I tried PME and then potential-shift-verlet for Van der Waals. The energy is still not constant in mixed precision gromacs 5.0.1 NVE run on 12 xeon cpu without gpu. Those were nearly the last few things that I knew that I can try in mixed precision. The system has ~165 amino acid, ~9000 water, and 3 Cl- ion in a truncated rhombic dodecahedron box. Total energy of the following simulation drift constantly down. This production run was a continuation of ~300ns npt and ~300ns nvt. The PME grid was 80x80x80.
; Run parameters integrator = md-vv ; leap-frog integrator nsteps = 150000000 ; 1 ns dt = 0.0005 ; 0.5 fs ;NVE ; nstcomm = 500 ; remove center of mass motion to reduce energy and temp. drift. (seems less drift without this) shake-tol = 0.000001 ; shake tolerance. verlet-buffer-tolerance = 0.00000005 ;Output control nstenergy = 20 ; save energies every 1.0 ps nstlog = 100 ; update log file every 0.2 ps ; Bond parameters continuation = yes ; Restarting after NVT constraint_algorithm = lincs ; holonomic constraints constraints = all-bonds ; all bonds (even heavy atom-H bonds) constrained lincs_iter = 2 ; accuracy of LINCS lincs_order = 4 ; also related to accuracy ; Neighborsearching ns_type = grid ; search neighboring grid cells nstlist = 5 ; 10 fs vdw-modifier = Potential-shift-Verlet; rlist = 1.1 ; short-range neighborlist cutoff (in nm) rcoulomb = 1.0 ; short-range electrostatic cutoff (in nm) rvdw = 1.0 ; short-range van der Waals cutoff (in nm) ; Electrostatics coulombtype = PME ; Particle Mesh Ewald for long-range electrostatics pme_order = 6 ; cubic interpolation fourierspacing = 0.1 ; grid spacing for FFT ; Temperature coupling is off tcoupl = no ; modified Berendsen thermostat ; Pressure coupling is off pcoupl = no refcoord_scaling = com ; Periodic boundary conditions pbc = xyz ; 3-D PBC ; Dispersion correction DispCorr = EnerPres ; account for cut-off vdW scheme ; Velocity generation gen_vel = no ; Velocity generation is off cutoff-scheme = Verlet I guess the energy leakage when no thermostat is used doesn't mean the distribution of conformation is incorrect when a thermostat is used. Since double precision run is stable while single precision is not, the energy leakage is caused by numerical error of the computer operation and algorithm instead of the physics or conformation of the setup. (only the precision changed. the mdp files were the same.) I also guess that the majority of simulations in published paper would have such constant drop/rise of total energy if the thermostat is switched off. ( Is this true??? ) When the Amber99ff-ILDN force field is made, NPT and NVT is used, instead of NVE [http://dx.doi.org/10.1002%2Fprot.22711] The Amber lipid 14 forcefield uses NPT to create the parameters. In one of the validations, they ran NPT followed by NVE to calculate diffusion coefficient. The tolerance was more stringent in the NVE part than the NPT part. The plot of energy v.s. time seems to be not in the paper. [ http://pubs.acs.org/doi/full/10.1021/ct4010307] Dynamics is another matter. Langevin thermostats removes momentum transfer and changes the rate dynamics, despite it samples the canonical ensemble ... if the system is ergodic in practice. Two different simulations can converge, if the string of random numbers used in langevin thermostats is the same. I'm not sure about V-rescaling. I agree that all thermostats and barostats likely cause some artifacts in dynamics. On Sat, Oct 11, 2014 at 5:29 PM, Dr. Vitaly Chaban <vvcha...@gmail.com> wrote: > Well, if it does not conserve total energy, than such a computational > experiment is not reliable. This is a sad truth. > > Philosophically, you have two options here. Either you accept that > energy leakage is significant and then you must fight against it > (smaller time-step, double precision, constraints, etc), or -- on the > contrary -- you say that energy loss can be neglected and proceed "as > is". > > My personal rule is if the system does not lose more than 2% of total > energy during an entire run, the result is acceptable. I can be > criticized for this voluntary assumption, obviously. > > Anyway, tricks with T-coupling constitute a very bad idea in this > case. You basically try to mask the energy leakage problem with the > thermostat. Thermostat never removes the problem itself. > > > Dr. Vitaly V. Chaban > > Виталий Витальевич ЧАБАН > > > On Sat, Oct 11, 2014 at 11:12 PM, Johnny Lu <johnny.lu...@gmail.com> > wrote: > > I tried NVE on single precision gromacs 4.6.7 with and without GPU. > > The energy either drift down or up, depending on the mdp file, and the > drift > > seems linear. > > > > Linear fit says that if I run 100 ns simulation, at the end the energy is > > doubled. > > > > If I use double precision gromacs, the drift seems insignificant over > 100 ns > > simulation by linear regression fitting on a 300 ps run. > > But double precision gromacs can't use gpu and is much slower than single > > precision gromacs. > > > > On Sat, Oct 11, 2014 at 4:33 PM, Dr. Vitaly Chaban <vvcha...@gmail.com> > > wrote: > >> > >> I also cannot get why not to just turn off the thermostat... > >> > >> If energy transfer phenomena are in focus, NVE is an ensemble of > >> choice, no matter which relaxation constant is used for the T-coupling > >> algorithm. > >> > >> > >> Dr. Vitaly V. Chaban > >> > >> Виталий Витальевич ЧАБАН > >> > >> > >> On Sat, Oct 11, 2014 at 6:32 PM, Antonio Baptista <bapti...@itqb.unl.pt > > > >> wrote: > >> > Then, run true NVE, for the reasons we already pointed. > >> > > >> > On Sat, 11 Oct 2014, Johnny Lu wrote: > >> > > >> >> For dynamics with correct rate and correct fluctuation. > >> >> > >> >> On Sat, Oct 11, 2014 at 8:09 AM, Dr. Vitaly Chaban < > vvcha...@gmail.com> > >> >> wrote: > >> >> > >> >>> what is it needed for? > >> >>> > >> >>> > >> >>> > >> >>> Dr. Vitaly V. Chaban > >> >>> > >> >>> Виталий Витальевич ЧАБАН > >> >>> > >> >>> > >> >>> On Fri, Oct 10, 2014 at 7:41 PM, Johnny Lu <johnny.lu...@gmail.com> > >> >>> wrote: > >> >>> > Hi. > >> >>> > > >> >>> > Is it a good idea to mimic NVE by a NVT simulation with a large > >> >>> temperature > >> >>> > coupling time constant, to reduce the effect of the thermostat ? > >> >>> > > >> >>> > If I use V-rescale thermostat, what artifacts will the simulation > >> >>> > get > >> >>> > if > >> >>> I > >> >>> > use a large coupling time (say, 500 ps) and single precision > gromacs > >> >>> > ? > >> >>> > > >> >>> > Thank you. > >> >>> > -- > >> >>> > 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 gmx-users-requ...@gromacs.org. > >> >>> -- > >> >>> 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 gmx-users-requ...@gromacs.org. > >> >>> > >> >> -- > >> >> 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 gmx-users-requ...@gromacs.org. > >> > > >> > > >> > -- > >> > 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 gmx-users-requ...@gromacs.org. > >> > > >> -- > >> 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 gmx-users-requ...@gromacs.org. > > > > > -- 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 gmx-users-requ...@gromacs.org.