1
$\begingroup$

Consider the Hilbert space $\mathcal{H}=l^2(\mathbb{Z})$ and define the left shift operator $\mathcal{L}:\mathcal{H} \rightarrow \mathcal{H}$ by $$ \mathcal L (a_n) = (b_n) \qquad \text{ where } \qquad b_n=a_{n+1} ,$$ so $\mathcal L$ is the right-shift operator.

My first issue is with notation, what does $\mathcal L -zI$ mean for $z \in \mathbb{C}$? Does this mean $$\mathcal L-I((a_n)) = (a_{n+1}-a_n)?$$

Moreover, I'm supposed to show $\mathcal L-zI$ is not invertible for $|z| =1$ by showing that $$(a_n) = \frac{1}{|n|+1}$$ is in $\ell^2(\mathbb{Z})$, yet not in the image of $\mathcal L$ for $z=1$, and then adjust for other values of $z$. Though, I don't quite understand how this shows that $\mathcal L -zI$ isn't surjective.. was this perhaps a typo? Should I show $(a_n)$ isn't in the image of $\mathcal L-I$ and then adjust for values of $z$? It doesn't seem like it would make sense because surely it's in the image of $\mathcal L$, as the left shift operator is $\mathcal L$'s adjoint and inverse, with both being norm-preserving, so this wouldn't make sense.

I don't quite see how to begin with this either way, so any hints would be wonderful. Indeed, this is homework. (The next part involves the construction of an inverse to $\mathcal L -z I$ using the power series for $(1-x)^{-1}$.)

$\endgroup$
2
$\begingroup$

More elementary way:

a. If $\lvert\,\lambda\rvert>1$, then $\lambda\not\in\sigma(\mathcal L)$. This is clear as the series $(\lambda-\mathcal L)^{-1}=\lambda^{-1}\sum_{n=0}^\infty (\lambda^{-1}\mathcal L)^n$ converges.

b. If $\lvert\,\lambda\rvert<1$, then $\lambda\not\in\sigma(\mathcal L)$. This is clear as the series $(\lambda-\mathcal L)^{-1}=-{\mathcal L}^{-1}\sum_{n=0}^\infty (\lambda\mathcal L^{-1})^n$ converges.

c. If $\lambda=\mathrm{e}^{i\vartheta}$ then $\lambda\in\sigma(\mathcal L)$. This can be shown as follows. For every $\varepsilon>0$, find a vector $\boldsymbol{a}=(a_k)_{k\in\mathbb Z}$, such that $\sum_{k\in\mathbb Z}\lvert a_k\rvert^2=1$, while $$ \|\mathcal L\boldsymbol{a}\|^2= \sum_{k\in\mathbb Z}\lvert a_{k+1}-\mathrm{e}^{i\vartheta}a_k\rvert^2<\varepsilon^2. $$ To achieve that, look for $\boldsymbol{a}=(a_k)_{k\in\mathbb Z}$, with elements the Fourier coefficients of a smooth $2\pi-$function, which is equal to 1 at $x=\vartheta$, and very small elsewhere. In particular, consider $p_n(x)=n^{-1/2}\sum_{\lvert k\rvert\le n}\mathrm{e}^{-ik\vartheta}\mathrm{e}^{ikx}$, which approximates the $\delta-$function at $x=\vartheta$.

$\endgroup$
  • $\begingroup$ Does the notation $\mathcal{L}^n$ mean apply $\mathcal L$ $n$ times, or $(a_{n+1}^n)$? That is, how do I apply $\frac{\mathcal L^k}{\lambda^k}$ to $(a_n)$? $\endgroup$ – Anthony Peter Nov 13 '15 at 5:40
  • $\begingroup$ @AnthonyPeter $\big({\mathcal L}^n(a_k)_{k\in\mathbb Z}\big)_j=a_{j+n}$. $\endgroup$ – Yiorgos S. Smyrlis Nov 13 '15 at 11:33
3
$\begingroup$

Answer. $S=\{z\in\mathbb C: \lvert\,z\rvert=1\}$.

Explanation. The space $\ell^2(\mathbb Z)$ is isometric to $L^2(\mathbb T)$, where $\mathbb T$ is the unit circle (equivalently, the domain of $2\pi-$periodic functions), and this is done by the transfomation $$ \varPhi\big((a_k)_{k\in\mathbb Z}\big)=\sum_{k\in\mathbb Z}a_k\mathrm{e}^{ikx}, $$ as $\|(a_k)_{k\in\mathbb Z}\|_{\ell^2}=\big\|\sum_{k\in\mathbb Z}a_k\mathrm{e}^{ikx}\big\|_{L^2}$. The shift $S$ in $L^2(\mathbb T)$ becomes a multiplication operator, by $\mathrm{e}^{ix}$, and hence, its the spectrum coincides with its essential range.

$\endgroup$
  • $\begingroup$ Thank you for this answer. Unfortunately, we don't have the tools of the trigonometric functions available at this time in the course, still, +1 $\endgroup$ – Anthony Peter Nov 12 '15 at 9:24
1
$\begingroup$

HINT.If $a_{n+1}-z a_n=b_n$ then $a_2=z a_1+b_1$, $a_3=z a_2+b_2=z^2 a_1+z b_1+b_2$ ,$a_4=z a_3+b_3=z^3a_1+z^2b_1+z b_2+b_3$, etc. If $|z|=1$ and $b_n= z^{n}/(1+|n|)$, what happens to $|a_n-z^{n-1}a_1|$ for large positive $n$?

$\endgroup$
  • $\begingroup$ I didn't mean to imply that there was anything wrong with anyone else's argument.I usually find the most difficult method. $\endgroup$ – DanielWainfleet Nov 12 '15 at 18:57
  • $\begingroup$ Actually, could you slightly further your hint? $\endgroup$ – Anthony Peter Nov 13 '15 at 3:18
  • $\begingroup$ Is it because $|a_n -z^{n-1}a_1|$ can grow arbitrarily large in modulus, and thus so can $a_n$? Since the series $|b_n|$ diverges? $\endgroup$ – Anthony Peter Nov 13 '15 at 3:41
  • $\begingroup$ But if this were correct what's the need for considering $|a_n-z^{n-1}a_1|$ at all? $\endgroup$ – Anthony Peter Nov 13 '15 at 4:00
  • $\begingroup$ @Anthony Peter . Hence $L-z I$ is not invertible because it's not a surjection because if $b_n=z^n/(|n|+1)$ there is no $a$ such that $(L-z I)a=b$. $\endgroup$ – DanielWainfleet Nov 13 '15 at 6:07
1
$\begingroup$

The operator $L$ maps $\{ a_n \}_{n=-\infty}^{\infty}$ to $\{ a_{n+1} \}_{n=-\infty}^{\infty}$. If $e_n$ is the standard basis element defined as a sequence of all $0$'s except for a $1$ in the $n$-th place, then $Le_n = e_{n-1}$, which is why $L$ would be considered to be a left shift. And this follows from the definition $L\{ a_n \} = \{ a_{n+1} \}$: the $k$ index spot in $\{ a_n \}$ becomes the $(k-1)$ index spot in $\{ a_{n+1} \}$. I prefer dealing with the basis elements because "left shift" makes sense in terms of the basis, and because I was taught to work in terms of the basis in Linear Algebra. The standard basis $\{ e_n \}_{n=-\infty}^{\infty}$ is a complete orthonormal basis of $\ell^2(\mathbb{Z})$. In terms of this basis, $$ L x = L \sum_{n=-\infty}^{\infty}(x,e_n)e_n = \sum_{n=-\infty}^{\infty}(x,e_n)e_{n-1}. $$ That makes $L$ look more like an operator from Linear Algebra, and it's rigorously correct.

The operator $(L-zI)$ means $$ (L-zI)\{ a_n \} = L\{ a_n\} - z\{ a_n \} = \{ a_{n+1}-za_n \}. $$ You can alternatively view the action of $(L-zI)$ on the standard basis elements: $$ (L-zI)e_n = e_{n-1}-ze_{n}. $$ Try showing that the following has no solution $x\in\ell^{2}$ for $|z|=1$: $$ (L-zI)x = \sum_{n=-\infty}^{\infty}\frac{\overline{z}^n}{|n|+1}e_n $$ If it did, \begin{align} \sum_{n=-\infty}^{\infty}(x,e_n)(e_{n-1}-ze_n)&=\sum_{n=-\infty}^{\infty}\frac{\overline{z}^{n}}{|n|+1}e_n \\ \sum_{n=-\infty}^{\infty}\{(x,e_{n+1})-z(x,e_{n})\}e_n &= \sum_{n=-\infty}^{\infty}\frac{\overline{z}^{n}}{|n|+1}e_n \\ (x,e_{n+1})-z(x,e_{n})&=\frac{1}{|n|+1} \\ (x,e_n)-z(x,e_{n-1})&=\frac{\overline{z}}{|n-1|+1} \\ (x,e_{n-1})-z(x,e_{n-2})&=\frac{\overline{z}^{2}}{|n-2|+1} \end{align} Multiply these equations by powers of $z$ and add them to obtain a telescoping series. Then, knowing $(x,e_{n-k})\rightarrow 0$ as $k\rightarrow\infty$, $$ (x,e_{n+1})=\sum_{k=0}^{\infty}\frac{1}{|n-k|+1} $$ This is a definite problem because the right side diverges. So $L-zI$ is not surjective for any unimodular $z$.

$\endgroup$

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.