[Paper Review] Computational methods in Coupled Electron-Ion Monte Carlo

Carlo Pierleoni, David M. Ceperley|ArXiv.org|Jan 4, 2005

Advanced Chemical Physics Studies3 citations

TL;DR

This paper introduces an improved sampling algorithm for Reptation Quantum Monte Carlo (RQMC) to enhance efficiency in Coupled Electron-Ion Monte Carlo (CEIMC) simulations. The method enables accurate sampling of proton degrees of freedom using variational electronic energy estimates, reducing noise and enabling reliable computation of electronic forces for ab initio molecular dynamics in metallic hydrogen, with energy differences insensitive to projection time.

ABSTRACT

In the last few years we have been developing a Monte Carlo simulation method to cope with systems of many electrons and ions in the Born-Oppenheimer (BO) approximation, the Coupled Electron-Ion Monte Carlo Method (CEIMC). Electronic properties in CEIMC are computed by Quantum Monte Carlo (QMC) rather than by Density Functional Theory (DFT) based techniques. CEIMC can, in principle, overcome some of the limitations of the present DFT based ab initio dynamical methods. Application of the new method to high pressure metallic hydrogen has recently appeared. In this paper we present a new sampling algorithm that we have developed in the framework of the Reptation Quantum Monte Carlo (RQMC) method chosen to sample the electronic degrees of freedom, thereby improving its efficiency. Moreover, we show here that, at least for the case of metallic hydrogen, variational estimates of the electronic energies lead to an accurate sampling of the proton degrees of freedom.

Motivation & Objective

• To develop a more efficient sampling algorithm for Reptation Quantum Monte Carlo (RQMC) in the context of Coupled Electron-Ion Monte Carlo (CEIMC).
• To enable accurate and efficient sampling of ionic (proton) degrees of freedom using variational estimates of electronic energy.
• To reduce noise and improve convergence in CEIMC by minimizing sensitivity to projection time in electronic energy differences.
• To validate that variational electronic energy estimates are sufficient for sampling proton configurations in metallic hydrogen.

Proposed method

• A new RQMC sampling scheme is proposed that improves scaling with imaginary time projection, requiring minimal code changes.
• The algorithm avoids pathological behavior seen in standard RQMC by modifying the resampling procedure of the path direction.
• Electronic degrees of freedom are sampled using RQMC with a Trotter breakup of the imaginary-time propagator into P time slices of size τₑ.
• Proton configurations are sampled using the energy difference between configurations, computed via variational Monte Carlo (VMC), to guide the acceptance/rejection in a Metropolis step.
• The method leverages the fact that energy differences between proton configurations are insensitive to the projection time βₑ, enabling stable sampling.
• The approach is implemented in CEIMC to compute forces and equation of state for metallic hydrogen at high pressure.

Experimental results

Research questions

• RQ1 Can a new RQMC sampling algorithm significantly improve the efficiency of electronic structure sampling in CEIMC simulations?
• RQ2 Is the energy difference between proton configurations robust to changes in the projection time βₑ, enabling stable sampling?
• RQ3 Can variational electronic energy estimates accurately represent the electronic contribution to the potential energy surface for proton dynamics?
• RQ4 How does the correlation time and variance of the energy difference scale with projection time in the new sampling scheme?
• RQ5 To what extent can VMC-based sampling of proton configurations yield results consistent with full RQMC in the electronic energy evaluation?

Key findings

• The new RQMC sampling algorithm shows favorable scaling with projection time, with correlation time growing as βₑ^0.15, indicating moderate noise increase with longer projection times.
• The energy difference ΔE/kBT between proton configurations is nearly independent of βₑ and τₑ, suggesting that variational energy estimates are sufficient for accurate sampling.
• The projected electronic energy is 1809 K/atom lower than the variational estimate, indicating a significant correction that must be included for accuracy.
• For metallic hydrogen at rs=1 and T=1000 K, RQMC yields an energy of -0.41114(8)H/atom, compared to -0.4087(1)H/atom from VMC, with pressure estimates in agreement within error bars.
• The average correlation time increases from 7.1 (VMC) to 16.5 (RQMC), and variance from 2.3 to 26.5, indicating that RQMC is more computationally demanding but necessary for accuracy.
• The method enables reliable CEIMC simulations with VMC-based sampling of proton configurations, validated at a single thermodynamic point, suggesting broader applicability if trial wave functions are accurate.