orthonormal sequence in $L^2[0,1]$ - how to prove these following equivalent terms?

1.4k Views Asked by At

I've been asked this following very interesting question and would like some help figuring out why it is true :)

Let $u_n$ be an orthonormal sequence in $L^2[0,1]$

Prove that the following are equivalent:

  1. $u_n$ is complete (orthonormal basis)

  2. for every $x\in [0,1]$ : $x=\sum_{n=1}^{\infty} |\int_{0}^{x} u_n(t)dt|^2 $

  3. $0.5=\sum_{n=1}^{\infty} \int_{0}^{1} |\int_{0}^{x} u_n(t)dt|^2 dx $

My thoughts so far:

a. if $u_n$ is complete, I could use Parseval's identity with functions $f=1_{[0,x]}$ and get condition 2, and integrate it to get condition 3 (p.s: why could I change the order of the inifite sum with that $\int_{0}^{1}$?)

b. The other way aronud is harder. I figured out that using the definition of orthonormal basis would be a little difficult, So probably I'm supposed to show the Parsevals' identity, or something like that. no luck there.

I would like some guide if you could.

Thanks!

Edit: I tried this question for a couple more hours and couldn't solve. A more direct hint would be appreciated.

1

There are 1 best solutions below

9
On BEST ANSWER

The span $S$ of the functions $\{ \chi_{[0,x]} : 0 \le x \le 1\}$ is a dense linear subspace of the Hilbert space $L^{2}[0,1]$. To see that the span is dense, it is enough to show that $(f,\chi_{[0,x]})=0$ for all $x$ and some $f \in L^{2}[0,1]$ implies $f=0$. By the Lebesgue differentiation theorem, the following holds a.e.: $$ \frac{d}{dx}(f,\chi_{[0,x]})=\frac{d}{dx}\int_{0}^{x}f(t)\,dt = f(x). $$ So it follows that $S$ is dense in $L^{2}[0,1]$.

Let $\{ u_{n} \}$ be an orthonormal subset of $L^{2}[0,1]$. Parseval's equality holds for a given $f$ with respect to this set iff $f$ is in the closed linear subspace $\mathcal{U}$ generated by this orthonormal subset, which means that $\{ u_{n} \}$ is a complete orthonormal subset iff $\mathcal{U}$ includes $S$. That is, $\{ u_{n} \}$ is complete iff $$ \begin{align} x & = \int_{0}^{x}|\chi_{[0,x]}(t)|^{2}\,dt \\ & = \sum_{n}^{\infty}|(\chi_{[0,x]},u_{n})|^{2} \\ & =\sum_{n} \left|\int_{0}^{x}u_{n}(t)\,dt\right|^{2},\;\;\; 0 \le x \le 1. \end{align} $$ So that takes care of (1) $\iff$ (2).

For (2) $\iff$ (3), note that Bessel's inequality always holds: $$ \sum_{n}\left|\int_{0}^{x}u_{n}(t)\,dt\right|^{2} \le \int_{0}^{1}|\chi_{[0,x]}(t)|^{2}\,dx = x. $$ Let $P$ be the orthogonal projection of $L^{2}[0,1]$ onto the closed linear span $\mathcal{U}$ of the $\{ u_{n} \}$. Then $x\mapsto P\chi_{[0,x]}$ is continuous because $x\mapsto \chi_{[0,x]}$ is continuous from $[0,1]$ into $L^{2}[0,1]$ (dominated convergence,) and because $P$ is continuous on $L^{2}[0,1]$. The left side above is $\|P\chi_{[0,x]}\|^{2}$, which must be continuous in $x$. The left side must equal $x$ for all $x$ in order for $\mathcal{U}$ to be complete. It follows that equality holds for all $x$ iff $$ \int_{0}^{1} \sum_{n}\left|\int_{0}^{x}u_{n}(t)\,dt\right|^{2}\,dx = \frac{1}{2}. $$ By monotone convergence, you are allowed to interchange summation and integration in order to conclude that the above holds iff $$ \sum_{n}\int_{0}^{1}\left|\int_{0}^{x}u_{n}(t)\,dt\right|^{2}\,dx = \frac{1}{2}. $$