HEIGHT OF VARIETIES OVER FINITELY GENERATED FIELDS ´ IGNACIO BURGOS GIL, PATRICE PHILIPPON, AND MART´IN SOMBRA JOSE Abstract. We show that the height of a variety over a finitely generated field of characteristic zero can be written as an integral of local heights over the set of places of the field. This allows us to apply our previous work on toric varieties and extend our combinatorial formulae for the height to compute some arithmetic intersection numbers of non toric arithmetic varieties over the rational numbers. Introduction In [Mor00, Mor01a], Moriwaki introduced a notion of height for cycles over a finitely generated extension of Q. Using this definition, it was possible to extend several central results about cycles over a number field to cycles over a finitely generated extension of Q. These results include Northcott’s theorem on the finiteness of cycles with bounded degree and height, the Manin-Mumford and the Bogomolov conjectures, Zhang’s theorem on successive algebraic minima, and the equidistribution properties of Galois orbits of points of small height [Mor00, Mor01a, Mor01b, YZ13]. This notion of height is defined as follows. Let B be an arithmetic variety, that is, a normal flat projective scheme over Spec(Z), of relative dimension b. Set K = K(B) for its function field, which is a finitely generated extension of Q of transcendence degree b. Let Hi , i = 1, . . . , b, be a family of nef Hermitian line bundles on B. Let π : X → B be a dominant morphism of arithmetic varieties and denote by X the fibre of π over the generic point of B, which is a variety over K. Let Y be a prime cycle of X of dimension d. Let Y be the closure of Y in X and Lj , j = 0, . . . , d, a family of semipositive Hermitian line bundles on X . Moriwaki defines the height of Y relative to this data as the arithmetic intersection number in the sense of Gillet-Soul´e given by hπ∗ H 1 ,...,π ∗ H ,L ,...,L 0 d b (Y), (0.1) see [Mor00, Mor01a] for details. It is well known that this arithmetic intersection number can be written as a sum over the places of Q of local heights of the fibre of Y over the generic point of Spec(Z), see for instance [BPS14, § 1.5]. However, for points in a projective space and the canonical metric, Moriwaki showed that this arithmetic intersection number is also equal to an integral of local heights over a measured set of places of K [Mor00, Proposition 3.2.2]. Date: November 28, 2014. 2010 Mathematics Subject Classification. Primary 14G40; Secondary 11G50, 14M25. Key words and phrases. Metrized line bundle, height of varieties, toric variety, Mahler measure. Burgos Gil was partially supported by the MICINN research project MTM2010-17389. Philippon was partially supported by the CNRS international project for scientific cooperation (PICS) “G´ eom´ etrie diophantienne et calcul formel” and the ANR research project “Hauteurs, modularit´ e, transcendance”. Sombra was partially supported by the MINECO research project MTM201238122-C03-02. 1 2 BURGOS GIL, PHILIPPON, AND SOMBRA In this paper, we extend Moriwaki’s result to a cycle Y of arbitrary dimension and general semipositive metrics, to show that the height of Y is equal to an integral of local heights over this set of places of K (Theorem 2.4). This allows us to apply our previous work on toric varieties in [BPS14] and extend our combinatorial formulae for the height to some arithmetic intersection numbers of non toric arithmetic varieties. More explicitly, let X → B be a dominant morphism of arithmetic varieties as above, and such that its generic fibre X is a toric variety over K of dimension n. For simplicity, suppose that L0 = · · · = Ln = L. The semipositive Hermitian line bundle L defines a polytope ∆ in a linear space of dimension n and, for each place w of K, a concave function ϑw : ∆ → R called the w-adic “roof function”, see § 3 for details and pointers to the literature. By [BPS14, Theorem 5.1.6], the w-adic local height of X is given by (n + 1)! times the integral over ∆ of this concave function. Combining this with Theorem 2.4, we derive a formula for the corresponding arithmetic intersection number as an integral of the function (w, x) 7→ ϑw (x) over the product of the polytope and the set of places of K (Corollary 3.1). Furthermore, we can define a global roof function ϑ : ∆ → R by integrating the local ones over the set of places of K. Then, in this case, the arithmetic intersection number (0.1) is also equal to (n + 1)! times the integral of ϑ over the polytope (Corollary 3.4). As an application, we give in § 4 an explicit formula for the case of translates of subtori of a projective space and canonical metrics (Corollary 4.2). The obtained integrals reduce, in some instances, to logarithmic Mahler measures of multivariate polynomials. Acknowledgements. Part of this work was done while the authors met at the Universitat de Barcelona, the Instituto de Ciencias Matem´aticas (Madrid) and the Institut de Math´ematiques de Jussieu (Paris). We are thankful for their hospitality. We also thank Julius Hertel and the referee for their useful comments. 1. Fields with product formula from arithmetic varieties In [Gub03, Example 11.22], Gubler observed that, for an arithmetic variety equipped with a family of nef Hermitian line bundles, one can endow its function field with a measured set of places satisfying the product formula. In this section, we explain the details of this construction and, as an example, we explicit it for the projective space and the universal line bundle equipped with the canonical metric. We refer to [BPS14, Chapter 1] and [BMPS12, § 3] for the background for this section on metrized line bundles and their associated measures and heights. Let B be an arithmetic variety, which means that B is a normal flat projective scheme over Spec(Z). We denote by b the relative dimension of B and by K = K(B) its function field, which is a finitely generated extension of Q of transcendence degree b. For i = 1, . . . , b, let Hi = (Hi , k · ki ) be a Hermitian line bundle on B, that is, a line bundle Hi on B equipped with a continuous metric on the complexification Hi,C over B(C), invariant under complex conjugation. We will furthermore assume that each Hi is nef in the sense of [Mor00, § 2] or [BMPS12, Definition 3.18(3)]. This amounts to the conditions: (1) the metric k · ki is semipositive, namely it is the uniform limit of a sequence of smooth semipositive metrics as in [Mai00, Definition 4.5.5] or [BPS14, Definition 1.4.1]; (2) the height of every integral one-dimensional subscheme of B with respect to Hi is nonnegative. Let B (1) denote the set of hypersurfaces of B, that is, the integral subschemes of B of codimension 1. Let V ∈ B (1) . By [Zha95, Theorem 1.4(a)] or [Mor00, HEIGHT OF VARIETIES OVER FINITELY GENERATED FIELDS 3 Proposition 2.3], the hypothesis that the Hi ’s are nef implies that the height of V with respect to these Hermitian line bundles, denoted by hH1 ,...,Hb (V), is nonnegative. Hence, we associate to V the non-Archimedean absolute value on K given, for γ ∈ K, by −h (V)ordV (γ) |γ|V = e H1 ,...,Hb , where ordV denotes the discrete valuation associated to the local ring OB,V . We denote by µfin the counting measure of B (1) . We define the set of generic points of B(C) as [ B(C)gen = B(C) \ V(C). V∈B(1) By definition, a point p ∈ B(C) belongs to B(C)gen if and only if, for all γ ∈ K× , this point does not lie in the analytification of the support of div(γ). Hence |γ(p)| is a well-defined positive real number, and we associate to p the Archimedean absolute value given, for γ ∈ K× , by |γ|p = |γ(p)|. (1.1) On B(C), we consider the measure µ∞ = c1 (H1 ) ∧ · · · ∧ c1 (Hb ) associated to the family of semipositive Hermitian line bundles Hi , i = 1, . . . , b, as in [BPS14, Definition 1.4.2]. By [CT09, Corollaire 4.2], the measure of each hypersurface of B(C) with respect to µ∞ is zero. Since the complement of B(C)gen is a countable union of hypersurfaces, it has measure zero. We will denote also by µ∞ the induced measure on B(C)gen . Put then (M, µ) = (B (1) , µfin ) t (B(C)gen , µ∞ ). (1.2) The set M is in bijection with a set of absolute values. Moreover, all the nonArchimedean absolute values in this set are associated to a discrete valuation. Example 1.1. Let B = PbZ with projective coordinates (x0 : · · · : xb ) and Hi = can O(1) , i = 1, . . . , b, the universal line bundle on PbZ equipped with the canonical metric as in [BPS14, Example 1.4.4]. We have that K = K(B) ' Q(z1 , . . . , zb ), with zi = xi /x0 . Consider the compact subtorus of PbZ (C) given by S = {(1 : z1 : · · · : zb ) ∈ PbZ (C) | |zi | = 1 for all i} ' (S 1 )b and the measure µS of Pb (C) given by the current 1 dz1 dzb ∧ ··· ∧ ∧ δS . (2πi)b z1 zb Namely, µS is the Haar probability measure on S. A hypersurface V of PbZ corresponds to an irreducible homogeneous polynomial PV ∈ Z[x0 , . . . , xb ]. The associated absolute value is given, for γ ∈ K× , by log |γ|V = −ordV (γ) m(PV ), where m(PV ) is the logarithmic Mahler measure of PV given by Z m(PV ) = log |PV (1, z1 , . . . , zb )| dµS . If PV is a irreducible polynomial of degree zero, then PV = p ∈ Z, a prime number. In this case m(PV ) = log(p) and V is the fibre over the point corresponding to p. The absolute value associated to a point of PbZ (C)gen is given by the Archimedean absolute value of the evaluation at this point as in (1.1). 4 BURGOS GIL, PHILIPPON, AND SOMBRA In this example, the measure µfin on (PbZ )(1) is the counting measure and the measure µ∞ is the restriction to PbZ (C)gen of µS . A function on a measured space is integrable (also called summable) if its integral is a well-defined real number. Proposition 1.2. For each γ ∈ K× , the function M → R given by w 7→ log |γ|w is µ-integrable. Furthermore, the “product formula” Z log |γ|w dµ(w) = 0 (1.3) M holds. Proof. Given γ ∈ K× , the set of hypersurfaces V such that |γ|V 6= 1 is contained in the set of components of the support of div(γ), when γ is viewed as a rational function on B. Hence this set is finite, and so the function on B (1) given by V 7→ log |γ|V is µfin -integrable. Moreover, by [CT09, Th´eor`eme 4.1], the function on B(C)gen given by p 7→ log |γ(p)| is µ∞ -integrable. Summing up, log |γ|w is µintegrable, which proves the first statement. For the second one, let O be the trivial metrized line bundle on B. Then Z Z X log |γ|w dµ(w) = −ordV (γ) hH1 ,...,Hb (V) + log |γ(p)| dµ∞ (p) M B(C)gen V∈B(1) = − hO,H1 ,...,Hb (B) by the arithmetic B´ezout formula, see for instance [BGS94, (3.2.2)] for the smooth case or [CT09, Th´eor`eme 1.4] for an adelic version in the general case. From the multilinearity of the height, it follows that hO,H1 ,...,Hb (B) = 0, which concludes the proof. Definition 1.3. Given γ = (γ0 , . . . , γn ) ∈ Kn+1 \ {0}, the size of γ with respect to (K, M, µ) is defined as Z tK,M,µ (γ) = log max(|γ0 |w , . . . , |γn |w ) dµ(w). (1.4) M Example 1.4. Let K ' Q(z1 , . . . , zb ) with the measured set of places (M, µ) as described in Example 1.1. Let γ ∈ K× given in reduced representation as γ = α/β with coprime α, β ∈ Z[z1 , . . . , zb ]. Using the product formula (1.3), the size of γ can be given in this case by Z tK,M,µ (γ) = log max(|α|w , |β|w ) dµ(w) M Since α and β are coprime, the contribution of the integral over the places of (PbZ )(1) is zero. Hence, Z 1 dz1 dzb tK,M,µ (γ) = log max(|α(z)|, |β(z)|) ∧ ··· ∧ . (1.5) b (2πi) (S 1 )b z1 zb Using Jensen’s formula, this size can be alternatively written as the logarithmic Mahler measure of the polynomial Pγ = α(z1 , . . . , zb )t1 − β(z1 , . . . , zb ) ∈ Z[t1 , z1 , . . . , zb ], (1.6) where t1 denotes an additional variable. The difference between this size and the logarithm of the maximum of the absolute values of the coefficients of α and β can be bounded by the maximum of their degrees times a constant depending only on b. HEIGHT OF VARIETIES OVER FINITELY GENERATED FIELDS 5 2. Relative arithmetic varieties In this section we prove our main result (Theorem 2.4), showing that the height of a cycle over the finitely generated extension K can be written as an integral of the local heights of this cycle over the measured set of places (M, µ). Let π : X → B be a dominant morphism of arithmetic varieties of relative dimension n ≥ 0 and L a Hermitian line bundle on X . We denote by X the fibre of π over the generic point of B. This is a variety over K of dimension n, and the line bundle L induces a line bundle on X, denoted by L. There is a collection of metrics on the analytifications of L for each absolute value of M, that we now describe. For each V ∈ B (1) , the local ring OB,V is a discrete valuation ring with field of fractions K. The scheme X and the line bundle L induce a projective model over Spec(OB,V ), denoted (XV , LV ), of the pair (X, L). Following Zhang, the model an (XV , LV ) induces a metric on the analytification Lan V over XV , see [Zha95] or [BPS14, Definition 1.3.5] for details. The map π also induces a map of complex analytic spaces X (C) → B(C), that we also denote by π. A point p ∈ B(C)gen induces an Archimedean absolute value | · |p on K and the analytification of the variety X with respect to | · |p can be identified with the fibre π −1 (p) ⊂ X (C), with its structure of real analytic space when the point p is real. The analytification of the line bundle L on X with respect to | · |p can also be identified with the restriction of LC to π −1 (p). Then the metric on Lan p is defined as the restriction of the metric on LC to this fibre. We then denote L = (L, (k · kw )w∈M ) (2.1) the obtained M-metrized line bundle on X. Let Y be a d-dimensional cycle on X and Li , i = 0, . . . , d, M-metrized line bundles on X as in (2.1). We assume that each Li is constructed from a DSP Hermitian line bundle Li on X . Recall that a DSP (difference of semipositive) Hermitian line bundle on X is the quotient of two semipositive ones as in [BPS14, Definition 1.4.1]. Given a collection of nonzero rational sections si of Li , i = 0, . . . , d, intersecting properly on Y and w ∈ M, we denote by hL0,w ,...,Ld,w (Y ; s0 , . . . , sd ) the local height of Y with respect to the family of w-adic metrized line bundles Li,w := (Li , k · ki,w ), i = 0, . . . , d. It is defined inductively on the dimension of Y by the arithmetic B´ezout formula hL0,w ,...,Ld,w (Y ; s0 , . . . , sd ) = hL0,w ,...,Ld,w (Y · div sd ; s0 , . . . , sd−1 ) Z d−1 ^ − log ksd kd,w c1 (Li,w ) ∧ δY , an Xw (2.2) i=0 see [BPS14, Definition 1.4.11]. Recall that the local height We will show in Theorem 2.4 below that the function M → R given, for w ∈ M, by w 7−→ h(L0 ,k·k0,w ),...,(Ld ,k·kd,w ) (Y ; s0 , . . . , sd ) (2.3) is µ-integrable. Definition 2.1. With notation as above, the global height of Y with respect to the M-metrized line bundles Li , i = 0, . . . , d, is defined as the integral of the function in (2.3), that is Z hL0 ,...,Ld (Y ) = h(L0 ,k·k0,w ),...,(Ld ,k·kd,w ) (Y ; s0 , . . . , sd ) dµ(w). M 6 BURGOS GIL, PHILIPPON, AND SOMBRA Thanks to the product formula, this notion does not depend on the choice of the sections si . Example 2.2. Let B be an arithmetic variety as above and (K, M, µ) the associated finitely generated field and measured set of places. Let X = PnB ' B × PnZ and can L = $∗ OPnZ (1) , can where $ denotes the projection B × PnZ → PnZ . Hence X = PnK and L = OPnK (1) For a point p = (γ0 : · · · : γn ) ∈ X(K) = PnK (K), we have hL (p) = tK,M,µ (γ), . (2.4) where tK,M,µ (γ) denotes the size of the vector γ = (γ0 , . . . , γn ) as in (1.4). This is the “naive height” in [Mor00, § 3.2]. The following projection formula for heights of schemes over Spec(Z) generalizes [Mor00, Proposition 1.3(1)]. Proposition 2.3. Let π : W → V be a morphism between two finitely generated projective schemes over Spec(Z) of relative dimensions d + b − 1 and b − 1, respectively, with b, d ≥ 0. Let Li , i = 1, . . . , d, and Hj , j = 1, . . . , b, be DSP Hermitian line bundles on W and V, respectively. Then hπ∗ H 1 ,...,π ∗ Hb ,L1 ,...,Ld (W) = degL1 ,...,Ld (Wη ) hH1 ,...,Hb (V), where Wη denotes the fibre of W over the generic point η of V. In particular, if π is not dominant, then hπ∗ H ,...,π∗ H ,L1 ,...,Ld (W) = 0. 1 b Proof. By linearity, we can reduce to the case when the Li ’s are ample and the metrics are semipositive. By continuity, we can also reduce to the case when the metrics in Li and Hj are smooth for all i, j. We proceed by induction on d. The case d = 0 is given by [Mor00, Proposition 1.3(2)] in the case when π is dominant and by [BPS14, Theorem 1.5.11(2)] in the general case. Let d ≥ 1 and choose a nonzero rational section sd of Ld . Let k · kd denote the metric of Ld . By the arithmetic B´ezout formula, hπ∗ H 1 ,...,π ∗ Hb ,L1 ,...,Ld (W) = hπ∗ H ,...,π∗ H ,L1 ,...,Ld−1 (div(sd )) 1 b Z b d−1 ^ ^ − log ksd kd c1 (π ∗ Hi ) ∧ c1 (Lj ). (2.5) W(C) i=1 j=1 Vb Since dim(V(C)) = b − 1, we have that i=1 c1 (Hi ) = 0. Hence, the measure in the integral in the right-hand side of (2.5) is zero, and so this integral is zero too. Decompose the divisor of sd into its horizontal and vertical components over Spec(Z) as div(sd ) = div(sd )hor + div(sd )vert . P Write div(sd )vert = p∈Spec(Z) Zp as a finite sum of schemes over the primes. We have that degπ∗ H1 ,...,π∗ H ,L1 ,...,Ld−1 (Zp ) = 0 because dim(π(div(sd )vert )) ≤ b − 1. b It follows that hπ∗ H 1 ,...,π ∗ Hb ,L1 ,...,Ld−1 (div(sd )vert ) = X log(p) degπ∗ H1 ,...,π∗ H b ,L1 ,...,Ld−1 (Zp ) = 0. p∈Spec(Z) By the inductive hypothesis, hπ∗ H 1 ,...,π ∗ H ,L ,...,L 1 d−1 b (div(sd )hor ) = degL1 ,...,Ld−1 (div(sd )hor,η ) hH1 ,...,Hb (V). Since degL1 ,...,Ld−1 (div(sd )hor,η ) = degL1 ,...,Ld (Wη ), we obtain the result. HEIGHT OF VARIETIES OVER FINITELY GENERATED FIELDS 7 Theorem 2.4. Let B be an arithmetic variety of relative dimension b and Hi , i = 1, . . . , b, nef Hermitian line bundles on B. Let K = K(B) be the function field of B and (M, µ) the associated measured set of places as in (1.2). Let π : X → B be a dominant morphism of arithmetic varieties of relative dimension n and X the fibre of π over the generic point of B. Let Y be a prime cycle of X of dimension d and Y its closure in X . Let Lj , j = 0, . . . , d, be DSP Hermitian line bundles on X and Lj , j = 0, . . . , d, the associated M-metrized line bundles as in (2.1). Let s0 , . . . , sd be rational sections of L0 , . . . , Ld respectively, intersecting properly on Y . Then the function M → R given, for w ∈ M, by w 7−→ h(L0 ,k·k0,w ),...,(Ld ,k·kd,w ) (Y ; s0 , . . . , sd ) (2.6) is µ-integrable. Moreover, hL0 ,...,Ld (Y ) = hπ∗ H 1 ,...,π ∗ H ,L ,...,L 0 d b (Y). (2.7) In other words, the integral of the function (2.6) coincides with the height of Y as defined in (0.1). Proof. By linearity, we reduce to the case when the line bundles Lj are ample, their metrics are semipositive and the sections are global sections. Moreover, since multiplying one of the metrics on Lj changes both sides of the equality (2.7) by the same additive constant, we can assume that the sections sj of Lj , j = 0, . . . , d are small, in the sense that supp∈X(C) ksj (p)kj ≤ 1. We proceed by induction on the dimension of Y . If dim(Y ) = −1, then Y = ∅ and so the local heights of Y are zero. Hence, these local heights are µ-integrable and, by Proposition 2.3, the equality in (2.7) is reduced to 0 = 0. We now assume that dim(Y ) = d ≥ 0. In this case, the restriction π|Y : Y → B is dominant. Since the height does not change by normalization, by restricting objects to Y and pulling back to its normalization, we may assume in the computations that follow that Y = X . In particular, Y = X and d = n = dim(X). Let s0 , . . . , sn be global sections of L0 , . . . , Ln respectively, that meet properly on X, and denote by ρ : M → R the local height function in (2.6). We have to show that this function is µ-integrable and that Z ρ(w) dµ(w) = hπ∗ H ,...,π∗ H ,L0 ,...,Ln (X ). 1 M b For each w ∈ M, by the definition of local heights in (2.2), we can write ρ(w) = ρ1 (w) − ρ2 (w) with ρ1 (w) = h(L0 ,k·k0,w ),...,(Ln−1 ,k·kn−1,w ) (div(sn ); s0 , . . . , sn−1 ), Z n−1 ^ ρ2 (w) = log ksn kn,w c1 (Lj , k · kj,w ). an Xw j=0 We decompose the cycle div(sn ) as div(sn ) = div(sn )hor/B + div(sn )vert/B , where div(sn )hor/B contains all the components that are dominant over B and div(sn )vert/B contains the remaining ones. Clearly, div(sn )hor/B is the closure of div(sn ) · X, and div(sn )vert/B contains all the components of div(sn ) that do not meet X. By the inductive hypothesis, the function w 7→ ρ1 (w) is µ-integrable and Z (2.8) ρ1 (w) dµ(w) = hπ∗ H ,...,π∗ H ,L0 ,...,Ln (div(sn )hor/B ). M 1 b 8 BURGOS GIL, PHILIPPON, AND SOMBRA Let now w = V ∈ B (1) . The local ring OB,V is a discrete valuation ring. The scheme X and the line bundle Li induce models XV and Li,V over Spec(OB,V ) of X and Li . Each component of the special fibre of XV is the localization WV = W × Spec(K(V)) V (1) of a hypersurface W ∈ X with π(W) = V. Since the metric over w is an algebraic metric coming from a model, by [BPS14, (1.3.6) and Remark 1.4.14], X ρ2 (V) = − hH1 ,...,Hb (V)ordW (sn ) degL0 ,...,Ln−1 (WV ). (2.9) W∈X (1) π(W)=V Since the number of components of div(sn ) is finite, we deduce from (2.9) that there is only a finite number of V ∈ B (1) with ρ2 (V) 6= 0. Thus ρ2 is integrable on B (1) with respect to the counting measure µfin as in (1.2). By Proposition 2.3, hH1 ,...,Hb (V) degL0 ,...,Ln−1 (WV ) = hπ∗ H 1 ,...,π ∗ Hb ,L0 ,...,Ln−1 (W). The same result implies that, if dim(π(W)) ≤ b − 1, then hπ∗ H 1 ,...,π ∗ Hb ,L0 ,...,Ln−1 (W) = 0. Since div(sn )vert/B = X ordW (sn )W (1) W∈X dim(π(W))≤b = X X ordW (sn )W + V∈B(1) W∈X (1) π(W)=V X ordW (sn )W, W∈X (1) dim(π(W))≤b−1 it follows from (2.9) that Z X ρ2 (w) dµfin (w) = ρ2 (V) B(1) V∈B(1) =− X X ordW (sn ) hπ∗ H 1 ,...,π ∗ H ,L ,...,L 0 n b (W) V∈B(1) W∈X (1) π(W)=V = − hπ∗ H 1 ,...,π ∗ Hb ,L0 ,...,Ln (div(sn )vert/B ). (2.10) We next consider the places associated to the points p ∈ B(C)gen . In this case, by the definition of ρ2 , we have that Z n−1 ^ ρ2 (p) = log ksn kn c1 (Lj,C |π−1 (p) ), π −1 (p) j=0 where π denotes the projection X → B and k · kn the metric in Ln . We have to Vb show that ρ2 is µ∞ -integrable with µ∞ = i=1 c1 (Hi ) and that Z Z n−1 b ^ ^ ρ2 (p) dµ∞ (p) = log ksn kn c1 (Lj ) ∧ c1 (π ∗ Hi ). (2.11) B(C)gen X (C) j=0 i=1 We first assume that, for each j = 0, . . . , n, the metric on the line bundle Lj is smooth, but that the metric on Hi , i = 1, . . . , b, is not necessarily smooth. By definition, there is a sequence of smooth semipositive metrics (k · ki,k )k≥0 on Hi,C that converge to k·ki . Set Hi,k = (Hi , k·ki,k ) and let µ∞,k be the measure associated to the differential form c1 (H1,k ) ∧ · · · ∧ c1 (Hb,k ). HEIGHT OF VARIETIES OVER FINITELY GENERATED FIELDS 9 By [CT09, Th´eor`eme 4.1], the measures µ∞,k converge weakly to µ∞ . By the same result, even if log ksn kn is not bounded, the equality Z log ksn kn lim k→∞ X (C) n−1 ^ c1 (Lj ) ∧ j=0 b ^ c1 (π ∗ Hi,k ) i=1 Z log ksn kn = n−1 ^ X (C) c1 (Lj ) ∧ j=0 b ^ c1 (π ∗ Hi ) i=1 holds. Let U ⊂ B(C) be a connected Zariski open subset such that the restriction π |π−1 (U ) is a proper smooth map. By Ehresmann’s fibration theorem, this restriction is a locally trivial proper differentiable fibration. Thus, that there exists a compact differentiable manifold F and an analytic open cover {Uα }α of U such that π −1 (Uα ) is diffeomorphic to F × Uα for every α. Let {να }α be a partition of unity subordinated to the open cover {Uα }α . Fix an α. To avoid burdening the notation, we identify π −1 (Uα ) with F × Uα through the above diffeomorphism. Let λF be a measure of F given by a volume form. Since the metrics k · kj are smooth, there is a smooth function g : F × Uα → R such that, for each u ∈ Uα , n−1 ^ j=0 c1 (Lj ) = g(·, u)λF . {u}×F Vb By [Bil68, Theorem 3.2], the measures λF × i=1 c1 (π ∗ Hi,k ) converge weakly to Vb the measure λF × i=1 c1 (π ∗ Hi ). By the unicity of weak limits of measures, n−1 ^ c1 (Lj ) ∧ j=0 b ^ i=1 c1 (π ∗ Hi ) = gλF × F ×Uα b ^ c1 (π ∗ Hi ). (2.12) i=1 Vn−1 Vb ∗ Hi ), by (2.12) Vb the function (να ◦ π) log ksn kn g is integrable with respect to λF × i=1 c1 (π ∗ Hi ). By Fubini’s theorem [Fed69, Theorem 2.6.2], the function Z (να ◦ π) log ksn kn gλF = να ρ2 Since log ksn kn is integrable with respect to j=0 c1 (Lj )∧ i=1 c1 (π F is µ∞ -integrable and Z Z (να ◦ π) log ksn kn gλF × να ρ2 (p) dµ∞ (p) = F ×Uα Uα (να ◦ π) log ksn kn π −1 (Uα ) c1 (π ∗ Hi ) i=1 Z = b ^ n−1 ^ j=0 c1 (Lj ) ∧ b ^ c1 (π ∗ Hi ), i=1 where the last equality follows from (2.12). Since the above holds for every α, Lebesgue’s monotone convergence theorem [Fed69, Corollary 2.4.8] and the fact that µ∞( B(C) \ U )) = 0 which follows from [CT09, Corollaire 4.2], imply that ρ2 is µ∞ -integrable and that (2.11) holds. Observe that we can apply Lebesgue’s monotone convergence theorem because we are assuming that the section sn is small, and so the function log ksn kn is nonpositive. We now assume that the metrics on Lj and Hi are not necessarily smooth, and choose sequences of smooth semipositive metrics (k · kj,kj )kj ≥0 on Lj that converge 10 BURGOS GIL, PHILIPPON, AND SOMBRA uniformly to k · kj . For p ∈ B(C))gen , write Z n−1 ^ ρ2,k0 ,...,kn (p) = log ksn (p)kn,kn c1 (Lj,kj ,C |π−1 (p) ). π −1 (p) (2.13) j=0 From equation (2.13) when i = n, and from equation (2.13) and Stokes’ theorem when i 6= n, one can prove that, for each ε > 0, there is a constant Ki that does not depend on p nor on kj , j 6= i, such that, for all ki , ki0 ≥ Ki , |ρ2,k0 ,...,ki ,...,kn (p) − ρ2,k0 ,...,ki0 ,...,kn (p)| ≤ ε. (2.14) For k ≥ 0, denote by ρ2,k the function in (2.13) for the choice of indices k0 = · · · = kn = k. We deduce from (2.14) that the diagonal sequence (ρ2,k )k≥0 converges uniformly to ρ2 . Since the measure µ∞ has finite total mass and, by the previous case, the functions ρ2,k are µ∞ -integrable, we deduce that ρ2 is µ∞ -integrable and that Z Z lim ρ2,k (p) dµ∞ (p) = ρ2 (p) dµ∞ (p). k→∞ B(C) B(C) Therefore, using (2.11) for the functions ρ2,k and [CT09, Th´eor`eme 4.1], we deduce that (2.11) also holds in the case when all the metrics are semipositive. In consequence, ρ = ρ1 − ρ2 is µ-integrable and, using (2.8), (2.10), (2.11), the arithmetic B´ezout theorem in (2.2) and the inductive hypothesis, Z ρ(w) dµ(w) = hπ∗ H ,...,π∗ H ,L0 ,...,Ln (div(sn )hor/B ) 1 M b + hπ∗ H ,...,π∗ H ,L0 ,...,Ln (div(sn )vert/B ) 1 b Z b n−1 ^ ^ c1 (π ∗ Hi ) − log ksn kn c1 (Lj ) ∧ X (C) = hπ∗ H 1 i=1 j=0 ,...,π ∗ Hb ,L0 ,...,Ln (X ), which concludes the proof. Example 2.5. Let K ' Q(z1 , . . . , zb ) with the measured set of places (M, µ) as in Examples 1.1 and 1.4. Let X = PbZ × P1Z and can L = $∗ OP1Z (1) , can where $ denotes the projection PbZ × P1Z → P1Z , and let L = OP1K (1) denote the canonical M-metrized line bundle structure on the universal line bundle of P1K as in Example 2.2. Let (1 : γ) ∈ P1Z (K) with γ ∈ K× . The closure Y of this point in X is the hypersurface defined by the bihomogenization of the polynomial Pγ in (1.6). In this case, Theorem 2.4 together with (2.4) and (1.5) gives hπ∗ H 1 ,...,π ∗ Hb ,L (Y) = hL (1 : γ) = tK,M,µ (γ) = m(Pγ ), where m(Pγ ) denotes the logarithmic Mahler measure of Pγ . 3. Height of toric varieties over finitely generated fields Using our previous work on toric varieties in [BPS14], we can give a “combinatorial” formula for the mixed height of a toric variety with respect to a family of M-metrized line bundles. As a consequence of Theorem 2.4, this formula also expresses an arithmetic intersection number, in the sense of Gillet-Soul´e, of a non toric arithmetic variety. HEIGHT OF VARIETIES OVER FINITELY GENERATED FIELDS 11 Let B be an arithmetic variety of relative dimension b and Hi = (Hi , k · ki ), i = 1, . . . , b, a family of nef Hermitian line bundles on B, as at the beginning of § 1. Let K = K(B) and (M, µ) the associated set of places of K as in (1.2). Let T ' Gnm be a split torus of dimension n over K. Let N = Hom(Gm , T) be the lattice of cocharacters of T, M = Hom(T, Gm ) = N ∨ the lattice of characters, and set NR = N ⊗ R and MR = M ⊗ R. Let X be a proper toric variety over K with torus T, described by a complete fan Σ on NR . A toric divisor on X is a Cartier divisor invariant under the action of T. Such a divisor D defines a “virtual support function”, that is, a function ΨD : NR → R whose restriction to each cone of the fan Σ is an element of M . The toric divisor D is nef if and only if ΨD is concave. One can also associate to D the polytope defined as ∆D = {x ∈ MR | x ≥ ΨD }. Now let π: X → B be a dominant morphism of arithmetic varieties of relative dimension n ≥ 0 and L a Hermitian line bundle on X . We assume that (X, L), the fibre of (X , L) over the generic point of B, is a toric variety over K with a line bundle associated to a toric divisor D on X. We consider the associated M-metrized line bundle L on X as in (2.1). For each place w ∈ M, we associate to the torus T an analytic space Tan w and we denote by Sw its compact subtorus. In the Archimedean case, it is isomorphic to (S 1 )n . In the non-Archimedean case, it is a compact analytic group, see [BPS14, § 4.2] for a description. Then, the M-metrized line bundle L = O(D) on X is toric if its w-adic metric k · kw is invariant with respect to the action of Sw for all w. Assume that L is toric and let s be the toric section of L with div(s) = D. an For each w ∈ M, denote by X0,w the analytification of the open principal subset X0 ⊂ X corresponding to the cone {0}, which is isomorphic to the torus T. Then an the function X0,w → R given by p 7→ log ks(p)kw is invariant under the action of Sw and induces a function ψL,s,w : NR → R as in [BPS14, Definition 4.3.5]. For shorthand, when L and s are fixed, we will denote ψL,s,w by ψw . We now further assume that the line bundle L is generated by global sections and that the Hermitian metric on L is semipositive. Hence, the line bundle L is also generated by global sections and, for each w ∈ M, the metric induced in Lan w by L is semipositive. In this case, by [BPS14, Theorem 4.8.1], for each w ∈ M, the function ψw is concave. We associate to it a concave function on ∆D , denoted by ϑL,s,w or ϑw for short, and called the w-adic roof function of the pair (L, s) as in [BPS14, Definition 5.1.4]. This function is defined as the Legendre-Fenchel dual of ψw , and so it is defined, for x ∈ ∆D , as ϑw (x) = inf (hx, ui − ψw (u)). u∈NR We denote by volM the measure on ∆D given by the restriction of the Haar measure on MR normalized so that the lattice M has covolume 1. Corollary 3.1. With notation as above, the function Z M −→ R, w 7−→ ϑL,s,w (x) dvolM (x) (3.1) ∆D is µ-integrable. Moreover, Z hπ∗ H1 ,...,π∗ Hb ,L,...,L (X ) = hL (X) = (n+1)! M Z ∆D ϑL,s,w (x) dvolM (x) dµ(w). (3.2) 12 BURGOS GIL, PHILIPPON, AND SOMBRA Proof. By [BPS14, Theorem 5.1.6], the quantity Z (n + 1)! ϑL,s,w (x) dvolM (x) ∆D is equal to the difference of local heights h(L0 ,k·k0,w ),...,(Ln ,k·kn,w ) (Y ; s0 , . . . , sn )−h(L0 ,k·k0,w,can ),...,(Ln ,k·kn,w,can ) (Y ; s0 , . . . , sn ), where k · ki,w,can is the canonical w-adic metric on Li as in [BPS14, PropositionDefinition 4.3.15]. By Theorem 2.4, both local heights are µ-integrable. Hence, so is the function in (3.1), which proves the first statement. For the second statement, the first equality follows from Theorem 2.4. By the discussion above, Z Z (n + 1)! ϑL,s,w (x) dvolM (x) dµ(w) = hL (X) − hLcan (X). M ∆D Using the argument in the proof of [BPS14, Proposition 5.2.4], it can be shown that hLcan (X) = 0, which proves the second equality in (3.2). Theorem 3.2. Let notation be as above. (1) For each x ∈ ∆D , the function M −→ R, w 7−→ ϑw (x) is µ-integrable. (2) The function Z ∆D −→ R, x 7−→ ϑw (x) dµ(w) M is concave and continuous on ∆D . (3) The function M × ∆D −→ R, (w, x) 7−→ ϑw (x) is (µ × volM )-integrable. Proof. Let σ ∈ Σn . The closure V (σ) of the orbit of X corresponding to σ is a point. By [BPS14, Proposition 4.8.9], for each w ∈ M, ϑι∗ L,ι∗ sσ ,w (0) = ϑL,sσ ,w (mσ ) = ϑw (mσ ), where ι denotes the inclusion V (σ) ,→ X. By Corollary 3.1, the function w 7→ ϑw (mσ ) in µ-integrable, and its integral coincides with the height of V (σ) with respect to L. Since ϑw is a concave function, for all x ∈ ∆D , min ϑw (mσ ) = min ϑw (y) ≤ ϑw (x). σ∈Σn y∈∆D (3.3) On the other hand, using again the concavity of ϑw , ϑw (x) − min ϑw (y) ≤ max ϑw (y) − min ϑw (y) y∈∆D y∈∆D y∈∆D Z n+1 ≤ ϑw (z) − min ϑw (y) dvolM (z) y∈∆D volM (∆D ) ∆D (3.4) It follows from (3.3) and (3.4) that, for all x ∈ ∆D , Z n+1 min ϑw (mσ ) ≤ ϑw (x) ≤ ϑw (z) dvolM (z) − n minn ϑw (mσ ). σ∈Σn σ∈Σ volM (∆D ) ∆D By Corollary 3.1 and the fact that ∆D has finite measure, we have that both the upper and the lower bound are integrable with respect to the measure µ × volM . The statements (1) and (3) follow directly from these bounds, while the statement HEIGHT OF VARIETIES OVER FINITELY GENERATED FIELDS 13 (2) follows from the same bounds and Lebesgue’s bounded convergence theorem [Fed69, Theorem 2.4.9]. Definition 3.3. With notations as above, the (global) roof function is the continuous concave function ϑL,s : ∆D → R given by Z ϑL,s (x) = ϑw (x) dµ(w). M Corollary 3.4. With the previous notations, Z hπ∗ H1 ,...,π∗ Hb ,L,...,L (X ) = hL (X) = (n + 1)! ∆D ϑL,s (x) dvolM (x) holds. Proof. This follows from Corollary 3.1 and Theorem 3.2(3) together with Fubini’s theorem [Fed69, Theorem 2.6.2]. Remark 3.5. More generally, when we have a family L0 , . . . , Ln of semipositive Hermitian line bundles on X such that the induced Hermitian line bundles L0 , . . . , Ln on X are toric, we can express hπ∗ H1 ,...,π∗ Hb ,L0 ,...,Ln (X ) = hL0 ,...,Ln (X) in terms of mixed integrals, similarly as in [BPS14, Theorem 5.2.5], 4. Canonical height of translated of subtori over finitely generated fields In this section, we particularize the formulae in § 3 to the case when X is the normalization of a translated of a subtori in the projective space. As before, let B be an arithmetic variety of relative dimension b and Hi = (Hi , k · ki ), i = 1, . . . , b, a family of nef Hermitian line bundles on B. Let also K = K(B) and (M, µ) the associated set of places of K as in (1.2). Let r ≥ 1 and consider the projective space PrB over B and the universal line bundle OPrB (1) on it. Since PrB = PrZ × B and OPrB (1) is the pull-back of OPrZ (1) Spec(Z) under the first projection, we can pull-back the canonical metric on OPrZ (1) to obtain a metric on OPrB (1), also called canonical. We denote by O(1) = OPrB (1) the obtained Hermitian line bundle. Choose mj ∈ Zn and fj ∈ K× , j = 0, . . . , r. For simplicity, we assume that m0 = 0 ∈ Zn and that the collection of vectors mj generates Zn as Abelian group. Consider the map Gnm,K −→ PrK , t 7−→ (f0 tm0 : · · · : fr tmr ), m m where fj tmj denotes the monomial fj t1 j,1 . . . tn j,n . We then denote by Y the closure in PrK of the image of this map. The projective space PrK is the fibre of PrB over the generic point of B. We denote by Y the closure of Y in PrB and by π : Y → B the dominant map obtained by restricting the projection PrB → B. In this setting, we want to give a formula for the arithmetic intersection number hπ∗ H,...,π∗ H,O(1),...,O(1) (Y). (4.1) The subvariety Y is not a toric variety over K because it is not necessarily normal. Indeed, it is a “translated toric subvariety” of PrK in the sense of [BPS14, Definition 3.2.6]. Let X be the normalization of Y, and X the corresponding variety over K. Let L be the pull-back of O(1) to X and L the associated M-metrized line 14 BURGOS GIL, PHILIPPON, AND SOMBRA bundle over X as in (2.1). Therefore, X is a toric variety over K with torus Gnm,K and the M-metrized line bundle L is toric and semipositive. In this case we can give an explicit description of the corresponding w-adic roof functions. Proposition 4.1. With notation as above, let s be the toric section of L determined by the section x0 of O(1). The polytope associated to the divisor D = div(s) on X is given by ∆ = conv(m0 , . . . , mr ) and, for w ∈ M, the w-adic roof function ϑw : ∆ → R is the function parameterizing e w ⊂ Rn × R given by the upper envelope of the extended polytope ∆ ( conv (mj , − hH1 ,...,Hb (V)ordV (fj ))j=0,...,r , if w = V ∈ B (1) , e ∆w = conv((mj , log |fj (p)|)j=0,...,r ), if w = p ∈ B(C)gen . Proof. This follows from [BPS14, Example 5.1.16]. Combining this result with Corollary 3.1, we obtain a formula for the arithmetic intersection number in (4.1), that we will particularize to concrete examples. can Corollary 4.2. With notation as above, set B = PbZ and Hi = H = OPrB (1) i = 1, . . . , b. Then hπ∗ H,...,π∗ H,O(1),...,O(1) (Y) is equal to ! ! Z Z XZ (n + 1)! ϑp (x) dvol(x) dµS + ϑV (x) dvol(x) , ∆ V∈C , ∆ (1) where C ⊂ B is the set of irreducible components of the divisors div(fj ), j = 0, . . . , r, vol denotes the Lebesgue measure on Rn , and µS is the Haar measure of the compact torus S as in Example 1.1. Proof. We have that hπ∗ H,...,π∗ H,O(1),...,O(1) (Y) = hπ∗ H,...,π∗ H,L,...,L (X ) by the invariance of the height under normalization. The formula then follows from Corollary 3.1, Proposition 4.1, the description of the measured set of places (M, µ) of the field K in Example 1.1 together with the fact that, for V ∈ B (1) \ C, the local roof function ϑV vanishes identically. Example 4.3. Consider the case n = 0. Thus m` = 0 for all ` and choose a collection f0 , . . . , fr ∈ Z[z1 , . . . , zb ] of coprime polynomials with integer coefficients. Hence ∆ = {0} ⊂ R0 and, for w ∈ M, the local roof function is given by ( max` log |f` (p)| if w = p ∈ PbZ (C)gen , ϑw (0) = − h(V) min` ordV (f` ) if w = V ∈ (PbZ )(1) . If V is not the hyperplane at infinity of Pb , then by the coprimality of the f` , we have min` ordV (f` ) = 0, while, if V is the hyperplane at infinity, h(V) = 0. Thus the finite contribution vanishes and, from Theorem 2.4 and Proposition 4.1 we deduce Z hπ∗ H,...,π∗ H,L (Y) = max(log |f` (p)|) dµS . ` In particular, if r = 1, this arithmetic intersection number agrees with the size of the element γ = f1 /f0 given in Example 1.4. For instance, consider the case when b = 1, f0 = α and f1 = βz1 + γ with α, β, γ coprime integers. Using (1.5) and (1.6), the corresponding intersection number is given by the logarithmic Mahler measure of the affine polynomial αt1 − βz1 − γ. By [Mai00, Proposition 7.3.1], this logarithmic Mahler measure can be computed in terms of the Bloch-Wigner dilogarithm. HEIGHT OF VARIETIES OVER FINITELY GENERATED FIELDS 15 Example 4.4. Let n = 1 and consider the case when mi = i, i = 0, . . . , r, and f0 , . . . , fr ∈ Z[z1 , . . . , zb ] is a family of coprime polynomial with integer coefficients with f0 = fr = 1. We have that ∆ = [0, r]. Let w ∈ M and ϑw : [0, r] → R the corresponding local roof function. By Proposition 4.1, if w = V ∈ (PbZ )(1) , then this function is zero. Also, if w = p ∈ PbZ (C)gen , this function is the minimal concave function on [0, r] whose values at the integers are given, for i = 0, . . . , r, by ϑw (i) = max 0≤j≤i≤`≤r j6=` `−i i−j log |fj (p)|w + log |f` (p)|w . `−j `−j In particular, ϑw (0) = log |f0 (p)|w = 0 and ϑw (r) = log |fr (p)|w = 0. It follows that Z r ϑw (x) dx = 0 δ−1 X i=1 max j≤i≤`,j6=` i−j `−i log |fj (p)| + log |f` (p)| `−j `−j (4.2) From Corollary 4.2 and (4.2), we deduce that hπ∗ H,...,π∗ H,L (Y) = 2 Z X r−1 i=1 max 0≤j≤i≤`≤r j6=` `−i i−j log |fj (p)| + log |fj (p)| `−j `−j dµS . Hence, this arithmetic intersection number can be expressed in terms of integrals over the compact torus, of maxima of logarithms of absolute values of polynomials. References [Bil68] [BGS94] P. Billingsley, Convergence of probability measures, John Wiley & Sons, 1994. J.-B. Bost, H. Gillet, and C. Soul´ e, Heights of projective varieties and positive Green forms, J. Amer. Math. Soc. 7 (1994), 903–1027. [BMPS12] J. I. Burgos Gil, A. Moriwaki, P. Philippon, and M. Sombra, Arithmetic positivity on toric varieties, e-print arXiv:1210.7692v1, 2012, to appear in J. Alg. Geom. [BPS14] J. I. Burgos Gil, P. Philippon, and M. Sombra, Arithmetic geometry of toric varieties. Metrics, measures and heights, Ast´ erisque, vol. 360, Soc. Math. France, 2014. [CT09] A. Chambert-Loir and A. Thuillier, Mesures de Mahler et ´ equidistribution logarithmique, Ann. Inst. Fourier 59 (2009), 977–1014. [Dem97] J.-P. Demailly, Complex analytic and differential geometry, downloadable from http: //www-fourier.ujf-grenoble.fr/~demailly/manuscripts/agbook.pdf, 1997. [Fed69] H. Federer, Geometric measure theory, Grund. Math. Wiss., Band 153, SpringerVerlag, 1969. [Gub03] W. Gubler, Local and canonical heights of subvarieties, Ann. Sc. Norm. Super. Pisa Cl. Sci. (5) 2 (2003), 711–760. [Mai00] V. Maillot, G´ eom´ etrie d’Arakelov des vari´ et´ es toriques et fibr´ es en droites int´ egrables, M´ em. Soc. Math. France, vol. 80, Soc. Math. France, 2000. [Mor00] A. Moriwaki, Arithmetic height functions over finitely generated fields, Invent. Math. 140 (2000), 101–142. [Mor01a] , The canonical arithmetic height of subvarieties of an abelian variety over a finitely generated field, J. Reine Angew. Math. 530 (2001), 33–54. [Mor01b] , A generalization of conjectures of Bogomolov and Lang over finitely generated fields, Duke Math. J. 107 (2001), 85–102. [YZ13] X. Yuan and S.-W. Zhang, The arithmetic Hodge index theorem for adelic line bundles II: finitely generated fields, e-print arXiv: 1304.3539, 2013. [Zha95] S.-W. Zhang, Small points and adelic metrics, J. Algebraic Geom. 4 (1995), 281–300. 16 BURGOS GIL, PHILIPPON, AND SOMBRA ´ ticas (CSIC-UAM-UCM-UCM3). Calle Nicola ´ s CabreInstituto de Ciencias Matema ra 15, Campus UAB, Cantoblanco, 28049 Madrid, Spain E-mail address: [email protected] URL: http://www.icmat.es/miembros/burgos ´ ´matiques de Jussieu – U.M.R. 7586 du CNRS, Equipe ´orie des Institut de Mathe de The Nombres. BP 247, 4 place Jussieu, 75005 Paris, France E-mail address: [email protected] URL: http://www.math.jussieu.fr/~pph ` ICREA & Universitat de Barcelona, Departament d’Algebra i Geometria. Gran Via 585, 08007 Barcelona, Spain E-mail address: [email protected] URL: http://atlas.mat.ub.es/personals/sombra

© Copyright 2019 ExploreDoc