OpenMath
Internal proofs; conditional height results

17 November 2026 · 19 PDF pages

Download standalone PDF

Results and proofs

Arithmetic Lefschetz and Integral Heights

Proofs and reductions on products of curves

17 November 2026

Abstract

For a product of pointed curves, the tensor Chow projectors turn arithmetic hard Lefschetz into an exact vanishing condition on the all-h1h^1 groups of its subproducts. A primitive sum-of-squares identity then reduces positivity and integral finite-height questions to the central degree. In that degree, a self-transpose Chow projector separates a Néron–Tate summand from a canonical Griffiths pairing. We prove that central integral Northcott is equivalent to injectivity of the algebraic Abel–Jacobi map and integral Northcott on this residual quotient. We also construct residual arithmetic regulators and prove criteria for the common integral lattice and for the fields of definition needed in a Northcott argument. The height statements concern a system satisfying the four explicit axioms below.

1. Introduction

A height pairing on homologically trivial cycles is expected to satisfy arithmetic analogues of hard Lefschetz and Hodge–Riemann. Finiteness of integral height balls asks for additional arithmetic control. On a product of curves, these questions can be separated using its exact tensor Chow decomposition: the position of a Lefschetz string is geometric, its signed height is central, and its integral coefficients have one fixed denominator bound.

Let FF be a number field and let X=∏i=1nCi,oi∈Ci(F),Hi=pi∗[oi],ℓ=∑i=1naiHi(ai∈Z>0).X=\prod_{i=1}^n C_i,\qquad o_i\in C_i(F),\qquad H_i=p_i^*[o_i],\qquad \ell=\sum_{i=1}^n a_iH_i \quad(a_i\in\mathbb Z_{>0}). Each CiC_i is smooth, projective and geometrically connected. For a finite extension E/FE/F, write Vr(XE)=CH⁡r(XE)hom,QV^r(X_E)=\operatorname{CH}^r(X_E)_{\mathrm{hom},\mathbb Q}. For S⊆{1,…,n}S\subseteq\{1,\ldots,n\}, put YS=∏i∈SCi,ΠS=⨂i∈S(ΔCi−oi×Ci−Ci×oi),Y_S=\prod_{i\in S}C_i,\qquad \Pi_S=\bigotimes_{i\in S} (\Delta_{C_i}-o_i\times C_i-C_i\times o_i), USc(E)=ΠS∗Vc(YS,E).U_S^c(E)=\Pi_{S*}V^c(Y_{S,E}). The empty subproduct has zero homologically trivial Chow groups. A nonempty core UScU_S^c is central when ∣S∣=2c−1|S|=2c-1.

1.1. Heights and integral classes

Fix an algebraic closure F‾\overline F of FF and an embedding into C\mathbb C. Homological triviality is defined by geometric rational Betti cohomology. All Chow groups below have rational coefficients unless otherwise indicated. We regard Vr(XE)V^r(X_E) as a subspace of Vr(XF‾)V^r(X_{\overline F}): restriction is injective because a rational equivalence descends to a finite extension and trace composed with restriction multiplies by its degree.

Definition 1.1. A curve-normalized height system consists of pairings on all smooth projective varieties used below, including the subproducts and the source curves of correspondences. For such a variety XX of dimension nn, these are hX:Vr(XE)×Vn+1−r(XE)⟶Rh_X:V^r(X_E)\times V^{n+1-r}(X_E)\longrightarrow\mathbb R over every finite extension E/FE/F, satisfying:

  1. bilinearity over Q\mathbb Q and symmetry under exchange of the complementary degrees;

  2. invariance under extension in the absolute normalization, and covariance under isomorphisms, including Galois conjugation;

  3. correspondence adjunction hX(Γ∗z,w)=hY(z,tΓ∗w);h_X(\Gamma_*z,w)=h_Y(z,{}^t\Gamma_*w);

  4. on a pointed curve CC, the form −hC-h_C on Pic⁡0(C)Q\operatorname{Pic}^0(C)_\mathbb Q is the positive Néron–Tate form for the canonical principal polarization, in the corresponding Poincaré normalization.

These hypotheses will be understood in every statement involving hh. A real positive form means positive definiteness after tensoring the rational group over a fixed number field with R\mathbb R. For codimension rr on XX, the symbol AJ⁡Q\operatorname{AJ}_\mathbb Q denotes the complex Abel–Jacobi map with target Jr(XC)(C)⊗ZQJ^r(X_\mathbb C)(\mathbb C)\otimes_\mathbb Z\mathbb Q.

Let ΛXr(E)\Lambda_X^r(E) be the image in Vr(XF‾)V^r(X_{\overline F}) of the integral-coefficient cycles over EE with zero geometric rational cycle class. Thus torsion and rational equivalence are removed before counting. For a fixed rational projector PP, we use its projected integral image P∗ΛXr(E)P_*\Lambda_X^r(E); for an integral projector this equals ΛXr(E)∩P∗Vr(XE)\Lambda_X^r(E)\cap P_*V^r(X_E).

Definition 1.2. For compatible integral images Λ(E)\Lambda(E) in one geometric rational group and a quadratic form qq, bounded-degree Northcott means that, for every B∈RB\in\mathbb R and integer D≥1D\ge1, {ξ∈⋃[E:F]≤DΛ(E):q(ξ)≤B}\{\xi\in\bigcup_{[E:F]\le D}\Lambda(E):q(\xi)\le B\} is finite. Fixed-field Northcott means the same assertion with EE fixed. Bounded denominators mean the union of d−1Λ(E)d^{-1}\Lambda(E) for 1≤d≤d01\le d\le d_0, with d0d_0 fixed.

Northcott for integral images implies the bounded-denominator version by scaling. A fixed initial extension F0/FF_0/F is harmless for bounded-degree assertions: after replacing EE by EF0EF_0, [EF0:F0]≤[E:F][EF_0:F_0]\le[E:F], the integral image remains integral, and the absolute height is unchanged.

1.2. The principal reduction

For an odd subproduct ∣S∣=2c−1|S|=2c-1, set AS,E=USc(E)∩CH⁡c(YS,E)alg,Q,GS,E=USc(E)/AS,E.A_{S,E}=U_S^c(E)\cap\operatorname{CH}^c(Y_{S,E})_{\mathrm{alg},\mathbb Q}, \qquad G_{S,E}=U_S^c(E)/A_{S,E}. Here algebraic triviality is geometric. The construction in Proposition 3.2 supplies actual self-transpose projectors eS≤ΠSe_S\le\Pi_S and RS=ΠS−eSR_S=\Pi_S-e_S. After one finite extension F0/FF_0/F, these projectors are defined for all odd subproducts. The eSe_S-summand is represented by a fixed abelian variety, and RSAS,E=ker⁡(AJ⁡Q∣AS,E),RSUSc(E)/RSAS,E≃GS,E.R_SA_{S,E}=\ker(\operatorname{AJ}_\mathbb Q|_{A_{S,E}}),\qquad R_SU_S^c(E)/R_SA_{S,E}\simeq G_{S,E}. Orthogonalization defines a canonical residual pairing hG,Sh_{G,S} on GS,EG_{S,E}; write qG,S(ξ)=(−1)chG,S(ξ,ξ)q_{G,S}(\xi)=(-1)^ch_{G,S}(\xi,\xi). Let ΛG,S(E)\Lambda_{G,S}(E) be the image of ΠS∗ΛYSc(E)\Pi_{S*}\Lambda_{Y_S}^c(E) in this quotient.

For 1≤r≤⌊(n+1)/2⌋1\le r\le\lfloor(n+1)/2\rfloor, define Pℓr(XE)=ker⁡ ⁣(ℓn+2−2r:Vr(XE)⟶Vn+2−r(XE)),qℓ,r(z)=(−1)rhX(z,ℓn+1−2rz).\begin{aligned} P_\ell^r(X_E)&=\ker\!\left( \ell^{n+2-2r}:V^r(X_E)\longrightarrow V^{n+2-r}(X_E)\right),\\ q_{\ell,r}(z)&=(-1)^rh_X(z,\ell^{n+1-2r}z). \end{aligned} Its integral image is ΛXr(E)∩Pℓr(XE)\Lambda_X^r(E)\cap P_\ell^r(X_E).

Theorem 1.3 (The product and central reductions). Let X,ℓ,F0X,\ell,F_0 be as above, and let hh be a curve-normalized height system. The following conditions are equivalent.

  1. For every finite E/F0E/F_0 and every 1≤r≤⌊(n+1)/2⌋1\le r\le\lfloor(n+1)/2\rfloor, the map ℓn+1−2r:Vr(XE)→∼Vn+1−r(XE)\ell^{n+1-2r}:V^r(X_E)\xrightarrow{\sim}V^{n+1-r}(X_E) is an isomorphism; and qℓ,rq_{\ell,r} has bounded-degree Northcott on the primitive integral images, in every such degree.

  2. The following three assertions hold for the subproducts of XX:

    1. USc(E)=0U_S^c(E)=0 for every finite E/F0E/F_0 whenever ∣S∣≠2c−1|S|\ne2c-1.

    2. For every ∣S∣=2c−1|S|=2c-1 and finite E/F0E/F_0, the rational Abel–Jacobi map is injective on AS,EA_{S,E}.

    3. For every ∣S∣=2c−1|S|=2c-1, qG,Sq_{G,S} has bounded-degree Northcott on ΛG,S\Lambda_{G,S}.

Either condition implies real positive definiteness of every primitive form and nondegeneracy of every complementary height pairing over each fixed number field.

The proof is completed in Section 7. The three assertions in (ii) are formulated separately as conjectures in the companion programme [1]. The theorem identifies their combined consequence on the whole product.

1.3. The proof and the remaining arithmetic inputs

The proof has three stages. First, the exact Chow projectors identify each summand with USc(E)⊗QQ[xi:i∉S]/(xi2).U_S^c(E)\otimes_\mathbb Q \mathbb Q[x_i:i\notin S]/(x_i^2). Multiplication by ℓ\ell acts on the square-zero factor. A noncentral core places the centre of its string in the wrong degree, forcing a failure of injectivity or surjectivity. This proves the Lefschetz equivalence in Subsection 2.3.

Second, adjunction pairs only complementary point and fundamental-class indices. In a central block, with m=n−∣S∣m=n-|S|, t=r−ct=r-c, I={1,…,n}∖SI=\{1,\ldots,n\}\setminus S, and zS=∑∣T∣=tuTxTz_S=\sum_{|T|=t}u_Tx_T primitive, it gives qℓ,r(zS)=(m−2t)!∏i∈Iai∑∣T∣=t(−1)chYS(uT,uT). q_{\ell,r}(z_S) =(m-2t)!\prod_{i\in I}a_i \sum_{|T|=t}(-1)^c h_{Y_S}(u_T,u_T).(1.1) The exact sign and factorial are proved in Subsection 2.4. This identity on a central block uses no off-central vanishing hypothesis.

Third, the central projector separates the classical Néron–Tate summand and the residual Griffiths group. The algebraic Abel–Jacobi kernel is precisely the algebraic height radical. One fixed correspondence detects the integral classes on the visible summand, and one denominator clears the residual projector. These facts prove the central Northcott criterion in Subsection 3.3. Integral Northcott itself forces real positive definiteness by the rounding argument of Lemma 5.1.

The later sections address the arithmetic inputs to residual Northcott. The actual residual projector isolates one Hochschild–Serre row and defines a regulator GE→H1(E,Vp)G_E\to H^1(E,V_p). Weight −1-1 makes the entire target finite-dimensional; scalar-extended injectivity would then give finite rational rank. The lattice criteria identify the further common-denominator condition. Finally, trace and Hermite–Minkowski show that fixed-field Northcott upgrades to bounded-degree Northcott exactly when the fields of the bounded-height classes ramify in one finite set.

The arithmetic Lefschetz and primitive positivity assertions fit the standard conjectural framework recorded in [10, Conjecture 2.3.1]. The regulator construction uses the continuous-cohomology framework of [8; 4]. The results below give the reductions, constructions and arithmetic criteria in this setting.

2. Lefschetz strings and primitive heights

The tensor projectors of the curves give exact Chow summands. In each summand, the product polarization acts on a square-zero polynomial algebra. Its Lefschetz strings detect the central degree.

Theorem 2.1 (Arithmetic Lefschetz reduction). Fix positive integers aia_i and ℓ=∑iaiHi\ell=\sum_i a_iH_i. For every finite E/FE/F, the following conditions are equivalent:

  1. USc(E)=0U_S^c(E)=0 for every S,cS,c with ∣S∣≠2c−1|S|\ne2c-1.

  2. For 1≤r≤⌊(n+1)/2⌋1\le r\le\lfloor(n+1)/2\rfloor, ℓn+1−2r:Vr(XE)→∼Vn+1−r(XE).\ell^{n+1-2r}:V^r(X_E)\xrightarrow{\sim}V^{n+1-r}(X_E).

In particular, condition (ii) for one choice of positive integers aia_i is equivalent to condition (ii) for every such choice, with the points oio_i fixed.

2.1. Exact Chow blocks

On a pointed curve put π0=o×C,π2=C×o,π1=ΔC−π0−π2.\pi_0=o\times C,\qquad \pi_2=C\times o,\qquad \pi_1=\Delta_C-\pi_0-\pi_2. They are orthogonal projectors, with tπ0=π2{}^t\pi_0=\pi_2 and tπ1=π1{}^t\pi_1=\pi_1. For a partition [n]=S⊔T⊔U[n]=S\sqcup T\sqcup U, define PS,T,U=⨂i∈Sπ1,i⊗⨂i∈Tπ2,i⊗⨂i∈Uπ0,i.P_{S,T,U}=\bigotimes_{i\in S}\pi_{1,i} \otimes\bigotimes_{i\in T}\pi_{2,i} \otimes\bigotimes_{i\in U}\pi_{0,i}. Let jTj_T insert the point [oi][o_i] on factors in TT and the fundamental class [Ci][C_i] on factors in UU. Its inverse on this projector image is the fixed correspondence bT(z)=ΠS∗pS∗(z⋅∏i∈UHi).b_T(z)=\Pi_{S*}p_{S*} \left(z\cdot\prod_{i\in U}H_i\right). Indeed bTjT=ΠSb_Tj_T=\Pi_S and jTbT=PS,T,Uj_Tb_T=P_{S,T,U}. Therefore Vr(XE)=⨁S,T,UjTUSr−∣T∣(E). V^r(X_E)=\bigoplus_{S,T,U}j_TU_S^{r-|T|}(E).(2.1) These identities are also valid on integral images. All intersections and pushforwards in bTb_T have integral coefficients.

Put I=[n]∖SI=[n]\setminus S, m=∣I∣m=|I|, and BI=Q[xi:i∈I]/(xi2),deg⁡xi=1.B_I=\mathbb Q[x_i:i\in I]/(x_i^2),\qquad \deg x_i=1. For T⊂IT\subset I, write aT=∏i∈Taia_T=\prod_{i\in T}a_i. The identification u⊗xT⟼aTjTu u\otimes x_T\longmapsto a_Tj_Tu(2.2) groups (2.1) as V∗(XE)=⨁S,cUSc(E)⊗BI[c],V^*(X_E)=\bigoplus_{S,c} U_S^c(E)\otimes B_I[c], where a vector in USc⊗BItU_S^c\otimes B_I^t has codimension c+tc+t. Multiplication by ℓ=∑iaiHi\ell=\sum_i a_iH_i becomes multiplication by EI=∑i∈IxiE_I=\sum_{i\in I}x_i on the second factor. On the other factors it vanishes because Hiπ1,i=0H_i\pi_{1,i}=0 as a Chow correspondence. This last identity follows by composing μ[oi]=[oi]×[oi]\mu_{[o_i]}=[o_i]\times[o_i] with the three curve projectors.

If zz is integral, its coefficient uTu_T in (2.2) satisfies aTuT∈ΛUSc(E)a_Tu_T\in\Lambda_{U_S^c}(E). Thus the single integer Na=∏i=1nai N_a=\prod_{i=1}^n a_i(2.3) clears every coefficient denominator, in every degree and over every finite extension.

2.2. A square-zero Lefschetz calculation

On BIB_I, use the monomial basis xTx_T and define ExT=∑i∉TxT∪{i},FxT=∑i∈TxT∖{i},HxT=(m−2∣T∣)xT.E x_T=\sum_{i\notin T}x_{T\cup\{i\}},\qquad F x_T=\sum_{i\in T}x_{T\setminus\{i\}},\qquad Hx_T=(m-2|T|)x_T. Then [F,E]=H[F,E]=H, [H,F]=2F[H,F]=2F, and [H,E]=−2E[H,E]=-2E. The operators E,FE,F are adjoint for the positive inner product in which the monomials are orthonormal. Let ∗xT=xI∖T*x_T=x_{I\setminus T}.

Lemma 2.2. For 0≤t≤m/20\le t\le m/2, Em−2t:BIt→∼BIm−t,ker⁡(Em−2t+1∣BIt)=ker⁡(F∣BIt).E^{m-2t}:B_I^t\xrightarrow{\sim}B_I^{m-t},\qquad \ker(E^{m-2t+1}|_{B_I^t})=\ker(F|_{B_I^t}). For every rational vector space MM and every z∈M⊗ker⁡(F∣BIt)z\in M\otimes\ker(F|_{B_I^t}), Em−2tz=(−1)t(m−2t)! ∗z. E^{m-2t}z=(-1)^t(m-2t)!\,*z.(2.4)

Proof. The displayed commutators give the standard finite-dimensional sl2\mathfrak{sl}_2 representation on BIB_I. More explicitly, if Fv=0Fv=0, v∈BItv\in B_I^t, then FEjv=j(m−2t−j+1)Ej−1v.FE^jv=j(m-2t-j+1)E^{j-1}v. Adjointness shows that the string generated by a nonzero such vv has nonzero terms through Em−2tvE^{m-2t}v and then terminates. Decomposing successively into ker⁡F\ker F and im⁡E\operatorname{im}E, which are orthogonal, gives the Lefschetz strings and the first two assertions.

For the exact constant, write z=∑∣T∣=tuTxTz=\sum_{|T|=t}u_Tx_T. From Fz=0Fz=0 one obtains, for every J⊂IJ\subset I with ∣J∣<t|J|<t, ∑T⊃J∣T∣=tuT=0.\sum_{\substack{T\supset J\\|T|=t}}u_T=0. This follows by looking at the coefficient of xJx_J in Ft−∣J∣zF^{t-|J|}z, which is (t−∣J∣)!(t-|J|)! times the sum. If K⊂IK\subset I has size m−tm-t and R0=I∖KR_0=I\setminus K, inclusion–exclusion gives ∑T⊂K∣T∣=tuT=∑J⊂R0(−1)∣J∣∑T⊃J∣T∣=tuT=(−1)tuR0.\sum_{\substack{T\subset K\\|T|=t}}u_T =\sum_{J\subset R_0}(-1)^{|J|} \sum_{\substack{T\supset J\\|T|=t}}u_T =(-1)^tu_{R_0}. The coefficient of xKx_K in Em−2tzE^{m-2t}z is (m−2t)!(m-2t)! times the left side. This proves (2.4), including t=0t=0. ◻

2.3. Proof of the arithmetic Lefschetz reduction

We prove Theorem 2.1. Suppose first that all the noncentral cores vanish. On a remaining block put k=∣S∣=2c−1k=|S|=2c-1, m=n−km=n-k, and t=r−ct=r-c. Then n+1−2r=m−2t,n+1-2r=m-2t, and the desired map is 1⊗Em−2t1\otimes E^{m-2t}. It is an isomorphism by Lemma 2.2. If t<0t<0, both its source and target in this block vanish. Taking the direct sum proves (ii).

Conversely, suppose USc(E)≠0U_S^c(E)\ne0 and put d=k+1−2cd=k+1-2c. The homologically trivial groups force 1≤c≤k1\le c\le k. If d>0d>0, take r=cr=c. The required map has exponent m+d>mm+d>m, so it kills USc(E)⊗1U_S^c(E)\otimes1; it cannot be injective. If d<0d<0, take r=k+1−cr=k+1-c. This satisfies 1≤r≤⌊(n+1)/2⌋1\le r\le\lfloor(n+1)/2\rfloor. The target has the nonzero component USc(E)⊗xIU_S^c(E)\otimes x_I in degree c+mc+m, whereas the source in this same block has polynomial degree r−c=d<0r-c=d<0 and is zero. Multiplication by ℓ\ell preserves the block, so surjectivity fails. Hence d=0d=0 for every nonzero core. This proves (i).

2.4. Primitive heights and their exact constants

Adjunction and tPS,T,U=PS,U,T{}^tP_{S,T,U}=P_{S,U,T} show that Chow blocks are height-orthogonal unless their point and fundamental-class indices are complementary and their SS indices agree. In the complementary case, hX(jTu,jI∖Tv)=hYS(u,v). h_X(j_Tu,j_{I\setminus T}v)=h_{Y_S}(u,v).(2.5) Indeed the transpose of the first embedding composed with the second integrates one degree-one point in each omitted factor.

Define Pℓr(XE)=ker⁡(ℓn+2−2r:Vr(XE)→Vn+2−r(XE))P_\ell^r(X_E)=\ker\bigl( \ell^{n+2-2r}:V^r(X_E)\to V^{n+2-r}(X_E)\bigr) for 1≤r≤⌊(n+1)/2⌋1\le r\le\lfloor(n+1)/2\rfloor.

Theorem 2.3 (Primitive sum of squares). Let ∣S∣=2c−1|S|=2c-1, m=n−∣S∣m=n-|S|, and 0≤t=r−c≤m/20\le t=r-c\le m/2. Write a vector in the corresponding block as zS=∑∣T∣=tuTxT,uT∈USc(E).z_S=\sum_{|T|=t}u_Tx_T, \qquad u_T\in U_S^c(E). It is primitive exactly when FzS=0Fz_S=0. For such a vector, qℓ,r(zS):=(−1)rhX(zS,ℓn+1−2rzS)=(m−2t)! aI∑∣T∣=tqS,c(uT), q_{\ell,r}(z_S):=(-1)^rh_X(z_S,\ell^{n+1-2r}z_S) =(m-2t)!\,a_I\sum_{|T|=t}q_{S,c}(u_T),(2.6) where aI=∏i∈Iaia_I=\prod_{i\in I}a_i and qS,c(u)=(−1)chYS(u,u)q_{S,c}(u)=(-1)^ch_{Y_S}(u,u).

Proof. The primitive exponent is m−2t+1m-2t+1, so the equivalence follows from Lemma 2.2. The same lemma gives ℓn+1−2rzS=(−1)t(m−2t)! ∗zS\ell^{n+1-2r}z_S=(-1)^t(m-2t)!\,*z_S. In pairing zSz_S with ∗zS*z_S, only the identical coefficient indices survive by (2.5). Their weights multiply to aTaI∖T=aIa_Ta_{I\setminus T}=a_I. The remaining sign is (−1)r+t=(−1)c(-1)^{r+t}=(-1)^c, since r=c+tr=c+t. This proves the formula. ◻

Corollary 2.4. Suppose all the noncentral subproduct cores vanish and all central forms qS,cq_{S,c} are real positive definite over a fixed number field. Then the complementary height pairings on XX are nondegenerate over that field, also after real extension, and qℓ,rq_{\ell,r} is real positive definite on PℓrP_\ell^r. Its associated bilinear form bℓ,rb_{\ell,r} satisfies ∣bℓ,r(z,w)∣2≤qℓ,r(z)qℓ,r(w).|b_{\ell,r}(z,w)|^2 \le q_{\ell,r}(z)q_{\ell,r}(w). If noncentral vanishing holds over every finite extension and each central integral image has bounded-degree Northcott, then so does ΛXr(E)∩Pℓr\Lambda_X^r(E)\cap P_\ell^r for qℓ,rq_{\ell,r}.

Proof. The decomposition into complementary blocks and the definiteness of their central forms prove nondegeneracy. Distinct SS are orthogonal. Formula (2.6) therefore proves primitive definiteness. Cauchy–Schwarz follows by applying positivity to qℓ,r(z+tw)q_{\ell,r}(z+tw), t∈Rt\in\mathbb R.

For the bounded-degree assertion, central Northcott gives nonnegativity over each finite extension: a nonzero rational class of nonpositive height would have infinitely many integral multiples in a bounded height ball. For an integral primitive zz of height at most BB, every coefficient satisfies qS,c(uT)≤B(m−2t)!aI,NauT∈ΛUSc(E).q_{S,c}(u_T)\le \frac{B}{(m-2t)!a_I},\qquad N_a u_T\in\Lambda_{U_S^c}(E). Thus central Northcott and (2.3) give finitely many possibilities for every coefficient. There are only finitely many blocks, and the decomposition is injective. ◻

3. Central projectors and finite height balls

The all-h1h^1 central group has a curve-visible summand with a Néron–Tate height. We first construct its Chow projector and identify the algebraic radical. The integral finiteness criterion then follows by orthogonal decomposition and one denominator bound.

3.1. Curve-invisible cycles

Lemma 3.1. Let CC be a smooth projective curve and Γ∈CH⁡r(C×X)Q\Gamma\in\operatorname{CH}^r(C\times X)_\mathbb Q, with dim⁡X=n\dim X=n. If Γ∗H1(C)=0\Gamma_*H^1(C)=0 in H2r−1(X)H^{2r-1}(X), then tΓ∗w=0in Pic⁡0(C)Q(w∈Vn+1−r(X)).{}^t\Gamma_*w=0\quad\text{in }\operatorname{Pic}^0(C)_\mathbb Q \qquad(w\in V^{n+1-r}(X)). Hence Γ∗Pic⁡0(C)Q\Gamma_*\operatorname{Pic}^0(C)_\mathbb Q is in the left height radical.

Proof. By Poincaré duality, the transpose map from H2n+1−2r(X)H^{2n+1-2r}(X) to H1(C)H^1(C) is zero. Functoriality of the Abel–Jacobi map under Chow correspondences therefore makes the Abel–Jacobi class of tΓ∗w{}^t\Gamma_*w zero. On a curve, the rationalized Abel–Jacobi map is the identification with Pic⁡0(C)Q\operatorname{Pic}^0(C)_\mathbb Q. Vanishing after complex base change descends for divisor classes with rational coefficients. The final assertion is adjunction. The correspondence functoriality used here is also proved in [6, Lemma 4.11]. ◻

Algebraically trivial cycles are generated by images of degree-zero divisors on smooth projective curves. For geometric algebraic triviality over EE, the finitely many data can be defined over a finite extension and then traced to EE. This convention is used throughout.

3.2. The maximal curve-visible projector

Let k=2c−1k=2c-1, Y=∏i=1kCiY=\prod_{i=1}^kC_i, and Π=⨂iπ1,i\Pi=\bigotimes_i\pi_{1,i}. Set UE=Π∗Vc(YE),AE=UE∩CH⁡c(YE)alg,Q,GE=UE/AE.U_E=\Pi_*V^c(Y_E),\qquad A_E=U_E\cap\operatorname{CH}^c(Y_E)_{\mathrm{alg},\mathbb Q},\qquad G_E=U_E/A_E. For the product polarization ℓY=∑ai[oi]\ell_Y=\sum a_i[o_i], the actual Chow identity ℓYΠ=0\ell_Y\Pi=0 follows from [oi]π1,i=0[o_i]\pi_{1,i}=0. Thus the geometric cohomology of Π\Pi is middle and primitive.

Let W⊂ΠHk(Y,Q)W\subset\Pi H^k(Y,\mathbb Q) be the span of all images H1(C)⟶ΠHk(Y,Q)H^1(C)\longrightarrow\Pi H^k(Y,\mathbb Q) of codimension-cc curve correspondences over finite extensions of FF. Choose finitely many correspondences spanning WW, and one field F0F_0 defining them and their source points. Write their matrix as Γ=(Γj)j\Gamma=(\Gamma_j)_j, with Γj=ΠΓjπ1,j\Gamma_j=\Pi\Gamma_j\pi_{1,j}, and put N=⨁jh1(Cj)N=\bigoplus_jh^1(C_j), J=∏jPic⁡0(Cj)J=\prod_j\operatorname{Pic}^0(C_j).

Proposition 3.2. There are actual Chow projectors e,Re,R over F0F_0, with e=te,e2=e,eΠ=Πe=e,R=Π−e,e={}^te,\quad e^2=e,\quad e\Pi=\Pi e=e,\quad R=\Pi-e, such that the following hold for every finite E/F0E/F_0.

  1. The cohomological action of ee is the orthogonal projection onto WW. There is a fixed abelian variety B/F0B/F_0 with eUE≃B(E)QeU_E\simeq B(E)_\mathbb Q, through fixed Chow correspondences. The form qe=(−1)chYq_e=(-1)^ch_Y on this summand is a positive Néron–Tate form.

  2. With KE=RAEK_E=RA_E, KE=ker⁡(AJ⁡Q∣AE)=AE∩rad⁡(hY∣UE)=rad⁡(hY∣AE).K_E=\ker(\operatorname{AJ}_\mathbb Q|_{A_E}) =A_E\cap\operatorname{rad}(h_Y|_{U_E}) =\operatorname{rad}(h_Y|_{A_E}).

  3. There is an orthogonal decomposition UE=eUE⊕RUEU_E=eU_E\oplus RU_E. The residual form on RUE/RAE≃GERU_E/RA_E\simeq G_E is independent of the choice of maximal curve-visible projector. Write it as hGh_G and set qG=(−1)chGq_G=(-1)^ch_G.

Proof. Set a=tΓΓa={}^t\Gamma\Gamma. Since 2c−k=12c-k=1, its entries are divisor correspondences, so a∈End⁡(N)=End⁡(J)Qa\in\operatorname{End}(N)=\operatorname{End}(J)_\mathbb Q. This realization is faithful, and transposition corresponds to Rosati adjunction for the product principal polarization. Geometric Hodge–Riemann on the primitive subspace WW implies that (−1)c−1a(-1)^{c-1}a is positive semidefinite and ker⁡a=ker⁡Γ\ker a=\ker\Gamma on H1H^1. In particular aa is semisimple. Choose a polynomial f∈Q[t]f\in\mathbb Q[t] equal to t−1t^{-1} on the nonzero spectrum and zero at zero, and set a+=f(a),p=a+a,e=Γa+tΓ.a^+=f(a),\qquad p=a^+a,\qquad e=\Gamma a^+{}^t\Gamma. Faithfulness on NN makes a+aa+=a+a^+aa^+=a^+, p2=p=tpp^2=p={}^tp actual Chow identities. These identities prove the assertions about ee and its cohomology. Let B=pJB=pJ, up to isogeny. The maps Γ\Gamma and a+tΓa^+{}^t\Gamma identify B(E)QB(E)_\mathbb Q with eUEeU_E. Adjunction gives hY(Γx,Γy)=hcurves(x,ay).h_Y(\Gamma x,\Gamma y)=h_{\mathrm{curves}}(x,ay). The curve normalization and the positivity of (−1)c−1a(-1)^{c-1}a on pJpJ give (i).

For every later curve correspondence Δ\Delta, the geometric H1H^1 action of RΔR\Delta is zero by maximality of WW. Curve generation and Lemma 3.1 give RAE⊆rad⁡(hY∣UE)RA_E\subseteq\operatorname{rad}(h_Y|_{U_E}). The Abel–Jacobi map is zero on RAERA_E by the same geometric action. Its restriction to eUEeU_E is injective after rationalization: the corresponding map from BB to the intermediate Jacobian has injective rational H1H_1, hence finite kernel. Since eUE⊆AEeU_E\subseteq A_E and its height is nondegenerate, all equalities in (ii) follow.

The self-transpose projectors give the orthogonal decomposition. The radical assertion just proved makes the residual form well defined on GEG_E. Two orthogonalized lifts of the same Griffiths class differ by an element of AEA_E orthogonal to AEA_E. By (ii), their difference is in the full radical. Thus the residual form is independent of the lift and of ee. ◻

The cohomological subspace WW and its orthogonal complement are defined independently of the chosen spanning correspondences. Betti and étale comparison transports them to each Qℓ0\mathbb Q_{\ell_0} realization. The algebraic construction of ee in the proof uses geometric Hodge–Riemann and the faithful realization of h1h^1 endomorphisms; it is available independently of the height data.

Lemma 3.3. The projected integral image in eUEeU_E has bounded-degree Northcott for qeq_e, allowing any fixed denominator bound.

Proof. Put D=a+tΓD=a^+{}^t\Gamma. These fixed curve correspondences satisfy De=DDe=D. One integer NN clears their denominators, so an integral cycle zz gives a genuine point NDz∈J(E)NDz\in J(E). The same point detects ezez, even though ezez may have rational coefficients.

On the fixed factor pJ(F‾)QpJ(\overline F)_\mathbb Q, the ambient Néron–Tate height of JJ is bounded above by a fixed positive multiple of qeq_e under the inverse DD. This is comparison of two polarizations on BB; passing to rational points removes torsion without changing either height. Thus a bound for qe(ez)q_e(ez) bounds the height of the actual point NDzNDz. Classical bounded-degree Northcott on JJ, followed by the injectivity of DD on eUEeU_E, proves the assertion. Fixed extra denominators are absorbed into NN. See [2, Sections 9–10] for the height and Mordell–Weil facts used here. ◻

3.3. The exact central Northcott reduction

Continue with the central notation of Subsection 3.2. Let ΛU(E)=Π∗ΛYc(E)\Lambda_U(E)=\Pi_*\Lambda_Y^c(E) and let ΛG(E)\Lambda_G(E) be its image in GEG_E. The quotient may be viewed in the geometric group: if an element defined over EE becomes geometrically algebraically trivial, it is in AEA_E by definition. In particular, these rational quotients inject under field extension.

Theorem 3.4. The following conditions are equivalent:

  1. The form qU(z)=(−1)chY(z,z)q_U(z)=(-1)^ch_Y(z,z) has bounded-degree Northcott on ΛU\Lambda_U.

  2. For every finite E/F0E/F_0, KE=ker⁡(AJ⁡Q∣AE)=0K_E=\ker(\operatorname{AJ}_\mathbb Q|_{A_E})=0, and qGq_G has bounded-degree Northcott on ΛG\Lambda_G.

There is also a fixed-field version, with all groups and the vanishing condition taken over one fixed EE.

Proof. Suppose (i). If 0≠z∈KE0\ne z\in K_E, multiply by one integer to make zz integral. All positive integer multiples of this class are distinct and have height zero, contradicting Northcott. Thus KE=0K_E=0.

Choose an integer M≥1M\ge1 such that MRMR is represented by an integral correspondence. For ξ∈ΛG(E)\xi\in\Lambda_G(E), choose z∈ΛU(E)z\in\Lambda_U(E) mapping to ξ\xi. Then MRz∈ΛU(E),qU(MRz)=M2qG(ξ),[MRz]=Mξ.MRz\in\Lambda_U(E),\qquad q_U(MRz)=M^2q_G(\xi),\qquad [MRz]=M\xi. Northcott for ΛU\Lambda_U bounds the possible classes MRzMRz, and hence the possible ξ\xi. This proves residual Northcott.

Conversely, assume (ii). Residual Northcott implies qG(ξ)>0q_G(\xi)>0 for every nonzero rational class ξ∈GE\xi\in G_E. Indeed one multiple of ξ\xi is in ΛG(E)\Lambda_G(E), and a nonpositive height would give infinitely many integral multiples of height at most zero. For every z∈UEz\in U_E, orthogonality gives qU(z)=qe(ez)+qG([z]). q_U(z)=q_e(ez)+q_G([z]).(3.1) Both terms are nonnegative. If zz is integral and qU(z)≤Bq_U(z)\le B, residual Northcott bounds the possibilities for [z][z]. The denominators of the fixed projector ee, followed by Lemma 3.3, bound the possibilities for ezez. Finally z⟼(ez,[z])z\longmapsto(ez,[z]) is injective: its kernel is RAE=KE=0RA_E=K_E=0. This proves (i). The argument with EE fixed is identical. ◻

The equivalence separates the algebraic Abel–Jacobi kernel from integral finiteness on the residual quotient. In fact, Northcott also forces real positive definiteness, by Lemma 5.1. This observation will combine the central equivalence with the primitive calculation in Theorem 1.3.

4. Residual arithmetic regulators

The residual Chow projector isolates a single geometric Hochschild–Serre row. This gives an arithmetic cycle map on the Griffiths quotient and a finite-dimensional target.

Theorem 4.1 (Residual arithmetic regulator). For every prime ℓ0\ell_0 and finite E/F0E/F_0, set Vℓ0=R∗Heˊt2c−1(YF‾,Qℓ0(c)).V_{\ell_0}=R_*H^{2c-1}_{\textnormal{ét}}(Y_{\overline F},\mathbb Q_{\ell_0}(c)). There is a natural Q\mathbb Q-linear map aℓ0,E:GE⟶H1(E,Vℓ0)a_{\ell_0,E}:G_E\longrightarrow H^1(E,V_{\ell_0}) induced by the continuous étale cycle class. It is independent of the curve-visible Chow lift under the common cohomological identification, and is compatible with restriction and trace. There is one finite set S0S_0 of places of F0F_0 such that inflation identifies H1(GE,SE,Vℓ0)≃H1(E,Vℓ0),H^1(G_{E,S_E},V_{\ell_0})\simeq H^1(E,V_{\ell_0}), where SES_E consists of places above S0S_0 and ℓ0\ell_0. This whole target is finite-dimensional over Qℓ0\mathbb Q_{\ell_0}.

4.1. The single-row construction

We use continuous étale cohomology and its cycle class map as in [8; 4]. Fix a rational prime ℓ0\ell_0, and put k=2c−1,Vℓ0=R∗Heˊtk(YF‾,Qℓ0(c)).k=2c-1,\qquad V_{\ell_0}=R_*H^k_{\textnormal{ét}}(Y_{\overline F},\mathbb Q_{\ell_0}(c)). The action of the actual Chow idempotent RR on geometric cohomology is concentrated in degree kk. Apply it to the correspondence-compatible Hochschild–Serre spectral sequence Hp(E,Heˊtq(YF‾,Qℓ0(c)))⟹Hcontp+q(YE,Qℓ0(c)).H^p\bigl(E,H^q_{\textnormal{ét}}(Y_{\overline F},\mathbb Q_{\ell_0}(c))\bigr) \Longrightarrow H^{p+q}_{\mathrm{cont}}(Y_E,\mathbb Q_{\ell_0}(c)). Only the row q=kq=k survives. Hence there is a canonical identification R∗Hcont2c(YE,Qℓ0(c))≃H1(E,Vℓ0). R_*H^{2c}_{\mathrm{cont}}(Y_E,\mathbb Q_{\ell_0}(c)) \simeq H^1(E,V_{\ell_0}).(4.1) On this summand the Hochschild–Serre filtration satisfies F1=HF^1=H and F2=0F^2=0.

Proposition 4.2. The assignment z↦cl⁡ℓ0(R∗z)z\mapsto\operatorname{cl}_{\ell_0}(R_*z), followed by (4.1), annihilates AEA_E. It therefore defines a map aℓ0,E:GE⟶H1(E,Vℓ0).a_{\ell_0,E}:G_E\longrightarrow H^1(E,V_{\ell_0}). The map is compatible with restriction, corestriction and correspondences preserving the indicated projectors. Under the common cohomological identification, it is independent of the maximal curve-visible projector ee.

Proof. For a degree-zero divisor DD on a pointed curve CC, its continuous cycle class lies in the π1,C\pi_{1,C}-part of Hcont2(C,Qℓ0(1))H^2_{\mathrm{cont}}(C,\mathbb Q_{\ell_0}(1)). This part has only the Hochschild–Serre term H1(E,H1(CF‾,Qℓ0(1)))H^1(E,H^1(C_{\overline F},\mathbb Q_{\ell_0}(1))). If Δ∈CH⁡c(C×Y)Q\Delta\in\operatorname{CH}^c(C\times Y)_\mathbb Q is a projected curve correspondence, RΔπ1,CR\Delta\pi_{1,C} has zero geometric H1H^1 action by Proposition 3.2. By functoriality its map on the only associated graded term is zero. Its image in the target of (4.1) is consequently in F2=0F^2=0.

Curve generation proves the vanishing for algebraically trivial cycles after a finite extension L/EL/E. It descends to EE because cor⁡L/Eres⁡L/E=[L:E]\operatorname{cor}_{L/E}\operatorname{res}_{L/E}=[L:E] on rational cohomology. Restriction is therefore injective. Compatibility with restriction, trace and the stated correspondences follows from the cycle map and the spectral sequence.

For independence, let e′e' be another maximal curve-visible lift. Its geometric action is the same orthogonal projector onto WW. The Π\Pi-part of total degree 2c=k+12c=k+1 also has only its first associated graded term. Thus equality of the geometric actions of e,e′e,e' gives equality of their actions on this whole continuous-cohomology group. The cycle classes of (Π−e)z(\Pi-e)z and (Π−e′)z(\Pi-e')z coincide. ◻

4.2. Arithmetic models and weight-minus-one cohomology

In the notation GE,SEG_{E,S_E}, all archimedean places are included in SES_E; displayed conditions on SES_E specify its finite places.

Proposition 4.3. There is a finite set S0S_0 of finite places of F0F_0 such that, for every prime ℓ0\ell_0 and finite E/F0E/F_0, aℓ0,E(GE)⊆H1(GE,SE,Vℓ0)⊆H1(E,Vℓ0),a_{\ell_0,E}(G_E) \subseteq H^1(G_{E,S_E},V_{\ell_0}) \subseteq H^1(E,V_{\ell_0}), where SES_E consists of places above S0S_0 or ℓ0\ell_0. The first space is finite-dimensional over Qℓ0\mathbb Q_{\ell_0}.

Proof. Spread YY to a smooth proper scheme over Spec⁡OF0[1/S0]\operatorname{Spec}\mathcal O_{F_0}[1/S_0]. Spread the finitely many cycle representatives and rational-equivalence witnesses defining the relative correspondence RR and the identity R2=RR^2=R. After enlarging S0S_0 once, these data all exist on this model. For a chosen ℓ0\ell_0, work over Uℓ0=Spec⁡OF0[1/(S0∪{v:v∣ℓ0})].U_{\ell_0}=\operatorname{Spec}\mathcal O_{F_0} [1/(S_0\cup\{v:v\mid\ell_0\})]. The higher direct images of the smooth proper morphism are lisse. Since their RR-parts vanish at the geometric generic point in degrees other than kk, they vanish everywhere in those degrees. The surviving part is a lisse sheaf Vℓ0\mathcal V_{\ell_0}. Applying RR to Leray yields R∗Hcont2c(Y,Qℓ0(c))≃H1(Uℓ0,Vℓ0).R_*H^{2c}_{\mathrm{cont}}(\mathcal Y,\mathbb Q_{\ell_0}(c)) \simeq H^1(U_{\ell_0},\mathcal V_{\ell_0}). Every generic-fibre cycle extends by closure to this regular model. Compatibility of cycle classes with restriction proves the factorization. Arithmetic étale cohomology finiteness gives finite Qℓ0\mathbb Q_{\ell_0}-dimension. Equivalently one may apply the model argument of [8, Proposition 2.5 and Corollary 2.6].

For a finite extension E/F0E/F_0, base change this same model to OE[1/SE]\mathcal O_E[1/S_E]. This is a finite flat base change, preserves smoothness, and preserves the relative correspondence identities. Taking closures of cycles over EE proves the assertion uniformly. In degree one, the étale cohomology of the base with this lisse sheaf identifies with H1(GE,SE,Vℓ0)H^1(G_{E,S_E},V_{\ell_0}). ◻

The residual representation has a further useful property: its geometric weight is 2c−1−2c=−12c-1-2c=-1. This strengthens the preceding image statement to one about the whole target.

Proposition 4.4 (Weight-minus-one cohomology). Let S0S_0 contain the bad-reduction places of Y/F0Y/F_0. For every finite E/F0E/F_0, every prime ℓ0\ell_0, and every finite v∉SEv\notin S_E, H1(Ev,Vℓ0)=0.H^1(E_v,V_{\ell_0})=0. Inflation is an isomorphism H1(GE,SE,Vℓ0)→∼H1(E,Vℓ0).H^1(G_{E,S_E},V_{\ell_0})\xrightarrow{\sim} H^1(E,V_{\ell_0}). In particular, the full group on the right is finite-dimensional. For every stable free Zℓ0\mathbb Z_{\ell_0}-lattice T⊂Vℓ0T\subset V_{\ell_0}, inflation also identifies H1(GE,SE,T)H^1(G_{E,S_E},T) with H1(E,T)H^1(E,T), and these groups are finitely generated over Zℓ0\mathbb Z_{\ell_0}.

Proof. At v∉SEv\notin S_E, smooth proper base change makes inertia act trivially on Vℓ0V_{\ell_0}. The eigenvalues of geometric Frobenius have complex absolute value qv−1/2q_v^{-1/2}, by [3, Theorem 1.6]. Consequently Vℓ0/(Frv−1)Vℓ0=0V_{\ell_0}/(\mathrm{Fr}_v-1)V_{\ell_0}=0. The ℓ0\ell_0-part of tame inertia is Zℓ0(1)\mathbb Z_{\ell_0}(1), and continuous homomorphisms to Vℓ0V_{\ell_0} kill wild inertia and the other tame factors. Therefore H1(Iv,Vℓ0)Frv=Vℓ0(−1)Frv=0,H^1(I_v,V_{\ell_0})^{\mathrm{Fr}_v} =V_{\ell_0}(-1)^{\mathrm{Fr}_v}=0, since the eigenvalues on the twist have absolute value qv1/2q_v^{1/2}. Local inflation–restriction proves the first assertion.

For a global continuous cocycle, the restriction to every such inertia group is zero. Indeed its local class is zero, and inertia acts trivially, so even the local coboundaries restrict to zero on inertia. The kernel of Gal⁡(F‾/E)→GE,SE\operatorname{Gal}(\overline F/E)\to G_{E,S_E} is the closed normal subgroup generated by these inertia groups. It acts trivially on Vℓ0V_{\ell_0}. If zz is the cocycle and ii is in one of these inertia groups, the cocycle identity gives z(gig−1)=(1−gig−1)z(g)=0z(gig^{-1})=(1-gig^{-1})z(g)=0. The restriction of zz to the kernel is thus a continuous homomorphism vanishing on all generators, hence on the whole kernel. It factors through GE,SEG_{E,S_E}. Inflation is injective, proving the isomorphism.

For TT, the inertia invariants are T(−1)Frv=0T(-1)^{\mathrm{Fr}_v}=0, as they inject into the rational ones. The same global cocycle argument proves integral inflation. The local integral group may be the nonzero finite group T/(Frv−1)TT/(\mathrm{Fr}_v-1)T; its restriction to inertia is still zero. Finally, in degree one the continuous-cohomology convention agrees with these continuous cocycles: compactness gives a bounded image in a scalar multiple of a stable lattice.

To prove finite generation, write G=GE,SEG=G_{E,S_E}. The groups Hi(G,T/ℓ0aT)H^i(G,T/\ell_0^aT) are finite by [7, Corollary I.4.15]. The inverse system of degree-zero groups is Mittag–Leffler, so H1(G,T)=lim←⁡aH1(G,T/ℓ0aT)H^1(G,T)=\varprojlim_a H^1(G,T/\ell_0^aT) is a compact pro-ℓ0\ell_0 module. The exact sequence for 0→T→ℓ0T→T/ℓ0T→00\to T\xrightarrow{\ell_0}T\to T/\ell_0T\to0 embeds its reduction modulo ℓ0\ell_0 in the finite group H1(G,T/ℓ0T)H^1(G,T/\ell_0T). The compact form of Nakayama’s lemma therefore makes H1(G,T)H^1(G,T) finitely generated. Every continuous cocycle into Vℓ0V_{\ell_0} takes values in ℓ0−aT\ell_0^{-a}T for some aa, and a coboundary can likewise be cleared of denominators. Hence H1(G,Vℓ0)=H1(G,T)⊗Zℓ0Qℓ0,H^1(G,V_{\ell_0})=H^1(G,T)\otimes_{\mathbb Z_{\ell_0}}\mathbb Q_{\ell_0}, which proves the rational finite-dimensionality. ◻

Proposition 4.2 and Proposition 4.4 prove Theorem 4.1. These ramification statements concern cohomology over EE, rather than the ramification of E/F0E/F_0. The latter enters separately in Section 6.

Corollary 4.5. If, for one prime ℓ0\ell_0, the scalar-extended map GE⊗QQℓ0⟶H1(E,Vℓ0) G_E\otimes_\mathbb Q\mathbb Q_{\ell_0} \longrightarrow H^1(E,V_{\ell_0})(4.2) is injective, then dim⁡QGE<∞\dim_\mathbb QG_E<\infty.

Proof. Its image lies in the finite-dimensional space of Proposition 4.3. Scalar extension preserves the dimension of a rational vector space. ◻

The scalar extension in (4.2) is essential: a finite-dimensional Qℓ0\mathbb Q_{\ell_0}-space has infinite dimension over Q\mathbb Q.

4.3. Local integral bounds

Proposition 4.6. For every fixed E,ℓ0E,\ell_0, the subgroup aℓ0,E(ΛG(E))a_{\ell_0,E}(\Lambda_G(E)) is contained in a finitely generated Zℓ0\mathbb Z_{\ell_0}-module in the rational cohomological target. If (4.2) is injective, then ΛG(E)⊗ZZℓ0\Lambda_G(E)\otimes_\mathbb Z\mathbb Z_{\ell_0} is finitely generated over Zℓ0\mathbb Z_{\ell_0}, and ΛG(E)/ℓ0ΛG(E)\Lambda_G(E)/\ell_0\Lambda_G(E) is finite.

Proof. Choose one integer MM clearing the Chow denominators of RR. For an integral representative ZZ, Mcl⁡ℓ0(RZ)=cl⁡ℓ0((MR)Z)M\operatorname{cl}_{\ell_0}(RZ)=\operatorname{cl}_{\ell_0}((MR)Z) is in the image of the integral continuous cycle class map. This image is a finitely generated Zℓ0\mathbb Z_{\ell_0}-module by [4, Theorem 2.4]. Let Mℓ0M_{\ell_0} be its image in rational continuous cohomology. The intersection Mℓ0∩R∗Hcont2c(YE,Qℓ0(c))M_{\ell_0}\cap R_*H^{2c}_{\mathrm{cont}}(Y_E,\mathbb Q_{\ell_0}(c)) is again finitely generated and contains Mcl⁡ℓ0(RZ)M\operatorname{cl}_{\ell_0}(RZ). Applying the single-row identification to this intersection and dividing by MM gives the claimed bound.

Under scalar-extended injectivity, the induced map on ΛG(E)⊗Zℓ0\Lambda_G(E)\otimes\mathbb Z_{\ell_0} is injective. Its source is torsion-free over Zℓ0\mathbb Z_{\ell_0}, and its localization is GE⊗Qℓ0G_E\otimes\mathbb Q_{\ell_0}, because ΛG(E)\Lambda_G(E) spans GEG_E. Its image is a submodule of a finitely generated Zℓ0\mathbb Z_{\ell_0}-module, hence is finitely generated. Reduction modulo ℓ0\ell_0 proves the last assertion. ◻

The integral image here lies in a rational quotient. Unrestricted integral continuous cycle class maps have torsion counterexamples over number fields in codimension three [9, Theorems 1.3–1.4]. Those torsion kernels and the torsion-free subgroup ΛG(E)\Lambda_G(E) are different objects.

5. Integral discreteness and height gaps

5.1. Real positivity forced by finiteness

Lemma 5.1. Let Λ\Lambda be a subgroup spanning a rational vector space VV, and let qq be a real-valued quadratic form on VV. If qq has Northcott on Λ\Lambda, its extension to V⊗QRV\otimes_\mathbb Q\mathbb R is positive definite.

Proof. It suffices to consider a finite-dimensional rational subspace and choose a rational basis v1,…,vsv_1,\ldots,v_s in Λ\Lambda. Let QQ be the Gram matrix. A negative real direction would, by rational approximation and clearing denominators, give a nonzero v∈∑iZviv\in\sum_i\mathbb Zv_i with q(v)<0q(v)<0. Its integer multiples contradict Northcott. Thus QQ is positive semidefinite.

If 0≠u∈ker⁡Q0\ne u\in\ker Q, round the coordinates of tutu to integers, obtaining zt∈Zsz_t\in\mathbb Z^s with uniformly bounded zt−tuz_t-tu. Then q(∑i(zt)ivi)=(zt−tu)tQ(zt−tu)q\Bigl(\sum_i(z_t)_iv_i\Bigr) =(z_t-tu)^tQ(z_t-tu) is uniformly bounded. The rounded vectors are unbounded as t→∞t\to\infty, hence take infinitely many distinct values. This again contradicts Northcott. Therefore every such finite Gram matrix is positive definite, which proves the assertion. ◻

In particular, residual integral Northcott already contains the real-positivity assertion for qGq_G. The converse requires an arithmetic condition on the integral image.

5.2. Finite-rank lattices

Theorem 5.2. Let Λ\Lambda be torsion-free of finite rank, V=Λ⊗QV=\Lambda\otimes\mathbb Q, and suppose that qq is real positive definite on V⊗RV\otimes\mathbb R. The following are equivalent:

  1. qq has Northcott on Λ\Lambda.

  2. There is ε>0\varepsilon>0 such that q(λ)≥εq(\lambda)\ge\varepsilon for every nonzero λ∈Λ\lambda\in\Lambda.

  3. Λ\Lambda is finitely generated.

  4. For a rational basis v1,…,vsv_1,\ldots,v_s of VV, there is an integer N≥1N\ge1 such that Λ⊂N−1∑iZvi\Lambda\subset N^{-1}\sum_i\mathbb Zv_i.

  5. The diagonal image of Λ\Lambda in V(Af)V(\mathbb A_f) is relatively compact.

The zero group satisfies these statements with the gap condition interpreted vacuously.

Proof. Finiteness in a unit height ball gives a positive minimum, or the bound 11 if that ball contains no nonzero point. This proves (i) implies (ii). The gap makes Λ\Lambda uniformly discrete for the Euclidean norm q\sqrt q. Choose a rational basis in Λ\Lambda and reduce elements of Λ\Lambda modulo its Z\mathbb Z-span into a bounded fundamental parallelepiped. Uniform discreteness makes the set of representatives finite. Hence this span has finite index in Λ\Lambda, proving (iii) and (iv). Conversely, a bounded subset of a Euclidean lattice is finite, so (iv) implies (i). Finite generation gives (iv) for any rational basis by clearing the finitely many coefficients.

Every compact subset of V(Af)V(\mathbb A_f) is contained in N−1Λ0⊗Z^N^{-1}\Lambda_0\otimes\widehat{\mathbb{Z}} for some integer NN and a fixed reference lattice Λ0\Lambda_0. This follows by a finite subcover from the open compact sets of that form. Conversely each such set is compact. Intersecting with V(Q)V(\mathbb Q) proves the equivalence with (v). ◻

If Λ⊂N−1Λ0\Lambda\subset N^{-1}\Lambda_0, and the least eigenvalue of the Gram matrix in a basis of Λ0\Lambda_0 is μ>0\mu>0, one obtains the explicit estimate #{λ∈Λ:q(λ)≤B}≤(1+2⌊NB/μ⌋)s(B≥0). \#\{\lambda\in\Lambda:q(\lambda)\le B\} \le\left(1+2\left\lfloor N\sqrt{B/\mu}\right\rfloor\right)^s \qquad(B\ge0).(5.1)

5.3. An all-prime saturation criterion

Theorem 5.3. Let VV be a finite-dimensional rational vector space, Λ0⊂V\Lambda_0\subset V a full lattice, and Λ⊂V\Lambda\subset V a subgroup. For every prime pp, let Dp:V⊗Qp↪WpD_p:V\otimes\mathbb Q_p\hookrightarrow W_p be an injective linear map to a finite-dimensional space, and let Mp⊂WpM_p\subset W_p be a full Zp\mathbb Z_p-lattice. Assume Dp(Λ)⊆Mpfor every p,Dp−1(Mp)=Λ0⊗Zpfor almost every p.D_p(\Lambda)\subseteq M_p\quad\text{for every }p, \qquad D_p^{-1}(M_p)=\Lambda_0\otimes\mathbb Z_p \quad\text{for almost every }p. Then Λ⊂N−1Λ0\Lambda\subset N^{-1}\Lambda_0 for one integer NN. Every real positive definite quadratic form has Northcott on Λ\Lambda.

Proof. Injectivity makes Lp=Dp−1(Mp)L_p=D_p^{-1}(M_p) a Zp\mathbb Z_p-lattice: it is open and compact in V⊗QpV\otimes\mathbb Q_p. For each exceptional prime choose bp≥0b_p\ge0 with Lp⊂p−bpΛ0⊗ZpL_p\subset p^{-b_p}\Lambda_0\otimes\mathbb Z_p, and put N=∏pbpN=\prod p^{b_p}. The coordinates of NλN\lambda lie in Zp\mathbb Z_p for every pp. Since Q∩∏pZp=Z\mathbb Q\cap\prod_p\mathbb Z_p=\mathbb Z, they are integers. Apply Theorem 5.2. ◻

For the residual regulator, a common multiple of the projector denominators can first be absorbed into DpD_p. Once scalar injectivity is available, Proposition 4.6 supplies local integral modules. The almost-everywhere saturation condition in Theorem 5.3 is the additional comparison with one common rational lattice.

Example 5.4. Let Λ=∑p mprimeZ 1p⊂Q,q(x)=x2.\Lambda=\sum_{p\ {m prime}}\mathbb Z\,\frac1p\subset\mathbb Q, \qquad q(x)=x^2. Its elements have squarefree denominators. For every prime pp, Λ⊗Zp=p−1Zp\Lambda\otimes\mathbb Z_p=p^{-1}\mathbb Z_p, which is finite free of rank one. Nevertheless q(1/p)=p−2→0q(1/p)=p^{-2}\to0, so Northcott fails. The integral maps Dp(x)=pxD_p(x)=px have inverse lattices larger than Zp\mathbb Z_p at infinitely many primes. Thus local finite generation at every prime does not provide the common denominator required by Theorem 5.2.

Already Z[1/p]\mathbb Z[1/p], with q(x)=x2q(x)=x^2, defeats any finite collection of integral detectors at primes different from pp. Another distinction is real definiteness: the form (a−2b)2(a-\sqrt2b)^2 on Z2\mathbb Z^2 is positive on nonzero rational vectors, but rounding along its real kernel produces infinitely many bounded-height lattice vectors.

5.4. A quantitative descent criterion

Theorem 5.5. Let Λ\Lambda be torsion-free and let q\sqrt q be the norm of a real positive quadratic form on Λ⊗R\Lambda\otimes\mathbb R. Suppose, for some integer m≥2m\ge2, s=∣Λ/mΛ∣<∞,q(λ)≥ε>0(0≠λ∈Λ).s=|\Lambda/m\Lambda|<\infty,\qquad q(\lambda)\ge\varepsilon>0\quad(0\ne\lambda\in\Lambda). Then Λ\Lambda is finitely generated and qq has Northcott. More precisely, if B≥0B\ge0, j∈Z≥0j\in\mathbb Z_{\ge0}, and m2jε>4Bm^{2j}\varepsilon>4B, then #{λ∈Λ:q(λ)≤B}≤sj. \#\{\lambda\in\Lambda:q(\lambda)\le B\}\le s^j.(5.2)

Proof. If x,yx,y are in the height-BB ball and x−y=mjzx-y=m^jz, then ∥z∥≤2B/mj<ε\|z\|\le2\sqrt B/m^j<\sqrt\varepsilon, so z=0z=0. Reduction modulo mjΛm^j\Lambda is injective on that ball. Torsion-freeness identifies the successive quotients miΛ/mi+1Λm^i\Lambda/m^{i+1}\Lambda with Λ/mΛ\Lambda/m\Lambda, proving (5.2).

Choose a finite set TT of representatives modulo mΛm\Lambda, and let C=max⁡t∈T∥t∥C=\max_{t\in T}\|t\|. Writing x=my+tx=my+t gives ∥y∥≤(∥x∥+C)/m\|y\|\le(\|x\|+C)/m. Repetition brings yy into any fixed ball of radius greater than C/(m−1)C/(m-1). This ball is finite by the first part; together with TT, it generates Λ\Lambda. This is the height-descent argument underlying weak Mordell–Weil; compare [2, Section 10.3]. ◻

Northcott by itself need not give finite rational rank. For example, Λ=⨁i≥1Zei\Lambda=\bigoplus_{i\ge1}\mathbb Ze_i with q(∑aiei)=∑2iai2q(\sum a_ie_i)=\sum2^ia_i^2 has finite height balls and infinite rank. Rational detection, integral discreteness and real positivity therefore retain distinct roles in the general program.

6. Fields of classes and bounded-degree finiteness

6.1. Descent by trace

Let G\mathcal G be a rational geometric Chow quotient with its discrete Gal⁡(F‾/F0)\operatorname{Gal}(\overline F/F_0)-action, and let Λ(E)⊂G\Lambda(E)\subset\mathcal G be compatible integral images. We assume that restriction and trace preserve these images in their usual directions. Fixed rational Chow projectors and quotient by geometric algebraic equivalence have this property. Every element considered is defined over a finite extension, so its stabilizer is open. Define its minimal class field by F0(ξ)=F‾Stab⁡Gal⁡(F‾/F0)(ξ).F_0(\xi)=\overline F^{\operatorname{Stab}_{\operatorname{Gal}(\overline F/F_0)}(\xi)}.

Lemma 6.1. If ξ∈Λ(E)\xi\in\Lambda(E) and K=F0(ξ)K=F_0(\xi), then K⊆E,[E:K]ξ∈Λ(K).K\subseteq E,\qquad [E:K]\xi\in\Lambda(K). In particular, if [E:F0]≤D[E:F_0]\le D, the required multiplier is at most DD.

Proof. The inclusion Gal⁡(F‾/E)⊆Stab⁡(ξ)\operatorname{Gal}(\overline F/E)\subseteq\operatorname{Stab}(\xi) gives K⊆EK\subseteq E. Take an integral cycle over EE representing ξ\xi in the indicated quotient and push it forward to KK. Over F‾\overline F, the resulting class is the sum over the KK-embeddings of EE. Every summand equals ξ\xi, so the sum is [E:K]ξ[E:K]\xi. Fixed projectors commute with trace, and algebraic equivalence is preserved by it. ◻

The lemma concerns the field of a class in a rational quotient. It supplies the precise denominator even when no chosen integral cycle representative descends to that field.

6.2. The ramification criterion

Let qq be a Galois-invariant quadratic form on G\mathcal G. For B≥0B\ge0 and an integer D≥1D\ge1, consider B(B,D)={ξ∈⋃[E:F0]≤DΛ(E):q(ξ)≤B}.\mathcal B(B,D)= \{\xi\in\bigcup_{[E:F_0]\le D}\Lambda(E):q(\xi)\le B\}.

Theorem 6.2. Assume qq has fixed-field Northcott on Λ(K)\Lambda(K) for every finite K/F0K/F_0. Then bounded-degree Northcott is equivalent to the following condition: for every B,DB,D, there is a finite set S(B,D)S(B,D) of finite places of F0F_0 such that each field F0(ξ)/F0F_0(\xi)/F_0, with ξ∈B(B,D)\xi\in\mathcal B(B,D), is unramified outside S(B,D)S(B,D).

Proof. Suppose the ramification condition holds. By the finite-ramification form of Hermite–Minkowski, there are only finitely many extensions K/F0K/F_0 in F‾\overline F of degree at most DD unramified outside S(B,D)S(B,D); see [5, Theorem 10.2.2]. This form also follows by bounding the local discriminant exponents in fixed degree and applying discriminant finiteness.

For ξ∈B(B,D)\xi\in\mathcal B(B,D), Lemma 6.1 gives mξ∈Λ(K)m\xi\in\Lambda(K), where K=F0(ξ)K=F_0(\xi) is among these finitely many fields and 1≤m≤D1\le m\le D. Moreover q(mξ)=m2q(ξ)≤D2B.q(m\xi)=m^2q(\xi)\le D^2B. Fixed-field Northcott leaves finitely many possibilities for mξm\xi. Multiplication by mm is injective in the rational group, and only finitely many mm occur. Thus the ball is finite. Conversely, if the ball is finite, the union of the ramification sets of its finitely many class fields is finite. ◻

An explicit discriminant estimate would give a quantitative route to the hypothesis. For example, with dK/F0\mathfrak d_{K/F_0} the relative discriminant ideal, an inequality log⁡NF0/QdF0(ξ)/F0≤ADq(ξ)+BD \log N_{F_0/\mathbb Q}\mathfrak d_{F_0(\xi)/F_0} \le A_D q(\xi)+B_D(6.1) on integral classes of degree at most DD would suffice. The qualitative ramification condition only asks for some finite set depending on B,DB,D; it is weaker than (6.1).

6.3. Why uniform fixed-field estimates do not suffice

Example 6.3. Choose infinitely many distinct quadratic extensions Ki/F0K_i/F_0 with characters χi\chi_i. Put G=⨁iQei,σei=χi(σ)ei,Λ(E)=⨁Ki⊆EZei,q(∑aiei)=∑ai2.\mathcal G=\bigoplus_i\mathbb Qe_i,\qquad \sigma e_i=\chi_i(\sigma)e_i,\qquad \Lambda(E)=\bigoplus_{K_i\subseteq E}\mathbb Ze_i, \qquad q\left(\sum a_ie_i\right)=\sum a_i^2. The lattices are compatible with restriction and trace. Every finite extension EE has at most [E:F0]−1[E:F_0]-1 distinct quadratic subfields: their compositum is multiquadratic, and the number of its quadratic subfields is one less than its degree. Consequently rank⁡Λ(E)≤[E:F0]−1,q(λ)≥1 (λ≠0),∣Λ(E)/mΛ(E)∣≤m[E:F0]−1.\operatorname{rank}\Lambda(E)\le[E:F_0]-1,\qquad q(\lambda)\ge1\ (\lambda\ne0),\qquad |\Lambda(E)/m\Lambda(E)|\le m^{[E:F_0]-1}. All fixed-field height balls are finite. Nevertheless the classes ei∈Λ(Ki)e_i\in\Lambda(K_i) all have degree two and height one. Thus bounded-degree Northcott fails despite uniform degree-dependent bounds on ranks, gaps and Kummer quotients.

The fixed-SS cohomological statement of Proposition 4.3 holds over a chosen field of definition. Theorem 6.2 instead bounds the fields themselves. This distinction is essential when using residual arithmetic regulators to study the conjectural Northcott property.

7. Proof of the principal reduction

Proof of Theorem 1.3. Suppose (ii). Assertion (A) and Theorem 2.1 give arithmetic hard Lefschetz. Assertions (B) and (C), together with Theorem 3.4, give central integral Northcott on every odd subproduct. By Lemma 5.1 the central forms are real positive definite. Corollary 2.4 now gives primitive integral Northcott and positivity on XX, as well as nondegeneracy of the complementary height pairings.

Conversely, assume (i). Arithmetic hard Lefschetz gives off-central vanishing by Theorem 2.1, proving (A). For an odd subproduct ∣S∣=2c−1|S|=2c-1, inflate u∈USc(E)u\in U_S^c(E) by fundamental classes on all omitted factors. In the notation of Subsection 2.1, this is the integral injection u⟼j∅uu\longmapsto j_\varnothing u in degree r=cr=c. With m=n−∣S∣m=n-|S|, it is primitive and Theorem 2.3 gives qℓ,c(j∅u)=m!∏i∉Sai qS,c(u).q_{\ell,c}(j_\varnothing u) =m!\prod_{i\notin S}a_i\,q_{S,c}(u). Thus primitive Northcott on XX gives central integral Northcott on every odd subproduct. By Theorem 3.4, this is equivalent to (B) and (C). The positivity and nondegeneracy assertions then follow from the first direction of the proof. ◻

References

[1] Arithmetic conjectures for Beilinson–Bloch heights on products of curves, companion research programme, 17 November 2026.

[2] B. Conrad, notes by T. Feng, Abelian varieties, Stanford lecture notes, Spring 2015. Author-hosted PDF.

[3] P. Deligne, La conjecture de Weil I, Publications Mathématiques de l’IHÉS 43 (1974), 273–307. Original article.

[4] B. Kahn, On the injectivity and non-injectivity of the ℓ\ell-adic cycle class maps, manuscript dated 10 March 2024. Author-hosted PDF.

[5] K. S. Kedlaya, Weil cohomology in practice, Chapter 10, Comparing Galois representations: the Faltings–Serre method, lecture of 30 October 2019. Author-hosted notes.

[6] J. Laga and A. Shnidman, Vanishing criteria for Ceresa cycles, Compositio Mathematica, published online 29 December 2025. Publisher’s full text.

[7] J. S. Milne, Arithmetic duality theorems, second edition, BookSurge, 2006. Author-hosted book.

[8] W. Raskind, Higher ℓ\ell-adic Abel–Jacobi mappings and filtrations on Chow groups, arXiv:alg-geom/9412001.

[9] F. Scavia and F. Suzuki, Non-injectivity of the cycle class map in continuous ℓ\ell-adic cohomology, arXiv:2208.06713.

[10] S.-W. Zhang, Standard conjectures and height pairings, arXiv:2009.07089v3. The numbering cited here refers to the version of 29 April 2022.

PDF and source

Download the paper

The PDF contains the full article, numbered statements, proofs, and references. The LaTeX download is self-contained, with the included proof sources expanded.

Edition and evidence

Full supplied internal arguments. Height-dependent statements assume the stated system H1–H4. “Internal proof” does not mean external peer review.

Source and PDF fingerprints

19 proof environments retained from the source. This checks document preservation, not mathematical validity.

Expanded LaTeX SHA-256

109418af9ce0c27f54b5b24a089a5c71ca33f2e6f8121b009c6dd9f3778a59ea

PDF SHA-256

5abc8eee5b6972dfd1230ac2d441ec9884dcd35b3374a8de3ab9fb17c59d1c4f

Permanent records and reader corrections

These links retain individual claim identities and revision-bound feedback. They are the reference index for this paper and its associated open problems.