390707 Automated Global and Local Optimization Methods for Atomistic Force Field Development

Tuesday, November 18, 2014: 2:35 PM
212 (Hilton Atlanta)
Dirk Reith1, Thorsten Köddermann2, Marco Hülsmann2, Andreas Krämer1 and Karl N. Kirschner1,2, (1)Simulation Engineering, Bonn-Rhein-Sieg University of Applied Sciences, Sankt Augustin, Germany, (2)Simulation Engineering, Fraunhofer Institute for Algorithms and Scientific Computing (SCAI), Sankt Augustin, Germany

Generally molecular dynamics simulations require the adaptation of different atom or molecular models to desired physcial or chemical properties. In this procedure the  appropriate choice of force fields and of corresponding parameters is a crucial issue.

In recent years, we developed modular program packages, which can be simply tuned to the different occurring optimization problems. Once started, the programs run mostly without user interaction until the force field ist generated. In this way, force field parameters can be determined for complicated atomistic models of any chemical matter. In this way, we address various inter-linked resolutions of molecular modeling to solve one of the primary goals of our research: To develop accurate and reliable molecular parameters and models in a reasonable time and as error-free as possible.  For intramolecular interactions, we have created a scientific “Workflow for force-field  optimization package”  (Wolf2Pack) that incorporates our approach for transfering knowledge  gained from QM to Newtonian- based models. We define a scientific workflow as a series of  independent steps that are linked together according to the data flow and the dependencies between them. For intermolecular interactions we developed a systematic optimization workflow, based on efficient gradient-based numerical algorithms called GROW. GROW is a modular tool kit of programs and scripts. It is a generic implementation and can be easily extended by  other developers. Very recently, GROW was extended with a global optimization module called COSMoS („Calibration/Optimization by Simultaneous Modeling of Simulated data“).

Both programs facilitate: a) the development and optimization of molecular parameters for a given simulation engine, b) the transfer of parameters from one software  package to another, and c) testing of the parameters using a standard test suite and protocol  via a semi-automated iterative parameterization process.

In the presentation, we will show how the programs are developed and set-up along the scientific aspects. Consequently, we will show some example molecules. The physical properties to be fitted are e.g. saturated liquid density, enthalpy of vaporization, and vapor pressure, simultaneously at different temperatures. The resulting force fields will be assessed with respect to their applicability to the prediction of other physical properties over a range of temperatures.


Extended Abstract: File Not Uploaded