7

I want to solve the following problem:

Suppose $H$ is a Hilbert space, if $u_{n}\rightharpoonup u$ in $L^{2}(0,T;H)$ and $u'_{n}\rightharpoonup v$ in $L^{2}(0,T;H^{'})$.

I want to prove that $v=u'$.

Thanks for your help.

JKnecht
  • 6,637
Miguel
  • 71

2 Answers2

5

Let's write down what we know: \begin{align} \int_0^T\langle u_n(t),h(t) \rangle\,dt \to &\ \int_0^T\langle u(t),h(t) \rangle\,dt \quad \forall h \in L^2(0,T;H') \tag 1\\ \int_0^T\langle u_n'(t),h(t) \rangle\,dt \to &\ \int_0^T\langle v(t),h(t) \rangle\,dt \quad \forall h \in L^2(0,T;H)\tag 2\\ \int_0^T\varphi'(t)u_n(t)\,dt = &\ -\int_0^T\varphi(t)u_n'(t)\,dt \quad \forall \varphi \in C^{\infty}_c((0,T)). \tag 3 \end{align}

Notice that for $\varphi \in C^{\infty}_c((0,T))$ and $w \in H$ we have that $t \mapsto \varphi(t)w \in L^2(0,T;H)$. In the following I am going to identify $H$ with its dual $H'$. If you prefer to treat them as different spaces you can modify the following argument introducing Riesz's linear isometry $J \colon H \to H'$. For $\varphi$ and $w$ be as above, we have:

\begin{align} \Big\langle\int_0^T\varphi'(t)u(t)\,dt,w\Big\rangle = &\ \int_0^T\langle\varphi'(t)u(t),w\rangle\,dt\\ = &\ \int_0^T\langle u(t),\varphi'(t) w\rangle\,dt\\ \overset{(1)}{=} &\ \lim_{n \to \infty} \int_0^T\langle u_n(t), \varphi'(t)w\rangle\,dt \\ = &\ \lim_{n \to \infty}\Big\langle\int_0^T\varphi'(t)u_n(t)\,dt,w\Big\rangle \\ \overset{(3)}{=} &\ \lim_{n \to \infty}\Big\langle-\int_0^T\varphi(t)u_n'(t)\,dt,w\Big\rangle \\ = &\ \lim_{n \to \infty} \int_0^T-\langle u_n'(t),\varphi(t)w\rangle\,dt \\ \overset{(2)}{=} &\ \int_0^T-\langle v(t),\varphi(t)w\rangle\,dt \\ = &\ \Big\langle-\int_0^T\varphi(t)v(t)\,dt,w\Big\rangle. \end{align}

Comparing the first and last step in the previous chain of equalities we get

$$\Big\langle\int_0^T\varphi'(t)u(t)\,dt + \int_0^T\varphi(t)v(t)\,dt ,w\Big\rangle = 0.$$ Since this holds for every $w \in H$ we can conclude that $$\int_0^T\varphi'(t)u(t)\,dt = -\int_0^T\varphi(t)v(t)\,dt.$$ Since this holds for every test function $\varphi$, we get the desired result.

Giovanni
  • 6,569
  • Many thanks for your answer. I am trying to do the proof with H and H' by separate, but I have a problem since I obtain for one side the duality product in H'-H and by the other one I obtain the product in H-H'. Also in (2) is $L^{2}(0,T;H)$ – Miguel Oct 21 '15 at 14:11
  • Thanks for catching the typo. I'll edit later to clarify that step :) – Giovanni Oct 21 '15 at 15:23
  • Great proof! Do you happen to have a reference for the theory of Bochner spaces? – user23793 Apr 17 '16 at 19:47
1

Your problem is a corollary of the following

Theorem: Let $Y$ be a Banach space such that $Y'$ has the Radon Nikodym property, $X$ a Banach space continuously embeded in $Y$ and $1\leq p,q\leq\infty$. If $$\left\{\begin{align}u_k\rightharpoonup u\quad&\mbox{in}\quad L^p(0,T;X),\\ u_k'\rightharpoonup v\quad&\mbox{in}\quad L^q(0,T;Y),\end{align}\right.$$ then $u'=v$.

Sketch of the proof:

Since $X\hookrightarrow Y$, we get $L^p(0,T;X)\hookrightarrow L^1(0,T;Y)$ and $L^q(0,T;Y)\hookrightarrow L^1(0,T;Y)$. Thus,

$$\left\{\begin{align}u_k\rightharpoonup u\quad&\mbox{in}\quad L^1(0,T;Y),\\ u_k'\rightharpoonup v\quad&\mbox{in}\quad L^1(0,T;Y).\end{align}\right.$$

As $Y'$ has the Radon Nikodym property, it follows that

$$\left\{\begin{align}\int_0^T u_k\varphi'\,dt&\rightharpoonup \int_0^T u\varphi'\,dt &\mbox{in}\quad Y,\\ \int_0^T u_k'\varphi\,dt&\rightharpoonup \int_0^Tv\varphi\,dt &\mbox{in}\quad Y,\end{align}\right.$$ for all $\varphi\in C_c^\infty(0,T)$. But, by definition of weak derivative, we have $$\int_0^T u_k\varphi'\,dt=-\int_0^T u_k'\varphi\,dt,\quad \forall\ \varphi\in C_c^\infty(0,T)$$ so that (by the uniqueness of the weak limit) $$\int_0^T u\varphi'\,dt=-\int_0^T v\varphi\,dt,\quad \forall\ \varphi\in C_c^\infty(0,T).$$

Hence, $u'=v$.


Addendum. To prove that the weak convergence in $L^1(0,T;Y)$ implies that weak convergence of integrals in $Y$, we need the surjectivity of a certain mapping from $L^\infty(0,T;Y')$ to $L^1(0,T;Y)'$ (as sketched here). This mapping is surjective if $Y'$ has the Radon Nikodym property (what happens, for example, if $Y$ is separable reflexive or if $Y$ is Hilbert).

Pedro
  • 19,965
  • 9
  • 70
  • 138
  • Please, notice that the Hilbert space in the question in not necessarily separable. – Giovanni Oct 19 '15 at 23:49
  • @Giovanni You are right. But we can replace "separable reflexive" by Hilbert (or something more general) in the theorem. See my edit and addendum. Is it ok? – Pedro Oct 20 '15 at 00:52
  • Looks good, +1. Also, thank you for editing my post :) – Giovanni Oct 20 '15 at 01:37
  • Many thanks for your help. Please, can you give me the reference of the book in which I can find the cited Theorem.? – Miguel Oct 21 '15 at 13:46
  • @Pedro Many thanks for your answer. Please can you give me the reference in which I can find the cited theorem? – Miguel Oct 21 '15 at 14:00
  • @Miguel A certain version of this theorem can be found in the Zeidler book's II/A, p. 419. In this book the theorem is stated for $X$ and $Y$ being only Banach spaces. However, in the proof (it seems to me that) the item (e) of Proposition 23.9 is used. So, we also need assume that $Y$ is separable and reflexive. Since your Hilbert space is not necessarily separable, your problem is not a consequence of this version of the theorem. However, you can mimic the proof. – Pedro Oct 21 '15 at 23:50
  • @Miguel To clarify: with respect to the space $L^2(0,T;H)$, the convergence $(1)$ in the Giovanni's answer is not automatic (at least according to my knowledge). To have them, we need some assumption on $H$. The most common (that I have seem) is "reflexivity and separability" because we usually have $L^p(0,T; W^{m,p})$ or $L^p(0,T;L^p)$ for suitable choices of $m$ and $p$. – Pedro Oct 22 '15 at 00:22
  • @Pedro Many thanks for your help. – Miguel Oct 22 '15 at 13:58