Accurate and reliable description of water.
Studying the complex interactions which take place in water and govern all its phases is not a simple endeavour: the number of possible minima grows combinatorially with the size of the cluster model and their relative energetics must be accurately predicted for the quantitative treatment of aqueous solvation of ions or molecules. Moreover, nuclear quantum effects (NQEs) governing the dynamics of hydrogen bonds must be included to have a meaningful description of the water structure entirely from first principles. Putting together these three necessary ingredients, namely accurate energetics, NQEs, and nuclear dynamics in a complex energy landscape, has made the full understanding of water behavior out of reach.
The exascale-enabled TREX software provides unprecedentedly accurate data for large water clusters and solutes, presently too expensive for traditional highly-correlated methods. Through multi-level grid combination techniques, connecting lower level quantum chemistry results to QMC (Quantum Monte Carlo) numbers, TREX builds predictive ML (Machine Learning) models of atomic energies which can be differentiated to obtain forces for subsequent ML driven geometry relaxations and molecular dynamics studies in bulk liquid water.
About the demonstrator
Since solvation of solutes is at the heart of any process occurring in aqueous medium and dramatically understudied at the correlated level, in this demonstrators TREX uses its software platform combining QMC and ML to train ML models on forces and energies of solvated species predicted by QMC for subsequent molecular dynamics simulations with much larger solvation shells.