[gmx-users] The Cut-off for coulombtype heat up the water system?

Mark Abraham Mark.Abraham at anu.edu.au
Sat Jun 20 06:55:39 CEST 2009

Florian Dommert wrote:
> * Mark Abraham <Mark.Abraham at anu.edu.au> [2009-06-20 11:54:46 +1000]:
>>> When I understood the idea of the reaction field correctly, I treat the
>>> electrostatic forces with a cutoff and relative dielectric permittivity
>>> != 1. With the mentionend Ewald methods I should be able to reproduce
>>> exactly the same circumstances like in a reaction-field setup. So at the
>>> moment I can imagine just one critical point, when using SPME/PME/PPPM
>>> or an Ewald sum is the big set of parameters that have to adapted in
>>> order to obtain an appropriate accuracy of the forces. In the reaction
>>> field method you just have two parameters: the cutoff and epsilon_r. The
>>> other algorithms require addtionally require the input of an appropriate
>>> size for used grid in Fourier space and in case of SPME/PME/PPPM also an
>>> interpolation order. Finally you need to set the splitting paramter
>>> correctly, otherwise you will obtain unaccurate forces. So there can be
>>> a very large error introduced, when applying the wrong parameters to the
>>> Ewald methods. The heat up of the water is also just related to 
>>> extremly inaccurate
>>> electrostatic forces, since with PBC an "infinite" system is 
>>> simulated  and just a very small amount of the electrostatic 
>>> interaction that is of
>>> long range nature is calculated. Therefore an large error is not  
>>> unexpected.
>>> Finally the only restriction of Ewald I see is the requirement of PBC,
>>> where I can reach any level of accuracy for the electrostatic force
>>> given by certain charge distribution, don't I ?
>> I really haven't understood you, sorry.
> I think that I a complete wrong idea of an simulation using a Reaction
> field, so I have to get a correct picture. Because when investigating a
> protein you require a physiological environment with corresponding ions
> to provide a certain pH value. Is this finally all contained in the
> force field parameters ?

In principle, yes, however not even in theory is this true for the 
commonly-used force fields. Typically they were parameterized to 
reproduce a range of experimental or quantum-chemical data, but the 
scale of this parameterization problem was large enough that considering 
solvents of non-pure water would have been too much (even if data was 
available). One might demonstrate post-factum that a force field does a 
reasonable job in such a case. One might also demonstrate that a force 
field does a reasonable job under a different electrostatic treatment.

> This would make things clear and enlight my
> foggy insight in this special way to treat electrostatic forces.
> Furthermore I assume no periodic boundary conditions are used then ?

One's electrostatic model need not be confounded with the boundary 
conditions of the simulation. For Ewald-family methods, PBC is required, 
introducing the potential for periodicity artefacts. For other methods 
(cut-off, fast multipole and variants) one has the option of choosing a 
different boundary condition (e.g. non-periodic (RF) vacuum containing a 
restrained spherical shell of water around free water, or a large 
protein complex in vacuo) and suffering artefects from those boundary 
conditions, rather than perhaps periodicity-induced ones.

In particular for RF, the assumption of homogeneity would suggest not 
using PBC. With enough solvent, in practice that assumption would be 
approximately true even under PBC.

> You just simulate a protein/polymer/molecule and assume that it is
> surrounded by a medium with a certain epsilon_r.

Sure, but the RF model as applied to each particle does not depend 
strongly on whether the system is periodic if the system has enough 
solvent per image.


More information about the gromacs.org_gmx-users mailing list