An inverse estimate in the finite element method

350 Views Asked by At

Let $\hat K= [0,1]^d$ and $\hat K=\hat K_1 \cup \hat K_2 $, with $\hat K_i$ closed connected, with a non-empty interior, and $\hat K_1^\circ \cap \hat K_2^\circ=\emptyset$ . We denote as $\mathbb Q_p (\hat K)$ the space of polynomials with degree at most $p$ in each variable (tensor product space). We want to prove that there exists $C>0$ such that $$ \| \varphi \|_{L^\infty (\hat K_1)} \le C \| \varphi \|_{L^2(\hat K_2)}\qquad\forall\ \varphi\in\mathbb Q_p(\hat K). $$ Moreover, we would like to know how $C$ depends on the relative measures of $\hat K_1$, $\hat K_2$ and on the polynomial degree $p$.

Here, we show how we proceed. Let $\bar x\in \hat K_1$ be such that $|\varphi (\bar x)| = \| \varphi \|_{L^\infty (\hat K_1)}$. There are two cases:

  1. $\| \varphi \|_{L^\infty (\hat K_1)} < \| \varphi \|_{L^\infty (\hat K_2)}$;
  2. $\| \varphi \|_{L^\infty (\hat K_1)} \ge \| \varphi \|_{L^\infty (\hat K_2)}$.

If 1. holds, then there exists $\tilde x \in \hat K_2$ such that $\| \varphi \|_{L^\infty (\hat K_2)} = |\varphi(\tilde x)| > |\varphi(\bar x)| = \| \varphi \|_{L^\infty (\hat K_1)}$. We claim that, from the continuity of $\varphi$, there exist $C_1>0$ and $\varepsilon >0$ such that $$ \begin{cases} \forall x \in B_{\varepsilon} (\tilde x) \qquad |\varphi (x)| > C_1 |\varphi (\bar x)|,\\ B_{\varepsilon} (\tilde x) \subset \hat K_2. \end{cases} $$ Thus, $$ \| \varphi \|_{L^2(\hat K_2)}^2 \ge \int_{B_{\varepsilon}(\tilde x)} |\varphi(x)|^2 > C_1 \int_{B_{\varepsilon}(\tilde x)} |\varphi(\bar x)|^2 = C_1 |B_{\varepsilon}(\tilde x)| \| \varphi \|_{L^\infty(\hat K_2)}^2 \ge C_1 |B_{\varepsilon}(\tilde x)| \| \varphi \|_{L^\infty(\hat K_1)}^2. $$ If 2. holds, then $|\varphi(\bar x)|=\| \varphi \|_{L^\infty (\hat K_1)} \ge \| \varphi \|_{L^\infty (\hat K_2)}$. We claim that, from the continuity of $\varphi$, there exist $\varepsilon >0$ and $C_1>0$ such that $$ \begin{cases} \forall x \in B_{\varepsilon} (\bar x) \qquad |\varphi (x)| > C_1 |\varphi (\bar x)|,\\ | B_{\varepsilon} (\tilde x) \cap \hat K_2| \ne 0. \end{cases} $$ Thus, $$ \| \varphi \|_{L^2(\hat K_2)}^2 \ge \int_{B_{\varepsilon}(\bar x)\cap \hat K_2} |\varphi(x)|^2 > C_1 \int_{B_{\varepsilon}(\bar x)\cap \hat K_2} |\varphi(\bar x)|^2 = C_1 |B_{\varepsilon}(\bar x)\cap \hat K_2| \| \varphi \|_{L^\infty(\hat K_1)}^2. $$ So, we are left with the choices of $\varepsilon$ in both cases. Do you think this is the right strategy? Any ideas on how to finish?

1

There are 1 best solutions below

1
On BEST ANSWER

This should follow from the equivalence of norms on finite dimensional spaces. What is interesting is that $\|\cdot\|_{L^\infty(K_1)}$ and $\|\cdot\|_{L^2(K_2)}$ are norms on $\mathbb Q_p(K)$. Let's sketch the proof that this is true for the $L^2$-norm. The norm axioms are trivial other than $\|v\|_{L^2(K_2)}=0$ implies $v=0$ on $K$ (note that this property is special to polynomials). If $\|v\|_{L^2(K_2)}=0$, then $v=0$ in $K_2$, and therefore is zero inside a $d$-dimensional ball in $K_2$. Therefore, there are $d$ Hyperplanes on which $v$, $\nabla v$, $D^2v$, and so on are all zero. We can then use the fact that if a polynomial $v$ of degree $p$ vanishes on a hyperplane $L:=L(x)$, then $v = Lv'$, where $v$ is of degree $p-1$. We can piece this together (also applying it to the derivatives) to find that $v$ must by identically zero (i.e., all the degrees of freedom of $v$ are exhausted). This holds everywhere, and therefore $\|\cdot\|_{L^2(K_2)}$ is a norm on $\mathbb Q_p(K)$. We can argue similarly for $\|\cdot\|_{L^\infty(K_1)}$.

Combining the fact that all norms are equivalent on finite dimensional spaces with a scaling argument, we obtain the desired estimate. To get the explicit dependence, on the volumes of $K_1$ and $K_2$, this comes directly from the scaling.