FRAGMENTABILITY AND REPRESENTATIONS OF FLOWS

MICHAEL MEGRELISHVILI
Abstract. Our aim is to study weak

continuous representations of semigroup actions

into the duals of good (e.g., reexive and Asplund) Banach spaces. This approach leads
to ow analogs of Eberlein and Radon-Nikodym compacta and a new class of functions
(Asplund functions) which intimately is connected with Asplund representations and in-
cludes the class of weakly almost periodic functions. We show that a ow is weakly almost
periodic i it admits suciently many reexive representations. One of the main technical
tools in this paper is the concept of fragmentability (which actually comes from Namioka
and Phelps) and widespreadly used in topological aspects of Banach space theory. We
explore fragmentability as a generalized equicontinuity of ows. This unied approach
allows us to obtain several dynamical applications. We generalize and strengthen some
results of Akin-Auslander-Berg, Shtern, Veech-Troallic-Auslander and Hansel-Troallic. We
establish that frequently, for linear G-actions, weak and strong topologies coincide on, not
necessarily closed, G-minimal subsets. For instance such actions are orbitwise Kadec.
1. Introduction
Every compact jointly continuous G-ow X admits a faithful weak

continuous Banach
representation. More precisely, X is G-embedded into the dual ball B(V

) as a weak

compact G-subset of some Banach space V , where the group G acts continuously on V by
linear isometries. Indeed, this is a standard fact (see Telemans paper [53], or for a more
detailed discussion, the survey [47]) for V = C(X), where one can identify x X with the
point mass
x
C(X)

. The geometry of C(X), in general, is bad. For example, a very

typical disadvantage here is the norm discontinuity of the dual action of G on C(X)

. One
of the results of [36] guarantees (see also Corollary 8.7 below) the norm continuity of the
dual action of the group G on V

provided that V is Asplund. Recall that a Banach space

V is Asplund i the dual A

is separable for every separable Banach subspace A of V .

The following general question arises: how good can a Banach space V be among all
possible w

-continuous faithful G-linearizations of X into V

? For instance when can V

be chosen Asplund or reexive ? We show that the reexive case (for second countable X)
can be reduced completely to the question if X is a weakly almost periodic (in short: wap)
ow.
Eberlein compact in the sense of Amir and Lindenstrauss [3] is a compact space which
can be embedded into (V, weak) for some Banach space V . It is well known [12] that a
compact space X is Eberlein i it can be embedded into the unit ball (B(V ), weak) of some
reexive space V . If X is a weak

compact subset in the dual V

of an Asplund space V
then, following Namioka [44], X is called Radon-Nikodym compact (in short: RN). Every
reexive Banach space is Asplund. Hence, every Eberlein compact is RN.
Now introduce map versions of these concepts. Let f : X X be a selfmap on a
compact space X. Let us say that f is an Eberlein (Radon-Nikodym) map if it admits a
weak

linearization into certain reexive (resp.: Asplund) Banach space. That is, there
Date: June, 2005.
Key words and phrases. fragmentability, ow, semigroup compactication, weakly almost periodic, Eber-
lein compact, Radon-Nikodym compacta, Asplund space, Asplund function, Kadec property.
2000 Math. Subject Classication. 54H15, 54H20, 43A60, 22A25.
1
2
exists a reexive (Asplund) Banach space V and a weak

embedding X B(V

) in such
a way that f : X X is a restriction of the adjoint F

: V

of some linear operator

F : V V which is non-expansive ( |F| 1). In this point of view, the space X is
Eberlein or RN i the identity mapping 1
X
: X X is Eberlein or RN, respectively.
Clearly, every metric compact space is Eberlein since it is a compact subset of the Hilbert
space l
2
. In contrast, even simple maps on metric compacta can be non-Eberlein. For
example, the f(x) = x
2
map on the closed interval [0, 1] is not Eberlein. The map
f : T
2
T
2
, f([a], [b]) = ([a +b], [b +

2])
dened on the torus X = T
2
is not even RN (see Example 7.19).
It is signicant that a compact metric cascade (Z, X) is wap (equivalently, is an Eberlein
ow, by virtue of Corollary 4.10) i the generating selfhomeomorphism f : X X leads
to a wap Markov operator T
f
: C(X) C(X) (see Downarowicz [14] and the references
there). The study of wap operators and corresponding cascades goes back to the 60s (K.
Jacobs, B. Jamison, M. Rosenblatt, R. Sine, J. Montgomery, E. Thomas and others).
The setting of maps and their linearizations admits a natural generalization in terms of
ow linearizations. We introduce below Eberlein and Radon-Nikodym ows and show that
a ow is weakly almost periodic in the sense of Ellis-Nerurkar [16] i it is a subdirect product
of Eberlein ows. Investigation of RN ows naturally leads also to a new class of functions
which we call Asplund functions. Our approach emphasizes more the similarities (rather
than the dierences) between wap and Asplund functions. We show that a function is wap
(Asplund) i it comes from a matrix coecient dened by a representation into reexive
(resp.: Asplund) spaces. In both cases our method is based on corresponding dualities and
a factorization procedure by Davis, Figiel, Johnson and Pelczynski [12]. In the Asplund
case the technical part uses a modication (using Asplund subsets instead of weakly
compact) which is due to Stegall [52].
Let us briey describe one of the ideas explored in the present paper. Suppose that X is a
subow (under some action by linear isometries) of a weak

compact dual ball B(V

) of some
Banach space V . One of the important questions in Banach space theory is a relationship
between norm and weak

topologies on X V

. In the absolute case of the coincidence,

we say that X is a Kadec subset of V

. In such cases, X, as a ow, is equicontinuous.

Conversely, every compact metric equicontinuous ow X admits a ow representation in
such a way that X becomes a norm compact subow (and hence a Kadec subset) of a
suitable B(V

). In general, as an attempt to measure the level of equicontinuity, we can

ask how close can two natural topologies on X inherited from V

be. A more concrete

and exible enough question is: for what ow representations is the natural mapping 1
X
:
(X, weak

) (X, norm) fragmented in the sense of [45, 30]. The latter means that every
nonempty subset of X admits relatively weak

open nonempty subsets with arbitrarily small

diameters.
The great advantage of Asplund spaces is the (weak

, norm)-fragmentability of bounded
subsets in their duals [45, 44]. Many modern investigations in Banach spaces concern
Asplund spaces, the notion of fragmentability and closely related Radon-Nikodym property
(see [9, 44, 13, 17, 6] and the references therein). In [36, 38] we study some dynamical
applications of fragmentability. In the present paper we examine further developments
exploring some ideas more familiar in the topological aspects of Banach space theory.
For the convenience of the reader we have tried to make the exposition self-contained.
Acknowledgments: I would like to express my gratitude to M. Fabian, E. Glasner, A.
Leiderman, V. Pestov and V. Uspenskij for helpful comments and suggestions. The main
results of this paper were presented at the 9th Prague Topological Symposium (August,
2001) and also at the Auckland Summer Topological Conference (July, 2002). I would
3
like to thank the organizers for their kind invitation and hospitality. This work is supported
by ISF grant no. 4699.
2. Preliminaries
The closure and the interior operators in topological spaces will be denoted by cl and
int, respectively. If A is a subset in a Banach space then sp(A) is the linear span of A.
Let be a uniform structure on a set X. Its induced topology on X will be denoted by
top(). A uniformity on a topological space (X, ) is said to be compatible if top() = .
A (left) ow (S, X) consists of a topologized semigroup S and a (left) action : SX X
on a topological space X. We reserve the symbol G for the case when S is a group. As
usual we write simply sx instead of (s, x) = s(x) = x(s). Action means that always
s
1
(s
2
x) = (s
1
s
2
)x. If S is a monoid, we assume that the identity e of S acts as the identity
transformation of X. Every x X denes an orbit map x : S X, s sx. Say that a
topologized semigroup S is: (a) left (right) topological; (b) semitopological; (c) topological
if the multiplication function S S S is left (right) continuous, separately continuous, or
jointly continuous, respectively. Let S be a semitopological semigroup. A left ow (S, X)
is said to be a semitopological ow if the action is separately continuous.
A right ow (X, S) can be dened analogously. If S
opp
is the opposite semigroup of S
with the same topology then (X, S) can be treated as a left ow (S
opp
, X) (and vice versa).
If not stated otherwise the ows below are assumed to be semitopological. Compact will
mean compact and Hausdor.
Let h : S
1
S
2
be a semigroup homomorphism, S
1
act on X
1
and S
2
on X
2
. A map
f : X
1
X
2
is said to be h-equivariant if f(sx) = h(s)f(x) for every (s, x) S
1
X
1
. For
S
1
= S
2
with h = 1
S
, we say S-map. The map h : S
1
S
2
is an antihomomorphism i
h : S
1
S
opp
2
(the same assignement) is a homomorphism.
An S-compactication of (S, X) is a continuous S-map : X Y with a dense range
( S-compactication map) into a compact S-ow Y. A (jointly continuous) ow (S, X)
is said to be (resp.:joint continuously) compactiable if there exists an S-compactication
: X Y into a (jointly continuous) S-ow Y such that is a topological embedding.
Following Junghenn [31] we dene a bicompactication m = (h, ) : (S
1
, X
1
) (S
2
, X
2
) as
a pair h : S
1
S
2
, : X
1
X
2
, where (S
2
, X
2
) is a semitopological ow with compact
S
2
and X
2
, the map h is a continuous homomorphism and is a continuous h-equivariant
map with a dense range.
Let V be a Banach space with the dual V

. Set
B(V ) = v V : [[v[[ 1 and (V ) = L(V, V ) : [[[[ 1.
In most cases we endow the sets B(V ), B

= B(V

) and (V ) with weak, weak

and
weak operator topologies, respectively. Sometimes we use the subscripts w and w

.
The subscript s will mean the strong operator topology. The pairs ((V )
w
, B(V )
w
)
and (B(V

)
w
, (V )
w
) are semitopological ows. The (V )
opp
s
-ow B(V

)
w
is jointly
continuous as it follows directly from Fact 2.2. It induces a right action of the isometry
group Is(V ) = g Aut(V ) : [[g[[ = 1. Alternatively, we have a left action dened by
Is(V ) B

, (gf)(v) = f(g
1
v)
Hence, (Is(V )
s
, B

w
) is a well dened jointly continuous action (see also Remark 3.4).
The Banach algebra of all continuous real valued bounded functions on a topological
space X will be denoted by C(X). The same set with the pointwise topology (p-topology) is
denoted by C
p
(X). Let X be a (left) S-ow then it induces the antihomomorphism h : S
C(X) and the corresponding (right) action C(X) S C(X) where (fs)(x) = f(sx). In
4
the case of a topological group S = G, we can dene a homomorphism and a left action by
(gf)(x) = f(g
1
x). While the translations are continuous, the orbit maps

f : S C(X)
are not necessarily (even weakly) continuous. Denote by RUC
S
(X) the set of all functions
f C(X) such that the orbit map

f is norm continuous. If we require only weak continuity,
then we get the denition of weakly right uniformly continuous functions (see [8]). Denote
the corresponding set by WRUC
S
(X) .
The proof of the following fact is straightforward.
Fact 2.1. If X is compact, then : SX X is jointly continuous i C(X) = RUC
S
(X).
For general (separately continuous) action , the set RUC
S
(X) is an S-invariant Banach
subalgebra and the corresponding Gelfand compactication u
R
: X X
R
is a universal
(maximal) jointly continuous S-compactication of X. If X = S with the left regular
action of S, then we simply write RUC(S). If S = G is a topological group, then RUC(G)
is the set of all usual right uniformly continuous functions. The algebra of all left uniformly
continuous functions (dened for the right regular action of S on S)) will be denoted by
LUC(S).
The classical Gelfand-Naimark 1-1 correspondence between Banach subalgebras of C(X)
and the compactications of X can be extended to the category of jointly continuous S-
ows using Banach S-subalgebras of RUC
S
(X) (like the well-known results for topological
group actions (see J. de Vries [57])). One of the ways to verify this is to use the following
fact which is a key idea of Telemans above-mentioned result, as well as in the paper of
Uspenskij [55].
Fact 2.2. Let V be a Banach space. Suppose that a topologized semigroup S acts on V
from the right by linear non-expansive operators. The following are equivalent:
(i) V S V is norm jointly continuous.
(ii) The induced ane action S B

w
B

w
is jointly continuous.
Proof. The dual action denes an injective antihomomorphism of (V ) into C(B

, B

).
Now observe that the strong operator topology on (V ) coincides with the compact open
topology inherited from C(B

, B

).
Recall the denition of weakly almost periodic functions and some relevant facts.
Denition 2.3. Let S be a semitopological semigroup and X be an S-ow.
(i) A function f C(X) is said to be weakly almost periodic, (wap, in short) if the orbit
fS = fs : s S is relatively weakly compact in C(X). Write f WAP
S
(X).
(ii) We say that X is S-wap, or, (S, X) is wap (otation: X [wap]
S
) if WAP
S
(X)
separates points and closed subsets of X.
(iii) We say that S is wap (and write: S [wap]) if the regular left action (S, S) is wap.
This general form of denition (i) can be found in the work of Junghenn [31]. For the
left action (S, S) we get the classical notion of wap functions on S (see Eberlein [15] and
de-Leeuw Glicksberg [34]). We use the notation WAP(S) instead of WAP
S
(S).
Replacing weakly compact in Denition 2.3 by norm compact we get the denitions
of almost periodic functions and corresponding S-algebras AP
S
(X), AP(S).
Grothendiecks criteria [22] for relative weak compactness leads to the following assertion.
Fact 2.4. (Grothendiecks DLP) A function f C(X) dened on some S-ow is wap i
the following Double Limit Property is satised:
(DLP) For every pair of sequences s
m
S and x
n
X
lim
m
lim
n
f(s
n
x
m
) = lim
n
lim
m
f(s
n
x
m
)
holds whenever both of the limits exist.
5
Recall also the following very useful fact.
Fact 2.5. (Grothendiecks Lemma) Let X be a compact space. Then a bounded subset A of
C(X) is w-compact i A is p-compact.
The set WAP
S
(X) is a Banach S-subalgebra in C(X). This is mentioned in [31]. The
proof can be done using Fact 2.5 and the Eberlein-Smulian theorem.
This implies that our general Denition 2.3(ii), for compact X, is equivalent to the
denition of wap ows in the sense of Ellis-Nerurkar [16]. Gelfands compactication
u
W
: X X
W
induced by the algebra WAP
S
(X) is the universal wap compactica-
tion of X (see for details [31, Theorem 3.1]). In particular, for the left regular action (S, S)
we get the universal wap semigroup compactication u
W
: S S
W
. It is important to note
that in this case S
W
is a compact semitopological semigroup and enjoys the corresponding
universality property. By our denitions, the ow X (or, the semigroup S) is wap i u
W
is
a topological embedding.
Ellis semigroup E(S, X) (or, simply: E(X)) for compact X is the pointwise closure of
the set of all s-translations s : X X : s S in the compact semigroup X
X
. Denote
by : S E(X), (s) = s the corresponding natural homomorphism. In general, E(X)
is only right topological, that is, only the right translations E(X) E(X), s sp are
necessarily continuous.
Fact 2.6. [16, 31] For a compact S-ow X the following are equivalent:
(1) X is wap (that is, C(X) = WAP
S
(X)).
(2) Each element of E(S, X) is continuous (quasiequicontinuous in terms of [8]).
(3) The pair (E(X), X) is a semitopological ow with the compact semitopological semi-
group E(X).
(4) There exists a bicompactication (h, ) : (S, X) (P, Y ) such that : X Y
is an embedding and h : S P is a semigroup compactication into a compact
semitopological semigroup P.
Proof. The principal implication (1) (2) directly follows from Proposition 4.3 and Re-
marks 4.4(b).
For (4) (1) it suces to show that C(Y ) = WAP
P
(Y ). Let f C(Y ) then the P-orbit
fP is bounded. Since P is compact then fP is pointwise compact in C(Y ). By Fact 2.5,
fP is even w-compact. Thus, f WAP
P
(Y ).
Other implications are trivial.
For compact X, Denition 2.3(ii) agrees with the item (1) in Fact 2.6, as it easily follows
by Stone-Weierstrass theorem.
If X [wap]
S
then Y [wap]
P
for every subsemigroup P of S and every P-subow Y of
X. Moreover, [wap]
S
is closed under subdirect products (subspaces of products). The class
of compact S-wap ows is closed also under quotients.
Fact 2.7. (i) WAP
S
(X) WRUC
S
(X) for every semitopological S-ow X. Hence,
WAP(S) WRUC(S) for every semitopological semigroup S.
(ii) If G is a semitopological group then WAP
G
(X) RUC
G
(X). In particular, WAP(G)
RUC(G) holds.
Proof. (i) The orbit map

f : S C(X) is clearly p-continuous. If f WAP
S
(X) then
cl
w
(fS) is weakly compact. Hence, (fS, w) = (fS, p). Therefore

f is also weakly continuous.
(ii) Follows from Theorem 2.5 and Remark 2.6 (a) of [38]. It can be seen easily also as a
corollary of Theorem 8.5 below.
6
The inclusion WAP(G) RUC(G) is well known (see for example [24] or [8, Theorem
4.10]). Another proof of the inclusion WAP
G
(X) RUC
G
(X) can be derived also by
results of [28].
For every reexive Banach space V the semigroup (V ) is a weakly compact semitopo-
logical semigroup [34]. Observe that for every vector v V with norm 1, the orbit (V )v of
v coincides with B(V ). This guarantees the converse: if (V )
w
is compact then B
w
is com-
pact , and, hence, V is necessarily reexive. For every reexive V , the ows ((V ), B(V ))
and ((V )
opp
, B(V

)) are semitopological and (bi)compact. Hence, wap by Fact 2.6.

One of our applications below (see section 8) provides a simple proof of the following
important theorem of Lawson [32] which in itself is a generalization of Ellis theorem.
Fact 2.8. (Ellis-Lawsons Joint Continuity Theorem).
Let G be a subgroup of a compact semitopological monoid S. Suppose that (S, X) is a
semitopological ow with compact X. Then the action G X X is jointly continuous
and G is a topological group.
A (not necessarily compact) G-ow X is said to be minimal if every orbit Gx is dense
in X. Equicontinuous compact ows are the simplest one in Topological Dynamics. Every
equicontinuous compact ow is wap. The converse is true for every minimal compact wap
G-ow X [54, 4]. Below we show (Theorem 6.10) that the compactness assumption is
superuous here. That is, every minimal wap (and even, RN-approximable), not necessarily
compact, G-ow is equicontinuous.
3. Banach representations and matrix coefficients
Let V be a Banach space with the canonical duality <, >: V V

R. If a semigroup
S acts from the right on V (equivalently: if we have an antihomomorphism S L(V, V ))
then it induces a left action of S on the dual V

such that < vs, >=< v, s > for every

v V and V

.
Denition 3.1. A (non-expansive) V -representation of a ow (S, X) is an equivariant pair
(h, ) : (S, X) ((V )
opp
, B

)
where h : S (V )
opp
is a weak continuos homomorphism (equivalently: antihomo-
morphism S (V )) and : X B

is weak

continuous and equivariant, that is

(sx) = h(s)(x).
We say that a representation is strongly continuous if h : S (V )
s
is continuous.
Topologically faithful (or, simply: faithful) will mean that : X (B

, w

) is a topological
embedding.
Let / BAN be a subclass of Banach spaces. We say that a ow (S, X) is:
(a) /-representable if there exists a faithful V -representation of (S, X) for some V /.
(b) /-approximable if there exists a system (h
i
,
i
) of representations of (S, X) in V
i
separating points and closed subsets in X with V
i
/ (equivalently, if X is a
subdirect product of /-representable S-ows).
(c) Eberlein if it is REFL-representable.
(d) Radon-Nikodym (in short: RN) if it is ASP-representable.
(e) RN-approximable if it is ASP-approximable.
In this denition REFL and ASP mean the classes of all reexive and Asplund spaces
respectively. Since REFL ASP, every Eberlein ow is RN. If S is a trivial monoid and
X is compact, then the denitions (c) and (d) give exactly the classical notions of Eberlein
and RN compacta mentioned in the introduction.
7
Remark 3.2. Sometimes weak continuous (anti)homomorphisms automatically are strongly
continuous. This happens for instance if either: (a) S is an arbitrary semitopological group
and V is reexive; (b) S is a locally compact Hausdor topological group; or (c) S is a
topological group metrizable by a complete metric. The rst assertion follows from [38,
Theorem 2.8] (or, from Corollary 8.2 below). For the last two assertions see [8]. For some
other results of the nature weak implies strong see also [24, 25, 33, 36, 38].
The following standard fact (see for example [53]) states actually that every jointly con-
tinuous action on compact spaces admits a faithful Banach representation.
Lemma 3.3. Let (S, X) be a jointly continuous semigroup action on a compact X. Then
there exists a Banach space V and a faithful strongly continuous representation (h, ) of
(S, X) into the jointly continuous ane action ((V )
opp
s
, B(V

)
w
).
Proof. Take V = C(X) and dene the antihomomorphism h : S (V ) induced by
the natural right action C(X) S C(X). This action is norm continuous by Fact 2.1
because RUC
S
(X) = C(X). Thus, h is strongly continuous by Fact 2.2. Finally dene the
natural weak

embedding : X (B(C(X)

) identifying each x X with the point mass

x
B(C(X)

).
For every weakly continuous antihomomorphism h : S L(V, V ) and every chosen pair
of vectors v V and V

, there exists a canonically associated (generalized) matrix

coecient m
v,
: S R, s < vs, >=< v, s >
S
m
v,
-
R
L(V, V )
h
?
v
-
V

6
Remark 3.4. In many important cases we can use homomorphisms instead of antihomomor-
phisms. Indeed, if S is a topological group (or, a semigroup with a continuous involution),
then we can dene a homomorphism h

: S L(V, V ), s h(s
1
) and redene the function
m
v,
by s < s
1
v, >.
It is natural to expect that matrix coecients reect good properties of ow representa-
tions (see, for example, [47]). We recall two well-known facts. The rst example is the case
of Hilbert representations. If h : G Is(H) is a group representation into Hilbert space H
and = v, then the corresponding map g < g
1
v, v > is a positive denite function on
G. The converse is also true: every continuous positive denite function comes from some
continuous Hilbert representation. Every positive denite function is wap (see [11]).
The second example comes from Eberlein [15] (see also [8, Examples 1.2.f]). If V is
reexive, then every bounded V -representation (h, ) and arbitrary pair (v, ) lead to a
weakly almost periodic function m
v,
on S. This follows easily by the (weak) continuity of
the natural operators dened by the following rule. For every xed V

(v V ) dene
introversion type operators by
L

: V C(S) and R
v
: V

C(S), where L

(v) = R
v
() = m
v,
.
We say that a vector v V is strongly (weakly) continuous if the corresponding orbit
map v : S V, v(s) = vs, dened through h : S (V ), is strongly (weakly) continuous.
Fact 3.5. Let h : S L(V, V ) be a weakly continuous antihomomorphism with the norm
bounded range. Then
(1) L

: V C(S) (and R
v
: V

C(S)) are linear bounded S-operators between right

(left) S-actions.
8
(2) If (resp.: v V ) is norm continuous, then m
v,
is left (resp.: right) uniformly
continuous on S.
(3) If V is reexive, then m
v,
WAP(S).
Proof. (1) Is straightforward.
(2) Since h(S) is norm bounded, sup[[vt[[ : t S = c < . Let be a norm continuous
vector. In order to establish that m
v,
LUC(S), observe that
[m
v,
(ts) m
v,
(ts
0
)[ = [ < vts, > < vts
0
, > [ =
[ < vt, s > < vt, s
0
> [ [[vt[[ [[s s
0
[[ c [[s s
0
[[.
Similar verication is valid for the second case.
(3) If the orbit vS is relatively weakly compact in V (e.g., if V is reexive), then the
same is true for L

(vS) = m
v,
S in C(S). Thus m
v,
is wap.

Fact 3.6. Let (h, ) : (S, X) ((V )

opp
, B

) be an equivariant pair with weak

continuous
but without no continuity assumptions on h.
(i) The map T : V C(X), v T(v), where T(v) : X R is dened by
T(v)(x) =< v, (x) >
is a linear S-operator (between right S-actions) with [[T[[ 1.
(ii) T(v
0
) RUC
S
(X) for every strongly continuous vector v
0
in V . Hence, if h is
strongly continuous then T(V ) RUC
S
(X).
(iii) If V is reexive, then T(V ) WAP
S
(X).
Proof. (i) is straightforward.
(ii) Observe that [[(x)[[ 1 for every x X. We get
[[T(v
0
)s T(v
0
)s
0
[[ = sup[ < v
0
s v
0
s
0
, (x) > [ : x X
[[v
0
s v
0
s
0
[[ [[(x)[[ [[v
0
s v
0
s
0
[[.
This implies that T(v
0
) RUC
S
(X).
(iii) If V is reexive, the orbit vS is relatively weakly compact for each v V . By the
(weak) continuity of the S-operator T, the same is true for the orbit of T(v) in C(X).
Therefore we get T(v) WAP
S
(X).
Proposition 3.7. For every S-ow X the following are equivalent:
(1) f RUC
S
(X).
(2) There exist: a Banach space V , a strongly continuous antihomomorphism h : S
(V ), a weak

continuous equivariant map : X B

, and a vector v V such

that f(x) =< v, (x) > (that is f = T(v).
Proof. (1) = (2) The function f belongs to an S-invariant Banach subalgebra / of
RUC
S
(X). The right action of S on V := / is jointly continuous. Then by Fact 2.2,
corresponding left action of S on the dual ball (B

, w

) is jointly continuous. Then the

naturally associated map : X B

and the vector v := f satisfy the desired property.

(1) = (2) Immediate by Fact 3.6 (ii).
Proposition 3.8. For every semitopological monoid S the following are equivalent:
(1) f RUC(S).
(2) There exist: a Banach space V , a strongly continuous antihomomorphism h : S
(V ), and a pair of vectors v V and V

such that f = m
v,
.
9
Proof. (1) =(2) Consider the Gelfand compactication u
R
: S S
R
dened by RUC(S) =
C(S
R
). Then the action S S
R
S
R
is jointly continuous by Fact 2.1. Now dene:
V := C(S
R
), corresponding strongly continuous h : S (V ) (induced by the right action
of S on C(S
R
)), v := f V and = u
R
(e) V

.
(1) = (2) Immediate by Fact 3.5.2.
As we already have seen a right uniformly continuous function can be represented as a
matrix coecient m
v,
of some strongly continuous Banach representation. We mentioned
also the well known case of Hilbert representations. A positive denite function on a topo-
logical group G is exactly a matrix coecients of some unitary representation. One of our
aims is to understand the role of matrix coecients for intermediate cases of reexive and
Asplund representations. We show that wap functions are exactly the reexive matrix co-
ecients. In the Asplund case this approach leads to a denition of Asplund functions
introduced in Section 7.
4. Reflexive representations of flows
Denition 4.1. A (bounded) duality is a separately continuous (resp., bounded) mapping
<, >: Y X R. We say that the duality is right strict if the corresponding continuous
map q
X
: X C
p
(Y ), q
X
(x) =< y, x > is a topological embedding (e.g., an injection if
X is compact).
The left version can be dened analogously. Then strict will mean left and right
strict simultaneously.
Let a semigroup S act on X and Y by the following actions:

X
: S X X,
Y
: Y S Y.
The duality is an S-duality (or, S-invariant) if < ys, x >=< y, sx >.
Consider two typical examples:
(1) Canonical reexive duality: rBB

R with compact spaces rB and B

(under
weak topologies) is dened for every reexive V , a positive number r > 0 and an
antihomomorphism h : S (V ). In particular, we can choose the natural action
of S = (V )
opp
. Observe that (V )
opp
= (V

) (compare Proposition 4.3).

(2) Let K V be a weakly compact S-invariant subset in a Banach space V with
respect to some antihomomorphism h : S (V ). Then K B(V

) R is a left
strict S-duality.
Lemma 4.2. Let <, >: Y X R be an S-duality.
(1) <, > is left strict i a net y
i
converges to y in Y exactly when < y
i
, x >< y, x >
in R for every x X. Similarly, <, > is right strict i a net x
i
converges to x in
X exactly when < y, x
i
>< y, x > in R for every y Y .
(2) Let <, > be a left (right) strict S-duality. Then all s-translations s : Y Y (resp.,
s : X X) are continuous.
(3) Let <, > be a strict S-duality. Then
Y
is separately continuous i
X
is separately
continuous.
(4) Let <, >: Y X R be a left strict S-duality. Then it can be reduced canonically
to the naturally associated strict S-duality <, >
q
: Y X
q
R. If
Y
is separately
continuous then
X
q
is also separately continuous.
10
Proof. (1) Follows from the net characterization of the product topology.
(2) We have to show that every s-translation
s
Y
: Y Y is continuous (the case of

s
X
is similar). Let y
i
y. In order to show that sy
i
sy, it suces by (1) to check
that < y
i
s, x >< ys, x > for each x X. Or, equivalently, we have to show that
< y
i
, sx >< y, sx >. The latter follows from the assumption y
i
y and the separate
continuity of <, >.
(3) Similar to the proof of (2).
(4) Consider the canonical continuous map q
X
: X C
p
(Y ) and the corresponding range
X
q
= q
X
(X) C
p
(Y ). Dene
<, >
q
: Y X
q
R, < y, q
X
(x) >
q
:=< y, x > .
It is easy to show that this is a well-dened strict duality. Moreover, the action of S on X
induces the natural action of S on X
q
such that q : X X
q
is S-equivariant and <, >
q
is
S-invariant. Apply (3) to <, >
q
. If
Y
is separately continuous then
X
q
is also separately
continuous by virtue of (3).

Proposition 4.3. Let X and Y be Hausdor S-ows such that X is compact and every
orbit closure cl(yS) in Y is compact. Assume that <, >: Y X R is a strict S-duality.
Then the Ellis semigroup E(S, X) = E(X) is semitopological and there exist separately
continuous actions E(X) X X on Y E(X) Y extending the original actions of S
and such that <, >: Y X R becomes an E(X)-duality .
Proof. By Lemma 4.2 (assertions (2) and (3)) we need only to check that there exists an
action of E(X) on Y which extends the original right action of S on Y in such a way that
the duality <, > becomes an E(X)-invariant.
Denote by : S E(S, X) the canonical semigroup compactication. We follow the
idea of [16, Proposition II.2]. Let p E(X). Choose arbitrarily a net s
i
S such that
(s
i
) converges to p in E(X). Let y Y . Using the compactness of cl(yS), one can pick
a subnet t
j
of s
i
such that yt
j
converges to some z Y . Dene yp := z. Then for every
x X we have
< z, x >=< lim(yt
j
), x >= lim < yt
j
, x >= lim < y, t
j
x >=
=< y, lim(t
j
x) >=< y, lim(t
j
)x >=< y, px > .
The element < y, px > does not depend on the choice of subnets in the denition of z. Since
the duality Y X R is (left) strict, we can conclude that such z is uniquely determined.
Thus, yp is well-dened. These computations show also that < yp, x >=< y, px > because
each of them is the limit of < yt
j
, x >=< y, t
j
x >. Since the given duality is strict it follows
that the function Y E(X) Y, (y, p) yp is a right action which extends the given
action of S on Y . Indeed, we can choose for p := (s) the constant net s
i
= s in the above
denition.

Remarks 4.4. (a) If Y is compact then it is easy to see that the Ellis semigroups E(S, X)
and E(Y, S) are antiisomorphic as semitopological semigroups.
(b) Proposition 4.3 provides a proof of the crucial implication (1) = (2) of Fact 2.6.
Indeed, endow the S-ow Y := C(X) = WAP
S
(X) with the pointwise topology
and apply Proposition 4.3 to the natural S-pair Y X R.
(c) As a corollary we get that the semitopological ow (E(X), X) of Proposition 4.3 is
wap. In fact this ow is even Eberlein as it follows by Theorem 4.5 below.
(d) Proposition 4.3 implies that a compact S-ow X is S-wap i B

= B(C(X)

) is
S-wap. Indeed, by Fact 2.6 it suces to show that the action of the Ellis semigroup
11
E(S, X) on B

is separately continuous. This follows from Lemma 4.2.3 because

by Fact 2.7 (i) the action of E(S, X) on B(C(X))
w
is separately continuous and
B(C(X))
w
B

w
R is a strict E(S, X)-duality. Therefore if X is a compact
S-wap then we can conclude that the S-subow P(X) B

of all probability
measures is wap, too. This fact was established earlier by Glasner [19]. Below we
show (Theorem 4.11) that P(X) is S-Eberlein if X is S-Eberlein.
Now we prove that all bounded S-dualities Y X R come as restrictions of canonical
reexive S-dualities. This fact seems to be interesting even in a purely topological context
(that is, for trivial S) which has been proved by Krivine and Maurey [27] for metrizable
compacta X and Y . In the proof we provide a modication for ows of the well-known
construction of Davis, Figiel, Johnson and Pelczynski [12].
Theorem 4.5. Let : Y X R be a bounded strict S-duality, with semitopological
compact S-spaces X and Y . Then there exist: a reexive Banach space V , a positive
number r > 0, a weakly continuous antihomomorphism h : S (V ), and weak embeddings

1
: X B

and
2
: Y rB such that the following diagram is commutative.
Y X
-
R
rB B

2
?

1
?
-
R
id
?
If the action S X X is jointly continuous then we can suppose that h : S (V ) is
strongly continuous.
Proof. By Proposition 4.3 we can suppose that S is a compact semitopological semigroup.
Adjoining the isolated identity e, one can assume even that S is a monoid and ex = x. The
map q
Y
: Y (C(X), w) is a topological embedding by Grothendiecks Lemma. We will
identify q
Y
(Y ) and Y . Denote by E the Banach subspace of C(X) topologically generated
by Y . That is, E = cl(sp(Y )).
Consider the right action C(X)S C(X), (fs)(x) := f(sx). Then every s-translation
s : C(X) C(X) is a contractive linear operator. The orbit map y : S C(X) is p-
continuous for every y Y C(X). By our assumption S is compact. Therefore, the
orbit yS is bounded p-compact, and, hence w-compact by Grothendiecks Lemma. Since
p-topology coincides with the w-topology on yS, it follows that y : S C(X) is w-
continuous. Then the same is true for every u E = cl(sp(Y )) (as it follows, for example,
from [8, Proposition 6.1.2]). By the Hahn-Banach Theorem, the weak topology of E is the
same as its relative weak topology as a subset of C(X). Therefore we get that ((E, w), S) is
a semitopological ow. Consider the convex hull co(Y Y ) = W. By the Krein-Smulian
Theorem, W is relatively weakly compact in E. Since W is also convex and symmetric, we
can apply a factorization procedure of [12]. For each natural n, set U
n
= 2
n
W +2
n
B(E).
Let | |
n
be the Minkowskis functional of the set U
n
. That is, |v|
n
= inf > 0

v U
n
.
Then | |
n
is a norm on E equivalent to the given norm of E. For v E, let
N(v) :=

n=1
|v|
2
n

1/2
and V := v E

N(v) < .
Denote by j : V E the inclusion map. Then:
(1) (V, N) is a reexive Banach space, j : V E is a continuous linear injection and
Y W B(V ).
12
(2) The restriction of j : V E on each bounded subset A of V induces a homeomor-
phism of A and j(A) in the weak topologies.
By our construction W and B(E) are S-invariant. Thus we get
(3) V is an S-subset of E and N(vs) N(v) for every v V and every s S.
(4) For every v V , the orbit map v : S V, v(s) = vs is weakly continuous.
Indeed, by (3), the orbit v(S) = vS is N-bounded in V . Our assertion follows from (2)
(for A = vS), taking into account that v : S E is weakly continuous.
By (3), for every s S, the translation map s : V V, v vs is a linear contraction of
(V, N). Therefore, we get the antihomomorphism h : S (V ), h(s) = s.
Now, directly from (4), we obtain the following assertion.
(5) h : S (V ) is a w-continuous monoid antihomomorphism.
By (1) and (2) the natural inclusion map
2
: Y B = B(V ) is a topological (weak)
embedding. Dene the weak star embedding
1
: X V

by
1
(x)(v) = j(v)(x) =< v, x >.
Clearly,
1
(x) = (i

)(x), where i

: C(X)

is the adjoint of the inclusion i : E

C(X). In particular, we get that
1
(X) is a bounded subset of V

. It is evident that
1
is S-
equivariant and weak

(=weak) continuous. On the other hand, <

2
(y),
1
(x) >=< y, x >.
Since the original duality is strict, we obtain that
1
is injective and hence a topological
(weak) embedding. As we already mentioned
1
(X) is norm bounded in V

. Therefore,

1
(X) rB

for some r > 0. By renorming V (dening the new norm as [[v[[

new
:= rN(v)
and observing that [[vs[[
new
[[v[[
new
for every s S), we can suppose without restricting
of generality that in fact
1
(X) B

and
2
(Y ) rB.
If SX X is jointly continuous, then the action C(X)S C(X) is jointy continuous
with respect to the norm. By the denition of the Banach space (V, N), it is straightforward
to show that all orbit maps v : S V are N-norm continuous (recall that each | |
n
is
equivalent to the norm of E). This will guarantee that h : S (V )
s
is continuous.
Now we can prove the representation theorem.
Theorem 4.6. (WAP Representation Theorem) Let X be a semitopological S-ow.
The following conditions are equivalent:
(i) f : X R is weakly almost periodic.
(ii) There exist: a representation (h, ) of (S, X) into reexive V with a weak continuous
antihomomorphism h : S (V ), weak continuous : X B

, and a vector v V
such that f(x) =< v, (x) >.
(iii) As in (ii) but with no continuity assumptions on h.
If either: a) S = G is a semitopological group; or b) X is compact and the action
S X X is jointly continuous, then in (ii) we can suppose that h is strongly continuous.
Proof. (i) = (ii) We can suppose that S is a monoid. For the desired representation of
f WAP
S
(X) by some reexive V , choose a left strict bounded duality
<, >: K
f
D
f
R
where K
f
= (cl
w
(fS), w) and D
f
= (B(WAP
S
(X)

), w

). The weak and pointwise

topologies coincide on K
f
. Therefore the action of S on K
f
is separately continuous.
Note, however, that the action of S on D
f
is not necessarily separately continuous. By
Lemma 4.2 we can pass to the naturally associated strict separately continuous S-duality
<, >
q
: K
f
(D
f
)
q
R. Lemma 4.2.3 guarantees that the action of S on (D
f
)
q
is also
separately continuous. Denote by t : X (D
f
)
q
the composition of two natural maps
X D
f
and D
f
(D
f
)
q
. Now, by Theorem 4.5, there exist: a reexive Banach space
13
V and a weakly continuous antihomomorphism h : S (V ) such that S-duality <, >
q
equivariantly can be realized as a part of a reexive duality rB B

R.
K
f
(D
f
)
q
-
R
rB B

2
?

1
?
-
R
id
?
Dene : X B

by (x) =
1
(t(x)) and pick v :=
2
(f). Then f(x) =< v, (x) >, as
desired.
(ii) = (iii) Is trivial.
(iii) = (i) Is immediate by Fact 3.6 (iii).
If S is a semitopological group, then every weakly continuous reexive (anti)representation
is automatically strongly continuous as we mentioned in Section 3 (see Remark 3.2). This
proves the case a). In the second case b), we can apply directly Theorem 4.5.
Now we easily obtain one of our main results.
Theorem 4.7. An S-ow X is wap i X is REFL-approximable.
Proof. If X is REFL-approximable then X is wap by Fact 3.6 (iii).
The nontrivial part follows from Theorem 4.6 because if X has suciently many wap
functions, then (S, X) has suciently many reexive representations.
Corollary 4.8. Every wap ow (S, X) is RN-approximable.
It is well known that a countable product of Eberlein (RN) compacta is again Eberlein
(resp.: RN). We show that the same is true for ows.
Lemma 4.9. The classes of Eberlein and RN S-ows are closed under countable products.
Proof. Let X
n
be a sequence of Eberlein (or, RN) S-ows. By the denition there exists a
sequence of reexive (Asplund) representations
(h
n
,
n
) : (S, X
n
) ((V
n
)
opp
, B(V

n
)).
We can suppose that each X
n
is compact and
n
(X
n
) 2
n
B(V

n
). Turn to the l
2
-sum of
representations. That is, consider
(h, ) : (S, X) ((V )
opp
, B(V

))
where V := (

n
V
n
)
l
2
, h(s)(v) =

n
h(s)(v
n
) for every v =

n
v
n
, and (x) =

n
(x
n
)
for every x = (x
1
, x
2
, )

n
X
n
. It is easy to show that (x) B(V

), is weak

continuous and injective (hence, a topological embedding). Now use the fact that the l
2
-
sum of reexive (Asplund) spaces is again reexive (Asplund) [17].
Corollary 4.10. (i) Every second countable wap ow is Eberlein.
(ii) Every second countable RN-approximable ow is an RN ow.
Proof. Assertion (ii) is immediate by Lemma 4.9. For (i), we need also Theorem 4.7.
The following theorem provides, in particular, a ow generalization of a result by Amir-
Lindenstrauss [3] which states that if X is an Eberlein compact then B

= (B(C(X)

), w

)
is Eberlein, too.
Theorem 4.11. Let X be a compact semitopological S-ow. The following are equivalent:
(i) X is S-Eberlein.
(ii) There exists a Banach space E, a homomorphism h : S (V ) (no continuity
assumptions on h), and an S-embedding : X (V, w).
14
(iii) There exists a compact space Y and a (right) strict S-duality Y X R .
(iv) There exists a sequence of S-invariant weakly compact subsets K
n
C(X) such that

nN
K
n
separates the points of X.
(v) There exists an S-invariant weakly compact subset M in C(X) such that cl(sp(M)) =
C(X).
(vi) B

is S-Eberlein.
(vii) P(X) is S-Eberlein.
Proof. (i) = (ii) By the denition there exists a faithful reexive V -representation. That
is, we can choose a weakly continuous homomorphism h : S (V )
opp
= (V

) and an
equivariant embedding : X B(V

). It suces to choose E := V

.
(ii) = (iii) By our assumption X is S-embedded into (E, w). Dene the right strict
S-duality Y X R as a restriction of the canonical duality where Y := (B(E

), w

).
(iii) = (iv) Use the right version of Lemma 4.2.4. Then our right strict Y X R
duality induces the strict S-duality Y
q
X R. We can suppose in addition that this
duality is bounded. Now dene simply K
n
:= Y
q
C(X) for each n and use Fact 2.5.
(iv) = (v) Look at K
n
as an S-subow of (C(X), w). We can suppose that K
n

B(C(X)). Following a method of Rosenthal [49], consider S-invariant set M
n
consisting of
the constant function equal to 1 on X and of all products of functions f
1
f
2
f
n
where
f
i
(
n
m=1
K
m
) 1. By Fact 2.5 and the Eberlein-Smulian theorem, it is easy to see
that each M
n
is weakly compact. Then M :=
nN
2
n
M
n
is also S-invariant and weakly
compact in C(X). By the Stone-Weierstrass theorem, sp(M) is dense in C(X).
(v) = (vi) We can suppose that M B(C(X)). The corresponding left strict S-duality
M B

[1, 1] is also right strict because cl(sp(M)) = C(X). Now we can apply
Theorem 4.5.
(vi) = (vii) and (vii) = (i) are trivial because P(X) is an S-subow of of B

and X
can be treated is an S-subow of P(X).

5. Reflexive representations of (semi)groups

Now we examine a particular but important case of the ows (S, S), left regular actions of
a semitopological semigroup S on itself by multiplication. Every compact semitopological
semigroup is wap. In general, S is wap i the universal semitopological compactication
S S
W
is an embedding i S is a subsemigroup of a compact semitopological semigroup.
Every locally compact Hausdor topological group G is wap being a subsemigroup of its
one-point compactication (which clearly is a compact semitopological semigroup). More-
over, it is well known that such G is even unitarily representable because it can be embedded
into the unitary group Is(H)
s
of the Hilbert space H = L
2
(G, m
Haar
), where m
Haar
is the
Haar measure on G.
It is also easy to show that every non-Archimedean (having a local base of open subgroups)
topological group is unitarily representable (and, hence wap). Distinguishing unitarily and
reexive representability (and answering a question of Shtern [51]), we show in [39] that
the additive group of L
4
[0, 1] is wap but not unitarily representable. The proof is based on
Grothendiecks double limit property for wap functions. It is still an open question if every
abelian Hausdor topological group (e.g., the additive group of a Banach space) is wap.
Not every topological (even Polish) group is wap. Indeed, the group G = Homeo
+
[0, 1]
of all orientation preserving selfhomeomorphisms of the closed interval is not wap [37]. In
fact we show that every wap function on such G is necessarily constant (conjectured by
Pestov). As a corollary this implies that the universal semitopological compactication G
W
15
of G is trivial (answering a question of Ruppert [50]) and every weakly continuous bounded
representation h : G Aut(V ) into a reexive space V is trivial. This example also
shows (answering a question of Milnes [42]) that there exists a nonprecompact Hausdor
topological group G such that WAP(G) = AP(G).
Turn again to the WAP Representation theorem. It implies that every wap function
comes from a reexive matrix coecient.
Theorem 5.1. For every semitopological monoid S the function f : S R is wap i f is
a matrix coecient of a weak continuous antihomomorphism S (V ) for a reexive V .
That is, there exist v V and V

such that f(s) =< vs, >.

Proof. Apply Theorem 4.6 to the ow (S, S). Then for f WAP(S) there exists a reexive
V and a representation h : S (V ), : S B(V

) such that f(s) =< v, (s) > for a

suitable v V . Denote by e the identity of S. Then f = m
v,
where = (e).
If we wish to get a homomorphism, just consider h : S (V )
opp
= (V

).
It is also easy now to establish the following result rst established by Shtern [51] (see
also [38]).
Fact 5.2. The following are equivalent:
(i) A semitopological semigroup S is wap (equivalently, S is embedded into a compact
semitopological monoid).
(ii) There exists a reexive space E such that S is embedded (as a semitopological sub-
semigroup) into (E)
w
.
Proof. (i) = (ii) We can suppose that S is a monoid. Consider X := S
W
the universal
semitopological compactication of S. Then the corresponding universal map u
W
: S S
W
is a topological embedding by (i) and hence, the action (S, S
W
) is left strict. That is, there
is no strictly coarser topology on S under which S is a semitopological semigroup and S
W
is still a semitopological S-ow. By Theorem 4.6 there exists a separating family (h
i
,
i
)
of reexive V
i
-representations (i I) of (S, S
W
). Then the l
2
-sum of these representations
dened on the Banach space V := (

i
V
iI
)
l
2
will induce a weakly continuous antihomo-
morphism h : S (V ). Since the original action is left strict, it is easy to show that h
must be a topological embedding. Dene E := V

. It is clear that the antihomomorphism

h denes the desired homomorphism h : (V )
opp
= (V

) = (E).
(ii) = (i) It is well known [34] that (V )
w
is a compact semitopological semigroup for
every reexive V .
By [38] (or, Corollary 8.3 below), Is(V )
s
= Is(V )
w
for every reexive V . Therefore we
obtain the following result.
Fact 5.3. [38] Let G be a topological group. The following are equivalent:
(i) G is wap.
(ii) G is a topological subgroup of the group Is(V )
s
(endowed with the strong operator
topology) of all linear isometries for a suitable reexive V .
Remarks 5.4. (i) Theorem 4.7 implies that every wap S-ow X is compactiable. More-
over, if S = G is a semitopological group and : X Y is a corresponding faithful
wap G-compactication (which exists by Theorem 4.7) then the action GY Y
is jointly continuous. This follows from Fact 2.8. Therefore every noncompactiable
in a joint continuous way G-space provides an example of a non-wap ow. Such ex-
amples can be found even for jointly continuous group actions of Polish topological
groups G on Polish spaces X (see [35, 40]).
16
(ii) It is well known (as noted for example in Arhangelskij [5] or Namioka-Wheeler [46])
that a compact space is Eberlein i it can be included into some right strict duality.
Theorem 4.5 provides a ow version.
(iii) Theorems 4.5 , 4.6 and 5.1 admit also almost periodic versions, replacing: sepa-
rately continuous S-dualities by jointly continuous, weakly compact by norm com-
pact and weak

continuous by norm continuous.

6. Fragmentability and flows
Denition 6.1. Let (X, ) be a topological space and be a uniformity on the set X.
Then X is (, )-fragmented if for each nonempty A X and each there exists a
-open subset O of X such that O A ,= and O A is -small.
This denition (for metrics) is explicitly dened by Jayne and Rogers [30] and implicitly it
appears even earlier in Namioka-Phelps [45] (see also [26]). There are several generalizations:
for covers (Bouziad [10]), for functions [29, 36]. Similar concepts are studied in many
contexts: cliquish (Thielman 1953), huskable (in French, epluchable) (Godefroy 1977).
The works [36, 38] are devoted to a systematic study of the fragmentability concept in
the context of (semi)group actions and topological dynamics.
Namiokas famous joint continuity theorem implies that every weakly compact subset of
a Banach space is norm-fragmented [44]. We need the following generalization for locally
convex spaces (V, ) where denotes the usual additive uniform structure on V .
Lemma 6.2. Every relatively weakly compact X V in a l.c.s. V is (weak, )-fragmented.
Proof. See [36, Proposition 3.5].
We will use the following useful observation.
Fact 6.3. Let (X, ) be a Baire space and d a pseudometric on the set X. If X is (, d)-
fragmented, then 1
X
: (X, ) (X, d) is continuous at the points of a dense G

subset D
of X.
Proof. Easily follows using the standard Baire arguments. See for example the proof of
Lemma 1.1 in [44] or [36, Lemma 3.2 (d)].
The following characterization of Asplund spaces (which is a result of many works) in
terms of fragmentability is very important in our setting.
Fact 6.4. A Banach space V is Asplund i every bounded subset A V

of the dual V

is
(weak

,norm)-fragmented.
Standard examples of Asplund spaces include: reexive spaces and c
0
() spaces. Let K
be compact. Then C(K) ASP i K is scattered (that is, every nonempty subspace of K
contains an isolated point).
Let be a uniformity on an S-ow X. We say:
a) z X is a point of equicontinuity (or, a Lyapunov stable) (denote z Equic

(S, X)
or, simply, z Equic) if there exists a compatible uniformity such that for all > 0
there exists a neighborhood U(z) of z such that (sx, sz) for every (x, s) U S.
b) (S, X) is (almost) -equicontinuous if (resp.: X = cl(Equic)) X = Equic.
c) (S, X) is uniformly -equicontinuous if for every there exists such that
(sx, sy) for every (x, y) and every s S.
d) A point z X is the point of local -equicontinuity in the sense of Glasner and
Weiss [21] if z Equic

(S, cl(Sz)) (we do not require that X be compact). If this

17
condition holds for every point in X, we say that (X, ) is locally equicontinuous
and write X LE.
e) (S, X) is (almost, locally) equicontinuous if X is (resp.: almost, locally) -equicontinuous
with respect to some compatible uniformity on X.
f) (S, (X, )) is not sensitive (see for example [20] and the references there) if for every
there exists a non-empty open subset O of X such that (sx, sy) for all
x, y O and s S.
Theorem 6.5. (i) Every RN (e.g., Eberlein) Baire ow (S, X) is almost equicontinuous.
(ii) Every RN-approximable (e.g., wap) Polish S-ow X is almost equicontinuous.
Proof. (i) There exists an Asplund representation h : S (V )
w
, : X B(V

).
Then according to Fact 6.4, f(X) is (weak

, norm)-fragmented. The action of (V )

opp
on
(B(V

), norm) is obviously uniformly equicontinuous. Every point z X of continuity of

the map 1
X
: (X, w

) (X, norm) is a point of equicontinuity in the S-ow (X, w

). Fact
6.3 guarantees that such points are dense in X. Therefore (S, X) is almost equicontinuous.
(ii) Follows from (i) because every RN-approximable second countable ow is RN (Corol-
lary 4.10).
As a conclusion of the part (ii) and Corollary 4.8 we get the following known result.
Corollary 6.6. (Akin-Auslander-Berg [1]) Let G be a topological group and X a metrizable
compact G-ow. Assume that the G-ow X is wap. Then X is almost equicontinuous.
The following denition is an important tool for our purposes.
Denition 6.7. Let (X, ) be an S-ow and a uniformity on the set X such that
top(). We say that the ow (X, ) is -equifragmented if X is (, )-fragmented, the action
of S on X is uniformly -equicontinuous and for some uniformity we have top() = .
We collect here some useful stability conditions for equifragmentability.
Lemma 6.8. (i) The class of equifragmented ows is preserved under subows.
(ii) Equifragmentability is preserved under products. More precisely, if X
i
is
i
-eqifragmented
then the product

X
i
of S-ows is

i
-equifragmented.
(iii) For every Asplund space V the ow ((V )
opp
, (B(V

), w

)) is

-equifragmented
where

is the norm uniformity of V

.
(iv) ((V ), (B(V ), w)) is

-equifragmented for every reexive V .
(v) Every RN-approximable (e.g., wap) ow (S, X) is equifragmented.
(vi) If a compact ow X is equifragmented then X is not sensitive. Therefore, every
RN-approximable S-ow X is not sensitive.
Proof. The assertion (i) is trivial, (ii) and (vi) are straightforward. For (iii) and (iv), we
can use Fact 6.4 and Lemma 6.2, respectively. In order to establish (v), combine (i), (ii)
and (iii).
Let a group G act on a topological space X. We say that:
a) a point z X is transitive (write: z Trans) if cl(Gz) = X. If Trans ,= , then, as
usual, X is called transitive.
b) a point z X is quasitransitive (write: z qTrans) if int(cl(Gz)) ,= .
c) X is quasiminimal if X = qTrans.
d) X is minimal if cl(Gz) = X for all z X.
Let (X, ) be a topological space and be a uniformity on X such that top(). We
say that a subset K X is (, )-Kadec if [
K
= top()[
K
. Denote by Cont(, ) the subset
of all points of continuity of the identity map 1
X
: (X, ) (X, ). Clearly, Cont(, ) is
an example of a (, )-Kadec set.
18
Theorem 6.9. Let a topologized group G act on a topological space (X, ) by homeomor-
phisms. If this action is -equifragmented (with respect to such that top() = )
then:
(i) qTrans Cont(, ) Equic

(G, X). In particular, every point of quasitransitiv-

ity of X is a point of -equicontinuity.
(ii) If a G-subow Y is quasiminimal (e.g., 1-orbit subset Y = Gz) then Y is a (, )-
Kadec set. Hence [
Y
is an compatible uniformity on Y and Y is a uniformly
[
Y
-equicontinuous G-ow.
Proof. (i) Let z qTrans. We have to show that for every there exists a -
neighborhood O(z) of z such that O is -small. Choose such that (gy
1
, gy
2
)
for every (y
1
, y
2
) and g G. Since z qTrans, the set A := int(cl(Gz)) is non-void.
Since X is (, )-fragmentable, we can pick a non-void -open subset W of X such that
W A and W is -small in X. Clearly, W Gz ,= . One can choose g
0
G such that
g
0
z W Gz. Denote by O the open subset g
1
0
W of (X, ). Then O is a -neighborhood
of z and is -small. This proves the inclusion qTrans Cont(, ). The second inclusion
Cont(, ) Equic

(G, X) is trivial because X is uniformly -equicontinuous.

(ii) By the quasiminimality of Y , qTrans(Y ) = Y . Therefore, the assertion (i) implies
that [
Y
= top()[
Y
.
Theorem 6.10. Let G be a semitopological group and X be an RN-approximable semi-
topological G-ow. Then X LE and every G-quasiminimal subspace (for instance, every
orbit) of X is equicontinuous.
Proof. By Lemma 6.8 (v), X is -equifragmented. For every xed z X consider the
S-subow Y := cl(Gz). Clearly, z is a point of quasitransitivity of Y . Then we can apply
Theorem 6.9 to (G, Y ) and conclude that z is a point of local equicontinuity of Y (and
hence of X).
Corollary 6.11. (Generalized Veech-Troallic-Auslander Theorem)
Every wap (not necessarily compact or metrizable) G-ow X is LE and every G-quasiminimal
subow of X is equicontinuous.
Proof. By Corollary 4.8 every wap ow is RN-approximable.
Remark 6.12. Troallic [54] and also Auslander [4] proved that every minimal compact wap
G-ow X is equicontinuous. Previously such a result was established for compact Eberlein
(in our terminology) G-ows by Veech [56].
Combining Theorem 4.7 and Corollary 6.11 for general G-ows, we can draw the following
diagram
Eberlein WAP = REFL
app
RN
app
LE
Consider the case of S = Z and metrizable compact cascades. The fact that WAP ,=
LE is discussed in [21]. The authors constructed (see main example in [21, page 350]) a
transitive cascade (Z, X) such that X is in LE but not wap and every point of transitivity
is recurrent. If we do not require the last assumption then there exists an elementary
example distinguishing wap and RN (and , hence also wap and LE). Namely, the two-point
compactication X of Z with the natural action of Z on X is transitive and contains two
xed points. Fact 6.13 implies that such (Z, X) can not be wap. On the other hand, X is
clearly scattered. Therefore (Z, X) is RN by Proposition 7.15 below.
Fact 6.13. Let X be a wap transitive compact G-ow. Then X contains a unique minimal
compact subow.
19
Proof. Let E = E(G, X) be the Ellis (semitopological) semigroup. By [16, Proposition
II.5] this semigroup contains a unique minimal ideal K which is closed in E. It follows by
transitivity that Et
0
= X for some t
0
X. Then the unique minimal compact subset of X
is Kt
0
.
If the group action (G, X) is RN then there exists a compatible uniformity on X
(the precompact uniformity of the corresponding weak star G-embedding of X into B(V

)
with Asplund V ) such that X is not sensitive (see Lemma 6.8 (vi)). Another observation
comes from Theorem 6.10. It implies that every RN-approximable 1-orbit group action is
equicontinuos. This provides an easy way producing examples of G-ows which fail to be
RN. Roughly speaking, RN G-ow cannot be too chaotic or too massive.
Let G be a topological group. Consider the natural action (call it a -action)

: (GG) G G, (s, t)x = sxt

1
.
It actually coincides with the coset G-space action (GG, (GG)/H), where H = :=
(g, g) : g G.
Lemma 6.14. Let G be a topological group such that (GG, G,

) is an RN-approximable
(e.g., wap) ow. Then G satises SIN (small invariant neighborhoods).
Proof. The given (1-orbit) -action is -equicontinuous, by Theorem 6.10, with respect to
some compatible uniformity on G. Let U(e) be an arbitrary neighborhood of the identity
in G. Choose such that the neighborhood (e) = x G : (e, x) is contained in
U(e). By the -equicontinuity of the -action at the point e one can choose a neighborhood
O(e) such that sOt
1
is -small for all (s, t) GG. Then gOg
1
(e) U(e) for every
g G. This is equivalent to the condition G SIN.
Now we can strengthen a result of Hansel and Troallic. Let G be a topological group.
Following [23] we say that a function f C(G) is strictly wap (notation: f sWAP(G))
if GfG is relatively w-compact in C(G). Denote by [swap] the class of groups such that
sWAP(G) separates points and and closed subsets of G. Denote by [WS] the class of
groups for which every wap function is strictly wap. Clearly, [wap] [WS] [swap].
Proposition 6.15. Let G [swap] then G SIN.
Proof. First observe that G [swap] i the -action (G G, G) is wap and hence RN-
approximable. Now Lemma 6.14 nishes the proof.
Corollary 6.16. (Hansel-Troallic [23]) Let G [wap] [WS]. Then G SIN.
7. Asplund functions and representations
Recall that a Radon-Nikodym compact space [43] is a compact subset in (V

, w

) for an
Asplund space V . We introduce a generalization for ows. Our approach synthesizes some
ideas from [56, 52, 44, 17].
The following denition goes back to Stegall [52] and Namioka [44].
Denition 7.1. Let M be a nonempty bounded subset of a Banach space V . Say that M
is an Asplund set in V if for every countable C M the pseudometric space (V

,
C
) is
separable, where

C
(, ) = sup[ < c, > < c, > [ : c C.
20
Generalizing slightly this denition, lets say that M is an Asplund set for K V

if the
pseudometric subspace (K,
C
) is separable for every countable C M.
We need the following lemma of Namioka in the form presented by Fabian.
Lemma 7.2. [17, Lemma 1.5.3] Let X be a compact space (canonically embedded into
C(X)

) and let M C(X) be a bounded subset. Assume that (X,

M
) is separable. Then
the pseudometric space (C(X)

,
M
) is also separable.
Corollary 7.3. M C(X) is an Asplund set for compact X i M is an Asplund set for
C(X)

.
Remark 7.4. The family of Asplund sets in V has nice properties being stable under taking
subsets, nite unions, closures, linear continuous images, nite linear combinations, etc.
Note also that if M
1
and M
2
are Asplund sets in C(X) for a compact X, then the subset
M
1
M
2
is also Asplund. For these and some other results we refer to [17].
We say that a bounded duality Y X R is an Asplund duality if q
Y
(Y ) is an Asplund
subset of C(X). Conversely, the subset M C(X) is an Asplund set i the corresponding
duality MX R is an Asplund duality where M is endowed with the pointwise topology
inherited from C
p
(X).
The following Lemma is a reformulation of a result of Namioka [44, Theorem 3.4].
Lemma 7.5. Let V be a Banach space and K be a compact subspace in the dual ball B

w
.
Suppose that K is (weak

, norm)-fragmented in V

. Then B = B(V ) is an Asplund set for

K and B K R is an Asplund duality for every topology on B such that : B R
is -continuous for every K.
By Fact 6.4 and Lemma 7.5, the strict duality rB
w
B

w
R is an Asplund duality
(call it: canonical Asplund duality) for every Asplund space V (and r > 0).
Denition 7.6. Let X be a compact S-ow with a separately continuous left action. We
say that X is w-admissible if C(X) = WRUC
S
(X).
This happens if either: a) (S, X) is jointly continuous (then by Fact 2.1 we have even
C(X) = RUC
S
(X)); b) (S, X) is wap (by Fact 2.7); or c) S is a k-space (use Fact 2.5).
Theorem 7.7. Let X be a compact w-admissible S-ow. Every Asplund strict S-duality
Y X R is an S-restriction of a canonical Asplund S-duality with respect to a weakly
continuous antihomomorphism h : S (V ). More precisely, there exist: a suitable
Asplund space V , a positive number r > 0 and equivariant maps:
1
: X B

w
and

2
: Y B such that the following diagram is commutative
Y X
-
R
rB B

2
?

1
?
-
R
id
?
where we require that
1
is a topological embedding and
2
is an injective map.
If the action S X X is jointly continuous, then we can suppose that h : S (V )
is strongly continuous.
Proof. Consider the natural continuous injective map q
Y
: Y C
p
(X) and denote by K
the subset q
Y
(Y ) C(X). Then K is an Asplund subset in C(X). Then K is an Asplund
subset also in the Banach subspace E = cl(sp(K)) of C(X). Following the method of
[52] and, especially, [17, Section 1.4], one can modify the proof of Theorem 4.5 using the
factorization procedure for Asplund S-sets (instead of weakly compact sets). We dene a
21
sequence | |
n
of norms on E each of them equivalent to the original norm. Namely, for
every natural n consider Minkowskis functional of the set
P
n
:= 2
n
co(K K) + 2
n
B(E).
It is important that the subset
nN
P
n
is Asplund. Moreover, by [17, Theorem 1.4.4] we
get a linear injective continuous mapping j : V E where V is an Asplund space. Since
K is an S-invariant subset of C(X), the same is true for E, B(E) and P
n
. Therefore, every
norm [[ [[
n
is S-nonexpansive. Then it follows by the construction that the corresponding
norm N on the Banach space (V, N) is also S-nonexpansive.
Dene
2
: Y B(V ) as a natural S-inclusion (of sets). Since X is w-admissible we
have WRUC
S
(X) = C(X). This guarantees that every orbit map z : S C(X) is weakly
continuous. Hence the action of S on (E, weak) is separately continuous.
On the other hand, by [17, Theorem 1.4.4], the adjoint map j

: E

has the norm

dense range. It follows that for every bounded subset A of V , the weak topology of V and
the weak topology of E, considering of A as a subset of E and C(X), are the same. In
particular, this implies that every orbit map v : S (V, w) is weakly continuous. Thus,
the antihomomorphism h : S (V ) is weakly continuous.
We get also that the dual (left) action of S on V

is weak* separately continuous. The

natural S-inclusion j : V C(X) is a linear continuous S-map. The adjoint j

: C(X)

is a weak

-weak

continuous S-operator. Denote by

1
the restriction of this map on
X C(X)

. Clearly, < y, x >=<

2
(y),
1
(x) >. Then
1
is injective (and hence a
topological embedding) because the original duality is (right) strict.
If (S, X) is a jointly continuous ow then, like Theorem 4.5, we can prove that h is
strongly continuous, too.

It is well known (see [44, 48]) that, similarly to the Eberlein case, a compact space X
is RN i the unit ball B

C(X)

(and hence P(X)) is RN. The following result provides,

in particular, a generalization for ows.
Theorem 7.8. Let (S, X) be a compact w-admissible (e.g., jointly continuous) ow. The
following conditions are equivalent:
(i) The ow (S, X) is RN.
(ii) There exists a representation (h, ) of (S, X) into a Banach space V such that the
antihomomorphism h : S (V ) is weakly continuous, : X V

is a bounded
weak

embedding, and (X) is (weak

, norm)-fragmented.
(iii) There exists a representation (h, ) of (S, X) into a Banach space V such that
h : S (V ) is an antihomomorphism (no continuity assumptions on h), :
X V

is a bounded weak

embedding, and (X) is (weak

, norm)-fragmented.
(iv) There exists a (right) strict Asplund S-duality Y X R.
(v) There exists a bounded S-invariant subset M C(X) such that M separates points
of X and M is an Asplund set for X (equivalently, for (C(X)

)).
(vi) There exists an S-invariant Asplund set Q in C(X) such that cl(sp(Q))=C(X).
(vii) (S, B(C(X)

) is RN.
(viii) (S, P(X)) is RN.
If (S, X) is jointly continuous, then in the assertions (i), (ii), (vii), (viii) we can suppose
in addition that the corresponding h : S (V ) is strongly continuous.
Proof. (i) = (ii) By denition of RN, there exists a faithful Asplund V -representation. By
Fact 6.4, (X) is (w

, norm)-fragmented in V

.
22
(iii) = (iv) The ball B(V ) is S-invariant and separates points of (X). It follows from
Lemma 7.5 that the right strict S-duality
B(V ) X [1, 1], (v, x) < v, (x) >
is an Asplund duality. In order to get a strict duality, pass to the associated reduced duality
<, >
q
: B
q
X [1, 1] (using the dual version of Lemma 4.2.4). Clearly, <, >
q
is also
an Asplund duality.
(iv) = (v) Take M := q
Y
(Y ) C(X) (and use Corollary 7.3).
(v) = (vi) Suppose that a set M satises assumptions of (v). As in the proof of
Theorem 4.11, produce inductively the sequence of subsets M
n
= M
1
M
1
M
1
, where
M
1
= M 1. Then it is well known that each M
n
is again an Asplund set (Remark
7.4). Moreover, even the set Q =
nN
2
n
M
n
is Asplund. On the other hand, by the
Stone-Weierstrass theorem cl(sp(Q)) = C(X). By the construction Q is S-invariant.
(vi) = (vii) Since Q is an Asplund set in C(X) we obtain that Q B

R is a (left
strict) Asplund S-duality. This duality actually is right strict (and hence strict) because
cl(sp(Q)) = C(X). Now we can conclude that B

is RN S-ow as it follows directly from

Theorem 7.7. The same result guarantees that h is strongly continuous provided that (S, X)
is a jointly continuous ow.
Other implications are trivial.

Denition 7.9. Let X be a semitopological compact S-ow. Let us say that a function f
C(X) is S-Asplund if the orbit fS is an Asplund set in C(X). Equivalently, if (fS)X R
is an Asplund duality. More explicitly, taking into account Corollary 7.3, we see that f is
Asplund i for every countable subset (equivalently, separable) C S the pseudometric
space (X,
C
) is separable, where

C
(x, y) = sup[f(sx) f(sy)[ : s C.
If S is separable then it is equivalent to check the separability of the single semimetric space
(X,
S
).
Denote by Asp
S
(X) the set of all S-Asplund functions on a compact X. The product
F = f
1
f
2
of two Asplund functions f
1
and f
2
on X is again Asplund. This follows from the
inclusion FG (f
1
G) (f
2
G) taking into account Remark 7.4. It is easy to show that in
fact Asp
S
(X) is a Banach S-subalgebra of C(X) for every compact S-ow X.
Lemma 7.10. WAP
S
(X) Asp
S
(X) for every semitopological compact S-ow X.
Proof. Every weakly compact subset of a Banach space is an Asplund set (see [17]). In
particular, fS C(X) is an Asplund set if fS is relatively weakly compact.
Theorem 7.11. (RN Representation Theorem) Let X be a compact w-admissible S-
ow. The following are equivalent:
(i) f Asp
S
(X).
(ii) (fS) X R is an Asplund S-duality.
(iii) There exist: a representation (h, ) of (S, X) into an Asplund V with a (weakly
continuous) antihomomorphism h and a vector v V such that f(x) =< v, (x) >.
(iv) There exist: a representation (h, ) of (S, X) into a Banach space V with a weak

: X B(V

) (no continuity assumptions on h) such that (X) is (w

, norm)-
fragmented and there exists a vector v V satisfying f(x) =< v, (x) > for every
x X.
If (S, X) is jointly continuous, weakly continuous can be replaced by strongly continu-
ous.
23
Proof. (i) (ii) Is trivial as it was mentioned earlier.
(ii) = (iii) Using Lemma 4.2.4 pass to the associated strict S-duality (fS) X
q
R
(which again is Asplund) and apply Theorem 7.7.
(iii) = (iv) is trivial by Fact 6.4.
(iv) = (i) Observe that the orbit vS is an Asplund set for (X) by Lemma 7.5. Now
observe that the orbit T(vS) = fS of f is an Asplund set for X (and hence for C(X)

by
Corollary 7.3).

Corollary 7.12. Let X be a compact w-admissible S-ow. The following are equivalent:
(i) X is RN-approximable S-ow.
(ii) C(X) = Asp
S
(X).
Proof. (i) = (ii) Let X be RN-approximable. Then by Theorem 7.11, Asp
S
(X) separates
points of X. On the other hand, Asp
S
(X) is a Banach subalgebra of C(X) containing the
constants. Therefore by the Stone-Weierstrass Theorem we have the coincidence Asp
S
(X) =
C(X).
(ii) = (i) The algebra C(X) = Asp
S
(X) separates points and closed subsets of X.
Hence, by Theorem 7.11 there are suciently many Asplund representations of (S, X).
Proposition 7.13. (1) Let X be a compact S-ow and q : X Y be an S-quotient.
A continuous bounded function f : Y R is S-Asplund i the composition F =
f q : X R is S-Asplund.
(2) F Asp(X) i it comes from an RN S-factor. That is, there exist an RN S-ow
Y , an S-factor q : X Y and a continuous function f C(Y ) such that F = f q.
(3) A factor Y of RN-approximable compact w-admissible S-ow X is again RN-
approximable and w-admissible. If Y is metrizable then (S, Y ) is an RN ow.
Proof. (1) For every pair x
1
, x
2
in X and every countable C G we have

FC
(x
1
, x
2
) = sup[F(sx
1
) F(sx
2
)[ : s C =
= sup[f(sq(x
1
)) f(sq(x
2
))[ : s C =
fC
(q(x
1
), q(x
2
)).
Thus, q : (X,
FC
) (Y,
fC
) is a surjective pseudometric-preserving map. In partic-
ular, (X,
FC
) is separable i (Y,
fC
) is separable.
(2) Combine the rst assertion, Corollary 7.12 and RN representation Theorem 7.11.
(3) First observe that f WRUC
S
(Y ) i F WRUC
S
(X) because q : X Y induces
the S-inclusion q

: C(Y ) C(X) of Banach S-algebras. This implies that Y is also w-

admissible. We have to show that Y is RN-approximable. By Theorem 7.11 it is equivalent
to check that C(Y ) = Asp
S
(Y ). The latter follows directly from (1).
If, in addition, Y is metrizable, then by Corollary 4.10, (S, Y ) is RN.
For every xed S, the class of all RN-approximable compact S-ows is closed under
subdirect products. Therefore, using a well-known method (see for example [31, 58]) we
obtain that for every compact S-ow X there exists a universal RN-approximable com-
pactication u
A
: X X
A
which is a topological embedding i Asp
S
(X) separates points
and closed subsets. Indeed, by Corollary 7.12 and Proposition 7.13.1, it is easy to see that
u
A
: X X
A
is a compactication of X associated to the algebra Asp
S
(X).
Proposition 7.14. For every compact RN-approximable S-ow X, its Ellis semigroup
E(X), as an S-ow, is also RN-approximable.
Proof. By the denition, E(X) is a an S-subow of X
X
. Hence, E(X) is a subdirect product
of RN-approximable S-ows (X many copies of the ow X).
Proposition 7.15. Every scattered compact jointly continuous S-ow X is RN.
24
Proof. It is well known that X is scattered i C(X) is Asplund. Hence the canonical
representation S (V )
s
, X B(V

)
w
into an Asplund space V := C(X) is the
desired.
Let a semitopological group G act joint continuously on compact X. The following
scheme gives some intuitive explanation about the real place of Asplund functions.
functions: WAP
G
(X) Asp
G
(X) C(X)
compactications: X
W
X
A
X
R
= X
representations: REFL ASP BAN
Now dene Asplund functions on a semitopological group G via the universal compact-
ication u
R
: G G
R
(we identify G with u
R
(G)). A continuous bounded function
f : G R is said to be an Asplund function (and write: f Asp(G)) if there exists an
Asplund function F : G
R
R on the G-ow G
R
such that f = F u
R
. It is equivalent
to say that the orbit fG is an Asplund set in the Banach space RUC(G). In particular,
Asp(G) RUC(G).
Dene by : C(G
R
) RUC(G) the natural isomorphism of G-algebras induced by the
compactication u
R
: G G
R
.
Proposition 7.16. Let G be a semitopological group.
(1) Asp(G) is a G-invariant Banach subalgebra of RUC(G) canonically G-isomorphic
to Asp
G
(G
R
). More precisely, Asp(G) = (Asp
G
(G
R
)).
(2) WAP(G) Asp(G).
(3) Denote by u
A
: G G
A
the G-compactication induced by the algebra Asp(G).
Then u
A
: G G
A
is the universal RN-approximable jointly continuous G-compactication
of G. More precisely, for every jointly continuous G-compactication : G Y
with RN-approximable compact G-ow Y, there exists a (necessarily unique) G-map
: G
A
Y such that u
A
= .
(4) G
A
is a right topological monoid naturally isomorphic to the Ellis semigroup E(G, G
A
)
and u
A
is a right topological semigroup compactication of G.
Proof. (1) Follows by the denition of Asp(G).
(2) If f WAP(G), then Fact 2.7(ii) guarantees that fG RUC(G). Then fG, being
a relatively weakly compact in RUC(G), is necessarily an Asplund set (see [17]).
(3) For joint continuity of the action of G on G
A
, recall that Asp(G) is a G-invariant
subalgebra of RUC(G). Universality follows from the fact that G
A
canonically can be
identied with (G
R
)
A
dened for the jointly continuous compact G-ow G
R
.
(4) Let i : G E(G
A
) be the natural homomorphism of G into the Ellis semigroup of the
G-ow G
A
. Consider the orbit map : E(G
A
) G
A
, (p) = p(u
A
(e)). Clearly, (i(g)) =
u
A
(g) for every g G. Therefore is a morphism between two compactications i : G
E(G
A
) and u
A
: G G
A
. It suces to show that is an isomorphism of these transitive G-
ows. By [58, D.2] we need the existence of a morphism of compactications in the reverse
direction. We can use Proposition 7.14 which states that E(G
A
) is RN-approximable. By
the universality property of u
A
, there exists a continuous G-map : G
A
E(G
A
) such
that u
A
= i. Hence, is the desired morphism between the compactications.
The G-algebra Asp(G) is m-admissible in the sense of [8] as it follows by Proposition 7.16
and [8, Theorem 3.1.7].
25
Theorem 7.17. Let G be a semitopological group. The following are equivalent:
(i) f Asp(G).
(ii) There exist: an Asplund space V , a strongly continuous antihomomorphism h : G
Is(V ), vectors v V , and V

such that f(g) =< v, g > (that is, f = m

v,
).
(iii) There exist a G-compactication : G Y with a jointly continuous RN G-ow
Y and a function F C(Y ) = Asp
G
(Y ) such that f = F .
(iv) There exist a G-compactication : G Y with a jointly continuous G-ow Y and
a function F Asp
G
(Y ) such that f = F .
Proof. (i) = (ii) Since the G-action on a compact space X = G
R
is jointly continuous,
we can use strongly continuous version of Theorem 7.11 obtaining strongly continuous
antihomomorphism h : G Is(V ) and a weak

continuous
0
: G
R
B(V

) such that
f(g) =< v,
0
(g) >. Then f = m
v,
where =
0
(e) and e is the identity of G.
(ii) = (iii) Dene Y as the weak

closure of cl
w
(G) of the orbit of G. By Fact 2.2,
Y is a jointly continuous G-ow. Moreover, Y is an RN G-ow (Denition 3.1). Then,
F : Y R, F(y) =< v, y > is the desired function by Corollary 7.12.
(iii) = (iv) is trivial.
(iv) = (i) By our assumption FG is an Asplund set in C(Y ). The natural G-embedding
(induced by ) of Banach algebras C(Y ) RUC(G) maps FG onto fG. Therefore, fG is
an Asplund set in RUC(G).
Proposition 7.18. Let G be a semitopological group and X be a compact minimal G-ow.
Then AP
G
(X) = WAP
G
(X) = Asp
G
(X).
Proof. In general, AP
G
(X) WAP
G
(X) Asp
G
(X) by Lemma 7.10. So we have only
to show that in our situation AP
G
(X) Asp
G
(X) holds. Let F Asp
G
(X). Then by
the universality of the canonical S-quotient u
A
: X X
A
, we have F = f u
A
for some
f Asp
G
(X
A
). Since (G, X
A
) is also minimal, (G, X
A
) is equicontinuous by Theorem 6.10.
Hence, f AP
G
(X
A
). Then, clearly F AP
G
(X).

Examples 7.19. (1) As it was mentioned above, the two-point compactication of Z, as

a cascade, is RN but not wap. Similarly the two-point compactication of R, as an
R-ow, is RN but not wap.
(2) Dene f : Z R by f(z) = 1 i z is a positive integer and f(z) = 0 otherwise.
Then f Asp(Z) WAP(Z).
Indeed, f Asp(Z) by Theorem 7.17 because f comes from the two-point com-
pactication Y of Z which is RN Z-ow by (1). On the other hand, f / WAP(Z)
because f does not satisfy Grothendiecks DLP (see Fact 2.4). Choose s
n
= n and
x
m
= m. Then lim
m
lim
n
f(n m) = 1 ,= 0 = lim
n
lim
m
f(n m).
(3) As in (2), it is easy to show that f Asp(R)WAP(R) for the functions f(x) =
x
1+|x|
and f(x) = arcsinx.
(4) The cascade (Z, [0, 1]) generated by the f(x) = x
2
map is not wap. Indeed, it
contains, as a subow, the two-point compactication of Z. Take, for example, the
Z-orbit of the point x =
1
2
. Together with the endpoints 0 and 1, we get the
closure of this orbit.
(5) Let X be a minimal compact jointly continuous G-ow which is not equicontinuous.
Then C(X) ,= Asp
G
(X) (X is not RN-approximable) and RUC(G) ,= Asp(G).
Indeed, by Theorem 6.10, X is not RN-approximable. Theorem 7.12 guarantees
that C(X) ,= Asp
G
(X). Another proof of the same fact follows from the equality
AP
G
(X) = Asp
G
(X) (Proposition 7.18).
26
Now we check that RUC(G) ,= Asp(G). Fix f C(X) Asp
G
(X) and a point
z X. Since z is a point of transitivity of X, there exists a continuous onto G-map
q : G
R
X such that q(u
R
(g)) = gz for every g G. Dene F : G
R
R as the
composition f q. Then F / Asp
G
(G
R
) by Proposition 7.13. Thus the restriction
F[
G
(g) = f(gz) of F on G satises F[
G
RUC(G) Asp(G).
As a concrete example consider the cascade on the two-dimensional torus T
2
=
(R/Z)
2
generated by the selfhomeomorphism (see [7, Example 5.1.7])

: T
2
T
2
,

([a], [b]) = ([a +b], [a +])

where is a given irrational number. Then the corresponding ow (Z, T
2
,

) is min-
imal but not equicontinuous. The minimality one can check by results of Fursten-
berg [18]. In particular, the cascade (Z, T
2
,

) is not RN. As an another corollary,

Asp(Z) ,= RUC(Z) = C(Z) and Z
R
is not RN-approximable.
Remarks 7.20. (i) A result from [41] states that a topological group G is precompact i
WAP(G) = RUC(G), previously obtained in [2] for monothetic groups. Is it true
the same assuming Asp(G) = RUC(G) ?
(ii) Namioka and Phelps [45] proved a generalized Ryll-Nardzewski xed-point theorem
for S-ows which are weak star compact convex subsets in the dual of an Asplund
space. Hence, this situation is a particular case of RN ows. It is interesting to
analyze possible applications for amenability context as well as for decomposition
theorems.
8. Kadec property: when does weak imply strong ?
Recall that a Banach space V has the Kadec property if the weak and norm topologies
coincide on the unit (or some other) sphere of V . Let us say that a subset X of a locally
convex space (l.c.s.) (V, ) is a Kadec subset (light subset in [38]) if the weak topology
coincides with the strong topology. Light linear subgroups G Aut(V ) (with respect to the
weak and strong operator topologies) can be dened Analogously. Clearly, if G is orbitwise
Kadec on V that is, all orbits Gv are light in V , then G is necessarily light. The simplest
examples are the spheres (orbits of the unitary group Is(H)) in Hilbert spaces H.
The following results show that linear actions frequently are orbitwise Kadec.
Theorem 8.1. Let a subgroup G Aut(V ) be equicontinuous, X be a bounded, (weak, )-
fragmented G-invariant subset of an l.c.s. V with the natural uniformity . Then every,
not necessarily closed, quasiminimal G-subspace (e.g., the orbits) Y of X is a Kadec subset.
Proof. The equicontinuity of the subgroup G Aut(V ) implies that the action of G on
the bounded subspace X V is uniformly -equicontinuous with respect to the natural
uniformity on V . Since X is (weak, )-fragmented we get that in fact the G-ow X is
(weak, [
X
)-equifragmented. Therefore we can apply Theorem 6.9.
We say that an l.c.s. V is boundedly fragmented (write: V BF) if every bounded subset
X V is (weak, )-fragmented, where , as above, is the natural uniformity of V .
Corollary 8.2. Let V BF. Then every equicontinuous G Aut(V ) is a light subgroup
and every orbit Gv is a light subset in V .
The class BF is large (see the relevant references in [38]) and includes among others:
Banach spaces with PCP (point of continuity property), semireexive l.c.s., Frechet spaces
with the Radon-Nikodym Property.
27
Corollary 8.3. [38] Let V be a Banach space with PCP. Then any bounded subgroup G of
Aut(V ) (e.g., Is(V)) is light.
Now, combining Corollary 8.3 and Theorem 4.7, one can obtain a transparent proof of
Fact 2.8.
Ellis-Lawsons Joint Continuity Theorem: Let G be a subgroup of a compact semitopological
monoid S. Suppose that (S, X) is a semitopological ow with compact X. Then the action
GX X is jointly continuous and G is a topological group.
Proof. (S, X) is wap by Fact 2.6. Therefore by Theorem 4.7 there exists an approximating
family (h
i
,
i
) of reexive V
i
-representations. It suces to prove the theorem for the canoni-
cal wap (V )
opp
-ow B

. Let G be a subgroup of (V )
opp
. Then by Corollary 8.3 the strong
operator topology on G coincides with the weak topology. In particular, G is a topological
group. Moreover, by Fact 2.2 the action of G on (B

, w

) is jointly continuous.
Recall that X is a Namioka space if for every compact space Y and a separately continuous
map : Y X R there exists a dense subset P X such that is jointly continuous at
every (y, p) Y P. A topological space is said to be

Chech-complete if it can be represented
as a G

-subset of a compact space. Every

Chech-complete (e.g., locally compact or Polish)
space is a Namioka space.
Proposition 8.4. Let G be a semitopological group, X be a semitopological G-ow and
f C(X). Suppose that (cl
w
(fG), weak) be a Namioka space. Then fG is light in C(X).
Proof. As in the proof of Lemma 6.2, it is easy to show that if a bounded subset of a Banach
space is a Namioka space under the weak topology then it is (weak-norm)-fragmented.
Therefore we can complete the proof by Theorem 8.1.
Theorem 8.5. Let G be a semitopological group. Then for every semitopological G-ow X
and every f WAP(X) the pointwise and norm topologies coincide on the orbit fG. In
particular, WAP
G
(X) RUC
G
(X).
Proof. Proposition 8.4 guarantees that fG is (weak, norm)-Kadec. On the other hand, the
weak and pointwise topologies coincide on the weak compact set cl
w
(fG).
Now we turn to the weak

version of the lightness concept. Let (V, ) be an l.c.s. with its

strong dual (V

). Denote by

the corresponding uniformity on V

. Lets say that a

subset A of V

is weak

light if weak

and strong topologies coincide on A. If G is a subgroup

of Aut(V

), then the weak

(resp., strong

) topology on G is the weakest topology which

makes all orbit maps

: G V

: V

weak

(resp., strong) continuous.

Following [36] we say that an l.c.s. V is a Namioka-Phelps space (V NP) if every
equicontinuous subset X V

is (w

)-fragmented. The class NP is closed under sub-

spaces, products and l.c. sums and includes: Asplund Banach spaces, semireexive l.c.s.
and Nuclear l.c.s.
Theorem 8.6. Suppose that V is an NP space, G Aut(V ) is an equicontinuous subgroup,
and X V

is an equicontinuous G-invariant subset.

(i) If (X, w

) is a quasiminimal (e.g., 1-orbit) G-subset, then X is weak

light.
(ii) The weak

and strong

operator topologies coincide on G.

Proof. (i) The strong topology on the dual space (V

) is the topology of bounded con-

vergence. Since G is an equicontinuous subgroup of Aut(V ), it is easy to show that the dual
action of G on V

is also equicontinuous. On the other hand, X V

is (w

)-fragmented
as it follows by the denition of NP spaces. We obtain in fact that G-ow X is (weak

[
X
)-equifragmented. Now use once again Theorem 6.9.
(ii) Directly follows from (i) because every G-orbit in V

is an equicontinuous subset.
28
The last result is useful in the context of continuity of dual actions (for more information
see [36] and the references there). More precisely, let V be an l.c.s. and h : G Aut(V ) be
a homomorphism such that h(G) is an equicontinuous subgroup of Aut(V ) and the action
GV V is jointly continuous. Then we can ask: is the dual action

: G(V

) (V

), (gf)(v) = f(g
1
v)
also jointly continuous ?
Since h(G) is equicontinuous, clearly (G, V

) is equicontinuous with respect to the dual

action

. Therefore it is equivalent to ask if the orbit maps

f : G (V

) are continuous
for all f V

. Since

f : G (V

, w

) is continuous, it suces to show that the orbits

Gf are (weak

, strong)-Kadec subsets of V

. This fact follows directly from Theorem 8.6

provided that V NP. Hence we obtain the following result.
Corollary 8.7. Let V NP (e.g., Asplund Banach space) and : GV V be a linear
jointly continuous equicontinuous action. Then the dual action

: G V

is also
jointly continuous.
Remark 8.8. Corollary 8.7 can be derived also from [36, section 6]. If V is an Asplund
Banach space then we can drop the condition about equicontinuity (in fact, boundedness)
as it follows by [36, Corollary 6.9].
Proposition 8.9. Let V be an Asplund Banach space, G be a semitopological group, and
h : G Aut(V ) be a bounded weakly continuous antihomomorphism. Assume that v V
and V

are some xed vectors. Then the corresponding matrix coecient m

v,
: G R
is left uniformly continuous. Moreover, if the vector v is norm-continuous, then m
v,
, in
addition, is right uniformly continuous.
Proof. The antihomomorphism h sends G into a norm bounded subgroup of Aut(V ). There-
fore by Fact 3.5 it suces to show that is a norm G-continuous vector. Since h : G
Aut(V ) is weak continuous, the orbit map

: G V

is weak star continuous. Since V

is Asplund (and hence NP), Theorem 8.6 implies that the weak star and norm topologies
coincide on the orbit G. Then

: G V

is even norm continuous.

Corollary 8.10. For every semitopological group G
WAP(G) Asp(G) LUC(G) RUC(G)
holds.
Proof. The inclusion WAP(G) Asp(G) is a part of Proposition 7.16. Let f Asp(G).
By Theorem 7.17 the function f coincides with a matrix coecient m
v,
for a suitable
strongly continuous antihomomorphism h : G Is(V )
s
. Now we can apply Proposition
8.9 to f = m
v,
.

References
1. E. Akin, J. Auslander and K. Berg, Almost equicontinuity and the enveloping semigroup, Contemp.
Math., 215 (1998), 75-81.
2. E. Akin and E. Glasner, Residual properties and almost equicontinuity, J. Anal. Math., 84 (2001),
243-286.
3. D. Amir and J. Lindenstrauss, The structure of weakly compact subsets in Banach spaces, Ann. of Math.,
88 (1968), 35-46.
4. J. Auslander, Minimal ows and their extensions, 153, North-Holland Math. St., 1988.
5. A.V. Arhangelski, On some topological spaces that occur in Functional Analysis, Russ. Math. Surveys,
31:5 (1976), 17-32.
29
6. Y. Benyamini and J. Lindenstrauss, Geometric Nonlinear Functional Analysis, vol. 1, Amer. Math. Soc.,
Colloquium Publications 48, Providence, Rhode Island, 2000.
7. J.F. Berglund, H.D. Junghenn and P. Milnes, Compact right topological semigroups and generalizations
of almost periodicity, Lecture Notes in Math., 663 (1978), Springer-Verlag.
8. J.F. Berglund, H.D. Junghenn and P. Milnes, Analysis on Semigroups, Wiley, New York, 1989.
9. R.D. Bourgin, Geometric aspects of of convex sets with the Radon-Nikodym property, Lecture Notes in
Math. 993, Springer-Verlag, 1983.
10. A. Bouziad, Continuity of separately continuous group actions in p-spaces, Topology Appl. 71 (1996),
no. 2, 119124.
11. R.B. Burckel, Weakly almost periodic functions on semigroups, Gordon and Breach Science Publishers,
New York-London-Paris, 1970.
12. W.J. Davis, T. Figiel, W.B. Johnson and A. Pelczynski, J. of Funct. Anal., 17 (1974), 311-327.
13. R. Deville, G. Godefroy and V. Zizler, Smoothness and renormings in Banach spaces, Pitman Mono-
graphs and Surveys in Pure and Applied Mathematics, 64, Longman Scientic Technical, 1993.
14. T. Downarowicz, Weakly almost periodic ows and hidden eigenvalues, Contemp. Math., 215 (1998),
101-120.
15. W.F. Eberlein, Abstract ergodic theorems and weak almost periodic functions, Trans. AMS, 67 (1949),
217-240.
16. R. Ellis and M. Nerurkar, Weakly almost periodic ows, Trans. AMS, 313 (1989), 103-119.
17. M. Fabian, Gateaux dierentiability of convex functions and topology. Weak Asplund spaces, Canadian
Mathematical Society Series of Monographs and Advanced Texts, Wiley, New York, 1997.
18. H. Furstenberg, Strict ergodicity and transformation of the torus, Amer. J. Math., 83 (1961), 573-601.
19. E. Glasner, M-Dynamical systems, preprint, 1998.
20. E. Glasner and B. Weiss, Sensitive dependence on initial conditions, Nonlinearity 6 (1993), 1067-1075.
21. E. Glasner and B. Weiss, Locally equicontinuous dynamical systems, Colloquium Mathematicum, part
2, 84/85 (2000), 345-361.
22. A. Grothendieck, Criteres de compacite dans les espaces functionelles generaux, Amer. J. Math., 74
(1952), 168-186.
23. G. Hansel and J.P. Troallic, Extension properties of WS-groups, Semigroup Forum, 45:1 (1992), 6370.
24. D. Helmer, Joint continuity of ane semigroup actions, Semigroup Forum, 21 (1980), 153-165.
25. D. Helmer, Continuity of semigroup actions, Semigroup Forum, 23 (1981), 153-188.
26. P. Kenderov, Dense strong continuity of pointwise continuous mappings, Pacic. J. Math., 89:1 (1980),
111-130.
27. J.L. Krivine and B. Maurey, Espaces de Banach stables, Israel J. Math. 39:4 (1981), 273295.
28. A. Jalilian and M.A. Pourabdollah, Transformation semigroup compactications and norm continuity of
weakly almost periodic functions, Proc. Indian Acad. Sci. Math. Sci. 110:1 (2000), 5559.
29. J.E. Jayne, J. Orihuela, A.J. Pallares and G. Vera, -fragmentability of multivalued maps and selection
theorems, J. Funct. Anal. 117 (1993), no. 2, 243273.
30. J.E. Jayne and C.A. Rogers, Borel selectors for upper semicontinuous set-valued maps, Acta Math., 155,
(1985), 41-79.
31. H.D. Junghenn, Almost periodic compactications of product ows, Semigroup Forum, 58 (1999), 296-
312.
32. J.D. Lawson, Joint continuity in semitopological semigroups, Illinois J. Math., 18 (1974), 275- 285.
33. J.D. Lawson, Points of continuity for semigroup actions, Trans. AMS 284 (1984), 183-202.
34. K. de Leeuw and I. Glicksberg, Applications of almost periodic compactications, Acta Math., 105
(1961), 63-97.
35. M. Megrelishvili, A Tychono G-space not admitting a compact G-extension or a G-linearization, Rus-
sian Math. Surv., 43:2 (1988), 177-178.
36. M. Megrelishvili, Fragmentability and continuity of semigroup actions, Semigroup Forum, 57 (1998),
101-126.
37. M. Megrelishvili, Every semitopological semigroup compactication of the group H
+
[0, 1] is trivial, Semi-
group Forum, 63:3 (2001), 357-370.
38. M. Megrelishvili, Operator topologies and reexive representability, In: Nuclear Groups and Lie
Groups, Research and Exposition in Math. Series, 24 (2001), Heldermann-Verlag, 197-208.
39. M. Megrelishvili, Reexively but not unitarily representable topological groups, Topology Proceedings,
25 (2002), 615-625.
40. M. Megrelishvili and T. Scarr, Constructing Tychono G-Spaces Which Are Not G-Tychono, Topology
and its Applications, 86:1 (1998), 69-81.
30
41. M. Megrelishvili, V. Pestov and V. Uspenskij, A note on the precompactness of weakly almost periodic
groups In: Nuclear Groups and Lie Groups, Research and Exposition in Math. Series, 24 (2001),
Heldermann-Verlag, 209-216.
42. P. Milnes, Compactications of semitopological semigroups, J. Austral. Math. Soc. 15 (1973), 488503.
43. I. Namioka, Separate continuity and joint continuity, Pacic. J. Math., 51 (1974), 515-531.
44. I. Namioka, Radon-Nikodym compact spaces and fragmentability, Mathematika, 34 (1987), 258-281.
45. I. Namioka and R.R. Phelps, Banach spaces which are Asplund spaces, Duke Math. J., 42 (1975),
735-750.
46. I. Namioka and R.F. Wheeler, Gulkos proof of the Amir-Lindenstrauss theorem, Contemp. Math., 52
(1986), 113-120.
47. V.G. Pestov, Topological groups: where to from here? Topology Proceedings, 24 (1999), 421-502.
http://arXiv.org/abs/math.GN/9910144.
48. O.I. Reynov, On a class of Hausdor compacts and GSG Banach spaces, Studia Math., 71 (1981),
113-126.
49. H.P. Rosenthal, The heredity problem for weakly compactly generated Banach spaces, Compositio Math.
28 (1974), 83111.
50. W. Ruppert, Compact semitopological semigroups: An intrinsic theory, Lecture Notes in Math., 1079
(1984), Springer-Verlag.
51. A. Shtern, Compact semitopological semigroups and reexive representability of topological groups, Rus-
sian J. of Math. Physics, 2 (1994), 131-132.
52. Ch. Stegall, The Radon-Nikodym property and conjugate Banach spaces II, Trans. AMS, 264 (1981),
507-519.
53. S. Teleman, Sur la representation lineare des groupes topologiques, Ann. Sci. Ecole Norm. Sup., 74
(1957), 319-339.
54. J.P. Troallic, Espaces Fonctionnels et theoremes de I. Namioka, Bull. Soc. Math. France, 107 (1979),
127-137.
55. V.V. Uspenskij, Universal topological groups with a countable basis, Functional Anal. Prilojen. (in Rus-
sian), 20:2 (1986), 86-87.
56. W. A. Veech, A xed point theorem free approach to weak almost periodicity, Trans. AMS, 177 (1973),
353-362.
57. J. de Vries, Equivariant embeddings of G-spaces, in: J. Novak (ed.), General Topology and its Relations
to Modern Analysis and Algebra IV, Part B, Prague, 1977, 485-493.
58. J. de Vries, Elements of Topological Dynamics, Kluwer Academic Publishers, Dordrect-Boston-London,
1993.
Bar-Ilan University, Israel
E-mail address: megereli@math.biu.ac.il
URL: http://www.math.biu.ac.il/

megereli