[gmx-users] restart again

wnoid at hec.utah.edu wnoid at hec.utah.edu
Fri Feb 24 23:16:30 CET 2006

hello all.

i'm sorry to bother everyone again, but i have determined a little more
information regarding my restart problem.  first of all i have found the
problem to occur on two different 32 bit platforms but not on the only 64
bit processor that i have tested.  moreover it occurs both for 1cpu and
muli-cpu jobs and irrespective of whether i compile with gnu/intel or
with/without ia32_sse loops.  the problem persists when i am running nve
simulations so thermo/baro stat coupling is not the problem.

it appears that when i use tpbconv to continue a simulation, it reads in
the positions and velocities with high accuracy.  if i try to extend a 1
step simulation with dt-->0, then i find that x are correct to within 1
part in 1e14 (using double or single precision).  however i find that the
forces are only correct to within 1 part in 1e3 (again independent of the
precision for which i tried to compile).  in fact if i look in the mdrun
log file i find that the LJ(SR) energy differ at 1 part in 1e5 between
calculations for the last config of the first simulation and the
calculation for the (0 step) first config of the continuation simulation. 
i would have expected these numbers to be exactly the same since they are
both calculated from what appears to be the same configuration.

why is there so much inconsistency in the force calculation and even in
the LJ(SR) energy?  is this to be expected?  am i missing something?

anyway thanks to anyone who took the time to look at this.  any help would
be greatly appreciated.


will noid

More information about the gromacs.org_gmx-users mailing list