3
$\begingroup$

I am currently learning about algebraic viewpoint on closed embedded subsupermanifolds. In particular, I am struggling with something which should be ''easy to see''. Namely, I refer to the lemma just above Proposition 3.2.9 in "Introduction to the Theory of Supermanifolds" by D. A. Leites (http://iopscience.iop.org/0036-0279/35/1/R01).

Remark: I know there is an answer below. It, however, does not work for supermanifolds (I am not able to generalize).

The setting: Suppose we have a map $\psi: L \rightarrow M$ of supermanifolds, such that it is transversal to a given closed embedded subsupermanifold $N \subseteq M$.

Now, to every closed embedded subsupermanifold, one may assign a unique ideal $\mathcal{J}_{N} \leq C^{\infty}(M)$ in the superalgebra of its global functions, defined by $\mathcal{J}_{N} = \{ f \in C^{\infty}(M) \; | \; j^{*}(f) = 0 \}$, where $j: N \rightarrow M$ is the embedding. It can be then shown that $j^{\ast}: C^{\infty}(M) \rightarrow C^{\infty}(N)$ is a superalgebra epimorphism and $\mathcal{J}_{N}$ is its kernel, whence $C^{\infty}(N) \cong C^{\infty}(M) / \mathcal{J}_{N}$

Next, by definition of the map of supermanifolds, we have a superalgebra morphism $\psi^{\ast}: C^{\infty}(M) \rightarrow C^{\infty}(L)$. We can thus consider a subset $\psi^{\ast}(\mathcal{J}_{N}) \subseteq C^{\infty}(L)$. As $\psi^{\ast}$ is usually not surjective, this is in general not an ideal. We can, however, consider the ideal $\mathcal{I} = \langle \psi^{\ast}(\mathcal{J}_{N}) \rangle \leq C^{\infty}(L)$ generated by this subset.

THE ACTUAL QUESTION: $\mathcal{I}$ is supposed to have the following property: Let $\{ f_{\mu} \}_{\mu \in J}$ be any collection of functions in $\mathcal{I}$, such that $\{ supp(f_{\mu}) \}_{\mu \in J}$ is locally finite. Then also their sum $\sum_{\mu \in J} f_{\mu}$ must be function in $\mathcal{I}$.

I am stuck at this very point. According to D.A. Leites, this should be easy to see.

For the sake of completeness, let me recall some definitions:

A collection $\{C_{\mu} \}_{\mu \in J}$ of subsets of any topological space is locally finite, if for every compact subset $K$, $C_{\mu} \cap K \neq \emptyset$ only for finitely many $\mu \in J$. On (ordinary) manifolds, this is equivalent to every point $m$ having a neighborhood $U_{m}$, such that $C_{\mu} \cap U_{m} \neq \emptyset$ only for finitely many $\mu \in J$.

For a function $f$ on a supermanifold $M$, its support $supp(f)$ is a set of points $m$ of the underlying manifold $|M|$, where the germ $[f]_{m}$ of $f$ does not vanish.

Some comments: (i) The ideal $\mathcal{J}_{N}$ has this very property. For each $U \subseteq M$ open, let $j^{\ast}_{U}: C^{\infty}(U) \rightarrow C^{\infty}(U \cap N)$ be the superalgebra morphism induced by the pullback by $j$. For each point $m \in M$, we can pick a precompact neighborhood $U_{m}$. If $f_{\mu} \in \mathcal{J}_{N}$ for every $\mu \in J$, we obtain

$(j^{\ast}( \sum_{\mu \in J} f_{\mu} ))|_{U_{m} \cap N} = j^{\ast}_{U_{m}}( (\sum_{\mu \in J} f_{\mu} )|_{U_{m}}) = \sum_{\mu \in J} j^{\ast}_{U_{m}}( f_{\mu}|_{U_{m}}) = \sum_{\mu \in J} (j^{\ast}(f_{\mu}))|_{U_{m} \cap N} = 0,$

where it was important that after restriction to $U_{m}$, the sum is finite. But $m$ was arbitrary and $\{ U_{m} \cap N \}_{m \in M}$ forms an open cover of $N$, which proves that $j^{\ast}( \sum_{\mu \in J} f_{\mu} ) = 0$, that is $\sum_{\mu \in J} f_{\mu} \in \mathcal{J}_{N}$.

(ii) I don't know whether the transversality of $\psi$ to $N$ is somewhat important at this point.

(iii) D. A. Leites only assumes that the indexing set $J$ is countable. This is not very important though, as for general $J$, one can always find a countable subset $J'$, such that $\sum_{\mu \in J} f_{\mu} = \sum_{\mu' \in J'} f_{\mu'}$.

$\endgroup$
1
$\begingroup$

Update: I added a previous version (Version 2) which does not use dimension theory and tubular neighborhoods and it might be possible to generalize it to supermanifolds. However, it requires that $N$ can be covered by finitely many coordinate balls.

VERSION 1: FOR SMOOTH MANIFOLDS AND ANY N AND $\psi$ (uses that $N$ can be covered by $\mathrm{dim}(N)+1$ charts and that there is a tubular neighborhood of $N$)

It holds $$ \mathcal{I} = \Bigl\{ \sum_{i=1}^k (g_i \circ \psi) h_i \ \Bigl|\ k\in\mathbb{N}, g_i\in C^\infty(M), h_i\in C^\infty(L): g_i(N) = 0 \Bigr\}. $$ Given a locally finite collection of smooth functions $(f_i\in \mathcal{I} \mid i\in I)$, we will show that $$ \sum_{i\in I} f_i \in \mathcal{I}. $$

Lemma 1: We can assume that $f_i = (g_i \circ \psi) h_i$ for all $i\in\mathcal{I}$.

Proof: Pick a partition of unity $(\eta_j \mid j\in \mathcal{J})$ on $L$ such that $\eta_j$ has compact support for every $j\in\mathcal{J}$. Let $\mathcal{A}:= \mathcal{I}\times\mathcal{J}$, and define $$ f_\alpha:= \eta_j f_i $$ for all $\alpha=(i,j)\in \mathcal{A}$. The system $(f_{\alpha}\mid \alpha\in\mathcal{A})$ is locally finite, and it holds $$ \sum_{\alpha\in\mathcal{A}} f_{\alpha} = \sum_{j\in \mathcal{J}} \eta_j \sum_{i\in\mathcal{I}} f_i = \sum_{i\in\mathcal{I}} f_i. $$ For every $i\in\mathcal{I}$, there is an $m_i\in \mathbb{N}$ such that $f_i = \sum_{l=1}^{m_i} (g_{il}\circ\psi)h_{il}$. It follows that for every $\alpha = (i,j)\in\mathcal{A}$, it holds $$ f_\alpha = \eta_j f_i = \sum_{l=1}^{m_i} (g_{il}\circ\psi)(\eta_j h_{il}) = \sum_{i=1}^{m_i} (g_{il}\circ\psi)h_{\alpha l} \in \mathcal{I}, $$ where we defined $$ h_{\alpha l} := \begin{cases} 0 & \text{if }f_\alpha=0, \\ \eta_j h_{il} & \text{otherwise.} \end{cases} $$ Let $x\in L$. There is an open neighborhood $U$ of $x$ and a finite subset $\mathcal{J}_0\subset \mathcal{J}$ such that $\mathrm{supp}(\eta_j) \cap U = 0$ for all $j\in \mathcal{J}\backslash\mathcal{J}_0$. Because $\sum_{j\in \mathcal{J}}\eta_j = 1$, it holds $U\subset \bigcup_{j\in \mathcal{J}_0} \{ x\in M \mid \eta_j(x)\neq 0\}$. Because $\bigcup_{j\in \mathcal{J}_0} \mathrm{supp}(\eta_j)$ is compact, there exists a finite subset $\mathcal{I}_0\subset \mathcal{I}$ such that $$ \mathrm{supp}(f_i)\cap \bigcup_{j\in \mathcal{J}_0} \mathrm{supp}(\eta_j) = \emptyset $$ for all $i\in \mathcal{I}\backslash\mathcal{I}_0$. Suppose that $\mathrm{supp}(h_{\alpha l})\cap U \neq 0$ for some $\alpha = (i,j)$ and $l\in \{1,\dotsc,m_i\}$. Because $\mathrm{supp}(h_{\alpha l})\subset\mathrm{supp}(\eta_j)$, it must hold $j\in \mathcal{J}_0$. By definition of $h_{\alpha l}$, it holds either $h_{\alpha l} = 0$ if $f_\alpha = 0$, which is equivalent to $\{x\in M \mid f_i(x)\neq 0\}\cap\{x\in M\mid \eta_j(x)\neq 0\}=\emptyset$ which is equivalent to $\mathrm{supp}(f_i)\cap\{x\in M\mid \eta_j(x)\neq 0\} = \emptyset$, or $h_{\alpha_l} = \eta_j h_{il}$. The second option can possibly occur only for $i\in\mathcal{I}_0$. This shows that the collection $$ ((g_{il}\circ\psi)h_{\alpha l} \mid \alpha=(i,j)\in\mathcal{A}, l\in\{1,\dotsc,m_i\}) $$ is locally finite. Its sum equals $\sum_{\alpha\in\mathcal{A}} f_\alpha$ and hence $\sum_{i\in \mathcal{I}}f_i$ by construction. QED

By Lemma 1, we can assume that $f_i = (g_i\circ\psi)h_i$ for $g_i\in C^\infty(M)$ with $g_i(N)=0$ and $h_i\in C^\infty(L)$ without lost of generality.

Denote $k:=\dim(N)$ and $n:=\dim(M)$. Pick a tubular neighborhood $\mathcal{N}(N)$ of $N$ in $M$. The $k$-dimensional manifold $N$ can be always covered by $k+1$ (not necessarily connected) charts $U_1$, $\dotsc$, $U_{k+1}$. Every chart $U_j$ on $N$ induces a submanifold chart $V_j = \mathcal{N}(U_j)$ on $M$. Let $V_0\subset M$ be an open subset disjoint from $N$ such that $M = \cup_{j=0}^{k+1} V_j$. Let $\lambda_0$, $\dotsc$, $\lambda_{k+1}$ be a subordinate partition of unity. Let $\mu$ be a bump function which equals $1$ on $\mathrm{supp}(\lambda_0)$ and vanishes on $N$.

Let $(x_j,y_j)\in \mathbb{R}^n$ be coordinates on $\mathcal{N}(U_j)$ such that $x_j = (x_j^1,\dotsc,x_j^k)$ gives coordinates on the base and $y_j = (y_j^1,\dotsc,y_j^{n-k})$ on fibers. An important feature of $\mathcal{N}(U_j)$ is that it contains the vertical line $\gamma(t) = (x_j,0) + t((x_j,y_j)-(x_j,0))$ connecting $(x_j,0)$ and $(x_j,y_j)$. The Fundamental theorem of calculus in the form $$ f(\gamma(1))-f(\gamma(0)) = \int_{0}^1 (\nabla f)(\gamma(t))\cdot\gamma'(t) \mathrm{d}t $$ then asserts that the following holds for all $i\in I$ and $j\in \{1,\dotsc,k+1\}$ on the entire $\mathcal{N}(U_j)$: $$ (\lambda_j g_i)(x_j,y_j) - \underbrace{(\lambda_j g_i)(x_j,0)}_{=0} = \sum_{a=1}^{n-k} y^a_j \underbrace{\int_{0}^1 \frac{\partial(\lambda_j g_i)}{\partial y^a_j}(x_j,ty_j) \mathrm{d}t}_{\displaystyle=:u_{i a}^j}. $$ Let $\tilde{y}^a_j$ and $\tilde{u}_{ia}^j$ be the smooth functions on $M$ obtained from $y^a_j$ and $u_{ia}^j$, respectively, by multiplication with a bump function which is $1$ on $\mathrm{supp} \lambda_j$ and $0$ on a neighborhood of the closure of the complement of $\mathcal{N}(U_j)$.

For all $i\in I$ and $j\in \{1,\dotsc,k+1\}$, we have the following relations on $M$: $$ \lambda_0 g_i = \mu \lambda_0 g_i\quad\text{and}\quad\lambda_j g_i = \sum_{a=1}^{n-k} \tilde{y}^a_j \tilde{u}_{ia}^j. $$ Using this, we compute \begin{align*} \sum_{i\in I} (g_i \circ \psi) h_i &= \sum_{i\in I} \sum_{j=0}^{k+1} (\lambda_j g_i \circ \psi) h_i \\ & = \sum_{i\in I} (\lambda_0 g_i \circ \psi) h_i + \sum_{i\in I} \sum_{j=1}^{k+1} \sum_{a=1}^{n-k} (\tilde{y}^a_j \circ \psi)(\tilde{u}_{ia}^j \circ \psi)h_i \\ & = (\mu\circ\psi)\sum_{i\in I}(\lambda_0 g_i \circ \psi) h_i+ \sum_{j=1}^{k+1} \sum_{a=1}^{n-k} (\tilde{y}^a_j\circ \psi) \sum_{i\in I} (\tilde{u}_{ia}^j \circ \psi)h_i\\ & = (G_0 \circ \psi) H_0 + \sum_{j=1}^{k+1} \sum_{a=1}^{n-k} (G_{ja}\circ\psi)H_{ja}, \end{align*} where we denoted $$ G_0:= \mu,\quad G_{ja}:=\tilde{y}^a_j,\quad H_0:=\sum_{i\in I}(\lambda_0 g_i \circ \psi) h_i,\quad H_{ja}:=\sum_{i\in I} (\tilde{u}_{ia}^j\circ \psi)h_i. $$ It holds $G_0$, $G_{ja}\in C^\infty(M)$, $G_0(N)=G_{ja}(N) = 0$, $H_0$, $H_{ja}\in C^\infty(L)$, and it follows that $\sum_{i\in I} (g_i \circ \psi) h_i \in \mathcal{I}$.

VERSION 2: PROOF WHEN $N$ CAN BE COVERED BY FINITELY MANY COMPATIBLE COORDINATE BALLS (not using dimension theory and tubular neighborhood)

Write $\mathbb{R}^n = \mathbb{R}^k\times \mathbb{R}^{n-k}$ with coordinates $(x,y)$. Let $f: \mathbb{R}^n\rightarrow \mathbb{R}$ be a smooth function vanishing at $\{(x,y) \mid x = 0\}$. Then the fundamental theorem of calculus asserts that the following holds for all $(x,y)\in \mathbb{R}^n$: $$ f(x,y) = \sum_{j=1}^{k} x^j \int_{0}^1 \frac{\partial f}{\partial x^j}(tx,y) dt. $$ Let $U_\alpha$ $(\alpha\in\mathcal{A})$ be a cover of $N$ by coordinate balls, and let $\lambda_\alpha$ $(\alpha\in\mathcal{A})$ be a subordinate partition of unity. Suppose that we are given $\sum_{i\in I} (g_i \circ \psi) h_i$ as above and we want to show that it lies in $\mathcal{I}$. We can even assume that the support lies in an arbitrary small neighborhood of $\psi^{-1}(N)$. Using the analytical fact above, there are smooth functions $x_{\alpha}^j$ vanishing on $N$ and smooth functions $u^{\alpha}_{ij}$ for all $i\in I$, $\alpha\in\mathcal{A}$ and $j\in\{1,\dotsc,k:=\mathrm{codim} N\}$ such that $$ \lambda_\alpha g_i = \sum_{j=1}^k x_\alpha^j u_{ij}^\alpha. $$ We compute \begin{align*} \sum_{i\in I} (g_i \circ \psi) h_i = \sum_{j=1}^k \sum_{\alpha\in\mathcal{A}} (\lambda_\alpha x^j_\alpha\circ\psi) \sum_{i\in I} (u_{ij}^\alpha \circ \psi) h_i. \end{align*} If $\mathcal{A}$ is finite, then we are done.

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ Your answer related to the original formulation of this question (I have edited it significantly). In a summary - you give an answer for the case where all the manifolds are ordinary manifolds. However, in supermanifolds, one does not have a full apparautus of tubular neighborhoods, or the fact that k-dimensional manifold can be covered by k+1 charts. $\endgroup$ – Jan Vysoky Apr 21 at 13:53
  • $\begingroup$ Version 2 can be significantly simplified. By the modification of Lemma 1 above, you can assume that $f_{i} = (g_{i} \circ \psi) h_{i}$, where $\{ supp(g_{i}) \}_{i \in I}$ is locally finite. It then suffices to cover $N$ by finitely many precompact sets $\{U_{\alpha} \}_{\alpha=1}^{m}$ (e.g. coordinate balls). Let $U_{0} := M - N$ and let $\{ \lambda_{\alpha} \}_{\alpha = 0}^{m}$ be the corresponding partition of unity. Note that $\lambda_{0} \in \mathcal{J}_{N}$. Now, as $U_{\alpha}$ are compact for $\alpha > 0$, there is a finite subset $I_{\alpha} \subseteq I$ such that $\endgroup$ – Jan Vysoky Apr 28 at 12:52
  • $\begingroup$ such that $\lambda_{\alpha} \cdot g_{i} \neq 0$ only for $i \in I_{\alpha}$. Also note that $\lambda_{\alpha} \cdot g_{i} \in \mathcal{J}_{N}$. Then $\sum_{i \in I} (g_{i} \circ \psi) h_{i} = \sum_{i \in I} ((\sum_{\alpha=0}^{m}\lambda_{\alpha}) \cdot g_{i} \circ \psi) h_{i} = (\sum_{i \in I} (g_{i} \circ \psi) h_{i}) \cdot (\lambda_{0} \circ \psi)$ $+ \sum_{\alpha=1}^{m} \sum_{i \in I_{\alpha}} (\lambda_{\alpha} \cdot g_{i} \circ \psi) h_{i}$. The first summand is in $\mathcal{I}$ as $\lambda_{0} \in \mathcal{J}_{N}$ and the rest is a finite sum of elements in $\mathcal{I}$. $\endgroup$ – Jan Vysoky Apr 28 at 13:02

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.