nalgebra

nalgebra is a general-purpose linear algebra library for Rust, supporting both statically-sized and dynamically-sized matrices and vectors. It provides the numeric foundation for graphics, robotics, physics, and scientific computing, with transformations, rotations, and a full suite of matrix decompositions.

Beyond raw matrix math, nalgebra offers geometric types such as points, quaternions, isometries, and projective transforms, along with optional sparse matrices, WebAssembly support, and a no_std mode for embedded targets.

What You Get

• Statically- and dynamically-sized matrices and vectors
• Geometric types: points, rotations, quaternions, isometries, and projections
• Matrix decompositions including LU, QR, Cholesky, SVD, and eigen solvers
• Optional sparse matrix support and BLAS-backed multiplication
• no_std and WebAssembly compatibility for embedded and web targets

Common Use Cases

• 3D graphics and game engine math for transforms and cameras
• Robotics and physics simulations requiring rigid-body transforms
• Scientific and numerical computing with matrix decompositions
• Computer-vision and SLAM pipelines needing linear solvers

Under The Hood

Architecture — nalgebra is organized into a base module holding the generic Matrix storage abstraction (parameterized over dimension and storage type), a geometry module for transforms and quaternions, a linalg module for decompositions, plus sparse, io, and proptest modules and third_party integrations. Tech Stack — Rust edition 2024 (MSRV 1.89) built on num-traits, num-complex, simba (SIMD abstraction), matrixmultiply, and approx, with optional serde, rand, and BLAS/LAPACK bindings behind feature flags. Code Quality — The project is large, mature, and actively developed, with property-based tests via a dedicated proptest module, extensive documentation on nalgebra.rs, and a long release history. API Design — The type system encodes matrix dimensions generically, which yields strong compile-time guarantees at the cost of complex signatures; the geometry types offer an ergonomic high-level API, so the learning curve is steeper for the generic core than for everyday transform usage.