Learning to Prepare Molecular Ground States with Transformer Models

The paper introduces ADAPT-GQE, a generative-AI pipeline for synthesizing molecular ground-state preparation circuits. It first uses ADAPT-VQE to produce reference circuits, trains models to generate and score circuits, and then applies reinforcement learning to exceed the accuracy of the training targets. The abstract reports an order-of-magnitude reduction in circuit-generation time relative to ADAPT-VQE while maintaining comparable or better state-preparation accuracy. The authors demonstrate the method on imipramine and execute generated circuits on Quantinuum Helios-1, positioning the work as an example of AI-assisted quantum circuit synthesis for quantum chemistry.