[gmx-users] question about .trr file

[Mark Abraham]

Anna Marabotti wrote:
> Dear all,
> at present, when I'm doing a simulation (with Gromacs 3.3.3), I'm keeping both .trr and .xtc files, the first
> one in case of a continuation of the simulation, and the second one for analysis. As I saw in the manual and
> in the gmx-users list, the .trr file is essential for an EXACT restart of the trajectory since the program
> tpbconv reads coordinates and velocities in full precision, and the file .xtc does not contains velocities,
> and coordinates are in reduced precision. But file .trr is also a HUGE file when trajectories are very long.

Sure, so don't store information that you don't need. One can choose the 
frequency with which various information is written. See the manual and 
http://wiki.gromacs.org/index.php/Doing_Restarts

> So I'm asking: how many steps are needed to correctly restart a simulation? 

The last step that has complete information - positions, velocities and 
(if needed) energies.

> For example, let's say I have to
> do 100 ns simulation. I'm obtaining my huge 1st.trr file. Then, I continue the simulation for 1 ns, and I will
> obtain the 2nd.trr file (smaller than 1st.trr, I suppose).

If each is using the same output frequencies, they will have an amount 
of data in them roughly proportional to the length.

> Then, if I want to continue for other 100 ns
> simulation, can I use the 2nd.trr file with tpbconv to produce the new .tpr file, and will the final result be
> the same as if I use the 1st.trr file with tpbconv and continue the simulation for other 101 ns?

Yes, if your machine and executable don't change.

> If the answer
> if yes, I could store only the 2nd.trr file and delete the 1st.trr file, with a lot of space savings. Am I
> correct? 

Perhaps. In any case, the thing to do is to either plan to store only 
what you need (using the .mdp options), or to reduce this quantity later 
with trjconv. Don't mess around with little simulations to do the same 
job worse.

Mark