## Group invariants in algebras of generalized functions

### Sanja Konjik * * Faculty of Agriculture, University of Novi Sad, Trg Dositeja Obradovića 8, 21000 Novi Sad, SCG. Electronic mail: kinjoki@neobee.net Michael Kunzinger † † Faculty of Mathematics, University of Vienna, Nordbergstr. 15, A-1090 Wien, Austria, Electronic mail: michael.kunzinger@univie.ac.at

Abstract
We study invariance properties of Colombeau generalized functions under actions of smooth Lie transformation groups. Several characterization results analogous to the smooth setting are derived and applications to generalized rotational invariance are given.
Mathematics Subject Classification (2000): Primary: 46F30; secondary: 58E40,46T30, 35A30 Keywords: Lie group actions, Colombeau generalized functions, invariants

1 Introduction

Extensions of classical Lie group analysis ([17, 18, 4) to spaces of generalized functions, first in the distributional setting and later in nonlinear theories of generalized functions have been developed by various authors, starting as early as the 1950ies ([13, 22, 23, 2, 3, 16, 20, 21, 9, 5, 14, 15). The main focus of the extension to Colombeau algebras of generalized functions so far has been the study of symmetry groups of differential equations involving singularities. In [14, 15(see also [6, ch. 4), however, M. Oberguggenberger initiated the study of group invariant Colombeau generalized functions. He studied invariance under translations and rotations and gave some applications to the determination of group invariant fundamental solutions. Such questions have been taken up in [19, 7and will also be one of the main themes of this contribution. Our aim is to analyze the action of smooth Lie transformation groups on elements of Colombeau algebras and to derive characterizations of invariants under such transformations which parallel the analogous constructions in the smooth setting.
The theory underlying our investigations is global analysis in Colombeau algebras, as presented in [6(see [8for a recent survey). In particular, our study of one-parameter transformation groups is based on the theory of flows of singular vector fields introduced in [10.
Our plan of exposition is as follows. In section  2 we present some fundamental results on generalized flows of distributional and Colombeau vector fields and compare the two approaches.
Finally, section  3 addresses the main topic of this article, namely the analysis of group invariance under smooth Lie group actions in the Colombeau setting.
To conclude this introduction we fix some notation and terminology to be employed throughout the paper. By $X$  we denote a smooth, connected, paracompact Hausdorff manifold of dimension $n$  .
For notations from Colombeau's theory of generalized functions we follow [6. Similarly, concerning terminology from symmetry group analysis our standard references are [17for the classical theory and again [6for the Colombeau setting.

2 Generalized flows

Lie group actions on differentiable manifolds are determined by their one-parameter subgroups which in turn are given as the flows of the corresponding infinitesimal generators. Therefore, in order to analyze generalized group actions we need a theory of ordinary differential equations with distributional or Colombeau generalized right hand side. These problems have been addressed, e.g., in [12for the distributional and in [10for the Colombeau case.
We begin with a purely distributional approach, as provided by Marsden [12, and consider the initial value problem:
 $\begin{array}{c}\begin{array}{ccc}\stackrel{˙}{x}\left(t\right)& =& \zeta \left(x\left(t\right)\right)\\ x\left({t}_{0}\right)& =& {x}_{0},\end{array}\end{array}$ (1)
where $\zeta \in {\mathcal{D}}^{\prime }\left(X,TX\right)$  is a distributional vector field. We immediately note that, in the linear theory of distributions, it is difficult to make sense of the above problem: it describes a prospective solution $x$  which ought to be distributional, yet take values in a manifold. Moreover, the right hand side of the equation involves the composition of two distributional quantities. To circumvent these conceptual problems, Marsden's approach is to first approximate $\zeta$  by a sequence of smooth vector fields ${\xi }_{\varepsilon }$  . The classical theory of ODEs then yields a flow ${\Phi }^{\varepsilon }$  of each ${\xi }_{\varepsilon }$  . Then, under certain assumptions on $\zeta$  and the regularizing sequence ${\xi }_{\varepsilon }$  , the limiting measurable function $\Psi ={lim}_{\varepsilon \to 0}{\Phi }^{\varepsilon }$  exists and is called the flow of $\zeta$  . To be more precise, let $\zeta \in {\mathcal{D}}^{\prime }\left(X,TX\right)$  be a distributional vector field on the manifold $X$  and let $\left({\xi }_{\varepsilon }{\right)}_{\varepsilon }$  be a net of smooth vector fields with complete flows ${\Phi }^{\varepsilon }\left(t,.\right)$  and ${\xi }_{\varepsilon }\to \zeta \in {\mathcal{D}}^{\prime }\left(X,TX\right)$  . $\zeta$  is called a vector field with measurable flow ${\Psi }_{t}$  if
• (i) ${\Phi }^{\varepsilon }\left(t,.\right)\to \Psi \left(t,.\right)$  almost everywhere on $X$  for all $t$  (in particular, ${\Psi }_{t}$  is measurable), and
• (ii) For each $t\in \mathbb{R}$  and each $C\subset \subset X$  there exists ${\varepsilon }_{0}\in I$  and $K\subset \subset X$  with $C\subseteq K$  such that ${\Phi }^{\varepsilon }\left(t,C\right)\subseteq K$  for all $\varepsilon$  .
It should be noted that in our terminology, (ii) says that ${\Phi }^{\varepsilon }\left(t,.\right)$  is c-bounded. It turns out, however, that (in contradiction to Thm. 6.2 in [12) the assumption on $\zeta$  to be a vector field with measurable flow ${\Psi }_{t}$  does not guarantee the flow properties of ${\Psi }_{t}$  almost everywhere (see [10, Prop. 4.1 for a counterexample).
Despite this seeming impasse, the basic approach of Marsden (i.e., regularizing the distributional vector field and considering the net of flows corresponding to these regularizations), can successfully be transferred to the Colombeau setting.
Recall that ${\mathcal{G}}_{0}^{1}\left(X\right)$  denotes the space of Colombeau generalized vector fields on $X$  . In order to derive existence and uniqueness theorems for flows of generalized vector fields we will need the following notions of boundedness in terms of Riemannian metrics on $X$  .
2.1 Definition. Let $\xi \in {\mathcal{G}}_{0}^{1}\left(X\right)$  .
• (i) We say that $\xi$  is locally bounded resp. locally of ${L}^{\infty }$  -log-type if for all $K\subset \subset X$  and one (hence every) Riemannian metric $h$  on $X$  we have for one (hence every) representative ${\xi }_{\varepsilon }$  ${sup}_{p\in K}\parallel {\xi }_{\varepsilon }{|}_{p}{\parallel }_{h}\le C\text{resp.}{sup}_{p\in K}\parallel {\xi }_{\varepsilon }{|}_{p}{\parallel }_{h}\le C|log\varepsilon |,$  where $\parallel {\parallel }_{h}$  denotes the norm induced on ${T}_{p}X$  by $h$  .
• (ii) $\xi$  is called globally bounded with respect to $h$  if for some (hence every) representative $\left({\xi }_{\varepsilon }{\right)}_{\varepsilon }$  of $\xi$  there exists $C>0$  with ${sup}_{p\in X}\parallel {\xi }_{\varepsilon }{|}_{p}{\parallel }_{h}\le C.$
Note that, contrary to local boundedness resp. local ${L}^{\infty }$  -log type, global boundedness obviously depends on the choice of Riemannian metric. We now have the following basic existence and uniqueness result for ordinary differential equations with generalized right hand side on a differentiable manifold:
2.2 Theorem. Let $\left(X,h\right)$  be a complete Riemannian manifold, ${\stackrel{~}{x}}_{0}\in {\stackrel{~}{X}}_{c}$  and $\xi \in {\mathcal{G}}_{0}^{1}\left(X\right)$  such that
• (i) $\xi$  is globally bounded with respect to $h$  .
• (ii) For each differential operator $P\in \mathcal{P}\left(X,TX\right)$  of first order $P\xi$  is locally of ${L}^{\infty }$  -log-type.
Then the initial value problem
 $\begin{array}{ccc}\stackrel{˙}{x}\left(t\right)& =& \xi \left(x\left(t\right)\right)\end{array}$ (2)
 $\begin{array}{ccc}x\left({t}_{0}\right)& =& {\stackrel{~}{x}}_{0}\end{array}$
has a unique solution $x$  in $\mathcal{G}\left[\mathbb{R},X\right]$  .
Proof. See [10, Th. 3.5. $\square$  Moreover, we have:
2.3 Theorem. Let $\left(X,h\right)$  be a complete Riemannian manifold and suppose that $\xi \in {\mathcal{G}}_{0}^{1}\left(X\right)$  satisfies conditions (i) and (ii) of Theorem  2.2 . Then there exists a unique generalized function $\Phi \in \mathcal{G}\left[\mathbb{R}×X,X\right]$  , the generalized flow of $\xi$  , such that
 $\begin{array}{ccc}\frac{d}{dt}\Phi \left(t,x\right)& =& \xi \left(\Phi \left(t,x\right)\right)\text{in}{\mathcal{G}}^{h}\left[\mathbb{R}×X,TX\right]\end{array}$ (3)
 $\begin{array}{ccc}\Phi \left(0,.\right)& =& i{d}_{X}\text{in}\mathcal{G}\left[X,X\right]\end{array}$ (4)
 $\begin{array}{ccc}\Phi \left(t+s,.\right)& =& \Phi \left(t,\Phi \left(s,.\right)\right)\text{in}\mathcal{G}\left[{\mathbb{R}}^{2}×X,X\right].\end{array}$ (5)
Proof. See [10, Th. 3.6. $\square$
2.4 Definition. Let $\xi \in {\mathcal{G}}_{0}^{1}\left(X\right)$  be a generalized vector field for which there exists a unique global generalized flow $\Phi \in \mathcal{G}\left[\mathbb{R}×X,X\right]$  satisfying ( 3 ), ( 4 ) and ( 5 ). Then both $\xi$  and its generalized flow $\Phi$  are called $\mathcal{G}$  -complete. $\Phi$  is called a generalized (one-parameter) group action on $X$  and $\xi$  is called the infinitesimal generator of $\Phi$  .
Once the existence of the generalized flow of an element $\xi$  of ${\mathcal{G}}_{0}^{1}\left(X\right)$  is secured, the question arises whether there exist distributional limits of the corresponding flows ${\Phi }_{\varepsilon }$  . For a detailed analysis of this question (which, in a certain sense provides a resolution to the problems encountered in the distributional modelling of generalized flows above) we refer to [10, sec. 6.

3 Group invariants in the Colombeau setting

If $\Phi$  is a generalized group action on $X$  we call $u\in \mathcal{G}\left(X\right)$  invariant under $\Phi$  if $u\left(\Phi \left(\stackrel{~}{\eta },\stackrel{~}{x}\right)\right)=u\left(\stackrel{~}{x}\right)\forall \stackrel{~}{\eta }\in {\stackrel{~}{\mathbb{R}}}_{c},\stackrel{~}{x}\in {\stackrel{~}{X}}_{c}$  . By the point value characterization of Colombeau generalized functions this condition is equivalent to $u\circ \Phi =u\circ {\pi }_{2}$  as elements of $\mathcal{G}\left(\mathbb{R}×X\right)$  (with ${\pi }_{2}:\mathbb{R}×X\to X$  the projection). The basic infinitesimal criterion for invariance is given in the following proposition, proved here for the sake of completeness (cf. e.g., [6, Th. 4.5.1).
3.1 Proposition. Let $u\in \mathcal{G}\left(X\right)$  and let $\Phi$  be a generalized group action on $X$  with infinitesimal generator $\xi$  . Then the following statements are equivalent:
• (i) $u$  is $\Phi$  -invariant.
• (ii) $\xi \left(u\right)=0$  in $\mathcal{G}\left(X\right)$  .
Proof. (i) $⇒$  (ii) Since $u$  is $\Phi$  -invariant we have $0=\frac{d}{d\eta }{|}_{{}_{0}}\left(u\left(\Phi \left(\eta ,x\right)\right)\right)=\xi \left(u\right){|}_{x}$  in $\mathcal{G}\left(X\right)$  .
(ii) $⇒$  (i) Conversely, let $\xi \left(u\right)=0$  in $\mathcal{G}\left(X\right)$  . Then $\frac{d}{d\eta }u\left(\Phi \left(\eta ,\stackrel{~}{x}\right)\right)=\xi \left(u\right){|}_{\Phi \left(\eta ,\stackrel{~}{x}\right)}=0\text{in}\mathcal{G}\left(\mathbb{R}\right)\forall \stackrel{~}{x}\in {\stackrel{~}{X}}_{c}.$  Therefore, for each $\stackrel{~}{x}$  the map $\eta ↦u\left(\Phi \left(\eta ,\stackrel{~}{x}\right)\right)$  is constant in $\mathcal{G}\left(\mathbb{R}\right)$  , so $u\circ \Phi =u\circ {\pi }_{2}$  , by [6, Th. 3.2.8. $\square$  To analyze the concept of group invariance for Colombeau generalized functions let us first consider the case of a classical (i.e., smooth) generator. If $\xi \in \mathfrak{X}\left(X\right)$  is $\mathcal{G}$  -complete then the generalized flow of $\xi$  coincides with the classical flow. Important examples of $\mathcal{G}$  -complete smooth vector fields include:
3.2 Examples.
• (i) Let $\left(X,g\right)$  be a complete Riemannian manifold and let $\xi \in \mathfrak{X}\left(X\right)$  be globally bounded w.r.t. $g$  . Then $\xi$  is $\mathcal{G}$  -complete. In fact, since ${\xi }_{\varepsilon }\equiv \xi$  is a representative of $\xi$  as an element of ${\mathcal{G}}_{0}^{1}\left(X\right)$  it is clear that for each first order differential operator $P\in \mathcal{P}\left(X,TX\right)$  , $P\xi$  is locally of ${L}^{\infty }$  -log-type. The claim therefore follows from Th.  2.3 .
• (ii) As a particular case of (i), choose $X={\mathbb{R}}^{n}$  with the standard Euclidean metric. It follows that if $\xi$  is a smooth vector field on ${\mathbb{R}}^{n}$  with globally bounded coefficients then $\xi$  is $\mathcal{G}$  -complete and its flow is just the classical smooth flow.
• (iii) Suppose that $X={\mathbb{R}}^{n}$  and the coefficients of $\xi$  are linear functions of $x$  . Then $\xi$  is $\mathcal{G}$  -complete (see, e.g., [11, Th. 3.1). For example, if $\xi ={x}_{i}{\partial }_{{x}_{j}}-{x}_{j}{\partial }_{{x}_{i}}$  is the smooth generator of a rotation in ${\mathbb{R}}^{n}$  then it follows that $\xi$  is $\mathcal{G}$  -complete.
For the base vector fields ${\partial }_{{x}_{i}}$  on ${\mathbb{R}}^{n}$  , the following result by M. Oberguggenberger gives a characterization of translational invariance:
3.3 Theorem. Let $u\in \mathcal{G}\left({\mathbb{R}}^{n}\right)$  . The following are equivalent:
• (i) $u\left({\stackrel{~}{x}}_{1}+\eta ,{\stackrel{~}{x}}_{2},...,{\stackrel{~}{x}}_{n}\right)=u\left(\stackrel{~}{x}\right)$  for all $\stackrel{~}{x}\in {\stackrel{~}{\mathbb{R}}}_{c}^{n},\eta \in {\stackrel{~}{\mathbb{R}}}_{c}$  .
• (ii) ${\partial }_{{x}_{1}}u=0$  in $\mathcal{G}\left({\mathbb{R}}^{n}\right)$  .
• (iii) $u$  has a representative $\left({u}_{\varepsilon }{\right)}_{\varepsilon }$  such that ${\partial }_{{x}_{1}}{u}_{\varepsilon }\equiv 0$  for all $\varepsilon$  .
Proof. See [15, Th. 2.2. $\square$
3.4 Remark.
• (i) It follows easily from the proof of [15, Th. 2.2. that an analogous statement is valid for iterated derivatives ${\partial }_{{x}_{{i}_{1}}}...{\partial }_{{x}_{{i}_{k}}}u$  .
• (ii) Until recently it was an open question whether (i)–(iii) in the above theorem is equivalent to
• (i') $u\left({\stackrel{~}{x}}_{1}+\eta ,{\stackrel{~}{x}}_{2},...,{\stackrel{~}{x}}_{n}\right)=u\left(\stackrel{~}{x}\right)$  for all $\stackrel{~}{x}\in {\stackrel{~}{\mathbb{R}}}_{c}^{n},\eta \in \mathbb{R}$  .
i.e., whether standard translations suffice to characterize translational invariance of Colombeau generalized functions. In [19, however, Pilipović, Scarpalezos and Valmorin were able to give an affirmative answer to this question.
Note that Th.  3.3 implies in particular that a generalized function $u$  is invariant under translations if and only if it possesses a distinguished representative $\left({u}_{\varepsilon }{\right)}_{\varepsilon }$  such that each ${u}_{\varepsilon }$  is a translation invariant smooth function. The following proposition locally extends the validity of this result to all smooth group actions which are regular in the sense that their infinitesimal generators are non-vanishing.
3.5 Proposition. Let $\xi \in \mathfrak{X}\left(X\right)$  be $\mathcal{G}$  -complete with flow $\Phi$  and suppose that $\xi \left({x}_{0}\right)\ne 0$  for some ${x}_{0}\in X$  . Then there exists a neighborhood $U$  of ${x}_{0}$  in $X$  such that $u\in \mathcal{G}\left(U\right)$  is invariant under $\Phi$  if and only if $u$  possesses a representative $\left({u}_{\varepsilon }{\right)}_{\varepsilon }$  with each ${u}_{\varepsilon }$  invariant under $\Phi$  .
Proof. By [1, Th. 2.1.9 we may “straighten out” $\xi$  around ${x}_{0}$  . Thus there exists a chart $\left(U,\phi \right)$  around ${x}_{0}$  with ${\phi }_{*}\xi ={\partial }_{{x}_{1}}$  . But then ${\phi }_{*}u{|}_{U}$  and hence $u{|}_{U}$  itself satisfy the claim by Th.  3.3 . $\square$  We next wish to generalize Prop.  3.5 from one-parameter groups to more general Lie group actions on manifolds. Thus let us assume that $G$  is a Lie group and $\Psi :G×X\supseteq \mathcal{V}\to X$  is a regular transformation group (i.e., all orbits have the same dimension as submanifolds and each point in $X$  has a base of neighborhoods whose elements intersect each orbit in a connected subset thereof, cf. [18, p. 41). Analogous to the smooth setting, we call an element $u$  of $\mathcal{G}\left(X\right)$  invariant under $\Psi$  if $u\circ \Psi =u\circ {\pi }_{2}\text{in}\mathcal{G}\left(\mathcal{V}\right),$  where ${\pi }_{2}:G×X\to X$  is the projection onto the second factor. The desired generalization of Prop.  3.5 then takes the following form:
3.6 Theorem. Let $\Psi :G×X\supseteq \mathcal{V}\to X$  be a regular Lie group action on $X$  with $s$  -dimensional orbits. Then each point ${x}_{0}\in X$  possesses a neighborhood $U$  such that the following statements are equivalent for each $u\in \mathcal{G}\left(U\right)$  :
• (i) $u$  is invariant under $\Psi$  .
• (ii) $u$  possesses a representative $\left({u}_{\varepsilon }{\right)}_{\varepsilon }$  with each ${u}_{\varepsilon }$  invariant under $\Psi$  .
Proof. By [18, Th. 2.23 we may choose a rectifying local chart $\phi :x↦\left({y}_{1},...,{y}_{s},{z}_{1},...,{z}_{n-s}\right)$  in a neighborhood $U$  of ${x}_{0}$  such that each group orbit intersects the coordinate chart in at most one slice $\left\{{z}_{1}={c}_{1},...,{z}_{n-s}$  $=$  ${c}_{n-s}\right\}$  (with ${c}_{i}$  constant for $1\le i\le n-s$  ). Denote by ${O}_{x}$  the orbit of $\Psi$  through $x$  . Then for each $x\in U$  the local vector fields ${\frac{\partial }{\partial {y}_{i}}|}_{x}$  ( $1\le i\le s$  ) form a basis of the tangent space of ${O}_{x}$  . It follows that in this neighborhood each infinitesimal generator of $\Psi$  is a unique $\mathcal{C}\infty$  -linear combinations of the $\frac{\partial }{\partial {y}_{i}}$  and vice versa. In $U$  , invariance under $\Psi$  therefore amounts to ${\partial }_{{y}_{i}}u$  being zero in $\mathcal{G}\left(U\right)$  . Therefore, by an application of Remark  3.4 (i) to ${\phi }_{*}u$  we reach the desired conclusion. $\square$  As an important concrete example of invariance of Colombeau generalized functions under smooth transformation groups let us consider in some detail the case of rotational invariance, following [14, 15, 7. Let $\mathsf{S}\mathsf{O}\left(n,\stackrel{~}{\mathbb{R}}\right)$  denote the special orthogonal group over the ring $\stackrel{~}{\mathbb{R}}$  of generalized numbers and $\mathsf{S}\mathsf{O}\left(n,\mathbb{R}\right)$  the usual special orthogonal group. Letting $\mathsf{S}\mathsf{O}\left(n,\stackrel{~}{\mathbb{R}}\right)$  act naturally on ${\mathbb{R}}^{n}$  it is well known that a basis of the Lie algebra of infinitesimal generators of this action is given by the set of all ${\xi }_{ij}={x}_{i}{\partial }_{{x}_{j}}-{x}_{j}{\partial }_{{x}_{i}}$  for $i  . For an element $u$  of ${\mathbb{R}}^{n}$  to be invariant under the flow ${\Phi }_{ij}$  of ${\xi }_{ij}$  means that $u\left(\Phi \left(\stackrel{~}{\eta },\stackrel{~}{x}\right)\right)=u\left(\stackrel{~}{x}\right)\forall \stackrel{~}{\eta }\in {\stackrel{~}{\mathbb{R}}}_{c}\forall \stackrel{~}{x}\in {\stackrel{~}{\mathbb{R}}}_{c}^{n}.$  Here, the action of ${\Phi }_{ij}\left(\stackrel{~}{\eta },.\right)$  is precisely a rotation in the $\left({x}_{i},{x}_{j}\right)$  -plane by the generalized angle $\stackrel{~}{\eta }$  , hence is given by the action of the corresponding element of $\mathsf{S}\mathsf{O}\left(n,\stackrel{~}{\mathbb{R}}\right)$  on $\stackrel{~}{x}$  . Conversely, as was shown in [15, Sec. 2, Lemma 3, each generalized rotation $A\in \mathsf{S}\mathsf{O}\left(n,\stackrel{~}{\mathbb{R}}\right)$  in the $\left({x}_{i},{x}_{j}\right)$  -plane is precisely of this form. (This structural relationship in fact reaches even further: by the same result of Oberguggenberger, the ${\xi }_{ij}$  also form a basis of the “Lie algebra” of $\mathsf{S}\mathsf{O}\left(n,\stackrel{~}{\mathbb{R}}\right)$  in the following sense: each $A\in \mathsf{S}\mathsf{O}\left(n,\stackrel{~}{\mathbb{R}}\right)$  is of the form $exp\left(v\right)$  for some generalized vector field $v={\sum }_{i  with ${\alpha }_{ij}\in \stackrel{~}{\mathbb{R}}$  for all $i  .) Consequently, the action of a generalized $\left(i,j\right)$  -rotation $A\in \mathsf{S}\mathsf{O}\left(n,\stackrel{~}{\mathbb{R}}\right)$  (which is an example of a generalized group action in the sense of section  2 ) can be viewed as the “nonstandardization” of the corresponding classical rotation which is obtained by replacing the real angle $\eta$  by the generalized angle $\stackrel{~}{\eta }$  . This, of course, is a direct result of the $\mathcal{G}$  -completeness of the smooth generators ${\xi }_{ij}$  (cf. Ex.  3.2 (iii)). Combining these observations with Th.  3.1 we obtain (see [15for an alternative direct proof ):
3.7 Proposition. Let $u\in \mathcal{G}\left({\mathbb{R}}^{n}\right)$  . The following are equivalent:
• (i) $u\circ A=u$  for all $A\in \mathsf{S}\mathsf{O}\left(n,\stackrel{~}{\mathbb{R}}\right)$  , i.e., $u$  is rotationally invariant.
• (ii) $\xi \left(u\right)=0$  for each infinitesimal generator $\xi$  of $\mathsf{S}\mathsf{O}\left(n,\mathbb{R}\right)$  .
Moreover, we have:
3.8 Proposition. For $u\in \mathcal{G}\left({\mathbb{R}}^{n}\\left\{0\right\}\right)$  , (i) and (ii) are further equivalent with
• (iii) $u$  possesses a representative consisting entirely of rotation invariant functions.
Proof. $\mathsf{S}\mathsf{O}\left(n,\stackrel{~}{\mathbb{R}}\right)$  acts freely on ${\mathbb{R}}^{n}\\left\{0\right\}$  , so we may employ Prop.  3.5 to establish the claim. $\square$  The restriction to ${\mathbb{R}}^{n}\\left\{0\right\}$  in the above result is grounded in the method of proof (application of Prop.  3.5 ) rather than in the subject matter itself. In fact, the equivalence is true on all of ${\mathbb{R}}^{n}$  (see [15).
The above chain of equivalences raises the question whether in the present context the analogue of Remark  3.4 (ii) can be established as well, i.e., whether (i)–(iii) are equivalent to
• (i') $u\circ A=u$  for all $A\in \mathsf{S}\mathsf{O}\left(n,\mathbb{R}\right)$  .
In fact, we have:
3.9 Proposition. For $u\in \mathcal{G}\left({\mathbb{R}}^{n}\\left\{0\right\}\right)$  , (i) and (i') are equivalent.
Proof. Since each ${\xi }_{ij}$  is nonzero on ${\mathbb{R}}^{n}\\left\{0\right\}$  it may be straightened out, in fact even globally on all of ${\mathbb{R}}^{n}\\left\{0\right\}$  (by using appropriate polar coordinates as charts), cf. the proof of Prop.  3.5 . This procedure reduces the proof to an application of Remark  3.4 . $\square$  A direct extension of the proof of Prop.  3.9 to the case $X={\mathbb{R}}^{n}$  is not possible: contrary to the smooth situation a Colombeau generalized function which is rotationally invariant on ${\mathbb{R}}^{n}\\left\{0\right\}$  need not be rotationally invariant on all of ${\mathbb{R}}^{n}$  . As an example, choose any test function $\phi$  whose support is not rotationally invariant and set $u=\left[\left(\phi \left(./\varepsilon \right){\right)}_{\varepsilon }\right]$  . Then $u$  is supported in $\left\{0\right\}$  yet it is clearly not rotationally invariant on ${\mathbb{R}}^{n}$  . Despite this technical complication, however, it turns out that the result can be extended to all of ${\mathbb{R}}^{n}$  , thereby providing an affirmative answer to a question raised by M. Oberguggenberger in [15:
3.10 Theorem. For $u\in \mathcal{G}\left({\mathbb{R}}^{n}\right)$  , (i) and (i') are equivalent.
Proof. This result was established in [7, Th. 7.6. We include a proof here for the reader's convenience. It clearly suffices to show that (i') implies (i). Let us first consider the case $n=2$  . Let $\stackrel{~}{A}\in \mathsf{S}\mathsf{O}\left(2,\stackrel{~}{\mathbb{R}}\right)$  . Then by the discussion preceding Prop.  3.7 there exists some $\stackrel{~}{\eta }\in {\stackrel{~}{\mathbb{R}}}_{c}$  such that $\stackrel{~}{A}=\left[{\left(\begin{array}{cc}cos\left({\eta }_{\varepsilon }\right)& -sin\left({\eta }_{\varepsilon }\right)\\ sin\left({\eta }_{\varepsilon }\right)& cos\left({\eta }_{\varepsilon }\right)\end{array}\right)}_{\varepsilon }\right].$  Given $\stackrel{~}{x}$  , $\stackrel{~}{y}\in {\stackrel{~}{\mathbb{R}}}_{c}$  we have to show that $u\left(\stackrel{~}{A}\cdot \left(\stackrel{~}{x},\stackrel{~}{y}{\right)}^{t}\right)=u\left(\stackrel{~}{x},\stackrel{~}{y}\right)$  in $\stackrel{~}{\mathbb{R}}$  . We may write $\left(\stackrel{~}{x},\stackrel{~}{y}\right)=\left[\left({r}_{\varepsilon }cos\left({\theta }_{\varepsilon }\right),{r}_{\varepsilon }sin\left({\theta }_{\varepsilon }\right)\right)\right]$  for suitable ${r}_{\varepsilon }\ge 0$  , ${\theta }_{\varepsilon }$  . Now set ${v}_{\varepsilon }:=\theta ↦{u}_{\varepsilon }\left({r}_{\varepsilon }cos\left(\theta \right),{r}_{\varepsilon }sin\left(\theta \right)\right)$  . Then $v=\left[\left({v}_{\varepsilon }{\right)}_{\varepsilon }\right]\in \mathcal{G}\left(\mathbb{R}\right)$  and by assumption $v\left(\stackrel{~}{\theta }+\eta \right)=v\left(\stackrel{~}{\theta }\right)$  in $\stackrel{~}{\mathbb{R}}$  for all $\stackrel{~}{\theta }\in {\stackrel{~}{\mathbb{R}}}_{c}$  and all $\eta \in \mathbb{R}$  . But then by Th.  3.3 it follows that $v$  is a generalized constant, thereby finishing the proof for $n=2$  . In the general case $n\ge 2$  we verify (ii) of Prop.  3.7 . Let $1\le i  and let ${\xi }_{ij}={x}_{i}{\partial }_{{x}_{j}}-{x}_{j}{\partial }_{{x}_{i}}$  as above be an infinitesimal generator of $\mathsf{S}\mathsf{O}\left(n,\mathbb{R}\right)$  . Fix compactly supported generalized numbers ${\stackrel{~}{x}}_{1}$  ,. . . , ${\stackrel{~}{x}}_{i-1}$  , ${\stackrel{~}{x}}_{i+1}$  ,. . . , ${\stackrel{~}{x}}_{j-1}$  , ${\stackrel{~}{x}}_{j+1}$  ,. . . , ${\stackrel{~}{x}}_{n}$  and consider the maps ${w}_{\varepsilon }:\left({x}_{i},{x}_{j}\right)↦{u}_{\varepsilon }\left({\stackrel{~}{x}}_{1},...,{\stackrel{~}{x}}_{i-1},{x}_{i},...,{\stackrel{~}{x}}_{j-1},{x}_{j},...,{\stackrel{~}{x}}_{n}\right).$  Then $w=\left[\left({w}_{\varepsilon }{\right)}_{\varepsilon }\right]\in \mathcal{G}\left({\mathbb{R}}^{2}\right)$  and from our assumption it follows that $w\circ A=w$  in $\mathcal{G}\left({\mathbb{R}}^{2}\right)$  for all $A\in \mathsf{S}\mathsf{O}\left({\mathbb{R}}^{2}\right)$  . By what we have already proved in the 2D-case and Prop.  3.7 it follows that ${\xi }_{ij}w=0$  in $\mathcal{G}\left({\mathbb{R}}^{2}\right)$  for each $i  , which finishes the proof. $\square$  In the smooth setting, the local structure of invariants of a group action is determined by so-called complete sets of functionally independent invariants (cf. [17, Sec. 2.1). A family of smooth functions ${f}_{1},...,{f}_{k}$  on $X$  is called functionally dependent if for each $x\in X$  there exists a neighborhood $U$  of $x$  and a smooth function $F$  which is not identically zero on any open subset of ${\mathbb{R}}^{k}$  such that $x↦F\left({f}_{1}\left(x\right),...,{f}_{k}\left(x\right)\right)$  vanishes identically on $U$  . ${f}_{1},...,{f}_{k}$  are called functionally independent if they are not functionally dependent on any open subset of $X$  . It is shown in [17, Th. 2.17 that if $\Psi$  acts (semi-)regularly with $s$  -dimensional orbits then in a neighborhood of any point ${x}_{0}$  of $X$  there exists a set ${f}_{1},...,{f}_{n-s}$  of functionally independent invariants such that any other local invariant $f$  of $\Psi$  is of the form $f\left(x\right)=F\left({f}_{1}\left(x\right),...,{f}_{n-s}\left(x\right)\right)$  for some smooth function $F$  . Such a family ${f}_{1},...,{f}_{n-s}$  is called a complete set of functionally independent invariants of $\Psi$  . Using Th.  3.6 we now show that an analogous characterization of generalized invariants of regular smooth group actions holds true.
3.11 Theorem. Let $\Psi :G×X\supseteq \mathcal{V}\to X$  be a regular Lie group action on $X$  with $s$  -dimensional orbits. Then for each ${x}_{0}\in X$  there exists a neighborhood $U$  of ${x}_{0}$  and a complete set of functionally independent invariants ${f}_{1},...,{f}_{n-s}$  on $U$  such that every $\Psi$  -invariant $u\in \mathcal{G}\left(U\right)$  is of the form $u\left(x\right)=v\left({f}_{1}\left(x\right),...,{f}_{n-s}\left(x\right)\right)$  for some $v\in \mathcal{G}\left(V\right)$  , where $V$  is an appropriate open subset of ${\mathbb{R}}^{n-s}$
Proof. Employing the notations of Th.  3.6 , the coordinates ${z}_{1}={f}_{1}\left(x\right)$  , . . . ${z}_{n-s}={f}_{n-s}\left(x\right)$  are in fact a complete set of functionally independent invariants for the group action $\Psi$  on $X$  (cf. the proof of Th. 2.17 in [17). Now if $u\in \mathcal{G}\left(U\right)$  is an invariant of $\Psi$  then by Th.  3.6 we may choose a representative $\left({u}_{\varepsilon }{\right)}_{\varepsilon }$  of $u$  such that each ${u}_{\varepsilon }$  is an invariant of $\Psi$  on $U$  . Then by [17, Th. 2.17, for each $\varepsilon$  there exists a smooth function ${v}_{\varepsilon }$  such that ${u}_{\varepsilon }\left(x\right)={v}_{\varepsilon }\left({f}_{1}\left(x\right),...,{f}_{n-s}\left(x\right)\right)$  . Denoting ${\phi }_{*}u$  by $\stackrel{~}{u}$  we have ${\stackrel{~}{u}}_{\varepsilon }\left(y,z\right)={v}_{\varepsilon }\left(z\right)$  , so $\left({v}_{\varepsilon }{\right)}_{\varepsilon }\in \mathcal{ℰ}M\left(V\right)$  , where $V$  is the projection of $\phi \left(U\right)$  onto the last $n-s$  components. It follows that the class $v=\left[\left({v}_{\varepsilon }{\right)}_{\varepsilon }\right]$  of $\left({v}_{\varepsilon }{\right)}_{\varepsilon }$  in $\mathcal{G}\left(V\right)$  is the desired generalized function. $\square$  References

1. Abraham, R., Marsden, J. E. Foundations of Mechanics. Benjamin/Cummings, Reading, MA, second edition, 1978.
2. Berest, Yu. Yu. Group analysis of linear differential equations in distributions and the construction of fundamental solutions. Diff. Equ., 29(11):1700–1711, 1993.
3. Berest, Yu. Yu., Ibragimov, N. H. Group theoretic determination of fundamental solutions. Lie Groups Appl., 1(2):65–80, 1994.
4. Bluman, G. W., Kumei, S. Symmetries and Differential Equations. Springer, New York, 1989.
5. Djapić, N., Kunzinger, M., Pilipović, S. Symmetry group analysis of weak solutions. Proc. London Math. Soc. (3), 84(3):686–710, 2002.
6. Grosser, M., Kunzinger, M., Oberguggenberger, M., Steinbauer, R. Geometric Theory of Generalized Functions, volume 537 of Mathematics and its Applications 537. Kluwer Academic Publishers, Dordrecht, 2001.
7. Konjik, S., Kunzinger, M. Generalized group actions in a global setting. Submitted.
8. Kunzinger, M. Nonsmooth differential geometry and algebras of generalized functions. J. Math. Anal. Appl., 297:456–471, 2004.
9. Kunzinger, M., Oberguggenberger, M. Group analysis of differential equations and generalized functions. SIAM J. Math. Anal., 31(6):1192–1213, 2000.
10. Kunzinger, M., Oberguggenberger, M., Steinbauer, R., Vickers, J. Generalized flows and singular ODEs on differentiable manifolds. Acta Appl. Math., 80:221–241, 2004.
11. Ligȩza, J. Remarks on generalized solutions of ordinary differential equations in the Colombeau algebra. Math. Bohem., 123(3):301–316, 1998.
12. Marsden, J. E. Generalized Hamiltonian mechanics. Arch. Rat. Mech. Anal., 28(4):323–361, 1968.
13. Methée, P. D. Sur les distributions invariantes dans le groupe des rotations de Lorentz. Comment. Math. Helv., 28:224–269, 1954.
14. Oberguggenberger, M. Symmetry groups, nonlinear partial differential equations, and generalized functions. In Lesley, J. A., Robart, T., editor, Geometrical Study of Differential Equations, volume 285 of Contemporary Mathematics, pages 101–110. Amer. Math. Soc., 2001.
15. Oberguggenberger, M. Rotationally invariant Colombeau functions. In Delcroix, A., Hasler, M., Marti, J.-A., Valmorin, V., editor, Nonlinear Algebraic Analysis and Applications, pages 227–236. Cambridge Scientific Publishers, 2004.
16. Oberguggenberger, M., Rosinger, E. E. Solution of Continuous Nonlinear PDEs through Order Completion. North Holland, Amsterdam, 1994.
17. Olver, P. J. Applications of Lie Groups to Differential Equations, volume 107 of Graduate Texts in Mathematics. Springer, New York, second edition, 1993.
18. Olver, P. J. Equivalence, Invariants, and Symmetry. Cambridge University Press, 1995.
19. Pilipović, S., Scarpalezos, D., Valmorin, V. Equalities in generalized function algebras. Preprint, 2004.
20. Rosinger, E. E., Rudolph, M. Group invariance of global generalized solutions of smooth nonlinear PDEs: a Dedekind order completion method. Lie Groups Appl., 1(1):203–215, 1994.
21. Rosinger, E. E., Walus, Y. E. Group invariance of generalized solutions obtained through the algebraic method. Nonlinearity, 7:837–859, 1994.
22. Szmydt, Z. On homogeneous rotation invariant distributions and the laplace operator. Ann. Pol. Math., 6:249–259, 1979.
23. Ziemian, B. On distributions invariant with respect to some linear transformations. Ann. Pol. Math., 36:261–276, 1979.