[gmx-developers] hacking the update algorithm

[Giorgos Karvounis]

Dear developers,
i would like to hack the update algorithm in order to reproduce the
exactly the same run  "b"  from an original run "a" but starting from a
random frame of the original run (i.e. not from the first step). If my
original run was without T and P coupling , i can always reproduce it,
at any time frame. First of all some general questions:
1) From previous answers , i get the impression that this is happening
due to T coupling only (although P coupling rescales co-ordinates and
box-size!). I tried to reproduce a run without T but with P coupling and
i couldnt do it ( gen_vel=no, unconstrainstart=yes e.t.c), maybe need to
change something else at the mdp file?
2) When i take a random frame from my original run using the same
velocities v(t) (grompp -t .trr   -time...), i have the velocities at
v(t). And this is an average of v(t-dt/2) and v(t+dt/2). I was wondering
if there is a problem there (as David suggested already that i need to
store the velocities of v(t) and v(t-dt))  because i assume that the
scaling factor will use the v(t-dt/2) to get the Ekin(t-dt/2) and from
there to get the T  at t-dt/2 ...or maybe not? Is this the reason why we
cannot reproduce the same run with T coupling?
3)I was also wondering, if i put a switch to the code , and turn the
coupling (T and/or P) off at the frames that i want (eg every 20000
steps) will this do the trick?
4) At the end of the day, is it worth to get to all this trouble instead
of making  a short 'equilibration' run and then continue the rest of my
simulation (2ns) without T/P coupling (some kind of NVE simulation)? So
far i used weak coupling and i dont know if it would be reasonable to do
without.
If you are not already dizzy, and if any of this makes sense, thanks!
:-)