Statistically Learning Dispersed New Physics at the LHC – SLDNP

Experiments at the LHC make a tremendous effort to search for new physics in a plethora of final states, exploiting many different analysis techniques. The analysis designs are often motivated by theoretical ideas of how to extend the so-called Standard Model of particle physics. So far, no clear sign of new physics has emerged in any of these searches, and thus limits on the masses of hypothetical new particles of “beyond the Standard Model” (BSM) scenarios are pushed higher and higher. There is a fundamental problem, however, with this channel-by-channel approach: while it is very powerful for discovering or excluding simple, clear-cut signals that show up in only one channel, dispersed signals (i.e. effects of new particles which are spread out over several search regions or final states) may easily be missed since in each individual analysis only a small part of the available data is used.

The problem is a burning one for two reasons. First of all, it remains clear that the Standard Model cannot be the last word. The question of the stability of the mass of the Higgs boson remains an unresolved puzzle, as do the nature of dark matter and the origin of the baryon asymmetry in the universe, to name but some of the major problems that still elude us to date and that may have their answers in (TeV-scale) BSM physics. Second, the LHC will be the world’s energy-frontier machine for the foreseeable future, with Run 3 taking place 2022-2024 and the High-Luminosity LHC phase scheduled to start in 2028. It goes without saying that the full exploitation of the LHC data is of primary importance to the field.

With the SLDNP project, we attempt a new and much more global approach to the problem. Concretely, we aim at developing a statistical learning algorithm that identifies potential dispersed signals in the slew of published LHC analyses (collected in a large database) while remaining compatible with the entirety of LHC constraints. The dispersed signals are contextualized with "proto-models" of new physics for further testing and mapping onto full (UV-complete or effective-field) theories. In case of a discovery, this model-independent contextualising may address the inverse problem of particle physics and help to eventually unravel the concrete underlying BSM theory.

SLDNP is thus a data-driven, global bottom-up approach to the quest of new physics. Its feasibility was recently demonstrated in a prototype study. The objective of this ANR project is to realize the full potential of SLDNP and develop it into a robust, reliable framework for elucidating effects of new physics in the LHC data.

The project is a tight collaboration of theorists and experimentalists from the "Laboratoire de Physique Subatomique et de Cosmologie'' (LPSC) in Grenoble and the "Institute of High Energy Physics'' (HEPHY) of the Austrian Academy of Sciences in Vienna. The researchers involved are experts in BSM collider physics and have long-standing experience with the reinterpretation of LHC results, data science, and the development of computational tools.