compute rdf command


compute ID group-ID rdf Nbin itype1 jtype1 itype2 jtype2 ... keyword/value ...
  • ID, group-ID are documented in compute command

  • rdf = style name of this compute command

  • Nbin = number of RDF bins

  • itypeN = central atom type for Nth RDF histogram (integer, type label, or asterisk form)

  • jtypeN = distribution atom type for Nth RDF histogram (integer, type label, or asterisk form)

  • zero or more keyword/value pairs may be appended

  • keyword = cutoff

    cutoff value = Rcut
      Rcut = cutoff distance for RDF computation (distance units)


compute 1 all rdf 100
compute 1 all rdf 100 1 1
compute 1 all rdf 100 * 3 cutoff 5.0
compute 1 fluid rdf 500 1 1 1 2 2 1 2 2
compute 1 fluid rdf 500 1*3 2 5 *10 cutoff 3.5


Define a computation that calculates the radial distribution function (RDF), also called \(g(r)\), and the coordination number for a group of particles. Both are calculated in histogram form by binning pairwise distances into Nbin bins from 0.0 to the maximum force cutoff defined by the pair_style command or the cutoff distance Rcut specified via the cutoff keyword. The bins are of uniform size in radial distance. Thus a single bin encompasses a thin shell of distances in 3d and a thin ring of distances in 2d.


If you have a bonded system, then the settings of special_bonds command can remove pairwise interactions between atoms in the same bond, angle, or dihedral. This is the default setting for the special_bonds command, and means those pairwise interactions do not appear in the neighbor list. Because this fix uses a neighbor list, it also means those pairs will not be included in the RDF. This does not apply when using long-range coulomb interactions (coul/long, coul/msm, coul/wolf or similar. One way to get around this would be to set special_bond scaling factors to very tiny numbers that are not exactly zero (e.g., \(1.0 \times 10^{-50}\)). Another workaround is to write a dump file, and use the rerun command to compute the RDF for snapshots in the dump file. The rerun script can use a special_bonds command that includes all pairs in the neighbor list.

By default the RDF is computed out to the maximum force cutoff defined by the pair_style command. If the cutoff keyword is used, then the RDF is computed accurately out to the Rcut \(> 0.0\) distance specified.


Normally, you should only use the cutoff keyword if no pair style is defined (e.g., the rerun command is being used to post-process a dump file of snapshots) or if you really want the RDF for distances beyond the pair_style force cutoff and cannot easily post-process a dump file to calculate it. This is because using the cutoff keyword incurs extra computation and possibly communication, which may slow down your simulation. If you specify Rcut \(\le\) force cutoff, you will force an additional neighbor list to be built at every timestep this command is invoked (or every reneighboring timestep, whichever is less frequent), which is inefficient. LAMMPS will warn you if this is the case. If you specify a Rcut > force cutoff, you must ensure ghost atom information out to Rcut + skin is communicated, via the comm_modify cutoff command, else the RDF computation cannot be performed, and LAMMPS will give an error message. The skin value is what is specified with the neighbor command. In this case, you are forcing a large neighbor list to be built just for the RDF computation, and extra communication to be performed every timestep.

The itypeN and jtypeN arguments are optional. These arguments must come in pairs. If no pairs are listed, then a single histogram is computed for \(g(r)\) between all atom types. If one or more pairs are listed, then a separate histogram is generated for each itype,jtype pair.

The itypeN and jtypeN settings can be specified in one of three ways. One or both of the types in the I,J pair can be a type label. Or an explicit numeric value can be used, as in the fourth example above. Or a wild-card asterisk can be used to specify a range of atom types. This takes the form “*” or “*n” or “m*” or “m*n”. If \(N\) is the number of atom types, then an asterisk with no numeric values means all types from 1 to \(N\). A leading asterisk means all types from 1 to n (inclusive). A trailing asterisk means all types from m to \(N\) (inclusive). A middle asterisk means all types from m to n (inclusive).

If both itypeN and jtypeN are single values, as in the fourth example above, this means that a \(g(r)\) is computed where atoms of type itypeN are the central atom, and atoms of type jtypeN are the distribution atom. If either itypeN and jtypeN represent a range of values via the wild-card asterisk, as in the fifth example above, this means that a \(g(r)\) is computed where atoms of any of the range of types represented by itypeN are the central atom, and atoms of any of the range of types represented by jtypeN are the distribution atom.

Pairwise distances are generated by looping over a pairwise neighbor list, just as they would be in a pair_style computation. The distance between two atoms \(I\) and \(J\) is included in a specific histogram if the following criteria are met:

  • atoms \(I\) and \(J\) are both in the specified compute group

  • the distance between atoms \(I\) and \(J\) is less than the maximum force cutoff

  • the type of the \(I\) atom matches itypeN (one or a range of types)

  • the type of the \(J\) atom matches jtypeN (one or a range of types)

It is OK if a particular pairwise distance is included in more than one individual histogram, due to the way the itypeN and jtypeN arguments are specified.

The \(g(r)\) value for a bin is calculated from the histogram count by scaling it by the idealized number of how many counts there would be if atoms of type jtypeN were uniformly distributed. Thus it involves the count of itypeN atoms, the count of jtypeN atoms, the volume of the entire simulation box, and the volume of the bin’s thin shell in 3d (or the area of the bin’s thin ring in 2d).

A coordination number \(\mathrm{coord}(r)\) is also calculated, which is the number of atoms of type jtypeN within the current bin or closer, averaged over atoms of type itypeN. This is calculated as the area- or volume-weighted sum of \(g(r)\) values over all bins up to and including the current bin, multiplied by the global average volume density of atoms of type jtypeN.

The simplest way to output the results of the compute rdf calculation to a file is to use the fix ave/time command, for example:

compute myRDF all rdf 50
fix 1 all ave/time 100 1 100 c_myRDF[*] file tmp.rdf mode vector

Output info

This compute calculates a global array in which the number of rows is Nbins and the number of columns is \(1 + 2N_\text{pairs}\), where \(N_\text{pairs}\) is the number of \(I,J\) pairings specified. The first column has the bin coordinate (center of the bin), and each successive set of two columns has the \(g(r)\) and \(\text{coord}(r)\) values for a specific set of itypeN versus jtypeN interactions, as described above. These values can be used by any command that uses a global values from a compute as input. See the Howto output page for an overview of LAMMPS output options.

The array values calculated by this compute are all “intensive”.

The first column of array values will be in distance units. The \(g(r)\) columns of array values are normalized numbers \(\ge 0.0\). The coordination number columns of array values are also numbers \(\ge 0.0\).


By default, the RDF is not computed for distances longer than the largest force cutoff, since the neighbor list creation will only contain pairs up to that distance (plus neighbor list skin). This distance can be increased using the cutoff keyword but this keyword is only valid with neighbor styles ‘bin’ and ‘nsq’.

If you want an RDF for larger distances, you can also use the rerun command to post-process a dump file, use pair style zero and set the force cutoff to be longer in the rerun script. Note that in the rerun context, the force cutoff is arbitrary and with pair style zero you are not computing any forces, and you are not running dynamics you are not changing the model that generated the trajectory.

The definition of \(g(r)\) used by LAMMPS is only appropriate for characterizing atoms that are uniformly distributed throughout the simulation cell. In such cases, the coordination number is still correct and meaningful. As an example, if a large simulation cell contains only one atom of type itypeN and one of jtypeN, then \(g(r)\) will register an arbitrarily large spike at whatever distance they happen to be at, and zero everywhere else. The function \(\text{coord}(r)\) will show a step change from zero to one at the location of the spike in \(g(r)\).


compute rdf can handle dynamic groups and systems where atoms are added or removed, but this causes that certain normalization parameters need to be re-computed in every step and include collective communication operations. This will reduce performance and limit parallel efficiency and scaling. For systems, where only the type of atoms changes (e.g., when using fix atom/swap), you need to explicitly request the dynamic normalization updates via compute_modify dynamic/dof yes


The keyword defaults are cutoff = 0.0 (use the pairwise force cutoff).