Note di Matematica 29, n. 2, 2009, 55–76.
Isomorphisms between spaces of multilinear mappings or homogeneous polynomials
Po Ling, Kuo
iDepartamento Acadˆemico de Matem´atica, Universidade Tecnol´ogica Federal do Paran´a Av. Sete de Setembro, 3165, CEP 80230-901, Curitiba, PR, Brazil
kuopoling@gmail.com
Received: 22/08/2008; accepted: 13/02/2009.
Abstract. LetE and F be Banach spaces. Our objective in this work is to find conditions under which, whenever the topological dual spacesEandFare isomorphic, the spaces of multilinear mappings (resp. homogeneous polynomials) onE and F are isomorphic as well.
We also examine the corresponding problem for the spaces of multilinear mappings (resp.
homogeneous polynomials) of a certain type, for instance of finite, nuclear, compact or weakly compact type.
Keywords:Banach space , Polynomials , Isomorphisms.
MSC 2000 classification:primary 46G20, secondary 46G25
Notation
Throughout the whole paper D,E, F and G always denote Banach spaces over the same field K, where K = R or C. N denotes the set of all positive integers. L(
mE; G) denotes the vector space of all continuous m-linear mappings from E
minto G. L(
mE; G) is a Banach space under its natural norm. If G = K, we write L(
mE; K) = L(
mE). If m = 1, we write L(
1E; G) = L(E; G). If m = 1 and G = K, we write L(E) = E
, the topological dual of E. The mapping
I
m: L(
m+nE; G) → L(
mE; L(
nE; G))
defined by I
mA(x)(y) = A(x, y) for all A ∈ L(
m+nE; G), x ∈ E
m, y ∈ E
n, is an isometric isomorphism. Likewise the mapping
T
t: A ∈ L(
mE; L(
nF ; G)) → A
t∈ L(
nF ; L(
mE; G))
defined by A
t(y)(x) = A(x)(y) for all A ∈ L(
mE; L(
nF ; G)), x ∈ E
m, y ∈ F
n, is an isometric isomorphism. Let L
s(
mE; G) denote the subspace of all A ∈ L(
mE; G) which are symmetric. Let P(
mE; G) denote the vector space of all continuous m − homogeneous polynomials from E into G. If G = K, we write
iThis work had financial support from CNPq(Brazil).
Note Mat. 29 (2009), n. 2, 55-76 ISSN 1123-2536, e-ISSN 1590-0932 DOI 10.1285/i15900932v29n2p55
http://siba-ese.unisalento.it, © 2009 Università del Salento
__________________________________________________________________
P(
mE; K) = P(
mE). For each A ∈ L(
mE; G) let ' A ∈ P(
mE; G) be defined by ' A(x) = Ax
mfor every x ∈ E. The mapping A → ' A induces a topological isomorphism between L
s(
mE; G) and P(
mE; G).
1 Definition. A mapping A ∈ L(
mE; G) is said to be of finite type if there exist c
1, . . . , c
n∈ G, ϕ
1i, . . . , ϕ
mi∈ E
, 1 ≤ i ≤ n such that A can be written in the form: A(x
1, . . . , x
m) =
ni=1
ϕ
1i(x
1) ···ϕ
mi(x
m)c
ifor all (x
1, . . . , x
m) ∈ E
m. 2 Definition. A polynomial P ∈ P(
mE; G) is said to be of finite type, if there exist c
1, . . . , c
n∈ G, ϕ
1, . . . , ϕ
n∈ E
such that P can be written of the form: P (x) =
ni=1
ϕ
i(x)
mc
ifor all x ∈ E.
3 Definition. A mapping A ∈ L(
mE; G) is said to be nuclear, if there exist sequence (ϕ
ji)
i∈Nin E
, 1 ≤ j ≤ m and (c
i)
i∈Nin G with
∞i=1
ϕ
1i· · ·
ϕ
mic
i< ∞ such that A(x
1, . . . , x
m) =
∞i=1
ϕ
1i(x
1) · · · ϕ
mi(x
m)c
ifor all (x
1, . . . , x
m) ∈ E
m.
4 Definition. A polynomial P ∈ P(
mE; G) is said to be nuclear, if there exist sequences (ϕ
i)
i∈Nin E
, and (c
i)
i∈Nin G with
∞i=1
ϕ
imc
i< ∞ such that P (x) =
∞i=1
ϕ
i(x)
mc
ifor all x ∈ E.
Let L
f(
mE; G) denote the space of all A ∈ L(
mE; G) which are of finite type. Let P
f(
mE; G) denote the space of all P ∈ P(
mE; G) which are of finite type. Let L
N(
mE; G) denote the space of all A ∈ L(
mE; G) which are nuclear endowed with the nuclear norm A
N= inf
∞i=1
ϕ
1i· · · ϕ
mic
i, where the infimum is taken over all sequences (ϕ
ji)
i∈Nand (c
i)
i∈Nwhich satisfy the definition. When G = K, we write L
Θ(
mE, K) = L
Θ(
mE), where Θ = f or N . Let P
N(
mE; G) denote the space of all P ∈ P(
mE; G) which are nuclear, endowed with the nuclear norm P
N= inf
∞i=1
ϕ
imc
i, where the infimum
is taken over all sequences (ϕ
i)
i∈Nand (c
i)
i∈Nwhich satisfy the definition. When
G = K, we write P
Θ(
mE, K) = P
Θ(
mE), where Θ = f or N . Let us recall that
A ∈ L(
mE; G) is a compact (resp. weakly compact) mapping if A(B
Em) is
relatively compact in G (resp. for the weak topology), where B
Emdenotes the
closed unit ball of E
m. Let L
K(
mE; G) ( resp. L
W K(
mE; G) ) denote the space
of all A ∈ L(
mE; G) which are compact (resp. weakly compact). Recall that
P ∈ P(
mE; G) is a compact (resp. weakly compact) polynomial if P (B
E) is
relatively compact in G (resp. for the weak topology), where B
Edenotes the
closed unit ball of E. Let P
K(
mE; G) (resp. P
W K(
mE; G)) denote the space
of all compact (resp. weakly compact) homogeneous polynomials from E into
G. We observe that in the case where G = K, all the continuous homogeneous
polynomials are compact (resp. weakly compact). For background information
on multilinear mappings and homogeneous polynomials we refer to the books
[7] and [12].
1 Isomorphisms between spaces of multilinear map- pings or homogeneous polynomials
In this work, we use the Nicodemi sequences defined in [8] to prove all the theorems. We recall the definition of the Nicodemi sequences.
5 Definition. [8] Given a continuous linear operator R
1: L(E; G) −→
L(F ; G), let R
m: L(
mE; G) −→ L(
mF ; G) be inductively defined by R
m+1A = I
m−1[R
m◦ (R
1◦ I
m(A))
t]
tfor all A ∈ L(
m+1E; G) and m ∈ N.
6 Example. [8, Example 1.2] Let T
1: E
→ E
be the natural embedding and let T
m: L(
mE) → L(
mE
) be the Nicodemi sequence of operators beginning with T
1. This sequence is precisely the sequence of operators constructed by Aron and Berner in [1, Proposition 2.1].
7 Definition. [8, Proposition 4.1] Given R
1∈ L(E
; F
), let R
1∈ L(L(E; G
), L(F ; G
)) be defined by R
1A(y)(z) = R
1(δ
z◦ A)(y) for all A ∈ L(E; G
), y ∈ F and z ∈ G, where δ
z: G
→ K is defined by δ
z(z
) = z
(z) for all z
∈ G
. If (R
m) and (( R
m) are the corresponding Nicodemi sequences, then R (
mA(y)(z) = R
m(δ
z◦ A)(y) for all A ∈ L(
mE; G
), y ∈ F
mand z ∈ G.
8 Example. Let T
m: L(
mE) −→ L(
mE
) be the Nicodemi sequence of operators beginning with the natural embedding T
1: E
→ E
, and let T
m: L(
mE; G
) −→ L(
mE
; G
) be the corresponding sequence for vector- valued multilinear mappings (See [8, Proposition 4.1]). We observe that as G
is a C
1−space, the sequence T
m: L(
mE; G
) −→ L(
mE
; G
) ⊂ L(
mE
; G
) coincides with the sequence of operators constructed by Aron and Berner in [1, Proposition 2.1].
The next theorem shows the relationship between an arbitrary Nicodemi sequence of scalar-valued mappings and the Nicodemi sequence beginning with the natural embedding E
→ E
.
9 Theorem. Let R
m: L(
mE) −→ L(
mF ) be a Nicodemi sequence, let T
m: L(
mE) −→ L(
mE
) be the Nicodemi sequence beginning with the natural embedding T
1= J
E: E
→ E
, and let J
F: F → F
be the natural embedding.
Then
R
mA(y
1, . . . , y
m) = T
mA(R
1(J
Fy
1), . . . , R
1(J
Fy
m)) for all A ∈ L(
mE), and y
1, . . . , y
m∈ F , where R
1is the transpose of R
1.
Proof. We will prove this theorem by induction on m. If A ∈ L(E) = E
, we have
R
1A(y) = J
Fy, R
1A = R
1(J
Fy), A = T
1A, R
1(J
Fy) = T
1A(R
1(J
Fy))
for all y ∈ F. Now let us assume that the identity in Theorem 9 is true for m−
linear forms. Let A ∈ L(
m+1E). We will prove that
T
m+1A(z
1, . . . , z
m, R
1(J
Fy)) = T
m[(R
1◦ I
mA)
t(y)](z
1, . . . , z
m) (1) for all z
1, . . . , z
m∈ E
and y ∈ F . From the definition of Nicodemi sequences, it follows that
T
m+1A(z
1, . . . , z
m, R
1(J
Fy)) = T
m[(T
1◦ I
mA)
t(R
1(J
Fy))](z
1, . . . , z
m). (2) Therefore, to get (1) comparing with (2), it is enough to prove that
(R
1◦ I
mA)
t(y) = (T
1◦ I
mA)
t(R
1(J
Fy)).
In fact,
(T
1◦ I
mA)
t(R
1(J
Fy))(x
1, . . . , x
m) = T
1[I
mA(x
1, . . . , x
m)](R
1(J
Fy))
= R
1(J
Fy), I
mA(x
1, . . . , x
m)
= J
Fy, R
1[I
mA(x
1, . . . , x
m)]
= R
1[I
mA(x
1, . . . , x
m)], y
= R
1[I
mA(x
1, . . . , x
m)](y)
= (R
1◦ I
mA)
t(y)(x
1, . . . , x
m)
for all x
1, . . . , x
m∈ E. Thus by the induction hypothesis and (1) it follows that for all y
1, . . . , y
m+1∈ F
R
m+1A(y
1, . . . , y
m+1) =R
m[(R
1◦ I
mA)
t(y
m+1)](y
1, . . . , y
m)
=T
m[(R
1◦ I
mA)
t(y
m+1)](R
1(J
Fy
1), . . . , R
1(J
Fy
m))
=T
m+1A(R
1(J
Fy
1), . . . , R
1(J
Fy
m), R
1(J
Fy
m+1)).
QED
We next extend Theorem 9 to the case of a Nicodemi sequence of vector- valued mappings. Let us recall that each Nicodemi sequence (R
m) for scalar- valued mappings yields a Nicodemi sequence (( R
m) for vector - valued mappings.
10 Theorem. Let R
m: L(
mE) −→ L(
mF ) be a Nicodemi sequence, and let T
m: L(
mE) −→ L(
mE
) be the Nicodemi sequence beginning with the natural embedding T
1= J
E: E
→ E
. Let
( R
m: L(
mE; G
) → L(
mF ; G
)
and T
m: L(
mE; G
) → L(
mE
; G
)
be the corresponding Nicodemi sequences for vector-valued multilinear mappings.
Then ( R
mA(y
1, . . . , y
m) = T
mA(R
1(J
Fy
1), . . . , R
1(J
Fy
m))
for all A ∈ L(
mE; G
), and y
1, . . . , y
m∈ F , where R
1is the transpose of R
1. Proof. We will prove this theorem by induction on m. If A ∈ L(E; G
), it follows that
R
1A(y)(z) =R
1(δ
z◦ A)(y)
= J
Fy, R
1(δ
z◦ A)
= R
1(J
Fy), (δ
z◦ A)
= T
1(δ
z◦ A), R
1(J
Fy)
= T
1A(R
1(J
Fy))(z)
for all y ∈ F and z ∈ G. Now let us assume that the identity is true for m−
linear forms. Let A ∈ L(
m+1E; G
). We prove initially that
T
m+1A(z
1, . . . , z
m, R
1(J
Fy)) = T
m[( R
1◦ I
mA)
t(y)](z
1, . . . , z
m) (3) for all z
1, . . . , z
m∈ E
and y ∈ F . From the definition of Nicodemi sequences, it follows that
T
m+1A(z
1, . . . , z
m, R
1(J
Fy)) = T
m[( T
1◦ I
mA)
t(R
1(J
Fy))](z
1, . . . , z
m). (4)
Therefore, to get (3) comparing with (4), it is enough to prove that ( R
1◦ I
mA)
t(y) = ( T
1◦ I
mA)
t(R
1(J
Fy)).
In fact,
( T
1◦ I
mA)
t(R
1(J
Fy))(x
1, . . . , x
m)(z) = T
1[I
mA(x
1, . . . , x
m)](R
1(J
Fy))(z)
=T
1[δ
z◦ I
mA(x
1, . . . , x
m)](R
1(J
Fy))
= R
1(J
Fy), δ
z◦ I
mA(x
1, . . . , x
m)
= J
Fy, R
1[δ
z◦ I
mA(x
1, . . . , x
m)]
= R
1[δ
z◦ I
mA(x
1, . . . , x
m)], y
=R
1[δ
z◦ I
mA(x
1, . . . , x
m)](y)
=( R
1◦ I
mA)
t(y)(x
1, . . . , x
m)(z)
for all x
1, . . . , x
m∈ E and z ∈ G. Thus, from the induction hypothesis and (3), it follows that for all y
1, . . . , y
m+1∈ F
R
m+1A(y
1, . . . , y
m+1) =( R
m[( R
1◦ I
mA)
t(y
m+1)](y
1, . . . , y
m)
= T
m[( R
1◦ I
mA)
t(y
m+1)](R
1(J
Fy
1), . . . , R
1(J
Fy
m))
= T
m+1A(R
1(J
Fy
1), . . . , R
1(J
Fy
m), R
1(J
Fy
m+1)).
QED
Recall that a Banach space E is said to be Arens - regular if all linear operator E −→ E
are weakly compact, and symmetrically Arens - regular if this is so for all symmetric linear operators. An operator T : E → E
is said to be symmetric if T x(y) = T y(x) for all x, y ∈ E(see [3] and [9]). Let us recall that if E is symmetrically Arens - regular, then T
mA ∈ L
s(
mE
) for all A ∈ L
s(
mE), where (T
m) is the Nicodemi sequence beginning with the natural embedding T
1= J
E: E
→ E
(see [2, Theorem 8.3]). Now, we are ready to study the theorems of preservation of symmetric multilinear mappings.
11 Theorem. Let R
m: L(
mE) −→ L(
mF ) be a Nicodemi sequence and let T
m: L(
mE) −→ L(
mE
) be the Nicodemi sequence beginning with the natural embedding T
1= J
E: E
→ E
. If T
mA is symmetric, then R
mA is also symmetric. In particular, if E is symmetrically Arens - regular, then R
mA ∈ L
s(
mF ) for all A ∈ L
s(
mE).
Proof. By Theorem 9 we have that
R
mA(y
1, . . . , y
m) = T
mA(R
1(J
Fy
1), . . . , R
1(J
Fy
m)).
Thus, if T
mA is symmetric, then R
mA is also symmetric. Now if E is sym- metrically Arens - regular, we have that the Aron - Berner extension T
mA is symmetric for all A ∈ L
s(
mE)(See [2, Proposition 8.3]). We conclude that
R
mA ∈ L
s(
mF ) for all A ∈ L
s(
mE).
QED12 Theorem. Let R
m: L(
mE) −→ L(
mF ) be a Nicodemi sequence and let ( R
m: L(
mE; G
) −→ L(
mF ; G
) be the corresponding Nicodemi sequence for vector-valued multilinear mappings. If E is symmetrically Arens - regular, then ( R
mA ∈ L
s(
mF ; G
) for all A ∈ L
s(
mE; G
).
Proof. By [8, Proposition 4.1], we have that
( R
mA(y)(z) = R
m(δ
z◦ A)(y) (5)
for all A ∈ L(
mE, G
), y ∈ F
mand z ∈ G. The identity (5) and Theorem 11
imply that if A is symmetric, then ( R
mA is also symmetric.
QEDIn Theorem 13 we will denote J
F(y) by y for all y ∈ F and J
E(x) by x for all x ∈ E.
13 Theorem. Let R
m: L(
mE) −→ L(
mF ) be a Nicodemi sequence, let T
m: L(
mE) −→ L(
mE
) be the Nicodemi sequence beginning with the natural embedding T
1= J
E: E
→ E
, and let Q
m: L(
mF ) −→ L(
mF
) be the Nicodemi sequence beginning with the natural embedding Q
1= J
F: F
→ F
. If T
mA is symmetric, then
Q
m◦ R
mA(w
1, . . . , w
m) = T
mA(R
1w
1, . . . , R
1w
m) for all A ∈ L(
mE), w
j∈ F
, j = 1, . . . , m, and m ∈ N.
Proof. We will prove by induction on k ∈ N that Q
m◦ R
mA(w
1, . . . , w
k, y
k+1, . . . , y
m) =
T
mA(R
1(w
1), . . . , R
1(w
k), R
1(y
k+1), . . . , R
1(y
m)) for all y
j∈ F and w
j∈ F
. Recall that (i) Q
mB and T
mA are weak
continuous in its first variable for all B ∈ L(
mF ) and A ∈ L(
mE) by [8, Proposition 5.1]; (ii) elements of F
and of J
F(F ) commute in variables of Q
mB by [8, Lemma 3.4];
(iii) R
1is σ(F
, F
)−σ(E
, E
) continuous; (iv) by [8, Proposition 2.1] , we have that T
mA(x
1, . . . , x
m) = A(x
1, . . . , x
m) for all A ∈ L(
mE) and x
1, . . . , x
m∈ E;
and Q
mB(y
1, . . . , y
m) = B(y
1, . . . , y
m) for all B ∈ L(
mF ) and y
1, . . . , y
m∈ F . If k = 1, by Goldstine´ s theorem, there is a net (y
α) ⊂ F such that y
α−→ w
1for the topology σ(F
, F
). Then by Theorem 9
Q
m◦ R
mA(w
1, y
2. . . , y
m) =Q
m(R
mA)(w
1, y
2. . . , y
m)
= lim
α
Q
m(R
mA)(y
α, y
2. . . , y
m)
= lim
α
(R
mA)(y
α, y
2. . . , y
m)
= lim
α
T
mA(R
1y
α, R
1y
2, . . . , R
1y
m)
=T
mA(R
1w
1, R
1y
2, . . . , R
1y
m).
Now assuming that the identity holds for k, we will prove that the identity holds for k + 1. By the induction hypothesis and T
mA being symmetric, it follows that Q
m◦ R
mA(w
1, . . . , w
k+1,y
k+2, . . . , y
m)
=Q
m(R
mA)(w
1, . . . , w
k+1, y
k+2, . . . , y
m)
= lim
α
Q
m(R
mA)(y
α, w
2, . . . , w
k+1, y
k+2, . . . , y
m)
= lim
α
Q
m(R
mA)(w
2, . . . , w
k+1, y
α, y
k+2, . . . , y
m)
= lim
α
T
mA(R
1w
2, . . . , R
1w
k+1, R
1y
α, R
1y
k+2, . . . , R
1y
m)
= lim
α
T
mA(R
1y
α, R
1w
2, . . . , R
1w
k+1, R
1y
k+2, . . . , R
1y
m)
=T
mA(R
1w
1, . . . , R
1w
k+1, R
1y
k+2, . . . , R
1y
m).
QED
Next we will prove that each isomorphism between E
and F
induces an isomorphism between L(
mE; G
) and L(
mF ; G
) for all m ∈ N. If E and F are symmetrically Arens - regular, then each isomorphism between E
and F
induces an isomorphism between P(
mE; G
) and P(
mF ; G
) for all m ∈ N.
Given a continuous linear operator
R
m: L(
mE; G) −→ L(
mF ; G), we define
U
m: A ∈ L(
nD; L(
mE; G)) −→ R
m◦ A ∈ L(
nD; L(
mF ; G)).
We observe that if R
mis an isomorphism then U
mis also an isomorphism, whose inverse
U
m−1: L(
nD; L(
mF ; G)) −→ L(
nD; L(
mE; G))
is defined by U
m−1(B) = R
−1m◦ B for all B ∈ L(
nD; L(
mF ; G)) where R
−1mis the inverse of R
m. Thus, with the previous notations, it is possible to rewrite the definition of the Nicodemi operators in the following way:
14 Lemma. Given an operator
R
m: L(
mE; G) −→ L(
mF ; G), the operator
R
m+1: L(
m+1E; G) −→ L(
m+1F ; G) is given by
R
m+1(A) = I
m−1[R
m◦ (R
1◦ I
m(A))
t]
t= I
m−1◦ T
t◦ U
m◦ T
t◦ U
1◦ I
m(A) for all A ∈ L(
m+1E; G).
The following theorem was obtained in [4]. But we will need our proof in the proof of the subsequent theorem.
15 Theorem. If E
and F
are isomorphic, then L(
mE) and L(
mF ) are
isomorphic for all m ∈ N.
Proof. Since E
and F
are isomorphic, there exists an isomorphism R
1: E
−→ F
. Let R
m: L(
mE) −→ L(
mF ) be the Nicodemi sequence beginning with R
1. We will prove by induction on m ∈ N that R
mis an isomorphism between L(
mE) and L(
mF ) for all m ∈ N. By hypothesis R
1: E
−→ F
is an isomorphism. Assuming that R
1and R
mare isomorphisms, we show that R
m+1is also an isomorphism. In fact, by Lemma 1 it is possible to rewrite
R
m+1= I
m−1◦ T
t◦ U
m◦ T
t◦ U
1◦ I
m.
Since R
1and R
mare isomorphisms, we have that U
1and U
mare also isomor- phisms. Thus R
m+1is an isomorphism between L(
m+1E) and L(
m+1F ) being a composite of isomorphisms. Therefore L(
mE) and L(
mF ) are isomorphic for
all m ∈ N.
QED16 Theorem. If E
and F
are isomorphic, then L(
mE; G
) and L(
mF ; G
) are isomorphic for all m ∈ N.
Proof. Since E
and F
are isomorphic, there exists an isomorphism R
1: E
−→ F
. Let R
m: L(
mE) −→ L(
mF ) be the Nicodemi sequence beginning with R
1. Let
R (
m: L(
mE, G
) −→ L(
mF, G
)
be the corresponding Nicodemi sequence for vector-valued multilinear mappings.
We observe that
(δ
z◦ ( R
mA) = R
m(δ
z◦ A) (6)
for all z ∈ G and A ∈ L(
mE; G
). In fact, by [8, Proposition 4.1],
(δ
z◦ ( R
mA)(y) = ( R
mA(y)(z) = R
m(δ
z◦ A)(y)
for all y ∈ F
m. It follows from the proof of Theorem 15 that R
mis an isomor- phism for all m ∈ N. Let S
m: L(
mF ) −→ L(
mE) denote the inverse of R
mfor all m ∈ N. We define
S
m: L(
mF ; G
) −→ L(
mE; G
)
by S
mB(x)(z) = S
m(δ
z◦B)(x) for all B ∈ L(
mF ; G
), x ∈ E
m, z ∈ G and m ∈ N.
Thus S
mis linear and continuous. We will show that ( R
mis an isomorphism
between L(
mE; G
) and L(
mF ; G
) for all m ∈ N. In fact, we have that by (6) S
m◦ ( R
mA(x)(z) = S
m(( R
mA)(x)(z)
=S
m(δ
z◦ ( R
mA)(x)
=S
m(R
m(δ
z◦ A))(x)
=[S
m◦ R
m(δ
z◦ A)](x)
=(δ
z◦ A)(x)
=A(x)(z)
for all A ∈ L(
mE, G
), x ∈ E
mand z ∈ G. In a similar way, we can get that (( R
m◦ S
m)B = B for all B ∈ L(
mF ; G
).
QED17 Theorem. If E and F are symmetrically Arens - regular, and E
and F
are isomorphic, then L
s(
mE) and L
s(
mF ) are isomorphic for all m ∈ N.
Proof. We write J
E(x) = x for all x ∈ E. Recall that by [8, Proposition 2.1] T
mA(x
1, . . . , x
m) = A(x
1, . . . , x
m) for all A ∈ L(
mE) and x
1, . . . , x
m∈ E.
Since E
and F
are isomorphic, there exists an isomorphism R
1: E
−→ F
. Let R
m: L(
mE) −→ L(
mF ) be the Nicodemi sequence beginning with R
1. Let S
1= R
−11: F
−→ E
be the inverse of R
1and let S
m: L(
mF ) −→ L(
mE) be the Nicodemi sequence beginning with S
1. Since F is symmetrically Arens - regular, we have by Theorem 11 that S
m(L
s(
mF )) ⊂ L
s(
mE). By Theorem 9 we have that S
mB(x
1, . . . , x
m) = Q
mB(S
1x
1, . . . , S
1x
m) for all B ∈ L(
mF ), and x
1, . . . , x
m∈ E, where S
1is the transpose of S
1. In particular, we have that S
m(R
mA)(x
1, . . . , x
m) = Q
m(R
mA)(S
1x
1, . . . , S
1x
m) (7) for all A ∈ L(
mE). On the other hand, since E is symmetrically Arens - regular, it follows from Theorem 11 that R
m(L
s(
mE)) ⊂ L
s(
mF ). Moreover, by Theorem 13 we have that
Q
m(R
mA)(S
1x
1, . . . , S
1x
m) = T
mA(R
1(S
1x
1), . . . , R
1(S
1x
m)), (8) for all A ∈ L
s(
mE). Therefore we conclude by (7) and (8) that
S
m(R
mA)(x
1, . . . , x
m) =Q
m(R
mA)(S
1x
1, . . . , S
1x
m)
=T
mA(R
1(S
1x
1), . . . , R
1(S
1x
m))
=T
mA(x
1, . . . , x
m)
=A(x
1, . . . , x
m), for all A ∈ L
s(
mE), that is
(S
m◦ R
m)A = A (9)
for all A ∈ L
s(
mE). In an analogous way, we can prove that (R
m◦ S
m)B = B
for all B ∈ L
s(
mF ).
QEDThe following Corollary 18 was proven by Lassalle - Zalduendo in [11] and by F. Cabello S´ anchez, J. Castillo and R. Garc´ıa in [4] by a different method.
18 Corollary. If E and F are symmetrically Arens - regular, and E
and F
are isomorphic, then P(
mE) and P(
mF ) are isomorphic for all m ∈ N.
19 Theorem. If E and F are symmetrically Arens - regular, and E
and F
are isomorphic, then L
s(
mE; G
) and L
s(
mF ; G
) are isomorphic for all m ∈ N.
Proof. Since E
and F
are isomorphic, there exists an isomorphism R
1: E
−→ F
. Let R
m: L(
mE) −→ L(
mF ) be the Nicodemi sequence beginning with R
1. Let S
1= R
−11: F
−→ E
be the inverse of R
1, and let S
m: L(
mF ) −→
L(
mE) be the Nicodemi sequence beginning with S
1. Let ( R
m: L(
mE; G
) −→
L(
mF ; G
) and S
m: L(
mF ; G
) −→ L(
mE; G
) be the corresponding Nicodemi sequence for vector-valued multilinear mappings. By Theorem 12 we have that R (
m(L
s(
mE; G
)) ⊂ L
s(
mF ; G
). (10)
and
S
m(L
s(
mF ; G
)) ⊂ L
s(
mE; G
). (11) It follows from the proof of Theorem 16 that ( R
mis an isomorphism between L(
mE; G
) and L(
mF ; G
) such that ( R
m−1
= S
m. By (10) and (11), we have that R (
m|
Ls(mE;G)is an isomorphism between L
s(
mE; G
) and L
s(
mF ; G
).
QEDThe following Corollary 20 was proven for Carando - Lassalle in [5] by a different method.
20 Corollary. If E and F are symmetrically Arens - regular, and E
and F
are isomorphic, then P(
mE; G
) and P(
mF ; G
) are isomorphic for all m ∈ N.
2 Isomorphisms between spaces of nuclear multilin- ear mappings or homogeneous polynomials.
We will use the following notation:
(ϕ
1⊗ ϕ
2⊗ c)(x
1, x
2) = ϕ
1(x
1)ϕ
2(x
2)c
for all ϕ
1, ϕ
2∈ E
, c ∈ F and x
1, x
2∈ E. The next lemma comes essentially
from Aron and Berner [1].
21 Lemma. [1, Proposition 2.2] Let T
m: L(
mE) −→ L(
mE
) be the Nicodemi sequence beginning with the natural embedding T
1= J
E: E
→ E
. Let A ∈ L
f(
mE) and let c
1, . . . , c
n∈ K, ϕ
1i, . . . , ϕ
mi∈ E
, 1 ≤ i ≤ n such that
A(x
1, . . . , x
m) =
n i=1ϕ
1i(x
1) · · · ϕ
mi(x
m)c
ifor all (x
1, . . . , x
m) ∈ E
m, then
T
m(A)(x
1, . . . , x
m) =
n i=1(T
1ϕ
1i)(x
1) · · · (T
1ϕ
mi)(x
m)c
ifor all (x
1, . . . , x
m) ∈ (E
)
m. In particular T
m(A) ∈ L
f(
mE
).
22 Lemma. Let T
m: L(
mE) −→ L(
mE
) be the Nicodemi sequence begin- ning with the natural embedding T
1= J
E: E
→ E
. Let A ∈ L
N(
mE) and let (ϕ
ji)
i∈Nin E
, 1 ≤ j ≤ m and (c
i)
i∈Nin K with
∞ i=1ϕ
1i· · · ϕ
mic
i< ∞
such that
A(x
1, . . . , x
m) =
∞ i=1ϕ
1i(x
1) · · · ϕ
mi(x
m)c
ifor all (x
1, . . . , x
m) ∈ E
m. Then T
m(A) has the following form:
T
m(A)(x
1, . . . , x
m) =
∞ i=1(T
1ϕ
1i)(x
1) · · · (T
1ϕ
mi)(x
m)c
ifor all (x
1, . . . , x
m) ∈ (E
)
m. In particular T
m(A) ∈ L
N(
mE
).
Proof. Let A
n=
ni=1
ϕ
1i⊗···⊗ϕ
mi⊗c
ifor all n ∈ N. Then A
n∈ L
f(
mE) and A
n→ A for the nuclear norm, that is
A
n− A
N−→ 0. (12)
By Lemma 21 T
mA
nhas the following form: T
mA
n=
ni=1
T
1ϕ
1i⊗···⊗T
1ϕ
mi⊗c
ifor all n ∈ N. It follows that T
mA
n−→
∞i=1
T
1ϕ
1i⊗ · · · ⊗ T
1ϕ
mi⊗ c
ifor the nuclear norm. In particular
T
mA
n−→
∞ i=1T
1ϕ
1i⊗ · · · ⊗ T
1ϕ
mi⊗ c
i(13)
pointwise. On the other hand, we have that
T
mA
n→ T
mA (14)
pointwise. Then by (12)
T
mA(x
1, . . . , x
m) − T
mA
n(x
1, . . . , x
m)
=(T
mA − T
mA
n)(x
1, . . . , x
m)
= T
m(A − A
n)(x
1, . . . , x
m)
≤T
mA − A
n(x
1, . . . , x
m)
≤T
mA − A
nN(x
1, . . . , x
m) −→ 0 for all x
1, . . . , x
m∈ E
. By (13) and (14), we have that
T
m(A)(x
1, . . . , x
m) =
∞ i=1(T
1ϕ
1i)(x
1) · · · (T
1ϕ
mi)(x
m)c
ifor all x
1, . . . , x
m∈ E
and therefore T
m(A) ∈ L
N(
mE
).
QEDNext we will see that the operators from the Nicodemi sequence preserve multilinear mappings of finite type and nuclear multilinear mappings. We con- sider first the case of scalar-valued multilinear mappings.
23 Theorem. Let R
m: L(
mE) −→ L(
mF ) be a Nicodemi sequence. Then R
mA ∈ L
Θ(
mF ) for all A ∈ L
Θ(
mE), where Θ = f or N .
Proof. We will write the proof in detail in the case Θ = N . In the case Θ = f , the proof is similar. Given A ∈ L
N(
mE), there exist sequences (ϕ
ji)
i∈Nin E
, 1 ≤ j ≤ m, and (c
i)
i∈Nin K, with
∞i=1
ϕ
1i· · · ϕ
mi|c
i| < ∞ such that A(x
1, . . . , x
m) =
∞i=1
ϕ
1i(x
1) · · · ϕ
mi(x
m)c
ifor all (x
1, . . . , x
m) ∈ E
m. Let T
m: L(
mE) −→ L(
mE
) be the Nicodemi sequence beginning with the natural embedding T
1= J
E: E
→ E
. By Theorem 9 we have that
R
mA(y
1, . . . , y
m) = T
mA(R
1(J
Fy
1), . . . , R
1(J
Fy
m))
for all A ∈ L(
mE), and y
1, . . . , y
m∈ F , where R
1is the transpose of R
1and by Lemma 22 we have that
T
m(A)(x
1, . . . , x
m) =
∞ i=1(T
1ϕ
1i)(x
1) · · · (T
1ϕ
mi)(x
m)c
ifor all (x
1, . . . , x
m) ∈ (E
)
m. Therefore
R
mA(y
1, . . . , y
m) =T
mA(R
1(J
Fy
1), . . . , R
1(J
Fy
m))
=
∞ i=1(T
1ϕ
1i)(R
1J
Fy
1) · · · (T
1ϕ
mi)(R
1J
Fy
m)c
i=
∞ i=1R
1ϕ
1i(y
1) · · · R
1ϕ
mi(y
m)c
ifor all y
1, . . . , y
m∈ F because < T
1ϕ, R
1(J
Fy) >=< J
Eϕ, R
1(J
Fy) >=
< R
1(J
Fy), ϕ >=< J
Fy, R
1ϕ >=< R
1ϕ, y >, for all ϕ ∈ E
and all y ∈ F . Therefore
R
mA =
∞ i=1R
1ϕ
1i⊗ · · · ⊗ R
1ϕ
mi⊗ c
i(15)
and then R
mA ∈ L
N(
mF ).
QEDThe next theorem considers the case of vector-valued multilinear mappings.
24 Theorem. Let R
m: L(
mE) −→ L(
mF ) be a Nicodemi sequence and let ( R
m: L(
mE; G
) −→ L(
mF ; G
) be the corresponding sequence for vector-valued multilinear mappings. Then ( R
mA ∈ L
Θ(
mF ; G
) for all A ∈ L
Θ(
mE; G
) where Θ = f or N .
Proof. We will only write the proof in the case Θ = N . In the case Θ = f , the proof is similar. Given A ∈ L
N(
mE; G
) there exist sequences (ϕ
ji)
i∈Nin E
, 1 ≤ j ≤ m, and (c
i)
i∈Nin G
, with
∞i=1
ϕ
1i· · · ϕ
mic
i< ∞ such that A =
∞i=1
ϕ
1i⊗ · · · ⊗ ϕ
mi⊗ c
i. Then δ
z◦ A =
∞i=1
ϕ
1i⊗ · · · ⊗ ϕ
mi⊗ c
i(z) for all z ∈ G, and clearly δ
z◦ A ∈ L
N(
mE). We have by (15) that
R
m(δ
z◦ A) =
∞ i=1R
1ϕ
1i⊗ · · · ⊗ R
1ϕ
mi⊗ c
i(z)
for all z ∈ G. Then
R (
mA(y
1, . . . , y
m)(z) =R
m(δ
z◦ A)(y
1, . . . , y
m)
=
∞ i=1R
1ϕ
1i(y
1) · · · R
1ϕ
mi(y
m)c
i(z)
for all z ∈ G and y
1, . . . , y
m∈ F. We conclude that ( R
mA =
∞ i=1R
1ϕ
1i⊗ · · · ⊗ R
1ϕ
mi⊗ c
i(16)
and then ( R
mA ∈ L
N(
mF ; G
).
QED25 Definition. Given a Nicodemi sequence R
m: L(
mE; G) −→ L(
mF ; G), we define
R '
m: P(
mE; G) −→ P(
mF ; G)
by ' R
mA = ' R
mA por every symmetric A ∈ L(
mE; G).
26 Lemma. Let T
m: L(
mE) −→ L(
mE
) be the Nicodemi sequence begin- ning with the natural embedding T
1= J
E: E
→ E
. Let P ∈ P
N(
mE) and let c
1, . . . , c
n∈ K, ϕ
1, . . . , ϕ
m∈ E
such that
P =
∞ i=1ϕ
mi⊗ c
i.
Then
T '
m(P ) =
∞ i=1(T
1ϕ
i)
m⊗ c
i.
In particular ' T
m(P ) ∈ P
N(
mE
).
The proof of Lemma 26 is similar to the proof of Lemma 22 and is omit- ted. Next we will see that the operators from the Nicodemi sequence preserve homogeneous polynomials of finite type and nuclear homogeneous polynomials.
We consider first the case of scalar-valued homogeneous polynomials.
27 Theorem. Let R
m: L(
mE) −→ L(
mF ) be a Nicodemi sequence. Then R '
mP ∈ P
Θ(
mF ) for all P ∈ P
Θ(
mE) where Θ = f or N .
Proof. We will only write the proof in the case Θ = N . In the case Θ = f , the proof is similar. Given P ∈ P
N(
mE), there exist (c
i)
i∈N∈ K, (ϕ
i)
i∈N∈ E
such that P can be written in the form: P (x) =
∞i=1
ϕ
i(x)
mc
ifor all x ∈ E. Let T
m: L(
mE) −→ L(
mE
) be the Nicodemi sequence beginning with the natural embedding T
1= J
E: E
→ E
. By Theorem 9 we have that
R
mA(y
1, . . . , y
m) = T
mA(R
1(J
Fy
1), . . . , R
1(J
Fy
m))
for all A ∈ L(
mE), and y
1, . . . , y
m∈ F , where R
1is the transpose of R
1and by Lemma 26 we have that ' T
m(P ) has the following form:
T '
m(P )(x
) =
∞ i=1(T
1ϕ
i)(x
)
mc
ifor all x
∈ E
. Therefore, A being the m-linear mapping associated with P , it
follows that
R '
mP (y) =R
mA(y, . . . , y ) *+ ,
m−times
)
=T
mA(R
1(J
Fy), . . . , R
1(J
Fy)
) *+ ,
m−times
)
= ' T
mP (R
1(J
Fy))
=
∞ i=1(T
1ϕ
i)(R
1(J
Fy))
mc
i=
∞ i=1(R
1ϕ
iy))
mc
ifor all y ∈ F . Therefore,
R '
mP =
∞ i=1(R
1ϕ
i)
m⊗ c
i(17)
and then ' R
mP ∈ P
N(
mF ).
QEDThe next theorem consider the case of vector-valued homogeneous polyno- mials.
28 Theorem. Let R
m: L(
mE) −→ L(
mF ) be a Nicodemi sequence, let ( R
m: L(
mE; G
) −→ L(
mF ; G
) be the corresponding Nicodemi sequence for vector- valued multilinear mappings. Then ' R
mP ∈ P
Θ(
mF ; G
) for all P ∈ P
Θ(
mE; G
) where Θ = f or N .
Proof. We will only write the proof in the case Θ = N . In the case Θ = f , the proof is similar. Given ' A ∈ P
N(
mE; G
), there exist sequences (ϕ
i)
i∈Nin E
, and (c
i)
i∈Nin G
with
∞i=1
ϕ
imc
i< ∞ such that ' A(x) =
∞i=1
ϕ
i(x)
mc
ifor all x ∈ E, where A ∈ L
s(
mE; G
). We observe that if B =
∞i=1
ϕ )
i⊗ · · · ⊗ ϕ *+ ,
i m−times⊗c
i, then B ∈ L
N(
mE; G
) ∩ L
s(
mE; G
) and ' B = ' A. Then
A = B from the injectivity of the canonical isomorphism L
s(
mE; G
) −→
P(
mE; G
). Thus, we get that A ∈ L
N(
mE; G
) and by (16) ( R
mA =
∞ i=1R
1ϕ
i⊗ · · · ⊗ R
1ϕ
i) *+ ,
m−times
⊗c
i.
Since ' R
mA = ' ( R
mA, it follows that ' R
mP =
n i=1(R
1ϕ
i)
m⊗ c
i(18)
and then ' R
mP ∈ P
N(
mF ; G
).
QEDWe remark that if E is a closed subspace of F , then Aron and Berner [1, Theorem 2.1] proved that the restriction mapping
P
N(
mF ; G) −→ P
N(
mE; G)
is surjective for each m ∈ N, but even in the case of the natural embedding J
E: E → E
, they did not study the problem of existence of a linear extension operator
T
m: P
N(
mE) −→ P
N(
mE
)
for each m ∈ N. Next we will see that each isomorphism between E
and F
induces an isomorphism between L
Θ(
mE; G
) and L
Θ(
mF ; G
) for each m ∈ N and induces also an isomorphism between P
Θ(
mE; G
) and P
Θ(
mF ; G
) for each m ∈ N, where Θ = f or N.
29 Theorem. If E
and F
are isomorphic, then L
Θ(
mE) and L
Θ(
mF ) are isomorphic, for all m ∈ N, where Θ = f or N.
Proof. We will only write the proof in the case Θ = N . In the case Θ = f , the proof is similar. We use the notations from the proof of Theorem 17 By Theorem 23 we have that R
m(L
N(
mE)) ⊂ L
N(
mF ). As in the proof of Theorem 23 we can prove that S
m(R
mA) =
∞i=1
S
1(R
1ϕ
1i) ⊗ · · · ⊗ S
1(R
1ϕ
mi) ⊗ c
i. Since S
1◦ R
1is the identity, we have that S
m(R
mA) =
∞i=1
ϕ
1i⊗ · · · ⊗ ϕ
mi⊗ c
i= A for all A ∈ L
N(
mE). On the other hand, in a similar way, we can get that S
m(L
N(
mF )) ⊂ L
N(
mE) and R
m(S
mB) = B for all B ∈ L
N(
mF ). We conclude that L
N(
mE) and L
N(
mF ) are isomorphic for all m ∈ N.
QED30 Theorem. If E
and F
are isomorphic, then L
Θ(
mE; G
) and L
Θ(
mF ; G
) are isomorphic for all m ∈ N, where Θ = f or N.
Proof. We will only write the proof in the case Θ = N . In the case Θ = f , the proof is similar. We use the notations from the proof of Theo- rem 19. By Theorem 24 we have that ( R
m(L
N(
mE; G
)) ⊂ L
N(
mF ; G
) and S
m(L
N(
mF ; G
)) ⊂ L
N(
mE; G
). As in the proof of Theorem 24 we can prove that S
m(( R
mA) =
∞i=1
S
1(R
1ϕ
1i) ⊗ · · · ⊗ S
1(R
1ϕ
mi) ⊗ c
i. Since S
1◦ R
1is the identity, we have that S
m(( R
mA) =
∞i=1
ϕ
1i⊗ · · · ⊗ ϕ
mi⊗ c
i= A for
all A ∈ L
N(
mE; G
). On the other hand, in a similar way, we can get that
R (
m( S
mB) = B for all B ∈ L
N(
mF ; G
). We conclude that L
N(
mE; G
) and
L
N(
mF ; G
) are isomorphic for all m ∈ N.
QED31 Theorem. If E
and F
are isomorphic, then P
Θ(
mE) and P
Θ(
mF ) are
isomorphic for all m ∈ N, where Θ = f or N.
Proof. We will only write the proof in the case Θ = N . In the case Θ = f , the proof is similar. We use the notations from the proof of Theorem 17.
By Theorem 27 we have that ' R
m(P
N(
mE)) ⊂ P
N(
mF ) and ' S
m(P
N(
mF )) ⊂ P
N(
mE). As in the proof of Theorem 27 we can prove that ' S
m( ' R
mP ) =
∞i=1
(S
1(R
1ϕ
i))
m⊗ c
i. Since S
1◦ R
1is the identity, we have that ' S
m( ' R
mP ) =
∞i=1
(ϕ
i)
m⊗ c
i= P for all P ∈ P
N(
mE). On the other hand, in a similar way, we can get that ' R
m( ' S
mQ) = Q for all Q ∈ P
N(
mF ). We conclude that P
N(
mE) and P
N(
mF ) are isomorphic for all m ∈ N.
QED32 Theorem. If E
and F
are isomorphic, then P
Θ(
mE; G
) and P
Θ(
mF ; G
) are isomorphic for all m ∈ N, where Θ = f or N.
Proof. We will only write the proof in the case of Θ = N . In the case of Θ = f the proof is similar. We use the notations from the proof of The- orem 19. By Theorem 28 we have that ' R
m( P
N(
mE; G
)) ⊂ P
N(
mF ; G
) and ' S
m( P
N(
mF ; G
)) ⊂ P
N(
mE; G
). As in the proof of Theorem 28 we can prove that ' S
m( ' R
mP ) =
∞i=1
(S
1(R
1ϕ
i))
m⊗ c
i. Since S
1◦ R
1is the identity, we have that ' S
m( ' R
mP ) =
∞i=1