For benchmarks showing ParserNG duking it out in the ring with other parsers, check ParserNg-wars
ParserNG 1.1.3 is a blazing-fast, nigh zero allocation(memory wise), pure Java, zero-native-dependencies math expression parser and evaluator.
It beats exp4J, and com.expression.parser on evaluation speed across every kind of expression — from simple algebra to heavy trig, matrices, and calculus; and manages to beat Janino, the gold standard on some, while rivalling it on a host of other expressions
The normal mode routinely does about 3-10 million evaluations per second while the new Turbo mode easily peaks at about 10 million to 90 million evaluations per second.
It goes far beyond basic parsing — offering symbolic differentiation, resilient numerical integration, full matrix algebra, statistics, equation solving, user-defined functions, 2D graphing, and more — all in one lightweight, cross-platform library.
Perfect for scientific computing, simulations, real-time systems, education tools, Android apps, financial models, and high-performance scripting.
1.1.3 introduces Turbo Scalar and Turbo Matrix compiled paths + massive speed improvements via strength reduction, constant folding, and O(1) frame-based argument passing.
- Speed champion — rivals Janino in most benchmarks, and beats exp4J, com.expression.parser and Parsii in every benchmark (see BENCHMARK_RESULTS.md)
- Turbo Mode — compile once, evaluate millions of times per second (Scalar + Matrix paths)
- Symbolic differentiation (
diff) + resilient numerical integration (intg) that handles difficult expressions - Full matrix algebra (
det,eigvalues,eigvec,adjoint,cofactor, matrix division,linear_sys, …) - Statistics (
avg,variance,cov,min,max,rms,listsum,sort, …) - Equation solvers: quadratic, Tartaglia cubic, numerical roots, linear systems
- User-defined functions (
f(x)=…or lambda@(x,y)…) + persistentFunctionManager/VariableManager - Variables with execution frames for ultra-fast loops
- 2D function & geometric plotting support
- Logical expressions (
and,or,==, …) - No external dependencies — runs on Java SE, Android, JavaME, …
<dependency>
<groupId>com.github.gbenroscience</groupId>
<artifactId>parser-ng</artifactId>
<version>1.1.3</version>
</dependency>Also available on Maven Central:
https://central.sonatype.com/artifact/com.github.gbenroscience/parser-ng/1.1.3
import com.github.gbenroscience.parser.MathExpression;
MathExpression me = new MathExpression("x=2;sin(3*x+2)+sec(5*x/4)");
double res = me.solveGeneric().scalar;import com.github.gbenroscience.parser.MathExpression;
import com.github.gbenroscience.parser.turbo.tools.FastCompositeExpression;String expr = "x*sin(x) + y*cos(y) + z^2";
MathExpression me = new MathExpression(expr, false); // prepare for turbo
FastCompositeExpression turbo = me.compileTurbo(); // compile once
int xSlot = me.getVariable("x").getFrameIndex();
int ySlot = me.getVariable("y").getFrameIndex();
int zSlot = me.getVariable("z").getFrameIndex();
double[] frame = new double[3];
for (double t = 0; t < 10_000_000; t += 0.001) {
frame[xSlot] = t;
frame[ySlot] = t * 1.5;
frame[zSlot] = t / 2.0;
double result = turbo.applyScalar(frame); // ← ultra-fast!
}OR in more modern versions of Turbo, JUST
double[] frame = new double[3];
for (double t = 0; t < 10_000_000; t += 0.001) {
frame[0] = t;
frame[1] = t * 1.5;
frame[2] = t / 2.0;
//the slots are auto-computed inside the applyScalar and other applyXXX methods
double result = turbo.applyScalar(frame); // ← ultra-fast!
}String expression="R=@(3,3)(5,1,3, 2,9,12, 1,5,18)"
MathExpression me = new MathExpression("eigvalues(R)");
FastCompositeExpression turbo = TurboEvaluatorFactory.getCompiler(me).compile();
Matrix result = turbo.applyMatrix(new double[0]); // works for: linear_sys, adjoint, cofactor, A/B, etc. String expression = "R=@(3,3)(5,1,3, 2,9,12, 1,5,18);A=@(3,3)(2,0,5, 8,9,13, 1,2,1);A+2*R-1/A;";
MathExpression me = new MathExpression(expression);
FastCompositeExpression turbo = TurboEvaluatorFactory.getCompiler(me).compile();
double[] vars = {};
MathExpression.EvalResult result = turbo.apply(vars);
System.out.println("result:"+result.toString()); // works for: linear_sys, adjoint, cofactor, A/B, etc. public static void benchmarkMatrixAlgebra() throws Throwable {
int n = 20;
Matrix t = new Matrix(n, n);
t.setName("T");
t.randomFill(35);
t.print();
System.out.println("T: After fill-----\n");
FunctionManager.add(new Function(t));
Matrix v = new Matrix(n, n);
v.setName("V");
v.randomFill(35);
v.print();
System.out.println("V: After fill-----\n");
FunctionManager.add(new Function(v));
String ex = "2*T+V";
MathExpression expr = new MathExpression(ex);
FastCompositeExpression turbo = expr.compileTurbo();
double[] vars = {};
MathExpression.EvalResult er = null;
long start = System.nanoTime();
for (int i = 0; i < 1_000_000; i++) {
er = turbo.apply(vars);
}
long duration = System.nanoTime() - start;
System.out.printf("Expression: %s%n", ex);
System.out.println("res: " + er);
System.out.printf("Speed: %.2f ns/op%n", duration / 1_000_000.0);
System.out.printf("Throughput: %.2f ops/sec%n", 1_000_000.0 / (duration / 1e9));
}MathExpression expr = new MathExpression("r = 5; 2 * pi * r");
System.out.println(expr.solve()); // ≈ 31.4159MathExpression expr = new MathExpression("x^2 + 5*x + sin(x)", false);
FastCompositeExpression turbo = expr.compileTurbo();
int xSlot = expr.getVariable("x").getFrameIndex();
double[] frame = new double[1];
for (double x = 0; x < 100_000; x += 0.1) {
frame[xSlot] = x;
double y = turbo.applyScalar(frame);
// plot or process y
}MathExpression expr = new MathExpression("f(x) = x^3 * ln(x); diff(f, 2, 1)");
System.out.println(expr.solveGeneric().scalar);MathExpression expr = new MathExpression("intg(@(x) 1/(x*sin(x)+3*x*cos(x)), 0.5, 1.8)");
System.out.println("∫ ≈ " + expr.solve());MathExpression expr = new MathExpression("""
M = @(3,3)(1,2,3, 4,5,6, 7,8,9);
det(M)
""");
System.out.println("Determinant = " + expr.solve());You may use the rot function to rotate functions, surfaces(plane or curved), lines and even raw points in 3D space.
To rotate any of these, you need the orbital center, the coordinates of the direction vector(a,b,c) and the angle of rotation.
The example below shows two ways to use the ParserNG library to rotate the point p and q about the orbital center (1,0,1)
with the direction vector(0,0,1). The angle of rotation is pi radians.
String expression = "p=@(1,3)(4,2,5);q=@(1,3)(12,3,-1);rot(p,q, pi, @(1,3)(1,0,1),@(1,3)(1,1,0))";
MathExpression interpreted = new MathExpression(expression);
MathExpression.EvalResult ev = interpreted.solveGeneric();
System.out.printf("Expression: %s%n", expression);
System.out.println("interpreted: " + ev);
// Compile to turbo
FastCompositeExpression compiled = new ScalarTurboEvaluator(interpreted, true).compile();
// Warm up turbo JIT
double[] vars = new double[0];
MathExpression.EvalResult evr = compiled.apply(vars);
System.out.println("turbo: " + evr);The result is an array of 6 elements. The first 3 represent the coordinates of the new P after rotation and the last coordinates represent the coordinates of Q after rotation.
The next example is even more interesting.
ParserNG can take a Matrix of N points in 3D space(so each point has x,y,z coordinates) and rotate them through an angle, about an origin (O) in a specified direction, D. The Matrix would hence be of dimensions (N x 3) and its output will be a Matrix of similar dimensions, and each row in the initial Matrix would have its rotated result in the corresponding rows of the output Matrix.
String expression = "M_IN=@(5,3)(3,1,4, 2,2,8, 7,1,3, 4,5,19, 8,7,21);O=@(1,3)(0,0,0);D=@(1,3)(0,0,1);";
MathExpression me = new MathExpression(expression);
MathExpression m = new MathExpression("rot(M_IN,pi,O,D)");
FastCompositeExpression fce = m.compileTurbo();
MathExpression.EvalResult evr = fce.apply(new double[0]);
System.out.println("turbo: " + evr);OR, you could do it with one object using anonymous syntax:
MathExpression m = new MathExpression(" rot(@(5,3)(3,1,4, 2,2,8, 7,1,3, 4,5,19, 8,7,21), pi, @(1,3)(0,0,0), @(1,3)(0,0,1) ) ");
FastCompositeExpression fce = m.compileTurbo();
MathExpression.EvalResult evr = fce.apply(new double[0]);
System.out.println("turbo: " + evr);java -jar parser-ng-1.1.3.jar "sin(x) + cos(x)"
java -jar parser-ng-1.1.3.jar "eigvalues(R=@(5,5)(...))"
java -jar parser-ng-1.1.3.jar help
java -jar parser-ng-1.1.3.jar -i # interactive mode| Category | Key Functions | Turbo Support |
|---|---|---|
| Arithmetic & Logic | + - * / ^ % and or == != |
Full |
| Trigonometry | sin cos tan asin … sinh |
Full |
| Calculus | diff (symbolic), intg (resilient) |
Yes |
| Equations | Quadratic, Tartaglia cubic, root, linear_sys |
Yes |
| Matrices | det, eigvalues, eigvec, adjoint, cofactor, A / B |
Excellent (Turbo Matrix) |
| Statistics | avg variance cov min max rms listsum sort |
Yes |
| Custom | @(x,y)… or named functions |
Full |
Full list: run help or new MathExpression("help").solve().
- BENCHMARK_RESULTS.md — full speed comparisons
- GRAPHING.md — plotting on Swing / JavaFX / Android
- LATEST.md — what’s new in 1.1.3
- Javadoc: https://javadoc.io/doc/com.github.gbenroscience/parser-ng/1.1.3
- Hello world and original readme — Original readme for pre-1.0 versions with a lot of, still valid, examples
ParserNG is built with love in my free time. If it helps you:
- ⭐ Star the repository
- 🐞 Report bugs or suggest features
- 💡 Share what you built with it
- ☕ Buy me a coffee
Apache License 2.0
ParserNG 1.1.3 — faster than the competition, stronger on matrices, and now with real Turbo Scalar + Turbo Matrix compiled power.
Happy parsing! 🚀
— GBENRO JIBOYE (@gbenroscience)
