Abstract: Hamiltonian engineering of quantum many-body systems is central to many quantum simulation and sensing protocols. This technique uses a set of operations to generate a target Hamiltonian with which to apply to a system. Specifically, in Nuclear Magnetic Resonance (NMR) spectroscopy of solids, magnetic dipolar interactions between the spins of the protons causes the signal we receive to rapidly decay without the implementation of Hamiltonian engineering. We used Hamiltonian engineering to design sequences of magnetic pulses to decouple these dipolar interactions to increase sample’s coherence time. Historically, Average Hamiltonian Theory (AHT) has been the methodology used to both design these pulses sequences as well as to characterize their sensitivity to error. Recently, reinforcement learning (RL) has emerged as another avenue to engineer Hamiltonians. RL is a type of machine learning and treats the system’s

dynamics as a black box but has the potential to provide robustness to experimen-tal imperfections. However, unconstrained RL algorithms have not outperformed conventional methods to date. In this thesis, we used theoretical insights into the quantum dynamics of the interacting spin systems to constrain the action space of the RL algorithm, while also tweaking the RL algorithm itself. For the problem of decoupling magnetic dipolar interactions in solid-state spin systems, this approach allowed us to design multiple pulse sequences which have fidelities at

a higher level than other sequences designed either with AHT or other RL algorithms.

Join Zoom Meeting
https://dartmouth.zoom.us/j/97104796822?pwd=U2lvYllEVmRNYVZOZXdRWjBpa0RGQT09

Meeting ID: 971 0479 6822
Passcode: Eskandari