QOBLIB: tracking progress in quantum optimization

IBM Research is trying to make quantum optimization less of a slogan and more of a scorecard. The company says its Quantum Optimization Benchmarking Library, or QOBLIB, now has a foundational paper in Nature Computational Science, a new website, and more than 2,000 submitted results from researchers across quantum and classical optimization.

That matters because the field has already crossed one big threshold: IBM and partners recently announced three experiments showing clear quantum advantage. The harder question is what to do with that advantage. Optimization is one of the main places researchers are looking, even though it was not the focus of those recent demonstrations.

QOBLIB is IBM’s answer to that problem. The library is an open-source, community-driven set of ten difficult optimization problem classes, along with the metrics, baselines, and tools needed to compare quantum and classical approaches on equal footing. IBM says the ten classes were chosen because they become hard for leading classical solvers at relatively small sizes, while still being within reach of near-term quantum hardware.

The new website turns that work into something more public and easier to use. It tracks more than 1,200 instances, with 500-plus already solved to optimality, and shows the best-known result for each one, who achieved it, and their affiliation. A Submission Builder walks contributors through the process and exports files ready for a pull request. The point is to lower the friction, and to make every result visible.

And the baseline keeps moving. IBM says the largest solved market split problem instances in the repository have grown from roughly 60 to 110 variables since launch. That is a larger hill for quantum methods to climb, but also a cleaner test of any future claim that they can really do something classical methods cannot.