Fixed-sample optimization in quantum Monte Carlo using a probability density function

R N Barnett, R N Sun, R N Lester

Published Jun 18, 2007• 0 citations

Abstract

Abstract The optimization of trial wavefunctions ‘I/Jr for use in QMC calculations can be accomplished with a fixed set of configurations selected with probabilities proportional to ‘1/Jf, often using Metropolis sampling and occasionally using other methods. By varying the coefficients of ‘I/Jr, one can minimize the variance in local energies for the original set of configurations. This is effective except when coefficients affecting the node locations are varied and occasional very high weighting factors ‘I/J}(new)/’1/J}(original) can give large fluctuations in the variance. In this paper, the authors propose the use of probability density functions p to produce nodeless distributions of points replacing the original ‘1/Jf distributions. This has two advantages: it avoids the unbounded weighting problem allowing nodes to be varied, and it facilitates sampling without a Metropolis walk and its associated serial correlation. To be effective, i.e., for high efficiency, p must be carefully selected, it must at least resemble ‘1/Jf, and its distributions must be easily sampled. Tests were performed for the case of CH with optimization of all parameters of a two-determinant-and-Jastrow trial function. These included basis set exponents, MO coefficients, CI coefficient, and SchmidtMoskowitz Jastrow parameters. The resulting variance in local energy was 0.19 h2, versus 0.25 h2 for optimization of the Jastrow coefficients only. More interesting is the improvement in diffusion QMC energy for the complete optimization including node movement. The energy was lowered by 0.004 hartree, or 2.5 kcal/mol, with recovery of correlation energy increased from 92.8 to 94.8%.

DOI: 10.1093/oso/9780195310108.003.00108

Related papers

Quantum chemistry by quantum Monte Carlo: Beyond ground-state energy calculations

PETER J. REYNOLDS, R. N. Barnett, B. L. Hammond +2 more

Abstract The paper describes several extensions of fixed-node diffusion in the areas of obtaining analytic derivatives of the energies of molecules, calculating expectation values for quantities other than the energy, and treating excited states. The calculation of energy derivatives is illustrated with a calculation of the first derivative of the energy of the ground state of H2 with respect to the internuclear distance, without using finite differences. With importance sampling the energy is obtained from a weighted average of local energies expressed in analytic form. These may be differentiated to obtain their derivatives with respect to any of several variables, including the internuclear distance, and their weighted averages may be used to determine the desired derivatives. The results for the H2 energy derivatives were in agreement with independent determinations. For calculation of expectation values of the energy, sampling of the (true wavefunction)(trial wavefunction) ‘1/J’l/Jr distribution is sufficient, but other quantities usually require sampling the ‘ljJ2 distribution, which is not so readily available in QMC calculations. Extrapolations of ‘!/Jf and ‘ljJ’l/Jr results, from variational and diffusion calculations, to ‘!jJ2 results are illustrated for various expectation values for the case of ground-state H2.

Jun 18, 20070 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

Quantum Monte Carlo studies of density functional theory

Maziar Nekovee, W. M. C. Foulkes, Richard J. Needs

Abstract It is difficult to imagine that density functional theory (DFT) could be useful in improving quantum Monte Carlo methods, but it is easy to imagine that QMC could be useful in improving density functional methods. Since QMC can produce accurate wavefunctions and since wavefunctions lead to all other properties one should be able, in principle, to use QMC to predict the exchangecorrelation energy functional Exc[n(r)], the key (unknown) quantity required in DFT. In practice, that is not so easy, but this paper shows how QMC can take steps in that direction by providing insights gained from comparisons of results from QMC and DFT. The authors considered the case of the strongly inhomogeneous electron gas using variational Q:tvIC to obtain wavefunctions 1/J>. and energies E[7/J>.] for the ground state of the Hamiltonian iI>- = T +AVee+V >. associated with the coupling constant A in the range 0 to 1. Optimization of a Slater-Jastrow many-body wavefunction was made by minimizing a penalty function given by a weighted sum of the variance in local energies and a sum of squares of electron density differences. The variational parameters in 7/J>. and v>. were repeatedly varied to reach convergence followed by additional calculations to obtain the local values of the exchange-correlation energies e c and the exchange-correlation holes n c from which Exe and nxc could be determined.

Jun 18, 20070 citations

Self-healing diffusion quantum Monte Carlo algorithms: Direct reduction of the fermion sign error in electronic structure calculations

Fernando Agustín Reboredo, Randolph Q. Hood, Paul R. C. Kent

We develop a formalism and present an algorithm for optimization of the trial wave function used in fixed-node diffusion quantum Monte Carlo (DMC) methods. The formalism is based on the DMC mixed estimator of the ground-state probability density. We take advantage of a basic property of the walker configuration distribution generated in a DMC calculation, to (i) project out a multideterminant expansion of the fixed-node ground-state wave function and (ii) to define a cost function that relates the fixed-node ground-state and the noninteracting trial wave functions. We show that (a) locally smoothing out the kink of the fixed-node ground-state wave function at the node generates a new trial wave function with better nodal structure and (b) we argue that the noise in the fixed-node wave function resulting from finite sampling plays a beneficial role, allowing the nodes to adjust toward the ones of the exact many-body ground state in a simulated annealing-like process. Based on these principles, we propose a method to improve both single determinant and multideterminant expansions of the trial wave function. The method can be generalized to other wave-function forms such as pfaffians. We test the method in a model system where benchmark configuration-interaction calculations can be performed and most components of the Hamiltonian are evaluated analytically. Comparing the DMC calculations with the exact solutions, we find that the trial wave function is systematically improved. The overlap of the optimized trial wave function and the exact ground state converges to 100% even starting from wave functions orthogonal to the exact ground state. Similarly, the DMC total energy and density converges to the exact solutions for the model. In the optimization process we find an optimal noninteracting nodal potential of density-functional-like form whose existence was predicted in a previous publication [Phys. Rev. B 77, 245110 (2008)]. Tests of the method are extended to a model system with a conventional Coulomb interaction where we show we can obtain the exact Kohn-Sham effective potential from the DMC data.

arXiv:0808.3417Physical Review BMay 19, 200930 citations

Optimization of quantum Monte Carlo wave functions using analytical energy derivatives

Xi Lin, H Zhang, Andrew M. Rappe

Abstract The optimization of parameters in trial wavefunctions by minimizing the variance in local energy was used for some of the earliest variationala and diffusionb QMC calculations, and it has been used ever since by most practitioners of QMC. Perhaps the best known examples are given in the paper by Schmidt and Moskowitzc for optimizing their nine-parameter Jastrow functions for first-row atoms. An alternative method of optimization, leading to lower variational energies for these atoms, is reported in this paper.

Jun 18, 20071 citation

Improving Variational Monte Carlo Optimization by Avoiding Statistically Difficult Parameters

Scott M. Garner, Eric Neuscamman

Modern quantum Monte Carlo (QMC) methods often capture electron correlation through both explicitly correlating Jastrow factors and small to mid-sized configuration interaction (CI) expansions. Here, we study the additional optimization difficulty created by including increasing numbers of CI parameters. We find evidence that the quality of Variational Mone Carlo (VMC) optimization can be limited by the ability to statistically resolve the CI parameters in the presence of a Jastrow factor. Although using larger statistical samples can mitigate this issue and bring an optimization closer to its true minimum, this approach is computationally intensive. We present evidence that similar gains to optimization quality can be had without increasing the sample size by avoiding CI parameters that are statistically the most difficult to resolve. Our findings suggest that, in addition to the value of using traditional selected configuration interaction (sCI) methods to prepare VMC wave functions, sCI-like methods can play an important role within VMC by improving the effectiveness of stochastic energy minimization.

arXiv:2302.03078Feb 6, 20232 citations