Introduction
We write P k \mathbb{P}^{k} P k for the complex projective space P k ( C ) \mathbb{P}^{k}(\mathbb{C}) P k ( C ) . Let F : P k → P k F : \mathbb{P}^{k} \to\mathbb{P}^{k} F : P k → P k be a holomorphic endomorphism of algebraic degree d ≥ 2 d \ge2 d ≥ 2 , with k ≥ 1 k \ge1 k ≥ 1 . Its topological entropy is k log d k \log d k log d [36 ] , and it admits a unique measure μ F \mu_F μ F of maximal entropy; see [33 , 40 , 41 ] when k = 1 k = 1 k = 1 , and [34 , 4 , 24 ] when k ≥ 2 k \ge2 k ≥ 2 . In every dimension, μ F \mu_F μ F is a Monge–Ampère measure [16 ] of a Hölder continuous potential; see for instance [24 ] . By [19 Theorem 1.1] , this is equivalent to μ F \mu_F μ F having a Hölder continuous super-potential. In particular, μ F \mu_F μ F is moderate and gives zero mass to pluripolar sets (and, as a consequence, to proper analytic subsets); see [20 , 19 ] . The super-potential of a measure is a functional on positive closed ( 1 , 1 ) (1,1) ( 1 , 1 ) -currents and can be identified with the integration functional on their normalized quasi-plurisubharmonic potentials; see [23 ] . The notion of super-potential generalizes to higher dimensions the classical notion of potential for measures on P 1 \mathbb{P}^{1} P 1 .
In any dimension, endomorphisms of P k \mathbb{P}^{k} P k admit large families of measures whose entropy is arbitrarily close to the maximal value k log d k \log d k log d , see for instance [30 UZ13, 6 , 7 ] . The above discussion naturally raises the question of whether sufficiently large entropy forces analogous regularity for invariant measures other than μ F \mu_F μ F . For instance, as a consequence of the variational principle and Gromov entropy bound argument [36 ] , every ergodic F F F -invariant probability measure of entropy strictly larger than ( k − 1 ) log d (k-1)\log d ( k − 1 ) log d gives zero mass to every proper analytic subset; see [24 Section 1.7] . However, global Hölder regularity cannot be expected, even for mixing measures with entropy arbitrarily close to the maximum. Our main result gives a Lusin-type conclusion: every ergodic invariant measure of entropy strictly greater than ( k − 1 ) log d (k-1)\log d ( k − 1 ) log d has Hölder continuous super-potentials after restriction to compact sets of measure arbitrarily close to one. As a consequence, every such measure also gives no mass to pluripolar subsets.
Hölder regularity on arbitrarily large sets
Let ω \omega ω be the Fubini–Study form on P k \mathbb{P}^{k} P k , normalized by ∫ P k ω k = 1 \int_{\mathbb{P}^{k}} \omega^{k} = 1 ∫ P k ω k = 1 . We write h μ ( F ) h_{\mu}(F) h μ ( F ) for the metric entropy of F F F with respect to an F F F -invariant probability measure μ \mu μ , and h t o p ( F ) h_{\mathrm{top}}(F) h top ( F ) for the topological entropy of F F F .
A quasi-plurisubharmonic function is locally the sum of a plurisubharmonic function and a smooth function, and is not identically − ∞ -\infty − ∞ . Set d d c : = i π ∂ ∂ ‾ dd^{c} := \frac{i}{\pi}\partial\overline{\partial} d d c := π i ∂ ∂ . We denote by PSH ( P k , ω ) \operatorname{PSH}(\mathbb{P}^{k},\omega) PSH ( P k , ω ) the quasi-plurisubharmonic functions u u u for which ω + d d c u \omega+dd^{c}u ω + d d c u is a positive current; these functions are often called ω \omega ω -plurisubharmonic ; see for instance [37 ] and [24 Appendix A] . For ν ∈ L 1 ( ω k ) \nu\in L^{1}(\omega^{k}) ν ∈ L 1 ( ω k ) ,
2020 Mathematics Subject Classification. Primary 37A35, 37F80; Secondary 32U05, 32U40.
Key words and phrases. Holomorphic endomorphisms, entropy, pluripolar sets, positive closed currents, inverse branches. set ∥ v ∥ 1 : = ∫ P k ∣ v ∣ ω k \lVert v\rVert_1 := \int_{\mathbb{P}^k} |v| \omega^k ∥ v ∥ 1 := ∫ P k ∣ v ∣ ω k . For a Borel set A ⊂ P k A \subset\mathbb{P}^k A ⊂ P k , let 1 A \mathbf{1}_A 1 A denote its indicator function and 1 A μ \mathbf{1}_A\mu 1 A μ the restriction of μ \mu μ to A A A . Our main result is the following.
Theorem 1.1. Let F : P k → P k F:\mathbb{P}^k \to\mathbb{P}^k F : P k → P k be a holomorphic endomorphism of algebraic degree d ≥ 2 d \ge2 d ≥ 2 , with k ≥ 1 k \ge1 k ≥ 1 , and let μ \mu μ be an ergodic F F F -invariant probability measure with h μ ( F ) > ( k − 1 ) log d h_\mu(F) > (k-1)\log d h μ ( F ) > ( k − 1 ) log d . There exists β ∈ ( 0 , 1 ) \beta\in(0,1) β ∈ ( 0 , 1 ) such that, for every 0 < θ < 1 0 < \theta< 1 0 < θ < 1 , there are a compact set E θ ⊂ P k E_\theta\subset\mathbb{P}^k E θ ⊂ P k with μ ( E θ ) > 1 − θ \mu(E_\theta) > 1-\theta μ ( E θ ) > 1 − θ and a constant C θ < ∞ C_\theta< \infty C θ < ∞ for which η θ : = 1 E θ μ \eta_\theta:= \mathbf{1}_{E_\theta}\mu η θ := 1 E θ μ integrates every quasi-plurisubharmonic function and satisfies
∫ P k ∣ u − v ∣ d η θ ≤ C θ max { ∥ u − v ∥ 1 , ∥ u − v ∥ 1 β } for every u , v ∈ PSH ( P k , ω ) . (1) \int_{\mathbb{P}^k} |u-v| \,d\eta_\theta\le C_\theta\max\left\{\lVert u-v\rVert_1,\lVert u-v\rVert_1^\beta\right\} \qquad\text{for every }u,v \in\operatorname{PSH}(\mathbb{P}^k,\omega). \tag*{(1)} ∫ P k ∣ u − v ∣ d η θ ≤ C θ max { ∥ u − v ∥ 1 , ∥ u − v ∥ 1 β } for every u , v ∈ PSH ( P k , ω ) . (1)
As mentioned above, the functional u ↦ ∫ P k u d η θ u \mapsto\int_{\mathbb{P}^k}u\,d\eta_\theta u ↦ ∫ P k u d η θ , acting on ω \omega ω -plurisubharmonic functions normalized by ∫ P k u ω k = 0 \int_{\mathbb{P}^k}u\omega^k=0 ∫ P k u ω k = 0 , represents the normalized super-potential of η θ \eta_\theta η θ under the identification u ↔ ω + d d c u u \leftrightarrow\omega+dd^cu u ↔ ω + d d c u ; see [23 ] and [19 Definition 3.4] . Estimate (1.1) makes this functional Hölder continuous with respect to the L 1 ( ω k ) L^1(\omega^k) L 1 ( ω k ) distance; see [19 Lemma 3.5] . An explicit admissible range for β \beta β is given in Remark 4.1. When k = 1 k=1 k = 1 , the entropy assumption reduces to h μ ( F ) > 0 h_\mu(F)>0 h μ ( F ) > 0 , and the conclusion follows directly from Mañé’s dimension formula [42 Theorem A] and classical potential theory; see § 4.3 .
The restrictions in Theorem 1.1 imply vanishing on pluripolar sets and the following entropy bounds. Recall that a set P ⊂ P k P\subset\mathbb{P}^k P ⊂ P k is locally pluripolar if, near each point, P P P is contained in the pole set of a plurisubharmonic function that is not identically − ∞ -\infty − ∞ on the connected neighbourhood. By the global Josefson theorem, this is equivalent to the existence of u ∈ PSH ( P k , ω ) u\in\operatorname{PSH}(\mathbb{P}^k,\omega) u ∈ PSH ( P k , ω ) such that P ⊂ { u = − ∞ } P\subset\{u=-\infty\} P ⊂ { u = − ∞ } ; see [3 ] .
Corollary 1.2. Let F : P k → P k F:\mathbb{P}^k\to\mathbb{P}^k F : P k → P k be a holomorphic endomorphism of algebraic degree d ≥ 2 d\ge2 d ≥ 2 , with k ≥ 1 k\ge1 k ≥ 1 . The following hold.
(i) Every ergodic F F F -invariant probability measure μ \mu μ with h μ ( F ) > ( k − 1 ) log d h_\mu(F)>(k-1)\log d h μ ( F ) > ( k − 1 ) log d gives zero mass to every Borel pluripolar set P ⊂ P k P\subset\mathbb{P}^k P ⊂ P k .
(ii) If an F F F -invariant probability measure ν \nu ν satisfies ν ( P ) = 1 \nu(P)=1 ν ( P ) = 1 for a Borel pluripolar set P ⊂ P k P\subset\mathbb{P}^k P ⊂ P k , then h ν ( F ) ≤ ( k − 1 ) log d h_\nu(F)\le(k-1)\log d h ν ( F ) ≤ ( k − 1 ) log d . In particular, every non-empty compact pluripolar set K K K with F ( K ) ⊂ K F(K)\subset K F ( K ) ⊂ K satisfies h t o p ( F ∣ K ) ≤ ( k − 1 ) log d h_{\mathrm{top}}(F|_K)\le(k-1)\log d h top ( F ∣ K ) ≤ ( k − 1 ) log d .
In dimension one, Corollary 1.2 is already a consequence of classical results; see also the observation following the questions in [27 Section 2] . Indeed, Mañé’s dimension formula gives positive pointwise dimension almost everywhere for every ergodic invariant measure of positive entropy, whereas polar sets have Hausdorff dimension zero [2 Theorem 5.9.6] . A related vanishing statement holds for holomorphic automorphisms of compact Kähler surfaces and, in the non-compact setting, for compactly supported ergodic invariant measures of positive entropy of polynomial automorphisms of Hénon type and invertible horizontal-like maps in complex dimension two [28 ] . These invertible maps are outside the large-topological-degree setting of Corollary 1.2. Although the conclusion follows from standard ingredients and is already known for the measures of maximal entropy in these settings [15 , 13 , 28 , 17 ] , we have not found it explicitly stated in the literature. We therefore include a short proof, combining the Ledrappier–Young theory with a pluripolarity criterion of Dujardin; see Theorem 5.1 .
The Hölder estimate (1.1) also gives exponential integrability and a Monge–Ampère representation for each restriction η θ \eta_\theta η θ . Set U : = { u ∈ PSH ( P k , ω ) : sup P k u = 0 } \mathcal{U}:=\{u\in\operatorname{PSH}(\mathbb{P}^k,\omega):\sup_{\mathbb{P}^k}u=0\} U := { u ∈ PSH ( P k , ω ) : sup P k u = 0 } . Every quasi-plurisubharmonic function can be normalized into U \mathcal{U} U by positive scaling and addition of a constant. A finite positive measure η \eta η is moderate if sup u ∈ U ∫ P k e − α u d η < ∞ \sup_{u\in\mathcal{U}}\int_{\mathbb{P}^k}e^{-\alpha u}\,d\eta<\infty sup u ∈ U ∫ P k e − α u d η < ∞ for some α > 0 \alpha>0 α > 0 ; see [20 ] .
Corollary 1.3. The restrictions η θ \eta_\theta η θ in Theorem 1.1 have the following properties.
(i) There is α > 0 \alpha>0 α > 0 , independent of θ \theta θ , such that sup u ∈ U ∫ P k e − α u d η θ < ∞ \sup_{u\in\mathcal{U}}\int_{\mathbb{P}^k}e^{-\alpha u}\,d\eta_\theta<\infty sup u ∈ U ∫ P k e − α u d η θ < ∞ for every 0 < θ < 1 0<\theta<1 0 < θ < 1 . In particular, every η θ \eta_\theta η θ is moderate.
(ii) For each θ \theta θ , there is a Hölder continuous function φ θ ∈ PSH ( P k , ω ) \varphi_\theta\in\operatorname{PSH}(\mathbb{P}^k,\omega) φ θ ∈ PSH ( P k , ω ) such that η θ = η θ ( P k ) ( ω + d d c φ θ ) k \eta_\theta= \eta_\theta(\mathbb{P}^k)(\omega+ dd^c\varphi_\theta)^k η θ = η θ ( P k ) ( ω + d d c φ θ ) k . The Hölder exponent of φ θ \varphi_\theta φ θ can be chosen independently of θ \theta θ .
The passage to the restrictions η θ \eta_\theta η θ in Theorem 1.1 and Corollary 1.3 is necessary in general: even a mixing invariant measure with entropy arbitrarily close to the maximum need not integrate every quasi-plurisubharmonic function; see Proposition 4.3.
Entropy bounds and extensions
We next bound the Bowen entropy of pluripolar subsets of uniformly expanding sets. For Z ⊂ P k Z \subset\mathbb{P}^k Z ⊂ P k , let h B ( F , Z ) h_{\mathrm{B}}(F,Z) h B ( F , Z ) denote the Bowen entropy, defined using covers by Bowen balls of variable lengths; see [32 ] and § 4.2 . We say that F F F is uniformly expanding on a compact set K K K in the ambient tangent bundle if there are constants c > 0 c > 0 c > 0 and ρ > 1 \rho> 1 ρ > 1 such that ∥ D ( F n ) x v ∥ ≥ c ρ n ∥ v ∥ \lVert D(F^n)_xv\rVert\ge c\rho^n\lVert v\rVert ∥ D ( F n ) x v ∥ ≥ c ρ n ∥ v ∥ for every x ∈ K x \in K x ∈ K , v ∈ T x P k v \in T_x\mathbb{P}^k v ∈ T x P k , and n ≥ 1 n \ge1 n ≥ 1 , with norms taken in the Fubini–Study metric.
Theorem 1.4. Let F : P k → P k F:\mathbb{P}^k \to\mathbb{P}^k F : P k → P k be a holomorphic endomorphism of algebraic degree d ≥ 2 d \ge2 d ≥ 2 , with k ≥ 1 k \ge1 k ≥ 1 . Let K ⊂ P k K \subset\mathbb{P}^k K ⊂ P k be a non-empty compact set with F ( K ) ⊂ K F(K) \subset K F ( K ) ⊂ K on which F F F is uniformly expanding in the ambient tangent bundle. For every Borel pluripolar set P ⊂ P k P \subset\mathbb{P}^k P ⊂ P k , we have h B ( F , P ∩ K ) ≤ ( k − 1 ) log d h_{\mathrm{B}}(F,P \cap K) \le(k-1)\log d h B ( F , P ∩ K ) ≤ ( k − 1 ) log d .
When P ∩ K P \cap K P ∩ K is compact and forward invariant, the bound follows from Corollary 1.2 and the variational principle for topological entropy. For a general Borel set P ∩ K P \cap K P ∩ K , however, an invariant probability measure supported on P ∩ K P \cap K P ∩ K need not exist, so the inverse branch distortion estimates for invariant measures cannot be applied directly. Our argument only requires contracting holomorphic inverse branches with suitable quantitative bounds; we assume uniform expansion for simplicity to ensure these bounds. In dimension one, the uniform expansion assumption is unnecessary. In fact, h B ( F , P ) = 0 h_{\mathrm{B}}(F,P)=0 h B ( F , P ) = 0 for every Borel polar set P ⊂ P 1 P \subset\mathbb{P}^1 P ⊂ P 1 ; see § 4.3 .
Theorems 1.1 and 1.4 and Corollaries 1.2 and 1.3 extend to polynomial-like maps in any dimension [22 ] , provided that d t > d k − 1 ∗ ( f ) d_t > d^*_{k-1}(f) d t > d k − 1 ∗ ( f ) , and to surjective holomorphic self-maps of connected compact Kähler manifolds of dimension k ≥ 1 k \ge1 k ≥ 1 , provided that λ k ( F ) > λ k − 1 ( F ) \lambda_k(F) > \lambda_{k-1}(F) λ k ( F ) > λ k − 1 ( F ) ; see Appendix A. In dimension two, together with Gromov’s entropy bound, this gives a negative answer to a question of de Thélin and Vigny [27 Section 2] . A related question of de Thélin and Vigny concerning the dimension of positive closed currents was studied, in a directional form, by Dupont and Rogue [21 ] .
Proof strategy
The key idea behind the proofs of Theorems 1.1 and 1.4 is a regularization argument, which was inspired by the work of F. Rodriguez Hertz, J. Rodriguez Hertz, and W. Wu [44 ] . They apply heat regularization to the indicator of the singular set and control the resulting loss of flow invariance through quantitative commutator estimates. We adapt this idea to measures of large entropy, using inverse branches in place of the flow, as in our earlier work on joining rigidity for rational maps [10 ] . Since integrability against quasi-plurisubharmonic functions is not known a priori for these measures, we construct regularizations and recover integrability for suitable restrictions by controlling the regularization errors.
Theorem 3.1 implements this idea through a regularization criterion that does not require invariance of the measure. We construct approximating measures by transporting radial averaging kernels along contracting holomorphic inverse branches [8 , 5 ] . To compare two consecutive approximations, we express both averages on a common coordinate domain. In the spirit of the Dinh–Sibony d d c dd^c d d c -method [22 , 24 ] , we decompose the pulled-back test function into a difference of local potentials of positive closed ( 1 , 1 ) (1,1) ( 1 , 1 ) -currents and a pluriharmonic function. The pluriharmonic term cancels, leaving a bound in terms of the masses of these currents and the densities of the averaging measures.
The entropy assumption comes into play when these local comparisons are summed, giving exponential decay of the comparison errors, after absorbing the small exponential losses from the shrinking domains. Combining this decay with an exponential density bound yields integrability and Hölder regularity on sets of arbitrarily large measure, proving Theorem 1.1. Corollaries 1.2 and 1.3 then follow from Theorem 1.1 by standard arguments and known results from pluripotential theory. For Theorem 1.4, uniform expansion provides the required inverse branches, while the dynamical Frostman lemma [32 ] gives the Bowen ball mass bounds.
Organization of the paper
Organization of the paper. Section 2 collects preliminary estimates, and Section 3 establishes the key regularization criterion. Section 4 proves Theorems 1.1 and 1.4 and Corollaries 1.2 and 1.3, gives a direct proof of Theorem 1.1 in dimension one, and discusses examples. Section 5 treats positive-entropy measures in the surface automorphism, Hénon, and horizontal-like settings. Finally, Appendix A treats the Kähler and polynomial-like extensions.
Acknowledgments
Acknowledgments. The authors would like to thank Jana Rodriguez Hertz for her talk on [44 ] at the Rigidity Theory of Diffeomorphisms and Flows Workshop at the Brin Center in October 2026, which was the inspiration for this work.
This project has received funding from the project ANR TIGerS (ANR-24-CE40-3604) and from the MIUR Excellence Department Project awarded to the Department of Mathematics of the University of Pisa, CUP I57G22000700001. The first author is affiliated to the GNSAGA group of INdAM. The second author is supported by National Science Foundation DMS-2554272 and by the Simons Foundation Travel Support for Mathematicians program.
AI Declaration
AI Declaration. The authors used ChatGPT 6 to assist with literature searches, to discuss and check mathematical arguments, and to suggest revisions to the exposition. More precisely, inspired by the talk by J. Rodriguez Hertz, the second author initially asked ChatGPT 6 whether the method used in [44 ] could be adapted to prove Corollary 1.2, using inverse branches in place of geodesic flows in a similar way as in our earlier work [10 ] . ChatGPT 6 then proposed a possible strategy based on a similar regularization procedure that required certain pluripotential-theoretic estimates. The authors subsequently developed these estimates and improved the argument to obtain the stronger Theorem 1.1.
Preliminaries
This section collects potential-theoretic estimates and uniform inverse branch and entropy estimates.
Potential-theoretic estimates
Potential-theoretic estimates. In this paper, distances, balls, and derivative norms on P k \mathbb{P}^{k} P k refer to the Fubini–Study metric; in C k \mathbb{C}^{k} C k , balls and norms are Euclidean. We write Lip ( G ) \operatorname{Lip}(G) Lip ( G ) for the Lipschitz constant of a map G G G with respect to these metrics. For a holomorphic map G : U → V G: U \to V G : U → V between open subsets of P k \mathbb{P}^{k} P k , its real Jacobian is Jac R G = ( G ∗ ω k ) / ω k \operatorname{Jac}_{\mathbb{R}} G = (G^{*}\omega^{k})/\omega^{k} Jac R G = ( G ∗ ω k ) / ω k . We set d c : = i 2 π ( ∂ ˉ − ∂ ) d^{c} := \frac{i}{2\pi}(\bar{\partial}-\partial) d c := 2 π i ( ∂ ˉ − ∂ ) .
For a positive closed ( 1 , 1 ) (1,1) ( 1 , 1 ) -current S S S on P k \mathbb{P}^{k} P k , write ∥ S ∥ = ∫ P k S ∧ ω k − 1 \|S\| = \int_{\mathbb{P}^{k}} S \wedge\omega^{k-1} ∥ S ∥ = ∫ P k S ∧ ω k − 1 for its mass and denote its cohomology class by [ S ] [S] [ S ] . On a domain V ⊂ C k V \subset\mathbb{C}^{k} V ⊂ C k , write instead ∥ S ∥ V = ∫ V S ∧ ω e u c l k − 1 \|S\|_{V} = \int_{V} S \wedge\omega_{\mathrm{eucl}}^{k-1} ∥ S ∥ V = ∫ V S ∧ ω eucl k − 1 , where ω e u c l = d d c ∣ z ∣ 2 \omega_{\mathrm{eucl}} = dd^{c}|z|^{2} ω eucl = d d c ∣ z ∣ 2 , and denote Lebesgue measure by d V dV d V .
Recall that we denote U : = { u ∈ PSH ( P k , ω ) : sup u = 0 } \mathcal{U} := \{u \in\operatorname{PSH}(\mathbb{P}^{k},\omega): \sup u = 0\} U := { u ∈ PSH ( P k , ω ) : sup u = 0 } . For u ∈ U u \in\mathcal{U} u ∈ U , the current T u : = ω + d d c u T_{u} := \omega+ dd^{c}u T u := ω + d d c u is positive closed and has mass one. Compactness of normalized ω \omega ω -plurisubharmonic functions in L 1 ( ω k ) L^{1}(\omega^{k}) L 1 ( ω k ) gives sup u ∈ U ∫ P k ( − u ) ω k < ∞ \sup_{u\in\mathcal{U}}\int_{\mathbb{P}^{k}}(-u)\,\omega^{k}<\infty sup u ∈ U ∫ P k ( − u ) ω k < ∞ ; see for instance [24 Appendix A] and [37 Proposition 2.7] .
The following is a direct consequence of [24 Proposition A.16] .
Lemma 2.1. Fix k ≥ 1 k \ge1 k ≥ 1 and r > 0 r > 0 r > 0 , and write B t : = B ( 0 , t ) ⊂ C k B_{t} := B(0,t) \subset\mathbb{C}^{k} B t := B ( 0 , t ) ⊂ C k . Suppose that σ , τ \sigma,\tau σ , τ are probability measures supported in B ‾ 2 r \overline{B}_{2r} B 2 r , absolutely continuous with respect to d V dV d V , with densities bounded by some M > 0 M > 0 M > 0 , d V dV d V -almost everywhere. Assume that
∫ H d σ = ∫ H d τ for every real pluriharmonic function H on B 3 r . (2) \int H\,d\sigma= \int H\,d\tau\qquad\text{for every real pluriharmonic function } H \text{ on } B_{3r}. \tag*{(2)} ∫ H d σ = ∫ H d τ for every real pluriharmonic function H on B 3 r . (2)
Let φ ∈ L l o c 1 ( B 4 r ) \varphi\in L^1_{\mathrm{loc}}(B_{4r}) φ ∈ L loc 1 ( B 4 r ) be real-valued and satisfy d d c φ = S + − S − dd^c\varphi= S^+ - S^- d d c φ = S + − S − , where S ± S^\pm S ± are positive closed ( 1 , 1 ) (1,1) ( 1 , 1 ) -currents of finite mass on B 4 r B_{4r} B 4 r . Then there exists a constant C > 0 C > 0 C > 0 , depending only on k k k , such that
∣ ∫ φ d ( σ − τ ) ∣ ≤ C M r 2 ( ∥ S + ∥ B 4 r + ∥ S − ∥ B 4 r ) . \left|\int\varphi\,d(\sigma-\tau)\right| \leq CMr^2\left(\|S^+\|_{B_{4r}}+\|S^-\|_{B_{4r}}\right). ∫ φ d ( σ − τ ) ≤ C M r 2 ( ∥ S + ∥ B 4 r + ∥ S − ∥ B 4 r ) .
Proof. By the local potential estimate in [24 Proposition A.16] , after rescaling to B 4 r B_{4r} B 4 r , there exist plurisubharmonic functions v ± v^\pm v ± on B 3 r B_{3r} B 3 r with d d c v ± = S ± dd^cv^\pm=S^\pm d d c v ± = S ± and ∥ v ± ∥ L 1 ( B 3 r ) ≤ C r 2 ∥ S ± ∥ B 4 r \|v^\pm\|_{L^1(B_{3r})}\leq Cr^2\|S^\pm\|_{B_{4r}} ∥ v ± ∥ L 1 ( B 3 r ) ≤ C r 2 ∥ S ± ∥ B 4 r . Since d d c ( φ − v + + v − ) = 0 dd^c(\varphi-v^++v^-)=0 d d c ( φ − v + + v − ) = 0 , Weyl’s lemma implies that the function φ − v + + v − \varphi-v^++v^- φ − v + + v − agrees d V dV d V -almost everywhere with a pluriharmonic function on B 3 r B_{3r} B 3 r . Absolute continuity and (2) imply that the integrals of φ − v + + v − \varphi-v^++v^- φ − v + + v − against σ \sigma σ and τ \tau τ agree. The density bounds applied to v + − v − v^+-v^- v + − v − give the conclusion. □
Fix r ∗ > 0 r_* > 0 r ∗ > 0 , C ∗ > 1 C_* > 1 C ∗ > 1 , and holomorphic charts Φ x : B ( 0 , r ∗ ) ⊂ C k → P k \Phi_x:B(0,r_*)\subset\mathbb{C}^k\to\mathbb{P}^k Φ x : B ( 0 , r ∗ ) ⊂ C k → P k for x ∈ P k x\in\mathbb{P}^k x ∈ P k , with Φ x ( 0 ) = x \Phi_x(0)=x Φ x ( 0 ) = x and ( x , z ) ↦ Φ x ( z ) (x,z)\mapsto\Phi_x(z) ( x , z ) ↦ Φ x ( z ) Borel measurable. Uniformly in x x x and z , z ′ ∈ B ( 0 , r ∗ ) z,z'\in B(0,r_*) z , z ′ ∈ B ( 0 , r ∗ ) , we have
C ∗ − 1 ∣ z − z ′ ∣ ≤ dist ( Φ x ( z ) , Φ x ( z ′ ) ) ≤ C ∗ ∣ z − z ′ ∣ and C ∗ − 1 d V ≤ Φ x ∗ ω k ≤ C ∗ d V . (3) C_*^{-1}|z-z'|\leq\operatorname{dist}(\Phi_x(z),\Phi_x(z'))\leq C_*|z-z'|
\quad\text{and}\quad
C_*^{-1}dV\leq\Phi_x^*\omega^k\leq C_*dV. \tag*{(3)} C ∗ − 1 ∣ z − z ′ ∣ ≤ dist ( Φ x ( z ) , Φ x ( z ′ )) ≤ C ∗ ∣ z − z ′ ∣ and C ∗ − 1 d V ≤ Φ x ∗ ω k ≤ C ∗ d V . (3)
Such charts can be chosen from a finite holomorphic atlas.
We will use radial averaging measures to obtain exact pluriharmonic cancellation. Fix a non-negative radial function ρ ∈ C c ∞ ( B ( 0 , 1 ) ) \rho\in C^\infty_c(B(0,1)) ρ ∈ C c ∞ ( B ( 0 , 1 )) with ∫ ρ d V = 1 \int\rho\,dV=1 ∫ ρ d V = 1 . For 0 < t < r ∗ 0<t<r_* 0 < t < r ∗ , set
κ x , t : = ( Φ x ) ∗ ( t − 2 k ρ ( z / t ) d V ( z ) ) . (4) \kappa_{x,t}:=(\Phi_x)_*\left(t^{-2k}\rho(z/t)\,dV(z)\right). \tag*{(4)} κ x , t := ( Φ x ) ∗ ( t − 2 k ρ ( z / t ) d V ( z ) ) . (4)
For every Borel set A ⊂ P k A\subset\mathbb{P}^k A ⊂ P k , the map ( x , t ) ↦ κ x , t ( A ) (x,t)\mapsto\kappa_{x,t}(A) ( x , t ) ↦ κ x , t ( A ) is Borel measurable; thus ( κ x , t ) (\kappa_{x,t}) ( κ x , t ) is a Borel probability kernel. Moreover, we have
supp κ x , t ⊂ B ( x , C ∗ t ) , 0 ≤ d κ x , t d ω k ≤ C t − 2 k , and ∫ H d κ x , t = H ( x ) , (5) \operatorname{supp}\kappa_{x,t}\subset B(x,C_*t),\qquad0\leq\frac{d\kappa_{x,t}}{d\omega^k}\leq Ct^{-2k},\qquad\text{and}\qquad\int H\,d\kappa_{x,t}=H(x), \tag*{(5)} supp κ x , t ⊂ B ( x , C ∗ t ) , 0 ≤ d ω k d κ x , t ≤ C t − 2 k , and ∫ H d κ x , t = H ( x ) , (5)
where the constant C > 0 C>0 C > 0 is independent of x x x and t t t , and the last identity holds for every function H H H which is pluriharmonic near Φ x ( B ( 0 , t ) ) \Phi_x(B(0,t)) Φ x ( B ( 0 , t )) , by the Euclidean mean value property for H ∘ Φ x H\circ\Phi_x H ∘ Φ x .
The next lemma is a standard telescoping regularity argument, stated in a dual form for measures. It balances approximation errors against density bounds to obtain integrability and an absolute Hölder estimate; compare for instance the regularity arguments in [24 Lemma 1.19] and [6 Lemma 2.11] .
Lemma 2.2. Fix an integer N ≥ 1 N\geq1 N ≥ 1 . Let ( ν n ) n ≥ N (\nu_n)_{n\geq N} ( ν n ) n ≥ N be finite positive measures on P k \mathbb{P}^k P k of equal mass, converging weakly to ν \nu ν . Suppose that a , b , C > 0 a,b,C>0 a , b , C > 0 satisfy
ν n ≤ C e b n ω k and ∣ ∫ P k u d ν n + 1 − ∫ P k u d ν n ∣ ≤ C e − a n for every n ≥ N , u ∈ U . (6) \nu_n\leq Ce^{bn}\omega^k
\quad\text{and}\quad
\left|\int_{\mathbb{P}^k}u\,d\nu_{n+1}-\int_{\mathbb{P}^k}u\,d\nu_n\right|\leq Ce^{-an}
\quad\text{for every }n\geq N,\ u\in\mathcal{U}. \tag*{(6)} ν n ≤ C e bn ω k and ∫ P k u d ν n + 1 − ∫ P k u d ν n ≤ C e − an for every n ≥ N , u ∈ U . (6) Then ν \nu ν integrates every quasi-plurisubharmonic function. More precisely, setting β : = a / ( a + b ) \beta:=a/(a+b) β := a / ( a + b ) , there is C ′ > 0 C'>0 C ′ > 0 , depending only on a , b , C , N a,b,C,N a , b , C , N , such that
∫ P k ∣ u − v ∣ d ν ≤ C ′ max { ∥ u − v ∥ 1 , ∥ u − v ∥ 1 β } for every u , v ∈ PSH ( P k , ω ) . \int_{\mathbb{P}^k}|u-v|\,d\nu\leq C'\max\left\{\|u-v\|_1,\|u-v\|_1^\beta\right\}
\quad\text{for every }u,v\in\operatorname{PSH}(\mathbb{P}^k,\omega). ∫ P k ∣ u − v ∣ d ν ≤ C ′ max { ∥ u − v ∥ 1 , ∥ u − v ∥ 1 β } for every u , v ∈ PSH ( P k , ω ) . </lemma>
Proof. We first compare integrals of smooth test functions against ν n \nu_n ν n and the limit measure ν \nu ν . Since the measures have equal mass, subtracting sup P k u \sup_{\mathbb{P}^k}u sup P k u does not change any successive difference of integrals. Thus the second hypothesis in (6) applies to every smooth u ∈ PSH ( P k , ω ) u\in\operatorname{PSH}(\mathbb{P}^k,\omega) u ∈ PSH ( P k , ω ) . For m > n m>n m > n , write ν m − ν n = ∑ ℓ = n m − 1 ( ν ℓ + 1 − ν ℓ ) \nu_m-\nu_n=\sum_{\ell=n}^{m-1}(\nu_{\ell+1}-\nu_\ell) ν m − ν n = ∑ ℓ = n m − 1 ( ν ℓ + 1 − ν ℓ ) . Summing the bounds in (2.5) and then letting m → ∞ m\to\infty m → ∞ by weak convergence gives, with C 0 : = C / ( 1 − e − a ) C_0:=C/(1-e^{-a}) C 0 := C / ( 1 − e − a ) ,
∣ ∫ u d ( ν − ν n ) ∣ ≤ ∑ ℓ = n ∞ ∣ ∫ u d ( ν ℓ + 1 − ν ℓ ) ∣ ≤ C 0 e − a n . (7) \left|\int u\,d(\nu-\nu_n)\right|
\leq\sum_{\ell=n}^{\infty}\left|\int u\,d(\nu_{\ell+1}-\nu_\ell)\right|
\leq C_0e^{-an}. \tag*{(7)} ∫ u d ( ν − ν n ) ≤ ℓ = n ∑ ∞ ∫ u d ( ν ℓ + 1 − ν ℓ ) ≤ C 0 e − an . (7)
We next prove integrability and extend (2.6) to every u ∈ PSH ( P k , ω ) u \in\operatorname{PSH}(\mathbb{P}^{k},\omega) u ∈ PSH ( P k , ω ) . Choose smooth functions u j ∈ PSH ( P k , ω ) u_j \in\operatorname{PSH}(\mathbb{P}^{k},\omega) u j ∈ PSH ( P k , ω ) decreasing to u u u ; see, for instance, [24 Proposition A.41] . The density bound gives u ∈ L 1 ( ν N ) u \in L^{1}(\nu_N) u ∈ L 1 ( ν N ) , and (2.6) applied to u j u_j u j yields
∫ u j d ν ≥ ∫ u d ν N − C 0 e − a N . \int u_j\,d\nu\ge\int u\,d\nu_N-C_0e^{-aN}. ∫ u j d ν ≥ ∫ u d ν N − C 0 e − a N .
The integrals ∫ u j d ν \int u_j\,d\nu ∫ u j d ν therefore have a finite uniform lower bound. Applying the monotone convergence theorem to the sequence u 1 − u j u_1-u_j u 1 − u j gives u ∈ L 1 ( ν ) u \in L^{1}(\nu) u ∈ L 1 ( ν ) . For each fixed n ≥ N n \ge N n ≥ N , the density bound also gives u ∈ L 1 ( ν n ) u \in L^{1}(\nu_n) u ∈ L 1 ( ν n ) ; passing to the limit in (2.6) for u j u_j u j proves that bound for u u u . Scaling gives integrability of every quasi-plurisubharmonic function.
Next, we obtain a Hölder bound for the integral of u − v u-v u − v by balancing the approximation error against the density bound. For u , v ∈ PSH ( P k , ω ) u,v \in\operatorname{PSH}(\mathbb{P}^{k},\omega) u , v ∈ PSH ( P k , ω ) and t : = ∥ u − v ∥ 1 > 0 t := \lVert u-v\rVert_1 > 0 t := ∥ u − v ∥ 1 > 0 , (2.6) applied to u u u and v v v , together with the first assumption in (2.5), gives
∣ ∫ ( u − v ) d ν ∣ ≤ C 1 ( e − a n + e b n t ) . (8) \left|\int(u-v)\,d\nu\right| \le C_1(e^{-an}+e^{bnt}). \tag*{(8)} ∫ ( u − v ) d ν ≤ C 1 ( e − an + e bn t ) . (8)
Choosing n = ⌈ log ( 1 / t ) / ( a + b ) ⌉ ≥ N n=\lceil\log(1/t)/(a+b)\rceil\ge N n = ⌈ log ( 1/ t ) / ( a + b )⌉ ≥ N for sufficiently small t > 0 t>0 t > 0 , and n = N n=N n = N for larger t t t , yields
∣ ∫ ( u − v ) d ν ∣ ≤ C 2 max { t , t β } . (9) \left|\int(u-v)\,d\nu\right| \le C_2\max\{t,t^\beta\}. \tag*{(9)} ∫ ( u − v ) d ν ≤ C 2 max { t , t β } . (9)
The constants C 1 , C 2 C_1,C_2 C 1 , C 2 depend only on a , b , C , N a,b,C,N a , b , C , N .
We finally pass from the integral of u − v u-v u − v to the integral of ∣ u − v ∣ \lvert u-v\rvert ∣ u − v ∣ as in the statement. Set w : = max ( u , v ) w:=\max(u,v) w := max ( u , v ) , which belongs to PSH ( P k , ω ) \operatorname{PSH}(\mathbb{P}^{k},\omega) PSH ( P k , ω ) . Since the functions w − u w-u w − u and w − v w-v w − v are non-negative, we have ∣ u − v ∣ = ( w − u ) + ( w − v ) \lvert u-v\rvert=(w-u)+(w-v) ∣ u − v ∣ = ( w − u ) + ( w − v ) , and both ∥ w − u ∥ 1 \lVert w-u\rVert_1 ∥ w − u ∥ 1 and ∥ w − v ∥ 1 \lVert w-v\rVert_1 ∥ w − v ∥ 1 are at most t t t . Applying (2.8) to the pairs ( w , u ) (w,u) ( w , u ) and ( w , v ) (w,v) ( w , v ) and adding gives the stated estimate with C ′ = 2 C 2 C'=2C_2 C ′ = 2 C 2 . □
Inverse branches and entropy estimates
Let μ \mu μ be an ergodic F F F -invariant probability measure on P k \mathbb{P}^{k} P k . To record inverse orbits, we use its natural extension ( X ^ , F ^ , μ ^ ) (\widehat{X},\widehat{F},\widehat{\mu}) ( X , F , μ ) , where
X ^ : = { ( x j ) j ∈ Z ∈ ( P k ) Z : F ( x j ) = x j + 1 } , F ^ ( ( x j ) j ∈ Z ) : = ( x j + 1 ) j ∈ Z . \widehat{X}:=\{(x_j)_{j\in\mathbb{Z}}\in(\mathbb{P}^{k})^{\mathbb{Z}}:F(x_j)=x_{j+1}\},\qquad\widehat{F}((x_j)_{j\in\mathbb{Z}}):=(x_{j+1})_{j\in\mathbb{Z}}. X := {( x j ) j ∈ Z ∈ ( P k ) Z : F ( x j ) = x j + 1 } , F (( x j ) j ∈ Z ) := ( x j + 1 ) j ∈ Z .
For a history x ^ = ( x j ) j ∈ Z ∈ X ^ \widehat{x}=(x_j)_{j\in\mathbb{Z}}\in\widehat{X} x = ( x j ) j ∈ Z ∈ X , write π 0 ( x ^ ) : = x 0 \pi_0(\widehat{x}):=x_0 π 0 ( x ) := x 0 , so F ∘ π 0 = π 0 ∘ F ^ F\circ\pi_0=\pi_0\circ\widehat{F} F ∘ π 0 = π 0 ∘ F . The measure μ ^ \widehat{\mu} μ is the unique F ^ \widehat{F} F -invariant probability measure satisfying ( π 0 ) ∗ μ ^ = μ (\pi_0)_*\widehat{\mu}=\mu ( π 0 ) ∗ μ = μ ; see [18 Chapter 10] .
For x ∈ P k x\in\mathbb{P}^{k} x ∈ P k , δ > 0 \delta>0 δ > 0 , and n ≥ 1 n\ge1 n ≥ 1 , let B n ( x , δ ) : = { y ∈ P k : dist ( F j y , F j x ) < δ for 0 ≤ j < n } B_n(x,\delta):=\{y\in\mathbb{P}^{k}:\operatorname{dist}(F^j y,F^j x)<\delta\text{ for }0\le j<n\} B n ( x , δ ) := { y ∈ P k : dist ( F j y , F j x ) < δ for 0 ≤ j < n } be the *Bowen ball of length n n n and radius δ \delta δ . A set is ( n , δ ) (n,\delta) ( n , δ ) -separated if every two distinct points have distance at least δ \delta δ under some iterate F j F^j F j with 0 ≤ j < n 0\le j<n 0 ≤ j < n .
If h μ ( F ) > ( k − 1 ) log d h_\mu(F)>(k-1)\log d h μ ( F ) > ( k − 1 ) log d , then all Lyapunov exponents of μ \mu μ are strictly positive. For k = 1 k=1 k = 1 , Ruelle’s entropy inequality gives a strictly positive Lyapunov exponent whenever h μ ( F ) > 0 h_\mu(F)>0 h μ ( F ) > 0 . For k ≥ 2 k\ge2 k ≥ 2 , positivity follows from [30 Theorem C] , which extends de Thélin’s entropy–Lyapunov estimate [26 ] without assuming integrability of log Jac R F \log\operatorname{Jac}_{\mathbb{R}}F log Jac R F . In either case, positivity rules out the value − ∞ -\infty − ∞ , while boundedness of D F DF D F rules out + ∞ +\infty + ∞ , so all Lyapunov exponents are finite. Consequently, the multiplicative ergodic theorem gives a finite limit for n − 1 log Jac R ( F n ) n^{-1}\log\operatorname{Jac}_{\mathbb{R}}(F^n) n − 1 log Jac R ( F n ) at μ \mu μ -almost every point. Since log Jac R F \log\operatorname{Jac}_{\mathbb{R}}F log Jac R F is bounded above, the Birkhoff ergodic theorem applied to its bounded lower truncations gives log Jac R F ∈ L 1 ( μ ) \log\operatorname{Jac}_{\mathbb{R}}F\in L^1(\mu) log Jac R F ∈ L 1 ( μ ) : otherwise, n − 1 log Jac R ( F n ) n^{-1}\log\operatorname{Jac}_{\mathbb{R}}(F^n) n − 1 log Jac R ( F n ) would tend to − ∞ -\infty − ∞ μ \mu μ -almost everywhere.
The next lemma combines the inverse branch estimates of [8 , 5 ] with the Brin–Katok entropy formula [11 ] on one set of histories of large measure.
Lemma 2.3. Suppose that h μ ( F ) > ( k − 1 ) log d h_\mu(F)>(k-1)\log d h μ ( F ) > ( k − 1 ) log d , and fix ( k − 1 ) log d < s < h μ ( F ) (k-1)\log d<s<h_\mu(F) ( k − 1 ) log d < s < h μ ( F ) and 0 < θ < 1 0<\theta<1 0 < θ < 1 . Let χ min > 0 \chi_{\min}>0 χ m i n > 0 denote the smallest Lyapunov exponent of μ \mu μ . For every 0 < ε < χ min / 2 0<\varepsilon<\chi_{\min}/2 0 < ε < χ m i n /2 , there exist a Borel set E ^ ⊂ X ^ \widehat{E}\subset\widehat{X} E ⊂ X , constants r 0 , δ , λ > 0 r_0,\delta,\lambda>0 r 0 , δ , λ > 0 , A ≥ 1 A\ge1 A ≥ 1 , and an integer N ≥ 1 N\ge1 N ≥ 1 such that μ ^ ( E ^ ) > 1 − θ \widehat{\mu}(\widehat{E})>1-\theta μ ( E ) > 1 − θ and the following hold for every x ^ ∈ E ^ \widehat{x}\in\widehat{E} x ∈ E .
(i) For every n ≥ 1 n \ge1 n ≥ 1 , there is an inverse branch g ^ x ^ , n : B ( x n , r 0 e − ε n ) → P k \hat{g}_{\hat{x},n}: B(x_n,r_0e^{-\varepsilon n}) \to\mathbb{P}^k g ^ x ^ , n : B ( x n , r 0 e − ε n ) → P k with F n ∘ g ^ x ^ , n = id F^n \circ\hat{g}_{\hat{x},n} = \operatorname{id} F n ∘ g ^ x ^ , n = id and g ^ x ^ , n ( x n ) = x 0 \hat{g}_{\hat{x},n}(x_n) = x_0 g ^ x ^ , n ( x n ) = x 0 . For all y ∈ B ( x n , r 0 e − ε n ) y \in B(x_n,r_0e^{-\varepsilon n}) y ∈ B ( x n , r 0 e − ε n ) and 0 ≤ j ≤ n 0 \le j \le n 0 ≤ j ≤ n , we have dist ( F j g ^ x ^ , n ( y ) , x j ) ≤ A e ε n e − λ ( n − j ) dist ( y , x n ) \operatorname{dist}(F^j\hat{g}_{\hat{x},n}(y),x_j) \le Ae^{\varepsilon n}e^{-\lambda(n-j)}\operatorname{dist}(y,x_n) dist ( F j g ^ x ^ , n ( y ) , x j ) ≤ A e ε n e − λ ( n − j ) dist ( y , x n ) .
(ii) For every n ≥ N n \ge N n ≥ N , we have μ ( B n ( x 0 , δ ) ) ≤ e − s n \mu(B_n(x_0,\delta)) \le e^{-sn} μ ( B n ( x 0 , δ )) ≤ e − s n .
Proof. For (i), we use the tempered inverse branch estimates of [8 Theorem 1.4] and [5 ] ; see also [9 Section 2.2] . Fix λ : = χ min / 2 \lambda:= \chi_{\min}/2 λ := χ m i n /2 . The radius and distortion functions given by these estimates are μ ^ \hat{\mu} μ ^ -measurable, possibly with respect to the completed σ \sigma σ -algebra. Since X ^ \hat{X} X ^ is a standard Borel space, they admit Borel versions. Choose a Borel set Y ^ 0 \hat{Y}_0 Y ^ 0 of full μ ^ \hat{\mu} μ ^ -measure on which these versions agree with the original functions and all the inverse branch and tempered estimates hold, and set Y ^ : = ⋂ q ∈ Z F ^ q ( Y ^ 0 ) \hat{Y} := \bigcap_{q\in\mathbb{Z}}\hat{F}^q(\hat{Y}_0) Y ^ := ⋂ q ∈ Z F ^ q ( Y ^ 0 ) . Then Y ^ \hat{Y} Y ^ is an invariant Borel set of full μ ^ \hat{\mu} μ ^ -measure. On Y ^ \hat{Y} Y ^ we obtain Borel functions r : Y ^ → ( 0 , 1 ] r:\hat{Y}\to(0,1] r : Y ^ → ( 0 , 1 ] and L : Y ^ → [ 1 , ∞ ) L:\hat{Y}\to[1,\infty) L : Y ^ → [ 1 , ∞ ) and, for y ^ = ( y j ) j ∈ Z ∈ Y ^ \hat{y}=(y_j)_{j\in\mathbb{Z}}\in\hat{Y} y ^ = ( y j ) j ∈ Z ∈ Y ^ , inverse branches b ^ y ^ , m \hat{b}_{\hat{y},m} b ^ y ^ , m of F m F^m F m sending y 0 y_0 y 0 to y − m y_{-m} y − m , defined on B ( y 0 , r ( y ^ ) ) B(y_0,r(\hat{y})) B ( y 0 , r ( y ^ )) and satisfying Lip ( b ^ y ^ , m ) ≤ L ( y ^ ) e − λ m \operatorname{Lip}(\hat{b}_{\hat{y},m})\le L(\hat{y})e^{-\lambda m} Lip ( b ^ y ^ , m ) ≤ L ( y ^ ) e − λm for m ≥ 1 m\ge1 m ≥ 1 . The functions r r r and L L L can be chosen so that r ( F ^ q y ^ ) ≥ e − ε q r ( y ^ ) r(\hat{F}^q\hat{y})\ge e^{-\varepsilon q}r(\hat{y}) r ( F ^ q y ^ ) ≥ e − εq r ( y ^ ) and L ( F ^ q y ^ ) ≤ e ε q L ( y ^ ) L(\hat{F}^q\hat{y})\le e^{\varepsilon q}L(\hat{y}) L ( F ^ q y ^ ) ≤ e εq L ( y ^ ) for every integer q ≥ 0 q\ge0 q ≥ 0 .
Choose r 0 > 0 r_0>0 r 0 > 0 and A ≥ 1 A\ge1 A ≥ 1 so that E ^ 1 : = { y ^ ∈ Y ^ : r ( y ^ ) ≥ r 0 , L ( y ^ ) ≤ A } \hat{E}_1:=\{\hat{y}\in\hat{Y}:r(\hat{y})\ge r_0,\ L(\hat{y})\le A\} E ^ 1 := { y ^ ∈ Y ^ : r ( y ^ ) ≥ r 0 , L ( y ^ ) ≤ A } satisfies μ ^ ( E ^ 1 ) > 1 − θ / 2 \hat{\mu}(\hat{E}_1)>1-\theta/2 μ ^ ( E ^ 1 ) > 1 − θ /2 . For x ^ ∈ E ^ 1 \hat{x}\in\hat{E}_1 x ^ ∈ E ^ 1 , we have r ( F ^ n x ^ ) ≥ e − ε n r ( x ^ ) ≥ r 0 e − ε n r(\hat{F}^n\hat{x})\ge e^{-\varepsilon n}r(\hat{x})\ge r_0e^{-\varepsilon n} r ( F ^ n x ^ ) ≥ e − ε n r ( x ^ ) ≥ r 0 e − ε n . Thus we can define g ^ x ^ , n \hat{g}_{\hat{x},n} g ^ x ^ , n as the restriction of b ^ F ^ n x ^ , n \hat{b}_{\hat{F}^n\hat{x},n} b ^ F ^ n x ^ , n to B ( x n , r 0 e − ε n ) B(x_n,r_0e^{-\varepsilon n}) B ( x n , r 0 e − ε n ) . For 0 ≤ j < n 0\le j<n 0 ≤ j < n , uniqueness of inverse germs gives F j ∘ g ^ x ^ , n = b ^ F ^ n x ^ , n − j F^j\circ\hat{g}_{\hat{x},n}=\hat{b}_{\hat{F}^n\hat{x},n-j} F j ∘ g ^ x ^ , n = b ^ F ^ n x ^ , n − j on this ball. The bound L ( F ^ n x ^ ) ≤ A e ε n L(\hat{F}^n\hat{x})\le Ae^{\varepsilon n} L ( F ^ n x ^ ) ≤ A e ε n now gives (i); the case j = n j=n j = n follows from F n ∘ g ^ x ^ , n = id F^n\circ\hat{g}_{\hat{x},n}=\operatorname{id} F n ∘ g ^ x ^ , n = id .
For (ii), the Brin–Katok entropy formula [11 ] and s < h μ ( F ) s<h_\mu(F) s < h μ ( F ) give δ > 0 \delta>0 δ > 0 , N ≥ 1 N\ge1 N ≥ 1 , and a Borel set E 0 ⊂ P k E_0\subset\mathbb{P}^k E 0 ⊂ P k with μ ( E 0 ) > 1 − θ / 2 \mu(E_0)>1-\theta/2 μ ( E 0 ) > 1 − θ /2 such that μ ( B n ( x , δ ) ) ≤ e − s n \mu(B_n(x,\delta))\le e^{-sn} μ ( B n ( x , δ )) ≤ e − s n for every x ∈ E 0 x\in E_0 x ∈ E 0 and n ≥ N n\ge N n ≥ N . Set E ^ : = E ^ 1 ∩ π 0 − 1 ( E 0 ) \hat{E}:=\hat{E}_1\cap\pi_0^{-1}(E_0) E ^ := E ^ 1 ∩ π 0 − 1 ( E 0 ) . Both conclusions hold on E ^ \hat{E} E ^ , and μ ^ ( E ^ ) > 1 − θ \hat{\mu}(\hat{E})>1-\theta μ ^ ( E ^ ) > 1 − θ since ( π 0 ) ∗ μ ^ = μ (\pi_0)_*\hat{\mu}=\mu ( π 0 ) ∗ μ ^ = μ . □
An inverse branch regularization criterion
The following criterion deduces Hölder regularity from the existence of contracting inverse branches and Bowen ball mass bounds. In the application to Theorem 1.1 , Ω \Omega Ω will be the good set of histories E ^ \hat{E} E ^ provided by Lemma 2.3 . The measure η \eta η is not assumed to be invariant. When ξ \xi ξ is fixed, we write x j x_j x j for x j ( ξ ) x_j(\xi) x j ( ξ ) .
Theorem 3.1. Let F : P k → P k F:\mathbb{P}^k\to\mathbb{P}^k F : P k → P k be a holomorphic endomorphism of algebraic degree d ≥ 2 d\ge2 d ≥ 2 , with k ≥ 1 k\ge1 k ≥ 1 . Let m m m be a finite positive measure on a standard Borel space Ω \Omega Ω , with 0 < m ( Ω ) ≤ 1 0<m(\Omega)\le1 0 < m ( Ω ) ≤ 1 , let p : Ω → P k p:\Omega\to\mathbb{P}^k p : Ω → P k be Borel, and set η : = p ∗ m \eta:=p_*m η := p ∗ m and x j ( ξ ) : = F j ( p ( ξ ) ) x_j(\xi):=F^j(p(\xi)) x j ( ξ ) := F j ( p ( ξ )) for ξ ∈ Ω \xi\in\Omega ξ ∈ Ω and j ≥ 0 j\ge0 j ≥ 0 . Fix s > ( k − 1 ) log d s>(k-1)\log d s > ( k − 1 ) log d and ε > 0 \varepsilon>0 ε > 0 with ( 8 k − 4 ) ε < s − ( k − 1 ) log d (8k-4)\varepsilon<s-(k-1)\log d ( 8 k − 4 ) ε < s − ( k − 1 ) log d . Suppose that there are r 0 , λ , δ , C F > 0 r_0,\lambda,\delta,C_F>0 r 0 , λ , δ , C F > 0 , A ≥ 1 A\ge1 A ≥ 1 , and an integer N ≥ 1 N\ge1 N ≥ 1 such that, for every ξ ∈ Ω \xi\in\Omega ξ ∈ Ω , the following hold.
(i) For every n ≥ 1 n\ge1 n ≥ 1 , there is a holomorphic map g ξ , n : B ( x n , r 0 e − ε n ) → P k g_{\xi,n}:B(x_n,r_0e^{-\varepsilon n})\to\mathbb{P}^k g ξ , n : B ( x n , r 0 e − ε n ) → P k such that F n ∘ g ξ , n = id F^n\circ g_{\xi,n}=\operatorname{id} F n ∘ g ξ , n = id , g ξ , n ( x n ) = x 0 g_{\xi,n}(x_n)=x_0 g ξ , n ( x n ) = x 0 , and
dist ( F j g ξ , n ( y ) , x j ) ≤ A e ε n e − λ ( n − j ) dist ( y , x n ) , 0 ≤ j ≤ n , y ∈ B ( x n , r 0 e − ε n ) . \operatorname{dist}(F^jg_{\xi,n}(y),x_j)\le Ae^{\varepsilon n}e^{-\lambda(n-j)}\operatorname{dist}(y,x_n),\qquad0\le j\le n,\quad y\in B(x_n,r_0e^{-\varepsilon n}). dist ( F j g ξ , n ( y ) , x j ) ≤ A e ε n e − λ ( n − j ) dist ( y , x n ) , 0 ≤ j ≤ n , y ∈ B ( x n , r 0 e − ε n ) . (ii) For every n ≥ N n\ge N n ≥ N , we have η ( B n ( x 0 , δ ) ) ≤ C F e − s n \eta(B_n(x_0,\delta))\le C_Fe^{-sn} η ( B n ( x 0 , δ )) ≤ C F e − s n .
Then η \eta η integrates every quasi-plurisubharmonic function, and there exist β ∈ ( 0 , 1 ) \beta\in(0,1) β ∈ ( 0 , 1 ) and C > 0 C>0 C > 0 such that
∫ P k ∣ u − v ∣ d η ≤ C max { ∥ u − v ∥ 1 , ∥ u − v ∥ 1 β } for all u , v ∈ PSH ( P k , ω ) . \int_{\mathbb{P}^k}|u-v|\,d\eta\le C\max\left\{\|u-v\|_1,\|u-v\|_1^\beta\right\}\qquad\text{for all }u,v\in\operatorname{PSH}(\mathbb{P}^k,\omega). ∫ P k ∣ u − v ∣ d η ≤ C max { ∥ u − v ∥ 1 , ∥ u − v ∥ 1 β } for all u , v ∈ PSH ( P k , ω ) . The exponent β \beta β can be chosen to depend only on s , ε , F s,\varepsilon,F s , ε , F . In particular, η \eta η has a Hölder continuous super-potential.
Observe that no measurability of the family ( g ξ , n ) ξ ∈ Ω (g_{\xi,n})_{\xi\in\Omega} ( g ξ , n ) ξ ∈ Ω is assumed.
Proof. Recall that we denote U = { u ∈ PSH ( P k , ω ) : sup P k u = 0 } \mathcal{U}=\{u\in\operatorname{PSH}(\mathbb{P}^{k},\omega):\sup_{\mathbb{P}^{k}}u=0\} U = { u ∈ PSH ( P k , ω ) : sup P k u = 0 } ; see § 2.1 . We will apply Lemma 2.2 by constructing measures η n \eta_n η n of mass m ( Ω ) m(\Omega) m ( Ω ) that converge weakly to η \eta η and satisfy the two estimates in (2.5) for suitable constants a a a , b b b , C > 0 C>0 C > 0 .
We use the charts Φ x : B ( 0 , r ∗ ) → P k \Phi_x:B(0,r_*)\to\mathbb{P}^{k} Φ x : B ( 0 , r ∗ ) → P k and their metric and volume comparison constant C ∗ C_* C ∗ fixed in Section 2.1. Throughout the proof, C C C denotes a positive constant that may change from line to line, but is independent of n n n , j j j , ξ \xi ξ , and u ∈ U u\in\mathcal{U} u ∈ U .
Step 1. Constructing regularizations and grouping branches. The goal of this step is to construct absolutely continuous approximations η n \eta_n η n of η \eta η that converge weakly to η \eta η and satisfy the first bound in (2.5). For each n ≥ N n\geq N n ≥ N , we also construct a finite Borel partition of Ω \Omega Ω into sets of m m m -mass O ( e − s n ) O(e^{-sn}) O ( e − s n ) , on each of which the corresponding restricted inverse branch is fixed. For each ξ \xi ξ , we will replace the point mass at x 0 ( ξ ) x_0(\xi) x 0 ( ξ ) by a radial averaging measure centred at x n ( ξ ) x_n(\xi) x n ( ξ ) and transported back near x 0 ( ξ ) x_0(\xi) x 0 ( ξ ) by g ξ , n g_{\xi,n} g ξ , n . To verify that the transported radial measures can be integrated over Ω \Omega Ω , we will first restrict the branches g ξ , n g_{\xi,n} g ξ , n to finitely many fixed domains.
Step 1.1. A Borel partition adapted to the inverse branches. Choose a > 0 a>0 a > 0 such that
4 a < r ∗ , 5 C ∗ a < r 0 / 2 , 2 C ∗ A a < δ 4a<r_*,\qquad5C_*a<r_0/2,\qquad2C_*Aa<\delta 4 a < r ∗ , 5 C ∗ a < r 0 /2 , 2 C ∗ A a < δ
and set R n : = a e − 2 ε n R_n:=ae^{-2\varepsilon n} R n := a e − 2 ε n . For each n n n , choose a maximal R n / ( 2 C ∗ ) R_n/(2C_*) R n / ( 2 C ∗ ) -separated set of centres c n , j c_{n,j} c n , j in P k \mathbb{P}^{k} P k , and set U n , j ( ℓ ) : = Φ c n , j ( B ( 0 , ℓ R n ) ) U_{n,j}^{(\ell)}:=\Phi_{c_{n,j}}(B(0,\ell R_n)) U n , j ( ℓ ) := Φ c n , j ( B ( 0 , ℓ R n )) for 1 ≤ ℓ ≤ 4 1\leq\ell\leq4 1 ≤ ℓ ≤ 4 and U n , j : = U n , j ( 4 ) U_{n,j}:=U_{n,j}^{(4)} U n , j := U n , j ( 4 ) . By maximality, the balls B ( c n , j , R n / ( 2 C ∗ ) ) B(c_{n,j},R_n/(2C_*)) B ( c n , j , R n / ( 2 C ∗ )) cover P k \mathbb{P}^{k} P k . Since each such ball is contained in U n , j ( 1 ) U_{n,j}^{(1)} U n , j ( 1 ) , the sets U n , j ( 1 ) U_{n,j}^{(1)} U n , j ( 1 ) also cover P k \mathbb{P}^{k} P k . Moreover, U n , j ⊂ B ( c n , j , 4 C ∗ R n ) U_{n,j}\subset B(c_{n,j},4C_*R_n) U n , j ⊂ B ( c n , j , 4 C ∗ R n ) . Hence, if x ∈ U n , j x\in U_{n,j} x ∈ U n , j , then c n , j ∈ B ( x , 4 C ∗ R n ) c_{n,j}\in B(x,4C_*R_n) c n , j ∈ B ( x , 4 C ∗ R n ) . Since the centres are R n / ( 2 C ∗ ) R_n/(2C_*) R n / ( 2 C ∗ ) -separated, the balls B ( c n , j , R n / ( 4 C ∗ ) ) B(c_{n,j},R_n/(4C_*)) B ( c n , j , R n / ( 4 C ∗ )) are pairwise disjoint. For all indices j j j such that x ∈ U n , j x\in U_{n,j} x ∈ U n , j , these balls are contained in B ( x , C R n ) B(x,CR_n) B ( x , C R n ) , where C C C is independent of n n n and x x x . Uniform volume comparison for small balls on P k \mathbb{P}^{k} P k therefore shows that every point belongs to at most a fixed number of the sets U n , j U_{n,j} U n , j , independently of n n n . Order the centres and set D n , j : = U n , j ( 1 ) ∖ ⋃ i < j U n , i ( 1 ) D_{n,j}:=U_{n,j}^{(1)}\setminus\bigcup_{i<j}U_{n,i}^{(1)} D n , j := U n , j ( 1 ) ∖ ⋃ i < j U n , i ( 1 ) . Then { D n , j } j \{D_{n,j}\}_j { D n , j } j is a finite Borel partition of P k \mathbb{P}^{k} P k .
Step 1.2. Measurability of the regularizations. For ξ \xi ξ with x n ( ξ ) ∈ D n , j x_n(\xi)\in D_{n,j} x n ( ξ ) ∈ D n , j , the estimates for Φ c n , j \Phi_{c_{n,j}} Φ c n , j in (2.2) give U ‾ n , j ⊂ B ( x n ( ξ ) , 5 C ∗ R n ) ⊂ B ( x n ( ξ ) , r 0 e − ε n ) \overline{U}_{n,j}\subset B(x_n(\xi),5C_*R_n)\subset B(x_n(\xi),r_0e^{-\varepsilon n}) U n , j ⊂ B ( x n ( ξ ) , 5 C ∗ R n ) ⊂ B ( x n ( ξ ) , r 0 e − ε n ) . Thus every such g ξ , n g_{\xi,n} g ξ , n is holomorphic on a neighbourhood of U ‾ n , j \overline{U}_{n,j} U n , j and restricts to an inverse branch of F n F^n F n on the same connected domain U n , j U_{n,j} U n , j .
Choose t 0 > 0 t_0>0 t 0 > 0 such that t 0 < r ∗ t_0<r_* t 0 < r ∗ and 2 A C ∗ 2 t 0 < a 2AC_*^2t_0<a 2 A C ∗ 2 t 0 < a , and set
t n : = t 0 e − 4 ε n and K ξ , n : = Φ x n ( B ( 0 , t n ) ) . t_n:=t_0e^{-4\varepsilon n}\qquad\text{and}\qquad K_{\xi,n}:=\Phi_{x_n}(B(0,t_n)). t n := t 0 e − 4 ε n and K ξ , n := Φ x n ( B ( 0 , t n )) .
If x n ∈ D n , j x_n\in D_{n,j} x n ∈ D n , j , the chart comparisons give dist ( x n , ∂ U n , j ( 2 ) ) ≥ R n / C ∗ \operatorname{dist}(x_n,\partial U_{n,j}^{(2)})\geq R_n/C_* dist ( x n , ∂ U n , j ( 2 ) ) ≥ R n / C ∗ and K ‾ ξ , n ⊂ B ‾ ( x n , C ∗ t n ) \overline{K}_{\xi,n}\subset\overline{B}(x_n,C_*t_n) K ξ , n ⊂ B ( x n , C ∗ t n ) . Since C ∗ t n < R n / ( 2 C ∗ ) C_*t_n<R_n/(2C_*) C ∗ t n < R n / ( 2 C ∗ ) , we have K ‾ ξ , n ⊂ U n , j ( 2 ) \overline{K}_{\xi,n}\subset U_{n,j}^{(2)} K ξ , n ⊂ U n , j ( 2 ) . In particular, g ξ , n g_{\xi,n} g ξ , n is defined on a neighbourhood of supp χ x n , t n \operatorname{supp}\chi_{x_n,t_n} supp χ x n , t n (see (2.3)), and we set
Q ξ , n : = ( g ξ , n ) ∗ κ x n , t n . Q_{\xi,n}:=(g_{\xi,n})_*\kappa_{x_n,t_n}. Q ξ , n := ( g ξ , n ) ∗ κ x n , t n .
For each fixed n n n , we now partition Ω \Omega Ω according to the index j j j such that x n ( ξ ) ∈ D n , j x_n(\xi)\in D_{n,j} x n ( ξ ) ∈ D n , j and the restriction of g ξ , n g_{\xi,n} g ξ , n to U n , j U_{n,j} U n , j . Let G n , j \mathcal{G}_{n,j} G n , j be the collection of restrictions g ξ , n ∣ U n , j g_{\xi,n}|_{U_{n,j}} g ξ , n ∣ U n , j with x n ( ξ ) ∈ D n , j x_n(\xi)\in D_{n,j} x n ( ξ ) ∈ D n , j and, for g ∈ G n , j g\in\mathcal{G}_{n,j} g ∈ G n , j , set
Ω n , j , g : = { ξ ∈ Ω : x n ( ξ ) ∈ D n , j , x 0 ( ξ ) = g ( x n ( ξ ) ) } . \Omega_{n,j,g}:=\{\xi\in\Omega:x_n(\xi)\in D_{n,j},\ x_0(\xi)=g(x_n(\xi))\}. Ω n , j , g := { ξ ∈ Ω : x n ( ξ ) ∈ D n , j , x 0 ( ξ ) = g ( x n ( ξ ))} .
For fixed n n n and j j j , choose a regular value y ∈ U n , j y\in U_{n,j} y ∈ U n , j of F n F^n F n . Evaluation at y y y defines an injective map
G n , j ⟶ ( F n ) − 1 ( y ) , g ↦ g ( y ) . \mathcal{G}_{n,j}\longrightarrow(F^n)^{-1}(y),\qquad g\mapsto g(y). G n , j ⟶ ( F n ) − 1 ( y ) , g ↦ g ( y ) .
Indeed, if two inverse branches agree at y y y , uniqueness of the local inverse germ and connectedness of U n , j U_{n,j} U n , j imply that they agree identically. Since the fibre ( F n ) − 1 ( y ) (F^n)^{-1}(y) ( F n ) − 1 ( y ) is finite, G n , j \mathcal{G}_{n,j} G n , j is finite.
If ξ ∈ Ω n , j , g \xi\in\Omega_{n,j,g} ξ ∈ Ω n , j , g , then g ξ , n ∣ U n , j g_{\xi,n}|_{U_{n,j}} g ξ , n ∣ U n , j and g g g are inverse branches of F n F^n F n on U n , j U_{n,j} U n , j with the same value at x n ( ξ ) x_n(\xi) x n ( ξ ) . Hence, we have g ξ , n ∣ U n , j = g g_{\xi,n}|_{U_{n,j}}=g g ξ , n ∣ U n , j = g . Moreover, every Ω n , j , g \Omega_{n,j,g} Ω n , j , g is Borel, since ξ ↦ x 0 ( ξ ) \xi\mapsto x_0(\xi) ξ ↦ x 0 ( ξ ) and ξ ↦ x n ( ξ ) \xi\mapsto x_n(\xi) ξ ↦ x n ( ξ ) are Borel and g g g is continuous. Every ξ ∈ Ω \xi\in\Omega ξ ∈ Ω belongs to one such set: the partition { D n , j } j \{D_{n,j}\}_j { D n , j } j determines j j j , and g ξ , n ∣ U n , j ∈ G n , j g_{\xi,n}|_{U_{n,j}}\in\mathcal{G}_{n,j} g ξ , n ∣ U n , j ∈ G n , j gives the corresponding branch g g g . If ξ \xi ξ belongs to both Ω n , j , g \Omega_{n,j,g} Ω n , j , g and Ω n , j , h \Omega_{n,j,h} Ω n , j , h , then g ( x n ) = h ( x n ) g(x_n)=h(x_n) g ( x n ) = h ( x n ) ; uniqueness of the inverse germ and connectedness of U n , j U_{n,j} U n , j give g = h g=h g = h . Thus, for every n n n , the sets { Ω n , j , g } j , g \{\Omega_{n,j,g}\}_{j,g} { Ω n , j , g } j , g form a finite Borel partition of Ω \Omega Ω .
On Ω n , j , g \Omega_{n,j,g} Ω n , j , g , we have Q ξ , n = g ∗ κ x n ( ξ ) , t n Q_{\xi,n}=g_*\kappa_{x_n(\xi),t_n} Q ξ , n = g ∗ κ x n ( ξ ) , t n . In particular, to check measurability, observe that for every Borel set A ⊂ P k A\subset\mathbb{P}^k A ⊂ P k we have Q ξ , n ( A ) = κ x n ( ξ ) , t n ( g − 1 ( A ) ) Q_{\xi,n}(A)=\kappa_{x_n(\xi),t_n}(g^{-1}(A)) Q ξ , n ( A ) = κ x n ( ξ ) , t n ( g − 1 ( A )) on Ω n , j , g \Omega_{n,j,g} Ω n , j , g . The Borel dependence of the radial kernels on their centres and the finiteness of the partition show that ( Q ξ , n ) ξ ∈ Ω (Q_{\xi,n})_{\xi\in\Omega} ( Q ξ , n ) ξ ∈ Ω is a Borel probability kernel. We may therefore set
η n : = ∫ Ω Q ξ , n d m ( ξ ) . \eta_n:=\int_{\Omega}Q_{\xi,n}\,dm(\xi). η n := ∫ Ω Q ξ , n d m ( ξ ) .
Step 1.3. Convergence and density bounds. We now prove weak convergence η n → η \eta_n\to\eta η n → η and bound the densities of η n \eta_n η n . Hypothesis (i), applied with j = 0 j=0 j = 0 , and (2.4) give, uniformly in ξ \xi ξ , supp Q ξ , n ⊂ B ( x 0 , C e − ( λ + 3 ε ) n ) \operatorname{supp}Q_{\xi,n}\subset B(x_0,Ce^{-(\lambda+3\varepsilon)n}) supp Q ξ , n ⊂ B ( x 0 , C e − ( λ + 3 ε ) n ) . Take φ ∈ C ( P k ) \varphi\in C(\mathbb{P}^k) φ ∈ C ( P k ) and denote by m φ m_\varphi m φ its modulus of continuity. Since η = p ∗ m \eta=p_*m η = p ∗ m , we have
∣ ∫ P k φ d η n − ∫ P k φ d η ∣ ≤ ∫ Ω ∫ P k ∣ φ ( y ) − φ ( x 0 ( ξ ) ) ∣ d Q ξ , n ( y ) d m ( ξ ) ≤ m ( Ω ) m φ ( C e − ( λ + 3 ε ) n ) . \left|\int_{\mathbb{P}^k}\varphi\,d\eta_n-\int_{\mathbb{P}^k}\varphi\,d\eta\right|
\leq\int_{\Omega}\int_{\mathbb{P}^k}|\varphi(y)-\varphi(x_0(\xi))|\,dQ_{\xi,n}(y)\,dm(\xi)
\leq m(\Omega)m_\varphi\left(Ce^{-(\lambda+3\varepsilon)n}\right). ∫ P k φ d η n − ∫ P k φ d η ≤ ∫ Ω ∫ P k ∣ φ ( y ) − φ ( x 0 ( ξ )) ∣ d Q ξ , n ( y ) d m ( ξ ) ≤ m ( Ω ) m φ ( C e − ( λ + 3 ε ) n ) .
which tends to zero as n → ∞ n\to\infty n → ∞ . Hence η n → η \eta_n\to\eta η n → η weakly. Moreover, all the measures η n \eta_n η n have mass m ( Ω ) m(\Omega) m ( Ω ) .
Write κ x n , t n = q ξ , n ω k \kappa_{x_n,t_n}=q_{\xi,n}\,\omega^k κ x n , t n = q ξ , n ω k , with 0 ≤ q ξ , n ≤ C t n − 2 k 0\leq q_{\xi,n}\leq Ct_n^{-2k} 0 ≤ q ξ , n ≤ C t n − 2 k . Since F n ∘ g ξ , n = id F^n\circ g_{\xi,n}=\operatorname{id} F n ∘ g ξ , n = id , changing variables along the inverse branch gives
d Q ξ , n d ω k ( z ) = { q ξ , n ( F n ( z ) ) Jac R ( F n ) ( z ) , z ∈ g ξ , n ( U n , j ) , 0 , otherwise , (10) \frac{dQ_{\xi,n}}{d\omega^k}(z)=
\begin{cases}
q_{\xi,n}(F^n(z))\operatorname{Jac}_{\mathbb{R}}(F^n)(z), & z\in g_{\xi,n}(U_{n,j}),\\
0, & \text{otherwise},
\end{cases} \tag*{(10)} d ω k d Q ξ , n ( z ) = { q ξ , n ( F n ( z )) Jac R ( F n ) ( z ) , 0 , z ∈ g ξ , n ( U n , j ) , otherwise , (10)
where j j j is determined by x n ∈ D n , j x_n\in D_{n,j} x n ∈ D n , j . Hence, we have Q ξ , n ≤ C t n − 2 k ∥ Jac R ( F n ) ∥ ∞ ω k Q_{\xi,n}\leq Ct_n^{-2k}\|\operatorname{Jac}_{\mathbb{R}}(F^n)\|_\infty\omega^k Q ξ , n ≤ C t n − 2 k ∥ Jac R ( F n ) ∥ ∞ ω k . Integrating over Ω \Omega Ω and using m ( Ω ) ≤ 1 m(\Omega)\leq1 m ( Ω ) ≤ 1 gives η n ≤ C t n − 2 k ∥ Jac R ( F n ) ∥ ∞ ω k \eta_n\leq Ct_n^{-2k}\|\operatorname{Jac}_{\mathbb{R}}(F^n)\|_\infty\omega^k η n ≤ C t n − 2 k ∥ Jac R ( F n ) ∥ ∞ ω k . Since t n = t 0 e − 4 ε n t_n=t_0e^{-4\varepsilon n} t n = t 0 e − 4 ε n and ∥ Jac R ( F n ) ∥ ∞ ≤ ∥ Jac R F ∥ ∞ \|\operatorname{Jac}_{\mathbb{R}}(F^n)\|_\infty\leq\|\operatorname{Jac}_{\mathbb{R}}F\|_\infty ∥ Jac R ( F n ) ∥ ∞ ≤ ∥ Jac R F ∥ ∞ , we obtain
η n ≤ C e b 0 n ω k , b 0 : = 8 k ε + log max { 1 , ∥ Jac R F ∥ ∞ } > 0. (11) \eta_n\leq Ce^{b_0n}\omega^k,\qquad b_0:=8k\varepsilon+\log\max\{1,\|\operatorname{Jac}_{\mathbb{R}}F\|_\infty\}>0. \tag*{(11)} η n ≤ C e b 0 n ω k , b 0 := 8 k ε + log max { 1 , ∥ Jac R F ∥ ∞ } > 0. (11)
Step 1.4. Mass bounds for the partition elements. We finally estimate the m m m -mass of each set Ω n , j , g \Omega_{n,j,g} Ω n , j , g in this partition. Fix n ≥ N n\geq N n ≥ N , a non-empty element Ω n , j , g \Omega_{n,j,g} Ω n , j , g of the partition, and ξ ∈ Ω n , j , g \xi\in\Omega_{n,j,g} ξ ∈ Ω n , j , g . For any ζ ∈ Ω n , j , g \zeta\in\Omega_{n,j,g} ζ ∈ Ω n , j , g , set y ℓ : = F ℓ ( p ( ζ ) ) y_\ell:=F^\ell(p(\zeta)) y ℓ := F ℓ ( p ( ζ )) . Since y 0 = g ( y n ) y_0=g(y_n) y 0 = g ( y n ) and dist ( y n , x n ) ≤ 2 C ∗ R n \operatorname{dist}(y_n,x_n)\leq2C_*R_n dist ( y n , x n ) ≤ 2 C ∗ R n , assumption (i) gives dist ( y ℓ , x ℓ ) ≤ 2 C ∗ A a e − ε n < δ \operatorname{dist}(y_\ell,x_\ell)\leq2C_*Aa e^{-\varepsilon n}<\delta dist ( y ℓ , x ℓ ) ≤ 2 C ∗ A a e − ε n < δ for 0 ≤ ℓ < n 0\leq\ell<n 0 ≤ ℓ < n , by the choice of a a a . Thus p ( Ω n , j , g ) ⊂ B n ( x 0 ( ξ ) , δ ) p(\Omega_{n,j,g})\subset B_n(x_0(\xi),\delta) p ( Ω n , j , g ) ⊂ B n ( x 0 ( ξ ) , δ ) . Consequently, assumption (ii) gives
m ( Ω n , j , g ) ≤ m ( p − 1 ( B n ( x 0 ( ξ ) , δ ) ) ) = η ( B n ( x 0 ( ξ ) , δ ) ) ≤ C F e − s n . (12) m(\Omega_{n,j,g})\leq m\left(p^{-1}(B_n(x_0(\xi),\delta))\right)=\eta(B_n(x_0(\xi),\delta))\leq C_F e^{-sn}. \tag*{(12)} m ( Ω n , j , g ) ≤ m ( p − 1 ( B n ( x 0 ( ξ ) , δ )) ) = η ( B n ( x 0 ( ξ ) , δ )) ≤ C F e − s n . (12)
Step 2. Comparing consecutive regularizations at fixed parameters. For u ∈ U u\in\mathcal{U} u ∈ U , set
I n ( ξ , u ) : = ∫ u d Q ξ , n and J n ( u ) : = ∫ u d η n = ∫ Ω I n ( ξ , u ) d m ( ξ ) . I_n(\xi,u):=\int u\,dQ_{\xi,n}\qquad\text{and}\qquad J_n(u):=\int u\,d\eta_n=\int_{\Omega}I_n(\xi,u)\,dm(\xi). I n ( ξ , u ) := ∫ u d Q ξ , n and J n ( u ) := ∫ u d η n = ∫ Ω I n ( ξ , u ) d m ( ξ ) .
The density bounds for Q ξ , n Q_{\xi,n} Q ξ , n and η n \eta_n η n , together with the uniform L 1 ( ω k ) L^1(\omega^k) L 1 ( ω k ) bound for U \mathcal{U} U , imply that I n ( ξ , u ) I_n(\xi,u) I n ( ξ , u ) and J n ( u ) J_n(u) J n ( u ) are finite.
In this step, we estimate ∣ I n + 1 ( ξ , u ) − I n ( ξ , u ) ∣ |I_{n+1}(\xi,u)-I_n(\xi,u)| ∣ I n + 1 ( ξ , u ) − I n ( ξ , u ) ∣ for each ξ ∈ Ω \xi\in\Omega ξ ∈ Ω , by expressing the two terms as integrals over a common domain and applying Lemma 2.1.
Step 2.1. An expression for I n + 1 ( ξ , u ) − I n ( ξ , u ) I_{n+1}(\xi,u)-I_n(\xi,u) I n + 1 ( ξ , u ) − I n ( ξ , u ) . Fix n ≥ N n\geq N n ≥ N , a non-empty set Ω n , j , g \Omega_{n,j,g} Ω n , j , g , ξ ∈ Ω n , j , g \xi\in\Omega_{n,j,g} ξ ∈ Ω n , j , g , and u ∈ U u\in\mathcal{U} u ∈ U . We express I n + 1 ( ξ , u ) I_{n+1}(\xi,u) I n + 1 ( ξ , u ) and I n ( ξ , u ) I_n(\xi,u) I n ( ξ , u ) as averages of u ∘ g u\circ g u ∘ g on U n , j U_{n,j} U n , j . We already have Q ξ , n = g ∗ κ x n , t n Q_{\xi,n}=g_*\kappa_{x_n,t_n} Q ξ , n = g ∗ κ x n , t n . To obtain the corresponding expression for Q ξ , n + 1 Q_{\xi,n+1} Q ξ , n + 1 , set b ξ , n : = F n ∘ g ξ , n + 1 b_{\xi,n}:=F^n\circ g_{\xi,n+1} b ξ , n := F n ∘ g ξ , n + 1 on B ( x n + 1 , r 0 e − ε ( n + 1 ) ) B(x_{n+1},r_0e^{-\varepsilon(n+1)}) B ( x n + 1 , r 0 e − ε ( n + 1 ) ) . Then F ∘ b ξ , n = id F\circ b_{\xi,n}=\operatorname{id} F ∘ b ξ , n = id and b ξ , n ( x n + 1 ) = x n b_{\xi,n}(x_{n+1})=x_n b ξ , n ( x n + 1 ) = x n , so b ξ , n b_{\xi,n} b ξ , n is an inverse branch of F F F . By Step 1.2, K ξ , n + 1 K_{\xi,n+1} K ξ , n + 1 is contained in the domain of b ξ , n b_{\xi,n} b ξ , n .
For y ∈ K ξ , n + 1 y\in K_{\xi,n+1} y ∈ K ξ , n + 1 , hypothesis (i) at intermediate time n n n gives
dist ( b ξ , n ( y ) , x n ) ≤ A C ∗ e ε ( n + 1 ) e − λ t n + 1 ≤ A C ∗ t 0 e − 3 ε n < R n / ( 2 C ∗ ) . \operatorname{dist}(b_{\xi,n}(y),x_n)\leq AC_*e^{\varepsilon(n+1)}e^{-\lambda t_{n+1}}\leq AC_*t_0e^{-3\varepsilon n}<R_n/(2C_*). dist ( b ξ , n ( y ) , x n ) ≤ A C ∗ e ε ( n + 1 ) e − λ t n + 1 ≤ A C ∗ t 0 e − 3 ε n < R n / ( 2 C ∗ ) .
Since dist ( x n , ∂ U n , j ( 2 ) ) ≥ R n / C ∗ \operatorname{dist}(x_n,\partial U_{n,j}^{(2)}) \ge R_n/C_* dist ( x n , ∂ U n , j ( 2 ) ) ≥ R n / C ∗ , we obtain b ξ , n ( K ξ , n + 1 ) ⊂ U n , j ( 2 ) b_{\xi,n}(K_{\xi,n+1}) \subset U_{n,j}^{(2)} b ξ , n ( K ξ , n + 1 ) ⊂ U n , j ( 2 ) . In particular, g ∘ b ξ , n g \circ b_{\xi,n} g ∘ b ξ , n is defined on K ξ , n + 1 K_{\xi,n+1} K ξ , n + 1 . Both g ∘ b ξ , n g \circ b_{\xi,n} g ∘ b ξ , n and g ξ , n + 1 g_{\xi,n+1} g ξ , n + 1 are inverse branches of F n + 1 F^{n+1} F n + 1 there and send x n + 1 x_{n+1} x n + 1 to x 0 x_0 x 0 . Uniqueness of inverse germs and connectedness of K ξ , n + 1 K_{\xi,n+1} K ξ , n + 1 therefore give g ξ , n + 1 = g ∘ b ξ , n g_{\xi,n+1}=g \circ b_{\xi,n} g ξ , n + 1 = g ∘ b ξ , n on K ξ , n + 1 K_{\xi,n+1} K ξ , n + 1 .
Set σ ξ , n : = κ x n , t n \sigma_{\xi,n}:=\kappa_{x_n,t_n} σ ξ , n := κ x n , t n and τ ξ , n : = ( b ξ , n ) ∗ κ x n + 1 , t n + 1 \tau_{\xi,n}:=(b_{\xi,n})_*\kappa_{x_{n+1},t_{n+1}} τ ξ , n := ( b ξ , n ) ∗ κ x n + 1 , t n + 1 . The measures σ ξ , n \sigma_{\xi,n} σ ξ , n and τ ξ , n \tau_{\xi,n} τ ξ , n are supported in U n , j ( 2 ) U_{n,j}^{(2)} U n , j ( 2 ) , and the identity g ξ , n + 1 = g ∘ b ξ , n g_{\xi,n+1}=g \circ b_{\xi,n} g ξ , n + 1 = g ∘ b ξ , n gives Q ξ , n = g ∗ σ ξ , n Q_{\xi,n}=g_*\sigma_{\xi,n} Q ξ , n = g ∗ σ ξ , n and Q ξ , n + 1 = g ∗ τ ξ , n Q_{\xi,n+1}=g_*\tau_{\xi,n} Q ξ , n + 1 = g ∗ τ ξ , n . Hence, we have
I n + 1 ( ξ , u ) − I n ( ξ , u ) = ∫ U n , j ( u ∘ g ) d ( τ ξ , n − σ ξ , n ) . I_{n+1}(\xi,u)-I_n(\xi,u)=\int_{U_{n,j}}(u\circ g)\,d(\tau_{\xi,n}-\sigma_{\xi,n}). I n + 1 ( ξ , u ) − I n ( ξ , u ) = ∫ U n , j ( u ∘ g ) d ( τ ξ , n − σ ξ , n ) .
*Step 2.2. Density bounds of σ ξ , n \sigma_{\xi,n} σ ξ , n and τ ξ , n . ∗ W e f i r s t b o u n d t h e d e n s i t i e s o f \tau_{\xi,n}.* We first bound the densities of τ ξ , n . ∗ W e f i r s t b o u n d t h e d e n s i t i eso f \sigma_{\xi,n}$ and τ ξ , n \tau_{\xi,n} τ ξ , n with respect to ω k \omega^k ω k . By (5) , we have d σ ξ , n / d ω k ≤ C t n − 2 k d\sigma_{\xi,n}/d\omega^k\le Ct_n^{-2k} d σ ξ , n / d ω k ≤ C t n − 2 k . Arguing as in (10) , with b ξ , n b_{\xi,n} b ξ , n in place of g ξ , n g_{\xi,n} g ξ , n , we obtain d τ ξ , n / d ω k ≤ C t n + 1 − 2 k ∥ Jac R F ∥ ∞ d\tau_{\xi,n}/d\omega^k\le Ct_{n+1}^{-2k}\|\operatorname{Jac}_{\mathbb{R}}F\|_\infty d τ ξ , n / d ω k ≤ C t n + 1 − 2 k ∥ Jac R F ∥ ∞ .
To compute the densities of σ ξ , n \sigma_{\xi,n} σ ξ , n and τ ξ , n \tau_{\xi,n} τ ξ , n in the coordinates of U n , j U_{n,j} U n , j , write Φ : = Φ c n , j ∣ B ( 0 , 4 R n ) \Phi:=\Phi_{c_{n,j}}|_{B(0,4R_n)} Φ := Φ c n , j ∣ B ( 0 , 4 R n ) and J Φ : = d ( Φ ∗ ω k ) / d V J_\Phi:=d(\Phi^*\omega^k)/dV J Φ := d ( Φ ∗ ω k ) / d V , and set σ ~ ξ , n : = ( Φ − 1 ) ∗ σ ξ , n \widetilde{\sigma}_{\xi,n}:=(\Phi^{-1})_*\sigma_{\xi,n} σ ξ , n := ( Φ − 1 ) ∗ σ ξ , n and τ ~ ξ , n : = ( Φ − 1 ) ∗ τ ξ , n \widetilde{\tau}_{\xi,n}:=(\Phi^{-1})_*\tau_{\xi,n} τ ξ , n := ( Φ − 1 ) ∗ τ ξ , n . The uniform volume comparisons give J Φ ≤ C ∗ J_\Phi\le C_* J Φ ≤ C ∗ . Thus, for d V dV d V -almost every z ∈ B ( 0 , 4 R n ) z\in B(0,4R_n) z ∈ B ( 0 , 4 R n ) , we have
d σ ~ ξ , n d V ( z ) = d σ ξ , n d ω k ( Φ ( z ) ) J Φ ( z ) ≤ C t n − 2 k ≤ C t n + 1 − 2 k , \frac{d\widetilde{\sigma}_{\xi,n}}{dV}(z)=\frac{d\sigma_{\xi,n}}{d\omega^k}(\Phi(z))J_\Phi(z)\le Ct_n^{-2k}\le Ct_{n+1}^{-2k}, d V d σ ξ , n ( z ) = d ω k d σ ξ , n ( Φ ( z )) J Φ ( z ) ≤ C t n − 2 k ≤ C t n + 1 − 2 k ,
d τ ~ ξ , n d V ( z ) = d τ ξ , n d ω k ( Φ ( z ) ) J Φ ( z ) ≤ C ∥ Jac R F ∥ ∞ t n + 1 − 2 k . \frac{d\widetilde{\tau}_{\xi,n}}{dV}(z)=\frac{d\tau_{\xi,n}}{d\omega^k}(\Phi(z))J_\Phi(z)\le C\|\operatorname{Jac}_{\mathbb{R}}F\|_\infty t_{n+1}^{-2k}. d V d τ ξ , n ( z ) = d ω k d τ ξ , n ( Φ ( z )) J Φ ( z ) ≤ C ∥ Jac R F ∥ ∞ t n + 1 − 2 k .
Absorbing the factor max { 1 , ∥ Jac R F ∥ ∞ } \max\{1,\|\operatorname{Jac}_{\mathbb{R}}F\|_\infty\} max { 1 , ∥ Jac R F ∥ ∞ } into the constant C C C , the densities of σ ~ ξ , n \widetilde{\sigma}_{\xi,n} σ ξ , n and τ ~ ξ , n \widetilde{\tau}_{\xi,n} τ ξ , n with respect to d V dV d V are bounded by C t n + 1 − 2 k Ct_{n+1}^{-2k} C t n + 1 − 2 k , with C C C independent of n n n , j j j , and ξ \xi ξ .
Step 2.3. Pluriharmonic cancellations. We next verify that the integrals of pluriharmonic functions against σ ξ , n \sigma_{\xi,n} σ ξ , n and τ ξ , n \tau_{\xi,n} τ ξ , n agree. Let H H H be a real pluriharmonic function on U n , j ( 2 ) U_{n,j}^{(2)} U n , j ( 2 ) . Since b ξ , n ( K ξ , n + 1 ) ⊂ U n , j ( 2 ) b_{\xi,n}(K_{\xi,n+1})\subset U_{n,j}^{(2)} b ξ , n ( K ξ , n + 1 ) ⊂ U n , j ( 2 ) , the function H ∘ b ξ , n H\circ b_{\xi,n} H ∘ b ξ , n is pluriharmonic on a neighbourhood of K ξ , n + 1 K_{\xi,n+1} K ξ , n + 1 . The radial mean value identity (5) gives
∫ H d τ ξ , n = H ( x n ) = ∫ H d σ ξ , n , (13) \int H\,d\tau_{\xi,n}=H(x_n)=\int H\,d\sigma_{\xi,n}, \tag*{(13)} ∫ H d τ ξ , n = H ( x n ) = ∫ H d σ ξ , n , (13)
which proves the claim.
Step 2.4. Current masses. We continue to denote Φ : = Φ c n , j ∣ B ( 0 , 4 R n ) \Phi:=\Phi_{c_{n,j}}|_{B(0,4R_n)} Φ := Φ c n , j ∣ B ( 0 , 4 R n ) . Set T u : = ω + d d c u T_u:=\omega+dd^c u T u := ω + d d c u and ω n , j : = ( Φ − 1 ) ∗ ω e u c l \omega_{n,j}:=(\Phi^{-1})^*\omega_{\mathrm{eucl}} ω n , j := ( Φ − 1 ) ∗ ω eucl , and define
M n , j , g ( u ) : = ∫ U n , j g ∗ ( T u + ω ) ∧ ω n , j k − 1 . (14) M_{n,j,g}(u):=\int_{U_{n,j}}g^*(T_u+\omega)\wedge\omega_{n,j}^{k-1}. \tag*{(14)} M n , j , g ( u ) := ∫ U n , j g ∗ ( T u + ω ) ∧ ω n , j k − 1 . (14)
Since g g g is biholomorphic onto its image, u ∘ g ∘ Φ u\circ g\circ\Phi u ∘ g ∘ Φ is locally integrable on B ( 0 , 4 R n ) B(0,4R_n) B ( 0 , 4 R n ) and d d c ( u ∘ g ∘ Φ ) = Φ ∗ g ∗ T u − Φ ∗ g ∗ ω dd^c(u\circ g\circ\Phi)=\Phi^*g^*T_u-\Phi^*g^*\omega d d c ( u ∘ g ∘ Φ ) = Φ ∗ g ∗ T u − Φ ∗ g ∗ ω . Since g g g is holomorphic on a neighbourhood of U n , j U_{n,j} U n , j , the positive currents Φ ∗ g ∗ T u \Phi^*g^*T_u Φ ∗ g ∗ T u and Φ ∗ g ∗ ω \Phi^*g^*\omega Φ ∗ g ∗ ω have finite mass on B ( 0 , 4 R n ) B(0,4R_n) B ( 0 , 4 R n ) , with
∥ Φ ∗ g ∗ T u ∥ B ( 0 , 4 R n ) + ∥ Φ ∗ g ∗ ω ∥ B ( 0 , 4 R n ) = M n , j , g ( u ) . \|\Phi^*g^*T_u\|_{B(0,4R_n)}+\|\Phi^*g^*\omega\|_{B(0,4R_n)}=M_{n,j,g}(u). ∥ Φ ∗ g ∗ T u ∥ B ( 0 , 4 R n ) + ∥ Φ ∗ g ∗ ω ∥ B ( 0 , 4 R n ) = M n , j , g ( u ) .
Step 2.5. Application of Lemma 2.1 and conclusion. Steps 2.2–2.4 show that the measures σ ~ ξ , n \widetilde{\sigma}_{\xi,n} σ ξ , n and τ ~ ξ , n \widetilde{\tau}_{\xi,n} τ ξ , n are supported in B ( 0 , 2 R n ) B(0,2R_n) B ( 0 , 2 R n ) , have densities at most C t n + 1 − 2 k Ct_{n+1}^{-2k} C t n + 1 − 2 k with respect to d V dV d V , and have equal integrals against every real pluriharmonic function H ~ \widetilde{H} H on B ( 0 , 3 R n ) B(0,3R_n) B ( 0 , 3 R n ) . The last property follows from (13) applied to H = H ~ ∘ Φ − 1 H=\widetilde{H}\circ\Phi^{-1} H = H ∘ Φ − 1 . Moreover, we have d d c ( u ∘ g ∘ Φ ) = Φ ∗ g ∗ T u − Φ ∗ g ∗ ω dd^c(u\circ g\circ\Phi)=\Phi^*g^*T_u-\Phi^*g^*\omega d d c ( u ∘ g ∘ Φ ) = Φ ∗ g ∗ T u − Φ ∗ g ∗ ω , and the sum of the masses of these two positive closed currents on B ( 0 , 4 R n ) B(0,4R_n) B ( 0 , 4 R n ) is M n , j , g ( u ) M_{n,j,g}(u) M n , j , g ( u ) . We may therefore apply Lemma 2.1 with r = R n r=R_n r = R n and M = C t n + 1 − 2 k M=Ct_{n+1}^{-2k} M = C t n + 1 − 2 k . This gives
∣ I n + 1 ( ξ , u ) − I n ( ξ , u ) ∣ ≤ C R n 2 t n + 1 − 2 k M n , j , g ( u ) ≤ C e ( 8 k − 4 ) ε n M n , j , g ( u ) . (15) \left|I_{n+1}(\xi,u)-I_n(\xi,u)\right|\le CR_n^2t_{n+1}^{-2k}M_{n,j,g}(u)\le Ce^{(8k-4)\varepsilon n}M_{n,j,g}(u). \tag*{(15)} ∣ I n + 1 ( ξ , u ) − I n ( ξ , u ) ∣ ≤ C R n 2 t n + 1 − 2 k M n , j , g ( u ) ≤ C e ( 8 k − 4 ) ε n M n , j , g ( u ) . (15)
Observe that the quantity M n , j , g ( u ) M_{n,j,g}(u) M n , j , g ( u ) depends on the fixed branch g g g , not on the choice of ξ ∈ Ω n , j , g \xi\in\Omega_{n,j,g} ξ ∈ Ω n , j , g .
Step 3. A bound for ∣ J n + 1 ( u ) − J n ( u ) ∣ |J_{n+1}(u)-J_n(u)| ∣ J n + 1 ( u ) − J n ( u ) ∣ . In this step we first prove that the quantities M n , j , g ( u ) M_{n,j,g}(u) M n , j , g ( u ) defined in (3.6) satisfy
∑ j ∑ g ∈ G n , j M n , j , g ( u ) ≤ C d ( k − 1 ) n (16) \sum_j\sum_{g\in\mathcal{G}_{n,j}} M_{n,j,g}(u) \le C d^{(k-1)n} \tag*{(16)} j ∑ g ∈ G n , j ∑ M n , j , g ( u ) ≤ C d ( k − 1 ) n (16)
uniformly for u ∈ U u\in U u ∈ U , and use this bound to estimate ∣ J n + 1 ( u ) − J n ( u ) ∣ |J_{n+1}(u)-J_n(u)| ∣ J n + 1 ( u ) − J n ( u ) ∣ .
Step 3.1. The current-mass bound (3.8). For fixed n , j n,j n , j , distinct branches g , h ∈ G n , j g,h\in\mathcal{G}_{n,j} g , h ∈ G n , j have disjoint images. Indeed, if g ( y ) = h ( z ) g(y)=h(z) g ( y ) = h ( z ) for some y , z ∈ U n , j y,z\in U_{n,j} y , z ∈ U n , j , applying F n F^n F n gives y = z y=z y = z . Thus g ( y ) = h ( y ) g(y)=h(y) g ( y ) = h ( y ) , and uniqueness of inverse germs and connectedness of U n , j U_{n,j} U n , j imply g = h g=h g = h on U n , j U_{n,j} U n , j . For every global positive closed ( 1 , 1 ) (1,1) ( 1 , 1 ) -current S S S , we have
∑ g ∈ G n , j g ∗ S ≤ ( F n ) ∗ S ∣ U n , j . (17) \sum_{g\in\mathcal{G}_{n,j}} g^*S\le(F^n)_*S|_{U_{n,j}}. \tag*{(17)} g ∈ G n , j ∑ g ∗ S ≤ ( F n ) ∗ S ∣ U n , j . (17)
Indeed, let Ψ \Psi Ψ be a smooth positive ( k − 1 , k − 1 ) (k-1,k-1) ( k − 1 , k − 1 ) -form compactly supported in U n , j U_{n,j} U n , j , extended by zero to P k \mathbb{P}^k P k . Since Ψ \Psi Ψ is compactly supported in U n , j U_{n,j} U n , j , each g ∗ Ψ g_*\Psi g ∗ Ψ has compact support in g ( U n , j ) g(U_{n,j}) g ( U n , j ) and extends smoothly by zero to P k \mathbb{P}^k P k . On g ( U n , j ) g(U_{n,j}) g ( U n , j ) , we have g ∗ Ψ = ( F n ) ∗ Ψ g_*\Psi=(F^n)^*\Psi g ∗ Ψ = ( F n ) ∗ Ψ . Since the images of distinct branches are disjoint, it follows that ∑ g g ∗ Ψ ≤ ( F n ) ∗ Ψ \sum_g g_*\Psi\le(F^n)^*\Psi ∑ g g ∗ Ψ ≤ ( F n ) ∗ Ψ . Therefore,
∑ g ⟨ g ∗ S , Ψ ⟩ = ⟨ S , ∑ g g ∗ Ψ ⟩ ≤ ⟨ S , ( F n ) ∗ Ψ ⟩ = ⟨ ( F n ) ∗ S , Ψ ⟩ , \sum_g\langle g^*S,\Psi\rangle=\left\langle S,\sum_g g_*\Psi\right\rangle\le\langle S,(F^n)^*\Psi\rangle=\langle(F^n)_*S,\Psi\rangle, g ∑ ⟨ g ∗ S , Ψ ⟩ = ⟨ S , g ∑ g ∗ Ψ ⟩ ≤ ⟨ S , ( F n ) ∗ Ψ ⟩ = ⟨( F n ) ∗ S , Ψ ⟩ ,
which gives (3.9). Observe that no smoothness of S S S is assumed.
Set S : = T u + ω S:=T_u+\omega S := T u + ω . Since ω n , j \omega_{n,j} ω n , j is uniformly comparable to ω \omega ω , (3.9) and the uniformly bounded overlap of the domains U n , j U_{n,j} U n , j give
∑ j ∑ g ∈ G n , j M n , j , g ( u ) ≤ C ∑ j ∫ U n , j ( ∑ g ∈ G n , j g ∗ S ) ∧ ω k − 1 ≤ C ∑ j ∫ U n , j ( F n ) ∗ S ∧ ω k − 1 ≤ C ∫ P k ( F n ) ∗ S ∧ ω k − 1 = 2 C d ( k − 1 ) n . \begin{aligned}
\sum_j\sum_{g\in\mathcal{G}_{n,j}}M_{n,j,g}(u)\le C\sum_j\int_{U_{n,j}}\left(\sum_{g\in\mathcal{G}_{n,j}}g^*S\right)\wedge\omega^{k-1} \\
&\le C\sum_j\int_{U_{n,j}}(F^n)_*S\wedge\omega^{k-1}\le C\int_{\mathbb{P}^k}(F^n)_*S\wedge\omega^{k-1}=2Cd^{(k-1)n}.
\end{aligned} j ∑ g ∈ G n , j ∑ M n , j , g ( u ) ≤ C j ∑ ∫ U n , j g ∈ G n , j ∑ g ∗ S ∧ ω k − 1 ≤ C j ∑ ∫ U n , j ( F n ) ∗ S ∧ ω k − 1 ≤ C ∫ P k ( F n ) ∗ S ∧ ω k − 1 = 2 C d ( k − 1 ) n .
The equality follows from the projection formula and cohomology, using [ T u + ω ] = 2 [ ω ] [T_u+\omega]=2[\omega] [ T u + ω ] = 2 [ ω ] , ( F n ) ∗ [ ω k − 1 ] = d ( k − 1 ) n [ ω k − 1 ] (F^n)^*[\omega^{k-1}]=d^{(k-1)n}[\omega^{k-1}] ( F n ) ∗ [ ω k − 1 ] = d ( k − 1 ) n [ ω k − 1 ] , and ∫ P k ω k = 1 \int_{\mathbb{P}^k}\omega^k=1 ∫ P k ω k = 1 . This proves (3.8). Observe that the sum is controlled by current growth, with no factor for the number of branches or domains U n , j U_{n,j} U n , j .
Step 3.2. Estimating ∣ J n + 1 ( u ) − J n ( u ) ∣ |J_{n+1}(u)-J_n(u)| ∣ J n + 1 ( u ) − J n ( u ) ∣ . We now combine (3.8) with (3.4) to bound ∣ J n + 1 ( u ) − J n ( u ) ∣ |J_{n+1}(u)-J_n(u)| ∣ J n + 1 ( u ) − J n ( u ) ∣ . Set a 0 : = s − ( k − 1 ) log d − ( 8 k − 4 ) ε > 0 a_0:=s-(k-1)\log d-(8k-4)\varepsilon>0 a 0 := s − ( k − 1 ) log d − ( 8 k − 4 ) ε > 0 . Integrating (3.7) over each element Ω n , j , g \Omega_{n,j,g} Ω n , j , g of the partition and summing, then using (3.4) and (3.8), yields, for every n ≥ N n\ge N n ≥ N ,
∣ J n + 1 ( u ) − J n ( u ) ∣ ≤ C e ( 8 k − 4 ) ε n ∑ j , g m ( Ω n , j , g ) M n , j , g ( u ) ≤ C e − a 0 n . (18) |J_{n+1}(u)-J_n(u)|\le Ce^{(8k-4)\varepsilon n}\sum_{j,g}m(\Omega_{n,j,g})M_{n,j,g}(u)\le Ce^{-a_0n}. \tag*{(18)} ∣ J n + 1 ( u ) − J n ( u ) ∣ ≤ C e ( 8 k − 4 ) ε n j , g ∑ m ( Ω n , j , g ) M n , j , g ( u ) ≤ C e − a 0 n . (18)
Step 4. Hölder regularity. Together with the equality of the masses and the weak convergence established in Step 1, the density bound (3.3) and the comparison estimate (3.10) show that the hypotheses of Lemma 2.2 hold with a = a 0 a=a_0 a = a 0 and b = b 0 b=b_0 b = b 0 . Therefore η \eta η integrates every quasi-plurisubharmonic function and satisfies (3.1) with
β = a 0 a 0 + b 0 ∈ ( 0 , 1 ) . \beta=\frac{a_0}{a_0+b_0}\in(0,1). β = a 0 + b 0 a 0 ∈ ( 0 , 1 ) .
Since a 0 = s − ( k − 1 ) log d − ( 8 k − 4 ) ε a_0=s-(k-1)\log d-(8k-4)\varepsilon a 0 = s − ( k − 1 ) log d − ( 8 k − 4 ) ε and b 0 = 8 k ε + log max { 1 , ∥ Jac R F ∥ ∞ } b_0=8k\varepsilon+\log\max\{1,\lVert\operatorname{Jac}_{\mathbb{R}}F\rVert_\infty\} b 0 = 8 k ε + log max { 1 , ∥ Jac R F ∥ ∞ } , the exponent has the desired dependence. The final statement about the super-potential follows from (3.1) under the identification recalled after Theorem 1.1. □
Applications of the regularization criterion
In this section, we apply Theorem 3.1 to prove Theorems 1.1 and 1.4 , and deduce Corollaries 1.2 and 1.3. We also give a direct proof in dimension one and an example showing that global quasi-plurisubharmonic integrability can fail.
Properties of large entropy measures
Proof of Theorem 1.1 . We first construct a submeasure of μ \mu μ to which Theorem 3.1 applies. Let χ min > 0 \chi_{\min}>0 χ m i n > 0 be the smallest Lyapunov exponent of μ \mu μ ; see § 2.2 . Choose ( k − 1 ) log d < s < h μ ( F ) (k-1)\log d<s<h_{\mu}(F) ( k − 1 ) log d < s < h μ ( F ) and 0 < ε < χ min / 2 0<\varepsilon<\chi_{\min}/2 0 < ε < χ m i n /2 with ( 8 k − 4 ) ε < s − ( k − 1 ) log d (8k-4)\varepsilon<s-(k-1)\log d ( 8 k − 4 ) ε < s − ( k − 1 ) log d . Fix 0 < θ < 1 0<\theta<1 0 < θ < 1 , apply Lemma 2.3 with mass error θ / 4 \theta/4 θ /4 , and let E ^ \widehat{E} E be the resulting Borel set. Set Ω : = E ^ \Omega:=\widehat{E} Ω := E , m : = μ ^ ∣ E ^ m:=\widehat{\mu}|_{\widehat{E}} m := μ ∣ E , p : = π 0 ∣ E ^ p:=\pi_{0}|_{\widehat{E}} p := π 0 ∣ E , and η : = p ∗ m \eta:=p_{*}m η := p ∗ m . Then η ≤ μ \eta\leq\mu η ≤ μ and η ( P k ) > 1 − θ / 4 \eta(\mathbb{P}^{k})>1-\theta/4 η ( P k ) > 1 − θ /4 . Lemma 2.3 (i) gives the branch estimates, and (ii) gives η ( B n ( x 0 , δ ) ) ≤ μ ( B n ( x 0 , δ ) ) ≤ e − s n \eta(B_{n}(x_{0},\delta))\leq\mu(B_{n}(x_{0},\delta))\leq e^{-sn} η ( B n ( x 0 , δ )) ≤ μ ( B n ( x 0 , δ )) ≤ e − s n . Theorem 3.1 gives integrability of quasi-plurisubharmonic functions and the absolute Hölder estimate (3.1) for η \eta η . Since s s s and ε \varepsilon ε were chosen independently of θ \theta θ , so is the exponent β \beta β .
We now choose a compact set E θ E_{\theta} E θ with μ ( E θ ) > 1 − θ \mu(E_{\theta})>1-\theta μ ( E θ ) > 1 − θ such that 1 E θ μ ≤ 2 η \mathbf{1}_{E_{\theta}}\mu\leq2\eta 1 E θ μ ≤ 2 η . Choose a Borel density f f f with η = f μ \eta=f\mu η = f μ and 0 ≤ f ≤ 1 0\leq f\leq1 0 ≤ f ≤ 1 , and denote A θ : = { f ≥ 1 / 2 } A_{\theta}:=\{f\geq1/2\} A θ := { f ≥ 1/2 } . Since 1 − f > 1 / 2 1-f>1/2 1 − f > 1/2 on A θ c A_{\theta}^{c} A θ c , we have μ ( A θ c ) ≤ 2 ∫ P k ( 1 − f ) d μ < θ / 2 \mu(A_{\theta}^{c})\leq2\int_{\mathbb{P}^{k}}(1-f)\,d\mu<\theta/2 μ ( A θ c ) ≤ 2 ∫ P k ( 1 − f ) d μ < θ /2 . Since μ ( A θ ) > 1 − θ / 2 \mu(A_{\theta})>1-\theta/2 μ ( A θ ) > 1 − θ /2 , inner regularity gives a compact set E θ ⊂ A θ E_{\theta}\subset A_{\theta} E θ ⊂ A θ with μ ( E θ ) > 1 − θ \mu(E_{\theta})>1-\theta μ ( E θ ) > 1 − θ . The domination η θ : = 1 E θ μ ≤ 2 η \eta_{\theta}:=\mathbf{1}_{E_{\theta}}\mu\leq2\eta η θ := 1 E θ μ ≤ 2 η gives integrability and ∫ ∣ u − v ∣ d η θ ≤ 2 ∫ ∣ u − v ∣ d η \int|u-v|\,d\eta_{\theta}\leq2\int|u-v|\,d\eta ∫ ∣ u − v ∣ d η θ ≤ 2 ∫ ∣ u − v ∣ d η . Thus the absolute Hölder estimate holds for every η θ \eta_{\theta} η θ with the same exponent β \beta β . □
Remark 4.1. Set Γ F : = log max { 1 , ∥ Jac R F ∥ ∞ } \Gamma_{F}:=\log\max\{1,\|\operatorname{Jac}_{\mathbb{R}}F\|_{\infty}\} Γ F := log max { 1 , ∥ Jac R F ∥ ∞ } . In Theorem 1.1 , every exponent
0 < β < h μ ( F ) − ( k − 1 ) log d h μ ( F ) − ( k − 1 ) log d + Γ F 0<\beta<\frac{h_{\mu}(F)-(k-1)\log d}{h_{\mu}(F)-(k-1)\log d+\Gamma_{F}} 0 < β < h μ ( F ) − ( k − 1 ) log d + Γ F h μ ( F ) − ( k − 1 ) log d
can be used for all 0 < θ < 1 0<\theta<1 0 < θ < 1 , with a possibly θ \theta θ -dependent constant C θ C_{\theta} C θ . Indeed, Step 4 of the proof of Theorem 3.1 gives the exponent β 0 = ( s − ( k − 1 ) log d − ( 8 k − 4 ) ε ) / ( s − ( k − 1 ) log d + 4 ε + Γ F ) \beta_{0}=(s-(k-1)\log d-(8k-4)\varepsilon)/(s-(k-1)\log d+4\varepsilon+\Gamma_{F}) β 0 = ( s − ( k − 1 ) log d − ( 8 k − 4 ) ε ) / ( s − ( k − 1 ) log d + 4 ε + Γ F ) . Choose s < h μ ( F ) s<h_{\mu}(F) s < h μ ( F ) sufficiently close to h μ ( F ) h_{\mu}(F) h μ ( F ) and then ε > 0 \varepsilon>0 ε > 0 sufficiently small, independently of θ \theta θ , so that the conditions on s s s and ε \varepsilon ε in the proof of Theorem 1.1 hold and β 0 > β \beta_{0}>\beta β 0 > β . The estimate with exponent β 0 \beta_{0} β 0 implies the estimate with exponent β \beta β , since max { t , t β 0 } ≤ max { t , t β } \max\{t,t^{\beta_{0}}\}\leq\max\{t,t^{\beta}\} max { t , t β 0 } ≤ max { t , t β } for t ≥ 0 t\geq0 t ≥ 0 .
Proof of Corollary 1.2. We prove (i) using integrability of the restrictions of μ \mu μ to sets of measure arbitrarily close to one. Let P P P be a Borel pluripolar set. Choose u ∈ U u\in\mathcal{U} u ∈ U with P ⊂ { u = − ∞ } P\subset\{u=-\infty\} P ⊂ { u = − ∞ } . Integrability of u u u with respect to the restrictions in Theorem 1.1 gives η θ ( P ) = 0 \eta_{\theta}(P)=0 η θ ( P ) = 0 and hence μ ( P ) ≤ ( μ − η θ ) ( P k ) < θ \mu(P)\leq(\mu-\eta_{\theta})(\mathbb{P}^{k})<\theta μ ( P ) ≤ ( μ − η θ ) ( P k ) < θ for every 0 < θ < 1 0<\theta<1 0 < θ < 1 . Letting θ → 0 \theta\to0 θ → 0 gives μ ( P ) = 0 \mu(P)=0 μ ( P ) = 0 .
We deduce (ii) by ergodic decomposition and the variational principle. If ν ( P ) = 1 \nu(P)=1 ν ( P ) = 1 , then almost every component ρ \rho ρ in the ergodic decomposition of ν \nu ν satisfies ρ ( P ) = 1 \rho(P)=1 ρ ( P ) = 1 , so (i) gives h ρ ( F ) ≤ ( k − 1 ) log d h_{\rho}(F)\leq(k-1)\log d h ρ ( F ) ≤ ( k − 1 ) log d . Integrating the component entropies gives h ν ( F ) ≤ ( k − 1 ) log d h_{\nu}(F)\leq(k-1)\log d h ν ( F ) ≤ ( k − 1 ) log d . If K K K is non-empty, compact, forward invariant, and pluripolar, every invariant probability measure for F ∣ K F|_{K} F ∣ K satisfies this entropy bound. The variational principle therefore gives h t o p ( F ∣ K ) ≤ ( k − 1 ) log d h_{\mathrm{top}}(F|_{K})\leq(k-1)\log d h top ( F ∣ K ) ≤ ( k − 1 ) log d . □
Proof of Corollary 1.3. We prove exponential integrability with an exponent independent of θ \theta θ , using the truncation argument in [19 proof of Proposition 3.9] . Skoda’s integrability theorem [45 ] , in the uniform form of [24 Theorem A.22] , gives constants c , C > 0 c,C>0 c , C > 0 such that ∥ max ( u , − t ) ∥ 1 ≤ C e − c t \| \max(u,-t)\|_{1}\leq Ce^{-ct} ∥ max ( u , − t ) ∥ 1 ≤ C e − c t for u ∈ U u\in\mathcal{U} u ∈ U and t ≥ 1 t\geq1 t ≥ 1 . Apply (1.1) to u u u and u t : = max ( u , − t ) u_{t}:=\max(u,-t) u t := max ( u , − t ) . For all sufficiently large t t t , we have
η θ ( { u < − t − 1 } ) ≤ ∫ P k ( u t − u ) d η θ ≤ C θ e − c β t . \eta_{\theta}(\{u<-t-1\})\leq\int_{\mathbb{P}^{k}}(u_{t}-u)\,d\eta_{\theta}\leq C_{\theta}e^{-c\beta t}. η θ ({ u < − t − 1 }) ≤ ∫ P k ( u t − u ) d η θ ≤ C θ e − c β t .
Integration of the sublevel bound gives sup u ∈ U ∫ P k e − α u d η θ < ∞ \sup_{u\in\mathcal{U}}\int_{\mathbb{P}^{k}}e^{-\alpha u}\,d\eta_{\theta}<\infty sup u ∈ U ∫ P k e − α u d η θ < ∞ for every 0 < α < c β 0<\alpha<c\beta 0 < α < c β . Both c c c and β \beta β are independent of θ \theta θ , so the same α \alpha α works for all restrictions.
We obtain the Monge–Ampère representation with a common Hölder exponent by applying [19 Theorem 1.1] to η θ / η θ ( P k ) \eta_\theta/\eta_\theta(\mathbb{P}^k) η θ / η θ ( P k ) . Normalization preserves the super-potential exponent β \beta β in (1.1). The exponent of the resulting potential depends only on k k k and β \beta β , so it can be chosen independently of θ \theta θ . □
Frostman measures and Bowen entropy
Since uniform expansion gives inverse branches on fixed-radius balls, a uniform Bowen ball mass bound suffices to apply Theorem 3.1 .
Proposition 4.2. Let F : P k → P k F:\mathbb{P}^k\to\mathbb{P}^k F : P k → P k be a holomorphic endomorphism of algebraic degree d ≥ 2 d\ge2 d ≥ 2 , with k ≥ 1 k\ge1 k ≥ 1 , and let K ⊂ P k K\subset\mathbb{P}^k K ⊂ P k be non-empty and compact, with F ( K ) ⊂ K F(K)\subset K F ( K ) ⊂ K . Suppose that F F F is uniformly expanding on K K K . Let σ \sigma σ be a Borel probability measure on P k \mathbb{P}^k P k , not necessarily invariant, with σ ( K ) = 1 \sigma(K)=1 σ ( K ) = 1 . If there are δ , C F > 0 \delta,C_F>0 δ , C F > 0 , an integer N ≥ 1 N\ge1 N ≥ 1 , and s > ( k − 1 ) log d s>(k-1)\log d s > ( k − 1 ) log d such that σ ( B n ( x , δ ) ) ≤ C F e − s n \sigma(B_n(x,\delta))\le C_F e^{-sn} σ ( B n ( x , δ )) ≤ C F e − s n for every x ∈ K x\in K x ∈ K and n ≥ N n\ge N n ≥ N , then σ \sigma σ has a Hölder continuous super-potential. In particular, it integrates every quasi-plurisubharmonic function and is moderate.
Proof. We first obtain uniform inverse branches along orbits in K K K . Uniform expansion gives an integer m ≥ 1 m\ge1 m ≥ 1 with sup x ∈ K ∥ ( D ( F m ) x ) − 1 ∥ < 1 \sup_{x\in K}\|(D(F^m)_x)^{-1}\|<1 sup x ∈ K ∥ ( D ( F m ) x ) − 1 ∥ < 1 . Compactness of K K K and the inverse function theorem give local inverse branches of F m F^m F m on balls of a common radius, with a common contraction factor strictly less than one. The inverse branches of the remaining iterates F r F^r F r , 0 ≤ r < m 0\le r<m 0 ≤ r < m , have uniformly bounded Lipschitz constants on smaller balls of a common radius. Writing n = q m + r n=qm+r n = q m + r , with q ≥ 0 q\ge0 q ≥ 0 and 0 ≤ r < m 0\le r<m 0 ≤ r < m , we compose the inverse branches of F m F^m F m with the remaining inverse branches of F r F^r F r . Absorbing the bounds for 0 ≤ r < m 0\le r<m 0 ≤ r < m into a constant gives r 0 , λ > 0 r_0,\lambda>0 r 0 , λ > 0 and A ≥ 1 A\ge1 A ≥ 1 such that g x , n g_{x,n} g x , n , the inverse branch sending F n x F^n x F n x to x x x , is defined on B ( F n x , r 0 ) B(F^n x,r_0) B ( F n x , r 0 ) and satisfies Lip ( F j ∘ g x , n ) ≤ A e − λ ( n − j ) \operatorname{Lip}(F^j\circ g_{x,n})\le Ae^{-\lambda(n-j)} Lip ( F j ∘ g x , n ) ≤ A e − λ ( n − j ) for x ∈ K x\in K x ∈ K and 0 ≤ j ≤ n 0\le j\le n 0 ≤ j ≤ n .
We now apply Theorem 3.1 with Ω = K \Omega=K Ω = K , parameter measure σ ∣ K \sigma|_K σ ∣ K , and p p p the inclusion. Choose ε > 0 \varepsilon>0 ε > 0 with ( 8 k − 4 ) ε < s − ( k − 1 ) log d (8k-4)\varepsilon<s-(k-1)\log d ( 8 k − 4 ) ε < s − ( k − 1 ) log d . Restricting the branches to B ( F n x , r 0 e − ε n ) B(F^n x,r_0e^{-\varepsilon n}) B ( F n x , r 0 e − ε n ) gives hypothesis (i), and the assumed Bowen ball bound gives (ii). Theorem 3.1 then gives integrability and the absolute Hölder estimate for σ \sigma σ , and [19 Proposition 3.9] gives moderateness. □
Following [32 ] , we use variable length covers to measure the orbit complexity of sets that need not be compact or invariant. For Z ⊂ P k Z\subset\mathbb{P}^k Z ⊂ P k , s ≥ 0 s\ge0 s ≥ 0 , δ > 0 \delta>0 δ > 0 , and N ≥ 1 N\ge1 N ≥ 1 , denote
M N , δ s ( Z ) : = inf { ∑ i e − s n i : Z ⊂ ⋃ i B n i ( x i , δ ) , n i ≥ N } . \mathcal{M}_{N,\delta}^{s}(Z):=\inf\left\{\sum_i e^{-sn_i}:Z\subset\bigcup_i B_{n_i}(x_i,\delta),\ n_i\ge N\right\}. M N , δ s ( Z ) := inf { i ∑ e − s n i : Z ⊂ i ⋃ B n i ( x i , δ ) , n i ≥ N } .
where the infimum is over countable covers with x i ∈ P k x_i\in\mathbb{P}^k x i ∈ P k . Set M δ s ( Z ) : = lim N → ∞ M N , δ s ( Z ) \mathcal{M}_{\delta}^{s}(Z):=\lim_{N\to\infty}\mathcal{M}_{N,\delta}^{s}(Z) M δ s ( Z ) := lim N → ∞ M N , δ s ( Z ) and define the Bowen entropy by h B ( F , Z ) : = lim δ → 0 inf { s ≥ 0 : M δ s ( Z ) = 0 } h_{\mathrm{B}}(F,Z):=\lim_{\delta\to0}\inf\{s\ge0:\mathcal{M}_{\delta}^{s}(Z)=0\} h B ( F , Z ) := lim δ → 0 inf { s ≥ 0 : M δ s ( Z ) = 0 } . Bowen entropy agrees with topological entropy on compact invariant sets and is countably stable. The use of variable lengths distinguishes Bowen entropy from upper capacity entropy, which uses a common orbit length; see [32 Section 2] .
Proof of Theorem 1.4. Suppose on the contrary that h B ( F , P ∩ K ) > ( k − 1 ) log d h_{\mathrm{B}}(F,P\cap K)>(k-1)\log d h B ( F , P ∩ K ) > ( k − 1 ) log d . Since P ∩ K P\cap K P ∩ K is Borel and h t o p ( F ) < ∞ h_{\mathrm{top}}(F)<\infty h top ( F ) < ∞ , the compact approximation theorem [32 Theorem 1.2 (ii)] gives a compact set L ⊂ P ∩ K L\subset P\cap K L ⊂ P ∩ K with h B ( F , L ) > ( k − 1 ) log d h_{\mathrm{B}}(F,L)>(k-1)\log d h B ( F , L ) > ( k − 1 ) log d . Choose ( k − 1 ) log d < s < s + γ < h B ( F , L ) (k-1)\log d<s<s+\gamma<h_{\mathrm{B}}(F,L) ( k − 1 ) log d < s < s + γ < h B ( F , L ) . For sufficiently small δ > 0 \delta>0 δ > 0 and sufficiently large N N N , we have M N , δ s + γ ( L ) > 0 \mathcal{M}_{N,\delta}^{s+\gamma}(L)>0 M N , δ s + γ ( L ) > 0 . Applying [32 Proposition 3.2] and [32 Lemma 3.4] , we obtain a Borel probability measure σ \sigma σ with σ ( L ) = 1 \sigma(L)=1 σ ( L ) = 1 and σ ( B n ( x , δ ) ) ≤ C F e − s n \sigma(B_n(x,\delta))\le C_F e^{-sn} σ ( B n ( x , δ )) ≤ C F e − s n for every x ∈ P k x\in\mathbb{P}^k x ∈ P k and n ≥ N n\ge N n ≥ N , for some C F > 0 C_F>0 C F > 0 .
Since σ ( K ) = 1 \sigma(K)=1 σ ( K ) = 1 , Proposition 4.2 implies that every u ∈ U u\in\mathcal{U} u ∈ U is σ \sigma σ -integrable. Choose u ∈ U u\in\mathcal{U} u ∈ U with P ⊂ { u = − ∞ } P\subset\{u=-\infty\} P ⊂ { u = − ∞ } . The equality σ ( L ) = 1 \sigma(L)=1 σ ( L ) = 1 , with L ⊂ P L\subset P L ⊂ P , forces ∫ P k ( − u ) d σ = ∞ \int_{\mathbb{P}^k}(-u)\,d\sigma=\infty ∫ P k ( − u ) d σ = ∞ , which is a contradiction. □
The one-dimensional case
In dimension one, Theorem 1.1 is an immediate consequence of classical dimension theory and potential theory. Let F : P 1 → P 1 F:\mathbb{P}^{1}\to\mathbb{P}^{1} F : P 1 → P 1 be a rational map of degree at least two and let μ \mu μ be an ergodic invariant probability measure with h μ ( F ) > 0 h_{\mu}(F)>0 h μ ( F ) > 0 . By Mañé’s dimension formula [42 Theorem A] , the measure μ \mu μ is exact-dimensional, with dim μ = h μ ( F ) / χ μ > 0 \dim\mu=h_{\mu}(F)/\chi_{\mu}>0 dim μ = h μ ( F ) / χ μ > 0 , where χ μ \chi_{\mu} χ μ is its Lyapunov exponent. Fix 0 < a < dim μ 0<a<\dim\mu 0 < a < dim μ . Exact dimensionality implies that, for every 0 < θ < 1 0<\theta<1 0 < θ < 1 , there are a compact set E θ ⊂ P 1 E_{\theta}\subset\mathbb{P}^{1} E θ ⊂ P 1 , with μ ( E θ ) > 1 − θ \mu(E_{\theta})>1-\theta μ ( E θ ) > 1 − θ , and a constant C θ > 0 C_{\theta}>0 C θ > 0 such that μ ( B ( x , r ) ) ≤ C θ r a \mu(B(x,r))\leq C_{\theta}r^{a} μ ( B ( x , r )) ≤ C θ r a for every x ∈ E θ x\in E_{\theta} x ∈ E θ and r > 0 r>0 r > 0 . For η θ : = 1 E θ μ \eta_{\theta}:=\mathbf{1}_{E_{\theta}}\mu η θ := 1 E θ μ , the same estimate, after changing C θ C_{\theta} C θ , holds for balls with arbitrary centers: if B ( y , r ) B(y,r) B ( y , r ) meets E θ E_{\theta} E θ , choose x ∈ E θ ∩ B ( y , r ) x\in E_{\theta}\cap B(y,r) x ∈ E θ ∩ B ( y , r ) and use B ( y , r ) ⊂ B ( x , 2 r ) B(y,r)\subset B(x,2r) B ( y , r ) ⊂ B ( x , 2 r ) .
By [19 Proposition 3.8] , the measure η θ \eta_{\theta} η θ has a Hölder continuous super-potential. Since the exponent in that proposition depends only on a a a , it can be chosen independently of θ \theta θ . The estimate (1.1) then follows from [19 Lemma 3.5] .
In dimension one, Theorem 1.4 holds without uniform expansion. Fix L > max { 1 , Lip ( F ) } L>\max\{1,\operatorname{Lip}(F)\} L > max { 1 , Lip ( F )} . Polar sets have Hausdorff dimension zero [2 Theorem 5.9.6] , and comparison of ordinary ball covers with Bowen ball covers gives h B ( F , P ) ≤ ( log L ) dim H P = 0 h_{B}(F,P)\leq(\log L)\dim_{H}P=0 h B ( F , P ) ≤ ( log L ) dim H P = 0 for every Borel polar set P ⊂ P 1 P\subset\mathbb{P}^{1} P ⊂ P 1 , where dim H \dim_{H} dim H denotes Hausdorff dimension.
Failure of global integrability
The following standard renewal example, included for completeness, shows that global integrability can fail already in dimension one, even for mixing measures with entropy arbitrarily close to the maximum.
Proposition 4.3. Let Q : P 1 → P 1 Q:\mathbb{P}^{1}\to\mathbb{P}^{1} Q : P 1 → P 1 be given by Q [ Z 0 : Z 1 ] = [ Z 0 2 : Z 1 2 ] Q[Z_{0}:Z_{1}]=[Z_{0}^{2}:Z_{1}^{2}] Q [ Z 0 : Z 1 ] = [ Z 0 2 : Z 1 2 ] and set
u ( [ Z 0 : Z 1 ] ) : = log ∣ Z 1 − Z 0 ∣ 2 ( ∣ Z 0 ∣ 2 + ∣ Z 1 ∣ 2 ) 1 / 2 . u([Z_{0}:Z_{1}]):=\log\frac{\lvert Z_{1}-Z_{0}\rvert}{\sqrt{2}\left(\lvert Z_{0}\rvert^{2}+\lvert Z_{1}\rvert^{2}\right)^{1/2}}. u ([ Z 0 : Z 1 ]) := log 2 ( ∣ Z 0 ∣ 2 + ∣ Z 1 ∣ 2 ) 1/2 ∣ Z 1 − Z 0 ∣ . Then u ∈ U u\in\mathcal{U} u ∈ U . For every ε > 0 \varepsilon>0 ε > 0 , there is a mixing Q Q Q -invariant probability measure ν \nu ν supported on the uniformly expanding repeller S 1 : = { w ∈ C : ∣ w ∣ = 1 } \mathbb{S}^{1}:=\{w\in\mathbb{C}:\lvert w\rvert=1\} S 1 := { w ∈ C : ∣ w ∣ = 1 } , where w = Z 1 / Z 0 w=Z_{1}/Z_{0} w = Z 1 / Z 0 , satisfying
h ν ( Q ) > log 2 − ε , ν { u = − ∞ } = 0 , and ∫ P 1 ( − u ) d ν = ∞ . h_{\nu}(Q)>\log2-\varepsilon,\qquad\nu\{u=-\infty\}=0,\qquad\text{and}\qquad\int_{\mathbb{P}^{1}}(-u)\,d\nu=\infty. h ν ( Q ) > log 2 − ε , ν { u = − ∞ } = 0 , and ∫ P 1 ( − u ) d ν = ∞.
Proof. The statement follows from a standard stationary-renewal construction; see [31 Chapter XI] for the equilibrium renewal process and [1 ] for the entropy formula for induced transformations. Denote N 0 : = { 0 , 1 , 2 , … } \mathbb{N}_{0}:=\{0,1,2,\ldots\} N 0 := { 0 , 1 , 2 , … } . For 0 < δ < 1 0<\delta<1 0 < δ < 1 , let q = ( q ℓ ) ℓ ≥ 0 q=(q_{\ell})_{\ell\geq0} q = ( q ℓ ) ℓ ≥ 0 be the probability distribution on N 0 \mathbb{N}_{0} N 0 defined by q ℓ : = ( 1 − δ ) 2 − ( ℓ + 1 ) + δ c ( ℓ + 1 ) − 3 q_{\ell}:=(1-\delta)2^{-(\ell+1)}+\delta c(\ell+1)^{-3} q ℓ := ( 1 − δ ) 2 − ( ℓ + 1 ) + δ c ( ℓ + 1 ) − 3 , where c − 1 : = ∑ ℓ ≥ 0 ( ℓ + 1 ) − 3 c^{-1}:=\sum_{\ell\geq0}(\ell+1)^{-3} c − 1 := ∑ ℓ ≥ 0 ( ℓ + 1 ) − 3 . Set
L ‾ : = ∑ ℓ ≥ 0 ( ℓ + 1 ) q ℓ . \overline{L}:=\sum_{\ell\geq0}(\ell+1)q_{\ell}. L := ℓ ≥ 0 ∑ ( ℓ + 1 ) q ℓ .
Let ( X j ) j ≥ 0 (X_{j})_{j\geq0} ( X j ) j ≥ 0 be the stationary Markov chain on N 0 \mathbb{N}_{0} N 0 with deterministic transitions j ↦ j − 1 j\mapsto j-1 j ↦ j − 1 for j ≥ 1 j\geq1 j ≥ 1 and transition probabilities Pr ( 0 ↦ ℓ ) = q ℓ \Pr(0\mapsto\ell)=q_{\ell} Pr ( 0 ↦ ℓ ) = q ℓ . Let ϱ δ \varrho_{\delta} ϱ δ be the law on { 0 , 1 } N 0 \{0,1\}^{\mathbb{N}_{0}} { 0 , 1 } N 0 of a j : = 1 { X j = 0 } a_{j}:=\mathbf{1}_{\{X_{j}=0\}} a j := 1 { X j = 0 } . This is a stationary binary renewal process in which the numbers of zeros between successive 1’s are independent and have distribution q q q . This chain is irreducible, aperiodic, and positive recurrent; hence it is mixing by [43 Theorem 1.8.3] . Therefore ϱ δ \varrho_{\delta} ϱ δ is mixing. For a = ( a j ) j ≥ 0 a=(a_{j})_{j\geq0} a = ( a j ) j ≥ 0 , set R ( a ) : = min { j ≥ 0 : a j = 1 } R(a):=\min\{j\geq0:a_{j}=1\} R ( a ) := min { j ≥ 0 : a j = 1 } , with R ( a ) = ∞ R(a)=\infty R ( a ) = ∞ if no such index exists. Standard renewal theory gives
ϱ δ ( { R = j } ) = 1 L ‾ ∑ ℓ ≥ j q ℓ , hence E ϱ δ R = 1 2 L ‾ ∑ ℓ ≥ 0 ℓ ( ℓ + 1 ) q ℓ = ∞ . \varrho_{\delta}(\{R=j\})=\frac{1}{\overline{L}}\sum_{\ell\geq j}q_{\ell},\qquad\text{hence}\qquad\mathbb{E}_{\varrho_{\delta}}R=\frac{1}{2\overline{L}}\sum_{\ell\geq0}\ell(\ell+1)q_{\ell}=\infty. ϱ δ ({ R = j }) = L 1 ℓ ≥ j ∑ q ℓ , hence E ϱ δ R = 2 L 1 ℓ ≥ 0 ∑ ℓ ( ℓ + 1 ) q ℓ = ∞.
On the other hand, Abramov’s formula [1 ] gives h ϱ δ ( σ ) = H ( q ) / L ‾ h_{\varrho_{\delta}}(\sigma)=H(q)/\overline{L} h ϱ δ ( σ ) = H ( q ) / L , where σ \sigma σ is the left shift on { 0 , 1 } N 0 \{0,1\}^{\mathbb{N}_{0}} { 0 , 1 } N 0 and H ( q ) : = − ∑ ℓ ≥ 0 q ℓ log q ℓ H(q):=-\sum_{\ell\geq0}q_{\ell}\log q_{\ell} H ( q ) := − ∑ ℓ ≥ 0 q ℓ log q ℓ . As δ → 0 \delta\to0 δ → 0 , we have L ‾ → 2 \overline{L}\to2 L → 2 . Moreover, the geometric distribution has entropy 2 log 2 2\log2 2 log 2 , and the polynomial-tail distribution has finite entropy. The entropy bounds for convex mixtures therefore give H ( q ) → 2 log 2 H(q) \to2\log2 H ( q ) → 2 log 2 . Hence we have h ϱ δ ( σ ) → log 2 h_{\varrho_\delta}(\sigma) \to\log2 h ϱ δ ( σ ) → log 2 . Let
Π ( a ) = exp ( 2 π i ∑ j ≥ 0 a j 2 − ( j + 1 ) ) \Pi(a)=\exp\left(2\pi i\sum_{j\geq0}a_j2^{-(j+1)}\right) Π ( a ) = exp ( 2 π i j ≥ 0 ∑ a j 2 − ( j + 1 ) )
be the binary coding and set ν δ : = Π ∗ ϱ δ \nu_\delta:=\Pi_*\varrho_\delta ν δ := Π ∗ ϱ δ . Then ν δ \nu_\delta ν δ is mixing for w ↦ w 2 w\mapsto w^2 w ↦ w 2 . The coding is one-to-one outside the dyadic ambiguities, which have ϱ δ \varrho_\delta ϱ δ -measure zero, and therefore we have h ν δ ( w ↦ w 2 ) = h ϱ δ ( σ ) h_{\nu_\delta}(w\mapsto w^2)=h_{\varrho_\delta}(\sigma) h ν δ ( w ↦ w 2 ) = h ϱ δ ( σ ) . Moreover, ν δ ( { 1 } ) = 0 \nu_\delta(\{1\})=0 ν δ ({ 1 }) = 0 , while ∣ Π ( a ) − 1 ∣ ≤ 2 π 2 − R ( a ) |\Pi(a)-1|\leq2\pi2^{-R(a)} ∣Π ( a ) − 1∣ ≤ 2 π 2 − R ( a ) implies
∫ S 1 log 2 ∣ w − 1 ∣ d ν δ ( w ) = ∞ . \int_{\mathbb{S}^1}\log\frac{2}{|w-1|}\,d\nu_\delta(w)=\infty. ∫ S 1 log ∣ w − 1∣ 2 d ν δ ( w ) = ∞.
By the Poincaré–Lelong formula, we have d d c u = δ 1 − ω dd^cu=\delta_1-\omega d d c u = δ 1 − ω , where δ 1 \delta_1 δ 1 is the Dirac measure at the point w = 1 w=1 w = 1 , so u ∈ PSH ( P 1 , ω ) u\in\operatorname{PSH}(\mathbb{P}^1,\omega) u ∈ PSH ( P 1 , ω ) and { u = − ∞ } = { 1 } \{u=-\infty\}=\{1\} { u = − ∞ } = { 1 } . Moreover, u ≤ 0 u\leq0 u ≤ 0 by the Cauchy–Schwarz inequality, with u ( [ 1 : − 1 ] ) = 0 u([1:-1])=0 u ([ 1 : − 1 ]) = 0 . Hence u ∈ U u\in\mathcal{U} u ∈ U . On S 1 \mathbb{S}^1 S 1 we have − u ( w ) = log ( 2 / ∣ w − 1 ∣ ) -u(w)=\log(2/|w-1|) − u ( w ) = log ( 2/∣ w − 1∣ ) . Choosing δ > 0 \delta>0 δ > 0 sufficiently small and setting ν : = ν δ \nu:=\nu_\delta ν := ν δ proves all the assertions. □
Positive-entropy measures in saddle two-dimensional settings
The following result is independent of the regularization criterion of Theorem 3.1 . Its proof combines standard consequences of the Ledrappier–Young theory [39 ] with a pluripolarity criterion of Dujardin [29 ] .
Theorem 5.1. Let F F F be in one of the following settings:
(i) F : X → X F:X\to X F : X → X is a holomorphic automorphism of a connected compact Kähler surface X X X ;
(ii) F : C 2 → C 2 F:\mathbb{C}^2\to\mathbb{C}^2 F : C 2 → C 2 is a polynomial automorphism of Hénon type;
(iii) F F F is an invertible horizontal-like map on a bidisk D ⊂ C 2 D\subset\mathbb{C}^2 D ⊂ C 2 [28 ] .
In case (iii), let K F ⋐ D K_F\Subset D K F ⋐ D denote the compact maximal invariant set, consisting of the points whose full orbits are defined and remain in D D D . Write Y : = X Y:=X Y := X , C 2 \mathbb{C}^2 C 2 , or D D D , respectively. Let μ \mu μ be an ergodic F F F -invariant probability measure with h μ ( F ) > 0 h_\mu(F)>0 h μ ( F ) > 0 , assumed to be compactly supported in case (ii) and supported on K F K_F K F in case (iii). Then μ \mu μ gives zero mass to every Borel pluripolar set in Y Y Y .
For the canonical measures of maximal entropy in these settings, the conclusion follows from their representation as wedge products of Green currents with continuous local potentials; see [17 Section 5.2] for compact Kähler surfaces, [15 , 13 ] for polynomial automorphisms of Hénon type, and [28 ] for invertible horizontal-like maps.
Proof. The argument is local near supp ( μ ) \operatorname{supp}(\mu) supp ( μ ) and applies in all three settings. In each case, F F F and F − 1 F^{-1} F − 1 are holomorphic on neighbourhoods of the compact support, so their derivatives are bounded there. Thus the integrability hypotheses needed for Ruelle’s inequality and Pesin theory hold.
Applying Ruelle’s entropy inequality to F F F and F − 1 F^{-1} F − 1 shows that μ \mu μ has Lyapunov exponents χ u > 0 > χ s \chi^u>0>\chi^s χ u > 0 > χ s , each of real multiplicity two. By Pesin theory, the local stable and unstable manifolds through μ \mu μ -almost every point are complex disks. Moreover, a set of full measure can be covered by countably many compact Pesin boxes in which the local stable and unstable plaques have uniform size, depend continuously on the base point, and form two measurable families of pairwise disjoint holomorphic graphs. We refer to [13 Sections 2.4–2.6 and Section 4] , [28 Section 5] , and [17 Sections 8.1.2–8.1.3] .
We first recall a standard consequence of the Ledrappier–Young theory. Applied to F F F and F − 1 F^{-1} F − 1 , its conditional dimension theorem gives positive pointwise dimension to the conditional measures of μ \mu μ along both unstable and stable Pesin manifolds; more precisely, the corresponding dimensions are h μ ( F ) / χ u h_\mu(F)/\chi^u h μ ( F ) / χ u and h μ ( F ) / ∣ χ s ∣ h_\mu(F)/|\chi^s| h μ ( F ) /∣ χ s ∣ , respectively; see [39 Proposition 7.3.1 and Theorem C] . Since polar subsets of a complex disk have Hausdorff dimension zero [2 Theorem 5.9.6] , these leafwise conditional measures give zero mass to polar sets. The same remains true after restricting μ \mu μ to a Pesin box: the resulting plaque conditionals are normalized restrictions of the corresponding leafwise conditional measures; see [13 Section 2.4, Proposition 2.5 and the preceding discussion] .
Suppose now that P ⊂ Y P \subset Y P ⊂ Y is Borel and pluripolar and that μ ( P ) > 0 \mu(P)>0 μ ( P ) > 0 . Choose a connected coordinate neighborhood U U U and a plurisubharmonic function φ ≢ − ∞ \varphi\not\equiv-\infty φ ≡ − ∞ on U U U such that μ ( P ∩ U ) > 0 \mu(P \cap U)>0 μ ( P ∩ U ) > 0 and P ∩ U ⊂ { φ = − ∞ } P \cap U \subset\{\varphi=-\infty\} P ∩ U ⊂ { φ = − ∞ } . We can then choose a compact Pesin box R ⊂ U R \subset U R ⊂ U such that μ ( P ∩ R ) > 0 \mu(P \cap R)>0 μ ( P ∩ R ) > 0 and all its local stable and unstable plaques are contained in U U U . Set ν : = 1 R μ / μ ( R ) \nu:= \mathbf{1}_{R}\mu/\mu(R) ν := 1 R μ / μ ( R ) , E : = P ∩ R E := P \cap R E := P ∩ R , and write W R u ( x ) W^{u}_{R}(x) W R u ( x ) and W R s ( x ) W^{s}_{R}(x) W R s ( x ) for the local holomorphic plaques through x x x . Denote by ν x u \nu^{u}_{x} ν x u and ν x s \nu^{s}_{x} ν x s the conditional probabilities of ν \nu ν on the partitions of R R R by the intersections with these plaques.
Disintegrating ν \nu ν along unstable plaques gives a set of positive quotient measure on which ν x u ( E ) > 0 \nu^{u}_{x}(E)>0 ν x u ( E ) > 0 . Choose Borel versions of the conditional probabilities that are constant on each partition element. Let N u ⊂ R N^{u}\subset R N u ⊂ R be a saturated Borel set with ν ( N u ) = 0 \nu(N^{u})=0 ν ( N u ) = 0 such that, outside N u N^{u} N u , the conditional probabilities vanish on polar sets. Then B : = { x ∈ R ∖ N u : ν x u ( E ) > 0 } B:=\{x\in R\setminus N^{u}:\nu^{u}_{x}(E)>0\} B := { x ∈ R ∖ N u : ν x u ( E ) > 0 } is measurable and has positive ν \nu ν -measure. For every plaque selected by B B B , the set E ∩ W R u ( x ) E\cap W^{u}_{R}(x) E ∩ W R u ( x ) is nonpolar. Since the restriction of φ \varphi φ to W R u ( x ) W^{u}_{R}(x) W R u ( x ) is subharmonic or identically − ∞ -\infty − ∞ and has E ∩ W R u ( x ) E\cap W^{u}_{R}(x) E ∩ W R u ( x ) in its pole set, this restriction is identically − ∞ -\infty − ∞ . Thus every full unstable plaque selected by B B B lies in { φ = − ∞ } \{\varphi=-\infty\} { φ = − ∞ } .
We now disintegrate ν \nu ν along stable plaques. There is a stable plaque D D D whose conditional measure gives positive mass to B B B and vanishes on polar subsets of D D D . Cover B ∩ D B\cap D B ∩ D by countably many smaller flow boxes in which D D D is transverse to the unstable plaques and these plaques are graphs over a common disk. At least one of these boxes meets B ∩ D B\cap D B ∩ D in a set of positive stable conditional measure. By inner regularity, there is a compact set A ⊂ B ∩ D A\subset B\cap D A ⊂ B ∩ D , contained in this box, with positive stable conditional measure. In particular, A A A is nonpolar in the complex disk D D D .
The unstable plaques through the points of A A A form a continuous family of pairwise disjoint holomorphic graphs, and their union is contained in { φ = − ∞ } \{\varphi=-\infty\} { φ = − ∞ } . After straightening the transversal D D D , [29 Lemma 6.5] implies that the parameter set A A A is polar. This contradicts the preceding paragraph. Therefore μ ( P ) = 0 \mu(P)=0 μ ( P ) = 0 . □
By ergodic decomposition, Theorem 5.1 also implies that every F F F -invariant probability measure satisfying the same support assumptions and giving full mass to a Borel pluripolar set has zero entropy. By the variational principle, every non-empty compact pluripolar set K ⊂ Y K\subset Y K ⊂ Y satisfying F ( K ) ⊂ K F(K)\subset K F ( K ) ⊂ K has h t o p ( F ∣ K ) = 0 h_{\mathrm{top}}(F|_{K})=0 h top ( F ∣ K ) = 0 . In case (iii), we assume in addition that K ⊂ K F K\subset K_{F} K ⊂ K F .
Remark 5.2. The proof of Theorem 5.1 uses that the stable and unstable Pesin manifolds are complex curves, on which polar sets have Hausdorff dimension zero. In higher dimension, positive conditional dimension alone does not exclude mass on pluripolar subsets of the Pesin manifolds, so the argument does not directly extend.
Appendix A. Compact Kähler manifolds and polynomial-like maps
Theorems 1.1 and 1.4 and Corollaries 1.2 and 1.3 extend to the following settings, with entropy threshold log Λ \log\Lambda log Λ in place of ( k − 1 ) log d (k-1)\log d ( k − 1 ) log d .
Compact Kähler manifolds. Let F : X → X F:X\to X F : X → X be a surjective holomorphic self-map of a connected compact Kähler manifold ( X , ω ) (X,\omega) ( X , ω ) of dimension k ≥ 1 k\ge1 k ≥ 1 , normalized by ∫ X ω k = 1 \int_{X}\omega^{k}=1 ∫ X ω k = 1 . Denote by λ j ( F ) \lambda_{j}(F) λ j ( F ) the dynamical degrees and assume λ k ( F ) > Λ : = λ k − 1 ( F ) \lambda_{k}(F)>\Lambda:=\lambda_{k-1}(F) λ k ( F ) > Λ := λ k − 1 ( F ) . The conclusions hold on X X X , with the positivity conditions and norms defined using ω \omega ω . For super-potentials on compact Kähler manifolds, see [25 ] . In dimension two, the remaining case λ 2 ( F ) ≤ λ 1 ( F ) \lambda_{2}(F)\le\lambda_{1}(F) λ 2 ( F ) ≤ λ 1 ( F ) is covered by Gromov’s entropy bound [36 ] . Thus, for every surjective holomorphic self-map of a compact Kähler surface, an invariant measure supported on a compact pluripolar set has entropy at most log λ 1 ( F ) \log\lambda_{1}(F) log λ 1 ( F ) , answering a question in [27 Section 2] .
Polynomial-like maps. Let f : U → V f:U\to V f : U → V be a proper holomorphic map, where U ⋐ V ⋐ C k U\Subset V\Subset\mathbb{C}^{k} U ⋐ V ⋐ C k and V V V is convex, and let K f : = ⋂ n ≥ 1 f − n ( U ) K_{f}:=\bigcap_{n\ge1}f^{-n}(U) K f := ⋂ n ≥ 1 f − n ( U ) be its filled Julia set. Assume that the topological degree satisfies d t > Λ : = d k − 1 ∗ ( f ) d_{t}>\Lambda:=d^{*}_{k-1}(f) d t > Λ := d k − 1 ∗ ( f ) , where d k − 1 ∗ ( f ) d^{*}_{k-1}(f) d k − 1 ∗ ( f ) is the ∗ * ∗ -dynamical degree controlling the growth of pushforwards of positive closed ( 1 , 1 ) (1,1) ( 1 , 1 ) -currents; see [22 , 24 ] . The conclusions hold for measures supported on K f K_f K f , viewed as measures on P k \mathbb{P}^k P k through the standard affine chart. Pluripolar sets are understood in V V V , and the expanding compact set in Theorem 1.4 is assumed to lie in K f K_f K f .
We explain the changes needed in the proof of Theorem 3.1 in both cases. The chart, covering, radial averaging, and local comparison arguments apply on a finite atlas of X X X , or on a fixed neighbourhood K f ⋐ W ⋐ U K_f \Subset W \Subset U K f ⋐ W ⋐ U . In the polynomial-like case, the transported kernels and their forward images remain in W W W , where the Jacobian is uniformly bounded.
The main change in both settings concerns the global current-mass estimate (16) . On X X X , cohomology gives ∥ ( F n ) ∗ ( T u + ω ) ∥ ≤ C δ δ n \|(F^n)_*(T_u+\omega)\| \leq C_\delta\delta^n ∥ ( F n ) ∗ ( T u + ω ) ∥ ≤ C δ δ n for every δ > Λ \delta> \Lambda δ > Λ , uniformly in u ∈ PSH ( X , ω ) u \in\operatorname{PSH}(X,\omega) u ∈ PSH ( X , ω ) . In the polynomial-like case, the definition of d k − 1 ∗ ( f ) d^*_{k-1}(f) d k − 1 ∗ ( f ) gives ∥ ( f n ) ∗ S ∥ W ≤ C δ δ n \|(f^n)_*S\|_W \leq C_\delta\delta^n ∥ ( f n ) ∗ S ∥ W ≤ C δ δ n for every δ > Λ \delta> \Lambda δ > Λ and positive closed ( 1 , 1 ) (1,1) ( 1 , 1 ) -currents S S S of uniformly bounded mass on U U U . By the same current domination argument in Step 3.1 of the proof of Theorem 3.1 , these bounds yield
∑ j ∑ g ∈ G n , j M n , j , g ( u ) ≤ C δ δ n , δ > Λ , \sum_j \sum_{g\in\mathcal{G}_{n,j}} M_{n,j,g}(u) \leq C_\delta\delta^n,\qquad\delta>\Lambda, j ∑ g ∈ G n , j ∑ M n , j , g ( u ) ≤ C δ δ n , δ > Λ ,
uniformly in u u u , which replaces (16) .
We next verify the inverse branch hypotheses for invariant measures of entropy greater than log Λ \log\Lambda log Λ . In the compact Kähler case, the proof of the entropy–Lyapunov estimate in [30 Section 6] still applies with minimal modifications: the backward graph transform is local, and [26 Lemma 5] gives the required mixed pullback volume estimates. By log-concavity of the dynamical degrees [38 , 46 , 35 ] , this shows that h μ ( F ) > log Λ h_\mu(F) > \log\Lambda h μ ( F ) > log Λ forces all Lyapunov exponents to be strictly positive. For polynomial-like maps, the corresponding conclusions follow from [14 Theorem 4.1] . The local inverse branch construction [8 , 5 , 9 ] and the Brin–Katok entropy formula then give the analogue of Lemma 2.3 . For the Bowen entropy statement, uniform expansion provides the required branches.
The remaining potential-theoretic arguments also apply. On X X X , compactness of normalized ω \omega ω -plurisubharmonic functions [37 Proposition 2.7] and smooth decreasing approximation [12 Theorem 1] give the analogue of Lemma 2.2 . By [37 Theorem 7.2] , every locally pluripolar set in X X X is contained in the pole set of a global ω \omega ω -psh function, so the integrability arguments give the pluripolar conclusions. The regularity consequences follow as in § 4.1 , using [19 ] .
Dipartimento di Matematica, Università di Pisa, Largo Bruno Pontecorvo 5, 56127 Pisa, Italy Email address: fabrizio.bianchi@unipi.it
Department of Mathematics, University of Oklahoma, Norman, OK 73019, USA Email address: he@ou.edu