LLM agent autonomously designs competitive quantum circuits
Researchers from Osaka University introduce a seven-component agentic LLM framework that autonomously designs variational quantum circuits, outperforming established quantum feature maps in image classification and matching chemically inspired ansätze in molecular ground state estimation.
Score breakdown
The framework demonstrates that an LLM-driven agent can replace human-expert circuit design and produce results competitive with — or exceeding — established quantum and classical baselines across both machine learning and quantum chemistry tasks.
- 01Authors: Kenya Sakka, Wataru Mizukami, and Kosuke Mitarai
- 02The framework has seven components: Exploration, Generation, Discussion, Validation, Storage, Evaluation, and Review
- 03The closed-loop workflow combines web-based knowledge acquisition, literature-grounded critique, executable code generation, and experimental feedback
Kenya Sakka, Wataru Mizukami, and Kosuke Mitarai present an autonomous agentic framework designed to remove the dependence on human expertise in quantum circuit design. The system integrates seven named components — Exploration, Generation, Discussion, Validation, Storage, Evaluation, and Review — into a closed-loop pipeline. This pipeline combines web-based knowledge acquisition, literature-grounded critique, executable code generation, and experimental feedback to iteratively refine circuit designs under explicit constraints.
The framework is evaluated on two distinct tasks: quantum feature map construction for quantum machine learning and ansatz generation for variational quantum eigensolver (VQE) applications in quantum chemistry.
The framework is evaluated on two distinct tasks: quantum feature map construction for quantum machine learning and ansatz generation for variational quantum eigensolver (VQE) applications in quantum chemistry. On image classification benchmarks, the best feature map produced by the system outperforms representative quantum feature maps and, when scaled to larger qubit counts, surpasses the classical radial basis function (RBF) kernel. In molecular ground state estimation tested across seven molecules, the generated ansätze attain competitive accuracy with both widely used chemically inspired constructions and hardware-efficient constructions, while satisfying the imposed scaling constraints.
The authors argue these results establish LLM-driven agentic systems as a viable paradigm for automated quantum circuit design, and illustrate how AI systems can participate in iterative scientific optimization workflows across scientific domains.
Key facts
- 01Authors: Kenya Sakka, Wataru Mizukami, and Kosuke Mitarai
- 02The framework has seven components: Exploration, Generation, Discussion, Validation, Storage, Evaluation, and Review
- 03The closed-loop workflow combines web-based knowledge acquisition, literature-grounded critique, executable code generation, and experimental feedback
- 04Evaluated on two tasks: quantum feature map construction for quantum machine learning and ansatz generation for VQE in quantum chemistry
- 05The best generated feature map outperforms representative quantum feature maps and surpasses the classical radial basis function kernel at larger qubit counts
- 06Generated ansätze achieve competitive accuracy with chemically inspired and hardware-efficient constructions across seven molecules
- 07The system operates under explicit design constraints, including scaling constraints
Topics
Summary and scoring are generated automatically from the original article. We always link back to the publisher and never republish images or paywalled content. Last processed Jun 12, 2026 · 10:05 UTC. How this works →