Ex-NNQMD: Extreme-Scale Neural Network Quantum Molecular Dynamics

Deep learning is revolutionizing countless scientific and engineering fields. In particular, SC20 Gordon Bell award represented a breakthrough in molecular simulation, i.e., 100-million-atom simulation with quantum-mechanical accuracy on the Summit supercomputer at ORNL, using deep potential molecular dynamics (MD). Moving forward, while these simulations were performed only in gentle equilibrium conditions, far-from-equilibrium MD simulation involving light-induced electronic excited states finds numerous scientific and engineering applications. However, it remains a challenge to perform such far-from-equilibrium simulations at larger spatiotemporal scales, where growing number of unphysical predictions of interatomic force prohibits simulations involving larger numbers of atoms for longer times. In this paper, we propose a physically-based inductive bias, maximally-preserved Maxwell-Boltzmann (MPMB), to overcome this fidelity-scaling problem. Along with hybrid divide-and-conquer parallelization and single-node level optimization using multithreading and data parallel SIMD, the resulting Ex-NNQMD (extreme-scale neural network quantum molecular dynamics) algorithm has achieved unprecedented scales of far-from-equilibrium simulations: 1) 5.1-billion atom system with a parallel efficiency of 0.94, and 2) a sustained performance of 6.4 nanoseconds/day for 10-million atom system both on 262,144 cores of the Theta supercomputer at Argonne Leadership Computing Facility. Extended fidelity scaling and efficient parallelization have allowed us for the first time to study light-induced ferroelectric switching under extreme electronic excitation at experimentally relevant spatiotemporal scales with accuracy.

[1]  Mordechai Kornbluth,et al.  A fast neural network approach for direct covariant forces prediction in complex multi-element extended systems , 2019, Nature Machine Intelligence.

[2]  E Weinan,et al.  End-to-end Symmetry Preserving Inter-atomic Potential Energy Model for Finite and Extended Systems , 2018, NeurIPS.

[3]  Yoshua Bengio,et al.  Inductive Biases for Deep Learning of Higher-Level Cognition , 2020, ArXiv.

[4]  S. Nosé.  A molecular dynamics method for simulations in the canonical ensemble , 1984 .

[5]  D. Mihailovic,et al.  Ultrafast Switching to a Stable Hidden Quantum State in an Electronic Crystal , 2014, Science.

[6]  Daniel W. Davies,et al.  Machine learning for molecular and materials science , 2018, Nature.

[7]  E Weinan,et al.  Pushing the Limit of Molecular Dynamics with Ab Initio Accuracy to 100 Million Atoms with Machine Learning , 2020, SC20: International Conference for High Performance Computing, Networking, Storage and Analysis.

[8]  Supa Hannongbua,et al.  Shift/collapse on neighbor list (SC-NBL): Fast evaluation of dynamic many-body potentials in molecular dynamics simulations , 2019, Comput. Phys. Commun..

[9]  D. Haar,et al.  Statistical Physics , 1971, Nature.

[10]  P. Ajayan,et al.  Optical Control of Non-Equilibrium Phonon Dynamics. , 2019, Nano letters.

[11]  Ying Li,et al.  Scalable Reactive Molecular Dynamics Simulations for Computational Synthesis , 2019, Computing in Science & Engineering.

[12]  Christopher T. Nelson,et al.  Spatially resolved steady-state negative capacitance , 2019, Nature.

[13]  Nongnuch Artrith,et al.  An implementation of artificial neural-network potentials for atomistic materials simulations: Performance for TiO2 , 2016 .

[14]  R. Batra,et al.  Physically informed artificial neural networks for atomistic modeling of materials , 2018, Nature Communications.

[15]  Z. Ye.  Handbook of advanced dielectric, piezoelectric and ferroelectric materials , 2008 .

[16]  J. Behler.  Atom-centered symmetry functions for constructing high-dimensional neural network potentials. , 2011, The Journal of chemical physics.

[17]  Razvan Pascanu,et al.  Relational inductive biases, deep learning, and graph networks , 2018, ArXiv.

[18]  Rajiv K. Kalia,et al.  A scalable parallel algorithm for dynamic range-limited n-tuple computation in many-body molecular dynamics simulation , 2013, 2013 SC - International Conference for High Performance Computing, Networking, Storage and Analysis (SC).

[19]  Rajiv K. Kalia,et al.  RXMD: A scalable reactive molecular dynamics simulator for optimized time-to-solution , 2020, SoftwareX.

[20]  A. Nakano,et al.  Guidelines for creating artificial neural network empirical interatomic potential from first-principles molecular dynamics data under specific conditions and its application to α-Ag2Se. , 2019, The Journal of chemical physics.