ATLAS. III. Dust Around Little Red Dots: Hydrogen Line Ratios beyond Dust-free Non-Case B Models

This paper, the third in the Archival and Theoretical study of LRDs with AGN comparison across Surveys (ATLAS) series, investigates broad hydrogen line ratios in little red dots (LRDs) to understand the physical conditions in their central engines, particularly the roles of dense optically thick gas versus dust attenuation.

Sample and Data: The authors assemble a sample of 20 LRDs, comprising 15 low-redshift sources (z = 0.1–0.9) from the literature (three from X. Lin et al. 2026a and twelve from X. Lin et al. 2026b) and five high-redshift sources (z = 2.3–7.0) from the DAWN JWST Archive (DJA) with JWST/NIRSpec medium/high-resolution grating spectra. All sources have broad Hβ detected at S/N ≥ 5. An additional supplementary source, RUBIES-UDS-40579, is included for Paschen-line analysis but lacks Hβ coverage. The sample spans log(L5100/erg s−1) = 42.6–44.5.

Key Findings:

1. Broad Balmer decrements are very high: The broad Hα/Hβ ratios range from 6 to 30, well above the Case B recombination value of 2.87. In contrast, narrow Hα/Hβ ratios are broadly consistent with Case B values. This broad-narrow contrast indicates that the physical conditions and/or attenuation affecting the broad-line-emitting gas differ from those affecting the narrow-line-emitting gas.

2. Paschen line ratios support non-Case B processes: The two local LRDs with reliable broad Paα and Paβ measurements (J1022+0841 and J1047+0739) have Paα/Paβ ratios (1.5) below the adopted Case B value (= 2.0). Since a foreground attenuation screen would increase this ratio, the Paschen measurements support non-Case-B hydrogen-line formation processes in an optically thick, high-density gas.

3. Cloudy model constraints: Using plane-parallel Cloudy photoionization models with total line emission from both sides of the slab, the authors find that dust-free models can reproduce elevated broad Balmer decrements through radiative-transfer and collisional effects, with the largest Hα/Hβ ratios reaching 13 at log(nH/cm−3) 10 and log(Φ/cm−2 s−1) 18. Broad Hα/Hβ ratios above this model value would require additional differential attenuation or physical conditions not included in the grid.

4. Dust extinction required for some objects: Among the three LRDs with joint Balmer–Paschen constraints, J1047+0739 can be approximately reproduced by dust-free dense-gas models, whereas J1022+0841 and the Rosetta Stone require additional attenuation with E(B − V) ≳ 0.2–1.0, even after accounting for non-Case B effects. Since the narrow Hα/Hβ ratios do not indicate such large dust extinction, these results demonstrate that the obscuring dust is spatially associated with the broad-line region.

5. Prevalence of extreme decrements: Two high-redshift and eight low-redshift LRDs in the sample have broad Hα/Hβ ratios above 13, exceeding the maximum of the dust-free Cloudy grid. If interpreted as dust attenuation, this implies that dust preferentially attenuates the broad-line emission along some lines of sight with E(B − V) ≳ 0.2–1.0, suggesting dust obscuration might be common among LRDs, occurring in at least about half of the population.

6. Mid-infrared consistency: The warm/hot dust components inferred from WISE observations of local LRDs and the stacked rest-frame MIR excess (λrest = 3 μm) reported for high-redshift LRDs are consistent with hot/warm dust (300–1000 K) near central engines of LRDs. The authors note that MIR-emitting dust need not be identical to the material responsible for broad-line attenuation.

Interpretation: The authors propose that LRDs contain dense, optically thick ionized gas near the central engine that produces non-Case-B broad hydrogen-line ratios, surrounded by or viewed through a low-column-density counterpart of the classical AGN torus, which they refer to as a proto-torus. This picture reconciles the weak hot-dust emission observed in LRDs. They also discuss possible Type-2-like LRDs (narrow-line objects without detectable broad Hα) as potential obscured counterparts, and present an illustrative multiphase geometry with ionized gas (104 K), atomic gas (5000 K), and molecular/dusty gas (300–2000 K) at increasing distances from the central engine.

Improvements to AI Systems:

1. Physics-aware spectral line ratio modeling: Enhance AI systems used for analyzing AGN spectra by integrating non-Case-B hydrogen recombination physics (e.g., dense-gas radiative transfer, collisional excitation) into their inference pipelines. The improved system can automatically distinguish between dust attenuation and intrinsic line-formation effects when interpreting broad Balmer/Paschen decrements, reducing false conclusions about extinction in high-redshift galaxies.

2. Multi-wavelength dust decomposition: Build AI models that jointly fit broad-line ratios (Hα/Hβ, Paα/Paβ) with mid-infrared photometry (WISE, MIR excess) to separate foreground dust attenuation from hot/warm dust emission near the central engine. The improved system can predict whether obscuring dust is spatially coincident with the broad-line region or distributed in a larger torus-like structure, as demonstrated by the paper's proto-torus scenario.

3. Automated classification of LRD subtypes: Develop a classifier that uses observed broad-line ratios, narrow-line ratios, and MIR colors to categorize LRDs into dust-free dense-gas dominated, dust-attenuated, or Type-2-like (obscured, no broad lines) populations. The improved system can flag objects requiring additional extinction corrections and estimate E(B−V) ranges (0.2–1.0) directly from line ratios, even when only Balmer lines are available.

4. Generative model for synthetic LRD spectra: Train a conditional generative model (e.g., diffusion or normalizing flow) on the Cloudy photoionization grid (log nH = 8–12, log Φ = 16–20) to produce realistic broad-line profiles and ratios. The improved system can simulate mock LRD observations for survey design, test detection limits, and provide priors for Bayesian spectral fitting of JWST/NIRSpec data.

5. Uncertainty-aware dust correction module: Implement a Bayesian module that, given broad Hα/Hβ and Paα/Paβ measurements, outputs a posterior distribution over E(B−V) and intrinsic line ratios, explicitly accounting for the degeneracy between non-Case-B physics and dust. The improved system can quantify whether an object's extreme decrement (Hα/Hβ > 13) requires dust or is consistent with the dust-free model grid, aiding in population-level statistical studies.

6. Cross-survey harmonization tool: Create an AI pipeline that standardizes line-ratio measurements across heterogeneous datasets (literature samples, DAWN JWST Archive, future surveys) by correcting for resolution, S/N, and redshift-dependent effects. The improved system can automatically assemble larger LRD samples with consistent physical parameter estimates, enabling robust demographic studies of dust prevalence and evolution with cosmic time.

We investigate broad hydrogen line ratios in little red dots (LRDs) using five high-redshift (z>2) sources from JWST/NIRSpec medium/high-resolution spectra in the DAWN JWST Archive and fifteen low-redshift sources (z=0.1 -- 0.9) from the literature, all with broad H beta detected at >5 sigma. After carefully measuring the broad-line fluxes while accounting for absorption features and neighboring emission lines, we find that the broad H alpha /H beta ratios are very high, ranging from 6 to 30, well above the Case B recombination value. Using plane-parallel Cloudy photoionization models with the total line emission from both sides of the slab, we investigate the physical origin of the broad hydrogen line ratios beyond Case B, jointly modeling the Balmer and Paschen line ratios for the subset of one high- z and two low- z LRDs with detected broad Paschen lines. We find that one low- z LRD is reproduced by a high gas number density ((n H/ cm-3) about10 --11) on the broad-line H alpha /H beta --Pa beta /Pa gamma plane, whereas the remaining two LRDs additionally require substantial dust extinction of E(B-V) 0.2 -- 1.0, even after accounting for non-Case B effects. Since the narrow H alpha /H beta ratios do not indicate such large dust extinction, these results demonstrate that the obscuring dust is spatially associated with the broad-line region. Even without Paschen-line measurements, two and eight LRDs in the high- and low-redshift samples, respectively, exhibit H alpha /H beta>13, which cannot be reproduced by the non-Case B models, suggesting that dust obscuration might be common among LRDs, occurring in at least about half of the population. Such dust may represent a lower-column-density counterpart of the dusty torus in AGNs, reconciling the weak hot-dust emission.