Photonic processor benchmarking for variational quantum process tomography

Vladlen Galetsky ∗ , Paul Kohl ∗ , Janis Nötzel ∗ · 2025

Read Photonic processor benchmarking for variational quantum process tomography by Vladlen Galetsky ∗ , Paul Kohl ∗ , Janis Nötzel ∗ (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-2507-08570.txt.

Introduction

Checking a quantum processor means learning how it transforms states, a task called process tomography that grows costly fast. Galetsky, Kohl, and Nötzel run a variational version of that check on a photonic processor, using classical one-hot encoding and unitary building blocks. They then race the optical setup against three public quantum machines, including a 127-qubit superconducting system.

The optical processor reached a process fidelity of 0.8 within nine rounds and held up better as circuits deepened, where noise hurt the superconducting rivals. Heat noise in the phase shifters was the main flaw on the optical side. The authors present photonics as a strong near-term platform.

Catalog record

CREATORVladlen Galetsky ∗ , Paul Kohl ∗ , Janis Nötzel ∗
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE38 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2507-08570

Browse the full library · About the Tower of Babel

Full text

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

Photonic processor benchmarking for variational quantum process tomography Preprint: APS/123-QED Vladlen Galetsky ∗ , Paul Kohl ∗ , Janis Nötzel ∗ Abstract We present a quantum-analogous experimental demonstration of variational quantum process tomography using an optical processor. This approach leverages classical one-hot encoding and unitary decomposition to perform the variational quantum algorithm on a photonic platform. We create the first benchmark for variational quantum process tomography evaluating the performance of the quantum-analogous experiment on the optical processor against several publicly accessible quantum computing platforms, including IBM’s 127-qubit Sherbrooke processor, QuTech’s 5-qubit Tuna-5 processor, and Quandela’s 12-mode Ascella quantum optical processor. We evaluate each method using process fidelity, cost function convergence, and processing time per iteration for variational quantum circuit depths of d = 3 d=3 and d = 6 d=6 . Our results indicate that the optical processors outperform their superconducting counterparts in terms of fidelity and convergence behavior reaching fidelities of 0.8 0.8 after 9 9 iterations, particularly at higher depths, where the noise of decoherence and dephasing affect the superconducting processors significantly. We further investigate the influence of any additional quantum optical effects in our platform relative to the classical one-hot encoding. From the process fidelity results it shows that the (classical) thermal noise in the phase-shifters dominates over other optical imperfections, such as mode mismatch and dark counts from single-photon sources. The benchmarking framework and experimental results demonstrate that photonic processors are strong contenders for near-term quantum algorithm deployment, particularly in hybrid variational contexts. This analysis is valuable not only for state and process tomography but also for a wide range of applications involving variational quantum circuit based algorithms. \DeclareMathOperator \sins sins \DeclareMathOperator \sinc sinc \DeclareMathOperator \E E \DeclareMathOperator \sincp sincp I Introduction Classical and quantum optical processors offer several advantages over alternative quantum hardware platforms, including ion traps Liu et al. 2025 , superconducting qubits Arute et al. 2019 , and Rydberg atom arrays Bluvstein et al. 2024 . One key advantage is their compatibility with both quantum and classical photonic infrastructures, especially those operating at telecom wavelengths ( e.g. the C-Band around 1550\text{\,}\mathrm{nm} ), making them promising candidates for distributed quantum computing Taballione et al. 2021 . While ion- and atom-based systems typically require cryogenic cooling at temperatures below 20\text{\,}\mathrm{mK} , photonic processors often can operate at room temperature. Moreover, since photons interact only weakly with their environment, photonic systems exhibit longer coherence times than superconducting systems, which typically suffer from decoherence and dephasing times on the order of 150\text{\,}\mathrm{\SIUnitSymbolMicro s} McKay et al. 2023 , often necessitating quantum error correction for complex algorithms. Photonic processors currently achieve high fidelities for Haar-random unitary implementations, reaching 97,4\text{\,}\mathrm{\%} and 98,6\text{\,}\mathrm{\%} in 12-mode and 20-mode devices, respectively, with corresponding photon losses per mode of approximately 43,8\text{\,}\mathrm{\%}54,3\text{\,}\mathrm{\%} . Taballione et al. 2021 ; de Goede et al. 2022 Recent developments in integrated photonics include the creation of the first Gottesman-Kitaev-Preskill (GKP) qubit by Xanadu Larsen et al. 2025 and a demonstration of quantum computational advantage via Gaussian boson sampling on the Borealis processor Madsen et al. 2022 . New applications continue to emerge, such as the proposal to simulate real-time dynamics of quantum field theories using an optical, measurement-based continuous-variable processor Abel et al. 2024 . A particularly promising application is in variational quantum computation, where hybrid quantum-classical approaches leverage parametrized quantum circuits ( i.e. Variational Quantum Circuits (VQCs) ) on Noisy Intermediate-Scale Quantum (NISQ) devices to perform optimization tasks in a hardware-in-the-loop manner. VQCs form the backbone of quantum machine learning frameworks, including quantum deep neural networks and kernel-based methods Beer et al. 2020 ; Jerbi et al. 2023 . They also play a central role in algorithms such as the Quantum Approximate Optimization Algorithm (QAOA) Farhi et al. 2014 , the Variational Quantum Eigensolver (VQE) Tilly et al. 2022 , and variational approaches for simulating many-body Hamiltonians in condensed matter physics Landman et al. 2022 . In the context of quantum state and process tomography, VQCs provide resource-efficient alternatives to traditional measurement-based methods Mohseni et al. 2008 . Vianna et al. 2013 first proposed variational quantum process tomography using an optical quantum processor Vianna et al. 2013 , approximating the process matrix \chi variationally to fit experimental data. Their method matched the accuracy of standard techniques while reducing the required number of measurements from 256 to just 55. In another application, Carolan et al. 2020 Carolan et al. 2020 used a quantum optical processor to variationally compile and invert an unknown unitary operation using a limited number of input states. More recently, Xue et al. 2022 demonstrated an entanglement-assisted approach to process tomography on an optical processor Xue et al. 2022 , reducing the number of required input states from 4^{n} to 2^{n-m} however requiring entanglement based state initialization as well as using m auxiliary qubits. II Our contributions In this work, we propose utilizing the architecture introduced in Galetsky et al. 2024 , adapted to a local optical processor employing one-hot encoding to emulate the behavior of a quantum optical processor, as illustrated in Fig. 2 . This approach reduces the number of required input state initializations from the standard 4^{n} to 2^{n} orthogonal states per parameter update. For an n -qubit system, this one-hot encoding scheme requires a total of 2^{t+1} optical modes, with t being the target qubit dimension. Due to the limited benchmarking available across different quantum hardware platforms, this study aims to be the first to evaluate the applicability of variational quantum process tomography across various NISQ devices. Specifically, we analyze cost function behavior, process fidelity, and estimated processing time for two different circuit depths applied to Haar-random two-qubit unitary processes. We consider d=6 as the optimal depth for this qubit size, following the definition in Galetsky et al. 2024 . The platforms evaluated include our local optical processor, the Tuna 5-qubit superconducting processor from QuTech Vallés-Sanclemente et al. 2025 , IBM’s 127-qubit Sherbrooke processor McKay et al. 2023 , and the Ascella quantum optical processor developed by Quandela Maring et al. 2024 . This comparison is not only relevant for process tomography but also provides insights into the general utility of other VQC based algorithms such as QAOA and VQE in NISQ devices. Furthermore, we investigate how effectively one-hot encoding implemented on a classical optical processor can approximate the behavior of a genuine quantum optical processor. Despite the presence of hidden quantum noise sources such as dark counts in single-photon detectors and mode mismatch, from the comparison of the cost function and process fidelity our analysis indicates that these effects are negligible as the noise effects are comparable between the two devices Rohde and Ralph 2006 . Our experimental results show that, for both quantum and classical optical processors, increasing the circuit depth from d=3 to d=6 improves process fidelity from 0,71\text{\,} to 0,80\text{\,} , while reducing the cost function from 0,35\text{\,} to 0,22\text{\,} . In contrast, the performance of superconducting processors did not improve beyond four iterations due to cumulative decoherence and dephasing effects introduced by the increased circuit depth. Additionally, the initial average process fidelity for these platforms degraded from 0,42\text{\,} to 0,30\text{\,} as depth increased, likely due to a larger parameter space of the process matrix being represented

Read the complete text