RM-Tools: Software for Analyzing Polarized Radio Spectra

TL;DR

RM-Tools addresses the need for robust, scalable analysis of polarized radio spectra by uniting RM-synthesis, RM-clean, and QU-fitting within a Python-based open-source toolkit. It supports Stokes I modeling, 1D and 3D data products, and introduces complexity metrics such as sigma_add and M2 to diagnose Faraday complexity, along with a comprehensive suite of utilities for planning, processing, and validating polarization analyses. The package is deployed in major surveys (e.g., POSSUM, VLASS) and enables standardized, reproducible polarization science, while benchmarking reveals reliable QU-fitting performance with recommended nested samplers. Looking forward, the work highlights memory-footprint challenges for 3D RM synthesis and invites community contributions to extend capabilities for the SKA-era data deluge and broader radio-polarization research. RM-synthesis transforms the observed polarization as a function of wavelength-squared, P( o) into the Faraday-depth domain F( ) via a Fourier-like relation, enabling non-parametric exploration of complex magnetic-field structures along the line of sight. This toolkit provides a gallery of models, diagnostics, and interfaces that together empower researchers to extract physically meaningful Faraday structures from noisy, bandwidth-limited data, with quantified uncertainties and explicit metrics for detecting complexity.

Abstract

Polarization observations using modern radio telescopes cover large numbers of frequency channels over broad bandwidths, and require advanced techniques to extract reliable scientific results. We present RM-Tools, analysis software for deriving polarization properties, such as Faraday rotation measures, from spectropolarimetric observations of linearly polarized radio sources. The software makes use of techniques such as rotation measure synthesis and QU-model fitting, along with many features to simplify and enhance the analysis of radio polarization data. RM-Tools is currently the main software that large-area polarization sky surveys such as POSSUM and VLASS deploy for science-ready data processing. The software code is freely available online and can be used with data from a wide range of telescopes.

Figures (6)

• Figure 1: Top-left: Residual $\boldsymbol{q}/\boldsymbol{\sigma}_{q}$ and $\boldsymbol{u}/\boldsymbol{\sigma}_{u}$ spectra of a simple (blue) and complex (red) source after a Faraday thin model derived from RM-synthesis has been subtracted. Dashed lines show the $\pm\,1\sigma$ limits. Bottom-left: Reduced $\chi^2$ as a function of $\sigma_{\rm add}$. Values $<1$ indicate that modeling the residuals with this much $\sigma_{\rm add}$ would make the uncertainties inconsistently large compared to the data. Middle column: The probability density distribution (top) and cumulative density (bottom) of the residual spectra, as compared to a normal distribution (black dashed line). A wider distribution here is indication of unmodeled structure in the residuals. Right column: Marginal posterior probability distributions for the additional scatter term $\sigma_{\rm add}$. Solid vertical lines show the mean of the distribution and dashed vertical lines show the $\pm\,1\sigma$ values. The simple source is best modeled with a very small $\sigma_{\rm add}$ value (consistent with zero) while the complex source requires a larger value.
• Figure 2: Complexity metric values resulting from the uniform slab (left) or turbulent screen (right) models for different S/N and Faraday thickness values. The complexity metrics shown are $\sigma_{add,C}$ (top row) and $M_2$ (bottom row). Each pixel was derived from a single polarized spectrum realization, so inter-pixel variations indicate the influence of spectral noise on the resulting metric value. Blank (white) values in the $M_2$ plots occur occasionally at low S/N when the resulting FDF contains no samples above the CLEAN threshold, and thus no clean components with which to compute the second moment.
• Figure 3: Error distributions (normalized by reported uncertainty) in the Faraday depth (left), polarized intensity (center), and de-rotated polarization angle (right), for RM-synthesis (top row) and QU-fitting (bottom row). All distributions appear well described by a Gaussian of unit width (dashed lines)
• Figure 4: Results of performing QU-fitting on the simulated Faraday simple source 1000 times. Each row represents a different nested sampling sampler used (top to bottom): dynesty, nestle, pymultinest, and ultranest, and the fitted parameters are divided into the three columns. The red dotted lines mark the parameters' input values, and the grey shaded regions show the median output $1\sigma$ uncertainty.
• Figure 5: Similar to Figure \ref{['fig:sampler_test_m1']}, but showing the results from the Faraday complex source. The first Faraday component is shown here, and the second component is shown in Figure \ref{['fig:sampler_test_m11_2']}. The ultranest sampler is omitted here, since it crashed without converging.