EDIT: Let $M$ and $N$ are two smooth manifold and suppose $N$ is compact but $M$ is not necessarily compact. For my purpose, I just need to consider the case $M=\mathbb R \times S^1$ or $\mathbb R \times [0,1]$.

Thanks to David's comments that help me a lot, the problem I really concern is just the following.


  1. Now that we know if $M$ is non-compact then $C^\infty(M,N)$ is modeled on spaces $C^\infty_c(M,N)$. Since we know $W^{k,p}_0(\Omega)$ is the closure of $C^\infty_c(\Omega) $ in $W^{k,p}(\Omega)$ for $\Omega\subset \mathbb R$, in analogy, can we say $W^{k,p}(M,N)$ is modeled on spaces $W^{k,p}_0(M,f^*TN)$? Any reference?
  2. Can we consider $W^{k,p}_0(M,f^*TN)$ as a completion (in some sense?) of $C^\infty_c(M,f^*TN)$, the spaces of smooth sections with compact support of pullback bundles along $f:M\to N$?

Previous post:

(1) Can we define $W^{1,p}_0(M,N)$ in a similar manner that $W^{1,p}_0(\mathbb R)$ is defined by the completions of $C^\infty_0$ or $C^\infty_c$?

  • For example, Floer in his paper (see Definition 2.1) actually discusses (roughly) $\mathcal W:=W_{loc}^{k,p}(\mathbb R \times [0,1], N)$ with a topology given by open sets as follows: $ \mathcal O_{u,\rho,\epsilon} =\{ v\in \mathcal W \mid v= \exp_u \xi ~\text{on}~ [-\rho,\rho]\times[0,1], ~\text{and}~ ||\xi||_{W^{k,p}} <\epsilon \} $ On the other hand, Audin and Damian in their book (see Definition 8.2.2) consider Banach manifolds $\widetilde {\mathcal W} =W^{k,p}(\mathbb R \times [0,1], N)$ (actually $[0,1]$ should be replaced by $S^1$) in a quite different manner. Here open sets are the space of maps of the form $v=\exp_u \xi$ where $\xi \in W^{k,p}(\mathbb R \times [0,1], N)\equiv W_0^{k,p}(\mathbb R \times [0,1], N)$ and where $u$ is smooth and converges in some decay at the infinity.

    Heuristically, $\mathcal W$ is like the completion of $C_c^\infty$ while $\widetilde {\mathcal W}$ is like that of $C_0^\infty$.

(2) Is $\mathcal W$ the same as $\widetilde{\mathcal W}$? Which one could be a better candidate for the definition of $W^{k,p}_0(M,N)$? Are they both Banach manifolds as in the case $M$ is compact?

  • Recently I notice that when considering (infinite-dimensional) manifolds of mapping, say $C^\infty(M,N)$, we usually require $M$ to be compact. (See Section 4.2 of this paper: The inverse function theorem of Nash and Moser). Notice that as long as the domain $M$ is compact, the questions (1), (2) and (3) become trivial.

(3) In general, if $M$ is non-compact, then is $C^\infty(M,N)$ still a Frechet manifold as in the case $M$ is compact?

  • $\begingroup$ $W^{1,p}(\mathbb{R})$ is not the completion of $C^\infty(\mathbb{R})$. It is the completion of $W^{1,p} \cap C^\infty$. $\endgroup$ Jun 11, 2017 at 16:09
  • $\begingroup$ I see, I have edited. $\endgroup$
    – Hang
    Jun 11, 2017 at 16:20
  • 1
    $\begingroup$ If M is non-compact, $C^\infty(M,N)$ is locally convex, but not necessarily complete, as the model spaces are of compactly supported smooth sections. $\endgroup$
    – David Roberts
    Jul 12, 2017 at 5:22
  • 1
    $\begingroup$ See arxiv.org/abs/1603.09127 $\endgroup$
    – David Roberts
    Jul 15, 2017 at 5:19
  • 1
    $\begingroup$ In answer to your question, without compact support one cannot define the relevant supremum seminorms on partial derivatives. Of course, the function spaces themselves contain all smooth functions. $\endgroup$
    – David Roberts
    Jul 15, 2017 at 6:20

2 Answers 2


Let $M$ and $N$ be Riemannian manifolds. In general, the space of Sobolev mappings $W^{k,p}(M,N)$ should not be defined as a completion of smooth mappings even if $k=1$ and manifolds are compact. The common definition (at least if $N$ is compact) is as follows. Take an isometric embedding of $N$ into a Euclidean space $\mathbb{R^\nu}$ and then define: $$ W^{k,p}(M,N)=\{ f\in W^{m,p}(M,\mathbb{R}^\nu):\, f(x)\in N \text{ a.e.} \}. $$ This space is equipped with the metric inherited from the Sobolev norm and in general smooth mappings are not dense [1]. There are some problems when $N$ is not compact. You can find more papers about higher order Sobolev mappings between manifolds, including the case of non-compact target, at the homepage of Van Schaftingen.

[1] P. Bousquet, A. C. Ponce, J. Van Schaftingen, Strong density for higher order Sobolev spaces into compact manifolds. J. Eur. Math. Soc. (JEMS) 17 (2015), no. 4, 763–817.


Too long for a comment: If $M$ is a manifold, you have no volume form, and you cannot give a meaning to integrability of a function. To express integrability, you have to use densities, which are smooth sections of the fiber bundle of densities over the manifold $M$. Note that you don't have this difficulty with your example, since you have then the form $dx\wedge d\theta$ on $\mathbb R\times\mathbb R /2π\mathbb Z$.

  • 2
    $\begingroup$ The issue is not with $M$, where you can simply restrict to functions that are period in the second input. The issue is how to define Sobolev spaces of maps into a manifold $N$. $\endgroup$
    – Deane Yang
    Jun 11, 2017 at 17:40

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy