Let $P:\ell^1(\mathbb{Z}^d) \rightarrow \ell^1(\mathbb{Z}^d)$ be given by

$$(Pz)(x)=\sum_{y \tilde \ x} \frac{1}{2d} z(y)$$

where the tilde indicates that $y$ is a neighboured vertex of $x.$

I would like to know: Let $\lambda \notin [-1,1]$ is it true that $\operatorname{ran}(P-\lambda)$ is dense in $\ell^1.$ In other words, does this operator have no residual spetrum outside $[-1,1]?$

Background: If we look at $P$ on $\ell^2(\mathbb{Z}^d)$ this operator is even self-adjoint and thus there is no residual spectrum at all.

  • you mean that each vertex has $2^d$ neighbours? It looks strange. – Fedor Petrov Jul 15 '17 at 17:09
  • What for $d=3$? – Fedor Petrov Jul 15 '17 at 17:27
up vote 2 down vote accepted

Yes, it is true (if we replace $2^d$ to $2d$). If the range of $P-\lambda$ is not dense, there exists a bounded linear functional $\eta\in \ell^\infty=(\ell^1)^*$ which vanishes on this range. In other words, there exists a not identically zero bounded function $\eta$ on $\mathbb{Z}^d$ which satisfies $P^*\eta=P\eta=\lambda\eta$. Consider two cases.

1) $|\lambda|>1$. Note that since $\eta(x)=\lambda^{-1}(2d)^{-1}\sum_{y\sim x} \eta(y)$, the sum of coeffcients on the right is $\lambda^{-1}$. Iterating this, we see that each $\eta(x)$ is a linear combination of other values of $\eta$ with sum of coefficients $\lambda^{-n}$ for any $n$. Since $\lambda^{-n}$ tends to 0 and $\eta$ is bounded, we conclude that $\eta\equiv 0$, a contradiction.

2) $\lambda$ is not real. Denote $\mu=2d \lambda$. Take large $N$ and denote $\Omega_N=[-N,N]^d\cap \mathbb{Z}^d$. Consider the sum $S_n=\sum_{x\in \Omega_N,y\sim x} \eta(y)\overline{\eta(x)}$. It equals, if we replace $\sum_{y\sim x} \eta(y)$ to $\mu \eta(x)$, to $S_n=\mu\sum_{x\in \Omega_n}|\eta(x)|^2$. On the other hand, all internal edges have real contribution to this sum. For boundary edges use an estimate $2|\eta(y)\overline{\eta(x)}|\leqslant |\eta(x)|^2+|\eta(y)|^2$. Then denoting $\mu=u+iv$ and taking the imaginary part, we get $|v|\sum_{x\in \Omega_n} |\eta(x)|^2\leqslant 2\sum_{x\in \Omega_{n+1}\setminus \Omega_{n-1}} |\eta(x)|^2$. Now denote $a_n=\sum_{x\in \Omega_n} |\eta(x)|^2$. This sequence is positive for large $n$, increases and satisfies an inequality $a_{n+1}-a_{n-1}\geqslant ca_n$ for certain positive $c=|v|/2$. It follows by induction that $a_n\geqslant c_0 \rho^n$, where $\rho >1$ is a root of $\rho ^2-1=c\rho$. On the other hand, since $\eta$ is bounded, $a_n=O(n^d)$. A contradiction.

  • thank you for the answer, but sorry, where does this estimate after "Then denoting $\mu=u+iv$, we get" come from. Would you mind spending a few more words on this? Otherwise, I could completely follow your thoughts. – BaoLing Jul 15 '17 at 18:38
  • It looks to me that the Wiener algebra on the torus is as good as on the circle, so the operator is actually invertible. Am I missing anything? – fedja Jul 15 '17 at 19:45
  • @fedja could I please ask you to elaborate on this in an answer, this does sound very interesting to me. I did suspect invertibility as well, but wanted to look for an easier conjecture first. – BaoLing Jul 15 '17 at 19:50
  • @fedja or would you mind just briefly outlining invertibility as a comment below my question? – BaoLing Jul 15 '17 at 22:40
  • @BaoLing Just google Wiener's theorem/Wiener algebra and mimic the proof you like most :-) – fedja Jul 16 '17 at 0:47

Your Answer


By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

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