[gmx-users] Tabscale

David van der Spoel spoel at xray.bmc.uu.se
Mon Sep 22 13:42:01 CEST 2003 

On Mon, 2003-09-22 at 13:22, Yuguang Mu wrote:
> I have tried to increase tabscale = 2000 in  single precision,
> But as Eric said, this does not change the results at all.
> 
> So the only thing I can ssy is that the FFTW program itself, has somewhat
> differences  between single and double precision.
> 
> And in some cases, I can only trust in the double precision.
> 
Could you please do some more tests? 
Maybe try a different grid spacings and also a higher pme_order (=6) so
that would be a total of 8 simulations, 2 grid spacings, pme_order = 4
and 6, and single and double precision.

It is important to sort this out, but right now both Erik and me have
little time...

> 
> Dr. Yuguang Mu
> Institute for Physical and Theoretical Chemistry
> J.W. Goethe University Frankfurt am Main
> Marie Curie Str. 11
> 60439 Frankfurt/Main, Germany
> Tel: +49-(0)69-798-29711
> 
> On 20 Sep 2003, David wrote:
> 
> > On Wed, 2003-09-17 at 14:33, Yuguang Mu wrote:
> > > Dear David,
> > > Generally speaking, you are right.
> > > But in my case, not !
> > > I use -rerun switch, it is exactly the same trajectory.
> > >
> > > I check the different energy terms, only the Coulomb (LR) (using PME)
> > > term differs evidently by 0.5% , but systematicly , the single precision
> > > values always larger (or smaller )  than double one by around 0.3-0.5% .
> > > Here I show two figtures about this terms coming from two differnt
> > > systems.
> > >
> > > You can see clearly that
> > > they comes from the same trajectory, but differ a certain amount.
> > >
> > > So it seems that the PME term differs a factor of 0.3-0.5% .
> > >
> > > can we improve it ?
> > > At least this Coulomb (LR) should be as good as other terms
> > > having a difference factor of 0.0003-0.0005% which the single precision
> > > promises.
> > >
> >
> > Could you maybe rerun your test in single precision after recompiling
> > with tabscale = 2000 (line 581 in tables.c)
> >
> > It could be that the difference lies in the Ewald corrections for
> > exclusions.
> >
> > >
> > >
> > >
> > > Dr. Yuguang Mu
> > > Institute for Physical and Theoretical Chemistry
> > > J.W. Goethe University Frankfurt am Main
> > > Marie Curie Str. 11
> > > 60439 Frankfurt/Main, Germany
> > > Tel: +49-(0)69-798-29711
> > >
> > > On 17 Sep 2003, David wrote:
> > >
> > > > On Wed, 2003-09-17 at 12:37, Yuguang Mu wrote:
> > > > > Dear Erik,
> > > > > The question I really want to ask is that why the virial calculated from
> > > > > the same trajectory but using different precisions are quite different.
> > > > >
> > > > > I have a trajectory simulated in single precision, then I use -rerun
> > > > > switch to recalculate energies in double precision.
> > > > >
> > > > > The potentials, kinetics, terms turn out nearly exactly same, but I found
> > > > > the virial, also, the pressure, specially the off-diagonal terms, deviate
> > > > > greatly.
> > > > >
> > > > > In other words, if we eqilibrate the system in single precision with
> > > > > pressure coupling (orthogonal box but with anisotropy coupling), the
> > > > > final shape of the box will be greatly different from that simulated with
> > > > > double precision using the same protocol.
> > > > >
> > > > > I donot think it tolerant.
> > > > >
> > > > > Is it possible that in the formula of virial calculations there happens
> > > > > somehow overflow inthe case of single precision ?
> > > >
> > > > I don't think so. It is just lack of precision which leads to another
> > > > trajectory. If you have anisotropic scaling you must increase tau_p to
> > > > at least 5 ps. Even then two different trajectories may give you
> > > > different final box size. Try e.g. to run two double precision runs with
> > > > different starting velocities and see how the box evolves.
> > > >
> > > > >
> > > > > Dr. Yuguang Mu
> > > > > Institute for Physical and Theoretical Chemistry
> > > > > J.W. Goethe University Frankfurt am Main
> > > > > Marie Curie Str. 11
> > > > > 60439 Frankfurt/Main, Germany
> > > > > Tel: +49-(0)69-798-29711
> > > > >
> > > > > On Tue, 16 Sep 2003, Erik Lindahl wrote:
> > > > >
> > > > > > Hi,
> > > > > >
> > > > > > Not without hacking the code. The tabscale is selected as 500 points/nm
> > > > > > in single and 2000 points/nm in double. With the cubic spline
> > > > > > interpolation, this essentially gives you full single precision
> > > > > > accuracy, and almost full double precision accuracy for all common
> > > > > > potential forms, respectively.
> > > > > >
> > > > > > Cheers,
> > > > > >
> > > > > > Erik
> > > > > >
> > > > > >
> > > > > > On Tuesday, September 16, 2003, at 09:26 AM, Yuguang Mu wrote:
> > > > > >
> > > > > > > Hi All,
> > > > > > > Inthe log file always printed out the following lines:
> > > > > > >
> > > > > > > Table routines are used for coulomb: TRUE
> > > > > > > Table routines are used for vdw:     FALSE
> > > > > > > Using a Gaussian width (1/beta) of 0.320163 nm for Ewald
> > > > > > > Cut-off's:   NS: 1   Coulomb: 1   LJ: 1
> > > > > > > Generated table with 2000 data points for COUL.
> > > > > > > Tabscale = 2000 points/nm
> > > > > > > Generated table with 2000 data points for LJ6.
> > > > > > > Tabscale = 2000 points/nm
> > > > > > > Generated table with 2000 data points for LJ12.
> > > > > > > Tabscale = 2000 points/nm
> > > > > > > Generated table with 3200 data points for Ewald.
> > > > > > > Tabscale = 2000 points/nm
> > > > > > > Generated table with 3200 data points for LJ6.
> > > > > > > Tabscale = 2000 points/nm
> > > > > > > Generated table with 3200 data points for LJ12.
> > > > > > > Tabscale = 2000 points/nm
> > > > > > >
> > > > > > > here i sthe question, could I change the Tabscale manually ?
> > > > > > >
> > > > > > >
> > > > > > > Dr. Yuguang Mu
> > > > > > > Institute for Physical and Theoretical Chemistry
> > > > > > > J.W. Goethe University Frankfurt am Main
> > > > > > > Marie Curie Str. 11
> > > > > > > 60439 Frankfurt/Main, Germany
> > > > > > > Tel: +49-(0)69-798-29711
> > > > > > >
> > > > > > >
> > > > > > > _______________________________________________
> > > > > > > gmx-users mailing list
> > > > > > > gmx-users at gromacs.org
> > > > > > > http://www.gromacs.org/mailman/listinfo/gmx-users
> > > > > > > Please don't post (un)subscribe requests to the list. Use the
> > > > > > > www interface or send it to gmx-users-request at gromacs.org.
> > > > > >
> > > > > > _______________________________________________
> > > > > > gmx-users mailing list
> > > > > > gmx-users at gromacs.org
> > > > > > http://www.gromacs.org/mailman/listinfo/gmx-users
> > > > > > Please don't post (un)subscribe requests to the list. Use the
> > > > > > www interface or send it to gmx-users-request at gromacs.org.
> > > > > >
> > > > >
> > > > > _______________________________________________
> > > > > gmx-users mailing list
> > > > > gmx-users at gromacs.org
> > > > > http://www.gromacs.org/mailman/listinfo/gmx-users
> > > > > Please don't post (un)subscribe requests to the list. Use the
> > > > > www interface or send it to gmx-users-request at gromacs.org.
> > > > --
> > > > Groeten, David.
> > > > ________________________________________________________________________
> > > > Dr. David van der Spoel, 	Dept. of Cell and Molecular Biology
> > > > Husargatan 3, Box 596,  	75124 Uppsala, Sweden
> > > > phone:	46 18 471 4205		fax: 46 18 511 755
> > > > spoel at xray.bmc.uu.se	spoel at gromacs.org   http://xray.bmc.uu.se/~spoel
> > > > ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
> > > >
> > > > _______________________________________________
> > > > gmx-users mailing list
> > > > gmx-users at gromacs.org
> > > > http://www.gromacs.org/mailman/listinfo/gmx-users
> > > > Please don't post (un)subscribe requests to the list. Use the
> > > > www interface or send it to gmx-users-request at gromacs.org.
> > > >
> > --
> > Groeten, David.
> > ________________________________________________________________________
> > Dr. David van der Spoel, 	Dept. of Cell and Molecular Biology
> > Husargatan 3, Box 596,  	75124 Uppsala, Sweden
> > phone:	46 18 471 4205		fax: 46 18 511 755
> > spoel at xray.bmc.uu.se	spoel at gromacs.org   http://xray.bmc.uu.se/~spoel
> > ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
> >
> > _______________________________________________
> > gmx-users mailing list
> > gmx-users at gromacs.org
> > http://www.gromacs.org/mailman/listinfo/gmx-users
> > Please don't post (un)subscribe requests to the list. Use the
> > www interface or send it to gmx-users-request at gromacs.org.
> >
> 
> _______________________________________________
> gmx-users mailing list
> gmx-users at gromacs.org
> http://www.gromacs.org/mailman/listinfo/gmx-users
> Please don't post (un)subscribe requests to the list. Use the 
> www interface or send it to gmx-users-request at gromacs.org.
-- 
Groeten, David.
________________________________________________________________________
Dr. David van der Spoel, 	Dept. of Cell & Mol. Biology
Husargatan 3, Box 596,  	75124 Uppsala, Sweden
phone:	46 18 471 4205		fax: 46 18 511 755
spoel at xray.bmc.uu.se	spoel at gromacs.org   http://xray.bmc.uu.se/~spoel
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

More information about the gromacs.org_gmx-users mailing list