Special replica syntax

(this part of the manual is based on Using special syntax for multiple replicas).

In many cases, we need to run multiple replicas with almost identical PLUMED files. These files might be prepared with cut-and-paste, which is very error prone, or could be set up with some smart bash or python script. Additionally, one can take advantage of the INCLUDE keyword so as to have a shared input file with common definitions and specific input files with replica-dependent keywords. However, as of PLUMED 2.4, we introduced a simpler manner to manipulate multiple replica inputs with tiny differences. Look at the following example:

 
#SETTINGS NREPLICAS=3
# Compute a distance
d: DISTANCE ATOMS=1,2

# Apply a restraint.
RESTRAINT ARG=d AT=@replicas:1.0,1.1,1.2 KAPPA=1.0
# On replica 0, this means:
#   RESTRAINT ARG=d AT=1.0 KAPPA=1.0
# On replica 1, this means:
#   RESTRAINT ARG=d AT=1.1 KAPPA=1.0
# On replica 2, this means:
#   RESTRAINT ARG=d AT=1.2 KAPPA=1.0

If you prepare a single plumed.dat file like this one and feeds it to PLUMED while using 3 replicas, the 3 replicas will see the very same input except for the AT keyword, that sets the position of the restraint. Replica 0 will see a restraint centered at 1.0, replica 1 centered at 1.1, and replica 2 centered at 1.2.

The @replicas: keyword is not special for RESTRAINT or for the AT keyword. Any keyword in PLUMED can accept that syntax. For instance, the following single input file can be used to setup a bias exchange metadynamics [88] simulations:

#SETTINGS NREPLICAS=2
# Compute distance between atoms 1 and 2
d: DISTANCE ATOMS=1,2

# Compute a torsional angle
t: TORSION ATOMS=30,31,32,33

# Metadynamics.
METAD ...
  ARG=@replicas:d,t
  HEIGHT=1.0
  PACE=100
  SIGMA=@replicas:0.1,0.3
  GRID_MIN=@replicas:0.0,-pi
  GRID_MAX=@replicas:2.0,pi
...
# On replica 0, this means:
#  METAD ARG=d HEIGHT=1.0 PACE=100 SIGMA=0.1 GRID_MIN=0.0 GRID_MAX=2.0
# On replica 1, this means:
#  METAD ARG=t HEIGHT=1.0 PACE=100 SIGMA=0.3 GRID_MIN=-pi GRID_MAX=pi

This would be a typical setup for a bias exchange simulation. Notice that even though variables d and t are both read in both replicas, d is only computed on replica 0 (and t is only computed on replica 1). This is because variables that are defined but not used are never actually calculated by PLUMED.

If the value that should be provided for each replica is a vector, you should use curly braces as delimiters. For instance, if the restraint acts on two variables, you can use the following input:

#SETTINGS NREPLICAS=3
# Compute distance between atoms 1 and 2
d: DISTANCE ATOMS=10,20

# Compute a torsional angle
t: TORSION ATOMS=30,31,32,33

# Apply a restraint:
RESTRAINT ...
  ARG=d,t
  AT=@replicas:{{1.0,2.0} {3.0,4.0} {5.0,6.0}}
  KAPPA=1.0,3.0
...
# On replica 0 this means:
#  RESTRAINT ARG=d AT=1.0,2.0 KAPPA=1.0,3.0
# On replica 1 this means:
#  RESTRAINT ARG=d AT=3.0,4.0 KAPPA=1.0,3.0
# On replica 2 this means:
#  RESTRAINT ARG=d AT=5.0,6.0 KAPPA=1.0,3.0

Notice the double curly braces. The outer ones are used by PLUMED to know there the argument of the AT keyword ends, whereas the inner ones are used to group the values corresponding to each replica. Also notice that the last example can be split in multiple lines exploiting the fact that within multi-line statements (enclosed by pairs of ...) newlines are replaced with simple spaces:

#SETTINGS NREPLICAS=3
d: DISTANCE ATOMS=10,20
t: TORSION ATOMS=30,31,32,33
RESTRAINT ...
  ARG=d,t
# indentation is not required (this is not python!)
# but makes the input easier to read
  AT=@replicas:{
    {1.0,2.0}
    {3.0,4.0}
    {5.0,6.0}
  }
  KAPPA=1.0,3.0
...

In short, whenever there are keywords that should vary across replicas, you should set them using the @replicas: keyword. As mentioned above, you can always use the old syntax with separate input file, and this is recommended when the number of keywords that are different is large.