A Caltech Library Service

Quantum Monte Carlo: Faster, More Reliable, and More Accurate


Anderson, Amos Gerald (2010) Quantum Monte Carlo: Faster, More Reliable, and More Accurate. Dissertation (Ph.D.), California Institute of Technology. doi:10.7907/KVTV-N754.


The Schroedinger Equation has been available for about 83 years, but today, we still strain to apply it accurately to molecules of interest. The difficulty is not theoretical in nature, but practical, since we're held back by lack of sufficient computing power. Consequently, effort is applied to find acceptable approximations to facilitate real time solutions. In the meantime, computer technology has begun rapidly advancing and changing the way we think about efficient algorithms. For those who can reorganize their formulas to take advantage of these changes and thereby lift some approximations, incredible new opportunities await.

Over the last decade, we've seen the emergence of a new kind of computer processor, the graphics card. Designed to accelerate computer games by optimizing quantity instead of quality in processor, they have become of sufficient quality to be useful to some scientists. In this thesis, we explore the first known use of a graphics card to computational chemistry by rewriting our Quantum Monte Carlo software into the requisite "data parallel" formalism. We find that notwithstanding precision considerations, we are able to speed up our software by about a factor of 6.

The success of a Quantum Monte Carlo calculation depends on more than just processing power. It also requires the scientist to carefully design the trial wavefunction used to guide simulated electrons. We have studied the use of Generalized Valence Bond wavefunctions to simply, and yet effectively, captured the essential static correlation in atoms and molecules. Furthermore, we have developed significantly improved two particle correlation functions, designed with both flexibility and simplicity considerations, representing an effective and reliable way to add the necessary dynamic correlation. Lastly, we present our method for stabilizing the statistical nature of the calculation, by manipulating configuration weights, thus facilitating efficient and robust calculations.

Our combination of Generalized Valence Bond wavefunctions, improved correlation functions, and stabilized weighting techniques for calculations run on graphics cards, represents a new way for using Quantum Monte Carlo to study arbitrarily sized molecules.

Item Type:Thesis (Dissertation (Ph.D.))
Subject Keywords:kmc; monte carlo; gvb; gpu; qmc
Degree Grantor:California Institute of Technology
Division:Chemistry and Chemical Engineering
Major Option:Chemistry
Thesis Availability:Public (worldwide access)
Research Advisor(s):
  • Goddard, William A., III
Thesis Committee:
  • Kuppermann, Aron (chair)
  • Miller, Thomas F.
  • McKoy, Basil Vincent
  • Goddard, William A., III
Defense Date:29 June 2009
Non-Caltech Author Email:nitroamos (AT)
Record Number:CaltechETD:etd-08122009-151332
Persistent URL:
Default Usage Policy:No commercial reproduction, distribution, display or performance rights in this work are provided.
ID Code:5276
Deposited By: Amos Anderson
Deposited On:17 Dec 2009 17:53
Last Modified:08 Nov 2019 18:06

Thesis Files

PDF - Final Version
See Usage Policy.


Repository Staff Only: item control page