# BeAR code runs exoplanet atmosphere retrievals on GPU: 17-parameter fit of WASP-77Ab in 14 minutes

- Published: 2026-11-06 05:06 UTC
- Section: Sky & Space › Astronomy
- Tags: BeAR retrieval code, exoplanet atmosphere retrieval, WASP-77Ab, high-resolution spectroscopy, joint retrieval, GPU-accelerated code, Gemini South IGRINS, JWST NIRSpec, cross-correlation likelihood
- URL: https://realtech.media/news/rt-531e8c8c9ac279db2d5701b8_bear_code_runs/

**Authors of the Bern code BeAR report GPU retrievals of high- and low-resolution exoplanet spectra; one forward model takes about 1 ms. Results for WASP-77Ab match published values.**

![BeAR code runs exoplanet atmosphere retrievals on GPU: 17-parameter fit of WASP-77Ab in 14 minutes](https://sgzwpqqelzmajkcubgkl.supabase.co/storage/v1/object/public/rt-media/2026/11/126d5f31be096367.jpg)
*LSST, Wikimedia Commons, CC BY 4.0*

A team describing the Bern Atmospheric Retrieval code (BeAR), the successor of Helios-r2, presented a GPU-based framework for retrieving exoplanet atmospheres from high-resolution, low-resolution and joint spectra. The work is an arXiv · preprint and has not yet been peer reviewed. The code is open source and available on GitHub [1].

The authors tested it on dayside spectra of the hot Jupiter WASP-77Ab, taken with Gemini South/IGRINS and JWST/NIRSpec [1].

## How the framework works

BeAR performs joint retrievals by adding the log-likelihoods of high- and low-resolution data, both computed from the same model atmosphere [1]. The high-resolution part implements established cross-correlation log-likelihoods, with the line-depth scaling either fitted analytically or sampled as a free parameter. It also includes their per-pixel generalization, fast model filtering, and model re-injection adapted to that filtering [1].

The forward model and the likelihood run entirely on a graphics processing unit. One forward model takes about 1 ms on an NVIDIA GH200, which, according to the authors, allows millions of models per minute on a GPU cluster [1]. For comparison, an earlier retrieval of the IGRINS data of WASP-77Ab took about three days on 24 CPU cores and one GPU [1].

## Results for WASP-77Ab

In separate retrievals of the IGRINS and NIRSpec spectra, BeAR reproduces published abundances, C/O ratios and metallicities to within 1.5 sigma. The preference for 13CO, the thermal structure and the line contrast agree to within 1 sigma [1]. The joint retrieval confirms a sub-solar metallicity and a near-solar C/O ratio [1].

A full IGRINS retrieval with 17 free parameters takes 27 to 42 minutes on a single NVIDIA GeForce RTX 5080, or about 14 minutes on eight RTX 4090 GPUs [1].

Injection–recovery tests show a limitation: the posteriors underestimate the uncertainties of abundance ratios by about a factor of two [1]. The authors note that high-resolution results depend on analysis choices, such as the removal of telluric and stellar lines, and therefore need to be checked with independent codes. Few open-source codes exist for this, and runtimes of hours to days make such checks expensive [1].

## Sources

1. [arXiv · препринт](https://arxiv.org/abs/2611.08489v1)

---

Source: RealTech, realtech.media — https://realtech.media/news/rt-531e8c8c9ac279db2d5701b8_bear_code_runs/