Adaptive time-stepping for transient solver using SUNDIALS #292
Conversation
…ve explicit RK integrator
simlapointe
added
enhancement
There was a problem hiding this comment.
Looking good. There's a few smaller structural issues, but the two bigger things I am thinking about
a) Can the RUNGE_KUTTA option use instead the mfem internal time integrator? That way the sundials connection is really only used for the adaptivity. This would slightly tidy the interface for adaptivity, along with making the non-adaptive slightly more fully featured. There are some SDIRK options in there that would probably be good choices.
b) Is there a way to compute B implicit to this process?
I have some of my suggestions on hughcars/transient-adapt-dt if you want to take a look.
cmake/ExternalSUNDIALS.cmake
Outdated
| if(PALACE_WITH_CUDA) | ||
| list(APPEND SUNDIALS_OPTIONS | ||
| "-DENABLE_CUDA=ON" | ||
| ) | ||
| endif() | ||
|
|
||
| if(PALACE_WITH_MAGMA) | ||
| list(APPEND SUNDIALS_OPTIONS | ||
| "-DENABLE_MAGMA=ON" | ||
| "-DMAGMA_DIR=${MAGMA_DIR}" | ||
| ) | ||
| endif() |
There was a problem hiding this comment.
What tests have you done with CUDA/MAGMA builds?
There was a problem hiding this comment.
I just tried building with CUDA/MAGMA and SUNDIALS' ENABLE_MAGMA was causing a build failure. I removed it and Palace builds without issues but I'm getting NaNs at step 0 of the transient solve. Will try to understand what's going wrong. It happens even with generalized alpha, so I don't think it's SUNDIALS related.
| En += E; | ||
| Curl->AddMult(En, B, -0.5 * dt); |
There was a problem hiding this comment.
Is there a better way to be computing B as part of this?
There was a problem hiding this comment.
I don't think so. We solve the ODE system for E and Edot, not sure how else to get B.
Add adaptive time-stepping capability for transient simulations. The default transient solver type remains a fixed time-stepping integrator, but the user can now use SUNDIALS implicit multistep (CVODE) and Runge-Kutta (ARKODE) integrators if desired.