Exploring visible signatures in the LDMX detector

Abstract

Axion-like particles are prime candidates for beyond the Standard Model physics, and their possible decay into photon pairs provides an experimentally accessible signature in calorimeters. This project investigates such signatures within the Light Dark Matter eXperiment detector using simulated events.

From detector-level information, a collection of physically motivated observables is constructed. This includes electromagnetic and hadronic energy deposition, energy asymmetry, and the spatial separation between leading energy deposits. The discrimination between signal events and background processes is based on these features. The analysis revealed clear differences between signal and background distributions, as well as on the axion-like particle mass. Larger masses led to greater spatial separation between leading energy deposits. A Boosted Decision Tree classifier was used to exploit these differences, achieving robust performance with an area under the curve of 0.95-0.97.

The analysis shows two central observations. Axion-like particle decays were designed to occur within the hadronic calorimeter, reflected in the observed energy distributions and the physical interpretation of detector signals. A relatively small number of physically motivated observables, combined with multivariate classification methods, is sufficient to discriminate between signal and background. This provides a basis for future analyses and potential applications to real experimental data.

Popular Abstract

Axion-like particle decay: a missing piece in the dark matter puzzle?

The universe: dark, cold, silent and mostly ”empty” space. Or is there more? The expansion of the universe and the stability of galaxies would not be possible according to humanity’s visible measurements and detectors. Yet galaxies like Andromeda and the Milky Way hold together, while the universe continues to expand. This suggests that something unseen is influencing these phenomena, providing additional energy or mass. This unknown component is referred to as dark matter. This exotic and undiscovered form of physics, does not emit, absorb or reflect electromagnetic radiation. Making dark matter invisible to conventional detectors! How can mankind then succeed in finding positive evidence for dark matter when it can’t be seen?

The Light Dark Matter Experiment (LDMX) is an experiment designed to search for light dark matter from an electron-nucleus collision. The LDMX consists of several subsystems that measure how particles interact and deposit energy. The two main subsystems are the electromagnetic calorimeter (ECal) and the hadronic calorimeter (HCal). When an axion- like particle (ALP) is produced in a collision, it can decay into two photons, the particles of light! These photons leave distinct energy deposits in the ECal forming what are known as clusters. In specific scenarios, the decay can occur deep in the detector. Making the HCal an important region for identifying signatures of such events. If how the energy is distributed between these subsystems is studied, it becomes possible to distinguish signal events from more frequent background activity!

However, the detector environment is the opposite to quiet! Because each collision produces a large number of particles, which creates a complex pattern of energy deposits. It then becomes difficult to determine which events originate from an ALP decay and which belong to the background. The goal of this project is to improve this separation, a LDMX variation! To accomplish this, simulated data is used to study measurable quantities, known as physical observables, e.g. how energy is shared between detector components. These observables capture key features of the event structure. Though, since no single observable can clearly separate the signal from the background, the project uses machine learning and especially a Boosted Decision Tree (BDT) to combine the observables. The BDT functions as it learns patterns in the data and classifies events based on how signal-like the events are. A piece of the larger puzzle can then be obtained with respect to the BDT training. As the BDT trains on the simulated signal and background data, it can make it possible to identify subtle differences that would otherwise be difficult to detect.

In other words, the project’s goal is to improve the identification of ALP decays by combining multiple observables using machine learning. By doing so, the newly trained BDT model can then be an important piece of the puzzle for the LDMX’s search for dark matter and discoveries of new truths about the universe!

• author: Holmberg, Anton
• supervisor: Ruth Pöttgen; Lene Bryngemark; Clay Barton
• organization: Department of Physics; Particle and nuclear physics
• course: FYSK04 20261
• year: 2026
• type: M2 - Bachelor Degree
• subject: Physics and Astronomy
• keywords: LDMX, ALPs, dark matter
• language: English
• id: 9250741
• date added to LUP: 2026-09-14 08:36:31
• date last changed: 2026-09-14 08:36:31