⚛️ high-energy experiments

Quantum artificial intelligence for pattern recognition at high-energy colliders: Tales of Three "Quantum's"

This article reviews the current status, advantages, and limitations of three distinct quantum computing technologies—quantum gates, quantum annealing, and quantum-inspired methods—in addressing the computationally intensive task of pattern recognition at high-energy colliders.

Original authors: Hideki Okawa

Published 2026-10-14

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Hideki Okawa

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

High-energy physics is the science of smashing particles together at speeds close to that of light to see what they are made of. To do this, scientists build massive machines called colliders, which are essentially giant microscopes that can see the smallest building blocks of the universe. These machines generate an overwhelming flood of data. Every time particles collide, they leave behind a trail of signals in detectors, much like a car leaving tire tracks on a wet road. The job of the physicist is to look at these millions of scattered tracks and figure out which ones belong to the same particle, and which ones are just random noise. This process, known as pattern recognition, is the most computationally expensive part of the entire experiment. As these machines become more powerful, the amount of data they produce is growing so fast that traditional computers are struggling to keep up, threatening to slow down the discovery of new physics.

In a recent review, Hideki Okawa from the Institute of High Energy Physics in Beijing examines a new way to tackle this problem: using quantum computing. Unlike the computers we use every day, which process information as a series of simple on-or-off switches, quantum computers use the strange rules of quantum mechanics to handle information in a much more complex way. This allows them to explore many possibilities at once. Okawa's article does not present a single new experiment but rather a comprehensive map of the current landscape, looking at three distinct types of quantum technology that are being tested to solve these pattern recognition puzzles. The first type uses quantum logic gates, similar to the circuits in a normal computer but built with quantum bits. The second type, called quantum annealing, is a specialized machine designed specifically to find the lowest energy state of a system, which is perfect for solving optimization problems like sorting tracks. The third type, known as quantum-inspired, uses classical computers to run algorithms that mimic the behavior of quantum machines.

The paper focuses on two specific tasks where these technologies are being applied: tracking charged particles and grouping them into jets. Tracking is the process of connecting individual hits in a detector to form a continuous line, revealing the path of a particle. Grouping, or jet clustering, happens when a high-energy particle breaks apart into a spray of other particles; the computer must figure out which spray belongs to which original particle. Okawa explains that scientists have tried to turn these tasks into mathematical puzzles called Ising or QUBO problems, which ask a computer to find the best possible arrangement of variables to minimize an error. In the past, researchers have used quantum annealers to solve these puzzles for small sets of data. For example, one study used a quantum machine with about 1,100 bits to reconstruct tracks from a simulated collision. The results were promising, matching the performance of the best traditional methods for up to 500 particles. However, when the number of particles grew larger, the quantum hardware hit a wall because it simply did not have enough connections between its bits to handle the complexity.

To get around this hardware limitation, researchers have turned to quantum-inspired algorithms. These run on standard supercomputers but use the logic of quantum mechanics to find solutions faster. One such method, called simulated bifurcation, has shown remarkable speed. In tests using a single graphics processing unit, this algorithm reconstructed tracks for events with nearly 10,000 particles in just 0.14 seconds. This is a massive improvement over the quantum annealer, which took 23 minutes to do the same job on a much smaller dataset. The paper notes that while these classical simulations are fast and accurate, they are not true quantum computers. They are a bridge, proving that the mathematical approach works even before the hardware catches up. For the task of grouping particles into jets, similar progress has been made. Researchers have successfully used quantum-inspired methods to reconstruct complex events with multiple jets, improving the precision with which they can measure the mass of heavy particles like the top quark.

The review also looks at more experimental approaches using true quantum circuits, which are the most versatile but also the most fragile. Scientists have tried to build hybrid systems where a quantum computer helps a classical one find the best track paths. In these simulations, the quantum methods performed as well as the classical ones, but they required a level of precision and error correction that current machines do not yet possess. The paper is clear that while the potential is there, the hardware is not ready for the real thing. The number of particles in a real collider event is far too high for today's quantum devices to handle directly. The author suggests that the most immediate path forward is not to replace classical computers, but to use these new quantum-inspired algorithms to handle the massive datasets that future colliders will produce.

Ultimately, the paper concludes that we are in a period of transition. The technology is not yet mature enough to run the full reconstruction of a real collision event on a quantum machine. However, the mathematical foundations are solid, and the speed gains from quantum-inspired methods are already being realized. As the next generation of colliders prepares to come online in the 2030s, these new tools offer a way to keep pace with the exploding volume of data. The story is not one of a sudden revolution, but of a steady, practical evolution where new ideas are being tested, refined, and integrated to ensure that the search for the secrets of the universe does not get bogged down by the sheer weight of its own data.

Technical Summary: Quantum Artificial Intelligence for Pattern Recognition at High-Energy Colliders

Problem Statement
High-energy physics (HEP) experiments, particularly at future colliders like the High Luminosity Large Hadron Collider (HL-LHC) and the Circular Electron Positron Collider (CEPC), face an exponential growth in data volume and computational costs. The annual computing cost is projected to increase by a factor of 10 to 20, driven largely by the need to handle exabyte-level datasets. A primary bottleneck in the experimental workflow is pattern recognition, specifically track reconstruction (associating detector hits to charged particle trajectories) and jet clustering (grouping particles to reconstruct quark and gluon kinematics). These tasks are computationally intensive and often map to combinatorial optimization problems, which are NP-complete. The paper investigates whether emerging quantum computing technologies can offer efficiency gains or fundamentally different approaches to solving these optimization challenges compared to existing classical methods.

Methodology
The paper reviews and categorizes three distinct quantum computing technologies applied to HEP pattern recognition:

Quantum Circuits (Gate-based): Universal quantum computers using logic gates (e.g., superconducting qubits). The review focuses on variational algorithms like the Variational Quantum Eigensolver (VQE) and Quantum Approximate Optimization Algorithm (QAOA), as well as hybrid quantum-classical Graph Neural Networks (QGNNs).
Quantum Annealing: Specialized hardware (e.g., D-Wave) that uses the adiabatic theorem to find the ground state of a Hamiltonian, specifically tailored for Ising and Quadratic Unconstrained Binary Optimization (QUBO) problems.
Quantum-Inspired Algorithms: Classical algorithms (e.g., Simulated Annealing, Simulated Bifurcation) that mimic quantum mechanical principles like tunneling or bifurcation to solve optimization problems on classical hardware (GPUs/FPGAs).

The methodology involves formulating track finding and jet clustering as Ising/QUBO problems. For track reconstruction, the problem is mapped to finding the optimal set of detector hit combinations (doublets or triplets) that minimize a specific energy function. For jet clustering, the goal is to partition particles into jets to maximize thrust or minimize a distance metric, also formulated as a QUBO.

Key Contributions and Results

Track Reconstruction:
- Global Optimization: Early studies formulated track finding as an Ising problem using the Denby-Peterson method. Using D-Wave 2X hardware, researchers achieved track reconstruction efficiency and purity comparable to Simulated Annealing (SA) for events with up to 500 particles. However, hardware limitations (qubit count and connectivity) prevented scaling to higher multiplicities.
- Sub-QUBO Approaches: To handle the high multiplicity of HL-LHC events (requiring ~0.1 million spins), "sub-QUBO" methods were employed, splitting the problem into smaller matrices. While this allowed for larger problem sizes, it introduced a significant computational speed penalty (orders of magnitude slower than classical methods) due to the iterative splitting and recombination process.
- Quantum-Inspired Success: A significant finding is that Simulated Bifurcation (SB) algorithms (specifically ballistic and discrete variants) running on a single GPU can handle large-scale simulated datasets. For the largest TrackML dataset event with 9,435 particle multiplicity, SB achieved efficiency and purity comparable to D-Wave Neal SA but with a four-order-of-magnitude speedup (reducing runtime from 23 minutes to 0.14 seconds). SB can effectively run multiple processing tasks with cutting-edge computing resources such as GPU and FPGA, demonstrating its potential and readiness for practical applications.
- Iterative and Neural Approaches: Hybrid Quantum-Classical GNNs (QGNNs) were tested, showing performance comparable to classical GNNs but currently limited by the need for quantum circuit simulators (up to 16 qubits). Iterative approaches using Grover search and Quantum Amplitude Amplification (QAA) demonstrated theoretical complexity reductions (e.g., $O(\sqrt{N})$ vs $O(N)$ ) for seeding and track building, though practical implementation on current hardware faces topological bottlenecks.
Jet Reconstruction:
- Iterative Clustering: Quantum algorithms were applied to thrust axis finding and K-means clustering. Quantum maximum finding algorithms (generalized Grover search) offered theoretical scaling improvements. Hybrid quantum-classical K-means showed reduced computational complexity ( $O(N \log N)$ ) compared to classical counterparts, though noise and lack of Quantum Random-Access Memory (QRAM) hindered execution on real hardware.
- Global Clustering: Jet clustering was formulated as a QUBO problem. Quantum annealing with tuned parameters achieved results within 1–3% of target values for dijet events. An angle-based QUBO formulation outperformed thrust-based approaches in consistency.
- Multijet Reconstruction: Extending to multijet events (e.g., $t\bar{t}$ events) is challenging for quantum annealing due to the need for one-hot encoding and long annealing times. However, the ballistic Simulated Bifurcation (bSB) algorithm successfully performed global multijet reconstruction. It found minimum energy predictions ~10 times lower than SA for all-hadronic $t\bar{t}$ events and improved jet energy resolution by 6–7% in invariant mass reconstruction.
Vertex Reconstruction:
- A quantum annealing approach was developed to cluster tracks into primary vertices. While performance was comparable to SA for small numbers of vertices, scaling to LHC/HL-LHC conditions requires hardware with significantly more qubits and connectivity.

Significance and Claims
The paper posits that while practical quantum advantage (outperforming classical methods on real-world, large-scale problems using actual quantum hardware) has not yet been fully established due to the Noisy Intermediate-Scale Quantum (NISQ) era limitations, the field is making rapid progress.

Quantum-Inspired Algorithms: The paper highlights that quantum-inspired algorithms, particularly Simulated Bifurcation, are currently the most promising for near-term application. They demonstrate the ability to handle large datasets with the speed and accuracy required for future colliders, effectively bridging the gap until error-tolerant quantum hardware matures.
Hardware Limitations: The review emphasizes that current quantum circuit and annealing hardware is constrained by qubit count, connectivity, and noise. While they can solve small-scale or simplified versions of track and jet problems, they struggle with the sub-million to million-level problem sizes inherent in full HEP workflows.
Future Outlook: The paper concludes that the "rich program" of quantum computing in HEP will likely progress through constructive competition among the three technologies. It suggests that while error-tolerant quantum hardware may not be ready for the full experimental workflow in the immediate future, the integration of quantum algorithms with High-Performance Computing (HPC) and the development of quantum-inspired techniques are critical for managing the computational demands of the HL-LHC and future colliders.

Technical Summary: Quantum Artificial Intelligence for Pattern Recognition at High-Energy Colliders

More like this