
Neutrinos from our galaxy, the Milky Way, could have been detected by the IceCube Neutrino Observatory, a large underground neutrino detector at the South Pole, using new artificial intelligence (AI) methods.
IceCube produces a constantly large amount of data, which includes a large majority of unwanted background. To utilize the data for studying astrophysical objects, as much background as possible must be removed from the measured events in the detector. Afterwards, the properties of the detected neutrinos have to be reconstructed using the cleaned data. The new AI tools improve the data analysis, especially the removal of unwanted background and the reconstruction of the neutrinos’ properties.
The Milky Way
The Milky Way is the galaxy where our solar system and Earth are located. It comprises a collection of billions of stars, along with gas, dust, and dark matter, forming a spiral shape with several arms swirling around its center. From Earth, it looks like a glowing, milky band stretching across the sky. This band is caused by the dense concentration of stars in the Milky Way. However, in everyday terms, the stars inside a galaxy are still very far apart from each other.
Our galaxy emits electromagnetic radiation with numerous different wavelengths. By detecting and studying it, we can learn about the structure of our galaxy and its different sources of electromagnetic radiation. And by studying one galaxy, scientists learn about how galaxies form in general.
But there is a problem using electromagnetic radiation to study the Milky Way. Electromagnetic radiation can interact with matter on its way to Earth so that it gets absorbed or changes its wavelength.
One solution to this problem is using neutrinos as messenger particles, which are expected to be created in the Milky Way in some of the same processes as electromagnetic radiation. Also, they have some advantages compared to electromagnetic radiation.
Neutrinos
Neutrinos are extremely light particles that rarely interact with matter. Therefore, they can escape from regions in the universe where even electromagnetic radiation does not and allow us to explore regions of our universe that cannot be studied in any other way. They also do not carry an electrical charge like electrons and protons, so they are not deflected by magnetic fields that they cross on their way to Earth, and point directly to their origin. Moreover, they are not getting absorbed like photons.
The IceCube Neutrino Observatory
The IceCube Neutrino Observatory is an experiment at the South Pole observing neutrinos since 2011. The detector is located underground in the Antarctic ice, reaching almost 3 km depth. It detects the light that is produced when neutrinos interact in the Antarctic ice.
When a neutrino crosses the detector, it interacts with the ice, producing secondary particles. These particles „fly“ nearly in the same direction as the neutrino. If they move very fast, the atoms around their trajectory emit light, which is detected by the experiment. If one incoming particle is detected through these processes, it is called an event.
Filtering the Astrophysical Neutrinos
An enormous number of neutrinos constantly pass through the detector, while only a few interact in the detector. At the same time, a huge number of other „background“ particles create massive numbers of events. It would be impossible to save each of these events, because it would take an incredible amount of storage and computational resources. Therefore, the relevant events, called signals, have to be filtered, and the irrelevant ones, which are called background, must be removed.
IceCube’s primary goal is to detect neutrinos coming from space, called astrophysical neutrinos, which are the signal in this case. A huge background of other particles is detected constantly, mainly muons and neutrinos produced by interactions of cosmic rays with the Earth’s atmosphere. They are known as atmospheric muons and atmospheric neutrinos. A muon is a heavier version of the electron that behaves similarly but decays quickly into other particles. To give you a sense of the dimensions, here are some numbers: There is one astrophysical neutrino in 100 million muons. To study astrophysical neutrinos, this one neutrino must be filtered out from all the muons and all the other detected background particles.
Reconstruction of the Observed Neutrinos
To gain information about the astrophysical neutrinos’ origin, their properties, like their energy and direction of origin, have to be reconstructed. This has to happen as soon as possible after detection to be able to combine information from different telescopes. However, at the South Pole, computational resources that are needed for quick reconstruction are limited.
New Deep Learning Tools
Researchers from the IceCube Collaboration at TU Dortmund University have now succeeded in advancing their analysis tools using machine learning, in particular, convolutional neural networks. These new tools are used to filter the signal events from the background events and to reconstruct the identified neutrinos’ energies and directions of origin.
Once trained, machine learning methods are computationally inexpensive, which means that less computer power is necessary to get results, and they run 100 to 1000 times faster compared to the standard reconstruction methods in IceCube. Due to this shorter runtime, more complex and advanced filtering strategies can be used, which retain more than 20 times more events than previous methods. That is an advantage, because we want to improve filtering out the background events and keeping as many signal events as possible. Therefore, it is important to distinguish as precisely as possible between signal and background events and not to „waste“ more signal events than necessary. The higher number of retaining events results from different effects, e.g., from keeping events closer to the edges of the detector.
The machine learning methods provide an angular resolution that is up to 2 times better than the previous one. Therefore, the neutrino’s direction of origin and other properties can be reconstructed much more accurately.
The same improvement in accuracy could have been otherwise only achieved by extending the observing time of the IceCube neutrino observatory by 75 years. However, this would have cost an additional $500 million compared to the current observing time. A major part of this money has been saved by using machine learning methods instead.
Results
Using these new methods, 10 years of data from the IceCube Neutrino Observatory, from May 2011 to May 2021, were analysed to search for astrophysical neutrinos originating from the Milky Way. Previous studies have been unable to measure such a signal to any significant extent. In this study, based on the achievements of the group of Prof. Dr. Dr. Wolfgang Rhode, Physics Chair at the Lamarr Institute, a neutrino flux originating from the Milky Way could finally be observed. This means, in other words, that we were finally able to see the Milky Way through the neutrino lens.