A gaussian-process reconstruction of gravitational-wave friction from dark standard sirens

Technical Summary: Gaussian-Process Reconstruction of Gravitational-Wave Friction from Dark Standard Sirens

Problem Statement
Gravitational-wave (GW) standard sirens offer a direct probe of the cosmic distance–redshift relation, providing a powerful test of GW propagation on cosmological scales. In extensions of General Relativity (GR), GW propagation may be modified by an anomalous friction term, altering the amplitude evolution and resulting in a GW luminosity distance ($d_L^{GW}$) that differs from its electromagnetic (EM) counterpart ($d_L^{EM}$). Existing constraints on modified propagation have largely relied on theory-motivated parametric forms, such as the $\Xi_0$–$n$ ansatz for the distance ratio $\Xi(z) \equiv d_L^{GW}(z)/d_L^{EM}(z)$ or specific functional assumptions for the friction function $\alpha_M(z)$ (e.g., tracking the dark energy density). These parametric approaches may not span the full theory space (including DHOST, extra-dimensional, and bigravity scenarios) and introduce functional assumptions that can bias constraints. The paper addresses the need for a non-parametric, weakly modelled reconstruction that allows the data to directly inform the redshift dependence of modified GW propagation without committing to a fixed functional form.

Methodology
The authors develop a Gaussian Process (GP) reconstruction tailored specifically to the GW friction function $\alpha_M(z)$, rather than modeling the distance ratio $\Xi(z)$ directly. The methodology involves several key technical components:

1. Physical Framework: The reconstruction operates within the luminal, massless, source-free sector of GW propagation. The propagation equation includes a friction term $\alpha_M(z)$, which relates to the distance ratio via $\Xi(z) = \exp\left[\frac{1}{2} \int_0^z \frac{\alpha_M(z')}{1+z'} dz'\right]$. The EM distance is assumed to follow a flat $w_0$CDM background.
2. Monotonicity Enforcement: Since GW observations constrain luminosity distance rather than redshift, the likelihood requires the inverse map $z(d_L^{GW})$. To ensure this map is well-defined, the authors enforce that $d_L^{GW}(z)$ is strictly monotonic. They achieve this by modeling the logarithmic derivative $q(z) \equiv \frac{d}{dz} \ln d_L^{GW}(z)$ using a positive reparametrization involving a latent GP $f(x)$ (where $x = \ln(1+z)$) and a softplus function. This guarantees $q(z) > 0$, thereby ensuring a monotonic distance and a valid $\alpha_M(z)$ bound.
3. Resolution-Adapted Prior: The baseline GP uses a non-stationary Gibbs kernel in the coordinate $x = \ln(1+z)$. The local correlation length $\ell_{eff}(x)$ is a function of the data's redshift resolution, derived from the distance posteriors of individual events in the catalog. This allows for shorter correlation lengths (higher flexibility) where the catalog has better resolution (nearby, loud events) and smoother behavior where resolution is poor.
4. Hierarchical Inference: The method is applied to the GWTC-5.0 binary black hole (BBH) catalog using hierarchical Bayesian inference. The model jointly infers:
   • Background cosmological parameters ($H_0, \Omega_m, w_0$).
   • The BBH population properties (mass distribution, merger rate density).
   • The modified propagation history (GP hyperparameters and latent coefficients).
   • The total parameter space includes 235 degrees of freedom.
5. Diagnostics and Calibration: To ensure robustness and interpretability, the authors introduce function-space diagnostics:
   • Prior Calibration: The GP amplitude prior is calibrated to match the integrated prior width of the standard $\Xi_0$–$n$ parameterization in event-binned log $\Xi$ space, ensuring fair comparison.
   • Contraction Metrics: Measures of prior-to-posterior contraction quantify how much information the data provides.
   • Function-Space Summaries: A summary variable $\Delta\langle \log \Xi \rangle_{high-low}$ is used to compare propagation-population degeneracies between parametric and non-parametric models.
   • Principal Component Analysis (PCA): Used to analyze the covariance structure of the posterior, determining if uncertainty is concentrated in a single coherent mode or distributed across multiple independent directions.

Key Results
Applying the method to the GWTC-5.0 catalog, the authors find:

• Consistency with GR: There is no evidence for departures from GR. The reconstructed friction function $\alpha_M(z)$ is consistent with 0, and the distance ratio $\Xi(z)$ is consistent with 1 across the redshift range probed by the catalog.
• Stability: The result is stable under variations of the GP kernel (Gibbs vs. Matérn-5/2), smoothness priors, resolution prescriptions, and likelihood treatments. While the local reconstruction of $\alpha_M(z)$ shows sensitivity to these choices, the accumulated distance ratio $\Xi(z)$ remains remarkably stable.
• Comparison with Parametric Models: The GP reconstruction is mutually consistent with the standard $\Xi_0$–$n$ parameterization. However, the models organize information differently:
  • The $\Xi_0$–$n$ posterior is nearly rank-one, meaning uncertainty is concentrated in a single global deformation mode.
  • The GP posterior retains a higher-rank structure (effective rank $\approx 1.56$), allowing for more flexible, local variations while still being constrained by the data.
  • After prior calibration, the GP posterior contracts more strongly (factor of 5.21) than the $\Xi_0$–$n$ posterior (factor of 2.82), indicating the data provides more specific information when the functional form is not overly restrictive.
• Degeneracies: The propagation–rate degeneracy (correlation between the redshift evolution of the merger rate and the GW distance ratio) persists in the GP reconstruction, though it is expressed in function space rather than a single parameter.

Significance and Claims
The paper claims to provide a framework for interpretable non-parametric tests of gravitational-wave propagation. Its significance lies in:

1. Physical Explicitness: By reconstructing the local friction $\alpha_M(z)$ first and deriving $\Xi(z)$ via integration, the method maintains an explicit connection between the propagation equation and the observed distance modification, unlike approaches that model $\Xi(z)$ directly.
2. Robustness Diagnostics: The work introduces a suite of tools (amplitude calibration, contraction metrics, PCA) that allow researchers to distinguish which features of a reconstruction are informed by the data versus those imposed by the function-space prior. This addresses the challenge of interpreting flexible, non-parametric models.
3. Validation: The method is validated on simulated catalogs, successfully recovering both the GR limit and a complex, non-standard DHOST-inspired propagation history that falls outside standard parametric forms.
4. Current Constraints: The application to GWTC-5.0 reinforces the conclusion that current data is consistent with GR, but does so with a more general strategy that does not rely on specific theoretical ansatzes for the redshift dependence of modified gravity.

The authors explicitly note that their analysis is restricted to luminal, massless propagation with standard GR waveform generation. They state that extending the analysis to non-luminal propagation (time-dependent $c_T(z)$) would require a joint model for mass redshifting and amplitude propagation, which is left for future work. Similarly, relaxing the monotonicity assumption would require a different likelihood formulation to handle non-invertible distance-redshift maps.