(637e) Benchmark of Topology Automated Force-Field Interactions (TAFFI)

Recent trends in the development of atomic force-fields include machine learning based potentials (MLP) and quantum chemistry derived force-fields. An important issue for these methods is the trade-off between the accuracy and complexity of the force field, where the high accuracy is achieved at the expense of transferability compared to classical force fields. Here we present an all-atom version of Topology Automated Force-Field Interactions (TAFFI), TAFFI-gen, which provide the development of force fields with the desired degrees of complexity. This is achieved by building an arbitrary extendable and transferable force field using self-consistent atom typing and DFT-based parameterization schemes. The performance of the derived parameters will be presented in terms of liquid property predictions revealing the representability limitation.