Spectral Energy Distribution Analysis for Habitable Worlds Observatory Target Stars

The paper presents a spectral energy distribution (SED) analysis for the 164 stars in the Habitable Worlds Observatory (HWO) Exoplanet Exploration Program (ExEP) target list. The authors use χ2-minimization fits of empirical spectral templates from the Pickles (1998) stellar spectral flux library to broadband photometric data spanning UV through mid-IR wavelengths. The SED fits provide direct measurements of bolometric flux (Fbol) for each star, which is described as the primary independent product of this work.

For effective temperatures, the authors adopt spectroscopic Teff values from the PASTEL catalog for 127 of the 164 stars (77% of the sample), with the median number of reported PASTEL determinations per matched star being 17 and the median error-weighted uncertainty on the adopted Teff being 40 K. For the remaining 37 stars without PASTEL determinations, they adopt Teff values from the HWO ExEP catalog. These adopted temperatures are combined with the SED-derived Fbol to calculate angular diameters (θ⋆) via the Stefan-Boltzmann equation.

As a consistency check, the authors compare the Pickles template Teff with the adopted spectroscopic values. They find agreement to within 3% for 65% of the sample, with a median absolute offset of ≈135 K (∼2.3% for a typical solar-type star). There is a systematic tendency for the templates to yield cooler values: 76% of the template values fall below the adopted value, with a median signed offset of −112 K. This bias is attributed to the coarse discretization of the Pickles template grid combined with metallicity and luminosity-class effects.

The authors identify 48 stars (29% of the sample) with Teff discrepancies exceeding 200 K. The most prominent cases are the B components of close binary systems, including HD 131156 B (ξ Bootis) and HD 165341 B (70 Ophiuchi), where photometric contamination from the brighter primary compromises the SED solution. For these systems, the SED-derived Fbol values for the B components are nearly identical to those of their respective A components, strongly suggesting contamination. The components of the 36 Ophiuchi system (HD 155885 and HD 155886) exhibit θ⋆ discrepancies of 40% and 37%, respectively. The paper also notes HD 4614 A (η Cassiopeiae A) as a counterexample where the SED-derived value (1.668 mas) is substantially closer to a direct interferometric measurement (1.623 ± 0.004 mas) than the ExEP catalog value (1.983 mas).

The paper discusses implications for Habitable Zone (HZ) boundary calculations, noting that a shift of 200 K in Teff at constant luminosity translates to a displacement of HZ boundaries by several percent in orbital distance, corresponding to 1–3 mas at 10 pc, comparable to the anticipated inner working angle of HWO's coronagraph. For example, the conservative HZ inner boundary for HD 69830 shifts outward by 2% when the SED template temperature is adopted, corresponding to 1.3 mas. The authors also discuss implications for atmospheric retrieval of directly imaged planets, noting that a 200 K offset for a solar-type star alters the assumed UV flux by 20–30% while affecting the near-infrared by only 5%, which can bias the inferred strengths of biosignature absorption features.

The paper concludes that this SED catalog constitutes a uniform set of empirically determined Fbol values and associated stellar parameters for the HWO target sample. The supplementary products include complete photometry files for each star and individual SED fits. The authors note that for targets with interferometric θ⋆ from the CHARA Array or other facilities, the Fbol values can be combined with measured diameters to yield fully empirical Teff determinations independent of the adopted spectroscopic temperature scale.

Improvements to AI Systems:

1. Stellar Parameter Inference with Uncertainty-Aware Template Fitting

• Improve AI models for SED fitting by incorporating Bayesian or Monte Carlo methods that account for template discretization bias (e.g., the systematic −112 K offset). The AI can learn to correct for metallicity and luminosity-class effects by training on the 48 discrepant cases, reducing template-induced temperature errors from >200 K to <50 K for binary-contaminated or metal-poor stars.

2. Binary Contamination Detection and Deblending

• Develop an AI classifier that flags photometric contamination in close binary systems (e.g., HD 131156 B, HD 165341 B) by comparing SED-derived Fbol ratios between components. The improved system can automatically reject or re-fit contaminated SEDs, preventing erroneous angular diameter and habitable zone calculations.

3. Cross-Validation of Stellar Parameters via Multi-Method Fusion

• Build an AI ensemble that fuses SED-derived Fbol, spectroscopic Teff (from PASTEL), and interferometric θ⋆ (e.g., CHARA) to produce self-consistent stellar parameters. The system can detect outliers (like HD 4614 A) and assign confidence scores, improving accuracy for targets lacking direct measurements.

4. Habitable Zone Boundary Prediction with Temperature Sensitivity

• Enhance AI models for HZ calculations to propagate Teff uncertainties (e.g., ±200 K) into orbital distance predictions. The improved system can output probabilistic HZ boundaries (e.g., inner edge shifts of 1–3 mas at 10 pc) and flag targets where template temperature biases would misplace a planet’s detectability.

5. Atmospheric Retrieval Bias Correction

• Train AI retrieval models to correct for stellar Teff-induced flux biases in exoplanet spectra. Specifically, the system can adjust UV flux assumptions by ±20–30% when Teff is uncertain, preventing false positives/negatives in biosignature absorption features (e.g., O2, O3) for directly imaged planets.

6. Automated Catalog Generation and Quality Control

• Create an AI pipeline that automatically generates uniform SED catalogs (like this one) for new target lists, including photometric data compilation, template fitting, and error propagation. The system can flag stars needing interferometric follow-up and produce ready-to-use Fbol and θ⋆ values for mission planning.

7. Template Grid Optimization

• Use machine learning to interpolate or generate synthetic spectral templates that fill gaps in the Pickles grid, reducing the systematic cool-temperature bias. The improved AI can create continuous Teff–metallicity–luminosity parameterizations, enabling sub-1% temperature accuracy for all spectral types.