IAGQ

Master thesis: Implementation and analysis of gradient-computation in quantum-algorithms

Installation

For this tool poetry is used as a python packaging tool.

Install poetry and run:

poetry install

in the main folder to create a virtual environment including all required python libraries.

Scripts

plots_script.py

• Generates all plots used in the thesis from scratch. Takes some time!

Experiment scripts

There are also four scripts provided to run different experiments.

gradient_sample_shotslist.py

• Computes error of gradient-approximations with forward, backward and central finite differences, SPSA as well as general and standard parameter-shift for a given list of shots run on a simulator.

gradient_variance_shotslist.py

• Computes expected error of gradient-approximations with forward, backward and central finite differences, SPSA as well as general and standard parameter-shift for a given list of shots by computing the exact one-shot variance of all methods.

FD_variance_shotslist.py

• Computes expected error of gradient-approximations central finite differences with different shift values for a given list of shots by computing the exact one-shot variance of all methods.

• Use the argument -cd_h to specify distances $h$ to be used in the experiments, for example: -cd_h 0.5 0.1 0.001

GPS_variance_shotslist.py

• Computes expected error of gradient-approximations general parameter-shift with different shift values as well as standard parameter-shift for a given list of shots by computing the exact one-shot variance of all methods.

• Use the argument -g to specify distances $\gamma$ to be used in the experiments, for example: -g 0.5 0.1 0.001

General Arguments

For the first two scripts (that run all the methods that were used in the experiments of the thesis):

• Use the --no-fd, --no-bd,--no-cd,--no-spsa,--no-ps, --no-gps flags if you want to not compute the gradient with a certain method

• Use --fd_h, --bd_h, --cd_h, --gps_gamma, --spsa_h,--spsa_count to specify hyperparameters for the methods

For all four experiment scripts:

• Use --seed to specify the numpy random seed

• Use --shots_start, --shots_stop, --shots_step to specify the list of shotcounts per measurement to use for the approximations. In gradient_sample_shotslist.py big shotcounts can rapidly increase the runtime!

• Use --ansatz to specify the ansatz. Possible values: ESU2 (Qiskits EfficientSU2 Ansatz), CRE (Cross Resonance + Euler Rotations Ansatz), LHE (Layered Hardware Efficient Ansatz)

• Use --observable to specify the observable. Possible values: Diagonal (Diagonal Observable with values $2i$ on main diagonal), LiH (LiH VQE Observable), HeH (HeH VQE Observable), H2 (H2 VQE Observable)

• Use --metric to specify the metric used to compute the gradient errors. Possible values: MSE (Mean squared error), MAE (Mean absolute error)

How to use:

Run:

poetry run .\src\scripts\<scriptname>

for the script of your choice, starting in the main folder.

The scripts will generate plots in the plots directory. Also you can look at all the plots and parameters of the experiments in the mlflow UI.

Run:

poetry run mlflow ui

Then open the displayed link (normally http://127.0.0.1:5000) in your browser.