About

Matrices underpin nearly every computational discipline: finite element analysis, machine learning weight initialization, graph adjacency representation, and Monte Carlo simulation all require structured random numeric grids. A poorly initialized matrix causes convergence failure in iterative solvers or introduces bias into stochastic models. This tool generates matrices of arbitrary dimension up to 100 × 100 with configurable value bounds a ≤ x_ij ≤ b, selectable number type (integer or floating-point with adjustable decimal precision), and sparsity control. It supports structural constraints: identity, diagonal, symmetric, upper triangular, lower triangular, and row-stochastic forms. The generator uses uniform distribution. It does not produce cryptographically secure values. For reproducible experiments, record the output. The tool approximates true randomness via the JavaScript PRNG, which is adequate for prototyping but not for production cryptography or certified statistical sampling.

Formulas

For a dense random matrix A of size m × n, each entry is sampled uniformly:

x_ij = a + rand() ⋅ (b − a)

where a is the minimum bound, b is the maximum bound, and rand() returns a uniform variate in [0, 1). For integer output, the formula becomes:

x_ij = floor(rand() ⋅ (b − a + 1)) + a

For a symmetric matrix, only the upper triangle is generated, then mirrored: x_ji = x_ij for all i < j. For a row-stochastic matrix, each row is generated with rand() values, then normalized by dividing by the row sum S_i:

x_ij = r_ijS_i , S_i = n∑j=1 r_ij

The determinant (for square matrices ≤ 10 × 10) is computed via Gaussian elimination with partial pivoting. The product of the diagonal after elimination gives det(A), adjusted by the sign of the permutation count.

Variable legend: m = row count, n = column count, a = minimum value, b = maximum value, x_ij = element at row i column j, r_ij = raw random value before normalization, S_i = row sum for normalization.

Reference Data

Matrix Type	Structure Rule	Typical Use Case	Square Required	Symmetry
Random (Dense)	All x_ij filled independently	Monte Carlo simulation, test data	No	None
Identity	x_ij = 1 if i = j, else 0	Neutral element in multiplication	Yes	Symmetric
Diagonal	x_ij = 0 if i ≠ j	Eigenvalue matrices, scaling transforms	Yes	Symmetric
Symmetric	x_ij = x_ji	Covariance matrices, graph Laplacians	Yes	Symmetric
Upper Triangular	x_ij = 0 if i > j	LU decomposition, solving linear systems	Yes	None
Lower Triangular	x_ij = 0 if i < j	Cholesky factorization	Yes	None
Row-Stochastic	∑_j x_ij = 1, all x_ij ≥ 0	Markov chains, transition probabilities	No	Rarely
Sparse Random	Sparsity % of entries forced to 0	Large-scale FEM, network adjacency	No	None
Skew-Symmetric	x_ij = −x_ji, diagonal = 0	Angular velocity tensors, Lie algebras	Yes	Anti-symmetric
Toeplitz	Constant along each diagonal	Signal processing, convolution	No	Can be symmetric
Hilbert	x_ij = 1i + j − 1	Numerical analysis (ill-conditioned test)	No	Symmetric
Vandermonde	x_ij = a_i^j−1	Polynomial interpolation	No	None
Permutation	Exactly one 1 per row and column	Reordering, pivoting	Yes	Orthogonal

Frequently Asked Questions

Sparsity is applied after the structural constraint. For a symmetric matrix, a cell selected for zeroing at position (i, j) also forces (j, i) to zero, preserving symmetry. For triangular matrices, sparsity only affects the non-zero triangle. Identity matrices ignore sparsity entirely since their structure is fixed.

Gaussian elimination runs in O(n³) time. For a 100×100 matrix that is 1,000,000 operations which, while feasible, produces floating-point drift that makes the result unreliable without arbitrary-precision arithmetic. The 10×10 cap keeps computation under 1,000 operations and maintains at least 10 significant digits of accuracy with IEEE 754 doubles.

They are pseudorandom, produced by the JavaScript engine's Math.random() PRNG (typically xorshift128+ or similar). The distribution is uniform over [0, 1). This is sufficient for prototyping, test data generation, and educational purposes. For cryptographic or certified statistical applications, use a CSPRNG such as crypto.getRandomValues().

The tool automatically swaps the values so that the lower bound is always min and the upper bound is always max. A toast notification informs you of the correction. This prevents empty or undefined output ranges.

No. Identity, diagonal, symmetric, skew-symmetric, upper triangular, lower triangular, and permutation types require square matrices by mathematical definition. If you select one of these types, the column count automatically locks to equal the row count. Dense random, sparse, row-stochastic, Hilbert, Vandermonde, and Toeplitz types support rectangular dimensions.

Each row is filled with independent uniform random values in (0, 1), then every element is divided by the row's total sum. This guarantees each row sums to exactly 1.0 (within floating-point precision, approximately 15 significant digits). All entries remain non-negative, satisfying both conditions of a stochastic matrix.