Skip to content

2025-2-17日习题课

约 1740 字大约 6 分钟

2025-02-28

problem 1

证明:秩为 rr 的对称阵可以写成 rr 个秩为 11 的对称阵之和。

proof

CC 是可逆矩阵,使得

CAC=diag{a1,,ar,0,,0}C^{\prime} A C=\operatorname{diag}\{a_1, \cdots, a_r, 0, \cdots, 0\}

其中 ai0(1ir)a_i \neq 0(1 \leq i \leq r) ,则

A=(C1)a1E11C1++(C1)arErrC1A=(C^{-1})^{\prime} a_1 E_{11} C^{-1}+\cdots+(C^{-1})^{\prime} a_r E_{r r} C^{-1}

其中 EiiE_{i i} 是第 (i,i)(i, i) 元素为 1 ,其余元素全为 0 的基础矩阵,从而每个 (C1)aiEiiC1(C^{-1})^{\prime} a_i E_{i i} C^{-1} 都是秩等于 1 的对称矩阵。

problem 2

证明:一个实二次型 ff 可以分解为两个实系数的一次齐次多项式的乘积当且仅当 r(f)=2r(f)=2 且符号差 s=0,s=0, 或者秩等于 1.1.

proof

必要性。如果实二次型 f(x1,x2,,xn)f(x_1, x_2, \cdots, x_n) 可分解成两个一次齐次多项式的乘积:

f(x1,x2,,xn)=(a1x1+a2x2++anxn)(b1x1+b2x2++bnxn)\begin{aligned} & f(x_1, x_2, \cdots, x_n) \\ = & (a_1 x_1+a_2 x_2+\cdots+a_n x_n)(b_1 x_1+b_2 x_2+\cdots+b_n x_n) \end{aligned}

有两种可能:1)

(b1,b2,,bn)=k(a1,a2,,an),k0 于是 f(x1,x2,,xn)=k(a1x1+a2x2++anxn)2\begin{aligned} & (b_1, b_2, \cdots, b_n)=k(a_1, a_2, \cdots, a_n), k \neq 0 \text{ 于是 } \\ & f(x_1, x_2, \cdots, x_n)=k(a_1 x_1+a_2 x_2+\cdots+a_n x_n)^2 \end{aligned}

因为 a1,a2,,ana_1, a_2, \cdots, a_n 不全为 0 ,以 a1,a2,,ana_1, a_2, \cdots, a_n 为第一行,作一个可逆矩阵 CC ,线性替换

Y=CXY=C X

是可逆的。并且 f(x1,x2,,xn)f(x_1, x_2, \cdots, x_n) 经过这个线性替换化为

f(x1,x2,,xn)=ky12(k0)f(x_1, x_2, \cdots, x_n)=k y_1^2 \quad(k \neq 0)

此时 f(x1,x2,,xn)f(x_1, x_2, \cdots, x_n) 的秩等于 1 。 2)(a1,a2,,an)(a_1, a_2, \cdots, a_n)(b1,b2,,bn)(b_1, b_2, \cdots, b_n) 线性无关。于是可以 (a1,a2,,an),(b1,b2,,bn)(a_1, a_2, \cdots, a_n),(b_1, b_2, \cdots, b_n) 为第 1,2 行作一个可逆矩阵 CC ,可逆线性替换

Y=CXY=C X

f(x1,x2,,xn)f(x_1, x_2, \cdots, x_n) 化为

f(x1,x2,,xn)=y1y2f(x_1, x_2, \cdots, x_n)=y_1 y_2

再令

{y1=z1+z2y2=z1z2y3=z3yn=zn\begin{cases} y_1=z_1+z_2 \\ y_2=z_1-z_2 \\ y_3=z_3 \\ \cdots \\ y_n=z_n \end{cases}

f(x1,x2,,xn)=z12z22f(x_1, x_2, \cdots, x_n)=z_1^2-z_2^2

所以在这种情况,ff 的秩等于 2 ,符号差为 0 。 充分性。1)如果 f(x1,x2,,xn)f(x_1, x_2, \cdots, x_n) 的秩为 1 ,则 f(x1,x2,,xn)f(x_1, x_2, \cdots, x_n) 可经可逆线性替换

{y1=c11x1+c12x2++c1nxny2=c21x1+c22x2++c2nxnyn=cn1x1+cn2x2++cnnxn\begin{cases} y_1= c_{11} x_1+c_{12} x_2+\cdots+c_{1n} x_n \\ y_2= c_{21} x_1+c_{22} x_2+\cdots+c_{2n} x_n \\ \cdots \\ y_n= c_{n1} x_1+c_{n2} x_2+\cdots+c_{nn} x_n \end{cases}

化为

f(x1,x2,,xn)=y12=(c11x1+c12x2++c1nxn)2f(x_1, x_2, \cdots, x_n)=y_1^2=(c_{11} x_1+c_{12} x_2+\cdots+c_{1n} x_n)^2

2)如果 f(x1,x2,,xn)f(x_1, x_2, \cdots, x_n) 的秩为 2 ,符号差为 0 ,则 f(x1,x2,,xn)f(x_1, x_2, \cdots, x_n) 可经适当的可逆线性替换

Y=CXY=C X

化为规范形

f=y12y22=(c11x1++c1nxn)2(c21x1++c2nxn)2=[(c11+c21)x1++(c1n+c2n)xn][(c11c21)x1++(c1nc2n)xn]\begin{aligned} f= & y_1^2-y_2^2=(c_{11} x_1+\cdots+c_{1n} x_n)^2-(c_{21} x_1+\cdots+c_{2n} x_n)^2 \\ = & [(c_{11}+c_{21}) x_1+\cdots+(c_{1n}+c_{2n}) x_n] \cdot \\ & [(c_{11}-c_{21}) x_1+\cdots+(c_{1n}-c_{2n}) x_n] \end{aligned}

这两种情形都说明 f(x1,x2,,xn)f(x_1, x_2, \cdots, x_n) 可以分解成两个实系数一次齐次多项式的乘积。

problem 3

设实二次型

f(x1,x2,,xn)=i=1s(ai1x1++ainxn)2f(x_1,x_2,\dots,x_n)=\sum^s_{i=1}(a_{i1}x_1+\dots+a_{in}x_n)^2

证明:r(f)=r(A),A=(aij)s×n\operatorname{r}(f)=\operatorname{r}(A),A=(a_{ij})_{s \times n}

proof

设实二次型

f(x1,x2,,xn)=i=1s(ai1x1++ainxn)2f(x_1, x_2, \dots, x_n) = \sum_{i=1}^s (a_{i1}x_1 + \dots + a_{in}x_n)^2

其中 A=(aij)s×nA = (a_{ij})_{s \times n}。需证 r(f)=r(A)\operatorname{r}(f) = \operatorname{r}(A)

步骤分析:

  1. 二次型的矩阵表示 每个平方项 (ai1x1++ainxn)2(a_{i1}x_1 + \dots + a_{in}x_n)^2 可表示为 (Aix)2(A_i \mathbf{x})^2AiA_iAA 的第 ii 行),因此:

    f(x)=i=1s(Aix)2=xAAxf(\mathbf{x}) = \sum_{i=1}^s (A_i \mathbf{x})^2 = \mathbf{x}^\top A^\top A \mathbf{x}

    对应二次型的矩阵为 B=AAB = A^\top A,故 r(f)=r(B)\operatorname{r}(f) = \operatorname{r}(B)

  2. 矩阵秩的性质 对任意实矩阵 AA,有:

    r(AA)=r(A)\operatorname{r}(A^\top A) = \operatorname{r}(A)

    证明思路

    • AAx=0    Ax=0A^\top A \mathbf{x} = 0 \iff A \mathbf{x} = 0(因 Ax2=xAAx\|A \mathbf{x}\|^2 = \mathbf{x}^\top A^\top A \mathbf{x})。
    • AAA^\top AAA 的零空间相同,因此 dimker(AA)=dimker(A)\dim \ker(A^\top A) = \dim \ker(A)
    • 由秩-零化度定理,r(AA)=r(A)\operatorname{r}(A^\top A) = \operatorname{r}(A)
  3. 结论 结合上述结果:

    r(f)=r(AA)=r(A)\operatorname{r}(f) = \operatorname{r}(A^\top A) = \operatorname{r}(A)

最终结论:

r(f)=r(A)\boxed{\operatorname{r}(f) = \operatorname{r}(A)}

problem 4

tt 取何值时,二次型

t(x12+x22+x32)+2x1x22x1x3+2x2x3t(x_1^2+x_2^2+x_3^2)+2x_1x_2-2x_1x_3+2x_2x_3

正定?

solution

考虑二次型

f(x1,x2,x3)=t(x12+x22+x32)+2x1x22x1x3+2x2x3f(x_1, x_2, x_3) = t(x_1^2 + x_2^2 + x_3^2) + 2x_1x_2 - 2x_1x_3 + 2x_2x_3

其对应的对称矩阵为:

A=[t111t111t]A = \begin{bmatrix} t & 1 & -1 \\ 1 & t & 1 \\ -1 & 1 & t \end{bmatrix}

正定性条件要求所有顺序主子式均大于零:

  1. 一阶主子式t>0t > 0
  2. 二阶主子式

t11t=t21>0    t>1(因 t>0)\begin{vmatrix} t & 1 \\ 1 & t \end{vmatrix} = t^2 - 1 > 0 \implies t > 1 \quad (\text{因 } t > 0)

  1. 三阶主子式(矩阵行列式):

t111t111t=t33t2=(t2)(t+1)2>0\begin{vmatrix} t & 1 & -1 \\ 1 & t & 1 \\ -1 & 1 & t \end{vmatrix} = t^3 - 3t - 2 = (t - 2)(t + 1)^2 > 0

  • t>2t > 2 时,(t2)>0(t - 2) > 0(t+1)2>0(t + 1)^2 > 0,故行列式为正;
  • t<2t < 2 时,行列式非正。

综合条件

  • t>0t > 0
  • t>1t > 1
  • t>2t > 2

结论: 当且仅当 t>2t > 2 时,二次型正定。

t>2\boxed{t > 2}

problem 5

实数 a,ba, b 满足什么条件时

(a1b110b01)\begin{pmatrix} a & 1 & b \\ 1 & -1 & 0 \\ b & 0 & -1 \end{pmatrix}

正定,负定,半正定,半负定,不定?

solution

λIA=0    (λ+1)(λ2(1a)λ+a+1)=0\begin{vmatrix} \lambda I-A \end{vmatrix}=0 \implies (\lambda+1)(-\lambda^2-(1-a)\lambda+a+1)=0

AA 必有特征值 1-1,于是 AA 不可能正定或者半正定。 若有 00 特征值,则等价于 a=b21a=-b^2-1。 若 AA 负定,则等价于

{a1<0(a+b2+1)<0    a<b2+1\begin{cases} a-1<0 \\ -(a+b^2+1)<0 \end{cases} \implies a<-b^2+1

AA 半负定,则等价于

{a10(a+b2+1)0    ab2+1\begin{cases} a-1 \leq 0 \\ -(a+b^2+1) \leq 0 \end{cases} \implies a\leq-b^2+1

problem 6

证明:任一实对称矩阵,都是两个正定矩阵的差。

proof
  1. 特征值性质 实对称矩阵 AA 的特征值均为实数。设其最大绝对特征值为 A=maxiλi\|A\| = \max_i |\lambda_i|

  2. 选择常数 ccc>Ac > \|A\|(例如 c=A+1c = \|A\| + 1),则对 AA 的任意特征值 λi\lambda_i,有:

    λi+c>0(因 c>λi)\lambda_i + c > 0 \quad (\text{因 } c > |\lambda_i|)

  3. 构造正定矩阵

    • 定义 P=A+cIP = A + cI,其特征值为 λi+c>0\lambda_i + c > 0,故 PP 正定;
    • 定义 N=cIN = cI,显然为正定矩阵。
  4. 验证分解

    A=PN=(A+cI)cIA = P - N = (A + cI) - cI

结论: 任意实对称矩阵 AA 均可表示为两个正定矩阵的差,即:

A=PN其中 P,N 为正定矩阵\boxed{A = P - N \quad \text{其中 } P,N \text{ 为正定矩阵}}

problem 7

设半正定矩阵 AAr(A)=1\operatorname{r}(A)=1, 证明:存在 nn 维非零实向量 α\alpha s.t. A=ααA=\alpha \alpha^{\prime}.

proof

可作加强命题: 设 AAnn 阶实对称阵,求证: AA 是秩为 rr 的半正定阵的充分必要条件是存在秩为 rrr×nr \times n 实矩阵 BB ,使 A=BBA=B^{\prime} B

以下给出证明: 若 A=BBA=B^{\prime} B ,则 r(A)=r(BB)=r(B)=r\mathrm{r}(A)=\mathrm{r}(B^{\prime} B)=\mathrm{r}(B)=r ,且对任意的 nn 维实列向量 α\alpha, αAα=αBBα=(Bα)(Bα)0\alpha^{\prime} A \alpha=\alpha^{\prime} B^{\prime} B \alpha=(B \alpha)^{\prime}(B \alpha) \geq 0 ,因此 AA 是秩为 rr 的半正定阵。

反之,若 AA 是秩为 rr 的半正定阵,则存在可逆矩阵 CC ,使得 A=Cdiag{1,,1,0,,0}CA=C^{\prime} \operatorname{diag}\{1, \cdots, 1,0, \cdots, 0\} C ,其中有 rr 个 1。令 B=(IrO)CB=(I_r \quad O)C ,则 BB 是秩等于 rrr×nr \times n 矩阵,且 A=BBA=B^{\prime} B

贡献者: Ming Wei