Skip to content

Matrix Norms#

Since matrices of the same dimensionality form a vector space, it is possible to define norms on them.

Definition: Submultiplicativity

A norm \(||\cdot||: F^{n \times n} \to \mathbb{R}\) on the square matrices in \(F^{n \times n}\) is submultiplicative if

\[||AB|| \le ||A||\,||B||\]

for all \(A, B \in F^{n \times n}\).

Definition: Compatibility

We say that a norm \(||\cdot||_{F^n}: F^n \to \mathbb{R}\) on the vectors in \(F^{n}\) is compatible with a norm \(||\cdot||_{F^{n\times n}}: F^{n \times n} \to \mathbb{R}\) on the matrices in \(F^{n \times n}\) if

\[||A v||_{F^n} \le ||A||_{F^{n\times n}} \cdot ||v||_{F^n}\]

for all \(v \in F^n\) and all \(A \in F^{n \times n}\).

Induced Norms#

Theorem: Vector Norm induces Matrix Norm

If \(||\cdot||_{F^n}: F^n \to \mathbb{R}\) is a norm on the vectors in \(F^{n}\), then \(||\cdot||_{F^{n \times n}}: F^{n \times n} \to \mathbb{R}\) defined as the supremum

\[||A||_{F^{n \times n}} \overset{\text{def}}{=} \sup_{\mathbf{v} \in F^{n} \setminus \{\mathbf{0}\}}\frac{||A\mathbf{v}||_{F^n}}{||\mathbf{v}||_{F^n}}\]

for all matrices \(A \in F^{n \times n}\) is a norm on \(F^{n \times n}\).

Definition: Induced Norm

We call \(||A||_{F^{n \times n}}\) the norm induced by \(||\cdot||_{F^n}: F^n \to \mathbb{R}\) on \(F^{n \times n}\).

Example: The Norm Induced by \(l^1\)

The norm induced on the real matrices in \(\mathbb{R}^{n \times n}\) by the \(l^1\)-norm of \(\mathbb{R}^n\) is the maximum sum of absolute values entries in a single column:

\[||A||_1 = \max_{1 \le i \le n}\sum_{j = 1}^n |a_{ji}|\]
Example: The Norm Induced by \(l^{\infty}\)

The norm induced on the real matrices in \(\mathbb{R}^{n \times n}\) by the maximum norm of \(\mathbb{R}^n\) is the maximum sum of absolute values of entries in a single row:

\[||A||_{\infty} = \max_{1 \le i \le n} \sum_{j = 1}^n |a_{ij}|\]
Proof

TODO

Theorem: Submultiplicativity of Induced Norms

All induced norms are submultiplicative.

Proof

TODO

Theorem: Compatibility of Induced Norms

If \(||\cdot||_M\) is induced by \(||\cdot||_V\), then \(||\cdot||_M\) and \(||\cdot||_V\) are compatible.

Proof

TODO