Chevalley restriction theorem

In the mathematical theory of Lie groups, the Chevalley restriction theorem describes functions on a Lie algebra which are invariant under the action of a Lie group in terms of functions on a Cartan subalgebra.

Statement

Chevalley's theorem requires the following notation:

assumptionexample
Gcomplex connected semisimple Lie groupSLn, the special linear group
𝔤the Lie algebra of G𝔰𝔩n, the Lie algebra of matrices with trace zero
ℂ[𝔤]Gthe polynomial functions on 𝔤 which are invariant under the adjoint G-action
𝔥a Cartan subalgebra of 𝔤the subalgebra of diagonal matrices with trace 0
Wthe Weyl group of Gthe symmetric group Sn
ℂ[𝔥]Wthe polynomial functions on 𝔥 which are invariant under the natural action of Wpolynomials f on the space {x1,…,xn,∑xi=0} which are invariant under all permutations of the xi

Chevalley's theorem asserts that the restriction of polynomial functions induces an isomorphism

ℂ[𝔤]G≅ℂ[𝔥]W.

Proofs

Humphreys (1980) gives a proof using properties of representations of highest weight. Chriss & Ginzburg (2010) give a proof of Chevalley's theorem exploiting the geometric properties of the map 𝔤~:=G×B𝔟→𝔤.

This theorem in fact more generally holds for any reductive Lie algebra, and can also be proved directly in the case that G=GLn. In this case, the Cartan subalgebra 𝔱 can be identified with diagonal matrices, and the Weyl group can be identified with the symmetric group acting on diagonal matrices by permutations. Therefore, ℂ[𝔱]≅ℂ[x1,...,xn] and ℂ[𝔱]W can be identified with the subring of ℂ[𝔱] generated by e1,...,en where each ei is an elementary symmetric polynomial.

Identify 𝔤 with the set of n×n matrices. For injectivity, if we are given a G-invariant polynomial on 𝔤 which vanishes on the set of diagonal matrices, by G-invariance, this polynomial vanishes on any matrix which is conjugate to a diagonal matrix. Since these elements are Zariski dense in 𝔤, this polynomial vanishes on a Zariski dense open subset of 𝔤 and thus must itself be the zero polynomial.

For surjectivity, it of course suffices to show that each ei lies in the image of this map. Using Newton's identities and an inductive argument, it suffices to prove that the function x1i+...+xni lies in the image of this map for every i∈{1,...,n}. However, in this case, one can explicitly construct an invariant function which restricts to x1i+...+xni on diagonal matrices: it is given by the function A↦trace(Ai).