Abstract
With reactive force fields it is possible to perform atomistic simulations that join the accuracy of quantum chemical treatments (including bond breaking and formation) with the ability to treat hundreds of thousands of atoms on time scales well into the nanosecond regime. To utilize this power in everyday applications requires (I) the assembly of a suitable reference data set of sufficient quality, and (II) a reliable fit of the huge and complex parameter set of a general reactive force field to these reference data. In this contribution, we show that genetic algorithms can be used to achieve goal (II). We discuss algorithm design and implementation aspects (including parallelization) and present an application to azobenzene as real-life example.