EJML: Efficient Java Matrix Library

What Is It

A linear algebra library for Java — math tools for working with matrices — written 100% in Java and released free under the Apache v2.0 license. It handles real, complex, dense, and sparse matrices.

Design Goals

• Efficiency: As computationally and memory-efficient as possible for both small and large matrices.
• Accessibility: Usable by novices and experts.
• It reaches these goals by picking the best algorithm at runtime, offering a clean API, and providing multiple interfaces.

Three Ways to Use It

1. Operations (Procedural) — lowest-level, gives almost complete control over memory, speed, and specific algorithms. Uses a procedural API (function calls on matrices).
2. SimpleMatrix — a simplified, easy-to-use object-oriented API inspired by Jama, with a flow-style interface.
3. Equations — a symbolic interface similar to Matlab; lets you write compact equation strings.

All three can do the same math (the README shows a Kalman gain example in each style).

Feature Highlights

• Basic operators — addition, multiplication, and more
• Matrix manipulation — extract, insert, combine, etc.
• Linear solvers — linear, least-squares, incremental
• Decompositions — LU, QR, Cholesky, SVD, Eigenvalue, and more
• Matrix features — rank, symmetry, definiteness, etc.
• Random matrix generation — covariance, orthogonal, symmetric
• Different internal formats — row-major, block, sparse
• Graph BLAS (semirings)
• Single-thread and concurrent implementations
• Kotlin extensions
• Unit testing of every algorithm

Quality & Performance

• Extensively unit-tested to ensure correctness
• Benchmarked internally and against the Java Matrix Benchmark
• Documentation at ejml.org (manual, JavaDoc)

Package & Modules

Available on Maven Central as org.ejml:ejml-all:0.46.0 (or pull in individual modules):

Module	Purpose
ejml-core	Core data structures and common code
ejml-fdense	Dense real 32-bit float algorithms
ejml-ddense	Dense real 64-bit float algorithms
ejml-cdense	Dense complex 32-bit float algorithms
ejml-zdense	Dense complex 64-bit float algorithms
ejml-fsparse	Sparse real 32-bit float algorithms
ejml-dsparse	Sparse real 64-bit float algorithms
ejml-simple	SimpleMatrix and Equations interfaces
ejml-java9module	Single unified jar for Java Module System (JPMS)

Java Module System note: don't put multiple ejml-* modules on the module path at once — you'll get split-package errors. Use the ejml-java9module jar instead.

Naming Convention for Matrix Types

Matrix class names follow a pattern like DMatrixRMaj:

• First char = data type: D (real double), F (real float), Z (complex double), C (complex float), B (binary)
• Middle = structure: RMaj (dense row-major), RBlock (block), NxN/N (fixed size), CSC (compressed sparse column), Triplet

Example: DMatrixRMaj = double, real, dense, row-major matrix.

Building From Source

• Built with Gradle; requires Java 17+, but produces Java 11 byte code.
• Commands: ./gradlew autogenerate then ./gradlew publishToMavenLocal.
• Other commands: createLibraryDirectory (save all module jars in ejml/libraries) and oneJar (compile all into a single EJML.jar).
• Pre-built jars are available on Maven Central — most users don't need to build.

Repo Layout

• docs/ — documentation (may be outdated; online is freshest)
• examples/ — usage examples
• main/ — library source code
• change.txt — version history

Community & Support

• Google Group: efficient-java-matrix-library-discuss
• Issue tracker: GitHub issues (github.com/lessthanoptimal/ejml/issues)
• Credits: the author thanks David Watkins (for "Fundamentals of Matrix Computations") and Timothy A. Davis (for "Direct Methods for Sparse Linear Systems" and CSparse, which seeded the sparse algorithms).