![Arxiv:1802.06227V2 [Math.FA] 11 Oct 2018 Ibr Pc R Omprle Fadol Fte R Ierydepend Linearly Are They If E Only Two and Furthermore, If Space](http://data.docslib.org/img/a2a4bae89ca6e1146fb70c3e5fce6963-1.webp)
arXiv:1802.06227v2 [math.FA] 11 Oct 2018 ibr pc r omprle fadol fte r ierydepend linearly are they if e only two and Furthermore, if space. norm-parallel Hilbert are a space is Hilbert space prod a underlying inner the with coincides if orthogonality onality Birkhoff-James of notion The o some for fPoesrGdda ir.D.Dbay anfeseae oack (R Tr MATRICS to India. project elated Mrs. of by and supported feels Govt. Chowdhury is DST, Sain Partha author SERB, third Debmalya Mr. the of couple, research Dr. lovable the of Misra. Fellopresence Postdoctoral Gadadhar Kothari S. Professor D. of Dr. by sponsored is Sain Debmalya x otx faBnc pc,BrhffJmsotooaiyi homog is orthogonality [ Birkhoff-James symmetric space, Banach a of context B pc vrtefield the over space cmat ieroeaosfrom operators linear (compact) of sphere K iko-ae [ Birkhoff-James X ( ssi ob omprle [ norm-parallel be to said is eerho h rtato sspotdb G,Gv.o India. of Govt. UGC, by supported is author first the of Research e od n phrases. and words Key 2010 ∗ Date nti ae,letters paper, this In X orsodn author. Corresponding = )) NSM EMTI RPRISO OPERATOR OF PROPERTIES GEOMETRIC SOME ON { , eevd xxx eie:yyy;Acpe:zzzzzz. Accepted: yyyyyy; Revised: xxxxxx; Received: : ahmtc ujc Classification. Subject Mathematics td eirtn prtr n rv that spaces) prove convex and spa (strictly operators (semi-rotund points semi-rotund exposed points study of semi-rotund notion operat of the linear concept generalizes bounded the as ch well the introduce as orthogona Using operators we Birkhoff-James linear study space. compact we Hilbert of space parallelism, a operator on of defined ization operators linear bounded of z aalls ftobuddlna prtr ewe omdlinear normed between operators linear bounded X two of parallelism ize opeeycaatrz aalls ftocmatlna operators linear compact spaces two linear ope of normed linear parallelism bounded characterize of space completely the in semi-rotundity and orthogonality Abstract. Y hc sntsrcl ovx if convex, strictly not is which x if and ssrcl convex. strictly is ∈ λ X X ∈ X 9 epciey Let respectively. = ,weesnr-aalls sbt ymti and symmetric both is norm-parallelism whereas ], Y : T RIAMAL, ARPITA . k where , Y eivsiaeprleimadapoiaeprleimi h space the in parallelism approximate and parallelism investigate We x 4 For . ≤ k ,witnas written ], nti ae esuysm emti rprislk parallelism, like properties geometric some study we paper this In K 1 } T X ,y x, omprleim rhgnlt,sm-oud omattainment. norm semi-rotund, orthogonality, parallelism, norm { ∈ and , = X Y 1 and ∈ { C eoenre iersae and spaces linear normed denote EMLASAIN, DEBMALYA S λ , B 17 x X X 1. ∈ ( Y R , = X ⊥ to ] assuming , K } X Introduction. x , X B { Also . Y , SPACES ssi ob rhgnlto orthogonal be to said is x to : y Y y ( ) if , | ∈ rte as written , λ rmr 62;Scnay47L05. Secondary 46B20; Primary r nt-iesoa aahsae and spaces Banach finite-dimensional are Y K | X 1 . 1 = k ( X x : ewrite We X X k ≤ k k , ∗ } x ob eeie eas character- also We reflexive. be to Y ewudlk ont hti the in that note to like would We . k 2 eoe h ulsaeof space dual the denotes )dnt h pc falbounded all of space the denote )) B n ALLPAUL KALLOL and 1 = ( x X x , + Y } k B λy sasm-oudspace semi-rotund a is ) eteui aladteunit the and ball unit the be ( y X k if , , f o MTR/2017/000059), no. ef. o all for si,udrtementorship the under wship, Y k = ) n a,i i ie The life! his in Das, ina x H + efurther We . oldetejoyous the nowledge λ R B h eerho Dr. of research The r.Finally, ors. eoe Hilbert a denotes y λy aos We rators. e)which ces) ( − iyi the in lity ∈ 1 X ntesneof sense the in nosbtnot but eneous ∗ k between homogeneous. aracter- ( ) K. = spaces c orthog- uct K eet of lements k Moreover, x ( X k X n.In ent. , + Let . Y k = ) y k 2 ARPITAMAL,DEBMALYASAINandKALLOLPAUL case of normed linear spaces, two linearly dependent vectors are norm-parallel, but the converse is not true in general. In ℓ2 , (1, 1) and (1, 0) are norm-parallel but linearly independent. In [13], Paul et∞ al. studied the notion of strong or- thogonality. For x, y X, x is said to be strongly orthogonal to y in the sense of ∈ Birkhoff-James, written as x SB y, if x < x + λy for all λ K 0 . Birkhoff-James orthogonality⊥ plays a veryk k importantk k role in the∈ stud\{y of} geom- etry of Banach spaces. It has been explored in various settings by many other mathematicians [1, 2, 3, 11, 18]. Furthermore, various generalizations of it have also been considered. Dragomir [8] defined approximate Birkhoff-James orthogo- nality as follows: Let ǫ [0, 1) and x, y X. Then x is said to be approximate Birkhoff-James orthogonal∈ to y if x + ∈λy (1 ǫ) x λ K. Later on, Chmieli´nskik [6] slightlyk ≥ − modifiedk k ∀ the∈ definition given by Dragomir and defined approximate Birkhoff-James orthogonality as follows:
x ǫ y x + λy √1 ǫ2 x λ K. ⊥D ⇐⇒ k k ≥ − k k ∀ ∈ Motivated by the notion of approximate Birkhoff-James orthogonality, Moslehian and Zamani [12] introduced the notion of approximate parallelism(ǫ-parallelism) in the setting of normed linear space. For x, y X and ǫ [0, 1), x is said to be approximate parallel to y, written as, x ǫ y, if∈ inf x +∈µy : µ K ǫ x . We would like to remark here that in generalk in a{k normedk linear∈ space,} ≤x k ǫ ky with ǫ = 0 implies that x y but not the other way round. As for example,k in ℓ2 , (1, 1) (1, 0) but (1, 1)k ǫ (1, 0) with ǫ =0. ∞ k 6k We will see later that the norm attainment set of a bounded linear operator plays an important role in the study of norm-parallelism and Birkhoff-James orthogonality of bounded linear operators between Banach spaces. For a bounded linear operator T B(X, Y), we define MT to be the set of all unit vectors in SX at which T attains∈ norm, i.e.,
M = x S : T x = T . T { ∈ X k k k k} The purpose of this paper is to study norm-parallelism and Birkhoff-James orthogonality in the space of bounded linear operators, from the point of view of operator norm attainment. Bottazzi et al.[5, Th. 4.24] proved that if X is a reflexive Banach space, Y is a smooth, strictly convex Banach space, T, A K(X, Y) and M is either connected or M = u for some unit vector u X∈, T T {± } ∈ then T A if and only if there exists a vector x MT MA such that T x Ax. In this paper,k we substantially improve upon the∈ result∩ to show that therek is no necessity to put additional restrictions on the norm attainment set MT and the codomain space Y. We next give a complete characterization of norm-parallelism of bounded linear operators defined between any two normed linear spaces. We further study and completely characterize norm parallelism in the space of bounded linear operators on a Hilbert space H. We also obtain a necessary condi- tion for approximate parallelism in B(H) and provide an example to illustrate the subtle difference between norm parallelism of operators (T A) and approximate k ON SOME GEOMETRIC PROPERTIES OF OPERATOR SPACES 3 operator parallelism (T ǫ A, with ǫ = 0). k In section 3, we explore Birkhoff-James orthogonality in the space of bounded linear operators by classifying them into two exclusive and exhaustive categories. We first study the case “T SB A” and obtain a characterization of strong Birkhoff-James orthogonality⊥ in the space of bounded linear operators between finite-dimensional normed linear spaces. We next study the case “T A but ⊥B T SB A” and obtain a necessary condition for the same. 6⊥Motivated by the operator theoretic results involving Birkhoff-James orthog- onality and strong Birkhoff-James orthogonality, we introduce a new geometric notion, that of a semi-rotund point, defined in the following way: Definition 1.1 (semi-rotund point). Let X be a normed linear space. An element θ = x X is said to be a semi-rotund point of X if there exists y X such that x 6 ∈y. ∈ ⊥SB Definition 1.2. A normed linear space X is said to be a semi-rotund space if for each non-zero x X, x is a semi-rotund point. ∈ Clearly, every exposed point of the unit ball of a normed linear space is a semi- rotund point. However, it is interesting to observe that the converse is not true if the dimension of the space is greater than two. In ℓ3 , the point (1, 1, 0) is a semi- rotund point but not an exposed point of the unit ball.∞ It is also immediate that a strictly convex space is a semi-rotund space but the converse is not necessarily 3 2 2 true. For example, consider the space X, where SX = (x, y, z) R : x + y = 1, z 1 (x, y, z) R3 : x2 + y2 +(z 1)2 = 1, {1 z ∈2 (x, y, z) R3| :|x ≤2 +}∪{y2 +(z + 1)∈2 = 1, 2 z −1 . In this case,≤ ≤ it is}∪{ easy to verify∈ that every non-zero element in−X ≤is a semi-rotund≤ − } point but not every point of SX is an exposed point of BX. We note that the notions of exposed point (strictly convex space) and semi-rotund point (semi-rotund space) are equivalent if the space is two-dimensional. Continuing our study of semi-rotund points, we prove that every non-zero compact linear operator from a reflexive Banach space to a strictly convex Banach space is a semi-rotund point of the corresponding operator space. Finally, we show that B(X, Y) is a semi-rotund space which is not strictly convex, if X and Y are finite-dimensional Banach spaces and in addition, Y is strictly convex. In particular, this illustrates that the concept of semi-rotundity is a proper generalization of the concept of strict convexity. It is well-known that several convexity conditions such as uniform convexity and local uniform rotundity, strictly stronger than that of strict convexity, are of great importance in the study of the geometry of normed linear spaces. We would like to end this section with the remark that to the best of our knowledge, this is the first instance of a convexity property in normed linear spaces which is strictly weaker than strict convexity.
2. Norm-parallelism of bounded linear operators We begin this section with an easy proposition on approximate parallelism. 4 ARPITAMAL,DEBMALYASAINandKALLOLPAUL
Proposition 2.1. Let X, Y be two normed linear spaces. Let T B(X, Y) and x M . Then for any ǫ [0, 1) and any y X, x ǫ y = T x ǫ T∈ y. ∈ T ∈ ∈ k ⇒ k Proof. Let x ǫ y. Then inf x + λy : λ K ǫ x . Thus, inf T x + λTy : λ Kk T inf x{k+ λy :k λ K ∈ ǫ}T ≤ xk k= ǫ T x . Therefore,{k T x kǫ Ty. ∈ }≤k k {k k ∈ } ≤ k kk k k k k In [10, Th. 2.1], James characterized Birkhoff-James orthogonality in terms of linear functionals. We next state a lemma that characterizes approximate parallelism in terms of linear functionals. We would like to remark that the lemma follows from a slight variation of [7, Cor. 6.8], and therefore its proof is omitted. Lemma 2.2. Let X be a normed linear space. Let x, y X and d = inf x + λy : λ K . Then for any ǫ [0, 1), x ǫ y if and only∈ if there exists a{k linear k ∈ } ∈ k functional f S ∗ such that f(x)= d ǫ x and f(y)=0. ∈ X ≤ k k Now, we obtain a complete characterization of norm-parallelism of compact linear operators defined on a reflexive Banach space, which substantially improves on [5, Th. 4.24]. Theorem 2.3. Let X be a reflexive Banach space and Y be any normed linear space. Let T, A K(X, Y). Then T A if and only if there exists x MT MA such that T x Ax∈. k ∈ ∩ k Proof. First we prove the necessary part of the theorem. Let T A. Then there exists λ T such that T + λA = T + A . The operator Tk + λA, being a compact∈ operator on a reflexivek Banachk k k space,k k attains its norm. Therefore, there exists x S such that T + λA = (T + λA)x . Thus, ∈ X k k k k T + A = T + λA k k k k k k = (T + λA)x k k T x + Ax ≤ k k k k T x + A ≤ k k k k T + A . ≤ k k k k This implies that T x + λAx = T x + Ax and T x = T . Similarly, k k k k k k k k k k Ax = A . Therefore, x MT MA and T x Ax. This completes the proof ofk thek necessaryk k part of the∈ theorem.∩ k For the sufficient part, suppose that there exists x MT MA such that T x Ax. Then there exists λ T such that T x + λAx =∈T x ∩+ Ax . Therefore,k ∈ k k k k k k T + A T + λA (T + λA)x = T x + Ax = T + A . k k k k≥k k≥k k k k k k k k k k Thus, T + λA = T + A , i.e., T A. This completes the proof of the theorem.k k k k k k k
We make note of the following remark that will be needed later in Theorem 3.6. ON SOME GEOMETRIC PROPERTIES OF OPERATOR SPACES 5
Remark 2.4. From the proof of Theorem 2.3, it is clear that T +λA = T + A k k k k k k for λ T if and only if there exists x MT MA such that T x + λAx = T x +∈ Ax . ∈ ∩ k k k k k k We next give an example to show that the compactness of T, A in Theorem 2.3 is essential. Example 2.5. Consider the right shift operator T : ℓ ℓ defined by T (x , x , 2 −→ 2 1 2 x3,...) = (0, x1, x2, x3,...). Let I be the identity operator on ℓ2. Then T = I = 1. Consider y = 1 (1, 1,..., 1, 0, 0,...) S for each n N.k Thenk k k n √n ∈ ℓ2 ∈ n 2 1 | {z } 2 1 (T + I)yn = n (1, 2, 2,..., 2, 1, 0, 0,...) = n [2 + 4(n 1)] 4. Thus k2 T + Ik Tk + I = 2 T + I k= T + I , i.e.,− T −→I. We claim ≤ k k≤k k k k ⇒ k k k k k k k that there does not exist any x MT MI such that T x Ix. For, if T x Ix for some x = (x , x , x ,...) ∈M ∩M then there existsk λ T suchk that 1 2 3 ∈ T ∩ I ∈ T x + λIx = T x + Ix and so T x Ix = Re λ T x, Ix λ T x, Ix = k k k k k k k kk k { h i}≤| h i| T x, Ix T x Ix . Then T x, Ix = T x Ix = T I , since x MT . 1 |h i| ≤ k kk k |h i| 2 k kk2 k 2k kk k2 2 ∈ 2 Then 0x1 + x1x2 + x2x3 + ... =1=(0 + x1 + x2 + ...) ( x1 + x2 + | 1 | | | | | | | | | x 2 + ...) 2 . Thus, by the equality condition of Cauchy-Schwarz inequality, we | 3| have, (0, x1, x2, x3,...) = λ(x1, x2, x3,...) for some λ K, which implies that x = 0, a contradiction. ∈ In the next theorem, we characterize norm-parallelism of bounded linear oper- ators between any two normed linear spaces. Theorem 2.6. Let X, Y be two normed linear spaces and T, A B(X, Y). Then T A if and only if there exists a sequence x in S such that∈ k { n} X lim T xn = T , lim Axn = A n n →∞ k k k k →∞ k k k k and
lim T xn + λAxn = T + A n →∞ k k k k k k for some λ T. ∈ Proof. First we prove the necessary part of the theorem. Let T A. Then there exists λ T such that T + λA = T + A . Now, there existsk a sequence ∈ k k k k k k xn in SX such that limn (T + λA)xn = T + λA . Since T xn and { Ax} are bounded sequences→∞ k of real numbers,k k withoutk loss of{k generalitk}y (if {k nk} necessary, passing onto a subsequence) we can assume that limn T xn and →∞ k k limn Axn exists. Therefore, →∞ k k T + A = T + λA k k k k k k = lim (T + λA)xn n →∞ k k lim T xn + lim Axn n n ≤ →∞ k k →∞ k k lim T xn + A n ≤ →∞ k k k k T + A . ≤ k k k k 6 ARPITAMAL,DEBMALYASAINandKALLOLPAUL
This implies that limn T xn + λAxn = limn T xn + limn Axn and →∞ k k →∞ k k →∞ k k limn T xn = T . Similarly, limn Axn = A . This completes the proof→∞ ofk the necessaryk k k part of the theorem.→∞ k k k k For the sufficient part of the theorem, assume that there exists a sequence x { n} in SX such that limn T xn = T , limn Axn = A and limn T xn + λAx = T + A →∞fork somek λ k Tk. Therefore,→∞ k k k k →∞ k nk k k k k ∈ T + A T + λA k k k k ≥ k k lim (T + λA)xn n ≥ →∞ k k = T + A . k k k k Thus, T + λA = T + A , i.e., T A. This completes the proof of the theorem.k k k k k k k Remark 2.7. We note that norm parallelism in operator space can also be char- acterized by [20, Th. 2.4] using the notion of Birkhoff-James orthogonality. We next give an easy characterization of strictly convex spaces in terms of norm parallelism. Theorem 2.8. A normed linear space X is strictly convex if and only if for any x, y X, x y x, y is linearly dependent. ∈ k ⇔{ } Proof. Let X be strictly convex. Let x, y be linearly dependent. Let y = αx. If { } α α =0, then clearly, x + y = x + y . Let α =0. Let λ = α . Then λ T and k k k k k k 6 | | ∈ x+λy = x+ α αx = x+ α x = x + y . Thus, x y. On the other hand, k k k α k k | | k k k k k k let x y. Then there| | exists λ T such that x + λy = x + y = x + λy . SincekX is strictly convex, x,∈ y is linearly dependent.k k k k k k k k k k Conversely, suppose that X{is not} strictly convex. Then there exists two linearly independent vectors x, y SX and t (0, 1) such that (1 t)x+ty =1= (1 t)x + ty . Therefore, it∈ follows from∈ the homogeneityk − propertyk of parallelismk − thatk x k y.k This completes the proof of the theorem. k In [5, Th. 4.14], Bottazzi et al. proved that if X is a locally uniformly convex m 1 m Banach space and A K(X) is such that A − = 0 and A = 0 for some m N then Ak Aj for every∈ 1 k 3 3 Example 2.12. Let X = ℓ1 and A, B B(ℓ1) be given by the following matrices (with respect to the standard ordered basis∈ of R3) 1 0 0 0 0 0 0 0 0 , 0 1 0 0 0 0 1 0 1 respectively. Clearly, A2 = A and B2 = B. It is easy to verify that (1, 0, 0) ∈ MA MB and A(1, 0, 0) B(1, 0, 0). Therefore, by Theorem 2.3, we have, A B. Clearly,∩ in this case, A(Xk) B(X)= (0, 0, 0) . k ∩ { } Next, we study approximate parallelism in the space of bounded linear opera- tors on an infinite-dimensional Hilbert space. Theorem 2.13. Let H be an infinite-dimensional Hilbert space. Let ǫ [0, 1) and T B(H). Then (i) (ii), where, ∈ (i) For∈ any A B(H), T⇒ǫ A there exists x M M such that T x ǫ Ax. ∈ k ⇔ ∈ T ∩ A k (ii) There exists a finite-dimensional subspace H0 of H such that MT = SH0 and T ⊥ < T . k kH0 k k Proof. Without loss of generality assume that T = 1. From [15, Th. 2.2], we k k have, MT = SH0 , where H0 is a subspace of H. We first show that H0 is finite- dimensional. If possible, suppose that H0 is infinite-dimensional. Then there exists a sequence e : n N of orthonormal vectors in H . Extend this sequence { n ∈ } 0 to a complete orthonormal basis = eα : α Λ N of H. For each eα H0 2 B2 { ∈ ⊇ } ∈ ∩B we have T ∗T = T = T e = T ∗T e , e T ∗T e e T ∗T , so k k k k k αk h α αi≤k αkk αk≤k k that by the equality condition of Schwarz’s inequality, we get T ∗T eα = λαeα for some scalar λα. Thus, T eα : eα H0 is a set of orthonormal vectors in H. Define A : H as{ follows: ∈ ∩B} B −→ 1 Ae = T e , n N n n2 n ∀ ∈ Ae = 0, e e : n N. α ∀ α ∈B\{ n ∈ } Since T eα : α H0 is orthonormal, A can be extended to a bounded linear { ∈ ∩B} λ operator on H. Now, for any λ K, T +λA (T +λA)en = 1+ n2 T en = λ ∈ k k≥k k | |k ǫk 1+ n2 1. Therefore, infλ K T + λA 1 > ǫ. This implies that T A. |But e | −→M M and clearly∈T ek ǫ Ae .k However, ≥ this clearly contradicts6k the 1 ∈ T ∩ A 1 k 1 hypothesis. Therefore, H0 must be finite-dimensional. Next, we show that T ⊥ < 1. If possible, suppose that T ⊥ = 1. Then k kH0 k kH0 there exists a sequence x in S ⊥ such that T x 1. Define A : H H { n} H0 k nk −→ −→ by Az = T x, where z = x + y, x H , y H⊥. It is easy to verify that ∈ 0 ∈ 0 A B(H). Now, for any λ K, T + λA (T + λA)xn = T xn 1. ∈ ∈ k k≥k ǫ k k k −→ Therefore, infλ K T + λA 1 > ǫ. This shows that T A but clearly, for any x M M , ∈T xk ǫ Ax. Oncek ≥ again, this contradicts the6k hypothesis. Therefore, ∈ T ∩ A k we must have, T ⊥ < 1. This establishes the theorem. k kH0 The following example shows that the condition (ii) of Theorem 2.13 does not imply condition (i) of Theorem 2.13, i.e., there exists T, A B(H) and a ∈ ON SOME GEOMETRIC PROPERTIES OF OPERATOR SPACES 9 finite-dimensional subspace H of H such that M = S and T ⊥ < T but 0 T H0 k kH0 k k T ǫ A T x ǫ Ax for some x M M does not hold. k ⇔ k ∈ T ∩ A Example 2.14. Consider T, A B(ℓ2) defined by ∈ x x x T (x , x , x ,...,x ,...) = (x , 2 , 3 ,..., n ,...), 1 2 3 n 1 2 2 2 A(x1, x2, x3,...,xn,...) = (x1, 0, 0,..., 0,...), respectively, where (x1, x2, x3,...,xn,...) ℓ2. It is easy to verify that MT = ∈ 1 1 S , where H = span (1, 0, 0,...) and T ⊥ = < T . Let ǫ [0, ). H0 0 { } k kH0 2 k k ∈ 2 Clearly, (1, 0, 0,...) M M and T (1, 0, 0,...) ǫ A(1, 0, 0,...). Now, since ∈ T ∩ A k A(0, 1, 0, 0,..., 0,...), (0, 1, 0, 0,..., 0,...) =0 h i and 1 T (0, 1, 0, 0,..., 0,...), (0, 1, 0, 0,..., 0,...) = , h i 2 1 sup Tξ,η : ξ = η =1, Aξ, η =0 2 > ǫ. Therefore, by [12, Th. 3.7], we have,{|h T i|ǫ Ak. k k k h i } ≥ 6k However, if we consider T A instead of T ǫ A (ǫ [0, 1)) then we have the following characterization. k k ∈ Theorem 2.15. Let H be an infinite-dimensional Hilbert space. Let T B(H). Then the following two conditions are equivalent. ∈ (i) For any A B(H), T A there exists x M M such that T x Ax. ∈ k ⇔ ∈ T ∩ A k (ii) There exists a finite-dimensional subspace H0 of H such that MT = SH0 and T ⊥ < T . k kH0 k k Proof. (ii) (i) follows from [20, Th. 2.18]. We only prove⇒ (i) (ii). Without loss of generality assume that T = 1. From ⇒ k k [15, Th. 2.2], we have, MT = SH0 , where H0 is a subspace of H. We first show that H0 is finite-dimensional. If possible, suppose that H0 is infinite-dimensional. Then there exists a sequence e : n N of orthonormal vectors in H . Extend { n ∈ } 0 this sequence to a complete orthonormal basis = eα : α Λ N of H. B2 { 2 ∈ ⊇ } For each e H we have T ∗T = T = T e = T ∗T e , e α ∈ 0 ∩ B k k k k k αk h α αi ≤ T ∗T e e T ∗T , so that by the equality condition of Schwarz’s inequality k αkk αk≤k k we get T ∗T eα = λαeα for some scalar λα. Thus, T eα : eα H0 is a set of orthonormal vectors in H. Let { ∈ ∩ B} Λ = α Λ: e (H ) e : n N , Λ = α Λ: e H . 1 { ∈ α ∈ 0 ∩ B \{ n ∈ }} 2 { ∈ α ∈B\ 0} If Λ = , then Λ can be well-ordered. Let α be the least element of Λ and for 1 6 ∅ 1 0 1 any α Λ1, s(α) be the successor of α. IfΛ1 has a greatest element say, β, then define ∈s(β)= α . Now, define A : H as follows: 0 B −→ Ae = T e , n N n n+1 ∀ ∈ Ae = T e , α Λ α s(α) ∀ ∈ 1 Ae = 0, α Λ . α ∀ ∈ 2 Since T eα : eα H0 is a set of orthonormal vectors, A can be extended to a bounded{ linear∈ operator∩B} on H. 10 ARPITAMAL,DEBMALYASAINandKALLOLPAUL Next, we show that T A but there does not exist any x M M such k ∈ T ∩ A that T x Ax. Let x = Σn N x, en en + Σα Λ1 x, eα eα + Σα Λ2 x, eα eα. Then k ∈ h i ∈ h i 2∈ h i 2 Ax = Σn N x, en T en+1+Σα Λ1 x, eα T es(α). Therefore, Ax = Σn N x, en + ∈ h 2 i 2 ∈ h i k k ∈ |h i| Σα Λ1 x, eα x . Thus, A 1. Again, Aen = T en+1 = 1 gives ∈ |h i| ≤ k k k 1 k ≤ k k k k that A = 1. Now, let xn = √n (e1 + e2 + ... + en). Then xn SH for all k k 1 ∈ n N. Now, (T + A)xn = √n (T e1 + 2T e2 + 2T e3 + ... + 2T en + T en+1). ∈ 2 1 Therefore, (T + A)xn = n [2 + 4(n 1)] = (T + A)xn 2. Thus, we have, 2 kT + A Tk + A =2=− T +⇒A k= T + Ak −→. This gives that T A. ≤k k≤k k k k ⇒k k k k k k k Next, let x MT MA. Let x = Σn N x, en en + Σα Λ1 x, eα eα. Let n0 be the ∈ ∩ ∈ h i ∈ h i 2 least positive integer such that x, en0 = 0. Then x =1= Σn∞=n0 x, en + 2 h i 6 k k ⇒ |h i| Σα Λ1 x, eα = 1. Now, ∈ |h i| T x, Ax |h i| = Σn∞=n0 x, en T en + Σα Λ1 x, eα T eα, Σn∞=n0 x, en T en+1 + Σα Λ1 x, eα T es(α) |h h i ∈ h i h i ∈ h i i| = Σn∞=n0 x, en x, en+1 + Σα Λ1 x, eα x, es(α) ∈ | h ih i h 1 ih i| 1 2 2 2 2 2 2 Σn∞=n0 x, en + Σα Λ1 x, eα Σn∞=n0+1 x, en + Σα Λ1 x, es(α) ≤ { |h i| ∈ |h i| } { |h i| ∈ |h i| } < 1= T x Ax . k kk k Thus, T A but there exists no x MT MA such that T x Ax. If Λ = kthen define A : H as∈ follows:∩ k 1 ∅ B −→ Ae = T e , n N n n+1 ∀ ∈ Ae = 0, α Λ . α ∀ ∈ 2 Proceeding as before, we can show that T A but there exists no x MT MA such that T x Ax. Therefore, H must bek finite-dimensional subspace∈ of H∩. k 0 Next, we show that T ⊥ < 1. If possible, suppose that T ⊥ = 1. Then k kH0 k kH0 there exists a sequence xn in S ⊥ such that T xn 1. Define A : H H { } H0 k k −→ −→ by Az = T y, where z = x + y, x H0, y H0⊥. Now, z = x + y =1= y 1. Thus, Az = Ty ∈1. Again,∈ Ax = kT xk k 1k gives that⇒ k k ≤ k k k k ≤ k nk k nk −→ A = 1. Now, (T + A)xn = 2 T xn 2. Therefore, 2 T + A kT k + A = 2 andk so T + Ak = Tk + kA −→, i.e., T A. From the≤ construction k k ≤ ofk kA itk followsk that Axk = 0,kif xk k Mk.k Since A kis non-zero it follows that ∈ T MT MA = and so there does not exist any x MT MA such that T x Ax. This∩ is in clear∅ contradiction with the hypothesis.∈ Therefore,∩ we mustk have, T ⊥ < 1. This establishes the theorem. k kH0 Remark 2.16. We note that T A does not imply T ǫ A with ǫ = 0. This difference between T A and T kǫ A with ǫ = 0 justifiesk the fact that conditions (i) and (ii) in Theoremk 2.13 arek not equivalent for ǫ = 0, whereas conditions (i) and (ii) in Theorem 2.15 are equivalent. ON SOME GEOMETRIC PROPERTIES OF OPERATOR SPACES 11 3. Birkhoff-James orthogonality of bounded linear operators We begin this section with an easy proposition on approximate Birkhoff-James orthogonality ( ǫ ). ⊥D Proposition 3.1. Let X, Y be two normed linear spaces. Let T B(X, Y) and x M . Then for any ǫ [0, 1) and y X, T x ǫ Ty = x ǫ y∈. ∈ T ∈ ∈ ⊥D ⇒ ⊥D Proof. Let T x ǫ Ty. Then for any λ K, T x + λTy √1 ǫ2 T x . ⊥D ∈ k k ≥ − k k Therefore, for any λ K, T x + λy T x + λTy √1 ǫ2 T x = ∈ k kk k≥ k k ≥ − k k √1 ǫ2 T x . Thus, for any λ K, x + λy √1 ǫ2 x . Therefore, x −ǫ y.k kk k ∈ k k ≥ − k k ⊥D In [10, Th. 2.1], James characterized Birkhoff-James orthogonality in terms of linear functionals. In the next lemma, we characterize approximate Birkhoff- James orthogonality ( ǫ ) in terms of linear functionals. ⊥D Lemma 3.2. Let X be a normed linear space and x, y X. Then for any ǫ ǫ ∈ ∈ [0, 1), x y if and only if there exists a linear functional f S ∗ such that ⊥D ∈ X f(x) √1 ǫ2 x and f(y)=0. ≥ − k k Proof. The necessary part follows from a slight variation of [7, Cor. 6.8]. Let us prove the sufficient part of the lemma. Suppose there exists a linear functional 2 f SX∗ such that f(x) √1 ǫ x and f(y) = 0. Then for any λ K, ∈ 2 ≥ − k k ∈ ǫ √1 ǫ x f(x) = f(x + λy) f x + λy = x + λy . Therefore, x D y. − k k ≤ ≤ k kk k k k ⊥ The idea of the following theorem is adapted from [10, Th. 2.1]. However, for the sake of completeness, we present a complete proof of the result. Theorem 3.3. Let X be a normed linear space. Let x be a nonzero element in X and H be a hyperspace in X. Then for ǫ [0, 1), x ǫ H if and only if there ∈ ⊥D exists f S ∗ such that f(x) √1 ǫ2 x , where H = ker(f). ∈ X ≥ − k k 2 Proof. First suppose that there exists f SX∗ such that f(x) √1 ǫ x , where H = ker(f). Then for any h H and∈ for any λ K, ≥ − k k ∈ ∈ √1 ǫ2 x f(x)= f(x + λh) f x + λh = x + λh . − k k ≤ ≤k kk k k k This implies that √1 ǫ2 x x + λh for any h H and for any λ K. ǫ − k k≤k k ∈ ∈ Thus, x D H. ⊥ ǫ Conversely, suppose that x D H. Then it is easy to check that X = span x, H . Define g : X K by g(ax⊥ + h) = a, where a K and h H. Clearly{ g }is −→ ∈ ax+h ∈ ǫ linear and ker(g)= H. Now, g(ax + h) = a k k , since x h. Thus, √1 ǫ2 x D | | | | ≤ − k k ⊥ 1 2 1 g 2 √1 ǫ g x 1 = g(x). Let f = g. Then clearly, √1 ǫ x g k k ≤ − k k ⇒ − k kk k ≤ k k 2 f SX∗ , f(x) √1 ǫ x and H = ker(g) = ker(f). This establishes the theorem.∈ ≥ − k k Our next objective is to study Birkhoff-James orthogonality in the space of bounded linear operators by classifying them into two different cases: “T B A but T A” and “T A”. In [15, Th. 2.1], Sain and Paul proved that⊥ if 6⊥SB ⊥SB 12 ARPITAMAL,DEBMALYASAINandKALLOLPAUL T is a bounded linear operator on a finite-dimensional real Banach space X and D is a non empty connected subset of SX such that MT = D ( D), then for any A B(X), T A if and only if there exists x D such∪ that− T x Ax. ∈ ⊥B ∈ ⊥B Note that, if there exists x D such that T x SB Ax then for any λ K 0 , T + λA (T + λA)x >∈ T x = T , i.e.,⊥T A. However, the∈ following\{ } k k≥k k k k k k ⊥SB example illustrates that if T SB A then there may not exist any x D such that T x Ax. ⊥ ∈ ⊥SB Example 3.4. Consider the two-dimensional real normed linear space X such 2 2 2 2 that SX = (x, y) R : x =1, y 1 (x, y) R : x +(y 1) =1, 1 y 2 ({x, y) ∈R2 : x|2 +(| y +1)| 2 =1|≤ , }∪{2 y ∈ 1 . Define T,− A B(X) by≤ T (1≤, 1)}∪{ = (1, 1),∈ T (0, 2) = (0, 0) and A(1,−1) =≤ (0≤, 2) −, A}(0, 2) = (1, 1).∈ Now, it − is easy to observe that T = 1 and MT = D ( D), where D = (1,y): 1 y 1 . Let λ> 0. Thenk kT +λA (T +λA∪)(1−, 1) = (1, 1+2λ){ > 1=−T ≤. ≤ } k k≥k k k λ 5λ k k k Next, let λ< 0. Then T +λA (T +λA)(1, 0) = (1 2 , 1+ 2 ) > 1= T . Therefore, T A.k Now,k≥k let (1,y) D. Thenk kT (1−,y) = (1,k1). Clearly,k k ⊥SB ∈ there does not exist any (x1,y1) X such that (1, 1) SB (x1,y1). Therefore, T (1,y) A(1,y) for any (1,y) ∈ D. ⊥ 6⊥SB ∈ In the following theorem, we characterize strong Birkhoff-James orthogonality (T SB A) in the space of all bounded linear operators between finite-dimensional Banach⊥ spaces. Theorem 3.5. Let X, Y be finite-dimensional Banach spaces. Let T, A B(X, Y). Then T A if and only if for any ǫ > 0, there exists λ > ∈0 ⊥SB ǫ such that for each λ < λǫ, there exists yλ ( x MT B(x, ǫ)) SX such that Ty + λAy > T| .| ∈ ∪ ∈ ∩ k λ λk k k Proof. First, we prove the easier sufficient part of the theorem. Let ǫ > 0 be given. Then there exists λ > 0 such that for each λ < λ , there ǫ | | ǫ exists yλ ( x MT B(x, ǫ)) SX such that Tyλ + λAyλ > T . Therefore, ∈ ∪ ∈ ∩ k k k k T + λA Tyλ + λAyλ > T . Now, by the convexity of the norm function, kit is easyk≥k to observe that Tk k Ak. ⊥SB For the necessary part, suppose that T SB A. Let ǫ > 0 be given. Now, using the compactness of S , it is easy to⊥ observe that sup T z : z S X {k k ∈ X \ ( x MT B(x, ǫ)) < T δ for some δ > 0. Let z SX ( x MT B(x, ǫ)). Then ∪T z∈ < T }δ. Therefore,k k − ∈ \ ∪ ∈ k k k k − T z + λAz T z + λ Az k k ≤ k k | |k k < T δ + λ A k k − | |k k < T (for some λ <λ ). k k | | ǫ Now, since T +λA > T ,(T +λA) does not attain norm in SX ( x MT B(x, ǫ) k k k k \ ∪ ∈ for each λ <λǫ. Hence there exists yλ SX ( x MT B(x, ǫ)) such that Tyλ + λAy =| |T + λA > T . This completes∈ the∩ proof∪ ∈ of the theorem. k λk k k k k Now, we study the case when “T B A but T SB A”, where T and A are compact linear operators defined from⊥ a reflexive Banach6⊥ space to any normed linear space. ON SOME GEOMETRIC PROPERTIES OF OPERATOR SPACES 13 Theorem 3.6. Let X be a reflexive Banach space and Y be any normed linear space. Let T, A K(X, Y) be such that T B A but T SB A. Then there exists x M such that∈ T x Ax. ⊥ 6⊥ ∈ T ⊥B Proof. Without loss of generality, we may and do assume that T = 1. Since k k T B A but T SB A, there exists µ K 0 such that T + µA = T = 1. Let⊥B = µA 6⊥ T . Then B = 1.∈Now,\{T } A givesk that T k (kB k+ T ). − − k k ⊥B ⊥B Therefore, from [10, Th. 2.1], we have, there exists F K(X, Y)∗ such that F =1, F (T )= T and F (T + B) = 0. Clearly, F ( B)=∈ T =1= B . Therefore,k k T + k k B = F (T )+ F ( B) = F (T− B) k kF T kB − k= T B kTk + k −B k. Thus, T B −= T + B− . Now,≤ k bykk Remark− k2.4, wek − have,k≤k therek existsk −x k M Mk such− k thatk Tk x kBx − k= T x + Bx . We ∈ T ∩ B k − k k k k − k note that there exists g SX∗ such that g(T x Bx) = T x Bx . Therefore, g(T x) + g( Bx) = g(T∈ x Bx) = T x Bx− T xk +− Bxk . Again, g(T x) −T x and g( Bx−) kBx −givesk≤k that g(Tk x) =k −T x k= 1 and g( Bx)≤ = k kBx = 1.− This≤ implies k − thatk g(T x + Bx) = 0. Therefore,k k by [10, − k − k Th. 2.1], we have, T x B (T x + Bx), i.e., T x B ( µAx). Hence T x B Ax. This completes the proof⊥ of the theorem. ⊥ − ⊥ Remark 3.7. From the above theorem and [14, Th. 2.1], it is clear that for T, A K(X, Y), where X is a reflexive Banach space and Y is any normed linear ∈ space, T B A implies there exists x MT such that T x B Ax if either T SB A or M =⊥D ( D)(D is a non-empty∈ compact connected⊥ subset of S ). 6⊥ T ∪ − X Combining Theorem 3.6 of this paper and [16, Th. 2.3], we obtain the following corollary. Corollary 3.8. Let X be a two-dimensional real normed linear space. Let T ∈ B(X). If for any A B(X), T B A implies that T SB A then MT cannot have more than two components.∈ ⊥ 6⊥ Proof. Let T B(X). Suppose for any A B(X), T A implies that T A. ∈ ∈ ⊥B 6⊥SB If possible, suppose that MT has more than two components. Then by [16, Th. 2.3], we have, there exists A B(X) such that T B A but T x B Ax for any x M . Therefore, by Theorem∈ 3.6, we have T ⊥ A, which6⊥ contradicts the ∈ T ⊥SB hypothesis of the corollary. Thus, MT cannot have more than two components. The following example shows that the converse of Corollary 3.8 is not true. Example 3.9. Consider T, A B(ℓ2 ) defined by T (x, y)=(x, x) and A(x, y)= ∈ ∞ ( x, x) respectively. Clearly, MT has only two components but T SB A, since −T + λA (T + λA)(1, 1) > T (1, 1) = T for all λ R 0 ⊥. k k≥k k k k k k ∈ \{ } We further study the case “T B A but T SB A”, where T and A are bounded linear operators defined between⊥ any two normed6⊥ linear spaces. For this, we need the following lemma. Lemma 3.10. Let X be a normed linear space. Suppose xn , yn are two bounded sequences of X such that { } { } lim xn + yn = lim xn + lim yn . n n n →∞ k k →∞ k k →∞ k k 14 ARPITAMAL,DEBMALYASAINandKALLOLPAUL Then for each k N, there exists f S ∗ such that ∈ nk ∈ X lim fnk (xnk ) = lim xnk k k →∞ →∞ k k and lim fnk (ynk ) = lim ynk . k k →∞ →∞ k k Proof. By Hahn-Banach Theorem, we have, for each n N, there exists f S ∗ ∈ n ∈ X such that fn(xn +yn)= xn +yn . Therefore, limn fn(xn +yn) = limn xn + k k →∞ →∞ k yn . Since xn , yn are bounded sequences of X, fn(xn) , fn(yn) are boundedk sequences{ } { of R} . This proves that f (x ) , f{ (y ) }have{ convergent} { n n } { n n } subsequences. Let limk fnk (xnk ) and limk fnk (ynk ) exists. Then →∞ →∞ lim fnk (xnk ) + lim fnk (ynk ) = lim fnk (xnk + ynk ) k k k →∞ →∞ →∞ = lim xnk + ynk k →∞ k k = lim xnk + lim ynk . k k →∞ k k →∞ k k Again, since limk fnk (xnk ) limk xnk and limk fnk (ynk ) limk ynk , we have, →∞ ≤ →∞ k k →∞ ≤ →∞ k k lim fnk (xnk ) = lim xnk k k →∞ →∞ k k and lim fnk (ynk ) = lim ynk . k k →∞ →∞ k k This completes the proof of the lemma. Let us now obtain the desired necessary condition for “T B A but T SB A”, by applying the previous lemma. ⊥ 6⊥ Theorem 3.11. Let X, Y be two normed linear spaces and T, A B(X, Y). Suppose T A but T A. Then either there exists a sequence ∈x in S ⊥B 6⊥SB { n} X such that T xn T , Axn 0 or there exists a sequence xn in SX and a sequencek ǫ kin −→R+ k suchk that−→T x T , ǫ 0 and T{ x } ǫn Ax . { n} k nk −→k k n −→ n ⊥D n Proof. Without loss of generality assume that T = 1. Since T A but k k ⊥B T SB A, there exists µ K 0 such that T + µA = T = 1. Let B 6⊥= µA T . Then B∈ =\ 1. {Now,} T Ak gives thatk Tk k (B + T ). − − k k ⊥B ⊥B Therefore, from [10, Th. 2.1], we have, there exists F B(X, Y)∗ such that F =1, F (T )= T and F (T + B) = 0. Clearly, F ( B)=∈ T =1= B . Therefore,k k T + k k B = F (T )+ F ( B) = F (T− B) k kF T kB − k= T B k Tk + k −B k. Thus, T B −= T + −B . Now,≤ k fromkk Theorem− k 2.6k ,− it isk≤k easy tok observek − k that therek exists− k a sequencek k k −x kin S such that { n} X lim T xn = T , lim Bxn = B n n →∞ k k k k →∞ k k k k and lim T xn Bxn = lim T xn + lim Bxn . n n n →∞ k − k →∞ k k →∞ k − k ON SOME GEOMETRIC PROPERTIES OF OPERATOR SPACES 15 Now, by Lemma 3.10, without loss of generality we may assume that for each n N there exists f S ∗ such that ∈ n ∈ Y lim fn(T xn) = lim T xn = T =1 n n →∞ →∞ k k k k and lim fn( Bxn) = lim Bxn = B =1. n n →∞ − →∞ k − k k k Clearly, fn(T xn) = rn and fn( Bxn) = sn, where rn 1, sn 1, rn 1 and s 1. Therefore, f (T− x + Bx ) = r |s ,|which ≤ | converges| ≤ to−→ 0 as n −→ n n n n − n n . If Axn has a subsequence converging to 0 then we are done. So assume that→∞Ax {has} no subsequence converging to 0. Without loss of generality we { n} may assume that infn Axn = c > 0. Now, it is easy to observe that 1 > 2 k2 k 1 ( rn c µ rn sn ) 0, since rn 1, rn 1 and rn sn 0. Let for − | | − | 2| | − | ≥ 2 | |→2 | | ≤ | − |→ each n N, ǫn =1 ( rn c µ rn sn ) . Then clearly ǫn [0, 1) for each n N ∈ − | | − | | | ǫ−n | 2∈ 2 ∈ and ǫn 0. We show that T xn D Axn. First, let λ c T xn Axn T xn . → ⊥ | | ≥ k k ≥ k k k k Then T x + λAx λ Ax T x T x 1 ǫ2 T x . Let k n nk≥| |k nk−k nk≥k nk ≥ p − nk nk λ < 2 T x 2 . Then | | c k nk ≤ c λ T x + λAx = T x (T x + Bx ) k n nk k n − µ n n k λ f T x (T x + Bx ) ≥ | n n − µ n n | λ = r (r s ) | n − µ n − n | λ r | | r s ≥ | n| − µ | n − n| | | 2 > r r s | n| − c µ | n − n| | | = 1 ǫ2 p − n 1 ǫ2 T x . ≥ p − nk nk Hence T x ǫn Ax . This completes the proof of the theorem. n ⊥D n In the following example we show that the conditions given in Theorem 3.11 are not sufficient to ensure that T A but T A. ⊥B 6⊥SB Example 3.12. Let X = ℓ (R). Define linear operators T, A : ℓ ℓ as follows: ∞ ∞ −→ ∞ T (x1, x2, x3,...) = (x1, x1, x1,...) A(x , x , x ,...) = ( x , x , x ,...) 1 2 3 − 2 2 2 Then it is easy to check that T, A B(ℓ ) with T = A = 1. Clearly for λ = 0, T + λA (T + λA)(1, 1, 0∈, 0, 0,...∞ ) = k(1k λ,k1+k λ, 1+ λ, . . .) = 6 k k≥k k k − k 1+ λ > 1 = T . Therefore, T SB A. But (1, 0, 0,...) MT and choosing y =| (1| , 0, 0,...k) wek get Ty ⊥T , Ay 0. ∈ n k nk→k k k nk→ Similarly, defining B : ℓ ℓ by B(x1, x2, x3,...)=( x1, x1, x1,...), we can ∞ −→ ∞ − 16 ARPITAMAL,DEBMALYASAINandKALLOLPAUL check that T SB A although there exists yn = (1, 0, 0,...) MT , ǫn = 0 such that Ty ⊥ T and Ty ǫn By . ∈ k nk→k k n ⊥D n Now, we turn our attention to the newly introduced notion of semi-rotundity of a normed linear space. As mentioned earlier, every exposed point (strictly convex space) is a semi-rotund point ( semi-rotund space) but not conversely, if dimension of the space is greater than two. The notions are equivalent if the dimension of the space is two. In this context, we first prove the following proposition which states that every isometry defined between finite-dimensional Banach spaces is a semi-rotund point in the operator space. Proposition 3.13. Let X, Y be finite-dimensional Banach spaces. Then every isometry from X to Y is a semi-rotund point in B(X, Y). Proof. Let T B(X, Y) be an isometry, where X, Y are finite-dimensional Banach ∈ spaces. Clearly, T is invertible. Let y be an exposed point of BY. Then there exists x S such that T x = y. Since y is an exposed point of B(Y), there exists ∈ X z Y such that y SB z. Define A B(X, Y) by Ax = z and Aw = 0 for all w ∈ H, where H is⊥ a hyperspace such∈ that x H. Then T x Ax. We also ∈ ⊥B ⊥SB observe that x MT , since T is an isometry. Therefore, T SB A. Hence, T is a semi-rotund point∈ of B(X, Y). ⊥ In the following theorem, we prove that every non-zero compact linear operator from a reflexive Banach space to a strictly convex Banach space is semi-rotund. Theorem 3.14. Let X be a reflexive Banach space and Y be a strictly convex Banach space. Then every non-zero T K(X, Y) is a semi-rotund point. ∈ Proof. Since X is reflexive and T is compact, T attains its norm. Let x MT . By [10, Cor. 2.2], there exists y S such that T x y. Now, since∈ Y is ∈ Y ⊥B strictly convex and T is non-zero, we must have, T x SB y. Define a linear operator A from X to Y by Ax = y and Az = 0 for all⊥ z H, where H is a hyperspace such that x H. Clearly, A K(X, Y). Now, for∈ any λ K 0 , ⊥B ∈ ∈ \{ } T +λA (T +λA)x = T x+λy > T x = T . Therefore, T SB A. This kproves thatk≥kT is a semi-rotundk k pointk andk completesk k thek proof of the theorem.⊥ Finally, using Theorem 3.14, we obtain the following corollary that illustrates that semi-rotundity is a strictly weaker property compared to strict convexity. Corollary 3.15. Let X, Y be finite-dimensional Banach spaces and in addition, let Y be strictly convex. Then B(X, Y) is not strictly convex but semi-rotund. Proof. Let x S . Define T B(X, Y) by T x = x and Ty = 0 for all y H, ∈ X ∈ ∈ where H is a hyperspace such that x BH. Then clearly T = 1. Now, 2 = I + T I +T Ix+T x = 2. This⊥ shows that T +k Ik = T +I , althoughk k Tk andk≥kI are linearlyk≥k independent.k Therefore, B(X, kY)k is notk k strictlyk convex.k Now, being finite-dimensional, X is reflexive. Since Y is finite-dimensional, it follows that B(X, Y)= K(X, Y). Therefore, by Theorem 3.14, we have, every T B(X, Y) is a semi-rotund point. This completes the proof of the corollary. ∈ ON SOME GEOMETRIC PROPERTIES OF OPERATOR SPACES 17 References 1. L. Arambaˇsi´c, R. Raji´c, The Birkhoff-James orthogonality in Hilbert C*-modules, Linear Algebra Appl., 437 (2012) 1913–1929. 2. C. Ben´ıtez, M. Fern´andez and M.L. Soriano, Orthogonality of matrices, Linear Algebra Appl., 422 (2007) 155–163. 3. R. Bhatia and P. Semrl,ˇ Orthogonality of matrices and distance problem, Linear Algebra Appl., 287 (1999) 77–85. 4. G. Birkhoff, Orthogonality in linear metric spaces, Duke Math. J. 1 (1935) 169–172. 5. T. Bottazzi, C. Conde, M. S. Moslehian, P. Wojcik and A. Zamani, Orthogonality and parallelism of operators on various Banach spaces, J. Austral. Math. Soc. (to appear). 6. J. Chmieli´nski, Linear mappings approximately preserving orthogonality, J. Math. Anal. Appl. 304 (2005) 158-169. 7. J. B. Conway, A Course in Functional Analysis, Springer-Verlag, New York, Inc., 1990. 8. S.S. Dragomir, On approximation of continuous linear functionals in normed linear spaces, An. Univ. Timisoara Ser. Stiint. Mat., 29 (1991), 51–58. 9. P. Ghosh, D. Sain and K. Paul, On symmetry of Birkhoff-James orthogonality of linear operators, Adv. Oper. Theory 2 (2017) 428-434. 10. R. C. James, Orthogonality and linear functionals in normed linear spaces, Trans. Amer. Math. Soc. 61 (1947) 265-292. 11. C. K. Li and H. Schneider, Orthogonality of matrices, Linear Algebra Appl., 47 (2002) 115–122. 12. M. S. Moslehian and A. Zamani, Exact and approximate operator parallelism, Canad. Math. Bull. 58 (2015) 207-224. 13. K. Paul, D. Sain and K. Jha, On strong orthogonality and strictly convex normed linear spaces, J. Inequal. Appl. 2013, 2013:242. 14. K. Paul, D. Sain and P. Ghosh, Birkhoff-James orthogonality and smoothness of bounded linear operators, Linear Algebra Appl. 506 (2016) 551-563. 15. D. Sain and K. Paul, Operator norm attainment and inner product spaces, Linear Algebra Appl. 439 (2013) 2448-2452. 16. D. Sain, K. Paul and S. Hait, Operator norm attainment and Birkhoff-James orthogonality, Linear Algebra Appl. 476 (2015) 85-97. 17. A. Seddik, Rank one operators and norm of elementary operators, Linear Algebra Appl. 424 (2007) 177–183. 18. T. Bhattacharyya, P. Grover, Characterization of Birkhoff-James orthogonality, J. Math. Anal. Appl. 407 (2013) 350–358. 19. P. W´ojcik, Orthogonality of compact operators, Expositiones Mathematicae, 35 (2017) 86-94 20. A. Zamani and M. S. Moslehian, Norm-Parallelism in the geometry of Hilbert C∗-modules, Indag. Math. (N.S.) 27 (2016) 266-281. 1Department of Mathematics, Jadavpur University, Kolkata 700032, West Bengal, India. E-mail address: [email protected]; [email protected] 2Department of Mathematics, Indian Institute of Science, Bengaluru 560012, Karnataka, India. E-mail address: [email protected]