If A is a symmetric matrix whose eigenvalues have absolute values that are less than $1$, then: $$\det(I-A)\neq0$$ where $I$ is identity matrix.
Why is that inequality correct?
$\begingroup$
$\endgroup$
6
-
$\begingroup$ This is false, for $A=I$. $\endgroup$– Dietrich BurdeMar 28, 2017 at 11:48
-
$\begingroup$ It is not correct as written. Choose $A=I$. This matrix is clearly symmetric, has some eigenvalues (equal to $1$), but $det(I-A)=0$. $\endgroup$– SimonMar 28, 2017 at 11:48
-
$\begingroup$ ok, I edited question $\endgroup$– mokebeMar 28, 2017 at 11:49
-
4$\begingroup$ This holds when $1$ is not an eigenvalue of $A$. $\endgroup$– StubbornAtomMar 28, 2017 at 11:51
-
2$\begingroup$ This is a simple consequence of the definition of eigenvalues: $\lambda I-A$ is singular iff $\lambda$ is an eigenvalue of $A$. $\endgroup$– amdMar 28, 2017 at 19:21
|
Show 1 more comment
2 Answers
$\begingroup$
$\endgroup$
1
If you see "symmetric" (even normal is enough), diagonalize! \begin{align} \det(I - A) &= \det(UIU^* - U\Lambda U^*) \\&= \det(U(I-\Lambda) U^*) \\&= \det(U)\det(I-\Lambda)\det(U^*) \\&= \det(I-\Lambda) \\&= \prod\limits_{j=1}^n (1-\lambda_j) \\\det(I - A) &= 0 \iff \exists j\colon\lambda_j = 1 \end{align}
answered Mar 28, 2017 at 12:14
-
$\begingroup$ Now I feel bad for not including that the last line is naturally true for any $A$, but I don't want to bump this post. $\endgroup$ Mar 28, 2017 at 17:03
$\begingroup$
$\endgroup$
Mosquito-nuking solution: $$\rho(A) < 1\implies \sum_{n=0}^{\infty}A^n\text{ is convergent}$$ (see Neumann series and spectral radius) and $$(I - A)^{-1} = \sum_{n=0}^{\infty}A^n\text{ exists}.$$
answered Mar 31, 2017 at 7:59
Martín-Blas Pérez PinillaMartín-Blas Pérez Pinilla
40.1k4 gold badges42 silver badges86 bronze badges