Representing quantum circuits as hypergraphs, which collapse commuting operations into single units and preserve qubit connections, sharply accelerates optimisation processes. The new method enables template matching, a key technique for simplifying complex calculations, to focus searches locally within the circuit instead of revisiting every gate; this maintains vital information regarding interactions between circuit components. A new instruction format for quantum computers sharply speeds up template matching, reducing complex calculations by finding and replacing circuit fragments with simpler equivalents.

The approach uses “hypergraphs” that combine operations executable in any order while accurately mapping qubit connections. Optimising quantum algorithms requires simplification before execution on hardware, often achieved through template matching, identifying and substituting common patterns with equivalent simplifications. Researchers have developed a representation using these “hypergraphs”, combining operations regardless of order but maintaining accurate qubit mappings.

Unlike existing methods based on directed acyclic graphs (DAGs), which show calculation steps without efficiently representing simultaneous actions, hypergraphs function like LEGO bricks with multiple attachment points allowing for more flexible representation. This preserves vital information about circuit interactions, enabling faster optimisation by focusing searches locally than globally. But will preserving locality truly unlock scalable quantum computation?

Hypergraphs unlock scalable quantum circuit optimisation via preserved gate commutativity

A new hypergraph matcher delivers speedups ranging from 11x to 606x over Qiskit’s directed acyclic graph (DAG)-based implementation for both random and arithmetic benchmarks. This performance leap overcomes limitations that previously hindered the optimisation of larger circuits. Previous state-of-the-art methods struggled with increasing circuit size because they discarded connections between gates, ignoring their ability to be performed in any order, a process known as commutativity, which forced exhaustive global searches during simplification.

Performance gains consistently reached up to 606x when working with circuits containing more than one hundred gates compared to Qiskit’s existing DAG method. Retaining connectivity information, previously discarded by DAG approaches, allows matches to extend within immediate circuit neighbourhoods instead of requiring exhaustive global searches.

Maintaining equivalent optimisation quality while utilising arithmetic operations demonstrated speedups ranging from 11x to 606x; the acceleration increases alongside growing circuit complexity, suggesting a pathway towards optimising larger quantum programs. However, these figures currently reflect benchmark results and do not yet demonstrate performance on real quantum hardware with its inherent noise and limitations.

Hypergraph optimisation shows promise but requires wider validation with varied quantum circuits

Demonstrably accelerating optimisation using random and arithmetic benchmarks, achieving speedups up to six hundred times over existing methods, Quantum Technology scientists acknowledge an important gap in their validation: they haven’t tested this representation’s performance across diverse circuit types beyond those specific examples. Will the benefits hold true when applied to more complex algorithms or designs tailored for particular quantum applications? This raises questions about generalizability. The team will next focus on expanding testing to include a broader range of circuit designs, beginning with error mitigation techniques.

Intelligent grouping preserves crucial connections between qubits during optimisation, allowing template matching to concentrate searches locally rather than globally.

This research introduced a hypergraph representation for optimizing quantum circuits and demonstrated its ability to simplify these circuits faster than existing methods. By intelligently collapsing commuting gates while retaining local connectivity, the new approach achieved speed-ups ranging from 11x to 606x compared to a directed acyclic graph implementation in Qiskit. This means larger and more complex quantum programs can potentially be optimized more efficiently. The authors plan to expand testing of this method with additional circuit designs, including those used for error mitigation techniques.

👉 More information
🗞 Fast Template Matching for Quantum Circuits Using Hypergraph
✍️ Xin Hong, Runhong He, Shenggang Ying and Mingsheng Ying (Chinese Academy of Sciences); Jintao Yu (Arclight Quantum Computing Inc); Sanjiang Li (University of Technology Sydney)
🧠 DOI: https://doi.org/10.1145/3770743.3804087

More like this