Bias from small-scale leakage in Pulsar Timing Array maps

Technical Summary: Bias from Small-Scale Leakage in Pulsar Timing Array Maps

Problem Statement
Pulsar Timing Array (PTA) experiments are approaching the capability to detect anisotropies in the nanohertz gravitational wave background (GWB). However, a fundamental limitation exists: PTA timing residuals are integrated measures of GW power across the entire sky, while the number of monitored pulsars is finite and sparse. Consequently, current analyses can only reconstruct the angular structure of the background at large scales (low multipoles, $\ell$).

The paper identifies a previously unaccounted systematic bias termed "small-scale leakage." This phenomenon arises because the true GWB sky contains power at angular scales smaller than the reconstruction limit ($\ell_{rec}^{max}$). Since the timing residuals integrate power from all scales, the unmodeled small-scale power ($\ell > \ell_{rec}^{max}$) is inevitably projected onto the reconstructed large-scale modes. This leakage artificially inflates the inferred angular power spectrum ($C_\ell$) of the anisotropic background, potentially by more than an order of magnitude, leading to a distorted interpretation of the GWB's origin and structure.

Methodology
The authors develop a complete analytical framework to quantify this bias, moving beyond numerical simulations to derive exact expressions for the leakage effect.

1. Formalism: The GWB anisotropy is decomposed into real spherical harmonic coefficients ($a_{\ell m}$). The measured correlated timing residuals ($R_{meas}$) are modeled as a linear combination of large-scale (LS) and small-scale (SS) contributions plus noise: $R_{meas} = \Gamma_{LS}a_{LS} + \Gamma_{SS}a_{SS} + n$.
2. Reconstruction Bias: The study analyzes the "map-making" process where one attempts to estimate only the LS coefficients ($a_{LS}$) assuming the SS contribution is zero or negligible.
   • Geometrical Intuition: Using a minimum least-squares estimator, the authors show that the recovered coefficients are contaminated by a term proportional to the true small-scale coefficients ($a_{SS}$), mediated by the Moore-Penrose inverse of the geometry matrix.
   • Statistical Intuition: In a likelihood-based framework (Gaussian likelihood), the maximum likelihood estimator shifts due to the incorrect assumption that the data covariance is generated solely by LS modes.
3. Regularization and Noise: The analysis extends to include:
   • Regularization: The impact of ridge regularization (introduced to handle ill-conditioned matrices due to pulsar geometry) on the leakage.
   • Pulsar Noise: The inclusion of realistic pulsar timing noise (red noise) to verify if the effect persists in non-ideal conditions.
4. Simulation: The framework is tested using 1000 realizations of a GWB sourced by a population of supermassive black hole binaries (SMBHBs) distributed in galaxies, with various pulsar configurations (isotropic, uniform, and NANOGrav 15-year dataset subsets).

Key Contributions and Results

• Analytical Derivation of Leakage: The paper derives an explicit expression for the bias in the reconstructed angular power spectrum:
  $$C_{\ell}^{rec} = C_{\ell}^{true} + C_{\ell}^{leakage}$$
  where the leakage term is a sum over unmodeled small-scale multipoles weighted by a mode-mixing function $M_{\ell \ell'}$ that depends exclusively on the pulsar geometry:
  $$C_{\ell}^{leakage} = \sum_{\ell'=\ell_{rec}^{max}+1}^{\ell_{GWB}^{max}} M_{\ell \ell'} C_{\ell'}^{true}$$
• Magnitude of the Bias: The analysis demonstrates that for realistic astrophysical backgrounds, the leakage term dominates the true signal. The reconstructed $C_\ell$ is overestimated by at least one order of magnitude across a wide range of scales, far exceeding the cosmic variance uncertainty.
• Independence of Mitigation Strategies: The authors prove that this systematic error is fundamentally independent of:
  • The specific geometry of the pulsar array.
  • The choice of regularization scheme (e.g., ridge or SVD).
  • The presence of pulsar noise.
  • The specific reconstruction method (spherical harmonics vs. pixel-based).
  • Crucially, increasing the regularization parameter to suppress leakage simultaneously suppresses the true signal, making it impossible to tune the regularization to recover the true power spectrum without prior knowledge of the small-scale power.
• Persistence with Noise: Even when realistic noise budgets (compatible with anisotropy detection) are included, the leakage effect remains significant. Noise alters the covariance matrix but does not eliminate the contamination from unmodeled small-scale power.

Significance and Claims
The paper asserts that small-scale leakage represents a fatal source of error for current and future PTA anisotropy searches if left unaddressed.

• Physical Interpretation: The distorted maps and power spectra caused by leakage are incompatible with large-scale structure patterns. Consequently, any cross-correlation analysis with galaxy surveys (intended to locate sources or study SMBHB clustering) will yield spurious correlations driven by this systematic effect rather than physical clustering.
• Limitations of Current Methods: The authors state that perfect reconstruction of the large-scale structure is impossible without prior knowledge of the small-scale power distribution. Simply increasing the number of pulsars reduces statistical uncertainty but does not eliminate this systematic bias, which will eventually dominate the error budget.
• Future Directions: The paper concludes that the theoretical formalism developed here is essential for estimating the magnitude of this uncertainty. It suggests that future estimators must explicitly model or marginalize over the small-scale power (e.g., using quadratic estimators with nuisance bins) to achieve unbiased reconstruction. The authors emphasize that this issue is general and applies to any truncated reconstruction space, regardless of the basis used.

In summary, the work provides the first rigorous analytical quantification of how unresolved small-scale GW power contaminates large-scale PTA maps, establishing that current reconstruction methods are inherently biased and requiring new statistical strategies to mitigate this leakage.