Introduction
Assume that X \mathcal{X} X is any normed space over the real or complex field F ( = R \mathbb{F}(=\mathbb{R} F ( = R or C ) \mathbb{C}) C ) with some orthogonality ⊥ \perp ⊥ . We say that A , B ∈ X A,B\in\mathcal{X} A , B ∈ X are mutual orthogonal to each other, denoted by A ⊥ ⊥ B A\perp\!\!\perp B A ⊥ ⊥ B , if A ⊥ B A\perp B A ⊥ B and B ⊥ A B\perp A B ⊥ A . For an orthograph G = Γ ( X , ⊥ ⊥ ) G=\Gamma(\mathcal{X},\perp\!\!\perp) G = Γ ( X , ⊥ ⊥ ) , the set of its vertices is V ( G ) = { [ A ] = F A : A ∈ X and ∥ A ∥ = 1 } V(G)=\{[A]=\mathbb{F}A:A\in\mathcal{X}\text{ and }\|A\|=1\} V ( G ) = {[ A ] = F A : A ∈ X and ∥ A ∥ = 1 } ; and the set of its edges is E ( G ) = { A B : A , B ∈ X and A ⊥ ⊥ B } E(G)=\{AB:A,B\in\mathcal{X}\text{ and }A\perp\!\!\perp B\} E ( G ) = { A B : A , B ∈ X and A ⊥ ⊥ B } . For the convenience of expression, write A A A instead of [ A ] [A] [ A ] . Two vertices A , B ∈ V ( G ) A,B\in V(G) A , B ∈ V ( G ) are said to be adjacent if A B ∈ E ( G ) AB\in E(G) A B ∈ E ( G ) , and A ∈ V ( G ) A\in V(G) A ∈ V ( G ) is called an isolated point if it is not adjacent to any other vertices in V ( G ) V(G) V ( G ) . A vertex sequence A 1 A 2 ⋯ A n A_{1}A_{2}\cdots A_{n} A 1 A 2 ⋯ A n is called a path if the vertices { A i } \{A_{i}\} { A i } are distinct and A i A i + 1 ∈ E ( G ) A_{i}A_{i+1}\in E(G) A i A i + 1 ∈ E ( G ) with i = 1 , 2 , ⋯ n − 1 i=1,2,\cdots n-1 i = 1 , 2 , ⋯ n − 1 ; and in this case, the number of edges is called the length of the path. The graph G G G is said to be connected if there exists a path from any vertex to any other vertex of G G G . A connected component of G G G is a maximal connected subgraph. The distance between two distint vertices is the length of the shortest path between them. The diameter of G G G is the maximum of distances between vertices for all pairs of vertices in G G G , denoted by Diam G \operatorname{Diam}G Diam G . For more details, see [6 ] .
In [8 ] , Birkhoff James orthogonality was introduced, where x ∈ X x\in\mathcal{X} x ∈ X is said to be Birkhoff James orthogonal to y ∈ X y\in\mathcal{X} y ∈ X , denoted by x ⊥ B y x\perp_{B}y x ⊥ B y , if ∥ x + λ y ∥ ≥ ∥ x ∥ \|x+\lambda y\|\geq\|x\| ∥ x + λ y ∥ ≥ ∥ x ∥ holds for all λ ∈ F \lambda\in\mathbb{F} λ ∈ F . So x ⊥ ⊥ B y x\perp\!\!\perp_{B}y x ⊥ ⊥ B y means that x ⊥ B y x\perp_{B}y x ⊥ B y and y ⊥ B x y\perp_{B}x y ⊥ B x . In [2 ] , the authors proved that the orthograph Γ ( X , ⊥ ⊥ B ) \Gamma(\mathcal{X},\perp\!\!\perp_{B}) Γ ( X , ⊥ ⊥ B )
2010 Mathematics Subject Classification. Primary 47A12, 47B47.
Key words and phrases. Orthograph, orthogonality, numerical radius orthogonality, strong numerical radius orthogonality.
This work is partially supported by National Natural Science Foundation of China (12571138, 12071336) and the Natural Science Basic Research Program of Shaanxi Province (S2025-JC-QN-1854). is connected when dim X ≥ 3 \dim\mathcal{X} \ge3 dim X ≥ 3 , and that Diam Γ ( X , ⊥ B ) = 2 \operatorname{Diam}\Gamma(\mathcal{X},\perp_{\!B})=2 Diam Γ ( X , ⊥ B ) = 2 when dim X ≥ 5 \dim\mathcal{X} \ge5 dim X ≥ 5 . Later, strong Birkhoff James orthogonality was introduced for C∗ ^{*} ∗ -algebras in [4 ] , where A ∈ A A \in\mathcal{A} A ∈ A (any C∗ ^{*} ∗ -algebra) is said to be strong Birkhoff James orthogonal to B ∈ A B \in\mathcal{A} B ∈ A , denoted by A ⊥ B s B A\perp_{B}^{s}B A ⊥ B s B , if ∥ A + B C ∥ ≥ ∥ A ∥ \lVert A+BC\rVert\ge\lVert A\rVert ∥ A + B C ∥ ≥ ∥ A ∥ holds for all C ∈ A C \in\mathcal{A} C ∈ A . In [1 ] , the authors discussed the connected components and the diameter of the orthograph Γ ( A , ⊥ B s ) \Gamma(\mathcal{A},\perp_{B}^{s}) Γ ( A , ⊥ B s ) when A \mathcal{A} A is commutative or A = B ( H ) \mathcal{A}=\mathcal{B}(H) A = B ( H ) (the algebra of all bounded linear operators on a complex Hilbert space H H H ), and they proved that A ∈ B ( H ) A \in\mathcal{B}(H) A ∈ B ( H ) is an isolated vertex of Γ ( B ( H ) , ⊥ B s ) \Gamma(\mathcal{B}(H),\perp_{B}^{s}) Γ ( B ( H ) , ⊥ B s ) if and only if A A A is right invertible. In [13 ] , the authors proved that, for any C∗ ^{*} ∗ -algebra A \mathcal{A} A , A ∈ A A \in\mathcal{A} A ∈ A is an isolated vertex of Γ ( A , ⊥ B s ) \Gamma(\mathcal{A},\perp_{B}^{s}) Γ ( A , ⊥ B s ) if and only if A A A is approximately right invertible; and except C \mathbb{C} C , C ⊕ C \mathbb{C}\oplus\mathbb{C} C ⊕ C and M 2 ( C ) \mathcal{M}_{2}(\mathbb{C}) M 2 ( C ) (the algebra of all 2 × 2 2\times2 2 × 2 complex matrices), all non isolated points make a single connected component of Γ ( A , ⊥ B s ) \Gamma(\mathcal{A},\perp_{B}^{s}) Γ ( A , ⊥ B s ) with the diameter of this component not exceed 4. In [18 ] , the orthograph Γ ( A , ⊥ B s ) \Gamma(\mathcal{A},\perp_{B}^{s}) Γ ( A , ⊥ B s ) with A \mathcal{A} A being any finite-dimensional C∗ ^{*} ∗ -algebra is discussed, and showed that the diameter of Γ ( A , ⊥ B s ) \Gamma(\mathcal{A},\perp_{B}^{s}) Γ ( A , ⊥ B s ) can be used to identify different types of finite-dimensional C∗ ^{*} ∗ -algebras, that is, Diam Γ ( C ⊕ C , ⊥ B s ) = 1 \operatorname{Diam}\Gamma(\mathbb{C}\oplus\mathbb{C},\perp_{B}^{s})=1 Diam Γ ( C ⊕ C , ⊥ B s ) = 1 ; Diam Γ ( C ⊕ M 2 ( C ) , ⊥ B s ) = 4 \operatorname{Diam}\Gamma(\mathbb{C}\oplus\mathcal{M}_{2}(\mathbb{C}),\perp_{B}^{s})=4 Diam Γ ( C ⊕ M 2 ( C ) , ⊥ B s ) = 4 and Diam Γ ( A , ⊥ B s ) = 3 \operatorname{Diam}\Gamma(\mathcal{A},\perp_{B}^{s})=3 Diam Γ ( A , ⊥ B s ) = 3 for all other cases.
The numerical radius, which is equivalent to the operator norm, is an important tool for studying operators and matrices, and has been extensively studied. Recall that the numerical range and the numerical radius of any operator A ∈ B ( H ) A \in\mathcal{B}(H) A ∈ B ( H ) are defined respectively W ( A ) = { ⟨ A x , x ⟩ : x ∈ H , ∥ x ∥ = 1 } W(A)=\{\langle Ax,x\rangle:x\in H,\lVert x\rVert=1\} W ( A ) = {⟨ A x , x ⟩ : x ∈ H , ∥ x ∥ = 1 } and w ( A ) = sup { ∣ λ ∣ : λ ∈ W ( A ) } w(A)=\sup\{|\lambda|:\lambda\in W(A)\} w ( A ) = sup { ∣ λ ∣ : λ ∈ W ( A )} . The authors in [15 ] replaced the operator norm in B-J orthogonality with the numerical radius and introduced the concept of numerical radius orthogonality. Recall that A ∈ B ( H ) A \in\mathcal{B}(H) A ∈ B ( H ) is said to be numerical radius orthogonal to B ∈ B ( H ) B \in\mathcal{B}(H) B ∈ B ( H ) , denoted by A ⊥ w B A\perp_{w}B A ⊥ w B , if w ( A + λ B ) ≥ w ( A ) w(A+\lambda B)\ge w(A) w ( A + λ B ) ≥ w ( A ) holds for all λ ∈ C \lambda\in\mathbb{C} λ ∈ C . Note that ⊥ B \perp_{B} ⊥ B and ⊥ w \perp_{w} ⊥ w are not comparable. For the study of numerical radius orthogonality, see [3 , 10 , 17 ] and the references therein. Recently, the authors in [19 ] introduced strong numerical radius orthogonality on B ( H ) \mathcal{B}(H) B ( H ) , where A ∈ B ( H ) A\in\mathcal{B}(H) A ∈ B ( H ) is strong numerical radius orthogonal to B ∈ B ( H ) B\in\mathcal{B}(H) B ∈ B ( H ) , denoted by A ⊥ w s B A\perp_{w}^{s}B A ⊥ w s B , if w ( A + B C ) ≥ w ( A ) w(A+BC)\ge w(A) w ( A + B C ) ≥ w ( A ) holds for all C ∈ B ( H ) C\in\mathcal{B}(H) C ∈ B ( H ) . It is easy to verify that A ⊥ w s B A\perp_{w}^{s}B A ⊥ w s B if and only if A ⊥ w B C A\perp_{w}BC A ⊥ w B C holds for all C ∈ B ( H ) C\in\mathcal{B}(H) C ∈ B ( H ) . The numerical radius orthographs Γ ( B ( H ) , ⊥ w ) \Gamma(\mathcal{B}(H),\perp_{w}) Γ ( B ( H ) , ⊥ w ) and Γ ( B s ( H ) , ⊥ w ) \Gamma(\mathcal{B}_{s}(H),\perp_{w}) Γ ( B s ( H ) , ⊥ w ) were discussed in [2 ] . It was shown (ref. [2 Theorem 4.1] ) that both graphs are connected, that is, they have no isolated vertices, for dim H ≥ 2 \dim H\ge2 dim H ≥ 2 . It was also shown that Diam Γ ( B ( H ) , ⊥ w ) = Diam Γ ( B s ( H ) , ⊥ w ) = 2 \operatorname{Diam}\Gamma(\mathcal{B}(H),\perp_{w})=\operatorname{Diam}\Gamma(\mathcal{B}_{s}(H),\perp_{w})=2 Diam Γ ( B ( H ) , ⊥ w ) = Diam Γ ( B s ( H ) , ⊥ w ) = 2 when dim H ≥ 3 \dim H\ge3 dim H ≥ 3 . Then an interesting question is to ask what are the diameters of these two graphs if dim H = 2 \dim H=2 dim H = 2 ?
The purpose of the present paper is to discuss the numerical radius orthograph Γ ( Y , ⊥ w ) \Gamma(\mathcal{Y},\perp_w) Γ ( Y , ⊥ w ) and strong numerical radius orthograph Γ ( Y , ⊥ w s ) \Gamma(\mathcal{Y},\perp_w^s) Γ ( Y , ⊥ w s ) , where Y \mathcal{Y} Y represents B ( H ) \mathcal{B}(H) B ( H ) or B s ( H ) \mathcal{B}_s(H) B s ( H ) (the Jordan algebra of all self-adjoint operators in B ( H ) \mathcal{B}(H) B ( H ) ). The rest of the paper is organized as follows. Section 2 provides some properties about the (strong) numerical radius orthogonality of bounded linear operators. Section 3 is devoted to answering the question above. We show that Diam Γ ( B ( H ) , ⊥ w ) = 3 \operatorname{Diam}\Gamma(\mathcal{B}(H),\perp_w)=3 Diam Γ ( B ( H ) , ⊥ w ) = 3 and Diam Γ ( B s ( H ) , ⊥ w ) = 2 \operatorname{Diam}\Gamma(\mathcal{B}_s(H),\perp_w)=2 Diam Γ ( B s ( H ) , ⊥ w ) = 2 for the case dim H = 2 \dim H=2 dim H = 2 (Theorem 3.1 ). For a subset S S S of a C∗ ^* ∗ -algebra A A A , denote by NI ( S ) \operatorname{NI}(S) NI ( S ) the set of all element in S S S that is not invertible in A A A . In Section 4, we consider the case of strong numerical radius orthogonality and characterize the connected components of Γ ( M n ( C ) , ⊥ w s ) \Gamma(\mathcal{M}_n(\mathbb{C}),\perp_w^s) Γ ( M n ( C ) , ⊥ w s ) (Theorems 4.2 and 4.5 ) and show that Diam Γ ( NI ( M n ( C ) ) , ⊥ w s ) ≤ 4 \operatorname{Diam}\Gamma(\operatorname{NI}(\mathcal{M}_n(\mathbb{C})),\perp_w^s)\leq4 Diam Γ ( NI ( M n ( C )) , ⊥ w s ) ≤ 4 for n ≥ 3 n\geq3 n ≥ 3 (Theorem 4.3 ) and Diam Γ ( NI ( M 2 ( C ) ) , ⊥ w s ) = 4 \operatorname{Diam}\Gamma(\operatorname{NI}(\mathcal{M}_2(\mathbb{C})),\perp_w^s)=4 Diam Γ ( NI ( M 2 ( C )) , ⊥ w s ) = 4 (Theorem 4.8 ). In Section 5, we discuss the orthograph Γ ( B s ( H ) , ⊥ w s ) \Gamma(\mathcal{B}_s(H),\perp_w^s) Γ ( B s ( H ) , ⊥ w s ) . We characterize the connected components of Γ ( B s ( H ) , ⊥ w s ) \Gamma(\mathcal{B}_s(H),\perp_w^s) Γ ( B s ( H ) , ⊥ w s ) for dim H ≥ 3 \dim H\geq3 dim H ≥ 3 (Theorem 5.3 ) and show that Diam Γ ( NI ( B s ( H ) ) , ⊥ w s ) = 3 \operatorname{Diam}\Gamma(\operatorname{NI}(\mathcal{B}_s(H)),\perp_w^s)=3 Diam Γ ( NI ( B s ( H )) , ⊥ w s ) = 3 for dim H ≥ 4 \dim H\geq4 dim H ≥ 4 and Diam Γ ( NI ( B s ( H ) ) , ⊥ w s ) = 4 \operatorname{Diam}\Gamma(\operatorname{NI}(\mathcal{B}_s(H)),\perp_w^s)=4 Diam Γ ( NI ( B s ( H )) , ⊥ w s ) = 4 for dim H = 3 \dim H=3 dim H = 3 (Theorem 5.2 ). The structure of connected components of Γ ( B s ( H ) , ⊥ w s ) \Gamma(\mathcal{B}_s(H),\perp_w^s) Γ ( B s ( H ) , ⊥ w s ) when dim H = 2 \dim H=2 dim H = 2 is very different which is described in Theorem 5.4 .
We close the section by fixing some other notions. In this paper we always assume that H H H is any complex Hilbert space with dim H ≥ 2 \dim H\geq2 dim H ≥ 2 . For A ∈ B ( H ) A\in\mathcal{B}(H) A ∈ B ( H ) , ker A \ker A ker A and ran A \operatorname{ran} A ran A respectively stand for the kernel and the range of A A A . For any subspace H ′ H' H ′ of H H H , P [ H ′ ] P_{[H']} P [ H ′ ] denotes the projection operator onto H ′ H' H ′ . Write sphere H = { x ∈ H : ∥ x ∥ = 1 } \operatorname{sphere} H=\{x\in H:\lVert x\rVert=1\} sphere H = { x ∈ H : ∥ x ∥ = 1 } . For any set M M M , co ( M ) \operatorname{co}(M) co ( M ) denotes its convex hull. If H H H is finite dimensional with dim H = n \dim H=n dim H = n , then B ( H ) \mathcal{B}(H) B ( H ) can be regarded as the algebra of n × n n\times n n × n complex matrices M n ( C ) \mathcal{M}_n(\mathbb{C}) M n ( C ) , and in this case, let H n H_n H n denote the Jordan algebra of all self-adjoint matrices in M n ( C ) \mathcal{M}_n(\mathbb{C}) M n ( C ) . For any α ∈ C \alpha\in\mathbb{C} α ∈ C , arg α \arg\alpha arg α denotes its argument.
Properties of (strong) numerical radius orthogonality
In this section, we list some properties of (strong) numerical radius orthogonality that are useful for our purpose.
The following proposition gives a characterization of numerical radius orthogonality, which plays a key role in this paper.
Proposition 2.1. ([15 ] ) Assume that T , A ∈ B ( H ) T,A\in\mathcal{B}(H) T , A ∈ B ( H ) . The following statements are equivalent.
T ⊥ w A T\perp_w A T ⊥ w A .
α T ⊥ w β A \alpha T\perp_w\beta A α T ⊥ w β A for all α , β ∈ C ∖ { 0 } \alpha,\beta\in\mathbb{C}\setminus\{0\} α , β ∈ C ∖ { 0 } .
For each θ ∈ [ 0 , 2 π ) \theta\in[0,2\pi) θ ∈ [ 0 , 2 π ) , there exists a sequence { x n θ } ⊆ sphere H = { x ∈ H : ∥ x ∥ = 1 } \{x_n^\theta\}\subseteq\operatorname{sphere} H=\{x\in H:\lVert x\rVert=1\} { x n θ } ⊆ sphere H = { x ∈ H : ∥ x ∥ = 1 } such that lim n → ∞ ∣ ⟨ T x n θ , x n θ ⟩ ∣ = w ( T ) \lim_{n\to\infty}|\langle T x_n^\theta,x_n^\theta\rangle|=w(T) lim n → ∞ ∣ ⟨ T x n θ , x n θ ⟩ ∣ = w ( T ) and lim n → ∞ Re { e − i θ ⟨ T x n θ , x n θ ⟩ ⟨ A x n θ , x n θ ⟩ } ≥ 0 \lim_{n\to\infty}\operatorname{Re}\{e^{-i\theta}\langle T x_n^\theta,x_n^\theta\rangle\langle A x_n^\theta,x_n^\theta\rangle\}\geq0 lim n → ∞ Re { e − i θ ⟨ T x n θ , x n θ ⟩ ⟨ A x n θ , x n θ ⟩} ≥ 0 .
The following several propositions give some sufficient (or necessary) conditions for the strong numerical radius orthogonality.
Proposition 2.2. ([19 Proposition 3.2] ) Assume that A , B ∈ B ( H ) A,B\in\mathcal{B}(H) A , B ∈ B ( H ) . If there exists a sequence { x n } ⊂ sphere H \{x_n\}\subset\text{sphere }H { x n } ⊂ sphere H such that lim n → ∞ ∣ ⟨ A x n , x n ⟩ ∣ = w ( A ) \lim_{n\to\infty}|\langle Ax_n,x_n\rangle|=w(A) lim n → ∞ ∣ ⟨ A x n , x n ⟩ ∣ = w ( A ) and lim n → ∞ ∥ B ∗ x n ∥ = 0 \lim_{n\to\infty}\|B^{*}x_n\|=0 lim n → ∞ ∥ B ∗ x n ∥ = 0 , then A ⊥ w s B A\perp_{w}^{s}B A ⊥ w s B .
Proposition 2.3. ([19 Proposition 3.3] ) Assume that A , B ∈ B ( H ) A,B\in\mathcal{B}(H) A , B ∈ B ( H ) with 0 ≠ A 0\ne A 0 = A normaloid (that is, ∥ A ∥ = w ( A ) \|A\|=w(A) ∥ A ∥ = w ( A ) ). Then the following statements are equivalent.
(i) A ⊥ w s B A\perp_{w}^{s}B A ⊥ w s B .
(ii) There exist a sequence { x n } ⊂ sphere H \{x_n\}\subset\text{sphere }H { x n } ⊂ sphere H such that lim n → ∞ ∣ ⟨ A x n , x n ⟩ ∣ = w ( A ) \lim_{n\to\infty}|\langle Ax_n,x_n\rangle|=w(A) lim n → ∞ ∣ ⟨ A x n , x n ⟩ ∣ = w ( A ) and lim n → ∞ ∥ B ∗ x n ∥ = 0 \lim_{n\to\infty}\|B^{*}x_n\|=0 lim n → ∞ ∥ B ∗ x n ∥ = 0 .
Proposition 2.3 does not hold if the assumption that A A A is normaloid is removed.
Proposition 2.4. ([19 Proposition 3.6] ) Assume that T , A ∈ M 2 ( C ) T,A\in\mathcal{M}_2(\mathbb{C}) T , A ∈ M 2 ( C ) with T ≠ 0 T\ne0 T = 0 .
(i) If T 2 = 0 T^2=0 T 2 = 0 , that is, T = λ e 1 ⊗ e 2 T=\lambda e_1\otimes e_2 T = λ e 1 ⊗ e 2 for some λ ∈ C \lambda\in\mathbb{C} λ ∈ C and e 1 , e 2 ∈ sphere C 2 e_1,e_2\in\text{sphere }\mathbb{C}^2 e 1 , e 2 ∈ sphere C 2 with ⟨ e 1 , e 2 ⟩ = 0 \langle e_1,e_2\rangle=0 ⟨ e 1 , e 2 ⟩ = 0 , and if A = u ⊗ v A=u\otimes v A = u ⊗ v is of rank one with u , v ∈ C 2 u,v\in\mathbb{C}^2 u , v ∈ C 2 such that ∣ ⟨ u , e 1 ⟩ ∣ < ∣ ⟨ u , e 2 ⟩ ∣ |\langle u,e_1\rangle|<|\langle u,e_2\rangle| ∣ ⟨ u , e 1 ⟩ ∣ < ∣ ⟨ u , e 2 ⟩ ∣ , then T ⊥ w s A T\perp_w^sA T ⊥ w s A , but for any x ∈ M T x\in M_T x ∈ M T , A ∗ x ≠ 0 A^*x\ne0 A ∗ x = 0 .
(ii) If T 2 = 0 T^2=0 T 2 = 0 , that is, T = λ e 1 ⊗ e 2 T=\lambda e_1\otimes e_2 T = λ e 1 ⊗ e 2 for some λ ∈ C \lambda\in\mathbb{C} λ ∈ C and e 1 , e 2 ∈ sphere C 2 e_1,e_2\in\text{sphere }\mathbb{C}^2 e 1 , e 2 ∈ sphere C 2 with ⟨ e 1 , e 2 ⟩ = 0 \langle e_1,e_2\rangle=0 ⟨ e 1 , e 2 ⟩ = 0 , and if A = u ⊗ v A=u\otimes v A = u ⊗ v is of rank one with u , v ∈ C 2 u,v\in\mathbb{C}^2 u , v ∈ C 2 such that ∣ ⟨ u , e 1 ⟩ ∣ ≥ ∣ ⟨ u , e 2 ⟩ ∣ |\langle u,e_1\rangle|\ge|\langle u,e_2\rangle| ∣ ⟨ u , e 1 ⟩ ∣ ≥ ∣ ⟨ u , e 2 ⟩ ∣ , then T ⊥ w s A T\perp_w^sA T ⊥ w s A if and only if there exists some x ∈ M T x\in M_T x ∈ M T such that A ∗ x = 0 A^*x=0 A ∗ x = 0 .
(iii) If T 2 ≠ 0 T^2\ne0 T 2 = 0 , then T ⊥ w s A T\perp_w^sA T ⊥ w s A if and only if there exists some x ∈ M T x\in M_T x ∈ M T such that A ∗ x = 0 A^*x=0 A ∗ x = 0 .
Here, M T = { x ∈ sphere C 2 : ∣ ⟨ T x , x ⟩ ∣ = w ( T ) } M_T=\{x\in\text{sphere }\mathbb{C}^2:|\langle Tx,x\rangle|=w(T)\} M T = { x ∈ sphere C 2 : ∣ ⟨ T x , x ⟩ ∣ = w ( T )} .
Particularly, for square-zero matrices, Proposition 2.4 can be reformulated as follows.
Corollary 2.5. Assume that 0 ≠ B ∈ M 2 ( C ) 0\ne B\in\mathcal{M}_2(\mathbb{C}) 0 = B ∈ M 2 ( C ) , x , y ∈ sphere C 2 x,y\in\text{sphere }\mathbb{C}^2 x , y ∈ sphere C 2 with ⟨ x , y ⟩ = 0 \langle x,y\rangle=0 ⟨ x , y ⟩ = 0 . Then the following statements are equivalent.
(1) x ⊗ y ⊥ w s B x\otimes y\perp_w^sB x ⊗ y ⊥ w s B .
(2) B ∗ x B^*x B ∗ x and B ∗ y B^*y B ∗ y are linearly dependent and ∥ B ∗ y ∥ ≥ ∥ B ∗ x ∥ \|B^*y\|\ge\|B^*x\| ∥ B ∗ y ∥ ≥ ∥ B ∗ x ∥ .
In fact, Corollary 2.5 also holds for any rank-one nilpotent matrices in M n ( C ) \mathcal{M}_n(\mathbb{C}) M n ( C ) .
Proposition 2.6. Let n ≥ 3 n\ge3 n ≥ 3 . Assume that 0 ≠ A ∈ M n ( C ) 0\ne A\in\mathcal{M}_n(\mathbb{C}) 0 = A ∈ M n ( C ) , x , y ∈ sphere C n x,y\in\text{sphere }\mathbb{C}^n x , y ∈ sphere C n with ⟨ x , y ⟩ = 0 \langle x,y\rangle=0 ⟨ x , y ⟩ = 0 . Then the following statements are equivalent.
(1) x ⊗ y ⊥ w s A x\otimes y\perp_w^sA x ⊗ y ⊥ w s A .
(2) A ∗ x A^*x A ∗ x and A ∗ y A^*y A ∗ y are linearly dependent and ∥ A ∗ y ∥ ≥ ∥ A ∗ x ∥ \|A^*y\|\ge\|A^*x\| ∥ A ∗ y ∥ ≥ ∥ A ∗ x ∥ .
Proof. Take any A , B ∈ M n ( C ) A,B\in\mathcal{M}_n(\mathbb{C}) A , B ∈ M n ( C ) with A ≠ 0 A\ne0 A = 0 and any x , y ∈ s p h e r e C n x,y\in\mathrm{sphere}\,\mathbb{C}^n x , y ∈ sphere C n with ⟨ x , y ⟩ = 0 \langle x,y\rangle=0 ⟨ x , y ⟩ = 0 . It is easy to verify that
{ z ∈ s p h e r e C n : ∣ ⟨ ( x ⊗ y ) z , z ⟩ ∣ = w ( x ⊗ y ) } = { 1 2 ( x + e i δ y ) : δ ∈ [ 0 , 2 π ) } . (1) \{z\in\mathrm{sphere}\,\mathbb{C}^n:|\langle(x\otimes y)z,z\rangle|=w(x\otimes y)\}=\left\{\frac{1}{\sqrt{2}}(x+e^{i\delta}y):\delta\in[0,2\pi)\right\}. \tag*{(1)} { z ∈ sphere C n : ∣ ⟨( x ⊗ y ) z , z ⟩ ∣ = w ( x ⊗ y )} = { 2 1 ( x + e i δ y ) : δ ∈ [ 0 , 2 π ) } . (1)
Write z δ = 1 2 ( x + e i δ y ) z_\delta=\frac{1}{\sqrt{2}}(x+e^{i\delta}y) z δ = 2 1 ( x + e i δ y ) , and then ⟨ ( x ⊗ y ) z δ , z δ ⟩ = 1 2 e i δ \langle(x\otimes y)z_\delta,z_\delta\rangle=\frac{1}{2}e^{i\delta} ⟨( x ⊗ y ) z δ , z δ ⟩ = 2 1 e i δ . A direct calculation yields
⟨ A B z δ , z δ ⟩ = 1 2 ( ⟨ A B x , x ⟩ + e − i δ ⟨ A B x , y ⟩ + e i δ ⟨ A B y , x ⟩ + ⟨ A B y , y ⟩ ) , \langle ABz_\delta,z_\delta\rangle=\frac{1}{2}\left(\langle ABx,x\rangle+e^{-i\delta}\langle ABx,y\rangle+e^{i\delta}\langle ABy,x\rangle+\langle ABy,y\rangle\right), ⟨ A B z δ , z δ ⟩ = 2 1 ( ⟨ A B x , x ⟩ + e − i δ ⟨ A B x , y ⟩ + e i δ ⟨ A B y , x ⟩ + ⟨ A B y , y ⟩ ) ,
which means
⟨ ( x ⊗ y ) z δ , z δ ⟩ ‾ ⟨ A B z δ , z δ ⟩ = 1 4 [ e − 2 i δ ⟨ B x , A ∗ y ⟩ + e − i δ ( ⟨ B x , A ∗ x ⟩ + ⟨ B y , A ∗ y ⟩ ) + ⟨ B y , A ∗ x ⟩ ] . \overline{\langle(x\otimes y)z_\delta,z_\delta\rangle}\langle ABz_\delta,z_\delta\rangle=\frac{1}{4}\left[e^{-2i\delta}\langle Bx,A^*y\rangle+e^{-i\delta}\left(\langle Bx,A^*x\rangle+\langle By,A^*y\rangle\right)+\langle By,A^*x\rangle\right]. ⟨( x ⊗ y ) z δ , z δ ⟩ ⟨ A B z δ , z δ ⟩ = 4 1 [ e − 2 i δ ⟨ B x , A ∗ y ⟩ + e − i δ ( ⟨ B x , A ∗ x ⟩ + ⟨ B y , A ∗ y ⟩ ) + ⟨ B y , A ∗ x ⟩ ] .
Define a function f B f_B f B on C \mathbb{C} C by
f B ( u ) = ⟨ B x , A ∗ y ⟩ u 2 + ( ⟨ B x , A ∗ x ⟩ + ⟨ B y , A ∗ y ⟩ ) u + ⟨ B y , A ∗ x ⟩ . f_B(u)=\langle Bx,A^*y\rangle u^2+\left(\langle Bx,A^*x\rangle+\langle By,A^*y\rangle\right)u+\langle By,A^*x\rangle. f B ( u ) = ⟨ B x , A ∗ y ⟩ u 2 + ( ⟨ B x , A ∗ x ⟩ + ⟨ B y , A ∗ y ⟩ ) u + ⟨ B y , A ∗ x ⟩ .
By Proposition 2.1 , one has that
x ⊗ y ⊥ w A B if and only if 0 ∈ co { f B ( e i θ ) : θ ∈ [ 0 , 2 π ) } . (2) x\otimes y\perp_w AB\ \text{if and only if}\ 0\in\operatorname{co}\{f_B(e^{i\theta}):\theta\in[0,2\pi)\}. \tag*{(2)} x ⊗ y ⊥ w A B if and only if 0 ∈ co { f B ( e i θ ) : θ ∈ [ 0 , 2 π )} . (2)
Let N B N_B N B be the number of elements in the set { u : f B ( u ) = 0 and ∣ u ∣ < 1 } \{u:f_B(u)=0\text{ and }|u|<1\} { u : f B ( u ) = 0 and ∣ u ∣ < 1 } . Note that, according to the argument principle, the change in the argument of f B ( u ) f_B(u) f B ( u ) is 2 π N B 2\pi N_B 2 π N B after the independent variable u u u makes a complete counterclockwise (positive direction) circuit along { e i θ : θ ∈ [ 0 , 2 π ) } \{e^{i\theta}:\theta\in[0,2\pi)\} { e i θ : θ ∈ [ 0 , 2 π )} . Thus, we get
0 ∈ co { f B ( e − i θ ) : θ ∈ [ 0 , 2 π ) } if and only if N B ≥ 1. (3) 0\in\operatorname{co}\{f_B(e^{-i\theta}):\theta\in[0,2\pi)\}\ \text{if and only if}\ N_B\ge1. \tag*{(3)} 0 ∈ co { f B ( e − i θ ) : θ ∈ [ 0 , 2 π )} if and only if N B ≥ 1. (3)
( 2 ) ⇒ ( 1 ) (2)\Rightarrow(1) ( 2 ) ⇒ ( 1 ) : Assume that A ∗ x A^*x A ∗ x and A ∗ y A^*y A ∗ y are linearly dependent. If ∥ A ∗ y ∥ > ∥ A ∗ x ∥ > 0 \|A^*y\|>\|A^*x\|>0 ∥ A ∗ y ∥ > ∥ A ∗ x ∥ > 0 , writing A ∗ x = λ A ∗ y A^*x=\lambda A^*y A ∗ x = λ A ∗ y with λ ∈ C \lambda\in\mathbb{C} λ ∈ C and 0 < ∣ λ ∣ < 1 0<|\lambda|<1 0 < ∣ λ ∣ < 1 , then
f B ( u ) = ⟨ B x , A ∗ y ⟩ u 2 + ( λ ‾ ⟨ B x , A ∗ y ⟩ + ⟨ B y , A ∗ y ⟩ ) u + λ ‾ ⟨ B y , A ∗ y ⟩ . f_B(u)=\langle Bx,A^*y\rangle u^2+\left(\overline{\lambda}\langle Bx,A^*y\rangle+\langle By,A^*y\rangle\right)u+\overline{\lambda}\langle By,A^*y\rangle. f B ( u ) = ⟨ B x , A ∗ y ⟩ u 2 + ( λ ⟨ B x , A ∗ y ⟩ + ⟨ B y , A ∗ y ⟩ ) u + λ ⟨ B y , A ∗ y ⟩ .
This implies f B ( − λ ‾ ) = 0 f_B(-\overline{\lambda})=0 f B ( − λ ) = 0 . Now, using ∣ − λ ‾ ∣ < 1 |-\overline{\lambda}|<1 ∣ − λ ∣ < 1 and Eqs. (2) –(3) , we infer that x ⊗ y ⊥ w A B x\otimes y\perp_w AB x ⊗ y ⊥ w A B for all B ∈ M n ( C ) B\in\mathcal{M}_n(\mathbb{C}) B ∈ M n ( C ) , which means x ⊗ y ⊥ w s A x\otimes y\perp_w^s A x ⊗ y ⊥ w s A .
If ∥ A ∗ x ∥ = ∥ A ∗ y ∥ ≥ 0 \|A^*x\|=\|A^*y\|\ge0 ∥ A ∗ x ∥ = ∥ A ∗ y ∥ ≥ 0 , then there exists some δ ∈ [ 0 , 2 π ) \delta\in[0,2\pi) δ ∈ [ 0 , 2 π ) such that A ∗ z δ = 1 2 ( A ∗ x + e i δ A ∗ y ) = 0 A^*z_\delta=\frac{1}{\sqrt{2}}(A^*x+e^{i\delta}A^*y)=0 A ∗ z δ = 2 1 ( A ∗ x + e i δ A ∗ y ) = 0 , which and Proposition 2.2 imply x ⊗ y ⊥ w s A x\otimes y\perp_w^s A x ⊗ y ⊥ w s A .
If A ∗ x = 0 A^*x=0 A ∗ x = 0 and A ∗ y ≠ 0 A^*y\ne0 A ∗ y = 0 , then f B ( u ) = ⟨ B x , A ∗ y ⟩ u 2 + ⟨ B y , A ∗ y ⟩ u f_B(u)=\langle Bx,A^*y\rangle u^2+\langle By,A^*y\rangle u f B ( u ) = ⟨ B x , A ∗ y ⟩ u 2 + ⟨ B y , A ∗ y ⟩ u , which yields f B ( 0 ) = 0 f_B(0)=0 f B ( 0 ) = 0 , which and Eqs. (2) –(3) imply x ⊗ y ⊥ w s A x\otimes y\perp_w^s A x ⊗ y ⊥ w s A .
( 1 ) ⇒ ( 2 ) (1)\Rightarrow(2) ( 1 ) ⇒ ( 2 ) : If A ∗ x A^*x A ∗ x and A ∗ y A^*y A ∗ y are linearly independent, then we can take B 0 B_0 B 0 such that ⟨ B 0 x , A ∗ y ⟩ = ⟨ B 0 x , A ∗ x ⟩ = ⟨ B 0 y , A ∗ y ⟩ = 0 \langle B_0x,A^*y\rangle=\langle B_0x,A^*x\rangle=\langle B_0y,A^*y\rangle=0 ⟨ B 0 x , A ∗ y ⟩ = ⟨ B 0 x , A ∗ x ⟩ = ⟨ B 0 y , A ∗ y ⟩ = 0 and ⟨ B 0 y , A ∗ x ⟩ ≠ 0 \langle B_0y,A^*x\rangle\ne0 ⟨ B 0 y , A ∗ x ⟩ = 0 . So $f_{B_0}(u)=\langle B_0y,A^*x\rangle\ne 0. By Eq. (2) , we have x ⊗ y ̸ ⊥ w A B 0 x \otimes y \not\perp_{w} AB_{0} x ⊗ y ⊥ w A B 0 , which implies x ⊗ y ̸ ⊥ w s A x \otimes y \not\perp_{w}^{s} A x ⊗ y ⊥ w s A . Thus A ∗ x A^{*}x A ∗ x and A ∗ y A^{*}y A ∗ y must be linearly dependent.
If A ∗ y = 0 A^{*}y=0 A ∗ y = 0 and A ∗ x ≠ 0 A^{*}x\ne0 A ∗ x = 0 , then f B ( u ) = ⟨ B x , A ∗ x ⟩ u + ⟨ B y , A ∗ x ⟩ f_{B}(u)=\langle Bx,A^{*}x\rangle u+\langle By,A^{*}x\rangle f B ( u ) = ⟨ B x , A ∗ x ⟩ u + ⟨ B y , A ∗ x ⟩ . By taking B 0 B_{0} B 0 such that ∣ ⟨ B 0 y , A ∗ x ⟩ ∣ > ∣ ⟨ B 0 x , A ∗ x ⟩ ∣ \lvert\langle B_{0}y,A^{*}x\rangle\rvert>\lvert\langle B_{0}x,A^{*}x\rangle\rvert ∣⟨ B 0 y , A ∗ x ⟩∣ > ∣⟨ B 0 x , A ∗ x ⟩∣ , one gets that x ⊗ y ̸ ⊥ w A B 0 x\otimes y\not\perp_{w} AB_{0} x ⊗ y ⊥ w A B 0 by Eqs. (2) –(3) . Hence x ⊗ y ̸ ⊥ w s A x\otimes y\not\perp_{w}^{s} A x ⊗ y ⊥ w s A . If A ∗ x A^{*}x A ∗ x and A ∗ y A^{*}y A ∗ y are linearly dependent and ∥ A ∗ x ∥ > ∥ A ∗ y ∥ > 0 \lVert A^{*}x\rVert>\lVert A^{*}y\rVert>0 ∥ A ∗ x ∥ > ∥ A ∗ y ∥ > 0 , then take B 0 B_{0} B 0 such that ⟨ B 0 x , A ∗ y ⟩ = ⟨ B 0 x , A ∗ x ⟩ = 0 \langle B_{0}x,A^{*}y\rangle=\langle B_{0}x,A^{*}x\rangle=0 ⟨ B 0 x , A ∗ y ⟩ = ⟨ B 0 x , A ∗ x ⟩ = 0 and 0 < ∣ ⟨ B 0 y , A ∗ y ⟩ ∣ < ∣ ⟨ B 0 y , A ∗ x ⟩ ∣ 0<\lvert\langle B_{0}y,A^{*}y\rangle\rvert<\lvert\langle B_{0}y,A^{*}x\rangle\rvert 0 < ∣⟨ B 0 y , A ∗ y ⟩∣ < ∣⟨ B 0 y , A ∗ x ⟩∣ . Thus f B 0 ( u ) = ⟨ B 0 y , A ∗ y ⟩ u + ⟨ B 0 y , A ∗ x ⟩ f_{B_{0}}(u)=\langle B_{0}y,A^{*}y\rangle u+\langle B_{0}y,A^{*}x\rangle f B 0 ( u ) = ⟨ B 0 y , A ∗ y ⟩ u + ⟨ B 0 y , A ∗ x ⟩ . It follows from Eqs. (2) –(3) that x ⊗ y ̸ ⊥ w A B 0 x\otimes y\not\perp_{w} AB_{0} x ⊗ y ⊥ w A B 0 . Hence x ⊗ y ̸ ⊥ w s A x\otimes y\not\perp_{w}^{s} A x ⊗ y ⊥ w s A . So we must have ∥ A ∗ x ∥ ≤ ∥ A ∗ y ∥ \lVert A^{*}x\rVert\le\lVert A^{*}y\rVert ∥ A ∗ x ∥ ≤ ∥ A ∗ y ∥ , as desired.
The proof is finished. □
The following two propositions characterize invertible operators by the strong numerical radius orthogonality.
Proposition 2.7. ([19 Proposition 4.2] ) Assume that B ∈ B ( H ) B\in\mathcal{B}(H) B ∈ B ( H ) . Then the following statements are equivalent.
(i) For any A ∈ B ( H ) A\in\mathcal{B}(H) A ∈ B ( H ) , we have A ⊥ w s B ⇒ A = 0 A\perp_{w}^{s}B\Rightarrow A=0 A ⊥ w s B ⇒ A = 0 .
(ii) 0 ∉ σ r ( B ) 0\notin\sigma_{r}(B) 0 ∈ / σ r ( B ) , that is, B B B is right invertible.
Proposition 2.8. ([19 Corollary 4.3] ) Assume that B ∈ B s ( H ) B\in\mathcal{B}_{s}(H) B ∈ B s ( H ) . Then the following statements are equivalent.
(i) For any A ∈ B s ( H ) A\in\mathcal{B}_{s}(H) A ∈ B s ( H ) , we have A ⊥ w s B ⇒ A = 0 A\perp_{w}^{s}B\Rightarrow A=0 A ⊥ w s B ⇒ A = 0 .
(ii) 0 ∉ σ ( B ) 0\notin\sigma(B) 0 ∈ / σ ( B ) , that is, B B B is invertible.
By [19 Proposition 2.4] , numerical radius orthogonality and B-J orthogonality are equivalent for self-adjoint operators. However, the following proposition shows that strong numerical radius orthogonality and strong B-J orthogonality are not equivalent, and the strong numerical radius orthogonality is stronger.
Proposition 2.9. If A , B ∈ B ( H ) A,B\in\mathcal{B}(H) A , B ∈ B ( H ) with A A A being normaloid, then
A ⊥ w s B ⇒ A ⊥ B s B A\perp_{w}^{s}B\Rightarrow A\perp_{B}^{s}B A ⊥ w s B ⇒ A ⊥ B s B In addition, there exist some A , B ∈ B s ( H ) A,B\in\mathcal{B}_{s}(H) A , B ∈ B s ( H ) such that A ⊥ B s B A\perp_{B}^{s}B A ⊥ B s B and A ̸ ⊥ w s B A\not\perp_{w}^{s}B A ⊥ w s B .
Proof. Assume that A ∈ B ( H ) A\in\mathcal{B}(H) A ∈ B ( H ) is normaloid. If A ⊥ w s B A\perp_{w}^{s}B A ⊥ w s B for some B ∈ B ( H ) B\in\mathcal{B}(H) B ∈ B ( H ) , by Proposition 2.3, there exists a sequence { x n } ⊂ \{x_{n}\}\subset { x n } ⊂ sphere H H H such that
lim n → ∞ ∣ ⟨ A x n , x n ⟩ ∣ = w ( A ) = ∥ A ∥ and lim n → ∞ ∥ B ∗ x n ∥ = 0. \lim_{n\to\infty}\lvert\langle Ax_{n},x_{n}\rangle\rvert=w(A)=\lVert A\rVert
\quad\text{and}\quad
\lim_{n\to\infty}\lVert B^{*}x_{n}\rVert=0. n → ∞ lim ∣⟨ A x n , x n ⟩∣ = w ( A ) = ∥ A ∥ and n → ∞ lim ∥ B ∗ x n ∥ = 0.
As
∥ A ∥ = lim n → ∞ ∣ ⟨ A x n , x n ⟩ ∣ ≤ lim n → ∞ ∥ A x n ∥ ≤ ∥ A ∥ , \lVert A\rVert=\lim_{n\to\infty}\lvert\langle Ax_n,x_n\rangle\rvert\le\lim_{n\to\infty}\lVert Ax_n\rVert\le\lVert A\rVert, ∥ A ∥ = n → ∞ lim ∣⟨ A x n , x n ⟩∣ ≤ n → ∞ lim ∥ A x n ∥ ≤ ∥ A ∥ ,
one gets lim n → ∞ ∣ ⟨ A x n , x n ⟩ ∣ = lim n → ∞ ∥ A x n ∥ = ∥ A ∥ \lim_{n\to\infty}\lvert\langle Ax_n,x_n\rangle\rvert=\lim_{n\to\infty}\lVert Ax_n\rVert=\lVert A\rVert lim n → ∞ ∣⟨ A x n , x n ⟩∣ = lim n → ∞ ∥ A x n ∥ = ∥ A ∥ . Write lim n → ∞ ⟨ A x n , x n ⟩ = λ \lim_{n\to\infty}\langle Ax_n,x_n\rangle=\lambda lim n → ∞ ⟨ A x n , x n ⟩ = λ with ∣ λ ∣ = w ( A ) \lvert\lambda\rvert=w(A) ∣ λ ∣ = w ( A ) . Then
∥ A x n − λ x n ∥ 2 = ⟨ A x n − λ x n , A x n − λ x n ⟩ = ∥ A x n ∥ 2 − λ ‾ ⟨ A x n , x n ⟩ − λ ⟨ x n , A x n ⟩ + ∣ λ ∣ 2 . \lVert Ax_n-\lambda x_n\rVert^2=\langle Ax_n-\lambda x_n,Ax_n-\lambda x_n\rangle=\lVert Ax_n\rVert^2-\overline{\lambda}\langle Ax_n,x_n\rangle-\lambda\langle x_n,Ax_n\rangle+\lvert\lambda\rvert^2. ∥ A x n − λ x n ∥ 2 = ⟨ A x n − λ x n , A x n − λ x n ⟩ = ∥ A x n ∥ 2 − λ ⟨ A x n , x n ⟩ − λ ⟨ x n , A x n ⟩ + ∣ λ ∣ 2 .
It follows that lim n → ∞ ∥ A x n − λ x n ∥ = 0 \lim_{n\to\infty}\lVert Ax_n-\lambda x_n\rVert=0 lim n → ∞ ∥ A x n − λ x n ∥ = 0 . Also note that lim n → ∞ ∥ B ∗ A x n ∥ = lim n → ∞ ∥ λ B ∗ x n ∥ = 0 \lim_{n\to\infty}\lVert B^*Ax_n\rVert=\lim_{n\to\infty}\lVert\lambda B^*x_n\rVert=0 lim n → ∞ ∥ B ∗ A x n ∥ = lim n → ∞ ∥ λ B ∗ x n ∥ = 0 . By [4 Proposition 2.8] , we have A ⊥ B s B A\mathrel{\perp^{s}_{B}}B A ⊥ B s B .
Now, take A = x ⊗ y + y ⊗ x A=x\otimes y+y\otimes x A = x ⊗ y + y ⊗ x , which is self-adjoint, and B = x ⊗ x B=x\otimes x B = x ⊗ x with x , y ∈ sphere H x,y\in\operatorname{sphere} H x , y ∈ sphere H and ⟨ x , y ⟩ = 0 \langle x,y\rangle=0 ⟨ x , y ⟩ = 0 . Then ∥ A x ∥ = ∥ A ∥ \lVert Ax\rVert=\lVert A\rVert ∥ A x ∥ = ∥ A ∥ and B ∗ A x = 0 B^*Ax=0 B ∗ A x = 0 . This, by [4 Proposition 2.8] , implies that A ⊥ B s B A\mathrel{\perp^{s}_{B}}B A ⊥ B s B . However, if ∣ ⟨ A u , u ⟩ ∣ = w ( A ) \lvert\langle Au,u\rangle\rvert=w(A) ∣⟨ A u , u ⟩∣ = w ( A ) for some u ∈ sphere H u\in\operatorname{sphere} H u ∈ sphere H , then u = 2 2 ( x + y ) u=\frac{\sqrt{2}}{2}(x+y) u = 2 2 ( x + y ) or 2 2 ( x − y ) \frac{\sqrt{2}}{2}(x-y) 2 2 ( x − y ) . So B ∗ u ≠ 0 B^*u\ne0 B ∗ u = 0 . By Proposition 2.3, we see that A ̸ ⊥ w s B A\mathrel{\not\perp^{s}_{w}}B A ⊥ w s B .
The diameter of Γ ( A , ⊥ w ) \Gamma(A,\perp_{w}) Γ ( A , ⊥ w ) with A = M 2 ( C ) A=M_2(\mathbb{C}) A = M 2 ( C ) or H 2 H_2 H 2
It is known from [2 ] that, for dim H ≥ 2 \dim H\ge2 dim H ≥ 2 , both the numerical radius orthographs Γ ( B ( H ) , ⊥ w ) \Gamma(B(H),\perp_{w}) Γ ( B ( H ) , ⊥ w ) and Γ ( B s ( H ) , ⊥ w ) \Gamma(B_s(H),\perp_{w}) Γ ( B s ( H ) , ⊥ w ) are connected, that is, they have no isolated vertices. It is also known that Diam Γ ( B ( H ) , ⊥ w ) = Diam Γ ( B s ( H ) , ⊥ w ) = 2 \operatorname{Diam}\Gamma(B(H),\perp_{w})=\operatorname{Diam}\Gamma(B_s(H),\perp_{w})=2 Diam Γ ( B ( H ) , ⊥ w ) = Diam Γ ( B s ( H ) , ⊥ w ) = 2 if dim H ≥ 3 \dim H\ge3 dim H ≥ 3 . The purpose of this section is to give the diameters of these two graphs for the case when dim H = 2 \dim H=2 dim H = 2 .
In fact, we have the following result.
Theorem 3.1. Diam Γ ( M 2 ( C ) , ⊥ w ) = 3 \operatorname{Diam}\Gamma(M_2(\mathbb{C}),\perp_{w})=3 Diam Γ ( M 2 ( C ) , ⊥ w ) = 3 and Diam Γ ( H 2 , ⊥ w ) = 2 \operatorname{Diam}\Gamma(H_2,\perp_{w})=2 Diam Γ ( H 2 , ⊥ w ) = 2 .
To prove this theorem, several propositions are needed.
Proposition 3.2. Assume that x , y , u , v ∈ sphere C 2 x,y,u,v\in\operatorname{sphere}\mathbb{C}^2 x , y , u , v ∈ sphere C 2 with ⟨ x , y ⟩ = ⟨ u , v ⟩ = 0 \langle x,y\rangle=\langle u,v\rangle=0 ⟨ x , y ⟩ = ⟨ u , v ⟩ = 0 . Then x ⊗ y ⊥ w u ⊗ v x\otimes y\mathrel{\perp_{w}}u\otimes v x ⊗ y ⊥ w u ⊗ v if and only if ∣ ⟨ u , x ⟩ ∣ ≤ ∣ ⟨ u , y ⟩ ∣ \lvert\langle u,x\rangle\rvert\le\lvert\langle u,y\rangle\rvert ∣⟨ u , x ⟩∣ ≤ ∣⟨ u , y ⟩∣ .
Proof. Take any x , y , u , v ∈ sphere C 2 x,y,u,v\in\operatorname{sphere}\mathbb{C}^2 x , y , u , v ∈ sphere C 2 with ⟨ x , y ⟩ = ⟨ u , v ⟩ = 0 \langle x,y\rangle=\langle u,v\rangle=0 ⟨ x , y ⟩ = ⟨ u , v ⟩ = 0 . Without loss of generality, we can write u = α x + β y u=\alpha x+\beta y u = α x + β y with α , β ≥ 0 \alpha,\beta\ge0 α , β ≥ 0 ; otherwise, consider x ⊗ e i ( δ − θ ) y x\otimes e^{i(\delta-\theta)}y x ⊗ e i ( δ − θ ) y and e − i θ u ⊗ e i δ v e^{-i\theta}u\otimes e^{i\delta}v e − i θ u ⊗ e i δ v instead, where θ = arg α \theta=\arg\alpha θ = arg α and δ = arg β \delta=\arg\beta δ = arg β . Then v = β x − α y v=\beta x-\alpha y v = β x − α y . With respect to the space decomposition C 2 = [ x , y ] \mathbb{C}^2=[x,y] C 2 = [ x , y ] , one has
x ⊗ y = ( 0 1 0 0 ) and u ⊗ v = ( α β − α 2 β 2 − α β ) . x\otimes y=\begin{pmatrix}0&1\\0&0\end{pmatrix}\quad\text{and}\quad u\otimes v=\begin{pmatrix}\alpha\beta&-\alpha^2\\\beta^2&-\alpha\beta\end{pmatrix}. x ⊗ y = ( 0 0 1 0 ) and u ⊗ v = ( α β β 2 − α 2 − α β ) .
By Eq. (1) , we get
{ z ∈ sphere C 2 : ∣ ⟨ ( x ⊗ y ) z , z ⟩ ∣ = w ( x ⊗ y ) } = { z ξ : = 1 2 ( x + e i ξ y ) : ξ ∈ [ 0 , 2 π ) } . \{z\in\operatorname{sphere}\mathbb{C}^2:\lvert\langle(x\otimes y)z,z\rangle\rvert=w(x\otimes y)\}=\{z_\xi:=\tfrac{1}{\sqrt{2}}(x+e^{i\xi}y):\xi\in[0,2\pi)\}. { z ∈ sphere C 2 : ∣⟨( x ⊗ y ) z , z ⟩∣ = w ( x ⊗ y )} = { z ξ := 2 1 ( x + e i ξ y ) : ξ ∈ [ 0 , 2 π )} .
Furthermore, a simple calculation gives
⟨ ( x ⊗ y ) z ξ , z ξ ⟩ ⟨ ( u ⊗ v ) z ξ , z ξ ⟩ ‾ = 1 4 ( β 2 e − 2 i ξ − α 2 ) . \overline{\langle(x\otimes y)z_{\xi},z_{\xi}\rangle\langle(u\otimes v)z_{\xi},z_{\xi}\rangle}=\frac{1}{4}\left(\beta^{2}e^{-2i\xi}-\alpha^{2}\right). ⟨( x ⊗ y ) z ξ , z ξ ⟩ ⟨( u ⊗ v ) z ξ , z ξ ⟩ = 4 1 ( β 2 e − 2 i ξ − α 2 ) .
Now, by using the fact
β ≥ α ⟺ 0 ∈ co { β 2 e − 2 i ξ − α 2 : ξ ∈ [ 0 , 2 π ) } , \beta\geq\alpha\Longleftrightarrow0\in\operatorname{co}\left\{\beta^{2}e^{-2i\xi}-\alpha^{2}:\xi\in[0,2\pi)\right\}, β ≥ α ⟺ 0 ∈ co { β 2 e − 2 i ξ − α 2 : ξ ∈ [ 0 , 2 π ) } ,
one obtains that
x ⊗ y ⊥ w u ⊗ v ⟺ 0 ∈ co { ⟨ ( x ⊗ y ) z ξ , z ξ ⟩ ⟨ ( u ⊗ v ) z ξ , z ξ ⟩ ‾ : ξ ∈ [ 0 , 2 π ) } x\otimes y\perp_{w}u\otimes v\Longleftrightarrow0\in\operatorname{co}\left\{\overline{\langle(x\otimes y)z_{\xi},z_{\xi}\rangle\langle(u\otimes v)z_{\xi},z_{\xi}\rangle}:\xi\in[0,2\pi)\right\} x ⊗ y ⊥ w u ⊗ v ⟺ 0 ∈ co { ⟨( x ⊗ y ) z ξ , z ξ ⟩ ⟨( u ⊗ v ) z ξ , z ξ ⟩ : ξ ∈ [ 0 , 2 π ) }
⟺ β ≥ α ⟺ ∣ ⟨ u , y ⟩ ∣ ≥ ∣ ⟨ u , x ⟩ ∣ . \Longleftrightarrow\beta\geq\alpha\Longleftrightarrow|\langle u,y\rangle|\geq|\langle u,x\rangle|. ⟺ β ≥ α ⟺ ∣ ⟨ u , y ⟩ ∣ ≥ ∣ ⟨ u , x ⟩ ∣.
The proof is finished. □
Proposition 3.3. Assume that A = ( a 1 a 2 a 3 a 4 ) ∈ M 2 ( C ) A=\begin{pmatrix}a_{1}&a_{2}\\a_{3}&a_{4}\end{pmatrix}\in\mathcal{M}_{2}(\mathbb{C}) A = ( a 1 a 3 a 2 a 4 ) ∈ M 2 ( C ) . If ∣ a 1 ∣ = w ( A ) |a_{1}|=w(A) ∣ a 1 ∣ = w ( A ) , then ∣ a 2 ∣ = ∣ a 3 ∣ ≤ ∣ a 1 ∣ |a_{2}|=|a_{3}|\leq|a_{1}| ∣ a 2 ∣ = ∣ a 3 ∣ ≤ ∣ a 1 ∣ .
Proof. Assume that ∣ a 1 ∣ = w ( A ) |a_{1}|=w(A) ∣ a 1 ∣ = w ( A ) . Then ∣ a 2 ∣ = ∣ a 3 ∣ |a_{2}|=|a_{3}| ∣ a 2 ∣ = ∣ a 3 ∣ by [23 Lemma 1.5.13] .
Still, we need to prove ∣ a 2 ∣ ≤ ∣ a 1 ∣ |a_{2}|\leq|a_{1}| ∣ a 2 ∣ ≤ ∣ a 1 ∣ . Note that W ( A ) W(A) W ( A ) is an ellipse with foci { λ 1 , λ 2 } \{\lambda_{1},\lambda_{2}\} { λ 1 , λ 2 } and a minor axis of the length ( ∣ a 1 ∣ 2 + ∣ a 2 ∣ 2 + ∣ a 3 ∣ 2 + ∣ a 4 ∣ 2 − ∣ λ 1 ∣ 2 − ∣ λ 2 ∣ 2 ) 1 2 \left(|a_{1}|^{2}+|a_{2}|^{2}+|a_{3}|^{2}+|a_{4}|^{2}-|\lambda_{1}|^{2}-|\lambda_{2}|^{2}\right)^{\frac{1}{2}} ( ∣ a 1 ∣ 2 + ∣ a 2 ∣ 2 + ∣ a 3 ∣ 2 + ∣ a 4 ∣ 2 − ∣ λ 1 ∣ 2 − ∣ λ 2 ∣ 2 ) 2 1 , where σ ( A ) = { λ 1 , λ 2 } \sigma(A)=\{\lambda_{1},\lambda_{2}\} σ ( A ) = { λ 1 , λ 2 } ([9]). Let the center of W ( A ) W(A) W ( A ) be α + i β \alpha+i\beta α + i β with α , β ∈ R \alpha,\beta\in\mathbb{R} α , β ∈ R , the major axis length be 2 a 2a 2 a with direction θ ∈ [ 0 , π ) \theta\in[0,\pi) θ ∈ [ 0 , π ) , the minor axis length be 2 b 2b 2 b , and the focal distance be 2 c 2c 2 c . A simple calculation gives
{ λ 1 , λ 2 } = { α + c cos θ + i ( β + c sin θ ) , α − c cos θ + i ( β − c sin θ ) } . \{\lambda_{1},\lambda_{2}\}=\{\alpha+c\cos\theta+i(\beta+c\sin\theta),\alpha-c\cos\theta+i(\beta-c\sin\theta)\}. { λ 1 , λ 2 } = { α + c cos θ + i ( β + c sin θ ) , α − c cos θ + i ( β − c sin θ )} .
Using the length of the minor axis, we obtain
∣ a 1 ∣ 2 + ∣ a 2 ∣ 2 + ∣ a 3 ∣ 2 + ∣ a 4 ∣ 2 = 2 a 2 + 2 b 2 + 2 α 2 + 2 β 2 . (4) |a_{1}|^{2}+|a_{2}|^{2}+|a_{3}|^{2}+|a_{4}|^{2}=2a^{2}+2b^{2}+2\alpha^{2}+2\beta^{2}. \tag*{(4)} ∣ a 1 ∣ 2 + ∣ a 2 ∣ 2 + ∣ a 3 ∣ 2 + ∣ a 4 ∣ 2 = 2 a 2 + 2 b 2 + 2 α 2 + 2 β 2 . (4)
As a 1 + a 4 = tr A = λ 1 + λ 2 = 2 ( α + i β ) a_{1}+a_{4}=\operatorname{tr}A=\lambda_{1}+\lambda_{2}=2(\alpha+i\beta) a 1 + a 4 = tr A = λ 1 + λ 2 = 2 ( α + i β ) , we may set a 1 = α + s + i ( β + t ) a_{1}=\alpha+s+i(\beta+t) a 1 = α + s + i ( β + t ) and a 4 = α − s + i ( β − t ) a_{4}=\alpha-s+i(\beta-t) a 4 = α − s + i ( β − t ) with s , t ∈ R s,t\in\mathbb{R} s , t ∈ R . By the fact ∣ a 2 ∣ = ∣ a 3 ∣ |a_{2}|=|a_{3}| ∣ a 2 ∣ = ∣ a 3 ∣ and Eq. (4) , we infer that ∣ a 2 ∣ 2 = a 2 + b 2 − s 2 − t 2 |a_{2}|^{2}=a^{2}+b^{2}-s^{2}-t^{2} ∣ a 2 ∣ 2 = a 2 + b 2 − s 2 − t 2 . This implies
∣ a 2 ∣ 2 ≤ a 2 + b 2 − b 2 = a 2 . (5) |a_{2}|^{2}\leq a^{2}+b^{2}-b^{2}=a^{2}. \tag*{(5)} ∣ a 2 ∣ 2 ≤ a 2 + b 2 − b 2 = a 2 . (5)
In addition, note that μ 1 : = α + a cos θ + i ( β + a sin θ ) , μ 2 : = α − a cos θ + i ( β − a sin θ ) ∈ W ( A ) \mu_{1}:=\alpha+a\cos\theta+i(\beta+a\sin\theta),\ \mu_{2}:=\alpha-a\cos\theta+i(\beta-a\sin\theta)\in W(A) μ 1 := α + a cos θ + i ( β + a sin θ ) , μ 2 := α − a cos θ + i ( β − a sin θ ) ∈ W ( A ) . It follows that
α 2 + β 2 + a 2 + 2 a ∣ α cos θ + β sin θ ∣ = max { ∣ μ 1 ∣ , ∣ μ 2 ∣ } ≤ w ( A ) , \sqrt{\alpha^{2}+\beta^{2}+a^{2}+2a|\alpha\cos\theta+\beta\sin\theta|}=\max\{|\mu_{1}|,|\mu_{2}|\}\leq w(A), α 2 + β 2 + a 2 + 2 a ∣ α cos θ + β sin θ ∣ = max { ∣ μ 1 ∣ , ∣ μ 2 ∣ } ≤ w ( A ) ,
and so a ≤ w ( A ) a\leq w(A) a ≤ w ( A ) . Hence ∣ a 2 ∣ ≤ w ( A ) = ∣ a 1 ∣ |a_{2}|\leq w(A)=|a_{1}| ∣ a 2 ∣ ≤ w ( A ) = ∣ a 1 ∣ by Eq. (5) . □
Proposition 3.4. Assume that A ∈ M 2 ( C ) A \in\mathcal{M}_{2}(\mathbb{C}) A ∈ M 2 ( C ) . If ∣ ⟨ A x , x ⟩ ∣ = w ( A ) |\langle Ax,x\rangle|=w(A) ∣ ⟨ A x , x ⟩ ∣ = w ( A ) for some x ∈ s p h e r e C 2 x \in\mathrm{sphere}\,\mathbb{C}^{2} x ∈ sphere C 2 , then both A ⊥ ⊥ w x ⊗ y A \perp\!\!\perp_{w} x \otimes y A ⊥ ⊥ w x ⊗ y and A ⊥ ⊥ w y ⊗ x A \perp\!\!\perp_{w} y \otimes x A ⊥ ⊥ w y ⊗ x hold for all y ∈ s p h e r e C 2 y \in\mathrm{sphere}\,\mathbb{C}^{2} y ∈ sphere C 2 with ⟨ x , y ⟩ = 0 \langle x,y\rangle=0 ⟨ x , y ⟩ = 0 .
Proof. Take any y ∈ s p h e r e C 2 y \in\mathrm{sphere}\,\mathbb{C}^{2} y ∈ sphere C 2 with ⟨ x , y ⟩ = 0 \langle x,y\rangle=0 ⟨ x , y ⟩ = 0 and z δ = 1 2 ( x + e i δ y ) z_{\delta}=\frac{1}{\sqrt{2}}(x+e^{i\delta}y) z δ = 2 1 ( x + e i δ y ) . We will complete the proof by considering two cases.
Case 1. A A A is normal.
With respect to the space decomposition C 2 = [ x , y ] \mathbb{C}^{2}=[x,y] C 2 = [ x , y ] , A A A can be written as A = ( a 1 0 0 a 2 ) A=\begin{pmatrix}a_{1}&0\\0&a_{2}\end{pmatrix} A = ( a 1 0 0 a 2 ) with ∣ a 2 ∣ ≤ ∣ a 1 ∣ = w ( A ) |a_{2}|\leq|a_{1}|=w(A) ∣ a 2 ∣ ≤ ∣ a 1 ∣ = w ( A ) . It is easy to verify that A ⊥ w x ⊗ y A \perp_{w} x\otimes y A ⊥ w x ⊗ y . In addition, we have
⟨ ( x ⊗ y ) z δ , z δ ⟩ ‾ ⟨ A z δ , z δ ⟩ = 1 4 e − i δ ( a 1 + a 2 ) , \overline{\langle(x\otimes y)z_{\delta},z_{\delta}\rangle}\langle Az_{\delta},z_{\delta}\rangle=\frac{1}{4}e^{-i\delta}(a_{1}+a_{2}), ⟨( x ⊗ y ) z δ , z δ ⟩ ⟨ A z δ , z δ ⟩ = 4 1 e − i δ ( a 1 + a 2 ) ,
which means 0 ∈ co { ⟨ ( x ⊗ y ) z δ , z δ ⟩ ‾ ⟨ A z δ , z δ ⟩ : δ ∈ [ 0 , 2 π ) } 0\in\operatorname{co}\{\overline{\langle(x\otimes y)z_{\delta},z_{\delta}\rangle}\langle Az_{\delta},z_{\delta}\rangle:\delta\in[0,2\pi)\} 0 ∈ co { ⟨( x ⊗ y ) z δ , z δ ⟩ ⟨ A z δ , z δ ⟩ : δ ∈ [ 0 , 2 π )} . It follows from Proposition 2.1 and Eq. (1) that x ⊗ y ⊥ w A x\otimes y\perp_{w} A x ⊗ y ⊥ w A , and thus A ⊥ ⊥ w x ⊗ y A\perp\!\!\perp_{w}x\otimes y A ⊥ ⊥ w x ⊗ y .
Similarly, one can prove that A ⊥ ⊥ w y ⊗ x A\perp\!\!\perp_{w}y\otimes x A ⊥ ⊥ w y ⊗ x .
Case 2. A A A is not normal.
With respect to the space decomposition C 2 = [ x , y ] \mathbb{C}^{2}=[x,y] C 2 = [ x , y ] , A A A can be written as A = ( a 1 a 2 a 3 a 4 ) A=\begin{pmatrix}a_{1}&a_{2}\\a_{3}&a_{4}\end{pmatrix} A = ( a 1 a 3 a 2 a 4 ) .
By Proposition 3.3, one has ∣ a 2 ∣ = ∣ a 3 ∣ ≤ ∣ a 1 ∣ = ∣ ⟨ A x , x ⟩ ∣ = w ( A ) |a_{2}|=|a_{3}|\leq|a_{1}|=|\langle Ax,x\rangle|=w(A) ∣ a 2 ∣ = ∣ a 3 ∣ ≤ ∣ a 1 ∣ = ∣ ⟨ A x , x ⟩ ∣ = w ( A ) . It is easy to verify that A ⊥ w x ⊗ y A\perp_{w}x\otimes y A ⊥ w x ⊗ y . In addition, we have
⟨ ( x ⊗ y ) z δ , z δ ⟩ ‾ ⟨ A z δ , z δ ⟩ = 1 4 [ a 3 e − 2 i δ + ( a 1 + a 4 ) e − i δ + a 2 ] . \overline{\langle(x\otimes y)z_{\delta},z_{\delta}\rangle}\langle Az_{\delta},z_{\delta}\rangle=\frac{1}{4}[a_{3}e^{-2i\delta}+(a_{1}+a_{4})e^{-i\delta}+a_{2}]. ⟨( x ⊗ y ) z δ , z δ ⟩ ⟨ A z δ , z δ ⟩ = 4 1 [ a 3 e − 2 i δ + ( a 1 + a 4 ) e − i δ + a 2 ] .
Let f ( u ) = a 3 u 2 + ( a 1 + a 4 ) u + a 2 f(u)=a_{3}u^{2}+(a_{1}+a_{4})u+a_{2} f ( u ) = a 3 u 2 + ( a 1 + a 4 ) u + a 2 . To prove x ⊗ y ⊥ w A x\otimes y\perp_{w}A x ⊗ y ⊥ w A , by Proposition 2.1, it suffices to show
0 ∈ co { f ( e i ξ ) : ξ ∈ [ 0 , 2 π ) } . (6) 0\in\operatorname{co}\{f(e^{i\xi}):\xi\in[0,2\pi)\}. \tag*{(6)} 0 ∈ co { f ( e i ξ ) : ξ ∈ [ 0 , 2 π )} . (6)
In fact, if there exists some ξ ∈ [ 0 , 2 π ) \xi\in[0,2\pi) ξ ∈ [ 0 , 2 π ) such that f ( e i ξ ) = 0 f(e^{i\xi})=0 f ( e i ξ ) = 0 , then Eq. (6) holds trivially. If there does not exist ξ ∈ [ 0 , 2 π ) \xi\in[0,2\pi) ξ ∈ [ 0 , 2 π ) such that f ( e i ξ ) = 0 f(e^{i\xi})=0 f ( e i ξ ) = 0 , then let { u : f ( u ) = 0 } = { u 1 , u 2 } \{u:f(u)=0\}=\{u_{1},u_{2}\} { u : f ( u ) = 0 } = { u 1 , u 2 } . Note that ∣ u 1 u 2 ∣ = ∣ a 2 a 3 ∣ = 1 |u_{1}u_{2}|=\left|\frac{a_{2}}{a_{3}}\right|=1 ∣ u 1 u 2 ∣ = a 3 a 2 = 1 . Then there exists i ∈ { 1 , 2 } i\in\{1,2\} i ∈ { 1 , 2 } such that ∣ u i ∣ < 1 |u_{i}|<1 ∣ u i ∣ < 1 . Through a discussion similar to that of Eq. (3) , one achieves that Eq. (6) holds. Hence A ⊥ ⊥ w x ⊗ y A\perp\!\!\perp_{w}x\otimes y A ⊥ ⊥ w x ⊗ y .
Similarly, one can show A ⊥ ⊥ w y ⊗ x A\perp\!\!\perp_{w}y\otimes x A ⊥ ⊥ w y ⊗ x . The proof is finished. □
Proposition 3.5. If A = ( 1 0 0 a ) ∈ H 2 A=\begin{pmatrix}1&0\\0&a\end{pmatrix}\in\mathcal{H}_{2} A = ( 1 0 0 a ) ∈ H 2 with ∣ a ∣ ≤ 1 |a|\leq1 ∣ a ∣ ≤ 1 and B θ = ( 0 e i θ e − i θ 0 ) B_{\theta}=\begin{pmatrix}0&e^{i\theta}\\e^{-i\theta}&0\end{pmatrix} B θ = ( 0 e − i θ e i θ 0 ) with θ ∈ [ 0 , 2 π ) \theta\in[0,2\pi) θ ∈ [ 0 , 2 π ) , then A ⊥ ⊥ w B θ A\perp\!\!\perp_{w}B_{\theta} A ⊥ ⊥ w B θ .
Proof. On the one hand, by taking x = ( 1 0 ) x = \begin{pmatrix} 1 \\ 0 \end{pmatrix} x = ( 1 0 ) , one has
∣ ⟨ A x , x ⟩ ∣ = w ( A ) and ∣ ⟨ A x , x ⟩ ‾ ⟨ B θ x , x ⟩ ∣ = 0 , |\langle Ax,x\rangle|=w(A)\text{ and }|\overline{\langle Ax,x\rangle}\langle B_{\theta}x,x\rangle|=0, ∣ ⟨ A x , x ⟩ ∣ = w ( A ) and ∣ ⟨ A x , x ⟩ ⟨ B θ x , x ⟩ ∣ = 0 ,
which means A ⊥ w B θ A\perp_w B_{\theta} A ⊥ w B θ .
On the other hand, by taking x 1 = 1 2 ( e i θ 1 ) x_1=\frac{1}{\sqrt{2}}\begin{pmatrix} e^{i\theta} \\ 1 \end{pmatrix} x 1 = 2 1 ( e i θ 1 ) and x 2 = 1 2 ( e i θ − 1 ) x_2=\frac{1}{\sqrt{2}}\begin{pmatrix} e^{i\theta} \\ -1 \end{pmatrix} x 2 = 2 1 ( e i θ − 1 ) , a direct calculation gives
∣ ⟨ B θ x 1 , x 1 ⟩ ∣ = ∣ ⟨ B θ x 2 , x 2 ⟩ ∣ = w ( B ) |\langle B_{\theta}x_1,x_1\rangle|=|\langle B_{\theta}x_2,x_2\rangle|=w(B) ∣ ⟨ B θ x 1 , x 1 ⟩ ∣ = ∣ ⟨ B θ x 2 , x 2 ⟩ ∣ = w ( B )
and
0 ∈ co { ⟨ B θ x 1 , x 1 ⟩ ‾ ⟨ A x 1 , x 1 ⟩ , ⟨ B θ x 2 , x 2 ⟩ ‾ ⟨ A x 2 , x 2 ⟩ } = co { 1 + a 2 , − 1 + a 2 } . 0\in\operatorname{co}\left\{\overline{\langle B_{\theta}x_1,x_1\rangle}\langle Ax_1,x_1\rangle,\overline{\langle B_{\theta}x_2,x_2\rangle}\langle Ax_2,x_2\rangle\right\}=\operatorname{co}\left\{\frac{1+a}{2},-\frac{1+a}{2}\right\}. 0 ∈ co { ⟨ B θ x 1 , x 1 ⟩ ⟨ A x 1 , x 1 ⟩ , ⟨ B θ x 2 , x 2 ⟩ ⟨ A x 2 , x 2 ⟩ } = co { 2 1 + a , − 2 1 + a } .
These imply B θ ⊥ w A B_{\theta}\perp_w A B θ ⊥ w A . Hence A ⊥ ⊥ w B θ A\perp\!\!\perp_w B_{\theta} A ⊥ ⊥ w B θ .
Now, we are at the position to give a proof of Theorem 3.1 .
Proof of Theorem 3.1 . Firstly, for any A , B ∈ M 2 ( C ) A,B\in\mathcal{M}_2(\mathbb{C}) A , B ∈ M 2 ( C ) , there exist unit vectors x , u ∈ H x,u\in H x , u ∈ H such that ∣ ⟨ A x , x ⟩ ∣ = w ( A ) |\langle Ax,x\rangle|=w(A) ∣ ⟨ A x , x ⟩ ∣ = w ( A ) and ∣ ⟨ B u , u ⟩ ∣ = w ( B ) |\langle Bu,u\rangle|=w(B) ∣ ⟨ B u , u ⟩ ∣ = w ( B ) . It follows from Proposition 3.4 that
A ⊥ w x ⊗ y , A ⊥ w y ⊗ x , B ⊥ w u ⊗ v and B ⊥ w v ⊗ u , A\perp_w x\otimes y,\quad A\perp_w y\otimes x,\quad B\perp_w u\otimes v\text{ and }B\perp_w v\otimes u, A ⊥ w x ⊗ y , A ⊥ w y ⊗ x , B ⊥ w u ⊗ v and B ⊥ w v ⊗ u ,
where y , v ∈ sphere C 2 y,v\in\text{sphere }\mathbb{C}^2 y , v ∈ sphere C 2 with ⟨ x , y ⟩ = ⟨ u , v ⟩ = 0 \langle x,y\rangle=\langle u,v\rangle=0 ⟨ x , y ⟩ = ⟨ u , v ⟩ = 0 . By Proposition 3.2 , if ∣ ⟨ x , u ⟩ ∣ ≤ ∣ ⟨ y , u ⟩ ∣ |\langle x,u\rangle|\leq|\langle y,u\rangle| ∣ ⟨ x , u ⟩ ∣ ≤ ∣ ⟨ y , u ⟩ ∣ , we have
A ⊥ w x ⊗ y ⊥ w u ⊗ v ⊥ w B ; A\perp_w x\otimes y\perp_w u\otimes v\perp_w B; A ⊥ w x ⊗ y ⊥ w u ⊗ v ⊥ w B ;
if ∣ ⟨ x , u ⟩ ∣ > ∣ ⟨ y , u ⟩ ∣ |\langle x,u\rangle|>|\langle y,u\rangle| ∣ ⟨ x , u ⟩ ∣ > ∣ ⟨ y , u ⟩ ∣ , one has
A ⊥ w y ⊗ x ⊥ w u ⊗ v ⊥ w B . A\perp_w y\otimes x\perp_w u\otimes v\perp_w B. A ⊥ w y ⊗ x ⊥ w u ⊗ v ⊥ w B .
Hence Diam Γ ( M 2 ( C ) , ⊥ ⊥ w ) ≤ 3 \operatorname{Diam}\Gamma(\mathcal{M}_2(\mathbb{C}),\perp\!\!\perp_w)\leq3 Diam Γ ( M 2 ( C ) , ⊥ ⊥ w ) ≤ 3 .
So, to prove Diam Γ ( M 2 ( C ) , ⊥ ⊥ w ) = 3 \operatorname{Diam}\Gamma(\mathcal{M}_2(\mathbb{C}),\perp\!\!\perp_w)=3 Diam Γ ( M 2 ( C ) , ⊥ ⊥ w ) = 3 , it suffices to check that there exist A , B ∈ M 2 ( C ) A,B\in\mathcal{M}_2(\mathbb{C}) A , B ∈ M 2 ( C ) such that the distance between them is greater than 2. To do this, let A = ( 5 i i 1 ) A=\begin{pmatrix} 5 & i \\ i & 1 \end{pmatrix} A = ( 5 i i 1 ) and
B = ( 2 1 0 4 ) . B=\begin{pmatrix} 2 & 1 \\ 0 & 4 \end{pmatrix}. B = ( 2 0 1 4 ) .
A simple calculation yields that W ( A ) W(A) W ( A ) is an ellipse with the center 3, the major axis length 4, and the minor axis length 2. So 0 ∉ W ( A ) 0\notin W(A) 0 ∈ / W ( A ) , w ( A ) = 5 w(A)=5 w ( A ) = 5 and
{ x ∈ sphere C 2 : ∣ ⟨ A x , x ⟩ ∣ = w ( A ) } = { e i δ x 0 : x 0 = ( 1 0 ) , δ ∈ [ 0 , 2 π ) } . \left\{x\in\text{sphere }\mathbb{C}^2:|\langle Ax,x\rangle|=w(A)\right\}=\left\{e^{i\delta}x_0:x_0=\begin{pmatrix} 1 \\ 0 \end{pmatrix},\ \delta\in[0,2\pi)\right\}. { x ∈ sphere C 2 : ∣ ⟨ A x , x ⟩ ∣ = w ( A ) } = { e i δ x 0 : x 0 = ( 1 0 ) , δ ∈ [ 0 , 2 π ) } .
It follows that ⟨ A x 0 , x 0 ⟩ ⟨ B x 0 , x 0 ⟩ = 10 \langle Ax_0,x_0\rangle\langle Bx_0,x_0\rangle=10 ⟨ A x 0 , x 0 ⟩ ⟨ B x 0 , x 0 ⟩ = 10 , which means
A ̸ ⊥ w B . (7) A\not\perp_w B. \tag*{(7)} A ⊥ w B . (7)
Claim. If A ⊥ w C ⊥ w B A\perp_w C\perp_w B A ⊥ w C ⊥ w B for some C ∈ M 2 ( C ) C\in\mathcal{M}_2(\mathbb{C}) C ∈ M 2 ( C ) , then C = 0 C=0 C = 0 .
Write C = ( c 1 c 2 c 3 c 4 ) C=\begin{pmatrix} c_1 & c_2 \\ c_3 & c_4 \end{pmatrix} C = ( c 1 c 3 c 2 c 4 ) . As A ⊥ w C A\perp_w C A ⊥ w C , we must have c 1 = 0 c_1=0 c 1 = 0 . Suppose on the contrary that C ≠ 0 C\ne0 C = 0 .
Case 1. The center of W ( C ) W(C) W ( C ) is not at the origin.
In this case, there exists some x 0 ∈ s p h e r e C 2 x_0\in\mathrm{sphere}\,\mathbb{C}^2 x 0 ∈ sphere C 2 such that ∣ ⟨ C x 0 , x 0 ⟩ ∣ = w ( C ) |\langle Cx_0,x_0\rangle|=w(C) ∣ ⟨ C x 0 , x 0 ⟩ ∣ = w ( C ) and
{ x ∈ s p h e r e C 2 : ∣ ⟨ C x , x ⟩ ∣ = w ( C ) } = { e i δ x 0 : δ ∈ [ 0 , 2 π ) } . \{x\in\mathrm{sphere}\,\mathbb{C}^2:|\langle Cx,x\rangle|=w(C)\}=\{e^{i\delta}x_0:\delta\in[0,2\pi)\}. { x ∈ sphere C 2 : ∣ ⟨ C x , x ⟩ ∣ = w ( C )} = { e i δ x 0 : δ ∈ [ 0 , 2 π )} .
As C ⊥ w A C\perp_w A C ⊥ w A , we have ⟨ A x 0 , x 0 ⟩ = 0 \langle Ax_0,x_0\rangle=0 ⟨ A x 0 , x 0 ⟩ = 0 by Proposition 2.1, a contradiction to 0 ∉ W ( A ) 0\notin W(A) 0 ∈ / W ( A ) .
Case 2. The center of W ( C ) W(C) W ( C ) is at the origin, but W ( C ) W(C) W ( C ) is not a circle.
In this case, we must have c 4 = 0 c_4=0 c 4 = 0 . Thus, C C C can be written as C = ( 0 c 2 c 3 e i θ 0 ) C=\begin{pmatrix} 0 & c_2 \\ c_3e^{i\theta} & 0 \end{pmatrix} C = ( 0 c 3 e i θ c 2 0 ) with c 2 , c 3 > 0 c_2,c_3>0 c 2 , c 3 > 0 and θ ∈ [ 0 , 2 π ) \theta\in[0,2\pi) θ ∈ [ 0 , 2 π ) , which yields w ( C ) = c 2 + c 3 2 w(C)=\frac{c_2+c_3}{2} w ( C ) = 2 c 2 + c 3 . Note that
{ x ∈ s p h e r e C 2 : ∣ ⟨ C x , x ⟩ ∣ = w ( C ) } = { e i ξ x 1 : ξ ∈ [ 0 , 2 π ) } ∪ { e i ζ x 2 : ζ ∈ [ 0 , 2 π ) } . \{x\in\mathrm{sphere}\,\mathbb{C}^2:|\langle Cx,x\rangle|=w(C)\}=\{e^{i\xi}x_1:\xi\in[0,2\pi)\}\cup\{e^{i\zeta}x_2:\zeta\in[0,2\pi)\}. { x ∈ sphere C 2 : ∣ ⟨ C x , x ⟩ ∣ = w ( C )} = { e i ξ x 1 : ξ ∈ [ 0 , 2 π )} ∪ { e i ζ x 2 : ζ ∈ [ 0 , 2 π )} .
with x 1 = 1 2 ( 1 e i θ / 2 ) x_1=\frac{1}{\sqrt{2}}\begin{pmatrix}1\\e^{i\theta/2}\end{pmatrix} x 1 = 2 1 ( 1 e i θ /2 ) and x 2 = 1 2 ( 1 − e i θ / 2 ) x_2=\frac{1}{\sqrt{2}}\begin{pmatrix}1\\-e^{i\theta/2}\end{pmatrix} x 2 = 2 1 ( 1 − e i θ /2 ) .
For A A A , one has
⟨ A x 1 , x 1 ⟩ ⟨ C x 1 , x 1 ⟩ ‾ = 1 2 e − i θ / 2 ( c 2 + c 3 ) ( 3 + i cos θ 2 ) \langle Ax_1,x_1\rangle\overline{\langle Cx_1,x_1\rangle}=\frac{1}{2}e^{-i\theta/2}(c_2+c_3)\left(3+i\cos\frac{\theta}{2}\right) ⟨ A x 1 , x 1 ⟩ ⟨ C x 1 , x 1 ⟩ = 2 1 e − i θ /2 ( c 2 + c 3 ) ( 3 + i cos 2 θ )
and
⟨ A x 2 , x 2 ⟩ ⟨ C x 2 , x 2 ⟩ ‾ = − 1 2 e − i θ / 2 ( c 2 + c 3 ) ( 3 − i cos θ 2 ) . \langle Ax_2,x_2\rangle\overline{\langle Cx_2,x_2\rangle}=-\frac{1}{2}e^{-i\theta/2}(c_2+c_3)\left(3-i\cos\frac{\theta}{2}\right). ⟨ A x 2 , x 2 ⟩ ⟨ C x 2 , x 2 ⟩ = − 2 1 e − i θ /2 ( c 2 + c 3 ) ( 3 − i cos 2 θ ) .
As C ⊥ w A C\perp_w A C ⊥ w A , we infer that 0 ∈ co { ⟨ A x 1 , x 1 ⟩ ⟨ C x 1 , x 1 ⟩ ‾ , ⟨ A x 2 , x 2 ⟩ ⟨ C x 2 , x 2 ⟩ ‾ } 0\in\operatorname{co}\{\langle Ax_1,x_1\rangle\overline{\langle Cx_1,x_1\rangle},\langle Ax_2,x_2\rangle\overline{\langle Cx_2,x_2\rangle}\} 0 ∈ co {⟨ A x 1 , x 1 ⟩ ⟨ C x 1 , x 1 ⟩ , ⟨ A x 2 , x 2 ⟩ ⟨ C x 2 , x 2 ⟩ } , which yields θ = π \theta=\pi θ = π .
Thus, for B B B , a calculation gives
⟨ B x 1 , x 1 ⟩ ⟨ C x 1 , x 1 ⟩ ‾ = 1 4 e i π / 2 ( c 2 + c 3 ) ( 6 + i ) \langle Bx_1,x_1\rangle\overline{\langle Cx_1,x_1\rangle}=\frac{1}{4}e^{i\pi/2}(c_2+c_3)(6+i) ⟨ B x 1 , x 1 ⟩ ⟨ C x 1 , x 1 ⟩ = 4 1 e iπ /2 ( c 2 + c 3 ) ( 6 + i )
and
⟨ B x 2 , x 2 ⟩ ⟨ C x 2 , x 2 ⟩ ‾ = − 1 4 e − i π / 2 ( c 2 + c 3 ) ( 6 − i ) , \langle Bx_2,x_2\rangle\overline{\langle Cx_2,x_2\rangle}=-\frac{1}{4}e^{-i\pi/2}(c_2+c_3)(6-i), ⟨ B x 2 , x 2 ⟩ ⟨ C x 2 , x 2 ⟩ = − 4 1 e − iπ /2 ( c 2 + c 3 ) ( 6 − i ) ,
which means 0 ∉ co { ⟨ B x 1 , x 1 ⟩ ⟨ C x 1 , x 1 ⟩ ‾ , ⟨ B x 2 , x 2 ⟩ ⟨ C x 2 , x 2 ⟩ ‾ } 0\notin\operatorname{co}\{\langle Bx_1,x_1\rangle\overline{\langle Cx_1,x_1\rangle},\langle Bx_2,x_2\rangle\overline{\langle Cx_2,x_2\rangle}\} 0 ∈ / co {⟨ B x 1 , x 1 ⟩ ⟨ C x 1 , x 1 ⟩ , ⟨ B x 2 , x 2 ⟩ ⟨ C x 2 , x 2 ⟩ } , contradicting to C ⊥ w B C\perp_w B C ⊥ w B .
Case 3. W ( C ) W(C) W ( C ) is a disc centered at the origin. At this point, up to a nonzero scalar multiple, either C = ( 0 1 0 0 ) C = \begin{pmatrix} 0 & 1 \\ 0 & 0 \end{pmatrix} C = ( 0 0 1 0 ) or C = ( 0 0 1 0 ) C = \begin{pmatrix} 0 & 0 \\ 1 & 0 \end{pmatrix} C = ( 0 1 0 0 ) .
Note that the center of W ( B ) W(B) W ( B ) is not at the origin. As B ⊥ w C B \perp_w C B ⊥ w C , there exists a unit vector x ∈ sphere C 2 x \in\operatorname{sphere}\mathbb{C}^{2} x ∈ sphere C 2 such that
∣ ⟨ B x , x ⟩ ∣ = w ( B ) and ⟨ C x , x ⟩ = 0. |\langle Bx,x\rangle|=w(B) \quad\text{and} \quad\langle Cx,x\rangle=0. ∣ ⟨ B x , x ⟩ ∣ = w ( B ) and ⟨ C x , x ⟩ = 0.
which implies x = ( 1 0 ) x = \begin{pmatrix} 1 \\ 0 \end{pmatrix} x = ( 1 0 ) or ( 0 1 ) \begin{pmatrix} 0 \\ 1 \end{pmatrix} ( 0 1 ) . This is impossible.
Therefore, the claim is true.
Now, by Claim, it is clear that Diam Γ ( M 2 ( C ) , ⊥ w ) = 3 \operatorname{Diam}\Gamma(\mathcal{M}_{2}(\mathbb{C}),\perp_w)=3 Diam Γ ( M 2 ( C ) , ⊥ w ) = 3 .
Next, let us check that Diam Γ ( H 2 , ⊥ w ) = 2 \operatorname{Diam}\Gamma(\mathbb{H}_{2},\perp_w)=2 Diam Γ ( H 2 , ⊥ w ) = 2 .
For any A , B ∈ H 2 A,B\in\mathbb{H}_{2} A , B ∈ H 2 , write
A = a 1 x 1 ⊗ x 1 + a 2 x 2 ⊗ x 2 and B = b 1 y 1 ⊗ y 1 + b 2 y 2 ⊗ y 2 . A=a_{1}x_{1}\otimes x_{1}+a_{2}x_{2}\otimes x_{2}\quad\text{and}\quad B=b_{1}y_{1}\otimes y_{1}+b_{2}y_{2}\otimes y_{2}. A = a 1 x 1 ⊗ x 1 + a 2 x 2 ⊗ x 2 and B = b 1 y 1 ⊗ y 1 + b 2 y 2 ⊗ y 2 .
where x 1 , x 2 , y 1 , y 2 ∈ sphere C 2 x_{1},x_{2},y_{1},y_{2}\in\operatorname{sphere}\mathbb{C}^{2} x 1 , x 2 , y 1 , y 2 ∈ sphere C 2 , ⟨ x 1 , x 2 ⟩ = ⟨ y 1 , y 2 ⟩ = 0 \langle x_{1},x_{2}\rangle=\langle y_{1},y_{2}\rangle=0 ⟨ x 1 , x 2 ⟩ = ⟨ y 1 , y 2 ⟩ = 0 , a 1 , a 2 , b 1 , b 2 ∈ R a_{1},a_{2},b_{1},b_{2}\in\mathbb{R} a 1 , a 2 , b 1 , b 2 ∈ R , ∣ a 1 ∣ = w ( A ) |a_{1}|=w(A) ∣ a 1 ∣ = w ( A ) and ∣ b 1 ∣ = w ( B ) |b_{1}|=w(B) ∣ b 1 ∣ = w ( B ) . With respect to the space decomposition C 2 = [ x 1 , x 2 ] \mathbb{C}^{2}=[x_{1},x_{2}] C 2 = [ x 1 , x 2 ] , take C θ = ( 0 e i θ e − i θ 0 ) C_{\theta}=\begin{pmatrix} 0 & e^{i\theta} \\ e^{-i\theta} & 0 \end{pmatrix} C θ = ( 0 e − i θ e i θ 0 ) with any θ ∈ [ 0 , 2 π ) \theta\in[0,2\pi) θ ∈ [ 0 , 2 π ) . By Proposition 3.5, one gets A ⊥ w C θ A\perp_w C_{\theta} A ⊥ w C θ .
Write y 1 = α x 1 + β x 2 y_{1}=\alpha x_{1}+\beta x_{2} y 1 = α x 1 + β x 2 with α , β ∈ C \alpha,\beta\in\mathbb{C} α , β ∈ C . Choose an appropriate θ 0 ∈ [ 0 , 2 π ) \theta_{0}\in[0,2\pi) θ 0 ∈ [ 0 , 2 π ) such that ⟨ C θ 0 y 1 , y 1 ⟩ = 2 Re e i θ 0 α β ‾ = 0 \langle C_{\theta_{0}}y_{1},y_{1}\rangle=2\operatorname{Re}e^{i\theta_{0}}\alpha\overline{\beta}=0 ⟨ C θ 0 y 1 , y 1 ⟩ = 2 Re e i θ 0 α β = 0 . Then under the space decomposition C 2 = [ y 1 , y 2 ] \mathbb{C}^{2}=[y_{1},y_{2}] C 2 = [ y 1 , y 2 ] , we have
B = ( b 1 0 0 b 2 ) and C θ 0 = ( 0 μ μ ‾ 0 ) B=\begin{pmatrix} b_{1} & 0 \\ 0 & b_{2} \end{pmatrix}\quad\text{and}\quad C_{\theta_{0}}=\begin{pmatrix} 0 & \mu\\ \overline{\mu} & 0 \end{pmatrix} B = ( b 1 0 0 b 2 ) and C θ 0 = ( 0 μ μ 0 )
with μ ∈ C \mu\in\mathbb{C} μ ∈ C . This and Proposition 3.5 imply that B ⊥ w C θ 0 B\perp_w C_{\theta_{0}} B ⊥ w C θ 0 .
Thus, we have proved that A ⊥ w C θ 0 ⊥ w B A\perp_w C_{\theta_{0}}\perp_w B A ⊥ w C θ 0 ⊥ w B , that is, Diam Γ ( H 2 , ⊥ w ) ≤ 2 \operatorname{Diam}\Gamma(\mathbb{H}_{2},\perp_w)\leq2 Diam Γ ( H 2 , ⊥ w ) ≤ 2 . However, for
S = ( 1 0 0 1 2 ) and T = ( 1 0 0 1 3 ) , S=\begin{pmatrix} 1 & 0 \\ 0 & \frac{1}{2} \end{pmatrix}\quad\text{and}\quad T=\begin{pmatrix} 1 & 0 \\ 0 & \frac{1}{3} \end{pmatrix}, S = ( 1 0 0 2 1 ) and T = ( 1 0 0 3 1 ) ,
clearly, S ̸ ⊥ w T S\not\perp_w T S ⊥ w T . Hence Diam Γ ( H 2 , ⊥ w ) = 2 \operatorname{Diam}\Gamma(\mathbb{H}_{2},\perp_w)=2 Diam Γ ( H 2 , ⊥ w ) = 2 .
This completes the proof. □
The strong numerical radius orthograph Γ ( M n ( C ) , ⊥ w s ) \Gamma(\mathcal{M}_{n}(\mathbb{C}),\perp_w^{s}) Γ ( M n ( C ) , ⊥ w s )
In this section, we discuss the strong numerical radius orthograph Γ ( M n ( C ) , ⊥ w s ) \Gamma(\mathcal{M}_{n}(\mathbb{C}),\perp_w^{s}) Γ ( M n ( C ) , ⊥ w s ) .
The following result says that the set NI ( M n ( C ) ) \operatorname{NI}(\mathcal{M}_{n}(\mathbb{C})) NI ( M n ( C )) of all singular matrices is connected in Γ ( M n ( C ) , ⊥ w s ) \Gamma(\mathcal{M}_{n}(\mathbb{C}),\perp_w^{s}) Γ ( M n ( C ) , ⊥ w s ) with n ≥ 3 n\geq3 n ≥ 3 .
Proposition 4.1. Assume that n ≥ 3 n\geq3 n ≥ 3 and A ∈ NI ( M n ( C ) ) A\in\operatorname{NI}(\mathcal{M}_{n}(\mathbb{C})) A ∈ NI ( M n ( C )) is a nonzero operator. Then there exist mutually orthogonal vectors x , y ∈ sphere C n x,y\in\operatorname{sphere}\mathbb{C}^{n} x , y ∈ sphere C n such that $x\otimes y\perp_w^{s} A.
Proof. Assume that 0 ≠ A ∈ NI ( M n ( C ) ) 0 \ne A \in\operatorname{NI}(\mathcal{M}_n(\mathbb{C})) 0 = A ∈ NI ( M n ( C )) . Then there exist vectors y , v ∈ sphere C n y,v \in\operatorname{sphere}\mathbb{C}^n y , v ∈ sphere C n such that ∣ ⟨ A y , y ⟩ ∣ = w ( A ) |\langle Ay,y\rangle|=w(A) ∣ ⟨ A y , y ⟩ ∣ = w ( A ) and A ∗ v = 0 A^*v=0 A ∗ v = 0 . Let A y = α y + β u Ay=\alpha y+\beta u A y = α y + β u and v = a y + b u + c w v=ay+bu+cw v = a y + b u + c w , where { y , u , w } \{y,u,w\} { y , u , w } are mutually orthogonal unit vectors in C n \mathbb{C}^n C n and α , β , a , b , c ∈ C \alpha,\beta,a,b,c \in\mathbb{C} α , β , a , b , c ∈ C . It follows that
0 = ⟨ y , A ∗ v ⟩ = ⟨ A y , v ⟩ = ⟨ α y + β u , a y + b u + c w ⟩ = α a ‾ + β ‾ b . (8) 0=\langle y,A^*v\rangle=\langle Ay,v\rangle=\langle\alpha y+\beta u,ay+bu+cw\rangle=\alpha\overline{a}+\overline{\beta}b. \tag*{(8)} 0 = ⟨ y , A ∗ v ⟩ = ⟨ A y , v ⟩ = ⟨ α y + β u , a y + b u + c w ⟩ = α a + β b . (8)
Under the space decomposition C n = [ y , u ] ⊕ [ y , u ] ⊥ \mathbb{C}^n=[y,u]\oplus[y,u]^\perp C n = [ y , u ] ⊕ [ y , u ] ⊥ , we have A = ( α a 12 A 13 β a 22 A 23 0 a 32 A 33 ) A=\begin{pmatrix}\alpha&a_{12}&A_{13}\\ \beta&a_{22}&A_{23}\\ 0&a_{32}&A_{33}\end{pmatrix} A = α β 0 a 12 a 22 a 32 A 13 A 23 A 33 . Write A ′ = ( α a 12 β a 22 ) A'=\begin{pmatrix}\alpha&a_{12}\\ \beta&a_{22}\end{pmatrix} A ′ = ( α β a 12 a 22 ) , and then W ( A ′ ) ⊆ W ( A ) W(A')\subseteq W(A) W ( A ′ ) ⊆ W ( A ) by [23 Problems 8.21] . So
∣ α ∣ ≤ w ( A ′ ) ≤ w ( A ) = ∣ α ∣ . |\alpha|\leq w(A')\leq w(A)=|\alpha|. ∣ α ∣ ≤ w ( A ′ ) ≤ w ( A ) = ∣ α ∣.
It follows that ∣ α ∣ = w ( A ′ ) |\alpha|=w(A') ∣ α ∣ = w ( A ′ ) . By Proposition 3.3, one gets ∣ β ∣ ≤ ∣ α ∣ |\beta|\leq|\alpha| ∣ β ∣ ≤ ∣ α ∣ , which and Eq. (8) imply ∣ a ∣ ≤ ∣ b ∣ |a|\leq|b| ∣ a ∣ ≤ ∣ b ∣ . Obviously,
∣ a ∣ ≤ 1 2 . (9) |a|\leq\frac{1}{\sqrt{2}}. \tag*{(9)} ∣ a ∣ ≤ 2 1 . (9)
Take x = 1 ∥ b u + c w ∥ ( b u + c w ) x=\frac{1}{\|bu+cw\|}(bu+cw) x = ∥ b u + c w ∥ 1 ( b u + c w ) . Then ⟨ x , y ⟩ = 0 \langle x,y\rangle=0 ⟨ x , y ⟩ = 0 and v = a y + ∥ b u + c w ∥ x v=ay+\|bu+cw\|x v = a y + ∥ b u + c w ∥ x . Note that A ∗ v = a A ∗ y + ∥ b u + c w ∥ A ∗ x = 0 A^*v=aA^*y+\|bu+cw\|A^*x=0 A ∗ v = a A ∗ y + ∥ b u + c w ∥ A ∗ x = 0 . So, ∣ a ∣ ∥ A ∗ y ∥ = ∥ b u + c w ∥ ∥ A ∗ x ∥ |a|\|A^*y\|=\|bu+cw\|\|A^*x\| ∣ a ∣∥ A ∗ y ∥ = ∥ b u + c w ∥∥ A ∗ x ∥ . Combining it with Eq. (9) , we obtain ∥ A ∗ y ∥ ≥ ∥ A ∗ x ∥ \|A^*y\|\geq\|A^*x\| ∥ A ∗ y ∥ ≥ ∥ A ∗ x ∥ . Now, by Proposition 2.6, one has x ⊗ y ⊥ w s A x\otimes y\mathrel{\perp_w^s} A x ⊗ y ⊥ w s A ; and by Proposition 2.2, one gets A ⊥ w s x ⊗ y A\mathrel{\perp_w^s}x\otimes y A ⊥ w s x ⊗ y . Hence x ⊗ y ⊥ w ⊥ s A x\otimes y\mathrel{\perp_w^{\perp s}} A x ⊗ y ⊥ w ⊥ s A .
By the above proposition, we can characterize the connected components of the graph Γ ( M n ( C ) , ⊥ w s ) \Gamma(\mathcal{M}_n(\mathbb{C}),\mathrel{\perp_w^s}) Γ ( M n ( C ) , ⊥ w s ) and give a upper bound of Diam Γ ( NI ( M n ( C ) ) , ⊥ w s ) \operatorname{Diam}\Gamma(\operatorname{NI}(\mathcal{M}_n(\mathbb{C})),\mathrel{\perp_w^s}) Diam Γ ( NI ( M n ( C )) , ⊥ w s ) with n ≥ 3 n\geq3 n ≥ 3 .
Recall that the general linear group, denote by GL ( n , C ) \operatorname{GL}(n,\mathbb{C}) GL ( n , C ) , is the group of all invertible matrices in M n ( C ) \mathcal{M}_n(\mathbb{C}) M n ( C ) . Thus we have GL ( n , C ) = M n ( C ) ∖ NI ( M n ( C ) ) \operatorname{GL}(n,\mathbb{C})=\mathcal{M}_n(\mathbb{C})\setminus\operatorname{NI}(\mathcal{M}_n(\mathbb{C})) GL ( n , C ) = M n ( C ) ∖ NI ( M n ( C )) .
Theorem 4.2. Assume that n ≥ 3 n\geq3 n ≥ 3 . A connected component of Γ ( M n ( C ) , ⊥ w s ) \Gamma(\mathcal{M}_n(\mathbb{C}),\mathrel{\perp_w^s}) Γ ( M n ( C ) , ⊥ w s ) is either an isolated vertice or Γ ( NI ( M n ( C ) ) , ⊥ w s ) \Gamma(\operatorname{NI}(\mathcal{M}_n(\mathbb{C})),\mathrel{\perp_w^s}) Γ ( NI ( M n ( C )) , ⊥ w s ) . Moreover,
{ A ∈ M n ( C ) : A is an isolated vertex } = GL ( n , C ) . \{A\in\mathcal{M}_n(\mathbb{C}): A\text{ is an isolated vertex}\}=\operatorname{GL}(n,\mathbb{C}). { A ∈ M n ( C ) : A is an isolated vertex } = GL ( n , C ) .
Proof. The connectivity of Γ ( NI ( M n ( C ) ) , ⊥ w s ) \Gamma(\operatorname{NI}(\mathcal{M}_n(\mathbb{C})),\mathrel{\perp_w^s}) Γ ( NI ( M n ( C )) , ⊥ w s ) is clear by Theorem 4.1 and Proposition 2.6. For any invertible A ∈ M n ( C ) A\in\mathcal{M}_n(\mathbb{C}) A ∈ M n ( C ) , if A ⊥ w s B A\mathrel{\perp_w^s}B A ⊥ w s B for some B ∈ M n ( C ) B\in\mathcal{M}_n(\mathbb{C}) B ∈ M n ( C ) , then B ⊥ w s A B\mathrel{\perp_w^s}A B ⊥ w s A . This and Proposition 2.7 imply B = 0 B=0 B = 0 . So A A A is isolated.
Theorem 4.3. Diam Γ ( NI ( M n ( C ) ) , ⊥ w s ) ≤ 4 \operatorname{Diam}\Gamma(\operatorname{NI}(\mathcal{M}_n(\mathbb{C})),\mathrel{\perp_w^s})\leq4 Diam Γ ( NI ( M n ( C )) , ⊥ w s ) ≤ 4 for n ≥ 3 n\geq3 n ≥ 3 .
Proof. Take any A 1 , A 2 ∈ NI ( M n ( C ) ) A_1,A_2 \in\operatorname{NI}(\mathcal{M}_n(\mathbb{C})) A 1 , A 2 ∈ NI ( M n ( C )) . By Proposition 4.1, there exist x , y , u , v ∈ sphere C n x,y,u,v \in\operatorname{sphere}\mathbb{C}^n x , y , u , v ∈ sphere C n with ⟨ x , y ⟩ = ⟨ u , v ⟩ = 0 \langle x,y\rangle=\langle u,v\rangle=0 ⟨ x , y ⟩ = ⟨ u , v ⟩ = 0 such that x ⊗ y ⊥ w s A 1 x\otimes y \mathrel{\perp_w^s} A_1 x ⊗ y ⊥ w s A 1 and u ⊗ v ⊥ w s A 2 u\otimes v \mathrel{\perp_w^s} A_2 u ⊗ v ⊥ w s A 2 . Take any unit vectors e 1 ∈ [ x , u ] ⊥ e_1\in[x,u]^\perp e 1 ∈ [ x , u ] ⊥ and e 2 ∈ [ e 1 ] ⊥ e_2\in[e_1]^\perp e 2 ∈ [ e 1 ] ⊥ . By Proposition 2.6, one achieves
A 1 ⊥ w s x ⊗ y ⊥ w s e 1 ⊗ e 2 ⊥ w s u ⊗ v ⊥ w s A 2 , A_1 \mathrel{\perp_w^s} x\otimes y \mathrel{\perp_w^s} e_1\otimes e_2 \mathrel{\perp_w^s} u\otimes v \mathrel{\perp_w^s} A_2, A 1 ⊥ w s x ⊗ y ⊥ w s e 1 ⊗ e 2 ⊥ w s u ⊗ v ⊥ w s A 2 ,
which implies Diam Γ ( NI ( M n ( C ) ) , ⊥ w s ) ≤ 4 \operatorname{Diam}\Gamma(\operatorname{NI}(\mathcal{M}_n(\mathbb{C})),\mathrel{\perp_w^s})\leq4 Diam Γ ( NI ( M n ( C )) , ⊥ w s ) ≤ 4 . □ \square □
We do not know if exactly Diam Γ ( NI ( M n ( C ) ) , ⊥ w s ) = 4 \operatorname{Diam}\Gamma(\operatorname{NI}(\mathcal{M}_n(\mathbb{C})),\mathrel{\perp_w^s})=4 Diam Γ ( NI ( M n ( C )) , ⊥ w s ) = 4 for all 3 ≤ n < ∞ 3\leq n<\infty 3 ≤ n < ∞ . Moreover, we do not know what will happen to the strong numerical radius orthograph Γ ( B ( H ) , ⊥ w s ) \Gamma(\mathcal{B}(H),\mathrel{\perp_w^s}) Γ ( B ( H ) , ⊥ w s ) nor what the diameter of Γ ( NI ( B ( H ) ) , ⊥ w s ) \Gamma(\operatorname{NI}(\mathcal{B}(H)),\mathrel{\perp_w^s}) Γ ( NI ( B ( H )) , ⊥ w s ) is if dim H = ∞ \dim H=\infty dim H = ∞ .
In the rest of this section, we discuss the orthograph Γ ( M 2 ( C ) , ⊥ w s ) \Gamma(\mathcal{M}_2(\mathbb{C}),\mathrel{\perp_w^s}) Γ ( M 2 ( C ) , ⊥ w s ) , which is very different from the case when n ≥ 3 n\geq3 n ≥ 3 .
Let
NI 1 = { A ∈ NI ( M 2 ( C ) ) : A 2 ≠ 0 } and NI 2 = { A ∈ NI ( M 2 ( C ) ) : A 2 = 0 } . \operatorname{NI}_1=\{A\in\operatorname{NI}(\mathcal{M}_2(\mathbb{C})):A^2\ne0\}\quad\text{and}\quad\operatorname{NI}_2=\{A\in\operatorname{NI}(\mathcal{M}_2(\mathbb{C})):A^2=0\}. NI 1 = { A ∈ NI ( M 2 ( C )) : A 2 = 0 } and NI 2 = { A ∈ NI ( M 2 ( C )) : A 2 = 0 } .
It is obvious that NI ( M 2 ( C ) ) = NI 1 ∪ NI 2 \operatorname{NI}(\mathcal{M}_2(\mathbb{C}))=\operatorname{NI}_1\cup\operatorname{NI}_2 NI ( M 2 ( C )) = NI 1 ∪ NI 2 .
We first give some properties of the elements in NI 1 \operatorname{NI}_1 NI 1 and NI 2 \operatorname{NI}_2 NI 2 .
Proposition 4.4. For any A ∈ NI i A\in\operatorname{NI}_i A ∈ NI i , there exists some B ∈ NI j B\in\operatorname{NI}_j B ∈ NI j such that A ⊥ w s B A\mathrel{\perp_w^s}B A ⊥ w s B , where i ≠ j ∈ { 1 , 2 } i\ne j\in\{1,2\} i = j ∈ { 1 , 2 } .
Proof. Take any A ∈ NI 1 A\in\operatorname{NI}_1 A ∈ NI 1 . Then A A A can be written as A = x ⊗ ( α x + β y ) A=x\otimes(\alpha x+\beta y) A = x ⊗ ( α x + β y ) , where x , y ∈ sphere C 2 x,y\in\operatorname{sphere}\mathbb{C}^2 x , y ∈ sphere C 2 , ⟨ x , y ⟩ = 0 \langle x,y\rangle=0 ⟨ x , y ⟩ = 0 , α , β ∈ C \alpha,\beta\in\mathbb{C} α , β ∈ C and ∣ α ∣ 2 + ∣ β ∣ 2 = 1 |\alpha|^2+|\beta|^2=1 ∣ α ∣ 2 + ∣ β ∣ 2 = 1 with α ≠ 0 \alpha\ne0 α = 0 . Without loss of generality, we may assume that α , β ≥ 0 \alpha,\beta\geq0 α , β ≥ 0 ; otherwise, replace A A A by e i θ A = ∣ α ∣ x ⊗ x + ∣ β ∣ x ⊗ e i ( δ − θ ) y e^{i\theta}A=|\alpha|x\otimes x+|\beta|x\otimes e^{i(\delta-\theta)}y e i θ A = ∣ α ∣ x ⊗ x + ∣ β ∣ x ⊗ e i ( δ − θ ) y with θ = arg α \theta=\arg\alpha θ = arg α and δ = arg β \delta=\arg\beta δ = arg β . Let B = u ⊗ z ∈ NI 2 B=u\otimes z\in\operatorname{NI}_2 B = u ⊗ z ∈ NI 2 with
u = 1 − α 2 x − 1 + α 2 y , z = 1 + α 2 x + 1 − α 2 y . u=\sqrt{\frac{1-\alpha}{2}}x-\sqrt{\frac{1+\alpha}{2}}y,\qquad z=\sqrt{\frac{1+\alpha}{2}}x+\sqrt{\frac{1-\alpha}{2}}y. u = 2 1 − α x − 2 1 + α y , z = 2 1 + α x + 2 1 − α y . A direct calculation gives ∣ ⟨ A z , z ⟩ ∣ = w ( A ) = 1 + α 2 |\langle Az,z\rangle|=w(A)=\frac{1+\alpha}{2} ∣ ⟨ A z , z ⟩ ∣ = w ( A ) = 2 1 + α and B ∗ z = ( z ⊗ u ) z = 0 B^*z=(z\otimes u)z=0 B ∗ z = ( z ⊗ u ) z = 0 , which implies A ⊥ w s B A\mathrel{\perp_w^s}B A ⊥ w s B by Proposition 2.2. As ⟨ x , u ⟩ = 1 − α 2 \langle x,u\rangle=\sqrt{\frac{1-\alpha}{2}} ⟨ x , u ⟩ = 2 1 − α and ⟨ x , z ⟩ = 1 + α 2 \langle x,z\rangle=\sqrt{\frac{1+\alpha}{2}} ⟨ x , z ⟩ = 2 1 + α , noting that α ≠ 0 \alpha\ne0 α = 0 , we arrive at ∣ ⟨ x , u ⟩ ∣ < ∣ ⟨ x , z ⟩ ∣ |\langle x,u\rangle|<|\langle x,z\rangle| ∣ ⟨ x , u ⟩ ∣ < ∣ ⟨ x , z ⟩ ∣ , which entails B ⊥ w s A B\mathrel{\perp_w^s}A B ⊥ w s A . Hence A ⊥ w s B A\mathrel{\perp_w^s}B A ⊥ w s B by Proposition 2.4.
Take any A ∈ NI 2 A\in\operatorname{NI}_2 A ∈ NI 2 . Then A = u ⊗ z A=u\otimes z A = u ⊗ z for some u , z ∈ sphere H u,z\in\operatorname{sphere}H u , z ∈ sphere H and ⟨ u , z ⟩ = 0 \langle u,z\rangle=0 ⟨ u , z ⟩ = 0 . Let B = x ⊗ ( α x + 1 − α 2 y ) ∈ NI 1 B=x\otimes(\alpha x+\sqrt{1-\alpha^2}y)\in\operatorname{NI}_1 B = x ⊗ ( α x + 1 − α 2 y ) ∈ NI 1 , where
x = 1 − α 2 u + 1 + α 2 z , y = 1 − α 2 z − 1 + α 2 u , α ∈ ( 0 , 1 ) . x=\sqrt{\frac{1-\alpha}{2}}u+\sqrt{\frac{1+\alpha}{2}}z,\qquad y=\sqrt{\frac{1-\alpha}{2}}z-\sqrt{\frac{1+\alpha}{2}}u,\qquad\alpha\in(0,1). x = 2 1 − α u + 2 1 + α z , y = 2 1 − α z − 2 1 + α u , α ∈ ( 0 , 1 ) . By the same argument as the above, one can show that A ⊥ w s B A\mathrel{\perp_w^s}B A ⊥ w s B . □ \square □
Now, by Proposition 4.4 and a similar argument to that of Theorem 4.2 , the following result is immediate.
Theorem 4.5. A connected component of Γ ( M 2 ( C ) , ⊥ w s ) \Gamma(\mathcal{M}_2(\mathbb{C}),\perp_w^s) Γ ( M 2 ( C ) , ⊥ w s ) is either an isolated vertex or the set Γ ( NI ( M 2 ( C ) ) , ⊥ w s ) \Gamma(\operatorname{NI}(\mathcal{M}_2(\mathbb{C})),\perp_w^s) Γ ( NI ( M 2 ( C )) , ⊥ w s ) . Moreover,
{ A ∈ M 2 ( C ) : A is an isolated vertex } = GL ( 2 , C ) . \{A \in\mathcal{M}_2(\mathbb{C}) : A\text{ is an isolated vertex}\}=\operatorname{GL}(2,\mathbb{C}). { A ∈ M 2 ( C ) : A is an isolated vertex } = GL ( 2 , C ) .
The remaining part of this section is devoted to deriving the diameter of Γ ( NI ( M 2 ( C ) ) , ⊥ w s ) \Gamma(\operatorname{NI}(\mathcal{M}_2(\mathbb{C})),\perp_w^s) Γ ( NI ( M 2 ( C )) , ⊥ w s ) .
Proposition 4.6. For any A , B ∈ sphere NI 1 A,B\in\operatorname{sphere}\operatorname{NI}_1 A , B ∈ sphere NI 1 , if A ⊥ w s B A\perp_w^s B A ⊥ w s B , then ∣ tr A ∣ = ∣ tr B ∣ |\operatorname{tr} A|=|\operatorname{tr} B| ∣ tr A ∣ = ∣ tr B ∣ .
Proof. Take any A , B ∈ sphere NI 1 A,B\in\operatorname{sphere}\operatorname{NI}_1 A , B ∈ sphere NI 1 . We can write
A = x ⊗ ( α x + β y ) and B = u ⊗ v , A=x\otimes(\alpha x+\beta y)\quad\text{and}\quad B=u\otimes v, A = x ⊗ ( α x + β y ) and B = u ⊗ v ,
where x , y , u , v ∈ sphere C 2 x,y,u,v\in\operatorname{sphere}\mathbb{C}^2 x , y , u , v ∈ sphere C 2 , ⟨ x , y ⟩ = 0 \langle x,y\rangle=0 ⟨ x , y ⟩ = 0 , α , β ∈ C ∖ { 0 } \alpha,\beta\in\mathbb{C}\setminus\{0\} α , β ∈ C ∖ { 0 } and ∣ α ∣ 2 + ∣ β ∣ 2 = 1 |\alpha|^2+|\beta|^2=1 ∣ α ∣ 2 + ∣ β ∣ 2 = 1 . Let α = e − i δ ∣ α ∣ \alpha=e^{-i\delta}|\alpha| α = e − i δ ∣ α ∣ and β = e − i θ ∣ β ∣ \beta=e^{-i\theta}|\beta| β = e − i θ ∣ β ∣ with δ , θ ∈ [ 0 , 2 π ) \delta,\theta\in[0,2\pi) δ , θ ∈ [ 0 , 2 π ) . Through a simple calculation, one obtains
{ z ∈ sphere C 2 : ∣ ⟨ A z , z ⟩ ∣ = w ( A ) } = { e i ξ z 0 : ξ ∈ [ 0 , 2 π ) } \{z\in\operatorname{sphere}\mathbb{C}^2:|\langle Az,z\rangle|=w(A)\}=\{e^{i\xi}z_0:\xi\in[0,2\pi)\} { z ∈ sphere C 2 : ∣ ⟨ A z , z ⟩ ∣ = w ( A )} = { e i ξ z 0 : ξ ∈ [ 0 , 2 π )}
with
z 0 = 1 + ∣ α ∣ 2 x + 1 − ∣ α ∣ 2 e i ( δ − θ ) y . (10) z_0=\sqrt{\frac{1+|\alpha|}{2}}x+\sqrt{\frac{1-|\alpha|}{2}}e^{i(\delta-\theta)}y. \tag*{(10)} z 0 = 2 1 + ∣ α ∣ x + 2 1 − ∣ α ∣ e i ( δ − θ ) y . (10)
Since A ⊥ w s B A\perp_w^s B A ⊥ w s B , we have B ∗ z 0 = ( v ⊗ u ) z 0 = 0 B^*z_0=(v\otimes u)z_0=0 B ∗ z 0 = ( v ⊗ u ) z 0 = 0 , and so ⟨ z 0 , u ⟩ = 0 \langle z_0,u\rangle=0 ⟨ z 0 , u ⟩ = 0 . Then
u = 1 − ∣ α ∣ 2 e − i ( δ − θ ) x − 1 + ∣ α ∣ 2 y . (11) u=\sqrt{\frac{1-|\alpha|}{2}}e^{-i(\delta-\theta)}x-\sqrt{\frac{1+|\alpha|}{2}}y. \tag*{(11)} u = 2 1 − ∣ α ∣ e − i ( δ − θ ) x − 2 1 + ∣ α ∣ y . (11)
Then v = λ u + γ z 0 v=\lambda u+\gamma z_0 v = λ u + γ z 0 with λ , γ ∈ C \lambda,\gamma\in\mathbb{C} λ , γ ∈ C and ∣ λ ∣ 2 + ∣ γ ∣ 2 = 1 |\lambda|^2+|\gamma|^2=1 ∣ λ ∣ 2 + ∣ γ ∣ 2 = 1 , that is, B = u ⊗ ( λ u + γ z 0 ) B=u\otimes(\lambda u+\gamma z_0) B = u ⊗ ( λ u + γ z 0 ) . Write λ = ∣ λ ∣ e − i δ ′ \lambda=|\lambda|e^{-i\delta'} λ = ∣ λ ∣ e − i δ ′ and γ = ∣ γ ∣ e − i θ ′ \gamma=|\gamma|e^{-i\theta'} γ = ∣ γ ∣ e − i θ ′ . Similarly, we have
{ w ∈ sphere C 2 : ∣ ⟨ B w , w ⟩ ∣ = w ( B ) } = { e i ξ w 0 : ξ ∈ [ 0 , 2 π ) } \{w\in\operatorname{sphere}\mathbb{C}^2:|\langle Bw,w\rangle|=w(B)\}=\{e^{i\xi}w_0:\xi\in[0,2\pi)\} { w ∈ sphere C 2 : ∣ ⟨ B w , w ⟩ ∣ = w ( B )} = { e i ξ w 0 : ξ ∈ [ 0 , 2 π )}
with
w 0 = 1 + ∣ λ ∣ 2 u + 1 − ∣ λ ∣ 2 e i ( δ ′ − θ ′ ) z 0 . w_0=\sqrt{\frac{1+|\lambda|}{2}}u+\sqrt{\frac{1-|\lambda|}{2}}e^{i(\delta'-\theta')}z_0. w 0 = 2 1 + ∣ λ ∣ u + 2 1 − ∣ λ ∣ e i ( δ ′ − θ ′ ) z 0 .
By Eqs. (10) –(11) , one gets
w 0 = ( e − i ( δ − θ ) 1 + ∣ λ ∣ 2 1 − ∣ α ∣ 2 + e i ( δ ′ − θ ′ ) 1 − ∣ λ ∣ 2 1 + ∣ α ∣ 2 ) x + ( − 1 + ∣ λ ∣ 2 1 + ∣ α ∣ 2 + e i ( δ ′ − θ ′ + δ − θ ) 1 − ∣ λ ∣ 2 1 − ∣ α ∣ 2 ) y . \begin{aligned}
w_0={}&\left(e^{-i(\delta-\theta)}\sqrt{\frac{1+|\lambda|}{2}}\sqrt{\frac{1-|\alpha|}{2}}+e^{i(\delta'-\theta')}\sqrt{\frac{1-|\lambda|}{2}}\sqrt{\frac{1+|\alpha|}{2}}\right)x\\
&+\left(-\sqrt{\frac{1+|\lambda|}{2}}\sqrt{\frac{1+|\alpha|}{2}}+e^{i(\delta'-\theta'+\delta-\theta)}\sqrt{\frac{1-|\lambda|}{2}}\sqrt{\frac{1-|\alpha|}{2}}\right)y.
\end{aligned} w 0 = ( e − i ( δ − θ ) 2 1 + ∣ λ ∣ 2 1 − ∣ α ∣ + e i ( δ ′ − θ ′ ) 2 1 − ∣ λ ∣ 2 1 + ∣ α ∣ ) x + ( − 2 1 + ∣ λ ∣ 2 1 + ∣ α ∣ + e i ( δ ′ − θ ′ + δ − θ ) 2 1 − ∣ λ ∣ 2 1 − ∣ α ∣ ) y .
As B ⊥ w s A B\perp_w^s A B ⊥ w s A , we infer that A ∗ w 0 = ( ( α x + β y ) ⊗ x ) w 0 = 0 A^*w_0=((\alpha x+\beta y)\otimes x)w_0=0 A ∗ w 0 = (( α x + β y ) ⊗ x ) w 0 = 0 , which yields ⟨ w 0 , x ⟩ = 0 \langle w_0,x\rangle=0 ⟨ w 0 , x ⟩ = 0 . So
e − i ( δ − θ ) 1 + ∣ λ ∣ 2 1 − ∣ α ∣ 2 + e i ( δ ′ − θ ′ ) 1 − ∣ λ ∣ 2 1 + ∣ α ∣ 2 = 0 , e^{-i(\delta-\theta)}\sqrt{\frac{1+|\lambda|}{2}}\sqrt{\frac{1-|\alpha|}{2}}+e^{i(\delta'-\theta')}\sqrt{\frac{1-|\lambda|}{2}}\sqrt{\frac{1+|\alpha|}{2}}=0, e − i ( δ − θ ) 2 1 + ∣ λ ∣ 2 1 − ∣ α ∣ + e i ( δ ′ − θ ′ ) 2 1 − ∣ λ ∣ 2 1 + ∣ α ∣ = 0 ,
which forces ∣ λ ∣ = ∣ α ∣ |\lambda|=|\alpha| ∣ λ ∣ = ∣ α ∣ . As ∣ tr A ∣ = ∣ tr ( x ⊗ ( α x + β y ) ) ∣ = ∣ α ∣ |\operatorname{tr} A|=|\operatorname{tr}(x\otimes(\alpha x+\beta y))|=|\alpha| ∣ tr A ∣ = ∣ tr ( x ⊗ ( α x + β y )) ∣ = ∣ α ∣ and ∣ tr B ∣ = ∣ tr ( u ⊗ ( λ u + γ z 0 ) ) ∣ = ∣ λ ∣ |\operatorname{tr} B|=|\operatorname{tr}(u\otimes(\lambda u+\gamma z_{0}))|=|\lambda| ∣ tr B ∣ = ∣ tr ( u ⊗ ( λ u + γ z 0 )) ∣ = ∣ λ ∣ , it follows that ∣ tr A ∣ = ∣ tr B ∣ |\operatorname{tr} A|=|\operatorname{tr} B| ∣ tr A ∣ = ∣ tr B ∣ .
Proposition 4.7. If A , B ∈ N I 2 A,B\in\mathrm{NI}_{2} A , B ∈ NI 2 , then either A ⊥ ⊥ w s B A\perp\!\!\perp_{w}^{s}B A ⊥ ⊥ w s B or A ⊥ ⊥ w s C ⊥ ⊥ w s B A\perp\!\!\perp_{w}^{s}C\perp\!\!\perp_{w}^{s}B A ⊥ ⊥ w s C ⊥ ⊥ w s B for some C ∈ N I 2 C\in\mathrm{NI}_{2} C ∈ NI 2 .
Proof. For any A , B ∈ N I 2 A,B\in\mathrm{NI}_{2} A , B ∈ NI 2 , without loss of generality, we may assume ∥ A ∥ = ∥ B ∥ = 1 \|A\|=\|B\|=1 ∥ A ∥ = ∥ B ∥ = 1 . Then A = x ⊗ y A=x\otimes y A = x ⊗ y and B = u ⊗ v B=u\otimes v B = u ⊗ v for some x , y , u , v ∈ sphere C 2 x,y,u,v\in\operatorname{sphere}\mathbb{C}^{2} x , y , u , v ∈ sphere C 2 with ⟨ x , y ⟩ = ⟨ u , v ⟩ = 0 \langle x,y\rangle=\langle u,v\rangle=0 ⟨ x , y ⟩ = ⟨ u , v ⟩ = 0 and u = α x + β y u=\alpha x+\beta y u = α x + β y with some α , β ∈ C \alpha,\beta\in\mathbb{C} α , β ∈ C . Also we may assume that α , β ≥ 0 \alpha,\beta\ge0 α , β ≥ 0 and thus v = β x − α y v=\beta x-\alpha y v = β x − α y .
Indeed, letting α = ∣ α ∣ e i θ \alpha=|\alpha|e^{i\theta} α = ∣ α ∣ e i θ and β = ∣ β ∣ e i δ \beta=|\beta|e^{i\delta} β = ∣ β ∣ e i δ , we have
e − i θ u = ∣ α ∣ x + ∣ β ∣ e i ( δ − θ ) y e^{-i\theta}u=|\alpha|x+|\beta|e^{i(\delta-\theta)}y e − i θ u = ∣ α ∣ x + ∣ β ∣ e i ( δ − θ ) y and
e i δ v = ∣ β ∣ x − ∣ α ∣ e i ( δ − θ ) y . e^{i\delta}v=|\beta|x-|\alpha|e^{i(\delta-\theta)}y. e i δ v = ∣ β ∣ x − ∣ α ∣ e i ( δ − θ ) y . Thus we may replace x ⊗ e i ( δ − θ ) y x\otimes e^{i(\delta-\theta)}y x ⊗ e i ( δ − θ ) y and e − i θ u ⊗ e i δ v e^{-i\theta}u\otimes e^{i\delta}v e − i θ u ⊗ e i δ v with A A A and B B B , respectively.
If α = 0 \alpha=0 α = 0 , then u = y u=y u = y , v = x v=x v = x , and so B = y ⊗ x B=y\otimes x B = y ⊗ x . It is obvious that A ⊥ ⊥ w s B A\perp\!\!\perp_{w}^{s}B A ⊥ ⊥ w s B .
If α = 1 \alpha=1 α = 1 , then u = x u=x u = x , v = − y v=-y v = − y , and so B = − x ⊗ y B=-x\otimes y B = − x ⊗ y . It follows that A ⊥ ⊥ w s C ⊥ ⊥ w s B A\perp\!\!\perp_{w}^{s}C\perp\!\!\perp_{w}^{s}B A ⊥ ⊥ w s C ⊥ ⊥ w s B with C = y ⊗ x C=y\otimes x C = y ⊗ x .
So, we may assume that 0 < α < 1 0<\alpha<1 0 < α < 1 . Take C = z ⊗ w ∈ N I 2 C=z\otimes w\in\mathrm{NI}_{2} C = z ⊗ w ∈ NI 2 with z , w ∈ sphere C 2 z,w\in\operatorname{sphere}\mathbb{C}^{2} z , w ∈ sphere C 2 , where z = λ x + γ y z=\lambda x+\gamma y z = λ x + γ y , w = γ x − λ y w=\gamma x-\lambda y w = γ x − λ y , λ > 0 \lambda>0 λ > 0 , γ < 0 \gamma<0 γ < 0 and ∣ λ ∣ < ∣ γ ∣ |\lambda|<|\gamma| ∣ λ ∣ < ∣ γ ∣ . As
lim λ → ( 1 2 ) − 4 λ γ γ 2 − λ 2 = − ∞ , \lim_{\lambda\to\left(\frac{1}{\sqrt{2}}\right)^{-}}\frac{4\lambda\gamma}{\gamma^{2}-\lambda^{2}}=-\infty, λ → ( 2 1 ) − lim γ 2 − λ 2 4 λγ = − ∞ , we can choose suitable λ \lambda λ such that 4 λ γ γ 2 − λ 2 < α 2 − β 2 α β \frac{4\lambda\gamma}{\gamma^{2}-\lambda^{2}}<\frac{\alpha^{2}-\beta^{2}}{\alpha\beta} γ 2 − λ 2 4 λγ < α β α 2 − β 2 . So ∣ ⟨ u , z ⟩ ∣ < ∣ ⟨ v , z ⟩ ∣ |\langle u,z\rangle|<|\langle v,z\rangle| ∣ ⟨ u , z ⟩ ∣ < ∣ ⟨ v , z ⟩ ∣ and ∣ ⟨ u , z ⟩ ∣ < ∣ ⟨ u , w ⟩ ∣ |\langle u,z\rangle|<|\langle u,w\rangle| ∣ ⟨ u , z ⟩ ∣ < ∣ ⟨ u , w ⟩ ∣ , which, together with Proposition 2.4, yields B ⊥ ⊥ w s C B\perp\!\!\perp_{w}^{s}C B ⊥ ⊥ w s C . On the other hand, as ∣ λ ∣ < ∣ γ ∣ |\lambda|<|\gamma| ∣ λ ∣ < ∣ γ ∣ , one gets that ∣ ⟨ z , x ⟩ ∣ < ∣ ⟨ z , y ⟩ ∣ |\langle z,x\rangle|<|\langle z,y\rangle| ∣ ⟨ z , x ⟩ ∣ < ∣ ⟨ z , y ⟩ ∣ and ∣ ⟨ x , z ⟩ ∣ < ∣ ⟨ x , w ⟩ ∣ |\langle x,z\rangle|<|\langle x,w\rangle| ∣ ⟨ x , z ⟩ ∣ < ∣ ⟨ x , w ⟩ ∣ , which entails A ⊥ ⊥ w s C A\perp\!\!\perp_{w}^{s}C A ⊥ ⊥ w s C by Proposition 2.4. Hence A ⊥ ⊥ w s C ⊥ ⊥ w s B A\perp\!\!\perp_{w}^{s}C\perp\!\!\perp_{w}^{s}B A ⊥ ⊥ w s C ⊥ ⊥ w s B .
The proof is finished.
Theorem 4.8. Diam Γ ( N I ( M 2 ( C ) ) , ⊥ ⊥ w s ) = 4 \operatorname{Diam}\Gamma(\mathrm{NI}(M_{2}(\mathbb{C})),\perp\!\!\perp_{w}^{s})=4 Diam Γ ( NI ( M 2 ( C )) , ⊥ ⊥ w s ) = 4 .
Proof. Take A = x ⊗ x A=x\otimes x A = x ⊗ x , B = u ⊗ u ∈ N I ( M 2 ( C ) ) B=u\otimes u\in\mathrm{NI}(M_{2}(\mathbb{C})) B = u ⊗ u ∈ NI ( M 2 ( C )) with x , u ∈ sphere C 2 x,u\in\operatorname{sphere}\mathbb{C}^{2} x , u ∈ sphere C 2 such that u = α x + β y u=\alpha x+\beta y u = α x + β y with y ∈ sphere C 2 y\in\operatorname{sphere}\mathbb{C}^{2} y ∈ sphere C 2 , ⟨ x , y ⟩ = 0 \langle x,y\rangle=0 ⟨ x , y ⟩ = 0 , α , β ∈ C \alpha,\beta\in\mathbb{C} α , β ∈ C and ∣ α ∣ > ∣ β ∣ ≠ 0 |\alpha|>|\beta|\ne0 ∣ α ∣ > ∣ β ∣ = 0 .
It is obvious that A ⊥̸ ⊥ w s B A\not\perp\!\!\perp_{w}^{s}B A ⊥ ⊥ w s B . So the distance between A A A and B B B is greater than 1.
Assume that there exists some C = z ⊗ w ∈ sphere ( N I ( M 2 ( C ) ) ) C=z\otimes w\in\operatorname{sphere}(\mathrm{NI}(M_{2}(\mathbb{C}))) C = z ⊗ w ∈ sphere ( NI ( M 2 ( C ))) with z , w ∈ sphere C 2 z,w\in\operatorname{sphere}\mathbb{C}^{2} z , w ∈ sphere C 2 such that A ⊥ ⊥ w s C ⊥ ⊥ w s B A\perp\!\!\perp_{w}^{s}C\perp\!\!\perp_{w}^{s}B A ⊥ ⊥ w s C ⊥ ⊥ w s B . By A ⊥ ⊥ w s C A\perp\!\!\perp_{w}^{s}C A ⊥ ⊥ w s C and B ⊥ ⊥ w s C B\perp\!\!\perp_{w}^{s}C B ⊥ ⊥ w s C , one obtains C ∗ x = C ∗ u = 0 C^{*}x=C^{*}u=0 C ∗ x = C ∗ u = 0 , which means ⟨ x , z ⟩ = ⟨ u , z ⟩ = 0 \langle x,z\rangle=\langle u,z\rangle=0 ⟨ x , z ⟩ = ⟨ u , z ⟩ = 0 , a contradiction. So the distance between A A A and B B B is greater than 2.
Now, assume that there exist some C 1 , C 2 ∈ sphere ( N I ( M 2 ( C ) ) ) C_{1},C_{2}\in\operatorname{sphere}(\mathrm{NI}(M_{2}(\mathbb{C}))) C 1 , C 2 ∈ sphere ( NI ( M 2 ( C ))) such that A ⊥ ⊥ w s C 1 ⊥ ⊥ w s C 2 ⊥ ⊥ w s B A\perp\!\!\perp_{w}^{s}C_{1}\perp\!\!\perp_{w}^{s}C_{2}\perp\!\!\perp_{w}^{s}B A ⊥ ⊥ w s C 1 ⊥ ⊥ w s C 2 ⊥ ⊥ w s B . If C 1 , C 2 ∈ sphere N I 1 C_{1},C_{2}\in\operatorname{sphere}\mathrm{NI}_{1} C 1 , C 2 ∈ sphere NI 1 , then by Proposition 4.6, we have ∣ tr C 1 ∣ = ∣ tr C 2 ∣ = ∣ tr ( A ) ∣ = 1 |\operatorname{tr}C_{1}|=|\operatorname{tr}C_{2}|=|\operatorname{tr}(A)|=1 ∣ tr C 1 ∣ = ∣ tr C 2 ∣ = ∣ tr ( A ) ∣ = 1 . Thus C 1 = z 1 ⊗ z 1 C_{1}=z_{1}\otimes z_{1} C 1 = z 1 ⊗ z 1 and C 2 = z 2 ⊗ z 2 C_{2}=z_{2}\otimes z_{2} C 2 = z 2 ⊗ z 2 for some z 1 , z 2 ∈ sphere C 2 z_{1},z_{2}\in\operatorname{sphere}\mathbb{C}^{2} z 1 , z 2 ∈ sphere C 2 . It follows that
⟨ x , z 1 ⟩ = ⟨ z 1 , z 2 ⟩ = ⟨ u , z 2 ⟩ = 0 \langle x,z_{1}\rangle=\langle z_{1},z_{2}\rangle=\langle u,z_{2}\rangle=0 ⟨ x , z 1 ⟩ = ⟨ z 1 , z 2 ⟩ = ⟨ u , z 2 ⟩ = 0 , which leads to ⟨ u , x ⟩ = 0 \langle u,x\rangle=0 ⟨ u , x ⟩ = 0 , a contradiction. If C 1 ∈ sphere N I 2 C_{1}\in\operatorname{sphere}\mathrm{NI}_{2} C 1 ∈ sphere NI 2 and C 2 ∈ sphere N I 1 C_{2}\in\operatorname{sphere}\mathrm{NI}_{1} C 2 ∈ sphere NI 1 , we can write C 1 = w ⊗ z C_{1}=w\otimes z C 1 = w ⊗ z with w , z ∈ sphere C 2 w,z\in\operatorname{sphere}\mathbb{C}^{2} w , z ∈ sphere C 2 and ⟨ w , z ⟩ = 0 \langle w,z\rangle=0 ⟨ w , z ⟩ = 0 . On the one hand, since A ⊥ w s C 1 A\perp_{w}^{s}C_{1} A ⊥ w s C 1 , one has ( z ⊗ w ) x = 0 (z\otimes w)x=0 ( z ⊗ w ) x = 0 , and thus ⟨ x , w ⟩ = 0 \langle x,w\rangle=0 ⟨ x , w ⟩ = 0 , which means w ∈ [ y ] w\in[y] w ∈ [ y ] . Combined with C 1 ∈ sphere N I 2 C_{1}\in\operatorname{sphere}\mathrm{NI}_{2} C 1 ∈ sphere NI 2 , we get C 1 = e i θ y ⊗ x C_{1}=e^{i\theta}y\otimes x C 1 = e i θ y ⊗ x for some θ ∈ [ 0 , 2 π ) \theta\in[0,2\pi) θ ∈ [ 0 , 2 π ) . On the other hand, as B ⊥ w s C 2 B\perp_{w}^{s}C_{2} B ⊥ w s C 2 , one has ∣ tr C 2 ∣ = ∣ tr B ∣ = 1 |\operatorname{tr}C_{2}|=|\operatorname{tr}B|=1 ∣ tr C 2 ∣ = ∣ tr B ∣ = 1 by Proposition 4.6 and C 2 ∈ sphere N I 1 C_{2}\in\operatorname{sphere}\mathrm{NI}_{1} C 2 ∈ sphere NI 1 . Thus C 2 = e i δ v ⊗ v C_{2}=e^{i\delta}v\otimes v C 2 = e i δ v ⊗ v with v ∈ sphere [ u ] ⊥ v\in\operatorname{sphere}[u]^{\perp} v ∈ sphere [ u ] ⊥ and δ ∈ [ 0 , 2 π ) \delta\in[0,2\pi) δ ∈ [ 0 , 2 π ) . It follows from C 2 ⊥ w s C 1 C_{2}\perp_{w}^{s}C_{1} C 2 ⊥ w s C 1 that ( x ⊗ y ) v = 0 (x\otimes y)v=0 ( x ⊗ y ) v = 0 , which means ⟨ v , y ⟩ = 0 \langle v,y\rangle=0 ⟨ v , y ⟩ = 0 , a contradiction with α , β ≠ 0 \alpha,\beta\ne0 α , β = 0 . If C 1 , C 2 ∈ sphere N I 2 C_{1},C_{2}\in\operatorname{sphere}\mathrm{NI}_{2} C 1 , C 2 ∈ sphere NI 2 , then we can write C 1 = w 1 ⊗ z 1 C_{1}=w_{1}\otimes z_{1} C 1 = w 1 ⊗ z 1 and C 2 = w 2 ⊗ z 2 C_{2}=w_{2}\otimes z_{2} C 2 = w 2 ⊗ z 2 , where w 1 , w 2 , z 1 , z 2 ∈ sphere C 2 w_{1},w_{2},z_{1},z_{2}\in\operatorname{sphere}\mathbb{C}^{2} w 1 , w 2 , z 1 , z 2 ∈ sphere C 2 and ⟨ w 1 , z 1 ⟩ = ⟨ w 2 , z 2 ⟩ = 0 \langle w_{1},z_{1}\rangle=\langle w_{2},z_{2}\rangle=0 ⟨ w 1 , z 1 ⟩ = ⟨ w 2 , z 2 ⟩ = 0 . As A ⊥ w s C 1 A\perp_{w}^{s}C_{1} A ⊥ w s C 1 and B ⊥ w s C 2 B\perp_{w}^{s}C_{2} B ⊥ w s C 2 , one gets ( z 1 ⊗ w 1 ) x = ( z 2 ⊗ w 2 ) u = 0 (z_{1}\otimes w_{1})x=(z_{2}\otimes w_{2})u=0 ( z 1 ⊗ w 1 ) x = ( z 2 ⊗ w 2 ) u = 0 , which implies ⟨ x , w 1 ⟩ = ⟨ u , w 2 ⟩ = 0 \langle x,w_{1}\rangle=\langle u,w_{2}\rangle=0 ⟨ x , w 1 ⟩ = ⟨ u , w 2 ⟩ = 0 . That is, w 1 ∈ [ y ] w_{1}\in[y] w 1 ∈ [ y ] and w 2 ∈ [ u ] ⊥ w_{2}\in[u]^{\perp} w 2 ∈ [ u ] ⊥ . Hence C 1 = e i θ 1 y ⊗ x C_{1}=e^{i\theta_{1}}y\otimes x C 1 = e i θ 1 y ⊗ x and C 2 = e i θ 2 v ⊗ u C_{2}=e^{i\theta_{2}}v\otimes u C 2 = e i θ 2 v ⊗ u with θ 1 , θ 2 ∈ [ 0 , 2 π ) \theta_{1},\theta_{2}\in[0,2\pi) θ 1 , θ 2 ∈ [ 0 , 2 π ) and v ∈ sphere [ u ] ⊥ v\in\operatorname{sphere}[u]^{\perp} v ∈ sphere [ u ] ⊥ . As C 1 ⊥ w s C 2 C_{1}\perp_{w}^{s}C_{2} C 1 ⊥ w s C 2 , we achieve ∣ ⟨ v , y ⟩ ∣ ≤ ∣ ⟨ v , x ⟩ ∣ |\langle v,y\rangle|\le|\langle v,x\rangle| ∣ ⟨ v , y ⟩ ∣ ≤ ∣ ⟨ v , x ⟩ ∣ by Proposition 2.4, contradicting to ∣ α ∣ > ∣ β ∣ |\alpha|>|\beta| ∣ α ∣ > ∣ β ∣ .
Combining the above discussion, we conclude that the distance between A A A and B B B is greater than 3. So Diam Γ ( N I ( M 2 ( C ) ) , ⊥ w s ) > 3 \operatorname{Diam}\Gamma(\mathrm{NI}(\mathcal{M}_{2}(\mathbb{C})),\perp_{w}^{s})>3 Diam Γ ( NI ( M 2 ( C )) , ⊥ w s ) > 3 .
Take any A , B ∈ N I ( M 2 ( C ) ) A,B\in\mathrm{NI}(\mathcal{M}_{2}(\mathbb{C})) A , B ∈ NI ( M 2 ( C )) .
If A , B ∈ N I 1 A,B\in\mathrm{NI}_{1} A , B ∈ NI 1 , by Proposition 4.4, there exist some C 1 , C 2 ∈ N I 2 C_{1},C_{2}\in\mathrm{NI}_{2} C 1 , C 2 ∈ NI 2 such that A ⊥ w s C 1 A\perp_{w}^{s}C_{1} A ⊥ w s C 1 and B ⊥ w s C 2 B\perp_{w}^{s}C_{2} B ⊥ w s C 2 . By Proposition 4.7, either C 1 ⊥ w s C 2 C_{1}\perp_{w}^{s}C_{2} C 1 ⊥ w s C 2 , or C 1 ⊥ w s C 3 ⊥ w s C 2 C_{1}\perp_{w}^{s}C_{3}\perp_{w}^{s}C_{2} C 1 ⊥ w s C 3 ⊥ w s C 2 for some C 3 ∈ N I 2 C_{3}\in\mathrm{NI}_{2} C 3 ∈ NI 2 . Thus, we get that either A ⊥ w s C 1 ⊥ w s C 2 ⊥ w s B A\perp_{w}^{s}C_{1}\perp_{w}^{s}C_{2}\perp_{w}^{s}B A ⊥ w s C 1 ⊥ w s C 2 ⊥ w s B , or A ⊥ w s C 1 ⊥ w s C 3 ⊥ w s C 2 ⊥ w s B A\perp_{w}^{s}C_{1}\perp_{w}^{s}C_{3}\perp_{w}^{s}C_{2}\perp_{w}^{s}B A ⊥ w s C 1 ⊥ w s C 3 ⊥ w s C 2 ⊥ w s B .
If A ∈ N I 2 A\in\mathrm{NI}_{2} A ∈ NI 2 and B ∈ N I 1 B\in\mathrm{NI}_{1} B ∈ NI 1 , by Proposition 4.4 and Proposition 4.7, one obtains that either A ⊥ w s C 1 ⊥ w s B A\perp_{w}^{s}C_{1}\perp_{w}^{s}B A ⊥ w s C 1 ⊥ w s B for some C 1 ∈ N I 2 C_{1}\in\mathrm{NI}_{2} C 1 ∈ NI 2 , or A ⊥ w s C 2 ⊥ w s C 1 ⊥ w s B A\perp_{w}^{s}C_{2}\perp_{w}^{s}C_{1}\perp_{w}^{s}B A ⊥ w s C 2 ⊥ w s C 1 ⊥ w s B for some C 1 , C 2 ∈ N I 2 C_{1},C_{2}\in\mathrm{NI}_{2} C 1 , C 2 ∈ NI 2 .
If A , B ∈ N I 2 A,B\in\mathrm{NI}_{2} A , B ∈ NI 2 , by Proposition 4.7, we still obtain that either A ⊥ w s B A\perp_{w}^{s}B A ⊥ w s B , or A ⊥ w s C ⊥ w s B A\perp_{w}^{s}C\perp_{w}^{s}B A ⊥ w s C ⊥ w s B for some C ∈ N I 2 C\in\mathrm{NI}_{2} C ∈ NI 2 .
The above arguments show that 3 < Diam Γ ( N I ( M 2 ( C ) ) , ⊥ w s ) ≤ 4 3<\operatorname{Diam}\Gamma(\mathrm{NI}(\mathcal{M}_{2}(\mathbb{C})),\perp_{w}^{s})\le4 3 < Diam Γ ( NI ( M 2 ( C )) , ⊥ w s ) ≤ 4 . Therefore, we must have Diam Γ ( N I ( M 2 ( C ) ) , ⊥ w s ) = 4 \operatorname{Diam}\Gamma(\mathrm{NI}(\mathcal{M}_{2}(\mathbb{C})),\perp_{w}^{s})=4 Diam Γ ( NI ( M 2 ( C )) , ⊥ w s ) = 4 , as desired. □
The strong numerical radius orthograph Γ ( B s ( H ) , ⊥ w s ) \Gamma(\mathcal{B}_{s}(H),\perp_{w}^{s}) Γ ( B s ( H ) , ⊥ w s )
In this section, we study the strong numerical radius orthograph Γ ( B s ( H ) , ⊥ w s ) \Gamma(\mathcal{B}_{s}(H),\perp_{w}^{s}) Γ ( B s ( H ) , ⊥ w s ) with 2 ≤ dim H ≤ ∞ 2\le\dim H\le\infty 2 ≤ dim H ≤ ∞ .
Our first result is a useful proposition.
Proposition 5.1. If dim H ≥ 3 \dim H\ge3 dim H ≥ 3 , then there exist two vertices A , B ∈ Γ ( N I ( B s ( H ) ) , ⊥ w s ) A,B\in\Gamma(\mathrm{NI}(\mathcal{B}_{s}(H)),\perp_{w}^{s}) A , B ∈ Γ ( NI ( B s ( H )) , ⊥ w s ) such that the distance between A A A and B B B is greater than 2. Particularly, if dim H = 3 \dim H=3 dim H = 3 , then there exist two vertices A , B ∈ Γ ( B s ( H ) , ⊥ w s ) A,B\in\Gamma(\mathcal{B}_{s}(H),\perp_{w}^{s}) A , B ∈ Γ ( B s ( H ) , ⊥ w s ) such that the distance between A A A and B B B is greater than 3.
Proof. Assume that dim H ≥ 3 \dim H \ge3 dim H ≥ 3 . Take e 1 , e 2 , x 1 , x 2 ∈ sphere H e_1,e_2,x_1,x_2 \in\operatorname{sphere} H e 1 , e 2 , x 1 , x 2 ∈ sphere H with ⟨ e 1 , e 2 ⟩ = 0 \langle e_1,e_2\rangle= 0 ⟨ e 1 , e 2 ⟩ = 0 , x 1 ∈ [ e 1 , e 2 ] ⊥ x_1 \in[e_1,e_2]^\perp x 1 ∈ [ e 1 , e 2 ] ⊥ and x 2 = λ e 1 + μ e 2 x_2 = \lambda e_1 + \mu e_2 x 2 = λ e 1 + μ e 2 , λ , μ ∈ C ∖ { 0 } \lambda,\mu\in\mathbb{C}\setminus\{0\} λ , μ ∈ C ∖ { 0 } . Let
A = e 1 ⊗ e 1 + 1 2 P [ e 1 , e 2 ] ⊥ and B = x 1 ⊗ x 1 + 1 2 P [ x 1 , x 2 ] ⊥ . A = e_1 \otimes e_1 + \frac{1}{2}P_{[e_1,e_2]^\perp}
\quad\text{and}\quad
B = x_1 \otimes x_1 + \frac{1}{2}P_{[x_1,x_2]^\perp}. A = e 1 ⊗ e 1 + 2 1 P [ e 1 , e 2 ] ⊥ and B = x 1 ⊗ x 1 + 2 1 P [ x 1 , x 2 ] ⊥ .
From Proposition 2.3 , it is easily seen that B ̸ ⊥ w s A B \not\perp_w^s A B ⊥ w s A , which means A ̸ ⊥ w s B A \not\perp_w^s B A ⊥ w s B .
Suppose that A ⊥ w s C ⊥ w s B A \perp_w^s C \perp_w^s B A ⊥ w s C ⊥ w s B for some C ∈ sphere ( B s ( H ) ) C \in\operatorname{sphere}(\mathcal{B}_s(H)) C ∈ sphere ( B s ( H )) . We claim C e 1 = 0 Ce_1 = 0 C e 1 = 0 . If, on the contrary, C e 1 ≠ 0 Ce_1 \ne0 C e 1 = 0 , take { u n } ⊆ sphere H \{u_n\} \subseteq\operatorname{sphere} H { u n } ⊆ sphere H such that w ( A ) = lim n → ∞ ∣ ⟨ A u n , u n ⟩ ∣ w(A) = \lim_{n\to\infty}|\langle Au_n,u_n\rangle| w ( A ) = lim n → ∞ ∣ ⟨ A u n , u n ⟩ ∣ . Write u n = α n e 1 + β n z n u_n = \alpha_n e_1 + \beta_n z_n u n = α n e 1 + β n z n with α n , β n ∈ C \alpha_n,\beta_n \in\mathbb{C} α n , β n ∈ C and z n ∈ sphere [ e 1 ] ⊥ z_n \in\operatorname{sphere}[e_1]^\perp z n ∈ sphere [ e 1 ] ⊥ . Without loss of generality, we may assume α n , β n ≥ 0 \alpha_n,\beta_n \ge0 α n , β n ≥ 0 ; otherwise, consider e − i ξ n u n e^{-i\xi_n}u_n e − i ξ n u n and e i ( ζ n − ξ n ) z n e^{i(\zeta_n-\xi_n)}z_n e i ( ζ n − ξ n ) z n instead, where ξ n = arg α n \xi_n = \arg\alpha_n ξ n = arg α n and ζ n = arg β n \zeta_n = \arg\beta_n ζ n = arg β n . So
1 = w ( A ) = lim n → ∞ ∣ α n 2 + β n 2 ⟨ A z n , z n ⟩ ∣ ≤ lim n → ∞ ( α n 2 + 1 2 β n 2 ) ≤ lim n → ∞ ( α n 2 + β n 2 ) = 1 , 1 = w(A) = \lim_{n\to\infty} \left|\alpha_n^2 + \beta_n^2\langle Az_n,z_n\rangle\right|
\le\lim_{n\to\infty}\left(\alpha_n^2 + \frac{1}{2}\beta_n^2\right)
\le\lim_{n\to\infty}\left(\alpha_n^2 + \beta_n^2\right) = 1, 1 = w ( A ) = n → ∞ lim α n 2 + β n 2 ⟨ A z n , z n ⟩ ≤ n → ∞ lim ( α n 2 + 2 1 β n 2 ) ≤ n → ∞ lim ( α n 2 + β n 2 ) = 1 ,
which implies lim n → ∞ ( α n 2 + 1 2 β n 2 ) = lim n → ∞ ( α n 2 + β n 2 ) \lim_{n\to\infty}(\alpha_n^2 + \frac{1}{2}\beta_n^2) = \lim_{n\to\infty}(\alpha_n^2 + \beta_n^2) lim n → ∞ ( α n 2 + 2 1 β n 2 ) = lim n → ∞ ( α n 2 + β n 2 ) . Hence lim n → ∞ β n = 0 \lim_{n\to\infty}\beta_n = 0 lim n → ∞ β n = 0 , and thus lim n → ∞ α n = 1 \lim_{n\to\infty}\alpha_n = 1 lim n → ∞ α n = 1 . A direct calculation gives lim n → ∞ ∥ e 1 − u n ∥ = 0 \lim_{n\to\infty}\|e_1-u_n\| = 0 lim n → ∞ ∥ e 1 − u n ∥ = 0 , and so lim n → ∞ ∥ C u n ∥ = ∥ C e 1 ∥ ≠ 0 \lim_{n\to\infty}\|Cu_n\| = \|Ce_1\| \ne0 lim n → ∞ ∥ C u n ∥ = ∥ C e 1 ∥ = 0 , a contradiction to A ⊥ w s C A \perp_w^s C A ⊥ w s C .
Similarly, by the relations C ⊥ w s A C \perp_w^s A C ⊥ w s A and C ⊥ w s B C \perp_w^s B C ⊥ w s B , one can show that C e 2 = η 1 e 2 Ce_2 = \eta_1e_2 C e 2 = η 1 e 2 , C x 1 = 0 Cx_1 = 0 C x 1 = 0 and C x 2 = η 2 x 2 Cx_2 = \eta_2x_2 C x 2 = η 2 x 2 with η 1 , η 2 ∈ { − 1 , 1 } \eta_1,\eta_2 \in\{-1,1\} η 1 , η 2 ∈ { − 1 , 1 } . It follows that
λ = ⟨ x 2 , e 1 ⟩ = ⟨ η 2 C x 2 , e 1 ⟩ = ⟨ η 2 x 2 , C e 1 ⟩ = 0 , \lambda= \langle x_2,e_1\rangle= \langle\eta_2 Cx_2,e_1\rangle= \langle\eta_2 x_2,Ce_1\rangle= 0, λ = ⟨ x 2 , e 1 ⟩ = ⟨ η 2 C x 2 , e 1 ⟩ = ⟨ η 2 x 2 , C e 1 ⟩ = 0 ,
a contradiction. Therefore, the above C C C does not exist and the distance between A A A and B B B is greater than 2 2 2 .
Now, assume that dim H = 3 \dim H = 3 dim H = 3 . Let [ e 1 , e 2 , e 3 ] [e_1,e_2,e_3] [ e 1 , e 2 , e 3 ] and [ x 1 , x 2 , x 3 ] [x_1,x_2,x_3] [ x 1 , x 2 , x 3 ] be any two orthonormal bases of H H H . Write x 1 = λ e 1 + μ e 2 + γ e 3 x_1 = \lambda e_1 + \mu e_2 + \gamma e_3 x 1 = λ e 1 + μ e 2 + γ e 3 and x 2 = λ ′ e 1 + μ ′ e 2 + γ ′ e 3 x_2 = \lambda' e_1 + \mu' e_2 + \gamma' e_3 x 2 = λ ′ e 1 + μ ′ e 2 + γ ′ e 3 with λ , λ ′ , μ , μ ′ , γ , γ ′ ∈ C ∖ { 0 } \lambda,\lambda',\mu,\mu',\gamma,\gamma' \in\mathbb{C}\setminus\{0\} λ , λ ′ , μ , μ ′ , γ , γ ′ ∈ C ∖ { 0 } . Take
A = e 1 ⊗ e 1 + 1 2 e 3 ⊗ e 3 and B = x 1 ⊗ x 1 + 1 2 x 3 ⊗ x 3 . A = e_1 \otimes e_1 + \frac{1}{2}e_3 \otimes e_3
\quad\text{and}\quad
B = x_1 \otimes x_1 + \frac{1}{2}x_3 \otimes x_3. A = e 1 ⊗ e 1 + 2 1 e 3 ⊗ e 3 and B = x 1 ⊗ x 1 + 2 1 x 3 ⊗ x 3 .
Claim 1. A ̸ ⊥ w s B A \not\perp_w^s B A ⊥ w s B .
If A ⊥ w s B A \perp_w^s B A ⊥ w s B , then B e 1 = 0 Be_1 = 0 B e 1 = 0 by Proposition 2.3 , contradicting to μ ′ ≠ 0 \mu' \ne0 μ ′ = 0 .
Claim 2. There does not exist C ∈ NI ( B s ( H ) ) ∖ { 0 } C \in\operatorname{NI}(\mathcal{B}_s(H))\setminus\{0\} C ∈ NI ( B s ( H )) ∖ { 0 } such that A ⊥ w s C ⊥ w s B A \perp_w^s C \perp_w^s B A ⊥ w s C ⊥ w s B .
If there exists such C C C , then by A ⊥ w s C A \perp_w^s C A ⊥ w s C , one gets C e 1 = 0 Ce_1 = 0 C e 1 = 0 and C e 2 = η e 2 Ce_2 = \eta e_2 C e 2 = η e 2 with η ∈ { − 1 , 1 } \eta\in\{-1,1\} η ∈ { − 1 , 1 } ; by B ⊥ w s C B \perp_w^s C B ⊥ w s C , one has C x 1 = 0 Cx_1 = 0 C x 1 = 0 and C x 2 = η ′ x 2 Cx_2 = \eta' x_2 C x 2 = η ′ x 2 with η ′ ∈ { − 1 , 1 } \eta' \in\{-1,1\} η ′ ∈ { − 1 , 1 } . It follows that
λ ′ = ⟨ x 2 , e 1 ⟩ = ⟨ η ′ C x 2 , e 1 ⟩ = η ′ ⟨ x 2 , C e 1 ⟩ = 0 , \lambda' = \langle x_2,e_1\rangle= \langle\eta' Cx_2,e_1\rangle= \eta'\langle x_2,Ce_1\rangle= 0, λ ′ = ⟨ x 2 , e 1 ⟩ = ⟨ η ′ C x 2 , e 1 ⟩ = η ′ ⟨ x 2 , C e 1 ⟩ = 0 ,
a contradiction.
Claim 3. There does not exist C 1 , C 2 ∈ NI ( B s ( H ) ) ∖ { 0 } C_1,C_2 \in\operatorname{NI}(\mathcal{B}_s(H))\setminus\{0\} C 1 , C 2 ∈ NI ( B s ( H )) ∖ { 0 } such that $A \perp_w^s C_1 \perp_w^s C_2 \perp_w^s B. Otherwise, an argument similar to that in Claim 2 gives
C 1 e 1 = C 2 x 1 = 0 , C 1 e 2 = η e 2 and C 2 x 2 = η ′ x 2 with η , η ′ ∈ { − 1 , 1 } . C_1e_1=C_2x_1=0,\quad C_1e_2=\eta e_2\text{ and }C_2x_2=\eta'x_2\text{ with }\eta,\eta'\in\{-1,1\}. C 1 e 1 = C 2 x 1 = 0 , C 1 e 2 = η e 2 and C 2 x 2 = η ′ x 2 with η , η ′ ∈ { − 1 , 1 } .
So we can set C 1 = η e 2 ⊗ e 2 + ξ e 3 ⊗ e 3 C_1=\eta e_2\otimes e_2+\xi e_3\otimes e_3 C 1 = η e 2 ⊗ e 2 + ξ e 3 ⊗ e 3 and C 2 = η ′ x 2 ⊗ x 2 + ξ ′ x 3 ⊗ x 3 C_2=\eta'x_2\otimes x_2+\xi'x_3\otimes x_3 C 2 = η ′ x 2 ⊗ x 2 + ξ ′ x 3 ⊗ x 3 with ξ , ξ ′ ∈ C \xi,\xi'\in\mathbb{C} ξ , ξ ′ ∈ C and ∣ ξ ∣ , ∣ ξ ′ ∣ ≤ 1 |\xi|,|\xi'|\le1 ∣ ξ ∣ , ∣ ξ ′ ∣ ≤ 1 . If ξ ′ ≠ 0 \xi'\ne0 ξ ′ = 0 , we have x 1 ∈ [ e 2 , e 3 ] x_1\in[e_2,e_3] x 1 ∈ [ e 2 , e 3 ] by C 1 ⊥ w s C 2 C_1\perp_w^s C_2 C 1 ⊥ w s C 2 , a contradiction to λ ≠ 0 \lambda\ne0 λ = 0 ; if ξ ≠ 0 \xi\ne0 ξ = 0 , we get e 1 ∈ [ x 2 , x 3 ] e_1\in[x_2,x_3] e 1 ∈ [ x 2 , x 3 ] by C 2 ⊥ w s C 1 C_2\perp_w^s C_1 C 2 ⊥ w s C 1 , contradicting to ⟨ x 1 , e 1 ⟩ = λ ≠ 0 \langle x_1,e_1\rangle=\lambda\ne0 ⟨ x 1 , e 1 ⟩ = λ = 0 ; if ξ = ξ ′ = 0 \xi=\xi'=0 ξ = ξ ′ = 0 , one has ⟨ e 2 , x 2 ⟩ = 0 \langle e_2,x_2\rangle=0 ⟨ e 2 , x 2 ⟩ = 0 by C 1 ⊥ w s C 2 C_1\perp_w^s C_2 C 1 ⊥ w s C 2 , a contradiction to μ ′ ≠ 0 \mu'\ne0 μ ′ = 0 .
Now, by Claims 1-3, we see that the distance between A A A and B B B must be greater than 3. □
Applying Proposition 5.1 , we can draw out the diameter of Γ ( NI ( B s ( H ) ) , ⊥ w s ) \Gamma(\operatorname{NI}(\mathcal{B}_s(H)),\perp_w^s) Γ ( NI ( B s ( H )) , ⊥ w s ) .
Theorem 5.2. If dim H ≥ 4 \dim H\ge4 dim H ≥ 4 , then Diam Γ ( NI ( B s ( H ) ) , ⊥ w s ) = 3 \operatorname{Diam}\Gamma(\operatorname{NI}(\mathcal{B}_s(H)),\perp_w^s)=3 Diam Γ ( NI ( B s ( H )) , ⊥ w s ) = 3 ; if dim H = 3 \dim H=3 dim H = 3 , then Diam Γ ( NI ( B s ( H ) ) , ⊥ w s ) = 4 \operatorname{Diam}\Gamma(\operatorname{NI}(\mathcal{B}_s(H)),\perp_w^s)=4 Diam Γ ( NI ( B s ( H )) , ⊥ w s ) = 4 .
Proof. Take arbitrary A , B ∈ NI ( B s ( H ) ) A,B\in\operatorname{NI}(\mathcal{B}_s(H)) A , B ∈ NI ( B s ( H )) . Without loss of generality, assume that w ( A ) = w ( B ) = 1 w(A)=w(B)=1 w ( A ) = w ( B ) = 1 , 1 ∈ σ ( A ) 1\in\sigma(A) 1 ∈ σ ( A ) and 1 ∈ σ ( B ) 1\in\sigma(B) 1 ∈ σ ( B ) . We will prove the theorem by considering three cases.
Case 1. dim H = ∞ \dim H=\infty dim H = ∞ .
Let A = ∫ z d E ( z ) A=\int z\mathrm{d}E(z) A = ∫ z d E ( z ) and B = ∫ z d F ( z ) B=\int z\mathrm{d}F(z) B = ∫ z d F ( z ) be the spectral decomposition of A A A and B B B , respectively.
For any ε i > 0 \varepsilon_i>0 ε i > 0 , i ∈ { 1 , 2 , 3 , 4 } i\in\{1,2,3,4\} i ∈ { 1 , 2 , 3 , 4 } , let
H 1 = { E ( [ 1 − ε 1 , 1 ] ) if 1 ∉ σ p ( A ) , [ e 1 ] if 1 ∈ σ p ( A ) , H 2 = { E ( [ − ε 2 , ε 2 ] ) if 0 ∉ σ p ( A ) , [ e 2 ] if 0 ∈ σ p ( A ) , H_1=
\begin{cases}
E([1-\varepsilon_1,1]) & \text{if }1\notin\sigma_p(A),\\
[e_1] & \text{if }1\in\sigma_p(A),
\end{cases}
\qquad
H_2=
\begin{cases}
E([-\varepsilon_2,\varepsilon_2]) & \text{if }0\notin\sigma_p(A),\\
[e_2] & \text{if }0\in\sigma_p(A),
\end{cases} H 1 = { E ([ 1 − ε 1 , 1 ]) [ e 1 ] if 1 ∈ / σ p ( A ) , if 1 ∈ σ p ( A ) , H 2 = { E ([ − ε 2 , ε 2 ]) [ e 2 ] if 0 ∈ / σ p ( A ) , if 0 ∈ σ p ( A ) ,
H 3 = { F ( [ 1 − ε 3 , 1 ] ) if 1 ∉ σ p ( B ) , [ e 3 ] if 1 ∈ σ p ( B ) , H 4 = { F ( [ − ε 4 , ε 4 ] ) if 0 ∉ σ p ( B ) , [ e 4 ] if 0 ∈ σ p ( B ) . H_3=
\begin{cases}
F([1-\varepsilon_3,1]) & \text{if }1\notin\sigma_p(B),\\
[e_3] & \text{if }1\in\sigma_p(B),
\end{cases}
\qquad
H_4=
\begin{cases}
F([-\varepsilon_4,\varepsilon_4]) & \text{if }0\notin\sigma_p(B),\\
[e_4] & \text{if }0\in\sigma_p(B).
\end{cases} H 3 = { F ([ 1 − ε 3 , 1 ]) [ e 3 ] if 1 ∈ / σ p ( B ) , if 1 ∈ σ p ( B ) , H 4 = { F ([ − ε 4 , ε 4 ]) [ e 4 ] if 0 ∈ / σ p ( B ) , if 0 ∈ σ p ( B ) .
where e 1 ∈ sphere E ( 1 ) e_1\in\operatorname{sphere}E(1) e 1 ∈ sphere E ( 1 ) , e 2 ∈ sphere E ( 0 ) e_2\in\operatorname{sphere}E(0) e 2 ∈ sphere E ( 0 ) , e 3 ∈ sphere F ( 1 ) e_3\in\operatorname{sphere}F(1) e 3 ∈ sphere F ( 1 ) , e 4 ∈ sphere F ( 0 ) e_4\in\operatorname{sphere}F(0) e 4 ∈ sphere F ( 0 ) , σ p ( X ) \sigma_p(X) σ p ( X ) denotes the point spectrum of X X X . Let ε i \varepsilon_i ε i be sufficiently small so that there exist unit vectors y ∈ [ H 1 , H 4 ] ⊥ y\in[H_1,H_4]^\perp y ∈ [ H 1 , H 4 ] ⊥ and z ∈ [ H 3 , H 2 , y ] ⊥ z\in[H_3,H_2,y]^\perp z ∈ [ H 3 , H 2 , y ] ⊥ . Let C 1 = P [ H 2 , y ] C_1=P_{[H_2,y]} C 1 = P [ H 2 , y ] and C 2 = P [ H 4 , z ] C_2=P_{[H_4,z]} C 2 = P [ H 4 , z ] . Then we have
A ⊥ w s C 1 ⊥ w s C 2 ⊥ w s B , A\perp_w^s C_1\perp_w^s C_2\perp_w^s B, A ⊥ w s C 1 ⊥ w s C 2 ⊥ w s B ,
which, together with Proposition 5.1 , implies Diam Γ ( NI ( B s ( H ) ) , ⊥ w s ) = 3 \operatorname{Diam}\Gamma(\operatorname{NI}(\mathcal{B}_s(H)),\perp_w^s)=3 Diam Γ ( NI ( B s ( H )) , ⊥ w s ) = 3 .
Case 2. 4 ≤ dim H < ∞ 4\le\dim H<\infty 4 ≤ dim H < ∞ .
In this case, the numerical range W ( X ) W(X) W ( X ) is a closed convex set for any X ∈ B ( H ) X\in\mathcal{B}(H) X ∈ B ( H ) . Thus, there exist e 1 , e 2 , x 1 , x 2 ∈ sphere H e_1,e_2,x_1,x_2\in\operatorname{sphere}H e 1 , e 2 , x 1 , x 2 ∈ sphere H such that
∣ ⟨ A e 1 , e 1 ⟩ ∣ = w ( A ) , A e 2 = 0 , ∣ ⟨ B x 1 , x 1 ⟩ ∣ = w ( B ) and B x 2 = 0. |\langle Ae_1,e_1\rangle|=w(A),\quad Ae_2=0,\quad|\langle Bx_1,x_1\rangle|=w(B)\text{ and }Bx_2=0. ∣ ⟨ A e 1 , e 1 ⟩ ∣ = w ( A ) , A e 2 = 0 , ∣ ⟨ B x 1 , x 1 ⟩ ∣ = w ( B ) and B x 2 = 0.
Take y ∈ [ e 1 , x 2 ] ⊥ y \in[e_1,x_2]^\perp y ∈ [ e 1 , x 2 ] ⊥ , z ∈ [ x 1 , e 2 , y ] ⊥ z \in[x_1,e_2,y]^\perp z ∈ [ x 1 , e 2 , y ] ⊥ , and let C 1 = P [ e 2 , y ] C_1=P_{[e_2,y]} C 1 = P [ e 2 , y ] and C 2 = P [ x 2 , z ] C_2=P_{[x_2,z]} C 2 = P [ x 2 , z ] . It follows that A ⊥ w s C 1 ⊥ w s C 2 ⊥ w s B A \mathrel{\perp_w^s} C_1 \mathrel{\perp_w^s} C_2 \mathrel{\perp_w^s} B A ⊥ w s C 1 ⊥ w s C 2 ⊥ w s B , which again implies Diam Γ ( NI ( B s ( H ) ) , ⊥ w s ) = 3 \operatorname{Diam}\Gamma(\operatorname{NI}(\mathcal{B}_s(H)),\mathrel{\perp_w^s})=3 Diam Γ ( NI ( B s ( H )) , ⊥ w s ) = 3 by Proposition 5.1 .
Case 3. dim H = 3 \dim H=3 dim H = 3 .
Take x , y ∈ sphere H x,y \in\operatorname{sphere} H x , y ∈ sphere H such that A x = B y = 0 Ax=By=0 A x = B y = 0 . Let C 1 = x ⊗ x C_1=x\otimes x C 1 = x ⊗ x , C 2 = y ⊗ y C_2=y\otimes y C 2 = y ⊗ y and C 3 = z ⊗ z C_3=z\otimes z C 3 = z ⊗ z with z ∈ sphere [ x , y ] ⊥ z \in\operatorname{sphere}[x,y]^\perp z ∈ sphere [ x , y ] ⊥ . It is easily checked that A ⊥ w s C 1 ⊥ w s C 3 ⊥ w s C 2 ⊥ w s B A \mathrel{\perp_w^s} C_1 \mathrel{\perp_w^s} C_3 \mathrel{\perp_w^s} C_2 \mathrel{\perp_w^s} B A ⊥ w s C 1 ⊥ w s C 3 ⊥ w s C 2 ⊥ w s B , which entails that Diam Γ ( NI ( B s ( H ) ) , ⊥ w s ) = 4 \operatorname{Diam}\Gamma(\operatorname{NI}(\mathcal{B}_s(H)),\mathrel{\perp_w^s})=4 Diam Γ ( NI ( B s ( H )) , ⊥ w s ) = 4 by Proposition 5.1 .
The proof is complete.
For the connected components of Γ ( B s ( H ) , ⊥ w s ) \Gamma(\mathcal{B}_s(H),\mathrel{\perp_w^s}) Γ ( B s ( H ) , ⊥ w s ) , we have the following result.
Theorem 5.3. For dim H ≥ 3 \dim H \ge3 dim H ≥ 3 , the connected components of Γ ( B s ( H ) , ⊥ w s ) \Gamma(\mathcal{B}_s(H),\mathrel{\perp_w^s}) Γ ( B s ( H ) , ⊥ w s ) are either isolated vertices or Γ ( NI ( B s ( H ) ) , ⊥ w s ) \Gamma(\operatorname{NI}(\mathcal{B}_s(H)),\mathrel{\perp_w^s}) Γ ( NI ( B s ( H )) , ⊥ w s ) . Moreover,
{ A ∈ B s ( H ) : A is an isolated vertex } = B s ( H ) ∖ NI ( B s ( H ) ) . \{A \in\mathcal{B}_s(H): A\text{ is an isolated vertex}\}=\mathcal{B}_s(H)\setminus\operatorname{NI}(\mathcal{B}_s(H)). { A ∈ B s ( H ) : A is an isolated vertex } = B s ( H ) ∖ NI ( B s ( H )) .
Proof. Take any A ∈ B s ( H ) A \in\mathcal{B}_s(H) A ∈ B s ( H ) .
If A A A is invertible and A ⊥ w s B A \mathrel{\perp_w^s} B A ⊥ w s B for some B ∈ B s ( H ) B \in\mathcal{B}_s(H) B ∈ B s ( H ) , we have B ⊥ w s A B \mathrel{\perp_w^s} A B ⊥ w s A , which and Proposition 2.8 imply B = 0 B=0 B = 0 . So A A A is an isolated vertex.
If A A A is not invertible, then 0 ∈ σ ( A ) 0 \in\sigma(A) 0 ∈ σ ( A ) and 0 ∈ W ( A ) = [ λ 1 , λ 2 ] 0 \in W(A)=[\lambda_1,\lambda_2] 0 ∈ W ( A ) = [ λ 1 , λ 2 ] , where λ 1 = min { λ : λ ∈ σ ( A ) } \lambda_1=\min\{\lambda:\lambda\in\sigma(A)\} λ 1 = min { λ : λ ∈ σ ( A )} and λ 2 = max { λ : λ ∈ σ ( A ) } \lambda_2=\max\{\lambda:\lambda\in\sigma(A)\} λ 2 = max { λ : λ ∈ σ ( A )} . Let A = ∫ z d E ( z ) A=\int z\,dE(z) A = ∫ z d E ( z ) be the spectral decomposition of A A A . We may assume ∣ λ 2 ∣ ≥ ∣ λ 1 ∣ |\lambda_2| \ge|\lambda_1| ∣ λ 2 ∣ ≥ ∣ λ 1 ∣ ; otherwise, consider − A -A − A instead. Then there exist ε , ε ′ ≥ 0 \varepsilon,\varepsilon' \ge0 ε , ε ′ ≥ 0 such that the nontrivial subspaces H 1 = E ( [ − ε , ε ] ) H_1=E([-\varepsilon,\varepsilon]) H 1 = E ([ − ε , ε ]) and H 2 = E ( [ λ 2 − ε ′ , λ 2 ] ) H_2=E([\lambda_2-\varepsilon',\lambda_2]) H 2 = E ([ λ 2 − ε ′ , λ 2 ]) are orthogonal to each other. Take B = P [ H 1 ] B=P_{[H_1]} B = P [ H 1 ] . Then A ⊥ w s B A \mathrel{\perp_w^s} B A ⊥ w s B and B ≠ 0 B\ne0 B = 0 . This means that A A A is not an isolated vertex.
Now, by the proof of Theorem 5.2 , Γ ( NI ( B s ( H ) ) , ⊥ w s ) \Gamma(\operatorname{NI}(\mathcal{B}_s(H)),\mathrel{\perp_w^s}) Γ ( NI ( B s ( H )) , ⊥ w s ) is connected.
The case where the dimension of H H H is 2 2 2 is different and needs to be discussed separately. In this case, B s ( H ) = H 2 \mathcal{B}_s(H)=H_2 B s ( H ) = H 2 .
Theorem 5.4. Every connected component of Γ ( H 2 , ⊥ w s ) \Gamma(H_2,\mathrel{\perp_w^s}) Γ ( H 2 , ⊥ w s ) is either an isolated vertice determined by an invertible matrix in H 2 H_2 H 2 , or a set of the form
S x = { u ⊗ u : u ∈ [ x ] ∪ [ x ] ⊥ } S_x=\{u\otimes u:u \in[x]\cup[x]^\perp\} S x = { u ⊗ u : u ∈ [ x ] ∪ [ x ] ⊥ } for some x ∈ sphere C 2 x \in\operatorname{sphere}\mathbb{C}^2 x ∈ sphere C 2 . Moreover, for each x ∈ sphere C 2 x \in\operatorname{sphere}\mathbb{C}^2 x ∈ sphere C 2 , we have Diam Γ ( S x , ⊥ w s ) = 1 \operatorname{Diam}\Gamma(S_x,\mathrel{\perp_w^s})=1 Diam Γ ( S x , ⊥ w s ) = 1 .
Proof. Obviously, every invertible matrix in H 2 H_2 H 2 is an isolated vertice of the orthgraph Γ ( H 2 , ⊥ w s ) \Gamma(H_2,\mathrel{\perp_w^s}) Γ ( H 2 , ⊥ w s ) . Assume A , B ∈ H 2 A,B \in H_2 A , B ∈ H 2 are not invertible. Without loss of generality, we may assume that both A A A and B B B are positive. Then A = x ⊗ x A=x\otimes x A = x ⊗ x and B = y ⊗ y B=y\otimes y B = y ⊗ y for some x , y ∈ C 2 x,y \in\mathbb{C}^2 x , y ∈ C 2 . It is easy to verify that A ⊥ w s B ⇔ ⟨ x , y ⟩ = 0 A \perp_{w}^{s} B \Leftrightarrow\langle x,y\rangle=0 A ⊥ w s B ⇔ ⟨ x , y ⟩ = 0 . So A A A is not an isolated vertice of Γ ( H 2 , ⊥ w s ) \Gamma(H_{2},\perp_{w}^{s}) Γ ( H 2 , ⊥ w s ) , and the connected components containing A = x ⊗ x A=x\otimes x A = x ⊗ x has the form S x = { u ⊗ u : u ∈ [ x ] ∪ [ x ] ⊥ } S_{x}=\{u\otimes u:u\in[x]\cup[x]^{\perp}\} S x = { u ⊗ u : u ∈ [ x ] ∪ [ x ] ⊥ } with x ∈ H x\in H x ∈ H .
Moreover, for any A , B ∈ S x A,B\in S_{x} A , B ∈ S x , we have that either A ⊥ w s B A\perp_{w}^{s}B A ⊥ w s B or A = λ B A=\lambda B A = λ B for some λ ∈ R \lambda\in\mathbb{R} λ ∈ R . Hence Diam Γ ( S x , ⊥ w s ) = 1 \operatorname{Diam}\Gamma(S_{x},\perp_{w}^{s})=1 Diam Γ ( S x , ⊥ w s ) = 1 .
The proof is completed.
(Sun) Fundamentals Department, Air Force Engineering University, Xi’an, 710051, P. R. China
(Hou) College of Mathematics, Taiyuan University of Technology, Taiyuan 030024, P.R. China
(Qi) School of Mathematics and Statistics, Shanxi University, Taiyuan 030006, P. R. China; Key Laboratory of Complex Systems and Data Science of Ministry of Education, Shanxi University, Taiyuan 030006, Shanxi, China
Email address: ssx15603406512@163.com; xiaofeiqisxu@aliyun.com; jinchuanhou@aliyun.com