[gmx-users] Precision of trr and checkpoint files

[Kevin C Chan]

Dear Users,

I am wondering how precision of the files affects the continuation of
simulations. I have compiled both single and double precision of Gromacs
5.0.6 but I used to work with the double precision one.

Usually I continue a simulation from a equilibration run (probably with
Berendsen schemes) to a production run (with parrinello-rahman). So I have
to use grommp instead of a tpbconv or -cpi option. I used to specify -c, -p
and that's all. I have checked my gro file and it contains corrdinates,
velocities and box size so it should be fine. However when I learnt from
the command manual and used -t previous.trr to feed the corrdinates,
velocities and box size, grommp just asked me for a conf.gro. Shouldn't a
trr file contains all these things?

Okay I still can go on with -c, -p plus -t (Though it is kind of
duplication for me), but this time the precision of the files is not
understandable. When I do not use -t, everything seems perfectly fine.
When I use -t previous.cpt, it gives

Reading Coordinates, Velocities and Box size from old trajectory
Will read whole trajectory
Precision mismatch for state entry box, code precision is double, file
precision is float
Precision mismatch for state entry box-rel, code precision is double, file
precision is float
Precision mismatch for state entry pres_prev, code precision is double,
file precision is float
Precision mismatch for state entry x, code precision is double, file
precision is float
Precision mismatch for state entry v, code precision is double, file
precision is float
Last frame         -1 time 1000.000
Using frame at t = 1000 ps
Starting time for run is 0 ps

When I use -t previous.trr, it gives

Reading Coordinates, Velocities and Box size from old trajectory
Will read whole trajectory
trn version: GMX_trn_file (single precision)
Last frame         50 time 1000.000
Using frame at t = 1000 ps
Starting time for run is 0 ps

I think both screen-out means neither the trajectory nor the cpt file is in
double precision, given that I have been used a double-precision Gromacs
nearly all the time. I have gone through the online manual and some mailing
threads but little has straightly answered my question.

Also although it is stated on the manual (
http://www.gromacs.org/Documentation/Terminology/Precision?highlight=precision)
and FAQ that single (mixed) precision of Gromacs is enough for production,
can anyone elaborate a little bit more on the accuracy of this "mixed"
precision to (usually very large) biological system simulations. As guys in
our group used to run with (at least namely) double precision codes but the
GPU speed-up was too attractive and Gromacs only provides single precision
compiles for that.

Thanks in advance,
Kevin
City University of Hong Kong