° Kyungpook Mathematical Journalc
Polynomial Numerical Index of l
p(1 < p < ∞)
Sung Guen Kim
Department of Mathematics, Kyungpook National University, Daegu 702-701, Korea e-mail : sgk317@knu.ac.kr
Abstract. We present some estimates for the polynomial numerical index of lp(1 < p <
∞).
1. Introduction
Given a Banach space E we write BEfor its unit ball and SEfor its unit sphere.
The dual space of E is denoted by E∗ and let
Π(E) = {(x, x∗) : x ∈ SE, x∗∈ SE∗, x∗(x) = 1}.
A mapping P : E → E is called a (continuous) k-homogeneous polynomial if there is a (continuous) k-linear mapping A : E × · · · × E → E such that P (x) = A(x, · · · , x) for every x ∈ E. Let P(kE : E) denote the Banach space of all k-homogeneous polynomials from E to itself, endowed with the polynomial norm kP k = supx∈BEkP (x)k. We refer to a book [6] by Dineen for background on polynomials. It is natural to generalize the concepts of numerical range and numerical radius of linear operators to homogeneous polynomials. The numerical range of P ∈ P(kE : E) is defined to be the set of scalars
V (P ) := {x∗(P x) : (x, x∗) ∈ Π(E)}
and the numerical radius of P is defined by
v(P ) := sup {|λ| : λ ∈ V (P )}.
Received May 26, 2014; accepted October 23, 2014.
2010 Mathematics Subject Classification: Primary 46A22, 46G20.
Key words and phrases: Homogeneous polynomial, numerical radius, polynomial numerical index.
This research was supported by the Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education, Science and Technology (2013R1A1A2057788).
615
Clearly, v(·) is a semi-norm on P(kE : E), and v(P ) ≤ kP k for every P ∈ P(kE : E).
As in the linear case, the author et al. [3] introduced the concept of the polynomial numerical index of order k of E to be the constant
n(k)(E) := inf{v(P ) : P ∈ P(kE : E), kP k = 1}
= sup{M ≥ 0 : kP k ≤ 1
Mv(P ) for all P ∈ P(kE : E)}.
Of course, n(1)(E) coincides with the usual numerical index of the space E. Note that 0 ≤ n(k)(E) ≤ 1, and n(k)(E) > 0 if and only if v(·) is a norm on P(kE : E) equivalent to the usual norm. It is obvious that if E1, E2 are isometrically isomorphic Banach spaces, then n(k)(E1) = n(k)(E2).
The concept of the numerical index was first suggested by G. Lumer in 1968 (see [14]). The author et al. [3] introduced and studied the concept of the polyno- mial numerical index of order k of a Banach space, generalizing to k-homogeneous polynomials the classical numerical index. Recently, the author et al. [11] com- puted that 12 = n(2)(c0) = n(2)(l1) = n(2)(l∞) for the real spaces c0, l1and l∞. Very recently, Garcia et al. [9] gave estimates for the polynomial numerical indices of C(K) and L1(µ). For general information and background on polynomial numerical index, we refer to ([1]–[4], [7]–[15]).
In this paper, we present some inequality for the norm of 2-homogeneous poly- nomials from the real space lp (1 < p < ∞) to itself in terms of their coefficients.
Using this we give a lower bound for n(2)(lp). We also present some upper bounds for n(k)(lp).
2. Norm and Numerical Radius of 2-Homogeneous Polynomial on lp
Theorem 2.1. Let 1 < p < ∞, 1 = 1p + 1q and P (x, y) = (P1(x, y), P2(x, y)) ∈ P(2l2p: l2p), where Pk(x, y) = akx2+ bky2+ ckxy for k = 1, 2 and (x, y) ∈ l2p. Then
(1) kP kp≥ sup
0≤λ≤1|P1(λ1p, (1 − λ)1p)|p+ |P2(λ1p, (1 − λ)1p)|p; (2) v(P ) ≥ sup
0≤λ≤1|λ1qP1(λ1p, (1 − λ)1p) + (1 − λ)1qP2(λ1p, (1 − λ)1p)|.
Corollary 2.2. Let 1 < p < ∞, 1 = p1 +1q and P (x, y) = (P1(x, y), P2(x, y)) ∈ P(2l2p: l2p), where Pk(x, y) = akx2+ bky2+ ckxy for k = 1, 2 and (x, y) ∈ l2p. Then
(1) kP kp≥ sup
0≤λ≤1,k=1,2
{|P1([λ + (1 − λ)|ak− bk| (|ck|p+ |ak− bk|p)1p]1p,
[1 − λ − (1 − λ)|ak− bk|
(|ck|p+ |ak− bk|p)1p]1p)|p+ |P2([λ + (1 − λ)|ak− bk| (|ck|p+ |ak− bk|p)1p]p1,
[1 − λ − (1 − λ)|ak− bk|
(|ck|p+ |ak− bk|p)1p]1p)|p};
(2) v(P ) ≥ sup
0≤λ≤1,k=1,2{|[λ + (1 − λ)|ak− bk| (|ck|p+ |ak− bk|p)1p]1q P1([λ + (1 − λ)|ak− bk|
(|ck|p+ |ak− bk|p)1p]1p, [1 − λ − (1 − λ)|ak− bk| (|ck|p+ |ak− bk|p)1p]1p) + [1 − λ − (1 − λ)|ak− bk|
(|ck|p+ |ak− bk|p)1p]1q × P2([λ + (1 − λ)|ak− bk| (|ck|p+ |ak− bk|p)1p]p1, [1 − λ − (1 − λ)|ak− bk|
(|ck|p+ |ak− bk|p)1p]p1)|}.
Corollary 2.3. Let 1 < p < ∞, 1 = 1p +1q and P (x, y) = (P1(x, y), P2(x, y)) ∈ P(2l2p: l2p), where Pk(x, y) = akx2+ bky2+ ckxy for k = 1, 2 and (x, y) ∈ lp2. Then (1) kP kp≥ sup
k=1,2
{|P1([1
p+ |ak− bk|
q(|ck|p+ |ak− bk|p)1p]1p, [1
q− |ak− bk|
q(|ck|p+ |ak− bk|p)1p]p1)|p +|P2([1
p+ |ak− bk|
q(|ck|p+ |ak− bk|p)1p]p1, [1
q − |ak− bk|
q(|ck|p+ |ak− bk|p)1p]1p)|p,
|P1([1
q − |ak− bk|
q(|ck|p+ |ak− bk|p)1p]1p, [1
p+ |ak− bk|
q(|ck|p+ |ak− bk|p)1p]1p)|p +|P2([1
q − |ak− bk|
q(|ck|p+ |ak− bk|p)1p]1p, [1
p+ |ak− bk|
q(|ck|p+ |ak− bk|p)1p]1p)|p};
(2) v(P ) ≥ sup
k=1,2{|[1
p+ |ak− bk|
q(|ck|p+ |ak− bk|p)1p]1qP1([1
p+ |ak− bk|
q(|ck|p+ |ak− bk|p)p1]1p, [1
q − |ak− bk|
q(|ck|p+ |ak− bk|p)1p]1p) + [1
q − |ak− bk|
q(|ck|p+ |ak− bk|p)1p]q1
×P2([1
p+ |ak− bk|
q(|ck|p+ |ak− bk|p)1p]1p, [1
q − |ak− bk|
q(|ck|p+ |ak− bk|p)1p]1p)|,
|[1
q− |ak− bk|
q(|ck|p+ |ak− bk|p)1p]1qP1([1
q− |ak− bk|
q(|ck|p+ |ak− bk|p)1p]p1, [1
p+ |ak− bk|
q(|ck|p+ |ak− bk|p)1p]1p) + [1
p+ |ak− bk|
q(|ck|p+ |ak− bk|p)1p]1q
×P2([1
q − |ak− bk|
q(|ck|p+ |ak− bk|p)1p]1p, [1
p+ |ak− bk|
q(|ck|p+ |ak− bk|p)1p]1p)|}
3. Estimates for the Polynomial Numerical Index of lp
Theorem 3.1. Let 1 < p < ∞, 1 = 1p +1q and P (x) = (Pk(x))∞k=1 ∈ P(2lp : lp), where
Pk(x) =X
i≤j
a(k)ij xixj (x = (xn)∞n=1∈ lp).
Then
kP k ≤ k ( X∞ k=1
(|a(k)11| +1 2
X
1<j
|a(k)1j |), X∞ k=1
(|a(k)22| +1
2|a(k)12| +1 2
X
2<j
|a(k)2j |), · · · , X∞
k=1
(|a(k)nn| + 1 2
X
m<n
|a(k)mn| +1 2
X
n<j
|a(k)nj|), · · · ) kq.
Proof. It follows that
kP k = sup
x=(xn)∈Slp
( X∞ k=1
|Pk(x)|p)1/p
≤ sup
x=(xn)∈Slp
( X∞ k=1
( X∞ i=1
|a(k)ii ||xi|2+X
i<j
|a(k)ij ||xi||xj| )p)1/p
≤ sup
x=(xn)∈Slp
X∞ k=1
( X∞ i=1
|a(k)ii ||xi|2+X
i<j
|a(k)ij ||xi||xj| )
≤ sup
x=(xn)∈Slp
X∞ k=1
( X∞ i=1
|a(k)ii ||xi|2+X
i<j
|a(k)ij |(|xi|2+ |xj|2
2 ))
= sup
x=(xn)∈Slp
[ X∞ k=1
(|a(k)11| +1 2
X
1<j
|a(k)1j |)|x1|2
+ X∞ k=1
(|a(k)22| +1
2|a(k)12| +1 2
X
2<j
|a(k)2j |)|x2|2
+ · · · + X∞ k=1
(|a(k)nn| + 1 2
X
m<n
|a(k)mn| +1 2
X
n<j
|a(k)nj|)|xn|2+ · · · ]
≤ sup
x=(xn)∈Slp
[ X∞ k=1
(|a(k)11| +1 2
X
1<j
|a(k)1j |)|x1|
+ X∞ k=1
(|a(k)22| +1
2|a(k)12| +1 2
X
2<j
|a(k)2j |)|x2|
+ · · · + X∞ k=1
(|a(k)nn| + 1 2
X
m<n
|a(k)mn| +1 2
X
n<j
|a(k)nj|)|xn| + · · · ]
= k ( X∞ k=1
(|a(k)11| +1 2
X
1<j
|a(k)1j |), X∞ k=1
(|a(k)22| +1
2|a(k)12| +1 2
X
2<j
|a(k)2j |), · · · , X∞
k=1
(|a(k)nn| +1 2
X
m<n
|a(k)mn| +1 2
X
n<j
|a(k)nj|), · · · ) kq. 2
Theorem 3.2. Let 1 < p < ∞, 1 = 1p+1q. Then
n(2)(lp) ≥
inf{ supn∈N(n1)1+ 1p|Pn
l=1 P
i≤ja(l)ij| k (Pn
k=1(|a(k)11| +12Pn
j=2|a(k)1j|), · · · ,Pn
k=1(|a(k)nn| +12P
m<n|a(k)mn|), 0, 0, · · · ) kq :
nonzero P (x) = (Pk(x))∞k=1∈ P(2lp: lp), Pk(x) =X i≤j
a(k)ij xixj}.
Proof. Suppose that P (x) = (Pk(x))∞k=1∈ P(2lp: lp) with Pk(x) =X
i≤j
a(k)ij xixj (x = (xn)∞n=1∈ lp).
Let n ∈ N, x0=Pn
k=1(n1)1/pek, and x∗0=Pn
k=1(n1)1/qek. Then (∗) v(P ) ≥ |x∗0(P (x0))| = (1
n)1+p1| Xn l=1
X
i≤j
a(l)ij|.
It follows that by (∗) and Theorem 3.1, n(2)(lp)
= inf{v(P )
kP k : P (x) = (Pk(x))∞k=1∈ P(2lp: lp), P 6= 0}
≥ inf{ supn∈N(n1)1+ 1p|Pn
l=1 P
i≤ja(l)ij| k ( Pn
k=1(|a(k)11| +12Pn
j=2|a(k)1j|), · · · ,Pn
k=1(|a(k)nn| +12P
m<n|a(k)mn|), 0, 0, · · · ) kq :
nonzero P (x) = (Pk(x))∞k=1∈ P(2lp: lp)}. 2 Corollary 3.3. We have
inf{ supn∈N(n1)32|Pn
l=1 P
i≤ja(l)ij| k (Pn
k=1(|a(k)11| +12Pn
j=2|a(k)1j|), · · · ,Pn
k=1(|a(k)nn| +12P
m<n|a(k)mn|), 0, 0, · · · ) k2 :
nonzero P (x) = (Pk(x))∞k=1∈ P(2l2: l2), Pk(x) =P
i≤ja(k)ij xixj} = 0.
Theorem 3.4. Let k ∈ N with k ≥ 2. Then n(2)(lk2) = inf{kQPkl2
k: P ∈ P(2l2k : lk2),
kP k = 1, P (x, y) = (ax2+ by2+ cxy, a0x2+ b0y2+ c0xy)}, where QP ∈ P(k+1l2k) with QP(x, y) = axk+1+ bxk−1y2+ cxky + a0x2yk−1 +b0yk+1+ c0xyk.
Theorem 3.5. Let 1 < p < ∞ and k ∈ N with k ≥ 2. Then n(k)(lp) ≤ n(k)(l2p)
≤ min{( p − 1
k + p − 1)1−1p ( k
k + p − 1)kp, (p + k − 1
2p + k − 2)p1(2p+k−2p+k−2)p+k−2p (p+k−1k−1 )k−1p }.
Proof. Let P1(x, y) = (xk, 0) and P2(x, y) = (xk−1y, 0) for (x, y) ∈ lp2.
Then Pj∈ P(kl2p: l2p) for j = 1, 2 and kP1k = 1. Some computation shows that kP2k = ( k − 1
p + k − 1)k−1p ( p p + k − 1)1p, v(P2) = (p + k − 2
2p + k − 2)p+k−2p ( p 2p + k − 2)p1, and v(P1) = ( p − 1
k + p − 1)1−1p ( k k + p − 1)kp. It follows that
n(k)(lp) ≤ n(k)(l2p) ≤ min{v(P1),v(P2) kP2k}
= min{( p − 1
k + p − 1)1−1p ( k
k + p − 1)kp, ( p + k − 1
2p + k − 2)1p(2p+k−2p+k−2)p+k−2p (p+k−1k−1 )k−1p },
which completes the proof. 2
Theorem 3.6. Let k ∈ N. Then n(2k+1)(l2k) = 0 for the real space l2k. Proof. Let
P0(x) := (−x1x2k2 , x2k1 x2, 0, 0, . . .) (x = (xj) ∈ l2k).
Then kP0k 6= 0. Note that if (x1, x2, 0, 0, . . .) ∈ Sl2k, then (x2k−11 , x2k−12 , 0, 0, . . .) ∈ S(l2k)∗= Sl 2k
2k−1
. It follows that v(P0)
= sup
x2k1 +x2k2 =1
| < (sign(x2k1 )x2k−11 , sign(x2k2 )x2k−12 , 0, . . .), P0(x1, x2, 0, . . .) > |
= sup{| < (x2k−11 , x2k−12 , 0, 0, . . .), P0(x1, x2, 0, 0, . . .) > | : x2k1 + x2k2 = 1}
= sup{| − x1x2k2 + x1x2k2 | : x2k1 + x2k2 = 1} = 0.
Thus
n(2k+1)(l2k) ≤ v(P0) kP0k = 0.
2 Corollary 3.7. Let k, s ∈ N with s ≥ 2k + 1. Then n(s)(l2k) = 0 for the real space l2k.
Notice that there exist 1 < p < q < ∞, N ≥ 2 such that n(N )(lp) = n(N )(lq).
In fact, take p = 2k, q = 2(k + 1), N = 2k + 3. By Corollary 3.7, n(N )(lp) = 0 = n(N )(lq).
We are in position to prove our main result.
Theorem 3.8. Let 1 < p < ∞. Then for the complex space lp, n(k)(lp) ≤ 21−kp (∀k ∈ N).
Proof. Simple computation shows that sup
|x1|p+|x2|p=1
(|x1|p+k−1+ |x2|p+k−1) = 21−kp = sup
|x1|p+|x2|p=1
(|x1|kp+ |x2|kp)1p.
For every b ∈ C, define
Qb(x) := (xk1+ bxk2, bxk1+ xk2, 0, 0, . . .) ∈ P(klp) (x = (xj) ∈ lp).
Then kQbk ≥ |Qb(1, 0)| = (1 + |b|p)1p and v(Qb)
= sup
|x1|p+|x2|p=1
| < (sign(xp1)xp−11 , sign(xp2)xp−12 , 0, 0, . . .), Qb(x1, x2, 0, 0, . . .) > |
≤ sup
|x1|p+|x2|p=1
(|x1|p+k−1+ |x2|p+k−1) +|b| sup
|x1|p+|x2|p=1
(|x1|p−1|x2|k+ |x2|p−1|x1|k)
≤ 21−kp + |b| sup
|x1|p+|x2|p=1
(|x1|(p−1)q+ |x2|(p−1)q)1q sup
|x1|p+|x2|p=1
(|x1|kp+ |x2|kp)1p (by Holder’s inequality)
= 21−kp + |b| sup
|x1|p+|x2|p=1
(|x1|p+ |x2|p)q1 sup
|x1|p+|x2|p=1
(|x1|kp+ |x2|kp)1p
= 21−kp + |b|21−kp
= 21−kp (1 + |b|).
It follows that
n(k)(lp) ≤ inf
b∈C
v(Qb) kQbk ≤ inf
b∈C
21−kp (1 + |b|)
(1 + |b|p)1p = 21−kp .
2 Corollary 3.9. Let 1 < p < ∞. Then
(1) n(k)(lp) ≤ 2−1p < 1 (∀k ≥ 2).
(2) n(k)(lp) ≤ 12 (∀k ≥ p + 1).
(3) 2+M2k
k(2k−2) ≤ n(k)(l1) ≤ (12)k−1, where k ≥ 1 and Mk=Pk
j=1 jk j!(k−j)!. (4) limk→∞n(k)(lp) = 0.
Theorem 3.10. Let p be an even number and k an odd number with k ≤ p − 1.
Then
n(k)(lp) ≤ ( k
p − 1)p−1−kk − ( k
p − 1)p−1−kp−1 , when lp is the real space.
Proof. We define
Q0(x) := (−xk2, xk1, 0, 0, . . .) ∈ P(klp) (x = (xj) ∈ lp).
Then kQ0k = 1 and v(Q0)
= sup
xp1+xp2=1
| < (xp−11 , xp−12 , 0, 0, . . .), Q0(x1, x2, 0, 0, . . .) > |
= sup
xp1+xp2=1
| − xp−11 xk2+ xk1xp−12 |
≤ sup
xp1+xp2=1
|x1x2|k− |x1x2|p−1
= ( k
p − 1)p−1−kk − ( k
p − 1)p−1−kp−1 , when |x1x2| = ( k
p − 1)p−1−k1 . Thus
n(k)(lp) ≤ v(Q0) kQ0k = ( k
p − 1)p−1−kk − ( k
p − 1)p−1−kp−1 .
2 Theorem 3.11. Let p and s be even numbers with s < p − 1. Then
n(s)(lp) ≤ min{(s − 1
p − 1)p−ss−1 − (s − 1
p − 1)p−1p−s, 2−ps}, when lp is the real space.
Proof. By Theorem 3.10,
n(s)(lp) ≤ n(s−1)(lp) ≤ (s − 1
p − 1)s−1p−s− (s − 1 p − 1)p−1p−s.
We define
Q0(x) := (−xs2, xs1, 0, 0, . . .) ∈ P(slp) (x = (xj) ∈ lp).
Then kQ0k = 1 and v(Q0)
= sup
xp1+xp2=1
| < (xp−11 , xp−12 , 0, 0, . . .), Q0(x1, x2, 0, 0, . . .) > |
= sup
xp1+xp2=1
| − xp−11 xs2+ xs1xp−12 |
= sup
xp1+xp2=1
|x1|p−1|x2|s+ |x1|s|x2|p−1
≤ sup
xp1+xp2=1
2|x1x2|s
= 2−sp. Thus
n(k)(lp) ≤v(Q0)
kQ0k = 2−sp. 2
References
[1] F. F. Bonsall and J. Duncan, Numerical Ranges of Operators on Normed Spaces and of Elements of Normed Algebras, London Math. Soc. Lecture Note Ser., 2, (Cambridge Univ. Press, 1971).
[2] F. F. Bonsall and J. Duncan, Numerical Ranges II, London Math. Soc. Lecture Note Ser., 10, (Cambridge Univ. Press, 1973).
[3] Y. S. Choi, D. Garcia, S. G. Kim, and M. Maestre, The polynomial numerical index of a Banach space, Proc. Edinburgh Math. Soc., 49(2006), 39-52.
[4] Y. S. Choi, D. Garcia, S. G. Kim, and M. Maestre, Composition, numerical range and Aron-Berner extension, Math. Scand., 103(2008), 97-110.
[5] Y. S. Choi and S. G. Kim, Norm or numerical radius attaining multilinear mappings and polynomials, J. London Math. Soc., 54(2)(1996), 135-147.
[6] S. Dineen, Complex Analysis on Infinite Dimensional Spaces, Springer-Verlag, London (1999).
[7] J. Duncan, C. M. McGregor, J. D. Pryce, and A. J. White, The numerical index of a normed space, J. London Math. Soc., 2(1970), 481-488.
[8] D. Garcia, B. Grecu, M. Maestre, M. Martin, and J. Meri, Two dimensional Banach spaces with polynomial numerical index zero, Linear Alg. Appl., 430(2009), 2488-2500.
[9] D. Garcia, B. Grecu, M. Maestre, M. Martin, and J. Meri, Polynomial numerical indices of C(K) and L1(µ), Proc. Amer. Math. Soc., 142(2014), 1229–1235.
[10] V. Kadets, M. Martin, and R. Paya, Recent progress and open questions on the nu- merical index of Banach spaces, Rev. R. Acad. Cien. Serie A. Mat., 2000(2006), 155–182.
[11] S. G. Kim, M. Martin, and J. Meri, On the polynomial numerical index of the real spaces c0, l1 and l∞, J. Math. Anal. Appl., 337(2008), 98-106.
[12] S. G. Kim, Three kinds of numerical indices of a Banach space, Math. Proc. Royal Irish Acad., 112A(2012), 21-35.
[13] H. J. Lee, Banach spaces with polynomial numerical index 1, Bull. Lond. Math. Soc., 40(2)(2008), 193–198.
[14] G. Lumer, Semi-inner-product spaces, Trans. Amer. Math. Soc., 100(1961), 29-43.
[15] M. Martin and J. Meri, A note on the numerical index of the Lpspace of dimension 2, Linear and Multilinear Algebra, 57(2009), 201–204.