[Paper Review] Removing non-physical structure in fitted Faraday rotated signals: non-parametric QU-fitting

Luke Pratley, M. Johnston‐Hollitt+1 more|arXiv (Cornell University)|Oct 15, 2020

Geophysics and Gravity MeasurementsEarth and Planetary Sciences44 references5 citations

TL;DR

This paper introduces a convex non-parametric QU-fitting algorithm that enforces zero flux for negative wavelength-squared (λ² ≤ 0), eliminating non-physical Faraday structures in fitted rotation measure spectra. By incorporating a non-negativity constraint on λ² ≤ 0 via primal-dual optimization, the method improves spectral fidelity, reduces root mean square error by up to 30% in simulations, and prevents spurious components that distort scientific interpretation in broadband radio surveys.

ABSTRACT

Next-generation spectro-polarimetric broadband surveys will probe cosmic magnetic fields in unprecedented detail, using the magneto-optical effect known as Faraday rotation. However, non-parametric methods such as RMCLEAN can introduce non-observable linearly polarized flux into a fitted model at negative wavelengths squared. This leads to Faraday rotation structures that are consistent with the observed data, but would be impossible or difficult to measure. We construct a convex non-parametric $QU$-fitting algorithm to constrain the flux at negative wavelengths squared to be zero. This allows the algorithm to recover structures that are limited in complexity to the observable region in wavelength squared. We verify this approach on simulated broadband data sets where we show that it has a lower root mean square error and that it can change the scientific conclusions for real observations. We advise using this prior in next-generation broadband surveys that aim to uncover complex Faraday depth structures. We provide a public Python implementation of the algorithm at \url{https://github.com/Luke-Pratley/Faraday-Dreams}.

Motivation & Objective

• To address the issue of non-physical Faraday structures introduced by unconstrained flux at negative wavelength-squared (λ² ≤ 0) in non-parametric fitting of Faraday rotation signals.
• To develop a convex optimization framework that enforces zero flux for λ² ≤ 0 while preserving sparsity priors used in RMCLEAN-like algorithms.
• To improve the accuracy and physical consistency of Faraday depth models in next-generation broadband spectro-polarimetric surveys.
• To demonstrate that constraining λ² ≤ 0 flux leads to more reliable scientific conclusions about the number and complexity of Faraday components.
• To provide a publicly available, efficient implementation for use in large-scale polarimetric data analysis.

Proposed method

• Formulates a non-parametric QU-fitting problem as a convex optimization task with a non-negativity constraint on flux at λ² ≤ 0.
• Uses a primal-dual algorithm based on Combettes et al. (2014) to solve the constrained minimization problem, enabling non-differentiable and non-continuous constraints.
• Applies the RMCLEAN-style sparsity prior to promote physically plausible, sparse Faraday depth structures.
• Employs Non-Uniform Fast Fourier Transforms (NUFFT) for efficient forward and adjoint operations (Φ and Φ†) in each optimization iteration.
• Implements the constraint by projecting the solution to enforce Pλ²≤0(λ²) = 0 at each iteration, ensuring only observable λ² > 0 contributes to the model.
• Uses a two-step reconstruction process: first fit the model with the constraint, then convolve with the response function to compare with observed data.

Experimental results

Research questions

• RQ1 Can non-physical Faraday structures introduced by unconstrained flux at λ² ≤ 0 be systematically removed from fitted rotation measure spectra?
• RQ2 Does enforcing zero flux at λ² ≤ 0 improve the accuracy and physical consistency of non-parametric QU-fitting in broadband polarimetry?
• RQ3 How does the constrained model compare to unconstrained models in terms of root mean square error and spectral fidelity on simulated and real data?
• RQ4 What is the computational cost of applying the λ² ≤ 0 constraint in large-scale surveys, and can it be scaled efficiently?
• RQ5 Can this method prevent incorrect scientific conclusions about the number of Faraday components in complex magnetic field structures?

Key findings

• The constrained model reduces root mean square error by up to 30% compared to unconstrained fitting in simulated broadband data sets.
• The method successfully suppresses non-physical structures that arise from flux contributions at λ² ≤ 0, which otherwise lead to spurious spectral features.
• In real observations of sources like lmc_c15 and cena_c1972, the constrained model produces a single-peak spectrum, while the unconstrained model introduces multiple spurious components.
• The algorithm converges in 100s to 1,000s of iterations, with each iteration taking less than a second on modern hardware, enabling reconstruction of 1,000,000 lines of sight in approximately one hour using 64 cores.
• The residuals in Q and U Stokes parameters remain consistent between constrained and unconstrained models, confirming that the improvement comes from better spectral structure, not noise amplification.
• The method enables more reliable interpretation of Faraday depth complexity, particularly in interferometric observations of extended sources where fractional polarization can otherwise become non-physical.