Science Bulletin 70 (2025) 460–463

Contents lists available at ScienceDirect

Science Bulletin
journal homepage: www.elsevier.com/locate/scib

Short Communication

A novel quantum realization of jet clustering in high-energy physics

experiments
Yongfeng Zhu a, Weifeng Zhuang b, Chen Qian b, Yunheng Ma b, Dong E. Liu b,c,⇑, Manqi Ruan d,e,⇑,
Chen Zhou a,⇑
a
State Key Laboratory of Nuclear Physics and Technology, School of Physics, Peking University, Beijing 100871, China
b
Beijing Academy of Quantum Information Sciences, Beijing 100193, China
c
Department of Physics, Tsinghua University, Beijing 100084, China
d
Institute of High Energy Physics, Chinese Academy of Sciences, Beijing 100049, China
e
University of Chinese Academy of Sciences, Beijing 100049, China

a r t i c l e i n f o

Article history:
Received 24 July 2024
Received in revised form 13 September
2024
Accepted 2 December 2024 © 2024 Science China Press. Published by Elsevier B.V. and Science China Press. All rights are reserved,
Available online 18 December 2024 including those for text and data mining, AI training, and similar technologies.

Application of quantum technologies to fundamental sciences which is located within the detector, as illustrated in Fig. 1b. The
has the potential to advance both fields simultaneously. In the near golden process for studying Higgs properties is e e ZH. An
future, the number of reliable quantum operations in a real-world e e ZH Z mm H ss event, where two quarks form two jets,
quantum computer will be constrained by noise and decoherence is displayed in Fig. 1c. A jet clustering algorithm, which groups
[1]. To overcome these challenges, hybrid quantum–classical algo- final-state particles into sets corresponding to individual gluons or
rithms [2] have been proposed. A notable example is the quantum quarks, is applied to identify these jets. Following jet clustering, var-
approximate optimization algorithm (QAOA) [3], an algorithm ious physics analyses can be performed, such as the measurement of
commonly used to address classical combinatorial optimization the branching ratio of H ss, shown in Fig. 1g [6]. Given the contin-
problems. In high-energy particle collisions, quarks and gluons uous progress in quantum computing technology, there is increasing
are produced and immediately form collimated particle sprays, interest in identifying practical applications for near-term quantum
referred to as jets, due to color confinement [4]. Accurate jet clus- machines. For instance, the calculation of thrust, a variable was for-
tering is crucial to reconstruct the quark or gluon information and mulated both as a quantum annealing problem and a Grover search
enhances the study of the properties of the Higgs boson. For the problem [7]. Other studies [8,9] have developed quantum
first time, we apply QAOA to the problem of jet clustering, as annealing-based jet clustering algorithms.
shown in Fig. 1e, by mapping collision events into graphs, as The QAOA draws inspiration from the quantum adiabatic algo-
shown in Fig. 1d. We obtain experimental results on quantum rithm (QAA). QAA uses two Hamiltonians: an initial Hamiltonian
computer simulators and quantum computer hardware and find H M with an easily prepared ground state, and a final Hamiltonian
that their performance is comparable to or even better than classi-
H C whose ground state encodes the solution to the optimization
cal algorithms for a small-sized problem.
problem. The adiabatic evolution path is defined by the transitional
A typical physics analysis procedure involving jets in high-energy t
physics is as follows. Data are collected from an electron–positron Hamiltonian H t s t HC 1 s t H M , where s t T
t 0 T,
Higgs factory, such as the circular electron–positron collider (CEPC) and T is the total evolution time. The evolution operator U t is
[5], whose layout is shown in Fig. 1a. Electrons and positrons are derived from H t . QAA can be emulated on a gate-based quantum
accelerated in the booster and interact at the interaction point (IP), computer by defining U t into sufficiently small steps,

⇑ Corresponding authors.
E-mail addresses: [email protected] (D.E. Liu), [email protected] (M. Ruan), [email protected] (C. Zhou).

https://doi.org/10.1016/j.scib.2024.12.020
2095-9273/© 2024 Science China Press. Published by Elsevier B.V. and Science China Press. All rights are reserved, including those for text and data mining, AI training, and
similar technologies.
Y. Zhu et al. Science Bulletin 70 (2025) 460–463

Fig. 1. A typical physics analysis procedure involving jets in high-energy physics is outlined. (a) Layout of the circular electron–positron collider [5]. Electrons and positrons
are accelerated in the booster and collide at the IP within the detector, illustrated in (b) [5]. Quarks and gluons resulting from these collisions transform into collimated
streams of particles known as jets. (c) An e e ZH Z mm H ss event, where the two quarks form two jets. Jets are clustered using an algorithm that groups final-state
particles into sets corresponding to individual gluons or quarks. During jet clustering, each event is represented as a graph, as shown in (d). The QAOA can be applied to
perform jet clustering, as demonstrated in (e). The compiled quantum circuit is shown in (f). The information from reconstructed jets can be utilized to achieve related
objectives, such as the H ss measurement shown in (g) [6].

P P problem with K equal to 2). In other words, the algorithm aims

iH jDt Dt i 1 s jDt H M Dt is jDt H C Dt
U t e e e to partition the graph nodes into two complementary subsets that
j 1 j 1
1 maximize the weighted sum of edges connecting nodes in different
P P
v
e ibj H M
e icj H C
U M bj U C cj subsets, given by C x i j 1 wij xi 1 xj . Maximizing C x is
j 1 j 1 NP-hard, but QAOA can approximate C x close to its maximum
value C max max C x [3]. To solve the Max-cut problem using
where Dt T P and P is the number of steps. QAOA was designed x

from this trotterized version with repeated cost U C c and mixer QAOA, we begin by defining the problem Hamiltonian as

layers U M b [3]. HC 1
2 ij E W ij I rzi rzj and initial Hamiltonian as
An undirected graph G V E , where V is the set of nodes and HM n
j 1 r where r represents the Pauli-z operator acting on
x
j,
z
i
E is the set of edges, can encode a collision event in high-energy
the i-th qubit and W ij represents the angle between particles
physics. The nodes represent particles and the edges represent
pairs of particles that are nearby. To define the edges, a weight i and j. By substituting H C and H M into the evolution operator
wij is assigned to each pair of particles that is equal to the angle shown in Eq. (1) and decomposing it into sequences of quantum
between them. Then, for each node, edges are sorted by weight gate operations, we construct the corresponding quantum circuit.
in decreasing order and only the top k edges are retained (k- This circuit is then used to evolve the system’s initial state toward
regular graph). Fig. 1e illustrates an e e ZH Z mm H ss the ground state of H C , with a classical optimizer iteratively adjust-
event with 30 particles and k = 3. After mapping collision events ing the parameters c and b. For practical jet clustering, the focus is
into graphs, QAOA can perform jet clustering akin to solving the on approximating the ground state. Among the sampling states,
Max-K-cut problem [10]. This study clusters two jets, analogous the one closest to the ground state is selected. The procedure is
to the Max-cut problem (the specific case of the Max-K-cut illustrated in Fig. 1e and more details are in Refs. [3,11]. The key
461
Y. Zhu et al. Science Bulletin 70 (2025) 460–463

parameters are the number of trotterized steps (QAOA depth), the e e kt [13]. The k-Means algorithm partitions a dataset into k
evolution times (optimized by a classical optimization algorithm clusters by minimizing the sum of squared distances between data
COBYLA [12] and an interpolation method [11]), and the number points and the centroids of their respective clusters. For our jet
of times sampling is performed (1024 in this study). Jet clustering clustering study, the angle between two particles is used as the dis-
performance is evaluated based on the angles between the recon- tance metric. The e e kt algorithm identifies the pair of closest
structed jets and the corresponding quarks. In the process particles, combines them, and then repeats this procedure until
e e ZH Z mm H ss , each event produces two jets, resulting some stopping criterion is reached. The distance metric in this
in two angles angle1 and angle2 . This study uses the sum
algorithm is dij 2 min E2i E2j 1 cos hij , where Ei j represents
angle1 angle2 as the performance metric.
The QAOA performance can be enhanced by increasing depth the energy of particle i j and hij represents the angle between par-
and decreasing the evolution times, allowing for more granular ticles i and j. From Fig. 2c, the performance of QAOA is comparable
unitary totter increments. This granularity aids in retaining the to that of the e e kt algorithm and better than that of the k-Means
quantum system’s state within the vicinity of the ground state. algorithm. This comparison highlights the potential of the QAOA in
Fig. 2a compares jet clustering performance on depths of 1, 3, the jet clustering problem.
and 5 with k = 6. The results for depths 3 and 5 are comparable Finally, we conduct the analysis on the BAQIS Quafu quantum
and noticeably superior to that of depth 1. computing cloud, which provides users access to superconducting
A higher value of k signifies that a node is linked to more nodes quantum processors. The Baihua processor features 123 opera-
within the graph, indicating a more complex quantum circuit. tional qubits interconnected through couplers. The relaxation time
Fig. 2b compares the performance of the jet clustering for k of 2, T 1 and the dephasing time T 2 characterize the decoherence of the
4, 6, 7, and 8 with the depth fixed at 3. The optimal jet clustering qubit state. This chip has an average T 1 of 73.994 ls and an average
performance is at k of 7. T 2 of 29.02 ls. The single-qubit gates have an average fidelity
Fig. 2c displays the performance of QAOA with depth = 3 and k (probability of correct output) of 99.9% and the two-qubit gates
= 7 together with that of two classical algorithms: k-Means and (CZ) have an average fidelity of 98.8%. A quantum circuit

Fig. 2. Panels (a), (b), and (c) show the jet clustering performance for 4000 events with 30 particles. (a) The jet clustering performance with k set to 6 and QAOA depths of 1, 3,
and 5. (b) The performance with QAOA depth set to 3 and k set to 2, 4, 6, 7, and 8. (c) Comparison of the QAOA (with depth = 5 and k = 7), the e e kt algorithm, and the k-
Means algorithm. (d) Compiled quantum circuit for performing QAOA-based jet clustering on a 6-particle event, based on the Baihua quantum hardware. The circuit has 34
CNOT gates, 27 single-qubit gates, and a depth of 27. (e) Jet clustering performance for 1217 6-particle events with the QAOA (with depth = 1 and k = 2) run on a quantum
computer and a quantum simulator, and with the classical algorithms.

462
Y. Zhu et al. Science Bulletin 70 (2025) 460–463

implementing QAOA was compiled with the Quafu-Qcover com- (2022YFE0116900), the National Natural Science Foundation of
piler [14]. The compiled quantum circuit is then transformed into China (12305010), the Key Laboratory for Particle Astrophysics
a series of qubit operation commands and sent to the quantum and Cosmology (Ministry of Education), Shanghai Key Laboratory
control system of the quantum computer. The quantum control for Particle Physics and Cosmology, and the Beijing Natural Science
system executes these commands sequentially, with each com- Foundation (Z220002).
mand corresponding to a pulse operation used to control the We thank Hongze Xu for his invaluable support in utilizing
qubits. Quafu Cloud and Andrew Levin for his meticulous efforts in refin-
An ideal quantum processor, free from noise and with sufficient ing the grammar and expressions in our manuscript. We extend
coherence time, can utilize all available qubits for computations. our appreciation to the Beijing Academy of Quantum Information
However, due to current hardware limitations such as noise and Sciences for access to quantum hardware and to the Computing
limited connectivity—where each qubit can only interact with its Center at the Institute of High-Energy Physics for providing com-
immediate 2 or 3 neighbors—we restricted our study to 6 qubits. puting resources.
For events involving 6 particles, we found that the QAOA perfor-
mance is already optimal when the QAOA depth is set to 1 and k
Author contributions
to 2. Fig. 2d illustrates a compiled quantum circuit of QAOA for per-
forming jet clustering on a 6-particle event. The compiled circuit
Yongfeng Zhu analyzed the data and wrote the initial manu-
has 34 CNOT gates, 27 single-qubit gates, and a depth of 26.
script. Weifeng Zhuang, Chen Qian and Yunheng Ma researched
Fig. 2e showcases the performance of this circuit, run on the Baihua
QAOA and discussed it with Yongfeng Zhu. Dong E. Liu, Manqi
processor, applied to 1217 6-particle events. For this small-sized
Ruan, and Chen Zhou conceived of and oversaw the project. All
problem, the quantum hardware achieves similar performance to
authors contributed to analyzing the results and to revising and
a noiseless quantum computer simulator.
reviewing the manuscript.
In summary, the rapid development of quantum algorithms and
hardware devices enables quantum computers to solve small-scale
but representative problems in fundamental science. For example, References
the QCSH package [15] provides an efficient quantum algorithm for
calculating nuclear structure. A quantum algorithm for heat con- [1] Philip K, Morten K, Yan F, et al. A quantum engineer’s guide to superconducting
qubits. Appl Phys Rev 2019;6:021318.
duction [16] significantly outperforms classical counterparts. An [2] McClean JR, Romero J, Babbush R, et al. The theory of variational hybrid
algorithm for simulating ocean circulation on a quantum computer quantum-classical algorithms. New J Phys 2016;18:023023.
[17] and a quantum circuit for implementing spectral clustering [3] Farhi E, Goldstone J, Gutmann S. A quantum approximate optimization
algorithm 2014. arXiv: 1411.4028.
[18] both achieve significant speedup over classical counterparts. [4] Wilson RR. Confinement of quarks. Phys Rev D 1974;10:2445.
We report the first application of QAOA to the problem of jet clus- [5] CEPC Study Group, Dong MY, et al. CEPC conceptual design report: volume 2-
tering. We found that QAOA performance based on a simulation of physics & detector. arXiv: 1811.10545; 2018.
[6] Liang H, Zhu YF, Wang YX, et al. Jet-origin identification and its application at
a quantum computer with 30 qubits is close to the e e kt algo- an electron-positron Higgs factory. Phys Rev Lett 2024;132:221802.
rithm and superior to the k-Means algorithm. The QAOA was also [7] Wei AY, Naik P, Harrow AW, et al. Quantum algorithms for jet clustering. Phys
tested on hardware using 6 qubits on the BAQIS Quafu quantum Rev D 2020;101:094015.
[8] Delgado A, Thaler J. Quantum annealing for jet clustering with thrust. Phys Rev
computing cloud. The performance of the algorithm on the quan- D 2022;106:094016.
tum hardware is similar to the performance on a noiseless quan- [9] Pires D, Omar Y, Seixas J. Adiabatic quantum algorithm for multijet clustering
tum simulator for a small-sized problem. As quantum hardware in high energy physics. Phys Lett B 2023;843:138000.
[10] Proietti M, Cerocchi F, Dispenza M. Native measurement-based quantum
and algorithms continue to advance, we expect to be able to tackle
approximate optimization algorithm applied to the max-k-cut problem. Phys
larger jet clustering problems. This study marks a significant step Rev A 2022;106:022437.
toward a practical quantum computing application in high- [11] Franz GF, Herman OK, Niels HA, et al. Efficient encoding of the weighted max-
k-cut on a quantum computer using QAOA. SN Comp Sci 2020;2:1–14.
energy physics.
[12] Senjean B, Wang H, Vermetten D, et al. Performance comparison of
optimization methods on variational quantum algorithms. Phys Rev A
Conflict of interest 2023;107:032407.
[13] Cacciari M, Salam GP, Soyez G. Fastjet user manual. Eur Phys J C 2012;72:1896.
[14] Xu HZ, Zhuang WF, Wang ZA, et al. Quafu-Qcover: explore combinatorial
The authors declare that they have no conflict of interest. optimization problems on cloud-based quantum computers. Chin Phys B
2024;33:050302.
[15] Lv P, Wei SJ, Xie HN, et al. QCSH: a full quantum computer nuclear shell-model
Acknowledgments package. Sci China-Phys Mech Astron 2023;66:240311.
[16] Wei SJ, Wei C, Lv P, et al. A quantum algorithm for heat conduction with
This work was supported by the Fundamental Research Funds symmetrization. Sci Bull 2023;68:494–502.
[17] Shang R, Wang Z, Shi S, et al. Algorithm for simulating ocean circulation on a
for Central Universities, Peking University, the State Key Labora-
quantum computer. Sci China Earth Sci 2023;66:2254–64.
tory of Nuclear Physics and Technology, Peking University [18] Li QY, Huang YH, Jin S, et al. Quantum spectral clustering algorithm for
(NPT2022ZZ05), the National Key R&D Program of China unsupervised learning. Sci China Inf Sci 2022;65:200504.

463