AI can now build efficient quantum chemistry algorithms

Quantinuum and Hiverge reported early research showing large language model systems can generate quantum chemistry algorithms that match or outperform human-designed methods while using fewer quantum resources.

The work targets a persistent constraint in quantum computing. Designing practical algorithms for chemistry and materials problems is difficult, and access to high-fidelity processors is limited.

The team used Hiverge’s Hive platform, described as a distributed evolutionary program synthesis system. In the approach, large language models act as mutation operators rather than decision-makers, and the system evolves executable programs across iterations.

Researchers provided a simple sketch of a quantum chemistry algorithm and a target molecule. From that minimal input, Hive produced an algorithm intended for near-term hardware, reaching chemical precision on benchmark systems and cutting circuit depth and gate counts by one to two orders of magnitude compared with standard approaches.

Quantinuum quantum chemistry expert Dr. David Zsolt Manrique said the evolved code used the perturbative method MP2 as a guide. He said it used MP2 to set initial circuit parameters, order excitations, and then refine MP2-inspired heuristics across many iterations.

The initial target problem was computing a molecule’s ground-state energy. The report described how classical methods can become overwhelmed as the number of interacting electron states grows rapidly.

The evolved method, called Hive-ADAPT, was evaluated in classical simulations. Results showed it reproduced energy behavior for molecules including water (H₂O) and beryllium hydride (BeH₂), and reached chemical precision across a wider range of molecular shapes than a baseline method.

Quantinuum said the Hive can produce inspectable code in languages such as Python and integrate libraries including InQuanto. The report also referenced Quantinuum’s Guppy language and described a goal of automated pipelines for designing and benchmarking quantum algorithms.