roots, the induced bundle
∗
w (Eβ ) is Ew−1 (β) (cf. [HTP, 1.6]). This shows (2). Assertion (3) comes now
from the naturality of Euler classes.
Item (1), which follows also from (2) (and (4.3)), indicates that the action of
w on G/T has the effect to carry the leaf sphere through y corresponding to a
root β diffeomorphically onto the leaf sphere through w(y) corresponding to the
root w(β).

4.3. 2-spherical represented involutions. Each root β ∈ Φ+ gives rise

to an involution σβ : G/T → G/T in the fashion of (4.4), and defines also the
subspace
S(β) = {(y, y1) ∈ G/T × G/T | y1 ∈ S(y; β)}.
Projection pβ : S(β) → G/T onto the first factor is easily seen to be a 2-sphere
bundle projection (with the leaf sphere S(y; β) as the fiber over y ∈ G/T ). The
map sβ : G/T → S(β) by sβ (y) = (y, y) furnishes pβ with a ready-made cross
section.
Let Jβ be the involution on S(β) given by the antipodal map on each fiber
sphere and let fβ : S(β) → G/T be the projection onto the second factor. Then,
as is clear,
fβ ◦ sβ = id : G/T → G/T ;
fβ ◦ Jβ = σβ ◦ fβ : S(β) → G/T.
That is, the map fβ : (S(β), Jβ ) → (G/T, σβ ) is a 2-spherical representation of
the involution (G/T, σβ ) (cf. 3.2).

14
 Lemma 4.3. Assume the same setting as the above.
1) As a H ∗ (G/T )-module, the cohomology of S(β) is given by
H ∗(S(β)) = H ∗ (G/T )[1, x]/ < x2 + eβ x >,
in which x ∈ H 2 (S(β)) is uniquely characterized by s∗β (x) = 0 and
< i∗ (x), [S(y; β)] >= 1 (for all y ∈ G/T ),
where i : S(y; β) → S(β) is the inclusion of the fiber over y ∈ G/T .
2) Write θβ : H ∗ (G/T ) → H ∗ (G/T ) to denote the divided difference associ-
ated to the 2-spherical representation fβ of the involution (G/T, σβ ). Then
(1) for any γ ∈ Φ, θβ (eγ ) = β ◦ γ( ∈ H 0 (G/T ) = Z);
(2) θβ ◦ θβ = 0;
(3) σβ∗ = Id − eβ θβ : H ∗ (G/T ) → H ∗ (G/T );
(4) θβ (z1 z2 ) = θβ (z1 )z2 + σβ∗ (z1 )θβ (z2 ), z1 , z2 ∈ H ∗ (G/T ), and
(5) for any w ∈ W , θw(β) = (w −1 )∗ θβ w ∗.
Proof. Since the normal bundle of the embedding sβ is Eβ , one has the
relation x2 + eβ x = 0 by (1) of Lemma 3.1. This verifies 1).
In 2) properties (3) and (4) corresponds to the items (3) and (4) in Lemma
3.2. (2) follows from (2) of Lemma 3.2 since H ∗ (G/T ) is torsion free [BS1 ]. It
remains to show (1) and (5).
Property (1) is verified by
β ◦ γ =< eγ , ϕβ∗ [S(y; β)] >(by Lemma 4.1)
=< ϕ∗β (eγ ), [S(y; β)] > (by the naturality of <, >)
=< i∗ ◦ fβ∗ (eγ ), [S(y; β)] > (since ϕβ = fβ ◦ i)
=< i∗ (p∗β (eγ ) + p∗β (θβ (eγ ))x), [S(y; β)] > (the definition of θβ in (3.3))
= θβ (eγ ) (by 1) of this Lemma),
where in the last equality we have applied the standard fact that
(4.5) the composition i∗ ◦ p∗β : H r (G/T ) → H r (S(y; β)) is zero in degree r > 0,
and is an isomorphism Z → Z if r = 0.
Finally we show (5). For an w ∈ W the self diffeomorphism w × w : G/T ×
G/T → G/T × G/T restricts to a diffeomorphism w e : S(β) → S(w(β)) (by (1)
of Lemma 4.2) that fits into the following two commutative diagrams
we e
w
S(β) → S(w(β)) S(β) → S(w(β))
pβ ↓ ↓ pw(β) and fβ ↓ ↓ fw(β)
w w
G/T → G/T G/T → G/T .
From the definition of θβ in (3.3) we have
fβ∗ (w ∗(z)) = p∗β (w ∗ (z)) + p∗β (θβ (w ∗ (z)))x, z ∈ H ∗ (G/T ).
On the other hand
fβ∗ (w ∗ (z)) = w e∗ fw(β)
∗
(z) (by the commutivity of the second diagram)
=w e (pw(β) (z) + pβ [θw(β) (z)]x′ ) (by the definition of θw(β) )
∗ ∗ ∗

= p∗β (w ∗ (z)) + p∗β (w ∗[θw(β) (z)])x (by the commutivity of the first diagram).
Comparing the coefficients of x ∈ H ∗ (S(β)) in the above two expressions of
fβ∗ (w ∗ (z)) yields w ∗ θw(β) = θβ w ∗ . This shows (5).

15
 Lemma 4.4. Given an ordered sequence β1 , · · · , βk ∈ ∆ of simple roots we
put w = σβ1 ◦ · · · ◦ σβk , θ(β1 ,··· ,βk ) = θβ1 ◦ · · · ◦ θβ1 .
(1) if l(w) < k, then θ(β1 ,··· ,βk ) = 0;
(2) if l(w) = k, then θ(β1 ,··· ,βk ) depends only on w and not on the ordered
sequence β1 , · · · , βk ∈ ∆. In this case we put θw = θ(β1 ,··· ,βk ) .
Proof. Furnished with properties (1)-(5) in Lemma 4.3 (in analogue with
3.3 Lemma in [BGG]), an argument parallel to the proof of [BGG, 3.4 Theorem]
verifies Lemma 4.4. To do so one needs only to replace b∗Q = b∗Z ⊗ Q in [BGG]
by H 2 (G/T ), where b∗Z is the group of weights of G [Hu. p.67], and to resort
to certain properties of Weyl groups from [BGG; §2]. For brevity we omit the
details (see also Proposition 6 in 7.5).

4.4. Bott-Samelson cycles and their cohomologies. Iterating the con-

struction of leaf spheres gives rise to Bott-Samelson cycles.

For a y ∈ G/T and an ordered sequence (β1 , · · · , βk ) of k roots (in which

repetitions like βi = βj for some 1 ≤ i < j ≤ k may occur), we construct an
oriented twisted product of 2-spheres S(y; β1, · · · , βk ) together with a smooth
map
ϕβ1 ,··· ,βk : S(y; β1, · · · , βk ) → G/T
by induction on k. To start with, S(y; β1) is the leaf sphere specified by (4.3)
with ϕβ1 the natural inclusion S(y; β1 ) ⊂ G/T . Assume next that the map
ϕβ1 ,··· ,βk−1 : S(y; β1, · · · , βk−1) → G/T has been defined. Let
pk−1 : S(y; β1, · · · , βk ) → S(y; β1, · · · , βk−1 )
be the induced bundle of pβk : S(βk ) → G/T via ϕβ1 ,··· ,βk−1 . The cross section
sβk of pβk defines a cross section sk−1 of pk−1 . The obvious bundle map ϕbβ1 ,··· ,βk−1
over ϕβ1 ,··· ,βk−1 followed by fβk gives rise to the desired map
(4.6) ϕβ1 ,··· ,βk = fβk ◦ ϕbβ1 ,··· ,βk−1 .
The final step in the above construction is illustrated by the diagram below.
bβ1 ,··· ,βk−1
ϕ
S(y; β1, · · · , βk ) → S(βk )
fβ k
(4.7) pk−1 ↓↑ sk−1 pβk ↓↑ sβk ց
ϕβ1 ,··· ,βk−1 fβk ◦sβk =id
S(y; β1, · · · , βk−1 ) → G/T → G/T .

Remark 5. Alternatively one has

S(y; β1, · · · , βi) = {(y1, · · · , yi) ∈ G/T × · · · × G/T |
y1 ∈ S(y; β1), · · · , yi ∈ S(y; βi)}.
The projection pi−1 (resp. the cross section si−1 ) is given by
pi−1 (y1 , · · · , yi) = (y1 , · · · , yi−1 )
(resp. si−1 (y1 , · · · , yi−1 ) = (y1 , · · · , yi−1 , yi−1)).
The map ϕβ1 ,··· ,βi : S(y; β1, · · · , βi ) → G/T is seen to be
ϕβ1 ,··· ,βi (y1 , · · · , yi ) = yi .

16
 Definition 6 (cf. 7.2). The map ϕβ1 ,··· ,βk : S(y; β1, · · · , βk ) → G/T is called
the Bott-Samelson cycle associated to the sequence β1 , · · · , βk of roots.

For a subset L = [i1 , · · · , ir ] ⊆ [1, · · · , k], consider the embedding iL :

S(y; βi1 , · · · , βir ) → S(y; β1, · · · , βk ) by
iL (zi1 , · · · , zir ) = (y1 , · · · , yk ),
where yt = zis if is ≤ t < is+1 (cf. Remark 5). Set
S(y; L) = Im{iL : S(y; βi1 , · · · , βir ) → S(y; β1, · · · , βk )}.
Note that the tower of smooth maps
pk−1 pk−2 p1
S(y; β1, · · · , βk ) → S(y; β1, · · · , βk−1) → · · · → S(y; β1),
together with the sections si , 1 ≤ i ≤ k − 1, furnishes the space S(y; β1, · · · , βk )
with the structure of a twisted product of 2-spheres.
Lemma 4.5. Let [S(y; L)] ∈ H2|L| (S(y; β1, · · · , βk )) be the fundamental class
of the cycle S(y; L) ⊂ S(y; β1, · · · , βk ). Then
(1) the set {[S(y; L)] | L ⊆ [1, · · · , k]} constitutes an additive basis for the
graded Z-module H∗ (M).
Further, let xi ∈ H 2 (S(y; β1, · · · , βk )),1 ≤ i ≤ k, be the classes Kronecker dual
to the basis {[S(y; i)] | 1 ≤ i ≤ k} of H2 (S(y; β1, · · · , βk )). Then
(2) the structure matrix of S(y; β1, · · · , βk ) (with respect to x1 , · · · , xk ) is
A = (aij )k×k , where
−βi ◦ βj if i < j;
aij = {
0 if i > j.
(3) the Kronecker pairing in S(y; β1, · · · , βk ) is given by
< xL , [S(y; K)] >= δL,K , L, K ⊆ [1, · · · , k],
where xL = xi1 · · · xir if L = [i1 , · · · , ir ].
Proof. (1) and (3) correspond, respectively, to the items (1) and (3) in
Lemma 3.3. It remains to show (2).
For an 2 ≤ j ≤ k the normal bundle of the section sj−1 : S(y; β1, · · · , βj−1 ) →
S(y; β1, · · · , βj ) is seen to be the the induced bundle ϕ∗β1 ,··· ,βj−1 Eβj , whose Euler
class ej ∈ H 2 (S(y; β1, · · · , βj−1 )) is ϕ∗β1 ,··· ,βj−1 (eβj ) by the naturality property of
Euler classes. That is
ej = ϕ∗β1 ,··· ,βj−1 (eβj ).
On the other hand, by assuming that the structure matrix of S(y; β1, · · · , βk ) is
A = (aij )k×k , we have the expression
ej = −a1,j x1 − · · · − aj−1,j xj−1
by definition 5, Section 3. Since the xt are Kronecker dual to the [S(y; [t]) we
have, for t < j, that
−at,j =< ϕ∗β1 ,··· ,βj−1 (eβj ), [S(y; [t])] >
=< eβj , ϕβ1 ,··· ,βj−1 ∗ i[t]∗ [S(y; βt )] > (by the naturality of <, >)
=< eβj , ϕβt ∗ [S(y; βt)] > (since ϕβ1 ,··· ,βj−1 ◦ i[t] = ϕβt )
= βt ◦ βj (by Lemma 4.1).
This completes the proof of Lemma 4.5.

17
5 The induced action of a Bott-Samelson cycle
Given a sequence β1 , · · · , βk of roots consider the induce ring map
ϕ∗β1 ,··· ,βk : H ∗ (G/T ) → H ∗ (S(y; β1, · · · , βk ))
of the Bott-Samelson cycle ϕβ1 ,··· ,βk . The product in H ∗ (S(y; β1, · · · , βk )) is well
understood (cf. Lemma 3.4 and Lemma 4.5). Our aim is to reduce calcula-
tions in the ring H ∗ (G/T ) (which has been posed to be in question) to that in
H ∗ (S(y; β1, · · · , βk )) via ϕ∗β1 ,··· ,βk .
Recall from Lemma 4.5 that H ∗ (S(y; β1, · · · , βk )) has the additive basis {xL |
L ⊆ [1, · · · , k]}. Therefore, for a u ∈ H 2r (G/T ), one has a unique expression
(5.1) ϕ∗β1 ,··· ,βk (u) = Σ aL (u)xL , aL (u) ∈ Z.
|L|=r;L⊆ [1,··· ,k]
The determination of ϕ∗β1 ,··· ,βk amounts to find the aL (u).

In the special cases where β1 , · · · , βk is a sequence of simple roots, the action

of ϕ∗β1 ,··· ,βk on Schubert classes Pw , w ∈ W , in G/T can be determined completely.
Lemma 5.1. If β1 , · · · , βk is a sequence of simple roots, then the induced
map ϕ∗β1 ,··· ,βk satisfies
P
ϕ∗β1 ,··· ,βk (Pw ) = (−1)l(w) xL .
|L|=l(w),σL =w
This section is devoted to a proof of Lemma 5.1.

5.1. As the first step we express the coefficient aL (u) in (5.1) in terms of the
divided differences θβ . For a subset L = [i1 , · · · , ir ] ⊆ [1, · · · , k] write θL for the
composition θβi1 ◦ · · · ◦ θβir .
Lemma 5.2. If u ∈ H 2r (G/T ), then for L ⊆ [1, · · · , k] with | L |= r
aL (u) = θL (u) ( ∈ H 0 (G/T ) = Z).
Proof. In term of the Kronecker pairing one has
aL (u) =< ϕ∗β1 ,··· ,βk (u), [S(y; L)] > (by (3) of Lemma 4.5)
=< i∗L ◦ ϕ∗β1 ,··· ,βk (u), [S(y; βi1 , · · · , βir )] > (by the naturality of <, >)
=< ϕ∗βi ,··· ,βir (u), [S(y; βi1 , · · · , βir )] > (since ϕβi1 ,··· ,βir = ϕβ1 ,··· ,βk ◦ iL ).
1
This reduces the proof to the special case L = [1, · · · , k]. This will be done by
induction on k. The case k = 1 is easily verified by
ϕ∗β1 (u) = i∗ ◦ fβ∗1 (u) = i∗ (p∗β1 (u) + p∗β1 (θβ1 (u))x)
= θβ1 (u)x (cf. (4.5)).
Assume, finally, that L = [1, · · · , k] (i.e. u ∈ H 2k (G/T )). We compute
ϕ∗β1 ,··· ,βk (u) = ϕ b∗β1 ,··· ,βk−1 (fβ∗k (u)) (by (4.6))
=ϕ b∗β1 ,··· ,βk−1 (p∗βk (u) + p∗βk (θβk (u))x) (by the definition of θβk in (3.3))
= p∗k−1(ϕ∗β1 ,··· ,βk−1 (u)) + p∗k−1 (ϕ∗β1 ,··· ,βk−1 (θβk (u)))xk (by the diagram (4.7))
= p∗k−1(ϕ∗β1 ,··· ,βk−1 (θβk (u)))xk ( ϕ∗β1 ,··· ,βk−1 (u) ∈ H 2k (S(y; β1, · · · , βk−1 )) = 0)
= p∗k−1(θ[1,··· ,k−1] (θβk (u))x1 · · · xk−1 ))xk (by the inductive hypothesis)
= θL (u)xL .
This finishes the proof.

18
 5.2. Bott-Samelson resolution of Xw . We refer to 7.3 and Definition 7
in 7.4 for two equivalent geometric descriptions of Schubert varieties Xw , w ∈ W ,
in G/T . What is really relevant to us is the desingularization of Xw , rather than
Xw itself, originated from Bott-Samelson in the following way (compare [BS2 ,
p.1000] with discussions in 7.1-7.4).

Picture W as the W -orbit {w(α) ∈ L(T ) | w ∈ W } of the regular point α.

For an w ∈ W let Cw be a straight line segment in L(T ) from the Weyl chamber
containing α to w(α) that crosses the planes in D(G) one at a time. Assume that
they are met in the order Lα1 , · · · , Lαk , αi ∈ Φ+ . We have l(w) = k and
w = σαk ◦ · · · ◦ σα1 ,
where σα is the reflection of L(T ) in Lα ∈ D(G) (cf. [Han]).
In term of the sequence of positive roots α1 , · · · , αk specified by the segment
Cw we form the composed map
(5.2) ϕw = w ◦ ϕα1 ,··· ,αk : S(α; α1, · · · , αk ) → G/T ,
where ϕα1 ,··· ,αk is the Bott-Samelson cycle associated to α1 , · · · , αk . The map ϕw
may be appropriately termed as a Bott-Samelson resolution of Xw by the next
result shown in [Du2 , Proposition 3].
Lemma 5.3. The map ϕw is a degree 1 map onto Xw .

¿From Lemma 5.3 we have, for a Schubert class Pw′ ∈ H 2k (G/T ), that
ϕ∗w (Pw′ ) = δw,w′ x1 · · · xk .
On the other hand one has
ϕ∗w (u) = θα1 ◦ · · · ◦ θαk [w ∗(u)]x1 · · · xk , u ∈ H 2k (G/T )
by Lemma 5.2. These imply that
Lemma 5.4. θα1 ◦ · · · ◦ θαk [w ∗ (Pw′ )] = δw,w′ .

5.3. The operator θw : H 2l(w) (G/T ) → H 0 (G/T ) = Z. Let w ∈ W be

with l(w) = k. Recall from Lemma 4.4 that if w = σβ1 ◦ · · · ◦ σβk is reduced
decomposition, then the composition
θw = θβ1 ◦ · · · ◦ θβk : H ∗ (G/T ) → H ∗−2k (G/T )
is well defined (i.e. not depending on the reduced decomposition w = σβ1 ◦· · ·◦σβk
chosen). This enables one to evaluate the operator θw by using any reduced
decomposition of w.

Lemma 5.5. With respect to the additive basis {Pw′ | w ′ ∈ W, l(w ′ ) = k} of

the 2k-dimensional cohomology H 2k (G/T ), the operator θw : H 2k (G/T ) → Z is
given by θw (Pw′ ) = (−1)l(w) δw,w′ .
Proof. Let ϕw = w◦ϕα1 ,··· ,αk : S(α; α1 , · · · , αk ) → G/T be the Bott-Samelson
resolution of Xw given in (5.2). As in Example 1 we put
(5.3) β1 = α1 , β2 = σα1 (α2 ), · · · , βk = σα1 ◦ · · · ◦ σαk−1 (αk ).
Then βi ∈ ∆, and w = σβ1 ◦ · · · ◦ σβk is a reduced decomposition of w. Conse-
quently,
(5.4) θβi = σα∗ 1 ◦ · · · ◦ σα∗ i−1 ◦ θαi ◦ σα∗ i−1 ◦ · · · ◦ σα∗ 1

19
by (5.3) and (5) of Lemma 4.3. From (2) of Lemma 4.4 we have
θw = θβ1 ◦ · · · ◦ θβk
= (θα1 ◦ σα∗ 1 ) ◦ (θα2 ◦ σα∗ 2 ) ◦ · · · ◦ (θαk ◦ σα∗ k ) ◦ w ∗ (by (5.4)).
Substituting in
θαi ◦ σα∗ i = θαi ◦ (Id − eαi θαi ) (by (3) of Lemma 4.3)
= −θαi (by (1), (4) and (2) of Lemma 4.3)
we get θw = (−1)l(w) θα1 ◦ · · · ◦ θαk ◦ w ∗ . The proof is completed by Lemma 5.4.

5.4. Proof of Lemma 5.1. For an ordered sequence {β1 , · · · , βk } of simple

roots consider the induced map
ϕ∗β1 ,··· ,βk : H 2r (G/T ) → H 2r (S(α; β1 , · · · , βk )).
For a Pw ∈ H 2l(w) (G/T ) we have
ϕ∗β1 ,··· ,βk (Pw ) = Σ θL (Pw )xL (by Lemma 5.2)
|L|=l(w);L⊆ [1,··· ,k]
= Σ θσL (Pw )xL (by Lemma 4.4)
l(σL )=|L|=l(w)
l(w)
P
= (−1) xL (by Lemma 5.5).
|L|=l(w),σL =w

6 Proof of the Theorem

The Schubert varieties in a generalized flag manifold G/H can be described by
those in G/T , where H is the centralizer of a 1-parameter subgroup in G.

Let ∆ = {β1 , · · · , βn } be the set of simple roots relative to the regular point
α ∈ L(T ) (cf. 2.1). Assume that b ∈ L(T )\{0} is a point lying in exactly d of
the singular hyperplanes Lβ1 , · · · , Lβn , say b ∈ Lβ1 ∩ · · · ∩ Lβd . We set Gb to be
the centralizer of the 1-parameter subgroup {exp(tb) | t ∈ R} in G. It is well
known that
(1) if d = 0, then Gb is the fixed maximal torus T ;
and in general
(2) every H is conjugated in G to one of the subgroups Gb .
By (2) we may assume that H is of the form Gb for some b taking as the above.
Consequently, T is also a maximal torus of H and the Weyl group W ′ of H,
generated by the reflections σβi , k + 1 ≤ i ≤ n, is a subgroup of W . As in Section
1, we identify the set W/W ′ of left cosets of W ′ in W with the subset of W :
W = {w ∈ W | I(w ′ ) ≥ I(w) for all w ′ ∈ wW ′}.

Consider the standard fibration p : G/T → G/H. From [BGG, §5] we have
Lemma 6.1. If w ∈ W , the map p restricts to a degree 1 map Xw → Xw (H)
between Schubert varieties; and if w ∈
/ W , p∗ [Xw ] = 0.
Consequently, the induced map p∗ : H ∗ (G/H) → H ∗ (G/T ) satisfies
p∗ [Pw (H)] = Pw , w ∈ W .

Proof of the Theorem. For a pair u, v ∈ W assume, in H ∗ (G/H), that

20
 P ′ ′
Pu (H) · Pv (H) = aw w
u,v Pw ′ (H), au,v ∈ Z.
l(w ′ )=l(u)+l(v),w ′ ∈W
∗
Applying the induced ring mapPp we get the equality
′ w′
(6.1) Pu · Pv = awu,v Pw ′ , au,v ∈ Z
l(w ′ )=l(u)+l(v),w ′ ∈W
∗
in H (G/T ) by Lemma 6.1.
Let w = σβ1 ◦ · · · ◦ σβk , βi ∈ ∆ be a reduced decomposition of an w ∈ W , k =
l(u)+l(v), and let Aw = (ai,j )k×k be the associated Cartan matrix of w. Consider
the Bott-Samelson cycle ϕβ1 ,··· ,βk : S(α; β1 , · · · , βk ) → G/T associated to the
sequence β1 , · · · , βk of simple roots. Applying the induced ring map ϕ∗β1 ,··· ,βk to
(6.1) yields in H ∗ (S(α; β1 , · · · , βk ))
Pthat
∗ ′ ∗
ϕβ1 ,··· ,βk [Pu · Pv ] = awu,v ϕβ1 ,··· ,βk [Pw ′ ]
l(w ′ )=l(u)+l(v),w ′ ∈W
= (−1)k aw
u,v x1 · · · xk ,
where the second equality follows from
(−1)k x1 · · · xk if w ′ = w;
ϕ∗β1 ,··· ,βk [Pw′ ] = {
0 if w ′ 6= w
by Lemma 5.1. On the other hand P P
ϕ∗β1 ,··· ,βk [Pu · Pv ] = [((−1)l(u) xL )((−1)l(v) xK )]
|L|=l(u) |K|=l(v)
σL =u σK =v
(again) by Lemma 5.1. Summarizing we get in P H 2k (S(α; βP
1 , · · · , βk )) that
k w l(u)+l(v)
(−1) au,v x1 · · · xk = (−1) ( xL )( xK ).
|L|=l(u) |K|=l(v)
σL =u σK =v
Evaluating both sides on the orientation class [S(α; β1 , · · · , βk )] and noting that
< x1 · · · xk , [S(α; β1, · · · , βk )] >= 1 and k = l(u) + l(v), we get by Lemma 3.4
that P P
aw
u,v = TAw [( xL )( xK )].
|L|=l(u) |K|=l(v)
σL =u σK =v
This completes the proof of the Theorem.

7 Historical remarks
7.1. K-cycles in a flag manifold G/T. In 1954 R. Bott constructed a Morse
function on G/T and showed that G/T was a cell complex with cells in the even
dimensions only. They turn out to be so-called K-cycles of Bott-Samelson [BS2 ]
formulated in the following plausible way.
For each root β ∈ Φ let Kβ ⊂ G be the stabilizer of the singular plane
Lβ ∈ D(G) under the adjoint action of G on L(G). For an ordered sequence
{β1 , · · · , βk } of roots one forms the products K(β1 , · · · , βk ) = Kβ1 × · · · × Kβk .
Since T ⊂ Kβi for each i, the group T (k) = T × · · · × T (k factors) acts on
K(β1 , · · · , βk ) from the right by
(c1 , · · · , ck ) · (t1 , · · · , tk ) = (c1 t1 , t−1 −1
1 c2 t2 , · · · , tk−1 ck tk ).

21
This defines K(β1 , · · · , βk ) as a T (k)-principal bundle, whose base manifold is
called Kβ1 ×T · · · ×T Kβk . The point in the base corresponding to (c1 , · · · , ck ) ∈
K(β1 , · · · , βk ) is denoted by [c1 , · · · , ck ]. The K-cycle associated to the sequence
{β1 , · · · , βk } of roots is the map fβ1 ,··· ,βk : Kβ1 ×T · · · ×T Kβk → G/T by
fβ1 ,··· ,βk [c1 , · · · , ck ] = Adc1 ···ck (α).
Certain K-cycles were selected to describe the stable manifolds of a perfect
Morse function on G/T , hence provide an explicit additive basis for the homology
H∗ (G/T ) [BS2 ]. Picture W as the W -orbit {w(α) ∈ L(T ) | w ∈ W } of the regular
point α. For each w ∈ W let Cw be a straight line segment in L(T ) from the
Weyl chamber containing α to w(α) that crosses the planes in D(G) one at a
time, and assume that they are met in the order Lα1 , · · · , Lαk , αi ∈ Φ+ . Let
Γw = Kα1 ×T · · · ×T Kαk and define gw : Γw → G/T to be the composition
w ◦ fα1 ,··· ,αk . It was shown in [BS2 ] that
Proposition 1. The set of cycles {gw∗ [Γw ] ∈ H∗ (G/T ) | w ∈ W } is a basis
of H∗ (G/T ).
7.2. Bott-Samelson cycles in an isoparametric submanifold. The
embedding ϕ : G/T → L(G) given by the adjoint representation at the beginning
of Section 4 defines G/T as an isoparametric submanifolds in the Euclidean space
L(G) [HPT].
In general, associated to any isoparametric submanifold M in an Euclidean
space RN there are also concepts like (finite) Coxeter group, root system and
Dynkin diagram (marked with multiplicities). In order to generalize Bott-Samelson’s
above cited result to such more general spaces which are also of historical inter-
ests in differential geometry, Hsiang-Palais-Terng introduced in [HPT] the space
S(y; α1, · · · , αk ) as well as the map ϕβ1 ,··· ,βk : S(y; β1, · · · , βk ) → M under the
name “Bott-Samelson cycles”in the same way as that given in Remark 5, Section
4. The construction of these cycles uses only the integrability of the tangent dis-
tributions on M (which are also indexed by sequences of positive roots relative
to a non-focal point α ∈ RN of M) while the groups Kβ ’s required to define
K-cycles no long always exist in this more general situation.
The idea of Bott-Samelson cycles does generalize the K-cycles of Bott-Samelson
in the following sense.
Proposition 2 (cf.[Du2 , Lemma 8]). If M = G/T , there is an orientation
preserving diffeomorphism g : Kβ1 ×T · · · ×T Kβk → S(α; β1, · · · , βk ) so that the
following mapping triangle commutes
Kβ1 ×T · · · ×T Kβk
fβ1 ,··· ,βk
g↓ ց
ϕβ1 ,··· ,βk
S(α; β1, · · · , βk ) → G/T .
7.3. Schubert varieties. Let K be a linear algebraic group over the field
C of complex numbers, and let B ⊂ K be a Borel subgroup. The homogeneous

22
variety K/B is a projective variety on which the group K acts by left translations.
Historically, Schubert varieties were introduced in term of the orbits of B action
on K/B.
Let T be a maximal torus containing in B and let N(T ) be the normalizer of
T in K. The Weyl group of K (relative to T ) is W = N(T )/T . For an w ∈ W
take an n(w) ∈ N(T ) such that its residue class mod T is w.
The following result was first discovered by Bruhat for classical Lie groups K
in 1954, and proved to be the case for all reductive algebraic linear groups by
Chevalley [Ch2 ] in 1958.
Proposition 3. One has the disjoint union decomposition
K/B = ∪ Bn(w) · B
w∈W
in which each orbit Bn(w) · B is isomorphic to an affine space of complex dimen-
sion l(w).
The Zariski closure of the open cell Bn(w) · B in K/B, denoted by Xw , is
called the Schubert variety associated to w.
7.4. Bott-Samelson desingularizations of Schubert varieties. For a
compact connected Lie group G with a maximal torus T let K be the complexi-
fication of G, and let B be a Borel subgroup in K containing T . It is well known
that the natural inclusion G → K induces an isomorphism G/T = K/B. Con-
versely, the reductive algebraic linear groups are exactly the complexifications of
the compact real Lie groups (cf. [Ho]).
It follows now from Proposition 1 and 3 that the homology H∗ (G/T ) has
two canonical additive bases: the first of these is given by the K-cycles of Bott-
Samelson; the second consists of Schubert varieties, and both of them are indexed
by the Weyl group of G.
Before thinking of the problem on the relationship between these two bases
of H∗ (G/T ) as “a natural one”, one should bear in mind that the first basis was
constructed to provide the stable manifolds of a perfect Morse function on G/T ,
while the second arose from the jumbled efforts through centuries of great many
mathematicians who have contributed to lay the fundation of algebraic inter-
section theory for the Schubert’s enumerative calculus [K1 ], [K2 ]. The following
result was obtained by Hansen in 1973 [Han].
Proposition 4. Under the natural isomorphism G/T = K/B, the K-cycle
gw : Γw → G/T of Bott-Samelson is a degree 1 map onto the Schubert variety
Xw .
Combining Proposition 2 with Proposition 4, we have the following alternative
definition of Schubert varieties in G/T (for compact G) without resorting to the
complexification of G.
Given an w ∈ W let Cw be a straight line segment in L(T ) from the Weyl
chamber containing α to w(α) that crosses the planes in D(G) one at a time.
Assume that they are met in the order Lα1 , · · · , Lαk , αi ∈ Φ+ .

23
 Definition 7. The Schubert variety in G/T associated to an w ∈ W is
Xw = Im ϕw , where ϕw is the composition
ϕw = w ◦ ϕα1 ,··· ,αk : S(α; α1 , · · · , αk ) → G/T .
This is the version of the descriptions of Schubert varieties that we have made
use of in Section 5 and 6.

7.5. Divided differences. The inclusion T → G of the maximal torus

induces a fibration
c
G/T ֒→ BT → BG,
where BH denotes the classifying space of a Lie group H , and where the fibre
inclusion c is equivariant with respect to the standard W -action on both G/T
and BT [B].
Let H ∗ (X; R) be the cohomology ring (resp. algebra) of a space X with
coefficients in the field R of reals. We recall the classical result due to Borel [B].
Proposition 5. The map c induces an W -equivariant surjective homomor-
phism of algebras
c∗ : H ∗ (BT ;R) → H ∗ (G/T ; R)
with kernel H + (BT ; R)W , the ideal in H ∗ (BT ; R) generated by W -invariants in
positive degrees.
The induced map c∗ , playing a key role in this result, is well known as the
Borel’s characteristic map.

The infinite complex projective space CP ∞ serves both as the classifying space
of the circle group S 1 and the Eilenberg-MacLane space K(Z, 2). Keeping this
in mind one get two ingredients from each root α ∈ Φ.
(1) The reflection σα on L(T ) in the hyperplane Lα preserves the unit lat-
tice Λ = exp−1 (e), hence induces in successive manner an automorphism of the
torus T = L(T )/Λ; a diffeomorphism of the classifying space BT , and finally, an
induced automorphism σα∗ of the ring H ∗ (BT ; R).
(2) the co-root α∗ : L(T ) → R related to the root α ∈ Φ in the fashion
α∗ (v) = (αi , v), v ∈ L(T ) satisfies α∗ (Λ) ⊂ Z, hence induces successively a
homomorphism T → S 1 , a map between classifying space BT → BS 1 = CP ∞ ,
an finally a 2-cocycle [α] ∈ H 2 (BT ; R).
In terms of these, an additive operation Aα : H ∗ (BT ; R) → H ∗ (BT ; R) of degree
−2 can be defined as follows.
∗
Aα (f ) = f −σ[α]α (f ) , f ∈ H ∗ (BT ; R).

The operator Aα on H ∗ (BT ; R), known as the divided difference operators

associated to the root α, was introduced independently by Bernstein et al [BGG]
and Demazure [De1 ] in 1973. These operators possess the same properties as
that of the operators θα on H ∗ (G/T ) (compare [BGG, 3.3 Lemma] with (2) of
Lemma 4.3 in Section 4). In fact, Borel’s characteristic map gives rise to a linkage
between these two sets of operators.

24
 Proposition 6. For each α ∈ Φ, the two operators Aα and θα satisfy the
commutative diagram
c∗
H ∗ (BT ; R) → H ∗ (G/T, R) = H ∗ (G/T ) ⊗ R
Aα ↓ ↓ θα ⊗ 1
c ∗
H ∗ (BT ; R) → H ∗ (G/T, R) = H ∗ (G/T ) ⊗ R.
However, by introducing the operators θα in the much more general context
of spherical represented involutions (cf. 3.2), and by developing their basic prop-
erties (Lemma 3.2) in a way independent of the Aα and the Borel characteristic
map, these operators not only can act directly on integral ring H ∗ (G/T ) (in
which we are actually interested), but also become applicable in the study of
intersection theory of isoparametric submanifolds, transnormal submanifolds, as
well as their focal manifolds [HPT], [R]. In the context of Morse theory, these
families of manifolds are seen as natural generalizations of flag manifolds in the
sense that generalized Schubert cycles (cf. [HPT, p.449]) can be constructed for
distance functions on these manifolds and which are also indexed by the cosets
of certain finite groups. In these general settings there are no analogue of Cartan
subalgebra (and consequently, Borel’s characteristic maps) available.
7.6. Positively multiplying Schubert classes (continuing from Remark
1, Section 2). Let S be a twisted products of 2-spheres of rank k with structure
matrix A, and let B = {xI | I ⊂ [1, · · · , k]} be the additive basis for H ∗ (S)
specified by Lemma 3.3. Product among the basis elements of complementary
dimensions yields the numbers cI,J ∈ Z with
xI · xJ = cI,J x1 · · · xk , | I | + | J |= k.
Equivalently, cI,J = TA (xI · xJ ).
The advantages to work with the basis B are obvious. It consists of the
elements Kronecker dual to the ready-made geometric cycles S(I) in S (Lemma
3.3). With respect to this basis the action of Bott-Samelson cycles admits a simple
description (Lemma 5.1). As a result the geometric essence of the Theorem is
transparent. On the other hand, the product in S between these basis elements
are not always positive in the sense that cI,J < 0 may occur (cf. Example 3).
This explains the reason that our formula falls short of positivity.
The classical Richardson-Littlewood rule for multiplying Schubert classes in
the Grassmannian has the merit to meet the standard of positivity. However, it
is difficult to summarize the rule into an explicit formula required by effective
computation. Its lengthy and technical statement (cf. [St, p.228]) form a sharp
contrast with the conciseness of the Theorem.
Concerning the geometric origin of the problem, effective computability should
be granted with the first priority among various standards on the multiplicative
rules [K1 ], [K2 ]. Therefore, it is natural to make inquires about a unified solution
to the problem in its deserved simplicity and natural generality.

25
 References

[B] A. Borel, Topics in the homology theory of fiber bundles, Berlin, Springer,
1967.
[BGG] I. N. Bernstein, I. M. Gel’fand and S. I. Gel’fand, Schubert cells and
cohomology of the spaces G/P, Russian Math. Surveys 28 (1973), 1-26.
[Bi] S. Billey, Kostant polynomials and the cohomology ring for G/B, Duke
J. Math. 96, No.1(1999), 205-224.
[BH] S. Billey and M. Haiman, Schubert polynomials for the classical groups,
Journal of AMS, 8(2)(1995), 443-483.
[BS1 ] R. Bott and H. Samelson, The cohomology ring of G/T, Nat. Acad.
Sci. 41 (7) (1955), 490-492.
[BS2 ] R. Bott and H. Samelson, Application of the theory of Morse to sym-
metric spaces, Amer. J. Math., Vol. LXXX, no. 4 (1958), 964-1029.
[CF] I. Ciocan-Fontanine, On quantum cohomology rings of partial flag vari-
eties, Duck Math. J., Vol.98, No. 3(1999), 485-524.
[Ch1 ] C. Chevalley, La théorie des groupes algébriques, Proc. 1958 ICM,
Cambridge Univ. Press, 1960, 53-68.
[Ch2 ] C. Chevalley, Sur les Décompositions Celluaires des Espaces G/B, in
Algebraic groups and their generalizations: Classical methods, W. Haboush ed.
Proc. Symp. in Pure Math. 56 (part 1) (1994), 1-26.
[De1 ] M. Demazure, Invariants symétriques entiers des groupes de Weyl et
torsion, Invent. Math. 21 (1973), 287-301.
[De2 ] M. Demazure, Désingularization des variétés de Schubert généralisées,
Ann. Sci. École. Norm. Sup. (4) 7(1974), 53-88.
[Du1 ] H. Duan, Some enumerative formulas on flag varieties, Communication
in Algebra, 29 (10) (2001), 4395-4419.
[Du2 ] H. Duan, The degree of a Schubert variety, to appear in Adv. in Math
(Available online: http://www.sciencedirect.com /science/journal/00018708).
[DZ1 ] H. Duan and Xuezhi Zhao, Steenrod operations on Schubert classes,
arXiv: math.AT/0306250.
[DZ2 ] H. Duan and Xuezhi Zhao, A program for multiplying Schubert classes,
in preparation.
[DZZ] H. Duan, Xu-an Zhao and Xuezhi Zhao, Cartan matrix and enumerative
calculus, preprint.
[FP] W. Fulton and P. Pragacz, Schubert Varieties and Degeneracy Loci,
LNM 1689, Springer, 1998.
[Han] H.C. Hansen, On cycles in flag manifolds, Math. Scand. 33 (1973),
269-274.
[Ho] G. Hochschild, The structure of Lie groups, Holden-Day, Inc., San Francis
co-London-Amsterdam, 1965.
[HPT] W. Y. Hsiang, R. Palais and C. L. Terng, The topology of isoparametric
submanifolds, J. Diff. Geom., Vol. 27 (1988), 423-460.

26
 [Hu] J. E. Humphreys, Introduction to Lie algebras and representation theory,
Graduated Texts in Math. 9, Springer-Verlag New York, 1972.
[K1 ] S. Kleiman, Problem 15. Rigorous fundation of the Schubert’s enumera-
tive calculus, Proceedings of Symposia in Pure Math., 28 (1976), 445-482.
[K2 ] S. Kleiman, Intersection theory and enumerative geometry: A decade in
review, Algebraic Geometry, Bowdoin 1985 (Spancer Bloch, ed.), Proc. Sympos.
Pure Math., vol. 46, Part 2, AMS. 1987, 321-370.
[KK] B. Kostant and S. Kumar, The nil Hecke ring and the cohomology of
G/P for a Kac-Moody group G, Adv. Math. 62(1986), 187-237.
[LR] D. E. Littlewood and A. R. Richardson, Group characters and algebra,
Philos. Trans. Roy. Soc. London. 233(1934), 99-141.
[LS] A. Lascoux and M.-P. Schützenberger, Polynômes de Schubert, C. R.
Acad. Sci. Paris 294(1982), 447-450.
[M] I. G. Macdonald, Symmetric functions and Hall polynomials, Oxford
Mathematical Monographs, Oxford University Press, Oxford, second ed., 1995.
[R] S. A. Robertson, Smooth curves of constant width and transnormality,
Bull. London Math. Soc. 16, no.3(1984), 264-274.
[S] F. Sottile, Four entries for Kluwer encyclopaedia of Mathematics, arXiv:
Math. AG/0102047.
[St] R. Stanley, Some combinatorial aspects of the Schubert calculus, Combi-
natoire et représentation du groupe symétrique, Strasbourg (1976), 217-251.

27

View publication stats