The roadmap towards AI-assisted pulse programming for solid-state NMR

Solid-state NMR (ssNMR) relies on sophisticated pulse sequences to obtain information on spin systems in solids or semi-solids. Analytical derivations based on average-Hamiltonian theory and numerical approaches based on gradient-based optimal control (OC) have greatly helped the development of ssNMR pulse sequences. However, these conventional strategies face intrinsic limitations when multiple strong interactions or hardware constraints are present. In this review, we discuss some emerging alternative approaches for ssNMR pulse sequence design, based on artificial intelligence (AI) that include: evolutionary algorithms, deep learning and reinforcement learning methods. We first highlight advantages and limitations of the OC methods, highlighting where they succeed and where they fail, then we discuss the potential of AI algorithms in overcoming key bottlenecks that are affecting the ssNMR pulse-sequence design.