EPS_CISS_STRATEGY_SPLIT#
Use a collection of independent KSP objects for the linear solves associated with integration points in the EPSCISS solver.
Notes#
For the linear solves \((A-z_i B)Y_i=BV\) at every integration point \(z_i\), this
strategy uses an array of linear solver objects, see EPSCISSGetKSPs().
The number of KSP solvers is equal to the number of integration points divided by
the number of partitions, see EPSCISSSetSizes(). This value is halved in the case
of real matrices with a region centered at the real axis.
This strategy uses more memory, compared to EPS_CISS_STRATEGY_USEST, but it
potentially has more parallelism. The memory needs grow with the number of integration
points, and the degree of parallelism increases with the number of partitions.
For large scale problems, this approach is more scalable, with two levels of parallelism
several groups of MPI processes handling different integration points simultaneously and solving systems in parallel within the subcommunicator.
See Also#
EPS: Eigenvalue Problem Solver, EPSCISSStrategy, EPSCISSSetStrategy(), EPSCISSGetKSPs(), EPSCISSSetSizes(), EPS_CISS_STRATEGY_USEST, EPS_CISS_STRATEGY_MULTISHIFT
Level#
advanced
Location#
Index of all EPS routines Table of Contents for all manual pages Index of all manual pages