From: Jim Phillips (jim_at_ks.uiuc.edu)
Date: Thu Mar 01 2012 - 14:47:48 CST
Hi all,
I should have some code to separate water and non-water temperatures
checked in tonight.
I still need to run more tests, but my current line of thought is that
with "langevinHydrogen off" and "rigidBonds water" the protein has a lower
level of temperature coupling per degree of freedom than the water, the
protein will conserve energy worse than the water because it has
high-frequency bonds to hydrogens, and the temperature coupling is turned
off specifically on those high-frequency bonds most likely to heat.
That said, the reported temperature difference seems unlikely, as the
above explanation would suggest that "langevinHydrogen off" without rigid
bonds should produce a similarly high temperature for the whole system.
I don't see how "rigidBonds water" would cause the protein to heat more.
Again, this is all speculation. Tests are running.
-Jim
On Tue, 28 Feb 2012, Aron Broom wrote:
> Dear NAMD users,
>
> I've looked around on the mailing list and couldn't find a clear answer to
> as to whether or not it is supposed to be ok to use rigidBonds water (say
> with TIP3P) but then leave the solute (say a protein) flexible with a 1 fs
> timestep and langevinHydrogen off in order to save on calculating
> collisions with the large number of rigid hydrogens in the sample. It
> might seem like one should not use langevinHydrogen off as long as there
> are any flexible hydrogens, but several sources, including an NAMD tutorial
> (
> http://www.ks.uiuc.edu/Training/Tutorials/science/channel/channel-tutorial-files/ABF/win1/win1A.conf)
> have exactly this setup.
>
> Well, I believe I have confirmed that you should NOT do this. In looking
> at a 60ns simulation of ~2000 protein atoms in an ~100,000 water atom box,
> and determining the temperature of the non-rigid parts of the system by
> calculating their kinetic energies (KE=0.5mv^2), binning them, and then
> fitting to the boltzmann distribution ( y = (2/sqrt(Pi * (KbT)^3 )) *
> sqrt(x) * exp(-x/KbT) ), I find that the temperature of the whole system is
> fine (set to 298 K, and comes out on average at 297 K), but the temperature
> of the protein alone is ~20-25 K higher than the surrounding solvent
> throughout the simulation.
>
> This is a big problem, especially for studying ligand binding, and I
> imagine for anything having a 20-25 K difference across the solvent-solute
> interface is bad, not to mention just being incorrect.
>
> I tested running an extra 4ns from the end of my 60 ns, with two changes in
> NAMD parameters:
>
> 1) just turn langevinHydrogen on, which comes with a 5% performance
> penalty, and
>
> 2) leave langevinHydrogen off, but change rigidBonds water to rigidBonds
> all, and change from a 1 fs timestep to a 2 fs timestep, in which you gain
> ~30% performance, but some people are sceptical of the results.
>
> Both of these changes alone are capable of bringing the system back to a
> reasonable configuration in which the protein and solvent have
> approximately the same temperatures (the protein was still slightly hotter
> (~5 K, but perhaps a gamma of 1/ps is not fast enough to see complete
> equilibrium in 4ns).
>
> Has anyone else seen this and can confirm that the setup of using
> ridigBonds water and langevinHydrogen off are not to be used together when
> you have a solute that has hydrogens, despite this being somewhat common
> practice? If anyone wants to test the temperatures of various parts of
> their systems from existing *.vel files, I just followed the instructions
> on this page (
> http://www.ks.uiuc.edu/Training/Tutorials/namd/namd-tutorial-unix-html/node13.html),
> but note that the units in the .vel files must changed since this was
> written, so you'll have to convert amu to kg, and angstrom/ps to m/s, and
> then use the correct boltzmann constant for those units.
>
> ~Aron
>
>
> --
> Aron Broom M.Sc
> PhD Student
> Department of Chemistry
> University of Waterloo
>
This archive was generated by hypermail 2.1.6 : Mon Dec 31 2012 - 23:21:16 CST