Demonstration and performance of an online data selection algorithm for liquid argon time projection chambers using MicroBooNE

This paper demonstrates the first successful application of an online, charge-based data selection algorithm in a liquid argon time projection chamber using MicroBooNE data, providing a proof-of-principle for real-time signal preservation in future large-scale experiments like DUNE.

The Gist

The Cosmic Needle in a Haystack: How MicroBooNE is Learning to "Filter" the Universe

Imagine you are standing on a beach during a massive, non-stop storm. The waves are crashing, the wind is howling, and millions of grains of sand are flying past your face every single second.

Now, imagine someone tells you: "Somewhere in that chaos, a single, tiny, glowing blue pearl is being tossed around by the waves. You need to catch it, but you can’t stop the storm, and you don't have enough buckets to catch all the sand just to find the pearl."

That is essentially the problem scientists face at the MicroBooNE experiment.

The Problem: Too Much "Noise"

MicroBooNE is a massive detector (a "Liquid Argon Time Projection Chamber") designed to catch tiny, rare particles from space. But because it sits on the surface of the Earth, it is constantly being bombarded by "cosmic rays"—basically a relentless rain of high-energy particles from space.

For the detector, these cosmic rays are like the billions of grains of sand in our storm. They create a mountain of data—about 33 gigabytes every single second. If scientists tried to save all that data to a hard drive, they would run out of space almost instantly. They need to find the "pearls" (rare, interesting physics events) without drowning in the "sand" (the common cosmic rays).

The Solution: The "Smart Sieve"

In this paper, the MicroBooNE team describes a new way to build a "Smart Sieve."

Instead of trying to catch everything and sorting it later, they have developed an algorithm (a set of mathematical instructions) that works "online." This means the sorting happens while the storm is still happening.

How the "Smart Sieve" Works (The Anatomy of a Search)

The scientists decided to train their sieve to look for a very specific pattern: a Michel electron.

Think of a Michel electron like a specific type of "glitter" that only appears when a certain type of "heavy pebble" (a muon) stops moving and suddenly breaks apart. To find this glitter, the algorithm follows a three-step dance:

1. The Scout (Trigger Primitives): Instead of looking at the whole massive wave of data, the algorithm looks at tiny "snapshots" of the energy. It asks: "Is there a sudden burst of energy here?" If yes, it marks that spot.
2. The Detective (Trigger Candidates): Now it looks closer at those marked spots. It looks for a specific shape: a straight line (the pebble) that suddenly takes a sharp, jagged turn (the glitter). This "kink" in the path is the smoking gun.
3. The Librarian (High-Level Trigger): Once the Detective is sure it found a "kink," the Librarian rushes in, grabs the relevant data from the surrounding area, and saves it to the permanent archives. Everything else is tossed back into the ocean.

Why This Matters: Preparing for the "Mega-Storm"

This isn't just about MicroBooNE. This paper is a "proof of concept."

In the future, scientists are building even bigger detectors, like DUNE (the Deep Underground Neutrino Experiment). If MicroBooNE is a beach storm, DUNE will be a massive hurricane. The data rates will be thousands of times higher.

By proving that they can use "intelligent" software to recognize shapes and patterns in real-time, the MicroBooNE team has shown that we don't need bigger buckets; we just need smarter sieves. They are teaching computers how to "see" the interesting parts of the universe in the middle of the chaos.

Technical Summary

Technical Summary: Demonstration and Performance of an Online Data Selection Algorithm for LArTPCs using MicroBooNE

1. Problem Statement

Liquid Argon Time Projection Chambers (LArTPCs), such as those used in the MicroBooNE and upcoming DUNE experiments, generate massive volumes of raw data. For instance, MicroBooNE’s charge readout system produced approximately 33 GB/s, while future detectors like DUNE are expected to reach rates of several TB/s.

A critical challenge arises when searching for rare, off-beam physics processes (e.g., supernova bursts, proton decay, or baryon number violation). These events occur randomly in time and lack external timing triggers. To capture them without missing signals, detectors must run with nearly 100% "livetime." However, storing all continuous data is impossible due to bandwidth and storage constraints. Therefore, there is an urgent need for intelligent, real-time (online) data selection algorithms capable of identifying specific physics topologies from the raw ionization charge data to reduce data rates while preserving rare signals.

2. Methodology

The authors developed and tested an "emulated online" data selection framework. Since the MicroBooNE beam operations had concluded, they used pre-recorded "Supernova (SN) stream" data to simulate a real-time environment. The framework targets a specific topology: Michel electrons produced by the decay of stopping cosmic-ray muons.

The methodology follows a three-stage sequential workflow:

• Process 1: Trigger Primitive (TP) Generation:
  • Raw waveforms are pre-processed (bit-flip correction, baseline subtraction, and noise removal).
  • The continuous stream is "stitched" into 2.3 ms "drift regions" to ensure causally connected signals are grouped together.
  • Summaries of each channel (integrated charge, maximum amplitude, and waveform width) are extracted as TPs.
• Process 2: Trigger Candidate (TC) Generation:
  • Step A (Bragg Peak Identification): The algorithm searches for localized energy increases (Bragg peaks) characteristic of stopping muons using specific integrated charge and amplitude thresholds.
  • Step B (Kink Angle Analysis): The algorithm searches for track-like patterns in both forward and backward directions from the Bragg peak. It calculates the "kink angle" ($\theta_{kink}$) between these two segments. A decay from a muon to an electron typically creates a visible change in track linearity (a "kink").
• Process 3: High-Level Trigger (HLT):
  • Once a candidate is identified, this process retrieves the relevant full-waveform data from disk storage to ensure the event is preserved for offline analysis.

3. Key Contributions

• First Demonstration: This represents the first time an online data selection algorithm has been demonstrated in a LArTPC using real data and ionization charge information exclusively (without relying on external light triggers).
• Topological Selection: Unlike simple threshold-based triggers, this algorithm uses complex spatial and temporal information (Bragg peaks and track kink angles).
• Scalable Architecture: The multi-stage, parallelized software design (using tools like zeroMQ) is specifically modeled after the proposed data selection strategies for the DUNE experiment.

4. Results

• Algorithm Performance: By optimizing the kink angle requirement to $>30^\circ$, the researchers achieved a balance between efficiency and background rejection. In simulations, this requirement provided a Michel electron selection efficiency of approximately 20.4% while significantly reducing false positives from crossing muons.
• Throughput and Latency: The "emulated online" system successfully kept up with the detector's data rate. The total processing time per drift region peaked at approximately $1.2\text{ ms}$, which is well within the maximum allowed latency of $2.3\text{ ms}$ (the duration of one drift window).
• Resource Utilization: The framework demonstrated stable operation on standard CPU hardware, with a mean CPU utilization of approximately 40% during continuous execution.

5. Significance

This work serves as a vital proof-of-principle for the next generation of neutrino physics. It proves that TPC-based topological triggers are computationally feasible for real-time deployment. The success of this framework provides a roadmap for:

1. DUNE and SBND: Implementing similar intelligent triggers to manage the unprecedented data rates of future large-scale LArTPC experiments.
2. Rare Event Searches: Enhancing the sensitivity to supernova neutrinos and other BSM (Beyond the Standard Model) physics by enabling high-precision, low-energy data selection.
3. AI/ML Integration: Providing a foundation for future "intelligent" triggers that utilize Deep Learning (CNNs or Graph Neural Networks) to perform even more sophisticated real-time pattern recognition.