[CPMD-list] problem with path integrals restart
Mark E Tuckerman
mark.tuckerman at nyu.edu
Fri Apr 4 08:00:00 CEST 2008
Hi Axel,
I thought I would respond to this as well. Generally, it does not make
sense to run NVE path integrals EVER, i.e., you should always use
the MASSIVE thermostatting option. Having said this, let me
qualify it just a little. Thermostatting should always be used for
equilibrium path integrals. If, however, you want to try to to
approximate path-integral dynamics via the new ring-polyer
molecular dynamics scheme of Mannolopoulos and coworkers, then
you have to perform an NVE path-integral simulation in primitive
variables (no STAGING or NORMAL modes). In this case, you would
just compute the instantaneous temperature as usual from the
kinetic energy. This output is really only useful to ensure that your
simulation is heating or exhibiting some other pathology.
Mark
====================================================
Mark E. Tuckerman
Director of Graduate Studies
Associate Professor of Chemistry and Mathematics
Dept. of Chemistry
and Courant Institute of Mathematical Sciences
New York University
100 Washington Square East
New York, NY 10003
Phone: (212) 998-8471
Fax: (212) 260-7905
E-mail: mark.tuckerman at nyu.edu
URL: http://homepages.nyu.edu/~mt33
http://www.nyu.edu/pages/chemistry/faculty
====================================================
----- Original Message -----
From: Axel Kohlmeyer <akohlmey at cmm.chem.upenn.edu>
Date: Thursday, April 3, 2008 3:03 pm
Subject: Re: [CPMD-list] problem with path integrals restart
To: Marek Naryniecki <marek at elrond.chem.uni.wroc.pl>
Cc: CPMD-list at cpmd.org
> marek,
>
> i had a closer look at the PI code (while debugging something related)
> and found that those "spurious" NaNs and Infinities can come from
> when you are not using a thermostat and thus the target temperature
> is not defined. if just you put a TEMPERATURE statement into
> the &CPMD section, even though it will be ignored when restarting
> with velocities, those denormalized (information only) numbers go away.
>
> now as a result this brings up the question: does it make sense at all
> to run an NVE path-integrals run and if yes, how to handle this case?
> should one compute the instantenous temperature? or just skip that output?
>
> cheers,
> axel.
> 2008/4/2 Marek Naryniecki <marek at elrond.chem.uni.wroc.pl>:
> >
> > Dear Axel and Serge
> >
> > I know DIPOLE shouldn't apeare in PI calculation. My mistake.
> > Calculation without DIPOLE produce the same problem.
> > I put 2 replicas for fast check of the problem on varius machines
> and
> > pseudos. This problem also occure with 8 replicas in 120K. I know
> it is
> > small number but more replicas is to expensive for me. I'll try
> some short
> > calculation with 32 replicas.
> >
> >
> > > On Wed, 2 Apr 2008, Marek Naryniecki wrote:
> > >
> > > MN> Dear All
> > >
> > > dear marek,
> >
> > >
> > > MN> ***************************************************************
> > > MN> * CHARACTERISTIC RADII
> *
> > > MN> ***************************************************************
> >
> > > MN> SPECIES R OF GYRATION FREE PARTICLE (IN ANGSTROM)
> > > MN> 1 P 0.06799 Infinity
> > > MN> 2 O 0.05446 Infinity
> > > MN> 3 K 0.08015 Infinity
> > > MN> 4 H 0.12800 Infinity
> > >
> >
> > > looks like a division by zero somewhere.
> > >
> > > [...]
> > >
> >
> > > MN> what is the reason of that? After that calculations go on and
> average
> > > MN> values for bonds seems to be good.
> > > MN> I looked into cpmd code but without succes.
> > >
> >
> > > you're neither giving enough information to reproduce
> >
> > > that run somewhere else (full input!) nor enough details
> >
> > I'm sorry, I thought it will be easier to read. Now I know it was stupid.
> >
> > I attached inputs for replicas generation and dynamic. if I stop dynamic
> >
> > and then restart it this problem occures.
> >
> > > to narrow it down. what machine are you running on?
> >
> > > does the error happen in serial or in parallel?
> >
> > it happen in serial and in parallel. I chacked on several machines
> i was
> >
> > able and the problem happen.
> > >
> >
> > > looking at the code itself may not be enough.
> >
> > > compile cpmd with debug info, find the code
> >
> > > where the properties from above are computed,
> >
> > > set a breakpoint and try to figure out where the
> >
> > > denormalized numbers are coming from...
> >
> > I'm not familar with fortran, but I will try
> > >
> >
> > > MN> 100.0
> > > MN> &END
> > > MN> ..
> > > MN> &PIMD
> > > MN> TROTTER DIMENSION
> > >
> >
> > > it has become a very bad habit on this list lately to
> >
> > > exclude vital information from inputs posted. if you
> >
> > > don't make it easy to reproduce your problem, you'll
> >
> > > have to solve it by yourself.
> >
> > ok, You're right.
> > >
> > > [...]
> > > MN>
> > > MN> Please help me.
> >
> > >
> > > quid pro quo. you have to help us to help you.
> > >
> >
> > Thank You very much for reply. I'm still trying to figure out what
> >
> > is wrong. I thought someone tell me what can this mean physically.
> >
> >
> > Many Thanks
> > marek
> > > cheers,
> > > axel.
> > >
> >
> > > MN>
> > > MN> Best Regards,
> > > MN>
> > > MN>
> >
> > > MN> Marek Naryniecki
> > > MN> Wydzial Chemii
> >
> > > MN> Uniwersytet Wroclawski
> > > MN>
> > > MN>
> > > MN>
> > > MN> _______________________________________________
> >
> > > MN> CPMD-list mailing list
> > > MN> CPMD-list at cpmd.org
> > > MN> http://cpmd.org/mailman/listinfo/cpmd-list
> > > MN>
> > >
> > > --
> >
> >
> > > =======================================================================
> > > Axel Kohlmeyer akohlmey at cmm.chem.upenn.edu http://www.cmm.upenn.edu
> > > Center for Molecular Modeling -- University of Pennsylvania
> > > Department of Chemistry, 231 S.34th Street, Philadelphia, PA 19104-6323
> > > tel: 1-215-898-1582, fax: 1-215-573-6233, office-tel: 1-215-898-5425
> > > =======================================================================
> > > If you make something idiot-proof, the universe creates a better
> idiot.
> > >
> >
> > _______________________________________________
> > CPMD-list mailing list
> > CPMD-list at cpmd.org
> > http://cpmd.org/mailman/listinfo/cpmd-list
> >
> >
>
>
>
> --
> =======================================================================
> Axel Kohlmeyer akohlmey at cmm.chem.upenn.edu http://www.cmm.upenn.edu
> Center for Molecular Modeling -- University of Pennsylvania
> Department of Chemistry, 231 S.34th Street, Philadelphia, PA 19104-6323
> tel: 1-215-898-1582, fax: 1-215-573-6233, office-tel: 1-215-898-5425
> =======================================================================
> If you make something idiot-proof, the universe creates a better idiot.
>
More information about the CPMD-list
mailing list