Massively-Parallel Mesoscale Hydrodynamics on Graphics Processing Units

Multiparticle collision dynamics (MPCD) is a particle-based mesoscale method for resolving fluctuating hydrodynamics in computer simulations. A fluid is effectively represented by point particles that periodically undergo stochastic, momentum-conserving multiparticle collisions that impart long-ranged hydrodynamic interactions. MPCD is particularly useful for soft matter, including colloid-polymer mixtures, where it is important to resolve microscopic structural details and thermal fluctuations. In principle, MPCD simulations can be significantly more computationally efficient than conventional explicit-solvent molecular dynamics, allowing for large length-scales and long time-scales with fully resolved hydrodynamics. However, typical MPCD simulations may still require tens of millions of MPCD particles or more, necessitating a parallel computational approach. We present our development of the first massively-parallel, open-source implementation of the MPCD method for graphics processing units in HOOMD-blue. Our implementation fully exploits computational resources from desktop workstations up to leadership-class supercomputers, and will significantly accelerate simulations of soft matter systems where hydrodynamic interactions are important.