warm start updates and tied contact patch test #1260
Conversation
|
/style |
|
This one should be ready to go! |
| contact_.update(cycle_, time_, dt); | ||
| if (contact_.haveLagrangeMultipliers()) { | ||
| J_offsets_ = mfem::Array<int>({0, displacement_.Size(), displacement_.Size() + contact_.numPressureDofs()}); | ||
| J_constraint_ = contact_.jacobianFunction(J_.release()); |
There was a problem hiding this comment.
this call does not take displacement. Are we sure that the contact_ has the updated value for displacement internally at this point?
There was a problem hiding this comment.
Yes, the updated displacements are provided when contact_.update() is called on L309. I added some preconditions to the doxygen comments in contact_data.hpp so it's clear you have to call update() with the current config if you want the Jacobian contributions to be accurate.
Yeah, this helps. It cuts down on the number of Newton iterations needed for all of the contact examples as compared to turning warm start off. |
| * For the displacement function, the argument is the input position, the second argument is the time, and the output | ||
| * is the value of the component of the displacement. | ||
| * | ||
| * @note This method must be called prior to completeSetup() |
There was a problem hiding this comment.
Can/should this be checked at runtime?
There was a problem hiding this comment.
Yeah, that's a good idea. @btalamini is doing a more complete overhaul of this now. Could this check be added to your Dirichlet BC overhaul, Brandon?
There was a problem hiding this comment.
We have these comments all over the place. At some point we can go through and have a check that people are not calling this after completeSetup(), but I think its out of the scope for this.
| const mfem::Vector res = (*residual_)(time_ + dt, shape_displacement_, displacement_, acceleration_, | ||
| *parameters_[parameter_indices].state...); | ||
|
|
||
| // TODO this copy is required as the sundials solvers do not allow move assignments because of their memory |
There was a problem hiding this comment.
Probably a TODO once mfem/mfem#3531 is addressed, but it doesn't look like that will happen in the near term. I'll change it to a NOTE
| StateManager::setMesh(std::move(mesh), "patch_mesh"); | ||
|
|
||
| // #ifdef SERAC_USE_PETSC | ||
| // LinearSolverOptions linear_options{ |
There was a problem hiding this comment.
Should this be removed or a note added why its commented out?
There was a problem hiding this comment.
Note's been added. We should add PETSc back in, but the convergence was slow with these settings.
| blt_list_append(TO physics_parallel_test_sources ELEMENTS | ||
| contact_patch.cpp | ||
| contact_patch_tied.cpp | ||
| contact_beam.cpp | ||
| IF TRIBOL_FOUND) |
There was a problem hiding this comment.
| blt_list_append(TO physics_parallel_test_sources ELEMENTS | |
| contact_patch.cpp | |
| contact_patch_tied.cpp | |
| contact_beam.cpp | |
| IF TRIBOL_FOUND) | |
| blt_list_append(TO physics_parallel_test_sources | |
| ELEMENTS | |
| contact_patch.cpp | |
| contact_patch_tied.cpp | |
| contact_beam.cpp | |
| IF TRIBOL_FOUND) |
This PR adds penalty and Lagrange multiplier contact to the warm start. Additionally, it adds a tied contact patch test that fails if warm start does not account for contact constraints.
Some other small changes:
ContactData::jacobianFunction()was not using the displacement vector, so that argument has been removed from the methodode_time_point_totime_inSolidMechanicsContactto match what's inSolidMechanics