What's the best way to implement high-fidelity unitary operators on a quantum system? Given a target effective Hamiltonian, what are good ways to achieve the target Hamiltonian (taking into account interactions between terms of the Hamiltonian, experimental constraints, and errors)? How do you even know if the operation is high-fidelity? These are some of the questions we're trying to answer.
This repository contains Python code to simulate unitary dynamics for a spin system in a magnetic field (see spin_simulation), as well as a mess of code to apply reinforcement learning algorithms to the Hamiltonian engineering problem.
Running run_alpha_zero saves a job directory with tensorboard data and network parameters saved periodically during training. To see the tensorboard data, run tensorboard --logdir . in the directory containing the job files. Also, candidate pulse sequences are written to standard output, which is saved in files that are read later.
To view the tensorboard output, run tensorboard --logdir . in the directory with output. It should launch a browser window with time series of loss functions, and distributions of rewards.
My ad-hoc process for pulling pulse sequences uses Get-Candidate-Sequences (in rl_pulse/eval/). It has more detail there.
Using Sphinx, sphinx-apidocs, and autodoc to create API docs from docstrings. I need to make sure I've fully documented the code though... And find a place to put the documentation once it's built.
To generate rst files with sphinx-apidocs, run sphinx-apidoc -o docs/ . from the root directory.
If you use this code or any ideas presented in the repository, please cite it as
@misc{RLPulse,
title = {{RL-Pulse}: A library for Hamiltonian Engineering using Reinforcement Learning},
author = "{Will Kaufman and Chandrasekhar Ramanathan}",
howpublished = {\url{https://github.com/wjkaufman/rl_pulse}},
url = "https://github.com/wjkaufman/rl_pulse",
year = 2020,
note = "[Online; accessed 01-May-2020]"
}