Why is that sometimes eigenvalue decomposition produces a rotational matrix, other times it does not

505 Views Asked by At

Suppose we want to perform eigenvalue decomposition on this matrix:

$$P = \begin{bmatrix} 1.5 && -0.5\\ -0.5 && 1.5 \end{bmatrix}$$

I obtain the corresponding eigenvectors as $\lambda_1 = 1, \lambda_2 = 2$, and eigen-vector $v_1 = [1, 1]^T$, $v_2 = [1, -1]^T$.

So our $Q$ as in $P = Q\Lambda Q^T$ is given by

$$Q = \begin{bmatrix} 1 && 1\\ 1 && -1 \end{bmatrix}$$

Which is not a rotational matrix after we normalize it!

But we can alternatively pick our eigenvectors as $v_1 = [1, 1]^T$, $v_2 = [-1, 1]^T$. Notice $v_2$ has changed by a sign

Then obviously $$Q = \begin{bmatrix} 1 && -1\\ 1 && 1 \end{bmatrix}$$

Is a rotational matrix after normalization. In fact, it is trivial to see it rotates by $45^\circ$

Can someone resolve this paradox?