Neutrinos are one of the fundamental particles in nature: in fact, they are the most common things in the universe aside from photons of light, outnumbering electrons or protons a billion to one. They are uncharged cousins to the electron, interacting only via the Weak Force. This makes them very hard to study, as they will pass through any detector built to study them nearly all the time. On those rare occasions when one hits an atom in a detector, the resulting spray of more "normal" charged particles can be studied to infer the incoming neutrino properties. Very large detectors (often underground) maximize the number of neutrinos observed, and in addition to looking for neutrinos from natural sources such as the sun or cosmic rays, also build intense beams of man-made neutrinos.
This neutrino research group in Duluth works primarily with detectors studying neutrinos produced by Fermilab's NuMI beam, and makes extensive use of Monte Carlo simulations of neutrino interactions in their detectors in order to make sense of the data being recorded in current experiments, as well as to help design the optimal detector for future experiments. These are computationally intensive, but also trivially parallizeable: that is, any given particle interaction does not depend on the others, so many cores can be put to work simultaneously to generate many simulated interactions. Tools used are based around the GENIE neutrino interaction generator, propagated through GEANT4 based detector simulations, and analyzed by ART based analysis software.
This group developed and trained CVN-based machine learning algorithms for distinguishing beam-based neutrino signal from cosmic ray background in the NOvA experiment. The descendants of that algorithm are now running on Argonne's Theta cluster as the group re-processes more than a Petabyte of archived NOvA data.