Fast, Scalable Algorithms for Simulating Complex BiofluidsComputer Science and Engineering
New York University
Large-scale suspensions of soft-particles in viscous fluids are ubiquitous in biology: red blood cells, platelets and proteins are examples of passive suspensions; bacteria and spermatozoa are examples of active suspensions. Advances in experiments, theory and simulations are essential for understanding and engineering these systems. Direct numerical simulations have been limited in scope because of the significant computational challenges posed by the complex moving geometries, the nonlinear interfacial forces and the non-local hydrodynamic interactions.
In this talk, I will present new computational schemes for simulating certain classes of complex fluids. They overcome the numerical stiffness of the associated governing equations, compute N-body hydrodynamic interactions in linear time and incorporate spectral boundary representations and low-cost preconditioners. These schemes scale up to 200 thousand processors simulating over 260 million soft-particles suspended in a viscous fluid, an improvement of several orders of magnitude over previous results. Overall, we have achieved 0.7 Petaflops of sustained performance on the Jaguar supercomputer. I will discuss new physical insights gained via simulations on the red blood cell dynamics in nonlinear flows, self-organization in confined geometries and shape transformations in gravity.