{"id":"e6dbd749-2060-4a09-8ce2-002af530ce83","arxiv_id":"2511.10464","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"At 100-pc scales in 19 PHANGS galaxies, CO(2-1) correlates log-linearly with JWST PAH and dust emission, with the intercept—not slope—varying bimodally with host star-formation strength.","lead":"This paper measures how carbon monoxide gas correlates with JWST tracers of dust and PAH molecules inside 19 nearby galaxies at ~100 parsec scales. It finds the relation is a straight line in log-log space, but that its vertical offset—not its slope—varies between galaxies in a way the authors tie to how strongly each galaxy is forming stars.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The claimed 'clear bimodality' of the intercept b rests on visual inspection of 19 galaxies and an arbitrary split at b=0, with no formal test that the distribution is actually bimodal.","rationale":"The reader identified the uncertainty-rescaling heuristic as the weakest assumption, but the paper already performs robustness checks on that heuristic and reports consistent key results. More fundamental is the lack of any formal test for bimodality—the central claim. The reader's rationale does mention the 'arbitrary split at b_KS=0' and 'only 2-sigma correlations,' so there is overlap, but the reader's stated weakest_assumption is about the noise model rather than the statistical significance of the bimodality itself. I agree with the CONDITIONAL verdict: the paper presents valuable data and a new regression tool, but the headline claim needs stronger statistical support before it can be accepted as established. The concrete test proposed would either validate the bimodality or reveal it as an artifact of visual classification. Thus, the verdict should remain CONDITIONAL, pending such a test.","tokens_in":33485,"tokens_out":4823,"duration_ms":49128,"concrete_test":"Apply Hartigan's dip test to the 19 b_KS values for F770W_PAH in the 'all' ionization condition (posterior medians, with weights reflecting uncertainties if available). Also fit a one-component vs. two-component Gaussian mixture model to these b_KS values and compare via BIC or AIC. If the dip-test p-value is >0.05 or the BIC favors a single Gaussian, the 'clear bimodality' claim is not statistically established. To further test the arbitrary b=0 threshold, repeat the high-b/low-b classification using a data-driven split (e.g., k-means or Gaussian mixture) and check whether the sSFR difference between the groups remains significant.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central novel claim is that the intercept b_KS of the CO–MIR scaling relation is bimodal and that this bimodality is tied to host-galaxy sSFR. This underpins the abstract's assertion that the CO-to-PAH/dust conversion normalization is not single-valued at ~100 pc. However, the paper establishes bimodality only by plotting b_KS values and visually splitting at b_KS=0 in Figure 4. No quantitative test for multimodality is presented. With only 19 galaxies, the apparent gap between 'high-b' and 'low-b' galaxies could be a sampling fluctuation of a continuous distribution. The authors themselves concede in §4.2.2 that 'it is also rather difficult to explain why it is a b_KS bimodality rather than a continuum.' Moreover, the relationship to sSFR is only at the 2σ level and primarily reflects the chosen split, as the paper notes within each subgroup the correlation weakens significantly. If the bimodality is not formally significant, the headline claim that the normalization is bimodally tied to sSFR collapses. This is more load-bearing than the uncertainty-rescaling heuristic, because even if the regression parameters are unbiased, the interpretation of their distribution as bimodal remains unsupported.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper applies the raddest regression method (Jing & Li 2025) to PSF-matched PHANGS-ALMA CO(2-1) and PHANGS-JWST MIR maps of 19 star-forming galaxies, fitting log-log-linear relations between I_CO and I_F770W,PAH, I_F1130W, and I_F2100W at ≈100 pc scales, separately for HII-, composite-, and AGN-like ionization conditions. The authors report that the fitted slope and intrinsic scatter vary modestly across galaxies, while the intercept b_KS varies strongly and appears bimodal (high-b versus low-b, split at b_KS=0). This bimodality is argued to be related to the host galaxy's overall star formation strength (sSFR/SFE). The paper also quantifies deviations from log-linearity in the brightest regions, mainly as slope flattening, and studies how k_KS, b_KS, σ_KS depend on spatial scale. The manuscript provides extensive parameter tables and a public implementation of the fitting code.","tokens_in":33848,"tokens_out":3004,"duration_ms":33586,"significance":"If the central claims hold, the paper would establish that the CO-to-PAH/dust conversion at ~100 pc is not single-valued: the slope is approximately universal, but the normalization is bimodally tied to host-galaxy sSFR, and previously reported sublinear F2100W slopes are a regression artifact. This would be an important result for interpreting cloud-scale CO/MIR tracers. The paper's strengths include the use of a regression method with external mock-data validation, public code, detailed comparison with previous PHANGS work (Leroy et al. 2023b; Chown et al. 2025), and a large set of fitted parameters per galaxy and ionization condition. However, the headline bimodality claim currently rests on visual inspection of 19 b_KS values with an arbitrary split at b_KS=0 and no formal multimodality test; this is a load-bearing issue that needs to be resolved before the central claim can be accepted.","major_comments":[{"comment":"The claim of a 'clear bimodality' in b_KS is not supported by any quantitative test. The high-b/low-b classification is made by a visual split at b_KS=0 in the F770W,PAH 'all' panel, and with only 19 galaxies the apparent gap can be a sampling fluctuation of a continuous distribution. No Hartigan dip test, Gaussian mixture comparison, or other multimodality test is presented. Because the abstract and §4.2.2 build the central interpretation on this bimodality, please add a formal test and report its significance, and show the sensitivity of the classification to the threshold choice. The authors' own admission in §4.2.2 that it is 'rather difficult to explain why it is a b_KS bimodality rather than a continuum' strengthens the need for such a test.","section":"§3.2, Figure 4"},{"comment":"The connection between b_KS and sSFR is presented as supporting the bimodality interpretation, but the reported correlations are only at the 2σ level, and no 3σ correlations are found between any global property and k_KS, b_KS, or σ_KS. Moreover, because the high-b/low-b split is defined directly from b_KS, the statement that the b_KS–sSFR correlation 'primarily arises from the bimodality' is close to a restatement of the chosen split. Please report the b_KS–sSFR correlation on the full sample without subgrouping, and quantify whether a continuous correlation model is actually disfavored relative to the two-group model.","section":"§3.2.1, Figure 6 and Figure 7"},{"comment":"The heuristic rescaling of x_err or y_err for ~10% of normalizing-flow cases is a potential source of bias in the fitted parameters, including b_KS. The authors state that 'all key results' remain consistent when restricting to non-problematic data or including all data without correction, but no such comparison is shown. Since the bimodality claim depends on the exact b_KS values, please provide the requested robustness test (e.g., a comparison of k_KS, b_KS, σ_KS with and without rescaling) or a mock-based validation of the rescaling procedure itself.","section":"§2.3"}],"minor_comments":[{"comment":"The abstract and §1 use 'log-log linear relations', while §2.2 and Equation (1) describe a 'log-linear' relation. Please make the terminology consistent (the model is linear in log I_CO vs. log I_X).","section":"Abstract and §2.2"},{"comment":"The last column 'high-b' contains 'Yes'/'No' but no definition or pointer to Figure 4. Please add a note explaining the criterion and that it is based on F770W,PAH, all ionization conditions.","section":"Table 1"},{"comment":"The caption says 'the contour showcase the distribution' and 'the bold line is corresponding best-fit result.' These should be 'contours show' and 'the bold line is the corresponding best-fit result.' Also check for missing articles elsewhere in the text.","section":"Figure 2 caption"},{"comment":"There are several typos in this subsection: 'radiation filed' should be 'radiation field', 'cloud coexist' should be 'could coexist', and 'Hiiregions' should be 'HII regions'. Similar spacing issues occur for 'low-b' and 'high-b' in several places.","section":"§4.2.2"},{"comment":"The text refers to 'low-band high-b galaxies' and 'low-b and high-b galaxies' in a way that is easy to confuse with the MIR bands. Consider using 'low-b/high-b subsamples' throughout and reserving 'band' for F770W,PAH, F1130W, and F2100W.","section":"§3.4 and Figure 11"}],"recommendation":"major_revision","confidential_remarks":"The paper contains a large amount of useful, reproducible analysis and the methodological comparison is valuable. The main risk is that the headline 'clear bimodality' is not statistically established; if formal multimodality tests fail or are inconclusive, the paper would need to be substantially reframed. The authors' own caveats in §4.2.2 and the Summary suggest they are aware of this fragility. I would encourage a revision that either provides a rigorous test of bimodality and its relation to sSFR or downgrades the claim to 'indications of possible bimodality' with appropriate caveats."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things to know. First, this paper does a genuinely better job than its predecessors at measuring the CO–MIR scaling relations at ~100 pc: it uses raddest, a public regression code that demonstrably recovers slopes under heteroscedastic errors and outliers, and it makes a credible case that the sublinear CO–F2100W slope reported by Leroy et al. (2023b) is an artifact of mODR. Second, the abstract over-sells the intercept bimodality: that claim is supported by eye-balling 19 galaxies and a split at b_KS=0, with no formal multimodality test, and the authors admit in §4.2.2 that they cannot explain why it would be a bimodality rather than a continuum.\n\nWhat is actually new and useful: the ionization-condition-separated fits (HII vs composite vs AGN) are a real step forward and show a clean ordering in slopes; the scale-dependence analysis (k, b, sigma vs resolution) is interesting, especially the disappearance of the b split at low resolution; the non-log-linear deviation statistics (Delta_KS, q0, R_Delta k) give a quantitative handle on flattening in the brightest regions; and the explicit comparison to Chown et al. (2025) using mLINMIX on the same sample isolates method from sample differences. The tables are extensive and the code is public, so the analysis is reproducible in principle.\n\nSoft spots, in order. The bimodality is the biggest one. It is the headline result, but it is a visual taxonomy rather than a fitted mixture; the sSFR connection is only 2-sigma and mostly appears between the two groups, not within them. A referee should ask for a dip test, a two-component Gaussian fit, or an equivalent formal test, or else the abstract should say 'a group of galaxies with b>0 and a group with b<0' rather than 'clear bimodality.' The uncertainty-rescaling heuristic in §2.3 (scaling x_err/y_err until NFs pass the 2D KS test in ~10% of cases) is a bit hand-wavy; the authors say the results hold without it, but the evidence for that is a one-line assertion. The visual classification of deviations into cases (a)-(c) is subjective, though the quantitative descriptors mitigate this.\n\nThe paper is honest, well-written, and the central regression result (superlinear F2100W slope) holds up. It deserves a serious referee. I would accept it with major/minor revisions: tone down or formally test the bimodality, and expand the uncertainty-rescaling robustness description. Worth citing for the CO-MIR slope work.","headline":"Solid statistical study that credibly revises the CO–F2100W slope, but the headline 'clear bimodality' of the intercept is visually suggested, not formally demonstrated.","tokens_in":34348,"tokens_out":3874,"would_cite":true,"duration_ms":35097,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"CO and mid-infrared emission follow two distinct scaling families at 100-pc scales, set by the host galaxy's star-formation rate.","keywords":["CO(2-1)","PAH emission","dust emission","scaling relations","interstellar medium","JWST","ALMA","regression with uncertainties"],"falsifier":"Take the same 19 galaxies and re-fit with a regression that does not rely on the KS-test-based likelihood or on rescaled errors (e.g., a fully Bayesian model with explicit outlier component), and check whether the bimodality in b survives at >3σ; or apply the same raddest pipeline to an independent set of 19 star-forming galaxies and ask whether the b values fall into the same two clusters with the same sSFR separation.","tokens_in":33353,"feed_emoji":"🔭","tokens_out":4513,"duration_ms":42730,"temperature":0.7,"pith_summary":"This paper claims that on ~100 pc scales, the relation between CO(2-1) emission and mid-infrared emission from PAHs and dust is a log-log linear relation with a nearly universal slope, but the normalization (intercept) is not single-valued: it splits galaxies into two families with higher or lower intercept. The split tracks the host galaxy's overall star-formation rate, with stronger star-formers showing lower intercepts. The paper further argues that the previously reported sublinear slope for the 21-micron dust band is an artifact of the regression method used, and that the true slope is similar to the PAH bands. If right, this changes how CO-to-dust conversion is calibrated at cloud scales and points to a discrete physical switch in the interstellar medium.","feed_headline":"Two distinct CO–mid-IR scaling families at 100 pc","feed_subtitle":"The normalization of the CO–PAH/dust relation splits with host star-formation rate, not the slope.","key_machinery":"The central object is the log-log linear scaling relation log I_CO = k log I_X + b + ε, with ε Gaussian intrinsic scatter, fitted by raddest, a regression method that builds a generative model of the observed fluxes and uncertainties using normalizing flows and then estimates k, b, σ through likelihood or a KS-test-based goodness-of-fit. The intercept b carries the paper's main result: its bimodality, not the slope, encodes the physical state separating the two galaxy families.","core_discovery":"Applying a regression technique that explicitly handles heteroscedastic uncertainties and outliers (raddest) to 19 nearby star-forming galaxies, the paper finds that log I_CO versus log I_MIR (F770W, F1130W, F2100W) is well described by a single straight line for the majority of spaxels at ≤100 pc scales, with slopes consistently superlinear and similar across bands. The dominant galaxy-to-galaxy variation is in the intercept b, which shows a clear bimodal distribution; the two groups (high-b and low-b) have similar slopes but different overall CO-to-MIR ratios and different intrinsic scatter. The bimodality is tied to the normalized star-formation rate of the host galaxy: galaxies with high","pith_inferences":["The two-family split might reflect a threshold in the diffuse UV background intensity that enhances PAH/dust emissivity without destroying the carriers; if so, the boundary between low-b and high-b groups should correlate with a measurable jump in PAH-to-dust ratio or dust temperature, which the paper does not test directly.","A direct extension would be to apply the same fitting technique to the full set of galaxies in the same survey; the bimodality should persist and the group separation should sharpen with better statistics.","The method's reliance on normalizing flows for the intrinsic distribution of the independent variable could be probed by rerunning the analysis with a simpler parametric model; if the bimodality remains, the result is not an artifact of the density estimator."],"forward_implications":["If the intercept bimodality is real, CO-to-dust and CO-to-PAH conversion factors at ~100 pc are not single-valued; using one calibration for all star-forming galaxies would bias molecular gas masses by the factor corresponding to the intercept offset.","The previously published sublinear CO–F2100W slope is attributed to the fitting method rather than to astrophysics; re-analysis of those data with uncertainty-aware regression should recover the same slopes as the PAH bands.","Because the bimodality disappears at spatial scales above ~100 pc, measurements at kpc scales will miss the dichotomy—this predicts that resolved studies at hundreds of pc are necessary to see it.","The flattening of the relation in bright regions is systematic and stronger for the dust band, which suggests that PAH and dust emission are enhanced relative to CO in the most intense radiation environments."],"fun_headline_variants":["CO–PAH scaling splits by star-formation rate at 100 pc","Two CO–MIR scaling families: high and low star formation","Bimodal CO–MIR scaling: host SFR sets the offset","Same CO–PAH slope, two normalizations set by galaxy SFR","CO–MIR intercept bimodality traces host star-formation rate"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The fitted parameters—and thus the claimed bimodality—assume that the raddest regression likelihood (including its normalizing-flow model of the true flux distributions and its treatment of uncertainties) is unbiased; for ~10% of datasets the uncertainties had to be rescaled by hand to pass a 2D KS test, and if that heuristic is wrong, the intercept split could be an artifact.","fun_headline_variants_meta":{"raw":{"variants":["CO–PAH scaling splits by star-formation rate at 100 pc","Two CO–MIR scaling families: high and low star formation","Bimodal CO–MIR scaling: host SFR sets the offset","Same CO–PAH slope, two normalizations set by galaxy SFR","CO–MIR intercept bimodality traces host star-formation rate"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000898,"raw_usage":{"total_tokens":3818,"prompt_tokens":968,"completion_tokens":2850,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":712,"completion_tokens_details":{"reasoning_tokens":2752}},"tokens_in":712,"tokens_out":2850,"duration_ms":18101,"temperature":1.0,"reasoning_tokens":2752,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-04T06:46:25.681523+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take the same 19 galaxies and re-fit with a regression that does not rely on the KS-test-based likelihood or on rescaled errors (e.g., a fully Bayesian model with explicit outlier component), and check whether the bimodality in b survives at >3σ; or apply the same raddest pipeline to an independent set of 19 star-forming galaxies and ask whether the b values fall into the same two clusters with the same sSFR separation.","supporting_citations":[],"review_version":1}