For $m>0$ we consider the ring $C^{\infty}(\mathbb{R}^{m})$ of smooth functions on $\mathbb{R}^{m}$. For $n>0$ we consider the projection $\mathbb{R}^{m+n}\to \mathbb{R}^{m}$ hence $C^{\infty}(\mathbb{R}^{m+n})$ is naturally a $C^{\infty}(\mathbb{R}^{m})$ module.

My question is: is it true that $C^{\infty}(\mathbb{R}^{m+n})$ is a flat module over $C^{\infty}(\mathbb{R}^{m})$? I'm not sure if we should take the topology on algebra into consideration.

  • $\begingroup$ Is it perhaps even a free module? I have no intuition one way or another. $\endgroup$ – Gro-Tsen Dec 15 '19 at 16:50
  • $\begingroup$ My suspicion is that this should work if one uses the topological variants of flatness etc set out in the work of Helemskii and his school -- there might be some relevant work of Ogneva -- but I am away from my reference sources right now $\endgroup$ – Yemon Choi Dec 15 '19 at 21:10
  • $\begingroup$ "Example-Lemma 4.30" of springer.com/gp/book/9780387955438 might be related. $\endgroup$ – Qfwfq Dec 15 '19 at 21:15
  • $\begingroup$ Actually, I see that in an answer to one of your previous questions mathoverflow.net/a/304075 Igor Khavkhine already mentioned Ogneva's work, with a link to the relevant paper(s) and some technical details $\endgroup$ – Yemon Choi Dec 15 '19 at 21:26

This is just a long comment. Let $X,Y$ be smooth manifolds (here in particular, resp. $\mathbb{R}^n$ and $\mathbb{R}^m$), and let $M$ be a $C^\infty(X)$-module. Then any pair $(m,\phi)\in M\times C^\infty(X\times Y)$ defines a function $Y\to M$, namely $Y\ni c\mapsto \phi(x,c)m\in M$ (which is well defined as for all $c\in Y$, $\phi(x,c)\in C^\infty(X)$). Since this is correspondence is bilinear $M\times C^\infty(X,Y) \to M^Y$, it produces a morphism of $C^\infty(X)$-modules $$M\otimes C^\infty(X,Y) \to M^Y.$$ Is this morphism injective? That is, can one see $M\otimes C^\infty(X,Y) $ as a space of $M$-valued functions on $Y$? If the answer is yes, then any morphism $f:M\to M'$ produces a $f\otimes 1:M\otimes C^\infty(X,Y)\to M'\otimes C^\infty(X,Y)$ which extends to the morphism $M^Y\ni u\mapsto f\circ u\in M'^Y$, which is injective whenever $f$ is injective. Therefore $ C^\infty(X,Y)$ would be a flat $ C^\infty(X)$-module.


PS: The answer below has gaps, and it is likely incorrect.

Yes, $C^{\infty}(\mathbb{R}^{m+n})$ is a flat $C^{\infty}(\mathbb{R}^{m})$ module. Or, following @Pietro 's comment, with more generality, if $X,Y$ are smooth manifolds, then $C^\infty(X\times Y)$ is a $C^\infty(X)$ flat module. To see this, it is enough to consider $C^\infty(X)$ linear, injective maps $\varphi: K \rightarrow L$, where $K = <r_1,\dots,r_d>_{C^\infty(X)}$ is a finitely generated $C^\infty(X)$ ideal, and $L=C^\infty(X)$ is the ring of scalars.

Notice that in that case, considering that $C^\infty(X)$ is a sub-ring of $C^\infty(X\times Y)$, the tensor operation $.\otimes_{C^\infty(X)} C^\infty(X\times Y) $ applied to an ideal such as $K$ is just the extension of scalars, i.e., in some natural way, $$ K\otimes_{C^\infty(X)} C^\infty(X\times Y) = <r_1,\dots,r_d>_{C^\infty(X\times Y)}, $$ and $$L\otimes_{C^\infty(X)} C^\infty(X\times Y) = C^\infty(X\times Y).$$ Moreover, with these representations, $\bar{\varphi} = \varphi \otimes Id$, the extension of $\varphi$ from $K$ to $K\otimes C^\infty(X\times Y)$, satisfies: $$ \bar{\varphi}( \sum_{i=1}^d r_i\ m_i) = \sum_{i=1}^d \varphi(r_i)\ m_i $$ for any $m_1,\dots, m_d \in C^\infty(X\times Y)$.

We want to verify that, under the assumption the $\varphi$ is injective, $\bar{\varphi}$ is injective. Now, if for some linear combination $\sum_i r_i \ m_i$ we have $\bar{\varphi}(\sum_i r_im_i) = 0$, then $\sum_i \varphi(r_i)\ m_i = 0$ in $C^\infty(X\times Y)$, and so, for all $y \in Y$, $$ \sum_i \varphi(r_i)\ m_i(.,y) = 0 $$ in $C^\infty(X)$, but $\varphi$ being $C^\infty(X)$ linear, we get: $$ \varphi( \sum_i r_i\ m_i(.,y) ) = 0. $$ By injectivity of $\varphi$, $ \sum_i r_i\ m_i(.,y) = 0$

(With @Pietro notation, this is the condition of an element of $M^Y$ having all 0 coordinates applied to the case $M = $ a finitely generated ideal of $C^\infty(X)$)

Since this holds for all $y$, then $\sum_i r_i\ m_i = 0$, which shows injectivity of $\varphi$.

  • $\begingroup$ I am outside my comfort zone when it comes to Frechet algebras and modules rather than Banach ones, but in extension-of-scalars arguments don't we need to ise a suitable completed tensor product? Even at the vector space level, my gut feeling is that $C^\infty({\bf R})\otimes C^\infty({\bf R})$ is all of $C^\infty({\bf R}^2)$, as this seems similar to claiming that all continuous linear maps $C^\infty({\bf R})\to C^\infty ({\bf R})'$ have finite rank... $\endgroup$ – Yemon Choi Dec 15 '19 at 21:09
  • $\begingroup$ @Yemon Choi: According to Nestruev (see my previous comment in the OP), $\mathcal{C}^\infty (M\times N)$ is isomorphic to the "smooth envelope" (3.36) of the tensor product over $\mathbb{R}$ of the two algebras. There, "$\mathcal{C}^\infty$-closed" is defined by means of a property (3.32) that reminds me of the $\mathcal{C}^\infty$-schemes by Joyce. $\endgroup$ – Qfwfq Dec 15 '19 at 21:49
  • 1
    $\begingroup$ Maybe I'm missing some obvious fact, but the last line is not quite clear to me: If, for all $y\in Y$, one has $\sum_i r_i m_i(.,y)=0$ as an element of $K$, that is $\sum_i r_i m_i$ induces the zero function, that is it is $0$ as an element of $K^Y$, why can we conclude that $\sum_i r_i m_i=0$ as an element of $K\otimes C^\infty(X\times Y)$? $\endgroup$ – Pietro Majer Dec 15 '19 at 23:50
  • 1
    $\begingroup$ Translating into algebraic language: Set $R := C(X), S := C(X\times Y), \varphi: K \hookrightarrow R$ for $K \triangleleft R$. Then $R \le S$ and what you are saying is that there is a commutative diagramm $$K \otimes_R S \xrightarrow{\varphi \otimes id} R\otimes_R S$$ $$\cong\, \downarrow \qquad\qquad \downarrow\ \,\cong$$ $$KS \quad \xrightarrow[\bar{\varphi}] \quad S$$ Why is the left vertical arrow an isomorphism? $\endgroup$ – tj_ Dec 16 '19 at 5:27
  • 4
    $\begingroup$ I don't see how your argument uses in any way properties of the ring of smooth functions. Is the corresponding question known for rings of continuous functions on (locally) compact Hausdorff spaces $X$ and $Y$? This is perhaps in the comfort zone of Yemon Choi. $\endgroup$ – Jochen Wengenroth Dec 16 '19 at 9:46

Let $X,Y$ be smooth manifolds; let $\Im = <r_1,\dots,r_d>_{C^\infty(X)}$ be a finitely generated ideal of $C^\infty(X)$. Then $\Im \otimes_{C^\infty(X)} C^\infty(X \times Y) = \{ \sum_{i=1}^d r_i \otimes g_i \ | \ g_1,\dots,g_d \in C^\infty(X \times Y)\}$. I claimed elsewhere that the condition $\sum_{i=1}^d r_i \otimes g_i = 0$ in $\Im \otimes C^\infty(X \times Y)$ is equivalent to $\sum_{i=1}^d r_i g_i = 0$ in $C^\infty(X \times Y)$. That is not the case. The correct statement is:

The following conditions are equivalent:

i) $\sum_{i=1}^d r_i g_i = 0$ in $C^\infty(X \times Y)$
ii) $k(\sum_{i=1}^d r_i \otimes g_i) = 0$ for all $k \in \Im$

Proof $i) \Rightarrow ii)$ If $\sum_{i=1}^d r_i g_i = 0$ and $k \in \Im$ then $0 = k \otimes (\sum_{i=1}^d r_i g_i ) = \sum_{i=1}^d k \otimes (r_i g_i) =$
$ \sum_{i=1}^d r_i \ (k \otimes g_i) = \sum_{i=1}^d (r_i k) \otimes g_i = k (\sum_{i=1}^d r_i \otimes g_i) $

$ii) \Rightarrow i)$ Take $k = r_j$, and assume $0 = r_j( \sum_{i=1}^d r_i \otimes \ g_i) =\sum_{i=1}^d r_i \otimes ( r_j \ g_i) $. Fix $a \in Y$, and consider the bilinear map $e : \Im \times C^\infty(X \times Y) \to C^\infty(X)$ defined as $ e(k,g) = k(x)g(x,a)$. If $\bar{e}$ is the corresponding map on $\Im \otimes C^\infty(X \times Y)$, then $0 = \bar{e}( \sum_{i=1}^d r_i \otimes ( r_j \ g_i) ) = \sum_{i=1}^d \bar{e}(r_i \otimes ( r_j \ g_i)) =$
$ \sum_{i=1}^d r_i \ r_j \ g_i(.,a) = r_j ( \sum_{i=1}^d r_i \ g_i(.,a) )$

Since this holds for all $j$, at any point $x \in X$ where some $r_j $ does not vanish, $\sum_{i=1}^d r_i(x) \ g_i(x,a) =0$, and the same holds trivially at any point $x$ where all $r_j$'s vanish.

  • $\begingroup$ As far as I see this does not answer question but rather shows that and why your previous answer is not conclusive. I would suggest that you edit your first answer and delete this one. $\endgroup$ – Jochen Wengenroth Dec 18 '19 at 8:27

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.