Evaluating two-electron-repulsion integrals over arbitrary orbitals using zero variance Monte Carlo: Application to full configuration interaction calculations with Slater-type orbitals

Abstract

A Monte Carlo method for evaluating multi-center two-electron-repulsion integrals over any type of orbitals (Slater, Sturmian, finite-range, numerical, etc.) is presented. The approach is based on a simple and universal (orbital-independent) gaussian sampling of the two-electron configuration space and on the use of efficient zero-variance Monte Carlo estimators. Quite remarkably, it is shown that the high level of accuracy required on two-electron integrals to make Hartree-Fock (HF) and configuration interaction (CI) calculations feasible can be achieved. A first zero-variance estimator is built by introducing a gaussian approximation of the orbitals and by evaluating the two-electron integrals using a correlated sampling scheme for the difference between exact and approximate orbitals. A second one is based on the introduction of a general coordinate transformation. The price to pay for this simple and general Monte Carlo scheme is the high computational cost required. However, we argue that the great simplicity of the algorithm, its embarrassingly parallel nature, its ideal adaptation to modern computational platforms and, most importantly, the possibility of using more compact and physically meaningful basis sets make nevertheless the method attractive. HF and near full CI (FCI) calculations using Slater-type orbitals (STO) are reported for Be, CH 4 and [H 2 N(CH)NH 2 ] + (a simple model of cyanine). To the best of our knowledge, our largest FCI calculation involving 18 active electrons distributed among 90 orbitals for the cyanine molecule, is the most extensive molecular calculation performed so far using pure STO orbitals (no gaussian approximation, even for the challenging four-center two-electron integrals).

Optimization of Large Determinant Expansions in Quantum Monte Carlo

“…In order to reduce the fluctuations on the dressing elements, we will present two improved estimators based on the so-called control variates technique. , The general idea is to combine the estimator of eq with a correlated function in the 3 N -dimensional space of which the integral can be evaluated in a deterministic way.…”

Monte Carlo MP2-F12 for Noncovalent Interactions: The C60 Dimer

“…In contrast to QMC, these methods have no bias, ECP, or Fermion sign problems (at second order at least), and can directly compute energy differences (such as correlation correction, ,, F12 correction, ,, electron binding energies, , or quasiparticle energy bands). Therefore, this method is an example of the burgeoning class of stochastic ab initio electronic structure methods. − …”

Optimization of large determinant expansions in quantum Monte Carlo

“…In order to reduce the fluctuations on the dressing elements, we will present two improved estimators based on the socalled control variates technique 65,66 . The general idea is to combine the estimator of Eq (26) with a correlated function in the 3N-dimensional space of which the integral can be evaluated in a deterministic way.…”