3
$\begingroup$

Let $X$ and $Y$ be two separable Hilbert spaces such that $X \subsetneq Y$ densely, continuously and the norms are related through $\|x\|_X^2=|x|_X^2+\|x\|_Y^2$ for all $x \in X$; e.g. take $X=H^1$, $Y=L^2$. Here, $|\cdot|_X$ denotes a semi norm on $X$. Further, let $$a: X \times Y \to \mathbb{R}$$ be a bounded bilinear form, i.e. there is a $c$ such that $|a(x,y)|\leq c \|x\|_X \|y\|_Y$ for all $x \in X$, $y \in Y$. Additionally $a(\cdot,\cdot)$ shall fulfill a inf-sup condition on $X \times Y$, i.e. there is a $\beta>0$ such that

$$ \inf_{x \in X} \sup_{y \in Y} \frac{|a(x,y)|}{\|x\|_X \|y\|_Y}\geq \beta>0. $$

There is a paper, which states in Remark 2.10 that in general one can't follow that an inf-sup condition on $X \times X$ holds i.e. that there is a $\beta_X>0$ such that

$$ \inf_{u \in X} \sup_{v \in X}\frac{|a(u,v)|}{\|u\|_X \|v\|_X}\geq\beta_X>0.$$

Now I am looking for a bounded bilinear form, which has inf-sup condition of $\boldsymbol{X \times Y}$, but not on $\boldsymbol{X \times X}$.


My try:

I tried with $X=H^1_0(0,1), Y=L^2(0,1)$. We have an inf-sup condition on $H_0^1 \times L^2$ with the bilinear form $a(u,v):=\int_0^1 u' \cdot v \, dx=(u',v)_{L^2}$ because

$$\sup_{v \in L^2} \frac{|(u',v)_{L^2}|}{ \|v\|_{L^2}}\geq \frac{|(u',u')_{L^2}|}{ \|u'\|_{L^2}} = \|u'\|_{L^2}=|u|_{H_0^1}\geq C \|u\|_{H^1}$$

for all $u \in H_0^1$ by Poincare's inequality (here $|\cdot|_{H_0^1}$ is even a full norm). Dividing by $\|u\|_{H^1}$ and taking the inf over all $u \in H_0^1$ yields the inf-sup condition on $H_0^1 \times L^2$.

Now it would be great if $a(\cdot,\cdot)$ has no inf-sup condition on $H_0^1 \times H_0^1$. So I have to show that there is no $\beta_X>0$ such that

$$\inf_{u \in H_0^1} \sup_{v \in H_0^1} \frac{|(u',v)_{L^2}|}{\|u\|_{H^1} \|v\|_{H^1}}\geq\beta_X>0.$$

Or equivalently I could show that for all $\beta_X>0$ there is a $u \in H_0^1(0,1)$ such that $$\sup_{v \in H_0^1} \frac{|(u',v)_{L^2}|}{\|v\|_{H^1}}<\beta_X \|u\|_{H^1} $$

Does someone have an idea? Up to now I had no success; somehow I have to choose this $u$ in a smart way - likely depending on $\beta_X$. Another example with different Hilbert spaces or a different bilinear form is also fine; since I don't know if my example even works.


As a response to H. H. Rugh's answer:

  • If we have $\|v\|_{H^1}\leq 1$, then $|v(x)|=|v(0)+\int_0^x v'(y)dy|\leq \|v'\|_{L^1}\leq \|v'\|_{L^2} \leq \|v\|_{H^1}\leq 1$ hence $|v(x)| \leq 1$
  • Let $u_\varepsilon$ be piecewise continuous such that $u_\varepsilon(\tfrac12 \pm \varepsilon)=0$ and $u_\varepsilon(\tfrac12)=\sqrt{\varepsilon}.$ Then $$u_\varepsilon'(x)=\begin{cases} 1/\sqrt{\varepsilon}, &x \in (\tfrac12 -\varepsilon, \tfrac12) \\ -1/\sqrt{\varepsilon}, &x \in (\tfrac12,\tfrac12 +\varepsilon) \end{cases}$$
  • Computing the norms we have $\|u_\varepsilon'\|_{L^1}=\tfrac{2}{\sqrt{\varepsilon}}(\tfrac12-\tfrac12+\varepsilon)=2\sqrt{\varepsilon}$ and $\|u_\varepsilon'\|_{L^2}=\sqrt{2}$.
  • We have $$\sup_{\|v\|_{H^1} \leq 1} \frac{|\int_0^1 u_\varepsilon' v \, dx|}{\|u_\varepsilon\|_{H^1}} \leq \sup_{\|v\|_{H^1} \leq 1} \frac{|\int_0^1 u_\varepsilon' \, dx|}{\|u_\varepsilon'\|_{L^2}}\leq \frac{\|u_\varepsilon'\|_{L^1}}{\|u_\varepsilon'\|_{L^2}}=\sqrt{2 \varepsilon} \xrightarrow{\varepsilon \to 0} 0$$ and hence for every constant $\beta_X>0$ we have a $u_\varepsilon$ (as chosen above) such that $$\underbrace{\underset{\|v\|_{H^1} \leq 1}{\sup}\frac{|\int_0^1 u_\varepsilon' v \, dx|}{\|u_\varepsilon\|_{H^1}}}_{\to 0} < \beta_X.$$
$\endgroup$
2
+50
$\begingroup$

For fixed $u\in H^1_0$: If $v\in H^1_0$ has norm one, then $|v(x)|\leq 1$ for all $x$ so the sup over such $v$'s is bounded by $$ \frac{1}{\|u\|_H} \int_0^1 |u'| dx$$ This can be made arbitrarily small by choosing a suitable spike for $u$, e.g. piecewise linear with $u(\frac12)={\sqrt{\epsilon}}$, $u(\frac12\pm \epsilon)=0$ and letting $\epsilon\rightarrow 0$. The H-norm stays bounded away from zero but the integral goes to zero.

$\endgroup$
  • $\begingroup$ Thank you very much for your answer. I think I got it, can you please look at my question again? I edited it and wrote down how I understood your answer. $\endgroup$ – Fritz Sep 3 '17 at 11:50
  • 1
    $\begingroup$ @MarvinF Since $v(0)=0$ (when in $H_0^1$) you have just $|v(x)|\leq \int|v'| \times 1 \leq \|v'\|_{L^2}$ (Cauchy-Schwarz). For the second part $u'$ is non-zero only on the interval $[-\epsilon,\epsilon]$ so the square-norm is just 2. A part from this, your interpretation is fine. $\endgroup$ – H. H. Rugh Sep 3 '17 at 13:53

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.