Multi-GPU-Enabled Hybrid Quantum-Classical Workflow in Quantum-HPC Middleware: Applications in Quantum Simulations

TL;DR

The paper addresses the challenge of scalable quantum simulations by bridging quantum hardware with classical HPC through a distribution-aware QCQ (Quantum-Classical-Quantum) workflow. It implements a VQE-based state preparation on QPUs, tensor-network inspired variational ansatzes, and a QCNN-based classifier, all accelerated by multi-GPU cuQuantum and the PennyLane Lightning backend. Key findings include up to 10x speedups and up to $99.5\%$ test accuracy in phase-transition classification for Hamiltonians such as $H_{TFIM}$ and $H_{XXZ}$, demonstrating effective distribution of computation across quantum and classical resources. The work positions QCQ as a practical quantum-HPC middleware that can scale with advancing quantum hardware, with potential impacts in materials science and condensed matter physics.

Abstract

Achieving high-performance computation on quantum systems presents a formidable challenge that necessitates bridging the capabilities between quantum hardware and classical computing resources. This study introduces an innovative distribution-aware Quantum-Classical-Quantum (QCQ) architecture, which integrates cutting-edge quantum software framework works with high-performance classical computing resources to address challenges in quantum simulation for materials and condensed matter physics. At the heart of this architecture is the seamless integration of VQE algorithms running on QPUs for efficient quantum state preparation, Tensor Network states, and QCNNs for classifying quantum states on classical hardware. For benchmarking quantum simulators, the QCQ architecture utilizes the cuQuantum SDK to leverage multi-GPU acceleration, integrated with PennyLane's Lightning plugin, demonstrating up to tenfold increases in computational speed for complex phase transition classification tasks compared to traditional CPU-based methods. This significant acceleration enables models such as the transverse field Ising and XXZ systems to accurately predict phase transitions with a 99.5% accuracy. The architecture's ability to distribute computation between QPUs and classical resources addresses critical bottlenecks in Quantum-HPC, paving the way for scalable quantum simulation. The QCQ framework embodies a synergistic combination of quantum algorithms, machine learning, and Quantum-HPC capabilities, enhancing its potential to provide transformative insights into the behavior of quantum systems across different scales. As quantum hardware continues to improve, this hybrid distribution-aware framework will play a crucial role in realizing the full potential of quantum computing by seamlessly integrating distributed quantum resources with the state-of-the-art classical computing infrastructure.

Figures (8)

• Figure 1: Pipeline of the Hybrid QSandwich Architecture.
• Figure 2: (a) Quantum circuit depicted as a tensor network with bonds of dimension 2. (b) Checkerboard tensor network circuit. cuQuantum Each green block refers to a two-qubit entangler circuit. (c) The entangled block in the checkerboard circuit
• Figure 3: The quantum circuit used in QCNN. In this circuit, a Hadamard gate is first applied to each qubit, followed by a rotational-Y gate. The angles of the rotational gates are the trainable parameters (Theta1, Theta2 and Theta3). At the end of this circuit, measurement is executed and the possibilities of finding 7 out of 8 quantum states (except "111") are the output of this circuit. Thus, when regarded as a layer in the QCNN, this circuit has 3 input channels and 7 output channels.
• Figure 4: The schematic represents a distributed quantum computing architecture, illustrating the hybrid QCQ framework for classifying phase transitions. This architecture combines CPU-based scheduling with quantum processing across multiple layers and employs GPUs for state encoding. The dashed lines represent omitted $n$ blocks in parallel due to size constraints of the image
• Figure 5: (a)Predicted phases as a function of h for the TFIM model. (b) The predicted probability of phases as a function of Jz for the XXZ model. Positive prediction of label II represents phase II, which is above the dashed lines. The theoretically phase II (disorder phase) is the areas on the right-hand side of the red lines (shown in red color).