Creating Cosmological Initial Conditions

Enzo does not, by itself, generate cosmological initial conditions. Instead, it relies on other packages to do so. Here we describe a variety of external packages, beginning with MUSIC, a widely used and flexible method for generating “uniform” or “zoomed” initial condition files that can be read by Enzo. We also describe the original mechanism, inits, has long been distributed with Enzo. It is exclusively serial. We also now distribute mpgrafic with modifications to support Enzo data formats.


The MUSIC package is described in Hahn & Abel (2011) and is available from a BitBucket repository here. After you download and compiled it (note that MUSIC has its own depenencies including FFTW and GSL), you should have the MUSIC executable.

To use this to generate an enzo initial condition file, you can either modify the MUSIC example parameter file, or use the sample one here: provided music_example.conf which can be used to set a small (16^3) unigrid run. To generate the initial conditions file, use:

./MUSIC music_example.conf

which should generate a directory called ic.enzo, containing a number of files. MUSIC will generate the baryon densities, velocities as well as dark matter particle displacements and velocities. Finally, it will also generate a rudimentary enzo parameter file parameter_file.txt. You can modify this (text) file to change or add any enzo parameters (although it does contain all the basic settings to set the grid and cosmology parameters correctly. To run the enzo simulation, you will need to copy enzo.exe into that directory and run enzo:

[mpirun] enzo.exe parameter_file.txt

A more complicated MUSIC initial parameter file example for a zoomed simulation (one in which one or more static refined regions are created from the beginning, designed to model a small region at high resolution) can be found here: music_example_zoom.conf

Using inits

Inits can also generate 2D cosmological initial conditions. The inits program uses one or more ASCII input files to set parameters, including the details of the power spectrum, the grid size, and output file names. Each line of the parameter file is interpreted independently and can contain only a single parameter. Parameters are specified in the form:

ParameterName = VALUE

Spaces are ignored, and a parameter statement must be contained on a single line. Lines which begin with the pound symbol (#) are assumed to be comments and ignored.

First, set the parameters in the file. There are a large number of parameters, but many don’t need to be set since reasonable default values are provided. Modifying a provided example (see the sample files below) is probably the easiest route, but for reference there is a list of the parameters, their meanings, and their default values.

Generating a single grid initialization (for simple Enzo runs) is relatively straightforward. Generating a multi-grid initialization for Enzo is somewhat more complicated, and we only sketch the full procedure here.

Single Grid Initialization

To run a single grid initialization, you must set at least the following parameters: Rank, GridDims, ParticleDims, as well as the appropriate Cosmology and Power Spectrum parameters. A sample parameter file is available, which sets up a small, single grid cosmology simulation (that is, single grid for the initial conditions, once Enzo is used, additional grids will be created). It is available here: gas_plus_dm.inits and the corresponding enzo parameter file that can be used to run a simulation with those initial conditions is here:

After creating or modifying a parameter file, and compiling inits, run the code with: gas_plus_dm_amr_adia.enzo

inits [-d] parameter_file

Where parameter_file is the name of your modified parameter file (the -d turns on a debug option). This will produce a number of HDF files containing the initial grids and particles, which are in the correct units for use in Enzo.

For example, to use the provided sample files, you would use the following commands:

inits gas_plus_dm.inits
[mpirun] enzo gas_plus_dm_amr_adia.enzo

Multiple-grid Initialization

New in version 2.1.

The multi-grid (or nested) initialization can be used to refine in a specific region, such as the Lagrangian sphere of a halo. We assume that you have first run a single-grid simulation and identified a region out of which a halo will form and can put this in the form of the left and right corners of a box which describes the region. Then you add the following parameters to the single-grid initialization code:

MaximumInitialRefinementLevel = 2
RefineRegionLeftEdge          = 0.15523 0.14551 0.30074
RefineRegionRightEdge         = 0.38523 0.37551 0.53074
NewCenterFloat                = 0.270230055 0.260508984 0.415739357
AutomaticSubgridBuffer        = 4

MaximumInitialRefinementLevel indicates how many extra levels you want to generate (in this case two additional levels, or 3 in total, including the root grid). The next two parameters (RefineRegionLeftEdge and RefineRegionRightEdge) describe the region to be refined. The fourth (optional) parameter re-centers the grid on the halo to be resimulated. The fifth parameter (AutomaticSubgridBuffer) indicates how many course cells should be added around each refined region.

Once you have added these parameters, run inits once on the new parameter file in the standard way:

inits [-d] MultiGridParameterFile

It will give you a progress report as it runs (note that if MaximumInitialRefinementLevel is large, this can take a long time), and generate all of the necessary files (e.g. GridDensity.0, GridDensity.1, etc.).

It will also generate a file called EnzoMultigridParameters which you can then copy directly into the enzo parameter file, and it specifies the positions of the new grids. You will still need to set a few other parameters in the enzo parameter file, including RefineRegionLeftEdge and RefineRegionRightEdge so that it only refines in the specified region (typically this should match the most refined initial grid). Also set the MaximumRefinementLevel parameter and the parameter controlling the density to be refined (MinimumOverDensityForRefinement – this also applies to the root grid, so it needs to be divided by 8^l where l is the value of MaximumInitialRefinementLevel).

Note that it is also possible to generate each level of initial conditions manually. This should not really be necessary, but a rough guideline is given here. To do this, prepare multiple parameter file describing the individual parameter regions, and then top grid can be generated with:

inits [-d] -s SubGridParameterFile TopGridParameterFile

The -s flag provides the name of the sub-grid parameter file, which is required by inits so that the particles are not replicated in the sub-grid region. The sub-grids are made with the usual command line:

inits [-d] SubGridParameterFile

!Subgrids with MaxDims of 512 or larger will take some time and require a fair amount of memory since the entire region is generated and then the desired section extracted.

Inits Parameter List

Cosmology Parameters

This is the contribution of all non-relativistic matter (including HDM) to the energy density at the current epoch (z=0), relative to the value required to marginally close the universe. It includes dark and baryonic matter. Default: 1.0
This is the contribution of the cosmological constant to the energy density at the current epoch, in the same units as above. Default: 0.0
This is the contribution due to warm dark matter alone. Ignored unless PowerSpectrumType = 13 or 14. Default: 0.0
This is the contribution due to hot dark matter alone. Default: 0.0
The baryonic contribution alone. Default: 0.06
The size of the volume to be simulated in Mpc/h (at z=0). Default: 64.0
The Hubble constant at z=0, in units of 100 km/s/Mpc. Default: 0.5
The redshift for which the initial conditions are to be generated. Default: 20.0

Power Spectrum Parameters


This integer parameter indicates the routine to be used for generating the power spectrum. Default: 1 The following are currently available:

  • 1 - CDM approximation from BBKS (Bardeen et al 1986) as modified by Peacock and Dodds (1994), to include, very roughly, the effect of baryons. This should not be used for high baryon universes or for simulations in which precision in the PS is important.
  • 2 - CHDM approximate PS from Ma (1996). Roughly good for hot fractions from 0.05 to 0.3.
  • 3 - Power-law (scale-free) spectra.
  • 4 - Reads in a power-spectrum from a file (not working).
  • 5 - CHDM approximate PS from Ma (1996), modified for 2 equal mass neutrinos.
  • 6 - A CDM-like Power spectrum with a shape parameter (Gamma), that is specified by the parameter PowerSpectrumGamma.
  • 11 - The Eisenstein and Hu fitting functions for low and moderate baryon fraction, including the case of one massive neutrino.
  • 12 - The Eisenstein and Hu fitting functions for low and moderate baryon fraction, for the case of two massive neutrinos.
  • 13 - A Warm Dark Matter (WDM) power spectrum based on the formulae of Bode et al. (2001 ApJ 556, 93). The WDM equivalent of the Eisenstein & Hu fitting function with one massive neutrino (so a WDM version of #11).
  • 14 - A Warm Dark Matter (WDM) power spectrum based on the formulae of Bode et al. (2001 ApJ 556, 93). The WDM equivalent of the CDM BBKS approximation of Bardeen et al 1986 (the WDM version of #1).
  • 20 - A transfer function from CMBFast is input for this option, based on the filenames described below.
The amplitude of the linear power spectrum at z=0 as specified by the rms amplitude of mass-fluctuations in a top-hat sphere of radius 8 Mpc/h. Default: 0.6
This is the index of the mass power spectrum before modification by the transfer function. A value of 1 corresponds to the scale-free primordial spectrum. Default: 1.0.
This is the initial seed for all random number generation, which should be negative. The random number generator (Numerical Recipes RAN3) is machine-independent, so the same seed will produce the same results (with other parameters unchanged). Note also that because the spectrum is sampled strictly in order of increasing k-amplitude, the large-scale power will be the same even if you increase or decrease the grid size. Default: -123456789
The spectrum is set to zero above this wavenumber (i.e. smaller scales are set to zero), which is in units of 1/Mpc. It only works for power spectrum types 1-6. A value of 0 means no cutoff. Default: 0.0
These two parameters control the range of the internal lookup table in wavenumber (units 1/Mpc). Reasonably sized grids will not require changes in these parameters. Defaults: kmin = 1e-3, kmax = 1e+4.
This sets the number of points in the PS look-up table that is generated for efficiency purposes. It should not require changing. Default: 10000.
For input power spectra, such as those from CMBFAST, two transfer functions are required: one at z=0 to fix the amplitude (via Sigma8) and the other at the initial redshift to give the shape and amplitude relative to z=0. No default.
see above.
The shape parameter (Omega*h); ignored unless PowerSpectrumType = 6.
The mass of the dark matter particle in KeV for the Bode et al. warm dark matter (WDM) case. Ignored unless PowerSpectrumType = 13 or 14. Default: 1.0.
The number of degrees of freedom of the warm dark matter particles for the Bode et al. warm dark matter model. Ignored unless PowerSpectrumType = 13 or 14. Default: 1.5.
The shape parameter (Omega*h); ignored unless PowerSpectrumType = 6.

Grid Parameters: Basic

Dimensionality of the problem, 1 to 3 (warning: not recently tested for Rank !=2). Default: 3
This sets the actual dimensions of the baryon grid that is to be created (and so it may be smaller than MaxDims in some cases). Example: 64 64 64 No default.
Dimensions of the particle grid that is to be created. No default.
Flag indicating if the baryon grids should be produced (set to 0 if inits is being run to generate particles only). Default: 1
Flag indicating if the particles should be produced (set to 0 if inits is being run to generate baryons only). Default: 1
This is the name of the particle position output file. This HDF file contains one to three Scientific Data Sets (SDS), one for dimensional component. Default: ParticlePositions
The particle velocity file name, which must(!) be different from the one above, otherwise the order of the SDS’s will be incorrect. Default: ParticleVelocities
This is the name of the particle mass file, which is generally not needed (enzo generates its own masses if not provided). Default: None
The name of the HDF file which contains the grid density SDS. Default: GridDensity
The name of the HDF file which contains the SDS’s for the baryonic velocity (may be the same as GridDensityName). Default: GridVelocity

Grid Parameters: Advanced

Used for multi-grid (nested) initial code generation. This parameter speciesi the level (0-based) that the initial conditions should be generated to. So, for example, setting it to 1 generates the top grid and one additional level of refinement. Note that the additional levels are nested, keeping at least one coarse cell between the edge of a coarse grid and its refined grid. Default: 0
RefineRegionLeftEdge, RefineRegionRightEdge
Species the left and right corners of the region that should be refined using the AutomaticSubgridGeneration method (see above parameter). Default: 0 0 0 - 1 1 1
Indicates that the final grid should be recenter so that this point is the new center (0.5 0.5 0.5) of the grid.
For multi-grid (nested) initial code generation (with the above parameters). This parameter controls how many coarse cells are added around each refined region as buffer zones. The value of 1 is probably ok, but larger values (4?) are probably safer. Default: 1
All dimensions are specified as one to three numbers deliminated by spaces (and for those familiar with the KRONOS or ZEUS method of specifying dimensions, the ones here do not include ghost zones). An example is: 64 64 64. MaxDims are the dimensions of the conceptual high-resolution grid that covers the entire computational domain. For a single-grid initialization this is just the dimension of the grid (or of the particle grid if there are more particles than grid points). For multi-grid initializations, this is the dimensions of the grid that would cover the region at the highest resolution that will be used. It must be identical across all parameter files (for multi-grid initializations). The default is the maximum of GridDims or ParticleDims, whichever is larger (in other words unless you are using a multi-grid initialization, this parameter does not need to be set). Confused yet?
This integer is the sampling, for the baryon grid, in each dimension, relative to MaxDims. For single-grid initializations, this is generally 1. For multi-grids, it is the refinement factor relative to the finest level. In other words, if the grid covered the entire computational region, then each value in MaxDims would equal GridDims times the GridRefinement factor. Default: 1
Similar function as above, but for the particles. Note that it can also be used to generate fewer particles than grids (i.e. the GridRefinement and ParticleRefinement factors do not have to be the same). Default: 1
For single-grid initializations, this should be the zero vector. For multi-grid initializations it specifies the index (a triplet of integers in 3D) of the left-hand corner of the grid to be generated. It is specified in terms of the finest conceptual grid and so ranges from 0 to MaxDims-1. Note also that for AMR, the start and end of a sub-grid must lie on the cell-boundary of it’s parent. That means that this number must be divisible by the Refinement factor. The end of the sub-grid will be at index: StartIndex + GridRefinement*GridDims. The co-ordinate system used by this parameter is always the unshifted one (i.e. it does not change if NewCenter is set).

Using mpgrafic

New in version 2.0.

This version of mpgrafic is a modified version of the public version of mpgrafic, found at

to produce files readable by Enzo. It has been modified to write HDF5 files in parallel.


  • HDF5 with parallel and FORTRAN support (flags –enable-parallel –enable-fortran)
  • FFTW v2 with MPI support and different single and double precision versions. It must be compiled once for single precision and another time for double precision. For the former, use the flags –enable-mpi –enable-type-prefix –enable-float. For double precision, use –enable-mpi –enable-type-prefix.


Non-nested initial conditions are created only using mpgrafic. However if the user wants nested initial conditions, a full-resolution grid (e.g. 2563 grid for a 643 top grid with 2 nested grids) must be created first and then post-processed with degraf to create a degraded top-level grid and cropped (and degraded if not the finest level) grids for the nested grids.

As with the original inits Enzo package, the baryon density and velocities are written in a 3 dimensional array. The original inits writes the particle data in 1-d arrays. In mpgrafic, only the particle velocities are written in a 3-d array. Enzo has been modified to create the particle positions from the Zel’dovich approximation from these velocities, so it is not needed to write the positions anymore. Also it does not create particles that are represented by a finer grid at the same position.

One big benefit of writing the particle velocities in a 3-d array is avoiding the use of the RingIO tool because each processor knows which subvolume to read within the velocity data.

As of HDF5 version 1.8.2, there exists a bug that creates corrupted datasets when writing very large (e.g. >20483) datasets with multiple components (4-d arrays). The HDF5 I/O in mpgrafic works around this bug by creating one file per velocity component for both the baryons and particles.

How to run

First the user needs to compile both mpgrafic and degraf. The configure / make systems are set up similarly.

Configure flags:

--enable-enzo turns on I/O for Enzo
 creates files in double precision
 creates one file per velocity component
 sets directory for parallel HDF5

If FFTW is not present in the user’s library path, the following variables must be also set

CFLAGS="-I ${FFTW_DIR}/include"
FCFLAGS="-I ${FFTW_DIR}/include"

To run in parallel, you can use FC=mpif90 and LD=h5pfc, which the compiler wrapper for parallel HDF5.

Example configure (for Mac OSX):

./configure LD="-bind_at_load" FC=mpif90 CC=mpicc --enable-enzo \
--enable-double --enable-onedim --with-hdf=/usr/local/hdf5/1.8.2p

Example configure scripts can be found in mpgrafic/mpgrafic-0.2/conf.*. After a successful configure, you can make mpgrafic or degraf by typing ‘make’.

After the programs are compiled, you make the initial conditions by using a python script,, in the top directory that simplifies the user input into mpgrafic and degraf and the moving of files. parameters

number of processors
box size in comoving Mpc (not Mpc/h)
top-level grid resolution
level of the finest nested grid
width of the finest nested grid
number of cells separating nested grids
random seed (must be 9 digits)
name of the data directory (saved in mpgrafic/data/name/)
how much to shift the data in order to center on a particular region.
whether to use a noise file from a lower-resolution run
noise file from that lower-resolution run
whether we’re using one file per velocity component
Omega matter
Omega lambda
Omega baryon
Hubble constant in units of [km/s/Mpc]
slope of power spectrum

After you set your parameters, run this script with


and it will re-compile mpgrafic and (for nested grids) degraf. Then it will run mpgrafic for the full-resolution box. If the user wants nested grids, it will copy the data files to mpgrafic/degraf and create the set of nested grid files.

The user cannot specify the initial redshift because mpgrafic determines it from the parameter sigstart that is the maximum initial density fluctuation. From this, mpgrafic calculates the initial redshift. This file is overwritten by the python script, so if you want to change this parameter, change it in the python script (routine write_grafic1inc).

The noise file is always kept in mpgrafic/mpgrafic-0.2/src and is named $seed_$resolution.dat, where $resolution is the top-level grid resolution. It can be re-used with LargeScaleFile if the user wants to re-simulate the volume at a higher resolution.

The data files are moved to mpgrafic/data/$name. If nested grids were created, degraf writes a set of parameters in enzo.params for copy-pasting into an Enzo parameter file. Now you can move the files to the simulation directory and start your Enzo cosmology simulation!