[gmx-users] un-coupled simulations and traj recovery

[Michel Cuendet]

David van der Spoel wrote:

>On Wed, 2003-05-14 at 11:08, Anton Feenstra wrote:
>
>>Giorgos Karvounis wrote:
>>
>>>If we wanted to save every possible step, we would need a really huge
>>>disk.
>>>
>>Yes - if you really need every step, we're taling Tera-Bytes.
>>
>>>...Until we decide to try  hacking the code (...again...).
>>>
>>Which parameters exactly would you need in addition to the ones already
>>in the output files (i.e. .trr and .edr)?
>>
>I think the point is that restarting is not 100 % compatible due to T
>coupling. We would therefore have to hack the code such that restarting
>will be 100 % identical even with T and P coupling.  We will have to
>store additional information about velocities at previous time steps.
>Since we are planning to make some kind of "checkpoint" file, we should
>take care that it is correct then. I can't promise anything right now
>though...
>
Maybe you just have to store the nose-Hoover momentum at the last step 
(one real number). In principle, a state of the system is entirely 
defined by one set of positions and velocities (one point in the phase 
space), so saving velocities at previous time steps is irrelevent. The 
Nose Hoover equations actually operate in an extended phase space 
composed of the pysical variables plus the Nose-Hoover momentum. I think 
a GROMACS re-run always resets the thermostat to zero, which is enough 
to produce entirely different dynamics (correct me if I'm wrong).

Cheers,

Michel