[gmx-users] strange temperature coupling problem

[Mark Abraham]

Nickle Fan wrote:
> 
>     3000ps is an extremely large coupling constant to want to use.
> 
>     In any case, "Simple" is wrong about the frequency of coupling. This
>     coupling constant is a parameter in an algorithm that is applied each
>     step that expresses how well-connected the group is to the thermal
>     reservoir. It is not a time interval between applications of the
>     algorithm.
> 
>     I suggest you both have a read of section 3.4.5 of the gromacs manual,
>     where it will tell you that "The Berendsen algorithm is stable up to
>     Delta T approx tau_t". Doubtless 3000ps is much larger than your
>     simulation timestep, and the instability is not surprising!
> 
>     Mark
> 
> Thanks for your kind response, but I do not quite agree with your point. 
> I have used tau_t up to 2ps in my other simulations, and they works 
> fine. In fact, from version 3.1, it is allowed to not couple certain 
> groups. I tried that too, and the temperature of water still rises like 
> crazy.
> 
> The manual say that the algorithm is stable upto \DeltaT = tau_t, and I 
> intepret that as "we can use tau_t as small as \Delta_t", i.e., it 
> actually pose a lower limit for the coupling constant, rather than an 
> upper limit. Correct if I am wrong.

True, it can certainly be read that way, and perhaps it should be. After 
all, the normal values for tau_t of 0.1-1ps are larger than normal 
timesteps of a few femtoseconds...

Does someone actually know what they're talking about? :-)

Mark