Re: [AMBER] Bug when using RESPA in REMD runs

From: Charo del Genio <>
Date: Sat, 30 Jan 2021 15:51:18 +0000

Dear all,
        thank you for your answers. Based on what Brian wrote before, I am fully convinced that the bug lies in the REMD exchanges being evaluated 1 step later than reported, and therefore at a time step
that is not a multiple of nrespa. As for the slowdown from not using it, currently it's about 15%, so not too bad.

Also thanks for the other suggestions and ideas. So far, I haven't yet tried to use mass repartitioning, and I'm still limited to CPU runs for various other reasons, mainly the big difficulty in the
UK in getting any funding for hardware and software outside of Russell-Group universities.

For what concerns the isokinetic integrators, I assume you are referring to those available with ntt=9 and ntt=10. If that's the case, I have played with them a bit in the past (particularly ntt=9),
but somehow the trajectories I got didn't really convince me. I didn't carry out a proper comparison study with "regular" Langevin dynamics, so I can't really bring a quantitative argument to my gut
feeling. Qualitatively, the motion of the system I tried them on seemed very "constrained", so to speak, as if it the solvent had a very high viscosity and the motion of the solute was very smoothed.

Anyway, thank you all again for the discussion.



Dr. Charo I. del Genio
Senior Lecturer in Statistical Physics
Applied Mathematics Research Centre (AMRC)
Design Hub
Coventry University Technology Park
Coventry CV1 5FB
AMBER mailing list
Received on Sat Jan 30 2021 - 08:00:02 PST
Custom Search