Abstract
We study a question on characterizing polynomials among rational functions of degree \(>1\) on the projective line over an algebraically closed field that is complete with respect to a non-trivial and non-archimedean absolute value, from the viewpoint of dynamics and potential theory on the Berkovich projective line.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
Let K be an algebraically closed field that is complete with respect to a non-trivial and non-archimedean absolute value \(|\cdot |\). The Berkovich projective line \({\textsf {P}}^1={\textsf {P}}^1(K)\) is, as a topological augmentation of the (classical) projective line \({\mathbb {P}}^1={\mathbb {P}}^1(K)=K\cup \{\infty \}\), a compact, locally compact, uniquely arcwise connected, and Hausdorff topological space. The set \({\textsf {H}}^1:={\textsf {P}}^1\setminus {\mathbb {P}}^1\) is called the Berkovich upper half space in \({\textsf {P}}^1\).
Let \(f\in K(z)\) be a rational function of degree \(d>1\). For every \(n\in {\mathbb {N}}\), set \(f^n:=f\circ f^{n-1}\), where \(f^0:={\mathrm {Id}}_{{\mathbb {P}}^1}\). The action of f on \({\mathbb {P}}^1\) uniquely extends to a continuous endomorphism on \({\textsf {P}}^1\), which is still open, surjective, and fiber-discrete, and preserves both \({\mathbb {P}}^1\) and \({\textsf {H}}^1\). Let us define the Berkovich Julia set \({{\mathsf {J}}}(f)\) of f by the set of all points \({\mathcal {S}}\in {\textsf {P}}^1\) such that for any open neighborhood U of \({\mathcal {S}}\) in \({\textsf {P}}^1\),
where the set \(E(f):=\{a\in {\mathbb {P}}^1:\#\bigcup _{n\in {\mathbb {N}}}f^{-n}(a)<\infty \}\) is called the (classical) exceptional set of f and is at most countable subset in \({\mathbb {P}}^1\). The local degree function \(\deg _{\, \cdot }f\) on \({\mathbb {P}}^1\) also canonically extends to \({\textsf {P}}^1\), and this extended local degree function \(\deg _{\,\cdot }(f)\) induces a canonical pullback operator \(f^*\) from the space of all Radon measures on \({\textsf {P}}^1\) to itself (see Sect. 2.2 below). Corresponding to the construction of the unique maximal entropy measure in complex dynamics (studied since Lyubich [20], Freire–Lopes–Mañé [15], Mañé [23]), the f-canonical measure \(\mu _f\) on \({\textsf {P}}^1\) has been constructed as the unique probability Radon measure \(\nu \) on \({\textsf {P}}^1\) such that
so in particular \(\mu _f\) is invariant under f in that \(f_*\mu _f=\mu _f\) on \({\textsf {P}}^1\). The support of \(\mu _f\) coincides with \({{\mathsf {J}}}(f)\) and is the minimal non-empty and closed subset in \({\textsf {P}}^1\) backward invariant under f [14]. The Berkovich Fatou set of f is defined by
and each component of \({\mathsf {F}}(f)\) is called a Berkovich Fatou component of f. We note that \(E(f)\subset {\mathsf {F}}(f)\). A Berkovich Fatou component of f is mapped properly to a Berkovich Fatou component of f under f, and the preimage of a Berkovich Fatou component of f under f is the union of at most d Berkovich Fatou components of f.
Notation 1.1
For every \(z\in {\mathsf {F}}(f)\cap {\mathbb {P}}^1\), let \(D_z=D_z(f)\) be the Berkovich Fatou component of f containing z.
For any \(z\in {\mathsf {F}}(f)\cap {\mathbb {P}}^1\), the compact subset \({\textsf {P}}^1\setminus D_z\) in \({\textsf {P}}^1\) is of logarithmic capacity \(>0\) with pole z, or equivalently, there is the unique equilibrium mass distribution \(\nu _{z,{\textsf {P}}^1\setminus D_z}\) on \({\textsf {P}}^1\setminus D_z\) with pole z, which is in fact supported by \(\partial D_z\subset {{\mathsf {J}}}(f)\) (we will recall some details on the logarithmic potential theory on \({\textsf {P}}^1\) in Sect. 2.4 below). If \(f(\infty )=\infty \in {\mathsf {F}}(f)\), then \(\nu _{\infty ,{\textsf {P}}^1\setminus D_\infty }\) is invariant under f in that
(see Lemma 4.7 below). If moreover \(f\in K[z]\) or equivalently \(f^{-1}(\infty )=\{\infty \}\), then \(\infty \in E(f)\), \(f^{-1}(D_\infty )=D_\infty \), and we can see
(since Brolin [9] in complex dynamics). Let \(\delta _{{\mathcal {S}}}\) be the Dirac measure on \({\textsf {P}}^1\) at \({\mathcal {S}}\in {\textsf {P}}^1\).
Our aim is to study whether polynomials can be characterized among rational functions of degree \(>1\) using potential theory in non-archimedean setting, corresponding to the studies [19, 21, 22, 25, 29, 30] in complex dynamics. Concretely, we study the following question on a characterization of polynomials among rational functions in non-archimedean dynamics.
Question
Let \(f\in K(z)\) be a rational function of degree \(>1\), and suppose that \(f(\infty )=\infty \in {\mathsf {F}}(f)\) (so in particular \(f(D_\infty )=D_\infty \)) and that \({{\mathsf {J}}}(f)\not \subset {\textsf {H}}^1\). Then, are the statements
equivalent?
The corresponding question in complex dynamics has been answered affirmatively (Lopes[21]).
Here are a few comments on this Question. We already mentioned that (i) implies (ii) (without assuming \({{\mathsf {J}}}(f)\not \subset {\textsf {H}}^1\)). It is not difficult to construct such \(f\in K(z)\setminus K[z]\) of degree \(>1\) that \(f(D_\infty )=D_\infty \), that \(f(\infty )\ne \infty \in {\mathsf {F}}(f)\), that \({{\mathsf {J}}}(f)\not \subset {\textsf {H}}^1\), and that \(\mu _f=\nu _{\infty ,{\textsf {P}}^1\setminus D_\infty }\) on \({\textsf {P}}^1\) (e.g., Remark 6.5 below). On the other hand, if \({{\mathsf {J}}}(f)\subset {\textsf {H}}^1\), then for any \(g\in K(z)\) of the same degree as that of f which is close enough to f (in the coefficients topology), both the Berkovich Julia set \({{\mathsf {J}}}(g)\) of g and the action of g on \({{\mathsf {J}}}(g)\) are same as those of f (cf. [14, Sect. 5.3]). Since there is \(f\in K[z]\) of degree \(>1\) satisfying \({{\mathsf {J}}}(f)\subset {\textsf {H}}^1\) (e.g., such f that has a potentially good reduction, see below a characterization of this condition), for any such f and any \(b\in K\), if \(0<|b|\ll 1\), then the small perturbation \(f_b(z):=f(z)/(bz+1)\in K(z)\setminus K[z]\) of \(f=f/1\) in K(z) is of the same degree as that of f and satisfies that \(f_b(\infty )=\infty \in {\mathsf {F}}(f_b)\), that \({{\mathsf {J}}}(f_b)={{\mathsf {J}}}(f)\subset {\textsf {H}}^1\), and that \(\mu _{f_b}=\nu _{\infty ,{\textsf {P}}^1\setminus D_\infty (f_b)}\) on \({\textsf {P}}^1\).
Recall that f has a potentially good reduction if and only if there exists a point \({\mathcal {S}}\in {\textsf {H}}^1\) such that
then \({{\mathsf {J}}}(f)=\{{\mathcal {S}}\}(\subset {\textsf {H}}^1\) so \(\infty \in {\mathsf {F}}(f)\)) and \(\mu _f=\nu _{\infty ,{{\textsf {P}}}^1\setminus D_\infty }=\delta _{{\mathcal {S}}}\) on \({{\textsf {P}}}^1\) (see also Remark 3.2 below). We say f has no potentially good reductions if f does not have a potentially good reduction.
We already mentioned that the total invariance \(f^{-1}(D_\infty )=D_\infty \) of \(D_\infty \) under f is a necessary condition for \(f\in K[z]\). Our first result is the following more general statement, under no potentially good reductions:
Theorem 1
Let K be an algebraically closed field that is complete with respect to a non-trivial and non-archimedean absolute value. Let \(f\in K(z)\) be a rational function of degree \(>1\). If \(\infty \in {{\mathsf {F}}}(f)\), \(f(D_\infty )=D_\infty \), \(\mu _f=\nu _{\infty ,{{\textsf {P}}}^{1}\setminus D_\infty }\) on \({{\textsf {P}}}^1\), and f has no potentially good reductions, then
Our second result is that even if we assume in addition \({{\mathsf {J}}}(f)\subset {\mathbb {P}}^1\), the latter statement (ii) does not necessarily imply the former (i) in Question.
Pick a prime number p. The p-adic norm \(|\cdot |_p\) on \({\mathbb {Q}}\) is normalized so that for any \(m,\ell \in {\mathbb {Z}}\setminus \{0\}\) not divisible by p and any \(r\in {\mathbb {Z}}\), \(\bigl |\frac{m}{\ell }p^r\bigr |_p=p^{-r}\). The completion \({\mathbb {Q}}_p\) of \(({\mathbb {Q}},|\cdot |_p)\) is still a field, and the extended norm \(|\cdot |_p\) on \({\mathbb {Q}}_p\) extends to an algebraic closure \(\overline{{\mathbb {Q}}_p}\) of \({\mathbb {Q}}_p\) as a norm. The completion \({\mathbb {C}}_p\) of \((\overline{{\mathbb {Q}}_p},|\cdot |_p)\) is still an algebraically closed field, and the extended norm \(|\cdot |_p\) on \({\mathbb {C}}_p\) is a non-trivial and non-archimedean absolute value on \({\mathbb {C}}_p\). The completion \({\mathbb {Z}}_p\) of \(({\mathbb {Z}},|\cdot |_p)\) is a complete discrete valued local ring and has the unique maximal ideal \(p{\mathbb {Z}}_p\), and coincides with the ring of \({\mathbb {Q}}_p\)-integers \(\{z\in {\mathbb {Q}}_p:|z|_p\le 1\}\). In particular, the residual field of \({\mathbb {Q}}_p\) is \({\mathbb {F}}_p\).
The following counterexample of the implication (ii)\(\Rightarrow \)(i) in Question is suggested to the authors by Juan Rivera-Letelier:
Theorem 2
Pick a prime number p, and set
If \(c\ne 0\) and (a, b, c, d) is close enough to (1, 0, 0, 1) in \(({\mathbb {Z}}_p)^4\), then there is an attracting fixed point \(z_A\) of \(f\circ A\) in \({\mathbb {C}}_p\setminus {\mathbb {Z}}_p\) (so \(z_A\in {{\mathsf {F}}}(f\circ A))\) such that
Then setting \(m_A(z):=1/(z-z_A)\in \mathrm {PGL}(2,{\mathbb {C}}_p)\), the rational function \(g_A(z):=m_A\circ (f\circ A)\circ m_A^{-1}\in {\mathbb {C}}_p(z)\) is of degree p and satisfies \(g_A\not \in {\mathbb {C}}_p[z]\), \(g_A(\infty )=\infty \in {{\mathsf {F}}}(g_A)\), \({{{\mathsf {J}}}}(g_A)\subset {\mathbb {P}}^{1}({\mathbb {C}}_p)\), and
1.1 Organization of this Article
In Sects. 2 and 3, we prepare background material from potential theory and dynamics, respectively. In Sect. 4, we make preparatory computations from potential theory and give a proof of the invariance of \(\nu _{\infty ,{{\textsf {P}}}^{1}\setminus D_\infty }\) under f when \(f(\infty )=\infty \in {{\mathsf {F}}}(f)\). In Sects. 5 and 6, we show Theorems 1 and 2, respectively.
2 Background from Potential Theory on \({\textsf {P}}^1\)
Let K be an algebraically closed field that is complete with respect to a non-trivial and non-archimedean absolute value \(|\cdot |\); in general, a norm \(|\cdot |\) on a field k is non-trivial if \(|k|\not \subset \{0,1\}\), and is non-archimedean if \(|\cdot |\) satisfies the strong triangle inequality
For the foundation of potential theory on \({{\textsf {P}}}^1={{\textsf {P}}}^1(K)\), see [5, Sects. 5, 8] , [12, Sect. 7], [13, Sect. 3], [33], and the survey [18, Sects. 1–4], and the book [6, Sect. 13]. In what follows, we adopt a presentation from [28, Sects. 2, 3].
Notation 2.1
Let
be the canonical projection such that
following the convention on coordinate of \({\mathbb {P}}^1\) from the book [16].
On \(K^2\), let \(\Vert (p_0,p_1)\Vert \) be the maximum norm \(\max \{|p_0|,|p_1|\}\). With the wedge product \((p_0,p_1)\wedge (q_0,q_1):=p_0q_1-p_1q_0\) on \(K^2\), the normalized chordal metric [z, w] on \({\mathbb {P}}^1\) is the function
on \({\mathbb {P}}^1\times {\mathbb {P}}^1\), where \(p\in \pi ^{-1}(z),q\in \pi ^{-1}(w)\).
2.1 Berkovich Projective Line \({\textsf {P}}^1\)
A (K-closed) disk in K is a subset in K written as \(\{z\in K:|z-a|\le r\}\) for some \(a\in K\) and some \(r\ge 0\). By the strong triangle inequality, two decreasing infinite sequences of disks in K either infinitely nest or are eventually disjoint. This alternative induces the cofinal equivalence relation among decreasing (or more precisely, nesting and non-increasing) infinite sequences of disks in K, and the set of all cofinal equivalence classes \({\mathcal {S}}\) of decreasing infinite sequences \((B_n)\) of disks in K together with \(\infty \in {\mathbb {P}}^1\) is, as a set, nothing but \({{\textsf {P}}}^1\) ([7, p. 17]); if \(B_{{\mathcal {S}}}:=\bigcap _n B_n\ne \emptyset \), then \(B_{{\mathcal {S}}}\) is itself a disk in K, and we also say \({\mathcal {S}}\) is represented by \(B_{{\mathcal {S}}}\). For example, the canonical (or Gauss) point \({\mathcal {S}}_{{\text {can}}}\) in \({{\textsf {P}}}^1\) is represented by the the ring of K-integers
and each \(z\in K\) is represented by the disk \(\{z\}\) in K. The above alternative between two (decreasing infinite sequences of) disks in K also induces a canonical ordering \(\succeq \) on \({{\textsf {P}}}^1\) so that \(\infty \) is the unique maximal element in \(({{\textsf {P}}}^1,\succeq )\) and that for every \({\mathcal {S}},{\mathcal {S}}'\in {{\textsf {P}}}^1\setminus \{\infty \}\) satisfying \(B_{{\mathcal {S}}},B_{{\mathcal {S}}'}\ne \emptyset \), \({\mathcal {S}}\succeq {\mathcal {S}}'\) iff \(B_{{\mathcal {S}}}\supset B_{{\mathcal {S}}'}\) (the description of \(\succeq \) is a little complicated unless \(B_{{\mathcal {S}}},B_{{\mathcal {S}}'}\ne \emptyset \)), and equips \({{\textsf {P}}}^1\) with a (profinite) tree structure. The topology of \({{\textsf {P}}}^1\) coincides with the weak (or observer) topology on \({{\textsf {P}}}^1\) as a (profinite) tree, so that \({{\textsf {P}}}^1\) is compact and uniquely arcwise-connected, and contains both \({\mathbb {P}}^1\) and \({\textsf {H}}^1\) as dense subsets. For the details on the tree structure on \({{\textsf {P}}}^1\), see e.g. [18, Sect. 2].
2.2 Action of Rational Functions on \({\textsf {P}}^1\)
Let \(h\in K(z)\) be a rational function. The action of h on \({\mathbb {P}}^1\) uniquely extends to a continuous endomorphism on \({{\textsf {P}}}^1\). Suppose in addition that \(\deg h>0\). Then the extended action of h on \({{\textsf {P}}}^1\) is surjective and open, has discrete (so finite) fibers, and preserves both \({\mathbb {P}}^1\) and \({\textsf {H}}^1\), and the local degree function \(z\mapsto \deg _zh\) on \({\mathbb {P}}^1\) also canonically extends to \({{\textsf {P}}}^1\) so that for every \({\mathcal {S}}\in {{\textsf {P}}}^1\),
The action of h on \({{\textsf {P}}}^1\) induces the push-forward operator \(h_*\) on the space of all continuous functions on \({{\textsf {P}}}^1\) to itself and, by duality, also the pullback operator \(h^*\) on the space of all Radon measures on \({{\textsf {P}}}^1\) to itself; for every continuous test function \(\phi \) on \({{\textsf {P}}}^1\), \((h_*\phi )(\cdot )=\sum _{{\mathcal {S}}'\in h^{-1}(\cdot )}(\deg _{{\mathcal {S}}'}h)\cdot \phi ({\mathcal {S}}')\) on \({{\textsf {P}}}^1\), and for every \({\mathcal {S}}\in {{\textsf {P}}}^1\), \(h^*\delta _{{\mathcal {S}}} =\sum _{{\mathcal {S}}'\in h^{-1}({\mathcal {S}})}(\deg _{{\mathcal {S}}'}h)\cdot \delta _{{\mathcal {S}}'}\) on \({{\textsf {P}}}^1\). For more details, see [5, Sect. 9], [14, Sect. 2.2].
2.3 Kernel Functions and the Laplacian on \({\textsf {P}}^1\)
The generalized Hsia kernel \([{\mathcal {S}},{\mathcal {S}}']_{{\text {can}}}\) on \({{\textsf {P}}}^1\) with respect to \({\mathcal {S}}_{{\text {can}}}\) is a unique upper semicontinuous and separately continuous extension of the chordal distance function \({\mathbb {P}}^1\times {\mathbb {P}}^1\ni (z,z')\mapsto [z,z']\) to \({{\textsf {P}}}^1\times {{\textsf {P}}}^1\).
More generally, for every \(z_0\in {\mathbb {P}}^1\), the generalized Hsia kernel
on \({{\textsf {P}}}^1\) with respect to \(z_0\) is a unique upper semicontinuous and separately continuous extension of the function \(({\mathbb {P}}^1\setminus \{z_0\})\times ({\mathbb {P}}^1\setminus \{z_0\})\ni (z,z')\mapsto [z,z']/([z,z_0]\cdot [z',z_0])\) as a function \({{\textsf {P}}}^1\times {{\textsf {P}}}^1\rightarrow [0,+\infty ]\). In particular, the function
on \({{\textsf {P}}}^1\times {{\textsf {P}}}^1\) extends the distance function \(K\times K\ni (z,z')\mapsto |z-z'|\) to \(({{\textsf {P}}}^1\setminus \{\infty \})\times ({{\textsf {P}}}^1\setminus \{\infty \})\), jointly upper semicontinuously and separately continuously, and the function
extends the norm function \(K\ni z\mapsto |z|\) to \({{\textsf {P}}}^1\setminus \{\infty \}\) continuously (see [13, Sect. 3.4], [5, Sect. 4.4]).
Let \(\Omega _{{\text {can}}}\) be the Dirac measure \(\delta _{{\mathcal {S}}_{{\text {can}}}}\) on \({{\textsf {P}}}^1\) at \({\mathcal {S}}_{{\text {can}}}\). The Laplacian \(\Delta \) on \({{\textsf {P}}}^1\) is normalized so that for each \({\mathcal {S}}'\in {{\textsf {P}}}^1\),
on \({{\textsf {P}}}^1\), and then, for every \(z_0\in {\mathbb {P}}^1\) and every \({\mathcal {S}}'\in {{\textsf {P}}}^1\setminus \{z_0\}\), \(\Delta \log [\cdot ,{\mathcal {S}}']_{z_0}=\delta _{{\mathcal {S}}'}-\delta _{z_0}\) on \({{\textsf {P}}}^1\). For the details on the construction and properties of \(\Delta \), see [5, Sect. 5], [12, Sect. 7.7], [14, Sect. 2.4], [33, Sect. 3]; in [5, 33], the opposite sign convention for \(\Delta \) is adopted.
2.4 Logarithmic Potential Theory on \({\textsf {P}}^1\)
For every \(z\in {\mathbb {P}}^1\) and every positive Radon measure \(\nu \) on \({{\textsf {P}}}^1\) supported by \({{\textsf {P}}}^1\setminus \{z\}\), the logarithmic potential of \(\nu \) on \({{\textsf {P}}}^1\) with pole z is the function
and the logarithmic energy of \(\nu \) with pole z is defined by
Then \(p_{z,\nu }:{{\textsf {P}}}^1\rightarrow [-\infty ,+\infty ]\) is upper semicontinuous, and in fact is strongly upper semicontinuous in that for every \({\mathcal {S}}\in {{\textsf {P}}}^1\),
([5, Proposition 6.12]).
For every non-empty subset C in \({{\textsf {P}}}^1\) and every \(z\in {\mathbb {P}}^1\setminus C\), we say C is of logarithmic capacity \(>0\) with pole z if
where \(\nu \) ranges over all probability Radon measures on \({{\textsf {P}}}^1\) supported by C; otherwise, we say C is of logarithmic capacity 0 with pole z. For every non-empty compact subset C in \({{\textsf {P}}}^1\) of logarithmic capacity \(>0\) with pole \(z\in {\mathbb {P}}^1\setminus C\), there is a unique probability Radon measure \(\nu \) on \({{\textsf {P}}}^1\), which is called the equilibrium mass distribution on C with pole z and is denoted by \(\nu _{z,C}\), such that \({\text {supp}}\nu \subset C\) and that \(I_{z,\nu }=V_z(C)\), and then (i) \(\nu _{z,C}(E)=0\) for any subset E in C of logarithmic capacity 0 with pole z, (ii) letting \(D_z\) be the component of \({{\textsf {P}}}^1\setminus C\) containing z, we have
where E is a possibly empty \(F_{\sigma }\)-subset in \(\partial D_z\) of logarithmic capacity 0 with pole z, (iii) if in addition \(p_{z,\nu _{z,C}}\) is continuous on \({{\textsf {P}}}^1\setminus {\{z\}}\), then
and (iv) for any probability Radon measure \(\nu '\) supported by C, we have
(see [5, Sects. 6.2, 6.3]).
We list a few observations:
Observation 2.2
For every \(a\in K\setminus \{0\}\) and every \(b\in K\), setting \(\ell (z):=az+b\in \mathrm {PGL}(2,K)\), we have \(\log |\ell ({\mathcal {S}})-\ell ({\mathcal {S}}')|_\infty =\log |{\mathcal {S}}-{\mathcal {S}}'|_\infty +\log |a|\) on \(K\times K\), and in turn on \({{\textsf {P}}}^1\times {{\textsf {P}}}^1\). In particular, for every non-empty compact subset C in \({{\textsf {P}}}^1\setminus \{\infty \}\) of logarithmic capacity \(>0\) with pole \(\infty \), we have \(I_{\infty ,\nu _{\infty ,\ell (C)}}=I_{\infty ,\nu _{\infty ,C}}+\log |a|\) and \(\ell _*(\nu _{\infty ,C})=\nu _{\infty ,\ell (C)}\) on \({{\textsf {P}}}^1\).
Observation 2.3
Since the involution \(\iota (z)=1/z\in \mathrm {PGL}(2,{\mathcal {O}}_K)\) acts on \(({\mathbb {P}}^1,[z,w])\) isometrically, for any \(z_0\in {\mathbb {P}}^1\), we have \([\iota ({\mathcal {S}}),\iota ({\mathcal {S}}')]_{\iota (z_0)}=[{\mathcal {S}},{\mathcal {S}}']_{z_0}\) on \({\mathbb {P}}^1\times {\mathbb {P}}^1\), and in turn on \({{\textsf {P}}}^1\times {{\textsf {P}}}^1\). Hence for any non-empty compact subset C in \({{\textsf {P}}}^1\) and any \(z\in {{\textsf {P}}}^1\setminus C\), if C is of logarithmic capacity \(>0\) with pole z, then \(V_z(C)=V_{\iota (z)}(\iota (C))\) and \(\iota _*(\nu _{z,C})=\nu _{\iota (z),\iota (C)}\) on \({{\textsf {P}}}^1\).
Observation 2.4
For every \(z\in {\mathbb {P}}^1\), the strong triangle inequality \([{\mathcal {S}},{\mathcal {S}}'']_{z}\le \max \{[{\mathcal {S}},{\mathcal {S}}']_{z},[{\mathcal {S}}',{\mathcal {S}}'']_{z}\}\) for \({\mathcal {S}},{\mathcal {S}}',{\mathcal {S}}''\in {{\textsf {P}}}^1\) still holds (see [5, Proposition 4.10]). Hence for every non-empty compact subset C in \({{\textsf {P}}}^1\setminus \{\infty \}\) and every \(z\in {\mathbb {P}}^1\setminus C\) so close to \(\infty \) that \([z,\infty ]<\inf _{{\mathcal {S}}\in C}[{\mathcal {S}},z]_{{\text {can}}}\), we have \([\cdot ,\infty ]_{{\text {can}}}=[\cdot ,z]_{{\text {can}}}\) on C, which yields \([{\mathcal {S}},{\mathcal {S}}']_\infty =[{\mathcal {S}},{\mathcal {S}}']_{z}\) on \(C\times C\), so if in addition C is of logarithmic capacity \(>0\) with pole \(\infty \), then \(V_{\infty }(C)=V_z(C)\) and \(\nu _{\infty ,C}=\nu _{z,C}\) on \({{\textsf {P}}}^1\).
2.5 Potential Theory with a Continuous Weight on \({\textsf {P}}^1\)
A continuous weight g on \({{\textsf {P}}}^1\) is a continuous function on \({{\textsf {P}}}^1\) such that
is a probability Radon measure on \({{\textsf {P}}}^1\). Then \(\mu ^g\) has no atoms on \({\mathbb {P}}^1\), or more strongly, \(\mu ^g(E)=0\) for any subset E in \({{\textsf {P}}}^1\) of logarithmic capacity 0 with some (indeed any) point in \({{\textsf {P}}}^1\setminus E\).
For a continuous weight g on \({{\textsf {P}}}^1\), the g-potential kernel on \({{\textsf {P}}}^1\) (the negative of an Arakelov Green kernel function on \({{\textsf {P}}}^1\) relative to \(\mu ^g\) [5, Sect. 8.10] ) is an upper semicontinuous function
For every Radon measure \(\nu \) on \({{\textsf {P}}}^1\), the g-potential of \(\nu \) on \({{\textsf {P}}}^1\) is the function
and the g-energy of \(\nu \) is defined by
The g-equilibrium energy \(V_g\) of (the whole) \({{\textsf {P}}}^1\) is the supremum of the g-energy functional \(\nu \mapsto I_{g,\nu }\), where \(\nu \) ranges over all probability Radon measures on \({{\textsf {P}}}^1\). Then \(V_g\in \mathbb {R}\) since \(I_{g,\Omega _{{\text {can}}}}>-\infty \). As in the logarithmic potential theory presented in the previous subsection, there is a unique probability Radon measure \(\nu ^g\) on \({{\textsf {P}}}^1\), which is called the g-equilibrium mass distribution on \({{\textsf {P}}}^1\), such that \(I_{g,\nu ^g}=V_g\). In fact
(see [5, Theorem 8.67, Proposition 8.70]).
A continuous weight g on \({{\textsf {P}}}^1\) is a normalized weight on \({{\textsf {P}}}^1\) if \(V_g=0\). For a continuous weight g on \({{\textsf {P}}}^1\), \(\overline{g}:=g+V_g/2\) is the unique normalized weight on \({{\textsf {P}}}^1\) satisfying \(\mu ^{\overline{g}}=\mu ^g\).
3 Background from Dynamics on \({\textsf {P}}^1\)
For a potential-theoretic study of dynamics of a rational function of degree \(>1\) on \({{\textsf {P}}}^1={{\textsf {P}}}^1(K)\), see [5, Sect. 10], [14, Sect. 3], [18, Sect. 5], and [6, Sect. 13]. In the following, we adopt a presentation from [28, Sect. 8.1].
3.1 Canonical Measure and the Dynamical Green Function of f on \({\textsf {P}}^1\)
Let \(f\in K(z)\) be a rational function of degree \(d>1\). We call \(F\in (K[p_0,p_1]_d)^2\) a lift of f if
on \(K^2\setminus \{(0,0)\}\), where for each \(j\in {\mathbb {N}}\cup \{0\}\), \(K[p_0,p_1]_j\) is the set of all homogeneous polynomials in \(K[p_0,p_1]\) of degree j, as usual. A lift \(F=(F_0,F_1)\) of f is unique up to multiplication in \(K\setminus \{0\}\). Setting \(d_0:=\deg F_0(1,z)\) and \(d_1:=\deg F_1(1,z)\) and letting \(c^F_0,c^F_1\in K\setminus \{0\}\) be the coefficients of the maximal degree terms of \(F_0(1,z),F_1(1,z)\in K[z]\), respectively, the homogeneous resultant
of F does not vanish, where \(R(P,Q)\in K\) is the usual resultant of \((P,Q)\in (K[z])^2\) (for the details on \({\text {Res}}F\), see e.g. [32, Sect. 2.4]).
Let F be a lift of f, and for every \(n\in {\mathbb {N}}\cup \{0\}\), set \(F^n=F\circ F^{n-1}\) where \(F^0:={\mathrm {Id}}_{K^2}\). Then for every \(n\in {\mathbb {N}}\), \(F^n\) is a lift of \(f^n\), and the function
on \(K^2\setminus \{(0,0)\}\) descends to \({\mathbb {P}}^1\) and in turn extends continuously to \({{\textsf {P}}}^1\), satisfying the equality \(\Delta T_{F^n}=(f^n)^*\Omega _{{\text {can}}}-d^n\cdot \Omega _{{\text {can}}}\) on \({{\textsf {P}}}^1\) (see, e.g., [26, Definition 2.8]). The dynamical Green function of F on \({{\textsf {P}}}^1\) is the uniform limit \(g_F:=\lim _{n\rightarrow \infty }T_{F^n}/d^n\) on \({{\textsf {P}}}^1\), which is a continuous weight on \({{\textsf {P}}}^1\). The energy formula
is due to DeMarco [11] for archimedean K by a dynamical argument, and due to Baker–Rumely [4] when f is defined over a number field; see Baker [2, Appendix A] or the present authors [29, Appendix] for a simple and potential-theoretic proof of this remarkable formula, for general K. The f-canonical measure is the probability Radon measure
The measure \(\mu _f\) is independent of the choice of the lift F of f, has no atoms in \({\mathbb {P}}^1\), and satisfies the f-balanced property \(f^*\mu _f=d\cdot \mu _f\) (so in particular \(f_*\mu _f=\mu _f\)) on \({{\textsf {P}}}^1\). For more details, see [5, Sect. 10], [10, Sect. 2], [14, Sect. 3.1].
The dynamical Green function \(g_f\) of f on \({{\textsf {P}}}^1\) is the unique normalized weight on \({{\textsf {P}}}^1\) such that \(\mu ^{g_f}=\mu _f\). By the above energy formula on \(V_{g_F}\) and
there is a lift F of f normalized so that \(V_{g_F}=0\) or equivalently that \(g_F=g_f\) on \({{\textsf {P}}}^1\), and such a normalized lift F of f is unique up to multiplication in \(\{z\in K:|z|=1\}\). By \(g_f=g_F=\lim _{n\rightarrow \infty }T_{F^n}/d^n\) on \({{\textsf {P}}}^1\) for a normalized lift F of f, for every \(n\in {\mathbb {N}}\), we have \(g_{F^n}=g_{f^n}=g_f\) on \({{\textsf {P}}}^1\) and \(\mu _{f^n}=\mu _f\) on \({{\textsf {P}}}^1\). We note that \(g_f\circ f=d\cdot \lim _{n\rightarrow \infty }T_{F^{n+1}}/d^{n+1}-T_F=d\cdot g_f-T_F\) on \({\mathbb {P}}^1\), that is,
on \({\mathbb {P}}^1\), and in turn on \({{\textsf {P}}}^1\) by the density of \({\mathbb {P}}^1\) in \({{\textsf {P}}}^1\) and the continuity of both sides on \({{\textsf {P}}}^1\) (cf. [27, Proof of Lemma 2.4]).
3.2 Fundamental Properties of \(\mu _f\)
Recall the definition of \({{\mathsf {J}}}(f)\) in Sect. 1. The characterization of \(\mu _f\) as the unique probability Radon measure \(\nu \) on \({{\textsf {P}}}^1\) such that \(\nu (E(f))=0\) and that \(f^*\nu =d\cdot \nu \) on \({{\textsf {P}}}^1\) is a consequence of the following equidistribution theorem: for every probability Radon measure \(\mu \) on \({{\textsf {P}}}^1\), if \(\mu (E(f))=0\), then
This foundational result is due to Favre and Rivera-Letelier [14] (for a purely potential-theoretic proof, see also Jonsson [18]) and is a non-archimedean counterpart to Brolin [9], Lyubich [20], Freire et al. [15].
Remark 3.1
The classical Julia set \({{\mathsf {J}}}(f)\cap {\mathbb {P}}^1\) of f coincides with the set of all points in \({\mathbb {P}}^1\) at each of which the family \(\bigl (f^n:({\mathbb {P}}^1,[z,w])\rightarrow ({\mathbb {P}}^1,[z,w])\bigr )_{n\in {\mathbb {N}}}\) is not locally equicontinuous (see, e.g., [5, Theorem 10.67] ).
The equality \({\text {supp}}\mu _f={{\mathsf {J}}}(f)\) holds; the inclusion \({{\mathsf {J}}}(f)\subset {\text {supp}}\mu _f\) follows from the definition of \({{\mathsf {J}}}(f)\), the balanced property \(f^*\mu _f=d\cdot \mu _f\) on \({{\textsf {P}}}^1\), and \({\text {supp}}\mu _f\not \subset E(f)\) (or more precisely, recalling that E(f) is an at most countable subset in \({\mathbb {P}}^1\) and that \(\mu _f\) has no atoms in \({\mathbb {P}}^1\)). The opposite inclusion \({\text {supp}}\mu _f\subset {{\mathsf {J}}}(f)\) follows from the definition of \({{\mathsf {J}}}(f)\) and the above equidistribution theorem.
Remark 3.2
(see, e.g., [5, Corollary 10.33]) If \(\mu _f\) has an atom in \({{\textsf {P}}}^1\), then f has a potentially good reduction, so in particular \({{\mathsf {J}}}(f)\) is a singleton in \({\textsf {H}}^1\).
For every \(n\in {\mathbb {N}}\), by \({\text {supp}}\mu _f={{\mathsf {J}}}(f)\) and \(\mu _{f^n}=\mu _f\) on \({{\textsf {P}}}^1\), we also have \({{\mathsf {J}}}(f^n)={{\mathsf {J}}}(f)\). For every \(m\in \mathrm {PGL}(2,K)\), we have \(m_*\mu _f=\mu _{m\circ f\circ m^{-1}}\) on \({{\textsf {P}}}^1\), \(m({{\mathsf {J}}}(f))={{\mathsf {J}}}(m\circ f\circ m^{-1})\), and \(m({{\mathsf {F}}}(f))={{\mathsf {F}}}(m\circ f\circ m^{-1})\).
3.3 Root Divisors on \({\mathbb {P}}^1\) and the Proximity Functions on \({\textsf {P}}^1\)
For any distinct \(h_1,h_2\in K(z)\), let \([h_1=h_2]\) be the effective (K-)divisor on \({\mathbb {P}}^1\) defined by all solutions to the equation \(h_1=h_2\) in \({\mathbb {P}}^1\) taking into account their multiplicities, which is also regarded as the Radon measure
on \({{\textsf {P}}}^1\). The function \({\mathbb {P}}^1\ni z\mapsto [h_1(z),h_2(z)]\) between \(h_1\) and \(h_2\) uniquely extends to a continuous function \({\mathcal {S}}\mapsto [h_1,h_2]_{{\text {can}}}({\mathcal {S}})\) on \({{\textsf {P}}}^1\) (see, e.g., [26, Proposition 2.9]), so that for every continuous weight g on \({{\textsf {P}}}^1\), (the \(\exp \) of) the function
is a unique continuous extension of (the \(\exp \) of) the function \({\mathbb {P}}^1\ni z\mapsto \Phi _g(h_1(z),h_2(z))\).
4 Potential-Theoretic Computations
Let \(f\in K(z)\) be a rational function of degree \(d>1\).
Lemma 4.1
(Riesz’s decomposition for the pullback of an atom) For every \({\mathcal {S}}\in {{\textsf {P}}}^1\),
Proof
Fix a lift F of f normalized so that \(g_F=g_f\) on \({{\textsf {P}}}^1\). Fix \(w\in {\mathbb {P}}^1\) and \(W\in \pi ^{-1}(w)\). Choose a sequence \((q_j)_{j=1}^{d}\) in \(K^2\setminus \{(0,0)\}\) such that \(F(p_0,p_1)\wedge W\in K[p_0,p_1]_d\) factors as \(F(p_0,p_1)\wedge W=\prod _{j=1}^{d}((p_0,p_1)\wedge q_j)\) in \(K[p_0,p_1]\). This together with (3.1) and the definition of \(T_F\) implies
so \(\Phi _{g_f}(f(\cdot ),w)-U_{g_f,f^*\delta _w}(\cdot )\equiv C\) on \({\mathbb {P}}^1\), and in turn on \({{\textsf {P}}}^1\) by the density of \({\mathbb {P}}^1\) in \({{\textsf {P}}}^1\) and the continuity of (the \(\exp \) of) both sides on \({{\textsf {P}}}^1\). Integrating both sides against \(\mu _f\) over \({{\textsf {P}}}^1\), since \(\int _{{{\textsf {P}}}^1}U_{g_f,f^*\delta _w}\mu _f =\int _{{{\textsf {P}}}^1}U_{g_f,\mu _f}(f^*\delta _w)=0\) (by \(U_{g_f,\mu _f}\equiv 0\)) and \(f_*\mu _f=\mu _f\), we have
This completes the proof of (4.1) in the case \({\mathcal {S}}=w\in {\mathbb {P}}^1\).
Fix \({\mathcal {S}}_0\in {\textsf {H}}^1\). By the density of \({\mathbb {P}}^1\) in \({{\textsf {P}}}^1\), we can choose a sequence \((w_n)\) in \({\mathbb {P}}^1\) tending to \({\mathcal {S}}_0\) as \(n\rightarrow \infty \). Then \(\lim _{n\rightarrow \infty }f^*\delta _{w_n}=f^*\delta _{{\mathcal {S}}_0}\) weakly on \({{\textsf {P}}}^1\) and, for every \(n\in {\mathbb {N}}\), applying (4.1) to \({\mathcal {S}}=w_n\in {\mathbb {P}}^1\), we have \(\Phi _{g_f}(f(\cdot ),w_n)=U_{g_f,f^*\delta _{w_n}}(\cdot )\) on \({{\textsf {P}}}^1\). Hence, for each \({\mathcal {S}}'\in {\textsf {H}}^1\), by the continuity of both \(\Phi _{g_f}(f({\mathcal {S}}'),\cdot )\) and \(\Phi _{g_f}({\mathcal {S}}',\cdot )\) on \({{\textsf {P}}}^1\), we have
This completes the proof of (4.1) by the density of \({\textsf {H}}^1\) in \({{\textsf {P}}}^1\) and the continuity of (the \(\exp \) of) both \(\Phi _{g_f}(f(\cdot ),{\mathcal {S}}_0)\) and \(U_{g_f,f^*\delta _{{\mathcal {S}}_0}}(\cdot )\) on \({{\textsf {P}}}^1\). \(\square \)
The following computation is an application of Lemma 4.1. We include a proof of it although it will not be used in this article.
Lemma 4.2
(Riesz’s decomposition for the fixed points divisor on \({\mathbb {P}}^1)\)
Proof
Fix a lift F of f normalized so that \(g_F=g_f\) on \({{\textsf {P}}}^1\). Choose a sequence \((q_j)_{j=1}^{d+1}\) in \(K^2\setminus \{(0,0)\}\) so that \((F\wedge {\mathrm {Id}}_{{\mathbb {P}}^1})(p_0,p_1)\in K[p_0,p_1]_{d+1}\) factors as \((F\wedge {\mathrm {Id}}_{{\mathbb {P}}^1})(p_0,p_1)=\prod _{j=1}^{d+1}((p_0,p_1)\wedge q_j)\) in \(K[p_0,p_1]\), which with (3.1) implies
on \({\mathbb {P}}^1\), and in turn on \({{\textsf {P}}}^1\) by the density of \({\mathbb {P}}^1\) in \({{\textsf {P}}}^1\) and the continuity of (the \(\exp \) of) both sides on \({{\textsf {P}}}^1\). Integrating both sides against \(\mu _f\) over \({{\textsf {P}}}^1\), since \(\int _{{{\textsf {P}}}^1}U_{g_f,[f={\mathrm {Id}}_{{\mathbb {P}}^1}]}\mu _f=\int _{{{\textsf {P}}}^1}U_{g_f,\mu _f}[f={\mathrm {Id}}_{{\mathbb {P}}^1}]=0\) (by \(U_{g_f,\mu _f}\equiv 0\)), we have \(C=\int _{{{\textsf {P}}}^1}\Phi (f,{\mathrm {Id}}_{{\mathbb {P}}^1})_{g_f}\mu _f\), so that we first have
Fix \(z_0\in {\mathbb {P}}^1\setminus ({\text {supp}}[f={\mathrm {Id}}_{{\mathbb {P}}^1}])\). Using the above equality twice, by \(f_*[f={\mathrm {Id}}_{{\mathbb {P}}^1}]=[f={\mathrm {Id}}_{{\mathbb {P}}^1}]\) on \({{\textsf {P}}}^1\) and (4.1), we have
and moreover, \(\int _{{{\textsf {P}}}^1}\Phi (f,{\mathrm {Id}}_{{\mathbb {P}}^1})_{g_f}(f^*\delta _{z_0}) =U_{g_f,f^*\delta _{z_0}}(z_0)=\Phi _{g_f}(f(z_0),z_0)\) by (4.1). Hence \((d-1)\int _{{{\textsf {P}}}^1}\Phi (f,{\mathrm {Id}}_{{\mathbb {P}}^1})_{g_f}\mu _f=0\), and in turn since \(d>1\),
This completes the proof. \(\square \)
From now on, we focus on the case where \(\infty \in {{\mathsf {F}}}(f)\). We adopt the following convention when no confusion would be caused:
Convention
For every probability Radon measure \(\nu \) supported by \({{\textsf {P}}}^1\setminus \{\infty \}\), we denote \(p_{\infty ,\nu }\) and \(I_{\infty ,\nu }\) by \(p_\nu \) and \(I_\nu \), respectively, for simplicity.
Since \({\text {supp}}\mu _f={{\mathsf {J}}}(f)\subset {{\textsf {P}}}^1\setminus D_\infty \), the equality (4.5) below implies that \({{\textsf {P}}}^1\setminus D_\infty \) is of logarithmic capacity \(>0\) with pole \(\infty \).
Lemma 4.3
Suppose that \(\infty \in {{\mathsf {F}}}(f)\). Then
Proof
Suppose \(\infty \in {{\mathsf {F}}}(f)\). Then we have \({\text {supp}}\mu _f={{\mathsf {J}}}(f)\subset {{\textsf {P}}}^1\setminus D_\infty \) and
so that \(I_{\mu _f}=2\cdot \int _{{{\textsf {P}}}^1}(g_f-\log [\cdot ,\infty ]_{{\text {can}}})\mu _f\), which with
yields (4.4). By (4.4) and \(\log [z,\infty ]=\log [z,0]-\log |z|\) on \({\mathbb {P}}^1\setminus \{\infty \}\), we have
so that (4.5) holds. By (4.4) and (4.5), we have \(\Phi _{g_f}(\cdot ,\infty )=\log [\cdot ,\infty ]_{{\text {can}}}-g_f-g_f(\infty ) =(-p_{\mu _f}+I_{\mu _f}/2)+I_{\mu _f}/2=-p_{\mu _f}+I_{\mu _f}\) on \({{\textsf {P}}}^1\), so (4.6) also holds. \(\square \)
Let \(F=(F_0,F_1)\in (K[p_0,p_1]_d)^2\) be a normalized lift of f, and \(c_0^F,c_1^F\in K\setminus \{0\}\) be the coefficients of the maximal degree terms of \(F_0(1,z),F_1(1,z)\in K[z]\), respectively. No matter whether \(\infty \in {{\mathsf {F}}}(f)\), by the equality \([z,\infty ]=1/\Vert (1,z)\Vert \) on \({\mathbb {P}}^1\) and the definition of \(T_F\), we have
on \({\mathbb {P}}^1\setminus (\{\infty \}\cup f^{-1}(\infty ))\), and in turn on \({{\textsf {P}}}^1\setminus (\{\infty \}\cup f^{-1}(\infty ))\) by the density of \({\mathbb {P}}^1\) in \({{\textsf {P}}}^1\) and the continuity of both sides on \({{\textsf {P}}}^1\setminus (\{\infty \}\cup f^{-1}(\infty ))\). By (3.1), this equality is rewritten as
on \({{\textsf {P}}}^1\setminus (\{\infty \}\cup f^{-1}(\infty ))\).
Lemma 4.4
(Pullback formula for \(p_{\mu _f}\) under f) If \(\infty \in {{\mathsf {F}}}(f)\), then
on \({{\textsf {P}}}^1\setminus (\{\infty \}\cup f^{-1}(\infty ));\) moreover, for every \({\mathcal {S}}'\in {{\textsf {P}}}^1\setminus \{\infty ,f(\infty )\}\),
and similarly
Proof
Suppose \(\infty \in {{\mathsf {F}}}(f)\). Then for every \({\mathcal {S}}'\in {{\textsf {P}}}^1\setminus \{\infty ,f(\infty )\}\), by (4.7) and (4.4), we have (4.8). Integrating both sides in (4.8) against \(f^*\delta _{{\mathcal {S}}'}/d\) over \({{\textsf {P}}}^1\), we have (4.9). Similarly, integrating both sides in (4.8) against \(\mu _f\) over \({{\textsf {P}}}^1\), also by \(f_*\mu _f=\mu _f\) and \(I_{\mu _f}:=\int _{{{\textsf {P}}}^1}p_{\mu _f}\mu _f\), we have
so (4.10) also holds. \(\square \)
If \(f(\infty )=\infty \), then \(F(0,1)=(0,c^F_1)\), so that by the homogeneity of F, for every \(n\in {\mathbb {N}}\), \(F^n(0,1)=(0,(c^F_1)^{(d^n-1)/(d-1)})\) and that
Lemma 4.5
If \(f(\infty )=\infty \in {{\mathsf {F}}}(f)\), then
and, for every \({\mathcal {S}}'\in {{\textsf {P}}}^1\),
Proof
Suppose that \(f(\infty )=\infty \in {{\mathsf {F}}}(f)\). Then by the above computation of \(g_f(\infty )\) and (4.5), we have (4.11). Moreover, for every \({\mathcal {S}}'\in {{\textsf {P}}}^1\setminus \{\infty \}\), using (4.6) twice and (4.1) (and the assumption \(f(\infty )=\infty \)), we compute
so (4.12) holds for \({\mathcal {S}}'\in {{\textsf {P}}}^1\setminus \{\infty \}\). Finally, (4.12) for \({\mathcal {S}}'=\infty \) holds by (4.10) and (4.11). \(\square \)
Let us now focus on \(\nu _\infty =\nu _{\infty ,{{\textsf {P}}}^1\setminus D_{\infty }}\) when \(\infty \in {{\mathsf {F}}}(f)\). Then \(f(\infty )\in {{\mathsf {F}}}(f)\) and, since \({\text {supp}}\nu _\infty \subset \partial D_\infty \subset {{\mathsf {J}}}(f)={\text {supp}}\mu _f\), we have
Lemma 4.6
Suppose that \(\infty \in {{\mathsf {F}}}(f)\). Then for every \({\mathcal {S}}'\in {{\textsf {P}}}^1\setminus \{\infty ,f(\infty )\}\),
and, if in addition \(\nu _\infty \) is invariant under f in that \(f_*\nu _\infty =\nu _\infty \) on \({{\textsf {P}}}^1\), then
Proof
Suppose that \(\infty \in {{\mathsf {F}}}(f)\). Then for every \({\mathcal {S}}'\in {{\textsf {P}}}^1\setminus \{\infty ,f(\infty )\}\), using (4.4) repeatedly and (4.1), we have
Moreover, by Fubini’s theorem and \(p_{\nu _\infty }\equiv I_{\nu _\infty }\) on \({{\textsf {P}}}^1\setminus D_\infty \), we also have
which completes the proof of (4.13).
If in addition \(f_*\nu _\infty =\nu _\infty \) on \({{\textsf {P}}}^1\), then by the identity \(p_{\nu _\infty }\equiv I_{\nu _\infty }\) on \({{\textsf {P}}}^1\setminus (D_\infty \cup E)\), where E is an \(F_\sigma \)-subset in \(\partial D_\infty \) of logarithmic capacity 0 with pole \(\infty \), and by the vanishing \(\mu _f(E)=0\) (from (4.5)), we also have
which completes the proof of (4.14). \(\square \)
Lemma 4.7
(Invariance of \(\nu _\infty \) under f) If \(f(\infty )=\infty \in {{\mathsf {F}}}(f)\), then \(f_*\nu _\infty =\nu _\infty \) on \({{\textsf {P}}}^1\) and, for every \({\mathcal {S}}'\in {{\textsf {P}}}^1\),
Proof
Suppose that \(f(\infty )=\infty \in {{\mathsf {F}}}(f)\). Then for every \({\mathcal {S}}'\in {{\textsf {P}}}^1\setminus \{\infty \}\), by (4.13) and (4.12), we have
We claim that
for, by the equality (4.13’) and \(p_{\nu _\infty }\ge I_{\nu _\infty }\) on \({{\textsf {P}}}^1\) (and Fubini’s theorem and (4.4)), we have
so that \(p_{f_*\nu _\infty }\equiv \int _{{{\textsf {P}}}^1}p_{\mu _f}(f_*\nu _\infty )\) \(\mu _f\)-a.e. on \({{\textsf {P}}}^1\). Hence the claim follows by the strong upper semicontinuity (2.1) of \(p_{f_*\nu _\infty }\) on \({{\textsf {P}}}^1\) and \({{\mathsf {J}}}(f)={\text {supp}}\mu _f\), also recalling Remark 3.2.
Once the identity (4.17) is at our disposal, using also the maximum principle for the subharmonic function \(p_{f_*\nu _\infty }\) and the latter inequality in (2.2), we have
and integrating both sides of this inequality against \(f_*\nu _\infty \), we have \(I_{f_*\nu _\infty }\ge I_{\nu _\infty }\) or equivalently
Then (4.16) holds for every \({\mathcal {S}}'\in {{\textsf {P}}}^1\setminus \{\infty \}\) by (4.14) and (4.12). Finally, integrating both sides in (4.8) against \(\nu _\infty \) over \({{\textsf {P}}}^1\), by (4.15) and Fubini’s theorem, we compute
which with (4.11) yields (4.16) for \({\mathcal {S}}'=\infty \). \(\square \)
Remark 4.8
All the computations in this Section are also valid for \(K={\mathbb {C}}\).
Remark 4.9
The f-invariance of \(\nu _{\infty }\) in Lemma 4.7 is a non-archimedean counterpart to Mañé and da Rocha [22, p. 253, before Corollary 1]. Their argument was based on solving Dirichlet problem using the Poisson kernel on \(D_\infty \cup \partial D_\infty \). A similar machinery has been only partly developed in the potential theory on \({{\textsf {P}}}^1\) (see [5, Sects. 7.3, 7.6]).
5 Proof of Theorem 1
Let \(f\in K(z)\) be a rational function of degree \(d>1\), and \(F=(F_0,F_1)\in (K[p_0,p_1]_d)^2\) be a normalized lift of f. When \(\infty \in {{\mathsf {F}}}(f)\), let us still denote \(\nu _{{{\textsf {P}}}^1\setminus D_\infty }=\nu _{\infty ,{{\textsf {P}}}^1\setminus D_\infty }\) by \(\nu _\infty \) for simplicity. If \(\mu _f=\nu _\infty \) on \({{\textsf {P}}}^1\), then not only \(p_{\mu _f}=p_{\nu _\infty }>I_{\nu _\infty }=I_{\mu _f}\) on \(D_\infty \) but, by the continuity of \(p_{\mu _f}\) on \({{\textsf {P}}}^1\setminus \{\infty \}\) (by (4.4)), also \(p_{\mu _f}=p_{\nu _\infty }\equiv I_{\nu _\infty }=I_{\mu _f}\) on \({{\textsf {P}}}^1\setminus D_\infty \).
Suppose that \(\infty \in {{\mathsf {F}}}(f)\), \(f(D_\infty )=D_\infty \) (so \(D_\infty \subset f^{-1}(D_\infty )\)), and \(\mu _f=\nu _\infty \) on \({{\textsf {P}}}^1\). Then by (4.8) and \(p_{\mu _f}\equiv I_{\mu _f}\) on \({{\textsf {P}}}^1\setminus D_\infty \), we have
Let \({\mathcal {S}}_0\) be the point in \({\textsf {H}}^1\) represented by the disk \(\{z\in K:|z|\le e^{I_0}\}\) in K.
Suppose also that \(f^{-1}(D_\infty )\setminus D_\infty \ne \emptyset \). Then \(\deg F_0(1,z)>0\). The subset
in \({{\textsf {P}}}^1\) is the component of \({{\textsf {P}}}^1\setminus (F_0(1,\cdot ))^{-1}({\mathcal {S}}_0)\) containing \(\infty \), and \(\partial U_\infty =(F_0(1,\cdot ))^{-1}({\mathcal {S}}_0)\). By (5.1), we have \(U_\infty \subset f^{-1}(D_\infty )\), and in turn
For every \(w\in f^{-1}(\infty )\setminus \{\infty \} =(F_0(1,\cdot ))^{-1}(0)\subset \{{\mathcal {S}}\in {{\textsf {P}}}^1:|F_0(1,{\mathcal {S}})|_\infty <e^{I_0}\}\), let \(D_w\) (resp. \(U_w\)) be the component of \(f^{-1}(D_\infty )\) (resp. the component of \(\{{\mathcal {S}}\in {{\textsf {P}}}^1:|F_0(1,{\mathcal {S}})|_\infty <e^{I_0}\}\)) containing w. Then \(U_w\) is the component of \({{\textsf {P}}}^1\setminus (F_0(1,\cdot ))^{-1}({\mathcal {S}}_0)\) containing w, and \(\partial U_w\) is a singleton in \((F_0(1,\cdot ))^{-1}({\mathcal {S}}_0)=\partial U_\infty \). For every \(w\in f^{-1}(\infty )\cap D_\infty \), \(D_w=D_\infty \).
We claim that \(\partial D_\infty \) is a singleton say \(\{{\mathcal {S}}_\infty \}\) in \({\textsf {H}}^1\) and, moreover, that for every \(w\in f^{-1}(\infty )\setminus D_\infty (\ne \emptyset \) under the assumption that \(f^{-1}(D_\infty )\setminus D_\infty \ne \emptyset \)),
indeed, for every \(w\in f^{-1}(\infty )\setminus D_\infty \), we not only have \(D_w\subset U_w\) (since otherwise, we must have \(\emptyset \ne D_w\cap U_\infty \subset D_w\cap D_\infty \) so \(D_w=D_\infty \), which contradicts \(w\not \in D_\infty \)) but also \(U_w\subset D_w\) (by (5.1)), so that \(U_w=D_w\). This together with \(\partial U_w\subset \partial U_\infty \) and \(U_\infty \subset D_\infty \) yields
(since otherwise, we must have \(\emptyset \ne U_w\cap D_\infty =D_w\cap D_\infty \) so \(D_w=D_\infty \), which contradicts \(w\not \in D_\infty \)). Hence the claim holds since \(f(\partial U_w)=f(\partial D_w)=\partial D_\infty \) is a singleton in \({\textsf {H}}^1\).
Once the claim is at our disposal, we compute
so f has a potential good reduction. \(\square \)
6 Proof of Theorem 2
Pick a prime number p, and let us denote \(|\cdot |_p\) by \(|\cdot |\) for simplicity. Set
If \(|c|<1\), then \(|ad-bc|=|ad|=1\), so that \(|a|=|d|=1\).
Let \({{\mathsf {J}}}(f\circ A)\) and \({{\mathsf {F}}}(f\circ A)\) denote the Berkovich Julia and Fatou sets in \({{\textsf {P}}}^1({\mathbb {C}}_p)\) of \(f\circ A\) as an element of \({\mathbb {C}}_p(z)\) of degree p, respectively.
6.1 Computing \({\mathsf {J}}(f\circ A)\)
The fact that \(\mathsf {J}(f)\) coincides with the classical Julia set of f (see Remark 3.1), which is \(\mathbb {Z}_p\), is well known (see e.g., [17, Example 4.11], [6, Example 5.30]). In this subsection, more general facts will be established.
Lemma 6.1
If \(|c|<1\), then \((f\circ A)^{-1}({\mathbb {Z}}_p)={\mathbb {Z}}_p\).
Proof
We first claim that for every \(z \in \mathbb {Z}\), \(p\cdot f(z)=z^p-z\equiv 0\) modulo \(p{\mathbb {Z}}\); indeed, when is obvious if \(z=0\) modulo \(p{\mathbb {Z}}\), and is the case by Fermat’s Little Theorem when \(z\ne 0\) modulo \(p{\mathbb {Z}}\). By this claim, we have \(f({\mathbb {Z}})\subset {\mathbb {Z}}\) (cf. [34]), and in turn \(f({\mathbb {Z}}_p)\subset {\mathbb {Z}}_p\) by the continuity of the action of f on \({\mathbb {Q}}_p\) and the density of \({\mathbb {Z}}\) in \({\mathbb {Z}}_p\). Next, we claim that \(f^{-1}({\mathbb {Z}}_p)\subset {\mathbb {Z}}_p\) or equivalently that for every \(w\in {\mathbb {Z}}_p\), \(f^{-1}(w)\subset {\mathbb {Z}}_p\); indeed, setting \(W(X):=X^p-X-pw\in {\mathbb {Z}}_p[X]\) of degree p, we have already seen that the reduction \(\overline{W}(X)=X^p-X\in {\mathbb {F}}_p[X]\) of W modulo \(p{\mathbb {Z}}_p\) has p distinct roots \(\overline{0},\ldots ,\overline{p-1}\) in \({\mathbb {F}}_p\). Hence by Hensel’s lemma (see, e.g., [24, Corollary 1 in Sect. 5.1], [8, Sect. 3.3.4, Proposition 3]), W(X) also has p distinct roots in \({\mathbb {Z}}_p\), and has no other roots in \(\overline{{\mathbb {Q}}_p}\), so the claim holds. We have seen that \(f^{-1}({\mathbb {Z}}_p)={\mathbb {Z}}_p\).
Suppose now that \(|c|<1\). Then for every \(z\in {\mathbb {Z}}_p\), we have \(|cz|<1=|d|\), so that \(|A(z)|=|az+b|/|cz+d|=|az+b|\le 1\). Hence \(A({\mathbb {Z}}_p)\subset {\mathbb {Z}}_p\), and similarly \(A^{-1}({\mathbb {Z}}_p)\subset {\mathbb {Z}}_p\) since \(A^{-1}(z)=(dz-b)/(-cz+a)\in \mathrm {PGL}(2,{\mathbb {Z}}_p)\) and \(|-c|=|c|<1\). Now we conclude that \((f\circ A)^{-1}({\mathbb {Z}}_p)=A^{-1}({\mathbb {Z}}_p)={\mathbb {Z}}_p\). \(\square \)
Lemma 6.2
If \(|b|\ll 1\) and \(|c|\ll 1\), then \(f\circ A\) has an attracting fixed point \(z_A\) in \({\mathbb {P}}^1({\mathbb {C}}_p)\setminus {\mathbb {Z}}_p\), which tends to \(\infty \) as \((a,b,c,d)\rightarrow (1,0,0,1)\) in \(({\mathbb {Z}}_p)^4\). Moreover, if in addition \(c\ne 0\), then \(z_A\in {\mathbb {C}}_p\setminus {\mathbb {Z}}_p\) and \((f\circ A)^{-1}(z_A)\ne \{z_A\}\).
Proof
Since \(f^{-1}(\infty )=\{\infty \}\) and \(\deg f=p>1\), the former assertion holds also noting that \(({\mathrm {Id}}_{{\mathbb {P}}^1({\mathbb {C}}_p)})'\equiv 1\ne 0\) and applying an implicit function theorem to the equation \((f\circ A)(z)=z\) near \((z,a,b,c,d)=(\infty ,1,0,0,1)\) in \({\mathbb {P}}^1({\mathbb {C}}_p)\times ({\mathbb {Z}}_p)^4\) (see, e.g., [1, (10.8)]). Moreover, since \(f'(z)=z^{p-1}-p^{-1}\) and \(f''(z)=(p-1)z^{p-2}\), the point \(A^{-1}(\infty )=-d/c\) is the unique point \(z\in {\mathbb {P}}^1({\mathbb {C}}_p)\) such that \(\deg _z(f\circ A)=p(=\deg (f\circ A))\), and on the other hand, if in addition \(c\ne 0\), then the point \(A^{-1}(\infty )\) is \(\ne \infty \) and is not fixed by \(f\circ A\). Hence the latter assertion holds also noting that \((f\circ A)(\infty )\ne \infty \) if in addition \(c\ne 0\). \(\square \)
Consequently, if \(|b|\ll 1\) and \(|c|\ll 1\), then
indeed, by Lemma 6.1 (and (3.2)), if \(|c|<1\), then \({{\mathsf {J}}}(f\circ A)\subset {\mathbb {Z}}_p\). If in addition \(|b|\ll 1\) and \(|c|\ll 1\), then by Lemma 6.2 (and \({\mathbb {Z}}_p\subset {\mathbb {C}}_p\)), we have \({{\mathsf {F}}}(f\circ A)=D_{z_A}(f\circ A)\), which is an (immediate) attractive basin of f (see [31, Théorème de Classification]) associated with \(z_A\in {\mathbb {P}}^1({\mathbb {C}}_p)\setminus {\mathbb {Z}}_p\), and in turn have \({{\mathsf {J}}}(f\circ A)={\mathbb {Z}}_p\) since \((f\circ A)({\mathbb {Z}}_p)\subset {\mathbb {Z}}_p\) by Lemma 6.1.
6.2 Computing Energies and Measures
Since
the pair
is a normalized lift of f. Noting that \(|{\text {Res}}(az_0+bz_1,cz_0+dz_1)|=|ad-bc|=1\) and using a formula for the homogeneous resultant of the composition of homogeneous polynomial maps (see, e.g., [32, Exercise 2.12]), we also have \(\bigl |{\text {Res}}\bigl (F(az_0+bz_1,cz_0+dz_1)\bigr )\bigr | =\bigl |({\text {Res}}F)^1\cdot ({\text {Res}}(az_0+bz_1,cz_0+dz_1))^{p^2}\bigr | =1\), so that
is a normalized lift of \(f\circ A\). For every \(n\in {\mathbb {N}}\), write
Lemma 6.3
If \(|b|<1\) and \(|c|<1\), then
Proof
Suppose that \(|b|<1\) and \(|c|<1\)(, and recall \(|p|=p^{-1}<1\)). Then for every \((z_0,z_1)\in {\mathbb {C}}_p^2\), if \(|z_0|<|z_1|\), then
so
Hence inductively, for every \(n\in {\mathbb {N}}\), we have \(|F_{A,0}^{(n)}(0,1)|<|F_{A,1}^{(n)}(0,1)|\), and moreover
as \(n\rightarrow \infty \). \(\square \)
Lemma 6.4
If (a, b, c, d) is close enough to (1, 0, 0, 1) in \(({\mathbb {Z}}_p)^4\), then
Proof
If \(|b|\ll 1\) and \(|c|\ll 1\), then by (6.1) and \({\mathbb {Z}}_p\subset {\mathbb {C}}_p\), we have
Then by (4.5) and Lemma 6.3, we have
and in particular, recalling \(\nu _{\infty ,{\mathbb {Z}}_p}=\mu _f\) on \({{\textsf {P}}}^1({\mathbb {C}}_p)\), also \(I_{\infty ,\nu _{\infty ,{\mathbb {Z}}_p}}=I_{\infty ,\mu _f}=\log p^{\frac{-1}{p-1}}\) (for a non-dynamical and more direct computation of \(I_{\infty ,\nu _{\infty ,{\mathbb {Z}}_p}}\), see [3]). Now the first equality holds by the uniqueness of the equilibrium mass distribution on the non-polar compact subset \({\mathbb {Z}}_p\) in \({{\textsf {P}}}^1({\mathbb {C}}_p)\). The second equality holds since \(z_A\) tends to \(\infty \) as \((a,b,c,d)\rightarrow (1,0,0,1)\) in \(({\mathbb {Z}}_p)^4\) (by Lemma 6.2), also recalling Observation 2.4. \(\square \)
Remark 6.5
If \(0<|c|\ll 1\) and \(|b|\ll 1\), then \((f\circ A)(\infty )\ne \infty \in {{\mathsf {F}}}(f\circ A)\), \((f\circ A)(D_\infty (f\circ A))=D_\infty (f\circ A)\), \({{\mathsf {J}}}(f\circ A)\not \subset {\textsf {H}}^1\) (indeed \({{\mathsf {J}}}(f\circ A)\subset {\mathbb {C}}_p\)), and \(\mu _{f\circ A}=\nu _{\infty ,{{\textsf {P}}}^1\setminus D_\infty }\) on \({{\textsf {P}}}^1\).
6.3 Conclusion
If \(|b|\ll 1\) and \(0<|c|\ll 1\), then setting \(m_A(z):=\frac{1}{z-z_A}\in \mathrm {PGL}(2,{\mathbb {C}}_p)\), the rational function
is of degree p and satisfies \(g_A(\infty )=\infty ,|g_A'(\infty )|<1, g_A^{-1}(\infty )\ne \{\infty \}\), and \(\infty \in m_A(D_{z_A}(f\circ A))=D_\infty (g_A)\). If moreover (a, b, c, d) is close enough to (1, 0, 0, 1) in \(({\mathbb {Z}}_p)^4\), then also recalling Observations 2.2 and 2.3, we have
Now the proof of Theorem 2 is complete. \(\square \)
References
Abhyankar, S. S.: Local Analytic Geometry. World Scientific Publishing Co. Inc., River Edge (2001) (reprint of the 1964 original)
Baker, M.: A finiteness theorem for canonical heights attached to rational maps over function fields. J. Reine Angew. Math. 626, 205–233 (2009)
Baker, M.H., Hsia, L.-C.: Canonical heights, transfinite diameters, and polynomial dynamics. J. Reine Angew. Math. 585, 61–92 (2005)
Baker, M.H., Rumely, R.: Equidistribution of small points, rational dynamics, and potential theory. Ann. Inst. Fourier (Grenoble) 56(3), 625–688 (2006)
Baker, M., Rumely, R.: Potential theory and dynamics on the Berkovich projective line, Mathematical Surveys and Monographs, vol. 159. American Mathematical Society, Providence (2010)
Benedetto, R.L.: Dynamics in one non-archimedean variable, Graduate Studies in Mathematics, vol. 198. American Mathematical Society, Providence (2019)
Berkovich, V.G.: Spectral theory and analytic geometry over non-Archimedean fields. In: Mathematical Surveys and Monographs, vol. 33. American Mathematical Society, Providence (1990)
Bosch, S., Güntzer, U., Remmert, R.: Non-Archimedean analysis. In: Grundlehren der Mathematischen Wissenschaften [Fundamental Principles of Mathematical Sciences], vol. 261. Springer, Berlin (1984) (a systematic approach to rigid analytic geometry)
Brolin, H.: Invariant sets under iteration of rational functions. Ark. Mat. 6, 103–144 (1965)
Chambert-Loir, A.: Mesures et équidistribution sur les espaces de Berkovich. J. Reine Angew. Math. 595, 215–235 (2006)
DeMarco, L.: Dynamics of rational maps: Lyapunov exponents, bifurcations, and capacity. Math. Ann. 326(1), 43–73 (2003)
Favre, C., Jonsson, M.: The valuative tree. In: Lecture Notes in Mathematics, vol. 1853. Springer, Berlin (2004)
Favre, C., Rivera-Letelier, J.: Équidistribution quantitative des points de petite hauteur sur la droite projective. Math. Ann. 335(2), 311–361 (2006)
Favre, C., Rivera-Letelier, J.: Théorie ergodique des fractions rationnelles sur un corps ultramétrique. Proc. Lond. Math. Soc. (3) 100(1), 116–154 (2010)
Freire, A., Lopes, A., Mañé, R.: An invariant measure for rational maps. Bol. Soc. Brasil. Mat. 14(1), 45–62 (1983)
Fresnel, J., Put, van der M.: Rigid analytic geometry and its applications. In: Progress in Mathematics, vol. 218. Birkhäuser Boston, Inc., Boston (2004)
Hsia, L.-C.: Closure of periodic points over a non-Archimedean field. J. Lond. Math. Soc. (2), 62(3) , 685–700 (2000)
Jonsson, M.: Dynamics on Berkovich spaces in low dimensions, Berkovich Spaces and Applications. pp. 205–366. Springer, New York (2015)
Lalley, S.P.: Brownian motion and the equilibrium measure on the Julia set of a rational mapping. Ann. Probab. 20(4), 1932–1967 (1992)
Ljubich, M.J.: Entropy properties of rational endomorphisms of the Riemann sphere. Ergod. Theory Dyn. Syst. 3(3), 351–385 (1983)
Lopes, A.O.: Equilibrium measures for rational maps. Ergod. Theory Dyn. Syst. 6(3), 393–399 (1986)
Mañé, R., da Rocha, L.F.: Julia sets are uniformly perfect. Proc. Am. Math. Soc. 116(3), 251–257 (1992)
Mañé, R.: On the uniqueness of the maximizing measure for rational maps. Bol. Soc. Brasil. Mat. 14(1), 27–43 (1983)
Narkiewicz, W.: Elementary and analytic theory of algebraic numbers, 3rd edn. Springer Monographs in Mathematics. Springer, Berlin (2004)
Oba, M.K., Pitcher, T.S.: A new characterization of the \(F\) set of a rational function. Trans. Am. Math. Soc. 166, 297–308 (1972)
Okuyama, Y.: Adelic equidistribution, characterization of equidistribution, and a general equidistribution theorem in non-archimedean dynamics. Acta. Arith. 161(2), 101–125 (2013)
Okuyama, Y.: Quantitative approximations of the Lyapunov exponent of a rational function over valued fields. Math. Z. 280(3–4), 691–706 (2015)
Okuyama, Y.: Effective divisors on the projective line having small diagonals and small heights and their application to adelic dynamics. Pac. J. Math. 280(1), 141–175 (2016)
Okuyama, Y., Stawiska, M.: Potential theory and a characterization of polynomials in complex dynamics. Conform. Geom. Dyn. 15, 152–159 (2011)
Okuyama, Y., Stawiska, M.: A generalization of the converse of Brolin’s theorem. Ann. Polon. Math. 123(1), 467–472 (2019)
Rivera-Letelier, J.: Dynamique, des fonctions rationnelles sur des corps locaux. Astérisque 287:xv, 147–230 (2003) (geometric methods in dynamics, II)
Silverman, J. H.: The arithmetic of dynamical systems. In: Graduate Texts in Mathematics, vol. 241. Springer, New York (2007)
Thuillier, A.: Théorie du potentiel sur les courbes en géométrie analytique non archimédienne. Applications à la théorie d’Arakelov. PhD thesis, Université Rennes 1 (2005)
Woodcock, C.F., Smart, N.P.: \(p\)-adic chaos and random number generation. Exp. Math. 7(4), 333–342 (1998)
Acknowledgements
The authors thank the referee for a very careful scrutiny and invaluable comments. The first author was partially supported by JSPS Grant-in-Aid for Scientific Research (C), 15K04924 and 19K03541 and (B), 19H01798.
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
About this article
Cite this article
Okuyama, Y., Stawiska, M. On a Characterization of Polynomials Among Rational Functions in Non-Archimedean Dynamics. Arnold Math J. 6, 407–430 (2020). https://doi.org/10.1007/s40598-020-00145-9
Received:
Revised:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s40598-020-00145-9