Unitary gate design with time-independent Hamiltonians

GND package

The package is in the folder gnd, see main.py for an example.

Target gates are defined as classes in the configs.py. The __str__ method returns the name of the folder saved under data and __dir__ gives the unique variables in the file name. The config also contains information such as the number of qubits.

The Pauli word basis is given by gnd.basis.PauliBasis with the number of qubits. Parameters can be projected onto the one- and two-body projection with the basis method .two_body_projection(parameters).

The optimizer is called from gnd.optimize, and requires a target unitary, basis, and some initial parameters.

The data from the optimizer is then stored in a gnd.data.OptimizationData class, which handles the saving and loading of data. The data stored is the "steps", "parameters", "fidelities", and "step_sizes". The data class takes a list of optimizers and stores them as a CSV file in the correct folder defined by the config. Data can also be loaded using the data class and the correct config.

Benchmarks

As comparison we provide two benchmarks,

1. First, we compare the with the work by Innocenti (2020):

Luca Innocenti, Leonardo Banchi, Alessandro Ferraro, Sougato Bose, Mauro Paternostro (2020). "Supervised learning of time-independent Hamiltonians for gate design". New J. Phys. 22 065001 (arXiv:1803.07119).

who kindly provided code at https://github.com/lucainnocenti/quantum-gate-learning-1803.07119 that we adapted for ou purposes. See the folder innocenti for details.

2. In addition to comparing with a supervised learning method, we compare with a naive gradient descent method in the Lie algebra. See the folder jax_gd for details.