Consistent Problem Definition Enables Portable Quantum Simulations Across HPC Systems

Variational quantum algorithms represent a promising pathway for utilising near-term quantum computers, but comparing their performance across different hardware and software platforms remains a significant challenge. Marco, alongside colleagues at their respective institutions, addresses this issue by developing a toolchain that consistently defines and ports quantum problems between various high-performance computing (HPC) systems and simulators. The team investigated three key applications, ground state calculations for the hydrogen molecule, the MaxCut problem, and the travelling salesman problem, to assess how runtime environment impacts simulation scalability and the reliability of physical results. This work demonstrates successful translation of problem definitions between simulators, while also revealing inherent limitations in scaling due to runtime and memory constraints, offering valuable insights for optimising future quantum simulations and exploring HPC performance

Quantum Software Benchmarking for Circuit Simulation

This extensive document details a comprehensive benchmarking study of various quantum computing simulation software packages, evaluating their performance on classical hardware when simulating quantum circuits. The study concentrates on state vector simulators, fundamental for verifying and debugging quantum algorithms, and assesses their performance on diverse hardware configurations, including CPUs and GPUs. The researchers benchmarked popular quantum software packages like Qiskit, PennyLane, Cirq, myQLM, and CUDA Quantum, using a diverse set of quantum circuits, including random circuits and those from algorithms like QAOA and VQE. Key metrics measured included execution time, memory usage, and scalability, assessing performance as circuit size increases.

Containerization and standardized software stacks were used to ensure reproducibility. The results show substantial performance differences between simulators, with GPU-based simulators generally outperforming CPU-based ones for larger circuits. Memory usage is a major limitation due to the exponential scaling of state vector simulation with the number of qubits, and the efficiency of underlying code significantly impacts performance. This research provides a valuable resource for quantum computing researchers and developers, offering a detailed comparison of simulation tools and highlighting key challenges and opportunities in the field.

Standardized Intermediate Representations for Quantum Simulations

Researchers developed a methodology to ensure fair comparisons of quantum computing simulations across diverse hardware and software platforms. Recognizing that subtle differences in defining problems, such as the Hamiltonian and ansatz, can skew results, they prioritized a standardized approach where the same computational task is enacted identically on each platform. This methodology uses intermediate representations (IRs) for both the Hamiltonian and ansatz. The Hamiltonian is translated into a concise list of numerical values, creating a platform-independent description of the quantum system’s energy, while the ansatz is expressed using the established OpenQASM v2.

These IRs act as a bridge, allowing the team to parse and translate problem definitions into the format required by each simulator. This approach addresses a critical challenge in quantum computing research: the lack of standardization. The team discovered inconsistencies even in seemingly identical simulations, such as variations in how molecular Hamiltonians are calculated. By using these IRs, they ensured each simulator operated on the same fundamental problem definition, enabling a truly comparative analysis. The software developed automatically generates the necessary code to run simulations, streamlining the process and ensuring consistency.

Seamless VQA Comparisons Across Platforms Demonstrated

Researchers have developed a toolchain to streamline the comparison of variational quantum algorithms (VQAs) across different simulators and high-performance computing (HPC) systems. This addresses the challenge of consistent comparisons due to the many choices involved in defining these algorithms, including the Hamiltonian, circuit structure, and optimization method. The team’s approach uses a parser to translate problem definitions seamlessly between simulators, ensuring comparable results. The toolchain was tested using three representative applications relevant to near-term quantum devices: calculating the ground state energy of a hydrogen molecule, solving the MaxCut optimization problem, and tackling the Traveling Salesman Problem.

By focusing on state vector simulators, the team avoided complexities of circuit compilation and transpilation, concentrating on the core performance of the algorithms. The results demonstrate the successful translation of problem definitions across different simulation packages, paving the way for more reliable benchmarking and performance analysis. The study also investigated the impact of deployment strategies on simulation performance, comparing runs on bare-metal installations with those within containerized environments.

Standardised Quantum Simulation and Problem Translation

This research presents a toolchain designed to facilitate the consistent translation of problem definitions, specifically Hamiltonians and ansatzes, across different quantum simulation platforms. This addresses the challenge of comparing results obtained using different simulators, which can be skewed by variations in how problems are defined and implemented. The toolchain aims to provide a standardized framework for evaluating the performance of quantum algorithms and identifying the most promising approaches for implementation on near-term quantum hardware. The toolchain was tested using a range of relevant use cases, including the calculation of ground state energies, optimization problems, and combinatorial challenges.

By focusing on state vector simulators, the researchers were able to isolate the core performance of the algorithms themselves, avoiding the complexities of circuit compilation and transpilation. The results demonstrate the successful translation of problem definitions across different simulation packages, paving the way for more reliable benchmarking and performance analysis. The study also investigated the impact of deployment strategies on simulation performance, comparing runs on bare-metal installations with those within containerized environments.