SCHUBERT CELLS AND COHOMOLOGY OF

THE SPACES G/P

I. N. Bernstein, I. M. Gel'fand, S. I. Gel'fand

We study the homological properties of the factor space G/P, where G is a complex semi-
simple Lie group and Ρ a parabolic subgroup of G. To this end we compare two descriptions
of the cohomology of such spaces. One of these makes use of the partition of G/P into
cells (Schubert cells), while the other consists in identifying the cohomology of G/P with
certain polynomials on the Lie algebra of the Cartan subgroup Η of G. The results obtained
are used to describe the algebraic action of the Weyl group W of G on the cohomology
of G/P.

Contents
Introduction 1
§ 1. Notation, preliminaries, and statement of the main results . . . 3
§ 2. The ordering on the Weyl group and the mutual disposition
of the Schubert cells 6
§ 3 . Discussion of the ring of polynomials on t) 10
§4. Schubert cells 19
§ 5. Generalizations and supplements 22
References 25

Introduction
Let G be a linear semisimple algebraic group over the' field C of complex
numbers and assume that G is connected and simply-connected. Let Β be
a Borel subgroup of G and X = G/B the fundamental projective space of G.
The study of the topology of X occurs, explicitly or otherwise, in a
large number of different situations. Among these are the representation
theory of semisimple complex and real groups, integral geometry and a
number of problems in algebraic topology and algebraic geometry, in which
analogous spaces figure as important and useful examples. The study of
the homological properties of G/P can be carried out by two well-known
methods. The first of these methods is due to A. Borel [ 1 ] and involves
the identification of the cohomology ring of X with the quotient ring of
the ring of polynomials on the Lie algebra i) of the Cartan subgroup
2 /. Ν. Bernstein, I. M. Gel'fand, S. I. Gel'fand

Η c G by the ideal generated by the ^-invariant polynomials (where W is

the Weyl group of G). An account of the second method, which goes back
to the classical work of Schubert, is in Borel's note [2] (see also [3]); it
is based on the calculation of the homology with the aid of the partition
of X into cells (the so-called Schubert cells). Sometimes one of these
approaches turns out to be more convenient and sometimes the other, so
naturally we try to establish a connection between them. Namely, we must
know how to compute the correspondence between the polynomials
figuring in Borel's model of the cohomology and the Schubert cells.
Furthermore, it is an interesting problem to find in the quotient ring of
the polynomial ring a symmetrical basis dual to the Schubert cells. These
problems are solved in this article. The techniques developed for this
purpose are applied to two other problems. The first of these is the
calculation of the action of the Weyl group on the homology of Λ' in a
basis of Schuberts cells, which turns out to be very useful in the study of
the representations of the Chevalley groups.
We also study the action of W on X. This action is not algebraic (it
depends on the choice of a compact subgroup of G). The corresponding
action of W on the homology of X can, however, be specified in algebraic
terms. For this purpose we use the trajectories of G in X X X, and we
construct explicitly the correspondences on X (that is, cycles in X X X)
that specify the action of W on Η^ (X, Z). The study of such correspon-
dences forms the basis of many problems in integral geometry.
At the end of the article, we generalize our results, to the case when Β
is replaced by an arbitrary parabolic subgroup Ρ C G. When G = GL(n)
and G/P is the Grassmann variety, analogous results are to be found in [4].
B. Kostant has previously found other formulae for a basis of
H*(X, Ζ), Χ = G/B, dual to the Schubert cells. We would like to express
our deep appreciation to him for drawing our attention to this series of
problems and for making his own results known to us.
The main results of this article have already been announced in [13].
We give a brief account of the structure of this article. At the
beginning of § 1 we introduce our notation and state the known results
on the homology of X = G/B that are used repeatedly in the paper. The
rest of § 1 is devoted to a statement of our main results.
In § 2 we introduce an ordering on the Weyl group W of G that arises
naturally in connection with the geometry of X, and we investigate its
properties.
§ 3 is concerned with the ring R of polynomials on the Lie algebra §
of the Cartan subgroup He G. In this section we introduce the functionals
Dw on R and the elements Pw in R and discuss their properties.
In §4 we prove that the elements Dw introduced in §3 correspond to
the Schubert cells of X.
 Schubert cells and cohomology of the spaces G/P 3

§ 5 contains generalizations and applications of the results obtained, in

particular, to the case of manifolds X{P) = G/P, where Ρ is an arbitrary
parabolic subgroup of G. We also study in § 5 the correspondences on X
and in particular, we describe explicitly those correspondences that specify
the action of the Weyl group W on the cohomology of X. Finally, in this
section some of our results are put in the form in which they were earlier
obtained by B. Kostant, and we also interpret some of them in terms of
differential forms on X.

§1. Notation, preliminaries, and statement of the main results

We introduce the notation that is used throughout the article.
G is a complex semisimple Lie group, which is assumed to be connected
and simply-connected;
Β is a fixed Borel subgroup of G;
X = G/B is a fundamental projective space of G;
Ν is the unipotent radical of B\
Η is a fixed maximal torus of G, Η c B;
@ is the Lie algebra of G; £) and 9i are the subalgebras of @ correspond-
ing to Η and N;
f>* is the space dual to E);
A C £)* is the root system of t) in @ ;
Δ+ is the set of positive roots, that is, the set of roots of i) in 9i,
Δ_ = -Δ+, Σ C Δ + is the system of simple roots;
W is the Weyl group of G; if γ ξ Δ, then σν : ί)*->- ή* is an element
of W, a reflection in the hyperplane orthogonal to y. For each element1
w ζ. W = Norm(H)/H, the same letter is used to denote a representative
of w in Norm (H) c G.
l(w) is the length of an element w e W relative to the set of generators
{σα, α ζ Σ} of W, that is, the least number of factors in the decomposition
(1) w = % a a 2 . . . σ α; , at ζ Σ.
The decomposition (1), with / = /(w), is called reduced; s € W is the
unique element of maximal length, r - l{s);
N_ - sNs'1 is the subgroup of G "opposite" to N.
For any w G W we put Nw = w Λ^νν"1 η TV.
HOMOLOGY AND COHOMOLOGY OF THE SPACE X. We give at this
point two descriptions of the homological structure of X. The first of
these (Proposition 1.2) makes use of the decomposition of X into cells,
while the second (Proposition 1.3) involves the realization of two-dimen-
sional cohomology classes as the Chern classes of one-dimensional bundles.
We recall (see [5]) that Nw = w N_w~* η Ν is a unipotent subgroup of

Norm Η is the normalizer of Η in G.

4 /. Ν. Bernstein, I. Μ. GeVfand, S. I. Gel'fand

G of (complex) dimension l(w).

1.1. PROPOSITION (see [5]), Let ο ζ X be the image of Β in X. The
open and closed subvarieties Xw - Nwo C X, w ζ W, yield a decompo-
sition of X into N-orbits. The natural mapping Nw -* Xw (n H- nwo) is
an isomorphism of algebraic varieties.
Let Xw be the closure 1 of Xw in X, [Xw] 6 Hii(W) 0*Ίι;,Ζ) the
fundamental cycle of the complex_ algebraic variety Xw and
sw € H2hw) (X,Z) the image of [Xw] under the mapping induced by the
embedding X~w ·—>· X.
1.2. PROPOSITION (see [2]). The elements sw form a free basis of

We now turn to the other approach to the description of the cohomo^

logy of X. For this purpose we introduce in Ij the root system
{Hv, y ς Δ} dual to Δ. (This means that oy\ = χ - χ (Hy)y for all
χ € ί ) * > τ € Δ ) . We denote by tjQ c= i) the vector space over Q spanned
by the Hy. We also set ί)ζ = {χ € ί) * Ι χ ( # γ ) 6 Ζ for all γ € Δ} and
i)Q - ί)2 β Z Q.
Let /? = S I(£)Q) be the algebra of polynomial functions on i)Q with
rational coefficients. We extend the natural action of H' on ή* to Λ. We
denote by / the subring of W-invariant elements in R and set
/+ ={/e/|/(0) = 0}, J = I+R.
We construct a homomorphism a: /?-»• H*{X, Q) in the following way.
First let χ 6 ^ζ· Since G is simply-connected, there is a character
θ 6 Mor (H, C*) such that θ (exp h) = exp χ(Λ), h G t). We extend θ
to a character of 5 by setting θ (η) = 1 for η ζ Ν. Since G-»-X is a
principal fibre space with structure group B, this θ defines a one-
dimensional vector bundle Εχ on X. We set α! (χ) = cx, where
cx € IP(X, Z) is the first Chern class of is x . Then a t is a homomorph-
ism of t|z into IP(X, Z), which extends naturally to a homomorphism
of rings a: R -* H*(X, Q).
Note that W acts on the homology and cohomology of X. Namely,
let Κ c G be a maximal compact subgroup such that Τ = Κ η // is a
maximal torus in AT. Then the natural mapping K/T^-X is a homeomorph-
ism (see [ 1 ]). Now W acts on the homology and cohomology of X in
the same way as on KIT.
1.3. PROPOSITION ([1], [8]). (i) The homomorphism a commutes
with the action of W on R and H* (X, Q).
(ii) Ker a = J, and the natural mapping a: R/J -* H* {X, Q) is an
isomorphism.
In the remainder of this section we state the main results of this article.
The integration formula. We have given two methods of describing the
1
As Xw is an open and closed variety, its closure in the Zariski topology is the same as in the ordinary
topology.
 Schubert cells and cohomology of the spaces G/P 5

cohomological structure of X. One of the basic aims of this article is to

establish a connection between these two approaches. By this we under-
stand the following. Each Schubert cell sw 6 //* (X, Z) gives rise to a
linear functional Dw on R according to the formula

Dw(f) = (sw, a (/))

(where < , > is the natural pairing of homology and cohomology). We

indicate an explicit form for Dw.
For each root γ 6 Δ, we define an operator Ay : R -+ R by the formula

(that is, Ayf(h) = [/(h) - f(oyh)\ly(h) for all h£\). Then we have the
following proposition.
PROPOSITION. Let w = oai . . . σ α ; , a f £ Σ. If l{w) < /, then
^a1 • • • Aai = 0. // l(w) - I, then the operator ΑΛι . . . Aa[ depends
only on w and not on the representation of w in the form
w = σαι . . . σ α/ ; we put Aw = AUi . . . Aaj.
This proposition is proved in § 3 (Theorem 3.4).
The functional Dw is easily described in terms of the Aw : we define for
each w e W another functional Dw on R by the formula Dw f = Aw /(0).
The following theorem is proved in §4 (Theorem 4.1).
THEOREM .Dw = Uw for all w e W.
We can give another more explicit description of Dw (and thus of Dw).
To do this, we write ιυλ %• w2, Wi., w2 6 H^, γ 6 Δ+, to express.the fact
that Wi = OyW2 and / (w2) = I (u)i) = / + 1.
THEOREM. Let w € W, l(w) = I.
(i) If f G R is a homogeneous polynomial of degree k Φ I, then
Dw{f) = 0.
(ii) If X l , . . ., χ, €&Q ,then Dw(Xl . . . . Xl ) = Σ x,(// 7 i ) . . . Χι(ΗΊι),
where the sum is taken over all chains of the form

1
e = w0 ^ ifi -^ . . . ^ w, = w'

(see Theorem 3.12 (i), (v)).

The next theorem describes the basis of H*{X, Q) dual to the basis
{sw | w 6 W) of H*(X, Z). We identify the ring R = R/J with H*(X, Q)
by means of the isomorphism α of Proposition 1.3. Let {Pw | w ζ W} be
the basis of R dual to the basis {sw \ w 6 W) of H*(X, Z). To specify Pw,
we note that the operators Aw : R -*• R preserve the ideal J c R (lemma
3.3 (v)), and so the operators Aw : R -> R are well-defined.
THEOREM, (i) Let s ζ W be the element of maximal length, r - l(s)
Then Ps = pT\r\ (mod /) = \W\~l [] Ί (mod / ) , (where ρ ζ 1)5 is half
6 /. Ν. Bernstein, I. M. Gel'fand, S. I. Gel'fand

the sum of the positive roots and \W\ is the order of W)

(ii) If w e W, then Pw = Aw.isPs (see Theorem 3.15, Corollary 3.16,
Theorem 3.14(i)).
Another expression for the Pw has been obtained earlier by B. Kostant
(see Theorem 5.9).
The following theorem gives a couple of important properties of the Pw.
THEOREM (i). Let χ 6 %, w ζ W. Then %-Pw ~Σ w%(Hy) Pw.

(see Theorem 3.14 (ii)).

(ii) Let & : H^X, Q) -*• H*(X, Q) be the Poincare duality. Then
& (sw) = α (Pws) (see Corollary 3.19).
THE ACTION OF THE WEYL GROUP. The action of W on H* (X, Q)
can easily be described using the isomorphism a: R/J -*• H*(X, Q), but we
are interested in the problem of describing the action of W on the basis
{,„) of H,(X, Q).
THEOREM. Let a 6 2 , » £ W. Then aasw = - sw if l(woa) = l(w) - 1
and aasw == — s^ + 21 wa(Hy) sw-, if l(woa) = l(w) + 1 (see
ν
to'—*-woa
Theorem 3.12 (iv)).
In § 5 we consider some applications of the results obtained. To avoid
overburdening the presentation, we do not make precise statements at
this point. We merely mention that Theorem 5.5 appears important to us,
in which a number of results is generalized to the case of the varieties
X(P) = G/P (P being an arbitrary parabolic subgroup of G), and also
Theorem 5.7, in which we investigate certain correspondences on X.

§2. The ordering on the Weyl group and the mutual

disposition of the Schubert cells
2.1 DEFINITION (i) Let n^ , w2 ξ W, y 6 Δ+. Then w^ ^ w2 indicates
the fact that aywt = w2 and l(w2) = l(wi) + 1.
(ii) We put w < w- if there is a chain

It is helpful to picture W in the form of a directed graph with edges drawn

in accordance with Definition 2.1 (i).
Here are some properties of this ordering.
2.2 LEMMA. Let w = aai . . . aai be the reduced decomposition of an
element w 6 W. We put 7,· = aai . . . σα._ι (ο,·). Then the roots
7 i , · · ·, 7i are distinct and the set {7i, · · ·, Vn} coincides with Δ+ η wA-.
This lemma is proved in [ 6 ] .
2.3 COROLLARY, (i) Let w = aai . . . a^ be the reduced decompo-
sition and let y e Δ+ be a root such that Wl γ e Δ-. Then for some i
(2) σ ν σ α ι . . . σ α . = σ α ι . . . σβ._4.
 Schubert cells and cohomology of the spaces G/P 7

(ii) Let w ζ W, y £ Δ+.Then I (w) < / (σyw),if and only ifw~xy ζ Δ + .

P R O O F (i) F r o m L e m m a 2.2 w e d e d u c e t h a t y = ααι . . . oa (a,) for
some i, a n d ( 2 ) follows.
(ii) If w~xy 6 Δ_, t h e n b y ( 2 ) ayw = σ β ι . . . aa._t σ α . + ι . . . σαν t h a t is
l(ayw) < l(w). I n t e r c h a n g i n g w a n d a 7 u ; , w e see t h a t if w~ly ζ Δ + , t h e n
l(ayw).
2.4 LEMMA. Letwu w2 ζ W, α ζ Σ, y ζ A+,and y Φ a. Let y = oay. If

then

σα»2·

Conversely, (3) follows from (4).

PROOF. Since α ξ, Σ and y Φ a, we have γ' = aay 6 Δ+. It is there-
fore sufficient to show that l{aaw2) > l(w2) ~ l(wl). This follows from
Corollary 2.3, because aaw2 = σ 7 ΌΊ and (σ α ^ 2 )- χ γ' = wl^ov' = w^y 6 Δ_
by (3). The second assertion of the lemma is proved similarly.
2.5 LEMMA. Let w, w' β W, a £ Σ and assume that w < w'. Then
a) either aaw < w or aaw < aaw',
b) either w < aaw or aaw < aaw'.
PROOF a) Let

w = Wl^w2-*- ... -v wh = H/.

We proceed by induction on &. If aaw < w or σα«; = w2 , the assertion is

obvious. Let w < aaw, oaw Φ w2- Then aaw < oaw2 by Lemma 2.4. We
obtain a) by applying the inductive hypothesis to the pair (w2, w').
b) is proved in a similar fashion.
2.6. COROLLARY. Let α ζ Σ, wt J ^ w[, w2 J ^ w'z. If one of the
elements Wy, w[ is smaller (in the sense of the above ordering) than one
of w2, w2, then wx < w2 < w'2 and W\ < w\ < w'2.
The property in Lemma 2.5 characterizes the ordering <. More precisely,
we have the following proposition:
2.7 PROPOSITION. Suppose that we are given a partial ordering w -\ w
on W with the following properties:
a) / / α ζ 2 ' w 6 W with l(aaw) = l(w) + 1, then w Η aaw.
b) / / w -\ w', a 6 Σ, then either aaw -f w' or aaw - | aaw'.
Then w -\ w if and only if w < w'.
8 /. Λ: Bernstein, I. M. Gel'fand, S. I. Gel'fand

PROOF. Let s be the element of maximal length in W. It follows from

a) that e —| w -\ s tor all w € W.
I. We prove that w < w' implies that w π w'. We proceed by induction
on /(«/). If /(«/) = 0, then w' = e, w = e and so ID Η w'. Let /(«/) > 0
and let α 6 Σ be a root such that l(aaw') = l(w') - 1. Then by Lemma
2.5 a), either aaw < aaw or w < aaw .
(i) w < σαΐί/ =* w -\ oaw (by the inductive hypothesis), => u; Η Η/.
(using a)).
(ii) oaw < oaw' => o a u; Η σ α ιυ' (by the inductive hypothesis), =» either
w; -I o a it/ or w -\ w' (applying b) to the pair (oaw, o^w')), => w -\ w'.
II. We now show that w -\ w' implies that w < w . We proceed by
backward induction on /(u;). If /(u;) = r - l(s), then w = s, w' = s, and so
w < IU'. Let /(ii>) < r and let a. be an element of Σ such that
l(o2w) = l(w) + 1. By b) either aaw -\ w or a^w -\ oaw .
(i) σαιυ -\ w =>• σαΐϋ < u/ (by the inductive hypothesis) => u; < w;'.
(ii) σαιιι Η σαΐί;' =* σαΐί; < aaw' =* w < w' (by Corollary 2.6).
Proposition 2.7 is now proved.
2.8 P R O P O S I T I O N . Let w 6 W and let w = aOi . . . σ α / be the reduced
decomposition of w.
a) // 1 < h < h < • • • < ik < I and

b) // u/ < in, i/?e« w can be represented in the form (5) for some
indexing set {ij).
c) // w -> w, then there is a unique index i, 1 < / < /, such that

(6) u> = σ α ι . . . σβί_

PROOF. Let us prove c). Let w' 3 u>. Then by Lemma 2.2 there is at
least one index i for which (6) holds. Now suppose that (6) holds for two
indices i, j , i < j . Then oa.+i • • • %· = σ α, • · · o a y _ i . Thus,
σ α . . . . σα. = σ α ; + ι . . . oay_, , which contradicts the assumption that the
decomposition w = aOi . . . oai is reduced.
b) follows at once from c) if we take into account the fact that the
decomposition (6) is reduced. We now prove a) by induction on /. We
treat two cases separately.
(i) ί'ι > 1. Then by the inductive hypothesis w' < aa% . . . σαι, that is,
w < aa w < w.
(ii) /i = 1. Then, by the inductive hypothesis,
σΟι w = aa. . . . oa. < aaiw - aa2 . . . oav B y C o r o l l a r y 2 . 6 , w' < w.
Proposition 2.8 yields an alternative definition of the ordering on W
(see [7J). The geometrical interpretation of this ordering is very interesting
 Schubert cells and cohomology of the spaces G/P

and useful in what follows.

2.9 THEOREM. Let V be a finite-dimensional representation of a Lie
algebra <& with dominant weight X. Assume that all the weights wX
w 6 W, are distinct and select for each w a non-zero vector fw ζ V of
weight wX. Then
' w . e U ( K ) fw

(where UC$V) is the enveloping algebra of the Lie algebra %l).

PROOF. For each root γ ζ Δ we fix a root vector Ey £ @ in such a
way that [Ey, E.y] = Hy. Denote by Sfv the subalgebra of @, generated by
Ε7, E^, and Hy. SIV is isomorphic to the Lie algebra s/ 2 (C). Let
w' -Λ- w and let V be the smallest Sfv-invariant subspace of V containing
fw··
2.10 LEMMA. Let η = w'\{Hy) G Ζ, η > 0. The elements
{Elvfw. |i « 0, 1, . . ., n)form a basis of V. Put f= E^fc. Then
E_~f = 0, £?/ = c'U· (c' Φ 0) and fw = cj (c Φ 0).
PROOF. By Lemma 2.2, w'~* y £ Δ + , hence Eyfw· =
= cEyw'fe — cw'EW'-iyfe = 0, that is, fw· is a vector of dominant weight
relative to SI V . All the assertions of the lemma, except the last, follow
from standard facts about the representations of the algebra §lv ^ sl2 (C).
Furthermore, / and /„, are two non-zero vectors of weight wX in V, and
since the multiplicity of wX in V is equal to 1, these vectors are propor-
tional. The lemma is now proved.
To prove Theorem 2.9 we introduce a partial ordering on W by putting
w' ~\ w if fw< G U{W)fw- Since all the weights if λ are distinct, the
relation -\ is indeed an ordering; we show that it satisfies conditions a)
and b) of Proposition 2.7.
a) Let α ζ Σ and l(aaw) = l(w) + 1. Then w %• aaw, and by Lemma
2.10, /„, 6 U (31) faaW, that is, w -\ aaw.
b) Let w Η w . We choose an a e Σ such that w ^ aaw. Replacing
w by aaw', if necessary, we may assume that aaw' -> u/. We prove that
aaw -\ w', that is, faaw(:U(%l) fw,. It follows from Lemma 2.10 that
= an =
E-afw 0 d /σοι» cE"afw · Let $βα be the subalgebra of @ generated
by 5R, ή andSTa. Since ιν Η w, fweU(9l)fw. and so / α β Β ( = cE\fw = Xfc,
where Χζϋ(^βα). Any element Jf of U(5pa) can be represented in the

form X = S ^ ^ i +YE-a, where Κ,€ί7(91), VJ 6 ^ ( ή ) , YeU($a).

Therefore* fσ<χψ =% Υ tY'i fw- = Σ CiYifw'eUWfw and Theorem 2.9 is
proved.
We use Theorem 2.9 to describe the mutual disposition of the
Schubert cells.
2.Π. THEOREM (Steinberg [7J_). Let w 6 W, Xw C Χ α Schubert cell,
and Xw its closure. Then Xw- C Xw if and only if w' < w.
10 /• Ν. Bernstein, I. M. Gel'fand, S. I. Gel'fand

To prove this theorem, we give a geometric description of the variety

Let V be a finite-dimensional representation of G with regular dominant

weight λ (that is, all the weights w\ distinct). As above, we choose for
each w e W a non-zero vector /„, 6 Κ of weight w\. We consider the
space P{V) of lines in V; if / € V, f Φ 0, then we denote by [f] 6 P(V)
a line passing through /. Since λ is regular, the stabilizer of the point
[/J 6 P(v) under the natural action of G on P(V) is B. The G-orbit of
[fe] in P{V) is therefore naturally isomorphic to X = G/B. In what follows,
we regard AT as a subvariety of i*(K).
For each w 6 PV we denote by <£„, the linear function on V given by
<t>w(fw) ~ 1> Φι«(/) = 0 if / £ Ρ is a vector of weight distinct from w\.
2.12 LEMMA. l e i / e V and if] £ X. 77ien

PROOF. We may assume that / = gfe for some g 6 G.

Let [/] C * „ , , that is, g 6 NwB. Then / = d exp (Y)wf. for some
Υ € St, hence / 6 Ε/(91)/ο, and φω (f) Φ 0.
On the other hand, it is clear that for each / € V there is at most one
w 6 W such that f ζ U (M) fw and <t>w{f) Φ 0. The Lemma now follows
from the fact thatX = [} Xw.
We now prove JTheoreifT2.11
a) Let Xw· C Xw. Then [/„-] 6 * „ , , and by Lemma 2.12, / „ . ζ ^(91)/ ω .
So u;' < w, by Theorem 2.9.
b) To prove the converse it is sufficient to consider the case w' -^ w.
Let η = w\(Hy) 6.Z. Just as in the proof of Theorem 2.9, a) we can show
that η > 0, £?/„, = c/ w ' and Ef* fw = 0.
Therefore Um r " exp (i£ 7 ) /„ =%fw', that is, [/„,-] 6 Xw. Hence,

§3. Discussion of the ring of polynomials on ή

In this section we study the rings R and R. For each w £ W we define
an element ^ έ Λ and a functional !)„, on R and investigate their proper-
ties. In the next section we shall show that the Dw correspond to Schubert
cells, and that the Pw yield a basis, dual to the Schubert cell basis, for the
cohomology of X.
3.1 DEFINITION, (i) R =® Rt is the graded ring of polynomial
functions on I)Q with rational coefficients. W acts on R according to the
rule wfth) = Aw-1 h).
(ii) / is the subring of W-invariant elements in R,

h = {/ e /1 / (0) = o}.
 Schubert cells and cohomology of the spaces G/P 11

(iii) / is the ideal of R generated by /+.

(iv) R = R/J.
3.2 DEFINITION. Let γ ζ Δ. We specify an operator Ay on R by the
rule

Ayf lies in R, since f — ayf - 0 on the hyperplane y = 0 in t)Q-

The simplest properties of the Ay are described in the following lemma.
3.3 LEMMA.(i) A_y = -Ay, A}, = 0.
(ii) wAyw'i = Awy.
(iii) ayAy = - Ayay = Ay, ay = —yAy + 1 = Ayy — 1.
(iv) Ayf = 0 ~ oyf = f.
(v) AyJ C /.
(vi) Let χ ς £)Q. Then the commutator of Ay with the operator of
multiplication by X has the form [Ay, χ] = x(Hy)ay.
PROOF, (i) — (iv) are clear. To prove (v), let f = fif2, where
/i 6 /+, /a 6 -ff· It is then clear that Ayf = fi.Ayf2 ζ /. As to (vi), since
°yX = X - x(Hyh, we have
[Ay, χ ] / = |
X —σ ν Χ
=— °vf=X(Hy)-oyf.
T h e following p r o p e r t y o f t h e A y is f u n d a m e n t a l in w h a t follows.
3.4 T H E O R E M . Let u , , . . ., α , € Σ , and put w = σ α ι . . . o a j ;
Aa, at) = Aai . . . Aar
a) Ifl(w) < I, then Ai0lj αΰ = 0.
b) // l(w) - I, then Aiai α;) depends only on w and not on the
set <*i, . . ., at. In this case we put Aw = A(a^ αχ).

The proof is by induction on /, the result being obvious when / - 1.

For the proof of a), we may assume by the inductive hypothesis that
l(aai . . . aaii) = / _ 1, consequently /(σΟ) . . . o ^ o , , , ) = 1 — 2.
Then σ^. σ α / + 1 . . . σαι1 = oa.+1 . . . oatl σ α; for some i ( we have applied
Corollary 2.3 to the case w = a^^ . . . aa , y = at). We show that
=
^4(α,·, . . ., aj) 0.
Since I - i < I, the inductive hypothesis shows that
AaAaui ... Ααι_^Αα.+ι ... AailAa{, and so by lemma 3.3 (i)
Aa^ . . . Aa^ = -^α; +1 · · · -Aa^Aa^ — U.
To prove b), we introduce auxiliary operators Bicti a;) , by setting
B
&1 ocj) = tfccj · · • % - < 4 ( a i a,)·

We put «;,· = σα/ . . . σα.. Then in view of Lemma 3.3 (ii, iii) we have
12 I. N. Bernstein, I. M. Gel'fand, S. I. Gel'fand

(where A™ stands for wAyw^ ).

3.5 LEMMA. Let %£i)Q. 77zi? commutator of BiOli ννζί/ζ r/ze
αι)

operator of multiplication by χ is given by the following formula:1

ι
(8) [£(βι α;), χ] = 2
2 _
PROOF. We have
(04. ...-, α;), χ] = [Α^Αζξ ... Ααν χ] =

By Lemma 3.3 (ϋ, vi), [Α%+\ χ] = χ(ιι; ί + 1 # β ί ) o W i + i e j .

Since σ Β) . +ι01 . = wi+1w? , we have
2
Γ, = χ («;I+Iff«f) < · • · ^ί_ 1 «' Ι + ,«'ϊ ι 4«ί*ί · • · ^ « r

We want to move the term w^wf* to the left. To do this we note

that for / < ι

Therefore,

By (7), applied to the sequence or roots (a 1 ; . . ., a,·, . . ., a,), we have

Tt = χ (wi+jff e j ) wMw?B(ai Sj ap ,
and Lemma 3.5 is proved.
If /(σ^ . . . σα. . . . σ αΐ ) < / - 1, then 7^ = 0 by the inductive
hypothesis. If /(σαι . . . σα[. . . . σαι) = / — 1, then, putting
w' = aUi . . . aa. . . . σαι and 7 = σ αι . . . σ,^,^ία,-), we see from Lemma
2.2 that w'^* w, and also
χ {wMHa.) = ιο'χ {w'wi+lHa.) = u/χ (σ β ΐ . . . oaiiHa.) = κ/
and

1
" indicates that the corresponding term must be omitted.
 Schubert cells and cohomology of the spaces G/P 13

Using Proposition 2.8 c) and the inductive hypothesis, (8) can be

rewritten in the following form:
W
[£(«, «,), %] = 2 't (By) W^Ay,.
ν
w' —>
The right-hand side of this formula does not depend on the represen-
tation of w in the form of a product σΟ[ . . . o a j . The proof of theorem
3.4 is thus completed by the following obvious lemma.
3.6. LEMMA. Let Β be an operator in R such that B(\) = 0 and
[Β, χ) = 0 for all % 6 ί)ο· Then Β = 0.
3.7. COROLLARY. The operators Aw satisfy the following commutator
relation:
[w-1Aw,%]= 2
ν
w' — > w
We put St = Rt* (where Rt C R is the space of homogeneous poly-
nomials of degree /) and S = © 5 , . We denote by ( , ) the natural pairing
S X R -+ Q. Then W acts naturally on S.
3.8 DEFINITION, (i) For any χ ζ f)Q we let χ* denote the transform-
ation of S adjoint to the operator of multiplication by χ in R.
(ii) We denote by Fy: S -*• S the linear transformation adjoint to
Ay: R -* R.
The next lemma gives an explicit description of the Fy.
3.9 LEMMA. Let y ζ Δ. For any D 6 S there is a D ζ S such that
j * (D) = D. If D is any such operator, then D — oyD = Fy(D), (in par-
ticular, the left-hand side of this equation does not depend on the choice
of D).
PROOF. The existence of D follows from the fact that multiplication by
γ is a monomorphism of R. Furthermore, for any / 6 R we have
(D - ay5, j) = 0,f- σν/) = (D, Avf • y) = (γ* (D), Avf) = (D, Avj),
hence D — oyD = Fy.
REMARK. It is often convenient to interpret S as a ring of differ-
ential operators on ή with constant rational coefficients. Then the pairing
( , ) is given by the formula (D, f) = (Of) (0), D £S,'f £R. Also, it is easy
to check that x*(D) = [D, χ ] , where X € ^Q and D € S are regarded as
operators on R.
Theorem 3.4 and Corollary 3.7 can be restated in terms of the operators
Fy.
3.10 THEOREM. Let au . . ., α, ζ Σ, w = σα . . . σα .
(i) // l(w) < I, then Fai . . . Fai = 0. '
(ii) // l(w) = I, then Fai . . . FUi depends only on w and not on
«i, . . ., a,. In this case the transformation Fai . . . FUi j s denoted by
Fw. (Note that Fw = A*).
14 /. Ν. Bernstein, I. M. GeVfand. S. I. GeVfand

(iii) [%*,FwW]= Σ w'%(Hy)Fw.w.

W' > ID

3.11. DEFINITION. We set Dw = Fw (1).

As we shall show in § 4, the functionals Dw correspond to the Schubert
cells in Ht {X, Q) in the sense that {Dw, f) = isw, «(/)> for all f 6 R.
The properties of the Dw are listed in the following theorem.
3.12. THEOREM, (i) Dw £ Shwy
(ii) Let we W, α ζ Σ. Then
0 if l{waa) = Kw) - 1,
=
>woa if l(w°u)
(iii) Letie^k- Then

ν
w' — > w
(iv) Let a 6 Σ. Then
j-Da, if
a*Dw = i -Dw+ 2 w'oL(Hy)Dw- if

(v) Let w e W, I (w) = Ζ, χ!, . . ., χ,

(A»> XI> · · ·> Xi) = Σ Χ ι ^ 7 ι ) · - - Χι(Ηι), where the summation extends
over all chains
yt v2 v;

PROOF, (i) and (ii) follow from the definition of Dw and Theorem
3.10 (i).
(iii) χ* (£>„) = x*Fww(\) = [χ*, Fww] (1) (since χ* (1) = 0), and (iii)
follows from Theorem 3.10 (iii).
It follows from Lemma 3.3 (iii) that oa - u*Fa — 1. Thus, (iv) follows
from (ii) and (iii).
(v) We put Dw =/)„,->. Then the Dw satisfy the relation

ν
w —>
Since (D, xf) = ( χ * φ ) , f), (v) is a consequence of (9) by induction on /.
Let SB be the subspace of 51 orthogonal to the ideal 7 c R. It follows
from Lemma 3.3 (vi) that ©$ is invariant with respect to all the Fy. It is
also clear that 1 ς <ffl. Thus, Dw ζ $β for a n w ς Η.
3.13. THEOREM. The functionals Dw, w 6 W, form a basis for SE.
PROOF, a) We first prove that the Dw are linearly independent. Let
s 6 W be the element of maximal length and r = l(s). Then, by Theorem
3.12 (v), Ds(pr) > 0 and so Ds Φ 0. Now let 2 cwDw = 0 and let w be
one of the elements of maximal length for which cw Φ 0. Put / = Z(S3).
 Schubert cells and cohomology of the spaces G/P 15

There is a sequence at, . . ., ar-t for which ΐύσαι . . . aarl = s. Let

F F F
= <xr., • • • a, • It follows from Theorem 3.10 that FD% = D, and
FDW = 0 if l(w) > I, w Φ w. Therefore F(^cwDw) = csD, φ 0.
b) We now show that the Dw spand^1. It is sufficient to prove that if
f Ε R and (Dw, f) = 0 for all w 6 W, then / 6 /. We may assume that /
is a homogeneous element of degree k. For k = 0 the assertion is clear.
Now let k > 0 and assume that the result is true for all polynomials /
of degree less than k. Then for all α 6 Σ and w ζ W,
(Dw, Aaf) = (FaDw, f) = 0, by Theorem 3.10 (i) and (ii). By the inductive
hypothesis, Aaf ζ J, that is, / — aaf - aAaf ζ J. Hence for all w ζ. W,
f Ξ wf (mod / ) . Thus, | W I" 1 2 wf = f (mod / ) . Since the left-hand side
w£W
belongs to /+, we see that / € J. Theorem 3.13 is now proved.
The form ( , ) gives rise to a non-degenerate pairing between R = R/J
andSS- Let {Pw} be the basis of R dual to {Dw}. The following properties
of the Pw are immediate consequences of Theorem 3.12.
3.14 THEOREM, (i) Let w β W, α ζ Σ. Then
f0 if l(waa) - l(w) + 1,
I rwaa if l(W0a ) =

V
w — ^ w'

(iii) Let α ζ Σ. Then

[ Pw if
" • = 1 Pw - Σ wa (Hv) Pw. if
ν

From (i) it is clear that all the Pw can be expressed in terms of the
Ps. More precisely, let w = οαχ . . . aav l{w) = r — I. Then

Pw = A*, • · · ΛαιΛ>.

To find an explicit form for the Pw it therefore suffices to compute the

Ps 6 Λ .
3.15 THEOREM. P, = | W7 | J [] γ (mod/).
PROOF. We divide the proof into a number of steps. We fix an element
h 6 i) such that all the wh, w ζ W, are distinct.
1. We first prove that there is a polynomial δ G Λ of degree r such
that
(10) Q{sh) = 1, <?(ΜΛ) = 0 for w Φ s.

For each if e fV we choose in R a homogeneous polynomial Pw of degree

16 /. JV. Bernstein, I. M. Gel'fand, S. I. GeVfand

l(w) whose image in R = R/J is Pw. Since {Pw} is a basis of R, any polynomial
/ e / ? can be written in the form / = Σ Pwfw. where .£,€ / (this is easily proved
by induction on the degree of f). Now let Q' 6 R be an arbitrary poly-
nomial satisfying (10) and let Q' ~ 2 Pwgw, iw£l. It is clear that
<2 = Σ gu) (A) i-'u, meets our requirements.
w

2. Let Q be the image of β in Λ, and let Q = 5J - U , ^ be the represen-

tation of Q in terms of the basis \PW} of R. We now prove that

_ _
To prove this we consider ASQ. On the one hand ASQ = cs, by Theorem
3.13 (i); on the other hand, ASQ is a constant, since Q is a polynomial of
degree r. Hence, ASQ = cs.
We now calculate ASQ, Let s = aai . . . aar be the reduced decompo-
sition. We put wt = aa. . . . oa (in particular, w0 - e), γ; = wf\ at,
Q, = ^ i + 1 · · · Λ , , β . '
LEMMA. Q,- is a polynomial of degree i,

Qi(wh) = 0 if w 9«(«>ίΑ)
^ wt. = ( - 1 Γ * . Π
PROOF. We prove the lemma by backward ^j>
induction on i. For / = r
we have wr = s, Qr - Q, and the assertion of the lemma follows from the
definition of Q.
We now assume the lemma proved for Qt, i > 0. In the first place, it
is clear that β,·_, = Λα.<2,- is a polynomial of degree ι — 1.
Furthermore,
Qi (wh) — Qi (aa.wh)
Q^ (wh) = Aafit (wh) = ^ ^ — •
If w = ίο,·., , then io < u>j, oa.w = wt a n £ i
aK^i-i h) = (wf-i α.) (Λ) = - (wr[ a^ (h) = - Ji(h). Therefore, using the
inductive hypothesis, we have

But if w %> Wi_!, Corollary 2.6 implies that w > wt and a tti to > w,·. So
Qi_, (ιοΛ) = 0, and the lemma is proved.
Note that by Lemma 2.2, as i goes from 1 to r, γ,· ranges over all the
positive roots exactly once. Therefore

3. Consider the polynomial Alt (Q) = Σ ( - D ' ^ ^ Q ; Alt (Q) is skew-

symmetric, that is, aa Alt(Q)= -Alt(Q) for all y e Δ. Therefore Alt(Q) is divisible
 Schubert cells and cohomology of the spaces G/P 17

(in R) by [f y. Since the degrees of Alt(Q) and Π 7 are equal (to r),

Alt«2) = λ Π γ. Furthermore, Alt(Q) (Λ) = ( - l ) r , so that

(11) Alt (<?) = ( - l ) r Π (VW)' 1 Π 7·

4. We put Alt (β) = 2 ( - 1 ) i ( l " ) " ' δ - By Theorem 3.14 (iii),

Alt(Ps) = 2 ( - I)*"» wPs = \W\ Ps. Therefore Alt(<2) = cs \W\ Ps + terms
of smaller degree. Since Alt(£2) is a homogeneous polynomial of degree r,
we have
(12) Alt (Q) = cs\W\Ps.

By comparing (11) and (12) we find that

Ps = \W\-i Π γ (mod./).
V£A+
The theorem is now proved.
3.16 COROLLARY. Let ρ be half the sum of the positive roots. Then
Ps = pr/r\ (mod / ) .
PROOF. For each χ ζ- ί)* we consider the formal power series exp χ on
i) given by
oo

βχρχ= 2 χη/η!.

Then we have (see [9])

Comparing the terms of degree r we see that

L = π Υ·
If p r (mod /) = \PS, λ € C, then (wp r ) (mod / ) = XwPs = λ ( - l)'^/*,·
Thus, i ^ 2 ( — i)Uw)iwp)r = XPS (mod J). The result now follows from
Theorem 3.15.
To conclude this section we prove some results on products of the
Pw in R.
3 . 1 7 . T H E O R E M , ( i ) L e t α ζΣ, ινζ W. Then

PoaPW= Σ 7.a(Hw-ly)Pw',
w — > w'

where Χα € ί)ζ is the fundamental dominant weight corresponding to the

root a (that is, χα (Hp) = 0 for α Φ β € Σ, Χΰί (Ηα) = 1).
(ii) Let wi,w3 € W, l(wl) + l(w2) = r. Then Pm PWJ = 0 for
18 /. Ν. Bernstein, I. M. Gel'fand, S. I. Gel'fand

W2 Φ WiS, PwlPwls= Ps_ ^

(iii) Let weW, f e R. Then f Pw = & cw'Pw>.
W =3=
W 3= It»
It»

(iv) //Wl 4 w2s, then PWIPW2 = 0.

PROOF, (i) By Theorem 3.12 (ν), ΡΟα = χ,, (mod / ) . Therefore (i)
follows from Theorem 3.14 (ii).
(ii) The proof goes by backward induction on l(w2 )• If l(w2) = r, then
w2 = s, Wi = e and PWl = 1.
To deal with the general case we find the following simple lemma useful,
which is an easy consequence of the definition of the Ay.
3.18 LEMMA. Let y ζ Δ, /, g € R. Then Ay(Ayf-g) = Ayf-Ayg.
Thus, let w2 € W, l(w2) - I < r, and choose α 6 Σ so that
w2 -»• aaw2. We consider two cases separately.
A) Wi 9* aaWi. We observe that the following equation holds for any
w € W

(13) l(ws) = r — l(w).

Since in our case l{aaw2) = / + 1 and l(aaWi) = r — I + 1, we see that

oaWiS Φ aaw2, and so WiS Φ w2. On the other hand, PWi = AaPOaw2 and
PWI = ^aPaawi by Theorem 3.14 (i). Therefore, an application of Lemma
3.18 shows that

Since /(σαΐι^) + /(u; 2 ) = r — / + 1 + / > /·, we have POaWlPW2 = 0.

Hence PWiPW2 = 0 as well.
B) oawl ^ Wi. In this case, Ρ σ α ω ι = ^ P ^ j and PW2 = ^ α Ρ σ α Ι 1 ) 1 , by
Theorem 3.14 (i). Again applying Lemma 3.18, we have
(14) Aa (PWiPW2) = Aa {PWi · i a P V 2 ) = AaPWi • AaPOaWz =
= Aa (AaPWl · P(Saw^) = ^la (^CTaWj' - ^ V V "
5
Since the i ^ form a basis of Λ, any element / of degree r in R has the
form / = \PS, λ 6 C. Furthermore, Α Λ = iO f f « ^ 0· But
deg PWIPW2 - deg POaWr POaU>2 - r. Therefore (14) is equivalent to
PwiPw2 = P^^iPo^^
Applying the inductive hypothesis to the pair {aaWi, aaw2), we obtain
part (ii) of the theorem.
(iii) is an immediate consequence of Theorem 3.14 (ii).
(iv) follows from (ii) and (iii).
We define the operator oP: R"-> S£ of Poincare duality by the formula
<0Y)(g) = Ds(fg), /, g £ R , g>fe m .
3.19. COROLLARY. 0>PW =DWS.
 Schubert cells and cohomology of the spaces G/P 19

§4. Schubert cells

We prove in this section that the functionals Dw, w β W introduced in
§ 3 correspond to Schubert cells sw, w 6 W.
Let sw ζ Η^{Χ, Q) be a Schubert cell. It gives rise to a linear functional
on H*(X, Q), which, by means of the homomorphism a: R -* H*(X, Q)
(see Theorem 1.3), can be regarded as a linear functional on R. This
functional takes the value 0 on all homogeneous components Pk with
k Φ l(w), and thus determines an element Dw ζ Snw).
4.1. THEOREM. Dw = Dw (cf. Definition 3.11).
This theorem is a natural consequence of the next two propositions.
PROPOSITION 1. De = 1, and for any X 6 i)z

(15) χ* ( 4 · ) - Σ w'i(Hy)Dw.
ν
Η"' > V)

PROPOSITION 2. Suppose that for each w 6 W we are given an

element Dw 6 Sl(w), with De = 1, for which (15) holds for any% ζ ill .
Then Dw = Dw.
Proposition 2 follows at once from Theorem 3.12 (iii) by induction
on l(w).
We turn now to the proof of Proposition 1.
We recall (see [10]) that for any topological space Υ there is a bilinear
mapping

Hl (Y, Q) χ IIj (Y, Q) ^ Ή,-i (Y, Q)

(the cap-product). It satisfies the condition:

(16) 1. (cny, z) = (y, c-z)

for all yeiIj(Y, Q), ζζΗ'-ι(Υ, Q), οζΙΡ (Y, Q).

2. Let / : Yx -> F 2 be a continuous mapping. Then

(17) U(f*cny)=cnUy

for a l l / y e # ; ( ^ , Q). Ρ 6 / / * ( Υ 2 , Q).

By virtue of (17) we have for any χ 6 51. /Gi?
(χ* (ΑΛ, /) = (A,, if) =te»,«! (χ)«(/)> = (%· η α, (χ), α (/)}.
Therefore (15) is equivalent to the following geometrical fact.
PROPOSITION 3. For all χ£ί)ζ
(18) sw η α, (χ) = Τ, w'% (Hy) sw..
7

2
We restrict the fibering Εχ to Xw c X and let cx 6 H (Xw, Q) be the
first Chern class of Ex. By (17) and the definition of the homomorphism
cti:£)i->// 2 (^, Q), it is sufficient to prove that
20 /. JV. Bernstein, I. M. Gel'fand, S. I. Gel'fand

(19) swncx= 2 w'%{Hy)sw..

V
w' —>• w
in H2i(wy-2{XW, Q).
To prove (19), we use the following simple lemma, which can be verified
by standard arguments involving relative Poincar6 duality.
4.2 LEMMA. Let Υ be a compact complex analytic space of dimension
n, such that the codimension of the space of singularities of Υ is greater
than 1. Let Ε be an analytic linear fibering on Y, and c 6 IP(Y, Q) the
first Chern class of E. Let μ be a non-zero analytic section of Ε and
^mtYi = div μ the divisor of μ. Then [Υ] η c = ^m,[Yt] € H2n-2{Y, Q),
where [Y] and [FJ are the fundamental classes of Υ and Yt.
Let w € W, and let Xw c X be the corresponding Schubert cell. From
Lemma 4.2 and Theorem 2.11 it is clear that to prove Proposition 3 it is
sufficient to verify the following facts.
4.3. PROPOSITION. Let w -»• w. Then Xw is non-singular at points
χ € Xw>.
4.4. PROPOSITION. There 4s a section μ of the fibering Ex over Xw
such that
<ϋνμ= >J wi\ny)Aw·.
ν

To verify these facts we use the geometrical description of Schubert

cells given in 2.9. We consider a finite-dimensional representation of G on
a space V with regular dominant weight λ, and we realize X as a subvariety
of P(V). For each w € W we fix a vector /„, € V of weight α>λ.
PROOF OF PROPOSITION 4.3. For a root γ 6 Δ + we construct a
three-dimensional subalgebra 2IY c; © (as in the proof of Theorem 2.9).
Let i: SL2(C) -+ G be the homomorphism corresponding to the embedding
S2IV_> @. Consider in SL2(Q the subgroups^' = -ji _x J !>, H' =

f/1 0\Ί / 0 1\
and Λ^'_ = I I I }· and the element σ = I . ^ I. We may assume that
c H, i(B') c B.
Let V be the smallest S[Y-invariant subspace of V containing /„,-. It is
clear that V is invariant under i(SL2 (C)), and that the stabilizer of the line
[/u/ ] is B'. This determines a mapping δ: SL? (C)/B'- -»· X. The space
SL2 (C)/5' is naturally identified with the projective line P 1 . Let o, °° e P 1
be the images of e, ο 6 SL2(C).
We define a mapping ξ: Λ/^' Χ Ρ 1 -»· Χ by the rule
(χ, ζ) ι-*- α:·δ(ζ).
4.5. LEMMA. The mapping ξ has the following properties:
x {o}) = XW', &NU. Χ ( Ρ ! \ ο)) cz Xw.
 Schubert cells and cohomology of the spaces G/P 21

(ii) The restriction of ξ to (Nw· χ ρ 1 \ °ο)) iS a n isomorphism onto a

certain open subset of X^,.
Proposition 4.3 clearly follows from this lemma.
PROOF OF LEMMA 4.5. The first assertion of (i) follows at once from
the definition of Xw·. Since the cell Xw is invariant under N, the proof of
the second assertion of (i) is reduced to showing that δ(ζ)·€ Xw for
ζ 6 Ρ 1 \ o. Let h 6 SL2(C) be an inverse image of z. Then h can be
written in the form h - bvab2, where bx,b2 ζ Β'. It is clear that
i(b2)fw' = clfw· and i(o)fw> = c2fw , where cx , c2 are constants. Therefore
i(h)fw' = Cic2i(bi)fw, that is, δ(ζ) 6 Xw.
To prove (ii), we consider the mapping
w'-ioliNv,- X ( P x \ oo)-vZ.
The space P1 \°° is naturally isomorphic to the one-parameter subgroup
NL C SL2(C).
The mapping ξ: Λ^-Χ NL -*• X is given by the rule
l(n, «i) = ni(ni) [fw·], η £_ Nw>, nt (j Ν'.
Thus,
w"1 ο ξ (η, rii) = (w'-i-nw1) {w'~H («,) w') [fe].
We now observe that w'~lNw'w' C N_ (by definition of ^ 0 , and
w1'1 i(N')w' ζ-Ν_ (since u/" 1 γ 6 Δ+). Furthermore, the intersection of the
tangent spaces to these subgroups consists only of 0, because Λ^- c N,
i{NL) C 7V_. The mapping N_ ->· X (n h- n[fe]) is an isomorphism onto an
open subset of X. Therefore (ii) follows from the next simple lemma,
which is proved in [ 5 ] , for example.
4.6. LEMMA. Let TVj and N2 be two closed algebraic subgroups of a
unipotent group Ν whose tangent spaces at the unit element intersect only
in 0. Then the product mapping Νλ X N2 -*• Ν gives an isomorphism of
Ni X N2 with a closed subvariety of N.
This completes the proof of Proposition 4.3.
PROOF OF PROPOSITION 4.4. Any element of t)S has the form
χ = λ - λ', where λ, λ' are regular dominant weights. In this case,
E% = E% (g> Ej}, and it is therefore sufficient to find a section μ with the
required properties in the case χ = λ.
We consider the space P{V), where V is a representation of G with
dominant weight λ. Let r? v be the linear fibering on P(V) consisting of
pairs (Ρ, φ), where φ is a linear functional on the line Ρ C V. Then
E\ = i*(vv), where /: X -+ P(V) is the embedding described in §2.
The linear functional φω on V (see the proof of Theorem 2.11) yields a
section of the bundle η. We shall prove that the restriction of μ to this
section on Xw is a section of the fibering Ελ having the requisite properties.
By Lemma 2.12, μ(χ) Φ 0 for all χ 6 Xw. The support of the divisor
div μ is therefore contained in XW\XW - U Xw·.
22 /. Ν. Bernstein, I. M. Gel'fand, S. I. Gel'fard

Since X^,· is an irreducible variety, we see that div μ = ZJ ayXw·, where

ay 6 Ζ, ay > 0. It remains to show that ay = w'\{Hy).

In view of Lemma 4.5 (i) and (ii), the coefficient ay is equal to the
multiplicity of zero of the section δ*(μ) of the fibering δ*(Εχ) on P 1 at
the point o, that is, the multiplicity of zero of the function
= 0u/((exp tE_y)fw·) for t = 0. It follows from Lemma 2.10 that
= ct", hence ay = η = w\{Hy). This completes the proof of Proposi-
tion 4.4 and with it of Theorem 4.1.

§5. Generalizations and supplements

1. Degenerate flag varieties. We extend the results of the previous sec-
tions to spaces X(P) = G/P, where Ρ is an arbitrary parabolic subgroup of
G. For this purpose we recall some facts about the structure of parabolic
subgroups Ρ c G (see [7]).
Let Θ be some subset of Σ, and Δ Θ the subset of Δ+ consisting of
linear combinations of elements of Θ. Let G© be the subgroup of G
generated by Η together with the subgroups Ny = {exp tEy \ t 6 C} for
y £ Δ© υ - Δ Θ , and let ./ν© be the subgroup of TV generated by the Ny
for γ f Δ + \Δ Θ . Then G© is a reductive group normalizing ΝΘ , and
P@ = GQNQ is a parabolic subgroup of G containing B.
It is well known (see [7], for example) that every parabolic subgroup
Ρ c G is conjugate in G to one of the subgroups P&. We assume in what
follows that Ρ = P&, where Θ is a fixed subset of Σ. Let W& be the Weyl
group of G©. It is the subgroup of W generated by the reflections aa,
a e Θ.
We describe the decomposition of X(P) into orbits under the action of B.
5.1. PROPOSITION, (i) X{P) = JJW Bwo, where ο 6 X(F) is the image
of Ρ in G/P.
(ii) The orbits B\x)\O and Bw2o are identical if W1W21 6 W& and other-
wise are disjoint.
(iii) Let W© be the set of w 6 W such that w® c Δ+. Then each coset
of W/WQ contains exactly one element of W@. Furthermore, the element
w 6 W@ is characterized by the fact that its length is less than that of
any other element in the coset wW&.
(iv) If w e W@, then the mapping Nw -»• X(P) (n -» nwo) is an iso-
morphism of Nw with the subvariety Bwo c X(P).
PROOF, (i)-(ii) follow easily from the Bruhat decomposition for G and
G©. The proof of (iii) can be found in [7], for example, and (iv) follows
at once from (iii) and Proposition 1.1.
Let w 6 R/©, XW{P) = Bwo, let XwiP) be the closure of XW{P) and
[XW(P)\ € H2Kw)(Xw(P), Z) its fundamental class. Let
sw(P) € H2Kw)(X(P), Z) be the image of [XW(P)] under the mapping
 Schubert cells and cohomology of the spaces G/P 23

induced by the embedding Xw (P)cz—>X(P). The next proposition is an

analogue of Proposition 1.2.
5.2. PROPOSITION ([2]). The elements sw{P), w ζ W^, form a free
basis in Η*(Χ(Ρ), Ζ).
5.3. COROLLARY. Let aP: X -» X(P) be the natural mapping. Then
{aP)*sw = 0 if w φ W&, (ctp^Su, = sw{P) if w e W1®.
5.4. COROLLARY, (ap)*: H*(X, Z) -* H*(X(P), Z) is an epimorphism,
and (aP)*: H*(X(P), Z) -»• H* (X, Z) is a monomorphism.
5.5. THEOREM, (i) Im(a P )* C H*(X, Z) = R coincides with the set of
W@-invariant elements of R.
(ii) Pw 6 Im(ap)* for w G W& and {(aP)*-1Pw}w£Wi is the basis in
H*(X(P), Z) dual to the basis { s J P ) } ^ ^ in HJX(P), Z).
PROOF. Let w G W©. Since <PW, sWi) = 0 for w, φ W&, Pw is orthog-
onal to Ker(ap)*, that is, Pw 6 Im(ap)*. Now (ii) follows from the fact
that {(aP)*Pw, sw'(P)> = (Pw, v > for w, w G• W&. To prove (i), it is
sufficient to verify that the Pw, w 6 W@ J form a basis for the space of
W@ -invariant elements of R. We observe that an element / G R is
We-invariant if and only if Aaf = 0 for all α G Θ. Since w G W® if and
only if l(waa) = l(w) + 1 for all a 6 Θ, (i) follows from Theorem 3.14(i).
2. CORRESPONDENCES. Let Υ be a non-singular oriented manifold.
An arbitrary element z G Η* (Υ Χ Υ, Ζ) is called a correspondence on Y.
Any such element ζ gives rise to an operator ζ%: H*(Y, Z) -»· H*(Y, Z),
according to
z
*( c ) = (^2)*((jti)*(^>c) π ζ)> ° G # * (y, z),

where π ι , π 2 : Υ Χ Υ -*• Υ are the projections onto the first and second
components, and cP is the Poincare duality operator. We also define an
operator z*\ H*(Y, Z) -> H*(Y, Z) by setting z*(I) = ίΓ>[(πΙ),((πϊ)·(|) Π ζ )]-
I f tf*(7, Z). It is clear that z^ and z* are adjoint operators.
Let ζ be assigned to a (possibly singular) submanifold Ζ C Υ Χ Υ, in
such a way that 2 is the image of the fundamental cycle [Z] G H^.{Z, Z)
under the mapping induced by the embedding Ζ c_>_ Υ Χ Υ. Then
2 c
*( ) = (P2)* ([Ζ] Π (pi)*^c),
where p j , p 2 : Ζ -* Υ are the restrictions of η χ , n2 to Ζ.
If, in this situation, p1 : Ζ -»· Υ is a fibering and c is given by a sub-
manifold C C Y, then the cycle
izi η (pi)*^ c
is given by the submanifold ρϊι (C) c Z.
We want to study correspondences in the case Υ - X = G/B.
5.6. DEFINITION. Let w G W. We put Zw_= {(gwo, go)} C X X X
and denote by zw the correspondence zw = [Zw ] C H*(X X X, Z).
5.7. THEOREM. (zw )* = Fw .
PROOF. We calculate (zw )„. (sw·).
24 /. Ν. Bernstein, I. M. Gel'fand, S. I. Gel'fand

Since the variety Zw is G-invariant and G acts transitively on X, the

mapping px : Zw -»• X is a fibering. Thus,

It is easily verified that p? (Xw·) = π? (Xw·) η Zw. We put

Υ = ττΓ1 (Xw·) η Zw c X X X. Then
(20) Υ = {(nw'o, nw'bwo) \n£N, b ζ 5 } .

Since the dimension of the fibre of pi : Z^ ->· Ζ is equal to 2l(w), we see

that dim Υ = 2l(w) + 2/(u/). It is clear from (20) that
p2(Y) = {«H/OWO | η £ iV, 6 £ B} = Bw'Bwo.
It is well known (see [6]. Ch. IV, §2.1 Lemma 1) that
Bw'Bwo == BW'UJO U ( υ

Thus, two cases can arise.

a) l(w'w) < l(w') + l(w). In this case, dim p2(Y) < 21(w') + 2l(w), and
so ( z ^ C v ) = 0>2)*lY] = 0.
b) l(w'w) = ί(ΐϋ') + /(u;). In this case, p2{Y) = Xw'w + X', where
dim X' < dim Xw>w = 2l(w') + 2l{w). Thus, (p2)m[Y] = [Xw'wh that is,
izw)*(sw') = sw'w • Comparing the formulae obtained with 3.12 (ii), we see
that (zw \ = Fw .
5.8. COROLLARY. zw = ^sw,s <g) sw.w, where the summation extends
over those w 6 W for which l(w'w) = l(w) + l(w').
In § 1 we have defined an action of W on H^iX, Z). This definition
depended on the choice of a compact subgroup K. Using Theorem 5.7 we
can find explicitly the correspondences giving this action.
In fact, it follows from Lemma 3.3 (iii) that aa = a*Fa - 1 for any
α e Σ. The transformation Fa is given by the correspondence ZOa. The
operator a* can also be given by a correspondence: if Ua = SCjf/j is a
divisor in X giving the cycle ef(a) 6 H2r^2{X, Z) (for example,
Ua = Σ α(Ηβ)Χσβ), then the cycle Ua = Σ^£/{, where
Uι = {(χ, χ) | χ ζ Ut} cz Χ Χ Χ, determines the correspondence that gives
the operator a*. The operator aa in H^(X, Z) is therefore given by the
correspondence Lfait.ZOa - 1 (where * denotes the product of correspon-
dences, as in [11]). Using the geometrical realization of the product of
correspondences (see [11]), we can explicitly determine the correspondence
Sa that gives the transformation 1 + aa in Η*(Χ, Ζ), namely, Sa = Σο,-ϋ;
r
where ty = {{x, y) 6 Χ χ Χ \ χ 6 Uit ~x- y ζ Ρ{α}} • In this expression,
3c, y 6 G are arbitrary representatives of x, y, and P{ay is the parabolic
subgroup corresponding to the root a.
3. B. Kostant has described the Pw in another way. We state his result.
 Schubert cells and cohomology of the spaces G/P 25

Let h £ i)q be an element such that a(h) > 0 for all a £ Σ. Let
Jh = {/ £ R I f(wh) = 0 for all we W} be an ideal of /?.
5.9. THEOREM, (i) Let w & W, l(w) = /. There is a polynomial Qw £ R
of degree I such that
(21) Qw{wh) = 1, Qw (w'h) = 0 if I {w') < Ζ (w), w' Φ w.
The Qw are uniquely determined by (21) to within elements of Jh. (ii) Let
Q% be the form of highest degree in the polynomial Qw. The image of (?£
in R is equal to Π (y(h))-1-Pw.
ν£Δ_ Γ) «!-1Δ +
The proof is analogous to that of Theorem 3.15.
4. We choose a maximal compact subgroup KcG such that Κ π Β c Η
(see § 1). The cohomology of X can be described by means of the
ΛΓ-invariant closed differential forms on X. For let χ G Ijz, and let Ex be the
corresponding one-dimensional complex G-fibering on X. Let ω χ be the
2-form on X which is the curvature form of connectedness associated with
the ΛΤ-invariant metric on Ex (see [12J). Then the class of the form
ω χ — ω χ i s c x £ IP(X, Z). The mapping χ -* ωχ extends to a mapping
Θ: R ->• Ω*υ(Χ), where Ω*^ is the space of differential forms of even degree
on X. One can prove the following theorem, which is a refinement of
Proposition 1.3 (ii) and Theorem 3.17.
5.1 £. THEOREM (i) Ker θ = J, that is, θ induces a homomorphism of
rings Θ: R -*• Ω,£υ(Χ). (ii) Let Wi, w2 £ W, wx 4 w2s. Then the restriction
of the form J{PWi ) to XWi is equal to 0. (iii) Let νυχ, w2 £ W, wx 4 w2s.
Then HPWi ) e(PWj) = 0.

References
[1] A. Borel, Sur la cohomologie des espaces fibres principaux et des espaces homogenes
des groupes de Lie compacts, Ann. of Math. (2) 57 (1953), 115-207; MR 14 #490.
Translation: in 'Rassloennyeprostranstva', Inost. lit., Moscow 1958
[2] A. Borel, Kahlerian coset spaces of semisimple Lie groups, Proc. Nat. Acad, Sci.
U.S.A. 40 (1954), 1147-1151; MR 17 # 1108.
[3] B. Kostant, Lie algebra cohomology and generalized Schubert cells, Ann. of Math. (2)
77 (1963), 72-144; MR 26 # 266.
[4] G. Horrocks, On the relations of S-functions to Schubert varieties, Proc. London
Math. Soc. (3) 7 (1957), 265-280; MR 19 #459.
[5] A. Borel, Linear algebraic groups, Benjamin, New York, 1969; MR #4273.
Translation: Lineinye algebraicheskiegruppy, 'Mir', Moscow 1972.
[6] N. Bourbaki, Groupes et algebras de Lie, Ch. 1-6, Elements de mathematique, 26, 34,
36, Hermann & Cie, Paris, 1960-72.
Translation: Gruppyialgebry Li, 'Mir', Moscow 1972.
[7] R. Steinberg, Lectures on Chevalley groups, Yale University Press, New Haven,
Conn. 1967.
[8] M. F. Atiyah and F. Hirzebruch, Vector bundles and homogeneous spaces, Proc.
26 /. Ν. Bernstein, I. M. Gel'fand, S. I. Gel'fand

Sympos. Pure Math.; Vol. Ill, 7-38, Amer. Math. Soc, Providence, R. I., 1961;
MR 25 #2617.
= Matematika 6: 2 (1962), 3-39.
[9] P. Cartier, On H. Weyl's character formula, Bull. Amer. Math. Soc. 67 (1961),
228-230; MR 26 #3828.
= Matematika 6: 5 (1962), 139-141.
[10] Ε. Η. Spanier, Algebraic topology, McGraw-Hill, New York 1966; MR 35 # 1007
Translation: Algebraicheskaya topologiya, 'Mir', Moscow 1971.
[11] Yu.I. Manin, Correspondences, motives and monoidal transformations, Matem.
Sb. 77 (1968), 475-507; MR 41 # 3482
= Math. USSR-Sb. 6 (1968), 439-470.
[12] S. S. Chern, Complex manifolds, Institute de Fisca e Matematica, Recife 1959;
MR 22 #1920.
Translation: Kompleksnye mnogoobraziya, Inost. Lit., Moscow 1961.
[13] I. N. Bernstein, I. M. Gel'fand and S. I. Gel'fand, Schubert cells and the cohomology
of flag spaces, Funkts. analiz 7: 1 (1973), 64—65.

Received by the Editors 13 March 1973

Translated by D. Johnson