21cmfast

All functionality of the original 21cmFAST v2 C-code has been implemented in this
version, including ``USE_HALO_FIELD`` and performing full integration instead of using
the interpolation tables (which are faster).

Added
~~~~~
* Updated the radiation source model: (i) all radiation fields including X-rays, UV
ionizing, Lyman Werner and Lyman alpha are considered from two seperated population
namely atomic-cooling (ACGs) and minihalo-hosted molecular-cooling galaxies (MCGs);
(ii) the turn-over masses of ACGs and MCGs are estimated with cooling efficiency and
feedback from reionization and lyman werner suppression (Qin et al. 2020). This can
be switched on using new ``flag_options`` ``USE_MINI_HALOS``.
* Updated kinetic temperature of the IGM with fully ionized cells following equation 6
of McQuinn (2015) and partially ionized cells having the volume-weightied temperature
between the ionized (volume: 1-xHI; temperature T_RE ) and neutral components (volume:
xHI; temperature: temperature of HI). This is stored in IonizedBox as
temp_kinetic_all_gas. Note that Tk in TsBox remains to be the kinetic temperature of HI.
* Tests: many unit tests, and also some regression tests.
* CLI: run 21cmFAST boxes from the command line, query the cache database, and produce
plots for standard comparison runs.
* Documentation: Jupyter notebook demos and tutorials, FAQs, installation instructions.
* Plotting routines: a number of general plotting routines designed to plot coeval
and lightcone slices.
* New power spectrum option (``POWER_SPECTRUM=5``) that uses a CLASS-based transfer
function. WARNING: If POWER_SPECTRUM==5 the cosmo parameters cannot be altered, they
are set to the Planck2018 best-fit values for now (until CLASS is added):
(omegab=0.02237, omegac= 0.120, hubble=0.6736 (the rest are irrelevant for the
transfer functions, but in case: A_s=2.100e-9, n_s=0.9649, z_reio = 11.357)
* New ``user_params`` option ``USE_RELATIVE_VELOCITIES``, which produces initial relative
velocity cubes (option implemented, but not the actual computation yet).
* Configuration management.
* global params now has a context manager for changing parameters temporarily.
* Vastly improved error handling: exceptions can be caught in C code and propagated to
Python to inform the user of what's going wrong.
* Ability to write high-level data (``Coeval`` and ``Lightcone`` objects) directly to
file in a simple portable format.

Changed
~~~~~~~
* ``POWER_SPECTRUM`` option moved from ``global_params`` to ``user_params``.
* Default cosmology updated to Planck18.

----------------------

Performance
~~~~~~~~~~~

* When reading in data from cache, only read the necessary fields.

Deprecations
~~~~~~~~~~~~

* Python <3.9 is no longer supported.

Fixed
~~~~~

* Fixed small issue in ``Lightcone.lightcone_coords``.
* Bug when setting ``NON_CUBIC_FACTOR != 1``.
* Using new ``matplotlib`` versions and registering ``cmap``.
* Small bug in lightcone interpolation.

Added
~~~~~

* New ``Lightconer`` base class with user-facing classes ``RectilinearLightconer`` and
``AngularLightconer``. These are now the recommended way to define how lightcones
should be created. Note that these are *not* the final lightcone object, but simply a
"definition" for how to create the lightcone.
* The ``Lightcone`` class still exists, but is now joined by an ``AngularLightcone``
class that is the output lightcone when using the ``AngularLightconer`` definition.
* New ``KEEP_3D_VELOCITIES`` user-param. This is *required* to be True if you want to
apply RSDs to an ``AngularLightcone``.
* Ability to check-point lightcone creation, so that it can be resumed later. Simply
pass the ``lightcone_fname`` argument to ``run_lightcone``.
* New ``cache_tools.get_boxes_at_redshift`` function to get all boxes at a given
redshift, regardless of the parameters used to create them.
* New ``CosmoParams.from_astropy()`` classmethod for constructing a cosmo params object
directly from an astropy cosmology.


Internals
~~~~~~~~~

* Refactored RSDs to the ``subcell_rsds.c`` C-module.
* Added CI testing of the docs tutorials.
* More precise debugging summaries of boxes in the C code.

Fixed
~~~~~

* Incorrect sign on adiabatic fluctuations.
* Spurious warnings about boxes being read.
* Spurious warnings about configuration files being out of date.
* Compilation for ``clang`` and newer ``gcc`` versions.

----------------------

Fixed
~~~~~

* Compilation of C code for some compilers (330)

----------------------

Internals
~~~~~~~~~

* Refactored setting up of inputs to high-level functions so that there is less code
repetition.

Fixed
~~~~~

* Running with ``R_BUBBLE_MAX`` too large auto-fixes it to be ``BOX_LEN`` (112)
* Bug in calling ``clear_cache``.
* Inconsistency in the way that the very highest redshift of an evolution is handled
between low-level code (eg. ``spin_temperature()``) and high-level code (eg. ``run_coeval()``).


Added
~~~~~

* New ``validate_all_inputs`` function that cross-references the four main input structs
and ensures all the parameters make sense together. Mostly for internal use.
* Ability to save/read directly from an open HDF5 File (170)
* An implementation of cloud-in-cell to more accurately redistribute the perturbed mass
across all neighbouring cells instead of the previous nearest cell approach
* Changed PhotonConsEndCalibz from z = 5 -> z = 3.5 to handle later reionisation
scenarios in line with current observations (305)
* Add in an initialisation check for the photon conservation to address some issues
arising for early EOR histories (311)
* Added ``NON_CUBIC_FACTOR`` to ``UserParams`` to allow for non-cubic coeval boxes (289)

----------------------

Changed
~~~~~~~

* Included log10_mturnovers(_mini) in lightcone class. Only useful when USE_MINI_HALOS

----------------------

Changed
~~~~~~~

* Floats are now represented to a specific number of significant digits in the hash of
an output object. This fixes problems with very close redshifts not being read from
cache (80). Note that this means that very close astro/cosmo params will now be read
from cache. This could cause issues when creating large databases with many random
parameters. The behaviour can modified in the configuration by setting the
``cache_param_sigfigs`` and ``cache_redshift_sigfigs`` parameters (these are 6 and
4 by default, respectively).
**NOTE**: updating to this version will cause your previous cached files to become
unusable. Remove them before updating.

Fixed
~~~~~

* Added a missing C-based error to the known errors in Python.