如果把矩阵化成对角矩阵,关于矩阵的函数计算问题就会大大简化。但一般的矩阵未必与对角矩阵相似。
矩阵的标准型有多重,Jordan (约当)标准型是最接近对角矩阵的形式,在控制理论中经常用到。

存在条件:

ACnx<script type="math/tex" id="MathJax-Element-7932"> A \in C^{n*x} </script>, 其特征多项式可以写成如下形式:

φ(λ)=(λλ1)m1(λλs)ms
<script type="math/tex; mode=display" id="MathJax-Element-2"> \varphi (\lambda) = (\lambda - \lambda_1 )^{m_1} \dots (\lambda - \lambda_s )^{m_s}</script>

其中:m1+m2++ms=n<script type="math/tex" id="MathJax-Element-3"> m_1 + m_2 + \dots + m_s = n </script>, 那么,矩阵 A<script type="math/tex" id="MathJax-Element-4">A</script> 可以经过相似变换,化成唯一的 Jordan 标准型 J 。即存在可逆矩阵 P<script type="math/tex" id="MathJax-Element-5">P</script>, 满足

P1AP=J
<script type="math/tex; mode=display" id="MathJax-Element-6"> P^{-1} A P = J </script>
A<script type="math/tex" id="MathJax-Element-7">A</script> 有Jordan 分解:
A=PJP1
<script type="math/tex; mode=display" id="MathJax-Element-8"> A = P J P^{-1} </script>

J=diag(J1(λ1),J2(λ2),,Js(λs))
<script type="math/tex; mode=display" id="MathJax-Element-4485"> J = diag( J_1 (\lambda _1),J_2(\lambda _2), \dots, J_s(\lambda _s)) </script>

Ji(λi),i=1,2,,s<script type="math/tex" id="MathJax-Element-4486"> J_i(\lambda _i), i=1,2, \dots,s </script> 被称为 Jardon 块。

对应的:

P=(P1,P2,,Ps)
<script type="math/tex; mode=display" id="MathJax-Element-4487">P=(P_1,P_2,\dots,P_s)</script>

Ji(λi)=diag(J1(λi),J2(λi),,Jki(λi),)
<script type="math/tex; mode=display" id="MathJax-Element-4464"> J_i(\lambda _i) = diag( J_1(\lambda_i), J_2(\lambda_i), \dots,J_{k_i}(\lambda_i), ) </script>

Jki(λi),i=1,2,,ki<script type="math/tex" id="MathJax-Element-4465"> J_{k_i}(\lambda _i), i=1,2, \dots,k_i </script> 被称为 Jardon 子块。

对应的:

Pi=(Pi(1),Pi(2),,Pi(ki)
<script type="math/tex; mode=display" id="MathJax-Element-4466">P_i=(P_i(1),P_i(2),\dots,P_i(k_i)</script>

Jki(λi)=λi0001λi0001λi00001λiCkiki
<script type="math/tex; mode=display" id="MathJax-Element-7923"> J_ {k_i} (\lambda_i) = \begin{bmatrix} \lambda _i & 1 & 0 & \dots & 0\\ 0 & \lambda _i & 1& \dots & 0\\ 0&0 & \lambda _i & \dots & 0\\ \dots & \dots & \dots & \dots & 1\\ 0 & 0 & 0 & \dots & \lambda _i \end{bmatrix} \in C^{k_i * k_i} </script>

求解方法:

1、求矩阵的特征值 λi<script type="math/tex" id="MathJax-Element-7924"> \lambda _i</script> 及每个特征值的重数 mi<script type="math/tex" id="MathJax-Element-7925"> m_i</script>。

计算特征值 λi<script type="math/tex" id="MathJax-Element-7926"> \lambda _i </script> 的指标 ki<script type="math/tex" id="MathJax-Element-7927">k_i</script>, 即 rank(AλiI)ki=rank(AλiI)ki+1<script type="math/tex" id="MathJax-Element-7928">rank(A- \lambda _i I)^{k_i} = rank(A- \lambda _i I)^{k_i+1} </script> 成立的最小正整数 ki<script type="math/tex" id="MathJax-Element-7929">k_i</script>,也就是 λi<script type="math/tex" id="MathJax-Element-7930"> \lambda _i</script>对应的约当块的最大阶数。

2、计算特征值 λi<script type="math/tex" id="MathJax-Element-7931"> \lambda _i</script>对应的Jardon 块的个数及阶数。

rt=rank(AλiI)t,t=0,1,2,,ki
<script type="math/tex; mode=display" id="MathJax-Element-7911">r_t = rank(A- \lambda _i I)^t, t=0,1,2,\dots,k_i</script>
δt=rt1+rt+12rt
<script type="math/tex; mode=display" id="MathJax-Element-7912">\delta _t = r_{t-1} + r_{t+1} - 2 r_{t}</script>
δt<script type="math/tex" id="MathJax-Element-7913">\delta _t</script> 为 λi<script type="math/tex" id="MathJax-Element-7914"> \lambda _i</script>对应的 t<script type="math/tex" id="MathJax-Element-7915">t</script> 阶约当块 的个数Jt(λi<script type="math/tex" id="MathJax-Element-7916"> J_t ( \lambda _i ) </script>

3、计算 P<script type="math/tex" id="MathJax-Element-7917">P</script> 矩阵。

先求 Pi,i=1,2,,s<script type="math/tex" id="MathJax-Element-7918">P_i, i=1,2, \dots,s</script> 。

先求 Pit,t=1,2,,ki<script type="math/tex" id="MathJax-Element-7919">P_i{t} ,t=1,2,\dots,k_i </script>

t<script type="math/tex" id="MathJax-Element-7920">t</script>阶约旦子块,求

(AλiI)tx=0
<script type="math/tex; mode=display" id="MathJax-Element-7921">(A- \lambda _i I )^t x=0 </script> 的非零解(唯一) x<script type="math/tex" id="MathJax-Element-7922">x</script>,

Pit=(x,(AλiI)x,,(AλiI)t1x)
<script type="math/tex; mode=display" id="MathJax-Element-7879">P_i{t}=(x,(A- \lambda _i I )x,\dots, (A- \lambda _i I )^{t-1}x)</script>

进过组合,就可以得到变换矩阵 P<script type="math/tex" id="MathJax-Element-7880"></script>

Logo

有“AI”的1024 = 2048,欢迎大家加入2048 AI社区

更多推荐