Hi Thomas. I have seen this very often when I use membrane systems, and I
am still wondering why this happens but, for what it's worth, here is what
I think and my solution.
I don't know what the volume or density variation pmemd.cuda expects before
halting is, although as you said it's not much, probably only a few
percent, and not noticeable if you look at it in VMD.
My solution, which I now do automatically, is to run several rounds of
short PT equilibrations with cuda (say 10 rounds, several ns each,
depending on system size), and when I do the production run, say 500 ns, I
run it in parts, 5x100 ns or so, so that I give pmemd the opportunity to
recalculate the box size if it is varying. This way, at least, you don't
waste time. As far as I remember, I had no problems following this protocol
for several different systems.
Of course, this shouldn't be needed, as we are supposed to be using an
equilibrated NPT ensemble, where V should be pretty constant, with small
variations of order 1/sqrt(V); but I'm not sure if that is smaller than the
few percent pmemd.cuda expects.
Long ago I compared this protocol to just using pmemd (non cuda), and
everything appeared to be normal. But it would be really nice if eventually
pmemd.cuda could recalculate the box on-the-fly...
Regards,
Ignacio
On Mon, May 15, 2023 at 5:26 AM Steinbrecher, Thomas via AMBER <
amber.ambermd.org> wrote:
> Hi Amber users,
>
> I see the following error message occurring seemingly at random (in about
> 20% of runs) in some repeats of a large membrane system simulation (running
> Amber22 pmemd.CUDA on RHEL with ff19SB+gaff2+lipid21+OPC water):
>
> -----------------------------------------------
> ERROR: Calculation halted. Periodic box dimensions have changed too much
> from their initial values.
> Your system density has likely changed by a large amount, probably from
> starting the simulation from a structure a long way from equilibrium.
>
> [Although this error can also occur if the simulation has blown up for
> some reason]
> ...
> ------------------------------------------------
>
> A quick search shows this message related to badly equilibrated systems,
> but my case looks quite stably equilibrated and neither the density nor
> periodic box change much (certainly less than 5%) during the simulations.
> Nothing I see in the output or any other error messages suggest the system
> is blowing up (no vlimit errors or so). I havent looked into the code in a
> long time, does anyone know what exactly has gone wrong that caused pmemd
> to halt?
>
> I use SHAKE, 2fs timestep, Langevon thermostat, berendsen (because of the
> lipid issue with MC) barostat with taup=2 and ntb=2, cutoff = 10.
>
> Nothing else is odd about the system, no extra forces, custom parameters,
> odd chemistry or anything weird, just a pretty large protein in a membrane
> with ca 100k particles total.
>
> Has anybody seen similar cases or any ideas what (if anything) I may be
> doing wrong here?
>
> Since the crashes/exits appear randomly, I assume they will not happen when
> I restart, but this a) costs time, b) needs human monitoring of long runs
> and c) makes me wonder if my settings are not suited for long membrane
> simulations somehow.
>
> Kind Regards,
>
> Thomas
> --
> *Dr. Thomas Steinbrecher*
> Principal Scientist CADD
>
> Roche Pharma Research and Early Development
> Roche Innovation Center Basel
> F. Hoffmann-La Roche Ltd
> Bldg. 092/3.92
> Grenzacherstrasse 124
> 4070 Basel
> Switzerland
>
> Phone +41 61 682 1319
> mailto: thomas.steinbrecher.roche.com
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Mon May 15 2023 - 05:30:02 PDT
