3
$\begingroup$

Let $U$ be an open manifold (say an open subset of $\mathbb{R}^n$ for simplicity). Denote by $\mathscr{S}(U)$ the space of Schwartz functions on $U$. Schwartz functions are defined as usual to be those decreasing faster than any polynomial with respect to some compact exhaustion.

Define a subcomplex in the complex of differential forms on $\mathbb{R}^n$ spanned by those forms which have coefficients from the space $\mathscr{S}(U)$. Is there a good description of the cohomology of this complex?

I suspect that the answer should be related to properties of $U$ at infinity, maybe to the topology of the end space of $U$? Also, I know that much can be said if one considers instead the space of bump functions. Then the answer will just be the cohomology with compact support. Moreover, bump functions are dense in the space $\mathscr{S}(U)$, but I don’t know how to go from there.

I am sorry if this question is well known, and I would be very satisfied if someone could just give me a reference.

$\endgroup$
5
  • 2
    $\begingroup$ "the ideal of Schwartz functions" how can Schwartz functions make up an ideal if there are a lot of units among them? or did you mean the ideal of Schwartz functions vanishing on $U$? $\endgroup$ Commented Aug 3 at 2:15
  • 2
    $\begingroup$ Have a look at this old answer by Denis Serre, which treats a special case of the Poincaré lemma for Schwartz functions. The method should generalize when $U = \mathbb{R}^n$. $\endgroup$ Commented Aug 3 at 8:07
  • 1
    $\begingroup$ @DmitriiKorshunov, yes, sorry, it’s just a subspace $\endgroup$ Commented Aug 3 at 18:21
  • $\begingroup$ I think that Theorem 3.3.1 in arxiv.org/pdf/0802.3305 answers your question? it shows that on a Nash manifold the Schwartz and compactly supported sections of the de Rham complex have the same cohomologies. $\endgroup$
    – S. carmeli
    Commented Aug 3 at 22:26
  • 2
    $\begingroup$ @S.carmeli I’ve seen this paper, and if I understand the gist correctly it deals only with semialgebraic open manifolds. Since Schwartz functions don’t form a sheaf, I don’t think their results are applicable in general. Nevertheless, thank you so much for the reference! $\endgroup$ Commented Aug 4 at 2:30

1 Answer 1

3
$\begingroup$

Consider the simplified situation where $U$ has a finite number of ends with disjoint end neighborhoods $U_i$, $i=1,\ldots,k$. Choose an open $U_0$ with compact closure to complete the cover $\bigcup_{i=0}^k U_i = U$. The OP is supposing that there is a compact exhaustion of $U$ that defines the Schwartz property for functions, hence defining by restriction the Schwartz at infinity property for each end neighborhood $U_i$, that is, the spaces $\mathcal{S}_\infty(U_i)$. Make another simplifying assumption, that there exists a rectifiable vector field $\xi_i$ in $U_i$, $i\ge 1$, such that integrating from infinity along the flow lines of $\xi$ is well-defined for all Schwartz functions on $U_i$ and hence determines the inverse $\mathcal{L}_\xi^{-1}$ for the Lie derivative $\mathcal{L}_\xi$ on $\mathcal{S}(U_i)$. The point of these simplifying assumptions is that they are automatically satisfied when the ends have a nice enough structure (each $U_i$ has the structure of a collar neighborhood) and they will be sufficient to show that the $\mathcal{S}(U)$ de Rham cohomology is isomorphic to the $\mathcal{D}(U)$ cohomology, the one with compact supports. In general, the answer might depend more subtly on the structure of the ends of $U$ and the precise way the Schwartz condition is defined. But then the OP might decide that these simplifying conditions are already sufficiently general.

The key idea is that the Mayer-Vietoris sequence applied to the cover $U = \left(U_0 \sqcup \bigsqcup_{i=1}^k U_i\right) / (\bigsqcup_{i=1}^k U_0 \cap U_i)$ puts the Schwartz cohomology into the long exact sequence $$ \cdots \to \bigoplus_{i=1}^k H^m(U_0\cap U_i) \overset{[d]}{\to} H^m_{\mathcal{S}}(U) \to H^m(U_0) \oplus \bigoplus_{i=1}^k H^m_{\mathcal{S}_\infty}(U_i) \to \bigoplus_{i=1}^k H^m(U_0\cap U_i) \to \cdots $$ In complete analogy, if we define the functions with compact support at infinity $\mathcal{D}_\infty(U_i)$ by restriction from the compactly supported functions $\mathcal{D}(U)$, then we also have the long exact sequence $$ \cdots \to \bigoplus_{i=1}^k H^m(U_0\cap U_i) \overset{[d]}{\to} H^m_{\mathcal{D}}(U) \to H^m(U_0) \oplus \bigoplus_{i=1}^k H^m_{\mathcal{D}_\infty}(U_i) \to \bigoplus_{i=1}^k H^m(U_0\cap U_i) \to \cdots $$ But the cohomology groups of the ends with the different function spaces are actually isomorphic $H^m_{\mathcal{S}_\infty}(U_i) \cong H^m_{\mathcal{D}_\infty}(U_i)$ (in fact, they both vanish). This can be seen from the cochain maps \begin{align*} \Xi=\operatorname{id} &\colon \mathcal{D}_\infty(U_i) \otimes \Lambda^m(dx) \subset \mathcal{S}_\infty(U_i) \otimes \Lambda^m(dx) , \\ \bar{\Xi}=0 &\colon \mathcal{S}_\infty(U_i) \otimes \Lambda^m(dx) \to \mathcal{D}_\infty(U_i) \otimes \Lambda^m(dx), \end{align*} which constitute an equivalence up to homotopy, as both $\Xi \circ \bar{\Xi}$ and $\bar{\Xi} \circ \Xi$ (on appropriate domains) are equal to $$ 0 = \operatorname{id} - \mathcal{L}_\xi^{-1} \mathcal{L}_\xi = \operatorname{id} - \mathcal{L}_\xi^{-1} (\mathrm{d} \iota_\xi + \iota_\xi \mathrm{d}) = \operatorname{id} - (\mathrm{d} \mathcal{L}_\xi^{-1} \iota_\xi + \mathcal{L}_\xi^{-1} \iota_\xi \mathrm{d}) . $$ So comparing the two long exact sequences, we see also the isomorphism $H^m_\mathcal{S}(U) \cong H^m_\mathcal{D}(U)$.

$\endgroup$
2
  • $\begingroup$ Thank you so much! This argument is quite illuminating. Unfortunately, I am interested in open subsets of $\mathbb{R}^n$ where the complement has an arbitrary topology. Thus, the case you treat is, in a sense, complimentary to what I'm usually considering. Thank you again! $\endgroup$ Commented Aug 24 at 21:18
  • $\begingroup$ It seems plausible because of your argument that the difference between cohomology with Schwartz coefficients and compactly supported cohomology obstructs the existence of a collar neighbourhood of the ends. $\endgroup$ Commented Aug 24 at 21:19

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .