Relaqs

Relqs is a Python framework that uses reinforcement learning to optimize pulse control to implement high-fidelity quantum gates in noiseless and noisy environments.

A key feature is the option to train an agent that generalizes over chaning noise parameters and/or target gates.

For our noise model we use relaxation rates (T1, T2), and detuning which are resampled at the start of each episode. The noise parameters are provided to the liouvillian function from QuTip to simulate the noisy dymanics of Hamiltonian evolution.

Scripts

The scripts folder contains example files to train an RL agent in our environments. Consider them templates that you can run and modify.

Some examples:

• SU(2).py Trains an agent to learn optimal pulses for random SU(2) gates, with the initial and target gates changing each episode. Can also be configured to train on a single fixed gate or on a set of gates (e.g., X, Y, Z, …). By default, inference runs automatically after training.
• two-qubts/single_gate.py Trains an agent on a specified two-qubit gate. By default, it runs I → CNOT (U_initial → U_target), but any valid two-qubit gates can be substituted. Supports SU(4) → SU(4) training where both gates remain fixed across episodes. Switching the environment to "AnalyticalNoisyTwoQubitEnv" outputs analytical performance under the same noise model as the agent. You may also specify a previously trained file path to reuse the exact same SU(4) gates.
• two-qubits/temp_infer.py Runs inference with a trained agent. The set of gates for inference can be modified. Available gates are listed in src/relaqs/api/gates.py

Environments

src/relaqs/environments

Single-qubit gates:

• single_qubit_env.py base noiseless environment for learning a fixed target gate
• noisy_single_qubit_env.py noisy environment for learning fixed target gate
• changing_target_gate.py contains classes for noiseless or noisy evolution with a changing target gate on each episode

Two-qubit gates:

• noisy_two_qubit.py noisy environment to learn a fixed 2-qubit target gate
• two_qubit_changing_gate.py noisy environment to learn changing 2-qubit target gates
• noisy_two_qubit_analytical_env.py noisy environment used to test the analytic solution (not RL) used for a baseline

Install relaqs

pip install -e .

Required Packages:

The require packages should be installed automatically by pip. Legacy install instructions are below:

One may install required packages in a number of ways. Either by manually installing, or by executing one of the commands below:

Direct install:

    pip install -U "ray[rllib]"
    pip install torch
    conda install tensorboardX
    pip install tensorflow-probability
    pip install qutip

Install with pip:

pip install -r requirements.txt

Install in a conda environment (verified on M1 Mac):

If you are using a conda environment running conda env create -n <ENVNAME> --file requirements.yml will install packages using a combination of pip and conda. This is useful on M1 Macs where some packages aren't available through pip.