[gmx-users] restarting jobs using tpbconv and grompp : with regards to gen_vel

[Mark Abraham]

himanshu khandelia wrote:
> 
>     Check out http://wiki.gromacs.org/index.php/Doing_Restarts
>     <http://wiki.gromacs.org/index.php/Doing_Restarts>
> 
> 
> Thanks very much.  I do understand the general restarting mechanism. It 
> is only the first restart that was a little bit confusing
> 
>      >
>      > I just wanted to confirm if what I am doing makes sense ? Thanks very
>      > much for the help !
> 
>     These two output .tpr files should be the same except for the initial
>     steps and times, unless your mol-2.mdp file took care of that. You can
>     use gmxcheck to test for the differences.
> 
> 
> Sorry I did not understand ? There is no mol-2.mdp file.

If you achieved this:

> In order to circumvent the problem, and avoid deshuffling and reshuffling, I wrote another .tpr file for the first dynamics run but with gen_vel set to no.

then you must have a second .mdp file.

> So, in effect 
> the first tpbconf command should work. The velocities will be obtained 
> from the .trr file and not the .tpr file, correct ? That is the main 
> issue bothering me, because the .tpr file contains the initial 
> velocities generated during the first run.
> 
> tpbconv -f mol-1.trr -s mol-1.tpr -e mol-1.edr -extend 200 -o mol-2.tpr

Read tpbconv -h... tpbconv writes a *continuation* run input file, not 
some crazy hybrid with final positions and initial velocities.

MD simulations have a whole bunch of parameters that describe the 
algorithm, which don't vary with time, and a bunch of state variables 
(like positions and velocities) which do. tpbconv takes the former from 
the old .tpr, modifies them according to the command line options, and 
tkaes the latter from the state information you provide in the .edr and 
.trr files. grompp does the same kind of thing, getting the parameters 
from the .mdp file, and the state variables from input .gro/.edr/.trr 
files, unless overridden by gen_vel then and there.

Mark