gf2 0.1.0__tar.gz

gf2-0.1.0/.gitignore

@@ -0,0 +1,217 @@
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[codz]
+*$py.class
+
+# C extensions
+*.so
+
+# Distribution / packaging
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+share/python-wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+MANIFEST
+
+# PyInstaller
+#  Usually these files are written by a python script from a template
+#  before PyInstaller builds the exe, so as to inject date/other infos into it.
+*.manifest
+*.spec
+
+# Installer logs
+pip-log.txt
+pip-delete-this-directory.txt
+
+# Unit test / coverage reports
+htmlcov/
+.tox/
+.nox/
+.coverage
+.coverage.*
+.cache
+nosetests.xml
+coverage.xml
+*.cover
+*.py.cover
+.hypothesis/
+.pytest_cache/
+cover/
+
+# Translations
+*.mo
+*.pot
+
+# Django stuff:
+*.log
+local_settings.py
+db.sqlite3
+db.sqlite3-journal
+
+# Flask stuff:
+instance/
+.webassets-cache
+
+# Scrapy stuff:
+.scrapy
+
+# Sphinx documentation
+docs/_build/
+
+# PyBuilder
+.pybuilder/
+target/
+
+# Jupyter Notebook
+.ipynb_checkpoints
+
+# IPython
+profile_default/
+ipython_config.py
+
+# pyenv
+#   For a library or package, you might want to ignore these files since the code is
+#   intended to run in multiple environments; otherwise, check them in:
+# .python-version
+
+# pipenv
+#   According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
+#   However, in case of collaboration, if having platform-specific dependencies or dependencies
+#   having no cross-platform support, pipenv may install dependencies that don't work, or not
+#   install all needed dependencies.
+#Pipfile.lock
+
+# UV
+#   Similar to Pipfile.lock, it is generally recommended to include uv.lock in version control.
+#   This is especially recommended for binary packages to ensure reproducibility, and is more
+#   commonly ignored for libraries.
+#uv.lock
+
+# poetry
+#   Similar to Pipfile.lock, it is generally recommended to include poetry.lock in version control.
+#   This is especially recommended for binary packages to ensure reproducibility, and is more
+#   commonly ignored for libraries.
+#   https://python-poetry.org/docs/basic-usage/#commit-your-poetrylock-file-to-version-control
+#poetry.lock
+#poetry.toml
+
+# pdm
+#   Similar to Pipfile.lock, it is generally recommended to include pdm.lock in version control.
+#   pdm recommends including project-wide configuration in pdm.toml, but excluding .pdm-python.
+#   https://pdm-project.org/en/latest/usage/project/#working-with-version-control
+#pdm.lock
+#pdm.toml
+.pdm-python
+.pdm-build/
+
+# pixi
+#   Similar to Pipfile.lock, it is generally recommended to include pixi.lock in version control.
+#pixi.lock
+#   Pixi creates a virtual environment in the .pixi directory, just like venv module creates one
+#   in the .venv directory. It is recommended not to include this directory in version control.
+.pixi
+
+# PEP 582; used by e.g. github.com/David-OConnor/pyflow and github.com/pdm-project/pdm
+__pypackages__/
+
+# Celery stuff
+celerybeat-schedule
+celerybeat.pid
+
+# SageMath parsed files
+*.sage.py
+
+# Environments
+.env
+.envrc
+.venv
+env/
+venv/
+ENV/
+env.bak/
+venv.bak/
+
+# Spyder project settings
+.spyderproject
+.spyproject
+
+# Rope project settings
+.ropeproject
+
+# mkdocs documentation
+/site
+
+# mypy
+.mypy_cache/
+.dmypy.json
+dmypy.json
+
+# Pyre type checker
+.pyre/
+
+# pytype static type analyzer
+.pytype/
+
+# Cython debug symbols
+cython_debug/
+
+# PyCharm
+#  JetBrains specific template is maintained in a separate JetBrains.gitignore that can
+#  be found at https://github.com/github/gitignore/blob/main/Global/JetBrains.gitignore
+#  and can be added to the global gitignore or merged into this file.  For a more nuclear
+#  option (not recommended) you can uncomment the following to ignore the entire idea folder.
+#.idea/
+
+# Abstra
+# Abstra is an AI-powered process automation framework.
+# Ignore directories containing user credentials, local state, and settings.
+# Learn more at https://abstra.io/docs
+.abstra/
+
+# Visual Studio Code
+#  Visual Studio Code specific template is maintained in a separate VisualStudioCode.gitignore 
+#  that can be found at https://github.com/github/gitignore/blob/main/Global/VisualStudioCode.gitignore
+#  and can be added to the global gitignore or merged into this file. However, if you prefer, 
+#  you could uncomment the following to ignore the entire vscode folder
+# .vscode/
+
+# Ruff stuff:
+.ruff_cache/
+
+# PyPI configuration file
+.pypirc
+
+# Cursor
+#  Cursor is an AI-powered code editor. `.cursorignore` specifies files/directories to
+#  exclude from AI features like autocomplete and code analysis. Recommended for sensitive data
+#  refer to https://docs.cursor.com/context/ignore-files
+.cursorignore
+.cursorindexingignore
+
+# Marimo
+marimo/_static/
+marimo/_lsp/
+__marimo__/
+
+# Ruff cache
+.ruff_cache/
+
+# macOS
+.DS_Store
+
+.kiro/
+# graphify knowledge graph (regenerate with: graphify update .)
+graphify-out/

gf2-0.1.0/CHANGELOG.md

@@ -0,0 +1,128 @@
+# Changelog
+
+All notable changes to this project are documented here. The format follows
+[Keep a Changelog](https://keepachangelog.com/en/1.1.0/) and this project
+adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
+
+## [Unreleased]
+
+## [0.1.0] - 2026-09-05
+
+First public release.
+
+### Renamed
+
+- The project was developed as `binpy`. That name belongs to an unrelated
+  digital-logic package on PyPI, so the import root is `gf2`. There was no
+  public `binpy` release, so no upgrade path is needed: use `import gf2`.
+
+### Added
+
+- `SparseGF2Matrix` / `DenseGF2Matrix` with automatic CSR / bit-packed storage
+  selection, and `create_sparse_matrix`.
+- Core arithmetic: `add`, `multiply`, `transpose`, `rank`, `det`, `trace`,
+  `is_invertible`, `reduced_row_echelon_form`, `lu_decomposition`,
+  `matrix_power`, `characteristic_polynomial`, `minimal_polynomial`.
+- Solvers: `solve`, `solve_multiple_rhs`, `nullspace`, `nullspace_bitwise`,
+  `nullspace_fast`, `inverse`, `kernel`, `image`, `rank_nullity_theorem`,
+  `iterative_refinement`.
+- Generators: `identity`, `zeros`, `ones`, `random_sparse`, `random_regular`,
+  `circulant`, `circulant_random`, `toeplitz`, `ldpc_matrix`,
+  `hamming_matrix`, `repetition_matrix`.
+- Quantum code constructions: `hypergraph_product` (Tillich-Zemor),
+  `surface_code_matrix` (planar, built as a hypergraph product of repetition
+  codes), `css_code_matrix`, `bicycle_codes`.
+- `py.typed`: the package is annotated throughout and ships its types.
+- Structural equality on both matrix types: `A == B` compares shape and
+  contents, independently of storage format, and a sparse matrix compares equal
+  to a dense one holding the same bits. Previously `==` fell through to
+  identity and returned False for equal matrices. Defining `__eq__` makes the
+  types unhashable, which is correct: they are mutable through `set_bit`.
+- Operations returning several values return NamedTuples -- `LUDecomposition`,
+  `RowEchelonForm`, `NullspaceVector`, `RankNullity` -- so a caller can write
+  `result.pivot_columns` instead of `result[1]`. They unpack positionally
+  exactly like the plain tuples they replace.
+- `lu_decomposition` returns `(L, U, perm)` with `A[perm] == L @ U`, plus
+  `LUDecomposition.permutation_matrix()` for the equivalent `P @ A == L @ U`
+  form.
+- A benchmark suite (`benchmarks/`) with a documented measurement methodology,
+  and a report generated from recorded measurements rather than written by hand.
+
+### Performance
+
+Relative to the pre-release development tree, same inputs and warm caches:
+
+- `multiply` 22.9x faster at n=512 (Method of Four Russians over packed uint64;
+  row-XOR accumulation for sparse operands).
+- `nullspace` basis extraction 49.3x faster on a rank-deficient 256x256.
+- `transpose` 2.6x, `to_dense` 7.8x, `rank` 1.8x at n=512.
+- `ldpc_matrix(method="progressive")` 49.6x faster; `ones` 508x.
+- `minimal_polynomial` moved from O(2^n) enumeration to O(n^4 / 64) word
+  operations: n=16 went from 38 s to 0.18 ms.
+- CSR row access no longer rebuilds each row per call; a full row sweep of a
+  300x300 matrix went from 3.33 ms to 0.014 ms.
+
+See `benchmarks/OPTIMIZATION_LOG.md`.
+
+### Fixed
+
+Defects found during pre-release review, all of which failed silently:
+
+- `set_bit` discarded the write on an empty-format matrix and left the
+  packed-row cache stale on the CSR path, so later reads saw the pre-write
+  matrix.
+- `hypergraph_product` returned an all-zero `H_z`, so the CSS commutation
+  condition held only vacuously.
+- `surface_code_matrix` declared more stabiliser rows than it populated and did
+  not satisfy `H_x @ H_z.T == 0`; it now produces a genuine planar surface code
+  with k = 1.
+- `characteristic_polynomial` returned a stub that was wrong for every n >= 3.
+- `nullspace_fast` accepted `include_packing_time` and ignored it.
+- `iterative_refinement` mutated a NumPy right-hand side in place.
+- Coordinate input now de-duplicates, bounds-checks, and handles zero
+  dimensions; packed rows are masked to the column count.
+- `random_regular` could emit under-weight rows; seeded generators no longer
+  reseed the global `random` module.
+- `lu_decomposition` swapped rows of U without recording the permutation or
+  applying it to L, so its output satisfied neither `A == L @ U` nor
+  `P @ A == L @ U`. It reconstructed A in roughly half of random cases and gave
+  the caller no way to recover the difference. It now returns the permutation
+  as a third element and reconstructs exactly, verified over 500 random
+  matrices. The existing test was named `..._reconstruction` but only asserted
+  the triangular shapes, which is how this survived.
+- The packed representation assumed the host was little-endian: it paired
+  `int.to_bytes(..., "little")` with native-order NumPy arrays. On a big-endian
+  host (NumPy ships s390x wheels) a single set bit in word 0 read back as
+  2**56, and `unpackbits` returned the wrong byte's bits - wrong answers rather
+  than an error. The storage dtype is now pinned to `"<u8"`, which is the
+  native dtype on little-endian hosts and therefore costs nothing there.
+
+### Removed
+
+Cut before the first release rather than published under a name that promises
+something else. None of these had users; all three were exported and would have
+been a breaking change to withdraw later.
+
+- `bch_matrix` was not a BCH parity check matrix. It applied a fixed arithmetic
+  mask (`(i+1)*(j+1) % 3 == 1`) with no coding-theoretic meaning and ignored
+  its error-correction parameter `t` entirely.
+- `vandermonde` reduced integer powers mod 2. Since the parity of a power
+  depends only on the parity of the base, every row came out all-ones or
+  `[1, 0, 0, ...]`, giving a matrix of rank at most 2 — not a Vandermonde
+  matrix over GF(2^m).
+- `color_code_matrix` did not satisfy `H_x @ H_z.T == 0`, so the pair it
+  returned was not a quantum code. Use `surface_code_matrix` or
+  `hypergraph_product`, which are exact by construction.
+
+### Known limitations
+
+- Dense GF(2) multiplication is competitive with `galois` only from n = 1024;
+  below that `galois` is faster. Use it or `m4ri` if dense multiplication
+  dominates your workload.
+- No extension-field GF(2^m) arithmetic. `vandermonde` and `bch_matrix` were
+  removed before release rather than shipped under names they did not earn
+  (see Removed); real versions need GF(2^m) and are deferred.
+- Pure Python plus NumPy: no compiled kernel.
+
+[Unreleased]: https://github.com/kkKaan/gf2/compare/v0.1.0...HEAD
+[0.1.0]: https://github.com/kkKaan/gf2/releases/tag/v0.1.0

gf2-0.1.0/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2025 Kaan Karaçanta
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

gf2-0.1.0/PKG-INFO

@@ -0,0 +1,245 @@
+Metadata-Version: 2.5
+Name: gf2
+Version: 0.1.0
+Summary: High-performance binary (GF(2)) matrix operations for Python
+Project-URL: Homepage, https://github.com/kkKaan/gf2
+Project-URL: Source, https://github.com/kkKaan/gf2
+Project-URL: Issues, https://github.com/kkKaan/gf2/issues
+Author: Kaan Karaçanta
+License: MIT License
+        
+        Copyright (c) 2025 Kaan Karaçanta
+        
+        Permission is hereby granted, free of charge, to any person obtaining a copy
+        of this software and associated documentation files (the "Software"), to deal
+        in the Software without restriction, including without limitation the rights
+        to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+        copies of the Software, and to permit persons to whom the Software is
+        furnished to do so, subject to the following conditions:
+        
+        The above copyright notice and this permission notice shall be included in all
+        copies or substantial portions of the Software.
+        
+        THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+        IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+        FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+        AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+        LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+        OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+        SOFTWARE.
+License-File: LICENSE
+Keywords: GF(2),LDPC,binary,bitwise,coding theory,linear algebra,matrix,sparse
+Classifier: Development Status :: 3 - Alpha
+Classifier: Intended Audience :: Developers
+Classifier: Intended Audience :: Science/Research
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Programming Language :: Python
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3 :: Only
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Topic :: Scientific/Engineering :: Mathematics
+Classifier: Topic :: Security :: Cryptography
+Classifier: Typing :: Typed
+Requires-Python: >=3.10
+Requires-Dist: numpy>=1.23
+Provides-Extra: bench
+Requires-Dist: galois>=0.3; (python_version >= '3.10') and extra == 'bench'
+Requires-Dist: matplotlib>=3.7; extra == 'bench'
+Requires-Dist: scipy>=1.10; extra == 'bench'
+Provides-Extra: dev
+Requires-Dist: mypy==2.3.1; extra == 'dev'
+Requires-Dist: pre-commit>=3.7; extra == 'dev'
+Requires-Dist: ruff==0.16.6; extra == 'dev'
+Provides-Extra: test
+Requires-Dist: hypothesis>=6.90; extra == 'test'
+Requires-Dist: psutil>=5.9; extra == 'test'
+Requires-Dist: pytest-cov>=5.0; extra == 'test'
+Requires-Dist: pytest>=8.0; extra == 'test'
+Description-Content-Type: text/markdown
+
+# gf2
+
+High-performance binary (GF(2)) matrix operations library for Python.
+
+[![CI](https://github.com/kkKaan/gf2/workflows/CI/badge.svg)](https://github.com/kkKaan/gf2/actions)
+[![Python 3.10+](https://img.shields.io/badge/python-3.10+-blue.svg)](https://www.python.org/downloads/)
+[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
+
+## Installation
+
+```bash
+# Install from source (development)
+git clone https://github.com/kkKaan/gf2.git
+cd gf2
+pip install -e ".[dev,test]"
+```
+
+## Quick Start
+
+```python
+import gf2
+
+# Create matrices
+A = gf2.identity(5)  # 5x5 identity matrix
+B = gf2.random_sparse(5, 5, density=0.3)  # Random sparse matrix
+C = gf2.zeros(3, 4)  # 3x4 zero matrix
+
+# Basic operations (all in GF(2))
+sum_matrix = gf2.add(A, B)  # XOR addition
+product = gf2.multiply(A, B)  # Binary matrix multiplication
+A_transpose = gf2.transpose(A)  # Matrix transpose
+
+# Linear algebra
+r = gf2.rank(A)  # Matrix rank
+det_A = gf2.det(A)  # Determinant (0 or 1)
+is_inv = gf2.is_invertible(A)  # Invertibility check
+
+# Solve linear systems Ax = b over GF(2)
+b = [1, 0, 1, 0, 1]
+x = gf2.solve(A, b)  # Exact solution
+null_space = gf2.nullspace(A)  # Null space basis
+
+# Matrix generators for coding theory
+H = gf2.hamming_matrix(3)  # Hamming code parity check
+ldpc = gf2.ldpc_matrix(100, 200, row_weight=4)  # LDPC code (m*row_weight must divide n)
+circ = gf2.circulant([1, 0, 1, 1])  # Circulant matrix
+```
+
+## Advanced Usage
+
+### Custom Sparse Matrices
+
+```python
+# Create from coordinates
+coords = [(0, 1), (1, 2), (2, 0)]  # (row, col) positions
+matrix = gf2.create_sparse_matrix(3, 3, coordinates=coords)
+
+# Different storage formats are automatically chosen
+dense_like = gf2.random_sparse(10, 10, density=0.8)  # Uses bit-packed storage
+very_sparse = gf2.random_sparse(1000, 1000, density=0.01)  # Uses CSR
+
+# Access internal representation
+print(matrix.memory_usage())  # Shows compression statistics
+```
+
+### Coding Theory Applications
+
+```python
+# Generate LDPC codes
+H = gf2.ldpc_matrix(m=500, n=1000, row_weight=6, method="progressive")
+
+# Classical codes
+hamming_H = gf2.hamming_matrix(r=4)  # [15,11,3] Hamming code
+rep_H = gf2.repetition_matrix(5)  # length-5 repetition code
+
+# Quantum codes (exact CSS commutation: H_x @ H_z.T == 0)
+H_x, H_z = gf2.surface_code_matrix(distance=3)  # planar surface code, k=1
+Q_x, Q_z = gf2.hypergraph_product(hamming_H, rep_H)  # Tillich-Zemor product
+
+# Structured matrices
+toeplitz_A = gf2.toeplitz([1, 0, 1], [1, 1, 0, 1])
+circ = gf2.circulant([1, 0, 1, 1])
+```
+
+## Performance
+
+gf2 stores rows bit-packed and does GF(2) arithmetic with whole-row bitwise
+operations, so a row XOR costs one machine word per 64 columns instead of one
+Python step per column.
+
+- **Elimination** (rank, nullspace, solve, inverse) runs on Python big
+  integers below n = 384 and switches to vectorised NumPy uint64 rows above
+  it, because the crossover between the two was measured, not assumed.
+- **Multiplication** uses the Method of Four Russians: one lookup table of
+  2^8 pre-combined rows of B serves every row of A, so the XOR count drops
+  from O(mn/2) to O(mn/8 + n/8 * 256).
+- **Storage** picks CSR or bit-packed automatically from the density.
+
+### Measured results
+
+Numbers, methodology, and the exact environment live in
+[`benchmarks/BENCHMARK_RESULTS.md`](https://github.com/kkKaan/gf2/blob/main/benchmarks/BENCHMARK_RESULTS.md), which is
+**generated from `benchmarks/results.json`** rather than written by hand.
+
+To reproduce:
+
+```bash
+python benchmarks/bench_gf2.py      # measure -> benchmarks/results.json
+python benchmarks/make_report.py    # results.json -> BENCHMARK_RESULTS.md
+```
+
+Summary against the fastest honest rival at each size, square matrices at 50%
+density, Python 3.11 / NumPy 2.2.6 / galois 0.4.6 on an arm64 Mac:
+
+| operation | n=128 | n=512 | n=1024 |
+|---|---|---|---|
+| **rank** | **2.3x faster** than packed-NumPy | par with packed-NumPy | 1.1x slower than packed-NumPy |
+| **nullspace vector** | **1.9x faster** than packed-NumPy | **1.5x faster** | **1.4x faster** |
+| **multiply** | 5.9x slower than galois | 1.3x slower than galois | par with galois; **85x faster** than NumPy |
+
+Against `galois`, gf2's rank is 12-27x faster and its nullspace 6-13x faster
+across this range. Against a *naive* element-wise NumPy loop gf2 looks 9-13x
+faster, but that comparison is not meaningful and the benchmark labels it as a
+strawman: the baseline that matters is bit-packed uint64 NumPy, which is the
+`numpy-packed` row in the report.
+
+### Honest limitations
+
+- **Dense matrix multiply is not gf2's strength.** `galois` is faster below
+  n = 1024 and gf2 only draws level there. If dense GF(2) multiplication
+  dominates your workload, use `galois` or `m4ri`.
+- **Peak memory during multiply is higher than NumPy's**, because the Four
+  Russians table and the unpacked selector are transient allocations. Bit
+  packing wins on *stored* size, not on scratch space.
+- Everything here is pure Python plus NumPy. A C or Cython kernel would move
+  the elimination crossover a long way down.
+
+## Development
+
+```bash
+# Clone and setup
+git clone https://github.com/kkKaan/gf2.git
+cd gf2
+python -m venv .venv
+source .venv/bin/activate  # On Windows: .venv\Scripts\activate
+pip install -e ".[dev,test]"
+
+# Run tests
+pytest tests/
+
+# Code quality
+ruff check .          # Linting
+ruff format .         # Formatting  
+mypy gf2/           # Type checking
+
+# Install pre-commit hooks
+pre-commit install
+```
+
+## Simon's Algorithm Postprocessing
+
+gf2 provides fast GF(2) nullspace routines used in Simon-style workflows:
+
+- High-level basis: `gf2.nullspace(A)` returns a basis as a list of 0/1 lists
+- Fast bitwise single solution: `gf2.nullspace_bitwise(A)` -> `(solution_bits: str, seconds: float)`
+- Zero-overhead raw input: `gf2.nullspace_fast(matrix)` -> `(solution_bits: str, seconds: float)`
+
+Example using `nullspace_fast` directly on list-of-lists:
+
+```python
+from gf2 import nullspace_fast
+
+matrix = [
+    [1, 0, 1, 0, 1],
+    [0, 1, 1, 0, 0],
+    [1, 1, 0, 1, 0],  # use n-1 rows for underdetermined system
+]
+
+solution_bits, elapsed = nullspace_fast(matrix)
+print(solution_bits, elapsed)
+```
+
+This returns a nontrivial nullspace vector as a binary string and the elapsed time, matching usage patterns in Simon postprocessing scripts.