Suppose $H$ is a Hilbert space with orthonormal basis $\{e_n\}$. Suppose that $\{\alpha_n\}$ is a sequence of complex numbers with $\lim_{n\rightarrow\infty} \alpha_n = 0$, and define $T: H\rightarrow H$ by $Tx = \sum_{n=1}^\infty \alpha_n\langle x, e_n\rangle e_n$.

When does the equation $(T-\alpha I)z = x$ have a solution $z$?

By comparing coefficients we see that if $z = \sum \beta_n e_n$, then we need $$\beta_n = \frac{1}{\alpha_n -\alpha} \langle x, e_n \rangle.$$

So provided that $\langle x, e_n\rangle = 0$ whenever $\alpha_n = \alpha$ then we're fine. At least, this is what I got at first, and is also essentially what the answer here Equation in Hilbert space says. However, what happens if $\alpha = 0$? Then the sequence $\{\beta_n\}$ is unbounded since $\alpha_n\rightarrow 0$, and hence not square summable, so the definition of $z$ doesn't converge.

This would suggest to me that the equation, which is now just $Tz = x$, only has a solution if $\langle x, e_n\rangle \neq 0$ for only finitely many $n$, in order for $z$ to converge. But this means that $T$ is finite rank, but that's certainly not necessary because it could be the case that all the $\alpha_n$ are nonzero, in which case $T$ is still a well defined and bounded operator.

What am I missing here?


First of all even if $Tz = x$ has only solutions if $\langle x,e_n\rangle\neq 0$ for only finitely many $n$ then still every $e_n$ for which the corresponding $\alpha_n$ fulfills $\alpha_n \neq 0$ is in the range of $T$ since $T\frac{1}{\alpha_n} e_n = e_n$. Hence the $\dim\operatorname{ran} T$ is infinite if there is no $n_0\in\mathbb{N}$ such that $\alpha_n = 0$ for all $n\geq n_0$.

Anyway there is something you forgot: $\beta_n = \frac{1}{\alpha_n}\color{red}{\langle x,e_n\rangle}$ so $\beta_n$ must not be unbounded as $n\to\infty$ since the red marked $\langle x,e_n\rangle$ could shrink faster than $\alpha_n$.

for example take $\langle x,e_n\rangle = \frac{\alpha_n}{n}$ then we receive $\beta_n = \frac{1}{n}$ for all $n\in\mathbb{N}$. I hope this solves your confusion.

  • $\begingroup$ So how do I ensure that my $z$ is a solution given the information that I have? $\endgroup$ – IAlreadyHaveAKey May 27 '17 at 14:16
  • $\begingroup$ Suppose that $\langle x, e_n\rangle = 0$ whenever $\alpha_n = 0$, in the case where $\alpha = 0$. Does the equation have a solution? Or do we need more assumptions on $x$. $\endgroup$ – IAlreadyHaveAKey May 27 '17 at 14:24
  • $\begingroup$ In this case you also need that $\frac{1}{\alpha_n} \langle x,e_n\rangle$ (this is well defined since we say $\frac{0}{0} =0$ for now) is quadratic summable. since we want $\sum_{n=1}^\infty \frac{1}{\alpha_n} \langle x,e_n\rangle e_n$ to converge. $\endgroup$ – Nathanael Skrepek May 27 '17 at 14:29

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.