The dynamics of a fluid on macroscopic and usually even on mesoscopic scales is governed by the Navier-Stokes equations. Solving them provides the velocity of a fluid at each point in space, where ?point? means in fact a region of space large compared to the microscopic scale of atoms and molecules. Solving the Navier-Stokes equation without attitional approximations (called direct numerical simulation, DNS) quickly reaches its limits due to necessity to resolve finer and finer scales as the flow becomes more turbulent. Why not use computer resources to attack the problem from the other side: a bottom-up approach using the full atomistic description of the fluid? Of course, the limits for atomistic computational fluid dynamics are even more restrictive than for DNS, but an atomistic approach promises new physical insight beyond the assumptions of the Navier-Stokes equations. We plan to assess the feasibility of at least two freely available molecular dynamics (MD) code, LAMMPS and NAMD, for large scale fluid dynamic simulations, evaluate their scale-up an size-up behavior on the some of the computer architectures provided by PRACE, and investigate the problem of handling the large amount of data generated, the parallel file IO, and come up with strategies to reduce the amount of data by on-the-fly analysis of the fluid dynamics generated by the MD simulation.
Forschungs-