Let $B=\begin{bmatrix} 0 & 0 & 0 & -8 \\ 1&0&0 & 16\\ 0 &1&0& -14\\ 0&0&1&6 \end{bmatrix}$
Consider B a real matrix. Find its Jordan Form. So, the characteristic polynomial for B is $(x^2-2x+2)(x-2)^2$. Suppose $B$ represents $T$ in the standard basis. By the primary decomposition theorem, we have that $V$ is the direct sum of $K=Ker(T^2-2T+2I)$ and $L=Ker[(T-2)^2]$. $L$ has dimension 2(direct computation here), so that $K$ should have dimension 2, too. Defining an operator $N$ over $Null(T-2)^2$ as $N=T-2$, we have that $N$ is nilpotent(it "dies" when is raised to two), and also $[T]_B=[N]_B+2[I]_B$, for every basis of L. In particular, exists a "cyclic base for N" in L, i.e., a basis for which $N$ can be represented as \begin{bmatrix} 0 & 0 \\ 1 & 0 \\ \end{bmatrix} And then, the matrix of T with respect to that same basis is
\begin{bmatrix} 2 & 0 \\ 1 & 2 \\ \end{bmatrix}
Since the rest can't be factored in linear factors, I guess we can just try to find a basis for K and the matrix will look sthg like this:
$D=\begin{bmatrix} 2 & 0 & 0 & 0 \\ 1&2&0 & 0\\ 0 &0&*& *\\ 0&0&*&* \end{bmatrix}$
The problem is, my friends, that when I compute $B^2-2B+2I$ through WA, and then row reduce it, it says that it is equivalent to the identity matrix, so I can't extract any basis for K. What's wrong here?
You are getting something wrong. The point is, that the eigenvalues of this expression are $2,1+i$ and $1-i$, of which the eigenvalue $2$ has multiplicity $2$, hence the Jordan canonical form $J$ will look like:$$ \begin{pmatrix} 2 \qquad 1\qquad 0\qquad 0 \\ 0\qquad 2 \qquad 0 \qquad 0 \\ 0\qquad 0 \ \quad 1-i \ \quad 0 \\ \ \ 0\qquad 0 \qquad 0 \quad {1+i} \end{pmatrix} $$ This is the Jordan form of the matrix. Now, all you need to do, is to find the following: one eigenvector of $1+i$, one eigenvector of $1-i$ and two linearly independent eigenvectors of $2$.
For this, you need to solve the equations:
$B\vec{x}=(i+1)\vec{x}$
$B\vec{x}=(1-i)\vec{x}$
$B\vec{x}=2\vec{x}$ (will give you two linearly independent solutions)
Once you normalize all the above vectors, then adjoining them as columns will give you the transformation matrix $S$, such that $B=S^{-1}JS$.