| Rev |
Age |
Author |
Path |
Log message |
| 2023 |
1070d 07h |
eromero |
/trunk/src/eps/impls/davidson/common/dvd_calcpairs.c |
- Improved the performance of dvd_calcpairs_selectPairs_qz
- Added the mpd support to davidson solvers
- Improved the checking of the parameters for the davidson solvers |
| 2022 |
1070d 09h |
eromero |
/trunk/src/eps/impls/davidson/common/dvd_calcpairs.c |
- Bug fixed at function dvd_compute_eigenvector |
| 2021 |
1072d 00h |
eromero |
/trunk/src/eps/impls/davidson/common/dvd_calcpairs.c |
- Removed checknullspace from STPrecond
- Postponed the default type of st->ksp->pc until the setfromoptions or setup in STPrecond
- CHKERRQified a lot of code
- bug fixed (although unimportant): enforced ldpX and ldpY to d->size_H in dvd_updateV_conv_gen and dvd_updateV_restart_gen
- bug fixed: davidson solvers modified eps->ispositive bothering SVD
- Added mpd and the left vectors checkings to davidson solvers |
| 2012 |
1073d 14h |
eromero |
/trunk/src/eps/impls/davidson/common/dvd_calcpairs.c |
- Adapted veccomp, precond, gd and jd to PETSc with c++ |
| 1992 |
1091d 06h |
eromero |
/trunk/src/eps/impls/davidson/common/dvd_calcpairs.c |
- Corrected the state of the vectors in functions in dvd_blas
- Removed many calls to PetscObjectStateIncrease in VecComp object because the PETSc interface for Vec does it
- Completed the code that frees all the allocated objects by the Davidson solver |
| 1989 |
1093d 07h |
eromero |
/trunk/src/eps/impls/davidson/common/dvd_calcpairs.c |
- Integrated the deflation space into the davidson solvers |
| 1987 |
1094d 01h |
eromero |
/trunk/src/eps/impls/davidson/common/dvd_calcpairs.c |
- Merging from branches/davidsones to trunk |
| 1985 |
1094d 14h |
eromero |
/branches/davidsones/src/eps/impls/davidson/common/dvd_calcpairs.c |
- Renamed EPS_DAVIDSONES to EPS_DAVIDSON
- Changed davidsones.* davidson.*
- Removed the custom fortran wrapper tags from the EPSGDGet* |
| 1976 |
1101d 09h |
eromero |
/branches/davidsones/src/eps/impls/davidson/common/dvd_calcpairs.c |
- Transformation of the experimental Davidson solver "DAVIDSONES" to a full integrated family of solvers "davidson". The first solver is "gd". The code shared by the "davidson" solvers is packaged by the virtual solver "DAVIDSON": EPSSetUp_DAVIDSON, EPSSolve_DAVIDSON...
- Partially fixed bug: the size of the starting searching subspace is set by eps->nini, but this value can be negative.
- Integration of "davidsones_harm" as "eps_extraction" |
| 1975 |
1106d 09h |
eromero |
es/dvd_calcpairs.c |
- Updated from trunk up to r1974
- Made davidsones compatible with EPSSetInitialSpace |
| 1968 |
1121d 08h |
eromero |
es/dvd_calcpairs.c |
- Added a routine for computing the associated eigenvector to a Schur decomposition
- Bug fixed: now nR starts with PETSC_MAX |
| 1966 |
1126d 10h |
eromero |
es/dvd_calcpairs.c |
- In real arithmetic, the conjugate complex eigenpairs are enforced to converge together |
| 1883 |
1234d 05h |
eromero |
es/dvd_calcpairs.c |
- Independent harmonic target and EPS target
- Added the 'fix' parameter for JD
- Homogeneous EPS target representation
- Fixed bug: error extracting the eigenvectors from the converged Schur decomposition for the complex case
- Fixed bug: harmonic eigenpairs weren't back-transformed
- Updated the Krylov initial subspace and added as new option
- Added 'calcpairs_proj_res' for custom residual computation
- Standard problems solved with harmonic are handled as generalized
- Fixed minor bugs in 'dvd_calcpairs_projeig_qz_gen', 'dvd_schm_basic_preconf', 'dvd_managementV_basic'
- 'dvd_improvex_jd_proj_uv_0' implements an advance correction equation, instead of a collection
- Bug fixed: 'dvd_jacobi_precond' test if MatGetDiagonal is available
- Bug fixed: corrected some harmonic transformations
- Changed HarmType_t to enum type
- Added function lists called in the starting and in the ending of EPSSolve
- Improved the support of preconditioners |
| 1867 |
1297d 01h |
eromero |
es/dvd_calcpairs.c |
First version of Jacobi-Davidson variant based on Manel Cruz (mbc@isep.ipp.pt) code |
| 1829 |
1344d 14h |
eromero |
es/dvd_calcpairs.c |
- Update from trunk up to r1827
- First version of the new variant form GHEP without B-orthonormalization. Present convergence problem with relatively small ncv |
| 1821 |
1357d 16h |
eromero |
es/dvd_calcpairs.c |
- Bug fixed: imaginary eigenvalues were not computed correctly. |
| 1820 |
1358d 14h |
eromero |
es/dvd_calcpairs.c |
- Bug fixed: there were some uninitialized variables.
- Corrected the memory requirements. |
| 1809 |
1360d 08h |
eromero |
es/dvd_calcpairs.c |
- Update of olsen variant for harmonics version |
| 1805 |
1360d 11h |
eromero |
es/dvd_calcpairs.c |
- Merge from trunk up to r1795: adaptation of IPOrthogonalize and EPSSortDenseSchur
- Bug fixed: when harmonics is used, new vectors in V are orthonormalized against cX, instead of cY. |
| 1795 |
1363d 11h |
eromero |
es/dvd_calcpairs.c |
- Rewriting of updating, restarting and converging routines
- First implementation of harmonics (only largest eigenvalues tested). For some ncv, the solver doesn't converge or converge to wrong eigenpairs. Moreover, it is unstable when the number of processors grows |