When unit groups of continuous inverse

algebras are regular Lie groups

Helge Glöckner and Karl-Hermann Neeb
arXiv:1202.1129v1 [math.FA] 6 Feb 2012

Abstract
It is a basic fact in infinite-dimensional Lie theory that the unit group
A× of a continuous inverse algebra A is a Lie group. We describe
criteria ensuring that the Lie group A× is regular in Milnor’s sense.
Notably, A× is regular if A is Mackey-complete and locally m-convex.

Classification: 22E65 (primary); 32A12, 34G10, 46G20, 46H05, 58B10.

Key words: Continuous inverse algebra, Q-algebra, Waelbroeck algebra, locally m-convex
algebra, infinite-dimensional Lie group, regular Lie group, regularity, left logarithmic
derivative, product integral, evolution, initial value problem, parameter dependence

Introduction and statement of the main result

A locally convex, unital, associative topological algebra A over K ∈ {R, C}
is called a continuous inverse algebra if its group A× of invertible elements
is open and the inversion map ι : A× → A, x 7→ x−1 is continuous (cf. [19]).
Then ι is K-analytic and hence A× is a K-analytic Lie group [6]. Our goal is
to describe conditions ensuring that the Lie group A× is well-behaved, i.e., a
regular Lie group in the sense of Milnor [15].
To recall this notion, let G be a Lie group modelled on a locally convex
space E, with identity element 1, its tangent bundle T G and the Lie algebra
g := T1 G ∼
= E. Given g ∈ G and v ∈ T1 G, let λg : G → G, x 7→ gx be left
translation by g and gv := T1 (λg )(v) ∈ Tg G. If γ : [0, 1] → g is a continuous
map, then there exists at most one C 1 -map η : [0, 1] → G such that

η ′ (t) = η(t)γ(t) for all t ∈ [0, 1], and η(0) = 1.

If such an η exists, it is called the evolution of γ. The Lie group G is

called regular if each γ ∈ C ∞ ([0, 1], g) admits an evolution ηγ , and the map
evol : C ∞ ([0, 1], g) → G, γ 7→ ηγ (1) is smooth (see [15] and [16], where also
many applications of regularity are described). If G is regular, then its
modelling space E is Mackey-complete in the sense that every Lipschitz curve

1
in E admits a Riemann integral1 (as shown in [9]). It is a notorious open
problem whether, conversely, every Lie group modelled on a Mackey-complete
locally convex space is regular ([16, Problem II.2]; cf. [15]).
As a tool for the discussion of A× , we let µn : An → A be the n-linear
map defined via µn (x1 , . . . , xn ) := x1 x2 · · · xn , for n ∈ N. Given seminorms
q
p, q : A → [0, ∞[, we define B 1 (0) := {x ∈ A : q(x) ≤ 1} and
q
kµn kp,q := sup{p(µn (x1 , . . . , xn )) : x1 , . . . , xn ∈ B 1 (0)} ∈ [0, ∞] .

Our regularity criterion now reads as follows:

Theorem. Let A be a Mackey-complete continuous inverse algebra, such

that the following condition is satisfied :

(∗) For each continuous seminorm p on A, there exists a continuous semi-

norm q on A and r > 0 (which may depend on p and q) such that
∞
X
r n kµn kp,q < ∞ .
n=1

Then A× is a regular Lie group in Milnor’s sense.

A× even satisfies certain stronger regularity properties (see Proposition 3.4).

Of course, by Hadamard’s formula for the radius
p of convergence of a power
2
series, condition (∗) is equivalent to lim sup kµn kp,q < ∞. It is also equiv-
n

n→∞
alent to the existence of M ∈ [0, ∞[ such that kµn kp,q ≤ M n for all n ∈ N.
Remark A. The authors do not know whether condition (∗) can be
omitted, i.e., whether A× is regular for every Mackey-complete continuous
inverse algebra A. Here are some preliminary considerations:
If A is a continuous inverse algebra, then the map πn : A → A, x 7→ xn is
a continuous homogeneous polynomial of degree n, for each n ∈ N0 . It is
×
known that the analytic
P∞ ninversion
P∞ map ι : A → A is given by Neumann’s
series, ι(1 − x) = n=0 x = n=0 πn (x), for x in some 0-neighbourhood of
1
See [12] for a detailed discussion of this property.
2
If kµn kp,q < ∞, then also kµk kp,q < ∞ for all k ∈ {1, . . . , n}. In fact, kµk kp,q ≤
q(1)n−k kµn kp,q since µk (x1 , . . . , xk ) = µn (1, . . . , 1, x1 , . . . , xk ).

2
A [6, Lemma 3.3]. Hence, for each continuous seminorm p on A, there exists
a continuous seminorm q on A and s > 0 such that
∞
X
sn kπn kp,q < ∞ ,
n=1
q
where kπn kp,q := sup{p(πn (x)) : x ∈ B 1 (0)} (cf. [2, Proposition 5.1]).3 Let Sn
sym n
be the symmetric group P of all permutations of {1, . . . , n} and µn : A → A,
1
(x1 , . . . , xn ) 7→ n! σ∈Sn xσ(1) · · · xσ(n) be the symmetrization of µn . Then
nn
πn (x) = µsym sym
n (x, . . . , x) and thus kµn kp,q ≤ n! kπn kp,q by the Polarization
n
Formula (in the form [10, p. 34, (2)]). Since limn→∞ √ n
n!
= e is Euler’s con-
stant (as a consequence of Stirling’s Formula), it follows that
X ∞
tn kµsym s
n kp,q < ∞ for each t ∈ ]0, e [. (1)
n=1
In general, it is not clear how one could give good estimates for kµn kp,q in
terms of kµsym
n kp,q . Hence, it does not seem to be clear in general whether
(1) implies the existence of some r > 0 with (∗).
However, (∗) is satisfied in important cases. Following [13], a topological
algebra A is called locally m-convex if its topology arises from a set of semi-
norms q which are submultiplicative, i.e., q(xy) ≤ q(x)q(y) for all x, y ∈ A.

Corollary. Let A be a Mackey-complete continuous inverse algebra. If A is

commutative or locally m-convex, then A× is a regular Lie group.

Proof. In fact, if A is commutative, then µn = µsymn , whence (∗) is satisfied

with any r ∈ ]0, s/e[ as in (1). Therefore the theorem applies.4
If A is locally m-convex, then for every continuous seminorm p on A, there is
a submultiplicative continuous seminorm q on A such that p ≤ q. Using the
submultiplicativity, we see that kµn kp,q ≤ kµn kq,q ≤ 1. Thus (∗) is satisfied
with any r ∈ ]0, 1[, and the theorem applies. ✷
It can be shown that every Mackey-complete, commutative continuous in-
verse algebra is locally m-convex (cf. [18]).
3
If K = R, we can apply the proposition to AC , which is a complex continuous inverse
algebra (see, e.g., [6, Proposition 3.4]).
4
Alternative proof: (A, +) is regular, as it is Mackey complete [16, Proposition II.5.6].
Since exp : A → A× is a homomorphism of groups (as A× is abelian) and a local diffeo-
morphism (see [6, Theorem 5.6]), it follows that also A× is regular [17, Proposition 3].

3
Remark B. We mention that there is a quite direct, alternative proof for the
corollary if A is locally m-convex and complete.5 The easier arguments fail
however if A is not complete, but merely sequentially complete or Mackey-
complete. By contrast, our more elaborate method does not require that A
be complete: Mackey-completeness suffices.
Remark C. Our theorem is a variant of the (possibly too optimistic)
Theorem IV.1.11 announced in the survey [16], and its proof expands the
sketch of proof given there. To avoid the difficulties described in Remark A,
we added condition (∗).
Remark D. It is known that unit groups of Mackey-complete continuous
inverse algebras are so-called BCH-Lie groups [6, Theorem 5.6], i.e., they
admit an exponential function which is an analytic diffeomorphism around 0
(see [5], [16], [17] for information on such groups). Inspiration for the studies
came from an article by Robart [17]. He pursued the (possibly too optimistic)
larger goal to show that every BCH-Lie group with Mackey-complete mod-
elling space is regular. However, there seem to be gaps in his arguments.6

Remark E. The following questions are open:

(a) Are there examples of Mackey-complete continuous inverse algebras

which satisfy (∗) but are not locally m-convex? Or even:

(b) Does every Mackey-complete continuous inverse algebra satisfy (∗) ?

1 Notation and preparatory results

In this section, we fix some notation and formulate preparatory results.
Basic notation. Let N = {1, 2, . . .} and N0 := N ∪ {0}. If X is a set and
n ∈ N, we write X n := X × · · · × X (with n factors). If f : X → Y is a map,
we abbreviate f n := f ×· · ·×f : X n → Y n , (x1 , . . . , xn ) 7→ (f (x1 ), . . . , f (xn )).
5
Then A = lim Aq is a projective limit of Banach algebras (where q ranges through the
←−
set of all submultiplicative continuous seminorms on A). Being a Banach-Lie group, each
A×q is regular [15]. Then C ([0, 1], A) = lim C ∞ ([0, 1], Aq ) and evolA× = lim evolA× is a
∞
←− ←− q
smooth evolution (cf. [1, Lemma 10.3]).
6
For example, it is unclear whether the limit γu constructed in the proof of [17, Propo-
sition 7] takes its values in Aut(L) (as observed by K.-H. Neeb), and no explanation is
given how a smooth curve g in the local group with Ad(g) = γu can be obtained.

4
If (E, k.kE ) and (F, k.kF ) are normed spaces and β : E n → F is a
continuous n-linear map, we write kβkop for its operator norm, defined as
usual as sup{kβ(x1 , . . . , xn )kF : x1 , . . . , xk ∈ E, kx1 kE , . . . , kxn kE ≤ 1}. If
E is a locally convex space, we let P (E) be the set of all continuous semi-
norms on E. If p ∈ P (E), we consider the factor space Ep := E/p−1 (0) as a
normed space with the norm k.kp given by kx + p−1 (0)kp := p(x). Then the
canonical map πp : E → Ep , x 7→ x + p−1 (0) is linear and continuous, with
kπp (x)kp = p(x).
Weak integrals. Recall that if E is a locally convex space, Ra ≤ b are reals
b
and γ : [a, b] → E a continuous map, then the weak integral a γ(s) ds (if it
Rb Rb
exists) is the unique element of E such that λ( a γ(s) ds) = a λ(γ(s)) ds for
each continuous linear functional λ on E. RIf α : E → F is a continuous linear
b
map between locally convex spaces and a γ(s) ds (as before) exists in E,
Rb
then also a α(γ(s)) ds exists in F and is given by
Z b Z b 
α(γ(s)) ds = α γ(s) ds (2)
a a

(see, e.g., [9] for this observation). If E is sequentially complete, then

Rb
a
γ(s) ds always exists (cf. [2, Lemma 1.1] or [10, 1.2.3]).
C r -curves. Let r ∈ N0 ∪ {∞}. As usual, a C r -curve in a locally con-
vex space E is a continuous function γ : I → E such that the derivatives
γ (k) : I → E of order k exist for all k ∈ N with k ≤ r, and are continuous
(see, e.g., [9] for more details). The C ∞ -curves are also called smooth curves.
Smooth maps. If E and F are real locally convex spaces, U ⊆ E is an
open subset and r ∈ N0 ∪ {∞}, then a function f : U → F is called C r
if is continuous, the iterated directional derivatives d(k) f (x, y1 , . . . , yk ) :=
(Dyk · · · Dy1 f )(x) exist for all k ∈ N such that k ≤ r, x ∈ U and y1 , . . . , yk ∈
E, and define continuous functions d(k) f : U × E k → F . If U is not open,
but a convex (or locally convex) subset of E with dense interior U 0 , we say
that f is C r if it is continuous, f |U 0 is C r and d(k) (f |U 0 ) : U 0 × E k → F has
a continuous extension d(k) f : U × E k → F for each k ∈ N such that k ≤ r.
The C ∞ -maps are also called smooth. Abbreviate df := d(1) f . It is known
that the Chain Rule holds in the form d(f ◦ g)(x, y) = df (g(x), dg(x, y)), and
that compositions of C r -maps are C r . Moreover, a C 0 -curve γ : I → E is a
C r -curve if and only if it is a C r -map, in which case γ ′ (t) = dγ(t, 1) (see [9]

5
for all of these basic facts; cf. also [14], [15], and [4]).
Analytic maps. If E and F are complex locally convex spaces and n ∈ N,
then a function p : E → F is called a continuous homogeneous polynomial
of degree n ∈ N0 if p(x) = β(x, . . . , x) for some continuous n-linear map
β : E n → F (if n = 0, this means a constant function). A map f : U → F on
an open set U ⊆ E is called complex analytic (or C-analytic) if it is continu-
ous and for each x ∈ U, there is a 0-neighbourhood Y ⊆ E with x + Y ⊆ U
and continuous homogeneous polynomials pn : E → F of degree n, such that
∞
X
(∀y ∈ Y ) f (x + y) = pn (y)
n=0

(see [2], [4] and [9] for further information). Following [15], [4] and [9] (but
deviating from [2]), given real locally convex spaces E, F , we call a function
f : U → F on an open set U ⊆ E real analytic (or R-analytic) if it extends
to a complex analytic map V → FC , defined on some open subset V ⊆ EC
of the complexification of E, such that U ⊆ V . For K ∈ {R, C}, it is known
that compositions of K-analytic maps are K-analytic. Every K-analytic map
is smooth (see, e.g., [9] or [4] for both of these facts).
We shall use the following lemma (proved in Appendix A):

Lemma 1.1 Let E and F be complex locally convex spaces, Fe be a comple-

tion of F such that F ⊆ Fe as a dense vector subspace, and pn : E → F be
continuous homogeneous polynomials of degree n, for n ∈ N0 . Assume that
X
f (x) := pn (x)
n∈N0

converges in Fe for all x in a balanced, open 0-neighbourhood U ⊆ E, and

f : U → Fe is continuous. If F is Mackey-complete, then f (x) ∈ F for all
x ∈ U and f : U → F is C-analytic.

Function spaces. If E is a locally convex space and r ∈ N0 ∪ {∞},

let C r ([0, 1], E) be the space of all C r -maps from [0, 1] to E. We endow
C r ([0, 1], E) with the locally convex vector topology defined by the semi-
norms k.kC k ,p given by

kγkC k ,p := max max p(γ (j) (t)),

j=0,...,k t∈[0,1]

6
for p in the set of continuous seminorms on E and k ∈ N0 with k ≤ r. We
abbreviate C([0, 1], E) := C 0 ([0, 1], E). Three folklore lemmas concerning
these function spaces will be used. (The proofs can be found in Appendix A).

Lemma 1.2 Let E and F be locally convex spaces α : E → F be a continuous

linear map, and r ∈ N0 ∪ {∞}. Then also the map

α∗ := C r ([0, 1], α) : C r ([0, 1], E) → C r ([0, 1], F ), γ 7→ α ◦ γ

is continuous and linear. If α is a topological embedding (i.e., a homeomor-

phism onto its image), then also α∗ is a topological embedding.

Lemma 1.3 If E is a locally convex space and r ∈ N0 ∪ {∞}, then the

topology on C r ([0, 1], E) is initial with respect to the mappings (πp )∗ :
C r ([0, 1], E) → C r ([0, 1], Ep ), γ 7→ πp ◦ γ for p ∈ P (E).

Lemma 1.4 If r ∈ N0 ∪ {∞} and E is a locally convex space E which is

complete (resp., Mackey-complete), then also C r ([0, 1], E) is complete (resp.,
Mackey-complete).

2 Picard iteration of paths in a topological

algebra
2.1 Let A be a locally convex topological algebra over C, i.e., a unital,
associative, complex algebra, equipped with a Hausdorff locally convex vector
topology making the map A × A → A, (x, y) 7→ xy continuous. We assume
that condition (∗) from Theorem A is satisfied.7

If E is a locally convex space, then a function γ : [0, 1] → E is a Lipschitz

curve if { γ(t)−γ(s)
t−s
: s 6= t ∈ [0, 1]} is bounded in E (cf. [12, p. 9]). For our
current purposes, we endow the space Lip([0, 1], E) of all such curves with
the topology OC 0 induced by C 0 ([0, 1], E).

7
Note that A is not assumed to be a continuous inverse algebra in this section.

7
Lemma 2.2 (Picard Iteration) Let A be as in 2.1. If A is sequentially
complete and γ ∈ C([0, 1], A), we can define a sequence (ηn )n∈N in C 1 ([0, 1], A)
via η0 (t) := 1,
Z t
ηn (t) = 1 + ηn−1 (tn )γ(tn ) dtn for t ∈ [0, 1] and n ∈ N.
0

Then the following holds:

(a) The limit η := ηγ := lim ηn exists in C 1 ([0, 1], A).

n→∞
Pn R t R tk−1 R t2
(b) ηn (t) = 1+ k=1 0 0 · · · 0
γ(t1 ) · · · γ(tk ) dt1 . . . dtk for each n ∈ N0
and t ∈ [0, 1], and thus
∞ Z tZ
X tn−1 Z t2
η(t) = 1 + ··· γ(t1 ) · · · γ(tn ) dt1 . . . dtn . (3)
n=1 0 0 0

(c) η ′ (t) = η(t)γ(t) and η(0) = 1.

(d) The map Φ : C([0, 1], A) → C 1 ([0, 1], A), γ 7→ ηγ is complex analytic.

If A is not sequentially complete, but Mackey-complete, then the (ηn )n∈N0 can
be defined and (a)–(c) hold for each γ ∈ Lip([0, 1], A). Moreover,

(d)′ Φ : (Lip([0, 1], A), OC 0 ) → C 1 ([0, 1], A), γ 7→ ηγ is complex analytic.

Proof. If A is sequentially complete, set X := C([0, 1], A); otherwise, set

X := Lip([0, 1], A). Let A e be a completion of A such that A ⊆ A. e Then
1 1 e
the inclusion map φ : C ([0, 1], A) → C ([0, 1], A) is a topological embed-
ding (Lemma 1.2) and C 1 ([0, 1], A) e is complete (Lemma 1.4). Hence also
the closure Y ⊆ C 1 ([0, 1], A) e of the image im(φ) is complete, and thus
φ : C 1 ([0, 1], A) → Y is a completion of C 1 ([0, 1], A).
Proof of (a) and (b), (d) and (d)′ : Let γ ∈ X. Then all integrals needed
to define σn (γ) := ηn exist, and each ηn is C 1 , by the Fundamental Theorem
of Calculus. A trivial induction shows that
n Z t Z tk−1
X Z t2
ηn (t) = 1 + ··· γ(t1 ) · · · γ(tk ) dt1 . . . dtk (4)
k=1 0 0 0

8
(as asserted in (b)). Likewise, if n ∈ N and γ1 , . . . , γn ∈ X, then the weak
integrals needed to define τn (γ1 , . . . , γn ) : [0, 1] → A,
Z t Z tn Z t2
t 7→ ··· γ1 (t1 ) · · · γn (tn ) dt1 · · · dtn
0 0 0

exist and τn (γ1 , . . . , γn ) is a C 1 -map. Since τn : C([0, 1], A)n → C 1 ([0, 1], A),
(γ1 , . . . , γn ) 7→ τn (γ1 , . . . , γn ) is n-linear and σn (γ) = τn (γ, . . . , γ), the map
σn : C([0, 1], A) → C 1 ([0, 1], A) is a homogeneous polynomial of degree n (and
this conclusion also holds if n = 0, as σ0 is constant). If p ∈ P (A), there is
q ∈ P (A) and M ∈ [0, ∞[ such that
(∀n ∈ N) kµn kp,q ≤ M n ,
as a consequence of condition (∗). Applying p to the iterated integral in
n nM n
(4), we deduce that p(σn (γ)(t)) ≤ tn! kµn kp,q kγknC0 ,q ≤ t n! kγknC0 ,q for each
t ∈ [0, 1] and thus
Mn
kσn (γ)kC 0 ,p ≤ kγknC 0 ,q . (5)
n!
Also, σ1 (γ)′ (t) = γ(t) and
Z t Z tn−1 Z t2
′
σn (γ) (t) = ··· γ(t1 ) · · · γ(tn−1 )γ(t) dt1 · · · dtn−1 (6)
0 0 0

if n ≥ 2, by the Fundamental Theorem of Calculus. Thus

(∀n ∈ N) ηn′ (t) = ηn−1 (t)γ(t) , (7)
tn−1
which will be useful later. By (6), also p(σn (γ)′ (t)) ≤ (n−1)!
kµn kp,q kγknC 0 ,q
and thus
Mn
kσn (γ)′ kC 0 ,p ≤ kγknC 0 ,q . (8)
(n − 1)!
Combining (5) and (8), we see that
Mn
kσn (γ)kC 1 ,p ≤ kγknC 0 ,q . (9)
(n − 1)!
Therefore σn : C([0, 1], A) → C 1 ([0, 1], A) is a continuous homogeneous
polynomial. Moreover, we obtain
∞
X ∞
X M n kγknC 0 ,q
kσn (γ)kC 1 ,p ≤ = MkγkC 0 ,q eM kγkC 0 ,q < ∞ .
n=1 n=1
(n − 1)!

9
 P
This estimate entails that the series ∞ n=0 σn (γ) converges absolutely in the
1
completion Y of C ([0, 1], A). In particular, the limit
∞
X
Φ(γ) := σn (γ)
n=0

exists in Y , and defines a function Φ : X → Y . We claim that Φ is continu-

ous. If this is true, then we can exploit that C 1 ([0, 1], A) is Mackey-complete
by Lemma 1.4, and each σn takes its values inside C 1 ([0, 1], A). Thus all hy-
potheses of Lemma 1.1 are satisfied, and we deduce that Φ(γ) ∈ C 1 ([0, 1], A)
for each γ (entailing (a) and (b)), and that the map Φ : X → C 1 ([0, 1], A) is
complex analytic (establishing (d) and (d)′ , respectively).
To establish the claim, we need only show that Φ is continuous as a map to
e Identify p ∈ P (A) with its continuous extension to a seminorm
C 1 ([0, 1], A).
on A.e Let πp : Ae → ((A)
e p , k.kp ) be the canonical map. By Lemma 1.3, Φ will
e p ) are continu-
be continuous if the maps h := (πp )∗ ◦ Φ : X → C 1 ([0, 1], (A)
ous. It suffices that h is continuous on the ball BR := {γ ∈ X : kγkC 0 ,q < R}
for each R > 0. However,
∞
X
h(γ) = πp ◦ σn (γ)
n=0

for γ ∈ BR , where
Mn Mn
kπp ◦ σn (γ)kC 1 ,k.kp = kσn (γ)kC 1 ,p ≤ kγknC 0 ,q ≤ Rn
(n − 1)! (n − 1)!

for n ∈ N, by (9). Hence

∞
X
sup{πp ◦ σn (γ) : γ ∈ BR } ≤ 1 + MReRM < ∞ ,
n=0
P
entailing that nk=0 ((πp )∗ ◦σn |BR ) → h|R uniformly. Thus h|BR is continuous,
being a uniform limit of continuous functions.
(c) Because ηn → η in C 1 ([0, 1], A), we know that ηn′ → η ′ uniformly (and
thus pointwise). Letting n → ∞ in (7), we deduce that η ′ (t) = η(t)γ(t). ✷

10
3 Proof of the theorem
We establish our theorem as a special case of a more general result
(Proposition 3.4). The latter deals with certain strengthened regularity
properties (as used earlier in [7] and [3]):
Definition 3.1 Let G be a Lie group modelled on a locally convex space,
with Lie algebra g, and k ∈ N0 ∪ {∞}.

(a) G is called strongly C k -regular if every γ ∈ C k ([0, 1], g) admits an

evolution Evol(γ) ∈ C 1 ([0, 1], G) and the map evol : C k ([0, 1], g) → G,
γ 7→ Evol(γ)(1) is smooth.

(b) G is called C k -regular if each γ ∈ C ∞ ([0, 1], g) has an evolution and the
map evol : (C ∞ ([0, 1], g), OC k ) → G, γ 7→ Evol(γ)(1) is smooth, where
OC k denotes the topology induced by C k ([0, 1], g) on C ∞ ([0, 1], g).
The reader is referred to [8] for a detailled discussion of these regularity
properties (and applications depending thereon). Both C ∞ -regularity and
strong C ∞ -regularity coincide with regularity in the usual sense. If k ≤ ℓ
and G is (strongly) C k -regular, then G is also (strongly) C ℓ -regular.

Remark 3.2 If A is a continuous inverse algebra, we identify the tangent

bundle T (A× ) of the open set A× with A× × A in the natural way. Let
η : [0, 1] → A× be a C 1 -curve and γ : [0, 1] → A be continuous. Then η ′ (t) =
η(t)γ(t) holds in T (A× ) (using η ′ : [0, 1] → T (A× ), and identifying the range
A of γ with {1}×A ⊆ T1 (A× )) if and only if η ′ (t) = η(t)γ(t) holds in A (where
the product simply refers to the algebra multiplication, and η ′ : [0, 1] → A is
the derivative of the A-valued C 1 -curve η).
The next lemma will help us to see that the A-valued map η associated to γ
in Lemma 2.2 actually takes its values in A× , if A is a continuous inverse
algebra. Hence η will be the evolution of γ, by Remark 3.2.
Lemma 3.3 Let A be a continuous inverse algebra, γ : [0, 1] → A be contin-
uous and η : [0, 1] → A as well as ζ : [0, 1] → A be C 1 -curves. Assume that
η(0) = ζ(0) = 1 and

η ′ (t) = η(t)γ(t) and ζ ′(t) = ζ(t)γ(t) for all t ∈ [0, 1]. (10)

If ζ([0, 1]) ⊆ A× , then η = ζ.

11
Proof. Recall from [6, proof of Lemma 3.1] that the differential of the
inversion map ι : A× → A is given by dι(a, b) = −a−1 ba−1 for a ∈ A× and
b ∈ A. As a consequence, the derivative of the C 1 -curve ι ◦ ζ : [0, 1] → A× ,
t 7→ ζ(t)−1 is given by

(ι ◦ ζ)′(t) = −ζ(t)−1 ζ ′(t)ζ(t)−1 . (11)

Now consider the C 1 -curve θ : [0, 1] → A, θ(t) := η(t)ζ(t)−1. Using the

Product Rule, (11) and (10), we obtain

θ′ (t) = η ′ (t)ζ(t)−1 − η(t)ζ(t)−1ζ ′(t)ζ(t)−1

= η(t)γ(t)ζ(t)−1 − η(t)ζ(t)−1 ζ(t)γ(t)ζ(t)−1
= η(t)γ(t)ζ(t)−1 − η(t)γ(t)ζ(t)−1 = 0 .

Hence θ(t) = θ(0) = η(0)ζ(0)−1 = 1 for all t ∈ [0, 1] and thus η = ζ. ✷

Proposition 3.4 Let A be a continuous inverse algebra over K ∈ {R, C},

which satisfies the condition (∗) described in the introduction.

(a) If A is sequentially complete, then A is strongly C 0 -regular and the map

Evol : C 0 ([0, 1], A) → C 1 ([0, 1], A× ) is K-analytic.

(b) If A is Mackey-complete, then A is C 0 -regular and strongly C 1 -regular.

Further, each γ ∈ Lip([0, 1], A) has an evolution Evol(γ) ∈ C 1 ([0, 1], A× ),
and Evol : (Lip([0, 1], A), OC 0 ) → C 1 ([0, 1], A× ) is K-analytic.

Proof. If A is sequentially complete, let X := C([0, 1], A); otherwise, let

X := (Lip([0, 1], A), OC 0 ).
We assume first that K = C. Let Φ : X → C 1 ([0, 1], A) be the mapping pro-
vided by Lemma 2.2. Note that C 1 ([0, 1], A× ) ⊆ C 1 ([0, 1], A) is an identity
neighbourhood, Φ(0) = 1 (cf. (3)) and Φ is C-analytic (see (d) or (d)′ ) and
hence continuous. Therefore, there exists an open 0-neighbourhood Ω ⊆ X
such that Φ(Ω) ⊆ C 1 ([0, 1], A× ). By Proposition 2.2 (c), Evol(γ) := Φ(γ)
is an evolution for γ ∈ Ω. Moreover, evol : Ω → A× , γ 7→ Evol(γ)(1) =
Φ(γ)(1) is C-analytic, since Φ and the continuous linear point evaluation
ev1 : C 1 ([0, 1], A) → A, ζ 7→ ζ(1) are C-analytic.
If A is sequentially complete, Proposition 1.3.10 in [3] now shows that A× is

12
strongly C 0 -regular.8
If A is Mackey-complete, we see as in the proof of [3, Proposition 1.3.10] that
each γ ∈ Lip([0, 1], A) has an evolution Evol(γ) ∈ C 1 ([0, 1], A× ).
In either case, we deduce with Lemmas 2.2 (c) and 3.3 that Evol = Φ. As a
consequence, Evol : X → C 1 ([0, 1], A× ) is C-analytic. Thus (a) is established.
In the situation of (b), note that also evol := ev1 ◦ Evol : Lip([0, 1], A) → A×
is C-analytic. The inclusion maps (C ∞ ([0, 1], A), OC 0 ) → (Lip([0, 1], A), OC 0 )
and C 1 ([0, 1], A) → (Lip([0, 1], A), OC 0 ) being continuous linear and hence C-
analytic, it follows that also the maps evol : (C ∞ ([0, 1], A), OC 0 ) → A× and
evol : C 1 ([0, 1], A) → A× are C-analytic and thus smooth. Hence A× is C 0 -
regular and strongly C 1 -regular.
If K = R, then also the complexification AC of A is a continuous inverse
algebra (see, e.g., [6, Proposition 3.4]) with the same completeness proper-
ties. In (a), we can identify XC with C 0 ([0, 1], AC ); in the situation of (b),
we can identify XC with Lip([0, 1], AC ). For p ∈ P (A), let pC ∈ P (AC ) be
the seminorm defined via
P P
pC (a + ib) := inf j |zj |p(xj ) : a + ib = j zj xj , xj ∈ A, zj ∈ C

for a, b ∈ A (which satisfies max{p(a), p(b)} ≤ pC (a+ib) ≤ p(a)+p(b)). Then

also AC satisfies (∗), as k(µn )C kpC ,qC = kµn kp,q . Let Φ : XC → C 1 ([0, 1], AC )
be the complex analytic map provided by Lemma 2.2 (applied to AC in place
of A). By the complex case just discussed,

Φ = Evol(AC )× : XC → C 1 ([0, 1], (AC )× ).

If γ ∈ X, then Φ(γ) takes only values in the closed vector subspace A of

AC = A ⊕ iA, as is clear from (3). Hence Φ(γ) ∈ C 1 ([0, 1], A) (see [9] or [1,
Lemma 10.1]) and thus Φ(γ) ∈ C 1 ([0, 1], A× ), using the fact that A∩(AC )× =
A× for any unital algebra.9 We deduce that Φ|X : X → C 1 ([0, 1], A× ) is
the evolution map EvolA× of A× . Note that EvolA× is R-analytic, because
Φ : XC → C 1 ([0, 1], A)C is a C-analytic extension of EvolA× . Since evaluation
ev1 : C 1 ([0, 1], A) → A, ζ 7→ ζ(1) is continuous linear and hence R-analytic,
also evolA× := ev1 ◦ EvolA× : X → A× is R-analytic (and hence smooth).
8
Compare already [12, p. 409] and [17, Lemma 3] for similar arguments.
9
If x, a, b ∈ A and x(a + ib) = (a + ib)x = 1, then xa + ixb = 1 and ax + ibx = 1. Hence
xa = ax = 1, i.e., x−1 = a ∈ A.

13
In the situation of (a), this completes the proof. In (b), compose evolA×
with the continuous linear inclusion maps C 1 ([0, 1], A) → Lip([0, 1], A) (resp.,
(C ∞ ([0, 1], A), OC 0 ) → Lip([0, 1], A)) to see that also the evolution map on
C 1 ([0, 1], A) (resp., on (C ∞ ([0, 1], A× ), OC 0 )) is R-analytic and hence C ∞ . ✷

A Proofs for the lemmas from Section 1

It is useful to recall that a locally convex space E is Mackey-complete (in
the sense recalled in the introduction) if and only if every Mackey-Cauchy
sequence in E converges, i.e., every sequence (xn )n∈N in E for which there
exists a bounded subset B ⊆ E and a double sequence (rn,m)n,m∈N of real
numbers rn,m ≥ 0 such that xn − xm ∈ rn,m B for all n, m ∈ N, and rn,m → 0
as both n, m → ∞ (cf. [12, Theorem 2.14]).
Proof of LemmaP 1.1. Given x ∈ U, there exists r ∈ ]1, ∞[ such that
rx ∈ U. Thus ∞ n=0 r n
p n (x) converges and hence C := {r n pn (x) : n ∈ N0 }
is a bounded subset of F . Then also the absolutely convex hull B of C is
bounded. For all n, m ∈ N0 , we have
n+m
X n
X n+m
X n+m
X
−n−1
pk (x) − pk (x) = pk (x) = r r n+1−k r k pk (x)
k=0 k=0 k=n+1 k=n+1
m−1
!
X r −n−1
∈ r −n−1 (1/r)j B⊆ B.
j=0
1 − 1r
Pn
Hence ( k=0 pk (x))n∈N0 is a Mackey-Cauchy sequence in F and hence conver-
gent. Thus f (x) ∈ F . Note that f is complex analytic as a map e
P∞ to1 Fn, by [2,
Theorems 5.1 and 6.1 (i)]. Hence, if x ∈ U, then f (x + y) = n=0 n! δx (f )(y)
for all y in some 0-neighbourhood, where δxn f (y) := d(n) f (x, y, . . . , y) is the
n-th Gâteaux differential of f at x. Given y ∈ E, there is s > 0 such that
x + zy ∈ U for all z ∈ C such that |z| ≤ s. For each n ∈ N0 , Cauchy’s
Integral Formula for higher derivatives now shows that
Z 2π
n n! f (x + seit y)
δx (f )(y) = sieit dt,
2πi 0 (seit )n+1
which lies in F since the integrand is a Lipschitz curve in F and F is Mackey-
complete.10 Hence each δxn (f ) is a continuous homogeneous polynomial from
10
The integrand is a C ∞ -curve in Fe and hence a Lipschitz curve in Fe, with image in F .

14
E to F and thus f is complex analytic as a map from E to F . ✷
Proof of Lemma 1.2. Let p be a continuous seminorm on F and k ∈ N0
such that k ≤ r. Then q := p ◦ α is a continuous seminorm on E. Let
γ ∈ C r ([0, 1], E). For each j ∈ N0 such that j ≤ k, we have (α◦γ)(j) = α◦γ (j)
and thus k(α ◦ γ)(j) kC 0 ,p = kα ◦ γ (j) kC 0 ,p = kγ (j) kC 0 ,p◦α = kγ (j) kC 0 ,q , entailing
that kα ◦ γkC k ,p = kγkC k ,q . Hence α∗ is continuous.
If α is an embedding and Q a continuous seminorm on C r ([0, 1], E), then
there exists k ∈ N0 such that k ≤ r and a continuous seminorm q on E
such that Q ≤ k.kC k ,q . Because α is an embedding, there exists a continuous
seminorm p on F such that p(α(x)) ≥ q(x) for all x ∈ E (because α−1 is
continuous linear). Hence k(α ◦ γ)(j) kC 0 ,p = kγ (j) kC 0 ,p◦α ≥ kγ (j) kC 0 ,q for each
j ∈ N0 such that j ≤ k and thus kα ◦ γkC k ,p ≥ kγkC k ,q ≥ Q(γ), entailing
that α∗ is a topological embedding. ✷
Proof of Lemma 1.3. Let p ∈ P (E) and k ∈ N0 such that k ≤ r. Since
p = k.kp ◦ πp , we have

k(πp ◦ γ)(j) kC 0 ,k.kp = kπp ◦ γ (j) kC 0 ,k.kp = kγ (j) kC 0 ,k.kp ◦πp = kγ (j) kC 0 ,p

for each γ ∈ C r ([0, 1], E) and j ∈ {0, 1, . . . , k}, entailing that kπp (γ)kC k ,k.kp =
kγkC k ,p . The assertion follows. ✷

Remark A.1 Before we turn to the proof of Lemma 1.4, it is useful to record
some simple observations.

(a) It is clear from the definitions that the map

h : C k ([0, 1], E) → C([0, 1], E) × C k−1 ([0, 1], E) , γ 7→ (γ, γ ′ )

is linear and a homeomorphism onto its image, for each k ∈ N.

(b) The
R t image im(h) of h consists of all (γ, η) such that γ(t) = γ(0) +
0
η(s) ds for each t ∈ [0, 1]. Since point evaluations and the linear maps
Rt Rt
η 7→ 0 η(s) ds (with p( 0 η(s) ds) ≤ kηkC 0 ,p ) are continuous, it follows
that im(h) is a closed vector subspace of C([0, 1], E) × C k−1 ([0, 1], E).

Proof of Lemma 1.4. Because direct products of Mackey-complete

locally convex spaces are Mackey-complete, and so are closed vector
subspaces, also projective limits of Mackey-complete locally convex spaces are

15
Mackey-complete. Since C ∞ ([0, 1], E) = lim C k ([0, 1], E) (with the
←−
respective inclusion maps as the limit maps), we therefore only need to prove
Mackey-completeness if k := r ∈ N0 . Likewise in the case of completeness.
The case k = 0. If E is complete, then also C([0, 1], E) is complete, as
is well known (cf. [11, Chapter 7, Theorem 10]). If E is merely Mackey-
complete, let E e be a completion of E which contains E. Then C([0, 1], E) e
is complete. The inclusion map φ : C([0, 1], E) → C([0, 1], E) e is a topolog-
ical embedding, by Lemma 1.2. If (γn )n∈N is a Mackey-Cauchy sequence
in C([0, 1], E), then (φ ◦ γn )n∈N = (γn )n∈N is a Mackey-Cauchy sequence in
e hence convergent to some γ ∈ C([0, 1], E).
C([0, 1], E), e For each t ∈ [0, 1],
the point evaluation εt : C([0, 1], E) e → E,
e η 7→ η(t) is continuous and linear.
Hence (γn (t))n∈N is a Mackey-Cauchy sequence in E and hence convergent
in E. Since γn (t) = εt (γn ) → εt (γ) = γ(t), we deduce that γ(t) ∈ E. Hence
γ ∈ C([0, 1], E) and it is clear that γn → γ in C([0, 1], E).
Induction: If C k−1 ([0, 1], E) is (Mackey-) complete, then also C k ([0, 1], E),
as it is isomorphic to a closed vector subspace of the (Mackey-) complete
direct product C([0, 1], E) × C k−1 ([0, 1], E) (see Remark A.1 (b)). ✷
Acknowledgement. The research was supported by DFG, grant GL 357/5-1.

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] Bochnak, J. and J. Siciak, Analytic functions in topological vector spaces,
Studia Math. 39 (1971), 77–112.
[3] Dahmen, R., “Direct Limit Constructions in Infinite-Dimensional Lie
Theory,” Doctoral Dissertation, Universität Paderborn, May 2011; see
urn:nbn:de:hbz:466:2-239
[4] Glöckner, H., Lie groups without completeness restrictions, Banach Center
Publ. 55 (2002), 43–59.
[5] Glöckner, H., Lie group structures on quotient groups and universal com-
plexifications for infinite-dimensional Lie groups, J. Funct. Anal. 194 (2002),
347–409.
[6] Glöckner, H., Algebras whose groups of units are Lie groups, Studia Math.
153 (2002), 147–177.

16
 [7] Glöckner, H., Direct limits of infinite-dimensional Lie groups, pp. 243–280 in:
K.-H. Neeb and A. Pianzola (Eds.), “Developments and Trends in Infinite-
Dimensional Lie Theory,” Progr. Math. 288, Birkhäuser, Boston, 2011.
[8] Glöckner, H., Regularity in Milnor’s sense for direct limits of infinite-
dimensional Lie groups, manuscript in preparation.
[9] Glöckner, H. and K.-H. Neeb, “Infinite-dimensional Lie Groups,” Vol. I, book
in preparation.
[10] Hervé, M., ”Analyticity in Infinite-Dimensional Spaces,” de Gruyter, 1989.
[11] Kelley, J. L., “General Topology,” Springer, New York; reprint of 1955 ed.
[12] Kriegl, A. and P. W. Michor, “The Convenient Setting of Global Analysis,”
AMS, Providence, 1997,
[13] Michael, E. A., Locally multiplicatively-convex topological algebras, Mem.
Amer. Math. Soc. 11 (1952), 79 pp.
[14] Michor, P. W., “Manifolds of Differentiable Mappings,” Shiva Publ., Orping-
ton, 1980.
[15] Milnor, J., Remarks on infinite-dimensional Lie groups, pp. 1007–1057 in:
B. S. DeWitt and R. Stora (eds.), “Relativité, groupes et topologie II,” North-
Holland, Amsterdam, 1984.
[16] Neeb, K.-H., Towards a Lie theory of locally convex groups, Jpn. J. Math. 1
(2006), 291–468.
[17] Robart, Th., On Milnor’s regularity and the path-functor for the class of
infinite dimensional Lie algebras of CBH type, Algebras Groups Geom. 21
(2004), 367–386.
[18] Turpin, Ph., Une remarque sur les algèbres à inverse continu, C. R. Acad.
Sci. Paris Sér. A-B 270 (1970), A1686–A1689.
[19] Waelbroeck, L., Les algèbres à inverse continu, C. R. Acad. Sci. Paris 238
(1954), 640–641.
[20] Weil, A., “L’integration dans les groupes topologiques et ses applications,”
Hermann, Paris, 2nd edition, 1951.

Helge Glöckner, Universität Paderborn, Institut für Mathematik,

Warburger Str. 100, 33098 Paderborn, Germany; [email protected]
Karl-Hermann Neeb, FAU Erlangen-Nürnberg, Department Mathematik,
Cauerstr. 11, 91058 Erlangen, Germany;
[email protected]

17