[gmx-users] Energy minimizations with gmx3.2

[Erik Lindahl]

Hi,

On May 17, 2004, at 5:10 PM, Marc Vogt wrote:

> Thanks Erik
>
> The minimization under GROMACS 3.2.1 using l-bfgs converges whether or 
> not
> I use switch interactions.  Any comments on that?

Well, it should converge to _something_ in both cases, but in theory 
the algorithm will be able to do a better work if you have a more 
well-defined hamiltonian. It's kind of hard to compare your energies, 
since they are expected to differ a bit between sharp cut-offs and 
shift/switch.

Anyway that's just the theory. If it works with cutoffs I wouldn't 
worry to much about it.

I might have forgotten to add a reference in the manual, but the L-BFGS 
(L is for low memory) algorithm was invented by Nocedal. Google is your 
friend :-)

Cheers,

Erik