It is the nb potential which is problematic. This seems so strange, the
functions seem fine and exactly what you would expect from smooth well
sampled RDFs. What is even more puzzling is that the nb energy breaks in
just one time step. I have printed out energy information every timestep:
Step Time Lambda
4 0.00400 0.00000
Energies (kJ/mol)
Tab. Bonds Tab. Angles Tab. Dih. LJ (SR) Coulomb (SR)
1.20852e+04 3.70247e+04 9.51537e+03 -7.67663e+05 0.00000e+00
Potential Kinetic En. Total Energy Temperature Pressure (bar)
-7.09038e+05 2.99711e+04 -6.79067e+05 3.16241e+02 -3.31874e+04
Step Time Lambda
5 0.00500 0.00000
Energies (kJ/mol)
Tab. Bonds Tab. Angles Tab. Dih. LJ (SR) Coulomb (SR)
1.20860e+04 3.70238e+04 9.51379e+03 -nan 0.00000e+00
So after integration step 4 the LJ energy is -7.67663e+05 but after the
next it is -nan. In past trials with this system today the LJ energy jumps
to ~10^213 in just one step.
If the configuration was bad then surely the energy would be very large
(and positive) from the first step?
Any ideas what might cause it to jump like this so suddenly? I would
understand if the LJ energy increased monotonically but it seems stable
(admittedly for an embarrassingly short amount of time) and then breaks.
Thanks again,
Dan.
On Wednesday, 28 January 2015 17:30:20 UTC, Christoph Junghans wrote:
>
> 2015-01-28 9:11 GMT-07:00 Daniel Allen <[email protected] <javascript:>>:
>
> > Hi Christoph,
> >
> > thanks for your prompt reply. Sorry I should have been more clear, the
> > system is blowing up on the very first CG run using the initial guesses
> for
> > potentials. I realise that you could say the initial configuration must
> be
> > bad however I don't think it'll be any worse than any other snapshot
> from
> > the reference simulation given that the bilayer is stable during this.
> That usually means one of the initial guesses or the
> Boltzmann-inverted potential (when no initial potentials are given) is
> bad.
> You can look at the energy file to see, which contribution to the
> energy blows up.
> Or replace the potential one by one by a Lennard Jones to find the bad
> one.
>
> Christoph
>
> >
> > Thanks for your help,
> >
> > Dan.
> >
> > On Wednesday, 28 January 2015 15:49:53 UTC, Christoph Junghans wrote:
> >>
> >> 2015-01-28 7:45 GMT-07:00 Daniel Allen <[email protected]>:
> >> > Hello,
> >> >
> >> > I am trying to perform IBI on a bilayer formed of some polymers. I
> have
> >> > obtained distributions for non-bonded and bonded interactions from
> >> > csg_stat
> >> > and feed these into csg_inverse in *dist.tgt files. I am not
> currently
> >> > iterating bonded interactions, just using csg_inverse to produce
> >> > tabulated
> >> > potentials (bonds, angles, dihedrals).
> >> >
> >> > My RDFs do not quite converge to 1 as my reference system is not
> >> > homogeneous
> >> > so I am cutting off interactions at the first minimum in the RDFs (at
> >> > approximately 0.75 nm).
> >> >
> >> > After about 10 integration steps in the first iteration, the
> simulation
> >> > blows up. The initial configuration is the last snapshot of my
> atomistic
> >> > reference simulation mapped onto CG representation.
> >> >
> >> > I have looked at the nb potentials (they seem reasonable) and have
> run a
> >> > script to determine the shortest distance between 2 non-bonded beads
> in
> >> > my
> >> > initial conf.gro that I feed in as the starting state. The shortest
> >> > distance
> >> > is 0.328 nm and sigma for that pairwise interaction is 0.41 nm. This
> >> > would
> >> > result in an interaction energy of ~12.8 kJ/mol which doesn't seem
> >> > ridiculously high?
> >> >
> >> > Any ideas how to move past this? I could try using a different
> starting
> >> > state but I fear that any snapshots from the reference system will
> have
> >> > pair
> >> > distances < sigma.
> >> It is really hard to say how to overcome this, but I have more general
> >> list of advice:
> >> 1.) Scale the update
> >> 2.) Change the update sequence if you have multiple interactions
> >> 3.) Use last configuration from the previous step ("laststep") as
> >> initial configuration
> >> 4.) Do a pre simulation in each iteration step (pre_simulation) to
> >> minimize the initial configuration
> >> 5.) Increase the min to only update parts of the potential
> >> 6.) Longer coarse-grained simulations
> >> 7.) Apply a transformation on the rdf before doing the update
> >> 8.) Use an initial guess (pot.in) on a bigger interval with premolded
> >> structure
> >>
> >> Christoph
> >>
> >> >
> >> >
> >> > Thanks in advance,
> >> >
> >> > Dan.
> >> >
> >> > --
> >> > You received this message because you are subscribed to the Google
> >> > Groups
> >> > "votca" group.
> >> > To unsubscribe from this group and stop receiving emails from it,
> send
> >> > an
> >> > email to [email protected].
> >> > To post to this group, send email to [email protected].
> >> > Visit this group at http://groups.google.com/group/votca.
> >> > For more options, visit https://groups.google.com/d/optout.
> >>
> >>
> >>
> >> --
> >> Christoph Junghans
> >> Web: http://www.compphys.de
> >
> > --
> > You received this message because you are subscribed to the Google
> Groups
> > "votca" group.
> > To unsubscribe from this group and stop receiving emails from it, send
> an
> > email to [email protected] <javascript:>.
> > To post to this group, send email to [email protected]
> <javascript:>.
> > Visit this group at http://groups.google.com/group/votca.
> > For more options, visit https://groups.google.com/d/optout.
>
>
>
> --
> Christoph Junghans
> Web: http://www.compphys.de
>
--
You received this message because you are subscribed to the Google Groups
"votca" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
To post to this group, send email to [email protected].
Visit this group at http://groups.google.com/group/votca.
For more options, visit https://groups.google.com/d/optout.
