2020 AMS Subject Classification: 47C15; 46E22.
Introduction
Let D \mathbb{D} D denote the unit disk, and let H ∞ ( D ) H^\infty(\mathbb{D}) H ∞ ( D ) be the algebra of bounded analytic functions on D \mathbb{D} D . Let H \mathcal{H} H be the Hardy space H 2 ( D ) H^2(\mathbb{D}) H 2 ( D ) or the Bergman space L a 2 ( D ) L_a^2(\mathbb{D}) L a 2 ( D ) , the latter consisting of all holomorphic functions f ( z ) = ∑ n = 0 ∞ a n z n f(z) = \sum_{n=0}^{\infty} a_n z^n f ( z ) = ∑ n = 0 ∞ a n z n on D \mathbb{D} D satisfying
∥ f ∥ 2 = ∫ D ∣ f ( z ) ∣ 2 d A ( z ) < ∞ , \|f\|^2 = \int_{\mathbb{D}} |f(z)|^2 \,dA(z) < \infty, ∥ f ∥ 2 = ∫ D ∣ f ( z ) ∣ 2 d A ( z ) < ∞ ,
where d A dA d A denotes the area measure on D \mathbb{D} D . Each function φ \varphi φ in H ∞ ( D ) H^\infty(\mathbb{D}) H ∞ ( D ) induces a bounded multiplication operator
M φ : H → H , f ↦ φ f . M_\varphi:\mathcal{H}\to\mathcal{H},\qquad f\mapsto\varphi f. M φ : H → H , f ↦ φ f .
A closed subspace M ⊆ H \mathcal{M} \subseteq\mathcal{H} M ⊆ H is reducing for M φ M_{\varphi} M φ if it is invariant under both M φ M_{\varphi} M φ and its adjoint M φ ∗ M_{\varphi}^{*} M φ ∗ . The operator M φ M_{\varphi} M φ is called reducible if it admits a nontrivial reducing subspace (i.e., one other than { 0 } \{0\} { 0 } and H \mathcal{H} H ); otherwise it is irreducible . Equivalently, M φ M_{\varphi} M φ is irreducible if and only if its commutant { M φ } ′ \{M_{\varphi}\}' { M φ } ′ consists only of scalar multiples of the identity; that is, the von Neumann algebra
V ∗ ( φ ) = { M φ , M φ ∗ } ′ \mathcal{V}^{*}(\varphi)=\{M_{\varphi},M_{\varphi}^{*}\}' V ∗ ( φ ) = { M φ , M φ ∗ } ′
satisfies V ∗ ( φ ) = C I \mathcal{V}^{*}(\varphi)=\mathbb{C}I V ∗ ( φ ) = C I . It is known that V ∗ ( φ ) \mathcal{V}^{*}(\varphi) V ∗ ( φ ) is generated by all orthogonal projections from H \mathcal{H} H onto reducing subspaces of M φ M_{\varphi} M φ .
Background
A central theme at the interface of function theory and operator theory is the study of commutants of multiplication operators M φ M_{\varphi} M φ (φ ∈ H ∞ ( D ) \varphi\in H^{\infty}(\mathbb{D}) φ ∈ H ∞ ( D ) ), their reducing subspaces, and the von Neumann algebras they generate. A fundamental problem is: When does M φ M_{\varphi} M φ have no nontrivial reducing subspace? That is, when is M φ M_{\varphi} M φ irreducible? By the inner–outer factorization of H 2 ( D ) H^{2}(\mathbb{D}) H 2 ( D ) -functions, it is of particular interest to consider inner functions φ \varphi φ ; By Frostman’s theorem [9 ] , this further reduces to Blaschke products. Though formulated in operator-theoretic language, this question lies at the heart of two seemingly distinct lines of inquiry.
First, the study of invariant subspaces, and in particular reducing subspaces, of operators on function spaces has driven major developments in analysis. Beurling’s theorem [2 ] fully characterizes M z M_z M z -invariant subspaces on H 2 ( D ) H^{2}(\mathbb{D}) H 2 ( D ) . On the Bergman space L a 2 ( D ) L_a^{2}(\mathbb{D}) L a 2 ( D ) , the full invariant-subspace structure remains open due to Aleman–Richter–Sundberg [1 ] . Moreover, the Invariant Subspace Problem for Hilbert-space operators is equivalent to the saturation of the lattice Lat ( M z ) \operatorname{Lat}(M_z) Lat ( M z ) of invariant subspaces of M z M_z M z on L a 2 ( D ) L_a^{2}(\mathbb{D}) L a 2 ( D ) . Attention has therefore turned to the study of reducing subspaces for multiplication operators M φ M_{\varphi} M φ with φ ∈ H ∞ ( D ) \varphi\in H^{\infty}(\mathbb{D}) φ ∈ H ∞ ( D ) , which form a distinguished subclass of invariant subspaces. Considerable progress has been made in this direction, particularly for finite Blaschke products B B B , as well as their counterparts on Hardy and Bergman spaces over domains in C d \mathbb{C}^{d} C d (see [3 , 4 , 7 , 6 , 10 , 17 , 19 , 21 , 28 ] ). The study of reducing subspaces of M B M_B M B on L a 2 ( D ) L_a^{2}(\mathbb{D}) L a 2 ( D ) was initiated in [27 , 31 ] , where M B M_B M B was shown to have finitely many minimal reducing subspaces when B B B is of order 2. Subsequent work in [7 , 17 , 28 ] revealed a key connection: the dimension of V ∗ ( B ) \mathcal{V}^{*}(B) V ∗ ( B ) equals the number of connected components of the Riemann surface S B S_B S B associated with B B B . Douglas, Putinar and Wang [6 ] later proved this number also equals the number of minimal reducing subspaces of M B M_B M B by establishing commutativity of { M B , M B ∗ } ′ \{M_B,M_B^{*}\}' { M B , M B ∗ } ′ for all finite Blaschke products B B B . Very recently, the first three authors and Luo [13 ] introduced the notion of terminal branched components of B B B , via critical point analysis, to classify reducing subspaces of M B M_B M B .
Second, a parallel line of inquiry, rooted in the group factor problem, further motivates the study of reducing subspaces and the von Neumann algebras V ∗ ( ϕ ) \mathcal{V}^*(\phi) V ∗ ( ϕ ) associated with multiplication operators M ϕ M_\phi M ϕ induced by Blaschke products ϕ \phi ϕ . Recall that for n ≥ 1 n \ge1 n ≥ 1 , F n \mathbb{F}_n F n denotes the free group on n n n generators; the Group Factor Problem asks whether L ( F n ) \mathcal{L}(\mathbb{F}_n) L ( F n ) is C ∗ C^* C ∗ -isomorphic to L ( F m ) \mathcal{L}(\mathbb{F}_m) L ( F m ) for distinct n , m ≥ 2 n,m \ge2 n , m ≥ 2 , where L ( F n ) \mathcal{L}(\mathbb{F}_n) L ( F n ) is the von Neumann factor generated by the left regular representation of F n \mathbb{F}_n F n on ℓ 2 ( F n ) \ell^2(\mathbb{F}_n) ℓ 2 ( F n ) . For k ≥ 2 k \ge2 k ≥ 2 , define Ω k = D \ { a 1 , a 2 , ⋯ , a k } \Omega_k = \mathbb{D}\backslash\{a_1,a_2,\cdots,a_k\} Ω k = D \ { a 1 , a 2 , ⋯ , a k } , where a 1 , a 2 , ⋯ , a k ∈ D \ { 0 } a_1,a_2,\cdots,a_k \in\mathbb{D}\backslash\{0\} a 1 , a 2 , ⋯ , a k ∈ D \ { 0 } are distinct points. Then there exists a holomorphic covering map ϕ k : D → Ω k \phi_k:\mathbb{D}\to\Omega_k ϕ k : D → Ω k that is an interpolating Blaschke product (see [14 Example 3.6] ). By [14 Theorem 1.3] ,
V ∗ ( ϕ k ) ≅ L ( π 1 ( Ω k ) ) ≅ L ( F k ) . \mathcal{V}^*(\phi_k) \cong\mathcal{L}(\pi_1(\Omega_k)) \cong\mathcal{L}(\mathbb{F}_k). V ∗ ( ϕ k ) ≅ L ( π 1 ( Ω k )) ≅ L ( F k ) .
Hence, the Group Factor Problem is equivalent to asking whether V ∗ ( ϕ n ) ≅ V ∗ ( ϕ m ) \mathcal{V}^*(\phi_n) \cong\mathcal{V}^*(\phi_m) V ∗ ( ϕ n ) ≅ V ∗ ( ϕ m ) for distinct n , m ≥ 2 n,m \ge2 n , m ≥ 2 ; see [14 ] for further discussion.
These two lines of inquiry converge naturally on the following problem:
Problem. For a given subclass C C C of Blaschke products, can we characterize the irreducibility of the multiplication operator M B M_B M B with B B B in C C C in terms of the function-theoretic or geometric properties of B B B ?
In this paper we focus on a distinguished subclass, namely thin Blaschke products, and investigate how the irreducibility of M B M_B M B on L a 2 ( D ) L_a^2(\mathbb{D}) L a 2 ( D ) relates to the primeness of B B B and the structure of the associated Riemann surface S B S_B S B .
Our main result
We begin with some preliminary notations.
For a Blaschke product B B B , a critical point of B B B is defined as a point in the unit disk D \mathbb{D} D where its derivative B ′ B' B ′ vanishes. Define
Z ( B ′ ) = { z ∈ D : B ′ ( z ) = 0 } , Z(B')=\{z\in\mathbb{D}:B'(z)=0\}, Z ( B ′ ) = { z ∈ D : B ′ ( z ) = 0 } ,
and its critical value set E B \mathcal{E}_B E B by
E B = { B ( z ) : z ∈ Z ( B ′ ) } . \mathcal{E}_B=\{B(z):z\in Z(B')\}. E B = { B ( z ) : z ∈ Z ( B ′ )} .
Recall that a thin Blaschke product is a Blaschke product whose zero sequence { z n } n = 1 ∞ \{z_n\}_{n=1}^{\infty} { z n } n = 1 ∞ satisfies
lim n → ∞ ∏ k ≥ 1 k ≠ n ∣ z k − z n 1 − z k ‾ z n ∣ = 1. \lim_{n\to\infty}\prod_{\substack{k\ge1\\ k\ne n}}\left|\frac{z_k-z_n}{1-\overline{z_k}z_n}\right|=1. n → ∞ lim k ≥ 1 k = n ∏ 1 − z k z n z k − z n = 1.
For a thin/Blaschke product B B B , E B \mathcal{E}_B E B is a discrete subset of D \mathbb{D} D (see [15 Proposition 2.1] for example). We define
S B : = { ( z , w ) ∈ D 2 : B ( z ) = B ( w ) ∉ E B } . S_B:=\{(z,w)\in\mathbb{D}^2:B(z)=B(w)\notin\mathcal{E}_B\}. S B := {( z , w ) ∈ D 2 : B ( z ) = B ( w ) ∈ / E B } .
which is a Riemann surface associated with B B B [15 ] . It is known that every thin Blaschke product B B B has an infinite critical set Z ( B ′ ) Z(B') Z ( B ′ ) (see Proposition 2.8). We say B B B is essentially injective on its critical set if there exists a finite subset F F F of Z ( B ′ ) Z(B') Z ( B ′ ) such that B B B is injective on Z ( B ′ ) ∖ F Z(B')\setminus F Z ( B ′ ) ∖ F .
K. Guo and the second author [15 ] gave a criteria for irreducibility of M B M_B M B when B B B is a thin Blaschke product.
Theorem 1.1. [15 Theorem 4.1] Suppose B B B is a thin Blaschke product satisfying the following:
B B B is essentially injective on its critical set;
There is a finite subset F F F of Z ( B ′ ) Z(B') Z ( B ′ ) such that each point in Z ( B ′ ) ∖ F Z(B')\setminus F Z ( B ′ ) ∖ F is a simple zero of B ′ B' B ′ .
Then V ∗ ( B ) \mathcal{V}^{*}(B) V ∗ ( B ) is trivial; equivalently, the multiplication operator M B M_B M B on L a 2 ( D ) L_a^2(\mathbb{D}) L a 2 ( D ) is irreducible.
Following [26 ] , an inner function f f f is called prime if for any given composition f = u ∘ v f=u\circ v f = u ∘ v with inner functions u u u and v v v , at least one of them is a Möbius transform. Motivated by Theorem 1.1 , we propose the following conjecture.
Conjecture 1.2. Let B B B be a thin Blaschke product. Then the following are equivalent.
B B B is prime;
The von Neumann algebra V ∗ ( B ) \mathcal{V}^{*}(B) V ∗ ( B ) is trivial; equivalently, M B M_B M B on L a 2 ( D ) L_a^2(\mathbb{D}) L a 2 ( D ) is irreducible;
The Riemann surface S B S_B S B has exactly two components.
Remark 1.3. We have some comments for Conjecture 1.2.
The essential injectivity of a thin Blaschke product implies its primeness (see Proposition 4.2). Theorem 1.1 is therefore a special case of Conjecture 1.2.
(2) Condition (2) is equivalent to the statement that every bounded operator on L a 2 ( D ) L_a^2(\mathbb{D}) L a 2 ( D ) can be approximated in the strong operator topology by non-commutative polynomials in M B M_B M B and M B ∗ M_B^* M B ∗ , by the von Neumann Bicommutant Theorem.
(3) In Section 4, we also construct an example of a product B B B of finite many thin Blaschke products, which is not prime, but V ∗ ( B ) \mathcal{V}^*(B) V ∗ ( B ) is trivial. This indicates that Conjecture 1.2 is sharp.
(4) For a thin Blaschke product B B B , [16 Theorem 5.7.2] established that M B M_B M B is irreducible if and only if S B S_B S B has no non-identity component of finite multiplicity; here the identity component is { ( z , z ) ∈ D 2 : B ( z ) ∉ E B } \{(z,z)\in\mathbb{D}^2:B(z)\notin E_B\} {( z , z ) ∈ D 2 : B ( z ) ∈ / E B } . This was first proved in [15 ] under an additional mild assumption.
Conjecture 1.2 refines the existing criterion in two respects: it strengthens the condition on S B S_B S B to exactly two components, and it introduces a purely function-theoretic condition, primeness.
We propose Conjecture 1.2 partially based on the implication chain (3) ⇒ \Rightarrow ⇒ (2) ⇒ \Rightarrow ⇒ (1) (see the proof of Proposition 4.3 ). It remains open whether the reverse implications (1) ⇒ \Rightarrow ⇒ (3) or (1) ⇒ \Rightarrow ⇒ (2) holds. The following conjecture, which asserts (1) ⇒ \Rightarrow ⇒ (3), is therefore equivalent to Conjecture 1.2.
Conjecture 1.4. Let B B B be a thin Blaschke product. If B B B is prime, then the Riemann surface S B S_B S B has exactly two components.
Our main result provides substantial support for Conjecture 1.4 . Moreover, it essentially generalized Theorem 1.1.
Theorem 1.5. Suppose B B B is a thin Blaschke product such that B B B is essentially injective on its critical set. Then the Riemann surface S B S_B S B has exactly two components. Consequently, V ∗ ( B ) \mathcal{V}^*(B) V ∗ ( B ) is trivial; equivalently, the multiplication operator M B M_B M B on L a 2 ( D ) L_a^2(\mathbb{D}) L a 2 ( D ) is irreducible.
For thin Blaschke products, primeness and essential injectivity are closely related. In fact, if B B B is a thin Blaschke product and B B B is essentially injective on its critical points, then B B B is prime (see Proposition 4.2). It seems that the two properties are not far apart. One reason is that essential injectivity carries a clear geometric intuition. In fact, we will show later (see Proposition 4.1 ) that for any thin Blaschke product B B B , the following are equivalent:
(I) The mapping B B B is not essentially injective on its critical set.
(II) There exist two sequences { z n } n = 1 ∞ \{z_n\}_{n=1}^{\infty} { z n } n = 1 ∞ and { w n } n = 1 ∞ \{w_n\}_{n=1}^{\infty} { w n } n = 1 ∞ in D \mathbb{D} D such that
B ( z n ) = B ( w n ) and z n ≠ w n for all n ; (1) B(z_n)=B(w_n)\ \text{and}\ z_n\ne w_n\ \text{for all }n; \tag*{(1)} B ( z n ) = B ( w n ) and z n = w n for all n ; (1)
B ′ ( z n ) = B ′ ( w n ) = 0 ; (2) B'(z_n)=B'(w_n)=0; \tag*{(2)} B ′ ( z n ) = B ′ ( w n ) = 0 ; (2)
B ( z n ) ≠ B ( z m ) if m ≠ n . (3) B(z_n)\ne B(z_m)\ \text{if }m\ne n. \tag*{(3)} B ( z n ) = B ( z m ) if m = n . (3)
By [11 Theorem 4.1] , if B B B is a thin Blaschke product with the decomposition B = u ∘ v B=u\circ v B = u ∘ v , u u u and v v v being inner functions, then either u u u is a Möbius map or v v v is a finite Blaschke product. Using this result, we can show that a non-prime thin Blaschke product B B B satisfies Condition (II); that is, B B B is not essentially injective on its critical set. This observation, along with Theorem 1.5 , lends some support to Conjecture 1.4.
It is tempting to conjecture that primeness and essential injectivity are equivalent for thin Blaschke products. If this were true, Conjectures 1.2 and 1.4 would follow immediately from Theorem 1.5 . However, in § 7 , a prime thin Blaschke product that fails to be essentially injective on its critical set is constructed. This example does not contradict Conjectures 1.2 or 1.4.
The strategy for our main theorem
Our main idea is as follows. A thin Blaschke product B B B can be viewed as the union of countably many finite Blaschke products. In fact, we can choose increasing t n ∈ ( 0 , 1 ) t_n\in(0,1) t n ∈ ( 0 , 1 ) and simply connected components Ω n \Omega_n Ω n of { ∣ B ∣ < t n } \{|B|<t_n\} { ∣ B ∣ < t n } with Ω n ⊊ Ω n + 1 \Omega_n\subsetneq\Omega_{n+1} Ω n ⊊ Ω n + 1 and ⋃ n = 1 ∞ Ω n = D \bigcup_{n=1}^{\infty}\Omega_n=\mathbb{D} ⋃ n = 1 ∞ Ω n = D . Each restriction B ∣ Ω n : Ω n → D B|_{\Omega_n}:\Omega_n\to\mathbb{D} B ∣ Ω n : Ω n → D is a proper holomorphic map, factoring as a finite Blaschke product conjugated by conformal maps. This reduces the analysis of B B B to that of the sequence { B ∣ Ω n } \{B|_{\Omega_n}\} { B ∣ Ω n } . In this way, the proof for the following theorem (the case of finite Blaschke product) serves as the core ingredient of Theorem 1.5 .
Theorem 1.6. Suppose B B B is a finite Blaschke product with at least two distinct critical points and B ∣ Z ( B ′ ) B|_{Z(B')} B ∣ Z ( B ′ ) is injective. Then S B S_B S B has exactly two components. In this case, the multiplication operator M B M_B M B on L a 2 ( D ) L_a^2(\mathbb{D}) L a 2 ( D ) has exactly two minimal reducing subspaces.
For a thin Blaschke product B B B , Theorem 1.5 gives a function-theoretic criterion for the irreducibility of M B M_B M B . Our perspective is to view B B B as the limit of an exhaustion by finite Blaschke products. The proof of Theorem 1.5 rests on two key ideas.
First, the proof of Theorem 1.5 proceeds by analyzing the critical points of B B B . Unlike the earlier result in Theorem 1.1 , Theorem 1.5 allows critical points of higher order. For finite/thin Blaschke products, the Riemann surface S B S_B S B can be studied via local inverses, where points in a fiber B − 1 ( w ) B^{-1}(w) B − 1 ( w ) are connected by analytic continuation. In the presence of higher-order critical points, however, it becomes difficult to determine which local inverses lie in the same equivalence class. A global argument is therefore needed, and finally we prove that S B S_B S B has exactly two components under the stated conditions. This global analysis is the the key new contribution of Theorem 1.5, compared with earlier results in [15 ] .
Second, our approach differs substantially from that of [13 ] , where only a single value t 0 ∈ ( 0 , 1 ) t_0 \in(0,1) t 0 ∈ ( 0 , 1 ) and the components of { z ∈ D : ∣ B ( z ) ∣ < t 0 } \{z \in\mathbb{D} : |B(z)| < t_0\} { z ∈ D : ∣ B ( z ) ∣ < t 0 } (the terminal branched components) were needed. Here, by contrast, we must treat { z ∈ D : ∣ B ( z ) ∣ < t } \{z \in\mathbb{D} : |B(z)| < t\} { z ∈ D : ∣ B ( z ) ∣ < t } for every t ∈ ( 0 , 1 ] t \in(0,1] t ∈ ( 0 , 1 ] . To this end, we partition ( 0 , 1 ] (0,1] ( 0 , 1 ] into finitely many subintervals and carry out an induction on t t t over these subintervals. We then prove that the sub-Riemann surfaces of S B S_B S B lying on { z ∈ D : ∣ B ( z ) ∣ < t } \{z \in\mathbb{D} : |B(z)| < t\} { z ∈ D : ∣ B ( z ) ∣ < t } admit exactly two components under suitable conditions, and finally show that S B S_B S B itself has precisely two components.
The paper is organized as follows. Section 2 presents preliminaries on finite and thin Blaschke products and establishes several lemmas. Section 3 gives the proofs of Theorems 1.5 and 1.6. Section 4 shows that the essential injectivity of a thin Blaschke product implies its primeness, and also establishes the implications (3) ⇒ \Rightarrow ⇒ (2) ⇒ \Rightarrow ⇒ (1) in Conjecture 1.2. Section 5 constructs a product B B B of two thin Blaschke products, showing that Conditions (1)–(3) in Conjecture 2 are not equivalent for this B B B . Finally, Section 6 constructs a prime thin Blaschke product that is not essentially injective on its critical set.
Throughout this paper, B B B denotes a finite/thin Blaschke product. For 0 < t < 1 0 < t < 1 0 < t < 1 , the set { ∣ B ∣ < t } \{|B| < t\} { ∣ B ∣ < t } denotes { z ∈ D : ∣ B ( z ) ∣ < t } \{z \in\mathbb{D} : |B(z)| < t\} { z ∈ D : ∣ B ( z ) ∣ < t } , and { ∣ B ∣ ≤ t } \{|B| \le t\} { ∣ B ∣ ≤ t } denotes { z ∈ D : ∣ B ( z ) ∣ ≤ t } \{z \in\mathbb{D} : |B(z)| \le t\} { z ∈ D : ∣ B ( z ) ∣ ≤ t } .
Some preliminaries
Finite and thin Blaschke products
In this subsection, we collect some facts about finite and thin Blaschke products.
By direct computations, the derivative of a finite Blaschke product B B B vanishes nowhere on the unit circle. Remind that a critical point of a Blaschke product B B B refers to a point in D \mathbb{D} D where B ′ B' B ′ vanishes. We have the following statement [8 ] .
Lemma 2.1. A Blaschke product of order n n n contains exactly n − 1 n-1 n − 1 critical points counting multiplicity in the unit disk D \mathbb{D} D .
The following lemma is from [15 Lemma 2.2] . Given two domains Ω 1 , Ω 2 \Omega_1, \Omega_2 Ω 1 , Ω 2 , a holomorphic map f : Ω 1 → Ω 2 f:\Omega_1 \to\Omega_2 f : Ω 1 → Ω 2 is called proper if the preimage of every compact subset of Ω 2 \Omega_2 Ω 2 is compact in Ω 1 \Omega_1 Ω 1 . Equivalently, for any sequence { z n } n = 1 ∞ \{z_n\}_{n=1}^{\infty} { z n } n = 1 ∞ satisfying z n → ∂ Ω 1 z_n \to\partial\Omega_1 z n → ∂ Ω 1 , we have f ( z n ) → ∂ Ω 2 f(z_n) \to\partial\Omega_2 f ( z n ) → ∂ Ω 2 .
Lemma 2.2. Suppose that ϕ \phi ϕ is a holomorphic function over D \mathbb{D} D and Δ \Delta Δ is a planar domain. If Δ 0 \Delta_0 Δ 0 is a component of ϕ − 1 ( Δ ) \phi^{-1}(\Delta) ϕ − 1 ( Δ ) such that Δ 0 ⊆ D \Delta_0 \subseteq\mathbb{D} Δ 0 ⊆ D , then the restriction ϕ ∣ Δ 0 \phi|_{\Delta_0} ϕ ∣ Δ 0 is a proper map onto Δ \Delta Δ .
Lemma 2.2 directly implies Proposition 2.3 below; see also [13 Proposition 2.7] .
Proposition 2.3. Let B B B be a finite Blaschke product, and let Ω \Omega Ω be a component of { ∣ B ∣ < t } \{|B|<t\} { ∣ B ∣ < t } for t ∈ ( 0 , 1 ) t \in(0,1) t ∈ ( 0 , 1 ) . Then Ω \Omega Ω is a simply connected domain. In this case, there exists a conformal map G G G from Ω \Omega Ω onto the unit disk, and a finite Blaschke product φ \varphi φ such that
B ∣ Ω = t φ ∘ G . B|_{\Omega}=t\varphi\circ G. B ∣ Ω = tφ ∘ G .
We include a proof of the following fact for completeness; see also [13 Remark 3.9(e)] .
Lemma 2.4. Let B B B be a finite Blaschke product. For each s s s in ( 0 , 1 ) (0,1) ( 0 , 1 ) , write
Ω s : = { z ∈ D : ∣ B ( z ) ∣ < s } . \Omega_s:=\{z\in\mathbb{D}:|B(z)|<s\}. Ω s := { z ∈ D : ∣ B ( z ) ∣ < s } . There exists a number t t t in ( 0 , 1 ) (0,1) ( 0 , 1 ) such that Ω s \Omega_s Ω s is connected for all s ≥ t s\ge t s ≥ t .
Proof. Let r r r be the maximum module of zeros of B B B , and let t t t be the maximum of B B B on the closed disk { z ∈ D : ∣ z ∣ ≤ r + 1 2 } \{z\in\mathbb{D}:|z|\le\frac{r+1}{2}\} { z ∈ D : ∣ z ∣ ≤ 2 r + 1 } . For s ≥ t s\ge t s ≥ t , let Ω 1 \Omega_1 Ω 1 be the component of Ω s \Omega_s Ω s containing all zeros of B B B . We will prove that Ω s = Ω 1 \Omega_s=\Omega_1 Ω s = Ω 1 to complete the argument. Suppose for contradiction that there exists another distinct component Ω 2 ⊆ Ω s \Omega_2\subseteq\Omega_s Ω 2 ⊆ Ω s . Then by Lemma 2.2 B ∣ Ω 2 : Ω 2 → s D B|_{\Omega_2}:\Omega_2\to s\mathbb{D} B ∣ Ω 2 : Ω 2 → s D is a proper map, and onto. In particular, B B B has at least one zero in Ω 2 \Omega_2 Ω 2 . This leads to a contradiction with the fact that Ω 1 \Omega_1 Ω 1 contains all zeros of B B B .
The next result is included in [13 Theorem 2.6] .
Proposition 2.5. Let ϕ \phi ϕ be a holomorphic map from D \mathbb{D} D to D \mathbb{D} D , and fix 0 < t < 1 0<t<1 0 < t < 1 . If Ω \Omega Ω is a component of { ∣ ϕ ∣ < t } \{|\phi|<t\} { ∣ ϕ ∣ < t } and Ω ‾ ⊆ D \overline{\Omega}\subseteq\mathbb{D} Ω ⊆ D , then Ω \Omega Ω is bounded by a Jordan curve.
Proof. The idea of the proof comes from that in [13 Theorem 2.6] . Since ϕ \phi ϕ is holomorphic on D \mathbb{D} D , by the Maximal Modulus Principle, it is routine to show that each component Ω \Omega Ω of the set { ∣ ϕ ∣ < t } \{|\phi| < t\} { ∣ ϕ ∣ < t } is simply connected. It remains to prove that ∂ Ω \partial\Omega ∂ Ω is a Jordan curve.
Assume for contradiction that ∂ Ω \partial\Omega ∂ Ω is not a Jordan curve. Then ∂ Ω \partial\Omega ∂ Ω contains a Jordan curve γ \gamma γ ; let Ω γ \Omega_\gamma Ω γ denote its interior. Choose Γ \Gamma Γ among all such curves so that Ω Γ \Omega_\Gamma Ω Γ is maximal with respect to inclusion. Since ∣ ϕ ∣ = t |\phi| = t ∣ ϕ ∣ = t on Γ \Gamma Γ and ∣ ϕ ∣ < t |\phi| < t ∣ ϕ ∣ < t on Ω Γ \Omega_\Gamma Ω Γ , and since Ω ⊆ { ∣ ϕ ∣ < t } \Omega\subseteq\{|\phi| < t\} Ω ⊆ { ∣ ϕ ∣ < t } , it follows that Ω ⊆ Ω Γ \Omega\subseteq\Omega_\Gamma Ω ⊆ Ω Γ . However, as Ω \Omega Ω is a component of { ∣ ϕ ∣ < t } \{|\phi| < t\} { ∣ ϕ ∣ < t } , we must have Ω = Ω Γ \Omega= \Omega_\Gamma Ω = Ω Γ . This implies that ∂ Ω = Γ \partial\Omega= \Gamma ∂ Ω = Γ , which contradicts the assumption that ∂ Ω \partial\Omega ∂ Ω is not a Jordan curve. Therefore, ∂ Ω \partial\Omega ∂ Ω must be a Jordan curve, as desired. □ \square □
The following thin Blaschke product version of Proposition 2.3 appears in [15 Proposition 2.3] ; we include a proof for completeness.
Proposition 2.6. If B B B is a thin Blaschke product, then for each t ∈ ( 0 , 1 ) t \in(0,1) t ∈ ( 0 , 1 ) , every component Ω \Omega Ω of { ∣ B ∣ < t } \{|B| < t\} { ∣ B ∣ < t } is compactly contained in D \mathbb{D} D , and B : Ω → t D B:\Omega\to t\mathbb{D} B : Ω → t D is a proper map between simply connected domains.
Define the pseudo-hyperbolic distance
d ( z , w ) = ∣ z − w 1 − z ‾ w ∣ . d(z,w) = \left|\frac{z-w}{1-\overline{z}w}\right|. d ( z , w ) = 1 − z w z − w .
To prove Proposition 2.6, we need the following lemma (see [12 , 20 ] , and [9 Lemma 4.2] ).
Lemma 2.7. For a thin Blaschke product B B B with zero sequence { z n } \{z_n\} { z n } and every δ ∈ ( 0 , 1 ) \delta\in(0,1) δ ∈ ( 0 , 1 ) , there exists an ε ∈ ( 0 , 1 ) \varepsilon\in(0,1) ε ∈ ( 0 , 1 ) such that ∣ B ( z ) ∣ ≥ δ |B(z)| \ge\delta ∣ B ( z ) ∣ ≥ δ whenever d ( z , z n ) ≥ ε d(z,z_n) \ge\varepsilon d ( z , z n ) ≥ ε for all n n n .
Proof of Proposition 2.6. The idea of the proof is from that of [15 Proposition 2.3] .
For each t ∈ ( 0 , 1 ) t \in(0,1) t ∈ ( 0 , 1 ) , let Ω \Omega Ω be a connected component of { ∣ B ∣ < t } \{|B| < t\} { ∣ B ∣ < t } . It is straightforward to verify that Ω \Omega Ω is simply connected. Our goal is to prove that B : Ω → t D B:\Omega\to t\mathbb{D} B : Ω → t D is a proper map.
Given that B B B is thin with the zero sequence { z n } n = 1 ∞ \{z_n\}_{n=1}^{\infty} { z n } n = 1 ∞ , by Lemma 2.7, there exists ε ∈ ( 0 , 1 ) \varepsilon\in(0,1) ε ∈ ( 0 , 1 ) satisfying
B − 1 ( Δ ( 0 , t ) ) ⊆ ⋃ n = 1 ∞ Δ ( z n , ε ) . (4) B^{-1}(\Delta(0,t)) \subseteq\bigcup_{n=1}^{\infty}\Delta(z_n,\varepsilon). \tag*{(4)} B − 1 ( Δ ( 0 , t )) ⊆ n = 1 ⋃ ∞ Δ ( z n , ε ) . (4)
For intersecting disks Δ ( z n , ε ) ∩ Δ ( z k , ε ) \Delta(z_n,\varepsilon)\cap\Delta(z_k,\varepsilon) Δ ( z n , ε ) ∩ Δ ( z k , ε ) containing a point z ′ z' z ′ , we apply [16 Lemma 1.4] to obtain
d ( z n , z k ) ≤ d ( z n , z ′ ) + d ( z ′ , z k ) 1 + d ( z n , z ′ ) d ( z ′ , z k ) ≤ d ε , (5) d(z_n,z_k) \le\frac{d(z_n,z')+d(z',z_k)}{1+d(z_n,z')d(z',z_k)} \le d_\varepsilon, \tag*{(5)} d ( z n , z k ) ≤ 1 + d ( z n , z ′ ) d ( z ′ , z k ) d ( z n , z ′ ) + d ( z ′ , z k ) ≤ d ε , (5)
where
d ε = max { x + y 1 + x y : 0 ≤ x , y ≤ ε } < 1. d_\varepsilon=\max\left\{\frac{x+y}{1+xy}:0\le x,y\le\varepsilon\right\}<1. d ε = max { 1 + x y x + y : 0 ≤ x , y ≤ ε } < 1.
By the thinness condition of B B B , we have
lim n → ∞ ∏ k ≥ 1 k ≠ n d ( z n , z k ) = 1. \lim_{n\to\infty}\prod_{\substack{k\ge1\\k\ne n}}d(z_n,z_k)=1. n → ∞ lim k ≥ 1 k = n ∏ d ( z n , z k ) = 1.
There exists N ≥ 1 N\ge1 N ≥ 1 such that for all n ≥ N n\ge N n ≥ N ,
d ( z n , z k ) ≥ 1 + d ε 2 > d ε . d(z_n,z_k)\ge\frac{1+d_\varepsilon}{2}>d_\varepsilon. d ( z n , z k ) ≥ 2 1 + d ε > d ε .
Consequently, by (2.1), Δ ( z n , ε ) \Delta(z_n,\varepsilon) Δ ( z n , ε ) has no intersection with other disks Δ ( z m , ε ) \Delta(z_m,\varepsilon) Δ ( z m , ε ) with m ≠ n m\ne n m = n . The union K = ⋃ n = 0 N Δ ( z n , ε ) ‾ K=\bigcup_{n=0}^{N}\overline{\Delta(z_n,\varepsilon)} K = ⋃ n = 0 N Δ ( z n , ε ) is compact. For every component Ω \Omega Ω of ∣ B ∣ < δ \lvert B\rvert<\delta ∣ B ∣ < δ , either Ω ⊂ K \Omega\subset K Ω ⊂ K , or Ω ⊂ Δ ( z n , ε ) \Omega\subset\Delta(z_n,\varepsilon) Ω ⊂ Δ ( z n , ε ) for some n ≥ N n\ge N n ≥ N . In both cases, Ω ‾ \overline{\Omega} Ω is compact. Furthermore, standard analysis shows B ( ∂ Ω ) ⊂ ∂ ( t D ) B(\partial\Omega)\subset\partial(t\mathbb{D}) B ( ∂ Ω ) ⊂ ∂ ( t D ) . Therefore, by continuity of B B B on Ω ‾ \overline{\Omega} Ω , B B B is a proper map as desired. □
The following result, while likely known in the literature, is included here for completeness.
Proposition 2.8. Every thin Blaschke product has infinitely many critical points in D \mathbb{D} D .
Proof. Let B B B be a thin Blaschke product with the zero sequence z 1 , z 2 , … z_1,z_2,\ldots z 1 , z 2 , … . For every n ≥ 2 n\ge2 n ≥ 2 , consider a disk D D D containing the first n n n zeros z 1 , … , z n z_1,\ldots,z_n z 1 , … , z n . There exists t ∈ ( 0 , 1 ) t\in(0,1) t ∈ ( 0 , 1 ) such that ∣ B ∣ < t \lvert B\rvert<t ∣ B ∣ < t on D ‾ \overline{D} D . Let Ω \Omega Ω be the component of { ∣ B ∣ < t } \{\lvert B\rvert<t\} {∣ B ∣ < t } containing D ‾ \overline{D} D . Then the restriction B ∣ Ω : Ω → D B|_{\Omega}:\Omega\to\mathbb{D} B ∣ Ω : Ω → D is a finite Blaschke product of degree at least n n n . More precisely, let F F F be a conformal map from the unit disk D \mathbb{D} D onto Ω \Omega Ω . Defining h = 1 t B ∣ Ω ∘ F h=\frac{1}{t}B|_{\Omega}\circ F h = t 1 B ∣ Ω ∘ F , we observe that h h h is a proper map from D \mathbb{D} D onto D \mathbb{D} D . Then by [24 7.3.1] , h h h must be a finite Blaschke product, and B ∣ Ω = t h ∘ F − 1 B|_{\Omega}=th\circ F^{-1} B ∣ Ω = t h ∘ F − 1 . Therefore, B ∣ Ω B|_{\Omega} B ∣ Ω inherits at least n − 1 n-1 n − 1 critical points from h h h , counting multiplicity. Since n n n is arbitrary, we conclude that B B B possesses infinitely many critical points in D \mathbb{D} D . □
Thin Blaschke products have many good properties. The following characterization is due to [30 Theorem 2.3] .
Theorem 2.9. A Blaschke product B B B is thin if and only if
lim ∣ z ∣ → 1 − [ ∣ B ( z ) ∣ 2 + ( 1 − ∣ z ∣ 2 ) ∣ B ′ ( z ) ∣ ] = 1. \lim_{|z|\to1^-}\left[|B(z)|^2+(1-|z|^2)|B'(z)|\right]=1. ∣ z ∣ → 1 − lim [ ∣ B ( z ) ∣ 2 + ( 1 − ∣ z ∣ 2 ) ∣ B ′ ( z ) ∣ ] = 1.
For w ∈ D w\in\mathbb{D} w ∈ D , define φ w ( z ) = w − z 1 − w ‾ z \varphi_w(z)=\frac{w-z}{1-\overline{w}z} φ w ( z ) = 1 − w z w − z , z ∈ D z\in\mathbb{D} z ∈ D . As a consequence of the above theorem, we have the following (see [11 Proposition 4.3] and [5 Proposition 4] ).
Proposition 2.10. Let B B B be a thin Blaschke product, w ∈ D w\in\mathbb{D} w ∈ D and φ \varphi φ be a finite Blaschke product. Then both φ w ∘ B \varphi_w\circ B φ w ∘ B and B ∘ φ B\circ\varphi B ∘ φ are also thin Blaschke products.
Local inverse and gluable property
In this subsection, we present the definitions of local inverse and analytic continuation. Also, we recall the gluable property of finite and thin Blaschke products from [15 ] .
To begin with, we recall the notions of analytic continuation and local inverse (see [25 Chapter 16] and [29 ] , respectively). A function element is an ordered pair ( f , D ) (f,D) ( f , D ) , where D D D is an open disk and f f f is a holomorphic function on D D D . A curve is defined as a continuous map γ : [ 0 , 1 ] → C \gamma:[0,1]\to\mathbb{C} γ : [ 0 , 1 ] → C . Consider a function element ( f 0 , D 0 ) (f_0,D_0) ( f 0 , D 0 ) and a curve γ \gamma γ with γ ( 0 ) ∈ D 0 \gamma(0)\in D_0 γ ( 0 ) ∈ D 0 . If there exists a partition
0 = s 0 < s 1 < ⋯ < s n = 1 0=s_0<s_1<\cdots<s_n=1 0 = s 0 < s 1 < ⋯ < s n = 1
of [ 0 , 1 ] [0,1] [ 0 , 1 ] , and a collection of function elements { ( f i , D i ) } 0 ≤ i ≤ n \{(f_i,D_i)\}_{0\leq i\leq n} {( f i , D i ) } 0 ≤ i ≤ n so that ( f j , D j ) (f_j,D_j) ( f j , D j ) and ( f j + 1 , D j + 1 ) (f_{j+1},D_{j+1}) ( f j + 1 , D j + 1 ) are direct continuations, and γ ( [ s j , s j + 1 ] ) ⊆ D j \gamma([s_j,s_{j+1}])\subseteq D_j γ ([ s j , s j + 1 ]) ⊆ D j for 0 ≤ j ≤ n − 1 0\leq j\leq n-1 0 ≤ j ≤ n − 1 with γ ( 1 ) ∈ D n \gamma(1)\in D_n γ ( 1 ) ∈ D n , then ( f n , D n ) (f_n,D_n) ( f n , D n ) is called an analytic continuation of ( f 0 , D 0 ) (f_0,D_0) ( f 0 , D 0 ) along γ \gamma γ , and we say ( f 0 , D 0 ) (f_0,D_0) ( f 0 , D 0 ) admits analytic continuation along γ \gamma γ . This relationship is denoted by f 0 ∼ f n f_0\sim f_n f 0 ∼ f n , which establishes an equivalence relation. For a finite/thin Blaschke product B B B , and a subdomain Δ \Delta Δ of D ∖ B − 1 ( E B ) \mathbb{D}\setminus B^{-1}(\mathcal{E}_B) D ∖ B − 1 ( E B ) , a holomorphic map ρ \rho ρ on Δ \Delta Δ is called a local inverse of B B B if B ∘ ρ = B B\circ\rho=B B ∘ ρ = B . For two local inverses ρ 1 \rho_1 ρ 1 and ρ 2 \rho_2 ρ 2 of B B B , we say ρ 1 ∼ ρ 2 \rho_1\sim\rho_2 ρ 1 ∼ ρ 2 if ρ 1 \rho_1 ρ 1 is an analytic continuation of ρ 2 \rho_2 ρ 2 along a curve in D ∖ B − 1 ( E B ) \mathbb{D}\setminus B^{-1}(\mathcal{E}_B) D ∖ B − 1 ( E B ) .
For every a , b ∈ D ∖ B − 1 ( E B ) a,b\in\mathbb{D}\setminus B^{-1}(\mathcal{E}_B) a , b ∈ D ∖ B − 1 ( E B ) , there are small neighborhoods V a V_a V a of a a a , V b V_b V b of b b b , and a disk D D D such that both B : V a → D B:V_a\to D B : V a → D and B : V b → D B:V_b\to D B : V b → D are biholomorphic maps. Then define
ρ a , b ( z ) = ( B ∣ V b ) − 1 ∘ B ( z ) , z ∈ V a . \rho_{a,b}(z)=(B|_{V_b})^{-1}\circ B(z),\qquad z\in V_a. ρ a , b ( z ) = ( B ∣ V b ) − 1 ∘ B ( z ) , z ∈ V a .
We have ρ a , b ( a ) = b \rho_{a,b}(a)=b ρ a , b ( a ) = b , and B ( ρ a , b ) = B B(\rho_{a,b})=B B ( ρ a , b ) = B . If ρ \rho ρ is a local inverse defined on a neighborhood V V V of a a a and ρ ( a ) = b \rho(a)=b ρ ( a ) = b , then it is routine to check that
ρ = ρ a , b \rho=\rho_{a,b} ρ = ρ a , b
on V ∩ V a V\cap V_a V ∩ V a . In this way, ρ ∼ ρ a , b \rho\sim\rho_{a,b} ρ ∼ ρ a , b .
The following notion introduced in [15 ] plays an important role in the proof of Theorem 1.6 .
Definition 2.11. Let ψ \psi ψ be an analytic function on a domain Ω \Omega Ω . For two distinct points z 1 , z 2 ∈ Ω \ ψ − 1 ( E ψ ) z_1,z_2\in\Omega\backslash\psi^{-1}(E_\psi) z 1 , z 2 ∈ Ω\ ψ − 1 ( E ψ ) with ψ ( z 1 ) = ψ ( z 2 ) \psi(z_1)=\psi(z_2) ψ ( z 1 ) = ψ ( z 2 ) , if there exist curves σ 1 \sigma_1 σ 1 and σ 2 \sigma_2 σ 2 such that
(1) σ i ( 0 ) = z i \sigma_i(0)=z_i σ i ( 0 ) = z i for i = 1 , 2 i=1,2 i = 1 , 2 , and σ 1 ( 1 ) = σ 2 ( 1 ) \sigma_1(1)=\sigma_2(1) σ 1 ( 1 ) = σ 2 ( 1 ) ;
(2) ψ ( σ 1 ( t ) ) = ψ ( σ 2 ( t ) ) \psi(\sigma_1(t))=\psi(\sigma_2(t)) ψ ( σ 1 ( t )) = ψ ( σ 2 ( t )) for 0 ≤ t ≤ 1 0\leq t\leq1 0 ≤ t ≤ 1 ;
(3) σ i ( t ) ∈ Ω \ ψ − 1 ( E ψ ) \sigma_i(t)\in\Omega\backslash\psi^{-1}(E_\psi) σ i ( t ) ∈ Ω\ ψ − 1 ( E ψ ) , i = 1 , 2 i=1,2 i = 1 , 2 and 0 ≤ t < 1 0\leq t<1 0 ≤ t < 1 ,
then z 1 z_1 z 1 and z 2 z_2 z 2 are said to be directly ψ \psi ψ -glued in Ω \Omega Ω , denoted by z 1 ∼ ψ z 2 z_1\overset{\psi}{\sim}z_2 z 1 ∼ ψ z 2 .
Roughly, two points are directly ψ \psi ψ -glued if they are connected by a ψ \psi ψ -equal curve. For example, set ψ ( z ) = z n \psi(z)=z^n ψ ( z ) = z n (n > 1 n>1 n > 1 ) and Ω = D \Omega=\mathbb{D} Ω = D . If z 1 ≠ z 2 z_1\ne z_2 z 1 = z 2 and z 1 n = z 2 n z_1^n=z_2^n z 1 n = z 2 n , then we have z 1 ∼ ψ z 2 z_1\overset{\psi}{\sim}z_2 z 1 ∼ ψ z 2 . In fact, defining σ 1 ( t ) = ( 1 − t ) z 1 \sigma_1(t)=(1-t)z_1 σ 1 ( t ) = ( 1 − t ) z 1 and σ 2 ( t ) = ( 1 − t ) z 2 \sigma_2(t)=(1-t)z_2 σ 2 ( t ) = ( 1 − t ) z 2 , we have ψ ( σ 1 ( t ) ) = ψ ( σ 2 ( t ) ) \psi(\sigma_1(t))=\psi(\sigma_2(t)) ψ ( σ 1 ( t )) = ψ ( σ 2 ( t )) for all 0 ≤ t ≤ 1 0\leq t\leq1 0 ≤ t ≤ 1 .
The following is from [16 Definition 4.4] .
Definition 2.12. Let Ω \Omega Ω be a subdomain of D \mathbb{D} D . If there exists a finite sequence z 0 , z 1 , ⋯ , z k z_0,z_1,\cdots,z_k z 0 , z 1 , ⋯ , z k in Ω \Omega Ω such that z j z_j z j and z j + 1 z_{j+1} z j + 1 are directly ψ \psi ψ -glued in Ω \Omega Ω for 0 ≤ j ≤ k − 1 0\leq j\leq k-1 0 ≤ j ≤ k − 1 , and all related curves σ i \sigma_i σ i in Definition 2.11 can be chosen in Ω \Omega Ω , then z 0 z_0 z 0 and z k z_k z k are said to be ψ \psi ψ -glued on Ω \Omega Ω .
A function ψ \psi ψ is called gluable if for every w ′ ∈ D \ ψ − 1 ( E ψ ) w'\in\mathbb{D}\backslash\psi^{-1}(E_\psi) w ′ ∈ D \ ψ − 1 ( E ψ ) , and every w ′ ′ ≠ w ′ w''\ne w' w ′′ = w ′ satisfying ψ ( w ′ ′ ) = ψ ( w ′ ) \psi(w'')=\psi(w') ψ ( w ′′ ) = ψ ( w ′ ) , the points w ′ w' w ′ and w ′ ′ w'' w ′′ are ψ \psi ψ -glued in D \mathbb{D} D .
The following is from [16 Propositions 4.5 and 4.6] .
Proposition 2.13. A finite/thin Blaschke product is gluable. Therefore, a proper map between two simply connected domains is gluable.
For a holomorphic proper map ϕ : Ω 1 → Ω 2 \phi:\Omega_1\to\Omega_2 ϕ : Ω 1 → Ω 2 , it is a standard fact that there exists an integer k ≥ 1 k\geq1 k ≥ 1 such that for every w ∈ Ω 2 w\in\Omega_2 w ∈ Ω 2 , the equation ϕ ( z ) = w \phi(z)=w ϕ ( z ) = w has exactly k k k solutions in Ω 1 \Omega_1 Ω 1 , counting multiplicities. In this case, we call ϕ : Ω 1 → Ω 2 \phi:\Omega_1\to\Omega_2 ϕ : Ω 1 → Ω 2 is of multiplicity k k k . Following [13 ] , we present the following notion.
Definition 2.14. For a holomorphic proper map ϕ : Ω 1 → Ω 2 \phi:\Omega_1 \to\Omega_2 ϕ : Ω 1 → Ω 2 , we say Ω 1 \Omega_1 Ω 1 is of multiplicity k k k with respect to ϕ \phi ϕ if ϕ : Ω 1 → Ω 2 \phi:\Omega_1 \to\Omega_2 ϕ : Ω 1 → Ω 2 is of multiplicity k k k .
The following result is from [13 Lemma 2.12] .
Lemma 2.15. Let ϕ \phi ϕ be a holomorphic function on a planar domain V V V . Let σ 1 , σ 2 : [ 0 , 1 ) → V \sigma_1,\sigma_2:[0,1) \to V σ 1 , σ 2 : [ 0 , 1 ) → V be two disjoint curves satisfying:
ϕ ( σ 1 ( s ) ) = ϕ ( σ 2 ( s ) ) , 0 ≤ s < 1 , and \phi(\sigma_1(s))=\phi(\sigma_2(s)), \quad0 \le s < 1, \quad\text{and} ϕ ( σ 1 ( s )) = ϕ ( σ 2 ( s )) , 0 ≤ s < 1 , and ϕ ′ ( σ 1 ( s ) ) ≠ 0 and ϕ ′ ( σ 2 ( s ) ) ≠ 0 , 0 ≤ s < 1. \phi'(\sigma_1(s)) \ne0 \quad\text{and} \quad\phi'(\sigma_2(s)) \ne0, \quad0 \le s < 1. ϕ ′ ( σ 1 ( s )) = 0 and ϕ ′ ( σ 2 ( s )) = 0 , 0 ≤ s < 1. For each s ∈ [ 0 , 1 ) s \in[0,1) s ∈ [ 0 , 1 ) , let ρ s \rho_s ρ s be the local inverse of ϕ \phi ϕ defined in a neighborhood of σ 1 ( s ) \sigma_1(s) σ 1 ( s ) such that
ρ s ( σ 1 ( s ) ) = σ 2 ( s ) for all 0 ≤ s < 1. \rho_s(\sigma_1(s))=\sigma_2(s) \quad\text{for all } 0 \le s < 1. ρ s ( σ 1 ( s )) = σ 2 ( s ) for all 0 ≤ s < 1. Then all these ρ s \rho_s ρ s extends analytically to each other along the curve σ 1 \sigma_1 σ 1 , that is, ρ 0 ∼ ρ s \rho_0 \sim\rho_s ρ 0 ∼ ρ s , 0 ≤ s < 1 0 \le s < 1 0 ≤ s < 1 .
We need Lemma 5.8.3 in [16 ] , restated as follows.
Lemma 2.16. Given z 0 z_0 z 0 , z 1 z_1 z 1 and z 2 z_2 z 2 in D ∖ B − 1 ( E B ) \mathbb{D}\setminus B^{-1}(\mathcal{E}_B) D ∖ B − 1 ( E B ) with B ( z 0 ) = B ( z 1 ) = B ( z 2 ) B(z_0)=B(z_1)=B(z_2) B ( z 0 ) = B ( z 1 ) = B ( z 2 ) , suppose there are three curves σ 0 , σ 1 , σ 2 \sigma_0,\sigma_1,\sigma_2 σ 0 , σ 1 , σ 2 such that
(1) σ i ( 0 ) = z i \sigma_i(0)=z_i σ i ( 0 ) = z i for i = 0 , 1 , 2 i=0,1,2 i = 0 , 1 , 2 , and σ 1 ( 1 ) = σ 2 ( 1 ) \sigma_1(1)=\sigma_2(1) σ 1 ( 1 ) = σ 2 ( 1 ) ;
(2) B ( σ 0 ( t ) ) = B ( σ 1 ( t ) ) = B ( σ 2 ( t ) ) B(\sigma_0(t))=B(\sigma_1(t))=B(\sigma_2(t)) B ( σ 0 ( t )) = B ( σ 1 ( t )) = B ( σ 2 ( t )) for 0 ≤ t ≤ 1 0 \le t \le1 0 ≤ t ≤ 1 , and σ i ( t ) ∈ D ∖ B − 1 ( E B ) \sigma_i(t) \in\mathbb{D}\setminus B^{-1}(\mathcal{E}_B) σ i ( t ) ∈ D ∖ B − 1 ( E B ) , i = 0 , 1 , 2 i=0,1,2 i = 0 , 1 , 2 , 0 ≤ t < 1 0 \le t < 1 0 ≤ t < 1 ;
(3) B ′ ( σ 0 ( 1 ) ) ≠ 0 B'(\sigma_0(1)) \ne0 B ′ ( σ 0 ( 1 )) = 0 .
Then ρ z 0 , z 1 ∼ ρ z 0 , z 2 \rho_{z_0,z_1} \sim\rho_{z_0,z_2} ρ z 0 , z 1 ∼ ρ z 0 , z 2 .
Remind that for a thin Blaschke product B B B , both E B \mathcal{E}_B E B and B − 1 ( E B ) B^{-1}(\mathcal{E}_B) B − 1 ( E B ) are discrete in D \mathbb{D} D (see [15 Proposition 2.1] ). For a thin/finite Blaschke product B B B , and for a subdomain Ω \Omega Ω of D \mathbb{D} D , we define
S B , Ω = { ( w , z ) ∈ Ω 2 : B ( z ) = B ( w ) ∉ E B } . S_{B,\Omega}=\{(w,z)\in\Omega^2:B(z)=B(w)\notin\mathcal{E}_B\}. S B , Ω = {( w , z ) ∈ Ω 2 : B ( z ) = B ( w ) ∈ / E B } .
Specifically, for Ω = D \Omega=\mathbb{D} Ω = D , the associated Riemann surface
S B , D = S B = { ( z , w ) ∈ D 2 : B ( z ) = B ( w ) ∉ E B } . S_{B,\mathbb{D}}=S_B=\{(z,w)\in\mathbb{D}^2:B(z)=B(w)\notin\mathcal{E}_B\}. S B , D = S B = {( z , w ) ∈ D 2 : B ( z ) = B ( w ) ∈ / E B } .
We also need the following. The former part is directly from Proposition 5.3.2 in [16 ] , and its later part is a consequence of Proposition 2.3.
Proposition 2.17. Let B B B be a finite Blaschke product. If S S S is a component of S B S_B S B , then S − : = { ( w , z ) ∈ D 2 : ( z , w ) ∈ S } S^{-}:=\{(w,z)\in\mathbb{D}^{2}:(z,w)\in S\} S − := {( w , z ) ∈ D 2 : ( z , w ) ∈ S } is also a component of S B S_B S B . Consequently, for each t ∈ ( 0 , 1 ) t\in(0,1) t ∈ ( 0 , 1 ) , if Ω \Omega Ω is a component of { ∣ B ∣ < t } \{|B|<t\} { ∣ B ∣ < t } and T T T is a component of S B , Ω S_{B,\Omega} S B , Ω , then T − : = { ( w , z ) ∈ Ω 2 : ( z , w ) ∈ T } T^{-}:=\{(w,z)\in\Omega^{2}:(z,w)\in T\} T − := {( w , z ) ∈ Ω 2 : ( z , w ) ∈ T } is a component of S B , Ω S_{B,\Omega} S B , Ω .
For every subdomain Ω \Omega Ω of D \mathbb{D} D , we define the identity component of
{ ( z , w ) ∈ Ω 2 : B ( z ) = B ( w ) ∉ E B } , \{(z,w)\in\Omega^{2}:B(z)=B(w)\notin\mathcal{E}_{B}\}, {( z , w ) ∈ Ω 2 : B ( z ) = B ( w ) ∈ / E B } ,
to be the set { ( z , z ) : z ∈ Ω ∖ B − 1 ( E B ) } \{(z,z):z\in\Omega\setminus B^{-1}(\mathcal{E}_{B})\} {( z , z ) : z ∈ Ω ∖ B − 1 ( E B )} .
Definition 2.19. Let B B B be a finite/thin Blaschke product, 0 < t < 1 0<t<1 0 < t < 1 and let Ω \Omega Ω be a component of { ∣ B ∣ < t } \{|B|<t\} { ∣ B ∣ < t } . If
{ ( z , w ) ∈ Ω 2 : B ( z ) = B ( w ) ∉ E B } \{(z,w)\in\Omega^2:B(z)=B(w)\notin\mathcal{E}_B\} {( z , w ) ∈ Ω 2 : B ( z ) = B ( w ) ∈ / E B } contains exactly two components, then we say Ω \Omega Ω is purely gluable with respect to B B B .
We have the following result.
Proposition 2.20. Let 0 < t < 1 0<t<1 0 < t < 1 . For a component Ω \Omega Ω of { ∣ B ∣ < t } \{|B|<t\} { ∣ B ∣ < t } , Ω \Omega Ω is purely gluable with respect to B B B if and only if excluding the identity map, every local inverse defined on a subdomain of Ω \Omega Ω can be analytically continuable to any other such non-identity local inverse along some curve in Ω \ B − 1 ( E B ) \Omega\backslash B^{-1}(\mathcal{E}_B) Ω\ B − 1 ( E B ) .
Proof. In fact, by Propositions 2.3 and 2.6, the component Ω \Omega Ω of { ∣ B ∣ < t } \{|B|<t\} { ∣ B ∣ < t } is simply connected and there exists a conformal map G : Ω → D G:\Omega\to\mathbb{D} G : Ω → D such that
B ∣ Ω = t φ ∘ G ( z ) , z ∈ Ω B|_{\Omega}=t\varphi\circ G(z),\quad z\in\Omega B ∣ Ω = tφ ∘ G ( z ) , z ∈ Ω
for some finite Blaschke product φ \varphi φ . In this case, the Riemann surface
{ ( z , w ) ∈ Ω 2 : B ( z ) = B ( w ) ∉ E B } = { ( z , w ) ∈ Ω 2 : t φ ( G ( z ) ) = t φ ( G ( w ) ) ∉ E B } \begin{aligned}
\{(z,w)\in\Omega^2:B(z)=B(w)\notin\mathcal{E}_B\} \\
&=\{(z,w)\in\Omega^2:t\varphi(G(z))=t\varphi(G(w))\notin\mathcal{E}_B\}
\end{aligned} {( z , w ) ∈ Ω 2 : B ( z ) = B ( w ) ∈ / E B } = {( z , w ) ∈ Ω 2 : tφ ( G ( z )) = tφ ( G ( w )) ∈ / E B }
is biholomorphic to
{ ( G ( z ) , G ( w ) ) ∈ D 2 : t φ ( G ( z ) ) = t φ ( G ( w ) ) ∉ E B } , \{(G(z),G(w))\in\mathbb{D}^2:t\varphi(G(z))=t\varphi(G(w))\notin\mathcal{E}_B\}, {( G ( z ) , G ( w )) ∈ D 2 : tφ ( G ( z )) = tφ ( G ( w )) ∈ / E B } ,
which simplifies to { ( z ~ , w ~ ) ∈ D 2 : φ ( z ~ ) = φ ( w ~ ) ∉ E φ } \{(\widetilde{z},\widetilde{w})\in\mathbb{D}^2:\varphi(\widetilde{z})=\varphi(\widetilde{w})\notin\mathcal{E}_{\varphi}\} {( z , w ) ∈ D 2 : φ ( z ) = φ ( w ) ∈ / E φ } . This shows that { ( z , w ) ∈ Ω 2 : B ( z ) = B ( w ) ∉ E B } \{(z,w)\in\Omega^2:B(z)=B(w)\notin\mathcal{E}_B\} {( z , w ) ∈ Ω 2 : B ( z ) = B ( w ) ∈ / E B } is biholomorphic to S φ S_{\varphi} S φ . Combining this with Remark 2.18(1), we immediately obtain the desired conclusion. □
The following proposition comes from [6 , 7 ] and [16 ] .
Proposition 2.21. For thin/finite Blaschke products, we have the following properties.
If B B B is a thin/finite Blaschke product and D \mathbb{D} D is purely gluable, then S B S_B S B has exactly two components.
If B B B is a finite Blaschke product and S B S_B S B has exactly two components, then M B M_B M B on L a 2 ( D ) L_a^2(\mathbb{D}) L a 2 ( D ) has exactly two minimal reducing subspaces.
(3) If B B B is a thin Blaschke product and S B S_B S B has exactly two components, then M B M_B M B on L a 2 ( D ) L_a^2(\mathbb{D}) L a 2 ( D ) has no nontrivial reducing subspaces.
Proof. By definition, S B S_B S B has only two components provided that D \mathbb{D} D is purely gluable. Thus (1) is straightforward.
For (2), just note that for a finite Blaschke product B B B , the number of components of S B S_B S B equals that of minimal reducing subspaces for M B M_B M B on L a 2 ( D ) L_a^2(\mathbb{D}) L a 2 ( D ) (see [6 , 7 ] ).
It remains to deal with (3). Now assume B B B is a thin Blaschke product. By Theorem 5.3.1 in [16 ] , for a thin Blaschke product B B B , M B M_B M B on L a 2 ( D ) L_a^2(\mathbb{D}) L a 2 ( D ) is irreducible if and only if V ∗ ( B ) = C I V^*(B)=\mathbb{C}I V ∗ ( B ) = C I , if and only if the Riemann surface S B S_B S B has a component of finite multiplicity, distinct from the identity component { ( z , z ) : B ( z ) ∉ E B } \{(z,z):B(z)\notin\mathcal{E}_B\} {( z , z ) : B ( z ) ∈ / E B } . When S B S_B S B has exactly two components, the non-identity component S S S must be of infinite multiplicity. Then M B M_B M B is irreducible.
Critical points of finite Blaschke products
This subsection presents properties related to the critical points of finite Blaschke products.
We say a curve γ \gamma γ is non-self-intersecting if γ \gamma γ is injective, that is, γ ( s ) ≠ γ ( t ) \gamma(s)\ne\gamma(t) γ ( s ) = γ ( t ) for 0 ≤ s < t ≤ 1 0\le s<t\le1 0 ≤ s < t ≤ 1 . Part (1) of the following theorem appears in [13 Lemma 2.8] .
Theorem 2.22. Let B B B be a finite Blaschke product of order n n n with its critical set E B \mathcal{E}_B E B . Then we have the following.
(1) If γ \gamma γ is a non-self-intersecting curve such that γ ( [ 0 , 1 ] ) ⊆ D ∖ E B \gamma([0,1])\subseteq\mathbb{D}\setminus\mathcal{E}_B γ ([ 0 , 1 ]) ⊆ D ∖ E B , then there exist exactly n n n curves σ 1 , ⋯ , σ n \sigma_1,\cdots,\sigma_n σ 1 , ⋯ , σ n satisfying:
B ( σ i ( t ) ) = γ ( t ) and σ i ( t ) ≠ σ j ( t ) for 0 ≤ t ≤ 1 ; i ≠ j . B(\sigma_i(t))=\gamma(t)\quad\text{and}\quad\sigma_i(t)\ne\sigma_j(t)\quad\text{for }0\le t\le1;\quad i\ne j. B ( σ i ( t )) = γ ( t ) and σ i ( t ) = σ j ( t ) for 0 ≤ t ≤ 1 ; i = j . (2) If γ \gamma γ is a non-self-intersecting curve such that
γ [ 0 , 1 ) ⊆ D ∖ E B and γ ( 1 ) ∈ E B , \gamma[0,1)\subseteq\mathbb{D}\setminus\mathcal{E}_B\quad\text{and}\quad\gamma(1)\in\mathcal{E}_B, γ [ 0 , 1 ) ⊆ D ∖ E B and γ ( 1 ) ∈ E B , then there exist curves σ i \sigma_i σ i (1 ≤ i ≤ n 1\le i\le n 1 ≤ i ≤ n ) satisfy B ( σ i ( t ) ) = γ ( t ) B(\sigma_i(t))=\gamma(t) B ( σ i ( t )) = γ ( t ) for 0 ≤ t ≤ 1 0\le t\le1 0 ≤ t ≤ 1 and σ i ( t ) ≠ σ j ( t ) \sigma_i(t)\ne\sigma_j(t) σ i ( t ) = σ j ( t ) for 0 ≤ t < 1 0\le t<1 0 ≤ t < 1 , i ≠ j i\ne j i = j . Furthermore, if p p p is a critical point of B B B with B ( p ) = γ ( 1 ) B(p)=\gamma(1) B ( p ) = γ ( 1 ) and of order k − 1 k-1 k − 1 , there exist exactly k k k distinct indices i 1 , ⋯ , i k i_1,\cdots,i_k i 1 , ⋯ , i k such that:
σ i j ( 1 ) = p , 1 ≤ j ≤ k . \sigma_{i_j}(1)=p,\quad1\le j\le k. σ i j ( 1 ) = p , 1 ≤ j ≤ k .
Recall that if a nonconstant holomorphic function f f f on a domain Δ \Delta Δ has a zero at a ∈ Δ a \in\Delta a ∈ Δ , then there exist a holomorphic function g g g on Δ \Delta Δ with g ( a ) ≠ 0 g(a) \ne0 g ( a ) = 0 and a positive integer m m m such that f ( z ) = ( z − a ) m g ( z ) f(z)=(z-a)^m g(z) f ( z ) = ( z − a ) m g ( z ) for z ∈ Δ z \in\Delta z ∈ Δ . The integer m m m is the order of the zero of f f f at a a a , denoted by order f ( a ) \operatorname{order}_f(a) order f ( a ) . For the proof of Theorem 2.22 , we need the following form of Böttcher’s theorem, see [22 Theorem 9.1] for example.
Theorem 2.23 (Böttcher). Let ψ \psi ψ be a holomorphic function on a neighborhood of a point p p p . If ψ ′ ( p ) = 0 \psi'(p)=0 ψ ′ ( p ) = 0 and ψ ≢ 0 \psi\not\equiv0 ψ ≡ 0 , then there exist a simply connected domain Δ \Delta Δ containing p p p , a number ε > 0 \varepsilon>0 ε > 0 , and a biholomorphic map g : Δ → ε D g:\Delta\to\varepsilon\mathbb{D} g : Δ → ε D so that
ψ ( z ) − ψ ( p ) = [ g ( z ) ] k , z ∈ Δ , \psi(z)-\psi(p)=[g(z)]^k,\qquad z\in\Delta, ψ ( z ) − ψ ( p ) = [ g ( z ) ] k , z ∈ Δ , where k = order ψ ′ ( p ) + 1 k=\operatorname{order}_{\psi'}(p)+1 k = order ψ ′ ( p ) + 1 . In this case, ψ ∣ Δ : Δ → ψ ( Δ ) \psi|_{\Delta}:\Delta\to\psi(\Delta) ψ ∣ Δ : Δ → ψ ( Δ ) is a holomorphic proper map of multiplicity k k k .
Proof of Theorem 2.22. (1) First we assume that γ \gamma γ is a non-self-intersecting curve such that γ [ 0 , 1 ] ⊆ D ‾ ∖ E B \gamma[0,1]\subseteq\overline{\mathbb{D}}\setminus\mathcal{E}_B γ [ 0 , 1 ] ⊆ D ∖ E B . For each t 0 ∈ [ 0 , 1 ] t_0\in[0,1] t 0 ∈ [ 0 , 1 ] with γ ( t 0 ) ∉ E B \gamma(t_0)\notin\mathcal{E}_B γ ( t 0 ) ∈ / E B , there exist n n n points z 1 , ⋯ , z n z_1,\cdots,z_n z 1 , ⋯ , z n such that B ( z i ) = γ ( t 0 ) B(z_i)=\gamma(t_0) B ( z i ) = γ ( t 0 ) for 1 ≤ i ≤ n 1\leq i\leq n 1 ≤ i ≤ n . Since B ′ ( z i ) ≠ 0 B'(z_i)\ne0 B ′ ( z i ) = 0 for each i i i , by the Inverse Function Theorem, there exist neighborhoods U i U_i U i of z i z_i z i such that the map B ∣ U i : U i → B ( U i ) B|_{U_i}:U_i\to B(U_i) B ∣ U i : U i → B ( U i ) is biholomorphic. On each U i U_i U i , we can define σ i \sigma_i σ i as:
σ i ( t ) = B − 1 ( γ ( t ) ) ( t ∈ [ t 0 − δ , t 0 + δ ] ) \sigma_i(t)=B^{-1}(\gamma(t))\qquad(t\in[t_0-\delta,t_0+\delta]) σ i ( t ) = B − 1 ( γ ( t )) ( t ∈ [ t 0 − δ , t 0 + δ ])
for some δ > 0 \delta>0 δ > 0 . By compactness of [ 0 , 1 ] [0,1] [ 0 , 1 ] , we can cover it with finitely many intervals Δ 1 , ⋯ , Δ N \Delta_1,\cdots,\Delta_N Δ 1 , ⋯ , Δ N of the form [ t 0 − δ , t 0 + δ ] [t_0-\delta,t_0+\delta] [ t 0 − δ , t 0 + δ ] . For each Δ j \Delta_j Δ j , there correspond n n n curves σ i , j \sigma_{i,j} σ i , j satisfying B ( σ i , j ( t ) ) = γ ( t ) B(\sigma_{i,j}(t))=\gamma(t) B ( σ i , j ( t )) = γ ( t ) for t ∈ Δ j t\in\Delta_j t ∈ Δ j with i = 1 , ⋯ , n i=1,\cdots,n i = 1 , ⋯ , n . By pasting these curves over overlapping intervals, we obtain the desired curves σ i \sigma_i σ i . It is routine to verify that the values σ i ( 0 ) \sigma_i(0) σ i ( 0 ) are distinct for i = 1 , 2 , ⋯ , n i=1,2,\cdots,n i = 1 , 2 , ⋯ , n , and by continuity, the curves σ i ( t ) \sigma_i(t) σ i ( t ) remain distinct for all t ∈ [ 0 , 1 ] t\in[0,1] t ∈ [ 0 , 1 ] .
(2) Next we assume γ \gamma γ is a non-self-intersecting curve such that
γ [ 0 , 1 ) ⊆ D ‾ ∖ E B and γ ( 1 ) ∈ E B . \gamma[0,1)\subseteq\overline{\mathbb{D}}\setminus\mathcal{E}_B\qquad\text{and}\qquad\gamma(1)\in\mathcal{E}_B. γ [ 0 , 1 ) ⊆ D ∖ E B and γ ( 1 ) ∈ E B .
For every t ∗ ∈ ( 0 , 1 ) t^*\in(0,1) t ∗ ∈ ( 0 , 1 ) , by the discussions in (1), there exist n n n curves σ i \sigma_i σ i defined on [ 0 , t ∗ ] [0,t^*] [ 0 , t ∗ ] such that B ( σ i ( s ) ) = γ ( s ) B(\sigma_i(s))=\gamma(s) B ( σ i ( s )) = γ ( s ) , 0 ≤ s ≤ t ∗ 0\leq s\leq t^* 0 ≤ s ≤ t ∗ . The idea is to find the preimages of γ ( s ) \gamma(s) γ ( s ) (t ∗ ≤ s ≤ 1 t^*\leq s\leq1 t ∗ ≤ s ≤ 1 ) when t ∗ t^* t ∗ is enough close to 1 1 1 .
Before continuing, let us make an observation. Let p p p be a critical point with B ( p ) = γ ( 1 ) B(p)=\gamma(1) B ( p ) = γ ( 1 ) and critical order k − 1 k-1 k − 1 . By Böttcher’s theorem, there exist a simply connected domain Δ \Delta Δ containing p p p , ε > 0 \varepsilon> 0 ε > 0 , and a biholomorphic map g : Δ → ε D g : \Delta\to\varepsilon\mathbb{D} g : Δ → ε D so that g ( 0 ) = 0 g(0) = 0 g ( 0 ) = 0 and
B ( z ) − B ( p ) = [ g ( z ) ] k z ∈ Δ . B(z) - B(p) = [g(z)]^k \qquad z \in\Delta. B ( z ) − B ( p ) = [ g ( z ) ] k z ∈ Δ.
For any curve γ 0 : [ 0 , 1 ] → C \gamma_0 : [0, 1] \to\mathbb{C} γ 0 : [ 0 , 1 ] → C in the ε 2 \frac{\varepsilon}{2} 2 ε -neighborhood of B ( p ) B(p) B ( p ) with γ 0 ( 1 ) = B ( p ) \gamma_0(1) = B(p) γ 0 ( 1 ) = B ( p ) , the preimage B − 1 ( γ 0 ) ∩ Δ B^{-1}(\gamma_0) \cap\Delta B − 1 ( γ 0 ) ∩ Δ consists of n n n distinct analytic arcs { Γ i } 1 ≤ i ≤ n \{\Gamma_i\}_{1 \leq i \leq n} { Γ i } 1 ≤ i ≤ n , and exactly k k k of these arcs terminate at p p p . Note that if p p p is a regular point, then the above analysis remains valid with k = 1 k = 1 k = 1 .
Let γ 0 = γ ∣ [ t ∗ , 1 ] \gamma_0 = \gamma|_{[t^*,1]} γ 0 = γ ∣ [ t ∗ , 1 ] where t ∗ t^* t ∗ is sufficiently close to 1. Then B − 1 ( γ ∣ [ 0 , t ∗ ] ) B^{-1}(\gamma|_{[0,t^*]}) B − 1 ( γ ∣ [ 0 , t ∗ ] ) splits into n n n curves, and each curve connects to one of the Γ i \Gamma_i Γ i . The conclusions in (2) hold, thereby completing the proof.
We also have a consequence of Theorem 2.22.
Corollary 2.24. Let B B B be a finite Blaschke product and fix 0 < t < 1 0 < t < 1 0 < t < 1 . Suppose Ω 1 , … , Ω N \Omega_1, \ldots, \Omega_N Ω 1 , … , Ω N are all the distinct components of { ∣ B ∣ < t } \{|B| < t\} { ∣ B ∣ < t } . For every point p ∈ ∂ Ω 1 p \in\partial\Omega_1 p ∈ ∂ Ω 1 , p p p is a critical point of B B B if and only if p p p lies in the union
⋃ 2 ≤ j ≤ N ∂ Ω 1 ∩ ∂ Ω j . \bigcup_{2 \leq j \leq N} \partial\Omega_1 \cap\partial\Omega_j. 2 ≤ j ≤ N ⋃ ∂ Ω 1 ∩ ∂ Ω j .
Figure 1. Lemma 2.25
The following lemma is included in [13 Corollary 2.9] .
Lemma 2.25. Let B B B be a finite Blaschke product with the critical set E B E_B E B , and fix 0 < t < 1 0 < t < 1 0 < t < 1 . Suppose Ω 1 , Ω 2 \Omega_1, \Omega_2 Ω 1 , Ω 2 are distinct components of { ∣ B ∣ < t } \{|B| < t\} { ∣ B ∣ < t } . For any point p ∈ ∂ Ω 1 ∩ ∂ Ω 2 p \in\partial\Omega_1 \cap\partial\Omega_2 p ∈ ∂ Ω 1 ∩ ∂ Ω 2 , let γ \gamma γ be a curve such that γ [ 0 , 1 ) ∈ t D \ E B \gamma[0, 1) \in t\mathbb{D}\backslash E_B γ [ 0 , 1 ) ∈ t D \ E B , and γ ( 1 ) = B ( p ) \gamma(1) = B(p) γ ( 1 ) = B ( p ) . Then there exists a unique pair of curves σ 1 , σ 2 \sigma_1, \sigma_2 σ 1 , σ 2 such that for 0 ≤ s < 1 0 \leq s < 1 0 ≤ s < 1 , σ i ( s ) ∈ Ω i \sigma_i(s) \in\Omega_i σ i ( s ) ∈ Ω i ,
B ( σ i ( s ) ) = γ ( s ) and σ i ( 1 ) = p , i = 1 , 2. B(\sigma_i(s)) = \gamma(s) \quad\text{and} \quad\sigma_i(1) = p,\qquad i = 1, 2. B ( σ i ( s )) = γ ( s ) and σ i ( 1 ) = p , i = 1 , 2.
Consequently, p p p is a critical point of B B B .
Furthermore, if in addition, γ \gamma γ is non-self-intersecting and there exists a curve σ \sigma σ such that B ( σ ( s ) ) = γ ( s ) B(\sigma(s))=\gamma(s) B ( σ ( s )) = γ ( s ) with σ ( s ) ∈ Ω 1 \sigma(s)\in\Omega_1 σ ( s ) ∈ Ω 1 for 0 ≤ s < 1 0\leq s<1 0 ≤ s < 1 , then either σ \sigma σ is disjoint from σ 1 \sigma_1 σ 1 , or σ 1 ( s ) = σ ( s ) \sigma_1(s)=\sigma(s) σ 1 ( s ) = σ ( s ) , 0 ≤ s ≤ 1 0\leq s\leq1 0 ≤ s ≤ 1 .
The following describes the structure of the boundaries of components of { ∣ B ∣ < t } \{|B|<t\} { ∣ B ∣ < t } for 0 < t < 1 0<t<1 0 < t < 1 . This theorem will be used in our induction step in proving Theorem 1.6 .
Theorem 2.26. Let B B B be a finite Blaschke product, and fix 0 < t < 1 0<t<1 0 < t < 1 . Denote by { Ω i } i = 1 k \{\Omega_i\}_{i=1}^{k} { Ω i } i = 1 k the components of the sublevel set { ∣ B ∣ < t } \{|B|<t\} { ∣ B ∣ < t } where k ≥ 2 k\geq2 k ≥ 2 . Then these components satisfy the following properties.
(1) Critical adjacency. For i ≠ j i\ne j i = j , the intersection ∂ Ω i ∩ ∂ Ω j \partial\Omega_i\cap\partial\Omega_j ∂ Ω i ∩ ∂ Ω j contains at most one point. Moreover, if ∂ Ω i ∩ ∂ Ω j \partial\Omega_i\cap\partial\Omega_j ∂ Ω i ∩ ∂ Ω j is nonempty, then it consists precisely of a critical point of B B B .
(2) Cyclic connectivity. Under a reordering { Ω i } 1 ≤ i ≤ k \{\Omega_i\}_{1\leq i\leq k} { Ω i } 1 ≤ i ≤ k satisfying ∂ Ω i ∩ ∂ Ω i + 1 ≠ ∅ \partial\Omega_i\cap\partial\Omega_{i+1}\ne\varnothing ∂ Ω i ∩ ∂ Ω i + 1 = ∅ for 1 ≤ i ≤ k − 1 1\leq i\leq k-1 1 ≤ i ≤ k − 1 , exactly one of the following holds:
(a) ∂ Ω k ∩ ∂ Ω 1 = ∅ \partial\Omega_k\cap\partial\Omega_1=\varnothing ∂ Ω k ∩ ∂ Ω 1 = ∅ (open chain configuration);
(b) there exists a unique point p p p in ⋂ 1 ≤ i < j ≤ k ∂ Ω i ∩ ∂ Ω j \bigcap_{1\leq i<j\leq k}\partial\Omega_i\cap\partial\Omega_j ⋂ 1 ≤ i < j ≤ k ∂ Ω i ∩ ∂ Ω j (star-shaped intersection).
Proof. By Lemma 2.25, if ∂ Ω i ∩ ∂ Ω j ≠ ∅ \partial\Omega_i\cap\partial\Omega_j\ne\varnothing ∂ Ω i ∩ ∂ Ω j = ∅ for i ≠ j i\ne j i = j , then every point in ∂ Ω i ∩ ∂ Ω j \partial\Omega_i\cap\partial\Omega_j ∂ Ω i ∩ ∂ Ω j is a critical point. We now prove that ∂ Ω 1 ∩ ∂ Ω 2 \partial\Omega_1\cap\partial\Omega_2 ∂ Ω 1 ∩ ∂ Ω 2 contains at most one point. To reach a contradiction, we assume that ∂ Ω 1 ∩ ∂ Ω 2 \partial\Omega_1\cap\partial\Omega_2 ∂ Ω 1 ∩ ∂ Ω 2 is not empty, and ∂ Ω 1 ∩ ∂ Ω 2 \partial\Omega_1\cap\partial\Omega_2 ∂ Ω 1 ∩ ∂ Ω 2 contains two distinct points p p p and q q q . By Lemma 2.5, ∂ Ω 1 \partial\Omega_1 ∂ Ω 1 is a Jordan curve homeomorphic to the unit circle, and then we may choose two adjacent critical points p p p and q q q on ∂ Ω 1 \partial\Omega_1 ∂ Ω 1 . Since both ∂ Ω 1 \partial\Omega_1 ∂ Ω 1 and ∂ Ω 2 \partial\Omega_2 ∂ Ω 2 are homeomorphic to the unit circle, there exist curves γ 1 ⊆ ∂ Ω 1 \gamma_1\subseteq\partial\Omega_1 γ 1 ⊆ ∂ Ω 1 and γ 2 ⊆ ∂ Ω 2 \gamma_2\subseteq\partial\Omega_2 γ 2 ⊆ ∂ Ω 2 , each joining p p p to q q q . We may further assume that the domain Ω \Omega Ω bounded by the closed curve γ 1 ∪ γ 2 \gamma_1\cup\gamma_2 γ 1 ∪ γ 2 is simply connected and satisfies the condition that
Ω ∩ Ω 1 = Ω ∩ Ω 2 = ∅ \Omega\cap\Omega_1=\Omega\cap\Omega_2=\varnothing Ω ∩ Ω 1 = Ω ∩ Ω 2 = ∅
(see Figure 2 ).
Figure 2. Excluded cases in Theorem 2.26
Since ∣ B ∣ = t |B|=t ∣ B ∣ = t holds both on γ 1 \gamma_1 γ 1 and γ 2 \gamma_2 γ 2 , by the maximum principle, on Ω \Omega Ω we have ∣ B ∣ < t |B|<t ∣ B ∣ < t . Note that ∂ Ω = γ 1 ∪ γ 2 \partial\Omega=\gamma_1\cup\gamma_2 ∂ Ω = γ 1 ∪ γ 2 and ∂ Ω ∩ ∂ Ω 1 \partial\Omega\cap\partial\Omega_1 ∂ Ω ∩ ∂ Ω 1 contains the curve γ 1 \gamma_1 γ 1 . Noting that Ω \Omega Ω is also a component of { ∣ B ∣ < t } \{|B|<t\} { ∣ B ∣ < t } , by Lemma 2.25 each point in ∂ Ω ∩ ∂ Ω 1 \partial\Omega\cap\partial\Omega_1 ∂ Ω ∩ ∂ Ω 1 is a critical point. In particular, each point on γ \gamma γ is a critical point. This gives that B B B has infinitely many critical points, which contradicts to Lemma 2.1. Therefore ∂ Ω 1 ∩ ∂ Ω 2 \partial\Omega_1\cap\partial\Omega_2 ∂ Ω 1 ∩ ∂ Ω 2 contains at most one point.
Figure 3. A case excluded by Cyclic connectivity (Theorem 2.26 )
Finally, we turn to the proof of (2). For this, assume that ∂ Ω i ∩ ∂ Ω i + 1 ≠ ∅ \partial\Omega_i\cap\partial\Omega_{i+1}\ne\varnothing ∂ Ω i ∩ ∂ Ω i + 1 = ∅ (1 ≤ i ≤ k − 1 1\le i\le k-1 1 ≤ i ≤ k − 1 ) and (b) fails; that is, for i ≠ j i\ne j i = j , not all of ∂ Ω i ∩ ∂ Ω j \partial\Omega_i\cap\partial\Omega_j ∂ Ω i ∩ ∂ Ω j equals a singlet { p } \{p\} { p } for a fixed point p p p . Suppose conversely that ∂ Ω k ∩ ∂ Ω 1 ≠ ∅ \partial\Omega_k\cap\partial\Omega_1\ne\varnothing ∂ Ω k ∩ ∂ Ω 1 = ∅ . Since each curve ∂ Ω i \partial\Omega_i ∂ Ω i (1 ≤ i ≤ k 1\le i\le k 1 ≤ i ≤ k ) is topologically a circle, a simple geometric observation shows that there exists a contour ℘ \wp ℘ contained in ⋃ 1 ≤ i ≤ k ∂ Ω i \bigcup_{1\le i\le k}\partial\Omega_i ⋃ 1 ≤ i ≤ k ∂ Ω i so that the interior Ω ℘ \Omega_\wp Ω ℘ of ℘ \wp ℘ has no intersection with any Ω i \Omega_i Ω i . But Ω ℘ \Omega_\wp Ω ℘ is a component of { ∣ B ∣ < t } \{|B|<t\} { ∣ B ∣ < t } . To complete the proof of Theorem 2.26 , we now show that B B B must have infinitely many critical points, which yields the desired contradiction. In fact, there exists some Ω i 0 \Omega_{i_0} Ω i 0 such that ∂ Ω P ∩ Ω i 0 \partial\Omega_{\mathcal{P}} \cap\Omega_{i_0} ∂ Ω P ∩ Ω i 0 contains a non-degenerate curve, and thus infinitely many distinct points. But by Lemma 2.25, every point in this intersection is a critical point of B B B . However, B B B has only finitely many critical points, a contradiction. This completes the proof of Theorem 2.26$.
As we will show, the following lemma, which establishes a key property for finite Blaschke products, plays a critical role in the proof of Theorem 1.5 by serving as the foundational step in our approximation argument for thin Blaschke products.
Lemma 2.27. Let B B B be a finite Blaschke product and let 0 < t < 1 0 < t < 1 0 < t < 1 . Suppose Ω 1 \Omega_1 Ω 1 and Ω 2 \Omega_2 Ω 2 are distinct components of { ∣ B ∣ < t } \{|B| < t\} { ∣ B ∣ < t } . If the sets { B ( z ) : z ∈ Ω 1 , B ′ ( z ) = 0 } \{B(z) : z \in\Omega_1, B'(z) = 0\} { B ( z ) : z ∈ Ω 1 , B ′ ( z ) = 0 } and { B ( z ) : z ∈ Ω 2 , B ′ ( z ) = 0 } \{B(z) : z \in\Omega_2, B'(z) = 0\} { B ( z ) : z ∈ Ω 2 , B ′ ( z ) = 0 } are disjoint, then for any z 1 , z 2 ∈ Ω 1 z_1,z_2 \in\Omega_1 z 1 , z 2 ∈ Ω 1 and w 0 ∈ Ω 2 w_0 \in\Omega_2 w 0 ∈ Ω 2 with
B ( z 1 ) = B ( z 2 ) = B ( w 0 ) ≠ E B , B(z_1) = B(z_2) = B(w_0) \ne\mathcal{E}_B, B ( z 1 ) = B ( z 2 ) = B ( w 0 ) = E B , we have ρ w 0 , z 1 ∼ ρ w 0 , z 2 \rho_{w_0,z_1} \sim\rho_{w_0,z_2} ρ w 0 , z 1 ∼ ρ w 0 , z 2 .
Proof. Let B B B be a finite Blaschke product. Suppose 0 < t < 1 0 < t < 1 0 < t < 1 , and let Ω 1 \Omega_1 Ω 1 and Ω 2 \Omega_2 Ω 2 be distinct components of { ∣ B ∣ < t } \{|B| < t\} { ∣ B ∣ < t } .
By Proposition 2.3 , both Ω 1 \Omega_1 Ω 1 and Ω 2 \Omega_2 Ω 2 are simply connected domains. Assume z 1 , z 2 ∈ Ω 1 z_1,z_2 \in\Omega_1 z 1 , z 2 ∈ Ω 1 and w 0 ∈ Ω 2 w_0 \in\Omega_2 w 0 ∈ Ω 2 such that
B ( z 1 ) = B ( z 2 ) = B ( w 0 ) ≠ E B . B(z_1) = B(z_2) = B(w_0) \ne\mathcal{E}_B. B ( z 1 ) = B ( z 2 ) = B ( w 0 ) = E B .
There are two cases to distinguish.
Case I. z 1 z_1 z 1 and z 2 z_2 z 2 are directly B B B -glued in Ω 1 \Omega_1 Ω 1 . In this case, we will prove that ρ w 0 , z 1 ∼ ρ w 0 , z 2 \rho_{w_0,z_1} \sim\rho_{w_0,z_2} ρ w 0 , z 1 ∼ ρ w 0 , z 2 .
By definition of directly B B B -gluedness, we have
σ i ( 0 ) = z i \sigma_i(0) = z_i σ i ( 0 ) = z i for i = 1 , 2 i = 1, 2 i = 1 , 2 , and σ 1 ( 1 ) = σ 2 ( 1 ) \sigma_1(1) = \sigma_2(1) σ 1 ( 1 ) = σ 2 ( 1 ) ;
B ( σ 1 ( s ) ) = B ( σ 2 ( s ) ) B(\sigma_1(s)) = B(\sigma_2(s)) B ( σ 1 ( s )) = B ( σ 2 ( s )) for 0 ≤ s ≤ 1 0 \le s \le1 0 ≤ s ≤ 1 ;
σ i ( s ) ∈ Ω 1 ∖ B − 1 ( E B ) \sigma_i(s) \in\Omega_1 \setminus B^{-1}(\mathcal{E}_B) σ i ( s ) ∈ Ω 1 ∖ B − 1 ( E B ) , i = 1 , 2 i = 1, 2 i = 1 , 2 for 0 ≤ s < 1 0 \le s < 1 0 ≤ s < 1 .
By Lemma 2.2 , B ∣ Ω 2 : Ω 2 → t D B|_{\Omega_2} : \Omega_2 \to t\mathbb{D} B ∣ Ω 2 : Ω 2 → t D is a proper map onto t D t\mathbb{D} t D , and B ( Ω 2 ) = B ( Ω 1 ) = t D B(\Omega_2) = B(\Omega_1) = t\mathbb{D} B ( Ω 2 ) = B ( Ω 1 ) = t D . Therefore, the image of B ∣ Ω 2 B|_{\Omega_2} B ∣ Ω 2 covers the image of B B B on the curve σ 1 \sigma_1 σ 1 . By Lemma 2.25, There exists a curve σ 0 \sigma_0 σ 0 satisfying $\sigma_0(0) = w_0 and σ 0 ( s ) ∈ Ω 2 \sigma_0(s) \in\Omega_2 σ 0 ( s ) ∈ Ω 2 for 0 ≤ s ≤ 1 0 \le s \le1 0 ≤ s ≤ 1 , with B ( σ 0 ( s ) ) = B ( σ 1 ( s ) ) = B ( σ 2 ( s ) ) B(\sigma_0(s)) = B(\sigma_1(s)) = B(\sigma_2(s)) B ( σ 0 ( s )) = B ( σ 1 ( s )) = B ( σ 2 ( s )) for 0 ≤ s ≤ 1 0 \le s \le1 0 ≤ s ≤ 1 . Note that by the above Condition (2) and the condition σ 1 ( 1 ) = σ 2 ( 1 ) \sigma_1(1) = \sigma_2(1) σ 1 ( 1 ) = σ 2 ( 1 ) in (1), σ 1 ( 1 ) \sigma_1(1) σ 1 ( 1 ) is a critical point. That is, B ′ ( σ 1 ( 1 ) ) = 0 B'(\sigma_1(1)) = 0 B ′ ( σ 1 ( 1 )) = 0 . By assumption, the critical value sets
{ B ( z ) : z ∈ Ω 1 , B ′ ( z ) = 0 } \{B(z): z \in\Omega_1, B'(z) = 0\} { B ( z ) : z ∈ Ω 1 , B ′ ( z ) = 0 }
and { B ( z ) : z ∈ Ω 2 , B ′ ( z ) = 0 } \{B(z): z \in\Omega_2, B'(z) = 0\} { B ( z ) : z ∈ Ω 2 , B ′ ( z ) = 0 } are disjoint, and B ( σ 0 ( 1 ) ) = B ( σ 1 ( 1 ) ) B(\sigma_0(1)) = B(\sigma_1(1)) B ( σ 0 ( 1 )) = B ( σ 1 ( 1 )) . This implies B ′ ( σ 0 ( 1 ) ) ≠ 0 B'(\sigma_0(1)) \ne0 B ′ ( σ 0 ( 1 )) = 0 . Thus the conditions in Lemma 2.16 are satisfied to force ρ z 0 , z 1 ∼ ρ z 0 , z 2 \rho_{z_0,z_1} \sim\rho_{z_0,z_2} ρ z 0 , z 1 ∼ ρ z 0 , z 2 .
Case II. General case.
In general, by Proposition 2.3, Ω 1 \Omega_1 Ω 1 is simply connected and it follows from Lemma 2.2 that the map B : Ω 1 → t D B : \Omega_1 \to t\mathbb{D} B : Ω 1 → t D is a proper map. A further application of Proposition 2.3 shows that after composing two conformal maps, B ∣ Ω 1 B|_{\Omega_1} B ∣ Ω 1 becomes a finite Blaschke product. Furthermore, by Lemma 2.13, B ∣ Ω 1 B|_{\Omega_1} B ∣ Ω 1 is gluable. In particular, z 1 z_1 z 1 and z 2 z_2 z 2 are B B B -glued on Ω 1 \Omega_1 Ω 1 . This means that there is a finite sequence z 1 = λ 1 , λ 2 , ⋯ , λ N = z 2 z_1 = \lambda_1, \lambda_2, \cdots, \lambda_N = z_2 z 1 = λ 1 , λ 2 , ⋯ , λ N = z 2 in Ω 1 \Omega_1 Ω 1 such that λ i \lambda_i λ i and λ i + 1 \lambda_{i+1} λ i + 1 are directly B B B -glued in Ω 1 \Omega_1 Ω 1 for each i i i (1 ≤ i ≤ N 1 \le i \le N 1 ≤ i ≤ N ). Then by the conclusion in Case I, we have ρ w 0 , λ i ∼ ρ w 0 , λ i + 1 \rho_{w_0,\lambda_i} \sim\rho_{w_0,\lambda_{i+1}} ρ w 0 , λ i ∼ ρ w 0 , λ i + 1 for 1 ≤ i ≤ N − 1 1 \le i \le N - 1 1 ≤ i ≤ N − 1 , which gives ρ w 0 , λ 1 ∼ ρ w 0 , λ N \rho_{w_0,\lambda_1} \sim\rho_{w_0,\lambda_N} ρ w 0 , λ 1 ∼ ρ w 0 , λ N . That is, ρ w 0 , z 1 ∼ ρ w 0 , z 2 \rho_{w_0,z_1} \sim\rho_{w_0,z_2} ρ w 0 , z 1 ∼ ρ w 0 , z 2 , as desired.
The proof of Lemma 2.28 is complete.
As a special case of Lemma 2.27, we get the following consequence.
Corollary 2.28. Let B B B be a finite Blaschke product that is injective on its critical value set E B E_B E B . For 0 < t < 1 0 < t < 1 0 < t < 1 , suppose Ω 1 \Omega_1 Ω 1 and Ω 2 \Omega_2 Ω 2 are distinct components of { ∣ B ∣ < t } \{|B| < t\} { ∣ B ∣ < t } . Then for any z 1 , z 2 ∈ Ω 1 z_1,z_2 \in\Omega_1 z 1 , z 2 ∈ Ω 1 and w 0 ∈ Ω 2 w_0 \in\Omega_2 w 0 ∈ Ω 2 such that
B ( z 1 ) = B ( z 2 ) = B ( w 0 ) ∉ E B , B(z_1) = B(z_2) = B(w_0) \notin E_B, B ( z 1 ) = B ( z 2 ) = B ( w 0 ) ∈ / E B , we have ρ w 0 , z 1 ∼ ρ w 0 , z 2 \rho_{w_0,z_1} \sim\rho_{w_0,z_2} ρ w 0 , z 1 ∼ ρ w 0 , z 2 .
Proof of our main result
Proof of Theorem 1.6
For the reader’s convenience, we restate Theorem 1.6 as below.
Theorem 3.1. Suppose B B B is a finite Blaschke product with at least two distinct critical points and B ∣ Z ( B ′ ) B|_{Z(B')} B ∣ Z ( B ′ ) is injective. Then S B S_B S B has exactly two components. In this case, the multiplication operator M B M_B M B on L a 2 ( D ) L_a^2(\mathbb{D}) L a 2 ( D ) has exactly two minimal reducing subspaces.
Some observations
For 0 < s < 1 0 < s < 1 0 < s < 1 , let n s n_s n s be the number of components of { z ∈ D : ∣ B ( z ) ∣ < s } \{z \in\mathbb{D} : |B(z)| < s\} { z ∈ D : ∣ B ( z ) ∣ < s } , and let N s N_s N s be the number of components of { z ∈ D : ∣ B ( z ) ∣ ≤ s } \{z \in\mathbb{D} : |B(z)| \le s\} { z ∈ D : ∣ B ( z ) ∣ ≤ s } . Using Proposition 2.5, it is direct to see that N s N_s N s also equals the number of components of { z ∈ D : ∣ B ( z ) ∣ = s } \{z \in\mathbb{D} : |B(z)| = s\} { z ∈ D : ∣ B ( z ) ∣ = s } .
by (1) , { ∣ B ∣ = t } \{|B| = t\} { ∣ B ∣ = t } contains at least one critical value. At least two of the boundaries ∂ Ω 0 , t , ⋯ , ∂ Ω k , t \partial\Omega_{0,t}, \cdots, \partial\Omega_{k,t} ∂ Ω 0 , t , ⋯ , ∂ Ω k , t share a common point on { ∣ B ∣ = t } \{|B| = t\} { ∣ B ∣ = t } . Let W = ⋃ i = 0 k Ω i , t W = \bigcup_{i=0}^{k}\Omega_{i,t} W = ⋃ i = 0 k Ω i , t , and set N t = l N_t = l N t = l , meaning that W W W has exactly l l l connected components V 1 , ⋯ , V l V_1,\cdots,V_l V 1 , ⋯ , V l , with each V j V_j V j containing at least one Ω i , t \Omega_{i,t} Ω i , t . We have l < k + 1 = n t l < k+1 = n_t l < k + 1 = n t . Since n s n_s n s is a decreasing integer-valued function, it has only finitely many discontinuities. Hence there exists δ > 0 \delta> 0 δ > 0 such that for any s ′ ∈ ( t , t + δ ) s' \in(t,t+\delta) s ′ ∈ ( t , t + δ ) , n s ′ ≡ N n_{s'} \equiv N n s ′ ≡ N for some integer N N N . For such s ′ s' s ′ , let Ω 0 , s ′ , ⋯ , Ω N − 1 , s ′ \Omega_{0,s'},\cdots,\Omega_{N-1,s'} Ω 0 , s ′ , ⋯ , Ω N − 1 , s ′ be all components of { ∣ B ∣ < s ′ } \{|B| < s'\} { ∣ B ∣ < s ′ } .
Since every Ω i , t \Omega_{i,t} Ω i , t is contained in some Ω j , s ′ \Omega_{j,s'} Ω j , s ′ , and every Ω j , s ′ \Omega_{j,s'} Ω j , s ′ contains at least one Ω i , t \Omega_{i,t} Ω i , t , it is easy to see l ≥ N l \geq N l ≥ N . Besides, when s ′ s' s ′ is greater than and sufficiently close to t t t , every component Ω j , s ′ \Omega_{j,s'} Ω j , s ′ of { ∣ B ∣ < s ′ } \{|B| < s'\} { ∣ B ∣ < s ′ } contains at most one component of W = ⋃ i = 0 k Ω i , s ′ W = \bigcup_{i=0}^{k}\Omega_{i,s'} W = ⋃ i = 0 k Ω i , s ′ . Therefore, N ≥ l N \geq l N ≥ l , forcing N = l N = l N = l . Thus, we have N t = N = lim s → t + n s = l < k + 1 = n t N_t = N = \lim_{s\to t+} n_s = l < k+1 = n_t N t = N = lim s → t + n s = l < k + 1 = n t . This proves the claim.
Proof of Theorem 1.6
Throughout this subsection, let B B B denote a fixed finite Blaschke product. For each s ∈ ( 0 , 1 ) s \in(0,1) s ∈ ( 0 , 1 ) , let { Ω j , s } j = 1 n s \{\Omega_{j,s}\}_{j=1}^{n_s} { Ω j , s } j = 1 n s be the finitely many components of the domain { z ∈ D : ∣ B ( z ) ∣ < s } \{z \in\mathbb{D}: |B(z)| < s\} { z ∈ D : ∣ B ( z ) ∣ < s } (note that the number n s n_s n s of these components decreases in s s s ). Then
⋃ j = 1 n s ∂ Ω j , s = { z ∈ D : ∣ B ( z ) ∣ = s } . \bigcup_{j=1}^{n_s}\partial\Omega_{j,s} = \{z \in\mathbb{D}: |B(z)| = s\}. j = 1 ⋃ n s ∂ Ω j , s = { z ∈ D : ∣ B ( z ) ∣ = s } .
For this fixed B B B , we may assume without loss of generality that one of its critical value is zero. Otherwise, let a a a be a critical value of B B B , and we replace B B B by φ a ( B ) \varphi_a(B) φ a ( B ) , where φ a ( z ) = a − z 1 − a ˉ z \varphi_a(z) = \frac{a-z}{1-\bar{a}z} φ a ( z ) = 1 − a ˉ z a − z , z ∈ D z \in\mathbb{D} z ∈ D . Let
Λ = { ∣ B ( z ) ∣ : B ′ ( z ) = 0 , z ∈ D } \Lambda= \{|B(z)|: B'(z) = 0, z \in\mathbb{D}\} Λ = { ∣ B ( z ) ∣ : B ′ ( z ) = 0 , z ∈ D }
denote the finite set of absolute critical values. Then Λ \Lambda Λ can be ordered as 0 = t 0 < t 1 < ⋯ < t N 0 = t_0 < t_1 < \cdots< t_N 0 = t 0 < t 1 < ⋯ < t N . Consider the intervals
( t 0 , t 1 ] , ( t 1 , t 2 ] , ⋯ , ( t N − 1 , t N ] , ( t N , t N + 1 ] = ( t N , 1 ] , (t_0,t_1], (t_1,t_2], \cdots, (t_{N-1},t_N], (t_N,t_{N+1}] = (t_N,1], ( t 0 , t 1 ] , ( t 1 , t 2 ] , ⋯ , ( t N − 1 , t N ] , ( t N , t N + 1 ] = ( t N , 1 ] ,
where we define t N + 1 = 1 t_{N+1} = 1 t N + 1 = 1 .
Proposition 3.3. With the above notations, let 0 ≤ i ≤ N 0 \leq i \leq N 0 ≤ i ≤ N . For s ′ < s ′ ′ s' < s'' s ′ < s ′′ and s ′ , s ′ ′ ∈ ( t i , t i + 1 ] s',s'' \in(t_i,t_{i+1}] s ′ , s ′′ ∈ ( t i , t i + 1 ] , if Ω 0 , s ′ \Omega_{0,s'} Ω 0 , s ′ and Ω 0 , s ′ ′ \Omega_{0,s''} Ω 0 , s ′′ are two components of { ∣ B ∣ < s ′ } \{|B| < s'\} { ∣ B ∣ < s ′ } and
{ ∣ B ∣ < s ′ ′ } \{|B| < s^{\prime\prime}\} { ∣ B ∣ < s ′′ } respectively, and Ω 0 , s ′ ⊆ Ω 0 , s ′ ′ \Omega_{0,s^{\prime}} \subseteq\Omega_{0,s^{\prime\prime}} Ω 0 , s ′ ⊆ Ω 0 , s ′′ , then these two components share the same multiplicity. In this case, if Ω 0 , s ′ \Omega_{0,s^{\prime}} Ω 0 , s ′ is purely gluable, then Ω 0 , s ′ ′ \Omega_{0,s^{\prime\prime}} Ω 0 , s ′′ is also purely gluable.
Proof. We begin with the following claim.
Claim. For 0 < s ′ < s ′ ′ < 1 0 < s^{\prime} < s^{\prime\prime} < 1 0 < s ′ < s ′′ < 1 , suppose Ω 0 , s ′ \Omega_{0,s^{\prime}} Ω 0 , s ′ and Ω 0 , s ′ ′ \Omega_{0,s^{\prime\prime}} Ω 0 , s ′′ share the same multiplicity and Ω 0 , s ′ ⊆ Ω 0 , s ′ ′ \Omega_{0,s^{\prime}} \subseteq\Omega_{0,s^{\prime\prime}} Ω 0 , s ′ ⊆ Ω 0 , s ′′ . If Ω 0 , s ′ \Omega_{0,s^{\prime}} Ω 0 , s ′ is purely gluable, then Ω 0 , s ′ ′ \Omega_{0,s^{\prime\prime}} Ω 0 , s ′′ is also purely gluable.
This property will be used when both s ′ s^{\prime} s ′ and s ′ ′ s^{\prime\prime} s ′′ lie within ( t i , t i + 1 ] (t_i,t_{i+1}] ( t i , t i + 1 ] for 0 ≤ i ≤ N 0 \leq i \leq N 0 ≤ i ≤ N .
To see this, let ϕ = B ∣ Ω 0 , s ′ \phi= B|_{\Omega_{0,s^{\prime}}} ϕ = B ∣ Ω 0 , s ′ and ψ = B ∣ Ω 0 , s ′ ′ \psi= B|_{\Omega_{0,s^{\prime\prime}}} ψ = B ∣ Ω 0 , s ′′ . Since ϕ \phi ϕ is essentially a finite Blaschke product, by Propositions 2.3 and 2.13 we have that ϕ \phi ϕ is purely gluable on Ω 0 , s ′ \Omega_{0,s^{\prime}} Ω 0 , s ′ . The set S ϕ = { ( z , w ) ∈ Ω 0 , s ′ 2 : B ( z ) = B ( w ) ∉ E B } S_{\phi} = \{(z,w) \in\Omega_{0,s^{\prime}}^2 : B(z) = B(w) \notin E_B\} S ϕ = {( z , w ) ∈ Ω 0 , s ′ 2 : B ( z ) = B ( w ) ∈ / E B } consists of exactly two components. In the language of local inverses, if z 1 , ⋯ , z k z_1,\cdots,z_k z 1 , ⋯ , z k map to the same point under B B B and k k k denotes the multiplicity of Ω 0 , s ′ \Omega_{0,s^{\prime}} Ω 0 , s ′ , then we have
ρ z 1 , z 2 ∼ ρ z 1 , z i , 2 ≤ i ≤ k . \rho_{z_1,z_2} \sim\rho_{z_1,z_i}, \qquad2 \leq i \leq k. ρ z 1 , z 2 ∼ ρ z 1 , z i , 2 ≤ i ≤ k .
This means the local inverse ρ z 1 , z 2 \rho_{z_1,z_2} ρ z 1 , z 2 can be analytically extended to ρ z 1 , z i \rho_{z_1,z_i} ρ z 1 , z i along a curve in Ω 0 , s ′ \Omega_{0,s^{\prime}} Ω 0 , s ′ (and consequently in Ω 0 , s ′ ′ \Omega_{0,s^{\prime\prime}} Ω 0 , s ′′ ). Given that k k k also matches the multiplicity of Ω 0 , s ′ ′ \Omega_{0,s^{\prime\prime}} Ω 0 , s ′′ , the points z 1 , ⋯ , z k z_1,\cdots,z_k z 1 , ⋯ , z k are the only points in B − 1 ( B ( z 1 ) ) ∩ Ω 0 , s ′ ′ B^{-1}(B(z_1)) \cap\Omega_{0,s^{\prime\prime}} B − 1 ( B ( z 1 )) ∩ Ω 0 , s ′′ . We thereby conclude that
S ψ = { ( z , w ) ∈ Ω 0 , s ′ ′ 2 : B ( z ) = B ( w ) ∉ E B } S_{\psi} = \{(z,w) \in\Omega_{0,s^{\prime\prime}}^2 : B(z) = B(w) \notin E_B\} S ψ = {( z , w ) ∈ Ω 0 , s ′′ 2 : B ( z ) = B ( w ) ∈ / E B }
consists of exactly two components, establishing that B B B remains purely gluable on Ω 0 , s ′ ′ \Omega_{0,s^{\prime\prime}} Ω 0 , s ′′ . The proof of the claim is complete.
By Remark 3.2 (1), for all s ∈ ( t i , t i + 1 ] s \in(t_i,t_{i+1}] s ∈ ( t i , t i + 1 ] , the quantity n s n_s n s is equal to a fixed positive integer K K K . By left-continuity, this extends to n s = K n_s = K n s = K for all s ∈ ( t i , t i + 1 ] s \in(t_i,t_{i+1}] s ∈ ( t i , t i + 1 ] . For any two values s ′ , s ′ ′ s^{\prime},s^{\prime\prime} s ′ , s ′′ with t i < s ′ < s ′ ′ ≤ t i + 1 t_i < s^{\prime} < s^{\prime\prime} \leq t_{i+1} t i < s ′ < s ′′ ≤ t i + 1 , the set { ∣ B ∣ < s ′ } \{|B| < s^{\prime}\} { ∣ B ∣ < s ′ } admits exactly K K K connected components, Ω 0 , s ′ , Ω 1 , s ′ , ⋯ , Ω K − 1 , s ′ \Omega_{0,s^{\prime}}, \Omega_{1,s^{\prime}}, \cdots, \Omega_{K-1,s^{\prime}} Ω 0 , s ′ , Ω 1 , s ′ , ⋯ , Ω K − 1 , s ′ . Similarly, the set { ∣ B ∣ < s ′ ′ } \{|B| < s^{\prime\prime}\} { ∣ B ∣ < s ′′ } also contains exactly K K K components, which we label as Ω 0 , s ′ ′ , Ω 1 , s ′ ′ , ⋯ , Ω K − 1 , s ′ ′ \Omega_{0,s^{\prime\prime}}, \Omega_{1,s^{\prime\prime}}, \cdots, \Omega_{K-1,s^{\prime\prime}} Ω 0 , s ′′ , Ω 1 , s ′′ , ⋯ , Ω K − 1 , s ′′ . Since every component of { ∣ B ∣ < s ′ ′ } \{|B| < s^{\prime\prime}\} { ∣ B ∣ < s ′′ } contains at least a component of { ∣ B ∣ < s ′ } \{|B| < s^{\prime}\} { ∣ B ∣ < s ′ } , and every component of { ∣ B ∣ < s ′ } \{|B| < s^{\prime}\} { ∣ B ∣ < s ′ } is contained in exactly one component of { ∣ B ∣ < s ′ ′ } \{|B| < s^{\prime\prime}\} { ∣ B ∣ < s ′′ } , there exists a permutation of the indices such that the inclusion relation
Ω i , s ′ ⊆ Ω i , s ′ ′ , 0 ≤ i ≤ K − 1 (6) \Omega_{i,s^{\prime}} \subseteq\Omega_{i,s^{\prime\prime}}, \qquad0 \leq i \leq K-1 \tag*{(6)} Ω i , s ′ ⊆ Ω i , s ′′ , 0 ≤ i ≤ K − 1 (6)
holds. Furthermore, the order of the Blaschke product B B B equals the sum of the multiplicities of the components { Ω i , s ′ } 0 ≤ i ≤ K − 1 \{\Omega_{i,s^{\prime}}\}_{0\leq i\leq K-1} { Ω i , s ′ } 0 ≤ i ≤ K − 1 , and also coincides with the sum of the multiplicities of the components { Ω i , s ′ ′ } 0 ≤ i ≤ K − 1 \{\Omega_{i,s''}\}_{0\le i\le K-1} { Ω i , s ′′ } 0 ≤ i ≤ K − 1 . This, together with (3.1), forces the multiplicity of Ω i , s ′ \Omega_{i,s'} Ω i , s ′ to be identical to that of Ω i , s ′ ′ \Omega_{i,s''} Ω i , s ′′ for every 0 ≤ i ≤ K − 1 0\le i\le K-1 0 ≤ i ≤ K − 1 , and in particular for i = 0 i=0 i = 0 . Combining this result with the previously established claim and the fact that Ω 0 , s ′ \Omega_{0,s'} Ω 0 , s ′ is purely gluable, we immediately conclude that Ω 0 , s ′ ′ \Omega_{0,s''} Ω 0 , s ′′ is also purely gluable.
Outline of the proof of Theorem 1.6. Let B B B be a finite Blaschke product with at least two distinct critical points, such that B ∣ Z ( B ′ ) B|_{Z(B')} B ∣ Z ( B ′ ) is injective. We prove that for any 0 < s ≤ 1 0<s\le1 0 < s ≤ 1 , every component of { ∣ B ∣ < s } \{|B|<s\} { ∣ B ∣ < s } containing at least two critical points is purely gluable.
We proceed by induction on the intervals ( t j , t j + 1 ] (t_j,t_{j+1}] ( t j , t j + 1 ] :
Base case: The claim holds for s ∈ ( t 0 , t 1 ] s\in(t_0,t_1] s ∈ ( t 0 , t 1 ] .
Inductive step: Assume the claim holds for s ∈ ( t k , t k + 1 ] s\in(t_k,t_{k+1}] s ∈ ( t k , t k + 1 ] ; we extend it to s ∈ ( t k + 1 , t k + 2 ] s\in(t_{k+1},t_{k+2}] s ∈ ( t k + 1 , t k + 2 ] in two parts:
First, we show there exists ε > 0 \varepsilon>0 ε > 0 such that the claim holds for all s ∈ ( t k + 1 , t k + 1 + ε ) s\in(t_{k+1},t_{k+1}+\varepsilon) s ∈ ( t k + 1 , t k + 1 + ε ) (the core technical step of the proof).
Second, we extend the result to the full interval ( t k + 1 , t k + 2 ] (t_{k+1},t_{k+2}] ( t k + 1 , t k + 2 ] . By Remark 3.2(1), we have n s ′ = n s ′ ′ n_{s'}=n_{s''} n s ′ = n s ′′ for any s ′ , s ′ ′ ∈ ( t k + 1 , t k + 2 ) s',s''\in(t_{k+1},t_{k+2}) s ′ , s ′′ ∈ ( t k + 1 , t k + 2 ) . Furthermore, due to the left continuity of n s n_s n s , this equality extends to s ′ , s ′ ′ ∈ ( t k + 1 , t k + 2 ] s',s''\in(t_{k+1},t_{k+2}] s ′ , s ′′ ∈ ( t k + 1 , t k + 2 ] . Observe that for such s ′ , s ′ ′ s',s'' s ′ , s ′′ with s ′ < s ′ ′ s'<s'' s ′ < s ′′ , every component of { ∣ B ∣ < s ′ } \{|B|<s'\} { ∣ B ∣ < s ′ } is contained in a unique component of { ∣ B ∣ < s ′ ′ } \{|B|<s''\} { ∣ B ∣ < s ′′ } , and conversely, every component of { ∣ B ∣ < s ′ ′ } \{|B|<s''\} { ∣ B ∣ < s ′′ } contains exactly one component of { ∣ B ∣ < s ′ } \{|B|<s'\} { ∣ B ∣ < s ′ } . This implies that the components of { ∣ B ∣ < s } \{|B|<s\} { ∣ B ∣ < s } are nested as s s s increases over ( t k + 1 , t k + 2 ] (t_{k+1},t_{k+2}] ( t k + 1 , t k + 2 ] . Consequently, by Proposition 3.3, for each s ∈ ( t k + 1 , t k + 2 ] s\in(t_{k+1},t_{k+2}] s ∈ ( t k + 1 , t k + 2 ] , every component of { ∣ B ∣ < s } \{|B|<s\} { ∣ B ∣ < s } containing at least two distinct critical points is purely gluable.
After finitely many steps, the claim holds for s ∈ ( t N , t N + 1 ] s\in(t_N,t_{N+1}] s ∈ ( t N , t N + 1 ] . Since t N + 1 = 1 t_{N+1}=1 t N + 1 = 1 , the only component of { ∣ B ∣ < 1 } \{|B|<1\} { ∣ B ∣ < 1 } is D \mathbb{D} D , so D \mathbb{D} D is purely gluable. By definition, S B S_B S B has exactly two components, which completes the proof.
To complete the proof of Theorem 1.6, it therefore suffices to establish the inductive claim for s ∈ ( t k + 1 , t k + 1 + ε ) s\in(t_{k+1},t_{k+1}+\varepsilon) s ∈ ( t k + 1 , t k + 1 + ε ) , rather than verifying it over the entire interval.
Proof of Theorem 1.6. Let B B B be a finite Blaschke product with at least two distinct critical points, and suppose that B ∣ Z ( B ′ ) B|_{Z(B')} B ∣ Z ( B ′ ) is injective. By Proposition 2.21, the proof is complete once we establish that for 0 < s ≤ 1 0<s\le1 0 < s ≤ 1 , each component V V V of { ∣ B ∣ < s } \{|B|<s\} { ∣ B ∣ < s } that contains at least two distinct critical points is purely gluable; that is,
{ ( z , w ) ∈ V 2 : B ( z ) = B ( w ) ∉ E B } \{(z,w)\in V^2:B(z)=B(w)\notin E_B\} {( z , w ) ∈ V 2 : B ( z ) = B ( w ) ∈ / E B }
contains exactly two components, the identity component S 0 \mathcal{S}_0 S 0 and its complement. When this is done, for s = 1 s=1 s = 1 , we have D = { ∣ B ∣ < 1 } \mathbb{D}=\{|B|<1\} D = { ∣ B ∣ < 1 } , and then
{ ( z , w ) ∈ D 2 : B ( z ) = B ( w ) ∉ E B } \{(z,w)\in\mathbb{D}^2:B(z)=B(w)\notin E_B\} {( z , w ) ∈ D 2 : B ( z ) = B ( w ) ∈ / E B }
contains exactly two components. That is, S B S_B S B has exactly two components. By Proposition 2.21 , M B M_B M B on L a 2 ( D ) L_a^2(\mathbb{D}) L a 2 ( D ) has precisely two minimal reducing subspaces.
We proceed via induction to verify this statement. Note that t 0 = 0 t_0=0 t 0 = 0 , t N + 1 = 1 t_{N+1}=1 t N + 1 = 1 , and ( 0 , 1 ] = ⋃ j = 0 N ( t j , t j + 1 ] (0,1]=\bigcup_{j=0}^{N}(t_j,t_{j+1}] ( 0 , 1 ] = ⋃ j = 0 N ( t j , t j + 1 ] . For s ∈ ( t 0 , t 1 ] = ( 0 , t 1 ] s\in(t_0,t_1]=(0,t_1] s ∈ ( t 0 , t 1 ] = ( 0 , t 1 ] , we will shortly prove that each component of { ∣ B ∣ < s } \{|B|<s\} { ∣ B ∣ < s } has at most one critical point. Under this condition, we will soon see that the desired statement immediately holds for s ∈ ( t 0 , t 1 ] s\in(t_0,t_1] s ∈ ( t 0 , t 1 ] . In fact, let λ 1 , ⋯ , λ N \lambda_1,\cdots,\lambda_N λ 1 , ⋯ , λ N be the zeros of B B B . There exists a sufficiently small δ > 0 \delta>0 δ > 0 such that each component of { ∣ B ∣ < δ } \{|B|<\delta\} { ∣ B ∣ < δ } lies in an ε \varepsilon ε -neighborhood of some λ i \lambda_i λ i for 1 ≤ i ≤ N 1\leq i\leq N 1 ≤ i ≤ N . By the discreteness of the critical set of B B B , each such domain contains at most one possible critical point, whose only candidate is the zero λ i \lambda_i λ i . When δ \delta δ is replaced by a number s ′ ∈ ( t 0 , t 1 ] s'\in(t_0,t_1] s ′ ∈ ( t 0 , t 1 ] , note that the components are nested as s ′ s' s ′ increases.
To finish the proof of this case, we claim that each component of { ∣ B ∣ < s ′ } \{|B|<s'\} { ∣ B ∣ < s ′ } contains at most one critical point. The desired conclusion then follows. Suppose, to reach a contradiction, that there exist some s ′ ∈ ( 0 , t 1 ] s'\in(0,t_1] s ′ ∈ ( 0 , t 1 ] and a component Ω \Omega Ω containing at least two critical points. Since every critical point p p p in Ω \Omega Ω satisfies
∣ B ( p ) ∣ < s ′ ≤ t 1 , |B(p)|<s'\leq t_1, ∣ B ( p ) ∣ < s ′ ≤ t 1 ,
it follows from the definition of t i t_i t i that B ( p ) = 0 B(p)=0 B ( p ) = 0 . But as the components of { ∣ B ∣ < t ′ } \{|B|<t'\} { ∣ B ∣ < t ′ } are nested as t ′ t' t ′ increases, each such component can contain at most one critical point. This contradiction completes the proof of this case.
Assume each component of { ∣ B ∣ < s } \{|B|<s\} { ∣ B ∣ < s } that contains at least two distinct critical points is purely gluable for s ∈ ( t j , t j + 1 ] s\in(t_j,t_{j+1}] s ∈ ( t j , t j + 1 ] with 0 ≤ j ≤ k 0\leq j\leq k 0 ≤ j ≤ k . We aim to prove this holds for s ∈ ( t k + 1 , t k + 2 ] s\in(t_{k+1},t_{k+2}] s ∈ ( t k + 1 , t k + 2 ] . By the Proposition 3.3 , it suffices to find ε ∈ ( 0 , t k + 2 − t k + 1 ) \varepsilon\in(0,t_{k+2}-t_{k+1}) ε ∈ ( 0 , t k + 2 − t k + 1 ) such that each component of { ∣ B ∣ < s } \{|B|<s\} { ∣ B ∣ < s } containing at least two distinct critical points is purely gluable for any s ∈ ( t k + 1 , t k + 1 + ε ) s\in(t_{k+1},t_{k+1}+\varepsilon) s ∈ ( t k + 1 , t k + 1 + ε ) .
Notation. t t t and t + t^+ t + . In the subsequent induction step, we rewrite t = t j t=t_j t = t j for each index j j j . Note that t j t_j t j denotes the time t t t at which at least two boundaries ∂ Ω i , t \partial\Omega_{i,t} ∂ Ω i , t intersect (see Remark 3.2 (1)).
Let us see what happens to the components of { ∣ B ∣ ≤ s } \{|B| \le s\} { ∣ B ∣ ≤ s } when s = t s=t s = t is replaced by s = t + ε s=t+\varepsilon s = t + ε . Assume ℓ ≥ 2 \ell\ge2 ℓ ≥ 2 . After relabeling the components Ω i , t \Omega_{i,t} Ω i , t , we may suppose that the union ⋃ i = 1 ℓ Ω ‾ i , t \bigcup_{i=1}^{\ell}\overline{\Omega}_{i,t} ⋃ i = 1 ℓ Ω i , t is connected, and that no other Ω ‾ j , t \overline{\Omega}_{j,t} Ω j , t intersects this union. Let ε > 0 \varepsilon>0 ε > 0 be a prescribed parameter, and define Ω \Omega Ω as the connected component of the set { ∣ B ∣ < t + ε } \{|B|<t+\varepsilon\} { ∣ B ∣ < t + ε } that contains ⋃ i = 1 ℓ Ω ‾ i , t \bigcup_{i=1}^{\ell}\overline{\Omega}_{i,t} ⋃ i = 1 ℓ Ω i , t . By choosing ε \varepsilon ε sufficiently small, we ensure that Ω \Omega Ω lies in a small neighborhood of ⋃ i = 1 ℓ Ω ‾ i , t \bigcup_{i=1}^{\ell}\overline{\Omega}_{i,t} ⋃ i = 1 ℓ Ω i , t , which contains no more critical points than the original union ⋃ i = 1 ℓ Ω ‾ i , t \bigcup_{i=1}^{\ell}\overline{\Omega}_{i,t} ⋃ i = 1 ℓ Ω i , t . Since the full union ⋃ j Ω ‾ j , t = { ∣ B ∣ ≤ t } \bigcup_j\overline{\Omega}_{j,t}=\{|B|\le t\} ⋃ j Ω j , t = { ∣ B ∣ ≤ t } consists of only finitely many connected components, there exists a common constant ε > 0 \varepsilon>0 ε > 0 such that this property holds for every corresponding component Ω \Omega Ω . We let t + t^+ t + denote an arbitrary number in ( t , t + ε ) (t,t+\varepsilon) ( t , t + ε ) . In the sequel, t + t^+ t + consistently keeps this meaning.
It suffices to consider every component of { ∣ B ∣ < t + } \{|B|<t^+\} { ∣ B ∣ < t + } containing at least two distinct critical points. We shall prove by induction on s s s that, for a fixed t = t k t=t_k t = t k , such components of { ∣ B ∣ < t + } \{|B|<t^+\} { ∣ B ∣ < t + } are purely gluable.
Step I. Flower model. Remind t = t k t=t_k t = t k . In this case, some of the boundaries ∂ Ω j , t \partial\Omega_{j,t} ∂ Ω j , t share a common point ∗ \ast ∗ , with any two of them intersecting only at ∗ \ast ∗ . For all remaining indices i i i , ∂ Ω i , t \partial\Omega_{i,t} ∂ Ω i , t is disjoint from every such ∂ Ω j , t \partial\Omega_{j,t} ∂ Ω j , t . Since the union of the ∂ Ω j , t \partial\Omega_{j,t} ∂ Ω j , t passing through ∗ \ast ∗ forms a petal-like shape, this set of boundaries sharing the common point ∗ \ast ∗ is defined as the flower model. By hypothesis, the union of the corresponding closed regions Ω ‾ j , t \overline{\Omega}_{j,t} Ω j , t contains at least two distinct critical points.
After reordered, assume that ∂ Ω 0 , t , ∂ Ω 1 , t , … , ∂ Ω k , t \partial\Omega_{0,t},\ \partial\Omega_{1,t},\ldots,\partial\Omega_{k,t} ∂ Ω 0 , t , ∂ Ω 1 , t , … , ∂ Ω k , t are petals of the flower. We will soon show that there exists at least one Ω j , t \Omega_{j,t} Ω j , t (1 ≤ j ≤ k 1\le j\le k 1 ≤ j ≤ k ) with multiplicity at least two. To prove this, we proceed by contradiction and assume conversely that each Ω j , t \Omega_{j,t} Ω j , t (1 ≤ j ≤ k 1\le j\le k 1 ≤ j ≤ k ) has multiplicity one, meaning that B B B is univalent on each such Ω j , t \Omega_{j,t} Ω j , t . Consequently, by Corollary 2.24 the only critical point on ⋃ i = 1 k Ω ‾ i , t \bigcup_{i=1}^{k}\overline{\Omega}_{i,t} ⋃ i = 1 k Ω i , t would be ∗ \ast ∗ . Let δ > 0 \delta>0 δ > 0 be a prescribed parameter and let Ω \Omega Ω denote the component of { ∣ B ∣ < t + δ } \{|B|<t+\delta\} { ∣ B ∣ < t + δ } containing ⋃ i = 1 k Ω ‾ i , t \bigcup_{i=1}^{k}\overline{\Omega}_{i,t} ⋃ i = 1 k Ω i , t . By selecting δ \delta δ sufficiently small, we ensure that Ω \Omega Ω lies in a small neighborhood N \mathcal{N} N of ⋃ i = 1 k Ω ‾ i , t \bigcup_{i=1}^{k}\overline{\Omega}_{i,t} ⋃ i = 1 k Ω i , t . Thus either Ω \Omega Ω or this neighborhood N \mathcal{N} N contains no more critical points than ⋃ i = 1 k Ω ‾ i , t \bigcup_{i=1}^{k}\overline{\Omega}_{i,t} ⋃ i = 1 k Ω i , t does. Consequently, N \mathcal{N} N contains exactly one critical point, ∗ \ast ∗ . However, by the induction hypothesis, Ω \Omega Ω must possess at least two distinct critical points. This contradiction completes the proof.
Without loss of generality, we assume that ∂ Ω 0 , t \partial\Omega_{0,t} ∂ Ω 0 , t is of multiplicity greater than 2. There are two situations to distinguish: either Ω 0 , t \Omega_{0,t} Ω 0 , t contains at least two critical points or not. In either case, we will assume s s s is a bit larger than t t t so that Ω 0 , s \Omega_{0,s} Ω 0 , s denote the component containing Ω 0 , t \Omega_{0,t} Ω 0 , t and the petal associated with it, but Ω 0 , s \Omega_{0,s} Ω 0 , s has no intersection with other Ω j , t \Omega_{j,t} Ω j , t (see Figure 4 ). Following our earlier convention we write t + t^{+} t + for this s s s . As follows, we now show that Ω 0 , t + \Omega_{0,t^{+}} Ω 0 , t + is purely gluable.
Figure 4. Petals associated with Ω 0 , t \Omega_{0,t} Ω 0 , t
Case I.1 Ω 0 , t \Omega_{0,t} Ω 0 , t contains at least two distinct critical points.
By induction on s s s with s = t s=t s = t , we conclude that Ω 0 , t \Omega_{0,t} Ω 0 , t is purely gluable since it contains at least two distinct critical points.
Let w 0 w_0 w 0 be a point in t D \ E B t\mathbb{D}\backslash E_B t D \ E B . Let a 1 , … , a n a_1,\ldots,a_n a 1 , … , a n be all preimages of w 0 w_0 w 0 in Ω 0 , t \Omega_{0,t} Ω 0 , t . For each fixed Ω j , t \Omega_{j,t} Ω j , t (j > 0 j>0 j > 0 ), there exists an integer k = k ( j ) k=k(j) k = k ( j ) so that
B − 1 ( w 0 ) ∩ Ω j , t (see Figure 5) B^{-1}(w_0)\cap\Omega_{j,t}\qquad\text{(see Figure 5)} B − 1 ( w 0 ) ∩ Ω j , t (see Figure 5)
contains k k k elements, b 1 , … , b k b_1,\ldots,b_k b 1 , … , b k . By Corollary 2.28, ρ a 2 , b 1 ∼ ρ a 2 , b i \rho_{a_2,b_1}\sim\rho_{a_2,b_i} ρ a 2 , b 1 ∼ ρ a 2 , b i along a curve in Ω j , t + \Omega_{j,t^{+}} Ω j , t + for each i i i . That is,
ρ a 2 , z ′ ∼ ρ a 2 , z ′ ′ , z ′ , z ′ ′ ∈ B − 1 ( w 0 ) ∩ Ω j , t , (7) \rho_{a_2,z'}\sim\rho_{a_2,z''},\quad z',z''\in B^{-1}(w_0)\cap\Omega_{j,t}, \tag*{(7)} ρ a 2 , z ′ ∼ ρ a 2 , z ′′ , z ′ , z ′′ ∈ B − 1 ( w 0 ) ∩ Ω j , t , (7)
and denote this relation by ρ a 2 , Ω j , t \rho_{a_2,\Omega_{j,t}} ρ a 2 , Ω j , t ; for a local inverse ρ \rho ρ of B B B , we write ρ ∼ ρ a 2 , Ω j , t \rho\sim\rho_{a_2,\Omega_{j,t}} ρ ∼ ρ a 2 , Ω j , t if there exists a point z ′ ∈ B − 1 ( w 0 ) ∩ Ω j , t z'\in B^{-1}(w_0)\cap\Omega_{j,t} z ′ ∈ B − 1 ( w 0 ) ∩ Ω j , t such that ρ ∼ ρ a 2 , z ′ \rho\sim\rho_{a_2,z'} ρ ∼ ρ a 2 , z ′ .
Remind that a curve γ \gamma γ is non-self-intersecting if γ \gamma γ is injective, that is, γ ( s ′ ) ≠ γ ( s ′ ′ ) \gamma(s')\ne\gamma(s'') γ ( s ′ ) = γ ( s ′′ ) for 0 ≤ s ′ < s ′ ′ ≤ 1 0\le s'<s''\le1 0 ≤ s ′ < s ′′ ≤ 1 . Let γ \gamma γ be a non-self-intersecting curve. connecting w 0 w_0 w 0 and B ( ∗ ) B(\ast) B ( ∗ ) , and
γ [ 0 , 1 ) ⊆ t D \ E B . \gamma[0,1) \subseteq tD\backslash E_B. γ [ 0 , 1 ) ⊆ t D \ E B .
Figure 5. Flower model
Then by Lemma 2.25, in each Ω j , t \Omega_{j,t} Ω j , t (j ≥ 0 j \ge0 j ≥ 0 ) there exists exactly one curve σ j \sigma_j σ j with endpoint ∗ \ast ∗ (σ j ( 1 ) = ∗ \sigma_j(1)=\ast σ j ( 1 ) = ∗ ), so that
B ( σ j ( u ) ) = γ ( u ) , 0 ≤ u ≤ 1 , j ≥ 0. B(\sigma_j(u))=\gamma(u), \qquad0\le u\le1,\qquad j\ge0. B ( σ j ( u )) = γ ( u ) , 0 ≤ u ≤ 1 , j ≥ 0.
Write λ j = σ j ( 0 ) \lambda_j=\sigma_j(0) λ j = σ j ( 0 ) . Note that σ 0 ⊆ Ω 0 , t \sigma_0\subseteq\Omega_{0,t} σ 0 ⊆ Ω 0 , t and σ 0 \sigma_0 σ 0 connects ∗ \ast ∗ to some point a i a_i a i . Without loss of generality, assume σ 0 ( 1 ) = a 1 \sigma_0(1)=a_1 σ 0 ( 1 ) = a 1 .
We next prove that
ρ a 2 , a 1 ∼ ρ a 2 , Ω j , t , j ≥ 1. \rho_{a_2,a_1}\sim\rho_{a_2,\Omega_{j,t}},\qquad j\ge1. ρ a 2 , a 1 ∼ ρ a 2 , Ω j , t , j ≥ 1.
To this end, let σ \sigma σ be a curve satisfying
B ( σ ( u ) ) = γ ( u ) , σ ( 0 ) = a 2 . B(\sigma(u))=\gamma(u),\qquad\sigma(0)=a_2. B ( σ ( u )) = γ ( u ) , σ ( 0 ) = a 2 .
We first claim that B ′ ( σ ( 1 ) ) ≠ 0 B'(\sigma(1))\ne0 B ′ ( σ ( 1 )) = 0 . Suppose, to the contrary, that B ′ ( σ ( 1 ) ) = 0 B'(\sigma(1))=0 B ′ ( σ ( 1 )) = 0 . By Lemma 2.25, we also have B ′ ( ∗ ) = 0 B'(\ast)=0 B ′ ( ∗ ) = 0 . Since
B ( σ ( u ) ) = B ( σ j ( u ) ) = γ ( u ) , 0 ≤ u ≤ 1 , B(\sigma(u))=B(\sigma_j(u))=\gamma(u),\qquad0\le u\le1, B ( σ ( u )) = B ( σ j ( u )) = γ ( u ) , 0 ≤ u ≤ 1 ,
it follows that B ( σ ( 1 ) ) = B ( σ j ( 1 ) ) B(\sigma(1))=B(\sigma_j(1)) B ( σ ( 1 )) = B ( σ j ( 1 )) . But σ \sigma σ ends at σ ( 1 ) \sigma(1) σ ( 1 ) , while σ j \sigma_j σ j ends at ∗ \ast ∗ ; hence the two critical points σ ( 1 ) \sigma(1) σ ( 1 ) and ∗ \ast ∗ are mapped to the same value under B B B . This contradicts the injectivity of B B B on Z ( B ′ ) Z(B') Z ( B ′ ) , so we conclude that B ′ ( σ ( 1 ) ) ≠ 0 B'(\sigma(1))\ne0 B ′ ( σ ( 1 )) = 0 .
Applying Lemma 2.16 now gives
ρ a 2 , a 1 ∼ ρ a 2 , λ j , j ≥ 1. \rho_{a_2,a_1} \sim\rho_{a_2,\lambda_j}, \qquad j \ge1. ρ a 2 , a 1 ∼ ρ a 2 , λ j , j ≥ 1.
Since λ j ∈ Ω j , t \lambda_j \in\Omega_{j,t} λ j ∈ Ω j , t , by (7) we have ρ a 2 , λ j ∼ ρ a 2 , Ω j , t \rho_{a_2,\lambda_j} \sim\rho_{a_2,\Omega_{j,t}} ρ a 2 , λ j ∼ ρ a 2 , Ω j , t . Combining this with the preceding equivalence yields
ρ a 2 , a 1 ∼ ρ a 2 , Ω j , t , j ≥ 1 , (8) \rho_{a_2,a_1} \sim\rho_{a_2,\Omega_{j,t}}, \qquad j \ge1, \tag*{(8)} ρ a 2 , a 1 ∼ ρ a 2 , Ω j , t , j ≥ 1 , (8)
as desired.
Finally, we will show that Ω 0 , t + \Omega_{0,t+} Ω 0 , t + is purely gluable. Since Ω 0 , t \Omega_{0,t} Ω 0 , t is purely gluable by induction, we have ρ a 2 , a 1 ∼ ρ a 2 , a j \rho_{a_2,a_1} \sim\rho_{a_2,a_j} ρ a 2 , a 1 ∼ ρ a 2 , a j for j ≠ 2 j \ne2 j = 2 . This, together with (8) , yields
ρ a 2 , a 1 ∼ ρ a 2 , z ′ \rho_{a_2,a_1} \sim\rho_{a_2,z'} ρ a 2 , a 1 ∼ ρ a 2 , z ′
for every z ′ ∈ B − 1 ( w 0 ) ∩ Ω 0 , t + z' \in B^{-1}(w_0) \cap\Omega_{0,t+} z ′ ∈ B − 1 ( w 0 ) ∩ Ω 0 , t + with z ′ ≠ a 2 z' \ne a_2 z ′ = a 2 , with the convention that ρ a 2 , a 2 = id \rho_{a_2,a_2} = \operatorname{id} ρ a 2 , a 2 = id . Thereby, it follows that on Ω 0 , t + ∖ B − 1 ( E B ) \Omega_{0,t+} \setminus B^{-1}(\mathcal{E}_B) Ω 0 , t + ∖ B − 1 ( E B ) , there are exactly two equivalence classes of local inverses of B B B . By Proposition 2.20 , the Riemann surface
{ ( z , w ) ∈ Ω 0 , t + : B ( z ) = B ( w ) ∉ E B } \{(z,w) \in\Omega_{0,t+}: B(z) = B(w) \notin\mathcal{E}_B\} {( z , w ) ∈ Ω 0 , t + : B ( z ) = B ( w ) ∈ / E B }
has exactly two components: the component containing ( a 2 , a 1 ) (a_2,a_1) ( a 2 , a 1 ) and the identity component S 0 S_0 S 0 . Hence Ω 0 , t + \Omega_{0,t+} Ω 0 , t + is purely gluable, which completes the proof in this case.
Case I.2. Rotation case. In this case, Ω 0 , t \Omega_{0,t} Ω 0 , t contains exactly one critical point p p p , called the rotation case .
Since B ∣ Ω 0 , t B|_{\Omega_{0,t}} B ∣ Ω 0 , t is a proper map of multiplicity, say n n n , Proposition 2.3 yields a conformal map F : Ω → D F:\Omega\to\mathbb{D} F : Ω → D and a finite Blaschke product ϕ \phi ϕ of order n n n such that
B ∣ Ω 0 , t = t ( ϕ ∘ F ) ∣ Ω 0 , t . B|_{\Omega_{0,t}} = t(\phi\circ F)|_{\Omega_{0,t}}. B ∣ Ω 0 , t = t ( ϕ ∘ F ) ∣ Ω 0 , t .
Since Ω 0 , t \Omega_{0,t} Ω 0 , t contains exactly one critical point p p p , ϕ \phi ϕ also has exactly one critical point. There exist a ∈ D a \in\mathbb{D} a ∈ D and a holomorphic automorphism φ \varphi φ of D \mathbb{D} D such that ϕ = φ ( ϕ n ) \phi= \varphi(\phi^n) ϕ = φ ( ϕ n ) . From the above factorization and the properties of ϕ \phi ϕ , we obtain n n n pairwise disjoint non-self-intersecting curves γ 1 , … , γ n \gamma_1,\ldots,\gamma_n γ 1 , … , γ n and n n n distinct points a 1 , … , a n a_1,\ldots,a_n a 1 , … , a n in Ω 0 , t ∖ B − 1 ( E B ) \Omega_{0,t}\setminus B^{-1}(\mathcal{E}_B) Ω 0 , t ∖ B − 1 ( E B ) such that
B ( γ 1 ( u ) ) = ⋯ = B ( γ n ( u ) ) , 0 ≤ u ≤ 1 B(\gamma_1(u)) = \cdots= B(\gamma_n(u)), \qquad0 \le u \le1 B ( γ 1 ( u )) = ⋯ = B ( γ n ( u )) , 0 ≤ u ≤ 1
with γ i ( 0 ) = a i \gamma_i(0) = a_i γ i ( 0 ) = a i and γ i ( 1 ) = a i + 1 \gamma_i(1) = a_{i+1} γ i ( 1 ) = a i + 1 for 1 ≤ i ≤ n 1 \le i \le n 1 ≤ i ≤ n (where a n + 1 ≡ a 1 a_{n+1} \equiv a_1 a n + 1 ≡ a 1 ). Applying Lemma 2.15 yields the following equivalence
ρ γ i ( 0 ) , γ j ( 0 ) ∼ ρ γ i ( 1 ) , γ j ( 1 ) , 1 ≤ i ≤ j ≤ n \rho_{\gamma_i(0),\gamma_j(0)} \sim\rho_{\gamma_i(1),\gamma_j(1)}, \qquad1 \le i \le j \le n ρ γ i ( 0 ) , γ j ( 0 ) ∼ ρ γ i ( 1 ) , γ j ( 1 ) , 1 ≤ i ≤ j ≤ n
For i = 1 i=1 i = 1 and j = 2 , … , n j=2,\ldots,n j = 2 , … , n , we have
ρ a 1 , a j ∼ ρ a 2 , a j + 1 , 2 ≤ j ≤ n . (9) \rho_{a_1,a_j}\sim\rho_{a_2,a_{j+1}},\qquad2\le j\le n. \tag*{(9)} ρ a 1 , a j ∼ ρ a 2 , a j + 1 , 2 ≤ j ≤ n . (9)
Here the case j = n j=n j = n yields ρ a 1 , a n ∼ ρ a 2 , a 1 \rho_{a_1,a_n}\sim\rho_{a_2,a_1} ρ a 1 , a n ∼ ρ a 2 , a 1 .
As follows, we will prove that ( a 2 , a i ) (a_2,a_i) ( a 2 , a i ) (where i ≠ 2 i\ne2 i = 2 ) lie in a same component of { ( z , w ) ∈ Ω 0 , t + : B ( z ) = B ( w ) ∉ E B } \{(z,w)\in\Omega_{0,t+}:B(z)=B(w)\notin E_B\} {( z , w ) ∈ Ω 0 , t + : B ( z ) = B ( w ) ∈ / E B } . For this, let λ 1 ∈ Ω 1 , t \lambda_1\in\Omega_{1,t} λ 1 ∈ Ω 1 , t be such that B ( λ 1 ) = B ( a 1 ) B(\lambda_1)=B(a_1) B ( λ 1 ) = B ( a 1 ) and λ 1 ≃ B a 1 \lambda_1\overset{B}{\simeq}a_1 λ 1 ≃ B a 1 , with a 1 , … , a n ∈ Ω 0 , t a_1,\ldots,a_n\in\Omega_{0,t} a 1 , … , a n ∈ Ω 0 , t (See Figure 5). Since B B B is injective on critical points, by Corollary 2.28 we obtain
ρ λ 1 , a 1 ∼ ⋯ ∼ ρ λ 1 , a n . \rho_{\lambda_1,a_1}\sim\cdots\sim\rho_{\lambda_1,a_n}. ρ λ 1 , a 1 ∼ ⋯ ∼ ρ λ 1 , a n .
By Remark 2.18(2), we have ρ a 1 , λ 1 ∼ ⋯ ∼ ρ a n , λ 1 \rho_{a_1,\lambda_1}\sim\cdots\sim\rho_{a_n,\lambda_1} ρ a 1 , λ 1 ∼ ⋯ ∼ ρ a n , λ 1 . In particular,
ρ a 2 , λ 1 ∼ ⋯ ∼ ρ a n , λ 1 . (10) \rho_{a_2,\lambda_1}\sim\cdots\sim\rho_{a_n,\lambda_1}. \tag*{(10)} ρ a 2 , λ 1 ∼ ⋯ ∼ ρ a n , λ 1 . (10)
Next we prove that ρ a 2 , λ 1 ∼ ρ a 2 , a 1 \rho_{a_2,\lambda_1}\sim\rho_{a_2,a_1} ρ a 2 , λ 1 ∼ ρ a 2 , a 1 . Since λ 1 ≃ B a 1 \lambda_1\overset{B}{\simeq}a_1 λ 1 ≃ B a 1 , λ 1 \lambda_1 λ 1 and a 1 a_1 a 1 are connected by B B B -equal curves ending at p p p ; that is, there exist curves ℘ 1 , ℘ 2 \wp_1,\wp_2 ℘ 1 , ℘ 2 whose ranges lie in Ω 0 , t + \Omega_{0,t+} Ω 0 , t + , satisfying ℘ 1 ( 0 ) = λ 1 \wp_1(0)=\lambda_1 ℘ 1 ( 0 ) = λ 1 , ℘ 2 ( 0 ) = a 1 \wp_2(0)=a_1 ℘ 2 ( 0 ) = a 1 , ℘ 1 ( 1 ) = ℘ 2 ( 1 ) = p \wp_1(1)=\wp_2(1)=p ℘ 1 ( 1 ) = ℘ 2 ( 1 ) = p and B ( ℘ 1 ( u ) ) = B ( ℘ 2 ( u ) ) B(\wp_1(u))=B(\wp_2(u)) B ( ℘ 1 ( u )) = B ( ℘ 2 ( u )) , 0 ≤ u ≤ 1 0\le u\le1 0 ≤ u ≤ 1 (in fact, for any t ′ > t t'>t t ′ > t , the ranges are contained in the closure of the component of { ∣ B ∣ < t ′ } \{|B|<t'\} { ∣ B ∣ < t ′ } which contains Ω 0 , t \Omega_{0,t} Ω 0 , t , as well as the nearby petals around Ω 0 , t \Omega_{0,t} Ω 0 , t ). Then by Lemma 2.25 p p p is a critical point. Again by Lemma 2.25, there exists another B B B -equal curve ℘ 0 \wp_0 ℘ 0 beginning from a 2 a_2 a 2 and ending at a distinct point v ≠ p v\ne p v = p ; that is, ℘ 1 ( 0 ) = v \wp_1(0)=v ℘ 1 ( 0 ) = v , ℘ 2 ( 0 ) = p \wp_2(0)=p ℘ 2 ( 0 ) = p , and
B ( ℘ 0 ( u ) ) = B ( ℘ 1 ( u ) ) = B ( ℘ 2 ( u ) ) , 0 ≤ u ≤ 1. B(\wp_0(u))=B(\wp_1(u))=B(\wp_2(u)),\qquad0\le u\le1. B ( ℘ 0 ( u )) = B ( ℘ 1 ( u )) = B ( ℘ 2 ( u )) , 0 ≤ u ≤ 1.
By the assumption of the flower model, p p p serves as the only critical point on ∂ Ω j , t \partial\Omega_{j,t} ∂ Ω j , t . Consequently, at the endpoint v = γ 1 ( 0 ) v=\gamma_1(0) v = γ 1 ( 0 ) , we must have B ′ ( v ) ≠ 0 B'(v)\ne0 B ′ ( v ) = 0 . Applying Lemma 2.16, we obtain ρ a 2 , λ 1 ∼ ρ a 2 , a 1 \rho_{a_2,\lambda_1}\sim\rho_{a_2,a_1} ρ a 2 , λ 1 ∼ ρ a 2 , a 1 , as desired.
By similar reasoning, ρ a j , λ 1 ∼ ρ a j , a 1 \rho_{a_j,\lambda_1}\sim\rho_{a_j,a_1} ρ a j , λ 1 ∼ ρ a j , a 1 for j ≥ 2 j\ge2 j ≥ 2 . This, combined with the identity (3.5), shows that
ρ a 2 , a 1 ∼ ⋯ ∼ ρ a n , a 1 . \rho_{a_2,a_1}\sim\cdots\sim\rho_{a_n,a_1}. ρ a 2 , a 1 ∼ ⋯ ∼ ρ a n , a 1 .
Therefore, by Proposition 2.17, we obtain ρ a 1 , a 2 ∼ ⋯ ∼ ρ a 1 , a n \rho_{a_1,a_2}\sim\cdots\sim\rho_{a_1,a_n} ρ a 1 , a 2 ∼ ⋯ ∼ ρ a 1 , a n . Combining this relation with (3.4), we obtain that
ρ a 2 , a 3 ∼ ρ a 2 , a 4 ∼ ⋯ ∼ ρ a 2 , a 1 . (11) \rho_{a_2,a_3}\sim\rho_{a_2,a_4}\sim\cdots\sim\rho_{a_2,a_1}. \tag*{(11)} ρ a 2 , a 3 ∼ ρ a 2 , a 4 ∼ ⋯ ∼ ρ a 2 , a 1 . (11)
As in Case I.1, we conclude that
ρ a 2 , a 1 ∼ ρ a 2 , Ω j , t , j ≥ 1. \rho_{a_2,a_1}\sim\rho_{a_2,\Omega_{j,t}},\qquad j\ge1. ρ a 2 , a 1 ∼ ρ a 2 , Ω j , t , j ≥ 1.
Combining this with (11) , we obtain
ρ a 2 , a 1 ∼ ρ a 2 , z ′ \rho_{a_2,a_1} \sim\rho_{a_2,z'} ρ a 2 , a 1 ∼ ρ a 2 , z ′
for every z ′ ∈ B − 1 ( w 0 ) ∩ Ω 0 , t + z' \in B^{-1}(w_0) \cap\Omega_{0,t+} z ′ ∈ B − 1 ( w 0 ) ∩ Ω 0 , t + with z ′ ≠ a 2 z' \ne a_2 z ′ = a 2 where ρ a 2 , a 2 = i d \rho_{a_2,a_2}=id ρ a 2 , a 2 = i d . Hence, on Ω 0 , t + ∖ B − 1 ( E B ) \Omega_{0,t+}\setminus B^{-1}(\mathcal{E}_B) Ω 0 , t + ∖ B − 1 ( E B ) , there are exactly two equivalence classes of local inverses of B B B . By Proposition 2.20 , the Riemann surface
{ ( z , w ) ∈ Ω 0 , t + : B ( z ) = B ( w ) ∉ E B } \{(z,w)\in\Omega_{0,t+}: B(z)=B(w)\notin\mathcal{E}_B\} {( z , w ) ∈ Ω 0 , t + : B ( z ) = B ( w ) ∈ / E B }
admits exactly two components: the one containing ( a 2 , a 1 ) (a_2,a_1) ( a 2 , a 1 ) and the identity component S 0 S_0 S 0 .
Step II. We proceed to address the general case, hereafter referred to as the Cactuses Case. The name will become clear as we see the shape of the branching diagram shown later.
As s s s grows bigger, boundaries of distinct components of { ∣ B ∣ < s } \{|B|<s\} { ∣ B ∣ < s } intersect at one or more common points. Theorem 2.26 guarantees the existence of a finite-vertex tree structure. After relabeling, let Ω 1 , ⋯ , Ω n \Omega_1,\cdots,\Omega_n Ω 1 , ⋯ , Ω n be the components of { ∣ B ∣ < t } \{|B|<t\} { ∣ B ∣ < t } whose boundaries intersect those of other components, chosen such that their union is maximal with respect to inclusion, subject to the condition that ⋃ i = 1 n ∂ Ω i \bigcup_{i=1}^{n}\partial\Omega_i ⋃ i = 1 n ∂ Ω i is connected. We now treat Ω 1 , ⋯ , Ω N \Omega_1,\cdots,\Omega_N Ω 1 , ⋯ , Ω N as abstract vertices. By Theorem 2.26, any nonempty intersection ∂ Ω i ∩ ∂ Ω j \partial\Omega_i\cap\partial\Omega_j ∂ Ω i ∩ ∂ Ω j (i ≠ j i\ne j i = j ), contains exactly one point P P P . Then we construct edges by connecting Ω 1 \Omega_1 Ω 1 and Ω 2 \Omega_2 Ω 2 for each such pair, and mark the intersection point P P P as a vertex. By Theorem 2.26, this procedure yields a connected graph without cycles. We denote the graph vertices as { Ω 1 , Ω 2 , ⋯ , Ω n } \{\Omega_1,\Omega_2,\cdots,\Omega_n\} { Ω 1 , Ω 2 , ⋯ , Ω n } and { P 1 , ⋯ , P m } \{P_1,\cdots,P_m\} { P 1 , ⋯ , P m } , where the latter denote a subset of the critical set of B B B (See Figure 6 ).
Figure 6. E-components and R-components
Before continuing, we introduce two definitions: E E E -component and R R R -component. Given components Ω 1 , ⋯ , Ω n \Omega_1,\cdots,\Omega_n Ω 1 , ⋯ , Ω n of { ∣ B ∣ < t } \{|B|<t\} { ∣ B ∣ < t } , if one component Ω k \Omega_k Ω k is such that there exist two distinct components Ω i \Omega_i Ω i and Ω j \Omega_j Ω j satisfying
∂ Ω i ∩ ∂ Ω k ≠ ∅ , ∂ Ω j ∩ ∂ Ω k ≠ ∅ \partial\Omega_i \cap\partial\Omega_k \ne\varnothing,\qquad\partial\Omega_j \cap\partial\Omega_k \ne\varnothing ∂ Ω i ∩ ∂ Ω k = ∅ , ∂ Ω j ∩ ∂ Ω k = ∅
and ∂ Ω i ∩ ∂ Ω k ≠ ∂ Ω j ∩ ∂ Ω k \partial\Omega_i \cap\partial\Omega_k \ne\partial\Omega_j \cap\partial\Omega_k ∂ Ω i ∩ ∂ Ω k = ∂ Ω j ∩ ∂ Ω k , then Ω k \Omega_k Ω k is called an E-component (edge-component). If a component Ω k \Omega_k Ω k is such that there is a component Ω i \Omega_i Ω i satisfying ∂ Ω i ∩ ∂ Ω k ≠ ∅ \partial\Omega_i \cap\partial\Omega_k \ne\varnothing ∂ Ω i ∩ ∂ Ω k = ∅ and Ω k \Omega_k Ω k is not an E-component, then Ω k \Omega_k Ω k is called an R-component (root component).
The resulting graph G G G contains at least two R-components ; here, a component Ω k \Omega_k Ω k is an R-component exactly when it acts like a root when regarded as a vertex in G G G . This usage is not to be confused with the standard notion of a root vertex in graph theory.
Case II.1. There exists one R-component of multiplicity no less than two.
Let Ω 0 , t \Omega_{0,t} Ω 0 , t be such an R-component . There are two situations under consideration: either Ω 0 , t \Omega_{0,t} Ω 0 , t contains at least two distinct critical points (and by induction it is purely gluable) or Ω 0 , t \Omega_{0,t} Ω 0 , t contains exactly one critical point.
We first consider the case where Ω 0 , t \Omega_{0,t} Ω 0 , t contains at least two distinct critical points; the remaining case is similar. As done in Case I.2, we can pick w 0 ∈ t D \ E B w_0 \in tD\backslash E_B w 0 ∈ t D \ E B and let a 1 , … , a n a_1,\ldots,a_n a 1 , … , a n be all preimages of w 0 w_0 w 0 in Ω 0 , t \Omega_{0,t} Ω 0 , t . Since Ω 0 , t \Omega_{0,t} Ω 0 , t is an R-component , there exist Ω 1 , t ∗ , … , Ω N , t ∗ \Omega^*_{1,t},\ldots,\Omega^*_{N,t} Ω 1 , t ∗ , … , Ω N , t ∗ (N ≥ 1 N \ge1 N ≥ 1 ) such that
∂ Ω j , t ∗ ∩ ∂ Ω 0 , t \partial\Omega^*_{j,t}\cap\partial\Omega_{0,t} ∂ Ω j , t ∗ ∩ ∂ Ω 0 , t
share a common critical point ∗ * ∗ for 1 ≤ j ≤ N 1 \le j \le N 1 ≤ j ≤ N . These are petals of the flower with critical point ∗ * ∗ . Let γ \gamma γ be a non-self-intersecting curve connecting w 0 w_0 w 0 and B ( ∗ ) B(*) B ( ∗ ) , and
γ ( [ 0 , 1 ) ) ⊆ t D \ E B . \gamma([0,1)) \subseteq tD\backslash E_B. γ ([ 0 , 1 )) ⊆ t D \ E B .
Then by Lemma 2.25, in each Ω j , t ∗ \Omega^*_{j,t} Ω j , t ∗ (1 ≤ j ≤ N 1 \le j \le N 1 ≤ j ≤ N ) there exists exactly one curve σ j \sigma_j σ j with endpoint ∗ * ∗ (σ j ( 1 ) = ∗ \sigma_j(1)=* σ j ( 1 ) = ∗ ), and
B ( σ j ( u ) ) = γ ( u ) , 0 ≤ u ≤ 1. B(\sigma_j(u))=\gamma(u),\qquad0\le u\le1. B ( σ j ( u )) = γ ( u ) , 0 ≤ u ≤ 1.
Write λ j = σ j ( 0 ) \lambda_j=\sigma_j(0) λ j = σ j ( 0 ) . In Ω 0 , t \Omega_{0,t} Ω 0 , t , the preimage B − 1 ( γ ) ∩ Ω 0 , t B^{-1}(\gamma)\cap\Omega_{0,t} B − 1 ( γ ) ∩ Ω 0 , t of γ \gamma γ contains a curve connecting a 1 a_1 a 1 and ∗ * ∗ . Again by Lemma 2.25, there exists a curve σ 0 \sigma_0 σ 0 with σ 0 ( 0 ) = a 1 \sigma_0(0)=a_1 σ 0 ( 0 ) = a 1 and σ 0 ( 1 ) = ∗ \sigma_0(1)=* σ 0 ( 1 ) = ∗ , such that
B ( σ 0 ( u ) ) = γ ( u ) , 0 ≤ u ≤ 1. B(\sigma_0(u))=\gamma(u),\qquad0\le u\le1. B ( σ 0 ( u )) = γ ( u ) , 0 ≤ u ≤ 1.
Therefore,
B ( σ 0 ( u ) ) = B ( σ j ( u ) ) = γ ( u ) , 0 ≤ u ≤ 1 , B(\sigma_0(u))=B(\sigma_j(u))=\gamma(u),\qquad0\le u\le1, B ( σ 0 ( u )) = B ( σ j ( u )) = γ ( u ) , 0 ≤ u ≤ 1 ,
meaning a 1 a_1 a 1 and λ j \lambda_j λ j are both connected to the point ∗ \ast ∗ via B B B -equal curves. Applying Lemma 2.16 , we obtain
ρ a 2 , a 1 ∼ ρ a 2 , λ j , 1 ≤ j ≤ N . (12) \rho_{a_2,a_1}\sim\rho_{a_2,\lambda_j},\qquad1\leq j\leq N. \tag*{(12)} ρ a 2 , a 1 ∼ ρ a 2 , λ j , 1 ≤ j ≤ N . (12)
Then by Corollary 2.28,
ρ a 2 , a 1 ∼ ρ a 2 , λ j ∼ ρ a 2 , Ω j , t ∗ , 1 ≤ j ≤ N ; \rho_{a_2,a_1}\sim\rho_{a_2,\lambda_j}\sim\rho_{a_2,\Omega_{j,t}^{*}},\qquad1\leq j\leq N; ρ a 2 , a 1 ∼ ρ a 2 , λ j ∼ ρ a 2 , Ω j , t ∗ , 1 ≤ j ≤ N ;
remind that for a set G G G , ρ a 2 , G \rho_{a_2,G} ρ a 2 , G means with w 0 = B ( a 2 ) w_0=B(a_2) w 0 = B ( a 2 ) ,
ρ a 2 , z ′ ∼ ρ a 2 , z ′ ′ , z ′ , z ′ ′ ∈ B − 1 ( w 0 ) ∩ G . \rho_{a_2,z'}\sim\rho_{a_2,z''},\qquad z',z''\in B^{-1}(w_0)\cap G. ρ a 2 , z ′ ∼ ρ a 2 , z ′′ , z ′ , z ′′ ∈ B − 1 ( w 0 ) ∩ G .
Furthermore, for certain domains Ω \Omega Ω (to be specified shortly), we will prove that
ρ a 2 , a 1 ∼ ρ a 2 , z , z ∈ B − 1 ( w 0 ) ∩ Ω . \rho_{a_2,a_1}\sim\rho_{a_2,z},\qquad z\in B^{-1}(w_0)\cap\Omega. ρ a 2 , a 1 ∼ ρ a 2 , z , z ∈ B − 1 ( w 0 ) ∩ Ω.
More precisely, these domains lie in a chain of components associated with some Ω j , t ∗ \Omega_{j,t}^{*} Ω j , t ∗ , as illustrated below. For instance, there exist components
Ω 2 , t , ⋯ , Ω k , t \Omega_{2,t},\cdots,\Omega_{k,t} Ω 2 , t , ⋯ , Ω k , t
with Ω k , t = Ω \Omega_{k,t}=\Omega Ω k , t = Ω such that
∂ Ω 1 , t ∗ ∩ ∂ Ω 2 , t ≠ ∅ , ∂ Ω 2 , t ∩ ∂ Ω 3 , t ≠ ∅ , ⋯ , ∂ Ω k − 1 , t ∩ ∂ Ω k , t ≠ ∅ . \partial\Omega_{1,t}^{*}\cap\partial\Omega_{2,t}\neq\varnothing,\ \partial\Omega_{2,t}\cap\partial\Omega_{3,t}\neq\varnothing,\ \cdots,\ \partial\Omega_{k-1,t}\cap\partial\Omega_{k,t}\neq\varnothing. ∂ Ω 1 , t ∗ ∩ ∂ Ω 2 , t = ∅ , ∂ Ω 2 , t ∩ ∂ Ω 3 , t = ∅ , ⋯ , ∂ Ω k − 1 , t ∩ ∂ Ω k , t = ∅ .
The sequence ( Ω 1 , t ∗ , Ω 2 , t , ⋯ , Ω k , t ) (\Omega_{1,t}^{*},\Omega_{2,t},\cdots,\Omega_{k,t}) ( Ω 1 , t ∗ , Ω 2 , t , ⋯ , Ω k , t ) is then called a chain (see Figure 7 ). By Corollary 2.28, we clearly have the relationships:
Figure 7. Chain
ρ a 2 , Ω 1 , t ∗ , ρ a 2 , Ω 2 , t , ⋯ , ρ a 2 , Ω k , t . \rho_{a_2,\Omega_{1,t}^{*}},\rho_{a_2,\Omega_{2,t}},\cdots,\rho_{a_2,\Omega_{k,t}}. ρ a 2 , Ω 1 , t ∗ , ρ a 2 , Ω 2 , t , ⋯ , ρ a 2 , Ω k , t .
We will verify that
ρ a 2 , Ω i , t ∼ ρ a 2 , Ω i + 1 , t , i ≥ 2 , \rho_{a_2,\Omega_{i,t}}\sim\rho_{a_2,\Omega_{i+1,t}},\qquad i\geq2, ρ a 2 , Ω i , t ∼ ρ a 2 , Ω i + 1 , t , i ≥ 2 ,
and also
ρ a 2 , Ω 1 , t ∗ ∼ ρ a 2 , Ω 2 , t . \rho_{a_2,\Omega_{1,t}^{*}}\sim\rho_{a_2,\Omega_{2,t}}. ρ a 2 , Ω 1 , t ∗ ∼ ρ a 2 , Ω 2 , t .
For this, Theorem 2.26 (1) implies that ∂ Ω i , t ∩ ∂ Ω i + 1 , t \partial\Omega_{i,t}\cap\partial\Omega_{i+1,t} ∂ Ω i , t ∩ ∂ Ω i + 1 , t contains exactly one critical point, denoted by Q Q Q . Let w 0 = B ( a 1 ) w_0=B(a_1) w 0 = B ( a 1 ) , and let ℘ \wp ℘ be a non-self-intersecting curve connecting w 0 w_0 w 0 to B ( Q ) B(Q) B ( Q ) such that
℘ ( u ) ∈ t D \ E B , 0 ≤ u < 1. \wp(u)\in t\mathbb{D}\backslash E_B,\qquad0\leq u<1. ℘ ( u ) ∈ t D \ E B , 0 ≤ u < 1.
By Lemma 2.25, there are two curves, lying in Ω i , t \Omega_{i,t} Ω i , t and Ω i + 1 , t \Omega_{i+1,t} Ω i + 1 , t respectively, connecting a point in B − 1 ( w 0 ) B^{-1}(w_0) B − 1 ( w 0 ) to Q Q Q . Let λ \lambda λ and μ \mu μ be the endpoints of these two curves lying in B − 1 ( w 0 ) B^{-1}(w_0) B − 1 ( w 0 ) ; then λ ≃ B μ \lambda\underset{B}{\simeq} \mu λ B ≃ μ . Applying Lemma 2.16 immediately yields
ρ a 2 , λ ∼ ρ a 2 , μ . \rho_{a_2,\lambda} \sim\rho_{a_2,\mu}. ρ a 2 , λ ∼ ρ a 2 , μ .
Combined with the equivalences
ρ a 2 , λ ∼ ρ a 2 , Ω i , t and ρ a 2 , μ ∼ ρ a 2 , Ω i + 1 , t , \rho_{a_2,\lambda} \sim\rho_{a_2,\Omega_{i,t}} \quad\text{and}\quad\rho_{a_2,\mu} \sim\rho_{a_2,\Omega_{i+1,t}}, ρ a 2 , λ ∼ ρ a 2 , Ω i , t and ρ a 2 , μ ∼ ρ a 2 , Ω i + 1 , t ,
we conclude that
ρ a 2 , Ω i , t ∼ ρ a 2 , Ω i + 1 , t . \rho_{a_2,\Omega_{i,t}} \sim\rho_{a_2,\Omega_{i+1,t}}. ρ a 2 , Ω i , t ∼ ρ a 2 , Ω i + 1 , t .
The argument for ρ a 2 , Ω 1 , t ∗ ∼ ρ a 2 , Ω 2 , t \rho_{a_2,\Omega_{1,t}^{*}} \sim\rho_{a_2,\Omega_{2,t}} ρ a 2 , Ω 1 , t ∗ ∼ ρ a 2 , Ω 2 , t is similar. Combining the above identities with (12) , we obtain
ρ a 2 , a 1 ∼ ρ a 2 , λ 1 ∼ ρ a 2 , Ω 1 , t ∗ ∼ ρ a 2 , Ω 2 , t ∼ ⋯ ∼ ρ a 2 , Ω k , t . (13) \rho_{a_2,a_1} \sim\rho_{a_2,\lambda_1} \sim\rho_{a_2,\Omega_{1,t}^{*}} \sim\rho_{a_2,\Omega_{2,t}} \sim\cdots\sim\rho_{a_2,\Omega_{k,t}}. \tag*{(13)} ρ a 2 , a 1 ∼ ρ a 2 , λ 1 ∼ ρ a 2 , Ω 1 , t ∗ ∼ ρ a 2 , Ω 2 , t ∼ ⋯ ∼ ρ a 2 , Ω k , t . (13)
Since Ω 0 , t \Omega_{0,t} Ω 0 , t is purely gluable, we have ρ a 2 , a 1 ∼ ρ a 2 , a j \rho_{a_2,a_1} \sim\rho_{a_2,a_j} ρ a 2 , a 1 ∼ ρ a 2 , a j for all j ≠ 2 j \ne2 j = 2 . This, together with (13) , shows that
ρ a 2 , a 1 ∼ ρ a 2 , z ′ . \rho_{a_2,a_1} \sim\rho_{a_2,z'}. ρ a 2 , a 1 ∼ ρ a 2 , z ′ .
for every z ′ ∈ B − 1 ( w 0 ) ∩ Ω 0 , t + z' \in B^{-1}(w_0) \cap\Omega_{0,t+} z ′ ∈ B − 1 ( w 0 ) ∩ Ω 0 , t + with z ′ ≠ a 2 z' \ne a_2 z ′ = a 2 . Hence, on Ω 0 , t + ∖ B − 1 ( E B ) \Omega_{0,t+} \setminus B^{-1}(\mathcal{E}_B) Ω 0 , t + ∖ B − 1 ( E B ) , there are exactly two equivalence classes of local inverses of B B B . By Proposition 2.20 , the Riemann surface
{ ( z , w ) ∈ Ω 0 , t + : B ( z ) = B ( w ) ∉ E B } \left\{(z,w) \in\Omega_{0,t+}: B(z)=B(w)\notin\mathcal{E}_B\right\} { ( z , w ) ∈ Ω 0 , t + : B ( z ) = B ( w ) ∈ / E B }
has exactly two components. That is, Ω 0 , t + \Omega_{0,t+} Ω 0 , t + is purely gluable. We now address the case where Ω 0 , t \Omega_{0,t} Ω 0 , t contains exactly one critical point. By the arguments developed above, we immediately derive (3.8). Following the reasoning in Rotation Case of Case II, we further establish the relation (3.6). As a result, the conclusion that Ω 0 , t + \Omega_{0,t+} Ω 0 , t + is purely gluable still holds.
All R R R -components are of multiplicity one.
To illustrate the idea, we begin with a simplified example (see Figure 8 ) before addressing the general case.
Figure 8. Case 2-1
A special case. We assume, in this simplified case, that all components are of multiplicity one, and that there exists only one E E E -component, denoted by Ω 3 \Omega_3 Ω 3 , whose boundary ∂ Ω 3 \partial\Omega_3 ∂ Ω 3 contains exactly two critical points P P P and Q Q Q . There are R R R -components Ω 1 1 , ⋯ , Ω n 1 \Omega_1^1,\cdots,\Omega_n^1 Ω 1 1 , ⋯ , Ω n 1 and R R R -components Ω 2 2 , ⋯ , Ω m 2 \Omega_2^2,\cdots,\Omega_m^2 Ω 2 2 , ⋯ , Ω m 2 such that
P ∈ ∂ Ω i 1 ∩ ∂ Ω 3 , 1 ≤ i ≤ n P \in\partial\Omega_i^1 \cap\partial\Omega_3,\qquad1 \leq i \leq n P ∈ ∂ Ω i 1 ∩ ∂ Ω 3 , 1 ≤ i ≤ n
and
Q ∈ ∂ Ω j 2 ∩ ∂ Ω 3 , 2 ≤ j ≤ m . Q \in\partial\Omega_j^2 \cap\partial\Omega_3,\qquad2 \leq j \leq m. Q ∈ ∂ Ω j 2 ∩ ∂ Ω 3 , 2 ≤ j ≤ m .
This labeling Ω 2 2 , ⋯ , Ω m 2 \Omega_2^2,\cdots,\Omega_m^2 Ω 2 2 , ⋯ , Ω m 2 is adopted for notational convenience in the sequel. By Theorem 2.26 , P P P and Q Q Q are the only possible intersection points of the boundaries of any two such domains (among those components Ω j i \Omega_j^i Ω j i for i = 1 , 2 i=1,2 i = 1 , 2 , and Ω 3 \Omega_3 Ω 3 ). Let w 0 w_0 w 0 be a point in t D \ E B tD\backslash E_B t D \ E B , and define:
a i a_i a i as the unique point in Ω i 1 ∩ B − 1 ( w 0 ) \Omega_i^1 \cap B^{-1}(w_0) Ω i 1 ∩ B − 1 ( w 0 ) for 1 ≤ i ≤ n 1 \le i \le n 1 ≤ i ≤ n ,
b j b_j b j as the unique point in Ω j 2 ∩ B − 1 ( w 0 ) \Omega_j^2 \cap B^{-1}(w_0) Ω j 2 ∩ B − 1 ( w 0 ) for 2 ≤ j ≤ m 2 \le j \le m 2 ≤ j ≤ m .
In the special case, b 1 b_1 b 1 is the only point in Ω 3 ∩ B − 1 ( w 0 ) \Omega_3 \cap B^{-1}(w_0) Ω 3 ∩ B − 1 ( w 0 ) (as illustrated in Figure 8).
We will soon obtain that
ρ b 2 , b 1 ∼ ρ b 2 , a i , 1 ≤ i ≤ n . (14) \rho_{b_2,b_1} \sim\rho_{b_2,a_i}, \qquad1 \le i \le n. \tag*{(14)} ρ b 2 , b 1 ∼ ρ b 2 , a i , 1 ≤ i ≤ n . (14)
To this end, fix 1 ≤ i ≤ n 1 \le i \le n 1 ≤ i ≤ n . The relation b 1 ≃ B a i b_1 \overset{B}{\simeq} a_i b 1 ≃ B a i implies the existence of curves ℘ 1 \wp_1 ℘ 1 and ℘ 2 \wp_2 ℘ 2 such that
B ( ℘ 1 ( u ) ) = B ( ℘ 2 ( u ) ) ∉ E B , 0 ≤ u < 1 , B(\wp_1(u)) = B(\wp_2(u)) \notin\mathcal{E}_B, \qquad0 \le u < 1, B ( ℘ 1 ( u )) = B ( ℘ 2 ( u )) ∈ / E B , 0 ≤ u < 1 ,
where ℘ 1 \wp_1 ℘ 1 connects b 1 b_1 b 1 to P P P , and ℘ 2 \wp_2 ℘ 2 connects a i a_i a i to P P P . By Lemma 2.25, there exists a curve ℘ 0 \wp_0 ℘ 0 such that
B ( ℘ 0 ( u ) ) = B ( ℘ 1 ( u ) ) = B ( ℘ 2 ( u ) ) , 0 ≤ u ≤ 1 , B(\wp_0(u)) = B(\wp_1(u)) = B(\wp_2(u)), \qquad0 \le u \le1, B ( ℘ 0 ( u )) = B ( ℘ 1 ( u )) = B ( ℘ 2 ( u )) , 0 ≤ u ≤ 1 ,
with ℘ 0 ( 0 ) = b 2 \wp_0(0) = b_2 ℘ 0 ( 0 ) = b 2 and ℘ 0 ( u ) ∈ Ω 2 2 \wp_0(u) \in\Omega_2^2 ℘ 0 ( u ) ∈ Ω 2 2 for 0 ≤ u < 1 0 \le u < 1 0 ≤ u < 1 . Let v = ℘ 0 ( 1 ) v = \wp_0(1) v = ℘ 0 ( 1 ) . We claim that B ′ ( v ) ≠ 0 B'(v) \ne0 B ′ ( v ) = 0 , which, together with Lemma 2.16, gives ρ b 2 , b 1 ∼ ρ b 2 , a i \rho_{b_2,b_1} \sim\rho_{b_2,a_i} ρ b 2 , b 1 ∼ ρ b 2 , a i , as desired. To prove the claim B ′ ( v ) ≠ 0 B'(v) \ne0 B ′ ( v ) = 0 , suppose for contradiction that v v v is a critical point of B B B . Since B B B is injective on its critical set Z ( B ′ ) Z(B') Z ( B ′ ) , we must have v = Q v = Q v = Q . But B ( ℘ 0 ( 1 ) ) = B ( ℘ 1 ( 1 ) ) B(\wp_0(1)) = B(\wp_1(1)) B ( ℘ 0 ( 1 )) = B ( ℘ 1 ( 1 )) implies B ( Q ) = B ( v ) = B ( P ) B(Q) = B(v) = B(P) B ( Q ) = B ( v ) = B ( P ) , contradicting the assumption that P P P and Q Q Q have distinct B B B -values. This finishes the proof of (3.9). By symmetry, we may exchange the positions of a i a_i a i and b j b_j b j to obtain
ρ a 1 , b 2 ∼ ⋯ ∼ ρ a 1 , b m . \rho_{a_1,b_2} \sim\cdots\sim\rho_{a_1,b_m}. ρ a 1 , b 2 ∼ ⋯ ∼ ρ a 1 , b m .
By Remark 2.18(2),
ρ b 2 , a 1 ∼ ⋯ ∼ ρ b m , a 1 . (15) \rho_{b_2,a_1} \sim\cdots\sim\rho_{b_m,a_1}. \tag*{(15)} ρ b 2 , a 1 ∼ ⋯ ∼ ρ b m , a 1 . (15)
By (3.9) ρ b 2 , b 1 ∼ ρ b 2 , a 1 \rho_{b_2,b_1} \sim\rho_{b_2,a_1} ρ b 2 , b 1 ∼ ρ b 2 , a 1 , and similarly, ρ b i , b 1 ∼ ρ b i , a 1 \rho_{b_i,b_1} \sim\rho_{b_i,a_1} ρ b i , b 1 ∼ ρ b i , a 1 , i ≥ 2 i \ge2 i ≥ 2 . This, combined with (3.10), gives
ρ b 2 , b 1 ∼ ⋯ ∼ ρ b m , b 1 . \rho_{b_2,b_1} \sim\cdots\sim\rho_{b_m,b_1}. ρ b 2 , b 1 ∼ ⋯ ∼ ρ b m , b 1 .
By Remark 2.18(2), we have
ρ b 1 , b 2 ∼ ⋯ ∼ ρ b 1 , b m . \rho_{b_1,b_2} \sim\cdots\sim\rho_{b_1,b_m}. ρ b 1 , b 2 ∼ ⋯ ∼ ρ b 1 , b m .
By applying the arguments developed in Case I.2, we obtain an analogue of (3.6), ρ b 2 , b 3 ∼ ⋯ ∼ ρ b 2 , b m ∼ ρ b 2 , b 1 \rho_{b_2,b_3} \sim\cdots\sim\rho_{b_2,b_m} \sim\rho_{b_2,b_1} ρ b 2 , b 3 ∼ ⋯ ∼ ρ b 2 , b m ∼ ρ b 2 , b 1 . This, together with (3.9), imply that
ρ b 2 , b i ∼ ρ b 2 , b 1 ∼ ρ b 2 , a j , i ≥ 1 , i ≠ 2 ; j ≥ 1. \rho_{b_2,b_i} \sim\rho_{b_2,b_1} \sim\rho_{b_2,a_j}, \qquad i \ge1,\ i \ne2;\ j \ge1. ρ b 2 , b i ∼ ρ b 2 , b 1 ∼ ρ b 2 , a j , i ≥ 1 , i = 2 ; j ≥ 1.
That is,
ρ b 2 , b 1 ∼ ρ b 2 , z ′ \rho_{b_2,b_1}\sim\rho_{b_2,z'} ρ b 2 , b 1 ∼ ρ b 2 , z ′
for every z ′ ∈ B − 1 ( w 0 ) ∩ Ω 0 , t + z' \in B^{-1}(w_0)\cap\Omega_{0,t+} z ′ ∈ B − 1 ( w 0 ) ∩ Ω 0 , t + with z ′ ≠ b 2 z'\ne b_2 z ′ = b 2 where ρ b 2 , b 2 = i d \rho_{b_2,b_2}=id ρ b 2 , b 2 = i d . Hence, on Ω 0 , t + ∖ B − 1 ( E B ) \Omega_{0,t+}\setminus B^{-1}(\mathcal{E}_B) Ω 0 , t + ∖ B − 1 ( E B ) , there are exactly two equivalence classes of local inverses of B B B . By Proposition 2.20, the Riemann surface S B , Ω 0 , t + S_{B,\Omega_{0,t+}} S B , Ω 0 , t + admits exactly two components, the identity component S 0 S_0 S 0 and the component S S S of S B , Ω 0 , t + S_{B,\Omega_{0,t+}} S B , Ω 0 , t + containing the point ( b 2 , b 1 ) (b_2,b_1) ( b 2 , b 1 ) .
The general case. Let Ω 1 \Omega_1 Ω 1 be an R R R -component whose boundary contains at least one critical point, say ∗ \ast ∗ . There are two situations to distinguish: Ω 1 \Omega_1 Ω 1 is the unique R R R -component whose boundary contains ∗ \ast ∗ or not.
We first treat the former case. After relabeling, let Ω 2 ′ , ⋯ , Ω N ′ \Omega'_2,\cdots,\Omega'_N Ω 2 ′ , ⋯ , Ω N ′ denote the maximal family of components of { ∣ B ∣ < t } \{|B|<t\} { ∣ B ∣ < t } , other than Ω 1 \Omega_1 Ω 1 , such that their boundaries intersect those of other components and the union ⋃ i = 2 N Ω i ′ ‾ ∖ { ∗ } \bigcup_{i=2}^{N}\overline{\Omega'_i}\setminus\{\ast\} ⋃ i = 2 N Ω i ′ ∖ { ∗ } is connected. This union is like a cactus pup growing from a single node { ∗ } \{\ast\} { ∗ } . Recall that in the penultimate paragraph of Case II.1, we discussed a chain of components. By the discussions there, by letting b 1 b_1 b 1 be the only point in B − 1 ( w 0 ) ∩ Ω 1 B^{-1}(w_0)\cap\Omega_1 B − 1 ( w 0 ) ∩ Ω 1 (which follows from the assumption that Ω 1 \Omega_1 Ω 1 is of multiplicity one), we have
ρ b 1 , Ω 2 ′ ∼ ⋯ ∼ ρ b 1 , Ω N ′ , \rho_{b_1,\Omega'_2}\sim\cdots\sim\rho_{b_1,\Omega'_N}, ρ b 1 , Ω 2 ′ ∼ ⋯ ∼ ρ b 1 , Ω N ′ ,
since each Ω j ′ \Omega'_j Ω j ′ (2 ≤ j ≤ N 2\le j\le N 2 ≤ j ≤ N ) is contained in some chain. Here, for each i i i , ρ b 1 , Ω i ′ \rho_{b_1,\Omega'_i} ρ b 1 , Ω i ′ means
ρ b 1 , z ′ ∼ ρ b 1 , z ′ ′ , z ′ , z ′ ′ ∈ B − 1 ( w 0 ) ∩ Ω i ′ . \rho_{b_1,z'}\sim\rho_{b_1,z''},\qquad z',z''\in B^{-1}(w_0)\cap\Omega'_i. ρ b 1 , z ′ ∼ ρ b 1 , z ′′ , z ′ , z ′′ ∈ B − 1 ( w 0 ) ∩ Ω i ′ .
In this way, we will have that { ( z , w ) ∈ Ω 0 , t + : B ( z ) = B ( w ) ∉ E B } \{(z,w)\in\Omega_{0,t+}:B(z)=B(w)\notin\mathcal{E}_B\} {( z , w ) ∈ Ω 0 , t + : B ( z ) = B ( w ) ∈ / E B } has exactly two components.
It remains to consider the case where Ω 1 \Omega_1 Ω 1 is not the only R R R -component whose boundary contains ∗ \ast ∗ .
Below, we write Ω i \Omega_i Ω i for Ω i , t \Omega_{i,t} Ω i , t to simplify notation. After relabeling, let Ω 1 , ⋯ , Ω m \Omega_1,\cdots,\Omega_m Ω 1 , ⋯ , Ω m be the R R R -components of { ∣ B ∣ < t } \{|B|<t\} { ∣ B ∣ < t } of multiplicity one whose boundaries contain ∗ \ast ∗ , and let Ω m + 1 , ⋯ , Ω N \Omega_{m+1},\cdots,\Omega_N Ω m + 1 , ⋯ , Ω N be the remaining components of { ∣ B ∣ < t } \{|B|<t\} { ∣ B ∣ < t } whose boundaries also contain ∗ \ast ∗ (these are E E E -components). No other component Ω i \Omega_i Ω i has ∗ \ast ∗ on its boundary. Let
W 0 = ⋃ i = 1 N Ω i ‾ . W_0=\bigcup_{i=1}^{N}\overline{\Omega_i}. W 0 = i = 1 ⋃ N Ω i .
Figure 9. Structure of W 0 W_0 W 0
and let W W W be the component of W 0 ∖ { q } W_0\setminus\{q\} W 0 ∖ { q } containing Ω ‾ 1 ∖ { q } \overline{\Omega}_1\setminus\{q\} Ω 1 ∖ { q } . By Theorem 2.26 , the boundaries ∂ Ω i \partial\Omega_i ∂ Ω i for 1 ≤ i ≤ m 1\le i\le m 1 ≤ i ≤ m intersect only at ∗ \ast ∗ :
∂ Ω i ∩ ∂ Ω j = { ∗ } , 1 ≤ i < j ≤ m . \partial\Omega_i\cap\partial\Omega_j=\{\ast\},\qquad1\le i<j\le m. ∂ Ω i ∩ ∂ Ω j = { ∗ } , 1 ≤ i < j ≤ m .
Moreover, each ∂ Ω k \partial\Omega_k ∂ Ω k (1 ≤ k ≤ m 1\le k\le m 1 ≤ k ≤ m ) intersects ∂ Ω j \partial\Omega_j ∂ Ω j only at ∗ \ast ∗ for all j ≥ m + 1 j\ge m+1 j ≥ m + 1 . Let γ \gamma γ be a non-self-intersecting curve connecting w 0 w_0 w 0 to B ( ∗ ) B(\ast) B ( ∗ ) such that
γ ( u ) ∈ t D ∖ E B , 0 ≤ u < 1. \gamma(u)\in tD\setminus E_B,\qquad0\le u<1. γ ( u ) ∈ t D ∖ E B , 0 ≤ u < 1.
By Lemma 2.25, for each 1 ≤ j ≤ N 1\le j\le N 1 ≤ j ≤ N , there is exactly one curve σ j \sigma_j σ j in Ω j \Omega_j Ω j satisfying
B ( σ j ( u ) ) = γ ( u ) , 0 ≤ u < 1 , σ j ( 1 ) = ∗ . B(\sigma_j(u))=\gamma(u),\qquad0\le u<1,\qquad\sigma_j(1)=\ast. B ( σ j ( u )) = γ ( u ) , 0 ≤ u < 1 , σ j ( 1 ) = ∗ .
Write b j = σ j ( 0 ) b_j=\sigma_j(0) b j = σ j ( 0 ) for each j j j .
Consider the components of W ∖ { ∗ } W\setminus\{\ast\} W ∖ { ∗ } . For m + 1 ≤ k ≤ N m+1\le k\le N m + 1 ≤ k ≤ N , let W k W_k W k be the component of W ∖ { ∗ } W\setminus\{\ast\} W ∖ { ∗ } containing Ω k \Omega_k Ω k . Since Ω m + 1 , … , Ω N \Omega_{m+1},\ldots,\Omega_N Ω m + 1 , … , Ω N are E E E -components, each such W k ∩ B − 1 ( w 0 ) W_k\cap B^{-1}(w_0) W k ∩ B − 1 ( w 0 ) contains at least two points. We make the following claims.
Claim 1. For m + 1 ≤ k ≤ N m+1\le k\le N m + 1 ≤ k ≤ N and for z ′ , z ′ ′ ∈ B − 1 ( w 0 ) ∩ W k z',z''\in B^{-1}(w_0)\cap W_k z ′ , z ′′ ∈ B − 1 ( w 0 ) ∩ W k , we have
ρ b 1 , z ′ ∼ ρ b 1 , z ′ ′ . \rho_{b_1,z'}\sim\rho_{b_1,z''}. ρ b 1 , z ′ ∼ ρ b 1 , z ′′ .
Then we write
ρ b 1 , b k ∼ ρ b 1 , W k , m + 1 ≤ k ≤ N . (16) \rho_{b_1,b_k}\sim\rho_{b_1,W_k},\qquad m+1\le k\le N. \tag*{(16)} ρ b 1 , b k ∼ ρ b 1 , W k , m + 1 ≤ k ≤ N . (16)
To verify this, by Corollary 2.28 or Lemma 2.27 , we have the relation ρ b 1 , Ω i , t \rho_{b_1,\Omega_{i,t}} ρ b 1 , Ω i , t for all Ω i , t ⊆ W k \Omega_{i,t} \subseteq W_k Ω i , t ⊆ W k . Fix one such component Ω j , t \Omega_{j,t} Ω j , t in W k W_k W k , and denote it by Ω k , t ∗ \Omega^*_{k,t} Ω k , t ∗ . For every component Ω = Ω i , t ⊆ W k \Omega= \Omega_{i,t} \subseteq W_k Ω = Ω i , t ⊆ W k , there exists a chain of components
Ω 1 , t ′ : = Ω k , t ∗ , Ω 2 , t ′ , ⋯ , Ω m , t ′ = Ω \Omega'_{1,t} := \Omega^*_{k,t}, \quad\Omega'_{2,t}, \quad\cdots, \quad\Omega'_{m,t} = \Omega Ω 1 , t ′ := Ω k , t ∗ , Ω 2 , t ′ , ⋯ , Ω m , t ′ = Ω
such that
∂ Ω 1 , t ′ ∩ ∂ Ω 2 , t ′ ≠ ∅ , ∂ Ω 2 , t ′ ∩ ∂ Ω 3 , t ′ ≠ ∅ , ⋯ , ∂ Ω m − 1 , t ′ ∩ ∂ Ω m , t ′ ≠ ∅ . \partial\Omega'_{1,t} \cap\partial\Omega'_{2,t} \ne\varnothing,\quad\partial\Omega'_{2,t} \cap\partial\Omega'_{3,t} \ne\varnothing,\quad\cdots,\quad\partial\Omega'_{m-1,t} \cap\partial\Omega'_{m,t} \ne\varnothing. ∂ Ω 1 , t ′ ∩ ∂ Ω 2 , t ′ = ∅ , ∂ Ω 2 , t ′ ∩ ∂ Ω 3 , t ′ = ∅ , ⋯ , ∂ Ω m − 1 , t ′ ∩ ∂ Ω m , t ′ = ∅ .
To prove Claim 1, it suffices to show that
ρ b 1 , Ω i , t ′ ∼ ρ b 1 , Ω i + 1 , t ′ , 1 ≤ i ≤ m − 1. \rho_{b_1,\Omega'_{i,t}} \sim\rho_{b_1,\Omega'_{i+1,t}}, \qquad1 \le i \le m-1. ρ b 1 , Ω i , t ′ ∼ ρ b 1 , Ω i + 1 , t ′ , 1 ≤ i ≤ m − 1.
To see this, note that by Theorem 2.26 (1), ∂ Ω i , t ′ ∩ ∂ Ω i + 1 , t ′ \partial\Omega'_{i,t} \cap\partial\Omega'_{i+1,t} ∂ Ω i , t ′ ∩ ∂ Ω i + 1 , t ′ contains exactly one critical point, denoted by Q ′ Q' Q ′ . Let w 0 = B ( b 1 ) w_0 = B(b_1) w 0 = B ( b 1 ) , and let γ \gamma γ be a non-self-intersecting curve connecting w 0 w_0 w 0 to B ( Q ′ ) B(Q') B ( Q ′ ) such that
γ ( u ) ∈ t D ∖ E B , 0 ≤ u < 1. \gamma(u) \in tD \setminus E_B,\qquad0 \le u < 1. γ ( u ) ∈ t D ∖ E B , 0 ≤ u < 1.
By Lemma 2.25, there are two curves, lying in Ω i , t ′ \Omega'_{i,t} Ω i , t ′ and Ω i + 1 , t ′ \Omega'_{i+1,t} Ω i + 1 , t ′ respectively, connecting a point in B − 1 ( w 0 ) B^{-1}(w_0) B − 1 ( w 0 ) to Q ′ Q' Q ′ . Let λ \lambda λ and μ \mu μ be the endpoints of these two curves lying in B − 1 ( w 0 ) B^{-1}(w_0) B − 1 ( w 0 ) ; then λ ≃ B μ \lambda\overset{B}{\simeq} \mu λ ≃ B μ . Applying Lemma 2.16 immediately yields
ρ b 1 , λ ∼ ρ b 1 , μ . \rho_{b_1,\lambda} \sim\rho_{b_1,\mu}. ρ b 1 , λ ∼ ρ b 1 , μ .
Combined with the equivalences ρ b 1 , λ ∼ ρ b 1 , Ω i , t ′ \rho_{b_1,\lambda} \sim\rho_{b_1,\Omega'_{i,t}} ρ b 1 , λ ∼ ρ b 1 , Ω i , t ′ and ρ b 1 , μ ∼ ρ b 1 , Ω i + 1 , t ′ \rho_{b_1,\mu} \sim\rho_{b_1,\Omega'_{i+1,t}} ρ b 1 , μ ∼ ρ b 1 , Ω i + 1 , t ′ , we conclude that
ρ b 1 , Ω i , t ′ ∼ ρ b 1 , Ω i + 1 , t ′ . \rho_{b_1,\Omega'_{i,t}} \sim\rho_{b_1,\Omega'_{i+1,t}}. ρ b 1 , Ω i , t ′ ∼ ρ b 1 , Ω i + 1 , t ′ .
Therefore ρ b 1 , Ω k , t ∗ ∼ ρ b 1 , Ω \rho_{b_1,\Omega^*_{k,t}} \sim\rho_{b_1,\Omega} ρ b 1 , Ω k , t ∗ ∼ ρ b 1 , Ω for every Ω ⊆ W k \Omega\subseteq W_k Ω ⊆ W k . This proves Claim 1.
Let λ \lambda λ be a point in B − 1 ( w 0 ) ∩ W N B^{-1}(w_0) \cap W_N B − 1 ( w 0 ) ∩ W N such that λ ≠ b N \lambda\ne b_N λ = b N . The same method as in the proof of Claim 1 yields the following.
Claim 2. For 1 ≤ i ≤ N − 1 1 \le i \le N-1 1 ≤ i ≤ N − 1 , and for z ′ , z ′ ′ ∈ B − 1 ( w 0 ) ∩ W N z', z'' \in B^{-1}(w_0) \cap W_N z ′ , z ′′ ∈ B − 1 ( w 0 ) ∩ W N , we have
ρ b i , z ′ ∼ ρ b i , z ′ ′ . \rho_{b_i,z'} \sim\rho_{b_i,z''}. ρ b i , z ′ ∼ ρ b i , z ′′ .
Consequently, ρ b i , λ ∼ ρ b i , b N \rho_{b_i,\lambda} \sim\rho_{b_i,b_N} ρ b i , λ ∼ ρ b i , b N .
Claim 3. We have the equivalence chain: ρ b 1 , λ ∼ ρ b 2 , λ ∼ ⋯ ∼ ρ b N , λ \rho_{b_1,\lambda} \sim\rho_{b_2,\lambda} \sim\cdots\sim\rho_{b_N,\lambda} ρ b 1 , λ ∼ ρ b 2 , λ ∼ ⋯ ∼ ρ b N , λ . To prove Claim 3 holds, observe w 0 = B ( b 1 ) = ⋯ = B ( b N ) w_0 = B(b_1) = \cdots= B(b_N) w 0 = B ( b 1 ) = ⋯ = B ( b N ) . By Lemma 2.25, there exist curves γ 0 , γ 1 , ⋯ , γ N \gamma_0, \gamma_1, \cdots, \gamma_N γ 0 , γ 1 , ⋯ , γ N such that γ i ( 0 ) = b i ; γ i ( 1 ) = ∗ \gamma_i(0) = b_i; \gamma_i(1) = * γ i ( 0 ) = b i ; γ i ( 1 ) = ∗ for 1 ≤ i ≤ N 1 \le i \le N 1 ≤ i ≤ N ,
B ( γ 0 ( u ) ) = B ( γ 1 ( u ) ) = ⋯ = B ( γ N ( u ) ) , 0 ≤ u ≤ 1 , B(\gamma_0(u)) = B(\gamma_1(u)) = \cdots= B(\gamma_N(u)), \qquad0 \le u \le1, B ( γ 0 ( u )) = B ( γ 1 ( u )) = ⋯ = B ( γ N ( u )) , 0 ≤ u ≤ 1 ,
and γ 0 ( 0 ) = λ \gamma_0(0)=\lambda γ 0 ( 0 ) = λ . Let λ 0 ( 1 ) = v \lambda_0(1)=v λ 0 ( 1 ) = v . We will soon see that B ′ ( v ) ≠ 0 B'(v)\ne0 B ′ ( v ) = 0 . Clearly v ≠ ∗ v\ne* v = ∗ (otherwise, since both γ 0 \gamma_0 γ 0 and γ N \gamma_N γ N lie in W N W_N W N , they would share the same endpoints and hence belong to the same component of { ∣ B ∣ < t } \{|B|<t\} { ∣ B ∣ < t } . But this contradicts Lemma 2.25, as λ 0 ( 0 ) = λ ≠ b N = λ N ( 0 ) \lambda_0(0)=\lambda\ne b_N=\lambda_N(0) λ 0 ( 0 ) = λ = b N = λ N ( 0 ) ). Since B ( v ) = B ( ∗ ) B(v)=B(*) B ( v ) = B ( ∗ ) and ∗ * ∗ is a critical point, the injectivity of B ∣ Z ( B ′ ) B|_{Z(B')} B ∣ Z ( B ′ ) forces B ′ ( v ) ≠ 0 B'(v)\ne0 B ′ ( v ) = 0 , as desired.
By applying Lemma 2.16 , we get
ρ λ , b 1 ∼ ρ λ , b 2 ∼ ⋯ ∼ ρ λ , b N . \rho_{\lambda,b_1}\sim\rho_{\lambda,b_2}\sim\cdots\sim\rho_{\lambda,b_N}. ρ λ , b 1 ∼ ρ λ , b 2 ∼ ⋯ ∼ ρ λ , b N .
Remark 2.18 (2) then yields ρ b 1 , λ ∼ ρ b 2 , λ ∼ ⋯ ∼ ρ b N , λ \rho_{b_1,\lambda}\sim\rho_{b_2,\lambda}\sim\cdots\sim\rho_{b_N,\lambda} ρ b 1 , λ ∼ ρ b 2 , λ ∼ ⋯ ∼ ρ b N , λ , finishing the proof of Claim 3.
We will use the Claims 1 and 2 to prove the following.
Claim 4. We have ρ b 1 , b 2 ∼ ρ b 1 , b i \rho_{b_1,b_2}\sim\rho_{b_1,b_i} ρ b 1 , b 2 ∼ ρ b 1 , b i for 2 ≤ i ≤ N 2\le i\le N 2 ≤ i ≤ N .
For this, we first show that Claim 4 is equivalent to prove
ρ b N , b 1 ∼ ρ b N , b i , 1 ≤ i ≤ N − 1. \rho_{b_N,b_1}\sim\rho_{b_N,b_i},\qquad1\le i\le N-1. ρ b N , b 1 ∼ ρ b N , b i , 1 ≤ i ≤ N − 1.
reminding that b i ∈ Ω i b_i\in\Omega_i b i ∈ Ω i for each i i i . In fact, by Theorem 2.22 (2), the points b 1 , … , b N b_1,\ldots,b_N b 1 , … , b N are precisely those around q q q satisfying
B ( b 1 ) = ⋯ = B ( b N ) = w 0 , B(b_1)=\cdots=B(b_N)=w_0, B ( b 1 ) = ⋯ = B ( b N ) = w 0 ,
and order B ′ ( q ) = N − 1 \operatorname{order}_{B'}(q)=N-1 order B ′ ( q ) = N − 1 . By Theorem 2.23 , after a suitable permutation, we may arrange these points in anticlockwise order; denote them by b 1 ′ = b 1 , … , b N ′ b'_1=b_1,\ldots,b'_N b 1 ′ = b 1 , … , b N ′ . Moreover, there exist pairwise disjoint non-self-intersecting curves σ 1 , … , σ N \sigma_1,\ldots,\sigma_N σ 1 , … , σ N in Ω 0 , t ∖ B − 1 ( E B ) \Omega_{0,t}\setminus B^{-1}(\mathcal{E}_B) Ω 0 , t ∖ B − 1 ( E B ) such that
B ( σ 1 ( u ) ) = ⋯ = B ( σ N ( u ) ) , 0 ≤ u ≤ 1. B(\sigma_1(u))=\cdots=B(\sigma_N(u)),\qquad0\le u\le1. B ( σ 1 ( u )) = ⋯ = B ( σ N ( u )) , 0 ≤ u ≤ 1.
with σ i ( 0 ) = b i ′ \sigma_i(0)=b'_i σ i ( 0 ) = b i ′ and σ i ( 1 ) = b i + 1 ′ \sigma_i(1)=b'_{i+1} σ i ( 1 ) = b i + 1 ′ for 1 ≤ i ≤ N 1\le i\le N 1 ≤ i ≤ N (where b N + 1 ′ ≡ b 1 ′ b'_{N+1}\equiv b'_1 b N + 1 ′ ≡ b 1 ′ ). Then by Lemma 2.15 , we obtain the following equivalence
ρ σ i ( 0 ) , σ j ( 0 ) ∼ ρ σ i ( 1 ) , σ j ( 1 ) , 1 ≤ i ≤ j ≤ N . \rho_{\sigma_i(0),\sigma_j(0)}\sim\rho_{\sigma_i(1),\sigma_j(1)},\qquad1\le i\le j\le N. ρ σ i ( 0 ) , σ j ( 0 ) ∼ ρ σ i ( 1 ) , σ j ( 1 ) , 1 ≤ i ≤ j ≤ N .
Using this, we obtain the following equivalent conditions:
ρ b 1 , b 2 ∼ ρ b 1 , b i \rho_{b_1,b_2}\sim\rho_{b_1,b_i} ρ b 1 , b 2 ∼ ρ b 1 , b i for 2 ≤ i ≤ N 2\le i\le N 2 ≤ i ≤ N ; that is, ρ b 1 ′ , b 2 ′ ∼ ρ b 1 ′ , b i ′ \rho_{b'_1,b'_2}\sim\rho_{b'_1,b'_i} ρ b 1 ′ , b 2 ′ ∼ ρ b 1 ′ , b i ′ for 2 ≤ i ≤ N 2\le i\le N 2 ≤ i ≤ N ;
for each j j j (1 ≤ j ≤ N 1\le j\le N 1 ≤ j ≤ N ) and m , n ≠ j m,n\ne j m , n = j , ρ b j ′ , b m ′ ∼ ρ b j ′ , b n ′ \rho_{b'_j,b'_m}\sim\rho_{b'_j,b'_n} ρ b j ′ , b m ′ ∼ ρ b j ′ , b n ′ .
Consequently, Claim 4 is equivalent to
ρ b N , b 1 ∼ ρ b N , b i , 1 ≤ i ≤ N − 1. \rho_{b_N,b_1}\sim\rho_{b_N,b_i},\qquad1\le i\le N-1. ρ b N , b 1 ∼ ρ b N , b i , 1 ≤ i ≤ N − 1.
To proceed, by Claims 2 and 3, we immediately have
ρ b 1 , b N ∼ ρ b 2 , b N ∼ ⋯ ∼ ρ b N − 1 , b N . \rho_{b_1,b_N} \sim\rho_{b_2,b_N} \sim\cdots\sim\rho_{b_{N-1},b_N}. ρ b 1 , b N ∼ ρ b 2 , b N ∼ ⋯ ∼ ρ b N − 1 , b N .
By Remark 2.18(2), this gives
ρ b N , b 1 ∼ ρ b N , b 2 ∼ ⋯ ∼ ρ b N , b N − 1 . \rho_{b_N,b_1} \sim\rho_{b_N,b_2} \sim\cdots\sim\rho_{b_N,b_{N-1}}. ρ b N , b 1 ∼ ρ b N , b 2 ∼ ⋯ ∼ ρ b N , b N − 1 .
Then ρ b 1 , b 2 ∼ ρ b 1 , b i \rho_{b_1,b_2} \sim\rho_{b_1,b_i} ρ b 1 , b 2 ∼ ρ b 1 , b i for 2 ≤ i ≤ N 2 \le i \le N 2 ≤ i ≤ N . The proof of Claim 4 is complete.
So far, we have completed the proofs of Claims 1–4. By Claim 4, we have ρ b 1 , b 2 ∼ ρ b 1 , b i \rho_{b_1,b_2} \sim\rho_{b_1,b_i} ρ b 1 , b 2 ∼ ρ b 1 , b i for 2 ≤ i ≤ N 2 \le i \le N 2 ≤ i ≤ N . By (3.11) in Claim 1,
ρ b 1 , b 2 ∼ ⋯ ∼ ρ b 1 , b m ∼ ρ b 1 , b k ∼ ρ b 1 , W k \rho_{b_1,b_2} \sim\cdots\sim\rho_{b_1,b_m} \sim\rho_{b_1,b_k} \sim\rho_{b_1,W_k} ρ b 1 , b 2 ∼ ⋯ ∼ ρ b 1 , b m ∼ ρ b 1 , b k ∼ ρ b 1 , W k
for m + 1 ≤ k ≤ N m+1 \le k \le N m + 1 ≤ k ≤ N . Hence, restricted on Ω 0 , t + \Omega_{0,t+} Ω 0 , t + , B B B admits exactly two equivalence classes of local inverses, the class containing ρ b 1 , b 2 \rho_{b_1,b_2} ρ b 1 , b 2 , and the identity map. Hence Ω 0 , t + \Omega_{0,t+} Ω 0 , t + is purely gluable.
The proof of Theorem 1.6 is finished. □ \square □
Proof of Theorem 1.5
We start with a lemma that follows directly from the definition of a thin Blaschke product.
Lemma 3.4. For a thin Blaschke product ϕ \phi ϕ , { z ∈ D : ϕ ′ ( z ) = 0 , ϕ ( z ) = a } \{z \in\mathbb{D}: \phi'(z) = 0, \phi(z) = a\} { z ∈ D : ϕ ′ ( z ) = 0 , ϕ ( z ) = a } is a finite set for every a a a in D \mathbb{D} D .
Proof. Suppose, to the contrary, that there exists a point a a a in D \mathbb{D} D so that the set { z ∈ D : ϕ ′ ( z ) = 0 , ϕ ( z ) = a } \{z \in\mathbb{D}: \phi'(z) = 0, \phi(z) = a\} { z ∈ D : ϕ ′ ( z ) = 0 , ϕ ( z ) = a } is infinite. Then there exists a sequence of distinct points { p n } n = 1 ∞ \{p_n\}_{n=1}^{\infty} { p n } n = 1 ∞ such that for n ≥ 1 n \ge1 n ≥ 1 ,
ϕ ′ ( p n ) = 0 and ϕ ( p n ) = a . \phi'(p_n) = 0 \qquad\text{and} \qquad\phi(p_n) = a. ϕ ′ ( p n ) = 0 and ϕ ( p n ) = a .
The points p 1 , p 1 , p 2 , p 2 , … p_1,p_1,p_2,p_2,\ldots p 1 , p 1 , p 2 , p 2 , … form a subsequence of zeros of φ a ∘ ϕ \varphi_a \circ\phi φ a ∘ ϕ . Consequently, φ a ∘ ϕ \varphi_a \circ\phi φ a ∘ ϕ cannot be a thin Blaschke product. On the other hand, since ϕ \phi ϕ is a thin Blaschke product, Proposition 2.10 implies that φ a ∘ ϕ \varphi_a \circ\phi φ a ∘ ϕ is also thin, a contradiction. Thus the set is finite for every a ∈ D a \in\mathbb{D} a ∈ D . □ \square □
We now proceed to the proof of Theorem 1.5.
Proof of Theorem 1.5. Suppose B B B is a thin Blaschke product such that B B B is essentially injective on its critical set. That is, there exists a finite subset F F F of Z ( B ′ ) Z(B') Z ( B ′ ) such that B B B is injective on Z ( B ′ ) ∖ F Z(B')\setminus F Z ( B ′ ) ∖ F . To prove that the multiplication operator M B M_B M B is irreducible, by Proposition 2.21 it suffices to show that S B S_B S B has exactly two components.
Case I. First assume that B B B is injective on the set Z ( B ′ ) Z(B') Z ( B ′ ) of its critical points. We will show that S B S_B S B has exactly two components. By Proposition 2.8 , a thin Blaschke product B B B always has infinitely many critical points. Let z 0 ∈ D \ B − 1 ( E B ) z_0 \in\mathbb{D}\backslash B^{-1}(E_B) z 0 ∈ D \ B − 1 ( E B ) and all points in B − 1 ( B ( z 0 ) ) B^{-1}(B(z_0)) B − 1 ( B ( z 0 )) are listed as
z 0 , z 1 , … , z n , … . z_0,z_1,\ldots,z_n,\ldots. z 0 , z 1 , … , z n , … .
For a fixed n n n , there exists an r ∈ ( 0 , 1 ) r \in(0,1) r ∈ ( 0 , 1 ) such that z 0 , … , z n z_0,\ldots,z_n z 0 , … , z n are contained in r D r\mathbb{D} r D , and moreover, we require that r D r\mathbb{D} r D contains at least two critical points of B B B . By Proposition 2.6, for every s s s in ( 0 , 1 ) (0,1) ( 0 , 1 ) , each component Ω \Omega Ω of { ∣ B ∣ < s } \{|B|<s\} { ∣ B ∣ < s } is mapped onto s D s\mathbb{D} s D by the proper restriction B ∣ Ω B|_{\Omega} B ∣ Ω . Let
r ′ = max { ∣ B ( z ) ∣ : ∣ z ∣ ≤ r } , r'=\max\{|B(z)|:|z|\le r\}, r ′ = max { ∣ B ( z ) ∣ : ∣ z ∣ ≤ r } ,
and let Ω 0 \Omega_0 Ω 0 be the component of { ∣ B ∣ < r ′ } \{|B|<r'\} { ∣ B ∣ < r ′ } containing z 0 , z 1 , … , z n z_0,z_1,\ldots,z_n z 0 , z 1 , … , z n . It is routine to check that Ω 0 \Omega_0 Ω 0 is simply connected. By Proposition 2.3 , the restriction B ∣ Ω 0 : Ω 0 → r ′ D B|_{\Omega_0}:\Omega_0\to r'\mathbb{D} B ∣ Ω 0 : Ω 0 → r ′ D is, up to composition with conformal maps, a finite Blaschke product. More precisely, there exist a conformal map F 1 : Ω 0 → D F_1:\Omega_0\to\mathbb{D} F 1 : Ω 0 → D , a finite Blaschke product φ \varphi φ , and a scaling map F 2 ( z ) = r ′ z F_2(z)=r'z F 2 ( z ) = r ′ z such that
B ∣ Ω 0 = F 2 ∘ φ ∘ F 1 . B|_{\Omega_0}=F_2\circ\varphi\circ F_1. B ∣ Ω 0 = F 2 ∘ φ ∘ F 1 .
Moreover, the map
σ ↦ F 1 ∘ σ ∘ F 1 − 1 \sigma\mapsto F_1\circ\sigma\circ F_1^{-1} σ ↦ F 1 ∘ σ ∘ F 1 − 1
gives a one-to-one correspondence between the local inverses of φ \varphi φ and those of B ∣ Ω 0 B|_{\Omega_0} B ∣ Ω 0 . By Theorem 1.6 , we have
ρ z 0 , z 1 ∼ ⋯ ∼ ρ z 0 , z n . \rho_{z_0,z_1}\sim\cdots\sim\rho_{z_0,z_n}. ρ z 0 , z 1 ∼ ⋯ ∼ ρ z 0 , z n .
By the arbitrariness of n n n , we have ρ z 0 , z 1 ∼ ρ z 0 , z j \rho_{z_0,z_1}\sim\rho_{z_0,z_j} ρ z 0 , z 1 ∼ ρ z 0 , z j , j ≥ 1 j\ge1 j ≥ 1 . Thus B B B admits exactly two equivalence classes of local inverses, the class containing ρ z 0 , z 1 \rho_{z_0,z_1} ρ z 0 , z 1 , and the identity map. By Remark 2.18 (1), S B S_B S B has exactly two components.
Case II. In general, we assume B B B is essentially injective on its critical set. That is, there exists a finite set F F F such that B ∣ Z ( B ′ ) \ F B|_{Z(B')\backslash F} B ∣ Z ( B ′ ) \ F is injective. Let p 1 , p 2 , … , p n , … p_1,p_2,\ldots,p_n,\ldots p 1 , p 2 , … , p n , … be all critical points of B B B . Since B B B is essentially injective on its critical set Z ( B ′ ) Z(B') Z ( B ′ ) , there exists a positive integer N 0 N_0 N 0 such that
B ( p n ) ≠ B ( p m ) , m > n ≥ N 0 . (17) B(p_n)\ne B(p_m),\qquad m>n\ge N_0. \tag*{(17)} B ( p n ) = B ( p m ) , m > n ≥ N 0 . (17)
By Lemma 3.4, the set { n : B ( p n ) = B ( p i ) } \{n : B(p_n) = B(p_i)\} { n : B ( p n ) = B ( p i )} is finite for each i i i with 1 ≤ i ≤ N 0 1 \le i \le N_0 1 ≤ i ≤ N 0 . For all sufficiently large n n n , we thus have
B ( p n ) ≠ B ( p i ) , i = 1 , … , N 0 . (18) B(p_n) \ne B(p_i), \quad i=1,\ldots,N_0. \tag*{(18)} B ( p n ) = B ( p i ) , i = 1 , … , N 0 . (18)
Combing (3.12) and (3.13), we may choose an integer N > N 0 N > N_0 N > N 0 such that
B ( p n ) ≠ B ( p k ) , n ≠ k , n ≥ N . (19) B(p_n) \ne B(p_k), \quad n \ne k,\quad n \ge N. \tag*{(19)} B ( p n ) = B ( p k ) , n = k , n ≥ N . (19)
Fix an arbitrary natural number n n n . We can select a closed disk D ‾ ⊆ D \overline{D} \subseteq\mathbb{D} D ⊆ D that contains all critical points p 1 , … , p N p_1,\ldots,p_N p 1 , … , p N and the finite zero sequence z 0 , … , z n z_0,\ldots,z_n z 0 , … , z n . Define
t 0 = sup { ∣ B ( z ) ∣ : z ∈ D ‾ } . t_0=\sup\{|B(z)|:z\in\overline{D}\}. t 0 = sup { ∣ B ( z ) ∣ : z ∈ D } .
Let Ω 1 , t 0 \Omega_{1,t_0} Ω 1 , t 0 denote the component of { ∣ B ∣ < t 0 } \{|B|<t_0\} { ∣ B ∣ < t 0 } that contains D ‾ \overline{D} D ; in particular, every point p 1 , p 2 , … , p N p_1,p_2,\ldots,p_N p 1 , p 2 , … , p N lies in Ω 1 , t 0 \Omega_{1,t_0} Ω 1 , t 0 . Choose a critical point Q ∗ Q^* Q ∗ that lies outside Ω 1 , t 0 \Omega_{1,t_0} Ω 1 , t 0 . We may then pick a value t ∗ t^* t ∗ with t 0 < t ∗ < 1 t_0<t^*<1 t 0 < t ∗ < 1 such that there exists a component Ω \Omega Ω of { ∣ B ∣ < t ∗ } \{|B|<t^*\} { ∣ B ∣ < t ∗ } containing both Ω 1 , t 0 \Omega_{1,t_0} Ω 1 , t 0 and Q ∗ Q^* Q ∗ . Observe that by Lemma 2.6, the restriction B ∣ Ω B|_{\Omega} B ∣ Ω is essentially a finite Blaschke product, with Ω \Omega Ω playing the role of the unit disk D \mathbb{D} D . Techniques for finite Blaschke products remain applicable for B ∣ Ω B|_{\Omega} B ∣ Ω . The following arguments, based on this restricted map, will prove that Ω \Omega Ω is purely gluable.
Consider the family of the components Ω j , t 0 \Omega_{j,t_0} Ω j , t 0 of { z ∈ Ω : ∣ B ( z ) ∣ < t 0 } \{z\in\Omega:|B(z)|<t_0\} { z ∈ Ω : ∣ B ( z ) ∣ < t 0 } indexed by j j j . By slightly increasing t 0 t_0 t 0 while keeping t 0 < t ∗ t_0<t^* t 0 < t ∗ , we may assume that ∂ Ω 1 , t 0 \partial\Omega_{1,t_0} ∂ Ω 1 , t 0 intersects at least one other boundary ∂ Ω k , t 0 \partial\Omega_{k,t_0} ∂ Ω k , t 0 of a component lying in Ω \Omega Ω . Such a t 0 t_0 t 0 exists; otherwise, for every s < t ∗ s<t^* s < t ∗ , the number n s n_s n s of components of { z ∈ Ω : ∣ B ( z ) ∣ < s } \{z\in\Omega:|B(z)|<s\} { z ∈ Ω : ∣ B ( z ) ∣ < s } would satisfy n s ≥ 2 n_s\ge2 n s ≥ 2 . However, by Proposition 2.3 and Lemma 2.4, we have n s = 1 n_s=1 n s = 1 for s s s sufficiently close to t ∗ t^* t ∗ , which yields a contradiction. Hence, such a t 0 t_0 t 0 must exist.
We then take the maximal sub-collection of these boundaries ∂ Ω j , t 0 \partial\Omega_{j,t_0} ∂ Ω j , t 0 whose union is connected and contains ∂ Ω 1 , t 0 \partial\Omega_{1,t_0} ∂ Ω 1 , t 0 . By the finite tree structure of these boundaries ∂ Ω j , t 0 \partial\Omega_{j,t_0} ∂ Ω j , t 0 , it follows that there exist at least two R R R -components. Let Ω 0 , t 0 \Omega_{0,t_0} Ω 0 , t 0 be such a R R R -component distinct from Ω 1 , t 0 \Omega_{1,t_0} Ω 1 , t 0 and remind that Ω 0 , t 0 ⊆ Ω \Omega_{0,t_0}\subseteq\Omega Ω 0 , t 0 ⊆ Ω (see Figure 10). We select Ω 0 , t 0 \Omega_{0,t_0} Ω 0 , t 0 as the R R R -component of maximal multiplicity among all R R R -components distinct from Ω 1 , t 0 \Omega_{1,t_0} Ω 1 , t 0 . Formally, if Ω i 1 , t 0 , … , Ω i K , t 0 \Omega_{i_1,t_0},\ldots,\Omega_{i_K,t_0} Ω i 1 , t 0 , … , Ω i K , t 0 denote all such R R R -components, then Ω 0 , t 0 \Omega_{0,t_0} Ω 0 , t 0 is chosen as any one of them that attains the maximal multiplicity. We now state a key claim.
Claim. The domain Ω 0 , t 0 + \Omega_{0,t_0^+} Ω 0 , t 0 + is purely gluable.
Recall that the notation t 0 + t_0^+ t 0 + refers to a number s > t 0 s>t_0 s > t 0 taken sufficiently close to t 0 t_0 t 0 . The claim can be restated formally as follows: there exists some
Figure 10. An R R R -component Ω 0 , t 0 \Omega_{0,t_0} Ω 0 , t 0 distinct from Ω 1 , t 0 \Omega_{1,t_0} Ω 1 , t 0
ε > 0 \varepsilon> 0 ε > 0 , such that for every s s s with t 0 < s < t 0 + ε t_0 < s < t_0 + \varepsilon t 0 < s < t 0 + ε , the component Ω 0 , s \Omega_{0,s} Ω 0 , s of { ∣ B ∣ < s } \{|B| < s\} { ∣ B ∣ < s } that contains Ω 0 , t 0 \Omega_{0,t_0} Ω 0 , t 0 is purely gluable.
Remind that Ω 0 , t 0 + \Omega_{0,t_0}^{+} Ω 0 , t 0 + contains the tree structure associated with Ω ‾ 0 , t 0 \overline{\Omega}_{0,t_0} Ω 0 , t 0 (or, with Ω ‾ 1 , t 0 \overline{\Omega}_{1,t_0} Ω 1 , t 0 ). We treat some distinct cases for Ω 0 , t 0 \Omega_{0,t_0} Ω 0 , t 0 , corresponding respectively to Step I and Step II in the proof of Theorem 1.6 , and verify the claim in each case.
Step I. Flower model.
Step II. The Cactuses Case.
There are exactly three cases under consideration: Ω 0 , t 0 \Omega_{0,t_0} Ω 0 , t 0 has at least two critical points; Ω 0 , t 0 \Omega_{0,t_0} Ω 0 , t 0 has exactly one critical point; and Ω 0 , t 0 \Omega_{0,t_0} Ω 0 , t 0 has no critical point. These cases are respectively addressed in Steps I and II of the proof of Theorem 1.5 .
We now point out the difference between the current general setting and the earlier special case (Case I of Theorem 1.5 ). The key observation is that under Condition (3.14), for λ , μ ∈ Ω 1 , t \lambda,\mu\in\Omega_{1,t} λ , μ ∈ Ω 1 , t and ω ∈ Ω 0 , t \omega\in\Omega_{0,t} ω ∈ Ω 0 , t with B ( ω ) = B ( λ ) = B ( μ ) B(\omega) = B(\lambda) = B(\mu) B ( ω ) = B ( λ ) = B ( μ ) , Lemma 2.27 yields ρ ω , λ ∼ ρ ω , μ \rho_{\omega,\lambda} \sim\rho_{\omega,\mu} ρ ω , λ ∼ ρ ω , μ . In other words, we only need to substitute Corollary 2.28 with Lemma 2.27 . Then the arguments presented in Steps I and II of Theorem 1.6 proceeds accordingly, ensuring the desired conclusion. That is, Ω 0 , t 0 + \Omega_{0,t_0}^{+} Ω 0 , t 0 + is purely gluable. The proof of the claim is compete.
For any other Ω j , t 0 \Omega_{j,t_0} Ω j , t 0 with closure lying outside the component of { ∣ B ∣ ≤ t 0 } \{|B| \le t_0\} { ∣ B ∣ ≤ t 0 } containing Ω ‾ 1 , t 0 \overline{\Omega}_{1,t_0} Ω 1 , t 0 , the arguments in Steps I and II of Theorem 1.6 apply without modification. Therefore, for such Ω j , t 0 \Omega_{j,t_0} Ω j , t 0 , the corresponding domain Ω j , t 0 + \Omega_{j,t_0}^{+} Ω j , t 0 + either contains at most one critical point or is purely gluable. In conclusion, the induction step at s = t 0 + s = t_0^{+} s = t 0 + holds for all Ω j , s \Omega_{j,s} Ω j , s lying in Ω \Omega Ω .
Following the same procedure as in Steps I and II in the proof of Theorem 1.6, we obtain a sequence t 0 < t 1 < t 2 < ⋯ < t d < t ∗ t_0 < t_1 < t_2 < \cdots< t_d < t^* t 0 < t 1 < t 2 < ⋯ < t d < t ∗ , where for i ≥ 1 i \ge1 i ≥ 1 each t i t_i t i denotes the absolute value of the critical value of B ∣ Ω B|_{\Omega} B ∣ Ω that is strictly larger than t 0 t_0 t 0 . This yields a finite collection of stages:
t 0 < s ≤ t 1 , ⋯ , t d − 1 < s ≤ t d , t d < s ≤ t ∗ . t_0 < s \le t_1,\quad\cdots,\quad t_{d-1} < s \le t_d,\quad t_d < s \le t^*. t 0 < s ≤ t 1 , ⋯ , t d − 1 < s ≤ t d , t d < s ≤ t ∗ .
At each stage, the arguments in Steps I and II of Theorem 1.6 remain applicable with no modification. This holds because, under condition (19) , the problem reduces to the case where B B B is injective on the remaining critical set as Ω 0 , t + \Omega_{0,t^+} Ω 0 , t + has absorbed the critical points p 1 , … , p N p_1,\ldots,p_N p 1 , … , p N . At s = t ∗ s=t^* s = t ∗ , all components merge into a single component Ω \Omega Ω , and the final simply connected domain Ω \Omega Ω is purely gluable. In particular, since z 0 , … , z n z_0,\ldots,z_n z 0 , … , z n lie in Ω \Omega Ω , ρ z 0 , z 1 ∼ ρ z 0 , z i \rho_{z_0,z_1}\sim\rho_{z_0,z_i} ρ z 0 , z 1 ∼ ρ z 0 , z i , 1 ≤ i ≤ n 1\le i\le n 1 ≤ i ≤ n . By the arbitrariness of n n n , ρ z 0 , z 1 ∼ ρ z 0 , z i \rho_{z_0,z_1}\sim\rho_{z_0,z_i} ρ z 0 , z 1 ∼ ρ z 0 , z i , i ≥ 1 i\ge1 i ≥ 1 . This fact in turn implies that the Riemann surface S B S_B S B contains exactly two components.
The proof of Theorem 1.5 is now complete.
Primeness and essential injectiveness
In this section, we show that the essential injectivity of a thin Blaschke product implies its primeness, and also establishes the implications (3) ⇒ \Rightarrow ⇒ (2) ⇒ \Rightarrow ⇒ (1) in Conjecture 1.2.
Remind that in the introduction, we mention the equivalence of the following conditions (I) and (II) for a thin Blaschke product B B B .
(I) The mapping B B B is not essentially injective on its critical set.
(II) There exist two sequences { z n } n = 1 ∞ \{z_n\}_{n=1}^{\infty} { z n } n = 1 ∞ and { w n } n = 1 ∞ \{w_n\}_{n=1}^{\infty} { w n } n = 1 ∞ in D \mathbb{D} D such that
(1) B ( z n ) = B ( w n ) B(z_n)=B(w_n) B ( z n ) = B ( w n ) , and z n ≠ w n z_n\ne w_n z n = w n for all n n n ;
(2) B ′ ( z n ) = B ′ ( w n ) = 0 B'(z_n)=B'(w_n)=0 B ′ ( z n ) = B ′ ( w n ) = 0 ;
(3) B ( z n ) ≠ B ( z m ) B(z_n)\ne B(z_m) B ( z n ) = B ( z m ) if m ≠ n m\ne n m = n .
Proposition 4.1. The equivalence between (I) and (II) holds for a thin Blaschke product B B B .
Proof. First, we prove that (II) implies (I). By (II), there exist two sequences { z n } n = 1 ∞ \{z_n\}_{n=1}^{\infty} { z n } n = 1 ∞ and { w n } n = 1 ∞ \{w_n\}_{n=1}^{\infty} { w n } n = 1 ∞ such that
B ′ ( z n ) = B ′ ( w n ) = 0 and B ( z n ) = B ( w n ) . B'(z_n)=B'(w_n)=0 \quad\text{and}\quad B(z_n)=B(w_n). B ′ ( z n ) = B ′ ( w n ) = 0 and B ( z n ) = B ( w n ) .
Assume, for contradiction, that B B B is essentially injective on its critical set. Then there exists a finite set F 0 F_0 F 0 such that B B B is injective on Z ( B ′ ) ∖ F 0 Z(B')\setminus F_0 Z ( B ′ ) ∖ F 0 . In particular, B B B is injective on
( { z n : n ≥ 1 } ∪ { w n : n ≥ 1 } ) ∖ F 0 , \bigl(\{z_n:n\ge1\}\cup\{w_n:n\ge1\}\bigr)\setminus F_0, ( { z n : n ≥ 1 } ∪ { w n : n ≥ 1 } ) ∖ F 0 ,
which is impossible. This contradiction proves that (II) implies (I).
Next we show that (I) implies (II).
By (I), for any finite subset F F F , B B B is not injective on Z ( B ′ ) ∖ F Z(B')\setminus F Z ( B ′ ) ∖ F . Also recall, by Proposition 2.8, that Z ( B ′ ) ∖ F Z(B')\setminus F Z ( B ′ ) ∖ F is infinite. Specifically, when F = ∅ F=\varnothing F = ∅ , there exist distinct points z 1 , w 1 ∈ D z_1,w_1\in\mathbb{D} z 1 , w 1 ∈ D satisfying
B ( z 1 ) = B ( w 1 ) and B ′ ( z 1 ) = B ′ ( w 1 ) = 0. B(z_1)=B(w_1) \quad\text{and}\quad B'(z_1)=B'(w_1)=0. B ( z 1 ) = B ( w 1 ) and B ′ ( z 1 ) = B ′ ( w 1 ) = 0.
By induction, we have constructed sequences z n z_n z n and w n w_n w n (1 ≤ n ≤ N 1\le n\le N 1 ≤ n ≤ N ) in D \mathbb{D} D with the following properties for 1 ≤ m , n ≤ N 1\le m,n\le N 1 ≤ m , n ≤ N :
B ( z n ) = B ( w n ) B(z_n)=B(w_n) B ( z n ) = B ( w n ) with z n ≠ w n z_n\ne w_n z n = w n ;
B ′ ( z n ) = B ′ ( w n ) = 0 B'(z_n)=B'(w_n)=0 B ′ ( z n ) = B ′ ( w n ) = 0 ;
B ( z n ) ≠ B ( z m ) B(z_n)\ne B(z_m) B ( z n ) = B ( z m ) whenever m ≠ n m\ne n m = n .
By Lemma 3.4, for each n ≥ 1 n\ge1 n ≥ 1 , the set
{ z ∈ D : B ′ ( z ) = 0 , B ( z ) = B ( z n ) } \{z\in\mathbb{D}:B'(z)=0,\ B(z)=B(z_n)\} { z ∈ D : B ′ ( z ) = 0 , B ( z ) = B ( z n )}
is finite. Let
F = { z ∈ D : B ′ ( z ) = 0 , B ( z ) = B ( z n ) for some 1 ≤ n ≤ N } , F=\{z\in\mathbb{D}:B'(z)=0,\ B(z)=B(z_n)\text{ for some }1\le n\le N\}, F = { z ∈ D : B ′ ( z ) = 0 , B ( z ) = B ( z n ) for some 1 ≤ n ≤ N } ,
and F F F is a finite set. Since B B B is not injective on Z ( B ′ ) ∖ F Z(B')\setminus F Z ( B ′ ) ∖ F , there exist critical points z ∗ , w ∗ ∈ Z ( B ′ ) ∖ F z^*,w^*\in Z(B')\setminus F z ∗ , w ∗ ∈ Z ( B ′ ) ∖ F satisfying
B ( z ∗ ) = B ( w ∗ ) and B ′ ( z ∗ ) = B ′ ( w ∗ ) = 0. B(z^*)=B(w^*) \quad\text{and}\quad B'(z^*)=B'(w^*)=0. B ( z ∗ ) = B ( w ∗ ) and B ′ ( z ∗ ) = B ′ ( w ∗ ) = 0.
Setting z N + 1 = z ∗ z_{N+1}=z^* z N + 1 = z ∗ and w N + 1 = w ∗ w_{N+1}=w^* w N + 1 = w ∗ extends the sequences to satisfy all conditions of (II) for n = 1 , … , N + 1 n=1,\ldots,N+1 n = 1 , … , N + 1 , which completes the inductive step. This concludes the proof.
The essential injectivity of a thin Blaschke product implies its primeness.
Proposition 4.2. Let B B B be a thin Blaschke product. If B B B is essentially injective on its critical points, then B B B is prime.
Proof. Suppose B B B is essentially injective on its critical set. To establish that B B B is prime, assume for contradiction that B B B is not prime. Let B = η ∘ u B=\eta\circ u B = η ∘ u , where η \eta η and u u u are inner functions and neither η \eta η nor u u u is a Möbius map. Then by [11 Theorem 4.1] , u u u is a finite Blaschke product of order N ≥ 2 N\ge2 N ≥ 2 . Since B ( z ) = η ∘ u ( z ) B(z)=\eta\circ u(z) B ( z ) = η ∘ u ( z ) ,
B ′ ( z ) = η ′ ( u ( z ) ) ⋅ u ′ ( z ) . (20) B'(z)=\eta'(u(z))\cdot u'(z). \tag*{(20)} B ′ ( z ) = η ′ ( u ( z )) ⋅ u ′ ( z ) . (20)
::: Since u ′ u' u ′ has only finitely many critical points and, by Theorem 2.8, B B B has infinitely many critical points, the function η \eta η must have infinitely many critical points, say, λ 1 , ⋯ , λ n , ⋯ \lambda_1,\cdots,\lambda_n,\cdots λ 1 , ⋯ , λ n , ⋯ . By (20) , for all but finitely many points in E u \mathcal{E}_u E u , each λ j \lambda_j λ j has exactly N N N preimages in u − 1 ( λ j ) u^{-1}(\lambda_j) u − 1 ( λ j ) . Without loss of generality, for j ≥ 1 j \ge1 j ≥ 1 we can pick z j ( 1 ) , z j ( 2 ) ∈ u − 1 ( λ j ) z_j^{(1)},z_j^{(2)} \in u^{-1}(\lambda_j) z j ( 1 ) , z j ( 2 ) ∈ u − 1 ( λ j ) such that z j ( 1 ) ≠ z j ( 2 ) z_j^{(1)} \ne z_j^{(2)} z j ( 1 ) = z j ( 2 ) and
B ′ ( z j ( 1 ) ) = B ′ ( z j ( 2 ) ) = 0 , B'(z_j^{(1)})=B'(z_j^{(2)})=0, B ′ ( z j ( 1 ) ) = B ′ ( z j ( 2 ) ) = 0 ,
B ( z j ( 1 ) ) = B ( z j ( 2 ) ) = η ( λ j ) , j ≥ 1. B(z_j^{(1)})=B(z_j^{(2)})=\eta(\lambda_j),\qquad j\ge1. B ( z j ( 1 ) ) = B ( z j ( 2 ) ) = η ( λ j ) , j ≥ 1.
Since B B B is a thin Blaschke product, using Lemma 3.4 , we obtain that { η ( λ j ) : j ≥ 1 } \{\eta(\lambda_j):j\ge1\} { η ( λ j ) : j ≥ 1 } is an infinite sequence. Passing to a subsequence of { λ j } j = 1 ∞ \{\lambda_j\}_{j=1}^{\infty} { λ j } j = 1 ∞ , we observe that B B B is not essentially injective on its critical points. This contradicts our assumption, thus completing the proof.
To conclude this subsection, we clarify the implications in Conjecture 1.2 stated below.
Proposition 4.3. In Conjecture 1.2, the implications (3) ⇒ \Rightarrow ⇒ (2) ⇒ \Rightarrow ⇒ (1) hold.
Proof. Let B B B be a thin Blaschke product. We will prove the implication chain (3) ⇒ \Rightarrow ⇒ (2) ⇒ \Rightarrow ⇒ (1).
(3) ⇒ \Rightarrow ⇒ (2). By Theorem 5.3.1 in [16 ] , for a thin Blaschke product B B B , V ∗ ( B ) = C I \mathcal{V}^*(B)=CI V ∗ ( B ) = C I if and only if the Riemann surface S B S_B S B has a component S S S of finite multiplicity, other than the identity component S 0 S_0 S 0 . Given that S B S_B S B has exactly two components, the component S S S distinct from S 0 S_0 S 0 must be of infinite multiplicity. Consequently, V ∗ ( B ) = C I \mathcal{V}^*(B)=CI V ∗ ( B ) = C I .
(2) ⇒ \Rightarrow ⇒ (1). Suppose V ∗ ( B ) \mathcal{V}^*(B) V ∗ ( B ) is trivial. We will prove that B B B is prime. Assume conversely that B B B is not prime. By [11 Theorem 4.1] , if B = η ∘ B 0 B=\eta\circ B_0 B = η ∘ B 0 where η \eta η and B 0 B_0 B 0 are inner functions distinct from Möbius maps, then B 0 B_0 B 0 must be a finite Blaschke product. Moreover, B 0 B_0 B 0 is not a Möbius map by non-primeness of B B B . Clearly, we have V ∗ ( B ) ⊇ V ∗ ( B 0 ) \mathcal{V}^*(B)\supseteq\mathcal{V}^*(B_0) V ∗ ( B ) ⊇ V ∗ ( B 0 ) . By [18 ] , V ∗ ( B 0 ) ≠ C I \mathcal{V}^*(B_0)\ne CI V ∗ ( B 0 ) = C I . Thus V ∗ ( B ) ≠ C I \mathcal{V}^*(B)\ne CI V ∗ ( B ) = C I , which is a contradiction.
:::
An example tightening Conjecture 1.2
The following example constructs thin Blaschke products with prescribed properties, and illustrates how to extract subsequences from thin sequences.
Example 5.1. Let { z k } k = 1 ∞ \{z_k\}_{k=1}^{\infty} { z k } k = 1 ∞ be a Blaschke sequence. As follows, we will extract a subsequence { λ k } k = 1 ∞ \{\lambda_k\}_{k=1}^{\infty} { λ k } k = 1 ∞ of { z k } k = 1 ∞ \{z_k\}_{k=1}^{\infty} { z k } k = 1 ∞ which forms a thin Blaschke Blaschke product B B B such that for n ≥ 1 n\ge1 n ≥ 1 , the multiplication operator M B n M_{B^n} M B n is irreducible, and the associated Riemann surface S B n S_{B^n} S B n has precisely n + 1 n+1 n + 1 components.
To define S B n S_{B^n} S B n , note that B n B^n B n is not a thin Blaschke product for n ≥ 2 n \ge2 n ≥ 2 . However, by routine computations, we see that ( B n ) − 1 ( E B n ) (B^n)^{-1}(\mathcal{E}B^n) ( B n ) − 1 ( E B n ) is still a discrete subset of D \mathbb{D} D . Hence S B n S_{B^n} S B n is well-defined.
This example shows that Conjecture 1.2, which concerns a single thin Blaschke product, does not extend to products of finitely many thin Blaschke products.
To demonstrate Example 5.1, we need a lemma from [15 Lemma 5.1] . For a ∈ D a \in\mathbb{D} a ∈ D , let
ψ a ( z ) = ∣ a ∣ a a − z 1 − a ‾ z , z ∈ D . \psi_a(z)=\frac{|a|}{a}\frac{a-z}{1-\overline{a}z},\qquad z\in\mathbb{D}. ψ a ( z ) = a ∣ a ∣ 1 − a z a − z , z ∈ D .
Lemma 5.2. Suppose that B B B is a finite Blaschke product with order greater than 2. Then for any r ∈ ( 0 , 1 ) r \in(0,1) r ∈ ( 0 , 1 ) , there exists s ∈ ( 0 , 1 ) s \in(0,1) s ∈ ( 0 , 1 ) such that for all λ ∈ D \lambda\in\mathbb{D} λ ∈ D with ∣ λ ∣ > s |\lambda|>s ∣ λ ∣ > s , the function B ψ λ B\psi_\lambda B ψ λ has a critical point w ∗ w^* w ∗ satisfying ∣ B ( w ∗ ) ψ λ ( w ∗ ) ∣ > r |B(w^*)\psi_\lambda(w^*)|>r ∣ B ( w ∗ ) ψ λ ( w ∗ ) ∣ > r .
Specifically, for λ \lambda λ with ∣ λ ∣ |\lambda| ∣ λ ∣ sufficiently close to 1, ψ λ \psi_\lambda ψ λ is uniformly close to a unimodular constant on each compact subset of D \mathbb{D} D . By an application of Rouché’s theorem, the critical points of B ψ λ B\psi_\lambda B ψ λ other than w ∗ w^* w ∗ lie in a small neighborhood of critical points of B B B . The “new” critical point w ∗ w^* w ∗ can attain values λ ∗ \lambda^* λ ∗ with ∣ λ ∗ ∣ |\lambda^*| ∣ λ ∗ ∣ arbitrarily close to 1.
The construction of B B B . The construction adopts an approach very similar to the technique employed in the proof of [15 Theorem 5.3] . For clarity, we present the key idea rather than the detailed steps.
Let { z k } k = 1 ∞ \{z_k\}_{k=1}^{\infty} { z k } k = 1 ∞ be an arbitrary Blaschke sequence. Without loss of generality, we may assume that { z k } k = 1 ∞ \{z_k\}_{k=1}^{\infty} { z k } k = 1 ∞ is a thin Blaschke sequence; otherwise, we extract a subsequence to replace { z k } k = 1 ∞ \{z_k\}_{k=1}^{\infty} { z k } k = 1 ∞ . We begin by picking λ 1 \lambda_1 λ 1 and λ 2 \lambda_2 λ 2 from the sequence { z k } k = 1 ∞ \{z_k\}_{k=1}^{\infty} { z k } k = 1 ∞ so that the order-two Blaschke product
B 1 = ψ λ 1 ψ λ 2 B_1=\psi_{\lambda_1}\psi_{\lambda_2} B 1 = ψ λ 1 ψ λ 2
has exactly one critical point distinct from zero. The key step involves constructing a sequence of finite Blaschke products { B n } n = 1 ∞ \{B_n\}_{n=1}^{\infty} { B n } n = 1 ∞ through the recursive relation:
B k + 1 = B k ⋅ ψ λ k B_{k+1}=B_k\cdot\psi_{\lambda_k} B k + 1 = B k ⋅ ψ λ k
where { λ k } k = 1 ∞ \{\lambda_k\}_{k=1}^{\infty} { λ k } k = 1 ∞ is a subsequence of the thin sequence { z k } k = 1 ∞ \{z_k\}_{k=1}^{\infty} { z k } k = 1 ∞ . Hence the limit of { B k } k = 1 ∞ \{B_k\}_{k=1}^{\infty} { B k } k = 1 ∞ exists as a thin Blaschke product. The sequence { B m } m = 1 ∞ \{B_m\}_{m=1}^{\infty} { B m } m = 1 ∞ is accompanied by an increasing sequence { K m } m = 1 ∞ \{K_m\}_{m=1}^{\infty} { K m } m = 1 ∞ of closed domains in D \mathbb{D} D :
K 1 ⊆ K 2 ⊆ K 3 ⊆ ⋯ with ⋃ m = 1 ∞ K m = D , K_1\subseteq K_2\subseteq K_3\subseteq\cdots\qquad\text{with}\qquad\bigcup_{m=1}^{\infty}K_m=\mathbb{D}, K 1 ⊆ K 2 ⊆ K 3 ⊆ ⋯ with m = 1 ⋃ ∞ K m = D ,
and for each m m m , K m K_m K m is contained in the interior of K m + 1 K_{m+1} K m + 1 ; it is required that the interior K m + 1 K_{m+1} K m + 1 is required to contain the first m + 1 m+1 m + 1 critical points of B m + 1 , B m + 2 , ⋯ B_{m+1}, B_{m+2}, \cdots B m + 1 , B m + 2 , ⋯ , and of the limit B B B . For each m m m , the limit product B B B can be written as
B = B m B ~ m , B = B_m\widetilde{B}_m, B = B m B m ,
for an infinite Blaschke product B ~ m \widetilde{B}_m B m . When { ∣ λ k ∣ } k = 1 ∞ \{|\lambda_k|\}_{k=1}^{\infty} { ∣ λ k ∣ } k = 1 ∞ tends rapidly to 1 1 1 , the quantity ∥ B ~ m − 1 ∥ K m , ∞ \|\widetilde{B}_m-1\|_{K_m,\infty} ∥ B m − 1 ∥ K m , ∞ becomes sufficiently small, and
∥ B − B m ∥ K m , ∞ ≤ ∥ B ~ m − 1 ∥ K m , ∞ . \|B-B_m\|_{K_m,\infty} \leq\|\widetilde{B}_m-1\|_{K_m,\infty}. ∥ B − B m ∥ K m , ∞ ≤ ∥ B m − 1 ∥ K m , ∞ .
This error ∥ B − B m ∥ K m , ∞ \|B-B_m\|_{K_m,\infty} ∥ B − B m ∥ K m , ∞ can be controlled so that on K m K_m K m , B B B has exactly the same number of critical points of B m B_m B m , counting multiplicity. By induction, we can ensure that B m B_m B m admits only simple critical points; thus passing to the limit, B B B retains the same property.
By rearranging the first m m m critical points of B i B_i B i (i > m i>m i > m ) in K m K_m K m , we can ensure that they lie in a sufficiently small neighborhood of the critical points of B m B_m B m . Let p m , 1 , ⋯ , p m , m p_{m,1},\cdots,p_{m,m} p m , 1 , ⋯ , p m , m be all critical points of B m B_m B m , and let p i , 1 , ⋯ , p i , m p_{i,1},\cdots,p_{i,m} p i , 1 , ⋯ , p i , m be the only critical points of B i B_i B i on K m K_m K m . By Lemma 5.2, we carefully choose { λ k } k = 1 ∞ \{\lambda_k\}_{k=1}^{\infty} { λ k } k = 1 ∞ so that
∣ B m ( p m , 1 ) ∣ < ∣ B m ( p m , 2 ) ∣ < ⋯ < ∣ B m ( p m , m ) ∣ . |B_m(p_{m,1})| < |B_m(p_{m,2})| < \cdots< |B_m(p_{m,m})|. ∣ B m ( p m , 1 ) ∣ < ∣ B m ( p m , 2 ) ∣ < ⋯ < ∣ B m ( p m , m ) ∣.
Then with fixed m m m , we can find constants c i ′ c'_i c i ′ and c i ′ ′ c''_i c i ′′ satisfying
∣ B m ( p m , 1 ) ∣ < c 1 ′ < c 1 ′ ′ < ∣ B m ( p m , 2 ) ∣ < c 2 ′ < c 2 ′ ′ < ⋯ < c m − 1 ′ < c m − 1 ′ ′ < ∣ B m ( p m , m ) ∣ . \begin{aligned}
|B_m(p_{m,1})| &< c'_1 < c''_1 < |B_m(p_{m,2})| \\
&< c'_2 < c''_2 < \cdots< c'_{m-1} < c''_{m-1} \\
&< |B_m(p_{m,m})|.
\end{aligned} ∣ B m ( p m , 1 ) ∣ < c 1 ′ < c 1 ′′ < ∣ B m ( p m , 2 ) ∣ < c 2 ′ < c 2 ′′ < ⋯ < c m − 1 ′ < c m − 1 ′′ < ∣ B m ( p m , m ) ∣.
The above inequalities holds if each p m , i p_{m,i} p m , i (1 ≤ i ≤ m 1\leq i\leq m 1 ≤ i ≤ m ) is replaced with a sufficiently small closed disk centered at p m , i p_{m,i} p m , i and entirely contained in K m K_m K m . Furthermore, for i > m i>m i > m , the above inequality is maintained:
∣ B i ( p i , 1 ) ∣ < c 1 ′ < c 1 ′ ′ < ∣ B i ( p i , 2 ) ∣ < c 2 ′ < c 2 ′ ′ < ⋯ < c m − 1 ′ < c m − 1 ′ ′ < ∣ B i ( p i , m ) ∣ , \begin{aligned}
|B_i(p_{i,1})| &< c'_1 < c''_1 < |B_i(p_{i,2})| \\
&< c'_2 < c''_2 < \cdots< c'_{m-1} < c''_{m-1} \\
&< |B_i(p_{i,m})|,
\end{aligned} ∣ B i ( p i , 1 ) ∣ < c 1 ′ < c 1 ′′ < ∣ B i ( p i , 2 ) ∣ < c 2 ′ < c 2 ′′ < ⋯ < c m − 1 ′ < c m − 1 ′′ < ∣ B i ( p i , m ) ∣ ,
where each p i , j p_{i,j} p i , j lie in the above closed disk centered at p m , j p_{m,j} p m , j for 1 ≤ j ≤ m 1\leq j\leq m 1 ≤ j ≤ m . By taking a limit, we have
∣ B ( p 1 ) ∣ < ∣ B ( p 2 ) ∣ < ⋯ < ∣ B ( p m ) ∣ , |B(p_1)| < |B(p_2)| < \cdots< |B(p_m)|, ∣ B ( p 1 ) ∣ < ∣ B ( p 2 ) ∣ < ⋯ < ∣ B ( p m ) ∣ ,
where p 1 , p 2 , ⋯ , p m p_1,p_2,\cdots,p_m p 1 , p 2 , ⋯ , p m denotes the only critical points of B B B on K m K_m K m . In such a way, the critical points of B B B can be listed as p 1 , p 2 , ⋯ p_1,p_2,\cdots p 1 , p 2 , ⋯ satisfying
0 < ∣ B ( p 1 ) ∣ < ∣ B ( p 2 ) ∣ < ⋯ . (21) 0<|B(p_1)|<|B(p_2)|<\cdots. \tag*{(21)} 0 < ∣ B ( p 1 ) ∣ < ∣ B ( p 2 ) ∣ < ⋯ . (21)
Consequently, B B B is injective on its critical set Z ( B ′ ) Z(B') Z ( B ′ ) .
Properties of B B B . Next we will show that B B B satisfies the required property by first analyzing S B n \mathcal{S}_{B^n} S B n and then establishing its connection to the irreducibility of M B n M_{B^n} M B n .
It is clear that B B B is injective on critical points. Let ρ \rho ρ be a local inverse of B n B^n B n defined on a subdomain of D ∖ E \mathbb{D}\setminus\mathcal{E} D ∖ E for some discrete set E \mathcal{E} E . To illustrate, we assume n = 2 n=2 n = 2 without loss of generality. Now for a local inverse of B 2 B^2 B 2 , consider
B 2 ( ρ ( z ) ) = B 2 ( z ) , B^2(\rho(z))=B^2(z), B 2 ( ρ ( z )) = B 2 ( z ) ,
which splits into B ( ρ ( z ) ) = B ( z ) B(\rho(z))=B(z) B ( ρ ( z )) = B ( z ) , and B ( ρ ( z ) ) = − B ( z ) B(\rho(z))=-B(z) B ( ρ ( z )) = − B ( z ) . By Theorem 1.5 , the local inverses of B B B (the case B ( ρ ( z ) ) = B ( z ) B(\rho(z))=B(z) B ( ρ ( z )) = B ( z ) ) fall into two equivalence classes, corresponding to the two components of S B \mathcal{S}_B S B : the identity component S 0 \mathcal{S}_0 S 0 and its complement S 1 \mathcal{S}_1 S 1 .
Now consider
B ( ρ ( z ) ) = − B ( z ) . B(\rho(z))=-B(z). B ( ρ ( z )) = − B ( z ) .
In this case, we can establish an analogue of Lemma 2.16.
Lemma 5.3. Let B B B be a thin Blaschke product. Given z 0 , z 1 z_0,z_1 z 0 , z 1 and z 2 z_2 z 2 in D ∖ B − 1 ( E B ) \mathbb{D}\setminus B^{-1}(\mathcal{E}_B) D ∖ B − 1 ( E B ) with − B ( z 0 ) = B ( z 1 ) = B ( z 2 ) -B(z_0)=B(z_1)=B(z_2) − B ( z 0 ) = B ( z 1 ) = B ( z 2 ) , suppose there are three curves σ 0 , σ 1 , σ 2 \sigma_0,\sigma_1,\sigma_2 σ 0 , σ 1 , σ 2 such that
(1) σ i ( 0 ) = z i \sigma_i(0)=z_i σ i ( 0 ) = z i for i = 0 , 1 , 2 i=0,1,2 i = 0 , 1 , 2 , and σ 1 ( 1 ) = σ 2 ( 1 ) \sigma_1(1)=\sigma_2(1) σ 1 ( 1 ) = σ 2 ( 1 ) ;
(2) − B ( σ 0 ( t ) ) = B ( σ 1 ( t ) ) = B ( σ 2 ( t ) ) -B(\sigma_0(t))=B(\sigma_1(t))=B(\sigma_2(t)) − B ( σ 0 ( t )) = B ( σ 1 ( t )) = B ( σ 2 ( t )) for 0 ≤ t ≤ 1 0\leq t\leq1 0 ≤ t ≤ 1 ;
(3) σ i ( t ) ∈ D ∖ B − 1 ( E B ) \sigma_i(t)\in\mathbb{D}\setminus B^{-1}(\mathcal{E}_B) σ i ( t ) ∈ D ∖ B − 1 ( E B ) , i = 0 , 1 , 2 i=0,1,2 i = 0 , 1 , 2 for 0 ≤ t < 1 0\leq t<1 0 ≤ t < 1 ;
(4) B ′ ( σ 0 ( 1 ) ) ≠ 0 B'(\sigma_0(1))\ne0 B ′ ( σ 0 ( 1 )) = 0 .
Then as local inverses of B 2 B^2 B 2 , ρ z 0 , z 1 ∼ ρ z 0 , z 2 \rho_{z_0,z_1}\sim\rho_{z_0,z_2} ρ z 0 , z 1 ∼ ρ z 0 , z 2 .
We now explain the existence of σ 0 , σ 1 , σ 2 \sigma_0,\sigma_1,\sigma_2 σ 0 , σ 1 , σ 2 in our setting. Here z 1 z_1 z 1 and z 2 z_2 z 2 are directly B B B -glued in a component Ω \Omega Ω of { ∣ B ∣ < t ∗ } \{|B|<t^*\} { ∣ B ∣ < t ∗ } , that is, z 1 ∼ B z 2 z_1\mathrel{\overset{B}{\sim}}z_2 z 1 ∼ B z 2 . Then there exist σ 1 , σ 2 \sigma_1,\sigma_2 σ 1 , σ 2 satisfying the conditions in Lemma 5.3 . In particular,
σ i ( u ) ∈ D ∖ B − 1 ( E B ) , 0 ≤ u < 1 ; i = 1 , 2. \sigma_i(u)\in\mathbb{D}\setminus B^{-1}(\mathcal{E}_B),\qquad0\leq u<1;\quad i=1,2. σ i ( u ) ∈ D ∖ B − 1 ( E B ) , 0 ≤ u < 1 ; i = 1 , 2.
Since B ∣ Ω B|_{\Omega} B ∣ Ω is a holomorphic proper map by Proposition 2.6, we can find σ 0 \sigma_0 σ 0 such that
B ( σ 0 ( u ) ) = B ( σ 1 ( u ) ) , 0 ≤ u ≤ 1. B(\sigma_0(u)) = B(\sigma_1(u)), \qquad0 \leq u \leq1. B ( σ 0 ( u )) = B ( σ 1 ( u )) , 0 ≤ u ≤ 1.
With a small perturbation if necessary, we may require
σ 0 ( u ) ∈ D \ B − 1 ( E B ) , 0 ≤ u < 1. \sigma_0(u) \in\mathbb{D}\backslash B^{-1}(\mathcal{E}_B), \qquad0 \leq u < 1. σ 0 ( u ) ∈ D \ B − 1 ( E B ) , 0 ≤ u < 1.
Let B B B be the thin Blaschke product constructed as above, and we will soon see that the condition B ′ ( σ 0 ( 1 ) ) ≠ 0 B'(\sigma_0(1)) \ne0 B ′ ( σ 0 ( 1 )) = 0 itself holds. To reach a contradiction, assume B ′ ( σ 0 ( 1 ) ) = 0 B'(\sigma_0(1)) = 0 B ′ ( σ 0 ( 1 )) = 0 . Either σ 1 ( 1 ) \sigma_1(1) σ 1 ( 1 ) and σ 0 ( 1 ) \sigma_0(1) σ 0 ( 1 ) are equal or not. If σ 1 ( 1 ) = σ 0 ( 1 ) \sigma_1(1) = \sigma_0(1) σ 1 ( 1 ) = σ 0 ( 1 ) , the identity
− B ( σ 0 ( u ) ) = B ( σ 1 ( u ) ) = B ( σ 2 ( u ) ) , 0 ≤ u ≤ 1. -B(\sigma_0(u)) = B(\sigma_1(u)) = B(\sigma_2(u)), \qquad0 \leq u \leq1. − B ( σ 0 ( u )) = B ( σ 1 ( u )) = B ( σ 2 ( u )) , 0 ≤ u ≤ 1.
at 1 gives − B ( σ 0 ( 1 ) ) = B ( σ 0 ( 1 ) ) = 0 -B(\sigma_0(1)) = B(\sigma_0(1)) = 0 − B ( σ 0 ( 1 )) = B ( σ 0 ( 1 )) = 0 , a contradiction to (21) . If σ 1 ( 1 ) \sigma_1(1) σ 1 ( 1 ) and σ 0 ( 1 ) \sigma_0(1) σ 0 ( 1 ) are distinct critical points, the identity
− B ( σ 0 ( u ) ) = B ( σ 1 ( u ) ) = B ( σ 2 ( u ) ) , 0 ≤ u ≤ 1. -B(\sigma_0(u)) = B(\sigma_1(u)) = B(\sigma_2(u)), \qquad0 \leq u \leq1. − B ( σ 0 ( u )) = B ( σ 1 ( u )) = B ( σ 2 ( u )) , 0 ≤ u ≤ 1.
at 1 gives ∣ B ( σ 0 ( 1 ) ) ∣ = ∣ B ( σ 1 ( 1 ) ) ∣ |B(\sigma_0(1))| = |B(\sigma_1(1))| ∣ B ( σ 0 ( 1 )) ∣ = ∣ B ( σ 1 ( 1 )) ∣ . Consequently, this leads to a contradiction with (21) . For each 0 < t < 1 0 < t < 1 0 < t < 1 ,
{ ∣ B 2 ∣ < t 2 } = { ∣ B ∣ < t } . \{|B^2| < t^2\} = \{|B| < t\}. { ∣ B 2 ∣ < t 2 } = { ∣ B ∣ < t } .
Let Ω \Omega Ω be a component of { ∣ B ∣ < t } \{|B| < t\} { ∣ B ∣ < t } and w 0 ∈ Ω w_0 \in\Omega w 0 ∈ Ω . Let w 1 , w 2 , ⋯ , w N w_1, w_2, \cdots, w_N w 1 , w 2 , ⋯ , w N be all points in Ω \Omega Ω such that
− B ( w 0 ) = B ( w 1 ) = ⋯ = B ( w N ) . -B(w_0) = B(w_1) = \cdots= B(w_N). − B ( w 0 ) = B ( w 1 ) = ⋯ = B ( w N ) .
Note that B ∣ Ω B|_{\Omega} B ∣ Ω behaves as a finite Blashke product (see Proposition 2.6). By gluable property of finite Blaschke product combined with Lemma 5.3 , we obtain the equivalence relations:
ρ w 0 , w 1 ∼ ρ w 0 , w 2 ∼ ⋯ ∼ ρ w 0 , w N . \rho_{w_0,w_1} \sim\rho_{w_0,w_2} \sim\cdots\sim\rho_{w_0,w_N}. ρ w 0 , w 1 ∼ ρ w 0 , w 2 ∼ ⋯ ∼ ρ w 0 , w N .
As the parameter t t t increases, the domain Ω \Omega Ω expands, yielding an infinite sequence { w k : k ≥ 1 } \{w_k : k \geq1\} { w k : k ≥ 1 } satisfying B ( w k ) = − B ( w 0 ) B(w_k) = -B(w_0) B ( w k ) = − B ( w 0 ) . Consequently, we derive the extended equivalence:
ρ w 0 , w 1 ∼ ρ w 0 , w i , i ≥ 1. \rho_{w_0,w_1} \sim\rho_{w_0,w_i}, \qquad i \geq1. ρ w 0 , w 1 ∼ ρ w 0 , w i , i ≥ 1.
In other words, the set
{ ( z , w ) : B ( z ) = − B ( w ) , z ∈ D \ E } \{(z,w) : B(z) = -B(w),\ z \in\mathbb{D}\backslash\mathcal{E}\} {( z , w ) : B ( z ) = − B ( w ) , z ∈ D \ E }
is connected as a component of S B 2 S_{B^2} S B 2 , denoted by S 2 S_2 S 2 . Moreover, there exist points { w i ′ } i = 1 ∞ ⊂ D \{w_i'\}_{i=1}^{\infty} \subset\mathbb{D} { w i ′ } i = 1 ∞ ⊂ D such that
− B ( w 0 ) = − B ( w 1 ′ ) = ⋯ = − B ( w N ′ ) = ⋯ . -B(w_0)=-B(w_1')=\cdots=-B(w_N')=\cdots. − B ( w 0 ) = − B ( w 1 ′ ) = ⋯ = − B ( w N ′ ) = ⋯ .
and no other points satisfy this relation. That is,
B ( w 0 ) = B ( w 1 ′ ) = ⋯ = B ( w N ′ ) = ⋯ . B(w_0)=B(w_1')=\cdots=B(w_N')=\cdots. B ( w 0 ) = B ( w 1 ′ ) = ⋯ = B ( w N ′ ) = ⋯ .
By the construction of B B B , B B B is essentially injective on its critical point set. Applying Theorem 1.5 , we obtain
ρ w 0 , w 1 ′ ∼ ρ w 0 , w i ′ , i ≥ 1 , \rho_{w_0,w_1'}\sim\rho_{w_0,w_i'},\qquad i\geq1, ρ w 0 , w 1 ′ ∼ ρ w 0 , w i ′ , i ≥ 1 ,
which are local inverses of B B B (and hence of B 2 B^2 B 2 ). This corresponds to the only two components of S B S_B S B mentioned before, the identity component S 0 S_0 S 0 and S 1 S_1 S 1 .
It remains to discuss irreducibility of M B 2 M_{B^2} M B 2 , which is equivalent to
V ∗ ( M B 2 ) = C I , \mathcal{V}^*(M_{B^2})=\mathbb{C}I, V ∗ ( M B 2 ) = C I ,
where V ∗ ( M B 2 ) = { M B 2 , M B 2 ∗ } \mathcal{V}^*(M_{B^2})=\{M_{B^2},M_{B^2}^*\} V ∗ ( M B 2 ) = { M B 2 , M B 2 ∗ } . For a thin Blaschke product ϕ \phi ϕ , each operator T ∈ V ∗ ( M ϕ ) T\in\mathcal{V}^*(M_\phi) T ∈ V ∗ ( M ϕ ) admits the representation
T f ( z ) = ∑ k = 0 ∞ c k f ( ρ k ( z ) ) ρ k ′ ( z ) Tf(z)=\sum_{k=0}^{\infty}c_kf(\rho_k(z))\rho_k'(z) T f ( z ) = k = 0 ∑ ∞ c k f ( ρ k ( z )) ρ k ′ ( z )
on D \mathbb{D} D minus a discrete subset. Here, the coefficients sequence { c k } k = 1 ∞ ∈ ℓ 2 \{c_k\}_{k=1}^{\infty}\in\ell^2 { c k } k = 1 ∞ ∈ ℓ 2 are uniquely determined, ρ k \rho_k ρ k are local inverses of ϕ \phi ϕ . When ϕ \phi ϕ is replaced by B 2 B^2 B 2 , the same reasoning applies. Following the approach in [15 ] or [16 Theorem 5.3.1] , V ∗ ( M B 2 ) ≠ C I \mathcal{V}^*(M_{B^2})\ne\mathbb{C}I V ∗ ( M B 2 ) = C I if and only if there is a local inverse ρ \rho ρ of finite multiplicity, if and only if S B 2 S_{B^2} S B 2 has a non-identity component of finite multiplicity. From our earlier analysis, V ∗ ( M B 2 ) = C I \mathcal{V}^*(M_{B^2})=\mathbb{C}I V ∗ ( M B 2 ) = C I . Equivalently, M B 2 M_{B^2} M B 2 is irreducible. The same argument extends to M B n M_{B^n} M B n .
The construction of Example 5.1 is finished. □
A prime thin Blaschke product failing essential injectivity
This section will present a thin Blaschke product B B B with the following properties: (1) All zeros of B B B tend to 1. With further analysis, this implies that B B B is prime.
(2) B B B has infinitely many pairs of distinct critical points with the same image, and hence B B B is not essentially injective on its critical set.
This construction yields the following result.
Theorem 6.1. There exists a prime thin Blaschke product B B B that is not essentially injective on its critical set.
Throughout this section, the notations η \eta η , λ \lambda λ , r r r , s s s and P P P , are used with meanings independent of their usage in earlier sections.
Some preparations
Define
η ( z ) = 1 + z 1 − z \eta(z)=\frac{1+z}{1-z} η ( z ) = 1 − z 1 + z
This is a biholomorphic map from D \mathbb{D} D onto the right half-plane
H = { η ∈ C : Re η > 0 } . \mathbb{H}=\{\eta\in\mathbb{C}:\operatorname{Re}\eta>0\}. H = { η ∈ C : Re η > 0 } .
Its inverse is
z = h ( η ) = η − 1 η + 1 z=h(\eta)=\frac{\eta-1}{\eta+1} z = h ( η ) = η + 1 η − 1
For A > 1 A>1 A > 1 , set
a = h ( A ) = A − 1 A + 1 ∈ ( 0 , 1 ) . a=h(A)=\frac{A-1}{A+1}\in(0,1). a = h ( A ) = A + 1 A − 1 ∈ ( 0 , 1 ) .
Then A → ∞ A\to\infty A → ∞ if and only if h ( A ) → 1 h(A)\to1 h ( A ) → 1 . Note that a = a ˉ a=\bar{a} a = a ˉ . The corresponding Blaschke factor then satisfies
a − z 1 − a z = A − η A + η (22) \frac{a-z}{1-az}=\frac{A-\eta}{A+\eta} \tag*{(22)} 1 − a z a − z = A + η A − η (22)
Thus zeros on the interval ( 0 , 1 ) (0,1) ( 0 , 1 ) may be parametrized in the right half-plane by positive real numbers A A A .
To construct a thin Blaschke product on D \mathbb{D} D with desired properties, it suffices to construct one on H \mathbb{H} H .
Unique critical point between consecutive positive real zeros
Assume that { A j } j = 1 ∞ \{A_j\}_{j=1}^{\infty} { A j } j = 1 ∞ is a Blaschke sequence in the right half-plane H \mathbb{H} H , and the positive constants A j A_j A j will be constructed later. Let
P ( η ) = ∏ j = 1 ∞ A j − η A j + η ( η ∈ H ) , P(\eta)=\prod_{j=1}^{\infty}\frac{A_j-\eta}{A_j+\eta}\qquad(\eta\in\mathbb{H}), P ( η ) = j = 1 ∏ ∞ A j + η A j − η ( η ∈ H ) ,
where
0 < A 1 < A 2 < ⋯ . 0<A_1<A_2<\cdots. 0 < A 1 < A 2 < ⋯ .
For η ≠ A j \eta\ne A_j η = A j and η ∈ H \eta\in\mathbb{H} η ∈ H ,
P ′ ( η ) P ( η ) = − ∑ j = 1 ∞ 2 A j A j 2 − η 2 . (23) \frac{P'(\eta)}{P(\eta)}=-\sum_{j=1}^{\infty}\frac{2A_j}{A_j^2-\eta^2}. \tag*{(23)} P ( η ) P ′ ( η ) = − j = 1 ∑ ∞ A j 2 − η 2 2 A j . (23)
In fact, for a finite product P P P , it is routine to verify that (6.2) holds. In general, since P P P and P ′ P' P ′ can be approximated by sequences of finite Blaschke products, (6.2) holds for every η ≠ A j \eta\ne A_j η = A j as well.
Denote the right-hand side of (6.2) by L ( η ) L(\eta) L ( η ) , and
L ( η ) = − ∑ j = 1 ∞ 2 A j A j 2 − η 2 = P ′ ( η ) P ( η ) . (24) L(\eta)=-\sum_{j=1}^{\infty}\frac{2A_j}{A_j^2-\eta^2}=\frac{P'(\eta)}{P(\eta)}. \tag*{(24)} L ( η ) = − j = 1 ∑ ∞ A j 2 − η 2 2 A j = P ( η ) P ′ ( η ) . (24)
Then
L ′ ( η ) = − ∑ j = 1 ∞ 4 A j η ( A j 2 − η 2 ) 2 < 0 , η > 0. (25) L'(\eta)=-\sum_{j=1}^{\infty}\frac{4A_j\eta}{(A_j^2-\eta^2)^2}<0,\qquad\eta>0. \tag*{(25)} L ′ ( η ) = − j = 1 ∑ ∞ ( A j 2 − η 2 ) 2 4 A j η < 0 , η > 0. (25)
Hence L L L is strictly decreasing on each interval ( A j , A j + 1 ) (A_j,A_{j+1}) ( A j , A j + 1 ) . Moreover,
L ( η ) → + ∞ ( η → A j + ) ; L ( η ) → − ∞ ( η → A j + 1 − ) . L(\eta)\to+\infty\quad(\eta\to A_j^+);\qquad L(\eta)\to-\infty\quad(\eta\to A_{j+1}^-). L ( η ) → + ∞ ( η → A j + ) ; L ( η ) → − ∞ ( η → A j + 1 − ) .
By the intermediate value theorem, there is a unique point c j ∈ ( A j , A j + 1 ) c_j\in(A_j,A_{j+1}) c j ∈ ( A j , A j + 1 ) such that L ( c j ) = 0 L(c_j)=0 L ( c j ) = 0 . Since P ( c j ) ≠ 0 P(c_j)\ne0 P ( c j ) = 0 , (6.3) gives P ′ ( c j ) = 0 P'(c_j)=0 P ′ ( c j ) = 0 . Thus there is exactly one real critical point of P P P between any two consecutive positive real zeros A j A_j A j and A j + 1 A_{j+1} A j + 1 .
Furthermore, by (6.4), at such a critical point,
P ′ ′ ( c j ) = P ( c j ) L ′ ( c j ) ≠ 0. P''(c_j)=P(c_j)L'(c_j)\ne0. P ′′ ( c j ) = P ( c j ) L ′ ( c j ) = 0.
Hence these critical points are simple.
A symmetric four-zero block
A four-zero block is a set of four zeros corresponding to a Blaschke product of order four.
Fix λ > 1 \lambda> 1 λ > 1 and consider
ϕ λ ( s , η ) = 1 − η 1 + η λ − η λ + η λ 2 − η λ 2 + η λ 3 s − η λ 3 s + η . (26) \phi_{\lambda}(s,\eta)=\frac{1-\eta}{1+\eta}\frac{\lambda-\eta}{\lambda+\eta}\frac{\lambda^{2}-\eta}{\lambda^{2}+\eta}\frac{\lambda^{3}s-\eta}{\lambda^{3}s+\eta}. \tag*{(26)} ϕ λ ( s , η ) = 1 + η 1 − η λ + η λ − η λ 2 + η λ 2 − η λ 3 s + η λ 3 s − η . (26)
where s > 0 s > 0 s > 0 . When s = 1 s = 1 s = 1 , the four zeros are
1 , λ , λ 2 , λ 3 . 1,\quad\lambda,\quad\lambda^{2},\quad\lambda^{3}. 1 , λ , λ 2 , λ 3 .
For j = 0 , 1 , 2 , 3 j = 0,1,2,3 j = 0 , 1 , 2 , 3 ,
λ j − ( λ 3 / η ) λ j + ( λ 3 / η ) = − λ 3 − j − η λ 3 − j + η , \frac{\lambda^{j}-(\lambda^{3}/\eta)}{\lambda^{j}+(\lambda^{3}/\eta)}
=-\frac{\lambda^{3-j}-\eta}{\lambda^{3-j}+\eta}, λ j + ( λ 3 / η ) λ j − ( λ 3 / η ) = − λ 3 − j + η λ 3 − j − η ,
and then by (26) we obtain
ϕ λ ( 1 , λ 3 η ) = ϕ λ ( 1 , η ) . (27) \phi_{\lambda}\left(1,\frac{\lambda^{3}}{\eta}\right)=\phi_{\lambda}(1,\eta). \tag*{(27)} ϕ λ ( 1 , η λ 3 ) = ϕ λ ( 1 , η ) . (27)
Let x λ x_{\lambda} x λ denote the unique critical point in ( 1 , λ ) (1,\lambda) ( 1 , λ ) . By (27) ,
y λ = λ 3 x λ ∈ ( λ 2 , λ 3 ) y_{\lambda}=\frac{\lambda^{3}}{x_{\lambda}}\in(\lambda^{2},\lambda^{3}) y λ = x λ λ 3 ∈ ( λ 2 , λ 3 )
is also a critical point, and satisfies
ϕ λ ( 1 , x λ ) = ϕ λ ( 1 , y λ ) . (28) \phi_{\lambda}(1,x_{\lambda})=\phi_{\lambda}(1,y_{\lambda}). \tag*{(28)} ϕ λ ( 1 , x λ ) = ϕ λ ( 1 , y λ ) . (28)
Thus a single four-zero block already produces two distinct critical points with the same critical value.
The role of the perturbation parameter s s s
If infinitely many four-zero blocks are multiplied together, the remaining blocks would perturb both the critical points and the critical values of any fixed block, thereby destroying the exact identity (28) in general. In the following, the parameter s > 0 s > 0 s > 0 , chosen sufficiently close to 1, is introduced to absorb this perturbation.
To be precise, for s > 1 λ s > \frac{1}{\lambda} s > λ 1 , let c 1 ( s ) c_{1}(s) c 1 ( s ) and c 3 ( s ) c_{3}(s) c 3 ( s ) be the unique real critical points of ϕ λ ( s , ⋅ ) \phi_{\lambda}(s,\cdot) ϕ λ ( s , ⋅ ) in the intervals
( 1 , λ ) and ( λ 2 , λ 3 s ) , (1,\lambda)\quad\text{and}\quad(\lambda^{2},\lambda^{3}s), ( 1 , λ ) and ( λ 2 , λ 3 s ) ,
respectively; the critical point c 2 ( s ) c_{2}(s) c 2 ( s ) in ( λ , λ 2 ) (\lambda,\lambda^{2}) ( λ , λ 2 ) is not needed here. The function ϕ λ ( s , ⋅ ) \phi_{\lambda}(s,\cdot) ϕ λ ( s , ⋅ ) is zero at each interval endpoint and negative in its interior. Hence these critical values are precisely the minima:
m 1 ( s ) = min η ∈ [ 1 , λ ] ϕ λ ( s , η ) , m 3 ( s ) = min η ∈ [ λ 2 , λ 3 s ] ϕ λ ( s , η ) . m_{1}(s)=\min_{\eta\in[1,\lambda]}\phi_{\lambda}(s,\eta),\qquad
m_{3}(s)=\min_{\eta\in[\lambda^{2},\lambda^{3}s]}\phi_{\lambda}(s,\eta). m 1 ( s ) = η ∈ [ 1 , λ ] min ϕ λ ( s , η ) , m 3 ( s ) = η ∈ [ λ 2 , λ 3 s ] min ϕ λ ( s , η ) .
Define
D λ ( s ) = ϕ λ ( s , c 1 ( s ) ) − ϕ λ ( s , c 3 ( s ) ) , D_{\lambda}(s)=\phi_{\lambda}(s,c_1(s))-\phi_{\lambda}(s,c_3(s)), D λ ( s ) = ϕ λ ( s , c 1 ( s )) − ϕ λ ( s , c 3 ( s )) ,
and we have
D λ ( s ) = m 1 ( s ) − m 3 ( s ) . D_{\lambda}(s)=m_1(s)-m_3(s). D λ ( s ) = m 1 ( s ) − m 3 ( s ) .
In particular, D λ ( 1 ) = 0 D_{\lambda}(1)=0 D λ ( 1 ) = 0 by (6.7).
Proposition 6.2. For every λ > 1 \lambda>1 λ > 1 , we have that
{ D λ ( s ) < 0 , if 1 λ < s < 1 , D λ ( s ) > 0 , if s > 1. \begin{cases}
D_{\lambda}(s)<0, & \text{if }\frac{1}{\lambda}<s<1,\\
D_{\lambda}(s)>0, & \text{if }s>1.
\end{cases} { D λ ( s ) < 0 , D λ ( s ) > 0 , if λ 1 < s < 1 , if s > 1. Proof. Rewrite a = λ 3 a=\lambda^3 a = λ 3 . For 0 < t < min { a , a s } 0<t<\min\{a,as\} 0 < t < min { a , a s } define
H s ( t ) = ( a s − t ) ( a + t ) ( a s + t ) ( a − t ) . H_s(t)=\frac{(as-t)(a+t)}{(as+t)(a-t)}. H s ( t ) = ( a s + t ) ( a − t ) ( a s − t ) ( a + t ) . Only the fourth factor λ 3 s − η λ 3 s + η \frac{\lambda^3s-\eta}{\lambda^3s+\eta} λ 3 s + η λ 3 s − η of (6.5) changes with s s s , and we have
ϕ λ ( s , t ) = H s ( t ) ϕ λ , 1 ( t ) . \phi_{\lambda}(s,t)=H_s(t)\phi_{\lambda,1}(t). ϕ λ ( s , t ) = H s ( t ) ϕ λ , 1 ( t ) . We first determine the monotonicity of H s H_s H s . Subtracting one gives
H s ( t ) − 1 = ( a s − t ) ( a + t ) − ( a s + t ) ( a − t ) ( a s + t ) ( a − t ) = 2 a t ( s − 1 ) ( a s + t ) ( a − t ) . H_s(t)-1=\frac{(as-t)(a+t)-(as+t)(a-t)}{(as+t)(a-t)}=\frac{2at(s-1)}{(as+t)(a-t)}. H s ( t ) − 1 = ( a s + t ) ( a − t ) ( a s − t ) ( a + t ) − ( a s + t ) ( a − t ) = ( a s + t ) ( a − t ) 2 a t ( s − 1 ) . By direct computations,
H s ′ ( t ) = d H s ( t ) d t = 2 a ( s − 1 ) ( a 2 s + t 2 ) ( a s + t ) 2 ( a − t ) 2 . H'_s(t)=\frac{dH_s(t)}{dt}=\frac{2a(s-1)(a^2s+t^2)}{(as+t)^2(a-t)^2}. H s ′ ( t ) = d t d H s ( t ) = ( a s + t ) 2 ( a − t ) 2 2 a ( s − 1 ) ( a 2 s + t 2 ) . Thus, throughout the domain ( 0 , min { a , a s } ) (0,\min\{a,as\}) ( 0 , min { a , a s }) , H s H_s H s is strictly increasing if s > 1 s>1 s > 1 and strictly decreasing if s < 1 s<1 s < 1 .
Suppose first that s > 1 s>1 s > 1 . Choose x ∈ ( 1 , λ ) x\in(1,\lambda) x ∈ ( 1 , λ ) with ϕ λ ( s , x ) = m 1 ( s ) \phi_{\lambda}(s,x)=m_1(s) ϕ λ ( s , x ) = m 1 ( s ) and set y = a x y=\frac{a}{x} y = x a . Then
λ 2 < y < a < a s and 0 < x < y < a . \lambda^2<y<a<as \quad\text{and}\quad0<x<y<a. λ 2 < y < a < a s and 0 < x < y < a . In particular, y y y lies in the third zero interval. Monotonicity gives H s ( x ) < H s ( y ) H_s(x)<H_s(y) H s ( x ) < H s ( y ) , while the equality (6.6) gives ϕ λ ( 1 , x ) = ϕ λ ( 1 , y ) < 0 \phi_{\lambda}(1,x)=\phi_{\lambda}(1,y)<0 ϕ λ ( 1 , x ) = ϕ λ ( 1 , y ) < 0 . Hence, by ϕ λ ( s , t ) = H s ( t ) ϕ λ ( 1 , t ) \phi_{\lambda}(s,t)=H_s(t)\phi_{\lambda}(1,t) ϕ λ ( s , t ) = H s ( t ) ϕ λ ( 1 , t ) we obtain
ϕ λ ( s , y ) < ϕ λ ( s , x ) . \phi_{\lambda}(s,y)<\phi_{\lambda}(s,x). ϕ λ ( s , y ) < ϕ λ ( s , x ) .
Therefore
m 3 ( s ) = min η ∈ [ λ 2 , a s ] ϕ λ ( s , η ) ≤ ϕ λ ( s , y ) < ϕ λ ( s , x ) = m 1 ( s ) . m_3(s)=\min_{\eta\in[\lambda^2,as]}\phi_\lambda(s,\eta)\leq\phi_\lambda(s,y)<\phi_\lambda(s,x)=m_1(s). m 3 ( s ) = η ∈ [ λ 2 , a s ] min ϕ λ ( s , η ) ≤ ϕ λ ( s , y ) < ϕ λ ( s , x ) = m 1 ( s ) .
Hence D λ ( s ) > 0 D_\lambda(s)>0 D λ ( s ) > 0 .
Now let 1 λ < s < 1 \frac{1}{\lambda}<s<1 λ 1 < s < 1 . Choosing y ∈ ( λ 2 , a s ) y\in(\lambda^2,as) y ∈ ( λ 2 , a s ) with ϕ λ ( s , y ) = m 3 ( s ) \phi_\lambda(s,y)=m_3(s) ϕ λ ( s , y ) = m 3 ( s ) and put x = a y x=\frac{a}{y} x = y a , we have
1 < 1 s < x < λ , 0 < x < y < a s < a . 1<\frac{1}{s}<x<\lambda,\qquad0<x<y<as<a. 1 < s 1 < x < λ , 0 < x < y < a s < a .
Thus both x x x and y y y lie in the domain ( 0 , min { a , a s } ) (0,\min\{a,as\}) ( 0 , min { a , a s }) where H s H_s H s is strictly decreasing. Consequently H s ( x ) > H s ( y ) H_s(x)>H_s(y) H s ( x ) > H s ( y ) . Multiplying again by the common negative value ϕ λ ( 1 , x ) = ϕ λ ( 1 , y ) \phi_\lambda(1,x)=\phi_\lambda(1,y) ϕ λ ( 1 , x ) = ϕ λ ( 1 , y ) gives
ϕ λ ( s , x ) < ϕ λ ( s , y ) . \phi_\lambda(s,x)<\phi_\lambda(s,y). ϕ λ ( s , x ) < ϕ λ ( s , y ) .
It follows that
m 1 ( s ) = min η ∈ [ 1 , λ ] ϕ λ ( s , η ) ≤ ϕ λ ( s , x ) < ϕ λ ( s , y ) = m 3 ( s ) , m_1(s)=\min_{\eta\in[1,\lambda]}\phi_\lambda(s,\eta)\leq\phi_\lambda(s,x)<\phi_\lambda(s,y)=m_3(s), m 1 ( s ) = η ∈ [ 1 , λ ] min ϕ λ ( s , η ) ≤ ϕ λ ( s , x ) < ϕ λ ( s , y ) = m 3 ( s ) ,
which immediately gives D λ ( s ) < 0 D_\lambda(s)<0 D λ ( s ) < 0 . □ \square □
For the construction, choose an increasing sequence λ k ≥ 2 \lambda_k\geq2 λ k ≥ 2 with λ k → ∞ \lambda_k\to\infty λ k → ∞ . Fix, once and for all,
ε 0 = 1 4 , I = [ 3 4 , 5 4 ] . \varepsilon_0=\frac{1}{4},\qquad I=\left[\frac{3}{4},\frac{5}{4}\right]. ε 0 = 4 1 , I = [ 4 3 , 4 5 ] .
Since 3 4 > 1 λ k \frac{3}{4}>\frac{1}{\lambda_k} 4 3 > λ k 1 for every k k k , Proposition 6.2 gives
D λ k ( 1 − ε 0 ) < 0 , D λ k ( 1 + ε 0 ) > 0 ( k ≥ 1 ) . (29) D_{\lambda_k}(1-\varepsilon_0)<0,\qquad D_{\lambda_k}(1+\varepsilon_0)>0\qquad(k\geq1). \tag*{(29)} D λ k ( 1 − ε 0 ) < 0 , D λ k ( 1 + ε 0 ) > 0 ( k ≥ 1 ) . (29)
This is one common ε 0 \varepsilon_0 ε 0 . Let { λ k } k = 1 ∞ \{\lambda_k\}_{k=1}^{\infty} { λ k } k = 1 ∞ be any increasing sequence tending to infinity with λ 1 > 2 \lambda_1>2 λ 1 > 2 . Then 0 < ε 0 < 1 − 1 λ k 0<\varepsilon_0<1-\frac{1}{\lambda_k} 0 < ε 0 < 1 − λ k 1 holds for all k k k .
Patching together infinitely many four-zero blocks
In what follows, we will construct a sequence of four-zero blocks to build an infinite Blaschke product satisfying our requirements.
Let the four zeros in the k k k th block J k J_k J k be
R k , R k λ k , R k λ k 2 , R k λ k 3 s k , (30) R_k,\qquad R_k\lambda_k,\qquad R_k\lambda_k^2,\qquad R_k\lambda_k^3s_k, \tag*{(30)} R k , R k λ k , R k λ k 2 , R k λ k 3 s k , (30)
with s k ∈ I s_k \in I s k ∈ I . Set
Λ k = λ k 3 ( 1 + ε 0 ) = 5 4 λ k 3 , and J k = [ R k , Λ k R k ] . \Lambda_k = \lambda_k^3(1+\varepsilon_0)=\frac{5}{4}\lambda_k^3,\qquad\text{and}\qquad J_k=[R_k,\Lambda_kR_k]. Λ k = λ k 3 ( 1 + ε 0 ) = 4 5 λ k 3 , and J k = [ R k , Λ k R k ] .
The interval J k J_k J k contains every possible zero of block k k k and is independent of the actual value of s k s_k s k . Later, the scales R k R_k R k will be chosen to increase sufficiently rapidly so that distinct blocks are widely separated. Remind that
ϕ λ ( s , η ) = 1 − η λ − η λ 2 − η λ 3 s − η 1 + η λ + η λ 2 + η λ 3 s + η , \phi_\lambda(s,\eta)=\frac{1-\eta\lambda-\eta\lambda^2-\eta\lambda^3s-\eta}{1+\eta\lambda+\eta\lambda^2+\eta\lambda^3s+\eta}, ϕ λ ( s , η ) = 1 + η λ + η λ 2 + η λ 3 s + η 1 − η λ − η λ 2 − η λ 3 s − η ,
and for each s = ( s 1 , s 2 , … ) ∈ I ∞ \mathbf{s}=(s_1,s_2,\ldots)\in I^\infty s = ( s 1 , s 2 , … ) ∈ I ∞ , define
P s ( η ) = ∏ k = 1 ∞ ϕ λ k ( s k , η R k ) , η ∈ H . (31) P_{\mathbf{s}}(\eta)=\prod_{k=1}^{\infty}\phi_{\lambda_k}\left(s_k,\frac{\eta}{R_k}\right),\qquad\eta\in\mathbb{H}. \tag*{(31)} P s ( η ) = k = 1 ∏ ∞ ϕ λ k ( s k , R k η ) , η ∈ H . (31)
We first fix the error term appearing in (6.8). Note that the numbers
μ k = 1 2 min { − D λ k ( 1 − ε 0 ) , D λ k ( 1 + ε 0 ) } > 0 (32) \mu_k=\frac{1}{2}\min\{-D_{\lambda_k}(1-\varepsilon_0),D_{\lambda_k}(1+\varepsilon_0)\}>0 \tag*{(32)} μ k = 2 1 min { − D λ k ( 1 − ε 0 ) , D λ k ( 1 + ε 0 )} > 0 (32)
are well defined. Put
β k = min { μ k 8 , 2 − k − 4 } > 0. \beta_k=\min\left\{\frac{\mu_k}{8},2^{-k-4}\right\}>0. β k = min { 8 μ k , 2 − k − 4 } > 0.
The errors μ k \mu_k μ k and β k \beta_k β k need not be bounded away from zero. The two intervals associated with block k k k are
J k , 1 = [ R k , R k λ k ] , J k , 3 = [ R k λ k 2 , R k λ k 3 s k ] , J_{k,1}=[R_k,R_k\lambda_k],\qquad J_{k,3}=[R_k\lambda_k^2,R_k\lambda_k^3s_k], J k , 1 = [ R k , R k λ k ] , J k , 3 = [ R k λ k 2 , R k λ k 3 s k ] ,
both contained in J k J_k J k . The functions ϕ λ k ( s k , η R k ) \phi_{\lambda_k}(s_k,\frac{\eta}{R_k}) ϕ λ k ( s k , R k η ) and P s P_{\mathbf{s}} P s vanish at the endpoints and are negative inside, since the other blocks have positive product there. By Subsection 6.1, P s P_{\mathbf{s}} P s has exactly one critical point in either open interval, and its value is the interval minimum. Let c 1 ( k ) ( s ) c_1^{(k)}(\mathbf{s}) c 1 ( k ) ( s ) and c 3 ( k ) ( s ) c_3^{(k)}(\mathbf{s}) c 3 ( k ) ( s ) denote the critical points of P s P_{\mathbf{s}} P s in J k , 1 J_{k,1} J k , 1 and J k , 3 J_{k,3} J k , 3 , respectively. Let Φ k ( s ) \Phi_k(\mathbf{s}) Φ k ( s ) denote the difference between the corresponding two critical values of P s P_{\mathbf{s}} P s :
Φ k ( s ) = P s ( c 1 ( k ) ( s ) ) − P s ( c 3 ( k ) ( s ) ) = min η ∈ J k , 1 P s ( η ) − min η ∈ J k , 3 P s ( η ) . \Phi_k(\mathbf{s})=P_{\mathbf{s}}\left(c_1^{(k)}(\mathbf{s})\right)-P_{\mathbf{s}}\left(c_3^{(k)}(\mathbf{s})\right)=\min_{\eta\in J_{k,1}}P_{\mathbf{s}}(\eta)-\min_{\eta\in J_{k,3}}P_{\mathbf{s}}(\eta). Φ k ( s ) = P s ( c 1 ( k ) ( s ) ) − P s ( c 3 ( k ) ( s ) ) = η ∈ J k , 1 min P s ( η ) − η ∈ J k , 3 min P s ( η ) .
We have the following key result.
Theorem 6.3. With a fixed increasing sequence { λ k } k = 1 ∞ \{\lambda_k\}_{k=1}^{\infty} { λ k } k = 1 ∞ satisfying λ 1 > 2 \lambda_1>2 λ 1 > 2 , there exists a carefully chosen infinite sequence { R k } k = 1 ∞ \{R_k\}_{k=1}^{\infty} { R k } k = 1 ∞ , associated with { λ k } k = 1 ∞ \{\lambda_k\}_{k=1}^{\infty} { λ k } k = 1 ∞ as in (30) , such that
\begin{aligned}
\Phi_k(\mathbf{s}) &\leq-\mu_k < 0 && \text{when } s_k=1-\varepsilon_0,\\
\Phi_k(\mathbf{s}) &\geq\mu_k > 0 && \text{when } s_k=1+\varepsilon_0,
\end{aligned}
\qquad\varepsilon_0=\frac{1}{4}. \tag*{(33)}\end{aligned}
where s = ( s 1 , s 2 , ⋯ ) ∈ I ∞ \mathbf{s}=(s_1,s_2,\cdots)\in I^\infty s = ( s 1 , s 2 , ⋯ ) ∈ I ∞ .
Note that μ k \mu_k μ k , defined in (32) , depends only on λ k \lambda_k λ k , not on R k R_k R k . Later, we will see that, by the construction, the infinite sequence (30) yields a thin Blaschke sequences.
To prove Theorem 6.3, we begin with two facts.
Fact 1. If 0 ≤ t j ≤ 1 0\leq t_j\leq1 0 ≤ t j ≤ 1 holds for each j j j , then
0 ≤ 1 − ∏ j = 1 m t j ≤ ∑ j = 1 m ( 1 − t j ) . (34) 0\leq1-\prod_{j=1}^{m}t_j\leq\sum_{j=1}^{m}(1-t_j). \tag*{(34)} 0 ≤ 1 − j = 1 ∏ m t j ≤ j = 1 ∑ m ( 1 − t j ) . (34) This also holds for m = ∞ m=\infty m = ∞ .
Fact 2. Let both f f f and g g g be continuous real-valued functions on a compact interval J J J . Then
∣ min J f − min J g ∣ ≤ sup J ∣ f − g ∣ . (35) \left|\min_J f-\min_J g\right|\leq\sup_J|f-g|. \tag*{(35)} J min f − J min g ≤ J sup ∣ f − g ∣. (35)
For completeness, note that (34) follows from the identity
1 − t 1 ⋯ t m = ( 1 − t 1 ) + t 1 ( 1 − t 2 ) + ⋯ + t 1 ⋯ t m − 1 ( 1 − t m ) , 1-t_1\cdots t_m=(1-t_1)+t_1(1-t_2)+\cdots+t_1\cdots t_{m-1}(1-t_m), 1 − t 1 ⋯ t m = ( 1 − t 1 ) + t 1 ( 1 − t 2 ) + ⋯ + t 1 ⋯ t m − 1 ( 1 − t m ) ,
and every coefficient of 1 − t j 1-t_j 1 − t j is at most one. To verify Fact 2, let z 0 z_0 z 0 be a point in J J J satisfying g ( z 0 ) = min J g g(z_0)=\min_J g g ( z 0 ) = min J g . Then
min J f − min J g ≤ f ( z 0 ) − g ( z 0 ) ≤ sup J ∣ f − g ∣ . \min_J f-\min_J g\leq f(z_0)-g(z_0)\leq\sup_J|f-g|. J min f − J min g ≤ f ( z 0 ) − g ( z 0 ) ≤ J sup ∣ f − g ∣.
Similarly, min J g − min J f ≤ sup J ∣ f − g ∣ \min_J g-\min_J f\leq\sup_J|f-g| min J g − min J f ≤ sup J ∣ f − g ∣ . We therefore obtain (35) .
Before proving Theorem 6.3, we first introduce the construction of the sequence { R k } k = 1 ∞ \{R_k\}_{k=1}^{\infty} { R k } k = 1 ∞ mentioned above. Given that { λ j } j = 1 ∞ \{\lambda_j\}_{j=1}^{\infty} { λ j } j = 1 ∞ is a fixed increasing sequence with λ 1 > 2 \lambda_1>2 λ 1 > 2 , we will choose R k R_k R k so that
R j ≪ R j λ j ≪ R j + 1 . R_j\ll R_j\lambda_j\ll R_{j+1}. R j ≪ R j λ j ≪ R j + 1 .
Precisely, take R 1 > 2 R_1 > 2 R 1 > 2 . Having chosen R 1 , … , R n − 1 R_1,\ldots,R_{n-1} R 1 , … , R n − 1 , choose R n R_n R n so that
R n > 2 n , R n > 2 Λ n − 1 R n − 1 , (36) R_n > 2^n,\qquad R_n > 2\Lambda_{n-1}R_{n-1}, \tag*{(36)} R n > 2 n , R n > 2 Λ n − 1 R n − 1 , (36)
8 R n ∑ j < n Λ j R j < β n , (37) \frac{8}{R_n}\sum_{j<n}\Lambda_jR_j < \beta_n, \tag*{(37)} R n 8 j < n ∑ Λ j R j < β n , (37)
8 Λ j R j R n < 2 − n β j ( 1 ≤ j < n ) , (38) \frac{8\Lambda_jR_j}{R_n} < 2^{-n}\beta_j \qquad(1 \le j < n), \tag*{(38)} R n 8 Λ j R j < 2 − n β j ( 1 ≤ j < n ) , (38)
where β n = min { μ n 8 , 2 − n − 4 } > 0 \beta_n=\min\left\{\frac{\mu_n}{8},2^{-n-4}\right\}>0 β n = min { 8 μ n , 2 − n − 4 } > 0 . At stage n n n there are only finitely many requirements, and every quantity other than R n R_n R n is already fixed. This choice is maintained throughout the remainder of Section 6.
Proof of Theorem 6.3. Since the proof of this theorem is relatively lengthy, we divide it into two steps.
Step 1. We first prove
sup s ∈ I ∞ sup η ∈ J k ∣ P s ( η ) − ϕ λ k ( s k , η R k ) ∣ < 2 β k . (39) \sup_{s\in I^\infty}\sup_{\eta\in J_k}\left|P_s(\eta)-\phi_{\lambda_k}\left(s_k,\frac{\eta}{R_k}\right)\right|<2\beta_k. \tag*{(39)} s ∈ I ∞ sup η ∈ J k sup P s ( η ) − ϕ λ k ( s k , R k η ) < 2 β k . (39)
Fix k k k and η ∈ J k \eta\in J_k η ∈ J k , We will estimate 1 − ϕ λ k ( s k , η R k ) 1-\phi_{\lambda_k}\left(s_k,\frac{\eta}{R_k}\right) 1 − ϕ λ k ( s k , R k η ) according to whether j < k j<k j < k or j > k j>k j > k . If j < k j<k j < k , all four zeros A A A of block j j j are less than η \eta η . Its four minus signs cancel, and thus
ϕ λ j ( s j , η R j ) = ∏ A in block j η − A η + A . \phi_{\lambda_j}\left(s_j,\frac{\eta}{R_j}\right)=\prod_{A\text{ in block }j}\frac{\eta-A}{\eta+A}. ϕ λ j ( s j , R j η ) = A in block j ∏ η + A η − A .
Each factor belongs to ( 0 , 1 ) (0,1) ( 0 , 1 ) and
1 − η − A η + A = 2 A η + A ≤ 2 Λ j R j R k . 1-\frac{\eta-A}{\eta+A}=\frac{2A}{\eta+A}\le\frac{2\Lambda_jR_j}{R_k}. 1 − η + A η − A = η + A 2 A ≤ R k 2 Λ j R j .
Hence, by (6.13),
0 ≤ 1 − ϕ λ j ( s j , η R j ) ≤ 8 Λ j R j R k ( j < k ) . (40) 0\le1-\phi_{\lambda_j}\left(s_j,\frac{\eta}{R_j}\right)\le\frac{8\Lambda_jR_j}{R_k}\qquad(j<k). \tag*{(40)} 0 ≤ 1 − ϕ λ j ( s j , R j η ) ≤ R k 8 Λ j R j ( j < k ) . (40)
If j > k j>k j > k , all its zeros satisfy A > η A>\eta A > η and
1 − A − η A + η = 2 η A + η ≤ 2 Λ k R k R j . 1-\frac{A-\eta}{A+\eta}=\frac{2\eta}{A+\eta}\le\frac{2\Lambda_kR_k}{R_j}. 1 − A + η A − η = A + η 2 η ≤ R j 2 Λ k R k .
Thus
0 ≤ 1 − ϕ λ j ( s j , η R j ) ≤ 8 Λ k R k R j ( j > k ) . (41) 0\le1-\phi_{\lambda_j}\left(s_j,\frac{\eta}{R_j}\right)\le\frac{8\Lambda_kR_k}{R_j}\qquad(j>k). \tag*{(41)} 0 ≤ 1 − ϕ λ j ( s j , R j η ) ≤ R j 8 Λ k R k ( j > k ) . (41)
All right-hand sides are independent of the values of the parameters s j s_j s j .
Write
Q k , s ( η ) = ∏ j ≥ 1 j ≠ k ϕ λ j ( s j , η R j ) Q_{k,s}(\eta)=\prod_{\substack{j\geq1\\ j\ne k}}\phi_{\lambda_j}\left(s_j,\frac{\eta}{R_j}\right) Q k , s ( η ) = j ≥ 1 j = k ∏ ϕ λ j ( s j , R j η )
and
P s ( η ) = ϕ λ k ( s k , η R k ) Q k , s . P_s(\eta)=\phi_{\lambda_k}\left(s_k,\frac{\eta}{R_k}\right)Q_{k,s}. P s ( η ) = ϕ λ k ( s k , R k η ) Q k , s .
On J k J_k J k , every factor in Q k , s Q_{k,s} Q k , s is positive and at most one. Applying (34) to finite products and passing to the limit, then using (6.19)–(6.20) and (6.16)–(6.17), gives
0 ≤ 1 − Q k , s ( η ) ≤ 8 R k ∑ j < k Λ j R j + ∑ j > k 8 Λ k R k R j < β k + β k ∑ j = k + 1 ∞ 2 − j = β k ( 1 + 2 − k ) ≤ 3 2 β k . \begin{aligned}
0\leq1-Q_{k,s}(\eta)\leq\frac{8}{R_k}\sum_{j<k}\Lambda_jR_j+\sum_{j>k}\frac{8\Lambda_kR_k}{R_j} \\
&<\beta_k+\beta_k\sum_{j=k+1}^{\infty}2^{-j} \\
&=\beta_k(1+2^{-k})\leq\frac{3}{2}\beta_k.
\end{aligned} 0 ≤ 1 − Q k , s ( η ) ≤ R k 8 j < k ∑ Λ j R j + j > k ∑ R j 8 Λ k R k < β k + β k j = k + 1 ∑ ∞ 2 − j = β k ( 1 + 2 − k ) ≤ 2 3 β k .
Since ∣ ϕ λ k ( s k , η R k ) ∣ ≤ 1 \left|\phi_{\lambda_k}\left(s_k,\frac{\eta}{R_k}\right)\right|\leq1 ϕ λ k ( s k , R k η ) ≤ 1 on the positive real axis, and
∣ P s ( η ) − ϕ λ k ( s k , η R k ) ∣ = ∣ ϕ λ k ( s k , η R k ) ∣ ∣ 1 − Q k , s ( η ) ∣ \left|P_s(\eta)-\phi_{\lambda_k}\left(s_k,\frac{\eta}{R_k}\right)\right|=\left|\phi_{\lambda_k}\left(s_k,\frac{\eta}{R_k}\right)\right|\left|1-Q_{k,s}(\eta)\right| P s ( η ) − ϕ λ k ( s k , R k η ) = ϕ λ k ( s k , R k η ) ∣ 1 − Q k , s ( η ) ∣
we immediately obtain (39) .
Step 2. We now give some estimates to prove (33) . The idea is as follows:
D λ k ( 1 − ε 0 ) < 0 , D λ k ( 1 + ε 0 ) > 0. D_{\lambda_k}(1-\varepsilon_0)<0,\qquad D_{\lambda_k}(1+\varepsilon_0)>0. D λ k ( 1 − ε 0 ) < 0 , D λ k ( 1 + ε 0 ) > 0.
These inequalities hold for every k k k . Using a perturbation of D λ k ( s k ) D_{\lambda_k}(s_k) D λ k ( s k ) given by Φ k ( s ) \Phi_k(s) Φ k ( s ) , we obtain the desired assertion (33) .
By noting
D λ k ( s k ) = min η ∈ J k , 1 ϕ λ k ( s k , η R k ) − min η ∈ J k , 3 ϕ λ k ( s k , η R k ) D_{\lambda_k}(s_k)=\min_{\eta\in J_{k,1}}\phi_{\lambda_k}\left(s_k,\frac{\eta}{R_k}\right)-\min_{\eta\in J_{k,3}}\phi_{\lambda_k}\left(s_k,\frac{\eta}{R_k}\right) D λ k ( s k ) = η ∈ J k , 1 min ϕ λ k ( s k , R k η ) − η ∈ J k , 3 min ϕ λ k ( s k , R k η )
and
J k , 1 = [ R k , R k λ k ] , J k , 3 = [ R k λ k 2 , R k λ k 3 s k ] , J_{k,1}=[R_k,R_k\lambda_k],\qquad J_{k,3}=[R_k\lambda_k^2,R_k\lambda_k^3s_k], J k , 1 = [ R k , R k λ k ] , J k , 3 = [ R k λ k 2 , R k λ k 3 s k ] ,
we immediately have
D λ k ( s k ) = min η ∈ [ 1 , λ k ] ϕ λ k ( s k , η ) − min η ∈ [ λ k 2 , λ k 3 s k ] ϕ λ k ( s k , η ) . D_{\lambda_k}(s_k)=\min_{\eta\in[1,\lambda_k]}\phi_{\lambda_k}(s_k,\eta)-\min_{\eta\in[\lambda_k^2,\lambda_k^3s_k]}\phi_{\lambda_k}(s_k,\eta). D λ k ( s k ) = η ∈ [ 1 , λ k ] min ϕ λ k ( s k , η ) − η ∈ [ λ k 2 , λ k 3 s k ] min ϕ λ k ( s k , η ) .
Hence D λ k ( s k ) D_{\lambda_k}(s_k) D λ k ( s k ) does not depend on R k R_k R k . This observation allows us to choose R k R_k R k freely with respect to the value D λ k ( s k ) D_{\lambda_k}(s_k) D λ k ( s k ) .
Applying (6.14) on the two intervals and using (6.18), we obtain
∣ Φ k ( s ) − D λ k ( s k ) ∣ ≤ ∣ min η ∈ J k , 1 P s ( η ) − min η ∈ J k , 1 ϕ λ k ( s k , η R k ) ∣ + ∣ min η ∈ J k , 3 P s − min η ∈ J k , 3 ϕ λ k ( s k , η R k ) ∣ < 4 β k ≤ μ k 2 < μ k . (42) \begin{aligned}
\left|\Phi_k(\mathbf{s})-D_{\lambda_k}(s_k)\right|
&\leq\left|\min_{\eta\in J_{k,1}} P_s(\eta)-\min_{\eta\in J_{k,1}}\phi_{\lambda_k}\left(s_k,\frac{\eta}{R_k}\right)\right|\\
&\quad+\left|\min_{\eta\in J_{k,3}} P_s-\min_{\eta\in J_{k,3}}\phi_{\lambda_k}\left(s_k,\frac{\eta}{R_k}\right)\right|\\
&<4\beta_k\leq\frac{\mu_k}{2}<\mu_k. \tag*{(42)}
\end{aligned} ∣ Φ k ( s ) − D λ k ( s k ) ∣ ≤ η ∈ J k , 1 min P s ( η ) − η ∈ J k , 1 min ϕ λ k ( s k , R k η ) + η ∈ J k , 3 min P s − η ∈ J k , 3 min ϕ λ k ( s k , R k η ) < 4 β k ≤ 2 μ k < μ k . (42)
The definition of μ k \mu_k μ k gives
D λ k ( 1 − ε 0 ) ≤ − 2 μ k , D λ k ( 1 + ε 0 ) ≥ 2 μ k . D_{\lambda_k}(1-\varepsilon_0)\leq-2\mu_k,\qquad D_{\lambda_k}(1+\varepsilon_0)\geq2\mu_k. D λ k ( 1 − ε 0 ) ≤ − 2 μ k , D λ k ( 1 + ε 0 ) ≥ 2 μ k .
If s k = 1 − ε 0 s_k=1-\varepsilon_0 s k = 1 − ε 0 , then by (6.21)
Φ k ( s ) < D λ k ( 1 − ε 0 ) + μ k ≤ − μ k < 0. \Phi_k(\mathbf{s})<D_{\lambda_k}(1-\varepsilon_0)+\mu_k\leq-\mu_k<0. Φ k ( s ) < D λ k ( 1 − ε 0 ) + μ k ≤ − μ k < 0.
If s k = 1 + ε 0 s_k=1+\varepsilon_0 s k = 1 + ε 0 , then
Φ k ( s ) > D λ k ( 1 + ε 0 ) − μ k ≥ μ k > 0. \Phi_k(\mathbf{s})>D_{\lambda_k}(1+\varepsilon_0)-\mu_k\geq\mu_k>0. Φ k ( s ) > D λ k ( 1 + ε 0 ) − μ k ≥ μ k > 0.
We thereby obtain (6.12), as desired. □
The construction of P s P_s P s
Some observations
We take the product topology on I ∞ = I × I × ⋯ I^\infty=I\times I\times\cdots I ∞ = I × I × ⋯ , where I = [ 3 4 , 5 4 ] I=\left[\frac{3}{4},\frac{5}{4}\right] I = [ 4 3 , 4 5 ] . Its neighborhood basis consists of sets
U 1 × U 2 × ⋯ × U k × I × I × ⋯ , U_1\times U_2\times\cdots\times U_k\times I\times I\times\cdots, U 1 × U 2 × ⋯ × U k × I × I × ⋯ ,
where k < ∞ k<\infty k < ∞ and each U j U_j U j (1 ≤ j ≤ k 1\leq j\leq k 1 ≤ j ≤ k ) is a relatively open subset of I I I . For s N \mathbf{s}^N s N and s ∗ \mathbf{s}^* s ∗ in I ∞ I^\infty I ∞ , if { s N } N = 1 ∞ \{\mathbf{s}^N\}_{N=1}^{\infty} { s N } N = 1 ∞ converges to s ∗ \mathbf{s}^* s ∗ in each coordinate I ∞ I^\infty I ∞ , then { s N } N = 1 ∞ \{\mathbf{s}^N\}_{N=1}^{\infty} { s N } N = 1 ∞ converges to s ∗ \mathbf{s}^* s ∗ in the product topology of I ∞ I^\infty I ∞ .
For each m < ∞ m<\infty m < ∞ ,
P s , m ( η ) = ∏ k = 1 m ϕ λ k ( s k , η R k ) . P_{s,m}(\eta)=\prod_{k=1}^{m}\phi_{\lambda_k}\left(s_k,\frac{\eta}{R_k}\right). P s , m ( η ) = k = 1 ∏ m ϕ λ k ( s k , R k η ) .
as the partial product of P s P_s P s . In the following, we abbreviate “Observation” as “Ob.”
(Ob.1) Let 1 ≤ m < ∞ 1 \le m < \infty 1 ≤ m < ∞ . For s N s^{N} s N and s ∗ s^{*} s ∗ in I ∞ I^{\infty} I ∞ , if { s N } N = 1 ∞ \{s^{N}\}_{N=1}^{\infty} { s N } N = 1 ∞ converges to s ∗ s^{*} s ∗ in each coordinate, then
P s , m ( s N , η ) → P s , m ( s ∗ , η ) P_{s,m}(s^{N},\eta) \to P_{s,m}(s^{*},\eta) P s , m ( s N , η ) → P s , m ( s ∗ , η ) uniformly for η \eta η in each compact subset K ⊆ H K \subseteq\mathbb{H} K ⊆ H .
This follows directly from the definition of P s , m P_{s,m} P s , m , which involves only finitely many factors of the form ϕ λ k ( s k , η R k ) \phi_{\lambda_k}\left(s_k,\frac{\eta}{R_k}\right) ϕ λ k ( s k , R k η ) , and only finitely many of s 1 , ⋯ , s m s_1,\cdots,s_m s 1 , ⋯ , s m are involved.
(Ob.2) For each compact subset K ⊆ H K \subseteq\mathbb{H} K ⊆ H , we have
sup ( s N , s N + 1 , … ) ∈ I ∞ sup η ∈ K ∣ ∏ k = N + 1 ∞ ϕ λ k ( s k , η R k ) − 1 ∣ → 0 ( N → ∞ ) \sup_{(s_N,s_{N+1},\ldots)\in I^\infty}\sup_{\eta\in K}\left|\prod_{k=N+1}^{\infty}\phi_{\lambda_k}\left(s_k,\frac{\eta}{R_k}\right)-1\right|\to0 \quad(N\to\infty) ( s N , s N + 1 , … ) ∈ I ∞ sup η ∈ K sup k = N + 1 ∏ ∞ ϕ λ k ( s k , R k η ) − 1 → 0 ( N → ∞ ) and
sup s ∈ I ∞ sup η ∈ K ∣ P s ( η ) − ∏ j = 1 N ϕ λ j ( s j , η R j ) ∣ → 0 , N → ∞ . \sup_{s\in I^\infty}\sup_{\eta\in K}\left|P_s(\eta)-\prod_{j=1}^{N}\phi_{\lambda_j}\left(s_j,\frac{\eta}{R_j}\right)\right|\to0,\quad N\to\infty. s ∈ I ∞ sup η ∈ K sup P s ( η ) − j = 1 ∏ N ϕ λ j ( s j , R j η ) → 0 , N → ∞. In fact, for a compact subset K ⊆ H K\subseteq\mathbb{H} K ⊆ H , put M K = max { ∣ η ∣ : η ∈ K } M_K=\max\{|\eta|:\eta\in K\} M K = max { ∣ η ∣ : η ∈ K } . For those R j > M K R_j>M_K R j > M K , each zero A A A in block j j j satisfies A ≥ R j A\ge R_j A ≥ R j , and
∣ A − η A + η − 1 ∣ ≤ 2 M K R j , ∣ A − η A + η ∣ ≤ 1. \left|\frac{A-\eta}{A+\eta}-1\right|\le\frac{2M_K}{R_j},\qquad\left|\frac{A-\eta}{A+\eta}\right|\le1. A + η A − η − 1 ≤ R j 2 M K , A + η A − η ≤ 1. Applying (6.13) gives
∣ ϕ λ j ( s j , η R j ) − 1 ∣ ≤ 8 M K R j . \left|\phi_{\lambda_j}\left(s_j,\frac{\eta}{R_j}\right)-1\right|\le\frac{8M_K}{R_j}. ϕ λ j ( s j , R j η ) − 1 ≤ R j 8 M K . Furthermore,
sup s ∈ I ∞ sup η ∈ K ∣ P s ( η ) − ∏ j = 1 N ϕ λ j ( s j , η R j ) ∣ = sup s ∈ I ∞ sup η ∈ K ∣ ∏ j = 1 N ϕ λ j ( s j , η R j ) ∣ ∣ ∏ j = N + 1 ∞ ϕ λ j ( s j , η R j ) − 1 ∣ ≤ sup s ∈ I ∞ sup η ∈ K ∣ ∏ j = N + 1 ∞ ϕ λ j ( s j , η R j ) − 1 ∣ ≤ 8 M K ∑ j > N 1 R j → 0 , N → ∞ . \begin{aligned}
\sup_{s\in I^\infty}\sup_{\eta\in K}\left|P_s(\eta)-\prod_{j=1}^{N}\phi_{\lambda_j}\left(s_j,\frac{\eta}{R_j}\right)\right|
&=\sup_{s\in I^\infty}\sup_{\eta\in K}\left|\prod_{j=1}^{N}\phi_{\lambda_j}\left(s_j,\frac{\eta}{R_j}\right)\right|\left|\prod_{j=N+1}^{\infty}\phi_{\lambda_j}\left(s_j,\frac{\eta}{R_j}\right)-1\right|\\
&\le\sup_{s\in I^\infty}\sup_{\eta\in K}\left|\prod_{j=N+1}^{\infty}\phi_{\lambda_j}\left(s_j,\frac{\eta}{R_j}\right)-1\right|\\
&\le8M_K\sum_{j>N}\frac{1}{R_j}\to0,\quad N\to\infty.
\end{aligned} s ∈ I ∞ sup η ∈ K sup P s ( η ) − j = 1 ∏ N ϕ λ j ( s j , R j η ) = s ∈ I ∞ sup η ∈ K sup j = 1 ∏ N ϕ λ j ( s j , R j η ) j = N + 1 ∏ ∞ ϕ λ j ( s j , R j η ) − 1 ≤ s ∈ I ∞ sup η ∈ K sup j = N + 1 ∏ ∞ ϕ λ j ( s j , R j η ) − 1 ≤ 8 M K j > N ∑ R j 1 → 0 , N → ∞. The formula ∑ j > N 1 R j → 0 \sum_{j>N}\frac{1}{R_j}\to0 ∑ j > N R j 1 → 0 ( N → ∞ ) (N\to\infty) ( N → ∞ ) follows from (6.15).
Claim: For s N s^N s N and s ∗ s^* s ∗ in I ∞ I^\infty I ∞ , if { s N } N = 1 ∞ \{s^N\}_{N=1}^\infty { s N } N = 1 ∞ converges to s ∗ s^* s ∗ in each coordinate, then
c 1 k ( s N ) → c 1 k ( s ∗ ) and c 3 k ( s N ) → c 3 k ( s ∗ ) , N → ∞ . c_1^k(s^N) \to c_1^k(s^*) \quad\text{and} \quad c_3^k(s^N) \to c_3^k(s^*), \qquad N \to\infty. c 1 k ( s N ) → c 1 k ( s ∗ ) and c 3 k ( s N ) → c 3 k ( s ∗ ) , N → ∞. Recall that c 1 k ( s ) c_1^k(s) c 1 k ( s ) and c 3 k ( s ) c_3^k(s) c 3 k ( s ) are the unique critical points of P s P_s P s in J k , 1 J_{k,1} J k , 1 and J k , 3 J_{k,3} J k , 3 , respectively.
In Subsection 6.1, by the arguments in the paragraph above titled “A symmetric four-zero block”, we have that P s N P_{s^N} P s N and P s ∗ P_{s^*} P s ∗ have exactly one critical point in J k , 1 J_{k,1} J k , 1 . For any ε ′ > 0 \varepsilon' > 0 ε ′ > 0 , we now show that there exists an integer N 0 ≥ 1 N_0 \ge1 N 0 ≥ 1 such that
∣ c 1 k ( s N ) − c 1 k ( s ∗ ) ∣ < ε ′ for all N ≥ N 0 . \left|c_1^k(s^N)-c_1^k(s^*)\right|<\varepsilon' \qquad\text{for all } N\ge N_0. c 1 k ( s N ) − c 1 k ( s ∗ ) < ε ′ for all N ≥ N 0 . For this, we will apply Rouché’s theorem on a closed elliptic disk K K K (K ⊆ H K \subseteq\mathbb{H} K ⊆ H ) whose interior contains J k , 1 J_{k,1} J k , 1 and J k , 3 J_{k,3} J k , 3 . There exists a δ > 0 \delta>0 δ > 0 such that if a holomorphic function h h h on K K K satisfies
sup η ∈ K ∣ h ( η ) − P s ∗ ( η ) ∣ < δ , \sup_{\eta\in K}|h(\eta)-P_{s^*}(\eta)|<\delta, η ∈ K sup ∣ h ( η ) − P s ∗ ( η ) ∣ < δ , then h h h has exactly one critical point lying in an ε ′ \varepsilon' ε ′ -neighborhood of c 1 k ( s ∗ ) c_1^k(s^*) c 1 k ( s ∗ ) . We take h = P s N h=P_{s^N} h = P s N , and the unique critical point is c 1 k ( s N ) c_1^k(s^N) c 1 k ( s N ) . We will shortly prove that
sup η ∈ K ∣ P s N ( η ) − P s ∗ ( η ) ∣ < δ \sup_{\eta\in K}|P_{s^N}(\eta)-P_{s^*}(\eta)|<\delta η ∈ K sup ∣ P s N ( η ) − P s ∗ ( η ) ∣ < δ for sufficiently large N N N .
To be precise, by (Ob.2), for all η \eta η on this compact set K K K ,
∣ P s N , m ( η ) − P s N ( η ) ∣ < δ 4 and ∣ P s ∗ , m ( η ) − P s ∗ ( η ) ∣ < δ 4 |P_{s^N,m}(\eta)-P_{s^N}(\eta)|<\frac{\delta}{4}
\quad\text{and}\quad
|P_{s^*,m}(\eta)-P_{s^*}(\eta)|<\frac{\delta}{4} ∣ P s N , m ( η ) − P s N ( η ) ∣ < 4 δ and ∣ P s ∗ , m ( η ) − P s ∗ ( η ) ∣ < 4 δ for some m m m . Since m m m is fixed, by (Ob.1), there exists an integer N 0 ≥ 1 N_0\ge1 N 0 ≥ 1 such that
∣ P s ∗ , m ( η ) − P s N , m ( η ) ∣ < δ 4 , η ∈ K |P_{s^*,m}(\eta)-P_{s^N,m}(\eta)|<\frac{\delta}{4},\qquad\eta\in K ∣ P s ∗ , m ( η ) − P s N , m ( η ) ∣ < 4 δ , η ∈ K for N ≥ N 0 N\ge N_0 N ≥ N 0 . Hence we obtain that for those N N N ,
∣ P s N ( η ) − P s ∗ ( η ) ∣ ≤ ∣ P s N ( η ) − P s N , m ( η ) ∣ + ∣ P s N , m ( η ) − P s ∗ , m ( η ) ∣ + ∣ P s ∗ , m ( η ) − P s ∗ ( η ) ∣ < 3 δ 4 < δ , η ∈ K , \begin{aligned}
|P_{s^N}(\eta)-P_{s^*}(\eta)|
&\le|P_{s^N}(\eta)-P_{s^N,m}(\eta)|+|P_{s^N,m}(\eta)-P_{s^*,m}(\eta)|\\
&\qquad+|P_{s^*,m}(\eta)-P_{s^*}(\eta)|\\
&<\frac{3\delta}{4}<\delta,\qquad\eta\in K,
\end{aligned} ∣ P s N ( η ) − P s ∗ ( η ) ∣ ≤ ∣ P s N ( η ) − P s N , m ( η ) ∣ + ∣ P s N , m ( η ) − P s ∗ , m ( η ) ∣ + ∣ P s ∗ , m ( η ) − P s ∗ ( η ) ∣ < 4 3 δ < δ , η ∈ K , @】【。
as desired. In this way, we can find an N 0 ≥ 1 N_0 \ge1 N 0 ≥ 1 such that
∣ c 1 k ( s N ) − c 1 k ( s ∗ ) ∣ < ε ′ for all N ≥ N 0 . \left|c_1^k(s^N)-c_1^k(s^*)\right|<\varepsilon' \text{ for all } N\ge N_0. c 1 k ( s N ) − c 1 k ( s ∗ ) < ε ′ for all N ≥ N 0 .
A similar argument applies to ∣ c 3 k ( s N ) − c 3 k ( s ∗ ) ∣ \left|c_3^k(s^N)-c_3^k(s^*)\right| c 3 k ( s N ) − c 3 k ( s ∗ ) .
Remind that
Φ k ( s ) = P s ( c 1 k ( s ) ) − P s ( c 3 k ( s ) ) . \Phi_k(s)=P_s(c_1^k(s))-P_s(c_3^k(s)). Φ k ( s ) = P s ( c 1 k ( s )) − P s ( c 3 k ( s )) .
For s N s^N s N and s ∗ s^* s ∗ in I ∞ I^\infty I ∞ , if { s N } N = 1 ∞ \{s^N\}_{N=1}^{\infty} { s N } N = 1 ∞ converges to s ∗ s^* s ∗ in each coordinate, then
Φ k ( s N ) → Φ k ( s ∗ ) , N → ∞ . \Phi_k(s^N)\to\Phi_k(s^*),\quad N\to\infty. Φ k ( s N ) → Φ k ( s ∗ ) , N → ∞.
To see this, note that
∣ Φ k ( s N ) − Φ k ( s ∗ ) ∣ = ∣ P s N ( c 1 k ( s N ) ) − P s N ( c 3 k ( s N ) ) − P s ∗ ( c 1 k ( s ∗ ) ) + P s ∗ ( c 3 k ( s ∗ ) ) ∣ ≤ ∣ P s N ( c 1 k ( s N ) ) − P s ∗ ( c 1 k ( s ∗ ) ) ∣ + ∣ P s N ( c 3 k ( s N ) ) − P s ∗ ( c 3 k ( s ∗ ) ) ∣ . \begin{aligned}
\left|\Phi_k(s^N)-\Phi_k(s^*)\right|
={}&\left|P_{s^N}(c_1^k(s^N))-P_{s^N}(c_3^k(s^N))-P_{s^*}(c_1^k(s^*))+P_{s^*}(c_3^k(s^*))\right|\\
\le{}&\left|P_{s^N}(c_1^k(s^N))-P_{s^*}(c_1^k(s^*))\right|+\left|P_{s^N}(c_3^k(s^N))-P_{s^*}(c_3^k(s^*))\right|.
\end{aligned} Φ k ( s N ) − Φ k ( s ∗ ) = ≤ P s N ( c 1 k ( s N )) − P s N ( c 3 k ( s N )) − P s ∗ ( c 1 k ( s ∗ )) + P s ∗ ( c 3 k ( s ∗ )) P s N ( c 1 k ( s N )) − P s ∗ ( c 1 k ( s ∗ )) + P s N ( c 3 k ( s N )) − P s ∗ ( c 3 k ( s ∗ )) .
For every ε > 0 \varepsilon>0 ε > 0 , it suffices to prove
∣ P s N ( c 1 k ( s N ) ) − P s ∗ ( c 1 k ( s ∗ ) ) ∣ < 3 ε \left|P_{s^N}(c_1^k(s^N))-P_{s^*}(c_1^k(s^*))\right|<3\varepsilon P s N ( c 1 k ( s N )) − P s ∗ ( c 1 k ( s ∗ )) < 3 ε
for large N N N , and similarly,
∣ P s N ( c 3 k ( s N ) ) − P s ∗ ( c 3 k ( s ∗ ) ) ∣ < 3 ε . \left|P_{s^N}(c_3^k(s^N))-P_{s^*}(c_3^k(s^*))\right|<3\varepsilon. P s N ( c 3 k ( s N )) − P s ∗ ( c 3 k ( s ∗ )) < 3 ε .
Recall that
P s , m ( η ) = ∏ k = 1 m ϕ λ k ( s k , η R k ) . P_{s,m}(\eta)=\prod_{k=1}^{m}\phi_{\lambda_k}\left(s_k,\frac{\eta}{R_k}\right). P s , m ( η ) = k = 1 ∏ m ϕ λ k ( s k , R k η ) .
Then
∣ P s N ( c 1 k ( s N ) ) − P s ∗ ( c 1 k ( s ∗ ) ) ∣ ≤ ∣ P s N ( c 1 k ( s N ) ) − P s N , m ( c 1 k ( s N ) ) ∣ + ∣ P s N , m ( c 1 k ( s N ) ) − P s ∗ , m ( c 1 k ( s ∗ ) ) ∣ + ∣ P s ∗ ( c 1 k ( s ∗ ) ) − P s ∗ , m ( c 1 k ( s ∗ ) ) ∣ . \begin{aligned}
\left|P_{s^N}(c_1^k(s^N))-P_{s^*}(c_1^k(s^*))\right|
\le{}&\left|P_{s^N}(c_1^k(s^N))-P_{s^N,m}(c_1^k(s^N))\right|\\
&+\left|P_{s^N,m}(c_1^k(s^N))-P_{s^*,m}(c_1^k(s^*))\right|\\
&+\left|P_{s^*}(c_1^k(s^*))-P_{s^*,m}(c_1^k(s^*))\right|.
\end{aligned} P s N ( c 1 k ( s N )) − P s ∗ ( c 1 k ( s ∗ )) ≤ P s N ( c 1 k ( s N )) − P s N , m ( c 1 k ( s N )) + P s N , m ( c 1 k ( s N )) − P s ∗ , m ( c 1 k ( s ∗ )) + P s ∗ ( c 1 k ( s ∗ )) − P s ∗ , m ( c 1 k ( s ∗ )) .
By (Ob.2), the first term ∣ P s N ( c 1 k ( s N ) ) − P s N , m ( c 1 k ( s N ) ) ∣ \left|P_{s^N}(c_1^k(s^N))-P_{s^N,m}(c_1^k(s^N))\right| P s N ( c 1 k ( s N )) − P s N , m ( c 1 k ( s N )) and the third term ∣ P s ∗ ( c 1 k ( s ∗ ) ) − P s ∗ , m ( c 1 k ( s ∗ ) ) ∣ \left|P_{s^*}(c_1^k(s^*))-P_{s^*,m}(c_1^k(s^*))\right| P s ∗ ( c 1 k ( s ∗ )) − P s ∗ , m ( c 1 k ( s ∗ )) can be made less than ε \varepsilon ε for some fixed m m m and all N ≥ 1 N\ge1 N ≥ 1 . Since P s , m P_{s,m} P s , m depends only on finitely many products of ϕ λ k ( s k , η R k ) \phi_{\lambda k}\left(s_k,\frac{\eta}{R_k}\right) ϕ λk ( s k , R k η ) , the middle term converges uniformly for s ∈ I ∞ s \in I^\infty s ∈ I ∞ , essentially for ( s 1 , … , s m ) ∈ I m (s_1,\ldots,s_m) \in I^m ( s 1 , … , s m ) ∈ I m . Then we have by (Ob.3) that
c 1 k ( s N ) → c 1 k ( s ∗ ) , N → ∞ , c_1^k(s^N) \to c_1^k(s^*), \qquad N \to\infty, c 1 k ( s N ) → c 1 k ( s ∗ ) , N → ∞ ,
and thus the middle term is less than ε \varepsilon ε for sufficiently large N N N , and then
∣ P s N ( c 1 k ( s N ) ) − P s ∗ ( c 1 k ( s ∗ ) ) ∣ < 3 ε , \left|P_{s^N}\left(c_1^k(s^N)\right)-P_{s^*}\left(c_1^k(s^*)\right)\right|<3\varepsilon, P s N ( c 1 k ( s N ) ) − P s ∗ ( c 1 k ( s ∗ ) ) < 3 ε ,
as desired.
In the following construction, we will specialize s N s^N s N and s ∗ s^* s ∗ in I ∞ I^\infty I ∞ .
Completion of the construction
We will seek a parameter vector s ∗ s^* s ∗ in I ∞ I^\infty I ∞ such that
Φ k ( s ∗ ) = 0 , k = 1 , 2 , ⋯ . (43) \Phi_k(s^*)=0,\qquad k=1,2,\cdots. \tag*{(43)} Φ k ( s ∗ ) = 0 , k = 1 , 2 , ⋯ . (43)
The result used here is the Poincaré–Miranda Theorem (also called Miranda’s Theorem), the multivariable analogue of the intermediate value theorem and a theorem equivalent to the Brouwer fixed point theorem (see [?]).
Theorem 6.4 (Poincaré–Miranda Theorem). Let
Q = ∏ j = 1 N [ a j , b j ] ⊂ R N , F = ( F 1 , ⋯ , F N ) : Q → R N Q=\prod_{j=1}^{N}[a_j,b_j]\subset\mathbb{R}^{N},\qquad F=(F_1,\cdots,F_N):Q\to\mathbb{R}^{N} Q = j = 1 ∏ N [ a j , b j ] ⊂ R N , F = ( F 1 , ⋯ , F N ) : Q → R N be continuous. Suppose that, for each j = 1 , ⋯ , N j=1,\cdots,N j = 1 , ⋯ , N ,
F j ( x ) ≤ 0 on the face x j = a j ; F j ( x ) ≥ 0 on the face x j = b j . F_j(x)\leq0\quad\text{on the face }x_j=a_j;\qquad F_j(x)\geq0\quad\text{on the face }x_j=b_j. F j ( x ) ≤ 0 on the face x j = a j ; F j ( x ) ≥ 0 on the face x j = b j . Then there exists x ∗ ∈ Q x^*\in Q x ∗ ∈ Q such that
F ( x ∗ ) = 0. F(x^*)=0. F ( x ∗ ) = 0.
We apply Theorem 6.4 to the present construction. Fix N N N . Freeze the tail parameters, for instance by setting
s N + 1 = s N + 2 = ⋯ = 1 , s_{N+1}=s_{N+2}=\cdots=1, s N + 1 = s N + 2 = ⋯ = 1 ,
and consider the finite-dimensional rectangle
Q N = ∏ j = 1 N [ 1 − ε 0 , 1 + ε 0 ] , with ε 0 = 1 4 . Q_N=\prod_{j=1}^{N}[1-\varepsilon_0,1+\varepsilon_0],\qquad\text{with }\varepsilon_0=\frac{1}{4}. Q N = j = 1 ∏ N [ 1 − ε 0 , 1 + ε 0 ] , with ε 0 = 4 1 .
Define
F ( N ) ( s 1 , ⋯ , s N ) = ( Φ 1 ( s ) , ⋯ , Φ N ( s ) ) . F^{(N)}(s_1,\cdots,s_N)=(\Phi_1(s),\cdots,\Phi_N(s)). F ( N ) ( s 1 , ⋯ , s N ) = ( Φ 1 ( s ) , ⋯ , Φ N ( s )) .
By (Ob.4), each function Φ 1 ( s ) , ⋯ , Φ N ( s ) \Phi_1(s),\cdots,\Phi_N(s) Φ 1 ( s ) , ⋯ , Φ N ( s ) is continuous in s s s , and hence F ( N ) F^{(N)} F ( N ) is continuous. By Theorem 6.3 for every j j j , the j j j -th component Φ j \Phi_j Φ j is negative on the face s j = 1 − ε 0 s_j=1-\varepsilon_0 s j = 1 − ε 0 and positive on the opposite face s j = 1 + ε 0 s_j=1+\varepsilon_0 s j = 1 + ε 0 , uniformly with respect to the remaining coordinates. Hence the Poincaré–Miranda Theorem applies and yields a point of Q N Q_N Q N at which
Φ 1 = ⋯ = Φ N = 0. \Phi_1=\cdots=\Phi_N=0. Φ 1 = ⋯ = Φ N = 0.
Appending the frozen tail coordinates produces an infinite parameter vector s N \mathbf{s}^N s N . Thus, for N = 1 , 2 , ⋯ N=1,2,\cdots N = 1 , 2 , ⋯ , we have
Φ 1 ( s 1 ) = 0 , \Phi_1(\mathbf{s}^1)=0, Φ 1 ( s 1 ) = 0 ,
Φ 1 ( s 2 ) = Φ 2 ( s 2 ) = 0 , \Phi_1(\mathbf{s}^2)=\Phi_2(\mathbf{s}^2)=0, Φ 1 ( s 2 ) = Φ 2 ( s 2 ) = 0 ,
Φ 1 ( s 3 ) = Φ 2 ( s 3 ) = Φ 3 ( s 3 ) = 0 , \Phi_1(\mathbf{s}^3)=\Phi_2(\mathbf{s}^3)=\Phi_3(\mathbf{s}^3)=0, Φ 1 ( s 3 ) = Φ 2 ( s 3 ) = Φ 3 ( s 3 ) = 0 ,
Φ 1 ( s k ) = ⋯ = Φ k ( s k ) = 0 , \Phi_1(\mathbf{s}^k)=\cdots=\Phi_k(\mathbf{s}^k)=0, Φ 1 ( s k ) = ⋯ = Φ k ( s k ) = 0 ,
For each k k k , every coordinate s k j s_k^j s k j lies in the compact interval I I I . By a diagonal subsequence argument, there exists a subsequence of { s N } N = 1 ∞ \{\mathbf{s}^N\}_{N=1}^{\infty} { s N } N = 1 ∞ that converges in each coordinate. Without loss of generality, suppose that { s N } N = 1 ∞ \{\mathbf{s}^N\}_{N=1}^{\infty} { s N } N = 1 ∞ converges in each coordinate; then there exists s k ∗ ∈ I s_k^*\in I s k ∗ ∈ I for each k k k such that
s k N → s k ∗ , N → ∞ . s_k^N\to s_k^*,\qquad N\to\infty. s k N → s k ∗ , N → ∞.
Set
s ∗ = ( s 1 ∗ , s 2 ∗ , ⋯ ) ∈ I ∞ . \mathbf{s}^*=(s_1^*,s_2^*,\cdots)\in I^\infty. s ∗ = ( s 1 ∗ , s 2 ∗ , ⋯ ) ∈ I ∞ .
For each fixed k k k , since Φ k ( s j ) = 0 \Phi_k(\mathbf{s}^j)=0 Φ k ( s j ) = 0 holds for j = k , k + 1 , ⋯ j=k,k+1,\cdots j = k , k + 1 , ⋯ , by (Ob.4) we immediately conclude that Φ k ( s ∗ ) = 0 \Phi_k(\mathbf{s}^*)=0 Φ k ( s ∗ ) = 0 . That is, we find an s ∗ ∈ I ∞ \mathbf{s}^*\in I^\infty s ∗ ∈ I ∞ such that
Φ k ( s ∗ ) = 0 for every k . \Phi_k(\mathbf{s}^*)=0\qquad\text{for every }k. Φ k ( s ∗ ) = 0 for every k .
In other words, in each block the infinite product P s ∗ P_{\mathbf{s}^*} P s ∗ has two distinct critical points ξ k , η k \xi_k,\eta_k ξ k , η k such that
P s ∗ ′ ( ξ k ) = P s ∗ ′ ( η k ) = 0 , P s ∗ ( ξ k ) = P s ∗ ( η k ) . (44) P_{\mathbf{s}^*}'(\xi_k)=P_{\mathbf{s}^*}'(\eta_k)=0,\qquad P_{\mathbf{s}^*}(\xi_k)=P_{\mathbf{s}^*}(\eta_k). \tag*{(44)} P s ∗ ′ ( ξ k ) = P s ∗ ′ ( η k ) = 0 , P s ∗ ( ξ k ) = P s ∗ ( η k ) . (44)
By definition, P s ∗ P_{\mathbf{s}^*} P s ∗ is not essentially injective on its critical points.
Remark 6.5. The Poincaré–Miranda Theorem itself presents no difficulty. The two points that require explicit justification are: first, the sign condition (6.12) must hold uniformly with respect to all remaining parameters; second, when one passes from finite-dimensional solutions to the diagonal limit, the continuity of each Φ k \Phi_k Φ k with respect to the infinite parameter vector must be specified.
Verification: thinness, primeness, essential injectivity
Thinness of the resulting infinite product
Let { A j } j = 1 ∞ \{A_j\}_{j=1}^{\infty} { A j } j = 1 ∞ be the union of the four-zero blocks in (30) :
R k , R k λ k , R k λ k 2 , R k λ k 3 s k . R_k,\qquad R_k\lambda_k,\qquad R_k\lambda_k^2,\qquad R_k\lambda_k^3s_k. R k , R k λ k , R k λ k 2 , R k λ k 3 s k .
To pass from a positive real half-plane zero A A A to the corresponding disk zero, we use the Cayley transform
a = h ( A ) = A − 1 A + 1 . a=h(A)=\frac{A-1}{A+1}. a = h ( A ) = A + 1 A − 1 .
In what follows, we write A A A and C C C for two distinct elements of { A j } j = 1 ∞ \{A_j\}_{j=1}^{\infty} { A j } j = 1 ∞ . For A , C > 0 A,C>0 A , C > 0 , a direct calculation gives
d ( h ( A ) , h ( C ) ) = ∣ A − C ∣ A + C , (45) d(h(A),h(C))=\frac{\lvert A-C\rvert}{A+C}, \tag*{(45)} d ( h ( A ) , h ( C )) = A + C ∣ A − C ∣ , (45)
where
d ( z , w ) = ∣ z − w 1 − z ‾ w ∣ . d(z,w)=\left\lvert\frac{z-w}{1-\overline{z}w}\right\rvert. d ( z , w ) = 1 − z w z − w .
Set
r = min ( A , C ) max ( A , C ) . r=\frac{\min(A,C)}{\max(A,C)}. r = max ( A , C ) min ( A , C ) .
Then
d ( h ( A ) , h ( C ) ) = 1 − r 1 + r . d(h(A),h(C))=\frac{1-r}{1+r}. d ( h ( A ) , h ( C )) = 1 + r 1 − r .
Thus ρ ( h ( A ) , h ( C ) ) \rho(h(A),h(C)) ρ ( h ( A ) , h ( C )) is close to 1 whenever r r r is small. For 0 ≤ r ≤ 1 2 0\le r\le\frac{1}{2} 0 ≤ r ≤ 2 1 ,
0 ≤ − log 1 − r 1 + r = ∫ 0 r 2 1 − t 2 d t ≤ 8 3 r ≤ 4 r . (46) 0\le-\log\frac{1-r}{1+r}=\int_0^r\frac{2}{1-t^2}\,dt\le\frac{8}{3}r\le4r. \tag*{(46)} 0 ≤ − log 1 + r 1 − r = ∫ 0 r 1 − t 2 2 d t ≤ 3 8 r ≤ 4 r . (46)
Choose λ k → ∞ \lambda_k\to\infty λ k → ∞ and let R k R_k R k grow sufficiently rapidly so that, for every zero A A A in the k k k -th block,
∑ { C : C ≠ A } min ( A , C ) max ( A , C ) ⟶ 0 ( k → ∞ ) . (47) \sum_{\{C:C\ne A\}}\frac{\min(A,C)}{\max(A,C)}\longrightarrow0\qquad(k\to\infty). \tag*{(47)} { C : C = A } ∑ max ( A , C ) min ( A , C ) ⟶ 0 ( k → ∞ ) . (47)
Once (6.26) is verified, then by (6.24)–(6.26),
− log ∏ { C : C ≠ A } d ( h ( A ) , h ( C ) ) = ∑ { C : C ≠ A } − log d ( h ( A ) , h ( C ) ) -\log\prod_{\{C:C\ne A\}} d(h(A),h(C))=\sum_{\{C:C\ne A\}}-\log d(h(A),h(C)) − log { C : C = A } ∏ d ( h ( A ) , h ( C )) = { C : C = A } ∑ − log d ( h ( A ) , h ( C ))
≤ 4 ∑ { C : C ≠ A } min ( A , C ) max ( A , C ) → 0. \le4\sum_{\{C:C\ne A\}}\frac{\min(A,C)}{\max(A,C)}\to0. ≤ 4 { C : C = A } ∑ max ( A , C ) min ( A , C ) → 0.
Therefore
∏ { C : C ≠ A } ρ ( h ( A ) , h ( C ) ) → 1 , \prod_{\{C:C\ne A\}}\rho(h(A),h(C))\to1, { C : C = A } ∏ ρ ( h ( A ) , h ( C )) → 1 ,
and the resulting Blaschke product is thin.
It remains to verify (6.26) uniformly in all parameters. Rewrite
S k ( A ) = ∑ { C : C ≠ A } min ( A , C ) max ( A , C ) , S_k(A)=\sum_{\{C:C\ne A\}}\frac{\min(A,C)}{\max(A,C)}, S k ( A ) = { C : C = A } ∑ max ( A , C ) min ( A , C ) ,
where k k k indicates that A A A lies in the k k k -th zero block. For two consecutive zeros in block k k k , the smaller-to-larger ratios are 1 λ k \frac{1}{\lambda_k} λ k 1 , 1 λ k \frac{1}{\lambda_k} λ k 1 , and 1 λ k s k \frac{1}{\lambda_k s_k} λ k s k 1 . They are all at most 4 3 λ k < 1 \frac{4}{3\lambda_k}<1 3 λ k 4 < 1 . Thus the three other zeros in the same block give terms summing to at most 4 λ k \frac{4}{\lambda_k} λ k 4 to S k ( A ) S_k(A) S k ( A ) .
For earlier blocks, C A ≤ λ j R j R k \frac{C}{A}\le\frac{\lambda_jR_j}{R_k} A C ≤ R k λ j R j and there are four zeros per block. The sum of their terms is at most
4 R k ∑ j < k λ j R j < β k 2 . \frac{4}{R_k}\sum_{j<k}\lambda_jR_j<\frac{\beta_k}{2}. R k 4 j < k ∑ λ j R j < 2 β k .
For later blocks, A C ≤ λ k R k R j \frac{A}{C}\le\frac{\lambda_kR_k}{R_j} C A ≤ R j λ k R k , and their terms sum to at most
∑ j > k 4 λ k R k R j ≤ β k 2 ∑ j > k 2 − j = β k 2 2 − k . \sum_{j>k}\frac{4\lambda_kR_k}{R_j}\le\frac{\beta_k}{2}\sum_{j>k}2^{-j}=\frac{\beta_k}{2}2^{-k}. j > k ∑ R j 4 λ k R k ≤ 2 β k j > k ∑ 2 − j = 2 β k 2 − k .
Consequently
S k ( A ) ≤ 4 λ k + β k 2 ( 1 + 2 − k ) → 0 , S_k(A)\le\frac{4}{\lambda_k}+\frac{\beta_k}{2}(1+2^{-k})\to0, S k ( A ) ≤ λ k 4 + 2 β k ( 1 + 2 − k ) → 0 ,
which proves (6.26).
Primeness
All half-plane zeros satisfy A → ∞ A\to\infty A → ∞ , and hence
h ( A ) = A − 1 A + 1 → 1. h(A)=\frac{A-1}{A+1}\to1. h ( A ) = A + 1 A − 1 → 1.
Thus the zeros have the unique boundary accumulation point 1. The following result then shows immediately that the resulting Blaschke product B B B on the unit disk is prime, where B ( z ) = P s ∗ ( η ( z ) ) B(z)=P_{s^*}(\eta(z)) B ( z ) = P s ∗ ( η ( z )) .
Theorem 6.6. [11 Theorem 4.5] Let B B B be a thin Blaschke product whose zeros accumulate at only one point of T \mathbb{T} T , say at 1. Then B B B is prime.
Failure of essential injectivity
The Cayley transform satisfies
η ( z ) = 1 + z 1 − z , d η d z = 2 ( 1 − z ) 2 ≠ 0. \eta(z)=\frac{1+z}{1-z}, \qquad\frac{d\eta}{dz}=\frac{2}{(1-z)^2}\ne0. η ( z ) = 1 − z 1 + z , d z d η = ( 1 − z ) 2 2 = 0.
Hence critical points of the disk function B B B correspond exactly to critical points of the associated half-plane function P s ∗ P_{s^*} P s ∗ :
B ′ ( z ) = 0 ⟺ P s ∗ ′ ( η ( z ) ) = 0. B'(z)=0 \quad\Longleftrightarrow\quad P_{s^*}'(\eta(z))=0. B ′ ( z ) = 0 ⟺ P s ∗ ′ ( η ( z )) = 0.
By (44) , for every k k k there exist distinct critical points z k , w k z_k,w_k z k , w k such that
B ′ ( z k ) = B ′ ( w k ) = 0 , B ( z k ) = B ( w k ) . B'(z_k)=B'(w_k)=0, \qquad B(z_k)=B(w_k). B ′ ( z k ) = B ′ ( w k ) = 0 , B ( z k ) = B ( w k ) .
The pairs arising from distinct blocks are disjoint. Thus by definition, B B B is not essentially injective on its critical set.
Remark 6.7. Remind that I = [ 3 4 , 5 4 ] I=[\frac{3}{4},\frac{5}{4}] I = [ 4 3 , 4 5 ] . By the construction, for a fixed increasing sequence { λ k } \{\lambda_k\} { λ k } with λ 1 > 2 \lambda_1>2 λ 1 > 2 , there exist a carefully chosen infinite sequence { R k } \{R_k\} { R k } (as in (6.15)–(6.17)) and at least one s ∗ ∈ I ∞ s^*\in I^\infty s ∗ ∈ I ∞ such that the corresponding product P s ∗ P_{s^*} P s ∗ on H H H defines a prime thin Blaschke product that is not essentially injective on its critical set.
Acknowledgements. The initial construction of the unique example in Section 6 was suggested by ChatGPT 5.6 Plus. The authors identified a gap in the proposed construction, which was subsequently repaired through further dialogue with the AI. The authors have verified all details. We note that this example occupies only Section 6 of the paper; the remaining sections are the authors’ own work.
School of Mathematical Sciences, Fudan University, Shanghai, 200433, China; E-mail: danniguomath@163.com
School of Mathematics, East China University of Science and Technology, Shanghai, 200237, China; E-mail: hshuang@ecust.edu.cn
Department of Mathematics, Jiangsu University of Technology, Changzhou, 273000, China; E-mail: lishan_math@163.com
College of Mathematics and Statistics, Chongqing University, Chongqing, 401331, China; E-mail: xianfengzhao@cqu.edu.cn