An affine Hecke algebra H contains a large abelian subalgebra A spanned by the Bernstein-Zelevinski-Lusztig basis elements θx, where x runs over (an extension of) the root lattice. The centre Z of H is the subalgebra of Weyl group invariant elements in A. The natural trace ('evaluation at the identity') of the affine Hecke algebra can be written as integral of a certain rational n-form (with values in the linear dual of H) over a cycle in the algebraic torus T = Spec(A). This cycle is homologous to a union of 'local cycles'. We show that this gives rise to a decomposition of the trace as an integral of positive local traces against an explicit probability measure on the spectrum W 0 \ T of Z. From this result we derive the Plancherel formula of the affine Hecke algebra.
ContentsOn the spectral decomposition of affine Hecke algebras(2.13)
The Iwahori-Hecke algebra as a Hilbert algebraMany of the results of this subsection are well known (see [31]). Let R be a root datum, and let q be a real number with q > 1. We assume that for all s ∈ S aff we are given a real number f s . Throughout this paper we use the convention that the labels as discussed in the previous subsection are defined by Convention 2.1. The labels are of the form(2.14)We write q := (q(s)) s∈S aff for the corresponding label function on S aff . The following theorem is well known.Theorem 2.2. There exists a unique complex associative algebra H = H(R, q) with C-basis (T w ) w∈W which satisfy the following relations.(a) If l(ww ) = l(w) + l(w ), then T w T w = T ww .(b) If s ∈ S aff , then (T s + 1)(T s − q(s)) = 0.Proof . This is elementary and well known. Use the Frobenius-Schur theorem for (iii), Dixmier's version of Schur's lemma for (iv) and Proposition 2.10.
Corollary 2.12 (see also [31]). Restriction to H induces an injection of the setĈ into the spaceĤ of finite-dimensional irreducible * -representations of H.Consequently, the C * -algebra C is of finite type I.
Proof .Because H ⊂ C is dense, it is clear that a representation of C is determined by its restriction to H and that (topological) irreducibility is preserved. Hence, by the previous corollary, all irreducible representations of C have finite dimension.We equip T and W 0 \ T with the analytic topology. Given π ∈Ĉ, we denote by W 0 t π ∈ W 0 \ T the character of Z such that χ π (z) = dim(π)z(t π ) (note that π(Z) can not vanish identically since 1 ∈ Z).By Proposition 2.9, the * -operator on Z is such that z * (t) = z(t −1 ). When π ∈Ĉ, we have χ π (x * ) = χ π (x). It follows that t −1 π ∈ W 0 t π for all π ∈Ĉ. Let us denote byProposition 2.13. The map p z :Ĉ → W 0 \ T defined by p z (π) = W 0 t π is continuous and finite. Its image S = p z (Ĉ) ⊂ W 0 \ T Herm is the spectrumẐ of the closureZ of Z in C. The map p z :Ĉ → S is closed.Proof . It is clear that the image is in W 0 \ T Herm and that the map is finite (by Proposition 2.10). Since W 0 \ T is Hausdorff andĈ is compact, the map p z is closed if it is continuous.