L ( l 2 2 )">
검색
Article Search

JMB Journal of Microbiolog and Biotechnology

OPEN ACCESS eISSN 0454-8124
pISSN 1225-6951
QR Code

Article

Kyungpook Mathematical Journal 2018; 58(1): 47-54

Published online March 23, 2018

Copyright © Kyungpook Mathematical Journal.

The Geometry of L ( l 2 2 )

Sung Guen Kim

Department of Mathematics, Kyungpook National University, Daegu 702-701, Korea, e-mail: sgk317@knu.ac.kr

Received: October 8, 2015; Accepted: January 18, 2016

We classify the extreme, exposed and smooth bilinear forms of the unit ball of the space of bilinear forms on l2.

Keywords: Bilinear forms, extreme points, exposed points, smooth points

We write BE for the closed unit ball of a real Banach space E and the dual space of E is denoted by E*. An element xBE is called an extreme point of BE if y, zBE with x=12(y+z) implies x = y = z. An element xBE is called an exposed point of BE if there is a fE* so that f(x) = 1 = ||f|| and f(y) < 1 for every yBE {x}. An element xBE is called a smooth point of BE if there is a unique fE* so that f(x) = 1 = ||f||. It is easy to see that every exposed point of BE is an extreme point. We denote by extBE, expBE and smBE the sets of extreme, exposed and smooth points of BE, respectively. A mapping P: E → ℝ is a continuous 2-homogeneous polynomial if there exists a continuous bilinear form L on the product E×E such that P(x) = L(x, x) for every xE. We denote by ℒ(2E) the Banach space of all continuous bilinear forms on E endowed with the norm ||L|| = sup||x||= ||y||=1 |L(x, y)|. ℘(2E) denotes the Banach space of all continuous 2-homogeneous polynomials from E into ℝ endowed with the norm ||P|| = sup||x||=1 |P(x)|. For more details about the theory of multilinear mappings and polynomials on a Banach space, we refer to [8].

In 1998, Choi et al. ([3], [4]) characterized the extreme points of the unit ball of P(l212) and P(l222). In 2007, the author [11] classified the exposed 2-homogeneous polynomials on P(l2p2) (1 ≤ p≤∞). Recently, the author ([14], [16], [20]) classified the extreme, exposed and smooth points of the unit ball of ℘(2d*(1, w)2), where d*(1, w)2 = ℝ2 with the octagonal norm of weight w.

In 2009, the author [13] classified the extreme, exposed and smooth points of the unit ball of Ls(l22). Recently, the author ([15], [17]–[19]) classified the extreme, exposed and smooth points of the unit balls of ℒs(2d*(1, w)2) and ℒ(2d*(1, w)2).

We refer to ([1]–[7], [9]–[25] and references therein) for some recent work about extremal properties of multilinear mappings and homogeneous polynomials on some classical Banach spaces. In this paper, we classify the extreme, exposed and smooth bilinear forms of the unit ball of L(l22).

Let T((x1,y1),(x2,y2))=ax1x2+by1y2+cx1y2+dx2y1L(l22) for some real numbers a, b, c, d. For simplicity we will write T((x1, y1), (x2, y2)) = (a, b, c, d). Let

T1((x1,y1),(x2,y2)):=T((y1,x1),(y2,x2))=(b,a,d,c),T2((x1,y1),(x2,y2)):=T((x1,y1),(x2,-y2))=(a,-b,-c,d),T3((x1,y1),(x2,y2)):=T((x1,y1),(y2,x2))=(c,d,a,b)T4((x1,y1),(x2,y2)):=T((y2,x2),(x1,y1))=(c,d,b,a).

Then ||Ti|| = ||T|| (i = 1, …, 4).

Theorem 2.1

LetT((x1,y1),(x2,y2))=(a,b,c,d)L(l22). Then,

T=max{a+b+c+d,a-b+c-d}.
Proof

Since {(1, 1), (1,−1), (−1, 1), (−1,−1)} is the set of all extreme points of the unit ball of l2 and T is bilinear,

T=max{T((1,1),(1,1)),T((1,-1),(-1,1)),T((1,1),(1,-1)),T((-1,1),(1,1))}=max{a+b+c+d,a-b+c-d}.

Note that if ||T|| = 1, then |a| ≤ 1, |b| ≤ 1, |c| ≤ 1 and |d| ≤ 1.

Theorem 2.2

LetT((x1,y1),(x2,y2))=(a,b,c,d)L(l22). Then, the followings are equivalent:

  • (1) T is extreme;

  • (2) (b, a, d, c) is extreme;

  • (3) (a,−b,−c, d) is extreme;

  • (4) (c, d, a, b) is extreme;

  • (5) (c, d, b, a) is extreme.

Proof

It follows from Theorem 2.1 and the above remark of Theorem 2.1.

Let

Norm(T)={((x1,y1),(x2,y2)){((1,1),(1,1)),((1,-1),(1,-1)),((1,1),(1,-1)),((1,-1),(1,1))}:T((x1,y1),(x2,y2))=T}.

We call Norm(T) the norming set of T.

Theorem 2.3

LetTL(l22)with ||T|| = 1. Then,TextBL(l22)if and only if

Norm(T)={((1,1),(1,1)),((1,-1),(1,-1)),((1,1),(1,-1)),((1,-1),(1,1))}.
Proof

Suppose that TL(l22) with ||T|| = 1. By Theorem 2.2, we may assume that a ≥ |b| and c ≥ |d|.

(⇐): Obviously,

1=T((1,1),(1,1))=T((1,-1),(1,-1))=T((1,1),(1,-1))=T((1,-1),(1,1)).

Hence, 1 = |a + b + c + d| = |a + bcd| = |abc + d| = |ab + cd|, so 1 = a + b + c + d = ab + cd, so c = 1− a, d = −b. Since 1 = |a + bcd| = |abc+d|, 1 = |(2a−1)+2b| = |(2a−1)−2b|. Hence, (2a−1)2b = 0, so a=12 or b = 0. If a=12, then b=±12 and T=(12,12,12,-12) or T=(12,-12,12,12). Note that (12,12,12,-12) and T=(12,-12,12,12) are extreme. If b = 0, then a = 0 or a = 1, so T = (0, 0, 1, 0) or T = (1, 0, 0, 0). Note that (0, 0, 1, 0) and (1, 0, 0, 0) are extreme. By Theorem 2.2, (-12,12,12,12),(12,12,-12,12), (0, 1, 0, 0), (0, 0, 0, 1) are extreme.

(⇒): Suppose that TextBL(l22) and the necessary condition is not true.

Then

Norm(T){((1,1),(1,1)),((1,-1),(1,-1)),((1,1),(1,-1))}   orNorm(T){((1,1),(1,1)),((1,-1),(1,-1)),((1,-1),(1,1))}   orNorm(T){((1,1),(1,1)),((1,1),(1,-1)),((1,-1),(1,1))}   orNorm(T){((1,-1),(1,-1)),((1,1),(1,-1)),((1,-1),(1,1))}.

Without loss of generality we may assume that

Norm(T)={((1,1),(1,1)),((1,-1),(1,-1)),((1,1),(1,-1))}   orNorm(T)={((1,1),(1,1)),((1,-1),(1,-1)),((1,-1),(1,1))}   orNorm(T)={((1,1),(1,1)),((1,1),(1,-1)),((1,-1),(1,1))}   orNorm(T)={((1,-1),(1,-1)),((1,1),(1,-1)),((1,-1),(1,1))}.

Case 1: Norm(T) = {((1, 1), (1, 1)), ((1,−1), (1,−1)), ((1, 1), (1,−1))}

Let T1=(a-1n,b+1n,c-1n,d+1n) and T2=(a+1n,b-1n,c+1n,d-1n) for a sufficiently large n ∈ ℕ so that ||Ti|| = 1 for i = 1, 2. Therefore, T is not extreme, which is a contradiction.

Case 2: Norm(T) = {((1, 1), (1, 1)), ((1,−1), (1,−1)), ((1,−1), (1, 1))}

Let T1=(a+1n,b-1n,c-1n,d+1n) and T2=(a-1n,b+1n,c+1n,d-1n) for a sufficiently large n ∈ ℕ so that ||Ti|| = 1 for i = 1, 2. Therefore, T is not extreme, which is a contradiction.

Case 3: Norm(T) = {((1, 1), (1, 1)), ((1, 1), (1,−1)), ((1,−1), (1, 1))}

Let T1=(a-1n,b-1n,c+1n,d+1n) and T2=(a+1n,b+1n,c-1n,d-1n) for a sufficiently large n ∈ ℕ so that ||Ti|| = 1 for i = 1, 2. Therefore, T is not extreme, which is a contradiction.

Case 4: Norm(T) = {((1,−1), (1,−1)), ((1, 1), (1,−1)), ((1,−1), (1, 1))}

Let T1=(a+1n,b+1n,c+1n,d+1n) and T2=(a-1n,b-1n,c-1n,d-1n) for a sufficiently large n ∈ ℕ so that ||Ti|| = 1 for i = 1, 2. Therefore, T is not extreme, which is a contradiction.

Theorem 2.4

extBL(l22)={±(1,0,0,0),±(0,1,0,0),±(0,0,1,0),±(0,0,0,0,1),±12(-1,1,1,1),±12(1,-1,1,1),±12(1,1,-1,1),±12(1,1,1,-1)}.
Proof

It follows from the proof of Theorem 2.3.

Theorem 3.1

LetfL(l22)*and α = f(x1x2), β = f(y1y2), δ = f(x1y2), γ = f(x2y1). Then,

f=max{α,β,δ,γ,12α-β+δ+γ,12α+β-δ+γ,12α+β+δ-γ}.
Proof

It follows from Theorem 2.3 and the fact that

f=maxTextBL(l22)f(T).

Note that if ||f|| = 1, then |α| ≤ 1, |β| ≤ 1, |δ| ≤ 1, |γ| ≤ 1.

Theorem 3.2

([18, Theorem 2.3]). Let E be a real Banach space such that extBE is finite. Suppose that xextBE satisfies that there exists an fE* with f(x) = 1 = ||f|| and |f(y)| < 1 for every yextBE{±x}. Then, xexpBE.

Theorem 3.3

LetT((x1,y1),(x2,y2))=(a,b,c,d)L(l22). Then, the followings are equivalent:

  • (1) T is exposed;

  • (2) (b, a, d, c) is exposed;

  • (3) (a,−b,−c, d) is exposed;

  • (4) (c, d, a, b) is exposed;

  • (5) (c, d, b, a) is exposed.

Proof

It follows from Theorem 2.1 and the above remark of Theorem 2.1.

Now we are in position to describe all the exposed points of the unit ball of L(l22).

Theorem 3.4. expBL(l22)=extBL(l22)

Proof

It is enough to show that extBL(l22)expBL(l22).

Claim: T = (1, 0, 0, 0) is exposed.

Let fL(l22)* be such that α = 1,0 = β = δ = γ. Then f(T) = 1, |f(S)| < 1 for every SextBL(l22){±T}. By Theorem 3.2, T is exposed. By Theorem 3.3, (0, 1, 0, 0), (0, 0, 1, 0), (0, 0, 0, 1) are exposed.

Claim: T=12(-1,1,1,1) is exposed.

Let fL(l22)* be such that α = 1,0 = β, δ=γ=12. Then f(T) = 1, |f(S)| < 1 for every SextBL(l22){±T}. By Theorem 3.2, T is exposed. By Theorem 3.3, 12(1,-1,1,1),12(1,1,-1,1),12(1,1,1,-1) are exposed.

Theorem 4.1

LetT((x1,y1),(x2,y2))=(a,b,c,d)L(l22). Then, the followings are equivalent:

  • (1) T is smooth;

  • (2) (b, a, d, c) is smooth;

  • (3) (a,−b,−c, d) is smooth;

  • (4) (c, d, a, b) is smooth;

  • (5) (c, d, b, a) is smooth.

Proof

It follows from Theorem 2.1 and the above remark of Theorem 2.1.

Theorem 4.2

LetT((x1,y1),(x2,y2))=(a,b,c,d)L(l22). Then,TsmBL(l22)if and only if (0 < |a + b| < 1, |a + b| + |c + d| = 1, |ab| + |cd| < 1) or (0 < |ab| < 1, |ab| + |cd| = 1, |a + b| + |c + d| < 1).

Proof

By Theorem 4.1, we may assume that a ≥ |b| and c ≥ |d|.

(⇒): Suppose that T is smooth. For (u1,v1),(u2,v2)Sl2, let δ(u1,v1),(u2,v2)L(l22)* such that

δ(u1,v1),(u2,v2)(L) = L((u1, v1), (u2, v2)) for LL(l22). Then ||δ(u1,v1),(u2,v2)|| ≤ 1. Note that, by Theorem 3.1, 1 = ||δ(1,1),(1,±1)|| = ||δ(1,−1),(−1,1)|| = ||δ(−1,1),(1,1)||.

Obviously,

δ(1,1),(1,1)(T)=(a+b)+(c+d),   δ(1,-1),(-1,1)(T)=(a+b)-(c+d),δ(1,1),(1,-1)(T)=(a-b)-(c-d),   δ(-1,1),(1,1)(T)=(a-b)+(c-d).

Hence, if TsmBL(l22), then ||T|| = 1, so, by Theorem 2.1, (|a+b|+|c+d| = 1, |ab| + |cd| < 1) or (|ab| + |cd| = 1, |a + b| + |c + d| < 1). Suppose that |a+b|+|c+d| = 1, |ab|+|cd| < 1. We will show that 0 < |a+b| < 1. Otherwise. Then a + b = 1 or a + b = 0. Suppose that a + b = 1. Let f1L(l22)* such that f1(x1x2) = 1 = f1(y1y2), f1(x1y2) = 0 = f1(x2y1), and let f2L(l22)* such that 1 = f2(x1x2) = f2(y1y2) = f2(x1y2) = f2(x2y1). Since f1f2,1 = ||fj|| = fj(T) for j = 1, 2, T is not smooth, which is a contradiction. Suppose that a + b = 0. Then c + d = 1. Let g1L(l22)* such that g1(x1x2) = 0 = g1(y1y2), g1(x1y2) = 1 = g1(x2y1). Since g1f2,1 = ||g1|| = g1(T), T is not smooth, which is a contradiction. Therefore, if |a + b| + |c + d| = 1, |ab| + |cd| < 1, then 0 < |a + b| < 1. Similarly as the case that |a + b| + |c + d| = 1, |ab| + |cd| < 1, if |ab| + |cd| = 1, |a + b| + |c + d| < 1, then we should have 0 < |ab| < 1.

(⇐): Let fL(l22)* such that 1 = ||f|| = f(T) with α = f(x1x2), β = f(y1y2), δ = f(x1y2), γ = f(x2y1).

Case 1: 0 < |a + b| < 1, |a + b| + |c + d| = 1, |ab| + |cd| < 1

By Theorem 2.1, ||T|| = 1. Note that

a+b+c+d=1=aα+bβ+cδ+dγ.

By Theorem 2.1, it follows that, for a sufficiently large n ∈ ℕ,

1ax1x2+(b±1n)y1y2+(c1n)x1y2+dx2y11ax1x2+(b±1n)y1y2+cx1y2+(d1n)x2y11(a±1n)x1x2+by1y2+(c1n)x1y2+dx2y1

From (**),1f(ax1x2+(b±1n)y1y2+(c1n)x1y2+dx2y1)=1+1nβ-δ, hence β = δ and 1f(ax1x2+(b±1n)y1y2+cx1y2+(d1n)x2y1)=1+1nβ-γ, hence β = γ. 1f((a±1n)x1x2+by1y2+(c1n)x1y2+dx2y1)=1+1nα-δ, hence α = δ. Therefore, by (*), 1 = α = β = δ = γ, hence f is uniquely determined. Therefore, T is smooth.

Case 2: 0 < |ab| < 1, |ab| + |cd| = 1, |a + b| + |c + d| < 1

By Case 1 and Theorem 4.1, T is smooth.

Remark

Nine months after the acceptance of this paper in Kyungpook Mathematical Journal, the author found that W. Cavalcante and D. Pellegrino had proved in [2], independently, Theorems 2.1, 2.4 and 3.4.

  1. Aron, RM, Choi, YS, Kim, SG, and Maestre, M (2001). Local properties of polynomials on a Banach space. Illinois J Math. 45, 25-39.
  2. Cavalcante, W, and Pellegrino, D (). Geometry of the closed unit ball of the space of bilinear forms on .
  3. Choi, YS, Ki, H, and Kim, SG (1998). Extreme polynomials and multilinear forms on l1. J Math Anal Appl. 228, 467-482.
    CrossRef
  4. Choi, YS, and Kim, SG (1998). The unit ball of. Arch Math (Basel). 71, 472-480.
    CrossRef
  5. Choi, YS, and Kim, SG (1998). Extreme polynomials on c0. Indian J Pure Appl Math. 29, 983-989.
  6. Choi, YS, and Kim, SG (1999). Smooth points of the unit ball of the space ℘(2l1). Results Math. 36, 26-33.
    CrossRef
  7. Choi, YS, and Kim, SG (2004). Exposed points of the unit balls of the spaces. Indian J Pure Appl Math. 35, 37-41.
  8. Dineen, S (1999). Complex Analysis on Infinite Dimensional Spaces. London: Springer-Verlag
    CrossRef
  9. Dineen, S (2003). Extreme integral polynomials on a complex Banach space. Math Scand. 92, 129-140.
    CrossRef
  10. Grecu, BC (2002). Geometry of 2-homogeneous polynomials on lp spaces, 1 < p < ∞. J Math Anal Appl. 273, 262-282.
    CrossRef
  11. Grecu, BC, Munoz-Fernandez, GA, and Seoane-Sepulveda, JB (2009). Unconditional constants and polynomial inequalities. J Approx Theory. 161, 706-722.
    CrossRef
  12. Kim, SG (2007). Exposed 2-homogeneous polynomials on. Math Proc R Ir Acad. 107A, 123-129.
  13. Kim, SG (2009). The unit ball of. Extracta Math. 24, 17-29.
  14. Kim, SG (2011). The unit ball of ℘(2d*(1, w)2). Math Proc R Ir Acad. 111A, 79-94.
  15. Kim, SG (2013). The unit ball of ℒs(2d*(1, w)2). Kyungpook Mathematical Journal. 53, 295-306.
    CrossRef
  16. Kim, SG (2013). Smooth polynomials of ℘(2d*(1, w)2). Math Proc R Ir Acad. 113A, 45-58.
    CrossRef
  17. Kim, SG (2013). Extreme bilinear forms of ℒ(2d*(1, w)2). Kyungpook Mathematical Journal. 53, 625-638.
    CrossRef
  18. Kim, SG (2014). Exposed symmetric bilinear forms of ℒs(2d*(1, w)2). Kyungpook Mathematical Journal. 54, 341-347.
    CrossRef
  19. Kim, SG (2015). Exposed bilinear forms of ℒ(2d*(1, w)2). Kyungpook Mathematical Journal. 55, 119-126.
    CrossRef
  20. Kim, SG (2015). Exposed 2-homogeneous polynomials on the 2-dimensional real predual of Lorentz sequence space. Mediterr J Math. 7, 1-13.
  21. Kim, SG, and Lee, SH (2003). Exposed 2-homogeneous polynomials on Hilbert spaces. Proc Amer Math Soc. 131, 449-453.
    CrossRef
  22. Lee, J, and Rim, KS (2005). Properties of symmetric matrices. J Math Anal Appl. 305, 219-226.
    CrossRef
  23. Munoz-Fernandez, GA, Revesz, S, and Seoane-Sepulveda, JB (2009). Geometry of homogeneous polynomials on non symmetric convex bodies. Math Scand. 105, 147-160.
    CrossRef
  24. Munoz-Fernandez, GA, and Seoane-Sepulveda, JB (2008). Geometry of Banach spaces of trinomials. J Math Anal Appl. 340, 1069-1087.
    CrossRef
  25. Ryan, RA, and Turett, B (1998). Geometry of spaces of polynomials. J Math Anal Appl. 221, 698-711.
    CrossRef