Version history

See below for a listing of the most important code and interface changes in Colossus, starting with version 1.1.0. You can download older versions from the PyPi History for Colossus.

Version 1.3.6 (released 07/11/2024)

• Updates to make Colossus compatible with newer python versions. Thanks to Steven Murray for the updates!

Version 1.3.5 (released 03/02/2024)

• Added kpcPerArcsec() function to cosmology.

Version 1.3.4 (released 02/09/2024)

• Added rho_k() function for density of curvature.

• Small fixes and improvements in documentation.

Version 1.3.3 (released 12/10/2023)

• Better error message when Colossus encouters recollapsing Universes (e.g., with high matter density). These Universes pose a fundamental challenge to all redshift or scale-factor based interpolation tables, since the x-axis in those tables becomes multi-valued. Such Universes currently must be treated using their symmetry over the point where they begin to recollapse.

• Various bug fixes.

Version 1.3.2 (released 07/30/2023)

• Added documentation about errors stemming from the negative correlation function at large radii (thanks to Antonio Ragagnin for bringing this to my attention!)

• Other small fixes to documentation.

Version 1.3.1 (released 08/02/2022)

This version fixes a bug in the version check of the storage system, which erroneously triggered file updates.

Version 1.3.0 (released 06/13/2022)

This update represents a major new version. The two essential changes are a) the inclusion of power spectra computed by the CAMB Boltzmann code and b) a totally reworked halo density profile module.

First, the cosmology module has been updated significantly to allow for the direct inclusion of CAMB power spectra. Some of the main changes are:

• The user can request a new camb power spectrum model that is evaluated internally via CAMB’s python interface. This is much simpler to use than manually created, tabulated power spectra because Colossus correctly sets the important cosmological parameters and CAMB options.

• The camb module is now an optional dependency that will only ever be important if required. If it is not installed, an error is thrown.

• To accommodate such generalized power spectrum models, the transferFunction() function has been deprecated and will be removed in a future release. Instead, power spectra or transfer functions can be evaluated with the powerSpectrum() function. However, in general the matterPowerSpectrum() function within the cosmology object should be used for simplicitiy and consistency.

• The parameters to the power spectrum have been further generalized as a ps_args dictionary. If any parameters are passed, this dictionary needs to be kept consistent between all functions. For the CAMB model, passed parameters are fed to the initialization of the CAMB object, allowing the user to set a wide range of options.

• The previous path argument is now also part of ps_args in all functions (it was already in some functions).

• The public functions powerSpectrumModelName() and powerSpectrumLimits() have been added to the power spectrum module, and matterPowerSpectrumNorm() has been added to the Cosmology class. The latter allows the user insight into the normalization used to keep power spectra consistent to a fixed value of sigma8.

Second, the density profiles module has been reworked entirely. Some of the following changes are unfortunately not backwards compatible, as discussed below. The general philosophy of the new structure is to generalize the creation of profiles as much as possible, including the addition of outer terms. The constructor function signatures have been radically simplified to mostly take arbitrary keyword arguments, which are parsed by the respective constructors and by the functions called in turn. Please feel free to get in touch if you have trouble migrating your code to the new version. The main changes are as follows:

• Generalized construction of profiles

  • The profile constructor was generalized to work with the keyword arguments given by the user. A derived class now only needs to list its parameter (and perhaps option) names, and the parent constructor attempts to construct a profile from these arguments. If not all native parameters are given, the constructor looks for a function called setNativeParameters that sets the parameters based on mass, concentration, and redshift. Aside from this function, the entire logic for creating profiles is now contained in the parent class.

  • Child classes now must implement a setNativeParameters() function that needs to be able to accept any mass definition unless the allowed_mdefs parameter is set when calling the parent constructor.

  • All derived classes must now contain a normalization rhos because this variable is used to renormalize the profile in the presence of outer terms. The exception are parameter-free profiles, such as splines.

  • The generalized constructor fixes a previous issue when creating profiles with outer (infalling) terms. If the profile parameters are determined from a mass and concentration, and the outer profile depends on a radius such as R200m, the process of finding the profile parameters is iterative. This iteration was not performed for all profiles.

• Easier construction of composite (inner + outer) profiles

  • The new compositeProfile() function allows the user to easily create any combination of inner and outer profiles using shortcodes.

  • The user is responsible for passing the appropriate parameters to this function; otherwise, respective constructors throw errors.

• The new Diemer 2023 (truncated exponential) profile has been added; this form separately describes the orbiting and infalling components and is now recommended over the DK14 profile.

• The DK14 profile has been reworked

  • In the DK14 profile, all options have been removed as they are only needed for the constructor.

  • The getDK14ProfileWithOuterTerms function has been removed from the DK14 profile module, and has been replaced by the general compositeProfile() function. The signature is similar, but the parameter names are now consistent with the constructors of the respective outer terms.

  • The function DK14Profile.M4rs has been removed. The result can easily be found by evaluating the enclosed mass within four scale radii.

• Changes in other profile modules

  • For the NFW profile, the fundamentalParameters function (which has now been replaced by setNativeParameters()) was a class method, meaning that it could be called without calling the constructor first. This routine has been renamed to nativeParameters().

• Fitting

  • The transformation between linear and log parameters has been radically simplified. All parameters, including those of the outer profiles, are now by default fit in log space. This can lead to slightly different results compared to previous versions. The user can change this behavior by overwriting certain functions.

  • The old scipy.optimize.leastsq function was replaced by the newer scipy.optimize.least_squares interface, which contains more advanced algorithms such trf (the new default fitter).

  • The user can now pass parameter bounds in least-squares fits.

  • MCMC fits can now also be performed in log space by default to ensure the positivity of the parameters.

• All profile documentation pages have been overhauled.

• The profile Tutorials have been improved and expanded.

• The unit test suite has been improved and expanded.

A few other changes:

• The cosmology setter functions setCosmology() and addCosmology() now support keyword arguments in addition to a dictionary, which makes setting and overwriting parameters more convenient.

• Small bug fix where evaluating the dark energy density for an array of redshifts sometimes returned a number instead of an array.

• Convert np.int to int and np.float to float to avoid deprecation warnings.

Version 1.2.19 (released 09/02/2021)

• Added the final version of the ishiyama21 concentration model, which was renamed from ishiyama20 to conform with the published version. This model now also contains fits for the 500c mass definition and for relaxed halos.

Version 1.2.18 (released 03/18/2021)

• Added the new seppi20 mass function model. Thanks to Riccardo Seppi for the implementation and tutorial!

• Fixed a bug in the power spectrum derivative at z > 0 (thanks to Michael Joyce for finding this bug!)

Version 1.2.17 (released 08/07/2020)

This version contains the new ishiyama20 concentration model, which is a recalibration of the diemer19 model based on the Uchuu simulation.

Version 1.2.16 (released 07/15/2020)

Changes in this version include:

• The cosmology module now supports conversions to and from Astropy, including a number of dark energy models. See the toAstropy() and fromAstropy() functions. Thanks to Steven Murray for the idea and code!

• A new splashback model, diemer20, was added and made the default splashback model. This model is a recalibration of the diemer17 model, with percent-level changes.

• The implementation of these splashback models has changed, with some interface changes to the convenience functions.

• The main splashbackModel() function does not provide a default definition for the diemer17 and diemer20 models any longer (such as the mean or higher percentiles of the particle splashback distribution). The definition matters quite a bit and should be provided by the user to avoid confusion. The function now throws an error if no definition is passed.

• A new mass function model, diemer20 has been added (not to be mistaken for the splashback radius and mass model of the same name). This model is the first to predict splashback mass functions.

• The integration in sigma() was made more robust in the case where the user has specified a lower or upper limit to the integration. In particular, the calculation of the tree integration limit was improved and the code now automatically increases the number of bins in the intepolation table because the solution oscillates near the cutoff.

• A number of functions in the cosmology and halo modules are now safe to input of integers instead of float. For example, the growthFactorUnnormalized function returned wrong values when “1” was given instead of “1.0” or “1.”. Thanks to Yucheng Zhang for pointing this out!

Version 1.2.15 (released 04/15/2020)

Changes in this version include:

• In self-similar cosmologies, the correlation function is now computed from analytical expressions rather than numerical integration (thanks to Michael Joyce for the analytical formulae).

• The variance sigma can be computed between user-defined lower and/or upper limits in k-space. This feature is useful when calculating the variance in a box of limited size, for example.

Version 1.2.14 (released 01/23/2020)

Changes in this version include:

• The user can now pass power spectrum arguments to the Diemer & Joyce 2019 concentration model, for example, in order to use a non-standard power spectrum.

• The code returns more informative error messages when tabulated power spectra are used.

• The normalization of the power spectrum for self-similar (power-law) cosmologies has been fixed for both the tophat and Gaussian filters, and the variance is now computed from the analytical expression rather than numerical integration (thanks to Michael Joyce for finding this bug and providing the analytical formulae!).

Version 1.2.13 (released 11/08/2019)

Colossus has migrated from mercurial (hg) to git, and this version simply updates the documentation and readme files. The reason for this migration is that BitBucket is retiring its mercurial support in 2020, but this decision is just a symptom of a broader trend.

Some may wonder why the code has not been migrated to GitHub instead of BitBucket, now that the repository system does not matter any more. The answer is that GitHub does not support the current development model, namely a private fork of the public repository. Moreover, previous issues and commit details cannot be transferred to GitHub and would be lost. Thus, Colossus is now a git repository, but is still hosted on BitBucket.

Version 1.2.12 (released 10/28/2019)

This version contains some minor bug fixes, namely:

• Improved error checking in Spline profile.

• Fixed bug when trying to compute outer profile for objects that have no outer profile.

• Some calculations relating to dark energy, including the growth factor, can fail at far-future times when the w0-wa dark energy model is active. This happens because dark energy grows exponentially, leading to some very large values. Now, the default redshift range is reduced from a=200 to a=10 for w0wa and user-defined cosmologies. Thank to Antonio Villareal for pointing out this bug!

Version 1.2.11 (released 08/12/2019)

Fixes a bug in the Halo mass function module, where redshift was not correctly passed to the sigma function.

Version 1.2.10 (released 08/05/2019)

The changes in this version were largely inspired by a detailed comparison with the Core Cosmology Library (CCL) by the LSST-DESC.

• Physical and astronomical constants were updated to IAU 2015 / PDG 2018 standard, including the definition of parsec/kpc/Mpc and the solar mass. Those changes translate into changes in the gravitational constant in astronomical units and the critical density of the universe, which in turn are used in numerous functions.

  Note

  This change affects most outputs from Colossus, but only by factors up to 1E-4 or less. All stored pickles will automatically be recomputed following this change.

• Added the sugiyama95 transfer function model.

• When manually changing cosmology, all derived parameters are now automatically updated. Previously, changes to T_CMB0 and Neff did not have any effect. Thanks to Sebastian Bocquet for pointing out this issue!

• The Halo mass function module now correctly passes additional arguments to the power spectrum, variance, and collapse overdensity functions. This only makes a difference to the results if the user passes additional parameters such as a tabulated power spectrum. Thanks to Wojciech Hellwing for finding this bug!

Version 1.2.9 (released 03/23/2019)

• Removed reference to packaging package by adding manual version comparison function.

• Added unit tests for versioning and storage.

• Added unit tests for derived constants.

• Added a new FAQ and troubleshooting page to the documentation.

Version 1.2.6 (released 03/01/2019)

• Fixed small discrepancy in the unit system. The gravitational constant was adjusted by a factor of 4E-5, leading to the same discrepancy in the critical density of the universe. Thanks to Tom McClintock for pointing out this bug!

  Note

  This change affects numerous outputs from Colossus, but only by factors of around 4E-5 (and much less in most cases).

• Added a system to automatically delete outdated storage files. If files older than a certain version are found, a warning is displayed, the file is deleted, and the computations will be done from scratch.

• Fixed bug in the Bocquet et al. 2016 mass function for the M200c and M500c mass definitions (thanks to Michelle Ntampaka for catching this!).

Version 1.2.5 (released 01/30/2019)

• Renamed the diemer18 concentration model to diemer19 to match the publication date.

• Changed the default concentration model from diemer15_orig to diemer19.

  Note

  This changes the output of all functions that use the default concentration model, namely concentration(), changeMassDefinitionCModel(), and splashbackRadius(). If the user has specified a concentration model (which is possible in all these functions), the output will not change.

• Fixed bug in wCDM growth factor calculation.

• Added the mass function model of Comparat et al 2017 to the Halo mass function module.

• Added the bias models of Bhattacharya et al 2011 and Comparat et al 2017 to the Halo Bias module. Thanks to Johan Comparat for the suggestion!

Version 1.2.4 (released 10/29/2018)

This version corresponds to the published version of the code paper.

• The Gaussian filter in the filterFunction() (used to compute the variance of the linear power spectrum, sigma()) was changed by a factor of two to adhere to the common definition.

  Note

  This change of the Gaussian filter represents a significant, not backward-compatible change. If you use the Gaussian filter in ANY of your calculations, please check your results – they will be affected. Before re-computing your results, please remove all temporary cosmology files in ~/.colossus/cache/cosmology to make sure that the change has taken effect.

  Note

  Due to the change in the Gaussian filter, the return of the peakCurvature() function has changed. If you use this function, please check your results (and follow the procedure described in the note above).

• Many small fixes to the documentation, thanks to Jerry Maggioncalda for his careful proofreading!

• Activated continuous integration (i.e., automatically running the unit test suite after every commit). Thanks to Joseph Kuruvilla for setting that up!

• The Diemer & Joyce 2018 concentration model is presented in its published form. The routine was sped up through a pre-computed, stored interpolation table.

• The xDelta() function in the Navarro-Frenk-White profile module was restructured completely. It now uses an interpolation table instead of root finding which means that it now allows numpy arrays as input and makes it orders of magnitude faster (depending on the size of the input). The accuracy of the interpolation is better than 1E-7. The function interface has two fewer parameters.

• The cosmology of the Multidark-Planck simulations was added.

Version 1.2.2 (released 07/31/2018)

This version fixes several bugs and adds new features. Changes in the cosmology module include:

• Major bug fix: the growth factor was incorrect for \(w \neq -1\) cosmologies, an error that has been rectified in this release (thanks to Lehman Garrison for catching this bug).

• The redshift interpolation tables in the cosmology module are now spaced equally in \(\ln(1 + z)\) rather than \(z\). This change reduces the interpolation errors slightly and, more importantly, leads to less ringing in the first derivatives of some quantities, namely the linear growth factor. The new interpolation tables carry different names than the old ones, meaning that old cache files do not need to be deleted as the two tables can co-exist. Due to the changed tables (and the changes to the growth factor), some cosmology functions can exhibit differences of the order 0.1% compared to the previous version.

• The Planck 2018 cosmology was added (and can be used by setting planck18 or planck18-only for the cosmology).

• The inverse option was removed from the angularDiameterDistance() function because the inverse is multi-valued and leads to an error.

Changes in the large-scale structure module:

• Three new bias models were added to the Halo Bias module, namely those of Jing 1998, Seljak & Warren 2004, and Pillepich et al. 2010.

• The function powerSpectrumSlope() was added to the Peaks in Gaussian random fields module. This function evaluates the slope of the power spectrum or variance at a given peak height and is used in the bias and concentration modules.

• Bug fix: the ps_args parameter was not used in the massFromPeakHeight() and peakCurvature() functions (thanks to Michael Joyce for catching this bug).

Changes in the halo module:

• The halo concentration models of Ludlow et al. 2016, Child et al. 2018, and Diemer and Joyce 2018 were added.

• The Diemer and Kravtsov 2015 model was updated according to Diemer and Joyce 2018.

• The default concentation model remains the original Diemer & Kravtsov 2015 model, without the improvements of Diemer and Joyce 2018. In a near-future release, the default concentration model will switch to their new model which will influence a few functions such as changeMassDefinitionCModel(). However, the numerical differences to the previous default model are small.

Other changes:

• The function plotChain was removed from the MCMC module to avoid including the matplotlib library. The function is still available as part of the MCMC tutorial.

• Numerous small improvements were made in the documentation.

Version 1.2.1 (released 12/13/2017)

Version 1.2.1 is the version that coincided with the first publication of the code paper on arXiv.org. The following major changes were made:

• The documentation was reworked entirely.

• All functions and parameters that were deprecated in 1.1.0 have been removed from the code (rather than outputting warnings).

• The qx and qy parameters in the halo.splashback module were renamed to q_in and q_out to conform with the rest of the code. A number of other small inconsistencies in splashback radius interface were fixed.

Version 1.1.0 (released 11/27/2017)

Version 1.1.0 presents a major change to the Colossus interface, documentation, and tutorial system. The most important changes are that

• A new top-level module for large-scale structure, LSS, has been added, including functions previously housed in the cosmology module, the old halo bias module, and a new module for the halo mass function. The LSS module covers funtions that deal with peaks or halos as a statistical ensemble so that the cosmology module does no longer “know” anything about halos. Conversely, the halo module covers functions that apply to individual halos.

• The demo scripts have been converted to much more extensive Jupyter notebook Tutorials.

• A number of interfaces have been made more homogeneous.

• Wherever possible, deprecated function interfaces are still present for backward compatibility but issue a warning. These functions and parameters will be removed in the next version.

• This documentation has been reorganized and improved, and its location has shifted to https://bdiemer.bitbucket.io/colossus.

The following functions are now housed in the LSS module:

• Cosmology.lagrangianR() is now lss.peaks.lagrangianR()

• Cosmology.lagrangianM() is now lss.peaks.lagrangianM()

• Cosmology.collapseOverdensity() is now lss.peaks.collapseOverdensity()

• Cosmology.peakHeight() is now lss.peaks.peakHeight()

• Cosmology.massFromPeakHeight() is now lss.peaks.massFromPeakHeight()

• Cosmology.nonLinearMass() is now lss.peaks.nonLinearMass()

• Cosmology.peakCurvature() is now lss.peaks.peakCurvature()

• The module halo.bias is now lss.bias.

• The LSS module contains a brand new module to compute the halo mass function, lss.mass_function.

The following changes apply to interfaces across modules:

• Any module that implements models (e.g., fitting functions for concentration), now features an ordered dictionary called models that contains class objects with the properties of the respective models (which vary from module to module). This change affects the power spectrum, bias, halo mass function, concentration, and splashback modules. These new model dictionaries replace the previous MODELS lists that were present in some of the modules.

• There is a new storage module as part of utilities. The storage parameter in the cosmology module was renamed to persistence, as was the global setting STORAGE (renamed to PERSISTENCE). The storage module can now be used by other modules or from outside of Colossus.

Changes in the cosmology module:

• Cosmology now allows for a non-constant dark energy equations of state. The implemented dark energy models include a fixed or varying equation of state (see Cosmology class for more information). As a result, the OL0, OL(), and rho_L() parameters and functions were renamed to Ode0, Ode(), and rho_de().

• The power spectrum models were extracted into a separate module, cosmology.power_spectrum. The names of the available models were changed from eh98 to eisenstein98 and from eh98_smooth to eisenstein98_zb to conform with other Colossus modules.

• The Pk_source parameter was renamed to model in the matterPowerSpectrum() function. In functions that call the power spectrum, the user can pass a ps_args dictionary containing kwargs that are passed to the power spectrum function.

• The matterPowerSpectrum() function now takes redshift as an optional parameter.

• The text_output option was removed from the cosmology object.

• The soundHorizon() function now returns the sound horizon in Mpc/h rather than Mpc in order to be consistent with the rest of the cosmology module.

Changes in the LSS module:

• The collapseOverdensity() function has been completely reworked. By default, it still returns the constant collapse overdensity threshold in an Einstein-de Sitter universe. If a redshift is passed, it applies small corrections based on the underlying cosmology. The previous parameters to this function will now cause an error. This change also affects all functions that rely on the collapse overdensity, such as peakHeight(), massFromPeakHeight(), nonLinearMass(), and peakCurvature(). These functions now accept dictionaries of parameters that are passed to the collapse overdensity and sigma() functions.

• The halo bias module was extended with two new models for halo bias.

• The input units to the twoHaloTerm() function are now in comoving Mpc/h rather than physical kpc/h in order to conform to the unit system of the LSS module.

Changes in the halo module:

• The interface of the SO changing functions in halo.mass_defs has changed. The function previously called pseudoEvolve is now called evolveSO() to reflect its more general nature. The pseudoEvolve() function is a wrapper for evolveSO, and has one fewer parameter than previously (no final mass definition).

• The DK14Profile constructor does not take R200m as an input any more and instead computes it self-consistently regardless of what the other inputs are. In this new version, the redshift always needs to be passed to the constructor. These changes fix a bug with outer profiles that themselves rely on R200m as an input. Furthermore, the normalization of power-law outer profiles is no longer adjusted in order to maintain a constant amplitude of R200m changes. It is up to the user to ensure that the behavior of the outer profile makes sense physically.

• The klypin14_nu and klypin14_m concentration models were renamed to klypin16_nu and klypin16_m to maintain compatibility with the publication date of their paper.