seemps 2.9.8

Matrix Product State library for quantum simulation and numerical analysis

SeeMPS

Introduction

SEEMPS is the second iteration of the SElf-Explaining Matrix-Product-State library.

The original library, still available here was a collection of Jupyter notebooks with a well documented implementation of matrix-product state algorithms.

The current iteration aims to be more useable and have better and more standard documentation, while preserving the same degree of accessibility of the algorithms.

Intended audience

The library is thought out as introduction to the world of Matrix Product States and DMRG-inspired algorithms. Its main goal is not performance, but rapid prototyping and testing of ideas, providing a good playground before dwelling in more advanced (C++, Julia) versions of the algorithms.

This said, the library as it stands has been used in some heavy-duty simulations involving tens and hundreds of qubits, and, in particular, its current iteration arises from the following works on quantum-inspired algorithms for numerical analysis:

• Quantum-inspired algorithms for multivariate analysis: from interpolation to partial differential equations, Juan José García-Ripoll, Quantum 5, 431 (2021), https://doi.org/10.22331/q-2021-04-15-431

• Global optimization of MPS in quantum-inspired numerical analysis, Paula García-Molina, Luca Tagliacozzo, Juan José García-Ripoll, https://arxiv.org/abs/2303.09430

• Chebyshev approximation and composition of functions in matrix product states for quantum-inspired numerical analysis, Juan José Rodríguez-Aldavero, Paula García-Molina, Luca Tagliacozzo, Juan José García-Ripoll https://arxiv.org/abs/2407.09609

• Pseudospectral method for solving PDEs using matrix product states, Jorge Gidi, Paula García-Molina, Luca Tagliacozzo, Juan José García-Ripoll https://arxiv.org/abs/2409.02916

Usage

The library is developed in a mixture of Python 3 and Cython, with the support of Numpy, Scipy and h5py. Installation instructions are provided in the documentation.

Authors:

• Juan José García Ripoll (Institute of Fundamental Physics)
• Paula García Molina (Institute of Fundamental Physics)
• Juan José Rodríguez Aldavero (Institute of Fundamental Physics)

Contributors:

• Jorge Gidi

Development

Environment

For optimal development the following is expected:

• uv from Astral is installed
• In Linux, if you wish to use a local version of Python, you might need to install the python-devel package or equivalent one. This also installs a C and C++ compilers.
• In Windows, you need to install a Visual Studio C++ (Community Edition) compiler to build SeeMPS.
• A copy of Visual Code with the Python extensions installed plus some additional recommended extensions:
  • Coverage Gutters
  • Mypy Type Checker
  • Ruff support

The environment is bootstrapped using

uv sync --dev

This installs both the SeeMPS library and libraries that it depends on, plus additional tools that are used for development:

• ruff, for code linting
• mypy and basedpyright, for type checking
• coverage, for code coverage

On top of this, please use

uv run scripts/make.py --install-hooks

to ensure type checkers and other tests are run before committing changes with git.

Testing

The library contains a rather complete set of unittests under the tests/ folder. The tests can be run using the standard unittest module, as in

uv run python -m unittest -v

The code coverage of the test suite exceeds 88%. To analyze test coverage you can open a terminal and run

uv run coverage run -m unittest -v && uv run coverage report

Alternatively, you can use

uv run coverage lcov

to create a coverage file that is interpreted by the "Coverage Gutters" Visual Code extension. There is a task (right-button option in the explorer) with the name "Run Tests with Coverage" that both runs the tests and automatically creates the reports using

uv run coverage run -m unittest -v && uv run coverage lcov

TODOs

• Update documentation.
• Many functions are declared to accept Interval, when they actually can only use RegularInterval or ChebyshevInterval