Facilitating usage of quantum–classical optimization software with QHyper library

Abstract

Context:

Quantum computing offers promising approaches to solving combinatorial optimization problems. While there are many software frameworks for these approaches, they are developed by different providers, requiring users to reimplement their problems separately in each environment to facilitate comparisons. Additionally, due to the current limitations of quantum hardware, classical solvers are often used as reference methods, further increasing implementation overhead. This lack of integration highlights the need for a unified API that seamlessly connects all approaches, enabling efficient problem formulation and comparison of different methods.

Objective:

This work aims to provide a research tool with a unified interface for specifying combinatorial optimization problems, selecting different solvers, managing problem hyperparameters, and standardizing the output for effortless analysis and comparison.

Methods:

We have developed an open-source research tool with a modular architecture that fulfills the aforementioned objective. QHyper’s design provides an independent representation of the problem along with a converter that supports various formulations compatible with available solvers. Among them are the Quantum Approximate Optimization Algorithm implemented in PennyLane, the D-Wave Advantage solver, and the Gurobi optimizer. For optimization, QHyper offers various methods, from local techniques based on gradient descent to custom global Monte Carlo methods for hyperparameter optimization.

Results:

QHyper’s overhead is within the statistical margin of error compared to the stand-alone use of solvers. The included illustrative example shows how to create a custom problem and easily switch between the chosen solvers. The paper also contains references to real-life scientific use cases for the presented tool.

Conclusion:

QHyper’s design ensures easy extensibility to new problems, solvers, and optimizers, and has proven its usefulness for several scientific use cases mentioned in the paper. With simple configuration options and easy management, such as through Jupyter Notebooks, this library can be useful for practitioners, engineers, and academics working on combinatorial optimization research.