Matrix Norm Calculator

Background.

A matrix norm answers the question 'how big is this matrix?', and there is more than one defensible answer. This calculator computes five of them at once for any real matrix up to 8 by 8 — Frobenius, the 1-norm, the infinity-norm, the entrywise max, and the spectral 2-norm — and lets you choose which one sits in the headline while the other four stay visible underneath.

Type or paste your matrix with one row per line, entries separated by commas or spaces. Semicolons work as row separators and the bracketed form [[1,2],[3,4]] is accepted. The matrix does not need to be square: norms are defined for any shape, and a 1 by n matrix behaves exactly like a vector.

The four norms LAPACK's own documentation defines are the ones to reach for by default. The Frobenius norm is the square root of the sum of every squared entry — treat the matrix as one long vector and take its Euclidean length. The 1-norm is the largest absolute column sum. The infinity-norm is the largest absolute row sum. And the 2-norm, also called the spectral norm, is the largest singular value: the greatest factor by which the matrix can stretch any vector at all. Those last three are induced or operator norms, meaning each is defined as the maximum of the ratio of an output vector length to an input vector length under a particular way of measuring vectors, and that is what makes them the right tool for bounding errors.

Column versus row is the trap. The 1-norm sums down columns and the infinity-norm sums across rows — not the other way round — and getting them backwards is by far the most common error in matrix-norm code. This calculator prints both simultaneously and labels each one explicitly, so the two can never be silently confused. A second naming trap: the entrywise max norm, the largest single absolute entry, is sometimes loosely called 'the max norm', which invites confusion with the infinity-norm. It is offered here because people genuinely want it, and labelled as not an induced norm, because LAPACK explicitly cautions against substituting it for the four proper norms when deriving error bounds.

The spectral norm is the expensive one and the interesting one. Since the singular values of A are the positive square roots of the eigenvalues of A transpose times A, the calculator forms that symmetric matrix and finds its largest eigenvalue using the cyclic Jacobi rotation method, then takes the square root. Jacobi was chosen deliberately over the faster QR iteration because it requires only arithmetic and square roots — no trigonometry — so the whole computation stays in exact 60-digit decimal arithmetic rather than dropping to floating point. Rotations continue until the off-diagonal mass falls to a thousandth of a billionth of a billionth of a billionth of the diagonal mass, twenty orders of magnitude below what is displayed.

What is deliberately not here: the nuclear norm, the general Schatten p-norms, and the general induced p-norms for p other than 1, 2 and infinity. The first two need the entire singular-value spectrum rather than just the largest, and the induced p-norm for a general p has no closed form at all — it is an optimisation problem. Shipping an approximation of any of them under a confident label would be worse than leaving them out.

A useful cross-check you can perform on this page: the spectral norm is always less than or equal to the Frobenius norm, with equality exactly when the matrix has rank 1. If you see those two numbers coincide, your matrix is a single outer product in disguise.

What is matrix norm calculator?

A matrix norm assigns a single non-negative number to a matrix, measuring its magnitude in a way that respects scaling and the triangle inequality. The Frobenius norm treats the matrix as a flattened vector: ||A||_F is the square root of the sum of the squares of all entries. The induced (operator) norms instead measure how much the matrix can stretch a vector, each with respect to a chosen vector norm: ||A||_1 turns out to equal the maximum absolute column sum, ||A||_inf the maximum absolute row sum, and ||A||_2 the largest singular value of A. The entrywise max norm, max over i and j of |a_ij|, is a valid norm on the space of matrices but is not induced by any vector norm and is not submultiplicative, so it does not obey ||AB|| <= ||A|| ||B||. Singular values are the positive square roots of the eigenvalues of A-transpose-A, which is symmetric and positive semi-definite, so they are always real and non-negative and the largest of them is well-defined for any shape of matrix. Standard relationships worth knowing: ||A||_2 is at most ||A||_F, with equality exactly when A has rank 1; ||A||_2 is at most the square root of the product of ||A||_1 and ||A||_inf; and ||A^T||_1 equals ||A||_inf.

How to use this calculator.

1. 01Type your matrix into the box, one row per line, with entries separated by commas or spaces. It does not have to be square.
2. 02Pick which norm you want in the headline. Every other norm is still computed and displayed below, so the choice costs you nothing.
3. 03For a general 'size of this matrix' answer, the Frobenius norm is the usual default.
4. 04For an error bound or a stability argument, use one of the induced norms — 1, infinity or 2 — because those are the ones with the operator interpretation.
5. 05Read the 1-norm as the largest column total and the infinity-norm as the largest row total. They are not interchangeable and this page labels both.
6. 06Use the spectral norm when you need the true worst-case stretching factor of A. It is the largest singular value.
7. 07Compare the spectral norm with the Frobenius norm: if they are equal, your matrix has rank 1.

The formula.

Four of the five norms are direct sums over the entries. The Frobenius norm squares every entry, adds them all, and takes the square root: for A = [[1,2],[3,4]] that is the square root of 1 + 4 + 9 + 16 = 30, which is 5.4772255751. The 1-norm adds the absolute values down each column and takes the largest of those column totals: columns give 1 + 3 = 4 and 2 + 4 = 6, so the 1-norm is 6. The infinity-norm does the same across rows: 1 + 2 = 3 and 3 + 4 = 7, so the infinity-norm is 7. The entrywise max norm simply scans for the largest absolute entry, which is 4. The spectral norm is the one that needs real work. It is defined as the largest singular value of A, and the singular values are the positive square roots of the eigenvalues of A-transpose-A, so the calculator first forms that Gram matrix — for the example, [[10,14],[14,20]] — and then finds its largest eigenvalue. Because A-transpose-A is symmetric and positive semi-definite, the cyclic Jacobi rotation method applies: for each off-diagonal pair (p,q) it computes tau = (S_qq - S_pp)/(2 S_pq), takes the smaller root t = sign(tau)/(|tau| + sqrt(tau squared + 1)) of the quadratic t squared + 2 tau t - 1 = 0, sets c = 1/sqrt(t squared + 1) and s = t times c, and applies the rotation that zeroes that off-diagonal entry. Sweeping repeatedly drives all off-diagonal entries to zero and leaves the eigenvalues on the diagonal. Here the characteristic polynomial of [[10,14],[14,20]] is lambda squared minus 30 lambda plus 4, whose larger root is (30 + sqrt(884))/2 = 29.8660687473, and the square root of that is 5.4649857042 — the spectral norm. Every step runs at 60 significant digits with rounding applied once, at the very end, to 10 decimal places.

A worked example.

Frequently asked questions.

References& sources.

1. [1]LAPACK Users' Guide, 3rd ed. (SIAM, 1999), 'How to Measure Errors'. Gives the four norms verbatim: ||A||_1 = max_j sum_i |a_ij| (max column sum), ||A||_inf = max_i sum_j |a_ij| (max row sum), ||A||_F = (sum |a_ij|^2)^(1/2), and ||A||_2 as the largest singular value — and warns against the naive maximum-element approach. Retrieved 2026-07-29.
2. [2]Encyclopedia of Mathematics (EMS Press), 'Singular value decomposition'. States that the singular values are 'the positive square roots of the eigenvalues of AA* (equivalently, of A*A)', which is the identity used to compute the spectral norm here. Retrieved 2026-07-29.
3. [3]LAPACK Users' Guide, 3rd ed. (SIAM, 1999), section 2.4 'The Singular Value Decomposition'. Establishes the ordering sigma_1 >= ... >= sigma_r >= 0 that makes 'the largest singular value' well-defined. Retrieved 2026-07-29.
4. [4]Golub, G. H. & Van Loan, C. F. (2013). Matrix Computations, 4th ed., Johns Hopkins University Press — section 2.3 for matrix norms and their inequalities, and algorithms 8.4.1 / 8.4.2 for the cyclic Jacobi eigenvalue method and its tau/t/c/s rotation. Print reference; publisher page linked.