GRID size

[mihir...@gmail.com]

Hi

I am trying to run a metadynamics simulation. In my plumed input file, I am using grid. 

DISTANCE LIST <g1_CG_pocket> <g2_CG_lig> SIGMA 0.3

TORSION LIST <g1_CG_alpha_F> <g2_CG_alpha_E> <g3_CG_Nter_beta> <g2_CG_lig> SIGMA 0.15

TORSION LIST <g1_Nter_alpha_C> <g2_Cter_alpha_C> <g3_CG_hinge> <g2_CG_lig> SIGMA 0.15

GRID CV 1 MIN 0 MAX 20 NBIN 1000

GRID CV 2 MIN -pi MAX +pi NBIN 200 PBC

GRID CV 3 MIN -pi MAX +pi NBIN 200 PBC

What should be the optimum grid size. Is there any way to calculate the size of grid based on sigma? or any rule of thumb, ball park calculation.

I have tried different grid sizes but after it runs for a while, it crashes with error.

!!!!! PLUMED ERROR: You are outside the GRID!. Please increase GRID size.

!!!!! ABORTING RUN 

GRID CV 1 MIN 0 MAX 20 NBIN 50

GRID CV 2 MIN -pi MAX +pi NBIN 50 PBC

GRID CV 3 MIN -pi MAX +pi NBIN 50 PBC

Then the grid size is too large compared to sigma. 

Thanks

[mihir...@gmail.com]

By the way, I forget to add that I have WALL on CV1 (distance) as LWALL 0 UWALL 15 kappa 50.0
Same is true for CV2 and CV3 (torsions). CV2 and CV3 have walls that range from -2.5 to +2.5 with kappa 50.0 too

[mihir...@gmail.com]

I checked PLUMED manual. It says

If the simulation goes out of the grid, the code will stop. Please increase MIN or MAX and restart metadynamics.

Now if the grid size or grid dimensions are bigger than the WALLs on the CV, then I should not need any bigger size of grid. Should I?

One more thing I observed and that is after I use sum_hills, some of the distance CV values in fes.dat are negative. Is that possible? or does it mean distance being measured in just opposite direction?  

The free energy at negative distance CV points is 0.

Any help or suggestions are welcome. 

Thanks

[mihir...@gmail.com]

Hi

any comment on this issue is welcome.

It is still not resolved and I am trying to figure out optimum GRID size. 

[Po-chia Chen]

 Hi,

 Can you post your full plumed.dat file and plumed version? In PLUMED v2.2 docs, it gives a starting figure of a grid spacing being around 0.2 of sigma. As an example in v2.2:
 SIGMA=1
 GRID_SPACING=0.2
 As for GRID_SIZE, your post is ambiguous. Is the size of a grid referring to its MAX-MIN, or its number of points?

 = = =
 The DISTANCE CV itself should not be negative, as far as I'm aware. The sum_hills calculations, however, is not aware that this is a scalar, so hills near 0 may generate points below 0. Instead, please check if any of the gaussians in HILLS have a negative distance CV (there should be none).

[mihir...@gmail.com]

Hi Chen,

Thanks for reply. Attached is the plumed.dat file 

In my post, by GRID size, I meant to ask about the error message I am getting. !!!!! PLUMED ERROR: You are outside the GRID!. Please increase GRID size.

I wanted to ask both of the things, When such error message is shown, does it mean the problem is with MIN-MAX limits of grid or it is with NBIN?

Thanks again

Mihir

[Po-chia Chen]

 Ok right, it looks like a v1.3 file. I haven't used this. So, if I get this correctly:
 - You can start the simulation without problems. The error pops up after a while (some period of time):
 - CV2 and CV3 are dihedrals and periodic, and so should never exit the grid. This leaves CV1.
 - Your KAPPAs... Ah, they may be too small. KAPPA is the force constant in V_wall=kappa(x-x0)^EXP, and is measured in the units of your MD engine for PLUMED 1.3. (e.g. kJ mol^-1 nm^-2 for linear ones, and radians for dihedral ones.) If you check the manual, the values can be 3000, 10000, etc. - this is typical of GROMACS, with nm as their internal unit. NAMD would use Angstroms, and are therefore 100x smaller.

 (PLUMED 2.0 standardises itself into nm unless you change it in the plumed.dat file.)

 The grid needs to be bigger than the walls anyway, because walls are one-sided polynomial potentials. The problem should be with MIN/MAX limits and not NBIN at all: can you check COLVARS just before the simulation stops? Is CV1 moving outside 0-20? Or CV2/CV3 moving outside +-pi? The best case scenario is that it's simply moving outside R=20, and you should increase KAPPA for CV1. The other possibilities means there's something direly wrong.

 - P.

[mihir...@gmail.com]

Thanks Po,
I looked at my COLVAR. The distance CV is not moving beyond R=20 at all. Infact it always stays within 0 < R <15.
I am looking at other properties. This GRIS is not the geometric grid same as the grid we would use to perform faster grid based calculations like PME or PB equation. Am I right?
Does protein translation affect this grid? I think it should not.