Asynchronous block-iterative methods for almost linear equations (Q806957)

On a distributed multiprocessor with asynchronous communication, a block relaxation method is considered for the solution of a large system of equations with linear and Lipschitz-continuous nonlinear part. For bounded communication delays the authors show how to transform the iteration into standard form; then a general sufficient convergence condition is formulated. The method has been implemented on a 4-processor transputer with direct connections between all processors. On a 44\(\times 44\) system (not shown but said to be from power network background), convergence times of synchronous and asynchronous iterations are compared and speedups reported (of about 2); \(\omega =1\) was observed to be optimal. The choice of the partition of the matrix is said to be crucial but difficult.