Exploring pulsar profiles at population scale from a graph theory perspective

The full Square Kilometer Array is expected to increase the population of currently known pulsars (~2,900) by more than 10-fold, a large fraction of which should already be found during Phase I. Now is therefore a good time to research and develop automated methods for knowledge extraction that require little to no human intervention. In this talk, we explore the task of automatically sequencing and classifying pulsar profiles by considering their morphology at different levels (total intensity and polarization). 

The integrated pulse profile of radio pulsars—composed of one or more components distributed over the rotational phase—represent one of their defining characteristics. The pulsar profile at a given observing frequency is generally stable over decades, and often evolves with frequency, a key to study the structure of its radio emitting magnetosphere. To date, efforts to classify and interpret pulsar profile morphology were performed to a large extent by eye, sometimes guided by supervised methods (e.g. evaluate the number of fitting components). Conversely, unsupervised methods have the potential to unravel faint features and possibly unknown scaling and morphological relations between different pulsars.

Here, we present the status of an on-going effort to develop an unsupervised analysis framework that automatically sequences and classifies integrated pulse profiles by applying principles of graph theory using the semi-structured European Pulsar Network database as a case study. In addition, our custom codebase now permits to easily handle a collection of pulsar data and metadata, making it possible to perform post-processing and analysis within one coherent and expandable framework. We discuss analysis results, their potential physical implications, and lessons learned.