[gmx-developers] Shake question?

[David van der Spoel]

On 6/21/10 8:57 AM, Shirts, Michael (mrs5pt) wrote:
> Hi, all-
>
> A couple questions about the shake implementation.
>
> Right now, it appears that shake tolerance (set as shake_tol) is the
> tolerance on the relative distance from the constraint squared -- from the
> manual, one might guess that shake_tol is the maximum relative deviation of
> a constrained bond.  I think expressing an the parameter in terms of
> distances instead of distances**2 would make more sense.
>
> Also, currently in the cshake code, the constraint is considered satisfied
> if iconv = 0, where:
>
> iconv   = fabs(diff)*tt[ll];
>
> Diff is the difference between the square bond length and the square desired
> bond length, and tt[ll] currently is 1/(2*tol) -- rather than the tol*tol
> which I think would be a bit more clear where tol is equal to shake_tol.
>
> If tt is too big, then its possible that there's a 1 in about 4 billion
> chance that iconv will be zero even if the tolerance isn't satisfied.
> Actually, this appears to be the case even with the current code - if
> ir->shake_tol is less than 10^-13, then you'd start seeing the looping
> around the integers in a long int.
>
> I'd propose changing the definition of 1/(2*tol) to 1/(tol*tol), and to
> avoid the problem of integer wrapping by capping shake_tol internally in
> readir.c , so that  the hardware limit for "long int" isn't violated.
>
> I can go ahead and make these changes, but I wanted to ask around and verify
> there wasn't something I was missing!
>

The routine that is being used is vec_shake which says:

     tol2=2.0*tol;
     tt[ll] = 1.0/(toler*tol2);



> Best,
> ~~~~~~~~~~~~
> Michael Shirts
> Assistant Professor
> Department of Chemical Engineering
> University of Virginia
> michael.shirts at virginia.edu
> (434)-243-1821
>


-- 
David van der Spoel, Ph.D., Professor of Biology
Dept. of Cell & Molec. Biol., Uppsala University.
Box 596, 75124 Uppsala, Sweden. Phone:	+46184714205. Fax: +4618511755.
spoel at xray.bmc.uu.se	spoel at gromacs.org   http://folding.bmc.uu.se