Extended Anomalous Foreground Emission in the WMAP 3-Year Data

TL;DR

WMAP foreground analysis using a template-based regression disentangles free-free, dust, and synchrotron components while revealing persistent anomalous emissions, notably spinning-dust and the Galactic-center haze. The study shows that free-free traced by H-alpha deviates from a simple power law and that the dust spectrum arises from a mix of spinning-dust and thermal-dust contributions; the haze corresponds to hard synchrotron emission and its inference is highly sensitive to CMB estimator bias. By performing both full-sky and regional analyses and introducing haze templates, the work exposes fundamental limitations in WMAP-era CMB cleaning and outlines strategies for Planck to minimize systematic biases in foreground spectra. Overall, the results highlight the complexity of Galactic microwave foregrounds and the need for broad frequency coverage to robustly separate CMB from foreground components.

Abstract

We study the spectral and morphological characteristics of the diffuse Galactic emission in the WMAP temperature data using a template-based multi-linear regression, and obtain the following results. 1. We confirm previous observations of a bump in the dust-correlated spectrum, consistent with the Draine & Lazarian spinning dust model. 2. We also confirm the "haze" signal in the inner Galaxy, and argue that it does not follow a free-free spectrum as first thought, but instead is synchrotron emission from a hard electron cosmic-ray population. 3. In a departure from previous work, we allow the spectrum of Halpha-correlated emission (which is used to trace the free-free component) to float in the fit, and find that it does not follow the expected free-free spectrum. Instead there is a bump near 50 GHz, modifying the spectrum at the 20% level, which we speculate is caused by spinning dust in the warm ionized medium. 4. The derived cross-correlation spectra are not sensitive to the map zero points, but are sensitive to the choice of CMB estimator. In cases where the CMB estimator is derived by minimizing variance of a linear combination of the WMAP bands, we show that a bias proportional to the cross-correlation of each template and the true CMB is always present. This bias can be larger than any of the foreground signals in some bands. 5. Lastly, we consider the frequency coverage and sensitivity of the Planck mission, and suggest linear combination coefficients for the CMB template that will reduce both the statistical and systematic uncertainty in the synchrotron and haze spectra by more than an order of magnitude.

Figures (10)

• Figure 1: Residual maps ${\rm\bf r}$ for three of the five bands and for FS7 and FS8 fits using CMB5 (the K, Ka, and Q band maps are stretched to $\pm 0.25$, $\pm 0.12$, and $\pm 0.08$ mK respectively; all maps are mean subtracted). The unsubtracted WMAP data are also shown for comparison. The FS7 fit removes much of the emission, however there is a remaining excess towards the Galactic center. This excess emission is particularly notable south of the Galactic center where obscuration by dust and gas is negligible. FS8 incorporates a simple spatial template for this haze and removes much of the residuals in that region.
• Figure 2: The same as Figure \ref{['fig:fullsky']} except for the GC fits ($l,b = [-45:45]$ degrees). Though GC7 explicitly fits the free--free, dust, and synchrotron spectra, the haze is still present in ${\rm\bf r}_{\rm GC7}$ indicating that it is morphologically dissimilar to the those templates. The quality of the fit is substantially improved with the inclusion of a haze template (GC8). The extended structure in the north-west GC is likely due to an imperfect haze template.
• Figure 3: Illustration of the 12 regions independently fit in the RG fits. Breaking the sky up in this way removes many of the over- and under-subtracted features in the residual maps, particularly the excess emission in the north-west GC.
• Figure 4: Residual maps for the RG7 and RG8 fits using CMB5, with the same stretch as Figure \ref{['fig:fullsky']}. The haze is still present in the ${\rm\bf r}_{\rm RG7}$ fits despite the relatively small size of the fitting regions. The bottom row shows ${\rm\bf r}_{\rm RG8}$ with the subtracted haze in each region added back in to that region. These maps are unsmoothed and no continuity constraints are placed on the region boundaries.
• Figure 5: Foreground spectra for the FS8 fits with our six CMB estimators: solid = CMB1, dotted = CMB2, short dashed = CMB3, dot-dashed = CMB4, dot-dot-dot-dashed = CMB5, and long dashed = CMB6. The error bars on the coefficients are the formal error bars on the fit (see text).