2
$\begingroup$

This exercise is from Rudin's Real and Complex analysis at chapter 15, problem 22.

$A=\{ f_{n} = t^n exp(-t)\}$ where $(0 \leq t \leq \infty, n \in \mathbb{N}).$ Prove that $A$ is dense in $L^{2}(0,\infty)$.

Hint : If $g \in L^{2}(0,\infty)$ is orthogonal to each $f_n$ and if $F(z) = \int_{0}^{\infty}exp(-tz)\overline{g(t)}dt,$ (Re $z>0$), then all derivatives of $F$ are $0$ at $z=1$. Consider $F(1+yi)$.

I understand the hint want me to use Hanh-Banach theorem, but I don't know how the $F(z)$ is related with the theorem. Even if I accept $F'(1)=0$, what can I do that? Could you give me more hints to approach this problem?

Edit: Thank you for all of you. I'll give a try for solving problem.

Let $g \in L^{2}(0,\infty)$ is orthogonal to each $f_n,$ e.g. $\int_{0}^{\infty}t^{n}e^{-t}\overline{g(t)}dt = 0$ for all $n=0,1,2, \cdots$. Define $F(z) = \int_{0}^{\infty}\exp(-tz)\overline{g(t)}dt$ where Re $z >0$.

1) F is well-defined since $(\int_{0}^{\infty}|exp(-tz)|^{2}dt)^{1/2} = \frac{1}{\sqrt{2 Re z}} <\infty$, so $e^{-tz}$, $g(t)$ are in $L^{2}$.

2) $F^{(n)}(w) = \int_{0}^{\infty}(-1)^{n}t^n e^{-tz} \overline{g(t)}dt$

since $\lim_{z \to w}\frac{F(z) - F(w)}{z-w} = \lim_{z \to w} \int_{0}^{\infty}\frac{e^{-tz}-e^{-tw}}{z-w}\overline{g(t)}dt $

and this part is what I do not know how to deal with, but I think I can apply dominated convergence theorem or any other theorem, to derive that

$\lim_{z \to w} \int_{0}^{\infty}\frac{e^{-tz}-e^{-tw}}{z-w}\overline{g(t)}dt = \int_{0}^{\infty}\lim_{z \to w} \frac{e^{-tz}-e^{-tw}}{z-w}\overline{g(t)}dt= \int_{0}^{\infty}-te^{-tz}\overline{g(t)}dt$, and by repeating such differentiating, we get desired result.

3) $F^{(n)}(1) = 0$ for any $n \in \mathbb{N}\cup \{0\}$ since $F^{(n)}(1)$ is just $<f_{n},g>$, which is 0 by assumption.

4) So $F$ is holomorphic on half plane where real part is greater than 0, and $F$'s power series representation at $z=1$ implies $F \cong 0$.

So, if we think $0= F(1+iy)$ for $y\in \mathbb{R}$, $F(i+y) = \int_{0}^{\infty}e^{-t}\overline{g(t)}e^{-iyt}dt$, a Fourier transform of $e^{-t}\overline{g(t)}$, call $\mathbb{F}(e^{-t}\overline{g(t)}$. Hence by fourier inversion formula, $e^{-t}\overline{g(t)} = \int_{0}^{\infty}\mathbb{F}(e^{-t}\overline{g(t)}e^{iyt} = \int 0 dt =0$. So $e^{-t}\overline{g(t)} \cong 0$. But $e^{-t}\neq 0$ for any $t \in (0,\infty)$, $g(t) = 0$ almost everywhere.

5) Conclusion : so any function which is not spanned by closure of span$\{f_n \}$ is zero almost everywhere. Hence the span is dense.

Re-question : You know that my proof was not completed, especially on interchanging limit and integral when I calculate $F^{(n)}$. Could you give some more idea?

$\endgroup$
1
$\begingroup$

Not only the first derivative, but also the evaluation and all higher derivatives are zero: $F(1)=F'(1)=F''(1)=\ldots=0.$ This implies that $F(z)$ is identically zero on its domain.

Now the last part of the hint invites you to write out $F(1+yi)$. For $y$ running through $\mathbb R$ this is actually a Fourier transform and by Fourier inversion $g$ must be zero almost everywhere on the positive reals.

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ What is $g?\,\,$ $\endgroup$ – zhw. Nov 29 '15 at 18:53
  • $\begingroup$ A hypothetical squarely integrable function class that is not in the closed span of the functions $f_n$ $\endgroup$ – Justpassingby Nov 29 '15 at 19:03
  • $\begingroup$ Yes, sorry, it's right there in the hypotheses. $\endgroup$ – zhw. Nov 29 '15 at 19:26
  • $\begingroup$ Oh, I see. Thank you. I will post my own answer this afternoon (GMT+9) $\endgroup$ – user124697 Nov 29 '15 at 23:53
  • $\begingroup$ @Justpassingby I posted my answer by editing this post. It would be helpful for me to get your comment on the answer. Thanks! $\endgroup$ – user124697 Nov 30 '15 at 6:40
1
$\begingroup$

An alternative approach. If we define Laguerre polynomials through the Rodrigues formula: $$ L_n(x) = \frac{e^x}{n!}\frac{d^n}{dx^n}\left(x^n e^{-x}\right) $$ we have: $$ \int_{0}^{+\infty} L_n(x)L_m(x)e^{-x}\,dx = \delta(n,m) $$ by integration by parts, hence Laguerre polynomials give an orthogonal base of $L^2(0,+\infty)$ with the inner product $\langle f,g\rangle = \int_{0}^{+\infty}f(x)g(x)e^{-x}\,dx$. The leading term of $L_n(x)$ is just $\frac{(-1)^n x^n}{n!}$ and the previous base is a complete base. Rudin's claim easily follows.

| cite | improve this answer | |
$\endgroup$
1
$\begingroup$

By convergence dominated theorem, the limit and integral commutes. Hint: you may use the following inequality to dominate the function sequence: \begin{equation*} |\exp(z)-z-1|\leq\frac{|z|^2\exp(|z|)}{2},\quad z\in \mathbb{C}. \end{equation*}

| 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.