ipie: A Python-Based Auxiliary-Field Quantum Monte Carlo Program with Flexibility and Efficiency on CPUs and GPUs

Abstract

We report the development of a python-based auxiliary-field quantum Monte Carlo (AFQMC) program, , with preliminary timing benchmarks and new AFQMC results on the isomerization of [Cu2O2]2+. We demonstrate how implementations for both central and graphical processing units (CPUs and GPUs) are achieved in . We show an interface of with as well as a straightforward template for adding new estimators to . Our timing benchmarks against other C++ codes, and , suggest that is faster or similarly performing for all chemical systems considered on both CPUs and GPUs. Our results on [Cu2O2]2+ using selected configuration interaction trials show that it is possible to converge the ph-AFQMC isomerization energy between bis­(μ-oxo) and μ-η2:η2 peroxo configurations to the exact known results for small basis sets with 105–106 determinants. We also report the isomerization energy with a quadruple-zeta basis set with an estimated error less than a kcal/mol, which involved 52 electrons and 290 orbitals with 106 determinants in the trial wave function. These results highlight the utility of ph-AFQMC and for systems with modest strong correlation and large-scale dynamic correlation.

Twenty Years of Auxiliary-Field Quantum Monte Carlo in Quantum Chemistry: An Overview and Assessment on Main Group Chemistry and Bond-Breaking

“…We hope that the numerical data presented in this work, as well as the existence of currently available open-source packages − will encourage many more developers and users to explore uncharted territory and contribute to method development within the ph-AFQMC framework. While the data provided here constitutes the most extensive data set for ph-AFQMC to date, it is important to continue producing data and comparisons for a wide class of systems.…”

“…In addition, its final energy is also statistical in nature, introducing additional barrier for users to use the method. It is, therefore, important to have and further develop open-source frameworks , and to provide straightforward user interfaces. Such efforts will enable the expansion of our understanding of the relative benefits and weaknesses of ph-AFQMC, and we hope this goal is taken up by the electronic structure community with enthusiasm in the coming years.…”

“…For the A24 set, we used aug-cc-pVTZ with counterpoise corrections, although the reference data (i.e., CCSD with triples and perturbative quadruples) were obtained in the basis set limit. For H 4 and N 2 dissociation data, we used cc-pVQZ and cc-pVTZ bases, respectively, without the frozen core approximation. Molecular integrals for ph-AFQMC were generated by PySCF, all deterministic single-reference quantum chemistry calculations were performed with Q-Chem, and AFQMC calculations were performed with ipie , and QMCPACK . Cholesky factorization and integral transformations necessary for AFQMC calculations were performed by a script in ipie , with a Cholesky threshold of 10 –5 .…”

“…Since its invention by Zhang and Krakauer in 2003, ph-AFQMC has gained popularity in the quantum chemistry community due to its relative computational affordability and its accuracy, as demonstrated in recent calculations including those performed on small but challenging molecular systems, ,− simple transition metal complexes, − and solids. ,− ph-AFQMC is highly flexible in that the accuracy of the approach may be systematically improved by the use of increasingly sophisticated trial wave functions. In addition, ph-AFQMC has become more accessible to users via the development of open-source codes. − …”

Data-Efficient Machine Learning Potentials from Transfer Learning of Periodic Correlated Electronic Structure Methods: Liquid Water at AFQMC, CCSD, and CCSD(T) Accuracy

“…Periodic CCSD and CCSD(T) calculations were performed using PySCF, 44,45 where electron-repulsion integrals are handled using the rangeseparated density fitting method, 46,47 and AFQMC calculations were performed using QMCPACK 48,49 and ipie. 50 These calculations all began with a periodic spin-restricted HF calculation also perfomed using PySCF. We provide brief details here, and further information can be found in sections IB and IC in the SI.…”