32-Qubit Emulator Advances Quantum Chemistry Simulations

Olivier Adjoua and colleagues at Sorbonne University, Université Paris-Saclay, and Collège de France have created Hyperion, a new software emulator that addresses limitations in current quantum hardware. The massive parallel, GPU-accelerated emulator overcomes classical memory constraints when simulating strongly correlated quantum chemistry. Hyperion achieves accurate State-Vector ADAPT-VQE simulations for up to 32 qubits and extends emulation to 36 to 40 qubits using a novel partitioned strategy combining sparse State-Vector and Matrix Product State methods. By offering a high-fidelity platform, Hyperion enables the development of new quantum algorithms and more accurate modelling of complex chemical systems, approaching the gold standard of Full Configuration Interaction/Complete Basis Set calculations.

Hyperion emulator enables accurate quantum chemistry simulations with expanded qubit counts

Quantum chemistry emulation now extends to 40 qubits, a four-qubit increase over previous state-of-the-art methods. Modelling beyond 32 qubits with comparable fidelity was previously intractable, surpassing a key barrier limiting the accurate simulation of complex molecular systems. The difficulty arises from the exponential growth of the Hilbert space with increasing qubit number, necessitating immense computational resources and memory capacity. Hyperion utilises a new Sparse Vector-Matrix Product State strategy, partitioning calculations to balance accuracy and computational demand, enabling simulations approaching the exacting Full Configuration Interaction/Complete Basis Set standard. This partitioning intelligently distributes the computational load, allowing for the emulation of larger systems without sacrificing precision. The State-Vector method provides high accuracy for smaller subsystems, while the Matrix Product State representation efficiently handles the entanglement present in larger portions of the molecule.

GPU-acceleration and efficient handling of sparse data sharply reduce resource requirements, facilitating the development of new quantum algorithms and more realistic chemical modelling. The emulator is specifically designed to exploit the parallel processing capabilities of Graphics Processing Units, significantly accelerating matrix-vector multiplications, a core operation in quantum simulations. Sparse data handling is crucial because the matrices describing quantum states are often overwhelmingly filled with zero values; Hyperion’s custom-optimised Sparse Matrix-Sparse Vector (SpMspV) kernels efficiently store and process only the non-zero elements, dramatically reducing memory footprint and computational time. Hydrogen chains served as the demonstration system, chosen for their rapidly expanding computational demands as size increases. These chains provide a well-defined benchmark for assessing the scalability of the emulator, as the number of possible molecular configurations grows rapidly with chain length. Simulations on the Jean Zay supercomputer, utilising NVIDIA Hopper GPUs and high-bandwidth networks, successfully modelled systems up to 40 qubits. The Jean Zay supercomputer provides the necessary computational power and interconnectivity to handle the massive data transfer requirements of these simulations.

ADAPT-VQE analysis revealed a linear scaling in computational cost per iteration for smaller systems, with a performance metric decreasing approximately linearly with qubit number. This indicates that the algorithm is relatively efficient for smaller molecular systems. However, simulations requiring a fixed number of iterations exhibited exponential growth in walltime with increasing qubits and allocated GPUs; 128 GPUs were needed for a 16-qubit hydrogen chain, compared to only 4 for the 14-qubit equivalent. This exponential scaling highlights the fundamental limitations imposed by the size of the Hilbert space. While Hyperion mitigates these limitations through efficient algorithms and hardware acceleration, the underlying exponential complexity remains a significant challenge. Despite advances in sparse representations and symmetry restrictions, the exponential growth of the underlying Hilbert space remains a fundamental limitation, demanding further optimisation of resource allocation and algorithmic efficiency. Symmetry restrictions, for example, reduce the number of configurations that need to be explicitly considered, further reducing computational cost.

Expanding simulation scale aids molecular energy calculations and algorithm validation

Both hardware advances and clever software workarounds are vital for the development of viable quantum computers, and Hyperion represents a strong stride in the latter. The current limitations of near-term quantum hardware, such as qubit coherence times and gate fidelities, necessitate the use of emulators for validating quantum algorithms and exploring potential applications. The discussion is explicitly limited to ADAPT-VQE, raising an important question about its broader utility. ADAPT-VQE (Adaptive Density Functional Theory Variational Quantum Eigensolver) is a specific variational quantum algorithm used to approximate the ground state energy of molecules. It remains to be seen whether Hyperion will prove equally adept at accelerating other variational methods, or if its architecture is tailored to a single approach, potentially hindering its impact on the wider quantum computing field. Other variational methods, such as the Variational Quantum Factoy (VQF), may require different optimisation strategies and data representations.

Strong emulation tools are vital even if their immediate application is narrow, establishing a foundation for broader testing as quantum hardware matures. Hyperion’s ability to accurately simulate up to 40 qubits represents a substantial leap forward, allowing algorithm refinement and performance benchmarking against future quantum devices. This capability is crucial for identifying potential bottlenecks and optimising algorithms before they are deployed on actual quantum hardware. A Sparse Matrix-Sparse Vector strategy reduces resource requirements and opens avenues for more realistic chemical modelling. This capability allows exploration of more complex molecular systems and refinement of understanding of chemical processes, providing valuable insights for future quantum algorithm development. For instance, accurate simulations of catalytic reactions or materials with strong electron correlation could be within reach, potentially leading to the discovery of new materials and technologies. The ability to model these systems with increasing accuracy will be essential for advancing the field of quantum chemistry and materials science. Furthermore, the emulator provides a platform for testing and validating new quantum error mitigation techniques, which are crucial for overcoming the limitations of noisy intermediate-scale quantum (NISQ) devices.

Hyperion, a new massively parallel quantum emulator, successfully simulated quantum systems with up to 40 qubits using a combination of State-Vector and Matrix Product State methods. This achievement matters because it allows researchers to test and refine quantum algorithms for chemistry without needing access to large, and currently limited, quantum computers. The emulator uses optimised calculations to reduce computational demands and maintain accuracy during simulations of complex molecules. The authors demonstrated this capability with the ADAPT-VQE algorithm, and further work will focus on assessing its performance with other variational methods.