Discovering formally verified quantum programs with AI
The authors report a 290-fold improvement over prior quantum-arithmetic baselines, with correctness and cost verified in Lean.
Their system uses language-model agents to build quantum programs against formal problem specifications. Mathematical proofs let them verify programs beyond the small or restricted cases classical computers can evaluate. The paper also reports improvements of up to 1.4× for parallel search and 11×, 8× and 2.5× for simulations relevant to fusion, photochemistry and drug metabolism, respectively. These are algorithmic results, rather than measured speedups on quantum hardware.
ArXiv · AI/CL/LG's note
Their system uses language-model agents to build quantum programs against formal problem specifications. Mathematical proofs let them verify programs beyond the small or restricted cases classical computers can evaluate. The paper also reports improvements of up to 1.4× for parallel search and 11×, 8× and 2.5× for simulations relevant to fusion, photochemistry and drug metabolism, respectively. These are algorithmic results, rather than measured speedups on quantum hardware.
ArXiv · AI/CL/LG's note
score 6