Little Red Dots: One Photometric Tag Concealing Diverse Spectroscopic Flavors of Massive Star Formation and Black Hole Activity

Abstract

We compile JWST/NIRSpec prism and MIRI data for 249 Little Red Dots (LRDs) at $2.3<z<9.3$, forming a representative spectroscopic subset of NIRCam-selected LRDs. We derive a median stacked spectrum covering rest-frame 0.09-1.2 $μ$m, with MIRI photometry extending the spectral energy distribution to 4 $μ$m. Four additional stacks for subsamples defined by optical-to-UV luminosity ratios show that LRDs form a heterogeneous population spanning diverse continuum slopes and line properties. Assuming LRDs host super-massive black holes (BHs) surrounded by dense gas clouds, and stars accompany this core, we infer masses of $M_{BH}\sim10^{6.0-6.5}$~M$_\odot$ and $M_\bigstar\sim10^{8.3}$ M$_\odot$, corresponding to BH-to-stellar mass ratios of 1-2%. The stacks show ubiquitous UV and optical FeII emission, indicating a direct view of the broad-line region and high (but sub-Eddington) accretion ($λ_{Edd}=0.6\pm0.2$). We find a significant stellar contribution in the far-UV, reaching $\sim80$% in the bluest systems. Possible Wolf-Rayet features (HeII$λ$4687, nitrogen lines) are identified, tracing a young (3-7 Myr) compact starburst event. We also detect strong Balmer breaks and atypical Balmer, Paschen, [OIII], and optical and near-infrared HeI line ratios, and an absorption at $\sim4550$ Angstrom (probably linked to FeII), all consistent with radiative-transfer effects in high-density gas with warm temperatures (4000-7000 K). We find a diversity of LRD flavors modulated by the luminosity ratio between between a short ($\lesssim20$ Myr) and intense phase of BH activity, the most extreme stage lasting $\sim3-7$ Myr, characterized by near-Eddington-limit radiation, and a nuclear and compact starburst dominated by massive stars (even super-massive, $\mathrm{M}_\mathrm{SMS}\sim10^{5}$ M$_\odot$), all embedded in dense gas with modest dust content producing a variety of optical depths.

Figures (19)

• Figure 1: Photometric selection of LRDs, and the spectroscopic sample used in this work. Top left: Color-color diagram introduced by 2024arXiv241201887B, with the selection boundary represented by the dotted lines. The LRD selection criterion used in 2024ApJ...968....4P based on F277W-F444W colors is shown with red symbols, LRDs identified in 2025ApJ...986..126K and 2024ApJ...968...38K are shown as orange and lime symbols, respectively, while blue symbols depict the LRDs identified solely by 2024arXiv241201887B. Our spectroscopic sample (with NIRSpec prism observations) is depicted with yellow symbols. Gray symbols show the general NIRCam sample (drawn from the ASTRODEEP–JWST catalogs presented in 2024AA...691A.240M). Top right: Color-magnitude plot with LRD selection criterion used in 2024ApJ...968....4P based on F277W-F444W colors (marked with a dotted line). We use the same symbols described for the previous panel. Middle left: F444W magnitude distribution of the different samples of LRDs. Middle right: F115W-F200W color distribution of LRDs. Bottom left: F200W-F444W color distribution of LRDs. Bottom right: F277W-F444W color distribution of LRDs. In all the panels with histograms, the median and quartiles for the photometric (black) and spectroscopic (gold) samples are shown.
• Figure 2: Left: Redshift distribution for the spectroscopic sample of LRDs characterized in this paper (gold histogram) and the general sample of photometrically selected LRDs presented in Figure \ref{['fig:selection']} (gray histogram, scaled down by a factor of $\sim4$ to make the comparison easier). Right: Same but for the different LRD subtypes defined in Section \ref{['sec:classes']}, based on optical-to-UV luminosity ratios, running from the reddest LRDs (top) to the bluest (bottom). The photometric sample plotted again in gray, and this time using normalized histograms. In all panels, the median and quartiles of the distributions are shown with dots and segments, the gray histograms refer to the photometric sample and the colored ones to the spectroscopic subsamples.
• Figure 3: The top panel shows the stacked spectrum of the 249 LRDs with NIRSpec prism spectroscopy selected in this paper (black line). The MIRI stacks in the same rest-frame wavelength range are shown (including a horizontal error bar enclosing the wavelength range for 68% of the MIRI points entering the stack), to demonstrate that they agree with the spectroscopic stack (see next figure for redder MIRI stacks). The spectrum is compared with QSO models from 2021MNRAS.508..737T for a pure AGN ($q_\mathrm{gal}=0$, green line) and 90% contribution from a host ($q_\mathrm{gal}=90$%, cyan line). In orange and red, we show the stacked spectra for LAEs and non-LAEs presented in 2024ApJ...976..193R, after applying $A^V=0.8$ and 5.1 mag attenuation, respectively, following a 2000ApJ...533..682C law. The QSO templates are normalized to the emission around the MgII$\lambda2798$ line, the LAE template is normalized to the emission around 0.28 $\mu$m, and the non-LAE template and median LRD SED are normalized at 0.55 $\mu$m. The panels below the main one show zoomed-in versions (in linear scale for the flux axis) in the UV ( middle left panel), optical around [OIII]+H$\beta$ and H$\alpha$ regions ( middle right and bottom left, respectively), and the near-IR ( bottom right). In these panels, the comparison spectra are normalized to the median in the spectral window covered by the plot. The Gaussians at the bottom of the panels show the spectral resolution at different wavelengths.
• Figure 4: Position of the LRD median (black hexagon) in a UV line diagnostic plot comparing C III] EW $vs.$ C III]/He II intensity ratio. We also show the measurements for different subtypes of LRDs (colored hexagons) based on optical-to-UV luminosity ratios, $\mathrm{L_{5100}/L_{2500}}$, as defined in Section \ref{['sec:classes']}. We compare with a compilation of measurements for $z>2$ galaxies, extracted from 2025arXiv251216365A. In gray, we depict individual sources at $5<z<7$ (from that same paper -dots-) and at $2<z<4$ (from 2022AA...659A..16L -diamonds-). In lime, magenta and deep pink, stacks for strong, weak and non-detection C III] emitters are shown with crosses 2025arXiv251216365A. Models from 2022MNRAS.513.5134N of AGN-dominated and star formation-dominated sources are shown with orange and blue symbols, also depicting the separation lines between them, as well as the hybrid region (dashed).
• Figure 5: Left: Plot showing the estimation of the continuum emission around the WR blue bump (marked with a blue shade) for the stack of all LRDs, scaled in rest-frame flux to match the optical luminosity density at 5100 Å of the full spectroscopic sample for $z=7$ (approximate median redshift of the photometric sample). The estimated continuum is shown with a red line. Several relevant spectral features are marked, including an absorption at 4550 Å (see Section \ref{['sec:absorption']} for details). Right: Spectral region around the WR blue bump. The stack of spectra for all LRDs (black line with errors represented by the gray shading) is fitted with Gaussians placed at the wavelengths of emission lines typical of this kind of massive stars and from AGN (marked with black vertical lines in the plot). The He II$\lambda4687$ line is allowed to have a narrow and a wide component (the latter defined as $FWHM>2000$ km s$^{-1}$), and the intrinsic widths of the remaining lines are free. The continuum is measured as shown in the left panel. In cyan, we also depict the wavelengths for the Fe II line forest, and in pink the [Fe II] lines. The plot gives the calculated total mass of a starburst explaining the He II emission in terms of WR stars (WN subtype).