Loading Data

This section contains information on how to load data into yt, as well as some important caveats about different data formats.

ART Data

ART data has been supported in the past by Christopher Moody and is currently cared for by Kenza Arraki. Please contact the yt-dev mailing list if you are interested in using yt for ART data, or if you are interested in assisting with development of yt to work with ART data.

To load an ART dataset you can use the yt.load command and provide it the gas mesh file. It will search for and attempt to find the complementary dark matter and stellar particle header and data files. However, your simulations may not follow the same naming convention.

import yt

ds = yt.load("D9p_500/10MpcBox_HartGal_csf_a0.500.d")

It will search for and attempt to find the complementary dark matter and stellar particle header and data files. However, your simulations may not follow the same naming convention.

For example, the single snapshot given in the sample data has a series of files that look like this:

10MpcBox_HartGal_csf_a0.500.d  #Gas mesh
PMcrda0.500.DAT                #Particle header
PMcrs0a0.500.DAT               #Particle data (positions,velocities)
stars_a0.500.dat               #Stellar data (metallicities, ages, etc.)

The ART frontend tries to find the associated files matching the above, but if that fails you can specify file_particle_header, file_particle_data, and file_particle_stars, in addition to specifying the gas mesh. Note that the pta0.500.dat or pt.dat file containing particle time steps is not loaded by yt.

You also have the option of gridding particles and assigning them onto the meshes. This process is in beta, and for the time being it’s probably best to leave do_grid_particles=False as the default.

To speed up the loading of an ART file, you have a few options. You can turn off the particles entirely by setting discover_particles=False. You can also only grid octs up to a certain level, limit_level=5, which is useful when debugging by artificially creating a ‘smaller’ dataset to work with.

Finally, when stellar ages are computed we ‘spread’ the ages evenly within a smoothing window. By default this is turned on and set to 10Myr. To turn this off you can set spread=False, and you can tweak the age smoothing window by specifying the window in seconds, spread=1.0e7*365*24*3600.

There is currently preliminary support for dark matter only ART data. To load a dataset use the yt.load command and provide it the particle data file. It will search for the complementary particle header file.

import yt

ds = yt.load("PMcrs0a0.500.DAT")

Important: This should not be used for loading just the dark matter data for a ‘regular’ hydrodynamical data set as the units and IO are different!


ARTIO data has a well-specified internal parameter system and has few free parameters. However, for optimization purposes, the parameter that provides the most guidance to yt as to how to manage ARTIO data is max_range. This governs the maximum number of space-filling curve cells that will be used in a single “chunk” of data read from disk. For small datasets, setting this number very large will enable more data to be loaded into memory at any given time; for very large datasets, this parameter can be left alone safely. By default it is set to 1024; it can in principle be set as high as the total number of SFC cells.

To load ARTIO data, you can specify a command such as this:

ds = load("./A11QR1/s11Qzm1h2_a1.0000.art")

Athena Data

Athena 4.x VTK data is mostly supported and cared for by John ZuHone. Both uniform grid and SMR datasets are supported.

Loading Athena datasets is slightly different depending on whether your dataset came from a serial or a parallel run. If the data came from a serial run or you have joined the VTK files together using the Athena tool join_vtk, you can load the data like this:

import yt
ds = yt.load("kh.0010.vtk")

The filename corresponds to the file on SMR level 0, whereas if there are multiple levels the corresponding files will be picked up automatically, assuming they are laid out in lev* subdirectories under the directory where the base file is located.

For parallel datasets, yt assumes that they are laid out in directories named id*, one for each processor number, each with lev* subdirectories for additional refinement levels. To load this data, call load with the base file in the id0 directory:

import yt
ds = yt.load("id0/kh.0010.vtk")

which will pick up all of the files in the different id* directories for the entire dataset.

yt works in cgs (“Gaussian”) units by default, but Athena data is not normally stored in these units. If you would like to convert data to cgs units, you may supply conversions for length, time, and mass to load using the units_override functionality:

import yt

units_override = {"length_unit": (1.0, "Mpc"),
                  "time_unit": (1.0, "Myr"),
                  "mass_unit": (1.0e14, "Msun")}

ds = yt.load("id0/cluster_merger.0250.vtk", units_override=units_override)

This means that the yt fields, e.g. ("gas","density"), ("gas","x-velocity"), ("gas","magnetic_field_x"), will be in cgs units, but the Athena fields, e.g., ("athena","density"), ("athena","velocity_x"), ("athena","cell_centered_B_x"), will be in code units.

Some 3D Athena outputs may have large grids (especially parallel datasets subsequently joined with the join_vtk script), and may benefit from being subdivided into “virtual grids”. For this purpose, one can pass in the nprocs parameter:

import yt

ds = yt.load("sloshing.0000.vtk", nprocs=8)

which will subdivide each original grid into nprocs grids.


Virtual grids are only supported (and really only necessary) for 3D data.

Alternative values for the following simulation parameters may be specified using a parameters dict, accepting the following keys:

  • Gamma: ratio of specific heats, Type: Float
  • geometry: Geometry type, currently accepts "cartesian" or "cylindrical"
  • periodicity: Is the domain periodic? Type: Tuple of boolean values corresponding to each dimension
import yt

parameters = {"gamma":4./3., "geometry":"cylindrical", "periodicity":(False,False,False)}

ds = yt.load("relativistic_jet_0000.vtk", parameters=parameters)


  • yt primarily works with primitive variables. If the Athena dataset contains conservative variables, the yt primitive fields will be generated from the conserved variables on disk.
  • Special relativistic datasets may be loaded, but are not fully supported. In particular, the relationships between quantities such as pressure and thermal energy will be incorrect, as it is currently assumed that their relationship is that of an ideal a \gamma-law equation of state.
  • Domains may be visualized assuming periodicity.
  • Particle list data is currently unsupported.


The old behavior of supplying unit conversions using a parameters dict supplied to load for Athena datasets is still supported, but is being deprecated in favor of units_override, which provides the same functionality.

BoxLib Data

yt has been tested with BoxLib data generated by Orion, Nyx, Maestro and Castro. Currently it is cared for by a combination of Andrew Myers, Chris Malone, Matthew Turk, and Mike Zingale.

To load a BoxLib dataset, you can use the yt.load command on the plotfile directory name. In general, you must also have the inputs file in the base directory, but Maestro and Castro will get all the necessary parameter information from the job_info file in the plotfile directory. For instance, if you were in a directory with the following files:


You would feed it the filename pltgmlcs5600:

import yt
ds = yt.load("pltgmlcs5600")

For Maestro and Castro, you would not need the inputs file, and you would have a job_info file in the plotfile directory.


  • yt does not read the Maestro base state (although you can have Maestro map it to a full Cartesian state variable before writing the plotfile to get around this). E-mail the dev list if you need this support.
  • yt does not know about particles in Maestro.
  • For Maestro, yt aliases either “tfromp” or “tfromh to” temperature depending on the value of the use_tfromp runtime parameter.
  • For Maestro, some velocity fields like velocity_magnitude or mach_number will always use the on-disk value, and not have yt derive it, due to the complex interplay of the base state velocity.

Pluto Data

Support for Pluto AMR data is provided through the Chombo frontend, which is currently maintained by Andrew Myers. Pluto output files that don’t use the Chombo HDF5 format are currently not supported. To load a Pluto dataset, you can use the yt.load command on the *.hdf5 files. For example, the KelvinHelmholtz sample dataset is a directory that contains the following files:


To load it, you can navigate into that directory and do:

import yt
ds = yt.load("data.0004.hdf5")

The pluto.ini file must also be present alongside the HDF5 file. By default, all of the Pluto fields will be in code units.

Enzo Data

Enzo data is fully supported and cared for by Matthew Turk. To load an Enzo dataset, you can use the yt.load command and provide it the dataset name. This would be the name of the output file, and it contains no extension. For instance, if you have the following files:


You would feed the load command the filename DD0010/data0010 as mentioned.

import yt
ds = yt.load("DD0010/data0010")


  • There are no major caveats for Enzo usage
  • Units should be correct, if you utilize standard unit-setting routines. yt will notify you if it cannot determine the units, although this notification will be passive.
  • 2D and 1D data are supported, but the extraneous dimensions are set to be of length 1.0 in “code length” which may produce strange results for volume quantities.

Exodus II Data


To load Exodus II data, you need to have the netcdf4 python interface installed.

Exodus II is a file format for Finite Element datasets that is used by the MOOSE framework for file IO. Support for this format (and for unstructured mesh data in general) is a new feature as of yt 3.3, so while we aim to fully support it, we also expect there to be some buggy features at present. Currently, yt can visualize quads, hexes, triangles, and tetrahedral element types at first order. Additionally, there is experimental support for the high-order visualization of 20-node hex elements. Development of more high-order visualization capability is a work in progress.

To load an Exodus II dataset, you can use the yt.load command on the Exodus II file:

import yt
ds = yt.load("MOOSE_sample_data/out.e-s010", step=0)

Because Exodus II datasets can have multiple steps (which can correspond to time steps, picard iterations, non-linear solve iterations, etc...), you can also specify a step argument when you load an Exodus II data that defines the index at which to look when you read data from the file. Omitting this argument is the same as passing in 0, and setting step=-1 selects the last time output in the file.

You can access the connectivity information directly by doing:

import yt
ds = yt.load("MOOSE_sample_data/out.e-s010", step=-1)

This particular dataset has two meshes in it, both of which are made of 8-node hexes. yt uses a field name convention to access these different meshes in plots and data objects. To see all the fields found in a particular dataset, you can do:

import yt
ds = yt.load("MOOSE_sample_data/out.e-s010")

This will give you a list of field names like ('connect1', 'diffused') and ('connect2', 'convected'). Here, fields labelled with 'connect1' correspond to the first mesh, and those with 'connect2' to the second, and so on. To grab the value of the 'convected' variable at all the nodes in the first mesh, for example, you would do:

import yt
ds = yt.load("MOOSE_sample_data/out.e-s010")
ad = ds.all_data()  # geometric selection, this just grabs everything
print(ad['connect1', 'convected'])

In this dataset, ('connect1', 'convected') is nodal field, meaning that the field values are defined at the vertices of the elements. If we examine the shape of the returned array:

import yt
ds = yt.load("MOOSE_sample_data/out.e-s010")
ad = ds.all_data()
print(ad['connect1', 'convected'].shape)

we see that this mesh has 12480 8-node hexahedral elements, and that we get 8 field values for each element. To get the vertex positions at which these field values are defined, we can do, for instance:

import yt
ds = yt.load("MOOSE_sample_data/out.e-s010")
ad = ds.all_data()
print(ad['connect1', 'vertex_x'])

If we instead look at an element-centered field, like ('connect1', 'conv_indicator'), we get:

import yt
ds = yt.load("MOOSE_sample_data/out.e-s010")
ad = ds.all_data()
print(ad['connect1', 'conv_indicator'].shape)

we instead get only one field value per element.

For information about visualizing unstructured mesh data, including Exodus II datasets, please see Unstructured Mesh Slices and Unstructured Mesh Rendering.

Displacement Fields

Finite element codes often solve for the displacement of each vertex from its original position as a node variable, rather than updating the actual vertex positions with time. For analysis and visualization, it is often useful to turn these displacements on or off, and to be able to scale them arbitrarily to emphasize certain features of the solution. To allow this, if yt detects displacement fields in an Exodus II dataset (using the convention that they will be named disp_x, disp_y, etc...), it will optionally add these to the mesh vertex positions for the purposes of visualization. Displacement fields can be controlled when a dataset is loaded by passing in an optional dictionary to the yt.load command. This feature is turned off by default, meaning that a dataset loaded as

import yt
ds = yt.load("MOOSE_sample_data/mps_out.e")

will not include the displacements in the vertex positions. The displacements can be turned on separately for each mesh in the file by passing in a a tuple of (scale, offset) pairs for the meshes you want to enable displacements for. For example, the following code snippet turns displacements on for the second mesh, but not the first:

import yt
ds = yt.load("MOOSE_sample_data/mps_out.e", step=10,
             displacements={'connect2': (1.0, [0.0, 0.0, 0.0])})

The displacements can also be scaled by an arbitrary factor before they are added in to the vertex positions. The following code turns on displacements for both connect1 and connect2, scaling the former by a factor of 5.0 and the later by a factor of 10.0:

import yt
ds = yt.load("MOOSE_sample_data/mps_out.e", step=10,
             displacements={'connect1': (5.0, [0.0, 0.0, 0.0]),
                            'connect2': (10.0, [0.0, 0.0, 0.0])})

Finally, we can also apply an arbitrary offset to the mesh vertices after the scale factor is applied. For example, the following code scales all displacements in the second mesh by a factor of 5.0, and then shifts each vertex in the mesh by 1.0 unit in the z-direction:

import yt
ds = yt.load("MOOSE_sample_data/mps_out.e", step=10,
              displacements={'connect2': (5.0, [0.0, 0.0, 1.0])})


FITS data is mostly supported and cared for by John ZuHone. In order to read FITS data, AstroPy must be installed. FITS data cubes can be loaded in the same way by yt as other datasets. yt can read FITS image files that have the following (case-insensitive) suffixes:

  • fits
  • fts
  • fits.gz
  • fts.gz

yt can read two kinds of FITS files: FITS image files and FITS binary table files containing positions, times, and energies of X-ray events.


AstroPy is necessary due to the requirements of both FITS file reading and WCS coordinates. Since new releases of PyFITS are to be discontinued, individual installations of this package and the PyWCS package are not supported.

Though a FITS image is composed of a single array in the FITS file, upon being loaded into yt it is automatically decomposed into grids:

import yt
ds = yt.load("m33_hi.fits")
level  # grids         # cells     # cells^3
  0       512          981940800       994
          512          981940800

yt will generate its own domain decomposition, but the number of grids can be set manually by passing the nprocs parameter to the load call:

ds = load("m33_hi.fits", nprocs=1024)

Making the Most of yt for FITS Data

yt will load data without WCS information and/or some missing header keywords, but the resulting field information will necessarily be incomplete. For example, field names may not be descriptive, and units will not be correct. To get the full use out of yt for FITS files, make sure that for each image the following header keywords have sensible values:

  • CDELTx: The pixel width in along axis x
  • CRVALx: The coordinate value at the reference position along axis x
  • CRPIXx: The reference pixel along axis x
  • CTYPEx: The projection type of axis x
  • CUNITx: The units of the coordinate along axis x
  • BTYPE: The type of the image
  • BUNIT: The units of the image

FITS header keywords can easily be updated using AstroPy. For example, to set the BTYPE and BUNIT keywords:

import astropy.io.fits as pyfits
f = pyfits.open("xray_flux_image.fits", mode="update")
f[0].header["BUNIT"] = "cts/s/pixel"
f[0].header["BTYPE"] = "flux"

FITS Coordinates

For FITS datasets, the unit of code_length is always the width of one pixel. yt will attempt to use the WCS information in the FITS header to construct information about the coordinate system, and provides support for the following dataset types:

  1. Rectilinear 2D/3D images with length units (e.g., Mpc, AU, etc.) defined in the CUNITx keywords
  2. 2D images in some celestial coordinate systems (RA/Dec, galactic latitude/longitude, defined in the CTYPEx keywords), and X-ray binary table event files
  3. 3D images with celestial coordinates and a third axis for another quantity, such as velocity, frequency, wavelength, etc.
  4. 4D images with the first three axes like Case 3, where the slices along the 4th axis are interpreted as different fields.

If your data is of the first case, yt will determine the length units based on the information in the header. If your data is of the second or third cases, no length units will be assigned, but the world coordinate information about the axes will be stored in separate fields. If your data is of the fourth type, the coordinates of the first three axes will be determined according to cases 1-3.


Linear length-based coordinates (Case 1 above) are only supported if all dimensions have the same value for CUNITx. WCS coordinates are only supported for Cases 2-4.

Fields in FITS Datasets

Multiple fields can be included in a FITS dataset in several different ways. The first way, and the simplest, is if more than one image HDU is contained within the same file. The field names will be determined by the value of BTYPE in the header, and the field units will be determined by the value of BUNIT. The second way is if a dataset has a fourth axis, with each slice along this axis corresponding to a different field. In this case, the field names will be determined by the value of the CTYPE4 keyword and the index of the slice. So, for example, if BTYPE = "intensity" and CTYPE4 = "stokes", then the fields will be named "intensity_stokes_1", "intensity_stokes_2", and so on.

The third way is if auxiliary files are included along with the main file, like so:

ds = load("flux.fits", auxiliary_files=["temp.fits","metal.fits"])

The image blocks in each of these files will be loaded as a separate field, provided they have the same dimensions as the image blocks in the main file.

Additionally, fields corresponding to the WCS coordinates will be generated. based on the corresponding CTYPEx keywords. When queried, these fields will be generated from the pixel coordinates in the file using the WCS transformations provided by AstroPy.

X-ray event data will be loaded as particle fields in yt, but a grid will be constructed from the WCS information in the FITS header. There is a helper function, setup_counts_fields, which may be used to make deposited image fields from the event data for different energy bands (for an example see FITS X-ray Images in yt).


Each FITS image from a single dataset, whether from one file or from one of multiple files, must have the same dimensions and WCS information as the first image in the primary file. If this is not the case, yt will raise a warning and will not load this field.

Additional Options

The following are additional options that may be passed to the load command when analyzing FITS data:


FITS image data may include NaNs. If you wish to mask this data out, you may supply a nan_mask parameter, which may either be a single floating-point number (applies to all fields) or a Python dictionary containing different mask values for different fields:

# passing a single float
ds = load("m33_hi.fits", nan_mask=0.0)

# passing a dict
ds = load("m33_hi.fits", nan_mask={"intensity":-1.0,"temperature":0.0})


Generally, AstroPy may generate a lot of warnings about individual FITS files, many of which you may want to ignore. If you want to see these warnings, set suppress_astropy_warnings = False.


For some applications, decomposing 3D FITS data into grids that span the x-y plane with short strides along the z-axis may result in a significant improvement in I/O speed. To enable this feature, set z_axis_decomp=True.


Often, the aspect ratio of 3D spectral cubes can be far from unity. Because yt sets the pixel scale as the code_length, certain visualizations (such as volume renderings) may look extended or distended in ways that are undesirable. To adjust the width in code_length of the spectral axis, set spectral_factor equal to a constant which gives the desired scaling, or set it to "auto" to make the width the same as the largest axis in the sky plane.

Miscellaneous Tools for Use with FITS Data

A number of tools have been prepared for use with FITS data that enhance yt’s visualization and analysis capabilities for this particular type of data. These are included in the yt.frontends.fits.misc module, and can be imported like so:

from yt.frontends.fits.misc import setup_counts_fields, PlotWindowWCS, ds9_region


This function can be used to create image fields from X-ray counts data in different energy bands:

ebounds = [(0.1,2.0),(2.0,5.0)] # Energies are in keV
setup_counts_fields(ds, ebounds)

which would make two fields, "counts_0.1-2.0" and "counts_2.0-5.0", and add them to the field registry for the dataset ds.


This function takes a ds9 region and creates a “cut region” data container from it, that can be used to select the cells in the FITS dataset that fall within the region. To use this functionality, the pyregion package must be installed.

ds = yt.load("m33_hi.fits")
circle_region = ds9_region(ds, "circle.reg")


This class takes a on-axis SlicePlot or ProjectionPlot of FITS data and adds celestial coordinates to the plot axes. To use it, the WCSAxes package must be installed.

wcs_slc = PlotWindowWCS(slc)
wcs_slc.show() # for the IPython notebook

WCSAxes is still in an experimental state, but as its functionality improves it will be utilized more here.



The following functionality requires the spectral-cube library to be installed.

If you have a spectral intensity dataset of some sort, and would like to extract emission in particular slabs along the spectral axis of a certain width, create_spectral_slabs can be used to generate a dataset with these slabs as different fields. In this example, we use it to extract individual lines from an intensity cube:

slab_centers = {'13CN': (218.03117, 'GHz'),
                'CH3CH2CHO': (218.284256, 'GHz'),
                'CH3NH2': (218.40956, 'GHz')}
slab_width = (0.05, "GHz")
ds = create_spectral_slabs("intensity_cube.fits",
                                  slab_centers, slab_width,

All keyword arguments to create_spectral_slabs are passed on to load when creating the dataset (see Additional Options above). In the returned dataset, the different slabs will be different fields, with the field names taken from the keys in slab_centers. The WCS coordinates on the spectral axis are reset so that the center of the domain along this axis is zero, and the left and right edges of the domain along this axis are \pm 0.5*slab_width.

Examples of Using FITS Data

The following IPython notebooks show examples of working with FITS data in yt, which we recommend you look at in the following order:


FLASH HDF5 data is mostly supported and cared for by John ZuHone. To load a FLASH dataset, you can use the yt.load command and provide it the file name of a plot file, checkpoint file, or particle file. Particle files require special handling depending on the situation, the main issue being that they typically lack grid information. The first case is when you have a plotfile and a particle file that you would like to load together. In the simplest case, this occurs automatically. For instance, if you were in a directory with the following files:

radio_halo_1kpc_hdf5_plt_cnt_0100 # plotfile
radio_halo_1kpc_hdf5_part_0100 # particle file

where the plotfile and the particle file were created at the same time (therefore having particle data consistent with the grid structure of the former). Notice also that the prefix "radio_halo_1kpc_" and the file number 100 are the same. In this special case, the particle file will be loaded automatically when yt.load is called on the plotfile. This also works when loading a number of files in a time series.

If the two files do not have the same prefix and number, but they nevertheless have the same grid structure and are at the same simulation time, the particle data may be loaded with the particle_filename optional argument to yt.load:

import yt
ds = yt.load("radio_halo_1kpc_hdf5_plt_cnt_0100", particle_filename="radio_halo_1kpc_hdf5_part_0100")

However, if you don’t have a corresponding plotfile for a particle file, but would still like to load the particle data, you can still call yt.load on the file. However, the grid information will not be available, and the particle data will be loaded in a fashion similar to SPH data.


  • Please be careful that the units are correctly utilized; yt assumes cgs by default, but conversion to other unit systems is also possible.

Gadget Data

yt has support for reading Gadget data in both raw binary and HDF5 formats. It is able to access the particles as it would any other particle dataset, and it can apply smoothing kernels to the data to produce both quantitative analysis and visualization. See SPH Particle Data for more details and Using yt to view and analyze Gadget outputs for a detailed example of loading, analyzing, and visualizing a Gadget dataset. An example which makes use of a Gadget snapshot from the OWLS project can be found at Using yt to view and analyze Gadget-OWLS outputs.

Gadget data in HDF5 format can be loaded with the load command:

import yt
ds = yt.load("snapshot_061.hdf5")

Gadget data in raw binary format can also be loaded with the load command. This is only supported for snapshots created with the SnapFormat parameter set to 1 (the standard for Gadget-2).

import yt
ds = yt.load("snapshot_061")

Units and Bounding Boxes

There are two additional pieces of information that may be needed. If your simulation is cosmological, yt can often guess the bounding box and the units of the simulation. However, for isolated simulations and for cosmological simulations with non-standard units, these must be supplied. For example, if a length unit of 1.0 corresponds to a kiloparsec, you can supply this in the constructor. yt can accept units such as Mpc, kpc, cm, Mpccm/h and so on. In particular, note that Mpc/h and Mpccm/h (cm for comoving here) are usable unit definitions.

yt will attempt to use units for mass, length and time as supplied in the argument unit_base. The bounding_box argument is a list of two-item tuples or lists that describe the left and right extents of the particles.

ds = GadgetDataset("snap_004",
        unit_base = {'length': ('kpc', 1.0)},
        bounding_box = [[-600.0, 600.0], [-600.0, 600.0], [-600.0, 600.0]])

Indexing Criteria

yt generates a global mesh index via octree that governs the resolution of volume elements. This is governed by two parameters, n_ref and over_refine_factor. They are weak proxies for each other. The first, n_ref, governs how many particles in an oct results in that oct being refined into eight child octs. Lower values mean higher resolution; the default is 64. The second parameter, over_refine_factor, governs how many cells are in a given oct; the default value of 1 corresponds to 8 cells. The number of cells in an oct is defined by the expression 2**(3*over_refine_factor).

It’s recommended that if you want higher-resolution, try reducing the value of n_ref to 32 or 16.

Also yt can be set to generate the global mesh index according to a specific type of particles instead of all the particles through the parameter index_ptype. For example, to build the octree only according to the "PartType0" particles, you can do:

ds = yt.load("snapshot_061.hdf5", index_ptype="PartType0")

By default, index_ptype is set to "all", which means all the particles. Currently this feature only works for the Gadget HDF5 and OWLS datasets. To bring the feature to other frontends, it’s recommended to refer to this PR for implementation details.

Field Specifications

Binary Gadget outputs often have additional fields or particle types that are non-standard from the default Gadget distribution format. These can be specified in the call to GadgetDataset by either supplying one of the sets of field specifications as a string or by supplying a field specification itself. As an example, yt has built-in definitions for default (the default) and agora_unlv. Field specifications must be tuples, and must be of this format:

default = ( "Coordinates",
            ("InternalEnergy", "Gas"),
            ("Density", "Gas"),
            ("SmoothingLength", "Gas"),

This is the default specification used by the Gadget frontend. It means that the fields are, in order, Coordinates, Velocities, ParticleIDs, Mass, and the fields InternalEnergy, Density and SmoothingLength only for Gas particles. So for example, if you have defined a Metallicity field for the particle type Halo, which comes right after ParticleIDs in the file, you could define it like this:

my_field_def = ( "Coordinates",
            ("Metallicity", "Halo"),
            ("InternalEnergy", "Gas"),
            ("Density", "Gas"),
            ("SmoothingLength", "Gas"),

To save time, you can utilize the plugins file for yt and use it to add items to the dictionary where these definitions are stored. You could do this like so:

from yt.frontends.gadget.definitions import gadget_field_specs
gadget_field_specs["my_field_def"] = my_field_def

Please also feel free to issue a pull request with any new field specifications, as we’re happy to include them in the main distribution!

Particle Type Definitions

In some cases, research groups add new particle types or re-order them. You can supply alternate particle types by using the keyword ptype_spec to the GadgetDataset call. The default for Gadget binary data is:

( "Gas", "Halo", "Disk", "Bulge", "Stars", "Bndry" )

You can specify alternate names, but note that this may cause problems with the field specification if none of the names match old names.

Header Specification

If you have modified the header in your Gadget binary file, you can specify an alternate header specification with the keyword header_spec. This can either be a list of strings corresponding to individual header types known to yt, or it can be a combination of strings and header specifications. The default header specification (found in yt/frontends/sph/definitions.py) is:

default      = (('Npart', 6, 'i'),
                ('Massarr', 6, 'd'),
                ('Time', 1, 'd'),
                ('Redshift', 1, 'd'),
                ('FlagSfr', 1, 'i'),
                ('FlagFeedback', 1, 'i'),
                ('Nall', 6, 'i'),
                ('FlagCooling', 1, 'i'),
                ('NumFiles', 1, 'i'),
                ('BoxSize', 1, 'd'),
                ('Omega0', 1, 'd'),
                ('OmegaLambda', 1, 'd'),
                ('HubbleParam', 1, 'd'),
                ('FlagAge', 1, 'i'),
                ('FlagMEtals', 1, 'i'),
                ('NallHW', 6, 'i'),
                ('unused', 16, 'i'))

These items will all be accessible inside the object ds.parameters, which is a dictionary. You can add combinations of new items, specified in the same way, or alternately other types of headers. The other string keys defined are pad32, pad64, pad128, and pad256 each of which corresponds to an empty padding in bytes. For example, if you have an additional 256 bytes of padding at the end, you can specify this with:

header_spec = ["default", "pad256"]

This can then be supplied to the constructor. Note that you can also do this manually, for instance with:

header_spec = ["default", (('some_value', 8, 'd'),
                           ('another_value', 1, 'i'))]

The letters correspond to data types from the Python struct module. Please feel free to submit alternate header types to the main yt repository.

Specifying Units

If you are running a cosmology simulation, yt will be able to guess the units with some reliability. However, if you are not and you do not specify a dataset, yt will not be able to and will use the defaults of length being 1.0 Mpc/h (comoving), velocity being in cm/s, and mass being in 10^10 Msun/h. You can specify alternate units by supplying the unit_base keyword argument of this form:

unit_base = {'length': (1.0, 'cm'), 'mass': (1.0, 'g'), 'time': (1.0, 's')}

yt will utilize length, mass and time to set up all other units.


GAMER HDF5 data is supported and cared for by Hsi-Yu Schive. You can load the data like this:

import yt
ds = yt.load("InteractingJets/jet_000002")

Currently GAMER does not assume any unit for non-cosmological simulations. To specify the units for yt, you need to supply conversions for length, time, and mass to load using the units_override functionality:

import yt
code_units = { "length_unit":(1.0,"kpc"),
               "time_unit"  :(3.08567758096e+13,"s"),
               "mass_unit"  :(1.4690033e+36,"g") }
ds = yt.load("InteractingJets/jet_000002", units_override=code_units)

This means that the yt fields, e.g., ("gas","density"), will be in cgs units, but the GAMER fields, e.g., ("gamer","Dens"), will be in code units.


  • GAMER data in raw binary format (i.e., OPT__OUTPUT_TOTAL = C-binary) is not supported.

Generic AMR Data

See Loading Generic Array Data and load_amr_grids() for more detail.

It is possible to create native yt dataset from Python’s dictionary that describes set of rectangular patches of data of possibly varying resolution.

import yt

grid_data = [
    dict(left_edge = [0.0, 0.0, 0.0],
         right_edge = [1.0, 1.0, 1.],
         level = 0,
         dimensions = [32, 32, 32],
         number_of_particles = 0)
    dict(left_edge = [0.25, 0.25, 0.25],
         right_edge = [0.75, 0.75, 0.75],
         level = 1,
         dimensions = [32, 32, 32],
         number_of_particles = 0)

for g in grid_data:
    g["density"] = np.random.random(g["dimensions"]) * 2**g["level"]

ds = yt.load_amr_grids(grid_data, [32, 32, 32], 1.0)

Particle fields are supported by adding 1-dimensional arrays and setting the number_of_particles key to each grid‘s dict:

for g in grid_data:
    g["number_of_particles"] = 100000
    g["particle_position_x"] = np.random.random((g["number_of_particles"]))


  • Some functions may behave oddly, and parallelism will be disappointing or non-existent in most cases.
  • No consistency checks are performed on the index
  • Data must already reside in memory.
  • Consistency between particle positions and grids is not checked; load_amr_grids assumes that particle positions associated with one grid are not bounded within another grid at a higher level, so this must be ensured by the user prior to loading the grid data.

Generic Array Data

See Loading Generic Array Data and load_uniform_grid() for more detail.

Even if your data is not strictly related to fields commonly used in astrophysical codes or your code is not supported yet, you can still feed it to yt to use its advanced visualization and analysis facilities. The only requirement is that your data can be represented as one or more uniform, three dimensional numpy arrays. Assuming that you have your data in arr, the following code:

import yt

data = dict(Density = arr)
bbox = np.array([[-1.5, 1.5], [-1.5, 1.5], [1.5, 1.5]])
ds = yt.load_uniform_grid(data, arr.shape, 3.08e24, bbox=bbox, nprocs=12)

will create yt-native dataset ds that will treat your array as density field in cubic domain of 3 Mpc edge size (3 * 3.08e24 cm) and simultaneously divide the domain into 12 chunks, so that you can take advantage of the underlying parallelism.

Particle fields are detected as one-dimensional fields. The number of particles is set by the number_of_particles key in data. Particle fields are then added as one-dimensional arrays in a similar manner as the three-dimensional grid fields:

import yt

data = dict(Density = dens,
            number_of_particles = 1000000,
            particle_position_x = posx_arr,
            particle_position_y = posy_arr,
            particle_position_z = posz_arr)
bbox = np.array([[-1.5, 1.5], [-1.5, 1.5], [1.5, 1.5]])
ds = yt.load_uniform_grid(data, arr.shape, 3.08e24, bbox=bbox, nprocs=12)

where in this example the particle position fields have been assigned. number_of_particles must be the same size as the particle arrays. If no particle arrays are supplied then number_of_particles is assumed to be zero.


  • Particles may be difficult to integrate.
  • Data must already reside in memory.

Semi-Structured Grid Data

See Loading Generic Array Data, hexahedral_connectivity(), load_hexahedral_mesh() for more detail.

In addition to uniform grids as described above, you can load in data with non-uniform spacing between datapoints. To load this type of data, you must first specify a hexahedral mesh, a mesh of six-sided cells, on which it will live. You define this by specifying the x,y, and z locations of the corners of the hexahedral cells. The following code:

import yt
import numpy

xgrid = numpy.array([-1, -0.65, 0, 0.65, 1])
ygrid = numpy.array([-1, 0, 1])
zgrid = numpy.array([-1, -0.447, 0.447, 1])

coordinates,connectivity = yt.hexahedral_connectivity(xgrid,ygrid,zgrid)

will define the (x,y,z) coordinates of the hexahedral cells and information about that cell’s neighbors such that the cell corners will be a grid of points constructed as the Cartesion product of xgrid, ygrid, and zgrid.

Then, to load your data, which should be defined on the interiors of the hexahedral cells, and thus should have the shape, (len(xgrid)-1, len(ygrid)-1, len(zgrid)-1), you can use the following code:

bbox = numpy.array([[numpy.min(xgrid),numpy.max(xgrid)],
data = {"density" : arr}
ds = yt.load_hexahedral_mesh(data,conn,coords,1.0,bbox=bbox)

to load your data into the dataset ds as described above, where we have assumed your data is stored in the three-dimensional array arr.


  • Integration is not implemented.
  • Some functions may behave oddly or not work at all.
  • Data must already reside in memory.

Unstructured Grid Data

See Loading Generic Array Data, load_unstructured_mesh() for more detail.

In addition to the above grid types, you can also load data stored on unstructured meshes. This type of mesh is used, for example, in many finite element calculations. Currently, hexahedral and tetrahedral mesh elements are supported.

To load an unstructured mesh, you need to specify the following. First, you need to have a coordinates array, which should be an (L, 3) array that stores the (x, y, z) positions of all of the vertices in the mesh. Second, you need to specify a connectivity array, which describes how those vertices are connected into mesh elements. The connectivity array should be (N, M), where N is the number of elements and M is the connectivity length, i.e. the number of vertices per element. Finally, you must also specify a data dictionary, where the keys should be the names of the fields and the values should be numpy arrays that contain the field data. These arrays can either supply the cell-averaged data for each element, in which case they would be (N, 1), or they can have node-centered data, in which case they would also be (N, M).

Here is an example of how to load an in-memory, unstructured mesh dataset:

import yt
import numpy
from yt.utilities.exodusII_reader import get_data

coords, connectivity, data = get_data("MOOSE_sample_data/out.e-s010")

This uses a publically available MOOSE <http://mooseframework.org/> dataset along with the get_data function to parse the coords, connectivity, and data. Then, these can be loaded as an in-memory dataset as follows:

mesh_id = 0
ds = yt.load_unstructured_mesh(data[mesh_id], connectivity[mesh_id], coords[mesh_id])

Note that load_unstructured_mesh can take either a single or a list of meshes. Here, we have selected only the first mesh to load.


  • Integration is not implemented.
  • Some functions may behave oddly or not work at all.
  • Data must already reside in memory.

Generic Particle Data

See Loading Generic Particle Data and load_particles() for more detail.

You can also load generic particle data using the same stream functionality discussed above to load in-memory grid data. For example, if your particle positions and masses are stored in positions and massess, a vertically-stacked array of particle x,y, and z positions, and a 1D array of particle masses respectively, you would load them like this:

import yt

data = dict(particle_position=positions, particle_mass=masses)
ds = yt.load_particles(data)

You can also load data using 1D x, y, and z position arrays:

import yt

data = dict(particle_position_x=posx,
ds = yt.load_particles(data)

The load_particles function also accepts the following keyword parameters:

The units used for particle positions.
The units of the particle masses.
The units used to represent times. This is optional and is only used if your data contains a creation_time field or a particle_velocity field.
The units used to represent velocities. This is optional and is only used if you supply a velocity field. If this is not supplied, it is inferred from the length and time units.
The bounding box for the particle positions.

Gizmo Data

Gizmo datasets, including FIRE outputs, can be loaded into yt in the usual manner. Like other SPH data formats, yt loads Gizmo data as particle fields and then uses smoothing kernels to deposit those fields to an underlying grid structure as spatial fields as described in Gadget Data. To load Gizmo datasets using the standard HDF5 output format:

import yt
ds = yt.load("snapshot_600.hdf5")

Because the Gizmo output format is similar to the Gadget format, yt may load Gizmo datasets as Gadget depending on the circumstances, but this should not pose a problem in most situations. FIRE outputs will be loaded accordingly due to the number of metallicity fields found (11 or 17).

For Gizmo outputs written as raw binary outputs, you may have to specify a bounding box, field specification, and units as are done for standard Gadget outputs. See Gadget Data for more information.

Halo Catalog Data

yt has support for reading halo catalogs produced by Rockstar and the inline FOF/SUBFIND halo finders of Gadget and OWLS. The halo catalogs are treated as particle datasets where each particle represents a single halo. For example, this means that the particle_mass field refers to the mass of the halos. For Gadget FOF/SUBFIND catalogs, the member particles for a given halo can be accessed by creating halo data containers. See Halo Data Containers for more information.

If you have access to both the halo catalog and the simulation snapshot from the same redshift, additional analysis can be performed for each halo using Halo Finding and Analysis.


Rockstar halo catalogs are loaded by providing the path to one of the .bin files. In the case where multiple files were produced, one need only provide the path to a single one of them. The field type for all fields is “halos”. Some fields of note available from Rockstar are:

Rockstar field yt field name
halo id particle_identifier
virial mass particle_mass
virial radius virial_radius
halo position particle_position_(x,y,z)
halo velocity particle_velocity_(x,y,z)

Numerous other Rockstar fields exist. To see them, check the field list by typing ds.field_list for a dataset loaded as ds. Like all other datasets, fields must be accessed through Data Objects.

import yt
ds = yt.load("rockstar_halos/halos_0.0.bin")
ad = ds.all_data()
# halo masses
print(ad["halos", "particle_mass"])
# halo radii
print(ad["halos", "virial_radius"])


Gadget FOF/SUBFIND halo catalogs work in the same way as those created by Rockstar, except there are two field types: FOF for friend-of-friends groups and Subhalo for halos found with the SUBFIND substructure finder. Also like Rockstar, there are a number of fields specific to these halo catalogs.

FOF/SUBFIND field yt field name
halo id particle_identifier
halo mass particle_mass
halo position particle_position_(x,y,z)
halo velocity particle_velocity_(x,y,z)
num. of particles particle_number
num. of subhalos subhalo_number (FOF only)

Many other fields exist, especially for SUBFIND subhalos. Check the field list by typing ds.field_list for a dataset loaded as ds. Like all other datasets, fields must be accessed through Data Objects.

import yt
ds = yt.load("gadget_fof_halos/groups_042/fof_subhalo_tab_042.0.hdf5")
ad = ds.all_data()
# The halo mass
print(ad["Group", "particle_mass"])
print(ad["Subhalo", "particle_mass"])
# Halo ID
print(ad["Group", "particle_identifier"])
print(ad["Subhalo", "particle_identifier"])
# positions
print(ad["Group", "particle_position_x"])
# velocities
print(ad["Group", "particle_velocity_x"])

Multidimensional fields can be accessed through the field name followed by an underscore and the index.

# x component of the spin
print(ad["Subhalo", "SubhaloSpin_0"])

Halo Data Containers

Halo member particles are accessed by creating halo data containers with the type of halo (“Group” or “Subhalo”) and the halo id. Scalar values for halos can be accessed in the same way. Halos also have mass, position, and velocity attributes.

halo = ds.halo("Group", 0)
# member particles for this halo
print halo["member_ids"]
# halo virial radius
print halo["Group_R_Crit200"]
# halo mass
print halo.mass

Subhalos containers can be created using either their absolute ids or their subhalo ids.

# first subhalo of the first halo
subhalo = ds.halo("Subhalo", (0, 0))
# this subhalo's absolute id
print subhalo.group_identifier
# member particles
print subhalo["member_ids"]


OWLS halo catalogs have a very similar structure to regular Gadget halo catalogs. The two field types are FOF and SUBFIND. See Gadget FOF/SUBFIND for more information. At this time, halo member particles cannot be loaded.

import yt
ds = yt.load("owls_fof_halos/groups_008/group_008.0.hdf5")
ad = ds.all_data()
# The halo mass
print(ad["FOF", "particle_mass"])

PyNE Data

PyNE is an open source nuclear engineering toolkit maintained by the PyNE developement team (pyne-dev@googlegroups.com). PyNE meshes utilize the Mesh-Oriented datABase (MOAB) and can be Cartesian or tetrahedral. In addition to field data, pyne meshes store pyne Material objects which provide a rich set of capabilities for nuclear engineering tasks. PyNE Cartesian (Hex8) meshes are supported by yt.

To create a pyne mesh:

from pyne.mesh import Mesh
num_divisions = 50
coords = linspace(-1, 1, num_divisions)
m = Mesh(structured=True, structured_coords=[coords, coords, coords])

Field data can then be added:

from pyne.mesh import iMeshTag
m.neutron_flux = IMeshTag()
# neutron_flux_data is a list or numpy array of size num_divisions^3
m.neutron_flux[:] = neutron_flux_data

Any field data or material data on the mesh can then be viewed just like any other yt dataset!

import yt
pf = yt.frontends.moab.data_structures.PyneMoabHex8Dataset(m)
s = yt.SlicePlot(pf, 'z', 'neutron_flux')


In yt-3.0, RAMSES data is fully supported. If you are interested in taking a development or stewardship role, please contact the yt-dev mailing list. To load a RAMSES dataset, you can use the yt.load command and provide it the info*.txt filename. For instance, if you were in a directory with the following files:


You would feed it the filename output_00007/info_00007.txt:

import yt
ds = yt.load("output_00007/info_00007.txt")

yt will attempt to guess the fields in the file. You may also specify a list of fields by supplying the fields keyword in your call to load.

SPH Particle Data

For all of the SPH frontends, yt uses cython-based SPH smoothing onto an in-memory octree to create deposited mesh fields from individual SPH particle fields.

This uses a standard M4 smoothing kernel and the smoothing_length field to calculate SPH sums, filling in the mesh fields. This gives you the ability to both track individual particles (useful for tasks like following contiguous clouds of gas that would be require a clump finder in grid data) as well as doing standard grid-based analysis (i.e. slices, projections, and profiles).

The smoothing_length variable is also useful for determining which particles can interact with each other, since particles more distant than twice the smoothing length do not typically see each other in SPH simulations. By changing the value of the smoothing_length and then re-depositing particles onto the grid, you can also effectively mimic what your data would look like at lower resolution.

Tipsy Data

See Using yt to view and analyze Tipsy outputs from Gasoline and SPH Particle Data for more details.

yt also supports loading Tipsy data. Many of its characteristics are similar to how Gadget data is loaded; specifically, it shares its definition of indexing and mesh-identification with that described in Indexing Criteria.

ds = load("./halo1e11_run1.00400")

Specifying Tipsy Cosmological Parameters and Setting Default Units

Cosmological parameters can be specified to Tipsy to enable computation of default units. For example do the following, to load a Tipsy dataset whose path is stored in the variable my_filename with specified cosmology parameters:

cosmology_parameters = {'current_redshift': 0.0,
                        'omega_lambda': 0.728,
                        'omega_matter': 0.272,
                        'hubble_constant': 0.702}

ds = yt.load(my_filename,

If you wish to set the unit system directly, you can do so by using the unit_base keyword in the load statement.

import yt

ds = yt.load(filename, unit_base={'length', (1.0, 'Mpc')})

See the documentation for the TipsyDataset class for more information.

Loading Cosmological Simulations

If you are not using a parameter file (i.e. non-Gasoline users), then you must use keyword cosmology_parameters when loading your data set to indicate to yt that it is a cosmological data set. If you do not wish to set any non-default cosmological parameters, you may pass an empty dictionary.

import yt
ds = yt.load(filename, cosmology_parameters={})