Yes, DAC told you right. With a 12.0A cutoff, you are getting
significantly larger Gaussian spreads than you would have with the typical
settings, which entail an 8.0A cutoff. For a given dsum_tol (which you did
not change), the Gaussian spread is nearly linear in the cutoff. With
12.0A you could even run 64 x 64 x 64 and probably not have a problem. In
fact, the reciprocal space errors arising from that procedure would
probably still be smaller than the direct space errors. My advice would be
to push it to 84 x 84 x 84, if you are using the GPU code, and take a 9A
cutoff with dsum_tol = 5.0e-6. That should run fast and give a pretty good
balance of direct and reciprocal space error.
Dave
On Sun, May 10, 2020 at 9:11 PM Yuliana Bosken <ydavi002.ucr.edu> wrote:
> Thank you!
>
> On Sun, May 10, 2020 at 6:05 PM David A Case <david.case.rutgers.edu>
> wrote:
>
> > On Sun, May 10, 2020, Yuliana Bosken wrote:
> >
> > >I have numerous short MD runs from previous group member which I need to
> > >use for energy calculations. For all runs, the nfft1 and nfft2 values
> are
> > >smaller than my box size. I am not sure if the trajectories would be
> still
> > >accurate for energy calculations or should we discard them.
> > >My box size is as follows:
> > >
> > >Ewald parameters:
> > >verbose = 0, ew_type = 0, nbflag = 1, use_pme = 1
> > >vdwmeth = 1, eedmeth = 1, netfrc = 1
> > >Box X = 82.822 Box Y = 82.668 Box Z = 86.327
> > >Alpha = 90.000 Beta = 90.000 Gamma = 90.000
> > >NFFT1 = 80 NFFT2 = 80 NFFT3 = 120
> > >Cutoff= 12.000 Tol =0.100E-04
> > >Ewald Coefficient = 0.22664
> > >Interpolation order = 4
> >
> > I don't think you should worry much. The usual procedure in Amber is to
> > set dsum_tol (the tolerance in the direct Ewald sum) and the cutoff, and
> > let the program figure out how many grid points to use. This usually
> > leads to grids that have *about* a 1. Ang spacing.
> >
> > Users can also decide to set nfft1,2,3 by hand. This is typically an
> > advanced option, related to a need to get speed in highly parallel
> > situations, to ensure reproducibility across different numbers of MPI
> > threads, or the like. But your grid spacings are all about 1 Ang, which
> > should be fine. There is nothing that says a 1 Ang. spacing is any kind
> > of magical maximum value.
> >
> > ...hope this helps....dac
> >
> >
> > _______________________________________________
> > AMBER mailing list
> > AMBER.ambermd.org
> > http://lists.ambermd.org/mailman/listinfo/amber
> >
>
>
> --
> Yuliana Bosken
> _______________________________________________
> 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 Sun May 10 2020 - 22:00:01 PDT
