Machine Learning the Total Width of the Higgs Boson at CMS

Particle physics seeks to understand the universe on the most fundamental level, and answer many of the deepest questions about nature, concerning, e.g., the origin of mass, the identity of dark matter, and the greater abundance of matter than antimatter in our universe. Our current theory, the Standard Model of particle physics (SM), is enormously successful, but is known to be an incomplete description of nature. To make progress, we must observe a crack in the SM. The Higgs boson was discovered in 2012, and it is of central importance in the SM, it is unique among the fundamental particles, and it may be the key to finding such a crack. The aim of this program is to probe the lifetime of the Higgs boson, using a so-far unpublished data analysis strategy, at the Compact Muon Solenoid (CMS) experiment at the CERN Large Hadron Collider (LHC). Various machine learning / AI techniques will be employed, and the program will seek to improve the performance of the CMS experiment in various ways that complement the core goals of the program and that will benefit CMS more broadly. These efforts will further the NSF’s goal to promote the progress of science. Moreover, contributions to STEM pipeline building, to supporting the careers of early-career researchers and students, and to outreach, education and training will be made. This will be achieved by integrating early-career researchers including students in the program, and by various dedicated activities, including a machine learning workshop, a new machine learning course at the University of Kansas, a new recurring academic career panel event, and local outreach. Approximately 10% of Higgs bosons produced at the LHC are off-shell, which means that their masses have unusual values due to quantum fluctuations. The central objectives of this proposal are to constrain the off-shell Higgs boson cross section and its total width (equivalently, its lifetime), using the CMS experiment at the CERN LHC. The off-shell Higgs boson cross section provides sensitivity to potential new physics at high energies, while the total width provides sensitivity to potential new physics in Higgs boson decays, since any new or modified decays would change the Higgs boson lifetime. To achieve these objectives, off-shell Higgs boson decays to pairs of W bosons in the semi-leptonic final state will be used. This as-yet unpublished approach will address the statistical uncertainty that was dominant in previous off-shell Higgs boson analyses. Machine learning will be used to probe the signal, and a statistical combination, which will include this analysis, will be performed. The final state of this analysis consists of one charged lepton, one neutrino, and a jet of hadrons from the decay of a boosted W boson. While the charged lepton is typically precisely reconstructed, the jet and MET (MET is missing momentum transverse to the LHC beam line, and is how neutrinos are typically reconstructed at hadron collider experiments) are typically associated with large uncertainties. To address these issues for this analysis and beyond, the performance of jet and MET reconstruction will be improved using an AI-based combined jets-and-MET reconstruction algorithm, which builds on existing efforts, and jet transverse momentum and MET calibrations will be performed. The LHC will soon be upgraded to the High-Luminosity LHC (HL- LHC), which will allow it to collect data 5–7 times faster. CMS will be upgraded accordingly, including with the installation of the MIP Timing Detector, which will provide timing information for charged particles. The ability to utilise this timing information in the jets-and-MET reconstruction algorithm described above will be introduced as part of this program, which will improve the rejection of particles from simultaneous proton–proton collisions at the HL-LHC. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.