Anderson acceleration has demonstrated significant benefits in accelerating fixed point solutions in a number of applications. The method, however, adds new synchronization points that can slow its use in parallel. In this presentation, we will examine the parallel communication requirements of Anderson acceleration and discuss its performance for parallel application. In addition, we will discuss use of Anderson acceleration on GPUs.
This work was performed under the auspices of the U.S. Department of Energy by Lawrence Livermore National Laboratory under Contract DE-AC52-07NA27344. Lawrence Livermore National Security, LLC. LLNL-ABS-681046.