Forward Modeling Tool for Spectral Analysis (ForMoSA)

Forward Modeling Tool for Spectral Analysis (ForMoSA)#

ForMoSA is an open-source Python package for modeling exoplanetary atmospheres using a forward modeling approach. It compares observed spectra and photometry against grids of atmospheric models via nested sampling to derive posterior distributions on physical parameters.

Quick links: Getting Started | Tutorials | Scaling | ForMoSA API | What’s New in v2.0.0

Features#

  • Class-based API — a single Analysis entry point manages the full lifecycle: grid adaptation, nested sampling, and plotting.

  • Multi-instrument support (MOSAIC) — fit spectroscopic and photometric data simultaneously from multiple instruments with per-instrument intercalibration.

  • Three nested-sampling back-ends — nestle, PyMultiNest, and UltraNest.

  • High-contrast mode — model stellar speckles and systematics alongside the companion signal.

  • Automatic photometry filters — filter curves retrieved and cached on the fly from the SVO Filter Profile Service.

  • Configurable priors — uniform, log-uniform, Gaussian, and constant priors for every fitted parameter.

  • Comprehensive plotting — corner plots, chain diagnostics, radar diagrams, best-fit spectra, CCFs, and \(\mathrm{RV}-v \sin i\) maps.

  • Flexible configuration — Python dataclasses or INI files; both load into the same objects.

Why ForMoSA?#

🔭

One tool, every instrument. From low-resolution photometry to high-resolution cross-correlation spectroscopy (GRAVITY, HiRISE, JWST NIRSpec), ForMoSA handles them all in a single fit via its MOSAIC multi-instrument mode.

🪐

Forward modeling on self-consistent grids. Rather than parameterising spectra on the fly, ForMoSA interpolates pre-computed, physically motivated atmospheric model grids (BT-Settl, Sonora, ATMO, ExoREM, …). This anchors retrieved parameters to established theory.

📐

Proper Bayesian uncertainties, not best-fit guesses. Nested sampling (PyMultiNest, UltraNest, nestle) delivers joint posteriors and log-evidence for rigorous model comparison — not just a χ² minimum.

⚡

Fast to set up, fast to run. Grid adaptation is parallelised out of the box (joblib / Dask / Ray). A typical fit on a laptop takes minutes to hours, not days. Cluster scaling via MPI is supported through PyMultiNest.

🔬

High-contrast ready. Native support for the HCHR (High-Contrast High-Resolution) regime: stellar speckle templates, telluric transmission weighting, and systematic component decorrelation are first-class features.

🛠

Hackable and extensible. Every stage — observation loading, grid adaptation, likelihood evaluation, plotting — is a clean Python class. Custom likelihoods, new parameters, and novel grid formats can be plugged in without touching core code.

Statement of Need#

Recent advances in ground- and space-based observatories now enable routine, high-quality observations of exoplanet atmospheres across a wide range of wavelengths and resolutions. ForMoSA was developed to bridge the gap between these observations and atmospheric models by providing a Bayesian framework to robustly compare the two.

Unlike retrieval approaches that generate spectra on the fly using highly parameterized models, ForMoSA adopts a forward-modeling approach based on pre-computed, self-consistent atmospheric model grids — enabling a direct, model-driven comparison with physically motivated theoretical predictions. The framework supports simultaneous analysis of heterogeneous datasets (different instruments, resolutions, and epochs) within a single generalized statistical framework, making it well-suited for the next generation of high-contrast, high-resolution instruments.

How Does ForMoSA Work?#

At its core, ForMoSA interpolates a pre-computed atmospheric model grid at each nested-sampling iteration, applies physical transformations (Doppler shift, rotational broadening, extinction, scaling), and evaluates a log-likelihood against the observations. The posterior distributions on the fitted parameters emerge naturally from the nested-sampling evidence accumulation.

ForMoSA workflow diagram

ForMoSA workflow. The right-hand shaded area shows the core modules required for nested sampling; the left dark-grey area contains utility and support functions. See the ForMoSA API for full module documentation.#

Example Results#

Attribution#

If you use ForMoSA in your research, please cite Petrus et al. (2023).

Version Track#

  • 2.0.0 Complete rewrite with a class-based API (Analysis), Python dataclass configuration, restructured package layout, automatic photometry filter retrieval, CCF / \(\mathrm{RV}-v \sin i\) analysis, structured logging, and typed error handling. Not backwards-compatible with v1.x. See What’s New in v2.0.0 for the full details.

  • 1.1.6 Addition of high-contrast models, UltraNest, and automatically generated config files.

  • 1.1.2 Multi-instrument (MOSAIC) and high-spectral-resolution support.

  • 1.0.13 First operational release, presented at Cloud Academy 3.

  • 1.0.5 First operational release.

Acknowledgements#

The authors express their sincere thanks to the Code/Astro Workshop, which provided the foundational training necessary to transform ForMoSA into a professional, open-source Python package.

We gratefully acknowledge the funding and support for the ForM-X workshops held in Nice (2023), Heidelberg (2024/2025), and Grenoble (2025), which were instrumental in the development and refinement of the code. We also thank the various laboratories and institutions — especially IPAG, Lagrange, and MPIA — for their continued support.

This work has been supported by the French National Research Agency (ANR) through the MIRAGES project (PI: A. Vigan, ANR-20-CE31-0017).

Issues#

Run into something unexpected? Please open an issue on GitHub.