I'm learning Lax - Milgram theorem (from L.C. Lawrance book) and I have a problem with understanding the proof.

Assume $B\colon H\times X \rightarrow \mathbb{R}$ ($H$ - Hilbert space) is bilinear and

$|B(u,v)|\leq \alpha\|u\| \|v\|$ for $u,v \in H$,

$\beta\|u\|^{2}\leq B(u,u)$ for $u\in H$.

Let $f\colon H\rightarrow \mathbb{R}$ be bounded linear functional on $H$. Then there exists unique element $u\in H$ such that $B(u,v)=(f,v)$ for all $v\in H$.

So I follow the standard pattern and introduce an operator $A:H\rightarrow H$ as in Evans book. I easily obtain that $\beta \|u\|\leq\|Au\|$. I make a use of the latter inequality to prove that $A$ is injective. In the book however there is a statement that the inequality implies that Range(A) is closed. That is the part I don't get.


I assume that $A$ is linear, if we suppose that $A$ is also continuous (or bounded, equivalently) we can show that $\mathcal{R}(A) $ is closed.

Indeed, let $(v_n)_n \in \mathcal{R}(A) $ be a sequence such that $\lim_{n} v_n = v \in H $, we want to show that $v \in \mathcal{R}(A)$, i.e $A \overline{v} = v$ for some $\overline{v} \in H$. Since $(v_n)_n \in \mathcal{R}(A) $, for every $n$ there exists $u_n \in H$ such that $Au_n = v_n $; since $v_n$ is convergent, $(Au_n)_n$ is a Cauchy sequence in $H$ and the following inequality holds for every $m,n \in \mathbb{N}$: $$ \beta\| u_m - u_n \| \le \| A(u_m-u_n) \|= \| Au_m - Au_n \| \xrightarrow[m,n ]{} 0 $$

whence also $(u_n)_n $ is a Cauchy sequence. By the completeness of $H$, $(u_n)_n$ converges, say to $u \in H$. Finally,

$$ v=\lim_{n} v_n = \lim_{n} Au_n \underbrace{=}_{\text{continuity}} A(\lim_{n} u_n )= Au. $$

Hence $v \in \mathcal{R}(A) $ and $\mathcal{R}(A)$ is closed (note that in an Hilbert space topological closedness is equivalent to sequential closedness).

  • $\begingroup$ Why do you claim that $(Au_{n})$ is Cauchy? $\endgroup$ – zorro47 Mar 29 '17 at 14:39
  • $\begingroup$ Because $Au_n = v_n $ (remeber $v_n$ is in the range of $A$) for every $n$ and $v_n$ is convergent, hence Cauchy. $\endgroup$ – GaC Mar 29 '17 at 14:41
  • $\begingroup$ I was thinking of proving that $(u_{n})$ is convergent straight from the fact that $v_{n}=Au_{n}$ is convergent but it can not be done. Now, I understand the aim of using Cauchy condition and the inequality I mentioned before. $\endgroup$ – zorro47 Mar 29 '17 at 14:50
  • $\begingroup$ There is one more doubt. We have that $\mathcal{R}(A)$ is closed, hence $A$ is surjective. Now, from the Riesz representation theorem there exists unique $w$ such that $f(v)=(w,v)$ for every $v$. Since $A^{-1}$ exists, so there is $u_{0}$ such that $Au_{0}=w$. $u_{0}$ is unique. On the oher hand, $B(u_{0},v)=(Au_{0},v)=(w,v)=f(v)$, so there is no need to show uniqueness, still it is a part of Evans proof. $\endgroup$ – zorro47 Mar 29 '17 at 16:54
  • $\begingroup$ Uh, maybe I should look at the complete proof to answer your last question. I'm going to the library next week and take a look at Evans' proof. $\endgroup$ – GaC Mar 30 '17 at 13:07

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.