矩阵杂记——最大最小特征值

证明: x T A x x T x ≤ λ m a x ( A + A T 2 ) \frac{x^TAx}{x^Tx} \leq \lambda_{max}(\frac{A+A^T}2) xTxxTAxλmax(2A+AT)
∀ A ∈ R n × n \forall A \in \R^{n \times n} ARn×n

原问题等价于: sup ⁡ ∣ ∣ x ∣ ∣ 2 = 1 x T A x ≤ λ m a x ( A + A T 2 ) \sup_{||x||_2=1} x^TAx \leq \lambda_{max}(\frac{A+A^T}2) x2=1supxTAxλmax(2A+AT)事实上取等号,即:
sup ⁡ ∣ ∣ x ∣ ∣ = 1 x T A x = sup ⁡ ∣ ∣ x ∣ ∣ = 1 x T ( A + A T 2 ) x + x T ( A − A T 2 ) x = sup ⁡ ∣ ∣ x ∣ ∣ = 1 x T ( A + A T 2 ) x = λ m a x ( A + A T 2 ) \sup_{||x||=1} x^TAx \\= \sup_{||x||=1} x^T(\frac{A+A^T}2)x+x^T(\frac{A-A^T}2)x \\= \sup_{||x||=1} x^T(\frac{A+A^T}2)x \\ = \lambda_{max}(\frac{A+A^T}2) x=1supxTAx=x=1supxT(2A+AT)x+xT(2AAT)x=x=1supxT(2A+AT)x=λmax(2A+AT)
证明也简单,考虑优化问题:
m a x i m i z e    x T S x s . t .          x T x = 1 maximize \;x^TSx \\ s.t.\;\;\;\;x^Tx = 1 maximizexTSxs.t.xTx=1
其中 S 为对称阵,拘束条件表示欧式长度为1。由于 S 对称,可对角化。 S = P T Λ P S = P^T\Lambda P S=PTΛP其中 P P P 为单位正交阵, Λ = d i a g { λ 1 , λ 2 , . . . λ n } \Lambda = diag\{\lambda_1,\lambda_2,...\lambda_n\} Λ=diag{λ1,λ2,...λn}。令 y = P x y = P x y=Px,则原问题等价于:
m a x i m i z e      y T Λ y = ∑ i = 1 n λ i y i 2 s . t .          y T y = ∑ i y i 2 = 1 maximize \;\;y^T\Lambda y =\sum_{i=1}^n \lambda_iy_i^2\\ s.t.\;\;\;\;y^Ty =\sum_i y_i^2= 1 maximizeyTΛy=i=1nλiyi2s.t.yTy=iyi2=1显然, ∑ i = 1 n λ i y i 2 ≤ m a x { λ i } ∑ i y i 2 = m a x { λ i } \sum_{i=1}^n \lambda_iy_i^2 \leq max\{\lambda_i\}\sum_i y_i^2= max\{\lambda_i\} i=1nλiyi2max{λi}iyi2=max{λi}当且仅当 ∣ y i ∣ |y_i| yi λ i \lambda_i λi 最大的那个分量上等于1时取等号。


由上述证明可知: λ m i n ( A + A T 2 ) ≤ x T A x x T x ≤ λ m a x ( A + A T 2 ) \lambda_{min}(\frac{A+A^T}2) \leq\frac{x^TAx}{x^Tx} \leq \lambda_{max}(\frac{A+A^T}2) λmin(2A+AT)xTxxTAxλmax(2A+AT)
更准确地来说, ∀ A ∈ R n × n \forall A \in \R^{n \times n} ARn×n
sup ⁡ x ≠ 0 x T A x x T x = λ m a x ( A + A T 2 ) inf ⁡ x ≠ 0 x T A x x T x = λ m i n ( A + A T 2 ) \sup_{x \neq 0}\frac{x^TAx}{x^Tx} = \lambda_{max}(\frac{A+A^T}2)\\\inf_{x \neq 0}\frac{x^TAx}{x^Tx} = \lambda_{min}(\frac{A+A^T}2) x̸=0supxTxxTAx=λmax(2A+AT)x̸=0infxTxxTAx=λmin(2A+AT)
特别地,当A对称时,
sup ⁡ x ≠ 0 x T A x x T x = λ m a x ( A ) inf ⁡ x ≠ 0 x T A x x T x = λ m i n ( A ) \sup_{x \neq 0}\frac{x^TAx}{x^Tx} = \lambda_{max}(A)\\\inf_{x \neq 0}\frac{x^TAx}{x^Tx} = \lambda_{min}(A) x̸=0supxTxxTAx=λmax(A)x̸=0infxTxxTAx=λmin(A)

你可能感兴趣的:(矩阵论)