elge Glöckner

## Discontinuous Non-Linear Mappings on Locally Convex Direct Limits

### November 27, 2006

Abstract
We show that the self-map $f:{C}_{c}^{\infty }\left(\mathbb{R}\right)\to {C}_{c}^{\infty }\left(\mathbb{R}\right)$  , $f\left(\gamma \right):=\gamma \circ \gamma -\gamma \left(0\right)$  of the space of real-valued test functions on the line is discontinuous, although its restriction to the space ${C}_{K}^{\infty }\left(\mathbb{R}\right)$  of functions supported in $K$  is smooth (and hence continuous), for each compact subset $K\subseteq \mathbb{R}$  . More generally, we construct mappings with analogous pathological properties on spaces of compactly supported smooth sections in vector bundles over non-compact bases.
The results are useful in infinite-dimensional Lie theory, where they can be used to analyze the precise direct limit properties of test function groups and groups of compactly supported diffeomorphisms. Subject classification: 46F05, 46T20; 46A13, 46M40, 22E65
Introduction
Let ${E}_{1}\subseteq {E}_{2}\subseteq \cdots$  be an ascending sequence of locally convex spaces which does not become stationary, and such that ${E}_{n+1}$  induces the given topology on ${E}_{n}$  , for each $n$  . It is a well-known phenomenon that the topology on $E:={\cup }_{n\in \mathbb{N}}{E}_{n}$  making $E$  the direct limit of the spaces ${E}_{n}$  in the category of locally convex spaces (and continuous linear maps) can be properly coarser than the topology making $E$  the direct limit of its subspaces ${E}_{n}$  in the category of topological spaces. For example, this phenomenon occurs whenever each ${E}_{n}$  is an infinite-dimensional Fréchet space (cf. [13,Prop.4.26(ii)). In particular, the locally convex direct limit topology on the space ${C}_{c}^{\infty }\left(\mathbb{R}\right)=lim⟶{C}_{\left[-n,n\right]}^{\infty }\left(\mathbb{R}\right)$  of test functions is properly coarser than the topology of direct limit topological space (cf. also [3,p.506).
So, for abstract reasons, discontinuous mappings on the space of test functions ${C}_{c}^{\infty }\left(\mathbb{R}\right)$  are known to exist whose restriction to ${C}_{\left[-n,n\right]}^{\infty }\left(\mathbb{R}\right)$  is continuous for each $n\in \mathbb{N}$  . In this article, we describe such a mapping explicitly, whose restriction to ${C}_{\left[-n,n\right]}^{\infty }\left(\mathbb{R}\right)$  is not only continuous but actually smooth (Proposition  2.2 ). More generally, for every $\sigma$  -compact, non-compact, finite-dimensional smooth manifold $M$  of positive dimension and locally convex space $E\ne \left\{0\right\}$  , we construct a discontinuous map $f:{C}_{c}^{\infty }\left(M,E\right)\to {C}_{c}^{\infty }\left(M,\mathbb{R}\right)$  whose restriction to ${C}_{K}^{\infty }\left(M,E\right)$  is smooth, for each compact subset $K$  of $M$  . An analogous result is obtained for the space ${C}_{c}^{\infty }\left(M,E\right)$  of compactly supported smooth sections in a bundle of locally convex spaces $E\to M$  over $M$  , with non-trivial fibre (Theorem  3.2 ).
Further developments. The preceding result is useful for the investigation of direct limit properties of infinite-dimensional Lie groups. As shown in [11, it entails that there are discontinuous (and hence non-smooth) mappings on the Lie group ${\text{Diff}}_{c}\left(M\right)={\cup }_{K}{\text{Diff}}_{K}\left(M\right)$  of compactly supported smooth diffeomorphisms of $M$  (as in [14or [10), whose restriction to ${\text{Diff}}_{K}\left(M\right):=\left\{\phi \in \text{Diff}\left(M\right):\phi {|}_{M\K}={\text{id}}_{M\K}\right\}$  is smooth, for each compact subset $K\subseteq M$  . A similar pathology occurs for the Lie group ${C}_{c}^{\infty }\left(M,G\right)={\cup }_{K}{C}_{K}^{\infty }\left(M,G\right)$  of compactly supported smooth maps with values in a non-discrete finite-dimensional Lie group (as in [5). In this way, we obtain one half of the following table, which describes whether ${\text{Diff}}_{c}\left(M\right)=lim⟶{\text{Diff}}_{K}\left(M\right)$  and ${C}_{c}^{\infty }\left(M,G\right)=lim⟶{C}_{K}^{\infty }\left(M,G\right)$  holds in the categories shown:
 category $\$  group ${C}_{c}^{\infty }\left(M,G\right)$ ${\text{Diff}}_{c}\left(M\right)$ Lie groups yes yes topological groups yes yes smooth manifolds no no topological spaces no no
For the proof, see [11(cf. also [18for related results).
The present constructions of pathological mappings are complemented by investigations in [8[10(cf. also [7). In these articles, a mild additional property is introduced which ensures that a map $f:{C}_{c}^{\infty }\left(M,E\right)\to {C}_{c}^{\infty }\left(N,F\right)$  between spaces of test functions (or compactly supported sections) satisfying this property (an “almost local” map) is indeed smooth if and only if it is smooth on ${C}_{K}^{\infty }\left(M,E\right)$  for each $K$  . In contrast to these mappings, the pathological examples presented here are extremely non-local.
In the final section, we describe examples of discontinuous bilinear mappings which are continuous (and hence analytic) on each step of a directed sequence of subspaces.

1 Preliminaries

In this article, we are working in the setting of infinite-dimensional differential calculus known as Keller's ${C}_{c}^{\infty }$  -theory, based on smooth maps in the sense of Michal-Bastiani (see [4, [12, [14, [16for further information).
Definition 1.1 Let $E$  , $F$  be locally convex spaces and $f:U\to F$  be a mapping, defined on an open subset $U$  of $E$  . We say that $f$  is of class ${C}^{0}$  if $f$  is continuous. If $f$  is a continuous map such that the two-sided directional derivatives $df\left(x,v\right)={lim}_{t\to 0}\frac{1}{t}\left(f\left(x+tv\right)-f\left(x\right)\right)$  exist for all $\left(x,v\right)\in U×E$  , and the map $df:U×E\to F$  so defined is continuous, then $f$  is said to be of class ${C}^{1}$  . Recursively, given $k\in \mathbb{N}$  we call $f$  a mapping of class ${C}^{k+1}$  if it is of class ${C}^{1}$  and $df$  is of class ${C}^{k}$  on the open subset $U×E$  of $E×E$  . We set ${d}^{k+1}f:=d\left({d}^{k}f\right)={d}^{k}\left(df\right):U×{E}^{{2}^{k+1}-1}\to F$  in this case. The function $f$  is called smooth (or of class ${C}^{\infty }$  ) if it is of class ${C}^{k}$  for each $k\in {\mathbb{N}}_{0}$  .
Definition 1.2 Let $M$  be a finite-dimensional, $\sigma$  -compact smooth manifold and $E$  be a locally convex topological vector space. We equip the vector space ${C}^{\infty }\left(M,E\right)$  of $E$  -valued smooth mappings $\gamma$  on $M$  with the topology of uniform convergence of ${\partial }^{\alpha }\left(\gamma \circ {\kappa }^{-1}\right)$  on compact subsets of $V$  , for each chart $\kappa :M\supseteq U\to V\subseteq {\mathbb{R}}^{d}$  of $M$  and multi-index $\alpha \in {\mathbb{N}}_{0}^{d}$  (where $d:=dim\left(M\right)$  ). Given a compact subset $K\subseteq M$  , we equip the vector subspace ${C}_{K}^{\infty }\left(M,E\right):=\left\{\gamma \in {C}^{\infty }\left(M,E\right):\gamma {|}_{M\K}=0\right\}$  of ${C}^{\infty }\left(M,E\right)$  with the induced topology. We give ${C}_{c}^{\infty }\left(M,E\right):={\cup }_{K}{C}_{K}^{\infty }\left(M,E\right)=lim⟶{C}_{K}^{\infty }\left(M,E\right)$  the locally convex direct limit topology. We abbreviate ${C}_{c}^{\infty }\left(M\right):={C}_{c}^{\infty }\left(M,\mathbb{R}\right)$  , ${C}^{\infty }\left(M\right):={C}^{\infty }\left(M,\mathbb{R}\right)$  , and ${C}_{K}^{\infty }\left(M\right):={C}_{K}^{\infty }\left(M,\mathbb{R}\right)$  . Further details can be found, e.g., in [5.

2 Example of a discontinuous mapping on ${\mathbit{C}}_{\mathbit{c}}^{\mathbsf{\infty }}\mathbf{\left(}\mathbb{R}\mathbf{\right)}$

We show that the map $f:{C}_{c}^{\infty }\left(\mathbb{R}\right)\to {C}_{c}^{\infty }\left(\mathbb{R}\right)$  , $\gamma ↦\gamma \circ \gamma -\gamma \left(0\right)$  is discontinuous, although its restriction to ${C}_{\left[-n,n\right]}^{\infty }\left(\mathbb{R}\right)$  is smooth, for each $n\in \mathbb{N}$  .
The following fact is essential for our constructions. It follows from [13,Cor.3.13and is also a special case of [8,Prop.11.3. For the convenience of the reader, we offer a direct, elementary proof as an appendix.
Lemma 2.1 The composition map $\Gamma :{C}^{\infty }\left({\mathbb{R}}^{n},{\mathbb{R}}^{m}\right)×{C}^{\infty }\left(M,{\mathbb{R}}^{n}\right)\to {C}^{\infty }\left(M,{\mathbb{R}}^{m}\right),\Gamma \left(\gamma ,\eta \right):=\gamma \circ \eta$  is smooth, for each finite-dimensional, $\sigma$  -compact smooth manifold $M$  and $m,n\in {\mathbb{N}}_{0}$  . $\square$
For the following proof, recall that the sets $\mathcal{V}\left(k,e\right):=\left\{\gamma \in {C}_{c}^{\infty }\left(\mathbb{R}\right):\left(\forall n\in \mathbb{Z}\right)\left(\forall j\in \left\{0,\dots ,{k}_{n}\right\}\right)\left(\forall x\in \left[n-\frac{1}{2},n+\frac{1}{2}\right]\right)|{\gamma }^{\left(j\right)}\left(x\right)|<{\varepsilon }_{n}\right\}$  form a basis of open zero-neighbourhoods for the topology on ${C}_{c}^{\infty }\left(\mathbb{R}\right)$  , where $k=\left({k}_{n}\right)\in \left({\mathbb{N}}_{0}{\right)}^{\mathbb{Z}}$  and $e=\left({\varepsilon }_{n}\right)\in \left({\mathbb{R}}^{+}{\right)}^{\mathbb{Z}}$  (cf. [17,§II.1; see [5,Prop.4.8).
Proposition 2.2 $f:{C}_{c}^{\infty }\left(\mathbb{R}\right)\to {C}_{c}^{\infty }\left(\mathbb{R}\right)$  , $\gamma ↦\gamma \circ \gamma -\gamma \left(0\right)$  has the following properties:
• (a) The restriction of $f$  to a map ${C}_{\left[-n,n\right]}^{\infty }\left(\mathbb{R}\right)\to {C}_{c}^{\infty }\left(\mathbb{R}\right)$  is smooth $\left($  and hence continuous $\right)$  , for each $n\in \mathbb{N}$  .
• (b) $f$  is discontinuous at $\gamma =0$  .
Proof. (a) Fix $n\in \mathbb{N}$  ; we have to show that $f{|}_{{C}_{\left[-n,n\right]}^{\infty }\left(\mathbb{R}\right)}:{C}_{\left[-n,n\right]}^{\infty }\left(\mathbb{R}\right)\to {C}_{c}^{\infty }\left(\mathbb{R}\right)$  is smooth.
The image of this map being contained in the closed vector subspace ${C}_{\left[-n,n\right]}^{\infty }\left(\mathbb{R}\right)$  of ${C}_{c}^{\infty }\left(\mathbb{R}\right)$  , which also is a closed vector subspace of ${C}^{\infty }\left(\mathbb{R}\right)$  (with the same induced topology), it suffices to show that the map ${C}_{\left[-n,n\right]}^{\infty }\left(\mathbb{R}\right)\to {C}^{\infty }\left(\mathbb{R}\right)$  , $\gamma ↦\gamma \circ \gamma -\gamma \left(0\right)$  is smooth (see [9,Prop.1.9or [1,La.10.1). Now $\gamma ↦\gamma \left(0\right)$  being a continuous linear (and thus smooth) map, it suffices to show that ${C}_{\left[-n,n\right]}^{\infty }\left(\mathbb{R}\right)\to {C}^{\infty }\left(\mathbb{R}\right)$  , $\gamma ↦\gamma \circ \gamma$  is smooth. This readily follows from Lemma  2.1 .
(b) Consider the zero-neighbourhood $V:=\mathcal{V}\left(\left(|n|{\right)}_{n\in \mathbb{Z}},\left(1{\right)}_{n\in \mathbb{Z}}\right)$  in ${C}_{c}^{\infty }\left(\mathbb{R}\right)$  . Let $k=\left({k}_{n}\right)\in \left({\mathbb{N}}_{0}{\right)}^{\mathbb{Z}}$  and $e=\left({\varepsilon }_{n}\right)\in \left({\mathbb{R}}^{+}{\right)}^{\mathbb{Z}}$  be arbitrary. We show that $f\left(\mathcal{V}\left(k,e\right)\right)⊈V$  . Since $f\left(0\right)=0$  , this entails that $f$  is discontinuous at $\gamma =0$  . It is easy to construct a function $h\in {C}_{c}^{\infty }\left(\mathbb{R}\right)$  such that $\text{supp}\left(h\right)\subseteq \right]-\frac{1}{2},\frac{1}{2}\left[$  and $h\left(x\right)={x}^{{k}_{0}+1}$  for all $x\in \left[-\frac{1}{4},\frac{1}{4}\right]$  . Then $rh\in \mathcal{V}\left(k,e\right)$  for some $r>0$  . For $m\in \mathbb{N}$  , we define ${h}_{m}\in {C}_{c}^{\infty }\left(\mathbb{R}\right)$  via ${h}_{m}\left(x\right):=\frac{r}{{m}^{{k}_{0}}}h\left(mx\right).$  Then $\text{supp}\left({h}_{m}\right)\subseteq \right]-\frac{1}{2m},\frac{1}{2m}\left[$  and thus ${h}_{m}\in \mathcal{V}\left(k,e\right)$  since, for all $j=0,\dots ,{k}_{0}$  and $x\in \left[-\frac{1}{2},\frac{1}{2}\right]$  , we have $|{h}_{m}^{\left(j\right)}\left(x\right)|=\frac{r{m}^{j}}{{m}^{{k}_{0}}}|{h}^{\left(j\right)}\left(mx\right)|<{\varepsilon }_{0}$  . We now choose $n\in \mathbb{N}$  such that $n\ge {k}_{0}+2$  . It is easy to construct a function $\psi \in {C}_{c}^{\infty }\left(\mathbb{R}\right)$  such that $\psi \left(x\right)=x-n$  for $x$  in some neighbourhood of $n$  in $\mathbb{R}$  , and $\text{supp}\left(\psi \right)\subseteq \right]n-\frac{1}{2},n+\frac{1}{2}\left[$  . Then $\phi :=s\cdot \psi \in \mathcal{V}\left(k,e\right)$  for suitable $s>0$  . Choosing $s$  small enough, we may assume that $\text{im}\left(\phi \right)\subseteq \left[-1,1\right]$  .
The supports of $\phi$  and ${h}_{m}$  being disjoint, we easily deduce from $\phi ,{h}_{m}\in \mathcal{V}\left(k,e\right)$  that also ${\gamma }_{m}:=\phi +{h}_{m}\in \mathcal{V}\left(k,e\right)$  . Then ${\gamma }_{m}\left(0\right)=0$  , and since $\text{im}\left(\phi \right)\subseteq \left[-1,1\right]$  , we have $f\left({\gamma }_{m}\right)\left(x\right)=\left({h}_{m}\circ \phi \right)\left(x\right)$  for all $x\in W:=\right]n-\frac{1}{2},n+\frac{1}{2}\left[$  . For $x\in W$  sufficiently close to $n$  , we have $\phi \left(x\right)=s\cdot \left(x-n\right)\in \left[-\frac{1}{4m},\frac{1}{4m}\right]$  and thus $f\left({\gamma }_{m}\right)\left(x\right)=r\cdot m\cdot {s}^{{k}_{0}+1}\cdot \left(x-n{\right)}^{{k}_{0}+1}$  , whence $f\left({\gamma }_{m}{\right)}^{\left({k}_{0}+1\right)}\left(n\right)=r\cdot m\cdot {s}^{{k}_{0}+1}\cdot \left({k}_{0}+1\right)!$  . Thus $f\left({\gamma }_{m}\right)\notin V$  for all $m\in \mathbb{N}$  such that $r\cdot m\cdot {s}^{{k}_{0}+1}\cdot \left({k}_{0}+1\right)!\ge 1$  , and so $f\left(\mathcal{V}\left(k,e\right)\right)⊈V$  . As $k$  and $e$  were arbitrary, (b) follows. $\square$
Note that $\text{supp}\left(f\left(\gamma \right)\right)\subseteq \text{supp}\left(\gamma \right)$  here, for all $\gamma \in {C}_{c}^{\infty }\left(\mathbb{R}\right)$  .
Remark 2.3 Although the map $f$  from Proposition  2.2 is discontinuous and thus not smooth in the Michal-Bastiani sense, it is easily seen to be smooth in the sense of convenient differential calculus (as any map $f$  on a “regular” countable strict direct limit $E=lim⟶{E}_{n}$  of complete locally convex spaces, all of whose restrictions $f{|}_{{E}_{n}}$  are smooth). $\text{1}$

3 Discontinuous mappings on ${\mathbit{C}}_{\mathbit{c}}^{\mathbsf{\infty }}\mathbf{\left(}\mathbit{M}\mathbf{,}\mathbit{E}\mathbf{\right)}$

In this section, we generalize our discussion of ${C}_{c}^{\infty }\left(\mathbb{R}\right)$  from Section  2 to the spaces ${C}_{c}^{\infty }\left(M,E\right)=lim⟶{C}_{K}^{\infty }\left(M,E\right)$  of compactly supported smooth mappings on a $\sigma$  -compact finite-dimensional smooth manifold $M$  with values in a locally convex space $E$  . We show:
Proposition 3.1 If $E\ne \left\{0\right\}$  , the manifold $M$  is non-compact, and $dim\left(M\right)>0$  , then there exists a mapping $f:{C}_{c}^{\infty }\left(M,E\right)\to {C}_{c}^{\infty }\left(M,\mathbb{R}\right)$  such that
• (a) The restriction of $f$  to ${C}_{K}^{\infty }\left(M,E\right)$  is smooth, for each compact subset $K$  of $M$  .
• (b) $f$  is discontinuous at $0$  .
In particular, the locally convex direct limit topology on ${C}_{c}^{\infty }\left(M,E\right)=lim⟶{C}_{K}^{\infty }\left(M,E\right)$  is properly coarser than the topology making ${C}_{c}^{\infty }\left(M,E\right)$  the direct limit of the spaces ${C}_{K}^{\infty }\left(M,E\right)$  in the category of topological spaces.
Instead of proving this proposition directly, we establish an analogous result for spaces of sections in bundles of locally convex spaces, which is no harder to prove. Noting that the function space ${C}_{c}^{\infty }\left(M,E\right)$  is topologically isomorphic to the space ${C}_{c}^{\infty }\left(M,M×E\right)$  of compactly supported smooth sections in the trivial bundle ${\text{pr}}_{M}:M×E\to M$  , clearly Proposition  3.1 is covered by the ensuing discussions for vector bundles. For background material concerning bundles of locally convex spaces and the associated spaces of sections, the reader is referred to [9(or also [8,Appendix F).
For the present purposes, we recall: if $\pi :E\to M$  is a smooth bundle of locally convex spaces over the finite-dimensional, $\sigma$  -compact smooth manifold $M$  , with typical fibre the locally convex space $F$  , then one considers on the space ${C}^{\infty }\left(M,E\right)$  of all smooth sections the initial topology with respect to the family of mappings ${\theta }_{\psi }:{C}^{\infty }\left(M,E\right)\to {C}^{\infty }\left(U,F\right),{\theta }_{\psi }\left(\sigma \right):={\sigma }_{\psi }:={\text{pr}}_{F}\circ \psi \circ \sigma {|}_{U}^{{\pi }^{-1}\left(U\right)},$  which take a smooth section $\sigma$  to its local representation ${\sigma }_{\psi }:U\to F$  with respect to the local trivialization $\psi :{\pi }^{-1}\left(U\right)\to U×F$  of $E$  . Given a compact subset $K\subseteq M$  , the subspace ${C}_{K}^{\infty }\left(M,E\right)\subseteq {C}^{\infty }\left(M,E\right)$  of sections vanishing off $K$  is equipped with the induced topology, and ${C}_{c}^{\infty }\left(M,E\right):={\cup }_{K}{C}_{K}^{\infty }\left(M,E\right)=lim⟶{C}_{K}^{\infty }\left(M,E\right)$  is given the locally convex direct limit topology.
Theorem 3.2 Let $M$  be a $\sigma$  -compact, non-compact, finite-dimensional smooth manifold of dimension $dim\left(M\right)>0$  , and $\pi :E\to M$  be a smooth bundle of locally convex spaces over $M$  , whose typical fibre is a locally convex topological vector space $F\ne \left\{0\right\}$  .
Then there exists a discontinuous mapping $f:{C}_{c}^{\infty }\left(M,E\right)\to {C}_{c}^{\infty }\left(M,\mathbb{R}\right)$  whose restriction to ${C}_{K}^{\infty }\left(M,E\right)$  is smooth, for each compact subset $K$  of $M$  .
Proof. Let $d:=dim\left(M\right)$  .
Since $M$  is non-compact, there exists a sequence $\left({U}_{n}{\right)}_{n\in {\mathbb{N}}_{0}}$  of mutually disjoint coordinate neighbourhoods ${U}_{n}\subseteq M$  diffeomorphic to ${\mathbb{R}}^{d}$  such that local trivializations ${\psi }_{n}:{\pi }^{-1}\left({U}_{n}\right)\to {U}_{n}×F$  of $E$  exist, and such that every compact subset of $M$  meets only finitely many of the sets ${U}_{n}$  . We define ${\theta }_{{\psi }_{n}}:{C}_{c}^{\infty }\left(M,E\right)\to {C}^{\infty }\left({U}_{n},F\right),{\theta }_{{\psi }_{n}}\left(\sigma \right):={\sigma }_{{\psi }_{n}}:={\text{pr}}_{F}\circ {\psi }_{n}\circ \sigma {|}_{{U}_{n}}^{{\pi }^{-1}\left({U}_{n}\right)}.$  By definition of the topology on ${C}_{c}^{\infty }\left(M,E\right)$  , the linear maps ${\theta }_{{\psi }_{n}}$  are continuous. For each $n\in {\mathbb{N}}_{0}$  , let ${\kappa }_{n}:{U}_{n}\to {\mathbb{R}}^{d}$  be a ${C}^{\infty }$  -diffeomorphism; define ${x}_{n}:={\kappa }_{n}^{-1}\left(0\right)$  . We choose a function $h\in {C}_{c}^{\infty }\left({\mathbb{R}}^{d},\mathbb{R}\right)$  such that $h{|}_{\left[-1,1{\right]}^{d}}=1$  ; we define ${h}_{n}\in {C}_{c}^{\infty }\left(M,\mathbb{R}\right)$  via ${h}_{n}\left(x\right):=h\left({\kappa }_{n}\left(x\right)\right)$  if $x\in {U}_{n}$  , ${h}_{n}\left(x\right):=0$  if $x\in M\{U}_{n}$  . Let ${K}_{n}:=\text{supp}\left({h}_{n}\right)\subseteq {U}_{n}$  . We choose a continuous linear functional $0\ne \lambda \in {F}^{\prime }$  , and pick $v\in F$  such that $\lambda \left(v\right)=1$  . Note that $A:={\cup }_{n\in \mathbb{N}}{K}_{n}$  is closed in $M$  , the sequence $\left({K}_{n}{\right)}_{n\in \mathbb{N}}$  of compact sets being locally finite. Let $\mu :\mathbb{R}×F\to F$  be the scalar multiplication. The eventual definition of the mapping $f$  we are looking for will involve the map $\Phi :E\to M×\mathbb{R}$  , defined via
 $\begin{array}{c}\Phi {|}_{{\pi }^{-1}\left({U}_{n}\right)}:=\left(\pi {|}_{{\pi }^{-1}\left({U}_{n}\right)},\lambda \circ \mu \circ \left(\left({h}_{n}\circ \pi \right){|}_{{\pi }^{-1}\left({U}_{n}\right)},{\text{pr}}_{F}\circ {\psi }_{n}\right)\right)\end{array}$ (1)
for $n\in \mathbb{N}$  , and $\Phi {|}_{E\{\pi }^{-1}\left(A\right)}:=\left(\pi {|}_{E\{\pi }^{-1}\left(A\right)},0\right)$  . Note that $\Phi$  is well-defined as the function in Equation ( 1 ) coincides with $\left(\pi ,0\right)$  on the set ${\cup }_{n\in \mathbb{N}}{\pi }^{-1}\left({U}_{n}\A\right)$  . Also note that $\Phi$  is a fibre-preserving mapping from $E$  into the trivial bundle $M×\mathbb{R}$  . Furthermore, it is readily verified that $\Phi$  is a smooth. By [9,Thm.5.9(or [8,Rem.F.25(a)), the pushforward ${C}_{c}^{\infty }\left(M,\Phi \right):{C}_{c}^{\infty }\left(M,E\right)\to {C}_{c}^{\infty }\left(M,M×\mathbb{R}\right),\sigma ↦\Phi \circ \sigma$  is smooth. For later use, we introduce the continuous linear map $\Lambda :={\theta }_{{\text{id}}_{M×\mathbb{R}}}:{C}_{c}^{\infty }\left(M,M×\mathbb{R}\right)\to {C}^{\infty }\left(M,\mathbb{R}\right).$  Let $\iota :\mathbb{R}\to {\mathbb{R}}^{d}$  denote the embedding $t↦\left(t,0,\dots ,0\right)$  . The mapping $f$  to be constructed will also involve the map $\Psi :{C}_{c}^{\infty }\left(M,E\right)\to {C}^{\infty }\left(\mathbb{R},\mathbb{R}\right)$  defined via $\Psi :={C}^{\infty }\left(\mathbb{R},\lambda \right)\circ {C}^{\infty }\left({\kappa }_{0}^{-1}\circ \iota ,F\right)\circ {\theta }_{{\psi }_{0}},$  where the pullback ${C}^{\infty }\left({\kappa }_{0}^{-1}\circ \iota ,F\right):{C}^{\infty }\left({U}_{n},F\right)\to {C}^{\infty }\left(\mathbb{R},F\right)$  , $\gamma ↦\gamma \circ {\kappa }_{0}^{-1}\circ \iota$  and the pushforward ${C}^{\infty }\left(\mathbb{R},\lambda \right):{C}^{\infty }\left(\mathbb{R},F\right)\to {C}^{\infty }\left(\mathbb{R},\mathbb{R}\right)$  , $\gamma ↦\lambda \circ \gamma$  are continuous linear mappings and thus smooth, by [5,La.3.3,La.3.7. Being a composition of smooth maps, $\Psi$  is smooth.
We now define the desired map $f:{C}_{c}^{\infty }\left(M,E\right)\to {C}_{c}^{\infty }\left(M,\mathbb{R}\right)$  via $f:=\Gamma \circ \left(\Psi ,\Lambda \circ {C}_{c}^{\infty }\left(M,\Phi \right)\right)-\lambda \circ {\text{ev}}_{{x}_{0}}\circ {\theta }_{{\psi }_{0}}$  (co-restricted from ${C}^{\infty }\left(M,\mathbb{R}\right)$  to ${C}_{c}^{\infty }\left(M,\mathbb{R}\right)$  ), where $\Gamma :{C}^{\infty }\left(\mathbb{R},\mathbb{R}\right)×{C}^{\infty }\left(M,\mathbb{R}\right)\to {C}^{\infty }\left(M,\mathbb{R}\right),\Gamma \left(\gamma ,\eta \right):=\gamma \circ \eta$  denotes composition, and ${\text{ev}}_{{x}_{0}}:{C}^{\infty }\left({U}_{0},F\right)\to F$  the evaluation map $\gamma ↦\gamma \left({x}_{0}\right)$  . Here $\lambda \circ {\text{ev}}_{{x}_{0}}\circ {\theta }_{{\psi }_{0}}$  is a continuous linear map and thus smooth. Explicitly, for $\sigma \in {C}_{c}^{\infty }\left(M,E\right)$
 $\begin{array}{ccc}f\left(\sigma \right)\left(x\right)& =& \left(\lambda \circ {\sigma }_{{\psi }_{0}}\circ {\kappa }_{0}^{-1}\circ \iota \right)\left(\lambda \left({h}_{n}\left(x\right){\sigma }_{{\psi }_{n}}\left(x\right)\right)\right)\end{array}$
 $\begin{array}{ccc}& =& \lambda \left({\sigma }_{{\psi }_{0}}\left({\kappa }_{0}^{-1}\left({h}_{n}\left(x\right)\cdot \lambda \left({\sigma }_{{\psi }_{n}}\left(x\right)\right),0\right)\right)\right)-\lambda \left({\sigma }_{{\psi }_{0}}\left({x}_{0}\right)\right)\end{array}$
if $x\in {U}_{n}$  ( $n\in \mathbb{N}$  ), whereas $f\left(\sigma \right)\left(x\right)=0$  if $x\in M\A$  .
Claim : The restriction of $f$  to ${C}_{K}^{\infty }\left(M,E\right)$  is smooth, for each compact subset $K$  of $M$  .
To see this, note that $f\left({C}_{K}^{\infty }\left(M,E\right)\subseteq {C}_{K}^{\infty }\left(M,\mathbb{R}\right)$  , where ${C}_{K}^{\infty }\left(M,\mathbb{R}\right)$  is a closed vector subspace of ${C}^{\infty }\left(M,\mathbb{R}\right)$  and ${C}_{c}^{\infty }\left(M,\mathbb{R}\right)$  . Thus, it suffices to show that $f{|}_{{C}_{K}^{\infty }\left(M,E\right)}$  is smooth as a map into ${C}^{\infty }\left(M,\mathbb{R}\right)$  ([9,Prop.1.9, or [1,La.10.1). But this follows from the Chain Rule, as $\Gamma$  is smooth by Lemma  2.1 and also the other constituents of $f$  are smooth.
Claim : $f$  is discontinuous at the zero-section $\sigma =0$  . To see this, consider the set $V$  of all $\gamma \in {C}_{c}^{\infty }\left(M,\mathbb{R}\right)$  such that, for all $n\in \mathbb{N}$  and multi-indices $\alpha \in {\mathbb{N}}_{0}^{d}$  of order $|\alpha |\le n$  , we have $|{\partial }^{\alpha }\left(\gamma \circ {\kappa }_{n}^{-1}\right)\left(0\right)|<1$  . It is easily verified that $V$  is a symmetric, convex zero-neighbourhood in ${C}_{c}^{\infty }\left(M,\mathbb{R}\right)$  . Let $U$  be any convex zero-neighbourhood in ${C}_{c}^{\infty }\left(M,E\right)$  ; we claim that $f\left(U\right)⊈V$  . To see this, set ${L}_{n}:={\kappa }_{n}^{-1}\left(\left[-1,1{\right]}^{d}\right)$  for $n\in {\mathbb{N}}_{0}$  . Then ${\rho }_{n}:{C}_{{L}_{n}}^{\infty }\left(M,E\right)\to {C}_{\left[-1,1{\right]}^{d}}^{\infty }\left({\mathbb{R}}^{d},F\right),\sigma ↦{\sigma }_{{\psi }_{n}}\circ {\kappa }_{n}^{-1}$  is a topological isomorphism (cf. [9,La.3.9,La.3.10or [8,La.F.9,La.F.15) whose inverse gives rise to a topological embedding ${j}_{n}:{C}_{\left[-1,1{\right]}^{d}}^{\infty }\left({\mathbb{R}}^{d},F\right)\to {C}_{c}^{\infty }\left(M,E\right)$  . The linear mapping $\phi :\mathbb{R}\to F$  , $t↦tv$  gives rise to a continuous linear map ${C}_{\left[-1,1{\right]}^{d}}^{\infty }\left({\mathbb{R}}^{d},\phi \right):{C}_{\left[-1,1{\right]}^{d}}^{\infty }\left({\mathbb{R}}^{d},\mathbb{R}\right)\to {C}_{\left[-1,1{\right]}^{d}}^{\infty }\left({\mathbb{R}}^{d},F\right)$  , $\gamma ↦\phi \circ \gamma$  . Then ${W}_{n}:=\left({j}_{n}\circ {C}_{\left[-1,1{\right]}^{d}}^{\infty }\left({\mathbb{R}}^{d},\phi \right){\right)}^{-1}\left(\frac{1}{2}U\right)$  is a convex zero-neighbourhood in ${C}_{\left[-1,1{\right]}^{d}}^{\infty }\left({\mathbb{R}}^{d},\mathbb{R}\right)$  . Thus, there exists ${k}_{n}\in {\mathbb{N}}_{0}$  and ${\varepsilon }_{n}>0$  such that ${W}_{{k}_{n},{\varepsilon }_{n}}\subseteq {W}_{n}$  , where ${W}_{{k}_{n},{\varepsilon }_{n}}$  is the set of all $\gamma \in {C}_{\left[-1,1{\right]}^{d}}^{\infty }\left({\mathbb{R}}^{d},\mathbb{R}\right)$  such that $sup\left\{|{\partial }^{\alpha }\gamma \left(x\right)|:x\in \left[-1,1{\right]}^{d}\right\}<{\varepsilon }_{n}$  for all $\alpha \in {\mathbb{N}}_{0}^{d}$  such that $|\alpha |\le {k}_{n}$  . We let $g\in {C}_{\left[-1,1{\right]}^{d}}^{\infty }\left({\mathbb{R}}^{d},\mathbb{R}\right)$  be a function such that $g\left({y}_{1},\dots ,{y}_{d}\right)={y}_{1}^{{k}_{0}+1}$  for all $y=\left({y}_{1},\dots ,{y}_{d}\right)\in \left[-\frac{1}{2},\frac{1}{2}{\right]}^{d}$  . Then $rg\in {W}_{{k}_{0},{\varepsilon }_{0}}$  for some $r>0$  . It is clear from the definition of ${W}_{{k}_{0},{\varepsilon }_{0}}$  that then also ${\gamma }_{m}\in {W}_{{k}_{0},{\varepsilon }_{0}}$  for all $m\in \mathbb{N}$  , where ${\gamma }_{m}:{\mathbb{R}}^{d}\to \mathbb{R},{\gamma }_{m}\left({y}_{1},\dots ,{y}_{d}\right):=\frac{r}{{m}^{{k}_{0}}}g\left(m{y}_{1},{y}_{2},\dots ,{y}_{d}\right).$  Thus ${\tau }_{m}:={j}_{0}\left(\phi \circ {\gamma }_{m}\right)\in \frac{1}{2}U$  .
Let $\ell :={k}_{0}+1$  ; we easily find $\eta \in {W}_{{k}_{\ell },{\varepsilon }_{\ell }}$  such that, for suitable $s>0$  , we have $\eta \left(y\right)=s\cdot {y}_{1}$  for $y=\left({y}_{1},\dots ,{y}_{d}\right)$  in some zero-neighbourhood in ${\mathbb{R}}^{d}$  . We define $\tau :={j}_{\ell }\left(\phi \circ \eta \right)\in \frac{1}{2}U$  .
Then ${\sigma }_{m}:={\tau }_{m}+\tau \in U$  by convexity of $U$  . Consider ${g}_{m}:=f\left({\sigma }_{m}\right)\circ {\kappa }_{\ell }^{-1}:{\mathbb{R}}^{d}\to \mathbb{R}$  . For $y\in \left[-1,1{\right]}^{d}$  sufficiently close to $0$  , we have $\eta \left(y\right)=s{y}_{1}$  and $m|\eta \left(y\right)|\le \frac{1}{2}$  . Thus ${g}_{m}\left(y\right)={\gamma }_{m}\left(\eta \left(y\right),0,\dots ,0\right)=r\cdot m\cdot {s}^{{k}_{0}+1}\cdot {y}_{1}^{{k}_{0}+1},$  entailing that $\frac{{\partial }^{{k}_{0}+1}{g}_{m}}{\partial {y}_{1}^{{k}_{0}+1}}\left(0\right)=r\cdot m\cdot {s}^{{k}_{0}+1}\cdot \left({k}_{0}+1\right)!$  . Hence $f\left({\sigma }_{m}\right)\notin V$  for each $m\in \mathbb{N}$  such that $r\cdot m\cdot {s}^{{k}_{0}+1}\cdot \left({k}_{0}+1\right)!\ge 1$  . We have shown that $f\left(U\right)⊈V$  for any $0$  -neighbourhood $U$  in ${C}_{c}^{\infty }\left(M,E\right)$  , although $f\left(0\right)=0$  . Thus $f$  is discontinuous at $\sigma =0$  . $\square$

4 Further examples

We describe various pathological bilinear mappings.
Proposition 4.1 Let $\mathbb{K}\in \left\{\mathbb{R},\mathbb{C}\right\}$  .
The pointwise multiplication map $\mu :{C}^{\infty }\left(\mathbb{R},\mathbb{K}\right)×{C}_{c}^{\infty }\left(\mathbb{R},\mathbb{K}\right)\to {C}_{c}^{\infty }\left(\mathbb{R},\mathbb{K}\right),\mu \left(\gamma ,\eta \right):=\gamma \cdot \eta$  is a hypocontinuous bilinear $\left($  and thus sequentially continuous $\right)$  mapping on the locally convex direct limit ${C}^{\infty }\left(\mathbb{R},\mathbb{K}\right)×{C}_{c}^{\infty }\left(\mathbb{R},\mathbb{K}\right)=lim⟶\left({C}^{\infty }\left(\mathbb{R},\mathbb{K}\right)×{C}_{\left[-n,n\right]}^{\infty }\left(\mathbb{R},\mathbb{K}\right)\right),$  whose restriction to ${C}^{\infty }\left(\mathbb{R},\mathbb{K}\right)×{C}_{\left[-n,n\right]}^{\infty }\left(\mathbb{R},\mathbb{K}\right)$  is continuous bilinear and thus $\mathbb{K}$  -analytic, for each $n\in \mathbb{N}$  . However, $\mu$  is discontinuous.
Proof. Using the Leibniz Rule for the differentiation of products of functions, it is easily verified that $\mu$  is separately continuous. $\text{2}$  The spaces ${C}^{\infty }\left(\mathbb{R},\mathbb{K}\right)$  and ${C}_{c}^{\infty }\left(\mathbb{R},\mathbb{K}\right)$  being barrelled, this entails that $\mu$  is hypocontinuous and thus sequentially continuous [19,Thm.41.2. The restriction of $\mu$  to ${C}^{\infty }\left(\mathbb{R},\mathbb{K}\right)×{C}_{\left[-n,n\right]}^{\infty }\left(\mathbb{R},\mathbb{K}\right)$  is a sequentially continuous bilinear mapping on a product of metrizable spaces and therefore continuous. To see that $\mu$  is discontinuous, consider the zero-neighbourhood $W:=\left\{\gamma \in {C}_{c}^{\infty }\left(\mathbb{R},\mathbb{K}\right):\left(\forall x\in \mathbb{R}\right)|\gamma \left(x\right)|<1\right\}$  in ${C}_{c}^{\infty }\left(\mathbb{R},\mathbb{K}\right)$  . If $U$  is any zero-neighbourhood in ${C}^{\infty }\left(\mathbb{R},\mathbb{K}\right)$  and $V$  any zero-neighbourhood in ${C}_{c}^{\infty }\left(\mathbb{R},\mathbb{K}\right)$  , then there exists a compact subset $K$  of $\mathbb{R}$  such that $\left(\forall \gamma \in {C}^{\infty }\left(\mathbb{R},\mathbb{K}\right)\right)\gamma {|}_{K}=0⇒\gamma \in U.$  Pick any ${x}_{0}\in \mathbb{R}\K$  . There is a function $\phi \in {C}_{c}^{\infty }\left(\mathbb{R},\mathbb{K}\right)$  such that $\phi \left({x}_{0}\right)\ne 0$  and $\text{supp}\left(\phi \right)\subseteq \mathbb{R}\K$  . Then $r\phi \in V$  for some $r>0$  , and $t\phi \in U$  for all $t\in \mathbb{R}$  . Choosing $t\ge \frac{1}{r\cdot |\phi \left({x}_{0}\right){|}^{2}}$  , we have $\left(t\phi ,r\phi \right)\in U×V$  but $|\mu \left(r\phi ,t\phi \right)\left({x}_{0}\right)|=rt|\phi \left({x}_{0}\right){|}^{2}\ge 1$  , entailing that $\mu \left(U×V\right)⊈W$  . Thus $\mu$  is discontinuous at $\left(0,0\right)$  . $\square$
Another instructive example is the following (compare also the examples in [2):
Example 4.2 Let ${E}_{1}\subset {E}_{2}\subset \cdots$  be a strictly ascending sequence of Banach spaces, such that ${E}_{n+1}$  induces the given topology on ${E}_{n}$  . Set $E:=lim⟶{E}_{n}$  and $F:={E}_{b}^{\prime }$  . For example, we can take ${E}_{n}:={L}^{2}\left[-n,n\right]$  , in which case $E={L}_{\text{comp}}^{2}\left(\mathbb{R}\right)$  and $F={L}_{\text{loc}}^{2}\left(\mathbb{R}\right)=lim⟵{L}^{2}\left[-n,n\right]$  . Then ${A}_{n}:=F×{E}_{n}×\mathbb{K}×\mathbb{K}$  is a Fréchet space (and reflexive in the example ${E}_{n}={L}^{2}\left[-n,n\right]$  ).
The evaluation map ${E}_{n}^{\prime }×{E}_{n}\to \mathbb{R}$  being continuous as ${E}_{n}$  is a Banach space, it is easy to see that ${A}_{n}$  becomes a unital associative topological algebra via
 $\begin{array}{c}\left({\lambda }_{1},{x}_{1},{z}_{1},{c}_{1}\right)\cdot \left({\lambda }_{2},{x}_{2},{z}_{2},{c}_{2}\right):=\left({c}_{1}{\lambda }_{2}+{c}_{2}{\lambda }_{1},{c}_{1}{x}_{2}+{c}_{2}{x}_{1},{c}_{1}{z}_{2}+{\lambda }_{1}\left({x}_{2}\right)+{z}_{1}{c}_{2},{c}_{1}{c}_{2}\right).\end{array}$ (2)
The multiplication can be visualized by considering $\left(\lambda ,x,z,c\right)\in {A}_{n}$  as the 3-by-3 matrix $\left(\begin{array}{ccc}c& \lambda & z\\ 0& c& x\\ 0& 0& c\end{array}\right).$  The topological algebras ${A}_{n}$  are very well-behaved: they have open groups of units, and inversion is a $\mathbb{K}$  -analytic map. We can also use Formula ( 2 ) to define a multiplication map $\mu :A×A\to A$  turning the direct limit locally convex space $A:=F×E×\mathbb{K}×\mathbb{K}=lim⟶{A}_{n}$  into a unital, associative algebra. However, although the restriction of $\mu$  to ${A}_{n}×{A}_{n}$  is a continuous bilinear map for each $n\in \mathbb{N}$  , $\mu :A×A=lim⟶\left({A}_{n}×{A}_{n}\right)\to A$  is discontinuous (since the evaluation map ${E}_{b}^{\prime }×E\to \mathbb{R}$  is discontinuous, the space $E$  not being normable).
We refer to [6,Section 10for more details.
Appendix: Proof of Lemma  2.1 We give a proof which is as elementary as possible, by reducing the assertion to the case $M={\mathbb{R}}^{d}$  . First, let $M$  be a finite-dimensional, $\sigma$  -compact smooth manifold, of dimension $d$  .
We choose an open cover $\left({U}_{j}{\right)}_{j\in J}$  of $M$  and ${C}^{\infty }$  -diffeomorphisms ${\kappa }_{j}:{U}_{j}\to {\mathbb{R}}^{d}$  . Then $\Phi :{C}^{\infty }\left(M,{\mathbb{R}}^{m}\right)\to {\prod }_{j\in J}{C}^{\infty }\left({\mathbb{R}}^{d},{\mathbb{R}}^{m}\right)=:P,\Phi \left(\gamma \right):=\left(\gamma \circ {\kappa }_{j}^{-1}{\right)}_{j\in J}$  is a topological embedding onto a closed vector subspace of the cartesian product $P$  (cf. [9,La.3.7). Therefore $\Gamma$  is smooth if and only if $\Phi \circ \Gamma$  is smooth ([9,Prop.1.9or [1,La.10.1), if and only if each component ${\text{pr}}_{j}\circ \Phi \circ \Gamma$  is smooth [1,La.10.3, where ${\text{pr}}_{j}:P\to {C}^{\infty }\left({\mathbb{R}}^{d},{\mathbb{R}}^{m}\right)$  is the projection onto the $j$  -coordinate. But ${\text{pr}}_{j}\left(\Phi \left(\Gamma \right)\right)\left(\gamma ,\eta \right)=\gamma \circ \eta \circ {\kappa }_{j}^{-1}=\stackrel{~}{\Gamma }\left(\gamma ,{C}^{\infty }\left({\kappa }_{j}^{-1},{\mathbb{R}}^{n}\right)\left(\eta \right)\right)$  for all $\gamma \in {C}^{\infty }\left({\mathbb{R}}^{n},{\mathbb{R}}^{m}\right)$  and $\eta \in {C}^{\infty }\left(M,{\mathbb{R}}^{n}\right)$  , where $\stackrel{~}{\Gamma }:{C}^{\infty }\left({\mathbb{R}}^{n},{\mathbb{R}}^{m}\right)×{C}^{\infty }\left({\mathbb{R}}^{d},{\mathbb{R}}^{n}\right)\to {C}^{\infty }\left({\mathbb{R}}^{d},{\mathbb{R}}^{m}\right)$  is the composition map and ${C}^{\infty }\left({\kappa }_{j}^{-1},{\mathbb{R}}^{n}\right):{C}^{\infty }\left(M,{\mathbb{R}}^{n}\right)\to {C}^{\infty }\left({\mathbb{R}}^{d},{\mathbb{R}}^{n}\right)$  , $\eta ↦\eta \circ {\kappa }_{j}^{-1}$  is continuous linear and thus smooth, by [5,La.3.7. Hence ${\text{pr}}_{j}\circ \Phi \circ \Gamma$  (and thus $\Gamma$  ) will be smooth if so is $\stackrel{~}{\Gamma }$  .
By the reduction step just performed, it only remains to prove Lemma  2.1 for $M={\mathbb{R}}^{d}$  , which we assume now. We show by induction on $k\in {\mathbb{N}}_{0}$  that $\Gamma$  is ${C}^{k}$  .
The case $k=0$  . Let $\gamma \in {C}^{\infty }\left({\mathbb{R}}^{n},{\mathbb{R}}^{m}\right)$  , $\eta \in {C}^{\infty }\left({\mathbb{R}}^{d},{\mathbb{R}}^{n}\right)$  and $\left({\gamma }_{i},{\eta }_{i}{\right)}_{i\in \mathbb{N}}$  be a sequence in ${C}^{\infty }\left({\mathbb{R}}^{n},{\mathbb{R}}^{m}\right)×{C}^{\infty }\left({\mathbb{R}}^{d},{\mathbb{R}}^{n}\right)$  converging to $\left(\gamma ,\eta \right)$  . We have to show that
 $\begin{array}{ccc}{\delta }_{i}& :=& \Gamma \left({\gamma }_{i},{\eta }_{i}\right)-\Gamma \left(\gamma ,\eta \right)={\gamma }_{i}\circ {\eta }_{i}-\gamma \circ \eta \end{array}$
 $\begin{array}{ccc}& =& \left({\gamma }_{i}-\gamma \right)\circ {\eta }_{i}+\left(\gamma \circ {\eta }_{i}-\gamma \circ \eta \right)\end{array}$ (3)
converges to $0$  in ${C}^{\infty }\left({\mathbb{R}}^{d},{\mathbb{R}}^{m}\right)$  . To see this, we first check convergence in ${C}^{0}\left({\mathbb{R}}^{d},{\mathbb{R}}^{m}\right)$  (equipped with the topology of uniform convergence on compact sets). Given a compact set $K\subseteq {\mathbb{R}}^{d}$  , the set ${\cup }_{i\in \mathbb{N}}{\eta }_{i}\left(K\right)$  is bounded and hence has compact closure $L$  in ${\mathbb{R}}^{d}$  . Now the first term in ( 3 ) converges uniformly to $0$  on $K$  since ${\gamma }_{i}-\gamma \to 0$  uniformly on $L$  as $i\to \infty$  . The second term converges uniformly to $0$  on $K$  since $\gamma {|}_{L}$  is uniformly continuous and ${\eta }_{i}\to \eta$  uniformly on $K$  . Using the Chain Rule, for each fixed multi-index $\alpha \in {\mathbb{N}}_{0}^{d}$  of order $\ge 1$  , we find polynomials ${P}_{\beta }\in \mathbb{R}\left[\left({X}_{\gamma }{\right)}_{\gamma \le \alpha }\right]$  in indeterminates ${X}_{\gamma }$  , for multi-indices $\beta \in {\mathbb{N}}_{0}^{n}$  of order $|\beta |\le |\alpha |$  , such that
 $\begin{array}{ccc}{\partial }^{\alpha }{\delta }_{i}& =& {\sum }_{|\beta |\le |\alpha |}\left(\left({\partial }^{\beta }{\gamma }_{i}-{\partial }^{\beta }\gamma \right)\circ {\eta }_{i}\right)\cdot {P}_{\beta }\left(\left({\partial }^{\gamma }{\eta }_{i}{\right)}_{\gamma \le \alpha }\right)\end{array}$
 $\begin{array}{ccc}& & +{\sum }_{|\beta |\le |\alpha |}\left({\partial }^{\beta }\gamma \circ {\eta }_{i}\right)\cdot \left({P}_{\beta }\left(\left({\partial }^{\gamma }{\eta }_{i}{\right)}_{\gamma \le \alpha }\right)-{P}_{\beta }\left(\left({\partial }^{\gamma }\eta {\right)}_{\gamma \le \alpha }\right)\right)\end{array}$
 $\begin{array}{ccc}& & +{\sum }_{|\beta |\le |\alpha |}\left({\partial }^{\beta }\gamma \circ {\eta }_{i}-{\partial }^{\beta }\gamma \circ \eta \right)\cdot {P}_{\beta }\left(\left({\partial }^{\gamma }\eta {\right)}_{\gamma \le \alpha }\right).\end{array}$
We easily deduce from this formula that ${\partial }^{\alpha }{\delta }_{i}$  converges to $0$  as $i\to \infty$  , uniformly on compact sets. We have shown that ${\delta }_{i}\to 0$  in ${C}^{\infty }\left({\mathbb{R}}^{d},{\mathbb{R}}^{m}\right)$  . Thus $\Gamma$  is continuous.
Induction step. Suppose that $\Gamma$  is of class ${C}^{k}$  , where $k\in {\mathbb{N}}_{0}$  . Given $\gamma ,{\gamma }_{1}\in {C}^{\infty }\left({\mathbb{R}}^{n},{\mathbb{R}}^{m}\right)$  , $\eta ,{\eta }_{1}\in {C}^{\infty }\left({\mathbb{R}}^{d},{\mathbb{R}}^{n}\right)$  , we have
 $\begin{array}{c}\frac{1}{t}\left(\Gamma \left(\gamma +t{\gamma }_{1},\eta +t{\eta }_{1}\right)-\Gamma \left(\gamma ,\eta \right)\right)=\frac{1}{t}\left(\gamma \circ \left(\eta +t{\eta }_{1}\right)-\gamma \circ \eta \right)+{\gamma }_{1}\circ \left(\eta +t{\eta }_{1}\right)\end{array}$ (4)
for $0\ne t\in \mathbb{R}$  . Given $t\in \mathbb{R}$  , define ${F}_{t}:{\mathbb{R}}^{d}\to {\mathbb{R}}^{m}$  via ${F}_{t}\left(x\right):={\int }_{0}^{1}H\left(x,st\right)ds,$  where $H:{\mathbb{R}}^{d}×\mathbb{R}\to {\mathbb{R}}^{m}$  , $H\left(x,r\right):=d\gamma \left(\eta \left(x\right)+r{\eta }_{1}\left(x\right);{\eta }_{1}\left(x\right)\right)$  . Clearly $H$  is smooth. It is easy to see that ${F}_{t}\left(x\right)\to {F}_{0}\left(x\right)$  uniformly for $x$  in a compact set, as $t\to 0$  . Furthermore, differentiating under the integral sign we find that ${\partial }^{\alpha }{F}_{t}\left(x\right)={\int }_{0}^{1}{\partial }^{\left(\alpha ,0\right)}H\left(x,st\right)ds$  for $\alpha \in {\mathbb{N}}_{0}^{d}$  , which converges uniformly for $x$  in a compact set to ${\partial }^{\alpha }{F}_{0}\left(x\right)$  as $t\to 0$  . Since ${F}_{t}=\frac{1}{t}\left(\gamma \circ \left(\eta +t{\eta }_{1}\right)-\gamma \circ \eta \right)$  for $t\ne 0$  , by the Mean Value Theorem, we see that the first term on the right hand side of ( 4 ) converges to ${F}_{0}=\left(d\gamma \right)\circ \left(\eta ,{\eta }_{1}\right)=\stackrel{~}{\Gamma }\left(d\gamma ,\left(\eta ,{\eta }_{1}\right)\right)$  in ${C}^{\infty }\left({\mathbb{R}}^{d},{\mathbb{R}}^{m}\right)$  as $t\to 0$  , where $\stackrel{~}{\Gamma }:{C}^{\infty }\left({\mathbb{R}}^{n}×{\mathbb{R}}^{n},{\mathbb{R}}^{m}\right)×{C}^{\infty }\left({\mathbb{R}}^{d},{\mathbb{R}}^{n}×{\mathbb{R}}^{n}\right)\to {C}^{\infty }\left({\mathbb{R}}^{d},{\mathbb{R}}^{m}\right)$  is the composition map.
To tackle the second term, define ${G}_{t}:={\gamma }_{1}\circ \left(\eta +t{\eta }_{1}\right)=\Gamma \left({\gamma }_{1},\eta +t{\eta }_{1}\right)$  for $t\in \mathbb{R}$  . Since $\Gamma$  is continuous by the above, we have ${G}_{t}\to {G}_{0}={\gamma }_{1}\circ \eta$  in ${C}^{\infty }\left({\mathbb{R}}^{d},{\mathbb{R}}^{m}\right)$  as $t\to 0$  . Thus the second term in Equation ( 4 ) converges to ${\gamma }_{1}\circ \eta$  .
Summing up, we have shown that $d\Gamma \left(\gamma ,\eta ;{\gamma }_{1},{\eta }_{1}\right)$  exists, and is given by
 $\begin{array}{c}d\Gamma \left(\gamma ,\eta ;{\gamma }_{1},{\eta }_{1}\right)=\stackrel{~}{\Gamma }\left(d\gamma ,\left(\eta ,{\eta }_{1}\right)\right)+\Gamma \left({\gamma }_{1},\eta \right).\end{array}$ (5)
The map ${C}^{\infty }\left({\mathbb{R}}^{n},{\mathbb{R}}^{m}\right)\to {C}^{\infty }\left({\mathbb{R}}^{n}×{\mathbb{R}}^{n},{\mathbb{R}}^{m}\right)$  , $\gamma ↦d\gamma$  is continuous linear (cf. [5,La.3.8), and $\Gamma$  , $\stackrel{~}{\Gamma }$  are ${C}^{k}$  , by induction. Hence Equation ( 5 ) shows that $d\Gamma$  is ${C}^{k}$  . Thus $\Gamma$  is ${C}^{k+1}$  , as required.
Acknowledgement. The research was partially supported by DFG, FOR 363/1-1. References

1. Bertram, W., H. Glöckner and K.-H. Neeb, Differential calculus over general base fields and rings, Expo. Math. 22 (2004), 213–282.
2. Dierolf, S. and J. Wengenroth, Inductive limits of topological algebras, Linear Topol. Spaces Complex Anal. 3 (1997), 45–49.
3. Dudley, R. M., On sequential convergence, Trans. AMS 112 (1964), 483–507.
4. Glöckner, H., Infinite-dimensional Lie groups without completeness restrictions, pp. 43–59 in: Strasburger, A. et al. (Eds.), “Geometry and Analysis on Finiteand Infinite-Dimensional Lie Groups,” Banach Center Publications, Vol. 55, Warsaw, 2002.
5. —–, Lie group structures on quotient groups and universal complexifications for infinite-dimensional Lie groups, J. Funct. Analysis 194 (2002), 347–409.
6. —–, Algebras whose groups of units are Lie groups, Studia Math. 153 (2002), 147–177.
7. —–, $\text{Diff}\left({\mathbb{R}}^{n}\right)$  as a Milnor-Lie group, 11 pp, to appear in Math. Nachr. (cf. TU Darmstadt Preprint 2387, March 2005).
8. —–, Lie groups over non-discrete topological fields, preprint, arXiv:math.GR/0408008.
9. —–, Differentiable mappings between spaces of sections, submitted.
10. —–, Patched locally convex spaces, almost local mappings, and diffeomorphism groups of non-compact manifilds, manuscript in preparation.
11. —–, Direct limits of Lie groups compared to direct limits in related categories, manuscript in preparation.
12. Hamilton, R., The inverse function theorem of Nash and Moser, Bull. AMS 7 (1982), 65–222.
13. Kriegl, A. and P. W. Michor, “The Convenient Setting of Global Analysis,” AMS, 1997.
14. Michor, P. W., “Manifolds of Differentiable Mappings,” Shiva Publishing, 1980.
15. Milnor, J., On infinite-dimensional Lie groups, preprint, Institute for Advanced Studies, Princeton, 1982.
16. —–, Remarks on infinite-dimensional Lie groups, pp. 1008–1057 in: DeWitt, B., and R. Stora (Eds.), “Relativity, Groups and Topology II,” North-Holland, 1983.
17. Schwartz, L., “Théorie des Distributions, Tome I,” Hermann, Paris, 1957.
18. Tatsuuma, N., H. Shimomura and T. Hirai, On group topologies and unitary representations of inductive limits of topological groups and the case of the group of diffeomorphisms, Kyoto J. Math. 38 (1998), 551–578.
19. Treves, F., “Topological Vector Spaces, Distributions and Kernels,” Academic Press, 1967.