[gmx-users] still about the opls methanol box

[Erik Lindahl]

Hi,

> which seems quite strqnge, I thought I had rcoulomb to 1.4 but maybe it's 
> irrelevant for PME?

Yes - for PME the long-range (infinite) coulomb forces are calculated 
every step in the reciprocal space.

> 
> Anyway the box size is about 4.8 so it's not a problem of having a cut-off 
> less than half the box edge.
> 
> The machine is a G4 macintosh under OS X version 10.1 and mdrun 
> uses altivec loops.
> 
> I tried then to test if the crash is reproducible. The previous run crashed at 
> 221ps and I recorded at the trr every 100ps, so I began from 200ps giving 
> the -t option in grompp. The run is now at 61ps, so it passed the previous 
> crash point by 40ps and continues OK. But it is not exactly 
> superimposable to the corresponding portion of the previous run. I think 
> this is normal when you continue a run?
> I attach a small png graph of the densities, in black points the first run 
> (crashed at 221ps) and in red line the second, continued from 200ps.
> 

That's normal; by reproducible I mean if the crash happens again if you 
start from exactly the same .tpr file (sorry if that will take a while).

Cheers,

Erik