Hi Wei,
Unfortunately not. The virial is NOT calculated for NVT (or NVE)
simulations in pmemd or pmemd.cuda (forget the pmemd.cuda.MPI I really
don't understand why anybody wants to complicate their life by running in
parallel across multiple GPUs for only a 20% speedup at best). Sander
computes it because sander is slow and not very well optimized. Your
options are:
1) Use sander.
2) Run with NPT.
3) Run with NPT and set the pressure relaxation constant VERY long (taup).
4) Modify the fortran code and run with pmemd.
5) Modify the fortran code AND the cuda code and run with pmemd.cuda
6) If you are insane do 5 but also modify all the mpi stuff for
pmemd.cuda.MPI.
Probably 3 is your easiest option if you aren't worried about getting NVT
performance.
Note, there will be a patch coming very soon that tweaks the constant
pressure algorithm slightly for the GPU code so for specifically
calculating surface tensions you might want to stick with the CPU code
until that patch is released since such measurements will be very
sensitive to the virial.
All the best
Ross
On 4/17/13 4:42 PM, "Wei Lin" <w5lin.ucsd.edu> wrote:
>Hi All,
>
>
>I am trying to compute the surface tension of an interface within NVT
>ensemble by running pmemd.cuda.MPI.
>The formula for the surface tension, gamma, is following:
>
>gamma=(hz/2)[<Pzz>-0.5(<Pxx>+<Pyy>)]
>
>Pzz is the normal pressure tensor component and Pxx, Pyy are the lateral
>pressure tensor components.
>
>The output of the virial(x,y,z) and pressure(x,y,z) in mden are all zero
>for NVT ensemble. I tried to print out the virials in mdout file by set
>verbose =3 in &ewald namelist. It works for sander.MPI engine, but it
>doesn't print any virial informations with pmemd.cuda.MPI engine. I have
>tried to print virial or pressure data through modifying runmd.F90 and
>pme_force.F90 in $AMBERHOME/src/pmemd/src. But the outputs are also all
>zero. I am wondering is it possible to obtained either pressure tensor
>components or virials directly from the pmemd.cuda.MPI simulation?
>
>Thanks,
>
>Wei
>_______________________________________________
>AMBER mailing list
>AMBER.ambermd.org
>http://lists.ambermd.org/mailman/listinfo/amber
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed Apr 17 2013 - 17:00:04 PDT
