Cohomology of Flag Varieties - MathOverflow

Cohomology of Flag Varieties

2010-04-17T09:19:47Z

For $K$ a compact Lie-group with maximal torus $T$ , I'd like to know the cohomology $\text{H}^{\ast}(K/T)$ of the flag variety $K/T$ .

If I'm not mistaken, this should be isomorphic to the algebra of coinvariants of the associated root system, according to the "classical Borel picture", as it is often called in the literature (sadly, often without a reference). Unfortunately, Borels original paper is quite long and so I have the following

Question: Does anybody know a short proof for the theorem?

For example, can it be proved using a direct computation of the Lie-algebra cohomology $\text{H}^{\ast}({\mathfrak k},{\mathfrak t})$ ?

As a side-question: Is it correct that for a complex semisimple Lie-group $G$ with Borel $B$ , compact real form $K$ and maximal torus $T$ the map $K/T\to\ G/B$ is a homotopy equivalence? What is a good reference for these things?

I'm sorry if this is too elementary for MO, but apart from Borel's original paper I couldn't find good sources.

Answer by Jim Humphreys for Cohomology of Flag Varieties

2010-04-17T11:48:22Z

Borel's lengthy 1953 Annals paper is essentially his 1952 Paris thesis. It was followed by work of Bott, Samelson, Kostant, and others, which eventually answers your side question affirmatively. For a readable modern account in the setting of complex algebraic groups rather than compact groups, try to locate a copy of the lecture notes: MR649068 (83h:14045) 14M15 (14D25 20F38 57N99 57T15) Hiller, Howard, Geometry of Coxeter groups. Research Notes in Mathematics, 54. Pitman (Advanced Publishing Program), Boston, Mass.-London, 1982. iv+213 pp. ISBN 0-273-08517-4. (This was based on his 1980 course at Yale. Eventually he left mathematics to work for Citibank.) The identification of the cohomology ring with the coinvariant algebra of the Weyl group has continued to be important for algebraic and geometric questions, for instance in the work of Beilinson-Ginzburg-Soergel. While Hiller's notes are not entirely self-contained, they are helpfully written. (But note that his short treatment of Coxeter groups has a major logical gap.)

ADDED: In Hiller's notes, Chapter III (Geometry of Grassmannians) is most relevant. For connections with Lie algebra cohomology, the classical paper is: MR0142697 (26 #266) 22.60 (17.30) Kostant, Bertram, Lie algebra cohomology and generalized Schubert cells. Ann. of Math. (2) 77 1963 72–144. Nothing in this rich circle of ideas can be made quick and easy; a lot depends on what you already know.

P.S. Keep in mind that Hiller tends to give explicit details just for the general linear group and grassmannians, but he also points out how the main results work in general, with references. I don't know a more modern textbook reference for this relatively old work. But the intuitive connection between the Borel picture and the Bott/Kostant cohomology picture is roughly this: The Lie subalgebra spanned by negative root vectors plays the role of tangent space to the flag manifold/variety. In the Lie algebra cohomology approach you get an explicit graded picture for each degree in terms of number of elements in the Weyl group of a fixed length, whereas the Borel description in terms of Weyl group coinvariants makes the algebra structure of cohomology more transparent. (What I don't know is whether a simpler proof of Borel's theorem can be derived using Lie algebra cohomology.)

Concerning the relationship between $K/T$ and $G/B$ , this goes back to the work around 1950 on topology of Lie groups (Iwasawa, Bott, Samelson): all the topology of a connected, simply connected Lie group comes from a maximal compact subgroup. So the two versions of the flag manifold are homeomorphic. In later times, emphasis has often shifted to treating $G$ as a complex algebraic group, so that $G/B$ is a projective variety. For me the literature is hard to compactify.

One more reference, which treats the Borel theorem in a semi-expository style: MR1365844 (96j:57051) 57T10 Reeder, Mark (1-OK), On the cohomology of compact Lie groups. Enseign. Math. (2) 41 (1995), no. 3-4, 181–200.

Answer by Allen Knutson for Cohomology of Flag Varieties

2010-04-17T15:46:15Z

Note that your statement is only true in rational cohomology. For example, $H^\ast(SO(5)/T)$ is not generated in degree $2$ (though it is rationally).

The easiest proof I know starts from equivariant cohomology:

$ H^\ast_T(K/T) = H^\ast_{T\times T}(K) = H^\ast_{T\times K\times T}(K\times K) = H^\ast_K(K/T \times K/T) $

So far this uses $H^\ast_F(X) = H^\ast(X/F)$ for free actions. Now use the equivariant Künneth formula:

$... = H^\ast_K(K/T) \otimes_{H^\ast_K} H_K(K/T) = H^\ast_T \otimes_{H^\ast_K} H^\ast_T$

Rationally, the base ring $H^\ast_K$ is $(H^\ast_T)^W$, the invariants. Since you didn't want equivariant but ordinary cohomology, kill the left factor, leaving ${\mathbb Q} \otimes_{(H^\ast_T)^W} H^\ast_T$, which is your desired ring of coinvariants.

(I'm having a bunch of trouble with $H$ vs. $H^\ast$ in typesetting here, sorry!)

Answer by Leonid Positselski for Cohomology of Flag Varieties

2010-04-17T21:22:05Z

Concerning the side question, the map K/T → G/B is actually an isomorphism of real manifolds, not just a homotopy equivalence. Not sure about the references, but this is essentially textbook material (unfortunately I forgot where I learned about this). Considering the case of U(n) ⊂ GL(n) acting on the flags in Cⁿ is instructive.