Axion Search via Machine Learning Reveals No Signal in Pulsar Data.

Research identified a specific frequency – half the axion mass – where axion-induced time delays in electromagnetic waves are most pronounced. A machine learning pipeline, achieving 95% accuracy, was applied to pulsar data (PSR J1933-6211) with no axion signal detected within current limits. Future advancements in clock precision and telescope bandwidth promise four orders of magnitude improvement in constraining axion properties.

The search for weakly interacting slim particles (WISPs), including axions, constitutes a significant endeavour in contemporary particle physics, potentially resolving longstanding mysteries concerning dark matter and the strong CP problem. Detecting these elusive particles requires innovative approaches due to their anticipated feeble interactions with ordinary matter. Researchers from the Xinjiang Astronomical Observatory, Guangxi University, and Yunnan Observatories, led by Haihao Shi, have developed a machine learning pipeline to identify subtle time delays in pulsar dispersion measurements that could indicate the presence of axions. Their work, entitled ‘A Machine Learning Pipeline for Hunting Hidden Axion Signals in Pulsar Dispersion Measurements’, details a method achieving 95% classification accuracy in simulated low signal-to-noise data, and applied to observations of the pulsar PSR J1933-6211, currently placing limits on axion properties. The team, comprising Zhenyang Huang, Qiyu Yan, Jun Li, Guoliang Lü, and Xuefei Chen, demonstrate the potential for future radio telescope facilities, such as the Qitai Radio Telescope, to substantially refine constraints on axion decay constants.

Machine Learning Refines Axion Detection via Pulsar Timing

Researchers have developed a machine learning pipeline to enhance the search for axions – hypothetical elementary particles proposed as a component of dark matter – by analysing subtle variations in the arrival times of radio waves from pulsars. Axions, if they exist, are predicted to interact with electromagnetic radiation, inducing frequency-dependent delays in signal propagation. Traditional detection methods struggle with the faintness and complexity of these signals.

This work demonstrates the application of machine learning to a challenging astrophysical problem, potentially improving the sensitivity of dark matter searches using pulsar timing arrays (PTAs). PTAs exploit the remarkably regular radio emissions from millisecond pulsars – rapidly rotating neutron stars – as galactic-scale clocks. Deviations from expected pulse arrival times can indicate the presence of gravitational waves or, as this research proposes, interactions with axions.

Applying the pipeline to data from the millisecond pulsar PSR J1933+2434, the team currently finds no evidence of axion-induced delays within the limits of their current instrumentation. However, they project substantial improvements are possible with future observations. Current constraints on axion properties are limited by the precision of atomic clocks used in radio telescopes. Advancements in clock technology, combined with the increased bandwidth of next-generation telescopes such as the Qitai Radio Telescope, could enhance sensitivity by four orders of magnitude for axions within the relevant mass range.

The developed pipeline achieves 95% accuracy in identifying potential axion signatures in simulations, even under conditions with low signal-to-noise ratios. This represents a methodological advancement in the search for these elusive dark matter candidates. Simulations indicate that this combination of improved instrumentation and machine learning could improve constraints on the axion decay constant – a key parameter defining the strength of its interaction with electromagnetic fields – by four orders of magnitude.

The research confirms a predicted dispersion signal – a characteristic pattern in the timing variations – when the frequency of electromagnetic waves equals half the axion mass. This provides a specific target for future observational strategies, optimising data analysis pipelines and increasing the probability of detection. Further investigation into the relationship between axion properties and observed pulsar timing variations remains a key area for ongoing research.