\(\renewcommand{\AA}{\text{Å}}\)

3.1. Prerequisites

Which software you need to compile and use LAMMPS strongly depends on which features and settings and which optional packages you are trying to include. Common to all is that you need a C++ and C compiler, where the C++ compiler has to support at least the C++17 standard (note that some compilers require a command-line flag to activate C++17 support).

3.1.1. CMake build system

If you are building with CMake, you need at least CMake version 3.20 and a compatible build tool (e.g. GNU make or ninja-build on Linux). The CMake scripting includes tests for required software and will auto-detect and auto-enable available tools and libraries for optional features (of course those can be disabled, if desired). If required software is missing, CMake will try to download the compile the missing parts automatically, if possible or stop with an error.

3.1.2. GNU build system

If you are building LAMMPS with the legacy GNU make based build system you need GNU make (other make variants are not going to work since the build system uses features unique to GNU make) and a Unix-like build environment with a Bourne shell, and shell tools like “sed”, “grep”, “touch”, “test”, “tr”, “cp”, “mv”, “rm”, “ln”, “diff” and so on. Parts of LAMMPS interface with or use Python version 3.6 or later.

3.1.3. Compiler and OS compatibility

The LAMMPS developers aim to keep LAMMPS very portable and usable - at least in parts - on most operating systems commonly used for running MD simulations. Please see the section on portablility for more details.

Warning: LLVM based Intel Compilers

In recent years, Intel switched their compilers from a proprietary code base to compilers based on LLVM (which is also the foundation of the Clang compilers). It has since stopped updating their legacy compilers. Some versions of those LLVM compilers creates incorrect binary code that results in incorrect behavior of LAMMPS; this is particularly common when using high optimization levels and vectorization. There have been several cases where people reported bugs in LAMMPS which turned out to be caused by bugs in the Intel LLVM compilers.

Unfortunately there is no simple way to detect whether a binary is working correctly outside of running the unit and regression tests, but those do not cover all of LAMMPS and and would be reliable only for no or moderate optimization anyway. For most of LAMMPS there is not much of a benefit (if any) to use the Intel compilers over the GCC or Clang compilers, except for the INTEL package (which can be compiled with other compilers, but most vectorization directives are inactive for those) or KOKKOS with SYCL.

Our recommendation is thus to compile with GCC or Clang unless you are using features that require the Intel compilers. In that case, it is recommended to have a backup GCC/Clang compiled binary available to validate correctness of the simulation. If you have access to multiple Intel compiler versions, please try the latest version first and avoid “20##.0” versions which seem to have problems most often.