pair_style hybrid command

Accelerator Variants: hybrid/kk

pair_style hybrid/overlay command

Accelerator Variants: hybrid/overlay/kk

pair_style hybrid/scaled command


pair_style hybrid style1 args style2 args ...
pair_style hybrid/overlay style1 args style2 args ...
pair_style hybrid/scaled factor1 style1 args factor2 style 2 args ...
  • style1,style2 = list of one or more pair styles and their arguments

  • factor1,factor2 = scale factors for pair styles, may be a variable


pair_style hybrid lj/cut/coul/cut 10.0 eam lj/cut 5.0
pair_coeff 1*2 1*2 eam niu3
pair_coeff 3 3 lj/cut/coul/cut 1.0 1.0
pair_coeff 1*2 3 lj/cut 0.5 1.2

pair_style hybrid/overlay lj/cut 2.5 coul/long 2.0
pair_coeff * * lj/cut 1.0 1.0
pair_coeff * * coul/long

pair_style hybrid/scaled 0.5 tersoff 0.5 sw
pair_coeff * * tersoff Si.tersoff Si
pair_coeff * * sw Si.sw Si

variable one equal ramp(1.0,0.0)
variable two equal 1.0-v_one
pair_style hybrid/scaled v_one lj/cut 2.5 v_two morse 2.5
pair_coeff 1 1 lj/cut 1.0 1.0 2.5
pair_coeff 1 1 morse 1.0 1.0 1.0 2.5


The hybrid, hybrid/overlay, and hybrid/scaled styles enable the use of multiple pair styles in one simulation. With the hybrid style, exactly one pair style is assigned to each pair of atom types. With the hybrid/overlay and hybrid/scaled styles, one or more pair styles can be assigned to each pair of atom types. The assignment of pair styles to type pairs is made via the pair_coeff command. The major difference between the hybrid/overlay and hybrid/scaled styles is that the hybrid/scaled adds a scale factor for each sub-style contribution to forces, energies and stresses. Because of the added complexity, the hybrid/scaled style has more overhead and thus may be slower than hybrid/overlay.

Here are two examples of hybrid simulations. The hybrid style could be used for a simulation of a metal droplet on a LJ surface. The metal atoms interact with each other via an eam potential, the surface atoms interact with each other via a lj/cut potential, and the metal/surface interaction is also computed via a lj/cut potential. The hybrid/overlay style could be used as in the second example above, where multiple potentials are superimposed in an additive fashion to compute the interaction between atoms. In this example, using lj/cut and coul/long together gives the same result as if the lj/cut/coul/long potential were used by itself. In this case, it would be more efficient to use the single combined potential, but in general any combination of pair potentials can be used together in to produce an interaction that is not encoded in any single pair_style file, e.g. adding Coulombic forces between granular particles. Another limitation of using the hybrid/overlay variant, that it does not generate lj/cut parameters for mixed atom types from a mixing rule due to restrictions discussed below.

If the hybrid/scaled style is used instead of hybrid/overlay, contributions from sub-styles are weighted by their scale factors, which may be fractional or even negative. Furthermore the scale factors may be variables that may change during a simulation. This enables switching smoothly between two different pair styles or two different parameter sets during a run in a similar fashion as could be done with fix adapt or fix alchemy.

All pair styles that will be used are listed as “sub-styles” following the hybrid or hybrid/overlay keyword, in any order. In case of the hybrid/scaled pair style, each sub-style is prefixed with a scale factor. The scale factor is either a floating point number or an equal style (or equivalent) variable. Each sub-style’s name is followed by its usual arguments, as illustrated in the examples above. See the doc pages of the individual pair styles for a listing and explanation of the appropriate arguments for them.

Note that an individual pair style can be used multiple times as a sub-style. For efficiency reasons this should only be done if your model requires it. E.g. if you have different regions of Si and C atoms and wish to use a Tersoff potential for pure Si for one set of atoms, and a Tersoff potential for pure C for the other set (presumably with some third potential for Si-C interactions), then the sub-style tersoff could be listed twice. But if you just want to use a Lennard-Jones or other pairwise potential for several different atom type pairs in your model, then you should just list the sub-style once and use the pair_coeff command to assign parameters for the different type pairs.


There is one exception to this option to list an individual pair style multiple times: GPU-enabled pair styles in the GPU package. This is because the GPU package currently assumes that only one instance of a pair style is being used.

In the pair_coeff commands, the name of a pair style must be added after the I,J type specification, with the remaining coefficients being those appropriate to that style. If the pair style is used multiple times in the pair_style command, then an additional numeric argument must also be specified which is a number from 1 to M where M is the number of times the sub-style was listed in the pair style command. The extra number indicates which instance of the sub-style these coefficients apply to.

For example, consider a simulation with 3 atom types: types 1 and 2 are Ni atoms, type 3 are LJ atoms with charges. The following commands would set up a hybrid simulation:

pair_style hybrid eam/alloy lj/cut/coul/cut 10.0 lj/cut 8.0
pair_coeff * * eam/alloy nialhjea Ni Ni NULL
pair_coeff 3 3 lj/cut/coul/cut 1.0 1.0
pair_coeff 1*2 3 lj/cut 0.8 1.3

As an example of using the same pair style multiple times, consider a simulation with 2 atom types. Type 1 is Si, type 2 is C. The following commands would model the Si atoms with Tersoff, the C atoms with Tersoff, and the cross-interactions with Lennard-Jones:

pair_style hybrid lj/cut 2.5 tersoff tersoff
pair_coeff * * tersoff 1 Si.tersoff Si NULL
pair_coeff * * tersoff 2 C.tersoff NULL C
pair_coeff 1 2 lj/cut 1.0 1.5

It is not recommended to read pair coefficients for a hybrid style from a “Pair Coeffs” or “PairIJ Coeffs” section of a data file via the read_data command, since those sections expect a fixed number of lines, either one line per atom type or one line pair pair of atom types, respectively. When reading from a data file, the lines of the “Pair Coeffs” and “PairIJ Coeffs” are changed in the same way as the pair_coeff command, i.e. the name of the pair style to which the parameters apply must follow the atom type (or atom types), e.g.

Pair Coeffs

1 lj/cut/coul/cut 1.0 1.0

PairIJ Coeffs

1 1 lj/cut/coul/cut 1.0 1.0

Note that the pair_coeff command for some potentials such as pair_style eam/alloy includes a mapping specification of elements to all atom types, which in the hybrid case, can include atom types not assigned to the eam/alloy potential. The NULL keyword is used by many such potentials (eam/alloy, Tersoff, AIREBO, etc), to denote an atom type that will be assigned to a different sub-style.

For the hybrid style, each atom type pair I,J is assigned to exactly one sub-style. Just as with a simulation using a single pair style, if you specify the same atom type pair in a second pair_coeff command, the previous assignment will be overwritten.

For the hybrid/overlay and hybrid/scaled styles, each atom type pair I,J can be assigned to one or more sub-styles. If you specify the same atom type pair in a second pair_coeff command with a new sub-style, then the second sub-style is added to the list of potentials that will be calculated for two interacting atoms of those types. If you specify the same atom type pair in a second pair_coeff command with a sub-style that has already been defined for that pair of atoms, then the new pair coefficients simply override the previous ones, as in the normal usage of the pair_coeff command. E.g. these two sets of commands are the same:

pair_style lj/cut 2.5
pair_coeff * * 1.0 1.0
pair_coeff 2 2 1.5 0.8

pair_style hybrid/overlay lj/cut 2.5
pair_coeff * * lj/cut 1.0 1.0
pair_coeff 2 2 lj/cut 1.5 0.8

Coefficients must be defined for each pair of atoms types via the pair_coeff command as described above, or in the “Pair Coeffs” or “PairIJ Coeffs” section of the data file read by the read_data command, or by mixing as described below.

For all of the hybrid, hybrid/overlay, and hybrid/scaled styles, every atom type pair I,J (where I <= J) must be assigned to at least one sub-style via the pair_coeff command as in the examples above, or in the data file read by the read_data, or by mixing as described below. Also all sub-styles must be used at least once in a pair_coeff command.


With hybrid pair styles the use of mixing to generate pair coefficients is significantly limited compared to the individual pair styles. LAMMPS never performs mixing of parameters from different sub-styles, even if they use the same type of coefficients, e.g. contain a Lennard-Jones potential variant. Those parameters must be provided explicitly. Also for hybrid/overlay and hybrid/scaled mixing is only performed for pairs of atom types for which only a single pair style is assigned.

Thus it is strongly recommended to provide all mixed terms explicitly. For non-hybrid styles those could be generated and written out using the write_coeff command and then edited as needed to comply with the requirements for hybrid styles as explained above.

If you want there to be no interactions between a particular pair of atom types, you have 3 choices. You can assign the pair of atom types to some sub-style and use the neigh_modify exclude type command. You can assign it to some sub-style and set the coefficients so that there is effectively no interaction (e.g. epsilon = 0.0 in a LJ potential). Or, for hybrid, hybrid/overlay, or hybrid/scaled simulations, you can use this form of the pair_coeff command in your input script or the “PairIJ Coeffs” section of your data file:

pair_coeff  2 3  none

or this form in the “Pair Coeffs” section of the data file:

3  none

If an assignment to none is made in a simulation with the hybrid/overlay or hybrid/scaled pair style, it wipes out all previous assignments of that pair of atom types to sub-styles.

Note that you may need to use an atom_style hybrid command in your input script, if atoms in the simulation will need attributes from several atom styles, due to using multiple pair styles with different requirements.

Different force fields (e.g. CHARMM vs. AMBER) may have different rules for applying exclusions or weights that change the strength of pairwise non-bonded interactions between pairs of atoms that are also 1-2, 1-3, and 1-4 neighbors in the molecular bond topology. This is normally a global setting defined the special_bonds command. However, different weights can be assigned to different hybrid sub-styles via the pair_modify special command. This allows multiple force fields to be used in a model of a hybrid system, however, there is no consistent approach to determine parameters automatically for the interactions between atoms of the two force fields, thus this approach this is only recommended when particles described by the different force fields do not mix.

Here is an example for combining CHARMM and AMBER: The global amber setting sets the 1-4 interactions to non-zero scaling factors and then overrides them with 0.0 only for CHARMM:

special_bonds amber
pair_style hybrid lj/charmm/coul/long 8.0 10.0 lj/cut/coul/long 10.0
pair_modify pair lj/charmm/coul/long special lj/coul 0.0 0.0 0.0

This input achieves the same effect:

special_bonds 0.0 0.0 0.1
pair_style hybrid lj/charmm/coul/long 8.0 10.0 lj/cut/coul/long 10.0
pair_modify pair lj/cut/coul/long special lj 0.0 0.0 0.5
pair_modify pair lj/cut/coul/long special coul 0.0 0.0 0.83333333
pair_modify pair lj/charmm/coul/long special lj/coul 0.0 0.0 0.0

Here is an example for combining Tersoff with OPLS/AA based on a data file that defines bonds for all atoms where - for the Tersoff part of the system - the force constants for the bonded interactions have been set to 0. Note the global settings are effectively lj/coul 0.0 0.0 0.5 as required for OPLS/AA:

special_bonds lj/coul 1e-20 1e-20 0.5
pair_style hybrid tersoff lj/cut/coul/long 12.0
pair_modify pair tersoff special lj/coul 1.0 1.0 1.0

For use with the various compute */tally computes, the pair_modify compute/tally command can be used to selectively turn off processing of the compute tally styles, for example, if those pair styles (e.g. many-body styles) do not support this feature.

See the pair_modify page for details on the specific syntax, requirements and restrictions.

The potential energy contribution to the overall system due to an individual sub-style can be accessed and output via the compute pair command. Note that in the case of pair style hybrid/scaled this is the unscaled potential energy of the selected sub-style.


Several of the potentials defined via the pair_style command in LAMMPS are really many-body potentials, such as Tersoff, AIREBO, MEAM, ReaxFF, etc. The way to think about using these potentials in a hybrid setting is as follows.

A subset of atom types is assigned to the many-body potential with a single pair_coeff command, using “* *” to include all types and the NULL keywords described above to exclude specific types not assigned to that potential. If types 1,3,4 were assigned in that way (but not type 2), this means that all many-body interactions between all atoms of types 1,3,4 will be computed by that potential. Pair_style hybrid allows interactions between type pairs 2-2, 1-2, 2-3, 2-4 to be specified for computation by other pair styles. You could even add a second interaction for 1-1 to be computed by another pair style, assuming pair_style hybrid/overlay is used.

But you should not, as a general rule, attempt to exclude the many-body interactions for some subset of the type pairs within the set of 1,3,4 interactions, e.g. exclude 1-1 or 1-3 interactions. That is not conceptually well-defined for many-body interactions, since the potential will typically calculate energies and foces for small groups of atoms, e.g. 3 or 4 atoms, using the neighbor lists of the atoms to find the additional atoms in the group.

However, you can still use the pair_coeff none setting or the neigh_modify exclude command to exclude certain type pairs from the neighbor list that will be passed to a many-body sub-style. This will alter the calculations made by a many-body potential beyond the specific pairs, since it builds its list of 3-body, 4-body, etc interactions from the pair lists. You will need to think carefully as to whether excluding such pairs produces a physically meaningful result for your model.

For example, imagine you have two atom types in your model, type 1 for atoms in one surface, and type 2 for atoms in the other, and you wish to use a Tersoff potential to compute interactions within each surface, but not between the surfaces. Then either of these two command sequences would implement that model:

pair_style hybrid tersoff
pair_coeff * * tersoff SiC.tersoff C C
pair_coeff 1 2 none

pair_style tersoff
pair_coeff * * SiC.tersoff C C
neigh_modify exclude type 1 2

Either way, only neighbor lists with 1-1 or 2-2 interactions would be passed to the Tersoff potential, which means it would compute no 3-body interactions containing both type 1 and 2 atoms.

Here is another example to use 2 many-body potentials together in an overlapping manner using hybrid/overlay. Imagine you have CNT (C atoms) on a Si surface. You want to use Tersoff for Si/Si and Si/C interactions, and AIREBO for C/C interactions. Si atoms are type 1; C atoms are type 2. Something like this will work:

pair_style hybrid/overlay tersoff airebo 3.0
pair_coeff * * tersoff SiC.tersoff.custom Si C
pair_coeff * * airebo CH.airebo NULL C

Note that to prevent the Tersoff potential from computing C/C interactions, you would need to modify the SiC.tersoff potential file to turn off C/C interaction, i.e. by setting the appropriate coefficients to 0.0.

Styles with a gpu, intel, kk, omp, or opt suffix are functionally the same as the corresponding style without the suffix. They have been optimized to run faster, depending on your available hardware, as discussed on the Accelerator packages page. The accelerated styles take the same arguments and should produce the same results, except for round-off and precision issues.

These accelerated styles are part of the GPU, INTEL, KOKKOS, OPENMP, and OPT packages, respectively. They are only enabled if LAMMPS was built with those packages. See the Build package page for more info.

You can specify the accelerated styles explicitly in your input script by including their suffix, or you can use the -suffix command-line switch when you invoke LAMMPS, or you can use the suffix command in your input script.

See the Accelerator packages page for more instructions on how to use the accelerated styles effectively.


Since the hybrid, hybrid/overlay, hybrid/scaled styles delegate computation to the individual sub-styles, the suffix versions of the hybrid and hybrid/overlay styles are used to propagate the corresponding suffix to all sub-styles, if those versions exist. Otherwise the non-accelerated version will be used. The individual accelerated sub-styles are part of the GPU, KOKKOS, INTEL, OPENMP, and OPT packages, respectively. They are only enabled if LAMMPS was built with those packages. See the Build package page for more info.

Mixing, shift, table, tail correction, restart, rRESPA info

Any pair potential settings made via the pair_modify command are passed along to all sub-styles of the hybrid potential.

For atom type pairs I,J and I != J, if the sub-style assigned to I,I and J,J is the same, and if the sub-style allows for mixing, then the coefficients for I,J can be mixed. This means you do not have to specify a pair_coeff command for I,J since the I,J type pair will be assigned automatically to the sub-style defined for both I,I and J,J and its coefficients generated by the mixing rule used by that sub-style. For the hybrid/overlay and hybrid/scaled style, there is an additional requirement that both the I,I and J,J pairs are assigned to a single sub-style. If this requirement is not met, no I,J coeffs will be generated, even if the sub-styles support mixing, and I,J pair coefficients must be explicitly defined.

See the pair_modify command for details of mixing rules. See the See the page for the sub-style to see if allows for mixing.

The hybrid pair styles supports the pair_modify shift, table, and tail options for an I,J pair interaction, if the associated sub-style supports it.

For the hybrid pair styles, the list of sub-styles and their respective settings are written to binary restart files, so a pair_style command does not need to specified in an input script that reads a restart file. However, the coefficient information is not stored in the restart file. Thus, pair_coeff commands need to be re-specified in the restart input script. For pair style hybrid/scaled also the names of any variables used as scale factors are restored, but not the variables themselves, so those may need to be redefined when continuing from a restart.

These pair styles support the use of the inner, middle, and outer keywords of the run_style respa command, if their sub-styles do.


When using a long-range Coulombic solver (via the kspace_style command) with a hybrid pair_style, one or more sub-styles will be of the “long” variety, e.g. lj/cut/coul/long or buck/coul/long. You must ensure that the short-range Coulombic cutoff used by each of these long pair styles is the same or else LAMMPS will generate an error.

Pair style hybrid/scaled currently only works for non-accelerated pair styles and pair styles from the OPT package.

When using pair styles from the GPU package they must not be listed multiple times. LAMMPS will detect this and abort.