This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

Arithmetic degree and its application to Zariski dense orbit conjecture

Yohsuke Matsuzawa  and  Junyi Xie Department of Mathematics, Graduate School of Science, Osaka Metropolitan University, 3-3-138, Sugimoto, Sumiyoshi, Osaka, 558-8585, Japan [email protected] Beijing International Center for Mathematical Research, Peking University, Beijing 100871, China [email protected]
Abstract.

We prove that for a dominant rational self-map ff on a quasi-projective variety defined over ¯\overline{\mathbb{Q}}, there is a point whose ff-orbit is well-defined and its arithmetic degree is arbitrary close to the first dynamical degree of ff. As an application, we prove that Zariski dense orbit conjecture holds for a birational map defined over ¯\overline{\mathbb{Q}} such that the first dynamical degree is strictly larger than the third dynamical degree. In particular, the conjecture holds for birational maps on threefolds with first dynamical degree larger than 11.

Key words and phrases:
Arithmetic dynamics, Arithmetic degree, Zariski dense orbit conjecture
2020 Mathematics Subject Classification:
Primary 37P15; Secondary 37P55

1. Introduction

For a dominant rational map f:XXf\colon X\dashrightarrow X on a projective variety defined over ¯\overline{\mathbb{Q}}, Kawaguchi-Silverman conjecture predicts that height growth rate along a Zariski dense orbit is equal to the first dynamical degree of ff. More precisely, let LL be an ample divisor on XX and let hLh_{L} be a Weil height function associated with LL (we refer [11, 16, 5] for the basics of height functions). For a point xX(¯)x\in X(\overline{\mathbb{Q}}), we say the ff-orbit is well-defined if

(1.1) fn(x)If,n0\displaystyle f^{n}(x)\notin I_{f},\quad n\geq 0

where IfI_{f} is the indeterminacy locus of ff. The set of such points is denoted by Xf(¯)X_{f}(\overline{\mathbb{Q}}):

(1.2) Xf(¯)={xX(¯)fn(x)If,n0}.\displaystyle X_{f}(\overline{\mathbb{Q}})=\{x\in X(\overline{\mathbb{Q}})\mid f^{n}(x)\notin I_{f},\ n\geq 0\}.

For xXf(¯)x\in X_{f}(\overline{\mathbb{Q}}),

(1.3) αf(x):=limnmax{1,hL(fn(x))}1n\displaystyle\alpha_{f}(x):=\lim_{n\to\infty}\max\{1,h_{L}(f^{n}(x))\}^{\frac{1}{n}}

is called the arithmetic degree of ff at xx, provided the limit exists. By the basic properties of height function, it is easy to see that the limit is independent of the choice of LL and hLh_{L}. The existence of the limit is proven for surjective self-morphisms on projective varieties [14, Theorem 3] (it is stated for normal projective varieties, but the general case easily follows from normal case by taking normalization), and for arbitrary dominant rational self-maps and points with generic orbit [19, Theorem 1.3]. (A orbit is generic if it converges to the generic point with respect to Zariski topology. More generally, the convergence of arithmetic degree is proven for orbits satisfying dynamical Mordell-Lang conjecture.)

For i=0,,dimXi=0,\dots,\dim X, the ii-th dynamical degree of ff is

(1.4) λi(f)=limndegi,L(fn)1n\displaystyle\lambda_{i}(f)=\lim_{n\to\infty}\deg_{i,L}(f^{n})^{\frac{1}{n}}

where the ii-th degree degi,L(fn)\deg_{i,L}(f^{n}) is defined as follows. Let ΓfnX×X\Gamma_{f^{n}}\subset X\times X be the graph of fnf^{n} and let pi:ΓfnXp_{i}:\Gamma_{f^{n}}\longrightarrow X be the projections (i=1,2i=1,2):

(1.5) Γfn{\Gamma_{f^{n}}}X{X}X{X}p1\scriptstyle{p_{1}}p2\scriptstyle{p_{2}}fn\scriptstyle{f^{n}}

Then we define

(1.6) degi,L(fn)=(p2Lip1LdimXi).\displaystyle\deg_{i,L}(f^{n})=\big{(}p_{2}^{*}L^{i}\cdot p_{1}^{*}L^{\dim X-i}\big{)}.

It is known that the limits exist and independent of the choice of LL (cf. [8, 7, 28]).

Now let us state Kawaguchi-Silverman conjecture.

Conjecture 1.1 (Kawaguchi-Silverman conjecture [26, 15]).

Let f:XXf\colon X\dashrightarrow X be a dominant rational map on a projective variety defined over ¯\overline{\mathbb{Q}}. Let xXf(¯)x\in X_{f}(\overline{\mathbb{Q}}). Then αf(x)\alpha_{f}(x) exists (i.e. the limit exists), and if the orbit Of(x)={x,f(x),f2(x),}O_{f}(x)=\{x,f(x),f^{2}(x),\dots\} is Zariski dense in XX, then αf(x)=λ1(f)\alpha_{f}(x)=\lambda_{1}(f).

We refer [18] for introduction and recent advances on this conjecture. It is known that for any xXf(¯)x\in X_{f}(\overline{\mathbb{Q}}), the limsup version of arithmetic degree is bounded above by the first dynamical degree [17, Theorem 1.4] [12, Theorem 3.11]:

(1.7) α¯f(x):=lim supnmax{1,hL(fn(x))}1nλ1(f).\displaystyle\overline{\alpha}_{f}(x):=\limsup_{n\to\infty}\max\{1,h_{L}(f^{n}(x))\}^{\frac{1}{n}}\leq\lambda_{1}(f).

Thus the conjecture asserts that the arithmetic degree would take its maximal value at points with dense orbit. Although there is no logical implications, it is natural to ask that if there is always a point xXf(¯)x\in X_{f}(\overline{\mathbb{Q}}) such that αf(x)=λ1(f)\alpha_{f}(x)=\lambda_{1}(f). The answer is yes for surjective morphisms on projective varieties [20, Theorem 1.6] (it is stated only for smooth projective varieties, but the proof works for any projective varieties; just find a point at which the nef canonical height does not vanish), and also for some classes of rational maps [13, Theorem 3]. See [21, 23, 24] for related works. In this paper, we prove the following.

Theorem 1.2.

Let XX be a projective variety over ¯\overline{\mathbb{Q}}. Let f:XXf\colon X\dashrightarrow X be a dominant rational map defined over ¯\overline{\mathbb{Q}}. Then for any ε>0\varepsilon>0, the set

(1.8) {xXf(¯)|αf(x) exists and αf(x)λ1(f)ε }\displaystyle\left\{x\in X_{f}(\overline{\mathbb{Q}})\ \middle|\ \text{$\alpha_{f}(x)$ exists and $\alpha_{f}(x)\geq\lambda_{1}(f)-\varepsilon$ }\right\}

is Zariski dense in XX.

Remark 1.3.

The set eq. 1.8 is actually dense in X(¯)X(\overline{\mathbb{Q}}) with respect to the adelic topology (in the sense of [29]). See Theorem 3.1.

Remark 1.4.

We prove the same statement for quasi-projective varieties (Theorem 3.1). The arithmetic degree and the dynamical degrees are defined as follows. Take a projective closure ι:XX\iota\colon X\hookrightarrow X^{\prime}, i.e. open immersion into a projective variety XX^{\prime} over ¯\overline{\mathbb{Q}}. Then a dominant rational map f:XXf\colon X\dashrightarrow X can be regraded as that of on XX^{\prime}, denoted by ff^{\prime}. Then Xf(¯)Xf(¯)X_{f}(\overline{\mathbb{Q}})\subset X^{\prime}_{f^{\prime}}(\overline{\mathbb{Q}}), and we define αf(x):=αf(x)\alpha_{f}(x):=\alpha_{f^{\prime}}(x) for xXf(¯)x\in X_{f}(\overline{\mathbb{Q}}) (cf. [18, Definition 2.3]). The well-definedness, i.e. independence of the embedding follows from [12, Lemma 3.8], the same trick as in Remark 2.2. The dynamical degrees are defined in the same way: λi(f):=λi(f)\lambda_{i}(f):=\lambda_{i}(f^{\prime}). By the birational invariance of dynamical degrees (cf. [8, 7, 28]), this definition is also independent of the embedding ι\iota.

In the proof of [12, Theorem 8.4], they find an application of arithmetic degree to the following Zariski dense orbit conjecture.

Conjecture 1.5 (Zariski dense orbit conjecture [22, Conjecture 7.14], cf. [31, Conjecture 4.1.6] as well).

Let XX be a projective variety over an algebraically closed field kk of characteristic zero, and let f:XXf:X\dashrightarrow X be a dominant rational self-map. If every ff-invariant rational function on XX is constant, then there exists xXf(k)x\in X_{f}(k) whose orbit Of(x)O_{f}(x) is Zariski dense in XX.

Here Xf(k)X_{f}(k) is the set of points with well-defined ff-orbit, defined in the same way as eq. 1.2. We refer [29] for the history of this conjecture and known results. We remark that the conjecture is proven when the ground field kk is uncountable [1, 2]. The conjecture remains open over countable fields, in particular over ¯\overline{\mathbb{Q}}.

The idea in [12, Theorem 8.4] is, roughly speaking, that a point xXf(¯)x\in X_{f}(\overline{\mathbb{Q}}) with αf(x)=λ1(f)\alpha_{f}(x)=\lambda_{1}(f) must have Zariski dense orbit under some conditions on the map ff. Using the same idea, in [21, Theorem C], the conjecture is proven for cohomologically hyperbolic birational self-maps on smooth projective threefolds. In this paper, we weaken the assumption “cohomologically hyperbolic” to “λ1(f)>1\lambda_{1}(f)>1”. More generally, we prove the following.

Theorem 1.6.

Let XX be a projective variety over ¯\overline{\mathbb{Q}}. Let f:XXf\colon X\dashrightarrow X be a birational map. If λ3(f)<λ1(f)\lambda_{3}(f)<\lambda_{1}(f), then Zariski dense orbit conjecture holds for ff. That is, if ff does not admit invariant non-constant rational functions, then there is a point xXf(¯)x\in X_{f}(\overline{\mathbb{Q}}) with Of(x)O_{f}(x) being Zariski dense.

Remark 1.7.

Under the assumption of Theorem 1.6, if ff does not admit invariant non-constant rational functions, then the set of points xXf(¯)x\in X_{f}(\overline{\mathbb{Q}}) with Zariski dense orbit is dense in X(¯)X(\overline{\mathbb{Q}}) with respect to the adelic topology (in the sense of [29]). See Theorem 4.1.

As a corollary, we have:

Corollary 1.8.

Let XX be a projective variety of dimension three over ¯\overline{\mathbb{Q}}. Let f:XXf\colon X\dashrightarrow X be a birational map with λ1(f)>1\lambda_{1}(f)>1. Then the Zariski dense orbit conjecture holds for ff.

Proof.

Since λ3(f)=1\lambda_{3}(f)=1, the assumption of Theorem 1.6 is satisfied. ∎

Idea of the proof.

The idea of the proof of Theorem 1.2 is as follows. By a recent work of the second author [30], we roughly have

(1.9) hL(fn+2(x))(1+ε)μhL(fn+1(x))\displaystyle h_{L}(f^{n+2}(x))-(1+\varepsilon)\mu h_{L}(f^{n+1}(x))
(1.10) (1ε)λ1(f)(hL(fn+1(x))(1+ε)μhL(fn(x)))\displaystyle\geq(1-\varepsilon)\lambda_{1}(f)(h_{L}(f^{n+1}(x))-(1+\varepsilon)\mu h_{L}(f^{n}(x)))

for some 0μ<λ1(f)0\leq\mu<\lambda_{1}(f) (after replacing ff with its iterate). The main problem is that we do not know in general if hL(fn+1(x))(1+ε)μhL(fn(x)>0h_{L}(f^{n+1}(x))-(1+\varepsilon)\mu h_{L}(f^{n}(x)>0 for some nn. To find such a point xx, we consider a curve CC such that the degrees of fn(C)f^{n}(C) grow as fast as possible, i.e. in the order of λ1(f)n\lambda_{1}(f)^{n}. Then for a point xC(¯)x\in C(\overline{\mathbb{Q}}), we expect inequality hL(f(x))λ1(f)hL(x)h_{L}(f(x))\geq\lambda_{1}(f)h_{L}(x) hold. This is justified for points with large height, but we also need some additional good properties of xx, including well-definedness of its ff-orbit. The latter property is satisfied for any points in some adelic open subset (in the sense of [29]). We ensure the existence of xC(¯)x\in C(\overline{\mathbb{Q}}) with all desired properties by proving that height function is unbounded on a non-empty adelic open subset Proposition 2.3. Once we find such a point, eq. 1.9 shows αf(x)(1ε)λ1(f)\alpha_{f}(x)\geq(1-\varepsilon)\lambda_{1}(f).

The idea of the proof of Theorem 1.6 is as follows. By Theorem 1.2, there is a point xx such that αf(x)>λ3(f)\alpha_{f}(x)>\lambda_{3}(f). It is known that if birational ff does not admit invariant non-constant rational function, then there are only finitely many totally invariant hypersurfaces. Thus we may assume the orbit closure Of(x)¯\overline{O_{f}(x)} is either XX or has codimension at least two. If it is XX, we are done. If it has codimension r2r\geq 2, then we can show roughly αf(x)λ1(f|Of(x)¯)λ1+r(f)λ3(f)\alpha_{f}(x)\leq\lambda_{1}(f|_{\overline{O_{f}(x)}})\leq\lambda_{1+r}(f)\leq\lambda_{3}(f), and this is contradiction.

Convention.

  • An algebraic scheme over a field kk is a separated scheme of finite type over kk.

  • A variety over kk is an algebraic scheme over kk which is irreducible and reduced.

  • For a self-morphism f:XXf\colon X\longrightarrow X of an algebraic scheme over kk and a point xx of XX (scheme point or kk^{\prime}-valued point where kk^{\prime} is a field containing kk), the ff-orbit of xx is denoted by Of(x)O_{f}(x), i.e. Of(x)={fn(x)n=0,1,2,}O_{f}(x)=\{f^{n}(x)\mid n=0,1,2,\dots\}. The same notation is used for dominant rational map f:XXf\colon X\dashrightarrow X on a variety XX defined over kk and xXf(k)={xX(k)fn(x)If,n0}x\in X_{f}(k)=\{x\in X(k)\mid f^{n}(x)\notin I_{f},\ n\geq 0\}. Here IfI_{f} is the indeterminacy locus of ff.

  • Let f:XXf\colon X\dashrightarrow X be a dominant rational map on a variety XX over a field kk. For a point xXf(k)x\in X_{f}(k), we say (X,f,x)(X,f,x) satisfies DML property if for any closed subset WXW\subset X, the return set {n0fn(x)W}\{n\geq 0\mid f^{n}(x)\in W\} is a finite union of arithmetic progressions.

  • Let kk be an algebraically closed field of characteristic zero. For a dominant rational map f:XXf\colon X\dashrightarrow X on a variety over kk, λi(f)\lambda_{i}(f) denotes the ii-th dynamical degree of ff for i=0,,dimXi=0,\dots,\dim X. The cohomological Lyapunov exponent is denoted by μi(f)=λi(f)/λi1(f)\mu_{i}(f)=\lambda_{i}(f)/\lambda_{i-1}(f) for i=1,,dimXi=1,\dots,\dim X. We set μdimX+1(f)=0\mu_{\dim X+1}(f)=0.

Acknowledgements.

The essential part of the work was done during the Simons symposium “Algebraic, Complex, and Arithmetic Dynamics (2024)”. The authors would like to thank Simons Foundation, Laura DeMarco, and Mattias Jonsson for hosting and organizing the symposium. The authors also would like to thank Joe Silverman and Long Wang for helpful comments. The first author is supported by JSPS KAKENHI Grant Number JP22K13903. The second author is supported by the NSFC Grant No.12271007.

2. Height unboundedness on adelic open sets

In this section, we prove that height function associated with an ample divisor is unbounded on a non-empty adelic open subset. For an algebraic scheme XX over ¯\overline{\mathbb{Q}}, the adelic topology is a topology on X(¯)X(\overline{\mathbb{Q}}) introduced by the second author in [29]. The definition involves several steps, so we do not write down it here and refer [29, section 3] for the definition and basic properties. The point of the topology is that it allows us to discuss analytic local properties of ¯\overline{\mathbb{Q}}-points (because it is defined by using pp-adic open sets) while keeping coarseness of Zariski topology; if XX is irreducible, then X(¯)X(\overline{\mathbb{Q}}) is irreducible with respect to the adelic topology.

Definition 2.1.

Let XX be a quasi-projective scheme over ¯\overline{\mathbb{Q}}. A subset AX(¯)A\subset X(\overline{\mathbb{Q}}) is said to be height bounded if the following condition holds. For any immersion i:XPi\colon X\hookrightarrow P into a projective scheme PP defined over ¯\overline{\mathbb{Q}}, any ample Cartier divisor HH on XX, and any logarithmic Weil height function hHh_{H} associated with HH, the subset

(2.1) {hH(i(x))|xA}\displaystyle\left\{h_{H}(i(x))\ \middle|\ x\in A\right\}\subset{\mathbb{R}}

is bounded.

Remark 2.2.

The set is always bounded below since so is hHh_{H}. The definition remains equivalent if we require the boundedness only for some i:XPi\colon X\hookrightarrow P, HH, and hHh_{H}. Indeed, if j:XPj\colon X\hookrightarrow P^{\prime} is another immersion to projective scheme, HH^{\prime} is ample Cartier divisor on PP^{\prime}, and hHh_{H^{\prime}} is a height associated with HH^{\prime}, form the following diagram:

(2.2) P{P}X{X}X¯{\overline{X}}P×P{P\times P^{\prime}}P{P^{\prime}}i\scriptstyle{i}j\scriptstyle{j}p1\scriptstyle{p_{1}}p2\scriptstyle{p_{2}}pr2\scriptstyle{\operatorname{pr}_{2}}pr1\scriptstyle{\operatorname{pr}_{1}}

where X¯\overline{X} is the scheme theoretic closure of (i,j)(X)(i,j)(X) in P×PP\times P^{\prime}. Take n1n\geq 1 so that

(2.3) p2(p1𝒪P(H)𝒪X¯p2𝒪P(nH))(p2p1𝒪P(H))𝒪P𝒪P(nH)\displaystyle{p_{2}}_{*}\big{(}p_{1}^{*}{\mathcal{O}}_{P}(-H){\otimes}_{{\mathcal{O}}_{\overline{X}}}p_{2}^{*}{\mathcal{O}}_{P^{\prime}}(nH^{\prime})\big{)}\simeq\big{(}{p_{2}}_{*}p_{1}^{*}{\mathcal{O}}_{P}(-H)\big{)}{\otimes}_{{\mathcal{O}}_{P^{\prime}}}{\mathcal{O}}_{P^{\prime}}(nH^{\prime})

is globally generated. Note that p21(j(X))=Xp_{2}^{-1}(j(X))=X. Then the base locus of np2Hp1Hnp_{2}^{*}H^{\prime}-p_{1}^{*}H is contained in X¯X\overline{X}\setminus X, and hence nhHhHO(1)nh_{H^{\prime}}-h_{H}\geq O(1) on X(¯)X(\overline{\mathbb{Q}}). Similarly, there is m1m\geq 1 such that mhHhHO(1)mh_{H}-h_{H^{\prime}}\geq O(1) on X(¯)X(\overline{\mathbb{Q}}). Thus we are done.

We use the notation and terminologies on adelic open subsets from [29, section 3].

Proposition 2.3.

Let XX be a quasi-projective variety over ¯\overline{\mathbb{Q}} with dimX1\dim X\geq 1. Let AX(¯)A\subset X(\overline{\mathbb{Q}}) be a non-empty adelic open subset in the sense of [29]. Then AA is not height bounded.

To prove this proposition, we prepare some terminologies and a lemma.

Definition 2.4.

Let K¯K\subset\overline{\mathbb{Q}} be a number field. For an algebraic scheme XX over KK and d1d\in{\mathbb{Z}}_{\geq 1}, we define

(2.4) X(d):=KL¯[L:K]dX(L)X(¯),\displaystyle X(d):=\bigcup_{\begin{subarray}{c}K\subset L\subset\overline{\mathbb{Q}}\\ [L:K]\leq d\end{subarray}}X(L)\subset X(\overline{\mathbb{Q}}),

where each X(L)X(L) is regarded as a subset of X(¯)X(\overline{\mathbb{Q}}) via the inclusion L¯L\subset\overline{\mathbb{Q}}.

Lemma 2.5.

Let XX be a quasi-projective variety over ¯\overline{\mathbb{Q}} with dimX1\dim X\geq 1. Let AX(¯)A\subset X(\overline{\mathbb{Q}}) be a non-empty basic adelic subset in the sense of [29, section 3]. Let K¯K\subset\overline{\mathbb{Q}} be a number field and XKX_{K} a model of XX over KK. Then there is d1d\in{\mathbb{Z}}_{\geq 1} such that AXK(d)A\cap X_{K}(d) is Zariski dense in XX.

This follows from the proof of [29, Proposition 3.9]. We include here a proof for the completeness.

Proof.

By replacing KK with a finite extension and replacing AA with an appropriate subset, we may assume AA is a basic adelic subset over KK with respect to XKX_{K}. Moreover, we may assume

(2.5) A=XK((τi,Ui),i=1,,m)\displaystyle A=X_{K}((\tau_{i},U_{i}),i=1,\dots,m)

where τi:Kpτi\tau_{i}\colon K\longrightarrow{\mathbb{C}}_{p_{\tau_{i}}} are field embeddings such that ||i:=|τi()|pτi|\ |_{i}:=|\tau_{i}(\ )|_{{\mathbb{C}}_{p_{\tau_{i}}}} are distinct absolute values on KK, and as usual UiXK(pτi)U_{i}\subset X_{K}({\mathbb{C}}_{p_{\tau_{i}}}) are non-empty pτip_{\tau_{i}}-adic open subsets. (cf. the beginning of the proof of [29, Proposition 3.9]. ) Let KpiK_{p_{i}} be the closure of τi(K)\tau_{i}(K) in pτi{\mathbb{C}}_{p_{\tau_{i}}}. By further replacing KK with a finite extension, we may assume UiXK(Kpi)U_{i}\cap X_{K}(K_{p_{i}})\neq\emptyset. Note that this in particular implies UiXK(Kpi)U_{i}\cap X_{K}(K_{p_{i}}) is Zariski dense in (XK)Kpi(X_{K})_{K_{p_{i}}}.

By Noether normalization, there is a non-empty open subscheme XKXKX_{K}^{\circ}\subset X_{K} with finite étale morphism

(2.6) π:XKV\displaystyle\pi\colon X_{K}^{\circ}\longrightarrow V

to an open subscheme V𝔸KdV\subset{\mathbb{A}}^{d}_{K} of an affine space. By taking a connected Galois étale covering of VV dominating XKX_{K}^{\circ} (cf. [10, Proposition 3.2.10]) and applying it to [25, Proposition 3.3.1], there is a thin subset ZV(K)Z\subset V(K) such that for all xV(K)Zx\in V(K)\setminus Z, the scheme theoretic inverse image π1(x)\pi^{-1}(x) is integral, i.e. it is of the form Spec(field)\operatorname{Spec}\text{(field)}.

Let Wi=π(UiXK(Kpi))W_{i}=\pi(U_{i}\cap X_{K}^{\circ}(K_{p_{i}})), which is a non-empty open subset of V(Kpi)V(K_{p_{i}}).

Claim 2.6.

The set

(2.7) (V(K)Z)i=1mWi\displaystyle(V(K)\setminus Z)\cap\bigcap_{i=1}^{m}W_{i}

is Zariski dense in VV.

Proof of Claim 2.6.

Suppose it is contained in a proper Zariski closed subset CVC\subset V. Let ψ:V(K)i=1mV(Kpi),x(x,,x)\psi\colon V(K)\longrightarrow\prod_{i=1}^{m}V(K_{p_{i}}),x\mapsto(x,\dots,x) be the diagonal embedding. Then we have

(2.8) ψ1(1im(VC)(Kpi)1imWi)(V(K)Z)=.\displaystyle\psi^{-1}\bigg{(}\prod_{1\leq i\leq m}(V\setminus C)(K_{p_{i}})\cap\prod_{1\leq i\leq m}W_{i}\bigg{)}\cap(V(K)\setminus Z)=\emptyset.

Since WiW_{i} are Zariski dense in VKpiV_{K_{p_{i}}},

(2.9) 1im(VC)(Kpi)1imWi\displaystyle\prod_{1\leq i\leq m}(V\setminus C)(K_{p_{i}})\cap\prod_{1\leq i\leq m}W_{i}

is a non-empty open subset of i=1mV(Kpi)\prod_{i=1}^{m}V(K_{p_{i}}). But by the same proof of [29, Lemma 3.11], ψ(V(K)Z)\psi(V(K)\setminus Z) is dense in i=1mV(Kpi)\prod_{i=1}^{m}V(K_{p_{i}}). Thus we get a contradiction. ∎

Let x(V(K)Z)i=1mWix\in(V(K)\setminus Z)\cap\bigcap_{i=1}^{m}W_{i}. Then π1(x)=SpecL\pi^{-1}(x)=\operatorname{Spec}L for some finite field extension LL of KK. Note that [L:K]degπ[L:K]\leq\deg\pi. Fixing a field embedding L¯L\to\overline{\mathbb{Q}} over KK and we get a point zXK(¯)XK(¯)z\in X_{K}^{\circ}(\overline{\mathbb{Q}})\subset X_{K}(\overline{\mathbb{Q}}):

(2.10) XK{X_{K}^{\circ}}SpecL{\operatorname{Spec}L}Spec¯{\operatorname{Spec}\overline{\mathbb{Q}}}V{V}SpecK.{\operatorname{Spec}K.}π\scriptstyle{\pi}x\scriptstyle{x}

Since xWi=π(UiXK(Kpi))x\in W_{i}=\pi(U_{i}\cap X_{K}^{\circ}(K_{p_{i}})), there is yiUiXK(Kpi)y_{i}\in U_{i}\cap X_{K}^{\circ}(K_{p_{i}}) such that π(yi)=x\pi(y_{i})=x. Then we get the following diagram

(2.11) SpecKpi{\operatorname{Spec}K_{p_{i}}}Specpτi{\operatorname{Spec}{\mathbb{C}}_{p_{\tau_{i}}}}XK{X_{K}^{\circ}}SpecL{\operatorname{Spec}L}Spec¯{\operatorname{Spec}\overline{\mathbb{Q}}}V{V}SpecK{\operatorname{Spec}K}yi\scriptstyle{y_{i}}π\scriptstyle{\pi}x\scriptstyle{x}

where the dashed arrow is induced by extending LKpipτiL\longrightarrow K_{p_{i}}\hookrightarrow{\mathbb{C}}_{p_{\tau_{i}}} to ¯pτi\overline{\mathbb{Q}}\longrightarrow{\mathbb{C}}_{p_{\tau_{i}}}. This embedding restricted on KK agrees with τi\tau_{i}. This means zXK(τi,UiXK(pτi))XK(τi,Ui)z\in X_{K}^{\circ}(\tau_{i},U_{i}\cap X_{K}^{\circ}({\mathbb{C}}_{p_{\tau_{i}}}))\subset X_{K}(\tau_{i},U_{i}). Therefore we proved zAXK(degπ)z\in A\cap X_{K}(\deg\pi). Since xx is arbitrary element of (V(K)Z)i=1mWi(V(K)\setminus Z)\cap\bigcap_{i=1}^{m}W_{i}, which is Zariski dense in VV, these zz’s are Zariski dense in XX and we are done.

Proof of Proposition 2.3.

We may assume AA is a general adelic subset, i.e. there is a flat morphism π:YX\pi\colon Y\longrightarrow X from a reduced algebraic scheme over ¯\overline{\mathbb{Q}} and a basic adelic subset BY(¯)B\subset Y(\overline{\mathbb{Q}}) such that A=π(B)A=\pi(B). By replacing YY with a small open affine subscheme of an irreducible component intersecting with BB, we may assume YY is a quasi-projective variety. Let K¯K\subset\overline{\mathbb{Q}} be a number field such that X,YX,Y, and π\pi are defined over KK. Let πK:YKXK\pi_{K}\colon Y_{K}\longrightarrow X_{K} be their model. Now suppose AA is height bounded. Then for all d1d\in{\mathbb{Z}}_{\geq 1}, AXK(d)A\cap X_{K}(d) are finite sets because of Northcott’s theorem. Since BYK(d)π1(AXK(d))B\cap Y_{K}(d)\subset\pi^{-1}(A\cap X_{K}(d)), π\pi is flat, and dimX1\dim X\geq 1, BYK(d)B\cap Y_{K}(d) is not Zariski dense in YY for all d1d\in{\mathbb{Z}}_{\geq 1}. This contradicts to Lemma 2.5. ∎

Remark 2.7.

The proof also shows the following. Let XX be a quasi-projective variety over ¯\overline{\mathbb{Q}} and let AX(¯)A\subset X(\overline{\mathbb{Q}}) be a non-empty adelic open subset. Let K¯K\subset\overline{\mathbb{Q}} be a number field and XKX_{K} a model of XX over KK. Then there is d1d\in{\mathbb{Z}}_{\geq 1} such that AXK(d)A\cap X_{K}(d) is Zariski dense in XX.

3. Arithmetic degree can be arbitrary close to dynamical degree

In this section, we prove Theorem 1.2. We show the following stronger statement.

Theorem 3.1.

Let XX be a quasi-projective variety over ¯\overline{\mathbb{Q}}. Let f:XXf\colon X\dashrightarrow X be a dominant rational map defined over ¯\overline{\mathbb{Q}}. Then for any ε>0\varepsilon>0, the set

(3.1) {xXf(¯)|αf(x) exists and αf(x)λ1(f)ε }\displaystyle\left\{x\in X_{f}(\overline{\mathbb{Q}})\ \middle|\ \text{$\alpha_{f}(x)$ exists and $\alpha_{f}(x)\geq\lambda_{1}(f)-\varepsilon$ }\right\}

is dense in X(¯)X(\overline{\mathbb{Q}}) with respect to the adelic topology.

Proof.

By replacing XX with its smooth locus, we may assume XX is smooth. Let us take a projective closure ι:XX\iota\colon X\hookrightarrow X^{\prime}, i.e. XX^{\prime} is a projective variety over ¯\overline{\mathbb{Q}} and ι\iota is an open immersion. By replacing XX^{\prime} with its normalization, we may assume XX^{\prime} is normal. Let LL be a very ample divisor on XX^{\prime}. We take LL so that the embedding X¯NX^{\prime}\hookrightarrow{\mathbb{P}}^{N}_{\overline{\mathbb{Q}}} by the complete linear system |L||L| is not an isomorphism. We regard ff as a dominant rational self-map on XX^{\prime}. Let us write λi=λi(f)\lambda_{i}=\lambda_{i}(f) and μi=μi(f)\mu_{i}=\mu_{i}(f). To prove the theorem, we may assume λ1>1\lambda_{1}>1. Take p{1,,dimX}p\in\{1,\dots,\dim X\} such that

(3.2) μ1==μp>μp+1.\displaystyle\mu_{1}=\cdots=\mu_{p}>\mu_{p+1}.

Let ε>0\varepsilon>0 be arbitrary positive number. Let AX(¯)A\subset X(\overline{\mathbb{Q}}) be an arbitrary non-empty adelic open subset. We will construct a point xXf(¯)Ax\in X_{f}(\overline{\mathbb{Q}})\cap A such that αf(x)λ1ε\alpha_{f}(x)\geq\lambda_{1}-\varepsilon.

Take ζ(0,1)\zeta\in(0,1), which is close to 11, such that

(3.3) μp+1ζ3μp<1,ζ2μp>1,ζ2λ1λ1ε.\displaystyle\frac{\mu_{p+1}}{\zeta^{3}\mu_{p}}<1,\quad\zeta^{2}\mu_{p}>1,\quad\zeta^{2}\lambda_{1}\geq\lambda_{1}-\varepsilon.

By [30, Remark 3.7], there is mζ1m_{\zeta}\geq 1 such that for all mmζm\geq m_{\zeta},

(3.4) (f2m)L+(μpμp+1)mL(ζμp)m(fm)L\displaystyle(f^{2m})^{*}L+(\mu_{p}\mu_{p+1})^{m}L-(\zeta\mu_{p})^{m}(f^{m})^{*}L

is big as elements of Pic~(X)\widetilde{\operatorname{Pic}}(X^{\prime})_{{\mathbb{R}}}. Here Pic~(X)\widetilde{\operatorname{Pic}}(X^{\prime})_{{\mathbb{R}}} is the colimit of Pic(X′′)\operatorname{Pic}(X^{\prime\prime})_{{\mathbb{R}}} where X′′X^{\prime\prime} runs over birational models of XX^{\prime}. See [30] for the detail. We fix an mmζm\geq m_{\zeta} so that

(3.5) ζ2mμpm+ζ2mμp+1mζmμpm\displaystyle\zeta^{2m}\mu_{p}^{m}+\zeta^{-2m}\mu_{p+1}^{m}\leq\zeta^{m}\mu_{p}^{m}

holds. Such mm exists because of eq. 3.3. Let us fix a Weil height function hL:X(¯)h_{L}\colon X^{\prime}(\overline{\mathbb{Q}})\longrightarrow{\mathbb{R}} associated with LL. We choose hLh_{L} so that hL1h_{L}\geq 1. By eq. 3.4, there are cc\in{\mathbb{R}} and a Zariski open dense subset VXV\subset X such that

(3.6) VIfm=VIf2m=\displaystyle V\cap I_{f^{m}}=V\cap I_{f^{2m}}=\emptyset
(3.7) hLf2m+(μpμp+1)mhL(ζμp)mhLfmcon V(¯).\displaystyle h_{L}\circ f^{2m}+(\mu_{p}\mu_{p+1})^{m}h_{L}-(\zeta\mu_{p})^{m}h_{L}\circ f^{m}\geq c\quad\text{on $V(\overline{\mathbb{Q}})$.}

Then by eq. 3.5, we have

(3.8) hLf2m+(ζ2mμpm)(ζ2mμp+1m)hL(ζ2mμpm+ζ2mμp+1m)hLfmc\displaystyle h_{L}\circ f^{2m}+(\zeta^{2m}\mu_{p}^{m})(\zeta^{-2m}\mu_{p+1}^{m})h_{L}-(\zeta^{2m}\mu_{p}^{m}+\zeta^{-2m}\mu_{p+1}^{m})h_{L}\circ f^{m}\geq c

or, equivalently

(3.9) hLf2mζ2mμp+1mhLfmζ2mμpm(hLfmζ2mμp+1mhL)+c\displaystyle h_{L}\circ f^{2m}-\zeta^{-2m}\mu_{p+1}^{m}h_{L}\circ f^{m}\geq\zeta^{2m}\mu_{p}^{m}(h_{L}\circ f^{m}-\zeta^{-2m}\mu_{p+1}^{m}h_{L})+c

on V(¯)V(\overline{\mathbb{Q}}). If we take c1c_{1}\in{\mathbb{R}} so that c1ζ2mμpmc1=cc_{1}-\zeta^{2m}\mu_{p}^{m}c_{1}=c, then we have

(3.10) hLf2mζ2mμp+1mhLfmc1\displaystyle h_{L}\circ f^{2m}-\zeta^{-2m}\mu_{p+1}^{m}h_{L}\circ f^{m}-c_{1}
(3.11) ζ2mμpm(hLfmζ2mμp+1mhLc1)\displaystyle\geq\zeta^{2m}\mu_{p}^{m}(h_{L}\circ f^{m}-\zeta^{-2m}\mu_{p+1}^{m}h_{L}-c_{1})

on V(¯)V(\overline{\mathbb{Q}}). This recursive inequality almost shows that the arithmetic degree is at least ζ2mμpm\zeta^{2m}\mu_{p}^{m}. What we need to show is that there is at least one initial point at which hLfmζ2mμp+1mhLc1h_{L}\circ f^{m}-\zeta^{-2m}\mu_{p+1}^{m}h_{L}-c_{1} is strictly positive. We will find such point on a curve whose forward iterates by fmf^{m} have maximal degree growth. But we first need to guarantee that there are plenty of points whose orbits are well-defined and have nice properties.

By [29, Proposition 3.24, Proposition 3.27] (cf. proof of [21, Proposition 3.2]), there is a non-empty adelic open subset AV(¯)A^{\prime}\subset V(\overline{\mathbb{Q}}) such that for all xAx\in A^{\prime}, we have

(3.12) xXf(¯),#Of(x)=,Of(x)V,and\displaystyle x\in X^{\prime}_{f}(\overline{\mathbb{Q}}),\ \#O_{f}(x)=\infty,\ O_{f}(x)\subset V,\ \text{and}
(3.13) (X,f,x)(X^{\prime},f,x) has DML property.

Here the last condition means that for any closed set WXW\subset X^{\prime}, the return set {n0fn(x)W}\{n\geq 0\mid f^{n}(x)\in W\} is a finite union of arithmetic progressions.

Now set g=fmg=f^{m}. Let d=dimXd=\dim X. Since λ1(g)=λ1(f)m=μ1m=μpm\lambda_{1}(g)=\lambda_{1}(f)^{m}=\mu_{1}^{m}=\mu_{p}^{m}, we have

(3.14) limn((gn)LLd1)1n=μpm.\displaystyle\lim_{n\to\infty}\big{(}(g^{n})^{*}L\cdot L^{d-1}\big{)}^{\frac{1}{n}}=\mu_{p}^{m}.

(Here (gn)L(g^{n})^{*}L is the one defined as an element of Pic~(X)\widetilde{\operatorname{Pic}}(X^{\prime})_{{\mathbb{R}}}. So is the intersection number. ) We choose η(0,1)\eta\in(0,1) close to 11 and l1l\in{\mathbb{Z}}_{\geq 1} large enough so that

(3.15) ((gl)LLd1)(ημpm)l\displaystyle\big{(}(g^{l})^{*}L\cdot L^{d-1}\big{)}\geq(\eta\mu_{p}^{m})^{l}
(3.16) ημpm>ζ2mμp+1m\displaystyle\eta\mu_{p}^{m}>\zeta^{-2m}\mu_{p+1}^{m}
(3.17) ημpm(2(Ld))1l>ζ2mμp+1m.\displaystyle\frac{\eta\mu_{p}^{m}}{(2(L^{d}))^{\frac{1}{l}}}>\zeta^{-2m}\mu_{p+1}^{m}.

We can choose such η\eta and ll because of eq. 3.3.

Let us pick a point aAAa\in A\cap A^{\prime} such that

  • XX^{\prime} is smooth at aa;

  • the projection from aa, pa:¯N¯N1p_{a}\colon{\mathbb{P}}^{N}_{\overline{\mathbb{Q}}}\dashrightarrow{\mathbb{P}}^{N-1}_{\overline{\mathbb{Q}}}, is generically finite on X{a}X^{\prime}\setminus\{a\}.

(Recall we chose LL so that the embedding X¯NX^{\prime}\hookrightarrow{\mathbb{P}}^{N}_{\overline{\mathbb{Q}}} defined by |L||L| is not an isomorphism. In particular, XX^{\prime} is not contained in a hyperplane of ¯N{\mathbb{P}}^{N}_{\overline{\mathbb{Q}}} and thus projection from a general point of XX^{\prime} is generically finite on XX^{\prime}. Such aa exists because a non-empty adelic open set is Zariski dense.) By the choice of AA^{\prime}, we have aIgla\notin I_{g^{l}}. Note also that codimIgl2\operatorname{codim}I_{g^{l}}\geq 2 since XX^{\prime} is normal projective. Let Γ|L|\Gamma\subset|L| be the sub-linear system consisting of all hypersurfaces passing through aa. If dimX2\dim X^{\prime}\geq 2, by Lemma 3.3, there is H1ΓH_{1}\in\Gamma such that H1H_{1} is irreducible and reduced, smooth at aa, dimH1Igl<dimIgl\dim H_{1}\cap I_{g^{l}}<\dim I_{g^{l}}, and pap_{a} is generically finite on H1{a}H_{1}\setminus\{a\}. If dimH12\dim H_{1}\geq 2, apply the same argument to the restriction of Γ\Gamma to H1H_{1} and get H2ΓH_{2}\in\Gamma such that H1H2H_{1}\cap H_{2} is irreducible and reduced, smooth at aa, dimH1H2Igl<dimIglH1\dim H_{1}\cap H_{2}\cap I_{g^{l}}<\dim I_{g^{l}}\cap H_{1}, and pap_{a} is generically finite on H1H2{a}H_{1}\cap H_{2}\setminus\{a\}. Repeat this and we get H1,,Hd1|L|H_{1},\dots,H_{d-1}\in|L| passing through aa such that

(3.18) C:=H1Hd1C:=H_{1}\cap\cdots\cap H_{d-1} is an irreducible and reduced curve;
(3.19) CIgl=.\displaystyle C\cap I_{g^{l}}=\emptyset.

Moreover, the local equations of H1,,Hd1H_{1},\dots,H_{d-1} form a regular sequence at each point of CC.

Let us consider

(3.20) Γgl{\Gamma_{g^{l}}}X{X^{\prime}}X{X^{\prime}}π\scriptstyle{\pi}G\scriptstyle{G}gl\scriptstyle{g^{l}}

where Γgl\Gamma_{g^{l}} is the graph of the rational map glg^{l}. Then we have

(3.21) ((gl)LLd1)=(GLπLd1)=(GLπ1(C))=deg(gl|C)L.\displaystyle\big{(}(g^{l})^{*}L\cdot L^{d-1}\big{)}=(G^{*}L\cdot\pi^{*}L^{d-1})=(G^{*}L\cdot\pi^{-1}(C))=\deg(g^{l}|_{C})^{*}L.

(Here for the second equality, we use the equality of schemes πH1πHd1=π1(C)\pi^{*}H_{1}\cap\cdots\cap\pi^{*}H_{d-1}=\pi^{-1}(C) to see that the cycle class c1(πL)d1[Γgl]c_{1}(\pi^{*}L)^{d-1}\cap[\Gamma_{g^{l}}] is represented by the cycle [π1(C)][\pi^{-1}(C)]. )

Thus we get (use [11, Theorem B.5.9] on the normalization of CC)

(3.22) hLgl|C\displaystyle h_{L}\circ g^{l}|_{C} =h(gl|C)L+O(1)=deg(gl|C)LdegL|ChL|C+O(hL|C)\displaystyle=h_{(g^{l}|_{C})^{*}L}+O(1)=\frac{\deg(g^{l}|_{C})^{*}L}{\deg L|_{C}}h_{L|_{C}}+O\big{(}\sqrt{h_{L}|_{C}}\big{)}
(3.23) (ημpm)l(Ld)hL|CchL|C\displaystyle\geq\frac{(\eta\mu_{p}^{m})^{l}}{(L^{d})}h_{L}|_{C}-c^{\prime}\sqrt{h_{L}|_{C}}

on C(¯)C^{\circ}(\overline{\mathbb{Q}}) where CC^{\circ} is the normal locus of CC and cc^{\prime} is a constant depends on C,(gl|C)L,L|C,hLC,(g^{l}|_{C})^{*}L,L|_{C},h_{L}.

By the construction of CC (namely aCa\in C), C(¯)AAC^{\circ}(\overline{\mathbb{Q}})\cap A\cap A^{\prime} is a non-empty adelic open set of C(¯)C(\overline{\mathbb{Q}}). By Proposition 2.3, there is a point xC(¯)AAx\in C^{\circ}(\overline{\mathbb{Q}})\cap A\cap A^{\prime} such that

(3.24) chL(x)(ημpm)l2(Ld);\displaystyle\frac{c^{\prime}}{\sqrt{h_{L}(x)}}\leq\frac{(\eta\mu_{p}^{m})^{l}}{2(L^{d})};
(3.25) hL(gi(x))Mi=0,,l,\displaystyle h_{L}(g^{i}(x))\geq M\quad i=0,\dots,l,

where M1M\in{\mathbb{R}}_{\geq 1} is a large constant that we choose below. Then we have

(3.26) hL(gl(x))(ημpm)l2(Ld)hL(x).\displaystyle h_{L}(g^{l}(x))\geq\frac{(\eta\mu_{p}^{m})^{l}}{2(L^{d})}h_{L}(x).

Thus there is i{0,,l1}i\in\{0,\dots,l-1\} such that

(3.27) hL(gi+1(x))ημpm(2(Ld))1/lhL(gi(x)).\displaystyle h_{L}(g^{i+1}(x))\geq\frac{\eta\mu_{p}^{m}}{(2(L^{d}))^{1/l}}h_{L}(g^{i}(x)).

Since xAx\in A^{\prime}, gi(x)=fmi(x)Vg^{i}(x)=f^{mi}(x)\in V. We have

(3.28) (hLfmζ2mμp+1mhLc1)(gi(x))\displaystyle(h_{L}\circ f^{m}-\zeta^{-2m}\mu_{p+1}^{m}h_{L}-c_{1})(g^{i}(x))
(3.29) =hL(gi+1(x))ζ2mμp+1mhL(gi(x))c1\displaystyle=h_{L}(g^{i+1}(x))-\zeta^{-2m}\mu_{p+1}^{m}h_{L}(g^{i}(x))-c_{1}
(3.30) (ημpm(2(Ld))1/lζ2mμp+1m)hL(gi(x))c1\displaystyle\geq\bigg{(}\frac{\eta\mu_{p}^{m}}{(2(L^{d}))^{1/l}}-\zeta^{-2m}\mu_{p+1}^{m}\bigg{)}h_{L}(g^{i}(x))-c_{1}
(3.31) (ημpm(2(Ld))1/lζ2mμp+1m)Mc1.\displaystyle\geq\bigg{(}\frac{\eta\mu_{p}^{m}}{(2(L^{d}))^{1/l}}-\zeta^{-2m}\mu_{p+1}^{m}\bigg{)}M-c_{1}.

If we chose MM so that this quantity is strictly positive, then we get

(3.32) (hLfmζ2mμp+1mhLc1)(gi(x))>0.\displaystyle(h_{L}\circ f^{m}-\zeta^{-2m}\mu_{p+1}^{m}h_{L}-c_{1})(g^{i}(x))>0.

Since xAx\in A^{\prime}, we have gn(x)=fmn(x)V(¯)g^{n}(x)=f^{mn}(x)\in V(\overline{\mathbb{Q}}). Thus by eq. 3.10, we get

(3.33) hL(gi+n(x))ζ2mμp+1mhL(gi+n1(x))c1\displaystyle h_{L}(g^{i+n}(x))-\zeta^{-2m}\mu_{p+1}^{m}h_{L}(g^{i+n-1}(x))-c_{1}
(3.34) (ζ2mμpm)n1(hL(gi+1(x))ζ2mμp+1mhL(gi(x))c1)\displaystyle\geq(\zeta^{2m}\mu_{p}^{m})^{n-1}(h_{L}(g^{i+1}(x))-\zeta^{-2m}\mu_{p+1}^{m}h_{L}(g^{i}(x))-c_{1})

for n1n\geq 1 and thus

(3.35) α¯g(gi(x))=lim infnhL(gi+n(x))1nζ2mμpm.\displaystyle\underline{\alpha}_{g}(g^{i}(x))=\liminf_{n\to\infty}h_{L}(g^{i+n}(x))^{\frac{1}{n}}\geq\zeta^{2m}\mu_{p}^{m}.

By [21, Lemma 2.7],

(3.36) α¯f(x)=α¯f(fmi(x))=α¯g(gi(x))1mζ2μp=ζ2λ1λ1ε.\displaystyle\underline{\alpha}_{f}(x)=\underline{\alpha}_{f}(f^{mi}(x))=\underline{\alpha}_{g}(g^{i}(x))^{\frac{1}{m}}\geq\zeta^{2}\mu_{p}=\zeta^{2}\lambda_{1}\geq\lambda_{1}-\varepsilon.

Since (X,f,x)(X^{\prime},f,x) satisfies DML property, by [19], the arithmetic degree αf(x)\alpha_{f}(x) exists, i.e. αf(x)=α¯f(x)\alpha_{f}(x)=\underline{\alpha}_{f}(x). Thus we are done. ∎

Remark 3.2.

In the setting of Theorem 3.1, the set

(3.37) {xXf(¯)| αf(x) exists, αf(x)-λ1(f)ε, (X,f,x) satisfies DML property }\displaystyle\left\{x\in X_{f}(\overline{\mathbb{Q}})\ \middle|\ \parbox{150.00023pt}{$\alpha_{f}(x)$ exists, $\alpha_{f}(x)\geq\lambda_{1}(f)-\varepsilon$, \\ $(X,f,x)$ satisfies DML property }\right\}

is dense in X(¯)X(\overline{\mathbb{Q}}) with respect to the adelic topology. This follows directly from the proof or [29, Proposition 3.27] and Theorem 3.1.

Lemma 3.3.

Let XX be a projective variety of dimension 2\geq 2 over an algebraically closed field of characteristic zero. Let LL be a very ample line bundle, aXa\in X a smooth closed point, and WXW\subset X a closed subset such that aWa\notin W. Suppose Γ|L|\Gamma\subset|L| is a sub-linear system of the complete linear system |L||L| consisting of hypersurfaces passing through aa. If

  • at least one member of Γ\Gamma does not contain any of the irreducible component of WW;

  • the base locus of Γ\Gamma is {a}\{a\};

  • at least one member of Γ\Gamma is smooth at aa;

  • the rational map XnX\dashrightarrow{\mathbb{P}}^{n} defined by Γ\Gamma is generically finite,

then a general member HΓH\in\Gamma satisfies:

  1. (1)

    HH is irreducible and reduced;

  2. (2)

    any irreducible component is not contained in HH;

  3. (3)

    HH is smooth at aa.

Proof.

First note that since containing an irreducible component of WW is a closed condition, general member of Γ\Gamma does not contain any irreducible component of WW.

Next, note that the restriction of Γ\Gamma to X{a}X\setminus\{a\} has no base point. By [9, Corollary 3.4.9], for a general member HΓH\in\Gamma, we have H{a}H\setminus\{a\} is reduced. Moreover, since being singular at aa is a closed condition, for a general HΓH\in\Gamma, HH is smooth at aa. In particular, general HΓH\in\Gamma is reduced.

Finally, since dimX2\dim X\geq 2, Γ\Gamma satisfies the assumption of [9, Theorem 3.4.10], and thus general member of Γ\Gamma is irreducible. ∎

Question 3.4.

Is it possible to remove ε\varepsilon from the statement of Theorem 3.1? That is, are there always points xXf(¯)x\in X_{f}(\overline{\mathbb{Q}}) such that αf(x)=λ1(f)\alpha_{f}(x)=\lambda_{1}(f)?

If there is a family of rational maps

(3.38) X{X}X{X}Y{Y}f\scriptstyle{f}

such that λ1(f)\lambda_{1}(f) is strictly larger than any of λ1(f|Xy)\lambda_{1}(f|_{X_{y}}), where XyX_{y} is the fiber over yY(¯)y\in Y(\overline{{\mathbb{Q}}}), then the answer to the above question is no. But we do not know if such rational map exists or not for now.

4. Zariski dense orbit conjecture for birational maps under certain conditions

In this section, we prove Theorem 1.6. We prove the following stronger statement.

Theorem 4.1.

Let XX be a projective variety over ¯\overline{\mathbb{Q}}. Let f:XXf\colon X\dashrightarrow X be a birational map. If λ3(f)<λ1(f)\lambda_{3}(f)<\lambda_{1}(f), then Zariski dense orbit conjecture holds for ff. That is, if ff does not admit invariant non-constant rational functions, then the set

(4.1) {xXf(¯)|Of(x) is Zariski dense in X}\displaystyle\left\{x\in X_{f}(\overline{\mathbb{Q}})\ \middle|\ \text{$O_{f}(x)$ is Zariski dense in $X$}\right\}

is dense in X(¯)X(\overline{\mathbb{Q}}) with respect to the adelic topology.

Proof.

Let us take non-empty Zariski open subsets U,VXU,V\subset X such that U,VU,V are smooth and ff induces an isomorphism UVU\xrightarrow{\sim}V:

(4.2) X{X}X{X}U{U}V{V}f\scriptstyle{f}{\subset}\scriptstyle{\sim}{\subset}

Let us consider the induced dominant rational self-map g:UVUVg\colon U\cap V\dashrightarrow U\cap V.

Suppose ff does not admit invariant non-constant rational function. Then gg does not also admit invariant non-constant rational function. By [3, Corollary 1.3] or [6, Theorem B], there are only finitely many totally invariant hypersurfaces of gg. Here a hypersurface means closed subset of pure codimension one, and a closed subset HUVH\subset U\cap V is said to be totally invariant under gg if

(4.3) H={Hg|UVIg1(H)¯irreducible component|H dominates an irreducible component of H via g}\displaystyle H=\bigcup\left\{H^{\prime}\subset\overline{g|_{U\cap V\setminus I_{g}}^{-1}(H)}\ \ \parbox{50.00008pt}{irreducible component}\ \middle|\ \parbox{100.00015pt}{$H^{\prime}$ dominates an irreducible component of $H$ via $g$}\right\}

as sets. Let HUVH\subset U\cap V be the union of all of such invariant hypersurfaces.

By Theorem 3.1, for any given non-empty adelic open subset AX(¯)A\subset X(\overline{\mathbb{Q}}), there is a point x(UV)g(¯)AHx\in(U\cap V)_{g}(\overline{\mathbb{Q}})\cap A\setminus H such that αg(x)>λ3(g)=λ3(f)\alpha_{g}(x)>\lambda_{3}(g)=\lambda_{3}(f). We prove Og(x)O_{g}(x) is Zariski dense. Let Z=Og(x)¯Z=\overline{O_{g}(x)} be the Zariski closure in UVU\cap V and suppose ZUVZ\neq U\cap V. Since Og(x)Ig=O_{g}(x)\cap I_{g}=\emptyset, ZIgZ\setminus I_{g} is dense in ZZ and g(ZIg)Zg(Z\setminus I_{g})\subset Z. As g|UVIg:UVIgUVg|_{U\cap V\setminus I_{g}}\colon U\cap V\setminus I_{g}\longrightarrow U\cap V is an open immersion, gg acts on the set of generic points of ZZ transitively. Thus ZZ is pure dimensional and totally invariant under gg. Since xHx\notin H, we have ZHZ\not\subset H and thus dimZd2\dim Z\leq d-2, where d=dimXd=\dim X.

Let us fix an irreducible component WZW\subset Z of ZZ containing xx and take m1m\geq 1 such that gm(WIgm)Wg^{m}(W\setminus I_{g^{m}})\subset W. Note that since αg(x)>λ3(g)1\alpha_{g}(x)>\lambda_{3}(g)\geq 1, dimW>0\dim W>0. Also gmg^{m} is isomorphic at the generic point of WW. Thus by [21, Lemma 2.3], we have

(4.4) λ1(gm|W)λ1+codimW(gm).\displaystyle\lambda_{1}(g^{m}|_{W})\leq\lambda_{1+\operatorname{codim}W}(g^{m}).

Since codimW2\operatorname{codim}W\geq 2 and λ1(gm)=λ1(g)m>λ3(g)m=λ3(gm)\lambda_{1}(g^{m})=\lambda_{1}(g)^{m}>\lambda_{3}(g)^{m}=\lambda_{3}(g^{m}), by the log concavity of dynamical degrees, we have λ1+codimW(gm)λ3(gm)\lambda_{1+\operatorname{codim}W}(g^{m})\leq\lambda_{3}(g^{m}). Thus we get

(4.5) λ1(gm|W)λ3(gm)<λ1(gm).\displaystyle\lambda_{1}(g^{m}|_{W})\leq\lambda_{3}(g^{m})<\lambda_{1}(g^{m}).

Then we get

(4.6) αg(x)m=αgm(x)=αgm|W(x)λ1(gm|W)λ3(gm)=λ3(g)m,\displaystyle\alpha_{g}(x)^{m}=\alpha_{g^{m}}(x)=\alpha_{g^{m}|_{W}}(x)\leq\lambda_{1}(g^{m}|_{W})\leq\lambda_{3}(g^{m})=\lambda_{3}(g)^{m},

where the first inequality follows from [12, Proposition 3.11]. This inequality contradicts to the choice of xx. ∎

Remark 4.2.

Under the assumption of Theorem 4.1, when ff does not admit invariant non-constant rational function, the proof actually shows the following: for any ε>0\varepsilon>0, the set

(4.7) {xXf(¯)|Of(x) is Zariski dense in X and αf(x)λ1(f)ε}\displaystyle\left\{x\in X_{f}(\overline{\mathbb{Q}})\ \middle|\ \text{$O_{f}(x)$ is Zariski dense in $X$ and $\alpha_{f}(x)\geq\lambda_{1}(f)-\varepsilon$}\right\}

is dense in X(¯)X(\overline{\mathbb{Q}}) with respect to the adelic topology. In particular, for any ε>0\varepsilon>0, there are xXf(¯)x\in X_{f}(\overline{\mathbb{Q}}) such that (X,f,x)(X,f,x) satisfies DML property, Of(x)O_{f}(x) is Zariski dense in XX, and αf(x)>λ1(f)ε\alpha_{f}(x)>\lambda_{1}(f)-\varepsilon (cf.  [29, Proposition 3.27]). In this case, the orbit Of(x)O_{f}(x) is generic. By [19, Theorem 2.2], αf(x)\alpha_{f}(x) can take only the values from {λ1(f)=μ1(f),μ2(f),1}\{\lambda_{1}(f)=\mu_{1}(f),\mu_{2}(f),1\}. Thus if we take ε\varepsilon small enough, our point xx satisfies αf(x)=λ1(f)\alpha_{f}(x)=\lambda_{1}(f).

Remark 4.3.

Long Wang pointed out us that Remark 4.2 and [4] give us an example of birational map with a ¯\overline{\mathbb{Q}}-point whose arithmetic degree is a transcendental number. Indeed, by [4], there is a birational map f:¯3¯3f\colon{\mathbb{P}}^{3}_{\overline{\mathbb{Q}}}\dashrightarrow{\mathbb{P}}^{3}_{\overline{\mathbb{Q}}} whose first dynamical degree is a transcendental number. This map ff does not admit non-constant rational function. Indeed, if it is the case, the first dynamical degree of ff is equal to the first relative dynamical degree with respect to a non-constant rational map to a curve, which is equal to the first dynamical degree of a very general fiber (take base change to {\mathbb{C}} to find such a fiber). Since the relative dimension is two, the first dynamical degree on the fiber is algebraic, as birational map on surfaces are always algebraically stable. This is a contradiction. Therefore, by Remark 4.2, there is a point x(¯3)f(¯)x\in({\mathbb{P}}^{3}_{\overline{\mathbb{Q}}})_{f}(\overline{\mathbb{Q}}) such that αf(x)=λ1(f)\alpha_{f}(x)=\lambda_{1}(f), which is a transcendental number.

We note that the existence of transcendental arithmetic degree is first proven in [21]. The map in the example was not birational, and finding an example with birational map was left as a problem. Such example was recently constructed by Sugimoto in [27]. The above argument gives another construction of such example.

Corollary 4.4.

Let XX be a projective variety over ¯\overline{\mathbb{Q}} of dimension four. Let f:XXf\colon X\dashrightarrow X be a birational map with λ1(f)λ3(f)\lambda_{1}(f)\neq\lambda_{3}(f). Then Zariski dense orbit conjecture holds for ff. More strongly, if ff does not admit invariant non-constant rational function, then the set {xXf(¯)Of(x) is Zariski dense in X}\{x\in X_{f}(\overline{\mathbb{Q}})\mid\text{$O_{f}(x)$ is Zariski dense in $X$}\} is dense in X(¯)X(\overline{\mathbb{Q}}) with respect to the adelic topology.

Proof.

If λ1(f)>λ3(f)\lambda_{1}(f)>\lambda_{3}(f), then this is exactly the same with Theorem 4.1. Suppose λ1(f)<λ3(f)\lambda_{1}(f)<\lambda_{3}(f). Since λi(f1)=λ4i(f)\lambda_{i}(f^{-1})=\lambda_{4-i}(f) for i=0,,4i=0,\dots,4, we have λ1(f1)=λ3(f)>λ1(f)=λ3(f1)\lambda_{1}(f^{-1})=\lambda_{3}(f)>\lambda_{1}(f)=\lambda_{3}(f^{-1}). Moreover, if ff does not admit invariant non-constant rational function, then neither does f1f^{-1}. Thus by Theorem 4.1, the set

(4.8) {xXf1(¯)Of1(x) is Zariski dense in X}\displaystyle\{x\in X_{f^{-1}}(\overline{\mathbb{Q}})\mid\text{$O_{f^{-1}}(x)$ is Zariski dense in $X$}\}

is dense in X(¯)X(\overline{\mathbb{Q}}) with respect to the adelic topology. As before, let us take non-empty Zariski open subsets U,VXU,V\subset X such that ff induces an isomorphism UVU\xrightarrow{\sim}V:

(4.9) X{X}X{X}U{U}V{V}f\scriptstyle{f}{\subset}\scriptstyle{\sim}{\subset}

Let us consider the induced dominant rational self-map g:UVUVg\colon U\cap V\dashrightarrow U\cap V. Then by [29, Proposition 3.27], the set

(4.10) (UV)g(¯)(UV)g1(¯)\displaystyle(U\cap V)_{g}(\overline{\mathbb{Q}})\cap(U\cap V)_{g^{-1}}(\overline{\mathbb{Q}})

contains non-empty adelic open subset of (UV)(¯)(U\cap V)(\overline{\mathbb{Q}}), hence of X(¯)X(\overline{\mathbb{Q}}). Therefore the set

(4.11) {x(UV)g(¯)(UV)g1(¯)|Og1(x) is Zariski dense in UV}\displaystyle\left\{x\in(U\cap V)_{g}(\overline{\mathbb{Q}})\cap(U\cap V)_{g^{-1}}(\overline{\mathbb{Q}})\ \middle|\ \text{$O_{g^{-1}}(x)$ is Zariski dense in $U\cap V$}\right\}

is dense in X(¯)X(\overline{\mathbb{Q}}) with respect to the adelic topology. Now we claim that any point xx in this set has the property xXf(¯)x\in X_{f}(\overline{\mathbb{Q}}) and Of(x)O_{f}(x) is Zariski dense in XX. The first property is obvious as x(UV)g(¯)Xf(¯)x\in(U\cap V)_{g}(\overline{\mathbb{Q}})\subset X_{f}(\overline{\mathbb{Q}}). By the choice of U,VU,V, we see that Og(x)(UV)g1(¯)O_{g}(x)\subset(U\cap V)_{g^{-1}}(\overline{\mathbb{Q}}). Let ZOg(x)¯Z\subset\overline{O_{g}(x)} be a top dimensional irreducible component with the property #{n0gn(x)Z}=\#\{n\in{\mathbb{Z}}_{\geq 0}\mid g^{n}(x)\in Z\}=\infty. Here the closure is taken in UVU\cap V. Then g1(ZIg1)Og(x)¯g^{-1}(Z\setminus I_{g^{-1}})\subset\overline{O_{g}(x)}. Thus g1(ZIg1)¯\overline{g^{-1}(Z\setminus I_{g^{-1}})} has the same property as ZZ. We can repeat this process and eventually end up with the original ZZ. Noticing Og(x)(UV)g1(¯)O_{g}(x)\subset(U\cap V)_{g^{-1}}(\overline{\mathbb{Q}}), we conclude

(4.12) gn(Og(x))Og(x)¯\displaystyle g^{-n}(O_{g}(x))\subset\overline{O_{g}(x)}

for all n0n\geq 0. Since Og1(x)O_{g^{-1}}(x) is Zariski dense in UVU\cap V, we have Og(x)O_{g}(x) is Zariski dense in UVU\cap V, and hence Zariski dense in XX. ∎

References

  • [1] Amerik, E., and Campana, F. Fibrations méromorphes sur certaines variétés à fibré canonique trivial. Pure Appl. Math. Q. 4, 2 (2008), 509–545.
  • [2] Bell, J., Ghioca, D., and Reichstein, Z. On a dynamical version of a theorem of Rosenlicht. Ann. Sc. Norm. Super. Pisa Cl. Sci. (5) 17, 1 (2017), 187–204.
  • [3] Bell, J., Moosa, R., and Topaz, A. Invariant hypersurfaces. J. Inst. Math. Jussieu 21, 2 (2022), 713–739.
  • [4] Bell, J. P., Diller, J., Jonsson, M., and Krieger, H. Birational maps with transcendental dynamical degree. Proc. Lond. Math. Soc. (3) 128, 1 (2024), Paper No. e12573, 47.
  • [5] Bombieri, E., and Gubler, W. Heights in Diophantine geometry, vol. 4 of New Mathematical Monographs. Cambridge University Press, Cambridge, 2006.
  • [6] Cantat, S. Invariant hypersurfaces in holomorphic dynamics. Math. Res. Lett. 17, 5 (2010), 833–841.
  • [7] Dang, N.-B. Degrees of iterates of rational maps on normal projective varieties. Proc. Lond. Math. Soc. (3) 121, 5 (2020), 1268–1310.
  • [8] Dinh, T.-C., and Sibony, N. Une borne supérieure pour l’entropie topologique d’une application rationnelle. Ann. of Math. (2) 161, 3 (2005), 1637–1644.
  • [9] Flenner, H., O’Carroll, L., and Vogel, W. Joins and intersections. Springer Monographs in Mathematics. Springer-Verlag, Berlin, 1999.
  • [10] Fu, L. Etale cohomology theory, revised ed., vol. 14 of Nankai Tracts in Mathematics. World Scientific Publishing Co. Pte. Ltd., Hackensack, NJ, 2015.
  • [11] Hindry, M., and Silverman, J. H. Diophantine geometry, vol. 201 of Graduate Texts in Mathematics. Springer-Verlag, New York, 2000. An introduction.
  • [12] Jia, J., Shibata, T., Xie, J., and Zhang, D.-Q. Endomorphisms of quasi-projective varieties–towards zariski dense orbit and kawaguchi-silverman conjectures. To appear in Math. Res. Lett. arXiv:2104.05339 (2021).
  • [13] Kawaguchi, S., and Silverman, J. H. Examples of dynamical degree equals arithmetic degree. Michigan Math. J. 63, 1 (2014), 41–63.
  • [14] Kawaguchi, S., and Silverman, J. H. Dynamical canonical heights for Jordan blocks, arithmetic degrees of orbits, and nef canonical heights on abelian varieties. Trans. Amer. Math. Soc. 368, 7 (2016), 5009–5035.
  • [15] Kawaguchi, S., and Silverman, J. H. On the dynamical and arithmetic degrees of rational self-maps of algebraic varieties. J. Reine Angew. Math. 713 (2016), 21–48.
  • [16] Lang, S. Fundamentals of Diophantine geometry. Springer-Verlag, New York, 1983.
  • [17] Matsuzawa, Y. On upper bounds of arithmetic degrees. Amer. J. Math. 142, 6 (2020), 1797–1820.
  • [18] Matsuzawa, Y. Recent advances on kawaguchi-silverman conjecture. To appear in the proceedings for the Simons Symposia on Algebraic, Complex, and Arithmetic Dynamics arXiv:2311.15489 (2023).
  • [19] Matsuzawa, Y. Existence of arithmetic degrees for generic orbits and dynamical Lang-Siegel problem. arXiv preprint arXiv:2407.03097 (2024).
  • [20] Matsuzawa, Y., Sano, K., and Shibata, T. Arithmetic degrees and dynamical degrees of endomorphisms on surfaces. Algebra Number Theory 12, 7 (2018), 1635–1657.
  • [21] Matsuzawa, Y., and Wang, L. Arithmetic degrees and zariski dense orbits of cohomologically hyperbolic maps. to appear in Trans. Amer. Math. Soc. (2022).
  • [22] Medvedev, A., and Scanlon, T. Invariant varieties for polynomial dynamical systems. Ann. of Math. (2) 179, 1 (2014), 81–177.
  • [23] Sano, K., and Shibata, T. Zariski density of points with maximal arithmetic degree for surfaces. to appear in RIMS Kôkyûroku Bessatsu.
  • [24] Sano, K., and Shibata, T. Zariski Density of Points with Maximal Arithmetic Degree. Michigan Math. J. 73, 2 (2023), 429–448.
  • [25] Serre, J.-P. Topics in Galois theory, second ed., vol. 1 of Research Notes in Mathematics. A K Peters, Ltd., Wellesley, MA, 2008. With notes by Henri Darmon.
  • [26] SILVERMAN, J. H. Dynamical degree, arithmetic entropy, and canonical heights for dominant rational self-maps of projective space. Ergodic Theory and Dynamical Systems 34, 2 (2012), 647–678.
  • [27] Sugimoto, Y. A 11-cohomologically hyperbolic birational map of 3\mathbb{P}^{3}, with a transcendental arithmetic degree. arXiv preprint arXiv:2401.09821 (2024).
  • [28] Truong, T. T. Relative dynamical degrees of correspondences over a field of arbitrary characteristic. J. Reine Angew. Math. 758 (2020), 139–182.
  • [29] Xie, J. The existence of Zariski dense orbits for endomorphisms of projective surfaces. J. Amer. Math. Soc. 1004.
  • [30] Xie, J. Algebraic dynamics and recursive inequalities. arXiv preprint arXiv:2402.12678 (2024).
  • [31] Zhang, S.-W. Distributions in algebraic dynamics. In Surveys in differential geometry. Vol. X, vol. 10 of Surv. Differ. Geom. Int. Press, Somerville, MA, 2006, pp. 381–430.