Population control bias with applications to parallel diffusion monte carlo

Jaron T. Krogel, David M. Ceperley

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

4 Scopus citations

Abstract

The future of scientific computing will be driven by highly distributed parallel machines with millions of compute nodes. In order to take advantage of this already arriving wave of computing capability we must identify and remove the remaining barriers to parallel scaling in the Diffusion Monte Carlo algorithm. To address these scaling issues in a simple way, we propose that a time delay be introduced into the population control feedback. In order to assess this algorithm, we investigate the behavior of population fluctuations and the population control bias (which will emerge into greater relevance with larger physical systems and requirements of higher accuracy) in a model system for both the standard and time delayed DMC algorithms. We then condense our findings into a simple set of recommendations to improve the scaling of DMC while managing the population control bias.

Original languageEnglish
Title of host publicationAdvances in Quantum Monte Carlo
PublisherAmerican Chemical Society
Pages13-26
Number of pages14
ISBN (Print)9780841227507
DOIs
StatePublished - 2012
Externally publishedYes

Publication series

NameACS Symposium Series
Volume1094
ISSN (Print)0097-6156
ISSN (Electronic)1947-5918

Fingerprint

Dive into the research topics of 'Population control bias with applications to parallel diffusion monte carlo'. Together they form a unique fingerprint.

Cite this