# Marked theta sections, horizontal boundary orders, and unitary period divisibility

## 1. The local comparison problem

The signed height estimate of [1, (6)] concerns one family of integral period maps. A word specifies a chain of $m$ theta lifts and a choice $p=T_u$ of split word prime. Its period map pairs an input of the prescribed weight and marking with a test form in the final unitary row. The estimate compares $\mu(C)+h_u$ with a geometric main term and a signed finite term $F_o$, where $C$ is the dual of the common row test lattice and $h_u$ bounds the mean logarithmic squared operator norm. The finite term uses the same local shifts as the integral columns of [1, (4)]. Two problems in making this estimate arithmetic are the identification of the integral boundary and the divisibility of the actual periods. [1, 2, 7]

For the first problem, Theorems 1 and 2 identify the boundary on the matched integral arc and give a sharp test for what can change under replacement of the model. On a complete coefficient DVR $V$ with parameter $\pi$, the two-dimensional base is $V[[u]]$. The generic marking, integral semiabelian data and matched period weights give the ideal $(u^{H(b)})$, where $b$ is the marked period form and $H$ is the certified boundary order. An explicit vertical blowup preserves the generic family and changes integral orders. The two torsion grids, the comparison of the odd projective metric with the Haar norm, and the level-six recovery theorem identify the corresponding scalar of the integral section. This supplies the boundary and theta contributions to the point height in the same models, with a separate comparison required for any other integral extension. [3–7]

For the second problem, Corollary 13c evaluates the period in its fitted theta and row frames. At each edge the joint character acts on the simultaneous moments of the marked kernel, with a common measure for all source dilations of each base term. The matched intermediate point and coset preserve that character through the chain. The order bound $b_{\chi,w}$ at one edge becomes at least $m b_{\chi,w}$ in the shifted period. This is additional divisibility of its entries. There are $(p-2)^f$ retained joint characters, all of equal rank. On the coefficient places seeing the selected factor of $p$, the local degrees sum to half the degree over $p$. The determinant comparison in (28) and (29) therefore gives an integral enlargement of the same rank with normalized slope gain at least $(mf/2)\log p$. For a translate whose words pass the common spectral row tests and Gram bounds, Corollary 13e uses the full product of the $f$ character digits, mixed over words in proportion to their ranks. The bottom quantile of this distribution gives the saving in (30a), with (30) as a bound using just the one-digit estimate. For a retained rank fraction of at least one half and a combined Gram error of at most one quarter, (30c) saves at least $(mf/8)\min_u\log T_u\,(1-\sqrt{11/(3f)})_+$ in the scaled row gap. The signed row-height estimate, the reference covolume comparison and the choice of a uniform compatible family still have to hold for these maps. [1, 23–28]

## 2. Matching the horizontal boundary

Let $V$ be a complete coefficient DVR of mixed characteristic, with uniformizer $\pi$. Put $R=V[[u]]$ and $S=\operatorname{Spec}R$. Work with a fixed marked arc in the selected component, with trivial auxiliary level and full principal level $M$ (the deep level of the component). Its actual smooth separated semiabelian family $G/S$ is abelian on $D(u)$. We use the following data of the arc in [7].

1. The characteristic-zero map to the principal-level PEL moduli scheme is given by a polarized isomorphism on the generic point. It identifies the full order $\mathcal O_E$ of the CM field $E$, its conjugation adjoint, the PEL determinant type, the selected cyclotomic coordinate and the labelled level. The datum has a self-dual lattice $L$ and signed alternating form $B$ in the release. Take Lan’s form to be $-(2\pi i)B$, with cyclotomic coordinate sending $2\pi i$ modulo $M$ to the specified positive root $\zeta_M$. Thus the ordered Weil value is $\zeta_M^{-B(x,y)}$. For a split half-level ideal $\mathfrak b$ with $\mathfrak b\bar{\mathfrak b}=(M)$ and $\mathfrak b+\bar{\mathfrak b}=\mathcal O_E$, the marked summand is $\bar{\mathfrak b}L/ML$. The two summands are isotropic by adjointness and perfectly cross-paired by self-duality. The displayed Weil value defines the complementary labels uniquely. We require the principal marking to be symplectically liftable and the prime-to-residue-characteristic level neat, as in [3]. The level labels are generic when the residue characteristic divides $M$.
2. The given family includes the semiabelian extension of its generic dual and the trace-one rigidified cubical line. The polarized formal identification along $u=0$ gives the Raynaud extension, its constant abelian part $B_0$ with principal line $L_0$, and the common torus-character and period lattice $\Lambda$ under the principal polarization. The maps to the abelian quotient are specified by $c_0:\Lambda\to B_0$ and the polarization. The actual Fourier coefficients use the indicated integral weight lines. The marked filtration comes from the torus and Raynaud $M$-torsion in the formal extension, with their compatible maps into $G[M]$ on all thickenings and with the specified generic labels. The monomial arc has symmetric positive period form $b$ on the degenerating lattice, with integral $Q(y)=b(y,y)/2$ and well-defined divided exponents $b(y,z)/M$ in the actual level lattice. Its inverse-line period section is $\psi_u(y)=u^{Q(y)}\psi_0(y)$. The constant evaluations are units in their rigidified integral lines.
3. The effective Cartier divisor $D_{\rm rel}$ on the release model comes with its integral order certificate. Its ideal on $D(u)$ is the unit ideal. On the specialization $u\mapsto\pi$ that certificate gives order $H(b)$, where $H$ is the final integral boundary function used by the certificate, with its scaling included. On the chosen Lan chart, require the same weights in the full-level character lattice $P$, represented on the cone $\sigma$ of the arc by a Cartier character $m_\sigma\in P$. More explicitly, $H$ gives $\langle m_\sigma,n_\rho\rangle$ on each primitive cocharacter $n_\rho$ of a ray of $\sigma$. The chart has its specified coefficient map and fan. We use Lan’s horizontal boundary $D_H$ with these weights.

**Theorem 1.** In the identity datum of [3, Example 2.3], the map of the marked generic point extends uniquely to the specified Lan model. The pulled-back polarized semiabelian family and trace-one rigidified cubical line are identified with those of the arc. On $S$ the two boundary ideals agree:
$$
 \mathcal I_{D_{\rm rel}}R=\mathcal I_{D_H}R=(u^{H(b)}).
 \tag{1}
$$
For a map $R\to W$ to a DVR on which $\pi u$ is nonzero, their common order in $W$ is $v_W(u)H(b)$, with integer-normalized valuation. This includes maps with a kernel. If the comparison uses a compatible refinement or a Cartier multiple, the hypotheses apply to the resulting chart and the corresponding multiple of both weighted divisors. [3, 4, 7a]

**Proof.** We use the setting of [3, Theorems 7.14 and 11.4]. The base is irreducible, normal and Noetherian over Lan’s $\operatorname{Spec}\mathcal O_{F_0,(p)}$, where $F_0$ is the chosen extension of the reflex field and $p$ is the residue characteristic. For each index $j$ in Lan’s collection $J$, Theorem 11.4 requires a degenerating family of the corresponding PEL type over the base, with its semiabelian group, degenerating polarization to the connected dual, action and generic $H_j$-level. Its generic map must be the composite with the specified isogeny-level map (2.1) of [3]. The identity datum has one index in $J$, the same lattice and form, and identity adelic element. The generic polarization and its inverse extend to the supplied group and its connected dual by the homomorphism extension theorem over a normal base [4, Proposition 3.3.1.5]. The adjoint and inverse identities extend by uniqueness. In Theorem 11.4 the auxiliary families of Theorem 7.14 are induced from the supplied indexed families by Proposition 5.24. For the cone test, at a fixed geometric centre all dominant complete DVR tests with algebraically closed residue field must give forms in a common cone in the corresponding marked cusp data. A finite extension of a test is permitted for semistable reduction.

We verify the form and the identification of the cusp data together. Write $P_{B_0}$ for the Poincaré biextension with the abelian variety in the first variable. Orient the period evaluations as functionals on the **positive** Poincaré line. In that orientation the second difference of $Q$ gives
$$
 \tau_u(y,z)=u^{b(y,z)}\tau_0(y,z),\qquad
 b(y,z)=Q(y+z)-Q(y)-Q(z).
 \tag{2}
$$
To see that this is the period of the actual family, the weight-$z$ coefficient of a section is in $L_0$ tensored with the rigidified second difference line at $c_0(z)$. In these lines both Fourier recurrences compare weights $z$ and $z+y$ by translation by $c_0(y)$ and multiplication by $\psi_u(y)\tau_u(y,z)$. This is Lan’s recurrence [4, (4.3.1.2)–(4.3.1.7)] and the release recurrence [7a]. The coefficient maps on the generic principal line are nonzero. Comparing at zero identifies $\psi_u(y)$, then comparing at arbitrary $z$ identifies $\tau_u(y,z)$. A section scalar cancels in both comparisons. Taking the fractional ideal of the evaluation on the integral line gives $u^{b(y,z)}$ times a unit ideal.

For a test centred on $u=0$, the formal identifications pull back to its thickenings, since $u$ lies in the maximal ideal of the test ring. They transport the same Raynaud lattices, and (2) gives order $v_W(u)b(y,z)$. The torsion filtration is obtained by pulling back the torus and Raynaud torsion subgroups through the **same** generic marking. Its maps are compatible with these tests: the formal maps from the finite flat torsion sources algebraize by proper formal existence applied to their graphs in a proper compactification of the target. The inverse image of the complement misses the closed fibre of each finite source, hence is empty. Their generic images are the Raynaud filtration, including the abelian torsion quotient and the period quotient. Thus the graded identifications give a common cusp label at the centre. On the lattice $P$ the undivided characters have finite-index image and $M[(y/M)\otimes z]=[y\otimes z]$. The cocharacter $n_b$ is consequently given by $b(y,z)$ on the latter and $b(y,z)/M$ on the former. This also follows by taking the $M$-th power of the divided evaluation in its integral multiplication line. The ordered pairing fixes the marking and the finite-index calculation fixes the valuation. It works in the normalized chart at residue primes dividing $M$.

Every form at this centre is $v_W(u)n_b$ in the marked lattice. Choose the closed cone containing the ray of $n_b$. A finite extension only multiplies the form by the ramification index. At a centre on $D(u)$, the cone is zero by good reduction of the test family. This proves the simultaneous cone condition. Lan’s theorem supplies the map, and separatedness gives uniqueness. The map factors through the closure of the specified generic component. If that coefficient model is further finitely normalized, the prescribed generic lift extends by normality of $R$, since all images of the finite algebra are integral over $R$ in its fraction field. On the pullback the group, action and polarization agree by homomorphism extension and uniqueness. If $w_0\in\mathcal O_E$ is the specified element with $w_0+\bar w_0=1$, the line is the rigidified pullback $(1,\lambda w_0)^*P_G$ of the biextension for the group and its connected dual. This construction gives the cubical comparison. A generic isogeny with polarization multiplier would require comparing different tensor powers. [3, Theorems 7.14, 11.2, 11.4] [4, §§3.3–3.4, 6.2]

On a normalized toroidal chart the equation of the weighted Cartier boundary in an integral weight frame is the character of $m_\sigma$. Nonnegative ray weights put that character in the dual monoid. On a common face, a difference of Cartier characters with the same ray values is invertible. By [3, Proposition 8.14 and (8.15)], the chart has invertible weight sheaves and the torus is dense fibrewise over the normalized cusp base. Hence the divisor of this character has the prescribed horizontal coefficients and has no coefficient at a vertical prime, whose generic point meets that torus. One can also descend the Cartier equation from the formal chart of [3, Theorem 10.13]. The divisorial ideal of the horizontal closure extends under the regular faithfully flat completion of the excellent local ring. At height one its coefficients are pulled back along the branches with multiplicity one, and the height-one primes over the generic point have coefficient zero. Thus the extended ideal equals the character ideal by reflexivity, and invertibility descends.

The character has order $\langle m_\sigma,n_b\rangle=H(b)$ at $(u)$. Its pullback is invertible on $D(u)$, where the arc is abelian. The same support condition holds for the released divisor. Each pulled-back Cartier equation is therefore a unit times a power of $u$ in the factorial ring $R$. For the release equation the specialization $u\mapsto\pi$ computes this power from input 3. Formula (1) follows, and taking image ideals under any of the stated maps to $W$ proves the valuation assertion. A certificate for a theta ideal only on the characteristic-zero compactification enters this proof through its matched weights and horizontal Cartier extension. The integral specialization on the release side uses the integral certificate in input 3. [3, §8 and Theorem 10.13] [7a]

## 3. Three valuations and a vertical discrepancy

**Theorem 2.** Let $J$ be a finitely generated fractional ideal of $R$ which is the unit ideal over $R[1/(\pi u)]$. The frame is part of the data. Let $a,c$ be its orders at $(u),(\pi)$, respectively, and let $v_*$ be the exceptional valuation of the ordinary blowup of the closed point of $S$. Write
$$
 J=u^a\pi^c I.
$$
Then $I=R$ or $I$ is $(\pi,u)$-primary. Let $\lambda$ be the least nonnegative integer with $(\pi,u)^\lambda\subseteq I$. For any DVR test as in Theorem 1, put $m=v_W(u)$ and $r=v_W(\pi)$. The discrepancy is
$$
 v_W(J_W)-mH(b)
   =m(a-H(b))+rc+v_W(I_W),
 \qquad 0\le v_W(I_W)\le\lambda\min(r,m).
 \tag{3}
$$
The last bound is sharp given $\lambda,r,m$. Equality $J=(u^{H(b)})$ holds exactly when
$$
 a=H(b),\qquad c=0,\qquad v_*(J)=H(b).
 \tag{4}
$$
For a Cartier ideal the first two tests suffice. If an effective Cartier boundary is supported on the nonabelian locus of the identified family on the arc, the second condition holds automatically.

**Proof.** All height-one orders away from $(\pi)$ and $(u)$ are zero. Dividing out the two indicated factors leaves an ideal inside $R$ by the height-one criterion for normality. Its support has codimension at least two. If it is proper it is therefore primary for the maximal ideal. The image of that maximal ideal on a test has order $\min(r,m)$, proving (3) and ensuring a nonzero evaluated ideal even if the map has a kernel. The ideal $(\pi,u)^\lambda$ attains the bound. The associated graded ring of $R$ at the closed point is the polynomial ring in the two initial parameters. Its degree valuation is the exceptional valuation $v_*$, with $v_*(u)=v_*(\pi)=1$. Thus $v_*(I)=0$ exactly for $I=R$ and $v_*(J)=a+c+v_*(I)$, proving (4). A principal residual ideal must be the unit ideal since it has no height-one support. Finally $(\pi)$ has its generic point in $D(u)$, so a Cartier equation supported on the nonabelian locus is a unit there. $\square$

For coordinates $(t_i)$ in a specified rational frame, their image ideal is $(t_i)$ and its order on a test is $\min_i v_W(t_i)$. A common scalar changes that minimum by its valuation and leaves the projective ratios unchanged. On a normal integral graph of two models, equality of pulled-back Cartier boundaries and their canonical sections is equivalent to equality of the generic divisors and all vertical height-one coefficients. This follows from the injection of Cartier into Weil divisors on a normal scheme. For two rational theta readings in an identified line, the corresponding statement for equality **up to a regular unit** requires the ratio to have zero order at every height-one point. Normality then makes it and its inverse regular. Exact equality of the prescribed readings requires this ratio to be $1$ in the specified identification. The two comparisons use different ideals.

For example, a numerator homogeneous of degree $n$ in the same theta section and divided by a $j$-th power of the boundary equation has valuation change $n v_W(q)-j(v_W(J_W)-mH(b))$ if the theta is rescaled by $q$ and the boundary is replaced by $J$. Here $J_W$ is principal, and the comparison is where both readings are defined and nonzero. The sign on the discrepancy follows from the denominator. Where the fitted rule uses a finite sum of products of $n$ fibre derivatives, homogeneity holds by base-linearity of translation and fibre differentiation [7d].

**Proposition 3.** Suppose the arc maps into a normal projective model $X$ over an excellent DVR with uniformizer $\varpi_0$, and an effective Cartier divisor $D$ on $X$ pulls back to $(u^{d_b})$ with $d_b=H(b)>0$. Fix positive integers $A,B$. Normalize the blowup of
$$
 \mathcal J=(\varpi_0^A)+\mathcal I_D^B ,
 \qquad f:X'\longrightarrow X.
$$
Write $\mathcal J\mathcal O_{X'}=\mathcal O(-F_{\rm exc})$. Then $D_N=f^*D+N F_{\rm exc}$, for $N\ge1$, has the same support as $f^*D$ and the same generic divisor. On every DVR test in Theorem 1 its order on the lifted test is
$$
 m d_b+N\min(A r_0,Bm d_b),\qquad r_0=v_W(\varpi_0).
 \tag{5}
$$
In particular the discrepancy on the fixed test $u\mapsto\pi$ is unbounded in $N$. The original surface arc has no lift to $X'$.

**Proof.** The normalization is finite by excellence, and the resulting projective modification is an isomorphism away from the intersection of the special fibre and $|D|$. The image of $\mathcal J$ is invertible, and the support of $F_{\rm exc}$ lies in $f^{-1}|D|$. A DVR test principalizes the nonzero image ideal, so it lifts by the universal property of the blowup and by normality with its prescribed generic lift. Its exceptional order is the smaller of the two orders of the generators of $\mathcal J_W$, giving (5). The generic model, full marking and rigidified line are unchanged, and the original semiabelian family and primitive section line can be pulled back. At $u=\pi$, both orders in the minimum are strictly positive. But on the surface the image ideal is $(\pi^{A v_V(\varpi_0)},u^{Bd_b})$, a proper ideal of height two. A nonzero invertible ideal is principal on $S$ and cannot have height two. A lift to $X'$ would give a lift to the blowup, which would make this ideal invertible. $\square$

This modification compares normal projective models with the same generic marked object. Retaining Lan’s valuative property or the release’s prescribed boundary certificate is an additional condition on such a model.

## 4. Two torsion grids on a good trait

Let $R_0$ be a DVR with fraction field $K$, and let $A_0/R_0$ be an abelian scheme of dimension $g$ with a symmetric, rigidified, relatively ample principal line $\mathcal L$. By Riemann–Roch, ample-line cohomology vanishing and base change, $H^0(A_0,\mathcal L)=R_0h$ with $h$ reducing to a nonzero section. Fix a nonzero rational section $s=a h$ in this **identified** line. Thus $(a)$ is the scalar ideal, independent of the integral generator $h$.

Let $m\ge2$ be invertible in $R_0$, and set $d=m^2$. Symmetry and the theorem of the cube give the isomorphism $j_m:[m]^*\mathcal L\simeq\mathcal L^{\otimes d}$ normalized by the rigidification at zero. For $P\in A_0[m](K)$ the powered cubical value is
$$
 T_m(s,P)=j_m^{-1}\bigl(s(P)^{\otimes d}\bigr)
           \quad\text{read in the rigidification at zero}.
$$
Explicitly, in an integral generator $b_P$ of $P^*\mathcal L$, let $j_m(1)=u_P b_P^{\otimes d}$ with $u_P$ a unit. Then $T_m(s,P)=u_P^{-1}(s(P)/b_P)^d$. This is the normalization by the value of the multiplication frame at the identity in the released height construction. For each $x\in A_0[d](K)$ also choose a finite-order theta-group lift $t_x^*\mathcal L^{\otimes d}\simeq\mathcal L^{\otimes d}$, after a finite extension if necessary. Indeed, the iterate by the order of $x$ of any lift acts by a scalar, and a root of its inverse rescales the lift to finite order. Use the prescribed symmetric lift when comparing an individual released coordinate. Let $c_x(s)$ be the value at zero, in the $d$-th tensor of the rigidification, of the translated section $j_m([m]^*s)$ using that lift. These are the linear coordinates used before symmetrization in the projective theta construction. [5, §2.3] [6] [7b]

**Theorem 4.** On a split good-reduction trait the two complete grids on $s$ have ideals
$$
 (T_m(s,P):P\in A_0[m])=(a)^d,\qquad
 (c_x(s):x\in A_0[d])=(a).
 \tag{6}
$$
If $B=d^{2g}$ and $e_i$ is the $i$-th elementary symmetric function in the $B$ coordinates $c_x(s)$, then
$$
 \bigl(e_i((c_x(s))_x)^{\,B!/i}:1\le i\le B\bigr)=(a)^{B!}.
 \tag{7}
$$
These equalities use the same section. The first equality descends as a content ideal over the original trait. Integer-normalized orders on a test extension are multiplied by its ramification index.

**Proof.** Work first on the geometric special fibre and use the same symbols for the fibre objects. The map $[m]$ is a torsor under the constant group $A_0[m]$. Since its order is invertible, the deck characters split $[m]_*\mathcal O$ into rank-one eigensheaves $P_\chi$. On a trivialized torsor this is the decomposition of the regular representation. The product of the eigensheaves follows character multiplication and their pullback is trivial. Each is $m$-torsion, hence lies in $\operatorname{Pic}^0(A_0)$ by torsion-freeness of the Néron–Severi group. Two characters cannot give the same line: an isomorphism of the descents would give an equivariant automorphism of the trivial line upstairs, whose global multiplier is a constant. The number of eigensheaves is $m^{2g}$, so they exhaust $\operatorname{Pic}^0(A_0)[m]$. The projection formula gives the summands $H^0(A_0,\mathcal L\otimes P_\chi)$ of $H^0(A_0,[m]^*\mathcal L)$. Each has dimension one, because the principal polarization represents each twist by a translate of $\mathcal L$.

The section $[m]^*\bar h$ generates the trivial summand. Translation by $x\in A_0[d]$ lifts to $[m]^*\mathcal L$ because the polarization kernel of $\mathcal L^{\otimes d}$ is $A_0[d]$. The translated section has the character of the descent twist $t_{[m]x}^*\mathcal L\otimes\mathcal L^{-1}$, up to inversion if one uses the inverse convention for deck characters. Indeed, before identifying the translated pullback with the original pullback it descends to $t_{[m]x}^*\mathcal L$, and the identification trivializes the pullback of the twist with its given descent character. As $x$ varies, $[m]x$ exhausts $A_0[m]$. The translates thus span the full space. The line $\mathcal L^{\otimes d}$ is base-point-free. For a point $z$, choose $d$ shifts summing to zero whose translated theta divisors miss $z$. The conditions on the first $d-1$ shifts and their negative sum are nonempty opens in an irreducible product, and the theorem of the square identifies the product of the translated lines with $\mathcal L^{\otimes d}$. In dimension zero use the rigidification. Evaluating the spanning translates at zero shows $\bar h(P)\ne0$ for some $P\in A_0[m]$.

Over the split trait this means one $h(P)$ generates its integral fibre. All other values of $h$ are integral. For the linear coordinates, finite-order generic theta lifts act integrally. An isomorphism over the generic fibre has divisor supported on the single integral special fibre of the regular group scheme, so a uniformizer power makes it an integral isomorphism. An iterate of this extended lift over the identity acts by a base unit, and the corresponding iterate of the original finite-order lift acts by $1$. Thus the correcting power is zero. In particular each $c_x(h)$ differs in its integral value frame from the evaluation at $[m]x$ by a unit. The point $P$ above occurs among the $[m]x$. Scaling the two sets of evaluations by $a^d$ and $a$, respectively, proves (6). After division of the raw coordinates by $a$, some reduced entry is nonzero, so the monic polynomial with these reduced entries as roots differs from $X^B$. Some elementary symmetric coefficient is a unit. All normalized coordinates of (7) have degree $B!$ in the raw coordinates, proving (7).

For descent of the powered value let $C$ be the finite free algebra of $A_0[m]$. The cubical reading gives an element of $C\otimes_{R_0}K$. Its content is the image of the evaluation map from $C^\vee=\operatorname{Hom}_{R_0}(C,R_0)$ into $K$. This commutes with flat base change and gives the entry ideal over a splitting trait, so faithful flatness proves the assertion. $\square$ [5, Lemma 2.4 and Proposition 2.5] [6] [7b]

A unit change of frame leaves (6)–(7) unchanged and can change the values. Identification of the section, the point labels and the rigidifications is required for comparing individual readings. For a fixed good place the projective and height grids can also have different orders. After a common splitting extension apply the theorem at each order to the same section and divide the two minima by their respective homogeneity degrees. Both recover $v(a)$. In [7b] the height order can be chosen sufficiently large and prime to any fixed finite set of residue characteristics, since the sampling hypothesis is eventual. The projective construction in [7e] uses a power of $3$, so its tame test applies at good places of residue characteristic different from $3$. This includes good places dividing the moduli level whenever the generic coordinates and integral line are identified there.

The full geometric $m$-grid of powered values also has the minimum in (6) for $m=p^a n$ with $n\ge2$ prime to the residue characteristic $p$, since the tame subgrid supplies a unit and $j_m$ remains integral. For pure $p$-power order the geometric values can all reduce to the zero locus of the section, for example for the origin divisor in supersingular elliptic reduction.

### 4.1. The odd theta map on the integral model

Use the matched identity family of §2 with the coefficient embedding into $F$. The arithmetic theta compactification in [7e] is over $o=\mathcal O_F[1/(3M d_H\,\operatorname{disc}E)]$, where $d_H$ is the host prime and $M$ is the principal level. Write $f:G\to S_0$ for the semiabelian family on its matched normal Lan model over this base and $\mathcal T=f_*\mathcal L$ for the primitive theta line. Here $\mathcal L$ has the specified origin rigidification and equals the pullback of the Poincaré biextension under $x\mapsto (x,\lambda w x)$, where $\lambda$ is the principal polarization and $w+\bar w=1$ in the integral CM order. Take the release level $m=81\cdot3^j>81^g g!$ for an abelian fibre of dimension $g$. Set $N=m^{4g}$, the number of labels in $A[m^2]$, and $B=N!$.

**Theorem 4a.** Let $\Phi$ be the map given generically by the symmetric coordinates in (7) with these $N$ labels. Its coordinates generate $(\mathcal T^\vee)^{\otimes B}$ at every point of $S_0$, and
$$
                 \Phi^*\mathcal O(1)\simeq(\mathcal T^\vee)^{\otimes B}.
                 \tag{7a}
$$
On a complete trait with abelian generic fibre a generator of $\mathcal T$ generates the stable lattice after passage to a split model. If $s=a h$ in this line, the raw coordinates have content $(a)$ over the splitting trait, the powered values have content $(a)^{m^2}$, and the descended symmetric coordinates have content $(a)^B$.

**Proof.** On a split trait let $Y$ be the period lattice, $\phi$ its identification with the character lattice and $t(y,\phi z)$ the period pairing in its integral Poincaré value line. Its valuation $b(y,z)$ is symmetric and satisfies the Rosati adjoint identity. In the descent of the Poincaré bundle, translating the two arguments separately by periods $y,x$ and then composing the translations contributes $t(y,x)$ at the identity. Pulling back along the graph of $\lambda w$ therefore gives the period factor $t(y,\phi wy)$. Its order is
$$
 b(y,wy)=b(y,\bar w y)=\tfrac12 b(y,y).
$$
The first equality uses symmetry and adjunction, and the sum is $b(y,y)$ by trace one. This proves that the tropical half-characteristic is zero, also in residue characteristic two. The equality takes place in the value line of the abelian quotient, so it applies to partial toric degeneration. [3, 7n, 9, 15]

The primitive weight-zero coefficient is a section of a principal line on the good abelian quotient. On each split completed chart its integral coefficient line is free of rank one, by the abelian base-change theorem and Riemann–Roch. The Fourier recursion translates this coefficient and multiplies it at weight $y$ by the period factor just computed. Its positive orders for $y\ne0$ give convergence and specialization to the weight-zero section. This identifies the primitive generator $h$ on the connected model in these charts. Equivalently the local direct image is generated by a section nonzero on the closed fibre, and a rational multiple of it extends integrally to a trait exactly when the multiplier is integral. On a component of tropical label $t_0$ the canonical order of every evaluation is at least $\min_{y\in Y} b(y+t_0,y+t_0)/2$ by the Fourier expansion with its cubical metric as in §5.5. This lower bound has minimum zero at the identity component. Its weight-zero coefficient survives there. Hence the connected and stable generators agree. [3, 7n, 9]

By the translate argument of Theorem 4 the reduced coefficient is nonzero at some $m$-torsion point on the abelian quotient. Lift that point to $m$-torsion of the semiabelian special fibre by torus division and then to the trait by étaleness, allowing a field extension. This supplies a unit value of $h$ and the preceding lower bound gives integrality of all the values, including those at generic torsion points outside the connected model. In the marked cubical frame the raw coordinate $c_x$ satisfies $c_x^{m^2}=T_m(h,[m]x)$. For descent use the actual symmetric theta lift: it has order dividing $m^2$ and inversion carries it to its inverse. Two such lifts differ by a scalar whose square and $m^2$-th power are both one, hence coincide. Galois therefore permutes the actual labelled values. Their elementary symmetric functions descend with their phases, and the monic-polynomial argument in Theorem 4 gives a unit symmetric coordinate. Scaling $h$ gives the three content formulas. [7b, 7n]

Near a point of the normal model, choose a generator $h$ of $\mathcal T$. The rational symmetric coordinates lie in the local ring by the trait calculation at its height-one primes. If they all vanished at the point, take a complete DVR dominating the local ring with its fraction field embedded in the test field. For a positive-dimensional excellent normal local domain such a DVR is obtained by completing at the generic point of a component over the closed point on the normalized blowup of the maximal ideal. Its generic fibre still maps to the abelian generic point of the model. All the coordinates would have positive order on this trait, contrary to the unit just found. Changing $h$ to $v h$ multiplies every coordinate by $v^B$. They consequently give the claimed generating sections of the dual power, and the projective quotient gives (7a). $\square$

**The metric of the odd map.** On an abelian point of the marked family use a nonzero section $s$ of its rigidified principal line, the actual linear constants $z_x=c_x(s)$ for $x\in A[m^2]$, and $U(s)=(e_i(z)^{B/i})_{1\le i\le N}$. The torsion frame has $m^2$-th tensor power $j_m([m]x)(\rho)$, with $j_m:[m]^*L\simeq L^{m^2}$ the rigidified isometry. Thus $|z_x|=\|s([m]x)\|$ in the canonical metric at each place, including for torsion outside the connected model. The unique odd symmetric lifts in the proof of Theorem 4a are used in $e_i(z)$. Put $V_s=\max_i |e_i(z)|^{1/i}$ and, at infinity, $J_s=(N^{-1}\sum_x|z_x|^2)^{1/2}$. A root of $\prod_x(T-z_x)$ has modulus at most $2V_s$, since for $|T|>2V_s$ the sum of the absolute values of the other terms divided by $|T|^N$ is less than one. Also $|e_i|\le(\sum_x|z_x|)^i/i!$ and $\sum_x|z_x|\le NJ_s$. Hence $J_s/2\le V_s\le NJ_s$. At a finite place the ultrametric expansion gives $V_s\le\max_x|z_x|$, and the leading monomial strictly dominates the rest at any larger argument than $V_s$, giving equality. In the dual section frame the projective norm taken to the $1/B$ power is $V_s^{-1}$, while the dual Haar norm is $b_s^{-1}$. [7b, 7s]

Write $b_s=\|s\|_2$ for the Haar probability norm. Each point of $A[m]$ occurs $m^{2g}$ times in the list $([m]x)_{x\in A[m^2]}$, so $J_s^2=b_s^2G_m(0)$ for the normalized squared norm average of §5.7. At one embedding let $\tau=X+iY$ and let $\alpha,\delta\in\frac12\mathbb Z^g/\mathbb Z^g$ be the characteristic in this period frame. The squared point norm in the classical frame is proportional to $e^{-2\pi y^{\mathsf T}Yy}|\theta[\alpha,\delta](x+\tau y)|^2$. The proportionality cancels on division by $b_s^2$. For the Fourier coefficient at $(p,q)$, integration in $x$ fixes the difference $p$ of the two theta indices. Unfolding the $y$-integral with their midpoint as Gaussian variable gives the transform at $q-Xp$, and the midpoint contributes $e^{2\pi i q^{\mathsf T}(\alpha-p/2)}$. Keeping frequencies in $m\mathbb Z^{2g}$ then gives
$$
 G_m(0)=\sum_{p,q\in\mathbb Z^g}
 e^{-\pi m^2\|q-\tau p\|_{\rm pol}^2/2}
 (-1)^{p^{\mathsf T}q}e^{2\pi i(p^{\mathsf T}\delta+q^{\mathsf T}\alpha)} .
 \tag{7c}
$$
Here $m$ is odd. The half-characteristic survives the averaging, and the mixed sign remains even at zero characteristic. Absolute convergence follows from positive definiteness. Write $r_\sigma$ for the shortest nonzero period length for the norm $\|z\|_{\rm pol}^2=\bar z^{\mathsf T}Y^{-1}z$. The Gaussian sum $Z_m$ obtained by omitting the phases bounds $G_m(0)$ from above, with $\frac12\log Z_m\le4g^2/(m^2r_\sigma^2)$ by the packing argument in Corollary 10b. [7s, 14, 19]

For a lower bound put $|v|_Y^2=v^{\mathsf T}Yv$ and $H_m(d)=\sqrt{\det(2Y)}\,m^{-g}\sum_{r\in\mathbb Z^g}e^{-2\pi |d+r/m|_Y^2}$. In the direct theta sum the real mesh keeps differences $mk$ and the imaginary mesh gives midpoint cosets $\alpha+\mathbb Z^g/m-mk/2$. The diagonal is $H_m(\alpha)$. Translating a midpoint coset by either $\pm k/(2m)$ gives the diagonal coset since $m$ is odd. Averaging these two translates inserts $e^{-\pi |k|_Y^2/(2m^2)}\cosh(2\pi v^{\mathsf T}Yk/m)$ into the Gaussian sum over the midpoint $v$. The hyperbolic cosine is at least one. In the original sum the index difference contributed $e^{-\pi m^2|k|_Y^2/2}$. The total off-diagonal absolute error is therefore at most $H_m(\alpha)E_m$, with $E_m=\sum_{k\ne0} e^{-\pi(m^2-m^{-2})|k|_Y^2/2}$. In a Siegel reduced frame the shortest squared $Y$-length is $Y_{11}\ge\sqrt3/2$. The bound follows from Minkowski reduction, $|X_{11}|\le1/2$, and $|\tau_{11}|\ge1$ by inversion in the first symplectic pair. Packing in shells with radii doubling from the shortest length gives at most $5^{g(j+1)}$ points in shell $j$. Its Gaussian mass is at most $e^{-4^jm^2/2}$ since $m^2\ge4g$. Thus $E_m\le\epsilon_m=2e^{-m^2/2}<1$. [11, 19]

By Poisson summation $H_m(0)\ge1$. Pairing $r,-r$ after replacing $m\alpha$ by a shortest representative modulo $\mathbb Z^g$ gives $H_m(\alpha)\ge H_m(0)e^{-2\pi d_Y(\alpha)^2/m^2}$, where $d_Y$ is distance to $\mathbb Z^g$ since $m\alpha\equiv\alpha$. Projection from the polarization lattice onto the $y$-coordinates contracts distances, so $d_Y(\alpha)$ is at most the polarization covering radius, bounded by $g/r_\sigma$ as in Corollary 10b. In particular $\frac12\log G_m(0)\ge\frac12\log(1-\epsilon_m)-\pi g^2/(m^2r_\sigma^2)$. [16, 19]

For the global comparison let $h_v$ generate the stable Moret–Bailly section line after a finite extension marking $A[m]$ and write $s=a_v h_v$. In integer valuation units put $d_v=\min_{P\in A[m]}\nu_v(h_v(P))$ for the canonical cubical order, and $\Delta_m=\sum_v\ell_v d_v$. The stable model gives $d_v\ge0$. Theorem 4a gives $d_v=0$ at the matched places over $o$. Outside that base the same vanishing requires the tame centered comparison, or another unit test. For example at a good supersingular elliptic place over $3$ all $3$-power torsion specializes to zero, so for the origin divisor the integral generator has positive minimum order. The finite projective term divided by $B$ is $-\ell_v(v(a_v)+d_v)$. Using the degree identity of Proposition 8 gives, for the absolute projective height $h_\infty$ with coordinate maximum,
$$
 B^{-1}h_\infty(\Phi(A))=F_A-\Delta_m+
   \operatorname{av}_\sigma\bigl(\tfrac12\log G_{m,\sigma}(0)+\log(V_{s,\sigma}/J_{s,\sigma})\bigr).
 \tag{7d}
$$
The log ratio is in $[-\log 2,\log N]$. At infinity this bounds $\log(V_s/b_s)$ between $\tfrac12\log G_m(0)-\log 2$ and $\tfrac12\log G_m(0)+\log N$. A Euclidean archimedean norm adds between $0$ and $(\log N)/(2B)$. All terms in the height comparison are invariant under base extension with normalized weights. [5, 7s, 19]

The release identifies the $8B$-th power of the projective line algebraically with $(\det\omega)^{4B^2}$. The metric transported from projective space thus has Hodge height $2B^{-1}h_\infty$, while the Deligne metric gives $h_F(A)$. Section 4.2 computes the separate Fourier-ratio ideal of the blowup. The release's cusp intersection height uses finite chart maxima and an archimedean chart correction with abelian and toric Gram factors. Substitution of (7d) there requires a comparison of the chart lattices and metrics. The Haar norm in (7d) is the abelian norm used directly in the point-slope estimate of §6. [7s, 19]

### 4.2. The arithmetic boundary ideal before blowup

Let $Z$ be the finite theta normalization of projective space in [7e]. Fix the arithmetic ideal $J_0$ and the positive integer $d_J$ that [7o] constructs on $Z$ for the chosen support function $\varphi$. Its input to the generic blowup is the restriction of this pair to $F$. The property used here is its **arithmetic** order formula. On a complete characteristic-zero DVR mapping to $Z$, with compatible marked abelian generic point, semiabelian family and monomial period form $b'$ in uniformizer units, the extended ideal has order $d_J\varphi(b')$. The construction obtains this formula by taking the minimum of orders of finitely many products of Fourier ratios on a theta cover, descending a common power of the ideal, and dividing the measured orders by the ramification index. The multiplier $d_J$ includes this descent. [7o]

**Theorem 4b.** On the marked arc $R=V[[u]]$ of §2 suppose the given coefficient map extends to $o\to V$. Give its undivided level-one period form $b$ the normalization in §2. There is a unique lift to $Z$ of its theta map with the given generic lift. On the original surface,
$$
                   J_0 R=(u^{h}),\qquad h=d_J\varphi(b).
                   \tag{7b}
$$
The arc therefore lifts to the normalized arithmetic blowup of $J_0$. On a Lan chart with the matching Cartier character of weight $d_J\varphi$, (7b) equals the pullback of its horizontal Cartier ideal.

**Proof.** Theorem 4a gives the morphism to projective space over $R$. Its generic lift comes from the marked abelian point. The finite algebra obtained by pulling $Z$ back to $\operatorname{Spec}R$ thus maps to $\operatorname{Frac}R$. Its image is integral over $R$ and lies in $R$ by normality, giving the unique lift. On $D(\pi u)$ that lift agrees by separatedness with the map through the marked moduli open.

First take the trait $\operatorname{Frac}(V)[[u]]$ with uniformizer $u$. Its generic marking is the one just identified. The multiplier of period $y$ in the pulled-back Raynaud family is $u^{b(y,y)/2}$ times the integral unit in the constant value line. Hence the arithmetic order formula gives order $h$. Every element of $J_0R$ has $u$-order at least $h$, so comparison of the power-series coefficients gives $J_0R=u^h I$ for an ideal $I\subset R$.

For the second trait take the continuous specialization $u\mapsto\pi$ into $V$. Its generic point lies in $D(\pi u)$. The specialized marked family has the same constant abelian and character lines. Its period multipliers are $\pi^{b(y,y)/2}$ times the same units. This specialization follows by pulling the polarized formal family back on the thickenings $V/(\pi^n)$ and retaining the marked generic comparison. It is the uniformizer specialization proved in [7o]. In particular it applies to this ring map with kernel $(u-\pi)$. The second trait thus gives exactly $h$ independently of the first test, so $IV=V$. Since $(\pi,u)$ maps into $(\pi)$, a proper ideal of $R$ cannot have unit image. This proves (7b). The resulting invertible image ideal gives the lift to the blowup. Its normalization is finite by excellence and its generic lift is specified on the normal abelian open, so the preceding finite-algebra argument applies again. The comparison with the Lan character follows from (1) with the indicated weights. $\square$

Equations (7a) and (7b) apply to the same marked arc. For example, in the primitive frame the fractional ideal of the symmetric coordinates of $s=a h_0$, divided by the $k$-th power of a generator of $J_0R$, is $(a^B u^{-kh})$. Thus the contribution at a further trait is $Bv(a)-khv(u)$ whenever $a$ and $u$ have defined nonzero images. On a split multivariable monomial chart with the same marked family and integral Cartier weight, the corresponding boundary equation is $\prod_i t_i^{d_J\varphi(\gamma_i)}$, where $\gamma_i$ are the period forms in the full-level lattice. Indeed the trait form is the sum of $\gamma_i$ with coefficients $v(t_i)$ and $\varphi$ is linear on the cone. Divisibility of each ideal generator by the monomial follows by the DVR criterion on the normal local chart. Any proper residual ideal would have positive order on a dominating DVR. The chart calculation needs the monomial family and its generic marking on these tests. The example in §3 describes the freedom on other integral models. [7o]

## 5. The signed contribution to the torsion mean

A rational theta section in this section means a nonzero element of $H^0(A,L)$ defined over a number field. It is regular on the generic abelian variety. Its extension to a given integral model may have a vertical scalar denominator.

Let $K'$ be a number field splitting the full generic $m$-torsion of the abelian variety and carrying the specified section $s$. Use integer-normalized $v_w$ and $\ell_w=\log Nw/[K':\mathbb Q]$ at a finite place $w$, with $Nw$ the residue cardinality. Put $I=\{P:T_m(s,P)\ne0\}$. For $s\ne0$ this set is nonempty by the characteristic-zero translate argument. Use the same algebraic section at each embedding $\sigma:K'\to\mathbb C$. In the release’s analytic frame the ordinary Gaussian point norm $n_\sigma(P)$ satisfies $|\sigma T_m(s,P)|=n_\sigma(P)^{m^2}$ [7b]. If the analytic sampling inequality has lower bound $c_\sigma$ for the sample mean of $-\log(n_\sigma(P)/b_\sigma)$ with positive norm parameter $b_\sigma$, the product formula gives
$$
 \operatorname{av}_\sigma c_\sigma
 \le \operatorname{av}_\sigma\log b_\sigma-D_m(s),
 \qquad
 D_m(s)=\sum_w\frac{\ell_w}{m^2|I|}\sum_{P\in I}v_w(T_m(s,P)).
 \tag{8}
$$
Here the embedding average counts all embeddings. At a matched good place covered by Theorem 4 the local minimum divided by $m^2$ is $v_w(a)$ and the mean is at least $v_w(a)$. In fact the local mean for $s=a h$ is the local mean for $h$ plus $v_w(a)$, since their zero sets agree. Formula (6) certifies the exact scalar clearing exponent $\max(0,-v_w(a))$. Other places need their own lower bounds in (8).

### 5.1. The semistable lower bound

Let $V$ be a complete DVR of characteristic-zero fraction field with integer-normalized valuation $v$, and let $A$ be a principally polarized abelian variety of dimension $g\ge1$ with split semistable reduction. Work with a specified symmetric rigidified principal line and its power-one Raynaud uniformization. The principal polarization identifies the period and character lattices with a lattice $X$ of rank $r$. Let $B$ be the period pairing in valuation units and $Q(x)=B(x,x)/2$, positive definite on $X_{\mathbb R}$ if $r>0$. We impose two conditions.

1. The **line and its rigidification** are identified with the symmetric pullback from the abelian quotient in the integral Raynaud splitting. The comparison preserves symmetry, the cube structure and the rigidification. Use the constant term normalization in which the period trivialization at $y\in X$ is of order $Q(y)$ in the inverse cubical line, and evaluates on the positive line. The second difference thus has order $B(x,y)$. The valuation uses this same uniformization and this same cubical multiplication frame on the entire generic torsion grid.
2. The specified integral section lattice has a generator $h$ whose weight-$x$ Fourier coefficient has order at least $Q(x)$ in the integral abelian coefficient lattice. This means that after division by a scalar of order $Q(x)$ the coefficient is a regular section of its abelian weight line. This is the adapted integral generator in the release. We write the actual nonzero generic section as $s=a h$ and keep $v(a)$ in the formula. For an integral $s$ in this lattice it is nonnegative.

These conditions use the formal Raynaud and cubical comparison of [4, §3.3.3] and the specific integral-section and evaluation data of [7f]. For separately chosen integral models they are comparison hypotheses.

**Theorem 5.** On $X_{\mathbb R}/X$ put $f(t)=\min_{x\in X}Q(x+t)$, let $M_Q$ be its Haar probability mean, and let $R_Q$ be its maximum. Set both constants to zero if $r=0$. For $m\ge2$ let $I_m=\{P\in A[m]:T_m(s,P)\ne0\}$ on the geometric generic fibre and set $\delta=1-(1-m^{-2})^g$. Then $I_m$ is nonempty and
$$
 \frac{1}{m^2|I_m|}\sum_{P\in I_m} v(T_m(s,P))
 \ \ge\ v(a)+
 \max\left(0,\frac{(1-m^{-2})M_Q-\delta R_Q}{1-\delta}\right).
 \tag{9}
$$
The last summand is at least $M_Q(1-(1+3g)/m^2)_+$, where $x_+=\max(x,0)$. If $r=1$ and $Q(x)=qx^2$ in a lattice basis, then $M_Q=q/12$.

**Proof.** In the Raynaud extension, let $z$ represent a geometric $m$-torsion point $P$ and satisfy $[m]z=\iota(y)$, where $\iota$ is the period injection. Changing $z$ by a period changes $y$ by an element of $mX$. For a fixed representative $y$ of its class, the lifts are a translate of the kernel of $[m]$ on the extension. The torus of rank $r$ and abelian quotient of dimension $g-r$ give $m^{2g-r}$ lifts, with no further period identification since $X$ is torsion-free. Thus $y/m\bmod X$ has uniform multiplicity over the whole geometric grid, including points that do not extend to the identity component.

In the matched torsion frame, unpowered evaluation of $h$ is the convergent sum of Fourier weight evaluations. The order of the abelian coefficient at weight $x$ is at least $Q(x)$ by condition 2 and properness of the abelian quotient. The normalized character contributes $B(x,y)/m$ since its $m$-fold multiplication recovers the period evaluation. The inverse fibre factor in the rigidified torsion frame contributes $+Q(y)/m^2$. Its $m^2$-th power has the order of the period trivialization, after cancellation by the origin rigidification. Choices of the root have unit ratio. In this frame the $m^2$-th power of the unpowered evaluation is $T_m(h,P)$, by the cubical multiplication identity. These are the character and frame factors of the integral evaluation formula [7f]. Thus the order of each unpowered term is at least
$$
 Q(x)+B(x,y)/m+Q(y)/m^2=Q(x+y/m).
$$
Taking the minimum over $x$, the ultrametric inequality for the convergent sum gives $m^{-2}v(T_m(s,P))\ge v(a)+f(y/m)$. The valuation is extended from $V$, with $v(0)=+\infty$. This uses no condition on $m$ relative to the residue characteristic.

For the full average let $D$ be the Voronoi cell of zero for $\sqrt Q$, with volume one in lattice-normalized measure. The function $f$ is $Q$ on $D$, and the translates of $D/m$ centred at the $m^r$ grid points tile the torus up to measure-zero boundaries. At a grid point $t$ choose a shortest representative $x_t$. For $z\in D/m$ we have $f(t+z)\le Q(x_t+z)=f(t)+B(x_t,z)+Q(z)$. Central symmetry cancels the linear term and the mean of $Q(z)$ over the small cell is $M_Q/m^2$. Integrating cell by cell gives a full grid mean at least $(1-m^{-2})M_Q$.

The removed points are the zeros of the principal theta divisor on the generic grid. Descend the polarized variety, divisor and torsion incidence to a finitely generated subfield of characteristic zero and embed it into $\mathbb C$. The geometric point count is invariant under algebraically closed field extension. Pareschi’s Theorem C [8] bounds this count by $m^{2g}-(m^2-1)^g$ for any translate of the principal theta divisor, including a reducible divisor. Thus the fraction removed is at most $\delta<1$. If it is $d_0$, each removed point costs at most $R_Q$ in the sum of $f$. The surviving average is at least $((1-m^{-2})M_Q-d_0R_Q)/(1-d_0)$, which decreases in $d_0$, and is also nonnegative. The scalar $v(a)$ is then added once. This proves (9).

For the uniform error take $u$ with $f(u)=R_Q$. Apply the parallelogram identity to shortest representatives of $t$ and $u-t$. Their sum represents $u$ and their difference represents $2t-u$. Thus $2f(t)+2f(u-t)\ge R_Q+f(2t-u)$. Doubling preserves Haar probability, so integration gives $R_Q\le3M_Q$, as in [11, Lemma 2.2]. Now use $\delta\le g/m^2$ in (9). In rank one the cell is $[-1/2,1/2]$ and integrating $qx^2$ gives $q/12$. The case $r=0$ uses $f=0$. $\square$

The line condition is essential. On the Tate curve with parameter $\mathfrak q=\varpi^2$ in odd residue characteristic, let the symmetric principal line be the origin divisor and rigidify using the torus frame $e(1)$. Its primitive section has coefficient $F(z)=(1-z)\prod_{j\ge1}(1-\mathfrak q^jz)(1-\mathfrak q^j/z)$ in the frame $e(z)$. The automorphy $F(\mathfrak q z)=-z^{-1}F(z)$ gives $e(\mathfrak q z)=-z e(z)$. The rigidified multiplication map sends $e(z^m)$ to $z^{m(m-1)/2}e(z)^{m^2}$, by this descent relation and evaluation at $z=1$. At $m=2$ and $z=\varpi$, the coefficient $F(z)$ is a unit and $v(T_2(h,P))/4=-1/4$, whereas $Q(t)=t^2$ gives $f(1/2)=1/4$. The other point in that period class has the same order, the nonzero point of the zero class has order zero, and the origin is deleted. The surviving mean is $-1/6$. Inversion multiplies the rigidified torus frame by $z$, and a change by a special-fibre unit $cz^k$ changes that multiplier to $z^{1-2k}$. Hence this symmetry cannot come from the abelian quotient. This example has a primitive section and falls outside the matched-line hypothesis. [4, §3.3.3] [10, Chapter V]

A separate comparison may identify a symmetric line with a two-torsion translate of a centered line, with tropical label $\xi\in(\tfrac12 X)/X$. Suppose it identifies the canonical metric up to a constant under translation and identifies the translated primitive lattice with order zero at the identity of the skeleton. The centered profile there is $f$, so the translated primitive profile is $f(t+\xi)-f(\xi)$, with the constant fixed by its value at zero. The skeletal order bounds the order at any point having that label [9, Propositions 8.3, 9.1 and 10.3]. The same Voronoi argument works for the shifted grid. One then subtracts $f(\xi)$ outside the positive part in (9), giving limiting excess $M_Q-f(\xi)$ as a lower bound. These metric and lattice identifications are additional hypotheses for this variant.

### 5.2. From the finite term to the archimedean bound

Over a fixed number field $F$ write $\ell_w=\log Nw/[F:\mathbb Q]$ and choose a finite set $S$ of places with Theorem 5’s comparison, after a fixed field extension if needed. For $w\notin S$ assume that the nonzero divided valuations are bounded below by $-b_w$, with $b_w\ge0$ of finite weighted sum, independently of $m$. The hypothesis on an extension of $w$ scales this lower bound by the ramification index. Define
$$
 L_{\rm fin}=\sum_{w\in S}\ell_w\bigl(v_w(a_w)+M_{Q_w}\bigr)
                  -\sum_{w\notin S}\ell_w b_w .
 \tag{10}
$$
Summing (9) yields an explicit finite-$m$ lower bound by replacing $M_{Q_w}$ in (10) with the last summand of (9). The splitting field for the grid may depend on $m$. The orders of $a_w$ and $Q_w$ scale by the ramification index, and the weights sum with residue degrees, giving the same normalized contribution over $F$. In particular $\liminf_m D_m(s)\ge L_{\rm fin}$. For a fixed family the analytic sampling inequality is eventual at the finitely many embeddings. Equation (8) therefore gives $\operatorname{av}_\sigma c_\sigma\le\beta-L_{\rm fin}$ with $\beta=\operatorname{av}_\sigma\log b_\sigma$. At a rank-one place the term from the degeneration is $\ell_w q/12$.

Here is the resulting refinement with the analytic inputs of [7g]. For each permitted complex frame, use the analytic radius $\rho>0$, write $Z=\rho^{-2}$ and set
$$
 E_g(Z)=g\log(1+\sqrt Z)+C_0g\log(2g),\qquad
 c_\sigma=\frac{\pi}{96}Z-E_g(Z).
$$
In these frames the supremum is bounded by $b_\sigma\exp(E_g(Z))$, with the same $b_\sigma$ as in (8). Put $\kappa=\pi/96$. For any selection of one frame per embedding, Jensen’s inequality with $Y=\operatorname{av}_\sigma Z$ gives
$$
 \kappa Y\le\beta-L_{\rm fin}+g\log 2+C_0g\log(2g)
                         +\tfrac g2\log(1+Y).
$$
Using $1+Y\le(1+g/\kappa)(1+\kappa Y/g)$ and $\log(1+x)\le x$, we get $Y\le C((\beta-L_{\rm fin})_++g\log(2g))$ for a constant $C$ depending only on the analytic constant $C_0$. This applies to every frame selection. The individual suprema of $E_g(Z)$ are finite, by fixing the frames at the other embeddings and varying the chosen one. Selecting frames arbitrarily close to those suprema and using Jensen again gives positive bounds $M_\sigma$ for the theta suprema in all the prescribed frames with
$$
 \operatorname{av}_\sigma\log M_\sigma
 \le \beta+C_0g\log(2g)+g\log 2+
 \tfrac g2\log\!\left(1+C((\beta-L_{\rm fin})_++g\log(2g))\right).
 \tag{11}
$$
The finite moment reduces the upper bound on the radius term through $\beta-L_{\rm fin}$. The linear norm term is still $\beta$, as the mean-square parameter in the initial supremum estimate is unchanged. If only $\beta\le\Phi$ is available, the right side of (11) is increasing in $\beta$ and can be evaluated at $\Phi$. This bound uses the same marked section throughout. At the terminal CM point with potentially good reduction of the Serre tensor, the toric moment after good reduction is zero. Related tropical moments occur in the Faltings-height formula of de Jong and Shokrieh [9].

### 5.3. The full torsion mean for a marked elliptic product

**Proposition 6.** Let $E$ have good reduction over a mixed-characteristic DVR. Pass to a finite extension over which the labelled nonzero $H\in E[2]$ and the full $m$-torsion are rational, with $m\ge2$. Write $v_w$ for its integer valuation and $c_w=i_w(0,H)$ for the intersection order of the two sections. Rigidify $\mathcal L_H=\mathcal O([H])\otimes\pi^*(0^*\mathcal O([H]))^{-1}$ at the origin. Its rational section $s_H$ with divisor $[H]$ on the generic fibre and $s_H(0)=1$ is unique. For $I_m=\{P\in E[m]:s_H(P)\ne0\}$ the local mean is
$$
 \frac{1}{m^2|I_m|}\sum_{P\in I_m}v_w(T_m(s_H,P))=
 \begin{cases}
 v_w(m)/(m^2-1)-c_w,&m\ \text{even},\\
 -(1-m^{-2})c_w,&m\ \text{odd}.
 \end{cases} \tag{12}
$$
The origin belongs to both averages. At odd residue characteristic $c_w=0$. At a dyadic place the three nonzero half-periods, over a common splitting field, satisfy $\sum_{H\ne0}i_w(0,H)=v_w(2)$.

**Proof.** The canonical divisor section is primitive over the DVR. Its value at the origin in an integral frame has order $c_w$. Dividing by this value in the rigidified line gives $s_H$. For $P\ne H$ the evaluation in an integral frame has order $i_w(P,H)-c_w$. The integral cubical isomorphism of §4 consequently gives
$$
 m^{-2}v_w(T_m(s_H,P))=i_w(P,H)-c_w
$$
also in residue characteristic dividing $m$. The translated odd theta function in [7c] has this divisor and this value at zero. These two properties identify the section.

On the smooth integral model the Cartier divisor $[m]^*[0]$ equals $\sum_{P\in E[m]}[P]$. Its kernel is finite flat, so it has no fibre component, and each generic torsion point has multiplicity one. At odd $m$, evaluate this equality on $H$. Since $[m]H=H$, the sum of the intersection orders is $c_w$ and $|I_m|=m^2$. At even $m$, translation by $-H$ identifies the surviving intersections with $i_w(0,Q)$ for all nonzero $Q\in E[m]$. Near the origin in a formal parameter $z$, a local equation for $[m]^*[0]-[0]$ is $[m](z)/z$, up to a unit. Its value at zero is $m$ times a unit because the differential of multiplication by $m$ is $m$. Thus the intersection sum is $v_w(m)$ and $|I_m|=m^2-1$. This proves (12), including the claim about the three half-periods by taking $m=2$. Away from $2$ the finite étale group $E[2]$ separates distinct torsion sections. $\square$

**Corollary 7.** Take an exterior product $s=\boxtimes_{i=1}^g s_{H_i}$ on $A=\prod_{i=1}^g E_i$ with the product rigidification, where the elliptic factors have everywhere potentially good reduction. Enlarge the number field to give good reduction everywhere and split the specified points and grids. Define $C_H=\sum_{i,w}\ell_w i_w(0,H_i)$. Then $0\le C_H\le g\log2$ and
$$
 D_m(s)=
 \begin{cases}
 g\log m/(m^2-1)-C_H,&m\ \text{even},\\
 -(1-m^{-2})C_H,&m\ \text{odd}.
 \end{cases} \tag{13}
$$
For the auxiliary CM product in the ordinary construction $C_H=0$. Its full finite contribution to (8) is $g\log m/(m^2-1)$ for even $m$ and zero for odd $m$.

**Proof.** The nonzero set is the product of the single-factor nonzero sets, and the logarithmic valuations add. Summing (12) uses $\sum_{w\mid p}\ell_w v_w(p)=\log p$. This identity also gives the bound on $C_H$ from the $m=2$ intersection sum. Under extension, intersection orders scale by the local ramification indices and the normalized global sums agree.

For the actual auxiliary curve the release uses $k_0=\mathbb Q(\sqrt{-d})$ with $d\equiv7\bmod16$, and lattice $\delta\mathcal O_{k_0}$ in its chosen complex embedding. The constructor takes $H=[\delta/2]$. Its marking requires $aH\ne0$ and $(1-a)H\ne0$ for $a=(1+\sqrt{-d})/2$. The polynomial of $a$ reduces to $X(X-1)$ modulo $2$. Thus $a$ and $1-a$ induce the two complementary idempotents of $\mathcal O_{k_0}/2$ and the marking has nonzero projection to each summand. [7h]

After good reduction the CM action extends to the abelian scheme. The splitting $\mathcal O_{k_0}\otimes\mathbb Z_2=\mathbb Z_2\times\mathbb Z_2$ decomposes its $2$-divisible group into two height-one factors, as the generic Tate module has rank one over this ring. Their dimensions add to one. The dimension-zero factor has étale special fibre, so its finite flat $2$-torsion is étale over the trait. The nonzero projection of $H$ to this factor remains nonzero on reduction. This proves $i_w(0,H)=0$ at every dyadic place. The same argument applies to each copy and each conjugate of the marked curve. $\square$

There is an exact norm comparison for (11) on this product. Give each section line the cubical metric with unit norm at the rigidification and use Haar probability in its $L^2$ norm. Set $\beta=\operatorname{av}_\sigma\log\|s\|_{L^2,\sigma}$ and $F_A=\tfrac12 h_F(A)+\tfrac g4\log(2\pi)$ with Deligne’s metric on the Hodge determinant. In a primitive local generator the section $s$ has scalar order $-\sum_i i_w(0,H_i)$. The normalized Arakelov degree of the direct image is therefore $-C_H-\beta$. The symmetric principal line on the good abelian scheme is a Moret-Bailly model after splitting the required torsion, and its direct image has normalized degree $-F_A$ by [5, Theorem 3.4(iii)–(v)]. Hence $\beta=F_A-C_H$, and the expression in the sampling bound is
$$
 \beta-D_m(s)=
 \begin{cases}
 F_A-g\log m/(m^2-1),&m\ \text{even},\\
 F_A-C_H/m^2,&m\ \text{odd}.
 \end{cases} \tag{14}
$$
For the selected product, the product formula also identifies (13) with the logarithmic Gaussian point norm averaged over all embeddings and all nonzero grid values. The product enters the construction before the fourth-tensor lattice transport. Applying a height estimate to its transported section requires the specified integral line, scalar and metrics for that section. [7b, 7h]

The supplement contains the [dyadic CM marking note](supplement.zip) (entry `dyadic_marked_CM_class.md`) and the [independent grid-mean replay](supplement.zip) (entry `semistable_grid_checks/cm_half_period_mean.py`).

### 5.4. Torsion evaluation and the Faltings height

**Proposition 8.** Let $A$ be an abelian variety of dimension $g\ge1$ over a number field, with a symmetric rigidified principal line $L$. Use the cubical point metric and Haar probability in the $L^2$ norm. For each nonzero rational section $s$ and $m\ge2$, let $I_m$ be the subset of $A[m]$ where $s$ is nonzero. There is a number field $K$ splitting the grid and a Moret-Bailly model over $\mathcal O_K$ on which all its points extend. Write $\mathcal P=\pi_*\mathcal L$ for its line of sections. For $P\in I_m$ let $b_w(P)$ be the length of the cokernel of the evaluation of $\mathcal P$ in the line $\varepsilon_P^*\mathcal L$ at the finite place $w$. Then
$$
 \beta-D_m(s)=F_A-
 \frac1{|I_m|}\sum_{P\in I_m}\sum_w\ell_w b_w(P)
 \ \le\ F_A . \tag{15}
$$
This holds at every reduction type. In particular the radius term in (11) can use $(F_A)_+$. After one finite extension one can choose a single rational section $s$ with $\beta=F_A$ and $D_m(s)\ge0$ for every $m\ge2$.

**Proof.** Use the existence theorem for a Moret-Bailly model with the prescribed finite set $A[m]$. Its semistable group scheme can have disconnected special fibres. Its cubist line has the canonical point metric, fixed by the rigidification and the translation-invariant curvature. The direct image carries the Haar probability $L^2$ metric used in $\beta$. Since $L$ is principal, the normalized Arakelov degree of $\mathcal P$ is $-F_A$. The normalized degree of $\varepsilon_P^*\mathcal L$ is the Néron–Tate height of $P$, which is zero for torsion. [5, Definition 3.1 and Theorem 3.4(i), (ii), (v)]

For each surviving $P$, the evaluation $\mathrm{ev}_P:\mathcal P\to\varepsilon_P^*\mathcal L$ is an integral map of lines and is nonzero on the generic fibre. Its finite orders are the nonnegative integers $b_w(P)$. Its norm at $\sigma$ is $n_\sigma(P)/\|s\|_{L^2,\sigma}$, where $n_\sigma(P)$ is the point norm of this same section. Computing the degree of the Hom line in its rational frame $\mathrm{ev}_P$ gives
$$
 F_A=\sum_w\ell_w b_w(P)+\operatorname{av}_\sigma
               \log\frac{\|s\|_{L^2,\sigma}}{n_\sigma(P)} .
$$
This evaluation argument uses the single surviving point even when the line of sections has a base divisor elsewhere. For a system generating a line on the whole generic fibre, its base-point orders enter the height formula of [5, §4]. Here the characteristic-zero torsion translate argument of §4 ensures $I_m$ is nonempty. The defining cubical identity $|\sigma T_m(s,P)|=n_\sigma(P)^{m^2}$ and the product formula give $D_m(s)=|I_m|^{-1}\sum_{P\in I_m}\operatorname{av}_\sigma\log n_\sigma(P)$. Averaging the displayed degree equality proves (15). At a good place one can read $b_w(P)$ as the intersection with the divisor of a primitive local section. At a bad place (15) uses the evaluations on the model marking the complete grid. The numerical equality concerns their globally weighted sum. [5, Theorem 3.4(iii), (iv)] [7b]

For the simultaneous choice of $s$, first take a Moret-Bailly model with the empty marked set. Its invertible direct image becomes free after finite extension. Indeed, a fractional ideal $J$ of class order $h$ satisfies $J^h=(u)$, and after adjoining an $h$-th root of $u$ the extended ideal is principal. Choose a generator $s$. Its degree calculation gives $\beta=F_A$. The direct image commutes with extension and agrees isometrically on any two Moret-Bailly models of the same generic data. A model with additional markings is still a model for the empty marked set. Thus for each $m$ the grid can be marked over a further field with this same generic $s$, and (15) gives $D_m(s)\ge0$. $\square$ [5, Theorem 3.4(iii), (iv)]

For a prescribed rational section, (15) bounds the logarithmic sampling mean relative to its Haar norm. The absolute period-operator norm in [1, (6)] also uses the integral column and the scale of the section in its prescribed row.

### 5.5. A finite-grid bound at all reduction places

Work over a number field $K$ of semistable reduction for $A$. At each finite place $v$, take the geometric period lattice $X_v$, its pairing $B_v$ in integer valuation units and the function $f_v(t)=\min_{x\in X_v}B_v(x+t,x+t)/2$ on its real torus $\Sigma_v$. Let $M_v$ be its Haar probability mean. At good reduction $\Sigma_v$ is a point and $M_v=0$. Define
$$
 \mathcal T(A)=\sum_{v<\infty}\frac{\log Nv}{[K:\mathbb Q]}M_v,\qquad
 \alpha_{g,m}=\left(\frac{1-m^{-2}-3\delta}{1-\delta}\right)_+,
 \qquad \delta=1-(1-m^{-2})^g .
$$
The sum is finite and is unchanged by extending $K$. Integer-normalized period pairings scale by the ramification index, and the global weights account for the residue degrees. In the convention of de Jong and Shokrieh, $M_v$ is half the tropical moment [9, Theorem B].

**Theorem 9.** For any nonzero rational global section of the symmetric rigidified principal line and every $m\ge2$,
$$
 \beta-D_m(s)\ \le\ F_A-\alpha_{g,m}\mathcal T(A),
 \qquad \alpha_{g,m}\ge\bigl(1-(1+3g)/m^2\bigr)_+ . \tag{16}
$$

**Proof.** Pass to a Moret-Bailly model marking $A[2m]$. At a place $w$ above $v$ use $w$-valuation units, so $|\varpi_w|=\exp(-1)$ in the local argument. Let $e$ generate the direct image over this trait. The evaluation length $b_w(P)$ in (15) equals $-\log\|e(P)\|_{\rm can}$ for a surviving point. Indeed the cubical metric gives the finite model order on a section extending to this model [9, Proposition 10.2]. Equivalently the integral multiplication isomorphism identifies the $m^2$-th power of its evaluation line with the rigidified origin line, giving $b_w(P)=m^{-2}v_w(T_m(e,P))$.

On the canonical skeleton the negative log norm of $e$ has the form $f_w(t+\eta)+d$. To obtain this expression, translate to the symmetric line of [9, Proposition 8.3], use its quadratic formula [9, (11.3)–(11.5)], and translate back. The metrics differ under this translation by a constant. The translating point is killed by two because both lines are symmetric and the polarization is principal. Hence $\eta\in\Sigma_w[2]$. The map $A[2]\to\Sigma_w[2]$ is onto by the Raynaud division argument in §5.1. The component at the minimum therefore occurs on the marked model, including at $w\mid 2$. At the skeletal point of each model component the negative log norm is its vertical order [9, Proposition 10.3]. These orders are all nonnegative for $e$, and their minimum is zero since $e$ generates the direct image [9, Proposition 10.6]. As the skeletal minimum occurs among these components, $d=0$. The supremum on each fibre of retraction is attained at its skeletal point [9, Corollary 10.4]. We have proved
$$
 b_w(P)\ \ge\ f_w(t_P+\eta).
$$
This normalization also explains the Tate example in §5.1. There the primitive connected-model section has skeletal minimum $-1/4$ in base units. The stable generator adds $1/4$ to its divided orders.

The full grid has equal multiplicities over $\Sigma_w[m]$. The translated cells centred on $\Sigma_w[m]+\eta$ give mean at least $(1-m^{-2})M_w$ by the cell integration in Theorem 5. Removing the vanishing points costs at most $3M_w$ per removal by the parallelogram bound, and their fraction is at most $\delta$ by [8, Theorem C]. Dividing by the surviving fraction gives mean length at least $\alpha_{g,m}M_w$. Summing (15) and descending the weighted moments proves (16). $\square$

**Exact elliptic case.** For an elliptic factor put $d_v=v(\Delta_{\min})$ at split multiplicative reduction and $d_v=0$ at good reduction. Its moment is $d_v/24$. Let $[H]$ be the divisor of the section, so $2H=0$, and let $\lambda_v$ denote the Néron function for $[0]$ with skeletal mean zero. In base valuation units the local mean of the lengths in (15) is
$$
 \frac{d_v}{24}+
 \begin{cases}
 v(m)/(m^2-1), & [m]H=0,\\
 \lambda_v(H)/m^2, & [m]H\ne0 .
 \end{cases} \tag{17}
$$
At good reduction $\lambda_v(Q)=i_v(0,Q)$ for $Q\ne0$. On the Tate curve take a representative $z$ of $Q$ with $0\le t=v(z)/d_v<1$. Then $\lambda_v(Q)=v(1-z)+(d_v/2)B_2(t)$ with $B_2(t)=t^2-t+1/6$ [10, Chapter VI]. Its skeletal minimum is $-d_v/24$. Translation by $H$ gives a local function for $[H]$, and the minimum-zero normalization in the proof above yields $b_v(P)=\lambda_v(P-H)+d_v/24$ after dividing lengths by the ramification index.

For $[m]H=0$ the differences $P-H$ range over all nonzero $m$-torsion points. At good reduction their intersection orders sum to $v(m)$ by Proposition 6. At split multiplicative reduction use representatives $\zeta\mathfrak q^{j/m}$, $0\le j<m$, $\zeta^m=1$. The identity $m\sum_{j=0}^{m-1}B_2((t+j)/m)=B_2(t)$ makes the Bernoulli sum at $t=0$ vanish after removal of the identity term. The unit terms sum to $v(m)$ since $\prod_{\zeta\ne1}(1-\zeta)=m$ over the nontrivial $m$-th roots. In the other case the differences form the full inverse image of $H$ under $[m]$. At good reduction the pullback divisor gives total $\lambda_v(H)$. On the Tate curve apply the same Bernoulli identity at the label of $H$ and $\prod_{\zeta^m=1}(1-\zeta z^{1/m})=1-z$ to its unit terms. Division by the corresponding grid size proves (17). The arguments apply at residue characteristics dividing $2m$. Exterior products add the means because their surviving sets are Cartesian products. In particular, for any product of $g$ elliptic curves with its product polarization and even $m$, (15) is exactly $F_A-\mathcal T(A)-g\log m/(m^2-1)$.

### 5.6. The archimedean radius in the point-slope proof

For a complex embedding set $I_\sigma=\int-\log(n_\sigma(P)/b_\sigma)\,d\mu_\sigma(P)$ on the complex torus, with Haar probability and the ordinary point and $L^2$ norms of Proposition 8. Write $\Theta=\operatorname{div}(s)$ and $h'_L(\Theta)$ for its normalized canonical cycle height [9, (14.1)]. In the Hodge normalization of §5.3 the height formula of [9, Theorems A and B] reads
$$
 S_A:=\operatorname{av}_\sigma I_\sigma
   =F_A-\mathcal T(A)-g h'_L(\Theta). \tag{18}
$$
Indeed the squared Deligne Hodge norm is $\pi^{-g}$ times the squared Hodge norm of [9, (12.1)]. The normalized degree increases by $(g/2)\log\pi$, so $F_A$ equals half the sum of their Faltings height and $g\log(\pi\sqrt 2)$. The Haar norms and normalized section-line degree agree in both conventions. The integrals are finite by local integrability of the logarithm of a holomorphic section. Jensen's inequality gives $I_\sigma\ge0$, and $h'_L(\Theta)\ge0$ by positivity of the canonical cycle height.

There is also a direct relation to the sampled quantities. Truncate $-\log(n_\sigma/b_\sigma)$ from above at a fixed real number, assigning that value on the divisor. The resulting function is continuous and bounded. Its full grid averages tend to its Haar integral by ordinary Riemann sums. Removing the vanishing points changes that limit by zero since their fraction is at most $\delta$. Letting the truncation increase and using (16) now gives
$$
 S_A\ \le\ \liminf_m(\beta-D_m(s))\ \le\ \limsup_m(\beta-D_m(s))
       \ \le\ S_A+g h'_L(\Theta). \tag{19}
$$
For the generally unbounded logarithm this argument yields precisely the indicated directions. When the divisor has canonical height zero the two bounds agree.

**Corollary 10.** Let $r_\sigma$ be the shortest nonzero period length in the polarization norm $\sqrt{H(\gamma,\gamma)}$ at $\sigma$. Set $U=(24/\pi)(S_A+(g/4)\log 2)$. The actual section in (11) satisfies
$$
 \operatorname{av}_\sigma r_\sigma^{-2}\le U,\qquad
 \operatorname{av}_\sigma\log\|s\|_{\sup,\sigma}
 \le \beta+g\log\!\left(1+\sqrt{8gU/\pi}\right). \tag{20}
$$
These estimates apply to every symmetric marking inducing the polarization.

**Proof.** Put $t=\min(r_\sigma,\sqrt{\pi/(3g)})$. Autissier's local estimate [11, §3.2 and Lemma 3.2] gives $2I_\sigma\ge\pi/(6t^2)+g\log t+(g/2)\log(3g/(\pi e))$. Substituting $x=\pi/(6gt^2)$ and using $\log x\le x-1$ gives $I_\sigma\ge\pi/(24r_\sigma^2)-(g/4)\log2$. Average and use (18).

At one embedding write the complex torus in period coordinates $z=x+\tau y$, with $\tau=X+iY$. Set $F(z)=\|s(z)\|^2/b_\sigma^2$ and $\Gamma=\mathbb Z^g+\tau\mathbb Z^g$. A half-characteristic translates the normalized squared norm of the zero-characteristic theta section. Its Fourier coefficient at $(p,q)\in\mathbb Z^{2g}$ therefore has modulus
$$
 |\widehat F(p,q)|=\exp\!\left(-\tfrac{\pi}{2}
 \bigl[p^{\mathsf T}Yp+(q-Xp)^{\mathsf T}Y^{-1}(q-Xp)\bigr]\right)
 =e^{-\pi\|q-\tau p\|_{\rm pol}^2/2}.
$$
Indeed the $x$-integral in the squared theta series fixes the difference of its indices to $p$. If the second index is $j$, the change $u=j+y+p/2$ joins the $y$-cubes into $\mathbb R^g$. The resulting Fourier transform of $e^{-2\pi u^{\mathsf T}Yu}$ at $q-Xp$ gives the second quadratic term. Its determinant factor cancels against $\widehat F(0,0)=1$ after normalization. The residual phase at zero characteristic is $(-1)^{p^{\mathsf T}q}$. Both the normalized series and its Fourier expansion converge absolutely. [11, §3.2] [14]

The mean of $F$ on $z+A[m]$ keeps precisely the frequencies in $m\mathbb Z^{2g}$. Its absolute value is bounded by $\sum_{\gamma\in\Gamma}e^{-\pi m^2\|\gamma\|_{\rm pol}^2/2}$. Write $N(u)$ for the number of periods of length at most $u$. Disjoint balls of radius $r_\sigma/2$ give $N(u)\le(1+2u/r_\sigma)^{2g}$. If $T$ has density $\pi t e^{-\pi t^2/2}$ on $t\ge0$, the Gaussian sum is $\mathbb E N(T/m)$. Since $\mathbb E T^{2g}=(2/\pi)^g g!$, the triangle inequality in $L^{2g}$ gives
$$
 \left(\frac{1}{m^{2g}}\sum_{P\in A[m]}F(z+P)\right)^{1/2}
 \le \left(1+\frac{\sqrt{8g/\pi}}{m r_\sigma}\right)^g.
$$
Take $m=1$ and apply concavity of $\log(1+\sqrt{8g x/\pi})$ at $x=r_\sigma^{-2}$ to obtain (20). $\square$

### 5.7. The integral theta lattice from a single torsion level

Let $m$ be a positive multiple of six. Over a Moret–Bailly field splitting $A[m^2]$, use the stable line of sections $\mathcal P$ of Proposition 8. For a nonzero rational section $s$ let $\mathfrak c(s)=s\mathcal P^\vee$ be its coefficient fractional ideal. Thus $\mathfrak c(s)_w=(a_w)$ if $s=a_we_w$ in a local generator of $\mathcal P$. For $Z\in A(\overline{\mathbb Q})$, let $\Theta_m(Z)$ be its image in the theta embedding attached to $L^{m^2}$. Its coordinates are the values of the translates of $[m]^*s$ indexed by representatives of $A[m^2]/A[m]$, with the finite-order lifts of [5, §§2.3, 5.1]. We use the coordinate maximum at finite places and the Euclidean norm at complex embeddings in the absolute projective height $h_2$. Put
$$
 G_{m,\sigma}(z)=\frac{1}{m^{2g}}\sum_{P\in A[m]}
            \frac{\|s(z+P)\|_\sigma^2}{\|s\|_{2,\sigma}^2}.
$$

**Theorem 10a.** Already at $m=6$ the stable integral lattice is determined by its powered values. More generally,
$$
 \bigl(T_m(s,P):P\in A[m]\bigr)=\mathfrak c(s)^{\,m^2}.
 \tag{20a}
$$
For every algebraic point $Z$, write $\tau_wZ$ for its image on the skeleton and use $f_w$ in the valuation units of §5.5. Then
$$
\begin{aligned}
 h_2(\Theta_m(Z))-m^2\widehat h_L(Z)-g\log m
  ={}&F_A-\frac{1}{m^2}\sum_{w<\infty}\ell_w f_w(m^2\tau_w Z)\\
    &+\frac12\operatorname{av}_\sigma\log G_{m,\sigma}([m]Z).
\end{aligned} \tag{20b}
$$
Here $\widehat h_L$ is the Néron–Tate height for the rigidified line. All terms can be evaluated after extending the field to define $Z$. At $Z=0$ the finite correction vanishes place by place and
$$
 G_{m,\sigma}(0)=\sum_{\gamma\in\Gamma_\sigma}
                  e^{-\pi m^2\|\gamma\|_{\rm pol}^2/2}.
 \tag{20c}
$$
In particular $h_2(\Theta_m(0))-g\log m$ decreases to $F_A$ along these levels. For fixed $(A,L)$ the difference between the left side of (20b) and $F_A$ lies between $-3\mathcal T(A)/m^2-O_A(e^{-c_A m^2})$ and $O_A(e^{-c_A m^2})$ for some $c_A>0$, uniformly in the algebraic point $Z$.

**Proof.** Work at one finite place in additive valuation units and write $\nu(v)=-\log\|v\|_{\rm can}$. The proof of (16) gives the exact skeletal profile of the stable generator $e_w$ as $f_w(t+\eta)$ with $2\eta=0$. Translate to the centered line as in §5.5 and split the Raynaud data. If $t$ lifts the centered label of any point $D$, the labels of $D+A[m]$ lift to the coset $t+m^{-1}X_w$ because $A[m]\to\Sigma_w[m]$ is onto. Choose a vector $u$ of least norm in this coset. Then $mu$ lies in the closed Voronoi cell of the origin for $X_w$. For $Q(y)=B_w(y,y)/2$, the Voronoi inequalities give
$$
 Q(y+u)-Q(u)\ \ge\ (1-1/m)Q(y)>0
 \qquad (0\ne y\in X_w).
$$
In particular the Fourier expansion on this chosen component has a unique leading weight, even if the original coset has several shortest vectors.

In the centered expansion take the coefficient of weight zero primitive on the good abelian quotient. Its reduction is a nonzero section of a principal line. The Fourier recursion at weight $y$ translates that section and multiplies it by the period factor of order $Q(y)$. The character at $u$ contributes $B_w(y,u)$ and the cubical metric contributes $Q(u)$. Thus the skeletal order $Q(u)$ is attained at a point of this component as soon as the reduced weight-zero coefficient is nonzero at its abelian coordinate. All higher terms then have strictly greater order by the displayed inequality and their series converges. For the original section the additive constant is its stable order by §5.5. [9, §§6, 8–11] [15, Proposition 3.19, Corollary 3.21 and Theorem 4.10]

Choose $n=3$ in residue characteristic two and $n=2$ at every other finite place. Within the chosen component the evaluation points can be varied by the $n$-torsion of the identity Raynaud extension. Its reduction projects onto the $n$-torsion of the abelian quotient. Indeed a lift of any abelian $n$-torsion point has $n$-multiple in the torus, which can be removed by torus division. Special-fibre $n$-torsion points lift after a finite extension by étaleness of multiplication by $n$ at these points. On the resulting abelian $n$-torsion coset the reduced coefficient is nonzero somewhere by Theorem 4 applied to its translate. In abelian dimension zero the primitive coefficient is already a unit. This proves attainment of the minimum on the entire coset.

Now put $D=[m]Z$. The minimum over its whole grid is
$$
 \min_{P\in A[m]}\nu\!\left(s([m]Z+P)\right)
 =v_w(a_w)+\min_{v\in\Sigma_w[m]}f_w(m\tau_wZ+\eta+v)
 =v_w(a_w)+m^{-2}f_w(m^2\tau_w Z).
$$
The last equality is homogeneity of $Q$ on the lattice divided by $m$, using $m\eta=0$. At $Z=0$ the cubical identity for $T_m$ gives (20a) after taking the local exponents. In particular a single complete level-six vector tests integrality, including at its own residue primes. The stable line matters here. The connected generator in the shifted Tate example of §5.1 has minimum $-1/4$ in base units.

To compute the projective height, take any rational frame $d$ of the fibre $L_Z^{m^2}$. The isometric translation and multiplication maps identify the local norms of the coordinate sections in this fibre with the point norms of $s$ at $[m]Z+P$ [5, Proposition 5.1(i)]. In the coordinates divided by $d$, the frame terms over all places sum to the degree $m^2\widehat h_L(Z)$. The finite maximum contributes the negative of the minimum just computed. At infinity the log Euclidean norm before division by the frame is $\log\|s\|_{2,\sigma}+g\log m+\tfrac12\log G_{m,\sigma}([m]Z)$. Since $\beta-\sum_w\ell_wv_w(a_w)=F_A$, summing gives (20b). This evaluates the finite base-point terms of the translated linear system in [5, Lemma 5.2 and Proposition 5.3] for these levels.

In the surviving Fourier frequencies $(mp,mq)$ the phase $(-1)^{m^2p^{\mathsf T}q}$ is one. The half-characteristic also contributes phase one since $m$ is even. This gives (20c). At any $z$,
$$
 |G_{m,\sigma}(z)-1|\le\sum_{0\ne\gamma\in\Gamma_\sigma}e^{-\pi m^2\|\gamma\|_{\rm pol}^2/2}.
$$
For a fixed polarized variety the right side is $O_A(e^{-c_A m^2})$, by separating half the exponent and using the shortest period at each of its finitely many embeddings. The same bound then holds for $|\log G_{m,\sigma}|$ for large $m$. The finite error in (20b) has the stated sign and size by $0\le f_w\le3M_w$ from Theorem 5. $\square$

**Corollary 10b (uniform distortion in the even linear coordinates).** Write $D_m(Z)=h_2(\Theta_m(Z))-m^2\widehat h_L(Z)-g\log m$. With the tropical moment $\mathcal T(A)$ in (18) and $U$ in Corollary 10, for every $6\mid m$ and every algebraic point $Z$,
$$
\begin{aligned}
0&\le D_m(0)-D_m(Z)\le (3\mathcal T(A)+\pi g^2U)/m^2,\\
0&\le D_m(0)-F_A\le 4g^2U/m^2 .
\end{aligned} \tag{20d}
$$
Equality $D_m(Z)=D_m(0)$ holds precisely on $A[m^2]$.

**Proof.** At a complex embedding let $\Gamma$ be the period lattice with its real polarization metric in dimension $2g$. In these coordinates the integral dual is $\Gamma^\vee=-i\Gamma$ and the covolume is one. Indeed for a period matrix $X+iY$, the ordinary real covolume is $\det Y$ and the metric contributes its inverse. At $z$ the Fourier coefficients in (20c) acquire the phases $\exp(2\pi i\langle\gamma,-imz\rangle)$. Thus $G_m(0)$ is the maximum, and every phase equals one there. Poisson summation expresses $G_m(z)$ as a constant times $\sum_{\xi\in\Gamma^\vee}\exp(-2\pi\|\xi-u\|^2/m^2)$, where $u$ is a shortest representative of $-imz$ modulo $\Gamma^\vee$. Pair $\xi$ and $-\xi$. Each pair contributes at least its value at zero times $\exp(-2\pi\|u\|^2/m^2)$, by $\cosh t\ge1$. Banaszczyk’s Euclidean transference bound $\mu(\Gamma^\vee)\lambda_1(\Gamma)\le g$ bounds $\|u\|$ by $g/r_\sigma$. Averaging and using $\operatorname{av}_\sigma r_\sigma^{-2}\le U$ gives the archimedean part of the first inequality. At finite places $0\le f_w\le3M_w$. The Fourier maximum forces $-imz\in\Gamma^\vee$. In (20b) one has $z=[m]Z$, so this condition is $[m^2]Z=0$. For such $Z$ all the finite corrections vanish as well. [11, 16]

For the estimate at zero put $q=mr_\sigma$. If $q\le2\sqrt g$, the packing estimate used in (20) gives
$$
\tfrac12 q^2\log G_m(0)
 \le gq^2\log(1+\sqrt{8g/\pi}/q)<4g^2 .
$$
If $q\ge2\sqrt g$, disjoint balls of radius $q/2$ bound the number of points of $m\Gamma$ with norm in $[2^jq,2^{j+1}q)$ by $(1+2^{j+2})^{2g}$. Its logarithm is at most $4^jq^2$, since $q^2\ge4g$. After including the Gaussian, the sum on these shells is at most $\sum_{j\ge0}e^{-4^jq^2/2}\le2e^{-q^2/2}$. Thus $\tfrac12\log G_m(0)\le q^{-2}$ in this range. The zero frequency gives the lower bound at the origin. Averaging proves (20d). $\square$

For the positive-degree branch in the nonterminal tangent estimate, write $G=f(p+q)$ for the dimension bound and $\Lambda$ for the size parameter there. The estimates $G\ge 8$, $\Lambda>24/5$, $3\mathcal T(A)\le2G\Lambda$ and $U\le(16/3)G\Lambda$ make $64G^3\Lambda/3$ an upper bound for both numerators in (20d). Thus the single level $6\lceil\sqrt{64G^3\Lambda/(3\epsilon)}/6\rceil$ gives both $D_m(0)-D_m(Z)\le\epsilon$ and $|D_m(Z)-F_A|\le\epsilon$ on this branch. This choice measures the even linear embedding of the abelian fibre. The odd map of §4.1 uses symmetric coordinates on the moduli space. [7g, 11, 16]

## 6. Scope of the height application

In the nonterminal point proof the first theta input is existential. Given its connected integral semiabelian model $G$ with principally polarized abelian generic fibre $A$, the proof asks after finite extension for an integral section of its power-one line with mean log Haar norm at most $\max(0,-\widehat{\deg}_{\rm nor}\operatorname{Lie}G)/2+Cg$ [7j]. Take the generator in Proposition 8. On every finite trait the connected semiabelian model is the identity component of the Néron model, as is the identity component of the Moret-Bailly model [9, §10]. Their two rigidified extensions of the same generic line agree on this component. Indeed the divisor of the generic identification could only be a multiple of its integral special fibre, and restricting to the rigidification forces this multiple to vanish. Thus the generator restricts to an integral section on the prescribed model. In periods $(I,Z)$ the point's tangent metric is $\pi(\operatorname{Im}Z)^{-1}$. Its dual determinant gives the squared norm $\det(\operatorname{Im}Z)/\pi^g$ to the volume differential. This is Deligne's Hodge metric, so $-\widehat{\deg}_{\rm nor}\operatorname{Lie}G=h_F(A)$. Proposition 8 supplies the stated input with $C=\log(2\pi)/4$, and (20) bounds the supremum of this very section. At a point generically on the horizontal boundary the slope argument calls this result on its principally polarized abelian quotient of positive dimension. The toric determinant is calculated separately [7j].

**Proposition 11.** Give the full integral tangent lattice of $A$ the point metric just described and let $\widehat\mu_{\max}$ be its maximal normalized Arakelov slope. Put
$$
 H_A=F_A+g\log\!\left(1+\sqrt{8gU/\pi}\right),
 \qquad W=2+H_A+g\log(2g).
$$
For every positive multiple $m$ of three,
$$
 \widehat\mu_{\max}\ \le\ \tfrac12\left(\sum_{j=1}^g j^{-1}-1\right)+
 \log\!\bigl(\sqrt{e/3}\,m^2\bigr)
       +\frac{6}{m^2}\left(F_A+g\log\!\left(m+\sqrt{8gU/\pi}\right)\right).
 \tag{21}
$$
In particular $\widehat\mu_{\max}\le\log W+\tfrac12\log g+C$ with an absolute constant $C$.

**Proof.** Work after a finite extension with a free Moret-Bailly generator $s$. On a further model marking $A[m]$, independence of the direct image preserves this generator [5, Theorem 3.4]. For each $x\in A[m]$, the symmetric cubical multiplication isomorphism identifies $L_x^{m^2}$ integrally with the rigidified origin line. After a further finite extension choose an $m^2$-th root of the origin frame in $L_x$. Its coefficient in every integral frame has valuation zero and its norm at every complex embedding is one.

Write $B(u,v)=L_{u+v}\otimes L_u^{-1}\otimes L_v^{-1}$ using the rigidification. The cubical biextension identifies $B([m]z,x)=B(z,[m]x)=1$. Consequently the normalized torsion frame and $[m]^*L=L^{m^2}$ identify the pullback of $s$ by $z\mapsto[m]z+x$ with an integral section $s_x$ of $L^{m^2}$ on the marked model. Restrict to its identity component. At infinity this identification is isometric since its curvature agrees on both sides and its norm ratio at zero is one. Hence $\|s_x\|_{\sup,\sigma}=\|s\|_{\sup,\sigma}$. Under the deck translations of $[m]$, these sections belong to the characters paired with $x$ by the polarization Weil pairing. Indeed $t_x^*L\otimes L^{-1}$ is the dual $m$-torsion label $\lambda_L(x)$, and its pullback trivialization transforms by that character. The characters are distinct by principality, and the sections are nonzero. There are $R=m^{2g}=h^0(A,L^{m^2})$ of them by Riemann–Roch, so they form a basis over the complex numbers. The construction uses the section on the model containing the torsion sections. Moret-Bailly’s restriction theorem VI.3.4(i) concerns the open subgroup containing the polarization kernel of $L^2$, which includes $A[2]$ in this case. A primitive section on the connected model can require the finite scalar correction illustrated in §5.5. [5, §§2.3 and 3] [6] [12, I.5.5 and VI.3.4]

Fix a nonzero integral tangent vector $v$, and put $n=m^2$, $k=n/3$ and $\ell_\sigma=\|v\|_\sigma$. Along the formal curve in this direction at zero take the ordinary derivatives of orders $0,\ldots,k$ of the $s_x$. They give $k+1$ independent rows in their coefficient space. Indeed $L^3$ is very ample on the abelian generic fibre, so it has sections $u_0,u_1$ with respective orders zero and one in the given direction. The sections $u_0^{k-i}u_1^i$ of $L^n$ have exact orders $i$ for $0\le i\le k$. This proves the required complex jet ranks.

The exterior product of the derivative rows is a nonzero integral vector in the dual exterior power of the coefficient lattice. Locally at a finite place take an integral line frame near the identity agreeing with the rigidification there. The invariant derivation associated to $v$ preserves regular functions, and all its iterates evaluated at zero are integral. Changing normalized frames acts on the first $k+1$ rows by a triangular matrix with diagonal one, by the product rule. Thus their exterior product agrees with the one in a rational frame and with the one computed in each complex theta frame. This argument uses ordinary derivatives, including at primes dividing $k!$.

For a coefficient vector $c$, Cauchy–Schwarz bounds the point norm of $\sum_x c_xs_x$ by $\|c\|_2(\sum_x\|s([m]z+x)\|^2)^{1/2}$. The grid estimate in the proof of (20) bounds the latter square root by $B_\sigma=b_\sigma(m+\sqrt{8g/\pi}/r_\sigma)^g$. In a holomorphic theta frame normalized at zero the scalar reading on the complex line $zv$ is therefore bounded by $B_\sigma\|c\|_2\exp(n\ell_\sigma^2|z|^2/2)$. Here the factor $\pi$ is included in the tangent metric. Apply Cauchy's derivative estimate at radius $\sqrt{i/n}/\ell_\sigma$ for $i>0$ and use $i!\le i^i$. The $i$-th row has operator norm at most $B_\sigma(\sqrt{e n k}\,\ell_\sigma)^i$, with the bound for $i=0$ given by evaluation. Hadamard's inequality and the product formula for a nonzero integral minor give
$$
 -\operatorname{av}_\sigma\log\ell_\sigma
 \ \le\ \log\sqrt{e n k}
       +\frac2k\operatorname{av}_\sigma\log B_\sigma.
$$
All normalized averages descend under field extension. For each saturated sublattice $E$ of rank $j>0$ and every $\epsilon>0$, the absolute Siegel lemma supplies after finite extension a saturated line of normalized degree at least $\widehat\mu(E)-(\sum_{i=1}^j1/i-1)/2-\epsilon$. Principalize the line and take an integral generator. The left side of the displayed inequality is then its degree. Jensen's inequality and (20) give $\operatorname{av}_\sigma\log B_\sigma\le F_A+g\log(m+\sqrt{8gU/\pi})$ for our stable generator. Let $\epsilon$ decrease to zero and take the supremum over $E$ to obtain (21). [7k] [13, Theorem 3.1 and Lemma 3.2]

Finally take $m=3\lceil\sqrt W/3\rceil$, so $W\le m^2\le16W$. The logarithm in the last term of (21) is at most $\log m+\log(1+\sqrt{8gU/\pi})$. Since $W/g\ge\log(2g)$, the ratio $g\log W/W$ is bounded absolutely. Substitution in (21) with $\sum_{i=2}^g1/i\le\log g$ proves the last assertion. $\square$

**The scalar in the ordinary period.** On the marked elliptic factor of §5.3 the released column uses the even theta section $\vartheta$ with divisor $[H]$ and rigidified value $\vartheta(0)=1$. At every finite place over a good-reduction field the divisor misses the origin, by Corollary 7, so $\vartheta$ itself generates the stable line. Write $Q_\sigma=\|\vartheta\|_{2,\sigma}^2$. In particular $\operatorname{av}_\sigma\log Q_\sigma=2F_E$ for this particular elliptic curve $E$. In the field and differential of the release the same number can be read explicitly. Let $-d$ be the fundamental discriminant of the inducing quadratic field and $\chi_{-d}$ its character. Then
$$
 \operatorname{av}_\sigma\log Q_\sigma
 =\tfrac12\left(\log\sqrt d-\gamma+
                \frac{L'(1,\chi_{-d})}{L(1,\chi_{-d})}\right).
 \tag{22}
$$
Indeed, in discriminant-one periods of area $A_\sigma$ write $\theta_e$ for the marked even theta constant. Gaussian integration gives $Q_\sigma=(2\operatorname{Im}\tau_\sigma)^{-1/2}|\theta_e(0)|^{-2}$. The discriminant identity gives $A_\sigma=4\pi^2\operatorname{Im}\tau_\sigma|\eta(\tau_\sigma)|^4$. With $e_H=\wp(H)$ the normalized Weber value $X=((12e_H^2-g_2)/4)^{-3}$ has $|\sigma X|=|\theta_e(0)/\eta(\tau_\sigma)|^{12}$. It is integral by $(X+16)^3-jX=0$ and is a unit away from two. At a dyadic place the connected nonzero half-period has parameter of order $v(2)$ in an ordinary good model, by the linear unit term in $[2](z)/z$. Its $x$-coordinate has order $-2v(2)$ and the other two are integral. Differences of $x$-coordinates change by a unit on passing to discriminant-one coordinates. The formula $1=16\prod_{i<j}(e_i-e_j)^2$ gives valuation $2v(2)$ for the difference of the two integral roots. Our $H$ is one of this pair. Thus the product of its two differences, $(12e_H^2-g_2)/4$, is a unit at two as well. The field mean of $\log|\sigma X|$ vanishes. Substitution in the preceding norm and area identities, followed by the Chowla–Selberg mean of $\log A_\sigma$, gives (22). [7h, 7l]

In a step of ranks $a,b$ the squared Haar norm of the class-pair kernel uses $ab$ elliptic factors per CM embedding. The normalized intertwiners are isometries there and the class factors have complex modulus one at all conjugates. Thus the log norm contributes $fab$ times (22) after the conjugate average, where $f$ counts CM embeddings in the chosen type. Its upper bound on the selected host is the per-lift allowance in [7l].

**Proposition 12 (signed finite orders of the CM lattice changes).** Let $\Lambda_1,\Lambda_2$ be two commensurate self-dual period lattices with the same alternating form. Assume their symmetric theta multipliers agree on $C=\Lambda_1\cap\Lambda_2$, and identify the pulled-back lines in the same rigidification. For the normalized intertwiner $I_{12}$ of [7p], let $r_i(s)$ denote the order of a section in its stable local lattice at $w$, and put $b_w=r_2(I_{12}s)-r_1(s)$. Write $m_i=[\Lambda_i:C]$. Then
$$
 -\tfrac12 v_w(m_2)\ \le b_w\le\ \tfrac12 v_w(m_1),
 \qquad \sum_w\ell_w b_w=F_{A_1}-F_{A_2}.
 \tag{23}
$$
For each lattice change in the CM class-pair construction, including its four-copy comparison, one has the stronger equality
$$
                      \sum_{w\mid p}\ell_w b_w=0
                      \qquad\text{for every rational prime }p.
                      \tag{24}
$$
This includes the prescribed class factors with modulus one at every conjugate.

**Proof.** Put $X=V_{\mathbb C}/C$, where $V_{\mathbb C}$ is the common complex vector space, and $A_i=V_{\mathbb C}/\Lambda_i$. The isogenies $X\to A_i$ pull the two lines back to one algebraic line with its origin frame. The two deck groups $G_i=\Lambda_i/C$ have the same order by self-duality and pair perfectly by the commutator of their lifts. With $j_i$ denoting pullback and $D_i$ the signed deck action, the actual operator in these frames satisfies
$$
                  j_2 I_{12}s=m_2^{-1/2}\sum_{y\in G_2}D_2(y)j_1s .
$$
The stable finite norm $e^{-r_i(s)}$, in units $|\pi_w|=e^{-1}$, is the supremum of its cubical point norm on geometric torsion by Theorem 10a and its proof. Pullback preserves this supremum and the deck actions are isometries. The ultrametric inequality proves the lower bound in (23). In the reverse composite the sum over $G_1$ kills every nontrivial character of $G_2$ and the identity term has coefficient one. The same bound for $I_{21}=I_{12}^{-1}$ gives the upper bound. At a complex embedding the vectors $D_2(y)j_1s$ belong to distinct $G_1$-characters and are orthogonal for the Haar probability metric. Each has the norm of $s$. This proves isometry, also at each conjugate since Galois transports the perfect character pairing. Computing the normalized degree $-F_{A_i}$ on $s$ and $I_{12}s$ now gives the signed equality in (23). [5, 7p]

For the actual CM lattices the maximal order $\mathcal O$ of the auxiliary imaginary quadratic field acts through the same complex embedding on every Lie coordinate. Indeed the matrix periods are generated by multiplying matrices by the CM elliptic ideal lattice, and the four-copy trace isometry is linear for this action. Any such lattice of $\mathcal O$-rank $g$ is isomorphic to $\mathcal O^{g-1}\oplus\mathfrak a$ and this isomorphism extends complex linearly. The underlying abelian variety is thus a product of CM elliptic curves with maximal order. They all have the same stable Faltings height by ideal-class conjugacy. Additivity gives the same $F_A$ for all these lattices of fixed rank. [5, 7h, 7p]

To isolate one rational prime, split both deck groups into their primary factors. On the common cover replace the $p$-part of $G_1$ by the $p$-part of $G_2$, one prime at a time. Factors of coprime orders have trivial commutator. The mixed subgroup is maximal isotropic, its lifted action descends the symmetric line, and its inverse image in the period space is self-dual and $\mathcal O$-stable. Consequently its abelian variety has the same height. The operators for the successive replacements compose to $I_{12}$ because the normalized sums commute and their positive square roots multiply. Each replacement has a $p$-power index, so its individual $b_w$ vanish away from $p$ by (23). Its total over $p$ is zero by the degree equality, proving (24). Any additional algebraic scalar $c$ with modulus one at every conjugate has $|N_{K/\mathbb Q}c|=1$. The valuation of this rational norm at $p$ shows that its orders also have weighted sum zero above $p$. $\square$

**Theorem 13 (the finite-flat trace and the primitive CM section).** Let $q_i:X\to A_i$ be isogenies of abelian schemes over a mixed-characteristic DVR $R$. Identify $M=q_1^*L_1=q_2^*L_2$ in the prescribed rigidifications, for principal ample lines $L_i$. Assume addition identifies the two finite flat kernels $G_1,G_2$ with complementary Lagrangians in the kernel of the polarization of $M$. Write $n$ for the rank of $G_2$ and normalize $v_w(\pi)=1$. The raw $G_2$ deck sum between the two primitive lines has scalar ideal $(\det dq_2)$. Consequently (in compatible units after adjoining $\sqrt n$),
$$
 b_w(I_{12})=\operatorname{length}_R(e^*\Omega^1_{G_2/R})
                   -\tfrac12v_w(n). \tag{25}
$$
For the class-pair section of the released construction, both determinant corrections and the four-copy normalization cancel their respective local orders at each finite place. The resulting section generates the stable line in its matched rigidification.

**Proof.** The two pullbacks give deck linearizations of $M$. Their commutator identifies $G_1$ with the Cartier dual of $G_2$. Put $A=\mathcal O(G_2)$, $\epsilon:A\to R$ for augmentation and $J=\ker\epsilon$. The action of $G_1$ on $V=H^0(X,M)$ makes $V$ an $A$-module. In these coordinates $G_2$ acts by translations on $\operatorname{Spec} A$, by the commutator relation. This gives faithfully flat descent along $G_2\to\operatorname{Spec}R$, because $(y,b)\mapsto(y,y+b)$ identifies the action with the descent relation. Hence $V=A\otimes_R V^{G_2}$ with its translation action. Sections descend along the isogenies, so $V^{G_2}=q_2^*H^0(A_2,L_2)$ and $V^{G_1}=q_1^*H^0(A_1,L_1)$. The first is free of rank one and the second identifies, after choosing a basis of the first, with $\operatorname{Ann}_A J$. Over a splitting field the latter consists of functions supported at the identity. The deck sum takes such a function to the constant given by its value there. Its scalar ideal on the integral generators is therefore $\epsilon(\operatorname{Ann}_A J)$. [12, 17]

Complete $R$ and work in the factor of $A$ at the closed identity. On the other factors $J$ is the unit ideal. In coordinates on the two smooth formal groups this identity factor is $S/(F_1,\ldots,F_g)$, where $S=R[[x_1,\ldots,x_g]]$, $F_i=q_2^*y_i=\sum_j B_{ij}(x)x_j$. The $F_i$ and $\pi$ form a system of parameters by finite flatness, so the $F_i$ form a regular sequence. The augmentation ideal is generated by the $x_j$. Tensor the Koszul complexes on $F$ and $x$ with degree-one bases $u_i,v_j$. The product of the cycles $u_i-\sum_jB_{ij}v_j$ maps to the generator of top homology after resolving $S/(x)=R$. Resolving $S/(F)$ instead sends it to $\pm\det B$ times the top basis vector. There the top homology is exactly the annihilator of $(x)$, since there are no higher boundaries. Its augmentation is thus $(\det B(0))$. This is the determinant of $dq_2$, and its valuation is the length of the conormal module by the same linear presentation. Dividing the deck sum by $\sqrt n$ proves (25).

For the application, let $\mathcal O$ be the maximal order of the auxiliary imaginary quadratic field $k_0$, with its one active embedding. Every period lattice in the two matrix moves and in the four-copy comparison is a projective $\mathcal O$-module of this pure type. Write $C=\Lambda_1\cap\Lambda_2$ and $C^\#=\Lambda_1+\Lambda_2$ for the dual of $C$, using self-duality of the endpoints. Then $C^\#/C=(\Lambda_1/C)\oplus(\Lambda_2/C)$. To see this direct sum on the integral models, choose a field of good reduction for the auxiliary CM elliptic curve $\mathcal E$, whose period ideal is $J_0$. The model for a projective period module $P$ is $(J_0^{-1}\otimes P)\otimes_{\mathcal O}\mathcal E$. It is an abelian scheme by realizing the projective module by an idempotent. The kernel of the map from the model for $C$ to that for $P$ is the first $\operatorname{Tor}$ in fppf sheaves of $J_0^{-1}\otimes(P/C)$ with $\mathcal E$, computed by $C\hookrightarrow P$. Additivity for the displayed direct sum proves integral complementarity. The kernel for $C^\#$ is the polarization kernel, by extension of the generic dual identification over the good trait. The rigidified pullback isomorphism extends too. A possible vertical divisor on the smooth model is supported on its single special fibre and the rigidification fixes its order. This verifies the hypotheses even when the residue characteristic divides $n$. [7p, 12, 18]

Put $\mathfrak p=w|_{k_0}$, $e=e(w/\mathfrak p)$, $t_{i,\mathfrak p}=\operatorname{length}_{\mathcal O_{\mathfrak p}}(\Lambda_i/C)_{\mathfrak p}$ and $\Delta=\det_{\mathcal O}\Lambda_2/\det_{\mathcal O}\Lambda_1$. The Lie map on the Serre tensors comes from $C\hookrightarrow\Lambda_2$ in the active embedding. Its determinant has order $e t_{2,\mathfrak p}$. The perfect pairing exchanges $\mathfrak p$ and $\bar{\mathfrak p}$, so $t_{1,\mathfrak p}=t_{2,\bar{\mathfrak p}}$. At a split prime $v_w(n)=e(t_{2,\mathfrak p}+t_{2,\bar{\mathfrak p}})$. At an inert or ramified prime it is $2e t_{2,\mathfrak p}$, using the residue degree or ramification index in the quadratic field respectively. Since $\operatorname{ord}_{\mathfrak p}\Delta=t_{1,\mathfrak p}-t_{2,\mathfrak p}$, (25) becomes
$$
 b_w(I_{12})=-\tfrac{e}{2}\operatorname{ord}_{\mathfrak p}\Delta. \tag{26}
$$

Let $I,I'$ be the relative determinant ideals of the two Hermitian class lattices over the host $E$, each divided by the determinant of its standard lattice. Write $s=\operatorname{ord}_{\mathfrak p}N_{E/k_0}I$, $s'=\operatorname{ord}_{\mathfrak p}N_{E/k_0}I'$. The Hom lattice transforms by $Z\mapsto g'Zg^{-1}$ in ranks $a,b$. Its two determinant ratios over $k_0$ are $(N_{E/k_0}I)^{-b}$ and $(N_{E/k_0}I')^{a}$. The bare orders in (26) are $ebs/2$ and $-eas'/2$. For either norm-one ideal the character on a principal power $(u)=I^k$ has $\chi(I)^k=\iota(N_{E/k_0}u)^{-1}$, where $\iota$ is the active embedding. The two prescribed factors $\chi(I)^{b/2}$ and $\chi(I')^{-a/2}$ have the opposite orders. [7q]

In the four-copy move let $\Delta_4$ be the image determinant divided by the elliptic product determinant. At each split $p$ the chosen local standardizer takes the image to the elliptic product at $p$, hence its determinant order at the designated prime is $y_p=-\operatorname{ord}_{\mathfrak p}\Delta_4$. The correction in the release is the product of $r_p^{-y_p}$, where $r_p^{2h}=\pi_p/\bar\pi_p$ and $(\pi_p)=\mathfrak p^h$. Its order is $+e\,\operatorname{ord}_{\mathfrak p}\Delta_4/2$ on either side, with the sign changing between the two sides. At nonsplit primes the determinant order and the correction both vanish. The product thus cancels (26) for the four-copy move, including at the split prime two. The marked elliptic generator used in (22) is primitive, and its tensor product is primitive. Taking the indicated fourth tensor powers and the two class moves proves the assertion for the normalized section. [7p, 7q] $\square$

**Corollary 13a (content of the weighted induced section).** Fix a word, a step with source and target ranks $a,b$, and a class pair in the released packet. Let $\Lambda$ be its class period lattice and $A_c,B_c$ the two self-dual half lattices at level $c=c_B$, where $c_B=T(u)c_A$ and $c_A$ is supported on the row primes. At a selected host prime $Q$ of exponent $j_Q=v_Q(c)$ the local lattices $A_c,\Lambda,B_c$ are $q_Q^{-j_Q}P_Q,P_Q,q_Q^{j_Q}P_Q$, with $P_Q$ the local rank-$ab$ Hom lattice and $q_Q$ the rational prime norm. At $\bar Q$ the two halves are exchanged, and away from these primes all three agree. The two sets of primes are disjoint. Put $C'=A_c\cap B_c$, $G=A_c/C'$, $H=B_c/C'$, $U=(\Lambda\cap A_c)/C'$, $V=(\Lambda\cap B_c)/C'$ and $\mathcal Q=H/V$. Use also these names for the finite flat groups on the good Serre models after a finite splitting extension of complete DVRs. Put $n=\operatorname{rk}\mathcal Q$ and $d_{\mathcal Q}=\operatorname{length}_R e^*\Omega^1_{\mathcal Q/R}$. The orders below use the extended integer valuation and rescale by ramification. [7t, 20, 21]

In the common line $M$ on $A_{C'}$, write $W=H^0(A_{C'},M)$ and let $\operatorname{ord}_{W}$ be the order of a vector in this lattice, with order $+\infty$ at zero. Let $\Xi$ be the actual weight array in the coefficient embedding, using zero for labels outside the box. For the weighted section $F_{\rm div}$ with the class factor $\chi_F$ divided out, take $r_\Lambda=-v_w(\chi_F)=e(bs-as')/2$ from Theorem 13. Here $s,s'$ denote the determinant ideal orders there. If $\mathcal B_G$ is any integral basis of $\mathcal O(G)$, then
$$
 \operatorname{ord}_{W}(F_{\rm div})=
 r_\Lambda+d_{\mathcal Q}-v_w(n)
  +\min_{f\in\mathcal B_G}
     v_w\!\left(\sum_{x\in G(K_w)}\Xi(x)f(-x)\right).
 \tag{26a}
$$
At the word prime $p=T(u)$ and for a nonzero weighted section this specializes to
$$
 \operatorname{ord}_{W}(F_{\rm div})=r_\Lambda
  -a v_w(T(u))\,\#\{Q\in\mathrm{vT}(u):Q\cap\mathcal O_{k_0}=\mathfrak p\}
  +\sum_{Q\in\mathrm{rowSG}}\min_{x_Q}v_w(\Xi_Q(x_Q)).
 \tag{26b}
$$
Here $\mathrm{vT}(u)$ is the set of selected host primes over $T(u)$, $\mathfrak p$ is the active CM side at $w$ defined below, and $\mathrm{rowSG}$ is the set of row primes indexing the other local weights $\Xi_Q$. Each active word factor contributes exactly $-a v_w(T(u))$. [7t, 21]

**Proof.** The local half lattices give $\Lambda/C'=U\oplus V$. On $A_{C'}/V$ the kernels toward $A_\Lambda,A_{B_c}$ are $U,\mathcal Q$. Complementarity on both covers extends over $R$ by the additive fppf kernel construction in Theorem 13. The half lattices use the same signed semicharacter in the common origin frame. In the regular representation $W=\mathcal O(H)\otimes_R H^0(A_{B_c},L_{B_c})$, choose a primitive generator on the target. The $V$-invariants are $\mathcal O(\mathcal Q)$ times that generator. The divided class input identifies with $t\delta_0$ there, because invariance under the dual group $U$ forces generic support at the identity. Its trace scalar has order $r_\Lambda+d_{\mathcal Q}$ by the augmentation calculation of Theorem 13. As a function on $H$ this input has support $V$, so the raw deck sum over $H/V$ gives the constant $t$. In the released Fourier formula the coset indicator leaves the zero class of the inner transform and its pairing is one. The two inverse square roots of the volume cancel the numerator volume. The resulting coefficient is $1/n$. For the external lift of $x$ the operator is the signed deck lift of $-x$, which multiplies by the character $\langle-x,h\rangle$ on $H$. The coset vector is therefore $(t/n)\langle-x,h\rangle$. The CRT gives same-level lifts at $c_B$ since $c_A\mid c_B$, with lift independence and indicator refinement identifying the section on the original finer cover. Faithfully flat pullback of the rigidified line preserves section order. [7t, 20, 21]

By Cartier duality $\mathcal O(H)=\mathcal O(G)^\vee$, and the character just obtained is evaluation at $-x$ in this dual. Over the splitting field these characters are a basis. Over $R$ the matrix of moments against $\mathcal B_G$ gives the integral coordinates, proving (26a), including the zero array. At a split level prime the active side means the prime of $k_0$ inducing the action on the Lie line of the CM elliptic scheme. Its height-one $p$-divisible factor is multiplicative and the other is étale, by the two CM idempotents and duality. Accordingly the factor of $G$ at $Q$ on the active side is multiplicative and its dual factor of $\mathcal Q$ is étale. The latter has degree $q_Q^{abj_Q}$, giving a trace correction $-abj_Q v_w(p)$ at $p=q_Q$. On the other side the conormal order of the multiplicative factor of $\mathcal Q$ cancels its degree order. Coordinate bases are monomials on the multiplicative factors and delta functions on the split étale ones, so (26a) uses unnormalized Fourier moments on the former with the latter labels fixed. The released local weight factors multiply for fixed decorations, frames and shared terminal coset. Tensor product bases then let their minimum orders add. [7t, 20, 21]

At $p=T(u)$ one has $j_Q=1$ and the word weight is the indicator of the affine subspace $f'\bar x=f$ in $P_Q/pP_Q$, for the nonzero source and target decoration functionals $f,f'$. The map $\bar x\mapsto f'\bar x$ is onto the $a$-dimensional source dual, so the subspace has $p^{ab-a}$ elements. Its Fourier moments are zero off the annihilator of its direction and otherwise $p^{ab-a}$ times a root of unity, including the trivial character. This adds $(ab-a)v_w(p)$ to the active trace correction. On the étale side the minimum is zero. Row factors have prime-to-$p$ group orders at this place and keep their own coefficient orders. At a row place (26a) uses the actual local table. The entry factor includes the inverse volume, the penultimate factor is evaluated at the shared coset, and the terminal factor multiplies the test and its Fourier weight before taking moments. In particular the terminal twist factor contains $q_Q^{-\mathrm{cst.M}\,a}$ on its support, while the corresponding type scalar is one. [7t, 21]

For the divided section the formula gives its scalar in the integral cover line. To evaluate an ordinary derivative of multiindex $e$ add the orders of the specified Lie scale $\prod_\tau c_D(\tau)^{n_\tau}$, the factorial $e!$, and the Taylor coefficient of the primitive section in the actual Gaussian and Lie frame, with $n_\tau$ the degree at $\tau$. The local determinant-character phase of the fitted class reading is a further scalar. An evaluation or derivative of a primitive section can vanish on reduction. The corrected formal model, the marked parameters and line frames, the eventual contractions, and the finite bounds on the common row keep their separate hypotheses. [7q, 7t, 20, 21]

**Corollary 13b (determinant on the full cover).** With the same-level cover of Corollary 13a, take every generic label of $G$ and a class input of stable order $r_\Lambda$. The resulting fractional span of single-coset sections has determinant order $(\operatorname{rk}G)r_\Lambda$ in $\det W$. In particular the determinant generates $\det W$ for the primitive undivided class section. [23]

**Proof.** In the regular representation of Corollary 13a each column is point evaluation at the inverse label times a scalar of order $r_\Lambda-d_U$. Here $d_P=\operatorname{length}_R e^*\Omega^1_{P/R}$ for a finite flat kernel $P$. Indeed $U$ and $\mathcal Q$ are Cartier dual, so $d_U+d_{\mathcal Q}=v_w(n)$. Use the character basis on the split multiplicative factor of $G$ and point idempotents on its étale factor. Evaluation is a direct sum of Fourier matrices. A Fourier matrix of order $n_0$ multiplied by the transpose of its inverse-character matrix gives $n_0 I$. Inversion permutes rows, so its determinant has order $n_0v_w(n_0)/2$. Thus the evaluation determinant has order $(\operatorname{rk}G)d_G/2$. The active half-lattice exponents give $d_G=2d_U$, proving the formula. The weighted word test in (26b) is a specified affine fibre within this full family. Its selected period also uses the stabilizers, the marked jets and the retained column lattice. [23, 24]

The row has a separate positive determinant factor $\kappa_\sigma$. If $H_\kappa=\operatorname{av}_\sigma\log\kappa_\sigma$, dualizing its squared metric gives $h_i=h_i^{\rm rec}-H_\kappa$ in the column estimate. The comparison between the reference and fitted rows uses the same $\kappa_\sigma$ and bounds $\mu(C)-H_\kappa$ with finite main term $2F^{\rm ref}$. With the prescribed identification $F_o=2F^{\rm ref}$ the signed left side of [1, (6)] is exactly
$$
 (\mu(C)-H_\kappa)+h_i^{\rm rec}-2F^{\rm ref}.
$$
To identify the two occurrences of $F_o$, fix one fitted word $u$ and its reference tuple $t_0$. At a used tuple $t$, write $P_t$ for the quotient of the test space and $\iota_t$ for the isomorphism to the common row space. The period functional at $t_0$ vanishes on $\ker P_{t_0}$, so it induces a unique pairing $B_u$ on the quotient. The tuple comparison has the form
$$
 B_u(\phi,\iota_t P_t F)=\zeta_{u,t}\operatorname{Pt}_{u,t}(\phi,F),
 \qquad \zeta_{u,t_0}=1,\qquad v_w(\zeta_{u,t(w)})=0 .
 \tag{27}
$$
The unit in (27) follows from the fitted see-saw witnesses. Use the selected geometry with its class and level comparisons, terminal descent and link identities, the nonzero bottom anchor on the same restricted word column, and the primitive rational row and adjoint descent hypotheses. Write $m_0=2h_0$ for the number of steps and use the see-saw constants $c_{I,j}$ for the bottom links, $c_{K,j}$ for the top column steps, $c_{V,2i}$ for the even pair comparisons, and $c_P$ for the terminal comparison. Set
$$
 R_u(t)=
 \frac{c_P(t)\prod_{i=0}^{h_0-1}c_{V,2i}(t)c_{K,2i}(t)c_{K,2i+1}(t)}
      {\prod_{j=0}^{m_0-1}c_{I,j}(t)}.
 \tag{27a}
$$
The matched up-chain gives a nonzero scalar $s_u$ independent of $t$. In the two-term see-saw relation divide the terminal test coefficients by their respective bottom constants $\prod_j c_{I,j}$. The bottom tests then descend as the unscaled row forms, and the top terms have coefficients $s_uR_u(t)$ multiplying the tuple periods in the common column. Unfolding transports this relation to the bottom inner product with one common measure factor. Choose a reference lift and a column input with nonzero bottom pairing $b_u$, using the onto bottom and column identifications. Its corresponding top evaluation $p_u$ is nonzero, since a relation with that term alone would force $b_u=0$. For any second test take the two coefficients to cancel the top terms. The bottom relation then reads $s_uR_u(t)\operatorname{perCol}_{u,t}\,b_u=p_u b_t$, where $b_t$ is its bottom pairing. Normalizing by the reference cancels $p_u/b_u$, the fixed column norm change and the common row isometry. The coefficient in the fitted embedding $\sigma_0$ is thus $Z_u(t)=\operatorname{cj}_s(R_u(t)/R_u(t_0))$, with $\operatorname{cj}_s$ the specified identity or complex conjugation. The primitive row comparisons identify the complex pairing on the rational quotient. In detail the difference between the complex test and its rational quotient representative is orthogonal to the rational good row. Its image lies in the fitted row, whose adjoint preimages lie in the complex span of the rational good row, so the image of the difference is zero. On the reference tuple the fitted pairing therefore factors through $P_{t_0}$, which gives the stated definition of $B_u$. [7u, 22]

For the arithmetic support, let $I_{t,j}$ be the product of the actual extra quadratic level indices at step $j$. The bottom volume identity is $c_{I,j}(t) I_{t,j}=c_{I,j}(t_0)$, since the extra local subgroup has that index in the common Haar measure. In each even $c_{V,j}$ the accompanying level factor is $(I_{t,j}I_{t,j+1})^{-1}$, so division by the two bottom constants cancels both indices exactly. The remaining even-step ratios are the fitted scalar and character-power ratios. The column class factors telescope through the two matched determinant-character readings on each edge, and the remaining factors in $c_{K,j}$ and $c_P$ satisfy the fitted root or rational-power laws. Thus a common positive power $b_0$ of $R_u(t)/R_u(t_0)$ is a nonzero rational $q_t$ with numerator and denominator supported under the selected extra host primes $Q(t)$. Complex conjugation fixes $q_t$. [7u, 22]

The rational readings satisfy $\sigma_0(B_u(\phi,\iota_tP_tF))=Z_u(t)\sigma_0(\operatorname{Pt}_{u,t}(\phi,F))$. A nonzero tuple entry determines the field scalar $\zeta_{u,t}$ by the ratio of these entries. The fitted scalar choice uses the preimage of $Z_u(t)$ whenever it exists, also on an identically zero tuple. If it has no preimage, the equality of complex readings forces both pairings to vanish identically, and the choice is $1$. At the reference $Z_u(t_0)=1$. For $w\mid p$ the served tuple $t(w)$ has no prime of $Q(t(w))$ above $p$. Indeed each such prime would require ramification of the auxiliary level character there. Off the twisted places the reference character is trivial. On a twisted place the local correction is trivial on units over $p$ and the selected deletion character switches between the two choices with distinct exceptional rational primes to be unramified at $p$. This contradicts the required ramification. In the field-reading branch $\zeta_{u,t(w)}^{b_0}=q_{t(w)}$ by injectivity, and $v_w(q_{t(w)})=e(w/p)v_p(q_{t(w)})=0$. The fallback scalar also has order zero. This gives the local unit in the actual integral frame of (27). [7u, 22]

Gluing the actual images of the integral test lattices through $\iota_{t(w)}$ defines the row lattice. Thus the finite lower bounds for the class-pair derivatives test $B_u$ against this row after the specified fractional-ideal shift of its column. In the covolume calculation the fitted-cover short sections bound the saturated reference cusp lattices. The quotient Gram of a test form $H_t$ is $(P_tH_t^{-1}P_t^*)^{-1}$. If a lattice has determinant order $d_w$ in its chosen coordinates, its Gram for a form $H$ has order $v_w(\det H)+d_w+d_{\bar w}$. When the two relative indices and the duality index are added, the two orders of the intermediate lattice cancel, including at $w=\bar w$. This leaves the transported form on the saturated lattice and the quotient determinant. [7m, 7r]

**Corollary 13c (word-prime divisibility on the fitted period).** Fix the released word and its fitted tuple, on the same column and row as (27). Let $p=T(u)$ and let $S_T$ denote the selected places of the host above $p$, defined by the divisor of $\pi_T$ with $\pi_T\bar\pi_T=p$. Work at a coefficient place $w$ on the side that sees $\pi_T$. Put $e_w=v_w(p)$ with $v_w(\pi_w)=1$. Write $M_w^{\rm ram}$ for the completed integral source in the intersection of the scalar norm kernels. The fitted unshifted word lattice is $H_wM_w^{\rm ram}$. Here $H_w=p^{A_w}U_0^m$ is the scaled up-operator in the word comparison, $m$ is the number of lifts, and $U_0$ is the prescribed product of up-neighbour operators at stage zero. Let $\sigma_{uw}=-e_wm E_{\rm atT}(u,w)$ be the original integral shift, where $E_{\rm atT}$ is the released word-prime loss per lift. For a source pullback character $\chi(\mathbf c)=\prod_{P\in S_T}\omega(c_P)^{-j_P}$ of $\mathcal D=\prod_{P\in S_T}\mathbf F_p^\times$, put $J_\chi=\sum_Pj_P$, with $1\le j_P\le p-2$ on the retained source, and let $\omega$ be the Teichmüller character. For $x\in M_w^{\rm ram}[\chi]$ and every integral fitted row test $F$ at $t(w)$,
$$
 v_w B_u\bigl(\pi_w^{-\sigma_{uw}}H_w x,\iota_{t(w)}P_{t(w)}F\bigr)
   \ \ge\ m\left\lceil\frac{e_wJ_\chi}{p-1}\right\rceil .
 \tag{26c}
$$
This is pointwise on the fitted period model. Its inputs are the simultaneous moment and rational jet comparisons with the standardized class and companion lattices, the integral Weyl and up-neighbour comparisons, the prescribed terminal branching and row test ties, and the size hypotheses giving unit stabilizers. In the released model the latter follow by embedding the arithmetic stabilizers, including those of the coupled classes, into the unitary lattice groups and using the word-prime size bound. The selection of one family with these identifications, the reference covolume bounds and the global spectral and numerical comparisons are separate inputs. [7v–7x, 25, 26]

**Proof.** At a fixed target point of an edge, the scalar dilation $\mathbf c$ acts on the source by $f_P\mapsto c_P f_P$ in the same Weyl fibre. The row weight, frames, and literal coset $Y$ are fixed. The word test is $f'_P\bar X=c_P f_P$, and its simultaneous covariance at both endpoints follows by scaling this equality. For a source class orbit with arithmetic stabilizer order $s_d$ and diamond isotropy $H_d$, its contribution to the class sum is the sum over $\mathcal D$ divided by $s_d|H_d|$. Equivariance cancels the arithmetic transport between the kernel and the input, and the commuting diamond bijection preserves the stabilizer. Both divisors are $w$-units. This applies at nonfree intermediate orbits as well. Reindexing the class sum when the target is dilated shows that the incoming pullback character is also the output character. For the last two edges the intermediate point includes $Y$. The two kernels use the same $Y$, held fixed under dilation, and the final target is evaluated at the prescribed extension of the row point. Thus the reindexing applies to the full terminal column before that evaluation. [7v, 26]

To filter the actual jets, use the period lattice $M$ of the standardized pair and the companion lattice $M''$ of its fixed target, cut out on the selected side by the zero target functional. In the coset regrouping of the released word box, take the auxiliary zero word-labelled cells. Lattice CRT gives shifts $\Delta_P$ with word label $f_P$ at $P$, zero at the other word places and zero row residue, and with residue zero modulo $pqM$ on the conjugate side, where $q$ is the denominator clearing integer below. Adding $\Delta(\mathbf c)=\sum_P[c_P]\Delta_P$ carries the common regrouped sum to the actual cells of the orbit. The normalized Weyl operator $W_x=\mathrm{alV}(x)\mathrm{opU}(x)$ uses the full cocycle $\mathrm{coc}(x,y)=(\sqrt d\,\operatorname{Re}H(x,y)+\operatorname{Im}H(x,y))/2$ in its product, with $H$ the prescribed Hermitian form and $d$ the auxiliary discriminant parameter. Integrality of the cocycle on the half lattices allows the outer shift $W_{-\Delta(\mathbf c)}$ to act on the common sum of companion terms $A_y W_{t_y}e''$, with the same coefficients of order at least $-e_w|S_T|N(i)$ at edge $i$. The companion calculation clears each $t_y$ into $M''$ by an integer $q$ prime to $p$. For each base term the point-moment theorem chooses one $k=N_0\bar\pi_T/\pi_T$, with $p\nmid N_0$. CRT in $qM''$ gives $z_P$ such that $-\Delta_P-kz_P\in pqM''$. At the selected primes $k$ has order $-1$, and at their conjugates the deep zero condition gives the congruence. Since $pkz_P\in M''$, the discrepancy has integral product cocycle with the combined $kz_P$ shift and trivial commutator with $t_y$. It fixes the normalized companion section after commuting past $W_{t_y}$. The common base coefficient is therefore valid for all shifts. [7w, 25, 26]

For a fixed base term there is one integral point measure for all the shifts and all the tested jet indices. At a moment point $x'$ on the companion lattice the character sum multiplies its table by
$$
 \prod_{P\in S_T}\sum_{c\in\mathbf F_p^\times}
       \omega(c)^{-j_P}\zeta_p^{[c h_P(x')]} ,
$$
where $h_P(x')\in\mathbf F_p$ is the bracket exponent. The product has order at least $e_wJ_\chi/(p-1)$. Indeed for $j_P>0$ the sum is zero at zero frequency, and at a nonzero frequency the first nonzero binomial coefficient modulo $p$ in $\zeta_p-1$ is in degree $j_P$, by summing the monomials on $\mathbf F_p^\times$. Thus the order gain survives the integral approximation and the linear Weyl contraction on the common admitted support. The rational jet sum contains $\chi_F^{-1}$ once and is multiplied by the ratio of the two class-frame determinant factors. That ratio and $\chi_F$ are $w$-units in the standardized calculation. The differential multipliers $\lambda(c_D(\tau))/\Omega$ in the point table also have norm one. The same target determinant power and lowest-alcove integral comparison therefore apply with the additional factor. With the released clearing power $p^{E_i}$ for the operated step, this gives
$$
 p^{E_i}U_{i+1}K_i\bigl(S_i(w)[\chi]\bigr)
       \subseteq \pi_w^{b_\chi} S_{i+1}(w)[\chi],
 \qquad b_\chi=\left\lceil e_wJ_\chi/(p-1)\right\rceil .
$$
Here $K_i$ is the actual class-summed edge, $S_i(w)$ is its completed stage lattice and $U_{i+1}$ is the fitted up-neighbour product. The edge exponent before the up operators is $|S_T|N(i)+\sum_\tau C.sDiv(i,\tau)$. The first term clears the moment denominators and the second comes from the target determinant power. The operated exponent $E_i$ adds the first target weight entry at each place omitted from $U_{i+1}$. The ceilings are valid in these completed coefficient lattices because the rational maps and the integral character projectors are defined over the completion. [7w, 7x, 26]

Up-neighbour incidence commutes with diamond pullback. In the moved place a simultaneous dilation of the two marks is compensated by the inverse unit scaling of the neighbour vector. The rational up-operator relation is $U_{i+1}K_i=p^{sh}K_iU_i$, where $s=(t+1)/2$ for the released odd rank parameter $t$, and $h$ is the number of places in the product. Induction starting at the integral preimage $x$ gives $m b_\chi$ after clearing the terminal column of $H_w x$ by $p^{\sum_i E_i+sh\,m(m+1)/2-A_w}$. This exponent is at most $m E_{\rm atT}(u,w)$ by the released weight comparison. Exact terminal evaluation, integral branching and the unit stabilizers pass it to the signed period, and the unit in (27) gives (26c). On the opposite side the word indicator has disjoint supports for distinct source scalings, so a zero-sum difference there can have a unit coordinate. On each free stage-zero orbit, identify the integral fibres by transport from one representative. The value representation is then a sum of regular representations. The norm kernels remove precisely the characters trivial at one or more word places. Each of the $(p-2)^{|S_T|}$ remaining joint characters has rank $d_u/(p-2)^{|S_T|}$, where $d_u=\operatorname{rk}M_w^{\rm ram}$. The full-rank map $H_w$ carries this direct sum to the actual word image. The elementary divisors of the framed Fourier table in [23] agree with the single-factor character order with forward-label character $\omega^{j_P}$ corresponding to the pullback character $\omega^{-j_P}$. The proof of (26c) uses joint moment tables and the actual contractions to reach the period. [7x, 23, 26]

**Corollary 13d (slope gain from all lifts).** Use the normalized squared-metric degrees of [1, §2.1] over the coefficient field $K$, with $\ell_w=\log Nw/[K:\mathbb Q]$. Let $N_u$ be the original shifted column of [1, (4)], and take $C$ to be the dual of the glued row test lattice of (27). Suppose $B_u:N_u\to C$ is integral, including at the places outside (26c). Denote by $O_w$ the completed coefficient ring. At a seen place use the direct sum in Corollary 13c and set
$$
 \widetilde N_{u,w}=\bigoplus_\chi
        \pi_w^{-m b_{\chi,w}}\pi_w^{-\sigma_{uw}}H_wM_w^{\rm ram}[\chi],
 \qquad b_{\chi,w}=\left\lceil e_wJ_\chi/(p-1)\right\rceil .
$$
Use $N_{u,w}$ elsewhere. The resulting global lattice $\widetilde N_u$ has the same rank $d_u$ and the same Hermitian metrics on its generic vector space. The map to $C$ is integral. Its determinant order relative to $M_w^{\rm ram}$ at a seen place is
$$
 L_{uw}-d_u\sigma_{uw}-m\sum_{\chi}d_\chi b_{\chi,w},\qquad
 L_{uw}=\operatorname{length}_{O_w}(M_w^{\rm ram}/H_wM_w^{\rm ram}),\qquad
 d_\chi=\frac{d_u}{(p-2)^f},\quad f=|S_T|.
 \tag{28}
$$
In particular the normalized slope, and also the slope adjusted by subtracting the unchanged operator height bound $h_u$, increases by
$$
 \Delta_u=2m\sum_{w\ {\rm seen}}\ell_w\frac{\sum_\chi b_{\chi,w}}{(p-2)^f}
       \ \ge\ \frac{mf}{2}\log p .
 \tag{29}
$$
Here the last inequality uses the full selected half of the local degree. The original up-image colength $L_{uw}$ remains in (28). [1, 24–26]

**Proof.** Take a fractional-ideal multiple of $N_u$ containing the indicated enlargements and take the inverse image of the chosen submodules in the finite quotient by $N_u$. At each completion every summand pairs integrally with the same row generators by (26c). The local quotient by $N_{u,w}$ has $d_\chi$ elementary divisors of exponent $m b_{\chi,w}$ for each $\chi$. This gives (28) and the degree shift in (29). At the selected split word prime the two coprime factors $\pi_T,\bar\pi_T$ of $p$ are interchanged by conjugation on the conjugation-stable coefficient field. Their divisors partition the places over $p$ in the fitted embeddings. Consequently $\sum_{w\ {\rm seen}} e_w\ell_w=(\log p)/2$, since $e_w\ell_w=[K_w:\mathbb Q_p]\log p/[K:\mathbb Q]$. The equal ranks make each $j_P$ uniform on $\{1,\ldots,p-2\}$, of mean $(p-1)/2$. Inserting $b_{\chi,w}\ge e_wJ_\chi/(p-1)$ proves (29). [7v, 23–26]

**The tested frame bound.** Fix one translate, with a nonempty word set of positive ranks $d_i$, a common number $m$ of lifts, and $f\ge1$ selected host places for every word. Write $p_i=T_i$ and $L_T=\min_i\log T_i$. Use the frame data of [1, (7)] at a single embedding. Thus $A_i:\mathbb C^{d_i}\to\mathbb C^{D_{\rm all}}$, the projection $\Pi$ has rank $D_{\rm ret}=\operatorname{rk}C\ge\theta D_{\rm all}>0$ with $\theta=1-\delta_\alpha$, and the positive weights $c_i$ give $Q_A=\sum_i c_i A_i A_i^*$. The two bounds are
$$
 \sum_i\|c_i A_i^*A_i-I\|_{\rm HS}^2\le e_\alpha D_{\rm col},\qquad
 \operatorname{tr}(Q_A^2)\le(1+\eta_\alpha)(\operatorname{tr}Q_A)^2/D_{\rm all},
 \qquad D_{\rm col}=\sum_i d_i .
$$
Use $0\le e_\alpha\le1/4$, $\eta_\alpha\ge0$ and $0\le\delta_\alpha\le1/2$ as in [1]. Fix a single isomorphism $Y_{\rm row}:\operatorname{im}\Pi\to C_{\mathbb C}$. It suffices that every complex row test annihilating $(B_i)_{\mathbb C}$ also annihilates $Y_{\rm row}\Pi A_i$ for each word. The matrix comparison in [1, §3.2] implies this. In the reference section of (27) the test can also be verified on a $K$-basis of the annihilator. Lift through the prescribed rational right inverse of $P_{t_0}$, use the singleton see-saw with coefficient one and unfold at the common finite placement. When the bottom links and measure factors are nonzero and the bottom column identification is onto, this gives zero on the projected spectral image. The pairing is complex-linear in the transported test after the common conjugate-linear row isometry, so the basis tests suffice. This also covers a zero word map. The $Y$ held fixed in (26c) is the intermediate coset, distinct from $Y_{\rm row}$. [1, §§3.2–3.3] [27]

Let $\lambda_{ij}$ be the descending Harder–Narasimhan slopes of the **original** $N_i$, counted with ranks and shifted by $-h_i$. Here $h_i$ is an upper bound for the mean logarithm of the squared operator norm of the same rational $B_i$ with the same Hermitian metrics. Let $s$ be the mean of all these slopes, $d_0=D_{\rm col}^{-1}\sum_{i,j}(s-\lambda_{ij})_+$, and $H_*>0$ satisfy $s-\mu_{\min}(C)\le H_*$. Then, with $\xi=\eta_\alpha+e_\alpha/(1-e_\alpha)$,
$$
 \theta(s-\mu(C))\ \le\
 3d_0+\xi H_*-\frac{mf L_T}{6}(\theta-\xi)_+^2 .
 \tag{30}
$$
This estimate uses a one-digit bound. [1, (9a)] [24–27]

**Corollary 13e (joint characters).** Under the same integral and frame hypotheses, put $a=(\theta-\xi)_+$ and $w_i=d_i/D_{\rm col}$. Choose word $i$ with probability $w_i$ and then choose its $f$ digits $J_P$ independently and uniformly in $\{1,\ldots,p_i-2\}$. Set $X=(\sum_P J_P)/(f(p_i-1))$ and let $Q_X(u)=\inf\{x:\mathbb P(X\le x)\ge u\}$ for $0<u\le1$. The gain integrates the bottom $a/3$ of the rank distribution, counting a fraction of a cutoff atom when needed.
$$
 \theta(s-\mu(C))\le 3d_0+\xi H_*-\mathcal S_\alpha,\qquad
 \mathcal S_\alpha=3mfL_T\int_0^{a/3} Q_X(u)\,du .
 \tag{30a}
$$
In particular,
$$
 \begin{aligned}
 \sigma^2=\operatorname{Var}(X)
   &=\sum_i w_i\,\frac{p_i-3}{12f(p_i-1)},\\
 \frac{\mathcal S_\alpha}{mfL_T}
   &\ge \max\left\{\frac{a^2}{6},\ \frac a2-\sigma\sqrt{a(3-a)}\right\}.
 \end{aligned}
 \tag{30b}
$$
For $\theta\ge1/2$ and $\xi\le1/4$ the saving $\mathcal S_\alpha$ in the scaled row inequality is at least
$$
 \frac{mfL_T}{8}\left(1-\sqrt{\frac{11}{3f}}\right)_+ .
 \tag{30c}
$$
The digit law comes from the fitted column ranks in (28). The use of that law in (30a) requires the common spectral row tests, the two Gram bounds and the bound on the minimum slope of that row. [1, (9a)] [23–27]

**Proof of (30) and Corollary 13e.** At each $w$ list the exponents of $\widetilde N_{i,w}/N_{i,w}$ in descending order with multiplicities as $\beta_{iwj}$, using zeros where appropriate. Put $\Gamma_{ij}=2\sum_w\ell_w\beta_{iwj}$. Intersect the two lattices with a rational subspace of dimension $k$. The finite quotient embeds locally in the full quotient and has at most $k$ generators. Its length is at most the sum of the $k$ largest local exponents. The $a$-th layer, the subgroup killed by $\pi_w$ in the image of $\pi_w^{a-1}$, injects in the full layer and has dimension at most $k$. Summing gives the bound. With unchanged subspace metrics, this shows that the enlarged slope polygon is bounded above by the polygon with slopes $\lambda_{ij}+\Gamma_{ij}$. Indeed, at its vertices use the saturated filtration subspaces and the original polygon bound, then use concavity between vertices. The two polygons have equal endpoints at full rank by (28). For the enlarged adjusted slopes $\widetilde\lambda_{ij}$ it follows that
$$
 \sum_j(z-\widetilde\lambda_{ij})_+
 \le \sum_j(z-\lambda_{ij}-\Gamma_{ij})_+ .
 \tag{31}
$$
For descending slopes the deficit at $z$ is the maximum over $k$ of $(d_i-k)z$ minus the sum of the last $d_i-k$ slopes. The upper polygon with the same endpoint has smaller tail sums, giving the stated direction.

For a threshold $z$, the high row step consists of slopes at least $z$. The image of the enlarged high column step, taken at slope $z+h_i$, lies in it by integral height, as in [1, §3.2]. In fact an image in the low row quotient would induce a same-rank integral injection from a high quotient to a low subspace with operator height at most $h_i$, contradicting the determinant slope inequality. The row test implies $\operatorname{im}(Y_{\rm row}\Pi A_i)\subseteq\operatorname{im}(B_i)_{\mathbb C}$. The preimage of the high row step under $Y_{\rm row}\Pi A_i$ therefore has codimension at most the number of adjusted enlarged slopes strictly below $z$ for word $i$. Use the orthogonal projections onto these preimages in the centered trace argument of [1, §3.3]. If $q$ is the rank fraction of low column slopes, the discarded trace fraction is at most $q+\sqrt{e_\alpha q/(1-e_\alpha)}$. If $c$ is the ambient codimension fraction of the resulting image span, the row Gram bound gives $c-\sqrt{\eta_\alpha c}\le$ the discarded trace fraction. Thus $c\le3q+\xi$ by $2\sqrt{ab}\le a+b$. Projection of this span lies in the high row step and can only decrease codimension. This proves $\theta$ times the low row fraction is at most $3q+\xi$, including at ties. [1, (7), §3.3]

For $b\ge0$ integrate this threshold bound from $s-H_*$ to $s+b$. It counts every positive row deficit at $s+b$. By (31) and the positive-part triangle inequality the integrated column fraction is at most $d_0+\langle(b-\Gamma_{ij})_+\rangle$, where brackets average over all column dimensions. It follows that
$$
 \theta(s-\mu(C))\le
 3d_0+\xi H_*-(\theta-\xi)b+3\langle(b-\Gamma_{ij})_+\rangle.
$$
The descending exponent lists at the seen places are ordered by the same $J_\chi$, allowing arbitrary order within ties. Hence on each word the $\Gamma_{ij}$ dominate the list of $m(\log T_i)J_\chi/(T_i-1)$ with equal rank multiplicities. Put $b=mfL_T z$ with $0\le z\le1$. By subadditivity, the mean of $(b-\Gamma_{ij})_+$ is at most $mfL_T$ times the mean of $(z-J/(p_i-1))_+$ on an individual uniform digit $J\in\{1,\ldots,p_i-2\}$, averaged by word ranks. This last mean is at most $z^2/2$. Partition $[0,1]$ into $p_i-2$ equal intervals, take their midpoints, and contract the midpoints about $1/2$ by $(p_i-2)/(p_i-1)$ to get the digit positions. Jensen's inequality on each interval and on the contraction bounds the mean by the uniform integral. Taking $z=(\theta-\xi)_+/3$ proves (30). It uses the spread of the original slopes through $d_0\le\max_{i,j}\lambda_{ij}-s$, the quantity $u$ of [1]. [1, (9a)] [23–26]

For (30a), keep the full digit sum in the same sorted lower list for $\Gamma_{ij}$. Each of the $(p_i-2)^f$ digit tuples occurs with rank $d_i/(p_i-2)^f$, so the product law is exactly the rank distribution of the digits. The conditional probability of sum $k$ is the coefficient of $t^k$ in $(t+\cdots+t^{p_i-2})^f$ divided by $(p_i-2)^f$. Using $\log p_i\ge L_T$, the rank mean of $(b-\Gamma_{ij})_+$ for $b=mfL_T z$ is at most $mfL_T\,\mathbb E[(z-X)_+]$. When $a=0$, take $b=0$ to get (30a). For $a>0$, the gain available from the threshold bound is $mfL_T$ times the maximum of $az-3\mathbb E[(z-X)_+]$ over $0\le z\le1$.

For any function $v$ of $X$ with $0\le v\le1$ and $\mathbb E v=a/3$, the inequality $v(z-X)\le(z-X)_+$ bounds $az-3\mathbb E[(z-X)_+]$ above by $3\mathbb E(vX)$. Take $z=Q_X(a/3)$, let $v=1$ below $z$ and $v=0$ above $z$, and choose $v(z)$ to fill exactly the remaining mass $a/3-\mathbb P(X<z)$. This gives equality for every value of $X$. At this choice $\mathbb E(vX)=\int_0^{a/3}Q_X(u)\,du$. The threshold bound therefore gives (30a), including when several words contribute to the cutoff atom.

Conditionally on word $i$, each digit has mean $(p_i-1)/2$ and variance $((p_i-2)^2-1)/12$. Independence gives $\mathbb E(X\mid i)=1/2$ and $\operatorname{Var}(X\mid i)=(p_i-3)/(12f(p_i-1))$, including variance zero when $p_i=3$. Since all conditional means agree, averaging the conditional variances gives $\sigma^2$. For the optimizing $v$, the bounds $0\le v\le1$ give $\operatorname{Var}(v)\le(a/3)(1-a/3)$. Cauchy–Schwarz applied to the covariance gives $3\mathbb E(vX)\ge a/2-\sigma\sqrt{a(3-a)}$. The one-digit bound used for (30) shows that the maximum is also at least $a^2/6$, by evaluating at $z=a/3$. This proves (30b). If $\theta\ge1/2$ and $\xi\le1/4$, then $a\ge1/4$, $\sigma^2\le1/(12f)$ and $(3-a)/a\le11$. The variance bound is at least $(a/2)(1-\sqrt{11/(3f)})$, while the quantile gain is nonnegative. These inequalities imply (30c). [1, (9a)] [23–27]

For the signed application use $F_o$ from [1, (4), (6)] on the original columns and the same $C$ as in (27). With the original slope and signed height estimates [1, (5)–(6)], the lower bound in [1, §4] is $s-\mu(C)\ge G-M_\alpha-K_0-\langle\mathrm{Lo}\rangle_\alpha$, where $G$ is the common arithmetic gain, $M_\alpha$ the geometric main term, $K_0$ the common error and $\langle\mathrm{Lo}\rangle_\alpha$ the rank mean of the column losses. Inserting (30a) gives
$$
 G-M_\alpha\le K_0+\langle\mathrm{Lo}\rangle_\alpha+
           \frac{3d_0+\xi H_*-\mathcal S_\alpha}{\theta}.
 \tag{31a}
$$
The estimates of $d_0$ by the original spread and of $H_*$ by the row allowance thus enter on the original data. One may use (30) directly in this signed comparison by taking its saving $mfL_T a^2/6$. In (31a) the character gain enters through the threshold comparison with the enlarged columns. The signed height and the reference covolume retain their original inputs and their identification with the fitted row. [1, §§3–4]

The main term in the resulting signed row bound is $2\sum_{j,y\in{\rm Tw}}\ell_y e_y F^{\rm ref}_{jy}$. The reference coefficients vanish off the twisted set, the column coefficients vanish off their prescribed support, and the lift and place identifications carry one coefficient to the other on their union. They preserve $\ell_y$ and the ramification index $e_y$. The row sum thus equals the $F_o$ occurring in the column shift, including its factor two. At infinity the level comparison identifies the two row scales $\kappa_\sigma$ embedding by embedding. Dualizing the metric then subtracts $H_\kappa$ in the same pairing (27). This proves the displayed cancellation under the stated row and column comparisons. It uses the fitted theta section for the reference covolume and the matrix theta derivatives for the period functional. [7m, 7r]

**The projected reference row.** At a twisted site use a splitting DVR with $v(p)=1$ and length $\operatorname{len}_p$ divided by the integer order of $p$. Let $\mathcal T$ be the integral split Hecke order, $M$ the original lattice, $X$ its generic labels, $Z$ the labels with the prescribed base-change description and $A=Z^{\rm good}$ the retained labels. The congruence index is the length of $Q(M)=p_A M/(M\cap V_A)$, where $p_A$ projects to the retained generic subspace. For the finite group $\mathcal G=\prod_{k=0}^1\mathrm{GL}_r(\mathbf F_{g_k})^f$ use its central projectors $e_\tau$ for tuples of rational Lusztig series. Write $h=|\mathcal G|$, $m_{a\tau}=\dim e_\tau V_a$, $r_A=\dim V_A$, $r_B=\dim V_{X\setminus A}$, and let $A_\tau,B_\tau$ be the retained and complementary supports of $e_\tau$. The integral centre action and the localized section on the fitted forms give $he_\tau M\subset M$. On this module the sum of the good projectors fixes $V_A$ and kills $V_{Z\setminus A}$. In particular, if $A_\tau\ne\varnothing$, then $B_\tau$ consists of exterior labels in $X\setminus Z$ **within that tuple**. Put $c_{a\tau}=v(\operatorname{ev}_a\ker(\mathcal T\to O^{B_\tau}))$ and $d_{ab}=\min_{T\in\mathcal T}v(T(a)-T(b))$. The bound of [28, (17a)] is
$$
 \operatorname{len}_p Q(M)\ \le\ \min(r_A,r_B)v(h)+
 \sum_\tau\sum_{a\in A_\tau}\min\left\{m_{a\tau}c_{a\tau},
     \sum_{b\in B_\tau}\min(m_{a\tau},m_{b\tau})\,d_{ab}\right\}.
 \tag{32}
$$
Indeed, after splitting by the central projectors the kernel of the map on congruence quotients is killed by $h$ and requires at most $\min(r_A,r_B)$ generators, using projection to either side. Within a tuple, the retained filtration is killed by operators vanishing on the complementary support. Refining it by the complementary filtration gives pieces on which $T(a)-T(b)$ kills at most $\min(m_{a\tau},m_{b\tau})$ generators. Both methods bound the same retained graded piece, giving the minimum. [28]

In the released cut the lattice is $M^S=p_S M$, where the good support $S$ contains $A$ and can include exterior labels. For every tuple meeting $A$ its projection on the original and cut lattices agrees, since $e_\tau p_S=e_\tau$. Restricting the order to $S$ therefore preserves that tuple's depths and evaluated ideals. A bound on the full cut ideal, formed by vanishing on $(X\setminus Z)\cap S$, bounds $c_{a\tau}$ on each contributing original tuple. The direct estimate (32) charges the original denominator once. Using a bound for the entire cut quotient instead requires the return estimate $\operatorname{len}_p Q(M)\le\operatorname{len}_p Q(M^S)+r_Av(h)$. The cut bound on the evaluated ideal supplies no bound for the original full evaluated ideal on $X\setminus Z$. For example, when all the contributing $B_\tau$ are empty, (32) bounds the original index by $\min(r_A,r_B)v_p(h)$ alone. With the fitted rank bound at each place, its weighted contribution at $p$ in one lift is then at most $v_p(h)\log p$ per reference rank in length units. The residual same-type term in general requires bounds on the actual spectra. [28]

**Uniform bound for the length-$t$ labels.** Use the ordinary fitted CM host $E$ with $1<r<t$ and $t$ odd. Work at a served residual site with a retained label, on the good-type cut with the fitted normalized Hecke action, common absolutely irreducible rank-$r$ residual block and the released local–global comparisons. In this argument write $A_{\rm loc}$ for the original lattice and $M=E_{\rm good}A_{\rm loc}$ for its image under the sum $E_{\rm good}$ of the good central type projectors. Work in one contributing type summand of $M$ and write its generic space as $U\oplus W_s\oplus W_n$. Use $O$ for the splitting DVR. Here $U$ is retained, $W_s$ consists of the surviving exterior labels whose base-change parameter contains a part of length $t$, and $W_n$ consists of the other surviving exterior labels. A type with $U=0$ contributes zero. [7ac, 7af, 28]

Write $\kappa$ for the $p$-adic cyclotomic character of $G_E$, and $\chi_1$ for the common first Galois character of the retained string. The centered parameter decomposition and the common residual constituents, of ranks $r,1,\ldots,1$ with $t$ ones, force any length-$t$ part to have cuspidal rank one and to exhaust the residual lines. The remainder is one length-one term of rank $r$, with the common rank-$r$ residual representation. If the string comes from a Hecke character $\nu_b$, its first Galois character has the same norm shift $-r/2$ as $\chi_1$, which realizes the prescribed $\mu_*=\mu_i\eta_T^{-1}$ at site $i$. The normalized string on $b$ therefore has characters $\psi_b\kappa^{-u}$ for $0\le u<t$, where $\psi_b$ realizes $\nu_b/\mu_*$.

At a selected CM embedding the fitted infinitesimal weights are $v_a-\tfrac12$, $-b_a-\tfrac12$, and $-e_j-\tfrac12+(t-1)/2-u$ for $0\le u<t$. Here $v_a,-b_a$ are the interlaced rank-$r$ weights and $e_j$ is the prescribed midpoint parameter for the lift. The conjugate embedding has the negative list. The gaps $e_j-b_j$ and, when present, $b_{j+1}-e_j$ are at least $(t+3)/2$. Thus the displayed string is separated from the other weights by at least two and is the unique length-$t$ string at each embedding. Its midpoint agrees with that of $\mu_*$, whose finite tuple twist leaves the weights unchanged. Hence the ratio has infinity type zero, is finite order and has Hodge–Tate weight zero. Both parameters are conjugate self-dual, so $\psi_b^c=\psi_b^{-1}$ for complex conjugation $c$. Matching the residual line characters as multisets gives $\bar\psi_b\in\langle\bar\kappa\rangle$ by matching the first character, and $\bar\psi_b^{\,t}=1$ by taking products. Conjugation fixes $\bar\kappa$ and therefore gives $\bar\psi_b^{\,2}=1$. Odd $t$ yields $\bar\psi_b=1$, so $\psi_b$ has $p$-power order. If $\psi_b=1$, rank-one reciprocity identifies $\nu_b$ with $\mu_*$, so the parameter contains the prescribed part $\mu_*[t]$ and the label belongs to $Z$. Thus each $\psi_b$ on $W_s$ is nontrivial. [7ac, 32]

The fitted hyperspecial comparison gives unramified local parameters away from the exceptions, including at a place over $p$ outside the type primes. The exceptions are the tuple twist set $Q_i$, the active depletion prime $\mathrm{pde}$ (or $\mathrm{pde}'$ when $p=\mathrm{pde}$), and the two type primes, with all their indicated places in $E$. At the twist and active depletion places, all away from $p$, the inertia polynomial at $\gamma$ is $(X-1)^r(X-\chi_1(\gamma))^t$. Cyclotomic twists are unramified there, so the string scalar, which occurs at least $t$ times, must be $\chi_1(\gamma)$ since $r<t$. At a type prime away from $p$, the good-type comparison gives a squarefree rank-$r$ factor disjoint from the marked root of multiplicity $t$ at the tame generator. The inertia comparison in G5 identifies the marked root with the value of $\chi_1$, using inverted roots at the conjugate place. The $t$ string values must give this root. Both finite local characters are tame, so equality at that generator gives equality on inertia. At a type prime over $p$ the finite unit characters are again tame by the good-type comparison and the fitted character data. Their ratio on units has order dividing the residue cardinality minus one and is of $p$-power order, hence is trivial. Rank-one reciprocity, with the common infinity shift canceled, now factors $\psi_b$ through the ordinary class group $\mathrm{Cl}(E)$. [7ad, 32]

Take the marked tame inertia element $\gamma_0$ at the active depletion prime, a prime different from $p$. Then $\kappa(\gamma_0)=1$ and all $\psi_b(\gamma_0)=1$. The marked inertia proof compares the common residual polynomial with $(X-1)^r(X-\bar\chi_1(\gamma_0))^t$ and gives $\bar\chi_1(\gamma_0)\ne1$. Reduction preserves the prime-to-$p$ part of the order-$15$ inertia character. Canceling the string factor in the residual polynomial leaves $(X-1)^r$ before normalization by $\bar\chi_1(\gamma_0)$, so the normalized rank-$r$ block has no root $1$. Put $\widetilde\rho_b=\chi_1^{-1}\rho_b$ for the normalized Galois representation on a label. For each label in $U$ or $W_s$, divide its monic characteristic polynomial at $\gamma_0$ by $(X-1)^t$. The quotient $G_b\in O[X]$ is monic of degree $r$ and $G_b(1)$ is a unit. Thus
$$
 P(X)=\prod_{b\text{ in }U\text{ or }W_s}\frac{G_b(X)}{G_b(1)}
$$
is integral, $P(1)=1$, and Cayley–Hamilton gives zero on every indicated rank-$r$ block at $\gamma_0$. We can compute traces on the characteristic-zero direct sum of the parameter representations by Brauer–Nesbitt. [7ac, 7ae]

Define the tuple of traces $C(g)_b=\operatorname{tr}\bigl(P(\widetilde\rho_b(\gamma_0))\widetilde\rho_b(g)\bigr)$ for $g\in G_E$. If $P(X)=\sum_i a_i X^i$, this is $\sum_i a_i\chi_1(\gamma_0^i g)^{-1}\operatorname{tr}\rho_b(\gamma_0^i g)$. Each tuple of traces is minus the degree-one reverse characteristic-polynomial coefficient in G1. The released proof lifts these coefficients to the integral Hecke order for every group element by split Frobenius density and closedness of the order. The scalars in this sum belong to $O$, and the normalized local comparison transports the order action to the lattice and its type projection. Consequently $C(g)$ acts integrally on $M$. Its scalar on $U$ is $s(g)=\sum_{u=0}^{t-1}\kappa(g)^{-u}$, and its scalar on a label of $W_s$ is $\psi_b(g)s(g)$. [7af]

Put $A_E=\mathrm{Cl}(E)/(1+c)\mathrm{Cl}(E)$. Since $\psi_b^c=\psi_b^{-1}$, the value of $\psi_b$ on each product of an ideal class with its conjugate is one. Each $\psi_b$ on $W_s$ thus factors through the $p$-primary quotient $B=A_E[p^\infty]$ via the canonical primary projection. Class reciprocity gives a surjection $G_E\to B$. The image of $\ker\bar\kappa$ has index dividing $p-1$ in this $p$-group and is therefore all of $B$. Each coset of the kernel maps onto $B$ as well. For every $z\in B$ we may therefore choose a lift $g_z$ with any fixed cyclotomic residue in the image of $\bar\kappa$. [32]

If $p\nmid t$, choose residue $1$, giving $s(g_z)\equiv t\pmod p$, a unit. This includes $p=2$. If $p\mid t$, it is odd, and the actual fitted ranks satisfy $q,d>4rt+1>t$. The conductor-one host is $E=\mathbb Q(\zeta_{2^{e+2}})^+(\sqrt{-q},\sqrt{-d})$ for the fitted dyadic exponent $e$. Only $2,q,d$ can ramify, so $E$ is unramified at $p$. Since $\mathbb Q(\mu_p)$ is totally ramified at $p$, any common subfield has degree one. Restriction from $G_E$ to $\mathbb Q(\mu_p)$ therefore gives the full cyclotomic residue group. Choose residue $-1$ for every $g_z$. The alternating sum of $t$ terms gives $s(g_z)\equiv1\pmod p$. In both cases $s(g_z)$ is a unit of $O$. [7ah]

The integral Hecke action
$$
 D=\sum_{z\in B}\frac{C(g_z)}{s(g_z)}
$$
acts as $|B|$ on $U$ and as zero on $W_s$ by orthogonality of the nontrivial characters of $B$. Its action on $W_n$ is unrestricted. All choices and actions are local to the residual site and type summand, and the bound uses only finite-place valuations. [32]

Let $M_n=\operatorname{pr}_{U\oplus W_n}M$. The retained numerator $\operatorname{pr}_U M$ is unchanged by this projection. The difference of the two congruence lengths is the length of $(M_n\cap U)/(M\cap U)$, which is isomorphic by projection onto $U$ to $[M\cap(U\oplus W_s)]/[(M\cap U)+(M\cap W_s)]$. On $M\cap(U\oplus W_s)$ the operator $D$ is $|B|\operatorname{pr}_U$, so $|B|$ kills the quotient. Projection to either factor identifies it with a quotient of a lattice of rank at most the dimension of that factor. Thus the loss in normalized length is at most
$$
 \min(\dim U,\dim W_s)\,v_p(|A_E|).
 \tag{32a}
$$
The normalization divides integer length by $\operatorname{ord}_O(p)$, including after a finite splitting extension.

At each served coefficient place $w$ of a lift, the fitted multiplicity comparison bounds the sum of the retained ranks across the localized sites by the rank of the row space at $w$. The rank tie identifies that rank with the reference rank $D_0$. Splitting each retained space by types preserves its total dimension. The weights $e_w\ell_w=[K_w:\mathbb Q_p]\log p/[K:\mathbb Q]$ for the coefficient field $K$ sum over $w\mid p$ to $\log p$. Since $|B|$ is the $p$-part of $|A_E|$, (32a) costs at most $D_0\log|A_E|\le D_0\log h_E$ per lift, where $h_E=|\mathrm{Cl}(E)|$. This sum uses the retained rank once across all labels and tuples at each place. The original-to-type-projection cost is separately at most $D_0\log h$ per lift by (32), using the agreement of the contributing type summands on the original and released cut lattices. Thus, if $\mathcal I_j$ sums the weighted original congruence lengths at lift $j$ and $\mathcal I_j^n$ sums the corresponding lengths on $M_n$, then
$$
 0\le\mathcal I_j-\mathcal I_j^n\le D_0\bigl(\log h+\log|A_E|\bigr).
 \tag{32b}
$$
[7ag, 7ai, 28]

If $W_n=0$ at every served site, $\mathcal I_j^n=0$. On the fitted ordinary family $m\log h\le b$, and the length window bounds $t$ by a fixed power of the cutoff $x$. The host degree is $2f$ with $f<t$, and $\log|\operatorname{disc}E|\le 2f x^\varepsilon$. Minkowski bounds the norm of an integral ideal representing each class by $\sqrt{|\operatorname{disc}E|}$. Counting ideals up to this bound by the Euler product at $2$ gives $h_E\le|\operatorname{disc}E|\zeta(2)^{[E:\mathbb Q]}$. Hence $\log|A_E|\le\log h_E$ is bounded by a polynomial in $x$, while $V=x^\varepsilon(B_0+N_c)\ge fmbx^{2\varepsilon}$ with $b=e^{x^\varepsilon}$ and $B_0=fmbx^\varepsilon$. Here $f$ is the CM-type size, $m$ counts the lifts and $N_c\ge0$ is the other fitted size term in the division allowance. This yields the chain allowance $\sum_{j=1}^m\mathcal I_j\le C_eD_0mV$ of Paper III (27) on that subcase. Its use for [1, (6), (21)] retains the other fitted coverage, comparison and family hypotheses. In general the same allowance reduces to a uniform bound on $\sum_j\mathcal I_j^n$. [7ag, 32]

Theorems 3.1, 3.3–3.4 and 4.1–4.2 of [29] aggregate component jet degrees, descend slopes and assemble the signed covolume bound from stated inputs. For the spanning-norm clause of [29, Table 1, B4], the fitted short-section calculation supplies integral spanning vectors in the cusp lattices at the prescribed degrees, with norm at most $e^\Lambda$ at every embedding (see the supplement, *Short integral theta sections*, (7)–(9)). On identifying the target lattice of the descent with the corresponding fitted lattice in the same Petersson metric, [29, Theorem 3.3(1)] gives $\mu_{\min}\ge-2\Lambda$. The component ranks and height links, the jet-degree floor and the integral descent map with its loss bounds still require their own comparisons. The signed telescope uses the paired metric scales and $F_o=2\,\mathrm{Fo}$, with $\mathrm{Fo}$ the finite sum in [29]. Its reference covolume, row passage, lift-height estimates and error budgets are the hypotheses of [29, Theorems 4.1–4.2 and Table 1, B5]. [7m, 7r, 29]

Theorem 10a supplies the stable scalar ideal and the theta evaluation term on the matched model. Theorem 4b computes the boundary order on the original marked arc. These supply local data for the height calculation entering B4, whose component jet-degree inequality concerns the full fitted system. At a word prime, (28)–(29) measure the enlargement of the shifted period column. The parahoric class count and initial full-lattice slope estimates in [29, Proposition 2.1 and Theorem 2.3] concern the unshifted data. The character determinant enters the signed application through (30a)–(31a), with the common row and Gram hypotheses there. [7o, 23–27, 29]

**Component degrees on the fitted cover.** On a fixed good nonterminal cover $Y$, let $D\ge1$ be its flag dimension and $T\ge3D+5$ the prescribed finite jet order. At a component $Z$, write $\delta_Z>0$ for the image degree of the quotient-weight cusp system at $u_1$ and $h_Z$ for its Chow height divided by $(D+1)\delta_Z$ in the section-lattice metric. Let $\delta'_Z>0$ be the top degree at the boundary weight $u'=u_1-3r\kappa_{\rm H}$, where $r$ is the unitary rank and $\kappa_{\rm H}$ is zero on the positive block and minus one on the negative block. Write $\mathbb E_\delta,\mathbb E_\partial$ for the expectations with weights proportional to $\delta_Z,\delta'_Z$, and set $R_\partial=\sum_Z\delta'_Z/\sum_Z\delta_Z$. Use the fitted boundary volume with slope lower bound $s_Z$ at radius $c_s\ge0$, with $h_Z+c_s\ge0$ from the short generators. Require the carrying-section comparison into the cusp system at $c_s+c_c+\epsilon$ for every $\epsilon>0$, with $c_c\ge0$ including the norm and finite-order costs, and the Chow upper volume at that radius. [30, (1)]

Take the section-count limit on each component. The lower volume is at least $\delta'_Z(s_Z+c_s)_+$, and the upper volume is at most $\delta_Z(h_Z+c_s+c_c+\epsilon)$. The height floor makes the latter expression nonnegative before truncation. Let $\epsilon\downarrow0$ and sum. If $\mathbb E_\partial s_Z\ge-X$ with $X\ge0$, convexity gives
$$
 \mathbb E_\delta h_Z
 \ge R_\partial\mathbb E_\partial(s_Z+c_s)_+-c_s-c_c
 \ge R_\partial(c_s-X)_+-c_s-c_c .
 \tag{33a}
$$
Only one ratio of totals is used. Component-dependent carrying costs give the same inequalities with their $\delta$-weighted mean. [30, (1)]

Let $U_Y$ be the full cusp lattice at the auxiliary weight $u_-$, with component ranks $r_Z$ and ordinary normalized degrees $d_Z$ in the component probability metrics. Put $Q_Z=\binom{T-1}{D}\delta_Z$, $S=\sum_Z r_Z$ and $\gamma=\sum_Z Q_Z/S$. The rank and component splitting comparisons give $\operatorname{rk}U_Y=S$, $0<\gamma\le1$ and $\mu(U_Y)\ge2\sum_Z d_Z/S$. In the degree comparison the full squared norm on each component is its probability squared norm times the component mass, which is at most one. The finite-degree determinant estimate gives $d_Z\ge Q_Z(Th_Z-\mathcal E_Z)$ on this same lattice. Summing it and using (33a) gives
$$
 \mu(U_Y)\ge 2\gamma\bigl(T[R_\partial(c_s-X)_+-c_s-c_c]
                           -\mathbb E_\delta\mathcal E_Z\bigr).
 \tag{33b}
$$
The factor two converts ordinary normalized degrees to the squared-metric slope. [30, (2)]

In the tower comparison the label $a=\kappa Z$ has volume $v_a$ and $\delta'_Z=v_a G$, with $G>0$ the common flag degree at $u'$. The normalized arithmetic intersection $J_a$ of the $D+1$ boundary factors gives $s_Z=J_a/((D+1)v_aG)-b_{\rm tr}$, with the common trace cost $b_{\rm tr}=2\,\mathrm{costTwo}\,\|u'\|_1$. Multiplying by $\delta'_Z$ cancels $v_a$ in the denominator of the intersection term and gives
$$
 \mathbb E_\partial s_Z=\frac{\sum_a J_a}{(D+1)G\sum_a v_a}-b_{\rm tr}.
 \tag{33c}
$$
The total intersection lower bound and the trace-cost bound at this weight supply $X$ on the selected tower. The identity uses the volume of each component and holds for unequal volumes as well. An application on such a tower requires the corresponding bound for its total intersection. [30, (3)]

At the released scale let $L=x^\varepsilon$, $b=e^L$ and $E_3=fmbL\log\log x\ge b$, with $f$ the CM-type size and $m$ the number of lifts. The fitted degree comparison gives $R_\partial\ge1-\rho_B$ with $0\le\rho_B<1$. Thus (33a) implies $\mathbb E_\delta h_Z\ge-X-c_c-\rho_B c_s$. Use $\mathcal E_Z\le250E_3$ from the finite-degree calculation and the fitted budgets $TX,Tc_c,c_s(6D+10+T(z+2\rho_B))\le C_H E_3$, where $z=7(D^2/T+D\rho_B)\le1/2$. Since $z+2\rho_B\ge(7D+2)\rho_B$, we obtain
$$
 T\rho_Bc_s\le\frac{C_HE_3}{7D+2},\qquad
 \mu(U_Y)\ge -2\left[\left(2+\frac1{7D+2}\right)C_H+250\right]E_3 .
 \tag{33d}
$$
Here $\gamma\le1$ is applied after bounding the bracket in (33b) below by a nonpositive number. This replaces the coefficient $2(3C_H+250)$ in the earlier cover estimate and removes the $Tzc_s$ loss in averaging. The finite-degree floor retains the rank shares $(1-z)r_Z\le Q_Z\le r_Z$, the adjoint section and multiplication comparisons in the integral frames, and the component flat transfer to $u_-$. This alternative to [29, Theorem 3.1] uses that floor and the matched volume comparisons, and supplies the cover slope before flat twisting and descent. The signed row comparison and the uniform family hypotheses below still apply. [30, (4)]

**Corollary (signed finite carrying cost).** Use the selected quotient $u_1$ on the good cover of lift $1\le j<m$ in (33a), with its boundary weight $u'$, its existing carrying section and its matched ordinary and Hodge–Tate tests over the fitted Galois CM host $E$. Put $p_j=k+j-1$, $q_j=m-j$, and $r=p_j+q_j$. For the fitted positive boundary gap $g$ use the release cutoff $B=4fr^2(1+\max_{\tau,a}|u'(\tau,a)|/g)$. Let $S_B$ be the rational primes $\ell\le B$ together with those dividing $2|\operatorname{disc}E|M'$, where $M'$ is the level modulus in the mixed tests. Write
$$
 D_\ell=\frac{v_\ell(|\operatorname{disc}E|)}{[E:\mathbb Q]},\qquad
 \beta_\ell=\frac1{\ell-1}+D_\ell,\qquad
 N_j^-=\sum_{\tau\in\Psi}\sum_{a=k-1}^{p_j-1} -u'(\tau,a).
$$
Here $\Psi$ is the chosen CM type and the coordinates are numbered from zero. The log-radius allowance in (33a) can use $c_c=c_{\rm mult}+c_-$, with $c_{\rm mult}$ the original boundary multiplier cost and
$$
 c_-=N_j^-\sum_{\ell\in S_B}\beta_\ell\log\ell
 \le \frac{j}{m}\,c_{\rm ht},\qquad
 c_{\rm ht}=\|u'\|_1\sum_{\ell\in S_B}\left(\frac1{\ell-1}+40 D_\ell\right)\log\ell .
 \tag{33e}
$$
The original absolute-value allowance $c_{\rm ht}$ uses the same lift and cutoff, with $\|u'\|_1=\sum_{\tau,a}|u'(\tau,a)|$. [7y, 7z, 7aa]

**Proof.** The divided weight is anchored at $a_0=k-2$. This coordinate is in the large block, so the shift by $-3r\kappa_{\rm H}$ leaves $u'(\tau,a_0)=0$. The fitted quotient gaps, including the wall gap after the shift, make $u'$ strictly decreasing. At scalar weight zero the large block reads the first $p_j$ entries in order. The small block reads the last $q_j$ entries in reversed order with opposite signs, hence all its exponents are positive. The only negative exponents of the rule are the $j$ entries $k-1,\ldots,p_j-1$ per embedding. They are the first $j$ entries of the length-$m$ negative tail of $u'$. Since the magnitudes increase, their mean is bounded by the full tail mean. Thus $N_j^-\le(j/m)\|u'\|_1$. [7y]

In the triangular change from the ordinary to the matched Hodge–Tate frame let $t_s$ be a nonscalar diagonal, and write $d_s=-\log|t_s|/\log|\ell|$ for the multiplicative absolute value of the test. The pinned diagonal bounds give $1\le |t_s|\le|\ell|^{-\beta_\ell}$ before the discriminant allowance is enlarged to $40D_\ell$. Thus $0\le d_s\le\beta_\ell$, including at $\ell=2$. In coefficient valuation units at a place $w\mid\ell$, write $e_{w/\ell}=v_w(\ell)$ and let $e_s$ be the signed exponent of the rule at label $s$. For a nonzero evaluation the frame character gives
$$
 V_{\rm ht}-V_{\rm ord}=-e_{w/\ell}\sum_s e_s d_s
       \le e_{w/\ell}\beta_\ell N_j^- .
 \tag{33f}
$$
The scalar exponent is zero. A zero evaluation stays zero in the paired frame. At tensor power $n=8h$, mixed integrality at a patched prime therefore gives $V_{\rm ord}\ge -n e_{w/\ell}\beta_\ell N_j^-$. At the other primes the ordinary tests already impose integrality. Multiply by the same boundary section to power $h$ and use its balanced scalar together with the rational clearing factor $\prod_{\ell\in S_B}\ell^{\lceil n\beta_\ell N_j^-\rceil}$. The factor supplies the missing order relative to the ordinary depth test and adds at most $n c_-+\sum_{\ell\in S_B}\log\ell$ to the log norm. The latter sum is fixed for this cover and is absorbed by $n\epsilon$ before the section-count limit. The map is injective by cancellation of the multiplier at a generic point of the component. This proves the shifted radius in (33a), and $\beta_\ell\le 1/(\ell-1)+40D_\ell$ gives (33e). [7z, 7aa]

On the fitted window, let $S$ and $\eta$ be the nonnegative cloud and node widths and $w_\tau$ the original weight at the anchor. The quotient floor for each of the first $j$ gaps after the anchor has divisor $T+D+1$ and numerator the anchored dividend gap plus $2$. The node bound, summed over those gaps, gives
$$
 TN_j^-\le \frac{T}{T+D+1}\,j\sum_{\tau\in\Psi}
          (w_\tau+\eta+S+r+j+1)
       \le (2+o(1))fjb .
$$
For the last step use $w_\tau\le2b+S+r$, $j+1\le r\le\eta$ and $\mathrm{PP}(S+\eta)\le100b$ with $\mathrm{PP}\ge(frL)^{12}$. The prime sum satisfies $\sum_{\ell\in S_B}\beta_\ell\log\ell\le(2+o(1))L$ under $1\le B\le132b$, $\log\operatorname{rd}_E\le L$, $\log|\operatorname{disc}E|\le4x^A$ for the fixed family exponent $A$, and at most fifteen level primes. Indeed the different terms sum to $\log\operatorname{rd}_E$. The released reciprocal-prime estimate up to $\max(B,x^{A+2})$ costs at most $L+O_A(\log x)$, and the remaining discriminant primes cost $O(x^{-2})$. Remaining level primes each cost at most one. Thus at fixed $\varepsilon>0$ the new allowance satisfies
$$
 \frac{Tc_-}{E_3}\le (4+o(1))\frac{j}{m\log\log x}.
$$
The released bounds $T\|u'\|_1\le834fmb$ and $c_{\rm ht}/\|u'\|_1\le100L$ already give $T c_{\rm ht}/E_3\le83400/\log\log x$, so the original allowance is $o(E_3/T)$. The boundary multiplier, the upstream trace and intersection bounds, the component determinant and the signed row comparisons retain their separate costs. [7aa, 7ab]

**Proposition 14 (finite rule valuation on the ordinary Lie flag).** Fix a selected good nonterminal cover, a component of its marked abelian open over its coefficient field $F$, and a scalar-zero weight $\lambda$ of the component determinant. Use the selected characteristic-zero toroidal flag, with smooth base, normal crossings boundary, fitted uniformizations and the trivial-character reading. Write $V_\lambda$ for the supported cusp rules. On this flag their evaluations are exactly the sections of the weight line vanishing on the reduced horizontal boundary. This uses the characteristic-zero comparison on the selected cover, including the host degree condition $[E:\mathbb Q]/2>2$. [4, 7, 31]

At a finite place $v$, take a normal proper marked component over $R=\mathcal O_{F,v}$ carrying the connected semiabelian group $G$ and the prescribed generic rigidified polarized identification with CM action. Let $\pi$ generate the maximal ideal of $R$. In the product of ordinary relative flag schemes of $\operatorname{Lie}(G)$, take the normalized closure $X$ of the generic eigenflags with **all** full labelled eigenspaces as their top steps. Let $P_{\tau,i}$ be its tautological graded lines and $L_\lambda=\bigotimes_{\tau,i}P_{\tau,i}^{-e_{\tau,i}}$, where a graded basis change by $a_{\tau,i}$ multiplies the rule value by $\prod_{\tau,i}a_{\tau,i}^{e_{\tau,i}}$. The conjugate block uses the negative reversed exponents. For each prime component $C$ of the special fibre put $m_C=\operatorname{ord}_C(\pi)$ and $a_C(s)=\operatorname{ord}_C(s)$ in the line $L_\lambda$. For an actual framed test $y$ over a complete DVR $A$ of finite residue field, put $e_y=v_A(\pi)$ with integer-normalized $v_A$. Then
$$
 q_v(s):=\inf_y \frac{v_A(\operatorname{ev}_y s)}{e_y}
       =\min_C\frac{a_C(s)}{m_C}.
 \tag{34}
$$
For each $C$ there is one admitted framed test $y_C$ realizing $a_C(s)/m_C$ simultaneously for every $s\in V_\lambda$. The tests may be chosen for any fixed finite list of the scalar-zero weights on this flag. In particular a finite set of tests attains (34) for every rule. The zero rule has valuation $+\infty$. The unit ball is $V_\lambda\cap H^0(X,L_\lambda)$ in the generic section space. [31]

**Proof.** The characteristic-zero extension follows from cusp decay on transverse boundary arcs and normality. Conversely a section vanishing on the reduced boundary reconstructs a natural rule on the flagged open, with zero on the other components. The fitted Fourier–Jacobi test reads its Laurent series along the degenerating isotropic flow in frames related by units. The series has positive order, and Koecher nonnegativity of the frequencies gives zero Fourier–Jacobi constant term. The fitted growth estimate then gives the bounded cusp reading. On the normalized graph between the two generic flag models, extension of semiabelian homomorphisms identifies the Lie bundles with the given generic identification. The steps of both flags agree as subbundles by separatedness of the relative flag schemes. The pulled-back section is regular there and descends as a regular section on $X_F$ by normality and proper birationality. [4, Proposition 3.3.1.5] [31]

For any admitted test on the component, properness extends its generic flag to $X$. The test and model groups agree over $A$ by extending the generic isomorphism and its inverse. Every pulled-back step is saturated in $\operatorname{Lie}(G)_A$, so it equals the intersection with its generic subspace. In particular a full top intersects in exactly the test's integral eigenmodule. The defining eigenrelations hold in the torsion-free Lie module as soon as they hold after inverting $\pi$. The test bases and bases adapted to the pulled-back steps therefore agree up to triangular changes with unit diagonals. The theta generator is required for admission and has exponent zero. Thus $v_A(\operatorname{ev}_y s)$ is the order of the ordinary line coefficient. The argument uses each label separately and allows coincident eigenvalues on reduction. [31]

For $s\ne0$, write the right side of (34) as $a/b$ with integers $a,b$ and $b>0$. The rational section $s^{\otimes b}/\pi^a$ has nonnegative orders at all height-one points, by generic regularity and the inequalities $b a_C(s)\ge a m_C$. Normality gives regularity on $X$. Pullback gives the lower bound at every test, with off-component evaluations zero by support. [31]

For simultaneous attainment fix $C$. Let $k_v=R/(\pi)$ and write
$$
 U_n=\{s\in V_\lambda:a_C(s)\ge n\},\qquad W_n=U_n/U_{n+1}.
$$
Multiplication by $\pi$ identifies the layers at $n$ and $n+m_C$. Each $W_n$ is a vector space over the finite field $k_v$, since $\pi U_n\subseteq U_{n+1}$. Its dimension is at most $\dim_F V_\lambda$. Indeed, after dividing any relation between lifts of independent residues by the common least power of $\pi$ in its coefficients, reduction in $W_n$ would give a nonzero relation among the residues.

Work near the generic point of $C$ where $X$ and the reduced $C$ are regular, $C$ has equation $t$, and $\pi=u t^{m_C}$ with $u$ a unit. Choose frames for the finitely many weight lines and remove the other vertical components and the closure of the complement of the marked generic open. A horizontal prime of positive codimension cannot specialize to the height-one generic point of $C$. In the chosen frame the coefficient of $s/t^n$ for $s\in U_n$ is regular throughout this open by normality. Its order is nonnegative at $C$, and $t$ is a unit along every horizontal prime, where $s$ is regular. Restriction to the reduced $C$ embeds $W_n$ into its function field. For $0\le n<m_C$, exclude the zeros of every nonzero residue in each layer, for all weights in the list. There are finitely many such residues because $k_v$ is finite. The remaining open is nonempty and any closed point $P$ there gives injective specialization of the layers. In particular $[\kappa(P):k_v]\ge\dim_{k_v}W_n$ for every tested layer.

Extend $t$ to regular parameters at $P$, quotient by the others and complete. The resulting DVR $A$ has residue field $\kappa(P)$, $v_A(\pi)=m_C$ and its generic flag in the marked open. For a section of order $n$ in the tested range, the coefficient of $s/t^n$ pulls back to a unit by the residue injection. Scaling by powers of $\pi$ treats every integer order. Admission uses the fitted frame existence theorem on positive-rank data after finite extension at the same marked abelian point. Retain its theta generator and choose bases adapted to the extended flag by lifting generators of the free graded lines. The new field-valued frame has the required test-ring presentation. The residue field stays finite, and ramification multiplies numerator and denominator of the normalized order by the same index. Thus $y_C$ has normalized value $a_C(s)/m_C$ for every section in the chosen spaces. [31]

The tests give finite scalar constraints for the valuation and the unit ball. Let $K_C$ be the fraction field of the admitted test DVR, a finite extension of $F_v$, and let $e_C=v_{K_C}(\pi)$. In the basis $\varpi_C^i u_b$ with $0\le i<e_C$, $\varpi_C$ a test uniformizer and the $u_b$ lifts of a residue basis over $k_v$, let $z_{C,i,b}$ be the coordinates of $\operatorname{ev}_{y_C}s$. Write $v$ for the valuation on $F_v$ with $v(\pi)=1$. Then
$$
 q_v(s)=\min_{C,i,b}\left(v(z_{C,i,b})+\frac{i}{e_C}\right).
 \tag{34a}
$$
Indeed, within a fixed $i$ the least order survives by residue independence after removing the common power of $\pi$. The orders for different $i$ are distinct modulo $e_C$. This gives independence of the proposed basis, and its cardinality is the local field degree. The coordinate maps form a finite matrix over $F_v$. All weights $i/e_C$ lie in $[0,1)$, so ordinary integrality of the unweighted coordinates computes the unit ball. The weights record the fractional valuation.

In a basis of the unit ball $\Lambda=V_\lambda\cap H^0(X,L_\lambda)$ the matrix is integral over $\widehat R$. Properness gives finite generation, and clearing the vertical orders of a field basis makes $\Lambda$ a full lattice. On primitive coordinate vectors the minimum in (34a) lies in $[0,1)$ because $q_v(\pi s)=q_v(s)+1$. Replacing the matrix entries by representatives in $R$ modulo $\pi$ preserves this minimum, as all changes have weighted order at least one. The formula extends to the completion by the same reduction on primitive vectors. For any admitted test the difference from a lift of such a vector has normalized order at least one, since the basis lies in its integral ball. The chosen tests still attain the common minimum. The centres and residue degrees depend on the fixed ordinary flag and its actual cusp system at $v$.

**Sharpness at a pointed model.** Take $X=\mathbb P^1_R$ with marked generic open $\mathbb A^1_F$, line $\mathcal O(d)$ and the full degree-$d$ section space, with $d\ge0$. There is one fibre component with multiplicity one, and the layer at order zero has dimension $d+1$. At any centre of degree at most $d$ the residue map to the line fibre has a kernel. A nonzero leading residue in that kernel gives strictly higher normalized order on any trait through the centre, including after further extension. On the affine chart one can take the homogenized minimal polynomial of the centre, and at infinity the section $Y^d$ in coordinates $[X:Y]$ for $d\ge1$. A closed affine point of degree $d+1$ works since no nonzero polynomial of degree at most $d$ vanishes there. Lifting its monic irreducible polynomial over $k_v$ to $\widehat R$ gives the unramified detecting trait. The generic point is in the marked affine open. Thus $d+1$ is the sharp residue degree for a single centre in this example.

For a second integral line $L_\lambda(D_v)$ on a common normal model with vertical Cartier divisor $D_v=\sum_C d_C C$, its integral sections have $a_C(s)\ge-d_C$. Multiplication by $\pi^{b_v}$ for $b_v=\lceil\max_C d_C/m_C\rceil$ puts them in the actual unit ball. In the component determinant this identifies the ordinary finite lattice used for the integral coefficient wedges. For a full-rank comparison with matching Petersson metrics, the vertical correction gives a lower bound for the ordinary normalized determinant degree by the model degree minus $r_Z\sum_v b_v\log Nv/[F:\mathbb Q]$, when the comparisons have finite support. The patched Hodge–Tate and boundary-height lines require their own comparison in the same generic frame and the corresponding uniform finite costs. The carrying section, complex metric and rank bounds in (33a)–(33d) remain inputs. [30, 31]

One can compute the signed output-degree correction in this last comparison. Use the rows $B_\nu$ of the actual evaluations in (34a), before approximation, in a fixed generic basis of rank $r_Z$. For $\nu=(C,i,b)$ put $q_\nu=\lfloor d_C/m_C+i/e_C\rfloor$ and define
$$
 M_v(q)=\min_{\lvert I\rvert=r_Z,\,\det B_I\ne0}
             \left(v(\det B_I)+\sum_{\nu\in I}q_\nu\right).
 \tag{34b}
$$
The empty minor has value one. With the same generic metrics, the degree of the twisted lattice minus the degree of the ordinary one is $\sum_v (M_v(q)-M_v(0))\log Nv/[F:\mathbf Q]$. Indeed the order condition $a_C(s)\ge-d_C$ makes each row $\pi^{q_\nu}B_\nu$ integral on $s$. These inequalities cut out a full lattice over the completed DVR since all shifts are finite and the unshifted lattice is full. Its intersection with the generic $F$-space is dense, so it is the completion of the twisted lattice. Its dual is the span of the shifted rows, and the maximal minors generate that dual's determinant. Dualizing gives the stated sign. In a basis of the ordinary unit ball, when each $q_\nu$ is zero or one, the loss in local valuation units equals $r_Z$ minus the residue rank of the rows with $q_\nu=0$. The shifted row span contains $\pi$ times the ordinary dual and has precisely that residue image.

For example, blow up $\mathbf P^1_R$ at $x=\pi=0$, pull back $\mathcal O(d)$ with its full section space, and twist by $K C_0$ for $K\ge0$, where $C_0$ is the original special fibre's strict transform. The two vertical orders of $\sum_{i=0}^d a_i x^i$ are $\min_i v(a_i)$ and $\min_i(v(a_i)+i)$. For the second, set $x=\pi z$ and reduce at the generic point of the exceptional divisor, where $z$ is transcendental. The twisted lattice has basis $\pi^{-\min(K,i)}x^i$. Its local determinant correction is $d(d+1)/2$ for every $K\ge d$, even as the scalar allowance $(d+1)K$ increases. On the fitted cover the use of (34b) requires the divisor and the evaluations in the specified generic identification.

The height application in [1, (6) and (21)] still requires the following comparisons on one family at each cutoff, with constants fixed before taking the arbitrarily late power intervals.

1. The selected covers, level and coefficient descent, and the normalized class maps must be compatible with the actual integral columns and common row. The geometric rank comparisons and their hypotheses in [2] remain separate.
2. The actual arithmetic ideal on the matching arc over $o$ satisfies (7b), so its vertical and residual contributions in (3) vanish. For an independent model they require bounds for $c$ and $\lambda$ and the stated horizontal match. Applying (9) on a separately chosen semistable model requires its rigidified symmetric Raynaud line and the adapted integral coefficient lattice, including the scalar for its section.
3. For the prescribed class-pair section, the finite-flat trace, the weighted sum and Corollary 13c use the stated rigidifications and the same chosen jets and contractions. Passage to the integral map requires the exact terminal evaluation and stabilizer weights, quotient descent with the adjoint tie, the unit at the served tuple, and integral tests giving the localized generators of the actual row. Integrality at all other places retains the other marked jet and contraction bounds.
4. For the signed height use the fitted short sections, reference covolume and finite Gram and quotient comparisons, including (32)–(32b) on the indicated original or cut record and a uniform estimate for any remaining quotient from $W_n$. Passage to the true row needs its common principal-level representation, descended integral level embeddings and character averages, reference and true type multiplicities and determinant comparisons, and the reference maximal-slope cap and quantitative reweighting bound. The metric scales $\kappa_\sigma$ must agree at each embedding, with the quotient metric from minimum-norm lifts. The reference signed coefficient sum must match the original column shifts on their union of supports, with the same ramification and place weights. [7m, 7r, 28]
5. The frame consequence requires one spectral placement and the tested inclusion on the same common row for every word, the two Gram bounds, the retained rank bound and the minimum-row-slope allowance in (30a). Applying the signed comparison (31a) also requires the remaining weight, height and loss hypotheses there on the selected family. The determinant gain in (29) is already present for a single word satisfying the fitted period hypotheses. [1, 27]

For example, if the full fitted cotangent $\omega'$ embeds in the Néron cotangent $\omega_G$ with the same generic metric, then $h_F(A)=\widehat{\deg}_{\rm nor}\det\omega'+\sum_v\ell_v\,\operatorname{length}_v(\omega_G/\omega')$. This follows by taking determinants over each DVR. Passing to just the selected eigenspaces in the common row also requires the degrees of their complements.

The matching of marked models determines the horizontal order and the integral section used in the height. The character calculation adds the finite gain (29) on the actual period, and the joint character law gives the quantile saving in the signed comparison (31a). The remaining signed height and uniform family inputs enter through the comparisons above.

## References

1. bramble_trace. *Heights of integral period maps and one-sided prime bounds on power intervals*. rStation 194, version 1 (2026), especially (6), (7), (9a) and (21). [Published paper](https://rstation.ai/problems/riemann_hypothesis/papers/194/v1)
2. bramble_trace. *Primitive Hermitian densities, signed orbital sums and the rank of unitary period columns*. rStation 196, version 1 (2026). [Published paper](https://rstation.ai/problems/riemann_hypothesis/papers/196/v1)
3. K.-W. Lan. *Compactifications of PEL-type Shimura varieties in ramified characteristics*. Forum of Mathematics, Sigma **4** (2016), e1. [DOI](https://doi.org/10.1017/fms.2015.31)
4. K.-W. Lan. *Arithmetic Compactifications of PEL-Type Shimura Varieties*. London Mathematical Society Monographs 36, Princeton University Press, 2013. Numbering follows the author’s revision. [DOI](https://doi.org/10.23943/princeton/9780691156545.001.0001) · [Author’s revision](https://www.kwlan.org/articles/cpt-PEL-type-thesis-revision.pdf)
5. F. Pazuki. *Theta height and Faltings height*. Bulletin de la Société Mathématique de France **140** (2012), 19–49. [DOI](https://doi.org/10.24033/bsmf.2623)
6. D. Mumford. *On the equations defining abelian varieties. I*. Inventiones Mathematicae **1** (1966), 287–354. [DOI](https://doi.org/10.1007/BF01389737)
7. L. Alpöge. Released development, commit e7c8972c701d3396947808ab20cc16dc14623d33. [Public source](https://github.com/Deicyde/riemann-hypothesis/tree/e7c8972c701d3396947808ab20cc16dc14623d33)
   - **7a.** [Monomial marked arc](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/ReciprocityTheory/ReductiveGroups/ArithmeticGeometry/BirationalGeometry/ToroidalCompactification/UniformisationConclusion.lean#L83-L221) · [Fourier recurrence](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/ReciprocityTheory/ReductiveGroups/ArithmeticGeometry/ExtensionDescent/ThickTransComp.lean#L73-L178) · [Integral divisor certificate](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/IwasawaDecomposition/SysExOfPol.lean#L27-L59)
   - **7b.** [Powered value](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AutomorphicSpectrum/Definitions/GlobalGeometry/IsThetaMeanSquareNorm.lean#L17-L24) · [Torsion mean proof](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AutomorphicSpectrum/Proofs/MetricProofs/ExistsThetaSupBoundLogMean.lean#L190-L311) · [Linear coordinate](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/DegenerationProofs/ExistsRigidificationPullbackCompatibleIso.lean#L19-L27) · [Full list and symmetrization](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Definitions/Divisors/SymTupOne.lean#L18-L55)
   - **7c.** [Half-period normalization](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Formalism/LocalToGlobal/TorsionIntegrality/TorsionTranslationPowers/NonsingularJetNorms/Taylor0ConstMul.lean#L195-L265)
   - **7d.** [Fitted homogeneous theta reading](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/CuspFormEvaluations/ShortValTotOfPinned.lean#L62-L97)
   - **7e.** [Projective theta construction and its auxiliary order](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/DivisorProofs/UnitaryIntegralCompactificationExistsViaTheta.lean#L24-L142)
   - **7f.** [Adapted generator and torsion frame](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AutomorphicSpectrum/Proofs/GeometryProofs/TorsionValueIntegral.lean#L425-L514) · [Character and coefficient orders](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/UniformizationProofs/ValuationsTorusPointsFramesFamilyTerms.lean#L111-L219) · [Convergent evaluation](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/UniformizationProofs/SectionValueAtTorsionPointEqSumTerms.lean#L305-L363) · [Local square](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/DegenerationProofs/ExistsWeightZeroSectionUnitIntegralTorsionPoint.lean#L39-L63)
   - **7g.** [Analytic sampling inputs and supremum estimate](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AutomorphicSpectrum/Proofs/MetricProofs/ExistsThetaSupBoundLogMean.lean#L452-L555)
   - **7h.** [Auxiliary lattice and marked period](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/IwasawaDecomposition/ExistsCurveLattice.lean#L77-L143) · [Two CM conditions and theta normalization](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Formalism/SharpRestriction/GlobalSchurBounds/GL3Theta.lean#L198-L207) · [Elliptic product and fourth tensor](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Formalism/LocalToGlobal/KernelLattices/LateralTheory.lean#L20-L35) · [Finite-place comparison at 2](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/FoundationProofs/CmThetaFunctionsOverTwoUnitAndIntegral.lean#L415-L449)
   - **7i.** [Shortest period in a symplectic basis](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AutomorphicSpectrum/Proofs/MetricProofs/RealPartGramInSymplecticGroupStatement.lean#L21-L33) · [Gaussian bound in that basis](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AutomorphicSpectrum/Proofs/MetricProofs/ThetaFunctionEvenBound.lean#L261-L283)
   - **7j.** [Haar norm and existential section](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AutomorphicSpectrum/Definitions/GlobalGeometry/IsThetaMeanSquareNorm.lean#L56-L126) · [Metric on the full Lie lattice](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AutomorphicSpectrum/Proofs/GeometryProofs/HodgeMetricFrameInner.lean#L18-L54) · [Supremum and translated sections in the line proof](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AutomorphicSpectrum/Proofs/GeometryProofs/LineOfKeyDegree.lean#L269-L301) · [Slope on the abelian part of the selected point](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/WeilBound/CoreStatements/PlurisubharmonicCones/HeightComparisonGlobalModel.lean#L150-L169)
   - **7k.** [Integral jet ranks and the arithmetic flag](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AutomorphicSpectrum/Proofs/GeometryProofs/LineOfKeyDegree.lean#L196-L267) · [The vector of a sublattice after finite extension](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AutomorphicSpectrum/Proofs/AdelicProofs/ShortVectorOfSubmodule.lean#L249-L305) · [Passage to maximal slope](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AutomorphicSpectrum/Proofs/GeometryProofs/SlopeBoundOfLargeDegreeLine.lean#L94-L143)
   - **7l.** [Conjugate vacuum norms](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/FoundationProofs/VacuumNormConjCmPointsAllEmbeddings.lean#L211-L238) · [Weber power of a marked half-period](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/GlobalInterpolation/ExistsIntegralPeriodElement.lean#L168-L232) · [CM area and Haar norm](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/DecoratedLatticeProofs/ThetaZeroMeanLogNormConjugatesOfWeber.lean#L26-L103) · [Column height at the selected level](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/DivisorProofs/KernelHalfChainNumbersAnyLevelPrime.lean#L274-L330)
   - **7m.** [Row determinant quotient](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Formalism/LocalRingCardinality/DataKap.lean#L125-L176) · [Inverse metric in the operator bound](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Formalism/LocalToGlobal/Divisibility/ConstantGeTwentyEight.lean#L123-L151) · [Reference and fitted rows in the signed inequality](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Formalism/LocalToGlobal/Divisibility/ConstantGeTwentyEight.lean#L210-L245)
   - **7n.** [Trace-one integral Raynaud identity](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/ReciprocityTheory/ReductiveGroups/ArithmeticGeometry/BirationalGeometry/ToroidalCompactification/UniformisationConclusion.lean#L43-L109) · [Primitivity on a complete trait](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/DivisorProofs/ThetaConstantsPrimitivePrincipallyPolarisedTrait.lean#L98-L148) · [Integral symmetric values on a normal local base](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/DivisorProofs/ThetaSymmetricFunctionsIntegralUnitNormalLocalDo.lean#L153-L191)
   - **7o.** [Construction of the arithmetic ideal](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/DivisorProofs/UnitaryCompactificationExistsBoundaryIdeal.lean#L19-L38) · [Precise trait order](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Definitions/Divisors/ExistsOverIff.lean#L20-L31) · [Uniformizer specialization](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/DivisorProofs/ExistsPointArcDataAtUniformiserCompleteDVR.lean#L18-L43) · [Monomial chart calculation](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/DivisorProofs/IdealAlongMonomialFamilyEqSpanMonomial.lean#L58-L112)
   - **7p.** [CM period lattices and normalized lattice change](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Formalism/LocalToGlobal/KernelLattices/ColumnData.lean#L54-L121) · [Character sum and inverse](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Formalism/LocalToGlobal/TorsionIntegrality/QuadraticSemicharacters/QuadraticSemicharacterIdentityReformulated2.lean#L25-L34) · [Locally standard four-copy comparison](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Formalism/LocalToGlobal/InertKernelTheory/InertCoprimeLemmas/KappaSelfDualFourthPowerLattice.lean#L302-L367) · [Transport of the actual normalized class pair](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Formalism/LocalToGlobal/InertKernelTheory/InertCoprimeLemmas/IntAsm.lean#L55-L77)
   - **7q.** [Relative determinants, the four-copy scalar and the class characters](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Formalism/LocalToGlobal/KernelLattices/LateralTheory.lean#L88-L178) · [Character on principal powers](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Formalism/LocalToGlobal/GlobalTheory/GL3/NormNatCoprime.lean#L115-L129) · [Character in the descended derivatives](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/BlockTheory/DivisorBase/TopPlaceLogTheta.lean#L66-L72)
   - **7r.** [Quotient pairing](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Formalism/LocalToGlobal/IndefiniteReal/CutTuples2.lean#L157-L175) · [Signed determinant and height calculation](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Formalism/LocalToGlobal/Divisibility/ConstantGeTwentyEight.lean#L197-L248) · [Support and factor of two](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Formalism/LocalToGlobal/Divisibility/SmallTies.lean#L20-L89) · [Identification of the selected rows and matrices](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AutomorphicArithmetic/ThetaCorrespondence/ThetaPairLifts/ThirdPlaceSeries/DIVRecBridgeSel.lean#L113-L166)
   - **7s.** [Odd lift and symmetric descent](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/DivisorProofs/ExistsThetaSymmetricFunctionValuesInGroundField.lean#L60-L111) · [Haar metric](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AutomorphicSpectrum/Definitions/GlobalGeometry/IsThetaMeanSquareNorm.lean#L26-L126) · [Hodge power](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/DivisorProofs/ThetaMapPullO1PowerStatement.lean#L12) · [Point section norm](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AutomorphicSpectrum/Proofs/GeometryProofs/KeyDegreeOfBadReduction.lean#L223-L300) · [Chart height](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AutomorphicSpectrum/Proofs/GeometryProofs/CuspIntersectionHeight.lean#L301-L426)
   - **7t.** [Half lattices](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Formalism/LocalToGlobal/KernelLattices/XiIndCClass.lean#L166-L233) · [Signed coset expansion](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/DivisorProofs/CosetIndicatorThetaTorsionTranslates.lean#L200-L248) · [Refinement and box sum](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/SatakeParameters/ThetaSupportInDivLattice.lean#L81-L111) · [Canonical moduli](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/IwasawaDecomposition/EvLgCanonical.lean#L55-L65) · [Split level primes](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/GlobalInterpolation/AbsNormEqOfSplit.lean#L65-L112) · [Word factor](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Formalism/LocalToGlobal/Columns/FiniteSumNonzeroLargeDepth.lean#L636-L648) · [Row and terminal factors](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Formalism/LocalToGlobal/Columns/FiniteSumNonzeroLargeDepth.lean#L1109-L1167)
   - **7u.** [Matched link relation](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AutomorphicArithmetic/ThetaCorrespondence/ThetaPairLifts/ThirdPlaceSeries/RelOfLinks.lean#L22-L90) · [Bottom anchor and normalization](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AutomorphicArithmetic/ThetaCorrespondence/ThetaPairLifts/ThirdPlaceSeries/MiStar2AtRealisation.lean#L48-L196) · [Level volumes and indices](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AutomorphicArithmetic/ThetaCorrespondence/ThetaPairLifts/ThirdPlaceSeries/CIMulIdx.lean#L21-L116) · [Ratio of see-saw factors](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/DecoratedLatticeProofs/SeeSawTotalConstantRatioOfFactorRatios.lean#L14-L74) · [Field reading and unit](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Formalism/LocalToGlobal/IndefiniteReal/IsometricPeriodsOfUnitHypotheses.lean#L21-L201) · [Served-place conductor argument](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AutomorphicArithmetic/ThetaCorrespondence/ThetaPairLifts/ThirdPlaceSeries/NotMemQBase.lean#L44-L86)
   - **7v.** [Word-prime orientation](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Formalism/LocalToGlobal/Columns/AtTLemmas.lean#L424-L455) · [Covariance of the actual kernels](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/GlobalInterpolation/DecoratedLatticePointMoveDiamond.lean#L11-L46) · [Coupled and ordinary stabilizers](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/GlobalInterpolation/NotDvdCardStabilizerForm.lean#L157-L205) · [Shared intermediate point and coset](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/IwasawaDecomposition/KernelAtPointPenultimateStepFirstCoupledKernel.lean#L254-L305) · [Norm of the selected prime factor](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Formalism/LocalToGlobal/Columns/BaseLemmas.lean#L650-L674)
   - **7w.** [Companion and admitted point support](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/GlobalInterpolation/MemSuppBarOfNe.lean#L267-L325) · [Regrouped companion sum](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/GlobalInterpolation/MomentsCosetSum.lean#L423-L491) · [Integral half-lattice cocycle](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/GlobalInterpolation/PeriodLatticeHalfLatticeInclusionsSplitPrime.lean#L224-L231) · [Joint point measure and differential scale](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/GlobalInterpolation/ContinuousToZModComp.lean#L342-L443) · [Rational jet phases](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/GlobalInterpolation/KmapOfOkOfVecP.lean#L231-L288) · [Class-frame ratio at the word prime](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/GlobalInterpolation/CharacterValueAtT.lean#L865-L879) · [Integral Weyl contraction](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/GlobalInterpolation/ExistsMulOfApprox.lean#L313-L398)
   - **7x.** [Free stage-zero scalar orbits](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Formalism/LocalToGlobal/Columns/AtTRank.lean#L247-L270) · [Up-neighbour comparison](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/GlobalInterpolation/UopIsExt.lean#L123-L188) · [Up-neighbour commutation with an edge](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/GlobalInterpolation/LevelChainUOperatorCommutationOfPointPairLaw.lean#L52-L94) · [Chain exponents](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/GlobalInterpolation/LevelChainSharpColumnValuationOfStability.lean#L75-L193) · [Retained word image](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Definitions/GlobalTwisting/LevelChainSharpColumnValuationOwnChainIff.lean#L618-L638)
   - **7y.** [Lift signatures](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Formalism/LocalRingCardinality/Params.lean#L6-L46) · [Anchor and quotient datum](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Definitions/Divisors/QuoIff.lean#L18-L39) · [Quotient floor and node bound](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Definitions/Divisors/DividedWeightUnique.lean#L44-L68) · [Strict decrease across the wall](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/DivisorProofs/PosDividedSlopeWeightsBaseHeight.lean#L39-L88) · [Conjugate block and scalar exponent](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Definitions/Foundations/IsProbabilityMeasureMuPart.lean#L174-L216)
   - **7z.** [Valuation normalization](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Definitions/Divisors/IsLatBasisHtLatIffSpanEq.lean#L75-L94) · [Lower diagonal bound](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/DivisorProofs/SumAbsExOfLe.lean#L278-L303) · [Upper diagonal bound](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/DivisorProofs/SumAbsExOfLe.lean#L316-L344) · [Frame covariance and test selection](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/DivisorProofs/SumAbsExOfLe.lean#L346-L387)
   - **7aa.** [Absolute-value allowance](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Definitions/Divisors/SemiposAtOfPositive.lean#L72-L86) · [Mixed tests and prime support](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Definitions/Divisors/IsLatBasisHtLatIffSpanEq.lean#L515-L537) · [Prime clearing](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/DivisorProofs/WeightAbsEntrySumNonneg.lean#L39-L69) · [Carrying injection](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/DivisorProofs/WeightAbsEntrySumNonneg.lean#L223-L263) · [Volume and height comparison](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AutomorphicArithmetic/ThetaCorrespondence/ThetaPairLifts/HodgeBundleWeights/HeightAddLogNonneg.lean#L29-L51) · [Released carrying budget](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/DivisorProofs/BudgetOfSizeAndWeight.lean#L246-L313)
   - **7ab.** [Original anchor](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/DivisorProofs/WeightBoundHolds.lean#L22-L37) · [Cloud weight bound](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/DivisorProofs/EntRCloudSub.lean#L148-L172) · [Selected weight window](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/DivisorProofs/ParameterPrecisionLeLengthAndRank.lean#L140-L199) · [Width and rank bounds](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/DivisorProofs/WeightBoundHolds.lean#L63-L77) · [Precision and unit](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Definitions/Divisors/QuoIff.lean#L41-L56) · [Precision scale](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Formalism/LocalRingCardinality/DataParamsT.lean#L9-L12) · [Root-discriminant bound](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/DivisorProofs/ParameterPrecisionLeLengthAndRank.lean#L29-L63) · [Fixed host exponent](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/DivisorProofs/HostBoundHolds.lean#L110-L128) · [Reciprocal-prime estimate](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/DivisorProofs/SumLogDivPredLeReal.lean#L12-L15)
   - **7ac.** [Weights with the half shift](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/InertiaEigenvalues/WeightInfinitesimalCharacterParamOfLevelCompatib.lean#L42-L78) · [Weight list](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/BlockTheory/TorsionBlocks/JumpDictionary.lean#L15-L40) · [Rank and gaps](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Formalism/LocalRingCardinality/Params.lean#L6-L69) · [Centered string and reference shift](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AutomorphicArithmetic/ThetaCorrespondence/ThetaPairLifts/ThirdPlaceSeries/HeckeCharEqOfLAdicRealisation.lean#L14-L59) · [Common residual comparison](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AutomorphicArithmetic/ThetaCorrespondence/ThetaPairLifts/ThirdPlaceSeries/ResidualCount.lean#L47-L156) · [Base-change definition and retained part](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Formalism/SharpRestriction/LevelLetterCocycles/DataN.lean#L1241-L1298) · [Good-type cut](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AutomorphicArithmetic/ThetaCorrespondence/ThetaPairLifts/ThirdPlaceSeries/ExistsGood.lean#L15-L54)
   - **7ad.** [Fitted character conductors](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Formalism/LocalToGlobal/UnitaryNorms/CharHost.lean#L26-L83) · [Tuple character and active depletion](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Formalism/LocalToGlobal/UnitaryNorms/CharTied.lean#L61-L138) · [Level exceptions](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Formalism/SharpRestriction/LevelLetterCocycles/DataN.lean#L137-L155) · [Unramified parameters](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AutomorphicArithmetic/Foundations/GaloisRepresentations/TypeConditions/ConstituentConditionAtDistinguishedPrime.lean#L34-L55) · [Twist and depletion inertia](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Formalism/SharpRestriction/LevelLetterCocycles/SumRepresentations.lean#L243-L266) · [Squarefree type and conjugate roots](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Formalism/SharpRestriction/LevelLetterCocycles/PairedCharacterCondition.lean#L359-L374) · [Good-type transport to all places](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AutomorphicArithmetic/Foundations/GaloisRepresentations/HeckeGaloisCriteria/ExistsUnitOneAddGenerator.lean#L143-L150) · [Inertia roots on the good types](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AutomorphicArithmetic/Foundations/GaloisRepresentations/TypeConditions/CardRootsBar.lean#L71-L140) · [Tameness](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AutomorphicArithmetic/Foundations/GaloisRepresentations/TypeConditions/AutomorphicGaloisAdmissible.lean#L15-L27) · [Rank-one local comparison](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AutomorphicArithmetic/Foundations/GaloisRepresentations/TypeConditions/AutomorphicGaloisAdmissible.lean#L133-L151) · [Proof matching the marked inertia root](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Formalism/SharpRestriction/LevelLetterCocycles/ExistsInertiaCharpolyFactorOfTameType.lean#L1241-L1286)
   - **7ae.** [Cyclotomic unramifiedness](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/InertiaEigenvalues/CyclotomicCharUnramifiedAwayFromEll.lean#L8-L17) · [Marked inertia witness](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AutomorphicArithmetic/ThetaCorrespondence/ThetaPairLifts/ThirdPlaceSeries/EigPdHolds.lean#L17-L65) · [Reduction of the order-fifteen character](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/SatakeParameters/EigenPeriodOfLocalGlobal.lean#L84-L132) · [Use in the selected cut](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AutomorphicArithmetic/ThetaCorrespondence/ThetaPairLifts/ThirdPlaceSeries/AdmissibleJOfRestricted.lean#L28-L44)
   - **7af.** [G1](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Formalism/SharpRestriction/LevelLetterCocycles/SumRepresentations.lean#L115-L118) · [Integral coefficient lift from split Frobenius](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Formalism/SharpRestriction/LevelLetterCocycles/ExistsInertiaCharpolyFactorOfTameType.lean#L111-L170) · [Continuity and closedness argument](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Formalism/PadicSubsetSums/StableGradedCounts/ContinuousCoeffCharpolyRev.lean#L56-L78) · [G1 on the cut](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AutomorphicArithmetic/ThetaCorrespondence/ThetaPairLifts/ThirdPlaceSeries/CFacts.lean#L23-L39) · [Normalized local generator transport](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Formalism/LocalToGlobal/NormalizedHeckeFactors/HeckeStdTwOfStd.lean#L86-L135) · [Integral lattice action](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/InertiaEigenvalues/HeckeStableLocalLatticeOfIntegral.lean#L31-L67)
   - **7ag.** [Original indices and summation at a lift](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Formalism/LocalRingCardinality/RhoDetPrimeEqMuTildeMulChiPow.lean#L275-L283) · [Place weights and division allowance](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Formalism/SharpRestriction/GlobalTheoremInputs.lean#L36-L63) · [Arithmetic scale](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Formalism/LocalRingCardinality/DataArith.lean#L189-L205) · [Length window](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Formalism/LocalRingCardinality/DataParams%E2%82%82.lean#L171-L172) · [Type group size bound](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Formalism/LocalToGlobal/Divisibility/EigenvalueSumLeClosedOf.lean#L717-L762) · [Host degree](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Formalism/LocalToGlobal/ProjectiveLattices/HostSize.lean#L18-L30) · [Host size and discriminant bound](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/HeightProofs/HostSzOfSizes.lean#L26-L42)
   - **7ah.** [Field in the conductor-one specialization](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AutomorphicArithmetic/CharacterBoxes/ExistsPageN.lean#L19-L25) · [Prime size proof for the actual ranks](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AutomorphicArithmetic/ShortNorms/MulLogLeRpowEventually.lean#L21-L75) · [Use in the fitted host construction](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AutomorphicArithmetic/ShortNorms/hostOfFieldOf.lean#L159-L185)
   - **7ai.** [Retained multiplicities across localized sites](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Development/AbelianVarietyTheory/Proofs/InertiaEigenvalues/GoodPointMultiplicityLeRankProof.lean#L47-L113) · [Reference rank comparison at a served place](https://github.com/Deicyde/riemann-hypothesis/blob/e7c8972c701d3396947808ab20cc16dc14623d33/src/Formalism/SharpRestriction/AdmissibleParameterJoins/NandDIVb.lean#L26-L50)

8. G. Pareschi. *Torsion points on theta divisors and semihomogeneous vector bundles*. Algebra & Number Theory **15** (2021), 1581–1592, Theorem C. [DOI](https://doi.org/10.2140/ant.2021.15.1581) · [Author’s preprint](https://arxiv.org/pdf/2007.12252v3#page=2)
9. R. de Jong and F. Shokrieh. *Faltings height and Néron–Tate height of a theta divisor*. Compositio Mathematica **158** (2022), 1–32. [DOI](https://doi.org/10.1112/S0010437X21007661)
10. J. H. Silverman. *Advanced Topics in the Arithmetic of Elliptic Curves*. Graduate Texts in Mathematics 151, Springer, 1994. [DOI](https://doi.org/10.1007/978-1-4612-0851-8)
11. P. Autissier. *Un lemme matriciel effectif*. Mathematische Zeitschrift **273** (2013), 355–361, Lemmas 2.2 and 3.2 and §3.2. [DOI](https://doi.org/10.1007/s00209-012-1008-x) · [Author’s preprint](https://arxiv.org/pdf/1104.5644)
12. L. Moret-Bailly. *Pinceaux de variétés abéliennes*. Astérisque **129** (1985). [Full text](https://www.numdam.org/item/AST_1985__129__1_0/)
13. É. Gaudron and G. Rémond. *Minima, pentes et algèbre tensorielle*. Israel Journal of Mathematics **195** (2013), 565–591. [DOI](https://doi.org/10.1007/s11856-012-0109-x) · [Author’s preprint](https://arxiv.org/pdf/1109.2812#page=6)
14. K. Gröchenig. *Foundations of Time-Frequency Analysis*. Birkhäuser, 2001, Chapter 9, for the Fourier coefficients of the squared Zak transform. [Chapter](https://doi.org/10.1007/978-1-4612-0003-1_9)
15. T. Foster, J. Rabinoff, F. Shokrieh and A. Soto. *Non-Archimedean and tropical theta functions*. Mathematische Annalen **372** (2018), 891–914. [DOI](https://doi.org/10.1007/s00208-018-1646-3) · [Preprint](https://arxiv.org/abs/1705.08212)
16. W. Banaszczyk. *New bounds in some transference theorems in the geometry of numbers*. Mathematische Annalen **296** (1993), 625–635. [DOI](https://doi.org/10.1007/BF01445125)
17. D. Mumford. *On the equations defining abelian varieties. I*. Inventiones Mathematicae **1** (1966), 287–354, §1. [DOI](https://doi.org/10.1007/BF01389737) · [Faithfully flat descent of modules, Stacks Project](https://stacks.math.columbia.edu/tag/03O8)
18. Z. Amir-Khosravi. *Serre’s tensor construction and moduli of abelian schemes*, §§1.1–1.3. [Preprint](https://arxiv.org/abs/1507.07607)

19. *The metric of the released odd symmetric theta coordinates*. [Supplement](supplement.zip) entry `odd_symmetric_theta_metric.md`.
20. *Integral trace for an induced theta coset*. [Supplement](supplement.zip) entry `induced_coset_theta_integral_trace.md`.
21. *Integral content of the weighted induced theta section*. [Supplement](supplement.zip) entry `weighted_induced_theta_content.md`.
22. *The relative period scalar on the fitted row*. [Supplement](supplement.zip) entry `fitted_relative_period_unit.md`.

23. *Determinants of the induced theta sections at a word prime*. [Supplement](supplement.zip) entry `word_prime_fourier_determinants.md`.

24. *Word-prime decorations in the released period*. [Supplement](supplement.zip) entry `word_decoration_lattice_in_period.md`.
25. *Source diamond augmentation in the first lift*. [Supplement](supplement.zip) entry `word_augmentation_in_first_lift.md`.
26. *Word augmentation along the shared chain*. [Supplement](supplement.zip) entry `word_augmentation_along_shared_chain.md`.

27. *Testing the spectral image on the arithmetic row*. [Supplement](supplement.zip) entry `spectral_row_tests.md`.
28. *The projected row index and the finite type cut*. [Supplement](supplement.zip) entry `projected_row_index.md`.
29. halasz_resonator. *Multi-Component Jet Amplification, Good/Bad Arakelov Slope Descent, and the Covolume–Division Telescope in the Alpöge–Claude Lean Argument for RH*. rStation 199, version 1 (2026). [Published paper](https://rstation.ai/problems/riemann_hypothesis/papers/199/v1)

30. *Component degrees and the boundary volume in the nonterminal slope estimate*. [Supplement](supplement.zip) entry `component_boundary_degree_mean.md`.

31. *Cusp extension and the finite lattice on the ordinary Lie flag*. [Supplement](supplement.zip) entry `ordinary_lie_flag_cusp_lattice.md`.

32. J. S. Milne. *Class Field Theory*, Chapters I and V, for finite-order characters, ray classes and local and global reciprocity. [Course notes](https://www.jmilne.org/math/CourseNotes/CFT.pdf)
