Fourier transform over semi-simple algebras and harmonic analysis for probabilistic algorithms (Q1893975)

A problem of considerable interest in applied probability is to compute the convolution powers of probability measures on finite groups and to give estimates for the rate of convergence to the uniform distribution. In particular, D. Aldous, P. Diaconis, and others have achieved lots of results for concrete random walks on cyclic groups, the hypercubes \(\mathbb{Z}^n_2\), the symmetric group \(S_n\) and so on by using the duals of these groups and Fourier analytic techniques. The paper under review has a more algebraic flair. The authors consider the group algebra \(\mathbb{C} [G]\) as well as suitable subalgebras which turn out to be commutative for many examples. They investigate the primitive idempotents of these semisimple algebras instead of irreducible group representations or characters. The main object of the paper are the descent algebras \(\Gamma [S_n] \subset \mathbb{C} [S_n]\) and \(\Gamma [B_n] \subset \mathbb{C} [B_n]\) where \(B_n\) is the hyperoctahedral group. In particular, based on some combinatorial identities of A. Garsia, explicit relations between the basis consisting of primitive idempotents and the standard basis of the algebras above are computed.