EQUIVALENCE OF n-NORMS ON THE SPACE OF p-SUMMABLE SEQUENCES | Mutaqin | Journal of the Indonesian Mathematical Society 27 97 1 PB

J. Indones. Math. Soc.
Vol. 16, No. 1 (2010), pp. 1–7.

EQUIVALENCE OF n-NORMS ON THE SPACE OF
p-SUMMABLE SEQUENCES
Anwar Mutaqin1 and Hendra Gunawan2
1

Department of Mathematics Education, Universitas Sultan Ageng
Tirtayasa, Serang, Indonesia, anwarmutaqin@gmail.com
2
Department of Mathematics, Institut Teknologi Bandung, Bandung,
Indonesia, hgunawan@math.itb.ac.id

Abstract. We study the relation between two known n-norms on ℓp , the space of
p-summable sequences. One n-norm is derived from G¨
ahler’s formula [3], while the
other is due to Gunawan [6]. We show in particular that the convergence in one
n-norm implies that in the other. The key is to show that the convergence in each
of these n-norms is equivalent to that in the usual norm on ℓp .
Key words: n-normed spaces, p-summable sequence spaces, n-norm equivalence.

Abstrak. Dalam makalah ini dipelajari kaitan antara dua norm-n di ℓp , ruang
barisan summable-p. Norm-n pertama diperoleh dari rumus G¨
ahler [3], sementara
norm-n kedua diperkenalkan oleh Gunawan [6]. Ditunjukkan antara lain bahwa
kekonvergenan dalam norm-n yang satu mengakibatkan kekonvergenan dalam normn lainnya. Kuncinya adalah bahwa kekonvergenan dalam masing-masing norm-n
tersebut setara dengan kekonvergenan dalam norm biasa di ℓp .
Kata kunci: ruang norm-n, ruang barisan summable-p, kesetaraan norm-n

1. Introduction
In [6], Gunawan introduced an n-norm on ℓp (1 ≤ p ≤
p-summable sequences (of real numbers), given by the formula
¯

¯ x1j1 · · · xnj1
¯
X
X
1
¯

..
..
kx1 , . . . , xn kp := 
abs ¯ ...
···
.
.
¯
n! j
jn
¯ x1j
1
·
·
·
x
njn
n

2000 Mathematics Subject Classification: 46B15

Received: 31-03-2010, accepted: 05-08-2010.

1

∞), the space of
¯p 1/p
¯
¯
¯ 
¯ 
¯
¯

A. Mutaqin and H. Gunawan

2

for 1 ≤ p < ∞, and
kx1 , . . . , xn k∞

¯

¯ x1j1

¯

¯
= sup sup · · · sup abs ¯ ...
¯
jn 
j1 j2

¯ xnj
1

···
..
.
···

x1jn
..
.
xnjn

¯
¯
¯
¯
¯ ,
¯
¯

where xi = (xij ), i = 1, . . . , n. For p = 2, the above formula may be rewritten as
¯
¯
¯ hx1 , x1 i · · · hx1 , xn i ¯1/2
¯
¯
¯

¯
..
..
..
kx1 , . . . , xn k2 = ¯
¯ ,
.
.
.
¯
¯
¯ hxn , x1 i · · · hxn , xn i ¯

where hxi , xj i denotes the usual inner product on ℓ2 . Here kx1 , . . . , xn k2 represents
the volume of the n-dimensional parallelepiped spanned by x1 , . . . , xn in ℓ2 .
In general, an n-norm on a real vector space X is a mapping k·, . . . , ·k : X n →
R which satisfies the following four conditions:
(N1) kx1 , . . . , xn k = 0 if and only if x1 , . . . , xn are linearly dependent;
(N2) kx1 , . . . , xn k is invariant under permutation;
(N3) kαx1 , . . . , xn k = |α| kx1 , . . . , xn k for α ∈ R;

(N4) kx1 + x′1 , x2 , . . . , xn k ≤ kx1 , x2 , . . . , xn k + kx′1 , x2 , . . . , xn k.
The theory of n-normed spaces was developed by G¨ahler in 1969 and 1970 [3, 4, 5].
The special case where n = 2 was studied earlier, also by G¨ahler, in 1964 [2].
Related work may be found in [1]. For more recent works, see [7, 8, 10].
If X is equipped with a norm k · k, then according to G¨ahler, one may define
an n-norm on X (assuming that X is at least n-dimensional) by the formula
¯
¯
¯ f1 (x1 ) · · · f1 (xn ) ¯
¯
¯
¯
¯
∗
..
..
..
kx1 , . . . , xn k :=
sup
¯.

¯
.
.
.
′
¯
¯
fi ∈X , kfi k≤1 ¯
fn (x1 ) · · · fn (xn ) ¯
i = 1,...,n
Here X ′ denotes the dual of X, which consists of bounded linear functionals on X.
′
For X = ℓp (1 ≤ p < ∞), we know that X ′ = ℓp with p1 + p1′ = 1. In this
case the above formula reduces to
¯ P
¯
P
¯
x1j z1j · · ·
x1j znj ¯¯

¯
¯
¯
∗
..
..
..
kx1 , . . . , xn kp :=
sup
¯
¯,
.
.
.
′
¯
¯
P
P
zi ∈ℓp , kzi kp′ ≤1 ¯

¯
x
z
·
·
·
x
z
nj
1j
nj
nj
i = 1,...,n
′

where k·kp′ denotes the usual norm on ℓp and each of the sums is taken over j ∈ N.
Thus, on ℓp , we have two definitions of n-norms, one is due to Gunawan and the
other is derived from G¨ahler’s formula. For p = 2, one may verify that the two
n-norms are identical.
The purpose of this paper is to study the relation between the two n-norms

on ℓp for 1 ≤ p < ∞. In particular, we shall show that the two n-norms are weakly
equivalent, that is, the convergence in one n-norm implies that in the other. Here

Equivalence of n-Norms on the Space of p-Summable Sequences

3

a sequence (x(m)) in an n-normed space (X, k·, . . . , ·k) is said to converge to x ∈ X
if kx(m) − x, x2 , . . . , xn k → 0 as m → ∞, for every x2 , . . . , xn ∈ X.
For convenience, we prove the result for n = 2 first, and then extend it to
any n ≥ 2.

2. Main Results
Recall that Gunawan’s definition of 2-norm on ℓp (1 ≤ p ≤ ∞) is given by

¯p 1/p
xk ¯¯ 
yk ¯



¯
¯ x
1 XX

kx, ykp =
abs ¯¯ j
yj
2 j
k

if 1 ≤ p < ∞, and

¯¾
xk ¯¯
.
yk ¯

¯
½
¯ x
kx, yk∞ = sup sup abs ¯¯ j
yj
j
k

Meanwhile, G¨ahler’s definition is given by
kx, yk∗p

=

¯ P
¯
¯ P xj zj
¯
yj zj
′ ≤1

sup

z,w∈ℓp′ , kzkp′ , kwkp

P
P xj wj
yj wj

By the same trick as in [6], one may obtain
kx, yk∗p

=
z,w∈ℓp′ ,

¯
1 X X ¯¯ xj
sup
¯ yj
kzkp′ , kwkp′ ≤1 2 j
k

¯¯
xk ¯¯ ¯¯ zj
yk ¯ ¯ wj

From the last expression, we have the following fact.

¯
¯
¯.
¯
¯
zk ¯¯
.
wk ¯

Fact 2.1. The inequality kx, yk∗p ≤ 21/p kx, ykp holds for every x, y ∈ ℓp .
Proof. By H¨older’s inequality for
¯
1 X X ¯¯ xj
¯ yj
2
j

k

¯¯
xk ¯¯ ¯¯ zj
yk ¯ ¯ wj

1
p

+

1
p′

= 1, we have


¯
¯
¯ xj
¯
zk ¯  1 X X
¯
abs
≤
¯
¯ yj
wk
2
j



×

k

1 XX

2

j

k


¯p 1/p
¯
xk ¯ 
yk ¯

¯
¯ z
abs ¯¯ j
wj

 ′
¯p′ 1/p
¯
zk ¯ 
wk ¯

A. Mutaqin and H. Gunawan

4

Now, observe that

¯
XX
¯ z

abs ¯¯ j
wj
j

k

1/p′
 ′ 
¯p′ 1/p
X X£
¤p′
zk ¯¯ 
≤
|zj wk | + |zk wj | 
wk ¯
j

k

1/p′

1/p′ 
X
X
XX
′
′
+
≤
|zk wj |p 
|zj wk |p 
j

j

k

k

= 2 kzkp′ kwkp′ .

But for kzkp′ , kwkp′ ≤ 1 we have

¯
XX
¯ z
1
abs ¯¯ j
wj
2 j
k

 ′
¯p′ 1/p
′
zk ¯¯ 
≤ 21−(1/p ) = 21/p .
wk ¯

This proves the inequality.
Note that for p = 1, H¨older’s inequality
¯
¯¯
1 X X ¯¯ xj xk ¯¯ ¯¯ zj zk
¯ yj yk ¯ ¯ wj wk
2 j
k

gives
¯
¯
¯ ≤ kx, yk . kz, wk .
1
∞
¯

But kz, wk∞ ≤ 2 kzk∞ kwk∞ (see [6]), and so taking the supremum over kzk∞ and
kwk∞ ≤ 1, we get kx, yk∗1 ≤ 2kx, yk1 .
¤
Corollary 2.2 If (x(m)) converges in k·, ·kp , then it also converges (to the same
limit) in k·, ·k∗p .
We shall show next that the convergence in k·, ·k∗p also implies the convergence
in k·, ·kp . We do so by showing that: (1) the convergence in k·, ·k∗p implies that in
k · kp , and (2) the convergence in k · kp implies that in k·, ·kp .
The second implication is already proved in [6] (using the inequality kx, ykp ≤
21−(1/p) kxkp kykp ). Hence it remains only to show the first implication.
Theorem 2.3 If (x(m)) converges in k·, ·k∗p , then it also converges (to the same
limit) in k · kp .
Proof. Let (x(m)) be a sequence in ℓp which converges to x ∈ ℓp in k·, ·k∗p . Then,
for any ǫ > 0, there exists an N ∈ N such that for m ≥ N we have
¯
¯
¯¯
1 X X ¯¯ xj (m) − xj xk (m) − xk ¯¯ ¯¯ zj zk ¯¯
¯ ¯ wj wk ¯ < ǫ
¯
yj
yk
2
j

k

′

for every y ∈ ℓp and z, w ∈ ℓp with kzkp′ , kwkp′ ≤ 1. [Notice here that, for each m,
we have x(m) = (xj (m)) ∈ ℓp .] In particular, if we take y := (1, 0, 0, . . . ), z = (zj )

Equivalence of n-Norms on the Space of p-Summable Sequences

with zj :=

sgn(xj (m)−xj )|xj (m)−xj |p−1
kx(m)−xkp−1
p

5

and w := (1, 0, 0, . . . ), then we have

∞
X
|xj (m) − xj |p
< ǫ.
kx(m) − xkp−1
p
j=2

[Here we are handling only the case where kx(m) − xkp 6= 0.] Next, if we take
y := (0, 1, 0, . . . ), z = (z1 , 0, 0, . . . ) with z1 :=

sgn(x1 (m)−x1 )|x1 (m)−x1 |p−1
kx(m)−xkp−1
p

and w :=

(0, 1, 0, . . . ), then we have
|x1 (m) − x1 |p
< ǫ.
kx(m) − xkp−1
p
Adding up, we get
kx(m) − xkp =

∞
X
|xj (m) − xj |p
< 2ǫ.
kx(m) − xkp−1
p
j=1

This shows that (x(m)) converges to x in k · kp .

¤

Corollary 2.4 A sequence is convergent in k·, ·k∗p if and only if it is convergent (to
the same limit) in k·, ·kp .
All these results can be extended to n-normed spaces for any n ≥ 2. As an
extension of Fact 2.1, we have:
Fact 2.5 The inequality kx1 , . . . , xn k∗p ≤ (n!)1/p kx1 , . . . , xn kp holds for every x1 , . . . ,
xn ∈ ℓp .
Corollary 2.6 If (x(m)) converges in k·, . . . , ·kp , then it converges (to the same
limit) in k·, . . . , ·k∗p .
Analogous to Theorem 2.3, we have:
Theorem 2.7 If (x(m)) converges in k·, . . . , ·k∗p , then it also converges (to the same
limit) in k · kp .
Proof. Let (x1 (m)) be a sequence in ℓp which converges to x1 = (x11 , x12 , . . . ) ∈ ℓp
∗
in k·, . . . , ·kp . Then, for any ǫ > 0, there exists an N ∈ N such that for m ≥ N we
have
¯
¯¯
¯
¯ x1j1 (m) − x1j1 · · · x1jn (m) − x1jn ¯ ¯ z1j1 · · · z1jn ¯
¯
¯
¯
¯
X
X
1
¯ ¯ ..
¯
.. ¯ < ǫ
..
..
..
..
···
¯
¯
¯
.
.
. ¯¯
.
.
¯¯ .
¯
n! j
jn ¯
¯
¯
1
znj1 · · · znjn ¯
···
xnjn
xnj1

for every x2 , . . . , xn ∈ ℓp and z1 , . . . , zn ∈ ℓp with kz1 k , . . . , kzn k ≤ 1. Now, take
xk = zk := (0, . . . , 0, 1, 0, . . . ) for every k = 2, . . . , n, where 1 is (n + 1 − k)-th

A. Mutaqin and H. Gunawan

6

′

term and z1 = (z11 , z12 , . . . ) ∈ ℓp with z1j :=
we have

sgn(x1j (m)−x1j )|x1j (m)−x1j |p−1
,
kx1 (m)−x1 kp−1
p

∞
p
X
|x1j1 (m) − x1j1 |

p−1

j1 =n

kx1 (m) − x1 kp

then

< ǫ.

Next, if we take xk = zk := (0, . . . , 0, 1, 0, . . . ) for every k = 2, . . . , n, where 1 is
p−1
11 )|x11 (m)−x11 |
k-th term, and z1 := (z11 , 0, 0, . . . ) with z11 := sgn(x11 (m)−x
, then
kx (m)−x kp−1
1

we have

1 p

p

|x11 (m) − x11 |

p−1

kx1 (m) − x1 kp

< ǫ.

Similarly, if we alter the position of the entry 1 in xk and zk for k = 2, . . . , n, and
change the nonzero entry of z1 accordingly, then we can get
|x12 (m) − x12 |

p

p−1

kx1 (m) − x1 kp
and so on until

0 there exists an N ∈ N such
that kx(l) − x(m), x2 , . . . , xn k < ǫ whenever l, m ≥ N , for every x2 , . . . , xn ∈ X.]
Since (ℓp , k·, . . . , ·kp ) is a Banach space [6], we conclude, by Theorem 2.7, that
(ℓp , k·, . . . , ·k∗p ) also forms an n-Banach space.
3. Concluding Remarks
As we have mentioned earlier, the case where p = 2 is of course special.
Here, the two n-norms k·, . . . , ·k2 and k·, . . . , ·k∗2 are identical. Indeed, by using
Cauchy-Schwarz inequality (see [9]), one may obtain
¯
¯
¯ hx1 , z1 i · · · hx1 , zn i ¯
¯
¯
¯
¯
..
..
..
kx1 , . . . , xn k∗2 =
sup
¯ ≤ kx1 , . . . , xn k2 .
¯
.
.
.
¯
¯
2
zi ∈ℓ , kzi k2 ≤1 ¯
hxn , z1 i · · · hxn , zn i ¯
i = 1,...,n

Equivalence of n-Norms on the Space of p-Summable Sequences

7

By taking z1 , . . . , zn to be the orthonormalized vectors obtained from x1 , . . . , xn
through Gram-Schmidt process, one can show that the above upper bound is actually attained. Hence we have
kx1 , . . . , xn k∗2 = kx1 , . . . , xn k2 .
For p 6= 2, things are not so simple and we have difficulties in proving the strong
equivalence between the two n-norms k·, . . . , ·k∗p and k·, . . . , ·kp . The research on
this problem, however, is still ongoing.
Acknowledgement. The research was carried out while the first author did his
master thesis at Faculty of Mathematics and Natural Sciences, Institut Teknologi
Bandung.

References
[1] Diminnie, C.R., G¨
ahler, S., and White, A., ”2-inner product spaces”, Demonstratio Math. 6
(1973), 525–536.
[2] G¨
ahler, S., ”Lineare 2-normierte r¨
aume”, Math. Nachr. 28 (1964), 1–43.
[3] G¨
ahler, S., ”Untersuchungen u
¨ber verallgemeinerte m-metrische R¨
aume. I”, Math. Nachr.
40 (1969), 165 - 189.
[4] G¨
ahler, S., ”Untersuchungen u
¨ber verallgemeinerte m-metrische R¨
aume. II”, Math. Nachr.
40 (1969), 229–264.
[5] G¨
ahler, S., ”Untersuchungen u
¨ber verallgemeinerte m-metrische R¨
aume. III”, Math. Nachr.
41 (1970), 23–26.
[6] Gunawan, H., ”The space of p-summable sequences and its natural n-norms”, Bull. Austral.
Math. Soc. 64 (2001), 137–147.
[7] Gunawan, H., ”On n-inner products, n-norms, and the Cauchy-Schwarz inequality”, Sci.
Math. Jpn. 55 (2002), 53–60.
[8] Gunawan, H. and Mashadi, ”On n-normed spaces”, Int. J. Math. Math. Sci. 27 (2001),
631–639.
[9] Gunawan, H., Neswan, O., and Setya-Budhi, W., ”A formula for angles between two subspaces of inner product spaces”, Beitr¨
a ge Algebra Geom. 46 (2005), 311–320.
[10] Misiak, A., ”n-inner product spaces”, Math. Nachr. 140 (1989), 299–319.