1. LAMMPS Library Interfaces
As described on the library interface to LAMMPS page, LAMMPS can be built as a library (static or shared), so that it can be called by another code, used in a coupled manner with other codes, or driven through a Python script. The LAMMPS standalone executable itself is essentially a thin wrapper on top of the LAMMPS library, which creates a LAMMPS instance, passes the input for processing to that instance, and then exits.
Most of the APIs described below are based on C language wrapper
functions in the files
it is also possible to use C++ directly. The basic procedure is
always the same: you create one or more instances of
LAMMPS, pass commands as strings or
from files to that LAMMPS instance to execute calculations, and/or
call functions that read, manipulate, and update data from the active
class instances inside LAMMPS to do analysis or perform operations
that are not possible with existing input script commands.
LAMMPS was initially not conceived as a thread-safe program, but over the years changes have been applied to replace operations that collide with creating multiple LAMMPS instances from multiple-threads of the same process with thread-safe alternatives. This primarily applies to the core LAMMPS code and less so on add-on packages, especially when those packages require additional code in the lib folder, interface LAMMPS to Fortran libraries, or the code uses static variables (like the COLVARS package).
Another major issue to deal with is to correctly handle MPI.
Creating a LAMMPS instance requires passing an MPI communicator, or
it assumes the
MPI_COMM_WORLD communicator, which spans all MPI
processor ranks. When creating multiple LAMMPS object instances
from different threads, this communicator has to be different for
each thread or else collisions can happen. Or it has to be
guaranteed, that only one thread at a time is active. MPI
communicators, however, are not a problem, if LAMMPS is compiled
with the MPI STUBS library, which implies that there is no MPI
communication and only 1 MPI rank.
1.1. LAMMPS C Library API
The C library interface is the most commonly used path to manage LAMMPS
instances from a compiled code and it is the basis for the Python and Fortran modules. Almost all
functions of the C language API require an argument containing a
“handle” in the form of a
void * type variable, which points to the
location of a LAMMPS class instance.
library.h header file by default does not include the
header file and thus hides the
lammps_open() function which
requires the declaration of the
MPI_comm data type. This is only
a problem when the communicator that would be passed is different from
MPI_COMM_WORLD. Otherwise calling
will work just as well. To make
you need to compile the code with
-DLAMMPS_LIB_MPI or add the line
#define LAMMPS_LIB_MPI before
Please note the
mpi.h file must usually be the same (and thus the
MPI library in use) for the LAMMPS code and library and the calling code.
The exception is when LAMMPS was compiled in serial mode using the
STUBS MPI library. In that case the calling code may be compiled
with a different MPI library so long as
is called to create a LAMMPS instance. In that case each MPI rank will
run LAMMPS in serial mode.
Errors versus exceptions
If the LAMMPS executable encounters an error condition, it will abort
after printing an error message. For a library interface this is
usually not desirable. Thus LAMMPS can be compiled to to throw
a C++ exception instead. If enabled, the library
functions will catch those exceptions and return. The error status
can be queried and an
message retrieved. We thus
recommend enabling C++ exceptions when using the library interface,
Using the C library interface as a plugin
Rather than including the C library directly and link to the LAMMPS
library at compile time, you can use the
file and the
liblammpsplugin.c C code in the
examples/COUPLE/plugin folder for an interface to LAMMPS that is
largely identical to the regular library interface, only that it will
load a LAMMPS shared library file at runtime. This can be useful for
applications where the interface to LAMMPS would be an optional
No checks are made on the arguments of the function calls of the C library interface. All function arguments must be non-NULL unless explicitly allowed, and must point to consistent and valid data. Buffers for storing returned data must be allocated to a suitable size. Passing invalid or unsuitable information will likely cause crashes or corrupt data.
- 1.1.1. Creating or deleting a LAMMPS object
- 1.1.2. Executing commands
- 1.1.3. System properties
- 1.1.4. Per-atom properties
- 1.1.5. Compute, fixes, variables
- 1.1.6. Scatter/gather operations
- 1.1.7. Neighbor list access
- 1.1.8. Configuration information
- 1.1.9. Utility functions
- 1.1.10. Extending the C API
1.2. LAMMPS Python APIs
The LAMMPS Python module enables calling the LAMMPS C library API from Python by dynamically loading functions in the LAMMPS shared library through the Python ctypes module. Because of the dynamic loading, it is required that LAMMPS is compiled in “shared” mode. The Python interface is object-oriented, but otherwise tries to be very similar to the C library API. Three different Python classes to run LAMMPS are available and they build on each other. More information on this is in the Use Python with LAMMPS section of the manual. Use of the LAMMPS Python module is described in The lammps Python module.
1.3. LAMMPS Fortran API
The LAMMPS Fortran module is a wrapper around calling functions from the LAMMPS C library API. This is done using the ISO_C_BINDING feature in Fortran 2003. The interface is object-oriented but otherwise tries to be very similar to the C library API and the basic Python module.
1.4. LAMMPS C++ API
It is also possible to invoke the LAMMPS C++ API directly in your code. It lacks some of the convenience of the C library API, but it allows more direct access to simulation data and thus more low-level manipulations. The following links provide some examples and references to the C++ API.