Paper (November 13 version) · ecd77611496309f8908e6f08a6347f531d6f5527
1 \documentclass[pdflatex,sn-mathphys-num]{sn-jnl} 2 3 \usepackage{amsmath,amssymb,amsfonts} 4 % sn-jnl draws its ORCID logo from an EPS file it does not ship; orcidlink supplies the logo. 5 \let\orcidlogo\undefined 6 \usepackage{orcidlink} 7 \usepackage{amsthm} 8 \usepackage{microtype} 9 \usepackage{mathtools} 10 \usepackage{comment} 11 \usepackage{enumitem} 12 \usepackage[title]{appendix} 13 \input{ai_review} 14 \hypersetup{colorlinks=true,citecolor=blue,linkcolor=blue,urlcolor=blue} 15 \numberwithin{equation}{section} 16 17 \theoremstyle{thmstyleone} 18 \newtheorem{theorem}{Theorem}[section] 19 \newtheorem{proposition}[theorem]{Proposition} 20 \newtheorem{lemma}[theorem]{Lemma} 21 \newtheorem{corollary}[theorem]{Corollary} 22 \newtheorem{definition}[theorem]{Definition} 23 \newtheorem{convention}[theorem]{Notation} 24 \theoremstyle{thmstyletwo} 25 \newtheorem{remark}[theorem]{Remark} 26 27 28 29 30 \newcommand{\Char}{\operatorname{Char}} 31 \newcommand{\Weyl}{A_n} 32 \newcommand{\ord}{\operatorname{ord}} 33 \newcommand{\gr}{\operatorname{gr}} 34 \newcommand{\Ann}{\operatorname{Ann}} 35 \newcommand{\Dir}{\operatorname{Dir}} 36 \newcommand{\Sp}{\operatorname{Sp}} 37 \newcommand{\Spec}{\operatorname{Spec}} 38 \newcommand{\Supp}{\operatorname{Supp}} 39 \newcommand{\Sing}{\operatorname{Sing}} 40 \newcommand{\sm}{\mathrm{sm}} 41 \newcommand{\bbA}{\mathbb{A}} 42 \newcommand{\bbP}{\mathbb{P}} 43 \newcommand{\bbN}{\mathbb{N}} 44 \newcommand{\bbZ}{\mathbb{Z}} 45 \newcommand{\bbQ}{\mathbb{Q}} 46 \newcommand{\Kbar}{\overline{k}} 47 \newcommand{\dd}{\mathrm{d}} 48 \DeclareMathOperator{\rad}{rad} 49 50 51 \begin{document} 52 \setlength{\parindent}{0pt} 53 \title[Stafford's conjecture on cyclicity of torsion modules]{Stafford's conjecture on cyclicity of torsion modules} 54 55 \author[1]{\fnm{Christopher} \sur{Albert}\,\orcidlink{0000-0003-4773-416X}}\email{albert@tugraz.at} 56 57 \author*[1]{\fnm{Johanna} \sur{Moser}\,\orcidlink{0009-0005-5289-6915}}\email{j.moser@tugraz.at} 58 59 \author[1]{\fnm{Maximilian} \sur{Philipp}}\email{philipp@student.tugraz.at} 60 61 \affil*[1]{\orgname{Graz University of Technology}, \orgaddress{\city{Graz}, 62 \country{Austria}}} 63 64 \abstract{% 65 We prove that over a Weyl algebra over a field of characteristic zero $A_n(k)$, every nonzero $d\in A_n(k)$ admits $F,R,S\in A_n(k)$ such that 66 $1=dR+FdS$, a conjecture Stafford proposed in 1978. As a consequence, every finitely generated torsion right $A_n(k)$-module is cyclic. 67 The proof maps $d$, by a symplectic change of variables, to a normal form $\tilde d$ 68 and shows that the cyclic module $Q=A_n(k)/(\tilde dA_n(k)+x_1^N\tilde dA_n(k))$ vanishes. 69 We first show that right multiplication by $x_1$ is surjective on $Q$. A comparison of lengths of localized modules 70 over the associated graded ring then gives $\operatorname{gr}Q=x_1\operatorname{gr}Q$, 71 so that the characteristic variety of $Q$ is disjoint from the hyperplane 72 $\{x_1=0\}$. Finally, Gabber's involutivity theorem together with an argument about conormal directions at infinity from algebraic geometry shows that such a characteristic variety must be empty and therefore, $Q = 0$. 73 } 74 75 \keywords{Weyl algebra, Stafford's conjecture, cyclic module, characteristic variety, involutivity, formal proof} 76 77 \pacs[MSC Classification]{16S32, 32C38, 14F10, 68V20} 78 79 \maketitle 80 81 82 \section{Introduction} 83 84 Let $k$ be a field of characteristic zero, and let 85 \[ A_n(k)=k\langle x_1,\ldots,x_n,\partial_1,\ldots,\partial_n\rangle\] 86 be the $n$th Weyl algebra, with $[\partial_i,x_j]=\delta_{ij}$ and 87 $[\partial_i,\partial_j]=[x_i,x_j]=0$. 88 89 Stafford's Conjecture~3.8 in \cite{Stafford1978} asks whether every nonzero $d\in A_n(k)$ admits $F\in A_n(k)$ with $A_n(k)=dA_n(k)+FdA_n(k)$ 90 \cite[p.~438]{Stafford1978}. Stafford shows that this formulation implies stable cyclicity for all finitely generated torsion right Weyl-algebra modules 91 \cite[pp.~437--438]{Stafford1978}. 92 We prove Stafford's conjecture: 93 \begin{theorem}\label{thm:main} 94 For every field of characteristic zero $k$, every $n\geq0$, and every 95 nonzero $d\in A_n(k)$, there exists an $F\in A_n(k)$ such that 96 \[ A_n(k)=dA_n(k)+FdA_n(k),\] 97 or equivalently, there exist $F, S, R\in A_n(k)$ such that 98 \[ 1=dR+FdS.\] 99 Moreover, if $n\ge1$, one may choose $F=\Phi_g^{-1}(x_1)^{\deg_Bd}$ for a linear 100 $\omega$-preserving automorphism $g$ as in Section~\ref{sec:monic}. 101 \end{theorem} 102 103 \AIadd{\formalref{\leandecl{Stafford38/Paper/Main.lean}{24}{Stafford38.Paper.theorem_1_1} proves \leandecl{Stafford38/ChallengeDefinitions.lean}{78}{Stafford38Challenge.UniversalStatement}; 104 \leandecl{Stafford38/Paper/Main.lean}{29}{Stafford38.Paper.theorem_1_1_fixedSource} proves \leandecl{Stafford38/ChallengeDefinitions.lean}{193}{Stafford38FixedSourceChallenge.UniversalFixedSourceStatement} 105 with \leandecl{Stafford38/ChallengeDefinitions.lean}{170}{Stafford38FixedSourceChallenge.IsLinearWeylCoordinate} and \leandecl{Stafford38/ChallengeDefinitions.lean}{155}{Stafford38FixedSourceChallenge.bernsteinDegree}; 106 Mathlib-only statements compared by Comparator: \leandecl{Stafford38/ChallengeDefinitions.lean}{291}{Stafford38Challenge.PaperTheorem_1_1}, 107 \leandecl{Stafford38/ChallengeDefinitions.lean}{78}{Stafford38Challenge.UniversalStatement}, 108 \leandecl{Stafford38/ChallengeDefinitions.lean}{193}{Stafford38FixedSourceChallenge.UniversalFixedSourceStatement}.}} 109 110 111 \textit{Proof sketch:} 112 We state some preliminaries in section \ref{sec:setting}, then start in section \ref{sec:monic} by defining a map $\Phi_g$ sending any chosen $d$ of Bernstein degree $N$ to a monic normal form $\tilde d$ with order symbol $P$ using Bernstein and order filtrations of $A_n(k)$. For $\tilde d$, we prove in sections \ref{sec:euler}-\ref{sec:proof} that the module $Q = A_n(k)/(\tilde d A_n(k) + x_1^N\tilde d A_n(k)) $ vanishes, and find the final form of Theorem \ref{thm:main} by applying the inverse map $\Phi_g^{-1}$ to $\tilde d A_n(k) + x_1^N\tilde d A_n(k).$ The main components of the proof are the following: in section \ref{sec:euler}, we show that $Q = Qx_1$ using the extended euclidean algorithm, and $\gr Q = x_1\gr Q,$ through two contradictory length estimations of localized modules. In section \ref{sec:conormal-review}, using mainly results from algebraic geometry, we show that a set $W$ fulfilling certain criteria (Zariski closed over an algebraically closed field of characteristic zero, fibre-conical, involutive, there exists a polynomial with $P(1,0\dots,0)\neq 0$ vanishing on $W$, and $W$ does not contain points with $x_1 = 0$) must always be empty (Theorem \ref{thm:coisotropic-exclusion}). Finally, we show in section \ref{sec:proof} that by defining $W$ to be the characteristic variety of the closed-field pendant to $Q$, the results of section \ref{sec:euler} as well as the definition of $P$ in section \ref{sec:monic} ensure that $W$ fulfills all criteria and must thus be empty. We conclude that therefore, also $Q$ itself must be zero, from which Theorem \ref{thm:main} follows directly. Lastly, we provide the proof by Stafford \cite{Stafford1978} that hence, every finitely generated torsion module is stably cyclic (Corollary \ref{cor:cyclic}). 113 114 115 \subsection{Related literature} 116 Dixmier showed that over the first Weyl algebra $A_1(k)$ with $k$ having characteristic zero, any right ideal can be generated by at most two elements \cite{dixmier}. In a seminal paper in 1978, Stafford extended this result to all $A_n(k)$ and moreover showed, that any finitely generated left torsion $A_n(k)$-module is a homomorphic image of a projective left ideal, and as such can also be (stably) generated by at most two elements \cite[Theorem 3.7]{Stafford1978}. 117 Hillebrand and Schmale \cite{HillebrandSchmale2001}, as well as Leykin \cite{Leykin2004} give an algorithm to find the two generators of left ideals, also valid for rational coefficient algebras $B_n(k)$. Caro and Levcovitz \cite{CaroLevcovitz2010} generalize the result for both left and right ideals to formal and to convergent Laurent series, and Quadrat and Robertz \cite{QuadratRobertz2014} provide a constructive proof method to calculate the two generators of torsion modules over the Weyl algebras, as well as rings with rational coefficients, formal or convergent Laurent series, and, for ordinary differential operators, with formal or locally convergent power series coefficients. 118 119 In the same paper \cite{Stafford1978}, Stafford states the Conjecture 3.8 (our Theorem \ref{thm:main}), and shows that this conjecture implies that any finitely generated right torsion module can be stably generated by one element, i.e. is stably cyclic (Corollary \ref{cor:cyclic}). He shows that cyclicity is true if $n = 1$ (since $A_1(k)$ has the restricted minimum condition which makes torsion modules Artinian and he proves that Artinian modules are cyclic) and also states that the conjecture is true if $d$ is monomial, but provides, as far as the authors are aware, no proof of this. This case is also contained in our proof. 120 For \textit{holonomic} modules over ``simple rings such that they do not have finite length over themselves as left modules'' (which all Weyl algebras are), Leykin \cite{Leykin2004} proved cyclicity constructively, and attributed the proof method to Stafford. However, torsion modules need not be holonomic: for $n\ge2$, Stafford constructed quotients $A_n(k)/dA_n(k)$ of Gelfand--Kirillov dimension $2n-1$ as counterexamples \cite{StaffordNonholonomic1985}, so the question about cyclicity remained open for non-holonomic torsion modules. A paper by Bellamy in 2026 states this explicitly (\cite{Bellamy2026} Conjecture 3.4). 121 122 We base our proof on the Weyl algebra and on filtrations and characteristic 123 varieties of $A_n(k)$-modules as treated in the textbooks 124 \cite{coutinho,Bjork,HTT,MR,SchnellNotes}, on the involutivity of 125 the characteristic variety due to Gabber \cite{Gabber1981,Gabber1982,SinghKumar2014}, on commutative algebra, in particular localization, length, valuation rings, Kähler differentials and completions 126 \cite{atiyah,matsumura,Stacks}, field theory and bilinear forms \cite{lang} and spectral sequences \cite{weibel}, and on 127 algebraic geometry, mostly based on \cite{hartshorne}. 128 129 \subsection{Formal verification with Lean}\label{sec:formal} 130 Theorem~\ref{thm:main} and 131 Corollary~\ref{cor:cyclic} are proved in Lean~4~\cite{Lean4} with 132 Mathlib~\cite{Mathlib2020} and AlgebraicAnalysis 133 v0.3.3~\cite{AlgebraicAnalysis033}. The proofs use only Lean's three 134 standard axioms: propositional extensionality (\texttt{propext}), the axiom 135 of choice (\texttt{Classical.choice}) and soundness of quotient types 136 (\texttt{Quot.sound}). Every intermediate result is proved in Lean as well, 137 including Gabber's involutivity theorem for cyclic $A_n(k)$-modules. The 138 proof in Sections~\ref{sec:monic}--\ref{sec:proof} follows the formal proof 139 step by step. The main theorem is registered 140 in Palomar~\cite{Stafford38Palomar2026}, and the Lean sources are 141 archived~\cite{Stafford38Zenodo140}. Tooling and the role of AI assistance 142 are described in Appendix~\ref{app:ai-workflow}. Section~\ref{sec:generalizations} states further consequences proved in Lean and two generalizations, to rings of differential operators on smooth affine varieties and to simple filtered algebras. 143 The supplementary 144 material~\cite{Stafford38Supplementary032} shows each formalized statement of 145 this paper next to its Lean version and proof. It can be browsed at 146 \url{https://itpplasma.github.io/stafford38-supplementary/}. 147 148 \section{Preliminaries}\label{sec:setting} 149 150 151 Throughout, $k$ is a field of characteristic zero and for the $n$-th Weyl algebra over $k$, we write 152 \[ A_n(k)=k\langle x_1,\dots,x_n,\partial_1,\dots,\partial_n\rangle, 153 \quad [\partial_i,x_j]=\delta_{ij},\quad [x_i,x_j]=[\partial_i,\partial_j]=0 .\] 154 All ideals are right ideals and all modules are right modules 155 unless the contrary is said explicitly. 156 \begin{definition}[e.g. \cite{coutinho}]\label{def:cyclic} 157 Let $M$ be a right $A_n(k)$-module. We say $M$ is \emph{torsion} if for every $m\in M$ there exists 158 $0\ne d\in A_n(k)$ with $md=0$, and \emph{cyclic} if $M=mA_n(k)$ for some $m$. 159 \end{definition} 160 \begin{convention}\label{conv:pair} 161 We single out the first Weyl pair $(x_1, \partial_1)$ and write 162 \[\qquad p:=\partial_1,\qquad 163 C:=A_{n-1}(k)[x_1],\qquad A_n(k)=C[p;\delta],\] 164 Here $\delta(x_1)=1$ and $\delta$ vanishes on $A_{n-1}(k)$, 165 so that $x_1$ is central in $C$, $[p,x_1] = 1$, and the \emph{transverse algebra} $A_{n-1}(k)=k\langle x_i,\partial_i : i\ge2\rangle$ commutes with $x_1$ and $p$. 166 Every element of $A_n(k)$ can uniquely be written as $\sum_j p^j c_j$ with $c_j\in C$. \cite[Prop. 1.2.1]{coutinho} 167 \end{convention} 168 \AIadd{\formalref{% 169 \leandecl{Stafford38/Paper/Section2/Weyl.lean}{98}{Stafford38.Paper.Section2.transverse}, 170 \leandecl{Stafford38/Paper/Section2/Weyl.lean}{103}{Stafford38.Paper.Section2.C}, 171 \leandecl{Stafford38/Paper/Section2/Weyl.lean}{179}{Stafford38.Paper.Section2.convention_2_2_relations}, 172 \leandecl{Stafford38/Paper/Section2/Weyl.lean}{199}{Stafford38.Paper.Section2.convention_2_2_exists}, 173 \leandecl{Stafford38/Paper/Section2/Weyl.lean}{228}{Stafford38.Paper.Section2.convention_2_2_unique}}} 174 175 176 \begin{definition}[{[Def.~2.6 and ~2.9 \cite{SchnellNotes}]}]\label{def:filtration} 177 A filtration $F_\bullet$ on $A_n(k)$ is a sequence of linear subspaces 178 \[\{0\} = F_{-1} \subseteq F_0 \subseteq F_1\subseteq...\subseteq A_n(k),\] 179 such that $F_i\cdot F_j \subseteq F_{i+j}$ and $A_n(k) = \bigcup F_j.$ 180 The associated graded algebra, is defined as 181 \[\mathrm{gr}^F A_n(k) = \bigoplus_{j = 0}^\infty F_j/F_{j-1}.\] 182 Let $a\neq 0 \in A_n(k)$ and let $j$ be the smallest integer, such that $a \in F_j$, called a \emph{filtration degree} $\deg_F a$. The class of $a$ in $F_j/F_{j-1}$ is called its \emph{symbol} $\sigma_j(a)$. A symbol, being the class of an element in a single graded piece, is automatically homogeneous of the corresponding 183 degree, i.e.\ contains only terms of that total degree. 184 \end{definition} 185 186 In the proof, we use two filtrations of $A_n(k)$ (for more details, see e.g. \cite[Chapter 7]{coutinho}) 187 \begin{itemize}[nosep,leftmargin=1.4em] 188 \item The \emph{Bernstein filtration} $B_\bullet$ assigns degree $1$ to every 189 $x_i$ and every $\partial_i$. The Bernstein degree of $d$ is 190 written $\deg_B d$. 191 \item The \emph{order filtration} $F_\bullet$ assigns degree $0$ to every $x_i$ 192 and degree $1$ to every $\partial_i$; $F_m$ is spanned by the monomials 193 $x^\alpha\partial^\beta$ with $|\beta|\le m$. %We write $\sigma_m(a)$ for the order-$m$ principal symbol. 194 \end{itemize} 195 \AIadd{\formalref[{}]{% 196 \leandecl{Stafford38/Paper/Section2/Filtration.lean}{79}{Stafford38.Paper.Section2.IsFiltration}, 197 \leandecl{Stafford38/Paper/Section2/Filtration.lean}{68}{Stafford38.Paper.Section2.gr}, 198 \leandecl{Stafford38/Paper/Section2/Filtration.lean}{86}{Stafford38.Paper.Section2.filtrationDegree}, 199 \leandecl{Stafford38/Paper/Section2/Filtration.lean}{95}{Stafford38.Paper.Section2.symbol}, 200 \leandecl{Stafford38/Paper/Section2/Filtration.lean}{104}{Stafford38.Paper.Section2.bernsteinFiltration}, 201 \leandecl{Stafford38/Paper/Section2/Filtration.lean}{109}{Stafford38.Paper.Section2.orderFiltration}, 202 \leandecl{Stafford38/Paper/Section2/Filtration.lean}{121}{Stafford38.Paper.Section2.bernsteinDegree}, 203 \leandecl{Stafford38/Paper/Section2/Filtration.lean}{185}{Stafford38.Paper.Section2.grEquivSymbolRing}}} 204 For both the order filtration and the Bernstein filtration, the associated graded algebra $\mathrm{gr} A_n(k)$ is simply the polynomial ring $k[x,\xi]=k[x_1,\dots,x_n,\xi_1,\dots,\xi_n]$ in $2n$ variables with $\xi_i$ being the symbol of $\partial_i$ \cite[Th. 7.3.1]{coutinho}. In particular, $\mathrm{gr} A_n(k)$ is commutative \cite[Prop.~2.10]{SchnellNotes}. 205 Good filtrations are treated systematically for example in \cite{Bjork,HTT}. 206 207 \begin{definition}[{\cite[Ch. 8.3]{coutinho}}]\label{def:good-filtration} 208 A filtration of a right $A_n$-module $M$ compatible with a filtration $F_\bullet$ 209 is an exhaustive increasing chain of $k$-subspaces $M_\bullet$ with 210 $M_iF_j\subseteq M_{i+j}$ and $M_i=0$ for $i<0$. The associated graded module is $\operatorname{gr}M=\bigoplus_iM_i/M_{i-1}$; it is a graded module over $\operatorname{gr}_FA_n(k)$ via the maps 211 $M_i/M_{i-1}\times F_j/F_{j-1}\to M_{i+j}/M_{i+j-1}$, $(\bar m,\bar a)\mapsto\overline{ma}$, 212 which are well defined because $M_{i-1}F_j\subseteq M_{i+j-1}$ and $M_iF_{j-1}\subseteq M_{i+j-1}$. 213 It is \emph{good} if the associated graded module is finitely generated over $\gr^FA_n$. 214 \end{definition} 215 As an example, the filtration $M_i:=qF_i$ of a cyclic module over $A_n$ induced by its generator $q$ is good, since $\gr Q$ is generated by the element $\bar{q}$ denoting the class of $q$ in $Q_0/Q_{-1}.$ 216 217 For a finitely generated right $A_n$-module $M$ with a good order filtration 218 $M_\bullet$, the graded module $\gr M$ is a finitely generated $k[x,\xi]$-module. 219 220 \begin{definition}[{\cite[Ch. 1]{atiyah}}]\label{def:spec} 221 Let $S$ be a commutative ring. The \emph{spectrum} $\operatorname{Spec}S$ is the set 222 of all prime ideals of $S$. We say that $f\in S$ vanishes at the point 223 $\mathfrak p\in\operatorname{Spec}S$ if $f\in\mathfrak p$. The \emph{vanishing locus} 224 of an ideal $J\subseteq S$ is 225 \[ V(J)=\{\mathfrak p\in\operatorname{Spec}S:\ \mathfrak p\supseteq J\},\] 226 the set of points at which all elements of $J$ vanish; these sets are the (Zariski) closed sets 227 of the Zariski topology on $\operatorname{Spec}S$. For a field $K$ and an ideal $J\subseteq K[z_1,\dots,z_N]$ we write 228 \[V_K(J)=\{a\in K^N : f(a)=0\ \forall f\in J\}\] 229 for its (classical) zero set, and for $S\subseteq K^N$ we write 230 $I(S)=\{f\in K[z]: f(a)=0\ \forall a\in S\}$ for the \emph{vanishing ideal} of $S$. 231 The sets $V_K(J)$ are the closed sets of the Zariski topology on $K^N$. 232 Since for $f$ in the maximal ideal $f\in\mathfrak m_a:=(z_1-a_1,\dots,z_N-a_N)$ iff $f(a)=0$, we have 233 $a\in V_K(J)$ if and only if $\mathfrak m_a\in V(J)$.\\ 234 For an ideal $J$ of $S$, its \emph{radical} is 235 $\sqrt J=\{f\in S:\ f^m\in J\text{ for some }m\ge1\}$. $J$ is called \emph{radical} if $J=\sqrt J$. Since a prime ideal contains $J$ 236 iff it contains $\sqrt J$, we have $V(J)=V(\sqrt J)$. 237 \end{definition} 238 239 \begin{definition}[{\cite[Appendix E]{HTT}}]\label{def:cotangent} 240 Let $\mathbb A^n_k$ be affine $n$-space over $k$ with coordinates $x_1,\dots,x_n$. 241 Its \emph{cotangent bundle} $T^*\mathbb A^n_k=\mathbb A^{2n}_k$ has the additional 242 fibre coordinates $\xi_1,\dots,\xi_n$ dual to $x_1,\dots,x_n$, so that a covector is 243 $\sum_i\xi_i\,dx_i$. Its coordinate ring is $k[x,\xi]=\gr A_n(k)$, and we regard 244 $\mathbb A^n_k=\operatorname{Spec}k[x]$ and $T^*\mathbb A^n_k=\operatorname{Spec}k[x,\xi]$ 245 as sets of prime ideals \cite[Tag 00DZ]{Stacks}. 246 \end{definition} 247 248 249 \AIadd{\formalref{\leandecl{Stafford38/Paper/Section2/Geometry.lean}{159}{Stafford38.Paper.Section2.CotangentBundle}, 250 \leandecl{Stafford38/Paper/Section2/Geometry.lean}{162}{Stafford38.Paper.Section2.proj}, 251 \leandecl{Stafford38/Paper/Section2/Geometry.lean}{166}{Stafford38.Paper.Section2.fibre}; 252 \leandecl{Stafford38/Characteristic/Polynomial.lean}{22}{Stafford38.Characteristic.SymbolRing} 253 (the ring $k[x,\xi]$; the first summand indexes base coordinates $x_i$, and the second indexes fibre coordinates $\xi_i$)}} 254 255 256 \begin{definition}[{\cite[Ch. 2.2]{HTT}}]\label{def:char-variety} 257 The \emph{characteristic variety} of a finitely generated $A_n$-module $M$ with a good order filtration 258 $M_\bullet$ is defined by 259 \[\Char(M):=V\bigl(\sqrt{\Ann_{k[x,\xi]}\gr M}\bigr) = V\bigl(\Ann_{k[x,\xi]}\gr M\bigr)\subseteq 260 T^*\mathbb A^n_k=\Spec k[x,\xi],\] 261 $\Ann_{k[x,\xi]}\gr M = \{f \in k[x,\xi]: f\cdot m = 0\,\forall m \in \gr M\}$ are the \emph{annihilators} of $\gr M$ over $k[x,\xi]$, and $V$ denotes their vanishing locus. 262 263 $\Char(M)$ is independent of the choice of the good filtration. \cite[Lemma D.3.1]{HTT} 264 \end{definition} 265 \AIadd{\formalref[{}]{% 266 \leandecl{Stafford38/Paper/Section2/CharacteristicVariety.lean}{54}{Stafford38.Paper.Section2.CharOf}, 267 \leandecl{Stafford38/Paper/Section2/CharacteristicVariety.lean}{83}{Stafford38.Paper.Section2.Char}, 268 \leandecl{Stafford38/Paper/Section2/CharacteristicVariety.lean}{61}{Stafford38.Paper.Section2.definition_2_7_independent}, 269 \leandecl{Stafford38/Paper/Section2/CharacteristicVariety.lean}{87}{Stafford38.Paper.Section2.Char_eq}}} 270 271 272 \begin{definition}[{\cite[Ch. E.1]{HTT}}]\label{def:symplectic} 273 A bilinear form $\omega$ on a $k$-vector space is \emph{symplectic} if it is 274 antisymmetric and nondegenerate. The \emph{standard symplectic form} on $k^{2n}$ 275 with basis $x_1,\dots,x_n,\partial_1,\dots,\partial_n$ is determined by 276 \[ \omega(x_i,\partial_j)=\delta_{ij},\qquad 277 \omega(x_i,x_j)=\omega(\partial_i,\partial_j)=0,\] 278 analogously to the Weyl relations. 279 The form is \emph{nondegenerate} if $\omega(u,\cdot)=0$ implies $u=0$. For a subspace $U$, we define 280 $U^\perp=\{v:\omega(u,v)=0\ \forall u\in U\}$. A \emph{symplectic basis} is a basis 281 $e_1,\dots,e_n,f_1,\dots,f_n$ with $\omega(e_i,f_j)=\delta_{ij}$ and 282 $\omega(e_i,e_j)=\omega(f_i,f_j)=0$. The \emph{Gram matrix} of $\omega$ in a basis 283 $(b_i)$ is $J=(\omega(b_i,b_j))_{ij}$. 284 \end{definition} 285 286 \begin{definition}[{\cite[Appendix D.3]{HTT}}]\label{def:poisson} 287 The \emph{Poisson bracket} on $k[x,\xi]$ is 288 \[ \{f,g\}=\sum_{i=1}^n\left( 289 \frac{\partial f}{\partial x_i}\frac{\partial g}{\partial \xi_i} 290 -\frac{\partial f}{\partial \xi_i}\frac{\partial g}{\partial x_i}\right).\] 291 An ideal $J$ is \emph{involutive} if $\{J,J\}\subseteq J$. 292 293 \end{definition} 294 295 \AIadd{\formalref[{}]{% 296 \leandecl{Stafford38/Paper/Section2/Symplectic.lean}{130}{Stafford38.Paper.Section2.poisson}, 297 \leandecl{Stafford38/Paper/Section2/Symplectic.lean}{137}{Stafford38.Paper.Section2.poisson_eq}}} 298 299 300 \AIadd{\formalref{\leandecl{Stafford38/Paper/Section2/Symplectic.lean}{141}{Stafford38.Paper.Section2.IsInvolutive}, 301 \leandecl{Stafford38/Paper/Section2/Symplectic.lean}{152}{Stafford38.Paper.Section2.IsInvolutiveSet}, 302 \leandecl{Stafford38/Paper/Section2/Symplectic.lean}{146}{Stafford38.Paper.Section2.isInvolutive_iff}; 303 negative control \leandecl{Stafford38/Geometry/GeneralCoisotropicSetsTest.lean}{164}{Stafford38.Geometry.GeneralCoisotropicSetsTest.zeroSectionIdealOne_not_isPoisson}}} 304 305 306 \begin{theorem}[Gabber {\cite{Gabber1981}\cite{SinghKumar2014}}{\cite[{Thm. D.3.4}]{HTT}}]\label{ithm:gabber} 307 Let $K$ be a field of characteristic zero and $M$ a finitely generated $A_n(K)$-module 308 with a good order filtration. Let $J = \sqrt{Ann_{K[x, \xi]}\gr M}$ be the radical of the annihilators of $M$. Then $J$ is involutive, i.e. \[\{J, J\} \subset J.\] 309 %original: Let $R$ be a filtered ring such that $\gr(R)$ is a Noetherian commutative algebra over $\mathbb Q$. Let $M$ be a finitely generated $R$ module. Let $J = \sqrt{Ann_{\gr R}\gr M}$ be the radical of the annihilators of $M$. Then $J$ is involutive, i.e. \[\{J, J\} \subset J.\] 310 \end{theorem} 311 \AIadd{\formalref[{}]{% 312 \leandecl{Stafford38/Paper/Section2/Symplectic.lean}{161}{Stafford38.Paper.Section2.theorem_2_11}, 313 \leandecl{Stafford38/Paper/Section2/Symplectic.lean}{174}{Stafford38.Paper.Section2.theorem_2_11_cyclic}}} 314 %\begin{theorem}[{Hilbert's Nullstellensatz, see e.g. \cite[Cor 7.10, Ex. 7.14]{atiyah}}]\label{thm:nss} 315 %Let $K$ be an algebraically closed field and $J\subseteq K[z_1,\dots,z_N]$ an 316 %ideal. Then 317 %\begin{enumerate}[label=(\alph*)] 318 % \item $V(J)=\varnothing\iff 1 \in J$ 319 % \item $I(V(J))=\sqrt J$ 320 %\end{enumerate} 321 %\end{theorem} 322 323 \begin{theorem}[{Hilbert's Nullstellensatz, see e.g. \cite[Thm. I.1.3A, Cor. I.1.4, Thm. I.3.2]{hartshorne} and \cite[Cor 7.10, Ex. 7.14]{atiyah}}]\label{thm:nss} 324 Let $K$ be an algebraically closed field and $J\subseteq K[z_1,\dots,z_N]$ an 325 ideal. Then 326 \begin{enumerate} 327 \item[(a)] the maximal ideals of $K[z]$ are exactly $\mathfrak m_a:=(z_1-a_1,\dots,z_N-a_N)$, 328 $a\in K^N$; hence $a\mapsto\mathfrak m_a$ identifies $K^N$ with the closed 329 points of $\operatorname{Spec}K[z]$, and $V_K(J)$ with the closed points of 330 $V(J)$; 331 \item[(b)] $V_K(J)=\emptyset$ if and only if $1\in J$; 332 \item[(c)] $I(V_K(J))=\sqrt J$. 333 \end{enumerate} 334 \end{theorem} 335 336 \section{Mapping to a normal form}\label{sec:monic} 337 Given a fixed nonzero $d \in A_n(k)$ of Bernstein degree $N$, our first goal is to find a map to replace $d$ by a normal form $\tilde{d}$ that is monic in $p$, i.e. of the form $\tilde d=p^N+\sum_{j<N}p^ja_j$ with coefficients $a_j$ in $C$. \\ 338 339 If $N=0$, then $d \in B_0 = k$ is a nonzero scalar, hence a unit, and $1 = dR+FdS$ holds trivially with $F=1$, $R=d^{-1}$, $S=0$. We therefore assume $N\geq1$. 340 341 342 We want to construct a well-behaved automorphism of $\gr^BA_n(k)=k[x,\xi]$ 343 from simpler linear algebra. We proceed in three stages: 344 define an automorphism of the vector space $\mathbb V=k^{2n}$, extend it to 345 an automorphism of $A_n(k)$ itself, and observe the induced action this 346 extension leaves on the associated graded ring. 347 348 We can identify $\mathbb V = k^{2n}$ with the Bernstein-degree-one part $B_1/B_0$ of 349 $A_n(k)$, i.e.\ linear combinations of $x_1,\ldots,x_n,\partial_1,\ldots, 350 \partial_n$; the coordinates of a vector are then the coefficients of 351 these generators. By this identification, $\mathbb V$ carries the standard symplectic form $\omega$ (Def. \ref{def:symplectic}). 352 353 \begin{lemma}\label{lem:sympcompl} 354 Let $(\mathbb V,\omega)$ be a symplectic $k$-space of dimension $2n$ and $0\neq v\in \mathbb V$. 355 Then $v$ is the first vector of a symplectic basis of $\mathbb V$. Consequently, there 356 exists a $k$-linear automorphism $g$ preserving $\omega$ (i.e. $\omega(gu,gv)=\omega(u,v)$ for all 357 $u,v\in\mathbb V$) with $g(e)=v$ for any prescribed nonzero vector $e$. 358 \end{lemma} 359 360 \begin{proof} 361 Since $\omega$ is nondegenerate and $v\neq0$, the functional $\omega(v,-)$ 362 is not identically zero, so there exists a $w'\in\mathbb V$ with 363 $\omega(v,w')=c\neq0$ and defining $w:=w'/c$ gives $\omega(v,w)=1$. 364 365 Let $U=\operatorname{span}(v,w)$. For $0\neq av+bw\in U$: if $a\neq0$, 366 $\omega(av+bw,w)=a\neq0$; if $b\neq0$, $\omega(av+bw,v)=-b\neq0$. Either 367 way $av+bw$ pairs nontrivially with something in $U$, so $\omega|_U$ is 368 nondegenerate. 369 370 Since $\omega|_U$ is nondegenerate, $U\cap U^\perp=0$, and as 371 $\dim U+\dim U^\perp=2n = \dim\mathbb V$ \cite[Prop. 11.2.1]{coutinho} for a nondegenerate form, this means 372 $\mathbb V=U\oplus U^\perp$. If $u\in U^\perp$ pairs to zero with all of 373 $U^\perp$, it also pairs to zero with $U$ (as $u\in U^\perp$), hence with 374 all of $\mathbb V$, forcing $u=0$; so $\omega|_{U^\perp}$ is nondegenerate 375 too. 376 377 By induction on $n$ (trivial for $n=1$, where $U=\mathbb V$), a 378 symplectic basis $e_2,\ldots,f_n$ of $U^\perp$ therefore exists \cite[XV.8.1 and XV.8.2]{lang}. Setting $e_1:=v$, 379 $f_1:=w$ gives a symplectic basis of $\mathbb V=U\oplus U^\perp$, since 380 cross-pairings between $U$ and $U^\perp$ vanish by definition of 381 $U^\perp$. 382 383 Consequently: given nonzero $v,e\in\mathbb V$, we can complete both vectors to a symplectic basis as above. We then define the linear map $g$ to send $e$ to $v$, and each subsequent $e$-basis vector to the corresponding $v$-basis vector, thus preserving all pairwise values of $\omega$ on basis vectors. Through the bilinearity of $\omega$, $g$ then preserves $\omega$ on all of $\mathbb V$; and $g(e)=v$. 384 \end{proof} 385 \AIadd{\formalref{% 386 \leandecl{Stafford38/Paper/Section3/Symplectic.lean}{269}{Stafford38.Paper.Section3.lemma_3_1}}} 387 Let $g$ be a $k$-linear automorphism of $\mathbb V$ that preserves $\omega$. 388 Since $\mathbb V$ is identified with the span of the generators of $A_n(k)$, 389 $g$ assigns to each generator $u$ a linear combination $g(u)$ of generators. 390 Because $A_n(k)$ is presented by generators and the Weyl relations 391 $[v,u]=\omega(u,v)$, and $g$ preserves $\omega$, the elements $g(u)$ satisfy 392 the same relations as the $u$. Hence the substitution $u\mapsto g(u)$ extends 393 to a unique $k$-algebra endomorphism $\Phi_{g}$ of $A_n(k)$ \cite[Th. 1.3.1, Ex. 7.6.9]{coutinho}. Because $g$ is an automorphism, so is $\Phi_g$, and we have $\Phi_g^{-1} = \Phi_{g^{-1}}.$ 394 395 396 \AIadd{\formalref{\leandecl{Stafford38/Paper/Section3/PhiG.lean}{68}{Stafford38.Paper.Section3.phiG}, 397 \leandecl{Stafford38/Paper/Section3/PhiG.lean}{95}{Stafford38.Paper.Section3.step_phi_g}}} 398 399 \begin{proposition}\label{prop:monic} 400 Let $0\neq d\in A_n(k)$ with $N:=\deg_B d\ge1$. There exists an automorphism $\Phi_g$ of $A_n(k)$ and 401 $c\in k^{\times}$ such that 402 \begin{equation} 403 \tilde d:=c^{-1}\Phi_g(d)=p^{\,N}+\sum_{j<N}p^{\,j}a_j, 404 \qquad a_j\in C,\quad \deg_B a_j\le N-j\label{eq:monic-d} 405 \end{equation} 406 in the Notation \ref{conv:pair}. $\tilde d$ itself has $\deg_B \tilde d = N.$ 407 \end{proposition} 408 409 \begin{proof} 410 Let $s = \sigma_N^B(d)\in k[x,\xi]$ be the Bernstein principal symbol of $d$, i.e. s is a nonzero 411 homogeneous polynomial of degree $N$ on $\mathbb V$. A characteristic-zero field is 412 infinite, so every nonzero polynomial has a nonvanishing value: we choose $v\in \mathbb V$ 413 with $s(v)\neq0$. Because $s$ is homogeneous of degree $N\geq 1$, $s(0)$ must vanish, therefore $v$ is nonzero. 414 Let $e_p\in\mathbb V$ be the standard basis vector corresponding to $p$. 415 By Lemma~\ref{lem:sympcompl} there is an $\omega$-preserving automorphism 416 $h$ of $\mathbb V$ with $h(e_p)=v$.\\ 417 If $h$ preserves $\omega$, then $h^{\mathsf T}Jh=J$, where $J$ is the Gram matrix 418 of $\omega$. In particular $h$ is invertible, and inverting gives 419 $h^{-1}J^{-1}(h^{\mathsf T})^{-1}=J^{-1}$, i.e.\ $J^{-1}=hJ^{-1}h^{\mathsf T}$. 420 Since $J^2=-\mathbb{I}$, i.e.\ $J^{-1}=-J$, this is $hJh^{\mathsf T}=J$, so 421 $g:=h^{\mathsf T}$ again preserves $\omega$. \\ 422 Let $\Phi_g$ be the corresponding automorphism of $A_n(k)$, extending $u\mapsto g(u)$ on 423 generators and let $M$ be the matrix whose rows are the images of the generators, i.e.\ 424 $g(u_i)=\sum_j M_{ij}u_j$. Since $\Phi_g$ sends each generator to a 425 degree-one element, it maps $B_m$ into $B_m$ for all $m$, and since 426 Bernstein symbols are multiplicative, a standard monomial of degree $N$ is 427 mapped to an element of $B_N$ whose class modulo $B_{N-1}$ is obtained by 428 substituting $u_i\mapsto\sum_j M_{ij}u_j$ in the corresponding commutative 429 monomial. Hence $\Phi_g(d)\equiv s(Mu)\bmod B_{N-1}$. \\ 430 As $M$ is invertible, 431 $s(Mu)$ is again a nonzero homogeneous polynomial of degree $N$, so 432 $\deg_B\Phi_g(d)=N$ and the Bernstein symbol of $\Phi_g(d)$ is the 433 function $s_g\colon a\mapsto s(Ma)$. Since the rows of $M$ are the coordinate vectors of the $g(u_i)$, the matrix of 434 $g$ is $M^T$; as $g=h^T$, so the matrix of $h$ is $M$, i.e.\ $Me_j=h(e_j)$. Hence the 435 value of $s_g$ at $e_p$ is 436 \[ s_g(e_p) = s(Me_p)=s(h(e_p))=s(v)\neq0 .\] 437 We now set $c:=s(v)$. Since $c$ is a nonzero scalar, it is a central unit of 438 $A_n(k)$, and $\tilde d:=c^{-1}\Phi_g(d)$ satisfies $\deg_B\tilde d=N$ and 439 $\sigma_N^B(\tilde d)(e_p)=1$. 440 Using the normal form $A_n(k)=\bigoplus_j p^{\,j}C$, we can write 441 $\tilde d=\sum_{j = 0}^N p^{\,j}a_j$ with $a_j\in C$. The Bernstein filtration is 442 compatible with this decomposition, so 443 $N=\deg_B\tilde d=\max_j\,(j+\deg_B a_j)$, and we have $\deg_B a_j\le N-j$ for 444 all $j$; in particular $a_j=0$ for $j>N$ and $a_N\in C\cap B_0=k$. 445 The symbol is 446 \[ 447 \sigma_N^B(\tilde d)=\sum_{j+\deg_B a_j=N}\sigma_1^B(p)^{\,j}\,\sigma_{N-j}^B(a_j), 448 \] 449 For $j<N$ the factors $\sigma_N^B(a_j)$ must have a positive degree and must therefore be non-constant in a coordinate other than $p$, so they vanish at $e_p$. Because we defined $\tilde d$ to be $c^{-1}\Phi_g(d)$ , evaluating at $e_p$ gives $a_N=\sigma_N^B(\tilde d)(e_p)=c^{-1}\sigma_N^B(\Phi_g(d))(e_p) = c^{-1}s_g(e_p) = c^{-1}s(v)=1$, and therefore 450 \[ \tilde d=p^{\,N}+\sum_{j<N}p^{\,j}a_j,\qquad a_j\in C,\quad 451 \deg_B a_j\le N-j. \] 452 \end{proof} 453 454 \AIadd{\formalref[{}]{% 455 \leandecl{Stafford38/Paper/Section3/Monic.lean}{243}{Stafford38.Paper.Section3.proposition_3_2}, 456 \leandecl{Stafford38/Paper/Section3/MonicForm.lean}{44}{Stafford38.Paper.Section3.IsMonicNormalForm}}} 457 458 459 \begin{remark}\label{rem:transport} 460 Since $\Phi_g$ is a ring automorphism, the identity $1=\tilde dR+x_1^N\tilde dS$ would transport back to 461 \[ 1=d\bigl(c^{-1}\Phi_g^{-1}(R)\bigr)+F\,d\bigl(c^{-1}\Phi_g^{-1}(S)\bigr), 462 \qquad F=\Phi_g^{-1}(x_1)^N.\] 463 \end{remark} 464 465 \AIadd{\formalref[{}]{\leandecl{Stafford38/Paper/Section3/Transport.lean}{77}{Stafford38.Paper.Section3.remark_3_3}, 466 \leandecl{Stafford38/Paper/Section3/Transport.lean}{92}{Stafford38.Paper.Section3.remark_3_3_of_Q_eq_zero}}} 467 468 We will now define the modules 469 \begin{equation} \mathcal I=\tilde dA_n+x_1^N\tilde dA_n,\qquad Q=A_n/\mathcal I. 470 \label{eq:I-Q}\end{equation} 471 472 473 474 We now use the order filtration on $A_n(k)$ (where coordinates have 475 degree zero and derivatives have degree one) to characterize $P = 476 \sigma^{\mathrm{ord}}_N(\tilde d)$, the order principal symbol of $\tilde 477 d$ and show that $P$ annihilates the graded module $\gr Q$. 478 479 \begin{definition}\label{def:monic-xi1} 480 We say a polynomial $P=\sum_{i=0}^Nc_i\,\xi_1^i\in 481 k[\xi_1,\ldots,\xi_n]$, written as a polynomial in $\xi_1$ with 482 coefficients $c_i \in k[\xi_2,\ldots,\xi_n]$, is \emph{monic of degree $N$ in 483 $\xi_1$} if $c_N=1$. 484 \end{definition} 485 \AIadd{\formalref{\leandecl{Stafford38/Paper/Section3/OrderSymbol.lean}{52}{Stafford38.Paper.Section3.IsMonicInXi1}}} 486 487 \begin{lemma}\label{lem:order-symbol} 488 Let $\tilde d=p^N+\sum_{j<N}p^{\,j}a_j$ with $a_j\in C$, $\deg_Ba_j\le N-j$, 489 as in \eqref{eq:monic-d}, and set $N=\deg_B\tilde d$. Then $\tilde d$ has 490 order $N$, and its order principal symbol is a polynomial in $\xi$ alone, 491 \begin{equation} \sigma_N^{\mathrm{ord}}(\tilde d)=P(\xi),\qquad 492 P\in k[\xi_1,\ldots,\xi_n],\qquad \deg P=N. 493 \label{eq:P}\end{equation} 494 $P$ is monic of degree $N$ in $\xi_1$ and in particular 495 \begin{equation} P(1,0,\ldots,0)=1. 496 \label{eq:P-e1}\end{equation} 497 \end{lemma} 498 499 \begin{proof} 500 Since $\deg_B\tilde d=N$, every standard (PBW) monomial $x^\alpha\partial^\beta$ 501 occurring in $\tilde d$ satisfies $|\alpha|+|\beta|\le N$. A monomial 502 contributing to the order-$N$ part has $|\beta|=N$, hence $|\alpha|=0$: it 503 is a pure $\partial^\beta$ with scalar coefficient. Thus $\tilde d$ has 504 order $N$, and its order-$N$ part is a $k$-linear combination of 505 $\partial^\beta$ with $|\beta|=N$, giving $\sigma^{\mathrm{ord}}_N(\tilde 506 d)=P(\xi_1, \dots, \xi_n)$, homogeneous of degree $N$ with coefficients in $k$. 507 508 The monomial $p^N=\partial_1^N$ occurs in $\tilde d$ with coefficient $1$. 509 Every other order-$N$ monomial comes from some $p^ja_j$ with $j<N$ and 510 $a_j\in C$ (hence involving no further factor of $p=\partial_1$), so it 511 carries a positive power of some transverse $\partial_i$, $i>1$. Such a 512 monomial contributes only to coefficients $c_i$ with $i<N$, never to 513 $c_N$; hence $c_N$ receives no contribution besides the coefficient of 514 $p^N$ itself, giving $c_N=1$ identically, i.e.\ $P$ is monic of degree $N$ 515 in $\xi_1$. Evaluating at $(1,0,\ldots,0)$ then gives $P(1,0,\ldots,0)=1$. 516 \end{proof} 517 \AIadd{\formalref{\leandecl{Stafford38/Paper/Section3/OrderSymbol.lean}{211}{Stafford38.Paper.Section3.lemma_3_5}}} 518 519 \begin{proposition}\label{prop:charP} 520 Equip $Q$ with the order filtration $Q_m:=qF_m=\operatorname{im}(F_m\to Q)$ 521 induced by the cyclic generator $q = [1]$, the class of $1$ in $Q$ and write $P=\sigma^{\mathrm{ord}}_N(\tilde d)$. Then 522 \begin{equation}\label{eq:P-annihilates} 523 P\in\Ann_{k[x,\xi]}\gr Q. 524 \end{equation} 525 \end{proposition} 526 \begin{proof} 527 $Q_\bullet$ is a good filtration because $Q$ is cyclic. Let $\bar q$ be the class of $q$ in $\gr Q$. We know that $q\,\tilde d=[\tilde d]=0$ in $Q$ (because $\tilde d \in \cal I$) and $\tilde d$ has order $N$ (Lemma \ref{lem:order-symbol}), so $q\tilde d \in Q_N$. Therefore $\bar q\cdot P$ , its image in $Q_N/Q_{N-1}$, must also be zero, and hence $P$ annihilates the class $\bar q$ of $q$ in $\gr Q$. 528 Since $\gr Q$ is generated over $k[x,\xi]$ by $\bar q$, we get $P\in\Ann_{k[x,\xi]}\gr Q$. 529 \end{proof} 530 531 \section{$x_1$ as a surjective map}\label{sec:euler} 532 The first goal of this section will be to show that for $Q$ defined as in Eq. \eqref{eq:I-Q}, we have $Q = Qx_1$ (Proposition \ref{prop:x-surjective}). We will then use this result, as well as Gabber's theorem (Theorem \ref{ithm:gabber}) to establish two contradictory length estimations for the graded module $\gr Q/(\gr Q)x_1$ that we have assumed to be nonzero, concluding that therefore, also $\gr Q = (\gr Q)x_1$ (Corollary \ref{cor:Uzero}).\\ 533 534 \AIadd{\formalref{\leandecl{Stafford38/Paper/Section3/Transport.lean}{35}{Stafford38.Paper.Section3.idealI} 535 $=\operatorname{span}\{d,x^Nd\}$ as a right ideal, 536 \leandecl{Stafford38/Paper/Section3/Transport.lean}{40}{Stafford38.Paper.Section3.Q}}} 537 538 We define $E=x_1p$. The commutator of $E$ defines a grading of $A_n$: 539 $x_1$ has Euler degree $1$, $p$ has Euler degree $-1$, and the transverse Weyl algebra 540 has Euler degree zero. Let $A_{\geq0}$ denote the nonnegative part of $A_n$ in this grading. In PBW form (with $x_i$ on the left of $\partial_i$), $A_{\geq0}$ is generated by the transverse algebra, $E$, and $x_1$. 541 542 543 By the commutation relations of the Weyl algebra, it holds that 544 \begin{align} 545 Ex_1=x_1(E+1),\qquad x_1E=(E-1)x_1,\nonumber \\ 546 Ep=p(E-1), \qquad pE=(E+1)p 547 \label{eq:nonnegative-pbw} 548 \end{align} 549 Additionally, $x_1$ commutes with the transverse algebra, so $x_1$ is normal in $A_{\geq 0}$ (i.e. for every $r \in A_{\geq 0}$, there exists $r'\in A_{\geq 0}$ such that $x_1r=r'x_1$ \cite[4.1.10]{MR}) . This also means that every element of Euler degree $\ge1$ lies in $A_{\ge0}x_1$. 550 \AIadd{\formalref[{}]{% 551 \leandecl{Stafford38/Paper/Section4/Euler.lean}{65}{Stafford38.Paper.Section4.eulerDegree}, 552 \leandecl{Stafford38/Paper/Section4/Euler.lean}{111}{Stafford38.Paper.Section4.nonnegativePart}, 553 \leandecl{Stafford38/Paper/Section4/Euler.lean}{129}{Stafford38.Paper.Section4.positivePart}, 554 \leandecl{Stafford38/Paper/Section4/Euler.lean}{637}{Stafford38.Paper.Section4.step_nonnegative_pbw}, 555 \leandecl{Stafford38/Paper/Section4/Euler.lean}{210}{Stafford38.Paper.Section4.step_coordinate_normal}}} 556 557 \begin{lemma}\label{lem:euler} 558 For every $m\ge1$, 559 \begin{equation}\label{eq:eulerprod} 560 x_1^{m}p^{\,m}=\prod_{i=0}^{m-1}(E-i), \qquad p^{\,m} x_1^{m}=\prod_{i=1}^{m}(E+i), 561 \end{equation} 562 the factors in each product commuting with one another. 563 \end{lemma} 564 565 \begin{proof} 566 Since both products in \eqref{eq:eulerprod} are polynomials in the single element $E$, their factors commute. 567 We induct on $m$; both identities are trivial for $m=1$. Assuming the 568 first identity for $m$, 569 \[\begin{aligned} 570 x_1^{m+1}p^{\,m+1} 571 &=x_1\Bigl(\prod_{i=0}^{m-1}(E-i)\Bigr)p 572 =\Bigl(\prod_{i=0}^{m-1}(E-1-i)\Bigr)x_1p\\ 573 &=\Bigl(\prod_{i=1}^{m}(E-i)\Bigr)E 574 =\prod_{i=0}^{m}(E-i), 575 \end{aligned}\] 576 using $x_1E=(E-1)x_1$ repeatedly and the definition of $E$. The second identity follows 577 identically from $Ep=p(E-1)$. 578 %\[p^{m+1}x_1^{m+1}=p\Bigl(\prod_{i=1}^m(E+i)\Bigr)x_1=\Bigl(\prod_{i=1}^m(E+1+i)\Bigr)px_1=\Bigl(\prod_{i=2}^{m+1}(E+i)\Bigr)(E+1).\] 579 \end{proof} 580 \AIadd{\formalref{\leandecl{Stafford38/Paper/Section4/Euler.lean}{348}{Stafford38.Paper.Section4.lemma_4_1}}} 581 582 \begin{lemma}\label{lem:euler-residue} 583 There exists an element $u_0\in A_{\geq0}$ such that 584 \begin{equation} 1+u_0x_1\in \mathcal I. 585 \label{eq:euler-residue}\end{equation} 586 \end{lemma} 587 588 \begin{proof} 589 From Lemma \ref{lem:euler} we define 590 \[p^Nx_1^N=\prod_{r=1}^N(E+r)=:R_+(E),\qquad 591 x_1^Np^N=\prod_{r=0}^{N-1}(E-r)=:R_-(E).\] 592 The roots of $R_+$ and $R_-$ are disjoint (the sets $\{-1,\ldots,-N\}$ and $\{0,\ldots,N-1\}$ of integers, embedded in the prime field 593 $\mathbb Q\subset k$), so $\gcd(R_+,R_-)=1$ in $\mathbb Q[E]$. By the 594 extended Euclidean algorithm, there exist $u,v\in\mathbb Q[E]\subset 595 k[E]$ with 596 \[ R_+(E)u(E)+R_-(E)v(E)=p^Nx_1^Nu(E)+x_1^Np^Nv(E)=1.\] 597 We write $\tilde d=p^N+\Sigma$ with $\Sigma = \sum_{j<N}p^{\,j}a_j$. Then 598 \[(\tilde d\,x_1^N)u(E)+(x_1^N\tilde d)v(E)=1+w\in \mathcal I,\] 599 where 600 \[w=\Sigma x_1^Nu(E)+x_1^N\Sigma v(E).\] 601 We can write each $a_j$ in $\Sigma$ in powers of $x_1$, with coefficients in the transverse Weyl algebra. A term $p^jx_1^lb$ coming from $p^ja_j$ has $l\geq0$ and $j<N$, and multiplication by $x_1^N$ gives that term Euler degree at least $N-j>0$. $u$ and $v$ only depend on $E$, so they have Euler degree 0. Therefore every term of $w$ has positive Euler degree, and thus $w\in A_{\geq0}x_1$ - or equivalently, there exists a $u_0\in A_{\geq0}$, such that $w=u_0x_1$. 602 \end{proof} 603 \AIadd{\formalref[{}]{% 604 \leandecl{Stafford38/Paper/Section4/Residue.lean}{242}{Stafford38.Paper.Section4.lemma_4_2}}} 605 606 607 \begin{proposition}\label{prop:x-surjective} 608 Right multiplication by $x_1$ on $Q$ is surjective. 609 \end{proposition} 610 \begin{proof} 611 Let $q$ be the class of $1$ in $Q$. By Equation \eqref{eq:euler-residue}, we have 612 $q=-qu_0x_1$, so $q\in(qA_{\geq0})x_1$. We use the normality of $x_1$ to show 613 \begin{equation} qA_{\geq0}=(qA_{\geq0})x_1. 614 \label{eq:normal-surj}\end{equation} 615 The inclusion $\supseteq$ holds because $x_1\in A_{\geq0}$ and 616 $A_{\geq0}$ is a ring. For $\subseteq$, let $r\in A_{\geq0}$ and choose 617 $r'$ with $x_1r=r'x_1$ (possible since $x_1$ is normal) then $qr=-(qu_0)x_1r=-(qu_0r')x_1\in(qA_{\geq0})x_1$ since $u_0r'\in A_{\geq0}$. 618 619 $qA_{\geq0}$ is stable under right multiplication by $p$: If $m=m'x_1$, then 620 \[ mp=m'(x_1p)=m'E\in qA_{\geq0}.\] 621 It is of course also stable under $C$, since $C=A_{n-1}(k)[x_1]\subseteq A_{\ge0}$ because no element of $C$ contains $p$. Since $C$ and $p$ together generate $A_n$, we have $qA_{\geq0}=qA_n=Q$. Equation \eqref{eq:normal-surj} is therefore $Q=Qx_1$. 622 \end{proof} 623 \AIadd{\formalref[{}]{\leandecl{Stafford38/Paper/Section4/Surjectivity.lean}{223}{Stafford38.Paper.Section4.proposition_4_3}}} 624 625 626 627 \begin{definition}[{\cite{Stacks}[Tag 00CM]}] \label{def:localization} 628 Let $R$ be a commutative ring and $S\subseteq R$ a multiplicative subset 629 (i.e.\ $1\in S$ and $S$ is closed under products). The \emph{localization} 630 $S^{-1}R$ is the ring of fractions $a/s$ ($a\in R$, $s\in S$), where 631 \[ \frac{a}{s}=\frac{a'}{s'}\iff t(s'a-sa')=0\ \text{ for some }t\in S,\] 632 with the usual addition and multiplication of fractions. Analogously, for an $R$-module 633 $N$, the \emph{localization} $S^{-1}N$ is the $S^{-1}R$-module of fractions 634 $y/s$ ($y\in N$, $s\in S$), with $y/s=y'/s'$ iff $t(s'y-sy')=0$ for some 635 $t\in S$. For a prime $\mathfrak p\subset R$ we write 636 $R_{\mathfrak p}=S^{-1}R$ and $N_{\mathfrak p}=S^{-1}N$ for 637 $S=R\setminus\mathfrak p$. Localization is exact: it sends exact sequences 638 of $R$-modules to exact sequences of $S^{-1}R$-modules \cite[Tag 00CS]{Stacks}. 639 \end{definition} 640 \begin{definition}[{\cite{Stacks}[Definition 10.40.1]}]\label{def:supp} 641 Let $R$ be a ring and let $M$ be an $R$-module. The \emph{support} of $M$ is the set 642 \[\Supp_R(M) = \{\mathfrak{p} \in \Spec(R) | M_{\mathfrak{p}} \neq 0\}.\] 643 \end{definition} 644 645 646 \begin{definition}[\mbox{\cite{coutinho, cohn}, \cite[Tags~02LY, 00J3, 00IV]{Stacks}}]\label{def:length} 647 The length $\ell$ of a $T$-module of finite length $N$ is defined as the number of steps in a composition series of $N$ such that 648 \[0 = N_0\subsetneq N_1\subsetneq \dots \subsetneq N_{\ell} = N\] 649 where each successive quotient $N_i/N_{i-1}$ is simple (i.e. has no proper nontrivial submodules). The length is well-defined, as it does not depend on the choice of the composition series \cite[Tag 00J3]{Stacks}. Lengths are additive for short exact sequences. 650 \end{definition} 651 \AIadd{\formalref{\leandecl{Stafford38/Paper/Section4/Support.lean}{59}{Stafford38.Paper.Section4.definition_4_6}; 652 \texttt{Module.length} (Mathlib); finite length of the localized modules: 653 \leanlib{AlgebraicAnalysis/Module/MinimalSupportKernelCokernelLengths.lean}{48}{AlgebraicAnalysis.localized_kernel_and_cokernel_isFiniteLength}}} 654 \begin{lemma}\label{lem:supp-criterion} 655 Let $R$ be a commutative ring, $N$ an $R$-module and $\mathfrak p\subset R$ a 656 prime. Then $\mathfrak p\in\operatorname{Supp}_R(N)$ if and only if there is 657 $y\in N$ with $\operatorname{Ann}_R(y)\subseteq\mathfrak p$. 658 \end{lemma} 659 \AIadd{\formalref{\leandecl{Stafford38/Paper/Section4/Support.lean}{111}{Stafford38.Paper.Section4.lemma_4_7}}} 660 \begin{proof} 661 In $N_{\mathfrak p}$, a 662 fraction is zero when it can be cleared by an element outside $\mathfrak p$. 663 For $y/1$ this means: there is $u\in R\setminus\mathfrak p$ with $uy=0$. 664 Such a $u$ lies in $\operatorname{Ann}_R(y)$ and not in $\mathfrak p$. So 665 \[y/1=0\iff\operatorname{Ann}_R(y)\not\subseteq\mathfrak p,\] 666 and, taking the contrapositive, 667 \[y/1\neq0\iff\operatorname{Ann}_R(y)\subseteq\mathfrak p.\] 668 \emph{$(\Leftarrow)$} Suppose $y\in N$ has 669 $\operatorname{Ann}_R(y)\subseteq\mathfrak p$. Then, $y/1\neq0$ in 670 $N_{\mathfrak p}$. So $N_{\mathfrak p}$ has a nonzero element, hence 671 $N_{\mathfrak p}\neq0$ and $\mathfrak p\in\operatorname{Supp}_R(N)$.\\ 672 \emph{$(\Rightarrow)$} Suppose $N_{\mathfrak p}\neq0$. Pick a nonzero 673 element of $N_{\mathfrak p}$. It has the form $y/s$ with $y\in N$ and 674 $s\notin\mathfrak p$. The element $s$ is invertible in $N_{\mathfrak p}$, so 675 \[y/s=0\iff y/1=0.\] 676 Since $y/s\neq0$, we get $y/1\neq0$, and therefore 677 $\operatorname{Ann}_R(y)\subseteq\mathfrak p$, which is the desired $y$. 678 \end{proof} 679 680 For the next part, we write $M=\operatorname{gr}Q$ and denote the right multiplication by $x_1$ by $f:M\rightarrow M$. We furthermore define 681 \[ 682 \begin{gathered} 683 V=\ker(f),\qquad U=\operatorname{coker}(f) =M/x_1M\\ 684 T=k[x_2,\ldots,x_n,\xi_2,\ldots,\xi_n], 685 \end{gathered} 686 \] 687 hence $T[x_1,\xi_1] = k[x,\xi]$. If $n=1$, we define $T=k$. 688 We want to prove that $U=0$ by comparing 689 two lengths over a localization of $T$. 690 691 692 \begin{lemma}\label{lem:tangential-finiteness} 693 The modules $U$ and $V$ are finite over $T$, and 694 $\operatorname{Supp}_T V\subseteq\operatorname{Supp}_T U$. 695 If $\mathfrak q$ is a minimal prime in $\operatorname{Supp}_T U$, then 696 $U_{\mathfrak q}$ and $V_{\mathfrak q}$ have finite length over 697 $T_{\mathfrak q}$. 698 \end{lemma} 699 700 \begin{proof} 701 The good quotient filtration makes $M = \operatorname{gr} Q$ finitely generated over $\operatorname{gr} A_n = k[x,\xi] = T[{x_1}, \xi_1]$. 702 Since $P$ annihilates $M$ (Proposition~\ref{prop:charP}), the action of 703 $T[x_1,\xi_1]$ on $M$ factors through $T[x_1,\xi_1]/(P)$. As $P$ is monic of 704 degree $N$ in $\xi_1$, in $T[x_1,\xi_1]/(P)$ we have $\xi_1^N=-\sum_{i<N}c_i\xi_1^i$ with 705 $c_i\in T$, so $T[x_1,\xi_1]/(P)$ is generated as a $T[x_1]$-module by $1,\xi_1,\dots,\xi_1^{N-1}$. 706 If $m_1,\dots,m_l$ generate $M$ over $T[x_1,\xi_1]/(P)$, then the products $\xi_1^im_s$ 707 ($0\le i<N$, $1\le s\le l$) generate $M$ over $T[x_1]$. Hence $M$ is finite over 708 $T[x_1]$. 709 Because $M$ is finite, and $T[{x_1}]$ is a Noetherian ring, the kernel $V$ and cokernel $U$ of multiplication in $M$ by ${x_1}$ are also finite over $T[{x_1}]$. Since they are annihilated by ${x_1}$, they are finite over $T$. %Noetherian theorem 3.5 https://www.cambridge.org/core/services/aop-cambridge-core/content/view/5DD13688B6CC1628C3FD7A318CD479FE/9781139171762c1_p1-19_CBO.pdf/commutative-rings-and-modules.pdf 710 711 Let $\mathfrak p\in\operatorname{Supp}_T(V)$. By Lemma \ref{lem:supp-criterion}, we can choose $v\in V=\ker f$ with 712 $\operatorname{Ann}_T(v)\subseteq\mathfrak p$. Let us now consider the annihilators of $v$ in $k[x,\xi]=T[x_1, \xi_1]$, defining $S$ as the image of $T\setminus \mathfrak p$ in $k[x,\xi]$: $\operatorname{Ann}_{k[x,\xi]}(v)\cap S = \varnothing$, because were $r \in T\setminus\mathfrak p$ and in $\operatorname{Ann}_{k[x,\xi]}(v)$, then $v$ would be annihilated by $r$, which in turn would imply that $r$ is also in $\operatorname{Ann}_T(v)$. Since $\operatorname{Ann}_T(v) \subseteq \mathfrak p$, but we assumed $r \in T\setminus\mathfrak p$, this is a contradiction. There must therefore exist a prime $\mathfrak Q \in k[x,\xi]$ with $\mathfrak Q \cap S = \varnothing$ and $\operatorname{Ann}_{k[x,\xi]}(v) \subseteq \mathfrak Q$. Since $v \in V = \ker f$, by Lemma \ref{lem:supp-criterion}, it follows that $\mathfrak Q \in \Supp_{k[x,\xi]}(\ker f).$\\ 713 From Vasconcelos' theorem \cite[Theorem~2.4]{matsumura} \cite{vasconcelos} follows that $\operatorname{Supp}_{k[x,\xi]}(\ker f)\subseteq \operatorname{Supp}_{k[x,\xi]}(\operatorname{coker}f)$: 714 We prove the contrapositive. Let $\mathfrak Q$ be a prime of ${k[x,\xi]}$ with 715 $\mathfrak Q\notin\operatorname{Supp}_{k[x,\xi]}(\operatorname{coker}f)$, i.e.\ 716 $(\operatorname{coker}f)_{\mathfrak Q}=0$. 717 Since localization is exact, $(\operatorname{coker}f)_{\mathfrak Q}= 718 \operatorname{coker}(f_{\mathfrak Q})$, so 719 $\operatorname{coker}(f_{\mathfrak Q})=0$: the endomorphism 720 $f_{\mathfrak Q}\colon M_{\mathfrak Q}\to M_{\mathfrak Q}$ is therefore surjective. 721 The localization $M_{\mathfrak Q}$ is a finite module over the Noetherian ring 722 ${k[x,\xi]}_{\mathfrak Q}$, and $f_{\mathfrak Q}$ is ${k[x,\xi]}_{\mathfrak Q}$-linear, because 723 $f$ is multiplication by $x_1\in {k[x,\xi]}$ and ${k[x,\xi]}$ is commutative. By Vasconcelos' 724 theorem, a surjective endomorphism of a finite module is injective. Hence 725 $f_{\mathfrak Q}$ is injective, and by exactness of localization, 726 $(\ker f)_{\mathfrak Q}=\ker(f_{\mathfrak Q})=0$, i.e.\ 727 $\mathfrak Q\notin\operatorname{Supp}_{k[x,\xi]}(\ker f)$. The contrapositive is therefore $\mathfrak Q\in\operatorname{Supp}_{k[x,\xi]}(\ker f)$, 728 then $\mathfrak Q\in\operatorname{Supp}_{k[x,\xi]}(\operatorname{coker}f) = \Supp_{k[x,\xi]}(U).$ By 729 Lemma~\ref{lem:supp-criterion} there therefore exists an $u\in U$ with 730 $\operatorname{Ann}_{k[x,\xi]}(u)\subseteq\mathfrak Q$. 731 To get back to the support over $T$, we define $r\in\operatorname{Ann}_T(u)$. Then $ru=0$, so 732 $r\in\operatorname{Ann}_{k[x,\xi]}(u)\subseteq\mathfrak Q$. If $r\notin\mathfrak p$, 733 the image of $r$ would lie in $\mathfrak Q\cap S=\varnothing$, which is 734 impossible. Hence $r\in\mathfrak p$, i.e.\ 735 $\operatorname{Ann}_T(u)\subseteq\mathfrak p$, and 736 Lemma~\ref{lem:supp-criterion} gives 737 $\mathfrak p\in\operatorname{Supp}_T(U)$. Since $\mathfrak p\in 738 \operatorname{Supp}_T(V)$ was arbitrary, 739 $\operatorname{Supp}_T(V)\subseteq\operatorname{Supp}_T(U)$. 740 741 Let now $\mathfrak q$ be a minimal prime in $\operatorname{Supp}_TU$. The primes of 742 $T_{\mathfrak q}$ correspond to the primes $\mathfrak p'\subseteq\mathfrak q$ of $T$\cite[Prop 3.11]{atiyah}, and 743 $(U_{\mathfrak q})_{\mathfrak p'}=U_{\mathfrak p'}$. Hence $\operatorname{Supp}(U_{\mathfrak q})$ 744 corresponds to the primes $\mathfrak p'\subseteq\mathfrak q$ lying in $\operatorname{Supp}_TU$; 745 by minimality of $\mathfrak q$ the only such prime is $\mathfrak q$ itself, so 746 $\operatorname{Supp}(U_{\mathfrak q})=\{\mathfrak qT_{\mathfrak q}\}$, the set consisting 747 only of the maximal ideal of $T_{\mathfrak q}$. The same argument applies to $V$: a prime 748 $\mathfrak p'\subseteq\mathfrak q$ in $\operatorname{Supp}_TV$ lies in $\operatorname{Supp}_TU$ 749 by the inclusion just proved, hence equals $\mathfrak q$, so 750 $\operatorname{Supp}(V_{\mathfrak q})\subseteq\{\mathfrak qT_{\mathfrak q}\}$. 751 752 Now $T_{\mathfrak q}$ is a Noetherian local ring with maximal ideal $\mathfrak qT_{\mathfrak q}$, 753 and $U_{\mathfrak q}$, $V_{\mathfrak q}$ are finite modules (localizations of the finite 754 $T$-modules $U$, $V$) whose support is contained in $\{\mathfrak qT_{\mathfrak q}\}$. For a 755 finite module $L$ over $T_{\mathfrak q}$ $\operatorname{Supp}L$ is equal to the vanishing locus of its annihilators \cite[Tag 00L2]{Stacks}, 756 so the only prime containing $\operatorname{Ann}L$ is the maximal ideal (if $L\neq0$). Hence 757 $T_{\mathfrak q}/\operatorname{Ann}L$ is a Noetherian ring in which every prime is maximal, 758 i.e.\ an Artinian ring, and a finite module over an Artinian ring has finite length \cite[Thm. 8.5]{atiyah}\cite[Tag 00KH]{Stacks}. Thus 759 $U_{\mathfrak q}$ and $V_{\mathfrak q}$ have finite length over $T_{\mathfrak q}$ 760 ($V_{\mathfrak q}$ possibly zero). 761 \end{proof} 762 \AIadd{\formalref{\leandecl{Stafford38/Paper/Section4/TangentialFiniteness.lean}{164}{Stafford38.Paper.Section4.lemma_4_8}}} 763 764 765 \begin{lemma}\label{lem:page-inequality} 766 For every minimal prime $\mathfrak q$ in $\operatorname{Supp}_T U$, 767 \[ 768 \ell(U_{\mathfrak q}) 769 \le \ell(V_{\mathfrak q}) 770 \] over $T_{\mathfrak q}.$ 771 \end{lemma} 772 773 \begin{proof} 774 To prove the inequality we set up a spectral sequence measuring how $\ker(f)$ and 775 $\operatorname{coker}(f)$ acting on $Q$ itself compare, order by order, to $V$ and 776 $U$, acting on $\gr Q$. 777 778 Consider the filtered complex $Q\xrightarrow{f}Q$, with decreasing filtration $G^p=F_{-p}Q$, where $F$ is 779 the order filtration. Since ${x_1}$ has order $0$, $f(G^p)\subseteq G^p$ for every 780 $p$; the induced map on $G^p/G^{p+1}$, summed over $p$, is exactly the action of 781 ${x_1}$ on $M=\operatorname{gr}Q$. This is the crudest approximation to $\ker_Q(f)$ and 782 $\operatorname{coker}_Q(f)$: it correctly identifies $V,U$ as approximations, but it 783 does not yet see elements that vanish to higher order than this naive count 784 predicts. The spectral sequence refines this approximation in successive stages, 785 called \emph{pages} (see also \cite[Construction 5.4.6]{weibel}): for each $r\ge0$, the source and target pages are 786 \[ 787 A_r=\bigoplus_{p\in\mathbb Z}\frac{G^p\cap f^{-1}(G^{p+r})}{G^{p+1}\cap f^{-1}(G^{p+r})}, 788 \qquad 789 B_r=\bigoplus_{p\in\mathbb Z}\frac{G^p}{(G^p\cap f(G^{p-r+1}))+G^{p+1}}, 790 \] 791 where the degree-$p$ summand of $A_r$ isolates elements of $G^p$ whose image under 792 $f$ drops $r$ extra filtration levels (candidates for the kernel, accurate to 793 order $r$), and the degree-$p$ summand of $B_r$ isolates the part of $G^p$ not yet 794 hit by $f$ from $r-1$ levels shallower. 795 796 Applying $f$ to a representative gives a well-defined map $d_r:A_r\to B_r$, 797 shifting the summand index by $r$; one checks directly from these definitions that 798 \[ A_0=B_0=M,\qquad d_0={x_1},\qquad A_{r+1}=\ker d_r,\qquad B_{r+1}=\operatorname{coker}d_r,\] 799 so that each page is obtained from the previous one by taking kernel and cokernel 800 of $d_r$, exactly as in the construction of a spectral sequence from a filtered 801 complex. 802 803 These pages carry a $T$-module structure. The pages $A_r,B_r$ are 804 subquotients of $Q$, which a priori carries an action of the transverse Weyl 805 algebra $A_{n-1}(k)$ by right multiplication. This action commutes with 806 $f$, since the transverse algebra commutes with $x_1$, so the maps 807 $d_r$ are compatible with it. Two transverse elements commute only up to a 808 commutator, which lowers the order filtration by one. On the page 809 quotients such lower-order terms vanish, so the induced actions commute, and 810 the commutative ring $T$ acts on every $A_r$ and $B_r$, via the symbols of 811 the transverse generators. The maps $d_r$ and the successor-page maps are 812 $T$-linear. In particular, this means that localizing at $T\setminus\mathfrak q$ and 813 measuring lengths over $T_{\mathfrak q}$ is a well-defined action for the pages. 814 815 Now we fix a minimal prime $\mathfrak q$ in $\operatorname{Supp}_TU$ and localize at 816 $T\setminus\mathfrak q$. We localize the $T$-modules 817 $A_r,B_r$ and their $T$-linear maps. Exactness of localization preserves the 818 successor-page kernel and cokernel identities. By 819 Lemma~\ref{lem:tangential-finiteness}, $(A_1)_{\mathfrak q}=V_{\mathfrak q}$ and 820 $(B_1)_{\mathfrak q}=U_{\mathfrak q}$ have finite length over $T_{\mathfrak q}$; since each 821 $(A_{r+1})_{\mathfrak q}\subseteq (A_r)_{\mathfrak q}$ and $(B_{r+1})_{\mathfrak q}$ is a quotient of $(B_r)_{\mathfrak q}$, all later pages 822 have finite length as well. 823 824 825 The maps $(B_1)_{\mathfrak q}\to (B_r)_{\mathfrak q}$ are surjective by construction. 826 Let $K_r:=\ker((B_1)_{\mathfrak q}\to (B_r)_{\mathfrak q})$. Since $(B_1)_{\mathfrak q}\to (B_{r+1})_{\mathfrak q}$ factors through $(B_1)_{\mathfrak q}\to (B_r)_{\mathfrak q}\to (B_{r+1})_{\mathfrak q}$, we have 827 $K_r\subseteq K_{r+1}$, so $(K_r)_{r\ge1}$ is an increasing chain of submodules of $(B_1)_{\mathfrak q}$. 828 This chain exhausts $(B_1)_{\mathfrak q}$: Before localization, take a homogeneous 829 class in $B_1$, represented by $z\in G^p$. Since $Q=Qx_1$ by Proposition \ref{prop:x-surjective}, choose one preimage 830 $z=wx_1$. Exhaustiveness of the filtration means that $w$ must be in $G^s$ for some $s$; choose $r\ge1$ with 831 $p-r+1\le s$. Then $w\in G^{p-r+1}$, so $z$ belongs to the denominator of 832 $B_r$ and its class vanishes. Every class is a finite sum of homogeneous 833 classes and hence vanishes on a sufficiently late page. The same holds for a 834 localized class $z/a$, since its numerator already vanishes on such a page. 835 As $z\in B_1$ was arbitrary, so is $z/a$ in $(B_1)_{\mathfrak q}$, and $\bigcup_{r\ge1}K_r=(B_1)_{\mathfrak q}$. 836 Since $(B_1)_{\mathfrak q}$ has finite length, it is a Noetherian $T_{\mathfrak q}$-module. By the ascending 837 chain condition, the increasing chain $K_1\subseteq K_2\subseteq\cdots$ stabilizes: there exists a finite $R$ with 838 $K_R=K_{R+i}$ for all $i\geq 0$. Combined with $\bigcup_rK_r=(B_1)_{\mathfrak q}$, stabilization forces $K_R=(B_1)_{\mathfrak q}$. 839 Finally, $K_R=(B_1)_{\mathfrak q}$ means $(B_1)_{\mathfrak q}\to (B_R)_{\mathfrak q}$ must be the zero map: by the first isomorphism theorem, 840 $(B_1)_{\mathfrak q}/K_R\cong\operatorname{im}((B_1)_{\mathfrak q}\to (B_R)_{\mathfrak q})$, and $K_R=(B_1)_{\mathfrak q}$ gives $(B_1)_{\mathfrak q}/K_R=0$, so 841 $\operatorname{im}((B_1)_{\mathfrak q}\to (B_R)_{\mathfrak q})=0$. But $(B_1)_{\mathfrak q}\to (B_R)_{\mathfrak q}$ is also surjective, so $(B_R)_{\mathfrak q}=0$.\\ 842 843 At each stage $r\ge1$ the differential $(d_r)_{\mathfrak q}$ gives an exact sequence 844 \[ 0\to (A_{r+1})_{\mathfrak q}\to (A_{r})_{\mathfrak q}\xrightarrow{(d_r)_{\mathfrak q}}(B_{r})_{\mathfrak q}\to (B_{r+1})_{\mathfrak q}\to0,\] 845 which splits at $\operatorname{im}((d_r)_{\mathfrak q})$ into two short exact sequences. 846 Additivity of length gives 847 $\ell((A_{r})_{\mathfrak q})=\ell((A_{r+1})_{\mathfrak q})+\ell(\operatorname{im}(d_r)_{\mathfrak q})$ and 848 $\ell((B_{r})_{\mathfrak q})=\ell(\operatorname{im}(d_r)_{\mathfrak q})+\ell((B_{r+1})_{\mathfrak q})$; eliminating 849 $\ell(\operatorname{im}(d_r)_{\mathfrak q})$ yields, for every $r\ge1$, 850 \begin{equation}\label{eq:additivity} 851 \ell((B_r)_{\mathfrak q})+\ell((A_{r+1})_{\mathfrak q})=\ell((A_{r})_{\mathfrak q})+\ell((B_{r+1})_{\mathfrak q}). 852 \end{equation} 853 854 Summing this identity over $r=1,\ldots,R-1$ and reindexing the sums, most terms cancel on both sides, leaving % $\sum_{r=1}^{R-1}\ell((A_{r+1})_{\mathfrak q})=\sum_{r=2}^{R}\ell((A_{r})_{\mathfrak q})$, 855 %$\sum_{r=1}^{R-1}\ell((B_{r+1})_{\mathfrak q})=\sum_{r=2}^{R}\ell((B_{r})_{\mathfrak q})$, the terms $\ell((A_{r})_{\mathfrak q}),\ell((B_{r})_{\mathfrak q})$ for $2\le r\le R-1$ 856 %cancel on both sides, leaving 857 \[\ell((B_1)_{\mathfrak q})+\ell((A_R)_{\mathfrak q})=\ell((A_{1})_{\mathfrak q})+\ell((B_{R})_{\mathfrak q}).\] 858 Since $(B_{R})_{\mathfrak q}=0$ and $(A_{1})_{\mathfrak q}=V_{\mathfrak q}$, $(B_{1})_{\mathfrak q}=U_{\mathfrak q}$, this reads 859 \[ \ell(U_{\mathfrak q})+\ell((A_{R})_{\mathfrak q})=\ell(V_{\mathfrak q}).\] 860 As $\ell((A_{R})_{\mathfrak q})\ge0$, we conclude 861 \[ \ell(U_{\mathfrak q})\le\ell(V_{\mathfrak q}).\] 862 \end{proof} 863 864 \AIadd{\formalref{\leandecl{Stafford38/Paper/Section4/PageInequality.lean}{173}{Stafford38.Paper.Section4.lemma_4_9}; 865 exhaustion \leanlib{AlgebraicAnalysis/Module/UniformBoundaryVanishing.lean}{106}{AlgebraicAnalysis.exists_uniform_subsingleton_localized}}} 866 867 868 \noindent On the other hand, we can also establish a different inequality: 869 870 \begin{lemma}\label{lem:strict-support-inequality} 871 For every minimal prime $\mathfrak q$ in $\operatorname{Supp}_T U$, 872 \[ 873 \ell(U_{\mathfrak q}) 874 > \ell(V_{\mathfrak q}) 875 \] over $T_{\mathfrak q}$. 876 \end{lemma} 877 \begin{proof} 878 We start by showing, that by Gabber's involutivity theorem, $x_1$ cannot lie in any minimal prime. We then define a module $K_{\infty}$ as the kernel of multiplication by a power of $x_1$ over the localized $M_\mathfrak q$, which allows us to split the length of $U_{\mathfrak q}$ into two parts, $\ell(U_{\mathfrak q}) = \ell(V_{\mathfrak q}) + \ell\bigl((M_{\mathfrak q}/K_{\infty})/({x_1}(M_{\mathfrak q}/K_{\infty}))\bigr)$. It remains to show that the second term is nonzero. For this we use the fact that $x_1$ does not lie in any minimal prime to deduce that $x_1$ must therefore be invertible at the minimal primes while lying in every prime of the support of $U_{\mathfrak q}$. Nakayama's lemma then shows that $(M_\mathfrak q/K_{\infty})/x_1(M_\mathfrak q/K_{\infty})\neq0$. \\ 879 880 881 Let us consider $\operatorname{Ann}_{k[x,\xi]} M$. By Gabber's Theorem (\ref{ithm:gabber}, \cite{Gabber1981}), $J = \sqrt{\operatorname{Ann}_{k[x,\xi]} M}$ is involutive. 882 If $\mathfrak p$ is a minimal prime of $J$, 883 then $\mathfrak p$ is itself involutive, i.e. $\{\mathfrak p,\mathfrak p\}\subseteq\mathfrak p$ \cite[Thm. D.3.4]{HTT}. 884 \AIadd{\formalref{% 885 \leandecl{Stafford38/Paper/Section4/StrictInequality.lean}{405}{Stafford38.Paper.Section4.lemma_4_10}}} 886 From the involutivity of $\mathfrak p$ follows that no $\mathfrak p$ contains ${x_1}$: Assume ${x_1}\in \mathfrak p$. Since $P \in \mathfrak p$, also the Poisson brackets 887 \[ \{{x_1}, P\}, \quad \{{x_1}, \{x_1, P\}\}, \dots\] 888 lie in $\mathfrak p$. $\{{x_1},-\}$ differentiates once with respect to $\xi_1$, and because $P$ is homogeneous of degree $N$ and monic in $\xi_1$, its $N$th derivative is the nonzero constant $N!$ - therefore the $N$th bracket would put $N!$ in $\mathfrak p$. Since $\mathfrak p$ is a proper prime ideal, no nonzero constant can be in $\mathfrak p$, so we must conclude ${x_1} \notin \mathfrak p$. 889 This persists after localizing at $S=T\setminus\mathfrak q$: write $k[x,\xi]_{\mathfrak q}:=S^{-1}k[x,\xi]$. Since $M$ is finite, $\operatorname{Ann}(M_{\mathfrak q})=S^{-1}\operatorname{Ann}(M)$ \cite[Prop 3.14]{atiyah}, and primes of $k[x,\xi]_{\mathfrak q}$ correspond to primes of $k[x,\xi]$ disjoint from $S$ \cite[Prop 3.11(iv)]{atiyah}. A prime minimal among those containing $J$ and disjoint from $S$ is minimal over $J$ (any smaller prime containing $J$ is again disjoint from $S$). Hence every minimal prime of $\operatorname{Supp}(M_{\mathfrak q})$ is of the form $S^{-1}\mathfrak p$ with $\mathfrak p$ a minimal prime of $J$, and does not contain ${x_1}$. 890 891 Consider the increasing chain of kernels of powers of the localization $f_{\mathfrak q}$ of right multiplication with $x_1$ 892 $\operatorname{ker}(f_{\mathfrak q})$ $\subseteq \operatorname{ker}(f^2_{\mathfrak q})$ $\subseteq \dots$. 893 Since $M_{\mathfrak q}$ is Noetherian, this chain stabilizes. Let 894 $K_{\infty}:=\ker(f^m_{\mathfrak q})$ at a stage where it stabilizes, so $f_{\mathfrak q}(K_{\infty}) = {x_1}K_{\infty}\subseteq K_{\infty}$ and 895 $\ker(f^{m+1}_{\mathfrak q})=\ker(f_{\mathfrak q}^m)=K_{\infty}$. Each quotient $\ker(f_{\mathfrak q}^{i+1})/\ker(f_{\mathfrak q}^i)$ maps to 896 $V_{\mathfrak q}=\ker(f_{\mathfrak q})$ injectively via $v\mapsto {x_1}^iv$, so, as $V_{\mathfrak q}$ has 897 finite length, so does each quotient and hence $K_{\infty}$ itself. 898 899 Two elementary facts about $K_{\infty}$: 900 \begin{itemize} 901 \item ${f_{\mathfrak q}}^{-1}K_{\infty}=K_{\infty}$: if $v\in K_{\infty}$ then $f_{\mathfrak q}(v)\in K_{\infty}$ (as $f_{\mathfrak q}(K_{\infty})\subseteq K_{\infty}$); if $f_{\mathfrak q}(v)\in K_{\infty}=\ker(f_{\mathfrak q}^m)$ 902 then $f_{\mathfrak q}^{m+1}(v)=0$, so $v\in\ker({f_{\mathfrak q}}^{m+1})=K_{\infty}$. 903 \item $K_{\infty}\cap {x_1}M_{\mathfrak q}={x_1}K_{\infty} = f_{\mathfrak q}(K_{\infty})$: "$\supseteq$" is trivial. For "$\subseteq$", if $w=f_{\mathfrak q}(v)\in K_{\infty}$ for 904 some $v\in M_{\mathfrak q}$, then $v\in f_{\mathfrak q}^{-1}K_{\infty}=K_{\infty}$, so $w\in f_{\mathfrak q}(K_{\infty}) = x_1K_{\infty}$. 905 \end{itemize} 906 907 For any endomorphism $\varphi$ of a finite-length module $L$, 908 additivity along $0\to\ker \varphi\to L\to \varphi(L)\to0$ and $0\to \varphi(L)\to L\to L/\varphi(L)\to0$ gives 909 $\ell(\ker \varphi)=\ell(L/\varphi(L))$. We can apply this to $\varphi=f_{\mathfrak q}|_{K_{\infty}}$: since $\ker(f_{\mathfrak q})\subseteq K_{\infty}$, we 910 have $\ker(f_{\mathfrak q}|_{K_{\infty}})=\ker(f_{\mathfrak q})=V_{\mathfrak q}$, so with $f_{\mathfrak q}(K_{\infty}) = x_1K_{\infty},$ 911 \[\ell(K_{\infty}/{x_1}K_{\infty})=\ell(V_{\mathfrak q}).\] 912 913 On the other hand, the natural surjection $M_{\mathfrak q}/({x_1}M_{\mathfrak q})\to (M_{\mathfrak q}/K_{\infty})/({x_1}(M_{\mathfrak q}/K_{\infty}))$ has 914 the kernel $(K_{\infty}+({x_1}M_{\mathfrak q}))/({x_1}M_{\mathfrak q})\cong K_{\infty}/(K_{\infty}\cap ({x_1}M_{\mathfrak q}))=K_{\infty}/({x_1}K_{\infty})$, giving (we now write $\tilde E = (M_{\mathfrak q}/K_{\infty})$ for better readability) the exact sequence 915 \[0\longrightarrow K_{\infty}/({x_1}K_{\infty})\longrightarrow U_{\mathfrak q}\longrightarrow 916 \tilde E/({x_1}\tilde E)\longrightarrow0,\] 917 so by additivity and using $\ell(K_{\infty}/({x_1}K_{\infty}))=\ell(V_{\mathfrak q})$, 918 \[\ell(U_{\mathfrak q}) = \ell(V_{\mathfrak q}) + \ell\bigl(\tilde E/({x_1}\tilde E)\bigr).\] 919 It remains to show the last term is positive. 920 921 Since $\mathfrak q\in\operatorname{Supp}_TU$, 922 $U_{\mathfrak q}=M_{\mathfrak q}/({x_1}M_{\mathfrak q})\ne0$. We choose $\mathfrak P\in\operatorname{Supp}_{k[x,\xi]_{\mathfrak q}}(M_{\mathfrak q}/({x_1}M_{\mathfrak q}))$ 923 and a prime $\mathfrak p'\subseteq\mathfrak P$ minimal in $\operatorname{Supp}(M_{\mathfrak q})$ \cite[Tag 00E0]{Stacks}. By the above, ${x_1}\notin\mathfrak p'$, so ${x_1}$ is a unit in the localization at $\mathfrak p'$; as $K_{\infty}$ is annihilated by a power of ${x_1}$, $(K_{\infty})_{\mathfrak p'}=0$. From 924 $0\to (K_{\infty})_{\mathfrak p'}\to (M_{\mathfrak q})_{\mathfrak p'}\to \tilde E_{\mathfrak p'}\to0$ we get 925 $\tilde E_{\mathfrak p'}\cong (M_{\mathfrak q})_{\mathfrak p'}\ne0$, so $\mathfrak p'$ lies in the support of $M_{\mathfrak q}/K_{\infty}$ in $k[x,\xi]_{\mathfrak q}$. 926 Since support is closed under passing to larger primes \cite[Tag 00L2]{Stacks} and 927 $\mathfrak p'\subseteq\mathfrak P$, also $\mathfrak P\in\operatorname{Supp}_{k[x,\xi]_{\mathfrak q}}\tilde E$, i.e. 928 $\tilde E_{\mathfrak P}\ne0$. 929 930 Since $\mathfrak P\in\operatorname{Supp}_{k[x,\xi]_{\mathfrak q}}(M_{\mathfrak q}/({x_1}M_{\mathfrak q}))$ and $M_{\mathfrak q}/({x_1}M_{\mathfrak q})$ is annihilated by 931 ${x_1}$, we have ${x_1}\in\mathfrak P$. If $({x_1}\tilde E_{\mathfrak P})=\tilde E_{\mathfrak P}$, then as 932 ${x_1}$ lies in the maximal ideal of the local ring $(k[x,\xi]_{\mathfrak q})_{\mathfrak P}$, Nakayama's lemma \cite{nakayama}, (see also \cite[Tag~00DV]{Stacks}) 933 would force $\tilde E_{\mathfrak P}=0$, a contradiction. Hence 934 $\bigl(\tilde E/({x_1}\tilde E)\bigr)_{\mathfrak P}\ne0$, so $\tilde E/({x_1}\tilde E)\ne0$ and $\ell\left(\tilde E/({x_1}\tilde E)\right) > 0$. 935 936 Therefore $\ell(U_{\mathfrak q})=\ell(V_{\mathfrak q})+\ell\bigl(\tilde E/{x_1}\tilde E\bigr) 937 >\ell(V_{\mathfrak q})$. 938 \end{proof} 939 \AIadd{\formalref{\leandecl{Stafford38/Paper/Section4/StrictInequality.lean}{405}{Stafford38.Paper.Section4.lemma_4_10}}} 940 941 942 943 \begin{corollary} 944 \label{cor:Uzero} 945 Multiplication by $x_1$ on $\operatorname{gr}Q$ is surjective, i.e.\ 946 $\gr Q=(\gr Q)\,x_1$. 947 \end{corollary} 948 \begin{proof} 949 We write $M = \gr Q$ and $U = M/{x_1}M$. If $U\ne0$, it has a minimal support prime $\mathfrak q$. 950 Lemma~\ref{lem:tangential-finiteness} makes the relevant lengths finite, 951 and Lemmas~\ref{lem:page-inequality} and 952 \ref{lem:strict-support-inequality} contradict one another. 953 Thus $U = M/{x_1}M=0$ and therefore $\gr Q=(\gr Q)\,x_1$. 954 \end{proof} 955 \AIadd{\formalref{\leandecl{Stafford38/Paper/Section4/Vanishing.lean}{45}{Stafford38.Paper.Section4.corollary_4_11}}} 956 957 958 \section{Avoiding the hyperplane $\{x_1 = 0\}$}\label{sec:conormal-review} 959 This section uses mainly vocabulary and concepts from algebraic geometry. We will define the most important quantities used here (for a comprehensive overview of the concepts of algebraic geometry, we refer to textbooks such as \cite{hartshorne}), then state the two main results: Theorem~\ref{thm:asymptotic-conormal-formal} 960 shows that for a closed irreducible subvariety $Y\subseteq X$ not meeting the hyperplane 961 $\{x_1=0\}$, the covector $dx_1$ is an asymptotic conormal direction of $Y$. 962 Theorem~\ref{thm:coisotropic-exclusion} uses it to show that no nonempty closed, 963 fibre-conical, involutive subset of the cotangent space can lie over the complement 964 of the hyperplane $\{x_1=0\}$ and be contained in the zero set of a polynomial 965 $P(\xi)$ with $P(1,0,\dots,0)\neq0$. 966 We prove the first theorem in Subsection~\ref{sub:asym}, with the lemmas stated and 967 proved where they are used, and the second in Subsection~\ref{sub:excl}. 968 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% 969 \subsection{Preliminaries and definitions from algebraic geometry}\label{sub:def_geom} 970 Throughout this section $K$ denotes an algebraically closed field of characteristic 971 zero, and we use the notation of Definition~\ref{def:cotangent} with $k=K$. A prime 972 ideal that is maximal is called a \emph{closed point}. By Theorem~\ref{thm:nss}(a), 973 the map $a\mapsto\mathfrak m_a$ identifies $K^n$ with the closed points of 974 $\mathbb A^n_K$, and $K^n\times(K^n)^*$ with the closed points of 975 $T^*\mathbb A^n_K=\operatorname{Spec}K[x,\xi]$. We write 976 $X:=K^n$ and $T^*X:=K^n\times(K^n)^*$ for these sets of closed points, with 977 \emph{affine coordinates} $x=(x_1,\dots,x_n)$. The \emph{projection} is $\pi\colon T^*X\to X$, $(y,\xi)\mapsto y$, and for 978 $y\in X$ the \emph{fibre} over $y$ is $T_y^*X=\pi^{-1}(y)\cong(K^n)^*$. 979 980 The closed subsets of $X$ and of $T^*X$ are the zero sets $V_K(J)$ of ideals 981 $J\subseteq K[x]$, resp.\ $J\subseteq K[x,\xi]$ (Definition~\ref{def:spec}); by 982 Theorem~\ref{thm:nss}(a), $V_K(J)$ is the set of closed points of $V(J)$. Via the dual coordinates 983 $\xi_1,\dots,\xi_n$ we regard polynomials $P\in K[\xi_1,\dots,\xi_n]$ as functions on 984 $(K^n)^*$; in particular $P(1,0,\dots,0)=P(dx_1)$. 985 986 We write $H:=V_K(x_1)=\{(0,y_2,\dots,y_n)\}\subseteq X$ for the hyperplane on 987 which the $x_1$-coordinate vanishes; then $\pi^{-1}(H)=H\times (K^n)^* = \{(y,\xi)\in T^*X:y_1 = 0\}\subseteq T^*X$. 988 The covector $dx_1$, i.e.\ $\xi=(1,0,\dots,0)$, is conormal to $H$: it vanishes 989 on every direction tangent to $H$. 990 991 The projective space $\mathbb P^n$ consists of nonzero vectors 992 $(q_0,\dots,q_n)\in K^{n+1}$ taken up to a common nonzero scalar (\emph{homogeneous 993 coordinates}); we embed $X\hookrightarrow\mathbb P^n$ by $x\mapsto(1,x_1,\dots,x_n)$. A 994 point with $q_0\neq0$ has affine coordinates $x_i=q_i/q_0$, and the points with 995 $q_0=0$ form the hyperplane at infinity. For a vector $q\in K^{n+1}$ representing a point 996 of $\mathbb P^n$ we call its entry $i$ its $i$-coordinate, entry $0$ being the 997 homogenizing coordinate. The \emph{affine part} of a row vector 998 $\ell=(\ell_0,\dots,\ell_n)$ is $(\ell_1,\dots,\ell_n)$. 999 \begin{definition}[{\cite[VIII.1, V]{lang}}]\label{def:fieldextension} 1000 If $K\subseteq L$ are fields, we call $L$ an extension of $K$ and write $L/K$. For $z\in L$ and $B\subseteq L$ we say: 1001 \begin{enumerate} 1002 \item $z$ is \emph{algebraic over $K$} if there exists a nonzero polynomial $g$ 1003 in one variable over $K$ with $g(z)=0$ 1004 \item $z$ is \emph{transcendental over $K$} if no such polynomial exists, i.e.\ 1005 $g(z)=0\Rightarrow g=0$ 1006 \item The \emph{minimal polynomial} of an algebraic $z$ over $K$ is the polynomial of 1007 least degree over $K$ with $g(z)=0$ and coefficient 1 for the highest degree term. 1008 \item $L/K$ is \emph{algebraic} if every element of $L$ is algebraic over $K$ 1009 \item $L/K$ is \emph{finitely generated} if $L=K(z_1,\dots,z_m)$ for 1010 finitely many $z_i$, and \emph{separable} if every $z\in L$ that is algebraic over 1011 $K$ has a minimal polynomial without repeated roots - in characteristic zero every 1012 algebraic extension is separable. 1013 \item $B$ is \emph{algebraically independent} over $K$ if 1014 no nonzero polynomial over $K$ in finitely many elements of $B$ vanishes. 1015 \item A maximal algebraically independent subset is a \emph{transcendence basis} of 1016 $L/K$. If $B$ is a transcendence basis, then $L$ is algebraic over the subfield 1017 $K(B)$ generated by $K$ and $B$. 1018 \item Any two transcendence bases of $L/K$ have the 1019 same cardinality, called the \emph{transcendence degree} $\operatorname{trdeg}_KL$. 1020 \end{enumerate} 1021 The elements of $L$ that are algebraic over $K$ form a subfield, and every 1022 finite set generating $L$ over $K$ contains a transcendence basis. 1023 \end{definition} 1024 1025 \begin{definition}[{\cite[§25]{matsumura}}]\label{def:omega} 1026 Let $\kappa/K$ be a field extension. 1027 \begin{enumerate}[label=(\alph*)] 1028 \item A \emph{$K$-derivation} of $\kappa$ is a $K$-linear map 1029 $D\colon\kappa\to\kappa$ with $D(ab)=aD(b)+bD(a)$. 1030 \item Let $\mathcal G$ be the free $\kappa$-vector space with basis symbols 1031 $\langle a\rangle$, $a\in\kappa$. The space of \emph{differentials} 1032 $\Omega_{\kappa/K}$ is the quotient of $\mathcal G$ by the subspace spanned by 1033 \[ 1034 \langle a+b\rangle-\langle a\rangle-\langle b\rangle,\quad 1035 \langle ab\rangle-a\langle b\rangle-b\langle a\rangle,\quad 1036 \langle c\rangle\qquad(a,b\in\kappa,\ c\in K), 1037 \] 1038 and $da\in\Omega_{\kappa/K}$ denotes the class of $\langle a\rangle$. 1039 \item A $K$-derivation $D$ induces the $\kappa$-linear map 1040 $\Omega_{\kappa/K}\to\kappa$, $da\mapsto D(a)$, since the linear map 1041 $\langle a\rangle\mapsto D(a)$ on $\mathcal G$ kills the relations. Conversely, a 1042 $\kappa$-linear map $\Omega_{\kappa/K}\to\kappa$ gives the $K$-derivation 1043 $a\mapsto(\text{its value on }da)$. Hence a basis $d\beta_1,\dots,d\beta_d$ of 1044 $\Omega_{\kappa/K}$ has dual derivations $D_1,\dots,D_d$ with 1045 $D_k(\beta_l)=\delta_{kl}$. 1046 \end{enumerate} 1047 \end{definition} 1048 \begin{definition}[{\cite[§10-§11]{matsumura}}]\label{def:valuation} 1049 Let $F$ be a field. 1050 \begin{enumerate}[label=(\alph*)] 1051 \item A subring $W\subseteq F$ is a \emph{valuation ring} of $F$ if for 1052 every $a\in F^\times$ we have $a\in W$ or $a^{-1}\in W$. It is a local ring with 1053 maximal ideal $\mathfrak m_W=W\setminus W^\times$. 1054 \item A \emph{discrete valuation ring} (DVR) is a Noetherian valuation ring $V\ne F$. Then $\mathfrak m_V$ is generated by some $\varpi\in V$, and every $f\in F^\times$ is $f=\varpi^ku$ with $k\in\mathbb Z$ 1055 and $u\in V^\times$, and $\operatorname{ord}_V(f):=k$. 1056 \item For a field $\kappa = V/\mathfrak m_V$ we write $\kappa[[t]]$ for the ring of formal power series, and $\kappa((t))$ for its fraction field (formal Laurent series). 1057 \end{enumerate} 1058 \end{definition} 1059 \begin{definition}\label{def:conormal} 1060 We say 1061 \begin{enumerate}[label=(\alph*)] 1062 \item A closed set $Y$ is \emph{irreducible} if it is not a union of two proper closed subsets, equivalently if $I(Y)$ is prime \cite[Cor I.1.4]{hartshorne}. 1063 \item For irreducible closed $Y$, the \emph{smooth locus} $Y_{\mathrm{sm}}$ is the set of $y\in Y$ at which the Jacobian matrix of generators of $I(Y)$ has rank $n-\dim Y$, i.e.\ the points where $Y$ has a well-defined tangent space of the expected dimension $\dim Y$ (i.e. without kinks etc.). It is open and dense in $Y$. \cite[Ch. I.5]{hartshorne} 1064 \item For $y\in Y$ the \emph{conormal space} is 1065 \[N^*_yY=\operatorname{span}_K\{\,df_y : f\in I(Y)\,\}\subseteq(K^n)^*,\qquad 1066 df_y=\textstyle\sum_i\partial_if(y)\,dx_i,\] 1067 \item $N(Y)=\bigcup_{y\in Y_{\mathrm{sm}}}N^*_yY\subseteq (K^n)^*$ denotes the \emph{conormal 1068 cone}. 1069 \end{enumerate} 1070 \end{definition} 1071 1072 1073 \begin{definition}\label{def:dir} 1074 For $Y$ as above, the set of \emph{conormal directions} is 1075 \[ \operatorname{Dir}(Y)=\overline{\{[\xi]\in\mathbb P((K^n)^*): 0\neq\xi\in N(Y)\}} 1076 \subseteq\mathbb P((K^n)^*),\] 1077 the Zariski closure of the projectivization of $N(Y)$. 1078 \end{definition} 1079 1080 \begin{definition}\label{def:fibre-can} 1081 A closed subset $W\subseteq T^*X$ is \emph{fibre-conical} if it is stable under the 1082 fibrewise scaling $(y,\xi)\mapsto(y,t\xi)$ for all $t\in K^\times$.\\ 1083 We say it is \emph{involutive} if $I(W)$ is. 1084 \end{definition} 1085 1086 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% 1087 \begin{theorem}\label{thm:asymptotic-conormal-formal} 1088 Let $Y\subseteq X$ be nonempty, closed, and irreducible with $Y\cap H=\varnothing$, and let 1089 $N(Y)=\bigcup_{y\in Y_{\mathrm{sm}}}N^*_yY\subseteq (K^n)^*$ be the conormal cone over the smooth locus $Y_{sm}$. 1090 \begin{enumerate}[label=(\roman*)] 1091 \item Every polynomial $P\in K[\xi_1,\dots,\xi_n]$ vanishing on $N(Y)$ satisfies $P(1,0,\dots,0)=0$. 1092 \item $[dx_1]\in\operatorname{Dir}(Y)$. 1093 \end{enumerate} 1094 \end{theorem} 1095 1096 \AIadd{\formalref[{}]{% 1097 \leandecl{Stafford38/Paper/Section5/SmoothLocus.lean}{645}{Stafford38.Paper.Section5.theorem_5_7_i_printed}, 1098 \leandecl{Stafford38/Paper/Section5/SmoothLocus.lean}{660}{Stafford38.Paper.Section5.theorem_5_7_ii_printed}}} 1099 1100 \begin{theorem}\label{thm:coisotropic-exclusion} 1101 Let $K$ be an algebraically closed field of characteristic zero, and let 1102 $W\subseteq T^*X$ be a Zariski closed subset that is fibre-conical and 1103 involutive. Let $P\in K[\xi_1,\dots,\xi_n]$ be 1104 a polynomial with $P(1,0,\dots,0)\neq0$. If $P$ vanishes on $W$ and $W\cap\pi^{-1}(H)=\varnothing$, then $W$ must be empty. 1105 \end{theorem} 1106 1107 \AIadd{\formalref[{}]{% 1108 \leandecl{Stafford38/Paper/Section5/Exclusion.lean}{447}{Stafford38.Paper.Section5.theorem_5_8}}} 1109 1110 \subsection{Proof of Theorem \ref{thm:asymptotic-conormal-formal}}\label{sub:asym} 1111 \begin{proof} 1112 1113 If $Y$ is nonempty, closed and irreducible, then $\mathfrak p:=I(Y)$ is a prime 1114 ideal and $Y=V_K(\mathfrak p)$. We will write $R=K[x]/\mathfrak p$, $F=\operatorname{Frac}(R)$ for the field of fractions, and $\bar x_i$ for the image of $x_i$ in $R$. \\ 1115 1116 If $\bar x_1$ is algebraic over $K$, there exists a nonzero polynomial $g$ over $K$ 1117 in one variable with $g(\bar x_1)=0$ (Def. \ref{def:fieldextension}), therefore \ $g(x_1)\in\mathfrak p$. Since $K$ 1118 is algebraically closed, $g(x_1)=\lambda\prod_i(x_1-c_i)$ with 1119 $\lambda\in K^\times$, and as $\mathfrak p$ is prime, some factor $x_1-c$ lies 1120 in $\mathfrak p$. Thus $x_1=c$ on $Y=V_K(\mathfrak p)$, and since $Y\cap H = \varnothing$ and $Y\neq\varnothing$, we have $c\neq0$. 1121 In this case $f=x_1-c$ lies in $\mathfrak p=I(Y)$ and has the differential 1122 $df_y=dx_1$ at every point $y\in Y_{\mathrm{sm}}$, so $dx_1\in N_y^*Y$, and its coordinate vector in the dual basis $(1,0,\dots , 0) \in N(Y)$. Every polynomial vanishing on $N(Y)$ therefore vanishes at $(1,0,\dots , 0)$, and 1123 $[dx_1]\in\operatorname{Dir}(Y)$. So we may assume from now on that 1124 $\bar x_1$ is transcendental over $K$.\\ 1125 1126 The rest of the proof follows this rough logic. We construct a discrete valuation 1127 ring $V$ of $F$ at infinity in which $x_1$ vanishes, and the corresponding 1128 generic point $y\in\kappa((t))^n$ of $Y$. The affine cone $\mathcal C$ of $Y$ over the projective tangent plane at 1129 $y$ has a basis $M$ of power series in $t$ that remains a basis at $t=0$, and 1130 whose reduction $M(0)$ has vanishing $1$-coordinate. Hence the row vector 1131 $u=(0,1,0,\dots,0)$, which reads off the $1$-coordinate, satisfies 1132 $uM(0)=0$, and we lift it to row vectors $\ell(t)$ annihilating $\mathcal C$, 1133 with $\ell(0)=u$. By construction the affine part $\tilde\ell(t)$ consists of 1134 conormal covectors to $Y$ at $y$, and $\tilde\ell(0)=dx_1$. A polynomial 1135 vanishing on $N(Y)$ therefore vanishes at $dx_1$, and so $[dx_1]\in\operatorname{Dir}(Y)$. 1136 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% 1137 1138 We set $\bar x_0:=1$. 1139 1140 \begin{lemma}\label{lem:W} 1141 Let $\bar x_1$ be transcendental over $K$. There exist a valuation ring $W$ with 1142 $K\subseteq W\subset F$, $\bar x_1\in\mathfrak m_W$, and an 1143 index $j$ with $\bar x_i/\bar x_j\in W$ for all $0\le i\le n$. 1144 \end{lemma} 1145 \begin{proof} 1146 Since $\bar x_1$ is transcendental over $K$, $S=K[\bar x_1]$ is a polynomial 1147 ring, and $\mathfrak q=\bar x_1S$ is a prime ideal of $S$. The localization 1148 $S_{\mathfrak q}\subseteq F$ is a local ring with maximal ideal 1149 $\bar x_1S_{\mathfrak q}$. By \cite[Tag~00IA]{Stacks} there exists a valuation ring 1150 $W$ of $F$ dominating $S_{\mathfrak q}$, i.e.\ $S_{\mathfrak q}\subseteq W$ and 1151 $\mathfrak m_{S_{\mathfrak q}}=\mathfrak m_W\cap S_{\mathfrak q}$. Hence $K\subseteq W$ and 1152 $\bar x_1\in\mathfrak m_{S_{\mathfrak q}}\subseteq\mathfrak m_W$. For nonzero $a,b\in F$ either $a/b\in W$ or $b/a\in W$, 1153 so among the finitely many nonzero $\bar x_i$ we can choose $j$ with 1154 $\bar x_i/\bar x_j\in W$ for all $i$ (note $\bar x_0=1\ne0$). 1155 \end{proof} 1156 1157 Let $W$ be the valuation ring of Lemma \ref{lem:W}. We define the column vector $q$ in homogeneous coordinates by 1158 \[ q_i:=\bar x_i/\bar x_j\qquad(0\le i\le n),\] 1159 so that $q_i\in W$, $q_j=1$, $q_0=\bar x_j^{-1}\neq0$, $\bar x_i=q_i/q_0$ and 1160 $q_1=q_0\bar x_1$. 1161 1162 \begin{lemma}\label{lem:place} 1163 For $q$ defined as above, there exists a discrete valuation ring $V$ with $K\subseteq V\subset F$, 1164 $\operatorname{Frac}V=F$, such that, with $\kappa=V/\mathfrak m_V$ and $\operatorname{res}(q_i)$ the 1165 residue of $q_i$ in $\kappa$: 1166 \begin{enumerate}[label=(\alph*)] 1167 \item $q_i\in V$ for all $i$, $\bar x_1\in\mathfrak m_V$ and $R\not\subseteq V$; 1168 \item $\kappa$ is algebraic over $L=K(\operatorname{res}(q_0),\dots, \operatorname{res}(q_n))$; 1169 \item $\Omega_{V/K}$ is a finitely generated $V$-module. 1170 \end{enumerate} 1171 \end{lemma} 1172 \begin{proof} 1173 Let $C_0=K[q_0,\dots,q_n,\bar x_1]\subseteq W$; it is finitely generated with 1174 $\operatorname{Frac}C_0=F$ since on one hand $\bar x_i = q_i/q_0$ so every $\bar x_i$ is in $\operatorname{Frac}C_0$ and on the other hand, $F$ is generated by the $\bar x_i$. Let $A$ be its integral closure in $F$. Since $C_0$ is a domain finitely generated over a field, $A$ is a finite $C_0$-module \cite[Thm. I.3.9A]{hartshorne}, hence a finitely generated, integrally 1175 closed Noetherian domain with $\operatorname{Frac}A=F$ and $q_i\in A$. As $W$ is a valuation ring and thereby is 1176 integrally closed \cite[Prop 5.18]{atiyah}, $A\subseteq W$. $\bar x_1 \in \mathfrak m_W$, so $x_1$ is a nonzero nonunit of $W$ and therefore also of $A$. Since $\bar x_1$ is a nonzero nonunit of the Noetherian domain $A$, the ideal 1177 $\bar x_1A$ is proper, so some prime $\mathfrak p_D$ is minimal among the primes 1178 containing $\bar x_1$. By Krull's principal ideal theorem 1179 \cite[Corr. 11.17]{atiyah}, $\operatorname{ht}\mathfrak p_D\le1$. Because $A$ 1180 is a domain, the only prime of height $0$ is $(0)$, and $\bar x_1\neq0$ lies in 1181 $\mathfrak p_D$; hence $\operatorname{ht}\mathfrak p_D=1$. 1182 1183 Put $V:=A_{\mathfrak p_D}$. This is a Noetherian local domain. It is normal, 1184 since a localization of an integrally closed domain is integrally closed \cite[Prop 5.12]{atiyah}, and 1185 its dimension equals $\operatorname{ht}\mathfrak p_D=1$. A Noetherian local 1186 normal domain of dimension one is a discrete valuation ring 1187 \cite[Thm. 11.2]{matsumura}, so $V$ is a discrete valuation ring. 1188 1189 It remains to check the listed properties. 1190 \begin{itemize} 1191 \item $K\subseteq A\subseteq V$: we have $K\subseteq C_0\subseteq A$, and the 1192 domain $A$ embeds into its localization $V$. 1193 \item $\operatorname{Frac}V=F$: since $A\subseteq V\subseteq F$ and 1194 $\operatorname{Frac}A=F$. 1195 \item $q_i\in V$ for all $i$: since $q_i\in A\subseteq V$. 1196 \item $\bar x_1\in\mathfrak m_V$: since $\mathfrak m_V=\mathfrak p_DA_{\mathfrak p_D}$ 1197 and $\bar x_1\in\mathfrak p_D$. $V_K(p, x_1) = Y \cap H = \varnothing$, so $1 \in (p, x_1)$ by Theorem \ref{thm:nss}, i.e. $1=p+gx_1$ for some $p\in\mathfrak p$, $g\in K[x]$. Modulo 1198 $\mathfrak p$ this gives $1=\bar g\,\bar x_1$, so $\bar x_1\in R^\times$. 1199 If $R\subseteq V$, then $\bar x_1\in R^\times\subseteq V^\times$, contradicting 1200 $\bar x_1\in\mathfrak m_V$. 1201 \end{itemize} 1202 This proves (a). 1203 1204 (b) Since $V=A_{\mathfrak p_D}$ and $\mathfrak m_V=\mathfrak p_DV$, we have 1205 $\kappa=V/\mathfrak m_V=(A/\mathfrak p_D)_{(0)}=\operatorname{Frac}(A/\mathfrak p_D)$, 1206 and the residue map $\operatorname{res}:V\to\kappa$ restricts to the quotient map $A\to A/\mathfrak p_D$. As $A$ is integral over $C_0$ and $\bar x_1\in\mathfrak p_D$, the ring $A/\mathfrak p_D$ is integral over the image of 1207 $C_0$ in $A_{\mathfrak p_D}$, which is 1208 $K[\operatorname{res}(q_0),\dots,\operatorname{res}(q_n)]\subseteq L$. So every element of 1209 $A/\mathfrak p_D$, and hence of its fraction field $\kappa$, is algebraic over $L$. 1210 1211 (c) $A$ is a finitely generated $K$-algebra, so $\Omega_{A/K}$ is a finite 1212 $A$-module, and $\Omega_{V/K}=V\otimes_A\Omega_{A/K}$ since $V$ is a localization of $A$ \cite[Prop II.8.2A]{hartshorne}. 1213 \end{proof} 1214 1215 By Lemma~\ref{lem:place} there exists a discrete valuation ring $V\subset F$ with 1216 $q_i\in V$ for all $i$, $\bar x_1\in\mathfrak m_V$ and $R\not\subseteq V$. Let $\varpi$ be a 1217 generator of $\mathfrak m_V$, then every nonzero $f \in F$ can be written as $f = \varpi^k\cdot u$ with $k \in \mathbb{Z}$ and $u$ a unit of $V$ - we say $\operatorname{ord}_V(f) = k \in \mathbb Z$ is the valuation of $f$ in $V$. We denote the residual field with $\kappa=V/\mathfrak m_V$. 1218 1219 Let $\widehat V$ be the completion of $V$. Since $V\supseteq K$ 1220 and $K$ is of characteristic zero, there exists a coefficient field of $\widehat V$ 1221 containing $K$ and mapping isomorphically onto $\kappa$ \cite[Thm. 28.3]{matsumura}; 1222 choosing one gives us an isomorphism $\widehat V\cong\kappa[[t]]$ under which 1223 $\varpi$ corresponds to the formal variable $t$. Since the natural map $V\to\widehat V$ is injective ($\bigcap_k\varpi^kV=0$ \cite[Corr. 10.20]{atiyah}) and extends to an embedding $F\hookrightarrow\kappa((t))$, we can identify elements of $F$ with 1224 Laurent series. The residue $\operatorname{res}(f)$ of $f \in V$ is then just the constant term $f(0)$ of its series; in particular $\operatorname{res}(q_i) = q_i(0).$ 1225 1226 Let $a=\operatorname{ord}_V(q_0)$ and $b=\operatorname{ord}_V(q_1)$. $a,b\geq 0$ since $q_0,q_1$ are elements of $V$, so that in $\kappa[[t]]$ 1227 \[q_0(t)=t^au_0(t),\qquad q_1(t)=t^bu_1(t),\] 1228 with $u_0(0)\neq0\neq u_1(0)$. Since $\bar x_1\in\mathfrak m_V$, $b-a=\operatorname{ord}_V(q_1/q_0) = \operatorname{ord}_V(\bar x_1)>0$. Moreover 1229 $a>0$: if $a=0$, then $q_0$ is a unit of $V$, hence all $\bar x_i=q_i/q_0$ 1230 lie in $V$, so $R\subseteq V$ since also $K\subseteq V$, contradicting Lemma~\ref{lem:place}(a). It follows that $j \notin \{0,1\}$ since $q_j = 1$ is a unit while $q_0,q_1$ are not. 1231 1232 Dividing by $q_0(t)\neq 0$, we define $y=(q_1(t)/q_0(t),\dots,q_n(t)/q_0(t))\in\kappa((t))^n$ in affine coordinates. Its 1233 coordinates are the images of $\bar x_1,\dots,\bar x_n$ under $F\hookrightarrow 1234 \kappa((t))$. Since ring homomorphisms commute with polynomial evaluation and 1235 this map is injective, for every polynomial $f\in K[x_1,\dots,x_n]$ we have 1236 \begin{equation} 1237 f(y)=0\iff f(\bar x_1,\dots,\bar x_n)=0 \iff f\in\mathfrak p. 1238 \label{eq:generic} 1239 \end{equation} 1240 1241 Let $df_y=\nabla f(y)\in\kappa((t))^n$ for a polynomial $f$. 1242 We define $\Theta_y$ to be the tangent space of $Y$ at the generic point $y$, 1243 \[ \Theta_y=\bigcap_{f\in\mathfrak p}\ker df_y\subseteq\kappa((t))^n,\] 1244 the directions $v$ with $df_y\cdot v=0$ for all equations $f$ of $Y$. We now pass to homogeneous coordinates by defining a map 1245 \[ \rho\colon\kappa((t))^{n+1}\to\kappa((t))^n,\qquad 1246 \rho(v(t))_i=\frac{v_i(t)q_0(t)-q_i(t)v_0(t)}{q_0^2(t)}\quad(1\le i\le n),\] 1247 which sends a homogeneous vector $v(t)$ to the affine direction it induces at $q(t)$ 1248 (by the quotient rule); its kernel is the line $\kappa((t))q(t)$. We write 1249 $\mathcal C=\rho^{-1}(\Theta_y)$ for the affine cone over the projective tangent 1250 plane. Our goal is to build a basis of $\mathcal C$ whose entries are power 1251 series in $t$ and which also remains a basis at $t=0$. 1252 We do so by building independent vectors from the tangent planes of $q(t)$. 1253 Since $0<a<b$, in $\kappa$ we have $q_0(0)=q_1(0)=0$ and $q_j(0)=1$. Let 1254 $L=K(q_0(0),\dots,q_n(0))\subseteq\kappa$; by Lemma~\ref{lem:place}(b), 1255 $\kappa/L$ is algebraic. 1256 1257 \begin{lemma}\label{lem:kappa} 1258 There exist indices $i_1,\dots,i_{d'}\notin\{0,1,j\}$ such that 1259 $d(q_{i_1}(0)),\dots,d(q_{i_{d'}}(0))$ form a basis of $\Omega_{\kappa/K}$. In 1260 particular $d' = \dim_\kappa\Omega_{\kappa/K}<\infty$. 1261 \end{lemma} 1262 \begin{proof} 1263 1264 Since $q_0(0)=q_1(0)=0$ and $q_j(0)=1$, the field $L$ is generated over $K$ by 1265 the finitely many $q_i(0)$ with $i\notin\{0,1,j\}$, so some subset 1266 $B=\{q_{i_1}(0),\dots,q_{i_{d'}}(0)\}$ of them is a transcendence basis of $L/K$ 1267 \cite[Thm. VIII.1.1]{lang}. As $L/K(B)$ and $\kappa/L$ are algebraic 1268 (Lemma~\ref{lem:place}(b)), so is $\kappa/K(B)$; hence $B$ is also a 1269 transcendence basis of $\kappa/K$. Since $K$ has characteristic zero, a 1270 transcendence basis of $\kappa/K$ is a differential basis, i.e.\ its 1271 differentials form a $\kappa$-basis of $\Omega_{\kappa/K}$ 1272 \cite[Thm.~26.5]{matsumura}. 1273 \end{proof} 1274 1275 1276 We write for all $q_i(t) = \sum_rc_{ir}t^r$. The derivative with respect to $t$ is then $q'_i(t) = \sum_rr\,c_rt^{r-1}$. This is a $K$-derivation of $\kappa((t))$. By Lemma \ref{lem:kappa} there exist $i_1,\dots,i_{d'}\notin\{0,1,j\}$ such that $\{d(q_{i_1}(0)),\dots,d(q_{i_{d'}}(0))\}$ form a basis of $\Omega_{\kappa/K}$. Let $D_1,\dots,D_{d'}$ be the dual $K$-derivations of this basis (Definition~\ref{def:omega}(c)). 1277 We apply $D_k$ coefficientwise to $q_i(t)=\sum_rc_{ir}t^r$ and define $Z_{ik}(t):=\sum_rD_k(c_{ir})t^r$. Since $D_k(q_{i_l}(0))=\delta_{kl}$, the reduction $Z(0)$ is the identity on the rows $i_1,\dots,i_{d'}$, and since $q_j(t) = 1$ is constant, $Z_{jk}(t)=0$. Rows $0$ and $1$ of $Z(t)$ have order at least $a$ and $b$, since $D_k$ does not differentiate with respect to t. 1278 1279 Let $\tilde Z(t)$ be the ${d'}\times {d'}$ submatrix of $Z(t)$ consisting of the rows 1280 $i_1,\dots,i_{d'}$ and all columns. Its reduction $\tilde Z(0)$ is the identity 1281 matrix, so $\tilde Z(t)$ is invertible over $\kappa[[t]]$. Hence there exists a unique vector 1282 $\lambda(t)=(\lambda_1(t),\dots,\lambda_{d'}(t))^{\mathsf T}\in\kappa[[t]]^{d'}$ with 1283 $\tilde Z(t)\lambda=(q'_{i_1}(t),\dots,q'_{i_{d'}}(t))^{\mathsf T}$. We define $w(t)=q'(t)-Z(t)\lambda(t)$, so 1284 that $w_{i_l}(t)=0$ for $l=1,\dots,{d'}$. 1285 Its $0$th entry is $w_0(t)=q_0'(t)-Z_{0\bullet}(t)\lambda(t)$. Here 1286 $q_0'(t)=at^{a-1}u_0(t)+t^au_0(t)'$ has order exactly $a-1$ (since the coefficient 1287 $au_0(0)$ of $t^{a-1}$ is nonzero in characteristic zero), 1288 while $Z_{0\bullet}(t)\lambda(t)$ has order at least $a$, because every entry of 1289 row $0$ of $Z(t)$ has order at least $a$ (since $q_0(t)$ 1290 has order $a$). Hence $\operatorname{ord}w_0(t)=a-1$. Let $c$ be the minimal order 1291 among the entries of $w(t)$; then $c\le a-1$, and $\tau(t):=t^{-c}w(t)$ is a vector 1292 over $\kappa[[t]]$ with $\tau(0)\neq0$ and $\tau_{i_l}(t)=0$. In row $1$, both 1293 $q_1'(t)$ and $Z_{1\bullet}(t)\lambda(t)$ have order at least $b-1$, so $\operatorname{ord}w_1(t)\ge b-1$, so 1294 $\operatorname{ord}\tau_1(t)\ge b-1-c\ge b-a>0$. Finally $q_j'(t)=0$ and 1295 $Z_{j\bullet}(t)=0$ (as $q_j(t)=1$), so $\tau_j(t)=0$. \\ 1296 1297 1298 We now show that $q(t), Z_{\bullet1}(t),\dots, Z_{\bullet {d'}}(t), \tau(t)$ span all of $\mathcal C$: 1299 Let $M(t)=(\,q(t), Z_{\bullet1}(t),\dots, Z_{\bullet {d'}}(t),\tau(t)\,)$, an 1300 $(n+1)\times({d'}+2)$ matrix over $\kappa[[t]]$. Its reduction $M(0)$ has full 1301 column rank: in row $j$ only $q(0)$ is nonzero ($q_j(0)=1$); in rows $i_1,\dots,i_{d'}$ 1302 the columns $Z_{\bullet k}(0)$ form an identity block while $\tau(0)$ vanishes 1303 there; and $\tau(0)\neq0$, i.e. we have ${d'}+2$ independent reduced column vectors. This also translates to the column vectors of $M(t)$ themselves: Since some $({d'}+2)\times({d'}+2)$ minor of $M(0)$ is nonzero, the corresponding 1304 minor of $M(t)$ is a unit of $\kappa[[t]]$; hence the columns of $M(t)$ are also 1305 linearly independent over $\kappa((t))$, so they span a $({d'}+2)$-dimensional 1306 subspace there. 1307 Furthermore, every column of $M(0)$ has a vanishing entry for the $1$-coordinate: 1308 $q_1(0)=0$ (as $b>0$), the entries $Z_{1k}(t)$ have order at least $b>0$, hence $Z_{1k}(0) = 0$, and 1309 $\tau_1(0)=0$. Thus the span of $q(0), Z_{\bullet1}(0),\dots, Z_{\bullet {d'}}(0), \tau(0)$ is contained in $ \{v \in \kappa^{n+1} : v_1 = 0\}$ and also has dimension ${d'}+2$. 1310 1311 To show that the columns of $M(t)$ really span all of $\mathcal C$, we must still show that a) the dimensions match and b) the columns actually lie in $\mathcal C$. The following Lemma will help us with that. 1312 \begin{lemma}\label{lem:tangent} 1313 Let $m = \operatorname{trdeg_K}F$ and $\Theta_y$, $\rho$ and $\mathcal C$ be defined as above. 1314 \begin{enumerate}[label=(\alph*)] 1315 \item $\dim_{\kappa((t))}\Theta_y = m \le {d'}+1$; hence $\dim_{\kappa((t))}\mathcal C = m + 1 \le {d'}+2$. 1316 \item For every $K$-derivation $\delta$ of $\kappa((t))$ the column 1317 $\delta q(t)=(\delta q_i(t))_i$ lies in $\mathcal C$; moreover $q(t)\in\mathcal C$. 1318 \end{enumerate} 1319 \end{lemma} 1320 \begin{proof} 1321 (a) For a finitely generated field extension $F/K$ in characteristic $0$, $\dim_F\Omega_{F/K}=\operatorname{trdeg}_KF = m$ \cite[Thm.~II.8.6A]{hartshorne}. 1322 The second fundamental exact sequence \cite[Theorem 25.2]{matsumura} for $K[x]\to R=K[x]/\mathfrak p$ is 1323 \[\mathfrak p/\mathfrak p^2\to\Omega_{K[x]/K}\otimes_{K[x]}R\to\Omega_{R/K}\to0 ,\] 1324 where $\Omega_{K[x]/K}$ is free on $dx_1,\dots,dx_n$, so the middle term is $R^n$ 1325 and the first map sends $f$ to $df(\bar x)$. Hence 1326 $\Omega_{R/K}=R^n/\operatorname{span}_R\{df(\bar x):f\in\mathfrak p\}$. 1327 For a multiplicative set $S\subseteq R$, $\Omega_{S^{-1}R/K}=S^{-1}\Omega_{R/K}$ \cite[Prop II.8.2A]{hartshorne}. In particular, this means for $F=S^{-1}R$ with 1328 $S=R\smallsetminus\{0\}$ $\Omega_{F/K}=F\otimes_R\Omega_{R/K}$ and since localization is exact, applying $F\otimes_R-$ to the description of 1329 $\Omega_{R/K}$ above gives 1330 \[ \Omega_{F/K}=F^n/\operatorname{span}_F\{df(\bar x):f\in\mathfrak p\}.\] 1331 Analogously, applying this for $S = V\setminus \{0\}$ and $F=S^{-1}V$ gives $F\otimes_V\Omega_{V/K}=\Omega_{F/K}$. 1332 $\operatorname{span}_F\{df(\bar x):f\in\mathfrak p\}$ has therefore dimension $n-m$, 1333 i.e.\ the Jacobian $(\partial_if(\bar x))_{f\in\mathfrak p}$ has rank $n-m$ over $F$. 1334 Rank does not change under the field extension $F\subseteq\kappa((t))$, so 1335 $\dim\operatorname{span}_{\kappa((t))}\{df_y:f\in\mathfrak p\}=n-m$, and therefore 1336 $\dim_{\kappa((t))}\Theta_y=\dim_{\kappa((t))}\bigcap_f\ker df_y=n-(n-m)=m$. 1337 1338 The second fundamental exact sequence for $V\to\kappa$ is 1339 \[ \mathfrak m_V/\mathfrak m_V^2\to\kappa\otimes_V\Omega_{V/K}\to\Omega_{\kappa/K}\to0 .\] 1340 Since $\dim_\kappa\mathfrak m_V/\mathfrak m_V^2=1$ and 1341 $\dim_\kappa\Omega_{\kappa/K}=d'$, we get 1342 $\dim_\kappa(\kappa\otimes_V\Omega_{V/K})\le d'+1$. As $\Omega_{V/K}$ is a finitely 1343 generated $V$-module (Lemma~\ref{lem:place}(c)), Nakayama's lemma shows that it is 1344 generated by at most $d'+1$ elements. Since $F\otimes_V\Omega_{V/K}=\Omega_{F/K}$, 1345 we obtain $m\le d'+1$, hence $\dim\Theta_y\le d'+1$. Finally, $\rho$ maps 1346 $\mathcal C$ onto $\Theta_y$ with one-dimensional kernel, so 1347 $\dim_{\kappa((t))}\mathcal C=\dim_{\kappa((t))}\Theta_y+1\le d'+2$. 1348 1349 1350 (b) Let $f\in\mathfrak p$ and $f = \sum_\alpha c_\alpha y^\alpha$. By \eqref{eq:generic}, $f(y)=0$, and $\delta$ vanishes on 1351 $K$, so 1352 \[\delta\bigl(f(y)\bigr) 1353 =\sum_\alpha\Bigl(\underbrace{y^\alpha\,\delta(c_\alpha)}_{=\,0}+c_\alpha\,\delta(y^\alpha)\Bigr) 1354 =\sum_{i=1}^n\frac{\partial f}{\partial x_i}(y)\,\delta y_i 1355 =df_y\cdot\delta y,\] hence 1356 $\delta y\in\Theta_y$. Since $y_i=q_i/q_0$, the quotient rule gives 1357 $\rho(\delta q)_i=(\delta q_i\,q_0-q_i\,\delta q_0)/q_0^2=\delta(q_i/q_0)=\delta y_i$. 1358 Thus $\rho(\delta q)=\delta y\in\Theta_y$, i.e.\ $\delta q\in\mathcal C$. Moreover, $\rho(q)_i=(q_iq_0-q_iq_0)/q_0^2=0$ for all $i$, so $\rho(q)=0\in\Theta_y$ 1359 and therefore $q\in\mathcal C$. 1360 1361 \end{proof} 1362 1363 The column vectors $q(t)$, $Z_{\bullet k}(t)$ and $q'(t)$ are of the form $\delta q(t)$, where 1364 $\delta$ is a $K$-derivation of $\kappa((t))$ (the $D_k$ extended 1365 coefficientwise, and $d/dt$). By Lemma~\ref{lem:tangent}(b) they lie in $\mathcal C$. Since 1366 $\tau(t)=t^{-c}(q'(t)-Z(t)\lambda(t))$ and $t^{-c}\in\kappa((t))$, $\tau(t)$ is a 1367 $\kappa((t))$-linear combination, hence also lies in $\mathcal C$. By Lemma~\ref{lem:tangent}(a), $\dim \mathcal C\le {d'}+2$, so the ${d'}+2$ independent columns of $M(t)$ must span $\mathcal C$.\\ 1368 1369 \begin{lemma}\label{lem:ell} 1370 If a row vector $\ell\in\kappa((t))^{n+1}$ annihilates $\mathcal C$, then 1371 $(\ell_1,\dots,\ell_n)\in\operatorname{span}\{df_y:f\in\mathfrak p\}$. 1372 \end{lemma} 1373 \begin{proof} 1374 Let $\ell$ annihilate $\mathcal C$. Since $q\in\mathcal C$, $\ell(q)=0$, i.e.\ 1375 $\sum_{i\ge1}\ell_iq_i=-\ell_0q_0$. For $v\in\mathcal C$ we have $\ell(v)=0$, i.e.\ 1376 $\sum_{i\ge1}\ell_iv_i=-\ell_0v_0$, so 1377 \[ \sum_{i\ge1}\ell_i\rho(v)_i 1378 =\frac{q_0\sum_{i\ge1}\ell_iv_i-v_0\sum_{i\ge1}\ell_iq_i}{q_0^2} 1379 =\frac{-q_0\ell_0v_0+v_0\ell_0q_0}{q_0^2}=0 .\] 1380 As $\rho(\mathcal C)=\Theta_y$, the vector $(\ell_1,\dots,\ell_n)$ annihilates 1381 $\Theta_y=\bigcap_f\ker df_y$, hence lies in 1382 $\Theta_y^\perp=\operatorname{span}\{df_y:f\in\mathfrak p\}$. 1383 \end{proof} 1384 1385 Let $u=(0,1,0,\dots,0)$ be the row vector reading off the $1$-coordinate; 1386 since $\operatorname{im}M(0)\subseteq \{v \in \kappa^{n+1} : v_1 = 0\}$, we know that $uM(0)=0$. 1387 We can lift $u$ to a row vector $\ell(t)\in\kappa[[t]]^{n+1}$ with $\ell(t) M(t)=0$ and 1388 $\ell(0)=u$: Since $M(0)$ has full column rank ${d'}+2$, there is a set $\iota$ of 1389 ${d'}+2$ row indices for which the square submatrix $M_\iota(0)$ of $M(0)$ is 1390 invertible; then $M_\iota$ is invertible over $\kappa[[t]]$, since its determinant 1391 has a nonzero constant term. Let $\upsilon$ be the complementary set of row indices, 1392 and define $\ell(t)$ by 1393 \[\ell_\upsilon(t)=u_\upsilon,\qquad \ell_\iota(t)=-u_\upsilon M_\upsilon (t)M_\iota^{-1}(t),\] 1394 where $M_\upsilon(t)$ denotes the submatrix of $M(t)$ with rows in $\upsilon$. Then 1395 $\ell(t) M(t)=\ell_\iota(t)M_\iota(t)+\ell_\upsilon(t)M_\upsilon(t)=-u_\upsilon M_\upsilon(t)+u_\upsilon M_\upsilon(t)=0$. Moreover 1396 $uM(0)=0$ gives $u_\iota=-u_\upsilon M_\upsilon(0)M_\iota(0)^{-1}$, hence $\ell(0)=u$. 1397 1398 Since the columns of $M(t)$ span $\mathcal C$, $\ell(t)$ annihilates $\mathcal C$, so by 1399 Lemma~\ref{lem:ell} its affine part 1400 $\tilde\ell(t)=(\ell_1(t),\dots,\ell_n(t))$ lies in 1401 $\operatorname{span}\{df_y:f\in\mathfrak p\}$, and since $\ell(0)=u$, 1402 $\tilde\ell(0)=(1,0,\dots,0)$. 1403 1404 Finally, let $P\in K[\xi_1,\dots,\xi_n]$ vanish on $N(Y)$. We will show 1405 $P(\tilde\ell(0))=0$: Since $\tilde\ell(t)\in\operatorname{span}_{\kappa((t))}\{dg_y:g\in 1406 \mathfrak p\}$, we can write 1407 \[\tilde\ell(t)=\sum_{k=1}^s\mu_k(t)\nabla g_k(y),\qquad g_k\in\mathfrak p,\ 1408 \mu_k(t)\in\kappa((t)),\] 1409 With some $s \in \mathbb{N}$. Here $\nabla g(x)=(\partial g/\partial x_1(x),\dots, 1410 \partial g/\partial x_n(x))$ denotes the coordinate vector of the covector 1411 $dg_x$. We furthermore define 1412 \[\Phi(x,\mu)=P\Bigl(\sum_{k=1}^s\mu_k\nabla g_k(x)\Bigr) 1413 =\sum_\alpha\phi_\alpha(x)\mu^\alpha\in K[x_1,\dots,x_n,\mu_1,\dots,\mu_s],\] 1414 with $\phi_\alpha\in K[x_1,\dots,x_n]$ and $\mu^\alpha = \mu_1^{\alpha_1}\dots\mu_s^{\alpha_s}$ for a multi-index $\alpha \in \mathbb{N}^s$. 1415 For a smooth point $y'\in Y_{\mathrm{sm}}$ and 1416 $\mu\in K^s$ we have $\sum_k\mu_kdg_k(y')\in N^*_{y'}Y\subseteq N(Y)$, so 1417 $\Phi(y',\mu)=0$. As $K$ is infinite, a polynomial in $\mu$ vanishing on all of 1418 $K^s$ is zero, so $\phi_\alpha(y')=0$ for all $\alpha$. Since $Y_{\mathrm{sm}}$ 1419 is dense in $Y$, every $\phi_\alpha$ vanishes on all of $Y$, i.e.\ 1420 $\phi_\alpha\in\mathfrak p=I(Y)$. By \eqref{eq:generic} (applied to the generic 1421 point $y\in\kappa((t))^n$), we therefore have $\phi_\alpha(y)=0$ for all 1422 $\alpha$. Hence $\Phi(y,\mu(t))=\sum_{\alpha}\phi_\alpha(y)\mu^\alpha(t)=\sum_\alpha 0\cdot \mu^\alpha(t) = 0$ for all $\mu(t)\in\kappa((t))^s$, and in 1423 particular 1424 \[ P(\tilde\ell(t))=\Phi\bigl(y,(\mu_1(t),\dots,\mu_s(t))\bigr)=0.\] 1425 Now $\tilde\ell(t)\in\kappa[[t]]^n$ and $P$ has coefficients in $K$, so 1426 $P(\tilde\ell(t))\in\kappa[[t]]$, and its constant term is $P(\tilde\ell(0))$. Since $P(\tilde\ell(t))=0$, we get 1427 $P(\tilde\ell(0))=P(1,0,\dots,0)=0$, which is statement~(i). 1428 1429 By statement~(i), every polynomial vanishing on $N(Y)$ vanishes at $(1,0,\dots,0)$. We 1430 use this to show $[dx_1]\in\operatorname{Dir}(Y)$. Recall that $\operatorname{Dir}(Y)$ is the Zariski closure of the set of classes $[\xi]$ with $0\ne\xi\in N(Y)$, and that a point of $\mathbb P((K^n)^*)$ lies in the closure of a set $S$ if and only if every homogeneous polynomial vanishing on $S$ also vanishes at that point. So let $G$ be a homogeneous polynomial vanishing on all classes $[\xi]$ with $0\ne\xi\in N(Y)$. Since also $G(t\xi)=t^{\deg G}G(\xi)$ by homogeneity, $G$ vanishes on every nonzero $\xi\in N(Y)$ (and at $0$ if $\deg G>0$), hence on $N(Y)$; if $\deg G=0$, $G$ is a constant that vanishes somewhere (such classes exist: if $N(Y)\subseteq\{0\}$, the first coordinate function vanishes on $N(Y)$ but not at $(1,0,\dots,0)$, contrary to~(i)), so $G=0$. By~(i), $G(1,0,\dots,0)=0$, 1431 i.e.\ $G$ vanishes at $[dx_1]=[1:0:\dots:0]$. As $G$ was arbitrary, 1432 $[dx_1]\in\operatorname{Dir}(Y)$. 1433 \end{proof} 1434 1435 \AIadd{\formalref{\leandecl{Stafford38/Paper/Section5/Proof.lean}{66}{Stafford38.Paper.Section5.theorem_5_7_i_paper}; 1436 \leandecl{Stafford38/Paper/Section5.lean}{51}{Stafford38.Paper.Section5.theorem_5_7_i}; 1437 \leandecl{Stafford38/Paper/Section5/Place.lean}{51}{Stafford38.Paper.Section5.lemma_5_9}; 1438 \leandecl{Stafford38/Paper/Section5/Place.lean}{155}{Stafford38.Paper.Section5.lemma_5_10}; 1439 \leandecl{Stafford38/Paper/Section5/ResidueBasis.lean}{33}{Stafford38.Paper.Section5.lemma_5_11}; 1440 \leandecl{Stafford38/Paper/Section5/KaehlerRankBound.lean}{139}{Stafford38.Paper.Section5.lemma_5_12_a_bound}; 1441 \leandecl{Stafford38/Paper/Section5/TangentCone.lean}{100}{Stafford38.Paper.Section5.lemma_5_12_a_cone}; 1442 \leandecl{Stafford38/Paper/Section5/TangentCone.lean}{142}{Stafford38.Paper.Section5.lemma_5_12_b}; 1443 \leandecl{Stafford38/Paper/Section5/TangentCone.lean}{171}{Stafford38.Paper.Section5.lemma_5_13}}} 1444 1445 1446 \subsection{Proof of Theorem \ref{thm:coisotropic-exclusion}}\label{sub:excl} 1447 \begin{proof} 1448 Suppose we have a \textit{nonempty} closed subset $W\subseteq T^*X$ that is fibre-conical and involutive. Suppose further, that $W\cap\pi^{-1}(H)=\varnothing$. 1449 1450 We will give a proof by contradiction, with this rough logic: Suppose $W\neq\varnothing$ and let $Y = \pi(W)$. Since $W$ is by assumption fibre-conical, we start by showing that $Y\times \{0\} = \pi(W)\times \{0\} \subseteq W$ and $Y \cap H = \varnothing$. Using involutivity, we then show that we can translate along $W$ in the direction of the fibres which means that all conormal covectors of every irreducible component $Y_0$ of $Y$ lie in $W$. This means $P$ vanishes on all of $W$, hence on $N(Y_0)$. This gives us the setup of Theorem \ref{thm:asymptotic-conormal-formal}, which implies $P(1,0,\dots,0) = 0,$ a contradiction to our assumption. \\ 1451 1452 1453 We define the (radical) vanishing ideal $J=I(W)\subseteq K[x,\xi]$ consisting of all polynomials that vanish at every point $(y,\xi)\in W$, and $B=J\cap K[x]$, consisting of the functions $f(x)$ with $f(y)=0$ for every $(y,\xi) \in W$. \\ 1454 Because $W$ is fibre-conical, the $\xi$-homogeneous components of any 1455 polynomial in $J$ again lie in $J$: if $g=\sum_dg_d\in J$ and 1456 $(y,\xi)\in W$, then $\sum_da^dg_d(y,\xi)=g(y,a\xi)=0$ for all $a\in K^\times$, 1457 and a polynomial in $a$ with infinitely many roots is zero, so every 1458 $g_d(y,\xi)=0$ and therefore $g_d \in J$. 1459 Let $g \in J$ be written in its $\xi$-homogeneous components form. In particular, $g_0 \in J$. Since $g_0$ has $\xi$-degree $0$, it is a polynomial in $x$ alone, so $g_0 \in B$. Let $Y = V_K(B)$. For $y \in Y$ we therefore have $g(y, 0) = g_0(y) = 0$, so every $g \in J$ vanishes at $(y,0)$, i.e. $(y,0)\in V_K(J)$. $V_K(J) = V_K(I(W)) = W$, since $W$ is closed \cite[Prop. I.1.2]{hartshorne}, hence $(y, 0) \in W$, or $Y \times \{0\} \subseteq W$. 1460 Furthermore, $Y$ is the projection of $W$ and is therefore per definition of $Y$ closed: $\pi(W)\subseteq Y$ since for ($y',\xi')\in W$ and $f \in B\subseteq J$, we have $f(y') = f(y',\xi') = 0$ so $y' \in V_K(B) = Y$, and $Y\subseteq \pi(W)$, since if $y \in Y$, then also $(y,0)\in W$, so $y = \pi(y,0) \in \pi(W)$.\\ 1461 Lastly, assume there exists a $y \in Y\cap H$, then $(y,0)\in W$ and $\pi(y,0) = y \in H$, which would put $(y,0)\in W\cap \pi^{-1}(H)$, which is wrong by our assumption. Therefore, $Y \cap H = \varnothing$.\\ 1462 1463 For the next argument, we also refer to \cite[Ch. E.2]{HTT}. Let $f\in B$. Then $f\circ\pi\in J$. 1464 The Hamiltonian derivation of $f$ applied to $g \in J$ is 1465 \[ H_f g=\sum_i\frac{\partial f}{\partial x_i} 1466 \frac{\partial g}{\partial\xi_i} = \{f,g\}\] 1467 since the second term of the Poisson bracket vanishes because $f$ does not depend on $\xi$. Since $f, g \in J$ and $\{J, J\} \subseteq J$ by the involutivity of $W$, it follows that $H_f g \in J$ and by induction also $H^m_fg \in J$. 1468 For a fixed $y \in Y$ and $g \in J$, we define the polynomial $\varphi(t)=g(y,t\,df_y)\in K[t]$. Its derivative is $\varphi'(t) = \sum_i\partial_{x_i}f(y)\frac{\partial g}{\partial \xi_i}{(y, tdf_y)} = H_fg(y, tdf_y)$, so by iteration, $\varphi^{(m)}(t) = H^m_fg(y, tdf_y)$ for $m\geq 0.$ 1469 Evaluating at $t = 0$, $\varphi^{(m)}(0) = H_f^mg(y,0).$ Since we have established that $(y,0)\in W$ and $H^m_fg \in J$, and all polynomials in $J$ vanish on $W$, we have $\varphi^{(m)}(0) = 0$ for all $m\geq 0$ -- so $\varphi(t)$ is a polynomial in $t$ all of whose derivatives vanish at $t = 0$. If we write $\varphi(t) = \sum_m c_mt^m$, then $\varphi^{(m)}(0) = m!c_m$ and since $m!\neq 0$ because $K$ has characteristic zero, $c_m = 0$ for all $m$, and therefore, $\varphi(t) = 0$ everywhere. Remembering the definition of $\phi = g(y,t\,df_y) = 0$, so $(y,t\,df_y) \in V_K(J) = W$. 1470 More generally, since any linear combination of elements in $B$ is again in $B$, we can apply this result to $f = \sum_a c_af_a \in B$ and evaluating at $t = 1$, it follows that 1471 \[(y, df_y) = (y, d(\sum_ac_af_a)_y) = (y, \sum_a c_ad(f_a)_y) \in W,\] 1472 so the whole linear span $\{y\}\times \operatorname{span}\{df_y:f \in B\}$ is a subset of $W$.\\ 1473 1474 1475 Since $V_K(B)=Y=\pi(W)\neq\varnothing$, $B \neq K[x]$, so $B$ must have at least one minimal prime. The ideal $B$ is radical (by construction since it is a vanishing ideal) in the Noetherian ring $K[x]$, so $B$ has only finitely many minimal primes $\mathfrak q_1,\dots,\mathfrak q_r$ and $B=\bigcap_i\mathfrak q_i$ \cite[Cor. I.1.4 and Prop. I.1.5]{hartshorne}. We choose $\mathfrak q=\mathfrak q_1$ and set $Y_0=V_K(\mathfrak q)$. Since $\mathfrak q$ is prime, $1\notin\mathfrak q$, so $Y_0\neq\varnothing$ by Theorem~\ref{thm:nss}; moreover $Y_0$ is closed and irreducible since $\mathfrak q$ is prime \cite[Cor.~I.1.4]{hartshorne}, and since primeness implies $\mathfrak q$ is radical, $\mathfrak q = I(Y_0)$ by Theorem \ref{thm:nss}. Additionally, $Y_0\subseteq Y$, hence 1476 $Y_0\cap H=\varnothing$, as we have already established $Y\cap H=\varnothing$. 1477 1478 Furthermore, if $r = 1$, we choose $s = 1$, then $sh \in \mathfrak q = B$, and if $r\geq 2$, choose for every $i\geq 2$ elements $s_i \in \mathfrak q_i \setminus \mathfrak q$. Then $s = s_2s_3\dots s_r$ does not lie in $\mathfrak q$ since no factor lies in $\mathfrak q,$ but does lie in every other $\mathfrak q_i$ since they are ideals and one of the factors lies in them. On the other hand, if $h \in \mathfrak q$, then $sh \in \mathfrak q$ since $h\in \mathfrak q$, and $sh \in \mathfrak q_i$ since $s \in \mathfrak q_i$ for all $i\geq 2$. Therefore, $sh \in \bigcap_i \mathfrak q_i = B,$ i.e. $s\mathfrak q \subseteq B.$ We now fix such a $h \in \mathfrak q.$ For a $y \in Y_0\cap\{s\neq0\}$, $(y, dh_y) \in W:$ 1479 Since $sh \in B,$ and $d(sh)_y = s(y)dh_y+h(y)ds_y = s(y)dh_y$ since $h(y) = 0$, so $(y, s(y)dh_y) \in W$ and since $s(y)\neq 0,$ also $(y, dh_y)\in W$. 1480 For $g\in J$, the polynomial $G(x)=g(x,dh_x)$ therefore vanishes on $Y_0\cap\{s\neq0\}$, and $sG$ vanishes on all of $Y_0$, so $sG \in \mathfrak q$ and therefore $G \in \mathfrak q$, which means that $G$ vanishes on all of $Y_0$. Since this is true for every $g \in J$, we have $g(y, dh_y) = G(y) = 0$ for all $g\in J$, so $(y,dh_y) \in V_K(J)= W.$ As before, we can argue that since $\mathfrak q$ is an ideal, it is closed under $K$-linear combinations, so for $h_1,\dots ,h_k \in \mathfrak q$ and $c_a \in K$, also $(y, \sum_a c_a (dh_a)_y) \in W$, or put differently, $\{y\} \times \operatorname{span}\{dh_y: h \in \mathfrak q\} \subseteq W$ for all $y \in Y_0$. At a smooth point $y$ of $Y_0$, by Definition \ref{def:conormal}, the conormal space is 1481 \[N_y^*Y_0 = \operatorname{span}\{df_y : f \in \mathfrak q\}\] 1482 and we have already established that $\mathfrak q = I(Y_0)$. We therefore have that $\{y\} \times N_y^*Y_0 \subseteq W$ for every smooth $y \in Y_0$, so $P$, which only depends on $\xi$, vanishes on $N(Y_0)$. Because we have established, that $Y_0$ is closed, irreducible, and $Y_0 \cap H = \varnothing $, $Y_0$ therefore fulfills all requirements of Theorem \ref{thm:asymptotic-conormal-formal}, so $P(1,0,\dots,0) = 0,$ contradicting our initial assumption. Therefore, $W$ must be empty. 1483 \end{proof} 1484 1485 1486 \AIadd{\formalref{% 1487 \leandecl{Stafford38/Paper/Section5/Exclusion.lean}{447}{Stafford38.Paper.Section5.theorem_5_8}; 1488 \leandecl{Stafford38/Paper/Section5/SmoothLocus.lean}{645}{Stafford38.Paper.Section5.theorem_5_7_i_printed}}} 1489 1490 1491 \section{Proof of Theorem~\ref{thm:main}}\label{sec:proof} 1492 1493 \AIadd{\formalref{\leandecl{Stafford38/Paper/Section3/OrderSymbol.lean}{290}{Stafford38.Paper.Section3.proposition_3_6}}} 1494 \begin{proposition}\label{prop:char-avoids} 1495 $\Char(Q)\cap V(x_1)=\varnothing,$ 1496 where $V(x_1)\subseteq \Spec k[x, \xi]$ is the set of primes containing $x_1$. 1497 \end{proposition} 1498 \begin{proof} 1499 Write $M=\gr Q$. Since $M$ is finite over $k[x,\xi]$, 1500 $\Supp_{k[x,\xi]}(M/x_1M)=\Supp_{k[x,\xi]}(M)\cap V(x_1)$ \cite[Tag~00L3]{Stacks}. By Corollary~\ref{cor:Uzero}, 1501 $M/x_1M=0$, so its support is empty, hence $\Supp_{k[x,\xi]}(M)\cap V(x_1)=\varnothing$. 1502 Since $M$ is finitely generated, $\Char(Q)=V(\Ann_{k[x,\xi]}M)=\Supp_{k[x,\xi]}(M)$ 1503 \cite[Tag~00L2]{Stacks}. Therefore $\Char(Q)\cap V(x_1)=\varnothing$. 1504 \end{proof} 1505 1506 \AIadd{\formalref[{}]{% 1507 \leandecl{Stafford38/Paper/Section6/Avoidance.lean}{71}{Stafford38.Paper.Section6.proposition_6_1}}} 1508 1509 1510 1511 \begin{lemma}\label{lem:base-change} 1512 Let $K/k$ be a field extension and put 1513 $\mathcal I_K=\tilde dA_n(K)+x_1^N\tilde dA_n(K)$, $Q_K=Q\otimes_kK=A_n(K)/\mathcal I_K$, with the 1514 order filtration $Q_{K,i}=q\,F_iA_n(K)$. Then 1515 \[ 1516 \gr Q_K\cong\gr Q\otimes_kK,\qquad 1517 \Ann_{K[x,\xi]}\gr Q_K=\bigl(\Ann_{k[x,\xi]}\gr Q\bigr)\,K[x,\xi], 1518 \] 1519 and $Q=0\iff Q_K=0$. In particular, $\Char(Q_K)$ is the preimage of $\Char(Q)$ 1520 under $\Spec K[x,\xi]\to\Spec k[x,\xi]$. 1521 \end{lemma} 1522 \begin{proof} 1523 By the PBW basis \cite[Prop 1.2.1]{coutinho}, $F_iA_n(K)=F_iA_n(k)\otimes_kK$ and $\mathcal I_K=\mathcal I\otimes_kK$, so 1524 $Q_{K,i}=Q_i\otimes_kK$. As $K$ is flat over $k$, 1525 $Q_{K,i}/Q_{K,i-1}=(Q_i/Q_{i-1})\otimes_kK$, and summing over $i$ gives 1526 $\gr Q_K=\gr Q\otimes_kK$. The module $\gr Q$ is finitely generated (cyclic, 1527 generated by the class of $q$) over the Noetherian ring $k[x,\xi]$, and 1528 $k[x,\xi]\to K[x,\xi]$ is flat, so the annihilator of the finite module 1529 commutes with this base change \cite[Tag~07T8]{Stacks}. Finally, $Q\otimes_kK=0$ 1530 implies $Q=0$, because $K$ is a nonzero free $k$-module. 1531 \end{proof} 1532 1533 Let $d\in A_n(k)$ be nonzero of Bernstein degree $N$ and let 1534 $\tilde d=c^{-1}\Phi_g(d)$ be its monic normal form. If $n=0$, then $A_0(k)=k$ and every nonzero $d$ is 1535 a unit anyways, so let $n\ge1$. As in Section \ref{sec:euler} we write 1536 \[ \mathcal I=\tilde dA_n(k)+x_1^N\tilde dA_n(k),\qquad Q=A_n(k)/\mathcal I .\] 1537 It suffices to show $Q=0$: then $1\in \mathcal I$, i.e.\ there are 1538 $R,S\in A_n(k)$ with $1=\tilde dR+x_1^N\tilde dS$. 1539 1540 Let $K=\bar k$, and let 1541 \[W = V_K\bigl(\operatorname{Ann}_{K[x,\xi]}\operatorname{gr}Q_K\bigr)\subseteq K^n\times (K^n)^*,\] 1542 be the set of closed points of $\operatorname{Char}(Q_K)$ (Theorem~\ref{thm:nss}), 1543 with vanishing ideal $I(W)=\sqrt{\operatorname{Ann}\operatorname{gr}Q_K}$. 1544 We check 1545 the hypotheses of Theorem~\ref{thm:coisotropic-exclusion} for $W$ and the 1546 polynomial $P=\sigma^{\mathrm{\ord}}_N(\tilde d)$ of Section~\ref{sec:monic}: 1547 1548 \begin{itemize} 1549 \item $W$ is Zariski closed, since it is the zero set of an ideal of polynomials. 1550 \item The module $\gr Q_K$ is graded by the order filtration, which assigns degree 1551 $0$ to the $x_i$ and degree $1$ to the $\xi_i$; so $\Ann\gr Q_K$ is homogeneous in $\xi$. Hence if $(y,\xi)\in W$ 1552 and $g\in\Ann\gr Q_K$, then $g(y,t\xi)=\sum_dt^dg_d(y,\xi)=0$ for every 1553 $t\in K^\times$, where $g_d$ denotes the $\xi$-homogeneous component of degree $d$ of $g$. So 1554 $(y,t\xi)\in W$, i.e.\ $W$ is fibre-conical. 1555 \item By Gabber's theorem (Theorem~\ref{ithm:gabber}) $I(W)$ is involutive. 1556 \item By eq.~\eqref{eq:P-annihilates}, $P\in 1557 \Ann_{k[x,\xi]}\gr Q$, hence $P\in\Ann_{K[x,\xi]}\gr Q_K$ by 1558 Lemma~\ref{lem:base-change}, so $P$ vanishes on $W$. 1559 \item By eq.~\eqref{eq:P-e1}, $P(1,0,\dots,0)=1\neq 0$. 1560 \item By Lemma~\ref{lem:base-change}, $\operatorname{Char}(Q_K)$ is the preimage of 1561 $\operatorname{Char}(Q)$ under $\operatorname{Spec}K[x,\xi]\to\operatorname{Spec}k[x,\xi]$, 1562 and $V(x_1)\subseteq\operatorname{Spec}K[x,\xi]$ is the preimage of 1563 $V(x_1)\subseteq\operatorname{Spec}k[x,\xi]$. Since preimages commute with 1564 intersections, Proposition~\ref{prop:char-avoids} gives 1565 $\operatorname{Char}(Q_K)\cap V(x_1)=\emptyset$. By Theorem~\ref{thm:nss}(a) and Definition~\ref{def:spec}, a closed point $\mathfrak m_{(y,\xi)}$ lies in $V(x_1)$ if and only if $y_1=0$, i.e.\ the closed points of $V(x_1)$ form $\pi^{-1}(H)=H\times(K^n)^*$. Hence the closed points of $\operatorname{Char}(Q_K)\cap V(x_1)$ are $W\cap\pi^{-1}(H)$, and so $W\cap\pi^{-1}(H)=\emptyset$. 1566 \end{itemize} 1567 By Theorem~\ref{thm:coisotropic-exclusion}, $W=\varnothing$. As $K$ is 1568 algebraically closed, by Theorem \ref{thm:nss} $1 \in \Ann_{K[x,\xi]}\gr Q_K$, 1569 so $\gr Q_K=0$. Since the filtration of $Q_K$ is exhaustive with $Q_{K,-1}=0$, 1570 induction on $m$ implies $Q_{K,m}=0$ for all $m$, so $Q_K=0$, and 1571 $Q=0$ by Lemma~\ref{lem:base-change}. Therefore, $A_n(k) = \tilde dA_n(k) + x_1^N\tilde dA_n(k)$, or equivalently $1=\tilde dR+x_1^N\tilde dS$ 1572 with $R,S\in A_n(k)$. 1573 1574 We apply the inverse of the automorphism $\Phi_g$ of 1575 Remark~\ref{rem:transport}: 1576 \[ 1577 1=d\bigl(c^{-1}\Phi_g^{-1}(R)\bigr)+F\,d\bigl(c^{-1}\Phi_g^{-1}(S)\bigr), 1578 \qquad F=\Phi_g^{-1}(x_1)^N, 1579 \] 1580 which proves Theorem~\ref{thm:main}. \qed 1581 1582 1583 \AIadd{\formalref{\leandecl{Stafford38/Paper/Main.lean}{24}{Stafford38.Paper.theorem_1_1}, 1584 \leandecl{Stafford38/Paper/Main.lean}{29}{Stafford38.Paper.theorem_1_1_fixedSource}, 1585 \leandecl{Stafford38/Paper/Section6/Proof.lean}{357}{Stafford38.Paper.Section6.theorem_1_1_paper}, 1586 \leandecl{Stafford38/Paper/Section6/Proof.lean}{318}{Stafford38.Paper.Section6.theorem_6_Q_eq_zero}}} 1587 1588 \AIadd{\formalref[{}]{% 1589 \leandecl{Stafford38/Paper/Section6/BaseChange.lean}{460}{Stafford38.Paper.Section6.lemma_6_2}}} 1590 1591 1592 Lastly, as Stafford already showed, from $1 = dR+FdS$ follows cyclicity of the torsion right module: 1593 \begin{corollary}[\cite{Stafford1978}]\label{cor:cyclic} 1594 Every finitely generated torsion right $A_n(k)$-module $T$ is cyclic, i.e. can be generated by one element. Moreover, it is stably cyclic, meaning that given any $l\geq 1$ and $a_1,...a_{l+1} \in T$ that generate $T$, there exist $f_i \in A_n$ such that $(a_i + a_{l+1}f_i)$ $i = 1,...l$ also generate $T$ \cite[p. 435]{Stafford1978}. 1595 \end{corollary} 1596 1597 \begin{proof} 1598 Suppose for every $d\neq 0 \in A_n$, there exists $F, R, S \in A_n$, such that $1 = dR + FdS$. Let us denote a torsion module by $T$. Stafford's theorem (the right module analogue to Theorem 3.7 \mbox{\cite[p.~438]{Stafford1978}}) established, that every finitely generated (right) torsion module can be generated by 2 generators, so let us denote the generators of $T$ with $t_1$ and $t_2$. We can define a homomorphism 1599 \[\phi:A_n \oplus A_n\rightarrow T \quad \phi(r,s)\mapsto t_1r+t_2s.\] 1600 This map is surjective, since $t_1, t_2$ generate all of $T$. By the first isomorphism theorem (for $\varphi:M\rightarrow N$, $im(\varphi) \cong M/ker(\varphi)$), it follows 1601 \[T \cong (A_n\oplus A_n)/M\] 1602 with $M = ker(\phi)$. \\ 1603 $T$ is torsion and therefore $\forall t\in T: \exists m \in A_n\backslash\{0\}: tm = 0.$ Specifically, let us denote with $d_1$ and $d_2$ the annihilators of $t_1$ and $t_2$, i.e. 1604 \[t_1d_1 = t_2d_2 = 0.\] 1605 $A_n$ is a right Ore domain \cite[Prop. 1.8.4]{Bjork}, which means it has the property that for any nonzero $d_1, d_2$ there exists elements $r_1, r_2 \in A_n$ such that $d_1r_1 = d_2r_2 \neq 0.$ 1606 Defining therefore $d := d_1r_1 = d_2r_2$ yields us a common annihilator of $t_1$ and $t_2$.\\ 1607 $(d, 0)$ and $(0,d)$ are elements in $M = ker(\phi)$, since $\phi(d, 0) = t_1d = 0$ and $\phi(0, d) = t_2d = 0.$\\ 1608 Because $(0,d), (d, 0) \in M$, also 1609 $(d, 0)S + (0, d)R = (dS, dR) = (dS, 1-FdS) =: (a, 1-Fa) \in M$, i.e. $(a, 1-Fa)$ is an element of $M$. \\ 1610 On the other hand, $A_n\oplus A_n$ can be generated by $(a, 1-Fa)$ and $(1, -F)$. We can check this by finding $u,v$ such that for all $(x,y)\in A_n\oplus A_n$ we can write 1611 \[(x,y) = (a, 1-Fa)u + (1, -F)v.\] 1612 Suitable choices are $u = y+Fx$ and $v = x-a(y+Fx)$.\\ 1613 Since $A_n\oplus A_n$ can be generated by $(a, 1-Fa)$ and $(1, -F)$ and $(a, 1-Fa)$ is an element of $M$, it follows that $T \cong (A_n\oplus A_n)/M$ can be generated by $\phi(1, -F)$, i.e. $T$ is cyclic. \\ 1614 For stable cyclicity we see that since $t_1, t_2$ were arbitrary generators, for $l = 1$ we can set $a_1 = t_1, a_2 = t_2$ and choose $f_1 = -F$, then $\phi(1, -F) = t_1 - t_2F$ generates $T$. For $l\geq 2$ we simply note that by the right module version of \cite[Theorem~3.7]{Stafford1978}, $T$ is already proven to be stably two-generated, therefore, T is stably cyclic. 1615 1616 \end{proof} 1617 \AIadd{\formalref{\leandecl{Stafford38/Paper/Main.lean}{67}{Stafford38.Paper.corollary_6_3}, 1618 \leandecl{Stafford38/Paper/Main.lean}{81}{Stafford38.Paper.corollary_6_3_stablyCyclic}, 1619 \leandecl{Stafford38/Paper/Section6/Cyclicity.lean}{219}{Stafford38.Paper.Section6.corollary_6_3_of}; 1620 Mathlib-only statements \leandecl{Stafford38/ChallengeDefinitions.lean}{302}{Stafford38Challenge.PaperCorollary_6_3}, 1621 \leandecl{Stafford38/PaperClaimDefinitions.lean}{33}{Stafford38Challenge.PaperCorollary_6_3_stablyCyclic}}} 1622 1623 \section{Generalizations and open questions}\label{sec:generalizations} 1624 The formalization in Lean discussed in Section~\ref{sec:formal} also proves the following consequences, which have no printed proof: the left-handed identity $1=Rd+SdF$, via the 1625 formal adjoint, which implies Corollary \ref{cor:cyclic} also for left torsion modules; the identity for finite-order differential operators on localizations of $k[x_1,\dots,x_n]$, including principal opens, 1626 Laurent-polynomial coefficients and the fraction field $k(x_1,\dots,x_n)$; and, independently 1627 of the geometric part of Section \ref{sec:conormal-review}, the identity with $F=x^r$ for 1628 $d=p^r-\sum_j a_jx^{m_j}$, $r\ge1$, in any $k$-algebra containing elements 1629 $x,p$ with $px=xp+1$, provided each $a_j$ commutes with $x$ and $p$. 1630 \formalref{\leandecl{Stafford38/Paper/Section7.lean}{38}{Stafford38.Paper.Section7.leftHandedIdentity}; 1631 \leandecl{Stafford38/Paper/Section7.lean}{72}{Stafford38.Paper.Section7.corollary_6_3_left}; 1632 \leandecl{Stafford38/Paper/Section7.lean}{225}{Stafford38.Paper.Section7.localizedDifferentialIdentity}; 1633 \leandecl{Stafford38/Paper/Section7.lean}{246}{Stafford38.Paper.Section7.evolutionaryIdentity}} 1634 1635 Two further results, also proved in Lean, give the identity $1=dR+FdS$ in larger classes of rings, though without the 1636 choice of $F$ in Theorem~\ref{thm:main}: Let $k$ be a field of 1637 characteristic zero. 1638 \begin{enumerate}[label=(\alph*),nosep,leftmargin=1.6em] 1639 \item Let $A$ be a smooth finitely generated $k$-algebra that is a domain, 1640 and let $D_k(A)$ be the ring of $k$-linear differential operators of finite 1641 order on $A$. For every nonzero $d\in D_k(A)$ there are $F,R,S\in D_k(A)$ 1642 with $1=dR+FdS$~\cite{GlobalStaffordPalomar2026,GlobalStaffordZenodo106}. 1643 For $A=k[x_1,\dots,x_n]$ this is the identity in $A_n(k)$. 1644 \formalref{\leanglobal{Solution.lean}{18}{GlobalStaffordChallenge.universalStatement}; 1645 statement \leanglobal{Challenge.lean}{70}{GlobalStaffordChallenge.UniversalStatement}} 1646 \item Let $U$ be an infinite-dimensional simple $k$-algebra with an 1647 exhaustive filtration $k=F_0U\subseteq F_1U\subseteq\cdots$ by 1648 finite-dimensional subspaces, $F_iU\cdot F_jU\subseteq F_{i+j}U$, whose 1649 associated graded ring is a commutative domain finitely generated over $k$. 1650 For all nonzero $d,a\in U$ and every $B\in U$ there are $H,R,S\in U$ with 1651 $1=dR+(B+Ha)dS$~\cite{SameDivisorFiltered}. The Bernstein filtration makes 1652 $A_n(k)$ such an algebra, and $B=0$, $a=1$ gives $1=dR+HdS$. 1653 \formalref{\leanfiltered{Solution.lean}{46}{SymplecticStaffordChallenge.universalStatement}; 1654 statement \leanfiltered{Challenge.lean}{54}{SymplecticStaffordChallenge.AffineSameDivisorStatement}} 1655 \end{enumerate} 1656 The proof of Theorem~\ref{thm:main} shows $Q=0$ without producing $R$ and 1657 $S$. Over a computable field, the fixed choice of $F$ nevertheless gives a terminating algorithm: 1658 enumerate the countably many pairs $(g,D)$ of linear symplectic changes of generators and degree 1659 bounds, and solve the linear system $1=dR+FdS$ with 1660 $F=\Phi_g^{-1}(x_1)^{\deg_Bd}$ for $R,S$ of Bernstein degree at most $D$. 1661 Theorem~\ref{thm:main} guarantees that some pair succeeds. We cannot provide bounds for 1662 $\deg_BR$ and $\deg_BS$ in terms of $\deg_Bd$, so the question for an efficient 1663 algorithm remains open. In (a) and (b), $F$ is not fixed, so this search does 1664 not apply. 1665 1666 1667 \backmatter 1668 1669 \section*{Declarations} 1670 1671 \bmhead{Acknowledgements} 1672 J.~Moser was funded by the Austrian Research Promotion Agency (FFG) within 1673 the programme AI for Green, project VENTUS (910263). The 1674 authors would like to acknowledge the use of HPC resources provided by the ZID 1675 of Graz University of Technology. We gratefully acknowledge support from NAWI 1676 Graz. 1677 1678 \bmhead{Competing interests} 1679 The authors have no competing interests to declare that are relevant to the 1680 content of this article. 1681 1682 \bmhead{Code availability} 1683 The Lean sources of the formal proof and the supplementary material are 1684 archived~\cite{Stafford38Zenodo140,Stafford38Supplementary032}. 1685 1686 \begin{appendices} 1687 1688 \section{AI-assisted research and formalization}\label{app:ai-workflow} 1689 1690 AI agents assisted with proof discovery, counterexample searches, 1691 review, formalization and writing. The work began with computations of torsion generators in 1692 Macaulay2 and Maple. Direct attempts to construct universal cofactors 1693 produced special cases and reductions, but no general construction. The 1694 decisive step was to study the quotient $Q=A_n/(dA_n+x_1^N dA_n)$ and to show 1695 that its characteristic support is empty. 1696 1697 The first complete argument was written by AI agents in the style of a paper 1698 proof. Agents then formalized it in Lean, and the formalization changed parts 1699 of the argument where the original steps could not be checked as written. 1700 After the Lean proof was complete and published, the paper proof was rewritten 1701 from the checked Lean proof, so that every statement corresponds to a Lean 1702 declaration, and the Lean proof was then again aligned with the printed proof 1703 step by step. The authors reviewed this correspondence and substantially revised the text for 1704 clarity and readability together with the agents, until neither the human review nor the AI 1705 checks raised further objections. 1706 1707 For finding the first complete argument, we proceeded as follows: 1708 One controlling agent kept the authoritative state of the proof. 1709 Parallel agents worked on separate approaches and returned evidence together 1710 with its scope, and independent reviewer agents checked each result before it was 1711 accepted. The first author steered the work with short instructions, and the 1712 agents wrote the detailed task prompts for each other. Advice from 1713 consultations counted only after an independent check. 1714 1715 The working rules developed together with the proof into a reusable 1716 agent skill for open mathematical problems. Early versions emphasized 1717 falsification and keeping conditional results apart from established ones. 1718 Once a candidate proof existed, the emphasis moved to repairing the first 1719 invalid step of a promising argument, and 1720 to auditing proofs. 1721 1722 Research and review used OpenAI GPT models from GPT-5.5 to GPT-6.1, 1723 Anthropic Claude models from Haiku~4.5 to Opus~5, and locally hosted Qwen 1724 and DeepSeek models. GPT and Claude models, most recently Claude Opus~5.5, 1725 helped draft and revise the manuscript and the supplementary material. The 1726 authors are responsible for the final text. 1727 1728 The formal proof uses Lean~4, Mathlib and AlgebraicAnalysis 1729 v0.3.3~\cite{Lean4,Mathlib2020,AlgebraicAnalysis033}. Conditional results 1730 were kept apart from proved ones until every hypothesis was discharged. 1731 Comparator checked that the proved theorems have exactly the statements of 1732 separate challenge files that use only Mathlib definitions, and both the Lean 1733 kernel and the independent checker NanoDa accepted the proofs 1734 \cite{Comparator,NanoDa,PalomarDocs}. Theorem~\ref{thm:main} and 1735 generalization~(a) of Section~\ref{sec:generalizations} are registered in the 1736 Palomar registry~\cite{Stafford38Palomar2026,GlobalStaffordPalomar2026}, and 1737 their Lean sources are archived~\cite{Stafford38Zenodo140,GlobalStaffordZenodo106}. 1738 1739 \end{appendices} 1740 1741 1742 \bibliography{references} 1743 1744 1745 1746 1747 1748 \end{document} 1749