Stochastic method for analytic continuation of quantum Monte Carlo data

A method for analytic continuation of quantum Monte Carlo data is presented. The spectrum A(ω) is parametrized as a sum of δ functions, the weights A_i of which are sampled according to a distribution p(A)~exp(-χ²/Θ). It is argued that the calculated entropy S provides a criterion for determining the Θ corresponding to the ``best'' averaged spectrum. The appearance of spurious structure is signaled by a sharp drop in <S(Θ)>, which in test cases is preceded by a local maximum. Results for the dynamic spin structure factor of a 16-site Heisenberg chain obtained at this maximum are in better agreement with exact results than ``classic'' maximum-entropy results.