quantum-classical orchestration

This work addresses the deep integration of quantum computing with high-performance computing (HPC). We conduct a systematic literature review of 107 studies and propose a seven-category classification framework spanning hardware architectures, software stacks, programming models, and scheduling mechanisms. Methodologically, we adopt “interface standardization” and “co-abstraction” as unifying conceptual lenses to map the technical landscape of quantum-classical hybrid systems. Our analysis identifies three critical challenges: heterogeneous communication latency, cross-layer scheduling inefficiency, and fragmented programming models. The contribution is a structured design pathway for hybrid computing architectures, enabling cross-platform toolchain development and informing future standardization efforts. By clarifying integration bottlenecks and abstraction requirements, this work advances quantum computing’s evolution from domain-specific accelerators toward interoperable, composable computing units within HPC ecosystems.

This work proposes the first cloud-native scheduling framework systematically designed for hybrid quantum-classical computing to address the challenge of efficiently orchestrating heterogeneous computational resources at scale. Built upon Kubernetes, Argo Workflows, and Kueue, the framework enables unified, resource-aware dynamic scheduling across CPUs, GPUs, and quantum processing units (QPUs), supporting multi-stage, reproducible, and observable hybrid workflows. The effectiveness of the framework is demonstrated through end-to-end collaborative experiments on distributed quantum circuit cutting tasks, which highlight its significant advantages in scalability, flexibility, and reproducibility.

This work addresses the resource scheduling challenges faced by hybrid quantum-classical applications in heterogeneous and dynamic computing environments, where existing high-performance computing (HPC) schedulers lack application semantics awareness and runtime adaptability. To overcome these limitations, the authors propose a four-layer middleware architecture that integrates an abstract execution model for hybrid applications, a Pilot-Quantum dynamic scheduling framework, and the Q-Dreamer performance modeling toolkit. This integrated approach enables application-aware, adaptive resource management and optimized quantum circuit partitioning. The system supports coordinated scheduling across CPU, GPU, and quantum processing unit (QPU) backends and has been validated on the Perlmutter and NVIDIA DGX platforms. Experimental results demonstrate that Q-Dreamer achieves an 82% accuracy rate in predicting optimal circuit-cut configurations.

This study addresses the lack of orchestration, recovery, and portability in driver scripts for hybrid quantum-classical optimization by modeling iterative decomposition-solving-aggregation loops as scientific workflows. A dedicated orchestration layer is constructed to uniformly manage task generation, data provenance, and fault recovery. The proposed workflow model incorporates termination predicates, subproblem-level recovery, and QPU-to-classical-backend failover, while integrating ADMM, hierarchical partitioning, speculative re-execution, and quantum-HPC middleware. Experiments quantify the orchestration overhead across different decomposition patterns, validate system robustness under fault injection, and perform cross-device latency analysis.

This work addresses the critical challenge of efficiently integrating quantum processing units (QPUs) into existing high-performance computing (HPC) systems to enable unified scheduling and协同 execution of classical and quantum resources. The paper proposes a Quantum-integrated High-Performance Computing (QHPC) architecture—the first HPC framework explicitly designed for quantum-classical convergence—featuring a layered design that holistically manages CPUs, GPUs, FPGAs, and QPUs. Key innovations include a user request abstraction layer offering a Slurm-like unified interface, a quantum-aware scheduling algorithm, a hybrid workflow engine, and a hierarchical execution model supported by high-speed interconnects. This architecture provides scalable heterogeneous computing support for emerging applications such as quantum chemistry, materials discovery, combinatorial optimization, and climate modeling, thereby laying the foundation for a future quantum-classical HPC ecosystem.

Quantum-classical hybrid applications face significant challenges in cloud environments, including complex scheduling, severe resource contention, and noise-induced fidelity degradation. To address these issues, this paper introduces Qonductor—the first cloud-native orchestration system designed specifically for hybrid quantum-classical computing. Its core contributions are: (1) a hardware-agnostic unified API for quantum and classical workloads; (2) a fidelity-aware heterogeneous resource estimation model; and (3) a multi-objective hybrid scheduler jointly optimizing user QoS (latency and fidelity) and operator resource efficiency. Built as a Kubernetes extension, Qonductor integrates dynamic fidelity modeling and load-aware scheduling. Evaluated on real IBM quantum processing units (QPUs), it reduces job completion time by 54%, increases QPU utilization to 66%, and maintains high scalability—supporting thousands of concurrent tasks—while incurring only a 6% average fidelity overhead.

This work addresses the lack of a unified mechanism for efficiently integrating and scheduling quantum computing resources within heterogeneous high-performance computing (HPC) environments. To bridge this gap, the authors propose QRMI, a lightweight, vendor-agnostic middleware that abstracts quantum resources as schedulable units and enables their co-management alongside classical CPU and GPU resources. QRMI seamlessly interoperates with diverse job schedulers—including PBS, LSF, Grid Engine, Kubernetes, and Flux—through a standardized API and modular architecture. Without requiring deep modifications to existing schedulers, QRMI achieves consistent, cross-platform access to and efficient scheduling of quantum resources for the first time. Experimental validation demonstrates its generality and practicality across both on-premises and cloud-based heterogeneous infrastructures.

This study addresses the challenges of quantum-classical heterogeneous integration and collaborative scheduling in supercomputing by proposing a two-tier scheduling architecture and an open-source Slurm plugin suite. Leveraging GRES mechanisms, QRM&CI just-in-time compilation, and SPANK modules, this approach transparently integrates quantum processors as accelerators into existing HPC scheduling frameworks, enabling efficient hybrid workflow orchestration without modifying the scheduler core. Experimental evaluations on next-generation Cray platforms demonstrate negligible latency overhead, confirming both portability and practicality. Consequently, this work provides a standardized solution for managing heterogeneous quantum computing resources, facilitating seamless integration within current high-performance computing environments while maintaining system integrity and performance efficiency.

This work addresses the challenge of coordinating classical scheduling queues with remote quantum device external queues when integrating quantum processing units (QPUs) into cloud-native orchestration systems. To this end, the authors propose Fluence, a Kubernetes scheduler plugin built upon the Fluxion graph scheduler. Fluence introduces quantum-aware scheduling into cloud-native environments for the first time, enabling efficient co-scheduling of hybrid workloads without requiring modifications to user containers. It achieves this through atomic gang scheduling, synchronization primitives, and a cost- and queue-aware backend auto-selection mechanism. Experimental results demonstrate that Fluence virtually eliminates resource waste under node contention, reduces idle time by over fivefold, lowers per-run costs by approximately 70×, and shortens job completion times from hours to under one minute.

This work addresses the poor energy efficiency of conventional quantum computing architectures and the absence of system-level integration strategies aligned with sustainability goals. It proposes a heterogeneous hybrid architecture that synergistically integrates edge, cloud, and high-performance computing through a three-tier design: a shared optical fiber physical layer, a user-managed control and orchestration layer, and an Adaptive Quantum-Classical Fusion (AQCF) application framework. By extending system integration from the cloud down to the cryogenic logic level, the architecture introduces the novel concept of “green performance advantage.” This approach substantially reduces thermal footprint and energy consumption, achieving measurable gains in energy efficiency—quantified by energy per problem solved—and thereby advancing the development of sustainable, quantum-enhanced computing.

This work addresses the tight coupling between circuit cutting logic and execution scheduling in existing quantum circuit partitioning frameworks, which hinders high-performance computing (HPC) systems from applying mature resource management strategies to NISQ workloads. To overcome this limitation, the authors propose DQR, a runtime framework that, for the first time, abstracts quantum circuit fragments as first-class schedulable units. By introducing backend-agnostic structured descriptors, a wave-based coordinator, and a non-blocking polling mechanism, DQR decouples cutting from scheduling. The framework enables transparent fault tolerance, heterogeneous backend integration, and hybrid local-cloud scheduling. Experimental results demonstrate significant reductions in job completion time on 32-qubit hardware-efficient ansatz circuits, with coordination overhead accounting for only 5% in deep circuits, and automatic migration of failed fragments to classical simulators, confirming its high scalability and flexibility.