Discrete probability distributions and moment problem: numerical aspects

摘要

Discrete probability distributions with finite range, generated by maximum entropy technique and constrained by first moments have been considered. Numerical aspects, as stability of parameter's computation, entropy variation, spectral properties of involved Hankel's matrices have been studied. The ill-conditioning of Hankel's matrices suggests minimization methods like Newton's should be unreliable as a high number of moments are kept, while methods relying upon relative error estimate between assigned and calculated moments should be possible candidates.