Thanks Rafal for your response.
My timestep is 2 fs. I hadn't thought about it, but I'll try running with 1
fs.
It happens at other lambdas too. I considered the possibility of my system
being out of equilibrium. But if I re-run the same simulation, it runs fine
some of the time ! Also the energy, temperature etc are totally fine when
the error occurs.
I think for now I'll just run a few times and keep the results from the
runs that finish without error.
Best,
Ali
On Mon, Dec 20, 2021 at 11:57 PM Rafał Madaj <rmadaj.cbmm.lodz.pl> wrote:
> Hi Ali,
>
> I encountered this problem when /lambda/ was close to boundary values
> (either 0 or 1).
> As far as I remember, it is due to zero appearing in denominator when
> calculating energy according to formula used in TI. As we have
> stochastic calculations it may appear that it sometimes does not
> encounter this problem.
>
> My workaround for this was just ignoring the faulty calculations and
> after all simulations either removing faulty /lambda/, or manually
> fitting the energy values by interpolation from polynomial regression
> (in excel or something similar).
>
> However, if it fails at other points too, make sure your system is in
> equilibrium.
>
> Is your timestep set to 1 or 2 fs? Including equilibration period.
>
> Regards,
>
> Rafał
>
> On 20.12.2021 22:50, Ali Morshedi wrote:
> > Hi all,
> >
> > When running TI MD simulations using pmemd.cuda, I sometimes get the
> > following error after some steps:
> > Error: an illegal memory access was encountered launching kernel
> > kClearTIKinEnergy_kernel
> >
> > What's strange is that it's not reproducible. Meaning if I run the
> > simulation again, it might finish without the error ! Also energy,
> > temperature etc are all normal at the step that the error occurs.
> >
> > I'd appreciate it if you let me know any information you might have
> > about this.
> >
> > Thanks,
> > Ali
> > _______________________________________________
> > 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
>
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Tue Dec 21 2021 - 00:30:02 PST
