Probing Cosmology through Higher-Order CMB Lensing Statistics

1. Problem Statement

Current analyses of Cosmic Microwave Background (CMB) lensing rely primarily on the two-point correlation function (angular power spectrum, $C_{\ell}^{\kappa\kappa}$). While effective, the power spectrum captures only Gaussian information and misses significant cosmological data encoded in the non-Gaussian features generated by nonlinear structure formation, particularly on small angular scales.

Upcoming experiments like the Simons Observatory (SO) will provide high-resolution, low-noise CMB maps. However, extracting non-Gaussian information is challenging due to:

• Reconstruction Noise: The process of reconstructing the lensing convergence field ($\kappa$) from CMB temperature and polarization introduces complex, higher-order noise biases (e.g., $N^{(1)}, N^{(2)}$).
• Modeling Complexity: Analytical perturbative expansions often fail to capture the full hierarchy of nonlinear effects, post-Born corrections, and reconstruction noise simultaneously.
• Lack of Comprehensive Assessment: Previous studies have not fully assessed the utility of various non-Gaussian statistics (e.g., Minkowski functionals, peak counts) under realistic SO-like noise conditions.

The paper aims to quantify the relative information gain provided by a suite of higher-order statistics beyond the standard power spectrum for an SO-like experiment, specifically targeting constraints on the matter density ($\Omega_m$), primordial amplitude ($A_s$), and the sum of neutrino masses ($M_\nu$).

2. Methodology

The authors employ a fully field-level forward-modeling pipeline, avoiding reliance on analytical approximations for noise or non-Gaussianity.

A. Simulations and Forward Modeling

• N-body Simulations: Utilized the MassiveNuS suite, which includes 101 cosmologies varying $\Omega_m$, $A_s$, and $M_\nu$. The simulations use $1024^3$ CDM particles in a $(512 \text{ Mpc}/h)^3$ box.
• Ray-Tracing: Generated 10,000 pseudo-independent lensing convergence maps ($\kappa$) from a single N-body realization using random shifts and rotations to maximize sample size.
• Lensed CMB Generation:
  1. Generated unlensed Gaussian CMB fields ($T, E, B$) based on a fiducial cosmology.
  2. Applied the lensing remapping using the simulated $\kappa$ maps.
  3. Added realistic noise realizations based on the expected Simons Observatory (SO) noise power spectra (after component separation).
• Lensing Reconstruction: Reconstructed the lensing potential ($\phi$) and convergence ($\kappa$) using the standard quadratic estimator. The pipeline naturally incorporates post-Born corrections and the full hierarchy of reconstruction noise biases without separate analytical treatment.

B. Statistics Analyzed

The study measured several non-Gaussian statistics from the reconstructed $\kappa$ maps (after Wiener filtering):

1. Angular Power Spectrum ($C_\ell$): The standard baseline.
2. Minkowski Functionals (MFs): $V_0$ (area), $V_1$ (perimeter), and $V_2$ (Euler characteristic/curvature) measured across excursion set thresholds.
3. Peak and Minima Counts (PC/MC): The number of local maxima and minima within specific threshold bins.
4. Low-Order Moments (MOM): Variance ($\sigma_0, \sigma_1$) and skewness/kurtosis-like combinations ($S_1, K_1$).
5. Wavelet Scattering Transform (WST): A translation-invariant, multi-scale representation capturing non-Gaussian phase information.

C. Emulation and Likelihood

• Gaussian Process (GP) Emulators: Trained on the 10,000 simulations to model the dependence of each statistic on $\{\Omega_m, A_s, M_\nu\}$. A Matérn kernel ($\nu=0.5$) was used to handle rapid fluctuations.
• Covariance Matrix: Constructed directly from the 10,000 simulations, including a correction for emulator interpolation uncertainty ($\hat{C}_{tot} = \hat{C}_{sim} + \hat{C}_{emu}$).
• Forecasting: Used Markov Chain Monte Carlo (MCMC) sampling (via emcee) to derive marginalized constraints on cosmological parameters.

3. Key Contributions

• First Field-Level Assessment: This is the first comprehensive study to evaluate higher-order CMB lensing statistics using a fully forward-modeled pipeline that includes realistic reconstruction noise, post-Born effects, and nonlinear structure formation simultaneously.
• Systematic Comparison: Provides a direct comparison of MFs, PC/MC, MOM, and WST under identical SO-like noise conditions, identifying which statistics are robust to noise.
• Neutrino Mass Constraints: Specifically quantifies the ability of non-Gaussian statistics to break degeneracies and improve constraints on the sum of neutrino masses ($M_\nu$), a notoriously difficult parameter to constrain.

4. Key Results

A. Performance of Individual Statistics

• Minkowski Functionals (MFs): The most powerful non-Gaussian probe. They capture the full morphological and topological information (filaments, voids, clusters) of the field.
  • Combined with $C_\ell$, MFs reduce uncertainties on $\Omega_m$ by 34% and $A_s$ by 32%.
  • They are uniquely sensitive to $M_\nu$, reducing the one-sided uncertainty on $M_\nu$ by 69% (from 0.101 eV to 0.032 eV).
• Peak/Minima Counts (PC/MC): Provide complementary information, improving constraints by ~15-20% on $\Omega_m$ and $A_s$. They are less effective than MFs for $M_\nu$ but still significant.
• Moments (MOM) and Wavelet Scattering (WST): Provided limited gains at SO noise levels.
  • Reason: These statistics rely on real-space averages or are highly sensitive to reconstruction noise. The high noise levels in CMB lensing (compared to galaxy weak lensing) degrade their signal-to-noise ratio, and their coefficients become highly correlated, reducing effective dimensionality.

B. Combined Constraints

• Total Information Gain: Combining all non-Gaussian statistics with the power spectrum yields an 88% improvement in the Figure of Merit (FoM) for the $(\Omega_m, A_s)$ plane.
• Uncertainty Reduction:
  • $\sigma(\Omega_m)$ reduced by 40%.
  • $\sigma(A_s)$ reduced by 38%.
  • One-sided uncertainty on $M_\nu$ reduced by 70%.
• Robustness: These gains persist even when the power spectrum is extended to larger scales ($\ell < 300$) and combined with primary CMB and BAO data. MFs alone provide an additional 11% improvement on $\sigma(M_\nu)$ and 35% on $\sigma(\Omega_m)$ beyond the extended power spectrum.

C. Parameter Degeneracies

• MFs exhibit degeneracy directions in the parameter space that are orthogonal to those of the power spectrum, allowing them to effectively break degeneracies (particularly between $\Omega_m$ and $M_\nu$).
• PC/MC have degeneracy directions more aligned with the power spectrum, offering less breaking power but still additive.

5. Significance

• Cosmological Potential: The study demonstrates that non-Gaussian statistics contain a wealth of "untapped" information, comparable to roughly half the information captured by the power spectrum itself.
• Neutrino Mass: The results suggest that future CMB surveys (SO, SPT-3G, PICO) can significantly tighten constraints on neutrino mass using morphology-based statistics, potentially reaching sensitivities where the neutrino mass hierarchy could be distinguished.
• Methodological Framework: The developed forward-modeling pipeline serves as a robust template for future analyses, showing that simulation-based inference is necessary to accurately forecast the performance of higher-order statistics in the presence of complex reconstruction noise.
• Future Applications: The methodology is broadly applicable to other large-scale structure observables, including galaxy weak lensing and line-intensity mapping.

In conclusion, the paper establishes that Minkowski functionals are the premier non-Gaussian statistic for CMB lensing, offering substantial, robust improvements in cosmological parameter constraints that are essential for maximizing the scientific return of next-generation CMB experiments.