Breaking the Million-Electron and 1 EFLOP/s Barriers: Biomolecular-Scale Ab Initio Molecular Dynamics Using MP2 Potentials

TL;DR

This study presents an innovative approach that enables biomolecular-scale ab initio molecular dynamics (AIMD) simulations at WF theory level by combining molecular fragmentation with MP2 perturbation theory and the introduction of asynchronous time steps.

Abstract

The accurate simulation of complex biochemical phenomena has historically been hampered by the computational requirements of high-fidelity molecular-modeling techniques. Quantum mechanical methods, such as ab initio wave-function (WF) theory, deliver the desired accuracy, but have impractical scaling for modelling biosystems with thousands of atoms. Combining molecular fragmentation with MP2 perturbation theory, this study presents an innovative approach that enables biomolecular-scale ab initio molecular dynamics (AIMD) simulations at WF theory level. Leveraging the resolution-of-the-identity approximation for Hartree-Fock and MP2 gradients, our approach eliminates computationally intensive four-center integrals and their gradients, while achieving near-peak performance on modern GPU architectures. The introduction of asynchronous time steps minimizes time step latency, overlapping computational phases and effectively mitigating load imbalances. Utilizing up to 9,400 nodes of Frontier and achieving 59% (1006.7 PFLOP/s) of its double-precision floating-point peak, our method enables us to break the million-electron and 1 EFLOP/s barriers for AIMD simulations with quantum accuracy.

Figures (8)

• Figure 1: Maximum system sizes of static energy evaluations and AIMD simulations achieved with varying levels of theory and corresponding average isomerization energy errors. Isomerization energy errors are obtained from Ref. Grimme2007. Exact sizes and corresponding references are listed in Table \ref{['tab:record_table']}.
• Figure 2: Overarching algorithmic scheme adopted for the AIMD/RI-HF+RI-MP2 calculations.
• Figure 3: Execution time of RI-MP2 gradients with and without the RI-HF approximation for varying length glycine chains on a single 40 GB NVIDIA A100 GPU. The cc-pVDZ/cc-pVDZ-RIFIT primary and auxiliary basis set were used. Data points are labelled with the speedup of the RI vs non-RI HF version.
• Figure 4: Schematic of the asynchronous time step workflow showing the major algorithmic components and data flow.
• Figure 5: Energy contributions of dimers and trimers for the initial geometry of the 6PQ5 system. Cutoff distances used in the calculation are also marked.