Evaluating Quantum-Classical Middleware using Quantum Mini-Apps

Abstract

As quantum computing evolves toward practical hybrid applications, middleware systems must support efficient coordination between quantum and classical resources. However, existing middleware platforms are often benchmarked using hardware-centric metrics, without standardized workloads or evaluation frameworks that reflect real-world application behavior. This thesis adopts a motif-driven evaluation strategy based on lightweight, reusable mini-applications. A mini-app is a small, self- contained program that models a specific hybrid quantum-classical execution pattern. Unlike traditional benchmarks that only focus pri- marily on low-level metrics such as qubit count or circuit depth, motif- based benchmarking targets execution motifs, which are recurring interaction and coupling patterns that characterize hybrid workflows. This approach enables system-aware evaluation of middleware by capturing key behaviors such as task orchestration, quantum–classical coupling, and scalability. Through a structured literature review, five critical execution motifs are identified: A base motif of Circuit Optimization (CO), and com- positional motifs of Variational Quantum Algorithms (VQA), Circuit Cutting, and Multistage Pipelines. These motifs represent recurring hybrid patterns such as iterative feedback loops, modular decomposi- tion, and noise-aware transformations. Each motif is implemented as a quantum mini-app using Qiskit or PennyLane, and executed on a Ray- and SLURM-based cluster via the Pilot-Quantum middleware. Evaluation metrics are categorized into three domains: application- oriented (for example, fidelity and Jensen–Shannon divergence), system-level (such as qubit count and circuit depth), and runtime per- formance (such as execution time and sampling overhead). Results demonstrate how motif design significantly influences middleware behavior, particularly in compositional workflows involving noise mitigation and circuit decomposition. This thesis demonstrates a systematic method for benchmarking middleware using motif-specific mini-apps and contributes reusable tools and metrics for evaluating orchestration, scalability, and fidelity in hybrid quantum-classical systems.