$\texttt{geko}$: A tool for modelling galaxy kinematics and morphology in JWST/NIRCam slitless spectroscopic observations

TL;DR

JWST/NIRCam slitless spectroscopy offers high-resolution kinematic information but is hampered by morphology–kinematics degeneracies in 2D grism maps. geko provides a forward-modeling, Bayesian framework that jointly infers emission-line morphology and gas kinematics by convolving a Sérsic morphology with an arctangent rotation curve through a full instrument model and using gradient-based sampling on GPUs. The paper demonstrates extensive mock-data validation across position angles, S/N, and morphologies, and applies geko to real FRESCO Hα emitters at z≈4–6, recovering both rotation- and dispersion-dominated systems and deriving v_rot(r_e), v/σ_0, v_circ, and M_dyn. This approach enables robust, scalable dynamical studies of galaxies in the early universe and provides a public tool for the community to perform statistical analyses of galaxy dynamics with JWST data.

Abstract

Wide-field slitless spectroscopy (WFSS) is a powerful tool for studying large samples of galaxies across cosmic times. With the arrival of JWST, and its NIRCAM grism mode, slitless spectroscopy can reach a medium spectral resolution of $(R\sim 1600)$, allowing it to spatially resolve the ionised-gas kinematics out to $z\sim 9$. However, the kinematic information is convolved with morphology along the dispersion axis, a degeneracy that must be modelled to recover intrinsic properties. We present the Grism Emission-line Kinematics tOol ($\texttt{geko}$), a Python package that forward-models NIRCam grism observations and infers emission-line morphologies and kinematics within a Bayesian framework. $\texttt{geko}$ combines Sérsic surface-brightness models with arctangent rotation curves, includes full point-spread function (PSF) and line-spread function (LSF) convolution, and leverages gradient-based sampling via $\texttt{jax}$/$\texttt{numpyro}$ for efficient inference. It recovers parameters such as effective radius, velocity dispersion, rotational velocity, rotational support, and dynamical mass, with typical run times of $\sim$20 minutes per galaxy on GPUs. We validate performance using extensive mock data spanning position angle, S/N, and morphology, quantifying where degeneracies limit recovery. Finally, we demonstrate applications to real FRESCO H$α$ emitters at $z\approx 4-6$, recovering both rotation- and dispersion-dominated systems. $\texttt{geko}$ opens the way to statistical studies of galaxy dynamics in the early Universe and is publicly available at https://github.com/angelicalola-danhaive/geko.

Figures (16)

• Figure 1: Flowchart of the geko inference process as described in Sec. \ref{['sec:geko-ingredients']}. The code takes as input continuum subtracted grism data, as well priors for the galaxy's morphology, and then outputs spatially resolved emission line maps and kinematic maps. In the inference process, a mock grism spectrum is constructed based on the current sampled parameters, and is compared to the observed spectrum to evaluate the likelihood.
• Figure 2: Schematic illustration of the position angles ($\rm PA$) as defined within the geko framework. The PA is measured as the angle between the negative x-axis and the galaxy's major axis, as shown in the $\rm PA = 45^{\circ}$ case.
• Figure 3: Test results for the recovery of kinematics, using noisy mock grism data, for a range of position angles, $\rm PA = 0^{\circ},15^{\circ},45^{\circ},75^{\circ}$, asymptotic velocities $V_a = 5,100,200,300$ km/s, and velocity dispersions $\sigma_0 = 5,25,70,100,200,500$ km/s. For clarity, we do not show the $\rm PA = 90^{\circ}$ case, but we show the average results on Fig. \ref{['fig:real_test_param_mean_curves']}.
• Figure 4: Mean inferred values for $\sigma_0$ and $v_{\rm re}$, for the corresponding input true values, for each position angle $\rm PA = 0-90^{\circ}$. The means are taken over all of the $V_a$-$\sigma_0$ combinations tested and shown on Fig. \ref{['fig:real_test_param']}. When the galaxy is parallel to the dispersion axis, $\rm PA = 90^{\circ}$, the kinematics cannot be well constrained due to the strong degeneracies.
• Figure 5: Test results for the recovery of kinematics, using noisy mock grism data, for a range of turn-around around radii $r_{\rm t} = 0.25, 0.5, 1,1.25,1.5$ px (which sets the effective radii $r_{\rm e} \approx 1, 2, 4,5,6 \rm px \approx 0.06,0.13,0.25, 0.38"$ following Eq. \ref{['eq:re-rt']}) and Sérsic indices $n=0.5,1,2,4,6,8$. The position angle of the mock galaxies is set to $\rm PA = 45^{\circ}$, and the kinematics to $V_a =200$ km/s and $\sigma_0 = 100$ km/s. The input kinematic values are fixed, but we spread them out on the plot artificially, for clarity purposes.