A function that assigns a nonnegative size to a matrix consistent with vector norms and the boundedness of the associated linear map; common examples include operator (induced) norms, the Frobenius norm, the 1-norm and infinity-norm.

Principle

Matrix norms respect submultiplicativity (||AB|| ≤ ||A||·||B||) when compatible with vector norms, and induced norms are defined by suprema ||A|| = sup_{x≠0} ||Ax||/||x||; choice of norm encodes what aspect of 'size' is most relevant (worst-case amplification, entrywise energy, column-sum, etc.).

Demonstration

The spectral (operator) 2-norm equals the largest singular value and measures worst-case amplification of unit vectors. The 1-norm equals the maximum absolute column sum and is useful for bounding columnwise effects. The Frobenius norm equals the square root of the sum of squares of entries and is the l2 norm of singular values, useful for elementwise error measurement.

Misapplication

Using an entrywise norm like Frobenius when the problem requires an induced/operator norm to control worst-case amplification (e.g., stability of iterative methods), or mixing norms without converting error measures consistently across analyses.

Consequence

A well-chosen matrix norm yields meaningful bounds on error propagation, convergence rates, and stability of linear algorithms; it allows comparing operators and designing preconditioners or normalization strategies aligned with the chosen measure of size.

Reversal

Replacing a full norm by a seminorm or ad hoc scalar that ignores submultiplicativity or compatibility with vector norms; the reversed notion loses the ability to compose bounds and reason about operator behaviour.

Boundary

Norms must be specified (induced vs entrywise) and are defined on linear operators between normed spaces; in infinite dimensions boundedness is required. Not all useful measures of matrix effect are norms (e.g., numerical rank is not a norm).

Semantic Tension

Tension exists between induced/operator norms (which capture worst-case vector amplification) and entrywise norms (which capture aggregate energy); choosing one over the other changes which properties are controlled and can lead to conflicting intuitions about 'size'.

Synthesis

A matrix norm is a norm on linear operators tied to a choice of vector norms whose algebraic properties (submultiplicativity, induced-sense) determine which aspects of operator size—worst-case amplification, energy, or entrywise magnitude—are being measured.