Niels Kristian Kjærgård Madsen
Niels Kristian Kjærgård Madsen
Hi again @jan-wassenberg I have now made a patch which passes the checks. I have tested the x86 targets locally, but not the arm, rvv, and wasm. Are those tested...
@jan-wassenberg Great! Thanks for all the helpful comments :-)
This is the case for `clang-tidy` as well. Any ideas on how to fix this?
Isn't the problem simply that you don't remove the `--version` argument from `args` in `Command.parse_args`.
It is correct that the salt screening is not implemented for the P16 kernel, which is why I have used the Still kernel in my tests (and also in the...
Thank you very much for the explanation @awvwgk !
@sef43 That makes sense. If I run the minimization in double precision, the algorithm converges to a much more reasonable result. It is, however, much slower than optimizing in single...
I am getting the error with mixed precision as well. I think the question is whether OpenMM should issue a warning if the energy minimization doesn't converge (or even raise...
@peastman As far as I can see from `LocalEnergyMinimizer::minimize`, the `param.xtol` is set to `1e-7` whenever the platform _supports_ double precision. But what happens if you use CUDA with single...
Okay, I see that the `xtol` is only used in the More-Thuente line search, and OpenMM uses backtracking. So that is not the source of error.