MEASURE OF WEAK NONCOMPACTNESS
AND REAL INTERPOLATION OF OPERATORS
ANDRZEJ KRYCZKA, STANISLAW PRUS, AND MARIUSZ SZCZEPANIK
Abstract. A new measure of weak noncompactness is introduced. A logarithmic convexity-type result on the behaviour of this measure applied to
bounded linear operators under real interpolation is proved. In particular, it
gives a new proof of the theorem showing that if at least one of operators
T : Ai → Bi , i = 0, 1 is weakly compact, then so is T : Aθ,p → Bθ,p for all
0 < θ < 1 and 1 < p < ∞.
1. Introduction
In 1960, M.A. Krasnoselskii [20] proved that under the hypotheses of the Riesz–
Thorin interpolation theorem (i.e. if T is a linear operator such that T : Lpi → Lqi
is bounded for i = 0, 1 where pi , qi ∈ [1, ∞]) and the additional assumption that
T : Lp0 → Lq0 is compact and q0 < ∞, it follows that T : Lp → Lq is also compact.
Here 1/p = (1 − θ)/p0 + θ/p1 , 1/q = (1 − θ)/q0 + θ/q1 and 0 < θ < 1.
This has initiated a series of theorems which refer to the question whether in
the above theorem the couples (Lp0 , Lp1 ), (Lq0 , Lq1 ) and the spaces Lp , Lq can
be replaced by couples of Banach spaces (A0 , A1 ), (B0 , B1 ) and real interpolation
spaces Aθ,p , Bθ,p . More generally, if T , viewed as a map from Aθ,p to Bθ,p , inherits
any compactness properties which it may possess as an element of L(Ai , Bi ).
Since the works of J.-L. Lions and J. Peetre [21], several authors have obtained
results of different degrees of generality. Finally, M. Cwikel [15] showed that if
T : A0 → B0 is compact and T : A1 → B1 is bounded, then T : Aθ,p → Bθ,p
is also compact for all 0 < θ < 1 and 1 < p < ∞. The similar problem for
weak compactness properties of T was studied as well. Here a key result is due
to B. Beauzamy [8], who proved that if the embedding I : A0 ∩ A1 → A0 + A1 is
weakly compact, then so is I : Aθ,p → Aθ,p . Next, S. Heinrich [18] extended this
result to closed operator ideals. Other generalizations of Beauzamy’s result are due
to M.Kh. Aizenstein and Yu.A. Brudnyi (see the book [9]), and to L. Maligranda
and A. Quevedo [22] (see also M. Mastylo [23]) who established the following: if
0 < θ < 1, 1 < p < ∞ and T : A0 ∩ A1 → B0 + B1 is weakly compact, then so is
T : Aθ,p → Bθ,p . In particular, this is the case if T : A0 → B0 or T : A1 → B1 is
weakly compact.
The Riesz–Thorin theorem gives also a logarithmically convex estimate for the
norms of an interpolated operator. This motivated some authors to find quantitative versions of the above-mentioned results on compact and weakly compact
operators. The main tools in such investigations are measures of noncompactness
1991 Mathematics Subject Classification. Primary 46B70; Secondary 46A50, 47B07.
Key words and phrases. The Riesz–Thorin theorem, interpolation space, measure of weak
noncompactness, w∗-cluster point of a sequence, free ultrafilter.
1
2
A. KRYCZKA, S. PRUS, AND M. SZCZEPANIK
and weak noncompactness. Measures of noncompactness appear in various contexts, so it was convenient to define many different measures (see [1]). Here let us
only mention the Hausdorff and separation measures. In contrast, very few measures of weak noncompactness can be found in the literature. Most authors use the
measure of weak noncompactness ω introduced by F.S. De Blasi [16], which can
be seen as a counterpart of the Hausdorff measure of noncompactness. De Blasi’s
measure was successfully applied to operator theory and to the theory of differential and integral equations (see [5] and the references given there). The measure
ω was also used by A.G. Aksoy and L. Maligranda [3] in order to obtain a quantitative version of the theorem on real interpolation of weakly compact operators
(similar results for the Hausdorff measure of noncompactness were earlier proved
in [27]). Recently the thorough study of this kind of estimates for operators under
real interpolation was undertaken by F. Cobos, P. Fernández-Martı́nez, A. Manzano and A. Martı́nez in the series of papers [10]–[14]. In particular, some essential
restrictions on interpolation of De Blasi’s measure are indicated (see [12], [14]).
In this paper we introduce a new measure of weak noncompactness γ, which
can be treated as a counterpart of the separation measure of noncompactness. Its
definition bases on an idea due to R.C. James [19]. In the general case, the measures
ω and γ are not equivalent. The measure γ appeals directly to the norm topology,
while in the definition of ω the weak topology is involved. This seems to determine
more geometric character of γ. However, γ can be expressed by various formulae.
In particular, we show that γ coincides with the function based on the double-limit
criterion of weak compactness, which was considered by K. Astala and H. -O. Tylli
in [4]. Our main result shows that Riesz–Thorin-type estimates hold for the measure
γ without any additional restrictions.
Throughout this paper, by a Banach space we mean a real one. However, the
results presented here can be easily restated and proved for complex Banach spaces.
We denote the open unit ball of a Banach space X by BX and its closure by BX .
Furthermore, conv A stands for the convex hull of a set A ⊂ X and the closure
of conv A will be denoted by Conv A. For simplicity of notation we use the same
letter to designate an element of a Banach space X and its canonical image in the
second dual X ∗∗ . The abbreviations with the beginning w∗ refer to the weak-star
topology.
2. Measures of weak noncompactness
2.1. Measure γ. Let (xn ) be a sequence in a Banach space X. We shall say that
(yn ) is a sequence of successive convex combinations, or scc, for (xn ) if there exists
pn+1
a sequence of integers 0 = p1 < p2 < p3 < . . . such that yn ∈ conv(xi )i=p
n +1
for every n. Similarly, vectors u1 , u2 are said to be a couple of scc for (xn ) if
u1 ∈ conv(xi )pi=1 , u2 ∈ conv(xi )∞
i=p+1 for some p ∈ N.
The following theorem, motivated by [25], will play a significant role in the sequel.
Theorem 2.1. Let (xn ) be a bounded sequence in a Banach space X. For every
ε > 0 there exists a sequence (yn ) of scc for (xn ) such that if u1 , u2 and v1 , v2 are
any couples of scc for (yn ), then |ku1 − u2 k − kv1 − v2 k| ≤ ε.
T∞
Proof. Let An = Conv(xi )∞
i=n for n ∈ N. Assume first that
n=1 An 6= ∅. Then we
can find a convergent sequence of scc for (xn ) and, by ignoring a finite
T∞ number of
terms if necessary, we get a sequence as required. Suppose now that n=1 An = ∅.
WEAK NONCOMPACTNESS AND INTERPOLATION
3
Reasoning similar to that in the proof of the corollary of Theorem 2 [25] (see also
[26]) gives a constant d > 0 such that for any ε > 0 there exists (yn ) of scc for (xn )
such that d − ε ≤ ku1 − u2 k ≤ d for any couple u1 , u2 of scc for (yn ). Clearly, (yn )
fulfills the assertion of the theorem.
Now, following [6], we introduce an axiomatic approach to the notion of a measure of weak noncompactness. Let µ be a real-valued function defined on the family
of all bounded and nonempty subsets of a Banach space X. We call µ a measure of
weak noncompactness on X, if the following conditions are satisfied for any subsets
A, B and c ∈ R:
(1) µ(A) = 0 if and only if A is a relatively weakly compact set;
(2) if A ⊂ B, then µ(A) ≤ µ(B);
(3) µ(Conv A) = µ(A);
(4) µ(A ∪ B) = max{µ(A), µ(B)};
(5) µ(A + B) ≤ µ(A) + µ(B);
(6) µ(cA) = |c| µ(A).
Definition 2.2. We introduce the function γ defined for every nonempty and
bounded set A ⊂ X by the formula
γ(A) = sup{csep(xn ) : (xn ) ⊂ conv A},
where
csep(xn ) = inf{ky1 − y2 k : y1 , y2 is a couple of scc for (xn )}.
Theorem 2.3. γ is a measure of weak noncompactness on X.
Proof. Conditions (2),(3) and (6) are easy to check. Implications in (1) for γ are
consequences of the James theorem [19] and Mazur theorem [24] respectively.
To prove (5), let us take bounded A, B ⊂ X and a sequence (xnP
) ⊂ conv(A +
mn n
B). For every n ∈ N there exist nonnegative λn1 , . . . , λnmn , where i=1
λi = 1,
P
m
n
n
n
n n
and elements
u
∈
A,
v
∈
B,
1
≤
i
≤
m
such
that
x
=
λ
(u
+ vin ).
n
n
i
i
i
i
i=1
Pmn n n
Pmn n n
Let yn =
i=1 λi ui , zn =
i=1 λi vi and fix ε > 0. By Theorem 2.1 there
exists a sequence (yn′ ) of scc for (yn ) such that |kw1 − w2 k − kw1′ − w2′ k| ≤ ε for
′
′
′
term in (yn′ ) takes the form
any couples
Pnk+1 w1 , wk2 and w1 , w2k of scc for (yn ). Each P
nk+1
′
k
′
yk =
i=nk +1 µi yi , where µi are nonnegative and
i=nk +1 µi = 1. Set zk =
P
nk+1
k
′′
′
i=nk +1 µi zi . Applying again Theorem 2.1 we get a sequence (zn ) of scc for (zn )
′′
′′
′′
such that |kv1 − v2 k − kz1 − z2 k| ≤ ε for any couple v1 , v2 of scc for (zn ). For such
obtained (zn′′ ) we can construct a sequence (yn′′ ) of scc for (yn′ ) in the similar way
as we have constructed (zn′ ) for a given (yn′ ). Then
ky1′′ − y2′′ k ≤ csep(yn′′ ) + ε
and
kz1′′ − z2′′ k ≤ csep(zn′′ ) + ε.
Let us notice that x′′1 = y1′′ + z1′′ , x′′2 = y2′′ + z2′′ form a couple of scc for (xn ) and
therefore
csep(xn ) ≤ kx′′1 − x′′2 k ≤ ky1′′ − y2′′ k + kz1′′ − z2′′ k ≤ csep(yn′′ ) + csep(zn′′ ) + 2ε
≤ γ(A) + γ(B) + 2ε.
An arbitrary choice of ε yields csep(xn ) ≤ γ(A) + γ(B). Thus γ(A + B) ≤ γ(A) +
γ(B).
To obtain (4), first we see from (2) that max{γ(A), γ(B)} ≤ γ(A ∪ B). For the
opposite inequality let us choose a sequence (xn ) in conv(A ∪ B). Each xn takes
the form xn = tn yn + (1 − tn )zn for some tn ∈ [0, 1], yn ∈ conv A and zn ∈ conv B.
4
A. KRYCZKA, S. PRUS, AND M. SZCZEPANIK
There exists a subsequence (tnk ) convergent to some t ∈ [0, 1]. The sets A, B are
bounded, so M = sup{kxk : x ∈ A ∪ B} < ∞ and for arbitrarily fixed ε > 0 we can
assume that |tnk − t| ≤ Mε+1 for all k. Then k(tnk − t)ynk k ≤ Mε+1 kynk k ≤ ε and
similarly k(1 − tnk )znk − (1 − t)znk k ≤ ε. It follows that kxnk − uk k ≤ 2ε, where
uk = tynk + (1 − t)znk . Since (uk ) is a sequence in t conv A + (1 − t) conv B, we have
csep(uk ) ≤ γ(t conv A + (1 − t) conv B) ≤ tγ(A) + (1 − t)γ(B)
≤ max{γ(A), γ(B)}.
But csep(xnk ) ≤ csep(uk ) + 4ε, which shows that
csep(xn ) ≤ csep(xnk ) ≤ max{γ(A), γ(B)} + 4ε.
Finally, γ(A ∪ B) ≤ max{γ(A), γ(B)}.
In the next two results we establish alternative formulae for the measure γ.
Theorem 2.4. Let A be a nonempty and bounded subset of a Banach space X.
Then
(2.1)
γ(A) = sup dist(x∗∗ , conv(xn )) ,
where the supremum is taken over all sequences (xn ) in conv A and all w∗-cluster
points x∗∗ ∈ X ∗∗ of a sequence (xn ).
Proof. Let us denote by γ ′ (A) the right-hand side of formula (2.1). In order to show
that γ ′ (A) ≥ γ(A) we deduce likewise in the proof of Lemma 2.1 in [26]. Indeed,
for ε > 0 let us find a sequence (xn ) in conv A such that γ(A) − ε ≤ csep(xn ) and
fix x ∈ conv(xn ). There exists m ∈ N such that if y ∈ conv(xi )∞
i=m , then x, y is
a couple of scc for (xn ) and therefore kx − yk ≥ csep(xn ) ≥ γ(A) − ε. Applying
a separation theorem we can find a functional x∗ ∈ X ∗ such that kx∗ k ≤ 1 and
∗∗
x∗ (y − x) ≥ γ(A) − ε for all y ∈ conv(xi )∞
∈ X ∗∗ be a w∗-cluster
i=m . Let x
∗∗ ∗
∗
point of (xn ). Then x (x ) = limk→∞ x (xnk ) for some subsequence (xnk ) and
consequently kx∗∗ − xk ≥ (x∗∗ − x)(x∗ ) = limk→∞ x∗ (xnk − x) ≥ γ(A) − ε. This
gives dist(x∗∗ , conv(xn )) ≥ γ(A) − ε and finally, γ ′ (A) ≥ γ(A).
The proof of the opposite inequality is a modification of reasoning in [19] (see also
[17]). For each ε > 0 there exists a sequence (xn ) in conv A and its w∗-cluster point
x∗∗ ∈ X ∗∗ such that dist(x∗∗ , conv(xn )) ≥ γ ′ (A) − ε. By a separation theorem
we obtain a functional x∗∗∗ ∈ X ∗∗∗ such that kx∗∗∗ k ≤ 1 and x∗∗∗ (x∗∗ − x) ≥
γ ′ (A)−ε for all x ∈ conv(xn ). We now choose by induction a sequence of functionals
(x∗k ) ⊂ X ∗ and a subsequence (xnk ) with the following properties: kx∗k k ≤ 1 +
ε for all k ∈ N, x∗k (xni ) ≥ x∗∗∗ (x∗∗ ) − ε if k ≤ i and x∗k (xni ) = x∗∗∗ (xni ) if
i < k. To construct these sequences, we apply the principle of local reflexivity
[17, p.33]. By this principle, there exists x∗1 ∈ X ∗ such that kx∗1 k ≤ 1 + ε and
x∗∗ (x∗1 ) = x∗∗∗ (x∗∗ ). Since x∗∗ is the w∗-cluster point of (xn ) we can find n1 ∈ N
satisfying |x∗1 (xn1 ) − x∗∗ (x∗1 )| ≤ ε. Let us assume now, that we have obtained
required x∗1 , . . . , x∗k−1 and xn1 , . . . , xnk−1 . There exists x∗k ∈ X ∗ such that kx∗k k ≤
1 + ε, x∗k (xni ) = x∗∗∗ (xni ) for i = 1, . . . , k − 1 and x∗∗ (x∗k ) = x∗∗∗ (x∗∗ ). We
choose a number nk > nk−1 to satisfy |x∗i (xnk ) − x∗∗ (x∗i )| ≤ ε for i = 1, . . . , k.
Let us consider now a couple u, v of scc for (xni ). There exists k > 1 such that
k−1
∗
∗∗∗
(u) and x∗k (v) ≥
, v ∈ conv(xni )∞
u ∈ conv(xni )i=1
i=k . Of course xk (u) = x
∗∗∗ ∗∗
∗∗∗ ∗∗
∗
x (x )−ε, and hence (1+ε) ku − vk ≥ xk (v −u) ≥ x (x −u)−ε ≥ γ ′ (A)−2ε.
WEAK NONCOMPACTNESS AND INTERPOLATION
′
, which gives γ(A) ≥
From this it follows that csep(xni ) ≥ γ (A)−2ε
1+ε
ε → 0 we obtain γ(A) ≥ γ ′ (A), and the proof is complete.
5
γ ′ (A)−2ε
.
1+ε
Letting
The measure γ is equal to the function measuring the deviation from relative
weak compactness based on the classical double-limit criterion. The function of
this type was considered also in [4].
Theorem 2.5. Let A be a nonempty and bounded subset of a Banach space X.
Then
(2.2)
γ(A) = sup{lim lim Fn (xk ) − lim lim Fn (xk ) :
n
k
k
n
(xk ) ⊂ conv A, (Fn ) ⊂ BX ∗ and the limits exist}.
Proof. Let us denote by e(A) the right-hand side of formula (2.2). From the proof
of Theorem 2.4 it follows that there exists a such that for every ε > 0 one can find
sequences (xk ) ⊂ conv A and (Fn ) ⊂ BX ∗ satisfying the following: Fn (xk ) ≥ a − ε
for n ≤ k and Fn (xk ) ≤ a − γ(A) + ε for n > k. Passing to a subsequence,
if necessary, we can assume that all the limits α1 = limn limk Fn (xk ) and α2 =
limk limn Fn (xk ) exist. Of course, γ(A) − 2ε ≤ α1 − α2 . Hence γ(A) − 2ε ≤ e(A)
and γ(A) ≤ e(A).
Let (xk ) ⊂ conv A and (Fn ) ⊂ BX ∗ be any sequences for which the limits β1 =
limn limk Fn (xk ) and β2 = limk limn Fn (xk ) exist. By Theorem 2.1 for arbitrary
ε > 0 we can find a sequence (zk ) of scc for (xk ) such that kzi − zj k − csep(zk ) ≤ ε
for every i 6= j. Then β1 = limn limk Fn (zk ) and β2 = limk limn Fn (zk ). Therefore
β1 − β2 ≤ lim inf j lim inf i kzi − zj k. Consequently, we get β1 − β2 ≤ csep(zk ) + ε ≤
γ(A) + ε. Hence γ(A) + ε ≥ e(A) and finally γ(A) ≥ e(A).
2.2. γ and De Blasi’s measure ω. The axiomatic approach to a measure of weak
noncompactness, presented in subsection 2.1, was preceded by a definition of the
measure ω introduced by F.S. De Blasi (see [16] and the references given there).
For a nonempty and bounded subset A of a Banach space X let
ω(A) = inf{t > 0 : A ⊂ C + tBX and C ⊂ X is a weakly compact set}.
Theorem 2.5 and a result of K. Astala and H.-O. Tylli (see [4]) show that, in
general, the measures γ and ω are not equivalent. Certainly, if X is reflexive, then
BX is weakly compact and ω(BX ) = 0. Otherwise it is shown [16] that ω(BX ) = 1.
Similarly, for reflexive spaces γ(BX ) = 0. The nonreflexive case for γ differs from
the one for ω.
Example 2.6. Consider the vectors en = (0, . . . , 0, 1, 0, . . .) ∈ l1 , n ∈ N, where 1
occupies the n-th place. It is evident that csep(en ) = 2 and therefore γ(Bl1 ) = 2.
Example 2.7. The case of c0 is more complicated. From the next theorem we will
see at once that γ(Bc0 ) = 1.
Theorem 2.8. Let A be a nonempty and bounded subset in c0 . Then
(2.3)
γ(A) = sup dist(x∗∗ , c0 ) ,
where the supremum is taken over all sequences (xn ) in conv A and all w∗-cluster
points x∗∗ ∈ c∗∗
0 of (xn ).
6
A. KRYCZKA, S. PRUS, AND M. SZCZEPANIK
Proof. Let us denote by γ ′ (A) the right-hand side of formula (2.3). In view of
Theorem 2.4 it is enough to prove that γ(A) ≤ γ ′ (A). Choose M such that A ⊂
M Bc0 and fix a sequence (xn ) ⊂ conv A. For each w∗-cluster point x∗∗ = (x∗∗ (k))
of (xn ) there exists a subsequence (xni ) such that x∗∗ = w∗- limi→∞ xni . Let
q = dist (x∗∗ , c0 ) = lim supk→∞ |x∗∗ (k)| and yi = xni for i ∈ N. Fix ε > 0 and
N
N +1
N ∈ N. Let us choose a subsequence (yni )i=1 and an increasing sequence (ki )i=1 of
∗∗
natural numbers such that |x (k)| < q + ε for k > k1 , and satisfying the following
conditions: |yni (k) − x∗∗ (k)| < ε for k ≤ ki and |yni (k)| < ε for k > ki+1 ,
PN
i = 1, . . . , N . Let us define x = N1 i=1 yni ∈ conv(xn ). Then for k > k1 we have
|x (k) − x∗∗ (k)| < |x∗∗ (k)| + ε +
M
M
< q + 2ε +
N
N
and for k ≤ k1
|x (k) − x∗∗ (k)| < ε.
Both the inequalities and an arbitrary choice of ε and N yields dist (x∗∗ , conv(xn )) ≤
q and hence γ (A) ≤ γ ′ (A).
Theorem 2.9. The measures γ and ω are equal in c0 .
Proof. Having q ≥ 0, we set
rq (α) =
0
α
α − q |α|
if |α| ≤ q
if |α| > q.
Next, let Rq : c0 → c0 be given by the formula Rq x = (rq (x (k))), where x =
(x(k)). We take a nonempty bounded set A ⊂ c0 and put q = γ(A). We will
show that γ (Rq (A)) = 0. Let (xn ) ⊂ A and suppose that there exists the limit
∗∗
w∗- limn→∞ xn = x∗∗ ∈ c∗∗
0 . By Theorem 2.8 lim supk→∞ |x (k)| ≤ q. Therefore
∗∗
∗∗
∗
lim sup w - lim Rq xn (k) = lim sup |rq (x (k))| = rq lim sup |x (k)| = 0.
k→∞
n→∞
k→∞
k→∞
In view of Theorem 2.8 we obtain γ(Rq (A)) = 0. It follows that Rq (A) is relatively
weakly compact. Since A ⊂ Rq (A) + qBc0 , we get ω (A) ≤ q = γ (A).
On the other hand, we have γ(A) ≤ γ(Bc0 )ω(A) = ω(A) (see [7]).
3. Real interpolation of operators
3.1. Measure of weak noncompactness for operators. By L(X, Y ) we denote
the space of all linear and bounded operators T : X → Y between Banach spaces X
and Y . Using the measure γ, introduced in the previous section, we can define the
measure of weak noncompactness for every operator T ∈ L(X, Y ). This measure
provides in addition a seminorm in L(X, Y ).
Definition 3.1. For every T ∈ L(X, Y ) the number Γ(T ) = γ(T (BX )) will be
called the measure of weak noncompactness of the operator T .
Let X be a Banach space. We shall denote by lp (X), 1 < p < ∞ the Banach
space of all sequences x = (x(i)) such that x(i) ∈ X for all i ∈ N and kxklp (X) =
P∞
p 1/p
( i=1 kx(i)k )
is finite. A standard verification shows that we can identify
∗
(lp (X)) withPlq (X ∗ ), where 1/p + 1/q = 1, and for each φ = (φ(i)) ∈ (lp (X))∗ we
∞
have φ(x) = i=1 φ(i)(x(i)) for all x ∈ lp (X) (see [21]). If X, Y are Banach spaces
WEAK NONCOMPACTNESS AND INTERPOLATION
7
and T ∈ L(X, Y ), then the operator Te : lp (X) → lp (Y ) given by Tex = (T x(i)) is
also bounded and kTek = kT k.
To deal with the measure γ, ultrafilters will be used in several cases. For more
details concerning filters we refer the reader for instance to [2]. We recall two
important facts.
Lemma 3.2. Let U be an ultrafilter on a nonempty set I and I1 ∪ I2 = I. Then
I1 or I2 belongs to U.
Lemma 3.3. An ultrafilter U is free if and only if U does not contain a finite set.
Let τ be a Hausdorff topology in a space E and U be an ultrafilter on the set of
positive integers N. An element x ∈ E is said to be the limit over U of a sequence
(xn ) in E, if {n ∈ N : xn ∈ V } ∈ U for every neighbourhood V of x. Then we
write x = τ -limU xn or simply x = limU xn . Let us recall that if E is compact, then
limU xn exists for each sequence (xn ) in E.
The next lemma, roughly speaking, can stand for a passing to a subsequence. In
the light of Lemma 3.3, the proof is straightforward.
Lemma 3.4. Let U be a free ultrafilter on N and N1 ∈ U. If f : N1 → N is the
bijection given by f (nk ) = k, then U1 = {f (N1 ∩ A) : A ∈ U} is a free ultrafilter on
N. Moreover, if limU xn = x in a topological space X, then limU1 xnk = x as well.
Lemma 3.5. Let Y be a Banach space and 1 < p < ∞. If y = (y(i)), yn =
(yn (i)) ∈ lp (Y ∗∗ ) for all n ∈ N and y = w∗- limU yn over some free ultrafilter U on
N, then y(i) = w∗- limU yn (i) for each i ∈ N.
Proof. Let us first recall that if f : X → Y is continuous function between topological spaces, (xi )i∈I ⊂ X and x = limU ′ xi over a free ultrafilter U ′ on a set of
indices I, then f (x) = limU ′ f (xi ). Fix i ∈ N and take v = (0, . . . , 0, v(i), 0, . . .) ∈
lq (Y ∗ ). The functional fv given by the formula fv (z) = z(v) = z(i)(v(i)) for
z ∈ lp (Y ∗∗ ) is w∗-continuous. Consequently, fv (yn ) = yn (i)(v(i)) and y(i)(v(i)) =
fv (y) = w∗- limU yn (i)(v(i)) for every v(i) ∈ Y ∗ which is equivalent to y(i) =
w∗- limU yn (i).
Theorem 3.6. Let X, Y be Banach spaces and 1 < p < ∞. If T ∈ L(X, Y )
and Te ∈ L(lp (X), lp (Y )) is defined by Tex = (T x(i)) for x = (x(i)) ∈ lp (X), then
Γ(T ) = Γ(Te).
Proof. Since T = Te|{(x, 0, 0, . . .)}, we see that Γ(T ) ≤ Γ(Te). For the proof of
Γ(T ) ≥ Γ(Te), assume that Γ(Te) > 0, otherwise at once Γ(T ) = Γ(Te). Fix 0 < ε <
Γ(Te) and let Γ1 = Γ(Te) − ε. There exists (xn ) ⊂ Blp (X) such that for yn = Texn we
have 0 < Γ1 ≤ dist(y, conv(yn )) ≤ dist(y, (yn )), where y ∈ lp (Y ∗∗ ) is a w∗-cluster
point of the sequence (yn ) and therefore y = w∗- limU yn over some free ultrafilter
U in N. By a separation theorem there exists φ ∈ lq (Y ∗∗∗ ) such
P∞ that kφk ≤q 1 and
φ(z) ≥ Γ1 for all z ∈ y −conv(yn ). Now, fix m ∈ N satisfying i=m+1 kφ(i)k < εq .
The boundedness of y − conv(yn ) gives a constant c > 0 such that
!1/q
!1/p
m
∞
∞
X
X
X
q
p
Γ1 ≤
φ(i)(y(i) − yn (i)) +
kφ(i)k
ky(i) − yn (i)k
i=1
≤
m
X
i=1
i=m+1
φ(i)(y(i) − yn (i)) + εc
i=m+1
8
A. KRYCZKA, S. PRUS, AND M. SZCZEPANIK
for every n ∈ N, where, by Lemma 3.5, y(i) = w∗- limU yn (i) for all i. Writing
I = {1 ≤ i ≤ m : φ(i) 6= 0} and
ψ(i) =
φ(i)
for i ∈ I,
kφ(i)k
αi = lim kxn (i)k ,
U
v(i) =
y(i)
ε ,
αi + m
vn (i) =
yn (i)
ε ,
αi + m
by Hölder’s and Minkowski’s inequalities we obtain
X
ε
kφ(i)k (αi + )ψ(i) (v(i) − vn (i))
Γ1 − εc ≤
m
i∈I
!1/p
!
1/q
m
m
X
X
ε p
q
max ψ(i) (v(i) − vn (i))
(αi + )
≤
kφ(i)k
i∈I
m
i=1
i=1
≤ (1 + εm1/p−1 ) max ψ(i) (v(i) − vn (i)) .
i∈I
We set Γ2 = (Γ1 − εc)(1 + εm
(3.1)
1/p−1 −1
)
. Then
Γ2 ≤ max ψ(i) (v(i) − vn (i))
i∈I
for all n ∈ N. For each 1 ≤ i ≤ m let Ni denote the set of all natural
S numbers n
for which the maximum in (3.1) is attained for i. It is clear that 1≤i≤m Ni = N
and, by Lemma 3.2, Nj ∈ U for some j. Let us apply Lemma 3.4 for the set
ε
} = {nk } ∈ U and change U for a free ultrafilter
Nj ∩ {n ∈ N : |kxn (j)k − αj | < m
U1 given by this lemma. It follows that
Γ2 ≤ ψ(j) (v(j) − vnk (j))
for k ∈ N. Considering convex combinations of the obtained sequence yields
Γ2 ≤ dist (v(j), conv(vnk (j))) .
But (vnk (j)) ⊂ T (BX ) and v(j) = w∗- limU1 vnk (j), which gives Γ2 ≤ Γ(T ). Letting
ε → 0 we conclude that Γ(Te) ≤ Γ(T ) and this finishes the proof.
Remark 3.7. We can consider a space lp (X) of sequences indexed by the set of
all integers Z. The same properties as stated at the beginning of this section,
Lemma 3.5 and Theorem 3.6 also hold in that case.
3.2. Real interpolation. We shall say that Banach spaces A0 and A1 are compatible if they are continuously embedded in a Hausdorff topological vector space
X. Then A0 ∩ A1 and A0 + A1 with norms
kakA0 ∩A1 = max{kakA0 , kakA1 } and
kakA0 +A1 =
inf
a=a0 +a1
{ka0 kA0 + ka1 kA1 }
respectively, are also Banach spaces. Let A = (A0 , A1 ) denote a couple of compatible Banach spaces A0 and A1 . A Banach space A is said to be an intermediate
space with respect to A if
A0 ∩ A1 ⊂ A ⊂ A0 + A1
and both inclusions are continuous. Let A = (A0 , A1 ) and B = (B0 , B1 ) be two
couples of compatible Banach spaces and T a linear operator from A0 + A1 into
B0 +B1 . We shall write T : A → B for brevity, if T ∈ L(A0 , B0 ) and T ∈ L(A1 , B1 ),
viewed as the restrictions of T . If A and B are intermediate spaces with respect to
A and B respectively, and T : A → B implies T ∈ L(A, B), then A and B are said
to be interpolation spaces with respect to A and B.
WEAK NONCOMPACTNESS AND INTERPOLATION
9
In the sequel, we restrict our considerations to one of the equivalent constructions
of so called real interpolation spaces, i.e. a discrete method introduced by J.-L.
Lions and J. Peetre [21]. Let 0 < θ < 1, 1 < p < ∞ and for each a ∈ A0 + A1
kakθ,p = inf max{k(2iθ a0 (i))klp (A0 ) , k(2i(θ−1) a1 (i))klp (A1 ) },
where the infimum is taken over all (a0 (i)) ⊂ A0 , (a1 (i)) ⊂ A1 such that a0 (i) +
a1 (i) = a for all i ∈ Z. Then Aθ,p = {a ∈ A0 + A1 : kakθ,p < ∞} is an intermediate
space with respect to A and moreover one can show [21] that for a ∈ Aθ,p
(3.2)
kakθ,p ≤ 2θ(1−θ)
inf
a0 (i)+a1 (i)=a
i(θ−1)
k(2iθ a0 (i))k1−θ
a1 (i))kθlp (A1 ) .
lp (A0 ) k(2
The spaces Aθ,p and Bθ,p , obtained by this method, are interpolation spaces with
respect to A and B. Furthermore, for every T : A → B
1−θ
kT kθ,p ≤ 2θ(1−θ) kT k0
θ
kT k1 ,
where kT kθ,p and kT ki , i = 0, 1 are norms of operators T : Aθ,p → Bθ,p and
T : Ai → Bi , i = 0, 1 respectively.
We prove an analogous inequality for our measure of weak noncompactness for
operators. Note that a similar result for De Blasi’s measure does not hold (see [12]).
Theorem 3.8. Let Aθ,p and Bθ,p be interpolation spaces with respect to A =
(A0 , A1 ) and B = (B0 , B1 ) obtained by the real method described above for some
0 < θ < 1 and 1 < p < ∞. Then for every T : A → B
Γθ,p (T ) ≤ 2θ(1−θ) Γ0 (T )1−θ Γ1 (T )θ ,
where Γθ,p and Γi , i = 0, 1 are measures of weak noncompactness for operators
T : Aθ,p → Bθ,p and T : Ai → Bi , i = 0, 1 respectively.
Proof. Fix ε > 0 and a sequence (an ) ⊂ BAθ,p . For each an there exist
(2iθ a0n (i))i∈Z ∈ Blp (A0 )
and (2i(θ−1) a1n (i))i∈Z ∈ Blp (A1 )
such that a0n (i) + a1n (i) = an for all i ∈ Z. Let yn = (2iθ T a0n (i))i∈Z , zn =
(2i(θ−1) T a1n (i))i∈Z and bn = T an for every n ∈ N. By a similar method as in
the proof of condition (5) of Theorem 2.3, we can obtain sequences (yn′′ ), (zn′′ ) of
scc forP
(yn ), (zn ) respectively. They
satisfy therefore the assertion of Theorem 2.1,
Pnk+1
nk+1
′′
k
k
yk′′ = j=n
λ
= n < n2 < n3 < . . .
y
and
z
=
j j
j=nk +1 λj zj for some 0
k
k +1
Pnk+11
k
k
and some nonnegative coefficients λnk +1 , . . . , λnk+1 with j=nk +1 λkj = 1. We set
Pnk+1
b′′k = j=n
λkj bj for k = 1, 2. Then
k +1
1−θ
θ
csep(bn ) ≤ kb′′1 − b′′2 kθ,p ≤ 2θ(1−θ) ky1′′ − y2′′ klp (B0 ) kz1′′ − z2′′ klp (B1 ) ,
the last inequality being a consequence of (3.2). But ky1′′ − y2′′ klp (B0 ) ≤ csep(yn′′ ) +
f0 (Bl (A ) ) and zn′′ ∈
ε and kz1′′ − z2′′ klp (B1 ) ≤ csep(zn′′ ) + ε. Moreover, yn′′ ∈ T
p
0
f1 (Bl (A ) ) for all n ∈ N, where Tei : lp (Ai ) → lp (Bi ), i = 0, 1 is defined as in
T
p
1
Theorem 3.6. Hence
f0 ) + ε)1−θ (Γ(T
f1 ) + ε)θ .
csep(bn ) ≤ 2θ(1−θ) (Γ(T
Finally, Theorem 3.6 with Remark 3.7 and an arbitrary choice of ε and (an ) lead
to the desired conclusion.
10
A. KRYCZKA, S. PRUS, AND M. SZCZEPANIK
Since Γ(T ) = 0 if and only if T is weakly compact, the above theorem brings a
new proof of the following: if T : A0 → B0 or T : A1 → B1 is weakly compact,
then so is T : Aθ,p → Bθ,p for all 0 < θ < 1 and 1 < p < ∞. Let us also formulate
another immediate consequence of Theorem 3.8.
Corollary 3.9. Let 0 < θ < 1, 1 < p < ∞ and Aθ,p be the interpolation space with
respect to A = (A0 , A1 ). Then
γ(BAθ,p ) ≤ 2θ(1−θ) γ(BA0 )1−θ γ(BA1 )θ .
In particular, if A0 or A1 is reflexive then Aθ,p is reflexive as well.
References
[1] R.R. Akhmerov, M.I. Kamenskii, A.S. Potapov, A.E. Rodkina and B.N. Sadovskii, Measures
of noncompactness and condensing operators (Birkhäuser Verlag, Basel Boston Berlin 1992).
[2] A.G. Aksoy and M.A. Khamsi, Nonstandard methods in fixed point theory (Springer-Verlag,
Berlin Heidelberg New York 1990).
[3] A.G. Aksoy and L. Maligranda, ‘Real interpolation and measure of weak noncompactness’,
Math. Nachr. 175 (1995), 5–12.
[4] K. Astala and H.-O. Tylli, ‘Seminorms related to weak compactness and to Tauberian operators’, Math. Proc. Camb. Phil. Soc. 107 (1990), 367–375.
[5] J. Banaś, ‘Applications of measure of weak noncompactness and some classes of operators in
the theory of functional equations in the Lebesgue space’, Proceedings of the Second World
Congress of Nonlinear Analysts, Part 6 (Athens, 1996), Nonlinear Anal. 30 (1997), 3283–
3293.
[6] J. Banaś and A. Martinón, ‘Measures of weak noncompactness in Banach sequence spaces’,
Portugaliae Math. 52 (1995), 131–138.
[7] J. Banaś and J. Rivero, ‘On measures of weak noncompactness’, Ann. Mat. Pura Appl., 151
(1988), 213–224.
[8] B. Beauzamy, Espaces d’interpolation réels: topologie et géométrie (Springer-Verlag, Berlin
Heidelberg New York 1978).
[9] Yu.A. Brudnyi and N.Ya. Krugljak, Interpolation functors and interpolation spaces. Vol. I
(North-Holland, Amsterdam New York Oxford Tokyo 1991).
[10] F. Cobos, ‘Interpolation theory and measures related to operator ideals’, Nonlinear analysis,
function spaces and applications, Vol. 6 (eds M. Krbec and A. Kufner, Olympia Press,
Prague, 1999).
[11] F. Cobos, P. Fernández-Martı́nez and A. Martı́nez, ‘Interpolation of the measure of noncompactness by the real method’, Studia Math. 135 (1999), 25–38.
[12] F. Cobos, A. Manzano and A. Martı́nez, ‘Interpolation theory and measures related to operator ideals’, Quart. J. Math. Oxford Ser. (2) 50 (1999) 401–416.
[13] F. Cobos and A. Martı́nez, ‘Extreme estimates for interpolated operators by the real method’,
J. London Math. Soc. (2) 60 (1999) 860–870.
[14] F. Cobos and A. Martı́nez, ‘Remarks on interpolation properties of the measure of weak
non-compactness and ideal variations’, Math. Nachr. 208 (1999) 93–100.
[15] M. Cwikel, ‘Real and complex interpolation and extrapolation of compact operators’, Duke
Math. J. 65 (1992), 333–343.
[16] F.S. De Blasi, ‘On a property of the unit sphere in a Banach space’, Bull. Math. Soc. Sci.
Math. R.S. Roumanie 21 (1977), 259–262.
[17] D. van Dulst, Reflexive and superreflexive Banach spaces (Mathematisch Centrum, Amsterdam 1978).
[18] S. Heinrich, ‘Closed operator ideals and interpolation’, J. Funct. Anal. 35 (1980), 397–411.
[19] R.C. James, ‘Weak compactness and reflexivity’, Israel J. Math. 2 (1964), 101–119.
[20] M.A. Krasnoselskii, ‘On a theorem of M. Riesz’, Dokl. Akad. Nauk SSSR 131 (1960), 246–248
(Russian).
[21] J.-L. Lions and J. Peetre, ‘Sur une classe d’espaces d’interpolation’, Inst. Hautes Études Sci.
Publ. Math. 19 (1964), 5–68.
WEAK NONCOMPACTNESS AND INTERPOLATION
11
[22] L. Maligranda and A. Quevedo, ‘Interpolation of weakly compact operators’, Arch. Math. 55
(1990), 280–284.
[23] M. Mastylo, ‘On interpolation of weakly compact operators’, Hokkaido Math. Jour. 22 (1993),
105–114.
[24] S. Mazur, ‘Über konvexe Mengen in linearen normierten Räumen’, Studia Math. 4 (1933),
70–84.
[25] D.P. Milman and V.D. Milman, ‘The geometry of imbeddings with empty intersection. The
structure of the unit sphere in a non-reflexive space’, Mat. Sbornik 66 (1965), 109–118 (Russian).
[26] V.D. Milman, ‘Geometric theory of Banach spaces. II. Geometry of the unit ball’, Uspehi Mat.
Nauk 26 (1971), 73–149 (Russian). English translation: Russian Math. Surv. 26 (1971), 79–
163.
[27] M.F. Teixeira and D.E. Edmunds, ‘Interpolation theory and measures of non-compactness’,
Math. Nachr. 104 (1981), 129–135.
Institute of Mathematics, Maria Curie-Sklodowska University, 20-031 Lublin, Poland
E-mail address: akryczka@golem.umcs.lublin.pl
Institute of Mathematics, Maria Curie-Sklodowska University, 20-031 Lublin, Poland
E-mail address: bsprus@golem.umcs.lublin.pl
Institute of Mathematics, Maria Curie-Sklodowska University, 20-031 Lublin, Poland
E-mail address: szczepan@golem.umcs.lublin.pl