On Wed, Nov 03, 2021, mish wrote:
>What I wonder most is why MD keeps running despite having NAN in energy
>terms and coordinates.
The GPU code is written to be highly efficient, and doesn't waste time
checking for very rare numerical errors. So one needs to be careful.
The most likely cause of your error lies in having two few non-bonded cells
in the unit cell. It's worth trying a smaller cutoff, and/or a bigger unit
cell. I'm not an expert here, but recall some discussions with Dave Cerutti
about such a problem. (You have a cutoff of 10, and don't ever report what
your unit cell size is (it's reported in the output file). If the ratio
gets to be less than 3 or 4, problems may arise.
....dac
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Fri Nov 05 2021 - 07:00:03 PDT
