Peter,
I didn't see anything in your inputs that looked unusual.
I have run several proteins and peptides with langevin dynamics
out to many nanoseconds with nothing unusual, but I don't
restart often. I think Dave Case's point about the repeat of the
sequence of random forces sounds very much like it could be
responsible. Maybe that's why my runs without restart are fine.
Carlos
Peter Varnai wrote:
>Dear Amber users,
>
>I continue to look into the problem of quasi-periodic behaviour in
>Langevin simulations in Amber8. I created a small system to test different
>input parameters and I could reproduce the following behaviour in a
>consistent manner. I used the following input:
>
> &cntrl
> ntx=5, ntpr=500, ntwr=500, ntwx=500, ntwe=500,
> nstlim=500000, nscm=1000, ntf=2, ntc=2, ntb=0, ntp=0,
> irest=1, ntt=3, temp0=300, gamma_ln=5, cut=99.0, rgbmax=99, dt=0.002,
> ntr=0, saltcon=0.1, gbsa=1, surften=0.005, igb=1 &end
>
>The result shows constant and large translational movement of the
>molecule. This was checked visually and by non-fitted rmsd along the
>merged trajectory. Apparently, the removal of trans/rot at default 2 ps
>cannot cope with it. The "KE Trans" values reported in the outputs are
>identical (eg: between run1.out run10.out), but the "KE Rot" shows
>quasi-periodic behaviour along trajectory too. This effect is clearly
>related to the restarting frequency of the simulation. Although increasing
>the trans/rot removal frequency to 2 fs (every step) stops the overall
>translation (so it works!), the molecule experiences small but identical
>COM trans/rot energies between different fragments of the trajectory.
>Quasi-periodicity is also seen in E_kin or rmsd values along the
>trajectory. That the problem is truly related to Langevin was checked by
>setting gamma_ln=0 and no problem was observed.
>
>Therefore, it seems to me that something is either wrong with the
>writing/reading of the restart file and/or each time sander restarts it
>assigns some constant, non-random forces to the system?!
>
>Before looking into the code, I wonder if someone has other idea what is
>wrong? If anyone needs I can send a test case or the analysis results
>described above. Lastly, has anyone carried out long and reasonable
>Langevin simulation with Amber 8?
>
>Thanks,
>Peter Varnai
>-----------------------------------------------------------------------
>The AMBER Mail Reflector
>To post, send mail to amber.scripps.edu
>To unsubscribe, send "unsubscribe amber" to majordomo.scripps.edu
>
>
-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber.scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo.scripps.edu
Received on Fri Apr 08 2005 - 20:53:00 PDT
