Artificial neural network correction for density-functional tight-binding molecular dynamics simulations
- PDF / 605,698 Bytes
- 7 Pages / 612 x 792 pts (letter) Page_size
- 103 Downloads / 228 Views
Artificial Intelligence Research Letter
Artificial neural network correction for density-functional tight-binding molecular dynamics simulations Junmian Zhu , Department of Chemistry, Grinnell College, Grinnell, IA, USA Van Quan Vuong , Bredesen Center for Interdisciplinary Research and Graduate Education, University of Tennessee, Knoxville, TN, USA Bobby G. Sumpter , Center for Nanophase Materials Sciences and Computational Sciences & Engineering Division, Oak Ridge National Laboratory, Oak Ridge, TN, USA Stephan Irle , Bredesen Center for Interdisciplinary Research and Graduate Education, University of Tennessee, Knoxville, TN, USA; Center for Nanophase Materials Sciences and Computational Sciences & Engineering Division, Oak Ridge National Laboratory, Oak Ridge, TN, USA; Chemical Sciences Division, Oak Ridge National Laboratory, Oak Ridge, TN, USA Address all correspondence to Stephan Irle at [email protected] (Received 21 February 2019; accepted 5 June 2019)
Abstract The authors developed a Behler–Parrinello-type neural network (NN) to improve the density-functional tight-binding (DFTB) energy and force prediction. The Δ-machine learning approach was adopted and the NN was designed to predict the energy differences between the density functional theory (DFT) quantum chemical potential and DFTB for a given molecular structure. Most notably, the DFTB-NN method is capable of improving the energetics of intramolecular hydrogen bonds and torsional potentials without modifying the framework of DFTB itself. This improvement enables considerably larger simulations of complex chemical systems that currently could not easily been accomplished using DFT or higher level ab initio quantum chemistry methods alone.
Introduction The density-functional tight-binding (DFTB) method[1–4] provides an efficient semi-empirical computational scheme to approximate density function theory (DFT) calculations. Using a minimal atomic orbital basis set for valence shell electrons and employing a two-center approximation for Hamiltonian matrix elements, the DFTB method efficiently reduces the computational cost of traditional DFT by 2–3 orders of magnitude.[5,6] This tremendous advance in computational speed can be exploited in long timescale molecular dynamics (MD) simulations of large-scale reactive processes.[6,7] However, due to the approximations used in DFTB, in particular the use of the minimal atomic orbital basis set,[2] the representation of the electron density by a simplified point charge model,[1,4] and the neglect of three-center terms, traditional DFTB computed energies and gradients often deviate from DFT values.[6] For instance, the monopolar charge approximation and the use of a minimal basis set make it impossible for DFTB to describe out-of-plane polarization for planar molecular systems, to accurately capture atomic hybridization effects, etc. These deficiencies often result in problems for DFTB in the prediction of potential energy surfaces of highly polar systems.[6–8] Conventional DFTB parameterization methods often place em
Data Loading...
