skip to main content
OSTI.GOV title logo U.S. Department of Energy
Office of Scientific and Technical Information

Title: Multiple Walkers in the Wang-Landau Algorithm

Abstract

The mean cost for converging an estimated density of states using the Wang-Landau algorithm is measured for the Ising and Heisenberg models. The cost increases in a power-law fashion with the number of spins, with an exponent near 3 for one-dimensional models, and closer to 2.4 for two-dimensional models. The effect of multiple, simultaneous walkers on the cost is also measured. For the one-dimensional Ising model the cost can increase with the number of walkers for large systems. For both the Ising and Heisenberg models in two-dimensions, no adverse impact on the cost is observed. Thus multiple walkers is a strategy that should scale well in a parallel computing environment for many models of magnetic materials.

Authors:
Publication Date:
Research Org.:
Oak Ridge National Lab. (ORNL), Oak Ridge, TN (United States)
Sponsoring Org.:
USDOE
OSTI Identifier:
885956
Report Number(s):
ORNL/TM-2005/1
TRN: US200617%%269
DOE Contract Number:
DE-AC05-00OR22725
Resource Type:
Technical Report
Country of Publication:
United States
Language:
English
Subject:
36 MATERIALS SCIENCE; 99 GENERAL AND MISCELLANEOUS//MATHEMATICS, COMPUTING, AND INFORMATION SCIENCE; ALGORITHMS; HEISENBERG MODEL; ISING MODEL; MAGNETIC MATERIALS

Citation Formats

Brown, G. Multiple Walkers in the Wang-Landau Algorithm. United States: N. p., 2005. Web. doi:10.2172/885956.
Brown, G. Multiple Walkers in the Wang-Landau Algorithm. United States. doi:10.2172/885956.
Brown, G. Wed . "Multiple Walkers in the Wang-Landau Algorithm". United States. doi:10.2172/885956. https://www.osti.gov/servlets/purl/885956.
@article{osti_885956,
title = {Multiple Walkers in the Wang-Landau Algorithm},
author = {Brown, G},
abstractNote = {The mean cost for converging an estimated density of states using the Wang-Landau algorithm is measured for the Ising and Heisenberg models. The cost increases in a power-law fashion with the number of spins, with an exponent near 3 for one-dimensional models, and closer to 2.4 for two-dimensional models. The effect of multiple, simultaneous walkers on the cost is also measured. For the one-dimensional Ising model the cost can increase with the number of walkers for large systems. For both the Ising and Heisenberg models in two-dimensions, no adverse impact on the cost is observed. Thus multiple walkers is a strategy that should scale well in a parallel computing environment for many models of magnetic materials.},
doi = {10.2172/885956},
journal = {},
number = ,
volume = ,
place = {United States},
year = {Wed Dec 28 00:00:00 EST 2005},
month = {Wed Dec 28 00:00:00 EST 2005}
}

Technical Report:

Save / Share:
  • We present a modified Wang-Landau algorithm for models with continuous degrees of freedom. We demonstrate this algorithm with the calculation of the joint density of states of ferromagnet Heisenberg models and a model polymer chain. The joint density of states contains more information than the density of states of a single variable-energy, but is also much more time consuming to calculate. We present strategies to significantly speed up this calculation for large systems over a large range of energy and order parameter.
  • We implement the Wang-Landau algorithm in the context of SU(N) lattice gauge theories. We study the quenched, reduced version of the lattice theory and calculate its density of states for N=20, 30, 40, 50. We introduce a variant of the original algorithm in which the weight function used in the update does not asymptote to a fixed function, but rather continues to have small fluctuations that enhance tunneling. We formulate a method to evaluate the errors in the density of states, and use the result to calculate the dependence of the average action density and the specific heat on themore » 't Hooft coupling {lambda}. This allows us to locate the coupling {lambda}{sub t} at which a strongly first-order transition occurs in the system. For N=20 and 30 we compare our results with those obtained using Ferrenberg-Swendsen multihistogram reweighting and find agreement with errors of 0.2% or less. Extrapolating our results to N={infinity}, we find ({lambda}{sub t}){sup -1}=0.3148(2). We remark on the significance of this result for the validity of quenched large-N reduction of SU(N) lattice gauge theories.« less
  • We introduce a simple method to compute free energy differences employing Jarzynski identity in conjunction with Wang – Landau algorithm. We demonstrate this method on Ising spin system by comparing the results with those obtained from canonical sampling.
  • Wang Landau sampling is implemented on the Graphics Processing Unit (GPU) with the Compute Unified Device Architecture (CUDA). Performances on three different GPU cards, including the new generation Fermi architecture card, are compared with that on a Central Processing Unit (CPU). The parameters for massively parallel Wang Landau sampling are tuned in order to achieve fast convergence. For simulations of the water cluster systems, we obtain an average of over 50 times speedup for a given workload.
  • The dissertation outlines development of a medical imaging machine which renders 3D images from voxel data within a MIMD multiprocessor architecture at interactive rates. Interactive performance is achieved using local dynamic selection of the optimum adaptive recursive hidden-surface removal algorithm. A survey of the medical imaging, graphics, and medical imaging modality literature is provided. A description of Computerized Technology, Magnetic Resonance Imaging, Positron Emission Tomography, Single Photon Emission Computed Tomography, and Ultrasound imaging modalities is presented Previous work in 3D volume rendering graphics techniques and data models is introduced. Eleven medical imaging machines are examined with emphasis on characterization ofmore » the major innovation(s) and performance of each machine. A five stage image processing pipeline is described.« less