1:45 PM - MT01.04.02
AGNI—A Machine Learning Platform for the Rapid Prediction of Atomistic Properties
James Chapman1,Rohit Batra1,Huan Tran1,Chiho Kim1,Anand Chandrasekaran1,Deepak Kamal1,Christopher Kuenneth1,Rampi Ramprasad1
Georgia Institute of Technology1
Show Abstract
Propelled partly by the Materials Genome Initiative, and partly by the algorithmic developments and the resounding successes of data-driven efforts in other domains, machine learning (ML) strategies are beginning to take shape within several subfields of materials science[1-4]. One area of enormous importance is atomic-level materials phenomena, which spans numerous fields from electronic structure theory to mechanical failure, and has been dominated by either quantum mechanics (QM) based methods [4-7]—which are time-intensive, but accurate and versatile—or semi-empirical/ classical methods[8,9]—which are fast but are significantly limited in veracity, versatility and transferability. ML methods have the potential to bridge the chasm between the two extremes and can combine the best of both worlds. We have created a platform for the rapid prediction of properties such as potential energy, atomic forces, stresses, charge density, and the electronic density of states. Our ML-models are trained on accurate QM reference data, and can reproduce the QM results with the same level of accuracy but several orders of magnitude faster [10-15]. The ML models can also be progressively improved in quality by periodically (or on-demand) exposing them to fresh QM data in regions of poor performance, a feature that currently is either impossible or daunting with modern empirical/classical methods. Our ML models may thus be used to perform large-scale and/or long-time simulations of important materials phenomena previously beyond the reaches of QM based methods. Here, we demonstrate the power and versatility of this new platform in correctly capturing electronic, thermodynamic, mechanical, and diffusive properties for a variety of systems, with the hope of ushering in a new era of atomic-level understanding of materials.
References:
[1] Chan, H.; Cherukara, M.; Narayanan, B.; Loeffler, T.; Benmore, C.; Gray, S.; Sankaranarayanan, S. Nat. Comm. 2019, 10, A397
[2] Behler, J.; Parrinello, M. Phys. Rev. Lett. 2007, 98, 146401
[3] Botu, V.; Ramprasad, R. Int. J. Quant. Chem. 2015, 115, 1074–1083
[4] Kolb, B.; Lentz, L.; Koplak, A. Scientific Reports, 2017, 7, A1192
[5] Jones, R. O. Rev. Mod. Phys. 2015, 87, 897-923
[6] Hohenburg, P.; Kohn, W. Phys. Rev. 1964, 136, B864-871
[7] Kohn, L., W; Sham, K. Phys. Rev. 1965 , 140, A1133-A1138
[8] Daw, M. S.; Baskes, M. I. Phys. Rev. B 1984 , 29 , 6443–6453
[9] Hrennikoff, A. IABSE, 1968
[10] Botu, V.;Batra, R.;Chapman, J.;Ramprasad, R. Jour. Phys. Chem. C, 2017, 121, 511–522
[11] Chandrasekaran, A.; Kamal, D.; Batra, R.; Kim, C.; Chen, L.; Ramprasad, R. npj Computational Materials, 2019, 5, 22
[12] Batra, R.; Chandrasekaran, A.; Chapman, J.; Kim, C.; Chen, L.; Huan, T.; Ramprasad, R. Jour. Phys. Chem. C, 2019
[13] Huan, T.; Batra, R.; Chapman, J.; Kim, C.; Chandrasekaran, A.; Ramprasad, R. (manuscript under review)
[14] Chapman, J.; Batra, R.; Ramprasad, R. (manuscript in preparation)