It is easy to show that if a bounded linear operator on a separable Hilbert space $ \mathcal{H}$ has:

(i) eigenvectors that constitute an orthonormal basis

(ii) eigenvalues that are real that go to zero

Then the operator must be compact and symmetric.

My question is, suppose I only know that the operator is linear. Can I still show that the operator is compact and symmetric?

My conjecture is no. Let $\mathcal{H}$ have basis $\{e_i\}_{i=1}^\infty$. Define our operator $T$ such that $Te_k = \frac{1}{k^{1/3}}e_k$. Then the eigenvalues tend to $0$.

Consider $h = \sum_{k=1}^\infty \frac{1}{k^{2/3}}e_k$. Then $h \in \mathcal{H}$ since $\lVert h \rVert^2 = \sum_{k=1}^\infty k^{-4/3} < \infty$.

Suppose T is continuous. Then $\lVert Th \rVert = \sum_{k=1}^\infty k^{-1}$ which diverges, so $T$ is not bounded. Does this argument work? Is $T$ a linear operator?


The operator is not necessarily bounded. The example you give is unclear, specifying the values on an orthonormal basis does not uniquely determine a linear operator (it does for a bounded linear operator).

To get unbounded examples you can proceed as follows. Complete $(e_k)_{k\in\mathbb{N}}$ to a Hamel basis $(e_i)_{i\in\mathbb{N}\sqcup I}$ of $\mathcal{H}$ with $\|e_i\|=1$ for all $i\in I$. If $\mathcal{H}$ is infinite-dimensional, then $I$ is infinite (in fact, uncountable). Let $\phi\colon \mathbb{N}\to I$ be an injective function and define $Te_k=\lambda_k e_k$ for $k\in\mathbb{N}$, $Te_{\phi(l)}=l e_1$ for $l\in\mathbb{N}$ and $Te_i=e_2$ for $i\in I\setminus\phi(\mathbb{N})$. Then $(e_k)$ is an orthonormal basis of eigenvectors, we can choose a null sequence $(\lambda_k)$ for the eigenvalues and yet the operator $T$ is not bounded.

The operator $T$ constructed above will not be symmetric. In fact, every symmetric everywhere defined linear operator on a Hilbert space is bounded.


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.