MIDIS: JWST/MIRI Reveals the Stellar Structure of ALMA-selected Galaxies in the Hubble Ultra Deep Field at Cosmic Noon

We present deep James Webb Space Telescope (JWST)/Mid-Infrared Instrument (MIRI) F560W observations of a flux-limited, Atacama Large Millimeter/submillimeter Array (ALMA)-selected sample of 28 galaxies at z = 0.5–3.7 in the Hubble Ultra Deep Field (HUDF). The data from the MIRI Deep Imaging Survey (MIDIS) reveal the stellar structure of the HUDF galaxies at rest-frame wavelengths of λ > 1 μm for the first time. We revise the stellar mass estimates using new JWST photometry and find good agreement with pre-JWST analyses; the few discrepancies can be explained by blending issues in the earlier lower-resolution Spitzer data. At z ∼ 2.5, the resolved rest-frame near-infrared (1.6 μm) structure of the galaxies is significantly more smooth and centrally concentrated than seen by the Hubble Space Telescope at rest-frame 450 nm (F160W), with effective radii of R e (F560W) = 1–5 kpc and Sérsic indices mostly close to an exponential (disk-like) profile (n ≈ 1), up to n ≈ 5 (excluding active galactic nuclei). We find an average size ratio of R e (F560W)/R e (F160W) ≈ 0.7 that decreases with stellar mass. The stellar structure of the ALMA-selected galaxies is indistinguishable from a HUDF reference sample of all galaxies with a MIRI flux density greater than 1 μJy. We supplement our analysis with custom-made, position-dependent, empirical point-spread function models for the F560W observations. The results imply that a smoother stellar structure is in place in massive gas-rich, star-forming galaxies at “Cosmic Noon,” despite a more clumpy rest-frame optical appearance, placing additional constraints on galaxy formation simulations. As a next step, matched-resolution, resolved ALMA observations will be crucial to further link the mass- and light-weighted galaxy structures to the dusty interstellar medium.


Introduction
Galaxy formation reached its high point around 10 billion years ago during the peak of cosmic star formation at "Cosmic Noon" (z ≈ 1-3; Madau & Dickinson 2014).The typical starforming galaxy, located on the galaxy main sequence, formed around 8× more stars during this epoch than a galaxy with similar stellar mass in the local Universe (e.g., Whitaker et al. 2014).This rise in the global star formation rates goes hand in hand with an increase in the global molecular gas content of galaxies (Walter et al. 2014(Walter et al. , 2016;;Riechers et al. 2019;Decarli et al. 2020;Boogaard et al. 2023).Galaxies at Cosmic Noon Original content from this work may be used under the terms of the Creative Commons Attribution 4.0 licence.Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
are significantly more gas rich than their local counterparts at fixed stellar mass, with a total cold gas mass that can exceed the total mass in stars (M mol /M * 1; e.g., Tacconi et al. 2013Tacconi et al. , 2018)).These large gas fractions are expected to have a significant impact on the conditions inside the interstellar medium, potentially leading to the apparently clumpy starforming structures in the rest-frame UV/optical and increased ionized-gas velocity dispersions (see Förster Schreiber & Wuyts 2020 for a review).
Yet, the older stellar populations-best traced at rest-frame wavelengths λ > 1 μm-which contain the bulk of the stellar mass, have remained unresolved due to the relatively large point-spread function (PSF) of the Spitzer/IRAC instrument (1 6-2 0, or 13-17 kpc at z ≈ 2; Fazio et al. 2004) and their fluxes have remained uncertain due to the complex deblending of neighboring sources (e.g., Labbé et al. 2015).As a result, the total stellar mass and structure of the galaxies remain uncertain, as well as the gas fraction and its potential impact on galaxy structure.This leaves open important questions, such as whether the galaxies are intrinsically clumpy, or whether their underlying stellar-mass distribution is smoother.
The Mid-Infrared Instrument (MIRI) on board the James Webb Space Telescope (JWST) can now provide sensitive, high-resolution imaging at wavelengths of 5.6 μm and above (Wright et al. 2023).This allows one to trace the rest-frame near-infrared (NIR) light of galaxies at z ≈ 2.5 (λ rest ≈ 1.6 μm) and beyond for the first time, with unprecedented spatial resolution (≈0 2 at 5.6 μm, or 1.7 kpc at z ≈ 2).
In this paper, we utilize deep JWST/MIRI observations to study a flux-limited sample of dust-continuum-and cold gasselected galaxies in the Hubble Ultra Deep Field (HUDF).The sample is taken from the Atacama Large Millimeter/submillimeter Array (ALMA) Spectroscopic Survey (ASPECS) Large Program (Walter et al. 2016;Decarli et al. 2019Decarli et al. , 2020) that performed flux-limited spectral scans in the 1.2 mm and 3 mm bands to detect molecular gas via 12 CO across cosmic time (Boogaard et al. 2019;González-López et al. 2019) while simultaneously obtaining extremely sensitive 1.2 mm dustcontinuum imaging over the HUDF (Aravena et al. 2020;González-López et al. 2020).
This paper is organized as follows: in Section 2 we present the JWST observations, including NIRCam imaging and slitless spectroscopy.In Section 3 we discuss the updated properties of the full ALMA sample in the JWST era.We then analyze the rest-frame NIR morphology focusing on the z 2 galaxies, now probed by the deep MIRI/F560W observations, in Section 4 and discuss the results in Section 5. We adopt a Chabrier (2003) initial mass function and a Planck Collaboration et al. (2020) cosmology (flat Lambda cold dark matter (ΛCDM) with H 0 = 67.7 km s −1 Mpc −1 , Ω m = 0.31, and Ω Λ = 0.69).We use log to denote log 10 and ln for the natural logarithm.We report magnitudes in the AB system (Oke & Gunn 1983).

JWST Observations
The MIRI Deep Imaging Survey (MIDIS) is a deep JWST/ MIRI survey of the HUDF conducted by the MIRI European Consortium GTO program (Program ID 1283, PI: G. Östlin).MIDIS was designed as a single F560W deep field, spread over six observations totaling 63 hr (48.8 hr net exposure time), with 100 groups, 100 integrations, and a 10-point random cycling dither pattern.The majority of the MIDIS observations (4/6) were executed between 2022 December 2 and 6, while the fifth observation was executed on 2022 December 20, at a slightly different central position and position angle, due to several safe-mode incidents of the observatory in the meantime.The data were reduced using the official JWST pipeline version 1.12.3 (CRDS pmap 1137; Bushouse et al. 2023).In addition to the default pipeline steps, we perform custom routines to deal with cosmic showers, background variations, and other wellknown MIRI instrument effects as well as the absolute and relative astrometric alignment of the exposures.These additional steps use a similar approach as those taken in other highredshift studies (e.g., Iani et al. 2022;Bagley et al. 2023;Rinaldi et al. 2023).The final mosaic consists of a total of 96 exposures with 50 unique dithers and has a total exposure time of 41.4 hr, reaching a 5σ point-source depth of 28.6 mag (accounting for correlated noise), and covers a total area of 4.7 arcmin 2 .Further details will be described in G. Östlin et al. (2024, in preparation).
In addition to the MIRI data, we use the publicly available JWST/NIRCam imaging over the HUDF, taken as part of the programs JADES (Program ID 1180; Rieke et al. 2023), JEMS (Program ID 1963;Williams et al. 2023), andFRESCO (Program ID 1895;Oesch et al. 2023).We use the combined medium-band and wide-band observations as released by the JADES team (Data Release 1) and the GRIZLI Image Release v7.0, 2023 July. 21WST/NIRCam slitless spectroscopy over the HUDF was taken as part of the FRESCO program and used to obtain redshift information of galaxies without prior secure redshift information.These observations were taken in the F444W filter (covering 3.9-5.0μm, limited to 4.4 μm for part of the survey area) and a single grism orientation (GrismR) with a resolution of R ∼ 1600. 2 hr of exposure time is obtained in eight exposures, designed to reach a 5σ point-source depth for an unresolved emission line of 2 × 10 −18 erg s −1 cm −2 .For more details on the spectral coverage and extraction we refer to Oesch et al. (2023).

Sample
The flux-limited ASPECS sample consists of all galaxies that are detected in the deep 1.2 mm dust-continuum image from ASPECS, which has an unprecedented 9.3 μJy beam −1 rms sensitivity and a resolution (beam size) of roughly 1 3 (Aravena et al. 2020;González-López et al. 2020).From this sample, we discard three sources (1mm.C27, 1mm.C29, and 1mm.C34) that remain without counterparts in the optical-far-infrared (FIR) and also the new JWST imaging and are potentially false positives, consistent with the fidelity estimates of the sample (González-López et al. 2020).The continuum-selected sample includes all of the 12 CO-selected galaxies (Boogaard et al. 2019;González-López et al. 2019), as discussed in Boogaard et al. (2020), apart from two sources detected in 12 CO only, which are included in Table 1 for completeness, but excluded from the analysis where relevant (one of these sources is part of the continuum-faint sample, see Section 4.2).The sample also encompasses all 1 mm continuum sources from shallower ALMA data in the HUDF area (e.g., Dunlop et al. 2017;Franco et al. 2018;Hatsukade et al. 2018).(This table is available in machine-readable form.) As the Rayleigh-Jeans tail of the dust-continuum emission is nearly always optically thin, the 1.2 mm selection is essentially a cold dust mass selection (the phase which contains most of the dust mass) and effectively also a cold, molecular gas mass selection (e.g., Hildebrand 1983;Scoville et al. 2014Scoville et al. , 2016;;see Aravena et al. 2020, their Appendix A, on the consistency between the dust-and 12 CO-based molecular gas mass estimates).
The complete sample consists of 35 galaxies, 28 of which are inside of the MIRI/F560W footprint of the HUDF, as shown in Figure 1 (see also Table 1), and span a redshift range of z = 0.45-3.71.We show MIRI/NIRCam/Hubble Space Telescope (HST) cutouts in Figure 2.

Redshifts
Most ASPECS sources already have spectroscopic redshifts from the deep MUSE HUDF and MXDF Surveys (Bacon et al. 2017(Bacon et al. , 2023;;see Boogaard et al. 2019 andAravena et al. 2020) and/or multi-J CO and C I emission (Boogaard et al. 2020).The few sources with missing spectroscopic redshifts lie primarily between z = 1.5 and 2.9 where there are no bright emission lines covered by MUSE, including 1.74 < z < 2.0, where there is also no low-J CO-coverage from ASPECS (see Boogaard et al. 2019, their Figure 1).This gap in spectroscopic redshift coverage is now mostly filled by FRESCO, which covers a range of IR emission lines (including He I 1.083 μm, [Fe II] 1.257 μm, Paγ 1.094 μm, and Paβ 1.282 μm lines between 2 < z < 3.6, Paα 1.876 μm between 1 < z < 1.6, and many fainter lines).
We extract FRESCO spectra for all ALMA sources, removing the continuum using a median filter, and search for bright emission lines in a Δz/(1 + z) = 0.1 window around the known redshift.We detect line emission in nearly all sources that have spectral coverage of the brightest emission lines.Notably, we confirm the earlier photometric redshifts for all sources (see Aravena et al. 2020).We detect [S III] λλ9071, 9533 in the highest-redshift sources and confirm the tentative MUSE redshifts for two known blended sources, 1mm.C08 and 1mm.C22 (which corresponds to the northern galaxy).For 1mm.C2 no lines are covered in the FRESCO wavelength range, but we report a spectroscopic redshift of z = 1.91 based on the detection of Hα in the NGDEEP slitless spectroscopy (Pirzkal et al. 2023;Bagley et al. 2024).The updated redshift information is given in Table 1.The FRESCO spectra of the aforementioned sources are shown in Figure 10 in Appendix B. Notably, all sources in the sample now have a spectroscopic redshift.

Spectral Energy Distributions
Photometry is performed for all sources in the F560W image in the MIRI, NIRCam and HST bands with THE FARMER (Weaver et al. 2022), using our F560W PSF model for MIRI (see Appendix A) and WEBBPSF models for NIRCam (Perrin et al. 2012(Perrin et al. , 2014)).In brief, we first detect and model the galaxies in the F560W MIRI band, and then perform forced photometry in the other multiwavelength bands, allowing the flux to vary, while keeping the structural parameters fixed (details will be presented in S. Gillman et al. 2024, in preparation).
We compare the F560W fluxes for all sources in the combined MIDIS and ASPECS field to the Spitzer IRAC/ Channel 3 (5.8μm) fluxes from CANDELS (Guo et al. 2013) in Figure 3, masking sources near the edge of the MIDIS image.Overall, the Spitzer fluxes agree well with the MIRI ones at flux densities 10 μJy, but show larger scatter below ≈1 μJy, consistent with the 5σ limit of the Spitzer photometry.Visual inspection shows that the strong outliers above the Spitzer limit are largely due to neighboring sources that are poorly (or not) deblended in the Spitzer photometry.The  strongest outlier from ASPECS is 1mm.C22, which is indeed heavily blended with a similarly bright neighboring galaxy (see Figure 2).Interestingly, toward the bright end there seems to be a increasing offset between the Spitzer fluxes compared to the MIRI fluxes (though the number of sources is limited), which has become more apparent with the updated MIRI F560W flux calibration released in 2023 September. 22Potential causes for the offset are discussed in more detail in G. Östlin et al. (2024, in preparation) and are potentially related to the significant changes in the intrinsic morphology between the HST/F160W model used to measure the IRAC fluxes and the actual morphology in F560W (see Section 4.1) that is used by THE FARMER in this work, as well as deblending issues of nearby faint galaxies; though the exact nature remains unclear at the time of writing.Discrepancies between the Spitzer IRAC and JWST fluxes at the bright end have also been reported in other works (e.g., Rieke et al. 2023;Yang et al. 2023) and at least in part attributed to the order-of-magnitude difference in the PSF between IRAC and JWST.
We model the spectral energy distributions using the high-z version of MAGPHYS (Da Cunha et al. 2008, 2015).We join the new JWST and HST photometry with the total fluxes from existing longer-wavelength photometry from Spitzer/IRAC at 24 μm (Whitaker et al. 2014), Herschel/PACS at 100 and 160 μm (Elbaz et al. 2011), and ALMA at 1 mm and 3 mm (Dunlop et al. 2017;González-López et al. 2019, 2020), including 5σ upper limits on the millimeter flux.The resulting stellar masses are listed in Table 1.The total stellar masses are on average very consistent (within 0.04 dex) with the pre-JWST determinations using the same models (see Aravena et al. 2020;Boogaard et al. 2020), though some individual sources show larger scatter (≈0.25 dex) likely due to differences in the photometry.We attribute the overall agreement in the total stellar masses to the relative consistency between the integrated fluxes from Spitzer and JWST and the large number of constraints on the shape of the spectral energy distribution across wavelengths available for the ALMA sources in the HUDF.

Structural Parameters
We measure the global structural parameters observed in the MIRI/F560W filter by modeling the galaxies with a single Sérsic profile (Sérsic 1963) using GALFIT (Peng et al. 2002).We create 200 × 200 pixel cutouts around each source from the background-subtracted F560W science image and error map at a pixel scale of 0 06.We adopt initial guesses on the magnitude, half light radius (R e ), Sérsic index (n), axis ratio (b/a), and position angle from THE FARMER catalog.We simultaneously model all sources in the cutout up to 2.5 mag fainter and within 3″ of the target source and mask fainter sources using the segmentation map.We slightly tweak these parameters for the background sources in a few individual cases to improve the overall fit.For the PSF, we use our empirical, position-dependent PSF model that is described in Appendix A. To take into account underestimated random and systematic errors as well as residual uncertainties in the PSF, we quadratically fold in a minimum relative uncertainty on the effective radius and Sérsic index of 5%, following van der Wel et al. (2012; based on the signal-to-noise ratio (S/N) of the faintest objects in the sample).
The best-fit parameters for all ASPECS galaxies are listed in Table 1.We discard the two AGN that are best fit with a pointsource template from the structural analysis (3mm.09 and 1mm.C08, see Figure 2).From the reference sample of galaxies in the HUDF (see Section 4.2), we remove all galaxies for which GALFIT returned bad flags or parameter values at the edge of the parameter space (indicative of bad fits).These galaxies are typically at the edge of the field and/or affected by strong gradients in the background.

4.1.
Rest-frame Near-infrared Morphology at z ≈ 2.5 The MIDIS observations at 5.6 μm resolve the rest-frame NIR light of galaxies at z ≈ 2.5, around λ rest ≈ 1.6 μm, for the first time.These wavelengths trace the bulk of the (older) stellar light, that remained inaccessible with HST, and are less affected by dust attenuation than observations at shorter wavelengths.
Figure 2 shows that the observed-frame 5.6 μm morphology of the galaxies at z ≈ 2.5 is markedly different from the morphology traced at shorter wavelengths.To examine the morphological differences at z ≈ 2.5 in more detail, we compare the HST/F160W, and now higher-resolution NIRCam/F150W at similar wavelengths, to the MIRI/F560W observations in Figure 4 (more extensive multiwavelength cutouts of all sources are shown in Appendix C).We also show the higher-redshift sources at z ≈ 3.7 in the figure, where F560W probes λ rest ≈ 1.2 μm.The GALFIT models and residuals for the galaxies are shown in Figure 12 in Appendix D.
The MIRI observations reveal that all z ≈ 2.5 sources have a centrally concentrated and relatively smooth light distribution at rest-frame 1.6 μm.This is in contrast to the significantly more substructured or clumpy appearance in HST and NIRCam at restframe 450 nm.We stress the smoother morphology is intrinsic and not primarily due to the MIRI/F560W PSF (with a FWHM ≈ 0 2).This can be seen by comparing to the HST/ F160W observations, which trace roughly the same wavelength as NIRCam/F182M, but have a similar PSF size (FWHM ≈ 0 15) as MIRI/F560W.The F560W images are on the other hand not perfectly smooth and (residual) structure (after subtracting a single Sérsic component) can be seen in Figure 12.However, in all cases, the galaxies show significantly more substructure in the F160W filter than in F560W.
As an aside, we note that the high-resolution NIRCam imaging reveals some striking details in the galaxies (see Figure 11).This includes rich substructure in the X-ray AGN 1mm.C01, which suggests merger activity may be triggering the starburst and AGN activity in this system.For the X-ray AGN 3mm.09, there is extended emission in the F182M filter not present in F210M (Figure 11).This is likely [O III] λλ4960, 5008 + Hβ emission in the medium band that could originate from a gas in/outflow or gas that is ionized by the AGN at larger distances.Notable are also the pronounced obscuring dust structures (lanes) visible in several galaxies across the entire redshift range of our sample (see Figure 2).We show the F560W fluxes of the dust-continuum-selected galaxies in the context of the HUDF galaxy population in Figure 5.The ASPECS galaxies trace the bright end of the galaxy population in F560W.In addition to the flux-limited ALMA sample, we also mark the fainter galaxies in the MIDIS footprint that were detected at 1.2 mm based on an optical-FIR prior (González-López et al. 2020).These galaxies mostly correspond to galaxies that lie just below the formal blind detection threshold of ASPECS.

Structural
The GALFIT-derived structural parameters are shown in Figure 5.The effective radii of the ALMA galaxies are between R e (F560W) = 1-5 kpc, slightly decreasing toward higher redshift on average, as expected from the evolution of the mass-size relation (van der Wel et al. 2014).The ALMA galaxies trace a similar distribution in effective radius as the other galaxies from the HUDF reference sample.The Sérsic indices are shown in the rightmost panel of Figure 5. Between 2 < z < 3, where MIRI traces rest-frame λ rest ≈ 1.6 μm, most galaxies have n ≈ 1, indicating an exponential profile that is typical for star-forming disks, while a few show higher, n 2, Sérsic indices.The higher-redshift galaxy at z = 3.6 (λ rest ≈ 1.2 μm) has n ≈ 2.6.Below z = 2, the ALMA galaxies show a range of Sérsic indices, mostly between 1 and 3. Overall, the distribution of Sérsic indices follows the HUDF reference sample, with potentially a slight trend toward lower Sérsic indices.
Figure 4. 3″ × 3″ images of the gas-and dust-mass-selected galaxies at z ≈ 2.5 at λ obs ≈ 1.6 μm (HST/WFC3 and JWST/NIRCam) and 5.6 μm (JWST/MIRI) to highlight the change in morphology between the rest-frame UV/optical (λ rest ≈ 450 nm) and rest-frame NIR at 1.6 μm, which is now resolved with MIRI for the first time.The difference in morphology is not due to the PSF, which is very similar between F160W (≈0 15) and F560W (≈0 2). Figure 6.Ratio of the sizes (effective radii) at 5.6 μm and 1.6 μm vs. stellar mass, color coded by cold gas mass (derived from the cold dust emission at 1.2 mm; see Section 3.1; top) and cold gas-to-stellar mass ratio (bottom) for the ALMA galaxies in the context of the HUDF reference sample (see Figure 5).The left and right panels show the redshift ranges 1.0 < z < 2.0 and 2.0 < z < 4.0, respectively.The size ratio anticorrelates clearly with stellar mass.It also shows a potential (anti-) correlation with gas-to-stellar mass ratio (gas mass), see Figure 7.
Next, we compare the structural parameters measured at λ obs = 5.6 μm by MIRI to the measurements at λ obs = 1.6 μm.As measuring the structural parameters requires accurate knowledge of the PSF, we do not measure the structural parameters from the NIRCam imaging, but instead refer to the established measurements from HST/F160W by van der Wel et al. (2012).This excludes 1mm.C12 (AGN) and 1mm.C22 as they are not properly deblended in the HST catalog.
The average effective radius at 5.6 μm is smaller than at 1.6μ m.For the ASPECS sources, R e (F560W) is on average about 35% and 30% smaller than R e (F160W) at 1 < z < 2 and 2 < z < 4, respectively.For the full reference sample at f F560W 1 μJy, R e (F560W) is about 20% smaller than R e (F160W) at both 1 < z < 2 and 2 < z < 4. We show the ratio of effective radii as a function of stellar mass at 1 < z < 2 and 2 < z < 4 in Figure 6.We find that the ratio decreases as a function of stellar mass, such that more massive galaxies have relatively more compact light profiles at longer wavelengths.A similar trend has been shown by Suess et al. (2022) comparing shorter-wavelength NIRCam 4.4 and 1.5 μm observations for a larger sample of galaxies out to z = 2.5 (see also Chen et al. 2022;Gillman et al. 2023).The Sérsic indices are, on average, roughly a factor ≈1.5 larger in F560W than in F160W, though with significant scatter and no clear trend with stellar mass or redshift.
We also investigate correlations between the JWST/HST size ratio and quantities related to the cold gas mass (or, equivalently, the cold dust mass, see Section 3.1) in Figure 7.We find that the size ratio weakly anticorrelates with cold gas (or cold dust) mass, but more strongly correlates with the cold gas-to-stellar mass ratio, with a slope that is close to the inverse of the trend found with stellar mass.

HST versus JWST Structure and Morphology
The MIRI observations imply that the mass-weighted stellar structure (λ rest ≈ 1.6 μm) of these galaxies is significantly smoother compared to the light-weighted structure in the restframe optical frame observed with HST and NIRCam at λ rest ≈ 450 nm.The more structured and clumpy rest-frame optical morphology can be due to a combination of different factors.If the underlying true distribution of young stars and star formation (traced by HST) was as smooth as seen in the stellar mass (traced by JWST/MIRI), the structure at 450 nm could be due to a patchy distribution of attenuating dust, leading to the more clumpy morphology.On the other hand, star formation is expected to occur in spatially separated clumps, i.e., the observed HST morphology may reflect the actual underlying distribution of young stars.Matched highresolution ALMA imaging of the dust distribution will be able to measure the relative impact of the two scenarios.

HST versus JWST Sizes and Color Gradients
Most galaxies in our sample at z ≈ 2.5 show large R e (F560W) ≈ 1-5 kpc, exponential (disk-like) structures (n ≈ 1).The fact that the ASPECS galaxies do not clearly stand out in their structural parameters compared to the other sources in the HUDF reference sample (Figure 5) implies that at the depth reached by ASPECS over a relatively small field, the galaxies mostly trace the massive end of the typical galaxy population at these redshifts (see Boogaard et al. 2019;González-López et al. 2020).
The inferred rest-frame NIR sizes are smaller compared to those observed in the rest-frame optical, implying a negative color gradient (i.e., the galaxies have redder centers).A similar trend has also been observed in other galaxy populations (e.g., Chen et al. 2022;Suess et al. 2022;Gillman et al. 2023;Magnelli et al. 2023), including local face-on spiral galaxies (e.g., Casasola et al. 2017), and in studies of stellar-mass maps (e.g., Wuyts et al. 2012;van der Wel et al. 2024), and can be driven by differences in the properties of the stars, such as their age, or by dust extinction.The results can thus be attributed to the presence or formation of an older, more centrally concentrated stellar population (such as a bulge) and/or stronger dust extinction in the center (potentially linked to compact central star formation), where both lead to a "flatter" distribution in the rest-frame optical bands.The trend with stellar mass can be linked to the same effects becoming stronger in more massive star-forming galaxies.Indeed, massive galaxies are known to show overall stronger extinction and a larger fraction of obscured star formation (e.g., Garn & Best 2010;Whitaker et al. 2017) and have stronger extinction toward their centers (e.g., Nelson et al. 2016;Matharu et al. 2023).
While the rest-frame 1.6 μm emission is primarily sensitive to the old stellar light, it may also contain a contribution from an AGN, which can make the profiles look more centrally concentrated (e.g., Prieto et al. 2010).While the number of AGN in the sample is limited, we do not see a clear difference in Figure 6 between the ALMA galaxies that are identified as AGN and those that are not (note this already excludes the two AGN best fit with a point-source model).In the same vein, the MIRI emission may also trace nebular continuum and/or high equivalent width emission lines in case of very vigorous young star formation, which can have a complicated impact on the morphology (e.g., Papaderos et al. 2023).This would most strongly impact galaxies with a low stellar mass and/or high specific star formation rate.Given the comparatively smooth light distribution over large scales, this however does not appear to have a major impact on the MIRI morphology of the relatively massive gas-rich galaxies studied here.

Trends with Total Gas/Dust Mass?
It would be interesting to assess whether the color and size trends above are correlated with any other galaxy property.As dust can be responsible for some of the observed trends, we can check for the influence of the total dust (or gas) mass, even in the absence of resolved dust imaging (see Figure 6).However, both the gas mass and the gas-to-stellar-mass fraction are known to independently correlate and anticorrelate with stellar mass, respectively (e.g., Tacconi et al. 2018;Aravena et al. 2019), and larger samples are needed to distinguish potential trends.Again, we would expect a relation with the resolved dust properties.The sizes of the dust in these massive starforming galaxies are often significantly smaller than the restframe optical sizes (Tadaki et al. 2020), though typically not extremely compact as seen in submillimeter galaxies (e.g., Gullberg et al. 2019).Based on the average submillimeter-tooptical size ratios, they may even be more compact than the stellar sizes now measured with MIRI.This centrally concentrated dust emission is often linked to a compact starburst, which may be responsible for building up bulges (e.g., Nelson et al. 2019;Tadaki et al. 2020).Though note the dust may also appear more compact due to dust temperature gradients, as also shown in simulations (e.g., Cochrane et al. 2019;Popping et al. 2022).Matched high-resolution ALMA imaging, especially in multiple bands, would help to differentiate between the presence (or formation) of a bulge and/or stronger extinction in the center, especially in combination with studies of the resolved color gradients and spectral energy distributions now possible with JWST (e.g., Miller et al. 2022;Pérez-González et al. 2023).

Summary and Conclusions
We present JWST/MIRI F560W observations of the stellar structure of gas-and dust-rich galaxies in the HUDF at restframe wavelengths of λ > 1 μm using MIDIS.
We select a complete, 1.2 mm continuum-flux-limited sample of 35 galaxies from ASPECS-encompassing all sources from shallower ALMA 1 mm continuum imaging in the HUDF-of which 28 lie within the 4.7 arcmin 2 footprint of MIDIS, at z = 0.5-3.7.Using JWST slitless spectroscopy, we determine spectroscopic redshifts for all of the few galaxies in the ASPECS sample that were still missing spectroscopic confirmation, in particular in the 2 < z < 3 range.
We find reasonable agreement between the MIRI/F560W flux densities and those previously determined from (deblended) Spitzer IRAC 5.8 μm observations for the ALMA sources, though with a potential systematic offset at the bright end.Subsequently, we revisit the stellar masses by modeling the spectral energy distributions with MAGPHYS models, finding good agreement with previous determinations.The reinforcement of the stellar-mass estimates for the ALMA galaxies at z = 1-4 implies that there are no fundamental changes to the previously reported gas and dust-mass fractions (for these bright and relatively massive systems), nor to how the properties of this population of gas and dust-rich galaxies evolve with stellar mass (Aravena et al. 2019(Aravena et al. , 2020;;Boogaard et al. 2019;González-López et al. 2020).
Figure 7. Ratio of the sizes (effective radii) at 5.6 μm and 1.6 μm vs. cold gas mass (derived from the cold dust emission at 1.2 mm; top; see Section 3.1) and gas-tostellar mass ratio (bottom), color coded by stellar mass for the ALMA galaxies.Upper limits on the gas mass and gas-to-stellar mass ratio are shown for the HUDF reference sample.The left and right panels show the redshift ranges 1.0 < z < 2.0 and 2.0 < z < 4.0, respectively.The gas-to-stellar mass ratio (gas mass; anti) correlates more with size ratio, connected to the correlation with stellar mass (see Figure 6).
We find that the rest-frame NIR light distribution at λ rest ≈ 1.6 μm of the ALMA galaxies at z ≈ 2.5-that can now be resolved with MIRI-is intrinsically more smooth and centrally concentrated compared to the more substructured or clumpy appearance in the rest-frame UV/optical at λ obs ≈ 450 nm as probed by HST/F160W.This is not a resolution effect as both observations have a similar PSF.We build a custom, position-dependent, empirical PSF model for the MIDIS observations (described in Appendix A) and use it to perform a structural analysis using GALFIT.We find the galaxies at z ≈ 2.5 have effective radii of R e (F560W) = 1-5 kpc and Sérsic indices mostly close to n = 1, consistent with an exponential (disk-like) profile, up to n ≈ 5 (excluding AGN).We find average size ratios between JWST and HST of R e (F560W)/R e (F160W) ≈ 0.75 and 0.7 at z ≈ 1.5 and z ≈ 2.5, respectively, which decrease with stellar mass.Overall the mass-weighted stellar structure of the ALMA-selected galaxies is indistinguishable from a reference sample of other galaxies in the HUDF.
The results imply that a smoother stellar structure is already in place in gas-rich, star-forming galaxies at Cosmic Noon and their clumpy rest-frame optical light is likely caused by a combination of intrinsically clumpy star formation and/or patchy dust extinction.The difference in the mass-weighted radial structure of the galaxies now traced by MIRI, compared to earlier observations with HST, can be explained both by the presence of stronger dust extinction in the center, potentially linked to compact central star formation, and/or the presence or formation of older central stellar populations, such as a bulge.Future matched-resolution, spatially resolved ALMA observations will be key to measure the actual extent (size) of the dust and its resolved structure and column densities on the same scale as MIRI, and link the dust properties to the resolved optical/NIR morphology that we can now for the first time characterize with JWST.
Association of Universities for Research in Astronomy, Inc., under NASA contract NAS 5-03127 for JWST; and from the European JWST archive (eJWST)23 operated by the ESDC.The observations can be accessed via doi:10.17909/gn5v-f189.

Appendix A MIRI/F560W Point-spread Function
The MIRI PSF at 5.6 μm is affected by a number of effects, including the nonlinearity of the detector, the brighter-fatter effect, and internal diffraction that occurs inside the detector below approximately 10 μm (Gáspár et al. 2021;Argyriou et al. 2023, see Wright et al. 2023), which are are not fully captured by the current WEBBPSF models. 24While the first two are mainly relevant in the case of very bright sources-of which there are none in the MIDIS field-the latter effect can scatter photons to large distances from the core of the PSF and is very relevant for observations at 5.6 μm.The net result of the internal diffraction is a broadening of the core of the PSF and a cruciform artifact out to large radii.The shape of the cruciform artifact is dependent on the angle of incidence and varies with position across the detector, with the spikes effectively bending inwards toward the center of the detector.
For these reasons, we construct a varying PSF model for the MIDIS field making use of empirical PSFs at different positions on the MIRI detector.The MIDIS image itself contains only two bright stars, of which one is in a highly crowded region.The other bright star is shown in Figure 8, where we have masked background sources.This star is in the very north of the field and only in 9% of all exposures, close to the top of the detector (see Figure 9), and the bending of the cruciform artifact can be clearly seen for this star.Hence, we cannot use this single star as an empirical PSF for the entire field.
Instead, we create empirical PSFs by stacking stars in the LMC images from the commissioning program ID 1473 and two calibration programs (IDs 1040 and 1024).The F560W frames in each program are reduced and resampled to a pixel scale of 55 mas (i.e., oversampled by a factor of 2) using the standard JWST pipeline.Given that we want to stack many different stars, the astrometry needs to be thoroughly checked in the image, but since these are well-known regions of the sky, we found the JWST pipeline astrometry to be sufficient.We then select a total of 64 stars in the stacked frames.In this selection we avoid saturated stars and stars in very crowded regions, i.e., stars which have other sources within ∼1″.A stamp of size ∼ 16″ is cut out for each star and a local background is subtracted.Faint sources are masked in each stamp using standard source detection techniques (photutils.detection).The images are resampled keeping the native detector orientation, which enables us to separately stack three bins of stars, situated in upper, middle, and lower detector positions.By doing so we take into account the strongest PSF variation in the field that occurs over the longest axis of the detector.The total number of stars in these bins is 31, 21, and 12, respectively.We scale the star stamps to have the same flux within a radius of 2 pixels (0 11, measured using photutils) and stack the stars in each detector position bin by using median stacking (dropping the two brightest and faintest data points).The resulting three empirical PSFs still have a few defects in the outskirts of the PSF image which are manually masked (flux set to zero).We inspect S/N maps (using the standard deviation of the stacked star stamps as noise), to make sure that the final PSF model is valid.There is an insufficient number of stars for a more detailed spatially varying model using this technique, but we note that with additional calibration data, and by stacking stars from individual data frames (before resampling/stacking the full images), PSF models with higher subsampling and similar (or even greater) depth can be derived, though this is beyond the scope of the present work.The resulting stacked PSFs and their radial profiles are shown in Figure 8.These clearly show the up-and downward bending of the cruciform artifact along the long axis of the detector and the significantly broader core compared to the WEBBPSF model.The LMC-based PSF model is available at https://github.com/jensmelinder/miripsfs.
We use the empirical PSFs to create a field varying PSF for the MIDIS field.Based on the relative orientation of each of the MIDIS exposures, we determine nine areas in the MIDIS field between which the PSF varies most strongly and mark a world coordinate system (WCS) coordinate (α, δ) in each area, as shown in Figure 9.In each of these areas we insert the appropriate empirical PSF, based on the corresponding image coordinates (x, y), into each of the 96 individual exposures (i.e., WCS-aligned cal files).We then run the Stage 3 Imaging Processing twice, once with the inserted stars and once without, and recover the varying PSF model by taking the difference.To ensure the pipeline produces identical results in the different runs, we turn off all steps except the resample step.This results in nine effective PSFs, shown in Figure 9, which model the largest PSF variations across the MIDIS field.The MIDIS PSF model is available at https:// github.com/lboogaard/midis_psf.The radial profile of the MIDIS PSF is very close to the profile of the only sufficiently bright and isolated real star in the MIDIS field (HUDF north star).The star as well as the (LMC-based) input PSF model and resulting MIDIS PSF are smoother in the core region compared to the WEBBPSF models, which is likely a result of (under)sampling compared to the perfectly (over)sampled WEBBPSF models.The other panels show the stacked PSF from the Large Magellanic Cloud (LMC) commissioning data for the top-, middle-, and bottom-third of the MIRI field (16 5 × 16 5 cutouts with logarithmic scaling).Right: radial profiles of the different PSFs, compared to the WEBBPSF models (v0.9 and v1.2.1, where the later includes detector effects), on the same pixel scale and normalized to have the same total flux within a 1 1 aperture.The LMC-based PSF is smoother and has more power in the wings than both the WEBBPSF models.The PSF from the star in the north of the field and the MIDIS PSF model (#5, see Figure 9) are in good agreement.The differences in core width are likely due to the (re)sampling of the intrinsically undersampled PSFs, compared to the perfectly (over)sampled WEBBPSF model.The PSF from the star reaches low S/N at large radii (beyond 3 5).

Appendix D GALFIT Residuals
We show the GALFIT residuals for the sources in Figure 4 in Figure 12.

Figure 2 .
Figure 2. MIRI/F560W, NIRCam/F182M, and HST/F814W (RGB) cutouts of the flux-limited ALMA/ASPECS sample in the MIDIS footprint.The cutouts are ordered by decreasing redshifts and are 4″ × 4″, except the last four galaxies at z < 1, which are 8″ × 8″ (as indicated by the scale bars).See Table1for more information on the source properties.

Figure 3 .
Figure 3. JWST/MIRI F560W fluxes compared to the Spitzer/IRAC Channel 3 (5.8μm) fluxes from CANDELS (Guo et al. 2013) for galaxies at 1 < z < 4. The vertical gray dashed line shows the Spitzer 5σ flux limit.The ALMA/ ASPECS sources are shown in red.The gray band shows a 20% relative flux difference.
Parameters at z = 1-4 with MIRI and HST We analyze the structural parameters of the ALMA sources in the context of a flux-limited reference sample, consisting of all galaxies at z = 1-4 in the HUDF galaxy population covered by MIRI and ALMA that have a flux density f F560W 1.0 μJy.The redshifts for the non-ALMA sources are taken from the MUSE (Bacon et al. 2023) or else 3D-HST (Momcheva et al. 2016) catalogs.

Figure 5 .
Figure 5. ALMA galaxies in context of the galaxy population in the HUDF covered by both MIDIS and ASPECS.The panels show the MIRI/F560W flux density (left), effective radius (center), and Sérsic index (right) as a function of redshift.The black box in the left panel denotes the HUDF reference sample of galaxies with a flux density in F560W 1.0 μJy, shown in the other panels.

Figure 8 .
Figure 8.The MIRI/F560W PSF.Left: the top left panel shows a single star in the north of the MIDIS field.The other panels show the stacked PSF from the Large Magellanic Cloud (LMC) commissioning data for the top-, middle-, and bottom-third of the MIRI field (16 5 × 16 5 cutouts with logarithmic scaling).Right: radial profiles of the different PSFs, compared to the WEBBPSF models (v0.9 and v1.2.1, where the later includes detector effects), on the same pixel scale and normalized to have the same total flux within a 1 1 aperture.The LMC-based PSF is smoother and has more power in the wings than both the WEBBPSF models.The PSF from the star in the north of the field and the MIDIS PSF model (#5, see Figure9) are in good agreement.The differences in core width are likely due to the (re)sampling of the intrinsically undersampled PSFs, compared to the perfectly (over)sampled WEBBPSF model.The PSF from the star reaches low S/N at large radii (beyond 3 5).

Figure 9 .
Figure 9. Model of the MIRI/F560 PSF variation over the MIDIS field.For each of the 96 exposures, the red, green, and blue areas in the left panel indicate where the three different stacked PSFs are applicable and the white star marks the location of the star in the image (see Figure 8).The background color map shows the nine unique areas where the pipeline-processed PSF, which is created by inserting the different PSF model in each exposure at the marked coordinates, is effective.The right panels show a 16 5 × 16 5 cutout of the pipeline-processed PSFs at the nine locations.

Figure 12 .
Figure 12.GALFIT modeling results for all sources.From left to right the three panels show the F560W data, the best-fit single Sérsic model, and the residual (data − model) image, all at the same linear scale.The first 10 panels correspond to the galaxies from Figure 4.Note 1mm.C01 has an intrinsic multicomponent structure, while for 1mm.C04 the residuals are driven by the model being affected by the second component to the southwest (treating this as one galaxy is consistent with earlier work).These cutouts are the same size as those in Figure 2, but on the native orientation of the MIDIS image at which the fitting is performed, such that north points 27°.83 clockwise, as indicated by the 1″ scale bar.