PulsarX: High-Performance Pulsar Folding

• PulsarX is a pulsar searching package that integrates RFI mitigation, dedispersion via the pruned FDMT, folding, and parameter optimization to efficiently process large candidate sets.
• It reduces computational load by reusing dedispersion computations, cutting operations by a factor of 50 for around 500 candidates in modern, wide field-of-view surveys.
• It has been successfully deployed in MeerKAT surveys, where it functions alongside tools like PRESTO to offer acceleration-aware and polarization-sensitive candidate analysis.

PulsarX is a pulsar searching package introduced as “a new pulsar searching package – I. A high performance folding program for pulsar surveys,” with its initial emphasis on the computational bottleneck of candidate folding in modern radio pulsar surveys (Men et al., 2023). It addresses the demands created by wide field-of-view surveys, radio interferometers, and real or quasi-real-time processing by providing a complete folding pipeline that includes radio frequency interference (RFI) mitigation, dedispersion, folding, and parameter optimization. Its principal methodological contributions are an optimized dedispersion strategy, the pruned Fast Discrete Dispersion Measure Transform (pFDMT), and a folding algorithm based on the Tikhonov-regularised least squares method (TLSM), with later work showing its integration into full-Stokes, acceleration-aware survey pipelines (Desvignes et al., 11 Jul 2025).

1. Problem domain and software scope

PulsarX was developed in response to the fact that pulsar surveys with modern radio telescopes are becoming increasingly computationally demanding, especially for wide field-of-view pulsar surveys with radio interferometers and for surveys conducted in real or quasi-real time (Men et al., 2023). In that setting, the cost of candidate folding can become a data-analysis bottleneck that limits the parameter space covered by a survey and reduces scientific return.

Within that context, PulsarX is not described merely as a single folding routine. The original paper presents it as a complete folding pipeline for large-scale pulsar surveys, including RFI mitigation, dedispersion, folding, and parameter optimization (Men et al., 2023). The emphasis on “candidate folding” is important: the package is aimed at the stage where large numbers of pulsar candidates, already generated by earlier search stages, must be evaluated efficiently and consistently.

This positioning distinguishes PulsarX from survey software focused primarily on Fourier-domain periodicity searches or single-pulse searches. A later Galactic Centre survey makes this division explicit: PulsarX was used together with PRESTO and TransientX, with PRESTO providing the Fourier search and TransientX the single-pulse search, while PulsarX handled dedispersion, folding, and candidate analysis in the periodic-search branch (Desvignes et al., 11 Jul 2025). This suggests that PulsarX is best understood as high-performance survey infrastructure centered on the folding stage, while remaining interoperable with broader pulsar-search ecosystems.

2. Pipeline structure and survey workflow

The original PulsarX paper defines a complete survey-facing workflow composed of RFI mitigation, dedispersion, folding, and parameter optimization (Men et al., 2023). That description is concise, but it establishes the intended scope: PulsarX is designed for bulk candidate processing rather than for isolated, manual follow-up of a small number of sources.

A later deployment in a survey of the 10 pc region around Sgr A* provides a detailed operational picture of how PulsarX can sit inside a larger processing chain (Desvignes et al., 11 Jul 2025). In that implementation, calibrated full-Stokes data were converted to PSRFITS search-mode format; after calibration, the periodic-search pipeline split into a PRESTO-based Fourier search and a PulsarX stage that performed dedispersion, some RFI filtering, folding, and polarization-aware candidate analysis.

In that survey, PulsarX dedispersed Stokes $I$ over $0 \leq \mathrm{DM} \leq 5000~\mathrm{pc~cm^{-3}}$ with a DM step of $2~\mathrm{pc~cm^{-3}}$, while simultaneously applying a zero-DM filter and a Kadane filter (Desvignes et al., 11 Jul 2025). Candidates identified by PRESTO were then folded by PulsarX using full-Stokes data and the acceleration and jerk solutions returned by the Fourier search. PulsarX also generated enhanced candidate plots with polarization diagnostics, including RM-synthesis output, for manual scoring.

This later workflow does not redefine the original package, but it shows that PulsarX was engineered to function as a survey component that bridges dedispersion, folding, and candidate inspection. A plausible implication is that its architecture was shaped not only by algorithmic speed, but also by the need to connect efficiently to existing calibration, search, and vetting toolchains.

3. Core algorithms: pFDMT and TLSM folding

The main algorithmic contribution in the original paper is the use of the Fast Discrete Dispersion Measure Transform (FDMT) in an optimized implementation termed the pruned FDMT, or pFDMT (Men et al., 2023). The package leverages the FDMT algorithm proposed by Zackay et al. (2017), but specifically emphasizes an implementation that is both optimized and cache-friendly.

The defining feature of pFDMT is that it efficiently reuses intermediate processing results and prunes unused computation paths, yielding a significant reduction in arithmetic operations (Men et al., 2023). The paper links this directly to the simultaneous folding of large numbers of candidates, where repeated brute-force dedispersion would otherwise scale with the number of candidates. For approximately 500 candidates, the theoretical number of dedispersion operations can be reduced by a factor of around 50 relative to brute-force dedispersion (Men et al., 2023).

The second named methodological contribution is a novel folding algorithm based on the Tikhonov-regularised least squares method, abbreviated TLSM (Men et al., 2023). The explicit claim made for TLSM is that it can improve the time resolution of the pulsar profile. The abstract does not provide the full mathematical development, but the formulation indicates that profile reconstruction is treated as a regularised inverse problem rather than as a purely binning-based accumulation step.

Taken together, pFDMT and TLSM address two different computational aspects of folding. pFDMT reduces the cost of preparing candidate-specific dedispersed streams, whereas TLSM targets the quality of the recovered folded profile (Men et al., 2023). This suggests a design philosophy in which throughput optimization and profile fidelity are treated as coupled requirements rather than as separate engineering problems.

4. MeerKAT deployment and performance claims

PulsarX was presented not only as a methodological proposal but also as an operational component of two MeerKAT projects: the MPIfR-MeerKAT Galactic Plane Survey (MMGPS) and the Transients and Pulsars with MeerKAT (TRAPUM) project (Men et al., 2023). The paper states that the performance of its real-world application is presented as an integral part of those survey efforts.

The key quantitative performance claim in the abstract concerns dedispersion reuse across multiple candidates. In the reported processing, for approximately 500 candidates, the theoretical number of dedispersion operations can be reduced by a factor of around 50 relative to brute-force dedispersion (Men et al., 2023). Because brute-force dedispersion scales with candidate count, this reduction is especially relevant when candidate lists are large.

The significance of this claim lies in the survey regime PulsarX targets. Large interferometric surveys and near-real-time pipelines are constrained by throughput, memory locality, and repeated processing of related candidates. The pFDMT strategy directly addresses that regime by exploiting overlap in intermediate computations (Men et al., 2023).

The original paper therefore places PulsarX within a broader transition in pulsar searching: improvements in telescope capability were making folding, rather than only search-stage transforms, a limiting factor. PulsarX’s contribution was to recast folding as a shared-computation problem across candidate sets rather than as a strictly candidate-by-candidate workflow.

5. Later extensions: full-Stokes folding, RM synthesis, and candidate scoring

A later survey around Sgr A* documents an expanded use of PulsarX in a demanding radio-search environment and shows capabilities not described in the original abstract, especially in polarization-aware candidate analysis (Desvignes et al., 11 Jul 2025). In that work, PulsarX was used on calibrated full-Stokes Effelsberg data between 4 and 8 GHz, wrapped around a PRESTO-based periodicity search.

For periodicity processing, the survey used a two-pass Fourier strategy in PRESTO, comprising a low-acceleration pass and an acceleration-plus-jerk pass. PulsarX then folded the remaining candidates using full-Stokes data and the acceleration and jerk information from PRESTO (Desvignes et al., 11 Jul 2025). The standard relations used in that survey were

$$z = \frac{a f T_\textrm{int}^2}{c}, \qquad w = \frac{j f T_\textrm{int}^3}{c},$$

with $f$ the spin frequency, $a$ the line-of-sight acceleration, $j$ the line-of-sight jerk, and $T_\textrm{int}=1~\mathrm{h}$ (Desvignes et al., 11 Jul 2025). In practical terms, this shows that PulsarX could fold candidates using nontrivial orbital-motion solutions rather than only static ephemerides.

The same survey also embedded Faraday RM synthesis into candidate diagnostics. For each folded candidate with a defined pulse window, PulsarX searched RM values from $-100{,}000$ to $+100{,}000~\mathrm{rad~m^{-2}}$ with an RM step of $0 \leq \mathrm{DM} \leq 5000~\mathrm{pc~cm^{-3}}$0, maximizing total linear polarization

$0 \leq \mathrm{DM} \leq 5000~\mathrm{pc~cm^{-3}}$1

after derotation across frequency (Desvignes et al., 11 Jul 2025). The resulting RM spectrum and best-RM estimate were shown in the candidate plots and used in manual scoring.

This polarization-aware extension is notable because the later survey states explicitly that polarization information was used in candidate scoring (Desvignes et al., 11 Jul 2025). PulsarX thus became, in that deployment, not just a high-performance folder but also a locus for astrophysically informed candidate vetting. The RM-synthesis implementation was validated on PSR J1746–2850, where PulsarX recovered an RM of $0 \leq \mathrm{DM} \leq 5000~\mathrm{pc~cm^{-3}}$2, consistent with $0 \leq \mathrm{DM} \leq 5000~\mathrm{pc~cm^{-3}}$3 from independent work (Desvignes et al., 11 Jul 2025).

6. Practical role, boundaries, and significance

PulsarX occupies a specific place in the pulsar-search software stack. It is a folding-centered survey package with explicit support for large candidate sets, shared dedispersion computation, and parameter optimization, rather than a monolithic replacement for all search software (Men et al., 2023). Later practice reinforces this interpretation: in the Galactic Centre survey, PRESTO remained the Fourier search engine and TransientX handled single-pulse searches, while PulsarX provided dedispersion, RFI filtering, folding, and full-Stokes candidate analysis for the periodic branch (Desvignes et al., 11 Jul 2025).

This division of labor addresses a common misconception that “pulsar search software” necessarily denotes a single end-to-end executable. In PulsarX’s documented deployments, the package is better understood as a specialized, high-throughput component designed to remove the folding bottleneck and to integrate tightly with adjacent packages (Men et al., 2023).

Its limitations are also visible in practice. In the Sgr A* survey, folding full-Stokes data with jerk information was memory-intensive, and only the first 210 candidates from each search pass were folded because of memory limits on the computing nodes (Desvignes et al., 11 Jul 2025). More broadly, that survey concluded that its lack of sensitivity to a population of millisecond pulsars was due to Galactic Centre background, scattering, and single-dish limitations rather than to PulsarX’s configuration (Desvignes et al., 11 Jul 2025). This indicates that software efficiency alone does not overcome propagation and system-noise constraints.

The enduring significance of PulsarX is therefore methodological and infrastructural. It reframed candidate folding as a major computational problem in its own right, proposed a dedispersion-sharing strategy through pFDMT, introduced a TLSM-based folding method to improve profile time resolution, and later demonstrated compatibility with full-Stokes, acceleration- and jerk-aware survey pipelines (Men et al., 2023). In large radio surveys, especially those seeking to preserve search breadth under strict compute budgets, that combination makes PulsarX a representative example of how folding software became a first-order determinant of survey efficiency rather than a downstream afterthought.