A parallel algorithm for long-timescale simulation of concentrated vesicle suspensions in three dimensions

We develop a parallel boundary integral method for simulating highly concentrated vesicle suspensions in a Stokesian fluid. This method is an extension of our previous work [1]. The simulation of high volume fraction vesicle suspensions, which are representative of real biological systems (such as blood with 35% ∼ 50% volume fraction for RBC) presents several challenges. It requires computing accurate vesicle-vesicle interactions at length scales where standard quadratures are too expensive. The inter-vesicle separation can become arbitrarily small leading to vesicle collisions. Numerical errors can accumulate over time, making long-timescale simulations inaccurate. We tackle these challenges by developing state-of-the-art parallel algorithms for efficient computation of boundary integrals and an adaptive time-stepping scheme. We have also developed algorithms for handling vesicle collisions, remeshing of vesicle surfaces and correcting drift in vesicle area and volume. We study the accuracy of our method when compared to a reference solution and show convergence with the spatial discretization order and the time step size. We visualize long-timescale simulations for periodic Taylor-Green vortex flow and sedimentation of polydisperse vesicle suspensions with thousands of vesicles. For these flows, we present strong and weak scaling results on thousands of CPU cores on the Stampede system at Texas Advanced Computing Center.