This seems to have nothing to do with shifted potentials or gpu. I run a tripeptide double precisiion NVE simulation on a machine with xeon cpu and without gpu. kill it, then continue it with the cpt file generated. That energy jump still exists.
On Tue, Nov 11, 2014 at 1:39 PM, Johnny Lu <[email protected]> wrote: > Going from non-shift to shifted potential changes the energy. So I was not > surprised by the sudden jump of energy between step 0 and step 10 for the > previous simulation (npt14). > > But the jump near step 0 for the current simulation (npt15) is not right. > > This is an equilibration run. > > I have just uploaded the files at http://redmine.gromacs.org/issues/1640 > > > On Tue, Nov 11, 2014 at 10:54 AM, Mark Abraham <[email protected]> > wrote: > >> On Tue, Nov 11, 2014 at 3:38 PM, Johnny Lu <[email protected]> >> wrote: >> >> > Hi. >> > >> > I ran 30ns NPT with Potential-shift-Verlet on both potentials, >> continuing >> > from a NPT without Potential-shift-Verlet. The energy dropped from -3e5 >> to >> > -4e5, which is fine. >> > >> > After that, I continue the 30ns NPT with the same mdp file, except with >> a >> > longer run time (300ns). >> > >> > The total energy was -4e5 at step 0, and then -3e5 at step 5000, and >> now at >> > about 25 ns, it goes back to -3.5e5. Why the energy suddenly changed >> after >> > step 0? >> > >> >> IIRC, the restart ought to compute the same energy on the initial >> coordinates as it did at the end of the previous run, and even if there's >> no re-computation on the same coordinates, obviously it should not jump >> by >> 1e5 in a few steps (or one? can't tell from the resolution). I would think >> that might be a bug in filling a data structure for the output, but >> probably not one in the simulation itself. Please file an issue at >> http://redmine.gromacs.org, tarballing all the relevant files, and I'll >> try >> to look at it. >> >> The long-time behaviour of both simulations is pretty suspect though. >> Since >> you're prepared to be ultra-conservative with your non-bonded and SETTLE >> settings, I would explore using a 0.5fs time step (in normal and >> ultra-conservative mode). Berendsen is not a good barostat algorithm for >> production work, too... but I would not expect such a problem from its >> use. >> Perhaps more importantly, why are you turning off the long-range >> dispersion >> correction? >> >> Mark >> >> I used the same machine and same mixed precision gromacs 4.6.7 mdrun and >> > grompp for both simulations. >> > >> > The plot of total energies for the two simulations: >> > http://oi60.tinypic.com/ddmejd.jpg >> > The two mdp files: http://oi58.tinypic.com/11980si.jpg >> > >> > The commands that I used for continuing the simulations: >> > >> > The previous simulation (npt14): ../../sofware/gromacs-4.6.7/bin/grompp >> -f >> > npt14.mdp -c npt13.tpr -t npt13_step22620750.cpt -o npt14.tpr >> > >> > The current simulation (npt15): ../../sofware/gromacs-4.6.7/bin/grompp >> -f >> > npt15.mdp -c npt14.tpr -t npt14_step30000000.cpt -o npt15.tpr >> > >> > Thanks. >> > -- >> > Gromacs Users mailing list >> > >> > * Please search the archive at >> > http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before >> > posting! >> > >> > * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists >> > >> > * For (un)subscribe requests visit >> > https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or >> > send a mail to [email protected]. >> > >> -- >> Gromacs Users mailing list >> >> * Please search the archive at >> http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before >> posting! >> >> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists >> >> * For (un)subscribe requests visit >> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or >> send a mail to [email protected]. >> > > -- Gromacs Users mailing list * Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before posting! * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists * For (un)subscribe requests visit https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or send a mail to [email protected].
