Atomic orbital-based SOS-MP2 with tensor hypercontraction. II. Local tensor hypercontraction.
journal of chemical physics
2017; 146 (3): 034104-?
In the first paper of the series [Paper I, C. Song and T. J. Martinez, J. Chem. Phys. 144, 174111 (2016)], we showed how tensor-hypercontracted (THC) SOS-MP2 could be accelerated by exploiting sparsity in the atomic orbitals and using graphical processing units (GPUs). This reduced the formal scaling of the SOS-MP2 energy calculation to cubic with respect to system size. The computational bottleneck then becomes the THC metric matrix inversion, which scales cubically with a large prefactor. In this work, the local THC approximation is proposed to reduce the computational cost of inverting the THC metric matrix to linear scaling with respect to molecular size. By doing so, we have removed the primary bottleneck to THC-SOS-MP2 calculations on large molecules with O(1000) atoms. The errors introduced by the local THC approximation are less than 0.6 kcal/mol for molecules with up to 200 atoms and 3300 basis functions. Together with the graphical processing unit techniques and locality-exploiting approaches introduced in previous work, the scaled opposite spin MP2 (SOS-MP2) calculations exhibit O(N(2.5)) scaling in practice up to 10 000 basis functions. The new algorithms make it feasible to carry out SOS-MP2 calculations on small proteins like ubiquitin (1231 atoms/10 294 atomic basis functions) on a single node in less than a day.
View details for DOI 10.1063/1.4973840
View details for PubMedID 28109237
Geometry optimization made simple with translation and rotation coordinates
JOURNAL OF CHEMICAL PHYSICS
2016; 144 (21)
The effective description of molecular geometry is important for theoretical studies of intermolecular interactions. Here we introduce a new translation-rotation-internal coordinate (TRIC) system which explicitly includes the collective translations and rotations of molecules, or parts of molecules such as monomers or ligands, as degrees of freedom. The translations are described as the centroid position and the orientations are represented with the exponential map parameterization of quaternions. When TRIC is incorporated into geometry optimization calculations, the performance is consistently superior to existing coordinate systems for a diverse set of systems including water clusters, organic semiconductor donor-acceptor complexes, and small proteins, all of which are characterized by nontrivial intermolecular interactions. The method also introduces a new way to scan the molecular orientations while allowing orthogonal degrees of freedom to relax. Our findings indicate that an explicit description of molecular translation and rotation is a natural way to traverse the many-dimensional potential energy surface.
View details for DOI 10.1063/1.4952956
View details for Web of Science ID 000378923500010
View details for PubMedID 27276946
Atomic orbital-based SOS-MP2 with tensor hypercontraction. I. GPU-based tensor construction and exploiting sparsity
JOURNAL OF CHEMICAL PHYSICS
2016; 144 (17)
We present a tensor hypercontracted (THC) scaled opposite spin second order Møller-Plesset perturbation theory (SOS-MP2) method. By using THC, we reduce the formal scaling of SOS-MP2 with respect to molecular size from quartic to cubic. We achieve further efficiency by exploiting sparsity in the atomic orbitals and using graphical processing units (GPUs) to accelerate integral construction and matrix multiplication. The practical scaling of GPU-accelerated atomic orbital-based THC-SOS-MP2 calculations is found to be N(2.6) for reference data sets of water clusters and alanine polypeptides containing up to 1600 basis functions. The errors in correlation energy with respect to density-fitting-SOS-MP2 are less than 0.5 kcal/mol for all systems tested (up to 162 atoms).
View details for DOI 10.1063/1.4948438
View details for Web of Science ID 000377711300015
View details for PubMedID 27155629
- Automated Code Engine for Graphical Processing Units: Application to the Effective Core Potential Integrals and Gradients JOURNAL OF CHEMICAL THEORY AND COMPUTATION 2016; 12 (1): 92-106
- Efficient implementation of effective core potential integrals and gradients on graphical processing units JOURNAL OF CHEMICAL PHYSICS 2015; 143 (1)
- Analytic first derivatives of floating occupation molecular orbital-complete active space configuration interaction on graphical processing units. journal of chemical physics 2015; 143 (1): 014111-?