I am doing some self-study in preparation for an exam, and in this problem I am given the following matrix in $R^{3×3}$:
$\begin{pmatrix} 1&0&1\\ 0&1&-1\\ 0&x&y\\ \end{pmatrix}$
I need to determine the set of x and y values so that the only eigenvalue of this matrix is one.
At first glance, we have x = 0 and y = 1, since that would make this into a unipotent matrix and I know that the only eigenvalue of a unipotent matrix equals 1.
Now consider the characteristic polynomial of the matrix, which I have determined to be
$p(\lambda) := -\lambda^3 + (y + 2)\lambda^2 - (x + 2x + 1)\lambda + (x+y)$.
We see that $a_2 = (y + 2)$ and $a_0 = (x+y)$, and that $a_2$ is the sum of eigenvalues and that $a_0$ is the product of eigenvalues, both of which must be one. Solving for x and y gives us
x = 2 and y = -1.
For the third method, consider the companion matrix of the characteristic polynomial.
Transforming $p(\lambda) := -\lambda^3 + (y + 2)\lambda^2 - (x + 2x + 1)\lambda + (x+y)$ to the monic polynomial $p(\lambda) := \lambda^3 - (y + 2)\lambda^2 + (x + 2x + 1)\lambda - (x+y)$ and then determining the companion matrix, we get:
C(p) :=
$\begin{pmatrix} 0&0&x+y\\ 1&0&-(x+2x+1)\\ 0&1&y+2\\ \end{pmatrix}$
If the sum of all columns equals 1, then one of the eigenvalues of this matrix equals 1. Summing the third column gives us a value of 1 for all x and y. Therefore, knowing that the eigenvalues of C(p) are the roots of $p(\lambda)$, we see that 1 is an eigenvalue of the original matrix for all x and y.
To see that it is the only real eigenvalue, I factor $(\lambda - 1)$ out of the original monic polynomial and solve for the remaining quadratic coefficients. I get:
$p(\lambda) := (\lambda - 1)(\lambda^2 - (y + 1)\lambda + (x+y))$
Solving for the quadratic coefficients gives us:
$\lambda_{2,3} = \frac{(y+1) ± \sqrt{(y^2 - 2y -4x + 1)}}{2}$
So if we find x and y in the real numbers so that the term under the radical is less than zero, we have found x and y where 1 is the only real eigenvalue.
From this point I am stuck.
In your second method, the sum of the eigenvalues shouldn't be $1$ but actually $3$, since we want $1$ to be an eigenvalue with multiplicity $3$ so sum of the eigenvalues is $1+1+1 = 3$.
Regarding the third method, the eigenvalues could be complex numbers but what we want is to demand that $\lambda_{2,3} = 1$. So we set
$$ \begin{align} y^2-2y-4x+1 &= 0 \\ y+1 &= 2. \end{align} $$
So we get only the solution $x=0, y=1$.