[gmx-users] implicit solvent

[Erik Lindahl]

Marc Baaden wrote:

>feenstra at chem.vu.nl said:
>  
>
>>>Is there any preliminary version already available somewhere that we
>>>could already use for testing purposes? 
>>>      
>>>
>
>feenstra at chem.vu.nl said:
>  
>
>>>We'd like at least to have a look at that. Could we get access to it? 
>>>      
>>>
>
>I would like to strongly back that up (at least concerning the
>implicit solvent part).
>
>  
>
I've started adding implicit solvent stuff to the input files in CVS. It 
is NOT working yet
(still haven't committed the inner loops), but it will get there.


Before you get too excited - let me stress that implicit solvent is 
usually not a good idea.
Due to the complexity of the interactions it is hard to do the innerloop 
in assembly or
to tabulate it, and it really isn't faster than simulating the system 
with water since you
have to perform three iterations over all pairs instead of one. You also 
need very large
cutoffs (2.0-2.5 nm) for calculating the Born radii.

The only thing it is really useful for is energy evaluations for single 
structures, since
we avoid the noise from the water, and perhaps energy minimizations.

This is also the reason why it takes some time - we have finished code 
that calculates the EXACT
Born interactions and surface area, so now we can start trying to 
optimize/approximate it.

Cheers,

Erik