arXiv subjects

Wes Armour

Publications and source records attributed to Wes Armour.

At least 19 recordsLinked to original sources

arXiv2026

Discovering Millisecond Pulsars in Globular Clusters with the GMRT (GCGPS)

The Globular Clusters GMRT Pulsar Search (GCGPS) project, launched in mid-May 2023, has emerged as one of the most successful pulsar surveys conducted with the upgraded Giant Metrewave Radio Telescope (uGMRT), leading to the discovery of several new millisecond pulsars (MSPs). The recently concluded Phase I of the survey resulted in the discovery of seven MSPs across four globular clusters (GCs), all of which previously had no known pulsars. These discoveries have enabled the precise determination of the dispersion measures (DMs) for the respective clusters for the first time. Consistent timing follow-up revealed that out of the seven MSPs, three are confirmed binaries, while two are isolated. This paper summarises the design and the implementation of Phase I of the GCGPS project, presents the key scientific results obtained so far, and outlines the strategy and progress of the recently initiated Phase II observations.

astro-ph.HE↗

arXiv2026

The SPOTLIGHT Pulsar Search Pipeline: A GPU-Accelerated FFT Approach

We present the pulsar search component of SPOTLIGHT (Survey for sPoradic radiO bursTs via a commensaL multI-beam Gpu-powered Hpc at the gmrT), a GPU-accelerated commensal backend operating at the upgraded Giant Metrewave Radio Telescope (uGMRT). While SPOTLIGHT is primarily designed for real-time detection and localisation of fast radio bursts (FRBs), it simultaneously records a subset of beamformed data products for periodicity searches without requiring dedicated telescope time. To process the large data volumes generated by the survey, we have developed a scalable FFT-based pulsar search pipeline that combines radio-frequency interference mitigation, GPU-accelerated dedispersion and periodicity searches, multi-beam candidate sifting, efficient folding and machine-learning classification. Using population synthesis and archival uGMRT observations, we estimate that a fully operational SPOTLIGHT survey with 160 PC and one IA beam could discover $\sim$ 450 new pulsars, probing both high-sky coverage and faint pulsars over three and a half years of commensal observations. The pipeline has been validated on GMRT Cycle 48 and 49 observations (i.e. April 2025 to Mar 2026), successfully re-detecting numerous known pulsars with a wide range of Period, DM and flux densities, and is currently operational for SPOTLIGHT commensal data processing. We describe the SPOTLIGHT observing system, pulsar survey design, search parameter space, candidate-selection strategy, current status, and future developments. SPOTLIGHT demonstrates the scientific potential of commensal pulsar surveys and serves as a pathfinder for real-time, large-scale pulsar and transient searches in the SKA era.

astro-ph.HE↗

arXiv2026

The SPOTLIGHT Multibeam Real-Time Transient Detection System

Fast Radio Bursts (FRBs) are among the most enigmatic transient phenomena in the Universe. In order to unravel the mystery behind these events, one requires instruments that possess the ability to search, detect, localise, and capture these events in high resolution over large fields-of-view in real-time. The SPOTLIGHT project is one such backend, leveraging the upgraded Giant Metrewave Radio Telescope (uGMRT) to conduct a commensal search for FRBs and other radio transients, using a dedicated high-performance computing facility, comprised of 90 NVIDIA A100 GPUs and 60 compute servers. Here we present the design, implementation, and performance of SPOTLIGHT's real-time transient search pipeline, a GPU-accelerated system capable of processing up to 2000 post-correlation beams in real time. The pipeline combines AstroAccelerate-powered brute-force dedispersion and single pulse search, with a multi-stage and robust candidate optimisation framework, as well as a triggering system for automatic capture of high-resolution visibility and baseband data. To ensure continuous validation of pipeline performance, we have also developed a real-time signal injection framework capable of injecting synthetic bursts directly into SPOTLIGHT's beamformed data stream. The system operates commensally with routine uGMRT observations, processing data streams in real-time while maintaining high sensitivity to ms-duration transients across dispersion measures extending up to 2000 pc cm$^{-3}$. During its initial deployment in uGMRT Cycle 49 and Cycle 50, the pipeline detected 2870 bursts from 42 known sources, and demonstrated sensitivity consistent with the predicted survey threshold of $\sim$ 0.2 Jy ms. The SPOTLIGHT system establishes a scalable framework for wide-field, low-frequency transient discovery and localisation, and provides a key technological foundation for next-generation radio transient surveys.

astro-ph.IM↗

arXiv2026

Second-Order Multi-Level Variance Correction for Modality Competition in Multimodal Models

Autoregressive next-token training offers a unified formulation for image generation and text understanding, but it also creates strong modality competition that destabilizes optimization and limits large-batch scaling. We show that first-order optimizers such as AdamW are vulnerable to cross-modality gradient heterogeneity, while second-order preconditioning, particularly SOAP, provides a more stable basis for multimodal alignment. Building on this insight, we propose \emph{ML-FOP-SOAP}, a second-order optimization framework with Multi-Level Variance Correction. Our Fisher-Orthogonal Projection suppresses variance-induced modality conflicts, reducing the trade-off between visual generation and textual understanding. To make this practical under large gradient accumulation, we introduce a hierarchical folding strategy that captures fine-grained variance with low micro-step overhead. Experiments on Janus and Emu3 show consistent gains across both modalities and stable training at batch size 8192. Compared with AdamW, our method improves sample efficiency by up to $1.4\times$ and accelerates wall-clock training by up to $1.5\times$, offering a robust optimizer for scaling multimodal foundation models.

cs.CV↗

arXiv2026

Runtime-Orchestrated Second-Order Optimization for Scalable LLM Training

Second-order methods offer an attractive path toward more sample-efficient LLM training, but their practical use is often blocked by the systems cost of maintaining and updating large matrix-based optimizer states. We introduce \textbf{Asteria}, a runtime system designed to remove this bottleneck by separating second-order optimization logic from the critical GPU training path. Rather than keeping all preconditioner state on the accelerator, Asteria dynamically distributes optimizer state across GPU memory, CPU memory, and optional NVMe storage according to architectural constraints and runtime pressure. It further uses training hooks to prepare shadow states in advance, allowing expensive inverse-root computations to proceed asynchronously on the host while GPU computation continues. For distributed training, Asteria employs a bounded-staleness protocol that limits synchronization frequency while preserving optimizer effectiveness through topology-aware coordination. We evaluate Asteria on both memory-constrained and distributed training settings. On a DGX Spark platform with a single GB10 GPU and 128GB unified memory, Asteria supports second-order training for a 1B-parameter language model. On multi-node GH200 systems, it lowers visible optimizer overhead, reduces recurring latency spikes, accelerates convergence in wall-clock time, and maintains the optimization advantages of SOAP and KL-Shampoo in a 7B-parameter language model. Our results suggest that second-order LLM training can be made practical not by simplifying the optimizer alone, but by rethinking how optimizer state, background computation, and distributed synchronization are managed at the runtime level.

cs.DC↗

arXiv2025

Globular Clusters GMRT Pulsar Search (GCGPS) II: Discovery of five MSPs in M69 and M70

This paper reports recent discoveries from the Globular Clusters GMRT Pulsar Search (GCGPS) survey, which aims to uncover pulsars in the globular clusters (GCs) of the Milky Way using the upgraded Giant Metrewave Radio Telescope (uGMRT). Utilising the Band-4 (550$-$750 MHz) and Band-3 (300$-$500 MHz) receivers, the survey targets GCs accessible to uGMRT ($-53^\circ\,<\,\delta\,<\,-17^\circ$), excluding the declination range that can be covered by the Five-hundred-meter Aperture Spherical radio Telescope (FAST). The survey focuses on GCs that have not previously been searched with comparable sensitivity in these radio frequencies. In this paper, we present the discovery of the five MSPs in two GCs, $-$ NGC~6637 (M69) and NGC~6681 (M70), each hosting MSPs identified here for the first time. Observations of M69 led to the discovery of two MSPs: J1831$-$3220A (M69A) and J1831$-$3220B (M69B), both of which we localize with arcsecond precision using interferometric imaging. Observations of M70 resulted in three new MSPs: J1843$-$3217A (M70A), J1843$-$3217B (M70B), and J1843$-$3217C (M70C). Although direct imaging did not yield precise localizations for these MSPs, we provide initial estimates based on uGMRT beam forming and imaging analysis. Additionally, we present preliminary imaging results for other observed GCs, and in cases of non-detections, we report upper limits on pulsed emission based on the rms noise levels in the image plane.

astro-ph.HE↗

arXiv2025

Globular Clusters GMRT Pulsar Search (GCGPS) I: Survey description, discovery and timing of the first pulsar in NGC 6093 (M80)

This paper describes the new Globular Clusters GMRT Pulsar Search (GCGPS) survey. This survey aims to find MSPs in the globular clusters (GCs) of the Milky Way using uGMRT. The observations use the uGMRT's Band-4 (550$-$750 MHz) and Band-3 (300$-$500 MHz) receivers, which are well suited for steep-spectral-index radio sources like MSPs; the survey will eventually cover the GCs accessible to the uGMRT sky (i.e. $\delta\:>\:\sim\:-\:53^\circ$), and that is South of $\delta = -17^\circ$ (FAST sky limit) and have not been targeted with the sensitivity of this survey. The observations started in May 2023, having so far resulted in seven new discoveries. In this paper, we present the discovery and follow-up study of the first pulsar from this survey, J1617$-$2258A, a 4.32 ms binary MSP that is also the first to be discovered in the globular cluster NGC 6093. We localised this MSP with arc-sec precision from imaging and obtained the unique timing solution from more than one year of timing observations with the Band-4 (550$-$750 MHz) receivers of the uGMRT. This revealed an unusual binary MSP, with a $\sim$ 19-hour, highly eccentric (e $\sim$ 0.54) orbit having a low-mass companion. This orbital eccentricity allowed the measurement of the rate of advance of periastron for this system, which led to the derivation of its total mass, $1.67 \, \pm \, 0.06 \, \rm M_{\odot}$; this together with the system's mass function implies, for the pulsar and the companion, $M_\mathrm{p} < 1.60 \, \rm M_{\odot}$ and $M_\mathrm{c} > 0.072 \, \rm M_{\odot}$. The system is likely a perturbed MSP-Helium WD system seen at a low orbital inclination.

astro-ph.HE↗

arXiv2022

Cutting the cost of pulsar astronomy: Saving time and energy when searching for binary pulsars using NVIDIA GPUs

Using the Fourier Domain Acceleration Search (FDAS) method to search for binary pulsars is a computationally costly process. Next generation radio telescopes will have to perform FDAS in real time, as data volumes are too large to store. FDAS is a matched filtering approach for searching time-domain radio astronomy datasets for the signatures of binary pulsars with approximately linear acceleration. In this paper we will explore how we have reduced the energy cost of an SKA-like implementation of FDAS in AstroAccelerate, utilising a combination of mixed-precision computing and dynamic frequency scaling on NVIDIA GPUs. Combining the two approaches, we have managed to save 58% of the overall energy cost of FDAS with a (<3%) sacrifice in numerical sensitivity.

astro-ph.IM↗

arXiv2021

Implementing CUDA Streams into AstroAccelerate -- A Case Study

To be able to run tasks asynchronously on NVIDIA GPUs a programmer must explicitly implement asynchronous execution in their code using the syntax of CUDA streams. Streams allow a programmer to launch independent concurrent execution tasks, providing the ability to utilise different functional units on the GPU asynchronously. For example, it is possible to transfer the results from a previous computation performed on input data n-1, over the PCIe bus whilst computing the result for input data n, by placing different tasks in different CUDA streams. The benefit of such an approach is that the time taken for the data transfer between the host and device can be hidden with computation. This case study deals with the implementation of CUDA streams into AstroAccelerate. AstroAccelerate is a GPU accelerated real-time signal processing pipeline for time-domain radio astronomy.

astro-ph.IM↗

arXiv2019

GREENBURST: a commensal fast radio burst search back-end for the Green Bank Telescope

We describe the design and deployment of GREENBURST, a commensal Fast Radio Burst (FRB) search system at the Green Bank Telescope. GREENBURST uses the dedicated L-band receiver tap to search over the 960$-$1920 MHz frequency range for pulses with dispersion measures out to $10^4$ pc cm$^{-3}$. Due to its unique design, GREENBURST will obtain data even when the L-band receiver is not being used for scheduled observing. This makes it a sensitive single pixel detector capable of reaching deeper in the radio sky. While single pulses from Galactic pulsars and rotating radio transients will be detectable in our observations, and will form part of the database we archive, the primary goal is to detect and study FRBs. Based on recent determinations of the all-sky rate, we predict that the system will detect approximately one FRB for every 2$-$3 months of continuous operation. The high sensitivity of GREENBURST means that it will also be able to probe the slope of the FRB source function, which is currently uncertain in this observing band.

astro-ph.IM↗

arXiv2018

Science Pipelines for the Square Kilometre Array

The Square Kilometre Array (SKA) will be both the largest radio telescope ever constructed and the largest Big Data project in the known Universe. The first phase of the project will generate on the order of 5 zettabytes of data per year. A critical task for the SKA will be its ability to process data for science, which will need to be conducted by science pipelines. Together with polarization data from the LOFAR Multifrequency Snapshot Sky Survey (MSSS), we have been developing a realistic SKA-like science pipeline that can handle the large data volumes generated by LOFAR at 150 MHz. The pipeline uses task-based parallelism to image, detect sources, and perform Faraday Tomography across the entire LOFAR sky. The project thereby provides a unique opportunity to contribute to the technological development of the SKA telescope, while simultaneously enabling cutting-edge scientific results. In this paper, we provide an update on current efforts to develop a science pipeline that can enable tight constraints on the magnetised large-scale structure of the Universe.

astro-ph.IM↗

arXiv2017

ALFABURST: A commensal search for Fast Radio Bursts with Arecibo

ALFABURST has been searching for Fast Radio Bursts (FRBs) commensally with other projects using the Arecibo L-band Feed Array (ALFA) receiver at the Arecibo Observatory since July 2015. We describe the observing system and report on the non-detection of any FRBs from that time until August 2017 for a total observing time of 518 hours. With current FRB rate models, along with measurements of telescope sensitivity and beam size, we estimate that this survey probed redshifts out to about 3.4 with an effective survey volume of around 600,000 Mpc$^3$. Based on this, we would expect, at the 99% confidence level, to see at most two FRBs. We discuss the implications of this non-detection in the context of results from other telescopes and the limitation of our search pipeline. During the survey, single pulses from 17 known pulsars were detected. We also report the discovery of a Galactic radio transient with a pulse width of 3 ms and dispersion measure of 281 pc cm$^{-3}$, which was detected while the telescope was slewing between fields.

astro-ph.IM↗

arXiv2017

Initial Results from the ALFABURST Survey

Here, we present initial results from the ALFABURST radio transient survey, which is currently running in a commensal mode with the ALFA receiver at the Arecibo telescope. We observed for a total of 1400 hours and have detected single pulses from known pulsars but did not detect any FRBs. The non-detection of FRBs is consistent with the current FRB sky rates.

astro-ph.HE↗

arXiv2017

SETIBURST: A Robotic, Commensal, Realtime Multi-Science Backend for the Arecibo Telescope

Radio astronomy has traditionally depended on observatories allocating time to observers for exclusive use of their telescopes. The disadvantage of this scheme is that the data thus collected is rarely used for other astronomy applications, and in many cases, is unsuitable. For example, properly calibrated pulsar search data can, with some reduction, be used for spectral line surveys. A backend that supports plugging in multiple applications to a telescope to perform commensal data analysis will vastly increase the science throughput of the facility. In this paper, we present 'SETIBURST', a robotic, commensal, realtime multi-science backend for the 305-m Arecibo Telescope. The system uses the 1.4 GHz, seven-beam Arecibo L-band Feed Array (ALFA) receiver whenever it is operated. SETIBURST currently supports two applications: SERENDIP VI, a SETI spectrometer that is conducting a search for signs of technological life, and ALFABURST, a fast transient search system that is conducting a survey of fast radio bursts (FRBs). Based on the FRB event rate and the expected usage of ALFA, we expect 0-5 FRB detections over the coming year. SETIBURST also provides the option of plugging in more applications. We outline the motivation for our instrumentation scheme and the scientific motivation of the two surveys, along with their descriptions and related discussions.

astro-ph.IM↗

arXiv2016

A polyphase filter for many-core architectures

In this article we discuss our implementation of a polyphase filter for real-time data processing in radio astronomy. We describe in detail our implementation of the polyphase filter algorithm and its behaviour on three generations of NVIDIA GPU cards, on dual Intel Xeon CPUs and the Intel Xeon Phi (Knights Corner) platforms. All of our implementations aim to exploit the potential for data reuse that the algorithm offers. Our GPU implementations explore two different methods for achieving this, the first makes use of L1/Texture cache, the second uses shared memory. We discuss the usability of each of our implementations along with their behaviours. We measure performance in execution time, which is a critical factor for real-time systems, we also present results in terms of bandwidth (GB/s), compute (GFlop/s) and type conversions (GTc/s). We include a presentation of our results in terms of the sample rate which can be processed in real-time by a chosen platform, which more intuitively describes the expected performance in a signal processing setting. Our findings show that, for the GPUs considered, the performance of our polyphase filter when using lower precision input data is limited by type conversions rather than device bandwidth. We compare these results to an implementation on the Xeon Phi. We show that our Xeon Phi implementation has a performance that is 1.47x to 1.95x greater than our CPU implementation, however is not insufficient to compete with the performance of GPUs. We conclude with a comparison of our best performing code to two other implementations of the polyphase filter, showing that our implementation is faster in nearly all cases. This work forms part of the Astro-Accelerate project, a many-core accelerated real-time data processing library for digital signal processing of time-domain radio astronomy data.

astro-ph.IM↗

arXiv2016

Commissioning of ALFABURST: initial tests and results

Fast Radio Bursts (FRBs) are apparently one-time, relatively bright radio pulses that have been observed in recent years. The origin of FRBs is currently unknown and many instruments are being built to detect more of these bursts to better characterize their physical properties and identify the source population. ALFABURST is one such instrument. ALFABURST takes advantage of the 7-beam Arecibo L-band Feed Array (ALFA) receiver on the 305-m Arecibo Radio Telescope in Puerto Rico, to detect FRBs in real-time at L-band (1.4 GHz). We present the results of recent on-sky tests and observations undertaken during the commissioning phase of the instrument. ALFABURST is now available for commensal observations with other ALFA projects.

astro-ph.IM↗

arXiv2015

ALFABURST: A realtime fast radio burst monitor for the Arecibo telescope

Fast radio bursts (FRBs) constitute an emerging class of fast radio transient whose origin continues to be a mystery. Realizing the importance of increasing coverage of the search parameter space, we have designed, built, and deployed a realtime monitor for FRBs at the 305-m Arecibo radio telescope. Named 'ALFABURST', it is a commensal instrument that is triggered whenever the 1.4 GHz seven-beam Arecibo $L$-Band Feed Array (ALFA) receiver commences operation. The ongoing commensal survey we are conducting using ALFABURST has an instantaneous field of view of 0.02 sq. deg. within the FWHM of the beams, with the realtime software configurable to use up to 300 MHz of bandwidth. We search for FRBs with dispersion measure up to 2560 cm$^{-3}$ pc and pulse widths ranging from 0.128 ms to 16.384 ms. Commissioning observations performed over the past few months have demonstrated the capability of the instrument in detecting single pulses from known pulsars. In this paper, I describe the instrument and the associated survey.

astro-ph.IM↗

arXiv2015

Strong Interaction Effects at a Fermi Surface in a Model for Voltage-Biased Bilayer Graphene

Monte Carlo simulation of a 2+1 dimensional model of voltage-biased bilayer graphene, consisting of relativistic fermions with chemical potential mu coupled to charged excitations with opposite sign on each layer, has exposed non-canonical scaling of bulk observables near a quantum critical point found at strong coupling. We present a calculation of the quasiparticle dispersion relation E(k) as a function of exciton source j in the same system, employing partially twisted boundary conditions to boost the number of available momentum modes. The Fermi momentum k_F and superfluid gap Delta are extracted in the limit j tends to zero for three different values of mu, and support a strongly interacting scenario at the Fermi surface with Delta of order O(mu). We propose an explanation for the observation mu < k_F in terms of a dynamical critical exponent z < 1.

cond-mat.str-el↗