Borel–Weil–Bott theorem

In mathematics, the Borel–Weil-Bott theorem is a basic result in the representation theory of Lie groups, showing how a family of representations can be obtained from holomorphic sections of certain complex vector bundles, and, more generally, from higher sheaf cohomology groups associated to such bundles. It is built on the earlier Borel–Weil theorem of Armand Borel and André Weil, dealing just with the section case, the extension being provided by Raoul Bott. One can equivalently, through Serre's GAGA, view this as a result in complex algebraic geometry in the Zariski topology.

Contents

Formulation

Let G be a semisimple Lie group or algebraic group over \mathbb C, and fix a maximal torus T along with a Borel subgroup B which contains T. Let λ be an integral weight of T; λ defines in a natural way a one-dimensional representation Cλ of B, by pulling back the representation on T = B/U, where U is the unipotent radical of B. Since we can think of the projection map GG/B as a principal B-bundle, for each Cλ we get an associated fiber bundle L on G/B (note the sign), which is obviously a line bundle. Identifying Lλ with its sheaf of holomorphic sections, we consider the sheaf cohomology groups H^i( G/B, \, L_\lambda ). Since G acts on the total space of the bundle L_\lambda by bundle automorphisms, this action naturally gives a G-module structure on these groups; and the Borel–Weil–Bott theorem gives an explicit description of these groups as G-modules.

We first need to describe the Weyl group action centered at \rho. For any integral weight \lambda and w in the Weyl group W, we set w*\lambda�:=  w( \lambda %2B \rho ) - \rho \,, where \rho denotes the half-sum of positive roots of G. It is straightforward to check that this defines a group action, although this action is not linear, unlike the usual Weyl group action. Also, a weight \mu is said to be dominant if \mu( \alpha^\vee ) \geq 0 for all simple roots \alpha. Let \ell denote the length function on W.

Given an integral weight \lambda, one of two cases occur: (1) There is no w \in W such that w*\lambda is dominant, equivalently, there exists a nonidentity w \in W such that w * \lambda = \lambda; or (2) There is a unique w \in W such that w * \lambda is dominant. The theorem states that in the first case, we have

H^i( G/B, \, L_\lambda ) = 0 for all i;

and in the second case, we have

H^i( G/B, \, L_\lambda ) = 0 for all i \neq \ell(w), while
H^{ \ell(w) }( G/B, \, L_\lambda ) is the dual of the irreducible highest-weight representation of G with highest weight \lambda.

It is worth noting that case (1) above occurs if and only if \lambda( \beta^\vee ) = 0 for some positive root \beta. Also, we obtain the classical Borel–Weil theorem as a special case of this theorem by taking \lambda to be dominant and w to be the identity element e \in W.

Example

For example, consider G = SL2(C), for which G/B is the Riemann sphere, an integral weight is specified simply by an integer n, and ρ = 1. The line bundle Ln is O(n), whose sections are the homogeneous polynomials of degree n (i.e. the binary forms). As a representation of G, the sections can be written as Symn(C2)*, and is canonically isomorphic to Symn(C2). This gives us at a stroke the representation theory of \mathfrak{sl}_2(\mathbf{C}): Γ(O(1)) is the standard representation, and Γ(O(n)) is its n-th symmetric power. We even have a unified description of the action of the Lie algebra, derived from its realization as vector fields on the Riemann sphere: if H, X, Y are the standard generators of \mathfrak{sl}_2(\mathbf{C}), then we can write

H = x\frac{d}{dx}-y\frac{d}{dy}
X = x\frac{d}{dy}
Y = y\frac{d}{dx}.

Positive characteristic

One also has a weaker form of this theorem in positive characteristic. Namely, let G be a semisimple algebraic group over an algebraically closed field of characteristic p > 0. Then it remains true that H^i( G/B, \, L_\lambda ) = 0 for all i if \lambda is a weight such that w*\lambda is non-dominant for all w \in W. However, the other statements of the theorem do not remain valid in this setting.

More explicitly, let \lambda be a dominant integral weight; then it is still true that H^i( G/B, \, L_\lambda ) = 0 for all i > 0, but it is no longer true that this G-module is simple in general, although it does contain the unique highest weight module of highest weight \lambda as a G-submodule. If \lambda is an arbitrary integral weight, it is in fact a large unsolved problem in representation theory to describe the cohomology modules H^i( G/B, \, L_\lambda ) in general. Unlike over \mathbb{C}, it need not be the case for a fixed \lambda that these modules are all zero except in a single degree i.

References

This article incorporates material from Borel–Bott–Weil theorem on PlanetMath, which is licensed under the Creative Commons Attribution/Share-Alike License.