% Single-file version of main.tex: the sections and the bibliography of % this directory are included verbatim, so this file alone produces the paper. \documentclass[11pt]{amsart} \usepackage[margin=1in]{geometry} \usepackage{amsmath,amssymb,amsthm,mathtools,booktabs,longtable,array,microtype} \usepackage{url} \usepackage{hyperref} \hypersetup{colorlinks=true,linkcolor=blue,citecolor=blue,urlcolor=blue, pdftitle={An Exponent of 1.04273 for the Unit Distance Problem}, pdfauthor={Eric Naslund}} \numberwithin{equation}{section} % Theorems and equations share one counter, so every number is unique. \newtheorem{theorem}[equation]{Theorem} \newtheorem{proposition}[equation]{Proposition} \newtheorem{lemma}[equation]{Lemma} \newtheorem{corollary}[equation]{Corollary} \theoremstyle{definition} \newtheorem{definition}[equation]{Definition} \theoremstyle{remark} \newtheorem{remark}[equation]{Remark} \DeclareMathOperator{\Gal}{Gal} \DeclareMathOperator{\Cl}{Cl} \DeclareMathOperator{\rd}{rd} \DeclareMathOperator{\Tr}{Tr} \DeclareMathOperator{\rank}{rank} \DeclareMathOperator{\Br}{Br} \DeclareMathOperator{\Pic}{Pic} \DeclareMathOperator{\Frob}{Frob} \DeclareMathOperator{\inv}{inv} \DeclareMathOperator{\Res}{Res} \DeclareMathOperator{\covol}{covol} \newcommand{\Ftwo}{\mathbb F_2} \newcommand{\Q}{\mathbb Q} \newcommand{\R}{\mathbb R} \newcommand{\C}{\mathbb C} \newcommand{\Z}{\mathbb Z} \newcommand{\deltastar}{0.04273} \newcommand{\thetastar}{65535/131072} \newcommand{\ceiling}{0.04871285} \setcounter{tocdepth}{1} \raggedbottom \title[The Unit Distance Problem]{An Exponent of 1.04273\\for the Unit Distance Problem} \author[Eric Naslund]{Eric Naslund\textsuperscript{*}} \email{naslund.math@gmail.com} \date{September 30, 2026} \begin{document} \begin{abstract} Let $u(U)$ count the unordered pairs at distance one in a finite planar set $U$. We construct finite sets $U_j$ with $|U_j|\to\infty$ and $u(U_j)/|U_j|^{1.04273}\to\infty$. The largest exponent previously claimed, $1.0358$, is in the author's unpublished manuscript. The method is the number-field construction of OpenAI and Sawin: unit distances come from elements of relative norm one in quadratic extensions, and the fields come from an infinite pro-$2$ class field tower. The new ingredients are quadratic extensions of mixed signature, with an exact average over their norm-one units, and a tower over the real quadratic field $\Q(\sqrt{241})$, in which $2$, $3$ and $5$ split. Its Golod--Shafarevich function contains two copies of the local conditions at these primes but only one constant term, and the extra room lets the tower be ramified only above $2$, $3$ and $5$; the root discriminant of its fields is about $286$. The relative zeta value is bounded through the zeta function of the degree-$512$ field generated over $\Q(\sqrt{241})$ by the square roots of its $\{2,3,5\}$-units, which is the product of the Dedekind zeta function of $\Q(\sqrt{241})$ and $255$ quadratic Hecke $L$-functions. Finite facts and numerical inequalities are certified by exact computation and interval arithmetic. A significant portion of this work was verified in Lean, reducing the result with exponent $1.0427$ to an explicit zeta function inequality (Palomar registry, PALOMAR-2026-10-01-000018, version~1). \end{abstract} \maketitle % AI Methodology statement, referenced by the asterisk on the author's name. {\noindent\small\textsuperscript{*}\textbf{AI Methodology.} This work was done with extensive AI use, with agents developing the mathematics and writing the paper. I supervised and prompted the AI agents that produced this work. While this has been edited to improve readability, I recommend the reader also make use of their AI model of choice to ask the paper questions. The construction over $\Q(\sqrt{241})$, the computations that certify it, the Lean formalization of the reduction of the proof of exponent $1.0427$ to one explicit zeta inequality (Remark~\ref{an:lean-hypothesis}), and this paper were produced with a combination of Claude Opus 5.5, ChatGPT Sol 5.6, ChatGPT Astra 6, working as agents in Claude Code and Codex. The construction adapts the author's previous manuscript \cite{Naslund0358}, which was also written with AI tools.\par} \medskip \section{Introduction}\label{intro:section} For a finite set $U\subset\R^2$, let \[ u(U)=\#\{\{x,y\}\subset U:|x-y|=1\},\qquad u(n)=\max_{|U|=n}u(U), \] where the pairs are unordered and the distance is Euclidean. The unit distance problem of Erd\H{o}s \cite{Erdos1946} asks for the order of growth of $u(n)$. Our main result is the following. \begin{theorem}\label{thm:main} There are finite sets $U_j\subset\R^2$, with $|U_j|\to\infty$, such that \[ \frac{u(U_j)}{|U_j|^{1.04273}}\longrightarrow\infty. \] In particular, $u(n)\ge n^{1.04273}$ for arbitrarily large integers $n$. \end{theorem} The largest exponent previously claimed, $1.0358324$, is in the author's unpublished manuscript \cite{Naslund0358}. Theorem~\ref{thm:main} uses the same number-field method with several new ingredients: fields of mixed signature, averaged over their norm-one units; an infinite pro-$2$ extension of the real quadratic field $\Q(\sqrt{241})$, with restricted ramification and a nonabelian local Galois group at the primes above $2$; an explicit bound for the relative zeta value (defined in Subsection~\ref{intro:outline}) from the quadratic $L$-functions of a Kummer field of degree $512$; and optimized \emph{profiles}: smooth or locally constant weight functions at the archimedean places and at finitely many primes, whose product determines, through its superlevel sets, the region in which lattice points are counted. Subsection~\ref{intro:new} describes them. The theorem concerns an unbounded sequence of cardinalities. Its proof uses no unproved hypothesis, such as the generalized Riemann hypothesis, but it is computer-assisted: finite facts about the tower and several numerical inequalities are verified by exact computation and interval arithmetic, as described in Subsection~\ref{intro:code}. \subsection{Background}\label{intro:background} Erd\H{o}s \cite{Erdos1946} showed that a suitable section of a scaled integer lattice gives \[ u(n)\ge n^{1+c_0/\log\log n} \] for some $c_0>0$ and arbitrarily large $n$, and he conjectured that this is essentially optimal \cite{Erdos1946,Erdos1994}. The best upper bound, \[ u(n)=O(n^{4/3}), \] is due to Spencer, Szemer\'edi and Trotter \cite{SpencerSzemerediTrotter1984}; Sz\'ekely \cite{Szekely1997} gave a proof by crossing numbers. In $2026$ a team at OpenAI \cite{OpenAI2026} disproved Erd\H{o}s's conjecture. They constructed sets with $u(U)\ge|U|^{1+\delta}$ for a fixed $\delta>0$ and arbitrarily large $|U|$, replacing the integer lattice by ideal lattices in number fields of large degree and small root discriminant. Alon, Bloom, Gowers, Litt, Sawin, Shankar, Tsimerman, Wang and Wood \cite{Alon2026} simplified the argument and obtained $\delta\approx6\cdot10^{-38}$. Sawin \cite{Sawin2026} made every step explicit and sharpened it, and reached the exponent $1.014114$. Improvements followed within days in online discussions. On the Erd\H{o}s Problems forum, mlewko \cite{ErdosProblems90} reached $1.031849$ in Sawin's criterion with finite data found by ChatGPT, and on MathOverflow spiderduckpig \cite{SpiderduckpigMO2026} adjusted these data to reach $1.031883$. In a comment the next day, spiderduckpig proposed a structural change: in the Golod--Shafarevich step of Sawin's argument \cite[Lemma~11]{Sawin2026}, replace the unweighted count of relations by a count weighted by degree in the Zassenhaus filtration (recalled in Section~\ref{tw:tower}). This was reported to give $\delta\ge0.0333487$ \cite{SpiderduckpigMO2026}. Weighted counts of this kind underlie the towers of \cite{Naslund0358} and of the present paper. The author's MathOverflow answer \cite{NaslundMO2026} then combined this count with covolume counting, dyadic ramification and Euler products over a fixed subfield. The manuscript \cite{Naslund0358} is a formal write-up of that construction, and it records $\delta=0.0358324$. Table~\ref{intro:table-bounds} lists the explicit exponents, and the repository \cite{TaoConstants84a} records further bounds posted online. \begin{table}[htbp] \centering \caption{Explicit lower bounds $u(n)\ge n^{1+\delta}$ for arbitrarily large $n$, in increasing order. The exponents are rounded down and constant factors are omitted. The last column gives the kind of source; apart from Erd\H{o}s's, none of these bounds had appeared in a refereed journal when this paper was written.} \label{intro:table-bounds} \small \begin{tabular}{@{}llll@{}} \toprule Exponent & Source & Main new input & Kind of source\\ \midrule $1+c_0/\log\log n$ & Erd\H{o}s \cite{Erdos1946} & integer lattice & journal\\ $1+\delta$, $\delta>0$ & OpenAI \cite{OpenAI2026} & ideal lattices from class field towers & manuscript\\ $1+6\cdot10^{-38}$ & Alon et al.\ \cite{Alon2026} & simplified argument & preprint\\ $1.014114$ & Sawin \cite{Sawin2026} & explicit and sharpened criterion & preprint\\ $1.015263$ & Emmerich \cite{Emmerich2026} & optimized parameters & preprint\\ $1.031185$ & Emmerich, Cordella \cite{EmmerichCordella2026} & extended range of primes & online certificate\\ $1.031849$ & mlewko \cite{ErdosProblems90} & finite data found by ChatGPT & forum post\\ $1.0333487$ & spiderduckpig \cite{SpiderduckpigMO2026} & weighted Zassenhaus count & online post\\ $1.0358324$ & Naslund \cite{Naslund0358,NaslundMO2026} & covolume counting, dyadic & manuscript,\\ & & ramification, fixed-subfield Euler products & online post\\ $1.04273$ & this paper & see Subsection~\ref{intro:new} & \\ \bottomrule \end{tabular} \end{table} \subsection{The construction in outline}\label{intro:outline} The construction carries Erd\H{o}s's lattice argument to number fields of large degree. Let $K$ be such a field, of degree $2d$, with an involution $\iota$ whose fixed field is $F$. We take the points of a translated ideal lattice of $K$ that lie in a bounded region of $K\otimes\R$. A complex embedding in which $\iota$ acts as complex conjugation maps them injectively to the plane. It sends every difference of relative norm one over $F$ to a unit vector, so two points whose difference has relative norm one are at distance one. The exponent is then set by a balance between four quantities, each normalized as $d^{-1}$ times the logarithm of a count. First, equal-norm choices: if a prime $\mathfrak p$ of $F$ splits in $K$ as $\mathfrak P\cdot\iota\mathfrak P$, the $k+1$ ideals $\mathfrak P^j(\iota\mathfrak P)^{k-j}$, $0\le j\le k$, all have relative norm $\mathfrak p^k$. When the construction uses these $k+1$ ideals, we say that $\mathfrak p$ is used with exponent $k$, and we call the primes so used the \emph{selected primes}. Second, ideal classes: a choice of exponents $0\le j_{\mathfrak p}\le k_{\mathfrak p}$, one at each selected prime $\mathfrak p$ (used with exponent $k_{\mathfrak p}$), gives the product of the ideals $\mathfrak P^{j_{\mathfrak p}}(\iota\mathfrak P)^{k_{\mathfrak p}-j_{\mathfrak p}}$. For the choices whose ideals $\prod_{\mathfrak p}\mathfrak P^{j_{\mathfrak p}}$ lie in one class of $\Cl(K)$ modulo the image of $\Cl(F)$, there are elements of relative norm one that generate these products times one fixed fractional ideal, and keeping the most frequent class loses at most a factor equal to the order of that quotient (Lemma~\ref{geo:selection}). Through the class-number formula, this loss is governed by the root discriminant $\lambda$ of $K$ and by the \emph{relative zeta value} $L_F(1)$, where $L_F(s)=\zeta_K(s)/\zeta_F(s)$. Third, the region: its shape decides how many pairs of points with difference of relative norm one lie in it, and its volume fixes the number of points. Fourth, density: using a prime $\mathfrak p$ with exponent $k$ also makes the ideal lattice denser by the factor $N\mathfrak p^k$; since the exponent $1+\delta$ is measured against the number of points, this costs $\delta k\log N\mathfrak p$ before division by $d$. Quantitatively, suppose that such fields exist with $d\to\infty$ and a fixed root discriminant $\lambda$. Let $(b,c)$ be the signature of $F$, so that $d=b+2c$, and put $\theta=c/d$; thus $2\theta$ is the proportion of non-real embeddings of $F$. Suppose also that the selected primes have \emph{uniform local types}: they are all the primes of $F$ above the primes $r$ of a fixed finite set $\mathcal R$, each of them splits in $K$, and all those above a given $r$ have the same absolute ramification index $e_r$ and residue degree $f_r$ (Definitions~\ref{tw:type} and~\ref{geo:uniform-types}). Then $F$ has exactly $d/(e_rf_r)$ primes above each $r$, and we use all of them with a common exponent $k_r$. If the points are taken from one translate of a fixed ideal lattice, with no further weighting at the selected primes (\emph{unweighted shell profiles}, Definition~\ref{fw:shell-profile}), the exponent $1+\delta$ is attained when \begin{equation}\label{intro:margin} \sum_{r\in\mathcal R}\frac{\log(k_r+1)-\delta k_rf_r\log r}{e_rf_r} -\Bigl(\frac12-\delta\Bigr)\log\lambda-C+\mathcal A_\delta(\theta)>0. \end{equation} Here $C$ is an upper bound for $d^{-1}\log L_F(1)$, and \[ \mathcal A_\delta(\theta)=(1-\delta)\log2+(1-\theta)\log\pi +(1-2\theta)J_\R+\theta J_\C \] is the archimedean contribution. In it, $J_\R$ and $J_\C$ are the functionals of the profiles at the real and at the complex places of $F$ (Definition~\ref{geo:profiles}); they compare how much a profile overlaps its translates by elements of relative norm one with its mass. We call the left side of \eqref{intro:margin} the \emph{margin}. Proposition~\ref{geo:transfer}, the geometric transfer, makes this precise for quadratic extensions $K/F$ with $K$ totally imaginary, $F$ with a real place and $K/F$ unramified at every finite place: under conditions on the profiles, if the margin is bounded below by a positive constant along such a family, then there are finite sets $U_j\subset\R^2$ with $|U_j|\to\infty$ and $u(U_j)/|U_j|^{1+\delta}\to\infty$. Section~\ref{fw:section} allows general \emph{shell profiles}, weighted sums of indicator functions of sets defined by valuations (Definition~\ref{fw:shell-profile} and Proposition~\ref{fw:transfer}). This replaces each summand of the sum over $\mathcal R$ in \eqref{intro:margin} by $\log\mathcal F_{\delta,r}/(e_rf_r)$, where $\mathcal F_{\delta,r}$ is the value of the local functional of Definition~\ref{fw:local-functional} at the shell profile used above $r$; we still call the resulting left side the margin (Corollary~\ref{fw:uniform-types}). Section~\ref{sec:certificate} proves a positive lower bound for it at $\delta=\deltastar$. The fields must form an infinite family of growing degree with fixed $\lambda$ and fixed types at the selected primes. Such families come from \emph{towers}, pro-$2$ extensions with restricted ramification and prescribed local Galois groups, when these are infinite. Golod--Shafarevich arguments \cite{GolodShafarevich} show that a tower is infinite when an explicit function of $t\in(0,1)$, built from the number of generators of its Galois group and from its local conditions, takes a negative value; here that function is the \emph{Golod--Shafarevich function} $P_B$ of \eqref{tw:PB}, and Proposition~\ref{tw:infinite} proves the criterion for it. The use of Frobenius conditions to control splitting in such towers goes back to Hajir, Maire and Ramakrishna \cite{HajirMaireRamakrishna2021Cutting}. \subsection{New ingredients}\label{intro:new} The construction differs from those of Sawin \cite{Sawin2026} and of \cite{Naslund0358} in four ways. \emph{Mixed signature.} In Sawin's construction, and in \cite{Naslund0358}, the field $K$ is a CM field: complex conjugation is central in $\Gal(K/\Q)$, $F$ is totally real, every norm-one element has modulus one at every archimedean place, and the norm-one units are finite. Here $K$ is totally imaginary and Galois over $B=\Q(\sqrt{241})$, and $F$ is the fixed field of one complex conjugation $\iota_1$, which is not central: its conjugacy class has at least $2^{15}$ elements, so $\theta\ge\theta_*=\thetastar$ (Theorem~\ref{tw:field-family}) and almost every place of $F$ is complex. Above a complex place of $F$ the two coordinates of a norm-one element can have moduli $e^u$ and $e^{-u}$, and the norm-one units, of rank $c$, move $u$ along a lattice. Averaging over a fundamental domain of that lattice counts the resulting pairs of points exactly, and the class-number formula cancels the regulator of these units together with the capitulation kernel, the kernel of $\Cl(F)\to\Cl(K)$ (Proposition~\ref{geo:mass}). Mixed signature therefore costs only a factor $\pi$ per complex place of $F$, and a profile that couples the two coordinates at that place more than repays it. \emph{A tower over $\Q(\sqrt{241})$.} Over a totally real base field in which every prime at which the tower is ramified splits completely, the Golod--Shafarevich function has a local term for every prime of the base above such a prime, but only one constant term (in \eqref{tw:PB}, the terms $2\psi_D$ and $4\psi_{C_2\times C_2}$, and the constant $1$). The field $B=\Q(\sqrt{241})$ is the real quadratic field of smallest discriminant in which $2$, $3$ and $5$ all split. Over $B$ we prescribe a nonabelian dyadic local Galois group of order $32$ at each of the two primes above $2$, a tame local group $C_2\times C_2$ at each of the four primes above $3$ and $5$, and conditions that make the Frobenius elements have order dividing $4$ at the two primes above $29$ and at the prime $7$, which is inert in $B$. The relations are counted through their images in the Zassenhaus filtration, and one dyadic relation is implied by the others through Hilbert reciprocity. The resulting tower is infinite, it is ramified only above $2$, $3$ and $5$, and its Galois group has $8$ generators. The fields of Theorem~\ref{tw:field-family} have root discriminant \[ \lambda=2^{9/4}\sqrt{3615}\approx286.0, \qquad 3615=3\cdot5\cdot241. \] Sawin's fields come from a pro-$2$ tower of unramified extensions of $\Q(\sqrt{3\cdot5\cdots43})$, so they are ramified over $\Q$ at $2$ and at the thirteen odd primes up to $43$, and their root discriminant is about $1.6\cdot10^8$. Theorem~\ref{tw:field-family} states the family of fields. \emph{The zeta value.} Every field $K$ of Theorem~\ref{tw:field-family} contains the Kummer field $E=B(\sqrt V)$, of degree $512$ over $\Q$, where $V$ is the group of $\{2,3,5\}$-units of $B$ (the units of $\mathcal O_B[1/30]$) modulo squares. Its Dedekind zeta function is the product of $\zeta_B$ and $255$ quadratic Hecke $L$-functions of $B$, which we evaluate rigorously with approximate functional equations (Sections~\ref{gf:section} and~\ref{an:section}). Together with refinements from the known types of primes in the fields of the tower (Definition~\ref{tw:type}), this gives the bound $C=\ceiling$ for $d^{-1}\log L_F(1)$ (Theorem~\ref{an:ceiling}). It replaces Louboutin's bound \cite{Louboutin2000}, which Sawin uses, and the Euler products over a fixed multiquadratic subfield of \cite{Naslund0358}. \emph{Profiles and shells.} Sawin counts points in a region bounded in the sup norm by packing, and \cite{Naslund0358} counts them by covolume in a Euclidean ball. We use a Gaussian profile at the real places of $F$; at the complex places, a Student--Bernstein profile, which couples the two coordinates (Definition~\ref{pf:student}); and shell profiles at the selected primes (Definition~\ref{fw:shell-profile}). We count points by Poisson summation (Sections~\ref{geo:section}--\ref{pf:section}). Table~\ref{intro:table-constructions} compares Sawin's example with the present construction. \begin{table}[htbp] \centering \caption{Sawin's example and the present construction. The pairs $(e,f)$ are absolute ramification indices and residue degrees of the selected primes of $F$; Sawin's residue degrees are at most $2$.} \label{intro:table-constructions} \begin{tabular}{lll} \toprule & Sawin \cite{Sawin2026} & this paper\\ \midrule base field of the tower & $\Q(\sqrt{3\cdot5\cdots43})$ & $\Q(\sqrt{241})$\\ primes ramified in the tower & $2$ and $3,5,\dots,43$ & those above $2,3,5$\\ fields $K$ & CM, $\theta=0$ & mixed signature, $\theta\ge\theta_*$\\ selected primes, $(e,f)$ & $22$ primes up to $179$ & $2{:}\,(8,4)$;\ $3,5{:}\,(2,2)$;\\ & & $7{:}\,(1,8)$;\ $29{:}\,(1,4)$\\ root discriminant $\lambda$ & about $1.6\cdot10^8$ & $2^{9/4}\sqrt{3615}\approx286.0$\\ $\ell=\log\lambda$ & $18.90$ & $5.656$\\ zeta term & $1+\log\log\lambda\approx3.94$ & $C=\ceiling$\\ exponent $1+\delta$ & $1.014114$ & $1.04273$\\ \bottomrule \end{tabular} \end{table} \subsection{Why the exponent improves}\label{intro:why} Write $\ell=\log\lambda$, \[ J=\sum_{r\in\mathcal R}\frac{\log(k_r+1)}{e_rf_r},\qquad H=\sum_{r\in\mathcal R}\frac{k_r\log r}{e_r}, \] so that the sum in \eqref{intro:margin} is $J-\delta H$. Sawin's criterion \cite[Proposition~10]{Sawin2026} applies to Galois CM fields $K$ of unbounded degree, with maximal totally real subfield $F$, whose relative root discriminant $(|\Delta_K|/|\Delta_F|)^{1/d}$ is $\lambda$, and counts points in a region bounded in the sup norm, with a parameter $R>1$. In his example, as for our fields, $K/F$ is unramified at every finite place, so that $|\Delta_K|=|\Delta_F|^2$ and $\lambda=\rd(K)$ (Definition~\ref{tw:rd}). Rearranged, his criterion gives arbitrarily large sets with $u(U)\gg|U|^{1+\delta}$ whenever \eqref{intro:sawin-margin} holds; inequality \eqref{intro:regrouped} is \eqref{intro:margin} with $\mathcal A_\delta(\theta)$ written out and regrouped: \begin{align} &(J-\delta H)+\bigl(\log2\pi-\tfrac12\ell-1-\log\ell\bigr) +\bigl(2\log(1-R^{-1})-2\delta\log(2R+e^{-H/2})\bigr)\ge0, \label{intro:sawin-margin}\\ &(J-\delta H)+\bigl(\log2+(1-\theta)\log\pi-\tfrac12\ell-C\bigr) +\bigl(\delta\ell-\delta\log2+(1-2\theta)J_\R+\theta J_\C\bigr)>0. \label{intro:regrouped} \end{align} In both, the three groups are the gain from equal-norm choices, the loss in selecting a common ideal class, and the count of points together with the shape of the region. \emph{The fields.} The dominant change is in $\ell$. Sawin's example has $\ell>18.9$; here $\ell<5.657$, so the term $-\frac12\ell$ improves by more than $6.6$ per unit of $d$. The price is paid in the first group: Sawin selects $22$ primes with residue degree at most $2$, and we select only the primes above $2$, $3$, $5$, $7$ and $29$, several with residue degree $4$ or $8$. \emph{Ideal-class selection.} Both criteria use the analytic class-number formula, which is exact in either signature; they differ in how they bound the zeta value. At $\theta=0$ the constants agree, and the groups differ only in $C$ against $1+\log\ell$. In Sawin's criterion, $1+\log\ell$ is the sum of $1-\log2+\log\ell$, which is Louboutin's bound \cite[Corollary~3]{Louboutin2000} for $d^{-1}\log L_F(1)$ read through the class-number formula, and $\log2$, for the factor $2^d$ allowed for units in \cite[Lemma~6]{Sawin2026}. The manuscript \cite{Naslund0358} removed the unit factor and sharpened Louboutin's bound with Euler products over a fixed subfield. Here Proposition~\ref{geo:mass} accounts for the unit index $[\mathcal O_F^\times:N_{K/F}\mathcal O_K^\times]$ exactly, and the Kummer field gives $C=\ceiling$, whereas $1+\log\ell$ would exceed $2.73$ even at our value of $\lambda$. Mixed signature costs $\theta\log\pi$ in this group, since each place of $F$ carries one factor $\pi$ in the class-number formula. \emph{Points and region.} Besides the term $-\delta H$, which both criteria share, Sawin's packing count of points contributes $-2\delta\log(2R+e^{-H/2})$ to \eqref{intro:sawin-margin}, which does not involve $\lambda$. Counting by covolume, as in \cite{Naslund0358} and here, contributes $\delta\ell-\delta\log2$ to \eqref{intro:regrouped}, because a larger discriminant makes the ideal lattice sparser. The term $2\log(1-R^{-1})$, which measures how much Sawin's region overlaps its translates by elements of relative norm one, and the volume of that region become the functionals $J_\R$ and $J_\C$ of the profiles. \emph{Signature.} CM fields have $\theta=0$. With the other data fixed, mixed signature adds $\theta(J_\C-2J_\R-\log\pi)$ to the left side of \eqref{intro:regrouped}. Section~\ref{sec:certificate} proves $J_\C-2J_\R-\log\pi>0.6324$ (Proposition~\ref{cert:enclosures}(h) and Lemma~\ref{cert:lower-margin}), so with $\theta\ge\theta_*$ this term exceeds $0.316$; at $\theta=0$ the lower bound for the margin proved there would be negative (Remark~\ref{cert:signature-remark}). Table~\ref{cert:tab-budget} lists, rounded, the terms of the lower bound for the margin that Section~\ref{sec:certificate} proves at $\delta=\deltastar$ and $\theta=\theta_*$; they sum to about $1.77\cdot10^{-4}$ (Proposition~\ref{cert:enclosures}). With the same complex-place profile, the real-place profile adjusted to $\delta$ as in Definition~\ref{cert:data}, and shell weights re-optimized in the same way, the corresponding lower bound at $\delta=0.0428$ is negative (Remark~\ref{cert:observations}). \subsection{Organization}\label{intro:organization} Section~\ref{tw:tower} constructs the tower over $B$ and proves Theorem~\ref{tw:field-family}, which supplies the fields. Section~\ref{gf:section} describes the Kummer field and its quadratic $L$-functions, and Section~\ref{an:section} proves the upper bound $d^{-1}\log L_F(1)0$, this contradicts $P_B(34/117)<0$ (Lemma~\ref{tw:PB-value}). \end{proof} \subsection{The Fields}\label{tw:fields-section} \begin{definition}\label{tw:rd} The \emph{root discriminant} of a number field $M$ with discriminant $\Delta_M$ is $\rd(M)=|\Delta_M|^{1/[M:\Q]}$. \end{definition} \begin{definition}\label{tw:type} For a number field $M$ and a prime $\mathfrak P$ of $M$ above a rational prime $p$, the \emph{absolute type} of $\mathfrak P$ is the pair $(e,f)$ of its ramification index and residue degree over $p$. For a Galois extension $M/A$ of number fields, all primes of $M$ above a prime $\mathfrak p$ of $A$ have the same ramification index $e$ and residue degree $f$ in $M/A$; the pair $(e,f)$ is the \emph{type} of $\mathfrak p$ in $M/A$. \end{definition} As in the introduction, let $\lambda=2^{9/4}\sqrt{3615}$, where $3615=3\cdot5\cdot241$. \begin{theorem}\label{tw:field-family} For every power of two $m\ge2^{49}$ there is a finite Galois extension $K/B$ of degree $m$, contained in $B_G$, such that $G_B\to\overline G_B$ factors through $\Gal(K/B)$. In particular $K$ contains the fixed field of $D_4G_B$, hence the fixed field of $D_3G_B$ and the Kummer field $E$. Every such $K$ is totally imaginary. Let $\iota_1$ also denote complex conjugation at any place of $K$ above $v_1$; these complex conjugations are conjugate in $\Gal(K/B)$, one of them is the image of the element $\iota_1$ of Definition~\ref{tw:local-generators}, and the statements below do not depend on the choice. Let $F=K^{\langle\iota_1\rangle}$ and $d=[F:\Q]=m$, let $(b,c)$ be the signature of $F$ and $\theta=c/d$, and put $\theta_*=\thetastar=1/2-2^{-17}$. Then $F$ has a real place, \[ \rd(K)=\rd(F)=\lambda,\qquad \theta_*\le\theta<\frac12 , \] and $b/d=1/(2N_\iota)$, where $N_\iota\ge2^{15}$ is the number of conjugates of $\iota_1$ in $\Gal(K/B)$. The extension $K/F$ is quadratic and unramified at every finite place. For $r\in\{2,3,5,7,29\}$, every prime of $F$ above $r$ splits in $K$, and the primes of $K$ and of $F$ above $r$ have the absolute type $(e_r,f_r)$ of Table~\ref{tw:types-table}; that is, $K/F$ has uniform local types $(e_r,f_r)_{r\in\{2,3,5,7,29\}}$ in the sense of Definition~\ref{geo:uniform-types}. The completions of $K$ at the primes above $2,3,5,7$ and $29$ are the local extensions listed in Table~\ref{tw:types-table}. \end{theorem} \begin{table}[htbp] \centering \caption{Absolute types of the places of $K$ and $F$ above $2$, $3$, $5$, $7$ and $29$, the corresponding local extensions of the completions of $B$, and the contributions of these primes to $\rd(K)$. The last row is the prime $241$, which ramifies in $B$; $K/B$ is unramified above it.} \label{tw:types-table} \begin{tabular}{llll} \toprule $r$ & $(e_r,f_r)$ & local extension over $B$ & factor of $\rd(K)$\\ \midrule $2$ & $(8,4)$ & $L_2/\Q_2$, group $D$ & $2^{9/4}$\\ $3$ & $(2,2)$ & $\Q_3(\sqrt{-1},\sqrt3)/\Q_3$, group $C_2\times C_2$ & $3^{1/2}$\\ $5$ & $(2,2)$ & $\Q_5(\sqrt2,\sqrt5)/\Q_5$, group $C_2\times C_2$ & $5^{1/2}$\\ $7$ & $(1,8)$ & unramified of degree $4$ over $B_{\mathfrak t_0}$ & $1$\\ $29$ & $(1,4)$ & unramified of degree $4$ over $\Q_{29}$ & $1$\\ \midrule $241$ & $e=2$ & unramified & $241^{1/2}$\\ \bottomrule \end{tabular} \end{table} \begin{proof} \emph{Existence.} By Proposition~\ref{tw:infinite}, $G_B$ has open normal subgroups of arbitrarily large index, and $D_4G_B$ is open because $G_B$ is finitely generated. Given $m$, choose an open normal subgroup $\mathcal U'\subset D_4G_B$ of $G_B$ of index at least $m$, and put $N'=D_4G_B/\mathcal U'$. A nontrivial normal subgroup of a finite $2$-group meets its center, so $N'$ contains normal subgroups of $G_B/\mathcal U'$ of every order dividing $|N'|$. Take one of order $[G_B:\mathcal U']/m$, which divides $|N'|=[G_B:\mathcal U']/2^{49}$. Its preimage $\mathcal U$ is open and normal in $G_B$, contained in $D_4G_B$, and of index $m$. Let $K$ be its fixed field in $B_G$. Conversely, every $K$ as in the statement is the fixed field of an open normal subgroup $\mathcal U\subset D_4G_B$. We fix such a $K$ and put $H=\Gal(K/B)=G_B/\mathcal U$. \emph{Complex places.} Since $\sqrt{-1}=\sqrt{\alpha_0}\in E\subset K$, the field $K$ is totally imaginary, and $[K:F]=2$. Let $j\in\{1,2\}$. The group $H$ permutes the places of $K$ above $v_j$ transitively. If an embedding $\sigma\colon K\to\C$ induces such a place and $\iota_\sigma\in H$ is its complex conjugation, so that $\bar\sigma=\sigma\circ\iota_\sigma$, then for $h\in H$ we have $\overline{\sigma\circ h}=(\sigma\circ h)\circ(h^{-1}\iota_\sigma h)$, so the complex conjugation at the place of $\sigma\circ h$ is $h^{-1}\iota_\sigma h$. By the definition of the local map at $v_j$, the image of $\iota_j$ in $H$ is the complex conjugation at the place of $K$ below the place of $B_S$ fixed in Definition~\ref{tw:local-generators}. Hence every complex conjugation at a place of $K$ above $v_j$ is conjugate in $H$ to the image of $\iota_j$, and has elementary image $\bar\iota_j$, since conjugation preserves elementary images. In particular, two choices of $\iota_1$ in the statement are conjugate in $H$, so an element of $H$ maps the field $F$ of one choice onto that of the other, and the statements of the theorem hold for both choices or for neither. Let $\sigma_1\colon K\to\C$ induce the chosen place, so that $\bar\sigma_1=\sigma_1\circ\iota_1$ and $\sigma_1(F)\subset\R$. The embeddings above $v_1$ are $\sigma_1\circ h$ with $h\in H$, and two of them agree on $F$ exactly when they differ by right multiplication by an element of $\langle\iota_1\rangle$. The restriction of $\sigma_1\circ h$ to $F$ is real exactly when $\sigma_1\circ\iota_1h$ and $\sigma_1\circ h$ agree on $F$, that is, when $h^{-1}\iota_1h\in\langle\iota_1\rangle$, that is, when $h\in C_H(\iota_1)$. Similarly, if $\iota'$ is complex conjugation for an embedding $\sigma_2$ above $v_2$, the restriction of $\sigma_2\circ h$ to $F$ is real exactly when $h^{-1}\iota'h=\iota_1$. This never happens, because conjugation preserves elementary images and $\bar\iota'=\bar\iota_2\ne\bar\iota_1$ (Lemma~\ref{tw:retained-quotient}(d)). Hence $b=|C_H(\iota_1)|/2\ge1$, so $F$ has a real place and $\theta=(1-b/d)/2<1/2$. Moreover $d=|H|$, so $b/d=|C_H(\iota_1)|/(2|H|)$ is half the reciprocal of the size $N_\iota$ of the conjugacy class of $\iota_1$ in $H$, that is, $b/d=1/(2N_\iota)$. This class maps onto the class in $\overline G_B$ of the element $\iota_1$ of Definition~\ref{tw:local-generators}, which has $2^{15}$ elements by Lemma~\ref{tw:retained-quotient}(c). Hence $N_\iota\ge2^{15}$, $b/d\le2^{-16}$ and $\theta\ge(1-2^{-16})/2=\theta_*$. \emph{Local structure and splitting.} Let $\mathfrak P$ be a prime of $K$ above $\mathfrak p\in S\cup\{\mathfrak t_0,\mathfrak t_1,\mathfrak t_2\}$. Its decomposition group in $H$ is conjugate to the image of the local map at $\mathfrak p$. Since $\mathcal U\subset D_3G_B$, Lemma~\ref{tw:retained-quotient}(b) shows that this image is $D$, $C_2\times C_2$ or $C_4$, the last one cyclic on the Frobenius, and that the prescribed local group injects into $H$. Therefore the completion $K_{\mathfrak P}$ is $L_2$ over $B_{\mathfrak p_j}=\Q_2$, the field $\Q_p(\sqrt u,\sqrt p)$ over $B_{\mathfrak p}=\Q_p$ for $p=3,5$, and the unramified extension of degree four of $B_{\mathfrak t_k}$. Since $B_{\mathfrak t_0}$ is unramified of degree two over $\Q_7$ and $B_{\mathfrak t_1}=B_{\mathfrak t_2}=\Q_{29}$, the places of $K$ above $2,3,5,7$ and $29$ have the absolute types of Table~\ref{tw:types-table}. Every element of the decomposition group has elementary image in the span of the images of the local generators at $\mathfrak p$, whereas every conjugate of $\iota_1$ has elementary image $\bar\iota_1$. By Lemma~\ref{tw:retained-quotient}(d), $\iota_1$ lies in no decomposition group of a prime above $S\cup\{\mathfrak t_0,\mathfrak t_1,\mathfrak t_2\}$, so the decomposition group of $\mathfrak P$ in $\Gal(K/F)=\langle\iota_1\rangle$ is trivial. Hence the places of $F$ above $2,3,5,7,29$ split in $K$, and they have the same absolute types. \emph{Ramification and root discriminant.} The extension $K/B$, and hence $K/F$, is unramified at every prime outside $S$, in particular at $(\sqrt{241})$. At primes above $S$ the inertia group of $K/F$ is trivial, because it lies in the decomposition group. So $K/F$ is unramified at every finite place, its relative discriminant is trivial, $|\Delta_K|=|\Delta_F|^2$, and $\rd(K)=\rd(F)$. To compute $\rd(K)$ we use $|\Delta_K|=241^{[K:B]}N\mathfrak d_{K/B}$. Since $K/B$ is Galois, for $\mathfrak p\in S$ all primes $\mathfrak P$ above $\mathfrak p$ have the same ramification index $e$, residue degree and different exponent $\operatorname{ord}_{\mathfrak P}(\mathfrak D_{K/B})$, where $\operatorname{ord}_{\mathfrak P}$ and $\operatorname{ord}_{\mathfrak p}$ are the normalized valuations at $\mathfrak P$ and $\mathfrak p$, and $\operatorname{ord}_{\mathfrak p}(\mathfrak d_{K/B})=[K:B]\,\operatorname{ord}_{\mathfrak P}(\mathfrak D_{K/B})/e$. For $\mathfrak p$ above $3$ or $5$ we have $e=2$ and $\operatorname{ord}_{\mathfrak P}(\mathfrak D_{K/B})=1$, since over the unramified extension $\Q_p(\sqrt u)$ the different of $\Q_p(\sqrt u,\sqrt p)$ is generated by $2\sqrt p$. For $\mathfrak p_1,\mathfrak p_2$, Lemma~\ref{tw:local-two}(d) gives $\operatorname{ord}_{\mathfrak P}(\mathfrak D_{K/B})/e=9/4$. Since $[K:\Q]=2[K:B]$, the base contributes $241^{1/2}$, each of the two primes above $3$ contributes $3^{1/4}$, each of the two above $5$ contributes $5^{1/4}$, and each dyadic prime contributes $2^{9/8}$. Hence $\rd(K)=2^{9/4}\,3^{1/2}\,5^{1/2}\,241^{1/2}=\lambda$. \end{proof} \begin{remark}\label{tw:lean-remark} The Lean development \cite{NaslundLean2026} formalizes a version of this construction; Remark~\ref{an:lean-hypothesis} states what it proves and the numerical hypothesis it assumes. \end{remark} \section{\texorpdfstring{The Kummer Field and Its Quadratic $L$-Functions}{The Kummer Field and Its Quadratic L-Functions}}\label{gf:section} The analytic estimate of Section~\ref{an:section} compares every field of Theorem~\ref{tw:field-family} with one fixed subfield, the Kummer field \[ E=B(\sqrt{\alpha_0},\ldots,\sqrt{\alpha_7})=B(\sqrt V) \] of Subsection~\ref{tw:global-section}, with $\alpha_0,\ldots,\alpha_7$ as in \eqref{tw:alpha} and $V$ as in Lemma~\ref{tw:base}(d). It is the maximal elementary abelian $2$-extension of $B$ unramified at the finite primes outside $S$, that is, the fixed field of the Frattini subgroup of $G_B$ (Lemma~\ref{gf:in-tower}). The Lean formalization calls $E$ the genus field; we do not use that name here. This section shows that $E$ lies in every field of Theorem~\ref{tw:field-family}. It then writes $\zeta_E$ as the product of $\zeta_B$ and $255$ quadratic Hecke $L$-functions of $B$, and determines their conductors, gamma factors, root numbers and Euler factors. For $v,e\in\Ftwo^8$ put $\langle v,e\rangle=\sum_{i=0}^7v_ie_i\in\Ftwo$; as in Section~\ref{tw:tower}, vectors are written as strings of their coordinates. For a nonzero ideal $\mathfrak a$ of a number field, $\mathrm N\mathfrak a$ denotes its absolute norm. The number of positive divisors of an integer $n\ge1$ is $\tau(n)$. \subsection{The Characters of the Kummer Field} By Lemma~\ref{tw:kummer-field}, $[E:B]=256$, and the elementary image $\bar g\in\Ftwo^8$ of Definition~\ref{tw:elementary-image}, given by $g(\sqrt{\alpha_i})=(-1)^{\bar g_i}\sqrt{\alpha_i}$, identifies $\Gal(E/B)$ with $\Ftwo^8$. \begin{definition}\label{gf:characters} For $e\in\Ftwo^8$ put \[ \alpha_e=\prod_{i=0}^7\alpha_i^{e_i},\qquad B_e=B(\sqrt{\alpha_e}), \] and let $\chi_e$ be the character $g\mapsto(-1)^{\langle\bar g,e\rangle}$ of $\Gal(E/B)$. \end{definition} The character $\chi_e$ takes the values $\pm1$; in the additive notation of Definition~\ref{tw:elementary-image}, $\chi_e(g)=(-1)^{\chi_{\alpha_e}(g)}$. Since $g(\sqrt{\alpha_e})=\chi_e(g)\sqrt{\alpha_e}$, the character $\chi_e$ cuts out $B_e$, and the $256$ characters $\chi_e$ are all the characters of $\Gal(E/B)$. For $e\ne0$ the extension $B_e/B$ is quadratic, because the classes of $\alpha_0,\ldots,\alpha_7$ are independent in $B^\times/B^{\times2}$ (Lemma~\ref{tw:base}(d)). \begin{lemma}\label{gf:in-tower} The fixed field of the Frattini subgroup $D_2G_B$ of $G_B$ is $E$. Consequently $E$ is contained in every field $K$ of Theorem~\ref{tw:field-family}. \end{lemma} \begin{proof} Let $G_S$ be the group of Subsection~\ref{tw:global-section}, so that $G_S/D_2G_S=\Gal(E/B)$ (Lemma~\ref{tw:kummer-field}), and let $G_B=G_S/N$ as in Definition~\ref{tw:GB}. By Lemma~\ref{tw:GB-basic}(b), $N\subseteq D_2G_S$, so $G_B/D_2G_B=G_S/D_2G_S=\Gal(E/B)$. A field $K$ of Theorem~\ref{tw:field-family} contains the fixed field of $D_4G_B\subseteq D_2G_B$, and hence contains $E$. \end{proof} \subsection{Local Data} \begin{definition}\label{gf:frobenius-vector} For a prime $\mathfrak p\notin S$ of $B$, the \emph{Frobenius vector} $\overline{\Frob}_{\mathfrak p}\in\Ftwo^8$ is the elementary image of the Frobenius of $\mathfrak p$ in $\Gal(E/B)$, that is, of the Frobenius element $\Frob_{\mathfrak p}$ of Subsection~\ref{tw:global-section}. \end{definition} For $\mathfrak p=\mathfrak t_k$ it is the vector of $\Frob_{\mathfrak t_k}$ in Table~\ref{tw:vector-table}. By Euler's criterion, $(\overline{\Frob}_{\mathfrak p})_i=1$ exactly when $\alpha_i$ is not a square in $\mathcal O_B/\mathfrak p$. For a place $\mathfrak p$ of $B$ we write $(\cdot,\cdot)_{\mathfrak p}$ for the quadratic Hilbert symbol of $B_{\mathfrak p}$. \begin{proposition}\label{gf:local-data} Let $e\in\Ftwo^8\setminus\{0\}$. For a prime $\mathfrak p$ of $B$ let $c_{\mathfrak p}(e)$ be the exponent of $\mathfrak p$ in the conductor of $\chi_e$, and if $c_{\mathfrak p}(e)=0$, let $\chi_e(\mathfrak p)\in\{\pm1\}$ be the value of $\chi_e$ at the Frobenius of $\mathfrak p$. \begin{enumerate} \item If $\mathfrak p\notin S$, then $c_{\mathfrak p}(e)=0$ and $\chi_e(\mathfrak p)=(-1)^{\langle\overline{\Frob}_{\mathfrak p},e\rangle}$. Let $p$ be the rational prime below $\mathfrak p$. If $p$ splits in $B$, then $\mathfrak p=(p,\sqrt{241}-r)$ with $r^2\equiv241\pmod p$, and $(\overline{\Frob}_{\mathfrak p})_i=1$ exactly when the image of $\alpha_i$ under $\sqrt{241}\mapsto r$ is a quadratic nonresidue modulo $p$. If $p$ is inert, then $(\overline{\Frob}_{\mathfrak p})_i=1$ exactly when $N_{B/\Q}(\alpha_i)$ is a quadratic nonresidue modulo $p$. If $p=241$, the same holds with $\sqrt{241}\mapsto0$. \item Let $\mathfrak p$ lie above $p\in\{3,5\}$, let $u=-1$ if $p=3$ and $u=2$ if $p=5$, and let $\alpha_i$ be the generator of $\mathfrak p$. If $(u,\alpha_e)_{\mathfrak p}=-1$, then $c_{\mathfrak p}(e)=1$. Otherwise $c_{\mathfrak p}(e)=0$ and $\chi_e(\mathfrak p)=(-\alpha_i,\alpha_e)_{\mathfrak p}$. The Frobenius generator $\varphi_{\mathfrak p}$ of Definition~\ref{tw:local-generators}, which fixes $\sqrt{\alpha_i}$, acts on square roots by $(-\alpha_i,\cdot)_{\mathfrak p}$ (Table~\ref{tw:vector-table}). \item Let $\mathfrak p=\mathfrak p_j$. If $(5,\alpha_e)_{\mathfrak p}=-1$, then $c_{\mathfrak p}(e)=3$. If $(5,\alpha_e)_{\mathfrak p}=1$ and $(-1,\alpha_e)_{\mathfrak p}=-1$, then $c_{\mathfrak p}(e)=2$. Otherwise $c_{\mathfrak p}(e)=0$ and $\chi_e(\mathfrak p)=(-2,\alpha_e)_{\mathfrak p}$. \item The real place $v_k$ becomes complex in $B_e$, that is, $\chi_e$ is nontrivial at $v_k$, exactly when $\alpha_e<0$ at $v_k$. \end{enumerate} \end{proposition} \begin{proof} For (1), $\mathfrak p$ is odd and $\alpha_e$ is a unit at $\mathfrak p$, so $\mathfrak p$ is unramified in $B_e$. By Euler's criterion the Frobenius multiplies $\sqrt{\alpha_e}$ by the quadratic residue symbol of $\alpha_e$ modulo $\mathfrak p$, which is $\prod_i(\alpha_i/\mathfrak p)^{e_i}$. For a split prime, $\mathcal O_B/\mathfrak p\cong\mathbb F_p$ via $\sqrt{241}\mapsto r$. For an inert prime, $\mathcal O_B/\mathfrak p\cong\mathbb F_{p^2}$ and the conjugation of $B$ induces the Frobenius $x\mapsto x^p$ of the residue field, so the residue of $N_{B/\Q}(\alpha)$ is $\bar\alpha^{p+1}$. An element of $\mathbb F_{p^2}^\times$ is a square exactly when $\bar\alpha^{(p^2-1)/2}=(\bar\alpha^{p+1})^{(p-1)/2}$ equals $1$, that is, when its norm is a square in $\mathbb F_p$. The prime above $241$ has residue field $\mathbb F_{241}$ and $\sqrt{241}\mapsto0$. For (2) and (3) we use local class field theory for $B_{\mathfrak p}=\Q_p$. The local Artin map sends $b\in\Q_p^\times$ to the automorphism multiplying $\sqrt{\alpha_e}$ by $(b,\alpha_e)_{\mathfrak p}$. Units map onto inertia, a uniformizer maps to a Frobenius, and the conductor exponent is the least $n\ge0$ such that $b\mapsto(b,\alpha_e)_{\mathfrak p}$ is trivial on $U^{(n)}$, where $U^{(0)}=\mathbb Z_p^\times$ and $U^{(n)}=1+p^n\mathbb Z_p$ \cite{MilneCFT}. For $p=3,5$, the unit group modulo squares is $\{1,u\}$, and $U^{(1)}$ consists of squares by Hensel's lemma. Hence $c_{\mathfrak p}(e)\le1$, with equality exactly when $(u,\alpha_e)_{\mathfrak p}=-1$. In the unramified case the Frobenius is the image of any uniformizer, such as $-\alpha_i$. For $p=2$, the squares of $\mathbb Z_2^\times$ form $U^{(3)}$. Moreover $U^{(2)}=U^{(3)}\cup5U^{(3)}$ and $U^{(0)}=U^{(1)}=\{\pm1,\pm5\}U^{(3)}$. The character is thus trivial on $U^{(3)}$. It is trivial on $U^{(2)}$ exactly when $(5,\alpha_e)=1$, and on $U^{(0)}$ exactly when also $(-1,\alpha_e)=1$. This gives the stated exponents; the exponent $1$ cannot occur since $U^{(0)}=U^{(1)}$. In the unramified case the Frobenius is the image of the uniformizer $-2$. For (4), the real place $v_k$ splits in $B_e$ exactly when $\alpha_e>0$ at $v_k$. \end{proof} Each condition in the proposition is linear in $e$: \begin{lemma}\label{gf:linear} Let $e\in\Ftwo^8$. Then $(5,\alpha_e)_{\mathfrak p_j}=(-1)^{\langle\bar x_j,e\rangle}$, $(-1,\alpha_e)_{\mathfrak p_j}=(-1)^{\langle\bar y_j,e\rangle}$ and $(-2,\alpha_e)_{\mathfrak p_j}=(-1)^{\langle\bar z_j,e\rangle}$. For a prime $\mathfrak p$ above $3$ or $5$, with $u$ as in Proposition~\ref{gf:local-data}(2) and $\alpha_i$ the generator of $\mathfrak p$, $(u,\alpha_e)_{\mathfrak p}=(-1)^{\langle\bar\tau_{\mathfrak p},e\rangle}$ and $(-\alpha_i,\alpha_e)_{\mathfrak p}=(-1)^{\langle\bar\varphi_{\mathfrak p},e\rangle}$. Finally, $\alpha_e<0$ at $v_k$ exactly when $\langle\bar\iota_k,e\rangle=1$. \end{lemma} \begin{proof} In each case $(b,\alpha_e)_{\mathfrak p}=(-1)^{\langle w,e\rangle}$, where $w$ is the elementary image of the local generator that acts on square roots by $(b,\cdot)_{\mathfrak p}$ in Table~\ref{tw:vector-table}: at $\mathfrak p_j$ the vectors $\bar x_j,\bar y_j,\bar z_j$ belong to $b=5,-1,-2$, and above $3$ and $5$ the images of the inertia and Frobenius generators belong to $b=u$ and $b=-\alpha_i$. Likewise the complex conjugation $\iota_k$ negates $\sqrt{\alpha_e}$ exactly when $\alpha_e<0$ at $v_k$ (Lemma~\ref{tw:local-groups}(a)), and it multiplies $\sqrt{\alpha_e}$ by $(-1)^{\langle\bar\iota_k,e\rangle}$. \end{proof} We put $\nu_k(e)=\langle\bar\iota_k,e\rangle\in\{0,1\}$ for $k=1,2$. By Proposition~\ref{gf:local-data}, the conductor of $\chi_e$ is $\mathfrak f_e=\prod_{\mathfrak p\in S}\mathfrak p^{c_{\mathfrak p}(e)}$. \begin{corollary}\label{gf:types} The type (Definition~\ref{tw:type}) of a prime $\mathfrak p$ of $B$ in $E/B$ is $(4,2)$ at $\mathfrak p_1,\mathfrak p_2$; $(2,2)$ at $\mathfrak q_1,\mathfrak q_2,\mathfrak r_1,\mathfrak r_2$; and $(1,1)$ or $(1,2)$ at $\mathfrak p\notin S$, according as $\overline{\Frob}_{\mathfrak p}=0$ or not. In particular $\mathfrak t_0,\mathfrak t_1,\mathfrak t_2$ have type $(1,2)$ in $E/B$. \end{corollary} \begin{proof} For $\mathfrak p\in S$, local class field theory identifies the decomposition group of $\mathfrak p$ in $\Gal(E/B)=\Ftwo^8$ with the image of $\Q_p^\times$ under the local Artin map, and the inertia group with the image of the units. By Lemma~\ref{gf:linear}, at $\mathfrak p_j$ these are spanned by $\bar x_j,\bar y_j,\bar z_j$ and by $\bar x_j,\bar y_j$; the three vectors are independent (Table~\ref{tw:vector-table}), so the decomposition group has order $8$ and inertia has order $4$. Above $3$ and $5$ the two elementary images are independent, which gives $(2,2)$. For $\mathfrak p\notin S$ the decomposition group is generated by $\overline{\Frob}_{\mathfrak p}$. The Frobenius vectors of $\mathfrak t_0,\mathfrak t_1,\mathfrak t_2$ are nonzero (Table~\ref{tw:vector-table}). \end{proof} \subsection{The Zeta Function of the Kummer Field} Let $\chi_B$ be the Dirichlet character $n\mapsto(n/241)$. \begin{proposition}\label{gf:factorization} For $\operatorname{Re}s>1$, \[ \zeta_E(s)=\zeta_B(s)\prod_{e\in\Ftwo^8\setminus\{0\}}L(s,\chi_e), \qquad \zeta_B(s)=\zeta(s)L(s,\chi_B), \] where $L(s,\chi_e)=\zeta_{B_e}(s)/\zeta_B(s)$ is the Hecke $L$-function of $\chi_e$. \end{proposition} \begin{proof} Compare Euler factors at a prime $\mathfrak p$ of $B$. Let $T_{\mathfrak p}\subseteq Z_{\mathfrak p}\subseteq G=\Gal(E/B)$ be its inertia and decomposition groups, and put $f_{\mathfrak p}=|Z_{\mathfrak p}/T_{\mathfrak p}|$ and $g_{\mathfrak p}=|G/Z_{\mathfrak p}|$. The Euler factor of $\zeta_E$ at $\mathfrak p$ is $(1-X^{f_{\mathfrak p}})^{-g_{\mathfrak p}}$ with $X=\mathrm N\mathfrak p^{-s}$. A character nontrivial on $T_{\mathfrak p}$ contributes the factor $1$. The characters trivial on $T_{\mathfrak p}$ are those of $G/T_{\mathfrak p}$; they restrict onto the $f_{\mathfrak p}$ characters of the cyclic group $Z_{\mathfrak p}/T_{\mathfrak p}$, each with $g_{\mathfrak p}$ extensions. Hence their factors multiply to $\prod_{\zeta^{f_{\mathfrak p}}=1}(1-\zeta X)^{-g_{\mathfrak p}} =(1-X^{f_{\mathfrak p}})^{-g_{\mathfrak p}}$. The same identity for $B_e/B$ identifies $L(s,\chi_e)$ with $\zeta_{B_e}/\zeta_B$. The second identity is the case of $B/\Q$. Since $241\equiv1\pmod8$, the prime $2$ splits in $B$, and quadratic reciprocity shows that the character of $B/\Q$ is $\chi_B$. \end{proof} Put $\Gamma_\R(s)=\pi^{-s/2}\Gamma(s/2)$ and $\Gamma_\C(s)=2(2\pi)^{-s}\Gamma(s)$, so that $\Gamma_\C(s)=\Gamma_\R(s)\Gamma_\R(s+1)$ by the duplication formula. For a number field $M$ with discriminant $\Delta_M$ and signature $(r_1(M),r_2(M))$, the completed Dedekind zeta function is \[ \Lambda_M(s)=|\Delta_M|^{s/2}\Gamma_\R(s)^{r_1(M)}\Gamma_\C(s)^{r_2(M)} \zeta_M(s). \] \begin{proposition}\label{gf:degree-two} Let $e\ne0$, write $L(s,\chi_e)=\sum_{n\ge1}a_n(e)n^{-s}$, and put \[ Q_e=241\,\mathrm N\mathfrak f_e,\qquad \Lambda(s,\chi_e)=Q_e^{s/2}\,\Gamma_\R(s+\nu_1(e))\,\Gamma_\R(s+\nu_2(e))\, L(s,\chi_e). \] \begin{enumerate} \item The coefficients $a_n(e)$ are integers with $|a_n(e)|\le\tau(n)$. The Euler factor at a rational prime $p$ is $\prod_{\mathfrak p\mid p}(1-\chi_e(\mathfrak p)\mathrm N\mathfrak p^{-s})^{-1}$, with $\chi_e(\mathfrak p)=0$ when $c_{\mathfrak p}(e)>0$. \item The function $\Lambda(s,\chi_e)$ equals $\Lambda_{B_e}(s)/\Lambda_B(s)$. It is entire of order at most one and satisfies $\Lambda(s,\chi_e)=\Lambda(1-s,\chi_e)$; that is, the root number of $L(s,\chi_e)$ is $1$. \item Exactly $63$ of the characters have $\nu_1(e)=\nu_2(e)=0$, exactly $64$ have $\nu_1(e)=\nu_2(e)=1$, and $128$ have $\nu_1(e)\ne\nu_2(e)$. \end{enumerate} \end{proposition} \begin{proof} The Euler factor in (1) is Proposition~\ref{gf:factorization} applied prime by prime. Above a rational prime there are at most two primes of $B$. So the local factor is either a product of at most two factors $(1-cp^{-s})^{-1}$, or a single factor $(1-cp^{-2s})^{-1}$, with $c\in\{0,\pm1\}$. In both cases the coefficient of $p^{-ks}$ has modulus at most $k+1=\tau(p^k)$. Multiplicativity gives (1). For (2), the completed Dedekind zeta functions satisfy $\Lambda_M(1-s)=\Lambda_M(s)$, and $L(s,\chi_e)$ extends to an entire function because $\chi_e$ is nontrivial \cite{Hecke1920,Tate1967}. The discriminants satisfy $|\Delta_{B_e}|=|\Delta_B|^2\,\mathrm N\mathfrak d_{B_e/B}$, where $\mathfrak d_{B_e/B}$ is the relative discriminant, and the conductor--discriminant formula gives $\mathfrak d_{B_e/B}=\mathfrak f_e$ for the quadratic extension $B_e/B$. Hence $|\Delta_{B_e}|/|\Delta_B|=241\,\mathrm N\mathfrak f_e=Q_e$. If $\nu_k(e)=0$, the place $v_k$ splits into two real places of $B_e$ and contributes $\Gamma_\R(s)^2/\Gamma_\R(s)=\Gamma_\R(s)$. If $\nu_k(e)=1$, it becomes one complex place and contributes $\Gamma_\C(s)/\Gamma_\R(s)=\Gamma_\R(s+1)$. Thus $\Lambda_{B_e}/\Lambda_B=\Lambda(s,\chi_e)$. It is the completed Hecke $L$-function of $\chi_e$, which is entire of order at most one \cite{Hecke1920,Tate1967}. For (3), the vectors $\bar\iota_1,\bar\iota_2$ are linearly independent. So $e\mapsto(\nu_1(e),\nu_2(e))$ maps $\Ftwo^8$ onto $\Ftwo^2$ with fibers of size $64$. Removing $e=0$ from the fiber over $(0,0)$ gives the counts. \end{proof} \begin{definition}\label{gf:pure-mixed} Let $e\ne0$. The gamma factor of $\chi_e$ is \emph{pure} if $\nu_1(e)=\nu_2(e)=\nu$, and then it equals $\Gamma_\R(s+\nu)^2$; otherwise it is \emph{mixed}, and then it equals $\Gamma_\R(s)\Gamma_\R(s+1)=\Gamma_\C(s)$. \end{definition} \begin{corollary}\label{gf:conductors} For $e\ne0$, $Q_e=241\cdot2^i\cdot m$, where $2^i=2^{c_{\mathfrak p_1}(e)+c_{\mathfrak p_2}(e)}\in\{1,4,8,16,32,64\}$ and $m=3^{c_{\mathfrak q_1}(e)+c_{\mathfrak q_2}(e)}5^{c_{\mathfrak r_1}(e)+c_{\mathfrak r_2}(e)}$ divides $225$. The number of characters $\chi_e$ with given $2^i$ and $m$ is given by the table \begin{equation}\label{gf:conductor-table} \begin{array}{c|ccccccccc|c} 2^i\backslash m&1&3&5&9&15&25&45&75&225&\text{total}\\\hline 1&0&2&2&1&4&1&2&2&1&15\\ 4&2&4&4&2&8&2&4&4&2&32\\ 8&4&8&8&4&16&4&8&8&4&64\\ 16&1&2&2&1&4&1&2&2&1&16\\ 32&4&8&8&4&16&4&8&8&4&64\\ 64&4&8&8&4&16&4&8&8&4&64\\\hline \text{total}&15&32&32&16&64&16&32&32&16&255 \end{array} \end{equation} In particular $723\le Q_e\le241\cdot14400=3470400$. \end{corollary} \begin{proof} Only the primes of $S$ divide the conductors, so $Q_e=241\,\mathrm N\mathfrak f_e$ with \[ \mathrm N\mathfrak f_e=2^{c_{\mathfrak p_1}(e)+c_{\mathfrak p_2}(e)}3^{c_{\mathfrak q_1}(e)+c_{\mathfrak q_2}(e)}5^{c_{\mathfrak r_1}(e)+c_{\mathfrak r_2}(e)}. \] By Proposition~\ref{gf:local-data}, $c_{\mathfrak p_j}(e)\in\{0,2,3\}$ and $c_{\mathfrak q_j}(e),c_{\mathfrak r_j}(e)\in\{0,1\}$, and by Lemma~\ref{gf:linear} these exponents are determined by the vectors of Table~\ref{tw:vector-table}; counting the $255$ vectors $e\ne0$ gives the table, which the supplementary program \texttt{check255.gp} also prints (Remark~\ref{gf:checks}). No nontrivial character has $\mathrm N\mathfrak f_e=1$, since $B$ has class number one and a unit of norm $-1$ (Lemma~\ref{tw:base}(a),(b)), so that its narrow class number is one. The range of $Q_e$ can be read off from the table. \end{proof} \begin{remark}\label{gf:checks} The supplementary program \texttt{lfun241.py} computes, for each $e\ne0$, the integer $Q_e$, the pair $(\nu_1(e),\nu_2(e))$ and the coefficients $a_n(e)$ for $n\le N_e=4\lfloor\sqrt{Q_e}\rfloor+1$, the truncation point used in Subsection~\ref{an:value-section}, from Propositions~\ref{gf:local-data} and~\ref{gf:degree-two}(1); these are the data used in Section~\ref{an:section}. The supplementary program \texttt{check255.gp} compares all $255$ of them with PARI/GP's own number fields and $L$-functions \cite{PARI}. For each $e$ it computes the maximal order of $B_e$, checks $|\Delta_{B_e}|=241\,Q_e$ and the pair $(\nu_1(e),\nu_2(e))$ against the signature of $B_e$, and checks every coefficient $a_n(e)$ with $n\le N_e$ against the Dirichlet series of $\zeta_{B_e}/\zeta_B$. All agree. It also prints the table \eqref{gf:conductor-table}, the counts of Proposition~\ref{gf:degree-two}(3) and the range of $Q_e$. These comparisons check the implementation; the identification of the $L$-functions rests on Propositions~\ref{gf:local-data} and~\ref{gf:degree-two}. \end{remark} \section{The Relative Zeta Value and an Explicit Upper Bound}\label{an:section} We bound the relative zeta value $L_F(1)$ needed in the geometric construction. Logarithms of Dedekind zeta functions are normalized by the degree: for a number field $A$ we work with $[A:\Q]^{-1}\log\zeta_A(s)$. \begin{definition}\label{an:family} Let $\mathcal K$ be the set of all finite Galois extensions $K/B$ contained in $B_G$, of degree a power of two at least $2^{49}$, such that $G_B\to\overline G_B$ factors through $\Gal(K/B)$ (notation of Definition~\ref{tw:GB}). These are the fields $K$ of Theorem~\ref{tw:field-family}. \end{definition} For $K\in\mathcal K$ we use the notation of Theorem~\ref{tw:field-family}: $\iota_1$ is a complex conjugation at a place of $K$ above $v_1$, $F=K^{\langle\iota_1\rangle}$, $d=[F:\Q]$, so that $[K:\Q]=2d$, $(b,c)$ is the signature of $F$, $\theta=c/d$, and $\lambda=2^{9/4}\sqrt{3615}$ is the root discriminant of both $K$ and $F$. Put \[ \ell=\log\lambda=\tfrac94\log2+\tfrac12\log3615=5.65600472355\ldots \] Different choices of $\iota_1$ are conjugate in $\Gal(K/B)$ and give conjugate fields $F$; hence $b$, $c$, $\theta$ and the function $L_F(s)=\zeta_K(s)/\zeta_F(s)$ depend only on $K$. The extension $K/F$ is quadratic and unramified at every finite place, $K$ is totally imaginary, and $b\ge1$. Let $\gamma$ be Euler's constant. For a prime $\mathfrak p$ of $B$, let $e_K(\mathfrak p)$ and $f_K(\mathfrak p)$ be its ramification index and residue degree in the Galois extension $K/B$, so that $(e_K(\mathfrak p),f_K(\mathfrak p))$ is the type of $\mathfrak p$ in $K/B$ (Definition~\ref{tw:type}). \begin{theorem}\label{an:ceiling} There is an $m_0$ such that every $K\in\mathcal K$ with $[K:B]\ge m_0$ satisfies \[ \frac1d\log L_F(1)1$ (Subsection~\ref{an:reduction}). The Tsfasman--Vl\u{a}du\c{t} inequality bounds the error of this shift (Subsection~\ref{an:budget-section}, Proposition~\ref{an:prime-budget}). Finally, $\zeta_K(\sigma)$ is bounded prime by prime by $\zeta_E(\sigma)$, for the Kummer field $E$ of Section~\ref{gf:section}, minus corrections at the primes above $2$, $7$ and $29$, whose types in $K/B$ are known, and at the primes of $\mathcal P_4$ (Definition~\ref{an:P4}), whose residue degree in $K/B$ is at least $4$ (Proposition~\ref{an:Ystar}); $\zeta_E(\sigma)$ is evaluated rigorously in Subsections~\ref{an:afe-section}--\ref{an:value-section} (Proposition~\ref{an:genus-value}). Subsection~\ref{an:census} computes these corrections and the lower bounds for the types in $K/B$ that Proposition~\ref{an:prime-budget} uses, and the last subsection combines the steps. Throughout, $\sigma$ denotes a real number greater than $1$ and $\varepsilon=\sigma-1$; in this section $\varepsilon$ is this positive number, not the unit $\alpha_1$ of \eqref{tw:alpha}. \subsection{\texorpdfstring{Reduction to $s>1$}{Reduction to s > 1}} \label{an:reduction} For a number field $A$, write $h_A$, $\operatorname{Reg}_A$, $w_A$ and $\Delta_A$ for its class number, regulator, number of roots of unity and discriminant. The residue of $\zeta_A$ at $s=1$ is $2^{r_1(A)}(2\pi)^{r_2(A)}h_A\operatorname{Reg}_A/(w_A|\Delta_A|^{1/2})$. Since $K/F$ is unramified at the finite places, $|\Delta_K|=|\Delta_F|^2$. Hence \begin{equation}\label{an:class-formula} L_F(1)=\lim_{s\to1}\frac{\zeta_K(s)}{\zeta_F(s)} =2^c\pi^{d-c}\lambda^{-d/2}\, \frac{h_K\operatorname{Reg}_Kw_F}{h_F\operatorname{Reg}_Fw_K}>0. \end{equation} Let $s>1$ be real. At a prime of $F$ of norm $q$, the Euler factor of $L_F$ is $(1-q^{-s})^{-1}$ if the prime splits in $K$ and $(1+q^{-s})^{-1}$ if it is inert. Its square is at most the Euler factor of $\zeta_K$, which is $(1-q^{-s})^{-2}$, respectively $(1-q^{-2s})^{-1}$. Consequently \begin{equation}\label{an:relative-euler} \frac{\log L_F(s)}d\le\frac{\log\zeta_K(s)}{2d} =\frac{\log\zeta_K(s)}{[K:\Q]}. \end{equation} The function $L_F$ is the Hecke $L$-function of the quadratic character of $K/F$. This character is unramified at every finite place and nontrivial at every real place of $F$, since $K$ is totally imaginary. With $\Gamma_\R$, $\Gamma_\C$ and $\Lambda_A$ as in Section~\ref{gf:section}, put \begin{equation}\label{an:completion} \mathcal L_F(s)=\frac{\Lambda_K(s)}{\Lambda_F(s)} =\lambda^{ds/2}\,\Gamma_\R(s+1)^b\,\Gamma_\C(s)^c\,L_F(s). \end{equation} By Hecke's theorem \cite{Hecke1920,Tate1967}, $\mathcal L_F$ is entire of order at most one, and it satisfies $\mathcal L_F(s)=\mathcal L_F(1-s)$ because both completed Dedekind zeta functions do. \begin{lemma}\label{an:monotonicity} The function $\mathcal L_F(s)$ is positive and nondecreasing on $[1,\infty)$. \end{lemma} \begin{proof} The Euler product shows that $\mathcal L_F$ has no zeros with $\operatorname{Re}s>1$, and the functional equation excludes $\operatorname{Re}s<0$. So every zero $\rho$ satisfies $0\le\operatorname{Re} \rho\le1$. There is no zero on $[1,\infty)$, by the Euler product and by \eqref{an:class-formula}. As $\mathcal L_F(s)>0$ for large real $s$, it is positive on $[1,\infty)$. Put $\Xi(z)=\mathcal L_F(\frac12+z)$. It is entire of order at most one, even, and real on the real axis. Its zeros $z=\rho-\frac12$ satisfy $|\operatorname{Re}z|\le\frac12$, and $\sum_{z\ne0}|z|^{-2}<\infty$. Group the nonzero zeros into pairs $\{z_0,-z_0\}$, and let $m\ge0$ be the order of $\Xi$ at $0$. By Hadamard's theorem for functions of order at most one, \[ \frac{\Xi'}{\Xi}(z)-\frac mz-\sum_{\{z_0,-z_0\}}\frac{2z}{z^2-z_0^2} \] is constant, where the series converges absolutely and locally uniformly away from the zeros. Since $\Xi$ is even, $\Xi'/\Xi$ is odd, so the constant is zero. For real $x\ge\frac12$ the logarithmic derivative is real, so it equals the sum of the real parts. If $z_0=a+iy$, then \[ \operatorname{Re}\frac{2x}{x^2-z_0^2} =\frac{2x(x^2-a^2+y^2)}{(x^2-a^2+y^2)^2+4a^2y^2}\ge0, \] because $|a|\le\frac12\le x$. Since also $m/x\ge0$, the logarithmic derivative is nonnegative, and $\log\Xi$ is nondecreasing on $[\frac12,\infty)$. \end{proof} For $0<\varepsilon\le1$ and $0\le\theta\le\frac12$ put \begin{equation}\label{an:gamma} \Upsilon(\varepsilon,\theta) =\frac\varepsilon2(\ell-\log\pi)-\varepsilon\theta\log2 +(1-2\theta)\log\Gamma\Bigl(1+\frac\varepsilon2\Bigr) +\theta\log\Gamma(1+\varepsilon). \end{equation} \begin{lemma}\label{an:shift} Let $K\in\mathcal K$ and $0<\varepsilon\le1$. Then \begin{equation}\label{an:finite-gamma} \frac{\log L_F(1)}d \le\frac{\log\zeta_K(1+\varepsilon)}{[K:\Q]}+\Upsilon(\varepsilon,\theta). \end{equation} Moreover $\Upsilon(\varepsilon,\theta)\to0$ as $\varepsilon\downarrow0$, uniformly for $0\le\theta\le\frac12$. \end{lemma} \begin{proof} By Lemma~\ref{an:monotonicity}, $\mathcal L_F(1)\le\mathcal L_F(1+\varepsilon)$. In \eqref{an:completion} the gamma factors change by \[ \frac{\Gamma_\R(2+\varepsilon)}{\Gamma_\R(2)} =\pi^{-\varepsilon/2}\,\Gamma\Bigl(1+\frac\varepsilon2\Bigr),\qquad \frac{\Gamma_\C(1+\varepsilon)}{\Gamma_\C(1)} =(2\pi)^{-\varepsilon}\,\Gamma(1+\varepsilon). \] Taking logarithms, dividing by $d$, and using $b/d=1-2\theta$, $c/d=\theta$ and \eqref{an:relative-euler}, we obtain \eqref{an:finite-gamma}. Since $\Upsilon$ is affine in $\theta$, $\Upsilon(\varepsilon,\theta)\to0$ as $\varepsilon\downarrow0$, uniformly for $0\le\theta\le\frac12$. \end{proof} The term $\Upsilon$ does not appear in the final estimate. With the value $\varepsilon=1/300$ used in the proof of Theorem~\ref{an:ceiling}, \eqref{an:finite-gamma} would add $\Upsilon(1/300,\theta)\approx0.0054$ for $\theta$ near $\frac12$. Instead, the proof of Proposition~\ref{an:prime-budget} applies \eqref{an:finite-gamma} with $s-1$ in place of $\varepsilon$ for each fixed $s\in(1,2]$, passes to the limit along a sequence of fields, and then lets $s\downarrow1$, so that $\Upsilon$ vanishes in the limit; the Tsfasman--Vl\u{a}du\c{t} inequality then bounds the change from $1+\varepsilon$ to $1$ by the term $\varepsilon(\kappa_\infty+\mathcal D)$ of that proposition, which is about $0.00215$ in the proof of Theorem~\ref{an:ceiling}. This limit is also the reason why $m_0$ is not effective. \subsection{\texorpdfstring{The Tsfasman--Vl\u{a}du\c{t} Inequality over $B$}{The Tsfasman-Vladut Inequality over B}} \label{an:budget-section} For $q>1$ and $s\ge1$ put $g_s(q)=-\log(1-q^{-s})$. \begin{lemma}\label{an:local-contribution} Let $M$ be a finite Galois extension of $B$ and $s>1$. For a prime $\mathfrak p$ of $B$ let $e_M(\mathfrak p)$ and $f_M(\mathfrak p)$ be its ramification index and residue degree in $M/B$. Then \begin{align} \frac{\log\zeta_M(s)}{[M:\Q]} &=\sum_{\mathfrak p}\frac{g_s(\mathrm N\mathfrak p^{f_M(\mathfrak p)})} {2e_M(\mathfrak p)f_M(\mathfrak p)},\label{an:local-sum}\\ -\frac1{[M:\Q]}\frac{\zeta_M'}{\zeta_M}(2) &=\frac1{[M:\Q]}\sum_{\mathfrak P}\frac{\log\mathrm N\mathfrak P} {\mathrm N\mathfrak P^2-1} =\sum_{\mathfrak p}\frac{\log\mathrm N\mathfrak p} {2e_M(\mathfrak p)(\mathrm N\mathfrak p^{2f_M(\mathfrak p)}-1)}, \label{an:local-debit} \end{align} where $\mathfrak P$ runs over the primes of $M$. Each summand on the right does not increase when $e_M(\mathfrak p)$ or $f_M(\mathfrak p)$ is replaced by a larger real number. If $M\subseteq M'$ are both Galois over $B$, then $e_M(\mathfrak p)\mid e_{M'}(\mathfrak p)$ and $f_M(\mathfrak p)\mid f_{M'}(\mathfrak p)$. \end{lemma} \begin{proof} Above $\mathfrak p$ there are $[M:B]/(ef)$ primes of $M$, each of norm $\mathrm N\mathfrak p^f$, and $[M:\Q]=2[M:B]$. This gives both formulas. The first equality in \eqref{an:local-debit} follows from $-\zeta_M'/\zeta_M(s)=\sum_{\mathfrak P}\log\mathrm N\mathfrak P/(\mathrm N\mathfrak P^s-1)$. For monotonicity in $f$, write $g_s(x^f)/f=\sum_{j\ge1}x^{-fjs}/(fj)$ for $x>1$; each term decreases in $f$. The divisibilities hold because ramification indices and residue degrees are multiplicative in towers. \end{proof} Put \[ \kappa_\infty=\frac{\ell-\gamma-\log(4\pi)}4,\qquad w(q)=2\log q\sum_{m\ge1}\frac1{q^m+1}. \] Here $2\kappa_\infty$ is the right side, and $w(q)$ the weight of the prime power $q$, in the Tsfasman--Vl\u{a}du\c{t} inequality \eqref{an:tv-budget} below. \begin{proposition}\label{an:prime-budget} Let $\sigma=1+\varepsilon$ with $0<\varepsilon\le1$, and let $Y_*$ be a real number with $[K:\Q]^{-1}\log\zeta_K(\sigma)\le Y_*$ for every $K\in\mathcal K$. For each prime $\mathfrak p$ of $B$, let $e^0_{\mathfrak p},f^0_{\mathfrak p}\ge1$ be integers with $e_K(\mathfrak p)\ge e^0_{\mathfrak p}$ and $f_K(\mathfrak p)\ge f^0_{\mathfrak p}$ for every $K\in\mathcal K$, that is, lower bounds for the type of $\mathfrak p$ in $K/B$, and let \[ \mathcal D\ge\sum_{\mathfrak p}\frac{\log\mathrm N\mathfrak p} {2e^0_{\mathfrak p}(\mathrm N\mathfrak p^{2f^0_{\mathfrak p}}-1)}. \] Then every $C'>Y_*+\varepsilon(\kappa_\infty+\mathcal D)$ has the following property: there is an $m_0$ such that $d^{-1}\log L_F(1)1$, $(2d_j)^{-1}\log\zeta_{K^{(j)}}(s)=\sum_q(N_q(K^{(j)})/2d_j)g_s(q)$ tends to $Z(s)$ by dominated convergence: for $q=p^f$ the terms are at most $g_s(q)/f\le2q^{-s}$, and $\sum_q q^{-s}<\infty$. By Lemma~\ref{an:shift} with $s-1$ in place of $\varepsilon$, for $10$ define the incomplete gamma and Bessel integrals \[ H_\mu(x)=\int_1^\infty u^{\mu-1}e^{-xu}\,du,\qquad I_\mu(y)=\int_1^\infty u^\mu K_0(yu)\,du, \] where $K_0$ is the modified Bessel function of the second kind of order zero, and their envelopes (upper bounds, by Lemma~\ref{an:kernels}(2)) \[ \mathcal V_1(x)=\frac{e^{-x}}x\Bigl(1+\frac1x\Bigr),\qquad \mathcal V_2(y)=\sqrt{\frac\pi2}\,\frac{e^{-y}}{y^{3/2}}\Bigl(1+\frac1y\Bigr). \] Numerical evaluation of $L$-functions from their functional equations is developed in \cite{Dokchitser2004}. Part (1) of the next lemma treats mixed gamma factors, and part (2), for pure gamma factors, uses the Bessel function $K_0$. We prove both. \begin{lemma}\label{an:balanced-afe} Let $a_n$ be real numbers with $|a_n|\le\tau(n)$, let $Q>0$ and $\nu_1,\nu_2\in\{0,1\}$ (integers, not places), and let $L(s)=\sum_{n\ge1}a_nn^{-s}$. Suppose that $\Lambda(s)=Q^{s/2}\Gamma_\R(s+\nu_1)\Gamma_\R(s+\nu_2)L(s)$ extends to an entire function of order at most one with $\Lambda(1-s)=\Lambda(s)$. Put $t=2\pi/\sqrt Q$ and let $\sigma>1$ be real. \begin{enumerate} \item If $\nu_1\ne\nu_2$, then \[ L(\sigma)=\frac{t^\sigma}{\Gamma(\sigma)}\sum_{n\ge1}a_n \bigl(H_\sigma(tn)+H_{1-\sigma}(tn)\bigr). \] \item If $\nu_1=\nu_2=\nu$, then \[ L(\sigma)=\frac{4(t/2)^{\sigma+\nu}}{\Gamma((\sigma+\nu)/2)^2} \sum_{n\ge1}a_nn^\nu \bigl(I_{\sigma+\nu-1}(tn)+I_{\nu-\sigma}(tn)\bigr). \] \end{enumerate} The series converge absolutely. \end{lemma} \begin{proof} In case (1) put $k(x)=e^{-x}$ and $\nu=0$, and in case (2) put $k(x)=K_0(x)$. Their Mellin transforms are $\Gamma(w)$ and $2^{w-2}\Gamma(w/2)^2$ for $\operatorname{Re}w>0$ \cite[10.43.19]{DLMF}. Let $\Theta(u)=\sum_na_nn^\nu k(tnu)$ for $u>0$. For $\operatorname{Re}w>1+\nu$, termwise integration gives \[ \widetilde\Theta(w)=\int_0^\infty\Theta(u)u^{w-1}\,du= \begin{cases} t^{-w}\Gamma(w)L(w)=\tfrac12\Lambda(w)&\text{in case (1)},\\ 2^{w-2}t^{-w}\Gamma(w/2)^2L(w-\nu)=\tfrac14Q^{\nu/2}\Lambda(w-\nu) &\text{in case (2)}. \end{cases} \] In case (1) we used $\Gamma_\R(w)\Gamma_\R(w+1)=\Gamma_\C(w)$ and $Q^{w/2}(2\pi)^{-w}=t^{-w}$. In both cases $\widetilde\Theta$ is entire and $\widetilde\Theta(w)=\widetilde\Theta(1+2\nu-w)$. The function $L=\Lambda/(Q^{s/2}\Gamma_\R(s+\nu_1)\Gamma_\R(s+\nu_2))$ of the complex variable $s$ is entire of finite order. Fix $\eta>0$. The function $L$ is bounded on $\operatorname{Re}s=1+\eta$, and, by the functional equation and Stirling's formula, of polynomial growth on $\operatorname{Re}s=-\eta$. By the Phragm\'en--Lindel\"of principle it has polynomial growth in $|\operatorname{Im}s|$ on every vertical strip. With Stirling's formula this shows that $\widetilde\Theta(w)$ decays exponentially in $|\operatorname{Im}w|$, uniformly on vertical strips. Mellin inversion on a line $\operatorname{Re}w=c>1+\nu$ and a shift to $\operatorname{Re}w=1+2\nu-c$ are therefore justified, and no residues occur. With the functional equation of $\widetilde\Theta$ they give \[ \Theta(u)=u^{-1-2\nu}\,\Theta(1/u)\qquad(u>0). \] In particular $\Theta(u)$ decays rapidly as $u\downarrow0$. Splitting the Mellin integral at $u=1$ and substituting $u\mapsto1/u$ below $1$ gives \[ \widetilde\Theta(\sigma+\nu)=\int_1^\infty\Theta(u) \bigl(u^{\sigma+\nu-1}+u^{\nu-\sigma}\bigr)\,du. \] Termwise integration, justified by the exponential decay of $k$, turns the right side into $\sum_na_nn^\nu$ times $H_\sigma(tn)+H_{1-\sigma}(tn)$ in case (1), and times $I_{\sigma+\nu-1}(tn)+I_{\nu-\sigma}(tn)$ in case (2). On the left, $\widetilde\Theta(\sigma)=t^{-\sigma}\Gamma(\sigma)L(\sigma)$ in case (1), and $\widetilde\Theta(\sigma+\nu)=2^{\sigma+\nu-2}t^{-\sigma-\nu} \Gamma((\sigma+\nu)/2)^2L(\sigma)$ in case (2). \end{proof} \begin{lemma}\label{an:kernels} Let $\mu\in\R$. \begin{enumerate} \item The functions $H_\mu$ and $I_\mu$ are Laplace transforms of positive measures on $[1,\infty)$. They are completely monotone on $(0,\infty)$, extend holomorphically to $\operatorname{Re}z>0$, and satisfy $|H(z)|\le H(\operatorname{Re}z)$ there, for $H=H_\mu$ or $H=I_\mu$. \item If $\mu\le2$, then $H_\mu\le\mathcal V_1$. If $\mu\le3/2$, then $I_\mu\le\mathcal V_2$. \item They satisfy \[ xH_\mu'(x)+\mu H_\mu(x)=-e^{-x},\qquad yI_\mu'(y)+(\mu+1)I_\mu(y)=-K_0(y), \] and $K_0$ satisfies $y^2K_0''+yK_0'-y^2K_0=0$ and $K_0'=-K_1$, where $K_1$ is the modified Bessel function of the second kind of order one. \end{enumerate} \end{lemma} \begin{proof} The definition exhibits $H_\mu$ as the Laplace transform of $u^{\mu-1}\,du$ on $[1,\infty)$. Taking the order $0$ in \cite[10.32.9]{DLMF} and substituting $v=\cosh r$ in its integral gives $K_0(x)=\int_1^\infty e^{-xv}(v^2-1)^{-1/2}\,dv$. Substituting this into $I_\mu$, and then $v=r/u$ in the inner integral, shows \[ I_\mu(y)=\int_1^\infty e^{-yr}\Bigl(\int_1^ru^\mu(r^2-u^2)^{-1/2}\,du\Bigr) \,dr, \] with a nonnegative inner integral. Differentiation under the integral gives complete monotonicity. Since the measures are positive and $|e^{-zr}|=e^{-r\operatorname{Re}z}$, we get $|H(z)|\le H(\operatorname{Re}z)$ for $\operatorname{Re}z>0$, which proves (1). For (2), $u^{\mu-1}\le u$ for $u\ge1$ and $\mu\le2$, and $\int_1^\infty ue^{-xu}\,du=\mathcal V_1(x)$. For $I_\mu$ we use $K_0(x)\le K_{1/2}(x)=\sqrt{\pi/(2x)}\,e^{-x}$ \cite[10.39.2]{DLMF}. It holds because the modified Bessel function of order $\eta\ge0$ is $\int_0^\infty e^{-x\cosh r}\cosh(\eta r)\,dr$ \cite[10.32.9]{DLMF}, which increases with $\eta$. Since $u^{\mu-1/2}\le u$ for $\mu\le3/2$, \[ I_\mu(y)\le\sqrt{\frac\pi{2y}}\int_1^\infty u^{\mu-1/2}e^{-yu}\,du \le\mathcal V_2(y). \] For (3), differentiate under the integral and integrate by parts: $xH_\mu'(x)=-\int_1^\infty u^\mu xe^{-xu}\,du=-e^{-x}-\mu H_\mu(x)$, and $yI_\mu'(y)=\int_1^\infty u^{\mu+1}\frac{d}{du}K_0(yu)\,du =-K_0(y)-(\mu+1)I_\mu(y)$. The last two identities are Bessel's equation and a standard derivative \cite[10.25.1, 10.29.3]{DLMF}. \end{proof} In the application $\sigma=301/300$, so $1<\sigma\le3/2$. The four exponents $\sigma$, $1-\sigma$, $\sigma+\nu-1$ and $\nu-\sigma$ in Lemma~\ref{an:balanced-afe} are then covered by the envelopes: $\mathcal V_1$ bounds $H_\sigma$ and $H_{1-\sigma}$, and $\mathcal V_2$ bounds $I_{\sigma+\nu-1}$ and $I_{\nu-\sigma}$. \begin{lemma}\label{an:tails} Let $1<\sigma\le3/2$, let $|a_n|\le\tau(n)$, let $t>0$, $N\ge1$ and $\omega>1$. \begin{enumerate} \item If $t(N+1)\ge\omega-1$, then for $\mu\in\{\sigma,1-\sigma\}$ \[ \sum_{n>N}|a_n|H_\mu(tn)\le\zeta(\omega)^2(N+1)^\omega \mathcal V_1\bigl(t(N+1)\bigr). \] \item If $\nu\in\{0,1\}$ and $t(N+1)\ge\omega+\nu-\frac32$, then for $\mu\in\{\sigma+\nu-1,\nu-\sigma\}$ \[ \sum_{n>N}|a_n|n^\nu I_\mu(tn)\le\zeta(\omega)^2(N+1)^{\omega+\nu} \mathcal V_2\bigl(t(N+1)\bigr). \] \end{enumerate} \end{lemma} \begin{proof} The logarithmic derivative of $x^\omega\mathcal V_1(x)=x^{\omega-1}e^{-x}(1+1/x)$ is $(\omega-1)/x-1-1/(x(x+1))<0$ for $x\ge\omega-1$. So $n^\omega\mathcal V_1(tn)\le(N+1)^\omega\mathcal V_1(t(N+1))$ for $n>N$. By Lemma~\ref{an:kernels}(2), \[ \sum_{n>N}|a_n|H_\mu(tn)\le\sum_{n>N}\frac{\tau(n)}{n^\omega}\, n^\omega\mathcal V_1(tn)\le(N+1)^\omega\mathcal V_1\bigl(t(N+1)\bigr) \sum_{n\ge1}\frac{\tau(n)}{n^\omega}, \] and the last sum is $\zeta(\omega)^2$. Part (2) is the same, since $y^{\omega+\nu}\mathcal V_2(y)$ is a multiple of $y^{\omega+\nu-3/2}e^{-y}(1+1/y)$, which decreases for $y\ge\omega+\nu-\frac32$. \end{proof} \subsection{\texorpdfstring{Enclosures of $H_\mu$ and $I_\mu$, and Finite Sums}{Enclosures of H and I, and Finite Sums}} \label{an:kernel-enclosures} An \emph{enclosure} of a real number is an interval that contains it. We enclose $H_\mu$ and $I_\mu$ on a geometric grid. Let $H$ be $H_\mu$ or $I_\mu$ with $\mu$ as in Lemma~\ref{an:tails}, and let $\mathcal V$ be its envelope, $\mathcal V_1$ or $\mathcal V_2$. For $J\ge0$ and $\xi>0$ put \[ \rho_J(\xi)=\mathcal V(\xi/2)\,\frac{16^{-J-1}}{1-1/16}. \] \begin{lemma}\label{an:taylor} Let $\xi>0$ and let $h_j=H^{(j)}(\xi)/j!$ be the Taylor coefficients of $H$ at $\xi$. \begin{enumerate} \item $(-1)^jh_j\ge0$ and $|h_j|\le\mathcal V(\xi/2)(2/\xi)^j$ for all $j\ge0$. \item For $|x-\xi|\le\xi/32$, $\bigl|H(x)-\sum_{j=0}^Jh_j(x-\xi)^j\bigr|\le\rho_J(\xi)$. For $x=31\xi/32$ the difference lies in $[0,\rho_J(\xi)]$. \item Let $k_j$ be the Taylor coefficients at $\xi$ of $e^{-x}$ if $H=H_\mu$, and of $K_0$ if $H=I_\mu$. If $H=H_\mu$, then $(j+1)\xi h_{j+1}=-k_j-(j+\mu)h_j$, and $k_j=(-1)^je^{-\xi}/j!$. If $H=I_\mu$, then $(j+1)\xi h_{j+1}=-k_j-(j+\mu+1)h_j$, where $k_0=K_0(\xi)$, $k_1=-K_1(\xi)$, and \[ (j+1)(j+2)\xi^2k_{j+2}=-(j+1)(2j+1)\xi k_{j+1}-(j^2-\xi^2)k_j +2\xi k_{j-1}+k_{j-2}, \] with $k_{-1}=k_{-2}=0$. \end{enumerate} \end{lemma} \begin{proof} Complete monotonicity gives the signs in (1). The disc $|z-\xi|\le\xi/2$ lies in $\operatorname{Re}z\ge\xi/2$, where $|H(z)|\le H(\xi/2)\le\mathcal V(\xi/2)$ by Lemma~\ref{an:kernels}. Cauchy's estimate gives the bound on $h_j$. For $|x-\xi|\le\xi/32$, the omitted terms have modulus at most $\mathcal V(\xi/2)16^{-j}$, and their sum over $j>J$ is at most $\rho_J(\xi)$. For $x<\xi$ every term $h_j(x-\xi)^j$ is nonnegative by (1), so the remainder is nonnegative. Part (3) follows by comparing coefficients of $(x-\xi)^j$ in the differential equations of Lemma~\ref{an:kernels}(3), written at $x=\xi+(x-\xi)$. \end{proof} The grid points are $\xi_i=64\,(31/32)^i$ for $0\le i\le i_1$, where $i_1=1878$ is the least index with $\xi_{i_1}\le2^{-80}$. Taylor polynomials are formed at $\xi_0,\ldots,\xi_{i_1-1}$. The initial values $e^{-\xi_i}$, $K_0(\xi_i)$ and $K_1(\xi_i)$ of part~(3) of Lemma~\ref{an:taylor} are enclosed with the rigorous exponential and Bessel functions of Arb \cite{Johansson2017}, which is now part of FLINT \cite{FLINT} and is called through its Python interface. \begin{lemma}\label{an:grid} Let $J\ge0$, and suppose that enclosures of $e^{-\xi_i}$, $K_0(\xi_i)$ and $K_1(\xi_i)$ are given for $0\le i0$, $\nu\in\{0,1\}$, integers $a_1,\ldots,a_N$ and $J\ge0$ be given, with $tN<\xi_0$ and $t>\xi_{i_1}$. For $0\le i0.0033$, because $Q_e\le3470400$ for every $e$ (Corollary~\ref{gf:conductors}). So every argument $tn$ exceeds $0.0033$, and the part of the grid below that point is not used. \subsection{\texorpdfstring{The Value of $\zeta_E$ at $\sigma=301/300$}{The Value of zeta(E) at sigma = 301/300}} \label{an:value-section} From now on $\sigma=301/300$. Let $e\in\Ftwo^8\setminus\{0\}$. By Proposition~\ref{gf:degree-two}, $L(s,\chi_e)$ satisfies the hypotheses of Lemma~\ref{an:balanced-afe} with $Q=Q_e$ and $\nu_k=\nu_k(e)$. Put $t=2\pi/\sqrt{Q_e}$ and \[ N_e=4\lfloor\sqrt{Q_e}\rfloor+1, \] and let $\Pi_e$ be the prefactor of that lemma: $\Pi_e=t^\sigma/\Gamma(\sigma)$ if $\nu_1(e)\ne\nu_2(e)$, and $\Pi_e=4(t/2)^{\sigma+\nu}/\Gamma((\sigma+\nu)/2)^2$ if $\nu_1(e)=\nu_2(e)=\nu$; in the first case put $\nu=0$. Let $\mathcal S_1$ and $\mathcal S_2$ be the sums $\sum_i\sum_{j=0}^{20}P_{i,j}M_{i,j}$ of Lemma~\ref{an:binning}, with $N=N_e$, $a_n=a_n(e)$ and $J=20$, for the two functions of the series of Lemma~\ref{an:balanced-afe}: $H=H_\sigma$ and $H=H_{1-\sigma}$ in the first case, and $H=I_{\sigma+\nu-1}$ and $H=I_{\nu-\sigma}$ in the second; their envelope is $\mathcal V=\mathcal V_1$ in the first case and $\mathcal V=\mathcal V_2$ in the second. For every $e$, $N_e$ lies between $105$ and $7449$, and $t(N_e+1)$ between $24.7$ and $25.5$ (the supplementary program \texttt{check255.gp} prints both ranges). Hence $tN_e<\xi_0$ and $t>\xi_{i_1}$, as Lemma~\ref{an:binning} requires, and the hypotheses of Lemma~\ref{an:tails} hold for every $\omega$ in the list $\frac{101}{100},\frac{21}{20},\frac{11}{10},\frac98,\frac65,\frac54, \frac43,\frac32,2$. The value of $\omega$ in this list is chosen to minimize the bound \[ \mathcal T_e=\sum_i\rho_{20}(\xi_i)A_i +\zeta(\omega)^2(N_e+1)^{\omega+\nu}\mathcal V\bigl(t(N_e+1)\bigr). \] \begin{lemma}\label{an:row} For $\sigma=301/300$ and every $e\in\Ftwo^8\setminus\{0\}$, \begin{equation}\label{an:row-enclosure} \bigl|L(\sigma,\chi_e)-\Pi_e(\mathcal S_1+\mathcal S_2)\bigr| \le2\Pi_e\mathcal T_e. \end{equation} \end{lemma} \begin{proof} Split each of the two series of Lemma~\ref{an:balanced-afe} at $N_e$. By Lemma~\ref{an:binning}, each finite part differs from its approximation $\mathcal S_1$ or $\mathcal S_2$ by at most $\sum_i\rho_{20}(\xi_i)A_i$; the same bound serves both series, since their two functions have the same envelope. By Lemma~\ref{an:tails}, each tail is at most $\zeta(\omega)^2(N_e+1)^{\omega+\nu}\mathcal V(t(N_e+1))$. Multiplying by $\Pi_e$ gives \eqref{an:row-enclosure}. \end{proof} \begin{proposition}\label{an:genus-value} For $\sigma=301/300$, \[ 0.08264460177456<\frac{\log\zeta_E(\sigma)}{512}<0.08264460807138. \] \end{proposition} \begin{proof}[Proof (computer-assisted)] By Proposition~\ref{gf:factorization}, \[ \log\zeta_E(\sigma)=\log\zeta(\sigma)+\log L(\sigma,\chi_B) +\sum_{e\ne0}\log L(\sigma,\chi_e). \] The coefficients $a_n(e)$ for $n\le N_e$ are computed exactly from the Euler factors of Proposition~\ref{gf:local-data} by the supplementary program \texttt{lfun241.py}. The supplementary program \texttt{afe241.py} evaluates \eqref{an:row-enclosure} in Arb midpoint--radius interval arithmetic \cite{Johansson2017,FLINT} with $256$ bits of precision, rounding every quantity outward. Its routines for Lemmas~\ref{an:balanced-afe}--\ref{an:binning} are taken unchanged from an earlier supplementary archive of the author (Subsection~\ref{intro:code}), and they implement these lemmas as stated here. Every enclosure of $L(\sigma,\chi_e)$ is positive. The factor $\zeta_B(\sigma)=\zeta(\sigma)L(\sigma,\chi_B)$ is evaluated with \[ L(\sigma,\chi_B)=241^{-\sigma}\sum_{a=1}^{240}\Bigl(\frac a{241}\Bigr)\zeta\Bigl(\sigma,\frac a{241}\Bigr), \] where $\zeta(s,x)=\sum_{n\ge0}(n+x)^{-s}$ is the Hurwitz zeta function, using Arb's rigorous implementation of it \cite{Johansson2015Hurwitz}. This gives $\zeta_B(\sigma)=725.51409164864\ldots$ and $L(\sigma,\chi_B)=2.41373420241\ldots$. Summing the logarithms gives the enclosure \[ [0.08264460177456168\ldots,\ 0.08264460807137072\ldots] \] of $\log\zeta_E(\sigma)/512$, which the proposition states rounded outward. So the inequalities follow from Lemmas~\ref{an:balanced-afe}--\ref{an:row}, the exact coefficients, and outward-rounded interval arithmetic. \end{proof} As a check independent of the approximate functional equation, the supplementary programs \texttt{yecheck.gp} and \texttt{lvalues255.gp} evaluate the same values numerically with PARI/GP \cite{PARI}: each of the $255$ values $L(\sigma,\chi_e)$ lies in its enclosure, and PARI's value $0.08264460492293\ldots$ of $\log\zeta_E(\sigma)/512$ lies in the enclosure above. \subsection{Frobenius Vectors and Lower Bounds for the Types} \label{an:census} \begin{lemma}\label{an:sigma2} The set $\Sigma_2$ of Definition~\ref{an:P4} has exactly $21$ elements: zero and the following twenty vectors, written as strings $c_0c_1\cdots c_7$ of their coordinates, as in Section~\ref{tw:tower}: \[ \begin{array}{ccccc} 00100000&11110010&11010010&00010000&10000001\\ 10010001&00001000&10100111&10101111&00000100\\ 11101011&11101111&00000010&01100101&01100111\\ 00000001&01011010&01011011&10111010&11000101. \end{array} \] In the notation of Table~\ref{tw:vector-table}, these twenty vectors are the elementary images of the local involutions $x_j$, $y_j$ and $x_jy_j$ for $j=1,2$, of the nonidentity elements of the four local groups $C_2\times C_2$ above $3$ and $5$, and of $\iota_1$ and $\iota_2$. \end{lemma} \begin{proof} Each of these images lies in $\Sigma_2$ by Lemma~\ref{tw:quadratic-layer}, since an involution has trivial square. The supplementary program \texttt{census241.py} computes $\Sigma_2$ by linear algebra over $\Ftwo$ from the relations of Lemma~\ref{tw:quadratic-layer}, and finds no other vectors. \end{proof} \begin{proposition}[Frobenius vectors of the primes of norm at most $10^6$]\label{an:census-prop} Among the $78616$ primes $\mathfrak p\notin S\cup\{\mathfrak t_0,\mathfrak t_1,\mathfrak t_2\}$ of $B$ with $\mathrm N\mathfrak p\le10^6$, exactly $246$ have $\overline{\Frob}_{\mathfrak p}=0$, exactly $5994$ have $\overline{\Frob}_{\mathfrak p}\in\Sigma_2\setminus\{0\}$, and the remaining $72376$ form $\mathcal P_4$. \end{proposition} \begin{proof} The supplementary program \texttt{census241.py} computes the Frobenius vector of each of these primes from Proposition~\ref{gf:local-data}(1) with exact modular arithmetic, and compares it with the elements of $\Sigma_2$ (Lemma~\ref{an:sigma2}). \end{proof} Below we use this classification of the primes, not the three counts. \begin{lemma}\label{an:corrections} For $\sigma=301/300$, \begin{equation}\label{an:census-values} \Delta_{2,7,29}(\sigma)>0.0344534299,\qquad \Delta_{\mathcal P_4}(\sigma)>0.0016298411. \end{equation} \end{lemma} \begin{proof} The supplementary program \texttt{ceiling241.py} uses the classification of Proposition~\ref{an:census-prop} and evaluates \eqref{an:sel} and \eqref{an:cen} in Arb interval arithmetic. It gives $\Delta_{2,7,29}(\sigma)=0.03445342991613\ldots$ and $\Delta_{\mathcal P_4}(\sigma)=0.00162984111205\ldots$, and the lemma states these values rounded down. \end{proof} For every prime $\mathfrak p$ of $B$ put \begin{equation}\label{an:floors} (e^0_{\mathfrak p},f^0_{\mathfrak p})= \begin{cases} (8,4)&\mathfrak p=\mathfrak p_1,\mathfrak p_2,\\ (2,2)&\mathfrak p=\mathfrak q_1,\mathfrak q_2,\mathfrak r_1,\mathfrak r_2,\\ (1,4)&\mathfrak p\in\{\mathfrak t_0,\mathfrak t_1,\mathfrak t_2\}\cup\mathcal P_4,\\ (1,2)&\mathrm N\mathfrak p\le10^6,\ \overline{\Frob}_{\mathfrak p}\in\Sigma_2\setminus\{0\},\\ (1,1)&\text{otherwise}, \end{cases} \end{equation} where $\mathfrak q_1,\mathfrak q_2$ and $\mathfrak r_1,\mathfrak r_2$ are the primes above $3$ and $5$ (Lemma~\ref{tw:base}). \begin{lemma}\label{an:floors-valid} For every $K\in\mathcal K$ and every prime $\mathfrak p$ of $B$, $e_K(\mathfrak p)\ge e^0_{\mathfrak p}$ and $f_K(\mathfrak p)\ge f^0_{\mathfrak p}$. Thus \eqref{an:floors} gives lower bounds for the types in $K/B$, as Proposition~\ref{an:prime-budget} requires. \end{lemma} \begin{proof} At $\mathfrak p_j$ and $\mathfrak t_k$ the types in $K/B$ are $(8,4)$ and $(1,4)$ by Theorem~\ref{tw:field-family}, as in the proof of Proposition~\ref{an:Ystar}. At $\mathfrak q_j$ and $\mathfrak r_j$ the type in $E/B$ is $(2,2)$ by Corollary~\ref{gf:types}, and if $\overline{\Frob}_{\mathfrak p}\ne0$, then $f_E(\mathfrak p)=2$ by the same corollary. Since $E\subseteq K$, the ramification indices and residue degrees in $E/B$ divide those in $K/B$ (Lemma~\ref{an:local-contribution}). On $\mathcal P_4$ use Lemma~\ref{an:census-degree}. The bounds $(1,1)$ hold trivially. \end{proof} \begin{lemma}\label{an:D} For the lower bounds \eqref{an:floors}, \begin{equation}\label{an:R} \sum_{\mathfrak p}\frac{\log\mathrm N\mathfrak p} {2e^0_{\mathfrak p}(\mathrm N\mathfrak p^{2f^0_{\mathfrak p}}-1)} <0.0085105513. \end{equation} Hence $\mathcal D=0.0085105513$ satisfies the hypothesis of Proposition~\ref{an:prime-budget}. \end{lemma} \begin{proof} The primes of norm greater than $10^6$ receive $(1,1)$. At most two primes of $B$ have a given norm, and $x\mapsto\log x/(x^2-1)$ decreases on $[2,\infty)$. So their contribution to the sum is at most \[ \sum_{m>10^6}\frac{\log m}{m^2-1}\le\int_{10^6}^\infty\frac{\log x}{x^2-1}\,dx \le\frac{\log10^6+1}{10^6}\,(1+10^{-6}). \] The supplementary program \texttt{ceiling241.py} sums the primes of norm at most $10^6$ in interval arithmetic and adds this bound for the remaining primes; the result, $0.0085105512261\ldots$, is less than $0.0085105513$. \end{proof} \subsection{\texorpdfstring{Proof of Theorem~\ref{an:ceiling}}{Proof of the Upper Bound}} \begin{corollary}\label{an:Ystar-value} Let $\sigma=301/300$. Every $K\in\mathcal K$ satisfies $[K:\Q]^{-1}\log\zeta_K(\sigma)\le Y_*$, where \[ Y_*=0.0826446081-0.0344534299-0.0016298411=0.0465613371. \] \end{corollary} \begin{proof} Combine Proposition~\ref{an:Ystar}, the upper bound of Proposition~\ref{an:genus-value} rounded up, and the lower bounds of Lemma~\ref{an:corrections}, which are rounded down. \end{proof} \begin{proof}[Proof of Theorem~\ref{an:ceiling}] Let $\sigma=301/300$ and $\varepsilon=1/300$. We apply Proposition~\ref{an:prime-budget} with $Y_*=0.0465613371$ (Corollary~\ref{an:Ystar-value}), with the lower bounds \eqref{an:floors} for the types (Lemma~\ref{an:floors-valid}), and with $\mathcal D=0.0085105513$ (Lemma~\ref{an:D}). Numerically $\kappa_\infty=(\ell-\gamma-\log(4\pi))/4=0.63694120292\ldots<0.6369412030$, so \[ Y_*+\varepsilon(\kappa_\infty+\mathcal D) <0.0465613371+\frac{0.6369412030+0.0085105513}{300} <0.04871284300$ for $x\in K^\times$, so $-1$ is not a norm from $K^\times$. If $\zeta\in K$ is a root of unity, then $N_{K/F}\zeta$ is a root of unity of $F$, hence $\pm1$ because $F$ has a real place, and it is positive at $v_0$. Thus $N_{K/F}\zeta=1$. \emph{Logarithmic embeddings.} For $x\in\mathcal O_K^\times$ and $y\in\mathcal O_F^\times$ let \[ \operatorname{Log}_K(x)=\bigl((2\log|\phi_vx|)_v,\,(2\log|\phi_j^+x|, 2\log|\phi_j^-x|)_j\bigr),\qquad \operatorname{Log}_F(y)=\bigl((\log|v(y)|)_v,\,(2\log|\tau_j(y)|)_j\bigr). \] Their images are lattices in the hyperplanes of coordinate sum zero, the kernel of $\operatorname{Log}_F$ is $\{\pm1\}$ \cite[Proposition~5.8]{MilneANT}, and $\operatorname{Reg}_K$ and $\operatorname{Reg}_F$ are the covolumes of the images after one coordinate is deleted; the choice of the deleted coordinate does not matter. \emph{The lattice $l(\mathcal O_K^1)$.} Let $\beta\in\mathcal O_K^1$. Then $|\phi_v(\beta)|=1$ and $|\phi_j^\pm(\beta)|=e^{\pm l_j(\beta)}$, so \[ \operatorname{Log}_K(\beta)=\bigl(0,\ (2l_j(\beta),-2l_j(\beta))_j\bigr). \] Hence $l(\beta)=0$ exactly when every conjugate of $\beta$ has absolute value one, that is, when $\beta$ is a root of unity \cite[Corollary~5.6]{MilneANT}. By the first step the kernel of $l$ on $\mathcal O_K^1$ has order $w_K$. The norm of $y\in\mathcal O_F^\times$ is $y^2$, so $N_{K/F}(\mathcal O_K^\times)$ has finite index in $\mathcal O_F^\times$ and rank $b+c-1$. By the unit theorem \cite[Theorem~5.1]{MilneANT}, $\mathcal O_K^\times$ has rank $d-1$, so $\mathcal O_K^1$ has rank $c$. As $\operatorname{Log}_K(\mathcal O_K^1)$ is discrete, so is $l(\mathcal O_K^1)$, and it is a full lattice in $\R^c$. \emph{The capitulation kernel.} The Herbrand quotient $h(M)=|\widehat H^0(\langle\iota\rangle,M)|/|H^1(\langle\iota\rangle,M)|$ \cite[Chapter~II, \S3]{MilneCFT} of the units satisfies $2\,h(\mathcal O_K^\times)=\prod_v[K_w:F_v]$, where $v$ runs over the archimedean places of $F$ and $w$ lies above $v$ \cite[Chapter~VII, Proposition~3.1]{MilneCFT}. Each real place contributes $2$ and each complex place $1$, so $h(\mathcal O_K^\times)=2^{b-1}$. Since $\widehat H^0(\langle\iota\rangle,\mathcal O_K^\times)=\mathcal O_F^\times/N_{K/F}(\mathcal O_K^\times)$ has order $I_N$, we get $|H^1(\langle\iota\rangle,\mathcal O_K^\times)|=I_N/2^{b-1}$. This group is $\mathcal O_K^1$ modulo the units $\gamma/\iota\gamma$ with $\gamma\in\mathcal O_K^\times$. If $\mathfrak a$ is a fractional ideal of $F$ with $\mathfrak a\mathcal O_K=x\mathcal O_K$, the class of the unit $x/\iota x$ depends neither on the choice of $x$ nor on the choice of $\mathfrak a$ in its ideal class. This defines a homomorphism from $\ker(\Cl(F)\to\Cl(K))$ to $H^1(\langle\iota\rangle,\mathcal O_K^\times)$. It is injective: if $x/\iota x=\gamma/\iota\gamma$, then $y=x/\gamma$ lies in $F$ and $\mathfrak a\mathcal O_K=y\mathcal O_K$, so $\mathfrak a=y\mathcal O_F$ by unique factorization. It is surjective: by Hilbert's Theorem~90 \cite[Chapter~II, Corollary~1.23]{MilneCFT}, an element of $\mathcal O_K^1$ has the form $x/\iota x$ with $x\in K^\times$, and then the fractional ideal $x\mathcal O_K=(\iota x)\mathcal O_K$ is invariant under $\iota$. By (G2), every $\iota$-invariant fractional ideal of $K$ is extended from $F$: at a split prime $\mathfrak P\,\iota\mathfrak P$ invariance forces equal exponents, an inert prime of $F$ stays prime, and no prime ramifies. Hence $\kappa=I_N/2^{b-1}$. \emph{Regulators.} Delete the coordinate of $v_0$ from both logarithmic embeddings, and let $\mathcal U_K\subset\R^{d-1}$ and $\mathcal U_F\subset\R^{b+c-1}$ be the resulting lattices, of covolumes $\operatorname{Reg}_K$ and $\operatorname{Reg}_F$. By \eqref{geo:norm-modulus}, $\operatorname{Log}_F(N_{K/F}x)$ is obtained from $\operatorname{Log}_K(x)$ by keeping the coordinates of the real places and adding the two coordinates of each pair. This induces a linear map $T\colon\R^{d-1}\to\R^{b+c-1}$ such that $T(\mathcal U_K)$ is the image of $N_{K/F}(\mathcal O_K^\times)$ in $\mathcal U_F$. The substitution $(x_j,y_j)\mapsto(x_j,x_j+y_j)$ on each pair has determinant one and turns $T$ into a coordinate projection whose kernel consists of the $c$ coordinates $x_j$. The covolume of a lattice in a product of two coordinate spaces is the covolume of its intersection with the first factor times the covolume of its projection to the second, whenever both are lattices. A unit $x\in\mathcal O_K^\times$ lies in the kernel of $T$ exactly when $\operatorname{Log}_F(N_{K/F}x)=0$, because the deleted coordinate is minus the sum of the others. Then $N_{K/F}x=\pm1$, and $N_{K/F}x=1$ by the first step. So the intersection of $\mathcal U_K$ with the kernel is the image of $\mathcal O_K^1$, which after the substitution is $2l(\mathcal O_K^1)$, of covolume $2^c\operatorname{Reg}^1$. Since $-1\notin N_{K/F}(\mathcal O_K^\times)$, the projection $T(\mathcal U_K)$ has index $[\mathcal O_F^\times:\{\pm1\}N_{K/F}(\mathcal O_K^\times)]=I_N/2$ in $\mathcal U_F$. Hence $\operatorname{Reg}_K=2^c\operatorname{Reg}^1\cdot\operatorname{Reg}_FI_N/2$. \emph{The class-number formula.} The analytic class-number formula \cite[Chapter~V, Theorem~2.4]{MilneCFT} (compare \eqref{an:class-formula}), applied to $K$ (with $r_1=0$ and $r_2=d$) and to $F$ (with $r_1=b$, $r_2=c$ and $w_F=2$), gives \[ L_F(1)=\frac{(2\pi)^dh_K\operatorname{Reg}_K}{w_K\lambda^d}\cdot \frac{2\lambda^{d/2}}{2^b(2\pi)^ch_F\operatorname{Reg}_F} =\frac{2^{c+1}\pi^{b+c}h_{\rm rel}}{w_K\lambda^{d/2}}\, \frac{\operatorname{Reg}_K}{\operatorname{Reg}_F}. \] Substituting $\operatorname{Reg}_K/\operatorname{Reg}_F=2^{c-1}I_N\operatorname{Reg}^1 =2^{b+c-2}\kappa\operatorname{Reg}^1$ gives \eqref{geo:mass-density}. \end{proof} \subsection{Norm-One Elements from Split Primes} \begin{definition}\label{geo:uniform-types} Let $\mathcal R$ be a finite set of rational primes. The extension $K/F$ has \emph{uniform local types} $(e_r,f_r)_{r\in\mathcal R}$ if, for every $r\in\mathcal R$, every prime of $F$ above $r$ splits in $K$ and has absolute type $(e_r,f_r)$ (Definition~\ref{tw:type}). \end{definition} If $K/F$ has uniform local types $(e_r,f_r)_{r\in\mathcal R}$, then $F$ has exactly $d/(e_rf_r)$ primes above $r$, each of norm $r^{f_r}$, since the products $e_{\mathfrak p}f_{\mathfrak p}$ of the absolute ramification indices and residue degrees of the primes $\mathfrak p$ of $F$ above $r$ add up to $[F:\Q]=d$. Given integers $k_r\ge0$, define \begin{equation}\label{geo:HJ} H=\sum_{r\in\mathcal R}\frac{k_r\log r}{e_r},\qquad J=\sum_{r\in\mathcal R}\frac{\log(k_r+1)}{e_rf_r}. \end{equation} \begin{lemma}\label{geo:selection} Assume that $K/F$ has uniform local types $(e_r,f_r)_{r\in\mathcal R}$, and fix integers $k_r\ge0$. There are a fractional ideal $I$ of $K$ with $N(I)=e^{-dH}$, an integer $t\ge e^{dJ}/(\kappa h_{\rm rel})$, and elements $\beta_1,\dots,\beta_t\in I$ of relative norm one whose cosets $\beta_i\mathcal O_K^1$ are pairwise disjoint. \end{lemma} \begin{proof} For each prime $\mathfrak p$ of $F$ above a prime $r\in\mathcal R$, write $\mathfrak p\mathcal O_K=\mathfrak P_{\mathfrak p}\,\iota\mathfrak P_{\mathfrak p}$ and $k_{\mathfrak p}=k_r$. For integer vectors $\boldsymbol j=(j_{\mathfrak p})$ with $0\le j_{\mathfrak p}\le k_{\mathfrak p}$ put \[ \mathfrak A_{\boldsymbol j}=\prod_{\mathfrak p} \mathfrak P_{\mathfrak p}^{j_{\mathfrak p}},\qquad \mathfrak J_{\boldsymbol j}=\prod_{\mathfrak p} \mathfrak P_{\mathfrak p}^{j_{\mathfrak p}} (\iota\mathfrak P_{\mathfrak p})^{k_{\mathfrak p}-j_{\mathfrak p}}. \] There are $\prod_{\mathfrak p}(k_{\mathfrak p}+1)=e^{dJ}$ such vectors. The quotient of $\Cl(K)$ by the image of $\Cl(F)$ has order $h_K\kappa/h_F=\kappa h_{\rm rel}$, so one class of this quotient contains the classes of $\mathfrak A_{\boldsymbol j}$ for a set of $t\ge e^{dJ}/(\kappa h_{\rm rel})$ vectors $\boldsymbol j$. Fix one of them, $\boldsymbol j_0$, and put $I=\mathfrak J_{\boldsymbol j_0}^{-1}$. For each $\boldsymbol j$ in the set, write $\mathfrak A_{\boldsymbol j}\mathfrak A_{\boldsymbol j_0}^{-1} =x\,\mathfrak a\mathcal O_K$ with $x\in K^\times$ and $\mathfrak a$ a fractional ideal of $F$, and put $\beta_{\boldsymbol j}=x/\iota x$. Then $N_{K/F}\beta_{\boldsymbol j}=1$, and since $\iota$ fixes $\mathfrak a\mathcal O_K$, \[ \beta_{\boldsymbol j}\mathcal O_K =\mathfrak A_{\boldsymbol j}\mathfrak A_{\boldsymbol j_0}^{-1} \bigl(\iota\mathfrak A_{\boldsymbol j}\bigr)^{-1} \iota\mathfrak A_{\boldsymbol j_0} =\mathfrak J_{\boldsymbol j}I. \] As $\mathfrak J_{\boldsymbol j}$ is integral, $\beta_{\boldsymbol j}\in I$. The ideals $\mathfrak J_{\boldsymbol j}I$ are distinct for distinct $\boldsymbol j$, and every element of $\beta_{\boldsymbol j}\mathcal O_K^1$ generates $\mathfrak J_{\boldsymbol j}I$, so the cosets are disjoint. Finally, since each $\mathfrak p$ splits in $K$, $N(I)^{-1}=\prod_{\mathfrak p}(N\mathfrak p)^{k_{\mathfrak p}} =\prod_r r^{f_rk_rd/(e_rf_r)}=e^{dH}$. \end{proof} \subsection{Unit Averaging and Translation} In this subsection and the next, $I$ is a fractional ideal of $K$ and $\beta_1,\dots,\beta_t\in I$ are elements of relative norm one whose cosets $\beta_i\mathcal O_K^1$ are pairwise disjoint; Lemma~\ref{geo:selection} provides such data. For $h\in\R^c$ let $W_h\colon\C^d\to\C^d$ multiply the coordinate $\phi_j^+$ by $e^{-h_j}$ and $\phi_j^-$ by $e^{h_j}$, and fix the compact coordinates. It is a real diagonal map of determinant one. Put $\Lambda_h=W_hI_\infty$. For a lattice $\Lambda\subset\C^d$, $\covol(\Lambda)$ denotes the volume of a fundamental domain. By \cite[Proposition~4.26]{MilneANT}, extended to fractional ideals by scaling and \cite[Proposition~4.2]{MilneANT}, \begin{equation}\label{geo:covolume} \covol(\mathfrak a_\infty)=2^{-d}|\Delta_K|^{1/2}N\mathfrak a \end{equation} for every fractional ideal $\mathfrak a$ of $K$. Hence $\covol(\Lambda_h)=2^{-d}\lambda^dN(I)$, independently of $h$; for the ideal of Lemma~\ref{geo:selection}, \begin{equation}\label{geo:determinant} \covol(\Lambda_h)=2^{-d}\lambda^de^{-dH}. \end{equation} If $N_{K/F}\beta=1$, then $W_h\beta_\infty$ has compact coordinates of modulus one and pair coordinates of moduli $e^{\pm(l_j(\beta)-h_j)}$. Let $\Pi^1$ be a fundamental parallelepiped of the lattice $l(\mathcal O_K^1)$; its volume is $\operatorname{Reg}^1$. When $c=0$, integrals over $\R^0$ and over $\Pi^1$ are evaluations at the single point. \begin{lemma}\label{geo:unfold-lemma} For every Borel function $\Phi\colon\R^c\to[0,\infty]$, \begin{equation}\label{geo:unfold} \frac1{\operatorname{Reg}^1}\int_{\Pi^1}\sum_{i=1}^t \sum_{\beta\in\beta_i\mathcal O_K^1}\Phi(l(\beta)-h)\,dh =\frac{tw_K}{\operatorname{Reg}^1}\int_{\R^c}\Phi(u)\,du . \end{equation} \end{lemma} \begin{proof} The map $l$ is a homomorphism on $K^\times$, so $l(\beta_i\mathcal O_K^1)=l(\beta_i)+l(\mathcal O_K^1)$, and by Proposition~\ref{geo:mass} each point of this translate of $l(\mathcal O_K^1)$ is the image of exactly $w_K$ elements of $\beta_i\mathcal O_K^1$. The translates $\Pi^1-x$, with $x\in l(\mathcal O_K^1)$, tile $\R^c$, so by Tonelli's theorem $\int_{\Pi^1}\sum_{x\in l(\mathcal O_K^1)}\Phi(l(\beta_i)+x-h)\,dh=\int_{\R^c}\Phi(u)\,du$ for each $i$. \end{proof} \begin{definition}\label{geo:window} A \emph{window} is a bounded Borel set $\Omega\subset\C^d$ of positive volume that is invariant under rotation of each coordinate separately. For $u\in\R^c$ let $\nu(u)\in\C^d$ have compact coordinates $1$ and pair coordinates $(e^{u_j},e^{-u_j})$, and for a Borel set $\Omega\subset\C^d$ put \[ \Phi_\Omega(u)=|\Omega\cap(\Omega-\nu(u))|. \] \end{definition} \begin{lemma}\label{geo:phi-omega} Let $\Omega$ be a window, let $h\in\R^c$, and let $\beta\in K$ satisfy $N_{K/F}\beta=1$. Then $|\Omega\cap(\Omega-W_h\beta_\infty)|=\Phi_\Omega(l(\beta)-h)$. \end{lemma} \begin{proof} If $x\in\C^d$ has compact coordinates of modulus one and pair coordinates of moduli $e^{\pm u_j}$, then a rotation of the coordinates preserves $\Omega$ and maps $\nu(u)$ to $x$, so $|\Omega\cap(\Omega-x)|=\Phi_\Omega(u)$. By the remark before Lemma~\ref{geo:unfold-lemma}, $x=W_h\beta_\infty$ is such a point, with $u=l(\beta)-h$. \end{proof} For a window $\Omega$, the function $\Phi_\Omega$ is continuous, bounded by $|\Omega|$, and vanishes when some $e^{|u_j|}$ exceeds the diameter of $\Omega$: if $x$ and $x+\nu(u)$ both lie in $\Omega$, then $e^{|u_j|}$, the modulus of a coordinate of their difference $\nu(u)$, is at most the diameter of $\Omega$. Let $\Omega$ be a window. Fix a $\Z$-basis of $I_\infty$, and for $y\in[0,1)^{2d}$ let $y_I\in\C^d$ be the point with coordinate vector $y$ in this basis. For $h\in\R^c$ let $\mathcal X(h,y)=(\Lambda_h+W_hy_I)\cap\Omega$ and $n(h,y)=|\mathcal X(h,y)|$, and let $\mathcal E(h,y)$ be the number of \emph{ordered edges} of $\mathcal X(h,y)$, that is, of pairs $(x,\beta)$ with $x\in\mathcal X(h,y)$, $\beta\in\bigcup_i\beta_i\mathcal O_K^1$ and $x+W_h\beta_\infty\in\Omega$. Since $\beta\in I$, the point $x+W_h\beta_\infty$ then lies in $\mathcal X(h,y)$. \begin{lemma}\label{geo:joint-average-lemma} Let $\Omega$ be a window and suppose $n(h,y)\le n_{\max}<\infty$ for all $(h,y)$. Then there is a pair $(h,y)$ with $n(h,y)>0$ and \begin{equation}\label{geo:joint-average} \frac{\mathcal E(h,y)}{n(h,y)}\ge \frac{tw_K}{\operatorname{Reg}^1}\, \frac{\int_{\R^c}\Phi_\Omega(u)\,du}{|\Omega|}. \end{equation} \end{lemma} \begin{proof} For fixed $x$, distinct $\beta$ give distinct points $x+W_h\beta_\infty\ne x$, so $\mathcal E\le n^2\le n_{\max}^2$. Both $n$ and $\mathcal E$ are countable sums of indicator functions of Borel sets in $(h,y)$. For fixed $h$, the map $y\mapsto W_hy_I$ has Jacobian $\covol(\Lambda_h)$ and maps $[0,1)^{2d}$ onto a fundamental domain of $\Lambda_h$, so tiling and Lemma~\ref{geo:phi-omega} give \[ \int_{[0,1)^{2d}}n(h,y)\,dy=\frac{|\Omega|}{\covol(\Lambda_h)},\qquad \int_{[0,1)^{2d}}\mathcal E(h,y)\,dy =\frac1{\covol(\Lambda_h)}\sum_{i}\sum_{\beta\in\beta_i\mathcal O_K^1} \Phi_\Omega(l(\beta)-h). \] Average over $h\in\Pi^1$ with respect to $dh/\operatorname{Reg}^1$ and apply \eqref{geo:unfold}. For the probability measure $dh\,dy/\operatorname{Reg}^1$ on $\Pi^1\times[0,1)^{2d}$ this gives $\mathbb E\,\mathcal E=\rho\,\mathbb En$ and $\mathbb En=|\Omega|/\covol(\Lambda_h)>0$, where $\rho$ is the right side of \eqref{geo:joint-average}. If $\mathcal E<\rho n$ held wherever $n>0$, then, since $\mathcal E=0$ where $n=0$ and $\{n>0\}$ has positive measure, we would get $\mathbb E\,\mathcal E<\rho\,\mathbb En$. \end{proof} \begin{lemma}\label{geo:planar} Let $\Omega$ be a window, let $(h,y)$ be a pair, and let $v_0$ be a real place of $F$. For $x\in\C^d$ let $x_{v_0}$ be its compact coordinate at $v_0$, and let $U=\{x_{v_0}:x\in\mathcal X(h,y)\}\subset\C=\R^2$. Then $|U|=n(h,y)$ and $u(U)\ge\mathcal E(h,y)/2$. \end{lemma} \begin{proof} Two distinct points of $\mathcal X(h,y)$ differ by $W_h\gamma_\infty$ with $0\ne\gamma\in I$, whose $v_0$-coordinate $\phi_{v_0}(\gamma)$ is nonzero, so $|U|=n(h,y)$. An ordered edge $(x,\beta)$ gives two points of $U$ at distance $|\phi_{v_0}(\beta)|=1$, and distinct ordered edges give distinct ordered pairs of points. Hence $u(U)\ge\mathcal E(h,y)/2$. \end{proof} \subsection{Uniform Lattice Counting} For a lattice $\Lambda\subset\C^d$ let $\Lambda^*=\{\xi\in\C^d:\langle\xi,x\rangle\in\Z\text{ for all }x\in\Lambda\}$ be its dual lattice, and let $\|\xi\|_*=\sum_w|\xi_w|$, a norm on $\C^d$. We use the Fourier transform $\widehat\Psi(\xi)=\int_{\C^d}\Psi(x)e^{-2\pi i\langle x,\xi\rangle}\,dx$. \begin{lemma}\label{geo:dual} Let $\mathfrak a$ be a fractional ideal of $K$ and $h\in\R^c$. Every nonzero $\xi$ in the dual lattice of $W_h\mathfrak a_\infty$ satisfies \[ \prod_w|\xi_w|\ge\frac{2^d}{\lambda^d(N\mathfrak a)^{1/2}}, \qquad \|\xi\|_*\ge\frac{2d}{\lambda(N\mathfrak a)^{1/(2d)}}. \] In particular, if $N(I)=e^{-dH}$, as for the ideal of Lemma~\ref{geo:selection}, every nonzero dual vector of $\Lambda_h$ satisfies \begin{equation}\label{geo:dual-product} \prod_w|\xi_w|\ge\mu^d,\qquad \mu=2e^{H/2-\ell},\qquad \|\xi\|_*\ge d\mu. \end{equation} \end{lemma} \begin{proof} First let $h=0$. For $x,y\in K$ we have $\langle x_\infty,\overline{y_\infty}\rangle =\operatorname{Re}\sum_w\phi_w(xy)=\tfrac12\Tr_{K/\Q}(xy)$, where the bar is coordinatewise complex conjugation and $\phi_w$ is the embedding of the coordinate $w$, because these embeddings and their complex conjugates are all the embeddings of $K$. Let $\mathfrak a^\vee=\{y\in K:\Tr_{K/\Q}(y\mathfrak a)\subset\Z\}$. It is an $\mathcal O_K$-module, and since the trace form is nondegenerate, the dual basis of a $\Z$-basis of $\mathfrak a$ is a $\Z$-basis of $\mathfrak a^\vee$. So $\mathfrak a^\vee$ is a fractional ideal, and the dual lattice of $\mathfrak a_\infty$ is $\overline{(2\mathfrak a^\vee)_\infty}$. Dual lattices have reciprocal covolumes, and complex conjugation preserves volume. With \eqref{geo:covolume} and $N(2\mathfrak a^\vee)=2^{2d}N(\mathfrak a^\vee)$, this gives $N(\mathfrak a^\vee)=\lambda^{-2d}(N\mathfrak a)^{-1}$. If $0\ne y\in\mathfrak a^\vee$, then $y\mathcal O_K\subset\mathfrak a^\vee$ and $|N_{K/\Q}y|=N(y\mathcal O_K)\ge N(\mathfrak a^\vee)$ \cite[Propositions~4.1 and~4.2]{MilneANT}. For $\xi=\overline{(2y)_\infty}$, \[ \prod_w|\xi_w|=|N_{K/\Q}(2y)|^{1/2} \ge\bigl(2^{2d}\lambda^{-2d}(N\mathfrak a)^{-1}\bigr)^{1/2}. \] For general $h$, the dual lattice of $W_h\mathfrak a_\infty$ is the image of the dual lattice of $\mathfrak a_\infty$ under $W_{-h}$, because $W_h$ is real diagonal, and $W_{-h}$ preserves $\prod_w|\xi_w|$. The bound for $\|\xi\|_*$ follows from the arithmetic-geometric mean inequality. For \eqref{geo:dual-product} take $\mathfrak a=I$. \end{proof} \begin{lemma}\label{geo:poisson} Let $\Lambda\subset\C^d$ be a lattice such that $\prod_w|\xi_w|\ge\mu^d$ for every nonzero $\xi\in\Lambda^*$, where $\mu>0$. Let $\Psi\colon\C^d\to[0,\infty)$ be smooth, bounded and integrable, with $\int\Psi>0$ and all partial derivatives bounded, and suppose that \[ |\widehat\Psi(\xi)|\le M^de^{-\sigma\|\xi\|_*}\int\Psi \qquad(\xi\in\C^d) \] for some $M\ge1$ and $\sigma>0$. If the \emph{Fourier condition} \begin{equation}\label{geo:poisson-condition} \sigma\mu\ge\log M+2\log5+2 \end{equation} holds, then for every $y\in\C^d$, \begin{equation}\label{geo:pointwise-poisson} \sum_{x\in\Lambda+y}\Psi(x)\le\frac2{\covol(\Lambda)}\int\Psi. \end{equation} \end{lemma} \begin{proof} A nonzero $\xi\in\Lambda^*$ has no zero coordinate, and the arithmetic-geometric mean inequality gives $\|\xi\|_*\ge d\mu$. Hence distinct points of $\Lambda^*$ are at $\|\cdot\|_*$-distance at least $d\mu$, and the open $\|\cdot\|_*$-balls of radius $d\mu/2$ about them are disjoint. For $j\ge1$, the balls about the points with $jd\mu\le\|\xi\|_*<(j+1)d\mu$ lie in the ball of radius $(j+3/2)d\mu$ about $0$. Comparing volumes in $\R^{2d}$ bounds the number of these points by $(2j+3)^{2d}\le5^{2dj}$. By the Fourier condition \eqref{geo:poisson-condition} and $M\ge1$, \[ M^d\sum_{\xi\in\Lambda^*\setminus\{0\}}e^{-\sigma\|\xi\|_*} \le\sum_{j\ge1}e^{d(\log M+2j\log5-j\sigma\mu)} \le\sum_{j\ge1}e^{-2dj}\le1 . \] For $\varepsilon>0$ put $\Psi_\varepsilon(x)=\Psi(x)e^{-\varepsilon|x|^2}$, a Schwartz function. Its Fourier transform is $\widehat\Psi*G_\varepsilon$, where $G_\varepsilon(\zeta)=(\pi/\varepsilon)^de^{-\pi^2|\zeta|^2/\varepsilon}$ is a probability density on $\R^{2d}$. Hence $|\widehat{\Psi_\varepsilon}(\xi)|\le m_\varepsilon M^de^{-\sigma\|\xi\|_*}\int\Psi$, with $m_\varepsilon=\int e^{\sigma\|\zeta\|_*}G_\varepsilon(\zeta)\,d\zeta$. The function $y\mapsto\sum_{x\in\Lambda}\Psi_\varepsilon(x+y)$ is smooth and $\Lambda$-periodic, its Fourier coefficient at $\xi\in\Lambda^*$ is $\widehat{\Psi_\varepsilon}(\xi)/\covol(\Lambda)$, and these coefficients are absolutely summable. This gives the Poisson summation formula \[ \sum_{x\in\Lambda}\Psi_\varepsilon(x+y) =\frac1{\covol(\Lambda)}\sum_{\xi\in\Lambda^*}\widehat{\Psi_\varepsilon}(\xi) e^{2\pi i\langle\xi,y\rangle}. \] Using $\widehat{\Psi_\varepsilon}(0)=\int\Psi_\varepsilon$ and the bound above, the sum is at most $(\int\Psi_\varepsilon+m_\varepsilon\int\Psi)/\covol(\Lambda)$. As $\varepsilon\to0$, we have $m_\varepsilon\to1$ by dominated convergence, since $G_\varepsilon$ is the image of $G_1$ under $\zeta\mapsto\sqrt\varepsilon\,\zeta$; moreover $\Psi_\varepsilon$ increases to $\Psi$. Monotone convergence gives \eqref{geo:pointwise-poisson}. \end{proof} The next bound is used in Section~\ref{pf:section} for the Gaussian real-place profile. \begin{lemma}\label{geo:gaussian} If $00$. \end{proof} \subsection{Profiles at Real and Complex Places} \begin{definition}\label{geo:profiles} Fix $0<\delta<1$ and put $p=2/(1+\delta)$, the \emph{Lebesgue exponent} (not a prime), so that $10$. The pair $(f_\R,g_\C)$ is \emph{admissible with Fourier constants $(M,\sigma)$} if: \begin{enumerate} \item[(P1)] $f_\R^p$ and $g_\C^p$ are smooth and bounded, and all their partial derivatives are bounded; \item[(P2)] $A_\R$, $I_\R$, $A_\C$ and $I_\C$ are finite and positive; \item[(P3)] $V_\R$ and $V_\C$ are bounded below and coercive, that is, their sublevel sets are bounded; \item[(P4)] $V_\R(z)$ and $V_\C(z,w)$ have finite second moments under the endpoint laws; \item[(P5)] $|\widehat{f_\R^p}(\xi)|\le Me^{-\sigma|\xi|}A_\R$ and $|\widehat{g_\C^p}(\xi_1,\xi_2)|\le M^2e^{-\sigma(|\xi_1|+|\xi_2|)}A_\C$ for all $\xi,\xi_1,\xi_2\in\C$. \end{enumerate} \end{definition} Section~\ref{pf:section} proves admissibility for the profiles we use. \begin{lemma}\label{geo:tensor} Let $(f_\R,g_\C)$ be admissible with Fourier constants $(M,\sigma)$, and let $\Psi=\prod_vf_\R^p\prod_jg_\C^p$ be the product function on $\C^d$, with one factor for each block. Then $\Psi$ satisfies the hypotheses that Lemma~\ref{geo:poisson} places on $\Psi$, other than the Fourier condition \eqref{geo:poisson-condition}: it is smooth and bounded, with bounded derivatives, $\int\Psi=A_\R^bA_\C^c$, and \[ |\widehat\Psi(\xi)|\le M^de^{-\sigma\|\xi\|_*}\int\Psi\qquad(\xi\in\C^d). \] \end{lemma} \begin{proof} The first properties follow from (P1) and (P2). The Fourier transform of $\Psi$ is the product of the transforms of the factors, so (P5) gives $|\widehat\Psi(\xi)|\le M^{b+2c}e^{-\sigma\|\xi\|_*}\int\Psi =M^de^{-\sigma\|\xi\|_*}\int\Psi$. \end{proof} \begin{lemma}\label{geo:concentration} Let $f_\R$ and $g_\C$ be profiles satisfying (P1), (P2) and (P4), and let $\eta>0$. Let $m_\R$ and $m_\C$ be the means of $V_\R(z)$ and $V_\C(z,w)$ under the endpoint laws, and let $\mathrm{Var}_\R$ and $\mathrm{Var}_\C$ be their variances. For $x\in\C^d$ with compact coordinates $x_v$ and pair coordinates $(x_j^+,x_j^-)$, let $V(x)=\sum_vV_\R(x_v)+\sum_jV_\C(x_j^+,x_j^-)$ be the total energy, put $\bar m=(bm_\R+cm_\C)/d$, and let \[ \Omega=\bigl\{x\in\C^d:V(x)\le d(\bar m+\eta)\bigr\}. \] Then, for all $d$ larger than a bound that depends only on $\eta$, $\mathrm{Var}_\R$ and $\mathrm{Var}_\C$, \[ \int_{\R^c}\Phi_\Omega(u)\,du\ge\tfrac12I_\R^bI_\C^ce^{2d(\bar m-\eta)}. \] \end{lemma} \begin{proof} By (P1), $f_\R$ and $g_\C$ are continuous, so $V$ is continuous and $\Omega$ is closed, hence a Borel set. The function $(x,u)\mapsto e^{-V(x)}e^{-V(x+\nu(u))}/(I_\R^bI_\C^c)$ is the product of the endpoint densities of the blocks, hence a probability density on $\C^d\times\R^c$. Under it, the energies $V(x)$ and $V(x+\nu(u))$ of the two endpoints are sums of independent terms, one for each block, and $V(x)$ has mean $d\bar m$. On a compact block the map $z\mapsto-z-1$, and on a pair block the map $(z,w,u_j)\mapsto(-z-e^{u_j},-w-e^{-u_j},u_j)$, preserves the endpoint density and carries the first endpoint to minus the second. Since the profiles are even, $V(x+\nu(u))$ has the same law as $V(x)$. By Chebyshev's inequality, each of the two energies differs from $d\bar m$ by more than $\eta d$ with probability at most $\max(\mathrm{Var}_\R,\mathrm{Var}_\C)/(\eta^2d)$. So the set $\mathcal Y\subset\C^d\times\R^c$ where both energies lie in $[d(\bar m-\eta),d(\bar m+\eta)]$ has probability at least $1/2$ for all $d$ larger than a bound that depends only on $\eta$, $\mathrm{Var}_\R$ and $\mathrm{Var}_\C$. On $\mathcal Y$ both endpoints lie in $\Omega$ and $e^{-V(x)}e^{-V(x+\nu(u))}\le e^{-2d(\bar m-\eta)}$. Therefore, with $|\mathcal Y|$ the Lebesgue measure of $\mathcal Y$, \[ \int_{\R^c}\Phi_\Omega(u)\,du\ge|\mathcal Y| \ge e^{2d(\bar m-\eta)}I_\R^bI_\C^c\,\mathbb P(\mathcal Y) \ge\tfrac12I_\R^bI_\C^ce^{2d(\bar m-\eta)}. \] \end{proof} \begin{proposition}[Geometric transfer]\label{geo:transfer} Let $0<\delta<1$ and $C\in\R$. Let $(K_i/F_i)_{i\ge1}$ be quadratic extensions satisfying (G1)--(G3) with a common $\lambda$ and with $d_i=[F_i:\Q]\to\infty$, such that $\log L_{F_i}(1)\le d_iC$ and such that every $K_i/F_i$ has the same uniform local types $(e_r,f_r)_{r\in\mathcal R}$. Fix integers $k_r\ge0$, let $H$ and $J$ be as in \eqref{geo:HJ}, and let $\mu=2e^{H/2-\ell}$. Let $(f_\R,g_\C)$ be admissible with Fourier constants $(M,\sigma)$ satisfying the Fourier condition \eqref{geo:poisson-condition}. Put \begin{equation}\label{geo:general-margin} \mathcal M(\theta)=J-\delta H-(1/2-\delta)\ell-C +(1-\delta)\log2+(1-\theta)\log\pi +(1-2\theta)J_\R+\theta J_\C. \end{equation} We call $\mathcal M(\theta)$ the \emph{margin}. If $\inf_i\mathcal M(\theta_i)>0$, where $\theta_i$ is the value of $\theta$ for $F_i$, then there are finite sets $U_i\subset\R^2$ with $|U_i|\to\infty$ and $u(U_i)/|U_i|^{1+\delta}\to\infty$. \end{proposition} \begin{proof} \emph{The window.} Let $m_\R$, $m_\C$, $\mathrm{Var}_\R$ and $\mathrm{Var}_\C$ be as in Lemma~\ref{geo:concentration}; the variances are finite by (P4). Fix $\eta>0$ with $4\eta<\inf_i\mathcal M(\theta_i)$, and consider one extension $K/F$ of the sequence, with $d$ large. Take $I$, $t$ and $\beta_1,\dots,\beta_t$ from Lemma~\ref{geo:selection}, so that $\Lambda_h$ is defined; the counts $n(h,y)$ and $\mathcal E(h,y)$ are those of the window $\Omega$ chosen next. With the total energy $V$ and the mean $\bar m$ of Lemma~\ref{geo:concentration}, let \[ \Omega=\bigl\{x\in\C^d:V(x)\le d(\bar m+\eta)\bigr\}. \] By (P1) and (P3) (see the proof of Lemma~\ref{geo:concentration}), this is a bounded Borel set, and it is invariant under coordinate rotations because the profiles are radial in each coordinate. By Lemma~\ref{geo:concentration}, \[ \int_{\R^c}\Phi_\Omega(u)\,du\ge\tfrac12I_\R^bI_\C^ce^{2d(\bar m-\eta)} \] for all large $d$, uniformly in the signature. In particular $|\Omega|>0$ for large $d$, so $\Omega$ is a window. \emph{Counting points.} With the product function $\Psi=\prod_vf_\R^p\prod_jg_\C^p=e^{-pV}$ of Lemma~\ref{geo:tensor}, we have $\mathbf 1_\Omega\le e^{pd(\bar m+\eta)}\Psi$, so \[ |\Omega|\le e^{pd(\bar m+\eta)}A_\R^bA_\C^c, \] and Lemmas~\ref{geo:dual}, \ref{geo:poisson} and~\ref{geo:tensor} give, for all $(h,y)$, \[ n(h,y)\le\frac{2e^{pd(\bar m+\eta)}A_\R^bA_\C^c}{\covol(\Lambda_h)} =2e^{dY_\eta},\qquad Y_\eta=H+\log2-\ell+(1-2\theta)\log A_\R+\theta\log A_\C +p(\bar m+\eta), \] by \eqref{geo:determinant}, $b/d=1-2\theta$ and $c/d=\theta$. \emph{Counting edges.} Now take a pair $(h,y)$ from Lemma~\ref{geo:joint-average-lemma}, applied with $n_{\max}=2e^{dY_\eta}$. By $t\ge e^{dJ}/(\kappa h_{\rm rel})$, Proposition~\ref{geo:mass}, $\log L_F(1)\le dC$ and $b+c=(1-\theta)d$, \[ \frac{tw_K}{\operatorname{Reg}^1}\ge \frac{e^{dJ}2^{d-1}\pi^{b+c}}{\lambda^{d/2}L_F(1)} \ge\tfrac12\exp\bigl(d[J+\log2+(1-\theta)\log\pi-\ell/2-C]\bigr). \] Combined with the bounds for $|\Omega|$ and for the integral of $\Phi_\Omega$, \eqref{geo:joint-average} shows that the set $\mathcal X=\mathcal X(h,y)$ has $n=|\mathcal X|>0$ points and at least $\tfrac14e^{dX_\eta}n$ ordered edges, where \[ \begin{aligned} X_\eta={}&J+\log2+(1-\theta)\log\pi-\ell/2-C +(1-2\theta)(\log I_\R-\log A_\R)\\ &+\theta(\log I_\C-\log A_\C)+(2-p)\bar m-(2+p)\eta. \end{aligned} \] Since $n\le2e^{dY_\eta}$ and $p(1+\delta)=2$, the terms in $\bar m$ cancel, and \[ \frac{\mathcal E(h,y)}{n^{1+\delta}} \ge\frac{e^{dX_\eta}}{4\cdot2^\delta e^{\delta dY_\eta}} =2^{-2-\delta}e^{d(\mathcal M(\theta)-4\eta)}. \] \emph{The planar set.} Finally let $v_0$ be a real place of $F$, and let $U=\{x_{v_0}:x\in\mathcal X\}\subset\C=\R^2$. By Lemma~\ref{geo:planar}, $|U|=n$ and $u(U)\ge\mathcal E(h,y)/2$, so \[ \frac{u(U)}{|U|^{1+\delta}} \ge2^{-3-\delta}e^{d(\mathcal M(\theta)-4\eta)}, \] which tends to infinity along the sequence. Since $u(U)\le|U|^2/2$ and $\delta<1$, also $|U|\to\infty$. These sets are the $U_i$. The conclusion concerns a sequence of cardinalities, not every large cardinality. \end{proof} \section{Shell Profiles at Split Finite Places} \label{fw:section} We now place locally constant weights at finitely many finite places of $F$ that split in $K$; in Section~\ref{sec:certificate} these are the places above $2$, $3$, $5$, $7$ and $29$. In Lemma~\ref{geo:selection}, the count at each prime of $F$ above $r\in\mathcal R$ corresponds to the indicator function of a product of two balls in the completion, the unweighted shell profile of Definition~\ref{fw:shell-profile}; shell profiles replace this indicator by weighted sums of indicators of products of shells (Corollary~\ref{fw:uniform-types}). We keep the hypotheses (G1)--(G3) and the notation of Section~\ref{geo:section}, and put \[ \covol_0=\covol\bigl((\mathcal O_K)_\infty\bigr)=(\lambda/2)^d . \] \subsection{The Local Functional} Let $L$ be a nonarchimedean local field with valuation ring $\mathcal O$, uniformizer $\varpi$ and residue field of cardinality $Q$. We use the additive Haar measure on $L$ with $|\mathcal O|=1$, the product measure on the plane $L^2=L\times L$, and the multiplicative Haar measure $d\omega$ on $\mathcal O^\times$ of total measure one. For $n\in\Z$ and $\omega\in\mathcal O^\times$ put \[ s(n,\omega)=(\varpi^n\omega,\ \varpi^{-n}\omega^{-1})\in L^2,\qquad \mathcal Tg(z)=\sum_{n\in\Z}\int_{\mathcal O^\times}g(z+s(n,\omega))\,d\omega . \] The two coordinates of $s(n,\omega)$ have product one; they model the finite coordinates of an element of relative norm one at a split place. If $g_1$ and $g_2$ are locally constant and compactly supported on $L^2$, only finitely many $n$ contribute to $\int_{L^2}g_1\,\mathcal Tg_2$: if $g_1(z)g_2(z+s(n,\omega))\ne0$, then both coordinates of $s(n,\omega)$ lie in the compact set $\operatorname{supp}g_2-\operatorname{supp}g_1$, which bounds $n$ from above and from below. \begin{definition}\label{fw:local-functional} For locally constant, compactly supported functions $g_1,g_2$ on $L^2$ put $\langle g_1,g_2\rangle_{\mathcal T}=\int_{L^2}g_1\,\mathcal Tg_2$. For a nonnegative, locally constant, compactly supported function $g$ on $L^2$ with $g\ne0$, and $0<\delta<1$, $p=2/(1+\delta)$, put \begin{equation}\label{fw:functional} A(g)=\int_{L^2}g^p,\qquad Z(g)=\langle g,g\rangle_{\mathcal T},\qquad \mathcal F_\delta(g)=\frac{Z(g)}{A(g)^{1+\delta}} . \end{equation} We call $\mathcal F_\delta$ the \emph{local functional}. \end{definition} For $i\in\Z$ let $\mathcal B_i=\varpi^{-i}\mathcal O$, a ball of measure $Q^i$. Write $[x]_+=\max(x,0)$. \begin{lemma}\label{fw:gram-lemma} For all integers $i_1,j_1,i_2,j_2$, \begin{equation}\label{fw:rectangle-gram} \bigl\langle\mathbf 1_{\mathcal B_{i_1}\times\mathcal B_{j_1}}, \mathbf 1_{\mathcal B_{i_2}\times\mathcal B_{j_2}}\bigr\rangle_{\mathcal T} =Q^{\min(i_1,i_2)+\min(j_1,j_2)}\, [\max(i_1,i_2)+\max(j_1,j_2)+1]_+. \end{equation} \end{lemma} \begin{proof} Two balls of an ultrametric space are disjoint or nested. Hence $\mathcal B_{i_1}\cap(\mathcal B_{i_2}-x)$ has measure $Q^{\min(i_1,i_2)}$ if $x\in\mathcal B_{\max(i_1,i_2)}$, and is empty otherwise. The left side of \eqref{fw:rectangle-gram} is therefore $Q^{\min(i_1,i_2)+\min(j_1,j_2)}$ times the measure of the set of $(n,\omega)$ with $\varpi^n\omega\in\mathcal B_{\max(i_1,i_2)}$ and $\varpi^{-n}\omega^{-1}\in\mathcal B_{\max(j_1,j_2)}$. These conditions say $-\max(i_1,i_2)\le n\le\max(j_1,j_2)$ and do not involve $\omega$, and $d\omega$ has total measure one. \end{proof} \begin{definition}\label{fw:shell-profile} Put $\mathcal S_0=\mathcal O$ and $\mathcal S_i=\mathcal B_i\setminus \mathcal B_{i-1}$ for $i\ge1$. These are the \emph{shells}, and their measures $m_0=1$ and $m_i=Q^i-Q^{i-1}$ are the \emph{shell measures}. Let $k\ge0$ be an integer, and let $w=(w_{ij})_{i,j\ge0}$ be nonnegative weights, finitely many of them nonzero, with $w_{00}>0$. The \emph{shell profile} with parameters $(Q,k,w)$ on $L^2$ is \[ g=\sum_{i,j\ge0}w_{ij}\mathbf 1_{\mathcal S_i\times\varpi^{-k}\mathcal S_j}. \] The shell profile with $w_{00}=1$ and $w_{ij}=0$ otherwise, that is, the indicator function of $\mathcal O\times\varpi^{-k}\mathcal O$, is the \emph{unweighted shell profile} with parameters $(Q,k)$. \end{definition} \begin{lemma}\label{fw:shell-invariance} A shell profile $g$ with parameters $(Q,k,w)$ is nonnegative, locally constant and compactly supported. It is invariant under $(x,y)\mapsto(\omega x,\omega'y)$ for all $\omega,\omega'\in\mathcal O^\times$; in particular it is even. It is also invariant under translation by the group \begin{equation}\label{fw:hard-period} \mathcal C=\mathcal O\times\varpi^{-k}\mathcal O,\qquad |\mathcal C|=Q^k. \end{equation} \end{lemma} \begin{proof} The first assertion holds because the shells are compact open sets and finitely many of the nonnegative weights are nonzero. Multiplication by a unit preserves valuations and hence each shell. Adding an element of $\mathcal O$ does not change the shell of a point, whatever the number of shells; this gives the invariance under $\mathcal C$, applied to the first coordinate and to $\varpi^k$ times the second. \end{proof} We call $\mathcal C$ the \emph{period} of $g$. Define the symmetric matrix $\Delta=(\Delta_{ii'})_{i,i'\ge0}$ by \[ \Delta_{ii'}=m_{\min(i,i')}\ (i\ne i'),\qquad \Delta_{00}=0,\qquad \Delta_{ii}=im_i-Q^{i-1}\ (i\ge1), \] and the matrix $\mathcal G$ by \[ \mathcal G_{(i,j),(i',j')}=(k+1)m_im_j\mathbf 1_{i=i',\,j=j'} +m_j\Delta_{ii'}\mathbf 1_{j=j'}+m_i\Delta_{jj'}\mathbf 1_{i=i'}. \] \begin{lemma}\label{fw:shell-lemma} The shell profile $g$ with parameters $(Q,k,w)$ satisfies \begin{equation}\label{fw:shell-kernel} A(g)=Q^k\sum_{i,j}m_im_jw_{ij}^p,\qquad Z(g)=Q^k\sum_{(i,j),(i',j')}w_{ij}\,\mathcal G_{(i,j),(i',j')}\,w_{i'j'} . \end{equation} All entries of $\mathcal G$ are nonnegative, and $Z(g)\ge(k+1)Q^kw_{00}^2>0$. In particular $\mathcal F_\delta(g)$ depends only on the parameters, and the unweighted shell profile has $\mathcal F_\delta(g)=(k+1)Q^{-\delta k}$. \end{lemma} \begin{proof} The formula for $A(g)$ holds because the sets $\mathcal S_i\times\varpi^{-k}\mathcal S_j$ are disjoint of measure $Q^km_im_j$. For $Z(g)$, write $\mathbf 1_{\mathcal S_0}=\mathbf 1_{\mathcal B_0}$ and $\mathbf 1_{\mathcal S_i}=\mathbf 1_{\mathcal B_i}-\mathbf 1_{\mathcal B_{i-1}}$ for $i\ge1$, and note $\varpi^{-k}\mathcal B_j=\mathcal B_{j+k}$. Only balls with nonnegative indices occur, so by \eqref{fw:rectangle-gram} and $k\ge0$ the positive part can be dropped, and $\langle\mathbf 1_{\mathcal B_{i_1}\times\mathcal B_{j_1+k}}, \mathbf 1_{\mathcal B_{i_2}\times\mathcal B_{j_2+k}}\rangle_{\mathcal T}$, for $i_1,j_1,i_2,j_2\ge0$, is \[ Q^k\bigl[(k+1)Q^{\min(i_1,i_2)}Q^{\min(j_1,j_2)} +\max(i_1,i_2)Q^{\min(i_1,i_2)}Q^{\min(j_1,j_2)} +Q^{\min(i_1,i_2)}\max(j_1,j_2)Q^{\min(j_1,j_2)}\bigr]. \] For a function $\varphi$ of two nonnegative integers put $(\partial\varphi)(i,i')=\varphi(i,i')-\varphi(i-1,i')-\varphi(i,i'-1) +\varphi(i-1,i'-1)$, where terms with an index $-1$ are omitted; this is the expansion of the shells into balls. Apply $\partial$ in $(i_1,i_2)$ and in $(j_1,j_2)$. A direct computation gives, for $i,i'\ge0$, that $\partial Q^{\min(i_1,i_2)}$ at $(i,i')$ is $m_i\mathbf 1_{i=i'}$, and that $\partial\bigl(\max(i_1,i_2)Q^{\min(i_1,i_2)}\bigr)$ is $\Delta_{ii'}$. For example, when $1\le i0$, so $g(z)g(z+s(n,\omega))/Z(g)$ is a probability density on $L^2\times\Z\times\mathcal O^\times$, with counting measure on $\Z$. As at the archimedean places (Definition~\ref{geo:profiles}), we call it the \emph{endpoint law} of $g$; its first endpoint is $z$ and its second endpoint is $z+s(n,\omega)$. \begin{corollary}\label{fw:product-weights} Let $x=(x_i)_{i\ge0}$ and $x'=(x'_j)_{j\ge0}$ be finitely supported vectors of nonnegative numbers with $x_0x'_0>0$, put \[ P_x=\sum_im_ix_i^p,\qquad S_x=\sum_im_ix_i^2,\qquad R_x=\sum_{i,i'}x_i\Delta_{ii'}x_{i'}, \] and define $P_{x'},S_{x'},R_{x'}$ in the same way. The shell profile $g$ with parameters $(Q,k,w)$ and product weights $w_{ij}=x_ix'_j$ satisfies \begin{equation}\label{fw:product-shells} A(g)=Q^kP_xP_{x'},\qquad Z(g)=Q^k\bigl[(k+1)S_xS_{x'}+R_xS_{x'}+S_xR_{x'}\bigr]. \end{equation} \end{corollary} \begin{proof} Substitute $w_{ij}=x_ix'_j$ in \eqref{fw:shell-kernel}. \end{proof} \begin{remark}\label{fw:one-shell} Suppose that only $\mathcal S_0$ and $\mathcal S_1$ carry weight, with $x=x'=(1,t)$ and $t\ge0$. Then $S_x=1+(Q-1)t^2$ and $R_x=2t+(Q-2)t^2$, and the difference between $\log\mathcal F_\delta(g)$ and the value $\log\bigl((k+1)Q^{-\delta k}\bigr)$ of the unweighted shell profile is \begin{equation}\label{fw:one-shell-gain} \log\frac{(k+1)S_x^2+2S_xR_x}{k+1}-2(1+\delta)\log\bigl(1+(Q-1)t^p\bigr). \end{equation} Since $p>1$, its right derivative at $t=0$ is $4/(k+1)>0$, so a sufficiently small weight on $\mathcal S_1$ always improves on the unweighted shell profile. \end{remark} Section~\ref{sec:certificate} uses the exact formulas \eqref{fw:product-shells}. \subsection{Finite Periods and Lattice Counting} Let $\mathcal P$ be a finite set of finite places of $F$ that split in $K$. For $\mathfrak p\in\mathcal P$ write $\mathfrak p\mathcal O_K=\mathfrak P_{\mathfrak p}\,\iota\mathfrak P_{\mathfrak p}$, where we also write $\mathfrak p$ for the prime ideal of $F$. The completion of $K$ at $\mathfrak P_{\mathfrak p}$ is $F_{\mathfrak p}$; for $x\in K$ let $x_{\mathfrak p}\in F_{\mathfrak p}$ be its image under the completion map. The map $x\mapsto(x_{\mathfrak p},(\iota x)_{\mathfrak p})$ records the completions at $\mathfrak P_{\mathfrak p}$ and at $\iota\mathfrak P_{\mathfrak p}$, and $\iota$ exchanges the two coordinates. If $N_{K/F}\beta=1$, then $\beta_{\mathfrak p}(\iota\beta)_{\mathfrak p}=1$. Let $\mathcal O_{\mathfrak p}$, $\varpi_{\mathfrak p}$ and $Q_{\mathfrak p}$ be the valuation ring, a uniformizer and the residue cardinality of $F_{\mathfrak p}$. \begin{definition}\label{fw:finite-coordinates} Put \[ \mathbb A_{\mathcal P}=\prod_{\mathfrak p\in\mathcal P} (F_{\mathfrak p}\times F_{\mathfrak p}), \] with the Haar measure that gives $\prod_{\mathfrak p}(\mathcal O_{\mathfrak p}\times\mathcal O_{\mathfrak p})$ measure one. For a point $(z,\zeta)$ of $\C^d\times\mathbb A_{\mathcal P}$ we call $z$ its \emph{archimedean part} and $\zeta$ its \emph{finite part}. Let $\Gamma=\mathcal O_{K,\mathcal P}$ be the ring of elements of $K$ that are integral at every prime other than the $\mathfrak P_{\mathfrak p}$ and $\iota\mathfrak P_{\mathfrak p}$. We embed $\Gamma$ in $\C^d\times\mathbb A_{\mathcal P}$ by $\gamma\mapsto(\gamma_\infty,\gamma_f)$, where $\gamma_f=(\gamma_{\mathfrak p},(\iota\gamma)_{\mathfrak p})_{\mathfrak p}$; thus $\gamma_\infty$ and $\gamma_f$ are the archimedean part and the finite part of the image of $\gamma$. A subgroup of $\mathbb A_{\mathcal P}$ of the form \[ \mathcal C_f=\prod_{\mathfrak p\in\mathcal P}\bigl(\varpi_{\mathfrak p}^{a_{\mathfrak p}}\mathcal O_{\mathfrak p}\times\varpi_{\mathfrak p}^{b_{\mathfrak p}}\mathcal O_{\mathfrak p}\bigr),\qquad a_{\mathfrak p},b_{\mathfrak p}\in\Z, \] is called \emph{rectangular}. \end{definition} A rectangular group has measure $|\mathcal C_f|=\prod_{\mathfrak p}Q_{\mathfrak p}^{-a_{\mathfrak p}-b_{\mathfrak p}}$. \begin{lemma}\label{fw:lattice} Let $\mathcal C_f$ be rectangular and put $\mathfrak L=\prod_{\mathfrak p}\mathfrak P_{\mathfrak p}^{a_{\mathfrak p}}(\iota\mathfrak P_{\mathfrak p})^{b_{\mathfrak p}}$. \begin{enumerate} \item[(a)] The elements $\gamma\in\Gamma$ with $\gamma_f\in\mathcal C_f$ form $\mathfrak L$, and $N\mathfrak L=|\mathcal C_f|^{-1}$, so $\covol(\mathfrak L_\infty)=\covol_0/|\mathcal C_f|$. \item[(b)] Every coset of $\mathcal C_f$ in $\mathbb A_{\mathcal P}$ contains $\gamma_f$ for some $\gamma\in\Gamma$. \item[(c)] $\Gamma$ is a lattice in $\C^d\times\mathbb A_{\mathcal P}$, that is, a discrete subgroup with a fundamental domain of finite measure. If $P_{\mathfrak L}$ is a fundamental parallelepiped of $\mathfrak L_\infty$, then $P_{\mathfrak L}\times\mathcal C_f$ is a fundamental domain for $\Gamma$, of measure $\covol_0$. \item[(d)] For $h\in\R^c$, the set $\Gamma_h=\{(W_h\gamma_\infty,\gamma_f):\gamma\in\Gamma\}$ is again a lattice, with fundamental domain $W_hP_{\mathfrak L}\times\mathcal C_f$ of measure $\covol_0$. Put $H_f=d^{-1}\log|\mathcal C_f|$. Every nonzero vector $\xi$ of the dual lattice of $W_h\mathfrak L_\infty$ satisfies $\prod_w|\xi_w|\ge\mu_f^d$, where $\mu_f=2e^{H_f/2-\ell}$. \end{enumerate} \end{lemma} \begin{proof} (a) The valuation of $\gamma_{\mathfrak p}$ is the exponent of $\mathfrak P_{\mathfrak p}$ in $\gamma$, and that of $(\iota\gamma)_{\mathfrak p}$ is the exponent of $\iota\mathfrak P_{\mathfrak p}$. Together with integrality at the other primes, $\gamma_f\in\mathcal C_f$ says exactly that $\gamma\in\mathfrak L$. As $N\mathfrak P_{\mathfrak p}=N(\iota\mathfrak P_{\mathfrak p})=Q_{\mathfrak p}$, we get $N\mathfrak L=\prod_{\mathfrak p}Q_{\mathfrak p}^{a_{\mathfrak p}+b_{\mathfrak p}}=|\mathcal C_f|^{-1}$, and the covolume follows from the covolume formula \eqref{geo:covolume}. (b) Choose $\tilde a_{\mathfrak p}\le a_{\mathfrak p}$ and $\tilde b_{\mathfrak p}\le b_{\mathfrak p}$ such that the given coset lies in $\tilde{\mathcal C}_f=\prod_{\mathfrak p}(\varpi_{\mathfrak p}^{\tilde a_{\mathfrak p}}\mathcal O_{\mathfrak p}\times\varpi_{\mathfrak p}^{\tilde b_{\mathfrak p}}\mathcal O_{\mathfrak p})$, and let $\tilde{\mathfrak L}\supset\mathfrak L$ be the corresponding ideal. The map $\gamma\mapsto\gamma_f$ induces a homomorphism $\tilde{\mathfrak L}/\mathfrak L\to\tilde{\mathcal C}_f/\mathcal C_f$. It is injective by (a), applied to $\mathcal C_f$. Both groups are finite of the same order: $|\tilde{\mathcal C}_f/\mathcal C_f|=|\tilde{\mathcal C}_f|/|\mathcal C_f|$, and $|\tilde{\mathfrak L}/\mathfrak L|=N\mathfrak L/N\tilde{\mathfrak L}$ \cite[Proposition~4.2]{MilneANT}, which is the same number by (a). So the map is bijective, and every coset of $\mathcal C_f$ in $\tilde{\mathcal C}_f$ contains $\gamma_f$ for some $\gamma\in\tilde{\mathfrak L}\subset\Gamma$. No principality is needed. (c) Given $(z,\zeta)\in\C^d\times\mathbb A_{\mathcal P}$, by (b) there is $\gamma\in\Gamma$ with $\zeta-\gamma_f\in\mathcal C_f$, and then a unique $\gamma'\in\mathfrak L$ with $z-(\gamma+\gamma')_\infty\in P_{\mathfrak L}$. If $(z,\zeta)$ had two such representations, the difference of the two elements of $\Gamma$ would have finite part in $\mathcal C_f$, hence lie in $\mathfrak L$ by (a), and then it would vanish because $P_{\mathfrak L}$ is a fundamental domain. Thus $P_{\mathfrak L}\times\mathcal C_f$ is a fundamental domain, of measure $(\covol_0/|\mathcal C_f|)\cdot|\mathcal C_f|=\covol_0$. Finally $\Gamma$ is discrete: for a bounded open set $Y\ni0$ in $\C^d$, its intersection with the neighborhood $Y\times\mathcal C_f$ of $0$ consists of the finitely many $\gamma\in\mathfrak L$ with $\gamma_\infty\in Y$. (d) The map $(z,\zeta)\mapsto(W_hz,\zeta)$ preserves measure, because $W_h$ is real diagonal of determinant one, and it carries $\Gamma$ to $\Gamma_h$ and $P_{\mathfrak L}\times\mathcal C_f$ to $W_hP_{\mathfrak L}\times\mathcal C_f$; so the first assertion follows from (c). By (a), $N\mathfrak L=e^{-dH_f}$, and Lemma~\ref{geo:dual} gives the bound for the dual vectors. \end{proof} Accordingly, the Fourier condition used in this section is \begin{equation}\label{fw:fourier-condition} \sigma\cdot2e^{H_f/2-\ell}\ge\log M+2\log5+2 . \end{equation} \begin{lemma}\label{fw:counting} Let $(f_\R,g_\C)$ be admissible with Fourier constants $(M,\sigma)$, and let $\Psi_\infty=\prod_vf_\R^p\prod_jg_\C^p$ on $\C^d$, with one factor for each real place $v$ and each complex place $j$ of $F$ (Lemma~\ref{geo:tensor}). Let $\Psi_f\ge0$ be a function on $\mathbb A_{\mathcal P}$ that is invariant under translation by a rectangular group $\mathcal C_f$ and vanishes outside a finite union of its cosets. If \eqref{fw:fourier-condition} holds, with $H_f$ defined by this $\mathcal C_f$, then for every $h\in\R^c$ and every $z_0\in\C^d\times\mathbb A_{\mathcal P}$ (a point of the whole space, not an archimedean part), \begin{equation}\label{fw:weighted-count} \sum_{(z,\zeta)\in\Gamma_h+z_0}\Psi_\infty(z)\Psi_f(\zeta) \le\frac2{\covol_0}\int\Psi_\infty\int\Psi_f . \end{equation} \end{lemma} \begin{proof} Group the points by the coset of $\mathcal C_f$ containing their finite part. If two points of $\Gamma_h+z_0$ have finite parts in the same coset, their difference comes from an element of $\mathfrak L$, by Lemma~\ref{fw:lattice}(a). So the archimedean parts of the points in one coset form a translate of $W_h\mathfrak L_\infty$, or the empty set. By Lemma~\ref{geo:poisson}, with the dual bound of Lemma~\ref{fw:lattice}(d) and the Fourier bound for $\Psi_\infty$ of Lemma~\ref{geo:tensor}, the sum of $\Psi_\infty$ over such a translate is at most $2|\mathcal C_f|\int\Psi_\infty/\covol_0$. The values of $\Psi_f$ on its finitely many nonzero cosets sum to $\int\Psi_f/|\mathcal C_f|$. \end{proof} \begin{remark}\label{fw:rescaling} Let $\alpha\in\mathbb A_{\mathcal P}$ have coordinates $(\varpi_{\mathfrak p}^{n_{\mathfrak p}},\varpi_{\mathfrak p}^{-n_{\mathfrak p}})$. Multiplication by $\alpha$ preserves the Haar measure, and for a rectangular group $\mathcal C_f$ the group $\alpha\mathcal C_f$ is rectangular, of the same measure. Hence, if $\Psi_f$ satisfies the hypotheses of Lemma~\ref{fw:counting} with $\mathcal C_f$, then $\Psi_f\circ\alpha$ satisfies them with the rectangular group $\alpha^{-1}\mathcal C_f$, which gives the same $H_f$, and $\int\Psi_f\circ\alpha=\int\Psi_f$. \end{remark} \subsection{Class Selection} For $u\in\R^c$ and $n=(n_{\mathfrak p})\in\Z^{\mathcal P}$ let $\nu(u,n)\in\C^d\times\mathbb A_{\mathcal P}$ have archimedean part $\nu(u)$ and finite part $(\varpi_{\mathfrak p}^{n_{\mathfrak p}},\varpi_{\mathfrak p}^{-n_{\mathfrak p}})_{\mathfrak p}$. \begin{lemma}\label{fw:selection} Let $\mathcal N\subset\Z^{\mathcal P}$ be finite, and let $\mathcal W\colon\mathcal N\to[0,\infty)$ have positive sum. There are a subset $\mathcal N'\subset\mathcal N$ with \[ \sum_{n\in\mathcal N'}\mathcal W(n)\ge\frac1{\kappa h_{\rm rel}} \sum_{n\in\mathcal N}\mathcal W(n), \] an element $n_0\in\mathcal N'$ with $\mathcal W(n_0)>0$, and elements $\beta_n\in\Gamma$, for $n\in\mathcal N'$, with $N_{K/F}\beta_n=1$ and \[ \beta_n\mathcal O_K=\prod_{\mathfrak p}\mathfrak P_{\mathfrak p}^{\,n_{\mathfrak p}-n_{0,\mathfrak p}} (\iota\mathfrak P_{\mathfrak p})^{-(n_{\mathfrak p}-n_{0,\mathfrak p})}. \] The cosets $\beta_n\mathcal O_K^1$ are pairwise disjoint, and $\beta_n\mathcal O_K^1$ is the set of all elements of relative norm one that generate this ideal. Moreover, let $\varpi_0\in\mathbb A_{\mathcal P}$ have coordinates $(\varpi_{\mathfrak p}^{n_{0,\mathfrak p}},\varpi_{\mathfrak p}^{-n_{0,\mathfrak p}})$. Then for $n\in\mathcal N'$ and $\beta\in\beta_n\mathcal O_K^1$, the point $\varpi_0\beta_f$ has coordinates $(\varpi_{\mathfrak p}^{n_{\mathfrak p}}\omega_{\mathfrak p},\varpi_{\mathfrak p}^{-n_{\mathfrak p}}\omega_{\mathfrak p}^{-1})$ with units $\omega_{\mathfrak p}\in\mathcal O_{\mathfrak p}^\times$. \end{lemma} \begin{proof} Map $n$ to the class of $\mathfrak A(n)=\prod_{\mathfrak p}\mathfrak P_{\mathfrak p}^{n_{\mathfrak p}}$ in $\Cl(K)$ modulo the image of $\Cl(F)$, a group of order $\kappa h_{\rm rel}$. Let $\mathcal N'$ be a fiber on which the sum of $\mathcal W$ is largest; this sum is positive, so we can choose $n_0\in\mathcal N'$ with $\mathcal W(n_0)>0$. For $n\in\mathcal N'$ write $\mathfrak A(n)\mathfrak A(n_0)^{-1}=x_n\mathfrak a_n\mathcal O_K$ with $x_n\in K^\times$ and $\mathfrak a_n$ a fractional ideal of $F$, and put $\beta_n=x_n/\iota x_n$. As in the proof of Lemma~\ref{geo:selection}, $N_{K/F}\beta_n=1$ and $\beta_n$ generates the displayed ideal, which is supported on the primes $\mathfrak P_{\mathfrak p},\iota\mathfrak P_{\mathfrak p}$; hence $\beta_n\in\Gamma$. Two elements of relative norm one generating the same ideal differ by an element of $\mathcal O_K^1$, and distinct $n\in\mathcal N'$ give distinct ideals. Finally, if $\beta\in\beta_n\mathcal O_K^1$, then $\beta_{\mathfrak p}$ has valuation $n_{\mathfrak p}-n_{0,\mathfrak p}$ and $(\iota\beta)_{\mathfrak p}=\beta_{\mathfrak p}^{-1}$, which gives the last assertion. \end{proof} \subsection{Transfer with Shell Profiles} \begin{proposition}[Transfer with shell profiles]\label{fw:transfer} Let $0<\delta<1$ and $C\in\R$. Let $(K_j/F_j)_{j\ge1}$ be quadratic extensions satisfying (G1)--(G3) with a common $\lambda$, with $d_j=[F_j:\Q]\to\infty$ and $\log L_{F_j}(1)\le d_jC$. For each $j$ let $\mathcal P_j$ be a finite set of finite places of $F_j$ that split in $K_j$, and for $\mathfrak p\in\mathcal P_j$ let $g_{\mathfrak p}$ be a shell profile on $F_{\mathfrak p}^2$ with parameters $(Q_{\mathfrak p},k_{\mathfrak p},w^{(\mathfrak p)})$, where the parameters range over a fixed finite set independent of $j$. Let $\mathcal F_{\delta,\mathfrak p}=\mathcal F_\delta(g_{\mathfrak p})$, let $(f_\R,g_\C)$ be admissible with Fourier constants $(M,\sigma)$, and suppose that \eqref{fw:fourier-condition} holds for every $j$, with $\mathcal C_f=\prod_{\mathfrak p\in\mathcal P_j}(\mathcal O_{\mathfrak p}\times\varpi_{\mathfrak p}^{-k_{\mathfrak p}}\mathcal O_{\mathfrak p})$ the product of the periods of the $g_{\mathfrak p}$, that is, with $H_f=d_j^{-1}\sum_{\mathfrak p\in\mathcal P_j}k_{\mathfrak p}\log Q_{\mathfrak p}$. Let $\theta_j$ be the value of $\theta$ for $F_j$, and put \begin{equation}\label{fw:margin} \begin{split} \mathcal M_{{\rm fin},j}={}& \frac1{d_j}\sum_{\mathfrak p\in\mathcal P_j}\log\mathcal F_{\delta,\mathfrak p} -(1/2-\delta)\ell-C+(1-\delta)\log2+(1-\theta_j)\log\pi\\ &+(1-2\theta_j)J_\R+\theta_jJ_\C ; \end{split} \end{equation} we call $\mathcal M_{{\rm fin},j}$ the \emph{margin} of $K_j/F_j$ with these shell profiles. If $\inf_j\mathcal M_{{\rm fin},j}>0$, then there are finite sets $U_j\subset\R^2$ with $|U_j|\to\infty$ and $u(U_j)/|U_j|^{1+\delta}\to\infty$. \end{proposition} Compared with \eqref{geo:general-margin}, the margin \eqref{fw:margin} has no separate term $-\delta H$: by \eqref{fw:shell-kernel}, each $\mathcal F_{\delta,\mathfrak p}$ contains the factor $Q_{\mathfrak p}^{-\delta k_{\mathfrak p}}$, and Corollary~\ref{fw:uniform-types} shows that for uniform local types and unweighted shell profiles \eqref{fw:margin} reduces to \eqref{geo:general-margin}. \begin{proof}[Proof of Proposition~\ref{fw:transfer}] Fix $\eta>0$ with $4\eta<\inf_j\mathcal M_{{\rm fin},j}$, and consider one extension $K/F$ of the sequence, with $d$ large. We drop the sequence index $j$ and write $\mathcal M_{\rm fin}$ for the margin of $K/F$; below, $j$ indexes the complex places of $F$. \emph{The profile and the probability measure.} On $\C^d\times\mathbb A_{\mathcal P}$ let $f=\prod_vf_\R\prod_{j=1}^cg_\C\prod_{\mathfrak p\in\mathcal P}g_{\mathfrak p}$, with one factor for each real place $v$ of $F$, each complex place $j$ of $F$ and each $\mathfrak p\in\mathcal P$, let $V=-\log f$ on the set where $f>0$, and put \[ A=\int f^p=A_\R^bA_\C^c\prod_{\mathfrak p\in\mathcal P}A(g_{\mathfrak p}),\qquad Z=I_\R^bI_\C^c\prod_{\mathfrak p\in\mathcal P}Z(g_{\mathfrak p}). \] Besides the points $\nu(u,n)$ defined before Lemma~\ref{fw:selection}, let $\nu(u,n,\omega)$, for $\omega=(\omega_{\mathfrak p})\in\prod_{\mathfrak p}\mathcal O_{\mathfrak p}^\times$, have archimedean part $\nu(u)$ and finite part $(\varpi_{\mathfrak p}^{n_{\mathfrak p}}\omega_{\mathfrak p},\varpi_{\mathfrak p}^{-n_{\mathfrak p}}\omega_{\mathfrak p}^{-1})_{\mathfrak p}$. Let $\mathbb P$ be the probability measure on $(\C^d\times\mathbb A_{\mathcal P})\times\R^c\times\Z^{\mathcal P}\times\prod_{\mathfrak p}\mathcal O_{\mathfrak p}^\times$ with density $f(x)f(x+\nu(u,n,\omega))/Z$ with respect to $dx\,du\,d\omega$ and counting measure in $n$, where $d\omega$ is the Haar probability measure. It is the product of the endpoint laws of the archimedean blocks (Definition~\ref{geo:profiles}) and of the endpoint laws of the $g_{\mathfrak p}$. Its endpoints are $x$ and $x+\nu(u,n,\omega)$. Their energies $V(x)$ and $V(x+\nu(u,n,\omega))$ have the same distribution and are sums of independent terms, one for each block and each $\mathfrak p\in\mathcal P$: at the archimedean blocks this is shown in the proof of Lemma~\ref{geo:concentration}, and at $\mathfrak p\in\mathcal P$ the map $(z,n,\omega)\mapsto(-z-s(n,\omega),n,\omega)$ preserves the endpoint law of $g_{\mathfrak p}$ and carries its first endpoint to minus the second, $g_{\mathfrak p}$ is even, and the energy $-\log g_{\mathfrak p}$ of the first endpoint takes finitely many values. Let $d\bar m$ be their common mean; this $\bar m$ includes the places of $\mathcal P$ and is not the $\bar m$ of Lemma~\ref{geo:concentration}. The $Q_{\mathfrak p}$ take finitely many values, since the parameters of the shell profiles do, and each is a power of the residue characteristic of $\mathfrak p$; so these characteristics lie in a fixed finite set, and $|\mathcal P|=O(d)$. Since the parameters range over a finite set, the variance of either energy is $O(d)$. Let $\chi$ be the indicator of the event that both energies lie in $[d(\bar m-\eta),d(\bar m+\eta)]$. By Chebyshev's inequality, as in the proof of Lemma~\ref{geo:concentration}, $\mathbb P(\chi=1)\ge1/2$ for large $d$. \emph{The set $\Omega$.} Let $\Omega$ be the set of points of $\C^d\times\mathbb A_{\mathcal P}$ where $f>0$ and $V\le d(\bar m+\eta)$; it plays the role of the window of Definition~\ref{geo:window}. It is a bounded Borel set: the archimedean energies are coercive and bounded below, and the finite part lies in the compact support of $\prod_{\mathfrak p}g_{\mathfrak p}$, on which the finite energies are bounded. It is invariant under rotations of the archimedean coordinates, and, by Lemma~\ref{fw:shell-invariance}, under $(x_{\mathfrak p},y_{\mathfrak p})\mapsto(\omega x_{\mathfrak p},\omega'y_{\mathfrak p})$ for units at each $\mathfrak p\in\mathcal P$ and under translation by $\mathcal C_f$. For $u\in\R^c$ and $n\in\Z^{\mathcal P}$ let $\Phi(u,n)=|\Omega\cap(\Omega-\nu(u,n))|$, and let $\mathcal W(n)=\int_{\R^c}\Phi(u,n)\,du$. Only finitely many $n$ have $\mathcal W(n)\ne0$. By these invariances, the measure of the intersection of $\Omega$ with its translate by any vector whose archimedean part has the moduli of $\nu(u)$ and whose finite part is that of some $\nu(u,n,\omega)$ equals $\Phi(u,n)$. Where $\chi=1$, both endpoints lie in $\Omega$ and $f(x)f(x+\nu(u,n,\omega))\le e^{-2d(\bar m-\eta)}$. Integrating over $x$, $u$ and $\omega$ and summing over $n$, we obtain \[ \begin{split} \sum_n\mathcal W(n) &=\sum_n\int\mathbf 1_\Omega(x)\,\mathbf 1_\Omega(x+\nu(u,n,\omega))\, dx\,du\,d\omega\\ &\ge e^{2d(\bar m-\eta)}\sum_n\int\chi\,f(x)f(x+\nu(u,n,\omega))\, dx\,du\,d\omega =e^{2d(\bar m-\eta)}Z\,\mathbb P(\chi=1) \ge\tfrac12Ze^{2d(\bar m-\eta)} . \end{split} \] Moreover $\mathbf 1_\Omega\le e^{pd(\bar m+\eta)}f^p$, so $|\Omega|\le Ae^{pd(\bar m+\eta)}$. \emph{Class selection and rescaling.} Apply Lemma~\ref{fw:selection} to the finitely many $n$ with $\mathcal W(n)>0$, obtaining $n_0$, $\mathcal N'$ and $\beta_n$, and let $\varpi_0$ be as in that lemma. Put $\Omega'=\varpi_0^{-1}\Omega$, where $\varpi_0$ acts on the finite part. Then $|\Omega'|=|\Omega|$, and $\Omega'$ is invariant under the rectangular group $\mathcal C'_f=\varpi_0^{-1}\mathcal C_f$, of measure $|\mathcal C_f|$. For $n\in\mathcal N'$, $\beta\in\beta_n\mathcal O_K^1$ and $h\in\R^c$, the element $\beta^{(h)}=(W_h\beta_\infty,\beta_f)\in \Gamma_h$ satisfies \[ |\Omega'\cap(\Omega'-\beta^{(h)})|=|\Omega\cap(\Omega-(W_h\beta_\infty, \varpi_0\beta_f))|=\Phi(l(\beta)-h,n), \] by the last assertion of Lemma~\ref{fw:selection} and the invariances of $\Omega$, as in Lemma~\ref{geo:phi-omega}. \emph{Points and edges.} Let $\mathfrak L'$ be the ideal attached to $\mathcal C'_f$ in Lemma~\ref{fw:lattice}. Fix a $\Z$-basis of $\mathfrak L'_\infty$, let $P_{\mathfrak L'}$ be the fundamental parallelepiped that it spans, and for $y\in[0,1)^{2d}$ let $y_{\mathfrak L'}\in P_{\mathfrak L'}$ be the point with coordinate vector $y$ in this basis. For $h\in\R^c$, $y\in[0,1)^{2d}$ and $\zeta\in\mathcal C'_f$ put $z_0=(W_hy_{\mathfrak L'},\zeta)$; these points form the fundamental domain $W_hP_{\mathfrak L'}\times\mathcal C'_f$ of $\Gamma_h$ of Lemma~\ref{fw:lattice}(d). Let $\mathcal X(h,z_0)=(\Gamma_h+z_0)\cap\Omega'$ and let $N(h,z_0)=|\mathcal X(h,z_0)|$ be the number of its points (the analogue of the count $n(h,y)$ of Section~\ref{geo:section}), and let $\mathcal E(h,z_0)$ count the pairs $(x,\beta)$ with $x\in\mathcal X(h,z_0)$, $\beta\in\bigcup_{n\in\mathcal N'}\beta_n\mathcal O_K^1$ and $x+\beta^{(h)}\in\Omega'$. Averages over $(h,z_0)$ are taken with respect to the probability measure $dh\,dy\,d\zeta/(\operatorname{Reg}^1|\mathcal C'_f|)$ on $\Pi^1\times[0,1)^{2d}\times\mathcal C'_f$. Since $\mathbf 1_{\Omega'}\le e^{pd(\bar m+\eta)}(f\circ\varpi_0)^p$ and $(f\circ\varpi_0)^p$ is the product of an archimedean factor and a function invariant under $\mathcal C'_f$, Lemma~\ref{fw:counting} and Remark~\ref{fw:rescaling} give \[ N(h,z_0)\le N_{\max}=\frac{2Ae^{pd(\bar m+\eta)}}{\covol_0} \] for all $(h,z_0)$. By tiling, for fixed $h$ the average of $N(h,z_0)$ over $(y,\zeta)$ is $|\Omega'|/\covol_0$, and the average of $\mathcal E(h,z_0)$ is $\covol_0^{-1}\sum_{n\in\mathcal N'}\sum_{\beta\in\beta_n\mathcal O_K^1} \Phi(l(\beta)-h,n)$. Averaging over $h\in\Pi^1$ as in Lemma~\ref{geo:unfold-lemma}, the average of $\mathcal E$ becomes \[ \frac{w_K}{\operatorname{Reg}^1\covol_0}\sum_{n\in\mathcal N'}\mathcal W(n) \ge\frac{w_K}{\kappa h_{\rm rel}\operatorname{Reg}^1\covol_0} \sum_n\mathcal W(n). \] Measurability and integrability hold as in Lemma~\ref{geo:joint-average-lemma}, since $\mathcal E\le N^2\le N_{\max}^2$. The argument of that lemma yields a pair $(h,z_0)$ with $N=N(h,z_0)>0$ and \[ \frac{\mathcal E(h,z_0)}N\ge\frac{w_K}{\kappa h_{\rm rel} \operatorname{Reg}^1}\,\frac{\sum_n\mathcal W(n)}{|\Omega|} \ge\frac{w_K}{\kappa h_{\rm rel}\operatorname{Reg}^1}\, \frac{Ze^{2d(\bar m-\eta)}}{2Ae^{pd(\bar m+\eta)}} . \] With $N\le N_{\max}$ and $p(1+\delta)=2$, \[ \frac{\mathcal E(h,z_0)}{N^{1+\delta}}\ge \frac{w_K\covol_0^\delta}{2^{1+\delta}\kappa h_{\rm rel} \operatorname{Reg}^1}\,\frac{Z}{A^{1+\delta}}\,e^{-4\eta d}. \] By \eqref{geo:mass-density} and $\log L_F(1)\le dC$, the first factor is at least $2^{-2-\delta}2^{d(1-\delta)}\pi^{(1-\theta)d}\lambda^{-(1/2-\delta)d}e^{-dC}$, and $Z/A^{1+\delta}=\exp\bigl(bJ_\R+cJ_\C+\sum_{\mathfrak p}\log\mathcal F_{\delta,\mathfrak p}\bigr)$. Hence $\mathcal E/N^{1+\delta}\ge2^{-2-\delta}e^{d(\mathcal M_{\rm fin}-4\eta)}$. \emph{The planar set.} Let $v_0$ be a real place of $F$ and let $U$ be the set of $v_0$-coordinates of the points of $\mathcal X(h,z_0)$. Distinct points differ by $(W_h\gamma_\infty,\gamma_f)$ with $0\ne\gamma\in\Gamma$, whose $v_0$-coordinate is nonzero, so $|U|=N$. Each counted pair $(x,\beta)$ gives two points at distance $|\phi_{v_0}(\beta)|=1$, and distinct pairs give distinct ordered pairs of points. As in the proof of Proposition~\ref{geo:transfer} (Lemma~\ref{geo:planar}), $u(U)\ge\mathcal E/2$, the ratio $u(U)/|U|^{1+\delta}$ tends to infinity, and the bound $u(U)\le|U|^2/2$ forces $|U|\to\infty$. \end{proof} \subsection{Uniform Local Types} \begin{corollary}[Uniform local types]\label{fw:uniform-types} Let $0<\delta<1$ and $C\in\R$. Let $(K_j/F_j)_{j\ge1}$ be quadratic extensions satisfying (G1)--(G3) with a common $\lambda$, with $d_j=[F_j:\Q]\to\infty$ and $\log L_{F_j}(1)\le d_jC$, and with the same uniform local types $(e_r,f_r)_{r\in\mathcal R}$ (Definition~\ref{geo:uniform-types}). For each $r\in\mathcal R$ fix an integer $k_r\ge0$ and weights $w^{(r)}$ as in Definition~\ref{fw:shell-profile}, let $\mathcal F_{\delta,r}$ be the value of the local functional $\mathcal F_\delta$ at the shell profile with parameters $(r^{f_r},k_r,w^{(r)})$, and let $H$ be as in \eqref{geo:HJ}. Let $(f_\R,g_\C)$ be admissible with Fourier constants $(M,\sigma)$ satisfying \eqref{geo:poisson-condition} with $\mu=2e^{H/2-\ell}$, and put \[ \begin{split} \mathcal M_{\mathcal R}(\theta)={}& \sum_{r\in\mathcal R}\frac{\log\mathcal F_{\delta,r}}{e_rf_r} -(1/2-\delta)\ell-C+(1-\delta)\log2+(1-\theta)\log\pi\\ &+(1-2\theta)J_\R+\theta J_\C . \end{split} \] If $\inf_j\mathcal M_{\mathcal R}(\theta_j)>0$, where $\theta_j$ is the value of $\theta$ for $F_j$, then there are finite sets $U_j\subset\R^2$ with $|U_j|\to\infty$ and $u(U_j)/|U_j|^{1+\delta}\to\infty$. For unweighted shell profiles, $\sum_r\log\mathcal F_{\delta,r}/(e_rf_r)=J-\delta H$, so that $\mathcal M_{\mathcal R}$ is the margin $\mathcal M$ of \eqref{geo:general-margin}. \end{corollary} \begin{proof} Apply Proposition~\ref{fw:transfer} with $\mathcal P_j$ the set of all places of $F_j$ above $\mathcal R$, which split in $K_j$, and with $g_{\mathfrak p}$ the shell profile with parameters $(r^{f_r},k_r,w^{(r)})$ for $\mathfrak p$ above $r$; these parameters take $|\mathcal R|$ values, and $\mathcal F_{\delta,\mathfrak p}=\mathcal F_{\delta,r}$. Since there are $d/(e_rf_r)$ places above $r$, each with $Q_{\mathfrak p}=r^{f_r}$, \begin{equation}\label{fw:uniform} \frac1d\sum_{\mathfrak p\in\mathcal P}\log\mathcal F_{\delta,\mathfrak p} =\sum_{r\in\mathcal R}\frac{\log\mathcal F_{\delta,r}}{e_rf_r}, \qquad H_f=\sum_{r\in\mathcal R}\frac{k_r\log r}{e_r}=H . \end{equation} In particular \eqref{fw:fourier-condition} coincides with \eqref{geo:poisson-condition}, whatever the shell weights, and the margin \eqref{fw:margin} of $K_j/F_j$ equals $\mathcal M_{\mathcal R}(\theta_j)$. For unweighted shell profiles, Lemma~\ref{fw:shell-lemma} gives $\log\mathcal F_{\delta,r}=\log(k_r+1)-\delta k_rf_r\log r$, and the sum in \eqref{fw:uniform} is $J-\delta H$. \end{proof} Thus Proposition~\ref{fw:transfer} contains Proposition~\ref{geo:transfer}, and the shell weights replace the term $J-\delta H$ of \eqref{geo:general-margin} by $\sum_r\log\mathcal F_{\delta,r}/(e_rf_r)$ without changing the Fourier condition. \section{Archimedean Profiles} \label{pf:section} We use a Gaussian at the compact coordinates, and at the pair coordinates a Student profile multiplied by a positive polynomial. \begin{definition}\label{pf:student} Let $s>0$ and $a>0$, and let $P$ be a real polynomial in two variables that is positive on $[0,1]^2$. For $(z,w)\in\C^2$ put \[ t=(1+a|z|^2)^{-1},\qquad t'=(1+a|w|^2)^{-1},\qquad g_\C(z,w)=t^s\,t'^{\,s}P(t,t'). \] We call $g_\C$ the \emph{Student--Bernstein profile} with parameters $(s,a,P)$. For $P\equiv1$ it is the \emph{Student profile} $t^st'^{\,s}$; for $s>1$, each factor $(1+a|z|^2)^{-s}$ is proportional to the density of a bivariate Student $t$-distribution. If $P$ has degree at most $m$ in each variable, its \emph{Bernstein coefficients} of degree $m$ are the numbers $\beta_{ij}$ with \[ P(t,t')=\sum_{i,j=0}^m\beta_{ij}\,b_i^m(t)\,b_j^m(t'),\qquad b_i^m(t)=\binom mi t^i(1-t)^{m-i}. \] \end{definition} The Bernstein polynomials $b_i^m$ are nonnegative on $[0,1]$ and sum to one, so $\min\beta\le P\le\max\beta$ on $[0,1]^2$, where $\min\beta$ and $\max\beta$ are the smallest and largest Bernstein coefficients. From now on, $g_\C$ denotes a Student--Bernstein profile with parameters $(s,a,P)$. Section~\ref{sec:certificate} uses \[ s=1.14600585,\qquad a=0.0000128633241758, \] and the polynomial $P$ of degree $m=3$ in each variable whose Bernstein coefficients form the symmetric matrix \begin{equation}\label{pf:bernstein-witness} (\beta_{ij})=\begin{pmatrix} 1&2.4129438977&2.7721799210&6.6857798061\\ 2.4129438977&3.8532983888&4.6371750119&8.1307168183\\ 2.7721799210&4.6371750119&5.4628959084&9.5512647588\\ 6.6857798061&8.1307168183&9.5512647588&13.8659430610 \end{pmatrix}. \end{equation} All finite decimals here are exact rational numbers. \emph{Heuristic.} The shape of $g_\C$ is adapted to the displacements at a pair of coordinates. By \eqref{geo:norm-modulus} they have moduli $(e^u,e^{-u})$, so as $u$ varies they run along a hyperbola. A profile of width $a^{-1/2}$ in each coordinate stays close to its translate for all $u$ with $e^{|u|}\lesssim a^{-1/2}$, a range of length about $\log(1/a)$; this produces the factor $\log(1/a)$ in Proposition~\ref{pf:overlap}, against the term $2\delta\log a$ in \eqref{pf:collected-functional}, which comes from the exponent $1+\delta$ on the mass $A_\C$. For Student--Bernstein profiles the overlap $I_\C$ reduces to beta integrals and digamma values (Lemmas~\ref{pf:laplace}--\ref{pf:pair-reduction}), which makes rigorous bounds possible, and the polynomial $P$ couples the two radii. In this section we prove that these profiles are admissible in the sense of Definition~\ref{geo:admissible}, and we derive bounds for the functionals $J_\R$ and $J_\C$ (Definition~\ref{geo:profiles}) that reduce to finite computations with rational data. Section~\ref{sec:certificate} carries out these computations at $\delta=\deltastar$. Throughout, $0<\delta<1$ and $p=2/(1+\delta)$. We write $\psi=\Gamma'/\Gamma$, $\gamma$ for Euler's constant, $(x)_n$ for the rising factorial, $H_n=\sum_{k=1}^n1/k$ with $H_0=0$, and $\mathrm{Beta}(\alpha,\alpha')$ for the beta distribution on $(0,1)$ with density proportional to $t^{\alpha-1}(1-t)^{\alpha'-1}$. \subsection{The Gaussian at the Compact Coordinates} \begin{lemma}\label{pf:gaussian} Let $a_\R=2p\delta$ and $f_\R(z)=e^{-a_\R|z|^2/p}$, so that $f_\R^p=e^{-a_\R|z|^2}$. Then \begin{equation}\label{pf:gaussian-compact} J_\R=\log(p/2)+\delta\log(a_\R/\pi)-a_\R/(2p) =\log(p/2)+\delta\bigl(\log(a_\R/\pi)-1\bigr), \end{equation} and $a_\R$ maximizes $J_\R$ over the Gaussians $e^{-a'|z|^2/p}$, $a'>0$. \end{lemma} \begin{proof} For $f=e^{-a'|z|^2/p}$ we have $A_\R=\pi/a'$ and, since $|z|^2+|z+1|^2=2|z+1/2|^2+1/2$, $I_\R=(\pi p/(2a'))e^{-a'/(2p)}$. Hence $J_\R=\log(p/2)+\delta\log(a'/\pi)-a'/(2p)$, a concave function of $\log a'$ with its maximum at $a'=2p\delta$. \end{proof} By \eqref{geo:general-margin}, the margin is affine in $\theta$ with slope $-\log\pi-2J_\R+J_\C$. This slope depends on the profiles; for ours it is positive (Section~\ref{sec:certificate}). \subsection{The Pair Convolution} \begin{lemma}\label{pf:laplace} Let $\alpha,\alpha'>0$ with $A=\alpha+\alpha'-1>0$, and let $h\in\C$. Then \begin{equation}\label{pf:laplace-convolution} \int_{\C}(1+|z|^2)^{-\alpha}(1+|z+h|^2)^{-\alpha'}\,dz =\frac\pi A\,\mathbb E\bigl(1+T(1-T)|h|^2\bigr)^{-A}, \qquad T\sim\mathrm{Beta}(\alpha,\alpha'). \end{equation} \end{lemma} \begin{proof} Insert $(1+|z|^2)^{-\alpha}=\Gamma(\alpha)^{-1}\int_0^\infty r^{\alpha-1}e^{-r(1+|z|^2)}\,dr$ and the analogous formula with a variable $r'$. Since $r|z|^2+r'|z+h|^2=(r+r')|z+r'h/(r+r')|^2+rr'|h|^2/(r+r')$, the integral of $e^{-r|z|^2-r'|z+h|^2}$ over $z\in\C$ is $\pi(r+r')^{-1}e^{-rr'|h|^2/(r+r')}$; the factors $e^{-r}$ and $e^{-r'}$ remain in the Laplace integrals. Substitute $r=RT$, $r'=R(1-T)$, with Jacobian $R$, and integrate over $R$. This gives \[ \frac{\pi\,\Gamma(A)}{\Gamma(\alpha)\Gamma(\alpha')} \int_0^1T^{\alpha-1}(1-T)^{\alpha'-1}\bigl(1+T(1-T)|h|^2\bigr)^{-A}dT, \] which equals the right side of \eqref{pf:laplace-convolution} because $\Gamma(\alpha+\alpha')=A\,\Gamma(A)$. All integrands are nonnegative, so Tonelli's theorem justifies the interchanges. \end{proof} For $A,B>0$ and $y>0$ define \[ Q_{AB}(y)=\int_\R(1+ye^{2v})^{-A}(1+ye^{-2v})^{-B}\,dv . \] \begin{lemma}\label{pf:hyperbola} Let $A,B>0$. \begin{enumerate} \item[(a)] For $y>0$, \begin{equation}\label{pf:elementary-hyperbola-error} \Bigl|Q_{AB}(y)-\log(1/y)+\frac{\psi(A)+\psi(B)+2\gamma}{2}\Bigr| \le\frac{(A+B)y}2 . \end{equation} \item[(b)] For $00$. Then \[ I_\C=\frac{\pi^2}{a^2}\sum_{(i,j,k,l)}w_{ijkl}\, \mathbb E\,Q_{A_{ik}B_{jl}}(y),\qquad y=a\sqrt{T(1-T)T'(1-T')}\le a/4, \] where the sum runs over all crosses and, for each cross, $T\sim\mathrm{Beta}(s+i,s+k)$ and $T'\sim\mathrm{Beta}(s+j,s+l)$ are independent. Each expectation is finite. \end{lemma} \begin{proof} Expand $g_\C(z,w)=\sum_{i,j}d_{ij}(1+a|z|^2)^{-s-i}(1+a|w|^2)^{-s-j}$ and multiply out $g_\C(z,w)g_\C(z+e^u,w+e^{-u})$. For one cross, the scaling $z\mapsto z/\sqrt a$ and Lemma~\ref{pf:laplace} give \[ \int_\C(1+a|z|^2)^{-s-i}(1+a|z+e^u|^2)^{-s-k}\,dz =\frac\pi{aA_{ik}}\,\mathbb E\bigl(1+aT(1-T)e^{2u}\bigr)^{-A_{ik}}, \] and similarly in $w$ with $e^{-2u}$, $B_{jl}$ and $T'$. By Tonelli's theorem, the $u$-integral of the product is $\mathbb E\int_\R(1+Ye^{2u})^{-A_{ik}}(1+Y'e^{-2u})^{-B_{jl}}du$ with $Y=aT(1-T)$ and $Y'=aT'(1-T')$, and the translation $u\mapsto u+\frac14\log(Y'/Y)$ turns the inner integral into $Q_{A_{ik}B_{jl}}(\sqrt{YY'})$. By \eqref{pf:elementary-hyperbola-error}, $Q_{AB}(y)$ is at most $\log(1/y)$ plus a constant, and $\log(1/(T(1-T)))$ is integrable under every beta law; so each term is finite and the finite signed sum is legitimate. \end{proof} \subsection{Admissibility} For the Fourier bound we assume that $P$ has positive Bernstein coefficients $\beta_{ij}$ of some degree $m\ge1$. \begin{definition}\label{pf:tube-constant} Let $\Delta_1$ and $\Delta_2$ be the forward difference operators in the two indices of $(\beta_{ij})$. For $0<\rho<1$, the width of the complex tube in the proof of Lemma~\ref{pf:tube}, put $\omega=(\rho+\rho^2)/(1-\rho^2)$, \[ E_{rr'}=\binom mr\binom m{r'}\max_{i,j}\bigl|\Delta_1^r\Delta_2^{r'} \beta_{ij}\bigr|,\qquad q_0=\frac1{\min\beta}\sum_{r+r'>0}E_{rr'}\,\omega^{r+r'} . \] We call $q_0$ the \emph{tube constant}. \end{definition} \begin{lemma}\label{pf:tube} Let $s>0$, $sp>1$ and $a>0$, let $P$ have positive Bernstein coefficients $\beta_{ij}$ of degree $m\ge1$, let $0<\rho<1$, and suppose that the tube constant satisfies $q_0<1$. Then for all $\xi_1,\xi_2\in\C$, \[ \bigl|\widehat{g_\C^p}(\xi_1,\xi_2)\bigr| \le K_\C e^{-\sigma_\C(|\xi_1|+|\xi_2|)}A_\C,\qquad K_\C=(1-\rho^2)^{-2sp}(1+q_0)^p,\qquad \sigma_\C=2\pi\rho/\sqrt a . \] \end{lemma} \begin{proof} Write $x=\sqrt a\,z\in\R^2$ and consider complex vectors $x+iy$ with $x,y\in\R^2$ and $|y|\le\rho$. Put $\mathcal D(x+iy)=1+(x_1+iy_1)^2+(x_2+iy_2)^2$. Then \[ \operatorname{Re}\mathcal D=1+|x|^2-|y|^2\ge(1-\rho^2)(1+|x|^2),\qquad \bigl|\mathcal D^{-1}-(1+|x|^2)^{-1}\bigr| \le\frac{2|x|\rho+\rho^2}{(1-\rho^2)(1+|x|^2)^2}\le\omega . \] So the complexified $t=\mathcal D^{-1}$ lies within $\omega$ of the real $t_0=(1+|x|^2)^{-1}\in(0,1]$, and likewise for $t'$. By the Bernstein derivative formula, $\partial_t^r\partial_{t'}^{r'}P/(r!\,r'!)=\binom mr\binom m{r'} \sum_{i,j}(\Delta_1^r\Delta_2^{r'}\beta)_{ij}\,b_i^{m-r}(t)b_j^{m-r'}(t')$, which is at most $E_{rr'}$ in absolute value on $[0,1]^2$. Taylor expansion of the polynomial $P$ about $(t_0,t'_0)$ therefore gives $|P(t,t')-P(t_0,t'_0)|\le q_0\min\beta\le q_0P(t_0,t'_0)$. Thus $P(t,t')$ lies in the right half-plane, and with principal branches, \[ \bigl|\mathcal D(x+iy)^{-sp}\mathcal D(x'+iy')^{-sp}P(t,t')^p\bigr| \le K_\C\,t_0^{sp}\,t_0'^{\,sp}P(t_0,t'_0)^p , \] where $x'+iy'$ is the variable attached to $w$. The same bounds hold for $|y|,|y'|<\rho'$ with some $\rho'>\rho$, since $q_0$ depends continuously on $\rho$, so the left side is holomorphic in each complex coordinate on a neighborhood of the closed tube. The function $g_\C^p$ is invariant under rotations of $z$ and of $w$. Rotate so that $\xi_1$ and $\xi_2$ are positive multiples of the first basis vector, and shift the first real coordinate of $x$ by $-i\rho$, then that of $x'$ by $-i\rho$. Each shift is justified by Cauchy's theorem in one variable: the integrand is holomorphic in the strip and bounded there by a constant times $(1+|x|^2)^{-sp}(1+|x'|^2)^{-sp}$, which tends to zero at infinity and is integrable because $sp>1$. After the shifts the Fourier factor is multiplied by $e^{-2\pi\rho(|\xi_1|+|\xi_2|)/\sqrt a}$, and the displayed bound gives the lemma. \end{proof} \begin{proposition}\label{pf:admissible} Let $s\ge1$, $a>0$, and let $P$ have positive Bernstein coefficients $\beta_{ij}$ of some degree $m\ge1$ in each variable. Let $0<\delta<1$ with $a_\R=2p\delta\le1$, let $f_\R$ be the Gaussian of Lemma~\ref{pf:gaussian}, and let $0<\rho<1$ with $q_0<1$. Then $(f_\R,g_\C)$ is admissible with Fourier constants \begin{equation}\label{pf:polynomial-fourier-contract} M=\max\{2,\sqrt{K_\C}\},\qquad \sigma=\min\{1,\sigma_\C\}. \end{equation} \end{proposition} \begin{proof} (P1) The functions $f_\R^p$ and $g_\C^p=(1+a|z|^2)^{-sp}(1+a|w|^2)^{-sp}P(t,t')^p$ are smooth, because $P\ge\min\beta>0$ on $[0,1]^2$, and they and all their derivatives are bounded. (P2) The integrals at the compact coordinates are computed in Lemma~\ref{pf:gaussian}. The mass $A_\C$ is finite because $sp>1$ (Proposition~\ref{pf:mass} below), and $00$ and $s$ and $A=2s-\tau-1>0$, this integral is a multiple of $\mathbb E\,Q_{A,2s-1}(y)$, which is finite by \eqref{pf:elementary-hyperbola-error}. The same holds for $w$. (P5) By Lemma~\ref{geo:gaussian} the Gaussian satisfies $|\widehat{f_\R^p}(\xi)|\le2e^{-|\xi|}A_\R\le Me^{-\sigma|\xi|}A_\R$, and by Lemma~\ref{pf:tube} the pair profile satisfies the bound with $K_\C e^{-\sigma_\C(|\xi_1|+|\xi_2|)}\le M^2e^{-\sigma(|\xi_1|+|\xi_2|)}$. \end{proof} \subsection{\texorpdfstring{A Lower Bound for the Overlap $I_\C$}{A Lower Bound for the Overlap}} \begin{definition}\label{pf:overlap-coefficients} For $n\ge0$ and integers $i,k\ge0$ put \[ K_n(i,k)=\frac{(s+i)_n(s+k)_n}{n!\,(\eta_0+i+k+n)_{n+1}},\qquad D_x(j)=\sum_{h=0}^{j-1}\frac1{x+h}\quad(D_x(0)=0), \] \[ r_n(i,k)=-\frac1{\eta_0}+\frac{H_n}2-\frac{D_{\eta_0}(i+k+n)}2 +D_{\eta_0}(i+k+2n+1)-\frac{D_s(i+n)+D_s(k+n)}2, \] $R_n(i,k)=K_n(i,k)r_n(i,k)$, and \[ G_n=\sum d_{ij}d_{kl}K_n(i,k)K_n(j,l),\qquad \widetilde G_n=\sum d_{ij}d_{kl}\bigl[R_n(i,k)K_n(j,l)+K_n(i,k)R_n(j,l)\bigr], \] where the sums run over all crosses. Put \begin{equation}\label{pf:student-constant} c_s=-2\psi(s)+\psi(2s)+(2s-1)^{-1}-\gamma ,\qquad y_0=a/4 . \end{equation} \end{definition} For rational $s$ and $d_{ij}$, the numbers $K_n(i,k)$, $r_n(i,k)$, $R_n(i,k)$, $G_n$ and $\widetilde G_n$ are rational. \begin{proposition}\label{pf:overlap} Let $s\ge1$ and $a>0$, and suppose that for every cross \begin{equation}\label{pf:cross-threshold} A_{ik}B_{jl}y_0^2<1,\qquad 2\log(4/a)\ge H_{\lceil A_{ik}\rceil-1}+H_{\lceil B_{jl}\rceil-1}, \qquad \log(4/a)>1/2 . \end{equation} Then for every integer $n_*\ge0$, \begin{equation}\label{pf:collected-overlap} \frac{I_\C}{\pi^2a^{-2}}\ge \sum_{n=0}^{n_*}a^{2n}\bigl\{G_n[\log(1/a)+c_s]+\widetilde G_n\bigr\} -E_{-,n_*}, \end{equation} where the \emph{remainder term} is \[ E_{-,n_*}=\log(4/a)\sum_{w_{ijkl}<0}(-w_{ijkl}) \frac{(A_{ik}B_{jl}y_0^2)^{n_*+1}}{1-A_{ik}B_{jl}y_0^2}. \] \end{proposition} \begin{proof} Start from Lemma~\ref{pf:pair-reduction}. Since $y\le y_0=a/4<1$, the hypotheses of Lemma~\ref{pf:hyperbola}(c) hold at every value of $y$, with $A=A_{ik}\ge1$ and $B=B_{jl}\ge1$. Split each $Q_{A_{ik}B_{jl}}(y)$ into the terms $n\le n_*$ of Lemma~\ref{pf:hyperbola}(b) and the remainder $\operatorname{Rem}_{n_*}(y)$. Consider the term of index $n$ for one cross, with $\alpha=s+i$, $\alpha'=s+k$, $A=A_{ik}=\alpha+\alpha'-1$, and similarly for $T'$. Its factor $y^{2n}$ is $a^{2n}(T(1-T))^n(T'(1-T'))^n$, and $\log(1/y)=\log(1/a)-\frac12\log(T(1-T))-\frac12\log(T'(1-T'))$. The beta integral gives $\mathbb E(T(1-T))^n=(\alpha)_n(\alpha')_n/(A+1)_{2n}$, so \[ \frac{(A)_n}{A\,n!}\,\mathbb E(T(1-T))^n =\frac{(\alpha)_n(\alpha')_n}{n!\,(A+n)_{n+1}}=K_n(i,k). \] Weighting by $(T(1-T))^n$ turns $\mathrm{Beta}(\alpha,\alpha')$ into $\mathrm{Beta}(\alpha+n,\alpha'+n)$, under which the mean of $\log(T(1-T))$ is $\psi(\alpha+n)+\psi(\alpha'+n)-2\psi(A+2n+1)$. The recurrence $\psi(x+j)=\psi(x)+D_x(j)$ and $\psi(1)=-\gamma$ give \[ \begin{gathered} \psi(n+1)=-\gamma+H_n,\qquad \psi(\alpha+n)=\psi(s)+D_s(i+n),\qquad \psi(\alpha'+n)=\psi(s)+D_s(k+n),\\ \psi(A+n)=\psi(\eta_0)+D_{\eta_0}(i+k+n),\qquad \psi(A+2n+1)=\psi(\eta_0)+D_{\eta_0}(i+k+2n+1),\\ \tfrac12c_s=-\psi(s)+\tfrac12\psi(\eta_0)+\eta_0^{-1}-\tfrac12\gamma, \end{gathered} \] the last one because $\psi(2s)=\psi(\eta_0)+1/\eta_0$. Substituting, \[ \tfrac12\psi(n+1)-\tfrac12\psi(A+n) -\tfrac12\bigl[\psi(\alpha+n)+\psi(\alpha'+n)\bigr]+\psi(A+2n+1) =\tfrac12c_s+r_n(i,k). \] Hence the expectation of the term of index $n$, multiplied by $w_{ijkl}$, is \[ a^{2n}d_{ij}d_{kl}K_n(i,k)K_n(j,l)\bigl[\log(1/a)+c_s+r_n(i,k) +r_n(j,l)\bigr]. \] Summing over crosses gives the displayed main terms. By Lemma~\ref{pf:hyperbola}(c), $0\le\operatorname{Rem}_{n_*}(y)\le\log(1/y)(ABy^2)^{n_*+1}/(1-ABy^2)$. The right side increases with $y$ on $(0,y_0]$, because the derivative of $y^{2n_*+2}\log(1/y)$ is $y^{2n_*+1}((2n_*+2)\log(1/y)-1)>0$ when $\log(1/y)>1/2$. So $0\le\operatorname{Rem}_{n_*}(y)\le\log(4/a)(ABy_0^2)^{n_*+1}/(1-ABy_0^2)$. For crosses with $w_{ijkl}>0$ we discard the remainder, and for crosses with $w_{ijkl}<0$ we subtract its upper bound. This proves \eqref{pf:collected-overlap}. \end{proof} With $n_*=1$ the bound keeps the exact term of order $a^2$; the terms it discards or subtracts are of order $a^4\log(1/a)$. It is a lower bound for $I_\C$ only. \subsection{\texorpdfstring{An Upper Bound for the Mass $A_\C$}{An Upper Bound for the Mass}} \begin{proposition}\label{pf:mass} Let $g_\C$ be a Student--Bernstein profile with parameters $(s,a,P)$ such that $q=sp-1>0$. Then \begin{equation}\label{pf:polynomial-mass} A_\C=\frac{\pi^2}{a^2q^2}\,\bar A_\C,\qquad \bar A_\C=\mathbb E\,P(T,T')^p, \end{equation} where $T$ and $T'$ are independent with density $qt^{q-1}$ on $(0,1)$. Partition $[0,1]^2$ into rectangles $\mathcal Q=[l,r]\times[l',r']$, and use on $\mathcal Q$ the coordinates $x=(T-l)/(r-l)$ and $x'=(T'-l')/(r'-l')$. Suppose that the Bernstein coefficients, of some degree, of the restriction of $P$ to $\mathcal Q$ in these coordinates lie in $[m_{\mathcal Q},M_{\mathcal Q}]$ with $m_{\mathcal Q}>0$, and put \[ c_{\mathcal Q}=\frac{m_{\mathcal Q}+M_{\mathcal Q}}2,\qquad \rho_{\mathcal Q}=\frac{M_{\mathcal Q}-m_{\mathcal Q}} {M_{\mathcal Q}+m_{\mathcal Q}},\qquad P=c_{\mathcal Q}(1+W_{\mathcal Q})\ \text{on }\mathcal Q . \] Define \begin{equation}\label{pf:subrectangle-moments} M_i(l,r)=\int_l^rqv^{q-1}\Bigl(\frac{v-l}{r-l}\Bigr)^idv =\frac q{(r-l)^i}\sum_{j=0}^i\binom ij(-l)^{i-j} \frac{r^{q+j}-l^{q+j}}{q+j}, \end{equation} and, writing $W_{\mathcal Q}^k=\sum_{i,j}c^{(k)}_{ij}x^ix'^{\,j}$, \[ \Sigma_{\mathcal Q,N}=c_{\mathcal Q}^p\sum_{k=0}^N\binom pk \sum_{i,j}c^{(k)}_{ij}M_i(l,r)M_j(l',r'). \] Then for every $N\ge1$, \begin{equation}\label{pf:subrectangle-mass} \Bigl|\bar A_\C-\sum_{\mathcal Q}\Sigma_{\mathcal Q,N}\Bigr| \le\sum_{\mathcal Q}c_{\mathcal Q}^p \frac{|\binom p{N+1}|\rho_{\mathcal Q}^{N+1}}{1-\rho_{\mathcal Q}} M_0(l,r)M_0(l',r'). \end{equation} \end{proposition} \begin{proof} The substitution $t=(1+a|z|^2)^{-1}$ gives $dz=\pi\,dt/(at^2)$ on $\C$. Hence \[ A_\C=\frac{\pi^2}{a^2}\int_0^1\!\!\int_0^1t^{sp-2}t'^{\,sp-2} P(t,t')^p\,dt\,dt', \] which is \eqref{pf:polynomial-mass} since $sp-2=q-1$. The Bernstein convex-hull property gives $|W_{\mathcal Q}|\le\rho_{\mathcal Q}<1$ on $\mathcal Q$. For $10$. \end{proof} If $P$ is symmetric and both axes are partitioned alike, a rectangle and its reflection contribute equally, so off-diagonal rectangles may be counted twice and diagonal ones once. Subsection~\ref{cert:computations} describes how the quantities in Proposition~\ref{pf:mass} are evaluated. \subsection{\texorpdfstring{A Lower Bound for the Functional $J_\C$}{A Lower Bound for the Functional}} \begin{corollary}\label{pf:functional} Under the hypotheses of Propositions~\ref{pf:overlap} and~\ref{pf:mass}, let $Z_*$ be the right side of \eqref{pf:collected-overlap} for some $n_*$, and suppose $Z_*>0$. If $\bar A_\C^*\ge\bar A_\C$, then \begin{equation}\label{pf:collected-functional} J_\C\ge\log Z_*+2(1+\delta)\log q-2\delta\log\pi +2\delta\log a-(1+\delta)\log\bar A_\C^* . \end{equation} \end{corollary} \begin{proof} By Proposition~\ref{pf:overlap}, $I_\C\ge\pi^2a^{-2}Z_*$, and by \eqref{pf:polynomial-mass}, $A_\C\le\pi^2a^{-2}q^{-2}\bar A_\C^*$. Substitute into $J_\C=\log I_\C-(1+\delta)\log A_\C$. \end{proof} Interval evaluation of the right side of \eqref{pf:collected-functional} gives a rigorous lower bound for $J_\C$; the upper endpoint of such an enclosure is not an upper bound for $J_\C$. For rational $s$, all digamma values in Proposition~\ref{pf:overlap} reduce to rational corrections and to $c_s=2/\eta_0+\psi(\eta_0)-2\psi(s)+\psi(1)$, since $\psi(2s)=\psi(\eta_0)+1/\eta_0$ and $\psi(1)=-\gamma$. The digamma function at rational arguments is enclosed as follows. \begin{lemma}\label{pf:digamma} For $x>0$ and integers $L\ge0$ and $k_*\ge1$, put $z=x+L$. Then \[ \Bigl|\psi(x)-\Bigl(\log z-\frac1{2z} -\sum_{k=1}^{k_*-1}\frac{\mathrm B_{2k}}{2kz^{2k}} -\sum_{j=0}^{L-1}\frac1{x+j}\Bigr)\Bigr| \le\frac{|\mathrm B_{2k_*}|}{2k_*z^{2k_*}}, \] where $\mathrm B_{2k}$ are the Bernoulli numbers. \end{lemma} \begin{proof} By Binet's formula \cite[Equation~(5.9.15)]{DLMF}, \[ \psi(z)=\log z-\frac1{2z} -2\int_0^\infty\frac{t\,dt}{(t^2+z^2)(e^{2\pi t}-1)} . \] Write \[ \frac1{t^2+z^2}=\sum_{k=1}^{k_*-1}(-1)^{k-1}\frac{t^{2k-2}}{z^{2k}} +(-1)^{k_*-1}\frac{t^{2k_*-2}}{z^{2k_*-2}(t^2+z^2)}, \] and use $\int_0^\infty t^{2k-1}(e^{2\pi t}-1)^{-1}dt=|\mathrm B_{2k}|/(4k)$ and $(-1)^{k-1}|\mathrm B_{2k}|=\mathrm B_{2k}$. The remaining integral has a fixed sign and is at most $|\mathrm B_{2k_*}|/(2k_*z^{2k_*})$ in absolute value, because $(t^2+z^2)^{-1}\le z^{-2}$. Finally apply the recurrence $\psi(x+1)=\psi(x)+1/x$ $L$ times. \end{proof} \section{Proof of Theorem~\ref{thm:main}} \label{sec:certificate} We prove Theorem~\ref{thm:main} by applying Corollary~\ref{fw:uniform-types} to the fields of Theorem~\ref{tw:field-family}, with the upper bound for the relative zeta value in Theorem~\ref{an:ceiling} and the data of Definition~\ref{cert:data}. Propositions~\ref{cert:exact} and~\ref{cert:enclosures} are verified by computer, in exact rational arithmetic and in outward-rounded interval arithmetic, as described in Subsection~\ref{cert:computations}; the other steps are proved in the text. The inequalities between decimals in Propositions~\ref{cert:exact} and~\ref{cert:enclosures} and in Remarks~\ref{cert:signature-remark} and~\ref{cert:observations}, and the entries of Tables~\ref{cert:tab-local} and~\ref{cert:tab-budget}, are computed and printed by the supplementary program \texttt{geom241.py}. In the inequalities the decimals are rounded outward from interval enclosures, and the table entries are rounded to ten decimals. No rounded decimal is substituted into a later calculation; in particular, $\bar A_\C^*$ in Definition~\ref{cert:lower-bounds} is the exact upper end of a computed enclosure, not its rounded decimal. \subsection{Parameters and Local Data} \begin{definition}\label{cert:data} The \emph{data} of this section are the following. \begin{enumerate} \item[(a)] $\delta=4273/100000=\deltastar$, $p=2/(1+\delta)=200000/104273$ and $C=4871285/10^8=\ceiling$; $\lambda=2^{9/4}\sqrt{3615}$ and $\ell=\log\lambda$, as in Theorem~\ref{tw:field-family}. \item[(b)] At the compact coordinates, $f_\R$ is the Gaussian of Lemma~\ref{pf:gaussian}, with $f_\R^p=e^{-a_\R|z|^2}$ and $a_\R=2p\delta=17092/104273<1$. \item[(c)] At the pair coordinates, $g_\C$ is the Student--Bernstein profile of Definition~\ref{pf:student} (the polynomial $P$ below is positive on $[0,1]^2$, since its Bernstein coefficients are positive by Proposition~\ref{cert:exact}(b)) with $s=1.14600585$, $a=0.0000128633241758$ and the polynomial $P$ of degree $m=3$ in each variable whose Bernstein coefficients are \eqref{pf:bernstein-witness}. We put $q=sp-1=12492817/10427300$. \item[(d)] In Definition~\ref{pf:tube-constant} and Lemma~\ref{pf:tube} we take $\rho=10^{-3}$, with the Bernstein coefficients of degree $m=3$ of (c). In Proposition~\ref{pf:mass} we partition $[0,1]^2$ into the $64$ squares of side $1/8$, use on each square the Bernstein coefficients of degree $3$, with $m_{\mathcal Q}$ and $M_{\mathcal Q}$ the smallest and largest of these coefficients, and take $N=18$. In Proposition~\ref{pf:overlap} we take $n_*=1$. \item[(e)] $\mathcal R=\{2,3,5,7,29\}$. For $r\in\mathcal R$, $(e_r,f_r)$ is the absolute type (Definition~\ref{tw:type}) of the primes of $F$ above $r$ given by Theorem~\ref{tw:field-family} (Table~\ref{tw:types-table}, repeated in Table~\ref{cert:tab-local}), the exponent $k_r$ is that of Table~\ref{cert:tab-local}, and $Q_r=r^{f_r}$. For a prime $\mathfrak p$ of $F$ above $r$, $g_r$ is the shell profile on $F_{\mathfrak p}^2$ with parameters $(Q_r,k_r,(w_{r,i}w_{r,j})_{i,j\ge0})$, where $w_{r,0}=1$, the weights $w_{r,i}$ with $1\le i0$ for every $r\in\mathcal R$ and $0\le i1.7518755886$ and $\mu=2e^{H/2-\ell}>17.8683654522$. \item[(h)] $0.6324418249848039<-\log\pi-2J_\R+J_\C^*<0.6324418249848040$. \item[(i)] The lower bound for the margin satisfies \begin{equation}\label{cert:margin} \mathcal M_*(\theta_*)>0.000176730033534598 . \end{equation} \end{enumerate} \end{proposition} \begin{proof} The supplementary program \texttt{geom241.py} evaluates these quantities in outward-rounded interval arithmetic from the exact data, as described in Subsection~\ref{cert:computations}: (a) by Proposition~\ref{pf:mass}; (b) from \eqref{pf:collected-overlap}, with Lemma~\ref{pf:digamma} for the digamma values; (c) from \eqref{pf:collected-functional}; (d) from \eqref{pf:gaussian-compact}; (e) from \eqref{cert:finite-factor}; (f) from the definitions of $H$ and $\ell$; (g) from Lemma~\ref{pf:tube} and the definition of $\mu$; and (h) and (i) from Definition~\ref{cert:lower-bounds}, where $\bar A_\C^*$ is the upper end of the enclosure computed for (a). \end{proof} In (b), the upper endpoint bounds the expression $Z_*$, not the overlap $I_\C$. The proof of Theorem~\ref{thm:main} uses Proposition~\ref{cert:exact} and, from Proposition~\ref{cert:enclosures}, only the fact, established by the computation for (a), that $\bar A_\C\le\bar A_\C^*$, the positivity of $Z_*$ given by (b), and (g), (h) and (i); the other bounds in Proposition~\ref{cert:enclosures} are not used in the proof. Table~\ref{cert:tab-budget} lists the terms of $\mathcal M_*(\theta_*)$. \begin{table}[htbp] \centering\small \caption{The terms of $\mathcal M_*(\theta_*)$, rounded to ten decimals.} \label{cert:tab-budget} \begin{tabular}{lr} \toprule Term & Value\\ \midrule $\sum_r\log\mathcal F_{\delta,r}/(e_rf_r)$ & $0.7214954821$\\ $-(1/2-\delta)\ell$ & $-2.5863212799$\\ $-C$ & $-0.0487128500$\\ $(1-\delta)\log2$ & $0.6635290015$\\ $(1-\theta_*)\log\pi$ & $0.5723736765$\\ $(1-2\theta_*)J_\R$ & $-0.0000032159$\\ $\theta_*J_\C^*$ & $0.6778159157$\\ \midrule $\mathcal M_*(\theta_*)$ & $0.0001767300$\\ \bottomrule \end{tabular} \end{table} \subsection{Conclusion of the Proof} \begin{corollary}\label{cert:fourier} The pair $(f_\R,g_\C)$ of Definition~\ref{cert:data} is admissible with Fourier constants $M=2$ and $\sigma=1$, and these satisfy \eqref{geo:poisson-condition} with $\mu=2e^{H/2-\ell}$. \end{corollary} \begin{proof} Proposition~\ref{pf:admissible} applies, since $s\ge1$, $a>0$, $a_\R\le1$ and $0<\rho<1$, the Bernstein coefficients \eqref{pf:bernstein-witness} of degree $m=3$ are positive, and $q_0<1$ (Proposition~\ref{cert:exact}(b)). It gives the Fourier constants $M=\max\{2,\sqrt{K_\C}\}$ and $\sigma=\min\{1,\sigma_\C\}$. By Proposition~\ref{cert:enclosures}(g), $K_\C<4$ and $\sigma_\C>1$, so $M=2$ and $\sigma=1$. The right side of \eqref{geo:poisson-condition} is then $\log50+2<5.9121$, while $\sigma\mu=\mu>17.8683654522$ by Proposition~\ref{cert:enclosures}(g). \end{proof} \begin{lemma}\label{cert:lower-margin} For the data of Definition~\ref{cert:data}, $J_\C\ge J_\C^*$. Consequently, for $\theta_*\le\theta<1/2$, the margin $\mathcal M_{\mathcal R}(\theta)$ of Corollary~\ref{fw:uniform-types} for these data satisfies \[ \mathcal M_{\mathcal R}(\theta)\ge\mathcal M_*(\theta)\ge\mathcal M_*(\theta_*)>0.000176730033534598 . \] \end{lemma} \begin{proof} Corollary~\ref{pf:functional} applies: the hypotheses of Propositions~\ref{pf:overlap} and~\ref{pf:mass} hold by Proposition~\ref{cert:exact}(c),(d), $Z_*>0$ by \eqref{cert:overlap}, and $\bar A_\C\le\bar A_\C^*$ by Proposition~\ref{cert:enclosures}(a) and Definition~\ref{cert:lower-bounds}. Hence $J_\C\ge J_\C^*$. Since $\mathcal M_{\mathcal R}(\theta)-\mathcal M_*(\theta)=\theta(J_\C-J_\C^*)$ and $\theta\ge0$, we get $\mathcal M_{\mathcal R}(\theta)\ge\mathcal M_*(\theta)$. The function $\mathcal M_*$ is affine in $\theta$ with slope $-\log\pi-2J_\R+J_\C^*$, which is positive by Proposition~\ref{cert:enclosures}(h); so $\mathcal M_*(\theta)\ge\mathcal M_*(\theta_*)$ for $\theta\ge\theta_*$, and \eqref{cert:margin} completes the proof. \end{proof} \begin{proof}[Proof of Theorem~\ref{thm:main}] For each $n\ge49$, let $K^{(n)}/B$ be a Galois extension with $[K^{(n)}:B]=2^n$ as in Theorem~\ref{tw:field-family}, let $\iota_1$ be a complex conjugation at a place of $K^{(n)}$ above the real place $v_1$ of $B$, put $F^{(n)}=(K^{(n)})^{\langle\iota_1\rangle}$ and $d_n=[F^{(n)}:\Q]$, and let $\theta_n$ be the value of $\theta$ for $F^{(n)}$. By Theorem~\ref{an:ceiling}, $d_n^{-1}\log L_{F^{(n)}}(1)0$ for $i0$ for all $n$. Corollary~\ref{fw:uniform-types} therefore gives finite sets $U_n\subset\R^2$ with $|U_n|\to\infty$ and $u(U_n)/|U_n|^{1+\delta}\to\infty$, where $1+\delta=1.04273$; these are the sets of Theorem~\ref{thm:main}, indexed by $n$. In particular $u(U_n)\ge|U_n|^{1.04273}$ for all large $n$, which gives the last assertion of Theorem~\ref{thm:main} at the cardinalities $|U_n|$. \end{proof} \subsection{Remarks} \begin{remark}\label{cert:signature-remark} The proof uses the bound $\theta\ge\theta_*$ only through Lemma~\ref{cert:lower-margin}, that is, through the positivity of $\mathcal M_*$ on $[\theta_*,1/2)$. The program \texttt{geom241.py} also encloses the value at $\theta=0$ and the zero $\theta_0$ of the affine function $\mathcal M_*$: \[ \mathcal M_*(0)<-0.316,\qquad \theta_0<0.49971293 . \] So $\mathcal M_*(\theta)>0$ for $\theta\ge0.49971293$; these enclosures are not used in the proof. By Theorem~\ref{tw:field-family}, $b/d=1/(2N_\iota)$, where $N_\iota$ is the number of conjugates of $\iota_1$ in $\Gal(K/B)$, so that $\theta=1/2-1/(4N_\iota)$. Hence $N_\iota\ge871$ would suffice, whereas that theorem gives $N_\iota\ge2^{15}$. On the other hand, at $\theta=0$ the margin $\mathcal M_{\mathcal R}(0)$ does not involve $J_\C$ and equals $\mathcal M_*(0)<0$: with these profiles, mixed signature is essential. \end{remark} \begin{remark}\label{cert:observations} The following numerical observations are not used in the proof. For the unweighted shell profiles with the same exponents $k_r$, the sum in Proposition~\ref{cert:enclosures}(e) is replaced by $J-\delta H<0.7165268434$. The proof of Proposition~\ref{fw:transfer} needs only some $\eta>0$ with $4\eta$ below the lower bound of the margins, where $\eta$ is the auxiliary parameter of that proof; for instance, $\eta=10^{-9}$ gives \begin{equation}\label{cert:final-margin} \mathcal M_*(\theta_*)-4\eta>0.000176726033534598>0 . \end{equation} The value $\mathcal M_*(\theta_*)$ is small compared with the individual terms in Table~\ref{cert:tab-budget}. Since $\mathcal M_*$ decreases with slope $1$ in $C$ and with slope $1/2-\delta$ in $\ell$, it would be absorbed by an increase of $1.77\cdot10^{-4}$ in $C$, or of $3.87\cdot10^{-4}$ in $\ell$. For $\delta=0.0428$, with the same pair profile $g_\C$ (the same $s$, $a$ and coefficients \eqref{pf:bernstein-witness}), the Gaussian of Lemma~\ref{pf:gaussian} for this $\delta$, the same $C$, and exponents and weights produced by the optimization program \texttt{shells241.py} for this $\delta$ (its exponents $k_r$ and numbers $n_r$ agree with Table~\ref{cert:tab-local}), the same evaluation gives $\mathcal M_*(\theta_*)<-0.0015$. \end{remark} \begin{remark}\label{cert:consistency} The weights $w_{r,i}$ are nonincreasing in $i$; this is an exact rational comparison. So $g_r$ is the nonnegative combination $\sum_{i,j}(w_{r,i}-w_{r,i+1})(w_{r,j}-w_{r,j+1})\mathbf 1_{\mathcal B_i\times\varpi^{-k_r}\mathcal B_j}$ of indicators of products of balls. As a consistency check, the rational number $Z(g_r)Q_r^{-k_r}$ is computed twice for each $r$: from \eqref{fw:product-shells}, and from this combination and \eqref{fw:rectangle-gram}. The two exact values agree. \end{remark} \subsection{Computations}\label{cert:computations} The program \texttt{geom241.py} evaluates the formulas of Sections~\ref{fw:section} and~\ref{pf:section} for the data of Definition~\ref{cert:data}: the local functionals \eqref{cert:finite-factor} from the exponents and weights of Tables~\ref{cert:tab-local} and~\ref{cert:tab-weights}, and the quantities of Propositions~\ref{pf:mass} and~\ref{pf:overlap}, of Lemma~\ref{pf:tube} and of Corollary~\ref{pf:functional} from the profile data. For these it uses, unchanged, three modules of an earlier supplementary archive of the author (Subsection~\ref{intro:code}), which implement exactly these formulas. In Proposition~\ref{pf:mass} it uses the symmetry of \eqref{pf:bernstein-witness}: each square off the diagonal is evaluated once and counted twice. The program checks the hypotheses listed in Proposition~\ref{cert:exact} (for (c), on the squares on or above the diagonal; the others follow by the symmetry of \eqref{pf:bernstein-witness}, since the restriction of $P$ to a reflected square has the transposed Bernstein coefficients), the conditions $a_\R\le1$, $\sigma_\C>1$, $\log K_\C<2\log2$ and $\mu>10$ (which implies \eqref{geo:poisson-condition} for $M=2$ and $\sigma=1$, since $\log50+2<10$), and the positivity of the slope in Proposition~\ref{cert:enclosures}(h); it also makes the consistency check of Remark~\ref{cert:consistency}. It then prints the enclosures stated in this section. The exponents and weights were produced by the separate optimization program \texttt{shells241.py}; they enter the proof only as the exact rational data of Table~\ref{cert:tab-weights}. Rational quantities are computed exactly: $q_0$, the restrictions of $P$ to the squares and their Bernstein coefficients, the coefficients of the powers $W_{\mathcal Q}^k$ and the binomial coefficients in Proposition~\ref{pf:mass}, $G_n$ and $\widetilde G_n$, and the shell sums $S_r$ and $R_r$; the polynomial arithmetic is done with integers after clearing denominators. The remaining quantities involve the logarithm, the exponential (also through the real powers $x^y=\exp(y\log x)$ in \eqref{pf:subrectangle-moments}, in $c_{\mathcal Q}^p$ and in $w_{r,i}^p$), the square root in $\sigma_\C$, the constant $\pi$, and the digamma function, which enters only through Lemma~\ref{pf:digamma}; that lemma is applied with exact rational arguments and Bernoulli numbers, $L=64$ and $k_*=16$, without rounding the rational parameters of the profile. These quantities are evaluated in the interval arithmetic of mpmath \cite{Mpmath}, which computes each arithmetic operation and each of these elementary functions with the lower endpoint rounded down and the upper endpoint rounded up, at a working precision of $80$ decimal digits; exact rational algebra followed by interval evaluation keeps the bound of Proposition~\ref{pf:mass} valid despite the cancellation in \eqref{pf:subrectangle-moments}. Hence every computed interval contains the exact value, and the decimals stated above, which are coarser than the working precision by more than fifty orders of magnitude, are rigorous bounds. \bibliographystyle{amsplain} \providecommand{\bysame}{\leavevmode\hbox to3em{\hrulefill}\thinspace} \providecommand{\MR}{\relax\ifhmode\unskip\space\fi MR } % \MRhref is called by the amsart/book/proc definition of \MR. \providecommand{\MRhref}[2]{% \href{http://www.ams.org/mathscinet-getitem?mr=#1}{#2} } \providecommand{\href}[2]{#2} \begin{thebibliography}{10} \bibitem{Alon2026} Noga Alon, Thomas~F. Bloom, W.~T. Gowers, Daniel Litt, Will Sawin, Arul Shankar, Jacob Tsimerman, Victor Wang, and Melanie~Matchett Wood, \emph{Remarks on the disproof of the unit distance conjecture}, 2026, arXiv:2605.20695v1; \url{https://arxiv.org/abs/2605.20695v1}. \bibitem{Lean4_2021} Leonardo de~Moura and Sebastian Ullrich, \emph{The {Lean 4} theorem prover and programming language}, Automated Deduction---CADE 28 (Cham) (Andr{\'e} Platzer and Geoff Sutcliffe, eds.), Lecture Notes in Computer Science, vol. 12699, Springer, 2021, pp.~625--635. \bibitem{Dokchitser2004} Tim Dokchitser, \emph{Computing special values of motivic {$L$}-functions}, Experimental Mathematics \textbf{13} (2004), no.~2, 137--149. \bibitem{Emmerich2026} Michael T.~M. Emmerich, \emph{Optimizing explicit unit-distance lower-bound certificates}, 2026, arXiv:2606.03419v5, June 9, 2026 (v1 June 2, 2026); certificate with $\delta=0.015263\ldots$; \url{https://arxiv.org/abs/2606.03419v5}. \bibitem{EmmerichCordella2026} Michael T.~M. Emmerich and Francesco Cordella, \emph{Optimized certificate for the unit distance problem with extended prime number range}, Zenodo record, June 5, 2026, Certificate and verification code. \url{https://doi.org/10.5281/zenodo.20551478}. \bibitem{Erdos1946} Paul Erd{\H{o}}s, \emph{On sets of distances of $n$ points}, The American Mathematical Monthly \textbf{53} (1946), no.~5, 248--250, \url{https://doi.org/10.1080/00029890.1946.11991674}. \bibitem{Erdos1994} \bysame, \emph{Some problems in number theory, combinatorics and combinatorial geometry}, Mathematica Pannonica \textbf{5} (1994), no.~2, 261--269. \bibitem{Ershov2012} Mikhail Ershov, \emph{{Golod--Shafarevich} groups: a survey}, International Journal of Algebra and Computation \textbf{22} (2012), no.~5, 1230001, 68 pages. \bibitem{Fox1953} Ralph~H. Fox, \emph{Free differential calculus. {I}: Derivation in the free group ring}, Annals of Mathematics \textbf{57} (1953), no.~3, 547--560. \bibitem{GolodShafarevich} E.~S. Golod and I.~R. Shafarevich, \emph{On the class field tower}, Izv. Akad. Nauk SSSR Ser. Mat. \textbf{28} (1964), no.~2, 261--272 (Russian), English translation: Amer. Math. Soc. Transl. (2) \textbf{48} (1965), 91--102; \url{https://www.mathnet.ru/eng/im2955}. \bibitem{HajirMaireRamakrishna2021Cutting} Farshid Hajir, Christian Maire, and Ravi Ramakrishna, \emph{Cutting towers of number fields}, Annales math{\'e}matiques du Qu{\'e}bec \textbf{45} (2021), no.~2, 321--345. \bibitem{Hecke1920} Erich Hecke, \emph{Eine neue {Art} von {Zetafunktionen} und ihre {Beziehungen} zur {Verteilung} der {Primzahlen}. {Zweite Mitteilung}}, Mathematische Zeitschrift \textbf{6} (1920), 11--51, \url{https://doi.org/10.1007/BF01202991}. \bibitem{Jennings} S.~A. Jennings, \emph{The structure of the group ring of a {$p$}-group over a modular field}, Transactions of the American Mathematical Society \textbf{50} (1941), no.~1, 175--185, \url{https://doi.org/10.2307/1989916}. \bibitem{Johansson2015Hurwitz} Fredrik Johansson, \emph{Rigorous high-precision computation of the {Hurwitz} zeta function and its derivatives}, Numerical Algorithms \textbf{69} (2015), no.~2, 253--270. \bibitem{Johansson2017} \bysame, \emph{{Arb}: Efficient arbitrary-precision midpoint-radius interval arithmetic}, IEEE Transactions on Computers \textbf{66} (2017), no.~8, 1281--1292, \url{https://doi.org/10.1109/TC.2017.2690633}. \bibitem{Louboutin2000} St{\'e}phane Louboutin, \emph{Explicit bounds for residues of {Dedekind} zeta functions, values of {$L$}-functions at {$s=1$}, and relative class numbers}, Journal of Number Theory \textbf{85} (2000), no.~2, 263--282, \url{https://doi.org/10.1006/jnth.2000.2545}. \bibitem{MilneEtale} James~S. Milne, \emph{Lectures on {\'e}tale cohomology}, 2013, Version 2.21, March 22, 2013; \url{https://www.jmilne.org/math/CourseNotes/LEC.pdf}. \bibitem{MilneANT} \bysame, \emph{Algebraic number theory}, 2020, Version 3.08, July 19, 2020; \url{https://www.jmilne.org/math/CourseNotes/ANT.pdf}. \bibitem{MilneCFT} \bysame, \emph{Class field theory}, 2020, Version 4.03, August 6, 2020; \url{https://www.jmilne.org/math/CourseNotes/CFT.pdf}. \bibitem{ErdosProblems90} {mlewko}, \emph{Post in the discussion thread of {Erd\H{o}s} {P}roblem \#90}, Erd\H{o}s Problems forum, May 21, 2026, Reports $\delta=0.031849981\ldots$ in Sawin's criterion, with finite data found by ChatGPT; accessed September 30, 2026; \url{https://www.erdosproblems.com/forum/thread/90}. \bibitem{NaslundMO2026} Eric Naslund, \emph{Answer to ``{W}hat is the unit distance exponent?''}, MathOverflow, answer 511576, 2026, Posted May 23, 2026, last edited June 9, 2026; accessed September 30, 2026; \url{https://mathoverflow.net/a/511576}. \bibitem{NaslundLean2026} \bysame, \emph{An exponent of 1.04273 for the unit distance problem: {L}ean formalization at exponent $1.0427$, conditional on one zeta inequality}, Palomar registry, PALOMAR-2026-10-01-000018, version~1, 2026, \url{https://palomar-registry.org/entry?id=PALOMAR-2026-10-01-000018&version=1}; Lean~4 source at \url{https://github.com/enaslund/unit-distance-bound-0.0427/tree/e0ac836176016dbefae28646bc91fc3a5aca30e8/lean}. \bibitem{Naslund0358} \bysame, \emph{An improved explicit lower bound for the unit distance problem}, Unpublished manuscript, 2026, Records exponent $1.0358324$. \url{https://github.com/enaslund/unit-distance-bound-0.0427/tree/8ab94da4ba4faf6f6d8fe43c298f04c3b284e796/papers/0.0358324}. \bibitem{NaslundCode2026} \bysame, \emph{Supplementary programs and data for ``{A}n exponent of 1.04273 for the unit distance problem''}, GitHub repository, directory \texttt{papers/0.04273/certificates}, 2026, \url{https://github.com/enaslund/unit-distance-bound-0.0427}. \bibitem{DLMF} {National Institute of Standards and Technology}, \emph{{NIST Digital Library of Mathematical Functions}}, 2026, Release 1.2.8, September 15, 2026; accessed September 27, 2026; \url{https://dlmf.nist.gov/}. \bibitem{NeukirchSchmidtWingberg2008} J{\"u}rgen Neukirch, Alexander Schmidt, and Kay Wingberg, \emph{Cohomology of number fields}, 2 ed., Grundlehren der mathematischen Wissenschaften, vol. 323, Springer, Berlin, Heidelberg, 2008. \bibitem{OpenAI2026} {OpenAI}, \emph{Planar point sets with many unit distances}, Research manuscript, May 20, 2026, \url{https://cdn.openai.com/pdf/74c24085-19b0-4534-9c90-465b8e29ad73/unit-distance-proof.pdf}. \bibitem{TaoConstants84a} \emph{Erd{\H{o}}s unit distance exponent}, Entry 84a of \emph{Optimization Constants in Mathematics}, a crowdsourced repository started by Terence Tao, 2026, Accessed September 30, 2026; \url{https://teorth.github.io/optimizationproblems/constants/84a.html}. \bibitem{Poonen} Bjorn Poonen, \emph{Rational points on varieties}, Graduate Studies in Mathematics, vol. 186, American Mathematical Society, Providence, RI, 2017, \url{https://math.mit.edu/~poonen/papers/Qpoints.pdf}. \bibitem{Quadrelli2021} Claudio Quadrelli, \emph{Pro-{$p$} groups with few relations and universal {Koszulity}}, Mathematica Scandinavica \textbf{127} (2021), no.~1, 28--42, \url{https://doi.org/10.7146/math.scand.a-123644}. \bibitem{Quillen} D.~G. Quillen, \emph{On the associated graded ring of a group ring}, Journal of Algebra \textbf{10} (1968), no.~4, 411--418, \url{https://doi.org/10.1016/0021-8693(68)90069-0}. \bibitem{Sawin2026} Will Sawin, \emph{An explicit lower bound for the unit distance problem}, 2026, arXiv:2605.20579v1; \url{https://arxiv.org/abs/2605.20579v1}. \bibitem{SerreCourse} Jean-Pierre Serre, \emph{A course in arithmetic}, Graduate Texts in Mathematics, vol.~7, Springer, New York, 1973. \bibitem{SpencerSzemerediTrotter1984} Joel Spencer, Endre Szemer{\'e}di, and William~T. Trotter, Jr., \emph{Unit distances in the {Euclidean} plane}, Graph Theory and Combinatorics (B{\'e}la Bollob{\'a}s, ed.), Academic Press, London, 1984, \url{https://trotter.math.gatech.edu/papers/44.pdf}, pp.~293--303. \bibitem{SpiderduckpigMO2026} {spiderduckpig}, \emph{Answer to ``{W}hat is the unit distance exponent?''}, MathOverflow, answer 511531, 2026, Posted May 21, 2026, with $1+\delta\ge1.03188306553964\ldots$; comment of May 22, 2026 proposing a weighted count in the Zassenhaus filtration in Sawin's Golod--Shafarevich step, with $\delta\ge0.0333487156372102$; last edited June 9, 2026; accessed September 30, 2026; \url{https://mathoverflow.net/a/511531}. \bibitem{Szekely1997} L{\'a}szl{\'o}~A. Sz{\'e}kely, \emph{Crossing numbers and hard {Erd{\H{o}}s} problems in discrete geometry}, Combinatorics, Probability and Computing \textbf{6} (1997), no.~3, 353--358, \url{https://doi.org/10.1017/S0963548397002976}. \bibitem{Tate1967} John~T. Tate, \emph{Fourier analysis in number fields and {Hecke}'s zeta-functions}, Algebraic Number Theory (J.~W.~S. Cassels and A.~Fr{\"o}hlich, eds.), Academic Press, London, 1967, Originally a Princeton University doctoral thesis, 1950; \url{https://www.lms.ac.uk/publications/algebraic-number-theory}, pp.~305--347. \bibitem{FLINT} {The FLINT team}, \emph{{FLINT}: {F}ast {L}ibrary for {N}umber {T}heory}, 2026, Version 3.6.0, used through python-flint 0.9.0; \url{https://flintlib.org}. \bibitem{Mathlib2020} {The mathlib Community}, \emph{The {Lean} mathematical library}, Proceedings of the 9th ACM SIGPLAN International Conference on Certified Programs and Proofs, ACM, 2020, pp.~367--381. \bibitem{Mpmath} {The mpmath development team}, \emph{{mpmath}: A {Python} library for arbitrary-precision floating-point arithmetic}, 2023, Version 1.3.0; \url{https://github.com/mpmath/mpmath/releases/tag/1.3.0}. \bibitem{PARI} {The PARI Group}, \emph{{PARI/GP}}, Universit{\'e} de Bordeaux, 2025, Version 2.17.2; \url{https://pari.math.u-bordeaux.fr/}. \bibitem{TsfasmanVladut2002} M.~A. Tsfasman and S.~G. Vl{\u a}du{\c t}, \emph{Infinite global fields and the generalized {Brauer--Siegel} theorem}, Moscow Mathematical Journal \textbf{2} (2002), no.~2, 329--402, \url{https://doi.org/10.17323/1609-4514-2002-2-2-329-402}. \end{thebibliography} \end{document}