Inverse operator being bounded and linear

120 Views Asked by At

So, here is my next edited attempt to solve this problem. conditions for the existence of the inverse of spectral operator as a bounded linear operator

Let $X$ be a Hilbert space with a countable orthonormal base $[e_1, e_2, \cdots]. $ Suppose $Ax = \sum_{n=1}^{\infty} \alpha_n \langle x,e_n \rangle e_n$. Find the conditions under which $A^{-1}$ exists as a bounded linear operation.

First, $e_n$ is an eigenvector of $A$ correspoding the eigenvalue $\alpha_n$. Now we must have $A^{-1}(A(x)) = x, \forall x \in X.$ Picking $x= e_n$ leads to $A^{-1}(\alpha_n e_n)=e_n$. Since $A^{-1}$ is supposed to be linear, we get $A^{-1}(e_n) = \alpha_n^{-1}e_n$, which makes $e_n$ an eigenvector of $A^{-1}$ as well, correspoding to $\alpha_n^{-1}$.

Now, to bound the norm of $A^{-1}$, a user suggested the following in the original post:

$$ \|A\|=\sup\{|\alpha_n|:\ n\}. \quad (*) $$ So $A^{-1}$ will be bounded if $$ \sup\{|\alpha_n|^{-1}:\ n\}<\infty. \quad (**) $$

Now, I understand the procedure leading to (*) as we have a spectral form for $A$. But where does (**) come from?

So far, we have found the eigenvalues of $A^{-1}$. How does it bound the norm of $A^{-1}$ from above? I mean, as far as I know, in general, the norm of an operator is greater than or equal to the largest absolute value of its eigenvalues. So, at the end of the day, I guess, I must prove $||A^{-1}|| \leq \sup\{|\alpha_n|^{-1}:\ n\}$. How is it done? I'd appreciate some elaboration on this.

Edit: From the very definition of the norm operator, I can see that $$||A^{-1} || \geq |\alpha_n^{-1}| \Rightarrow ||A^{-1} || \geq \sup \{ |\alpha_n^{-1}| \} $$ But then bounding $|\alpha_n^{-1}| $ from above does not necessarily bound $||A^{-1} ||.$ So, I guess I need to show the reverse inequality. But how?

1

There are 1 best solutions below

0
On

As a consequence of the "Bounded inverse theorem" it suffices to prove that $A$ is bounded and bijective.

Boundedness: $$||Ax||^2 = \sum_{n = 1}^\infty |\alpha_n|^2|\langle x,e_n \rangle|^2 \leq (\sup_n|\alpha_n|)^2 ||x||^2$$ So $A$ is bounded if $\{\alpha_n\}_n$ is bounded.

Injectivity: $$||Ax||^2 = 0 = \sum_{n = 1}^\infty |\alpha_n|^2|\langle x,e_n \rangle|^2$$ gives $|\alpha_n||\langle x,e_n \rangle| = 0$ for every $n$ and $x$ = 0 if $\alpha_n \neq 0$ for every $n$.

Surjectivity: let $y \in X$ then we want to define $x \in X$ such that $Ax = y$ as follow $$x := \sum_{i = 1}^\infty \frac{1}{\alpha_n} \langle y,e_n \rangle e_n$$ which make sense if $\{\frac{1}{\alpha_n}\}_n$ is bounded.

Summarizing you need $\{\alpha_n\}_n$ to be a bounded sequence that never vanish for which $0$ is not an accumulation point.