5
$\begingroup$

I'm having a really hard time with nearly every part of this proof, any help would be appreciated.

Schur's Theorem: In $\ell^1$ weak convergence of $x_n$ is the same as convergence in the norm.

Definition: For $x_n \in \ell^1$ convergence in the norm: $$ x_n \to x \iff \|\ x_n - x \|_{\ell^1} \to 0.$$

Definition: For $x_n \in \ell^1$, weak convergence: $$x_n \rightharpoonup x \iff \phi x_n \to \phi x \hspace{1cm} \forall x_n \in \ell^1 , \space\ \space\ \forall \phi \in \ell^{1*}.$$

This problem is from Muscat's "Functional Analysis" text. It breaks Schur's Theorem up into the following parts:

1) If the statement were false there would be unit $x_n = (a_{ni}) \in \ell^1$ such that $x_n \rightharpoonup 0$

Proof: Taking a unit $(a_{ni}) \in \ell^1$ we have that $\sum_{i=1}^{\infty} |a_{ni}| = 1$. Then, for some $\phi \in \ell^{1*}$ we (somehow?) obtain that $\phi x_n \to \phi x = 0$.

2) For each $n$ there is an $N_n$, such that $\sum_{i=1}^{N_n} |a_{ni}| > \frac{4}{5}$.

Proof: Since $(a_{ni})$ is convergent in $\ell^1$ as a unit, we have that $$\sum_{i=1}^{\infty} |a_{ni}| = 1 \hspace{1cm} \forall n$$ as a requirement for convergent series is that their tail sequence goes to $0$, so that for $\epsilon > 0$ there exists some $N \in \mathbb{N}$ where $$\sum_{i=N}^{\infty} |a_{ni}| < \epsilon.$$ Hence, it follows that we may find an $N_n$ where $$\sum_{i=1}^{N_n} |a_{ni}| > \frac{4}{5} \hspace{1cm} \forall n$$

3) Each coefficient converges to $0$ as $n \to \infty$, so $\forall k , \exists M, n \geq M \Rightarrow \sum_{i < k} |a_{ni} | < \frac{1}{5}$.

Proof: Mirroring what we did above in part (2) and choosing $\epsilon = \frac{1}{5}$.

4) A subsequence of $\left\{ x_n \right\}$ exists with $$\sum_{i < N_n - 1} |a_{ni}| < \dfrac{1}{5} , \sum_{i=N_{n-1}}^{N_n} |a_{ni} | > \dfrac{3}{5}, \sum_{i > N_n} |a_{ni} | < \dfrac{1}{5} .$$

Proof:

5) Let $y := (|a_{ni}| / a_{ni}) \in \ell^{\infty}$ where for each $i$, $n$ is such that $N_{n-1} \leq i < N_n$. Show $|y \cdot x_n | \geq \frac{1}{5}$ to obtain a contradiction.

Proof:

$\endgroup$
10
$\begingroup$

The crux of the matter is the idea of the "gliding hump". Intuitively, if our sequence of $\ell^1$ sequences doesn't converge in norm to $0$, we can construct a functional on $\ell^1$ such that the $\ell^1$ mass of our sequence is always carried down the line, like a hump gliding to infinity. Put another way, if we have $x_n \rightharpoonup 0$, but $\|x_n\| \not\to 0$, there will be a subsequence with $\|x_m\|> \epsilon$ for all $m$, and some $\epsilon$. Then we will find a functional which preserves most of this "mass" (like, say, 98% of the $\epsilon$), and then we will contradict weak convergence. So here we go.

Suppose we have a sequence $\{x^n\}_{n \in \mathbb{N}}$ such that $x^n \rightharpoonup 0$ but $\|x^n\| \not\to 0$. Then there exists an $\epsilon >0$ such that $\|x^m\| > \epsilon$ for infinitely many $m$. Pass then to the subsequence $\{x^m\}_{m \in \mathbb{N}}$. Of course it suffices to consider $x^n \rightharpoonup 0$ otherwise if the weak limit is $x$ we pass to $x^n - x$. Now, choose $N_1, M_1, N_2$ and $M_2$ in the following way:

\begin{align} &N_1 \Rightarrow \sum_{k=N_1}^\infty |x_k^1| < \frac{\epsilon}{100}&\\ &M_1 \Rightarrow \sum_{k=1}^{N_1} |x_k^{M_1}| < \frac{\epsilon}{100}&\\ &N_2 \Rightarrow \sum_{k=N_2}^\infty |x_k^{M_1}| < \frac{\epsilon}{100}&\\ &M_2 \Rightarrow \sum_{k=1}^{N_2}|x_k^{M_2}|< \frac{\epsilon}{100} \end{align} where choices of $N_i$ follow from the fact that the tail ends of $x^{M_{i-1}}$ can be made small, and choices of $M_i$ can be made by the termwise convergence (note these are finite sums) given by weak convergence. In general, take \begin{align} &N_i \Rightarrow \sum_{k=N_i}^\infty |x_k^{M_{i-1}}| < \frac{\epsilon}{100}&\\ &M_i \Rightarrow \sum_{k=1}^{N_i} |x_k^{M_i}| < \frac{\epsilon}{100}& \end{align}

where it goes without saying (but we say it anyway) that $\Rightarrow$ fulfills the role of "such that". Now, we see that in this construction the bulk of the $\ell^1$ mass of $x^{M_j}$ lies in the $x_k^{M_j}$ with $N_j + 1 \le k \le N_{j+1}$. More precisely, 98 percent of the mass lies in the $x_k^{M_j}$ in that interval.

Now define the function $f \in (\ell^1)^*$ corresponding to the bounded sequence $y \in \ell^\infty$ given by $y_k =$ sgn$(x_k^{M_j})$ for $N_{j-1} \le k < N_j$. Then $\displaystyle f(x^{M_J}) = \sum_{k=1}^\infty y_kx^{M_j}_k = \sum_{k=0}^{N_{j-1}}y_kx_k^{M_j} + \sum_{k=N_{j-1}+1}^{N_j}y_kx_k^{M_j} + \sum_{k=N_j +1}^{\infty}y_kx_k^{M_j}$
so then $\displaystyle \left|\sum_{k=1}^\infty y_kx^{M_j}_k \right| \ge \left| \sum_{k=N_{j-1}+1}^{N_j}y_kx_k^{M_j} - \left|\left|\sum_{k=0}^{N_{j-1}}y_kx_k^{M_j}\right| - \left|\sum_{k=N_j +1}^{\infty}y_kx_k^{M_j}\right|\right|\right| \ge | \frac{98}{100}\epsilon - \frac{1}{100}\epsilon - \frac{1}{100}\epsilon |= \frac{24}{25}\epsilon > 0$

for all $j$. But then this contradicts weak convergence to 0 along the subsequence $\{x^{M_j}\}_{j \in \mathbb{}}$. So therefore weak convergence implies strong convergence.

I'll leave proving strong convergence implies weak convergence to you.

$\endgroup$
  • $\begingroup$ The last part of this proof is not clear at all. $\endgroup$ – Diego Fonseca Nov 18 '17 at 3:12
  • $\begingroup$ It was probably worded poorly. We know that $\ell^\infty = (\ell^1)^*$ so we construct a functional $f$ on $\ell^1$ such that $f(e_i) = y_i$, where $y$ is the sequence in $\ell^\infty$ detailed above. Then we find a subsequence $\{x^{M_j}\}_{j \in \mathbb{N}}$ such that $f(x^{M_j})> \displaystyle \frac{24\varepsilon}{25}$ for all $j$, which contradicts weak convergence to $0$ (since $f$ is a bounded linear functional. $\endgroup$ – David Bowman Nov 18 '17 at 16:10
  • $\begingroup$ The indexing at the end of the proof seems mistyped, when you define y_k shouldnt it be assigned in the region N_j + 1 to N_{j+1}? $\endgroup$ – gen May 13 '18 at 8:03

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.