[gmx-users] epsilon r

[Erik Lindahl]

jozef hritz wrote:

Hi Jozef,

> I have probably naive question. Very often we simulate protein in big box
> of explicit water molecules with epsilon r=1. It would be correct if we
> have polarizability in force field, but we haven't. Are there some
> correction from this point of view in charges of water molecule. If not
> isn't it better to use higher value of epsilon r? The question is what
> value, when for water it's 78 and for proteins around 4. Is it possible to
> put different epsilon r into different regions (e.g. inside and outside of
> protein)? 

No - even if most forcefields don't include atomic polarizabilities 
there is still molecular polarizability, i.e. the water molecules can 
screen charges by their orientation. The effective dielectricity 
coefficient isn't 1 for water models like SPC or TIP3 but rather in the 
range 60-80. (TIP5P has 81 or something).

Unless the forcefield was explicitly designed for a different epsilon 
you shouldn't use anything but 1.0.


> 
> Next thing is concerning with implicit solvent. In mailing list I red that
> Erik had problems with effective calculation of protein surface so there
> was problem with dynamics (how does it look now?). Is it possible to
> implement implicit solvent for single point calculations?  

The problem is getting the exact derivatives of the surface area when 
atom spheres intersect. We have code that does this now based on 
Delaynay triangulation (it just has to be implemented), and Vijay 
Pande's group at stanford are implementing a simpler model (perturbation 
-based, less exact) that they are programming in Gromacs.

It should appear in 4.0, I think....


Cheers,

Erik