MathOverflow is a question and answer site for professional mathematicians. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Let $H$ be an infinite dimensional separable Hilbert space.

Definition: The commutant $\mathcal{S}'$ of a subset $\mathcal{S} \subset B(H)$ is $ \{A \in B(H) : AB=BA \ , \ \forall B \in \mathcal{S} \} $.

Definitions : An operator $A \in B(H)$ is :

  • Irreducible (Halmos 1968) if its commutant $\{ A\}'$ does not contain projections other than $0$ and $I$ ($A \ne A_{1} \oplus A_{2}$, $A$ generates $B(H)$ as von Neumann algebra : $\{A,A^{*}\}''=B(H)$).

  • Nonnormal if $\{ A\}'$ does not contain $A^{*}$ (i.e. $AA^{*} \ne A^{*}A$).

  • Noncompact commuting if $\{ A\}'$ does not contain a compact operator.

Definition : The spectrum $\sigma(A)$ of $A \in B(H)$ is $\{\lambda \in \mathbb{C} : A - \lambda I \text{ not bijective} \}$.
It decomposes as follows:

  • Point spectrum: $\sigma_{p}(A) = \{\lambda \in \mathbb{C} : A - \lambda I \text{ not injective} \}$
  • Continuous spectrum: $\sigma_{c}(A) = \{\lambda \in \mathbb{C} : A - \lambda I \ \text{injective, dense nonclosed range} \}$
  • Residual spectrum: $\sigma_{r}(A) = \{\lambda \in \mathbb{C} : A - \lambda I \ \text{ injective, nondense range} \}$

The spectrum of $A$ is strictly continuous if $\sigma(A) = \sigma_{c}(A)$.


  • Let $S$ be the bilateral shift defined on $H = l^{2}(\mathbb{Z})$ by $S.e_{n} = e_{n+1} $.
    Its spectrum is strictly continuous : $\sigma(S) = \sigma_{c}(S) = \mathbb{S}^{1}$.
    It's also a unitary operator ($SS^{*} = S^{*}S = I$), so a fortiori a normal operator.
    It is noncompact commuting and reducible.
  • Let $T$ be the unilateral shift defined on $H = l^{2}(\mathbb{N})$ by $T.e_{n} = e_{n+1} $.
    Its spectrum is not strictly continuous because $0 \in \sigma_{r}(T)$.
    It's a nonnormal operator because $[T^{*},T].e_{0} = e_{0}$.
    It is noncompact commuting and irreducible.
  • Let $V$ the Volterra operator defined on $H= L^{2}[0,1]$ by $(V.f)(t)=\int_0^tf(x)dx$.
    Its spectrum is strictly continuous $\sigma(V) = \sigma_{c}(V) = \{ 0\}$.
    It is compact, irreducible and nonnormal (see here).
  • Let $p$ be a non-constant polynomial (see here).
    Then $p(V)$ is nonnormal, compact commuting and irreducible.
    Its spectrum is strictly continuous $\sigma(p(V)) = \sigma_{c}(p(V)) = \{ p(0)\}$.
    It's compact commuting, nonnormal and irreducible.
  • Let $S \oplus V$ defined on $l^{2}(\mathbb{Z}) \oplus L^{2}[0,1]$.
    It is reducible, compact commuting, nonnormal and with spectrum strictly continuous.

If you find a mistake, thank you let me know in comment.

The main question: Is there an irreducible, noncompact commuting and nonnormal operator, with spectrum strictly continuous ?

Bonus questions : How classify these operators ?

share|cite|improve this question
One more possible example (though not an answer to your question): I think that if you take any non-zero element of the von Neumann algebra of a discrete group $\Gamma$, thought of as an operator on $\ell^2(\Gamma)$, then I think you get noncompact commuting and strictly continuous spectrum, and clearly you can arrange for non-normal. But irreducibility is not clear to me right now – Yemon Choi Jul 19 '13 at 18:56
@YemonChoi : Thank you for this comment. What's not clear with irreducibility ? If the group $\Gamma$ is ICC (infinite conjugacy classes), it generates a $II_{1}$ factor, and there are plenty of projections in its commutant, so there is no irreducibility for a single element. Next, for any discrete group $\Gamma$, if we take an irreducible representation $H$, the von Neumann algebra it generates is the whole $B(H)$. – Sebastien Palcoux Jul 19 '13 at 20:12
Could the "downvoter" and "closer" indicate what's wrong ? Thank you. – Sebastien Palcoux Jul 20 '13 at 7:54
Cross-posted on MSE. Hi Sébastien, you are supposed to let people know when you ask the same question on both sites. – 1015 Jul 21 '13 at 13:43
Definitely a question for MO. Don't cross post unless you say so and give the URL for the other post. – Bill Johnson Jul 23 '13 at 19:12
up vote 6 down vote accepted

Essentially I think weighted shifts should be a sufficiently rich class of operators. Consider, for instance, the following example.

Take the doubly infinite sequence $$ w_k=\left\{\begin{array}{ll} 2 & \text{if } k<0 \\ 1 & \text{if } k\geq 0 \end{array}\right. $$ and let $W$ be the weighted shift on $\ell^2(\mathbb{Z})$ defined by $$ We_j=w_je_{j+1} \qquad \forall j\in\mathbb{Z}\qquad (\text{here }(e_j)\text{ is the usual basis}).$$ Then

  1. W has strictly continuous spectrum. This is an almost identical argument the usual bilateral shift.

  2. W is not normal. The adjoint is given by $W^\ast e_j=w_{j-1}e_{j-1}$ and so $$W^\ast We_j=w_j^2e_{j},\quad \text{but}\quad WW^\ast e_j = w_{j-1}^2e_j.$$

  3. $W$ is irreducible. This follows from Corollary 2 of R. Gellar, Operators commuting with a weighted shift, Proc. Amer. Math. Soc. 23 (1969), 538-545.

  4. I do not seem to be able to show $W$ is non-compact-commuting. See Bill's comment below.

share|cite|improve this answer
It is non-compact-commuting. First, $\|Wx\| \ge \|x\|$, so if $S$ is compact and $S e_n \not= 0$ then $\|W^k S e_n\|$ is bounded away from zero. However, $(W_k e_n)_k$ tends weakly to zero, so $ (SW_k e_n)_k$ tends in norm to zero. – Bill Johnson Jul 23 '13 at 18:20
Thanks, I think it would've taken me a while to think of something so slick! – Ollie Margetts Jul 24 '13 at 5:16

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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