Performance of Low Synchronization Orthogonalization Methods in Anderson Accelerated Fixed Point Solvers

10/19/2021
by   Shelby Lockhart, et al.
0

Anderson Acceleration (AA) is a method to accelerate the convergence of fixed point iterations for nonlinear, algebraic systems of equations. Due to the requirement of solving a least squares problem at each iteration and a reliance on modified Gram-Schmidt for updating the iteration space, AA requires extra costly synchronization steps for global reductions. Moreover, the number of reductions in each iteration depends on the size of the iteration space. In this work, we introduce three low synchronization orthogonalization algorithms into AA within SUNDIALS that reduce the total number of global reductions per iteration to a constant of 2 or 3, independent of the size of the iteration space. A performance study demonstrates the reduced time required by the new algorithms at large processor counts with CPUs and demonstrates the predicted performance on multi-GPU architectures. Most importantly, we provide convergence and timing data for multiple numerical experiments to demonstrate reliability of the algorithms within AA and improved performance at parallel strong-scaling limits.

READ FULL TEXT

page 7

page 8

research
04/02/2021

Low-Synch Gram-Schmidt with Delayed Reorthogonalization for Krylov Solvers

The parallel strong-scaling of Krylov iterative methods is largely deter...
research
09/16/2018

Low synchronization GMRES algorithms

Communication-avoiding and pipelined variants of Krylov solvers are crit...
research
05/26/2022

An Acceleration of Fixed Point Iterations for M/G/1-type Markov Chains by Means of Relaxation Techniques

We present some accelerated variants of fixed point iterations for compu...
research
03/03/2022

Improved convergence of the Arrow-Hurwicz iteration for the Navier-Stokes equation via grad-div stabilization and Anderson acceleration

We consider two modifications of the Arrow-Hurwicz (AH) iteration for so...
research
02/10/2022

Non-stationary Anderson acceleration with optimized damping

Anderson acceleration (AA) has a long history of use and a strong recent...
research
05/04/2019

Predict-and-recompute conjugate gradient variants

The standard implementation of the conjugate gradient algorithm suffers ...
research
10/22/2021

The convergence analysis of an accelerated iteration for solving algebraic Riccati equations

The discrete-time algebraic Riccati equation (DARE) have extensive appli...

Please sign up or login with your details

Forgot password? Click here to reset