Codimension one attracting sets in Pk(C) Sandrine Daurat Johan Taflin ∗ 5 1 0 2 Abstract n a J We are interested in attracting sets of Pk(C) which are of small 9 topological degree and of codimension 1. We first show that there 1 exists a large family of examples. Then we study their ergodic and pluripotential theoretic properties. ] S D . 1 Introduction h t a m The aim of this article is to study a special class of attracting sets for holo- [ morphic endomorphisms of the complex projective space Pk. Let f be such an endomorphism, of algebraic degree d 2. Recall that an attracting set 1 ≥ v of f is a compact set A Pk admiting an open neighborhood U which is 1 a trapping region i.e. f(U⊂) ⋐ U and A := fn(U). These sets play an 2 n≥0 4 important role in dynamical systems and they have been studied by many 4 T authors, see e.g. [FS98, FW99, JW00, Din07, R12] for the complex setting. 0 . In [Dau14], the first author introduced a notion of dimension for at- 1 0 tracting sets and the notion of attracting sets of small topological degree, see 5 Definition 2.1 and Definition 2.2 below. In this framework several interesting 1 dynamical and geometric properties have been pointed out. Among other : v things, it was shown in this paper that such attracting sets are non-algebraic i X and also that such examples are quite abundant in P2. Our purpose here is r a to answer questions that were left open. Our first result is a generalization of the construction [Dau14] to arbi- trary dimensions k 2. To be more specific, let k 2 be an integer and for ≥ ≥ d 2, let F denote the set of pairs f = (f ,R) where R is an homogeneous d ∞ ≥ polynomial of degree d in Ck and f = (F ,...,F ) is a k-tuple of homo- ∞ 0 k−1 geneous polynomials which defines a holomorphic endomorphism of Pk−1 of ∗ SandrineDauratwaspartiallysupportedbytheFMJH(GovernementProgram: ANR- 10-CAMP-0151-02) and the ANR project LAMBDA (Governement Program: ANR-13- BS01-0002) 1 degree d. The set F is a quasi-projective variety. If f = (f ,R) F and d ∞ d ∈ ε C then we denote by f the endomorphism of Pk defined by ε ∈ [z : : z : z ] = [z′ : z ] [f (z′) : zd +εR(z′)]. 0 ··· k−1 k k 7→ ∞ k When ε = 0, these maps admit the hyperplane z = 0 as an attracting set, k { } which is algebraic and thus cannot be of small topological degree. However, a generic perturbation f has an attracting set with this property. ε Theorem 1.1. There exists a non empty Zariski open set Ω F , such that d ⊂ if f Ω then for ε C with ε = 0 small enough, the maps f is of small ε ∈ ∈ | | 6 topological degree on an attracting set A close to the hyperplane z = 0 . ε k { } The idea of the construction of the Zariski open set Ω is the same as in [Dau14]. In particular, the proof uses the fact that the maps f preserve the ε pencil of lines passing through [0 : : 0 : 1]. The new delicate part here is ··· to provide examples in any dimension and any degree to ensure that Ω is not empty. Notice that the property to possess an attracting set of small topological degree is stable under small perturbations. It is also the case for all our other assumptions below. Hence, Theorem 1.1 provides a lot of attracting sets of small topological degree of codimension 1, i.e. they carry non-trivial positive closed currents of bidegree (1,1) and they are not equal to the whole space Pk. It would be interesting to understand the nature of the open set in the whole parameter space where the small topological degree condition holds. Our other results are related to the ergodic and pluripotential theoretic properties of such attracting sets. Let A be an attracting set of codimension 1 with a trapping region U. Let C (U) be the set of all positive closed (1,1)- 1 currents S supported in U and of mass 1. Under a mild assumption on U, which will be referred to as (HD), Dinh constructed in [Din07] an attracting current τ C (U) such that any continuous form S C (U) satisfies 1 1 ∈ ∈ lim ΛnS = τ, n→+∞ where Λ := d1−kf is the normalized push forward operator acting on C (U). ∗ 1 In particular τ is invariant under Λ. He also obtained that if T denotes the Green current of f then the measure ν := τ Tk−1 is mixing and of entropy ∧ (k 1)logd. − The first author showed that when f is of small topological degree on A then τ has bounded potential and asked whether this potential was contin- uous. The answer is positive, provided that the geometric assumption (HD) holds. 2 Theorem 1.2. Let f be a holomorphic endomorphism of Pk. If f is of small topological degree on an attracting set A of codimension 1 satisfying (HD) then the attracting current τ on A has continuous potential. More precisely, we are going to prove that there exist constants A > 0 and α > 0 such that for all x,y Pk u(x) u(y) A logdist(x,y) −α ∈ | − | ≤ | | where u is such that τ ω = ddcu. − This result gives information on the size of the attracting set in the sense that the moreregular thepotential ofτ is, thebigger itssupport is. However, we were not able to obtain Hölder continuity which would give a lower bound on the Hausdorff dimension of A. Another interesting question about currents in C (U) is whether τ is the 1 unique current in C (U) invariant by Λ. Actually, under the same assump- 1 tions, this is indeed the case and moreover we obtain the following equidis- tribution result towards τ with uniform exponential speed. Theorem 1.3. Let f and A be as in Theorem 1.2. There exist constants 0 < δ < 1 and c 0 such that ≥ ΛnS τ,φ cδn φ , C2 |h − i| ≤ k k for all S C (U) and test forms φ of class 2. In particular, τ is the unique 1 ∈ C current in C (U) invariant by Λ. 1 Both theorems can be reformulated as speed of convergence results for potentials, inthe sup normfor the first one andin theL1 norm forthe second one. Besides the assumption of small topologicaldegree, the proofs are based on the “geometric” properties of the space C (U) which come from (HD). 1 Our last result is about the hyperbolicity of certain invariant measures. In [deT08], de Thélin obtained a powerful result which relates the entropy of an invariant measure, its Lyapunov exponents and the dynamical degrees of f. For an endomorphism of Pk, the dynamical degree of order i is just di. The last one, dk, is the topological degree of f on Pk. We will substitute it by a local version to prove the following result. Theorem 1.4. Let ν be an ergodic measure and of entropy h(ν) = (k − 1)logd. If there exists an open neighbourhood V of supp(ν) such that 1/n d := limsup (fn)∗(ωk) < dk−1, k,loc n→+∞ (cid:18)ZV (cid:19) then ν is hyperbolic and its Lyapunov exponents satisfy 1 1 d t χ χ logd > 0 > log χ . 1 ≥ ··· ≥ k−1 ≥ 2 2 dk−1 ≥ k ≥ −∞ (cid:18) (cid:19) 3 The assumption d < dk−1 holds in all examples known by the authors. k,loc Moreover, to be of small topological degree on A is much stronger than d < dk−1. The first one is a uniform pointwise property while the latter k,loc deals with an average on V. Hence, Theorem1.4appliestothemeasureν constructedbyDinh, aslong as d < dk−1, so in particular for all the examples obtained in Theorem k,loc 1.1. To conclude, let us underline here that results similar to Theorem 1.3 and Theorem 1.4 have been obtained by the second author in [Taf13] in a different setting well adapted to perturbations of super-attractive attracting sets of any codimension, which are all the know examples. However, this set of examples is quite reduced and we believe these theorems, or at least there proofs, could be applied to a very large class of codimension 1 attracting sets. Notice furthermore that, in Theorem 1.4, the support of the measure ν is not necessary contained in an attracting set. The paper is organized as follows. In Section 2, we recall Dinh’s setting (HD) and the notions of dimension and of small topological degree attached to attracting sets. Section 3 is devoted to the proof of Theorem 1.1 on the existence ofexamplesinanydimension. Then, inSection4weproveTheorem 1.2 and Theorem 1.3 about the attracting current τ. Finally, we discuss in Section 5 the hyperbolicity of measures of entropy (k 1)logd. − We thank H. de Thélin for having pointed out to us that we should be able to extend his result [deT08] to our case. 2 Attracting sets in Pk In this section, we summarize several notions and results about attracting sets. For 0 p k and E Pk, let denote by C (E) the set of all positive p ≤ ≤ ⊂ closed (p,p)-currents of mass 1 and supported on E. Here, the mass of a (p,p)-current S is S := S,ωk−p where ω is the standard Fubini-Study k k h i form on Pk. Following [Dau14], we define the dimension of an attracting set. Definition 2.1. Let A be an attracting set. We say that A has dimension k p, or equivalently codimension p, if C (A) = ∅ and C (A) = ∅. p p−1 − 6 The whole space Pk is always an attracting set of maximal dimension k and it is the only one. In the same way, the attracting periodic points corre- spond exactly to attracting sets of dimension 0. Between these two extremes, 4 the possibilities for attracting sets of dimension p with 0 < p < k are much more varied. In the sequel we will focus on the codimension 1 case. The key notion in this paper is the following. Definition 2.2. Let f be an endomorphism of Pk of degree d admitting an attracting set A of dimension s. The endomorphism f is said to be of small topological degree on A if there exists a trapping region U such that limsup(♯f−n(x) U)1/n < ds, ∩ n→+∞ for any x Pk. ∈ Notice that if the inequality above holds for a trapping region U, it also holds for any other trapping region. Sometimes, we will abbreviate this into “A is an attracting set of small topological degree”. This notion is stable under small deformations and it ensures that A is not algebraic (cf. [Dau14]). To conclude this section, let us introduce a set of the geometric assump- tion on U used by Dinh [Din07], which imply good path connectedness prop- erties for C (U). We say that an attracting set A with a trapping region U p satisfy the assumptions (HD) if the following holds: There exist an integer 1 p k and two projective subspaces I and L of ≤ ≤ dimension p 1 and k p respectively such that I U = ∅ and L U. Since − − ∩ ⊂ I L = ∅, for each x L there exists a unique projective subspace I(x) of ∩ ∈ dimension p which contains I and such that L I(x) = x . We ask that for ∩ { } each x L, the set U I(x) is star-shaped in x as a subset of I(x) I Cp. ∈ ∩ \ ≃ It is easy to check that in this situation A has codimension p. Under these assumptions, Dinh showed in [Din07] that continuous forms in C (U) equidistribute under push-forward. To be more precise, let Λ be p the operator on C (U) defined by p ΛS,φ := dp−kf S,φ := S,dp−kf∗φ . ∗ h i h i h i Then, the following result holds. Theorem 2.3 ([Din07]). Let f, A and U be as above. There exists a current τ C (U) such that if S is a continuous form in C (U) then p p ∈ lim ΛnS = τ. n→+∞ Moreover τ is woven, invariant under Λ and extremal relative to this last property. 5 3 Existence of attracting sets of small topolog- ical degree in Pk This section is devoted to the proof of Theorem 1.1. We assume that the dimension k is at least 2. From now on and in the rest of the paper, we will always denote by s the dimension of our attracting sets, i.e. s = k 1 1. − ≥ Recall from the introduction that if d 2, then F denotes the set of pairs d ≥ f = (f ,R) where R is an homogeneous polynomial of degree d in Ck and ∞ f = (F ,...,F ) is a k-tuple of homogeneous polynomials which defines a ∞ 0 s holomorphic endomorphism of Ps of degree d. By an abuse of notation, we will also denote by f this endomorphism. Moreover, if f F and ε C ∞ d ∈ ∈ f stands for the endomorphism of Pk defined by ε [z : : z : z ] = [z′ : z ] [f (z′) : zd +εR(z′)]. 0 ··· s k k 7→ ∞ k When ε is small enough, f admits an attracting set near the hyperplane ε | | z = 0 . Indeed, for ε = 0 small, we will use the following trapping region k { } | | 6 U = [z : : z : z ] Pk; z c ε max( z , , z ) , ε 0 s k k 0 s { ··· ∈ | | ≤ | | | | ··· | | } where c > 0 is a small number depending on f F , see [Dau14, Theorem d ∈ 4.1 Step 1.] for more details. It satisfies the assumptions of Dinh (HD), and we denote by A := fn(U ) the associated attracting set. ε ε ε n∈N The proof of TheoTrem 1.1 will be given in the next two subsections. In the first one, we will define the Zariski open set Ω F and show that if d ⊂ f Ω then f is of small topological degree on A for ε = 0 small enough. ε ε ∈ | | 6 This part is just an adaptation of [Dau14] to higher dimension. The second subsection is devoted to show that Ω is not empty and there- fore, it is a dense Zariski open set of F . Even if it relies on elementary d arguments, the construction is quite delicate when k 3. ≥ 3.1 An algebraic sufficient condition All the maps f preserve the pencil of lines passing through [0 : : 0 : 1], ε ··· which is naturally parametrized by Ps. Moreover, via this parametrization, the dynamics induced by f on this pencil is exactly f . The four following ε ∞ sets of lines playanimportant roleinorder to controltonumber ofpreimages which are in an attractive region (cf. [Dau14]). Denote by 6 X (f) := z Ps; ♯(f+)−1(f+(z)) ds−1 +1 = dk−2 +1 , −1 { ∈ ∞ ∞ ≥ } X(f) := f (X (f)), ∞ −1 Y (f) := z Ps; R(z) = 0 −2 { ∈ }∪ z Ps; z′ Ps,f+(z) = f+(z′), f (z) = f (z′),f f+(z) = f f+(z′) , { ∈ ∃ ∈ ∞ 6 ∞ ∞ ∞ ◦ ∞ ◦ ∞ } Y(f) := f2 (Y (f)). ∞ −2 Here f+ is the map from Ps to Pk defined by ∞ f+(z) = [f (z) : R(z)]. ∞ ∞ Observe that these sets only depend on f F , not on ε. When no confusion d ∈ is possible, we shall just write X, Y, etc. Notice that in the definition of X (f) the cardinality is counted with multiplicity and −1 f+ −1 f+(z) = z′ (f )−1(f (z)) R(z′) = R(z) . ∞ ∞ { ∈ ∞ ∞ | } The fol(cid:0)low(cid:1)ing r(cid:0)esult gi(cid:1)ves a sufficient condition on X and Y to ensure the existence of an attracting set with small topological degree. It is analogous to [Dau14, Proposition 4.3]. Proposition 3.1. Let Z(f) := X(f) Y(f). If f F is such that d ∪ ∈ X(f) and Y(f) are disjoint, • there exists n N such that n f−i(Z(f)) = ∅, • ∈ i=0 ∞ then for ε C with ε = 0 small eTnough, f is of small topological degree on ε ∈ | | 6 A . ε Denote by Ω the subset of pairs (f ,R) F satisfying the assumptions ∞ d ∈ of this proposition. Before proving the result, let us point out the important fact that the set Ω is a Zariski open set. Indeed, the sets X and Y are −1 −2 analytic subsets of Ps, and so the same holds for X, Y, Z and f−i(Z), for ∞ all i 1. When f varies in F , it is easy to check that they form analytic d ≥ subsets of Ps F . The set Ω is exactly the complement of the projection d × on F of an analytic subset obtained by intersecting and taking finite union d of such sets, see Step 3 of proof of [Dau14, Theorem 4.1] for more details. Proof. The proof is essentially the same than the one of [Dau14, Proposition 4.3]. The fact that the pencil of lines is parametrized by Ps instead of P1 doesn’t affect the arguments. However, for the convenience of the reader we recall the main ideas. 7 Throughout the proof, by a line we mean a line passing through [0 : : ··· 0 : 1]. We will identify such a line l with the associated point in Ps, i.e. the unique point in the intersection of l with the hyperplane at infinity. The image of l by f is f (l), for all ε C. With this identification, the sets X ε ∞ ∈ and Y can be seen as sets of points in Ps or sets of lines in Pk. As in [Dau14, Step 2. Theorem 4.1] we reduce the estimation of the num- ber of preimages staying in f (U ) to a combinatorial problem by analyzing ε ε the geometry of f (U ), as well as its self-crossings, and the behaviour of the ε ε preimages. A generic line has ds preimages and a generic point p include a line ℓ has ds+1 = dk preimages (in Pk), d in each line ℓ f−1(ℓ). So we may : −1 ∈ control the number of lines ℓ that contain preimages staying in U . −1 ε • Foreachlineℓ , boundthenumber ofpreimagesthatbelongtof (U ). −1 ε ε • As in [Dau14, Step 2. Theorem 4.1], this done by defining sets A,B of “good” lines. The set A is the set of lines l such that the preimages of a point x ∈ f (U ) l, which lie in U , are contained in at most ds−1 lines. Here and in ε ε ε ∩ the sequel, everything is counted with multiplicity. We say that l is in B if for all x l f2(U ) and each l f−1(l) the ∈ ∩ ε ε −1 ∈ ∞ point x has at most one preimage in l f (U ). −1 ε ε ∩ To summarize : If x l A then there are at most ds−1 lines intersecting f−1(x) U , • ∈ ∈ ε ∩ ε in at most d points for each of them. If x l B then there are at most ds lines intersecting f−1(x) f (U ), • ∈ ∈ ε ∩ ε ε in at most 1 point for each of them. If x l A B then there are at most ds−1 lines intersecting f−1(x) • ∈ ∈ ∩ ε ∩ f (U ), in at most 1 point for each of them. ε ε In the two first cases, f−1(x) f (U ) contains at most ds points. In the last ε ∩ ε ε one, it contains at most ds−1 points. By our assumptions, there exists r > 0 such that n Xr Yr = ∅, and f−i(Zr) = ∅, ∩ ∞ i=0 \ where Xr (resp. Yr,Zr) denotes the r-neighbourhood of X (resp. Y,Z) in Ps. 8 Notice that the sets A and B depend on ε. Indeed, a key point is that the lines in X (resp. Y) are exactly those who are not in A (resp. B) for any ε C. This fact is illustrated by the following result which can be proved in ∈ the same way than [Dau14, Lemma 4.4]. Lemma 3.2. Fix r as above. Then if ε = 0 is small enough, we have that | | 6 Ac Xr, Bc Yr and X Y = Z (A B)c Zr. (3.1) ⊂ ⊂ ∪ ⊂ ∩ ⊂ Now we have reduced the problem to a combinatorial problem. We have to make the same disjunction of cases as in the end of the proof of [Dau14, Proposition 4.3], except that there are now n step instead of 3. By the previous lemma : If x l / Xr then x has at most d preimages in at most ds−1 lines. • ∈ ∈ If x l / Yr then x has at most 1 preimages in at most ds lines. • ∈ ∈ The condition Xr Yr = ∅ imply that for all x f2(U ): ε ∩ ∈ either there are at most ds−1 lines intersecting f−1(x) U , in at most • ε ∩ ε d points for each of them, or there are up to ds lines l intersecting f−1(x) f (U ), but in each • −1 ε ∩ ε ε such line there is at most 1 preimage of x. So the number of preimages of a point staying in f (U ) is less than ds. Since ε ε f−i(Zr) = ∅, there exists one step, between 0 and n, such that the ∞ i=0,..,n prTeimages of a line are in A B so have less than ds−1 preimages. ∩ This is summarized in the following diagram : ♯(p∩fε(Uε)) inA∩B1❘❘❘❘❘❘❘❘n❘o❘t❘❘in❘❘A❘∩❘❘B❘ ♯(f−1(p) f (U )) ds−1 ds ε ∩ ε ε ≤ ≤ ◆◆◆◆◆n◆o◆t◆i◆n◆◆A∩B ♯(f−2(p) f (U )) ds.d(s−1) ds.ds−1 d2s ε ∩ ε ε ≤ ≤ ≤ ♯(f−(n−1)(p) f (U )) d(n−2)s.ds−1 d(n−2)s.ds−1 d(n−1)s ε ∩ ε ε ≤ ≤ ≤ ❑❑❑❑❑ Imposs❑i❑b❑le❑❑ ♯(f−n(p) f (U )) d(n−1)s.ds−1 d(n−1)s.ds−1 d(n−1)s.ds−1 > ε ∩ ε ε ≤ ≤ ≤ Thus in each case ♯(f−n(p) f (U )) d(n−1)s.ds−1 < dns andwe conclude ε ∩ ε ε ≤ that fn is of small topological degree in f (U ), hence on A too. This finish ε ε ε ε the proof. 9 3.2 Construction of examples In order to exhibit an element f Ω F , the first step is to find an d ∈ ⊂ example for which X(f) is finite. To this aim, until the end of this section, we only consider pairs f = (f ,R) F such that only monomials of the ∞ d ∈ form zd appear in the expression of f . i ∞ Let ξ C be a primitive d-th root of unity. Let G :=< ξ > Z/dZ. The ∈ ≃ group Gs acts on Ps by (ξ ,...,ξ ).[z : z : : z ] = [z : ξ z : : ξ z ]. 1 s 0 1 s 0 1 1 s s ··· ··· The special form of f ensures that if z is a preimage of w Ps then ∞ ∈ f−1(w) = Gs. z . Therefore, in this case, the set X consists of points ∞ { } −1 z Ps such that there exists A Gs satisfying ♯A = ds−1 +1, Id A and ∈ ⊂ ∈ R(g.z) = R(h.z) for all g,h A. It is easy to see that the s + 1 points in ∈ Ps which have only one non-zero coordinate are fixed by all the elements of Gs. In particular, they are all in X . We call these s+1 points the extremal −1 points of Ps. The following result exhibits an example where X , and therefore X, is −1 finite. Proposition 3.3. If R = ( s z )d then X is finite. i=0 i −1 Proof. We have to prove thPat if A Gs satisfied ♯A = ds−1 +1 then the set ⊂ X of points z Ps such that R(g.z) = R(h.z), g,h A, is finite. A ∈ ∀ ∈ The choice of R provides two advantages. The first one is that if R(z) = R(w) then there exists η C such that s z = η s w , which is a ∈ i=0 i i=0 i linear equation. Hence, the problem turns out to be reduce to basic linear P P algebra. The second one is that the restriction of R to z = 0 gives the i { } counterpart of R in dimension s 1 on z = 0 Ps−1. It allows us to i − { } ≃ process by induction on s. For s = 0, there is nothing to prove since P0 is reduced to a single point. Now, let s 1 and assume that the result holds for s 1. To be more ≥ − specific, assume that for any subset A′ Gs−1 with ♯A′ = ds−2 +1, there is ⊂ only a finite set X′ of points z′ Ps−1 such that R′(g.z′) = R′(h.z′) for all A′ ∈ g,h A′, where R′(z′) := ( s−1z′)d. ∈ i=0 i LetA Gs with♯A = ds−1+1.ThefirststepistoshowthatX z = 0 A i ⊂ P ∩{ } is finite. Let 1 i s and α G. If A denotes the set of points (ξ ,...,ξ ) A i,α 1 s ≤ ≤ ∈ ∈ such that ξ = α then ♯A = ds−1+1. Hence, there exists at least one i α∈G i,α α˜ G such that ♯A ds−2+1. Since the projection i,α˜ ∈ P≥ π: (ξ ,...,ξ ,ξ ,ξ ,...,ξ ) (ξ ,...,ξ ,ξ ,...,ξ ) 1 i−1 i i+1 s 1 i−1 i+1 s 7→ 10