1. Vector
- L0 norm: the number of non-zero elements in the vector, also called 0 norm
- L1 norm: It is the sum of absolute values, also called the norm or 1 norm.
- L2 norm: module in the usual sense, also called 2 norm
- p norm: that is, pp of the absolute value of the vector element1 / p 1/pof the sum of p powers1/ p power
- ∞ \infty∞ norm: take the maximum value of the vector
- − ∞ -\infty− ∞ norm: take the minimum value of the vector
Suppose there is a vector x = ( x 1 , x 2 , … , xn ) T x = (x_1,x_2,\dots,x_n)^Tx=(x1,x2,…,xn)T , the norm of the vector is:
∣ ∣ x ∣ ∣ 1 = ∣ x 1 ∣ + ∣ x 2 ∣ + ⋯ + ∣ xn ∣ ||x||_1 = |x_1| + |x_2| + \dots + |x_n |∣∣x∣∣1=∣x1∣+∣x2∣+⋯+∣xn∣
∣ ∣ x ∣ ∣ 2 = ( ∣ x 1 ∣ + ∣ x 2 ∣ + ⋯ + ∣ x n ∣ ) 1 2 ||x||_2 = (|x_1| + |x_2| + \dots + |x_n|)^{\frac{1}{2}} ∣∣x∣∣2=(∣x1∣+∣x2∣+⋯+∣xn∣)21
∣ ∣ x ∣ ∣ p = ( ∣ x 1 ∣ + ∣ x 2 ∣ + ⋯ + ∣ x n ∣ ) 1 p ||x||_p = (|x_1| + |x_2| + \dots + |x_n|)^{\frac{1}{p}} ∣∣x∣∣p=(∣x1∣+∣x2∣+⋯+∣xn∣)p1
∣ ∣ x ∣ ∣ ∞ = max { ∣ x 1 ∣ + ∣ x 2 ∣ + ⋯ + ∣ x n ∣ } ||x||_{\infty} = \max\{|x_1| + |x_2| + \dots + |x_n|\} ∣∣x∣∣∞=max{ ∣x1∣+∣x2∣+⋯+∣xn∣}
∣ ∣ x ∣ ∣ − ∞ = min { ∣ x 1 ∣ + ∣ x 2 ∣ + ⋯ + ∣ x n ∣ } ||x||_{-\infty} = \min\{|x_1| + |x_2| + \dots + |x_n|\} ∣∣x∣∣−∞=min{ ∣x1∣+∣x2∣+⋯+∣xn∣}
2. Matrix
There are mainly three types of matrix norms. Remember, the three types of norms are different! !
2.1. Induced norm
Also called operator norm, it is defined that matrix A is located in the matrix space R m × n \mathbb{R}^{m\times n}Rm×n 上
∣ ∣ x ∣ ∣ = max { ∣ ∣ A x ∣ ∣ ; x ∈ R n , ∣ ∣ x ∣ ∣ = 1 } = m a x { ∣ ∣ A x ∣ ∣ ∣ ∣ x ∣ ∣ , x ∈ R n , x ≠ 0 } ||x||= \max\{||Ax||;x\in \mathbb{R}^n,||x||=1 \} = max\left\{ \frac{||Ax||}{||x||}, x\in \mathbb{R}^n,x\ne 0 \right\} ∣∣x∣∣=max{
∣∣Ax∣∣;x∈Rn,∣∣x∣∣=1}=max{
∣∣x∣∣∣∣Ax∣∣,x∈Rn,x=0 }
The commonly used induced norm is the p norm, also calledL p L_pLp 范数
∣ ∣ A ∣ ∣ p = m a x { ∣ ∣ A x ∣ ∣ p ∣ ∣ x ∣ ∣ p , x ≠ 0 } ||A||_p = max\left\{ \frac{||Ax||_p}{||x||_p} ,x\ne 0 \right\} ∣∣A∣∣p=max{
∣∣x∣∣p∣∣Ax∣∣p,x=0 }
0 norm: the number of non-zero elements in the matrix.
1 norm is also called column norm
∣ ∣ A ∣ ∣ 1 = max ∑ i = 1 m ∣ aij ∣ , 1 ≤ j ≤ n ||A||_1 = \max \sum_{i=1}^m |a_{ij}|,1 \le j \le n∣∣A∣∣1=maxi=1∑m∣aij∣,1≤j≤The n-
infinity norm is also called the row norm
∣ ∣ A ∣ ∣ ∞ = max ∑ j = 1 n ∣ aij ∣ , 1 ≤ i ≤ n ||A||_\infty = \max \sum_{j=1 }^n |a_{ij}|,1 \le i \le n∣∣A∣∣∞=maxj=1∑n∣aij∣,1≤i≤The n
2 norm is also called the spectral norm. The spectral norm of the matrix is the maximum singular value of the matrix
∣ ∣ A ∣ ∣ spec = ∣ ∣ A ∣ ∣ 2 = σ max ( A ) = λ max ( ATA ) ||A ||_{spec} = ||A||_{2} = \sigma_{max}(A) = \sqrt{\lambda_{max}(A^TA)}∣∣A∣∣spec=∣∣A∣∣2=pmax(A)=lmax(AT A)
σ m a x \sigma_{max} pmaxIt means finding the maximum singular value of the matrix, spec means spectrum.
2.2. Element norm
Let m × nm \times nm×The n matrix is first arranged into amn × 1 mn \times 1mn×1 vector, and then use the norm definition of the vector, that is, the norm of the matrix
∣ ∣ A ∣ ∣ P = ( ∑ i = 1 m ∑ j = 1 n ∣ aij ∣ p ) 1 p ||A||_P = \ left( \sum_{i=1}^m\sum_{j=1}^n |a_{ij}|^p \right)^\frac{1}{p}∣∣A∣∣P=(i=1∑mj=1∑n∣aij∣p)p1
1 norm is also called sum norm or L1 norm
∣ ∣ A ∣ ∣ 1 = ∑ i = 1 m ∑ j = 1 n ∣ aij ∣ ||A||_1 = \sum_{i=1}^m\ sum_{j=1}^n |a_{ij}|∣∣A∣∣1=i=1∑mj=1∑n∣aij∣
2 norm is also called Forbenius norm∣
∣ A ∣ ∣ F = ∣ ∣ A ∣ ∣ 2 = ( ∑ i = 1 m ∑ j = 1 n ∣ aij ∣ 2 ) 1 2 = tr ( AHA ) ||A ||_F = ||A||_2 = \left( \sum_{i=1}^m\sum_{j=1}^n |a_{ij}|^2 \right)^\frac{1}{ 2} = \sqrt{tr({A^HA})}∣∣A∣∣F=∣∣A∣∣2=(i=1∑mj=1∑n∣aij∣2)21=tr(AHA)
∞ \infty∞ norm is also called the maximum norm
∣ ∣ A ∣ ∣ p = max { ∣ aij ∣ } ||A||_p = \max\{ |a_{ij}| \}∣∣A∣∣p=max{
∣aij∣}
2.3. Schatten norm
Using the norm defined by the singular values of the matrix, let the singular values of the matrix form a vector σ = [ σ 1 σ 2 … σ k ] \sigma = \begin{bmatrix} \sigma_1 & \sigma_2 & \dots & \sigma_k \end {bmatrix}p=[p1p2…pk] , k = min ( m , n ) k=\min(m,n) k=min(m,n)
p
∣ ∣ ∣ A ∣ ∣ p = ∣ ∣ σ ∣ ∣ p = ( ∑ i = 1 k σ ip ) 1 p ||A||_p = ||\sigma||_p = (\sum_{i =1}^k \sigma_i^{p})^\frac{1}{p}∣∣A∣∣p=∣∣σ∣∣p=(i=1∑kpip)p1
p=1, also called the nuclear norm, is the sum of the singular values of the matrix
∣ ∣ A ∣ ∣ 1 = ∑ i = 1 k σ i = tr ( AHA ) ||A||_1 =\sum_{i=1} ^k \sigma_i = tr(\sqrt{A^HA})∣∣A∣∣1=i=1∑kpi=tr(AHA)
p=2, Schattern norm is equivalent to Frobenius
∣ ∣ A ∣ ∣ 1 = ( ∑ i = 1 k σ i 2 ) 1 2 = t r ( A H A ) = ( ∑ i = 1 m ∑ j = 1 n ∣ a i j ∣ 2 ) 1 2 ||A||_1 =(\sum_{i=1}^k \sigma_i^{2})^\frac{1}{2} = \sqrt{tr(A^HA)} = \left(\sum_{i=1}^m\sum_{j=1}^n |a_{ij}|^2 \right)^\frac{1}{2} ∣∣A∣∣1=(i=1∑kpi2)21=tr(AHA)=(i=1∑mj=1∑n∣aij∣2)21
p= ∞ \infty ∞ , the Schattern norm is the same as the spectral norm
∣ ∣ A ∣ ∣ ∞ = σ m a x ( A ) ||A||_\infty = \sigma_{max}(A) ∣∣A∣∣∞=pmax(A)