[gmx-users] dispersion correction and heterogeneous systems

[Erik Lindahl]

Hi,

On Feb 2, 2007, at 7:37 AM, Michael Shirts wrote:

>
> It can, actually.  The theory that was developed to deal with PME can
> be used for any r^-k potential -- in fact, in the original paper
> (Essman,  J. Chem. Phys., 103, 1995, 8577--8592 -- I think) it's
> derived for a general potential.  However, implementing it, and tuning
> the speed so it wasn't a big hit is a non-trivial task that it's not
> 100% clear is worth it.   There are other workarounds, but they are
> obviously difficult as well.

It can be done for any 1/r^n potential.

The problem has to do with the combination rules. First, any force  
field with "exceptions" from the strict combination rules (e.g.  
GROMOS96) is a no-go. Second, the standard combination rules in OPLS/ 
Amber will lead to six separate fourier transforms, convolutions, and  
inverse transforms, so it is _way_ more expensive than standard PME.


Cheers,

Erik