The accuracy of ab initio calculations without ab initio calculations for charged systems: Kriging predictions of atomistic properties for ions in aqueous solutions

Abstract

Using the machine learning method kriging, we predict the energies of atoms in ion-water clusters, consisting of either Cl or Na surrounded by a number of water molecules (i.e., without NaCl interaction). These atomic energies are calculated following the topological energy partitioning method called Interacting Quantum Atoms (IQAs). Kriging predicts atomic properties (in this case IQA energies) by a model that has been trained over a small set of geometries with known property values. The results presented here are part of the development of an advanced type of force field, called FFLUX, which offers quantum mechanical information to molecular dynamics simulations without the limiting computational cost of ab initio calculations. The results reported for the prediction of the IQA components of the energy in the test set exhibit an accuracy of a few kJ/mol, corresponding to an average error of less than 5%, even when a large cluster of water molecules surrounding an ion is considered. Ions represent an important chemical system and this work shows that they can be correctly taken into account in the framework of the FFLUX force field.

Contributions of IQA electron correlation in understanding the chemical bond and non-covalent interactions

“…Three different types of ML techniques, that is, Random Forest (RF), Support Vector Regression (SVR), and Kriging [39,56,[66][67][68][69], were applied to predict correlation energies of three systems: the water monomer, the H 2 -He complex and the water dimer. In the first system, intramolecular-interatomic correlation energies were investigated, while in the last two, the intermolecular correlation energies are in the spotlight, since they are correlated with intermolecular dispersion forces.…”

Interacting Quantum Atoms—A Review

“…Using this methodology, they have been able to predict intra- [ 143 ] and interatomic energies [ 144 , 145 ], as well as correlation energy at the MP2 approximation for a previously unknown molecular geometries [ 146 ] Using. the models produced in this manner, they were able to accurately carry out geometry optimisations with chemical accuracy [ 147 ] for biomolecules [ 148 ], charged systems [ 149 ] and non-covalent interactions [ 150 ].…”

Guest Editorial: Special Topic on Data-Enabled Theoretical Chemistry

“…Another type of ML method is GPR or Kriging, and Di Pasquale et al 43 use it to predict energies of ions solvated in water. Energies are based on atomic energies obtained from the topological partitioning called interacting quantum atoms.…”