Quantum supercharger library: Hyper-parallelism of the Hartree-Fock method

Abstract

We present here a set of algorithms that completely rewrites the Hartree-Fock (HF) computations common to many legacy electronic structure packages (such as GAMESS-US, GAMESS-UK, and NWChem) into a massively parallel compute scheme that takes advantage of hardware accelerators such as Graphical Processing Units (GPUs). The HF compute algorithm is core to a library of routines that we name the Quantum Supercharger Library (QSL). We briefly evaluate the QSL's performance and report that it accelerates a HF 6-31G Self-Consistent Field (SCF) computation by up to 20 times for medium sized molecules (such as a buckyball) when compared with mature Central Processing Unit algorithms available in the legacy codes in regular use by researchers. It achieves this acceleration by massive parallelization of the one- and two-electron integrals and optimization of the SCF and Direct Inversion in the Iterative Subspace routines through the use of GPU linear algebra libraries. © 2015 Wiley Periodicals, Inc.

Quantum supercharger library: Hyper-parallel integral derivatives algorithms forab initioQM/MM dynamics

“…Moreover, the auxiliary functions are related by the same recursion relationship as shown previously. [2] An advantage of the Hermite form of the Gaussian functions is that the evaluation of the derivatives is similar to the evaluation of higher angular momentum integrals.…”

“…[2] Here, we show how to evaluate the integral derivatives using the MD approach. In the MD scheme, the product of two Gaussians can be expanded in terms of Hermite Gaussian functions as shown in equation 32 of the accompanying paper.…”

“…(4) the first derivative of the twoelectron integral can be written in terms of the auxiliary functions, expansion coefficients and other variables defined before, [2] as…”

“…[1] In the companion paper to this work, we described the acceleration of the one-and two-electron integrals and the self-consistent field (SCF) routines that are central to computing the energy of a molecular system. [2] Key molecular processes that underpin, for example, chemical reactivity can be understood through the net forces and appropriately partitioned components of the net forces on the nuclei in a molecular system. [3] Optimizing small molecular systems in vacuum at 0 K with a view to understanding the effect of geometry and conformation on enthalpically based macroscopic properties forms the kernel of research in many laboratories.…”

Double‐buffered, heterogeneous CPU + GPU integral digestion algorithm for single‐excitation calculations involving a large number of excited states

“…The specific bottleneck depends upon the method in question, but these two potential bottlenecks have each been targeted for GPU acceleration. ERI generation was originally tackled on GPUs by Yasuda and by Martínez and coworkers, then later by others . In early work, there was difficulty extending GPU‐based ERI algorithms to basis sets with high angular momentum, as the intermediates required for computing high angular momentum shells were too large to store in GPU cache and registers, while the recursive nature of the integrals generation algorithm became rather complex for the “same instruction, multiple data” (SIMD)‐style operations where GPUs excel.…”