1110.6285 (Alexander Rothkopf)

Improved Maximum Entropy Analysis with an Extended Search Space    [PDF]

Alexander Rothkopf

I propose an improvement to the implementation of the well established Maximum Entropy Method by departing from Bryan's concept of singular search space. Working with numerical data from Lattice QCD I found that in cases where prior information is scarce and the number of data-points small, extension of the search space can dramatically improve the reconstructed spectra. The reason for the inadequacy of Bryan's method and a remedy are discussed based on the shape of the basis functions of the search space and results from a mock data analysis. In order to adequately approach problems where an exponentially damped Kernel is used, I provide an implementation using the C/C++ language that utilizes high precision arithmetic adjustable at run-time. The LBFGS algorithm is included in the code in order to attack problems without the need to resort to a particular search space restriction.

View original: http://arxiv.org/abs/1110.6285