🔭 astrophysics

ForMoSA: Forward Modeling tool for Spectral Analysis

This paper introduces ForMoSA v2.0, an open-source Python package developed through international collaboration that utilizes a Bayesian framework and self-consistent atmospheric models to robustly fit spectroscopic and photometric data for directly imaged young planetary-mass objects and exoplanets.

Original authors: ForMoSA Collaboration, Simon Petrus, Paulina Palma-Bifani, Matthieu Ravet, Allan Denis, Bhavesh Rajpoot, Mickaël Bonnefoy, Gaël Chauvin, Arthur Vigan, Alice Radcliffe, Pablo Requeijo, Kevin Hoy

Published 2026-07-30

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: ForMoSA Collaboration, Simon Petrus, Paulina Palma-Bifani, Matthieu Ravet, Allan Denis, Bhavesh Rajpoot, Mickaël Bonnefoy, Gaël Chauvin, Arthur Vigan, Alice Radcliffe, Pablo Requeijo, Kevin Hoy

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine the universe as a giant, cosmic library where every star and planet is a book written in a language of light. For decades, astronomers have been trying to read these books, but the pages are often blurry, torn, or written in a code that changes depending on how the light travels through space. This is the world of exoplanet science: the hunt to understand the atmospheres of planets orbiting other stars. To do this, scientists use two main tricks. The first is "free retrieval," which is like trying to guess the ingredients of a soup by tasting it and calculating the chemistry on the fly; it's flexible but takes a massive amount of brainpower and time. The second is "forward modeling," which is more like having a massive cookbook of pre-made recipes. You don't cook from scratch; instead, you flip through the pages to find the recipe that looks most like your soup. This paper introduces a new, super-smart tool designed specifically to help astronomers flip through that cookbook much faster and more accurately than before.

The paper presents ForMoSA (Forward Modeling tool for Spectral Analysis), a new open-source software package that acts like a high-speed, super-organized librarian for these cosmic cookbooks. Think of the "cookbook" as a giant grid of pre-computed atmospheric models—thousands of theoretical recipes for what a young, hot planet or brown dwarf might look like. In the past, matching a real observation to these recipes was like trying to find a specific needle in a haystack while wearing blindfolded gloves. ForMoSA changes the game by using a clever statistical method called "nested sampling." Instead of guessing randomly, it systematically explores the "recipe book," narrowing down the possibilities until it finds the perfect match for the data.

The creators of ForMoSA, a team of scientists from France, Germany, the USA, and Chile, built this tool to handle the messy reality of modern astronomy. Real observations are rarely perfect; they come from different telescopes, at different times, and with different levels of clarity. ForMoSA is designed to be a "data blender," capable of mixing these diverse datasets together without getting confused. It can take a blurry photo from one telescope and a sharp spectrum from another, blend them, and tell you exactly what the planet's atmosphere is made of. It's like being able to combine a sketch, a photo, and a video of a suspect to get a perfect description, even if the sources don't perfectly agree.

The paper highlights that ForMoSA is particularly good at handling "heterogeneous" data—meaning data that is all over the place. It uses a special module called MOSAIC to fix the little errors that happen when you combine data from different instruments, ensuring the final picture isn't distorted by the tools used to take it. The software is also incredibly fast. While older methods might take forever to crunch the numbers, ForMoSA uses modern computing tricks to zoom through the possibilities. The authors tested its speed and found that the time it takes to solve a problem depends heavily on how detailed the data is (spectral resolution) and how noisy the signal is. They showed that for complex problems with many variables, it works best when paired with a powerful engine called PyMultiNest, which is like giving the librarian a jetpack instead of a bicycle.

Since its start in 2020, ForMoSA has already been used to study a roster of famous "celebrity" planets, including HIP 65426 b, AB Pic b, and VHS 1256 b. It has helped scientists figure out the atmospheric properties of these distant worlds, from their temperatures to the presence of clouds and dust. The tool is so effective that it has allowed researchers to analyze huge libraries of brown dwarfs and exoplanets all at once, spotting patterns that were previously hidden. The team emphasizes that this is an active, evolving project. Just as a video game gets new updates, ForMoSA is being constantly improved to handle data from future, even more powerful telescopes. The authors are clear that while the tool is powerful, it is a means to an end: a way to turn the chaotic noise of starlight into a clear story about what these alien worlds are made of. They have made the code free for everyone to use, hoping that other scientists will join in to refine the recipes and explore the universe together.

Technical Summary of ForMoSA: Forward Modeling tool for Spectral Analysis

Problem Statement
The paper addresses the challenge of bridging the gap between high-quality, multi-wavelength observations of exoplanet atmospheres (ranging from broadband photometry to high-resolution spectroscopy) and pre-computed atmospheric models. While "free retrieval" frameworks exist that compute radiative transfer on the fly, they are often computationally expensive. Conversely, existing forward modeling tools often lack the modularity to handle heterogeneous datasets from multiple instruments or the flexibility to incorporate specific observational uncertainties and systematics. The authors identify a need for a robust, Bayesian forward modeling tool capable of efficiently exploring parameter spaces using self-consistent atmospheric model grids, specifically tailored for directly imaged young planetary-mass brown dwarfs and exoplanets.

Methodology
ForMoSA (FORward MOdeling tool for Spectral Analysis) is an open-source Python package designed to fit spectroscopic and photometric observations using a Bayesian framework. The core methodology relies on nested sampling (Skilling, 2004) to explore the parameter space of pre-computed, self-consistent atmospheric model grids.

Key technical components include:

Architecture: Version 2.0 features a class-based architecture organized into modules for configuration, observation handling, filters, grid management, parameter definition, and nested sampling. It utilizes xarray for fast interpolation and manipulation of large model grids.
Workflow: The user provides an atmospheric model grid, observational data, and a configuration file (.ini). A central Analysis class manages sub-grids tailored to specific observations. During nested sampling, random parameters are drawn from priors, interpolated to generate model spectra, and compared to observations via a log-likelihood function.
Likelihood and Systematics: The code supports flexible likelihood mappings to account for specific observational uncertainties. It includes a dedicated MOSAIC module to jointly fit multi-instrument, heterogeneous datasets. This module uses customized likelihood scaling and parametric covariance models to mitigate inter-calibration offsets and complex noise structures.
High-Contrast Module: A specific module incorporates the method of Landman et al. (2024) to fit contrast-limited data by including reference stellar, atmospheric transmission, and systematics spectra.
Parameterization: Beyond standard atmospheric parameters, the tool allows for the inclusion of extra-grid parameters such as radial velocity, rotational broadening, and scaling factors.

Key Contributions
The paper presents ForMoSA v2.0, which introduces several technical advancements over previous iterations and competing tools:

Refactored Architecture: The code has been restructured into a modular, class-based system with user-friendly features, moving away from previous implementations to better handle data heterogeneity.
MOSAIC Framework: A novel approach for the simultaneous analysis of diverse, overlapping datasets from different instruments and times, addressing biases introduced by combining heterogeneous data.
Performance Optimization: The tool leverages xarray for efficient grid interpolation and offers a choice between nested sampling algorithms (PyMultiNest and Nestle), with benchmarks showing PyMultiNest generally provides superior scaling for high-dimensional fits.
Specialized Modules: The inclusion of High-Contrast (HCHR) and MOSAIC modules specifically targets the complexities of direct imaging data, including stellar contamination and calibration offsets.

Results and Performance
The paper provides a performance analysis demonstrating that computational cost is primarily driven by spectral resolution, signal-to-noise ratio (S/N), the number of live points, and the dimensionality of the parameter space.

Spectral Resolution: Higher resolution increases the number of data points, directly increasing the cost of likelihood evaluations.
S/N: Higher S/N leads to sharper likelihood functions, requiring the sampler to explore the parameter space with finer precision, thus increasing inversion time.
Dimensionality: Inversion time scales significantly with the number of free parameters due to the growing volume of the prior space.
Algorithm Comparison: Benchmarks indicate that while Nestle is a convenient pure-Python implementation, PyMultiNest (utilizing the Fortran MultiNest library) demonstrates superior scaling and efficiency, particularly for fits with more than five parameters.

Significance and Claims
The authors claim that ForMoSA has become a highly adaptable tool for the atmospheric characterization of substellar objects since its initiation in 2020. Its primary significance lies in its ability to:

Jointly Fit Heterogeneous Data: By maximizing wavelength coverage through the simultaneous fitting of multi-instrument datasets, ForMoSA enables the accurate characterization of benchmark companions such as HIP 65426 b, AB Pic b, VHS 1256 b, AF Lep b, and $\beta$ Pic b.
Enable Large-Scale Analysis: The code's speed facilitates the homogeneous analysis of large spectral libraries (e.g., brown dwarfs and planetary-mass companions), aiding in the identification of parameter trends and structural systematics in atmospheric models.
Extract Precision Constraints: The High-Contrast module allows for the extraction of precise radial velocities and detailed orbital constraints from contrast-limited, stellar-contaminated data (e.g., VLT/HiRISE).

The paper positions ForMoSA as a specialized, modular inversion framework optimized for data heterogeneity and statistical flexibility, complementing existing "free retrieval" codes by offering a faster, grid-interpolated alternative. The authors note that the software is an active project, with future developments planned to accommodate data from upcoming observatories (ELTs, HWO) and updated atmospheric physics.

More like this