Dresden 2009 – scientific programme
Parts | Days | Selection | Search | Downloads | Help
TT: Fachverband Tiefe Temperaturen
TT 24: Correlated Electrons: (General) Theory 2
TT 24.4: Talk
Wednesday, March 25, 2009, 10:15–10:30, HSZ 301
Analytic Continuation of Quantum Monte Carlo Data by Stochastic Analytic Inference — •Sebastian Fuchs1,2, Mark Jarrell2, and Thomas Pruschke1 — 1Institut für Theoretische Physik, Georg-August-Universität Göttingen, Friedrich-Hund-Platz 1, 37077 Göttingen — 2Center for Computation and Technology, Louisiana State University, Baton Rouge, LA 70803, USA
The maximum entropy method is the standard tool for the analytic continuation of imaginary-time quantum Monte Carlo data. It uses arguments of Bayesian logic to obtain the most probable energy spectrum given the imaginary-time input data.
In the past efforts where made to provide an alternative to this standard approach [2]. It was proposed to perform an average over a wide range of spectra using Monte Carlo techniques instead of selecting a single spectrum. So far, the method lacked a rigorous rule to eliminate a free regularization parameter inherent in the algorithm.
We propose an algorithm that is based on Bayesian inference. It utilizes Monte Carlo simulations to both calculate a weighted average of possible spectra and to provide a strict criterion for the elimination of the regularization parameter.
Our implementation is based on the libraries of the ALPS project [3]. ALPS is an open source effort providing libraries and simulation codes for strongly correlated quantum mechanical systems.
[1] M. Jarrell, G. E. Gubernatis, Phys. Rep. 269, 133 (1996).
[2] A. Sandvik, PRB 57, 10287 (1998); K. Beach, cond-mat/0403055
[3] http://alps.comp-phys.org