AMBER: PMEMD BUG ALERT!!!

From: Robert Duke <rduke.email.unc.edu>
Date: Fri, 22 Aug 2003 16:50:02 -0400

Dear Amber Gang -
Turns out that the problem reported by James Tomomi MacDonald is indeed a
bug in PMEMD, probably introduced as a typo in the 3.00 Release. Basically
COM motion removal has not been working, and if you have the "ewald block of
flying ice" problem in your simulation, it does not get fixed by COM
velocity removal. In really long simulations with COM motion, this is
eventually not good. The fixes that need to be applied are to runmd_cit.f90
and runmd_cit7.f90, as shown by the diff's below:

runmd_cit.f90 (Release 3.01):

734c734
< do j = 1, my_atm_lo, my_atm_hi
---
>       do j = my_atm_lo, my_atm_hi
This is line 732 in Release 3.00
runmd_cit7.f90 (Release 3.01):
767c767
<       do j = 1, my_atm_lo, my_atm_hi
---
>       do j = my_atm_lo, my_atm_hi
This is line 765 in Release 3.00
Basically, the 1, got left in by mistake when COM velocity correction was
being effectively parallelized (this stuff used to be done by all nodes on
all atoms).
If your simulations have not been blowing up with the "block of flying ice
phenomenon", you are basically okay, but I would sure pick up the fix.  I
will try to get a tested 3.02 out to the web site soon, and will build new
executables for the folks that I do this for.  Very sorry about that!
Regards - Bob
-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber.scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo.scripps.edu
Received on Fri Aug 22 2003 - 21:53:01 PDT
Custom Search