|
Written by Dominik Domin
|
|
Thursday, 10 February 2005 |
This section will describe how to run variational Monte Carlo (VMC) calculations using Zori.
Short VMC calculations are necessary after creating walkers using the create.xml input file and to equilabrate walkers according to trial wave function distribution.
Implemented VMC Algorithms
Metropolis, Rosenbluth, Rosenbluth, Teller and Teller (a.k.a Metropolis
Monte Carlo) N. METROPOLIS, A. W. ROSENBLUTH, M. N. ROSENBLUTH, N. M. TELLER, and E. TELLER, J. Chem. Phys. 21, 1087 (1953).
Langevin accelerated Metropolis C. UMRIGAR, Phys. Rev. Lett. 71, 408 (1993).
Langevin accelerated Metropolis with single electron moves. C. UMRIGAR, Phys. Rev. Lett. 71, 408 (1993).
In the current implementation of Zori the Metropolis VMC algorithm has been found to be more robust than Langevin variational Monte Carlo for trial wav efunctions without a correlation function. Other Accelerated variational Monte Carlo are possible by turning off branching in diffusion Monte Carlo algorithms (VMC=”TRUE” option).To run VMC:
zori -i vmc.xml -p psi.xml -r walkers
Recommended algorithm: Metropolis seems to be more robust with a larger time step. The other Langevin algorthims as currently implemented in Zori seem to be less black box in their application.
VMC Input file options
Algorithm
Type= M(RT)2 or Langevin or LangevinSingle
Length
Steps= number of steps to be moved
Block=number of steps in a block
TimeStep= time step size (how far one can move in a step)
Population
This Tag has the same properties as described before
WalkersPerProcessor= number of walkers on each processor
Punch= print the walkers to file (Default="True")
PunchType= Single each processor has its own walker file
PunchInterval= print walkers to every PunchInterval number of steps
Parameters
Most important thing is for the number of particles to match the number of electrons.
|
|
Last Updated ( Tuesday, 29 March 2005 )
|
