1. 向量的范数
定义 如果向量(一组数) x ⃗ ∈ R n \vec{x}\in \mathbb{R}^n x∈Rn 的某个实函数 N ( x ⃗ ) = ∣ ∣ x ⃗ ∣ ∣ N(\vec{x})=||\vec{x}|| N(x)=∣∣x∣∣,满足如下条件:
(1) ∣ ∣ x ⃗ ∣ ∣ ≥ 0 ||\vec{x}||\ge0 ∣∣x∣∣≥0,当且仅当 x ⃗ = 0 ⃗ \vec{x}=\vec{0} x=0 时取等号(正定条件);
(2) ∀ α ∈ R , ∣ ∣ α x ⃗ ∣ ∣ = ∣ α ∣ ⋅ ∣ ∣ x ⃗ ∣ ∣ \forall\ \alpha\in\mathbb{R},||\alpha\vec{x}||=|\alpha|\cdot||\vec{x}|| ∀ α∈R,∣∣αx∣∣=∣α∣⋅∣∣x∣∣;
(3) ∣ ∣ x ⃗ + y ⃗ ∣ ∣ ≤ ∣ ∣ x ⃗ ∣ ∣ + ∣ ∣ y ⃗ ∣ ∣ ||\vec{x}+\vec{y}||\le||\vec{x}||+||\vec{y}|| ∣∣x+y∣∣≤∣∣x∣∣+∣∣y∣∣(三角不等式);
则称 N ( x ⃗ ) N(\vec{x}) N(x) 为 R n \mathbb{R}^n Rn 上的一个向量范数。
根据向量范数的三角不等式有:
∣
∣
x
⃗
−
y
⃗
∣
∣
+
∣
∣
y
⃗
∣
∣
≥
∣
∣
x
⃗
∣
∣
⟹
∣
∣
x
⃗
−
y
⃗
∣
∣
≥
∣
∣
x
⃗
∣
∣
−
∣
∣
y
⃗
∣
∣
∣
∣
x
⃗
−
y
⃗
∣
∣
+
∣
∣
x
⃗
∣
∣
=
∣
∣
y
⃗
−
x
⃗
∣
∣
+
∣
∣
x
⃗
∣
∣
≥
∣
∣
y
⃗
∣
∣
⟹
∣
∣
x
⃗
−
y
⃗
∣
∣
≥
∣
∣
y
⃗
∣
∣
−
∣
∣
x
⃗
∣
∣
\begin{aligned} &||\vec{x}-\vec{y}||+||\vec{y}||\ge||\vec{x}||\Longrightarrow ||\vec{x}-\vec{y}||\ge||\vec{x}||-||\vec{y}|| \\\\ &||\vec{x}-\vec{y}||+||\vec{x}||=||\vec{y}-\vec{x}||+||\vec{x}||\ge||\vec{y}||\Longrightarrow ||\vec{x}-\vec{y}||\ge||\vec{y}||-||\vec{x}|| \end{aligned}
∣∣x−y∣∣+∣∣y∣∣≥∣∣x∣∣⟹∣∣x−y∣∣≥∣∣x∣∣−∣∣y∣∣∣∣x−y∣∣+∣∣x∣∣=∣∣y−x∣∣+∣∣x∣∣≥∣∣y∣∣⟹∣∣x−y∣∣≥∣∣y∣∣−∣∣x∣∣
结合上述两式与三角不等式,便有:
∣
∣
∣
x
⃗
∣
∣
−
∣
∣
y
⃗
∣
∣
∣
≤
∣
∣
x
⃗
−
y
⃗
∣
∣
≤
∣
∣
x
⃗
∣
∣
+
∣
∣
y
⃗
∣
∣
|\ ||\vec{x}||-||\vec{y}||\ |\le||\vec{x}-\vec{y}||\le||\vec{x}||+||\vec{y}||
∣ ∣∣x∣∣−∣∣y∣∣ ∣≤∣∣x−y∣∣≤∣∣x∣∣+∣∣y∣∣
2. 常用的向量范数
-
向量的 ∞ \infty ∞-范数(最大范数):
∣ ∣ x ⃗ ∣ ∣ ∞ ≜ max 1 ≤ i ≤ n ∣ x i ∣ ||\vec{x}||_{\infty}\triangleq \max\limits_{1\le i\le n}|x_i| ∣∣x∣∣∞≜1≤i≤nmax∣xi∣
条件 (1):
∣ ∣ x ⃗ ∣ ∣ ∞ = max 1 ≤ i ≤ n ∣ x i ∣ ≥ 0 ||\vec{x}||_{\infty}=\max\limits_{1\le i\le n}|x_i|\ge0 ∣∣x∣∣∞=1≤i≤nmax∣xi∣≥0
上述不等式取等号时,当且仅当
max 1 ≤ i ≤ n ∣ x i ∣ = 0 ⟺ x ⃗ = 0 \max\limits_{1\le i\le n}|x_i|=0\Longleftrightarrow \vec{x}=0 1≤i≤nmax∣xi∣=0⟺x=0
条件 (2):
∣ ∣ α x ⃗ ∣ ∣ ∞ = max 1 ≤ i ≤ n ∣ α x i ∣ = ∣ α ∣ max 1 ≤ i ≤ n ∣ x i ∣ = ∣ α ∣ ⋅ ∣ ∣ x ⃗ ∣ ∣ ∞ ||\alpha\vec{x}||_{\infty}=\max\limits_{1\le i\le n}|\alpha x_i|=|\alpha|\max\limits_{1\le i\le n}|x_i|=|\alpha|\cdot||\vec{x}||_{\infty} ∣∣αx∣∣∞=1≤i≤nmax∣αxi∣=∣α∣1≤i≤nmax∣xi∣=∣α∣⋅∣∣x∣∣∞
条件 (3):
∣ ∣ x ⃗ + y ⃗ ∣ ∣ ∞ = max 1 ≤ i ≤ n ∣ x i + y i ∣ : = ∣ x k + y k ∣ ≤ ∣ x k ∣ + ∣ y k ∣ ≤ max 1 ≤ i ≤ n ∣ x i ∣ + max 1 ≤ i ≤ n ∣ y i ∣ = ∣ ∣ x ⃗ ∣ ∣ ∞ + ∣ ∣ y ⃗ ∣ ∣ ∞ ||\vec{x}+\vec{y}||_{\infty}=\max\limits_{1\le i\le n}|x_i+y_i|:=|x_k+y_k|\le|x_k|+|y_k|\le\max\limits_{1\le i\le n}|x_i|+\max\limits_{1\le i\le n}|y_i|=||\vec{x}||_{\infty}+||\vec{y}||_{\infty} ∣∣x+y∣∣∞=1≤i≤nmax∣xi+yi∣:=∣xk+yk∣≤∣xk∣+∣yk∣≤1≤i≤nmax∣xi∣+1≤i≤nmax∣yi∣=∣∣x∣∣∞+∣∣y∣∣∞ -
向量的 1-范数:
∣ ∣ x ⃗ ∣ ∣ 1 ≜ ∑ i = 1 n ∣ x i ∣ ||\vec{x}||_{1}\triangleq \sum_{i=1}^n|x_i| ∣∣x∣∣1≜i=1∑n∣xi∣
条件 (1):
∣ ∣ x ⃗ ∣ ∣ 1 = ∑ i = 1 n ∣ x i ∣ ≥ 0 ||\vec{x}||_{1}=\sum_{i=1}^n|x_i|\ge0 ∣∣x∣∣1=i=1∑n∣xi∣≥0
上述不等式取等号时,当且仅当
∣ x i ∣ = 0 ( i = 1 , 2 , … , n ) ⟺ x ⃗ = 0 |x_i|=0\ (i=1,2,\dots,n)\Longleftrightarrow \vec{x}=0 ∣xi∣=0 (i=1,2,…,n)⟺x=0
条件 (2):
∣ ∣ α x ⃗ ∣ ∣ 1 = ∑ i = 1 n ∣ α x i ∣ = ∣ α ∣ ∑ i = 1 n ∣ x i ∣ = ∣ α ∣ ⋅ ∣ ∣ x ⃗ ∣ ∣ 1 ||\alpha\vec{x}||_{1}=\sum_{i=1}^n|\alpha x_i|=|\alpha|\sum_{i=1}^n|x_i|=|\alpha|\cdot||\vec{x}||_{1} ∣∣αx∣∣1=i=1∑n∣αxi∣=∣α∣i=1∑n∣xi∣=∣α∣⋅∣∣x∣∣1
条件 (3):
∣ ∣ x ⃗ + y ⃗ ∣ ∣ 1 = ∑ i = 1 n ∣ x i + y i ∣ ≤ ∑ i = 1 n ( ∣ x i ∣ + ∣ y i ∣ ) = ∣ ∣ x ⃗ ∣ ∣ 1 + ∣ ∣ y ⃗ ∣ ∣ 1 ||\vec{x}+\vec{y}||_{1}=\sum_{i=1}^n|x_i+y_i|\le\sum_{i=1}^n(|x_i|+|y_i|)=||\vec{x}||_{1}+||\vec{y}||_{1} ∣∣x+y∣∣1=i=1∑n∣xi+yi∣≤i=1∑n(∣xi∣+∣yi∣)=∣∣x∣∣1+∣∣y∣∣1 -
向量的 2-范数(欧氏范数):
∣ ∣ x ⃗ ∣ ∣ 2 ≜ ∣ x ⃗ ∣ = ∑ i = 1 n x i 2 ||\vec{x}||_{2}\triangleq |\vec{x}|=\sqrt{\sum_{i=1}^n x_i^2} ∣∣x∣∣2≜∣x∣=i=1∑nxi2
条件 (1):
∣ ∣ x ⃗ ∣ ∣ 2 = ∣ x ⃗ ∣ ≥ 0 ||\vec{x}||_{2}=|\vec{x}|\ge0 ∣∣x∣∣2=∣x∣≥0
上述不等式取等号时,当且仅当
x ⃗ = 0 \vec{x}=0 x=0
条件 (2):
∣ ∣ α x ⃗ ∣ ∣ 2 = ∣ α x ⃗ ∣ = ∣ α ∣ ⋅ ∣ x ⃗ ∣ = ∣ α ∣ ⋅ ∣ ∣ x ⃗ ∣ ∣ 2 ||\alpha\vec{x}||_{2}=|\alpha \vec{x}|=|\alpha|\cdot|\vec{x}|=|\alpha|\cdot||\vec{x}||_{2} ∣∣αx∣∣2=∣αx∣=∣α∣⋅∣x∣=∣α∣⋅∣∣x∣∣2
条件 (3):
∣ ∣ x ⃗ + y ⃗ ∣ ∣ 2 = ∣ x ⃗ + y ⃗ ∣ ≤ ∣ x ⃗ ∣ + ∣ y ⃗ ∣ = ∣ ∣ x ⃗ ∣ ∣ 2 + ∣ ∣ y ⃗ ∣ ∣ 2 ||\vec{x}+\vec{y}||_{2}=|\vec{x}+\vec{y}|\le|\vec{x}|+|\vec{y}|=||\vec{x}||_{2}+||\vec{y}||_{2} ∣∣x+y∣∣2=∣x+y∣≤∣x∣+∣y∣=∣∣x∣∣2+∣∣y∣∣2 -
向量的 p-范数:
∣ ∣ x ⃗ ∣ ∣ p ≜ ( ∑ i = 1 n ∣ x i ∣ p ) 1 p , p ∈ [ 1 , ∞ ) ||\vec{x}||_{p}\triangleq\left(\sum_{i=1}^n |x_i|^p\right)^{\frac{1}{p}},p\in[1,\infty) ∣∣x∣∣p≜(i=1∑n∣xi∣p)p1,p∈[1,∞)
条件 (1):
∣ ∣ x ⃗ ∣ ∣ p = ( ∑ i = 1 n ∣ x i ∣ p ) 1 p ≥ 0 ||\vec{x}||_{p}=\left(\sum_{i=1}^n |x_i|^p\right)^{\frac{1}{p}}\ge0 ∣∣x∣∣p=(i=1∑n∣xi∣p)p1≥0
上述不等式取等号时,当且仅当
∣ x i ∣ = 0 ( i = 1 , 2 , … , n ) ⟺ x ⃗ = 0 |x_i|=0\ (i=1,2,\dots,n)\Longleftrightarrow \vec{x}=0 ∣xi∣=0 (i=1,2,…,n)⟺x=0
条件 (2):
∣ ∣ α x ⃗ ∣ ∣ p = ( ∑ i = 1 n ( ∣ α ∣ ⋅ ∣ x i ∣ ) p ) 1 p = ∣ α ∣ ( ∑ i = 1 n ∣ x i ∣ p ) 1 p = ∣ α ∣ ⋅ ∣ ∣ x ⃗ ∣ ∣ p ||\alpha\vec{x}||_{p}=\left(\sum_{i=1}^n (|\alpha|\cdot|x_i|)^p\right)^{\frac{1}{p}}=|\alpha|\left(\sum_{i=1}^n |x_i|^p\right)^{\frac{1}{p}}=|\alpha|\cdot||\vec{x}||_{p} ∣∣αx∣∣p=(i=1∑n(∣α∣⋅∣xi∣)p)p1=∣α∣(i=1∑n∣xi∣p)p1=∣α∣⋅∣∣x∣∣p
条件 (3):
∑ i = 1 n ∣ x i + y i ∣ p = ∑ i = 1 n ∣ x i + y i ∣ p − 1 ∣ x i + y i ∣ ≤ ∑ i = 1 n ∣ x i + y i ∣ p − 1 ( ∣ x i ∣ + ∣ y i ∣ ) = ∑ i = 1 n ∣ x i + y i ∣ p − 1 ∣ x i ∣ + ∑ i = 1 n ∣ x i + y i ∣ p − 1 ∣ y i ∣ ≤ [ ∑ i = 1 n ∣ x i + y i ∣ q ( p − 1 ) ] 1 q ( ∑ i = 1 n ∣ x i ∣ p ) 1 p + [ ∑ i = 1 n ∣ x i + y i ∣ q ( p − 1 ) ] 1 q ( ∑ i = 1 n ∣ y i ∣ p ) 1 p ( p q = p q − q ; p , q > 1 ) = [ ( ∑ i = 1 n ∣ x i ∣ p ) 1 p + ( ∑ i = 1 n ∣ y i ∣ p ) 1 p ] [ ∑ i = 1 n ∣ x i + y i ∣ q ( p − 1 ) ] 1 q = [ ( ∑ i = 1 n ∣ x i ∣ p ) 1 p + ( ∑ i = 1 n ∣ y i ∣ p ) 1 p ] [ ∑ i = 1 n ∣ x i + y i ∣ p ] 1 q \begin{aligned} & \quad\sum_{i=1}^n |x_i+y_i|^p \\\\ & =\sum_{i=1}^n |x_i+y_i|^{p-1}|x_i+y_i| \\\\ & \le\sum_{i=1}^n |x_i+y_i|^{p-1}(|x_i|+|y_i|) \\\\ & =\sum_{i=1}^n |x_i+y_i|^{p-1}|x_i|+\sum_{i=1}^n |x_i+y_i|^{p-1}|y_i| \\\\ & \le\left[\sum_{i=1}^n |x_i+y_i|^{q(p-1)}\right]^{\frac{1}{q}}\left(\sum_{i=1}^n |x_i|^p\right)^{\frac{1}{p}}+\left[\sum_{i=1}^n |x_i+y_i|^{q(p-1})\right]^{\frac{1}{q}}\left(\sum_{i=1}^n |y_i|^p\right)^{\frac{1}{p}} (pq=pq-q;p,q>1) \\\\ & =\left[\left(\sum_{i=1}^n |x_i|^p\right)^{\frac{1}{p}}+\left(\sum_{i=1}^n |y_i|^p\right)^{\frac{1}{p}}\right]\left[\sum_{i=1}^n |x_i+y_i|^{q(p-1)}\right]^{\frac{1}{q}} \\\\ & =\left[\left(\sum_{i=1}^n |x_i|^p\right)^{\frac{1}{p}}+\left(\sum_{i=1}^n |y_i|^p\right)^{\frac{1}{p}}\right]\left[\sum_{i=1}^n |x_i+y_i|^{p}\right]^{\frac{1}{q}} \end{aligned} i=1∑n∣xi+yi∣p=i=1∑n∣xi+yi∣p−1∣xi+yi∣≤i=1∑n∣xi+yi∣p−1(∣xi∣+∣yi∣)=i=1∑n∣xi+yi∣p−1∣xi∣+i=1∑n∣xi+yi∣p−1∣yi∣≤[i=1∑n∣xi+yi∣q(p−1)]q1(i=1∑n∣xi∣p)p1+[i=1∑n∣xi+yi∣q(p−1)]q1(i=1∑n∣yi∣p)p1(pq=pq−q;p,q>1)=⎣ ⎡(i=1∑n∣xi∣p)p1+(i=1∑n∣yi∣p)p1⎦ ⎤[i=1∑n∣xi+yi∣q(p−1)]q1=⎣ ⎡(i=1∑n∣xi∣p)p1+(i=1∑n∣yi∣p)p1⎦ ⎤[i=1∑n∣xi+yi∣p]q1
上述证明过程前后应用了绝对值不等式与Holder不等式,进一步:
[ ∑ i = 1 n ∣ x i + y i ∣ p ] 1 − 1 q = [ ∑ i = 1 n ∣ x i + y i ∣ p ] 1 p ≤ [ ( ∑ i = 1 n ∣ x i ∣ p ) 1 p + ( ∑ i = 1 n ∣ y i ∣ p ) 1 p ] ( p > 1 ) \left[\sum_{i=1}^n |x_i+y_i|^{p}\right]^{1-\frac{1}{q}} =\left[\sum_{i=1}^n |x_i+y_i|^{p}\right]^{\frac{1}{p}} \le\left[\left(\sum_{i=1}^n |x_i|^p\right)^{\frac{1}{p}}+\left(\sum_{i=1}^n |y_i|^p\right)^{\frac{1}{p}}\right](p>1) [i=1∑n∣xi+yi∣p]1−q1=[i=1∑n∣xi+yi∣p]p1≤⎣ ⎡(i=1∑n∣xi∣p)p1+(i=1∑n∣yi∣p)p1⎦ ⎤(p>1)
当 p = 1 p=1 p=1 时由1-范数的三角不等式知上述不等式同样成立,综上得证Minkowski不等式(闵可夫斯基不等式):
[ ∑ i = 1 n ∣ x i + y i ∣ p ] 1 p ≤ [ ( ∑ i = 1 n ∣ x i ∣ p ) 1 p + ( ∑ i = 1 n ∣ y i ∣ p ) 1 p ] ( p ≥ 1 ) \left[\sum_{i=1}^n |x_i+y_i|^{p}\right]^{\frac{1}{p}} \le\left[\left(\sum_{i=1}^n |x_i|^p\right)^{\frac{1}{p}}+\left(\sum_{i=1}^n |y_i|^p\right)^{\frac{1}{p}}\right](p\ge1) [i=1∑n∣xi+yi∣p]p1≤⎣ ⎡(i=1∑n∣xi∣p)p1+(i=1∑n∣yi∣p)p1⎦ ⎤(p≥1)
即,
∣ ∣ x ⃗ + y ⃗ ∣ ∣ p ≤ ∣ ∣ x ⃗ ∣ ∣ p + ∣ ∣ y ⃗ ∣ ∣ p ||\vec{x}+\vec{y}||_p\le||\vec{x}||_p+||\vec{y}||_p ∣∣x+y∣∣p≤∣∣x∣∣p+∣∣y∣∣p
3. 最大范数、1-范数、欧氏范数与p-范数的关系
四者之间的关系:最大范数、1-范数、欧氏范数是p-范数的特殊情况,1-范数、欧氏范数与p-范数的关系根据定义是一目了然的,现在说明
p
=
∞
p=\infty
p=∞ 与最大范数之间的关系:
∣
∣
x
⃗
∣
∣
∞
=
lim
p
→
∞
(
∑
i
=
1
n
∣
x
i
∣
p
)
1
p
||\vec{x}||_{\infty}=\lim_{p\rightarrow\infty}\left(\sum_{i=1}^n |x_i|^p\right)^{\frac{1}{p}}
∣∣x∣∣∞=p→∞lim(i=1∑n∣xi∣p)p1
采用夹逼定理求上述极限,由于
0
≤
max
1
≤
i
≤
n
∣
x
i
∣
p
≤
∑
i
=
1
n
∣
x
i
∣
p
≤
n
(
max
1
≤
i
≤
n
∣
x
i
∣
p
)
0\le\max\limits_{1\le i\le n}|x_i|^p \le\sum_{i=1}^n |x_i|^p\le n\left(\max\limits_{1\le i\le n}|x_i|^p\right)
0≤1≤i≤nmax∣xi∣p≤i=1∑n∣xi∣p≤n(1≤i≤nmax∣xi∣p)
又
lim
p
→
∞
(
max
1
≤
i
≤
n
∣
x
i
∣
p
)
1
p
=
max
1
≤
i
≤
n
∣
x
i
∣
lim
p
→
∞
[
n
(
max
1
≤
i
≤
n
∣
x
i
∣
p
)
]
1
p
=
max
1
≤
i
≤
n
∣
x
i
∣
lim
p
→
∞
n
1
p
=
max
1
≤
i
≤
n
∣
x
i
∣
\begin{aligned} &\lim_{p\rightarrow\infty}\left(\max\limits_{1\le i\le n}|x_i|^p\right)^{\frac{1}{p}}=\max\limits_{1\le i\le n}|x_i|\\\\ &\lim_{p\rightarrow\infty}\left[n\left(\max\limits_{1\le i\le n}|x_i|^p\right)\right]^{\frac{1}{p}}=\max\limits_{1\le i\le n}|x_i|\lim_{p\rightarrow\infty}n^{\frac{1}{p}}=\max\limits_{1\le i\le n}|x_i| \end{aligned}
p→∞lim(1≤i≤nmax∣xi∣p)p1=1≤i≤nmax∣xi∣p→∞lim[n(1≤i≤nmax∣xi∣p)]p1=1≤i≤nmax∣xi∣p→∞limnp1=1≤i≤nmax∣xi∣
故
∣
∣
x
⃗
∣
∣
∞
=
lim
p
→
∞
(
∑
i
=
1
n
∣
x
i
∣
p
)
1
p
=
max
1
≤
i
≤
n
∣
x
i
∣
||\vec{x}||_{\infty}=\lim_{p\rightarrow\infty}\left(\sum_{i=1}^n |x_i|^p\right)^{\frac{1}{p}}=\max\limits_{1\le i\le n}|x_i|
∣∣x∣∣∞=p→∞lim(i=1∑n∣xi∣p)p1=1≤i≤nmax∣xi∣
4. 向量范数的等价性
定义 设
R
n
\mathbb{R}^n
Rn 中的向量序列
{
x
⃗
(
k
)
}
\{\vec{x}^{(k)}\}
{x(k)},向量
x
⃗
∗
∈
R
n
\vec{x}^*\in\mathbb{R}^n
x∗∈Rn,若:
lim
k
→
∞
x
i
(
k
)
=
x
i
∗
(
i
=
1
,
2
,
…
,
n
)
\lim_{k\rightarrow\infty}x^{(k)}_i=x^*_i\qquad(i=1,2,\dots,n)
k→∞limxi(k)=xi∗(i=1,2,…,n)
则称
{
x
⃗
(
k
)
}
\{\vec{x}^{(k)}\}
{x(k)}收敛于
x
⃗
∗
\vec{x}^*
x∗,记作:
lim
k
→
∞
x
⃗
(
k
)
=
x
⃗
∗
\lim_{k\rightarrow\infty}\vec{x}^{(k)}=\vec{x}^*
k→∞limx(k)=x∗
引理
lim
k
→
∞
x
⃗
(
k
)
=
x
⃗
∗
⟺
lim
k
→
∞
∣
∣
x
⃗
(
k
)
−
x
⃗
∗
∣
∣
=
0
\lim_{k\rightarrow\infty}\vec{x}^{(k)}=\vec{x}^*\Longleftrightarrow\lim_{k\rightarrow\infty}||\vec{x}^{(k)}-\vec{x}^*||=0
k→∞limx(k)=x∗⟺k→∞lim∣∣x(k)−x∗∣∣=0
定理 向量范数是向量的连续实值函数
证明:选定线性空间的基
e
⃗
i
(
i
=
1
,
2
,
…
,
n
)
\vec{e}_i\ (i=1,2,\dots,n)
ei (i=1,2,…,n),则:
0
≤
∣
∣
∣
x
⃗
∣
∣
−
∣
∣
y
⃗
∣
∣
∣
≤
∣
∣
x
⃗
−
y
⃗
∣
∣
=
∣
∣
∑
i
=
1
n
(
x
i
−
y
i
)
e
⃗
i
∣
∣
≤
∑
i
=
1
n
∣
x
i
−
y
i
∣
⋅
∣
∣
e
⃗
i
∣
∣
≤
∣
∣
x
⃗
−
y
⃗
∣
∣
∞
∑
i
=
1
n
∣
∣
e
⃗
i
∣
∣
0\le|\ ||\vec{x}||-||\vec{y}||\ |\le||\vec{x}-\vec{y}||=||\sum\limits_{i=1}^n(x_i-y_i)\vec{e}_i||\le\sum\limits_{i=1}^n|x_i-y_i|\cdot||\vec{e}_i||\le||\vec{x}-\vec{y}||_{\infty}\sum\limits_{i=1}^n||\vec{e}_i||
0≤∣ ∣∣x∣∣−∣∣y∣∣ ∣≤∣∣x−y∣∣=∣∣i=1∑n(xi−yi)ei∣∣≤i=1∑n∣xi−yi∣⋅∣∣ei∣∣≤∣∣x−y∣∣∞i=1∑n∣∣ei∣∣
故
lim
x
⃗
→
y
⃗
(
∣
∣
x
⃗
∣
∣
−
∣
∣
y
⃗
∣
∣
)
=
0
(
证毕
)
\lim_{\vec{x}\rightarrow\vec{y}}(||\vec{x}||-||\vec{y}||)=0\quad(证毕)
x→ylim(∣∣x∣∣−∣∣y∣∣)=0(证毕)
定义 设
∣
∣
x
⃗
∣
∣
a
||\vec{x}||_a
∣∣x∣∣a 与
∣
∣
x
⃗
∣
∣
b
||\vec{x}||_b
∣∣x∣∣b 为
R
n
\mathbb{R}^n
Rn 上的任意两种范数,若存在正常数
c
1
,
c
2
c_1,c_2
c1,c2 使得
∀
x
⃗
∈
R
n
,
c
1
∣
∣
x
⃗
∣
∣
a
≤
∣
∣
x
⃗
∣
∣
b
≤
c
2
∣
∣
x
⃗
∣
∣
a
\forall \vec{x}\in\mathbb{R}^n,\ c_1||\vec{x}||_a\le||\vec{x}||_b\le c_2||\vec{x}||_a
∀x∈Rn, c1∣∣x∣∣a≤∣∣x∣∣b≤c2∣∣x∣∣a
则称
∣
∣
x
⃗
∣
∣
a
||\vec{x}||_a
∣∣x∣∣a 与
∣
∣
x
⃗
∣
∣
b
||\vec{x}||_b
∣∣x∣∣b等价。
引理 范数的等价具有传递性。
证明:若
∣
∣
x
⃗
∣
∣
a
||\vec{x}||_a
∣∣x∣∣a 与
∣
∣
x
⃗
∣
∣
b
||\vec{x}||_b
∣∣x∣∣b等价且
∣
∣
x
⃗
∣
∣
b
||\vec{x}||_b
∣∣x∣∣b 与
∣
∣
x
⃗
∣
∣
c
||\vec{x}||_c
∣∣x∣∣c等价,则
∀
x
⃗
∈
R
n
,
{
c
1
∣
∣
x
⃗
∣
∣
b
≤
∣
∣
x
⃗
∣
∣
a
≤
c
2
∣
∣
x
⃗
∣
∣
b
c
3
∣
∣
x
⃗
∣
∣
c
≤
∣
∣
x
⃗
∣
∣
b
≤
c
4
∣
∣
x
⃗
∣
∣
c
,
其中
c
1
、
c
2
、
c
3
、
c
4
∈
R
\forall \vec{x}\in\mathbb{R}^n, \begin{cases} \ c_1||\vec{x}||_b\le||\vec{x}||_a\le c_2||\vec{x}||_b\\\\ \ c_3||\vec{x}||_c\le||\vec{x}||_b\le c_4||\vec{x}||_c \end{cases} ,其中\ c_1、c_2、c_3、c_4\in \mathbb{R}
∀x∈Rn,⎩
⎨
⎧ c1∣∣x∣∣b≤∣∣x∣∣a≤c2∣∣x∣∣b c3∣∣x∣∣c≤∣∣x∣∣b≤c4∣∣x∣∣c,其中 c1、c2、c3、c4∈R
那么,
c
1
c
3
∣
∣
x
⃗
∣
∣
c
≤
∣
∣
x
⃗
∣
∣
a
≤
c
2
c
4
∣
∣
x
⃗
∣
∣
c
c_1c_3||\vec{x}||_c\le||\vec{x}||_a\le c_2c_4||\vec{x}||_c
c1c3∣∣x∣∣c≤∣∣x∣∣a≤c2c4∣∣x∣∣c
故
∣
∣
x
⃗
∣
∣
a
||\vec{x}||_a
∣∣x∣∣a 与
∣
∣
x
⃗
∣
∣
c
||\vec{x}||_c
∣∣x∣∣c等价。
定理 任意两个向量范数等价。
证明:
当 x ⃗ = 0 ⃗ \vec{x}=\vec{0} x=0 时,显然任意范数等价。
当
x
⃗
≠
0
⃗
\vec{x}\ne\vec{0}
x=0 时,考虑集合
S
=
{
x
⃗
∣
∣
∣
x
⃗
∣
∣
∞
=
1
}
S=\{\vec{x}|\ ||\vec{x}||_{\infty}=1\}
S={x∣ ∣∣x∣∣∞=1}(超立方体表面,有界闭集),并且任意向量均可通过归一化映射为S的元素,即
∀
x
⃗
∈
R
n
,
x
⃗
∣
∣
x
⃗
∣
∣
∞
∈
S
\forall\vec{x}\in\mathbb{R}^n,\ \frac{\vec{x}}{||\vec{x}||_{\infty}}\in S
∀x∈Rn, ∣∣x∣∣∞x∈S
定义函数:
f
(
x
⃗
)
=
∣
∣
x
⃗
∣
∣
x
⃗
∣
∣
∞
∣
∣
f(\vec{x})=\left|\left|\frac{\vec{x}}{||\vec{x}||_{\infty}}\right|\right|
f(x)=∣
∣∣
∣∣∣x∣∣∞x∣
∣∣
∣
由于
f
(
x
)
f(x)
f(x)是S上的连续函数,故有界,同时根据范数的正定性有:
0
<
c
1
<
∣
∣
x
⃗
∣
∣
x
⃗
∣
∣
∞
∣
∣
<
c
2
0<c_1<\left|\left|\frac{\vec{x}}{||\vec{x}||_{\infty}}\right|\right|<c_2
0<c1<∣
∣∣
∣∣∣x∣∣∞x∣
∣∣
∣<c2
那么
c
1
∣
∣
x
⃗
∣
∣
∞
<
∣
∣
x
⃗
∣
∣
<
c
2
∣
∣
x
⃗
∣
∣
∞
c_1||\vec{x}||_{\infty}<||\vec{x}||<c_2||\vec{x}||_{\infty}
c1∣∣x∣∣∞<∣∣x∣∣<c2∣∣x∣∣∞
这说明任何向量范数均与最大范数等价,进一步根据向量范数等价的传递性知任意两个向量范数等价。
5. 几个常用向量范数之间的等价关系
- 欧式范数与1-范数:
∣ ∣ x ⃗ ∣ ∣ 2 ≤ ∣ ∣ x ⃗ ∣ ∣ 1 ≤ N ∣ ∣ x ⃗ ∣ ∣ 2 , x ⃗ ∈ R N ||\vec{x}||_2\le||\vec{x}||_1\le\sqrt{N}||\vec{x}||_2,\vec{x}\in\mathbb{R}^N ∣∣x∣∣2≤∣∣x∣∣1≤N∣∣x∣∣2,x∈RN
证明:
∣ ∣ x ⃗ ∣ ∣ 2 2 = x 1 2 + x 2 2 + ⋯ + x N 2 ≤ ( ∣ x 1 ∣ + ∣ x 2 ∣ + ⋯ + ∣ x N ∣ ) 2 = ∣ ∣ x ⃗ ∣ ∣ 1 2 ||\vec{x}||_2^{\ 2}=x_1^2+x_2^2+\dots+x_N^2\le(|x_1|+|x_2|+\dots+|x_N|)^2=||\vec{x}||_1^{\ 2} ∣∣x∣∣2 2=x12+x22+⋯+xN2≤(∣x1∣+∣x2∣+⋯+∣xN∣)2=∣∣x∣∣1 2
即,
∣ ∣ x ⃗ ∣ ∣ 2 ≤ ∣ ∣ x ⃗ ∣ ∣ 1 ||\vec{x}||_2\le||\vec{x}||_1 ∣∣x∣∣2≤∣∣x∣∣1
又
∣ x 1 ∣ + ∣ x 2 ∣ + ⋯ + ∣ x N ∣ = c 1 x 1 + c 2 x 2 + ⋯ + c N x N = c ⃗ T x ⃗ |x_1|+|x_2|+\dots+|x_N|=c_1x_1+c_2x_2+\dots+c_Nx_N=\vec{c}^{\ T}\vec{x} ∣x1∣+∣x2∣+⋯+∣xN∣=c1x1+c2x2+⋯+cNxN=c Tx
其中
c i = { − 1 ( x i < 0 ) 1 ( x i > 0 ) c_i= \begin{cases} -1&(x_i<0)\\\\ 1&(x_i>0) \end{cases} ci=⎩ ⎨ ⎧−11(xi<0)(xi>0)
由 Cauchy-Bunjakovski 不等式(Cauchy–Schwarz 不等式):
∣ x 1 ∣ + ∣ x 2 ∣ + ⋯ + ∣ x N ∣ = ∣ c ⃗ T x ⃗ ∣ ≤ ∣ c ⃗ T ∣ ⋅ ∣ x ⃗ ∣ = N ∣ ∣ x ⃗ ∣ ∣ 2 (证毕) |x_1|+|x_2|+\dots+|x_N|=|\vec{c}^{\ T}\vec{x}|\le|\vec{c}^{\ T}|\cdot|\vec{x}|=\sqrt{N}||\vec{x}||_2(证毕) ∣x1∣+∣x2∣+⋯+∣xN∣=∣c Tx∣≤∣c T∣⋅∣x∣=N∣∣x∣∣2(证毕) - 最大范数与欧式范数:
∣ ∣ x ⃗ ∣ ∣ ∞ ≤ ∣ ∣ x ⃗ ∣ ∣ 2 ≤ N ∣ ∣ x ⃗ ∣ ∣ ∞ , x ⃗ ∈ R N ||\vec{x}||_{\infty}\le||\vec{x}||_2\le\sqrt{N}||\vec{x}||_{\infty},\vec{x}\in\mathbb{R}^N ∣∣x∣∣∞≤∣∣x∣∣2≤N∣∣x∣∣∞,x∈RN
证明:
max 1 ≤ i ≤ N ∣ x i ∣ = ( max 1 ≤ i ≤ N ∣ x i ∣ ) 2 ≤ ∣ x 1 ∣ 2 + ∣ x 2 ∣ 2 + ⋯ + ∣ x N ∣ 2 ≤ N ( max 1 ≤ i ≤ N ∣ x i ∣ ) 2 \max_{1\le i\le N}|x_i|=\sqrt{\left(\max_{1\le i\le N}|x_i|\right)^2}\le\sqrt{|x_1|^2+|x_2|^2+\dots+|x_N|^2}\le\sqrt{N\left(\max_{1\le i\le N}|x_i|\right)^2} 1≤i≤Nmax∣xi∣=(1≤i≤Nmax∣xi∣)2≤∣x1∣2+∣x2∣2+⋯+∣xN∣2≤N(1≤i≤Nmax∣xi∣)2
(证毕) - 最大范数与1-范数:
∣ ∣ x ⃗ ∣ ∣ ∞ ≤ ∣ ∣ x ⃗ ∣ ∣ 1 ≤ N ∣ ∣ x ⃗ ∣ ∣ ∞ , x ⃗ ∈ R N ||\vec{x}||_{\infty}\le||\vec{x}||_1\le N||\vec{x}||_{\infty},\vec{x}\in\mathbb{R}^N ∣∣x∣∣∞≤∣∣x∣∣1≤N∣∣x∣∣∞,x∈RN
通过欧式范数与1-范数、欧式范数与最大范数间的关系可知其正确性。