Authors and citation

Who wrote lzcomplexity, where it came from, how to cite it, and where to report bugs.

The people

• Efrén Aragón Pérez — principal author and creator. lzcomplexity exists because of him. He wrote the original C++ implementation that everything here descends from, and he holds the copyright on the MIT licence.
• Daniel Estévez Moya — author of the Rust rewrite that is the current production backend: the lzcomplexity-core crate, the PyO3 bindings, and the two command-line tools shipped in 1.0.
• Ernesto Estévez Rams — from the research lineage of the original C++ implementation.

That order is the one recorded in Cargo.toml, pyproject.toml, README.md and the BibTeX entry below.

Note

The published package metadata on PyPI and crates.io still carries the older two-name author list until the next release goes out. The repository is the current record.

Where the library came from

lzcomplexity started as a C++ library for complexity analysis using entropic magnitudes — entropy density, effective measure complexity and information distance, all derived from Lempel–Ziv 76 factorization. That implementation still lives on the repository's main branch, with Python bindings built on nanobind.

The rewrite landed in 0.11.0 and the API settled at 1.0.0. The Python bindings moved to PyO3, the CLI was rebuilt on the same core, and the shuffle PRNG stopped drawing from a clock-seeded mt19937: it is now a per-block ChaCha8Rng seeded from a hash of the sequence itself, so the same input returns the same emc on every run and at any thread count. There is no user-facing seed parameter.

Factor counts, factor boundaries and nid match the C++ backend exactly; h matches at equal log base, and emc now matches too — the C++ backend adopted both the projected-ladder estimator and the content-seeded ChaCha8 shuffle, so the two draw the same surrogates. See Rust vs C++ and Determinism.

Spectral analysis (spectral.psd, spectral.entropy, spectral.semc) was removed in the same 0.11.0 change and moved to a separate package. It is not part of the 1.0 API, and the CI wheel smoke test asserts that spectral and metrics stay absent.

---

Citing lzcomplexity

Cite the software with the entry below, replacing the version with the one you actually ran.

@software{lzcomplexity_2026,
  title   = {lzcomplexity: LZ76 complexity, entropy rate and information distance
             for symbolic sequences},
  author  = {Arag{\'o}n P{\'e}rez, Efr{\'e}n and
             Est{\'e}vez Moya, Daniel and
             Est{\'e}vez Rams, Ernesto},
  url     = {https://github.com/pleros-ai/lzcomplexity},
  version = {1.0.1},
  year    = {2026}
}

The accents are written as BibTeX escapes so the entry survives a classic 8-bit bibtex run. With biblatex and a UTF-8 .bib file you can write Aragón Pérez, Efrén directly.

Plain text:

Aragón Pérez, E., Estévez Moya, D., & Estévez Rams, E. (2026). lzcomplexity: LZ76 complexity, entropy rate and information distance for symbolic sequences (version 1.0.1) [Computer software]. https://github.com/pleros-ai/lzcomplexity

The installed version is on the module, so a methods section can quote the exact build that produced its numbers:

>>> import lzcomplexity as lz
>>> lz.__version__
'1.0.1'

Quote the version, not just the name. emc values from the C++ line do not compare with 1.x, and the Releases page records what changed when.

Note

A citation to this software does not credit the method. The factor count, h and the information distance come from published results by other people; the library implements them. Cite Lempel & Ziv (1976) for the factorization, and the relevant entropy-rate and distance papers alongside the software — all of them are on References.

Licence

lzcomplexity is MIT licensed. The copyright line reads, verbatim, Copyright (c) 2024 Efren Aragon Perez — unaccented in the file.

The full text is in LICENSE at the repository root, and every wheel carries a copy at lzcomplexity-<version>.dist-info/licenses/LICENSE. You may use, modify and redistribute the library, including commercially, provided the copyright notice and licence text travel with it. There is no warranty.

Reporting bugs

Open an issue at https://github.com/pleros-ai/lzcomplexity/issues.

Both binaries print Send bug reports to estevez@fisica.uh.cu or efrenaragon96@gmail.com. in their --help banner. The issue tracker is the better route: it is public, searchable, and linked from the package metadata.

A report that can be reproduced gets fixed faster. Include:

Item	How to get it
Version	lz.__version__, or --version on either binary
Platform	OS, architecture, and Python version
Input	The sequence, or a script that regenerates it
Expected vs actual	The number you got and the number you expected

The two CLIs report their versions in different formats:

$ lzcomplexity --version
 [ Info ] v1.0.1
$ lzdistance --version
 [ Info ] Version of lzdistance: v1.0.1

The discrete part of every computation is deterministic: the same bytes give the same factor count, the same factor boundaries and the same shuffle permutation, in every process and on every platform. Floats carry one caveat — ln comes from the platform's libm, and the CLI and Python entropy paths order their arithmetic differently, so the last bit of h or emc can differ by a few ULP. Report a discrepancy larger than that. Determinism sets out exactly what is guaranteed.

Release notes and the changelog for every version are on Releases.