1
$\begingroup$

Let $A_i:\ell^2\rightarrow \ell^2$ be two operators given as follows:

$A_1x=(0,x_1,0,\frac{x_2}{2},0,\frac{x_3}{3},...)$ and $A_2x=(x_1,x_1,x_2,x_2,x_3,x_3,...)$

Compute the norm and the adjoint operator and decide whether these operators are compact or not.

1)

Norm: We have obviously $||A_1x||_{\ell^2}\leq ||x||_{\ell^2}$ for all $x\in \ell^2$, hence $||A_1||<1$. But for $x=(0,0,0,..)$ we have equality, hence $||A_1||=1$.

Adjoint operator: $A_1^*$ must fulfill $<A_1x,y>=<x,A_1^*y>$ for all $x,y\in \ell^2$. So we must have $\sum_{n=1}^{\infty}|(A_1x)_ny_n|=\sum_{n=1}^{\infty}|x_n(A_1^*y)_n|$.

But I don't know how to continue from here..

Compactness: The sequence of operators $A_nx=(0,x_1,0,\frac{x_2}{2},0,...,\frac{x_n}{n},0,0,..)$ is a sequence of finite rank operators, which converges to $A_1$ because:

$||A-A_n||^2=\sup_{||x||_{\ell^2}\leq1}\sum_{m=n+1}^{\infty}(|\frac{x_m}{m}|)^2\leq \sup_{||x||_{\ell^2}\leq1}(|\frac{1}{n+1}|)^2\sum_{m=n+1}^{\infty}|x_m|^2\leq(\frac{1}{n+1})^2\rightarrow 0$

2)

Norm: Each addend of $||A_2x||$ appears exactly twice in $||x||$ and both values are of course finite, hence $||A_2||=\frac{1}{2}$

Adjoint: No idea as in 1)

Compactness: The method from 1) did not work here for me

I hope someone can go through it and help me for the points which are left. Thanks :)

$\endgroup$
  • $\begingroup$ $\|A_{2}x\|^{2}=2\sum_{n=1}^{\infty}|x_{n}|^{2}=2\|x\|^{2}$. $\endgroup$ – DisintegratingByParts Feb 24 '15 at 19:39
  • $\begingroup$ For the adjoint of $A_2$, start with $A_2 = \sum_{n=1}^{\infty}\frac{1}{n}(x,e_n)e_{2n}$ and rewrite $(A_2x,y)$. $\endgroup$ – DisintegratingByParts Feb 24 '15 at 19:42
2
$\begingroup$

For every $x \in \ell^2$ we have \begin{eqnarray} \|A_1x\|^2&=&\sum_{n=1}^\infty\frac{x_n^2}{n^2} \le \sum_{n=1}^\infty x_n^2=\|x\|^2 \quad \mbox{ since } \quad n \ge 1,\\ \|A_2x\|^2&=&x_1^2+x_1^2+x_2^2+x_2^2+x_3^2+x_3^2+\ldots=2\sum_{n=1}^\infty x_n^2=2\|x\|^2, \end{eqnarray} and so
\begin{eqnarray} \|A_1\|&=&\sup_{\|x\|=1}\|A_1x\|\le \sup_{\|x\|=1}\|x\|=1,\\ \|A_2\|&=&\sup_{\|x\|=1}\|A_2x\|=\sup_{\|x\|=1}\sqrt{2}\|x\|=\sqrt{2}. \end{eqnarray} Because the vector $e_1=(1,0,\ldots)$ belongs to the unit ball of $\ell^2$ and $\|A_1e_1\|=1$, we deduce that $\|A_1\|=1$.

Unlike $A_1$, the operator $A_2$ is not compact. In fact, the unit ball $B_1=\{x\in \ell^2:\quad \|x\|=1\}$ is bounded, and its image through $A_2$ is the ball $B_{\sqrt{2}}=\{x\in\ell^2:\quad \|x\|=\sqrt{2}\}$ which, obviously is not relatively compact. As a matter of fact, the sequence $\{\sqrt{2}e_n\}_{n\ge 1}$, where $e_1=(1,0,\ldots),e_2=(0,1,0,\ldots),\ldots$, belongs to $B_{\sqrt{2}}$ but it has no convergent subsequence.

The idea is to find an operator $A_i^*$ such that $$ \langle A_ix,y\rangle=\langle x,A_i^*y\rangle \quad \forall x,y\in \ell^2. $$ For every $x,y\in \ell^2$ we have \begin{eqnarray} \langle A_1x,y\rangle&=&x_1y_2+\frac12x_2y_4+\frac13x_3y_6=\sum_{n=1}^\infty\frac{x_ny_{2n}}{n}=\langle(y_2,\frac{y_{4}}{2},\ldots),x\rangle,\\ \langle A_2x,y\rangle&=&x_1(y_1+y_2)+x_2(y_3+y_4)+x_3(y_5+y_6)+\ldots=\sum_{n=1}^\infty x_n(y_{2n-1}+y_{2n})\\ &=&\langle(y_1+y_2,y_3+y_4,\ldots),x\rangle, \end{eqnarray} hence $$ A_1^*x=2(\frac{x_2}{2},\frac{x_4}{4},\frac{x_6}{6},\ldots),\quad A_2^*x=(x_1+x_2,x_3+x_4,x_5+x_6,\ldots) \quad \forall x \in \ell^2. $$

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ Hello. I got a short question: You said that "$\{x\in \ell^2: ||x||=\sqrt{2}\}$ is obviously not relatively compact.". I would like to understand why this is obvious. Thanks:) $\endgroup$ – Duke Feb 25 '15 at 10:30
  • $\begingroup$ Second question: I showed compactness of $A_1$ by approximation with finite rank operators. Sometimes this does not work. Imagine that this would not work for $A_1$ could you provide a prove of compactness of $A_1$ with a different method? I would really appreciate it :) $\endgroup$ – Duke Feb 25 '15 at 11:44
  • $\begingroup$ I gave a short proof. $\endgroup$ – Mercy King Feb 25 '15 at 16:57
2
$\begingroup$

$A_1$:

Norm: $||A_1||=1$, because it is reached with $x=(1,0,0,\ldots)$ (note that for $x$ the null vector is not relevant. maybe it was a typo..) Adjoint: \begin{align} \langle A_1 x, y \rangle &= \langle (0,x_1,0,x_2/2,0,x_3/3,0,\ldots), (y_1,y_2,y_3,\ldots) \rangle \\ &= x_1 \bar{y_2} + \frac{x_2}{2} \bar{y_4} + \frac{x_3}{3} \bar{y_6} + \ldots \\ &= \langle (x_1,x_2,x_3,\ldots),(y_2,y_4/2,y_6/3,\ldots) \rangle, \end{align} hence $A_1^* x = (x_2,x_4/2,x_6/3,\ldots)$. Compactness: your argument works.

$A_2$:

Norm: $||A_2||=\sqrt{2}$, it's commented by T.A.E. Adjoint: a similar calculation as for $A_1$, it gives $A_2^* x = (x_1+x_2,x_3+x_4,\ldots)$. Compactness: No. Recall that $A$ is compact iff for every bounded sequence $(x_n)_n$, $(Ax_n)_n$ has a convergent subsequence. But the sequence $x_n=e_n$, $n\geq 1$ is bounded but $(Ax_n)_n$ does not have any convergent subsequence (two one's moving to the right). (Note aside: this operator is somehow similar to the identity operator, which is never compact in an infinite dimensional Hilbert space).

| cite | improve this answer | |
$\endgroup$

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.