Critique of "Three-dimensional, grain-aware magnetic modeling of polarized dust for CMB B-mode cleaning"

1. Factual / physics errors

Wrong attribution of the key simulation result. The draft cites "Sponseller & Kogut 2023" and links MNRAS 519, 4370. That paper is McBride, Bull and Hensley 2023, "Characterizing line-of-sight variability of polarized dust emission with future CMB experiments". Everything the draft summarizes from it, namely that a single MBB "can remain approximately unbiased in the absence of polarization-angle mixing" and fails once line-of-sight depolarization with misaligned fields enters, is McBride et al., obtained for a PICO-like 21-band setup. Sponseller and Kogut is a different paper, ApJ 936, 8 (2022), which estimated line-of-sight dispersions from Planck temperature maps, about 1.6 K in T and 0.045 in β for 30 degree disks, and argued that the resulting single-MBB bias needs a moment expansion. The two papers pull in different directions, and the draft has fused them under one wrong name.
Wrong citation for the complex moments. The "complex moment expansion" with moments like Σ A_i(β_i − β̄)e^{2iψ_i} is the spin-moment formalism of Vacher et al. 2023, A&A 669, A5. Vacher et al. 2022 expands cross-spectra in β and T for LiteBIRD and contains no complex moments. The draft also says the method was "developed by Mangilli et al. and Vacher et al."; it originates with Chluba, Hill and Abitbol 2017.
The grain-size section contradicts itself. It lists "which grain sizes dominate total and polarized emission" and "the frequency dependence of p(ν)" as things dn/da determines, then concedes that at CMB frequencies grains are in the Rayleigh regime. In that regime the absorption cross-section per unit volume is size-independent, so the millimetre SED of a given composition is insensitive to dn/da except through the weak size dependence of equilibrium temperature and the aligned mass fraction. "Stochastic temperatures" are irrelevant here: stochastically heated grains contribute nothing at ν < 400 GHz and are not aligned.
"Most tracers constrain only the plane-of-sky orientation, not field inclination" ignores Faraday rotation, Zeeman splitting and the polarization-fraction route via p/p_max. The correct and stronger statement is that the line-of-sight tracers sample ionized or dense gas rather than the diffuse neutral gas carrying the polarized dust.
A null result presented without its sensitivity. The draft says Planck 2018 XI "found no significant broad-sky decorrelation at its sensitivity" but never states that Planck itself frames that limit as adequate only for r ≈ 0.01 at the recombination peak. For the r ≈ 0.001 targets that motivate the essay, the Planck limit is not a constraint.
"Vector radiative-transfer model" is a misnomer. The model is an optically thin emission sum with no transfer, and the phrase implies a rigor the proposal does not have.

2. Missing physics

The σ_β to decorrelation mapping. The community's "spectral-index decorrelation" is the Gaussian-dispersion model in Planck Int. L and Vansyngel et al. 2017, ρ(ν1,ν2) = exp[−½ σ_β² ln²(ν1/ν2)], and BICEP/Keck's Δ_d with its ℓ and frequency scalings. The draft calls the phrase imprecise instead of giving the model. Without it the essay cannot say what "too much" variation is. A specialist would immediately note that Pelgrims' 15% cloud-to-cloud 353/217 ratio variation corresponds to Δβ ≈ 0.3 between clouds, far above the 0.045 dispersion Sponseller and Kogut inferred from temperature maps, and would demand that the draft reconcile the two.
Map-level versus spectrum-level decorrelation. The centrepiece is line-of-sight angle rotation. But the quantity Planck and BICEP/Keck measure, the BB correlation ratio, drops for any across-sky variation of the effective SED with no misaligned layers at all. The draft mentions this in one sentence and never says which ingredient addresses which channel. Three-dimensional maps and field tomography target the line-of-sight channel; the across-sky channel needs per-pixel SED freedom or moments.
Empirical evidence that layers have different SEDs. Planck Early XXIV (2011) and Planck Int. XVII (2014) decomposed high-latitude dust emission by H I velocity component and found intermediate-velocity dust warmer with lower emissivity per hydrogen atom. This is the observational foundation of the whole layer model and is absent.
The Local Bubble. Above |b| ≈ 30 degrees the dominant polarized layer is the Local Bubble shell. Skalidis and Pelgrims 2019 quantified its contribution, and Pelgrims et al. 2020 fitted a shell magnetic-field model to Planck 353 GHz using exactly the 3D extinction maps the draft discusses. Planck Int. XLIV's mean field toward the southern cap, at l ≈ 70°, b ≈ 24°, and its N-layer turbulent model with turbulent-to-mean ratio 0.9 ± 0.1 are the existing multi-layer description of the high-latitude sky. None of this appears, so the draft's "identify low-risk fields dominated by a single nearby structure" reads as if that work had not been done.
The bridge from starlight polarization to emission. Planck Int. XXI measured P_353/p_V ≈ 5.4 MJy/sr, and Fanciullo et al. 2017 studied its scatter. Without this ratio and its scatter, stellar tomography cannot set a per-layer polarized amplitude at 353 GHz. The draft never mentions it.
Post-Planck grain constraints. Planck found β_P > β_I, so p falls toward millimetre wavelengths. Planck Int. XXII and Hensley and Draine 2021 used this to disfavor silicate-only alignment, leading to the single-composition astrodust model of Hensley and Draine 2023. The draft cites Draine and Fraisse 2009 and Guillet 2018 as if the model family were still open. Radiative-torque alignment, which ties R(a) to grain size and radiation field, is also absent although it is the only physical link between dn/da, the 3D radiation field and polarization efficiency.
Large-area detection of frequency-dependent angles. Ritacco et al. 2023 found spatial variation of the polarized SED and a frequency dependence of polarization angles in Planck HFI residual maps over large sky areas, with a direct effect on the E/B decomposition. The draft claims detection only "in selected environments".
H I-based 3D Stokes modeling. Clark and Hensley 2019 built velocity-resolved Q and U maps from H I with no free parameters, correlating with Planck 353 GHz at r > 0.75 and reproducing the E/B asymmetry. Panopoulou and Lenz 2020 mapped the number of clouds per sightline, and Lenz, Hensley and Doré 2017 showed H I traces reddening below 4×10²⁰ cm⁻². The draft's "magnetically aligned H I structures" has no citation and no quantification.
Current 3D maps and footprints. Edenhofer et al. 2024 and Vergely et al. 2022 are the relevant all-sky products. Bayestar19, the draft's example, stops at declination −30 degrees, which excludes the BICEP/Keck and SPT fields.
Ground-based decorrelation limits. BICEP/Keck 2018 fits Δ_d with no external dust priors and finds no evidence for decorrelation at σ(r) = 0.009. Missing.
ℓ-dependence. The residual enters as BB power at ℓ ≈ 80 and, for LiteBIRD, at ℓ < 10. The proposal is per-pixel and says nothing about how modeling errors map into ℓ space.
Prior art on physical-model mismatch. Hensley and Bull 2018 tested physical dust models against MBB fits; PySM's d12 is the Martínez-Solaeche model in operational form; Hervías-Caimapo and Huffenberger 2022 built filament-based 3D models with line-of-sight decorrelation.

3. Overclaims

"Replace an uncontrolled, phenomenological decorrelation parameter with a physically structured ... model whose degrees of freedom can be constrained, validated and marginalized." No published analysis has shown a reduction in r bias from including Gaia or starlight tomography, and the draft offers no forecast.
"Prevent one cloud's fitted β_d from being interpreted as a property of the entire column." Microwave data can assign an SED per layer only if the layers have distinguishable angular morphologies at the scales that matter. A 3D extinction map supplies templates, not SEDs. Identifiability is asserted, never demonstrated.
"Magnetic tomography constrains their complex phases." The phase of Σ A_i(β_i − β̄)e^{2iψ_i} depends on the SED deviations weighted by the angles. Knowing ψ_i fixes nothing unless the β_i − β̄ are also known.
"Identify low-risk fields dominated by a single nearby structure." A single structure is not a single SED. The Local Bubble wall is thick, its field has a turbulent component comparable to the mean, and its temperature varies with the local radiation field. Planck Int. L's contested trend was that decorrelation increased at lower column density.
"Priors on inclination." No source is given. For diffuse neutral gas γ_i is essentially unconstrained.
"Tested against frequencies excluded from the fit." Simons Observatory and CMB-S4 have two to four dust-dominated bands and LiteBIRD about six. Holding one out removes the constraint you are trying to validate.
The framing treats "3D maps of thermal dust emission" as existing products. The draft concedes they are extinction maps but never states what constructing an emission map would require: 3D extinction, a 3D radiation field, and a per-cloud emissivity and temperature.

4. Weak links in the pipeline

3D map to layer amplitudes. At high latitude a cloud contributes E(B−V) ≈ 0.01 mag, below the per-star precision of Bayestar. Edenhofer's map excludes the inner 69 pc and degrades with distance. Converting a layer's E(B−V) to 353 GHz needs τ_353/E(B−V) and a temperature per layer, neither of which is in the map.
Grain size to β_P. Rayleigh-limit emission makes β a composition and porosity property. The one observable that ties size to polarization, the Serkowski λ_max, needs reddened stars and is unmeasurable in the low-column fields the proposal targets.
Projected field to per-layer angles. The 353 GHz angle is integrated over the sightline. Starlight gives per-layer angles only if the star density per beam and the polarimetric precision suffice, with p_V ≲ 0.3% at E(B−V) = 0.02. H I gives angles per velocity component, not per distance, and misses dark gas. None of the tracers gives inclination, and extinction-weighted and emission-weighted angles differ once layers have different temperatures.
Layer SEDs. Even with perfect A_i and ψ_i, each layer's S^P_bi must come from the microwave data. That is at least two SED parameters plus p_i per layer per pixel. With Simons Observatory-like bands the layer SEDs are entirely prior-driven, and the draft admits incorrect priors "can themselves bias r" without testing it.
Grain-model priors. Guillet et al. 2018 showed the composition, shape, porosity and alignment degeneracies are strong. A principal-component library inherits them, and the draft has no misspecification test.
Moment compression. The pivot β̄ and the truncation order set the bias. The 3D data do not tell you the order, and McBride et al. show the temperature-and-β-only failure mode gives equally good fits for right and wrong models, so goodness of fit cannot select the order either.

5. Referee demands

Fix the citations: McBride, Bull and Hensley 2023 for MNRAS 519, 4370; Sponseller and Kogut 2022, ApJ 936, 8, for the Planck-derived dispersions; Vacher et al. 2023, A&A 669, A5, for spin-moments; Chluba et al. 2017 for the moment expansion.
Add the Gaussian-dispersion decorrelation formula and the BICEP/Keck Δ_d parametrization, then compute the expected ρ(150×353) from σ_β ≈ 0.045, σ_T ≈ 1.6 K, and separately from Pelgrims' 15% cloud-to-cloud ratio with the Panopoulou and Lenz cloud statistics. Propagate each to δr for a BICEP/Keck-like patch and for LiteBIRD.
State the accuracy target: residual BB power at ℓ ≈ 80 relative to the r = 0.001 tensor signal in the target fields, in μK².
Count degrees of freedom per pixel against bands for Simons Observatory, CMB-S4 and LiteBIRD, and state which layer parameters are prior-only in each case.
Demonstrate identifiability: show that Gaia or H I layer templates have distinct morphologies at ℓ = 30 to 300 and recover per-layer SEDs in simulations with realistic noise.
Quantify tomographic feasibility in the actual fields: footprint coverage, stars per beam per distance bin, E(B−V) per layer, polarimetric precision, and the resulting uncertainty on each ψ_i.
Specify the emission-to-extinction bridge with its scatter and its environmental dependence, and how the per-layer polarized amplitude prior is set from it.
Separate the line-of-sight and across-sky decorrelation channels and state which ingredient constrains which.
Address inclination explicitly: state it is unconstrained, adopt p_max-based priors or rotation-measure priors where available, and show the effect of marginalizing over it.
Run a prior-misspecification test with deliberately wrong grain-model priors and report the r bias.
Validate against Ritacco et al. 2023's frequency-dependent angle maps and against the Clark and Hensley E/B ratio, not only against EE, BB, EB and TB power spectra.
Add the Local Bubble shell as the baseline single layer for |b| > 30 degrees and show what the multi-layer model adds beyond it.
Replace "radiative-transfer" and define "3D thermal dust emission map" honestly, with the construction steps it requires.

6. Severity list

blocker No quantitative link from the three data products to residual B-mode power or r bias; no forecast, no target.
blocker Identifiability of per-layer SEDs from microwave data is assumed, never shown; without distinct morphologies the 3D map adds nothing to the SED problem.
blocker The starlight-to-emission bridge is absent, so stellar tomography cannot set layer polarized amplitudes as the strategy requires.
major MNRAS 519, 4370 attributed to Sponseller and Kogut instead of McBride, Bull and Hensley; the two papers' conclusions are conflated.
major Complex moments cited to Vacher 2022 rather than Vacher 2023; moment expansion origin misattributed.
major Line-of-sight and across-sky decorrelation conflated; no allocation of which ingredient fixes which.
major Grain-size section internally inconsistent with the Rayleigh limit, and silent on post-Planck discrimination among grain models.
major Local Bubble shell, Planck Int. XLIV layer model, and Planck Early XXIV per-component SEDs all missing.
major Ritacco 2023 large-area angle detection and BICEP/Keck Δ_d limits omitted; Planck null result quoted without its r ≈ 0.01 sensitivity.
major Bayestar19 offered as the example map although it excludes the southern B-mode fields.
major Inclination unconstrained yet invoked as a prior; degrees-of-freedom accounting absent.
major No ℓ-space treatment of residuals.
minor "Radiative-transfer" misnomer; stochastic heating irrelevant at these frequencies.
minor Faraday rotation and Zeeman omitted from the line-of-sight field discussion.
minor PASIPHAE survey paper, Clark and Hensley 2019, Panopoulou and Lenz 2020, Edenhofer 2024, Hensley and Bull 2018 uncited.

Sources checked: Pelgrims et al. 2021, McBride, Bull & Hensley 2023, Sponseller & Kogut 2022, Fanciullo et al. 2017, Vacher et al. 2022, Vacher et al. 2023 E/B, Planck 2018 XI, Planck Int. L, Ritacco et al. 2023, BICEP/Keck XIII, Hensley & Draine 2021, Planck Int. XLIV, Pelgrims et al. 2020, Clark & Hensley 2019.