Skip to content

Utilization of Reinforcement Learning in Variational Quantum State Diagonalization.

Notifications You must be signed in to change notification settings

iitis/RL_for_VQSD_ansatz_optimization

Repository files navigation

DOI

Code for "Enhancing variational quantum state diagonalization using reinforcement learning techniques"


This is a repository for the code used to calculate the numerical results presented in the manuscript:

Akash Kundu, Przemysław Bedełek, Mateusz Ostaszewski, Onur Danaci, Yash J. Patel, Vedran Dunjko, Jarosław A. Miszczak, Enhancing variational quantum state diagonalization using reinforcement learning techniques, arXiv:2306.11086 (2023).

@Misc{kundu2023enhancing,
  author    = {Kundu, Akash and Bedełek, Przemysław and Ostaszewski, Mateusz and Danaci, Onur and Patel, Yash J. and Dunjko, Vedran and Miszczak, Jarosław A.},
  title     = {Enhancing variational quantum state diagonalization using reinforcement learning techniques},
  year      = {2023},
  copyright = {Creative Commons Attribution 4.0 International},
  doi       = {10.48550/ARXIV.2306.11086},
  keywords  = {Quantum Physics (quant-ph), Artificial Intelligence (cs.AI), Machine Learning (cs.LG), FOS: Physical sciences, FOS: Physical sciences, FOS: Computer and information sciences, FOS: Computer and information sciences},
  publisher = {arXiv},
}

Software Installation

The code was used on Ubuntu GNU/Linux 22.04.

For this project, we use Anaconda which can be downloaded from https://www.anaconda.com/products/individual.

To install activate the environment please do the following:

conda env create -f rl-vqsd.yml
conda activate rl-vqsd 

Alternatively, you could run

conda create -n rl-vqsd python=3.8.5
conda activate rl-vqsd
pip install qiskit==0.31.0 qiskit-aqua
pip install torch

Additionally, for running Jupyter notebooks you should also install

pip install jupyter
pip install mpl-axes-aligner

and Conda environment with all packages can be created using

conda env create -f rl-vqsd-full.yml

How to generate quantum state to diagonalize?

The utils.py python script contains two important functions random_state_gen(...) and ground_state_reduced_heisenberg_model(...) corresponding to the generation of (1) An arbitrary quantum state sampled from the Haar measure. (2) The reduced ground state of the Heisenberg model. You just need to run

python utils.py --state state_type --max_dim maxdim --seed seedno

To generate (1) : state_type (str) is replaced by mixed, maxdim (int) can be any upper limit to the size of the quantum state to be produced, and seedno (int) specifies the seed for the quantum state. Example

python utils.py --state mixed --max_dim 4 --seed 1

To generate (2) : state_typ (str) is replaced by reduced-heisenberg, maxdim (int) is either 3 or 4 and seedno (int) is a redundant variable. Example

python utils.py --state reduced-heisenberg --max_dim 3 --seed 342425

How to run RL-VQSD?

To diagonalize a quantum state we can just run the main.py python script using the following line of code:

python main.py --seed seedagent --config config_file --experiment_name "global_COBYLA/"

In the above, the seedagent (int) corresponds to the different initialization to the Neural Network (NN) and the config_file (str) is the configuration corresponding to the state that need to be diagonalized, the hyperparameters of the NN and the agent configuration and it looks like: h_s_2_rank_4_1 where 2 corresponds to the number of qubit of the state, 4 is the rank of the state and 1 is the seed used to generate the state. Example:

python main.py --seed 1 --config h_s_2_rank_4_1 --experiment_name "global_COBYLA/"

All the possible configurations can be found in the folder configuration_files. To run the reduced Heisenberg model:

python main.py --seed 102 --config h_s_3_reduced_heisenberg --experiment_name "global_COBYLA/"

Diagonalizing using random search:

python main.py --seed 100 --config h_s_3_reduced_heisenberg --experiment_name "random_search/"

How to reproduce the results?

The results of the above will be saved in the results folder. The reproduce the 2-qubit eigenvalue convergence (Fig. 6a in article) : You can run one of the jupyter notebooks titled eigenvalue_analysis.ipynb and run each cell. Similarly, The reproduce the 3-qubit eigenvalue convergence (Fig. 9 in article) : You can run one of the jupyter notebooks titled eigenvalue_analysis_reduced_heisenberg.ipynb and run each cell. Constant structure RL-ansatz statistics (Fig. 8 in article) : You just need to run the constant_structure_VQSD.ipynb to first load the RL-ansatz of your choice and then use this ansatz to diagonalize N arbitrary quantum states of same dimension. This is utilized to plot in the last couple of cells of the notebook. The reproduce the 2-qubit eigenvalue error (Fig. 6b in article) : You first need to produce the diagonalization results using Layered Hardware Efficient Ansatz (LHEA) which can be done using and by running the LHEA_VQSD.ipynb file. Then utilizing the LHEA results in LHEA_plot_analysis.ipynb we produce Figure 6b. Comparison with random search (Fig. 12a, 12b and 13) Both the plots for 2 and 3 qubits to compare the DDQN with random search can be generated just by running the plot_analysis_random_search.ipynb file.

About

Utilization of Reinforcement Learning in Variational Quantum State Diagonalization.

Resources

Stars

Watchers

Forks

Packages

No packages published

Languages