[Math] $\operatorname{span}\{x_n:n\in \Bbb N\}$is dense if $\sum_{n=1}^\infty \|x_n-e_n\|^2<1$

functional-analysishilbert-spaces

Let $H$ be a Hilbert space with orthonormal basis $\{e_1,e_2,\cdots\}$. Suppose $(x_n)$ is a sequence in $H$ with $\sum_{n=1}^\infty \|x_n-e_n\|^2<1$.

Claim: The span of the $x_n$ is dense in $H$.

I don't have a very good attempt, as I don't know how to use the sum condition. We can express any $x\in H$ as $x=\sum_{k=1}^\infty \langle x,e_k\rangle e_k$. Now we need to find an approximating sequence of $x$ in $\operatorname{span}\{x_1,x_2,\cdots\}$. In a first attempt, I tried using $y_n=\sum_{k=1}^n \langle x,e_k\rangle x_k\in \operatorname{span}\{x_1,x_2,\cdots\}$ as the approximating sequence but it didn't pan out.
How could we proceed?

Best Answer

Define the operator $S : H\to H$ by $Sx = \sum_k\langle x,e_k\rangle x_k$. From \begin{align*} \left\|\sum\langle x,e_k\rangle x_k\right\| &\le\left\|\sum\langle x,e_k\rangle (x_k-e_k)\right\| + \left\|\sum\langle x,e_k\rangle e_k\right\|\\ &\le\sum |\langle x,e_k\rangle| \|x_k-e_k\| + \left\|\sum\langle x,e_k\rangle e_k\right\|\\ &\le \left(\sum |\langle x,e_k\rangle|^2\right)^{1/2}\left(\sum \|x_k-e_k\|^2\right)^{1/2} + \left(\sum |\langle x,e_k\rangle|^2\right)^{1/2}\\ &\le 2\left(\sum |\langle x,e_k\rangle|^2\right)^{1/2}, \end{align*} we see that $\sum_{k=1}^n\langle x,e_k\rangle x_k$ is a Cauchy sequence. So, $S$ is well defined. Moreover, if we set $\delta := \left(\sum_k\|x_k-e_k\|^2\right)^{1/2} < 1$, then $$ \|(S-I)x\| = \left\|\sum_k\langle x,e_k\rangle (x_k-e_k)\right\|\le\delta\|x\|. $$ So, $\|S-I\| < 1$ which implies that $S$ is invertible. Now it should be easy for you to show that $(x_k)_k$ is dense in $H$. For this, note that $Se_k = x_k$.