Quantum Algorithm Scales Optimisation with HPC and GPU Acceleration.

The Distributed Approximate Optimisation Algorithm (DQAOA) demonstrates improved scalability and efficiency in solving complex combinatorial optimisation problems. Advanced problem decomposition and parallel execution on the Frontier supercomputer achieve up to a tenfold speedup compared to CPU-based simulations, facilitating practical deployment for hybrid classical-quantum applications. Integration with the Framework (QFw) supports future high-performance computing systems.

Combinatorial optimisation problems, prevalent in fields ranging from logistics and finance to materials science and machine learning, often present computational challenges that exceed the capabilities of classical algorithms. Researchers are increasingly exploring hybrid quantum-classical approaches to tackle these complexities, with the Distributed Approximate Optimisation Algorithm (DQAOA) emerging as a promising technique. DQAOA addresses high-dimensional, dense problems by leveraging both quantum and classical computational resources. A collaborative team comprising Zhihao Xu, Srikar Chundury, Seongmin Kim, Amir Shehata, Xinyi Li, Ang Li, Tengfei Luo, Frank Mueller, and In-Saeng Suh, affiliated with institutions including Oak Ridge National Laboratory, North Carolina State University, Pacific Northwest National Laboratory, and the University of Notre Dame, detail their work on enhancing the scalability and efficiency of DQAOA in a paper entitled ‘GPU-Accelerated Distributed QAOA on Large-scale HPC Ecosystems’. Their research focuses on advanced problem decomposition and parallel execution utilising the Frontier supercomputer, demonstrating significant performance gains through GPU acceleration and efficient workload management.

Oak Ridge National Laboratory researchers are advancing quantum optimisation through a distributed computational approach, tackling complex, high-dimensional combinatorial problems with a refined Distributed Approximate Quantum Optimisation Algorithm (DQAOA). DQAOA is a variational quantum algorithm designed to find approximate solutions to optimisation problems, employing a quantum circuit with adjustable parameters. The team demonstrates significant improvements in scalability and efficiency by strategically decomposing these problems and executing them in parallel on the Frontier CPU/GPU supercomputer, potentially enabling practical applications of hybrid quantum-classical algorithms.

Researchers address limitations inherent in existing quantum optimisation techniques by focusing on problem decomposition, a crucial step in dividing complex challenges into smaller, more manageable sub-problems suitable for quantum annealing. They optimise these decomposition strategies, coupling them with simulations accelerated by Graphics Processing Units (GPUs) to demonstrably improve DQAOA’s performance and expand its capabilities. Benchmarking reveals a substantial speedup, with the optimised DQAOA achieving up to ten times faster execution compared to simulations run solely on Central Processing Units (CPUs).

The study emphasises the importance of efficient workload management within a hybrid computing environment, intelligently distributing computational tasks between classical and quantum processors to minimise bottlenecks and maximise overall performance. Researchers utilise message passing, facilitated by tools like MPI (Message Passing Interface), a standardised protocol for communication between computer processes, to enable effective communication and data exchange between processing units, streamlining the computational process. Furthermore, the integration of the Quantum Framework (QFw) signifies ongoing efforts to support future High Performance Computing (HPC) systems and facilitate seamless integration of quantum resources, building a foundation for future advancements.

Researchers demonstrate the potential of combining advanced algorithmic techniques with powerful HPC infrastructure to unlock the capabilities of quantum-inspired optimisation algorithms, paving the way for advancements in fields such as logistics, finance, and materials science. They achieve substantial performance gains, highlighting the viability of using GPU systems for large-scale hybrid quantum-classical applications. The ongoing development of frameworks like Qfw further solidifies the foundation for future research and deployment of quantum-enhanced computing solutions.

The study establishes that strategic problem decomposition, coupled with parallel execution on the Frontier CPU/GPU architecture, markedly improves DQAOA’s scalability and efficiency, addressing a key limitation of previous approaches. Researchers facilitate effective workload management by distributing large problem instances across both classical and quantum resources, optimising resource utilisation. Experimental results confirm substantial performance gains, with the enhanced decomposition strategies and GPU-accelerated simulations achieving up to a tenfold increase in speed compared to CPU-based simulations.

Researchers focus on refining the problem decomposition strategies to further enhance scalability and explore the potential of incorporating more advanced quantum algorithms within the DQAOA framework, pushing the boundaries of quantum optimisation. They investigate the performance of DQAOA on various problem instances, identifying areas for further improvement and optimisation. This iterative process of refinement and optimisation ensures that DQAOA remains at the forefront of quantum optimisation research.

Researchers build upon a diverse range of publications, from conference proceedings and technical reports to established books on artificial intelligence, demonstrating a thorough understanding of the existing literature and a commitment to building upon established knowledge. They leverage established software tools like Qiskit, an open-source framework for quantum computing, ensuring reproducibility and facilitating collaboration within the research community. This commitment to open science and collaboration accelerates the pace of innovation in the field of quantum optimisation.