This talk is for a more sophisticated audience, going over Perelman's noncollapsing estimate for the adjoint heat equation and my nonconic estimate.

We consider random matrices associated to random walks on the complete
graph with random weights. When the weights have finite second moment we
find Wigner-like behavior for the empirical spectral density. If the
weights have finite fourth moment we prove convergence of extremal
eigenvalues to the edge of the semi-circle law. The case of weights with
infinite second moment is also considered. In this case we prove
convergence of the spectral density on a suitable scale and the limiting
measure is characterized in terms certain Poisson weighted infinite
trees associated to the starting graph. Connections with recent work on
random matrices with i.i.d. heavy-tailed entries and several open
problems are also discussed. This is recent work in collaboration with
D. Chafai and C. Bordenave (from Univ. P.Sabatier, Toulouse - France).