Finalizing AMR version (Vlasiator 5) - Githubissues

fmihpc / vlasiator

Vlasiator - ten letters you can count on

https://www.helsinki.fi/en/researchgroups/vlasiator

Other

45 stars 37 forks source link

Finalizing AMR version (Vlasiator 5) #396

Closed galfthan closed 3 years ago

galfthan commented 5 years ago

Initial implementation

[x] Finish field gridglue (sebastian,tuomas)
- Fields are moved, verified in non-AMR case
[x] io from fsgrid (bulk) (urs)
- Enables removal of many glue functions
[x] Max dt for fieldsolver computed on fsgrid (urs)
- Verified for magentosphere test
[x] Setup boundary flags directly on fsgrid (yann,tuomas)
- Compute at same time as dccrg flags, get's rid of one glue, OR implement glue

Cleaning & refactoring after initial implementation:

[x] Remove old coupling functions (tuomas)
[x] Remove unused variables from spatial cell class (tuomas)
[x] Remove coupling code from fsgrid (urs,tuomas?)
[ ] Reorganize fsgrids in container? (urs?)
[x] Re-implement Ehall datareducer https://github.com/fmihpc/vlasiator/blob/dev/datareduction/datareducer.cpp#L542

Validation & verification:

[x] Analysis in (larger) flowthrough tests
[x] Analysis in small Magnetosphere tests
[x] Scalability on Sisu
[x] Thread scalability
[x] vlsvdiff including fsgrid variables

For production

[x] Restart writing of fsgrid data
[x] Restart reading of fsgrid data

Bugs

[x] bvol derivatives are not averaged correctly (see #bug-bvolderivative-avg-in-glue in flowdock)
[x] Diffs in translation algorithm with size >1 pencils
[x] TBA

Numerical improvements

[x] May need interpolation in regridding moments dccrg => fsgrid.
- One option is to use same mapping as in vlasov solver
[ ] Move to a tight coupling at fieldolver dt between fieldsover and VDF transformation.
- See slideset in vlasiator-dev at #fs-vdftransform-coupling
- Not needed for initial V5 release, additional improvement.

ursg commented 3 years ago

Given the state of the AMR runs, this can be considered done! Rejoice!