前置知识
正惯性指数:正特征值个数
负惯性指数:负特征值个数
合同:如果存在可逆矩阵 P \mathbf{P} P,使得 P T A P = B \mathbf{P}^T\mathbf{A}\mathbf{P}=\mathbf{B} PTAP=B,则 A \mathbf{A} A合同于 B \mathbf{B} B,记为 A ≃ B \mathbf{A}\simeq \mathbf{B} A≃B
合同不改变矩阵的惯性指数
赫尔维茨定理
A
\mathbf{A}
A是实对称矩阵
A
≻
0
\mathbf{A}\succ 0
A≻0的充要条件是顺序主子式大于0
证明:
记
A
k
=
(
a
11
a
12
⋯
a
1
k
a
21
a
22
⋯
a
2
k
⋯
⋯
⋯
⋯
a
k
1
a
k
2
⋯
a
k
k
)
\mathbf{A}_k =\begin{pmatrix} a_{11}&a_{12}&\cdots & a_{1k}\\ a_{21}&a_{22}&\cdots & a_{2k}\\ \cdots & \cdots & \cdots & \cdots\\ a_{k1}&a_{k2}&\cdots & a_{kk}\\ \end{pmatrix}
Ak=
a11a21⋯ak1a12a22⋯ak2⋯⋯⋯⋯a1ka2k⋯akk
Δ
k
=
∣
A
k
∣
\mathbf{\Delta}_k=\left| \mathbf{A}_k\right|
Δk=∣Ak∣
必要性:
设
x
k
=
(
x
1
,
x
2
,
…
,
x
k
)
T
≠
0
\mathbf{x}_k=(x_1,x_2,\dots,x_k)^T \neq 0
xk=(x1,x2,…,xk)T=0
x
=
(
x
1
,
x
2
,
…
,
x
k
,
…
,
x
n
)
T
\mathbf{x}= (x_1,x_2,\dots,x_k,\dots,x_n)^T
x=(x1,x2,…,xk,…,xn)T,其中
x
k
+
1
=
⋯
=
x
n
=
0
x_{k+1}=\cdots=x_{n}=0
xk+1=⋯=xn=0
A
≻
0
⇒
∀
x
≠
0
,
x
T
A
x
=
x
k
T
A
k
x
k
>
0
⇒
A
k
≻
0
\mathbf{A}\succ 0\Rightarrow \forall \mathbf{x}\neq \mathbf{0} , \mathbf{x}^T \mathbf{A} \mathbf{x}= \mathbf{x}_k^T \mathbf{A}_k \mathbf{x}_k>0\Rightarrow \mathbf{A}_k \succ 0
A≻0⇒∀x=0,xTAx=xkTAkxk>0⇒Ak≻0
所以
A
k
≻
0
\mathbf{A}_k\succ 0
Ak≻0
由行列式等于特征值的乘积,以及正定矩阵的特征值大于0,所以行列式也大于0
Δ
k
=
∣
A
k
∣
>
0
\mathbf{\Delta}_k=\left|\mathbf{A}_k\right| >0
Δk=∣Ak∣>0
充分性:
方法1比较暴力,好懂,
方法2,感觉稍微花了点,但是也还好
方法1:
当
n
=
1
n=1
n=1时,显然成立
假设
n
=
k
−
1
n=k-1
n=k−1时成立
A
=
(
A
n
−
1
α
α
T
a
n
n
)
\mathbf{A}=\begin{pmatrix} \mathbf{A}_{n-1}& \mathbf{\alpha} \\ \mathbf{\alpha}^T & a_{nn}\\ \end{pmatrix}
A=(An−1αTαann)
A
n
−
1
\mathbf{A}_{n-1}
An−1是一个实对称矩阵,并且根据假设所有的顺序主子式大于0
所以一定合同于单位矩阵
即存在可逆
C
\mathbf{C}
C
使得
C
T
A
n
−
1
C
=
I
n
−
1
\mathbf{C}^T \mathbf{A}_{n-1}\mathbf{C} =\mathbf{I}_{n-1}
CTAn−1C=In−1
设
P
=
(
I
n
−
1
−
A
n
−
1
−
1
α
0
T
1
)
\mathbf{P}=\begin{pmatrix} \mathbf{I}_{n-1} & -\mathbf{A}_{n-1}^{-1}\mathbf{\alpha}\\ \mathbf{0}^T& 1\\ \end{pmatrix}
P=(In−10T−An−1−1α1),显然可逆
P
T
A
P
=
(
A
n
−
1
0
0
T
a
n
n
−
α
T
A
n
−
1
−
1
α
)
\mathbf{P}^T \mathbf{A}\mathbf{P}=\begin{pmatrix} \mathbf{A}_{n-1} &\mathbf{0}\\ \mathbf{0}^T& a_{nn}-\mathbf{\alpha}^T A_{n-1}^{-1} \mathbf{\alpha} \\ \end{pmatrix}
PTAP=(An−10T0ann−αTAn−1−1α)
设
b
=
a
n
n
−
α
T
A
n
−
1
−
1
α
b=a_{nn}-\mathbf{\alpha}^T \mathbf{A}_{n-1}^{-1} \mathbf{\alpha}
b=ann−αTAn−1−1α
⇒
A
≃
(
A
n
−
1
0
0
T
b
)
\Rightarrow \mathbf{A} \simeq \begin{pmatrix} \mathbf{A}_{n-1} &\mathbf{0}\\ \mathbf{0}^T& b \\ \end{pmatrix}
⇒A≃(An−10T0b)
同取行列式
∣
A
∣
=
∣
P
T
∣
∣
A
∣
∣
P
∣
=
∣
P
T
A
P
∣
=
∣
A
n
−
1
∣
b
>
0
⇒
b
>
0
\left|\mathbf{A}\right|=\left|\mathbf{P}^T\right|\left|\mathbf{A}\right|\left|\mathbf{P}\right|=\left|\mathbf{P}^T \mathbf{A}\mathbf{P}\right| =\left| \mathbf{A}_{n-1}\right| b>0 \Rightarrow b>0
∣A∣=
PT
∣A∣∣P∣=
PTAP
=∣An−1∣b>0⇒b>0
(
C
0
0
T
1
)
T
(
A
n
−
1
0
0
T
b
)
(
C
0
0
T
1
)
=
(
I
n
−
1
0
0
T
b
)
\begin{pmatrix} \mathbf{C}& \mathbf{0}\\ \mathbf{0}^T & 1 \end{pmatrix}^T \begin{pmatrix} \mathbf{A}_{n-1} &\mathbf{0}\\ \mathbf{0}^T& b \\ \end{pmatrix} \begin{pmatrix} \mathbf{C}& \mathbf{0}\\ \mathbf{0}^T & 1 \end{pmatrix}=\begin{pmatrix} \mathbf{I}_{n-1} &\mathbf{0}\\ \mathbf{0}^T& b \\ \end{pmatrix}
(C0T01)T(An−10T0b)(C0T01)=(In−10T0b)
(
A
n
−
1
0
0
T
b
)
≃
(
I
n
−
1
0
0
T
b
)
\begin{pmatrix} \mathbf{A}_{n-1} &\mathbf{0}\\ \mathbf{0}^T& b \\ \end{pmatrix} \simeq \begin{pmatrix} \mathbf{I}_{n-1} &\mathbf{0}\\ \mathbf{0}^T& b \\ \end{pmatrix}
(An−10T0b)≃(In−10T0b)
由合同传递性
A
\mathbf{A}
A的正惯性指数为
n
n
n,
A
≻
0
\mathbf{A}\succ 0
A≻0
方法2:
当
n
=
1
n=1
n=1时,显然成立
假设
n
=
k
−
1
n=k-1
n=k−1时成立
当
n
=
k
n=k
n=k时
f
(
x
1
,
x
2
,
…
,
x
k
)
=
∑
i
=
1
k
∑
j
=
1
k
a
i
j
x
i
x
j
f(x_1,x_2,\dots,x_k)=\sum_{i=1}^{k}\sum_{j=1}^{k}a_{ij}x_{i}x_{j}
f(x1,x2,…,xk)=i=1∑kj=1∑kaijxixj
由假设,
a
11
>
0
a_{11}>0
a11>0
f
t
=
f
(
x
1
,
…
,
x
t
,
0
,
0
,
…
,
0
)
=
∑
i
=
1
t
∑
j
=
1
t
a
i
j
x
i
x
j
(
1
<
t
≤
k
)
f_t=f(x_1,\dots , x_t,0,0,\dots,0)=\sum_{i=1}^{t}\sum_{j=1}^{t}a_{ij}x_{i}x_{j}(1<t\le k)
ft=f(x1,…,xt,0,0,…,0)=i=1∑tj=1∑taijxixj(1<t≤k)
由假设,
a
11
>
0
a_{11}>0
a11>0
f
t
=
1
a
11
(
a
11
x
1
+
a
12
x
2
+
x
2
⋯
+
a
1
t
x
t
)
2
+
∑
i
=
2
t
∑
j
=
2
t
b
i
j
x
i
x
j
f_t=\frac{1}{a_{11}}(a_{11} x_{1} +a_{12}x_{2}+x_{2}\dots + a_{1t} x_{t})^2+\sum_{i=2}^{t}\sum_{j=2}^{t}b_{ij}x_{i}x_{j}
ft=a111(a11x1+a12x2+x2⋯+a1txt)2+i=2∑tj=2∑tbijxixj
其中
b
i
j
=
a
i
j
−
a
1
i
a
1
j
a
11
b_{ij}=a_{ij}-\frac{a_{1i} a_{1j}}{a_{11}}
bij=aij−a11a1ia1j
a
i
j
=
a
j
i
⇒
b
i
j
=
b
j
i
a_{ij}=a_{ji}\Rightarrow b_{ij}=b_{ji}
aij=aji⇒bij=bji
当
t
=
k
t=k
t=k的时候
可以看出前半部分时大于等于0的
现在证明后半部分大于0
Δ
t
=
∣
a
11
a
12
⋯
a
1
t
a
21
a
22
⋯
a
2
t
⋯
⋯
⋯
⋯
a
t
1
a
t
2
⋯
a
t
t
∣
=
i
=
2
,
3
,
…
,
t
r
i
−
a
i
1
a
11
r
1
∣
a
11
a
12
⋯
a
1
t
0
b
22
⋯
b
2
t
⋯
⋯
⋯
⋯
0
b
t
2
⋯
b
t
t
∣
=
a
11
∣
b
22
⋯
b
2
t
⋯
⋯
⋯
b
t
2
⋯
b
t
t
∣
>
0
\begin{aligned} \mathbf{\Delta}_t &=\left| \begin{array}{cccc} a_{11}&a_{12}&\cdots & a_{1t}\\ a_{21}&a_{22}&\cdots & a_{2t}\\ \cdots & \cdots & \cdots & \cdots\\ a_{t1}&a_{t2}&\cdots & a_{tt}\\ \end{array}\right|\\ &\xlongequal[i=2,3,\dots,t]{r_{i} -\frac{a_{i1}}{a_{11}}r_1} \left| \begin{array}{cccc} a_{11}&a_{12}&\cdots & a_{1t}\\ 0&b_{22}&\cdots & b_{2t}\\ \cdots & \cdots & \cdots & \cdots\\ 0&b{t2}&\cdots & b_{tt}\\ \end{array}\right|\\ &=a_{11}\left| \begin{array}{cccc} b_{22}&\cdots & b_{2t}\\ \cdots & \cdots & \cdots\\ b_{t2}&\cdots & b_{tt}\\ \end{array}\right|\\ &>0 \end{aligned}
Δt=
a11a21⋯at1a12a22⋯at2⋯⋯⋯⋯a1ta2t⋯att
ri−a11ai1r1i=2,3,…,t
a110⋯0a12b22⋯bt2⋯⋯⋯⋯a1tb2t⋯btt
=a11
b22⋯bt2⋯⋯⋯b2t⋯btt
>0
⇒
∣
b
22
⋯
b
2
t
⋯
⋯
⋯
b
k
2
⋯
b
t
t
∣
>
0
\Rightarrow \left| \begin{array}{cccc} b_{22}&\cdots & b_{2t}\\ \cdots & \cdots & \cdots\\ b_{k2}&\cdots & b_{tt}\\ \end{array}\right|>0
⇒
b22⋯bk2⋯⋯⋯b2t⋯btt
>0
显然对于所有的
t
=
2
,
3
,
…
,
k
t=2,3,\dots,k
t=2,3,…,k都成立
设
B
=
(
b
22
⋯
b
2
t
⋯
⋯
⋯
b
t
2
⋯
b
t
t
)
\mathbf{B}=\begin{pmatrix} b_{22}&\cdots & b_{2t}\\ \cdots & \cdots & \cdots\\ b_{t2}&\cdots & b_{tt}\\ \end{pmatrix}
B=
b22⋯bt2⋯⋯⋯b2t⋯btt
因为
B
\mathbf{B}
B是一个
k
−
1
k-1
k−1阶实对称矩阵,
B
\mathbf{B}
B所有顺序主子式大于0
由假设
B
≻
0
\mathbf{B}\succ 0
B≻0
所以
A
≻
0
\mathbf{A}\succ 0
A≻0
由数学归纳法,成立