Hd209458b Retrieval Eval

hd209458b-retrieval-eval

A Detailed Investigation of HD 209458 b HST & JWST Transmission Spectra with SANSAR — Avinash Verma et al. (arXiv:2505.04413, 2025)

What this evaluates

Evaluates atmospheric retrieval models on exoplanet transmission spectra to constrain chemical abundances, temperature-pressure profiles, and cloud properties. It probes the model's ability to disentangle spectral features across multi-wavelength observations and quantify detection significance of trace gases.

Datasets

• JWST NIRCam transmission spectra — total ?; splits: test (-1)
• HST WFC3 transmission spectra — total ?; splits: test (-1)
• HST STIS transmission spectra — total ?; splits: test (-1)

Metrics

• reduced chi-squared (χ²_red) (primary) — range: other
  • Measures goodness-of-fit between the model's predicted transmission spectrum and the observed data, normalized by degrees of freedom. Values near 1 indicate a good fit.
• detection significance — range: other
  • Calculated by comparing the Bayesian evidence of a retrieval model including a specific molecule against a model excluding it. Reported in units of standard deviations (σ).

Input / output format

Input: Multi-wavelength transmission spectra observations (flux vs. wavelength) from JWST NIRCam (2.3–5.1 μm), HST WFC3 (1.1–1.7 μm), and HST STIS (0.3–0.8 μm), along with fixed planetary/stellar parameters (radius, temperature, gravity, distance).

Output: Posterior distributions for atmospheric parameters: logarithmic mixing ratios of trace gases (H2O, CO2, Na, K, etc.), pressure-temperature profile parameters, cloud/haze parameters, and reference radius. Reported as median values with 1σ uncertainties.

Scoring recipe

1. Run Bayesian retrieval (e.g., PyMultiNest) to sample posterior distributions for all free parameters.
2. Compute χ²_red = Σ[(obs_i - model_i)² / σ_i²] / DOF for the best-fit model.
3. For detection significance: run two retrievals per molecule (one with, one without the molecule's free abundance parameter).
4. Calculate Δln(Z) = ln(Z_with) - ln(Z_without) using Bayesian evidence.
5. Convert Δln(Z) to detection significance in σ using standard Bayesian model comparison thresholds.

Common pitfalls

• NIRCam-only retrievals overestimate metallicity and C/O ratios due to insufficient spectral coverage; optical/UV baseline data are critical.
• Large uncertainties in HST WFC3 observations dominate H2O abundance constraints, creating strong degeneracies with CO2 and cloud parameters.
• Detection significance is derived from Bayesian evidence ratios, not frequentist p-values, requiring careful prior specification.

Evidence (verbatim from paper)

The detection significance of H2O, CO2, Na and K are 12.6σ, 7.8σ, 6.9σ and 4.9σ, respectively and other stastical parameters are provided in Table 3. This detection significance has been calculated by comparing the Bayesian evidence from the retrieval run with all the parameters and that without the respective molecule.

Citation

@misc{verma2025sansasr,
  title={A Detailed Investigation of HD 209458 b HST & JWST Transmission Spectra with SANSAR},
  author={Avinash Verma et al.},
  year={2025},
  note={arXiv:2505.04413}
}

• arXiv: 2505.04413