3
$\begingroup$

In $\ell^2$, a sequence converges weakly if it is bounded and each component converges. My question is: How can I think about weak convergence in $L^2[0,1]$?

$\endgroup$
4
+50
$\begingroup$

As $\ell_2(\mathbb N)$ is isometrically isomorphic to any separable infinite-dimensional Hilbert space $\mathcal H$ (such as $L^2[0,1]$), the $\ell_2$-condition for weak convergence of a bounded sequence transfers to every such $\mathcal H$ where

$$ \text{"each component"} $$

gets replaced by

$$ \text{"each }(\langle e_k,x_n\rangle)_{n\in\mathbb N}\text{ where }(e_k)_{k\in\mathbb N}\text{ is any orthonormal basis of }\mathcal H", $$

e.g. see here (Lemma 8). Then, you can take any orthonormal basis of $L^2[0,1]$ (e.g. $\exp(2\pi inx)$, or the Legendre polynomials on $[0,1]$, or $\ldots$) you like to characterize weak convergence in it.

$\endgroup$
  • $\begingroup$ Thanks for your answer. If I work in the real version of $L^2[0,1]$, what should I do if I have a concrete sequence of functions? Are there any illuminating examples? The isometric identification won't help me much I fear. Thanks - how would I deal with concrete examples, say $\sin(\alpha_nt)$, where $(\alpha_n)$ is a sequence? $\endgroup$ – max_zorn Mar 13 '18 at 3:35
3
$\begingroup$

A sequence $(u_n)$ in $L^2[0,1]$ converges weakly iff it is bounded and for all intervals $(a,b)\subset [0,1]$ the integrals $\int_a^b u_n$ are converging. Instead of 'each component' converges, we have now each integral on an arbitrary interval converges.

$\endgroup$
  • $\begingroup$ Cool, can you add a reference please? $\endgroup$ – max_zorn Mar 6 '18 at 8:42
  • $\begingroup$ Thanks - how would I deal with concrete examples, say $\sin(\alpha_nt)$, where $(\alpha_n)$ is a sequence? $\endgroup$ – max_zorn Mar 13 '18 at 3:35
1
$\begingroup$

The space $L^2([0,1])$ equipped with the inner product $$\langle f,g\rangle:=\int^1_0f(x)\overline{g(x)}\,dx$$ is a Hilbert space. By definition of weak convergence we say a sequence $(f_n)\subset L^2([0,1])$ converges weakly to $f\in L^2([0,1])$ whenever $$\lim_n\langle f_n,g\rangle=\langle f,g\rangle\hspace{0.2cm}\text{for all}\hspace{0.2cm}g\in L^2([0,1])$$ On the other hand the spaces $l^2(\mathbb{N})$ and $L^2([0,1])$ are isometric isomorphic to each other. To see this note that if $(e_n)\subset L^2([0,1])$ is a complete orthonormal basis for $L^2([0,1])$ then every $f\in L^2([0,1])$ admits a representation of the form $$f=\sum_{n=1}^{\infty}\langle f,e_n\rangle e_n\Rightarrow ||f||^2_{L^2}=\int_0^1f(x)\overline{f(x)}\,dx=\langle f,f\rangle=\sum^{\infty}_{n=1}|\langle f,e_n\rangle|^2:=||f_n||^2_{l^2}$$ implies $(f_n)\in l^2(\mathbb{N})$ where $f_n:=\langle f,e_n\rangle$. Clearly the operator $A:L^2([0,1])\mapsto l^2(\mathbb{N})$ defined by $Af:=(f_n)$ is a linear bounded operator which preserves the norms (as we can see for the above relation). Now to come back to the initial statement that if $(f^k_n)$ is a bounded sequence for each $k$ and for each $n$ we have $|f^k_n-f_n|\to 0$ as $k\to\infty$ then $(f^k_n)$ converges weakly to $(f_n)\in l^2(\mathbb{N})$. Again by definition $(f^k_n)$ converges weakly to $(f_n)$ whenever $$\lim_k\langle f^k,g\rangle=\langle f,g\rangle$$ where the inner product for $l^2(\mathbb{N})$ is defined by $$\langle f,g\rangle:=\sum_nf_n\overline{g_n}$$ Boundedness of each $(f^k_n)$ implies $(f^k_n)\in l^2(\mathbb{N})$ for each $k$ so their inner product with any other element $g$ of $l^2(\mathbb{N})$ is finite and well defined. On the other hand component wise convergence, $|f^k_n-f_n|\to 0$ as $k\to\infty$ for each $n$, implies $$|\langle f^k,g\rangle-\langle f,g\rangle|=|\langle f^k-f,g\rangle=\Big|\sum_n(f^k_n-f_n)\overline{g_n}\Big|\leqslant \sum_n|f^k_n-f_n||\overline{g_n}|\to 0$$ as $k\to\infty$. Therefore the same definition for both spaces is valid.

$\endgroup$
  • $\begingroup$ Thanks - how would I deal with concrete examples, say $\sin(\alpha_nt)$, where $(\alpha_n)$ is a sequence? $\endgroup$ – max_zorn Mar 13 '18 at 3:35
  • $\begingroup$ depends also on your sequence $\alpha_n$. if $\alpha_n=\theta$ for all $n$ then $\sin(\alpha_nt)$ converges trivially to $\sin(\theta t)$. $\endgroup$ – Arian Mar 15 '18 at 11:40

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.