COSMIX — COSmological Modular Inference eXplorer

DOI Tests

COSMIX is a modular, likelihood-driven Bayesian inference framework for testing modified gravity and dark energy cosmological models against observational data. It supports multiple samplers (emcee, Dynesty, PolyChord) and a growing library of likelihoods (CC, Pantheon+, DESI DR2 BAO, RSD, GW standard sirens, DES-SN5YR).

Installation

From PyPI

pip install pycosmix

From source (for development)

git clone https://github.com/AmeyaKolhatkar/COSMIX.git
cd COSMIX
pip install -e .

For a specific tagged release (recommended for reproducible research):

git clone --branch v1.5.1 https://github.com/AmeyaKolhatkar/COSMIX.git
cd COSMIX
pip install -e .

Alternatively, a stable archived version is available via Zenodo: DOI

Optional: build documentation locally

pip install -e ".[docs]"

Optional: PolyChord nested sampler

Warning! The PolyChord nested sampler has not been tested extensively. If you still wish to use it, do

pip install -e src/cosmix/data/PolyChordLite

On Windows you will need WinLibs/MinGW with gfortran and ensure <mingw64>/bin is on your PATH (or set POLYCHORD_DLL_DIR).

Quick Start

1. Configure a run

Edit input.yaml (or copy it) to specify the model, datasets, sampler, and output options:

run:
  name: "MyModel_test"
  run_id: "my_model_test_1"

model:
  name: MyModel          # LCDM | fQ_Hybrid | fQ_EHybrid | fQ_LSR | fQ_LSR_IDE | fQ_Hybrid_IDE

likelihoods:
  - name: cosmicchronometers        # Cosmic Chronometers
  - name: desidr2bao                # DESI DR2 BAO
  - name: pantheonplus              # Pantheon+ (SN Ia)
  - name: rsd                       # Redshift Space Distortions

sampler:
  name: dynesty                     # emcee | dynesty | polychord
  init:
    nlive: 500

outputs:
  plots:
    trace: true
    corner: true
    residual: true
  archive: true

2. Run

Once installed, the cosmix command is available on your PATH:

cosmix input.yaml

Results are saved to runs/<run_id>/ including the chain, diagnostics, information criteria (AIC, BIC, DIC), and publication-quality figures.

3. Override parameters (optional)

Every parameter declared by the model or a likelihood can be changed per run in a parameters: block, without editing code:

parameters:
  lambda0: {fixed: 1.5}                       # fix a free parameter
  rd:      {fixed: 147.05}                    # re-value an already fixed one
                                              #   or {free: true, prior: {type: uniform, low: 130, high: 160}}
  M:       {fixed: -19.25}                    # fix a marginalized one
                                              #   or {free: true} / {marginalized: true}
  H0:      {prior: {type: uniform, low: 60, high: 80}, proposal: 0.3}
  omega_b: {prior: {type: gaussian, mean: 0.02237, sigma: 0.00015}}

Keys: fixed, free, prior (implies free), proposal, marginalized. fixed, free/prior and marginalized are mutually exclusive. Unknown parameter names or keys stop the run with a suggestion. The applied settings are saved in each run’s manifest.yaml.

Project Structure

COSMIX/
├── pyproject.toml            # Package metadata, dependencies, build config
├── input.yaml                # Template run configuration
├── src/
│   └── cosmix/
│       ├── run_cosmix.py     # Main entry point — reads input.yaml and runs the pipeline
│       ├── Constants.py      # Physical constants (c, Omegar0)
│       ├── core/             # Framework internals (pipeline, parameter management, caching)
│       ├── theory/           # Cosmological models (LCDM, f(Q) variants)
│       │   ├── Solvers_/     # ODE/root solvers (RK4, analytical, Numba JIT)
│       │   └── Background_/  # Background state layout
│       ├── likelihoods/      # Observational likelihoods
│       ├── samplers/         # Sampler wrappers (emcee, Dynesty, PolyChord)
│       ├── drivers/          # Multi-chain convergence strategies
│       ├── postprocessing/   # Results, diagnostics, visualization, archival
│       │   └── Archive_/     # Run manifest, serialization (YAML/JSON/NumPy)
│       └── data/             # Observational data files
│           └── PolyChordLite/  # Bundled PolyChord source (build separately)
└── runs/                     # Output directory (gitignored contents)

Available Models

Key

Description

LCDM

Standard ΛCDM

CPL

w₀w_aCDM (CPL equation of state)

MG_muSig

μ–Σ modified growth on a ΛCDM background (μ₀ = Σ₀)

fQ_LCDM

f(Q) with ΛCDM background plus the λ₀√(QQ₀) growth coupling

fQ_LCDM_Curved

fQ_LCDM with spatial curvature

fQ_Hybrid

f(Q) Hybrid model (analytical solution)

fQ_EHybrid

Extended f(Q) Hybrid model (ODE solver)

fQ_Hybrid_Curved

f(Q) Hybrid model with spatial curvature

fQ_LSR

f(Q) Log-Square-Root model

fQ_LSR_IDE

f(Q) LSR with Interacting Dark Energy

fQ_Hybrid_IDE

f(Q) Hybrid with Interacting Dark Energy

fQ_Squared_Curved

Squared f(Q) model with spatial curvature

Available Likelihoods

Key

Dataset

Reference

cosmicchronometers

Cosmic Chronometers (correlated + uncorrelated)

Moresco et al.

pantheonplus

Pantheon+ (SN Ia, no SH0ES prior)

Brout et al. 2022

pantheonplusshoes

Pantheon+SH0ES (SN Ia + H₀ prior)

Brout et al. 2022

desidr2bao

DESI DR2 BAO (full GC combination)

DESI Collaboration 2025

rsd

Redshift Space Distortions (fσ₈)

Various

rsdwigglez

RSD compilation with the WiggleZ covariance

Various

sdssdr16

SDSS DR16 BAO, BAO+fσ₈ or RSD (mode option)

eBOSS Collaboration 2021

planckaspriorcamb

Planck ln(10¹⁰A_s) prior, A_s computed per sample with CAMB

Planck 2018

planckasprior

Deprecated; use planckaspriorcamb

—

gw

GW Standard Sirens (GW170817)

LIGO/Virgo (Test implementation only)

desy5

DES-SN5YR (SN Ia with probabilistic classification)

DES Collaboration

desdovekie

DES-Dovekie Likelihood

DES Collaboration

shoes

SH0ES H₀ Gaussian prior

Riess et al.

trgb

TRGB H₀ prior

Freedman et al.

holicow

H0LiCOW H₀ prior

Wong et al.

compressedcmb

Compressed CMB

Z. Zhai and Y. Wang

egstatistic

Eg Statistic

G. Alestas et al.

Adding a New Model

  1. Create src/cosmix/theory/MyModel.py inheriting from cosmix.core.CosmologyModelBase.

  2. Implement declare_parameters(), check_physicality(), and get_requirements().

  3. Register it through the Registry.py utility

    from cosmix.core.Registry import cosmix_registry

@cosmix_registry.register_model("MyModel")
class MyModel(...)

  4. Reference it in input.yaml as model: { name: MyModel }.

Adding a New Likelihood

  1. Create src/cosmix/likelihoods/MyLikelihood.py inheriting from cosmix.core.LikelihoodBase_.

  2. Implement declare_parameters(), get_requirements(), and lnlike().

  3. Register it through the Registry.py utility

    from cosmix.core.Registry import cosmix_registry

@cosmix_registry.register_likelihood("MyLikelihood")
class MyLikelihood(...)

Under Development

  • The PolyChord sampler has not been fully tested yet. User discretion is adviced.

  • The multi_auto mode for emcee sampler is not fully developed. The goal is to implement the auto stop feature found in most Bayesian inference codes like Cobaya.

  • The MetropolisHastings.py sampler is premature. Avoid using it for serious projects.

Coming Soon

  • Full CAMB-based CMB power spectrum likelihoods (preliminary interface already in place).

Requirements

  • Python ≥ 3.10

  • See pyproject.toml for the full list of dependencies.

  • A Fortran compiler (gfortran) is only needed if building PolyChordLite.

License

This project is licensed under the GNU General Public License v3.0. See LICENSE for details.

Citation

If you use COSMIX in your research, please cite the following:

@software{kolhatkar_cosmix_2026,
  author       = {Kolhatkar, Ameya},
  title        = {{COSMIX: Cosmological Modular Inference Explorer}},
  year         = 2026,
  publisher    = {Zenodo},
  version      = {v1.5.1},
  doi          = {10.5281/zenodo.19791571},
  url          = {https://doi.org/10.5281/zenodo.19791571}
}

AI Assistance

Parts of the code and documentation were written with the assistance of Claude Code (Anthropic). All the design decisions, physics and validation are the author’s, and all the code was reviewed by the author.

Contact

If there are any issues or suggestions for the code please feel free to contact me at
kolhatkarameya1996@gmail.com