Development history and contributions

Development history and contributions#

This section summarizes the development history of the sparse-intermediate representation (sparse-IR) ecosystem and acknowledges the main contributors.

Predecessor: irbasis#

Before sparse-ir and SparseIR.jl, the irbasis library [Chikano et al., 2018] provided precomputed, tabulated intermediate-representation (IR) basis functions. In irbasis, the IR basis functions were obtained by solving the integral equation for each choice of the cutoff parameter \(\Lambda = \beta \omega_\mathrm{max}\) with high precision, and the results were stored in a database. Users could then load these tabulated basis functions and use them directly in their calculations.

The sparse-ir family of libraries generalizes this idea by computing the basis functions on the fly for arbitrary \(\Lambda\), without relying on a fixed precomputed database.

First implementation of on-the-fly basis construction (sparse-ir 1.x, SparseIR.jl 1.x)#

The original Python implementation sparse-ir 1.x was mainly developed by Markus Wallerberger. It was the first implementation that constructed the IR basis functions on the fly for arbitrary cutoff \(\Lambda\), without relying on a precomputed database. This provided a practical and widely used realization of the intermediate representation for imaginary-time Green’s functions.

The Julia port SparseIR.jl 1.x was then developed primarily by Samuel Badr and Markus Wallerberger. Both the Python and Julia implementations received contributions from Hiroshi Shinaoka.

The tutorials for these libraries were developed collaboratively by the three authors above and many physicists listed in the contributor list below.

Portable and universal low-level implementation#

Since around 2024, the focus has shifted towards building a portable and universal low-level implementation that can be used from multiple languages via a common C-compatible API.

An important application that motivated this direction is the work of [Mori et al., 2024], where IR-based quantities were computed with sparse-ir (Python) and the resulting data were saved to files and consumed by the EPW code (EPW website). This workflow made it clear that having a low-level library which can compute IR basis functions on the fly in any language via a stable C API would be highly beneficial.

As a first step, a C++ prototype library libsparseir was developed by Satoshi Terasaki and Hiroshi Shinaoka. This attempt explored how to expose IR functionality through a C API, but was not released as a long-term, production-quality solution.

The project then migrated to Rust as sparse-ir-rs, again led by Satoshi Terasaki and Hiroshi Shinaoka. The new versions

  • SparseIR.jl 2.x (Julia)

  • sparse-ir 2.x (Python)

are designed as high-level wrappers around this Rust implementation. The move from C++ to Rust was motivated by improved portability and maintainability, while still exposing a C-compatible API that can be used from C, C++, Fortran, Python, Julia, and other languages.

For high-precision computations, sparse-ir-rs uses the emulated quadruple-precision backend provided by the Rust crate xprec-rs, which was developed in parallel by Markus Wallerberger.

The Fortran bindings to the Rust implementation were developed by Hitoshi Mori in collaboration with Satoshi Terasaki and Hiroshi Shinaoka.

Contributors#

The sparse-IR ecosystem (libraries and tutorials) has benefitted from the contributions of many researchers. The following list collects the main contributors in alphabetical order (by family name):

  • Samuel Badr (TU Wien)

  • Shintaro Hoshino (Saitama Univ.)

  • Fumiya Kakizawa (Saitama Univ.)

  • Takashi Koretsune (Tohoku Univ.)

  • Hitoshi Mori (Riken)

  • Yuki Nagai (JAEA, Riken)

  • Kosuke Nogaki (Kyoto Univ.)

  • Takuya Nomoto (Univ. Tokyo)

  • Junya Otsuki (Okayama Univ.)

  • Soshun Ozaki (Univ. Tokyo)

  • Rihito Sakurai (Saitama Univ.)

  • Hiroshi Shinaoka (Saitama Univ.)

  • Satoshi Terasaki (AtelierArith)

  • Constanze Vogel (TU Wien)

  • Markus Wallerberger (TU Wien)

  • Niklas Witt (Univ. Hamburg)

  • Kazuyoshi Yoshimi (ISSP)

If you find that your contribution is missing or incorrectly described, please feel free to open an issue or pull request on the documentation repository.

Contents