2

Given a bounded linear operator: $$ T:PC[0,1]\to PC[0,1]$$ and its completion $$ \hat{T}:L^2[0,1] \to L^2[0,1]$$ (defined by $\hat{T}x:=\lim Tx_n$ where $x_n \to x$).

If $T$'s image is $C[0,1]$ (all continuous functions), how can I show that the image of $\hat{T}$ is all of $L^2[0,1]$? (note, I may be wrong about this. Also - if it helps, it's a statement based on this question).

(note, $PC$ stands for piecewise-continuous functions, which are dense in $L^2[0,1]$. Also $C[0,1]$ is dense in $L^2[0,1]$).

I have tried to claim the following:

Let $f\in L^2[0,1]$. We will show that there exists a $g\in L^2[0,1]$ such that $\hat{T}g=f$. Since $C$ is dense in $L^2$, there is a sequence $f_n \in C[0,1]$ such that $f_n \to f$. Since $f_n \in Img(T)$, there exist $g_n \in PC[0,1]$ such that $Tg_n = f_n$ for all $n$. Taking the limit we obtain that $\lim Tg_n = \lim f_n$. Since the RHS is $f$, if I could only claim that $g_n$ must a converging sequence, converging to some $g \in L^2$ then I would be able to say, by the definition of the completion $\hat{T}$, that: $\hat{T}g=\lim Tg_n=\lim f_n=f$ so that we have found our $g$.    

I am unable to prove this last claim, i.e. that $g_n$ converges to $g$. So - either this is the wrong way to go about this, or I'm missing something.

EDIT: I believe the claim holds if we replace $C[0,1]$ with $PC[0,1]$, since then we have that $T$ is a bijection and then it is possible to claim that each $f \in L^2$ has a distinct image in $L^2$ as such: $f,h \in L^2[0,1]$, with sequences $f_n, h_n \in PC[0,1]$ such that $f_n \to f$ and $h_n \to h$ then $\hat{T}f = \lim Tf_n \neq \lim Th_n = \hat{T}h$

EDIT 2: This last comment on $PC$ is actually also incorrect since I'm basically stating that injective means surjective here, and that's not necessarily true.

Any ideas?

Thanks.

Anon
  • 1,757
  • Consider the case of $Tf=\int_0^xf(t)dt$. – Disintegrating By Parts May 13 '22 at 20:58
  • @DisintegratingByParts Do you mind elaborating how this is a counter example? Specifically, why is the image of $T$ precisely $C$ (I realize it's in $C$, but why all of $C$?). Also - is the image of its completion not all of $L^2$ because $F(x)$ is continuous by the fundamental theorem? – Anon May 13 '22 at 21:43
  • The operator provided by @DisintegratingByParts maps $C$ into the functions of class $\mathscr C^1$, hence a proper subset of $C$. – Ruy May 13 '22 at 22:57

1 Answers1

1

Here is an example of a bounded operator $T$ on $L^2[0,1]$ which

  1. maps $C[0,1]$ onto itself,

  2. maps $PC[0,1]$ onto itself,

and yet it does not map $L^2[0,1]$ onto itself.

For $f$ in $L^2[0,1]$, set $$ T(f)(x) = f(\sqrt x), \quad \forall x\in [0,1]. $$ Then $T$ is bounded because $$ \|T(f)\|^2 = \int_0^1|f(\sqrt x)|^2\, dx = \int_0^12y|f(y)|^2\, dy \leq 2\int_0^1|f(y)|^2\, dy = 2 \|f\|^2. $$

Points (1) and (2) above are easy to verify since the correspondence $x\mapsto \sqrt x$ is an increasing homeomorphism on $[0,1]$.

To prove that $T$ is not onto $L^2[0,1]$, pick a real number $\beta $ in the interval $(-\frac 1 2, -\frac 14]$, and let $$ g(x) = x^\beta , \quad \forall x\in [0,1]. $$ The clearly $g$ lies in $L^2[0,1]$, because $\beta >-\frac12$, but I claim that $g$ is not in the range of $T$. Indeed, if $g=T(f)$, for some $f$, then $$ x^\beta = g(x) = T(f)(x) = f(\sqrt x), $$ whence $$ f(x) = f(\sqrt {x^2}) = x^{2\beta }, $$ which is not in $L^2[0,1]$, because $2\beta \leq -\frac12$.

Ruy
  • 19,160
  • 1
  • 15
  • 37