RE: Solvent model other than TIP3 for use with CHARMM FF in NAMD??

From: Blake Charlebois (
Date: Tue Jun 21 2005 - 15:41:19 CDT

Here is a quote from Alexander D. MacKerell (the PI in charge of the CHARMM
force field, if I am not mistaken) from page 22 of A.D. MacKerell Jr.,
"Atomistic Models and Force Fields," in Computational biochemistry and
biophysics, O.M. Becker Ed. 2001, pp. 7-38.

"The SPC/E water model is known to yield better pure solvent properties than
the TIP3P model; however, this has been achieved by overestimating the
water-dimer interaction energy (i.e., the solvent-solvent interactions are
too favorable). Although this overestimation is justifiable considering the
omission of explicit electronic polarizability from the force field, it will
cause problems when trying to produce a balanced force field due to the need
to overestimate the solute-solvent and solute-solute interaction energies in
a compensatory fashion. Owing to this limitation, the TIP3P model is
suggested to be a better choice for the development of a balanced force
field. It is expected that water models that include electronic polarization
will allow for better pure solvent properties while having the proper
solvent-solvent interactions to allow for the development of balanced force
fields. It is important when applying a force field to use the water model
for which that particular force field was developed and tested. Furthermore,
extensions of the selected force field must maintain compatibility with the
originally selected water model."

I interpret this to mean that changing the water model is dangerous unless
you really know what you are doing.

-----Original Message-----
From: [] On Behalf
Of Marc Q. Ma
Sent: June 21, 2005 3:12 PM
To: Leonardo Sepulveda Durán
Cc:; namd-l
Subject: Re: namd-l: Solvent model other than TIP3 for use with CHARMM FF in

Dear all,

My understanding was that TIP3P rigid water model is just one of many
existing force fields for liquid water. FF for liquid water is
independent of any FF for proteins and DNAs and lipids and ... It just
happened that TIP3P is the most popular one to go with CHARMM for
solvated biomolecule simulations. And, probably a preference, a
shortcut, or a "mistake" by NAMD developers not to leave water models
as a choice. For example, when you specify rigid waters, NAMD assumes
TIP3P model automatically and apply SHAKE or RATTLE iterations to make
the 3 point water molecules rigid.

Computing self diffusion coefficients can be very tricky in using any
FF for liquid water. You got to use block averaging, and weather you
choose overlapping or non-overlapping blocks, block sizes, and the
region that you "think" is the "true" linear region according to
Einstein's equation all contribute to the accuracy/inaccuracy of
computation of this "observable." Arguably true, if you use small time
steps such as 1fs, and keep all long range forces, you should get the
best estimates of self diffusion coefficient as long as you run long
enough simulations, or use hybrid Monte Carlo to improve the sampling
efficiency and egodicity of simulations.

My experience with computing RDF for TIP3P water model is that the
peaks and valleys are all clearly shown, and the 1st and 2nd peaks (1.9
aa and 3.3 aa) for O-H RDF indicate the 1st and 2nd strongest H-bond
distance. From the same 400ps simulations, the self diff coeff is
estimated to be 3.69+/0.01 (10^-5 cm^2 s) for 300 K temperature. I
believe the experimental value is around 2.5.

I hope one day NAMD would have full support of many different water
models, including the various implicit solvent models. That will leave
serious developers lots of option to do verification and discovery.
Right now, researchers have to go from software to software to take
advantage of the best features of each one.

On Jun 21, 2005, at 1:42 PM, Leonardo Sepulveda Durán wrote:

> Yesterday I read the 1997 F3C Levitt's paper. In table 3 a comparison
> to other published 3 point models. TIP3P is far one of the worst
> models, it have a Diffusion coefficient double than experimental, and
> the rdf don't show the two long range peaks, suggesting the lack of
> water tetrahedral structure. In the other hand, SPC/E (or even SPC)
> shows features very similar to F3C.
> In the paper they argue consitency in molecular representation is an
> important issue, so if Angle and bond degrees of freedom are allowed
> in solute, it must be in water too. So as CHARMM and ENCAD have fairly
> the same type of armonic potentials, and in F3C parametrization the
> only ENCAD related issues they use were different cutoff methods, F3C
> would be independent of ForceField and its implementation would depend
> on cutoff or algoritmic issues (as long as the FF would be like AMBER,
> If CHARMM forcefield was parametrized with TIP3P and that is important
> to the solute parameters, then may be some problems...but I hope
> Brian's research could succeesfully solve them to have a broader
> variety of solvents for NAMD/CHARMM use.
> Leonardo

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:40:52 CST