Low Northern Variance in CMB

• Low Northern Variance is a phenomenon where the northern ecliptic hemisphere shows a significant deficit in CMB temperature fluctuations compared to the ΛCDM model.
• The analysis uses regional pixel-space variance estimators, comparing hemispherical data with Monte Carlo simulations to assess statistical significance.
• Empirical results from WMAP and Planck indicate extremely low p-values for the northern anomaly, while polarization tests demonstrate consistency with ΛCDM predictions.

Low Northern Variance is the well-established finding that the Cosmic Microwave Background temperature field has anomalously low large-scale variance in the northern Ecliptic hemisphere, while the southern Ecliptic hemisphere appears consistent with the best-fitting $\Lambda$CDM model. The anomaly is a real-space manifestation of a broader hemispherical asymmetry in power that is nearly aligned with the Ecliptic North/South division. It was first noted with WMAP and confirmed with Planck, and it has become a focal point in tests of statistical isotropy, mask dependence, low-$\ell$ anomaly phenomenology, and temperature–polarization consistency within $\Lambda$CDM (O'Dwyer et al., 2016, O'Dwyer et al., 2019).

1. Phenomenology and geometric definition

The anomaly is defined by partitioning the sky into northern and southern Ecliptic hemispheres and computing a regional variance statistic on the CMB map in each half. In this construction, the northern Ecliptic hemisphere exhibits a deficit of temperature variance, whereas the southern hemisphere is ordinary. The preferred axis is nearly aligned with the Ecliptic, which immediately raises the possibility of solar-system foregrounds or scanning-related systematics; however, the effect has persisted across independent pipelines and sky maps (O'Dwyer et al., 2016).

In the temperature literature summarized here, “Low Northern Variance” refers specifically to the northern Ecliptic hemisphere rather than to an arbitrary dipolar modulation axis. This is important because some apparently related statistics, such as local-variance dipoles, probe spatial modulation in a different way and need not have identical significance properties under masking or limited sky coverage. The southern hemisphere’s consistency with $\Lambda$CDM is a recurrent empirical point: the anomaly is not a symmetric north–south suppression but a localized deficit in the Ecliptic north (O'Dwyer et al., 2016).

A central interpretive issue is whether the temperature result is a statistical fluke within $\Lambda$CDM or evidence for a physical or instrumental mechanism. The papers considered here argue that polarization variance provides the decisive complementary test because the observed temperature anomaly does not, by itself, imply a corresponding low-variance polarization realization under standard TE correlations (O'Dwyer et al., 2016, O'Dwyer et al., 2019).

2. Variance estimators, power-spectrum relations, and masking

The core observable is a regional pixel-space variance. One form used in the cited work subtracts the regional mean,

$$\sigma^2_R = \frac{1}{N_R} \sum_{p \in R} \big(T_p - \bar T_R\big)^2, \qquad \bar T_R = \frac{1}{N_R} \sum_{p \in R} T_p.$$

A related estimator used for a map $X(\hat n)$, with $X=T$ for temperature or a polarization-derived field, is

$$\hat{\sigma}_{X,R}^2 \equiv \frac{1}{N_R}\sum_{p\in R} X^2(p).$$

Both are used operationally to characterize regional power deficits on masked hemispheres (O'Dwyer et al., 2016, O'Dwyer et al., 2019).

For statistically isotropic Gaussian fields on the full sky, the variance is determined by the angular power spectrum. For temperature,

$$\langle T^2 \rangle = \sum_{\ell} \frac{2\ell+1}{4\pi} C_\ell^{TT}.$$

For polarization, the full-sky expressions summarized in the literature include

$\ell$0

and, for the rotationally invariant amplitude $\ell$1,

$\ell$2

One of the cited polarization forecasts instead adopts a rotationally invariant statistic based on the variance of $\ell$3, noting that in $\ell$4CDM, $\ell$5 when $\ell$6 and $\ell$7 are uncorrelated (O'Dwyer et al., 2016, O'Dwyer et al., 2019).

On a masked or limited sky, the variance is no longer a simple diagonal sum over multipoles. A schematic approximation is

$\ell$8

or, more generally for a regional estimator,

$\ell$9

where the window function or mask induces mode coupling. Accordingly, the cited analyses calibrate expectations and p-values with pseudo-$\Lambda$0 logic and Monte Carlo ensembles rather than with a closed-form variance formula alone (O'Dwyer et al., 2019).

Methodologically, the anomaly is sensitive to coordinate rotation, mask degradation, and low-$\Lambda$1 treatment. Temperature maps and masks are rotated from Galactic to Ecliptic coordinates before evaluation. In the low-$\Lambda$2 study, maps are degraded to $\Lambda$3 with multipoles restricted to $\Lambda$4, after deconvolving the high-resolution pixel window and the $\Lambda$5 Gaussian beam and reconvolving the $\Lambda$6 pixel window. In the broader-band study, maps are prepared at HEALPix $\Lambda$7 and smoothed or compensated with a $\Lambda$8 arcmin FWHM Gaussian beam prior to downgrading for variance estimation, with a truncation at $\Lambda$9 to avoid Doppler-induced dipolar modulation that becomes measurable above $\Lambda$0 (O'Dwyer et al., 2016, O'Dwyer et al., 2019).

3. Empirical status in WMAP and Planck temperature data

The anomaly was first noted with WMAP and later confirmed with Planck. In one Planck SMICA analysis with the Planck Common mask, the Ecliptic North temperature variance has p-value $\Lambda$1 relative to unconditioned $\Lambda$2CDM realizations, while the Ecliptic South has p-value $\Lambda$3, indicating that the southern hemisphere is unremarkable (O'Dwyer et al., 2016).

A complementary low-$\Lambda$4 analysis based on $\Lambda$5 reports comparably small northern-tail probabilities but with values that depend on release and mask choice. Using SMICA temperature maps and Common masks matched to each Planck release, the measured northern Ecliptic hemispherical variance gives p-values of approximately $\Lambda$6 for PR2 (2015) and $\Lambda$7 for PR3 (2018). Using the PR2 mask on the PR3 SMICA map yields $\Lambda$8; the shift from $\Lambda$9 to $\Lambda$0 is described as mostly mask-driven, with a small residual contribution from best-fit cosmology differences (O'Dwyer et al., 2019).

Configuration	North p-value	Note
Planck SMICA + Common mask, full Ecliptic North	$\Lambda$1	Ecliptic South: $\Lambda$2
Atacama-visible subset of Ecliptic North	$\Lambda$3	Reduced sky area broadens the distribution
PR2 low-$\Lambda$4 analysis	$\Lambda$5	$\Lambda$6
PR3 low-$\Lambda$7 analysis	$\Lambda$8	PR3 map with PR2 mask: $\Lambda$9

The broader anomaly literature summarized in the same sources includes related but distinct diagnostics. Fitting a dipole to local variance maps yields amplitudes that only $$\sigma^2_R = \frac{1}{N_R} \sum_{p \in R} \big(T_p - \bar T_R\big)^2, \qquad \bar T_R = \frac{1}{N_R} \sum_{p \in R} T_p.$$0 in $$\sigma^2_R = \frac{1}{N_R} \sum_{p \in R} \big(T_p - \bar T_R\big)^2, \qquad \bar T_R = \frac{1}{N_R} \sum_{p \in R} T_p.$$1 Planck Full Focal Plane simulations match, and only $$\sigma^2_R = \frac{1}{N_R} \sum_{p \in R} \big(T_p - \bar T_R\big)^2, \qquad \bar T_R = \frac{1}{N_R} \sum_{p \in R} T_p.$$2 in $$\sigma^2_R = \frac{1}{N_R} \sum_{p \in R} \big(T_p - \bar T_R\big)^2, \qquad \bar T_R = \frac{1}{N_R} \sum_{p \in R} T_p.$$3 FFP simulations have lower Ecliptic-North variance than observed (O'Dwyer et al., 2016). However, neither the variance difference nor the variance ratio between hemispheres is itself highly anomalous, which underscores that the primary effect is the low absolute variance of the Ecliptic north rather than a singularly extreme north–south contrast (O'Dwyer et al., 2016).

Several caveats are established. The significance depends on mask choice and multipole range, and the inclusion of the lowest multipoles is essential. For Planck 2015, multipoles as low as $$\sigma^2_R = \frac{1}{N_R} \sum_{p \in R} \big(T_p - \bar T_R\big)^2, \qquad \bar T_R = \frac{1}{N_R} \sum_{p \in R} T_p.$$4 need inclusion to see the hemispherical variance anomaly; for Planck 2018, the low-variance feature is not robust if multipoles below $$\sigma^2_R = \frac{1}{N_R} \sum_{p \in R} \big(T_p - \bar T_R\big)^2, \qquad \bar T_R = \frac{1}{N_R} \sum_{p \in R} T_p.$$5 are excluded (O'Dwyer et al., 2019). This makes the anomaly a large-angle phenomenon in a strict sense.

4. Polarization as the decisive complementary test

The central question is whether the observed low temperature variance in the Ecliptic north should propagate into polarization under $$\sigma^2_R = \frac{1}{N_R} \sum_{p \in R} \big(T_p - \bar T_R\big)^2, \qquad \bar T_R = \frac{1}{N_R} \sum_{p \in R} T_p.$$6CDM. The answer given by both cited studies is no. Even though temperature and E-mode polarization are correlated through $$\sigma^2_R = \frac{1}{N_R} \sum_{p \in R} \big(T_p - \bar T_R\big)^2, \qquad \bar T_R = \frac{1}{N_R} \sum_{p \in R} T_p.$$7, constrained polarization realizations conditioned on the observed temperature map do not generically produce anomalously low northern polarization variance (O'Dwyer et al., 2016, O'Dwyer et al., 2019).

The construction is based on the joint Gaussian distribution of $$\sigma^2_R = \frac{1}{N_R} \sum_{p \in R} \big(T_p - \bar T_R\big)^2, \qquad \bar T_R = \frac{1}{N_R} \sum_{p \in R} T_p.$$8 and $$\sigma^2_R = \frac{1}{N_R} \sum_{p \in R} \big(T_p - \bar T_R\big)^2, \qquad \bar T_R = \frac{1}{N_R} \sum_{p \in R} T_p.$$9 with covariance

$X(\hat n)$0

Conditioning on the observed temperature coefficients yields

$X(\hat n)$1

and

$X(\hat n)$2

One forecast uses this framework to generate E-mode realizations up to $X(\hat n)$3, split into parts correlated and uncorrelated with temperature, and then evaluates pixel-space polarization variance on masked hemispheres (O'Dwyer et al., 2016).

The result is structurally robust. Conditioning on the anomalous temperature map produces only a slight overall suppression of polarization variance, and this suppression affects both hemispheres similarly. The correlated E-mode component does show some suppression, but it is subdominant to the uncorrelated E-mode component, which sets the width of the variance distribution. Consequently, there is no predicted north–south asymmetry in polarization variance under the fluke hypothesis (O'Dwyer et al., 2016).

The low-$X(\hat n)$4 parameter study reaches the same conclusion from a complementary direction. Its covariance analysis reports that the Pearson correlation between full-sky temperature variance and polarization variance is weak, with $X(\hat n)$5, consistent with modest TE correlation (O'Dwyer et al., 2019). A plausible implication is that the temperature anomaly does not automatically forecast an analogous polarization anomaly even when the temperature map is taken as given.

This makes polarization variance an unusually clean falsification test of the fluke interpretation. The cited forecast states that a fractional suppression of polarization variance in the Ecliptic North by approximately $X(\hat n)$6, comparable to the temperature suppression, would fall far below $X(\hat n)$7 in the constrained $X(\hat n)$8CDM distributions, even with only partial northern coverage. Detecting such a low-North polarization variance would therefore provide strong evidence against the hypothesis that the temperature anomaly is merely a statistical fluctuation within $X(\hat n)$9CDM (O'Dwyer et al., 2016).

5. Dependence on cosmological parameters and the role of reionization optical depth

Unlike temperature variance, polarization variance is noticeably sensitive to present uncertainties in cosmological parameters. The dominant effect comes from the reionization optical depth $X=T$0, because the large-scale E-mode “reionization bump” at low multipoles is sourced by Thomson scattering during reionization and scales approximately as

$X=T$1

with $X=T$2 the scalar amplitude. Since the variance tests of interest use low $X=T$3, the expected polarization-variance distribution inherits a substantial dependence on $X=T$4 (O'Dwyer et al., 2019).

The paper quantifies this using Planck MCMC chains. For PR2 (2015) TT,TE,EE + lowP, the parameter constraints quoted are $X=T$5 and $X=T$6. For PR3 (2018) TT,TE,EE + low$X=T$7 + lowE, the quoted constraints are $X=T$8 and $X=T$9 (O'Dwyer et al., 2019). The improved PR3 determination of $$\hat{\sigma}_{X,R}^2 \equiv \frac{1}{N_R}\sum_{p\in R} X^2(p).$$0 narrows and shifts the predicted polarization-variance distribution relative to PR2.

A key comparison is between a best-fit fixed-cosmology distribution and a chain-marginalized distribution that incorporates parameter uncertainty. For temperature, these are nearly identical for both PR2 and PR3, so temperature variance is effectively cosmic-variance-limited at current parameter precision. For polarization, by contrast, the chain distribution is appreciably broader. In the full-sky $$\hat{\sigma}_{X,R}^2 \equiv \frac{1}{N_R}\sum_{p\in R} X^2(p).$$1 analysis, the PR3 polarization-variance distribution is about $$\hat{\sigma}_{X,R}^2 \equiv \frac{1}{N_R}\sum_{p\in R} X^2(p).$$2 wider than the best-fit, cosmic-variance-limited benchmark (O'Dwyer et al., 2019).

To isolate the impact of tighter parameter knowledge, the study introduces a forecast chain constructed by uniformly dividing all parameter covariances by four, thereby reducing $$\hat{\sigma}_{X,R}^2 \equiv \frac{1}{N_R}\sum_{p\in R} X^2(p).$$3 errors by a factor of two while preserving positive semi-definiteness. In this forecast, the width of the full-sky polarization-variance distribution is reduced to approximately $$\hat{\sigma}_{X,R}^2 \equiv \frac{1}{N_R}\sum_{p\in R} X^2(p).$$4 above the best-fit case, with

$$\hat{\sigma}_{X,R}^2 \equiv \frac{1}{N_R}\sum_{p\in R} X^2(p).$$5

This indicates that improved $$\hat{\sigma}_{X,R}^2 \equiv \frac{1}{N_R}\sum_{p\in R} X^2(p).$$6 measurements can bring hemispherical polarization-variance expectations close to the cosmic-variance-limited distribution, substantially sharpening the anomaly test (O'Dwyer et al., 2019).

6. Sky coverage, methodological caveats, and observational implications

Sky coverage is a first-order determinant of statistical power. The CMB-S4 forecast considers two northern-hemisphere scenarios: full Ecliptic-North coverage and the Atacama-accessible part of the Ecliptic north. After masking, the full-sky foreground-cleaned coverage is quoted as $$\hat{\sigma}_{X,R}^2 \equiv \frac{1}{N_R}\sum_{p\in R} X^2(p).$$7, with the northern half used for the hemispherical test. For an Atacama site on the high Chilean plateau, the observable celestial declinations run from $$\hat{\sigma}_{X,R}^2 \equiv \frac{1}{N_R}\sum_{p\in R} X^2(p).$$8 South to $$\hat{\sigma}_{X,R}^2 \equiv \frac{1}{N_R}\sum_{p\in R} X^2(p).$$9 North, yielding an Atacama-North coverage of approximately $$\langle T^2 \rangle = \sum_{\ell} \frac{2\ell+1}{4\pi} C_\ell^{TT}.$$0 of the northern hemisphere after masking (O'Dwyer et al., 2016).

Within $$\langle T^2 \rangle = \sum_{\ell} \frac{2\ell+1}{4\pi} C_\ell^{TT}.$$1CDM with temperature conditioning but no anomaly model, the full-North and Atacama-North polarization-variance measurements have the same mean and are correlated, but the Atacama-North distribution is broader because of the smaller sky area. A joint distribution built from $$\langle T^2 \rangle = \sum_{\ell} \frac{2\ell+1}{4\pi} C_\ell^{TT}.$$2 constrained realizations shows that low variance in one scenario predicts low variance in the other, but reduced sky coverage generically inflates p-values; for example, a p-value of $$\langle T^2 \rangle = \sum_{\ell} \frac{2\ell+1}{4\pi} C_\ell^{TT}.$$3 in the full Ecliptic North corresponds to p-values “up to a few percent” in Atacama-North (O'Dwyer et al., 2016).

This is one reason the simple hemispherical variance statistic is favored under fragmented coverage. The cited forecast compares it with the local-variance-dipole method using $$\langle T^2 \rangle = \sum_{\ell} \frac{2\ell+1}{4\pi} C_\ell^{TT}.$$4 disks and finds that, for Atacama-North temperature, the local-variance-dipole significance drops to $$\langle T^2 \rangle = \sum_{\ell} \frac{2\ell+1}{4\pi} C_\ell^{TT}.$$5, much weaker than the hemispherical variance approach (O'Dwyer et al., 2016). The methodological lesson is not that local variance is uninformative, but that the hemispherical variance is more robust and non-parametric when the observed region is geometrically restricted.

Several caveats remain. The anomaly is partly a posteriori because the Ecliptic partition was motivated by observed temperature features. Its significance depends on mask handling, coordinate rotation, and low-$$\langle T^2 \rangle = \sum_{\ell} \frac{2\ell+1}{4\pi} C_\ell^{TT}.$$6 selection. Current polarization maps have low signal-to-noise at large scales; one summary notes that Planck 2018 isotropy analyses reported an E-mode variance dipole with p-values below $$\langle T^2 \rangle = \sum_{\ell} \frac{2\ell+1}{4\pi} C_\ell^{TT}.$$7 in amplitude or direction for some component-separated maps, but also cautioned that anisotropic noise remains a concern (O'Dwyer et al., 2019). The CMB-S4 forecast itself does not model instrumental noise or foreground residuals; it targets the theoretical posterior under $$\langle T^2 \rangle = \sum_{\ell} \frac{2\ell+1}{4\pi} C_\ell^{TT}.$$8CDM given temperature (O'Dwyer et al., 2016).

The broader implication is sharply formulated by the cited work. If future large-scale polarization data from experiments such as CLASS, LiteBIRD, or CMB-S4 show ordinary northern Ecliptic polarization variance, that would support the interpretation of the temperature anomaly as a statistical fluke. If instead they show anomalously low northern polarization variance, especially at a suppression level comparable to temperature, that outcome is not expected under $$\langle T^2 \rangle = \sum_{\ell} \frac{2\ell+1}{4\pi} C_\ell^{TT}.$$9CDM even after conditioning on the observed temperature and would weigh against the fluke hypothesis (O'Dwyer et al., 2016, O'Dwyer et al., 2019). A plausible implication is that polarization will determine whether Low Northern Variance remains a temperature-specific large-angle curiosity or becomes evidence for a deeper anisotropic mechanism.