HMath

HMath is a mathematics and numerical computing library with Python bindings, written in Haskell. It also provides native Haskell, C, and Wave interfaces.

HMath targets Linux x86-64. The Python distribution is named hmath-python and is imported as hmath. Version 1.0.1 includes the native library and runtime; GHC is not needed to use its Python wheel.

Python quick start

Install the Python package on Linux x86-64:

pip install hmath-python

from hmath import Tensor, add

print(add(2, 3))  # 5.0

with Tensor.matrix([1, 2, 3, 4], 2, 2) as matrix:
    with matrix @ matrix as product:
        print(product.values())  # [7.0, 10.0, 15.0, 22.0]

Use with to release native resources automatically. Invalid operations raise HMathError. Install hmath-python[numpy,cffi] for optional NumPy and cffi support.

See the documentation for installation requirements, tutorials, API reference, and C/Wave integration.

Features

• Core math: checked scalar operations, numeric types, constants, units, number theory, and polynomials.
• Linear algebra: dense and sparse matrices, decompositions, iterative solvers, and CPU tensors.
• Scientific computing: differentiation, integration, ODE/PDE solvers, optimization, interpolation, FFTs, and signal processing.
• Statistics and learning: reproducible sampling, probability distributions, inference, automatic differentiation, and machine learning primitives.
• Discrete and symbolic math: combinatorics, graphs, numerical data structures, symbolic calculus, and a bounded expression evaluator.
• Applied math: complex numbers, rotations, computational geometry, time series, and numerical image processing.
• Native interface: checked scalar, vector, matrix, and tensor operations; explicit ownership and errors; Python/NumPy and Wave bindings.
• Interchange: typed binary and text arrays, CSV/TSV, Matrix Market, NPY/NPZ, and CPU DLPack.

The Haskell library contains the full numerical API; Python currently exposes scalar addition and binary64 vector, matrix, and tensor construction, arithmetic, and reductions.

Build from source

Requires GHC, Cabal, a C11 compiler, Meson >= 1.3, Ninja, and Python 3. Zlib enables compressed interchange. BLAS/LAPACKE are optional runtime backends; the full spectral test suite needs LAPACKE.

git clone https://github.com/wavefnd/hmath.git
cd hmath
meson setup builddir
meson compile -C builddir
meson test -C builddir --print-errorlogs

This builds builddir/libhmath.so and runs the Haskell, C, and Python tests. Wave tests also run when wavec is available.

Haskell

Save the following as Main.hs in the checkout. Run cabal build lib:hmath, then cabal exec -- runghc -package=hmath Main.hs:

import HMath

main :: IO ()
main = do
    print (dotChecked [1, 2, 3 :: Int] [4, 5, 6]) -- Right 32
    print $ do
        matrix <- fromListChecked 2 2 [1, 2, 3, 4 :: Int]
        pure (toList (transposeMatrix matrix))   -- Right [1,3,2,4]

Checked operations return Either MathError for invalid shapes or lengths. More examples are available in examples/haskell.

C and Wave

The C header defines errors, ownership, and runtime lifetime. See the C example and Wave bindings for usage.

Contributing

Report bugs or discuss changes through GitHub issues. Submit pull requests from a fork against master; see ROADMAP.md for planned work.

License

Mozilla Public License 2.0.