On 3/4/2021 10:37 AM, Gabella, William E wrote:
Present: Peter D, Bill G, Roland H, Atul K, Miguel G, Steve B, Karim S
Chair: Peter, Minutes: Bill
- SPEC benchmark contribution
[https://bitbucket.org/einsteintoolkit/tickets/issues/2470 task] [RH, SB]
Roland, Sent the Makefile to the person at NVidia and a second version that is not sent yet, it is a little nicer. PUGH run with 8th order finite difference and seems to run really slow. Not sure about how long the run the SPEC want. Decided as a test case not to use Minkowski ST too man zeros, so put in a shifted gauge wave. Started with Maclachlan with gamma driver and 1+log lapse (black hole driver) and periodic BCs and seems unstable, the amplitude of the wave grows. Benchmark parameter file chooses the number of cells and the time steps, did satisfy the Courant condition, it is smaller than 0.4 (that is the gallery example), even smaller shows the same growth.
This may not be an original thought, but it occurs to me that figuring out a better gauge condition might be a good application for machine learning, since it's not terribly important that we understand how the gauge works, just that it behaves the way we want.
--Steve