[gmx-users] weird tabulated potential problem

LuLanyuan lulanyuan at msn.com
Wed Oct 14 21:25:02 CEST 2009

Some months ago I reported a strange conflict between the GMX 4 new option nblist=-1 and tabulated potentials. At that time I contacted Berk and he could run my test system without any problem. So I thought I had something wrong for my system configurations. However, recently I re-checked the problem and found I got similar issues from two different software/hardware configurations. It seems the problem is sort of general. I'm wondering if any developer or people who are familiar with the code can help me to figure out the problem.
The problem is like the follows. When I use nblist=-1 option with tabulated nonbond interation, sometimes I got a segmatation fault. I used nonebond tables with space 0.004 nm and I smoothed the curves so no complain of wrong numerical derivatives from GMX were found. I tested the simulation on three machines:
A: Intel i7 gcc 4.1.2
B: Intel Xeon icc 10.1
C: AMD Barcelona icc 10.1
All tests were done using one CPU core. It seems system A/B always gave me troubles and C was fine.
I first used rlist =1.3 and rvdw=1.2 to give a 0.1 buffer. And I found the potential energys from A/B were weired with values sometimes as 1.0e29. And the potential energy with C is always around 1.0e3. When I set rlist=1.28, I can optimize the "Average neighborlist lifetime" about 11, as Berk onece suggested. But in this buffer size, simulations with A/B got segmentation fault immediately. Interestingly, I tried to try many conventional force field simulations/CG simulations with tabulated potentials before with A/B, and all of them seemed to be fine if I didn't use nblist=-1 with tabulated potentials.
Can anyone give me any suggestions?
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-users/attachments/20091014/50bf3f72/attachment.html>

More information about the gromacs.org_gmx-users mailing list