A black hole star at cosmic noon: Extreme Balmer break, photospheric continuum, and broad absorption by thick winds in a Little Red Dot at z = 1.7

Torralba Torregrosa, Alberto; Matthee, Jorryt J; Weibel, Andrea; Naidu, Rohan P.; Ma, Yilun; Cloonan, Aidan P.; Desai, Aayush A; De Graaff, Anna; Greene, Jenny E.; Jespersen, Christian Kragh; Kramarenko, Ivan; Mascia, Sara; Oesch, Pascal A.; Sun, Wendy Q.; Williams, Christina C.

A black hole star at cosmic noon: Extreme Balmer break, photospheric continuum, and broad absorption by thick winds in a Little Red Dot at z = 1.7

Torralba Torregrosa A, Matthee JJ, Weibel A, Naidu RP, Ma Y, Cloonan AP, Desai AA, De Graaff A, Greene JE, Jespersen CK, Kramarenko I, Mascia S, Oesch PA, Sun WQ, Williams CC. 2026. A black hole star at cosmic noon: Extreme Balmer break, photospheric continuum, and broad absorption by thick winds in a Little Red Dot at z = 1.7. The Astrophysical Journal Letters. 1005(2), L37.

Download

2026_AstrophysicalJourLetters_Torralba.pdf 5.42 MB [Published Version]

DOI

10.3847/2041-8213/ae7bfd

Journal Article | Published | English

Scopus indexed

Author

Torralba Torregrosa, Alberto^ISTA ; Matthee, Jorryt J^ISTA ; Weibel, Andrea; Naidu, Rohan P.; Ma, Yilun; Cloonan, Aidan P.; Desai, Aayush A^ISTA; De Graaff, Anna; Greene, Jenny E.; Jespersen, Christian Kragh; Kramarenko, Ivan^ISTA ; Mascia, Sara^ISTA
All

Corresponding author has ISTA affiliation

Department

Matthee Group
Caiazzo Group
Graduate School

Grant

Young galaxies as tracers and agents of cosmic reionization

Abstract

Recent studies at high redshift have revealed an enigmatic class of little red dots (LRDs) with extreme Balmer breaks, stronger than in any stellar atmosphere. However, it is unclear whether such objects exist at lower redshift, especially given the low number of LRDs reported at z ≲ 2. Here, we report the discovery of PAN-BH*-1, an LRD with an extreme Balmer break at z = 1.73, identified from JWST/NIRCam pure-parallel imaging taken by the PANORAMIC survey, and confirmed by deep VLT/X-Shooter spectroscopy. The rest-optical to near-infrared spectral energy distribution of PAN-BH*-1 is consistent with a photospheric continuum with effective temperature Teff ≈ 4800 K. The broad Hα emission line shows remarkably deep absorption, stronger than previously measured in any LRD. The absorption trough spans from −520 to +267 km s−1 with respect to the systemic redshift. The presence of blue- and red-shifted absorption suggests complex dynamics of the obscuring gas along the line of sight. We speculate that the absorption trough can be produced by a thick wind launched from a thick, rotating photospheric disk, the latter being the source of the red optical continuum. While the source is unresolved in the rest-optical JWST data (reff < 47 pc), the rest-near-UV Hubble Space Telescope imaging shows an extended morphology with (formular displayed) kpc, which we interpret as a host galaxy with a stellar mass of ∼10^8 M⊙, in line with the narrow Hα emission. The discovery of this object at cosmic noon highlights the feasibility of systematic searches for extreme LRDs with wide-area facilities such as Euclid and Roman.

Publishing Year

2026

Date Published

2026-07-10

Journal Title

The Astrophysical Journal Letters

Publisher

IOP Publishing

Acknowledgement

IOP Science home The Astrophysical Journal Letters The American Astronomical Society, find out more. The following article isOpen access A Black Hole Star at Cosmic Noon: Extreme Balmer Break, Photospheric Continuum, and Broad Absorption by Thick Winds in a Little Red Dot at z = 1.7 Alberto Torralba, Jorryt Matthee, Andrea Weibel, Rohan P. Naidu, Yilun Ma, Aidan P. Cloonan, Aayush Desai, Anna de Graaff, Jenny E. Greene, Christian Kragh JespersenShow full author list Published 2026 June 30 • © 2026. The Author(s). Published by the American Astronomical Society. The Astrophysical Journal Letters, Volume 1005, Number 2 Citation Alberto Torralba et al 2026 ApJL 1005 L37 DOI 10.3847/2041-8213/ae7bfd PDFOpens in a new tab.ePub Authors Figures Tables References Article data PDFOpens in a new tab.ePub Article metrics 122 Total downloads Share this article Article information Abstract Recent studies at high redshift have revealed an enigmatic class of little red dots (LRDs) with extreme Balmer breaks, stronger than in any stellar atmosphere. However, it is unclear whether such objects exist at lower redshift, especially given the low number of LRDs reported at z ≲ 2. Here, we report the discovery of PAN-BH*-1, an LRD with an extreme Balmer break at z = 1.73, identified from JWST/NIRCam pure-parallel imaging taken by the PANORAMIC survey, and confirmed by deep VLT/X-Shooter spectroscopy. The rest-optical to near-infrared spectral energy distribution of PAN-BH*-1 is consistent with a photospheric continuum with effective temperature Teff ≈ 4800 K. The broad Hα emission line shows remarkably deep absorption, stronger than previously measured in any LRD. The absorption trough spans from −520 to +267 km s−1 with respect to the systemic redshift. The presence of blue- and red-shifted absorption suggests complex dynamics of the obscuring gas along the line of sight. We speculate that the absorption trough can be produced by a thick wind launched from a thick, rotating photospheric disk, the latter being the source of the red optical continuum. While the source is unresolved in the rest-optical JWST data (reff < 47 pc), the rest-near-UV Hubble Space Telescope imaging shows an extended morphology with kpc, which we interpret as a host galaxy with a stellar mass of ∼108 M⊙, in line with the narrow Hα emission. The discovery of this object at cosmic noon highlights the feasibility of systematic searches for extreme LRDs with wide-area facilities such as Euclid and Roman. Export citation and abstract BibTeXRIS Previous article in issue Next article in issue Original content from this work may be used under the terms of the Creative Commons Attribution 4.0 licence. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI. 1. Introduction The unprecedented sensitivity of JWST has enabled the discovery of a new, abundant population of objects at redshifts z ≈ 3–9 nicknamed the “little red dots” (LRDs). These are characterized by their compact rest-frame optical morphology, broad emission lines, and a characteristic rest-UV to optical “V-shape” in their spectral energy distributions (SED; e.g., D. D. Kocevski et al. 2023; V. Kokorev et al. 2024; J. Matthee et al. 2024; I. Labbe et al. 2025). The nature of LRDs is highly debated (see K. Inayoshi & L. C. Ho 2025, for a recent overview) as the LRDs show systematic differences with respect to other types of active galactic nuclei (AGN), such as faintness in X-rays (e.g., T. T. Ananna et al. 2024; M. Yue et al. 2024), mid-to-far-infrared dust emission (e.g., G. C. K. Leung et al. 2025; C. C. Williams et al. 2024; I. Delvecchio et al. 2025; D. J. Setton et al. 2025; M. Xiao et al. 2025), and radio (e.g., G. Mazzolari et al. 2026; M. A. Latif et al. 2025; K. Perger et al. 2025, see A. J. Gloudemans et al. 2025). A recurring spectral feature of LRDs is the presence of a strong Balmer break (e.g., D. J. Setton et al. 2025; B. Wang et al. 2024; R. E. Hviding et al. 2025; W. Q. Sun et al. 2026), in some cases stronger than any star or stellar population can produce. The two most prominent examples known to date are The Cliff at z ≈ 3.5 (A. de Graaff et al. 2025a) and MoM-BH* at z ≈ 7.8 (R. P. Naidu et al. 2025). The joint appearance of strong Balmer lines as well as strong Balmer breaks has been modeled as being due to absorption by a dense, neutral gas with a high column density in the line of sight to a highly ionizing source (K. Inayoshi & R. Maiolino 2025; X. Ji et al. 2025; A. Sneppen et al. 2026; A. Torralba et al. 2026). These observations have sparked the development of new theoretical models, ranging from a spherical envelope analogous to stellar atmospheres (e.g., M. C. Begelman & J. Dexter 2026; D. Kido et al. 2025; H. Liu et al. 2025; D. Nandal & A. Loeb 2026) or a thick accretion disk (e.g., H. Liu et al. 2025, 2026; K. Inayoshi et al. 2025; Y.-X. Chen et al. 2026). Besides their spectral features, the evolution of the LRD number densities is also in stark contrast to other types of AGNs (e.g., K. Inayoshi 2025). At 4 ≲ z ≲ 7, LRDs represent a few percent of the galaxy population (e.g., D. D. Kocevski et al. 2023, 2025; J. E. Greene et al. 2024; V. Kokorev et al. 2024; X. Lin et al. 2024; R. Maiolino et al. 2024; J. Matthee et al. 2024), with number densities of ≳10−5 Mpc−3. The number density does not appear to drop quickly beyond z > 5 (e.g., J. Zhang et al. 2026), with various LRDs having been confirmed at z > 8 (V. Kokorev et al. 2023; A. J. Taylor et al. 2025; R. Tripodi et al. 2025), well beyond the quasar redshift record (F. Wang et al. 2021). Photometric LRD candidates exist beyond z > 10 (T. S. Tanaka et al. 2025). In turn, the number density of LRDs seems to decline steeply at z < 4 (e.g., Y. Ma et al. 2026), with estimates of a number density of ∼10−6 cMpc−3 at z ∼ 2 and even ∼10−10 cMpc−3 at z ≈ 0.3 (X. Lin et al. 2026). While it is challenging to ensure a uniform selection function across such a large redshift baseline and dedicated spectroscopic follow-up of such lower redshift candidates has only just started, it is challenging to attribute five orders of magnitude to such effects. Motivated by the discovery of rare objects with extreme Balmer breaks at z > 3 and the very small number of known LRDs at lower redshift, we performed a dedicated search for extreme Balmer break objects using a template-match approach on a large compilation of JWST NIRCam data over ≈0.3 deg2 and z ≈ 1.5–7.0. This survey is presented in A. Weibel et al. (2026a). As part of an ongoing ground-based spectroscopic campaign of LRD candidates at z ∼ 2 (Y. Ma et al. 2026), we followed up the most luminous candidate with a photometric redshift of z ≈ 2 with the X-Shooter spectrograph on the Very Large Telescope (VLT). In this Letter, we present the discovery and spectroscopic confirmation of PAN-BH*-1, a luminous LRD at z = 1.73 with an extreme Balmer break comparable to the strongest observed in any LRD (and, in general, any astrophysical source). The low redshift of this source enables high-resolution spectroscopy from ground-based observatories that is otherwise impossible to obtain at high redshift. Throughout this Letter, we use a ΛCDM cosmology with Ωm = 0.31, ΩΛ = 0.69, and h = 0.677 as described by Planck Collaboration et al. (2020). All the magnitudes are given in the AB system (J. B. Oke & J. E. Gunn 1983). 2. Observations 2.1. Photometry and Source Selection We identified PAN-BH*-1 (ID: PAN-1115, RA, DEC: 40.015835, −1.659363 J2000) as part of a systematic search across ≈0.3 deg2 of JWST NIRCam legacy imaging comprising at least six filters of coverage (A. Weibel et al. 2026b). Notably, this dataset includes the Cycle 1 pure parallel survey PANORAMIC (PID: 2514, PIs: Williams & Oesch; C. C. Williams et al. 2025) that contributes 28 of the 35 independent lines of sight, thereby enabling the discovery of rare objects such as PAN-BH*-1 across diverse large-scale structure environments. Specifically, this source was identified in the footprint j024000m0142 of the PANORAMIC DR1,15 which is adjacent to the A370 field (G. O. Abell et al. 1989), where archival images by the Hubble Space Telescope (HST) are available from the BUFFALO survey (C. L. Steinhardt et al. 2020). The HST/ACS images were processed with grizli and also released as part of the PANORAMIC dataset. PAN-BH*-1 is in the outskirts of the A370 lensing cluster, but the magnification is only μ ≈ 1.05 according to the models from A. Niemiec et al. (2023). Throughout the rest of the paper, we report the uncorrected flux measurements, since the effect of magnification (∼5%) is negligible given the uncertainties in the observations and the lensing model. The search strategy and full photometric selection are described in a companion paper (A. Weibel et al. 2026a). Briefly, that work presents a new selection of LRDs as a combination of a “black hole star” template (BH*; R. P. Naidu et al. 2025) embedded in a host galaxy, instead of the typically used “V-shaped” selections (e.g., D. D. Kocevski et al. 2025; V. Kokorev et al. 2024). The host galaxies are modeled using eazy’s blue_sfhz templates. The BH*s are modeled using a novel template set comprising empirical luminosity-based stacks constructed in W. Q. Sun et al. (2026), the cloudy template from R. P. Naidu et al. (2025), and by using spectra of prominent LRDs spanning the observed effective temperature range (I. Labbe et al. 2024; A. de Graaff et al. 2025a; B. Wang et al. 2026). PAN-BH*-1 stood out as one of the few sources where the BH* template effectively dominated all the light over the full wavelength range covered by NIRCam (hence the name). The redshift of PAN-BH*-1 was estimated to be zphot = 1.85. Follow-up VLT/X-Shooter spectroscopy confirmed the redshift as zspec = 1.731 (see Section 3.2). PAN-BH*-1 is also covered by archival data from the VLT with the HAWK-I camera in the Ks band (G. B. Brammer et al. 2016) and in data from the Spitzer Space Telescope in IRAC bands 1 and 3 (3.6 and 5.7 μm), and MIPS 24 μm (P. Capak 2019). PAN-BH*-1 is detected in the Ks band and in the two IRAC filters. Performing Spitzer photometry of this source is challenging due to the large point spread function (PSF) and a neighboring source, especially in the MIPS band. However, the NIRCam photometry of the neighboring source suggests it has a limited contribution to the IRAC fluxes. The details of the photometry extraction are described in Appendix A, and the measured magnitudes in Table 2. 2.2. VLT/X-Shooter Spectroscopy PAN-BH*-1 was observed for 5.8 ks with the X-Shooter spectrograph (J. Vernet et al. 2011) on the VLT as a bright backup target for program 116.294D (PI: Matthee) in visitor mode on 2025 December 17. The main aim of this program was to confirm candidate LRDs at cosmic noon (Y. Ma et al. 2026). These observations confirmed the redshift through the detection of Hα at z = 1.731. A DDT program (ID 116.2AQ0; PI: Matthee) obtained additional follow-up data of PAN-BH*-1 in service mode for 26.2 ks during 2026 January 10–26, yielding a total exposure time of 8.9 hr. X-Shooter observes with three arms simultaneously, UVB, VIS, and near-infrared (NIR), covering rest-frame wavelengths of ≈0.14–0.9 μm, albeit hampered by skyline emission and telluric absorption, primarily in the rest-frame optical. The observing conditions were clear, with a seeing ranging from 05 to 07 (median 06). The service mode observations were primarily conducted during dark nights, with some gray (FLI = 0.03–0.6, median 0.1), and a typical airmass of 1.35. We used UVB, VIS, and NIR slits with widths 10, 09, and 09, yielding a nominal resolution of R = 5400, 8900, and 5600, respectively (FWHM ∼53 km s−1 for NIR). The target acquisition was done using blind offsets from a reference star, due to the target being too faint for direct acquisition. We used a standard nodding on the slit pattern, with 4″ nod throws in an ABBA pattern, and 1″ jitters in the NIR arm to improve the sky subtraction. In each observing block of ≈1 hr, the exposure times were 700, 655, and (2×)365 s for the three arms at each nod position. The reduction of the X-Shooter data uses a combination of EsoRex libraries16 and Python code based on the reduction pipeline employed in J. Matthee et al. (2021). Each observing block was reduced separately. We used standard stars taken during the observing night for a first-pass flux calibration. Telluric corrections were applied using the molecfit tool (A. Smette et al. 2015) implemented in the X-Shooter EsoRex pipeline. Telluric stars were observed during the visitor nights, but they were not always observed during the service mode observations in January. For those observations, we took the telluric star that was observed at the closest observing date. Based on the variation in telluric absorption among the reference stars taken during this period, we estimate the variation in the transmission and propagate the uncertainty in the telluric correction. For each observing block, we then extracted an optimally extracted 1D spectrum using the spatial profile of the Hα line, thus accounting for seeing variations and (more importantly) minor errors in the accuracy of the slit pointing. Before median combining these spectra, we normalize them by the median Hα flux of all observations to account for variations in slit losses and flux calibrations. Besides Hα (integrated S/N = 75) and Hβ (integrated S/N = 6; Section 3.2), we also detect continuum emission in the best regions in the H and K bands at 1.6 μm and 2.1 μm, respectively, with a low signal-to-noise ratio (S/N) of ∼1 per resolution element. Unfortunately, the [O iii] λλ4960, 5008 doublet is undetectable because the observed wavelengths are impacted by very strong telluric absorption. No other lines or continuum are detected in the X-Shooter spectrum. 3. Properties of PAN-BH*-1 3.1. Spectral Shape: A Photospheric Continuum with Strong Hα Emission The photometric SED of PAN-BH*-1 has remarkable similarities with The Cliff (Figure 1): luminous in the rest optical, with a sudden drop toward the rest-UV around the Balmer limit, and very weak near-to-mid infrared continuum emission. With a rough extrapolation of the two HST photometric points using a power-law fit (fλ ∝ λβ), we obtain a UV slope of β = −0.1 ± 1.2, and MUV = −16.7 ± 0.7. For the rest-frame optical to NIR data, we fit a Planck blackbody law to the JWST data points, after subtracting the measured Hα flux (see Section 3.2) from the F200W photometry. The rest-optical and NIR photometry of PAN-BH*-1 is remarkably well described by a single temperature blackbody with T = 4204 K (with a best-fit ). We measure the strength of the Balmer break from the fν ratio F115W/F814W = 7 ± 1, in line with the Balmer break strengths of The Cliff (; A. de Graaff et al. 2025a)17 and MoM-BH* (7.8 ± 1.8; R. P. Naidu et al. 2025), measured from JWST/NIRSpec PRISM spectra as fν,4000–4100/fν,3620−3720. In Figure 2, we compare the Balmer break strength with the spectroscopic sample of A. de Graaff et al. (2025b), showing that out of 134 sources, only two have breaks significantly above 5. This suggests that PAN-BH*-1 has among the most extreme Balmer breaks known, although we caution that our value is derived from wide-band photometry with pivot wavelengths corresponding to 4212 and 3042 Å, respectively, rather than from spectroscopy. Zoom InZoom OutReset image size Figure 1. SED of PAN-BH*-1 Top: cutouts from all the HST and JWST images in which PAN-BH*-1 is covered. It shows a remarkably compact morphology in all the wavelengths, resolved only in the HST F606W and F814W bands (Section 3.3). Bottom: photometry from JWST/NIRCam (blue squares), HST/ACS (purple pentagons), and Spitzer/IRAC+MIPS (red hexagons, and red triangle for the 5σ upper limit). The empty square is the F200W flux after subtracting the Hα flux measured from X-Shooter spectroscopy. We show the spectrum of The Cliff for comparison (gray line), shifted to z = 1.73 and normalized to the F150W flux of PAN-BH*-1. We also show the best-fitting blackbody spectrum (blue dashed line) and the best model from the synthetic LRD atmosphere models from H. Liu et al. (2026), shifted to z = 1.73 (green line), undersampled by a factor of 500 for clarity. Download figure: Standard imageHigh-resolution image Zoom InZoom OutReset image size Figure 2. Spectroscopic sample of LRDs by redshift and Balmer break strength. We plot the redshift and Balmer break strength of PAN-BH*-1, and the JWST sample from A. de Graaff et al. (2025b) (purple diamonds), and three local LRDs in X. Lin et al. (2026), for comparison. We also highlight three sources with a particularly strong Balmer break: The Cliff (A. de Graaff et al. 2025a), MoM-BH* (R. P. Naidu et al. 2025), and CAPERS-LRDz9 (A. J. Taylor et al. 2025). The Balmer break strength of the JWST spectroscopic sample is computed as fν,4000–4100/fν,3620–3720, whereas the value for PAN-BH*-1 is directly obtained from the F115W/F814W photometry. Download figure: Standard imageHigh-resolution image 3.2. Hα and Hβ Emission Lines The Hα profile appears as a complex combination of a broad line with strong absorption close to the systemic redshift. We fit the Hα emission line with a similar model as the one used in A. Torralba et al. (2026) and J. Matthee et al. (2026). The Hα model consists of two Gaussian emission components (with narrow and intermediate line widths), and a broad symmetric exponential convolved with the intermediate profile and parameterized as in F. D’Eugenio et al. (2025a). The absorption is implemented as an opacity law defined as e−τ(λ), where τ(λ) also follows a single Gaussian velocity distribution (F. D’Eugenio et al. 2025a, 2025b; A. Torralba et al. 2026). For simplicity, we assume a covering factor of Cf = 1 for the absorbing gas. In previous works, the width of the narrow component is tied to that of [O iii], assuming both components come from the same region, often interpreted as the interstellar medium (ISM) of the host galaxy. In this case, we have no information about [O iii] due to this doublet falling in a wavelength range heavily affected by strong telluric absorption. We fit the Hα line after masking relevant skylines and strong telluric absorption bands. The fitted Hα parameters are listed in Table 1 and the best-fit model is shown in Figure 3. The absorption feature is notably strong, with an equivalent width of EWabs = −148 ± 12 Å with respect to the fitted continuum and 12.2 ± 0.2 Å if including the broad emission component. The absorption corresponds to a Balmer optical depth at the line center of , reaching roughly the continuum level. The FWHM of the single Gaussian fitted to the absorber is 283 ± 8 km s−1, and is offset from the systemic redshift by −94 ± 4 km s−1. We note that this parameterization is somewhat arbitrary, and we discuss in detail the absorber properties in Section 4. Zoom InZoom OutReset image size Figure 3. Hα spectrum, and the best fit to our fiducial model. We show the X-Shooter R ∼ 5600 spectrum of the Hα line of PAN-BH*-1, along with the best-fit to the model described in Section 3.2; total model (red solid line) and individual components (discontinuous color lines). The red wing of the line is severely affected by telluric absorption, thus the large uncertainties. Download figure: Standard imageHigh-resolution image Table 1. Properties of PAN-BH*-1 Parameter Value Unit Width (FWHM; Hα) Exponential 1257 ± 27 km s−1 Intermediate 687 ± 43 km s−1 Narrow 184 ± 12 km s−1 Absorption 283 ± 8 km s−1 Flux (Hα) Exponential 643 ± 7 10−18 erg s−1 cm−2 Intermediate 19 ± 5 10−18 erg s−1 cm−2 Narrow 38 ± 3 10−18 erg s−1 cm−2 Total 522 ± 7 10−18 erg s−1 cm−2 General properties (LHα/erg s−1) 43.046 ± 0.006 ⋯ EW0(Hα) 520 ± 20 Å SFR(Hα, narrow)a 2.1 ± 0.2 M⊙ yr−1 SFR(Hα, narrow)b 3.3 ± 0.3 M⊙ yr−1 reff,UV (F606W+F814W) kpc reff,opt (F200W) <0.047 kpc Hα/Hβ (total) >9.4 ⋯ Hα/Hβ (narrow) 5 ± 1 ⋯ Notes. aCalibration from I. G. Kramarenko et al. (2026). bCalibration from R. C. Kennicutt & N. J. Evans (2012). SFR values calculated assuming no dust attenuation. Download table as: ASCIITypeset image The Hβ line is marginally detected. After undersampling the spectrum by a factor 5, a hint of a weak narrow component can be identified (Figure 4), along with a tentative absorption at the same mean velocity as in Hα. We fit the best Hα model to the Hβ spectrum, only rescaling it by a multiplicative factor, and adding a flat continuum component. By doing this, we find an Hβ flux of (47 ± 8) × 10−18 erg s−1 cm−2 (S/N ≈ 6). Conservatively, we obtain a Balmer decrement of Hα/Hβ > 9.4 (at a 3σ confidence level), in line with the high decrements found for the LRD population (e.g., G. P. Nikopoulos et al. 2026; A. de Graaff et al. 2025b; J. Matthee et al. 2026). In Figure 5, we show the Hβ spectrum compared to the rescaled Hα model. By matching the best-fit Hα profile with the data at the expected observed wavelength for Hβ (±5000 km s−1), we obtain a better agreement (, BIC = 1537) than fitting a flat continuum only (, BIC = 1658) with ΔBIC = 121 ≫ 10, strongly favoring a detection of a broad Hβ emission line, and securing the spectroscopic redshift. Similarly, we fit a narrow Gaussian to Hβ with the same width and velocity as the Hα best-fit model, assuming a completely saturated absorption. We obtain a Balmer decrement for the narrow component of Hα/Hβ = 5 ± 1, which would imply a dust extinction of using a J. A. Cardelli et al. (1989) attenuation law, under the assumption of case B recombination. However, due to the low S/N of Hβ this result is only tentative, and compatible with a standard Case B value within ∼2σ. Zoom InZoom OutReset image size Figure 4. X-Shooter spectrum of Hα and Hβ of PAN-BH*-1 (blue). We compare to the spectrum of The Cliff (gray; data from JWST DDT #9433), normalized in each panel to the flux of PAN-BH*-1 in the range v ∈ (−3000, −2000) km s−1. Due to the low S/N, the Hβ spectrum of PAN-BH*-1 is rebinned to a coarser grid by a factor 5, after masking the most relevant skylines. Download figure: Standard imageHigh-resolution image Zoom InZoom OutReset image size Figure 5. Hβ spectrum. The spectrum is rebinned by a factor of 10 with inverse variance flux weighting for visual clarity, due to the low S/N. We compare to the best-fit Hα model, scaled by a factor of 0.112. In the bottom panel, we show the χ residuals between the spectrum and the rescaled Hα model in black, and for only the continuum in pink (ΔBIC = 121 strongly favoring the presence of a broad Hβ line). Download figure: Standard imageHigh-resolution image 3.3. Spatial Morphology In order to assess whether PAN-BH*-1 is spatially resolved, we use the Bayesian profile fitting software pysersic (I. Pasha & T. B. Miller 2023)18 to fit a single Sérsic profile to the JWST and HST imaging data of PAN-BH*-1. For JWST/NIRCam, we choose F200W as the filter with the highest S/N in the short wavelength channel, benefiting from a high spatial resolution and probing rest-frame optical wavelengths. To model its PSF, we use version 2.2.0 of the stpsf software (formerly webbpsf, M. D. Perrin et al. 2014). For the two HST bands F606W and F814W, we instead construct empirical PSFs from public imaging data in the GOODS-S field following A. Weibel et al. (2024). In all three bands, we sample the posterior with the No U-turn sampler in two chains with 1000 warm-up and 2000 sampling steps each. We find that PAN-BH*-1 is unresolved with NIRCam in F200W where the effective radius converges toward the edge of the prior at 0.25 pixels. Using the 95th percentile of the posterior chains as an upper limit on the effective radius, we find a rest-optical size of reff < 47 pc. PAN-BH*-1 appears to be resolved in the HST images corresponding to rest-frame pivot 0.2 and 0.3 μm, respectively. Due to the low signal-to-noise of the F606W and F814W photometry, we fit both bands simultaneously fixing all the morphological parameters in both images. We measure physical effective radii of kpc (see Appendix B). The modest stretching by the foreground A370 lensing cluster could imply a correction of ∼10% to the measured radius (A. Niemiec et al. 2023), which we disregard given the uncertainties. These measured sizes are consistent with the typical sizes for galaxies with a stellar mass ≲ 109 M⊙ at z = 1.75 (A. van der Wel et al. 2014). These findings are consistent with the scenario of a compact LRD “engine” dominating the rest-optical light embedded in a host galaxy, whose contribution becomes significant blueward of the Balmer break (see A. P. Cloonan et al. 2026, for a relevant discussion). 4. Absorber Kinematics As described in Section 3.2, the velocity distribution of the absorber is empirically modeled with a Gaussian, which we find has a central velocity of −94 ± 4 km s−1 relative to the redshift of the narrow emission component (adopted as systemic). The absorption trough extends from negative to positive velocities with respect to the redshift of the narrow component, but also with respect to the center of the symmetric exponential wings. However, there are several degeneracies between the shape of the absorber and other components of the emission line, such as the narrow central emission (see Section 3.2). Furthermore, direct interpretation of the absorber center velocity shift is challenging in an optically thick gas with presumably complex dynamics, and it does not necessarily trace bulk motion. A more robust, physically motivated pair of quantities is the minimum and maximum absorber velocities. We define them as the values where the transmission of the Balmer absorber increases to 99%, km s−1 and km s−1. These values trace the largest velocities in the line of sight of gas with significant Balmer absorption. The absorbing trough extends over 787 ± 17 km s−1 under this definition. The values of and are relatively agnostic to the choice of the shape of the absorber, since they are determined by the wavelength where the line profile deviates from a broad, symmetric exponential profile. In Figure 6, we illustrate three proposed configurations of the velocity distribution of the absorbing gas that could explain the shape of the observed Balmer absorption, and we discuss these scenarios below. Zoom InZoom OutReset image size Figure 6. Geometric configurations for the absorber. We illustrate three scenarios that could give rise to the observed Balmer absorption in PAN-BH*-1. In scenario (a), the obscuring agent is a thick screen of gas with a certain bulk velocity, and turbulent motions produce the broadening of the absorption trough. In (b), there are two (or more) absorbers with opposite velocities in the line of sight. These first two scenarios are dynamically unstable; therefore, variability is expected in the absorption. Lastly, in (c), we observed an extended source through a disk wind with a rotational component (vϕ) in addition to the poloidal (nonazimuthal) velocity (vp). In the last scenario, the redshifted absorption is produced by streamlines that oppose the observer when projected along the line of sight, despite the fact that the gas is outflowing from the central source. Download figure: Standard imageHigh-resolution image 4.1. Unstable Gas Flows? The fact that there is significant absorption at both negative and positive velocities with respect to the systemic redshift cannot be simply explained by an axisymmetric outflowing or inflowing wind. In the case of observing a compact object through a spherically symmetric, nonturbulent bulk flow, a classical P Cygni profile is expected, with a purely blueshifted absorption (or redshifted if the wind is infalling). The fact that we also see redshifted absorption rules out this simple scenario. In principle, turbulent motions could also produce broadening of the absorbing medium (scenario a in Figure 6). However, the required turbulent velocity dispersion σturb ≈ 120 km s−1 (from the Gaussian fit in Section 3.2) is comparable to the mean velocity of the absorption trough, meaning that turbulence dominates the gas flow. In such a case, strong variability of the absorption profile would be expected, given the typical dynamical crossing times (see Sect. 4.1 in F. D’Eugenio et al. 2025b). For example, for a radius of 1016 cm (e.g., A. Torralba et al. 2026) and a mass of 106 M⊙, the dynamical freefall time is yr. Moreover, the turbulent velocity would be highly supersonic, and the dissipation timescale would be comparable to the dynamical time (e.g., M.-M. Mac Low 1999). Alternatively, in the context of a strong Balmer absorber at z ∼ 7, F. D’Eugenio et al. (2026) recently discussed a “breathing mode” scenario with cyclic inflows and outflows along the same line of sight, with the gas being in different phases at different depths (scenario b in Figure 6; see also K. Park et al. 2017). In this case, the same arguments regarding the stability of the absorber would apply, and absorber variability is expected on observed timescales of ∼5 yr (for a source at z = 1.7), which is testable with future observations. 4.2. The Case for the Disk Wind Hypothesis An alternative, dynamically stable scenario is a disk wind configuration (scenario c in Figure 6). Here, the wind would be launched from a thick disk near the central engine, which we speculate could be the source of the optical continuum emission (e.g., H. Liu et al. 2025, 2026; L. Zwick et al. 2025; Y.-X. Chen et al. 2026). A rotating disk would imprint to the wind an azimuthal velocity component (vϕ). Observations at specific lines of sight, particularly for high inclination angles (close to edge-on) where the rotational component dominates the poloidal velocity, can give rise to both blueshifted and redshifted absorption features (D. Proga et al. 2000; P. B. Hall et al. 2002, 2013; D. Proga & T. R. Kallman 2004; M. Giustini & D. Proga 2012). Most observed LRDs have blueshifted P Cygni–like absorbers (J. Matthee et al. 2026), which can be naively interpreted as a uniformly expanding shell. The low incidence of redshifted Balmer absorbers in LRD spectra (e.g., I. Labbe et al. 2024; A. de Graaff et al. 2025a; F. D’Eugenio et al. 2025b, 2026; Y. Ma et al. 2026) can therefore be explained by the requirement of high inclination angles to observe such features (see also A. Sneppen et al. 2026). Such a picture is broadly in line with disk wind models for AGN with broad absorption lines (e.g., P. B. Hall et al. 2002; H. Zhou et al. 2019) and around stars with circumstellar disks (e.g., J. Erkal et al. 2022), such as accreting T Tauri stars (S. Edwards et al. 2006) or cataclysmic variables (D. Proga 2003). 4.3. Implications of Rotating Winds for the Emission Lines of LRDs The disk wind hypothesis would imply that a photosphere in the shape of a rotating disk is the source of the optical continuum emission, and drives winds that can explain the observed absorption trough. Emission lines originating in a thin rotating disk would have a double-peaked profile in the idealized case (for most inclination angles), but this is not necessarily true if the disk is not sufficiently thin (e.g., N. Murray & J. Chiang 1997), for instance, in the case of a puffed-up disk associated with super-Eddington accretion (e.g., H. Liu et al. 2026). In addition, most line emission would not be produced directly at the base of the disk, but slightly outside (e.g., via collisional cooling or residual recombination; A. Torralba et al. 2026), where the rotational velocity is lower, and the dynamics are complex (e.g., G. A. Shields 1977). The Balmer lines of most LRDs are dominated by broad, symmetric exponential components that are associated with broadening by electron scattering (e.g., V. Rusakov et al. 2026; J. Matthee et al. 2026). For PAN-BH*-1, the Hα line profile of PAN-BH*-1 is compatible with a broad exponential profile emerging through a dense wind where the absorption trough is produced. In dense gas with a large column density of neutral hydrogen, and optically thick to Balmer transitions (NHI,2s ≳ 1014 cm−2), resonant scattering effects become important. Crucially, resonant scattering impacts Hα and Hβ differently (e.g., S.-J. Chang et al. 2026), hence the 3D radiative transfer and photon redistribution of both lines may produce different profiles (see, e.g., Figure 2 in D. Proga 2003). Therefore, the empirical fitting and interpretation of the absorption profiles becomes nontrivial. Dedicated radiative transfer modeling is necessary to study such effects, and they can be tested in other emission lines with high optical depth, such as He i λ10830 Å, or resonant lines like C iv λ1550. 5. Implications for the Galaxy and Black Hole Masses 5.1. Properties of the Host Galaxy Assuming that the narrow component of Hα corresponds to ISM emission in the host galaxy, we compute the associated star formation rate using the local calibration from R. C. Kennicutt & N. J. Evans (2012) and assuming no dust attenuation. We obtain SFR(Hα) = 3.3 ± 0.3 M⊙ yr−1. A somewhat lower value of SFR(Hα) = 2.1 ± 0.2 M⊙ yr−1 is obtained using the high-redshift (z ≳ 4) calibrations in I. G. Kramarenko et al. (2026), which might be more appropriate for a young dwarf galaxy with a bursty star formation history. The star formation rates are low, but in line with a main-sequence galaxy with (extrapolating the relation from J. S. Speagle et al. 2014). Assuming zero dust attenuation, the UV absolute magnitude (MUV = −16.7 ± 0.7; Section 3.1) would imply SFR(UV) = 0.18 ± 0.12 M⊙ yr−1 (R. C. Kennicutt & N. J. Evans 2012). The discrepancy between the UV and Hα inferred star formation rate suggests there is some amount of dust attenuation in the host galaxy. We derive a dynamical mass from the width of the narrow component Hα line and the estimated UV size as , adopting the empirical virial correction K(n)K(q) from A. van der Wel et al. (2022), where K(n) and K(q) are functions of the best-fit ellipticity and Sérsic index (see Appendix B). Adopting a Mdyn/M* factor of 40 as found by A. de Graaff et al. (2024) for dwarf galaxies at high redshift, we infer a stellar mass of . However, the Mdyn/M* is very uncertain in this regime, and the uncertainty can span over 1 dex (A. Saldana-Lopez et al. 2025). We advise caution in interpreting this result, as there are large uncertainties in the measurements of the narrow Hα component, the HST morphology, and the empirical relations used. As discussed in Section 4, the absorption profile is compatible with broadening by a rotating disk wind, and numerical modeling of such configurations often predicts a narrow component arising from increased transmission due to purely kinematic effects in the wind geometry (D. Proga et al. 2000; D. Proga 2003; D. Proga & T. R. Kallman 2004). This would be an alternative explanation for at least part of the narrow component flux. On the other hand, most LRDs present narrow [O iii] emission that is often associated with the host galaxy. Indeed, the ionized gas producing [O iii] emission should have associated emission in the Hα and higher-order Balmer lines. However, constraining this component largely depends on the assumptions on dust attenuation or ISM conditions, and requires very high S/N and resolution data. Deep, space-based follow-up observations of PAN-BH*-1 would be very constraining for the wind kinematics (e.g., by the joint analysis of Hβ) and to assess whether a narrow component comes from a host galaxy (e.g., by comparing to a narrow Hβ component or [O iii] λλ4960, 5008). 5.2. Black Hole Mass From Photosphere Models The general physical setup of LRDs is an open debate, and their masses are a major unknown. Due to the multiple differences with respect to the classical AGN population, the validity of standard virial calibrations has been questioned (e.g., V. Rusakov et al. 2026; J. E. Greene et al. 2026; A. Sneppen et al. 2026; A. Torralba et al. 2026, although see, e.g., M. Brazzini et al. 2025, 2026; J. Scholtz et al. 2026 for an alternative interpretation). One can obtain a mass estimate assuming a system in radiative equilibrium with Lbol/LEdd = 1 (e.g., H. Umeda et al. 2026); this yields a total mass of ≈106 M⊙, using the bolometric luminosity from integrating the best-fit blackbody in Section 3.1. Recently, H. Liu et al. (2026) developed a synthetic spectral library of LRD atmosphere models. In these models, the density of the photosphere is regulated by the net surface gravity of an optically thick atmosphere, enabling constraints on the mass of the system. We fit the JWST photometry of PAN-BH*-1 using the models from H. Liu et al. (2026), assuming a negligible contribution from a host galaxy to the optical continuum. The best-fit model has effective temperature Teff = 4800 K, surface gravity , and metallicity (; see Figure 1). The best-fit implies a total mass of the system (BH plus gas) of (Equation (6) in H. Liu et al. 2026, assuming hydrostatic equilibrium). For the second and third best fits, we obtain and −2, respectively (, respectively; with the same metallicity and effective temperature), which would imply lower limits to the system mass between and 4. The bolometric luminosity of PAN-BH*-1 (from the integral of the best-fit green curve in Figure 1) implies an Eddington luminosity ratio of L/LEdd ≲ 13, assuming the best-fit mass from the H. Liu et al. (2026) models. The elevated Eddington ratio is in line with the hypothesis of a radiation-driven wind discussed in Section 4, and allows for somewhat larger system masses. The low masses obtained with this model, combined with the stellar mass inferred from dynamical arguments for the host galaxy (Section 3.3) set lower limits to the BH-to-stellar mass ratio of MBH/M* ≳ 10−4–10−2, which are compatible with the relations observed in the Local Universe, within the large uncertainties (A. E. Reines & M. Volonteri 2015). 6. Conclusions In this Letter, we presented the discovery and spectroscopic confirmation of PAN-BH*-1, an LRD with an extreme Balmer break at z = 1.731. The strength of the Balmer break (F115W/F814W = 7 ± 1) is comparable to the most extreme LRDs known, The Cliff (A. de Graaff et al. 2025a) and MoM-BH* (R. P. Naidu et al. 2025). We summarize the observations and our main conclusions as follows. 1. We obtained deep VLT/X-Shooter spectroscopy of PAN-BH*-1. The Hα emission line is luminous and broad (LHα = 1043 erg s−1), and has an unusually strong absorption. Hβ is detected with an S/N ≈ 6, and we conservatively estimate a lower limit for the Balmer decrement of Hα/Hβ > 9.4 (at a 3σ confidence level), in line with other LRDs in the literature (e.g., A. de Graaff et al. 2025b; G. P. Nikopoulos et al. 2026). 2. The absorption trough spans from −520 to 267 km s−1 (at a transmission level of 99%). We interpret the presence of blue- and redshifted absorption as produced by a disk wind, analogous to those analyzed in the context of broad absorption line quasars or accreting stars. This hypothesis would imply that the source of the optical continuum is likely a thick photospheric disk. 3. We detect a narrow Hα component (FWHM = 184 ± 12 km s−1), which we interpret as probing a host galaxy with M* ≈ 108 M⊙ and SFR = 2–3 M⊙. This interpretation is in line with the extended rest-NUV morphology measured in the HST bands ( kpc). 4. By fitting the synthetic atmosphere models of H. Liu et al. (2026), we estimate a system mass (BH+envelope) of 104–106 M⊙. The inferred masses, together with the stellar mass inferred from morphology and narrow emission line dynamics, imply BH-to-stellar mass ratios of 10−2–10−4, close to the extrapolated trend in the local Universe (A. E. Reines & M. Volonteri 2015). 5. The confirmation of this source at cosmic noon (magnitude of ≈22 in the K band, Hα flux ≈5 × 10−16 erg s−1 cm−2) proves the feasibility of detecting extreme LRDs at such epochs with wide-area spectroscopic surveys like Euclid or the forthcoming Nancy Grace Roman Space Telescope. Acknowledgments A.T. thanks Debasish Dutta and Tamara Bogdanović for useful conversations about stellar and AGN winds. We thank the scientific referee for the useful and constructive feedback, which helped improve the quality of this paper. J.M. and A.T. acknowledge funding by the European Union (ERC, AGENTS, 101076224). The work of CCW is supported by NOIRLab, which is managed by the Association of Universities for Research in Astronomy (AURA) under a cooperative agreement with the National Science Foundation. A.P.C. warmly acknowledges the support of the National Science Foundation through the NSF Graduate Research Fellowship Program. A.d.G. acknowledges support from a Clay Fellowship awarded by the Smithsonian Astrophysical Observatory. Based on observations made with ESO Telescopes at the Paranal Observatory under program IDs 116.294D and 116.2AQ0. This work is based in part on observations made with the NASA/ESA/CSA James Webb Space Telescope. The data were obtained from the Mikulski Archive for Space Telescopes at the Space Telescope Science Institute, which is operated by the Association of Universities for Research in Astronomy, Inc., under NASA contract NAS 5-03127 for JWST. These observations are associated with programs #2514 and #9433. C.C.W. gratefully acknowledges support for program JWST-GO-2514 provided by NASA through a grant from the Space Telescope Science Institute, which is operated by the Association of Universities for Research in Astronomy, Inc., under NASA contract NAS 5-03127. The authors acknowledge the team led by co-PIs R. Maiolino and F. D’Eugenio for developing their observing program with a zero-exclusive-access period. This research is based on observations made with the NASA/ESA Hubble Space Telescope obtained from the Space Telescope Science Institute, which is operated by the Association of Universities for Research in Astronomy, Inc., under NASA contract NAS 5–26555. These observations are associated with program #15117. The JWST and HST data presented in this article were obtained from the Mikulski Archive for Space Telescopes (MAST) at the Space Telescope Science Institute. The specific observations analyzed can be accessed via doi:10.17909/ydwx-st06. This work is based in part on observations made with the Spitzer Space Telescope, which was operated by the Jet Propulsion Laboratory, California Institute of Technology under a contract with NASA. The Spitzer data used in this work can be found in doi:10.26131/IRSA3. This work was supported by the International Space Science Institute (ISSI) in Bern, through ISSI International Team project #25-659 “Little Red Dots, Big Open Questions.” JWST cartoon in Figure 6, credit: NASA. Facilities: VLT:Kueyen - Very Large Telescope (Kueyen) (X-Shooter), VLT:Yepun (HAWK-I), JWST - James Webb Space Telescope (NIRCam, NIRspec), HST - Hubble Space Telescope satellite (ACS), Spitzer - Spitzer Space Telescope satellite (IRAC, MIPS) - . Software: astropy (Astropy Collaboration et al. 2013, 2018; Astropy Collaboration et al. 2022), NumPy (C. R. Harris et al. 2020), SciPy (P. Virtanen et al. 2020), pysersic (I. Pasha & T. B. Miller 2023), stpsf (M. D. Perrin et al. 2014), lmfit (M. Newville et al. 2014), EsoRex (ESO CPL Development Team 2015), Claude (used for Python coding; https://claude.ai/), SEP (K. Barbary 2016).

Volume

1005

Issue

Article Number

L37

ISSN

2041-8205

eISSN

2041-8213

IST-REx-ID

22263

Cite this

Torralba Torregrosa A, Matthee JJ, Weibel A, et al. A black hole star at cosmic noon: Extreme Balmer break, photospheric continuum, and broad absorption by thick winds in a Little Red Dot at z = 1.7. The Astrophysical Journal Letters. 2026;1005(2). doi:10.3847/2041-8213/ae7bfd

Torralba Torregrosa, A., Matthee, J. J., Weibel, A., Naidu, R. P., Ma, Y., Cloonan, A. P., … Williams, C. C. (2026). A black hole star at cosmic noon: Extreme Balmer break, photospheric continuum, and broad absorption by thick winds in a Little Red Dot at z = 1.7. The Astrophysical Journal Letters. IOP Publishing. https://doi.org/10.3847/2041-8213/ae7bfd

Torralba Torregrosa, Alberto, Jorryt J Matthee, Andrea Weibel, Rohan P. Naidu, Yilun Ma, Aidan P. Cloonan, Aayush A Desai, et al. “A Black Hole Star at Cosmic Noon: Extreme Balmer Break, Photospheric Continuum, and Broad Absorption by Thick Winds in a Little Red Dot at z = 1.7.” The Astrophysical Journal Letters. IOP Publishing, 2026. https://doi.org/10.3847/2041-8213/ae7bfd.

A. Torralba Torregrosa et al., “A black hole star at cosmic noon: Extreme Balmer break, photospheric continuum, and broad absorption by thick winds in a Little Red Dot at z = 1.7,” The Astrophysical Journal Letters, vol. 1005, no. 2. IOP Publishing, 2026.

Torralba Torregrosa, Alberto, et al. “A Black Hole Star at Cosmic Noon: Extreme Balmer Break, Photospheric Continuum, and Broad Absorption by Thick Winds in a Little Red Dot at z = 1.7.” The Astrophysical Journal Letters, vol. 1005, no. 2, L37, IOP Publishing, 2026, doi:10.3847/2041-8213/ae7bfd.

All files available under the following license(s):

Creative Commons Attribution 4.0 International Public License (CC-BY 4.0):