FAST MAXIMUM ENTROPY ALGORITHM FOR ILL-POSED PROBLEMS

The fast algorithm of the Maximum Entropy (MaxEnt) numerical solution of the linear inverse problem is described. The minimization of a general functional intrinsic to the MaxEnt approach is reduced to an iteration procedure with each step being a constrained least-squares problem (minimization of a quadratic functional with linear inequality constraints). The algorithm is structurally simple and can be assembled from blocks available in standard program libraries. The algorithm is tested on “toy” tasks with exponential kernel, as well as on practical problems of the recovery of the spectral density of strongly correlated quantum systems from the imaginary time Green’s functions obtained by Monte Carlo.