AI Discovers Hidden Extragalactic X-ray Flash in Chandra’s Archival Data

We welcome Steven Dillmann as a guest blogger. Steven is a PhD student in Computational Mathematics at Stanford University, and previously graduated from the University of Cambridge with an MPhil in Data Intensive Science. His research interests revolve around leveraging modern advances in artificial intelligence (AI) to accelerate and enable scientific discovery in data-intensive disciplines like astronomy. Previous experiences include internships at the European Space Agency on studying the impact of satellite constellations on Hubble astronomy and at NASA JPL on studying cloud populations on Mars with citizen science and machine learning.

Together with Rafael Martínez-Galarza (Deputy Director of AstroAI at the Center for Astrophysics | Harvard & Smithsonian and formerly of the Chandra X-ray Center), Steven developed the first representation learning approach to search for rare high-energy transients in the Chandra archive which led to the discovery of the new extragalactic fast X-ray transient XRT 200515. The result was published in a recent paper and was the subject of a press release from the Royal Astronomical Society. More details about the discovery are explained in this post.

Have you ever flipped through old photo albums and suddenly found something fascinating hidden in the background of a picture that no one had ever noticed before? Now imagine doing that on a cosmic scale.

Using a novel machine learning approach, we looked back through over 20 years of archived observations from NASA's Chandra X-ray Observatory and discovered a remarkable, powerful X-ray flash from an unknown object outside our own galaxy that had gone unnoticed for years within the vast Chandra archive — a true needle in the haystack event.

On May 15th 2020, while observing the remains of an exploded star in the Large Magellanic Cloud (a small galaxy neighboring our Milky Way), Chandra accidentally captured something remarkable: an extremely bright and fast X-ray flash from an unknown origin. This flash appeared and disappeared within a few seconds. It went unnoticed during the initial observation and then simply got stored in the large Chandra archive. Unlike traditional approaches, our novel machine learning method managed to uncover this so-called extragalactic fast X-ray transient (FXT), which we named XRT 200515 according to the day it was detected by Chandra.

This cosmic flash is particularly interesting because of its unusual characteristics that are different to any of the other extragalactic (that is, outside the Milky Way) FXTs that have previously been detected by Chandra. It produced an incredibly energetic initial burst that lasted for only 10 seconds, whereas others lasted for minutes or hours. This was followed by a longer, less energetic afterglow lasting for a few minutes. A video of the flash is available on YouTube.

Because Chandra and no other telescope has ever recorded this source before or since this burst, its true nature remains a puzzle. In our study we propose three possible explanations:

1. It could be the first X-ray burster ever discovered in the Large Magellanic Cloud.

These are systems involving two stars: one small and super-dense dead star (called a neutron star) and a normal companion star that orbits around it. The neutron star is like a cosmic vacuum cleaner – its powerful gravity pulls gas off its companion star. When enough gas builds up on the neutron star's surface, it triggers a massive thermonuclear explosion that releases an intense burst of X-ray radiation.

2. It might be a rare flare from one of the rarest and most mysterious objects in the universe: a magnetar.

Magnetars are neutron stars with extremely strong magnetic fields. Giant flares from magnetars are some of the rarest and most explosive events in the cosmos releasing a huge amount of gamma rays in a very short time. If XRT 200515 is an X-ray counterpart to such an event, it would be the very first giant magnetar flare observed at these X-ray energy levels.

3. Lastly, it could be something entirely new — a previously unknown type of cosmic explosion that could reveal new insights about the universe.

This discovery reminds us that space is dynamic and ever-changing, with exciting phenomena occurring constantly. It also demonstrates the value of using artificial intelligence for scientific discovery in archived astronomical data — there might be countless other discoveries waiting to be found in observations we've already made.

We identified thousands of other flares, many from young stars (common), but others may be similar mysterious events outside our galaxy that warrant further investigation. Our team is currently fine-tuning our method to search for signs of planets outside our galaxy, building on previous work led by some of the co-authors of the present work.

As we enter the era of big data, machine learning emerges as an invaluable tool to handle massive, high-dimensional datasets in modern astronomy. It is increasingly used for the detection and characterization of interesting astronomical objects and phenomena in multi-wavelength data sets, and for inference of cosmological parameters from both simulations and astronomical data. Finally, machine learning techniques are also setting a new paradigm for solving complicated numerical problems by replacing expensive simulations with machine learning emulators.