[gmx-users] Re: Temperature coupling problem: reference T is different to given T in mdp file

[Justin Lemkul]

On 1/22/13 9:27 AM, Andrey Frolov wrote:
> Dear Justin,
>
>> Considering all data points will include early frames that are not
>> necessarily
>> anywhere close to being equilibrated.  You could try block averaging to see
> if
>> the results converge to the correct result over the latter part of the
>> simulation, or at least see if the system is approaching the target
> temperature.
>
> Temperature jumps to the "wrong" target temperature at the beginning of the
> simulation (see the figure attached). So, this is not the reason of the
> problem.
>
> <http://gromacs.5086.n6.nabble.com/file/n5004788/v-rescale_temperature_evolution.png>
>
>
>> It would be better to diagnose the system using PME.  Even if the results
>> are
>> "similar," rounding errors are compounded when using plain cutoffs, which
>> render
>> the results inherently unreliable.
>
> This is true. But, the problem persists even with PME. I have simulations
> with slightly different setup: more CO2 molecules and higher CO2
> concentration, umbrella windows for different separations between
> paracetamol molecules, 24 ns simulation time, with PME electrostatics. Here
> it gives me the "wrong" average temperature 478.486 K, while i set v-rescale
> thermostat to 423.15 K. So, electrostatics method seems to play no role in
> this.
>
>
> Moreover, i tested the same simulation setups as in the initial posts with
> Nose-Hoover thermostat. Single Nose-Hoover with number of chains = 1 gives
> correct temperatures (the same as a set in mdp file). Even if i set three
> thermostats to three subsystems (eventhough, i agree that this approach is
> valid only for testing purposes) the resulting temperatures are correct.
>
> But with N chains = 10, the resulting temperatures are again wrong (up to ~
> 20 K). (In this case integrator was md-vv).
>

As additional tests, can you set dt = 0.001 with the current setup, and then 
also try constraining all bonds rather than just hbonds (with dt = 0.002)?

-Justin

-- 
========================================

Justin A. Lemkul, Ph.D.
Research Scientist
Department of Biochemistry
Virginia Tech
Blacksburg, VA
jalemkul[at]vt.edu | (540) 231-9080
http://www.bevanlab.biochem.vt.edu/Pages/Personal/justin

========================================