A 3D view on the local gravitational instability of cold gas discs in star-forming galaxies at $0\lesssim\mathrm{z}\lesssim 5$

We consider the typical case of a rotating gas disc in the gravitational potential of a galaxy. This whole system is symmetric with respect to the rotation axis and the midplane. Therefore, the total gravitational potential of the galaxy is axisymmetric: $\Phi_{\mathrm{tot}}=\Phi_{\mathrm{tot}}(R,z)$, where $R$ and $z$ are the cylindrical coordinates. The unperturbed gas disc is assumed to stationary rotate with $V_{\mathrm{rot}}=V_{\mathrm{rot}}(R)$ and has no radial or vertical motions. We further assume that the gas disc is in vertical hydrostatic equilibrium in the galactic potential. We also take the gas pressure to be $P=\rho\sigma^{2}$ with $\rho$ the gas volume density. For simplicity, we assume that the velocity dispersion is isotropic and independent of $z$, thus $\sigma=\sigma(R)$. The gas vertical distribution at a given radius is then (e.g. Olling, 1995)

where $\rho(R,0)$ and $\Phi_{\mathrm{tot}}(R,0)$ are, respectively, the gas volume density and the total gravitational potential evaluated in the midplane. In particular, $\Phi_{\mathrm{tot}}$ can be calculated via numerical integration assuming a model of the mass distribution of the galaxy.

The main mass components of a star-forming galaxy are the dark matter (DM) halo, the stellar disc (or discs), the stellar bulge (when present), and the cold gas mostly consisting of atomic gas and molecular gas, which are usually distributed in discs. We rely on models of the mass distribution available in the literature (see Sect. 3 for details). The functional forms used for each mass component are described in the following sections.

Nipoti (2023) has shown that a criterion sufficient for the LGI of a gaseous disc is

where $\kappa$ is the epicycle frequency (Eq. 2), $h_{z}$ is the gas disc thickness defined as the height of an infinitesimal vertical strip centred at $z=0$ and containing $\approx 70$% of the mass per unit surface, and $\nu$ is a frequency related to vertical gradients of the gas pressure and density

with the last equality holding under the assumptions $\sigma=\sigma(R)$ and $P=\rho\sigma^{2}$. We note that $\rho=\rho(R,z)$, $\kappa=\kappa(R)$ (as we assumed cylindrical rotation), and $h_{z}=h_{z}(R)$ by definition, thus $Q_{\mathrm{3D}}(R,z)$. In Eq. 12, we have substituted the gas sound speed $c_{\mathrm{s}}=\sqrt{\partial P/\partial\rho}$ in Nipoti (2023) with $\sigma$, implicitly assuming that the perturbations occur at constant $\sigma$. We stress that Eq. 12 is a sufficient criterion for instability, hence it does not guarantee that the disc is stable where $Q_{\rm 3D}>1$³³3Eq. 12 in Nipoti (2023) provides a sufficient criterion for stability, however this is never verified in our case as $\partial\sigma/\partial z=0$ and $\partial P=\sigma^{2}\partial\rho$.. In particular, this criterion was derived assuming axisymmetric perturbations, but it cannot inform us about the stability of the disc against non-axisymmetric perturbations (see Sect. 5.3 for discussion).

For our gas disc in vertical hydrostatic equilibrium, Eq. 12 can be evaluated once $V_{\mathrm{rot}}$, $\sigma$, and $\Phi_{\mathrm{tot}}$ are known. The first two ingredients, $V_{\mathrm{rot}}$ and $\sigma$, can be measured from emission line observations of gas tracers (see Sect. 3 for details), while $\Phi_{\mathrm{tot}}$ can be calculated via numerical integration of the mass models described in Sect. 2.1. However, to obtain a self-consistent 3D model of the gas disc, the self-gravity of the gas disc must be taken into account. We explain our methodology in the following section.

To calculate $Q_{\rm 3D}$ using Eq. 12, we need the gas volume density as a function of $R$ and $z$, and the radial profile of the gas disc thickness. Under the assumptions described in Sect. 2.1, the vertical distribution of our gas disc is regulated by the hydrostatic equilibrium between the gravitational potential of the galaxy $\Phi_{\mathrm{tot}}$ and the gas pressure. Crucially, $\Phi_{\mathrm{tot}}$ includes not only the stellar and DM potentials, but also the gas potential $\Phi$, which depends on the 3D distribution of the gas itself.

Galpynamics⁴⁴4https://gitlab.com/iogiul/galpynamics (Iorio, 2018) is an ideal tool to obtain $\rho(R,z)$ taking into account the whole potential. This is a Python module for galaxy dynamics that includes an iterative algorithm to account for the gas self-gravity (see also Abramova & Zasov, 2008; Banerjee et al., 2011). Given a set of mass components, Galpynamics can be used to calculate, via numerical integration, the gravitational potential of a galaxy, the 3D density distribution of a gas disc in vertical hydrostatic equilibrium in such potential, and the radial profile of gas scale height. The gas in the disc has a user-defined velocity dispersion, which we model using the exponential profile

with $\sigma_{0}$ being the gas velocity dispersion at the galaxy centre and $R_{\sigma}$ a scale radius. These parameters are obtained by fitting Eq. 14 to the radial profile of the gas velocity dispersion measured from observations (see Sect. 3). The observed velocity dispersion is used in the rare cases when Eq. 14 cannot describe the observed profile (e.g. non-monotonic trends).

For a galaxy model including a DM halo, a stellar component, and a gas disc, Galpynamics applies the following iterative procedure. In a preliminary stage, the software calculates the external and fixed potential of DM and stars (i.e $\Phi_{\mathrm{ext}}=\Phi_{\mathrm{DM}}+\Phi_{\star}$) on a user-defined grid. Initially, the gas disc is assumed infinitesimally thin to estimate its gravitational potential $\Phi$ by numerical integration. The total galactic potential is then set to $\Phi_{\mathrm{tot}}=\Phi_{\mathrm{ext}}+\Phi$. A first estimate of the gas 3D distribution is obtained through Eq. 4. An initial guess of the gas scale height ($z_{\mathrm{gas}}$) is also derived by fitting the gas vertical distribution at each $R$ using either the Gaussian profile $\exp[-z^{2}/(2z_{\mathrm{gas}}^{2})]$ or the function $\operatorname{sech}^{2}(z/z_{\mathrm{gas}})$. The profile that is most suitable to fit the gas vertical profile is chosen as the one that minimises the residuals between the best-fit curve and $\rho(R,z)$ calculated numerically through Eq. 4. The resulting scale height is adopted to re-evaluate $\Phi$, which is then used to update $\Phi_{\mathrm{tot}}$. This procedure is iterated until two successive computations of the scale height differ by less than a tolerance factor chosen by the user. When convergence is reached, the gas volume density is calculated on the user-defined grid and the gas disc thickness is obtained as $h_{z}=2z_{\mathrm{gas}}$. Thus, $h_{z}$ contains $\approx 68$% and $\approx 76$% of the mass per unit surface for the Gaussian profile and the $\operatorname{sech}^{2}$ distribution respectively, which is within the fiducial range $\approx 60-80$% recommended by Nipoti (2023).

In the case of present-day galaxies hosting an atomic gas disc and a molecular gas disc (see Sect. 3), we follow the same approach as in Bacchini et al. (2019a). We first calculate $\rho(R,z)$ and $h_{z}$ for the disc of atomic gas and we then do the same for the molecular gas disc but including also the potential of the atomic gas disc using the model obtained in the previous step. Therefore, the distribution of the atomic gas disc is not influenced by the molecular gas. This choice does not significantly affect the results, as the molecular gas disc is subdominant with respect to the other mass components, at least for the present-day galaxies studied here (for a thorough study, see Mancera Piña et al., 2022a).

It is worth noting that the procedure used here to derive the 3D distribution of galactic gas discs is a standard methodology that has been applied to several nearby galaxies (e.g. Iorio, 2018; Bacchini et al., 2019a, b, 2020a, 2020b; Mancera Piña et al., 2022a, but also Banerjee et al. 2011 for a similar approach). As explained in the following section, our sample includes not only present-day galaxies, but also systems at high redshift for which this procedure to derive the 3D gas distribution has never been applied before. We extend the application of this methodology to high-z galaxies relying on the fact that the properties of their gas discs are quite similar to those of galaxies in the present-day Universe (e.g. the gas kinematics is rotation-dominated, see Rizzo et al., 2020, 2021; Lelli et al., 2021; Roman-Oliveira et al., 2023).

We collected the data for 31 galaxies in the present-day Universe consisting of the following sub-samples.

We collected the data for 13 galaxies at $\mathrm{z}\gtrsim 1$, consisting in the following three sub-samples.

Figure 2 shows the scale height profiles $z_{\mathrm{gas}}$ of the gas discs for galaxy in the sample (see Figs. 7 for the scale height profiles plotted in linear scale). The top panels are for the atomic gas discs in the present-day Universe, while the bottom panels are for the molecular gas discs in nearby spirals (left) and for cold gas discs of SFMS galaxies (center) and starbursts (right) at high redshift. We note that the scale heights of nearby spirals (including the MW) and of some dwarfs were already displayed in previous works (Bacchini et al., 2019a, b, 2020b; Mancera Piña et al., 2022a), but this is the first time that the radial profiles of the gas disc scale heights are shown for the high-z sample. The scale heights of the atomic gas discs in present-day galaxies (top panels in Fig. 2) tend to be a few hundreds parsecs in the inner regions and then increase up to about 1 kpc in the outskirts, with bulge hosts having very thin discs in their innermost regions because the stellar mass is centrally concentrated. The same happens for the molecular gas discs at $\mathrm{z}\approx 0$ (bottom right panel in Fig. 2), which tend to be thinner than the atomic gas discs due to the lower velocity dispersion of the molecular gas (Bacchini et al., 2020b). As pointed out by Mancera Piña et al. (2022a), despite the radial profiles of the atomic and molecular gas scale heights share some similarities, there is little evidence for the ”universal” profile proposed by Patra (2019, 2020a, 2020b). In these works, the author postulated the existence of a universal linear profile for the gas scale height when normalised to a characteristic scale. However, the precise shape of $z_{\mathrm{gas}}(R)$ depends not only on the relative contribution of the stellar disc and the DM halo at a given radius, but also on the bulge contribution and, most importantly, on the gas velocity dispersion (for futher discussion, see Mancera Piña et al., 2022a). All these factors can be significantly different from galaxy to galaxy, creating a variety of profiles that is difficult to describe with a simple functional form.

The galaxies at cosmic noon (zC400569 and zC488879) host flaring molecular gas discs very similar to those of present-day galaxies (see bottom central panel in Fig. 2). We note that, since the CO velocity dispersion in these systems is an upper limit (which may be itself uncertain, see Lelli et al., 2023), the molecular gas scale height may be overestimated. Nevertheless, the scale heights of zC400569 and zC488879 in Fig. 2 are very similar to those of molecular gas discs in present-day galaxies, which were obtained using $\sigma$ measured from high resolution observations. In the case of gas discs observed via [CII] emission (bottom central and right panels in Fig. 2), the scale heights tend to be similar, on average, to those of present-day galaxies, which is consistent with the idea that the [CII] emission can trace both the atomic and molecular gas (e.g. Carilli & Walter, 2013; Zanella et al., 2018; Vizgan et al., 2022b). We do not find striking differences between the scale heights of main-sequence and starburst galaxies at high redshift. In the latter, the strong supernova feedback is expected to enhance the gas turbulence with respect to moderately star-forming galaxies, increasing the gas disc thickness. The lack of a dependence of the gas scale heights on the SFR is not straightforward to interpret, as the gas scale height depends not only on the gas velocity dispersion, but also on the galaxy potential. We note though that BRI1135-0417 is one of the galaxies with the highest SFR (see Table 1) and hosts the thickest gas disc ($\approx 2$ kpc). Overall, these results indicate that the scale height of gas discs at massive high-z galaxies is of the order of 100 pc.

For the majority of the gas discs, the scale height profiles increase monotonically for increasing galactocentric distance, indicating a flaring. However, the scale heights of the UDG (AGC114905) and nine high-z galaxies increase with $R$ in the central parts up to a peak value and decreases outward. The non-monotonic behaviour in the outer parts can be understood from Eq. 4: the scale height drops when $\sigma^{2}$ decreases more rapidly with $R$ than the difference between the gravitational potential in the midplane and at $|z|>0$. In practice, the gas pressure is not sufficient to support the flaring and the radial profile of the gas scale height starts decreasing. Typically, the gas scale height does not increase monotonically outwards in the galaxies in which the contribution of the DM halo to the galactic potential is subdominant with respect to the baryonic component (see § 4.6.2 in Cimatti et al., 2019).

Roman-Oliveira et al. (2023) performed some experiments to obtain a tentative estimate of the gas disc scale height in J81740, SGP38326-1, SGP38326-2, and BRI1135-0417 using the Python routine Cannubi⁷⁷7https://www.filippofraternali.com/cannubi. This was designed to estimate geometrical parameters of a rotating disc performing an iterative 3D modelling of the observations using ^3DBarolo . Roman-Oliveira et al. (2023) report scale heights of the order of 1 kpc, while we obtain $z_{\mathrm{gas}}\lesssim 0.2$ kpc using Galpynamics. The only exception is BRI1135-0417: we find $z_{\mathrm{gas}}\approx 1-1.5$ kpc, which is close to the values reported by Roman-Oliveira et al. (2023). The origin of the discrepancy between the scale heights for J81740, SGP38326-1, and SGP38326-2 is not easy to understand, as the methods used are not fully comparable: Cannubi assumes that the thickness is constant within the disc, while Galpynamics derives a radial profile using a dynamical modelling based on the vertical hydrostatic equilibrium. As pointed out by Roman-Oliveira et al. (2023), further tests are needed to understand how to break the degeneracy between thickness and inclination of the observed discs, in particular when the spatial scale resolved by the data is much larger than the gas scale height (as in the case of these systems). We therefore decided to use the profile of $z_{\mathrm{gas}}$ obtained with Galpynamics in our analysis and leave further investigations for future works.

We derive the radial profiles of $Q$ (Eq. 1) using $V_{\mathrm{rot}}$, $\sigma$, and $\Sigma$ ⁸⁸8The atomic gas surface density of galaxies at $\mathrm{z}\approx 0$ includes a multiplicative factor of 1.36 to account for helium. measured from observations, searching for gas discs with $Q<1$ in some radial interval. The resulting profiles are shown in Fig. 3 (see also Figs. 7 for $Q(R)$ shown in linear scale): the top panels are for atomic gas discs in present-day galaxies, while the bottom panels are for molecular gas discs in local spirals (left) and for high-z galaxies on the SFMS (center) and the SBS (right). If we classify as potentially unstable only galaxies with $Q<1$ in a region larger than the resolution element of the observations, we find that 12 galaxies host potentially unstable gas discs according to the 2D stability criterion. However, the unstable regions in DDO 50 and NGC 3198 (see Figs. 7) are most likely due to fluctuations in the observed profiles rather than a real feature of the disc. Indeed, the uncertainties on observables ($\sigma$ in particular) are large in these cases (see also Sect. 5.3). We thus conclude that, at $\mathrm{z}\approx 0$, only the atomic gas disc of the UDG AGC114905 can be classified as locally unstable across a significantly large region disc. Unsurprisingly, the atomic gas disc of AGC114905 is also unstable if we use the data from Mancera Piña et al. (2022b), who measure lower $\sigma$ with respect to Mancera Piña et al. (2024). For AGC114905, the radial extent of the unstable region where $Q\lesssim 1$ is $L_{Q}\approx 4$ kpc (see Table 2). In the high-z sample, we find that nine out of 13 galaxies (i.e. $\approx 70$%) have $Q<1$ beyond a certain galactocentric distance, while only ALESS073.1, BRI1335-0417, SPT0345-47, and zC488879 have no unstable region. Typically, the radial extent of the unstable regions ranges from $\sim 1$ kpc to a few kpc when the 2D instability criterion is adopted.

We estimate the fraction of the unstable gas as

where $M_{\mathrm{gas}}$ is the total mass of the gaseous disc and $M_{\mathrm{inst}}$ is the mass of the unstable gas obtained by integrating the gas surface density in the region where $Q(R)<1$. We find a broad range of mass fractions ($0.1\lesssim f_{\mathrm{inst}}\lesssim 1.0$) with median and $1\sigma$ uncertainty of $\langle f_{\mathrm{inst}}\rangle=0.6^{+0.2}_{-0.3}$. Table 2 reports the total gas mass, the properties of the unstable regions (mass and radial extent), and the fraction of unstable gas for the galaxies with $Q<1$.

We then apply the 3D instability criterion by Nipoti (2023) using Eq. 12, aiming to understand whether there is any gas disc that is unstable (or partially unstable) when its vertical structure is taken into account. For each galaxy, we obtain $\rho(R,z)$ using the procedure described in Sect. 2.3 and we calculate Eqs. 2 and 13 using the central difference method. The resulting maps of $Q_{\rm 3D}(R,z)$ are shown in Fig. 4 for the galaxies with $Q<1$ (see Figs. 7 for the whole sample). Interestingly, we find that two galaxies with $Q<1$ in some radial interval, namely AGC114905 and SPT2146-55, have $Q_{\rm 3D}>1$ everywhere throughout the disc. Therefore, no galaxy in our sample at $\mathrm{z}\approx 0$ host unstable gas discs, considering both the atomic gas and the molecular gas. We note that, using the data from Mancera Piña et al. (2022b), we would find that AGC114905 is locally unstable, as the stabilising effect of the gas vertical structure is less effective since $\sigma$ is slightly lower than in Mancera Piña et al. (2024). Only eight galaxies out of a sample of 44 ($\approx 18$%) host locally unstable gas discs according to the 3D criterion. These are all at redshift $\mathrm{2}\lesssim\mathrm{z}\lesssim 5$ and account for $\approx 60$% of the whole sample of galaxies at high redshift. The gas tracer is [CII] for all the locally unstable systems, except zC400569, which is traced by CO.

Figure 4 shows that the vertical extent of the unstable region is always smaller than the gas scale height (grey dot-dashed curve). This confirms that the disc is more prone to LGI near the midplane (Nipoti, 2023). Figure 4 also shows that the radial extent $L_{Q_{\rm 3D}}$ of an unstable region according to $Q_{\rm 3D}$ is smaller than $L_{\mathrm{Q}}$. Table 2 compares the values of $L_{Q_{\rm 3D}}$ and $L_{Q}$. If we consider only the galaxies that are unstable based on the 3D criterion, the median value of $L_{Q_{\rm 3D}}/L_{Q}$ is $\approx 0.8\pm 0.1$ (the uncertainty corresponds to $1\sigma$). Hence, the instability region is $\approx 20$% smaller when the vertical stratification of the gas disc is taken into account using $Q_{\rm 3D}$ rather than $Q$.

The mass of the gas that is located in the unstable regions ($M_{\mathrm{inst}}$) can be calculated by integrating $\rho(R,z)$ in $R$ and $z$ where $Q_{\rm 3D}(R,z)<1$, which is the region encompassed by the white contour in Fig. 4. The resulting values can then be used in Eq. 15 to calculate the fraction of unstable gas $f_{\mathrm{inst}}$. The results are reported in Table 2: we find $0.1\lesssim f_{\mathrm{inst}}\lesssim 0.5$ with median and $1\sigma$ error $\langle f_{\mathrm{inst}}\rangle=0.3\pm 0.1$. Hence, $f_{\mathrm{inst}}$ based on the 3D criterion is reduced by a factor of 2 on overage and covers a narrower range of values compared to the estimates resulting from the 2D analysis. These results also indicate that a small fraction of the total gas in the disc is potentially involved in the disc fragmentation.