Combined shared and distributed memory ab-initio computations of molecular-hydrogen systems in the correlated state: process pool solution and two-level parallelism

TL;DR

This paper presents an efficient hybrid shared and distributed memory computational scheme for ab-initio calculations of correlated molecular hydrogen systems, achieving a 300-fold speedup through hierarchical parallelism.

Contribution

It introduces a novel process-pool solution combined with two-level parallelism for ab-initio computations of correlated systems.

Findings

Achieved a 300-fold increase in calculation speed.

Successfully applied hierarchical parallelism using OpenMP and MPI.

Demonstrated effective ground state energy optimization for hydrogen chains.

Abstract

An efficient computational scheme devised for investigations of ground state properties of the electronically correlated systems is presented. As an example, $(H_{2})_{n}$ chain is considered with the long-range electron-electron interactions taken into account. The implemented procedure covers: (i) single-particle Wannier wave-function basis construction in the correlated state, (ii) microscopic parameters calculation, and (iii) ground state energy optimization. The optimization loop is based on highly effective process-pool solution - specific root-workers approach. The hierarchical, two-level parallelism was applied: both shared (by use of Open Multi-Processing) and distributed (by use of Message Passing Interface) memory models were utilized. We discuss in detail the feature that such approach results in a substantial increase of the calculation speed reaching factor of $300$ for the fully parallelized solution.