Philipp Marienhagen (Thermodynamics, HSU)
Recently, it has become possible to combine accurate ab initio potentials with Monte Carlo simulations which allows the theoretical prediction of thermodynamic properties of fluids with an accuracy challenging the best experimental results. This requires the use of a combination of pair and nonadditive three-body potentials with additional corrections for quantum effects to describe the intermolecular interactions. The three-body algorithms thereby scale with the order of N^3 instead of standard simulations which only consider pair potentials (order N^2) limiting the number of particles in the investigated system.
For state points in the critical region, high densities in the liquid and supercritical phases as well as when investigating phase equilibria pronounced finite size effects can be observed. This may even lead to the artificial metastability of different phases restricting theoretical predictions. To accurately investigate thermodynamic properties in these regions significantly larger particle numbers are necessary than currently possible in our parallelized Monte Carlo code which has been optimized specifically for cpu’s considering scalar and vectorization related optimizations. In this performance engineering project we will develop scalable algorithms for the calculation of three-body potentials utilizing gpu’s. Thereby initially the existing concepts for the vectorized cpu implementation can be adapted leading to the further development of new optimization strategies. This would then allow the investigation of the described regions of the phase diagram.