6.8 The Primary Decomposition Theorem

这个定理（Theorem 12）可能是非计算导向的线性代数课中比较重要的定理之一，也许比它重要的就是Spectral Theory了。这一定理从一个operator $T$的minimal polynomial出发，根据最小多项式的因式（primes）出发，得到一个对应的direct sum且invariant under $T$的对$V$的分解，并且在每一个分解出的子空间中，$T$在其上的限制得到的operator的最小多项式恰是对应的prime。也就是说，根据$T$最小多项式的因式情况，将$V$分解为由每个因式独立支配（作为minimal polynomial）的subspace。这一定理比之前的讨论更进一步，是因为其并不再要求minimal polynomial的因式都是1阶的，可以更高阶，之前在最小多项式因式都是1阶且次幂为1时，我们可以将$T$对角化，也就是将$V$分解为由特征向量组成的子空间，如果1阶但次幂不为1，则可以将$T$三角化，但无法得到direct sum的子空间。
在这一定理的证明过程中，可以得到与primary decomposition 对应的投影$E_i$是$T$的一个多项式，因此如果$U$和$T$可交换，则primary decomposition出的每个子空间在$U$下也是invariant的。
接下来讨论一个特殊情况，即$T$的minimal polynomial的因式都是1阶（此时相当于可三角化但不一定对角化），通过Theorem 13可知，$T$可以分解为一个可对角化的operator和一个幂零的operator的和，且这两个operator是可交换的、唯一的，并且都可表示为$T$的多项式。
实际上越读到后面越发现，多项式是一个神奇的东西。

Exercises

1.Let $T$ be a linear operator on $R^3$ which is represented in the standard ordered basis by the matrix
$$\left[\begin{array}{ccc}6& -3& -2\\ 4& -1& -2\\ 10& -5& -3\end{array}\right].$$
Express the minimal polynomial $p$ for $T$ in the form $p=p_1{p}_2$, where $p_1$ and $p_2$ are monic and irreducible over the field of real numbers. Let $W_i$ be the null space of $p_i(T)$. Find basis ${\mathfrak{B}}_i$ for the spaces $W_1$ and $W_2$. If $T_i$ is the operator induced on $W_i$ by $T$, find the matrix of $T_i$ in the basis ${\mathfrak{B}}_i$.
Solution: We let $A$ be the matrix above and then
$$\begin{array}{rl}\det (xI-A)& =\left|\begin{array}{ccc}x-6& 3& 2\\ -4& x+1& 2\\ -10& 5& x+3\end{array}\right|=\left|\begin{array}{ccc}x-2& 2-x& 0\\ -4& x+1& 2\\ -10& 5& x+3\end{array}\right|\\ & =\left|\begin{array}{ccc}x-2& 0& 0\\ -4& x-3& 2\\ -10& -5& x+3\end{array}\right|\\ & =(x-2)(x^2+1)\end{array}$$
Since $A-2I\ne 0$ and $A^2+I\ne 0$, the minimal polynomial of $A$ is $(x-2)(x^2+1)$, thus $p_1=x-2$ and $p_2=x^2+1$.
For $W_1$ we have
$$2I-A=\left[\begin{array}{ccc}-4& 3& 2\\ -4& 3& 2\\ -10& 5& 5\end{array}\right]\to \left[\begin{array}{ccc}-4& 3& 2\\ -2& 1& 1\\ 0& 0& 0\end{array}\right]\to \left[\begin{array}{ccc}-2& 1& 1\\ 0& 1& 0\\ 0& 0& 0\end{array}\right]$$
Thus we can let ${\mathfrak{B}}_1=(1,0,2)$.
For $W_2$ we have
$$A^2+I=\left[\begin{array}{ccc}6& -3& -2\\ 4& -1& -2\\ 10& -5& -3\end{array}\right]\left[\begin{array}{ccc}6& -3& -2\\ 4& -1& -2\\ 10& -5& -3\end{array}\right]+I=\left[\begin{array}{ccc}5& -5& 0\\ 0& 0& 0\\ 10& -10& 0\end{array}\right]$$
Thus we can let B 2 = { ( 1 , 1 , 0 ) , ( 1 , 1 , 1 ) } \mathfrak B_2=\{(1,1,0),(1,1,1)\} B2​={ (1,1,0),(1,1,1)}.
The matrix $T_1$ of ${\mathfrak{B}}_1$ is $2$, since it is the subspace of characteristic vectors associated with the characteristic value $2$, or a direct computation shows that
$$A\left[\begin{array}{c}1\\ 0\\ 2\end{array}\right]=\left[\begin{array}{ccc}6& -3& -2\\ 4& -1& -2\\ 10& -5& -3\end{array}\right]\left[\begin{array}{c}1\\ 0\\ 2\end{array}\right]=\left[\begin{array}{c}2\\ 0\\ 4\end{array}\right]=2\left[\begin{array}{c}1\\ 0\\ 2\end{array}\right]$$
The matrix $T_2$ of ${\mathfrak{B}}_2$ can be computed with:
$$\begin{gathered} A\left[\begin{array}{c}1\\ 1\\ 0\end{array}\right]=\left[\begin{array}{ccc}6& -3& -2\\ 4& -1& -2\\ 10& -5& -3\end{array}\right]\left[\begin{array}{c}1\\ 1\\ 0\end{array}\right]=\left[\begin{array}{c}3\\ 3\\ 5\end{array}\right]=-2\left[\begin{array}{c}1\\ 1\\ 0\end{array}\right]+5\left[\begin{array}{c}1\\ 1\\ 1\end{array}\right] \\ A\left[\begin{array}{c}1\\ 1\\ 1\end{array}\right]=\left[\begin{array}{ccc}6& -3& -2\\ 4& -1& -2\\ 10& -5& -3\end{array}\right]\left[\begin{array}{c}1\\ 1\\ 1\end{array}\right]=\left[\begin{array}{c}1\\ 1\\ 2\end{array}\right]=-\left[\begin{array}{c}1\\ 1\\ 0\end{array}\right]+2\left[\begin{array}{c}1\\ 1\\ 1\end{array}\right] \end{gathered}$$
Thus the matrix $T_2$ in the basis ${\mathfrak{B}}_2$ is $\left[\begin{array}{cc}-2& -1\\ 5& 2\end{array}\right]$.

2.Let $T$ be the linear operator on $R$