| #
df90af56
|
| 24-Oct-2018 |
Matthew G. Knepley <knepley@gmail.com> |
Merge branch 'master' into arcowie-rem/feature-error-logging
* master: (393 commits)
Bib: Update reference
Mat: Doc fix
Bib: Updated ref
PetscDS: Doc fixes
PC+LU: Do not try to refactor an
Merge branch 'master' into arcowie-rem/feature-error-logging
* master: (393 commits)
Bib: Update reference
Mat: Doc fix
Bib: Updated ref
PetscDS: Doc fixes
PC+LU: Do not try to refactor an already factored matrix
Mat: Small fix for checking and docs
Mat: Added MatSetFactorType() - Needed it when making a shell matrix look factored
PetscDS: Added PetscDSUpdateBoundary() - Lets the user change the boundary condition
single precision produces different convergence history
p4est: has a dependency on zlib - so handle it correctly
Add -mat_mffd_complex to use the Lyness complex number trick to compute J_u * v instead of differencing.
Replace VecWAXPY by VecAXPY if needed
Revert "Replace VecWAXPY by VecAXPY if needed"
Replace VecWAXPY by VecAXPY if needed
MatHermitianTransposeGetMat and MatCreateVecs
ex19: Updated test filter to avoid false positives.
DM: Improved Global-To-Natural docs Suggested-by: Josh L <ysjosh.lo@gmail.com>
Do not build shared openblas when doing static build
Upgrade OpenBLAS to 0.3.3
test: add alt output file for changes in OSX-10.14 and Xcode-10.0 [perhaps related to ML]
...
Conflicts:
src/vec/is/utils/vsectionis.c
show more ...
|
| #
bf67e7b2
|
| 03-Oct-2018 |
Karl Rupp <me@karlrupp.net> |
Merge branch 'denera/tao-gauss-newton-wrapper' [PR #1151]
* denera/tao-gauss-newton-wrapper:
Regularized Gauss-Newton algorithm/wrapper
New BRGN algorithm added for regularized Gauss-Newton formulat
Merge branch 'denera/tao-gauss-newton-wrapper' [PR #1151]
* denera/tao-gauss-newton-wrapper:
Regularized Gauss-Newton algorithm/wrapper
New BRGN algorithm added for regularized Gauss-Newton formulation.
This algorithm is a thin wrapper that relies on TaoSetResidualRoutine() and TaoSetResidualJacobianRoutine() interfaces to accept the residual and Jacobian associated with a least-squares problem,
and then construct the internal objective, gradient and Hessian evaluation functions for the regularized problem formulation.
Currently only Tikhonov regularization is available, but other smooth regularizers are planned in the future.
The resulting regularized problem is passed onto an existing bound-constrained algorithm (e.g.: BQNLS or BNTL), controlled by the -tao_brgn_subsolver flag.
show more ...
|