Tupek/trust region subspace solver #1288
Conversation
|
/style |
tupek2
force-pushed
the
tupek/trust_region_subspace_solver
branch
from
c8d6425 to
92e28a2
Compare
|
/style |
1 similar comment
|
/style |
|
/style |
|
im not too familiar with petsc/ mfem, but could a lot of the petsc wrapper stuff use mfem's petsc wrapper classes? https://github.com/mfem/mfem/blob/master/linalg/petsc.hpp ? also https://github.com/mfem/mfem/blob/master/linalg/slepc.hpp |
|
The mfem wrappers are intended for 'parallel' across processor things. The new wrapper is for local, small, dense things. In general, many coding experts recommend creating internal abstractions to external library wrappers anyways, but in this case, the mfem one doesn't even really fit the purpose. |
tupek2
force-pushed
the
tupek/trust_region_subspace_solver
branch
from
0858469 to
f7164f3
Compare
|
/style |
Subspace trust region solver using petsc.
… info on algorithmic counts.
tupek2
force-pushed
the
tupek/trust_region_subspace_solver
branch
from
0da6a06 to
b0e61ee
Compare
| auto eigh(const DenseMat& Adense) | ||
| { | ||
| auto [isize, jsize] = Adense.size(); | ||
| SLIC_ERROR_IF(isize != jsize, "Eig must be called for symmetric matrices"); |
There was a problem hiding this comment.
| SLIC_ERROR_IF(isize != jsize, "Eig must be called for symmetric matrices"); | |
| SLIC_ERROR_IF(isize != jsize, "'eigh' must be called for symmetric matrices"); |
?
| * @brief compute the quadratic energy from small dense matrices and vectors | ||
| * @param A The stiffness matrix | ||
| * @param b The rhs vector | ||
| * @param x The current solution vector |
There was a problem hiding this comment.
Does the doxygen check not throw a warning/error on missing the return parameter documentation?
There was a problem hiding this comment.
it does not seem to error. Presumably our functions should be named such that the return is often obvious.
|
/style |
No description provided.