---
DOAJ_listed: '1'
OA_place: publisher
OA_type: gold
_id: '22296'
abstract:
- lang: eng
  text: 'The detection of strong Balmer breaks and absorption features in Little Red
    Dots (LRDs) suggests they host active galactic nuclei embedded within dense gas
    envelopes, potentially powered by super-Eddington accretion. We present GLIMPSE-17775,
    a luminous (Lbol ∼ 1045 erg s−1) LRD at z = 3.501 behind Abell S1063 (μ ∼ 2),
    observed with deep JWST/NIRCam and a ∼20 hr (80 hr delensed) NIRSpec G395M spectrum.
    The data reveal over 40 emission and absorption features, including a rich forest
    of low-ionization Fe ii lines and numerous broad hydrogen recombination transitions.
    We use this depth to test the dense-gas interpretation through five independent
    diagnostics. Nearly all permitted lines show exponential wings with consistent
    FWHM, a signature of Thomson scattering requiring ne ≳ 108 cm−3. Adopting this
    width yields MBH ∼ 106.7M⊙, a factor of 10 lower than Gaussian fits, and λEdd
    ∼ 1.8. Additional diagnostics support the same picture: a pronounced Balmer break
    (fν,4050/fν,3670 = 2.0 ± 0.1), enhanced He i λ7065 and λ10830 with P-Cygni absorption,
    Bowen-fluorescent O i λ8446–λ11290 emission requiring Lyβ pumping, and 16 Fe ii
    lines matching fluorescence models. These features indicate a dense (n ∼ 108 cm−3),
    partially ionized cocoon where scattering and fluorescence dominate line formation,
    providing strong evidence that at least some LRDs are powered by super-Eddington
    black hole growth in the early Universe.'
acknowledgement: "IOP Science home\r\nThe Astrophysical Journal\r\nThe American Astronomical
  Society, find out more.\r\n\r\nThe following article isOpen access\r\nThe Deepest
  GLIMPSE of a Dense Gas Cocoon Enshrouding a Little Red Dot\r\nVasily Kokorev, John
  Chisholm, Rohan P. Naidu, Seiji Fujimoto, Hakim Atek, Gabriel Brammer, Steven L.
  Finkelstein, Hollis B. Akins, Danielle A. Berg, Lukas J. FurtakShow full author
  list\r\n\r\nPublished 2026 June 10 • © 2026. The Author(s). Published by the American
  Astronomical Society.\r\nThe Astrophysical Journal, Volume 1004, Number 2\r\nCitation
  Vasily Kokorev et al 2026 ApJ 1004 153\r\nDOI 10.3847/1538-4357/ae4ed7\r\n\r\nPDFOpens
  in a new tab.ePub\r\nAuthors\r\nFigures\r\nTables\r\nReferences\r\nArticle data\r\nPDFOpens
  in a new tab.ePub\r\nArticle metrics\r\n5138 Total downloads\r\n\r\n22 total citations
  on Dimensions.\r\nShare this article\r\nArticle information\r\nAbstract\r\nThe detection
  of strong Balmer breaks and absorption features in Little Red Dots (LRDs) suggests
  they host active galactic nuclei embedded within dense gas envelopes, potentially
  powered by super-Eddington accretion. We present GLIMPSE-17775, a luminous (Lbol
  ∼ 1045 erg s−1) LRD at z = 3.501 behind Abell S1063 (μ ∼ 2), observed with deep
  JWST/NIRCam and a ∼20 hr (80 hr delensed) NIRSpec G395M spectrum. The data reveal
  over 40 emission and absorption features, including a rich forest of low-ionization
  Fe ii lines and numerous broad hydrogen recombination transitions. We use this depth
  to test the dense-gas interpretation through five independent diagnostics. Nearly
  all permitted lines show exponential wings with consistent FWHM, a signature of
  Thomson scattering requiring ne ≳ 108 cm−3. Adopting this width yields MBH ∼ 106.7M⊙,
  a factor of 10 lower than Gaussian fits, and λEdd ∼ 1.8. Additional diagnostics
  support the same picture: a pronounced Balmer break (fν,4050/fν,3670 = 2.0 ± 0.1),
  enhanced He i λ7065 and λ10830 with P-Cygni absorption, Bowen-fluorescent O i λ8446–λ11290
  emission requiring Lyβ pumping, and 16 Fe ii lines matching fluorescence models.
  These features indicate a dense (n ∼ 108 cm−3), partially ionized cocoon where scattering
  and fluorescence dominate line formation, providing strong evidence that at least
  some LRDs are powered by super-Eddington black hole growth in the early Universe.\r\n\r\nExport
  citation and abstract\r\nBibTeXRIS\r\n\r\nPrevious article in issue\r\nNext article
  in issue\r\nRelated links\r\n\r\nOriginal 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.\r\n\r\n1. Introduction\r\nOne of the most enticing puzzles
  brought about by the launch of the James Webb Space Telescope (JWST) has been the
  discovery of red, compact objects called “Little Red Dots” (LRDs; J. Matthee et
  al. 2024). Previously invisible to the Hubble Space Telescope (HST) due to their
  extreme faintness in the optical and lack of near-infrared (NIR) coverage, LRDs
  have emerged in abundance (V. Kokorev et al. 2024a; H. B. Akins et al. 2025a; D.
  D. Kocevski et al. 2025; G. Barro et al. 2026) thanks to the unprecedented NIR sensitivity
  of JWST.\r\n\r\nTheir unusual properties—such as compact morphologies in rest-optical
  and distinctive “V-shaped” spectral energy distributions (SEDs)—make LRDs easy to
  identify in JWST fields; however, this is where the simplicity ends. Explaining
  LRDs as either evolved or dusty compact galaxies proves difficult: The former scenario
  requires massive stellar populations that strain Lambda cold dark matter (ΛCDM)
  predictions (M. Boylan-Kolchin 2023; I. Labbé et al. 2023), while the latter implies
  significant dust emission, yet none is observed (C. M. Casey et al. 2024; G. C.
  K. Leung et al. 2025; H. B. Akins et al. 2025a; I. Labbé et al. 2025).\r\n\r\nOver
  time, the accumulating detections of broad Balmer-series emission lines (D. D. Kocevski
  et al. 2023, 2025; J. Matthee et al. 2024, just to name a few), often accompanied
  by signatures of extreme ionization (e.g., V. Kokorev et al. 2023), have begun to
  clarify the physical origin of LRDs, pointing increasingly toward active galactic
  nuclei (AGN) as the underlying power source. In parallel, the much needed advent
  of NIRSpec Micro Shutter Assembly (MSA) programs based on red targets selected from
  JWST imaging (e.g., A. de Graaff et al. 2025b) has provided critical confirmation:
  Nearly all point sources exhibiting “V-shaped” SEDs reveal broad emission lines
  upon spectroscopic follow-up (R. E. Hviding et al. 2025), solidifying their AGN
  interpretation.\r\n\r\nThe nature of the spectral inflection point in LRDs, typically
  located near ∼3600 Å, has also undergone significant revision. Initially interpreted
  as a stellar Balmer break (I. Labbé et al. 2023), this feature implied implausibly
  high stellar masses (M*) far too early in cosmic history (M. Boylan-Kolchin 2023;
  N. Sabti et al. 2024). A second hypothesis invoked differential dust attenuation
  and host-galaxy contamination in the rest-UV to explain the sharp discontinuity
  (M. Volonteri et al. 2025). However, this explanation was soon ruled out by D. J.
  Setton et al. (2025) and Y. Ma et al. (2025), who demonstrated that the break generally
  lies around the Balmer limit, albeit with a fundamentally different origin than
  initially proposed.\r\n\r\nMore recently, a series of theoretical and observational
  works have converged on a new picture in which LRDs may host accreting black holes
  enshrouded in exceptionally dense, partially ionized gas cocoons, often referred
  to as “black hole stars” (hereafter BH*; A. de Graaff et al. 2025b; K. Inayoshi
  & R. Maiolino 2025; R. P. Naidu et al. 2025; A. J. Taylor et al. 2025). Here, BH*
  is used in an empirical sense to denote cocooned black holes whose emergent spectra
  exhibit a combination of black-hole-like and star-like features, such as broad permitted
  lines, steep Balmer-limit breaks, and Balmer absorption, as observed in the archetypal
  BH* sources (A. de Graaff et al. 2025b; R. P. Naidu et al. 2025). This usage does
  not assume a specific geometry or formation channel: The underlying physical picture
  involves an AGN power-law ionizing spectrum processed through very dense gas, and
  similar phenomenology can arise in a range of dense-gas configurations (e.g., K.
  Inayoshi & R. Maiolino 2025; D. Kido et al. 2025; H. Liu et al. 2025; A. Sneppen
  et al. 2026). In such environments, suppressed X-ray and radio emission (e.g., K.
  Inayoshi & R. Maiolino 2025; R. P. Naidu et al. 2025) arise naturally from the high
  optical depth, while the UV emission and forbidden optical lines (e.g., [O iii]
  λλ4959, 5007) may originate in the more extended host (e.g., A. de Graaff et al.
  2025d).\r\n\r\nThe dense-gas interpretation also offers a natural explanation for
  the P-Cygni-like Balmer and helium profiles frequently seen in high-signal-to-noise-ratio
  (S/N) spectra (J. Matthee et al. 2024), the non-Gaussian, exponentially winged line
  shapes predicted by radiative-transfer models of scattering in ionized gas (S.-J.
  Chang et al. 2026; V. Rusakov et al. 2026), and Lyα-like resonantly scattered shapes
  (R. P. Naidu et al. 2025; S.-J. Chang et al. 2026). Intriguingly, if electron and
  resonant scattering indeed dominate the broad-line widths, then the true virial
  velocities (traced by intrinsically narrower Gaussian cores)—and hence black hole
  masses—could be lower by an order of magnitude, alleviating the apparent tension
  between LRD black hole masses and their compact host galaxies (V. Rusakov et al.
  2026).\r\n\r\nSo far, however, the BH*/dense-gas interpretation of LRDs has relied
  largely on the Balmer break, occasional absorption features (X. Lin et al. 2026),
  and, to a limited extent, the shapes of broad emission lines (V. Rusakov et al.
  2026). While these features are consistent with the presence of dense, partially
  ionized gas, they stop short of providing direct spectroscopic confirmation of the
  physical conditions expected by the BH* scenario. What has been missing is an unambiguous
  demonstration—through emission-line physics—that LRDs indeed host dense, optically
  thick cocoons surrounding rapidly accreting black holes. Such evidence would directly
  tie the observed line formation, excitation, and radiative transfer to dense, stratified
  envelopes of gas surrounding the central engine.\r\n\r\nIn this work, we present
  precisely such a case: an exceptionally deep, ∼20 hr (equivalent to 80 hr without
  lensing magnification) JWST/NIRSpec G395M spectrum of a luminous LRD at z = 3.50102.
  The target, GLIMPSE-17775, lies in a highly magnified region of the massive galaxy
  cluster Abell S1063 (hereafter AS1063) and benefits from the combined power of JWST/NIRCam
  imaging from the GLIMPSE GO program (PID 3293; PIs: H. Atek & J. Chisholm) and recent
  NIRSpec spectroscopy from the GLIMPSEDirector's Discretionary Time (DDT) (hereafter
  GLIMPSE-D) program (PID 9223; PIs: S. Fujimoto & R. Naidu). This synergy of ultradeep
  spectroscopy and strong gravitational lensing enables an unprecedented view of GLIMPSE-17775,
  revealing rest-frame optical and NIR features at a level unseen before in any LRD.
  We detect over 40 emission and absorption features, including 16 Fe ii transitions
  that form a dense “iron forest.” The remarkable richness of this spectrum makes
  GLIMPSE-17775 a uniquely powerful laboratory for dissecting the dense, radiation-dominated
  environments that accompany early black hole growth.\r\n\r\nThis paper is organized
  as follows. In Section 2, we present the GLIMPSE-D NIRSpec dataset alongside the
  photometric datasets used in this study. In Section 3, we calculate the spectroscopic
  redshift, describe the identification of prominent emission/absorption features,
  and describe bespoke fitting of various line complexes. Section 4 describes the
  measurement of the source morphology, dust attenuation, and black hole masses. In
  Section 5, we comment on the various properties of the emission features in the
  our target. Finally, we discuss our findings in Section 6.\r\n\r\nThroughout this
  work, we assume a flat ΛCDM cosmology with Ωm,0 = 0.3, ΩΛ,0 = 0.7, and H0 = 70 km
  s−1 Mpc−1, and a G. Chabrier (2003) initial mass function between 0.1 and 100M⊙.
  All magnitudes are expressed in the AB system (J. B. Oke 1974).\r\n\r\n2. Observations
  and Data\r\nThe target, GLIMPSE-17775, was originally identified as a bright LRD
  candidate in JWST/NIRCam imaging captured by the GLIMPSE (PID: 3293; PIs: H. Atek
  & J. Chisholm) survey (H. Atek et al. 2025) of the lensed AS1063 Hubble Frontier
  Field (HFF; J. M. Lotz et al. 2017). The highly magnified area of AS1063 has already
  successfully yielded z  >  16 galaxy candidates (V. Kokorev et al. 2025), potential
  Population III (Pop III) hosts (S. Fujimoto et al. 2025a), numerous faint and high-redshift
  galaxies (I. Chemerynska et al. 2026), new constraints on reionization (D. Korber
  et al. 2025), identification of intermediate-mass black holes (Q. Fei et al. 2025),
  and a wide variety of enigmatic LRDs, some moderately lensed. The LRD selection
  was based on the standard compactness plus “V-shape” criteria already laid out in
  J. E. Greene et al. (2024), V. Kokorev et al. (2024a), H. B. Akins et al. (2025a),
  and I. Labbé et al. (2025) using photometric redshifts derived with eazy (G. B.
  Brammer et al. 2008), a technique that has proven successful at consistently identifying
  many exciting sources (see, e.g., H. B. Akins et al. 2025b; A. de Graaff et al.
  2025b; V. Kokorev et al. 2023; R. P. Naidu et al. 2025; A. J. Taylor et al. 2025).
  Located at zphot = 4.2 ± 0.1, GLIMPSE-17775 stood out in particular due to its extreme
  rest-optical brightness with mf444w ∼ 23.6 mag and a modest lensing magnification
  μ ∼ 2. While very little could be done with photometry alone, this marked GLIMPSE-17775
  as a promising target for any future spectroscopic follow-up.\r\n\r\n2.1. Photometry\r\nWe
  use photometry both to select targets for the GLIMPSE-D NIRSpec observations and,
  during spectroscopic modeling, to constrain the overall shape of the SED and correct
  for NIRSpec slit losses. In addition to JWST/NIRCam data, we incorporate deep HST
  Advanced Camera for Surveys and Wide Field Camera 3 imaging from the HFF (J. M.
  Lotz et al. 2017) and BUFFALO (C. L. Steinhardt et al. 2020) programs. Our reprocessed
  HST mosaics are based on Gaia-aligned images from the CHArGE archive (V. Kokorev
  et al. 2022), hosted on the Dawn JWST Archive (F. Valentino et al. 2023). More descriptions
  of the image reduction and source extraction procedure can be found in R. Endsley
  et al. (2025), as well as in the GLIMPSE overview paper (H. Atek et al. 2025); we
  briefly summarize the latter procedure below.\r\n\r\nPhotometry is performed on
  point-spread function (PSF)-homogenized HST and JWST images, convolved to the resolution
  of the F480M filter. Source detection is carried out using SExtractor (E. Bertin
  & S. Arnouts 1996) through two parallel steps. We construct two inverse-variance-weighted
  detection images: one from the short-wavelength (SW; F090W, F115W, F150W, F200W)
  bands to preserve spatial resolution, and one from the long-wavelength (LW; F277W,
  F356W, F444W) bands to ensure sensitivity to red or dusty sources (e.g., LRDs).
  Unlike the science images, these have not been PSF-matched. These SW and LW detection
  catalogs are subsequently merged into a single, combined catalog. Photometry is
  then measured using photutils (L. Bradley et al. 2020) in a range of circular apertures
  (–). Photometric uncertainties are estimated by placing random apertures in empty
  regions around each source. Unless stated otherwise, we adopt total fluxes measured
  within a aperture throughout this work.\r\n\r\n2.2. NIRSpec Observations\r\nGLIMPSE-D
  NIRSpec data were obtained during a campaign (DDT #9223; PIs: S. Fujimoto & R. Naidu)
  to study a promising Pop III galaxy candidate (S. Fujimoto et al. 2025a). GLIMPSE-D
  obtained a total of three G395M/F295LP MSA pointings, totaling 29.78 hr of total
  exposure time. The pointing center differed for each of the three configurations
  to maximize the total object yield. In total, the GLIMPSE-D sample contains 384
  spectra, with depths varying from 9.2 hr to ∼30 hr. All of the planned observations
  were successfully executed between 2025 June 30 and July 2. For each MSA configuration,
  GLIMPSE-D employed a standard three-point nod pattern at an aperture position angle
  PA_V3 = 2705, using 494 groups per integration with the NRSIRS2 readout mode. The
  full details of the target selection, prioritization, and MSA planning will be presented
  in a forthcoming survey paper (S. Fujimoto et al. 2025b).\r\n\r\n2.3. G395M Data
  Reduction and Calibration\r\nThe GLIMPSE-D MSA spectra were uniformly reduced using
  msaexp (v0.9.8; G. Brammer 2022). This procedure starts with the level-2 calibrated
  products obtained from MAST and applies a number of corrections that include 1/f
  noise, artifact detection and removal, and a bias correction in individual exposures
  (see J. Rigby et al. 2023 for more details). Together with the JWST pipeline, msaexp
  sets the slit World Coordinate System, performs flat-fielding, and computes an initial
  pass-loss correction. Each one of the two-dimensional shutters is then drizzled
  onto a common pixel grid. Using the standard approach, the background subtraction
  is performed locally using stacked, source-free shutters. The one-dimensional spectra
  are then obtained via the optimal extraction method (e.g., P. Arrabal Haro et al.
  2023; A. de Graaff et al. 2024), in which the center and width of the extraction
  “aperture” vary depending on the best-fit Gaussian model (K. Horne 1986), similar
  to the methodology widely adopted by a variety of other works (e.g., B. Wang et
  al. 2023; J. E. Greene et al. 2024; V. Kokorev et al. 2024b).\r\n\r\nThe absolute
  flux calibration of MSA spectra can be influenced by several factors, including
  the position of the source within the shutter, calibration and astrometric uncertainties,
  and the intrinsic morphology of the source. To correct for these effects and derive
  an overall slit-loss correction, we rescale the extracted one-dimensional spectra
  by convolving them with all available NIRCam filters and comparing the resulting
  flux densities to the total photometry from the GLIMPSE catalog. A wavelength-dependent
  correction is then obtained by fitting a second-order polynomial to these differences.\r\n\r\nThe
  target in this paper was observed in all three configurations; however, one was
  severely contaminated by sources in the same row, making the data recovery for that
  configuration unfeasible. Despite this, GLIMPSE-17775 is securely detected in two
  configurations (33,262 s and 40,703 s), yielding a total integration time of 20.55
  hr. We combine the extracted and separately calibrated one-dimensional spectra using
  inverse-variance weighting. The stacked spectrum is shown in Figure 1. We also make
  the spectrum publicly available.17\r\n\r\nZoom InZoom OutReset image size\r\nFigure
  1. Top: JWST/NIRCam and HST 20 stamps and red, green, blue short-wavelength (SW)
  and long-wavelength (LW) color images, comprising the F115W, F150W, and F200W and
  F277W, F356W, and F444W bands, respectively. MSA shutters for both configurations
  covering GLIMPSE-17775 are shown in blue and red, respectively. The source morphology
  is resolved and extended up to ∼2 μm, and then appears to transition to a more PSF-dominated
  and compact shape, echoing a growing sample of LRDs with extended rest-UV morphology
  (I. Juodžbalis et al. 2024; I. Labbe et al. 2024; J. Matthee et al. 2024; P. Rinaldi
  et al. 2025a). This likely hints at the presence of the host galaxy in the filters
  covering the rest-UV. In each panel, we show the total AB magnitude as presented
  in the GLIMPSE catalog (H. Atek et al. 2025; V. Kokorev et al. 2025). The source
  is exceptionally bright (M444 ∼ 23.6) and is detected in most JWST bands at >100σ.
  Middle: two-dimensional MSA G395M spectra covering GLIMPSE-17775. Bottom: combined
  one-dimensional spectrum (see, e.g., V. Kokorev et al. 2023; A. de Graaff et al.
  2025a for the extraction method) of the LRD in the observed frame. We show the data
  in black, and the uncertainty as a black shaded region. Fixing the systemic redshift
  to the [S iii] λ9071 line (zspec = 3.50102 ± 0.00019), we show the positions and
  label the prominent emission with significant (≥3σ) detections as solid vertical
  lines. Iron lines are shown separately in green. Due to the sheer number of features,
  not all could be labeled; we therefore display all features in Figure 2. Emission
  lines for which only upper limits are obtained are shown with dashed lines.\r\n\r\nDownload
  figure:\r\n\r\nStandard imageHigh-resolution image\r\n3. Emission-line Analysis\r\n3.1.
  Spectroscopic Redshift\r\nThe spectrum of GLIMPSE-17775 reveals a remarkably wide
  variety of significantly detected emission and absorption lines. Before focusing
  on individual features and their complex components, we perform an initial tally
  of all detectable lines using a heavily modified version of msaexp (G. Brammer 2022;
  V. Kokorev et al. 2024b). The key modifications allow varying the line widths and
  fitting multiple components, albeit tied to the same redshift.\r\n\r\nWe fit the
  full spectrum using msaexp, adopting Gaussian profiles for emission lines and a
  three-segment cubic spline to model the continuum. This minimal spline structure,
  appropriate given the high average S/N of GLIMPSE-17775 (≳30), helps prevent overfitting,
  particularly of features that may be slightly offset in wavelength or velocity space
  from the average redshift. Emission-line positions are fixed; narrow components
  are allowed a FWHM between 150 and 800 km s−1, and permitted transitions may include
  a broad component (800–5000 km s−1).\r\n\r\nThis yields a redshift of zspec = 3.5010
  ± 0.0001 (see Table 1). However, several strong lines exhibit significant pixel-level
  offsets of 100–200 km s−1 relative to this value. These discrepancies are statistically
  robust and appear in both stacked and individual spectra, suggesting that the derived
  redshift reflects a weighted average dominated by high-S/N lines. To properly assess
  velocity structure, a consistent systemic redshift is required. In the following
  sections describing the detailed line analysis, we adopt the centroid of a narrow
  forbidden line as the systemic reference frame (specifically [S iii] λ9071; see
  Section 3.3).\r\n\r\nUsing the updated redshift alongside the strong-lensing model
  of AS1063 (A. Zitrin et al. 2015; L. Furtak et al. 2026, in preparation), we recalculate
  the lensing magnification to be μ = 2.04 ± 0.21, consistent with previous estimates
  based on zphot. With redshift and magnification now fixed, we turn to a deeper analysis
  of the emission-line properties.\r\n\r\n3.2. Line Identification\r\nBeyond velocity
  offsets, the initial msaexp fit highlights several notable features. Broad hydrogen
  lines, ubiquitous in LRDs (e.g., D. D. Kocevski et al. 2023; V. Kokorev et al. 2023;
  L. J. Furtak et al. 2024; J. Matthee et al. 2024; A. de Graaff et al. 2025b; R.
  E. Hviding et al. 2025; R. P. Naidu et al. 2025; A. J. Taylor et al. 2025), are
  strongly favored across the Balmer and Paschen series, with Δχ2 > 10. These include
  Hα and Pa10 through Paγ. Although Paβ falls near the edge of the detector, we nonetheless
  observe a prominent broad wing at its expected location (see Figure 1).\r\n\r\nBroad
  components are also detected in several helium and oxygen lines: He i λ6680, He
  i λ7065, O i λ8448, He i λ10830, and O i λ11290. Notably, the He i λ10830 line shows
  a classic P-Cygni profile with a clear blueshifted absorption trough. While Balmer
  absorption is becoming a familiar sight in LRDs (R. Maiolino et al. 2024; J. Matthee
  et al. 2024; A. de Graaff et al. 2025b; D. D. Kocevski et al. 2025; R. P. Naidu
  et al. 2025), helium absorption remains rare, with only a few published cases at
  high z (e.g., I. Juodžbalis et al. 2024; R. P. Naidu et al. 2024; B. Wang et al.
  2025). Curiously, this was also recently reported in local analogs of LRDs (X. Lin
  et al. 2026).\r\n\r\nThe most striking feature is the detection of an extensive
  Fe ii emission forest in the rest-NIR—at least 16 distinct features are identified
  by msaexp and highlighted in green in Figure 1. Previous detections of permitted
  Fe ii in LRDs have been mostly limited to rest-UV/optical wavelengths using low-resolution
  PRISM spectra (R. Tripodi et al. 2025), with only few examples in the NIR (e.g.,
  see broad Fe iiλ9200 I. Labbe et al. 2024). Iron has also been identified in local
  LRD analogs (X. Lin et al. 2026; X. Ji et al. 2026). Only recently have a few Fe
  ii emitters been reported in medium-resolution (F. D’Eugenio et al. 2025) and high-resolution
  spectra (A. Torralba et al. 2026). The richness of these lines in GLIMPSE-17775
  hints at a dense, partially shielded gas phase, where processes such as Lyβ fluorescence
  and continuum pumping may be enhancing Fe ii emission (T. A. A. Sigut & A. K. Pradhan
  1998, 2003; T. A. A. Sigut et al. 2004), offering a new diagnostic window into early
  AGN environments.\r\n\r\nIn summary, GLIMPSE-17775 exhibits an exceptionally rich
  and kinematically complex array of emission and absorption features. While the initial
  msaexp modeling provides a solid foundation for redshift estimation and line identification,
  it lacks the flexibility to capture the full diversity of line profiles in this
  high-S/N dataset. In the following sections, we peel back the spectral layers with
  increasing precision.\r\n\r\n3.3. Line Fitting\r\nThe exceptional depth of the G395M
  spectrum for GLIMPSE-17775 enables detailed modeling of both multiple kinematic
  components and potential velocity offsets between line species. Unless noted otherwise,
  each line is fit with a single Gaussian narrow component (FWHM = 100−800 km s−1).
  For permitted lines flagged by msaexp as potentially broad, we follow the approach
  of V. Rusakov et al. (2026) and model the profile as an intrinsic Gaussian core
  broadened by electron scattering, implemented via convolution with an exponential-wing
  component. The intrinsic Gaussian FWHM is allowed to vary between 500 and 5000 km
  s−1. The characteristic scattering width (i.e., the strength of the exponential
  wings, W) is treated as a free parameter and is allowed to vary independently between
  different transitions. For each broad line, we also perform an alternative fit with
  the W parameter fixed to zero (pure Gaussian) and compare the resulting goodness
  of fit to assess whether the exponential wings are statistically required. Line
  centers are allowed to vary independently unless otherwise specified.\r\n\r\nAbsorption
  lines are modeled using a Gaussian optical-depth profile, where the observed flux
  is Fobs = Fem e−τ(λ), where . The free parameters are then the central optical depth
  τ0, velocity width Δvabs, and redshift zabs (see I. Juodžbalis et al. 2024). Widths
  and redshifts (for both narrow and broad components) are typically free, with redshift
  limited to ± 0.1 from the msaexp value. Local continua are fit using first-order
  polynomials.\r\n\r\nAll individual model components are first initialized and coadded
  on an oversampled wavelength grid. To take into account the wavelength-dependent
  resolution of the grating, we interpolate our model onto a variable step grid while
  making sure that the total integrated flux is preserved. Further, we increase the
  nominal spectral resolution by a factor of 1.7, as it has been shown that the spectral
  resolution for a pointlike source falling within a shutter is higher than that of
  a uniformly illuminated slitlet (A. de Graaff et al. 2024). Fitting uses nonlinear
  χ2 minimization, with uncertainties derived from multivariate resampling of the
  covariance matrix.\r\n\r\nTable 1. Source Properties\r\n\r\nParametera\tGLIMPSE-17775\r\nID\t17775\r\nR.A.
  [deg]\t342.20080\r\nDecl. [deg]\t–44.54366\r\nzphot (eazy)\t4.2 ± 0.1\r\nzspec ([S
  iii] λ9071)\t3.50102 ± 0.00019\r\nμ\t2.04 ± 0.21\r\nMUV [AB mag]\t−17.27 ± 0.05\r\nreff,UV
  [pc]\t1000 ± 200\r\nreff,opt [pc]\t<300\r\nβ\t–0.69 ± 0.12\r\nlog10(MBH/M⊙)\t6.65
  ± 0.15\r\nLbol [erg s−1]\t(1.06 ± 0.14)×1045\r\nλedd\t1.86 ± 0.25\r\nlog10(M*/M⊙)\t<7.5\r\nAV\t0.1
  ± 0.3\r\nfν,4050Å/fν,3670Å\t2.02 ± 0.10\r\nNote. aPhysical parameters are corrected
  for the lensing magnification.\r\n\r\nDownload table as: \r\nASCIITypeset image\r\n\r\nGiven
  the complexity, the spectrum is divided into six windows grouped by line species
  or proximity (see Figure 2). Below, we describe the assumptions adopted for each
  window. Final fluxes and equivalent widths of the narrow and broad lines are listed
  in Tables 2 and 3, respectively, while kinematics are reported in Tables 4 and 5.
  We show detailed line fits in Figure 2.\r\n\r\nZoom InZoom OutReset image size\r\nFigure
  2. A staggering abundance of spectral lines in GLIMPSE-17775 at z = 3.501. For each
  spectral window defined in Section 3.3, we show the data (black), the best-fit narrow
  and broad components (dark purple and blue), and the total best-fit model including
  the continuum (red). All broad components were fit with models allowing exponential
  wings. The ΔBIC between exponential and Gaussian fits is reported in the top right
  of each panel; negative values indicate a preference for an exponential profile.
  The lower panels display the uncertainty-weighted residuals for each fit.\r\n\r\nDownload
  figure:\r\n\r\nStandard imageHigh-resolution image\r\nTable 2. Fluxes of Narrow
  Emission Lines and Their Rest-frame Equivalent Widthsa\r\n\r\nLine\tλrest\tFlux\tEW0\r\n
  \t(Å)\t(10−20 erg s−1 cm−2)\t(Å)\r\nNarrow Emission Lines\r\nO i\t6302.0\t54.0 ±
  21.2\t3.3 ± 1.2\r\n[N ii]\t6549.0\t25.6 ± 10.3\t1.6 ± 0.7\r\nHα\t6562.8\t2622.9
  ± 200.7\t167.7 ± 12.8\r\n[N ii]\t6584.0\t77.1 ± 37.1\t4.9 ± 1.6\r\nHe i\t6680.0\t14.7
  ± 11.1\t0.8 ± 0.7\r\n[S ii]\t6717.0\t22.2 ± 0.1\t1.5 ± 0.7\r\n[S ii]\t6731.0\t21.4
  ± 0.1\t1.4 ± 0.5\r\nHe i\t7065.0\t461.7 ± 210.2\t31.0 ± 12.2\r\n[Ar iii]\t7138.0\t12.6
  ± 3.4\t1.0 ± 0.3\r\nFe ii\t7156.0\t26.2 ± 7.2\t1.8 ± 0.2\r\n[O ii]\t7323.0\t26.0
  ± 23.0\t1.7 ± 1.5\r\n[O ii]\t7325.0\t16.9 ± 1.8\t1.2 ± 0.1\r\n[O ii]\t7332.0\t26.2
  ± 9.1\t1.8 ± 0.6\r\nFe ii\t8228.0\t23.7 ± 4.2\t2.0 ± 0.3\r\nFe ii\t8239.0\t7.2 ±
  4.0\t0.6 ± 0.3\r\nFe ii\t8289.0\t6.3 ± 3.7\t0.5 ± 0.3\r\nFe iii\t8306.0\t5.7 ± 3.6\t0.5
  ± 0.3\r\nO i\t8448.0\t67.6 ± 2.4\t6.1 ± 1.3\r\nFe ii\t8470.0\t16.2 ± 5.0\t1.4 ±
  0.4\r\nFe ii\t8490.0\t26.7 ± 4.3\t2.4 ± 0.4\r\nPa10\t9015.0\t16.5 ± 5.0\t1.8 ± 0.5\r\n[S
  iii]\t9071.0\t27.2 ± 2.3\t3.5 ± 0.3\r\nFe ii\t9075.0\t30.0 ± 4.7\t3.4 ± 0.5\r\nFe
  ii\t9125.0\t16.8 ± 5.2\t1.9 ± 0.6\r\nFe ii\t9134.0\t36.9 ± 5.0\t4.1 ± 0.6\r\nFe
  ii\t9179.0\t95.0 ± 5.5\t10.6 ± 0.6\r\nFe ii\t9204.0\t50.1 ± 6.1\t5.6 ± 0.7\r\nPa9\t9229.0\t19.4
  ± 7.0\t2.2 ± 0.8\r\nFe ii\t9394.0\t17.3 ± 3.9\t2.0 ± 0.5\r\n[S iii]\t9533.0\t84.7
  ± 5.9\t10.2 ± 0.7\r\nPa8\t9545.0\t44.8 ± 11.3\t5.4 ± 1.4\r\nFe ii\t9997.0\t29.5
  ± 4.5\t3.8 ± 0.6\r\nPaδ\t10049.0\t104.2 ± 12.2\t13.7 ± 1.7\r\nFe ii\t10490.0\t8.0
  ± 2.7\t1.1 ± 0.4\r\nFe ii\t10501.0\t24.4 ± 5.9\t3.1 ± 0.8\r\nHe i\t10830.0\t163.0
  ± 96.0\t22.6 ± 13.3\r\nPaγ\t10938.0\t146.3 ± 25.5\t20.9 ± 3.8\r\nFe ii\t11128.0\t12.3
  ± 3.7\t1.8 ± 0.6\r\nO i\t11290.0\t49.2 ± 4.1\t7.7 ± 2.3\r\nNote. aNot corrected
  for lensing magnification.\r\n\r\nDownload table as: \r\nASCIITypeset image\r\n\r\nTable
  3. Fluxes of Broad Emission Lines and Their Rest-frame Equivalent Widthsa\r\n\r\nLine\tλrest\tFlux\tEW0\r\n
  \t(Å)\t(10−20 erg s−1 cm−2)\t(Å)\r\nBroad Emission Lines\r\nHα\t6562.8\t14803.5
  ± 198.6\t946.3 ± 15.6\r\nHe i\t6680.0\t144.6 ± 44.7\t9.4 ± 2.9\r\nHe i\t7065.0\t824.2
  ± 210.0\t55.4 ± 15.1\r\nPa10\t9015.0\t53.7 ± 11.1\t5.8 ± 1.3\r\nPa9\t9229.0\t119.0
  ± 13.5\t13.6 ± 1.5\r\nPa8\t9545.0\t191.2 ± 17.0\t23.1 ± 2.1\r\nPaδ\t10049.0\t370.6
  ± 15.0\t48.8 ± 2.3\r\nHe i\t10830.0\t2294.9 ± 212.6\t318.8 ± 30.3\r\nPaγ\t10938.0\t652.3
  ± 14.0\t92.9 ± 2.4\r\nO i\t8448.0\t93.7 ± 7.9\t8.4 ± 0.7\r\nO i\t11290.0\t106.2
  ± 63.9\t16.6 ± 10.4\r\nNote. aNot corrected for lensing magnification.\r\n\r\nDownload
  table as: \r\nASCIITypeset image\r\n\r\nTable 4. Kinematic Properties of Narrow
  Lines\r\n\r\nLine\tλrest\tFWHM\tΔv\r\n \t(Å)\t(km s−1)\t(km s−1)\r\nNarrow Emission
  Lines\r\n[O i]\t6300.3\t455 ± 175\t−101 ± 81\r\nHα\t6562.8\t304 ± 8\t+145 ± 13\r\n[N
  ii]\t6583.4\t347 ± 117\t+84 ± 47\r\n[S ii]\t6716.4\t232 ± 62\t+82 ± 33\r\n[S ii]\t6731\t232
  ± 62\t+82 ± 33\r\nHe i\t6678.2\t87 ± 14\t+40 ± 14\r\nHe i\t7065.2\t408 ± 644\t+112
  ± 396\r\n[Ar iii]\t7135.8\t89 ± 50\t−26 ± 20\r\nFe ii\t7155.2\t96 ± 50\t+15 ± 15\r\n[O
  ii]\t7320.0\t400 ± 120\t−37 ± 72\r\nFe ii\t8228.0\t290 ± 48\t+21 ± 5\r\nFe ii\t8239.0\t290
  ± 48\t+21 ± 5\r\nFe ii\t8289.0\t290 ± 48\t+21 ± 5\r\nFe iii\t8306.0\t290 ± 48\t+21
  ± 5\r\nO i\t8446.4\t235 ± 30\t−43 ± 19\r\nFe ii\t8470.0\t290 ± 48\t+21 ± 5\r\nFe
  ii\t8490.0\t290 ± 48\t+21 ± 5\r\nPa10\t9014.9\t259 ± 36\t−46 ± 19\r\n[S iii]\t9071.0\t230
  ± 13\t⋯\r\nFe ii\t9075.0\t486 ± 24\t−21 ± 17\r\nFe ii\t9125.0\t486 ± 24\t−21 ± 17\r\nFe
  ii\t9134.0\t486 ± 24\t−21 ± 17\r\nFe ii\t9179.0\t486 ± 24\t−21 ± 17\r\nFe ii\t9204.0\t486
  ± 24\t−21 ± 17\r\nPa9\t9229.0\t259 ± 35\t−46 ± 19\r\nFe ii\t9394.0\t486 ± 24\t−21
  ± 17\r\nPa8\t9545.6\t259 ± 36\t−46 ± 19\r\n[S iii]\t9533\t230 ± 13\t+23 ± 13\r\nFe
  ii\t9997.0\t331 ± 55\t+90 ± 25\r\nPaδ\t10049.4\t332 ± 22\t+21 ± 15\r\nHe i\t10830.3\t287
  ± 46\t+112 ± 26\r\nPaγ\t10938.1\t267 ± 25\t+26 ± 17\r\nFe ii\t10490.0\t332 ± 96\t+44
  ± 43\r\nFe ii\t10501.0\t332 ± 96\t+44 ± 43\r\nFe ii\t11128.0\t153 ± 116\t+37 ± 30\r\nO
  i\t11290.0\t260 ± 58\t−33 ± 24\r\nNote. Velocity offsets (Δv) calculated relative
  to [S iii] λ9071 at z = 3.50102.\r\n\r\nDownload table as: \r\nASCIITypeset image\r\n\r\nTable
  5. Kinematic Properties of Broad and Absorption Lines\r\n\r\nLine\tλrest\tFWHM\tW\tΔv\r\n
  \t(Å)\t(km s−1)\t(km s−1)\t(km s−1)\r\nBroad Emission Lines\r\nHα\t6562.8\t1024
  ± 21\t3010 ± 450\t−7 ± 16\r\nHe i\t6678.2\t1041 ± 792\t3890 ± 2083\t−137 ± 151\r\nHe
  i\t7065.2\t473 ± 116\t1000 ± 600\t−136 ± 82\r\nO i\t8446.4\t562 ± 331\t1120 ± 590\t+39
  ± 80\r\nPa10\t9014.9\t766 ± 202\t2100 ± 600\t−155 ± 39\r\nPa9\t9229.0\t766 ± 202\t2100
  ± 600\t−155 ± 39\r\nPa8\t9545.6\t766 ± 202\t2100 ± 600\t−155 ± 39\r\nPaδ\t10049.4\t1672
  ± 458\t3500 ± 400\t+23 ± 44\r\nHe i\t10830.3\t536 ± 45\t2760 ± 220\t+154 ± 22\r\nPaγ\t10938.1\t1035
  ± 322\t2400 ± 260\t−75 ± 31\r\nO i\t11287.0\t1188 ± 1099\t4300 ± 2000\t+245 ± 214\r\nAbsorption
  Lines\r\nHα\t6562.8\t1000 ± 123\t⋯\t−200 ± 32\r\nHe i\t7065.2\t587 ± 1024\t⋯\t−52
  ± 83\r\nHe i\t10830.3\t794 ± 21\t⋯\t+45 ± 22\r\nNote. Velocity offsets (Δv) calculated
  relative to [S iii] λ9071 at z = 3.50102.\r\n\r\nDownload table as: \r\nASCIITypeset
  image\r\n\r\n3.3.1. Hα Complex\r\nHα is modeled with independent narrow, broad,
  and absorption components. [O i] λ6302 and He i λ6680 are modeled with narrow and
  broad profiles. [N ii] lines are fixed at a 1:3 ratio and share kinematics; [S ii]
  lines have independent amplitudes but tied velocities and widths.\r\n\r\nWe find
  that broad Hα and He i λ6680 are strongly favored to have exponential wings (ΔBIC
  ∼ −500), yielding a FWHM (of the intrinsic Gaussian core) of ∼1000 km s−1—significantly
  narrower than the ∼3000 km s−1 derived from pure Gaussians. Although no distinct
  Hα absorption component is explicitly detected, likely due to insufficient resolution,
  the overall asymmetry of the line is best reproduced when a weak, blueshifted absorber
  is included in the fit. This may indicate subtle self-absorption or partial obscuration
  within the dense cocoon. We return to the implications of this profile shape later.\r\n\r\n3.3.2.
  He i λ7065\r\nThe He i λ7065 line shows a prominent blueshifted absorption component.
  We fit it with narrow, broad, and absorption profiles. The rest of the lines, including
  [Ar iii] λ7138, Fe ii λ7156, and the [O ii] triplet (λλ7323, 7325, 7332), are modeled
  as independent narrow lines.\r\n\r\nHe i λ7065 shows a strong preference for an
  exponential profile (ΔBIC ∼ − 16) with a FWHM = 473 ± 116 km s−1. A residual bump
  spanning from ∼λ7290 to λ7320 is likely a combination of multiple Fe ii, such as
  λ7290, λ7308, and potentially Fe iii λ7319.65 lines. Curiously, the latter line
  is often labeled as “hazy” in various emission-line libraries (e.g., A. Kramida
  et al. 2024), reflecting its undefined shape as a result of pressure broadening
  or scattering in very dense gas.\r\n\r\n3.3.3. O i λ8448 and Iron Lines\r\nIn this
  part of the spectrum the O i λ8448 is fit with narrow and broad components. All
  of the iron lines, including Fe ii λλ8228, 8239, 8289, 8470, 8490 as well as a potential
  Fe iii λ8306, are fit as narrow lines with shared kinematics. O i components are
  fit independently.\r\n\r\nWe find that the broad O i component is best described
  by a standard Gaussian profile, with a stronger statistical preference over exponential
  wings (ΔBIC ∼ 4). However, we caution that a trough at ∼3.77 μm, likely an artifact,
  may affect the reliability of the fit in this region.\r\n\r\n3.3.4. Iron Forest
  and Paschen Lines\r\nThis spectral region is among the most complex, due to the
  dense clustering of emission features. The Paschen lines Pa10–Pa8 are modeled with
  narrow and broad components. To reduce the number of free parameters, we tie all
  narrow Paschen components together kinematically, and do the same for the broad
  components. Fe ii λλ9075, 9125, 9134,9179, 9204, 9394 are fit with a shared centroid
  and width. [S iii] lines are fit independently.\r\n\r\nThe [S iii] λ9071 line is
  used to define the systemic redshift: z = 3.50102 ± 0.00019. This is done because
  forbidden lines, such as [S iii] or [O iii], have lower critical densities that
  are incompatible with the dense-gas envelopes giving rise to the other lines we
  observe. These lines may instead come from the host galaxy itself (e.g., R. Maiolino
  et al. 2024), and we therefore choose this line to define the overall redshift of
  the system and to measure all the offsets relative to it. All velocity offsets discussed
  in the subsequent sections are computed relative to this reference frame. The Paschen
  lines show strong preference for exponential wings (ΔBIC ∼ −38), with a FWHM = 766
  ± 202 km s−1—narrower than Hα, but consistent within 2σ.\r\n\r\n3.3.5. Paδ\r\nThe
  Fe ii λ9997 and Paδ complex is modeled with an independent narrow component for
  Fe ii and both narrow and broad components for Paδ. There is no evidence for absorption.
  Exponential wings are again preferred (ΔBIC ∼ −26). The resulting FWHM of 1672 ±
  458 km s−1 is notably larger than those of both higher-order Paschen lines and Hα,
  though still consistent with the latter within 2σ. It is also possible that the
  line shape is impacted by Fe ii λ9997 and the undetected (but likely present) Fe
  ii λλ10131, 10173 lines.\r\n\r\n3.3.6. He i λ10830 and Paγ\r\nThe final window contains
  a complicated blend of broad He i λ10830 and Paγ lines, with a clear blueshifted
  absorption in the former. We fit three components (narrow, broad, and absorption)
  to He iλ10830 and narrow plus broad components to Paγ, keeping everything kinematically
  independent. We fit narrow Fe ii lines at 10490, 10501, and 11128 Å with fixed redshift
  and line width. Finally, O i λ11290 is fit with independent narrow and broad components.
  Exponential wings are again preferred over a pure Gaussian profile.\r\n\r\n4. Data
  Analysis\r\n4.1. Morphology\r\nLRDs are unresolved in the rest-optical by definition,
  with measured sizes consistent with the PSF HWHM ( in F444W), and in some cases
  even smaller when dithers align favorably (e.g., I. Labbé et al. 2025). Gravitational
  lensing can further push constraints on their intrinsic sizes to ≲100 pc (L. J.
  Furtak et al. 2024).\r\n\r\nIn the rest-UV, however, a more complex picture is emerging.
  Several studies have now reported faint, extended, asymmetric components adjacent
  to the compact core (V. Kokorev et al. 2024b; I. Labbe et al. 2024; J. Matthee et
  al. 2024; P. Rinaldi et al. 2025a, 2025b), often suppressed by surface-brightness
  dimming. As shown in Figure 1, GLIMPSE-17775 likewise consists of two components
  out to F200W (rest ∼ 4000 Å), coincident with the Balmer break. However, we note
  that the break itself is much weaker than found in objects with very similar spectra
  (e.g., B. Wang et al. 2024; see Figure 3). At longer wavelengths, a point-source
  (PS) morphology dominates, consistent with a black-hole-dominated core. To quantify
  this transition, we model the NIRCam imaging using PYSERSIC (I. Pasha & T. B. Miller2023)
  with a minimal configuration: a fixed-center PS plus a freely offset Sérsic profile.
  Normalizations, n, and reff are all allowed to vary, and uncertainties are drawn
  from the Markov Chain Monte Carlo posteriors (e.g., V. Kokorev et al. 2024a).\r\n\r\nZoom
  InZoom OutReset image size\r\nFigure 3. The diversity of Balmer breaks in LRDs.
  The black points show the HST and JWST GLIMPSE photometry of GLIMPSE-17775. The
  blue line shows best-fit EAZY SED fit to the photometry only, fixing the redshift
  to the zspec. The maroon line show the combined and photometry-corrected G395M spectrum.
  While the red color in F200W–F277W is partially influenced by a bright Hα line,
  the Balmer break between F150W and F200W is still prominent. We further show spectra
  of various other LRDs (I. Labbe et al. 2024; B. Wang et al. 2024; A. de Graaff et
  al. 2025b; R. P. Naidu et al. 2025; A. J. Taylor et al. 2025), all shifted to z
  = 3.501 and normalized at 5100 Å. Finally, we show HST (blue) and JWST (orange)
  filter transmission curves below.\r\n\r\nDownload figure:\r\n\r\nStandard imageHigh-resolution
  image\r\nOur fits show that the host galaxy dominates up to λrest ∼ 4000 Å, where
  the host and nucleus contribute roughly equally, beyond which the PS prevails to
  λrest ∼ 1 μm. Because the NIRCam LW detector has ≳2× poorer resolution than the
  SW, a clean separation of host and nucleus in the rest-optical is impractical (see,
  e.g., K. E. Whalen et al. 2026). Fortunately, F200W lies at the transition where
  both the resolution and flux ratio are favorable, so we use that band to illustrate
  our galaxy/LRD decomposition. We illustrate this in Figure 4, which also shows the
  increasing PS fraction with wavelength, while acknowledging the uncertainties at
  longer wavelengths.\r\n\r\nZoom InZoom OutReset image size\r\nFigure 4. Top: two-component
  Sérsic+point-source (PS) fit to the F200W morphology of GLIMPSE-17775. Panels show,
  from left to right, the data, best-fit model, model with the extended component
  removed, and residuals. Bottom: fractional PS contribution vs. rest-frame wavelength.
  Shaded regions mark SW (blue) and LW (red) detectors, where the lower resolution
  of LW filters hinders reliable two-component decomposition.\r\n\r\nDownload figure:\r\n\r\nStandard
  imageHigh-resolution image\r\nAfter applying the lensing correction, we find that
  the extended component dominating the UV light has a radius of reff ∼ 1000 pc, whereas
  the actual compact LRD, dominant in the rest-optical, is fully consistent with the
  PSF and reff < 300 pc.\r\n\r\n4.2. Dust Attenuation\r\nThe ratio between observed
  emission-line fluxes is commonly used to estimate dust extinction. For LRDs, this
  has typically been achieved through the Balmer decrement, provided that multiple
  Balmer lines are available. In our case, only Hα is detected. While the Paschen
  series, of which five lines are observed, could, in principle, be used to estimate
  extinction, the likely presence of stratified dense gas in LRDs (A. de Graaff et
  al. 2025b; R. P. Naidu et al. 2025) implies that the intrinsic line ratios may deviate
  significantly from standard Case B recombination expectations (S.-J. Chang et al.
  2026). Further, at the wavelengths of the Paschen series, the attenuation curves
  are practically flat (e.g., D. Calzetti et al. 2000). Small offsets of the Case
  B ratios would have a significant effect on the derived dust attenuation, but in
  return require extreme precision on measured line fluxes to be meaningful. Our uncertainties
  on, e.g., Paγ/Paδ are small, roughly 10%–15%, yet this still leads to an impractically
  high uncertainty on AV. To circumvent this limitation, we instead employ the ratio
  between the permitted O i λ8448 and O i λ11290 lines.\r\n\r\nThese transitions share
  a common energy level, and under the assumption of Bowen (Lyβ) fluorescence, all
  transitions through the λ11290 transition also cascade through the λ8448. This makes
  their intrinsic intensity set solely by the ratio of the inverse of their wavelengths.\r\n\r\nTo
  ensure we are comparing the same physical gas component, we isolate the narrow emission
  peaks in both transitions, which show consistent widths of FWHM ∼ 250 km s−1. This
  similarity supports the use of their flux ratio as a dust probe. We adopt the Small
  Magellanic Cloud (SMC) attenuation curve (K. D. Gordon et al. 2003), widely used
  for high-redshift galaxies and reddened AGN (P. F. Hopkins et al. 2004; P. L. Capak
  et al. 2015; N. A. Reddy et al. 2015, 2018; V. Kokorev et al. 2023; A. J. Taylor
  et al. 2025). Assuming an intrinsic ratio of (8448/11290)int = 1.336 (D. E. Osterbrock1989),
  we infer AV = 0.1 ± 0.3 mag, consistent with negligible extinction.\r\n\r\nWe caution,
  however, that this ratio probes only the narrow-line-emitting gas, which could originate
  either in the host galaxy or within a narrow-line region associated with the AGN.
  If the latter is the case, then the red optical continuum slope (βopt ∼ 0.35) would
  suggest that the observed SED is intrinsically red rather than reddened by dust.
  If instead the narrow lines arise predominantly in the host galaxy, this constraint
  does not directly inform the origin of the continuum emission. Similar Case B–consistent
  narrow-line ratios have been reported in other LRDs (e.g., M. Brooks et al. 2025;
  G. P. Nikopoulos et al. 2025), where the broad lines were found to show significant
  departures from Case B.\r\n\r\nAlthough the wavelength coverage of the red grating
  does not allow a Balmer decrement test for the broad-line emission (only Hα is detected),
  we can perform a consistency check using the brightest broad Paschen lines (Paγ,
  Paδ, and Pa8). Adopting intrinsic Case B ratios from P. J. Storey & D. G. Hummer
  (1995; Paγ/Paδ ≃ 1.5 and Paδ/Pa8 ≃ 1.8), and comparing these to our measured values
  of Paγ/Paδ = 1.77 ± 0.15 and Paδ/Pa8 = 1.94 ± 0.20, we find the broad Paschen lines
  to be consistent with Case B within ≃2σ, particularly when systematic uncertainties
  from blending and profile decomposition are taken into account. If Case B is applicable
  in the dense environments surrounding the central AGN, this would be consistent
  with a low-dust or dust-free origin for the broad-line-emitting gas.\r\n\r\n5. Line
  Properties\r\nOur fitting procedure yields over 40 emission and absorption features,
  most of which are detected at a high (S/N> 3) significance. With all the pieces
  in place, we now comment on the line profiles, kinematics, and the specific line
  species that we identify.\r\n\r\n5.1. Exponential Wings\r\nAs already noted throughout
  Section 4, and further highlighted in Figure 2, all permitted lines in the spectrum,
  with the exception of O iλ8448, which is likely impacted by data-quality issues,
  are better fit when the broad component is convolved with an exponential profile.
  The statistical preference for this model (as measured by ΔBIC) is especially strong
  for our brightest line, Hα. To further demonstrate this, and to compare this fit
  with a more standard, Gaussian-only approach, we show both models in Figure 5. When
  the broad line is fit using only a Gaussian profile, large portions of the line
  are underfit at a ≳3σ significance level, whereas a model that includes exponential
  wings shows a much smoother, albeit not perfect, residual plot.\r\n\r\nZoom InZoom
  OutReset image size\r\nFigure 5. Exponential wings are required. Comparison of a
  Gaussian×exponential model (red) and a single Gaussian profile (blue) for Hα. Uncertainties
  are shown as vertical lines in each spectral bin, although they are too small to
  be visible. The exponential model provides a far superior fit, with smoother residuals
  and a strongly preferred ΔBIC, highlighting the necessity of exponential wings to
  capture the broad-line shape.\r\n\r\nDownload figure:\r\n\r\nStandard imageHigh-resolution
  image\r\nThis strong preference for exponential wings is consistent with expectations
  from electron scattering in dense, ionized gas. Unlike Gaussian profiles, which
  arise from Doppler broadening due to thermal and bulk motions around the supermassive
  black hole, exponential profiles arise in environments dominated by electron scattering
  (e.g., S.-J. Chang et al. 2026; V. Rusakov et al. 2026). The absence of strong broadening
  in the forbidden lines (e.g., [S ii], [S iii], and [N ii]) already suggests that
  the densities where these electron-scattered profiles originate must already exceed
  multiple times the critical densities of these transitions (ne > 106). Combined
  with the fact that these broad profiles are non-Gaussian in shape, both the likely
  volume and column densities are likely even higher, at ne = 108 cm−3 and Ne ∼ 1024
  cm−2, respectively, as suggested by both K. Inayoshi & R. Maiolino (2025) and V.
  Rusakov et al. (2026).\r\n\r\nRecent work (e.g., M. Brazzini et al. 2025) argues
  that, in classical broad-line region (BLR) environments, an electron-scattering
  origin for broad wings requires similar intrinsic and scattering widths across recombination
  lines. We therefore examine these criteria for the hydrogen transitions and find
  that, within the uncertainties of our fits, both the intrinsic and exponential widths
  are broadly comparable among the recombination lines. However, unlike classical
  BLRs, LRDs are expected to be strongly stratified, with different recombination
  transitions forming at different depths and probing different effective electron
  columns (e.g., A. de Graaff et al. 2025d; A. Sneppen et al. 2026). In such a structure,
  strict equality of widths is not expected even if a common scattering kernel shapes
  the wings. The modest variations observed here are therefore consistent with electron
  scattering in a dense, stratified cocoon.\r\n\r\nIt is important to note that such
  detailed profile decomposition is only possible due to the exceptionally high S/N
  per pixel achieved in the line wings of our NIRSpec G395M spectrum. As emphasized
  in V. Rusakov et al. (2026), these features would be impossible to distinguish with
  shallower data or lower-resolution modes such as NIRSpec/PRISM, or even G395M exposures
  lacking comparable depth.\r\n\r\n5.2. Line Profiles\r\nFurther, in Figure 6, we
  compare the best-fit models for all broad lines. As noted previously, with the exception
  of O i λ8448, every broad line is better described by exponential wings, resulting
  in very similar overall line shapes. The main differences arise in their widths.\r\n\r\nZoom
  InZoom OutReset image size\r\nFigure 6. Best-fit profiles of all broad lines, oversampled
  and shifted to a common center. The top panel shows hydrogen recombination (Hα,
  black, and Paschen series), the middle panel shows O i, and the bottom panel shows
  He i. Shaded regions indicate 1σ uncertainties for the least-constrained lines (Paδ,
  O i λ11290, and He i λ6680); uncertainties for the remaining lines are omitted for
  clarity. We note that hydrogen and O i lines show largely similar widths, consistent
  with their coupling through charge exchange. By contrast, He i lines are systematically
  narrower, likely reflecting their distinct metastable triplet physics and formation
  in a less dense outer region.\r\n\r\nDownload figure:\r\n\r\nStandard imageHigh-resolution
  image\r\nDespite some fits having sizable uncertainties, a trend emerges across
  line species. Hydrogen recombination lines (with the exception of Paδ) and O i transitions
  consistently show comparable intrinsic FWHM values of ∼800 km s−1. This agreement
  is not coincidental. Neutral oxygen and ionized hydrogen have nearly identical ionization
  potentials, enabling rapid charge-exchange coupling between O+ + H0⇌ O0 + H+ (D.
  E. Osterbrock & G. J. Ferland 2006). This continual electron exchange tightly locks
  the spatial distribution, ionization state, and kinematics of neutral oxygen to
  ionized hydrogen (and vice versa), so their broad-line profiles are naturally expected
  to track one another. The observed similarity therefore provides an important consistency
  check, reinforcing that the O i emission originates in the same dense, ionized gas
  as the hydrogen recombination lines. In contrast, the Fe ii lines exhibit significantly
  narrower widths, matching the narrow cores of the permitted transitions rather than
  their exponential wings. This implies that the Fe ii emission arises in a cooler,
  less turbulent zone exterior to the scattering-dominated region, but still closely
  coupled to the AGN continuum source.\r\n\r\nBy contrast, He i emission is systematically
  narrower. He i lines such as λ7065 and λ10830 do not couple to hydrogen via charge
  exchange, and their lower metastable level (23S) has distinct population physics,
  being fed by both collisional and recombination pathways. Further, one has to take
  into account the radiative-transfer effects, which make it possible to radiatively
  excite metastable ground-state electrons. As a result, He i emission may arise from
  a somewhat different spatial or kinematic region than H and O. We return to this
  point in more detail below.\r\n\r\nTaken together, the similar FWHM of hydrogen
  and oxygen broad-line profiles, combined with the near-universal exponential wings
  seen across all permitted transitions, strongly suggests that the line shapes are
  set by a common, line-independent scattering kernel rather than by transition-specific
  processes. Electron (Thomson) scattering in a dense, ionized cocoon provides a natural
  explanation: It produces exponential wings of nearly identical form across species,
  with widths set primarily by the electron temperature and column density (typically
  ∼1024 cm−2), and leaves only secondary variations from species-specific excitation
  or optical-depth effects (e.g., S.-J. Chang et al. 2026; V. Rusakov et al. 2026).
  In this framework, the narrower He i lines reflect stratification within the cocoon,
  where He i emission arises from an outer region that has a lower density, temperature,
  or column density, while the bulk hydrogen and oxygen emission share a common kinematic
  imprint within an inner, denser region.\r\n\r\n5.3. Line Kinematics\r\nBefore we
  proceed to discussing the individual features in more detail, we would like to comment
  on the systematic velocity offsets between various line species in the spectrum
  of GLIMPSE-17775. As mentioned previously, we choose the redshift of the narrow
  forbidden [S iii] λ9701 as the systemic reference, due to it being relatively bright
  and isolated. We place all velocity offsets in context by plotting all values from
  Tables 4 and 5 in Figure 7.\r\n\r\nZoom InZoom OutReset image size\r\nFigure 7.
  Velocity offsets (relative to [S iii]λ9071) and FWHM for all detected lines. Narrow
  permitted lines are shown as maroon circles (open for forbidden), broad lines as
  blue, absorption as gold, and Fe ii lines as green diamonds. The shaded band marks
  the velocity uncertainty set by the median spectral resolution. Narrow lines align
  closely with the systemic redshift, while absorption features show moderate blueshifts
  of ∼150 km s−1.\r\n\r\nDownload figure:\r\n\r\nStandard imageHigh-resolution image\r\nWhile
  every fit formally yields a centroid, the significance of any offset is constrained
  by the spectral resolution and S/N of the data. To gauge which velocity shifts are
  meaningful, we adopt a conservative uncertainty of ΔV ∼ med(R)/5, with R = λ/Δλ.
  This corresponds to ∼60 km s−1, below which offsets are unlikely to be statistically
  significant.\r\n\r\nQualitatively, we do not find systematic shifts of either permitted
  or forbidden narrow lines relative to the systemic redshift. Iron lines also appear
  consistent with the systemic velocity. In contrast, the permitted broad lines, together
  with their associated absorption components, show a consistent blueshift of order
  ∼100 km s−1.\r\n\r\nSuch blueshifts are commonly interpreted as signatures of outflowing
  gas in the BLR, where scattering and absorption occur preferentially along outflowing
  sight lines. In the context of GLIMPSE-17775, this modest but systematic offset
  strengthens the case for a stratified cocoon, while the narrow lines trace gas in
  the outer regions at systemic velocity, the BLR and absorbing layers appear to be
  participating in bulk outflows. Confirming the detailed velocity structure, however,
  will require higher-resolution follow-up with the NIRSpec H-gratings (e.g., A. Saldana-Lopez
  et al. 2025; A. Torralba et al. 2026).\r\n\r\n5.4. Hydrogen Absorption Lines\r\nAlthough
  no strong hydrogen absorption lines are explicitly detected, the Hα profile (Figure
  2) shows a noticeable asymmetry, with suppressed flux on the blue side of the line.
  This feature is best modeled by a blueshifted (∼−200 km s−1) absorption component,
  as already noted in Section 4. Similar nonresonant absorption signatures (e.g.,
  Balmer lines) are now routinely observed in LRDs across a wide range of redshifts
  (e.g., J. E. Greene et al. 2024; V. Kokorev et al. 2024b; I. Labbe et al. 2024;
  X. Lin et al. 2024; A. J. Taylor et al. 2025). The presence of such absorption implies
  high gas densities of order n ∼ 109 cm−3 (P. B. Hall 2007; K. Inayoshi & R. Maiolino
  2025). Interestingly, only atoms in the n = 2 state appear significantly populated,
  as we detect no evidence for Paschen absorption in any of the five lines present
  in our spectrum. This is not an instrumental effect of the M-grating, as we clearly
  detect an absorption feature in He i λ10830 but not in the immediately adjacent
  Paγ. This indicates that the n = 3 state is comparatively underpopulated.\r\n\r\nAnother
  key manifestation of the same physical mechanism is the Balmer break, corresponding
  to the n = ∞ → n = 2 transition limit. A prominent break is common in many LRD spectra
  (D. J. Setton et al. 2025), though not ubiquitous (e.g., V. Kokorev et al. 2023;
  R. Tripodi et al. 2025). While our spectral coverage does not extend to the Balmer
  limit itself, we observe a discontinuity of ∼1 mag between the F200W and F150W filters.
  Due to the F200W coverage of the SED, this flux jump cannot be attributed to line
  boosting from Hβ and the [O iii] doublet. Adopting the break parameterization of
  R. P. Naidu et al. (2025), we use our best-fit SED and measure fν,4050Å/fν,3670Å
  = 2.02 ± 0.10. Although weaker than the extreme values (∼4–7) reported for LRDz9
  (A. J. Taylor et al. 2025), MoM BH*-1 (R. P. Naidu et al. 2025), or “the Cliff”
  (A. de Graaff et al. 2025b), this value lies close to the maximum achievable by
  evolved stellar populations (I. Labbe et al. 2024; B. Wang et al. 2024). However,
  this measurement is based on the total (PS+host) photometry. Given that the UV continuum
  is spatially resolved, as we have shown in Section 4.1, host-galaxy contamination
  may dilute the intrinsic break strength of the central component. While uncertainties
  on the PS/host decomposition increase significantly toward the rest-optical, we
  find that using the PS-only photometry produces a noticeably sharper break, increasing
  the inferred strength by ∼30%–50% to fν,4050Å/fν,3670Å ≈ 2.6–3.0. Although this
  does not allow us to robustly conclude whether the intrinsic PS break reaches the
  most extreme BH*-type objects (A. de Graaff et al. 2025b; R. P. Naidu et al. 2025;
  A. J. Taylor et al. 2025), it suggests that the Balmer discontinuity becomes stronger
  once the host contribution is removed, consistent with recent findings that host
  light can suppress the apparent break amplitude (e.g., A. de Graaff et al. 2025d;
  W. Q. Sun et al. 2026).\r\n\r\nTaken together, the presence of asymmetry in the
  Hα profile, likely caused by blueshifted Balmer absorption, and a Balmer break adds
  to the evidence, complementing the exponential line wings, that the emission in
  GLIMPSE-17775 arises from a dense, ionized cocoon of gas. The lack of a detectable
  Paschen break is consistent with this picture, as high densities and optical depths
  result in Thomson scattering while also naturally washing out higher-order continuum
  edges (e.g., B. Wang et al. 2025). Assuming LTE, an extremely low ratio n3/n2 ≲
  0.01 corresponds to an electron temperature of Te ∼ 5000 K or below, consistent
  with the warm, partially ionized gas expected in dense LRD cocoons.\r\n\r\nWe also
  observe moderately prominent and extremely prominent blueshifted absorption, respectively,
  in He i λ7065 and He i λ10830. The mechanism for this is similar, but not identical
  to, the hydrogen lines. We also note the prominent blueshifted absorption seen in
  both He i λ7065 and He i λ10830. However, the physical mechanism driving these helium
  features differs fundamentally from that of the hydrogen lines; we return to this
  in the following section.\r\n\r\n5.5. Helium Lines\r\nBoth He i λ7065 and He i λ10830
  show blueshifted (∼70−80 km s−1 from the narrow-line center) absorption components
  with widths of ∼600−700 km s−1. Unlike the Balmer series, helium transitions are
  resonantly scattered, so the arguments regarding n = 2 state abundances do not apply.
  Before addressing the absorption itself, it is useful to consider what the observed
  line strengths already tell us.\r\n\r\nIn the spectrum of GLIMPSE-17775, we detect
  three He i lines: λ6680, λ7065, and λ10830. These originate from different physical
  mechanisms. The λ6680 singlet line is produced via recombination or radiative pumping,
  and is only weakly dependent on density. By contrast, λ7065 and λ10830 belong to
  the triplet system (along with λ3889, not covered by G395M), where the lower level
  is populated due to its very long lifetime. Electrons in this “ground state” can
  be collisionally or radiatively excited into higher energies, and therefore the
  triplet-line strengths depend strongly on density and, to a lesser extent, temperature.
  As emphasized by D. A. Berg et al. (2026), He i λ10830 is the most sensitive density
  diagnostic among these transitions, followed by λ7065. The fact that both are much
  stronger than λ6680 indicates that the region where they form is very dense. The
  prominence of λ10830 in particular already points to a high-density environment
  with a temperature T ≳ 104 K. Further, the broad λ6680 line has a noticeably larger
  FWHM than the triplet lines but is consistent with hydrogen recombination lines,
  which again makes sense since both originate from the same mechanism.\r\n\r\nWhat
  about absorption? Its presence further reinforces this picture. The lower level
  (23S) of the triplet system is a metastable, long-lived state that effectively acts
  as a ground state in dense or partially ionized gas. As a result, photons from the
  23P → 23S transition can be resonantly absorbed and reemitted many times, analogous
  to Lyα or Mg ii. This mechanism naturally produces strong absorption features when
  the He i column density is high. The fact that He i λ10830 not only dominates the
  helium spectrum in emission but also shows the deepest absorption in the entire
  spectrum provides compelling evidence for a dense, ionized cocoon of gas enshrouding
  GLIMPSE-17775.\r\n\r\n5.6. Oxygen Lines\r\nAbove, we noted that O i λ8446 and λ11290
  share very similar profiles with the hydrogen recombination lines. This is naturally
  expected because neutral oxygen is tightly coupled to hydrogen via charge exchange,
  which rapidly equilibrates the O/H ionization states in dense gas. When charge exchange
  is fast (high nH), the O i/O ii ratio tracks the H i/H ii ratio, so the O i-emitting
  gas shares the same kinematics as the hydrogen recombination region (B. T. Draine
  2011). The observed agreement in FWHM between H and O i lines is therefore an important
  consistency check that both species originate in the same dense, partially ionized
  phase.\r\n\r\nA second, independent clue of high gas density and radiation field
  intensity comes from the excitation mechanism. The O iλ8446–λ11287 pair is the classic
  signature of Lyβ (Bowen) fluorescence: Lyβ pumps O i from the ground state and the
  ensuing cascade preferentially populates the levels that emit at 1.129 μm, which
  then cascade down through the 8446 Å line. In this channel, the line emissivity
  scales with both the neutral oxygen (or equivalently neutral hydrogen, via charge
  exchange) column and the local Lyβ radiation field. Therefore, producing strong
  O i fluorescence requires both a very bright Lyβ source and a dense neutral (or
  partially ionized) cocoon.\r\n\r\nFurther, we do not detect (5σ upper limit <12/[10−20
  erg s−1 cm−2]) any of the other permitted O iλ7774, λ7254, and λ7790 lines, and
  the [O i] λ6302/O i λ8446 ratio is weak. If collisional excitation or pure recombination
  dominated, these lines would be comparatively strong; their absence strongly favors
  Bowen fluorescence as the primary driver of the observed O i emission.\r\n\r\nFinally,
  the velocity structure adds a natural stratification: hydrogen recombination and
  O i (fluorescent, charge-exchange coupled) share similar widths and profiles, while
  He i lines are systematically narrower. Since He i triplet transitions emerge from
  a metastable 23S level (with distinct, density-sensitive population pathways and
  resonant transfer), they likely trace a kinematically distinct layer within the
  same cocoon.\r\n\r\n5.7. Lyα Fluorescence in Iron Lines\r\nPermitted iron emission,
  primarily Fe ii and Fe iii (e.g., I. Labbe et al. 2024; R. Tripodi et al. 2025;
  A. Torralba et al. 2026), but in some cases extending to highly ionized species
  such as Fe vii (E. Lambrides et al. 2025; M. Tang et al. 2025) and Fe x (L. J. Furtak
  et al. 2024), has become a recurring feature in LRD spectra. In GLIMPSE-17775, the
  spectrum is exceptionally rich in NIR iron lines; we identify 16 Fe ii and one Fe
  iii transition. Understanding the origin of this emission is key to interpreting
  the dense-gas environment in LRDs.\r\n\r\nThe physics of Fe ii emission has long
  been a challenge for BLR photoionization models (M. Joly 1993). Thick, high-column-density
  (∼1025 cm−2) gas at the edges of the accretion disk has been invoked as a potential
  source (M. Joly 1987; S. Collin-Souffrin et al. 1988), where the scattering and
  absorption of the hard X-ray photons ionize the gas. Similarly, this would also
  enhance the Balmer and Paschen line luminosities. Further, extensive theoretical
  work has shown that Lyα fluorescence is fundamental in reproducing the observed
  Fe ii strengths (T. A. A. Sigut & A. K. Pradhan 1998, 2003). Lyα pumping not only
  boosts the UV and optical Fe ii emission, but also predicts strong lines in the
  NIR.\r\n\r\nThis expectation aligns closely with our observations. In Figure 8,
  we compare our measured Fe ii flux ratios (normalized to Fe ii λ9075) against the
  Lyα-pumped model predictions of T. A. A. Sigut & A. K. Pradhan (2003). With the
  exception of Fe ii λ9179, affected by blending with Fe ii λ9204 and Pa9, we find
  remarkably good agreement across the suite of detected features. In particular,
  the dense forest of Fe ii transitions spanning λλ9000–9200 (center-right panel in
  Figure 2) is reproduced almost exactly by the theoretical spectrum (see Figure 11
  in T. A. A. Sigut & A. K. Pradhan 2003). We overlay the T. A. A. Sigut & A. K. Pradhan
  (2003) model directly on our continuum-subtracted fit in Figure 8, demonstrating
  that the observed Fe ii emission is fully consistent with fluorescence in dense
  gas near the BLR. The implication is that the majority of the Fe ii emission in
  GLIMPSE-17775 arises through Lyα pumping, directly tracing an extremely dense medium,
  ionized by an extremely luminous source, likely an accreting black hole. Given that
  the Fe ii lines are somewhat narrower than other permitted features, it is likely
  that the emitting region is located farther from the BLR—something that has already
  been shown in a classic example of a narrow Fe ii emitter, I Zw 1 (R. J. Rudy et
  al. 2000; A. O. M. Marinello et al. 2016).\r\n\r\nZoom InZoom OutReset image size\r\nFigure
  8. Lyα-pumped iron emission in GLIMPSE-17775. Left: observed Fe ii flux ratios (black
  points), normalized to Fe iiλ9075, compared with predictions from Lyα fluorescence
  models of T. A. A. Sigut & A. K. Pradhan (2003, blue). The close correspondence
  across ∼14 lines indicates a common excitation mechanism driven by Lyα pumping in
  dense, partially shielded gas. Right: zoom in on the λλ9000–9400 complex, showing
  the remarkable agreement between the modeled continuum-subtracted Fe ii spectrum
  (black) and the T. A. A. Sigut & A. K. Pradhan (2003) model (blue). We show air
  wavelengths of each Fe ii line directly from T. A. A. Sigut & A. K. Pradhan (2003),
  including the blended features. Together, these comparisons support an origin in
  the dense inner regions of the cocoon surrounding the accreting black hole.\r\n\r\nDownload
  figure:\r\n\r\nStandard imageHigh-resolution image\r\nTogether with the Balmer absorption,
  helium triplet features, and Bowen fluorescence in O i, the iron forest adds another
  piece of evidence that points to a dense cocoon enshrouding the black hole.\r\n\r\n5.8.
  Source Properties\r\nBased on the previous discussion, we now present the likely
  physical properties of the black hole and its host. The coexistence of broad permitted
  and narrow forbidden lines indicates, as in most LRDs (R. E. Hviding et al. 2025),
  that the broad components originate in an AGN BLR. Following J. E. Greene & L. C.
  Ho (2005), we derive MBH from the luminosity and width of the broad Hα line, assuming
  the exponential-wing profile demonstrated in Figures 2 and 5. With a FWHMHα = 1024
  ± 21 km s−1, negligible dust, and μ = 2.04, we obtain , where the dominant uncertainty
  arises from the J. E. Greene & L. C. Ho (2005) calibration. A purely Gaussian fit,
  though statistically disfavored, would yield a FWHM 3 times larger and a black hole
  mass nearly a dex higher. Throughout this work, we adopt the MBH from the exponential
  model.\r\n\r\nFurther, by assuming that the bolometric luminosity (Lbol) scales
  as Lbol = 130 × LHα (G. T. Richards et al. 2006), we find Lbol = (1.06 ± 0.14) ×
  1045 erg s–1. Using our MBH derived by assuming exponential wings, we find that
  this object is accreting at a super-Eddington rate, Lbol/Ledd = 1.86 ± 0.25. This
  is higher than the vast majority of LRDs at all redshifts (e.g., V. Kokorev et al.
  2023; L. J. Furtak et al. 2024; I. Juodžbalis et al. 2024; R. Maiolino et al. 2024;
  H. B. Akins et al. 2025b; D. D. Kocevski et al. 2025; A. J. Taylor et al. 2025),
  found to accrete at sub-Eddington rates. However, recent works examining LRDs with
  extreme Balmer breaks derive accretion rates that exceed the Eddington limit (E.
  Lambrides et al. 2024; A. de Graaff et al. 2025b; R. P. Naidu et al. 2025).\r\n\r\nA
  recent work examining multiwavelength LRD data suggests that the bolometric luminosity
  emerges predominantly from the rest-frame optical, with X-ray and radio contributions
  being largely subdominant (J. E. Greene et al. 2026). Given the suggested dominance
  of optical light in LRDs, the much lower bolometric corrections (×7–10 lower) would
  imply lower black hole masses and total bolometric luminosities (e.g., those derived
  from Hα) than standard AGN prescriptions (e.g., J. E. Greene & L. C. Ho 2005) might
  suggest. This adjustment would lower both the inferred black hole masses and bolometric
  luminosities. Because these two quantities scale together, the implied super-Eddington
  nature of GLIMPSE-17775 would remain unchanged, although the black hole mass could
  decrease to ∼105.5–105.8M⊙. This interpretation is consistent with dense-gas (BH*)
  models, which generically predict Eddington or super-Eddington accretion in heavily
  obscured environments (e.g., D. Kido et al. 2025; H. Liu et al. 2025; A. Sneppen
  et al. 2026). A more detailed reassessment of bolometric corrections for LRDs is
  clearly warranted, but such an analysis lies beyond the scope of this work. For
  consistency with previous studies, we adopt standard AGN bolometric corrections
  throughout, with all derived parameters listed in Table 1.\r\n\r\nFinally, we estimate
  an upper limit on the stellar mass in GLIMPSE-17775. Previous studies have modeled
  LRDs using joint galaxy+AGN SED decomposition (e.g., V. Kokorev et al. 2023; L.
  J. Furtak et al. 2024), and more recently, within the BH* framework, have assumed
  that most of the rest-UV flux arises from the host galaxy (R. P. Naidu et al. 2025;
  A. J. Taylor et al. 2025). Our spatial decomposition (Figure 4) supports this assumption:
  The extended component contributes >80% of the rest-UV light, implying that the
  bulk of the stellar mass resides in this resolved structure. Given this, we adopt
  a simple empirical approach using the MUV–M* relation (e.g., D. P. Stark et al.
  2009; I. Labbé et al. 2013) and obtain a conservative upper limit on the stellar
  mass of M* ≲ 107.5M⊙. This in turn gives us a black hole-to-host mass ratio of ≲0.14,
  which is significantly elevated from local expectations (J. E. Greene & L. C. Ho
  2005), but is not as extreme as some other LRDs reported in the literature (V. Kokorev
  et al. 2023; L. J. Furtak et al. 2024; R. Maiolino et al. 2024).\r\n\r\n6. Discussion
  and Summary\r\n6.1. Dense and Ionized Gas Surrounding the AGN\r\nOur NIRSpec G395M
  observations of GLIMPSE-17775 reveal a remarkably consistent picture: Across independent
  tracers, the line emission requires an environment of extremely high density and
  partial ionization.\r\n\r\nFirst, the broad permitted lines are universally better
  fit by exponential wings, a hallmark of electron scattering in gas with ne ≳ 108
  cm−3 and column densities approaching Ne ∼ 1024 cm−2. Such profiles are not reproduced
  by Doppler broadening alone and point to an ionized scattering medium enveloping
  the source. The fact that hydrogen and O i lines share consistent FWHM values further
  anchors this interpretation, as charge exchange tightly couples neutral oxygen to
  the ionization and kinematics of hydrogen.\r\n\r\nSecond, the detection of blueshifted
  Balmer absorption and a significant Balmer break both require high n = 2 populations
  and densities n ∼ 109 cm−3. Helium transitions provide a complementary view: The
  triplet lines λ7065 and λ10830 are both strongly enhanced relative to the singlet
  states and show deep blueshifted absorption, consistent with resonant scattering
  from the metastable 23S level. Their systematically narrower widths compared to
  hydrogen and oxygen suggest stratification, with helium arising from denser, more
  compact layers of the cocoon.\r\n\r\nThird, the O i λ8446–λ11290 pair confirms Lyβ
  fluorescence, requiring both a bright Lyβ radiation field and a dense reservoir
  of neutral gas. The absence of other permitted O i lines not fed by Lyβ strengthens
  this conclusion. Finally, the detection of 16 Fe ii lines, forming an incredibly
  rich iron forest in this LRD, matches predictions from Lyα fluorescence models,
  again demanding an intense radiation field and very high densities.\r\n\r\nBecause
  the BLR is unresolved in essentially all AGN, broad-line widths are traditionally
  interpreted as virial tracers of the black hole potential. In GLIMPSE-17775, the
  virial story alone is insufficient: The issue is not how broad the lines are, but
  how they broaden. The profiles exhibit extended, nearly linear wings in velocity
  space that are incompatible with a Gaussian. Instead, the lines are systematically
  and significantly better described by a model consisting of a narrow Gaussian core
  (virial motion) plus exponential wings, the hallmark of Thomson scattering in a
  dense ionized medium. Thus, the line shape encodes both gravitational kinematics
  and radiative-transfer physics in the surrounding cocoon. Our schematic (Figure
  9) summarizes this revised view: In LRDs, broad-line profiles are not set by dynamics
  alone, but by the scattering environment through which the photons escape.\r\n\r\nZoom
  InZoom OutReset image size\r\nFigure 9. Physical picture of the dense cocoon around
  GLIMPSE-17775. The inner region (1) corresponds to the dense, highly ionized broad-line
  region (BLR) where electron scattering (orange lines) produces exponential wings,
  H and O I share coupled kinematics (via charge exchange and Bowen fluorescence),
  and strong Balmer absorption/break signatures arise. The outer layer (2) represents
  an intermediate-density ionized medium, where He I triplet lines show resonant absorption
  from the metastable 23S level and Fe II emission is driven by Lyα fluorescence.
  Together, these zones form a stratified cocoon enshrouding the accreting black hole.
  The bottom-right panel shows stacked line profiles of H, O i, He i, and Fe ii.\r\n\r\nDownload
  figure:\r\n\r\nStandard imageHigh-resolution image\r\nSuch conditions—high optical
  depths, large column densities, and evidence for radiation-dominated gas—are precisely
  those expected in super-Eddington accretion flows. In this regime, radiation pressure
  inflates the inner accretion structure, driving powerful winds and forming the very
  dense, partially ionized envelope we infer here. The low X-ray luminosities and
  weak radio emission commonly observed in LRDs (e.g., T. T. Ananna et al. 2024; M.
  Yue et al. 2024; H. B. Akins et al. 2025a; M. Kokubo & Y. Harikane 2025) are consistent
  with this picture: The X-rays are likely absorbed or thermalized within the optically
  thick cocoon, while dust cannot survive in such an intense radiation field (e.g.,
  E. Lambrides et al. 2024; A. de Graaff et al. 2025b; K. Inayoshi & R. Maiolino 2025;
  R. P. Naidu et al. 2025; A. J. Taylor et al. 2025). Thus, the spectroscopic signatures
  observed in GLIMPSE-17775—exponential wings, absorption features, and fluorescence-driven
  metal lines—fit naturally into a scenario where super-Eddington accretion onto a
  low-mass black hole powers the luminous yet heavily reprocessed emission.\r\n\r\nTaken
  together, the exponential wings, Balmer and helium absorption, Bowen oxygen lines,
  and Lyα-pumped iron forest all converge on the same physical picture: GLIMPSE-17775
  is enshrouded in a dense, partially ionized cocoon of gas.\r\n\r\n6.2. Final Remarks\r\nUsing
  a combination of the intrinsically deepest NIRCam photometry and NIRSpec G395M observations
  of the lensed AS1063 field, we present a detailed investigation of a LRD at z =
  3.501. The spectrum of GLIMPSE-17775 is exceptionally rich, with more than 40 detected
  features, allowing us to probe the physical conditions of the gas with unprecedented
  detail. Multiple independent diagnostics converge on the presence of a dense, partially
  ionized cocoon heated by a powerful ionizing source.\r\n\r\nTypical of other LRDs,
  GLIMPSE-17775 is extremely compact in the rest-optical (reff < 300 pc) but exhibits
  more extended structure in the rest-UV (reff ∼ 1000 pc; Figure 1). It shows unmistakable
  AGN signatures through broad permitted lines (e.g., D. D. Kocevski et al. 2023;
  V. Kokorev et al. 2023; J. E. Greene et al. 2024; L. J. Furtak et al. 2024). The
  uniquely deep G395M spectrum further reveals clues to the physical origin of both
  its continuum and line-emission properties: We detect exponential broad-line wings,
  Balmer and helium absorption, Bowen-pumped O i, Fe ii emission produced by Lyα fluorescence,
  and evidence for rapid, potentially super-Eddington growth.\r\n\r\nTogether, these
  diagnostics provide one of the clearest cases yet for the BH* “dense cocoon” scenario
  (A. de Graaff et al. 2025b; K. Inayoshi & R. Maiolino 2025; R. P. Naidu et al. 2025;
  A. J. Taylor et al. 2025). The same ingredients we observe—exponential broad wings,
  Balmer-break absorption, He i absorption, and rich Fe ii and O i fluorescence—have
  been detected individually in several other bright LRDs with medium- and high-resolution
  spectroscopy (e.g., L. J. Furtak et al. 2024; I. Juodžbalis et al. 2024; I. Labbe
  et al. 2024; F. D’Eugenio et al. 2025; E. Lambrides et al. 2025; B. Wang et al.
  2025; A. Torralba et al. 2026), suggesting that dense, optically thick gas may be
  a common feature of the population rather than an anomaly. What distinguishes GLIMPSE-17775
  is that all signatures are captured simultaneously and at high S/N, allowing a self-consistent
  physical interpretation. If such cocoons are widespread, then super-Eddington accretion
  may be a typical pathway for black hole growth in LRDs, especially among the most
  luminous systems. Establishing how these signatures vary with luminosity and redshift
  will be essential for determining whether dense cocoons represent a dominant mode
  of early black hole assembly.\r\n\r\nThe convergence of five independent diagnostics—exponential
  scattering wings, Balmer-limit absorption, helium triplet physics, and two distinct
  fluorescence channels—leaves little doubt: GLIMPSE-17775 hosts a dense (n ∼ 108−9
  cm−3), optically thick (Ne ∼ 1024 cm−2) cocoon of partially ionized gas surrounding
  a super-Eddington accreting black hole. This represents some of the most direct
  and comprehensive spectroscopic evidence to date for the dense cocoon scenario in
  LRDs.\r\n\r\nAcknowledgments\r\nThe authors would like to thank Aaron Sigut and
  Anil Pradhan for their help with understanding iron emission in active galactic
  nuclei. The authors would like to acknowledge the National Institute of Standards
  and Technology (NIST) database of spectral lines (A. Kramida et al. 2024), which
  made identification of less-known emission features possible. V.K., J.C., S.F.,
  D.B., L.F., T.H., and J.M. acknowledge support from the University of Texas at Austin
  Cosmic Frontier Center. A.Z. acknowledges support by the Israel Science Foundation
  grant No. 864/23. This work is based on observations made with the NASA/ESA/CSA
  James Webb Space Telescope, obtained at the Space Telescope Science Institute, which
  is operated by the Association of Universities for Research in Astronomy, Incorporated,
  under NASA contract NAS5-03127. The JWST 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:10.17909/4byn-fe55
  and doi:10.17909/zq0c-8t87. These observations are associated with programs GO #3293
  and DDT #9223.\r\n\r\nFacilities: JWST - James Webb Space Telescope, HST - Hubble
  Space Telescope satellite.\r\n\r\nSoftware: EAZY (G. B. Brammer et al. 2008), grizli
  (G. Brammer 2023), msaexp (G. Brammer 2022), photutils (L. Bradley et al. 2020),
  pysersic (I. Pasha & T. B. Miller 2023), sep (K. Barbary 2016), SExtractor (E. Bertin
  & S. Arnouts1996)."
article_number: '153'
article_processing_charge: Yes
article_type: original
arxiv: 1
author:
- first_name: Vasily
  full_name: Kokorev, Vasily
  last_name: Kokorev
- first_name: John
  full_name: Chisholm, John
  last_name: Chisholm
- first_name: Rohan P.
  full_name: Naidu, Rohan P.
  last_name: Naidu
- first_name: Seiji
  full_name: Fujimoto, Seiji
  last_name: Fujimoto
- first_name: Hakim
  full_name: Atek, Hakim
  last_name: Atek
- first_name: Gabriel
  full_name: Brammer, Gabriel
  last_name: Brammer
- first_name: Steven L.
  full_name: Finkelstein, Steven L.
  last_name: Finkelstein
- first_name: Hollis B.
  full_name: Akins, Hollis B.
  last_name: Akins
- first_name: Danielle A.
  full_name: Berg, Danielle A.
  last_name: Berg
- first_name: Lukas J.
  full_name: Furtak, Lukas J.
  last_name: Furtak
- first_name: Qinyue
  full_name: Fei, Qinyue
  last_name: Fei
- first_name: Tiger Yu-Yang
  full_name: Hsiao, Tiger Yu-Yang
  last_name: Hsiao
- first_name: Ivo
  full_name: Labbé, Ivo
  last_name: Labbé
- first_name: Jorryt J
  full_name: Matthee, Jorryt J
  id: 7439a258-f3c0-11ec-9501-9df22fe06720
  last_name: Matthee
  orcid: 0000-0003-2871-127X
- first_name: Julian B.
  full_name: Muñoz, Julian B.
  last_name: Muñoz
- first_name: Pascal A.
  full_name: Oesch, Pascal A.
  last_name: Oesch
- first_name: Richard
  full_name: Pan, Richard
  last_name: Pan
- first_name: Pierluigi
  full_name: Rinaldi, Pierluigi
  last_name: Rinaldi
- first_name: Alberto
  full_name: Saldana-Lopez, Alberto
  last_name: Saldana-Lopez
- first_name: Daniel
  full_name: Schaerer, Daniel
  last_name: Schaerer
- first_name: Marta
  full_name: Volonteri, Marta
  last_name: Volonteri
- first_name: Adi
  full_name: Zitrin, Adi
  last_name: Zitrin
citation:
  ama: Kokorev V, Chisholm J, Naidu RP, et al. The Deepest GLIMPSE of a dense gas
    cocoon enshrouding a Little Red Dot. <i>The Astrophysical Journal</i>. 2026;1004(2).
    doi:<a href="https://doi.org/10.3847/1538-4357/ae4ed7">10.3847/1538-4357/ae4ed7</a>
  apa: Kokorev, V., Chisholm, J., Naidu, R. P., Fujimoto, S., Atek, H., Brammer, G.,
    … Zitrin, A. (2026). The Deepest GLIMPSE of a dense gas cocoon enshrouding a Little
    Red Dot. <i>The Astrophysical Journal</i>. IOP Publishing. <a href="https://doi.org/10.3847/1538-4357/ae4ed7">https://doi.org/10.3847/1538-4357/ae4ed7</a>
  chicago: Kokorev, Vasily, John Chisholm, Rohan P. Naidu, Seiji Fujimoto, Hakim Atek,
    Gabriel Brammer, Steven L. Finkelstein, et al. “The Deepest GLIMPSE of a Dense
    Gas Cocoon Enshrouding a Little Red Dot.” <i>The Astrophysical Journal</i>. IOP
    Publishing, 2026. <a href="https://doi.org/10.3847/1538-4357/ae4ed7">https://doi.org/10.3847/1538-4357/ae4ed7</a>.
  ieee: V. Kokorev <i>et al.</i>, “The Deepest GLIMPSE of a dense gas cocoon enshrouding
    a Little Red Dot,” <i>The Astrophysical Journal</i>, vol. 1004, no. 2. IOP Publishing,
    2026.
  ista: Kokorev V, Chisholm J, Naidu RP, Fujimoto S, Atek H, Brammer G, Finkelstein
    SL, Akins HB, Berg DA, Furtak LJ, Fei Q, Hsiao TY-Y, Labbé I, Matthee JJ, Muñoz
    JB, Oesch PA, Pan R, Rinaldi P, Saldana-Lopez A, Schaerer D, Volonteri M, Zitrin
    A. 2026. The Deepest GLIMPSE of a dense gas cocoon enshrouding a Little Red Dot.
    The Astrophysical Journal. 1004(2), 153.
  mla: Kokorev, Vasily, et al. “The Deepest GLIMPSE of a Dense Gas Cocoon Enshrouding
    a Little Red Dot.” <i>The Astrophysical Journal</i>, vol. 1004, no. 2, 153, IOP
    Publishing, 2026, doi:<a href="https://doi.org/10.3847/1538-4357/ae4ed7">10.3847/1538-4357/ae4ed7</a>.
  short: V. Kokorev, J. Chisholm, R.P. Naidu, S. Fujimoto, H. Atek, G. Brammer, S.L.
    Finkelstein, H.B. Akins, D.A. Berg, L.J. Furtak, Q. Fei, T.Y.-Y. Hsiao, I. Labbé,
    J.J. Matthee, J.B. Muñoz, P.A. Oesch, R. Pan, P. Rinaldi, A. Saldana-Lopez, D.
    Schaerer, M. Volonteri, A. Zitrin, The Astrophysical Journal 1004 (2026).
das_tickbox: '1'
dataavailabilitystatement: "The specific observations analyzed can be accessed via
  doi:10.17909/4byn-fe55 and doi:10.17909/zq0c-8t87. These observations are associated
  with programs GO #3293 and DDT #9223.\r\n\r\nFacilities: JWST - James Webb Space
  Telescope, HST - Hubble Space Telescope satellite.\r\n\r\nSoftware: EAZY (G. B.
  Brammer et al. 2008), grizli (G. Brammer 2023), msaexp (G. Brammer 2022), photutils
  (L. Bradley et al. 2020), pysersic (I. Pasha & T. B. Miller 2023), sep (K. Barbary
  2016), SExtractor (E. Bertin & S. Arnouts1996)."
date_created: 2026-07-13T09:48:38Z
date_published: 2026-06-10T00:00:00Z
date_updated: 2026-07-13T13:25:09Z
day: '10'
ddc:
- '520'
department:
- _id: JoMa
doi: 10.3847/1538-4357/ae4ed7
external_id:
  arxiv:
  - '2511.07515'
file:
- access_level: open_access
  checksum: 464e60013bf14d087eb0968e9c81a269
  content_type: application/pdf
  creator: dernst
  date_created: 2026-07-13T13:23:11Z
  date_updated: 2026-07-13T13:23:11Z
  file_id: '22313'
  file_name: 2026_AstrophysicalJour_Kokorev.pdf
  file_size: 2435643
  relation: main_file
  success: 1
file_date_updated: 2026-07-13T13:23:11Z
has_accepted_license: '1'
intvolume: '      1004'
issue: '2'
language:
- iso: eng
month: '06'
oa: 1
oa_version: Published Version
publication: The Astrophysical Journal
publication_identifier:
  eissn:
  - 1538-4357
  issn:
  - 0004-637X
publication_status: published
publisher: IOP Publishing
quality_controlled: '1'
researchdata_availability: no
scopus_import: '1'
status: public
supplementarymaterial: no
title: The Deepest GLIMPSE of a dense gas cocoon enshrouding a Little Red Dot
tmp:
  image: /images/cc_by.png
  legal_code_url: https://creativecommons.org/licenses/by/4.0/legalcode
  name: Creative Commons Attribution 4.0 International Public License (CC-BY 4.0)
  short: CC BY (4.0)
type: journal_article
user_id: 2DF688A6-F248-11E8-B48F-1D18A9856A87
volume: 1004
year: '2026'
...