Passage to H*(G/B)

Arun Ram
Department of Mathematics and Statistics
University of Melbourne
Parkville, VIC 3010 Australia
aram@unimelb.edu.au

Last update: 21 February 2013

Passage to H*(G/B)

Let us explain how our results in K(G/B) are related to the cohomology H*(G/B) and Schubert polynomials. The transfer is by way of the Chern character ch.

If X is a finite CW complex then the Chern character (see [Mac1968, Ch. 10], [Hir1995, §23-24], [Har1977, App. A])

ch:Kvb(X)⊗ ℚ⟶H*(X;ℚ)

is a natural ring isomorphism, i.e. if f:X→Y is continuous then ch(f!(x))= f*(ch(x)). If f:X→Y is a morphism of nonsingular projective varieties then the Grothendieck-Riemann-Roch theorem [Har1977, App. A Theorem 5.3], henceforth G-R-R, says

ch(f!(x))= f*(ch(x)td(𝒯f)),

where td(𝒯f) is Todd class of the relative tangent sheaf of f. If ℒ is a line bundle on X with first Chern class λ∈H2(X;ℚ) then the Chern character and the Todd class of ℒ are the elements of H*(X) given by

ch(ℒ)=eλ= ∑k≥0 λk k! andtd(ℒ)= λ1-e-λ, respectively.

The expression eλ is finite sum since λk=0 in H*(X) whenever k>dim(X).

H*(G/B) as the quotient of a polynomial ring

Let X=G/B. Let 𝔥 be the Lie algebra of T and let S(𝔥*) be the ring of polynomials on 𝔥* (over ℚ). This is a polynomial ring in the n variables α1,…,αn (the simple roots). Let ε^:S(𝔥*)→ℚ be the homomorphism given by ε^(λ)=0 for all λ∈𝔥*. If f∈S(𝔥*) then ε^(f) is the constant term of f. It is a classical theorem of Borel (see [BGG1973, Prop. 1.3]) that

H*(G/B;ℚ) ≅S(𝔥*)ℐ^, (5.1)

where ℐ^ is the ideal of S(𝔥*) generated by { f∈S(𝔥*)W  ∣ f-ε^ (f)=0 } . Proposition 1.5 is the K-theory analogue of Borel’s theorem.

Let -λ∈Λ. The element -λ determines a character of T, denoted by eλ∈R(T) (see section 1). Let c1(ℒ-λ)∈H2(G/B) be the first Chern class of the line bundle ℒ-λ=ϕ(eλ) where ϕ:R(T)→K(G/B) is the map in (1.2). Because of the isomorphism in (5.1) we often abuse notation and write λ=c1(ℒ-λ)∈H*(G/B). All of the maps in the following commutative diagram are isomorphisms. (Recall that we can identify K(G/B) and Kvb(G/B).)

R(T)/ℐ⊗ℚ ⟶ϕ Kvb (G/B) ⊗ℚ ↓ ch ↓ ch S(𝔥*) /ℐ^ ⟶ϕ^ H* (G/B;ℚ) where eλ ⟶ϕ [ℒ-λ] ↓ch ↓ch eλ ⟶ϕ^ ec1(ℒ-λ) (5.2)

The left hand ch map is obtained by viewing R(T) and S(𝔥*) as subsets of K(ℬT) and H*^(ℬT), respectively, where ℬT is the classifying space of T.

The relation between Tα and ∂α

Let α be a simple root and let Pα be the parabolic subgroup with Lie algebra 𝔭α spanned by 𝔟 and the root space 𝔤-α. Let fα:G/B→ G/Pα corresponding ℙ1-bundle.

A proof of Proposition 3.3:

Since K(G/B) is torsion free, we can check the identity (fα)! (fα)! (x)=Tα (x) by applying ch to both sides and checking the result in cohomology.

The G-R-R for the map fα says

ch((fα)!(x))= (fα)* (ch(x)td(𝒯fα)), (5.3)

where 𝒯fα is the bundle of tangents along the fibres, which is the line bundle associated to the ad(B)-module 𝔭α/𝔟. Since this module has weight -α, c1(𝒯fα)= α∈H2 (G/B:ℤ) and so (5.3) becomes

ch((fα)!(x)) =(fα)* ( ch(x) α1-e-α ) . (5.4)

The BGG operator ∂α= (fα)* (fα)* is explicitly given by the formula

∂α(z)= z-sα(z)α, (5.5)

see [Dem1974]. Thus, by applying (fα)* to both sides of (5.4) we obtain

ch ( (fα)! (fα)! (x) ) = (fα)* (fα)* ( ch(x) α1-e-α ) =∂α ( ch(x) α1-e-α ) . (5.6)

The strategy now is to manipulate the right hand side of (5.6) using the “skew-Leibniz” rule satisfied by ∂α to obtain ch(Tα(x)). For convenience, let y=ch(x). Then the right side of (5.6) is

∂α (yα1-e-α)= y∂α (α1-e-α) +sα (α1-e-α) ∂α(y)

and we claim

(a)sα (α1-e-α)= αeα-1, (b)∂α (α1-e-α)= 1.

The first equality is trivial and the second can be proved by formal computation (carefully done!) or by applying the G-R-R again. Using (a) and (b) we obtain

∂α (yα1-e-α)= y+αeα-1 (y-sα(y)α).

Now cancelling the α's' in the second term on the right and recalling that y=ch(x) we find

ch ( (fα)! (fα)! (x) ) =ch(Tα(x)).

□

We see that

ch(Tα(x))= ∂α ( ch(x) α1-e-α ) (5.7)

which relates to Tα to ∂α in an explicit way. In fact, if one wished one could reverse the argument and derive the formula (5.5) for ∂α from the formula for Tα.

A proof of Proposition 1.6

Proposition 1.6. If f:G/B→G/P is the natural projection then the induced map f!:K(G/P)→K(G/B) is an injection.

Proof.

Since K(G/P) is torsion free there is a natural injection K(G/P)↪ K(G/P)⊗ℚ→ H*(G/P;ℚ) and so it suffices to check that the pull-back f* in rational cohomology is injective. Since the odd cohomology groups of the base and fiber are zero the Serre spectral sequence of the bundle P/B→G/B→G/P shows that f* is injective even for H*(FP ;ℤ).

□

Dictionary between K(G/B) and H*(G/B)

In summary, the Chern character gives an isomorphism

R(T)/ℐ ≅ K(G/B) ⟶ch H*(G/B) ≅ S(𝔥*)/ℐ^ eλ ⟼ eλ,

where

ℐ=ideal generated by  { f∈R(T)W ∣  f-ε(f)=0 } , ε: R(T) ⟶ ℤ eλ ⟼ 1, ℐ^=ideal generated by  { f∈S(𝔥*)W ∣  f-ε^(f)=0 } , ε^: S(𝔥*) ⟶ ℤ λ ⟼ 0.

Let [Xw]∈H*(G/B) be the element which is Poincaré dual to the fundamental cycle of Xw in H*(G/B). This element is called a Schubert polynomial. Then

K(G/B) has basis  { [𝒪Xw] ∣  w∈W } andH* (G/B) has basis  {[Xw] ∣ w∈W} .

From a general fact (see [Ful1984, Ex. 15.2.16] or [CGi1433132, 5.8.13(i) and p. 289])

ch([𝒪Xw])= [Xw]+higher degree terms.

where deg([Xw])= dim(G/B)- dim(Xw)=N-ℓ (w), where N  is the number of positive roots for 𝔤.

If α is a simple root and fα:G/B→G/Pα is the corresponding ℙ1-bundle, then

Tα(x)= (fα)! (fα)!(x)= eαx-sα(x) eα-1 and∂α(x)= (fα)* (fα)*(x)= x-sα(x)α

in K(G/B) and H*(G/B), respectively. As illustrated in (5.7) each of these two formulas can be derived from the other via the use of the Grothendieck-Riemann-Roch Theorem. This means that the following formulas (Proposition 3.4 and [BGG1973, Th. 3.14])

Tα([𝒪Xw])= { [𝒪Xwsα], if wsα>w, [𝒪Xw], if wsα<w, and∂α ([Xw])= { [Xwsα], if wsα>w, [Xw], if wsα<w,

are equivalent.

Our new Pieri-Chevalley formula in K(G/B), Theorem 4.3, and Chevalley’s classical Pieri formula in H*(G/B), [Che1994, Prop. 10], are

eλ[𝒪Xw]= ∑η∈𝒯wλ [𝒪Xv(η,w)] andλ·[Xw]= ∑αv→w ⟨λ,α∨⟩ [Xv],

where the sum is over all v∈W such that ℓ(v)=ℓ(w)-1 and there is a root α such that v=sαw. Chevalley’s formula can be obtained from ours formula by subtracting [𝒪Xw] from each side, applying the Chern character ch, and comparing the lowest degree terms on each side.

Notes and References

This is an excerpt of the paper entitled A Pieri-Chevalley formula for K(G/B) authored by H. Pittie and Arun Ram (preprint May 19, 1998).

Research supported in part by National Science Foundation grant DMS-9622985.

page history