Let $\Omega, \Omega' \subset \mathbb{R}^n$ be two open subsets and let $\psi : \Omega \rightarrow \Omega'$ be a $C^k$ diffeomorphism. Then, $\psi$ induces by pullback a linear isomorphism $$u \mapsto u \circ \psi$$ between the Sobolev spaces $W^{k,p}_{\text{loc}}(\Omega')$ and $W^{k,p}_{\text{loc}}(\Omega)$. I assume that there is an analogous result for the Hilbert-Sobolev spaces $H^s_{\text{loc}}(\Omega)$ with real exponent $s$. Can someone please provide me a reference where this is discussed?

  • $\begingroup$ I guess the result should be obvious once you write down the definition for $H^s_{loc}(\Omega)$. But if you insist on a reference, I would recommend Friedlander's Introduction to the theory of distributions. $\endgroup$ – Hui Yu Nov 11 '12 at 3:32
  • $\begingroup$ Not for me, and the reference you provide doesn't help at all... $\endgroup$ – levap Nov 12 '12 at 17:10
  • $\begingroup$ Really? Can you tell me what did you get and where are you stuck? $\endgroup$ – Hui Yu Nov 12 '12 at 17:13
  • $\begingroup$ Say I work with the Fourier transform definition. So $u \in H^s(\Omega)$ if for all $\phi \in C^{\infty}_c(\Omega)$, we have $u\phi \in H^s(\mathbb{R}^n)$, where $H^s(\mathbb{R}^n)$ is defined using Fourier transform conditions. I have no idea how to estimate the Fourier transform of the composition $u \circ \psi$ of $u$ with a diffeomorphism in terms of the Fourier transform of $u$. $\endgroup$ – levap Nov 12 '12 at 17:38
  • $\begingroup$ I think it's better to consider $H^s$ as the Besov space $B^s_{2,2}$ and use their local characterization. The invariance under diffeomorphisms ought to be someone in the 1st or 2nd volume of Triebel's Function Spaces but this is only a guess, and I don't have access to the book now. $\endgroup$ – user53153 Dec 22 '12 at 2:40

Use the theory of operator interpolation (see for instance, the book Interpolation Spaces by Jöran Bergh, Jörgen Löfström).

The idea is as follows. Let $\Psi$ be the linear operator defined by $\Psi u = u \circ \psi$. Prove the following: $$ \Psi \in \mathcal{L}(H^k(\Omega),H^k(\Omega)) \text{ and } \Psi \in \mathcal{L}(H^{k-1}(\Omega),H^{k-1}(\Omega))$$ by using the chain rule and change of variables. Then interpolation theory provides the estimate $$ \Psi \in \mathcal{L}(H^s(\Omega),H^s(\Omega)) \text{ for } s \in (k-1,k)$$ since $H^s$ is the interpolation between $H^{k-1}$ and $H^{k}$. Moreover, we get an estimate $$ \lVert \Psi \rVert_{H^s \to H^s} \le C \lVert \Psi \rVert_{H^{k-1} \to H^{k-1}}^\theta \lVert \Psi \rVert_{H^k \to H^k}^{1-\theta} $$ where $\theta \in (0,1)$ is determined by $s$ and $k$.


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.