⚛️ phenomenology

Light Dark Matter Discovery Potential and Model Selection at LDMX

This paper evaluates the Light Dark Matter eXperiment's (LDMX) projected 5 $\sigma$ discovery potential and model selection capabilities for light dark matter mediated by a dark photon, demonstrating that a two-dimensional likelihood-based analysis of recoil electron distributions can effectively distinguish between competing dark sector hypotheses and accurately infer model parameters.

Original authors: Adam Berger, Riccardo Catena, Jan Conrad, Taylor R. Gray

Published 2026-07-28

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Adam Berger, Riccardo Catena, Jan Conrad, Taylor R. Gray

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine the universe is a giant, bustling party where we can only see a tiny fraction of the guests. We know the "Standard Model" of physics describes the visible crowd—electrons, protons, and photons—but there's a massive, invisible section of the dance floor we can't see. This missing crowd is called "Dark Matter." We know it's there because it has gravity; it holds galaxies together like invisible glue. But we've never seen a single particle of it. Scientists have been throwing "search parties" for decades, using giant detectors underground to wait for a dark matter particle to bump into a normal atom, or using massive particle accelerators to smash particles together and hope a dark particle pops out. The big question is: what does this invisible guest look like? Is it heavy and slow, or light and speedy? And how do we catch it without seeing it?

This paper focuses on a specific, exciting idea: that dark matter might be "light," meaning it's much lighter than an atom, and it interacts with our world through a new, invisible force carrier called a "dark photon." Think of the dark photon as a secret messenger that can talk to both our visible world and the dark world. If we can create these messengers in a lab, they might decay into dark matter particles that fly away unseen, leaving behind a "missing energy" signature. The paper simulates a future experiment called LDMX (Light Dark Matter eXperiment), which is designed to be the ultimate detective for these light, invisible particles. The researchers built a sophisticated statistical toolkit to ask: If LDMX runs, will it find the dark matter? And if it finds a signal, can it tell us exactly what kind of dark matter it is, or if it's just a background noise?

The authors of this paper didn't just guess; they ran thousands of computer simulations to see how well the LDMX experiment would perform. They set up a "digital twin" of the experiment, feeding it data from four different scenarios where dark matter might exist, based on the most popular theories about how the universe formed. They also tested two different "noise" levels: one where the experiment is very quiet (only 1 background event expected) and one where it's a bit noisier (100 background events).

Here is what their simulations found:

The Discovery Potential
The team found that if the dark matter exists at certain specific "sweet spots" (called benchmark points), LDMX has a very strong chance of finding it. For two of the scenarios they tested (where the dark matter is a specific ratio of mass to the dark photon, known as R = 2.5), the experiment is projected to see a signal so clear that it would be a 5-sigma discovery. In science-speak, "5-sigma" is the gold standard; it means there's less than a one-in-a-million chance the signal is a fluke. However, for the other two scenarios (where the ratio is R = 2.2, which are harder to find), the experiment might fall just short of that 5-sigma mark, especially if the background noise is high. This tells us that while LDMX is a powerful tool, finding the "lighter" or more elusive types of dark matter will require keeping the background noise as low as possible.

The Detective Work: Measuring the Mystery
Finding the particle is only step one. The next challenge is figuring out its properties. The paper shows that by looking at the energy and the sideways motion (transverse momentum) of the electron that recoils when the dark photon is created, the experiment can actually measure the mass of the dark photon and how strongly it mixes with our world. In their simulations, when the signal was strong (the "easy" scenarios), the computer was able to reconstruct the true values of these properties with high precision, getting them right within about 10% uncertainty. But when the signal was weak, the measurements became fuzzy and less reliable, showing that a stronger signal is needed to really understand the particle's nature.

The Model Selection: Who is the Culprit?
Perhaps the most playful part of the study is the "model selection" test. The researchers asked: If we see a signal, can we tell if it's caused by a standard "kinetic mixing" dark photon, or if it's caused by more exotic interactions like "magnetic dipoles" or "anapole moments"? These are just fancy names for different ways the dark photon could interact with matter.

Using a statistical tool called the "Bayes factor," they compared these different theories against each other. They found that if they use the full two-dimensional data (both energy and sideways motion), the experiment can distinguish between these different theories with high confidence. It's like having a high-resolution photo that clearly shows the suspect's face, rather than a blurry silhouette. However, if they only looked at the energy (a one-dimensional analysis), the ability to tell the theories apart dropped significantly. This proves that using all the available data is crucial for solving the mystery.

The Verdict
In short, this paper simulates the future of the LDMX experiment and concludes that it is a highly promising hunt for light dark matter. It suggests that the experiment will likely be able to rule out large chunks of the "dark matter map" if it finds nothing, and if it does find something, it will have the statistical power to not only confirm the discovery but also to measure the particle's properties and distinguish between different types of dark sector theories. The study emphasizes that while the math is complex, the strategy is sound: by carefully measuring the energy and momentum of recoil electrons and using advanced statistics, we might finally get a glimpse of the invisible guests at the universe's party.

Technical Summary: Light Dark Matter Discovery Potential and Model Selection at LDMX

Problem and Motivation
Accelerator-based searches offer a competitive avenue for detecting light, sub-GeV dark matter (DM), complementing direct detection efforts. The Light Dark Matter eXperiment (LDMX) is designed to probe this regime by measuring the energy and transverse momentum of recoil electrons in high-intensity electron-nucleus collisions, looking for a missing momentum signature indicative of dark photon ( $A'$ ) production and subsequent invisible decay into a DM pair. While LDMX aims to improve constraints on the kinetic mixing parameter $\epsilon$ by orders of magnitude, a rigorous statistical framework is required to assess its discovery potential, parameter inference capabilities, and ability to distinguish between competing dark sector hypotheses. This is particularly critical given the low expected event counts in signal regions, which render standard asymptotic statistical approximations unreliable.

Methodology
The authors develop a comprehensive likelihood-based statistical framework for LDMX, applicable to both frequentist and Bayesian analyses. The core of the methodology involves a binned likelihood function constructed over the two-dimensional space of recoil electron energy ( $E_{e^-}$ ) and transverse momentum ( $|p_T|_{e^-}$ ).

Signal and Background Modeling: Signal events are simulated using MadGraph5_aMCNLO for dark bremsstrahlung processes ( $e^- N \to e^- N A', A' \to \chi \chi^*$ ). Backgrounds are modeled based on photo-nuclear interactions and bremsstrahlung photons, assuming uncorrelated energy and transverse momentum distributions for the background. Both signal and background distributions are smeared to account for detector resolution effects.
Statistical Framework:
- Frequentist: The study employs profile likelihood ratios to determine projected exclusion limits (90% C.L.), discovery significance ( $5\sigma$ ), and maximum likelihood parameter estimation. Due to low bin counts, the authors reject asymptotic approximations (Wilks' theorem) in favor of generating full toy Monte Carlo pseudo-experiments to derive test statistic distributions.
- Bayesian: The framework utilizes dynamic nested sampling (via dynesty) to compute posterior distributions and Bayesian evidence. Nuisance parameters (signal and background normalization) are marginalized over using log-normal priors, with the background uncertainty width treated as a hyperparameter.
Model Comparison: To assess model selection power, the authors compare the standard kinetic mixing (KM) model against Dark Electromagnetic Moment (DEM) models, including magnetic/electric dipoles (M, E), charge radius (C), and anapole moment (A). Model discrimination is quantified using the Bayes factor ( $K$ ), evaluated with and without a prior constraint on the relic density.

Key Contributions

Robust Statistical Framework: The paper establishes a likelihood framework specifically designed for LDMX that incorporates systematic uncertainties and avoids unreliable asymptotic approximations by using pseudo-experiments. This framework is intended for direct application to real LDMX data.
Two-Dimensional Analysis: The study demonstrates the critical importance of utilizing the full two-dimensional recoil distribution ( $E_{e^-}, |p_T|_{e^-}$ ) rather than energy-only analyses. The shape information in the transverse momentum dimension significantly enhances discriminating power.
Benchmark Evaluation: The analysis evaluates LDMX's sensitivity at four benchmark points along the thermal relic target for complex scalar DM, specifically focusing on mass ratios $R \equiv m_{A'}/m_{DM} = 2.5$ and $2.2$, which remain viable after recent direct detection constraints (e.g., DAMIC-M, PANDAX-4T) that excluded $R \gtrsim 3$ .

Results

Exclusion and Discovery:
- In the absence of a signal, LDMX is projected to exclude the thermal relic target for $R \gtrsim 2.2$ across the mass range $m_{A'} \in [10^{-3}, 0.2]$ GeV.
- For discovery, benchmarks on the $R=2.5$ target (Benchmarks 1 and 3) show strong projected significance ( $\gg 5\sigma$ ) across both low ( $N_{BG}=1$ ) and high ( $N_{BG}=100$ ) background scenarios.
- Benchmarks on the more challenging $R=2.2$ target (Benchmarks 2 and 4) yield median significances just below $5\sigma$ in the low-background scenario and significantly lower in the high-background scenario, indicating that $5\sigma$ discovery is not guaranteed for these points.
Parameter Estimation:
- For $R=2.5$ benchmarks, LDMX can recover the kinetic mixing $\epsilon$ and mediator mass $m_{A'}$ with high accuracy and precision (relative uncertainties $\sim 10\%$ ) in both background scenarios.
- For $R=2.2$ benchmarks, parameter inference is less reliable due to small signal yields and Poisson fluctuations, resulting in larger variances and, in some cases, uninformative constraints.
Model Discrimination:
- The two-dimensional analysis provides substantially greater discriminating power (approximately an order of magnitude higher in $\ln K$ ) compared to energy-only analyses.
- While models with identical spectral shapes (M/E and C/A pairs) are indistinguishable, the KM model is distinguishable from DEM models.
- However, when models share the same spectral shape but differ in normalization (e.g., KM vs. C/A), they are statistically indistinguishable via the Bayes factor alone. Applying a relic density prior on the coupling strength renders these previously indistinguishable hypotheses statistically separable.
- The study highlights that model misspecification can lead to large biases in parameter inference even when the Bayes factor is near zero, emphasizing the need for correct model identification.

Significance and Claims
The paper claims that the developed statistical framework is essential for translating raw event counts into robust physics conclusions for LDMX. It asserts that LDMX Phase II will not only probe the viable thermal relic parameter space for light complex scalar DM with strong exclusion and discovery sensitivity but will also be capable of characterizing the underlying dark sector interaction if an excess is observed. The authors emphasize that their tools are designed to be directly applicable to real LDMX data and can be extended to other missing momentum experiments (e.g., Lohengrin, DarkSHINE) and combined likelihood analyses across the broader light DM experimental program. The work underscores that accurate modeling of signal and background distributions, along with a thorough understanding of normalization uncertainties, are critical for future LDMX analyses.

More like this