In quantum physics, tuning a time-dependent term(controlling protocal) in the Hamiltonian of a system to prepare a desired quantum state is an important task in the context of quantum computing and/or quantum information. A simple example is to get a state with nearly 50% probabilties for the ground state and the first excited state in a quantum well, by tuning the strength of a barrier described by Dirac function. This problem is studied recently in the paper "Deep reinforcement learning for robust quantum optimization"(arXiv:1904.04712), in which several different methods are used. Especially, the authors used deep Q-learning(DQL) as well as deep deterministic policy gradient(DDPG) algorithms to attack the problem. The protocal obtained by these reinforcement learning approches bear a greatly desired property, i.e., they show robustness with respect to certain geometrical randomness(in the position of the Dirac barrier) of the quantum well. Magic!
In this repo, I present the code to solve the eigenstates and useful wave function overlaps of the problem. Besides that, I also worked on the base code of DQL from the first authors' repo and chained these two parts together to get an independent repo to solve the problem.
A basic DQL code is ready. The DDPG code is under development.
Run Main.ipynb only