[gmx-users] New force fields

Erik Lindahl lindahl at stanford.edu
Fri May 3 18:59:47 CEST 2002


>Although not needed for supporting other force-fields, one feature that
>I find  essential is the capability of using different potentials for
>different atom types. At present all the atom types must use the same
>potential, albeit with different parameters.
Yes, we should definitely add this as a formatting option to the force 
fields, but it will be much
more difficult to actually implement it. The main problem is what we 
should do with the cross
terms; If atom A uses LJ and atom B Buckingham, what do we do with the 
A-B interaction?

>In addition, the procedure for adding a custom potential (that is, with
>a different formula) should be documented (as opposed to adding a potential
>defined by a table, that is well documented). More that two parameters
>should be allowed.
I've been thinking of that; It's not too hard to document it, although 
some of the code in the
innerloop generator could use a little cleaning-up. One problem is of 
course that your code
would only be C or Fortran (never SSE/SSE2/3DNow or Altivec assembly), 
so you would lose
a significant amout of speed (say ~40% of your performance). Since the 
table loops are
written in assembly they will always be faster when you can use them.

Oh well, maybe it's time to go out and buy a book and make a real 
primitive compiler
out of the innerloop generator :-)



More information about the gromacs.org_gmx-users mailing list