Machine Learning for Quantum Chemistry

This repository contains the codebase for a series of research papers exploring the intersection of Machine Learning, Quantum Chemistry, and Quantum Algorithms The work is organized into multiple distinct projects, each corresponding to a specific publication.

Projects and Publications

1. Learning Minimal Representations of Fermionic Ground States

This project introduces an unsupervised framework to discover the minimal degrees of freedom in quantum many-body ground states.

• Optimal Compression: Utilizes an autoencoder architecture to identify a sharp reconstruction threshold at $L-1$ latent dimensions for $L$-site Fermi-Hubbard models, matching the system's intrinsic physical dimensions.
• Interpretable Representations: Analyzes the Jacobian of the encoder to quantify the relative importance of physical features like density, nearest-neighbor correlations, and on-site interactions.
• Variational Ansatz: Employs the trained decoder as a differentiable manifold for direct energy minimization, bypassing the $N$-representability problem by restricting optimization to physically valid states.

👉 Explore the Code & Data

2. Learning density functionals from noisy quantum data

This project focuses on the robustness of learning density functionals when trained on noisy data generated by near-term quantum algorithms.

• Noise Robustness: Successfully generalizes from small, noisy datasets typical of NISQ algorithms, demonstrating that the learning procedure can effectively filter out unbiased sampling noise.

• Bias Replication: Shows that while sampling noise is filtered, ML models trained on data with VQE-style expressibility or optimization errors tend to replicate those underlying biases.

• Kohn-Sham Optimization: Applies trained functionals to new problem instances via an automatically differentiable Kohn-Sham-like density optimization scheme.

👉 Explore the Code & Data

Repository Structure

The code is structured as follows:

• ./src/dftqml: Core Python package (functionals, architectures).
• ./compression-main: Scripts for the Fermionic Ground States paper.
• ./noisy-dft-main: Scripts for the Noisy DFT paper.

Getting Started

Installation

git clone git@github.com:StefanoPolla/DFTQML.git
cd DFTQML
pip install .

If you want to train or evaluate models, you will additionally need to install TensorFlow (used in noisy-dft-main) or PyTorch (used in compression-main). Tensorflow can be installed by pip install tensorflow, while for pytorch check the website (https://pytorch.org/get-started/locally).