[gmx-users] free energy with TIP4P bug fixed

[chris.neale at utoronto.ca]

Berk, David, and other developers,

I believe that all the relevant information has already been posted,  
but since I use TIP4P ubiquitously, I would like to get some  
clarification regarding simulations that may be affected. I currently  
understand that:

1. Utilization of TIP4P in the absence of the free_energy code is  
currently believed to be without error in gromacs 3.3 through 4.0.3
2. Combining TIP4P and the free-energy code may lead to the errors noted.
3. Combining TIP4P and the free-energy code with at least 3 energy  
groups (TIP4P, perturbed atoms, rest of system) will not lead to the  
errors noted. This would only require 2 energy groups if the system  
was composed of only TIP4P and the perturbed atoms.

It would be very useful for me if somebody could validate the  
statements above or state any incorrections in those statements.

Thank you,
Chris.





-- original message --

All,

I just wanted to pass along the below e-mail, previously sent to the
mailing list, and highlight the potential implications.

If I understand properly, there was a bug in all Gromacs versions
beginning prior to 3.3 (probably not 3.1.4) and through 4.0.3 that
affected the TIP4P and TIP4P-Ew water models, *unless* you were using
separate energy groups for water and the rest of your system. Berk
notes below that the bug "would cause a few tip4p-tip4p charge
interactions to be missing". I want to emphasize that this is
potentially an extremely serious problem.

In my own testing I found that with this bug, densities for TIP4P-Ew
still came out close to the literature values, but it could result in
substantial errors in other properties. For example, in computing
hydration free energies with acetamide, I found my hydration free
energies changed by more than 2 kcal/mol when I changed rcoulomb by
0.1 nm, even when using PME and long-ish real-space cutoffs, which
should not be the case.

Secondarily, I found that my computed hydration free energies with
either set of cutoffs disagreed with the "correct" values (as obtained
by other simulation packages) by 0.7 to 3 kcal/mol depending on my
choice of real-space cutoff. In contrast, with other water models, I
could reproduce the expected values to within my computed
uncertainties (+/-0.05 to 0.1 kcal/mol).

I am still doing some testing, but I anticipate that I will ultimately
have to submit errata for two different papers I've published in which
I used the TIP4P-Ew water model in combination with GROMACS. I urge
you to check your own studies that have used TIP4P/TIP4P-Ew and see if
they are affected.

David Mobley
Assistant Professor
Department of Chemistry
University of New Orleans