Rev Mat Complut (2010) 23: 113137

DOI 10.1007/s13163-009-0003-1
On stability of K-monotonicity of Banach couples
Sergey V. Astashkin Konstantin E. Tikhomirov
Received: 3 November 2008 / Accepted: 31 March 2009 / Published online: 4 November 2009
Revista Matemtica Complutense 2009
Abstract Suppose that E is a separable Banach lattice of two-sided real sequences
such that e
n
= 1 (n N), where {e
n
}
nZ
is the standard basis. One of the main aims
of this paper is a characterization of couples

E =(E, E(2
k
)) whose K-monotonicity
is stable when multiplying the weight by a constant. It is shown that such a property
holds only for a couple

E constructed upon a shift-invariant lattice. We construct
also a non-trivial example of shift-invariant separable Banach lattice E such that the
couple

E is K-monotone. The last result contrasts with the following well-known
theorem due to Kalton: if E is a separable symmetric Banach lattice such that the
couple

E is K-monotone then either E =l
p
(1 p <) or E =c
0
.
Keywords Banach lattices Interpolation of operators Peetre K-functional Real
method of interpolation K-monotone Banach couples Shift-invariant spaces
Caldern-Lozanovski

i spaces
Mathematics Subject Classication (2000) 46M35 46E30
1 Introduction
A pair (X
0
, X
1
) of Banach spaces X
0
and X
1
is called a (compatible) Banach
couple if there is a Hausdorff topological vector space in which each of X
0
and
X
1
is continuously embedded. Let

X = (X
0
, X
1
) and

Y = (Y
0
, Y
1
) be two Banach
The rst author was partly supported by the Russian Fund of Basic Research grant 07-01-96603-a.
S.V. Astashkin () K.E. Tikhomirov
Department of Mathematics and Mechanics, Samara State University, Samara 443011, Russia
e-mail: astashkn@ssu.samara.ru
K.E. Tikhomirov
e-mail: ktikhomirov@yandex.ru
114 S.V. Astashkin, K.E. Tikhomirov
couples and X and Y be Banach spaces such that X
0
X
1
X X
0
+ X
1
and
Y
0
Y
1
Y Y
0
+ Y
1
. The spaces X and Y are said to be C-interpolation spaces
with respect to

X and

Y if every linear operator T bounded from X
0
into Y
0
and from
X
1
into Y
1
is bounded from X into Y and
T
XY
Cmax(T
X
0
Y
0
, T
X
1
Y
1
).
For an arbitrary Banach couple

X =(X
0
, X
1
), any x X
0
+X
1
and t >0 we dene
the Peetre K-functional as follows:
K(t, x;

X) = inf{x
0

X
0
+t x
1

X
1
: x =x
0
+x
1
, x
i
X
i
}.
Banach couples

X = (X
0
, X
1
) and

Y = (Y
0
, Y
1
) are called (uniformly) relatively
K-monotone with constant > 0 if for any C > 0 and arbitrary C-interpolation
spaces X and Y the inequality
K(t, y;

Y) K(t, x;

X),
fullled for some x X, y Y
0
+ Y
1
and all t > 0, implies: y Y and y
Y

Cx
X
. In particular, if

X =

Y and X = Y then we get the denition of a (uni-
formly) K-monotone couple with constant , or Caldern couple. K-monotone Ba-
nach couples are very important in interpolation theory because it is possible to give
a complete description of all interpolation spaces for such a couple. Specically,
if X is an interpolation space with respect to a K-monotone couple

X = (X
0
, X
1
)
then X = (X
0
, X
1
)
K
E
, where (X
0
, X
1
)
K
E
is a space of the real K-method of inter-
polation constructed by some Banach lattice E of two-sided real sequences and
consisting of all x X
0
+ X
1
such that (K(2
k
, x;

X))

k=
E with the norm
x := (K(2
k
, x;

X))
k

E
[4, Theorem 4.4.5].
Here, mainly, we will consider separable Banach lattices E of two-sided real
sequences with the Fatou property and such that e
n

E
= 1, where {e
n
}
nZ
is
the canonical vector basis. We recall that E has the Fatou property if x
k
E,
x
k

E
1 and x
k
x (coordinate-wise) imply x E and x
E
1. Every such
lattice generates the Banach couple

E = (E, E(2
k
)), where E(v
k
) is the weighted
space consisting of all sequences a = (a
k
)

k=
such that a v E, with the norm
a
E(v
k
)
:= a v
E
. K-monotonicity of such a couple is connected with some highly
specic geometric properties of E which we dene below.
If x = (x
k
)

k=
is a real sequence then its support is dened by supp x = {k
Z : x
k
= 0}. If A Z and B Z then the inequality A < B means that a < b for
arbitrary a A, b B. Let {x
n
}
m
n=1
, {y
n
}
m
n=1
be two families of sequences. The pair
(x
n
, y
n
)
m
n=1
is interlaced if supports of x
n
, y
n
are nite and
supp x
n
<supp y
n
(1 n m),
supp y
n
<supp x
n+1
(1 n m1).
We say that a Banach sequence space E has the right-shift property (RSP) if there
exists C
RS
>0 such that for any interlaced pair (x
n
, y
n
)
m
n=1
with y
n

E
x
n

E
= 1
On stability of K-monotonicity of Banach couples 115
(1 n m) and for all a
n
R we have
_
_
_
_
_
m

n=1
a
n
y
n
_
_
_
_
_
E
C
RS
_
_
_
_
_
m

n=1
a
n
x
n
_
_
_
_
_
E
.
Analogously, E has the left-shift property (LSP) if for some C
LS
> 0, any interlaced
pair (x
n
, y
n
)
m
n=1
, x
n

E
y
n

E
= 1 (1 n m) and all a
n
R
_
_
_
_
_
m

n=1
a
n
x
n
_
_
_
_
_
E
C
LS
_
_
_
_
_
m

n=1
a
n
y
n
_
_
_
_
_
E
.
Kalton showed [13, Theorem 4.5] that a couple

E = (E, E(2
k
)) is K-monotone if
and only if E has (RSP) and (LSP).
One of the main results of this paper is a characterization of couples

E =
(E, E(2
k
)) whose K-monotonicity is stable when multiplying the weight by a con-
stant. Specically, it is shown that such a property holds only for a couple

E con-
structed upon a shift-invariant lattice. It may be interesting to compare this result
with the well-known Cwikel-Nilsson theorem [8] which implies, in particular, that if
E is separable and K-monotonicity of a couple

E = (E, E(2
k
)) is stable for any
replacement of the weight then E coincides either with l
p
(1 p < ) or with c
0
.
At the same time, we show that a class of K-monotone couples

E given by separa-
ble shift-invariant Banach lattices is not trivial. Specically, we construct a separable
shift-invariant lattice F which has the properties (RSP), (LSP) and such that F = l
p
(1 p < ), F = c
0
. Let us note that the situation changes completely if we con-
sider symmetric sequence spaces. As Kalton showed [13, Theorem 4.6], in that case
a couple

E is K-monotone if and only if either E =l
p
(1 p <) or E =c
0
.
2 K
u
-monotone Banach couples
Let X be a lattice of measurable functions dened on some measure space with a
-nite measure, v(t ) > 0 be a weighted function, and X(v) be the corresponding
weighted space consisting of all functions f (t ) such that f v X, with the norm
f
X(v)
= f v
X
. By the Sedaev result [17], if X = L
p
(1 p ) then for any
weights v
0
, v
1
, w
0
, w
1
couples (X(v
0
), X(v
1
)) and (X(w
0
), X(w
1
)) are relatively K-
monotone with a constant depending on p only. The converse statement was proved
by Cwikel and Nilsson in [8]. To be more precise, they showed that in the case when
X is a separable lattice of functions with the Fatou property, the following conditions
are equivalent:
(1) for any weights v
0
, v
1
, w
0
, w
1
the couples (X(v
0
), X(v
1
)) and (X(w
0
), X(w
1
))
are relatively K-monotone with a constant >0 independent of weights;
(2) there are a weight v and p [1, ) such that X coincides with L
p
(v) with
equivalence of norms.
In this article we are concerned with Banach couples whose K-monotonicity is sta-
ble concerning to a weaker disturbance of weight, namely, multiplying by a constant.
116 S.V. Astashkin, K.E. Tikhomirov
Specically, let us suppose that (X
0
, X
1
) is a couple of Banach lattices of measur-
able functions dened on some space with a -nite measure. We will consider the
question when the couples (X
0
, X
1
(u)) and (X
0
, X
1
) are relatively K-monotone with
a constant > 0 independent of a real number u > 0? As it is well known (see, for
example, [10, Chap. 15] or [4]), the answer is positive if and only if there exists a
constant > 0 such that the following is true: if x, y X
0
+ X
1
,

> and u > 0

then the inequality
K(t, y; X
0
, X
1
) K(ut, x; X
0
, X
1
) (t >0) (2.1)
implies that T x = y for some linear operator T : X
i
X
i
(i = 0, 1) such that
T
X
0
X
0

and T
X
1
X
1

u. If a couple (X
0
, X
1
) satises the last con-
dition we will call it K
u
-monotone. Let us remark that this notion makes sense for
arbitrary Banach couples as well.
Example 1 Showthat the Banach couple (L
1
, L

) of function spaces on the semiaxis

(0, ) with Lebesgue measure is K
u
-monotone. Indeed, if f, g L
1
+L

and u >0
then in view of [3, Theorem 5.2.1] inequality (2.1) can be rewritten as follows:
_
t
0
g

(s) ds
_
ut
0
f

(s) ds (t >0)
or, after change of variables,
_
t
0
g

(s)
u
ds
_
t
0

1/u
f

(s) ds (t >0),
where
v
f (s) :=f (s/v). Then, by the Caldern-Mityagin theorem (see [5] or [15]),
for an arbitrary small > 0 there exists a linear operator T bounded in L
1
and in
L

such that T
L
1
L
1
1 + , T
L

1 + and g = uT (
1/u
f ). Since

L
1
L
1
= v and
v

= 1 (v > 0), the operator T

u
h := u T (
1/u
h)
satises the conditions T
u

L
1
L
1
1 + , T
u

u(1 + ). Moreover,
T
u
f =g. Therefore, the couple (L
1
, L

) is K
u
-monotone with constant 1.
The next simple statement generalizes the preceding example. We recall that a
Banach couple

X = (X
0
, X
1
) is a partial retract of a Banach couple

Y = (Y
0
, Y
1
) if
every x X
0
+X
1
is orbitally equivalent to some y Y
0
+Y
1
. This means that there
are linear operators U :

X

Y and V :

Y

X such that Ux = y and Vy = x. If
there is a constant C > 0 independent of x X
0
+X
1
such that for any C

> C we
can choose operators U and V which satisfy additionally
U

Y
= max
i=0,1
{U
X
i
Y
i
} C

,
V

Y

X
= max
i=0,1
{V
Y
i
X
i
} C

,
then

X is called a uniform partial retract of

Y (with constant C).
On stability of K-monotonicity of Banach couples 117
Proposition 2.1 Let a Banach couple

X = (X
0
, X
1
) be a uniform partial retract of
a weighted L
p
-couple. Then

X is K
u
-monotone.
Proof By hypothesis,

X is a uniform partial retract of a couple

L = (L
p
0
(w
0
),
L
p
1
(w
1
)), where 1 p
0
, p
1
, w
0
>0, w
1
>0.
Suppose that x, y X
0
+ X
1
and u > 0 satisfy (2.1). Then for some C > 0 and
any >0 there are f, g L
p
0
(w
0
) +L
p
1
(w
1
) and linear operators U
i
:

X

L and
V
i
:

L

X such that U
i

L
C + , V
i

L

X
C + (i = 1, 2), U
1
x = f,
U
2
y =g, V
1
f =x, V
2
g =y. From the denition of the Peetre K-functional we get
(C +)
1
K(t, x;

X) K(t, f ;

L) (C +)K(t, x;

X) (t >0)
and
(C +)
1
K(t, y;

X) K(t, g;

L) (C +)K(t, y;

X) (t >0).
So, (2.1) implies
K(t, g;

L) (C +)
2
K(t u, f ;

L) =(C +)
2
K(t, f ;

L
u
) (t >0),
where

L
u
:= (L
p
0
(w
0
), L
p
1
(uw
1
)). Since

L
u
and

L are weighted L
p
-couples,
then there exists > 0 independent of u and a linear operator W :

L
u


L
such that W
L
p
0
(w
0
)L
p
0
(w
0
)
(C + )
2
, W
L
p
1
(uw
1
)L
p
1
(w
1
)
(C + )
2
and g = Wf (see [18] and [9]). Consequently, W
L
p
0
(w
0
)L
p
0
(w
0
)
(C + )
2
,
W
L
p
1
(w
1
)L
p
1
(w
1
)
(C + )
2
u and the operator T := V
2
WU
1
satises all con-
ditions from the denition of K
u
-monotonicity with constant (C +)
4
.
It is well known (see [11] or [7]) that the conditions of the last proposition are
fullled for couples of the form (

X

0
,p
0
,

X

1
,p
1
) (0 <
0
,
1
< 1, 1 p
0
, p
1
),
where (X
0
, X
1
)
,p
:=(X
0
, X
1
)
K
l
p
(2
k
)
, i.e., the space with the norm
x
(X
0
,X
1
)
,p
=
_

k=
(K(2
k
, x; X
0
, X
1
)2
k
)
p
_
1/p
(with appropriate modication when p = ) [3]. Thus, we come to
Corollary 2.2 For an arbitrary Banach couple

X =(X
0
, X
1
) and any 0 <
0
,
1
<1
and 1 p
0
, p
1
the Banach couple (

X

0
,p
0
,

X

1
,p
1
) is K
u
-monotone.
3 Couples given by shift-invariant Banach lattices
Suppose that the shift operator P
k
(a
j
) := (a
k+j
)

j=
is bounded in a Banach lat-
tice of two-sided real sequences E for any k Z. We call E a shift-invariant lattice
if sup
kZ
P
k

EE
< . Later on by {e
n
}
nZ
we denote the canonical basis in se-
quence spaces, i.e., e
n
= (. . . , 0, 1
.,,.
n
, 0, . . . ), and by E

a Banach lattice E(2

k
)
( (0, 1)).
118 S.V. Astashkin, K.E. Tikhomirov
Theorem 3.1 For an arbitrary Banach lattice E of two-sided sequences such that
e
n

E
= 1 (n Z) the following conditions are equivalent:
(a) the couple

E =(E, E(2
k
)) is K
u
-monotone;
(b) the couple

E is K-monotone and, in addition, there exists C > 0 such that for
any (0, 1) the inequality
K(t, b;

E) K(ut, a;

E) (t >0), (3.1)
where a E

, b E +E(2
k
) and u >0, implies that b E

and
b
E

Cu

a
E

; (3.2)
(c) the couple

E is K-monotone and, in addition, there exist C > 0 and (0, 1)
such that inequality (3.1), fullled for some a E

, b E +E(2
k
) and u >0,
implies b E

and (3.2);
(d) the couple

E is K-monotone and the space E is shift-invariant.
Proof (a) (b). First of all, a direct examination shows that E

isometrically coin-
cides with the Caldern-Lozanovski

i space E
1
E(2
k
)

. Recall that it consists of

all
a =(a
k
)

k=
such that
|a
k
| C|a
0
k
|
1
|a
1
k
|

(k Z), (3.3)
for some C > 0, a
0
= (a
0
k
) E, a
1
= (a
1
k
) E(2
k
), a
0

E
1, a
1

E(2
k
)
1.
Moreover, a
E
1
E(2
k
)
= inf C, where the inmum is taken over all C > 0 satis-
fying (3.3).
Since E is separable then from [14, Theorem 4.1.14 and remarks after it] it follows
that the spaces E
u

:= E
1
(uE(2
k
))

and E

are 2-interpolation with respect to

the couples

E
u
:= (E, uE(2
k
)) and

E := (E, E(2
k
)), for all (0, 1) and u > 0
(after the standard complexication these spaces isometrically coincide with corre-
sponding 1-interpolation spaces of the complex method of interpolation). Therefore,
inequality (3.1) implies
b
E

2a
E
u

, (3.4)
for some >0. We remark that (3.3) is equivalent to the following inequality:
|a
k
| Cu

|a
0
k
|
1

a
1
k
u

(k Z),
and also a
1
/u
uE(2
k
)
= a
1

E(2
k
)
. Thus, a
E
u

= u

a
E

and (3.4) leads

to (3.2).
Implication (b) (c) is evident, and so we turn to proving that (c) implies (d).
First, since the couple

E =(E, E(2
k
)) is (uniformly) K-monotone, Theorem 4.3
from [13] implies that E has (RSP), (LSP), and then, by [2, Lemma 3], P
n

EE

On stability of K-monotonicity of Banach couples 119
C

(n +1) (n 0) and P
n

EE
C

(1 n) (n <0). Hence, the Caldern operator

dened by
Q(
k
)
n
:=

k=
min(1, 2
nk
)
k
is bounded in the Banach lattice E

for every (0, 1), which implies that any linear

operator bounded from l
1
into l

and from l
1
(2
k
) into l

(2
k
) is bounded in E

.
Moreover, since e
n
= 1 (n Z), we have
l
1
E l

. (3.5)
Thus,
(l
1
, l
1
(2
k
))
K
E

=(E, E(2
k
))
K
E

=(l

, l

(2
k
))
K
E

=E

(3.6)
(more general results in this direction can be found in [1]).
Suppose that E is not shift-invariant. Then, there exist a sequence of integers
{n
k
}

k=1
, converging either to + or to , and a sequence a
k
= (a
k
j
)

j=
E
such that a
k

E
= 1 (k = 1, 2, . . . ) and
lim
k
P
n
k
a
k

E
= . (3.7)
If we set b
k
:=(a
k
j
2
j
)
jZ
, then b
k
E

, b
k

= 1 (k = 1, 2, . . . ) and
P
n
k
b
k
=(a
k
jn
k
2
(jn
k
)
)
jZ
= 2
n
k

((P
n
k
a
k
)
j
2
j
)
jZ
.
Consequently, P
n
k
b
k

= 2
n
k

P
n
k
a
k

E
and in view of (3.7)
lim
k
2
n
k

P
n
k
b
k

= . (3.8)
Let us denote c
k
j
:= K(2
j
, b
k
;

E) and c
k
= (c
k
j
)
jZ
. Taking into consideration
(3.5), we get:
|b
k
j
| sup
iZ
min(1, 2
ji
)|b
k
i
| =K(2
j
, b
k
; l

, l

(2
i
))
K(2
j
, b
k
;

E) =c
k
j
(j Z , k = 1, 2, . . . ).
By that, if d
k
:=P
n
k
c
k
, then d
k
P
n
k
(|b
k
|), hence, having in mind (3.8),
lim
k
2
n
k

d
k

= . (3.9)
Let us dene a sequence of functions
f
k
(t ) :=K(2
n
k
t, b
k
;

E) (t >0), k = 1, 2, . . . .
120 S.V. Astashkin, K.E. Tikhomirov
Since E is separable, the intersection E E(2
k
) is dense in the sum E +E(2
k
).
Therefore, f
k
belongs to
0
(k = 1, 2, . . . ), where
0
is the set of all increasing
concave functions dened on the semiaxis (0, ) and such that
lim
t 0+
(t ) = lim
t
(t )
t
= 0
(see, for example, [16, p. 386]).
Next, we need an auxiliary statement which, in fact, is already known [16, p. 398].
Nevertheless, we give its proof for the sake of completeness as well as to emphasize
universality of constants appearing.
Lemma 3.2 For any function
0
there exists a sequence = (
i
)
iZ

l
1
+l
1
(2
k
) such that
1
2
(t ) K(t, ; l
1
, l
1
(2
k
)) (t ) (t >0).
Proof As it is well known (see, for example, [3, p. 152]), a function
0
has the
following integral representation:
(t ) =
_

0
min(t, s) d(

(s)) (t >0), (3.10)

with the usual Riemann-Stieltjes integral on the right (since is concave, derivative

decreases). Recall also the following well-known formula:

K(t, (c
i
); l
1
, l
1
(2
k
)) =

i=
min(1, t 2
i
)|c
i
| (t >0).
So, if
i
:= 2
i
(

(2
i
)

(2
i+1
)) and =(
i
), then in view of (3.10)
(t ) =

i=
_
2
i+1
2
i
min(t, s) d(

(s))

i=
min(t, 2
i
)(

(2
i
)

(2
i+1
))
= K(t, ; l
1
, l
1
(2
k
)) (t >0),
whence, in particular, l
1
+l
1
(2
k
). At the same time,
(t )

i=
min(t, 2
i+1
)(

(2
i
)

(2
i+1
))
2K(t, ; l
1
, l
1
(2
k
)) (t >0),
and the proof is completed.
On stability of K-monotonicity of Banach couples 121
Continuation of the proof of implication (c) (d). Since f
k

0
then, by
Lemma 3.2, for each k = 1, 2, . . . there exists a sequence
k
l
1
+ l
1
(2
k
) such
that
1
2
f
k
(t ) K(t,
k
; l
1
, l
1
(2
k
)) f
k
(t ) (t >0). (3.11)
From (3.5) it follows that
k
E +E(2
k
). Having in mind the denition of f
k
we
get
K(t,
k
;

E) K(t,
k
; l
1
, l
1
(2
k
)) K(2
n
k
t, b
k
;

E) (t >0).
Therefore, the sequences
k
, b
k
and u = 2
n
k
satisfy condition (3.1), so it follows
that
k
E

and

C2
n
k

b
k

=C2
n
k

(k = 1, 2, . . . ).
But then the denition of d
k
and relations (3.11) and (3.6) imply that
d
k

= (K(2
jn
k
, b
k
;

E))
j

2(K(2
j
,
k
; l
1
, l
1
(2
k
)))
j

2C
1

2C
1
C2
n
k

(k = 1, 2, . . . ),
which contradicts (3.9). Thus, our assumption is wrong and E is shift-invariant.
(d) (a). We may assume that P
n

EE
= 1, n Z (this can be easily achieved
by replacing the norm in E with the following equivalent one x
E
:= sup
n
P
n
x
E
).
Suppose that inequality (3.1) is true, for some a, b E+E(2
k
) and u >0. We have
to nd a linear operator T :

E

E such that T
EE
, T
E(2
k
)E(2
k
)
u
and T x =y, where a constant is independent of a, b and u. Note, that it is enough
to prove the last statement only in the case when u = 2
k
(k Z).
In accordance with terminology of [13], the couple

E is exponentially separated
and (n) := e
n

E
/e
n

E(2
k
)
= 2
n
(n Z). Therefore, in view of [13, Lemma 4.1]
there exists C
2
>0 such that for any x E +E(2
k
)
K(t, x;

E) x (, n]
E
+t x (n, )
E(2
k
)
C
2
K(t, x;

E), (3.12)
if 2
n
t 2
n+1
(for any x = (x
j
)

j=
and V Z we put x V = (x V)
j
, where
(x V)
j
=x
j
if j V and (x V)
j
= 0 if j V).
Let u = 2
k
, k Z. If 2
n
t 2
n+1
, then 2
n+k
ut 2
n+k+1
, and, by concavity
of K-functional and relations (3.12) and (3.1), we get
K(t, b;

E) 2(a (, n +k]
E
+2
n+k
a (n +k, )
E(2
k
)
). (3.13)
Since E is a shift-invariant lattice, then
a (, n +k]
E
=
_
_
_
_
_
n+k

i=
a
i
e
i
_
_
_
_
_
E
=
_
_
_
_
_
n

i=
a
i+k
e
i
_
_
_
_
_
E
= (P
k
a) (, n]
E
,
122 S.V. Astashkin, K.E. Tikhomirov
and analogously
2
k
a (n +k, )
E(2
k
)
= 2
k
_
_
_
_
_

i=n+k+1
a
i
2
i
e
i
_
_
_
_
_
E
=
_
_
_
_
_

i=n+1
a
i+k
2
i
e
i
_
_
_
_
_
E
= (P
k
a) (n, )
E(2
i
)
.
Thus, (3.13) can be rewritten in the following way:
K(t, b;

E) 2((P
k
a) (, n]
E
+2
n
(P
k
a) (n, )
E(2
i
)
).
Since the last inequality holds for all 2
n
t 2
n+1
, n Z, then applying (3.12) once
again we get
K(t, b;

E) 2C
2
K(t, P
k
a;

E) (t >0). (3.14)
By hypotheses, the Banach couple

E is (uniformly) K-monotone. So, there exists a
linear operator T

:

E

E satisfying T

EE
C
3
, T

E(2
k
)E(2
k
)
C
3
and
T

(P
k
a) =b. Now we put T :=T

P
k
. Since E is shift-invariant, then
T
EE
T

EE
P
k

EE
C
3
and
T
E(2
k
)E(2
k
)
T

E(2
k
)E(2
k
)
P
k

E(2
k
)E(2
k
)
C
3
2
k
.
Moreover, T a =b. Hence,

E is K
u
-monotone and the proof is completed.
Now we can give examples of K-monotone but not K
u
-monotone Banach couples.
Example 2 We consider an arbitrary Banach lattice of sequences G having the fol-
lowing properties: (a) G is separable; (b) e
k

G
= 1; (c) G has (LSP) and (RSP);
(d) G is not shift-invariant. Note that such lattices exist. The rst example is the
direct sum l
q
(Z

) l
r
(Z
+
) (1 q =r <) with the norm
(x
k
) =
_

k=1
|x
k
|
q
_
1/q
+
_

k=0
|x
k
|
r
_
1/r
.
The second one is the well-known Tsirelson space T and its versions [6]. Right- and
left-shift properties for the former of these spaces are checked directly, for the latter
see [6, Proposition II.4].
In view of (c) and [13, Theorem 4.5] every such Banach couple (G, G(2
k
)) is
K-monotone. At the same time, since Gis not shift-invariant, by Theorem3.1, the last
couple is not K
u
-monotone. In particular, according to Proposition 2.1 (G, G(2
k
))
is not a uniformly partial retract of any weighted L
p
-couple.
On stability of K-monotonicity of Banach couples 123
A natural question arises in connection with the preceding examples, namely, does
there exist a separable shift-invariant Banach lattice with (RSP), (LSP) which does
not coincide with l
p
(1 p < ) and c
0
? In view of [13] and Theorem 3.1 the
question can be reformulated this way: does there exist a K
u
-monotone couple of
the form (E, E(2
k
)), where E is a separable Banach lattice, e
k

E
= 1, E = l
p
(1 p < ), E = c
0
? This question is also interesting in connection with the well-
known Kalton result [13, Theorem 4.6] saying that for a symmetric sequence space
E such a couple is K-monotone if and only if E =l
p
(1 p <) or E =c
0
. In the
next part of the article such an example, interesting as we hope in other respects as
well, is constructed.
4 A shift-invariant Banach lattice with (RSP) and (LSP)
Our construction of a shift-invariant Banach lattice F possessing the properties (RSP)
and (LSP) is somewhat similar to the one of the dual T

for Tsirelsons space pro-

posed by Figiel and Johnson [12]. The choice of the recursive norm in F is caused
by results stated in [6] where, in particular, it is proved that T

has (RSP) and (LSP).

However, the proof of these properties in the case of F signicantly differs from
the one in the case of T

. In this connection it may be interesting to compare our

Lemma 4.7 with Lemma I.11 from [6].
Denition 4.1 If a, b Z, a < b, then [a; b] := {a, a + 1, . . . , b}. By a partition
we will mean a nite ordered set of intervals {[a
j
; a
j+1
1]}
p
j=1
, where a
j
Z and
a
j+1
>a
j
for all j = 1, 2, . . . , p.
We denote by C a set of all partitions. Let c C. We use next the following nota-
tion:
cardc is number of intervals in the partition c;
c[j]j-th interval in c (numeration begins with 1);
cardc[j] = maxc[j] minc[j] +1.
Denition 4.2 A partition c is called admissible if
card{j : cardc[j] k} 14k
for any k >0. The set of all admissible partitions will be denoted by A.
At rst, we dene the sequence of the norms {
m
}

m=0
on the space c
00
of all
real sequences which are eventually 0. If x =

n
e
n
c
00
then
_
x
0
= max
n
|
n
|;
x
m+1
= max(x
m
,
1
2
max
cA

cardc
j=1
c[j]x
m
), m 0.
It is easy to see that the space c
00
with the norm
m
is a shift-invariant lattice, and
for all x c
00
x
c
0
x
m
x
l
1
.
124 S.V. Astashkin, K.E. Tikhomirov
Since x
m+1
x
m
for any x, there exists a limit x
A
= lim
m
x
m
. Let
F be the
A
-completion of c
00
. The simplest properties of F are listed in the
following proposition.
Proposition 4.1
(i) F is a separable shift-invariant Banach lattice with the Fatou property.
(ii) For any x F
x
A
= max
_
x
c
0
,
1
2
sup
cA
cardc

j=1
c[j]x
A
_
. (4.1)
(iii) For any sequence (y
j
)

j=1
of vectors y
j
c
00
such that
suppy
j
[b
j
; b
j+1
1], b
j+1
=b
j
+
_
j
14
_
, j = 1, 2, . . .
we have
_
_
_
_
_

j=1
y
j
_
_
_
_
_
A

1
2

j=1
y
j

A
.
Proposition 4.2 F does not coincide with l
p
(1 p <) and c
0
.
Proof According to property (iii) from Proposition 4.1, F contains a basic sequence
equivalent to the canonical basis in l
1
. Thus, it is enough to showthat F =l
1
. Suppose
the contrary. Let
K := sup
xF
x
l
1
x
A
<.
We choose vector y
1
=

i

i
e
i
c
00
such that
y
1

A
<
8
7K
y
1

l
1
,
and dene
y
n
=

i=

i
e
i+(n1)d
, d = diamsuppy
1
+1, n >1,
where here and next diamU := maxU minU (U Z). Now, using (4.1), let us
estimate norm of the vector y =

N
n=1
y
n
, where N = 28(2d 1) + 4. If y
A
=
y
c
0
, then we immediately come to contradiction. Therefore,
y
A
=
1
2
cardc

j=1
c[j]y
A
On stability of K-monotonicity of Banach couples 125
for some partition c. Let
H = {n : card{j : c[j]y
n
= 0} >1}.
Since the interval I = [minsuppy
minH
; maxsuppy
maxH
] contains at least cardH 1
intervals from c, we get
Nd cardI

c[j]I
cardc[j] =

k=1
card{c[j] I : cardc[j] k}

k=1
max(0, cardH 1 14(k 1))
2d

k=1
(cardH 1 14(k 1))
= 2d(cardH 1) 7(1 +(2d 1))(2d 1),
whence, it follows
cardH
N
2
+7(2d 1) +1 =
3N
4
.
If n G:= [1; N] \ H, then
cardc

j=1
c[j]y
n

A
y
n

A
.
Thus,
y
A
=
1
2
cardc

j=1
_
_
_
_
_
c[j]
N

n=1
y
n
_
_
_
_
_
A

1
2
cardc

j=1
_
_
_
_
c[j]

nH
y
n
_
_
_
_
A
+
1
2
cardc

j=1
_
_
_
_
c[j]

nG
y
n
_
_
_
_
A

nH
y
n

A
+
1
2

nG
y
n

A

7
8
Ny
1

A
<
1
K
y
l
1
.
We came to contradiction which proves the statement.
Now we formulate the main result of this part of the article.
Theorem 4.3 F possesses the properties (RSP) and (LSP).
Corollary 4.4 The Banach couple (F, F(2
k
)) is K
u
-monotone and it is not a uni-
form partial retract of any weighted L
p
-couple.
Proof K
u
-monotonicity of this couple is an immediate result of applying Theo-
rems 3.1, 4.3, Proposition 4.1 and [13, Theorem 4.5]. The second part follows from
Proposition 4.2 and the proof of Theorem 1 from [2].
126 S.V. Astashkin, K.E. Tikhomirov
We split the proof of Theorem 4.3 in two stages. First, we show that F satises
(RSP) and (LSP) for interlaced pairs of families consisting of vectors e
n
(the weak
shift properties (WRSP), (WLSP)). Then, by comparing norms of arbitrary linear
combinations of nite vectors and corresponding linear combinations of e
n
we obtain
the desired result.
Denition 4.3 We say that a Banach lattice E has the weak right-shift property
(WRSP), if there is C
WRS
> 0 such that for any interlaced pair (e
n
i
, e
m
i
)
k
i=1
(i.e.,
n
i
<m
i
<n
i+1
) and arbitrary
i
R
_
_
_
_
_
k

i=1

i
e
m
i
_
_
_
_
_
E
C
WRS
_
_
_
_
_
k

i=1

i
e
n
i
_
_
_
_
_
E
.
Analogously, if
_
_
_
_
_
k

i=1

i
e
n
i
_
_
_
_
_
E
C
WLS
_
_
_
_
_
k

i=1

i
e
m
i
_
_
_
_
_
E
,
for any
i
R, then E has the weak left-shift property (WLSP).
Proposition 4.5 F has (WRSP), (WLSP) and C
WRS
=C
WLS
2.
First, we introduce a number of denitions and prove some statements which will
form a basis for further discussion.
Denition 4.4 Let c = {[a
j
; a
j+1
1]}
p
j=1
be a partition. We call a partition c

en-
largement of c if c

= {[a
j
l
; a
j
l+1
1]}
r
l=1
for some 1 =j
1
<j
2
< <j
r+1
=p+1.
The constant of the enlargement c

is the value
En(c

, c) = max
l=1,...,r
(j
l+1
j
l
).
In other words, En(c

, c) is the maximal number of intervals from c which are joined

together in one interval of c

.
Denition 4.5 An ordered set
S = {c
1
; c
2
; . . . ; c
r
}, c
l
C (l = 1, . . . , r),
is called a system of partitions, if
maxc
l
[cardc
l
] = minc
l+1
[1] 1 (l = 1, . . . , r 1).
Denition 4.6 We say that a system of partitions S

= {c

1
; c

2
; . . . ; c

r
} is an enlarge-
ment of a system S = {c
1
; c
2
; . . . ; c
r
}, if c

l
is an enlargement of c
l
for all 1 l r.
We also dene the constant of the enlargement S

as follows:
En(S

, S) = max
l=1,...,r
En(c

l
, c
l
).
On stability of K-monotonicity of Banach couples 127
Denition 4.7 We will say that a partition c is given by a system of partitions S =
{c
1
; c
2
; . . . ; c
r
}, if
c =
r
_
l=1
c
l
.
Now, we are ready to state the rst auxiliary statement.
Lemma 4.6 Let c

1
, c

2
, . . . , c

h
be admissible partitions (h 14), S = {c
1
; c
2
; . . . ; c
h
}
be a system of partitions (not necessarily admissible) such that cardc
l
= cardc

l
for
any 1 l h and
j
2

j=j
1
cardc
l
[j]
1
3
j
2
2

j=j
1
cardc

l
[j]
if
2 j
1
<j
2
cardc
l
1, j
2
j
1
+2.
Then there exists a systemS
e
being an enlargement of S with constant En(S
e
, S) 14
such that the partition c
e
given by the system S
e
is admissible.
Proof It is enough to show that for any 1 l h we can construct an enlargement
c
e
l
of the partition c
l
with a constant satisfying En(c
e
l
, c
l
) 14 and such that
card{r <cardc
e
l
: cardc
e
l
[r] k} <k (4.2)
for all k >0. We will successively join intervals from c
l
, i.e., we set
R
l
=
_
cardc
l
14
_
1;
c
e
l
[r] =
14r
_
j=14(r1)+1
c
l
[j], r = 1, . . . , R
l
;
c
e
l
[R
l
+1] =
cardc
l
_
j=14R
l
+1
c
l
[j].
Now let us assume that there is a k >0 which fails (4.2), i.e.,
cardc
e
l
[r
v
] k, 1 r
v
R
l
, v = 1, . . . , k.
Let
j
v
1
= 14(r
v
1) +2 2;
j
v
2
= 14r
v
cardc
l
1.
128 S.V. Astashkin, K.E. Tikhomirov
According to the restrictions on c
l
,
j
v
2
2

j=j
v
1
cardc

l
[j] 3
j
v
2

j=j
v
1
cardc
l
[j] <3cardc
e
l
[r
v
] 3k.
Summing by v, we get
k

v=1
j
v
2
2

j=j
v
1
cardc

l
[j] <3k
2
. (4.3)
Let M = {j : j [j
v
1
; j
v
2
2]; v [1; k]}. Then cardM = 11k. As c

l
A,

jM
cardc

l
[j] =

k
1
=1
card{j M : cardc

l
[j] k
1
}

k
1
=1
max(0, 11k 14(k
1
1)) =

11k
14
+1

k
1
=1
(11k 14(k
1
1))
= 11k
__
11k
14
_
+1
_
14
1
2
__
11k
14
_
+1
__
11k
14
_

11k
14
_
11k 7
_
11k
14
__

121k
2
28
>3k
2
.
The obtained estimate contradicts (4.3). Consequently, our assumption is wrong, and
the lemma is proved.
Denition 4.8 A partition c is called a standard partition of a vector x c
00
, if
(1) c[j]x = 0 for all 1 j cardc and (2) suppx

cardc
j=1
c[j].
Denition 4.9 Let
x =
k

i=1
a
i
e
n
i
, n
i
<n
i+1
(i = 1, . . . , k 1),
y =
k

i=1
a
i
e
m
i
, m
i
<m
i+1
(i = 1, . . . , k 1)
(4.4)
be two nite vectors and c
1
, c
2
be two standard partitions of x and y, respectively.
We say that c
1
and c
2
are equivalent with respect to x and y, if
n
i
c
1
[j] m
i
c
2
[j]
for all possible i, j. Certainly, equivalent partitions have the same number of inter-
vals.
On stability of K-monotonicity of Banach couples 129
Let us introduce a binary relation P on the set of nite vectors. For two vectors x
and y representable in the form (4.4), the pair <x, y >belongs P if and only if for all
1 i
1
i
2
k and for any standard admissible partition c

of the vector

i
2
i=i
1

i
e
m
i
there exists an equivalent partition c of the vector

i
2
i=i
1

i
e
n
i
such that
j
2

j=j
1
cardc[j]
1
3
j
2
2

j=j
1
cardc

[j] (4.5)
if
2 j
1
<j
2
cardc 1, j
2
j
1
+2. (4.6)
Lemma 4.7 Let <x, y > P. Then
x
m+1

1
2
cardc

l=1
c

[l]y
m
,
for any partition c

C such that cardc

14 and for arbitrary m 0.

Proof We prove this statement by induction. We may (and will) assume that c

is a
standard partition of y. Let c be an equivalent partition of x. Since cardc 14, then
c is admissible and, consequently, the statement is true for m= 0. Now we x m>0.
We dene the sets M
0
and M
m1
as follows:
M
0
= {l : c

[l]y
m
= c

[l]y
0
};
M
m1
= [1; cardc

] \ M
0
.
It is easy to see that if l M
m1
then
c

[l]y
m
=
1
2
cardc

j=1
c

l
[j]c

[l]y
m1
,
for some standard admissible partition c

l
of the vector c

[l]y. Next, we dene the

following system of partitions S = {c
l
}
cardc
l=1
:
(i) if l M
0
then c
l
= {c[l]};
(ii) if l M
m1
then c
l
is a partition of c[l]x, equivalent to the partition c

l
of the
vector c

[l]y such that c

l
and c

l
satisfy (4.5) and
minc
l
[1] = minc[l];
maxc
l
[cardc
l
] = maxc[l].
(4.7)
Note that equalities (4.7) can be achieved by means of a suitable choice of cardc
l
[1]
and cardc
l
[cardc
l
]. Since lengths of the rst and the last intervals are missing in
formula (4.5) and <x, y > P, then a needed partition c
l
exists.
130 S.V. Astashkin, K.E. Tikhomirov
According to Lemma 4.6, there is an enlargement S
e
of the system S with constant
at most 14 such that the partition c
e
given by the system S
e
is admissible. Conse-
quently,
x
m+1

1
2
cardc
e

r=1
c
e
[r]x
m
.
Let
R
m1
=
_
r : l M
m1
, c
e
[r]
cardc
l
_
j=1
c
l
[j]c[l]
_
.
It is obvious that
_
rR
m1
c
e
[r] =
_
lM
m1
cardc
l
_
j=1
c
l
[j]c[l].
Taking into account that each of the intervals c
e
[r] is an union of at most 14 intervals
c
l
[j]c[l], we can use the inductive hypothesis. Therefore,

rR
m1
c
e
[r]x
m

1
2

lM
m1
cardc

j=1
c

l
[j]c

[l]y
m1
=

lM
m1
c

[l]y
m
.
If r / R
m1
then c
e
[r] =c[l] for some l M
0
and
c
e
[r]x
m
c
e
[r]x
0
= c

[l]y
0
= c

[l]y
m
.
Finally, we get
x
m+1

1
2
cardc

l=1
c

[l]y
m
,
and the proof is completed.
Now we are ready for proving that F has (WRSP) and (WLSP).
Proof of Proposition 4.5 We prove only the property (WRSP). Let (e
n
i
; e
m
i
)
k
i=1
be
an interlaced pair,
x =
k

i=1

i
e
n
i
and y =
k

i=1

i
e
m
i
.
Having in mind Lemma 4.7, we have to establish only that the pair <x, y > belongs
to P. We consider an arbitrary standard partition c

of the vector y and construct an

equivalent partition c of the vector x in the following way:
c[1] = [minsuppx; max{n
i
: m
i
c

[1]}];
On stability of K-monotonicity of Banach couples 131
c[j] = [max{n
i
: m
i
c

[j 1]} +1; max{n

i
: m
i
c

[j]}], 2 j cardc.
Let j
1
, j
2
satisfy inequalities (4.6). Then
j
2

j=j
1
cardc[j] = max{n
i
: m
i
c

[j
2
]} max{n
i
: m
i
c

[j
1
1]}
=n
i
M
n
i
m
,
where i
M
= max{i : m
i
c

[j
2
]}, i
m
= max{i : m
i
c

[j
1
1]}. Then, as m
i

m
i1
<n
i+1
n
i1
(2 i k 1),
j
2
2

j=j
1
cardc

[j] <m
i
M
1
m
i
m
=
i
M
1

i=i
m
+1
(m
i
m
i1
)
<
i
M
1

i=i
m
+1
(n
i+1
n
i1
) <2(n
i
M
n
i
m
).
From the last formula we conclude
j
2

j=j
1
cardc[j] >
1
2
j
2
2

j=j
1
cardc

[j].

Now we are going to show that the expansion operator is bounded in F.

Proposition 4.8 For any nite vector

i

i
e
n
i
(n
i
<n
i+1
)
_
_
_
_

i
e
2n
i
_
_
_
_
A
2
_
_
_
_

i
e
n
i
_
_
_
_
A
.
Proof Let x =

k
i=1

i
e
n
i
, y =

k
i=1

i
e
2n
i
. Then we need to prove that x, y P.
Consider an arbitrary standard partition c

= {[a

j
; a

j+1
1]}
p
j=1
of y. If
c = {[a
j
; a
j+1
1]}
p
j=1
, where a
j
=
_
a

j
2
_
,
then
a

j+1
2
>
a

j
2
. Indeed, otherwise, a

j+1
=a

j
+1, and a

j
1 (mod 2). It would
mean that the interval [a

j
; a

j+1
1] does not contain even numbers and therefore the
partition c

is not standard for y. Thus, a

j+1
>a
j
(j = 1, . . . , p1), and c is dened
correctly. It is easy to see that 2n
i
[a

j
; a

j+1
1] implies n
i
[a
j
; a
j+1
1], that
is, c is equivalent to c

with respect to the vectors x and y. Let us show that for any j
_
a

j+1
2
_

_
a

j
2
_

1
3
(a

j+1
a

j
).
132 S.V. Astashkin, K.E. Tikhomirov
If a

j+1
1 (mod 2) or both numbers a

j+1
, a

j
are even then the inequality is evident.
We consider the case when a

j+1
0 (mod 2), a

j
1 (mod 2). Since there is an
even number in the interval [a

j
; a

j+1
1], then a

j+1
a

j
+3. Consequently,
_
a

j+1
2
_

_
a

j
2
_
=
a

j+1
2

a

j
+1
2

a

j+1
a

j
2

1
2
a

j+1
a

j
3
=
a

j+1
a

j
3
.
Thus, c and c

satisfy (4.5), and the statement is proved.

Nowwe turn to comparing norms of suitable linear combinations of arbitrary nite
vectors and vectors of the standard basis.
Lemma 4.9 Let x =

N
n=1
u
n
, u
n
=

i
u
n
i
e
i
c
00
, suppu
n
< suppu
n+1
, n =
1, . . . , N 1, and for all n [1; N 1]
maxsuppu
n
+diamsuppu
n
minsuppu
n+1
.
Then
_
_
_
_
_
N

n=1
u
n

A
e
minsuppu
n
_
_
_
_
_
A
2
_
_
_
_
_
N

n=1
u
n
_
_
_
_
_
A
.
Proof Let us denote
y =
N

n=1
u
n

A
e
minsuppu
n ; u
n,2
=

i=
u
n
i
e
2i
; x
2
=
N

n=1
u
n,2
.
Taking into account Proposition 4.8, it is enough to prove that
y
A
x
2

A
.
Show that for any standard admissible partition c

of the vector y there exists a stan-

dard admissible partition c of the vector x
2
such that cardc = cardc

and
minsuppu
n
c

[j] suppu
n,2
c[j]. (4.8)
We dene c in the following way:
c = {[a
j
; a
j+1
1]}
cardc

j=1
;
a
j
= minsuppu
n
j
+a

j
, j = 1, . . . , cardc

, a
cardc+1
= 2a

cardc

+1
;
n
j
= min{n : minsuppu
n
c

[j]}, j = 1, . . . , cardc

.
First of all, a
j+1
a
j
>a

j+1
a

j
, j = 1, . . . , cardc, i.e., the partition c is admissi-
ble. Let a

j
minsuppu
n
for some j, n. Then
n
j
= min{h : minsuppu
h
a

j
} n
On stability of K-monotonicity of Banach couples 133
and
a
j
minsuppu
n
j
+minsuppu
n
2minsuppu
n
= minsuppu
n,2
.
Suppose, on the contrary, a

j
>minsuppu
n
. Then n
j
n +1, and, by hypotheses,
minsuppu
n
j
maxsuppu
n
+diamsuppu
n
,
whence it follows that
a
j
2maxsuppu
n
minsuppu
n
+a
j
>maxsuppu
n,2
.
Thus, (4.8) holds for c and c

. Then we use recursive formula (4.1) and induction

by N, nishing the proof.
Proposition 4.10 Let x =

N
n=1
u
n
, suppu
n
< suppu
n+1
, n = 1, . . . , N 1. Then
there exist i
n
suppu
n
(n = 1, . . . , N) such that
_
_
_
_
_
N

n=1
u
n

A
e
i
n
_
_
_
_
_
A
8
_
_
_
_
_
N

n=1
u
n
_
_
_
_
_
A
.
Proof We represent each vector u
n
as a sum
u
n
=u
1,n
+u
2,n
,
where u
1,n
and u
2,n
are dened as follows:
(a) if diamsuppu
n
= 0 then u
1,n
=u
n
, u
2,n
= 0;
(b) if diamsuppu
n
>0 then suppu
1,n
<suppu
2,n
,
diamsuppu
1,n
=
diamsuppu
n
2
.
Then
minsuppu
1,n+1
>maxsuppu
n
2diamsuppu
1,n
+minsuppu
1,n
= maxsuppu
1,n
+diamsuppu
1,n
.
Applying Lemma 4.9 to vectors u
1,n
, we get
_
_
_
_
_
N

n=1
u
1,n

A
e
minsuppu
n
_
_
_
_
_
A
2
_
_
_
_
_
N

n=1
u
1,n
_
_
_
_
_
A
.
Analogous estimate holds also for vectors u
2,n
. Combining it with the equality
minsupp(u
2,n
) = maxsuppu
2,n
, we obtain that
_
_
_
_
_
N

n=1
u
2,n

A
e
maxsuppu
n
_
_
_
_
_
A
2
_
_
_
_
_
N

n=1
u
2,n
_
_
_
_
_
A
.
134 S.V. Astashkin, K.E. Tikhomirov
As a result,
8
_
_
_
_
_
N

n=1
u
n
_
_
_
_
_
A
4
__
_
_
_
_
N

n=1
u
1,n
_
_
_
_
_
A
+
_
_
_
_
_
N

n=1
u
2,n
_
_
_
_
_
A
_
2
_
_
_
_
_
N

n=1
u
1,n

A
e
i
1,n
_
_
_
_
_
A
+2
_
_
_
_
_
N

n=1
u
2,n

A
e
i
2,n
_
_
_
_
_
A
2
_
_
_
_
_
N

n=1
max(u
1,n

A
, u
2,n

A
)e
i
n
_
_
_
_
_
A

_
_
_
_
_
N

n=1
u
n

A
e
i
n
_
_
_
_
_
A
,
where i
1,n
= minsuppu
n
, i
2,n
= maxsuppu
n
, i
n
{i
1,n
; i
2,n
}.
Later on we will use the notation: if a, b Z, a b and, as usual, [a; b] = {x
Z : a x b}, then
I[a; b] :=(a, b) = {x Z : a <x <b} and J[a; b] := {a; b}.
Proposition 4.11 For an arbitrary sequence of nite vectors {u
n
}
N
n=1
, suppu
n
<
suppu
n+1
(n = 1, . . . , N 1), and for any m 0 there exist i
n
suppu
n
(n =
1, . . . , N) such that
_
_
_
_
_
N

n=1
u
n
_
_
_
_
_
m

_
_
_
_
4

nI[1;N]
u
n

m
e
i
n
+2

nJ[1;N]
u
n

m
e
i
n
_
_
_
_
m
. (4.9)
Proof We carry on the proof by induction. If m = 0 the statement is evident. Let
m>0, x =

N
n=1
u
n
and
x
m
=
1
2
cardc

j=1
c[j]x
m1
,
for some standard admissible partition c.
We divide the interval [1; N] into four sets (some of them may be empty):
H
1
:= {n : {n
1
: c[j]u
n
1
= 0} = {n} for some j}
(that is, there exists an interval c[j] which intersecs only with support of u
n
),
H
2
:= {n / H
1
: c[j]u
n
= 0, c[j +1]u
n
= 0 for some j}
(corresponding vectors u
n
do not belong to the rst set and their supports intersect
with two intervals c[j] and c[j +1]),
H
3
:= {n I[1; N] \ H
1
: suppu
n
c[j] for some j}
On stability of K-monotonicity of Banach couples 135
(corresponding vectors u
n
again do not belong to the rst set and their supports are
contained within one of the intervals, excluding u
1
and u
N
),
H
4
:=J[1; N] \ H
1
.
It follows easily that the sets of indices H
l
, 1 l 4, are mutually disjoint and
their union coincides with [1; N]. Moreover, we introduce the sets G
1
, G
2
:
G
1
= {j [1; cardc] : card{n : c[j]u
n
= 0} = 1};
G
2
= {j [1; cardc] : card{n : c[j]u
n
= 0} >1}.
At rst,
cardc

j=1
_
_
_
_
_
c[j]
N

n=1
u
n
_
_
_
_
_
m1

cardc

j=1
_
_
_
_
c[j]

nH
1
u
n
_
_
_
_
m1
+

jG
2
_
_
_
_
c[j]

nH
2
H
3
H
4
u
n
_
_
_
_
m1
=S
1
+S
2
.
According to the denition of the norm
m
, we have
S
1

nH
1
cardc

j=1
c[j]u
n

m1
2

nH
1
u
n

m
=

jG
1
_
_
_
_
c[j]

nH
1
2u
n

m
e
i
n
_
_
_
_
m1
,
where i
n

jG
1
suppc[j]u
n
, n H
1
.
Now let us estimate the sum S
2
. If n H
2
and c[j]u
n
= 0 for some j, then
n J[min{n
1
: c[j]u
n
1
= 0}; max{n
1
: c[j]u
n
1
= 0}].
Analogously, in the case when n H
4
we have
n J[min{n
1
: c[1]u
n
1
= 0}; max{n
1
: c[1]u
n
1
= 0}] or
n J[min{n
1
: c[cardc]u
n
1
= 0}; max{n
1
: c[cardc]u
n
1
= 0}].
Then, by the inductive hypothesis,
S
2

jG
2
_
_
_
_
c[j]
_

nH
2
2u
n

m1
(e
i
1,n
+e
i
2,n
)
+

nH
3
4u
n

m1
e
i
n
+

nH
4
2u
n

m1
e
i
n
__
_
_
_
m1
,
where i
n
suppc[j]u
n
, i
1,n
suppc[j]u
n
, i
2,n
suppc[j + 1]u
n
for correspond-
ing j. Let us note that for
m1
, as well as any other norm, the inequality

i
z
i
1
+z
i
2
+z
i
3

m1
max
_

i
2z
i
1
+z
i
3

m1
,

i
2z
i
2
+z
i
3

m1
_
136 S.V. Astashkin, K.E. Tikhomirov
holds for arbitrary z
i
1
, z
i
2
, z
i
3
. Consequently, by replacing i
1,n
or i
2,n
with i
n
, we get
S
2

jG
2
_
_
_
_
c[j]
_

nH
2
4u
n

m
e
i
n
+

nH
3
4u
n

m
e
i
n
+

nH
4
2u
n

m
e
i
n
__
_
_
_
m1
,
where i
n
suppu
n
. As a result we have
2x
m
S
1
+S
2

cardc

j=1
_
_
_
_
c[j]
_

nH
1
2u
n

m
e
i
n
+

nH
2
4u
n

m
e
i
n
+

nH
3
4u
n

m
e
i
n
+

nH
4
2u
n

m
e
i
n
__
_
_
_
m1
2
_
_
_
_

nH
1
2u
n

m
e
i
n
+

nH
2
4u
n

m
e
i
n
+

nH
3
4u
n

m
e
i
n
+

nH
4
2u
n

m
e
i
n
_
_
_
_
m
.
Taking into account the embedding {1; N} H
1
H
4
, we get inequality (4.9).
Proof of Theorem 4.3 We apply Propositions 4.5, 4.10 and 4.11. Note that the con-
stant C
RSP
=C
LSP
2 8 4 = 64.
References
1. Astashkin, S.V.: A description of interpolation spaces between (l
1
(w
0
), l
1
(w
1
)) and
(l

(w
0
), l

(w
1
)). Mat. Zametki 35(4), 497503 (1984) (in Russian). English transl.: Math.
Notes 35(4), 261265 (1984)
2. Astashkin, S.V.: K-monotone weighted pairs generated by a space non-invariant concerning to shift.
In: Trudy of 2nd Intern. Semin. Differ. Equations and their Appl., Samara, pp. 1925 (1998) (in
Russian)
3. Bergh, J., Lofstrom, J.: Interpolation Spaces. An Introduction. Springer, Berlin (1976)
4. Brudny, Yu.A., Krugljak, N.Ya.: Interpolation Functors and Interpolation Spaces. North-Holland,
Amsterdam (1991)
5. Caldern, A.P.: Spaces between L
1
and L

and the theorem of Marcinkiewicz. Stud. Math. 26(3),

273299 (1966)
6. Casazza, P.G., Shura, T.J.: Tsirelsons Space. Lect. Notes Math., vol. 1363, pp. 1204. Springer, Berlin
(1989)
7. Cwikel, M.: Monotonicity properties of interpolation spaces 2. Arch. Math. 19(1), 123136 (1981)
8. Cwickel, M., Nilsson, P.: The coincidence of real and complex interpolation methods for couples of
weighted Banach lattices. In: Lect. Notes Math., vol. 1070, pp. 5465. Springer, Berlin (1984)
9. Dmitriev, V.I.: On interpolation of operators in L
p
spaces. Dokl. Akad. Nauk SSSR 260, 10511054
(1981) (in Russian). English transl.: Sov. Math. Dokl. 24, 373376 (1981)
10. Dmitriev, V.I., Krein, S.G., Ovchinnikov, V.I.: Fundamentals of the theory of interpolation of linear
operators. In: Geometry of Linear Spaces and Operator Theory, pp. 3174. Yaroslavl Gos. Univ.,
Yaroslavl (1977) (in Russian)
On stability of K-monotonicity of Banach couples 137
11. Dmitriev, V.I., Ovchinnikov, V.I.: On the real method spaces. Dokl. Akad. Nauk SSSR. 246(4), 794
797 (1979) (in Russian). English transl.: Sov. Math. Dokl. 20, 538542 (1979)
12. Figiel, T., Johnson, W.B.: A uniformly convex Banach space which contains no l
p
. Compos. Math.
29, 179190 (1974)
13. Kalton, N.J.: Caldern couples. Stud. Math. 106(3), 233277 (1993)
14. Krein, S.G., Petunin, Ju.I., Semenov, E.M.: Interpolation of Linear Operators. Am. Math. Soc., Prov-
idence (1982)
15. Mityagin, B.S.: An interpolation theorem for modular spaces. Mat. Sb. 66(4), 472482 (1965) (in
Russian). English transl.: Lect. Notes Math., vol. 1070, pp. 1023. Springer, Berlin (1984)
16. Ovchinnikov, V.I.: The method of orbits in interpolation theory. Math. Rep. 1, 349515 (1984)
17. Sedaev, A.A.: Description of interpolation spaces for the couple (L
p
a
0
, L
p
a
1
) and some related prob-
lems. Dokl. Akad. Nauk SSSR 209(4), 798800 (1973) (in Russian). English transl.: Sov. Math. Dokl.
14, 538541 (1973)
18. Sparr, G.: Interpolation of weighted L
p
-spaces. Stud. Math. 62, 229271 (1978)