Impact of Circuit Depth versus Qubit Count on Variational Quantum Classifiers for Higgs Boson Signal Detection

Fatih Maulana · 2026

Read Impact of Circuit Depth versus Qubit Count on Variational Quantum Classifiers for Higgs Boson Signal Detection by Fatih Maulana (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2601-11937.txt.

Introduction

Fatih Maulana asks how circuit depth and qubit count trade off in quantum classifiers. The test task is to spot Higgs boson signals in crash data. Such data sets are vast and strain plain machines. Quantum learning may help, but today's quantum chips are noisy and small.

He varies depth and qubit count to see which matters more. Deep circuits can express more, yet build more noise. More qubits hold more data traits at once. The study maps that trade for one real physics task on near-term chips. The maps help pick chip size for such search tasks.

Catalog record

CREATORFatih Maulana
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE23 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2601-11937

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Impact of Circuit Depth versus Qubit Count on Variational Quantum Classifiers for Higgs Boson Signal Detection Fatih Maulana Abstract High-Energy Physics (HEP) experiments, such as those at the Large Hadron Collider (LHC), generate massive datasets that challenge classical computational limits. Quantum Machine Learning (QML) offers a potential advantage in processing high-dimensional data; however, finding the optimal architecture for current Noisy Intermediate-Scale Quantum (NISQ) devices remains an open challenge. This study investigates the performance of Variational Quantum Classifiers (VQC) in detecting Higgs Boson signals using the ATLAS Higgs Boson Machine Learning Challenge 2014 experiment dataset. We implemented a dimensionality reduction pipeline using Principal Component Analysis (PCA) to map 30 physical features into 4-qubit and 8-qubit latent spaces. We benchmarked three configurations: (A) a shallow 4-qubit circuit, (B) a deep 4-qubit circuit with increased entanglement layers, and (C) an expanded 8-qubit circuit. Experimental results demonstrate that increasing circuit depth significantly improves performance, yielding the highest accuracy of 56.2% (Configuration B), compared to a baseline of 51.9%. Conversely, simply scaling to 8 qubits resulted in a performance degradation to 50.6% due to optimization challenges associated with Barren Plateaus in the larger Hilbert space. These findings suggest that for near-term quantum hardware, prioritizing circuit depth and entanglement capability is more critical than increasing qubit count for effective anomaly detection in HEP data. Keywords: Quantum Machine Learning, Higgs Boson, Variational Quantum Classifier, Dimensionality Reduction, High-Energy Physics. 1 Introduction The discovery of the Higgs Boson by the ATLAS and CMS collaborations at CERN marked a milestone in particle physics. However, identifying the Higgs signal from the overwhelming background noise of Standard Model processes is a computationally intensive task. As the luminosity of the Large Hadron Collider (LHC) increases, the volume of data is expected to reach exabytes, posing significant challenges for classical Deep Learning algorithms in terms of training time and energy consumption. Quantum Machine Learning (QML) has emerged as a promising paradigm to address these challenges. Quantum algorithms utilizing Hilbert space can potentially identify complex correlations in high-dimensional data more efficiently than classical counterparts [ 1 , 2 , 3 ] . In the current era of Noisy Intermediate-Scale Quantum (NISQ) computing, Variational Quantum Algorithms (VQA), specifically the Variational Quantum Classifier (VQC), are the most viable candidates [ 8 , 9 , 10 ] for practical application. However, NISQ devices are limited by qubit count and coherence time. This research aims to address a critical architectural question: To improve classification performance on HEP data, is it more effective to increase the number of qubits (width) or the complexity of the circuit (depth)? By utilizing the ATLAS Higgs Challenge dataset [ 4 ] , similar to recent benchmarks in HEP QML [ 5 , 6 , 15 ] , this study benchmarks the trade-offs between dimensionality reduction (PCA) and quantum circuit architecture, providing guidelines for designing efficient QML models for particle physics. 2 Methodology 2.1 Classical Data Representation and Preprocessing The dataset D=\{(\mathbf{x}_{i},y_{i})\}_{i=1}^{N} consists of N events, where \mathbf{x}_{i}\in\mathbb{R}^{30} represents the feature vector containing physical quantities (e.g., DER_mass_MMC , momentum p_{T} ) derived from the particle collisions, and y_{i}\in\{0,1\} represents the class label (Background or Higgs Signal). To prepare the classical data for quantum encoding, we first apply normalization to constrain the features within the domain [0,1] , ensuring compatibility with the rotation angles of quantum gates: \mathbf{x}^{\prime}_{i}=\frac{\mathbf{x}_{i}-\min(\mathbf{x})}{\max(\mathbf{x})-\min(\mathbf{x})} (1) 2.2 Dimensionality Reduction via PCA Given the limited qubit count ( n_{q} ) of current NISQ devices (where n_{q}\ll 30 ), we employ Principal Component Analysis (PCA) [ 18 ] to project the high-dimensional feature space into a reduced latent space. Let \Sigma be the covariance matrix of the standardized data. We compute the eigendecomposition: \Sigma=V\Lambda V^{T} (2) where V contains the eigenvectors (principal components) and \Lambda is the diagonal matrix of eigenvalues. We select the top k eigenvectors corresponding to the largest eigenvalues to form the projection matrix W_{k}\in\mathbb{R}^{30\times k} . The reduced feature vector \mathbf{z}_{i} is obtained by: \mathbf{z}_{i}=\mathbf{x}^{\prime}_{i}W_{k},\quad\text{where }k\in\{4,8\} (3) This transformation retains the maximum variance of the physical properties while satisfying the constraint k=n_{q} . 2.3 Quantum Variational Classifier (VQC) Formulation The classification is performed using a parameterized quantum circuit (PQC). The process consists of three mathematically distinct stages: 1. Data Encoding (Feature Map) The classical latent vector \mathbf{z} is mapped into the Hilbert space \mathcal{H} using a unitary operation U_{\Phi}(\mathbf{z}) . We utilize the ZZFeatureMap [ 7 ] , which induces entanglement to capture non-linear correlations. For a qubit j , the initial state |0\rangle^{\otimes n} is transformed as: |\psi(\mathbf{z})\rangle=U_{\Phi}(\mathbf{z})|0\rangle^{\otimes n}=\left(\prod_{j=1}^{n}e^{i\phi(z_{j})Z_{j}}\prod_{(j,k)}e^{i\phi(z_{j},z_{k})Z_{j}Z_{k}}\right)H^{\otimes n}|0\rangle^{\otimes n} (4) where H is the Hadamard gate, Z is the Pauli-Z operator, and \phi(\cdot) describes the phase encoding function. Figure 1 : High-level schematic of the Variational Quantum Circuit (VQC) architecture. The circuit is composed of a ZZFeatureMap for data encoding and a RealAmplitudes ansatz for the trainable variational layers. 2. Variational Ansatz (The Learning Model) To separate the classes in Hilbert space, we apply a trainable unitary circuit W(\theta) , parameterized by vector \theta . We employ the RealAmplitudes ansatz, characterized by layers of parameterized R_{y} rotations and CNOT entangling gates. The rotation gate is defined as: R_{y}(\theta_{j})=\exp\left(-i\frac{\theta_{j}}{2}Y\right)=\begin{pmatrix}\cos(\frac{\theta_{j}}{2})&-\sin(\frac{\theta_{j}}{2})\\ \sin(\frac{\theta_{j}}{2})&\cos(\frac{\theta_{j}}{2})\end{pmatrix} (5) The depth of the circuit ( D ) corresponds to the number of repetitions (reps) of these entanglement layers. A deeper circuit increases the expressibility of W(\theta) [ 11 , 12 ] , allowing the model to approximate more complex decision boundaries. 3. Measurement and Prediction The final quantum state |\Psi_{final}\rangle=W(\theta)|\psi(\mathbf{z})\rangle is measured to obtain a classical prediction. The probability of the event belonging to the Signal class ( y=1 ) is derived from the expectation value of the Pauli-Z operator (parity) on the measured qubits: P(\hat{y}=1)=\frac{1}{2}\left(1+\langle\Psi_{final}|\hat{Z}^{\otimes n}|\Psi_{final}\rangle\right) (6) The optimal parameters \theta^{*} are found by minimizing a loss function (e.g., Cross-Entropy) utilizing the COBYLA optimizer implemented using Qiskit SDK [ 19 , 20 ] . 2.4 Experimental Hyperparameter Tuning Strategy To achieve optimal performance on Noisy Intermediate-Scale Quantum (NISQ) devices, we designed a three-stage tuning strategy that explores the trade-off between Expressibility and Trainability . • Stage I: The Baseline (Shallow Circuit) Objective: To establish a performance baseline using a minimal circuit configuration. Configuration: Input Dimension ( n=4 ), Circuit Depth ( d=1 ). Analysis: Number of parameters N_{\theta}=n(d+1)=8 . We hypothesized that this configuration might be too linear to capture the complex decision boundary, leading to underfitting. • Stage II: Depth Optimization (Deep Circuit) Objective: To enhance the model’s expressibility. Configuration: Input Dimension ( n=4 ), Circuit Depth ( d=2 ). Analysis: Parameter space expands to N_{\theta}=12 . Increasing depth ( d ) increases the frequency spectrum available to the model, forming more curved decision boundaries. • Stage III: Width Expansion (High-Dimensional Space) Objective: To test incorporating more physical information (8 PCA features). Configuration: Input Dimension ( n=8 ), Circuit Depth ( d=1 ). Analysis: The Hilbert space expands exponentially to 2^{8}=256 . This configuration is susceptible to the Barren Plateau phenomenon [ 13 , 14 ] , where the variance of the cost function gradients vanishes exponentially with the number of qubits ( n ): \text{Var}\left(\frac{\partial\mathcal{L}}{\partial\theta_{k}}\right)\propto\frac{1}{2^{n}} (7) For n=8 , this variance becomes negligible, causing the optimization landscape to become effectively flat. 3 Experimental Results The models were implemented using Qiskit SDK and trained using

Read the complete text