University of Cambridge > Talks.cam > Lennard-Jones Centre > Machine Learning Force Field for Organic Liquids: EC/EMC Binary Solvent

Machine Learning Force Field for Organic Liquids: EC/EMC Binary Solvent

Add to your list(s) Download to your calendar using vCal

If you have a question about this talk, please contact Dr Christoph Schran.

Machine learning methods have been employed successfully to model small molecules in vacuum where the highly directional bonding interactions dominate, as well as inorganic solids and liquids where the interactions are weaker but homogeneous. The molecular condensed phase presents unique challenges to ML owing to a large energy scale separation between their intra- and inter-molecular interactions. Molecular liquids with mixed compositions present the additional complication of large dimensionality imbalances between intra-/inter- molecular environments. In practice this makes it difficult to fit the inter- contribution which underlies the thermodynamics of the liquid state. Typical loss functions are on total energies and forces and since inter-molecular interactions are weaker, the intra-molecular force dominates by about an order of magnitude. Therefore, a typical fit would get good intra- and poor inter- relative accuracy. Previous attempts to model molecular liquids with ML have employed several strategies to tackle the intra-/inter- imbalance. One approach is to separate the interaction scales by creating independent force fields for the molecule and for the liquids, respectively. This of course solves the problem of scale separation neatly; however, it also makes it impossible to extend such a model to the reactive regime where the bonds can break and reform allowing chemical reactions to take place. Here we demonstrate that by crafting a sufficiently diverse training set through iterative training and by carefully testing the accuracy of our models on the relevant inter-molecular scale, we can fit a general purpose potential which describes the EC:EMC binary solvent liquid. This paves the way for a more general full reactive force field which will be needed to study the electrode-electrolyte interface with ab initio accuracy.

This talk is part of the Lennard-Jones Centre series.

Tell a friend about this talk:

This talk is included in these lists:

Note that ex-directory lists are not shown.

 

© 2006-2022 Talks.cam, University of Cambridge. Contact Us | Help and Documentation | Privacy and Publicity