thanks. I also was thinking of the time step, but it is hard to test since
it seems to take tens of ns to start seeing the problems. We will start some
tests on that. Never a problem with TIP3P doing the same things. Might be
the off center charge?
carlos
On 4/16/07, David A. Case <case.scripps.edu> wrote:
>
> On Mon, Apr 16, 2007, Carlos Simmerling wrote:
>
> > we're seeing a lot of non-fatal vlim errors with tip4p that we don't get
> in
> > comparable
> > sander simulations with tip3p. These are at 320K, NPT, 1atm, 2fs time
> step,
> > truncated oct PBC.
> > has anyone else seen this?
>
> My experience has been to see no vlimit errors with tip4p (using the
> default
> vlimit of 20), but with different conditions:
>
> 300K (not 320), NVE (not NPT), 1 fs time step (not 2fs).
>
> This is over more than 200 nsec of two or three small proteins.
>
> I suspect that the principal difference is the time step, since a larger
> step size would be more likely to lead to some bad forces that violate the
> vlimit upper bound.
>
> [Not sure if this helps very much, but it *is* one more data point. If
> you
> are getting "a lot" of such errors, it might not take long to see if they
> go
> away with the shorter time step.]
>
> ...dac
>
>
Received on Wed Apr 18 2007 - 06:07:21 PDT
