University of Michigan Logo

Official Binaries

Official binaries of HOOMD-blue are available via conda through the glotzer channel. Conda is a python-centric packaging system targeted at computational researchers. The default package database includes many useful packages such as scipy, ipython and scikit-learn. Many more user-provided packages are available through channels on

To install HOOMD-blue, first download and install miniconda following conda's instructions. Then add the glotzer channel and install HOOMD-blue:

$ conda config --add channels glotzer
$ conda install hoomd

If you have already installed hoomd in conda, you can upgrade to the latest version:

$ conda update --all

If you find HOOMD-blue useful in your research, please cite it. See the citation page for instructions.


  • On Linux (64-bit):
    • A recent linux distribution (such as CentOS/RHEL 6, ubuntu 14.04, or newer)
    • [optional] NVIDIA drivers 375.26 or newer
    • [optional] CUDA capable NVIDIA GPU, compute 3.0 or newer.
  • On Mac:
    • OSX 10.8 or newer
    • Note: Mac builds are not MPI or GPU enabled.


Use git to clone from HOOMD-blue's bitbucket repository.

See the user manual for system requirements and build instructions.

Change Log


Released 2017/10/04

Bug fixes

  • Add special pair headers to install target
  • Fix a bug where hpmc.integrate.convex_polyhedron, hpmc.integrate.convex_spheropolyhedron, hpmc.integrate.polyedron, hpmc.integrate.faceted_sphere, hpmc.integrate.sphere_union and hpmc.integrate.convex_polyhedron_union produced spurious overlaps on the GPU


Released 2017/09/08

New features

  • General:

    • Add hoomd.hdf5.log to log quantities in hdf5 format. Matrix quantities can be logged.
    • dump.gsd can now save internal state to gsd files. Call dump_state(object) to save the state for a particular object. The following objects are supported:
      • HPMC integrators save shape and trial move size state.
    • Add dynamic argument to hoomd.dump.gsd to specify which quantity categories should be written every frame.
    • HOOMD now inter-operates with other python libraries that set the active CUDA device.
    • Add generic capability for bidirectional ghost communication, enabling multi body potentials in MPI simulation.
  • MD:

    • Added support for a 3 body potential that is harmonic in the local density.
    • force.constant and can now apply torques.
    • quiet option to nlist.tune to quiet the output of the embedded run() commands.
    • Add special pairs as exclusions from neighbor lists.
    • Add cosine squared angle potential md.angle.cosinesq.
    • Add md.pair.DLVO() for evaluation of colloidal dispersion and electrostatic forces.
    • Add Lennard-Jones 12-8 pair potential.
    • Add Buckingham (exp-6) pair potential.
    • Add Coulomb 1-4 special_pair potential.
    • Check that composite body dimensions are consistent with minimum image convention and generate an error if they are not.
    • md.integrate.mode.minimize_fire() now supports anisotropic particles (i.e. composite bodies)
    • md.integrate.mode.minimize_fire() now supports flexible specification of integration methods
    • md.integrate.npt()/md.integrate.nph() now accept a friction parameter (gamma) for damping out box fluctuations during minimization runs
    • Add new command integrate.mode_standard.reset_methods() to clear NVT and NPT integrator variables
  • HPMC:

    • hpmc.integrate.sphere_union() takes new capacity parameter to optimize performance for different shape sizes
    • hpmc.integrate.polyhedron() takes new capacity parameter to optimize performance for different shape sizes
    • hpmc.integrate.convex_polyhedron and convex_spheropolyhedron now support arbitrary numbers of vertices, subject only to memory limitations (max_verts is now ignored).
    • HPMC integrators restore state from a gsd file read by init.read_gsd when the option restore_state is True.
    • Deterministic HPMC integration on the GPU (optional): mc.set_params(deterministic=True).
    • New hpmc.update.boxmc.ln_volume() move allows logarithmic volume moves for fast equilibration.
    • New shape: hpmc.integrate.convex_polyhedron_union performs simulations of unions of convex polyhedra.
    • hpmc.field.callback() now enables MC energy evaluation in a python function
    • The option depletant_mode='overlap_regions' for hpmc.integrate.* allows the selection of a new depletion algorithm that restores the diffusivity of dilute colloids in dense depletant baths


  • HPMC: hpmc.integrate.sphere_union() no longer needs the max_members parameter.
  • HPMC: hpmc.integrate.convex_polyhedron and convex_spheropolyhedron no longer needs the max_verts parameter.
  • The static argument to hoomd.dump.gsd should no longer be used. Use dynamic instead.

Bug fixes

  • HPMC:

    • hpmc.integrate.sphere_union() and hpmc.integrate.polyhedron() missed overlaps.
    • Fix alignment error when running implicit depletants on GPU with ntrial > 0.
    • HPMC integrators now behave correctly when the user provides different RNG seeds on different ranks.
    • Fix a bug where overlapping configurations were produced with hpmc.integrate.faceted_sphere()
  • MD:

    • charge.pppm() with order=7 now gives correct results
    • The PPPM energy for particles excluded as part of rigid bodies now correctly takes into account the periodic boundary conditions
  • EAM:

    • metal.pair.eam now produces correct results.

Other changes

  • Optimized performance of HPMC sphere union overlap check and polyhedron shape
  • Improved performance of rigid bodies in MPI simulations
  • Support triclinic boxes with rigid bodies
  • Raise an error when an updater is given a period of 0
  • Revised compilation instructions
  • Misc documentation improvements
  • Fully document constrain.rigid
  • -march=native is no longer set by default (this is now a suggestion in the documentation)
  • Compiler flags now default to CMake defaults
  • ENABLE_CUDA and ENABLE_MPI CMake options default OFF. User must explicitly choose to enable optional dependencies.
  • HOOMD now builds on powerpc+CUDA platforms (tested on summitdev)
  • Improve performance of GPU PPPM force calculation
  • Use sphere tree to further improve performance of hpmc.integrate.sphere_union()


Released 2017/08/22

Bug fixes

  • Fix a bug where the log quantity momentum was incorrectly reported in MPI simulations.
  • Raise an error when the user provides inconsistent charge or diameter lists to md.constrain.rigid.
  • Fix a bug where pair.compute_energy() did not report correct results in MPI parallel simulations.
  • Fix a bug where make rigid bodies with anisotropic constituent particles did not work on the GPU.
  • Fix hoomd compilation after the rebase in the cub repository.
  • deprecated.dump.xml() now writes correct results when particles have been added or deleted from the simulation.
  • Fix a critical bug where charge.pppm() calculated invalid forces on the GPU


Released 2017/07/19

Bug fixes

  • init.read_getar now correctly restores static quantities when given a particular frame.
  • Fix bug where many short calls to run() caused incorrect results when using md.integrate.langevin.
  • Fix a bug in the Saru pseudo-random number generator that caused some double-precision values to be drawn outside the valid range [0,1) by a small amount. Both floats and doubles are now drawn on [0,1).
  • Fix a bug where coefficients for multi-character unicode type names failed to process in Python 2.

Other changes

  • The Saru generator has been moved into hoomd/Saru.h, and plugins depending on Saru or SaruGPU will need to update their includes. The SaruGPU class has been removed. Use hoomd::detail::Saru instead for both CPU and GPU plugins.


Released 2017/05/11

Bug fixes

  • Fix PPM exclusion handling on the CPU
  • Handle r_cut for special pairs correctly
  • Fix tauP reference in NPH documentation
  • Fixed constrain.rigid on compute 5.x.
  • Fixed random seg faults when using sqlite getar archives with LZ4 compression
  • Fixed XZ coupling with integration
  • Fixed aspect ratio with non-cubic boxes in hoomd.hpmc.update.boxmc


Released 2017/04/12

Bug fixes

  • Document hpmc.util.tune_npt
  • Fix dump.getar.writeJSON usage with MPI execution
  • Fix a bug where integrate.langevin and integrate.brownian correlated RNGs between ranks in multiple CPU execution
  • Bump CUB to version 1.6.4 for improved performance on Pascal architectures. CUB is now embedded using a git submodule. Users upgrading existing git repositories should reinitialize their git submodules with git submodule update --init
  • CMake no longer complains when it finds a partial MKL installation.


Released 2017/03/09

Bug fixes

  • Fixed a compile error on Mac


Released 2017/03/09

Bug fixes

  • Fixed a bug re-enabling disabled integration methods
  • Fixed a bug where adding particle types to the system failed for anisotropic pair potentials
  • scipy is no longer required to execute DEM component unit tests
  • Issue a warning when a subsequent call to context.initialize is given different arguments
  • DPD now uses the seed from rank 0 to avoid incorrect simulations when users provide different seeds on different ranks
  • Miscellaneous documentation updates
  • Defer initialization message until context.initialize
  • Fixed a problem where a momentary dip in TPS would cause walltime limited jobs to exit prematurely
  • HPMC and DEM components now correctly print citation notices


Released 2017/02/07

Bug fixes

  • Fixed a bug where the WalltimeLimitReached was ignored


Released 2017/01/11

Bug fixes

  • (HPMC) Implicit depletants with spheres and faceted spheres now produces correct ensembles
  • (HPMC) Implicit depletants with ntrial > 0 now produces correct ensembles
  • (HPMC) NPT ensemble in HPMC (hpmc.update.boxmc) now produces correct ensembles
  • Fix a bug where multiple nvt/npt integrators caused warnings from analyze.log.
  • update.balance() is properly ignored when only one rank is available
  • Add missing headers to plugin install build
  • Fix a bug where charge.pppm calculated an incorrect pressure

Other changes

  • Drop support for compute 2.0 GPU devices
  • Support cusolver with CUDA 8.0


Released 2016/10/23

Bug fixes

  • Fix memory allocation bug
  • Quiet Python.h warnigns when building (python 2.7)
  • Allow multi-character particle types in HPMC (python 2.7)
  • Enable dump.getar.writeJSON in MPI
  • Allow the flow to change directions in md.update.mueller_plathe_flow
  • Fix critical bug in MPI communication when using HPMC integrators


Released 2016/10/04

New features

  • enable/disable overlap checks between pairs of constituent particles for hpmc.integrate.sphere_union()
  • Support for non-additive mixtures in HPMC, overlap checks can now be enabled/disabled per type-pair
  • Add md.constrain.oned to constrain particles to move in one dimension
  • hpmc.integrate.sphere_union() now takes max_members as an optional argument, allowing to use GPU memory more efficiently
  • Add md.special_pair.lj() to support scaled 1-4 (or other) exclusions in all-atom force fields
  • md.update.mueller_plathe_flow(): Method to create shear flows in MD simulations
  • use_charge option for md.pair.reaction_field
  • md.charge.pppm() takes a Debye screening length as an optional parameter
  • md.charge.pppm() now computes the rigid body correction to the PPPM energy


  • HPMC: the ignore_overlaps flag is replaced by hpmc.integrate.interaction_matrix

Other changes

  • Optimized MPI simulations of mixed systems with rigid and non-rigid bodies
  • Removed dependency on all boost libraries. Boost is no longer needed to build hoomd
  • Intel compiler builds are no longer supported due to c++11 bugs
  • Shorter compile time for HPMC GPU kernels
  • Include symlinked external components in the build process
  • Add template for external components
  • Optimized dense depletant simulations with HPMC on CPU

Bug fixes

  • fix invalid mesh energy in non-neutral systems with md.charge.pppm()
  • Fix invalid forces in simulations with many bond types (on GPU)
  • fix rare cases where analyze.log() would report a wrong pressure
  • fix possible illegal memory access when using md.constrain.rigid() in GPU MPI simulations
  • fix a bug where the potential energy is misreported on the first step with md.constrain.rigid()
  • Fix a bug where the potential energy is misreported in MPI simulations with md.constrain.rigid()
  • Fix a bug where the potential energy is misreported on the first step with md.constrain.rigid()
  • md.charge.pppm() computed invalid forces
  • Fix a bug where PPPM interactions on CPU where not computed correctly
  • Match logged quantitites between MPI and non-MPI runs on first time step
  • Fix md.pair.dpd and md.pair.dpdlj set_params
  • Fix diameter handling in DEM shifted WCA potential
  • Correctly handle particle type names in lattice.unitcell
  • Validate is consistent across MPI ranks


Released 2016/08/30

  • hpmc.util.tune now works with particle types as documented
  • Fix pressure computation with pair.dpd() on the GPU
  • Fix a bug where dump.dcd corrupted files on job restart
  • Fix a bug where HPMC walls did not work correctly with MPI
  • Fix a bug where stdout/stderr did not appear in MPI execution
  • HOOMD will now report an human readable error when users forget context.initialize()
  • Fix syntax errors in frenkel ladd field


Released 2016/08/09

  • Support CUDA Toolkit 8.0
  • group.rigid()/nonrigid() did not work in MPI simulations
  • Fix builds with ENABLE_DOXYGEN=on
  • Always add -std=c++11 to the compiler command line arguments
  • Fix rare infinite loops when using hpmc.integrate.faceted_sphere
  • Fix hpmc.util.tune to work with more than one tunable
  • Fix a bug where dump.gsd() would write invalid data in simulations with changing number of particles
  • replicate() sometimes did not work when restarting a simulation


Released 2016/07/15

Bug fixes

  • Fix acceptance criterion in mu-V-T simulations with implicit depletants (HPMC).
  • References to disabled analyzers, computes, updaters, etc. are properly freed from the simulation context.
  • Fix a bug where init.read_gsd ignored the restart argument.
  • Report an error when HPMC kernels run out of memory.
  • Fix ghost layer when using rigid constraints in MPI runs.
  • Clarify definition of the dihedral angle.


Released 2016/06/22

HOOMD-blue v2.0 is released under a clean BSD 3-clause license.

New packages

  • dem - simulate faceted shapes with dynamics
  • hpmc - hard particle Monte Carlo of a variety of shape classes.

Bug fixes

  • Angles, dihedrals, and impropers no longer initialize with one default type.
  • Fixed a bug where integrate.brownian gave the same x,y, and z velocity components.
  • Data proxies verify input types and vector lengths.
  • dump.dcd no longer generates excessive metadata traffic on lustre file systems

New features

  • Distance constraints constrain.distance - constrain pairs of particles to a fixed separation distance
  • Rigid body constraints constrain.rigid - rigid bodies now have central particles, and support MPI and replication
  • Multi-GPU electrostatics charge.pppm - the long range electrostatic forces are now supported in MPI runs
  • context.initialize() can now be called multiple times - useful in jupyter notebooks
  • Manage multiple simulations in a single job script with SimulationContext as a python context manager.
  • util.quiet_status() / util.unquiet_status() allow users to control if line status messages are output.
  • Support executing hoomd in Jupyter (ipython) notebooks. Notice, warning, and error messages now show up in the notebook output blocks.
  • analyze.log can now register python callback functions as sources for logged quantities.
  • The GSD file format ( is fully implemented in hoomd
    • dump.gsd writes GSD trajectories and restart files (use truncate=true for restarts).
    • init.read_gsd reads GSD file and initializes the system, and can start the simulation from any frame in the GSD file.
    • data.gsd_snapshot reads a GSD file into a snapshot which can be modified before system initialization with init.read_snapshot.
    • The GSD file format is capable of storing all particle and topology data fields in hoomd, either static at frame 0, or varying over the course of the trajectory. The number of particles, types, bonds, etc. can also vary over the trajectory.
  • applies an active force (optionally with rotational diffusion) to a group of particles
  • update.constrain_ellipsoid constrains particles to an ellipsoid
  • integrate.langevin and integrate.brownian now apply rotational noise and damping to anisotropic particles
  • Support dynamically updating groups. group.force_update() forces the group to rebuild according to the original selection criteria. For example, this can be used to periodically update a cuboid group to include particles only in the specified region.
  • pair.reaction_field implements a pair force for a screened electrostatic interaction of a charge pair in a dielectric medium.
  • force.get_energy allows querying the potential energy of a particle group for a specific force
  • init.create_lattice initializes particles on a lattice.
    • lattice.unitcell provides a generic unit cell definition for create_lattice
    • Convenience functions for common lattices: sq, hex, sc, bcc, fcc.
  • Dump and initialize commands for the GTAR file format (
    • GTAR can store trajectory data in zip, tar, sqlite, or bare directories
    • The current version stores system properties, later versions will be able to capture log, metadata, and other output to reduce the number of files that a job script produces.
  • integrate.npt can now apply a constant stress tensor to the simulation box.
  • Faceted shapes can now be simulated through the dem component.

Changes that require job script modifications

  • context.initialize() is now required before any other hoomd script command.
  • init.reset() no longer exists. Use context.initialize() or activate a SimulationContext.
  • Any scripts that relied on undocumented members of the globals module will fail. These variables have been moved to the context module and members of the currently active SimulationContext.
  • bonds, angles, dihedrals, and impropers no longer use the set_coeff syntax. Use bond_coeff.set, angle_coeff.set, dihedral_coeff.set, and improper_coeff.set instead.
  • hoomd_script no longer exists, python commands are now spread across hoomd,, and other sub packages.
  • integrate.\*_rigid() no longer exists. Use a standard integrator on group.rigid_center(), and define rigid bodies using constrain.rigid()
  • All neighbor lists must be explicitly created using nlist.\*, and each pair potential must be attached explicitly to a neighbor list. A default global neighbor list is no longer created.
  • Moved cgcmm into its own package.
  • Moved eam into the metal package.
  • Integrators now take kT arguments for temperature instead of T to avoid confusion on the units of temperature.
  • phase defaults to 0 for updaters and analyzers so that restartable jobs are more easily enabled by default.
  • dump.xml (deprecated) requires a particle group, and can dump subsets of particles.

Other changes

  • CMake minimum version is now 2.8
  • Convert particle type names to str to allow unicode type name input
  • __version__ is now available in the top level package
  • boost::iostreams is no longer a build dependency
  • boost::filesystem is no longer a build dependency
  • New concepts page explaining the different styles of neighbor lists
  • Default neighbor list buffer radius is more clearly shown to be r_buff = 0.4
  • Memory usage of nlist.stencil is significantly reduced
  • A C++11 compliant compiler is now required to build HOOMD-blue


  • Removed integrate.bdnvt: use integrate.langevin
  • Removed mtk=False option from integrate.nvt - The MTK NVT integrator is now the only implementation.
  • Removed integrate.\*_rigid(): rigid body functionality is now contained in the standard integration methods
  • Removed the global neighbor list, and thin wrappers to the neighbor list in nlist.
  • Removed PDB and MOL2 dump writers.
  • Removed init.create_empty


  • Deprecated
  • Deprecated dump.xml.
  • Deprecated dump.pos.
  • Deprecated init.read_xml.
  • Deprecated init.create_random.
  • Deprecated init.create_random_polymers.

Older versions

Older change log entries are available in the source file