The dark side of early galaxies: $\texttt{geko}$ uncovers dark-matter fractions at $z\sim4-6$

TL;DR

This study uses JWST/NIRCam grism data to model H$\alpha$ kinematics in 163 galaxies at $z\sim4-6$ with the geko forward-modeler, deriving dynamical masses and baryonic contents to estimate $f_{\rm gas}$ and $f_{\rm DM}$ within the central $r_{\mathrm{e}}$. The authors find high median gas and DM fractions ($\langle f_{\rm gas}\rangle\approx0.77$, $\langle f_{\rm DM}\rangle\approx0.73$), with ~67% of systems DM-dominated in the inner $\sim0.5-1$ kpc, and a strong negative correlation between $f_{\rm DM}$ and baryonic surface density $\Sigma_{\rm bar}$. The dynamical masses produce a broad Tully–Fisher relation with large intrinsic scatter and a substantial zero-point offset to local and cosmic-noon relations, indicating the transition to stable disks is underway but not yet established at $z\sim5$. By comparing with empirical models and simulations, the work suggests a mix of cuspy and cored inner DM profiles driven by baryonic physics, bursts of star formation, and potential adiabatic contraction, with implications for the growth of overmassive black holes in the early Universe. Overall, these results place high-redshift DM content in a broader evolutionary context, linking progenitors of present-day baryon-dominated systems and informing models of galaxy–halo co-evolution.

Abstract

JWST/NIRCam slitless spectroscopy enables dynamical mass measurements for typical star-forming galaxies only a billion years after the Big Bang. We model the H$α$ morpho-kinematics of 163 galaxies at redshift $z\approx4$-6 from FRESCO and CONGRESS (with JADES imaging), using the $\texttt{geko}$ code, and infer rotational velocities and dispersions within $r_{\rm e}$. Our sample spans $\log M_{\star}\approx7$-10 and $\log M_{\rm dyn}\approx9$-11. Gas masses are estimated via scaling relations, yielding baryonic masses and dark-matter (DM) fractions $f_{\rm DM}(r<r_{\rm e})$ within the H$α$ half-light radius. We find high median fractions of $\langle f_{\rm gas}\rangle=0.77$ and $\langle f_{\rm DM}\rangle=0.73$, where $f_{\rm gas}$ is measured with respect to the baryonic mass and $f_{\rm DM}$ with respect to the DM+baryonic mass. About two-thirds of systems are DM-dominated within $r_{\rm e}\sim0.5-1$ kpc. Both $f_{\rm gas}$ and $f_{\rm DM}$ decrease with stellar mass, consistent with simulations. The stellar Tully-Fisher relation shows a tentative offset to higher $v_{\rm circ}$ at fixed $M_{\star}$ and substantial intrinsic scatter, suggesting that the relation is only beginning to emerge at $z\sim5$. We measure a negative correlation between $f_{\rm DM}$ and baryonic surface density $Σ_{\rm bar}$, weaker but broadly consistent with trends at cosmic noon and at $z\sim0$. Qualitatively comparing with modified NFW profiles coupled to an empirical stellar-to-halo mass relation suggests that the lowest $f_{\rm DM}$ ($\lesssim0.4$) require cored inner DM profiles, while the highest fractions favour cuspier profiles, potentially reflecting adiabatic contraction. Overall, the elevated $f_{\rm gas}$ and $f_{\rm DM}$ at $z\gtrsim4$ are compatible with progenitors of baryon-dominated systems at $z\sim2$ and naturally anticipate overmassive black holes at fixed $M_{\star}$.

Figures (8)

• Figure 1: Comparison of the dynamical masses ($M_{\mathrm{dyn}}$) inferred from our modelling of grism data and the stellar masses ($M_{\star}$) inferred from SED fitting. We colour-code our galaxies based on their offset from the main sequence ($\Delta\rm MS$) as defined by Simmonds:2025aa. The majority of our systems lie below the one-to-one relation (dashed line), consistent with a significant contribution to the dynamical mass from gas and/or DM. Six galaxies lie on the relation, highlighting discrepancies in the different mass estimates. We compare our values to ionised gas measurements from de-Graaff:2024ab and Saldana-Lopez:2025aa at similar redshifts.
• Figure 2: The stellar Tully-Fisher plane for our sample, where the best-fit linear relation for galaxies with $\log(M_{\star}\xspace ~\rm [\rm M_{\odot}\xspace])\xspace>8$ is shown in solid purple (dotted red for the full sample). The circular velocity ($v_{\mathrm{circ}}\xspace$) is evaluated at $r_{\mathrm{e}}\xspace$ with an asymmetric‑drift correction. We find that most of our galaxies lie along the relation, albeit with a large scatter, but the lowest-mass galaxies drop off. We compare our measurements at $z\sim 5$ to the relation found for the KMOS$^{\rm 3D}$$z=0.9-2.3$ star-forming galaxies from Ubler:2017aa (dashed orange line), and the medians from the RC100 sample Nestor-Shachar:2023aa. We also include the TFR at $z\sim 0$ from Reyes:2011aa for reference (dot-dashed tan curve). We fit our data with the same fixed slope as these works, $a = 3.60$, but computing our own zero-point $b$. We find $\Delta b \approx -1.3$ dex between from $z=0.9-2.3$ to $z=4-6$. The large intrinsic scatter around the relation ($\sigma_{\rm int} = 0.49 \pm 0.06$; shaded region) indicates that it only beginning to emerge at $z\sim5$.
• Figure 3: Gas (top) and DM (bottom) fractions (within $r_{\mathrm{e}}\xspace$) as a function of stellar mass, with the characteristic uncertainty shown in grey. The $f_{\text{gas}} (r<r_{\mathrm{e}}\xspace)$ (Eq. \ref{['eq:fgas']}) is computed using the sSFR, redshift, and stellar mass of each galaxy following Tacconi:2020aa. The $f_{\text{DM}} (r<r_{\mathrm{e}}\xspace)$ is computed using $M_{\text{gas}}$ along with $M_{\text{dyn}}$ and $M_{\star}$ as shown in Eq. \ref{['eq:fdm']}. We fit our data with a power-law (solid purple line) for both relations, showing the intrinsic scatter (shaded region) only for $f_{\mathrm{DM}}\xspace$. We cannot constrain $\sigma_{\rm int}$ for $f_{\mathrm{gas}}\xspace$ due to the inferred uncertainties being significantly larger than the scatter in the outputs of the Tacconi:2020aa scaling relation for our sample (see Sec. \ref{['sec:gas-fracs']}). We compare our results with measurements from Genzel:2020aa (green stars) and Nestor-Shachar:2023aa (RC100; blue diamonds) at cosmic noon and de-Graaff:2024abde-Graaff:2024aa (orange pentagons) and Lee:2025aa (pink hexagons) at high redshift. For most of our sample, we find high gas and DM fractions $f > 0.5$. We find relatively good agreement with median trends from the thesan-zoom simulations McClymont:2025ac and the tng50 simulations de-Graaff:2024aa.
• Figure 4: DM fraction as a function of baryonic surface density ($\Sigma_{\rm bar}$) for our sample (circles) and samples at lower Nestor-Shachar:2023aa and similar to higher de-Graaff:2024aa redshift. We colour-code points by stellar mass (left panel) and gas fractions (right panel). We show the characteristic uncertainty for our sample in grey. We find a strong negative correlation between $f_{\mathrm{DM}}\xspace$ and $\Sigma_{\rm bar}$ ($\rho = -0.551, p < 0.001$), with a best-fit relation (solid purple line) steeper than Wuyts:2016aa at cosmic noon (dashed brown line), and with an offset. We also plot the best-fit relation obtain when fitting our points with those from Nestor-Shachar:2023aa (dashed-dotted red line). For comparison, we plot medians from ETGs at $z\sim 0$ from the ATLAS3D survey Cappellari:2013aa and LTGs from the DiskMass Survey Martinsson:2013aa. Our galaxies agree with Nestor-Shachar:2023aa at low fractions $f_{\mathrm{DM}}\xspace<0.5$, but lie above the relation at high fractions. The galaxies in our sample are smaller, less massive, and more gas rich than those probed at cosmic noon, driving the apparent shift in the relation.
• Figure 5: DM fraction as a function of redshift for our sample (circles) and samples from the literature Nestor-Shachar:2023aade-Graaff:2024abde-Graaff:2024aa. We plot the evolution measured in Nestor-Shachar:2023aa (dashed blue line), and its extrapolation to $z>2.5$ (dotted blue line). Our sample medians lie well above those at cosmic noon, although our sample shows large diversity in $f_{\mathrm{DM}}\xspace$ and probes $\sim 2-3$ dex lower stellar masses. Qualitatively, we compare our medians ($\log(M_{\star}\xspace ~\rm [\rm M_{\odot}\xspace])\xspace\approx 9$) with the median evolution of $f_{\mathrm{DM}}\xspace(r<r_{\mathrm{e}}\xspace)$ for tng50 galaxies de-Graaff:2024aa selected at $z=6$ with $8<\log(M_{\star}\xspace ~\rm [\rm M_{\odot}\xspace])\xspace<9$ (thick line with $16^{\rm th}$ and $84^{\rm th}$ percentiles shown in the grey shaded region). Similarly, we plot the median evolution for thesan-zoom galaxies McClymont:2025ac with $\log(M_{\star}\xspace ~\rm [\rm M_{\odot}\xspace])\xspace = 7 \pm 0.5$ at $z=9$ (thick line with $16^{\rm th}$ and $84^{\rm th}$ percentiles shown in the green shaded region). Although we expect redshift evolution of $f_{\mathrm{DM}}\xspace(<r_{\mathrm{e}}\xspace)$ due to the evolution of sizes with redshift, the strong dependence of stellar mass is evident here.