Quantinuum uses AI agents to discover new quantum optimization algorithms

Quantinuum has partnered with Hiverge, Amazon Web Services and NVIDIA to explore the use of artificial intelligence for discovering algorithms that address combinatorial optimization problems. The work builds on an earlier collaboration with Hiverge that applied AI to quantum chemistry algorithms. The current effort examines how large language models can accelerate quantum algorithm research. A system called the Hive uses a collective of large language model agents that edit the same codebase. The agents include Gemini, ChatGPT, Claude, Llama and NVIDIA Nemotron accessed through AWS Bedrock. The Hive takes a researcher sketch of an algorithm along with stated constraints and evolves improved versions. Output appears in familiar languages such as Guppy or NVIDIA CUDA Q. The process requires no new model training. Each candidate algorithm runs and receives a fitness score based on how well it solves the problem. Stronger versions advance while weaker ones are discarded. The cycle repeats until researchers select the fittest result for further testing on additional problem instances. The project targets heuristic quantum optimization algorithms. These aim to improve efficiency in areas such as power grid dispatch and storage, nuclear fuel arrangement and molecular design for drug, material and chemical discovery. The algorithms are produced as readable code rather than being embedded inside a neural network. This format supports easier interpretation and deployment on new cases. The team relied on the NVIDIA CUDA Q platform running on graphics processing units supplied by AWS. Researchers first used this classical accelerated computing setup to simulate the algorithms and evaluate their fitness before any use on quantum hardware.