(167a) GROW: a GRadient-Based Optimization Workflow for Model Building Beyond Manual Intervention

Nowadays, molecular simulations are widely used to support the development process for new materials. Through simulations researchers are capable of predicting general trends quite well. But the key to quantitatively correct property predictions is the accuracy of a simulation's foundation, the force field. A force field describes the intra- and intermolecular interactions by a semi-empirical equation and its associated parameters. While the equation's functional form is usually clear, the force field parameterization is often tedious. Thereby manual adjustment and optimization is, at best, extremely time consuming. Hence, an automated parameterization scheme is essential in our pursuit to create tailor-made models for specific investigations in a timely fashion. This has been realized and implemented into a Gradient-based Optimization Workflow (GROW) for the automated development of molecular models.

GROW is a program tool kit to facilitate the use of gradient-based numerical optimization of force field parameters. Its components include various optimization algorithms (including Quasi Newton algorithms or Trust Region methods), analysis scripts, and I/O-handling. GROW can be ?attached? to various standard MD simulation engines, making it a powerful companion for many application fields.

In this presentation, we will show how various algorithms perform and discuss their advantages and disadvantages. Experimental target values, used for verification purposes only, can be typically matched by GROW within a few percent for common properties (e.g. density, thermal and electrical conductivity, diffusion, and vapor pressure). This benefit will be illustrated by some atomistic force field models for small solvent molecules, polymer precursors, and ionic liquids.