Quantum chemistry by random walk: A faster algorithm

James B. Anderson

Published Jun 18, 2007• 0 citations

Abstract

Abstract This little paper describes two procedures for achieving higher speeds in fixed-node diffusion QMC calculations with importance sampling. The first is an iterative procedure for reducing timestep error, thereby allowing larger and fewer time steps to obtain a given accuracy. The movement of a walker is divided into two parts: a diffusion step selected at random from an appropriate distribution and a drift step calculated from the drift velocity at the initial location. These give a temporary step to a new location for which the drift velocity is again determined. The average of the two velocities together with the original diffusion step is used to determine a permanent step. In tests, the procedure was found to reduce time-step error by a factor of 5 to 10. The second procedure makes use of two trial functions for importance sampling, the first a simple function for walker motion and the second a more complex and more accurate function for determining energies. These allow the more computationally expensive function to be used at large intervals for samples having a lower serial correlation.

DOI: 10.1093/oso/9780195310108.003.0035

Related papers

Quantum chemistry by random walk: Higher accuracy

James B. Anderson

Abstract In an earlier papera a method for the direct calculation of the difference between a true wavefunction and a trial wavefunction was described for simple diffusion calculations. This paper reports further increases in accuracy obtained by a combination of this difference method with importance sampling. The combination allows the calculation of corrections to the product ΨΨT of true and trial wavefunctions, rather than to the wavefunction Ψ itself. Using a new function g = ΨΨT- Ψ2T, the original diffusion equation with drift and local energies is altered by replacement of ΨΨT with g and by addition of a distributed source term. Since walkers fed by sampling the source term may be positive or negative, cancellation is required, but both sets of walkers proceed to the same distribution and age-based cancellation may be used. Node locations are not changed and remain at their original locations as specified by the trial function. The corrected wavefunction and the energy obtained are the same as for a fixed-node diffusion calculation using the same trial function for importance sampling. The method is useful in reducing statistical error, and because the statistical error occurs only in the difference terms, it is most effective when the trial wavefunction is already reasonably accurate. In test calculations for the hydrogen atom and the helium atom, both nodeless, the statistical error in the QMC calculations was greatly reduced. Thus, for hydrogen a trial function with an expectation value of -0.4998 hartrees led to a QMC energy of -0.499995(4) hartrees.

Jun 18, 20070 citations

Quantum chemistry by random walk: A faster algorithm

James B. Anderson

Two extensions of the fixed-node random walk method of solving the Schrödinger equation are described. A simple iterative procedure is found to reduce time-step error and the use of two trial wave functions in place of one for importance sampling is found to reduce computation effort. In combination the two modifications reduce computation effort for a fixed accuracy by a factor of about 10.

The Journal of Chemical PhysicsMar 15, 198520 citations

Quantum chemistry by random walk: Importance sampling for H+3

Frank Mentch, James B. Anderson

Abstract In this paper the combination of importance sampling° with diffusion QMC was shown to be remarkably effective in improving the accuracy of electonic structure calculations for molecular systems. The key to success was the use of a simple, fairly accurate trial wavefunction ΨT for which the function itself, its derivatives and the local energy Eioc may be easily computed. This, used in combination with the diffusion-with-drift expression for walkers corresponding to the product f = ΨΨT according to allows the energy to be evaluated as an average for all walkers of every step after reaching a fluctuating steady state. Since there is a time-step error associated with the simulation of diffusion, drift, and multiplication, varying the time-step and an extrapolation to zero step size are required. The result for H3+ in its triangular configuration of side length 1.6500 bohr was -1.3439 ± 0.0002 hartrees, a value well below the lowest variational energy levels at the time and having an error bar straddling the best modern value of -1.343 835 hartrees. The use of importance sampling along with higher computer speeds resulted in an uncertainty about a factor of 60 lower than that of the previous calculation.b

Jun 18, 20071 citation

A quantum Monte Carlo calculation of the ground state energy of the hydrogen molecule

Carol A. Traynor, James B. Anderson, Bruce M. Boghosian

Abstract Since the wavefunction for the ground state of the hydrogen molecule has no nodes, its energy can be determined in QMC without node location error. This paper reports such calculations for the full four-body (two-electron, two-proton) problem to obtain the energy without the use of the Born-Oppenheimer approximation or any other adiabatic approximations. The calculations were carried out with diffusion QMC and with Green’s function QMC. Importance sampling was incorporated into each of the calculations. For the diffusion calculations, extrapolation of energies obtained at several different time-step sizes was required. For the Green’s function calculations there was no time-step error, and sign changes for walkers were avoided by use of a potential energy offset and a minor approximation to eliminate positive potential energies. The energies so determined were not exact but had extremely small systematic errors. The large differences in electron and proton masses led to extensive computation requirements due to the slow equilibration and serial correlation induced by the heavier protons. The calculations were executed on one of the first massively parallel computers, a Thinking Machines CM-2 with 65,536 processors. Frequent communication among the processors was required to balance the number of walkers treated in each.

Jun 18, 20070 citations

Quantum Monte Carlo : Reduction of time‐step error with the wave function correction method

Hae‐Won Kim, James B. Anderson

Abstract We compare the accuracy and precision of the wave function correction method to that of traditional diffusion quantum Monte Carlo (traditional DQMC) simulations. Direct calculation of corrections to trial wave functions reduces time step error for Be and Ne electronic ground state energies when compared to traditional DQMC results. The fixed‐node method is utilized for these fermionic systems, so error due to incorrect node locations is not addressed. This correction DQMC method achieves more accurate energies at larger time steps, promising decreased computational cost to reach a given accuracy. For an optimized Be trial wave function, the statistical uncertainty per unit of computational effort is also reduced relative to traditional DQMC. Correction DQMC utilizes walker feed integrals and growth factors, as reported previously. We illustrate the method in a simple harmonic oscillator example, and we report high accuracy energies for Be and Ne at several time steps.

International Journal of Quantum ChemistryNov 14, 20220 citations

Reptation quantum Monte Carlo: A method for unbiased ground-state averages and imaginary-time correlations

Stefano Baroni A, Saverio Moroni

Abstract This paper introduces a successful new method for extending diffusion QMC with importance sampling to allow the direct calculation of pure ‘¢ 2 distributions in place of mixed ‘¢ 1/Jr distributions. The method removes any bias introduced with a trial function (except that due to node locations) in the calculation of observable quantities which do not commute with the Hamiltonian. It has the advantage, relative to the descendent weighting or forward walking technique,a of a controlled walker population which reduces statistical errors. The authors have named it the reptation quantum Monte Carlo (RQMC) method. Its basic object is a path or sequence of configurations called a “reptile” which may be altered by an action called “reptation,” terms adopted from related work in the polymer area. In conventional diffusion QMC with importance sampling walkers are subjected to diffusion, drift, and a local energy-based multiplication to produce a 1/J’I/Jr distribution. Without multiplication a ‘l/Jr’l/Jr distribution is produced. In RQMC multiplication is replaced by a Metropolis rejection procedure, also based on local energies, which produces a 1/J’!/J distribution of configurations in a reptile. The generation of a new reptile, by removing the tail and adding a new head, is based on diffusion and drift for proposed moves coupled with an acceptance test. Overall, the implementation is quite similar to that for variational QMC. As described, the method was tested first for the hydrogen atom, followed by the successful demonstration of a calculation for a helium fluid, 64 4He atoms in a cubic box with periodic boundary conditions. The advantages of RQMC relative to branching diffusion QMC were pointed out for treatment of clusters, films, and superfluids as well as for estimation of electronic forces.

Jun 18, 20070 citations