betaFeedback

WorksMathematics

Characterization of the boundary zero set of stable polynomials in B2\mathbb B_2

Abstract

Given a polynomial p∈C[z,w]p\in \mathbb{C}[z,w] whose zeros lie outside the unit ball B2\mathbb{B}_2 of C2,\mathbb{C}^2, we study its local Weierstrass polynomial–obtained via the Weierstrass Preparation Theorem–and provide a strong version of the Nash curve selection lemma. These results lead to a characterization of the intersection of the zero set of pp with the unit sphere: Z(p)∩S2.\mathcal{Z}(p)\cap\mathbb{S}_{2}.

Introduction

This paper is about the study of the zero set of polynomials in C2\mathbb{C}^2 that satisfy a stability condition with respect to the unit ball.

In particular, we obtain a stronger version of the so called Nash curve selection lemma for polynomials p∈C[z,w]p \in\mathbb{C}[z,w] whose zeros lie outside the Euclidean unit ball B2\mathbb{B}_2 of C2\mathbb{C}^2 and achieve a characterization of the intersection of the zero set of pp with the unit sphere: Z(p)∩S2\mathcal{Z}(p) \cap\mathbb{S}_2. Such polynomials are often called stable polynomials, and the set of their zeros lying on the unit sphere is called boundary zero set.

These results also serve as a different approach to recent breakthroughs by G. Knese, J.E. Pascoe, and A. Sola [9 Proposition 1.3, Theorem 1.5], who gave a complete local characterization and parameterization of the (complex) algebraic curves—and their finitely many Newton-Puiseux branches—contained within the local zero set of a stable polynomial in C2\mathbb{C}^2.

The Nash curve selection lemma is a foundational result in semi-algebraic geometry; consequently, establishing stronger variants is a matter of considerable importance. Recall the injective version as it is stated by K. Kurdyka in [12]. See also [1 Proposition 8.1.13, Theorem 9.7.10] which is a semi-algebraic consequence of the statements formulated in [2] by F. Bruhat, H. Cartan and [14] by J. Milnor.

Lemma. Let A⊂RNA \subset\mathbb{R}^N be a semi-algebraic set and let α∈A‾\alpha\in\overline{A} be an accumulation point of AA. Then there exists an injective Nash curve γα:(−1,1)→RN\gamma_\alpha: (-1,1) \to\mathbb{R}^N such that

γα((0,1))⊂Aandγα(0)=α.\gamma_\alpha((0,1)) \subset A \quad\text{and} \quad\gamma_\alpha(0) = \alpha.

It is possible that for each such curve γα\gamma_\alpha and each ϵ>0\epsilon> 0, the set γα((−ϵ,ϵ))\gamma_\alpha((-\epsilon,\epsilon)) is not a (smooth) submanifold in RN\mathbb{R}^N; for instance consider the algebraic set A:={(t2,t3):t∈R}A := \{(t^2,t^3) : t \in\mathbb{R}\} in R2\mathbb{R}^2 and the point α=(0,0)\alpha= (0,0).

Given a polynomial p∈C[z,w]p \in\mathbb{C}[z,w], the set Z(p)∩S2\mathcal{Z}(p) \cap\mathbb{S}_2 may be regarded as an algebraic set in R4\mathbb{R}^4. Hence, it admits a decomposition Z(p)∩S2=⋃iMi\mathcal{Z}(p) \cap\mathbb{S}_2 = \bigcup_i M_i, where MiM_i are finitely many, pairwise disjoint Nash submanifolds of real dimension at most 1, see [19 Theorem 1]. In particular, the decomposition ⋃iMi\bigcup_i M_i may be chosen so that each one-dimensional component

2020 Mathematics Subject Classification. Primary: 14P20, 32A08, 32A60.

Key words and phrases. stable polynomials, unit ball, zero set, Nash curves. Mi‾\overline{M_i} has exactly two points on its boundary, see the discussion right after [18 Proposition 4.1]. According to [9] and [19], we are led to the following question:

Question. Given a stable polynomial p∈C[z,w]p \in\mathbb{C}[z,w] with Z(p)∩S2=⋃iMi\mathcal{Z}(p) \cap\mathbb{S}_2 = \bigcup_i M_i, and any fixed accumulation point α∈Mi‾∖Mi\alpha\in\overline{M_i} \setminus M_i, does there always exist a Nash curve ωα:(−1,1)→Z(p)∩S2\omega_\alpha: (-1,1) \to\mathcal{Z}(p) \cap\mathbb{S}_2, that is a diffeomorphism onto its image and satisfies ωα((0,1))⊂Mi\omega_\alpha((0,1)) \subset M_i, ωα(0)=α\omega_\alpha(0) = \alpha?

We shall prove that the answer to the Question is yes thereby characterizing the boundary zero set Z(p)∩S2\mathcal{Z}(p) \cap\mathbb{S}_2.

The main goal is to apply Weierstrass preparation theorem and Puiseux Parametrization theorem, together with arguments and ideas coming up in [6] by F.R. Harvey and R.O. Wells, [5] by J.E. Fornaess and M. Ovrelid, [7 Lemma C.3] by G. Knese, [8] by G. Knese L. Kosiński, T. Ransford and A. Sola, [9] by G. Knese, J.E. Pascoe, A. Sola, [10] by L. Kosiński et al., [15] by A. Noell, B. Stensønes, [16 Proposition 10.7.5-Theorem 10.7.6] by W. Rudin, [18] by J. Wu, K. Zarvalis et al., and [19] by K. Zarvalis et al.

The answer to the Question is no in the setting Bn\mathbb{B}_n, for n≥3n \ge3 (see [18]); hence, the ambient C2\mathbb{C}^2 is essential.

The aforementioned works provide the primary motivation for the present study. Readers may certainly consider natural extensions of this problem such as changing the domain of stability to one where the topological boundary is an algebraic set or real analytic set.

Preliminary Remarks

Denote the unit ball of C2\mathbb{C}^2 by B2:={(z,w)∈C2:∥(z,w)∥<1}\mathbb{B}_2 := \{(z,w) \in\mathbb{C}^2 : \|(z,w)\| < 1\} and its topological boundary, the unit sphere, by S2:={(z,w)∈C2:∥(z,w)∥=1}\mathbb{S}_2 := \{(z,w) \in\mathbb{C}^2 : \|(z,w)\| = 1\}, where ∥(z,w)∥=∣z∣2+∣w∣2\|(z,w)\| = \sqrt{|z|^2 + |w|^2} is the Euclidean norm induced by the Euclidean inner product ⟨(z,w),(u,v)⟩=zu‾+wv‾\langle(z,w),(u,v)\rangle= z\overline{u} + w\overline{v} in C2\mathbb{C}^2. Denote the disk with center z∈Cz \in\mathbb{C} and radius ϵ>0\epsilon> 0 by Δ(z,ϵ):={z′∈C:∣z−z′∣<ϵ}\Delta(z,\epsilon) := \{z' \in\mathbb{C} : |z-z'| < \epsilon\}.

For the zero set of a function f∈C[z,w]f \in\mathbb{C}[z,w], we use the notation Z(f):={(z,w)∈C2:f(z,w)=0}\mathcal{Z}(f) := \{(z,w) \in\mathbb{C}^2 : f(z,w) = 0\}. In particular, a polynomial p∈C[z,w]p \in\mathbb{C}[z,w] is said to be stable (with respect to the unit ball) if pp is zero-free in the unit ball, that is, Z(p)∩B2=∅\mathcal{Z}(p) \cap\mathbb{B}_2 = \varnothing. The set of the zeros of a stable polynomial that lie on the unit sphere is called boundary zero set, and it is written as Z(p)∩S2\mathcal{Z}(p) \cap\mathbb{S}_2.

Any set A⊂CnA \subset\mathbb{C}^n may be regarded as a subset of R2n\mathbb{R}^{2n}, thus we define its real dimension by

dim⁡RA:=max⁡{dim⁡(Γ):Γ⊂A,Γ submanifold of R2n}∈{−1,0,1,…,2n}\dim_{\mathbb{R}} A := \max\{\dim(\Gamma) : \Gamma\subset A, \Gamma\text{ submanifold of }\mathbb{R}^{2n}\} \in\{-1,0,1,\ldots,2n\}

The case dim⁡RA=−1\dim_{\mathbb{R}} A = -1 is devoted to the instance when A=∅A = \varnothing.

Recall the following definitions on semi-algebraic geometry; see [1] for a concise presentation. See also [11] on theory of real analytic functions. Fix N∈NN \in\mathbb{N}. A set A⊂RNA \subset\mathbb{R}^N is said to be semi-algebraic set provided there exists a finite number of real polynomials pip_i, i=1,…,Ii = 1,\ldots,I, and qi,jq_{i,j}, j=1,…,Jj = 1,\ldots,J such that

A=⋃i=1I⋂j=1J{x∈RN:pi(x)=0,qi,j(x)>0}.A = \bigcup_{i=1}^{I}\bigcap_{j=1}^{J}\{x \in\mathbb{R}^N : p_i(x) = 0, q_{i,j}(x) > 0\}.
  1. Let A⊂RMA \subset\mathbb{R}^M and B⊂RNB \subset\mathbb{R}^N be two semi-algebraic sets. A mapping f:A→Bf : A \to B is characterized as semi-algebraic if its graph is a semi-algebraic set of RM+N\mathbb{R}^{M+N}.

  1. Let A⊂RNA \subset\mathbb{R}^N be semi-algebraic. A semi-algebraic mapping f:A→Rf : A \to\mathbb{R} of class C∞C^\infty is called a Nash function. Moreover, given two semi-algebraic sets A,B⊂RNA,B \subset\mathbb{R}^N, a semi-algebraic bijection f:A→Bf : A \to B of class C∞C^\infty is called a Nash diffeomorphism.

(iii) A semi-algebraic subset MM of RN\mathbb{R}^{N} is said to be a Nash submanifold of RN\mathbb{R}^{N} of dimension dd if for every point x∈Mx \in M, there exists a Nash diffeomorphism ϕ\phi from an open semi-algebraic neighborhood Ω\Omega of the origin in RN\mathbb{R}^{N} onto an open semi-algebraic neighborhood Ω′\Omega' of xx in RN\mathbb{R}^{N} such that ϕ(0)=x\phi(0)=x and ϕ((Rd×{0})∩Ω)=M∩Ω′\phi((\mathbb{R}^{d} \times\{0\}) \cap\Omega)=M \cap\Omega'.

(iv) Two Nash submanifolds MM and NN are considered to be Nash diffeomorphic if there exists a bijection f:M→Nf:M \to N such that both ff and f−1f^{-1} are Nash functions. The class of Nash functions coincide with the class of semi-algebraic real analytic functions.

Writing p=Re⁡p+iIm⁡pp=\operatorname{Re}p+i\operatorname{Im}p, we see that Z(p)∩S2\mathcal{Z}(p) \cap\mathbb{S}_{2} is an semi-algebraic set in R4\mathbb{R}^{4}; indeed, it is the following intersection: {Re⁡p=0}∩{Im⁡p=0}∩{∣z∣2+∣w∣2=1}\{\operatorname{Re}p=0\} \cap\{\operatorname{Im}p=0\} \cap\{|z|^{2}+|w|^{2}=1\}.

Recall the following definitions in the setting of the unit ball B2\mathbb{B}_{2}. We refer the reader to [16] for background on interpolation theory. Let I⊂RI \subset\mathbb{R} be an interval on the real line. A C1C^{1} curve γ:I→S2\gamma:I \to\mathbb{S}_{2} is said to be a complex tangential curve when

⟨γ′(t),γ(t)⟩=0,for all t∈I.\langle\gamma'(t),\gamma(t)\rangle=0,\qquad\text{for all }t\in I.

According to [19 Theorem 1], if p∈C[z,w]p \in\mathbb{C}[z,w] is stable and dim⁡R(Z(p)∩S2)>0\dim_{\mathbb{R}}(\mathcal{Z}(p) \cap\mathbb{S}_{2})>0, then Z(p)∩S2=⋃iMi\mathcal{Z}(p) \cap\mathbb{S}_{2}=\bigcup_i M_i, where MiM_i are pairwise disjoint Nash submanifolds, each equivalent to (−1,1)mi(-1,1)^{m_i}, mi=0,1m_i=0,1, by a complex tangential Nash diffeomorphism. The term ”equivalent” means that there exists a complex tangential Nash diffeomorphism ϕi:(−1,1)mi→Mi\phi_i:(-1,1)^{m_i}\to M_i. By the discussion right after [18 Proposition 4.1], we may assume that each one-dimensional component MiM_i contains exactly two points on its boundary Mi‾∖Mi\overline{M_i}\setminus M_i. Henceforth, the family {Mi}i\{M_i\}_i refers to a such decomposition. If dim⁡R(Z(p)∩S2)=0\dim_{\mathbb{R}}(\mathcal{Z}(p) \cap\mathbb{S}_{2})=0, then Z(p)∩S2\mathcal{Z}(p) \cap\mathbb{S}_{2} consists of finitely many points in the unit sphere. Consequently, the more involved scenario arises when the boundary zero set contains at least one accumulation point.

For the moment, the description provided above doesn’t rule out the geometric case where the union of the closures of two distinct one-dimensional components, namely, Mi‾∪Mj‾\overline{M_i}\cup\overline{M_j}, i≠ji\ne j, build a ”cusp” at a accumulation point α∈Mi‾∩Mj‾\alpha\in\overline{M_i}\cap\overline{M_j}. At this point and throughout the paper, by the term ”cusp” at α\alpha, we mean that for α∈Mi‾∖Mi\alpha\in\overline{M_i}\setminus M_i, and each injective Nash curve γα:(−1,1)→R4\gamma_\alpha:(-1,1)\to\mathbb{R}^{4} derived from Nash curve selection lemma (see § 1), there is no one-dimensional (smooth) submanifold N⊂R4N\subset\mathbb{R}^{4} such that γα([0,ϵ))⊂N\gamma_\alpha([0,\epsilon))\subset N, for all ϵ>0\epsilon>0.

Let us suppose that dim⁡R(Z(p)∩S2)=1\dim_{\mathbb{R}}(\mathcal{Z}(p) \cap\mathbb{S}_{2})=1 with Z(p)∩S2=⋃iMi\mathcal{Z}(p) \cap\mathbb{S}_{2}=\bigcup_i M_i. In this specific case, according to [18], for each accumulation point α∈Mi‾∖Mi\alpha\in\overline{M_i}\setminus M_i–for simplicity assume that α=(0,1)\alpha=(0,1); otherwise make use of a unitary matrix–there exists a Nash curve γα:(−1,1)→Z(p)∩S2\gamma_\alpha:(-1,1)\to\mathcal{Z}(p)\cap\mathbb{S}_{2}, which is homeomorphism onto its image, complex tangential, and is classified by

γα(t)=(βνtν+o(tν),1−∣βν∣22t2ν+o(t2ν)),\gamma_\alpha(t)=\left(\beta_\nu t^\nu+o(t^\nu),1-\frac{|\beta_\nu|^{2}}{2}t^{2\nu}+o(t^{2\nu})\right),

for some fixed integer ν≥2\nu\ge2 and \beta_\nu\ne 0. See [16] on unitary matrices. In particular, \gamma_\alpha(0)=\alpha$ and γα′(0)=0\gamma'_\alpha(0)=0; hence, removing the singularity at α\alpha via parameterization may not be possible. We shall show that this description can be further refined.

For instance, consider the well known stable polynomial p(z)=1−z2−w2p(z)=1-z^{2}-w^{2} and α=(0,1)∈Z(p)∩S2\alpha=(0,1)\in\mathcal{Z}(p)\cap\mathbb{S}_{2}; the Nash curve γα(t)=(tν,1−t2ν)\gamma_\alpha(t)=(t^\nu,\sqrt{1-t^{2\nu}}), t∈(−1,1)t\in(-1,1), ν=3,5,…\nu=3,5,\ldots, is a homeomorphism onto its image, complex tangential and satisfies γα((−1,1))⊂Z(p)∩S2\gamma_\alpha((-1,1))\subset\mathcal{Z}(p)\cap\mathbb{S}_{2}, γα(0)=α\gamma_\alpha(0)=\alpha. In particular, it can be parametrized as follows: γα(t)=(t,1−t2)\gamma_\alpha(t)=(t,\sqrt{1-t^{2}}) which yields a submanifold lying in Z(p)∩S2\mathcal{Z}(p)\cap\mathbb{S}_{2}. See [9] for crucial and novel examples of stable polynomials.

The results below examine whether the Nash curve γα\gamma_\alpha may be chosen to satisfy γα′(0)≠0\gamma'_\alpha(0)\ne0.

Main results

The following lemma is a stronger version of the Nash curve selection lemma for the set Z(p)∩S2\mathcal{Z}(p) \cap S_2 of a stable polynomial p∈C[z,w]p \in\mathbb{C}[z,w]. The proof describes also the Weierstrass polynomial of the local Weierstrass factorization of the polynomial.

Lemma 1. Let p∈C[z,w]p \in\mathbb{C}[z,w] be a stable polynomial with Z(p)∩S2=⋃iMi\mathcal{Z}(p)\cap S_2 = \bigcup_i M_i, and let α∈Mi‾∖Mi\alpha\in\overline{M_i}\setminus M_i, for any fixed ii. There exists a Nash curve ωα:(−1,1)→Z(p)∩S2\omega_\alpha: (-1,1) \to\mathcal{Z}(p) \cap S_2 that is a diffeomorphism onto its image and satisfies ωα((0,1))⊂Mi\omega_\alpha((0,1)) \subset M_i, ωα(0)=α\omega_\alpha(0) = \alpha.

Proof. For simplicity assume that pp is irreducible and α:=(0,1)\alpha:= (0,1). By the Nash curve selection lemma (see the Lemma in § 1), the identity principle and the arguments in the proof of [19 Theorem 1], there exists a Nash curve γα=(γ1,γ2):(−1,1)→Z(p)∩S2\gamma_\alpha= (\gamma_1,\gamma_2) : (-1,1) \to\mathcal{Z}(p) \cap S_2, which is homeomorphism onto its image, complex tangential, and γα((0,1))⊂Mi\gamma_\alpha((0,1)) \subset M_i, γα(0)=α\gamma_\alpha(0) = \alpha.

Recall the discussion concerning the Weierstrass Preparation Theorem (see [16 14.1.3. Theorem]) and the application of Puiseux’s Parametrization Theorem (see [4 pp. 136–137] and [13 Corollary, pp. 171–171]) in [8 pp. 33–34] and [10] [349, 350]. We may expand

p(z,w)=A(z,w)(1−h1(z)w)⋯(1−hN(z)w),(z,w)∈Δ(0,ϵ)×Δ(1,ϵ),p(z,w) = A(z,w)(1-h_1(z)w)\cdots(1-h_N(z)w), \qquad(z,w) \in\Delta(0,\epsilon)\times\Delta(1,\epsilon),

where AA is holomorphic and zero free on the polydisk Δ(0,ϵ)×Δ(1,ϵ)\Delta(0,\epsilon)\times\Delta(1,\epsilon), and each function hjh_j is well defined, singled-valued holomorphic function on cut-branch subdomains of Δ(0,ϵ)\Delta(0,\epsilon) with respect to 0.

In particular, there exists δ>0\delta> 0, a non-negative integer k>0k > 0, and holomorphic functions Φ1,…,ΦN∈Hol⁡(Δ(0,δ))\Phi_1,\ldots,\Phi_N \in\operatorname{Hol}(\Delta(0,\delta)) such that hj(z)=Φi(j)(z1/k)h_j(z) = \Phi_{i(j)}(z^{1/k}), for a properly chosen branch and zz close to 0. Fix jj and i(j):=ii(j) := i. Since pp is a stable polynomial, we obtain that

∣Φi(z)∣2≤11−∣z∣2k,or,∣z∣2k+1∣Φi(z)∣2−1≥0,(1)|\Phi_i(z)|^2 \le\frac{1}{1-|z|^{2k}}, \qquad\text{or,} \qquad|z|^{2k}+\frac{1}{|\Phi_i(z)|^2}-1 \ge0, \tag*{(1)}

for all zz close to 0.

Then there exists jj (for simplicity, put j=i(j)=ij=i(j)=i) and ϵ>0\epsilon> 0 such that

hi(γ1(t))=1γ2(t),(2)h_i(\gamma_1(t)) = \frac{1}{\gamma_2(t)}, \tag*{(2)}

for all t∈(0,ϵ)t \in(0,\epsilon) and fixed ϵ>0\epsilon> 0.

The conditions above yield the following classification:

hi(z)=Φi(z1/k)=1+λi0z2+λi1z2+1/k+λi2z2+2/k⋯=1+λi0z2+o(z2),(3)h_i(z) = \Phi_i(z^{1/k}) = 1+\lambda_i^0z^2+\lambda_i^1z^{2+1/k}+\lambda_i^2z^{2+2/k}\cdots= 1+\lambda_i^0z^2+o(z^2), \tag*{(3)}

where ∣λi0∣=1/2|\lambda_i^0|=1/2, see [10]. For the moment, hih_i may be not holomorphic.

Let us suppose first that h:=hih := h_i is a holomorphic function on Δ(0,ϵ)\Delta(0,\epsilon), for some ϵ>0\epsilon> 0 such that 1/h1/h is well defined and holomorphic function on Δ(0,ϵ)\Delta(0,\epsilon) as well. This case can be handled utilizing the techniques in the proof of [18 Proposition 4.6]; however, for completeness, we provide the proof. Indeed, define φ:Δ(0,ϵ)→R\varphi: \Delta(0,\epsilon) \to\mathbb{R} by

φ(z):=∣z∣2+∣h(z)∣−2−1\varphi(z) := |z|^2+|h(z)|^{-2}-1

Stability of the polynomial yields φ(z)≥0\varphi(z) \ge0, for all zz. A straightforward computation shows that ∂zz‾φ(0)>0\partial_{z\overline{z}}\varphi(0)>0. Whence, there exists a real analytic curve ω~:(−1,1)→C\widetilde{\omega} : (-1,1) \to\mathbb{C} which is diffeomorphism onto its image, ω~(0)=0\widetilde{\omega}(0)=0, and there exists ϵ′>0\epsilon' > 0 such that Z(φ)∩Δ(0,ϵ′)⊂ω~((−1,1))\mathcal{Z}(\varphi)\cap\Delta(0,\epsilon') \subset\widetilde{\omega}((-1,1)), see [6] (or, the proof of [16 Theorem 10.7.6]; apply the real analytic version of implicit function theorem, see [11]). Define ωα:(−1,1)→C2\omega_\alpha: (-1,1) \to\mathbb{C}^2 by ωα(t):=(ω~(t),1/h(ω~(t))\omega_\alpha(t):=(\widetilde{\omega}(t),1/h(\widetilde{\omega}(t)). Then ωα\omega_\alpha is a real analytic curve, diffeomorphism onto its image, and ωα(0)=α\omega_\alpha(0)=\alpha. (2) yields that Z(φ)∩Δ(0,ϵ)\mathcal{Z}(\varphi)\cap\Delta(0,\epsilon) contains a specific set with an accumulation point, hence, ω~((−1,1))\widetilde{\omega}((-1,1)) contains this set too. Summing up, p(ωα(t))=0p(\omega_\alpha(t))=0, and ∥ωα(t)∥2=1\lVert\omega_\alpha(t)\rVert^2=1, on a set having an accumulation point. Identity principle ensures that ωα:(−1,1)→Z(p)∩S2\omega_\alpha:(-1,1)\to\mathcal{Z}(p)\cap S_2. By the nature of the curves γα\gamma_\alpha, ωα\omega_\alpha, and the conditions that satisfy, we have ωα((0,ϵ))⊂Mi\omega_\alpha((0,\epsilon))\subset M_i for some fixed ϵ>0\epsilon>0, and ωα(0)=α\omega_\alpha(0)=\alpha. In particular, ωα((0,ϵ))∩Mi\omega_\alpha((0,\epsilon))\cap M_i is connected and open in the relative topology of MiM_i, α∈Mi∩ωα((0,ϵ))\alpha\in M_i\cap\omega_\alpha((0,\epsilon)), and ωα((−ϵ,ϵ))\omega_\alpha((-\epsilon,\epsilon)) has exactly two points on its boundary. Ergo, utilizing ωα\omega_\alpha we derive the desired Nash curve.

We shall prove that stability forces hh to be always a holomorphic function and the assertion will follows from the previous step. The idea is to apply again the implicit function theorem (as in the proof of [16 Theorem 10.7.6]) together with the stability condition in order to eliminate the terms that makes hh a non-holomorphic function. We slightly switch the stability inequality (1) to an equivalent one.

Put Φ:=Φi\Phi:=\Phi_i and λli=λl\lambda_l^i=\lambda_l, l=0,1,…l=0,1,\ldots. If k=1k=1, then indeed it is holomorphic. Let us suppose in contrary that k>1k>1, and there exists at least one l∉{0,k,2k,3k,…}l\notin\{0,k,2k,3k,\ldots\} such that λl≠0\lambda_l\ne0. Note that Φ(0)=1\Phi(0)=1, hence, there exists a function Ψ\Psi which is holomorphic close to 00, and such that Φ(z)=eΨ(z)\Phi(z)=e^{\Psi(z)}. The inequality (1) yields the equivalent one

Re⁡Ψ(z)≤σ(∣z∣2k),(4)\operatorname{Re}\Psi(z)\leq\sigma(|z|^{2k}), \tag*{(4)}

for all zz close to 00, where σ(θ):=−12log⁡(1−θ)=θ2+θ24+…\sigma(\theta):=-\frac{1}{2}\log(1-\theta)=\frac{\theta}{2}+\frac{\theta^2}{4}+\ldots, ∣θ∣<1|\theta|<1. In particular, Ψ\Psi has a Taylor series expansion of the form

Ψ(z)=log⁡(1−(1−Φ(z)))=−∑q=1+∞1q(1−Φ(z))q=λ0z2k+o(z2k).(5)\Psi(z)=\log(1-(1-\Phi(z)))=-\sum_{q=1}^{+\infty}\frac{1}{q}(1-\Phi(z))^q=\lambda_0z^{2k}+o(z^{2k}). \tag*{(5)}

The next goal is to remove the ”non-holomorphic part” of the series expansion of Ψ(z1/k)\Psi(z^{1/k}). Define the function FF by

F(zk):=1k∑j=0k−1Ψ(ζjz),(6)F(z^k):=\frac{1}{k}\sum_{j=0}^{k-1}\Psi(\zeta^jz), \tag*{(6)}

where ζ:=e2πi1k\zeta:=e^{2\pi i\frac{1}{k}}. We shall show that FF is a well defined holomorphic function close to 00. Indeed, it is enough to check the following finite sums:

1k∑j=0k−1∑q=1Q(−1)q+1q(∑l=0Lλl(ζjz)2k+l)q,(7)\frac{1}{k}\sum_{j=0}^{k-1}\sum_{q=1}^{Q}\frac{(-1)^{q+1}}{q}\left(\sum_{l=0}^{L}\lambda_l(\zeta^jz)^{2k+l}\right)^q, \tag*{(7)}

where 1−Φ(z)=−∑l=0+∞λlz2k+l1-\Phi(z)=-\sum_{l=0}^{+\infty}\lambda_lz^{2k+l}. Recall the average equality

1k∑j=0k−1ζjm={1,if k Z-divides m,0,otherwise.(8)\frac{1}{k}\sum_{j=0}^{k-1}\zeta^{jm}= \begin{cases} 1, & \text{if } k\ \mathbb{Z}\text{-divides }m,\\ 0, & \text{otherwise}. \end{cases} \tag*{(8)}

The multinomial formula and a straightforward computation yield

1k∑j=0k−1(∑l=0Lλl(ζjz)2k+l)q=1k∑j=0k−1∑∣ι∣=qq!ι!λ0ι0⋯λLιL(ζjz)2kι0⋯(ζjz)(2k+L)ιL=∑∣ι∣=qq!ι!λ0ι0⋯λLιL[1k∑j=0k−1ζj(2kι0+(2k+1)ι1+⋯+(2k+L)ιL)]z2kι0+(2k+1)ι1+⋯+(2k+L)ιL=∑∣ι∣=qq!ι!λ0ι0⋯λLιL[1k∑j=0k−1ζj(2kq+ι1+⋯+LιL)]z2kq+ι1+⋯+LιL.\begin{aligned} \frac{1}{k}\sum_{j=0}^{k-1}\left(\sum_{l=0}^{L}\lambda_l(\zeta^jz)^{2k+l}\right)^q =\frac{1}{k}\sum_{j=0}^{k-1}\sum_{\lvert\iota\rvert=q}\frac{q!}{\iota!}\lambda_0^{\iota_0}\cdots\lambda_L^{\iota_L}(\zeta^jz)^{2k\iota_0}\cdots(\zeta^jz)^{(2k+L)\iota_L} \\ &= \sum_{|\boldsymbol{\iota}|=q} \frac{q!}{\boldsymbol{\iota}!}\lambda_0^{\iota_0}\cdots\lambda_L^{\iota_L}\left[\frac{1}{k}\sum_{j=0}^{k-1}\zeta^{j(2k\iota_0+(2k+1)\iota_1+\cdots+(2k+L)\iota_L)}\right]z^{2k\iota_0+(2k+1)\iota_1+\cdots+(2k+L)\iota_L} \\ &= \sum_{|\boldsymbol{\iota}|=q} \frac{q!}{\boldsymbol{\iota}!}\lambda_0^{\iota_0}\cdots\lambda_L^{\iota_L}\left[\frac{1}{k}\sum_{j=0}^{k-1}\zeta^{j(2kq+\iota_1+\cdots+L\iota_L)}\right]z^{2kq+\iota_1+\cdots+L\iota_L}. \end{aligned}

An application of the equality (8) shows that the non-zero coefficients of the monomials z2kq+ι1+⋯+LιLz^{2kq+\iota_1+\cdots+L\iota_L} in the computation above, are those that kk Z\mathbb{Z}-divides the positive integer 2kq+ι1+⋯+LιL2kq+\iota_1+\cdots+L\iota_L. Summing up, we reach a classification of the holomorphic function FF:

F(z)=λ0z2+o(z2),andF((ζjw)k)=F(wk),j=0,1,…,k−1,(9)F(z)=\lambda_0z^2+o(z^2),\qquad\text{and}\qquad F((\zeta^jw)^k)=F(w^k),\qquad j=0,1,\ldots,k-1, \tag*{(9)}

for all z,wz,w close to 00, where o(z2)o(z^2) is a holomorphic function and its power series expansion consists of terms cmzmc_mz^m, m≥3m\ge3. The second equality in (9) follows immediately from the classification of FF and the condition ζjmk=1\zeta^{jmk}=1, for all m=1,2,…m=1,2,\ldots Let us point out that the series expansion of the difference Ψ(z)−F(zk)\Psi(z)-F(z^k) consists only of terms clz2k+lc_lz^{2k+l}, where l∉{0,k,2k,…}l\notin\{0,k,2k,\ldots\}; averaging Ψ\Psi actually cancels the rest of the terms in this difference.

Next, define the real analytic function uu and the holomorphic function δ\delta, close to 00, by

u(z):=σ(∣z∣2)−Re⁡F(z),andδ(z):=Ψ(z)−F(zk),(10)u(z):=\sigma(|z|^2)-\operatorname{Re}F(z),\qquad\text{and}\qquad\delta(z):=\Psi(z)-F(z^k), \tag*{(10)}

respectively. We obtain,

u(z)=1k∑j=0k−1σ(∣ζjw∣2k)−1k∑j=0k−1Re⁡Ψ(ζjw)=1k∑j=0k−1[σ(∣ζjw∣2k)−Re⁡Ψ(ζjw)]≥0,u(z)=\frac{1}{k}\sum_{j=0}^{k-1}\sigma(|\zeta^jw|^{2k})-\frac{1}{k}\sum_{j=0}^{k-1}\operatorname{Re}\Psi(\zeta^jw)=\frac{1}{k}\sum_{j=0}^{k-1}\left[\sigma(|\zeta^jw|^{2k})-\operatorname{Re}\Psi(\zeta^jw)\right]\ge0,

where wk=zw^k=z, j=0,1,…,k−1j=0,1,\ldots,k-1, and zz close to 00; the non-negativity comes from (4). Moreover, applying (9) and (4), we see that

u(z)=σ(∣ζjw∣2k)−Re⁡Ψ(ζjw)+Re⁡Ψ(ζjw)−Re⁡F(wk)≥Re⁡[Ψ(ζjw)−F((ζjw)k)]=Re⁡δ(ζjw).(11)\begin{aligned} u(z)&=\sigma(|\zeta^jw|^{2k})-\operatorname{Re}\Psi(\zeta^jw)+\operatorname{Re}\Psi(\zeta^jw)-\operatorname{Re}F(w^k)\\ &\ge\operatorname{Re}\left[\Psi(\zeta^jw)-F((\zeta^jw)^k)\right]\\ &=\operatorname{Re}\delta(\zeta^jw). \tag*{(11)} \end{aligned}

We shall classify uu. Note that ∂zzˉu(0)>0\partial_{z\bar z}u(0)>0, hence, one can therefore rotate coordinates in C\mathbb{C} so that

u(z)=ax2+by2+o(∣z∣2),u(z)=ax^2+by^2+o(|z|^2),

where z:=x+iyz:=x+iy and b>0b>0, see [6]. Whence, ∂yyu(0)=2b≠0\partial_{yy}u(0)=2b\ne0, and the real analytic version of implicit function theorem applied to ∂yu\partial_yu, yields a unique real analytic function b0(x)b_0(x) with b0(0)=b0′(0)=0b_0(0)=b'_0(0)=0, see [11]. In particular, b0(x)=O(x2)b_0(x)=O(x^2). Set a0(x):=u(x,b0(x))≥0a_0(x):=u(x,b_0(x))\ge0. By the Taylor expansion with respect to a0a_0 and b0b_0, we obtain

u(z)=a0(x)+(y−b0(x))2ϵ(z),(12)u(z)=a_0(x)+(y-b_0(x))^2\epsilon(z), \tag*{(12)}

for all zz close enough to 00, where ϵ\epsilon is real analytic, and ϵ(0)=b\epsilon(0)=b.

We shall show that δ≡0\delta\equiv0 utilizing inequality (11). This identification leads to Ψ(z)≡Φ(zk)\Psi(z)\equiv\Phi(z^k), thereby proving that hh is actually an holomorphic function.

Indeed, let us suppose in contrary that δ\delta is not identically zero. The Taylor power series expansions of Ψ\Psi and FF, together with the assumptions k>1k>1 and λl≠0\lambda_l\ne0, for at least one l∉{0,k,2k,3k,…}l \notin\{0,k,2k,3k,\ldots\}, yield

δ(z)=λ∗zℓ+o(zℓ),(13)\delta(z)=\lambda_*z^\ell+o(z^\ell), \tag*{(13)}

where λ∗∈{λl:l∉{0,k,2k,3k,…}}\lambda_* \in\{\lambda_l:l \notin\{0,k,2k,3k,\ldots\}\}, λ∗≠0\lambda_* \ne0, and the positive integer ℓ\ell cannot be Z\mathbb{Z}-divided by kk and it satisfies ℓ≥2k+1\ell\ge2k+1. Set r:=ℓ/k∈Qr:=\ell/k \in\mathbb{Q}, and

M:=max⁡j=0,…,k−1Re⁡(λ∗e2πijr),M~:=max⁡j=0,…,k−1Re⁡(λ∗eiπre2πijr).M:=\max_{j=0,\ldots,k-1}\operatorname{Re}(\lambda_*e^{2\pi ijr}),\qquad\widetilde{M}:=\max_{j=0,\ldots,k-1}\operatorname{Re}(\lambda_*e^{i\pi r}e^{2\pi ijr}).

Then M>0M>0 or M~>0\widetilde{M}>0. Indeed, the average equality (8) applied for e2πijr=ζjℓe^{2\pi ijr}=\zeta^{j\ell}, shows that

1k∑j=0k−1Re⁡(λ∗e2πijr)=1k∑j=0k−1Re⁡(λ∗eiπre2πijr)=0;(14)\frac{1}{k}\sum_{j=0}^{k-1}\operatorname{Re}(\lambda_*e^{2\pi ijr})=\frac{1}{k}\sum_{j=0}^{k-1}\operatorname{Re}(\lambda_*e^{i\pi r}e^{2\pi ijr})=0; \tag*{(14)}

this forces MM to be non-negative. Assume that M=0M=0; we shall show that M~>0\widetilde{M}>0. If Re⁡(λ∗e2πijr)≠0\operatorname{Re}(\lambda_*e^{2\pi ijr})\ne0, for some jj, then, by (14), there exists at least one j′j' (possibly j′=jj'=j) such that Re⁡(λ∗e2πij′r)>0\operatorname{Re}(\lambda_*e^{2\pi ij'r})>0, which contradicts the assumption M=0M=0. Hence, M=0M=0 yields Re⁡(λ∗e2πijr)=0\operatorname{Re}(\lambda_*e^{2\pi ijr})=0 for all j=0,…,k−1j=0,\ldots,k-1. In particular, for j=0j=0: λ∗=Im⁡λ∗≠0\lambda_*=\operatorname{Im}\lambda_* \ne0, and for j=1j=1: Re⁡(λ∗e2πir)=Im⁡λ∗sin⁡(2πr)=0\operatorname{Re}(\lambda_*e^{2\pi ir})=\operatorname{Im}\lambda_*\sin(2\pi r)=0. Whence, r=q/2r=q/2, for some q=1,3,5,…q=1,3,5,\ldots, and eiπr=±ie^{i\pi r}=\pm i. Ergo, equation (14) yields M~>0\widetilde{M}>0.

As we have already mentioned, we shall utilize (11) to get contradiction, so we should study the behaviour of uu and δ\delta near 0∈C0\in\mathbb{C}.

First, we deal with Re⁡δ(e2πij/kz1/k)\operatorname{Re}\delta(e^{2\pi ij/k}z^{1/k}), as z→0z\to0 through a proper path. This proper path comes from the application of implicit function theorem. Recall that b0(x)=O(x2)b_0(x)=O(x^2), and put z=z(x):=x+ib0(x)=x(1+O(x))z=z(x):=x+ib_0(x)=x(1+O(x)). Letting x>0x>0, we may substitute ζjz1/k=(e2πijz)1/k\zeta^{j}z^{1/k}=(e^{2\pi ij}z)^{1/k}, j=0,1,…,k−1j=0,1,\ldots,k-1, to the Taylor expansion (13) of the holomorphic function δ\delta (picking first a proper branch-cut around zero; this is possible since (x,b0(x))(x,b_0(x)), x≥0x\ge0, yields a real analytic curve on the half complex plain reaching the end point 0∈C0\in\mathbb{C}), namely,

δ(ζjz1/k)=xr[λ∗e2πijr(1+O(x))r+o(xs)],(15)\delta(\zeta^jz^{1/k})=x^r[\lambda_*e^{2\pi ijr}(1+O(x))^r+o(x^s)], \tag*{(15)}

for all x>0x>0 close to 00, and some s>0s>0. Similarly, if x<0x<0, then x=∣x∣eiπx=|x|e^{i\pi}, and thus,

δ(ζjz1/k)=∣x∣r[λ∗eiπre2πijr(1+O(∣x∣))r+o(∣x∣s)],(16)\delta(\zeta^jz^{1/k})=|x|^r[\lambda_*e^{i\pi r}e^{2\pi ijr}(1+O(|x|))^r+o(|x|^s)], \tag*{(16)}

for all x<0x<0 close to 00. Recall that M>0M>0 or M~>0\widetilde{M}>0. Without loss of generality, let us suppose that M>0M>0 (we may follow the same procedure for M~\widetilde{M} since ∣x∣>0|x|>0.) Then there exists j∈{0,…,k−1}j\in\{0,\ldots,k-1\} such that Re⁡(λ∗e2πijr)>0\operatorname{Re}(\lambda_*e^{2\pi ijr})>0. For this particular jj, we have then

Re⁡(λ∗e2πijr(1+O(x))r+o(xs))⟶Re⁡(λ∗e2πijr)>0,as x→0+.\operatorname{Re}\bigl(\lambda_*e^{2\pi ijr}(1+O(x))^r+o(x^s)\bigr)\longrightarrow\operatorname{Re}(\lambda_*e^{2\pi ijr})>0,\qquad\text{as }x\to0^+.

Equations (15) and (16), together with the fact M>0M>0 or M~>0\widetilde{M}>0, yield

Re⁡δ(ζjz1/k)≥c∣x∣r,(17)\operatorname{Re}\delta(\zeta^jz^{1/k})\ge c|x|^r, \tag*{(17)}

for some fixed jj, and for all xx close to 00, either by the side x≥0x\ge0, or the side x≤0x\le0. The constant c>0c>0 is uniform with respect to xx.

In what follows, each constants c,C>0c,C>0 are uniform with respect to the variables z,w,x,yz,w,x,y. Summing up, the inequalities (11), (17) hold for all z=x+ib0(x)z=x+ib_0(x), wk=zw^k=z, where either x≥0x\ge0, or x≤0x\le0, close enough to 00. For simplicity choose the side x≥0x\ge0. Since a0(x)=u(x,b0(x))≥0a_0(x)=u(x,b_0(x))\ge0, we have

a0(x)≥c∣x∣r,(18)a_0(x)\ge c|x|^r, \tag*{(18)}

for all x>0x>0 close to 0. This shows that the real analytic function a0a_0 is not identically zero. Let a0(x)=dmxm+dm+1xm+1+⋯a_0(x)=d_mx^m+d_{m+1}x^{m+1}+\cdots, where dm≠0d_m\ne0, and m=1,2,…m=1,2,\ldots. Then

a0(x)≥c∣x∣m,(19)a_0(x)\ge c|x|^m, \tag*{(19)}

for all xx close to 0 (that is, in both sides x≥0x\ge0 and x≤0x\le0). More specifically, mm must be even due to the positivity of a0a_0, dm>0d_m>0, and the inequality (18) together with the condition r∈Qr\in\mathbb{Q}, yield m<rm<r.

Furthermore, setting β(x,y):=y−b0(x)\beta(x,y):=y-b_0(x), and recalling that ϵ(0)=b>0\epsilon(0)=b>0, we obtain, by (19) and (12), that

u(z)=a0(x)+β(x,y)2ϵ(z)≥c(∣x∣m+β(x,y)2),(20)u(z)=a_0(x)+\beta(x,y)^2\epsilon(z)\ge c\bigl(|x|^m+\beta(x,y)^2\bigr), \tag*{(20)}

for all z=x+iyz=x+iy close to 0.

On the other hand, (13) yields ∣δ(w)∣≤c∣z∣r|\delta(w)|\le c|z|^r, for all z=wkz=w^k. In particular, since y=b0(x)+β(x,y)y=b_0(x)+\beta(x,y), b0(x)=O(x2)b_0(x)=O(x^2), and thus

z=x+iy=x+ib0(x)+iβ(x,y)=x(1+O(x))+iβ(x,y),z=x+iy=x+ib_0(x)+i\beta(x,y)=x(1+O(x))+i\beta(x,y),

we get

∣δ(w)∣≤C(∣x∣r+∣β(x,y)∣r),(21)|\delta(w)|\le C\bigl(|x|^r+|\beta(x,y)|^r\bigr), \tag*{(21)}

for all ww close enough to 0. Recall that r>mr>m and r>2r>2.

Summarizing, the definitions (10) yield the equality

u(wk)−Re⁡δ(w)=σ(∣w∣2k)−Re⁡F(wk)−Re⁡Ψ(w)+Re⁡F(wk)=σ(∣w∣2k)−Re⁡Ψ(w),(22)\begin{aligned} u(w^k)-\operatorname{Re}\delta(w)&=\sigma(|w|^{2k})-\operatorname{Re}F(w^k)-\operatorname{Re}\Psi(w)+\operatorname{Re}F(w^k)\\ &=\sigma(|w|^{2k})-\operatorname{Re}\Psi(w), \tag*{(22)} \end{aligned}

for all wk=zw^k=z close to 0, and the inequalities (20), (21) yield

u(wk)−Re⁡δ(w)≥u(wk)−∣δ(w)∣≥c(∣x∣m+β(x,y)2)−C(∣x∣r+∣β(x,y)∣r),u(w^k)-\operatorname{Re}\delta(w)\ge u(w^k)-|\delta(w)|\ge c\bigl(|x|^m+\beta(x,y)^2\bigr)-C\bigl(|x|^r+|\beta(x,y)|^r\bigr),

for all z=x+iyz=x+iy and z=wkz=w^k, where xx and yy close enough to 0∈R0\in\mathbb{R}. Since r>mr>m and r>2r>2, we obtain that

u(wk)−Re⁡δ(w)>0,(23)u(w^k)-\operatorname{Re}\delta(w)>0, \tag*{(23)}

for all non-zero zz and z=wkz=w^k sufficiently close to 0.

Last, the proofs of (2), (4), (22), and (23) lead to contradiction on the assumption that δ\delta is not identically 0.

It turns out that δ≡0\delta\equiv0, and hence, Ψ(z)=F(zk)\Psi(z)=F(z^k), for all zz in a fixed neighbourhood of 0. In particular,

h(z)=eF(z),h(z)=e^{F(z)},

namely, there exists ϵ>0\epsilon>0 such that h∈Hol⁡(Δ(0,ϵ))h\in\operatorname{Hol}(\Delta(0,\epsilon)). The assertion follows.

In light of the Lemma 1, we achieve a characterization of the boundary zero set of a stable polynomial in B2\mathbb{B}_2 (which is a stronger version of [19 Theorem 1]). See also [9 Proposition 1.3-Theorem 1.5] for a different approach.

Theorem 2. Let p∈C[z,w]p \in\mathbb{C}[z,w] be a stable polynomial such that Z(p)∩S2≠∅Z(p) \cap\mathbb{S}_2 \ne\varnothing. Then Z(p)∩S2Z(p) \cap\mathbb{S}_2 is either a finite set or, the non-isolated part of Z(p)∩S2Z(p) \cap\mathbb{S}_2 coincide with ⋃iMi\bigcup_i M_i, where MiM_i, i=1,2,…,ki=1,2,\ldots,k, k≥2k \ge2, are Nash submanifold lying in S2\mathbb{S}_2, each equivalent to (−1,1)(-1,1) by a complex tangential Nash diffeomorphism. Each MiM_i contains exactly two points on its boundary Mi‾∖Mi\overline{M_i} \setminus M_i. In particular, for each ii and each α∈Mi‾∖Mi\alpha\in\overline{M_i} \setminus M_i, there exists j≠ij \ne i such that α∈Mj\alpha\in M_j, α∈Mi‾∩Mj\alpha\in\overline{M_i} \cap M_j, and Mi∩MjM_i \cap M_j is connected open set in the relative topology of MiM_i. Furthermore, each Nash diffeomorphism ϕi:(−1,1)→Mi\phi_i : (-1,1) \to M_i—under the application of a unitary matrix UiU_i with respect to α=(0,1)\alpha= (0,1)—is locally, around 0∈(−1,1)0 \in(-1,1), classified as follows:

Uiϕi(t)=(βit+o(t),1−∣βi∣22t2+o(t2)),U_i\phi_i(t) = \left(\beta_i t + o(t),1-\frac{|\beta_i|^2}{2}t^2+o(t^2)\right),

for all t∈(−ϵ,ϵ)t \in(-\epsilon,\epsilon), where βi∈C∖{0}\beta_i \in\mathbb{C} \setminus\{0\}, and ϵ>0\epsilon> 0 fixed.

Proof. By the discussion in Section 2, if dim⁡RZ(p)∩S2=1\dim_{\mathbb{R}} Z(p) \cap\mathbb{S}_2 = 1, then the non-isolated part coincides with ⋃iMi\bigcup_i M_i, where MiM_i, are finitely many, pairwise disjoint, complex tangential Nash submanifold lying in S2\mathbb{S}_2, each Nash diffeomorphic to (−1,1)mi(-1,1)^{m_i}, mi=0,1m_i = 0,1. In particular, each one-dimensional component MiM_i has exactly two points on its boundary Mi‾∖Mi\overline{M_i} \setminus M_i. Since the components are homeomorphic to (−1,1)(-1,1), there exists at least one Mi0M_{i_0} with mi0=0m_{i_0}=0. This case mi0=0m_{i_0}=0 yields just an accumulation point, namely, the compact manifold Mi0={α}M_{i_0}=\{\alpha\}. In particular, α∈Mj0‾∖Mj0\alpha\in\overline{M_{j_0}} \setminus M_{j_0}, for some fixed j0≠i0j_0 \ne i_0. Applying the proof of Lemma 1 to this accumulation point α\alpha and the arguments concerning the complex tangential property in the proof of [19 Theorem 1], we obtain a Nash submanifold which is equivalent to (−1,1)(-1,1) by a complex tangential Nash diffeomorphism, lies in Z(p)∩S2Z(p) \cap\mathbb{S}_2, α∈Nα\alpha\in N_\alpha, α∈Nα‾∩Mj0‾\alpha\in\overline{N_\alpha} \cap\overline{M_{j_0}}, and Nα∩Mj0N_\alpha\cap M_{j_0} is connected open set in the relative topology of Mj0M_{j_0}. Last, for each MiM_i with mi=1m_i=1, we have that ⋃i,mi=1(Mi‾∖Mi)\bigcup_{i,m_i=1}(\overline{M_i} \setminus M_i) is a finite subset of Z(p)∩S2Z(p) \cap\mathbb{S}_2. Each point α∈⋃i,mi=1(Mi‾∖Mi)\alpha\in\bigcup_{i,m_i=1}(\overline{M_i} \setminus M_i) lies on a finite number of Mi‾∖Mi\overline{M_i} \setminus M_i (probably more than two), and hence, applying the previous argument to each one of them separately, it lies on a finite number of such NαN_\alpha. Add the finitely many NαN_\alpha on the partition {Mi}i,mi=1\{M_i\}_{i,m_i=1}. Last, the classification of the Nash diffeomorphisms follows from the complex-tangential property, see [18]. □

Further discussion

Remark 3. We note that [9 Theorem 1.5 (1)] singles out the curve-type boundary zero case, establishing that the underlying (complex) algebraic curve intersects the boundary S2\mathbb{S}_2 along a real one-dimensional analytic curve. Hence, Lemma 1 serves as an alternative, independent proof of the existence of such a curve lying in the intersection of a complex algebraic curve (associated with a stable polynomial) with the boundary sphere. In particular, Lemma 1 explicitly guarantees that this real analytic curve is non-singular (diffeomorphism onto its image), providing additional geometric regularity to the boundary trace.

Remark 4. Roughly speaking Theorem 2 says that an one-dimensional boundary zero set of a stable polynomial is either a union of finitely many isolated points and finitely many one-dimensional compact manifolds in S2\mathbb{S}_2 or, a union of finitely many isolated points and finitely many one-dimensional submanifolds in S2\mathbb{S}_2 that ”cross each other” (i.e., ”cusps” cannot occur; see Section 2 about the term ”cusp”). By the term ”cross each other”, we mean that locally—around a singular point α∈Z(p)∩S2\alpha\in Z(p) \cap\mathbb{S}_2—the boundary zero set mirrors the novel example i−1+2z2−(i+1)z22+z12i-1+2z_2-(i+1)z_2^2+z_1^2 of [9] around its singular point; with probably an arbitrary large number of finitely many half-branches and not just four. This example serves as a crucial stable polynomial; beyond the well known polynomial 1−z2−w21-z^2-w^2–and its variations by a unitary matrix–whose boundary zero set is just a compact manifold in S2S_2. The existence of such polynomials is essential and goes back to the discussion in [16 Chapter 10] and the references within.

Remark 5. Denote the open ball with center α∈C2\alpha\in\mathbb{C}^2 and radial ϵ>0\epsilon> 0 by B(α,ϵ):={(z,w)∈C2:∥(z,w)−α∥<ϵ}B(\alpha,\epsilon) := \{(z,w) \in\mathbb{C}^2 : \lVert(z,w)-\alpha\rVert< \epsilon\}. Note that Lemma 1 together with Theorem 2 rule out the existence of a hypothetical closed (real) algebraic curve γα([−1,1])⊂Z(p)∩S2\gamma_\alpha([-1,1]) \subset\mathcal{Z}(p) \cap S_2 and ϵ0>0\epsilon_0 > 0, such that γα(0)=α\gamma_\alpha(0)=\alpha, and for each 0<ϵ<ϵ00 < \epsilon< \epsilon_0, the curve satisfies γα([−1,1])∩B(α,ϵ)=(M0∪M1∪M2∪M3∪M4)∩B(α,ϵ)\gamma_\alpha([-1,1]) \cap B(\alpha,\epsilon) = (M_0 \cup M_1 \cup M_2 \cup M_3 \cup M_4) \cap B(\alpha,\epsilon), where M0={α}M_0 = \{\alpha\}, MiM_i are pairwise disjoint submanifolds with dim⁡RMi=1\dim_{\mathbb{R}} M_i = 1, i=1,2,3,4i=1,2,3,4, and ⋂iMi=M0\bigcap_i M_i = M_0, and M0∪M1∪M2M_0 \cup M_1 \cup M_2 yields a submanifold while M0∪M3∪M4M_0 \cup M_3 \cup M_4 does not (i.e. M3∪M4M_3 \cup M_4 builds a ”cusp” at α\alpha). Although the point α\alpha lies on the submanifold M0∪M1∪M2M_0 \cup M_1 \cup M_2 which is contained in Z(p)∩S2\mathcal{Z}(p) \cap S_2, it also lies on M0∪M3∪M4M_0 \cup M_3 \cup M_4 which is contained in Z(p)∩S2\mathcal{Z}(p) \cap S_2. Hence, Lemma 1 together with Theorem 2 serve as a stronger result than proving that for each α∈Z(p)∩S2\alpha\in\mathcal{Z}(p) \cap S_2 there exists a Nash curve γα:(−1,1)→Z(p)∩S2\gamma_\alpha: (-1,1) \to\mathcal{Z}(p) \cap S_2, that is diffeomorphism onto it image and satisfies γα(0)=α\gamma_\alpha(0)=\alpha. See Remark 9.5.8 together with Subsections 9.3 and 9.5 of [1] on the half-branches of algebraic curves. For a local description, Theorem 2 says that for each point α∈Z(p)∩S2\alpha\in\mathcal{Z}(p) \cap S_2, there exist finitely many one dimensional Nash submanifolds N1,…,Nk⊂Z(p)∩S2N_1,\ldots,N_k \subset\mathcal{Z}(p) \cap S_2 and ϵ0>0\epsilon_0 > 0 such that α∈Ni\alpha\in N_i, for all i=1,…,ki=1,\ldots,k, and Z(p)∩S2∩B(α,ϵ)=(N1∪⋯∪Nk)∩B(α,ϵ)\mathcal{Z}(p) \cap S_2 \cap B(\alpha,\epsilon) = (N_1 \cup\cdots\cup N_k) \cap B(\alpha,\epsilon), for all 0<ϵ<ϵ00 < \epsilon< \epsilon_0.

Remark 6. Recalling the [18 Definition 4.3] of essential singularities and weak essential singularities (these definitions coincide in the setting B2\mathbb{B}_2), Theorem 2 proves that given a stable polynomial p∈C[z,w]p \in\mathbb{C}[z,w], the set Z(p)∩S2\mathcal{Z}(p) \cap S_2 has no essential singularities. As we have already mentioned in the introduction, Lemma 1 and Theorem 2 cannot be extended to the setting Bn\mathbb{B}_n, for n≥3n \ge3, see [18]. Recall also that a compact set K⊂S2K \subset S_2 is called a peak set for A∞(B2)A^\infty(\mathbb{B}_2) (the space of all holomorphic functions in B2\mathbb{B}_2 that are extended smoothly on B‾2\overline{\mathbb{B}}_2), if there exists a function f∈A∞(B2)f \in A^\infty(\mathbb{B}_2) such that f(z)=0f(z)=0 for all z∈Kz \in K, and Re⁡f(z)>0\operatorname{Re} f(z)>0, for all z∈B‾2∖Kz \in\overline{\mathbb{B}}_2 \setminus K, see [16 Chapter 10] for more information. The theory established in [18] together with Theorem 2 shows that Z(p)∩S2=⋃iMi\mathcal{Z}(p) \cap S_2 = \bigcup_i M_i, where each Mi‾\overline{M_i} is a peak set for A∞(B2)A^\infty(\mathbb{B}_2). Thus, Theorem 2 together with [16 Proposition 10.7.5-Theorem 10.7.6] yield, additionally, an equivalent characterization of Z(p)∩S2\mathcal{Z}(p) \cap S_2 in terms of peak sets for A∞(B2)A^\infty(\mathbb{B}_2). Topics concerning peak sets, the boundary zero set of stable polynomials, and characterization of cyclicity of multipliers, are also discussed in [17]. The stability condition in the proof of Lemma 1 may be formulated in the switched setting coming up in [9]. Theorem 2 together with established theory in [17] can be applied in the cyclicity problem of polynomials in Dirichlet-type spaces in B2\mathbb{B}_2, see also [10]. Lemma 1 should be also compared with [15 1.5 Lemma].

Remark 7. The involvement of Nash curve selection lemma in proof of Lemma 1 can be omitted; the hypotheses that α∈Mi‾∖Mi\alpha\in\overline{M_i} \setminus M_i is enough. We use it to stress the distinction.

Remark 8. One may consider the same problem for polynomials that are stable with respect to domains whose topological boundary is a an algebraic set or a real analytic set. Recall the semi-analytic curve selection lemma in [3 Lemma 2.28] as it is stated by Z. Denkowska and M.P. Denkowski. A simple example is the ”ellipsoides” Ω:={(z,w)∈C2:σ(z,w)<1}\Omega:= \{(z,w) \in\mathbb{C}^2 : \sigma(z,w) < 1\}, where σ(z,w):=∣z∣2p+∣w∣2p\sigma(z,w) := |z|^{2p} + |w|^{2p}, where p>1p > 1 is integer.

Acknowledgements

The author would like to thank M. Denkowski, J.F. Fernando Galván, L. Kosiński, W. Kucharz for enlightenments. The conceptualisation of the main ideas took place during the Bench Math Session workshop, Kraków, February 2025 together with K. Zarvalis. The author would like also to thank D. Bianchi for help with verifying the arguments.

Statements and Declarations

Funding

Not applicable.

Competing Interests

The author has no competing interest to declare that is relevant to the content of this article.

Data Availability Statement

Not applicable.

AI Declaration

ChatGPT 6 assisted for verifying the arguments concerning the classification of the functions hjh_j in the proof of Lemma 1. The text, the ideas, and the strategies are based on human and the references contained herein. The author take full responsibility for the content of this manuscript. School of Mathematics (Zhuhai), Sun Yat-Sen University, Zhuhai, Guangdong, 519082, P. R. China

Email address: [email protected]

References

  1. [1]J. Bochnak, M. Coste and M.-F. Roy, Real Algebraic Geometry, Ergebnisse der Mathematik und ihrer Grenzgebiete, Springer-Verlag, Berlin, 36, 1998
  2. [2]F. Bruhat, H. Cartan, Sur les composantes irréductibles d’un sous-ensemble analytique reel, C.R. Acad. Sci. Paris, 244: 1123–1126, 1957
  3. [3]Z. Denkowska and M.P. Denkowski, A long and winding road to definable sets, Journal of Singularities, Proc. of Geometry and Topology of Singular Spaces, CIRM, 2012, 13: 57–86, 2015DOI
  4. [4]G. Fischer, Plane algebraic curves, [Translated from German by Leslie Kay], Student Mathematical Library, American Mathematical Society, Providence, RI, 15, 2001DOI
  5. [5]J.E. Fornaess, M. Ovrelid, Finitely generated ideals in A(ω), Annales de l’Institut Fourier, 33(2): 77–85, 1983DOI
  6. [6]F.R. Harvey, R.O. Wells, Zero sets of non-negative strictly plurisubharmonic functions, Math. Ann., 201: 165–170, 1973DOI
  7. [7]G. Knese, Integrability and regularity of rational functions, Proc. Lond. Math. Soc., 111(3), no. 6: 1261–1306, 2015
  8. [8]G. Knese, L. Kosiński, T. Ransford and A. Sola, Cyclic polynomials in anisotropic Dirichlet spaces, J. Anal. Math., 138(1): 23–47, 2019arxiv.org/abs/1512.04871
  9. [9]G. Knese, J.E. Pascoe, A. Sola, Stable polynomials and bounded rational functions in the unit ball, arXiv:2602.21051, [math.CV, math.AG], 2026arxiv.org/abs/2602.21051
  10. [10]Ł. Kosiński and D. Vavitsas, Cyclic polynomials in Dirichlet-type spaces in the unit ball of C², Constr. Approx., 58(2): 343–361, 2023arxiv.org/abs/2212.12013
  11. [11]S.G. Krantz and H.R. Parks, A Primer of Real Analytic Functions, Birkhäuser Boston, Inc., Boston, MA, 2nd ed., 2002
  12. [12]K. Kurdyka, Ensembles semi-algébriques symétriques par arcs, Math. Ann., 282: 445–462, 1988
  13. [13]S. Łojasiewicz, Introduction to Complex Analytic Geometry, [Translated from Polish by Maciej Klimek], Birkhäuser Verlag, Basel, 1991DOI
  14. [14]J. Milnor, Singular points of complex hypersurfaces, Princeton: Princeton Univ. Press, 1968
  15. [15]A. Noell, B. Stensønes, Proper holomorphic maps from weakly pseudoconvex domains, Duke Math. J., 60(2): 363–388, 1990DOI
  16. [16]W. Rudin, Function Theory in the Unit Ball of Cⁿ, Grundlehren der Mathematischen Wissenschaften, Springer-Verlag, New York-Berlin, 241, 1980DOI
  17. [17]P. Torkinejad Ziarati, Cyclicity of Multipliers on the Unit Ball of Cⁿ: A Corona-Based Approach, Anal. Math. Phys., 16(73), 2026
  18. [18]D. Vavitsas, J. Wu, K. Zarvalis, Boundary zeros of stable polynomials in the unit ball, arXiv:2607.05974v2 [math.CV]arxiv.org/abs/2607.05974
  19. [19]D. Vavitsas, K. Zarvalis, Non-cyclicity and polynomials in Dirichlet-type spaces of the unit ball, Bull. London Math. Soc., 56(120): 3905–3919, 2024arxiv.org/abs/2405.03020

Paper details

vlnc:W_znjagmxf
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents