Testing the inference of kinematics from mock JWST NIRSpec/MSA observations of TNG50 galaxies at $z\sim2-6$

Technical Summary: Testing the Inference of Kinematics from Mock JWST NIRSpec/MSA Observations of TNG50 Galaxies at $z \sim 2-6$

Problem Statement
Understanding the dynamical evolution of galaxies requires measuring the rotational velocities ($v$) and velocity dispersions ($\sigma$) of ionised gas across cosmic time. While hydrodynamical simulations like IllustrisTNG provide 6-dimensional spatial and kinematic data, they lack explicit radiation transfer solutions for specific observational setups. Conversely, observational studies using instruments like JWST's NIRSpec Micro-Shutter Array (MSA) rely on forward modelling to infer kinematics from 2D spectra. These models typically assume galaxies are axisymmetric, thin, rotating discs with Sérsic intensity profiles and arctangent velocity curves. However, high-redshift galaxies ($z > 2$) often exhibit complex morphologies (spheroidal, clumpy) and non-circular motions (inflows/outflows) that may violate these assumptions. There is a critical lack of statistical validation regarding whether simple parametric models can accurately recover intrinsic kinematics for typical high-redshift galaxies, particularly given the limited spatial resolution and slit-based nature of MSA observations compared to Integral Field Units (IFUs).

Methodology
The authors utilize the high-resolution TNG50 cosmological simulation to generate an end-to-end mock observation pipeline for over 4,000 star-forming galaxies ($M_* = 10^8 - 10^{11.5} M_\odot$) at redshifts $z = 2, 3, 4, 6$.

1. Intrinsic Measurements: The authors define "intrinsic" kinematic quantities ($v_{max}$ and $\sigma_e$) within the effective radius ($r_e$) of the H$\alpha$-emitting gas. They explicitly test the sensitivity of these intrinsic values to measurement choices, including aperture size, spatial binning (0.5 vs. 5 ckpc), and weighting schemes (unweighted vs. Star Formation Rate weighted).
2. Mock Observations:
   • Imaging: Mock NIRCam images are generated by projecting star-forming gas cells, converting Star Formation Rates (SFR) to H$\alpha$ luminosity via the Kennicutt (1998) relation, and convolving with NIRCam PSFs. Noise is added to mimic deep JADES-like exposures.
   • Spectroscopy: Mock NIRSpec/MSA 2D spectra are created by projecting gas cells along a random axis, assigning line-of-sight velocities, and convolving with the NIRSpec Line Spread Function (LSF) and PSF. Gaussian noise is applied to achieve a signal-to-noise ratio (S/N) comparable to the deepest existing JWST data (S/N $\sim$ 50).
3. Inference Pipeline:
   • Morphology: Mock NIRCam images are fitted with pysersic to derive morphological priors (effective radius $r_e$, Sérsic index $n_s$, axis ratio $q$, position angle PA).
   • Kinematics: Mock NIRSpec spectra are fitted using msafit, assuming a thin, rotating disc model with an arctangent velocity profile and constant velocity dispersion. The morphological parameters derived from the imaging are used as strong priors. The fitting employs a weighted $\chi^2$ minimization, including a non-parametric row-wise flux scaling to account for clumpy light distributions.

Key Contributions and Results
The study compares the best-fit kinematic parameters derived from the mock spectra against the intrinsic simulation values to quantify observational biases.

• Recovery Accuracy: For aligned and resolved sources at $z=3$, the inferred rotational velocities ($v$) and velocity dispersions ($\sigma$) generally agree with intrinsic measurements within a factor of $\sim 2$ and $\sim 1.5$, respectively. The scatter in $v$ is larger than in $\sigma$.
• Systematic Biases:
  • Morphology Dependence: The recovery is robust for smooth, disc-like systems. However, for intrinsically elongated or spheroidal systems (large intrinsic axis ratios $C/A$), $v$ tends to be overestimated and $\sigma$ underestimated by more than a factor of 2. This occurs because the thin-disc assumption forces the model to interpret non-rotational broadening as inclination effects.
  • Redshift Evolution: The scatter in recovery accuracy increases at higher redshifts ($z > 3$) as galaxies deviate more strongly from the thin-disc assumption.
  • Individual vs. Population: While individual galaxy kinematics ($v/\sigma$) can be misclassified by a factor of 2–3 (e.g., a dispersion-supported system appearing rotationally supported), the population trends are robustly recovered. The weak redshift evolution of $\sigma$ and $v/\sigma$, as well as the dependence of $\sigma$ on global SFR, are preserved in the mock-inferred data.
• Scaling Relations: The inferred $\sigma$-SFR relation aligns well with the "mass-transport + feedback" theoretical models (Krumholz et al. 2018), suggesting that despite individual uncertainties, kinematic modelling can reliably constrain the physical drivers of turbulence (feedback vs. gravity) in galaxy populations.
• Failure Modes: Approximately 5–10% of sources fail to fit the thin-disc model, often due to irregular morphologies, high stellar mass systems with steep central velocity gradients not captured by arctangent profiles, or sources where the intrinsic $\sigma$ is below the model's lower bound ($10$ km s$^{-1}$).

Significance and Claims
The paper claims to provide the first end-to-end framework for comparing NIRSpec MSA observations to cosmological simulations to quantify biases in measuring ionised gas kinematics. The authors conclude that while simple thin-disc models are insufficient for determining the precise "diskiness" ($v/\sigma$) of individual high-redshift galaxies due to large uncertainties, they are effective for recovering statistical population trends.

The work highlights that observational inferences of $v/\sigma$ for individual systems at $z \ge 2$ should be treated with caution, as they may be biased by the assumption of thin-disc geometry and the presence of non-circular motions. The authors argue that to improve constraints on the dynamical state of early galaxies, future work must move beyond 2D slit spectroscopy toward 3D IFU observations or slit-stepping strategies, and develop dedicated kinematic models that account for the complex 3D morphologies and clumpiness inherent to high-redshift galaxies. The study emphasizes that current limitations are not merely instrumental but stem from the mismatch between simple parametric models and the intrinsic complexity of galaxy formation physics.