Hunting Dark Matter Using Artificial Intelligence

From the rotation of galaxies, to the deflection of light as it travels across the cosmos, astronomers have concluded that there is more matter out there than can be seen. But whilst it dominates gravitational pulls across the universe, we simply don’t know what this dark matter is. 

With only gravitational interactions with normal matter, there’s no obvious place for dark matter within modern science, but physicists think it is a particle, an odd member of the family that contains quarks and electrons. Around the globe, experimenters and theorists are striving to uncover dark matter’s secrets. But perhaps the answer will come from the skies?

In 2009, the Fermi Gamma-Ray Space Telescope detected an unexpected signal from the centre of the Milky Way, an excess of high energy gamma rays. Dark matter hunters got very excited as the Galactic Centre should contain a lot of dark matter and they suggested the signal could be coming from annihilation, with the collision of dark matter particles creating the photons. Perhaps we were seeing the tell-tale sign of dark matter!

Others were less excited. The Galactic Centre is a complex, messy place, with other potential sources of gamma-ray emission, with the leading candidate being millisecond pulsars, super dense, rapidly rotating dead star hearts. 

Pulsars earn their rotation when they first form due to the conservation of angular momentum - the in-falling core matter during those last few moments of a supernova increases the rotational energy of the remnant object, causing it to rapidly spin. The original stellar object's magnetic field is also intensely squashed into the remaining pulsar, which is why they are so magnetically powerful.

Millisecond pulsars rotate much faster than regular pulsars - rotating many hundreds of times on their axis per second (as their name suggests). These pulsars, however, have had their rotational velocity increased due to them stealing matter from a nearby companion (increasing angular momentum in the process of mass accretion) - which makes these pulsars much older than regular pulsars. 

Through their powerful, rotating magnetic fields, particles are accelerated to tremendous energies within the pulsar’s magnetosphere - which in turn produce gamma-ray emissions.

Millisecond pulsars that emit gamma-rays have been found distributed across the sky in all directions by orbiting space observatories like Fermi and help scientists learn more about the processes occurring in these supernova relics (including their surrounding environments).

It should be straight-forward to tell these two possibilities apart, as millisecond pulsars should emit gamma rays from discrete points, whereas dark matter annihilation would smoothly emit gamma rays across the Galactic Centre. Unfortunately, gamma rays are difficult to detect, and Fermi’s view is blurry compared to the sharp optical images seen with Hubble. It is not obvious if the emission is smooth or made of points.

Since its discovery, the question of whether the Galactic Centre gamma-ray excess is due to dark matter has become an argument over modelling and statistics, with differing assumptions resulting in different conclusions. With no definitive answer one way or the other, astronomers are now trying something new.

Florian List, a Ph.D. student at the University of Sydney, and Nick Rodd, a researcher at Berkeley, have brought the power of Neural Networks to study the gamma rays from the Galactic Centre. The networks, developed in the search for artificial intelligence, are growing in popularity in science due to their ability to discriminate and classify. With growing computational power, these nets are ideal for scouring immense datasets, searching for astronomical discoveries. 

Neural nets need training data to learn their tasks, and the team generated more than half a million synthetic datasets, accounting for the Galactic disk and bulge and immense regions of emission known as Fermi Bubbles. They also sprinkled in dark matter and emission from pulsars, telling the network the truth of the mix. The network was tested on synthetic images it hadn’t seen before, successfully determining the presence of dark matter or pulsars. 

When fully trained, they fed the actual Fermi data into the network and asked it what it thought was in the image. Alas, the results were not conclusive, but the network decided that a smooth component was present, preferring the existence of dark matter over millisecond pulsars. But like all interesting mysteries in astronomy, more data will be needed before we can truly tell if we are seeing the signature of dark matter, but undoubtedly artificial intelligence will be there to uncover the signal in the noise.