2
$\begingroup$

Let $X$ and $Y$ be normed spaces, either both real or both complex; let $T \colon X \to Y$ be a linear mapping. If there is a real number $r > 0$ such that $$ \lVert T(x) \rVert_Y \leq r \lVert x \rVert_X \qquad \mbox{ for all } x \in X, $$ then $T$ is said to be a bounded linear operator.

One example of this operator is the following:

Let $X$ denote the normed space of all the real polynomials defined everywhere on the closed interval $[0, 1]$, with the maximum norm, and let $T \colon X \to X$ be the map $x \mapsto x^\prime$. Then this linear operator is unbounded.

Can we find any other examples of unbounded linear operators?

I know that every linear operator whose domain is a finite-dimensional normed space is bounded.

$\endgroup$
  • $\begingroup$ In $\ell^2$, send the $n$th standard unit vector $\mathbf{e}_n$ to $n\mathbf{e}_n$ and extend linearly. $\endgroup$ – max_zorn Dec 28 '17 at 6:52
  • $\begingroup$ @max_zorn, this is the same as differentiation via Fourier series. $\endgroup$ – fourierwho Dec 28 '17 at 8:05
  • $\begingroup$ How so, fourierwho? $\endgroup$ – max_zorn Dec 28 '17 at 8:34
  • $\begingroup$ Not quite the same. Taking Fourier series in reverse has the effect of sending $e_n \in \ell^2(\mathbb{Z})$ to $\exp(2 \pi i n x) \in \ell^2(S^1)$, and taking the derivative multiplies this by $2 \pi i n$. $\endgroup$ – Qiaochu Yuan Dec 28 '17 at 10:34
  • $\begingroup$ In any case, writing down unbounded operators is easy. Consider any operator $T$ defined on polynomials by sending $x^n$ to $r_n x^n$. This operator is bounded iff the sequence $r_n$ is bounded (exercise). $\endgroup$ – Qiaochu Yuan Dec 28 '17 at 10:35
3
$\begingroup$

First a word of caution: If people speak about an unbounded operator from $X$ to $Y$, they mostly mean a (unbounded) linear map $T\colon D \rightarrow Y$, where $D$ is linear subspace of $X$ (see here). This allows to deal with differential operators in a context which is more general than that of polynomials. For example take $X=Y=L^2[0,1]$, $D=C^1[0,1]$ and $Tx=x'$ as before.

Of course you can always extend such a map $T\colon D \rightarrow Y$ to a linear map $\tilde T\colon X \rightarrow Y$ by simply setting $\tilde T\vert_{\tilde D} = 0 $ on some complementary subspace $\tilde D$ (i.e. a subspace with $X= D \oplus \tilde D$). This will also be unbounded, but does neither contain any new information nor respect the norm of $X$.

Using this notion of unbounded operators, at least on function spaces there is a huge class of examples, which generalizes differential operators. For the sake of simplicity let us consider $X=Y=L^2\mathbb{R}$.

1) Multiplication operators: Given $m\in L^1_{\text{loc}}\mathbb{R}$, put $D=\{x\in L^2\mathbb{R}\vert ~mx\in L^2\mathbb{R}\}$ and $Tx= mx$ for $x\in D$. This is a bounded operator iff $m$ is bounded.

2) Fourier Multiplication operators: The Fourier transform is a bounded linear map $F\colon L^2\mathbb{R} \rightarrow L^2\mathbb{R}$ (even an isometric isomorphism if you use the right normalization) and can be used to generate operators of the form $$ Tx=F^{-1}\left(m\cdot(Fx)\right), \quad D=\{x\in L^2\mathbb{R}\vert~F^{-1}\left(m\cdot(Fx)\right) \in L^2\mathbb{R}\} $$ Since $\Vert Tx \Vert = \Vert m \cdot (Fx)\Vert $ and $\Vert Fx\Vert = \Vert x \Vert$, this will again be bounded iff $m$ is a bounded function. Taking $m(\xi)=i\xi$ gives as assoicated operator exactly the differentiation operator $Tx=x'$.

3) Pseudodifferential operators: Take the same definition of $T$ as before, but allow $m$ to both depend on the spatial variable $t$ and the Fourier variable $\xi$. This generalizes the concepts above, $m(t,\xi)=m(t)$ gives a multiplication operator and $m(t,\xi)=m(\xi)$ gives a Fourier multiplication operator (see here for details).

$\endgroup$
1
$\begingroup$

If $X$ is an infinite-dimensional normed space, then you can choose a Hamel basis $\{ x_{\alpha} \}$ and define a linear functional $F$ by setting $F(x_{\alpha})=s_{\alpha}$, where $\{ s_{\alpha} \}$ is a set of scalars. $F$ will be unbounded if $\sup_{\alpha} |s_{\alpha}|/\|x_{\alpha}\|=\infty$. Then you can extend the argument to obtain $T: X \rightarrow Y$ by choosing a finite set of discontinuous functionals $F_n$ on $X$, choosing an independent set of vectors $\{ y_n \} \subset Y$, and then setting $F(x)=\sum_n F_n(x)y_n$.

$\endgroup$
  • $\begingroup$ thank you for your answer, but I would be even more grateful if you could please expand your post by elaborating upon the claims made therein. For example, what a Hamel basis is, how to show that our $F$ is indeed unbounded? $\endgroup$ – Saaqib Mahmood Jan 1 '18 at 10:59
  • $\begingroup$ A Hamel basis is a maximal linearly independent set. Every $x\in X$ can be written as a finite linear combination of elements in the Hamel basis. To show that $F$ is unbounded, use Hahn-Banach to find a continuous linear functional $\phi_n$ such that $\phi_n(y_k)=\delta_{n,k}$ and conclude that $\phi\circ F = F_n$ is unbounded, which would force $F$ to be unbounded. $\endgroup$ – DisintegratingByParts Jan 1 '18 at 15:22

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.