\documentclass[11pt,reqno]{amsart} \usepackage{jaswal} \Urlmuskip=0mu\relax \usepackage{tikz} \newcommand{\C}{\mathbb{C}} \newcommand{\Q}{\mathbb{Q}} \newcommand{\R}{\mathbb{R}} \newcommand{\Z}{\mathbb{Z}} \newcommand{\code}[1]{\texttt{#1}} \DeclareUrlCommand\lean{\def\UrlFont{\ttfamily\upshape}\def\UrlBreaks{\do\/\do\.\do\_}} \DeclareUrlCommand\file{\def\UrlFont{\ttfamily\upshape}\def\UrlBreaks{\do\/\do\.\do\_}} \makeatletter \def\LT@makecaption#1#2#3{% \LT@mcol\LT@cols c{\hbox to\z@{\hss\parbox[t]{\dimexpr\textwidth-6pc\relax}{% \normalfont\normalsize \sbox\@tempboxa{#1{{\@captionheadfont #2}.\enspace}#3}% \ifdim\wd\@tempboxa>\hsize #1{{\@captionheadfont #2}.\enspace}#3% \else \hbox to\hsize{\hfil\box\@tempboxa\hfil}% \fi \endgraf\vskip 6pt}% \hss}}} \makeatother \newcommand{\locsep}{\par} \newcommand{\locseplast}{\par\nobreak} \newcommand{\locs}[1]{\par\nopagebreak\vspace{2pt}{\footnotesize\raggedright\leftskip=2.6em\parindent=-1.6em\parskip=0pt\relax #1\par}\smallskip} \setlength{\emergencystretch}{3em} \newcommand{\ragged}{\rightskip=0pt plus 4em\relax} \newcommand{\tnode}[2]{\par\noindent\hangindent=\dimexpr1.2em*#1+2.6em\relax\hangafter=1\hspace*{\dimexpr1.2em*#1+1em\relax}#2} \newcommand{\Star}{\ensuremath{(\star)}} \DeclareMathOperator{\Real}{Re} \DeclareMathOperator{\Ima}{Im} \DeclareMathOperator{\ASK}{ASK} \DeclareMathOperator{\MAIN}{MAIN} \DeclareMathOperator{\Gal}{Gal} \DeclareMathOperator{\Sha}{Sha} \makeatletter \def\@settitle{\begin{center}\baselineskip18\p@\relax\normalfont\Large\bfseries\boldmath\@title\end{center}} \def\@setauthors{\begin{center}\normalfont\normalsize\authors\\[3pt]\small Toronto, Canada\quad\texttt{manvirjaswal@icloud.com}\quad\url{https://manvirjaswal.github.io}\\[3pt]\@date\end{center}} \def\@adminfootnotes{} \def\@secnumfont{\bfseries} \def\section{\@startsection{section}{1}\z@{.7\linespacing\@plus\linespacing}{.5\linespacing}{\normalfont\bfseries\boldmath}} \makeatother \pagestyle{plain} \begin{document} \title[The Riemann hypothesis {[after L.~Alp\"oge]}]{The Riemann hypothesis for Dirichlet $L$-functions\\[2pt]{\normalsize\mdseries [after Levent Alp\"oge]}} \author{Manvir Jaswal} \date{Version 1, November 21, 2026} \hypersetup{pdftitle={The Riemann hypothesis for Dirichlet L-functions [after Levent Alpöge]},pdfauthor={Manvir Jaswal}} \maketitle \contents \section{Introduction}\label{sec:intro} L.~Alp\"oge announced on 20 November 2026 that he believes he has found a proof of the Riemann hypothesis, working with Claude, and presented it only as a formalization in Lean~4 \cite[pp.~1--2]{AlpNote}, \cite{AlpArchive}. This guide follows the chain of declarations from the five theorems of the archive's file \file{Main.lean} back to its inputs and says what each step means in classical terms, with files and lines. In ordinary notation the five theorems say the following. \begin{theorem}[Alp\"oge]\label{thm:main} Let $\chi$ be a Dirichlet character modulo $N\ge1$ and $L(s,\chi)$ its $L$-function, continued to~$\C$. \begin{enumerate} \item If $\zeta(s)=0$, $s\ne1$ and $s\notin\{-2,-4,-6,\dots\}$, then $\Real s=\tfrac12$. \item If $\Real s>\tfrac12$, then $L(s,\chi)\ne0$, unless $\chi$ is the trivial character modulo $N$ and $s=1$. \item If $0<\Real s<1$ and $L(s,\chi)=0$, then $\Real s=\tfrac12$. \end{enumerate} \end{theorem} In \file{Main.lean}, part~(a) is \lean{riemannHypothesis} (line~4), and \lean{riemannZeta_eq_zero_re} (line~6) states it again in the words of \lean{RiemannZetaZerosOnLine} (\file{Statements.lean:3}). Part~(b) is \lean{LFunction_ne_zero_of_half_lt_re} (line~8). Part~(c) is \lean{LFunction_eq_zero_re} (line~10) and, for primitive $\chi$, \lean{LFunction_eq_zero_re_of_isPrimitive} (line~12). The five statements are Mathlib's \lean{RiemannHypothesis} and four definitions in \file{Statements.lean}, a file that imports only Mathlib. As in Mathlib, $L(s,\chi)$ for imprimitive $\chi$ is the continuation of $\sum\chi(n)n^{-s}$ with $\chi(n)=0$ for $(n,N)>1$, so its zeros include those of the Euler factors $1-\chi^*(p)p^{-s}$ of the primitive character $\chi^*$ inducing~$\chi$, at the primes dividing the modulus but not the conductor; these zeros lie on $\Real s=0$. Mathlib's $\zeta$ and the $L$-functions of trivial characters are meromorphically continued and given a value at their pole $s=1$, which is why $s=1$ is set aside in (a) and~(b). Part~(a) is Riemann's hypothesis for~$\zeta$ \cite{Riemann}. Its analogue for curves over finite fields is Hasse's theorem for elliptic curves \cite{Hasse36} and Weil's for all curves \cite{Weil48}. Stepanov's method \cite{Stepanov}, extended to all curves by W.~Schmidt \cite{Schmidt} and by Bombieri \cite[\S II, Thm~1, and \S III]{Bombieri}, proves it with an auxiliary function that vanishes to high order at the rational points outside a possible pole. Deligne proved the analogue for all smooth projective varieties \cite[Thm~(1.6)]{Deligne}. Alp\"oge describes his proof as a descendant of Stepanov's argument in which multiplicity, hard to find over the integers, comes from a chain of theta lifts, each lift mimicking multiplicity \cite[p.~1]{AlpNote}. The comparison behind it is Chebyshev's, with a period in place of a ratio of factorials. A nonzero integral period has divisibility at most its size, the logarithm of its norm, by the product formula. In the simplest model the period is $(D^a\vartheta)(0)/\Omega^a$, for a theta function $\vartheta$ and the invariant derivation $D$ on a CM curve. Its divisibility at the primes inert in $\Q(\sqrt{-d})$ is, up to $O(a)$, the sum over prime powers $n\le a$ prime to~$d$ of $\Lambda(n)(1-\chi_{-d}(n))\,a/n$, its size is about $\log(a!\,\Omega^{-a})$, and the two have the same main term $a\log a$. So what remains of ``divisibility at most size'' is an upper bound for a centered ramp sum of $\Lambda(n)$ by the same sum twisted by $\chi_{-d}$. In a chain of $m$ lifts the gains in divisibility add up, and at the scale~$x$ of the ramp sums the error of the comparison is smallest at $m\approx\sqrt x$, one of the two places from which the exponent~$\frac12$ comes (\cref{rem:half}). The subsection ``The idea'' below works the model out. The proof runs as follows, and \cref{fig:chain} draws the chain. \begin{enumerate}[(1)]\ragged \item Part~(b) for primitive characters implies the rest of \cref{thm:main}, by the Euler product and the functional equation (\cref{lem:strip,lem:imprimitive}). \item For a primitive $\psi$ and the primitive characters $\xi$ that induce the powers of~$\psi$, Alp\"oge uses the centered ramp sums $G(\xi,x)=\sum_{n\le x}\Lambda(n)\xi(n)(x/n-1)+x\frac{L'}L(1,\xi)$ for $\xi\ne1$ and $G(x)=\sum_{n\le x}\Lambda(n)(x/n-1)-x\log x+(1+\gamma)x$, and a positive combination $G_\psi(x)$ of their real parts. A window form of Landau's theorem excludes zeros of the $L(s,\xi)$ with real part above~$\theta$, for $0<\theta<1$, when $G_\psi(x)\le x^\theta$ for all $x$ in windows $[Y,Y^\lambda]$, $\lambda>1$, with $Y$ arbitrarily large (\cref{thm:landau}). \item For every small $\varepsilon>0$, Alp\"oge bounds $G_\psi(x)$ by $x^{1/2+O(\varepsilon)}$ for all $x$ in windows $[Y,Y^{3/2}]$ with $Y$ arbitrarily large. The bound comes from an inequality \Star{} comparing divisibility with size on certain abelian CM fields, the host fields (\cref{sec:hosts,prop:core}), and \cref{rem:half} says where the exponent $\frac12$ enters. \item The inequality \Star{} is proved along a chain of $m\approx\sqrt x$ lifts, from divisibility at the inert primes, size, a slope inequality, near-orthogonality and a covolume bound (\cref{sec:star}). \item The host fields come from a bound for Tate--Shafarevich groups of Tate twists, \lean{VWin}, which also enters \Star{} through the loss of integrality at the twisted places. \Cref{sec:vwin} states the bound and sketches how the archive obtains it. \end{enumerate} \begin{figure}[ht] \centering\small \begin{tikzpicture}[x=1cm,y=1cm,box/.style={draw,rounded corners=2pt,align=center,inner sep=3pt},lab/.style={font=\scriptsize,midway,right}] \node[box] (M) at (0,8.6) {\cref{thm:main}, (a)--(c)\\[-1pt]{\scriptsize\file{Main.lean}}}; \node[box] (B) at (0,7.3) {$L(s,\chi)\ne0$ for $\Real s>\frac12$ if $\chi\ne1$ or $s\ne1$: part (b)\\[-1pt]{\scriptsize\lean{everyCharacter}}}; \node[box] (P) at (0,6.0) {$L(s,\psi)\ne0$ for $\Real s>\frac12$, every primitive $\psi$\\[-1pt]{\scriptsize\lean{zeroFreeAll_of_topG}, \cref{thm:topg}}}; \node[box] (W) at (0,4.7) {$G_\psi(x)\le x^{1/2+O(\varepsilon)}$ for $x$ in windows $[Y,Y^{3/2}]$\\[-1pt]{\scriptsize\cref{prop:core}\textsuperscript{\textdagger}}}; \node[box] (S) at (0,3.4) {\Star: divisibility at most size, along a chain of $m\approx\sqrt x$ lifts\\[-1pt]{\scriptsize\lean{ChainTied}, \cref{sec:star}}}; \node[box] (L) at (-3.4,1.55) {divisibility at inert primes, size, slope\\ inequality, near-orthogonality, covolume bound\\[-1pt]{\scriptsize\lean{HypsAlt}, \cref{sec:star}\textsuperscript{\textdagger}}}; \node[box] (V) at (3.6,1.55) {host fields on windows; a bound for\\ Tate--Shafarevich groups of Tate twists\\[-1pt]{\scriptsize\lean{VWin}, \cref{sec:vwin}\textsuperscript{\textdagger}}}; \node[box] (G) at (0,-0.35) {(i) Galois representations, weak form; (ii) local parameters; (iii) follows from (i)\\[-1pt]{\scriptsize\cref{thm:three}, \lean{Row0b}, \lean{RecTExists}}}; \draw[->] (B) -- node[lab]{\cref{lem:strip}} (M); \draw[->] (P) -- node[lab]{\cref{lem:imprimitive}} (B); \draw[->] (W) -- node[lab]{Landau on windows, \cref{thm:landau}} (P); \draw[->] (S) -- node[font=\scriptsize,midway,right=5pt]{thin, thick and size estimates, \cref{prop:core}} (W); \draw[->] (S) -- node[font=\scriptsize,midway,left=5pt]{hosts from \lean{VWin}; \lean{TopG} and its class} (W); \draw[->] (L) -- (S); \draw[->] (V) -- node[font=\scriptsize,midway,above right,inner sep=1.5pt]{loss at the twisted places} (S); \draw[->] (G) -- node[font=\scriptsize,midway,below left,inner sep=1.5pt]{bundles, \cref{sec:top}} (L); \draw[->] (G) -- node[font=\scriptsize,midway,below right,inner sep=1.5pt]{(iii)} (V); \end{tikzpicture} \caption{The chain from the inputs up to \texttt{Main.lean}, as in \cref{sec:top,sec:analytic,sec:star,sec:vwin}; an arrow points from a statement to what it is used to prove. A dagger marks a box that holds statements for which we know of no reference; \cref{tab:new} lists these statements and \cref{tab:known} the classical inputs.}\label{fig:chain} \end{figure} A declaration is \emph{on the route}, or \emph{on the chain}, if the proof of one of the five theorems of \file{Main.lean} refers to it, directly or through the declarations it refers to. Three points should be kept in mind from the start. First, for two of the twenty steps listed in Alp\"oge's note \cite[pp.~3--4]{AlpNote}, steps~1 and~4, we found no counterpart on the chain: no Stepanov determinant, and no determinant giving the positivity of step~4 (\cref{sec:steps}). Second, the periods on the route are defined from the kernels. The kernels follow Alp\"oge's plan and are built from derivatives at~$0$ of theta functions (\cref{sec:datum}). Step~12 of the note appears as an identity between the periods and inner products with Poincar\'e series (\lean{MIid}, \lean{BesselId}), proved by passing linear relations through Bessel periods of the row forms (\lean{JointCondition}, \cref{sec:onelift}). Third, two inputs are weaker in the archive than the published theorems, namely the Galois representations \lean{Row0b} and the local parameters of \lean{RecTExists}. The laws \lean{IsRecT} required of the second are satisfied by a map that, at each finite place $w$ of a number field~$E$, sends every supercuspidal representation of $\mathrm{GL}_n(E_w)$ outside an explicit class to a real unramified twist of a single parameter $J_{w,n}$. This map is not the local Langlands correspondence (\cref{sec:top,sec:weak}). The inputs of the proof were built by many people. \begin{itemize}\ragged \item Analytic number theory. Landau's theorem on Dirichlet series with coefficients of one sign \cite[\S2]{Landau}, and Kronecker's limit formula summed over ideal classes, as by Chowla and Selberg \cite{ChowlaSelberg}. \item Automorphic forms and the local theory. The theta correspondence and the Siegel--Weil formula \cite{Weil65,Ichino,Yamana}, with Langlands's continuation of Eisenstein series \cite{Langlands}; a unitary mass formula in the tradition of Siegel's for quadratic forms \cite{Siegel}; local class field theory, Zelevinsky's classification for $p$-adic $\mathrm{GL}_n$ \cite[Thm~6.1]{Zelevinsky} and the depth-zero part of the local Langlands correspondence for $\mathrm{GL}_n$ over~$\Q_p$ \cite{HarrisTaylor,Henniart}; Galois representations attached to polarized regular algebraic cuspidal automorphic representations of $\mathrm{GL}_n$ over CM fields \cite[Thm~2.1.1]{BLGGT}, in the conjugate self-dual case \cite{HarrisTaylor}, \cite[Thm~1.2]{Shin}, \cite[Thm~3.2.3]{ChenevierHarris}, \cite[Thm~1.1]{Caraiani}. \item Arakelov geometry. Bost's slope inequality and arithmetic intersection theory \cite{Bost,GilletSoule,Moriwaki,Yuan}, with Hironaka's resolution of singularities in characteristic zero \cite{Hironaka}. \item Iwasawa theory and Galois cohomology. Kato's Euler system \cite[\S\S8--9]{Kato} and its Selmer consequence \cite[Thm~14.2(1)]{Kato}, the theorem of Ferrero and Washington \cite{FW} and Poitou--Tate duality \cite[(8.6.10)]{NSW}. \end{itemize} \Cref{tab:new} lists the statements of the chain for which we know of no reference, and \cref{sec:known} compares the classical inputs with their archive forms. For the idea, read this introduction, \cref{sec:ramp,sec:landau,sec:star-def,sec:balance}, \cref{rem:half} and \cref{tab:new}. For the inputs, read \cref{sec:star,sec:vwin,sec:known}. A Lean user who follows the chain from \file{Main.lean} can start with \cref{sec:top}, \cref{fig:chain} and the call tree of \cref{sec:reduction}, and finds the file and line of each step in the section that treats it. \Cref{sec:steps} sets Alp\"oge's twenty steps against the chain, and \cref{sec:else} records other results in the archive. \subsection*{The archive and how it is read here} The archive we read is the one posted at \url{https://alpo.ge} on 20 November 2026, in fourteen parts. Joined, the parts form \file{lean.tar.gz}, with SHA-256 \texttt{9631b4f020199e580ba86178cb56c6e9}\allowbreak\texttt{b09f690c2369a80dc4334aa93f49c7fc}. It uses Lean~4.33.1 and Mathlib at commit \texttt{0df444a3}, tagged v4.33.1 \cite{AlpArchive,Mathlib}. We give statements and the order of the argument, with sketches, and complete proofs only for the elementary links. For the weak Galois representations through theta lifts, the covolume bound and the near-orthogonality bound, Theorem~S (\cref{sec:definite}), we give the statements and at most a sketch. We also include some parts of the archive off the chain (\cref{sec:vandiver,sec:steps,sec:else}). We read the statements of the declarations cited and selected proof terms, not the proofs below them, and we have not compiled the archive. Where we say how a proof goes, we name the lemmas its proof term applies, and the steps between them are our reconstruction. Corrections and comments are welcome. Constants and quantifiers are simplified throughout. ``For large $x$'' means beyond a threshold depending on the family, on~$\varepsilon$ and on the character. Families indexed by the scale and by their members are described one member at a time. A hypothesis bundle is described by what its fields assert, and a field with no short classical statement is named. But the files and lines cited lead to the archive's exact statements. A.~Sutherland has compiled the archive and reports that the five theorems of \file{Main.lean} depend only on the axioms \lean{propext}, \lean{Classical.choice} and \lean{Quot.sound}, and that every module passes Lean's kernel and lean4lean. The independent kernels nanoda and con-leche accept the exported proof, and con-leche's comparison mode matches the five theorems with the challenge statements, which import only Mathlib \cite{Sutherland}. Other independent checkers accept the same export \cite{ZulipRH}. The source contains no \lean{sorry}, no \lean{native_decide} and no declared axiom \cite{Sutherland,MirrorIssue2}. Some declarations are generated by copying others with \lean{addDecl} (\file{.../CutCptCopy.lean:78}, run at \file{.../CutMainCpt.lean:945}), and two tactics, \lean{kexactB} and \lean{kcopy}, assign proof terms without the elaborator's check. But in Sutherland's replay Lean's kernel checks these declarations like the rest. \subsection*{Notation} $\Lambda$ is the von Mangoldt function and $\gamma$ Euler's constant. In ramp sums $\psi$ and $\xi$ are primitive Dirichlet characters; for other characters the group or the modulus is given. For primitive $\psi$ we write $\langle\psi\rangle$ for the set of primitive characters that induce the powers of $\psi$, and $o=|\langle\psi\rangle|$ for the order of~$\psi$. For a prime $d\equiv3\pmod 4$, $\chi_{-d}$ is the character of $\Q(\sqrt{-d})$. The letter $x$ is a scale tending to infinity and $\varepsilon>0$ is small; $x^{O(\varepsilon)}$ stands for $x^{C\varepsilon}$ with $C$ fixed before $\varepsilon$ is chosen. In the final bound $C=9$ (\cref{sec:assembly}). \Cref{tab:names}, at the end, lists the names that \cref{sec:analytic,sec:star,sec:vwin} use far from their definitions. \subsection*{The idea} We model the note's steps~2 and~9 \cite[pp.~3--4]{AlpNote} by the following heuristic computation. \Cref{sec:analytic,sec:star} give the archive's exact form, with weights, truncations and errors. \emph{Divisibility.} Let $K=\Q(\sqrt{-d})$ with $d\equiv3\pmod4$ prime, $\chi=\chi_{-d}$, and $p$ a prime inert in~$K$, so that $\chi(p^i)=(-1)^i$. Legendre's formula $v_p(a!)=\sum_{i\ge1}\lfloor a/p^i\rfloor=(a-s_p(a))/(p-1)$, with $s_p(a)$ the sum of the digits of $a$ in base~$p$, gives \[ v_p(a!)-\frac a{p^2-1}=\sum_{i\ \mathrm{odd}}\frac a{p^i}-\frac{s_p(a)}{p-1}. \] Step~2 of the note states that $(D^a\vartheta)(0)/\Omega^a$, for a theta function $\vartheta$ and the invariant derivation $D$ on a CM curve, is divisible by $p$ to the order $a/p+a/p^3+\cdots-O(\log a)$, which is this quantity up to $O(\log a)$. For the Weierstrass function of one CM curve a bound with this exponent is a theorem of Katz \cite[\S IV]{Katz81}; \cref{sec:divis} places it. Weight $p$ by $2\log p$, the logarithm of the norm of $p\mathcal O_K$, and sum over the inert primes. As $1-\chi(p^i)$ is $2$ for odd $i$ and $0$ for even $i$, and $0$ at the split primes, \[ \sum_{p\ \mathrm{inert}}2\log p\sum_{i\ \mathrm{odd}}\frac a{p^i}=\sum_{(n,d)=1}\Lambda(n)\bigl(1-\chi(n)\bigr)\frac an . \] With a common cutoff $p^i=n\le B$ the two sides agree, and both diverge as $B$ increases. The archive adds the ramified prime~$d$, where $1-\chi=1$, and inserts a ramp, counting $(a/n-w)_+$ for a weight~$w$ in place of $a/n$ (\cref{sec:star-def}). With $w=1$ the divisibility is $\mathcal D(a)=\sum_{n\le a}\Lambda(n)(1-\chi(n))(a/n-1)$. \emph{Size.} By Chebyshev's identity and Stirling's formula, $\log a!=\sum_n\Lambda(n)\lfloor a/n\rfloor=a\log a-a+O(\log a)$, so the divisibility of $a!$ at all primes together is its size. We model the size of the period, the logarithm of its norm from~$K$, that is of its squared absolute value, by $\log(a!\,\Omega^{-a})$, for a period $\Omega$ of Chowla--Selberg shape. In the archive $\log\Omega=\gamma+\log(2\pi/\mathrm{rd}_E)-\frac{L'}L(1,\chi)$, with $\mathrm{rd}_E$ the root discriminant of the host field~$E$ below. The terms $a\log(\mathrm{rd}_E/2\pi)$ that this puts in the sizes of the $m$ lifts behind a word~$\omega$ (the index of a chain, \cref{sec:tied}) cancel together against $-a\log T_\omega$, for a prime $T_\omega$ attached to~$\omega$ with $\log T_\omega\ge m\log(\mathrm{rd}_E/2\pi)$ up to terms in the weights and an error $O(\log(a\,\mathrm{rd}_E))$. What remains is the size per lift $M(a)=a\bigl(\log a-1-\gamma+\frac{L'}L(1,\chi)\bigr)$ (\cref{sec:tied}). \emph{Comparison.} A nonzero integral period has divisibility at most its size, by the product formula, so $\mathcal D(a)\le M(a)$ up to errors. This is Chebyshev's argument, with a period in place of a ratio of factorials. Here, as the identity below shows, the two sides have the same main term. Since $\sum_{n\le a}\Lambda(n)(a/n-1)=a\log a-(1+\gamma)a+G(a)$, the terms $a\log a$ cancel, and by the definitions of the centered ramp sums \[ \mathcal D(a)-M(a)=G(a)-G(\chi,a). \] So ``divisibility at most size'' is a one-sided bound for a difference of centered ramp sums. For a host field, where $2$ and the auxiliary prime~$q$ of \cref{sec:hosts} split in~$K$, with the cutoff $N$ of $\ASK$ at least~$a$ and every weight set to~$1$, $\mathcal D(a)$ and $M(a)$ are the archive's $\ASK$ and $\MAIN$ (\cref{sec:star-def}). The model is a toy one, since in the archive's inequality~\Star{} the weights exceed~$1$ at the thick places (\cref{sec:star-def}). In the model, $\mathcal D(a)\le M(a)$ bounds $G(a)$ above by $G(\chi,a)$, so an upper bound for the twisted sum bounds $G(a)$ from above. The archive keeps its twisted sums small by its choice of $d$ and~$q$. A mean square over the moduli gives them square-root size on average, and by Markov's inequality most pairs are good (\cref{sec:supply}). \emph{From $\chi_{-d}$ to $\psi$.} In the archive the characters are those of $\Gal(E/\Q)$ for a CM field $E$, the host field, the compositum of $\Q(\zeta_{2^{e+2}})^+$, the field of~$\psi$, $\Q(\sqrt{-q})$ and~$K$, with $2^e$ tied to $x$ (\cref{sec:hosts}). The weights vary with the place, and $(a/n-w)_+=w\,(y/n-1)_+$ is a ramp at $y=a/w$. At the thick places the weights are near $b=\exp(x^\varepsilon)$, which with $a\approx xb$ puts the ramp at $y\approx x$. Orthogonality on $\Gal(E/\Q)$ makes $\ASK-\MAIN$ a weighted sum of $G(\theta,y)$ over the characters~$\theta$ (\cref{prop:balance}). After the thin places are estimated and the thick weights frozen, averaging over translates by the subgroup $A_0$ in the $2$-power direction leaves $4o$ of them: those of $\langle\psi\rangle$, which the shares of the shifts combine into $G_\psi$ up to an error, and $3o$ twists of them by $\chi_{-q}$, $\chi_{-d}$ and $\chi_{qd}=\chi_{-q}\chi_{-d}$, which are bounded separately (\cref{sec:core}). \emph{Why $m\approx\sqrt x$.} The gains of $m$ lifts add up, and the archive bounds the gain of the $m$ lifts together from below (\cref{sec:tied}). The comparison is then a sum of $m$ comparisons like the one above, with different weights. In \Star{}, with $TT$ the set of shifts~$t$ (\cref{sec:star-def}), the error has a part $|TT|\,a$ that does not grow with~$m$ and parts that do. Dividing by $m\,b\,|TT|$ leaves $x^{O(\varepsilon)}(x/m+m+\sqrt x)$ (\cref{sec:core}). This error is smallest at $m\approx\sqrt x$ (\cref{rem:half}), the note's chain of theta lifts of length $\sqrt X$ \cite[p.~1]{AlpNote}. The archive takes $x^{1/2-\varepsilon/16}\le m\le2x^{1/2-\varepsilon/16}$ (\lean{RankVal}, \cref{sec:datum}) and obtains $G_\psi(x)\le x^{1/2+O(\varepsilon)}$. \emph{In the archive} (\cref{sec:star}) the divisible object is not one number but a map $\mathrm{Bk}_\omega\colon N_\omega\to RR^\vee$ of Hermitian lattices whose entries are periods, one for each word~$\omega$, produced by a chain of $m$ steps that Alp\"oge describes as theta lifts between definite unitary groups \cite[p.~4]{AlpNote}. At inert $p\ge5$ the archive has integrality of corrected kernels (\lean{IntInert}) and, for every power series of the form this gives, a bound on its derivatives at~$0$ with an exponent of terms $a/p^e$, $e$ odd, with caps and some terms omitted, less a loss (\lean{kernel_bounds_generic}). Together they make the actual periods divisible by the lattice ideals that carry the gains (\cref{sec:divis}). A lower bound for the gains of the lifts together is proved separately, without the theta kernels. Divisibility raises the slopes of the $N_\omega$. A slope inequality for all the maps at once (\lean{SlopeIneq}) bounds the rank-weighted mean of their slopes, shifted by heights $h(\omega)$, by the slope of $RR^\vee$, which the size bound (\lean{DIV}, \eqref{eq:size}) compares with the size. A covolume bound and an approximate orthogonality of the columns control the losses. For each lift~$j$, \lean{LinkN} bounds the weighted loss of integrality at the twisted places by a dimension $d_{Z,0}(j)$ times the sum of an explicit term and a weighted sum of lengths of Tate--Shafarevich groups of Tate twists over the CM fields, when these are finite and the sum converges. The bound for that sum, \lean{VWin}, is proved from the weak Galois inputs, Kato's Euler system and, in one of several cases, congruences between automorphic forms on~$U(3)$ (\cref{sec:vwin}). \section{From the half-plane to the critical line}\label{sec:top} This section derives \cref{thm:main} from its part~(b) (\cref{lem:strip}) and part~(b) from the case of primitive characters (\cref{lem:imprimitive}), states the three inputs from which the archive proves~(b) (\cref{thm:three}), and traces the declarations that connect these statements. Lean declarations are in typewriter type. Files and lines, under the archive's folder \file{src/}, follow a declaration's name or, where a paragraph names several, stand in small type after the paragraph. Paths use these abbreviations: \begin{center}\small \begin{tabular}{@{}ll@{\quad}ll@{}} \code{DA} & \file{Development/AutomorphicArithmetic} & \code{AVT} & \file{Development/AbelianVarietyTheory}\\ \code{CB} & \file{DA/CharacterBoxes} & \code{WB} & \file{Development/WeilBound}\\ \code{PC} & \file{DA/PrimitiveCosets} & \code{FSR} & \file{Formalism/SharpRestriction}\\ \code{LT} & \file{DA/DedekindLFunctions/LandauTheorems} & \code{FLG} & \file{Formalism/LocalToGlobal}\\ \code{TPS} & \file{DA/ThetaCorrespondence/ThetaPairLifts/} & \code{TGC} & \file{FSR/RamificationFrobenius/}\\ & \quad\file{ThirdPlaceSeries} & & \quad\file{TotientGaloisClasses}\\ \code{AV} & \file{DA/AutomorphicVarieties} & \code{GTI} & \file{FSR/GlobalTheoremInputs.lean}\\ \code{SWB} & \file{Formalism/SizeWindowBrackets} & \code{LRC} & \file{Formalism/LocalRingCardinality} \end{tabular} \end{center} Lean namespaces are dropped where no confusion arises. A path beginning \file{.../} omits its leading folders, and a line number alone is in the file last named. \subsection{One term behind five theorems} The five theorems of \file{Main.lean} all come from a single proof of part~(b) of \cref{thm:main}, the statement for the half-plane $\Real s>\frac12$. Parts (a) and~(c), on the critical line, follow from~(b) by the functional equation and the Euler product (\cref{lem:strip}). The term \lean{rhOfAll} of the first two theorems has Mathlib's type \lean{RiemannHypothesis}, and the terms of the next two have the archive's types \lean{UpperTheory.EveryCharacter} and \lean{Claim.Strip}. Each type unfolds by definition to the statement of its theorem. {\ragged The terms that prove the five theorems are in one file, imported through two others. There \lean{allChi} is \lean{UpperTheory.FinalTheorems.everyCharacter}, whose type \lean{UpperTheory.EveryCharacter} is part~(b) of \cref{thm:main}. The other four theorems are derived from it: \lean{rhOfAll} applies \lean{Formalism.VW.rh_of_every}, \lean{critAll} applies \lean{Formalism.VW.critLine}, and \lean{claimAll} applies \lean{Claim.strip} and \lean{Claim.zeta}. Besides (b), these derivations use short lemmas of the archive proved from Mathlib alone, and from Mathlib the functional equations of $\zeta$ and of primitive Dirichlet $L$-functions, the Euler factors relating $L(s,\chi)$ to $L(s,\chi^*)$ and, in \lean{Claim.zeta}, the nonvanishing of $\zeta$ on $\Real s\ge1$. They are the classical argument, which we give. \locs{\lean{rhOfAll}~\file{Supplements/Kernels/MainTheorems.lean:9}, \file{Main.lean:4}, \file{Main.lean:6} \locsep \lean{UpperTheory.EveryCharacter}~\file{CB/EveryCharacter.lean:13} \locsep \lean{Claim.Strip}~\file{CB/HalfPlane.lean:6} \locsep the~statements~\file{Statements.lean} \locsep one~file~\file{Supplements/Kernels/MainTheorems.lean} \locsep imported~through~\file{MainTheoremsAll}, \file{MainTheoremsCritical} \locsep \lean{allChi}~\file{Supplements/Kernels/MainTheorems.lean:7}, \file{Main.lean:8} \locsep \lean{critAll}~\file{Supplements/Kernels/MainTheorems.lean:11}--\file{13} \locsep \lean{claimAll}~\file{Supplements/Kernels/MainTheorems.lean:15}--\file{16}, \file{Main.lean:10} \locsep \lean{Claim.strip}~\file{CB/ClaimStrip.lean:59} \locseplast \lean{Claim.zeta}~\file{CB/ClaimStrip.lean:75}}\par} \begin{lemma}\label{lem:strip} Part~(b) of \cref{thm:main} implies parts (a) and~(c). \end{lemma} \begin{proof} Let $\chi$ be primitive modulo $N\ge1$ (for $N=1$, $L(s,\chi)=\zeta(s)$), let $a\in\{0,1\}$ with $\chi(-1)=(-1)^a$, and let $\Lambda(s,\chi)=(N/\pi)^{(s+a)/2}\Gamma(\tfrac{s+a}2)L(s,\chi)$. Then $\Lambda(1-s,\bar\chi)=\epsilon(\chi)\Lambda(s,\chi)$ with $|\epsilon(\chi)|=1$, and for $N=1$ the function $\Lambda(s)=\pi^{-s/2}\Gamma(s/2)\zeta(s)$ satisfies $\Lambda(1-s)=\Lambda(s)$ \cite[Ch.~8, 9]{Davenport}. \emph{(c) for primitive $\chi$.} Let $0<\Real s<1$ and $L(s,\chi)=0$. If $\Real s>\frac12$, this contradicts (b), since $s\ne1$. If $\Real s<\frac12$, the factors $\Gamma(\frac{s+a}2)$ and $\Gamma(\frac{1-s+a}2)$ are finite and nonzero, so $\Lambda(s,\chi)=0$, hence $\Lambda(1-s,\bar\chi)=0$ and $L(1-s,\bar\chi)=0$ with $\Real(1-s)>\frac12$ and $1-s\ne1$; this contradicts~(b). So $\Real s=\frac12$. \emph{(c) in general.} If $\chi$ modulo $N$ is induced by the primitive $\chi^*$, then $L(s,\chi)=L(s,\chi^*)\prod_{p\mid N}(1-\chi^*(p)p^{-s})$, and the factors do not vanish for $\Real s>0$. A zero of $L(s,\chi)$ with $0<\Real s<1$ is therefore a zero of $L(s,\chi^*)$, and the primitive case applies. \emph{(a).} Let $\zeta(s)=0$ with $s\ne1$ and $s\notin\{-2,-4,\dots\}$. By (b) with $N=1$, $\Real s\le\frac12$, and by (c) $\Real s=\frac12$ if $\Real s>0$. Suppose $\Real s\le0$. As $\zeta(0)=-\frac12$, $s\ne0$, so $\Gamma(s/2)$ is finite and nonzero and $\Lambda(s)=0$. Then $\Lambda(1-s)=0$; since $\Real(1-s)\ge1$, $\Gamma(\frac{1-s}2)$ is finite, so $\zeta(1-s)=0$ with $\Real(1-s)>\frac12$ and $1-s\ne1$, against~(b). \end{proof} {\ragged In the archive, \lean{critLine} runs the primitive case of this argument, and \lean{rh_of_every} gives (a) through \lean{Formalism.VW.rh_of_right}, which treats all of $\Real s<\frac12$ at once by Mathlib's \lean{riemannZeta_one_sub}. \locs{\lean{critLine}~\file{Formalism/GammaFactorNonvanishing/CritLine.lean:11} \locsep \lean{rh_of_every}~\file{Formalism/GammaFactorNonvanishing/CritLine.lean:29} \locsep \lean{Formalism.VW.rh_of_right}~\file{Formalism/GammaFactorNonvanishing/RhOfRight.lean:5} \locseplast \lean{riemannZeta_one_sub}, applied~at~\file{Formalism/GammaFactorNonvanishing/RhOfRight.lean:28}}\par} \subsection{Primitive characters} {\ragged Alp\"oge reduces (b) to primitive characters as \lean{UpperTheory.Top.imprimitive}. Its hypothesis \lean{ZeroFreeAll} says that $L(s,\psi)\ne0$ for every primitive character $\psi$, the trivial one included, and every $s$ with $\Real s>\frac12$; at $s=1$ and $\psi=1$ this concerns Mathlib's value of $\zeta(1)$, which is not zero. The proof is the following lemma, through Mathlib's \lean{DirichletCharacter.LFunction_changeLevel}. \locs{\lean{UpperTheory.Top.imprimitive}~\file{CB/Imprimitive.lean:5} \locseplast \lean{ZeroFreeAll}~\file{TGC/ExplicitFormulaErrorOfHostConditions.lean:18}}\par} \begin{lemma}\label{lem:imprimitive} If $L(s,\psi)\ne0$ for every primitive $\psi$ and every $s$ with $\Real s>\frac12$, $(\psi,s)\neq(1,1)$, then part~(b) of \cref{thm:main} holds. \end{lemma} \begin{proof} Let $\Real s>\frac12$, with $\chi\ne1$ or $s\ne1$, and let $\chi$ modulo $N$ be induced by~$\chi^*$. If $\chi^*$ is trivial, then $\chi$ is the trivial character modulo~$N$, so $s\ne1$; in either case $(\chi^*,s)\ne(1,1)$, and $L(s,\chi^*)\ne0$. Then $L(s,\chi)=L(s,\chi^*)\prod_{p\mid N}(1-\chi^*(p)p^{-s})$, and $|\chi^*(p)p^{-s}|\le p^{-\Real s}<1$. (This application of Mathlib's \lean{LFunction_changeLevel} requires $\chi^*\ne1$ or $s\ne1$.) \end{proof} \subsection{Three inputs} Part~(b) follows from three inputs: Galois representations attached to polarized regular algebraic cuspidal automorphic representations over CM fields, local parameters for $\mathrm{GL}_n$ at the finite places of number fields, and a bound \lean{VWin} for Tate--Shafarevich groups of Tate twists. All three are proved in the archive, the first two in forms weaker than the published theorems. {\ragged The archive proves (b) as \lean{everyCharacter}, which is \lean{everyCharacter_of_row} applied to \lean{UpperTheory.Gal.Int.galoisLocalGlobalStrong_recParameter}, a proof of~(i) below. The theorem \lean{everyCharacter_of_row} has the single hypothesis~(i). It applies \lean{every_character_of_literature_of_vWin}, which, with its bundle unfolded, reads as follows.\looseness=-1 \locs{\lean{everyCharacter}~\file{CB/MainConclusion.lean:10} \locsep \lean{everyCharacter_of_row}~\file{CB/MainConclusion.lean:7} \locsep \lean{UpperTheory.Gal.Int.galoisLocalGlobalStrong_recParameter}~\file{DA/Foundations/GaloisRepresentations/IntegralStructures/RowSrc.lean:6} \locseplast \lean{every_character_of_literature_of_vWin}~\file{TPS/EveryCharacterOfLiteratureOfVWin.lean:810}}\par} \begin{theorem}[Alp\"oge]\label{thm:three} Suppose that \begin{enumerate}[(i)] \item \lean{Row0b}~\lean{IsRecT} holds, that is, for every CM field $E$, every prime $\ell$, every isomorphism $\iota\colon\bar\Q_\ell\cong\C$ and every regular algebraic cuspidal automorphic representation $\pi$ of $\mathrm{GL}_a(\mathbb A_E)$ polarized by a Hecke character of~$E^+$, there is a continuous semisimple $r\colon\Gal(\bar E/E)\to\mathrm{GL}_a(\bar\Q_\ell)$ with the properties of \cref{sec:weak} for every local parameter map $\mathrm{rc}$ satisfying \lean{IsRecT}; \item \lean{RecTExists} holds, that is, at every finite place $w$ of every number field $E$ there is a map from isomorphism classes of smooth irreducible complex representations of $\mathrm{GL}_n(E_w)$, all~$n$, to Weil--Deligne representations, satisfying \lean{IsRecT} (\cref{sec:weak}); \item \lean{VWin} holds (\cref{sec:vwin-statement}). \end{enumerate} Then part~(b) of \cref{thm:main} holds. \end{theorem} {\ragged The bundle \lean{OpenLiteratureStatements} has three fields, two copies of~(i) and one of~(ii). \lean{everyCharacter_of_row} fills it with (i), \lean{recTExists} and~(i), and supplies (iii) by \lean{vwin_of_row}, which proves (iii) from~(i). The archive proves (ii) as \lean{recTExists}. So \lean{everyCharacter} has no hypotheses. Two further theorems also conclude \lean{EveryCharacter}, namely \lean{plugAll} and \lean{everyE}$_2$, which applies \lean{plugAll}. Neither is on the route, as the proof map of \cite{MirrorIssue3} also finds. In input~(i) the local--global comparisons \lean{PropsB} hold only under stated conditions, and input~(ii) asserts injectivity and the laws of \lean{IsRecT} for smooth twists other than real powers of the norm, central characters beyond their absolute value, duals and depth zero only on representations built from characters and, over~$\Q_p$, from twists of depth-zero supercuspidals, whose inertia and zero monodromy it fixes (\cref{sec:weak}). \locs{\lean{OpenLiteratureStatements}~\file{TPS/EveryCharacterOfLiteratureOfVWin.lean:215} \locsep \lean{recTExists}~\file{DA/Foundations/LocalLanglands/LocalCentralUnits/RecTExists.lean:20} \locsep \lean{vwin_of_row}~\file{AV/ModularityArgument/ThresholdArguments/DenseThresholdCases/VwinOfRow.lean:4} \locsep \lean{plugAll}~\file{CB/EveryCharacter.lean:16} \locseplast \lean{everyE}$_2$~\file{DA/ThetaCorrespondence/FockModels/EveryCharacterE.lean:67}}\par} \subsection{The reduction}\label{sec:reduction} The proof of \cref{thm:three} derives from the three inputs the two hypotheses of \cref{thm:topg}, namely host fields on windows for each primitive~$\psi$ and the interface \lean{TopG}, which turns such hosts into a family that satisfies the conditions of \cref{sec:tied}, and with them the inequality~\Star{}, on its windows. For every primitive~$\psi$, \cref{thm:topg} then excludes zeros of $L(s,\psi)$ with $\Real s>\frac12$, and \cref{lem:imprimitive} gives~(b). The proof term of part~(b) unfolds as follows. Under each theorem stand the terms from which its proof term is built, and in the last two lines three of the arguments of \lean{forall_goal_of_shape_of_hyps}; a line number alone is in the file of the term it is indented under. The fifteen arguments of \lean{topCovolSlim2_use_alt_alt}, the covolume inputs of \cref{sec:known}, are not listed. {\footnotesize\raggedright\parskip=0pt\relax \tnode{0}{\lean{everyCharacter}~\file{CB/MainConclusion.lean:10}\enspace{\rmfamily part~(b)}} \tnode{1}{\lean{galoisLocalGlobalStrong_recParameter}~\file{.../IntegralStructures/RowSrc.lean:6}\enspace{\rmfamily input~(i)}} \tnode{1}{\lean{everyCharacter_of_row}~\file{:7}} \tnode{2}{\lean{recTExists}~\file{.../LocalCentralUnits/RecTExists.lean:20}\enspace{\rmfamily input~(ii)}} \tnode{2}{\lean{vwin_of_row}~\file{.../DenseThresholdCases/VwinOfRow.lean:4}\enspace{\rmfamily input~(iii), from~(i)}} \tnode{2}{\lean{every_character_of_literature_of_vWin} \file{TPS/EveryCharacterOfLiteratureOfVWin.lean:810}\enspace{\rmfamily \cref{thm:three}}} \tnode{3}{\lean{published_results_eigen}~\file{:666}\enspace{\rmfamily the first bundle}} \tnode{3}{\lean{published_results_first}~\file{:727}\enspace{\rmfamily the second bundle}} \tnode{3}{\lean{theta_integrals_agree_and_local_matching_of_literature}~\file{:565}\enspace{\rmfamily the two theta statements}} \tnode{3}{\lean{topCovolSlim2_use_alt_alt}~\file{TPS/TopCovolSlim2Use.lean:16}\enspace{\rmfamily the covolume bound}} \tnode{3}{\lean{local_langlands_gl_recT_exists_of_literature}~\file{:313}\enspace{\rmfamily \lean{RecTExists}, from~(ii)}} \tnode{3}{\lean{every_character_of_theta_integrals_local_matching} \file{TPS/SiegelEisensteinSecondTermHypotheses.lean:69}} \tnode{4}{\lean{green_of_regular}~\file{.../SumRangeEqRoots.lean:225}\enspace{\rmfamily Green's parametrization}} \tnode{4}{\lean{boundAtTwoThree_linked_of_recT}~\file{TPS/At23OfAll.lean:24}\enspace{\rmfamily the bound at $\ell=2,3$ (\cref{sec:definite})}} \tnode{4}{\lean{nonempty_hyps_definite_side_of_closed_bound}~\file{:44}\enspace{\rmfamily an element of \lean{HypsAlt}}} \tnode{4}{\lean{every_character_of_green_v_win}~\file{TPS/TopPlug.lean:16}} \tnode{5}{\lean{imprimitive}~\file{CB/Imprimitive.lean:5}\enspace{\rmfamily \cref{lem:imprimitive}}} \tnode{5}{\lean{zero_free_all_of_green_v_win}~\file{:12}} \tnode{6}{\lean{zeroFreeAll_of_topG}~\file{CB/ZeroFreeAllOfTopG.lean:26}\enspace{\rmfamily \cref{thm:topg} with $\lambda=\frac32$}} \tnode{6}{\lean{supplyChi}~\file{CB/SupplyChi.lean:8}\enspace{\rmfamily host fields on windows, from~(iii)}} \tnode{6}{\lean{forall_goal_of_shape_of_hyps}~\file{TPS/ShapeOf.lean:31}\enspace{\rmfamily the interface \lean{TopG}}} \tnode{7}{\lean{shape_closed_link_n_of_green}~\file{TPS/TopShape.lean:33}} \tnode{7}{\lean{trivial}, \lean{trivial}~\file{TPS/TopPlug.lean:15}\enspace{\rmfamily hypotheses of \lean{TopShapeAlt} set to \lean{True}}} \par}\smallskip {\ragged The two bundles are \lean{UnitaryGroupThirdPlaceHypotheses} and \lean{ThetaLocalInputs}, most of whose fields are named after published theorems; the first is filled from archive theorems together with (i) and~(ii), the second from archive theorems that use neither. One field of the first bundle, \lean{siegelEisensteinSimplePole}, is named after the simple pole of the Siegel Eisenstein series but has type \lean{UnusedSlot72}, which is \lean{True}, so it asserts nothing. From another field the archive proves Green's parametrization of the cuspidal irreducible characters of $\mathrm{GL}_d(\mathbb F_p)$, $p$ prime, by the monic irreducible polynomials of degree~$d$ over~$\mathbb F_p$ with nonzero constant term \cite{Green} (\lean{GlobalTheory.GLq.Green}). The two theta statements, \lean{ThetaIntegralsAgree} and \lean{LocalMatching}, compare theta integrals from two spaces into one unitary group and are proved from \lean{EisBL} and \lean{SiegelWeilStd} (\cref{sec:onelift}). The covolume bound \lean{CovolumeBoundOfX} (\cref{sec:covolume}) is proved from the covolume inputs of \cref{sec:known}. The element of \lean{HypsAlt}, the definite side (\cref{sec:definite}), is built from the bundles, the theta statements, the covolume bound, (ii) and the bound at $\ell=2,3$, which \lean{boundAtTwoThree_linked_of_recT} proves from the field \lean{nLink} of the first bundle. Green's parametrization and the element of \lean{HypsAlt} are both used in the proof of the interface \lean{TopG}: the first is the argument of \lean{shape_closed_link_n_of_green}, the second the last argument of \lean{forall_goal_of_shape_of_hyps}. \locs{\lean{UnitaryGroupThirdPlaceHypotheses}~\file{TPS/BernsteinUnitaryProductIrreducible.lean:22} \locsep \lean{siegelEisensteinSimplePole}~\file{TPS/BernsteinUnitaryProductIrreducible.lean:74} \locsep \lean{UnusedSlot72}~\file{.../DiscreteSpectrumResidualClassification.lean:7} \locsep \lean{GlobalTheory.GLq.Green}~\file{AVT/Definitions/NonGaloisDescent/LabelDatum.lean:108} \locsep \lean{TopG}~\file{CB/ZeroFreeAllOfTopG.lean:12} \locseplast \lean{TopShapeAlt}~\file{TPS/TopShapeAlt.lean:24}}\par} \begin{table}[ht] \centering\scriptsize\renewcommand{\arraystretch}{1.15} \caption{Statements of the chain for which we know of no reference, each with what it is needed for.}\label{tab:new} \begin{tabular}{@{}>{\raggedright\arraybackslash}p{0.38\textwidth}>{\raggedright\arraybackslash}p{0.28\textwidth}>{\raggedright\arraybackslash}p{0.28\textwidth}@{}} \toprule Statement & Declaration & Needed for\\ \midrule A one-sided inequality ``divisibility at most size'' on a $2$-power cyclotomic tower, turned into a bound for ramp sums & \lean{ch_balance}, \lean{core_gm} (\cref{prop:balance,prop:core}) & the window bound for $G_\psi$\\\addlinespace Theta integrals from two spaces on a unitary group that is not quasi-split are proportional, with the change of the constant under hyperbolic planes & \lean{ThetaIntegralsAgree} (\cref{sec:onelift}) & the reduction (\cref{sec:top})\\\addlinespace Seesaw identities along the whole chain & \lean{unfoldings_of_chain}, \lean{iter_colPer}, \lean{MainX} (\cref{sec:onelift}) & the Bessel period of the iterated lift as an iterated integral of column periods, and the transport of linear relations between periods from the top of the chain to the bottom\\\addlinespace Integrality of the corrected theta kernels at inert $p\ge5$, over CM fields of growing degree (one curve, one variable: Katz and others, \cref{sec:divis}) & \lean{IntInert} (\cref{sec:divis}) & divisibility of the actual periods by the lattice ideals, \lean{LinkDIVt}\\\addlinespace Near-orthogonality between different chains of lifts, off the degenerate orbits (Theorem~S) & \lean{TheoremSSelected} (\cref{sec:definite}) & \lean{LinkS}\\\addlinespace A point-sum bound for the cost of all $m$ lifts of a word & \lean{StepThmBG} (\cref{sec:definite}) & the diagonal term of each word\\\addlinespace A bound for the loss of integrality at the twisted places by lengths of Tate--Shafarevich groups & \lean{LinkN} (\cref{sec:definite}) & \lean{DIVinN}, and so \lean{LinkDIV}\\\addlinespace A lower bound for mean heights in signature $(p,q)$, $q\ge2$, $p\ge2q+1$ & \lean{VolumeTheoremX} (\cref{thm:volume}) & the covolume bound, \cref{thm:covol}\\\addlinespace A bound for plus parts polynomial in the modulus and the twist & \lean{CorGK} (\cref{sec:vwin}) & Count~I of \lean{VWin}\\\addlinespace A bound for $\#H^2(\mathcal O_L[1/\ell],\Z_\ell(e+1))$ for totally complex $L$ containing $\mu_\ell$ ($\mu_4$ if $\ell=2$), so that $\ell\le[L:\Q]+1$ & Lemma~K, \lean{lemmaK} (\cref{sec:vwin-proof}) & \lean{ThmBab}, and so the windows of \lean{VWin}; Counts~I, II, III and~V of the plus part\\ \bottomrule \end{tabular} \end{table} \section{Divisibility at most size excludes zeros right of \texorpdfstring{$\frac12$}{1/2}}\label{sec:analytic} This section follows the archive from the inequality \Star{} to \lean{ZeroFreeAll}. \Crefrange{sec:ramp}{sec:landau} reduce the absence of zeros right of~$\frac12$ to a bound $G_\psi(x)\le x^{1/2+O(\varepsilon)}$ on windows, \crefrange{sec:hosts}{sec:core} derive that bound from \Star{} on host fields, and \cref{sec:supply,sec:assembly} supply the host fields and put the steps together. Every step of the deduction is proved in the analytic part of the archive, with three exceptions: \Star{} comes from the chain of kernels of \cref{sec:star}, the host fields of \cref{sec:supply} come from the bound \lean{VWin} of \cref{sec:vwin}, and the families on windows from the interface \lean{TopG} and its class (\cref{sec:assembly}). \subsection{Ramp sums}\label{sec:ramp} By the explicit formula the ramp sums below are sums over the nontrivial zeros of $L$-functions, up to $O(\log y)$ (\cref{rem:explicit}). An upper bound $x^\theta$ on windows for the positive combination $G_\psi$ of their real parts (\cref{sec:level}) then excludes zeros of these $L$-functions with real part above~$\theta$ (\cref{thm:landau}). Alp\"oge works with the ramp sums \begin{align} G(y)&=\sum_{n\le y}\Lambda(n)\Bigl(\frac yn-1\Bigr)-y\log y+(1+\gamma)y,\label{eq:G}\\ G(\xi,y)&=\sum_{n\le y}\Lambda(n)\xi(n)\Bigl(\frac yn-1\Bigr)+y\,\frac{L'}{L}(1,\xi)\qquad(\xi\ne1\text{ primitive}),\label{eq:Gxi} \end{align} with $G(1,y)=G(y)$. In the archive they are \lean{Gz} and \lean{An.gChi}. \locs{\lean{Gz}~\file{Formalism/StatementOC.lean:13} \locseplast \lean{An.gChi}~\file{TGC/Analytic.lean:23}} \begin{remark}[classical]\label{rem:explicit} Since $\frac1{2\pi i}\int_{(2)}u^s\frac{ds}{s(s-1)}=\max(u-1,0)$ for $u>0$, the sum $\sum_{n\le y}\Lambda(n)\xi(n)(y/n-1)$ is $\frac1{2\pi i}\int_{(2)}-\frac{L'}{L}(s,\xi)\,\frac{y^s\,ds}{s(s-1)}$. Moving the line to the left, the residue at $s=1$ is $y\log y-(1+\gamma)y$ for $\xi=1$ and $-y\frac{L'}{L}(1,\xi)$ for $\xi\ne1$, and the residue at a nontrivial zero $\rho$ of multiplicity $m_\rho$ is $m_\rho y^\rho/(\rho(1-\rho))$. So, as $y\to\infty$, $G(\xi,y)=\sum_\rho m_\rho\,y^\rho/(\rho(1-\rho))+O_\xi(\log y)$, the sum converging absolutely \cite[Ch.~17, 19]{Davenport}. The corrections in \eqref{eq:G} and \eqref{eq:Gxi} remove the contribution of $s=1$. \end{remark} \subsection{The level of a character}\label{sec:level} Landau's argument needs one function in which the zeros of all the $L(s,\xi)$, $\xi\in\langle\psi\rangle$, appear with positive coefficients, and $G_\psi$ is that function. For a primitive $\psi$, Alp\"oge's \lean{Level.cyc} keeps the characters $\langle\psi\rangle$, with $o$ their number, and its \lean{Level.Gm} takes $G$ with coefficient~$1$ and the nontrivial ones with the weight \lean{Level.wt}, which is $1/(3o+1)$ here because \lean{Level.cyc} has no exceptional character (\lean{psK} is \lean{none}): \begin{equation}\label{eq:Gpsi} G_\psi(y)=G(y)+\frac1{3o+1}\sum_{\xi\in\langle\psi\rangle,\ \xi\ne1}\Real G(\xi,y). \end{equation} By \cref{rem:explicit}, $G_\psi(y)$ is a sum of the terms $\Real\bigl(y^\rho/(\rho(1-\rho))\bigr)$ over the nontrivial zeros of the $L(s,\xi)$, $\xi\in\langle\psi\rangle$, with positive coefficients, up to $O_\psi(\log y)$. \locs{\lean{Level.cyc}~\file{TGC/ExplicitFormulaErrorOfHostConditions.lean:121} \locsep \lean{Level.Gm}~\file{TGC/ExplicitFormulaErrorOfHostConditions.lean:38} \locseplast \lean{Level.wt}~\file{TGC/ExplicitFormulaErrorOfHostConditions.lean:36}} \subsection{Landau's theorem on windows}\label{sec:landau} {\ragged Landau's theorem turns an upper bound for $G_\psi$ into a zero-free half-plane. The archive needs it with the bound assumed only on windows $[Y,Y^\lambda]$, since the bound \lean{VWin} behind the host fields holds on windows that depend on the pair of primes $(q,d)$ from which the fields are built (\cref{sec:hosts,sec:supply}). The archive proves the following as \lean{Comparison}$_2$\lean{.formB}, through \lean{Deduction.zeroFree_of_rpow_window}. The statement is \lean{FormB} with \lean{HasWindows}. \locs{\lean{Comparison}$_2$\lean{.formB}~\file{PC/Imports.lean:8} \locsep \lean{Deduction.zeroFree_of_rpow_window}~\file{LT/ZeroFreeOfRpowWindow.lean:11} \locsep \lean{FormB}~\file{TGC/Window.lean:30,33} \locseplast \lean{HasWindows}~\file{TGC/Window.lean:7}}\par} \begin{theorem}[Landau's theorem, in the archive's window form]\label{thm:landau} Let $\psi$ be primitive, $0<\theta<1$ and $\lambda>1$. Suppose that for every $y$ there is $Y\ge y$ with $G_\psi(x)\le x^\theta$ for all real $x\in[Y,Y^\lambda]$. Then for every $\xi\in\langle\psi\rangle$ the function $L(s,\xi)$ has no zeros with $\Real s>\theta$. \end{theorem} \begin{proof}[Sketch of proof] Put $g(u)=G_\psi(e^u)$ and $\Phi(s)=\int_0^\infty g(u)e^{-su}\,du$ for $\Real s>2$. What the proof uses about $\Phi$ is the archive's \lean{PoleData}: $g$ is measurable with $|g(u)|\le Ce^{2u}$ for $u\ge0$; $\Phi$ continues holomorphically to $\Real s>0$ off a set $Z$ of points with positive real part and no accumulation point in that half-plane; each point of $Z$ is a simple pole with nonzero residue, real and positive when the pole is real. For $g$ coming from $G_\psi$, $Z$ is the set of zeros of the $L(s,\xi)$ in $\Real s>0$ and their conjugates, so not the trivial zeros, and the residue at $\rho$ is a positive combination of multiplicities times $1/(\rho(1-\rho))$, as in \cref{rem:explicit}. The proof uses no sign at a nonreal pole, but a positive residue at a real pole right of~$\theta$. The window hypothesis enters as \lean{WindowHypU}, which asks, for every $\eta>0$ and arbitrarily large $U$, for a majorant $R\ge0$ with $g\le R$ on $[U,(1+\delta)U]$ and $\int_U^{(1+\delta)U}R(u)e^{-\theta u}\,du\le e^{\eta U}$. Here $\delta=\lambda-1$ and $R(u)=e^{\theta u}$. \locs{\lean{PoleData}~\file{DA/DedekindLFunctions/LandauZeroFreeCriterion.lean:15} \locseplast \lean{WindowHypU}~\file{DA/DedekindLFunctions/LandauZeroFreeCriterion.lean:25}} The proof compares Taylor coefficients of $\Phi$. The $k$-th coefficient at a $\sigma$ chosen large in terms of $\delta$ is $c_k(\sigma)=\int g(u)\,u^ke^{-\sigma u}\,du/k!$ up to the sign $(-1)^k$, and for $k=\lceil(1+\delta/2)\sigma U\rceil$ the kernel $u^ke^{-\sigma u}/k!$ is concentrated on the window, the tails being bounded through $|g|\le Ce^{2u}$. On the window $|g|\le 2R-g$, so there the integral of $|g|$ against the kernel, which bounds the coefficient at every $\sigma+i\gamma$, is at most twice that of $R$ minus that of $g$, the window part of $c_k(\sigma)$. The result is \lean{landau_window_key}, which gives one $\sigma>2$ and one $C$ that serve every $\eta>0$. For each $\eta$ there are $U$ arbitrarily large and $k\ge U$ such that, for every real~$\gamma$, \[ \Bigl\|\frac{(-1)^k\Phi^{(k)}(\sigma+i\gamma)}{k!}\Bigr\|+\Real\frac{(-1)^k\Phi^{(k)}(\sigma)}{k!}\le\frac{Ce^{\eta U}}{(\sigma-\theta)^k}. \] Now suppose a pole has real part above~$\theta$. By \lean{landau_dichotomy} either a real pole $\beta>\theta$ is the only pole within some distance $r\le\sigma-\theta$ of~$\sigma$, or for some $D\le\sigma-\theta$ the open disc of radius $D$ about $\sigma$ has no pole and a pole $\rho_1$ has $\sigma-\Real\rho_10$, $t\in A$, a weight $w>0$ on $\Phi$ and a cutoff $N\ge1$, the archive defines \begin{align*} \ASK(a,t,w)&=\sum_{p}2\log p\sum_{\substack{1\le i\le N\\ \mathrm{Frob}_p^i\notin\Phi}}\max\Bigl(\frac a{p^i}-w\bigl(c\,\mathrm{Frob}_p^i\,\tau_t\bigr),0\Bigr)+\log d\cdot\mathrm{ASK}_d,\\ \MAIN(a,w)&=a\Bigl(\log a-1-\gamma+\frac{L'}{L}(1,\chi_{-d})-\frac1{|\Phi|}\sum_{g\in\Phi}\log w(g)\Bigr), \end{align*} where $\mathrm{Frob}_p$ is a Frobenius element, at a ramified prime a chosen lift, $p$ runs over the primes up to $N$ other than $2$, $q$ and $d$ with $\mathrm{Frob}_p\notin\Phi$ (\lean{IsInert}), $\tau_t$ is the element that \lean{topPl} attaches to~$t$, with $g\mapsto cg\tau_t$ the archive's \lean{partner}, and $\mathrm{ASK}_d$ the corresponding term at the ramified prime~$d$. The archive writes $q_1$ for~$q$ (\lean{arith}). The primes $p$ in the sum are those inert in $K$, and since $\Phi$ has index~$2$, $\mathrm{Frob}_p^i\notin\Phi$ means that $i$ is odd. The value $\frac{L'}{L}(1,\chi_{-d})$ is $\ell(\chi_C)$ for a function $\ell$ on characters that, at every nontrivial~$\theta$, is $\frac{L'}{L}(1,\cdot)$ of the matching primitive character (\lean{HostAddCharCompatible}). \locs{$\ASK$ and $\MAIN$~\file{SWB/FirstPrimeNeSecondPrime.lean:19}--\file{89} \locsep \lean{IsInert}~\file{SWB/FirstPrimeNeSecondPrime.lean:72} \locsep \lean{partner}~\file{SWB/FirstPrimeNeSecondPrime.lean:19} \locsep \lean{arith}~\file{PC/RecH.lean:23} \locseplast \lean{HostAddCharCompatible}~\file{PC/CoreAtH.lean:18}--\file{20}} At the scale $x$ put $b=\exp(x^\varepsilon)$ and $X'=\exp(x^{\varepsilon/3})$. The family consists of a set $TT\subset A$ with $|TT|\ge x^{-\varepsilon/8}|A|/2$ and sizes $a_t$, $|a_t/b-x|\le1$; a chain length $m$ and an integer $k$ with \[ x^{1/2-\varepsilon}\le m\le x^{1/2+\varepsilon/2},\qquad xm\le k\le|A|^2x^{1/2-\varepsilon}; \] a set $T_2\subset A_0$ with $|T_2|\ge x^{-\varepsilon}|A|$; and, for each $u\in A_0$ and $j0$, $w>0$ on $\Phi$ and $N\ge a/\min w$, \[ \ASK(a,t,w)-\MAIN(a,w)=\frac1{|\Phi|}\sum_{g\notin\Phi}\ \sum_\theta\overline{\theta(g)}\,w(g')\,G\bigl(\theta,a/w(g')\bigr),\qquad g'=c\,g\,\tau_t, \] the inner sum over all characters $\theta$ of $\Gal(E/\Q)$. \end{proposition} \begin{proof}[Sketch of proof] For fixed $g$, the kernel $|\Phi|^{-1}\sum_\theta\overline{\theta(g)}\theta(p^i)$, an average over the inertia coset $\mathrm{Frob}_p^iI_p$, is $2/|I_p|$ on that coset and $0$ off it. Since $w(g')(y/p^i-1)=a/p^i-w(g')$ for $y=a/w(g')$, the prime-power parts of the right side are the sums of $\log p\,(a/p^i-w(g'))_+$ over the $g\notin\Phi$ in these cosets, with weight $2/|I_p|$. At $p\notin S\cup\{d\}$ the coset is $\{\mathrm{Frob}_p^i\}$, with weight $2$; at $p\in S$ it lies in $\Phi$ and gives nothing; at $d$ it has one element outside~$\Phi$, with weight $1$. Adding these gives $\ASK$. Of the main terms, those of $\theta=1$ give $-a\log a+a\,|\Phi|^{-1}\sum_\Phi\log w+(1+\gamma)a$ (as $g\mapsto g'$ maps the complement of $\Phi$ onto $\Phi$), those of $\theta=\chi_C$ give $-a\,\ell(\chi_C)$, and those of the other $\theta$ cancel, since $\sum_{g\notin\Phi}\overline{\theta(g)}=0$ unless $\theta$ is trivial on~$\Phi$. Together they are $-\MAIN$. \end{proof} The identity is elementary. It is also where the Euler product enters. For the Davenport--Heilbronn function~$f$, which has zeros with real part above~$\frac12$, the ramp sum of $-f'/f$, once centered by adding $y\,f'(1)/f(1)$, has the properties of \lean{PoleData} used in \cref{thm:landau}, but $f$ has no Euler product (the coefficient of $-f'/f$ at~$6$ is not zero), and the orthogonality step above fails for it. \subsection{From \Star{} to the window inequality}\label{sec:core} The window inequality $G_\psi(x)\le x^{1/2+O(\varepsilon)}$, from which \cref{thm:landau} excludes zeros right of~$\frac12$, follows from \Star{} on the host fields. By the balance identity, averaging the left side of \Star{} over $A_0$ and over the shifts leaves, up to errors, $G_\psi(x)$ and $3o$ twisted sums that are bounded separately. The statement in the archive is \lean{core_gm}, proved through \lean{ch_core}. \locs{\lean{core_gm}~\file{PC/CoreGm.lean:12} \locseplast \lean{ch_core}~\file{LT/ChCore.lean:343}} \begin{proposition}[Alp\"oge]\label{prop:core} For every primitive $\psi$, all real $e_D$, $e_L$ and $C$, all values of three further real parameters described below, and every $\varepsilon>0$, there is $x_1$ such that for every real $x\ge x_1$, every frame, every host $H$ satisfying \lean{HostAx} and every subset $TT$ of~$A$, if \lean{EAtH} holds with the constant~$C$ and \lean{CoreAtH} with the five other parameters, for these data at scale~$x$, then $G_\psi(x)\le x^{1/2+(e_D+e_L+7)\varepsilon}$. \end{proposition} \lean{CoreAtH} asserts that there exist the data of \cref{sec:star-def} on~$H$, with $b$, $X'$ and~$k$ any real numbers that satisfy the conditions there, and a record of real numbers of the chain at the scale~$x$, with the constants of its conditions, identified with $\ASK$ and $\MAIN$ (\lean{IdentU}, \lean{SizeData}$_2$, \lean{ErrWindows}, \lean{BNBoundsU}). The constants of the chain's conditions are thus part of what is asserted to exist, and may change with $x$ and with the host; the parameters of the proposition only bound them. The exponents satisfy $e_{\mathrm{Div}}\le e_D$ and $e_{\mathrm{Loss}}\le e_L$; of the three further parameters, one bounds the exponent $c$ in the chain's height bound $x^{c}b$ (\lean{HHBound}), one is an exponent of~$x$ that bounds~$k$, and one bounds seven nonnegative constants of the conditions, $c_{\mathrm{Div}}$ among them. So $x_1$ depends only on $\psi$, $\varepsilon$ and the six parameters. \Star{} is not among the conditions: for $x\ge x_1$ and $u\in A_0$, \lean{ch_top} derives \Star{} at~$u$ from them by real algebra, and \lean{core_gm} enlarges the exponent to $(e_D+e_L+3)\varepsilon$. Among the conditions \lean{EAtH}, three relate $A_0$ to~$\psi$ (\lean{o_eq}, \lean{kept}, \lean{ps}), and two matter here. The first, on the shares, says that for every $g\in A$ the number of $t\in TT$ with $t-g\in A_0$ is $\lambda(g)|TT|$ up to $x^{-1+\varepsilon}|TT|$, where $\lambda(g)=4/(3o+1)$ on $A_0$ and $3/(3o+1)$ off it (\lean{Frame.lam}, \lean{Frame.Shares}). The second gives the bounds $|G(\theta,x)|\le x^{1/2+\varepsilon}$ and $|\frac{L'}{L}(1,\theta)|\le C\log x$ for the characters trivial on $A_0$ that are not kept. \locs{\lean{HostAx}~\file{TGC/ConjEMemHostFIff.lean:174} \locsep \lean{CoreAtH}~\file{PC/CoreAtH.lean:29} \locsep \lean{IdentU}~\file{DA/DedekindLFunctions/SplitInertia.lean:67} \locsep \lean{BNBoundsU}~\file{DA/DedekindLFunctions/SplitInertia.lean:74} \locsep the parameters~\file{PC/CoreAtH.lean:35}, \file{SWB/ErrorFormsWeakWindowsFalse.lean:23}--\file{45} \locsep \lean{HHBound}~\file{Formalism/RankGapBounds/MeanBoundSecondForm.lean:158} \locsep \Star{}~derived~\file{PC/CoreGm.lean:57}--\file{60} \locsep \lean{ch_top}~\file{LT/ChTop.lean:37} \locsep enlarges the exponent~\file{PC/CoreGm.lean:38}--\file{60} \locsep \lean{EAtH}~\file{TGC/ExplicitFormulaErrorOfHostConditions.lean:144}, its fields \lean{o_eq}, \lean{kept}, \lean{ps}~\file{145}--\file{147} \locsep \lean{Frame.lam}~\file{TGC/ExplicitFormulaErrorOfHostConditions.lean:135} \locseplast \lean{Frame.Shares}~\file{TGC/ExplicitFormulaErrorOfHostConditions.lean:139}} \begin{proof}[Sketch of proof] By \cref{prop:balance}, \Star{} bounds from above a weighted sum of ramp sums $G(\theta,y)$, at $y=a_t/w_{u,j}(g')$. At the thin places $y=a_t/w\ge X'/3$, and \lean{ThinScales}, one of the conditions \lean{CoreAtH}, gives $|G(\theta,y)|\le y\,\theta_0$ for a $\theta_0$ with $2|\Phi|(x+1)\theta_0\le1/x$ (\lean{CostWindows}). On the actual hosts, which lie in $\Q(\zeta_{Q_Sqd})$ with $Q_S$ prime to $qd$, the bound holds with $\theta_0=\exp(-cx^{\varepsilon/8})$ by \lean{ThinH}, which the archive proves as \lean{thinH} for $0<\varepsilon<\frac1{52}$, a bounded odd part of $Q_S$, $\log(Q_Sqd)\le2x^{\varepsilon/5}\log^2x$ and \lean{RealBox}, which says that no real character of $E$ whose conductor is divisible by $q$ or $d$ has an $L$-function with a real zero above $1-\frac1{400}$. The bound of \lean{thinH} is the prime number theorem for characters, through $G(\chi,y)=-y\int_y^\infty\Psi(u,\chi)\,u^{-2}\,du$, $\Psi(u,\chi)=\sum_{n\le u}\Lambda(n)\chi(n)$, less $u$ when $\chi=1$, and de la Vall\'ee Poussin's zero-free region, since $\log y/\log(Q_Sqd)\gg x^{2\varepsilon/15}/\log^2x$. \lean{RealBox} excludes an exceptional zero when $q$ or $d$ divides the conductor, and the other real characters have bounded conductor. The total over the thin places is absorbed in the error. At the thick places $w\in[b,b+\eta+1]$, so replacing $w$ by $b$ costs an error that is absorbed (\lean{LipW}; \lean{CostWindows}, also in \lean{CoreAtH}). Summing over $u\in A_0$ then leaves only the $4o$ characters trivial on $\{0\}\times A_0$ (\lean{sum_Th}). A thick place $g'=c\,\iota_d(s+u)$ comes from $g=\iota_d(s+u-t)$, so the thick part becomes $|A_0|\,b\,(\delta_2/2)\sum_\theta\theta(\iota_dt)\,G(\theta,a_t/b)$ with $\delta_2=|T_2|/|A|$, and $\theta(\iota_dt)=\theta(t)$ for $\theta\in\langle\psi\rangle$, which are unramified at~$d$. Placing the thick places in the coset $c\,\iota_dA$ removes the factor $\theta(c)=\psi(-1)^j$ that complex conjugation brings when $\psi$ is odd; with the thick places in $A$, the trivial coefficient would stay~$1$ and that of a nontrivial $\psi^j$ would be within $o\,x^{-1+\varepsilon}$ of $\psi(-1)^j/(3o+1)$. The archive applies \Star{} for each $u$, sums over $A_0$ and divides by $|A_0|$, and combines this with a separate lower bound in which $a_t/b$ is moved to $x$ (\lean{move}; \lean{one_char}), at a cost absorbed in the term $4x^{1/2+\varepsilon}$ below. Summing over $t\in TT$ and dividing by $|TT|$, the trivial character keeps the weight $1$, and the shares give the other kept characters weights within $o\,x^{-1+\varepsilon}$ of $1/(3o+1)$, the Fourier transform of $\lambda$ on $A/A_0$ at a nontrivial character being $\frac4{3o+1}-\frac3{3o+1}$. The discrepancies go to the error, and the sum with the ideal weights is $G_\psi$ (\lean{gmA_eq}). This step, like the averaging, is elementary. The $3o$ characters trivial on $A_0$ that are not kept are bounded in absolute value, $\mathrm{SigA}\le4o\,x^{1/2+\varepsilon}$ (\lean{sigA_le}). The result is \[ |TT|\,m\,b\,\frac{\delta_2}2\bigl(G_\psi(x)-\mathrm{SigA}-4x^{1/2+\varepsilon}\bigr)\le\text{right side of \Star}+\frac{2|TT|mb}x, \] and dividing out, the sizes of \cref{sec:star-def} turn the right side into $x^{O(\varepsilon)}(x/m+m+\sqrt x)$, which is $x^{1/2+O(\varepsilon)}$ on the range of~$m$. In the exponent the losses add up to $(e_D+e_L+7)\varepsilon$.\qedhere \locs{\lean{ThinScales}~\file{SWB/GrowthBoundGivesRiemannHypothesis.lean:55} \locsep \lean{CostWindows}~\file{SWB/SChain.lean:50} \locsep the actual hosts~\file{TGC/ConjEMemHostFIff.lean:21,33,54} \locsep \lean{ThinH}~\file{CB/PairGood.lean:36}--\file{39} \locsep \lean{thinH}~\file{DA/MeanSquareBounds/GaussSumFunctionEq.lean:55} \locsep \lean{LipW}~\file{SWB/GrowthBoundGivesRiemannHypothesis.lean:58} \locsep \lean{sum_Th}~\file{LT/ChCore.lean:56} \locsep divides~by~$|A_0|$~\file{LT/ChCore.lean:283} \locsep \lean{move}~\file{LT/ChCore.lean:154} \locsep \lean{one_char}~\file{LT/ChCore.lean:202} \locsep \lean{gmA_eq}~\file{PC/GmAEq.lean:61} \locseplast \lean{sigA_le}~\file{PC/SigALe.lean:38}} \end{proof} \subsection{Pairs and host fields on windows}\label{sec:supply} The window inequality needs a host field at every scale~$x$ of a window, built from a pair $(q,d)$ for which the twisted ramp sums are small. The archive obtains the host fields from the bound \lean{VWin} of \cref{sec:vwin}, and the pairs from a mean square over the moduli and Markov's inequality. Alp\"oge's \lean{supplyChi} turns \lean{VWinAt}~$\psi$~$401$ (\cref{sec:vwin-statement}) into host fields on windows, \lean{HostExW}~$\frac32$~(\lean{GChi}~$\psi$)\lean{.req}. This says that for small $\varepsilon$ there are windows $[Z,Z^{3/2}]$ arbitrarily far out on which every real $x$ has ranks and a host field at scale $x$ meeting \lean{RankVal} and \lean{HostGood} with common constants, the conditions \lean{EAtH} of \cref{sec:core} and the bound $\sqrt b$ for the weighted lengths of \cref{sec:vwin} over $1\le e1-\frac1{400}$ and $|\Ima s|$ at most a height depending on $M,\varepsilon,Y,j$ (\lean{ZeroBox}). Their existence, with $X=Y^A$, for every $Y$ sufficiently large in terms of $M\ge1$, $0<\varepsilon<\frac1{52}$ and $A\ge1$ (\lean{PairEx}), is \lean{pairEx}. Since $G(\chi,y)=-\sum_n\Lambda(n)\chi(n)\min(1,y/n)$, the proof cuts the sum at a fixed power $Y^{c_0}$ of~$Y$. For $n\le Y^{c_0}$, a mean square of the sums of $\Lambda(n)\min(1,y/n)\kappa(n)\,(n/q)$, $|\kappa(n)|\le1$, over prime moduli $q$ far larger than~$Y^{c_0}$ (\lean{meanSqK}), and Markov's inequality (\lean{bad_share}) leave few bad moduli, first among the~$q$, then among the~$d$. The tail $n>Y^{c_0}$ is small when $L(s,\chi)$ has no zero with $\Real s>1-\frac1{400}$ up to a height, and \lean{count_bad} shows that few moduli have one, through a zero-density estimate proved in the archive: at most $CQ^{1/4}$ primitive characters of conductor at most~$Q\ge3$ have a zero with $\Real s>1-\frac1{400}$ and $|\Ima s|\le Q$ (\lean{zero_density_box}), an estimate of the classical kind \cite{Montgomery}. The primes are counted in their progressions, and the sums of Jacobi symbols over them bounded (\lean{twistK}), by a prime number theorem for progressions with a term for an exceptional zero (\lean{PNTAPRelW}), Page's theorem (\lean{page_across}) and the bound $\beta\le1-c(\varepsilon)q^{-\varepsilon}$ for real zeros of real characters modulo~$q$ (\lean{ExcBound}), which the archive derives from Siegel's lower bound for $L(1,\chi)$ (\lean{siegelLower}). \locs{\lean{PairGood}~\file{CB/PairGood.lean:18} \locsep \lean{PairEx}~\file{CB/PairGood.lean:28} \locsep \lean{pairEx}~\file{DA/MeanSquareBounds/pairEx.lean:112} \locsep \lean{meanSqK}~\file{DA/MeanSquareBounds/meanSqK.lean:51} \locsep \lean{count_bad}~\file{DA/MeanSquareBounds/CountBad.lean:9} \locsep \lean{zero_density_box}~\file{Formalism/GL3MeanSquares/GenusMeanAsymptotics/JacobiCharacterSums/CountZAll.lean:13} \locsep \lean{twistK}~\file{DA/MeanSquareBounds/twistK.lean:148} \locsep \lean{PNTAPRelW}~\file{Formalism/GL3MeanSquares/GenusMeanAsymptotics/JacobiCharacterSums/MeanSquareHypWIff.lean:38} \locsep \lean{page_across}~\file{Formalism/GL3MeanSquares/GenusMeanAsymptotics/JacobiCharacterSums/MSqPage.lean:37} \locsep \lean{ExcBound}~\file{Formalism/LandauZeroFreeBounds/LogDerivRegionChiOf.lean:50} \locseplast \lean{siegelLower}~\file{Formalism/LandauZeroFreeBounds/FinalSiegel.lean:17}} The set $T_2$ of thick places is the image under $s\mapsto s\,a_0$, for a generator $a_0$ of $A_0$ of order~$n_0$, of a set of natural numbers without short relations. \lean{hostThickExists} gives one with \lean{NoRel}, that is, with no relation $j(s_2-s_1)=i(s_3-s_1)$ among distinct elements $s_1,s_2,s_3$ of the set, for integers $i\ne j$ with $0<|i|,|j|\le L$, where $L\ge2$ is a given natural number, and with $4Ls\frac12$. In the archive the assembly is \lean{zeroFreeAll_of_topG} (\file{CB/ZeroFreeAllOfTopG.lean:26}). \par} \begin{theorem}[Alp\"oge]\label{thm:topg} Let $\lambda>1$. Assume \lean{TopG}, and that for every primitive $\psi$ host fields meeting \lean{GChi}~$\psi$ exist, with common constants, on windows of exponent~$\lambda$ \textup{(}\lean{HostExW}~$\lambda$~\textup{(}\lean{GChi}~$\psi$\textup{)}\lean{.req}\textup{)}. Then for every primitive $\psi$ the function $L(s,\psi)$ has no zeros with $\Real s>\frac12$. \end{theorem} The interface \lean{TopG} says that for every $\lambda>1$ and every requirement \lean{FReq} on the scale, the frame, the data \lean{HostAx}, $TT$ and the rank parameter, if hosts meeting it exist on windows and every family in the requirement's class satisfies \lean{HypV} (the bound $\sqrt b$ of \lean{GChi} at the family's hosts), then \lean{FamilyClassConnected} holds. This says that there are $e_D,e_L$ and $\varepsilon_1>0$ such that for every $0<\varepsilon<\varepsilon_1$ some family of the class satisfies \lean{ChainTied} (\cref{sec:tied}) for all large $x$ in its windows, with $e_{\mathrm{Div}}\le e_D$ in \eqref{eq:size} and $e_{\mathrm{Loss}}\le e_L$ for the losses. The other constants of \lean{ChainTied} may depend on~$\varepsilon$ but move only $x_1$ in \cref{prop:core}. The archive proves \lean{TopG} with $e_D=2$ and $e_L=0$, so the bound is $G_\psi(x)\le x^{1/2+9\varepsilon}$. \locs{\lean{TopG}~\file{CB/ZeroFreeAllOfTopG.lean:12} \locsep \lean{FReq}~\file{CB/Supply.lean:10} \locsep \lean{HypV}~\file{CB/DivinNOfHypV.lean:10}--\file{11} \locsep \lean{FamilyClassConnected}~\file{CB/DivinNOfHypV.lean:13} \locsep $e_{\mathrm{Div}}\le e_D$ and $e_{\mathrm{Loss}}\le e_L$~\file{FSR/GlobalSchurBounds/TieGlobal.lean:107} \locseplast $e_D=2$ and $e_L=0$~\file{TPS/TopShape.lean:52}} \begin{proof}[Sketch of proof] Fix $\psi$ and $s$ with $\Real s>\frac12$, and take $\theta=\frac12+\frac12\min(\Real s-\frac12,\frac12)$ and $\varepsilon$ with $(e_D+e_L+7)\varepsilon\le\frac12\min(\Real s-\frac12,\frac12)$ (\lean{level_of_tiedWin}). By \lean{TopG}, applied to the requirement \lean{GChi}~$\psi$, the supplied hosts and \lean{hypV_of_cls}, there is for $0<\varepsilon<\varepsilon_1$ a family of the class with \lean{ChainTied} on its windows, and the condition \lean{GChi}~$\psi$ that its class imposes at every large scale gives it \lean{EAtH} (\lean{eFamC_of_cls}). \Cref{prop:core} then gives $G_\psi(x)\le x^{1/2+(e_D+e_L+7)\varepsilon}\le x^\theta$ there (\lean{level_gm_le_on_family_scales}, of type \lean{ChebyshevInequalityOnWindows}), and \cref{thm:landau} shows that $L(s,\xi)$ has no zeros with $\Real s>\theta$. In the archive this is \lean{zeroFreeOn_of_tiedWin_cyc} applied to \lean{Level.cyc}~$\psi$.\qedhere \locs{\lean{zeroFreeOn_of_tiedWin_cyc}~\file{TGC/TopWin.lean:90} \locsep \lean{level_of_tiedWin}~\file{TGC/TopWin.lean:68} \locsep \lean{hypV_of_cls}~\file{CB/ZeroFreeAllOfTopG.lean:16} \locsep \lean{eFamC_of_cls}~\file{CB/ZeroFreeAllOfTopG.lean:19} \locsep \lean{level_gm_le_on_family_scales}~\file{PC/LevelGmLeOnFamilyScales.lean:28} \locseplast \lean{ChebyshevInequalityOnWindows}~\file{TGC/TopWin.lean:26}} \end{proof} \needspace{4\baselineskip} \begin{remark}[Where the exponent $\frac12$ enters]\label{rem:half} The exponent $\frac12$ comes from two places. The first is the length of the chain. With $m$ lifts the right side of \Star{}, divided by $|TT|\,m\,b$, is $x^{O(\varepsilon)}(x/m+m+\sqrt x)$: $x/m$ from the term $|TT|\,a(1+m^2/k)\le2|TT|\,a$, which does not grow with~$m$; $m$ from the error $B_0\le2|A|\,m\,b\,x^\varepsilon$ that each of the $m$ lifts pays in \eqref{eq:size}; $\sqrt x$ from the term in~$k$, as $k\le|A|^2x^{1/2-\varepsilon}$. The first two balance at $m\approx\sqrt x$, which makes it $x^{1/2+O(\varepsilon)}$ (\cref{prop:core}). The second is the twisted ramp sums. \lean{EAtH} asks $|G(\theta,x)|\le x^{1/2+\varepsilon}$ for the $3o$ characters that are not kept, the bound that the Riemann hypothesis for their $L$-functions would give at this scale. Under that hypothesis \cref{rem:explicit}, made uniform in the conductor~$Q$, gives $|G(\theta,x)|\ll\sqrt x\log Q+\log x$, and here $Q$ divides $Q_Sqd$, so that $\log Q\le2x^{\varepsilon/5}\log^2x$. The archive obtains the bound of \lean{EAtH} not for every character but for the pairs $(q,d)$ of \lean{PairGood}, from a mean square over the moduli and Markov's inequality (\cref{sec:supply}). \Cref{thm:landau} then turns the bound on windows into zero-free half-planes with no loss in the exponent. The window bound is sharp up to the factor $x^{O(\varepsilon)}$. Indeed, since $1\in\langle\psi\rangle$ and $\zeta$ has zeros on the line $\Real s=\frac12$, \cref{thm:landau} shows that for every $0<\theta<\frac12$ and $\lambda>1$ the bound $G_\psi(x)\le x^\theta$ fails at some $x$ in every window $[Y,Y^\lambda]$ with $Y$ large. \end{remark} \begin{remark}\label{rem:analytic-verdicts} We know of no reference for the deduction as a whole, in which a one-sided inequality ``divisibility at most size'' over a family on a $2$-power cyclotomic tower of CM fields is turned, by an identity and an averaging, into a one-sided bound for the positive combination $G_\psi$ of ramp sums. \end{remark} \section{Where divisibility at most size comes from}\label{sec:star} This section follows the archive from its chain of kernels, which Alp\"oge describes as theta lifts and which are built from theta functions (\cref{sec:datum}), to \Star{}. \Cref{sec:tied} gives the interface with \cref{sec:analytic} and the conditions behind it, \cref{sec:datum} the datum and its kernels, \cref{sec:onelift,sec:divis} the lifts and their integrality at the inert primes, \cref{sec:definite} what supplies the conditions, and \cref{sec:covolume} the covolume bound. \subsection{The tied chain}\label{sec:tied} In the archive the divisible object is, for each word~$\omega$, a map $\mathrm{Bk}_\omega\colon N_\omega\to RR^\vee$ of Hermitian lattices whose entries are periods. This subsection describes these lattices and three of the conditions on them, and states the hypotheses from which Alp\"oge's \lean{ch_top} derives \Star{}. The interface with \cref{sec:analytic} is \lean{ChainTied}. At the scale $x$ it gives a host field with the data of \cref{sec:star-def} and, for each $u\in A$, a member of the family. A member carries a totally complex Galois number field $LL$, a finite set of \emph{words}~$\omega$ (the columns), Hermitian $\mathcal O_{LL}$-lattices $N_\omega$ and $RR$, linear maps $\mathrm{Bk}_\omega$ from the ambient space of $N_\omega$ to the dual of that of $RR$, integral by \lean{LinkDIV} below, and real numbers $h(\omega)$ (\lean{DivData}). The clauses \lean{TieLocal} identify the numbers that \lean{ChainTied} speaks about with the ranks, slopes and maximal slopes of these lattices and with the $h(\omega)$, and \lean{SizeData}$_2$ lists eleven conditions on them. The archive computes slopes with squared norms, so that $\mu=2\widehat\mu$ for Bost's normalized slope~$\widehat\mu$. \locs{\lean{ChainTied}~\file{FSR/GlobalSchurBounds/TieGlobal.lean:97} \locsep \lean{DivData}~\file{LRC/DataArith.lean:139} \locsep \lean{TieLocal}~\file{FSR/GlobalSchurBounds/TieGlobal.lean:63} \locseplast \lean{SizeData}$_2$~\file{SWB/SChain.lean:17}} We describe three of the conditions. \begin{enumerate} \item The size bound (\lean{DIV}) says that for every word $\omega$, \begin{equation}\label{eq:size} \mu(RR^\vee)+h(\omega)-F_o\le\MAIN_\omega+c_{\mathrm{Div}}|TT|\,a\log(a\,\mathrm{rd}_E)+c_{\mathrm{Div}}\,m\,(B_0+N_c)\,x^{e_{\mathrm{Div}}\varepsilon}, \end{equation} where $\MAIN_\omega=\sum_{t\in TT}\bigl[m\,(a_t\log a_t-a_t-a_t\log\Omega)-a_t\log T_\omega\bigr]$, $T_\omega$ is a prime attached to~$\omega$, $B_0$ and $N_c$ are the two error sizes of \lean{BNBounds}, and $F_o$ is the size of a correction that the archive fixes place by place from the weights and the Fock degrees of the lifts (\lean{Fval}). \locs{\lean{DIV}~\file{Formalism/RankGapBounds/MeanBoundSecondForm.lean:118} \locsep \lean{BNBounds}~\file{SWB/SChain.lean:33} \locsep $F_o$~\file{LRC/DataParams}$_2$\file{.lean:218} \locseplast \lean{Fval}~\file{LRC/DataForced.lean:126}--\file{132}} \item The slope inequality (\lean{SlopeIneq}) says that if $s$ is the mean of $\mu(N_\omega)-h(\omega)$ weighted by rank, $u_F=\max_\omega(\mu_{\max}(N_\omega)-h(\omega))-s$, and $H$ is the member's height bound, with $00$, with $L\ne0$, the classical inequality is, in Bost's slopes, $\widehat\mu_{\max}(L)\le\widehat\mu_{\max}(M)+\log C$ \cite[Prop.~4.3(1)]{Bost}, which the archive proves as \lean{slope_inequality_general}. Alp\"oge's \lean{ch_top} derives \Star{} by real algebra from \lean{SizeData}$_2$, the error windows, the bounds \lean{BNBoundsU}, the largeness condition \lean{LargeX} and the identification \lean{IdentU} (\lean{Ident}, with a map of index sets) of the archive's $\ASK$ and $\MAIN$ with those of \cref{sec:star-def}. This is the computation of ``The idea'' (\cref{sec:intro}), done with weights and errors. For a fixed member one takes the rank-weighted mean of the shifted slopes of the $N_\omega$, uses the comparison of slopes and the counting estimate for $\log T_\omega$, and bounds the gains of the lifts together from below. These estimates bound the left side of \Star{} above (\lean{ch_top}), and \cref{prop:balance} expresses each of its terms $\ASK-\MAIN$ by centered ramp sums, the ramified contribution included. \locs{\lean{slope_inequality_general}~\file{Formalism/HarderNarasimhan/HarderNarasimhanProofs/SlopeIneq.lean:507} \locsep statement~\file{Formalism/HarderNarasimhan/Statement.lean:91} \locsep \lean{ch_top}~\file{LT/ChTop.lean:37} \locsep \lean{Ident}~\file{SWB/SChain.lean:26} \locseplast map of index sets~\file{DA/DedekindLFunctions/SplitInertia.lean:67}} In $\MAIN_\omega$, by \lean{logOm} and \lean{Host.v}, \[ \log\Omega=\gamma+\log\frac{2\pi}{\mathrm{rd}_E}-\frac{L'}L(1,\chi_{-d}),\qquad \mathrm{rd}_E=|\mathrm{disc}\,E|^{1/[E:\Q]}. \] The shape of $\log\Omega$ is that of the Chowla--Selberg constant of~$K$. The archive proves Kronecker's first limit formula \cite[Thm~1]{DIT} averaged over the class group of~$K$ (\cref{tab:known}), as \lean{classMean_log_norm_disc_chowlaSelberg}, in the form \[ \frac1{h_K}\sum_{c}\log\bigl(N(\mathfrak a_c)^6|\Delta(L_c)|\bigr)=12\log2\pi-6\log d+6\gamma-6\frac{L'}L(1,\chi_{-d}) \] for $K=\Q(\sqrt{-d})$ of discriminant $-d$, where $d>4$ is the host prime of a datum (so that $d\equiv7\bmod16$), $L_c=\iota(\mathfrak a_c)$ is the embedded ideal of a representative of the class~$c$, and $\Delta(L_c)$ is its Weierstrass discriminant. Chowla and Selberg obtain this class sum from the factorization of $\zeta_K$ and then express $L'(1,\chi_{-d})$ through values of~$\Gamma$ \cite[\S12, pp.~109--110]{ChowlaSelberg}. The functional equation of $L(s,\chi_{-d})$ gives $\gamma+\log(2\pi/d)-\frac{L'}L(1,\chi_{-d})=\frac{L'}L(0,\chi_{-d})$, so that $\log\Omega=\frac{L'}L(0,\chi_{-d})-\log(\mathrm{rd}_E/d)$. \locs{\lean{logOm}~\file{LRC/DataParams}$_2$\file{.lean:224} \locsep \lean{Host.v}~\file{LRC/DataHost.lean:140}--\file{143} \locseplast \lean{classMean_log_norm_disc_chowlaSelberg}~\file{AVT/Proofs/DivisorProofs/ClassMeanLogNormDiscChowlaSelbergStatement.lean:9}} \lean{ChainTied} is produced by Alp\"oge's \lean{LinkGLOBAL}, which says that for a family satisfying the bundle of conditions \lean{ChosenE}, among them the bounds $|G(\chi,x)|\le x^{1/2+\varepsilon}$ and $|\frac{L'}L(1,\chi)|\le C\log x$ for $\chi=\chi_{-q}$, $\chi_{-d}$ and $\chi_{qd}$, the five conditions \lean{LinkG}, \lean{LinkS}, \lean{LinkIRR}, \lean{LinkDIV}, \lean{LinkMu} (\lean{GlobalTheoremInputs}) give constants and $x_0$ such that \lean{ChainTied} holds for every $x\ge x_0$ in the family's windows. The five conditions are bounds from above and below on the gains at the inert primes (\lean{LinkG}), near-orthogonality of the column spaces (\lean{LinkS}), a bound on the defect of the slope inequality (\lean{LinkIRR}), the size bound with the inclusion $\mathrm{Bk}_\omega(N_\omega)\subset RR^\vee$ (\lean{LinkDIV}) and a comparison of the slope of each $N_\omega$ with that of a reference lattice and the gains (\lean{LinkMu}). For the families used, \lean{link_gl} proves \lean{LinkGLOBAL} under two size conditions, from the condition \lean{Duty} on every family of the class, the real-place conditions, a volume ratio and a genus mass (\cref{sec:definite}). \locs{\lean{LinkGLOBAL}~\file{GTI:162} \locsep \lean{ChosenE}~\file{FSR/GlobalSchurBounds/Chosen.lean:83} \locsep the bounds~\file{FSR/GlobalSchurBounds/Chosen.lean:42} \locsep \lean{GlobalTheoremInputs}~\file{GTI:158} \locseplast \lean{link_gl}~\file{AVT/Proofs/GlobalInterpolation/LinkGl.lean:14}} \lean{LinkG} bounds the gains at the places $w$ over primes $p\le X''=x\,n\exp(x^{\varepsilon/3})$ inert in~$K$, with $f=2n$ the size of the CM type (\cref{sec:datum}). It gives $0\le\mathrm{gain}_j(w)\le4|TT|a/p$ and \[ \sum_{j}\sum_{t\in TT_w}\bigl(\textstyle\sum_{i\ \mathrm{odd}}\max\bigl(0,\,a_t/p^i-w_{j,t,p,i}\bigr)-\mathrm{lo}\bigr)-m\,\mathrm{fl}\le\sum_j\mathrm{gain}_j(w)-\tau(w), \] the inner sum over odd $i<\lfloor\log_pa_t\rfloor+2$, where the $w_{j,t,p,i}\ge1$ are real numbers given with the size data (its field \lean{wPart}), $TT_w\subset TT$ is a set of shifts, all of $TT$ when $p\ge X'$ (\lean{TTw}), $\mathrm{lo}$ and $\mathrm{fl}$ are the explicit losses \lean{loG} and \lean{flG}, and $\tau(w)$ is a loss at~$w$ that the total gain \lean{Ga} also subtracts, so that the lifts together gain at least $\sum_j\sum_t(a_t/p+a_t/p^3+\cdots)$ at~$p$, less the weights and explicit losses (\lean{vonMangoldtClassSum}). In \lean{LinkG} each lift alone is bounded only as above. \locs{\lean{LinkG}~\file{GTI:122}, \file{GTI:138} \locsep \lean{loG}, \lean{flG}~\file{GTI:109}, \file{100} \locsep \lean{vonMangoldtClassSum}~\file{LRC/DataParams}$_2$\file{.lean:240} \locseplast \lean{wPart}~\file{LRC/DataParams}$_2$\file{.lean:178}--\file{180}} \lean{LinkMu} says $\bigl|\mu(N_\omega)-\mu(N^H_\omega)-\sum_w\mathrm{sz}_w\mathrm{sg}_w\bigr|\le CX''$ (\lean{NprimeMuR}). Here $N^H_\omega$ is a reference lattice, $w$ runs over the finite set of places of $LL$ given with the size data, $\mathrm{sz}_w=2e_w\log N\mathfrak P_w/[LL:\Q]$ with $e_w$ the ramification index of $\mathfrak P_w$ over~$\Q$, and $\mathrm{sg}_w=\sum_{j=1}^m\bigl(\mathrm{gain}'_j(w)+\mathrm F_j(w)-\mathrm E_{\omega,j}(w)\bigr)$. In this sum $\mathrm{gain}'_j(w)$ is $\mathrm{gain}_j(w)$, replaced by~$0$ when the prime below $w$ lies in a set of dropped primes given with the size data, and $\mathrm F_j(w)$ and $\mathrm E_{\omega,j}(w)$ are real numbers given with the size data (its fields \lean{F} and \lean{E}), which the conditions \lean{FDef} and \lean{EDef} equate with the explicit \lean{Fval} and \lean{Eval}; $\sum_j\sum_w\mathrm{sz}_w\mathrm F_j(w)$ is the $F_o$ of \eqref{eq:size}. By definition the lattices carry the gains. The lattice $N_\omega$ is spanned by the $ay$ with $a\in I_\omega=\prod_w\mathfrak P_w^{-\lfloor e_w\,\mathrm{sg}_w\rfloor}$ and $y$ in the base lattice $N^H_\omega$ (\lean{NprimeDef}), and \lean{LinkMu} follows from this and \lean{ZeroAbove} when the $N_\omega$ are finitely generated, span their ambient spaces and have positive rank, and $\log T_\omega$ and $\log D_\Gamma$ (\cref{sec:datum}) are at most a constant times $X''$ (\lean{linkMu_of_linkDIV}). \locs{\lean{LinkMu}~\file{GTI:74} \locsep \lean{NprimeMuR}~\file{GTI:69} \locsep \lean{NprimeDef}~\file{LRC/DataDIVDefs.lean:17}--\file{24} \locsep \lean{linkMu_of_linkDIV}~\file{FSR/AdmissibleParameterJoins/SumLogPrimeFactorsLe.lean:775} \locsep \lean{sz}, \lean{gainD}, \lean{sg}~\file{LRC/DataParams}$_2$\file{.lean:205}, \file{212}, \file{215} \locsep $e_w$~\file{LRC/DataArith.lean:39}--\file{42} \locsep \lean{FDef}~\file{LRC/DataForced.lean:134} \locseplast \lean{EDef}, \lean{Eval}~\file{LRC/DataErr.lean:171}, \file{159}} \lean{LinkS}, the near-orthogonality, says that if $C_0^{-1}\le d_\omega\Gamma^2/D^*\le C_0$ for every $\omega$, where $d_\omega$ is the dimension of the stage-zero column space of~$\omega$ and $\Gamma=\exp(Cm^5fx^\varepsilon)$, then the kernel between the column spaces of two words $\omega\ne\omega'$ has $\|K_{\omega\omega'}\|_{HS}^2\le(1+\eta_0)d_\omega d_{\omega'}/D^*$, the diagonal defect is at most $e_0d_\omega$ and the total dimension at most $(1+\eta_1)D^*$, with $\eta_0,\eta_1,\sqrt{e_0}\le\exp(-x^{\varepsilon/8})$. \lean{LinkIRR} bounds the defect $\delta$ of \lean{SlopeIneq}, given by $D_{\mathrm{rows}}=(1-\delta)D_{\mathrm{all}}$ with $D_{\mathrm{rows}}$ the dimension of the space of rows and $D_{\mathrm{all}}$ the total dimension just bounded (the datum's fields \lean{Drows}, \lean{Dall}), by $\exp(-x^{\varepsilon/(7J)})$, with $J\ge1$ a constant of the family. Here $D^*$ is the product of a volume, an index of level subgroups at the two auxiliary primes and the formal degrees of the real discrete series of the weights (\lean{DstarD}; \lean{dsigTie}). \locs{\lean{LinkS}~\file{GTI:145}--\file{152} \locsep \lean{LinkIRR}~\file{GTI:55} \locsep \lean{DstarD}~\file{FSR/GlobalSchurBounds/GeneralLinearActions/GapAtLeast.lean:97} \locsep \lean{dsigTie}~\file{LRC/DataTies.lean:17} \locseplast \lean{Drows}, \lean{Dall}~\file{LRC/DataRows.lean:66}--\file{71}} \lean{LinkDIV} carries the size bound \eqref{eq:size}, with the inclusion $\mathrm{Bk}_\omega(N_\omega)\subset RR^\vee$, the height bound $h(\omega)$ for $\mathrm{Bk}_\omega\ne0$, a bound $x^{C_H}b$ for $s+\mu_{\max}(RR)$, with $s$ the mean of the slope inequality above and $C_H$ a real parameter of the condition, a condition \lean{ZeroAbove} on the gains above $X''$, ties between lifts and words, and the identity \lean{BesselId} (\cref{sec:onelift}). It is used in the form \lean{LinkDIVt}, which derives it from \lean{LinkEta} and \lean{DIVinN}. \lean{DIVinN} bounds the loss at the twisted places, those at which the tuple $\mu_w$ of Hecke characters is not the reference tuple~$\mu_0$ (\lean{etaSum}), by $D_0Cx^\varepsilon(B_0+N_c)$, with $D_0$ the common value of the dimensions $d_{Z,0}(j)$ of \lean{LinkN} (\lean{SmallLevelRankEq}). \locs{\lean{LinkDIV}~\file{GTI:77}, \file{GTI:91} \locsep \lean{LinkDIVt}~\file{GTI:94} \locsep \lean{LinkEta}~\file{GTI:49} \locseplast \lean{DIVinN}~\file{GTI:62}} \subsection{The datum}\label{sec:datum} Each member of the family of \cref{sec:tied} is a datum: ranks, Hermitian spaces over the host field~$E$ and a chain of kernels on finite index sets. Theta functions enter through the kernels. In the archive a datum is a \lean{D7}. Its ranks, given by \lean{Ranks}, are integers $k,m$ even and $t$ odd with $2\le m$, $m+3\le k$, $r+5\le t$ where $r=k+m-1$, $3ma'$ and $W$ skew-Hermitian of rank $a+a'$, with Weil representations attached to Hecke characters $\alpha,\beta$ of large weight difference and with $V$, $V'$ nearly definite, among the further conditions of \lean{IsTriple}. The theta integrals $I(\Phi)(h)=\int_{[U(V)]}\theta(g,h;\Phi)\,dg$ and $I'(\Phi')(h)$ converge on the Schwartz space and on its primed subspace, $I'$ is not identically zero, and there is a constant $c\ne0$ with $I(\Phi)=c\,I'(\Phi')$ on $U(W)(\mathbb A)$ for matched $\Phi,\Phi'$ (\lean{Decay}, \lean{OneConst}, \lean{Coinv}, \lean{Match}). Removing a hyperbolic plane from $V'$ and from $W$, compatible with the lattices, with the Cartan involutions and with the chosen bases at a finite set of finite places of~$F$ (\lean{Plane.IsAdapted}), which lowers $a'$ by~$2$ and leaves $V$ unchanged, multiplies the constant by $C\,L^{S_0}(\alpha/\beta,\frac{a-a'}2+1)$ times local factors that are explicit up to three signs, the same for all fields, where $a$ and $a'$ are the ranks before the plane is removed and $L^{S_0}$ is the $L$-function of the Hecke character $\alpha/\beta$ of~$E$ without its Euler factors above a finite set $S_0$ of places of~$F$ given with the characters (\lean{StepBeta}). The constant $C\ne0$ depends only on the base, which fixes the weights of $\alpha$ and $\beta$ at the infinite places, on the spaces before and after the removal with the coordinates, vacuum vectors, Haar measures and fundamental domains chosen for them (\lean{Geo}), on the planes and on the matching maps at the infinite places; it does not depend on $\alpha$ and $\beta$, on the Weil representations or on the matching maps at the finite places. Removing a hyperbolic plane from $V$ and from $W$, which lowers $a$ by~$2$ and leaves $V'$ unchanged, where $V'$ is definite and $V$ is not, divides the constant by $C\,L^{S_0}(\alpha/\beta,\frac{a-a'}2)$ times factors of the same kind, with a constant $C\ne0$ of the same kind (\lean{StepAlpha}). None of these constants is explicit. \lean{LocalMatching} says that the local maps used in the matching exist. \locs{\lean{ThetaIntegralsAgree}~\file{AVT/BlockTheory/DivisorBase/WeilTensorModels.lean:585} \locsep \lean{IsTriple}~\file{AVT/BlockTheory/DivisorBase/WeilTensorModels.lean:388} \locsep \lean{Decay}, \lean{OneConst}, \lean{Coinv}, \lean{Match}~\file{AVT/BlockTheory/DivisorBase/WeilTensorModels.lean:483}--\file{507} \locsep \lean{StepBeta}~\file{AVT/BlockTheory/DivisorBase/WeilTensorModels.lean:559} \locsep \lean{StepAlpha}~\file{AVT/BlockTheory/DivisorBase/WeilTensorModels.lean:571} \locsep \lean{Geo}~\file{AVT/BlockTheory/DivisorBase/WeilTensorModels.lean:314} \locsep \lean{Plane.IsAdapted}~\file{AVT/BlockTheory/DivisorBase/WeilTensorModels.lean:532} \locsep $L^{S_0}$~\file{AVT/Definitions/Foundations/NotIsSplitOfIsInertAt.lean:369} \locsep weights~in~the~base~\file{AVT/BlockTheory/DivisorBase/WeilTensorModels.lean:146}--\file{148} \locseplast \lean{LocalMatching}~\file{AVT/BlockTheory/DivisorBase/WeilTensorModels.lean:581}} The inputs are the Siegel--Weil formula, as \lean{SiegelWeilStd}, and a weak form of the meromorphic continuation of Siegel Eisenstein series on $U(n,n)$ (\lean{EisBL}). \lean{SiegelWeilStd} says that under its hypotheses, for $A$ definite of rank $m$, $Y$ of rank $n$ and a $K$-finite Schwartz function, the Siegel Eisenstein series on $U(n,n)$ at $s_0=(m-n)/2$ is $\kappa$ times the normalized average of the theta function over $[U(A)]$, with $\kappa=1$ for $n2n$ \cite{Weil65}, to Ichino for $n0$ (\lean{divinN_of_hypV}). \locs{\lean{HypV}~\file{CB/DivinNOfHypV.lean:10}--\file{11} \locsep \lean{vvRecord}~\file{CB/DivinNOfHypV.lean:17} \locseplast \lean{divinN_of_hypV}~\file{CB/DivinNOfHypV.lean:117}} \subsection{The covolume bound}\label{sec:covolume} The slope of $RR^\vee$ in \eqref{eq:size} is bounded through the covolumes of lattices of automorphic forms. The archive states the following as \lean{CovolumeBoundOfX} and proves it as \lean{topCovolSlim2_use_alt_alt} from the fifteen covolume inputs listed in \cref{sec:known}, all theorems of the archive. \locs{\lean{CovolumeBoundOfX}~\file{TPS/LiftLambdaBound.lean:60} \locseplast \lean{topCovolSlim2_use_alt_alt}~\file{TPS/TopCovolSlim2Use.lean:16}} \begin{theorem}[Alp\"oge]\label{thm:covol} For all small $\varepsilon$ there are largeness thresholds such that, for every family satisfying \lean{FamHyp} and every $C_H$ that together meet them, there is $C_0$ such that, for all large $x$, every member and every point at the member (defined below), every $C\ge C_0$ and every $j=1,\dots,m$, \[ -\mu(M_j)\le\sum_\rho(p_j-q_j)\,e_j(\rho)\log\Omega+C\,fmb\,x^\varepsilon\log\log x . \] \end{theorem} Here $W_j$ is the $j$-th seesaw space, of signature $(p_j,q_j)=(k+j-1,m-j)$; $M_j$ is the row lattice in its frame on $U(W_j)$, which the archive compares with the lattice of integral automorphic forms of the good Hecke type (\lean{RecAt}); $\rho$ runs over an index set of $f$ elements, each sent by the archive to an element of the CM type~$\Phi$ (\lean{RPl}, \lean{tauPl}), and $e_j(\rho)$ is the $j$-th entry of the part $e$ of the datum's Harish-Chandra parameter there (\lean{HCParam}); a \emph{point} at a member is the archive's record of the choices made there (\lean{PtAt}), and the theorem is about the points that satisfy the conditions \lean{RowsTAlt}; and $\mu=-\mathrm{lcv}/\mathrm{rank}$ is the archive's slope (\lean{LcvMu}). The bound enters once, as the field \lean{l3}, of type \lean{L3}, used by \lean{IndefSide.boundsMu}, which bounds $\mu(RR^\vee)$ by the sum over the lifts of these bounds, the theta indices, ramification terms and errors. \locs{\lean{RowsTAlt}~\file{TPS/LiftLambdaBound.lean:17}--\file{19} \locsep \lean{PtAt}~\file{AVT/Definitions/Divisors/TupleReadBaseTupleEqFamily.lean:145} \locsep main~term~\file{FLG/Divisibility/ZscOfNum.lean:64}--\file{66} \locsep \lean{RPl}~\file{FLG/Divisibility/Numbers.lean:13}, \file{160}--\file{163} \locsep \lean{tauPl}~\file{LRC/DataForced.lean:15}, \file{FLG/Divisibility/NumbersLemmas.lean:604}--\file{613} \locsep \lean{HCParam}~\file{LRC/Params.lean:50} \locsep \lean{RecAt}~\file{AVT/Definitions/Divisors/CovolXOfCovol.lean:57} \locsep \lean{LcvMu}~\file{Formalism/MeanHeightDescent/MuMaxMinkowski.lean:13} \locsep \lean{L3}~\file{FLG/IndefiniteForms/IndefSide.lean:315} \locsep \lean{IndefSide.boundsMu}~\file{FLG/IndefiniteForms/IndefUse.lean:271} \locsep the~sum~\file{FLG/IndefiniteForms/IndefUse.lean:286}--\file{288} \locseplast \lean{FamHyp}~\file{AVT/Definitions/Divisors/TupleReadBaseTupleEqFamily.lean:503}} For $j0$ satisfying \lean{WeightHyp}$(u,g)$ and every cutoff $B\ge128$ with $B+1>\mathrm{bigB}_0(u,g)$, where $\mathrm{bigB}_0(u,g)=4f(p+q)^2(1+\|u\|/g)$ with $\|u\|$ the norm of $u$ in the archive's weight space \textup{(}\lean{bigB0}\textup{)}, the mean slope satisfies the bound \lean{MeanSlope} \[ \frac1{\#\mathrm{Comp}}\sum_c\widehat h_{B,c}(u)\ge-C\Bigl[\Bigl(\log(B\,\mathrm{rd}_E)+\Bigl|\frac{L'}L(1,\chi_{-d})\Bigr|\Bigr)\ell_1(u)+q\,\mathrm{spread}(u)\Bigr]. \] \end{theorem} Here $\widehat h_{B,c}(u)$ is the archive's slope of the weight $u=(u_{\tau,i})$, $\tau$ in the CM type and $i16a^2$; at $w\mid\ell$, given the archive's tied period data (\lean{LocTied}) and a compatible local bridge, $r$ is crystalline when $\pi_w$ is spherical, and potentially semistable in the abelian and tame cases, with its Weil--Deligne representation compared with $\mathrm{rc}(\sigma)$ as at $w\nmid\ell$; in these three cases the multiset gives the Hodge--Tate jumps at the embeddings induced by the bridge. No de Rham assertion is made at the other $w\mid\ell$. \locs{\lean{Row0b}~\file{DA/Foundations/GaloisRepresentations/TypeConditions/WeightsConjBis.lean:69} \locseplast $\sigma$~\file{DA/Foundations/GaloisRepresentations/TypeConditions/WeightsConjBis.lean:59}} For polarized $\pi$ the published theorem gives local--global compatibility up to Frobenius semisimplification at $w\nmid\ell$ \cite[Thm~2.1.1(2)]{BLGGT} and, when $\pi_w$ has an Iwahori-fixed vector, at $w\mid\ell$ \cite[Thm~2.1.1(4)]{BLGGT}. For cohomological conjugate self-dual $\pi$ of rank at least~$2$, Shin proves existence with a slight regularity hypothesis in even rank \cite[Thm~1.2]{Shin}, which Chenevier and Harris remove \cite[Thm~3.2.3]{ChenevierHarris}. For such $\pi$ Caraiani proves the same compatibility at $w\mid\ell$ \cite[Thm~1.1]{Caraiani14} and at all $w\nmid\ell$ \cite[Thm~1.1]{Caraiani}. Taylor and Yoshida had proved the compatibility at $w\nmid\ell$ when $\pi$ is square integrable at a finite place \cite[Thm~1.2]{TaylorYoshida}. \lean{Row0b} is a weak form of these theorems. At a place $w\nmid\ell$ where $\pi_w$ is unramified it says as much as they do, since the laws of \lean{IsRecT} make $\mathrm{rc}(\sigma)$ diagonal with the inverse Satake parameters of~$\sigma$ (\lean{SphLaw.sat}), so that $r$ is unramified at~$w$ with these Frobenius eigenvalues (\lean{AwayB}, field \lean{tor}). \locs{\lean{SphLaw.sat}~\file{AVT/Proofs/InertiaEigenvalues/IsSphericalIffUnrOfLawsT.lean:73}} The archive proves it as \lean{galoisLocalGlobalStrong_recParameter} from two theorems, combined by \lean{galoisRepresentations_of_thetaLift}. The first, \lean{GeoOne}, gives for restricted data of odd rank auxiliary Galois representations~$\rho$, continuous and semisimple, such that at almost all split $w\nmid\ell$ over primes $p$ with $-D$ a square in $\Q_p$, for an integer $D$ with $-D$ a square in~$E$, at which the data have Satake parameters, $\rho$ is unramified and $\rho(\mathrm{Frob}_w^{-1})$ is annihilated by the polynomial whose roots are the normalized Satake parameters (\lean{VBase}, \lean{ESAt}). In its proof they are twists of graded pieces of the Galois action on a common eigenspace of finite \'etale correspondences in the $\ell$-adic cohomology of a scheme (\lean{RepEx}, \lean{HTower}). The second, \lean{TLiftK}, gives under its hypotheses, for an automorphic representation of the standard definite unitary group $U(n)$ and every $d$ beyond a threshold, one of $U(d)$ with the weights of a theta lift and, at each split place, local components linked by a nonzero map from the Weil representation. \locs{\lean{galoisLocalGlobalStrong_recParameter}~\file{.../IntegralStructures/RowSrc.lean:6} \locsep \lean{galoisRepresentations_of_thetaLift}~\file{.../GaloisRepresentationsOfThetaLift.lean:6} \locsep \lean{RepEx}~\file{.../GeometricFlagBases/HodgeTrAt.lean:103} \locsep \lean{HTower}~\file{AVT/Definitions/Divisors/IdTower.lean:9} \locsep \lean{TLiftK}~\file{.../HalfIndexLattices/TLiftK.lean:12} \locseplast proved~\file{.../ThetaKernels/TLiftK.lean:4}} Input~(ii), \lean{RecTExists}, asserts at every finite place a local Artin map and a map from irreducible representations of $\mathrm{GL}_n$ to Weil--Deligne representations that is compatible with local class field theory on characters, with the Langlands classification, with twists by real powers of the norm and with the absolute value of the central character. Compatibility with other smooth twists, central characters and duals, injectivity, and the passage from depth zero to tame parameters are asserted only at representations whose supercuspidal support consists of characters, or of twisted depth-zero cuspidals when the field is some $\Q_p$. It gives depth-zero cuspidals over $\Q_p$ their Green-parametrized inertia and zero monodromy, and it has no $L$- or $\varepsilon$-factors. The archive proves it as \lean{recTExists} (\cref{sec:top}) from local class field theory, Zelevinsky's classification \cite[Thm~6.1]{Zelevinsky} and the depth-zero parametrization over $\Q_p$. \locs{\lean{RecTExists}~\file{DA/Foundations/LocalLanglands/IsRecTOfIsRec.lean:78}} The map it exhibits satisfies \lean{IsRecT} and, for a finite place $w$ of a number field $E$, sends all supercuspidal representations of $\mathrm{GL}_n(E_w)$ outside \lean{GoodCusp}, the class of the characters and, when $E_w$ is some $\Q_p$, of the twists by smooth characters of representations of depth zero, to real unramified twists of a single parameter $J_{w,n}$ (\lean{cuspEx}). Each $J_{w,n}$ is irreducible of dimension $n\ge2$ and, when $E_w$ is some $\Q_p$, not a twist of a tame parameter (\lean{JunkEx}). When $\sigma=\pi_w\otimes|\det|^{(1-a)/2}$ is supercuspidal and outside \lean{GoodCusp}, \lean{Row0b} asks nothing of $r|_{G_w}$, whichever map satisfying \lean{IsRecT} is taken: the laws make $\mathrm{rc}(\sigma)$ irreducible of dimension $a\ge2$, which excludes the abelian case (\lean{dim_one_of_dvd}), and when $E_w$ is some $\Q_p$ a supercuspidal representation with tame parameter has depth zero (\lean{dz_of_cusp}) and so lies in \lean{GoodCusp}, which excludes the tame case. The exhibited map is not the local Langlands correspondence \cite{HarrisTaylor,Henniart}. \locs{\lean{GoodCusp}~\file{DA/Foundations/LocalLanglands/IsRecTOfIsRec.lean:20} \locsep \lean{cuspEx}~\file{DA/Foundations/LocalLanglands/LocalCentralUnits/CuspEx.lean:168} \locsep $J_{w,n}$~\file{DA/Foundations/LocalLanglands/LocalCentralUnits/CuspEx.lean:176}--\file{177} \locsep \lean{JunkEx}~\file{DA/Foundations/LocalLanglands/RecTTransport.lean:244} \locseplast \lean{dim_one_of_dvd},~\lean{dz_of_cusp}~\file{DA/Foundations/GaloisRepresentations/IntegralStructures/InertCharpoly/CharPolOfInert.lean:39},~\file{114}} A full form, with existence and uniqueness of a map satisfying all the laws, bijectivity, $\varepsilon$-factors of pairs and compatibility between fields, but no $L$-factors, is stated as \lean{LocalLanglandsGL} and proved only conditionally, on \lean{LitLoc} and \lean{LitAut}. It is carried as a hypothesis, directly or derived conditionally from fields of a bundle \lean{LitN} of about $65$ published theorems, among them \lean{llc}. Since \lean{everyCharacter} has no hypotheses and the archive has no unconditional proof of \lean{LocalLanglandsGL}, the five theorems do not rest on it. \locs{\lean{LocalLanglandsGL}~\file{AVT/PublishedTheorems/AutomorphicForms}$_2$\file{/LocalLanglandsGL.lean:96} \locsep proved conditionally~\file{AVT/Proofs/IwasawaDecomposition/LocalLanglandsOfLiterature.lean:21} \locsep \lean{LitN}~\file{AVT/Definitions/AutomorphicMembership/Brauer.lean:133} \locseplast \lean{llc}~\file{AVT/Definitions/AutomorphicMembership/Brauer.lean:76}} \subsection{A specialization over \texorpdfstring{$\Q$}{Q}}\label{sec:vandiver} The estimate \lean{CorGK} behind Count~I of the plus part also gives, over~$\Q$, for positive even~$e$ and primes $\ell\ge5$ with $\ell-1\nmid2e$, the vanishing of the $\omega^{-e}$-eigenspace of the class group of $\Q(\zeta_\ell)$ modulo~$\ell$, with $\omega$ the Teichm\"uller character of $\Gal(\Q(\zeta_\ell)/\Q)$, once $\log\ell>C(e+1)^A$, for constants $C$ and $A$. The specialization is off the chain. Specialized to $K=\Q(\sqrt{-3})$, $N=2$ and $\eta=1$, the estimate \lean{CorGK} becomes Alp\"oge's \lean{UpperTheory.V.SmallClassNumbers.bound}, which says that there are constants $C,A$ such that for every positive even~$e$ and every finite set of primes $\ell\ge5$ with $\ell-1\nmid2e$, \[ \sum_\ell\log\ell\cdot\mathrm{length}\,S_\ell(e)\le C(e+1)^A \] (for coefficients $\Z_\ell$; the statement allows finite extensions, with normalized lengths), where $S_\ell(e)\subset H^1(\Q,\Q_\ell/\Z_\ell(-e))$ is the group of classes unramified everywhere and trivial at~$2$ and~$\ell$. \locs{\lean{UpperTheory.V.SmallClassNumbers.bound}~\file{AV/SmallClassNumbers/CorGKOf.lean:27} \locseplast statement~\file{AV/SmallClassNumbers/ClassNumberThree.lean:16}} The statement \lean{Large}, which follows from it by \lean{large_of}, says that $S_\ell(e)=0$ when $\log\ell>C(e+1)^A$. This is the vanishing of the $\omega^{-e}$-eigenspace of the class group of $\Q(\zeta_\ell)$ modulo~$\ell$. With $n=e+1$, Soul\'e proved such vanishing for $\log\ell>n^{224n^4}$ and odd $n\ge5$ \cite{Soule99}, and Kurihara for $n=3$ with no condition on~$\ell$ \cite{Kurihara}. The statement \lean{Bound} also asserts that $S_\ell(e)$ is finite for positive even $e$, $\ell\ge5$ and $\ell-1\nmid2e$. Since $C$ is not explicit, the vanishing test $\log\ell>C(e+1)^A$ is not effective at a given prime. \locs{\lean{Large}~\file{AV/SmallClassNumbers/LenEqZeroIff.lean:65} \locsep \lean{large_of}~\file{AV/SmallClassNumbers/LenEqZeroIff.lean:70} \locsep \lean{Bound}~\file{AV/SmallClassNumbers/ClassNumberThree.lean:16} \locseplast $S_\ell(e)$~finite~\file{AV/SmallClassNumbers/ClassNumberThree.lean:19}} \section{Classical inputs and their archive forms}\label{sec:known} \Cref{tab:known} sets the classical inputs on the chain beside their archive forms, all proved in the archive, with two further statements that the archive does not prove unconditionally. Most entries open with one word for how the two compare: ``Full'' for the published statement, ``Special'' for a special case, ``Weak'' for a weaker statement and ``Different'' for a statement of another shape. Green's parametrization (\cref{sec:top}), the slope inequality (\cref{sec:tied}), \lean{LitThin} and the mass formulas (\cref{sec:definite}) are also proved in the archive. The statements of resolution, of principalization and of the Riemann hypothesis over finite fields are on Mathlib's schemes and their morphisms, but blowups, defined by their universal property, simple normal crossings and \'etale cohomology are the archive's own definitions. {\scriptsize \begin{longtable}{@{}>{\raggedright\arraybackslash}p{0.17\textwidth}>{\raggedright\arraybackslash}p{0.41\textwidth}>{\raggedright\arraybackslash\scriptsize}p{0.36\textwidth}@{}} \caption{Classical inputs on the chain, and two statements that the archive does not prove unconditionally.}\label{tab:known}\\ \toprule Theorem & In the archive & Declaration\\ \midrule \endfirsthead \toprule Theorem & In the archive & Declaration\\ \midrule \endhead Local Langlands for $\mathrm{GL}_n$ \cite{HarrisTaylor,Henniart} & Weak (\cref{sec:weak}): a map with the rank-one and segment laws and Green-parametrized depth-zero inertia and zero monodromy over~$\Q_p$; the laws hold for a map sending the supercuspidals of $\mathrm{GL}_n(E_w)$, $w$ finite, outside \lean{GoodCusp} to real unramified twists of a single parameter $J_{w,n}$; no bijectivity, no $L$- or $\varepsilon$-factors & \lean{recTExists}, \file{DA/Foundations/LocalLanglands/LocalCentralUnits/RecTExists.lean:20}\\\addlinespace Galois representations over CM fields \cite{BLGGT,Shin,ChenevierHarris,Caraiani,Caraiani14} & Weak (\cref{sec:weak}), against $\mathrm{rc}(\sigma)$, $\sigma=\pi_w\otimes|\det|^{(1-a)/2}$: at $w\nmid\ell$, equality when $\mathrm{rc}(\sigma)$ is abelian with $N=0$ (so at every unramified $\pi_w$), and $N=0$ and the characteristic polynomials on inertia when it is tame with $N=0$, its eigenvalues of multiplicity at least two on a generator modulo wild inertia have order dividing $Nw-1$, and $E_w\cong\Q_p$ for a prime $p$, with $a=1$ or $p>16a^2$; at $w\mid\ell$, given tied period data and a compatible bridge, crystalline when $\pi_w$ is spherical, the comparison in the abelian and tame cases, and in all three cases the prescribed Hodge--Tate jumps at every embedding induced by the bridge & \lean{galoisLocalGlobalStrong_recParameter}, \file{DA/Foundations/GaloisRepresentations/IntegralStructures/RowSrc.lean:6}\\\addlinespace Resolution of singularities in characteristic $0$, with principalization \cite{Hironaka} & Resolution without the assertion that the centers are smooth; principalization with smooth centers & \lean{resolutionCharZero}, \file{WB/CoreStatements/ResolutionCharZero/ResolutionCharZero.lean:6}; \lean{principalisationCharZero_closed}, same folder, \file{PrincipalisationCharZero}\allowbreak\file{Closed.lean:7}\\\addlinespace Arithmetic Hilbert--Samuel \cite[Thm~8]{GilletSoule}, \cite[Thm~(1.4)]{Zhang} & Different: \lean{HSUse} gives upper bounds and the arithmetic Kleiman inequality; the asymptotic for nef bundles is proved separately, as \lean{ArithHilbertSamuel}, the field \lean{q9} of the Hodge-bundle inputs (\file{TPS/EveryCharacterOfLiteratureOfVWin.lean:803}) & \lean{hsUse_outright}, \file{WB/CoreStatements/AutomorphicKernels/MoriwakiBigness/MoriwakiBigTheorem/Outright.lean:13}; \lean{ArithHilbertSamuel}, \file{AVT/PublishedTheorems/Divisors}$_2$\file{/SmallSecTensTriv.lean:14}\\\addlinespace Arithmetic bigness \cite[Thms~A, B]{Moriwaki}, \cite[Thm~2.2]{Yuan} & For $X$ in a product of projective spaces with the restricted metric, under the stated semipositivity and nefness assumptions: the arithmetic volume is at least the top self-intersection & \lean{moriwakiBig_outright}, \file{WB/CoreStatements/AutomorphicKernels/MoriwakiBigness/MoriwakiBigTheorem/Outright.lean:9}\\\addlinespace Siegel--Weil for unitary dual pairs \cite{Weil65,Ichino,Yamana} & Special: $V$ definite, with self-dual lattices and a $K$-finite Schwartz function; $E/F$ unramified at all finite places and Galois over~$\Q$, its inert places not above~$2$ and of norm $>3$, the additive character unramified there & \lean{siegelWeilStd}, \file{WB/CoreStatements/AutomorphicKernels/SiegelWeilFormula/SiegelWeilStandard/SiegelWeilStd.lean:8}\\\addlinespace Continuation of Siegel Eisenstein series on $U(n,n)$ \cite{Langlands} & Weak: meromorphic continuation after pole-clearing, against the specified functionals, with the moderate-growth bound; no functional equation; over the base of the Siegel--Weil row & \lean{eisBL}, \file{WB/RepresentationTheoryGL/EisensteinSeries/SiegelEisensteinHalfPlane/SchwartzBruhatMajorants/SiegelMultOf.lean:101}\\\addlinespace Kronecker's limit formula, summed over the classes \cite[Thm~1]{DIT}, \cite[\S12, pp.~109--110]{ChowlaSelberg} & Special: for $\Q(\sqrt{-d})$ with $d$ the host prime, $d\equiv7\bmod16$, in terms of ideal lattices and $\frac{L'}L(1,\chi_{-d})$ (\cref{sec:tied}); no evaluation by values of~$\Gamma$ & \lean{classMean_log_norm_disc_chowlaSelberg}, \file{AVT/Proofs/DivisorProofs/ClassMeanLogNorm}\allowbreak\file{DiscChowlaSelberg}\allowbreak\file{Statement.lean:9}\\\addlinespace Discrete series \cite{HarishChandra} and lowest $K$-types \cite{HechtSchmid} & Special: $U(p,q)$ only; the lowest $K$-type with multiplicity one, and a bound on the other $K$-types that allows all positive roots, not only the noncompact ones & \lean{discreteSeriesClassification}, \file{CompanionTheory/RealGroups/DiscreteSeries/DiscreteSeries}\allowbreak\file{Classification/DiscreteSeries}\allowbreak\file{Exhaustion/DsExhaust.lean:17}\\\addlinespace Ferrero--Washington \cite{FW} & Bounded $\ell$-rank up the cyclotomic tower, for abelian $K$ containing $\mu_\ell$ ($\mu_4$ if $\ell=2$) & \lean{FerreroWashington}, \file{AV/PublishedTheorems/FerreroWashington.lean:7}, proved as \lean{ferreroWashington}, \file{AV/GaloisLibrary/FerreroWashingtonIndex/FerreroWashington.lean:104}\\\addlinespace Kato's Euler system \cite[\S\S8--9]{Kato} and Selmer finiteness \cite[Thm~14.2(1)]{Kato} & Different: the cocycles, and the vanishing of rational Selmer groups of $W$ and of a twist $W\otimes\theta$ by the quadratic character of an imaginary quadratic field, under \lean{Used}, a newform attachment and \lean{WitAQ} & \lean{zetaUr}, \file{AV/ModularityArgument/KatoQuadraticCharacters/UnramifiedZeta/ZetaUr.lean:9}; \lean{KatoTwoW}, \file{.../SphericalDecomposition/KatoTwoW.lean:10}; \lean{katoTwoW_of_printsH}, \file{AV/ModularityArgument/ThresholdArguments/DenseThresholdCases/KatoTwoWOfPublishedH.lean:48}\\\addlinespace Poitou--Tate \cite[(8.6.10)]{NSW} & Special: finite modules killed by $\ell^r$, at the allowed places, $\ell$ odd or the field totally complex & \lean{poitouTateNormalised_holds}, \file{WB/CoreStatements/PublishedTheorems/PoitouTateDuality/PoitouTateNineTerm/PoitouTateNormalisedHolds.lean:18}\\\addlinespace Local Tate duality and local Euler characteristic \cite[(7.2.6), (7.3.1)]{NSW} & Special: perfect cup-product pairings with an isomorphism $H^2(\mu_{\ell^r})\cong\Z/\ell^r$ whose normalization is not specified, and the Euler characteristic formula, for finite modules killed by $\ell^r$ at the finite places & \lean{localTateDuality}, \file{AVT/Proofs/DivisorProofs/ExistsMulEquiv.lean:115}; \lean{loceuler_holds}, \file{AVT/Proofs/DivisorProofs/LoceulerHolds.lean:8}\\\addlinespace Artin reciprocity \cite{Artin} & Full & \lean{globalArtinReciprocity_holds}, \file{AVT/Proofs/IwasawaDecomposition/GlobalArtinReciprocityHolds.lean:5}\\\addlinespace Chebotarev's density theorem \cite{Chebotarev} & Full, in Dirichlet density & \lean{hasDirichletDensity_chebotarevSet}, \file{GroundTheory/GlobalFields/TateThesis/Chebotarev.lean:198}\\ \midrule The Riemann hypothesis over finite fields \cite{Weil48}, \cite[Thm~(1.6)]{Deligne} & Stated, and no other declaration names them: Deligne's purity bound for smooth projective varieties of every dimension (Weil's for curves) & \lean{WeilOneD}, \lean{WeilOneP}, \file{DA/CommutativeAlgebra/EtaleCohomology/WeilOneD.lean:17,21}\\\addlinespace Local Langlands, the full correspondence & Stated; proved conditionally on \lean{LitLoc} and \lean{LitAut} (\lean{localLanglandsGL_and_prN_of_lit}), not unconditionally; the main route uses \lean{RecTExists} & \lean{LocalLanglandsGL}, \file{AVT/PublishedTheorems/AutomorphicForms}$_2$\allowbreak\file{/LocalLanglandsGL.lean:96}; \file{AVT/Proofs/IwasawaDecomposition/LocalLanglandsOfLiterature.lean:21}\\ \bottomrule \end{longtable}} At the call site, the covolume inputs of \cref{sec:covolume} are: (1)~resolution and (2)~principalization in characteristic~$0$; (3)~an integrated Poincar\'e--Lelong formula; (4)~arithmetic Hilbert--Samuel bounds with arithmetic Kleiman and (5)~arithmetic bigness, as in \cref{tab:known}; (6)~under its hypotheses, among them mixed characteristic, for a $p$-divisible group, $\ker(\mathrm{HT})=(\ker\mathrm{dlog})^\perp$ under the Weil pairing (middle exactness of the rational Hodge--Tate sequence is \cite[Prop.~4.3.6]{ScholzeWeinstein}), which the archive derives from its form of prismatic Dieudonn\'e theory over $\mathcal O_C$, a functor from $p$-divisible groups to finite free $A_{\inf}$-modules with Frobenius and Hodge filtration (\lean{PrismaticDieudonneHodge}, by \lean{farguesExact_of_hodge}); (7)~Zarhin's trick \cite[Lemma~2.5]{Zarhin}, in the form that every polarized abelian scheme over such a valuation ring is a retract of a principally polarized one; (8)~asymptotic Riemann--Roch for nef weights; (9)~toroidal compactification of Siegel modular varieties over $\Z[\zeta_n,1/n]$ for odd $n\ge3$ \cite{FaltingsChai} and (10)~of unitary ones on the generic fiber; (11)~seminegativity of the Lie algebra of a semi-abelian scheme over a smooth projective geometrically connected curve in characteristic~$0$; (12)~under positivity and klt hypotheses, a section vanishing at exactly one of two given points (compare \cite[Thm~5.9]{Kollar}; simultaneous surjectivity onto both fibers, for smooth varieties, is \cite[Thm~0.3, Cor.~0.4]{AngehrnSiu}); (13)~birational invariance of arithmetic degrees, with the Chern--Weil degree formula; (14)~$28$ further inputs to the Hodge-bundle argument, among them Kawamata--Viehweg vanishing for a nef and big line bundle on a smooth projective variety over an algebraically closed field of characteristic~$0$ (\lean{KawamataViehweg}, field \lean{q6}), which the archive proves from Nadel vanishing for pairs over~$\C$ (\lean{NadelVanishingPair}; \lean{kawamata_viehweg_of_literature}); and (15)~deformation of abelian schemes and their homomorphisms over nilpotent divided-power thickenings, given a reference lift \cite{Messing,MazurMessing}. \locs{the~call~site~\file{TPS/EveryCharacterOfLiteratureOfVWin.lean:812} \locsep \lean{PrismaticDieudonneHodge}~\file{AVT/PublishedTheorems/Divisors}$_2$\file{/InjectiveOfSmulMemRange.lean:62} \locsep \lean{farguesExact_of_hodge}~\file{WB/CoreStatements/TemperedRepresentationTheory/FarguesExactness/FarguesExactOfPrismatic.lean:56} \locseplast \lean{kawamata_viehweg_of_literature}~\file{TPS/EveryCharacterOfLiteratureOfVWin.lean:447}} Other published theorems enter the route under their usual names. The archive proves the main theorem of complex multiplication for elliptic curves with multiplication by the maximal order of an imaginary quadratic field, in idelic form \cite{Shimura71} (\lean{main_theorem_complexMultiplication}), which serves the first bundle of \cref{sec:reduction}. Base change is used in several forms, among them Arthur--Clozel's cyclic base change of prime degree over totally complex fields, for cuspidal representations of $\mathrm{GL}_a$ not isomorphic to their twists by the nontrivial characters of the extension \cite{ArthurClozel} (\lean{ArthurClozelStrongTC}), in the proof of input~(i) (\cref{sec:weak}), and weak base change from definite unitary groups to $\mathrm{GL}_n$, for a CM field unramified over its real subfield \cite{Labesse} (\lean{DefiniteN}), in the first bundle. Both theorems of Jacquet and Shalika are used over totally complex fields: strong multiplicity one for isobaric sums \cite{JacquetShalikaII} (\lean{JacquetShalikaRigidity}) and the bound $Nw^{-1/2}<\lvert\alpha\rvert{\raggedright\arraybackslash}p{0.04\textwidth}>{\raggedright\arraybackslash}p{0.23\textwidth}>{\raggedright\arraybackslash}p{0.56\textwidth}>{\raggedright\arraybackslash\scriptsize}p{0.07\textwidth}@{}} \caption{The note's steps \cite[pp.~3--4]{AlpNote} and the chain.}\label{tab:steps}\\ \toprule Step & The note & On the chain & \S\\ \midrule \endfirsthead \toprule Step & The note & On the chain & \S\\ \midrule \endhead 1 & A Stepanov determinant & Not found. & --\\\addlinespace 2 & The invariant derivation $D$ on a CM curve at an inert $p$; $p$-divisibility of $(D^a\vartheta)(0)/\Omega^a$ & \lean{IntInert} and \lean{kernel_bounds_generic}. The first is integrality of the corrected theta kernels at inert $p\ge5$ after substituting the formal logarithm, with \lean{CM2} from Artin reciprocity. The second is a bound for the derivatives at~$0$ of every power series of that form, with exponent a sum of terms $\max(a_s/p^e-c,0)$, $e$ odd, some omitted, less a loss. Together they bound the derivatives of the kernel series of the datum at the inert places of residue characteristic at most~$X''$, under conditions (\lean{EntriesAtInert}) and make the actual periods divisible by the lattice ideals. The lower bound for the gains of the lifts together is \lean{LinkG}, proved without the theta kernels. & \ref{sec:tied}, \ref{sec:divis}\\\addlinespace 3 & An identity $qN=M^k+R$ with a path quadric & In part: the unweighted count identity occurs by specialization of \lean{count_mul_card} (\file{CompanionTheory/SpectralGeometry/LocalGlobalProofs/AdelicAnalysis/FiniteFieldQuadraticFormPointCount.lean:84}); the full weighted match was not found. & --\\\addlinespace 4 & A positivity $W(1_S)\ge0$ from a determinant & Not found. The sign on the chain is that of \lean{ch_top}, by which divisibility minus size is at most an error, with ramp test functions. & \ref{sec:star-def}, \ref{sec:tied}\\\addlinespace 5 & The weighted sum $\sum_{n\le X}\Lambda(n)(X/n-1)-X\log X+(1+\gamma)X$; ``Guinand-Weil had the sign'' & In part: \lean{Gz} is exactly this function. The ``variance test function trick'' and the integral over twists were not found. & \ref{sec:ramp}\\\addlinespace 6 & $W\bigl(1_{(0,1]}+({\sim}X^\eta)\cdot1_{(0,1/X]}\bigr)\ge0$ ``gives ${>}\,1-\eta$'' (in our reading, no zeros with $\Real s>1-\eta$) & By role: Landau's theorem on windows, with exponent $\frac12+C\varepsilon$ and $\varepsilon$ taken small for each $\Real s>\frac12$. The positivity heuristic is the note's. & \ref{sec:landau}\\\addlinespace 7 & Determinants containing Gan--Gross--Prasad periods; interlacing weights & In part: \lean{comult_unitary} (\file{AVT/Proofs/FoundationProofs/BranchesAlongOfIrreducible.lean:123}), the space of intertwiners for $U(n)\subset U(n+1)$ has dimension one for interlacing weights and zero otherwise. & --\\\addlinespace 8 & Iterated theta lifts between definite unitary groups, multiplicity $m$, with $m$ like $\sqrt X$ & The column chains of a datum (\lean{D7}): $m$ kernel maps, each raising the definite rank by $t+1$; \lean{RankVal} fixes $x^{1/2-\varepsilon/16}\le m\le2x^{1/2-\varepsilon/16}$. & \ref{sec:datum}\\\addlinespace 9 & Norms of the lifts, cancellation of the terms of order $m$, divisibility; Chowla--Selberg, Weyl, Siegel's mass & Kronecker's limit formula averaged over the class group, through $\log\Omega$; ratios of Siegel masses (\lean{SiegelMassRatioCM}). The size estimates bound the left side of \Star{} above, and \cref{prop:balance} expresses each term $\ASK-\MAIN$ by centered ramp sums, which we read as the note's cancellation. & \ref{sec:tied}, \ref{sec:definite}, \ref{sec:covolume}\\\addlinespace 10 & A lower bound for the divisibility of the $m$-fold lift & \lean{LinkG} bounds the gains of the lifts together from below and is proved without the theta kernels; \lean{IntInert} and the bound for a single kernel series (\cref{sec:divis}) make the actual periods divisible by the lattice ideals (\lean{LinkDIVt}), and \lean{LinkMu} compares slopes; at the level primes, the denominator bound \lean{aP10}, of exponent $\lceil e_wmE_T\rceil$, with $E_T$ growing with~$m$ at the seen places and $0$ at the others. & \ref{sec:tied}, \ref{sec:definite}\\\addlinespace 11 & The Iwasawa tower, by analogy with a pigeonhole on $\mathbb F_{q^n}/\mathbb F_q$ & The host fields, composita with $\Q(\zeta_{2^{e+2}})^+$; the average over the subgroup $A_0$. & \ref{sec:hosts}, \ref{sec:core}\\\addlinespace 12 & Seesaw: the period of the iterated lift is a Bessel period, independent of the Hecke characters; Rallis & In part: an identity $\zeta_{\omega,t}\,\mathrm{per}_{\omega,t}(\varphi,F)=a_\omega\langle V\Theta_tF,P_\omega(U_\omega\varphi)\rangle$ between the periods and inner products with Poincar\'e series, with $a_\omega\ne0$ and $\zeta_{\omega,\mu_0}=1$ (\lean{MIstar2}), proved by passing linear relations through Bessel periods of the row forms (\lean{JointCondition}); \lean{BesselId}, the comparison of this shape for $\mathrm{Bu}_\omega$; in the adelic model, the Bessel period of the iterated lift as an iterated integral of column periods (\lean{iter_colPer}); \lean{SeeLast}; theta integrals compared through Siegel--Weil (\lean{ThetaIntegralsAgree}). The condition \lean{LocalQuadraticParts.adm1}, which states the independence in terms of $\mathrm{Bu}_\omega$, and the tie \lean{TieBm} are not on the route (\cref{sec:onelift}). & \ref{sec:onelift}\\\addlinespace 13 & Bost's slope inequality & \lean{slope_inequality_general}; in the covolume bound, the separate statement \lean{DescentSlope}. & \ref{sec:tied}, \ref{sec:covolume}\\\addlinespace 14 & The cost of each lift; stopping at $\frac23$ a priori; the need for division & By role: the point-sum bound \lean{StepThmBG}, with factor $b^{C_Bm^5f}$. The $\frac23$ and the transfer to a function field were not found. & \ref{sec:definite}\\\addlinespace 15 & An integral structure across Hecke characters, for extra averaging & The integral structure: \lean{RR_spec} (\file{FLG/IndefiniteForms/IndefUse.lean:219}) specifies every localization of the lattice $RR$, which glues the row lattices selected by the Hecke-character tuples $\mu(w)$. The weighted sum over the characters generated by~$\psi$ (\lean{Gm}, \file{TGC/ExplicitFormulaErrorOfHostConditions.lean:38}), with weight $1$ at the trivial character and $1/(3o+1)$ at the others, $o$ the order of~$\psi$, is separate. & \ref{sec:tied}, \ref{sec:core}\\\addlinespace 16 & Relative trace formula; projected Poincar\'e series; approximate orthogonality of the columns & In part: Theorem~S, through the double-coset expansion of the Gram operator of projected Poincar\'e series (\lean{kExpTiedS_holds}, \file{FLG/SiegelLatticeTheory/FlagStabilizerTensors/TheoremSConclusion.lean:84}), for unitary groups and Siegel lattices; a sixfold cover of $\mathrm{GL}_5$ realizing the note's ``automorphy of~$\mu$'' was not found. & \ref{sec:definite}\\\addlinespace 17 & Classes in $H^1(\mathcal O_E[1/\ell],\Q_\ell/\Z_\ell(-e))$ not killed by twisting; a weakening of Vandiver & \lean{VWin}, proved as \lean{vwin_of_row}, bounds weighted lengths of localization kernels on selected windows; \lean{CorGK} contributes on the plus side. & \ref{sec:vwin}\\\addlinespace 18 & Covolumes, the index of the theta series, congruences between lifts and forms & \lean{CovolumeBoundOfX}, an upper bound for the normalized log covolume and so a lower bound for the slope; \lean{ON} (\file{FLG/IndefiniteForms/IndefSide.lean:318}) bounds the signed indices of the theta images, summed over the lifts and the conjugate places. A Hecke congruence module also occurs. & \ref{sec:covolume}\\\addlinespace 19 & The choice of tower, using Behrend & \lean{WinPair} constrains the primes; inside the given frame, \lean{hostThickExists} selects the thick set $T_2$, built on Mathlib's \lean{Behrend.sphere} and \lean{Behrend.map}. & \ref{sec:hosts}, \ref{sec:supply}\\\addlinespace 20 & GRH & \lean{everyCharacter}, \lean{imprimitive}, \lean{rh_of_every}. & \ref{sec:top}\\ \bottomrule \end{longtable}} \section{Other results in the archive}\label{sec:else} Several results of the archive are of interest on their own, among them the following. \begin{itemize}\ragged \item Ferrero--Washington in Iwasawa's form, as in \cref{tab:known}, and the class-number specialization of \lean{CorGK} (\cref{sec:vandiver}). \item Split primes in short intervals. \lean{normPrimesImagQuad_short_interval_lower_bound} says that there are $A,c,c'>0$ such that, for every prime $d\equiv3\pmod4$ with $d\ge100$, every $M\ge3$, and all $X\ge1$ and $h$ with $\log(4dM^2)\le c\sqrt{\log X}$ and $e^{-c\sqrt{\log X}}\le h\le1$, the number of primes $p\in(X,X(1+h)]$ with $4p=a^2+db^2$, $M\mid b$ and $2M\mid a-2-b$ is at least $c'hX/\bigl((4dM^2)^A\log X\bigr)$. These are the primes $p=N\alpha$ with $\alpha\in\mathcal O_K$, $K=\Q(\sqrt{-d})$, $\alpha\equiv1\pmod{M\mathcal O_K}$. Thorner and Zaman prove a lower bound of this shape for the prime ideals of degree one in a ray class of any number field \cite[Thm~3.1]{ThornerZaman17}. The archive's bound is its analogue for short intervals, in the narrower range $\log(4dM^2)\le c\sqrt{\log X}$. The statement gives no values for the constants. The theorem is used on the chain (\lean{shortNorm}). \locs{\lean{normPrimesImagQuad_short_interval_lower_bound}~\file{AVT/Proofs/GlobalInterpolation/NormPrimesImagQuadShortIntervalLowerBoundStatement.lean:8} \locseplast \lean{shortNorm}~\file{DA/ShortNorms/Primes.lean:5}} \item Bernoulli units for $2$-power twists. One file states two results. The first (proved as \lean{lemmaX}) says that $B_{1,\varphi\xi}$ is an $\ell$-adic unit when $\ell$ is prime, $r\ge1$ satisfies $4\mid\ell^r-1$, $\varphi$ is a primitive character of odd conductor~$f$ and order dividing $\ell^r-1$, and $\xi$ is a primitive character of conductor $2^n\ge f\cdot4^{v_2(\ell^r-1)}$ with $\varphi\xi$ odd. The second (proved as \lean{residual_bernoulli_nonvanishing}) is a vanishing criterion. For odd $\ell$ and a character $\chi\ne1$ modulo~$N$, not necessarily primitive, of order prime to~$\ell$, whose reduction $\bar\chi$ differs from the reduction of the cyclotomic character, if the Bernoulli number at level~$N$, $\sum_{a=1}^N\chi^{-1}(a)B_1(a/N)$, is an $\ell$-adic unit, then every cocycle of $G_\Q$ with values in $\bar\chi$ that is unramified everywhere, in the form \lean{UnrEv}, is a coboundary. \lean{corX} combines the two, and is used on the chain, in Counts~II and~III of \cref{sec:vwin} and in Count~V. \locs{two~results~\file{AV/LevelCohomology/UnramifiedCocycleIsCoboundaryOfBernoulliUnit.lean} \locsep the~first~\file{AV/LevelCohomology/UnramifiedCocycleIsCoboundaryOfBernoulliUnit.lean:24} \locsep \lean{lemmaX}~\file{AV/LevelCohomology/LemmaX.lean:78} \locsep the~second~\file{AV/LevelCohomology/UnramifiedCocycleIsCoboundaryOfBernoulliUnit.lean:43} \locsep \lean{residual_bernoulli_nonvanishing}~\file{AV/LevelCohomology/ResidualBernoulliNonvanishing.lean:174} \locsep \lean{UnrEv}~\file{AV/LevelCohomology/UnramifiedCocycleIsCoboundaryOfBernoulliUnit.lean:37} \locsep \lean{corX}~\file{AV/LevelCohomology/ChrInvNeg.lean:197} \locsep Counts~II~and~III~\file{AV/ModularityArgument/ThresholdArguments/DenseThresholdCases/AdjointLiftingOfCohomology.lean:33}--\file{37} \locseplast Count~V~\file{.../DenseThresholdCases/LinkedUnlE.lean:61}} \item Dedekind and Hecke $L$-functions over any number field $K$. The Dedekind zeta function continues meromorphically to~$\C$, holomorphic away from $s=1$, where it has a simple pole with the residue of the class number formula. Its completion satisfies $\Lambda_K(1-s)=\Lambda_K(s)$ for $s\ne0,1$, and $\zeta_K(s)/s^{r_1+r_2-1}\to-hR/w$ as $s\to0$. For a unitary Hecke character not of the form $|\cdot|^{i\tau}$, the $L$-function continues to an entire function and does not vanish on $\Real s\ge1$, and its completion satisfies $\Lambda(s,\chi)=\varepsilon(\chi)\Lambda(1-s,\chi^{-1})$. These are classical \cite{Hecke18,Hecke20,Tate}. Chebotarev's theorem is in \cref{tab:known}. \locs{Dedekind~and~Hecke~\file{GroundTheory/GlobalFields/TateThesis} \locsep continues~meromorphically~\file{.../TateThesis/DedekindZetaContinuation.lean:47}--\file{93} \locsep $\Lambda_K(1-s)=\Lambda_K(s)$~\file{.../TateThesis/DedekindContinuation.lean:181} \locsep $\zeta_K(s)/s^{r_1+r_2-1}$~\file{GroundTheory/GlobalFields/AnalyticClassNumber/ClassNumberFormula.lean:40} \locsep entire~function~\file{.../TateThesis/HeckeTheorem.lean:47} \locsep does~not~vanish~\file{.../TateThesis/Nonvanishing.lean:193} \locseplast $\Lambda(s,\chi)=\varepsilon(\chi)\Lambda(1-s,\chi^{-1})$~\file{.../TateThesis/HeckeTheoremFE.lean:86}} \item Quadratic fields. For quadratic $K$ the archive proves $\zeta_K(s)=\zeta(s)L(s,\chi_K)$ for $s\ne1$ (\lean{continuedDedekindZeta_eq_mul}, \file{GroundTheory/GlobalFields/AnalyticClassNumber/Quadratic.lean:165}). We deduce by \cref{thm:main} that the zeros of $\zeta_K$ in $0<\Real s<1$ lie on the critical line. Beyond~$\Q$ the archive states no Riemann hypothesis for a Dedekind zeta function, and the factorization after continuation only for quadratic fields. Alp\"oge expects the extended hypothesis to follow with more time \cite[p.~1]{AlpNote}. \item A frame for the Weil conjectures. The archive states, without proof, a conditional goal for the Frobenius weights, \lean{EWeil.Goal}, under about $35$ hypotheses \lean{Base} on an \'etale cohomology theory, and a separate affine Fermat estimate, \lean{EWeil.FermatCount} (\lean{LateralTheory.EWeil}, \file{Development/EtaleCohomology/WeilConjectures/FermatCount.lean:75,13,170}). Neither is on the route. The two theorems named \lean{weil_bound} are bounds for theta and Haar integrals, and the archive does not prove the Weil bound (\cref{tab:known}). \end{itemize} The bundles \lean{GeneralHyps}, \lean{GeneralHypU} and \lean{GeneralHypV} of published statements are declared in the archive, and the five theorems do not refer to them. \locs{\lean{GeneralHyps}~\file{AVT/Definitions/GlobalTwisting/GeneralHyps.lean:10} \locsep \lean{GeneralHypU}~\file{AVT/Definitions/GlobalTwisting/FinitelyManyWithInfinitesimalCharacter.lean:67} \locseplast \lean{GeneralHypV}~\file{AVT/Definitions/GlobalTwisting/SiegelWeilUnitaryConvergent2.lean:6}} \section*{Index of names} {\scriptsize \begin{longtable}{@{}>{\raggedright\arraybackslash}p{0.11\textwidth}>{\raggedright\arraybackslash}p{0.74\textwidth}>{\raggedright\arraybackslash}p{0.09\textwidth}@{}} \caption{Names used in \cref{sec:analytic,sec:star,sec:vwin}, what each says, and the section that defines it.}\label{tab:names}\\ \toprule Name & What it says & \S\\ \midrule \endfirsthead \toprule Name & What it says & \S\\ \midrule \endhead \lean{WinPair} & Primes $q\equiv3\bmod8$, $d\equiv7\bmod16$, $\log q\asymp Y^{\varepsilon/5}$, $\log d\asymp Y^{\varepsilon/5}\log^2Y$; $q$ and every odd prime $p\le Y^{\varepsilon/5}$ divide $d+1$ & \ref{sec:hosts}\\\addlinespace \lean{EAtH} & Shares of $TT$ in $A_0$-cosets; $|G(\theta,x)|\le x^{1/2+\varepsilon}$, $|\frac{L'}L(1,\theta)|\le C\log x$ for the $\theta$ trivial on $A_0$ that are not kept & \ref{sec:core}\\\addlinespace \lean{GChi} & Host requirement: \lean{EAtH} for $\langle\psi\rangle$, $C=4$; weighted $\Sha$ lengths $\le\sqrt b$ over $1\le e