[gmx-users] Problems with the SHAKE-algorithm -> is a problem of energy conservation

[Mark Abraham]

Björn Fischer wrote:
> David van der Spoel wrote:
> 
>> At high temperature you need a shorter time step. Are you sure you 
>> don't want to use T-coupling?
>>
> Thanks for the advice,
> 
> it didn't solve the problem, but it took me one step further to the 
> solution. I should have been more specific about my system:
> 
> Its an argon/nonane (2:1) system at 600 K. The normal timestep is 1.0 
> fs, I also used 0.5 fs and 0.1 fs for testing purposes.
> Only the argon is coupled to the heatbath.
> 
> The Shake-problem is only a symptom, the real problem is that the energy 
> is not conserved (checked via a NVE-run).
> The total energy is conserved for a time (for example about 60000 
> timesteps) , then jumps upward fourfold over about 1000 timesteps.
> After that the SHAKE-problems starts. Making the timestep smaller 
> doesn't solve the problem.
> 
> Do you have any suggestions, why the energy is not conserved (apart from 
> a wrongly implemented force field) ?

You've only coupled a small minority of your atoms in the simulation to 
the heat bath. I can't support my opinion, but that strikes me as asking 
for trouble. I just think the transfer of heat between argon and nonane 
won't be good (enough) to stabilize the system. I think you're right 
about the SHAKE problem being a symptom, though.

Mark