Getting Started

August 21, 2026 · View on GitHub

Model Independent Chemical Module. MICM can be used to configure and solve atmospheric chemistry systems.

Note MICM 3.x.x is part of a refactor and may include breaking changes across minor revision numbers and partially implemented features

Getting Started

Installing MICM locally

To build and install MICM locally, you must have CMake installed on your machine.

Open a terminal window, navigate to a folder where you would like the MICM files to exist, and run the following commands:

git clone https://github.com/NCAR/micm.git
cd micm
mkdir build
cd build
ccmake ..
sudo make install -j 8

To run the tests:

make test

If you would later like to uninstall MICM, you can run sudo make uninstall from the build/ directory.

There are multiple options for running micm. You can use our solvers on CPUs, cuda-based solvers to solve chemistry on GPUs, or Kokkos-based solvers for performance portability across CPUs and GPUs (-DMICM_ENABLE_KOKKOS=ON; see the Kokkos guide). Please read our docs to learn how to enable these options.

Third-party components fetched at build time (Kokkos, GoogleTest) are listed in NOTICE along with their licenses.

Running a MICM Docker container

You must have Docker Desktop installed and running. With Docker Desktop running, open a terminal window. To build the latest MICM release, run the following command to start the MICM container:

docker run -it ghcr.io/ncar/micm:release bash

To build the latest pre-release version of MICM, instead run:

git clone https://github.com/NCAR/micm.git
cd micm
docker build -t micm -f docker/Dockerfile .
docker run -it micm bash

Inside the container, you can run the MICM tests from the /build/ folder:

cd /build/
make test

Using the MICM API

The following example solves the fictitious chemical system:

foo       --k1--> 0.8 bar + 0.2 baz
foo + bar --k2--> baz

The k1 and k2 rate constants are for Arrhenius reactions. See the MICM documentation for details on the types of reactions available in MICM and how to configure them.

To solve this system save the following code in a file named foo_chem.cpp:

#include <micm/process/chemical_reaction_builder.hpp>
#include <micm/process/rate_constant/arrhenius_rate_constant.hpp>
#include <micm/solver/rosenbrock.hpp>
#include <micm/solver/solver_builder.hpp>

#include <iomanip>
#include <iostream>

using namespace micm;

int main(const int argc, const char *argv[])
{
  auto foo = Species{ "Foo" };
  auto bar = Species{ "Bar" };
  auto baz = Species{ "Baz" };

  Phase gas_phase{ "gas", std::vector<PhaseSpecies>{ foo, bar, baz } };

  System chemical_system{ gas_phase };

  Process r1 = ChemicalReactionBuilder()
                   .SetReactants({ foo })
                   .SetProducts({ StoichSpecies(bar, 0.8), StoichSpecies(baz, 0.2) })
                   .SetRateConstant(ArrheniusRateConstantParameters{ .A_ = 1.0e-3 })
                   .SetPhase(gas_phase)
                   .Build();

  Process r2 = ChemicalReactionBuilder()
                   .SetReactants({ foo, bar })
                   .SetProducts({ StoichSpecies(baz, 1) })
                   .SetRateConstant(ArrheniusRateConstantParameters{ .A_ = 1.0e-5, .C_ = 110.0 })
                   .SetPhase(gas_phase)
                   .Build();

  std::vector<Process> reactions{ r1, r2 };

  auto solver = micm::CpuSolverBuilder<micm::RosenbrockSolverParameters>(micm::RosenbrockSolverParameters::ThreeStageRosenbrockParameters())
                    .SetSystem(chemical_system)
                    .SetReactions(reactions)
                    .Build();

  State state = solver.GetState();

  state.conditions_[0].temperature_ = 287.45;  // K
  state.conditions_[0].pressure_ = 101319.9;   // Pa
  state.conditions_[0].CalculateIdealAirDensity();
  state[foo] = 20.0;                           // mol m-3

  state.PrintHeader();
  for (int i = 0; i < 10; ++i)
  {
    solver.UpdateStateParameters(state);
    auto result = solver.Solve(500.0, state);
    state.PrintState(i * 500);
  }

  return 0;
}

To build and run the example using GNU (assuming the default install location):

g++ -o foo_chem foo_chem.cpp -I/usr/local/micm-3.13.0/include -std=c++20
./foo_chem

Output:

 time,        Foo,        Bar,        Baz
    0,   1.18e+01,   5.90e+00,   1.91e+00
  500,   6.79e+00,   9.05e+00,   3.32e+00
 1000,   3.83e+00,   1.07e+01,   4.21e+00
 1500,   2.14e+00,   1.17e+01,   4.74e+00
 2000,   1.19e+00,   1.22e+01,   5.04e+00
 2500,   6.58e-01,   1.24e+01,   5.21e+00
 3000,   3.64e-01,   1.26e+01,   5.31e+00
 3500,   2.01e-01,   1.27e+01,   5.36e+00
 4000,   1.11e-01,   1.27e+01,   5.39e+00
 4500,   6.13e-02,   1.28e+01,   5.41e+00

Performance

Every push to main records the benchmark and publishes the history as a chart. Instruction counts come from callgrind and are deterministic, so they show a hot-path change even when the wall-clock time is noisy.

Two mechanisms run. Chapman has 7 reactions and shows per-call overhead. TS1 has 547 reactions and shows how the solver scales with mechanism size.

chart	mechanism	backend	grid cells	steps	machine
Instruction counts	Chapman	CPU	2000	5	`ubuntu-latest`
Wall-clock timing	Chapman	CPU	10000	30	`ubuntu-latest`
Instruction counts	TS1	CPU	2000	5	`ubuntu-latest`
Wall-clock timing	TS1	CPU	10000	30	`ubuntu-latest`
Wall-clock timing	Chapman and TS1	CUDA	10000	30	CIRRUS a10 GPU runner
Wall-clock timing	Chapman and TS1	Kokkos	10000	30	CIRRUS a10 GPU runner

Every step advances the solver by 30 s. The callgrind charts use a smaller grid and fewer steps, because valgrind runs far slower than a native run. The vector128 ordering pads its last group, so it solves 2048 cells rather than 2000, and 10112 rather than 10000.

Each pull request also gets a commit comment that compares its Chapman numbers against the latest main values. See docs/performance.md to run the benchmark yourself.

Citation

MICM is part of the MUSICA project and can be cited by reference to the MUSICA vision paper. The BibTeX entry below can be used to generate a citation for this.

@Article { acom.software.musica-vision,
    author = "Gabriele G. Pfister and Sebastian D. Eastham and Avelino F. Arellano and Bernard Aumont and Kelley C. Barsanti and Mary C. Barth and Andrew Conley and Nicholas A. Davis and Louisa K. Emmons and Jerome D. Fast and Arlene M. Fiore and Benjamin Gaubert and Steve Goldhaber and Claire Granier and Georg A. Grell and Marc Guevara and Daven K. Henze and Alma Hodzic and Xiaohong Liu and Daniel R. Marsh and John J. Orlando and John M. C. Plane and Lorenzo M. Polvani and Karen H. Rosenlof and Allison L. Steiner and Daniel J. Jacob and Guy P. Brasseur",
    title = "The Multi-Scale Infrastructure for Chemistry and Aerosols (MUSICA)",
    journal = "Bulletin of the American Meteorological Society",
    year = "2020",
    publisher = "American Meteorological Society",
    address = "Boston MA, USA",
    volume = "101",
    number = "10",
    doi = "10.1175/BAMS-D-19-0331.1",
    pages= "E1743 - E1760",
    url = "https://journals.ametsoc.org/view/journals/bams/101/10/bamsD190331.xml"
}

Community and contributions

We welcome contributions and feedback from anyone, everything from updating the content or appearance of the documentation to new and cutting edge science.

Collaboration
- Anyone interested in scientific collaboration which would add new software functionality should read the MUSICA software development plan.
Contributor's guide
- Before submiitting a PR, please thouroughly read this to you understand our expectations. We reserve the right to reject any PR not meeting our guidelines.

Documentation

Please see the MICM documentation for detailed installation and usage instructions.

License

Apache 2.0