Norms | Sahithyan's Notes

A norm is a real-valued function measuring the size of a vector or a matrix.

Vector Norm

For $\vec{x} = (x_1, x_2, \ldots, x_n)^T$ , 3 standard vector norms:

1-Norm

Sum of the absolute values of $\vec{x}$ ‘s components. Aka. taxicab norm, since it measures distance travelled along grid lines rather than in a straight line.

\lVert \vec{x} \rVert_1 = \sum_{i=1}^{n} |x_i|

2-Norm

Aka. Euclidean norm. Straight-line distance from $\vec{x}$ to the origin.

\lVert \vec{x} \rVert_2 = \left( \sum_{i=1}^{n} x_i^2 \right)^{1/2}

Infinity Norm

Largest absolute value among $\vec{x}$ ‘s components.

\lVert \vec{x} \rVert_\infty = \max_{1 \leq i \leq n} |x_i|

Equivalent Norms

2 norms $\lVert \cdot \rVert_{v1}, \lVert \cdot \rVert_{v2}$ on $\mathbb{R}^n$ are equivalent iff there exist positive constants $c_1, c_2$ such that for all $\vec{x} \in \mathbb{R}^n$ ,

c_1 \lVert \vec{x} \rVert_{v1} \leq \lVert \vec{x} \rVert_{v2} \leq c_2 \lVert \vec{x} \rVert_{v1}

$\lVert \cdot \rVert_1, \lVert \cdot \rVert_2, \lVert \cdot \rVert_\infty$ are equivalent on $\mathbb{R}^n$ .

\lVert \vec{x} \rVert_\infty \leq \lVert \vec{x} \rVert_2 \leq \sqrt{n} \, \lVert \vec{x} \rVert_\infty

\lVert \vec{x} \rVert_2 \leq \lVert \vec{x} \rVert_1 \leq \sqrt{n} \, \lVert \vec{x} \rVert_2

\lVert \vec{x} \rVert_\infty \leq \lVert \vec{x} \rVert_1 \leq n \, \lVert \vec{x} \rVert_\infty

Matrix Norm

A matrix norm is a norm on the set of $n \times n$ matrices, additionally satisfying submultiplicativity: $\lVert AB \rVert \leq \lVert A \rVert \, \lVert B \rVert$ for all $n \times n$ matrices $A, B$ .

Induced Matrix Norm

Aka. operator norm, subordinate norm.

The matrix norm induced by a vector norm $\lVert \cdot \rVert_v$ on $\mathbb{R}^n$

\lVert A \rVert_M = \sup_{\vec{x} \neq \vec{0}} \frac{\lVert A\vec{x} \rVert_v}{\lVert \vec{x} \rVert_v} = \sup_{\lVert \vec{x} \rVert_v = 1} \lVert A\vec{x} \rVert_v

For $A = (a_{ij})$ , $1 \leq i, j \leq n$ :

\lVert A \rVert_\infty = \max_{1 \leq i \leq n} \sum_{j=1}^{n} |a_{ij}| \quad \text{(maximum row sum)}

\lVert A \rVert_1 = \max_{1 \leq j \leq n} \sum_{i=1}^{n} |a_{ij}| \quad \text{(maximum column sum)}

Spectrum

The spectrum of $A$ , $\sigma(A)$ , is the set of all eigenvalues of $A$ .

\sigma(A) = \{ \lambda \; ; \; A\vec{v} = \lambda \vec{v}, \vec{v} \in \mathbb{R}^n, \lambda \text{ is an eigenvalue of } A \}

Spectral Radius

The spectral radius of $A$ , $\rho(A)$ , is the maximum absolute value among the eigenvalues of $A$ .

\rho(A) = \max \{ |\lambda| \; ; \; \lambda \in \sigma(A) \}

The matrix norm induced by the 2-norm

\lVert A \rVert_2 = \sqrt{\rho(A^TA)}