Dust Emission Polarization Spectrum

• Dust emission polarization spectrum is the wavelength-dependent variation of dust polarization measured across different astrophysical environments, defining both the spectral energy distribution and angular power spectrum.
• It reveals key insights into dust grain properties, alignment mechanisms, and magnetic field geometries, which are critical for effective CMB foreground subtraction.
• Practical studies span from microwave to submillimeter regimes, informing models for diffuse ISM, molecular clouds, protoplanetary disks, and low-frequency spinning dust.

Dust emission polarization spectrum denotes the dependence of polarized dust emission on wavelength or frequency, usually expressed through polarized intensity $P(\lambda)$ or fractional polarization $p(\lambda)$, and, in CMB foreground studies, also through the angular power spectra $C_\ell^{EE}$ and $C_\ell^{BB}$ of polarized dust maps [(Ashton et al., 2017); (Collaboration et al., 2014)]. Across astrophysical environments, the term therefore covers two related but distinct observables: the polarized spectral energy distribution (SED), often modeled with a modified blackbody in thermal-emission regimes, and the scale dependence of polarization fluctuations on the sky. In the diffuse interstellar medium and at intermediate and high Galactic latitude, polarized thermal dust is the dominant foreground above $100\,\mathrm{GHz}$, and its spectrum is central to CMB $B$-mode foreground subtraction [(Collaboration et al., 2014); (Collaboration et al., 2018)]. In molecular clouds and star-forming regions, the shape of $p(\lambda)$ constrains grain alignment, temperature structure, and radiative transfer (Ashton et al., 2017, Shariff et al., 2018, Cox et al., 16 Sep 2025). In circumstellar disks, the polarization spectrum is additionally mechanism-dependent, because self-scattering, thermal emission from aligned grains, and dichroic extinction can dominate at different wavelengths (Hull et al., 2018, Lin et al., 2021).

1. Diffuse Galactic dust and the polarized modified-blackbody spectrum

In the diffuse ISM, Planck and WMAP established that the polarized dust SED from microwave to submillimeter frequencies is well fit by a single-temperature modified blackbody over the thermal-dust-dominated range, with $T_d=19.6\,\mathrm{K}$ and a polarization spectral index close to $\beta_d^P \simeq 1.53$–1.59 depending on the analysis and data release [(Collaboration et al., 2018); (Collaboration et al., 2014)]. In the Planck 2018 foreground analysis, the SED of polarized dust emission was fit well by a single-temperature modified blackbody emission law from 353 GHz to below 70 GHz, with mean spectral index $\beta_{\rm d}^{P} = 1.53\pm0.02$ for $p(\lambda)$0 (Collaboration et al., 2018). In the earlier cross-correlation analysis with Planck and WMAP, the mean values were $p(\lambda)$1 for polarization and $p(\lambda)$2 for intensity, for a mean dust temperature of 19.6 K, with the difference quoted as $p(\lambda)$3 (Collaboration et al., 2014).

The standard thermal-dust form is

$p(\lambda)$4

and the same functional form describes the polarization SED over the high-frequency regime in these studies [(Collaboration et al., 2014); (Collaboration et al., 2018)]. Planck Intermediate Results XXII also found that the polarization fraction of the dust emission decreases by $p(\lambda)$5 from 353 to 70 GHz (Collaboration et al., 2014). The paper explicitly noted that this decrease could indicate differences in polarization efficiency among components of interstellar dust, for example carbon versus silicate grains (Collaboration et al., 2014). This suggests that even when intensity and polarization are each fit by an MBB, the aligned-grain subset need not be spectrally identical to the total-emitting population.

Below $p(\lambda)$6, the mean SED of the microwave emission correlated with the 353 GHz dust templates rises for decreasing frequencies in both intensity and polarization (Collaboration et al., 2014). In polarization, that low-frequency rise was interpreted as consistent with a synchrotron component correlated with dust, with no need for any polarization of the anomalous microwave emission (Collaboration et al., 2014). This is important because it separates the thermal-dust polarization spectrum from the broader dust-correlated microwave foreground spectrum.

A complementary single-frequency anchor was provided by the comparison of Planck 353 GHz polarized emission with visible interstellar polarization. For diffuse sightlines, the measured ratios were

$p(\lambda)$7

with statistical and systematic uncertainties $p(\lambda)$8 and $p(\lambda)$9, and

$C_\ell^{EE}$0

with uncertainties $C_\ell^{EE}$1 and $C_\ell^{EE}$2 (Collaboration et al., 2014). The estimate of $C_\ell^{EE}$3 was compatible with diffuse-ISM dust models, but $C_\ell^{EE}$4 was not, being too high by a factor of about 2.5 relative to then-current predictions (Collaboration et al., 2014). A plausible implication is that the optical properties of the aligned grain population, especially its submillimeter emissivity, require revision even when the broad spectral shape is acceptable.

2. Angular power spectra, $C_\ell^{EE}$5 asymmetry, and CMB foreground implications

For CMB foreground studies, the dust emission polarization spectrum is also an angular spectrum. Planck measured the polarized thermal dust angular power spectra $C_\ell^{EE}$6 and $C_\ell^{EE}$7 over large high-latitude sky fractions and found that they are well described by power laws in multipole (Collaboration et al., 2014). Using the rescaled form

$C_\ell^{EE}$8

the diffuse dust spectra satisfy

$C_\ell^{EE}$9

with fitted slopes at 353 GHz of

$C_\ell^{BB}$0

and adopted common mean slope

$C_\ell^{BB}$1

over $C_\ell^{BB}$2 (Collaboration et al., 2014). Planck 2018 XI extended this picture to lower multipoles, reported statistically significant variations of the exponents over sky regions, and found for the largest sky region LR71

$C_\ell^{BB}$3

with weighted means $C_\ell^{BB}$4 and $C_\ell^{BB}$5 across the six large retained regions (Collaboration et al., 2018).

A major empirical result is the systematic asymmetry between dust $C_\ell^{BB}$6- and $C_\ell^{BB}$7-modes. Planck Intermediate Results XXX found

$C_\ell^{BB}$8

summarized in the abstract as $C_\ell^{BB}$9 (Collaboration et al., 2014). Planck 2018 XI reported the corresponding weighted mean as

$100\,\mathrm{GHz}$0

(Collaboration et al., 2018). The 2014 analysis stressed that this was unexpected and not reproduced by then-current template models, implying that the geometry of interstellar filaments and Galactic magnetic fields is more ordered and anisotropic than simple equal-$100\,\mathrm{GHz}$1/$100\,\mathrm{GHz}$2 dust models (Collaboration et al., 2014).

The dust power amplitudes scale strongly with mean dust brightness. Planck found

$100\,\mathrm{GHz}$3

with

$100\,\mathrm{GHz}$4

so approximately

$100\,\mathrm{GHz}$5

(Collaboration et al., 2014). The same work emphasized that this is only an average law, with individual regions deviating by $100\,\mathrm{GHz}$6 or more, especially at low brightness, because dust polarization depends not only on column density but also on magnetic-field structure and depolarization effects (Collaboration et al., 2014). This underlies the practical conclusion that dust contamination in polarization cannot be predicted accurately from total intensity alone.

For primordial $100\,\mathrm{GHz}$7-mode searches, the implication is stringent. Even in the faintest $100\,\mathrm{GHz}$8 patches, there are no completely clean windows where dust subtraction can be ignored (Collaboration et al., 2014). In the BICEP2-like field, direct extrapolation of Planck 353 GHz $100\,\mathrm{GHz}$9 to 150 GHz gave

$B$0

over $B$1, with statistical uncertainty

$B$2

and extrapolation uncertainty

$B$3

a dust level of the same order as the $B$4-mode power reported by BICEP2 in that range (Collaboration et al., 2014). The paper explicitly concluded that subtraction of polarized dust emission is essential for detecting primordial $B$5-modes at $B$6 or below (Collaboration et al., 2014).

3. Frequency decorrelation, line-of-sight complexity, and statistical dust modeling

A central controversy in polarized foreground analysis concerns whether dust polarization maps decorrelate between frequencies. Planck 2018 XI examined the correlation of dust polarization maps across frequency by fitting multi-frequency cross-spectra and found no evidence for decorrelation (Collaboration et al., 2018). Over $B$7, the multi-frequency analysis yielded $B$8 values close to unity, for example $B$9 for LR71, and the paper concluded that if the Planck limit for the largest sky region applies to the smaller sky regions observed by sub-orbital experiments, decorrelation might not be a problem for CMB experiments aiming at $p(\lambda)$0 at the recombination peak (Collaboration et al., 2018). Likewise, Planck Intermediate Results XXX reported no evidence for significant decorrelation in an appendix test, finding for example

$p(\lambda)$1

for high-latitude patches and $p(\lambda)$2 for the large regions, both consistent with unity (Collaboration et al., 2014).

At the same time, physically motivated three-dimensional modeling predicts that line-of-sight variation in dust density, temperature, emissivity index, and polarization angle should generically induce some decorrelation and spectral complexity. A six-layer Milky Way model constructed from extinction-based layers and Planck large-scale templates wrote the polarized emission as

$p(\lambda)$3

$p(\lambda)$4

with $p(\lambda)$5, explicitly showing that frequency dependence in polarization comes from both the local MBB factor and vector summation of differently oriented layers (Martínez-Solaeche et al., 2017). In that model, the average SED of the multilayer sky is flatter at low frequency than a 2-D single-MBB model, and the predicted decorrelation is modest: close to 100% on large scales for 217–353 GHz and around 97% on small scales, while 143–353 GHz decorrelation is of order 1–2% over the large-scale multipole range discussed in the paper (Martínez-Solaeche et al., 2017). This suggests that current Planck non-detections of decorrelation do not eliminate physically plausible line-of-sight complexity; rather, they bound it at the current sensitivity level.

Statistical simulation work reached a related conclusion from a different direction. A phenomenological all-sky dust-foreground simulation constrained to Planck $p(\lambda)$6, $p(\lambda)$7, and $p(\lambda)$8 spectra modeled the Galactic magnetic field as a mean field plus a random component with

$p(\lambda)$9

and found that the observed $T_d=19.6\,\mathrm{K}$0 and $T_d=19.6\,\mathrm{K}$1 slopes are matched for

$T_d=19.6\,\mathrm{K}$2

(Vansyngel et al., 2016). The same framework proposed a harmonic-space covariance method for multifrequency simulations, including imposed decorrelation through

$T_d=19.6\,\mathrm{K}$3

for $T_d=19.6\,\mathrm{K}$4 (Vansyngel et al., 2016). The model was not a physical explanation of $T_d=19.6\,\mathrm{K}$5 and $T_d=19.6\,\mathrm{K}$6 asymmetry, but it showed that plausible multifrequency dust skies can be generated with realistic covariance and spatial variance (Vansyngel et al., 2016).

A separate methodological development focused on recovering the angular power spectrum and non-Gaussian structure of dust polarization at 353 GHz from noisy Planck data. A wavelet phase harmonics denoising scheme optimized

$T_d=19.6\,\mathrm{K}$7

on the complex field $T_d=19.6\,\mathrm{K}$8, and in simulation it recovered the true signal power spectrum down to scales where the noise power is an order of magnitude larger than that of the signal (Blancard et al., 2021). This is not a frequency-spectrum result, but it underscores a key point: the dust polarization spectrum relevant for CMB analyses is not exhausted by second-order Gaussian statistics, because real dust maps are filamentary and strongly non-Gaussian (Blancard et al., 2021).

4. Far-infrared and submillimeter polarization spectra in clouds

Outside the diffuse ISM, the dust emission polarization spectrum is commonly measured as normalized fractional polarization, $T_d=19.6\,\mathrm{K}$9, across FIR and submillimeter bands. In a translucent molecular cloud observed by BLASTPol at 250, 350, and 500 $\beta_d^P \simeq 1.53$0 and combined with Planck 850 $\beta_d^P \simeq 1.53$1, the first submillimeter polarization spectrum of this class of cloud was found to be approximately flat from 250 to 850 $\beta_d^P \simeq 1.53$2 (Ashton et al., 2017). The principal measurements were

$\beta_d^P \simeq 1.53$3

$\beta_d^P \simeq 1.53$4

$\beta_d^P \simeq 1.53$5

all consistent with unity within the quoted uncertainties (Ashton et al., 2017). The paper argued that this near-flatness disfavors Draine & Fraisse models in which only silicate grains are aligned and the shorter submillimeter bands are dominated by a warmer unaligned graphite population (Ashton et al., 2017).

The Carina Nebula yielded a similar result. BLASTPol plus Planck data showed a submillimeter spectrum flat to within $\beta_d^P \simeq 1.53$6 relative to 350 $\beta_d^P \simeq 1.53$7, with no evidence for a pronounced minimum near 350 $\beta_d^P \simeq 1.53$8 (Shariff et al., 2018). For the preferred Far-subtracted case, the medians were

$\beta_d^P \simeq 1.53$9

and a power-law fit

$\beta_{\rm d}^{P} = 1.53\pm0.02$0

gave

$\beta_{\rm d}^{P} = 1.53\pm0.02$1

for the same case (Shariff et al., 2018). The paper further found no strong evidence that the spectral shape depends on Planck-derived dust temperature or optical depth across the cloud (Shariff et al., 2018). Taken together with the translucent-cloud result, this supports a class of cloud-averaged submillimeter polarization spectra that are much flatter than older ground-based V-shaped compilations.

By contrast, far-infrared HAWC+ studies of star-forming clouds find falling spectra. In OMC-1, M17, and W3, the cloud-averaged 89–214 $\beta_{\rm d}^{P} = 1.53\pm0.02$2 polarization percentage decreases with increasing wavelength (Cox et al., 16 Sep 2025). The normalized medians were

$\beta_{\rm d}^{P} = 1.53\pm0.02$3

for OMC-1,

$\beta_{\rm d}^{P} = 1.53\pm0.02$4

for M17, and

$\beta_{\rm d}^{P} = 1.53\pm0.02$5

for W3 (Cox et al., 16 Sep 2025). The same paper fit the local spectrum as

$\beta_{\rm d}^{P} = 1.53\pm0.02$6

and found a critical column density below which a falling spectrum is not observed, roughly $\beta_{\rm d}^{P} = 1.53\pm0.02$7 to $\beta_{\rm d}^{P} = 1.53\pm0.02$8, summarized as order $\beta_{\rm d}^{P} = 1.53\pm0.02$9 or $p(\lambda)$00 (Cox et al., 16 Sep 2025). The authors interpreted this as support for the Hildebrand hypothesis that regions shielded from near-IR radiation are required to produce a sharply falling polarization spectrum (Cox et al., 16 Sep 2025).

Radiation-magnetohydrodynamic modeling of a massive star-forming cloud provided a physical route to such falling FIR spectra. In synthetic HAWC+-band observations of a $p(\lambda)$01 core, neither homogeneous grain alignment nor collisional dealignment produced the observed falling spectrum; both yielded flat or rising behavior (Lee et al., 2024). A temperature-dependent alignment efficiency model,

$p(\lambda)$02

with $p(\lambda)$03 did (Lee et al., 2024). Its global spectrum was

$p(\lambda)$04

with median slope

$p(\lambda)$05

qualitatively matching OMC-1 (Lee et al., 2024). The paper also found no significant correlation between the best-fit slope and fitted dust temperature for the favored model, but a strong positive correlation between column density and the slope parameter, which it suggested could result from wavelength-dependent polarization by absorption (Lee et al., 2024).

5. Dense filaments, opacity evolution, and suppression of polarized emission

In dense star-forming filaments, the dust emission polarization spectrum is often inferred indirectly by combining single-band polarization with multi-band dust SED information. In the massive IRDC G035.39-00.33, JCMT/POL-2 and Planck data at 850 $p(\lambda)$06 were combined with Herschel and SCUBA-2 total-intensity maps to characterize both the dust emission spectrum and the polarization behavior (Juvela et al., 2018). The dense filament has minimum dust temperature $p(\lambda)$07, peak column density

$p(\lambda)$08

mass of some $p(\lambda)$09 for the area $p(\lambda)$10, and average dust opacity spectral index

$p(\lambda)$11

(Juvela et al., 2018). The ratio

$p(\lambda)$12

is more than four times the typical diffuse-medium value, which the paper interpreted as strong evidence of dust evolution (Juvela et al., 2018).

At the same time, the polarization fraction decreases as a function of column density to $p(\lambda)$13 in the central filament, with the observed decrease significant only at $p(\lambda)$14 (Juvela et al., 2018). Modeling suggested that constant alignment is insufficient to explain the low dense-filament polarization, whereas the data can be explained with a complete loss of alignment at densities above $p(\lambda)$15 or using radiative torque alignment predictions, although uncertainty in field geometry and SCUBA-2 spatial filtering prevents strong conclusions (Juvela et al., 2018). This suggests that the polarized-emitting grain population in dense filaments can differ sharply from the population dominating the unpolarized continuum SED, even when both arise from the same cold dust reservoir.

A related but broader 353 GHz view of the environmental dependence of polarized thermal dust emission was provided by Planck 2018 XII. At $p(\lambda)$16 resolution, the maximum polarization fraction was

$p(\lambda)$17

with the uncertainty dominated by the total-intensity zero level (Collaboration et al., 2018). Over the diffuse ISM, $p(\lambda)$18 falls by a factor of about 3–4 between $p(\lambda)$19 and $p(\lambda)$20, but the product $p(\lambda)$21 decreases by only about 25% (Collaboration et al., 2018). The paper interpreted the inverse relation

$p(\lambda)$22

as evidence that much of the observed decrease in polarization fraction is caused by beam and line-of-sight depolarization from magnetic-field structure, not by a large decline in grain alignment efficiency (Collaboration et al., 2018). A plausible implication is that spectral changes in polarized emission with environment must be disentangled from geometric depolarization before they are attributed to grain physics.

6. Circumstellar disks and mechanism-dependent polarization spectra

In protoplanetary disks, the dust emission polarization spectrum is mechanism-dependent in a more direct sense than in the diffuse ISM. The Class II disk IM Lup provides a resolved single-band anchor at 870 $p(\lambda)$23: ALMA observations at mean frequency 343.479 GHz showed polarization vectors predominantly aligned with the disk minor axis, a polarization fraction that increases toward the center and reaches a peak of about $p(\lambda)$24, and a morphology consistent with self-scattering of submillimeter-wave emission from an optically thin inclined disk (Hull et al., 2018). The authors explicitly tested the four Band 7 spectral windows spanning $p(\lambda)$25–350.5 GHz and found no significant changes in polarized intensity, polarization fraction, or polarization angle across those sub-bands, but also stated that this does not constitute a meaningful polarization-spectrum measurement (Hull et al., 2018). The importance of the result is interpretive: at 870 $p(\lambda)$26, IM Lup lies in a scattering-dominated regime, so any multi-band polarization spectrum model must reproduce that point (Hull et al., 2018).

HL Tau demonstrates the full multiwavelength transition. A plane-parallel slab treatment of aligned, non-spherical grains with scattering showed that the polarization morphology changes from scattering-dominated at $p(\lambda)$27, through a mixed regime at $p(\lambda)$28, to an aligned-grain thermal-emission regime at $p(\lambda)$29 (Lin et al., 2021). The key physical point is that this transition can be driven by optical depth. In the optically thin limit, direct thermal polarization from aligned grains dominates; in the optically thick limit, dichroic extinction suppresses thermal polarization and scattering dominates (Lin et al., 2021). The exact optically thick thermal-emission limit was written as

$p(\lambda)$30

making explicit the competition between intrinsic thermal polarization and dichroic extinction (Lin et al., 2021).

For observational decomposition, the paper introduced

$p(\lambda)$31

and more generally

$p(\lambda)$32

where $p(\lambda)$33 is the scattering component and $p(\lambda)$34 the attenuated thermal aligned-grain component (Lin et al., 2021). Applying this to HL Tau, the authors found that with increasing wavelength the fractional polarization spectrum of the scattering component decreases, while the thermal component increases, exactly as expected if optical depth decreases with wavelength (Lin et al., 2021). This establishes a disk-specific usage of “polarization spectrum” in which the frequency trend traces not only grain size but the competition among scattering, aligned-grain emission, and dichroic extinction.

7. Microwave spinning dust, nanosilicates, and low-frequency polarization

At microwave frequencies, dust emission polarization can arise from spinning nanoparticles rather than thermal vibrational emission. A model constrained by UV extinction and polarization for two sightlines with a 2175 Å polarization feature predicted that the degree of polarization of spinning dust emission is about $p(\lambda)$35 at $p(\lambda)$36 and declines to below $p(\lambda)$37 for frequencies above $p(\lambda)$38 (Hoang et al., 2013). The same framework predicted the degree of polarization of thermal dust emission at 353 GHz to be $p(\lambda)$39 and $p(\lambda)$40 for the two lines of sight (Hoang et al., 2013). This places spinning-dust polarization in a distinct low-polarization, low-frequency regime relative to the thermal-dust spectrum.

A separate model considered ultrasmall silicate nanoparticles as AME carriers. For a log-normal size distribution,

$p(\lambda)$41

and polarized emissivity

$p(\lambda)$42

the paper found that the polarization fraction of spinning dust emission increases with decreasing dipole moment per atom $p(\lambda)$43 and can reach

$p(\lambda)$44

for $p(\lambda)$45 at grain temperature 60 K (Hoang et al., 2016). It identified a parameter space

$p(\lambda)$46

in which rotational emission from spinning silicate nanoparticles can reproduce both observed AME and the polarization of the AME without violating ultraviolet extinction and polarization constraints (Hoang et al., 2016). This does not imply that such grains dominate the observed low-frequency polarized dust foreground, but it establishes that the dust emission polarization spectrum below tens of GHz is a diagnostic of nanoparticle composition, abundance, and alignment, not merely an extrapolation of the thermal modified blackbody.

Across environments, these results indicate that “dust emission polarization spectrum” is not a single universal observable but a family of related diagnostics. In diffuse Galactic emission, it is well approximated at high frequency by a modified blackbody with stable angular statistics and no detected inter-frequency decorrelation at Planck sensitivity [(Collaboration et al., 2018); (Collaboration et al., 2014)]. In molecular clouds, flat submillimeter spectra and falling far-infrared spectra coexist, implying that temperature structure, shielding, and alignment heterogeneity all matter (Ashton et al., 2017, Shariff et al., 2018, Cox et al., 16 Sep 2025). In disks, the spectrum can map the transition among scattering, aligned-grain emission, and dichroic extinction (Hull et al., 2018, Lin et al., 2021). At low microwave frequencies, spinning dust adds a separate polarization component whose amplitude is typically small but strongly mechanism-dependent [(Hoang et al., 2013); (Hoang et al., 2016)]. The common lesson is that a polarization spectrum cannot be interpreted from emissivity alone: magnetic geometry, optical depth, line-of-sight mixing, and the identity of the polarized-emitting grain population are all inseparable parts of the observable.