Training a Gaussian process regression model of formamide for use in molecular simulations Matthew Brown , Jonathan M Skelton, Paul L A Popelier Department of Chemistry, The University of Manchester, Oxford Road, Manchester, M13 9PL, Britain Traditional force fields offer a way of simulating larger systems for longer timescales than methods such as ab initio molecular dynamics can feasibly provide, making them a necessary tool in many projects. However, the accuracy of these simulations is often limited by point charge electrostatics and approximate bonding potentials. FFLUX [1-3] is a novel force field that utilises Gaussian process regression (GPR) machine learning models trained on atomic energies and multipole moments up to the hexadecapole obtained from the interacting quantum atoms (IQA) scheme [4]. IQA is a quantum chemical topology method that unambiguously partitions a wavefunction energy into atomic contributions. These models can predict atomic energies and atomic multipole moments on-the-fly in molecular dynamics simulations of fully flexible molecules and replace the need for parameterised bonded terms. The intramolecular potential energy surface in FFLUX therefore more closely resembles that of the quantum-mechanical method used for training. This talk will present recent results from FFLUX simulations on formamide, which we have chosen as an example to validate the FFLUX methodology due to its size and rigidity making it a relatively easy candidate for accurate modelling. A monomeric formamide GPR model has been constructed at the B3LYP/aug-cc-pVTZ level of theory and has been combined with Lennard-Jones parameters in molecular dynamics simulations of formamide dimers, giving results consistent with the B3LYP-D3/aug-cc-pVTZ. In the near future, these preliminary models will be extended to cover more complex simulations of crystal structures and liquids. References 1. Popelier, P.L.A., Molecular Simulation by Knowledgeable Quantum Atoms. Phys.Scr., 2016. 91 : p. 033007. 2. Hughes, Z.E., et al., Description of Potential Energy Surfaces of Molecules Using FFLUX Machine Learning Models. J.Chem.Theor.Comp., 2019. 15 : p. 116-126. 3. Burn, M.J. and P.L.A. Popelier, Creating Gaussian Process Regression Models for Molecular Simulations Using Adaptive Sampling J.Chem.Phys., 2020. 153 : p. 054111. 4. Blanco, M.A., A. Martín Pendás, and E. Francisco, Interacting Quantum Atoms: A Correlated Energy Decomposition Scheme Based on the Quantum Theory of Atoms in Molecules. J.Chem.Theor. Comp., 2005. 1 : p. 1096-1109.
P07DD
Made with FlippingBook Learn more on our blog