arXiv:2012.14529v1 [math.NT] 28 Dec 2020 iudai omon form biquadratic hn nescinnme fmetrized of number intersection Zhang field ucin soitdt dlclymetrized adelically to associated functions vrtefnto field function the over fibers uvswt zero with curves surface rjciesurface, projective l endover defined all LITCSRAE N RTMTCEUDSRBTO FOR EQUIDISTRIBUTION ARITHMETIC AND SURFACES ELLIPTIC nti ril,w td h emtyadaihei fteset the of arithmetic and geometry the study we article, this In Date Suppose K E aur ,2021. 1, January : n epoea qiitiuintermfrteehihs est We heights. these for theorem equidistribution an prove we and , litccreover curve elliptic [ E fteAaeo-hn nescinnumbers intersection Arakelov-Zhang the of r ierydpnetin dependent linearly are o mohpoetv curve projective smooth for aehih ucin ntefibers the on functions height Tate R h aecurve base the oe ubrfil) xedn h qiitiuinterm fC of theorems [ equidistribution Yuan the and extending field), number a (over lopoeagnrleudsrbto hoe o dlclymetrized adelically for theorem equidistribution general a prove also perturbations height Abstract. aifisa least at satisfies MZ1 E t sacneune ehv h olwn oooo-yestateme Bogomolov-type following the have we consequence, a As . ( k r smrhcover isomorphic are sa litccurve elliptic an as with ) , → E MZ2 K CL1 n fBror n aun [ Capuano and Barroero of and ] m O B Suppose ota l u ntl ayfibers many finitely but all that so , t B esyta h litcsurface elliptic the that say We . ≥ , B sa litcsraedfie vranme field number a over defined surface elliptic an is two E Th sasot rjciecre n hr xssasection a exists there and curve, projective smooth a is ,teeeitannrpaigifiiesequence infinite non-repeating a exist there 2, AR EMROADNK YT MAVRAKI MYRTO NIKI AND MARCO DE LAURA vranme ed o each for field, number a over ( k k , k needn ierrltosfrall for relations linear independent nti ril,w osrc dlclymetrized adelically construct we article, this In . P ) Yu → E when i,t ⊗ ′ ,aduigrslso oiai[ Moriwaki of results using and ], n R ∈ R E E B erlt hstermt xsigrsls n rvd,for provide, and results, existing to theorem this relate We . K ( E DVSR NCURVES ON -DIVISORS B k vrtefield the over sanniorva litcsraedfie vranme field, number a over defined surface elliptic non-isotrivial a is → E t Let . Let . .Ti ie e ro frslso asradZannier and Masser of results of proof new a gives This ). n ( Q 1. fspecializations of ) k E B R k t Introduction eoetefnto field function the denote ( dvsr n rv hti nue non-degenerate a induces it that prove and -divisors Q sntiorva.T hsed ecnie height consider we end, this To isotrivial. not is eoetefnto field function the denote of ) R BC 1 dvsr ntebs curve base the on -divisors k E D . n xed u ale eut [ results earlier our extends and ] X over X · ∈ D t P Y E → E n i,t ∈ sabqartcfr on form biquadratic a as , ( fadol ftepoints the if only and if , E k n Mo2 B ) t ota h set the that so for , ⊗ ( B Q ]. R :gvnpoints given ): siorva falo h smooth the of all if isotrivial is epoenon-degeneracy prove We . Q t ( B ∈ abr-or Thuillier, hambert-Loir, t and ) n B R K ∈ ( E dvsr ncurves on -divisors K R tfrteN´eron- the for nt ( { B ( B E P -divisors ,aesot elliptic smooth are ), k ( K 1 ′ ;w love the view also we ); Q frtoa points rational of ) ,t h associated the P n htis, That . n small- and ) d h Arakelov- the udy B P , . . . , 1 P P , . . . , DM 1 vranumber a over P , . . . , E D O .We ]. ( m,t X ′ k m ) : on n ⊗ m ∈ B } E E → sa is , ELLIPTIC SURFACES AND ARITHMETIC EQUIDISTRIBUTION FOR R-DIVISORS ON CURVES 2 example, a new proof of results of Masser and Zannier and of Barroero and Capuano on linear relations between specializations of independent sections. 1.1. Heights and the Arakelov-Zhang intersection of points in E(k). Assume that
E → B is not isotrivial. Let hˆE denote the N´eron-Tate canonical height on E(k), associated ˆ to the choice of divisor O on E; let hEt denote the corresponding canonical height on the smooth fibers Et(K) for (all but finitely many) t ∈ B(K). By non-isotriviality, a point ˆ P ∈ E(k) satisfies hE(P ) = 0 if and only if it is torsion on E. We denote the specializations of P by Pt in the fiber Et. Tate showed in [Ta] that the canonical height function ˆ (1.1) hP (t) := hEt (Pt) is a Weil height on the base curve B(K), up to a bounded error. More precisely, there exists ˆ a Q-divisor DP on B of degree equal to hE(P ) so that hP (t)= hDP (t)+ O(1), where hDP is a Weil height on B(K) associated to DP . In [DM], we showed that we can also understand the small values of the function (1.1), from the point of view of equidistribution. Assume that hˆE(P ) > 0 (so that the function hP is nontrivial) and that, as a section, P : B →E is defined over the number field K. Building on work of Silverman [Si2, Si4, Si5], we showed
that hP is the height induced by an ample line bundle on B (with divisor DP ) equipped with a continuous, adelic metric of non-negative curvature defined over K, denoted by DP and satisfying
DP · DP =0 for the Arakelov-Zhang intersection number introduced in [Zh4]. In particular, we can then apply the equidistribution theorems of [CL1, Th, Yu] to deduce that the Gal(K/K)-orbits
of points tn ∈ B(K) with height hP (tn) → 0 are uniformly distributed on B(C) with respect to the curvature distribution ωP for DP at an archimedean place of K. A similar equidistri- an bution occurs at each place v of K to a measure ωP,v on the Berkovich analytification Bv [DM, Corollary 1.2]. As a consequence of our main result in [DM], and combined with the results of Masser and Zannier [MZ1, MZ2], we have
(1.2) DP · DQ ≥ 0 for all P, Q ∈ E(k), and
DP · DQ =0 ⇐⇒ either P or Q is torsion, or
∃ α> 0 such that hP (t)= α hQ(t) for all t ∈ B(K) ⇐⇒ ∃ (n, m) ∈ Z2 \{(0, 0)} such that nP = mQ
In particular, by the non-degeneracy of the N´eron-Tate bilinear form h·, ·iE on E(k), we have ˆ ˆ 2 (1.3) DP · DQ =0 ⇐⇒ hE(P )hE(Q)= hP, QiE for all P, Q ∈ E(k). ELLIPTIC SURFACES AND ARITHMETIC EQUIDISTRIBUTION FOR R-DIVISORS ON CURVES 3 The main result of this article is the proof of a stronger version of (1.3): Theorem 1.1. Let E → B be a non-isotrivial elliptic surface defined over a number field K. Let E be the corresponding elliptic curve over the field k = K(B). There exists a constant c> 0 so that ˆ ˆ 2 −1 ˆ ˆ 2 c hE(P )hE(Q) −hP, QiE ≤ DP · DQ ≤ c hE(P )hE(Q) −hP, QiE for all P, Q ∈ E(k), where h·, ·iE is the N´eron-Tate bilinear form on E(k).
Remark 1.2. The upper bound on DP · DQ in Theorem 1.1 is relatively straightforward. The difficulty lies in the lower bound; in Section 7, we observe that this is equivalent to
proving that DX · DY > 0 for all independent X,Y ∈ E(k) ⊗ R. 1.2. Motivation and context. Theorem 1.1 was inspired by the statements and proofs of the Bogomolov Conjecture [Ul, Zh3, SUZ], extending Raynaud’s theorem that settled the Manin-Mumford Conjecture [Ra].
Theorem 1.1 can be seen as a Bogomolov-type bound. The intersection number DP · DQ ˆ ˆ is related to the small values of the heights hEt (Pt)+ hEt (Qt) in the fibers Et(K). Indeed, as a consequence of Zhang’s Inequality [Zh4, Theorem 1.10] applied to the sum DP + DQ, and the fact that hP (t) ≥ 0 at all points t ∈ B(K) for every P ∈ E(k)[DM, Proposition 4.3], we have
1 DP · DQ (1.4) ess.min. (hP + hQ) ≤ ≤ ess.min. (hP + hQ) 2 hˆE(P )+ hˆE(Q) for every pair of non-torsion P, Q ∈ E(k). The essential minimum is defined by ess.min.(f)=
supF infB\F f over all finite sets F in B(K). Bogomolov-type bounds have found many applications in problems of unlikely intersections. In Section 7, we explain that Theorem 1.1 is equivalent to the following: Theorem 1.3. Let E → B be a non-isotrivial elliptic surface defined over a number field, and let π : E m → B be its m-th fibered power with m ≥ 2. Let E m,{2} denote the union of flat subgroup schemes of E m of codimension at least 2, and consider the tubular neighborhood
m,{2} m ′ m,{2} ′ ′ T (E , ǫ)= P ∈E : ∃P ∈E (Q) with π(P )= π(P ) and hˆ m (P − P ) ≤ ǫ Eπ(P ) Then, for any irreduciblen curve C in E m, defined over a number field and dominatingoB, there exists ǫ> 0 such that C ∩ T (E m,{2}, ǫ) is contained in a finite union of flat subgroup schemes of positive codimension. See, e.g., [BC, Lemma 2.2] for definitions and a classification of flat subgroup schemes. ELLIPTIC SURFACES AND ARITHMETIC EQUIDISTRIBUTION FOR R-DIVISORS ON CURVES 4 Remark 1.4. The conclusion of Theorem 1.3 with ǫ = 0 is a result of Barroero and Capuano [BC, Theorem 2.1]: using techniques involving o-minimality and transcendence theory, simi- lar to those of [MZ1, MZ2] (which treated the intersections of curves C with T (E m,{m}, 0), the torsion subgroups), they show that C ∩ T (E m,{2}, 0) is contained in a finite union of flat sub- group schemes of positive codimension. Thus Theorem 1.3 may be seen as a Bogomolov-type extension of [BC, Theorem 2.1], while providing a new proof of results in [BC, MZ1, MZ2]. Our main result in [DM] treated the intersections of C with the smaller tube T (E m,{m}, ǫ). In [Zh2], Zhang proposed the investigation of a function on the base curve B that detects drops in rank of the specializations of a subgroup of E(k): given a finitely-generated subgroup
Λ of E(k) of rank m ≥ 1, if the torsion-free part of Λ is generated by S1,...,Sm, let
(1.5) hΛ(t) := det(hSi,Sjit)i,j ≥ 0
on B(K), where defined, where h·, ·it is the N´eron-Tate bilinear form on the specialization Λt in the fiber Et. We propose the following result as the analog of [Zh2, §4 Conjecture] for elliptic surfaces; Zhang’s conjecture was formulated for geometrically simple families of abelian varieties A→ B of relative dimension > 1, and it does not hold as stated for elliptic surfaces [Zh2, §4 Remark 3]. Theorem 1.5. Let E → B be a non-isotrivial elliptic surface defined over a number field K, and let E be the corresponding elliptic curve over the field k = K(B). Let Λ ⊂ E(k) be a subgroup of rank m ≥ 2, with torsion-free part generated by S1,...,Sm ∈ E(k). For each i = 1,...,m, let Λi ⊂ Λ be generated by {S1,...,Sm}\{Si}. There is a constant ǫ = ǫ(Λ) > 0 so that
{t ∈ B(K): hΛ1 (t)+ ··· + hΛm (t) ≤ ε} is finite. In Section 7, we observe that Theorem 1.5 is equivalent to Theorems 1.1 and 1.3.
Remark 1.6. Note that, for rank 1 groups Λ, the value hΛ(t) is the canonical height of the ˆ generating point St in Et. In general, recall that the N´eron-Tate height hEt on a smooth fiber defines a positive definite quadratic form in Et(Q) ⊗ R; see e.g. [Si6, Ch. VIII, Prop. 9.6]. Thus, hΛ will vanish at t ∈ B(K) if and only if rank Λt < rank Λ. The sum hΛ1 (t)+ ··· + hΛm (t) will be zero if and only if the points S1,t,...,Sm,t satisfy (at least) two independent linear relations over Z in the fiber Et.
Remark 1.7. The independence of the points S1,...,Sm ⊂ Λ in Theorem 1.5 is necessary for the finiteness statement to hold. Indeed, suppose that Sm is a linear combination of S1,...,Sm−1, and suppose that {tn} ⊂ B(K) is any infinite non-repeating sequence for which ELLIPTIC SURFACES AND ARITHMETIC EQUIDISTRIBUTION FOR R-DIVISORS ON CURVES 5 hSm (tn) → 0 (for example, we can take tn where Sm,tn is torsion; see e.g. [DM, Proposition
6.2]). As we will see in Proposition 5.3, we then have hΛ1 (tn)+ ··· + hΛm (tn) → 0. If the elliptic surface E → B is isotrivial, the conclusion of Theorem 1.3 with ǫ = 0 was established by Viada [Vi1] and Galateau [Ga]. Moreover, in this isotrivial setting, Viada proved the analogue of Theorem 1.3 (for positive effective ǫ) in [Vi2, Theorem 1.4], [Vi1, Theorem 1.2], providing in particular new proofs of instances of earlier results by Poonen [Po] and Zhang [Zh1] and extending the work in [RV]. It is worth pointing out that the aforementioned results invoked a different Bogomolov-type bound than the one in Theorem 1.1, established by Galateau [Ga]. Our techniques provide an alternative approach to their results, as long as no complex multiplication is involved. The isotrivial case for ǫ = 0 is completed for curves in arbitrary abelian varieties in [HP] using amongst others techniques n from o-minimality. In the setting of the multiplicative group Gm, Habegger established results of this flavor in arbitrary dimension [Ha1], generalizing a result of Bombieri-Masser- Zannier [BMZ]. Finally, we remark that analogues of Theorem 1.1, Theorem 1.3 and Theorem 1.5 can be formulated for products of non-isogenous elliptic surfaces, as we did in [DM, Theorem 1.4]. Our methods here would yield these results and, in particular, Barroero-Capuano’s [BC2, Theorem 1.1]. We omit them in this article to simplify our exposition.
1.3. Metrized R-divisors on curves. For each t ∈ B(K) with Et smooth, the canonical ˆ height hEt induces a positive definite quadratic form on Et(K) ⊗ R. The height functions hP on B(K), defined by (1.1) for P ∈ E(k), therefore make sense for elements of the finite-dimensional vector space E(k) ⊗ R. In Theorem 4.6, we prove that every nonzero
element X ∈ E(k) ⊗ R gives rise to a continuous, adelic, semi-positive metrization DX of an ample R-divisor on the base curve B, defined over a number field K, with height function ˆ hX (t)= hEt (Xt) for t ∈ B(K) when Et is smooth, satisfying DX · DX = 0. Consequently, we are able to employ results of Moriwaki [Mo2] in our proof of Theorem 1.1. Specifically, we use his arithmetic Hodge index theorem for adelically metrized R-divisors on
curves defined over a number field [Mo2, Corollary 7.1.2] to understand when DX · DY =0 for X,Y ∈ E(k) ⊗ R. Although not needed for Theorem 1.1, we provide another application of Moriwaki’s arith- metic Hodge index theorem [Mo2, Theorem 7.1.1], combined with the continuity of the arith- metic volume function [Mo2, Theorem 5.3.1] and a generalization of Zhang’s fundamental inequality [CLT, Lemme 5.1]. We prove that metrizations of R-divisors on curves with these nice properties (continuous, adelic, semipositive) obey an equidistribution law. This extends the equidistribution theorems of Chambert-Loir, Thuillier, and Yuan [CL1, Th, Yu] and follows a proof strategy originating in [SUZ]: ELLIPTIC SURFACES AND ARITHMETIC EQUIDISTRIBUTION FOR R-DIVISORS ON CURVES 6 Theorem 1.8. Let B be a smooth projective curve defined over a number field K. Fix an ample R-divisor D on B, equipped with an adelic, semipositive and normalized metrization D over K. Let M be any admissible adelic metrization on an R-divisor M over K. For any
(non-repeating) sequence xn ∈ B(K) with hD(xn) → 0, we have D · M h (x ) → . M n deg D Corollary 1.9. Fix any ample R-divisor D on B, equipped with an adelic, semipositive and normalized metrization D over K. For each place v of K and for any (non-repeating) infinite
sequence xn ∈ B(K) with hD(xn) → 0, the discrete probability measures 1 µn = δy | Gal(K/K) · xn| y∈Gal(XK/K)·xn an 1 converge weakly in Bv to the probability measure µD,v = deg D ωD,v. an Here, Bv denotes the Berkovich analytification of the curve B over the complete and alge- braically closed field Cv. Returning to the setting of E → B and its sections, Corollary 1.9 applies to sequences in the base curve B where the specializations of points in E(k) satisfy non-trivial linear relations. For example, generalizing [DM, Corollary 1.2], we show: Theorem 1.10. Let E → B be a non-isotrivial elliptic surface defined over a number field K, and let E be the corresponding elliptic curve over the field k = K(B). Suppose that
P1,...,Pm is a collection of m ≥ 1 linearly independent points in E(k), also defined over K as sections of E → B. Suppose that {tn} ⊂ B(K) is an infinite non-repeating sequence where
(1.6) a1,nP1,tn + a2,nP2,tn + ··· + am,nPm,tn = Otn m−1 for ai,n ∈ Z, with [a1,n : ··· : am,n] → [x1 : ··· : xm] in RP as n →∞. Set
X = x1P1 + ··· + xnPn ∈ E(k) ⊗ R. Then
hX (tn) → 0 for the height function associated to the metrized R-divisor DX . Moreover, for each place v an of K, the Gal(K/K)-orbits of tn in B(K) are uniformly distributed on Bv with respect to the curvature distribution
2 ωX,v = xi − xixj ωPi,v + xixj ωPi+Pj ,v. i ! i Remark 1.11. For nonzero X ∈ E(k) ⊗ R, the height hX will have only finitely many zeros unless a positive real multiple cX is represented by an element of E(k); see Proposition 5.5. On the other hand, there is always an infinite sequence {tn} for which (1.6) is satisfied; see Proposition 5.1. 1.4. Measures at an archimedean place. In [DM], the curvature measures ωP for the metrized divisor DP associated to P ∈ E(k), at an archimedean place, are computed as the pullback by P of a certain (1, 1)-form on E(C), via a dynamical construction. In [CDMZ], it is shown that ωP = db1 ∧ db2 in the Betti coordinates (b1, b2) of P . We explain in Section 8 that elements X ∈ E(k) ⊗ R are also represented by holomorphic curves in the surface E, and the Betti coordinates of X are real linear combinations of the Betti coordinates of points Pi ∈ E(k). We use this to prove that the measure ωX , at a single archimedean place of the number field K, is enough to uniquely determine the pair of points X and −X: Theorem 1.12. Fix X and Y in E(k) ⊗ R and an archimedean place of the number field K. Let ωX and ωY denote the curvature distributions on B(C) at this place for the adelically metrized R-divisors DX and DY . Then ωX = ωY ⇐⇒ X = ±Y. We are grateful to Lars K¨uhne for helping us with the proof of Theorem 1.12: we use the holomorphic-antiholomorphic trick of Andr´e, Corvaja, and Zannier [ACZ, §5] and a transcendence result of Bertrand [Be, Th´eor`eme 5]. A special case of Theorem 1.12 was proved by a different method in [DWY, Proposition 1.9]. 1.5. Example. Let Et be the Legendre elliptic curve defined by y2 = x(x − 1)(x − t) for t ∈ Q\{0, 1}. By filling in the family over t =0, 1, ∞, we obtain an elliptic surface E → B with B = P1 defined over Q. Here k = Q(t). It is easy to see that rank E(k) = 0. However, 1 by choosing any collection of m distinct points x1, x2,...,xm ∈ P (Q) \{0, 1, ∞}, we can construct an elliptic surface E ′ → B′ with rank E′(k′) ≥ m where k′ = Q(B′). Indeed, we let Pxi be a point with constant x-coordinate equal to xi. As the points xi are distinct, the structure of the field extensions ki/k, determined by each Pxi , implies that the points must ′ be independent. We pass to a branched cover B → B so that each Pxi defines a section over B′ and set k′ = Q(B′). These examples were first considered in [MZ1] and the associated ′ C P1 measures ωPxi on B ( ) (or rather, their projections to B = ) were computed in [DWY]. 1.6. Outline of the article. In Section 3, we introduce metrizations on R-divisors on curves defined over a number field and define their intersection numbers. We then prove the equidistribution results, Theorem 1.8 and Corollary 1.9. In Section 4, we prove that each ELLIPTIC SURFACES AND ARITHMETIC EQUIDISTRIBUTION FOR R-DIVISORS ON CURVES 8 nonzero element X ∈ E(k) ⊗ R induces a continuous, adelic, semipositive metrization DX on an ample R-divisor on the base curve B. In Section 5, we study the sequences of small points for the height function hX on B(Q) associated to DX . In Section 6 we lay out the basic properties of the intersection number (X,Y ) 7→ DX · DY as a biquadratic form on the vector space E(k) ⊗ R. In Section 7, we analyze the significance of DX · DY = 0 for nonzero X,Y ∈ E(k) ⊗ R, and we explain how to relate Theorems 1.1, 1.3, and 1.5. Section 8 contains a proof of Theorem 1.12, and we complete the proofs of Theorems 1.1, 1.3, and 1.5 in Section 9. 1.7. Acknowledgements. We spoke with many people about this work, and we are grate- ful to all of them for helpful discussions. Special thanks go to Fabrizio Barroero, Daniel Bertrand, Laura Capuano, Gabriel Dill, Philipp Habegger, Lars K¨uhne, Harry Schmidt, Xinyi Yuan, and Umberto Zannier. 2. Notation Throughout this article, K denotes a number field. We let MK denote its set of places, with absolute values |·|v satisfying the product formula: [Kv:Qv] (2.1) |x|v =1 v∈YMK for all nonzero x in K. Here Kv denotes the completion of K with respect to |·|v. We set [K : Q ] (2.2) r := v v . v [K : Q] For each place v ∈ MK , we let Cv denote the completion of an algebraic closure of Kv. We let B denote a smooth projective curve defined over a number field K. For each an v ∈ MK , we let Bv denote the Berkovich analytification of B over the field Cv. We let DivZ(B) denote the group of divisors on B. Throughout, k denotes the function field K(B). Its places are in one-to-one correspondence with the elements t ∈ B(K), with absolute values given by |f|t = exp(− ordt(f)) for each nonzero f ∈ K(B). 3. R-divisors and equidistribution In this section, we introduce metrizations on R-divisors on curves, following Moriwaki [Mo2]. We show that an equidistribution theorem holds for sequences of small height on a curve B defined over a number field, for adelic semipositive metrizations D associated to an ELLIPTIC SURFACES AND ARITHMETIC EQUIDISTRIBUTION FOR R-DIVISORS ON CURVES 9 R-divisor, extending the equidistribution results for curves of Chambert-Loir, Thuillier, and Yuan [CL1, Th, Yu]. The proofs of Theorem 1.8 and Corollary 1.9 follow a known strategy for equidistribution. We mimic the presentation of Chambert-Loir and Thuillier in [CLT]; they themselves appeal to results of Yuan [Yu] and Zhang [Zh4], building on the ideas that originally appeared in [SUZ]. The key ingredient for passing from Q-divisors to R-divisors is the continuity of the arithmetic volume function on the space of metrized of R-divisors, as proved by Moriwaki [Mo2]. We provide the details for completeness. 3.1. Metrizations of R-divisors on curves. Let B be a smooth projective curve defined over a number field K. Let D = i aiDi be an ample R-divisor on B, with ai ∈ R and D ∈ Div (B) with support in B(K), invariant under the action of Gal(K/K). By rewriting i Z P the sum if necessary, we may assume that each Di is associated to an ample line bundle Li an that extends over the Berkovich analytification Bv for each place v of K. A continuous, adelic metrization for D is a collection of continuous functions an gv : Bv \ supp D → R for v ∈ MK, such that (1) For each v, the locally-defined function ψv := gv + i ai log |fi|v extends continuously to the support of D, where f is a local defining equation for D defined over K; i P i (2) there exists a model (B, D) of (B,D) over the ring of integers OK so that gv is the associated model function for all but finitely many v, or equivalently, the function ψv ≡ 0 at all but finitely many places v for the associated {fi} near each element of supp D. See [Mo2, §0.2] and [CL2, §1.3.2] for the definition of model functions. an The metrization is semipositive if each gv is subharmonic on Bv \ supp D. An R-divisor an D on B and collection of continuous functions gv : Bv \ supp D → R, for v ∈ MK, is said to be admissible if D = D1 − D2 and gv = gv,1 − gv,2 for two adelic, semipositive metrizations on R-divisors Di =(Di, {gi,v}). We write D = D1 −D2 for the data (D, {gv}). An associated height function is given by hD = hD1 − hD2 . We denote this data by D = (D, {gv}). Moriwaki calls D an adelic arithmetic R-divisor of (C0 ∩ PSH)-type on B [Mo2]. It extends Zhang’s notion of an adelic, semipositive metric on a line bundle to R-divisors [Zh4]. Indeed, for D an ample divisor on B associated to a line bundle L, equipped with an adelic metric {k·kv}v, and s a meromorphic section of L with (s)= D, we put gv = − log kskv at each place v of the number field K. an We let ωD,v denote the curvature distribution on Bv associated to the metrization D; by definition, this is a positive measure of total mass deg D, equal to the Laplacian of gv away from supp D. See, for example, [BR] for more information about the distribution-valued ELLIPTIC SURFACES AND ARITHMETIC EQUIDISTRIBUTION FOR R-DIVISORS ON CURVES 10 Laplacian on Berkovich curves. The associated probability measure is denoted by 1 µ := ω . D,v deg D D,v There is an associated height function on B(K) defined by rv ′ (3.1) hD(x) := gv(x ), v∈M | Gal(K/K) · x| ′ XK x ∈Gal(XK/K)·x for x 6∈ supp D. Recall that rv was defined in (2.2). For any rational function φ on B defined over K, and for any real a ∈ R, note that rv ′ hD(x)= (gv − a log |φ|v)(x ) | Gal(K/K) · x| ′ vX∈MK x ∈Gal(XK/K)·x away from (supp D)∪(supp(φ)), from the product formula (2.1). This allows definition (3.1) to extend to the points x ∈ supp D, by choosing any φ so that x ∈ supp(φ) and a so that ′ gv − a log |φ|v extends continuously at x for every v. For an R-divisor D = i bi [xi] with support in B(K), we will write P ′ hD(D ) := bi hD(xi). i X 3.2. Intersection. For divisors D1,D2 ∈ DivZ(B) associated to line bundles L1 and L2, respectively, equipped with continuous, adelic metrics D1 and D2, the arithmetic intersection number is defined in [Zh4] (see also [CL2]) as (3.2) D1 · D2 := hD1 ((s2)) + rv (− log ks2kD2,v) dωD1,v an v∈M Bv XK Z = hD1 ((s2)) + hD2 ((s1)) + rv (− log ks2kD2,v)(dωD1,v − δ(s1)) an v∈M Bv XK Z = hD1 ((s2)) + hD2 ((s1)) + rv (− log ks2kD2,v) ∆(− log ks1kD1,v) an Bv vX∈MK Z = D2 · D1, where si is a meromorphic section of Li defined over K, for i =1, 2, with divisors (s1) and (s2) of disjoint support. For the continuous, adelic metrizations of R-divisors, we extend by R-linearity, so that (3.3) D1 · D2 = hD1 (D2)+ rv gD2,v dωD1,v = D2 · D1. an v∈M Bv XK Z ELLIPTIC SURFACES AND ARITHMETIC EQUIDISTRIBUTION FOR R-DIVISORS ON CURVES 11 Remark 3.1. The intersection number (3.3) coincides with deg(D1 · D2) of [Mo2]. Indeed, [Mo2, Theorem 4.1.3] states that each D can be uniformly approximated by metrizations associated to models, and it is known that the intersection numbersd coincide for these model metrics [Mo1, Proposition 2.1.1]. We say D is normalized if its self-intersection number satisfies D · D =0. Note that any continuous, adelic metrization can be normalized by adding a constant to gv at some place. For each a ∈ R and D = (D, {gv}), we write aD for the pair (aD, {agv}). Normalized metrized divisors D1 and D2 on B are isomorphic if D1 − D2 is principal, meaning that there are rational functions φ1,...,φm ∈ K(B) and real numbers a1,...,am, so that m D1 − D2 = ai (φi), {− log |φi|v}v∈MK . i=1 X Note that by the product formula the height function hD depends only on the isomorphism class of D. We formulate Moriwaki’s arithmetic Hodge-index theorem as follows: Theorem 3.2. [Mo2, Corollary 7.1.2] Suppose D1 and D2 are normalized continuous semi- positive adelic metrizations on ample R-divisors with deg D1 = deg D2. Then D1 · D2 ≥ 0, and D1 · D2 =0 if and only if D1 and D2 are isomorphic. Proof. Set D = D1 − D2, so that the underlying divisor D has degree 0, and D · D = −2 D1 · D2 From [Mo2, Corollary 7.1.2], we have that D·D ≤ 0 with equality if and only if D is principal, up to the addition of a constant to the metrization gv at some place. The normalization of D1 and D2 implies that this constant must be 0. 3.3. Essential minima. Following [Zh4], the essential minimum of the height hD is defined as (3.4) e1(D) := sup inf hD(x), F x∈B(K)\F over all finite subsets F of B(K), and we put e2(D) := inf hD(x). x∈B(K) ELLIPTIC SURFACES AND ARITHMETIC EQUIDISTRIBUTION FOR R-DIVISORS ON CURVES 12 Theorem 3.3. [Zh4, Theorem 1.10] For any adelic, semipositive metrization D of an ample R-divisor D, we have D · D 1 e (D) ≥ ≥ e (D)+ e (D) 1 2 deg D 2 1 2 Proof. Zhang proved the result for ample line bundles equipped with adelic, semipositive metrics [Zh4, Theorem 1.10]. It holds also for metrizations of R-divisors because the height function associated to an R-divisor is a uniform limit of heights associated to Q-divisors, and the intersection number is a bilinear form on metrized divisors. Using the upper bound on D · D in Theorem 3.3, we can extend Theorem 3.2 to: Theorem 3.4. Suppose D1 and D2 are normalized semipositive adelic metrizations on ample R-divisors of the same degree, and suppose the essential minimum of at least one of the Di is 0. Then the following are equivalent: (1) D1 · D2 =0 (2) D1 and D2 are isomorphic (3) hD1 = hD2 on B(K) (4) hD1 = hD2 at all but finitely many points of B(K) (5) there exists an infinite non-repeating sequence tn in B(K) for which lim hD (tn)+ hD (tn) =0. n→∞ 1 2 Proof. We have (1) ⇐⇒ (2) from Theorem 3.2. The definition of the height function, in view of the product formula, implies that (2) =⇒ (3), and we clearly have (3) =⇒ (4). The essential minimum being 0 for D1 or for D2 gives (4) =⇒ (5). Finally, assume (5). Theorem 3.3 implies that e1(Di) ≥ 0, for i =1, 2, because Di is normalized. Therefore, we also have e1(D1 + D2) ≥ 0 for the essential minimum of the sum hD1 + hD2 . The existence of the sequence {tn} thus implies that e1(D1 + D2) = 0. As D1 · D2 ≥ 0 from Theorem 3.2 and Di · Di = 0 for i = 1, 2 by assumption, we apply Zhang’s inequality (Theorem 3.3) to D1 + D2 to obtain 2 D1 · D2 0= e1 D1 + D2 ≥ ≥ 0, deg D1 + deg D2 which allows us to deduce condition (1). We will use the equivalences of Theorem 3.4 repeatedly in our proofs of Theorems 1.1, 1.3, and 1.5. ELLIPTIC SURFACES AND ARITHMETIC EQUIDISTRIBUTION FOR R-DIVISORS ON CURVES 13 3.4. Arithmetic volume. The rest of this section is devoted to the proof of the equidistri- bution results, Theorem 1.8 and Corollary 1.9. For any R-divisor D on B defined over K, we set H0(X,D)= {φ ∈ K(B):(φ)+ D ≥ 0}. For ample D ∈ DivZ(B), the volume of D is 1 vol(D) = lim dim H0(B, kD) = deg D k→∞ k from Riemann-Roch. For a Q-divisor D, the volume can be defined by taking the limit along sequences where kD ∈ Div(B) and again we have vol(D) = deg D when D is ample. The arithmetic volume of an adelically-metrized R-divisor D is defined as follows. We first fix a family of norms on H0(B,D) by −gv(x) kφksup,v = sup |φ(x)|ve , an x∈Bv \supp D for each place v of K. Set µ((H0(B,D) ⊗ A )/H0(B,D)) χ(D)= − log K µ( v Uv) 0 where AK is the ring of adeles, µ is a Haar measureQ on H (B,D) ⊗ AK, and Uv is the unit 0 ball in H (B,D) ⊗ Cv in the induced norm. Then χ(kD) volχ(D) := lim sup 2 . k→∞ k /2 In [Mo2, Theorem 5.2.1], Moriwakic proves that volχ defines a continuous function on a space of continuous, adelic metrizations on R-divisors. As a consequence, he shows that for semipositive metrizations, we have volχ(D) = Dc· D [Mo2, Theorem 5.3.2]. Therefore, Zhang’s inequality (Theorem 3.3) implies that c (3.5) e1(D) ≥ volχ(D)/(2 deg D). for all continuous, semipositive, adelic metrizations of R-divisors on B. c Remark 3.5. This volume function volχ is defined differently than the one studied by Moriwaki in [Mo2], but they coincide. See, for example, Appendix C.2 of [BG] and the discussion on page 615, for the comparisonc of an adelic volume to a Euclidean volume. Proposition 3.6. For all admissible adelic metrizations on an ample R-divisor D on a curve, we have vol (D) e (D) ≥ χ 1 2 deg D c ELLIPTIC SURFACES AND ARITHMETIC EQUIDISTRIBUTION FOR R-DIVISORS ON CURVES 14 Proof. From (3.5), the inequality holds for semipositive metrics. For admissible metrics, we write D = D1 − D2 for semipositive Di and approximate each Di with semipositive adelic metrics on Q-divisors Di,n as n → ∞. Because D is ample, we can assume that D1,n − D2,n is ample for all n. In that setting, we apply [CLT, Lemme 5.1], recalling that vol(D1,n − D2,n) = deg(D1,n − D2,n). The result then follows by uniform convergence of the resulting height functions, so that e1 is continuous, and by continuity of the volume function volχ [Mo2, Theorem 5.2.1]. c 3.5. Proof of Equidistribution. Proof of Theorem 1.8. Fix an ample R-divisor D, equipped with an adelic, semipositive and normalized metrization D. Let xn ∈ B(K) be any (non-repeating) sequence xn ∈ B(K) with hD(xn) → 0. Assume first that M is an adelic, semi-positive metrization on an ample R-divisor M, not necessarily normalized. For each positive integer m, by Zhang’s inequality (Theorem 3.3) applied to (mD)+ M, we have m2 D · D +2m D · M + M · M 2m D · M + M · M lim inf (mhD(xn)+ hM (xn)) ≥ = . n→∞ 2(m deg D + deg M) 2(m deg D + deg M) As the sequence xn is small for D, this gives 2m D · M + M · M lim inf hM (xn) ≥ n→∞ 2(m deg D + deg M) for all m. Letting m go to ∞, we obtain D · M (3.6) lim inf hM (xn) ≥ . n→∞ deg D For the reverse inequality, we choose m large enough so that deg(mD − M) > 0. We can therefore apply Proposition 3.6 to obtain volχ mD − M lim inf(mhD(xn) − hM (xn)) ≥ n→∞ 2 deg( mD − M) c so that volχ mD − M − lim sup hM (xn) ≥ n→∞ 2 deg( mD − M) c ELLIPTIC SURFACES AND ARITHMETIC EQUIDISTRIBUTION FOR R-DIVISORS ON CURVES 15 On the other hand, again assuming that deg(mD − M) > 0, Moriwaki’s [Mo2, Theorem 7.1.1] implies that volχ mD − M ≥ mD − M · mD − M 2 = m (D · D)− 2m D · M + M · M c = −2m D · M + M · M, with the last equality because D is normalized. (Compare [Yu, Theorem 2.2] and [CLT, Proposition 5.3].) Therefore, 2m D · M − M · M lim sup hM (xn) ≤ n→∞ 2 deg(mD − M) for all sufficiently large m. As deg(mD − M) ≈ m deg D as m →∞, we obtain the desired upper bound of D · M (3.7) lim sup hM (xn) ≤ . n→∞ deg D Putting the two inequalities (3.6) and (3.7) together, we have D · M lim hM (xn)= . n→∞ deg D Now suppose that M is admissible. By definition, we can write M = M 1 − M 2 for semipositive M i. By adding and subtracting any choice of ample divisor with semipositive metric, we can assume that each Mi is ample, and we apply the result above to each M i. We have hM = hM 1 − hM 2 and D · M = D · M 1 − D · M 2. The theorem is a consequence of this linearity. Proof of Corollary 1.9. Fix a place v ∈ MK , and let φ be a smooth real-valued function on an Bv . By density as in [Mo2, Theorem 3.3.3] it is enough to consider these functions. We denote by Oφ the trivial divisor on B equipped with the metrization given by gv = φ and gw = 0 for all w =6 v in MK . This metrization is clearly admissible. an Let µn denote the probability measure in Bv supported uniformly on the Galois conjugates of xn. Note that hOφ (xn)= rv φdµn, an ZBv by the definition of the height function. From (3.3), we have D · Ov(φ)= rv φdωD,v. an ZBv ELLIPTIC SURFACES AND ARITHMETIC EQUIDISTRIBUTION FOR R-DIVISORS ON CURVES 16 Applying Theorem 1.8 to M = Ov(φ), we get that 1 [Kv : Qv] [Kv : Qv] lim hOv(φ)(xn)= φdωD,v = φdµD,v, n→∞ deg D [K : Q] an [K : Q] an ZBv ZBv an demonstrating weak convergence of µn to µD,v in Bv . 4. A metrized R-divisor for each element of E(k) ⊗ R Throughout this section, we let E → B be a non-isotrivial elliptic surface defined over a number field K, and let E be the corresponding elliptic curve over the field k = K(B). We denote the zero by O ∈ E(k). As E(k) is finitely generated, we enlarge K if needed so that all sections of E → B are defined over K. Recall that points P1,...,Pm are independent in E(k) if there is no nontrivial relation of the form a1P1 + ··· + amPm = O in E(k) with ai ∈ Z. In this section, we show that each nonzero element X ∈ E(k) ⊗ R naturally gives rise to an adelic, semipositive continuous metrization DP associated to an ample R-divisor DX ; see Theorem 4.6. For P ∈ E(k), these metrizations on R-divisors coincide with the adelically metrized line bundles on B that we studied in [DM]. 4.1. N´eron-Tate heights. Let F be a number field or a function field in characteristic 0. Let E/F be an elliptic curve with origin O, expressed in Weierstrass form as 2 3 2 E = {y + a1xy + a3y = x + a2x + a4x + a6} with discriminant ∆. Denote by hˆE : E(F) → [0, ∞) a N´eron-Tate canonical height function; it can be defined by 1 h(x(nP )) hˆE(P )= lim 2 n→∞ n2 where h is the naive Weil height on P1 and x : E → P1 is the degree 2 projection to the x-coordinate. For each v ∈ MF , recall that Fv denotes the completion of F with respect to |·|v and Cv de- note the completion of the algebraic closure of Fv. The canonical height has a decomposition into local heights, as 1 hˆE(P )= rv λˆE,v(Q) | Gal(F/F) · P | Q∈Gal(XF/F)·P vX∈MF ELLIPTIC SURFACES AND ARITHMETIC EQUIDISTRIBUTION FOR R-DIVISORS ON CURVES 17 for all P ∈ E(F) \{O}, with rv defined by (2.2) in the number field case, and rv = 1 for ˆ function fields. The local heights λE,v are characterized by the three properties [Si3, Chapter 6, Theorem 1.1]: (1) λˆE,v is continuous on E(Cv) \{O} and bounded on the complement of any v-adic neighborhood of O; ˆ 1 (2) the limit of λE,v(P ) − 2 log |x(P )|v exists as P → O in E(Cv); and (3) for all P =(x, y) ∈ E(Cv) with 2 P =6 O, 1 (4.1) λˆ (2 P )=4λˆ (P ) − log |2y + a x + a | + log |∆| . E,v E,v 1 3 v 4 v Property (3) may be replaced with the quasi-parallelogram law ˆ ˆ ˆ ˆ λE,v(P + Q)+ λE,v(P − Q)=2λE,v(P )+2λE,v(Q) (4.2) 1 − log |x(P ) − x(Q)| + log |∆| v 6 v ˆ under the assumption that none of P , Q, P + Q, nor P − Q is equal to O. Note that λE,v is independent of the choice of Weierstrass equation for E over F. It is useful to recall also the triplication formula; if 3P =6 O, then 2 (4.3) λˆ (3P )=9λˆ (P ) − log |(3x4 + b x3 +3b x2 +3b x + b )(P )| − log |∆| , E,v E,v 2 4 6 8 v 3 v where bi are the usual Weierstrass quantities; see e.g. [Si3, pg. 463]. 4.2. Metrized divisors for elements of E(k). Fix non-torsion P ∈ E(k). Define ˆ DP := λE,ordγ (P )[γ]. γ∈XB(K) ˆ We remark that λE,ordγ (P ) ∈ Q [La, Chapter 11, Theorem 5.1], so DP is a Q-divisor on B. As P is defined over K, the divisor is Gal(K/K)-invariant. In [DM, Theorem 1.1] we established that DP can be equipped with an adelic, semipositive, continuous and normalized metrization (4.4) DP := (DP , {λP,v}v∈MK ) ˆ an over the number field K, where λP,v denotes the extension of t 7→ λEt,v(Pt) to Bv . It follows that the associated height functions satisfy ˆ hP (t) := hDP (t)= hEt (Pt) for all t ∈ B(K) for which Et is smooth. Both minima e1(DP ) and e2(DP ) (defined in §3.3) are equal to 0 [DM, Proposition 4.3]. ELLIPTIC SURFACES AND ARITHMETIC EQUIDISTRIBUTION FOR R-DIVISORS ON CURVES 18 For O ∈ E(k), we set DO := (0, 0), the trivial divisor with all functions gv = 0. For torsion points T =6 O ∈ E(k), the metrized ˆ divisor DT can also be defined by (4.4), with λT,v(t) := λEt,v(Tt) for all t ∈ B(K) with Et smooth ◮ and Tt =6 O ◭ . The following proposition is key for the passage from E(k) to E(k) ⊗ R. Proposition 4.1. The metrized divisor DP is well defined for P in E(k)/E(k)tors, up to isomorphism. Moreover, for each m ≥ 1 and any set of independent points P1,...,Pm ∈ E(k) and integers a1,...,am, the following metrized divisors are isomorphic: m 2 (4.5) Da1P1+···+amPm ≃ ai − ai aj DPi + aiajDPi+Pj i=1 ! 1≤i Lemma 4.3. Fix P ∈ E(k). For each nonzero m ∈ Z with |m| ≥ 2 such that mP =6 O, there exist h ∈ K(B) and constant c ∈ Q so that ˆ 2 ˆ λEt,v(mPt)= m λEt,v(Pt)+ c log |h(t)|v at every place v and for each t ∈ B(K) such that Et is smooth. If P is torsion of order m ≥ 2, we have ˆ λEt,v(Pt)= c log |h(t)|v, for some c ∈ Q and h ∈ K(B). Proof. Upon replacing P by −P , it suffices to prove the statement for m ≥ 2. The duplication formula (4.1) provides the desired result for m = 2, assuming that 2P =6 O. Now fix any m ≥ 3 and P ∈ E(k), and assume that mP =6 O, (m − 1)P =6 O and (m − 2)P =6 O. Then the quasi-parallelogram law (4.2) implies ˆ ˆ ˆ ˆ (4.6) λEt,v(mPt) = 2 λEt,v((m − 1)Pt)+2 λEt,v(Pt) − λEt,v((m − 2)Pt) 1 − log |x((m − 1)P ) − x(P )| + log |∆ | , t t v 6 t v ELLIPTIC SURFACES AND ARITHMETIC EQUIDISTRIBUTION FOR R-DIVISORS ON CURVES 19 for each t ∈ B(Cv) such that Et is smooth and mPt =6 Ot,(m−1)Pt =6 Ot and (m−2)Pt =6 Ot ˆ and therefore for all t ∈ B(Cv) by the continuity of the local height t 7→ λEt,v(Pt) ∈ R∪{±∞}. The desired relation, for all non-torsion points and for all m ≥ 3, then follows from (4.6) by an easy induction. Now suppose that 2P = O with P =6 O. Then we have 3P = P =6 O, and the triplication formula (4.3) implies that ˆ ˆ ˆ λEt,v(3Pt)=9λEt,v(Pt) − c log |h(t)|v = λEt,v(Pt), for a constant c ∈ Q and h ∈ K(B) and for all but finitely many t. The equation then ˆ holds for all t ∈ B(Cv) by the continuity of the local heights, and it implies that λEt,v(Pt)= c 8 log |h(t)|v. For a torsion point P of order 3, we have 2P = −P =6 O, so we may apply the duplication formula (4.1) to see that ˆ ˆ ˆ ˆ λEt,v(2Pt)=4λEt,v(Pt) − c log |h(t)|v = λEt,v(−Pt)= λEt,v(Pt), ˆ c for a constant c ∈ Q and h ∈ K(B). It follows that λEt,v(Pt)= 3 log |h(t)|v. Finally, suppose that P is torsion of order n ≥ 4, and note that (n − 1)P = −P =6 O, (n − 2)P =6 O and (n − 3)P =6 O. We infer from (4.6) with 3 ≤ m ≤ n − 1 inductively that ˆ 2ˆ ˆ ˆ λEt,v((n − 1)Pt)=(n − 1) λEt,v(Pt) − c log |h(t)|v = λEt,v(−Pt)= λEt,v(Pt) ˆ c for a rational function h ∈ K(B) and c ∈ Q, so that λEt,v(Pt)= n2−2n log |h(t)|v. Proof of Proposition 4.1. Lemma 4.3 implies that DP ≃ DO for every torsion point P ∈ E(k). Furthermore, for any non-torsion point P , Lemma 4.3 also implies that 2 DaP ≃ a DP for all a ∈ Z, demonstrating (4.5) for m = 1. Therefore, if P is non-torsion and Q is torsion of order n ≥ 2, we have 1 1 D ≃ D = D ≃ D . P +Q n2 n(P +Q) n2 nP P This proves that the metrized divisors depend only on the class in E(k)/E(k)tors, up to isomorphism. Now fix any m ≥ 2, and any collection of independent points P1,...,Pm ∈ E(k) and integers a1,...,am. Define a divisor on B by m ′ 2 D = ai − ai aj DPi + aiaj DPi+Pj , i=1 ! 1≤i at all places v of K and for all but finitely many t ∈ B(Cv). Lemma 4.4. Let P, Q, R ∈ E(k) be independent points defined over K. Then, there is a rational function fP,Q,R ∈ K(B) such that ˆ ˆ ˆ ˆ λEt,v(Pt + Qt + Rt) = λEt,v(Pt + Rt)+ λEt,v(Pt + Qt)+ λEt,v(Qt + Rt) ˆ ˆ ˆ −λEt,v(Pt) − λEt,v(Qt) − λEt,v(Rt) − log |fP,Q,R(t)|v. for all t ∈ B(K) such that Et is smooth and all v ∈ MK . Proof. The proof follows by applying the quasi-parallelogram law (4.2) for the pairs {P + R, Q}, {P, R−Q}, {P +Q, R} and {R, Q} and taking an alternating sum as in [Si6, Theorem 9.3]. Lemma 4.5. Fix independent P, Q ∈ E(k). For each (a, b) ∈ Z2 \{(0, 0)}, there is rational function ha,b ∈ K(B) such that ˆ 2 ˆ ˆ λEt,v(aPt + bQt) = (a − ab)λEt,v(Pt)+ abλEt,v(Pt + Qt)+ 2 ˆ (b − ab)λEt,v(Qt) − log |ha,b|v, for all t ∈ B(K) such that Et is smooth and all v ∈ MK . Proof. The proof follows from the quasi-parallelogram law by an easy induction. Lemma 4.3 provides the desired result if one of a or b is 0. Next we will show that for each n ∈ Z there is a rational function g ∈ K(B) so that ˆ 2 ˆ ˆ λEt,v(nPt + Qt)=(n − n)λEt,v(Pt)+ nλEt,v(Pt + Qt) (4.8) ˆ + (1 − n)λEt,v(Qt) − log |g|v. Replacing P by −P we may assume that n ≥ 1. For n = 1 the statement is clear. For n ≥ 1, the quasi-parallelogram law (4.2) implies that ˆ ˆ λEt,v((n + 1)Pt + Qt) = λEt,v(nPt +(P + Q)t) ˆ ˆ ˆ = 2λEt,v(nPt)+2λEt,v(Pt + Qt) − λEt,v((n − 1)Pt − Qt) 1 − log |x(nP ) − x(P + Q )| + log |∆| t t t v 6 v ELLIPTIC SURFACES AND ARITHMETIC EQUIDISTRIBUTION FOR R-DIVISORS ON CURVES 21 and (4.8) follows inductively from Lemma 4.3. Using (4.8) we now have a rational function h ∈ K(B) so that ˆ 2 ˆ ˆ (4.9) λEt,v(aPt + bQt) = (a − a)λEt,v(Pt)+ aλEt,v(Pt + bQt)+ ˆ (1 − a)λEt,v(bQt) − log |h|v. The lemma then follows by another application of (4.3) and (4.8), exchanging the roles of P and Q. Finally, a simple induction using Lemmas 4.4 and 4.5 implies that for any m ≥ 2, and for any integers a1,...,am, the equality (4.7) holds, for some rational f. This completes the proof. 4.3. Metrized divisors for elements of E(k) ⊗ R. Fix nonzero X ∈ E(k) ⊗ R. Choose independent points P1,...,Pm ∈ E(k) that define a basis for E(k) ⊗ R, and write X = x1P1 + ··· + xmPm with xi ∈ R. With a slight abuse of notation, we identify the two isomorphic metrized divisors in Proposition 4.1 and define an adelically metrized R-divisor on B(K), over the number field K, by m 2 (4.10) DX := xi − xi xj DPi + xixj DPi+Pj i=1 1≤i Theorem 4.6. Fix nonzero X ∈ E(k)⊗R. The metrized divisor DX of (4.10) is continuous, adelic, semipositive and normalized. The degree of the underlying R-divisor DX is hˆE(X) > 0. Its associated height function satisfies ˆ (4.12) hX (t)= hEt (Xt) for all t ∈ B(K) with smooth fiber Et. Further, up to isomorphism, DX is independent of the choice of basis for E(k). Proof. Fix x1,...,xm ∈ R and choose sequences of rational numbers an,i/an,0 → xi for i =1,...,m. From Proposition 4.1 we know that the functions 1 m a2 − a a λ (t)+ a a λ (t) a2 n,i n,i n,j Pj ,v n,i n,j Pi+Pj ,v n,0 i=1 1≤i Dan,1P1+···+an,mPm · Dan,1P1+···+an,mPm =0, for all n ∈ N. In view of Proposition 4.1 we then have 2 1 m a2 − a a D + a a D =0, a4 n,i n,i n,j Pj n,i n,j Pi+Pj n,0 i=1 1≤i section, we constructed metrized R-divisors DX and associated height functions hX for each element X ∈ E(k) ⊗ R. In this section, we look at the sets of “small” points for the height hX . We conclude the section with a proof of Theorem 1.10. 5.1. Small sequences exist. For an adelic, continuous, semipositive, and normalized metrized R-divisor D, an infinite sequence {tn} ⊂ B(K) is said to be small if hD(tn) → 0 as n →∞. Proposition 5.1. For every nonzero X ∈ E(k) ⊗ R, there exist small sequences for DX , so that the essential minimum is e1(DX )=0. More precisely, write X = x1P1 + ···+ xmPm for xi ∈ R and independent Pi ∈ E(k), and choose integers ai,n for i = 1,...,m and n ∈ N so m−1 that [a1,n : ··· : am,n] → [x1 : ··· : xm] as n → ∞ in the real projective space RP . Then there exists an infinite non-repeating sequence of points tn ∈ B(K) at which (a1,nP1 + ··· + am,nPm)tn is torsion in the fiber Etn (K). Moreover, for any such sequence {tn} ⊂ B(K) we have hX (tn) → 0 as n →∞. To prove Proposition 5.1, we begin with a well-known statement that follows from Silver- man’s specialization theorem [Si1]. Lemma 5.2. Fix any set of independent points P1,...,Pm in E(k), and let h be any Weil height function on B associated to a divisor of degree 1. The set of all t for which there exist integers a1,...,am, not all zero, such that a1P1,t + ··· + amPm,t = Ot in Et has bounded h-height. Proof. For each non-torsion point Q ∈ E(k) we have [Si1] ˆ h(Qt) ˆ lim = hE(Q) > 0. h(t)→∞ h(t) Because det(hPi,PjiE)i,j > 0, it follows that the set R(P1,...,Pm)= {t ∈ B(K) : det(hPi,t,Pj,tit)=0} ELLIPTIC SURFACES AND ARITHMETIC EQUIDISTRIBUTION FOR R-DIVISORS ON CURVES 24 also has bounded height. This set R(P1,...,Pm) clearly contains the set of t at which the points become linearly dependent. Proof of Proposition 5.1. Write X = x1P1 + ··· + xmPm for independent P1,...,Pm ∈ E(k) and x1,...,xm ∈ R. Fix a sequence of positive integers Mn → ∞ as n → ∞. For i = 1,...,m, choose any sequence of integers ai,n so that ai,n/Mn → xi as n →∞, and set Qn = a1,nP1 + ... + am,nPm ∈ E(k), so that 1 Q → X in E(k) ⊗ R. Mn n Consider the set Tor(Qn)= {t ∈ B(K): Qn,t is torsion in Et} For each n, the set Tor(Qn) is infinite; in fact, it is dense in B(C)[DM, Proposition 6.2] [Za, §III.2 and Notes to Chapter III]. Moreover, from Lemma 5.2, this set has bounded height in the base curve B with respect to any chosen Weil height h, and the height is bounded independent of n. Therefore, from [Si1, Theorem A], we can find H > 0 so that hPi (t) ≤ H and hPi+Pj (t) ≤ H for all t ∈ n Tor(Qn) and for all i, j. From the formula for the height h given in (4.11) and the formula for the height of Q S X n appearing in Proposition 4.1, we have the following. For any given ε> 0, there exists N > 0 so that m 2 1 2 ai,n ai,n aj,n hX (t) − 2 hQn (t) = xi − xi xj − 2 + hPi (t) Mn Mn Mn Mn ! i=1 j6=1 j6=i X X X ai,naj,n + xixj − 2 hPi+Pj (t) <ε Mn ! 1≤i hX (tn) → 0 as n →∞. 5.2. Characterization of small sequences. Here, we observe that small sequences for real points X ∈ E(k) ⊗ R always arise from a construction similar to that of Proposition 5.1, where relations between the generators are “almost” satisfied. We will use this next proposition in the proof of Theorem 7.1. ELLIPTIC SURFACES AND ARITHMETIC EQUIDISTRIBUTION FOR R-DIVISORS ON CURVES 25 Proposition 5.3. Let M be a torsion-free subgroup of E(k) of rank m, generated by S1,...,Sm. Set hM (t) = det(hSi,t,Sj,tit), for the N´eron-Tate bilinear form h·, ·it on the fiber Et(K). For a non-repeating infinite sequence tn ∈ B(K), the following are equivalent (1) lim infn→∞ hM (tn) = 0; (2) there is a non-zero X ∈ M ⊗ R such that lim infn→∞ hX (tn)=0. (3) there are sequences of points si,n ∈ Etn (K), for i =1,...,m, satisfying ˆ lim inf max hEt (si,n) =0 n→∞ i n and so that the points S1,tn − s1,n,..., Sm,tn − sm,n satisfy a linear relation over Z in Etn (K). This proposition relies heavily on Silverman’s specialization results [Si1, Theorem A and Theorem B]. We point out that [Si1, Theorem B] holds for real points X ∈ E(k) ⊗ R by the bilinearity of the N´eron-Tate pairing. We begin with a lemma. Lemma 5.4. Assume we are in the setting of Proposition 5.3. Assume further that there are sequences of points si,n ∈ Etn (K), for i =1,...,m, satisfying ˆ sup max hEtn (si,n) < ∞ n i for which the points S1,tn − s1,n,..., Sm,tn − sm,n satisfy a linear relation over Z in Etn (K). Then the sequence {tn} will have bounded height in B(K) with respect to any Weil height on B. Proof. Fix any Weil height h on B(K) of degree 1. Consider the m × m matrix An := (hSi,tn − si,n, Sj,tn − sj,nitn )i,j where h·, ·itn is the N´eron-Tate inner product on the fiber Etn (K). Our assumption implies that det An =0 for all n. Assume that h(tn) → ∞. Then by Silverman’s specialization theorem [Si1, Theorem B] we have hSi,tn ,Sj,tn itn →hSi,SjiE, h(tn) ELLIPTIC SURFACES AND ARITHMETIC EQUIDISTRIBUTION FOR R-DIVISORS ON CURVES 26 as n →∞ for all i, j =1,...,m. On the other hand, the bounded height of the perturbations si,n and the Cauchy-Schwarz inequality for h·, ·itn implies that ˆ ˆ hs ,s i hEtn (si,n)hEtn (sj,n) i,n j,n tn ≤ → 0. h(t ) q h(t ) n n Using Silverman’s specialization [Si1 , Theorem A] we also have ˆ ˆ hS ,s i hEtn (Si,tn )hEtn (sj,n) i,tn j,n tn ≤ → 0. h(t ) q h(t ) n n Combining these estimates, we obtain det An 0= m −→ det(hSi,SjiE)i,j =06 , (h(tn)) which is a contradiction. Proof of Proposition 5.3. Assume condition (2). Let X ∈ M ⊗ R be nonzero and {tn} a sequence for which lim inf hX (tn)=0. Write X = x1S1 + ··· + xℓSℓ, for xi ∈ R not all equal n→∞ to 0. After reordering the points Si we may assume that x1 =6 0. Notice that ˆ hEtn (Xtn ) hXtn ,S2,tn i ··· hXtn ,Sℓ,tn i hS ,X i hˆ (S ) ··· hS ,S i 2,tn tn Etn 2,tn 2,tn ℓ,tn 2 det . . . = x1 hM (tn), . . . hS ,X i hS ,S i ··· hˆ (S ) ℓ,tn tn ℓ,tn 2,tn Etn ℓ,tn which easily follows by subtracting from the first column the sum of xi times the j-th column over all j =2,...,ℓ and then subtracting from the first row the sum of xi times the i-th row over all i =2,...,ℓ. Expanding the determinant along the first column we get ℓ 2 (5.1) x1 hM (tn)= hX (tn) f1,n + hSj,tn ,Xtn ifj,n, j=2 X where for all n ∈ N the fj,n are polynomial functions of the quantities hSj,tn ,Xtn i and hSj,tn ,Sk,tn i for j, k = 2,...,ℓ. Passing to a subsequence of {tn} we have lim hX (tkn ) = 0. n→∞ In particular, since X is non-trivial, [Si1, Theorem B] yields that {h(tkn )}n∈N is a bounded sequence. Using then [Si1, Theorem A], the functoriality of heights and the Cauchy-Schwarz inequality we get (5.2) max{|f1,kn |,..., |fℓ,kn |} ≤ L, for some L> 0. Moreover, for all j =2,...,ℓ and all n ∈ N we have 2 ˆ ˆ (5.3) |hSj,t ,Xt i| ≤ hEt (Sj,t )hEt (Xt ) ≤ L hX (tkn ) → 0. kn kn kn kn kn kn ELLIPTIC SURFACES AND ARITHMETIC EQUIDISTRIBUTION FOR R-DIVISORS ON CURVES 27 Our assumption on X together with (5.1), (5.2) and (5.3) yield lim inf hM (tn)=0, n→∞ proving that condition (1) holds. Now assume (1). Let At =(hSi,t,Sj,ti)i,j, so that hM (t) = det At, and consider the family of quadratic forms ˆ qt(~z) := hEt (z1S1,t + ··· zmSm,t) m 2 ˆ ⊤ = zk hEt (Sk,t)+2 zizjhSi,t,Sj,ti = ~zAt ~z , i lim inf det Atn =0, n→∞ so, if λn is the smallest eigenvalue of Atn , then lim inf λn =0. n→∞ Let ~vn =(v1,n,...,vm,n) =6 0 be an eigenvector of Atn corresponding to λn. Then ⊤ 2 ~vn Atn ~vn = λn||~vn|| , so that ~vn (5.4) lim inf qtn = lim inf λn =0. n→∞ ||~v || n→∞ n Passing to a subsequence of the {tn}, we have limn→∞ hM (tn) = 0, and passing to a further subsequence, we may set ~v ~x := lim n ∈ Rm \{~0}. n→∞ ||~vn|| By [Si1, Theorem B], the height of {tn} is bounded with respect to any choice of Weil height on B (because det(hSi,SjiE) =6 0). In view of [Si1, Theorem A], we get that the sequences {hSi,tn ,Sj,tn i}n for i, j =1,...,ℓ are bounded. Thus (5.4) yields lim qtn (~x)=0. n→∞ In other words, for X = x1S1 + ··· + xmSm we have limn→∞ hX (tn) = 0, providing condition (2). Assuming (2), we now prove (3). Reordering the points and rescaling X if necessary, we may assume that x1 = 1. Passing to a subsequence, we have ˆ (5.5) hEtn (S1,tn + x2S2,tn + ··· + xmSm,tn ) → 0, ELLIPTIC SURFACES AND ARITHMETIC EQUIDISTRIBUTION FOR R-DIVISORS ON CURVES 28 as n →∞. Let a2,n,...,am,n be infinite sequences of integers satisfying ai,n/n → xi for each ˆ i =2,...,m. As hE(X) =6 0, we have by Silverman specialization [Si1, Theorem B] that the sequence {tn} has bounded height in B. Invoking [Si1, Theorem A] we get that all sequences ˆ {hSi,tn ,Sj,tn iEtn }n∈N are bounded. Using the fact that each hEtn (·) defines a quadratic form on Etn (K), line (5.5) yields 1 (5.6) hˆ S + (a S + ··· + a S ) → 0. Etn 1,tn n 2,n 2,tn ℓ,n ℓ,tn Since K is algebraically closed we may find sn ∈ Etn (K) so that (5.7) nsn = a2,nS2,tn + ··· + aℓ,nSℓ,tn . Letting s1,n := S1,tn + sn and si,n := Otn for all i =2,...,m, equation (5.6) yields ˆ hEtn (si,n) → 0 for each i =1,...,m. Moreover by (5.7) we have that the set {S1,tn − s1,n,S2,tn ,...,Sℓ,tn } is linearly dependent in Etn for every n. Last, we assume condition (3) and prove (2). Pass to a subsequence so that ˆ lim max hEt (si,n) =0. n→∞ i n Choose sequences of integers ai,n for i =1,...,m, not all 0, so that a1,n(S1,tn − s1,n)+ ··· + am,n(Sm,tn − sm,n)= Otn for all n. Now, letting Mn = maxi ai,n, we can pass to a further subsequence so that ai,n → xi ∈ R Mn as n →∞ for each i, with at least one xi nonzero. This implies that (5.8) 1 1 hˆ (a S + ··· + a S ) = hˆ (a s + ··· + a s ) → 0 Etn M 1,n 1,tn m,n m,tn Etn M 1,n 1,n m,n m,n n n as n →∞. Finally set X = x1S1 + ... + xmSm. From Lemma 5.4, we know that the sequence {tn} has bounded height and by [Si1, Theorem ˆ A] we get that the sequences {hEtn (Si,tn )}n are bounded. Therefore, from the definition of hX in (4.11), line (5.8) implies that hX (tn) → 0. ELLIPTIC SURFACES AND ARITHMETIC EQUIDISTRIBUTION FOR R-DIVISORS ON CURVES 29 5.3. Proof of Theorem 1.10. From Theorem 4.6, we know that DX is a continuous, adelic, semipositive, and normalized metrization on an ample R-divisor. Thus, Corollary 1.9 applies to sequences with small height for hX . From Proposition 5.1, we have hX (tn) → 0 along any sequence tn for which i ri,nPi,t = Ot with ri,n ∈ Q satisfying ri,n → xi. The formula for ω at each place follows from the definition of D in (4.10). This completes the proof. X,v P X 5.4. Height 0. As we shall see, it follows from Theorem 1.1 that although small sequences exist as in Proposition 5.1, we don’t always have sequences with height 0: Proposition 5.5. Fix nonzero X ∈ E(k) ⊗ R. There exist infinitely many t ∈ B(K) for which hX (t)=0 if and only if there exists a real c > 0 so that cX is represented by an element of E(k). Proof. Suppose first that cX is represented by an element P ∈ E(k) for some real c > 0. 1 Then hX (t)= c2 hP (t) at all t, so that hX (t) = 0 whenever Pt is torsion in Et. This holds at infinitely many points t ∈ B(K). (See, e.g., [DM, Proposition 6.2].) For the converse, write X = x1P1 + ··· + xmPm for independent P1,...,Pm ∈ E(k) and xi ∈ R, and assume that hX (t) = 0 for infinitely many t. We can rewrite X as X = α1Q1 + ··· + αsQs for α1,...,αs ∈ R a basis for the span of {x1,...,xm} over Q and Q1,...,Qs ∈ E(k). For s = 1, we see that are we back in the setting where a multiple of X is represented by an element of E(k), so we may assume s > 1. But, for each t where hX (t) = 0, we must have that Xt = 0 in Et(Q)⊗R. By the choices of the αi, this means that each of the specializations Qi,t must be 0 in Et(Q) ⊗ R. (Compare [Mo2, Lemma 1.1.1].) In other words, the points Q1,...,Qs are simultaneously torsion at infinitely many t. From Theorem 1.1, combined with (1.4), this implies that each pair Qi and Qj is linearly related. (Alternatively, here one could use the main results of [MZ1, MZ2].) Thus we infer that X = c Q for some c ∈ R and Q ∈ E(k). 6. The intersection number as a biquadratic form on E(k) ⊗ R Let E → B be a non-isotrivial elliptic surface defined over a number field K, and let E be the corresponding elliptic curve over the field k = K(B). Recall that we can enlarge the number field K, if necessary, to ensure that each section P : B → E is defined over K. For each P ∈ E(k), a metrized divisor DP is defined by (4.4). We extended this definition to elements X ∈ E(k) ⊗ R with the definition (4.10). In this section, we study the basic properties of the Arakelov-Zhang intersection number (X,Y ) 7→ DX · DY ELLIPTIC SURFACES AND ARITHMETIC EQUIDISTRIBUTION FOR R-DIVISORS ON CURVES 30 as a biquadratic form on the finite-dimensional vector space E(k) ⊗ R. For X,Y ∈ E(k) ⊗ R, consider the metrized divisor 1 (6.1) D := D − D − D , hX,Y i 2 X+Y X Y of degree equal to the N´eron-Tate inner product of X and Y , 1 hX,Y i = hˆ (X + Y ) − hˆ (X) − hˆ (Y ) . E 2 E E E Our goal in this section is to prove: Proposition 6.1. Fix X,Y,Z ∈ E(k) ⊗ R. The following hold: (1) DX · DY = DY · DX ≥ 0, (2) DX · DY +Z + DX · DY −Z =2 DX · DY +2 DX · DZ, (3) DX · DX+Y = DX · DY , and 2 (4) for all x ∈ R, we have DX · DxY = x DX · DY . For each X ∈ E(k) ⊗ R the map Y 7→ DX · DY defines a positive semidefinite quadratic form on E(k) ⊗ R, induced by the bilinear form (Y,Z) 7→ DX · DhY,Zi. We begin with a lemma. Lemma 6.2. We have DX · DY ≥ 0 for all X,Y ∈ E(k) ⊗ R. Proof. From Theorem 4.6, both DX and DY are normalized, semipositive, continous adelic metrized divisors on B over K, so the lemma follows immediately from Theorem 3.2. Or we can see it as a consequence of Theorem 1.8, because the height functions satisfy hX , hY ≥ 0 at all points of B(K). The following lemma is a version of the Cauchy-Schwarz inequality. Lemma 6.3. For each X ∈ E(k) ⊗ R, the intersection (Y,Z) 7→ DX · DhY,Zi is bilinear in Y,Z ∈ E(k) ⊗ R. Moreover, 2 (DX · DhY,Zi) ≤ (DX · DY )(DX · DZ ) for all X,Y,Z ∈ E(k) ⊗ R. ELLIPTIC SURFACES AND ARITHMETIC EQUIDISTRIBUTION FOR R-DIVISORS ON CURVES 31 Proof. Fix X,Y,Z,W ∈ E(k) ⊗ R, and a, b ∈ R. It is easy to see that DX · DhY,Zi = DX · DhZ,Y i. Then definitions (6.1) and (4.10) give 1 D · D = D · D − D · D − D · D X hY,aZ+bW i 2 X Y +aZ+bW X Y X aZ+bW 1 = D · (1 − a − b)D +(a2 − a − ab)D +(b2 − b − ab)D 2 X Y Z W