I wish to prove that it is possible to relate the Sobolev norms on an arbitrary triangle $K$ to Sobolev norms on a reference triangle $\hat{K}$.


To this end: Let $F \colon \hat{K} \to K$ be an invertible affine map given by $F(\hat{x}) = B\hat{x} + c$. For a function $\hat{v} \in C^2(\hat{K})$, we define the corresponding function $v \in C^2(K)$ by $$ v(x) = (\hat{v}\circ F^{-1})(x). $$

I am interested in giving a bound for the Sobolev semi-norm $|v|_{2, K}$ in terms of $|\hat{v}|_{2, \hat{K}}$ and the matrix $B$.

I use the definitions

$$ |v|_{2, K} = \left(\int_{K} \sum_{|\alpha| = 2} |\partial^\alpha v(x)|^2 dx \right)^{1/2} $$ where $\partial^\alpha v$ is the the mixed partial derivatives of order $|\alpha| = 2$.

What I have tried

I started by trying to bound $|\partial^\alpha v(x)|$ in terms of the directional derivatives

$$ |\partial^\alpha v(x)| \leq \sup_{\|\xi_1\|, \|\xi_2\| \leq 1} |\nabla ((\nabla v(x))\cdot \xi_1)\cdot \xi_2)| $$ and then use that since $v(x) = \hat{v}(\hat{x})$ this equals

$$ |\partial^\alpha v(x)| \leq \sup_{\|\xi_1\|, \|\xi_2\| \leq 1} |\nabla ((\nabla v(x))\cdot \xi_1)\cdot \xi_2)| \\ = \sup_{\|\xi_1\|, \|\xi_2\| \leq 1} |\nabla ((\nabla v(x))\cdot B^{-1}\xi_1)\cdot B^{-1}\xi_2)| \\ \leq \sup_{\|\xi_1\|, \|\xi_2\| \leq 1} |\nabla ((\nabla v(x))\cdot \xi_1)\cdot \xi_2)|\| B^{-1} \|^{2}, $$

however - I get lost in the notation. Am I on the right path, and is there any better notation for working with derivatives in this fashion?


Yes, it is possible to estimate $|v|_{2,K}$ with $|\hat v|_{2,\hat K}$. It is even true for the general case $|v|_{m,K}$ and in literature it is known as the "Transformation rule", for example see "Finite Elements" by D. Braess.

$$|v|_{m,K} \leq C \|B^{-1}\|^m |\det B|^{1/2} |\hat v|_{m,\hat K}, \qquad \forall v \in H^m(K)$$

We have by the chain rule $$|\partial^\alpha v|_{0,\hat K} \leq C \|B^{-1}\|^m \sum_{|\beta|=m} |\partial^\beta \hat v ~\circ ~F^{-1}|_{0,\hat K} \leq C \|B^{-1}\|^m |\det B|^{1/2} |\partial^\alpha \hat v|_{0,K}$$ and summing over $\alpha$ for $|\alpha|=m$ gives the desired result. Your notation is fine (I am not aware of a better one) and indeed you were on the right track.


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.