The project was a collaboration with Harvard and other researchers using the OpenEvolve tool to search circuit-generating programs for applications in chemistry, materials science, and number theory. In one instance, the AI developed the first reversible implementation of divsteps, a classical algorithm that was previously impractical for reversible arithmetic, by deriving new algebraic identities and formalizing the claim in Lean. These efficiencies reduce the hardware resources needed for quantum algorithms, which could accelerate the viability of quantum computing applications.
