combining metadynamics options

[micb]

Hi,

I calculate the PMF between two relatively large molecules
using gromacs 4.5.1 (in fact 4.5.3 but the patch seems to work
there as well)

if find that the following works:

HILLS W_STRIDE 1000 HEIGHT 0.5
MULTIPLE_WALKERS HILLS_DIR /home/micb/fold/hills R_STRIDE 10000
NWALKERS 6 ID 0
WELLTEMPERED SIMTEMP 300 BIASFACTOR 10
DISTANCE LIST 241 482 SIGMA 0.05 DIR Z
UWALL CV 1 LIMIT 2.8 KAPPA 1000.0
GRID CV 1 MIN 0.0 MAX 3.5 NBIN 175
ENDMETA

However, as convergence is an issue here i tried to accelaerate
exploration of the configurational space by applying additional
CVs that cover the (relative) rotational degrees of freedom
of the two molecules.
doing this:

HILLS W_STRIDE 1000 HEIGHT 0.5
WELLTEMPERED SIMTEMP 300 BIASFACTOR 10
DISTANCE LIST 241 482 SIGMA 0.05 DIR Z
ANGLE LIST 241 482 262 SIGMA 0.25
ANGLE LIST 482 241 87 SIGMA 0.25
TORSION LIST 87 241 482 262 SIGMA 0.25
UWALL CV 1 LIMIT 2.8 KAPPA 1000.0
GRID CV 1 MIN 0.0 MAX 3.50 NBIN 175
GRID CV 2 MIN 0.0 MAX 3.15 NBIN 175
GRID CV 3 MIN 0.0 MAX 3.15 NBIN 175
GRID CV 4 MIN -3.15 MAX 3.15 NBIN 175
ENDMETA

my runs die without giving any warning or error message.
(in fact the jobs keep running but do not produce any output,
as if they got stuck in an endless loop right at the beginning of
the simulation run)
the above WITHOUT the WELLTEMPERED directive seems
to work, but then i found that not using WELLTEMPERED
makes, again, convergence very slow ...

so my question is: can I combine all of GRID, WELLTEMPERED,
MULTIPLE_WALKERS AND more than one CVs or are any of
these options mutually exclusive?

thanks!
Michael

[Davide Branduardi]

Hi Michael,

i am not an expert of grids but a grid of 175 point in 4 dimensions seems pretty huge to be allocated.
For such a high dimensional space try to use standard hills.
Probably the programs get stuck when trying to allocate and initialize the grid.
Hope it helps

Davide

> --
> You received this message because you are subscribed to the Google Groups "PLUMED users" group.
> To post to this group, send email to plumed...@googlegroups.com.
> To unsubscribe from this group, send email to plumed-users...@googlegroups.com.
> For more options, visit this group at http://groups.google.com/group/plumed-users?hl=en.
>

*********************************************
Davide Branduardi
Postdoc
Theoretical Molecular Biophysics Group
Max Planck Institute for Biophysics
Max-von-Laue-Straße 3
60438 Frankfurt am Main
Phone: +49 069 63031505
Fax: +49 069 63031502
davide.b...@biophys.mpg.de
davide.b...@gmail.com
*********************************************

[micb]

On Jun 28, 5:51 pm, Davide Branduardi <davide.brandua...@gmail.com>
wrote:

> Hi Michael,
>
> i am not an expert of grids but a grid of 175 point in 4 dimensions seems pretty huge to be allocated.
> For such a high dimensional space try to use standard hills.
> Probably the programs get stuck when trying to allocate and initialize the grid.

I agree, the grid is quite big ... but that doesn\t seem to be the
problem
i got gromacs and plumed to work with even larger grids ...

and i should correct what i wrote above, it doesn't seem to work
even without the WELLTEMPERED directive:
the following:

HILLS W_STRIDE 1000 HEIGHT 0.5

MULTIPLE_WALKERS HILLS_DIR /home/micb/aggr/pmf/gmx/implicit/metadyn-4d/

fold/hills R_STRIDE 10000 NWALKERS 6 ID 0

DISTANCE LIST 241 482 SIGMA 0.05 DIR Z
ANGLE LIST 241 482 262 SIGMA 0.25
ANGLE LIST 482 241 87 SIGMA 0.25
TORSION LIST 87 241 482 262 SIGMA 0.25
UWALL CV 1 LIMIT 2.8 KAPPA 1000.0
GRID CV 1 MIN 0.0 MAX 3.50 NBIN 175
GRID CV 2 MIN 0.0 MAX 3.15 NBIN 175
GRID CV 3 MIN 0.0 MAX 3.15 NBIN 175
GRID CV 4 MIN -3.15 MAX 3.15 NBIN 175
ENDMETA

gives me a gromacs run that dies ... the gromacs log file contains
the output until just before getting to the PLUMED stuff ...
and then in hangs indefinitely ...
if i use only ONE CV (the DISTANCE) then a combination
of WELLTEMPERED, GRID, and MULTIPLE_WALKERS seems to work ...

i'd be grateful for any other hints!

cheers
michael

> > For more options, visit this group athttp://groups.google.com/group/plumed-users?hl=en.

>
> *********************************************
> Davide Branduardi
> Postdoc
> Theoretical Molecular Biophysics Group
> Max Planck Institute for Biophysics
> Max-von-Laue-Straße 3
> 60438 Frankfurt am Main
> Phone: +49 069 63031505
> Fax:      +49 069 63031502

> davide.brandua...@biophys.mpg.de
> davide.brandua...@gmail.com
> *********************************************

[Massimiliano Bonomi]

Dear Michael,
in principle the options you used are perfectly compatible.
As Davide said, your grid seems to be pretty big, and
you don't really need such a dense grid for your calculation.
A bin size equal to half the Gaussian sigma should be enough.
As a side note, you may want to use +pi and -pi to set the MIN and MAX
of the grid for ANGLE and/or TORSION, and do not forget to add PBC
to the TORSION line, since this CV is periodic.
Finally, an estimate of the memory usage for grid allocation should be
printed in the output. Try to look for "GRID MEMORY USAGE" in the output.

Best,
Max

[Davide Branduardi]

Hi Michael,

Have you tried to remove the wall? (I'm thinking if, by chance, the energy of the wall is too high and the program hangs
for some strange numerical reason). Another test: if you remove any source of bias (Namely with hills height to 0 and without the wall or KAPPA set to zero)
does it still hung?

D

> For more options, visit this group at http://groups.google.com/group/plumed-users?hl=en.
>

*********************************************
Davide Branduardi
Postdoc
Theoretical Molecular Biophysics Group
Max Planck Institute for Biophysics
Max-von-Laue-Straße 3
60438 Frankfurt am Main
Phone: +49 069 63031505
Fax: +49 069 63031502

davide.b...@biophys.mpg.de
davide.b...@gmail.com
*********************************************

[micb]

thanks everybody for the feedback!

it seems as if the grid size was indeed the problem ...
however reducing the grid unfortunately doesn't seem to
help either, i now use the dat file quoted below
Massimiliano suggested using a grid size with bins of half the
Gaussian sigma), and the grid i use now is actually bigger than that
but I keep getting:
!!!!! PLUMED ERROR: GRID bin size is too large compared to HILLS
sigma.
!!!!! ABORTING RUN
after a while
can it be that the WELLTEMPERED algorithm changes not only the update
frequency but also the sigmas ? is there any way to control that?

HILLS W_STRIDE 1000 HEIGHT 0.5

PRINT W_STRIDE 100
MULTIPLE_WALKERS HILLS_DIR /home/micb/work/aggr/pmf/plumed/ful/
metadyn-4d-wt/hills R_STRIDE 10000 NWALKERS 6 ID 0

WELLTEMPERED SIMTEMP 300 BIASFACTOR 10
DISTANCE LIST 241 482 SIGMA 0.05 DIR Z

ANGLE LIST 241 482 262 SIGMA 0.25
ANGLE LIST 482 241 87 SIGMA 0.25
TORSION LIST 87 241 482 262 SIGMA 0.25
UWALL CV 1 LIMIT 2.8 KAPPA 1000.0

GRID CV 1 MIN 0.0 MAX 3.50 NBIN 150
GRID CV 2 MIN 0.0 MAX +pi NBIN 50
GRID CV 3 MIN 0.0 MAX +pi NBIN 50
GRID CV 4 MIN -pi MAX +pi NBIN 50 PBC
ENDMETA

[Massimiliano Bonomi]

Hi!
Can you try the same input file without the MULTIPLE_WALKERS keyword?

Thanks,
Max

[micb]

On Jun 28, 8:14 pm, Massimiliano Bonomi

<massimiliano.bon...@gmail.com> wrote:
> Hi!
> Can you try the same input file without the MULTIPLE_WALKERS keyword?

actually i found it does work with ALL the options ... after playing
around with
the grid dimensions. It turned out that the room between a too coarse,
and a too large
grid is rather small with 4 CVs ... the input file included below does
work ...
however, i also see that at 4 dimensions metadynamics slows down the
runs
considerably (my system only contains about 500 atoms) ... e.g., for
the system I use here
the simulation on a single node takes about 1.5 hours per nano second
with one CV and
about 5 hours with 4 CVs.
In the manual, section 3.4.5, it says about the GRID option: "In this
way the computational
cost of metadynamics becomes constant all over the simulation" .... I
cannot confirm this,
the projected simulation time keeps growing in fact (logarithmically),
with or without GRID,
only without GRID it doesn't grow as fast.
Given this fact I assume things would go faster if
1) i used more MULTIPLE_WALKERS
2) or if i did a lot of short runs with single walkers
I assume the MULTIPLE_WALKERS would, again, entail i high penalty
due to the required communication time, so i'd rather go for the
second option, but
here i am not sure how can i combine the results of many short
runs ...
with MULTIPLE_WALKERS i just concatenate all HILLS files into a single
one and use the result with sum_hills ... but if just run a lot of
single walker
runs can i still do this, or will that give me a wrong free energy
profile??

thanks!
Michael

HILLS W_STRIDE 1000 HEIGHT 0.5
PRINT W_STRIDE 100
MULTIPLE_WALKERS HILLS_DIR /home/micb/work/aggr/pmf/plumed/ful/
metadyn-4d-wt/hills R_STRIDE 10000 NWALKERS 6 ID 0
WELLTEMPERED SIMTEMP 300 BIASFACTOR 10

DISTANCE LIST 241 482 SIGMA 0.10 DIR Z

ANGLE LIST 241 482 262 SIGMA 0.25
ANGLE LIST 482 241 87 SIGMA 0.25
TORSION LIST 87 241 482 262 SIGMA 0.25
UWALL CV 1 LIMIT 2.8 KAPPA 1000.0

GRID CV 1 MIN 0.0 MAX 3.50 NBIN 100
GRID CV 2 MIN 0.0 MAX +pi NBIN 40
GRID CV 3 MIN 0.0 MAX +pi NBIN 40
GRID CV 4 MIN -pi MAX +pi NBIN 80 PBC
ENDMETA