New local ROM approaches for Rayleigh-Taylor instability

siuwuncheung commented 3 years ago

Add an implementation of using penetration distance and using the parameter-time rather than physical time as the indicator the trust-region of local ROM for 2D Rayleigh-Taylor instability.

siuwuncheung commented 3 years ago

@chldkdtn @dylan-copeland @kevinhkhuynh Ready for review. Undergoing regression tests.

dylan-copeland commented 3 years ago

@siuwuncheung Are there any new tests? I don't see any changes to the regression tests. Should there be a short regression test for maintaining the new capability? Also, is there a longer, more interesting test we can try? Does it work in parallel?

siuwuncheung commented 3 years ago

@siuwuncheung Are there any new tests? I don't see any changes to the regression tests. Should there be a short regression test for maintaining the new capability? Also, is there a longer, more interesting test we can try? Does it work in parallel?

I will add new short regression tests for this capability.
The command line options and the results for long time simulation with final time is 4.5 is tabulated on Confluence.
The penetration distance is in the context of Rayleigh-Taylor instability only, but in general problems there can be a small set of easy-to-evaluate functional which marks or is highly correlated to the position of the shock, the same idea should apply (probably the Sedov Blast explosion or the Riemann problem).
I think the merge phase does not work in parallel and therefore the online phase either, but I may be wrong. In that case we need to fix that, especially if we want to use higher order discretizations on fine meshes to show a good speed up.

chldkdtn commented 3 years ago

@siuwuncheung what is the status of this PR?

siuwuncheung commented 3 years ago

@siuwuncheung what is the status of this PR?

There are some new capabilities that I want to add to enable parametric distance windowing for long-time simulation. Right now the offset problem is resolved and some file I/O settings are being added. After that I think this PR should be reviewed or merge, or otherwise it will grow too large and complicated.

CEED / Laghos

New local ROM approaches for Rayleigh-Taylor instability #149