Mean Ergodic Operators in Frechet Spaces

MEAN ERGODIC OPERATORS IN FRCHET SPACES
A. A. ALBANESE, J. BONET
*
AND W. J. RICKER
Abstract. Classical results of A. Pelczynski and of M. Zippin concern-
ing bases in Banach spaces are extended to the Frchet space setting;
thus answering a question posed by N. Kalton almost 40 years ago.
Equipped with these results, we prove that a Frchet space with a basis
is reexive (resp. Montel) if and only if every power bounded operator
is mean ergodic (resp. uniformly mean ergodic). New techniques are
developed and many examples in classical Frchet spaces are exhibited.
1. Introduction and statement of the main results.
A continuous linear operator T in a Banach space X (or a locally convex
Hausdor space, briey lcHs) is called mean ergodic if the limits
Px := lim
n
1
n
n
m=1
T
m
x, x X, (1.1)
exist (in the topology of X). J. von Neumann (1931) proved that unitary
operators in Hilbert space are mean ergodic. Ever since, intensive research
has been undertaken concerning mean ergodic operators and their applica-
tions; for the period up to the 1980s see [19, Ch. VIII Section 4], [25, Ch.
XVIII], [33, Ch. 2], and the references therein. A continuous linear operator
T in X (the space of all such operators is denoted by L(X)) is called power
bounded if sup
m0
|T
m
| < . Such a T is mean ergodic if and only if
X = u X : u = Tu Im(I T), (1.2)
where Im(I T) denotes the range of I T and the bar denotes closure in
X. In general, the right-hand side of (1.2) is precisely the set of all x X
for which the sequence
1
n
n
m=1
T
m
x converges in X.
It quickly became evident that there was an intimate connection between
geometric properties of the underlying Banach space X and mean ergodic
operators on X. The space X itself is called mean ergodic if every power
bounded operator T L(X) satises (1.1). As a sample, F. Riesz (1938)
showed that all L
p
-spaces (1 < p < ) are mean ergodic. In 1939 E.R. Lorch
Key words and phrases. Mean ergodic operator, power bounded, Frchet space, basis,
Schauder decomposition.
Mathematics Subject Classication 2000: Primary 46A04, 46A35, 47A35 Secondary
46G10.
*
Research partially supported by MEC and FEDER Project MTM 2007-62643 and the
net MTM2007-30904-E (Spain).
1
2 A. A. Albanese, J. Bonet and W. J. Ricker
proved that all reexive Banach spaces are mean ergodic. L. Sucheston, [51],
posed the following question, concerning the converse of Lorchs result: If
every contraction in a Banach space X is mean ergodic, is X reexive?
In 1997, E. Emelyanov showed that every mean ergodic Banach lattice is
reexive, [20]. A major breakthrough came in 2001 in the penetrating paper
of V.P. Fonf, M. Lin and P. Wojtaszczyk, [23]. By using the theory of bases
in Banach spaces, they were able to establish (amongst other things) the
following characterizations for a Banach space X with a basis:
(i) X is nite dimensional if and only if every power bounded operator is
uniformly ergodic (i.e. the limit (1.1) exists uniformly on the unit ball
of X).
(ii) X is reexive if and only if every power bounded operator is mean er-
godic (i.e. if and only if X is mean ergodic).
Essential to their arguments are two classical results on bases in Banach
spaces. Namely, a result of A. Pelczynski, [15, p.54], [39, p.77], stating that
if X is non-reexive, then it possesses a non-reexive (separable), closed
subspace with a basis and, a result of M. Zippin, [55], stating that if a non-
reexive Banach space has a basis, then it has a non-shrinking basis. The
rst result for (a special class of) power bounded operators T on certain
lcHs X is due to M. Altman, [4]. The restriction on T that Altman imposed
(a weak compactness condition) was later removed by K. Yosida [54, Ch.
VIII]. For quasi-compact operators and R-endomorphisms in lcHs, a mean
ergodic theorem due to A. Pietsch is also available, [42]. Our aim is to
present several results in Frchet spaces X (some being in the spirit of [23])
which connect geometric/analytic properties of the underlying space X to
mean ergodicity of operators acting on X. To be more precise, T L(X) is
called power bounded if T
m
m=0
is an equicontinuous subset of L(X). Since
the requirement (1.1) is not dependent on X being normable, an operator
T L(X) is again called mean ergodic whenever it satises (1.1).
Technical terms concerning lcHs X and various types of bases will be
formally dened later. A general reference is [40]. Let us recall that if
X
is a system of continuous seminorms generating the topology of X, then the
strong operator topology
s
in L(X) is determined by the seminorms
q
x
(S) := q(Sx), S L(X), (1.3)
for each x X and q
X
(in which case we write L
s
(X)), and the uniform
operator topology
b
in L(X) is dened by the family of seminorms
q
B
(S) := sup
xB
q(Sx), S L(X), (1.4)
for each q
X
and bounded set B X (in which case we write L
b
(X)).
One refers to
b
as the topology of uniform convergence on the bounded sets
of X. If X is a Banach space, then
b
is precisely the operator norm topology
on L(X). If
X
is countable and X is complete, then X is called a Frchet
space. The result of A. Pelczynski and that of M. Zippin mentioned above,
MEAN ERGODIC OPERATORS IN FRCHET SPACES 3
both crucial if any headway is to be made in the non-normable setting, have
been extended, each one being of independent interest.
Theorem 1.1. Every non-reexive Frchet space contains a non-reexive,
closed subspace with a basis.
Theorem 1.2. A complete, barrelled lcHs with a basis is reexive if and only
if every basis is shrinking if and only if every basis is boundedly complete.
Theorem 1.2 clearly implies Zippins result in the Banach space setting. It
should be noted that Theorem 1.2 provides a positive answer to a question
posed by N. Kalton almost 40 years ago, [28, p.265].
Given any T L(X), let us introduce the notation
T
[n]
:=
1
n
n
m=1
T
m
, n N, (1.5)
for the Cesro means of T (which already arose in (1.1)). Then T is mean
ergodic precisely when T
[n]
n=1
is a convergent sequence in L
s
(X). If
T
[n]
n=0
happens to be convergent in L
b
(X), then T will be called uni-
formly mean ergodic. The space X itself is called uniformly mean ergodic if
every power bounded operator on X is uniformly mean ergodic. Equipped
with Theorems 1.1 and 1.2 it is possible to establish the following two facts.
Theorem 1.3. Let X be a Frchet space with a basis. Then X is Montel if
and only if every power bounded, mean ergodic operator on X is uniformly
mean ergodic, that is, if and only if X is uniformly mean ergodic.
Theorem 1.3 can be interpreted as an extension of (i) above. The following
extension of (ii) above is also presented.
Theorem 1.4. A Frchet space X with a basis is reexive if and only if
every power bounded operator on X is mean ergodic.
The mean ergodicity and uniform mean ergodicity of such classical Frchet
spaces as the Kthe echelon spaces
0
(A) and
p
(A), for 1 p , the
sequence spaces
p+
:=
q>p
q
for 1 p < , the Lebesgue spaces L
p
:=
1r<p
L
r
([0, 1]) for p (1, ) and the spaces L
p
oc
(R) of locally integrable
functions, for 1 p < , are completely determined.
For the following fact in the Banach spaces, see [23, Theorem 1].
Theorem 1.5. Let X be any Frchet space which admits a non-shrinking
Schauder decomposition. Then there exists a power bounded operator on X
which is not mean ergodic (i.e. X is not mean ergodic).
Our nal sample result is the following one. For the Banach space case it
follows from a result of E. Emelyanov and M. Wol, [21, Theorem 3.1].
Theorem 1.6. Let X be a Frchet space which contains an isomorphic copy
of the Banach space c
0
. Then there exists a power bounded operator on X
which is not mean ergodic (i.e. X is not mean ergodic).
2. Preliminary Results and some Examples.
Given any lcHs X and T L(X) we always have the identities
(I T)T
[n]
= T
[n]
(I T) =
1
n
(T T
n+1
), n N, (2.1)
and also (setting T
[0]
:= I, the identity operator on X) that
1
n
T
n
= T
[n]

(n 1)
n
T
[n1]
, n N. (2.2)
Some authors prefer to use
1
n
n1
m=0
T
m
in place of T
[n]
; since
T
[n]
= T
_
1
n
n1
m=0
T
m
_
=
1
n
(T
n
I) +
1
n
n1
m=0
T
m
, n N,
this leads to identical results. Recall the notation ker(T) := x X : Tx =
0. The following result is due to K. Yosida, [54, Ch. VIII, 3].
Proposition 2.1. Let X be a lcHs and T L(X) be power bounded. Then
Im(I T) = x X : lim
n
T
[n]
x = 0 (2.3)
and so, in particular,
Im(I T) ker(I T) = 0. (2.4)
A subset A of a lcHs X is called relatively sequentially (X, X
)-compact if
every sequence in A contains a subsequence which is (X, X
)-convergent to
some element of X. As a consequence of Proposition 2.1, Yosida established
the following mean ergodic theorem, [54, Ch. VIII; 3].
Theorem 2.2. Let X be a lcHs and T L(X) be a power bounded operator
such that
T
[n]
x
n=1
is relatively sequentially (X, X
)-compact, x X. (2.5)
Then T is mean ergodic and the operator P = lim
n
T
[n]
(limit exist-
ing in L
s
(X)) is a projection satisfying Im(P) = ker(I T) and ker(P) =
Im(I T) with
X = Im(I T) ker(I T). (2.6)
For X a Banach space, the above result is the customary one, that is, T
is required to satisfy sup
n
|T
n
| < , which clearly implies
lim
n
1
n
T
n
= 0, in L
s
(X). (2.7)
By the principle of uniform boundedness, every mean ergodic operator T
necessarily satises
sup
n
|T
[n]
| < , (2.8)
that is, T is Cesro bounded (in the terminology of [33, Ch. 2]). On the
other hand, (2.8) is also sucient for mean ergodicity of T whenever T
additionally satises (2.5) and (2.7), [33, p.72]. E. Hille exhibited a classical
kernel operator T in L
1
([0, 1]) which fails to be power bounded (actually,
|T
n
| = 0(n
1/4
)) but, nevertheless, is mean ergodic (and hence, satises
(2.8)), [24, 6]. Actually, for this example, it is shown in [24] that T
[n]
0
in L
s
(X) as n which then implies, via (2.2), that (2.7) holds. There also
exist operators T L(X) which satisfy (2.8) but, fail to satisfy (2.7), even
with X a nite dimensional space, [22, Example 4.7]. In view of such features
(see also [33, p.85], [43], for further relevant comments), it is not surprising
that some authors take the viewpoint that it is not the power boundedness
of T but, rather the existence of the limit (2.7), which is essential in ergodic
theory; this viewpoint is explicitly stated in [18, p.186], [38, p.214], for
example. The proof of the following version of Proposition 2.1 is routine.
Proposition 2.3. Let X be a barrelled lcHs. Let the operator T L(X)
satisfy (2.7) and the condition
T
[n]
x
n=1
is a bounded set in X, x X. (2.9)
Then T satises both (2.3) and (2.4).
Given T L(X), its dual operator T
t
: X
is dened by Tx, x
) =
x, T
t
x
) for all x X and x
. Let us denote X equipped with its weak

topology (X, X
) by X
.
Theorem 2.4. Let X be a barrelled lcHs and T L(X). Then T is mean
ergodic if and only if it satises both (2.5) and (2.7).
Setting P := lim
n
T
[n]
(existence of the limit in L
s
(X)), the operator
P is a projection which commutes with T and satises Im(P) = ker(I T)
and ker(P) = Im(I T). Moreover, X satises (2.6).
Proof. If T is mean ergodic, then it follows from (2.2) that (2.7) holds. More-
over, (2.5) is also satised. Conversely, suppose that T satises (2.5) and
(2.7). Fix x X. By (2.5) there exists x
0
X and a subsequence such that
T
[n
k
]
x x
0
in X
. It follows from (2.1) and (2.7) that

lim
n
(TT
[n]
T
[n]
)x = lim
n
_
(n + 1)
n

1
(n + 1)
T
n+1
x
1
n
Tx
_
= 0, (2.10)
in X, and hence, that (TT
[n
k
]
T
[n
k
]
)x 0 in X
. So, TT
[n
k
]
x x
0
in X
.
According to (2.10), for each x
, we then have
lim
k
TT
[n
k
]
x, x
) = lim
k
T
[n
k
]
x, x
) = x
0
, x
).
Using the identities
TT
[n
k
]
x, x
) = T
[n
k
]
x, T
t
x
), k N,
it follows that x
0
, x
) = x
0
, T
t
x
) = Tx
0
, x
). Since x
is arbitrary,
we have Tx
0
= x
0
and hence, via (1.5), that
T
[n]
x
0
= x
0
, n N. (2.11)
Accordingly, we have
x x
0
= x lim
k
T
[n
k
]
x = lim
k
(x T
[n
k
]
x) (2.12)
in X
. Now, since x T
[n]
x = (I T)
n
m=1
1
n
(I + T + ... + T
m1
)x
Im(I T), n N, we conclude that x T
[n
k
]
x
k=1
Im(I T). That is,
x x
0
belongs to the closure of Im(I T) in X
which, of course, coincides

with Im(I T), [31, p.236]. Since (2.5) implies (2.9), Proposition 2.3 is
available and hence, (2.3) holds. It follows, because xx
0
Im(I T), that
T
[n]
(x x
0
) 0 in X. Combined with (2.11) this yields lim
n
T
[n]
x = x
0
(in X). So, we can dene a linear map P : X X by
Px := lim
n
T
[n]
x, x X. (2.13)
Since X is barrelled, actually P L(X), [32, p.141]; in particular, T
[n]
P
in L
s
(X), that is, T is mean ergodic.
Note that Px = x
0
implies that TPx = Px for each x X, that is,
TP = P. It then follows that T
[n]
P = P for all n N and hence, that
P
2
x = lim
n
T
[n]
(Px) = Px, for x X, and P is a projection. On the
other hand, (2.1) and (2.7) imply that T
[n]
T
[n]
T =
1
n
(T T
n+1
) 0 in
L
s
(X), that is, P = PT (see (2.13)). So, P and T commute.
According to (2.3) and (2.13) we have Im(I T) = ker(P). It is routine
to verify that ker(I T) = Im(P). Since P is a projection, (2.6) follows.
Corollary 2.5. Let X be a barrelled lcHs such that each bounded set in X
is relatively sequentially (X, X
)-compact. If T L(X) satises (2.7) and

(2.9), then T is mean ergodic.
Proof. Under the given hypothesis on X, condition (2.9) implies (2.5) and
hence, Theorem 2.4 applies.
Remark 2.6. (i) In any lcHs X, conditions (2.7) and (2.9) are satised
by every power bounded operator T L(X). Indeed, x q
X
. By the
assumed equicontinuity of T
n
n=0
there exists p
X
satisfying
q(T
n
x) p(x), x X, n N. (2.14)
It follows that q(T
[n]
x)
1
n
n
m=1
q(T
m
x) p(x), x X, n N, which
clearly implies (2.9). Moreover, (2.14) gives that q(
1
n
T
n
x)
1
n
p(x) for all
x X and n N which yields (2.7).
(ii) Reexive lcHs which are either Frchet, (DF)- or (LF)-spaces satisfy
the hypotheses of Corollary 2.5; see [10, Theorem 11, Examples 1,2].
Corollary 2.7. Every reexive lcHs which is a (DF)- or (LF)-space (in
particular, any reexive Frchet space) is mean ergodic.
Proposition 2.8. Let X be any (DF)- or (LF)-space (in particular, any
Frchet space) which is Montel. Then X is uniformly mean ergodic.
Proof. Let T L(X) be power bounded. Since X is reexive, [31, p.369],
Corollary 2.7 implies that T is mean ergodic. Set P := lim
n
T
[n]
, with
the limit existing in L
s
(X). Since X is barrelled and H := P T
[n]
n=1
is bounded in L
s
(X), it follows that H is an equicontinuous subset of L(X),
[32, (2) p.137]. So,
s
and the topology
c
in L(X) of uniform convergence
on the precompact sets of X coincide on H, [32, (2) p.139]. But, bounded
sets in Montel spaces are relatively compact (per denition) and so T
[n]
P
in L
b
(X), that is, T is uniformly mean ergodic.
Let be an index set, always assumed to be countable. Any increasing
sequence A = (a
n
)
n
of functions a
n
: (0, ) is called a Kthe matrix
on , where by increasing we mean 0 < a
n
(i) a
n+1
(i) for all i and
n N. To each p [1, ) is associated the linear space
p
(A) := x C
: q
(p)
n
(x) :=
_
i
[a
n
(i)x
i
[
p
_
1/p
< , n N. (2.15)
We also require the linear space
(A) := x C
: q
()
n
(x) := sup
i
a
n
(i)[x
i
[ < , n N (2.16)
and its closed subspace (equipped with the relative topology)
0
(A) := x C
: lim
i
a
n
(i)x
i
= 0, n N.
Elements x C
are denoted by x = (x
i
). The spaces
p
(A), for p [1, ],
are called Kthe echelon spaces (of order p); they are all Frchet spaces
(separable if p ,= and reexive if p ,= 1, ) relative to the increasing
sequence of seminorms q
(p)
1
q
(p)
2
. . .. For the general theory of such
spaces we refer to [31], [40], for example.
Proposition 2.9. Let A be a Kthe matrix and 1 < p < . The reexive
Frchet space
p
(A) is Montel if and only if it is uniformly mean ergodic.
Proof. Suppose that
p
(A) is not Montel. According to the proof of Proposi-
tion 2.5(ii) in [7] there exists a direct sum decomposition
p
(A) = Y Z with
Z isomorphic to the Banach space
p
. M. Yahdi exhibited a power bounded
operator on
2
which fails to be uniformly mean ergodic, [53, Example 2.4].
It is routine to check that the same operator, say S, now considered to be
acting in
p
, is also power bounded but, not uniformly mean ergodic. Denote
the operator S, when transferred to Z, by R. Then T : Y Z Y Z
given by T(y, z) := (0, Rz), for (y, z) Y Z, is a power bounded operator
on
p
(A) which fails to be uniformly mean ergodic. In particular the Kthe
echelon space
p
(A) is not uniformly mean ergodic.
Recall the separable Frchet spaces L
p
:=

1r<p
L
r
([0, 1]), for p
(1, ), equipped with the seminorms
q
p,(m)
(f) :=
_
_
1
0
[f(t)[
(m)
dt
_
1/(m)
, f L
p
, (2.17)
for any increasing sequence 1 (m) p as m . These spaces, which
are all reexive, have been studied in [11].
Lemma 2.10. Each Frchet space L
p
, for 1 < p < , contains a comple-
mented subspace which is a Banach space isomorphic to
2
.
Proof. Consider the Rademacher functions (r
i
)
i=0
on [0, 1] which are or-
thonormal in L
2
([0, 1]). For every 1 p < , Khinchines inequality ensures
the existence of constants A
p
, B
p
> 0 such that
A
p
_
n
i=0
[
i
[
2
_
1/2
_
_
1
0
i=0
i
r
i
(t)
p
dt
_
1/p
B
p
_
n
i=0
[
i
[
2
_
1/2
, (2.18)
for all choices of scalars (
i
)
n
i=0
and n N
0
, [16, pp.13-14], [36, Theorem
2.b.3]. It follows from (2.18) that on the closed subspace Z, of L
p
, which
is spanned by (r
i
)
i=0
the relative topology from L
p
is a norm topology and
that Z endowed with this topology is canonically isomorphic to
2
.
Fix p (1, ). Given f L
p
, set f, r
i
) :=
_
1
0
f(t)r
i
(t) dt for i =
0, 1, 2, . . .. Then, in the notation of (2.17), we have for each n N
0
(by
(2.18) and Hlders inequality) that
n
i=0
[f, r
i
)[
2
=
_
1
0
f(t)
_
n
i=0
f, r
i
)r
i
(t)
_
dt
q
q,(m)
(f)
_
_
_
n
i=0
f, r
i
)r
i
_
_
_
L
(m)
([0,1])
q
p,(m)
(f)B
(m)

_
n
i=0
[f, r
i
)[
2
_
1/2
,
where 1 < (m)
< satises
1
(m)
+
1
(m)
= 1. Then for n large enough,

_
n
i=0
[f, r
i
)[
2
_
1/2
B
(m)
q
p,(m)
(f)
and hence, by letting n , we see that
_

i=0
[f, r
i
)[
2
_
1/2
B
(m)
q
p,(m)
(f), f L
p
. (2.19)
Accordingly, the linear map P : L
p
L
p
dened by
Pf :=
i=0
f, r
i
)r
i
, f L
p
,
is well dened. To see that P is continuous, x m N. Then, for each
f L
p
, it follows from (2.18) and (2.19) that
q
p,(m)
(Pf) = q
p,(m)
_

i=0
f, r
i
)r
i
_
B
(m)
_

i=0
[f, r
i
)[
2
_
1/2
B
(m)
B
(m)
q
p,(m)
(f).
That P is a projection follows from the fact that (r
i
)
i=0
is an orthonormal
sequence. So, P L(L
p
) is a projection of L
p
onto Z and hence, Z is a
complemented subspace of L
p
which is a Banach space isomorphic to
2
.
Since
2
is not uniformly ergodic, an argument as in the proof of Propo-
sition 2.9 yields the following result.
Proposition 2.11. Let 1 < p < . Then the reexive Frchet space L
p
fails to be uniformly mean ergodic.
As usual will denote C
N
when equipped with the seminorms
q
n
(x) := max
1in
[x
i
[, x , (2.20)
for each n N, and
s = x C
N
: p
n
(x) := sup
iN
i
n
[x
i
[ < , n N (2.21)
is the space of all rapidly decreasing sequences. Both are nuclear Frchet
spaces and hence, also Montel. For A := (a
n
)
n
with a
n
(i) := i
n
, for i N, it
is known that s =
p
(A) for all p [1, ], [40, p.360]. For each p [1, ),
the sequence space
p+
:=

q>p
q
is a separable, reexive Frchet space
when equipped with the seminorms
q
n,p
(x) :=
_

i=1
[x
i
[
(n)
_
1/(n)
, x
p+
, (2.22)
where (n) := p +
1
n
for n N. This family of spaces was studied in [41].
For Banach spaces, the following criterion is due to R.C. Sine, [49].
Theorem 2.12. Let X be a lcHs and T L(X) be power bounded with
ker(I T) = 0. Then T is mean ergodic if and only if ker(I T
t
) = 0.
Proof. Let T be mean ergodic. Suppose that ker(I T
t
) ,= 0. Then there
exist u ker(I T
t
) X
and x X satisfying u, x) = 1. It follows from

u = T
t
u that T
m
x, u) = x, (T
t
)
m
u) = x, u) for all m N and hence, via
(1.5), that T
[n]
x, u) = x, u) = 1 for all n N. That is, lim
n
T
[n]
x, u) =
1. The mean ergodicity of T ensures that x
0
:= lim
n
T
[n]
x exists. It
follows from (1.5) that
Tx
0
= lim
n
1
n
n
m=1
T
m+1
x = lim
n
_
T
[n]
x
1
n
Tx +
1
n
T
n+1
x
_
.
Clearly
1
n
Tx 0 and also
1
n
T
n+1
x 0 as T is mean ergodic; see (2.2). So,
x
0
= Tx
0
and hence, x
0
= 0 because ker(I T) = 0. That is, T
[n]
x 0
and hence, T
[n]
x, u) 0; contradiction. So, ker(I T
t
) = 0.
Now suppose that ker(I T
t
) = 0. If X ,= Im(I T), then there exists
x
0 whose restriction to Im(I T) is zero, that is,

x, (I T
t
)x
) = (I T)x, x
) = 0, x X.
Accordingly, x
is a non-zero element of ker(I T

t
); contradiction. So X =
Im(I T). According to Proposition 2.1 we have
X = Im(I T) = x X : lim
n
T
[n]
x = 0,
that is, T
[n]
0 in L
s
(X). Hence, T is mean ergodic.
Proposition 2.13. Let A be any Kthe matrix. Then the following asser-
tions are equivalent.
(i)
(A) is mean ergodic.

(ii)
(A) is uniformly mean ergodic.

(iii)
(A) is a Montel space.

(iv)
(A) does not contain an isomorphic copy of
.
(v)
0
(vi)
0
(vii)
0
(A) is a Montel space.
(viii)
0
(A) does not contain an isomorphic copy of c
0
.
(ix)
1
(x)
1
(xi)
1
(A) is Montel.
(xii)
1
(A) does not contain an isomorphic copy of
1
.
Proof. (iii) (ii) by Proposition 2.8 and (ii) (i) is obvious.
Suppose that
(A) is not Montel. According to [9, Proposition 2.3] we

have
(A) = Y Z with Z isomorphic to
. Hence there exists a power

bounded operator on
(A) which is not mean ergodic. This establishes (i)

(iii) and also (iv) (iii) was established along the way.
(iii) (iv) is trivial.
(iii) (iv) is known and is equivalent to
0
(A) =
(A); see for example,

[40, Theorems 27.9 and 27.15].
Again (vii) (vi) by Proposition 2.8 and (vi) (v) is clear.
Suppose that
0
(A) is not Montel. Then
0
(A) contains a sectional sub-
space which is complemented in
0
(A) and is isomorphic to c
0
; see, for ex-
ample, the proof of Proposition 2.5(ii) in [7] which also applies when p = 0
there. Thus, we have established that (v) (vii) and (viii) (vii).
(vii) (viii). Since c
0
is not reexive, the same argument used for the
proof of (iii) (iv) is again applicable.
(iii) (xi) is immediate from [40, Proposition 27.9].
(xi) (xii). Since
1
is not reexive, the same argument used for the
proof of (iii) (iv) is again applicable.
Suppose that
1
(A) is not Montel. Then
1
(A) contains a sectional sub-
space which is complemented in
1
(A) and is isomorphic to
1
. This implies
(ix) (xi) and (xii) (xi).
Finally, (xi) (x) by Proposition 2.8 and(x) (ix) is clear.
Remark 2.14. (i) A lcHs which contains an isomorphic copy of
contains
it as a complemented subspace, [27, Corollary 7.4.6]. Therefore every lcHs
which contains a copy of
is not mean ergodic.

(ii) An analogue of part (i) is also available for c
0
and is based on the
following version of
Sobczyks Theorem: Let X be a separable lcHs which contains an
isomorphic copy of c
0
. Then X contains a complemented copy of c
0
.
Its proof proceeds as follows: one modies in an obvious way the last part
of the proof of the classical Banach space version of Sobczyks theorem, as
presented in [27, p.160], by replacing there the use of Theorem 7.4.4 of [27]
with Corollary 7.4.5 of [27].
Each Frchet space
p+
, for 1 p < , has no closed subspace isomorphic
to any (innite dimensional) Banach space, [41, p.10]. So, an argument as
in the proof of Proposition 2.11 cannot be used for
p+
.
The following class of multiplication operators will be useful. Let X denote
any one of the sequence spaces , s,
p
, c
0
,
p+
or
p
(A). Given any sequence
of numbers 0 <
i
< 1, for i N, dene a linear operator T
()
L(X) by
T
()
x := (
i
x
i
), x X. (2.23)
Direct calculation shows, for each n N, that
T
()
[n]
x =
1
n
_

i
(1
i
)
(1
n
i
)x
i
_
i
x X. (2.24)
Suppose, in addition, that
i
1. Then there exists an increasing sequence
of integers n
i
satisfying n
i

1
(1
i
)
< 1 +n
i
, for i N, and hence,
1
n
i
i
(1
i
)
(1
n
i
i
)
1
_
1
_
1 (n
i
+ 1)
1
_
n
i
_
, i N. (2.25)
Proposition 2.15. For each 1 p < , the reexive Frchet space
p+
is
mean ergodic but, not uniformly mean ergodic.
Proof. First note that the Banach space
p
is continuously included in
p+
and hence, the unit ball B of
p
is a bounded set in
p+
. Let 0 <
i
1 and
dene the power bounded operator T
()
L(
p+
) via (2.23). Let q denote
any one of the seminorms (2.22). If e
i
B denotes the standard i-th unit
basis vector of
p
, then it follows from the denition of q and (2.24) that
q
_
T
()
[n]
e
i
_
=
1
n
i
(1
i
)
(1
n
i
), i, n N.
Since
p+
is mean ergodic, there is P L(
p+
) with T
()
[n]
P in L
s
(
p+
)
as n . By the above identities, P(e
i
) = 0 for all i N. But, (e
i
)
iN
is a basis for
p+
, [41, p.8], and so P = 0. By (2.25) and the fact that its
right-hand side converges to
1
(1 e
1
) > 0 we can conclude that
lim
n
sup
xB
q
_
T
()
[n]
x
_
,= 0,
that is,
_
T
()
[n]
_
fails to converge to 0 in L
b
(
p+
) as n . Hence, T
()
is
not uniformly mean ergodic.
For X a Banach space, the next result is due to M. Lin, [35].
Proposition 2.16. Let X be a Frchet space and T L(X) satisfy both
ker(I T) = 0 and
1
n
T
n
0 in L
b
(X). Consider the following statements.
(i) I T
[n]
is surjective for some n N.
(ii) I T is surjective.
(iii) T
[n]
0 in L
b
(X) as n .
Then (i) (ii) (iii). If X is a Banach space, then also (iii) (i).
Proof. (i) (ii). Suppose that I T
[n]
is surjective. It is routine to verify
that I T
[n]
= (I T)g
n
(T), where g
n
(T) :=
1
n
n1
r=0
(
r
j=0
T
j
). Given
y X there is, by hypothesis, x X such that (I T
[n]
)x = y and hence,
(I T)g
n
(T)x = y. So, (I T) is surjective.
(ii) (iii). Fix a bounded set B X and q
X
. Since (I T) : X X
is continuous and bijective and X is a Frchet space, it follows that (I T)
is a bicontinuous isomorphism. So, C := (I T)
1
B is a bounded set in X
and B = (I T)C. In the notation of (1.4) it follows from (2.1) that
q
B
(T
[n]
) = sup
xC
q
_
(I T)T
[n]
_
= sup
xC
q
_
1
n
(T T
n+1
)x
_
1
n
q
C
(T) + 2q
C
_
1
(n + 1)
T
n+1
_
.
Since we are assuming that
1
n
T
n
0 in L
b
(X), it follows that q
B
(T
[n]
) 0.
But, q and B are arbitrary and so T
[n]
0 in L
b
(X). Hence, (iii) holds.
Suppose now, in addition, that X is a Banach space. Given (iii), it follows
that |T
[n]
| < 1 for some n N. It is then classical that (I T
[n]
) is invertible
in L(X), [19, Ch. VII, 3], and so (i) holds.
We now show that (iii) (ii) may fail in the non-normable setting.
Example 2.17. Dene T L(s) by
Tx :=
_
(1 2
i
)x
i
_
, x = (x
i
) s.
It is routine to check that ker(I T) = 0 and that y := (2
i
) s does
not belong to Im(I T), that is, (I T) is not surjective. So (ii) of Propo-
sition 2.16 fails to hold.
Since T
m
x =
_
(1 2
i
)
m
x
i
_
for x s and all m N, it follows from
(2.21) (with the notation from there) that
p
n
(T
m
x) p
n
(x), x s, (2.26)
for all n, m N. Given any bounded set B s, it follows that
(p
n
)
B
_
1
m
T
m
_
= sup
xB
p
n
_
1
m
T
m
x
_
1
m
sup
xB
p
n
(x), m N.
Hence, (p
n
)
B
_
1
m
T
m
_
0 as m . That is,
1
m
T
m
0 in L
b
(s).
It remains to verify condition (iii) of Proposition 2.16. The dual space of
s is given by
s
= C
N
: k N with sup
i
[
i
[
i
k
<
and direct calculation shows that
T
t
=
_
(1 2
i
)
i
_
, s
.
It follows that ker(I T
t
) = 0 and, by (2.26), we see that T is power
bounded. According to Theorem 2.12 the operator T is mean ergodic. Since
s is Montel, T is actually uniformly mean ergodic (see Proposition 2.8). So,
there exists P L(s) such that T
[n]
P in L
b
(s). In particular, T
[n]
P
in L
s
(s). Fix r N and let e
r
denote the element of s with a 1 in the r-th
coordinate and 0s elsewhere. Then
T
n
e
r
= (1 2
r
)
n
e
r
0 as n
and hence, also T
[n]
e
r
=
(1)
n
n(1)
e
r
0 as n (where := 1 2
r
).
But, T
[n]
e
r
Pe
r
as n and we conclude that Pe
r
= 0. Since r N
is arbitrary, it follows that P = 0 and hence, T
[n]
0 in L
b
(s) which is
precisely condition (iii).
We conclude this section by exhibiting an interesting family of power
bounded operators. Fix 1 < p < . A Borel measurable function : [0, 1]
C denes a continuous multiplication operator M
L(L
p
) via f f,
for f L
p
, if and only if
1q<
L
q
([0, 1]), [6, Proposition 18].
Proposition 2.18. Let 1 < p < and
1q<
L
q
([0, 1]). The follow-
ing assertions for the multiplication operator M
L(L
p
) are equivalent.
(i) L
([0, 1]) with ||
1.
(ii) M
is power bounded.
(iii) M
is mean ergodic.
(iv) M
is uniformly mean ergodic.

Proof. Observe that always (M
)
n
= M
n, for n N, and so we have

q
p,m
((M
)
n
f) = q
p,m
(
n
f), f L
p
, (2.27)
for each m N and all n N (the notation is from (2.17)).
(i) (ii). Fix m N. It follows from (2.27) and (2.17) that q
p,m
((M
)
n
f)
|
n
|
q
p,m
(f) for all n N and f L
p
. Since (i) implies that |
n
|
1
for all n N, we see that
q
p,m
((M
)
n
f) q
p,m
(f), f L
p
, n N,
that is, (M
)
n
n=1
is equicontinuous and so M
is power bounded.
(ii) (iii). This is immediate from the reexivity of L
p
and Corollary 2.7.
(iii) (iv). Set T := M
. By assumption there exists Q L(L

p
) such
that T
[n]
Q in L
s
(L
p
) and hence, (2.2) implies that
1
n
T
n
=
1
n
M
n 0
in L
s
(L
p
) as n . For the choice f :=
[0,1]
L
p
we can conclude that
n
n
0 in L
p
. But, L
p
is continuously included in L
1
([0, 1]) and so

n
n
0
in L
1
([0, 1]). Accordingly, there exists a subsequence

n(k)
n(k)
0 a.e. in [0, 1]
as k , which implies that [[ 1 a.e. For each z C with [z[ 1, it is
routine to check that
1
n
n
k=1
z
k
1 if z = 1 and
1
n
n
k=1
z
k
0 otherwise
(as n ). Hence, ||
1 implies that
1
n
n
k=1
k

A
a.e. on [0, 1], (2.28)
as n , where A := t [0, 1] : (t) = 1.
For each Borel set E [0, 1], dene the projection P(E) := M
L(L
p
). Then E P(E) is a spectral measure in L
p
(see Section 4 for the
denition) which is boundedly -additive, [6, Proposition 6(iii)]. Since the
lcHs L
p
(L
p
) in which P takes its values is quasicomplete, [32, p.144], it is
known that all bounded measurable functions are P-integrable, [34, p.161].
In particular,
1
n
n
k=1
k
is P-integrable for each n N. It follows from
(2.28) and the dominated convergence theorem for the L
b
(L
p
)-valued vector
(= spectral) measure P, [34, Theorem 2.2], that
_
[0,1]
(
1
n
n
k=1
k
) dP
P(A) in L
b
(L
p
), that is, T
[n]
P(A) = Q in L
b
(L
p
) as n . Hence,
M
= T is uniformly mean ergodic.

(iv) (i). If M
is uniformly mean ergodic, then it is mean ergodic.

Hence, L
([0, 1]) with ||
1; see the proof of (iii) (iv) above.

Since the containment L
([0, 1])

1q<
L
q
([0, 1]) is proper, Propo-
sition 2.18 implies that there exist multiplication operators M
L(L
p
)
which fail to be power bounded.
3. On Bases and Reflexivity in Lchs
The main aim of this section is to establish Theorems 1.1 and 1.2.
A sequence (x
n
)
n
in a lcHs X is called a basis if, for every x X, there is
a unique sequence (
n
)
n
C such that the series

n=1
n
x
n
converges to
x in X. By setting f
n
(x) :=
n
we obtain a linear form f
n
: X C which is
called the n-th coecient functional associated to (x
n
)
n
. The functionals f
n
,
n N, are uniquely determined by (x
n
)
n
and (x
n
, f
n
)
n=1
is a biorthogonal
sequence (i.e. x
n
, f
m
) =
mn
for m, n N). For each n N, the map
P
n
: X X dened by
P
n
: x
n
i=1
f
i
(x)x
i
=
n
i=1
x, f
i
)x
i
, x X, (3.1)
is a linear projection with range equal to the nite dimensional space span(x
i
)
n
i=1
.
If, in addition, (f
n
)
n
X
, then the basis (x

n
)
n
is called a Schauder basis
for X. In this case, (P
n
)
n
L(X) and each dual operator
P
t
n
: x
i=1
x
i
, x
)f
i
, x
, (3.2)
for n N, is a projection with range equal to span(f
i
)
n
i=1
. Moreover, for
every x
the series

i=1
x
i
, x
)f
i
converges to f in X
(the space X
equipped with the topology (X
, X)). For this reason, (f

n
)
n
is also referred
to as the dual basis of the Schauder basis (x
n
)
n
. The terminology X has
a Schauder basis will also be abbreviated simply to saying that X has a
basis. A sequence (x
n
)
n
in a lcHs X is called a basic sequence if it is a
Schauder basis for the closed linear hull span(x
n
)
n
of (x
n
)
n
in X.
Let X be a Frchet space with a fundamental sequence of seminorms
X
= (q
n
)
n
. Then X
n
denotes the local Banach space generated by q
n
, that
is, X
n
is the completion of the quotient normed space (X/q
1
n
(0), q
n
). Let
n
: X X
n
denote the canonical map. Then X = proj
n
X
n
is the (reduced)
projective limit of the sequence of Banach spaces (X
n
)
n
, [31, p.232].
Now the promised extension of Pelcynskis result. We wish to thank Prof.
J.C. Daz for some useful discussions on this topic.
Proof of Theorem 1.1. Let X be a non-reexive Frchet space and
X
= (q
n
)
n
be an increasing, fundamental sequence of seminorms. If X
contains an isomorphic copy of
1
, then it surely contains a non-reexive,
closed subspace with a basis.
So, suppose that X does not contain an isomorphic copy of
1
. According
to [31, p.303 & pp.312-313], the non-reexivity of X ensures the existence
of a bounded sequence in X with no weakly convergent subsequence and
hence, by Rosenthals dichotomy theorem for Frchet spaces [13, Lemma 3],
X contains a sequence (x
k
)
k
which is Cauchy but, not convergent in X
. For
each n N, the sequence (
n
(x
k
))
k
is (X
n
, X
n
)-Cauchy in the Banach space
X
n
. Moreover, there exists n(0) N such that (
n
(x
k
))
k
is not convergent
in (X
n
)
for all n n(0). Indeed, if the contrary were the case, then there
would exist positive integers n(m) such that (
n(m)
(x
k
))
k
converges in
(X
n(m)
)
, for all m N. Since X = proj

m
X
n(m)
, it follows that (x
k
)
k
is
convergent in X
which contradicts the choice of (x

k
)
k
.
Now, (
n(0)
(x
k
))
k
is Cauchy but, not convergent in (X
n(0)
)
. According to
[15, p.54 Ex. 10(ii)], there exists a subsequence (
n(0)
(x
0
k
))
k
which is a basic
sequence in X
n(0)
. Then the sequence (
1+n(0)
(x
0
k
))
k
X
1+n(0)
is Cauchy
but, not convergent in (X
1+n(0)
)
. If it were convergent then, being a subse-

quence of the weak Cauchy sequence (
1+n(0)
(x
k
))
k
, also this latter sequence
would converge in (X
1+n(0)
)
; contradiction! So, we can again apply [15, p.54

Ex. 10(ii)] to select a subsequence (
1+n(0)
(x
1
k
))
k
of (
1+n(0)
(x
0
k
))
k
which is
a basic sequence in X
1+n(0)
. Continue this procedure inductively and, -
nally, select the diagonal subsequence (x
k
k
)
kn(0)
. According to [13, Lemma
1], we see that (x
k
k
)
kn(0)
is a basic sequence in X. Since this sequence is
(X, X
)-Cauchy but, has no convergent subsequence in X
, it follows that
span(x
k
k
)
kn(0)
is not reexive.
The strong topology on X (resp. X
) is denoted by (X, X
) (resp.
(X
, X)) and we write X
(resp. X
). For a Schauder basis (x

n
)
n
X,
we denote by H the subspace of X
consisting of all x
such that
P
t
n
x
in X
as n and endow it with the topology (H, X) in-

duced by (X
, X). If H = X
(i.e. the dual basis (f

n
)
n
is a Schauder
basis for X
), then (x
n
)
n
is called shrinking. A Schauder basis (x
n
)
n
X is
called boundedly complete if the series

n=1
n
x
n
converges in X whenever
(
n
i=1
i
x
i
)
n
is a bounded sequence in X. For Banach spaces, these notions
are due to R.C. James, [26], and for lcHs they are due to E. Dubinsky and
J.R. Retherford, [17], [44].
Corollary 3.1. Let X be a non-reexive Frchet space. Then X contains a
closed subspace with a basis which is not shrinking.
Proof. According to Theorem 1.1, X contains a basic sequence (x
n
)
n
such
that E := span(x
n
)
n
is non-reexive. Then [28, Theorem 3.3(ii)] implies that
E (hence, also X) has a basic sequence (u
k
)
k
which is not shrinking. So,
span(u
k
)
k
is a closed subspace of X with a basis which is not shrinking.
A Schauder basis (x
n
)
n
of a lcHs X is called regular if there exists a
neighbourhood V of zero such that x
n
/ V for all n N. Equivalently, there
exists q
X
such that inf
n
q(x
n
) > 0. If, in addition, the sequence (x
n
)
n
is bounded in X, then it is called normalized. Even if X is a Frchet space,
it does not necessarily follow that a given basis can be regularized, [30].
Proposition 3.2. Let X be a Frchet space which is not Montel. Then X
contains a closed subspace which is not Montel and has a basis.
Proof. According to [5, Proposition 2.2], the space X possesses a (innite)
normalized basic sequence, say (x
n
)
n
; a proof of this result which avoids
non-standard analysis is also available (see [5, p.205 Remark (2)]). Hence,
Y := span(x
n
)
n
is a Frchet space with a normalized Schauder basis. But, it
is routine to check that such a basis cannot exist in any innite dimensional
Montel Frchet space. So, Y is not Montel.
Given a lcHs X, a biorthogonal sequence (x
n
, f
n
)
n=1
in X X
is said
to be of type P if there exists a neighbourhood V of zero in X such that
x
n
/ V , for all n N, and if the sequence of partial sums (
n
i=1
x
i
)
n
is
bounded in X. We say (x
n
, f
n
)
n
is of type P
if (x
n
)
n
is bounded in X
and (
n
i=1
f
i
)
n
is bounded in X
. Bases and basic sequences of types P

and P
were introduced and studied in Banach spaces by I. Singer, [50].

The denition was extended by E. Dubinsky and J.R. Retherford, [17], to
bases and basic sequences in lcHs and by N. Kalton, [28], to biorthogonal
sequences.
The behavior of bases and basic sequences in reexive Banach spaces has
been completely characterized. Recall that R.C. James proved that a Ba-
nach space with a Schauder basis is reexive if and only if the basis is both
shrinking and boundedly complete, [26]. M. Zippin showed, for a Banach
space X with a basis, that if every basis is boundedly complete or if every
basis is shrinking, then X is reexive, [55]. Jamess characterization of re-
exivity was generalized to lcHs by Dubinsky and Retherford [17], [44] (see
also [27, Theorem 14.5.1]). Furthermore, Kalton extended both Singers and
Zippins results to cover lcHs more general than Banach spaces, [28]. In par-
ticular, he proved that a sequentially complete lcHs with a Schauder basis
in which every basic sequence is boundedly complete or every basic sequence
is shrinking is necessarily semi-reexive. He also showed that a complete
barrelled lcHs with a normalized Schauder basis in which every normalized
basis is boundedly complete, or in which every normalized basis is shrink-
ing, is reexive. As noted in the Introduction, Kalton raised the question of
whether the last result remains true without the restriction of normalization
on the basis. By using ideas of [28] and [55], we now proceed to show that
the answer is positive. As usual, we begin with some auxiliary results.
Remark 3.3. Let X be a complete barrelled lcHs and (x
n
)
n
be a sequence
in X 0 for which X = span(x
n
)
n
. Then (x
n
)
n
is a Schauder basis for X
if and only if for each p
X
there exist M
p
> 0 and q
X
such that
p
_
r
i=1
i
x
i
_
M
p
q
_
s
i=1
i
x
i
_
(3.3)
for arbitrary positive integers r s and arbitrary scalars
1
, . . . ,
s
; (see
[45, Theorems 3.1-3.2] or [27, Theorem 14.3.6]). By setting
p(x) := sup
rN
p
_
r
i=1
x, f
i
)x
i
_
,
we obtain from (3.3) that
p(x) p(x) M
p
q(x) M
p
q(x), x X, (3.4)
for all p
X
. Hence,

X
:= p : p
X
is also a system of continuous
seminorms generating the topology of X and we have
p
_
r
i=1
i
x
i
_
p
_
s
i=1
i
x
i
_
for every p

X
and for arbitrary positive integers r s and arbitrary
scalars
1
, . . . ,
s
. That is, (x
n
)
n
is a monotone basis with respect to

X
.
Lemma 3.4. Let X be a complete barrelled lcHs with a Schauder basis (x
n
)
n
.
Then X = proj
iI
X
i
is the projective limit of a system (X
i
)
iI
of local
Banach spaces such that (
i
(x
n
))
n
is a Schauder basis in X
i
for every i I,
where
i
denotes the canonical projection of X into X
i
.
Proof. According to Remark 3.3, we can select a system
X
of continuous
seminorms dening the topology of X with respect to which the basis (x
n
)
n
is monotone, that is, for every p
X
and for arbitrary positive integers
r s and arbitrary scalars
1
, . . . ,
s
we have
p
_
r
i=1
i
x
i
_
p
_
s
i=1
i
x
i
_
. (3.5)
In the case that X admits a continuous norm, we can assume that each
p
X
is a norm. Then the completion X
p
of the normed space (X, p) is
a Banach space. Clearly, X = proj
p
X
p
with X continuously embedded
in every X
p
, i.e.,
p
(x) = x. In particular, by (3.5) and Remark 3.3 we see
that (x
n
)
n
is a Schauder basis in every X
p
.
Suppose now that X does not admit a continuous norm. Denote by (f
n
)
n
the dual basis of (x
n
)
n
. For each p
X
, the space Ker(p) is then innite
dimensional and
Ker(p) = x X : p(x) = 0 =
nJ(p)
Ker(f
n
) , (3.6)
where J(p) := n N : p(x
n
) ,= 0. Indeed, if n J(p) (hence, p(x
n
) ,= 0),
then we obtain from (3.5) that
[f
n
(x)[ =
1
p(x
n
)
p(x, f
n
)x
n
)
2
p(x
n
)
p(x), x X,
thereby implying that, if x Ker(p), then x, f
n
) = 0 for all n J(p)
and hence, x
nJ(p)
Ker(f
n
). Conversely, if x
nJ(p)
Ker(f
n
), then
x =

iJ(p)
x, f
i
)x
i
and hence, 0 = p(
n
i=1,iJ(p)
x, f
i
)x
i
) p(x), i.e.,
p(x) = 0 because (x
n
)
nJ(p)
Ker(p).
Denote by
p
the canonical quotient map from X onto X/ ker(p) and by
p the quotient norm on X/ ker(p) given by
p(
p
(x)) := infp(y) :
p
(y) =
p
(x) .
Then the local Banach space X
p
is the completion of the normed space
(X/ ker(p), p) and X = proj
p
X
X
p
. It remains to show that (
p
(x
n
))
nN
=
(
p
(x
n
))
nJ(p)
is a Schauder basis in X
p
.
Suppose that J(p) = (n(i))
i
, where (n(i))
i
is either a nite sequence or
an increasing sequence of positive integers. If (n(i))
i
is a nite sequence,
then it is routine to check that (
p
(x
n(i)
))
i
is a linearly independent set and
X/ ker(p) = span(x
n(i)
)
i
.
Suppose then that (n(i))
i
is an increasing sequence of positive integers.
Let p
_
m+r
i=1

i
p
(x
n(i)
)
_
= 1. For any > 0 there exists x =

n=1
n
x
n
such that p(x) 1 + and
p
(x) =

n=1
p
(x
n
) =

m+r
i=1

i
p
(x
n(i)
).
This means that
n=1
n
x
n

m+r
i=1
i
x
n(i)
Ker(p) =
i
Ker(f
n(i)
).
In view of the biorthogonality, we obtain that
n(i)
= 0 if i > m + r, and
n(i)
= a
i
if i 1, . . . , m+r. Therefore, by (3.5) we have
p
_
m
i=1
p
(x
n(i)
)
_
= p
_
m
i=1
n(i)
p
(x
n(i)
)
_
p
_
m
i=1
n(i)
x
n(i)
_
p
_
m+r
i=1
n(i)
x
n(i)
_
= p
_

n=1
n
x
n
_
1 + . (3.7)
The last equality follows from p
_
m+r
i=1

n(i)
x
n(i)
_
= p (
s
n=1
n
x
n
) p(x),
which is a consequence of
p
_
s
n=1
n
x
n
_
= p
_
_
s
n=1,nJ(p)
n
x
n
+
s
n=1,nJ(p)
n
x
n
_
_
= p
_
_
s
n=1,nJ(p)
n
x
n
_
_
because p(x
n
) = 0 if n , J(p). Since (3.7) holds for all m, r and , we can
conclude that (
p
(x
n(i)
))
i
is a Schauder basis for the Banach space X
p
. In
particular, (f
n(i)
)
i
is the dual basis of (
p
(x
n(i)
))
i
.
Let X be a lcHs. For each p
X
, we set U
p
:= x X : p(x) 1 and
dene the dual seminorm p
of p by
p
(u) := sup[x, u)[ : p(x) 1 = sup[x, u)[ : p(x) = 1, u X
,
that is, p
is the gauge of the polar U
p
in X
. Let X
p
:= u X
: p
(u) <
. Then (X
p
, p
) is a Banach space and the transpose map

t
p
of the
canonical quotient map
p
is an isometry from the strong dual of the Banach
space X
p
(i.e. the completion of (X/ ker(p), p)) onto (X
p
, p
). Therefore,
every v (X/ ker(p), p)
denes a continuous linear functional u = v

p
X
with p
(u) < .
We now formulate a useful criterion for turning algebraic bases into Schauder
bases. This criterion is due to M. Zippin in the Banach space setting, [55,
Lemma 2], and was later extended by W.B. Robinson to complete barrelled
lcHs, [48, Theorem 1.1].
Lemma 3.5. Let X be a complete barrelled lcHs and (x
n
)
n
be a Schauder
basis of X.
For each k N, let (y
i
)
p(k+1)
i=p(k)+1
be a basis of span(x
i
)
p(k+1)
i=p(k)+1
, where
(p(k))
k
is an increasing sequence of positive integers with p(1) = 0. Sup-
pose, for each q
X
, that there exist M
q
> 0 and r
X
such that
q
_
_
m
i=p(k)+1
i
y
i
_
_
M
q
r
_
_
n
i=p(k)+1
i
y
i
_
_
(3.8)
for all k N, for all integers m, n with p(k) < m n p(k +1) and for all
scalars (
i
)
n
i=p(k)+1
.
Then the sequence (y
i
)
i
is a Schauder basis of X.
Remark 3.6. If the basis (x
n
)
n
in Lemma 3.5 is monotone relative to
X
and (3.8) holds with q in the place of r in its right-hand side, then it can be
shown that, for each q
X
, there exists C
q
> 0 such that
q
_
r
i=1
i
y
i
_
C
q
q
_
s
i=1
i
y
i
_
(3.9)
1
, . . . ,
s
.
We now present a criterion for extending Schauder block sequences to
Schauder bases for the whole space.
Lemma 3.7. Let X be a complete barrelled lcHs and (x
n
)
n
be a Schauder
basis of X.
Assume that 0 ,= y
k
=

p(k+1)
i=p(k)+1
i
x
i
where (p(k))
k
is an increasing
sequence of positive integers with p(1) = 0 so that, for some p
0

X
,
inf
kN
p
0
(y
k
) = d > 0 (3.10)
and, for every q
X
,
sup
kN
q(y
k
) M
q
< . (3.11)
Then there exists a Schauder basis (z
i
)
i
in X such that, for each k N,
we have z
i
= y
k
for some i p(k) + 1, . . . , p(k + 1).
Proof. As X is barrelled, we can suppose that (x
n
)
n
is a monotone basis with
respect to
X
. Indeed, by Remark 3.3 there exists a system
X
of continuous
seminorms dening the topology of X such that (x
n
)
n
is a monotone basis
with respect to

. Therefore, conditions (3.10) and (3.11) continue to be
hold. In particular, condition (3.10) is satised with p
0

X
; see (3.4).
According to Lemma 3.4, (
p
0
(x
n
))
n
is a Schauder basis in the local Banach
space X
p
0
. In particular, by setting J := n N : p
0
(x
n
) ,= 0 we have
(
p
0
(x
n
))
n
= (
p
0
(x
n
))
nJ
and
p
0
(x
n
) = 0 for all n , J. Moreover, by (3.6)
and (3.10) we have 0 ,=
p
0
(y
k
) =
p(k+1)
i=p(k)+1,iJ

i
p
0
(x
i
).
Fix k N and set E
k
= span(x
i
)
p(k+1)
i=p(k)+1
, in which case
p
0
(E
k
) =
span(
p
0
(x
i
))
p(k+1)
i=p(k)+1,iJ
. Since 0 ,=
p
0
(y
k
)
p
0
(E
k
), there exists i
k

p(k) + 1, . . . , p(k + 1) J so that
p
0
(y
k
) , span(
p
0
(x
i
))
p(k+1)
i=p(k)+1,iJ,i=i
k
.
To see this, observe that if for every i p(k) + 1, . . . , p(k + 1) J
we have
p
0
(y
k
) span(
p
0
(x
j
))
p(k+1)
j=p(k)+1,jJ,j=i
, then
i
= 0 for all i
p(k) + 1, . . . , p(k + 1) J, thereby implying that
p
0
(y
k
) = 0 and hence,
that p
0
(y
k
) = p
0
(
p
0
(y
k
)) = 0 which contradicts (3.6).
So, the set
p
0
(y
k
)
p
0
(x
i
)
i{p(k)+1,...,p(k+1)}J,i=i
k
is linearly inde-
pendent in the local Banach space X
p
0
. Accordingly, there exists w
k
X
p
0
for which
p
0
(y
k
), w
k
) = 1 and
p
0
(x
i
), w
k
) = 0 for all i p(k) +
1, . . . , p(k + 1) J and i ,= i
k
. Note that p
0
(w
k
) 1/d as p
0
(y
k
) > d.
In particular, by (3.6) we also have
p
0
(x
i
), w
k
) = 0 for all i p(k) +
1, . . . , p(k + 1) J. Clearly, the set dened by
z
i
=
_
x
i
if i p(k) + 1, . . . , p(k + 1) and i ,= i
k
y
k
if i = i
k
,
is also linearly independent. Thus, (z
i
)
p(k+1)
i=p(k)+1
is a basis of E
k
.
Dene a linear map P
k
: E
k
span(z
i
)
p(k+1)
i=p(k)+1
by P
k
z = z, v
k
)z
i
k
, where
v
k
:= w
k

p
0
X
so that p
0
(v
k
) 1/d. Then P
k
z
i
= 0 if i ,= i
k
and
P
k
z
i
k
= z
i
k
. Moreover, by (3.11) we obtain, for each q
X
with q p
0
,
that
q(P
k
z) p
0
(v
k
)p
0
(z)q(z
i
k
)
M
q
d
q(z) . (3.12)
For such a q, x p(k) < r < s p(k + 1) and scalars (
i
)
p(k+1)
i=p(k)+1
. If
r i
k
, then (3.5) and (3.12) imply that
q(
r
i=p(k)+1
i
z
i
) q
_
_
r
i=p(k)+1,i=i
k
i
x
i
_
_
+q(
i
k
z
i
k
)
q
_
_
s
i=p(k)+1,i=i
k
i
x
i
_
_
+q(
i
k
z
i
k
)
q
_
_
s
i=p(k)+1
i
z
i
_
_
+ 2q(
i
k
z
i
k
) (3.13)
= q
_
_
s
i=p(k)+1
i
z
i
_
_
+ 2q
_
_
P
k
_
_
s
i=p(k)+1
i
z
i
_
_
_
_
_
1 + 2
M
q
d
_
q
_
_
s
i=p(k)+1
i
z
i
_
_
.
If r satises p(k) < r < i
k
s, then again by (3.5) and (3.12) we have
q
_
_
r
i=p(k)+1
i
z
i
_
_
q
_
_
s
i=p(k)+1,i=i
k
i
z
i
_
_
(3.14)
q
_
_
s
i=p(k)+1
i
z
i
_
_
+q(
i
k
z
i
k
)
_
1 +
M
q
d
_
q
_
_
s
i=p(k)+1
i
z
i
_
_
.
Finally, if r, s satisfy p(k) < r < s < i
k
, then by (3.5) we have
q
_
_
r
i=p(k)+1
i
z
i
_
_
q
_
_
s
i=p(k)+1
i
z
i
_
_
. (3.15)
Inequalities (3.13), (3.14) and (3.15) allow us to conclude that there exists
M
q
:= (1 + 2
M
q
d
), for each q
X
, such that the sequence (z
i
)
i
satises the
inequality (3.8) for every q
X
. According to Lemma 3.5, (z
i
)
i
is then a
Schauder basis of X.
Theorem 1.2 will now be presented as two separate results.
Theorem 3.8. Let X be a complete barrelled lcHs with a Schauder basis
(x
n
)
n
. Assume that all the bases in X are shrinking. Then X is reexive.
Proof. Suppose that X is not reexive. Then, by a result of Retherford [44,
Theorem 2.3] (see, also, [12, Theorem 4] or [27, Theorem 14.5.1]), (x
n
)
n
is not boundedly complete. Let
X
be a system of continuous seminorms
dening the topology of X with respect to which the Schauder basis (x
n
)
n
is monotone. Since (x
n
)
n
is not boundedly complete, there exists a sequence
of scalars (
i
)
i
such that
sup
nN
q
_
n
i=1
i
x
i
_
= M
q
< , q
X
, (3.16)
and

i=1
i
x
i
does not converge. Accordingly, there exist p
0

X
and an
increasing sequence (p(k))
k
of positive integers with p(1) = 0 satisfying
inf
kN
p
0
_
_
p(k+1)
i=p(k)+1
i
x
i
_
_
= d > 0 (3.17)
and
sup
kN
q
_
_
p(k+1)
i=p(k)+1
i
x
i
_
_
2M
q
< , q
X
. (3.18)
For each k N, let y
k
=
p(k+1)
i=p(k)+1
i
x
i
. Then, by (3.17), p
0
(y
k
) d for all
k N and, by (3.16),
q
_
k
s=1
y
s
_
= q
_
_
p(k+1)
i=1
i
x
i
_
_
M
q
, k N, q
X
. (3.19)
By (3.17) and (3.18) we can apply Lemma 3.7 (and Remark 3.6) to con-
clude that there exist an increasing sequence (i
k
)
k
of integers with i
k

p(k) +1, . . . , p(k +1), for k N, and a Schauder basis (z
i
)
i
in X given by
z
i
=
_
x
i
if i ,= i
k
for all k,
y
k
if i = i
k
(3.20)
such that, for every q
X
with q p
0
, there exists C
q
> 0 such that
q
_
r
i=1
i
z
i
_
C
q
q
_
s
i=1
i
z
i
_
, (3.21)
1
, . . . ,
s
.
Denote by (f
i
)
i
the dual basis of (z
i
)
i
. Observe also that, if (v
n
)
n
denotes
n
)
n
, then span(f
i
)
i=1
= span(v
n
)
n=1
by (3.20). It follows
from (3.17), (3.19) and (3.21) that the sequence (z
i
k
, f
i
k
)
k
is of type P and
p
0
(f
i
k
) 2
C
p
0
d
, k N. (3.22)
For each i N, dene
u
i
=
_
z
i
if i ,= i
k
for all k,
k
r=1
z
i
r
if i = i
k
(3.23)
and
g
i
=
_
f
i
if i ,= i
k
for all k,
f
i
k
f
i
k+1
if i = i
k
.
(3.24)
Then span(u
i
)
i=1
= X and (u
i
, g
i
)
i
is a biorthogonal sequence; see, [50,
Proposition 2], for example. Actually, (u
i
)
i
is also a Schauder basis in X.
To show this we can proceed either as in [28, Theorem 3.1] or as follows. Let
U
n
(x) =
n
i=1
x, g
i
)u
i
for any x X and n N. If p(k) n < p(k + 1) for
some k N, then (3.19) and (3.21), (3.22) imply that
q (U
n
(x)) = q
_
_
n
i=1,i{i
1
,i
2
,...,i
k
}
x, f
i
)z
i
+
k
r=1
x, f
i
k
f
i
k+1
)
r
s=1
z
i
s
_
_
= q
_
_
n
i=1,i{i
1
,i
2
,...,i
k
}
x, f
i
)z
i
+
k
s=1
z
i
s
k
r=s
x, f
i
k
f
i
k+1
)
_
_
= q
_
_
n
i=1,i{i
1
,i
2
,...,i
k
}
x, f
i
)z
i
+
k
s=1
x, f
i
s
)z
i
s

k
s=1
x, f
i
k+1
)z
i
s
_
_
q
_
n
i=1
x, f
i
)z
i
_
+q
_
k1
s=1
x, f
i
k+1
)z
i
s
x, f
i
k+1
)z
i
k
_
C
q
q(x) +[x, f
i
k+1
)[q
_
k
s=1
z
i
s
_
(C
q
+p
0
(f
i
k+1
)M
q
)q(x) (C
q
+ 2
C
p
0
d
M
q
)q(x) ,
for all q
X
with q p
0
and all x N. As X is a complete lcHs, this
means that (u
i
)
i
is indeed a Schauder basis of X (with dual basis (g
i
)
i
). In
particular, the sequence (u
i
k
, g
i
k
)
k
is of type P
as the sequences (u
i
k
)
k
and
(
k
h=1
g
i
h
)
k
= (f
i
1
f
i
k+1
)
k
are bounded by (3.19) and (3.22).
But, for every k N, we have that
u
i
k
, f
i
1
) =
k
r=1
z
i
r
, f
i
1
) = 1,
which implies that (u
i
)
i
is not shrinking. For, if (u
i
)
i
is shrinking, then f
i
1
=
i=1
u
i
, f
i
1
)g
i
in X
because f
i
1
X
. Hence, u
i
, f
i
1
)g
i
0 in X
and so
also u
i
k
, f
i
1
)g
i
k
0, thereby implying that 1 = sup
x(u
i
s
)
s
[x, g
i
k
)[ 0 as
(u
i
k
)
k
is bounded. So, (u
i
)
i
is not shrinking and the theorem is proved.
Theorem 3.9. Let X be a complete barrelled lcHs with a Schauder basis
(x
n
)
n
. If all the bases in X are boundedly complete, then X is reexive.
Proof. Denote by (f
n
)
n
n
)
n
. Recall that H denotes the
subspace of X
consisting of all f X
such that lim

n
n
i=1
x
i
, f)f
i
= f
in X
, endowed with the topology (H, X) induced by (X
, X). Hence,
(f
n
)
n
is a Schauder basis for H.
By assumption (x
n
)
n
is boundedly complete. So, by [29, Lemma 6.2]
applied to the space H = span(f
n
)
n=1
, with the closure taken in X
, we
can conclude that (X
, X) = (H, X) = (H, H
) and, by [29, Proposition

5.3 and Theorem 6.3], H is barrelled with H
= (H
, (H
, H)) = X alge-
braically and topologically. Moreover, (f
n
)
n
is a shrinking basis for H whose
dual basis is clearly (x
n
)
n
; see [29, Corollary 3] or [17, Theorem 1.6].
Suppose that X is not reexive. Again by a result of Retherford [44,
Theorem 2.3] (see also [12, Theorem 4] or [27, Theorem 14.5.1]), (x
n
)
n
is
not shrinking in X and so (f
n
)
n
is not a boundedly complete basis for H,
[29, Proposition 5.5]. Since H is a complete barrelled lcHs and (f
n
)
n
is
not a boundedly complete Schauder basis of H, we can proceed as in the
proof of Theorem 3.8 to establish the following fact: the space H has an-
other Schauder basis (g
i
)
i
, with dual basis (w
i
)
i
, such that (g
i
)
i
contains a
biorthogonal subsequence (g
i
k
, w
i
k
)
k
of type P and
span(w
i
)
i=1
= span(x
n
)
n=1
. (3.25)
Therefore the sequence (h
i
)
i
in H dened by
h
i
=
_
g
i
if i ,= i
k
for all k,
k
r=1
g
i
r
if i = i
k
(3.26)
is a non shrinking Schauder basis for H with dual basis (t
i
)
i
given by
t
i
=
_
w
i
if i ,= i
k
for all k,
w
i
k
w
i
k+1
if i = i
k
.
(3.27)
In particular, the biorthogonal sequence (h
i
k
, t
i
k
)
k
is of type P
. Thus, (t
i
)
i
is a Schauder basis for span(t
i
)
i=1
in H
= X.
Now, set t
0
= w
i
1
. By (3.27) and (3.25) we obtain that
span(t
i
)
i=0
= span(w
i
)
i=1
= span(x
n
)
n=1
.
If t
0
= w
i
1
span(t
i
)
i=1
, then w
i
1
=

k=1
t
i
k
with the series converg-
ing in H
= X. But, this is impossible because (t

i
k
)
k
does not (H
, H)-
converge to 0 (indeed, 1 = sup
x(h
i
s
)
s
[x, t
i
k
)[, where (h
i
s
)
s
is (H, H
)-
bounded because (h
i
k
, t
i
k
)
k
is of type P
). Thus, X = H
= span(x
n
)
n=1
=
[t
0
] span(t
i
)
i=1
which implies that the extension (t
i
)
i=0
of (t
i
)
i=1
by the
element t
0
is also a Schauder basis of X. Hence, (t
i
)
i=0
must be bound-
edly complete. However, (
k
h=1
t
i
h
)
k
is bounded and does not converge in
X = H
; contradiction! The proof is thereby complete.

Finally, the validity of the converse of both Theorem 3.8 and Theorem 3.9
has already been noted earlier (see e.g. [27, Theorem 14.5.1]). The proof of
Theorem 1.2 is thereby completely established.
4. Schauder decompositions
A decomposition of a lcHs X is a sequence (E
n
)
n
of closed, non-trivial
subspaces of X such that each x X can be expressed uniquely in the form
x =

i=1
y
i
with y
i
E
i
for each i N. This induces a sequence (Q
n
)
n
of
projections dened by Q
n
x := y
n
where x =

i=1
y
i
with y
i
E
i
for each
i N. These projections are pairwise orthogonal (i.e. Q
n
Q
m
= 0 if n ,= m)
and Q
n
(X) = E
n
for n N. If, in addition, each Q
n
L(X), for n N,
then we speak of a Schauder decomposition of X. For a basis (x
n
)
n
of X, the
projection Q
n
takes the form Q
n
x = x, f
n
) with (f
n
)
n
being the sequence of
coecient functionals associated to (x
n
); see Section 3. Then Q
n
L(X) if
and only if f
n
X
. So, for the 1-dimensional spaces E

n
:= x
n
: C,
for n N, we see that (E
n
)
n
is a Schauder decomposition of X if and only
if (x
n
)
n
is a Schauder basis for X.
Let (E
n
)
n
be a Schauder decomposition for a lcHs X. As observed above,
(E
n
)
n
induces a sequence (Q
n
)
n
L(X) of non-zero projections satisfying
Q
n
Q
m
= 0 (if n ,= m) and x =

n=1
Q
n
x for each x X. Conversely, if
(Q
n
)
n
L(X) is any sequence of projections satisfying these two conditions,
then (Q
n
(X))
n
is a Schauder decomposition of X. By setting P
n
:=
n
i=1
Q
i
,
for n N, we arrive at the following equivalent denition, preferred by some
authors in certain situations; see [9], [37], [38]. A sequence (P
n
)
n
L(X) of
projections is called a Schauder decomposition of X if it satises:
(S1) P
n
P
m
= P
min{m,n}
for all m, n N,
(S2) P
n
I in L
s
(X) as n , and
(S3) P
n
,= P
m
whenever n ,= m.
By setting Q
1
:= P
1
and Q
n
:= P
n
P
n1
for n 2 we arrive at the more
traditional formulation of a Schauder decomposition as given above. Let
(P
n
)
n
L(X) be a Schauder decomposition of X. Then the dual projections
(P
t
n
)
n
L(X
) always form a Schauder decomposition of X
, [29, p.378].
Note that necessarily (P
t
n
)
n
L(X
), [32, p.134]. If, in addition, (P

t
n
)
n
is
a Schauder decomposition for X
, then the original sequence (P

n
)
n
is called
shrinking, [29, p.379]. Since (S1) and (S3) clearly hold for (P
t
n
)
n
, this means
precisely that P
t
n
I in L
s
(X
); see (S2).
In dealing with the uniform mean ergodicity of operators the following
notion, due to J.C. Daz and M.A Miarro, [14, p.194], is rather useful. A
Schauder decomposition (P
n
)
n
in a lcHs X is said to have property (M) if
P
n
I in L
b
(X) as n . Since every non-zero projection P in a Banach
space satises |P| 1, it is clear that no Schauder decomposition in any
Banach space can have property (M). For non-normable spaces we will see
below that the situation is quite dierent.
Remark 4.1. Let (P
n
)
n
L(X) be a Schauder decomposition of X with
property (M). If X is quasi-barrelled, then it is routine to verify that P
t
n
I
in L
b
(X
). In particular, P
t
n
I in L
s
(X
). That is, (P
n
)
n
is necessarily a
shrinking Schauder decomposition of X and (P
t
n
)
n
is a Schauder decompo-
sition of X
with property (M).

We now formulate a result which provides a systematic method for pro-
ducing examples of Schauder decompositions without property (M).
Let X be a lcHs and (, ) be a measurable space. A nitely additive
set function P : L(X) is called a spectral measure in X if it satises
P() = 0 and P() = I, is multiplicative (i.e. P(A B) = P(A)P(B) for
all A, B ) and is -additive in L
s
(X), i.e. A
n
in implies that
P(A
n
) 0 in L
s
(X). If, in addition, P is also -additive in L
b
(X), then it
is called boundedly -additive. Non-trivial examples of boundedly -additive
spectral measures can only occur in non-normable spaces. For examples of
spectral measures in classical Frchet spaces, some of which are boundedly
-additive and others which are not, we refer to [6], [7], [8], [9], [46], [47].
We denote by B(X) the collection of all bounded subsets of a lcHs X.
Proposition 4.2. Let X be a Frchet space. Suppose that there exists a
spectral measure in X which fails to be boundedly -additive. Then X admits
a Schauder decomposition without property (M).
Proof. Let P : L(X) be any spectral measure which fails to be bound-
edly -additive. Let q
1
q
2
. . . be a sequence of seminorms generating
the topology of X and having the property that
q
k
(P(A)x) q
k
(x), x X, A , (4.1)
for each k N; see Proposition 2.3 in [52] and the discussion following its
proof. Since P fails to be boundedly -additive, there exists a sequence
(A
n
)
n
with A
n
such that P(A
n
) , I in L
b
(X). It follows that
there exists a sequence of positive integers n(k) such that P(A
n(k)
) ,=
P(A
n(k+1)
) for k N. Set P
k
:= P(A
n(k)
) for k N, in which case it
is routine to check (using the -additivity of P in L
s
(X)) that (P
k
)
k
is a
Schauder decomposition of X.
The proof is completed by showing that P
k
, I in L
b
(X). On the con-
trary, suppose that P
k
I in L
b
(X) as k . According to (1.4), the
topology of L
b
(X) is determined by the seminorms
q
B,k
(T) := sup
xB
q
k
(Tx), T L(X),
for all k N and B B(X). So, x such a k and B. Given > 0, there
exists M N such that
sup
xB
q
k
((I P(A
n(m)
)x) = q
B,k
(I P
m
) , m N. (4.2)
Let n n(M). Then there exists m M satisfying n(m) n < n(m + 1).
Hence, by property (S1) of a Schauder decomposition we have I P(A
n
) =
(I P(A
n(m)
))(I P(A
n
)). It follows from this identity, (4.1) and (4.2) that
sup
xB
q
k
((I P(A
n
))x) , that is,
q
B,k
((I P(A
n
))) , n n(M).
This shows that P(A
n
) I in L
b
(X); contradiction to the choice of (P(A
n
))
n
! Hence, P
k
, I in L
b
(X).
Remark 4.3. Explicit examples of Frchet spaces which admit spectral
measures which fail to be boundedly -additive are surely known. For the
spaces L
p
oc
(R), 1 p < , we refer to [6, Proposition 6.2(ii)] and for
p+
,
1 p < , see [7, Corollary 3.2(i)]. For those Kthe echelon spaces
p
(A)
with p 0 [1, ) which are not Montel see [7, Corollary 3.2(ii)]. The
same conclusion holds for Kthe function spaces over non-atomic measure
spaces, [8, Proposition 2.14].
We conclude with two technical results needed later. The rst result is an
extension of a lemma in [23, p.149] to the Frchet space setting.
Lemma 4.4. Let X be a Frchet space which admits a non-shrinking Schauder
decomposition. Then there exist a Schauder decomposition (P
j
)
j
L(X)
of X, a functional X
and a bounded sequence (z

j
)
j
X with z
j

(P
j+1
P
j
)(X) such that [z
j
, )[ >
1
2
for all j N.
Proof. Let (R
n
)
n
L(X) be a non-shrinking Schauder decomposition of
X. It follows that there exist a set B B(X), a functional X
and a
sequence of positive integers n(j) such that
sup
xB
[x, (I R
t
n(j)
))[ > 1, j N. (4.3)
Select q
X
such that
[x, )[ q(x), x X. (4.4)
By (4.3), there exists x
1
B with [(I R
n(1)
)x
1
, )[ > 1. Set m(1) :=
n(1). Since R
n
x
1
x
1
in X as n , we can select m(2) (n(k))
k
with m(2) > n(1) such that q((I R
m(2)
)x
1
) <
1
2
. Moreover, (4.4) then
implies that [(I R
m(2)
)x
1
, )[ <
1
2
. Dene z
1
:= (R
m(2)
R
m(1)
)x
1
and
set P
1
:= R
m(1)
and P
2
:= R
m(2)
. Observe that
[z
1
, )[ = [(I R
m(1)
)x
1
(I R
m(2)
)x
1
, )[
[(I R
m(1)
)x
1
, )[ [(I R
m(2)
)x
1
, )[ >
_
1
1
2
_
=
1
2
.
According to (4.3), there exists x
2
B with [(I R
m(2)
)x
2
, )[ > 1. Select
m(3) (n(k))
k
with m(3) > m(2) such that q((I R
m(3)
)x
2
) <
1
2
. Again
(4.4) implies that [(I R
m(3)
)x
2
, )[ <
1
2
. Dene z
2
:= (R
m(3)
R
m(2)
)x
2
and set P
3
:= R
m(3)
. As above,
[z
2
, )[ = [(I R
m(2)
)x
2
(I R
m(3)
)x
2
, )[ >
1
2
.
Proceeding in this way we get m(j) and a sequence (x
j
)
j
B such
that [z
j
, )[
1
2
for all j N, where z
j
:= (R
m(j+1)
R
m(j)
)x
j
. Setting
P
j
:= R
m(j)
, for j N, we see that (P
j
)
j
is a Schauder decomposition of X
and that z
j
(P
j+1
P
j
)(X) for all j N. Since B B(X) and (P
j+1
P
j
)
j
is equicontinuous in L(X), it follows that D :=

jN
(P
j+1
P
j
)(B) is a
bounded set in X. But, (z
j
)
j
D and hence, (z
j
)
j
is bounded in X.
Lemma 4.5. Let X be a Frchet space which admits a Schauder decom-
position without property (M). Then there exist a Schauder decomposition
(P
j
)
j
of X, a seminorm q
X
j
)
j
X with
z
j
(P
j+1
P
j
)(X) such that q(z
j
) >
1
2
for all j N.
Proof. Let (R
n
)
n
L(X) be a Schauder decomposition of X which does not
have property (M). Hence, there exist a set B B(X), a seminorm q
X
and positive integers n(j) such that
sup
xB
q((I R
n(j)
)x) > 1, j N. (4.5)
Select x
1
B with q((I R
n(1)
)x
1
) > 1. Set m(1) := n(1). Since R
n
x
1
x
1
in X as n , there exists m(2) (n(k))
k
such that q((I R
m(2)
)x
1
) <
1
2
.
Set z
1
:= (R
m(2)
R
m(1)
)x
1
and dene P
1
:= R
m(1)
and P
2
:= R
m(2)
. As in
the proof of Lemma 4.4 we can conclude that
q(z
1
) = q
_
(I R
m(1)
)x
1
(I R
m(2)
)x
1
_
>
1
2
.
According to (4.5), there exists x
2
B with q((I R
m(2)
)x
2
) > 1. Select
m(3) (n(k))
k
with m(3) > m(2) such that q((I R
m(3)
)x
2
) <
1
2
. Dene
z
2
:= (R
m(3)
R
m(2)
)x
2
and P
3
:= R
m(3)
and note that q(z
2
) >
1
2
. Proceed
in this way to get m(j) and a sequence (x
j
)
j
B such that z
j
:=
(R
m(j+1)
R
m(j)
)x
j
satises q(z
j
) >
1
2
, for each j N. Put P
j
:= R
m(j)
, for
j N, and then complete the argument as in the proof of Lemma 4.4.
5. Mean Ergodic Operators
Using the results of previous sections we can now establish Theorems 1.3
1.6. Various examples and consequences of these results are also given.
We begin with a useful observation. Let (P
n
)
n
L(X) be any Schauder
decomposition in the Frchet space X. As noted in [14, p.192], there exists
an increasing sequence of seminorms (q
k
)
k
determining the topology of X
such that
q
k
(P
j
x) q
k
(x), x X, j N, (5.1)
for each k N. Indeed, if (p
m
)
m
is any increasing sequence of seminorms
determining the topology of X, then the seminorms q
k
(x) := sup
jN
p
k
(P
j
x),
for x X and k N, have the desired property (after noting that X is
barrelled and hence, (P
n
)
n
L(X) is equicontinuous).
Proof of Theorem 1.5 Let (P
j
)
j
L(X) denote a Schauder decompo-
sition as given by Lemma 4.4 and dene projections Q
j
:= P
j
P
j1
(with
P
0
:= 0) and closed subspaces X
j
:= Q
j
(X) of X, for j N. By Lemma 4.4
there also exist a bounded sequence (z
j
)
j
X with z
j
X
j+1
and X
such that [z
j
, )[ >
1
2
, for all j N. Set e
j
:= z
j
/z
j
, ) Q
j+1
(X), in
which case (e
j
)
j
is a bounded sequence in X with e
j
, ) = 1, for all j N.
Let (q
k
)
k
be any increasing sequence of seminorms giving the topology of X,
satisfying (5.1) and such that [x, )[ q
1
(x) for all x X.
As in [23, p.150], take an arbitrary sequence of positive numbers a = (a
j
)
j
with
j=1
a
j
= 1 and set A
n
:=
n
j=1
a
j
, for n N. Let x X. For integers
m > n 2 we have
m
k=n
A
k
Q
k
x =
_
n1
j=1
a
j
__
m
k=n
Q
k
x
_
+
m
j=n
a
j
_
m
k=j
Q
k
x
_
.
Since

k=1
Q
k
x = x, we see that (
m
k=1
A
k
Q
k
x)
m
is Cauchy and hence,
convergent in X. Moreover, for each s N, we have (c.f.(5.1))
q
s
_
m
k=1
A
k
Q
k
x
_
= q
s
_
m
j=1
a
j
(P
m
P
j1
)x
_
m
j=1
a
j
(q
s
(P
m
x) +q
s
(P
j1
x)) 2q
s
(x),
for all m N. Dene a linear map T
a
: X X by
T
a
x :=
k=1
A
k
Q
k
x +
j=2
P
j1
x, )a
j
e
j
, x X. (5.2)
Using the previous inequalities we see that
q
s
(T
a
x) 2q
s
(x) +
j=2
a
j
q
1
(P
j1
x)q
s
(e
j
), x X.
Setting M
s
:= sup
j
q
s
(e
j
) < (recall that (e
j
)
j
is bounded in X) and noting
that q
1
(P
j1
x) q
1
(x) q
s
(x), it follows that
q
s
(T
a
x) (2 +M
s
)q
s
(x), x X, s N,
with M
s
independent of a. In particular, T
a
L(X).
To show that T
a
is power bounded it suces to show, for arbitrary se-
quences a = (a
j
)
j
and b = (b
j
)
j
of positive numbers with

j=1
a
j
= 1 =
j=1
b
j
, that the composition T
a
T
b
is also of the same type (say, T
c
for an
appropriate c = (c
j
)
j
). But, this is precisely the Claim on p.150 in [23]
which is proved there (on p.151) by purely algebraic computations and
hence, carries over to our setting here. So, T
a
is indeed power bounded.
To deduce that T
a
is not mean ergodic we apply Theorem 2.12 by verifying
that ker(I T
a
) = 0 and that X
belongs to ker(I T
t
a
).
Let x ker(I T
a
). It follows from (5.2) and x =
k=1
Q
k
x that
k=1
Q
k
x = x =
k=1
A
k
Q
k
x +
k=2
P
k1
x, )a
k
e
k
. (5.3)
Moreover, (5.2) together with the identities
Q
k
Q
j
= 0, k ,= j, (5.4)
which imply that Q
1
e
k
= 0 for k 2, yield
Q
1
x = Q
1
T
a
x = A
1
Q
1
x.
Since 0 < A
1
= a
1
< 1, we see that Q
1
x = 0. For k > 1, apply Q
k
to (5.3)
to conclude that
Q
k
x = A
k
Q
k
x +P
k1
x, )a
k
e
k
. (5.5)
Now, P
1
x = Q
1
x +P
0
x = 0 and so, by substituting k = 2 into (5.5), we see
that Q
2
x = A
2
Q
2
x with 0 < A
2
< 1. Hence, Q
2
x = 0. Proceed inductively
(via (5.5) and the formula P
k1
x =

k1
j=1
Q
j
x) to conclude that Q
k
x = 0
for all k N. Then x =
k=1
Q
k
x = 0, that is, ker(I T
a
) = 0.
It remains to verify that ker(I T
t
a
). To this end, x k N and
y X
k
. By (5.2), (5.4), the equalities e
j
, ) = 1 for all j N, and the
identities
P
j1
y = P
j1
Q
k
y =
_
0 for 1 j k
y for j > k,
we have that
y, T
t
a
) = T
a
y, ) = A
k
y +
j=k+1
y, )a
j
e
j
, )
= y, )
_
A
k
+
j=k+1
a
j
_
= y, ).
Hence, (I T
a
)y, ) = 0 for all k N and y X
k
. In view of the decompo-
sition X =
k=1
X
k
we conclude that y, (I T
t
a
)) = 0 for all y X, that
is, ker(I T
t
a
).
Proposition 5.1. Let X be a Frchet space. Then X is reexive if and only
if every closed subspace of X is mean ergodic.
Proof. Suppose that X is not reexive. By Theorem 1.1 there exists a non-
reexive, closed subspace Y with a basis. According to Theorem 1.2, the
space Y must have some non-shrinking Schauder basis. In particular, Y
admits a non-shrinking Schauder decomposition and hence, Theorem 1.5
implies that Y is not mean ergodic.
Proof of Theorem 1.4 Let X be a Frchet space with a basis. If X
is non-reexive, then Theorem 1.2 shows that X admits a non-shrinking
Schauder basis (and hence, a non-shrinking Schauder decomposition). By
Theorem 1.5 we can conclude that X is not mean ergodic.
Theorem 5.2. Let X be a Frchet space which admits a Schauder decom-
position without property (M). Then there exists a power bounded, mean
ergodic operator in L(X) which fails to be uniformly mean ergodic.
Proof. Let (P
j
)
j
L(X) denote a Schauder decomposition of X as given
by Lemma 4.5 and dene projections Q
j
:= P
j
P
j1
(with P
0
= 0) and
closed subspaces X
j
:= Q
j
(X) of X, for j N. By Lemma 4.5 there exist a
seminorm q
X
j
)
j
X with z
j
X
j+1
such
that q(z
j
) >
1
2
, for all j N. Choose a sequence of increasing seminorms
(q
k
)
k
generating the topology of X which satisfy (5.1) and such that q q
1
.
Take any sequence of positive numbers a = (a
j
)
j
with

j=1
a
j
= 1 and
set A
n
:=
n
j=1
a
j
for n N. Dene a linear map T
a
: X X by
T
a
x :=
k=1
A
k
Q
k
x, x X; (5.6)
observe that this corresponds to (5.2) for the case when = 0. As in the
proof of Theorem 1.5 given above the operator T
a
is well dened, satises
q
s
(T
a
x) 2q
s
(x), x X, s N, (5.7)
and is power bounded.
Given x ker(I T
a
) we have x = T
a
x and so, from (5.6), it follows
that x =
k=1
A
k
Q
k
x. Using (5.4) it follows that Q
j
x = A
j
Q
j
x and hence,
since 0 < A
j
< 1, that Q
j
x = 0 for all j N. That is, x = 0 and so
ker(I T
a
) = 0. To conclude that T
a
is mean ergodic it suces, by
Theorem 2.12, to show that ker(I T
t
a
) = 0. So, suppose that u X
satises T
t
a
u = u. Let x
j
X
j
be arbitrary and note that x
j
, u) = T
a
x
j
, u)
for all j N. It follows form (5.6) and (5.7) that
x
j
, u) =
k=1
A
k
Q
k
x
j
, u) =
k=1
A
k
Q
k
Q
j
x
j
, u) = A
j
x
j
, u)
for j N; see also (5.4). Since 0 < A
j
< 1, we conclude that x
j
, u) = 0.
So, we have shown that y, u) = 0 whenever y X
j
for some j N. Since
j=1
Q
j
= I (in L
s
(X)), it follows that each x X has a decomposition
x =

j=1
Q
j
x with Q
j
x X
j
for all j N. By continuity of u it follows
that u = 0. So, ker(I T
t
a
) = 0 and hence, T
a
is mean ergodic.
The proof is completed by showing, for the choice a
j
:= 2
j
for j N,
that T
a
is not uniformly mean ergodic. For ease of notation, set T := T
a
.
Note that A
k
= 1 2
k
for each k N. Moreover, from (5.6), (5.7) and
j=1
Q
j
= I (in L
s
(X)), it follows that
T
m
x =
k=1
A
m
k
Q
k
x, x X, m N.
Then (1.5) and direct calculation yields
T
[n]
x =
1
n
k=1
A
k
(1 A
k
)
(1 A
n
k
)Q
k
x, x X, n N. (5.8)
Since T is mean ergodic, there exists P L(X) with T
[n]
P in L
s
(X) as
n . For j N xed and x X
j
, it follows from (5.4) that Q
k
x =
kj
x
for all k N hand hence, (5.8) implies that
T
[n]
x =
1
n
A
j
(1 A
j
)
(1 A
n
j
)x, n N. (5.9)
Since 0 < (1 A
n
j
) < 1, for each n N, it follows that
q
k
(T
[n]
x)
1
n
A
j
(1 A
j
)
q
k
(x), k, n N.
Accordingly, q
k
(T
[n]
x) 0 as n (for each k N). But, T
[n]
P in
L
s
(X) as n and so Px = 0. That is, Py = 0 for every y

j=1
X
j
.
We have already seen that

j=1
X
j
is dense in X and so P = 0, that is,
T
[n]
0 in L
s
(X) as n .
Suppose that T is uniformly mean ergodic, in which case necessarily T
[n]

0 in L
b
(X). In particular, since (z
j
)
j
is a bounded sequence in X, we have
lim
n
sup
jN
q(T
[n]
z
j
) = 0. (5.10)
Fix j N. It follows from (5.9) with x = z
j
X
j
and n = 2
j
(together with
A
j
= 1 2
j
) that
q(T
2
j z
j
) = (1 2
j
) (1 (1 2
j
)
2
j
)q(z
j
).
Using the inequalities q(z
j
) >
1
2
and (1 2
j
)
1
2
we conclude that
q(T
2
j z
j
) >
1
4
(1 (1 2
j
)
2
j
).
But, lim
j
(1 2
j
)
2
j
= e
1
and we have a contradiction to (5.10). Ac-
cordingly, T = T
a
is not uniformly mean ergodic.
Remark 5.3. (i) As noted in Section 4, every Schauder decomposition in a
Banach space fails to have property (M). So, for Banach spaces Theorem 5.2
reduces to [23, Theorem 2]. However, the proof of Theorem 2 given in [23] is
based on Lins criterion, namely Proposition 2.16 above which, as noted in
Section 2, fails to hold in non-normable spaces (in general). So, the Banach
space proof of [23] does not apply in Frchet spaces.
(ii) For each p [1, ), let L
p
oc
(R) denote the space of all (equivalence
classes of) Lebesgue measurable functions f dened on R which satisfy
q
(n)
p
(f) :=
_
_
n
n
[f(t)[
p
dt
_
1/p
< ,
for all n N. Each space L
p
oc
(R) is a separable Frchet space (reexive if
p ,= 1) when equipped with the seminorms q
(1)
p
q
(2)
p
. . .. This class of
spaces has been intensively studied in [1], [2], [3]. According to Remark 4.3
(see [6, Proposition 6.2(ii)]), there exists a spectral measure in L
p
oc
(R) which
fails to be boundedly -additive. Hence, Proposition 4.2 implies that L
p
oc
(R)
admits a Schauder decomposition without property (M). Then Theorem 5.2
shows that L
p
oc
(R) is not uniformly mean ergodic. For each 1 < p < , the
space L
p
oc
(R) is mean ergodic because of its reexivity (c.f. Corollary 2.7).
The case p = 1 is dierent. Since the Banach space L
1
([0, 1]) is isomorphic
to a complemented subspace of L
1
oc
(R), to show that L
1
oc
(R) is not mean
ergodic it suces to show (by an argument as in the Proof of Proposition 2.9)
that L
1
([0, 1]) is not mean ergodic. But, L
1
([0, 1]) has a Schauder basis [36,
p.3] and is non-reexive. Then [23, Corollary 1] implies that L
1
([0, 1]) is,
indeed, not mean ergodic. Or, one can appeal to [20, Theorem 2].
The following result has no counterpart in Banach spaces.
Theorem 5.4. For a Frchet space X the following assertions are equivalent.
(i) X is a Montel space.
(ii) Every closed subspace of X is uniformly mean ergodic.
(iii) Every power bounded, mean ergodic operator dened on a closed sub-
space of X is uniformly mean ergodic.
Proof. (i) (ii). This follows from Proposition 2.8 and (ii) (iii) is obvious.
Suppose that X is not Montel. According to Proposition 3.2, X contains a
closed subspace Y which is not Montel and has a basis. Then the Schauder
decomposition (P
n
)
n
L(Y ) induced by this basis has the property that
each (1-dimensional) space Q
n
(Y ) := (P
n
P
n1
)(Y ), for n N, is Montel
(with P
0
:= 0). By [14, Proposition 4], the Schauder decomposition (P
n
)
n
cannot have property (M) and hence, Theorem 5.2 guarantees the existence
of a power bounded, mean ergodic operator in L(Y ) which fails to be uni-
formly mean ergodic. This establishes (iii) (i).
Proof of Theorem 1.3 Let X be a Frchet space with a basis. If X is
Montel, then it is uniformly mean ergodic by Proposition 2.8. On the other
hand, if X is not Montel, then we can choose Y = X in the proof of (iii)
(i) in Theorem 5.4 to conclude that X is not uniformly mean ergodic.
Proof of Theorem 1.6 Initially we proceed along the lines of [23, Corol-
lary 4]. Let X be a Frchet space which contains an isomorphic copy of c
0
,
say via an isomorphism J : c
0
X. Let (e
n
)
n
be the standard unit basis
vectors of c
0
, in which case the sequence (y
n
)
n
with y
n
:= Je
n
, for n N,
is a Schauder basis of Y := J(c
0
). Let | |
c
0
denote the norm in c
0
and
X
= (q
k
)
k
be increasing. Then, for each k N, there exists M
k
> 0 satis-
fying q
k
(Jx) M
k
|x|
c
0
, for x c
0
, and there exist k
1
N and K > 0 such
that |x|
c
0
Kq
k
1
(Jx), for all x c
0
. By omitting q
j
, for 1 j < k
1
, and
relabelling (if necessary), we may assume that there exists M
0
> 0 with
|x|
c
0
M
0
q
1
(Jx), x c
0
, (5.11)
and for each k N an M
k
> 0 satisfying
q
k
(Jx) M
k
|x|
c
0
, x c
0
. (5.12)
Moreover, (q
k
)
k
is still increasing and determines the topology of X. In
particular, for each u = (u
j
) =
j=1
u
j
e
j
in c
0
we have
sup
jN
[u
j
[ M
0
q
1
_

j=1
u
j
y
j
_
= M
0
q
1
_

j=1
u
j
Je
j
_
and
q
k
_

j=1
u
j
y
j
_
M
k
sup
jN
[u
j
[. (5.13)
Let (f
n
)
1
= c
0
denote the dual basis of (e
n
)
n
c
0
. For each n N,
dene y
n
Y
by y
n
:= f
n
J
1
, in which case (y
n
)
n
is the dual basis of the
Schauder basis (y
n
)
n
of Y . Since [y, y
n
)[ M
0
q
1
(y), for y Y and n N,
we see that (y
n
)
n
is an equicontinuous subset of Y
. By the Hahn-Banach
theorem, for each n N there exists
n
X
satisfying
n
[
Y
= y
n
and
[x,
n
)[ M
0
q
1
(x), x X. (5.14)
Dene x
n
:=

n
j=1
y
j
Y and g
n
:= (
n

n+1
) X
for each n N.
Direct calculation yields
y
n
,
k
) =
kn
and x
n
, g
k
) =
kn
, k, n N.
Dene projections P
n
: X X, for each n N, by
P
n
x :=
n
k=1
x, g
k
)x
k
, x X,
with the range of P
n
equal to span(x
j
)
n
j=1
= span(y
j
)
n
j=1
Y . Clearly
P
n
P
m
= P
min{m,n}
. Set h :=
1
X
, in which case
x
n
, h) =
n
j=1
y
j
,
1
) = 1, n N.
Since (P
n
P
n1
)x
n
= x
n
(with P
0
:= 0), we have x
n
(P
n
P
n1
)(X)
for all n N. Moreover, (x
n
)
n
X is a bounded sequence, because x
n
=
J(
n
j=1
e
j
) with |
n
j=1
e
j
|
c
0
= 1 for all n N implies (via (5.12)) that
q
k
(x
n
) = q
k
_
J(
n
j=1
e
j
)
_
M
k
, k, n N.
Moreover, with
0
:= M
1
0
we see from (5.11) that
0
q
1
(x
n
), n N.
On the other hand, the identities
P
n
x = x,
1
)y
1
x,
n+1
)x
n
+
n
k=2
x,
k
)(x
k
x
k1
)
=
n
k=1
(x,
k
) x,
n+1
)) y
k
,
valid for all n N and x X, imply (via (5.13) and (5.14)) that
q
k
(P
n
x) M
k
sup
1kn
[x,
k
) x,
n+1
)[
2M
k
M
0
q
1
(x), x X; k, n N. (5.15)
Accordingly, (P
n
)
n
L(X) is equicontinuous.
Let a = (a
j
)
j
be any sequence of positive numbers satisfying
j=1
a
j
= 1
and set A
n
:=
n
j=1
a
j
, for n N. As in the statement of Theorem 3 in [23]
dene
S
a
x := x
n=2
a
n
P
n1
x +
n=2
a
n
P
n1
x, h)x
n
, (5.16)
for each x X. To verify that S
a
L(X), x k N. From the denition of
S
a
and the inequalities (5.13) and (5.15) we can conclude that
q
k
(S
a
x) q
k
(x) + 2M
k
M
0
q
1
(x) +M
k
sup
n2
[P
n1
x, h)[,
for each x X. But, (5.14) and (5.15) yield
[P
n1
x, h)[ = [P
n1
x,
1
)[ M
0
q
1
(P
n1
x) 2M
2
0
M
1
q
1
(x)
and hence,
q
k
(S
a
x)
_
1 + 2M
k
M
0
+ 2M
k
M
2
0
M
1
_
q
k
(x), x X,
with the right-hand side independent of a. So, S
a
L(X).
The fact that S
a
is power bounded follows from the Claim on p.156 of [23],
stating that S
a
S
b
= S
c
for an appropriate c (expressed in terms of apriori
given a and b). The argument in [23] is of a pure algebraic computational
nature and so, also applies here.
Next we verify that S
a
is not mean ergodic. To this end, rst dene
Y
n
:= P
n
(X) in which case the property P
n
P
m
= P
min{m,n}
implies that
Y
n
Y
m
whenever n m. Let y Y . Since (
n
j=1
e
j
)
n
is a Schauder basis
of c
0
, there exist scalars (
n
)
n
such that J
1
y =
n=1
n
(
n
j=1
e
j
), that is,
y =
j=1
j
x
j
. But,

m
j=1
j
x
j
P
m
(X) = Y
m
for all m N and so
Y =
m=1
Y
m
, (5.17)
with the closure formed in X.
To see that S
a
(Y ) Y it suces to verify that S
a
(Y
m
) Y for each
m N. According to (5.16), for a given x X we have
S
a
(P
m
x) = P
m
x
n=2
a
n
P
n1
P
m
x +
n=2
a
n
P
n1
P
m
x, h)x
n
.
Because of the property P
r
P
s
= P
min{r,s}
we have
n=2
a
n
P
n1
P
m
x =
m
n=2
a
n
P
n1
x +
n>m
a
n
P
m
x.
For the same reason we also have
n=2
a
n
P
n1
P
m
x, h)x
n
=
m
n=2
a
n
P
n1
x, h)x
n
+P
m
x, h)
n>m
a
n
x
n
.
Accordingly,
S
a
(P
m
x) = P
m
x +
m
n=2
a
n
P
n1
x +
_

n>m
a
n
_
P
m
x +
+
m
n=2
a
n
P
n1
x, h)x
n
+P
m
x, h)
n>m
a
n
x
n
.
It then follows from (5.17) that S
a
(P
m
x) Y .
Now, lim
n
P
n
y = y for all y Y
m
because y = P
m
y and so
lim
n
P
n
P
m
y = lim
nm
P
n
P
m
y = P
m
y.
Since (P
n
)
n
L(X) is equicontinuous, it follows from (5.17) that
y = lim
n
P
n
y, y Y.
Dene Q
1
:= P
1
and Q
k
:= (P
k
P
k1
) for k 2. Arguing as in the proof
of Theorem 3 in [23], it turns out that each Q
k
is a projection with Q
k
Q
j
= 0
whenever k ,= j and P
k
=

k
j=1
Q
j
. Hence, E
k
:= Q
k
(X) = Q
k
(Y ), for
k N, is a Schauder decomposition of Y .
Finally, with T
a
L(X) as dened by (5.2), it turns out (see the proof
of Theorem 3 in [23] where the formulae (4) and (7) used there are also
available here) that S
a
[
Y
= T
a
. Observe that x
k
belongs to E
k
and satises
x
k
, h) = 1 for all k N. Moreover, (x
k
)
k
is a bounded sequence in Y . By
the proof of Theorem 1.5 given above (applied to S
a
in Y ) there exists y Y
such that ((S
a
)
[n]
y)
n
does not converge in Y . In particular, T
a
L(X) is
not mean ergodic.
References
[1] Albanese, A.A., Primary products of Banach spaces, Arch. Math. (Basel), 66 (1996),
397-405.
[2] Albanese, A.A., and Moscatelli, V.B., Complemented subspaces of sums and products
of copies of L
1
([0, 1]), Rev. Mat. Univ. Complut. Madrid 9 (1996), 275-287.
[3] Albanese, A.A., On subspaces of the spaces L
p
oc
and of their strong duals, Math.
Nachr. 197 (1999), 5-18.
[4] Altman, M., Mean ergodic theorem in locally convex linear topological spaces, Studia
Math. 13 (1953), 190-193.
[5] Bellenot, S.F., Basic sequences in non-Schwartz-Frchet spaces, Trans. Amer. Math.
Soc. 258 (1980), 199-216.
[6] Bonet J. and Ricker, W.J., Spectral measures in classes of Frchet spaces, Bull. Soc.
Roy. Sci. Lige, 73 (2004), 99-117.
[7] Bonet, J. and Ricker, W.J., The canonical spectral measure in Kthe echelon spaces,
Integral Equations Operator Theory, 53 (2005), 477-496.
[8] Bonet, J., Okada, S. and Ricker, W.J., The canonical spectral measure and Kthe
function spaces, Quaestiones Math. 29 (2006), 91-116.
[9] Bonet, J. and Ricker. W.J., Schauder decompositions and the Grothendieck and
Dunford-Pettis properties in Kthe echelon spaces of innite order, Positivity 11
(2007), 77-93.
[10] Cascales B. and Orihuela, J., On compactness in locally convex spaces, Math. Z. 195
(1987), 365-381.
[11] Castillo, J.M.F., Daz, J.C. and Motos, J., On the Frchet space L
p
, Manuscripta
Math. 96 (1998), 219-230.
[12] Cook, T.A., Schauder decompositions and semi-reexive spaces, Math. Ann. 182
(1969), 232-235.
[13] Daz, J.C., Montel subspaces in the countable projective limits of L
p
()-spaces,
Canad. Math. Bull. 32 (1989), 169-176.
[14] Daz, J.C. and Miarro, M.M., Distinguished Frchet spaces and projective tensor
product, Doga-Tr. J. Math. 14 (1990), 191-208.
[15] Diestel, J., Sequences and Series in Banach Spaces, Springer Verlag, New York, 1984.
[16] Diestel J., Jarchow H. and Tonge A., Absolutely Summing Operators, Cambridge
Univ. Press, Cambridge (U.K.), 1995.
[17] Dubinsky, E. and Retherford J.R. , Schauder basis and Kthe sequence spaces, Trans.
Amer. Math. Soc. 130 (1968), 265-280.
[18] Dunford, N., Spectral theory I, Convergence to projections, Trans. Amer. Math. Soc.
54 (1943), 185-217.
[19] Dunford, N. and Schwartz, J.T., Linear Operators I: General Theory, 2nd Ed., Wiley-
Interscience, New York, 1964.
[20] Emelyanov, E., Banach lattices on which every power-bounded operator is mean
ergodic, Positivity 1 (1997), 291-296.
[21] Emelyanov, E. and Wol M.P.H., Mean ergodicity on Banach lattices and Banach
spaces, Arch. Math. (Basel), 72 (1999), 214-218.
[22] Emilion, R., Mean-bounded operators and mean ergodic theorems, J. Funct. Anal.
61 (1985), 1-14.
[23] Fonf, V.P., Lin, M. and Wojtaszczyk, P., Ergodic characterizations of reexivity in
Banach spaces, J. Funct. Anal. 187 (2001), 146-162.
[24] Hille, E., Remarks on ergodic theorems, Trans. Amer. Math. Soc., 57 (1945), 246-269.
[25] Hille, E. and Phillips, R.S., Functional Analysis and Semigroups, 4th printing of
Revised Ed., Amer. Math. Soc., Providence, 1981.
[26] James, R.C. , Bases and reexivity of Banach spaces, Ann. of Math. 52 (1950),
518527.
[27] Jarchow, H., Locally Convex Spaces, Teubner, Stuttgart, 1981.
[28] Kalton, N.J., Schauder bases and reexivity, Studia Math. 38 (1970), 255-266.
[29] Kalton, N.J., Schauder decompositions in locally convex spaces, Proc. Camb. Phil.
Soc. 68 (1970), 377-392.
[30] Kalton, N.J., Normalization properties of Schauder bases, Proc. London Math.. Soc.
22 (1971), 91-105.
[31] Kthe, G., Topological Vector Spaces I, 2nd Rev. Ed., Springer Verlag, Berlin-
Heidelberg-New York, 1983.
[32] Kthe, G., Topological Vector Spaces II, Springer Verlag, Berlin-Heidelberg-New York,
1979.
[33] Krengel, U., Ergodic Theorems, de Gruyter, Berlin, 1985.
[34] Lewis, D.R., Integration with respect to a vector measure, Pacic J. Math. 33 (1970),
157-165.
[35] Lin M., On the uniform ergodic theorem, Proc. Amer. Math. Soc. 43 (1974), 337-340.
[36] Lindenstrauss, J. and Tzafriri, L., Classical Banach Spaces I, Springer Verlag, Berlin-
Heidelberg-New York, 1977.
[37] Lotz, H.P., Tauberian theorems for operators on L
and similar spaces, in Functional

Analysis: Surveys and Recent Results III , K.-D. Bierstedt and B. Fuchssteiner (Eds.),
North Holland, Amsterdam (1984), pp. 117-133.
[38] Lotz, H.P., Uniform convergence of operators on L
and similar spaces, Math. Z.

190 (1985), 207-220.
[39] Marti, J.T., Introduction to the Theory of Bases, Springer Verlag, Berlin-Heidelberg-
New York, 1969.
[40] Meise, R.G. and Vogt, D., Introduction to Functional Analysis, Clarendon Press,
Oxford, 1997.
[41] Metafune, G. and Moscatelli, V.B., On the space
p+
=
q>p
q
, Math. Nachr. 147
(1990), 7-12.
[42] Pietsch, A., Quasi-prkompakte Endomorphismen und ein Ergodensatz in lokalkon-
vexen Vektorrumen, J. Reine Angew. Math. 207 (1961), 16-30.
[43] Radjavi, H., Tam, P.-K. and Tan, K.-K., Mean ergodicity for compact opertors, Studia
Math. 158 (2003), 207-217.
[44] Retherford, J.R., Bases, basic sequences and reexivity of linear topological spaces,
Math. Ann. 164 (1966), 280-285.
[45] Retherford, J.R. and McArthur, C.W., Some remarks on bases in linear topological
spaces, Math. Ann. 164 (1966), 38-41.
[46] Ricker, W.J., Spectral measures, boundedly -complete Boolean algebras and appli-
cations to operator theory, Trans. Amer. Math. Soc. 304 (1987), 819-838.
[47] Ricker, W.J., Operator algebras generated by Boolean algebras of projections in Mon-
tel spaces, Integral Equations Operator Theory, 12 (1989), 143-145.
[48] Robinson, W.B., Extensions of basic sequences in Frchet spaces, Studia Math. 45
(1973), 1-14.
[49] Sine, R.C., A mean ergodic theorem, Proc. Amer. Math. Soc. 24 (1970), 438-439.
[50] Singer I., Basic sequences and reexivity of Banach spaces, Studia Math. 21 (1962),
351-369.
[51] Sucheston, L., in Probability in Banach Spaces, Lecture Notes in Mathematics, Vol.
526, pp. 285-290, Springer Verlag, Berlin, New York, 1976.
[52] Walsh B., Structure of spectral measures on locally convex spaces, Trans. Amer.
Math. Soc. 120 (1965), 295-326.
[53] Yahdi, M., Super-ergodic operators, Proc. Amer. Math. Soc. 134 (2006), 2613-2620.
[54] Yosida, K., Functional Analysis, Springer Verlag, Berlin, Heidelberg, 1965.
[55] Zippin, M., A remark on bases and reexivity in Banach spaces, Israel J. Math. 6
(1968), 74-79.
A.A. Albanese, Dipartimento di Matematica, Universit - C.P.193,
I-73100 Lecce, Italy
E-mail address: angela.albanese@unile.it
J. Bonet, ETS Arquitectura, Departamento de Matemtica Aplicada, Uni-
versidad Politcnia de Valencia, E-46071 Valencia, Spain
E-mail address: jbonet@mat.upv.es
W.J. Ricker, Math.-Geogr. Fakultt, Katholische Universitt Eichsttt-
Ingolstadt, D-85072 Eichsttt, Germany
E-mail address: werner.ricker@ku-eichstaett.de

Mean Ergodic Operators in Frechet Spaces

Transféré par

Informations du document

Description originale:

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

Mean Ergodic Operators in Frechet Spaces

Transféré par

Droits d'auteur :

Formats disponibles

MEAN ERGODIC OPERATORS IN FRCHET SPACES

) for all x X and x

. Let us denote X equipped with its weak

. It follows from (2.1) and (2.7) that

which, of course, coincides

)-compact. If T L(X) satises (2.7) and

= 1. Then for n large enough,

and x X satisfying u, x) = 1. It follows from

0 whose restriction to Im(I T) is zero, that is,

is a non-zero element of ker(I T

(A) is mean ergodic.

(A) is uniformly mean ergodic.

(A) is a Montel space.

(A) does not contain an isomorphic copy of

(A) is not Montel. According to [9, Proposition 2.3] we

(A) = Y Z with Z isomorphic to

. Hence there exists a power

(A) which is not mean ergodic. This establishes (i)

(A); see for example,

is not mean ergodic.

([0, 1]) with ||

is uniformly mean ergodic.

n, for n N, and so we have

. By assumption there exists Q L(L

= T is uniformly mean ergodic.

is uniformly mean ergodic, then it is mean ergodic.

([0, 1]) with ||

1; see the proof of (iii) (iv) above.

, then the basis (x

equipped with the topology (X

, X)). For this reason, (f

, for all m N. Since X = proj

which contradicts the choice of (x

. If it were convergent then, being a subse-

; contradiction! So, we can again apply [15, p.54

)-Cauchy but, has no convergent subsequence in X

, X)) and we write X

). For a Schauder basis (x

as n and endow it with the topology (H, X) in-

(i.e. the dual basis (f

. Bases and basic sequences of types P

were introduced and studied in Banach spaces by I. Singer, [50].

(u) := sup[x, u)[ : p(x) 1 = sup[x, u)[ : p(x) = 1, u X

is the gauge of the polar U

) is a Banach space and the transpose map

denes a continuous linear functional u = v

such that lim

, endowed with the topology (H, X) induced by (X

) and, by [29, Proposition

= X. But, this is impossible because (t

; contradiction! The proof is thereby complete.

. So, for the 1-dimensional spaces E

) always form a Schauder decomposition of X

), [32, p.134]. If, in addition, (P

, then the original sequence (P

with property (M).

and a bounded sequence (z

and similar spaces, in Functional

and similar spaces, Math. Z.

Vous aimerez peut-être aussi