Clifford-informed initialization and loss landscape exploration for Variational Quantum Eigensolvers (VQEs)

This project investigates how the algebraic structure of Clifford circuits can be leveraged to improve the optimisation of VQEs on near-term quantum hardware. Loss landscape analysis, stabiliser simulation insights, and practical warm-start strategies are demonstrated.

• Baseline VQE implementation with IBM Qiskit
• Loss landscape exploration via Monte Carlo sampling, and neural network approximation.
• Clifford circuit identification, and runtime analysis (using Stim for stabiliser simulation).
• Warm-start optimisation from promising Clifford configurations.
• Comparison against conventional random initialisation across multiple ansatze and Hamiltonians.

• Clifford parameter configurations often provide competitve (sometimes superior) starting points compared to random initialisation.
• Significant simulation speed-ups are possible when the number of non-Clifford rotations remains low.
• Landscape reconstruction methods can yield useful insights even with sparse sampling (but this is often circuit specific, and requires training data from the real version).

To view this notebook, begin by installing the uv package manager via the following command.

pip install uv

Then, proceed to run the notebook, and install all of its dependencies by using the following.

uv run main.ipynb