...
- The second option is to use Krylov subspace method with different preconditionerfloat version of the sparse direct solver.
Code Block EigenSolver: { NumEigenvalues: 1 FrequencyShift: 10.e9 Preconditioner: MPMUMPSFLOAT //use the float version. memory //thisusage usereduced p-version of multilevel preconditionerinto half. }
- The code will choose either CG (real matrices) or GMRES (complex matrices) and the p-version
of multilevel precondtioner as the solver for shifted linear systems.The third option is to use out-of-core sparse direct solver. third option is to use Krylov subspace method with different preconditioner.
The code will choose either CG (real matrices) or GMRES (complex matrices) and the p-versionCode Block Code Block EigenSolver: { NumEigenvalues: 1 FrequencyShift: 10.e9 MemoryPreconditioner: 1000MP //if the memory usage of//this theuse matrixp-version factorof inmultilevel anypreconditioner. process }
of multilevel precondtioner as the solver for shifted linear systems.
- The fourth option is to use out-of-core sparse direct solver (an experimental feature).
The fourth option is to use float version of the sparse direct solver.Code Block EigenSolver: { is larger than 1000MBytes, NumEigenvalues: 1 FrequencyShift: //switch to use out-of-core solver. }
Code Block EigenSolver: { NumEigenvalues: 1 10.e9 Memory: 1000 //if the memory usage of the matrix factor in any process is larger than 1000MBytes, FrequencyShift: 10.e9 Preconditioner: MUMPSFLOAT //use the float version. memory//switch usageto reduced into halfuse out-of-core solver. }