Random incidence matrices: Moments of the spectral density

We study numerically and analytically the spectrum of incidence matrices of random labeled graphs on N vertices: any pair of vertices is connected by an edge with probability p. We give two algorithms to compute the moments of the eigenvalue distribution as explicit polynomials in iv and p. For large N and fixed p, the spectrum contains a large eigenvalue at Np and a semicircle of "small" eigenvalues. For large N and fixed average connectivity pN (dilute or sparse random mall ices limit) we show that the spectrum always contains a discrete component. An anomaly in the spectrum near eigenvalue 0 for connectivity close to c is observed. We develop recursion relations to compute the moments as explicit polynomials in pN. Their growth is slow enough so that they determine the spectrum. The extension of our methods to the Laplacian matrix is given in Appendix.