5. Other kinds of output

So far, the result of any simulation has been power spectra, though several other types of output are available, as exemplified by the below parameter file:

param/tutorial-5

# Non-parameter variable used to control the size of the simulation
_size = 64

# Input/output
initial_conditions = {
    'species': 'matter',
    'N'      : _size**3,
}
output_dirs = {    'snapshot' : f'{path.output_dir}/{param}',    'powerspec': ...,    'bispec'   : ...,    'render2D' : ...,    'render3D' : ...,}
output_times = {    'snapshot' : 0.1,    'powerspec': [a_begin, 0.3, 1],    'bispec'   : ...,    'render3D' : ...,    'render2D' : logspace(log10(a_begin), log10(1), 15),}powerspec_select = {    'matter': {'data': True, 'corrected': True, 'linear': True, 'plot': True},}bispec_select = {    'matter': {'data': True, 'reduced': True, 'tree-level': True, 'plot': True},}render2D_select = {    'matter': {'data': False, 'image': True, 'terminal image': True},}render3D_select = {    'matter': {'image': True},}
# Numerics
boxsize = 128*Mpc
potential_options = 2*_size

# Cosmology
H0      = 67*km/(s*Mpc)
Ωb      = 0.049
Ωcdm    = 0.27
a_begin = 0.02
 # Graphicsrender2D_options = {    'gridsize': {        'matter': _size,    },    'terminal resolution': {        'matter': 80,    },    'colormap': {        'matter': 'inferno',    },}render3D_options = {    'color': {        'matter': 'inferno',    },    'background': {        'matter': 'black',    },    'resolution': {        'matter': 640,    },}

Run a simulation using the above parameters, e.g. by saving them to param/tutorial-5 and executing

./concept \
    -p param/tutorial-5 \
    -n 4

This will take a few minutes. You may read along in the meantime.

We see that besides power spectra, we now have snapshots, bispectra and renders, the latter of which comes in a 2D and a 3D version. The ellipses (...) used above in e.g. output_dirs indicate that we want all kinds of output to go to the same directory.

For the output_times, different values are given for three of the output types, while 'bispec' and 'render3D' are set to use the same times as the output just above, i.e. that of 'powerspec'. For 'render2D', we’ve specified 15 outputs spaced logarithmically equidistant between \(a = a_{\text{begin}} = 0.02\) and \(a = 1\).

Among the new parameters introduced are powerspec_select, in which we have specified that we want the usual data files and plots, including the linear theory predictions. A new kind of power spectrum output, 'corrected', is also specified. This is the non-linear simulation power spectrum, but corrected for noise stemming from the current realisation (cosmic variance) as well as from the binning procedure of the spectrum. The power spectrum plots should show that the corrected non-linear power spectrum is a closer match to the linear power spectrum at the lower \(k\), at least at early times.

5.1. Bispectra

The output directory will be populated with a large number of output files. The ones with names beginning with bispec are the bispectrum files, analogous to the by-now familiar power spectrum files. Looking at the bispectrum plots, we see that they show both the bispectrum \(B\) as well as its reduced version \(Q\), and even the perturbative tree-level prediction. We get all of these as they are enabled in the bispec_select parameter.

For the first bispectrum plot at \(a = a_{\text{begin}}\), the coloured simulation \(B\) line is mostly dashed, indicating negative values, which usually means that the bispectrum signal present in the simulation is too small compared to the noise. Indeed, the later plots show the simulation \(B\) as a full line, indicating positive, proper values.

The bispectrum data files are very analogous to the power spectrum files, though the number of columns contained within them is usually much greater.

Tip

One way to comfortably view wide text files (e.g. output/tutorial-5/bispec_a=1.00) in the terminal is to use

less -S output/tutorial-5/bispec_a=1.00

You can then scroll both horizontally and vertically with the arrow keys. Press Q to quit.

The specific bispectrum that has been measured is the equilateral bispectrum. A later section of this tutorial is dedicated to the bispectrum, including other configurations.

5.2. 3D renders

You will find other image files in the output directory with names that begin with render3D. These are — unsurprisingly — the 3D renders.

In the render3D_select parameter we’ve specified that we want 3D renders of the 'matter' component, whereas specifics of the 3D renders are given in render3D_options. Here we’ve set the colour of the matter particles to be 'inferno', which Matplotlib recognises as a colormap. This colormap will be used to map colours to regions of the box depending on the local density.

Tip

The available colormaps can be viewed here. A single colour value may also be given, either as an RGB tuple (each value ranging from 0 to 1) or as a named colour.

We also specify the background colour, as well as the resolution (height and width) of the image, in pixels.

5.3. 2D renders

The 2D renders show the particle configuration projected along one of the axes of the box.

In the render2D_select parameter we’ve specified that we want images as well as terminal images, but no data. Here, images refer to the 2D render image files you see in the output directory. Terminal images are rendered directly in the terminal as part of the printed output, as you have probably noticed. If you turn on the data output, the 2D render data will further be dumped to HDF5 files.

The options for the 2D renders are collected in the render2D_options parameter. Here gridsize sets the resolution of the cubic grid onto which the particles are interpolated in order to produce the render. The height and width of the image files (in pixels) are then equal to gridsize. The terminal image is produced by resizing the interpolation grid to match the resolution given in 'terminal resolution'. The terminal image is then displayed using one character slot per grid cell, i.e. the terminal render will be 'terminal resolution' (80) characters wide. Since the terminal render is constructed from the original 2D render, this does not show more detail even though the resolution is higher (80 vs. 64).

Also available through render2D_options is the colormap to use.

The play utility

For this next trick, the simulation needs to have finished, and we need to know its job ID.

Tip

To grab the job ID from a power spectrum (or bispectrum) data file, you can do e.g.

grep -o "job [0-9]*" output/tutorial-5/powerspec_a=1.00

With the job ID at hand, try the following:

./concept -u play <ID>  # replace <ID> with job ID number

You should see a nice animation of the evolution of the large-scale structure, playing out right in the terminal! The animation is produced from the terminal images stored in the log file job/<ID>/log.

The -u option to the concept script signals CONCEPT to start up a utility rather than running a simulation. These utilities are handy (and in this case goofy) side programs baked into CONCEPT. You are encouraged to play around with the options to the play utility, a brief overview of which gets printed by

./concept -u play -h

Another such utility — the info utility — is encountered just below, and we will encounter others in later sections of the tutorial. To see the available utilities, do .e.g

ls util

For focused documentation on the individual utilities, consult Utilities.

5.4. Snapshots

Snapshots are raw dumps of the simulated system, in this case the positions and momenta of all \(N = 64^3\) particles. By default, CONCEPT uses its own snapshot format, which is simply a well-structured HDF5 file.

Tip

For a great graphical tool to explore HDF5 files in general, check out ViTables. If you encounter problems viewing HDF5 files produced by CONCEPT, check that you are using ViTables 3.

You can install ViTables as part of the CONCEPT installation via

(source concept && $python -m pip install vitables)

after which you can run it by e.g. doing

(source concept && $python -m vitables &)

If ViTables is unable to start, it might be due to Qt 5 not being installed on your system. If you’re on a Debian/Ubuntu-like system (and have root privileges), you can install Qt 5 by

sudo apt install qtbase5-dev

It might also help to install the master version rather than the latest release:

(source concept && $python -m pip uninstall -y vitables)
(source concept && $python -m pip install git+https://github.com/uvemas/ViTables.git)

Such snapshots are useful if you want to process the raw data using some external program. You can also initialise a simulation from a snapshot, instead of generating initial conditions from scratch. To try this, redefine the initial conditions to simply be the path to the snapshot produced by the simulation you just ran:

initial_conditions = f'{path.output_dir}/{param}/snapshot_a=0.10.hdf5'

Also, you should change a_begin to be 0.1 as to comply with the time at which the snapshot was dumped. Finally, before rerunning the simulation starting from the snapshot, you should probably comment out at least the 'render2D' output_times, as to not clutter up the output directory too heavily.

If you forget to correct a_begin, a warning will be emitted. The same goes for other obvious inconsistencies between the parameter file and the snapshot, like if boxsize or Ωcdm is wrong. To be able to do this, some meta data about the cosmology and numerical setup is stored in the snapshot as well.

Note

Warnings are shown in red bold text, easily distinguishable from the main output text. As warnings (as opposed to errors) results from non-fatal issues, CONCEPT will continue running. A warning emitted during the simulation may hint that something has gone wrong, meaning that the results perhaps should not be trusted. To make sure that no warnings go unnoticed, CONCEPT will notify you at the end of the simulation. A separate error log, job/<ID>/log_err, containing just warnings and error messages, will also be present.

To generate a warning and error log file, try wrongly specifying e.g. Ωb = 0.05. Once the simulation has completed, check out the error log file.

If you intend to run many simulations using the same initial conditions, it might be worthwhile to initialise these from a common snapshot. To produce such an initial snapshot, simply set output_times = {'snapshot': a_begin}, in which case CONCEPT will exit right after the snapshot has been dumped at the initial time, without doing any simulation. An ic directory should already exist within your CONCEPT installation, the purpose of which is to hold such initial condition snapshots. To dump snapshots to this directory, set the 'snapshot' entry in output_dirs to path.ic_dir. We can achieve both without even altering the parameter file:

./concept \
    -p param/tutorial-5 \
    -c "output_times = {'snapshot': a_begin}" \
    -c "output_dirs = {'snapshot': path.ic_dir}" \
    -n 4

You may also want to use CONCEPT purely as an initial condition generator, and perform the actual simulation using some other code. If so, the default CONCEPT snapshot format is of little use. To this end, CONCEPT also supports the binary Fortran format of GADGET, which is understood by many other simulation codes and tools. To use this snapshot format in place of the CONCEPT format, add snapshot_type = 'gadget' to your parameter file.

The info utility

We mentioned ViTables as a great way to peek inside the default CONCEPT (HDF5) snapshots. It would be nice to have a general tool which worked for the supported GADGET snapshots as well. Luckily, CONCEPT comes with just such a tool: the info utility. To try it out, simply do

./concept -u info output/tutorial-5

or

./concept -u info ic

The content of all snapshots — CONCEPT (HDF5) or GADGET format — in the given directory will now be printed to the screen. Should you want information about just a specific snapshot, simply provide its entire path.