Limitations on Morphological Fitting for JWST "Little Red Dots"

Kelly E. Whalen, Kimberly A. Weaver, Ryan C. Hickox, Erini Lambrides

First listed 2025-09-25 | Last updated 2025-09-25

Abstract

Early results from JWST uncover a peculiar class of objects referred to as ``little red dots'' (LRDs). The extremely compact morphology of LRDs is often invoked to point towards an AGN-dominated picture in the context of their conflicting multiwavelength properties. In this work, we assess the capability of pysersic and GALFIT -- commonly used tools in LRD morphological studies -- to recover input parameters for a simulated suite of LRD-like objects in the F444W band. We find that: 1) these tools have difficulty recovering input parameters for simulated images with SNR $\lesssim 25$; 2) estimated PSF fraction could be a more robust physically-motivated description of LRD compactness; and 3) almost all permutations of modeled LRDs with SNR $\lesssim 50$ cannot be differentiated from a point source, regardless of intrinsic extent. This has serious implications on how we interpret morphological results for increasingly large photometric samples of LRDs, especially at extremely high-$z$ or in relatively shallow fields. We present results of Sersic and two-component fitting to a sample of observed LRDs to compare with our mock sample fitting. We find that $\sim85\%$ of observed LRDs are PSF-dominated, consistent with the AGN-dominated interpretation. The remaining $\sim15\%$ have low estimated PSF fractions (two-component fit) and sizes $\gtrsim 150$ pc (Sersic). This morphological diversity of LRDs suggests that that the population likely is not homogeneous. It possibly has a primary subset of sources consistent with the AGN-dominated hypothesis, and a secondary population of sources more consistent with arising perhaps from extremely compact starbursts.

Short digest

Tests on mock LRD-like cutouts in NIRCam/F444W show that common morphology codes (pysersic, GALFIT) struggle: below SNR ≲ 25, Sérsic parameters are poorly recovered, and even up to SNR ≲ 50 most models are indistinguishable from a point source. Two-component fits can still robustly estimate the unresolved PSF fraction, but they do not reliably recover the host Sérsic size/shape, especially for small effective radii or steep profiles. Applying the same fitting to real LRDs, ~85% appear PSF-dominated, while the remaining ~15% have low PSF fractions and sizes ≳150 pc, indicating a non-homogeneous population. The takeaway for large, shallow high-z samples: lean on PSF fraction for compactness and treat “resolved” sizes with caution.

Key figures to inspect

• Figure 1: Inspect the input–output comparisons and residuals to see where single-component Sérsic fits begin to converge only at higher SNR, and how uncertainties explode below SNR ≲ 25.
• Figure 2: Compare recovery performance across Sérsic n and effective radius; note the systematic failures for steeper (higher-n) and very compact profiles that bias size inferences.
• Figure 3: From the two-component decompositions, focus on how PSF fraction is well constrained over a wide SNR range while the host Sérsic parameters (re, n) remain unconstrained or biased—especially the overestimation trend for small re at low SNR.
• Figure 4: Read the reduced-chi2 landscape versus SNR to verify that all simulated models with SNR ≲ 50 are PSF-consistent, and only low-n/large-re profiles at higher SNR are securely extended.