Analytical gradients for tensor hypercontracted MP2 and SOSMP2 on graphical processing units
Analytic energy gradients for tensor hypercontraction (THC) are derived and implemented for secondorder MøllerPlesset perturbation theory (MP2), with and without the scaledoppositespin (SOS)MP2 approximation. By exploiting the THC factorization, the formal scaling of MP2 and SOSMP2 gradient calculations with respect to system size is reduced to quartic and cubic, respectively. An efficient implementation has been developed that utilizes both graphics processing units and sparse tensor techniques exploiting spatial sparsity of the atomic orbitals. THCMP2 has been applied to both geometry optimization and ab initio molecular dynamics (AIMD) simulations. Furthermore, the resulting energy conservation in microcanonical AIMD demonstrates that the implementation provides accurate nuclear gradients with respect to the THCMP2 potential energy surfaces.
 Authors:

^{[1]};
^{[1]}
 Stanford Univ., Stanford, CA (United States); SLAC National Accelerator Lab., Menlo Park, CA (United States)
 Publication Date:
 Grant/Contract Number:
 AC0276SF00515; ACI1429830; CHE1565249; SGF
 Type:
 Accepted Manuscript
 Journal Name:
 Journal of Chemical Physics
 Additional Journal Information:
 Journal Volume: 147; Journal Issue: 16; Journal ID: ISSN 00219606
 Publisher:
 American Institute of Physics (AIP)
 Research Org:
 SLAC National Accelerator Lab., Menlo Park, CA (United States)
 Sponsoring Org:
 USDOE
 Country of Publication:
 United States
 Language:
 English
 Subject:
 72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS
 OSTI Identifier:
 1410666
Song, Chenchen, and Martinez, Todd J. Analytical gradients for tensor hypercontracted MP2 and SOSMP2 on graphical processing units. United States: N. p.,
Web. doi:10.1063/1.4997997.
Song, Chenchen, & Martinez, Todd J. Analytical gradients for tensor hypercontracted MP2 and SOSMP2 on graphical processing units. United States. doi:10.1063/1.4997997.
Song, Chenchen, and Martinez, Todd J. 2017.
"Analytical gradients for tensor hypercontracted MP2 and SOSMP2 on graphical processing units". United States.
doi:10.1063/1.4997997. https://www.osti.gov/servlets/purl/1410666.
@article{osti_1410666,
title = {Analytical gradients for tensor hypercontracted MP2 and SOSMP2 on graphical processing units},
author = {Song, Chenchen and Martinez, Todd J.},
abstractNote = {Analytic energy gradients for tensor hypercontraction (THC) are derived and implemented for secondorder MøllerPlesset perturbation theory (MP2), with and without the scaledoppositespin (SOS)MP2 approximation. By exploiting the THC factorization, the formal scaling of MP2 and SOSMP2 gradient calculations with respect to system size is reduced to quartic and cubic, respectively. An efficient implementation has been developed that utilizes both graphics processing units and sparse tensor techniques exploiting spatial sparsity of the atomic orbitals. THCMP2 has been applied to both geometry optimization and ab initio molecular dynamics (AIMD) simulations. Furthermore, the resulting energy conservation in microcanonical AIMD demonstrates that the implementation provides accurate nuclear gradients with respect to the THCMP2 potential energy surfaces.},
doi = {10.1063/1.4997997},
journal = {Journal of Chemical Physics},
number = 16,
volume = 147,
place = {United States},
year = {2017},
month = {8}
}