EXPOSED 2-HOMOGENEOUS POLYNOMIALS ON HILBERT ...

5 downloads 0 Views 256KB Size Report
May 17, 2002 - ous polynomials on a separable real Hilbert space is its exposed point ... We recall that a unit vector x in a real Banach space E is exposed if ...
PROCEEDINGS OF THE AMERICAN MATHEMATICAL SOCIETY Volume 131, Number 2, Pages 449–453 S 0002-9939(02)06544-9 Article electronically published on May 17, 2002

EXPOSED 2-HOMOGENEOUS POLYNOMIALS ON HILBERT SPACES SUNG GUEN KIM AND SANG HUN LEE (Communicated by Jonathan M. Borwein)

Abstract. We show that every extreme point of the unit ball of 2-homogeneous polynomials on a separable real Hilbert space is its exposed point and that the unit ball of 2-homogeneous polynomials on a non-separable real Hilbert space contains no exposed points. We also show that the unit ball of 2homogeneous polynomials on any infinite dimensional real Hilbert space contains no strongly exposed points.

We recall that a unit vector x in a real Banach space E is exposed if there is a unit vector f ∈ E ∗ so that f (x) = 1 and f (y) < 1 for each y ∈ BE with y 6= x, where BE is the closed unit ball of E. It is easy to see that every exposed point of BE is an extreme point. A unit vector x in a real Banach space E is strongly exposed if there is a unit vector f ∈ E ∗ so that f (x) = 1 and given any sequence (xk ) in BE with f (xk ) → 1 we can conclude that xk → x in norm. It is easy to see that every strongly exposed point of BE is its exposed point. We denote by extBE , expBE and sexpBE the sets of extreme points, exposed points and strongly exposed points of BE , respectively. Let P(2 H) be the Banach space of continuous 2-homogeneous polynomials on a real Hilbert space H. Recently many authors (see [1]-[7]) studied extremal problems for polynomials on a Banach space. The extreme points of the unit ball of this space has been characterized by Grecu [6]. The object of this note is to determine the exposed and strongly exposed points of the unit ball of P(2 H). Grecu showed that for a real Hilbert space H, P ∈ extBP(2 H) if and only if there exists an orthogonal decomposition of H = H1 ⊕ H2 such that P (x) = kπ1 (x)k2 − kπ2 (x)k2 , where πj : H → Hj are the orthogonal projections of H onto Hj (j = 1, 2). Using this result we show the following results: (1) If H is a separable real Hilbert space, then every extreme point of the unit ball of P(2 H) is exposed. (2) If H is a non-separable real Hilbert space, then the unit ball of P(2 H) contains no exposed points. (3) If H is an infinite dimensional real Hilbert space, then the unit ball of P(2 H) contains no strongly exposed points. Received by the editors January 15, 2001 and, in revised form, September 10, 2001. 2000 Mathematics Subject Classification. Primary 46B20, 46E15. The first author wishes to acknowledge the financial support of the Korea Research Foundation (KRF-2000-015-DP0012). The second author wishes to acknowledge the financial support by KOSEF research No. (20011-10100-007). c

2002 American Mathematical Society

449

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

450

SUNG GUEN KIM AND SANG HUN LEE

Theorem 1. Let H be a separable real Hilbert space. Then every extreme point of the unit ball of P(2 H) is exposed. Proof. Since expBP(2 H) ⊂ extBP(2 H) , it suffices to show that if P ∈ extBP(2 H) , then P ∈ expBP(2 H) . Let P ∈ extBP(2 H) . By Theorem 1.6 of [6], P (x) = kπ1 (x)k2 − kπ2 (x)k2 where H = H1 ⊕ H2 and πj : H → Hj are the orthogonal projections of H onto Hj (j = 1, 2). Clearly kπj k = 1. Let {eα }α∈A and {tβ }β∈B be orthonormal bases of H1 and H2 , respectively. It is clear that {eα , tβ }α∈A,β∈B is an orthonormal basis of H. Then for each x ∈ H, we have X X hx, eα ieα + hx, tβ itβ x= α∈A

and P (x) =

β∈B

X

X

hx, eα i2 −

α∈A

2

hx, tβ i .

β∈B

Note that P (eα ) = 1 for all α ∈ A and P (tβ ) = −1 for all β ∈ B. Let {aα }α∈A and {bβ }β∈B be collections of reals such that aα > 0, bβ < 0 and X X aα − bβ = 1. α∈A

β∈B



Define f ∈ P(2 H) such that for each Q ∈ P(2 H), X X Q(eα )aα + Q(tβ )bβ . f (Q) = α∈A

β∈B

Then kf k = 1. Indeed, for each Q ∈ P( H) with kQk = 1, we have |Q(eα )| ≤ 1 for all α ∈ A, |Q(tβ )| ≤ 1 for all β ∈ B and X X X X |Q(eα )|aα + |Q(tβ )|(−bβ ) ≤ 1 · aα − bβ = 1 |f (Q)| ≤ 2

α∈A

and f (P ) =

X

α∈A

β∈B

P (eα )aα +

α∈A

X

P (tβ )bβ =

β∈B

X

1 · aα −

α∈A

1·β∈B

X

1 · bβ = 1.

β∈B

We will show that this functional f exposes the polynomial P. Let Q ∈ P(2 H) be such that f (Q) = 1 = kQk. We claim that Q(eα ) = 1 for all α ∈ A and Q(tβ ) = −1 for all β ∈ B. Indeed, X X X X Q(eα )aα + Q(tβ )bβ = 1 · aα − 1 · bβ = 1, 1 = f (Q) = α∈A

β∈B

α∈A

β∈B

so Q(eα ) = 1 for all α ∈ A and Q(tβ ) = −1 for all β ∈ B because of aα > 0, bβ < 0. ˆ be the corresponding continuous symmetric bilinear form to the polynoLet Q mial Q. We claim: ˆ α , eα0 ) = 0 (α 6= α0 ∈ A); (1) Q(e ˆ β , tβ 0 ) = 0 (β 6= β 0 ∈ B); (2) Q(t ˆ (3) Q(eα , tβ ) = 0 (α ∈ A, β ∈ B).

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

EXPOSED 2-HOMOGENEOUS POLYNOMIALS ON HILBERT SPACES

451

Proof of (1). It follows that 1 = kQk ≥

sup x2α +x2α0 =1

=

sup x2α +x2α0 =1

=

sup x2α +x2α0 =1

=

sup x2α +x2α0 =1

=

sup x2α +x2α0 =1

|Q(xα eα + xα0 eα0 )| ˆ α eα + xα0 eα0 , xα eα + xα0 eα0 )| |Q(x ˆ α , eα0 )xα xα0 | |Q(eα )x2α + Q(eα0 )x2α0 + 2Q(e ˆ α , eα0 )xα xα0 | |x2α + x2α0 + 2Q(e ˆ α , eα0 )xα xα0 | |1 + 2Q(e

ˆ α , eα0 ) = 0. which implies Q(e Proof of (2). By the similar proof of (1), we have 1 = kQk ≥

sup x2β +x2β 0 =1

|Q(xβ tβ + xβ 0 tβ 0 )| =

sup x2β +x2β 0 =1

ˆ β , tβ 0 )xβ xβ 0 | |1 + 2Q(t

ˆ β , tβ 0 ) = 0. which implies Q(t Proof of (3). By the similar proof of (1), we have 1 = kQk ≥ =

sup x2α +x2β =1

sup x2α +x2β =1

=

sup x2α +x2β =1

|Q(xα eα + xβ tβ )|

ˆ α , tβ )xα xβ | |Q(eα )x2α + Q(tβ )x2β + 2Q(e ˆ α , tβ )xα xβ |. |x2α − x2β + 2Q(e

ˆ α , tβ ) = 0. For x ∈ H, it follows that By Lemma 2.1 of [3], we have Q(e ˆ x) Q(x) = Q(x, X X X X ˆ hx, eα ieα + hx, tβ itβ , hx, eα ieα + hx, tβ itβ ) = Q( α∈A

=

X

=

X X

α∈A

X

β∈B

ˆ β , tβ 0 ) hx, tβ ihx, tβ 0 iQ(t

β,β 0 ∈B

X

ˆ α , tβ ) hx, eα ihx, tβ iQ(e

α∈A,β∈B

Q(eα )hx, eα i2 +

α∈A

=

α∈A

ˆ α , eα0 ) + hx, eα ihx, eα0 iQ(e

α,α0 ∈A

+2

β∈B

hx, eα i2 −

X

X

Q(tβ )hx, tβ i

2

(by claims (1)-(3))

β∈B 2

hx, tβ i = P (x),

β∈B

showing that f exposes P. Therefore P ∈ expBP(2 H) . Theorem 2. Let H be an infinite dimensional real Hilbert space. Then the unit ball of P(2 H) contains no strongly exposed points. Proof. Since sexpBP(2 H) ⊂ extBP(2 H) , it suffices to show that sexpBP(2 H) ∩ extBP(2 H) = ∅.

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

452

SUNG GUEN KIM AND SANG HUN LEE

Let P ∈ extBP(2 H) . By Theorem 1.6 of [6] P (x) = kπ1 (x)k2 − kπ2 (x)k2 where H = H1 ⊕ H2 and πj : H → Hj are the orthogonal projections of H onto Hj (j = 1, 2). Clearly kπj k = 1. Without loss of generality, assume that dim(H1 ) = ∞. ∞ Let {eα }α∈A be an orthonormal P basis of H1 . Let {αj }j=1 ⊂ A. It is clear that for each x ∈ H, we have π1 (x) = α∈A hπ1 (x), eα ieα . Then X hπ1 (x), eα i2 − kπ2 (x)k2 . P (x) = α∈A ∗

Suppose that P ∈ sexpBP(2 H) . Then there is an f ∈ P(2 H) such that kf k = 1 = f (P ) and given any sequence {Pj } in BP(2 H) with f (Pj ) → 1, we have kPj −P k → 0. For each αj , we have X hπ1 (x), eα i2 − hπ1 (x), eαj i2 − kπ2 (x)k2 ) < 1 f( α6=αj

= f(

X

hπ1 (x), eα i2 − kπ2 (x)k2 ),

α∈A

so f (hπ1 (x), eαj i ) > 0. We will show f (hπ1 (x), eαj i2 ) → 0 as j → ∞. For each n, X X f (hπ1 (x), eαj i2 ) = f ( hπ1 (x), eαj i2 ) 2

1≤j≤n

≤ kf kk

X

1≤j≤n

1≤j≤n

=

sup

X

hπ1 (x), eαj i k = k 2

X

kxk=1 1≤j≤n

hπ1 (x), eαj i2 k

1≤j≤n

hπ1 (x), eαj i = sup kπ1 (x)k2 = kπ1 k2 = 1. 2

kxk=1

P Thus 1≤j 0 for every α ∈ A and similarly, f (hx, eα i2 ) < 0 for every α ∈ B. But X X 2 2 f (hx, tα i ) = f ( hx, tα i ) ≤ 1, α∈A

α∈A

and hence A must be countable. Similarly, B is countable. Thus H is separable.

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

EXPOSED 2-HOMOGENEOUS POLYNOMIALS ON HILBERT SPACES

453

Acknowledgement The authors wish to thank the referee for pointing out Theorem 3 and several useful remarks. References [1] C. Boyd and R.A. Ryan, Geometric theory of spaces of integral polynomials and symmetric tensor products, J. Funct. Anal. 179 (2001), 18-42. MR 2002b:46029 [2] Y.S. Choi, H. Ki and S.G. Kim, Extreme polynomials and multilinear forms on l1 , J. Math. Anal. Appl. 228 (1998), 467-482. MR 99k:46077b [3] Y.S. Choi and S.G. Kim, The unit ball of P(2 l22 ), Arch. Math. 71 (1998), 472-480. MR 2000c:46085 [4] Y.S. Choi and S.G. Kim, Exposed points of the unit balls of P(2 lp2 ) (p = 1, 2, ∞), (Preprint). [5] S. Dineen, Extreme integral polynomials on a complex Banach space, (Preprint). [6] B.C. Grecu, Extreme polynomials on Hilbert spaces, (Preprint). [7] R.A. Ryan and B. Turett, Geometry of spaces of polynomials, J. Math. Anal. Appl. 221 (1998), 698-711. MR 99g:46015 Department of Mathematics, Kyungpook National University, Daegu, Korea (702701) E-mail address: [email protected] Department of Mathematics, Kyungpook National University, Daegu, Korea (702701) E-mail address: [email protected]

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use