PLUMED Grid Issue

- Problem

I recently noted one of the systems we are simulating suffered from decreasing performance as the simulation progressed. Further investigation revealed it is not the actual performance of the simulation that was lower, but the fact that PLUMED needed increasingly large amounts of time to pre-process the HILLS file for each replica.

This is not an issue for a short-running simulation but for one where the simulation time runs into several μs (such as Coarse-grained simulations), it can completely hinder the trajectory from progressing past a certain point in time.

- Solution

Comparisons with other systems and similar issues mentioned in the issues of the PLUMED github repository, as well as in the PLUMED mailing list, pointed in the direction of a grid-related setting being the cause of the delay, specifically, the fineness/coarseness of the grid. 

The two files below (CORRECT PLUMED INPUT FILE & OLD PLUMED INPUT FILE) highlight a different way of specifying the grid density.

In the `old` file, the grid spacing was manually specified with the GRID_SPACING argument in the METAD block.

In the `correct` file, that line is absent from the file. In the absence of a GRID_SPACING or a GRID_BIN argument, PLUMED is going to use a grid spacing value equal to 1/5 of the Collective Variable (CV) width, for each CV, espectively.

Further testing is required but this value appears to be a robust default and has solved this issue in this particular instance.


CORRECT PLUMED INPUT FILE

C1: RMSD REFERENCE=rmsd_reference.pdb TYPE=OPTIMAL

COM1: CENTER ATOMS=1-122
COM2: CENTER ATOMS=123-244

D1: DISTANCE ATOMS=COM1,COM2

METAD ...
ARG=C1,D1
SIGMA=0.4,0.8
HEIGHT=0.005
PACE=100
LABEL=meta
BIASFACTOR=2.0
TEMP=300
GRID_MIN=0,1
GRID_MAX=4.5,6
... METAD
UPPER_WALLS ARG=C1 AT=4 KAPPA=300.0 EXP=2 EPS=1 OFFSET=0 LABEL=uwall
UPPER_WALLS ARG=D1 AT=5.5 KAPPA=300.0 EXP=2 EPS=1 OFFSET=0 LABEL=u2wall

# monitor the two variables and the metadynamics bias potential
PRINT STRIDE=10000 ARG=C1,D1,meta.bias FILE=COLVAR
​


OLD INPUT PLUMED FILE (do not use!)

C1: RMSD REFERENCE=rmsd_reference.pdb TYPE=OPTIMAL

COM1: CENTER ATOMS=1-122
COM2: CENTER ATOMS=123-244

D1: DISTANCE ATOMS=COM1,COM2

METAD ...
ARG=C1,D1
SIGMA=0.4,0.8
HEIGHT=0.005
PACE=100
LABEL=meta
BIASFACTOR=2.0
TEMP=300
GRID_MIN=0,1
GRID_MAX=4.5,6
GRID_SPACING=0.01,0.01
... METAD
UPPER_WALLS ARG=C1 AT=4 KAPPA=300.0 EXP=2 EPS=1 OFFSET=0 LABEL=uwall
UPPER_WALLS ARG=D1 AT=5.5 KAPPA=300.0 EXP=2 EPS=1 OFFSET=0 LABEL=u2wall

# monitor the two variables and the metadynamics bias potential
PRINT STRIDE=10000 ARG=C1,D1,meta.bias FILE=COLVAR