Machine-Learning-Enhanced Optimization of Noise-Resilient Variational Quantum Eigensolvers

Lena Funcke · 2025

Read Machine-Learning-Enhanced Optimization of Noise-Resilient Variational Quantum Eigensolvers by Lena Funcke (2025) 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-2501-17689.txt.

Introduction

Kim Nicoli, Luca Wagner, and Lena Funcke report on a machine learning upgrade for variational quantum eigensolvers, a leading method for finding the ground state energy of a quantum system. These hybrid algorithms run partly on quantum hardware, whose noise corrupts their results, and partly on classical computers that steer the search. Earlier work proposed guiding that search with Gaussian processes and Bayesian optimization, which cope well with noisy scores.

These proceedings paper adds experiments on how hardware noise and error mitigation change the picture. The team tested the method in classical simulations that mimic real quantum devices, including noise benchmarks not covered before. The enhanced optimizer beat strong standard baselines. The authors frame the work as groundwork for running such solvers on real hardware and for problems in lattice field theory.

Catalog record

CREATORLena Funcke
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE38 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2501-17689

Browse the full library · About the Tower of Babel

Full text

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

Machine-Learning-Enhanced Optimization of Noise-Resilient Variational Quantum Eigensolvers Lena Funcke Abstract Variational Quantum Eigensolvers (VQEs) are a powerful class of hybrid quantum-classical algorithms designed to approximate the ground state of a quantum system described by its Hamiltonian. VQEs hold promise for various applications, including lattice field theory. However, the inherent noise of Noisy Intermediate-Scale Quantum (NISQ) devices poses a significant challenge for running VQEs as these algorithms are particularly susceptible to noise, e.g., measurement shot noise and hardware noise. In a recent work, it was proposed to enhance the classical optimization of VQEs with Gaussian Processes (GPs) and Bayesian Optimization, as these machine-learning techniques are well-suited for handling noisy data. In these proceedings, we provide additional insights into this new algorithm and present further numerical experiments. In particular, we examine the impact of hardware noise and error mitigation on the algorithm’s performance. We validate the algorithm using classical simulations of quantum hardware, including hardware noise benchmarks, which have not been considered in previous works. Our numerical experiments demonstrate that GP-enhanced algorithms can outperform state-of-the-art baselines, laying the foundation for future research on deploying these techniques to real quantum hardware and lattice field theory setups. [a,b]Kim A. Nicoli [a,b]Luca J. Wagner 1 Introduction Quantum hardware with several hundreds of qubits can already be harnessed to outperform classical computers on specific tasks (see, e.g., Ref. [ 1 ] ). These tasks, however, currently have no practical applications and are specifically designed to be challenging for classical computers. As we have entered the Noisy Intermediate-Scale Quantum (NISQ) era of quantum computing, the primary challenge is to develop algorithms for NISQ devices that may demonstrate quantum advantage in tasks of practical relevance, such as quantum chemistry [ 2 ] or quantum field theories [ 3 , 4 ] . Particularly relevant for NISQ devices are hybrid quantum-classical algorithms, such as the Variational Quantum Eigensolver (VQE) [ 5 , 6 ] . VQEs can approximate the ground state of quantum Hamiltonians by classical variational optimization of a parametrized quantum circuit. The quantum computer measures the energy of the quantum system for a given set of parameters, i.e., the objective of the minimization problem, while a classical optimization routine is used to find a better set of parameters. Efficient exploration of the parameter space is crucial for successful optimization. Nakanishi et al. [ 7 ] demonstrated that, under certain conditions, the functional form of the VQE’s objective function can be explicitly derived and utilized to develop a more efficient optimization routine. The resulting Nakanishi-Fuji-Todo (NFT) algorithm [ 7 ] uses two measurements to sequentially optimize one parameter of the quantum circuit at a time, by fitting the objective in a one-dimensional subspace. Indeed, this constitutes a sequential minimal optimization (SMO) [ 8 ] , which differs from global optimization protocols such as gradient-based approaches. Despite its efficiency, the NFT algorithm faces several challenges, primarily due to the noise of current quantum hardware. In particular, the algorithm relies on noisy measurements to identify the global minimum in the one-dimensional subspace, which often hinders convergence and complicates optimization. Machine learning offers a promising approach to addressing the challenge of noisy measurements and enhancing the NFT algorithm with more flexibility. In a recent work [ 9 ] , Nicoli et al. proposed an optimization algorithm called NFT-with-EMICoRe (EMICoRe for short), which extends the NFT framework by incorporating Gaussian Process Regression (GPR) and Bayesian Optimization (BO) for a data-driven and potentially more noise-resilient SMO scheme. However, this work [ 9 ] focused solely on measurement shot noise, neglecting the impact of quantum hardware noise, such as decoherence, depolarization, and crosstalk errors. In these proceedings, we build on prior works [ 7 , 9 ] and demonstrate that EMICoRe outperforms the NFT baseline in the presence of simulated quantum hardware noise. We start in Sec. 2 by reviewing the theoretical foundations of VQE, Gaussian Processes, Bayesian Optimization. In Sec. 3 , we introduce the EMICoRe algorithm. This is followed by numerical experiments in Sec. 4 , which involve classical simulations of noisy quantum hardware, comparing the results with and without error mitigation. We conclude and provide an outlook in Sec. 5 . 2 Theoretical Foundations 2.1 Variational Quantum Eigensolver (VQE) Computing the ground state energy of a quantum system described by a Hamiltonian H and the corresponding ground state wave function \ket{\psi_{\mathrm{GS}}} is a key challenge in many fields of physics. The Hamiltonian of an N -qubit system can be represented as H=\sum_{\alpha=1}^{T}h_{\alpha}P_{\alpha}, (1) where P_{\alpha}\in\{X,Y,Z,I\}^{\otimes N} is a tensor product of single-qubit Pauli operators X,Y,Z and the identity operator I . The h_{\alpha} are real coefficients, and T is the number of individual Pauli terms. In general, it is challenging to compute the ground state wave function \ket{\psi_{\mathrm{GS}}} because it requires diagonalizing the Hamiltonian H , which is a \mathbb{C}^{2^{N}\times 2^{N}} -dimensional matrix for an N -qubit system and thus scales exponentially with the number of qubits. An alternative approach to exact diagonalization is to compute a variational approximation of the ground state. This can be achieved, for instance, by using the Variational Quantum Eigensolver (VQE) [ 5 , 6 ] , where a parametric quantum state |\psi(\boldsymbol{\theta})\rangle is prepared on a quantum computer. This quantum state, which depends on D angular parameters \boldsymbol{\theta}\in[0,2\pi)^{D} , is obtained through a sequence of D^{\prime} ( \geq D ) unitary quantum gate operations U(\boldsymbol{\theta})=U_{1}\cdots U_{D^{\prime}} acting on an initial state |\psi_{0}\rangle . A subset D of these operations are Pauli rotation gates, U_{d^{\prime}(d)}(\theta_{d})=R_{P_{d}}(\theta_{d})=\exp(-i\theta_{d}P_{d}) , where P_{d}\in\{X,Y,Z\} are the Pauli matrices and \theta_{d} are the angular parameters. The full variational state can thus be written as |\psi(\boldsymbol{\theta})\rangle=U(\boldsymbol{\theta})|\psi_{0}\rangle . After the preparation of the quantum state |\psi(\boldsymbol{\theta})\rangle on the quantum computer, the state is measured in the basis of the individual Pauli strings to obtain the corresponding measured energy given by \displaystyle\tilde{E}(\boldsymbol{\theta}) \displaystyle=E^{*}(\boldsymbol{\theta})+\varepsilon,\quad (2) \displaystyle\textrm{where}\quad E^{*}(\boldsymbol{\theta}) \displaystyle=\left\langle\psi(\boldsymbol{\theta})|H|\psi(\boldsymbol{\theta})\right\rangle=\left\langle\psi_{0}\left|U(\boldsymbol{\theta})^{\dagger}HU(\boldsymbol{\theta})\right|\psi_{0}\right\rangle\text{. } (3) The energy E^{\ast}(\boldsymbol{\theta}) is the true energy that would be obtained if the quantum computer were noise-free. Typically, a quantum computer is subject to both Gaussian noise, arising from the statistical nature of repeated measurements (i.e., measurement shot noise ), and hardware noise , which stems from imperfections in qubits, gates, and measurements [ 10 ] . Both types of noise are encoded in the noise term \varepsilon in Eq. 2 . As shown in the green box of Fig. 1 , the VQE protocol can be divided into the following steps: 1. Initialize a random quantum state with random parameters \boldsymbol{\theta} . 2. Measure the expectation value \tilde{E}(\boldsymbol{\theta})=\left\langle\psi(\boldsymbol{\theta})|H|\psi(\boldsymbol{\theta})\right\rangle+\varepsilon on the quantum computer. 3. Use a classical non-linear optimizer, such as gradient-based or gradient-free methods, to determine new parameters {\boldsymbol{\hat{\theta}}} that minimize \tilde{E}(\boldsymbol{\theta}) . 4. Iterate to step 2 until \tilde{E}(\boldsymbol{\theta}) converges to the ground state energy. Figure 1 : Illustration of the VQE workflow. For more details, see the text. Under mild assumptions for the above described variational circuit, Nakanishi et al. derived an analytical expression of the VQE’s objective functional [ 7 ] , E^{\ast}(\boldsymbol{\theta})=\boldsymbol{b}^{\top}\cdot\left[\bigotimes^{D}_{d=1}\left(\begin{array}[]{c}\cos{(\theta_{d})}\\ \sin{(\theta_{d})}\\ 1\end{array}\right)\right]\textrm{ , }\quad\forall\boldsymbol{\theta}\in[0,2\pi)^{D}, (4) where \boldsymbol{b} is a set of arbitrary coefficients and \theta_{d} is a D -dimensional vector of angular parameters for the variational quantum circuit. In particular, this functional form leads to the following observation: when keeping

Read the complete text