Reduction type of smooth quartics
Extracted source text · arXiv 1803.05816
\documentclass[reqno, 10pt, a4paper]{amsart}
\usepackage{fullpage}
\linespread{1.3}\selectfont
\usepackage[utf8]{inputenc}
\usepackage[OT2,T1]{fontenc}
\usepackage[english]{babel}
\usepackage{amsmath}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsthm}
\usepackage{mathrsfs}
\usepackage{listings}
\lstset{basicstyle=\ttfamily}
\usepackage[mathcal]{euscript}
\usepackage{bbm}%
\usepackage{enumitem}
\usepackage{multirow}
\usepackage[dvipsnames]{xcolor}
\usepackage{graphicx}
\usepackage{tikz}
\usetikzlibrary{matrix}
\usepackage[all]{xy}
\usepackage{colortbl}
\definecolor{mygray}{gray}{0.92}
\newcommand{\ctbl}{\cellcolor{mygray}}
\usepackage{array}
\newcolumntype{C}[1]{>{\centering\arraybackslash$}p{#1}<{$}}
\usepackage{breqn}
\newcounter{myequation}[equation]
\newenvironment{BREQequation}{%
\rm\setcounter{myequation}{\theequation}%
}{%
\setcounter{equation}{\themyequation}
}
\usepackage{float}
\restylefloat{table}
\usepackage{multirow}
\usepackage{hyperref}
\theoremstyle{plain}
\newtheorem{theorem}{Theorem}[section]
\newtheorem{conjecture}[theorem]{Conjecture}
\newtheorem{question}[theorem]{Question}
\newtheorem{proposition}[theorem]{Proposition}
\newtheorem{lemma}[theorem]{Lemma}
\newtheorem{corollary}[theorem]{Corollary}
\newtheorem{scholium}[theorem]{Scholium}
\newtheorem{algorithm}[theorem]{Algorithm}
\theoremstyle{definition}
\newtheorem{definition}[theorem]{Definition}
\theoremstyle{remark}
\newtheorem{remark}[theorem]{Remark}
\newtheorem{example}[theorem]{Example}
\newtheorem{fact}[theorem]{Fact}
\numberwithin{equation}{section}
\renewcommand{\theenumi}{\roman{enumi}}
\renewcommand{\labelenumi}{(\theenumi)}
\def\defi{\textsf}
\def\ext{\!\mid\!}
\def\limdir{\varinjlim}
\def\liminv{\varprojlim}
\def\eps{\varepsilon}
\def\epsilon{\varepsilon}
\def\thetac{\theta}
\def\thetar{\vartheta}
\def\tilde{\widetilde}
\newcommand{\trv}[3]{( #1,#2 )_{#3}}
\def\Belyi{Bely\u{\i}}
\def\LiE{\textsc{LiE}}
\def\Magma{\textsc{Magma}}
\def\SageMath{\textsc{SageMath}}
\def\NB{\textbf{N.B.}}
\def\cf{\textit{cf.}}
\def\etal{\textit{et al.}}
\def\etc{\textit{etc.}}
\def\eg{\textit{e.g.}}
\def\ie{i.e.}
\def\infra{\textit{infra}}
\def\loccit{\textit{loc.\ cit.}}
\def\PHM{toggle model}
\DeclareMathOperator{\alg}{alg}
\DeclareMathOperator{\ab}{ab}
\DeclareMathOperator{\ad}{ad}
\DeclareMathOperator{\an}{an}
\DeclareMathOperator{\AS}{AS}
\DeclareMathOperator{\Aut}{Aut}
\DeclareMathOperator{\BT}{BT}
\DeclareMathOperator{\Char}{char}
\DeclareMathOperator{\chr}{char}
\DeclareMathOperator{\Cl}{Cl}
\DeclareMathOperator{\coker}{coker}
\DeclareMathOperator{\cris}{cris}
\DeclareMathOperator{\cts}{cts}
\DeclareMathOperator{\Disc}{Disc}
\DeclareMathOperator{\disc}{disc}
\DeclareMathOperator{\Div}{div}
\DeclareMathOperator{\DOv}{\underline{DO}}
\DeclareMathOperator{\DO}{DO}
\DeclareMathOperator{\et}{\acute{e}t}
\DeclareMathOperator{\End}{End}
\DeclareMathOperator{\Ext}{Ext}
\DeclareMathOperator{\Frac}{Frac}
\DeclareMathOperator{\Frob}{Frob}
\DeclareMathOperator{\Gal}{Gal}
\DeclareMathOperator{\GL}{GL}
\DeclareMathOperator{\GO}{GO}
\DeclareMathOperator{\GSp}{GSp}
\DeclareMathOperator{\Hom}{Hom}
\DeclareMathOperator{\im}{im}
\DeclareMathOperator{\id}{id}
\DeclareMathOperator{\Ind}{Ind}
\DeclareMathOperator{\intt}{int}
\DeclareMathOperator{\Isom}{Isom}
\DeclareMathOperator{\Jac}{Jac}
\DeclareMathOperator{\JI}{JI}
\DeclareMathOperator{\Mon}{Mon}
\DeclareMathOperator{\mult}{mult}
\DeclareMathOperator{\new}{new}
\DeclareMathOperator{\Nm}{Nm}
\DeclareMathOperator{\Norm}{Nm}
\DeclareMathOperator{\NS}{NS}
\DeclareMathOperator{\old}{old}
\DeclareMathOperator{\Olive}{Olive}
\DeclareMathOperator{\ord}{ord}
\DeclareMathOperator{\Out}{Out}
\DeclareMathOperator{\pgcd}{pgcd}
\DeclareMathOperator{\Pic}{Pic}
\DeclareMathOperator{\PGL}{PGL}
\DeclareMathOperator{\PSL}{PSL}
\DeclareMathOperator{\PU}{PU}
\DeclareMathOperator{\proj}{proj}
\DeclareMathOperator{\Proj}{Proj}
\DeclareMathOperator{\res}{res}
\DeclareMathOperator{\Res}{Res}
\DeclareMathOperator{\rig}{rig}
\DeclareMathOperator{\sgn}{sgn}
\DeclareMathOperator{\Sh}{S}
\DeclareMathOperator{\Shv}{\underline{S}}
\DeclareMathOperator{\SL}{SL}
\DeclareMathOperator{\SO}{SO}
\DeclareMathOperator{\Sp}{Sp}
\DeclareMathOperator{\ssim}{ss}
\DeclareMathOperator{\Spec}{Spec}
\DeclareMathOperator{\Spf}{Spf}
\DeclareMathOperator{\Sym}{Sym}
\DeclareMathOperator{\Tor}{Tor}
\DeclareMathOperator{\tr}{tr}
\DeclareMathOperator{\Tr}{Tr}
\DeclareMathOperator{\Span}{Span}
\DeclareMathOperator{\tors}{tors}
\DeclareMathOperator{\rk}{rk}
\DeclareMathOperator{\Sel}{Sel}
\DeclareMathOperator{\Twist}{Twist}
\DeclareMathOperator{\vnorm}{v^{\textrm{\tiny norm}}}
\newcommand{\viff}{\rotatebox[origin=c]{-90}{$\iff$}}
\newcommand{\vrightarrow}{\rotatebox[origin=c]{-90}{$\Rightarrow$}}
\newcommand{\dq}{{/\kern -3pt/}}
\newcommand{\ds}{{\mathrm{ss}}}
\def\l{\ell}
\def\a{\mathfrak{a}}
\def\b{\mathfrak{b}}
\def\f{\mathfrak{f}}
\def\m{\mathfrak{m}}
\def\n{\mathfrak{n}}
\def\p{\mathfrak{p}}
\def\q{\mathfrak{q}}
\def\r{\mathfrak{r}}
\def\A{\mathbb{A}}
\def\B{\mathcal{B}}
\def\C{\mathbb{C}}
\def\D{\mathcal{D}}
\def\E{\text{E}}
\def\F{\mathbb{F}}
\def\FF{\mathcal{F}}
\def\G{\mathbb{G}}
\def\H{\mathcal{H}}
\def\L{\mathcal{L}}
\def\N{\mathbb{N}}
\def\O{\mathcal{O}}
\def\P{\mathbb{P}\,}
\def\Q{\mathbb{Q}}
\def\R{\mathbb{R}}
\def\T{\mathbb{T}}
\def\Z{\mathbb{Z}}
\def\Phat{\widehat{P}}
\def\Xhat{\widehat{X}}
\def\Yhat{\widehat{Y}}
\def\Zhat{\widehat{Z}}
\def\abar{\overline{a}}
\def\bbar{\overline{b}}
\def\fbar{\overline{f}}
\def\kbar{\overline{k}}
\def\xbar{\overline{x}}
\def\Cbar{\overline{C}}
\def\Ebar{\overline{E}}
\def\Fbar{\overline{\F}}
\def\Gbar{\overline{G}}
\def\Kbar{\overline{K}}
\def\Qbar{\overline{Q}}
\def\xbar{\overline{x}}
\def\Xbar{\overline{X}}
\def\ybar{\overline{y}}
\def\Ybar{\overline{Y}}
\def\Zbar{\overline{Z}}
\def\phibar{\overline{\varphi}}
\def\el{\mathfrak{e}}
\def\fl{\mathfrak{f}}
\def\hl{\mathfrak{h}}
\def\sll{\mathfrak{sl}}
\usepackage{bm}
\def\gammab{\bm{\gamma}}
\def\Deltab{\bm{\Delta}}
\def\etab{\bm{\eta}}
\def\rhob{\bm{\rho}}
\def\sigmab{\bm{\sigma}}
\def\taub{\bm{\tau}}
\def\psib{\bm{\psi}}
\def\chib{\bm{\chi}}
\def\Cb{\bm{C}}
\def\Fb{\bm{F}}
\def\Hb{\bm{H}}
\def\Ib{\bm{I}}
\def\Jb{\bm{J}}
\def\Mb{\bm{M}}
\def\Nb{\bm{N}}
\def\Pb{\bm{P}}
\def\Qb{\bm{Q}}
\def\Sb{\bm{S}}
\def\Tb{\bm{T}}
\def\Xb{\bm{X}}
\def\Yb{\bm{Y}}
\def\Zb{\bm{Z}}
\def\bb{\bm{b}}
\def\fb{\bm{f}}
\def\gb{\bm{g}}
\def\hb{\bm{h}}
\def\jb{\bm{j}}
\def\qb{\bm{q}}
\def\sb{\bm{s}}
\def\ub{\bm{u}}
\def\ff{\bm{f}}
\def\jj{\bm{j}}
\def\ss{\bm{s}}
\def\qq{\bm{q}}
\def\Cs{\mathsf{C}}
\def\xs{\mathsf{x}}
\def\Xs{\mathsf{X}}
\def\Ys{\mathsf{Y}}
\def\ys{\mathsf{y}}
\def\Zs{\mathsf{Z}}
\def\Oc{\mathcal{O}}
\def\Cc{\mathcal{C}}
\def\Jc{\mathcal{J}}
\def\Lc{\mathcal{L}}
\def\Mc{\mathcal{M}}
\def\xcal{{\scriptstyle \mathcal{X}}}
\def\xfrak{{\scriptstyle \mathfrak{X}}}
\def\av{\underline{a}}
\def\bv{\underline{b}}
\def\bbv{\underline{\bm{b}}}
\def\dv{\underline{d}}
\def\ivv{\underline{\iota}}
\def\iv{\iota}
\def\Iv{\underline{I}}
\def\Jv{\underline{J}}
\def\vv{\underline{v}}
\def\wv{\underline{w}}
\def\xv{\underline{x}}
\newcommand{\car}[2]{%
\left[{\substack{#1\\#2}}\right]}
\newcommand{\newcar}[2]{%
\left[%
\begin{smallmatrix}%
\displaystyle#1\\%
\displaystyle#2\end{smallmatrix}\right]}
\hypersetup{
pdfauthor = {Lercier, Liu, Lorenzo Garc\'ia, Ritzenthaler},
pdftitle = {Reduction type of smooth quartics},
pdfsubject = {},
pdfkeywords = {},
backref=true, pagebackref=true, hyperindex=true, colorlinks=true,
breaklinks=true, urlcolor=blue, linkcolor=blue, citecolor=blue,
bookmarks=true, bookmarksopen=true}
\begin{document}
\title{Reduction type of smooth quartics}
\date{\today}
\begin{abstract}
%
Let $C/K$ be a smooth plane quartic over a discrete valuation field. We give
a characterization of the type of reduction (\ie~smooth plane quartic,
hyperelliptic genus 3 curve or bad) over $K$ in terms of the existence of a
special plane quartic model and, over $\bar{K}$, in terms of the valuations
of the Dixmier-Ohno invariants of $C$ (if the characteristic of the residue
field is not $2,\,3,\,5$ or $7$). When the reduction is (potentially) good
we also provide an equation for the special fiber of a generic quartic. On
the way, we gather general results on geometric invariant theory over an
arbitrary ring $R$ in the spirit of \cite{Se}. For instance when $R$ is a
discrete valuation ring, we show the existence of a homogeneous system of
parameters over $R$ and we exhibit precise ones for ternary quartic forms
under the action of $\SL_{3,R}$ depending only on the characteristic of the
residue field.
\end{abstract}
\author[Lercier]{Reynald Lercier}
\address{%
Reynald Lercier,
DGA \& Univ Rennes, %
CNRS, IRMAR - UMR 6625, F-35000
Rennes, %
France. %
}
\email{reynald.lercier@m4x.org}
\author[Liu]{Qing Liu}
\address{%
Qing Liu,
Institut de mathématiques de Bordeaux, CNRS - UMR 5251,
Université de Bordeaux,
F-33405 Talence, %
France. %
}
\email{qing.liu@u-bordeaux.fr}
\author[Lorenzo]{Elisa Lorenzo Garc\'ia}
\address{%
Elisa Lorenzo Garc\'ia,
Univ Rennes, CNRS, IRMAR - UMR 6625, F-35000
Rennes, %
France. %
}
\email{elisa.lorenzogarcia@univ-rennes1.fr}
\author[Ritzenthaler]{Christophe Ritzenthaler}
\address{%
Christophe Ritzenthaler,
Univ Rennes, CNRS, IRMAR - UMR 6625, F-35000
Rennes, %
France. %
}
\email{christophe.ritzenthaler@univ-rennes1.fr}
\thanks{The authors would like to thank Jeroen Sijsling for his careful reading of an earlier version of the paper, Matthieu Romagny for useful references and Bianca Viray and Armand Brumer for inspiring discussions. The second named author is partly supported by Xiamen University and the NSFC grant no. 11031004.
}
\subjclass[2010]{11G20, 14Q05, 14D10, 14D20, 14H25}
\keywords{smooth plane quartic, reduction, hyperelliptic, invariants, valuation}
\maketitle
\section{Introduction and main results}
Let $K$ be a discrete valuation field with valuation $v$, valuation ring $\O$
and a uniformizer $\pi$. Let $k=\Oc/\langle \pi \rangle$ be the residue field
of characteristic $p \geq 0$. When $F$ is an integral polynomial, \ie~with
coefficients in $\O$, we denote by $\bar{F}$ its reduction modulo $\pi$. A
smooth projective geometrically connected curve of genus $3$ is either a plane
quartic or a hyperelliptic curve depending on whether the canonical divisor is
very ample or not. Let $C/K$ be a {(smooth)} plane quartic.
\begin{itemize}
\item We say that $C$ \emph{has good (resp. good quartic, resp. good
hyperelliptic) reduction} over $K$ if $C$ has a projective smooth model
$\Cc$ over $\O$ (resp. whose special fiber is a smooth plane quartic,
resp. a hyperelliptic curve).
\item We say that $C$ \emph{has potentially good (resp. potentially good
quartic reduction, resp. potentially good hyperelliptic reduction)} if for
some finite separable extension $(K', v')$ of $(K, v)$, $C_{K'}/K'$ has good
(resp. good quartic, resp. good hyperelliptic) reduction over $K'$.
\item We say that $C$ \emph{has bad reduction} over $K$ if it does not have
good reduction over $K$. We say that it \emph{has not potentially good
reduction} if it has bad reduction over any finite separable extensions
$(K', v')$ of $(K, v)$.
\end{itemize}
Note that the type of the potential reduction does not depend on the choice of
$(K', v')$. Hence, in the following, the expression ``after a possible finite
extension of $K$'', or \emph{potentially}, means that we are allowed to take a
finite extension of $K$ and still call $K, \O, v, \pi$ the corresponding
notions.
Given such a curve $C/K$, we want to determine the reduction type of $C$ among
the above possibilities. In the first two cases, we also look for an explicit
equation of the special fiber. In the first case, it will be again a smooth
plane quartic over $k$, whereas in the second case it will be isomorphic over
$\bar{k}$ to $y^2=f(x,z)$ where $f$ is a binary octic with no multiple roots.
\begin{example}
Let us consider the plane smooth quartic $C/\Q : F=0$ where
$$F=(x_2 +x_3) x_1^3-(2 x_2^2 + x_3 x_2) x_1^2+(x_2^3-x_3 x_2^2+2 x_3^2 x_2-x_3^3) x_1-2 x_3^2 x_2^2+3 x_3^3 x_2.$$
This curve is an equation of $X_{ns}(13) \simeq X_s(13) \simeq X_0^{+}(169)$
which is studied in \cite[Cor.6.8]{BDMTV17}. By using a general result
from~\cite{edixhoven90}, they prove that this curve has good quartic
reduction away from $13$ and potentially good hyperelliptic reduction at 13
after a ramified extension of degree 84. By using the characterizations we
are giving in this article (see Example~\ref{ex:bass}) and our
implementations, we can easily check that the stable reduction at 13 is the
hyperelliptic curve of affine equation $y^2=x^7-1$.
\end{example}
The most complicated task will be to distinguish good hyperelliptic reduction
from bad reduction. By identifying a plane quartic with its $15$ coefficients
up to a multiplicative constant, we can view the set $\Xb$ of smooth plane
quartics as a subset of $\P^{14}$. Since isomorphisms between smooth plane
quartics are linear, $\Xb/\SL_3(\bar{K})$ gives a description of the locus
$\Mb_3^{\textrm{qrt}}$ of plane quartics inside the moduli space $\Mb_3$ of
all curves of genus $3$. Looking at the locus of hyperelliptic curves as ``a
boundary'' of the locus of $\Mb_3^{\textrm{qrt}}$ inside $\Mb_3$, the
challenge is to understand whether a family of plane smooth quartics has limit
in this boundary. Similar situations have been studied to describe the
boundary of the compactification of $\Mb_3$ and there is a rich literature on
this subject \cite{artebani, glass, hassett, hyeon, kang, kondo, mum-ens,
yukie}. Surprisingly, the mentioned problem has never been fully addressed
before (still, see \cite[Prop.14 and Prop.15]{howe00} in the case of Ciani
quartics). Known results about the relations between plane quartics and
hyperelliptic genus 3 curves are basically contained in the following
proposition \cite[p.156]{clemens}, \cite{harriscm},
\cite[p.134]{harris-moduli}.
\begin{proposition}\label{2->1} Suppose $p \ne 2$.
Let $s >0$ be an integer, $G \in \O[x_1,x_2,x_3]$ be a primitive (\ie~the
gcd of its coefficients is 1) quartic form and $Q \in \O[x_1,x_2,x_3]$ be a
primitive quadratic form. Let us assume that $\bar{Q}$ is irreducible and
that $\bar{Q}=0$ intersects $\bar{G}=0$ transversely in $8$ distinct
$\overline{k}$-points. Let us denote $\Cc/\O$ the subscheme of the weighted
projective space $\P^{(1,1,1,2)}$ defined by
$$\begin{cases}
y^2 + G = 0, & \\
\pi^s y - Q =0. &
\end{cases}$$
Then the generic fiber is isomorphic to the plane smooth quartic
$C/K : Q^2 +\pi^{2s} G=0$ which has good hyperelliptic reduction. The
special fiber of $\Cc$ is isomorphic to the double cover of $\bar{Q}=0$
ramified over the 8 distinct intersection $\overline{k}$-points of
$\bar{Q}=0$ with $\bar{G}=0$.
\end{proposition}
This motivates the following definitions when $p \ne 2$ (see
Theorem~\ref{th:is2} for the case $p=2$).
\begin{definition}\label{def:HGRM}
Let $C/K$ be a plane smooth quartic. We say that $C$ admits a \emph{\PHM} if
there exist an even integer $s>0$, a primitive quartic form
$G \in \O[x_1,x_2,x_3]$ and a primitive quadric $Q \in \O[x_1,x_2,x_3]$ with
$\bar{Q}$ irreducible such that $Q^2+\pi^s G=0$ is $K$-isomorphic to
$C$. If, moreover, $\bar{Q}=0$ intersects $\bar{G}=0$ transversely in $8$
distinct $\overline{k}$-points, the model $Q^2+\pi^s G=0$ is a \emph{good
\PHM} of $C$.
\end{definition}
When $C$ admits a good \PHM, it has good hyperelliptic reduction. The converse
holds as well, this is the main result of Section~\ref{sec:hyperphm}.
\begin{theorem}[See Theorems~\ref{th:goodmodel-f} and
\ref{th:is2}] \label{th:goodmodel} Let $C/K$ be a plane smooth quartic.
Then $C$ admits good hyperelliptic reduction if and only if $C$ has a good
{\PHM} over $K$.
\end{theorem}
By using this result, the methods of \cite{kollar} and \cite{elsenhans-stoll}
and some ingredients from invariant theory, the question of the reduction type
over $K$ of a smooth quartic may be answered (see remark~\ref{rem:kollar}). We
turn now to the determination of the type when one allows extensions of $K$.
In the spirit of the work of \cite{liug2} on genus 2, we aim at getting a
characterization of the potential reduction type in terms of the valuations of
certain invariants of $C/K$ under the action of $\PGL_3(\bar{K})$. Let us
start by recalling some results on the algebra of invariants of ternary
quartic forms.
In \cite{dixmier}, Dixmier worked out a list of 7 homogeneous polynomial
invariants for the equivalence of ternary quartic forms under the action of
$\SL_3(\C)$, denoted $I_3, I_6, I_9,I_{12},I_{15},I_{18}$ and $I_{27}$. They
form a \emph{homogeneous system of parameters} (HSOP), which means that the
ring of invariants is a module of finite type over the ring generated by these
invariants (see also Definition~\ref{def:hsop} in a more general
setting). Later, this list was completed by \cite{ohno} (see also
\cite{elsenhans}) into a list of 13 generators, defined over
$\Z[\frac{1}{2 \cdot 3 \cdot 5 \cdot 7}]$, for the $\C$-algebra of invariants,
which are now called the \emph{Dixmier-Ohno invariants}. We denote $\DOv$ the
list of these invariants, $\DOv(F) \in K^{13}$ their values at a ternary
quartic form $F$ and when at least one of these values is not zero, we denote
$\DO(F)$ the corresponding point in the weighted projective space with weights
$3,6,9,9,12,12,15,15,18,18,21,21,27. $
Two smooth quartics $C_i:\,F_i=0$ are isomorphic over $\C$ if and only if
$\DO(F_1)=\DO(F_2)$.
This result over $\C$ is clearly not sufficient for our purposes as we want to
deal with fields of arbitrary characteristic and with reduction
properties. Fortunately, Seshadri in \cite{Se} has generalized the Geometric
Invariant Theory (GIT) of Mumford over an arbitrary ring $R$. For the
convenience of the reader, we include in Section~\ref{sec:appendix} some
general results on this topic and proofs we were unable to find in the
literature. We also prove, under some assumptions on $R$\footnote{the
assumptions are satisfied for instance for local rings and rings of integers
of global fields.}, that there exists a HSOP over $R$, \ie~a finite list of
invariants defined over $R$ being a HSOP after base change to all residue
fields of $R$. It is a bit of a surprise as one knows that there exist
invariants over fields of small characteristics which do not come from
reducing invariants in characteristic $0$ (see \cite{basson} for binary forms
and remark~\ref{rem:nonlift} for ternary quartics). In
Theorem~\ref{th:domodp}, we determine an explicit HSOP over $\O$ for ternary
quartic forms which depends only on the characteristic of the residue
field. Notice that, there exists one even over $\Z$ but the degrees may be too
large to be practical. Still, we show that $\DOv$ always contains a
HSOP\footnote{and \emph{not} the Dixmier invariants alone even for some large
$p$ like for instance $p=523$.} for $p \geq 7$, and we give\footnote{see
the \Magma~\cite{Magma}
\href{https://perso.univ-rennes1.fr/christophe.ritzenthaler/programme/reduction-quartics/}{code}
used to check this.} analogous lists when the residue field is of
characteristic $2$, $3$ and $5$. With the normalization of
\cite[p.426]{gelfand}, \cite{demazure}, the \emph{discriminant} $D_{27}$ of a
ternary quartic is a degree $27$ polynomial with coefficients in $\Z$ and
always belongs to our HSOPs (up to a multiplicative unit). For instance, it is
related to the Dixmier invariant $I_{27}$ by $2^{40} \cdot I_{27}=D_{27}$. A
plane quartic $F=0$ over any field is smooth if and only if $D_{27}(F) \ne 0$.
As we want to deal with valuations of the invariants, we need to introduce the
following notation before stating our results. Let
$\dv=(d_0,\ldots,d_n)\in\Z_{>0}^{n+1}$. Given an element
$\xv=(x_0,\ldots,x_n) \in K^n \setminus \{0\}$, we denote by
$x=(x_0:\ldots:x_n)$ the corresponding point in the weighted projective space
$\P^{\dv}(K)$. After a possible extension of $K$, one can always find a
representative $(X_0,\ldots,X_n) \in \O^n$ of $x$ such that one of the $X_i$'s
has valuation $0$. We call such a representative a \emph{minimal
representative} and denote it $\xv^{\min}$. It is not uniquely defined but
two minimal representatives differ by the action of a unit and their
valuations are coordinate-wise equal.
If $y \in K$ and $e \in \Z_{>0}$, we call the \emph{normalized valuation of
degree $e$ of $y$ with respect to $\xv$} the valuation $v_{x}(y)=v(Y)/e$
where $(X_0,\ldots,X_n,Y)$ is a minimal representative of
$(x_0:\ldots:x_n:y) \in \P^{\dv,e}$ (by convention $v(0)=\infty$). In other
words
$$v_x(y) = v(y)/e - \min \{v(y)/e,\,v(x_i)/d_i:\,i=0,\dots,n\} \in \Q_{\geq 0} \cup \{\infty\}.$$
Let $ \Iv =(I_0, \ldots, I_n)$ be a list of homogeneous polynomials of degree
$\underline{d}$ in $K[a_1,\ldots,a_m]$. For $F \in K^m$, if
$\Iv(F) = (I_0(F), \ldots, I_n(F))$ is not the zero vector, and if $J$ is a
homogeneous polynomials of degree $e$ we denote by $v_{I}(J(F))$ the
normalized valuation of degree $e$ of $J(F)$ with respect to $\Iv$. Finally,
if $F_0 \in \O^m$ and $\Iv(F_0) = \Iv(F_0)^{\min}$ we say that $F_0$ is a
\emph{minimal form}.
In the context of invariants of ternary quartic forms $F$, a natural question,
and a key argument for the sequel, is to know if, after a possible extension
of $K$, there exists a minimal form $\GL_3(K)$-equivalent to $F$. We show in
Proposition~\ref{prop:ssred} that this is the case when $\O$ is excellent,
$F=0$ is stable over $K$ in the GIT sense and the list $\Iv$ contains a HSOP
over $\O$. Using this result, we obtain the following theorem.
\begin{theorem}[See Theorem~\ref{th:gnhr}] \label{th:GNHR} Let $\Iv$ be a list
of invariants containing a HSOP for ternary quartic forms under the action
of $\SL_{3,\O}$ (when $p \geq 7$, one can take $\Iv=\DOv$). The smooth plane
quartic $C/K : F=0$ has potentially good quartic reduction if and only if
$v_{I}(D_{27}(F))=0$ (in other words, for any invariant $I_i$ of degree
$d_i$ in the list $\Iv$, we have $I_i(F)^{27}/D_{27}(F)^{d_i}\in \O$).
\end{theorem}
This theorem is proved in Section~\ref{sec:Shah}. As a direct application, we
give a characterization of the primes of geometrically bad reduction for
Picard curves (see Theorem~\ref{ex:picard}).\\
From Section~\ref{sec:comphyp} and beyond, we restrict to a DVR $\O$ with
residue field $k$ of characteristic $p=0$ or $p>7$ (see a discussion about
this restriction in remark~\ref{rem:p>7}). We focus there on distinguishing
potentially hyperelliptic reduction and bad reduction. In
Proposition~\ref{prop:MHR}, we first prove that $C/K : F=0$ potentially admits
a (not necessarily good) \PHM{} if and only if any minimal representative of
$\DO(F)$ reduces modulo $\pi$ to $\DO(Q_0^2)$ where $Q_0=x_2^2-4 x_1 x_3$ is a
fixed non-degenerate quadric (the choice of this particular one will become
clearer later). This uses the characterization of strict semi-stable quartics
with minimal orbit in \cite{artebani}.
It remains to characterize in terms of the invariants when a \PHM{}
$Q_0^2+\pi^s G=0$ is good, \ie~the transversal intersection of $\bar{Q}_0=0$
with $\bar{G}=0$. This is the content of
Section~\ref{sec:char-hyper-reduct}. In order to do so, we introduce a
well-known $\SL_2(\bar{K})$-equivariant linear morphism $b_8$ from the space
of ternary quartic forms $F$ to the space of binary octic forms. The action
on the ternary quartics is given through a morphism
$h : \SL_2(\bar{K}) \to \SL_3(\bar{K})$ whose image denoted $\SO_3(\bar{K})$
is the stabilizer of $Q_0$. Starting from a good \PHM{} $F=Q_0^2+\pi^s G$ of
$C/K$, a quick computation shows that $b_8(F)=\pi^s b_8(G)$ and that
$y^2=\overline{b_8(G)}$ is an affine equation of its special fiber. The
algebra of invariants of binary octics under the action of $\SL_2(\bar{K})$ is
generated by the classical Shioda invariants $j_2,\ldots,j_{10}$, see
\cite{shioda67}. The invariants $j_2,\ldots,j_7$ still forms a HSOP over $\O$,
see Corollary~\ref{cor:Shiodas} and the discussion before. Evaluating the
Dixmier-Ohno invariants on a \PHM{} $Q_0^2+\pi^s G=0$ and the Shioda
invariants $j_2,\ldots,j_7$ on $b_8(G)$ gives algebraic relations between the
two sets and considering their $\pi$-expansion reveals a list
$\iota=(\iota_6,\iota_9,\iota_{12},\iota_{15},\iota_{18},\iota_{21})$ of six
explicit invariants for ternary quartic forms, see
equation~\eqref{eq:DOshort}. Moreover we get a similar
congruence~\eqref{eq:trompe} between the discriminant of binary octics,
$D_{14}$, and $I_3^5\, I_{27}$ (which is an invariant of degree $42$). For a
good \PHM{}, one has $v_{\iota}(I_3^5(F)\, I_{27}(F)) =
v(D_{14}(b_8(G)))=0$. Actually, we have the following theorem.
\begin{theorem} \label{th:main} Let us assume that $\O$ is a DVR with residue
field $k$ has characteristic $p \ne 2,3,5,7$. Let $C/K$ be a plane smooth
quartic defined by $F=0$. Then $C$ has potentially good hyperelliptic
reduction if and only if
$$v_{\DO}(I_3(F))=0, \quad v_{\DO}(I_{27}(F))>0 \; \textrm{and} \;
v_{\iv}(\,I_3(F)^5\, I_{27}(F)\,)=0.$$
Under these conditions, one can also obtain an explicit equation for the
special fiber.
\end{theorem}
To prove that the conditions of Theorem~\ref{th:main} are also sufficient, one
first observes that they imply the existence of a \PHM{} $F=Q_0^2+\pi^s G$ for
$C$. Then, a crucial step is to prove that we can choose it such that
$\overline{b_8(G)}$ is semi-stable in the GIT sense. In order to do this, we
rigidify a \PHM{} in the following way: there is, up to a multiplicative
constant, a unique contravariant $\rho$ of order 2 and degree 4, which can be
seen as a quadratic form whose coefficients are degree $4$ polynomials in the
coefficients of $F$ (see \cite{dixmier} and \cite{LRS16}). One can always find
a \PHM{} $F$ such that $\rho(F) = \rho(Q_0^2)$. This extra-information forces
the integrality of the \PHM{} we are looking for (see
Lemma~\ref{a>=b}).\\
Let us point out that the conditions in Theorem~\ref{th:main} may give the
wrong idea that the study of the valuations of $I_{3}$ and $I_{27}$ would be
enough to characterize the reduction. But this is not true since we need to
consider them inside the broader lists of invariants $\DO$ and $\iota$, and
normalize them accordingly to the valuations of all the invariants of these
lists. Notice also the expression $I_3(F)^5I_{27}(F)$: in a forthcoming
article, we will show that this is the pull-back of the Siegel modular form
$\chi_{28}$ to the space of quartics. Now, over $\C$, the non-vanishing of
this modular form together with the vanishing of the modular form $\chi_{18}$
(which is known to be related to $I_{27}^2$) characterizes the locus of
hyperelliptic Jacobians among abelian varieties (see \cite{tsuyumine-g3}). A
natural guess is to expect that they would be the key forms to find the
reduction type of a quartic. However, their expressions in terms of invariants
seem to lead to the absurd statement that the vanishing of $\chi_{18}$ always
implies the vanishing of $\chi_{28}$. Within this article, one sees that this
is because these invariants have to be considered inside broader lists to
become significant.
One of the motivations of this work was, in the continuation of \cite{WINE1},
\cite{WINE1-2} and \cite{KLLRSS17}, to understand the primes dividing the
discriminant $I_{27}$ of a plane smooth quartic with complex multiplication
over a number field only in terms of the CM-field. We hope that the present
work can be useful for this task. Another direction is to refine the study of
the bad reduction type in the spirit of \cite{BW}, \cite{maistret} to describe
more precisely the degenerated cases.
\section{Hyperelliptic reduction in terms of \PHM s} \label{sec:hyperphm}
\newcommand{\caract}{\mathrm{char}} \newcommand{\cO}{{\mathcal O}}
\newcommand{\cC}{{\mathcal C}} \newcommand{\W}{{\mathcal W}}
\newcommand{\cZ}{{\mathcal Z}} \newcommand{\ol}{\overline}
Before proving Theorem~\ref{th:goodmodel}, we need some preliminary results.
\subsection{Some general facts} \label{general-facts}
In this section we prove the existence of suitable models when $C$ has good
hyperelliptic reduction over $K$ (see Theorems \ref{th:goodmodel-f} and
\ref{th:is2}).
We start with some geometric observations. Suppose $C$ has good reduction over
$K$. Let $\cC$ be a smooth projective model over $\cO$. Let
$\omega_{\cC/\cO}$ be the dualizing sheaf ($=\Omega^1_{\cC/\cO}$ as $\cC$ is
smooth). Then $\omega_{\cC/\cO}$ is a base-point free invertible sheaf and
induces a morphism
$$ f: \cC \to \P(H^0(\cC, \omega_{\cC/\cO}))\simeq \P^2_\cO. $$
On the generic fiber, $f$ is a closed immersion because $C$ is a plane
quartic. Let $\cZ=f(\cC)$ be the image of $f$ endowed with the reduced
subscheme structure. Then $f: \cC\to \cZ$ is a birational finite morphism. As
$\cC$ is normal, $f$ is just the normalization morphism.
\begin{fact} \label{fact} If we have a finite birational morphism
${\W}\to \cZ$, and if ${\W}$ is normal, then $\W \to \cZ$ is the
normalization morphism (\cite[Proposition 4.1.22]{liu-book}), hence ${\W}$
is isomorphic to $\cC$. \end{fact}
If $\omega_{\cC/\cO}$ is very ample, then $f$ is a closed immersion and $\cC$
is a smooth quartic in $\P^2_\cO$. Suppose from now on that $\omega_{\cC/\cO}$
is not very ample. Then $f_k : \cC_k \to \P^2_{k}$ is a finite generically
\'etale morphism of degree $2$ into its image $(\cZ_k)_{\mathrm{red}}$ which
is a smooth conic over $k$. The scheme $\cZ$ is defined by an irreducible
quartic form $F\in \cO[x_1, x_2,x_3]$ as
\begin{equation} \label{eq:defZ} \cZ=V_+(F)\subset \P^2_\cO.
\end{equation}
Let $q\in k[x_1,x_2,x_3]$ be a nondegenerate quadratic form defining
$(\cZ)_{\mathrm{red}}$:
\begin{equation}
\label{eq:conic}
\cZ_{\mathrm{red}}=V_+(q) \subset \P^2_k.
\end{equation}
In the sequel, $q$ will be fixed once for all. As $V_+(\ol{F})=\cZ_k$, the
reduction $\overline{F}$ is equal to a multiple of $q^2$. Scaling $F$ by a
unit of $\cO$, we can suppose that
\begin{equation}
\label{eq:Fq}
\ol{F}=q^2.
\end{equation}
Given such an $F$, equation~\eqref{eq:Fq} implies that there exist a positive
integer $r>0$, a quadratic form $Q \in \cO[x_1,x_2,x_3]$ and a quartic form
$G\in \cO[x_1,x_2,x_3]$ such that
\begin{equation} \label{eq:F-Q2} F=Q^2+\pi^{r} G,
\end{equation}
with $\overline{Q}=q$ and $\overline{G}\ne 0$. Note that if $\hat{\cO}$ is the
completion of $\cO$, then $F$ is not a square in $\hat{\cO}[x_1,x_2,x_3]$
because $C$ is geometrically reduced. So the biggest possible power $\pi^r$
dividing $F-Q^2$ in $\cO[x_1,x_2,x_3]$ for all liftings $Q$ of $q$ exists.
\medskip
We endow the polynomial ring $\cO[x_1,x_2,x_3,y]$ with the grading
$\deg x_i=1$ and $\deg y=2$. The following lemma will be repeatedly used:
\begin{lemma}\label{normality-proj} Let $B$ be a graded $\cO$-algebra of the
following form
$$ B= \cO[x_1,x_2, x_3, y]/(Q-\pi^s y, y^2+\pi^{\epsilon} T), \quad T\in \cO[x_1,x_2,x_3]+\cO[x_1, x_2, x_3]y$$
with $s\ge 0$, $\epsilon\in \{0, 1\}$ and $2s+\epsilon>0$. Then the
following properties hold:
\begin{enumerate}[label=\emph{(\alph*)}]
\item $B$ is flat over $\cO$.
\item Suppose that $B\otimes K$ is normal. If $B\otimes k$ is reduced when
$\epsilon=0$ and $T\notin (\pi, Q, y)$ when $\epsilon=1$, then $B$ is
normal.
\item If furthermore, $F\in (y^2+\pi^{\epsilon}T)K^*$ in
$K[x_1, x_2, x_3,y]/(Q-\pi^s y)$, then $\Proj B\simeq \cC$.
\end{enumerate}
\end{lemma}
\begin{proof} (a) The flatness is equivalent to $B$ being
$\pi$-torsion-free. The latter follows from a straightforward computation.
(b) As $\dim B\otimes K=\dim B\otimes k=2$, $\Spec B$ is equidimensional. It
is then easy to see that $B$ is a local complete intersection
(\cite{liu-book}, Exercise 6.3.4(c)). By Serre's criterion
(\cite{liu-book}, Corollary 8.2.24), it is then enough to check that $B_\p$
is normal for height $1$ prime ideals $\p$ of $B$. This follows from the
hypothesis on $B\otimes K$ if $\pi\notin\p$.
Suppose that $\pi\in \p$. Then $\p$ is a minimal prime ideal over $\pi B$.
Let us show that $B_\p$ is a discrete valuation ring, or equivalently, that
$\p B_\p$ is principal. If $\epsilon=0$, as $B\otimes k$ is reduced,
$\p B_\p$ is generated by $\pi$. If $\epsilon=1$, then $\p=(\pi, y)$ is the
unique prime ideal over $\pi B$. This prime ideal is the image in $B$ of
the prime ideal $(\pi, Q, y)$ of $\cO[x_1, x_2, x_3,y]$. By hypothesis,
$T\notin \p$ in $B$, therefore $\p B_\p$ is generated by $y$.
(c) The canonical map $\cO[x_1, x_2, x_3] \to B$ is finite. The hypothesis
on $F$ easily implies that $F$ belongs to the ideal
$(Q-\pi^sy, y^2+\pi^{\epsilon}T)$ in $\cO[x_1, x_2, x_3, y]$. So we get a
canonical map $\phi : \cO[x_1,x_2,x_3]/(F)\to B$ which is an isomorphism
after tensoring with $K$, thus $\phi$ is finite and injective and the
corresponding morphism $\Proj B\to \cZ$ is finite and birational. We
conclude by Fact~\ref{fact}.
\end{proof}
\begin{remark}\label{rk:singular} When $C$ has irreducible stable (not
necessarily smooth) reduction over $K$, the situation is very similar to the
good reduction case. Let $\cC$ be the stable model of $C$ over $\cO$. Then
$\omega_{\cC/\cO}$ is still base-point free and induces a morphism
$$ f: \cC \to \P(H^0(\cC, \omega_{\cC/\cO}))\simeq \P^2_\cO. $$
The reduced image $\cZ=f(\cC)$ is a quartic and $f: \cC\to \cZ$ is a
birational finite morphism, equal to the normalization morphism. If
$\omega_{\cC/\cO}$ is very ample, then $\cC\simeq \cZ$ is a stable quartic
in $\P^2_\cO$. If $\omega_{\cC/\cO}$ is not very ample, then one can show
that $(\cZ_k)_{\mathrm{red}}$ is a smooth conic over $k$ and
$f_k : \cC_k \to (\cZ_k)_{\mathrm{red}}$ is a finite generically \'etale
morphism of degree $2$. The reduction is a singular hyperelliptic curve.
\end{remark}
Next we examine when the special fiber of $\Proj B$ is reduced or semi-stable
(in the Deligne-Mumford sense). A point in a scheme over $k$ is said to be
\emph{separable over $k$} if its residue field is a separable extension of
$k$.
\begin{lemma} \label{lm:reduced} Suppose $\caract(k)\ne 2$. Let
$g\in k[x_1, x_2, x_3]$ be a quartic form and let
$$W=\Proj (k[x_1,x_2,x_3,y]/(q, y^2+g))$$
with $\deg x_i=1$ and $\deg y=2$. Then the following properties are true:
\begin{enumerate}[label=\emph{(\alph*)}]
\item $W$ is reduced if and only if $g\not\equiv 0 \mod q$;
\item $W$ is semi-stable (respectively smooth) if and only if the curves
$q=0$ and $g=0$ in $\P^2_k$ intersect at separable points over $k$, with
multiplicity at most $2$ (respectively equal to $1$).
\end{enumerate}
\end{lemma}
\begin{proof} (a) The condition is clearly necessary. Suppose that the
condition is satisfied. Let us prove that $B_k:=k[x_1,x_2,x_3,y]/(q, y^2+g)$
is reduced. Let $A=k[x_1,x_2,x_3]/(q)$. This is an integrally closed domain
and we have $B_k=A\oplus Ay$. By hypothesis $g\ne 0$ in $A$. So
$\Frac(A)[y]/(y^2+g)$ is separable over $\Frac(A)$. Hence,
$$B_k=A\oplus Ay\subset \Frac(A)\oplus \Frac(A)y=\Frac(A)[y]/(y^2+g)$$ is reduced.
(b) The natural projection $\phi: W\to V_+(q)$ is finite and \'etale away
from the subscheme $D:=V_+(q, g)$. Let $w_0\in W$ be a point lying over some
point of $D$. Then $\phi$ is ramified at $w_0$. Let us show that $w_0$ is a
separable point over $k$. When $W$ is smooth at $w_0$, this is proved for
instance in \cite{LiLo}, Lemma 3.3. The case when $W$ is singular at $w_0$
is proved in \cite{liu-book}, Proposition 10.3.7(b).
For the rest of (b), we can suppose that $k$ is algebraically closed. A
local equation of $W$ at a point lying over an intersection point $z\in D$
is $y^2=x^ra$ where $r$ is the intersection number at $z$ and
$a\in \cO_{V_+(q), z}^{*}$. The condition for $W$ to be semi-stable or
smooth is then clear.
\end{proof}
\begin{lemma} \label{lm:reduced2} Let $\caract(k)=2$. Let
$h, g\in k[x_1, x_2, x_3]$ be respectively a quadratic and a quartic
forms. Let
$$B_k=k[x_1,x_2,x_3,y]/(q, y^2+hy+g)$$
and let $W=\Proj B_k$ with $\deg x_i=1$ and $\deg y=2$. Then the following
properties are true:
\begin{enumerate}[label=\emph{(\alph*)}]
\item $W$ is reduced if and only if $h$ is prime to $q$ or, $h$ is divisible
by $q$ and $g$ is not a square mod $q$;
\item if $h$ is divisible by $q$, then $W$ is not semi-stable;
\item if the curves $V_+(q)$ and $V_+(h)$ intersect transversely at
separable points of $\P^2_k$, then $W$ is semi-stable.
\end{enumerate}
\end{lemma}
\begin{proof}
(a) The condition is clearly necessary. Let us show the converse. Let $A$ be
defined as in the proof of Lemma~\ref{lm:reduced}(a). If $h$ is prime to
$q$, then the proof is similar to that of Lemma~\ref{lm:reduced}(a).
Suppose now that $h \equiv 0 \pmod{q}$. If $b=a_0+a_1y\ne 0$ is nilpotent in
$B_k$, then so is $b^2=a_0^2+a_1^2g\in A$. Therefore $a_0^2+a_1^2g=0$ in
$A$. As the latter is integrally closed, $a_0/a_1\in A$ and $g$ is a square
in $A$, contradiction.
(b) Denote by $\phi: W\to Z:=V_+(q)$ the natural projection. If $h$ is
divisible by $q$, then $\phi$ is purely inseparable, so the singularities of
$W$ are all cuspidal. On the other hand, $W$ is not smooth as its arithmetic
genus is $3$ while its geometric genus is $0$. So $W$ is not semi-stable.
(c) We can suppose $k$ is separably closed. The above cover $W\to V_+(q)$ is
\'etale away from $D$. So $W\setminus \phi^{-1}(D)$ is smooth. The local
equation of $W$ at a point lying over $z\in D$ is $y^2+tu(t)y+g(t)$ where
$t$ is a local parameter of $Z\simeq \P^1_k$ at $z$, with $u(0)\ne 0$. After
a translation of $y$ by some constant in $\ol{k}$, the local equation has
the shape $y^2+tu(t)y+t^rf(t)$ with $r\ge 1$ and $f(0)\ne 0$. If $r=1$, the
point is smooth, otherwise it is a node.
\end{proof}
\begin{remark}
The condition in (c) is not necessary. We can have inseparable
non-transverse intersection points with $W$ smooth.
\end{remark}
\begin{remark} The smoothness condition is less easy to express than in the
$\caract(k)\ne 2$ case. Keep the notation of the proof of Part (b). Let
$w_0\in W$. We know that $w_0$ is a smooth point if
$w_0\notin \phi^{-1}(D)$. Suppose that $\phi(w_0)=z_0\in D$. The Jacobian
criterion implies that $W$ is smooth at $w_0$ if and only if the matrix
$$
\left(
\begin{matrix}
\partial q/\partial x_1\hfill && \partial q/\partial x_2 \hfill
&& \partial
q/\partial x_3 \\
y_0\partial h/\partial x_1+\partial g/\partial x_1 && y_0\partial
h/\partial x_2+\partial g/\partial x_2 && y_0\partial h/\partial
x_3+\partial g/\partial x_3
\end{matrix}
\right),
$$
where $y_0\in \bar{k}$ is a square root of $g(z_0)$, has rank 2 at $z_0$.
\end{remark}
\subsection{Proof of Theorem~\ref{th:goodmodel}}
In this subsection we prove the following result.
\begin{theorem} \label{th:goodmodel-f} Suppose $\caract(k)\ne 2$ and $C$ has
good hyperelliptic reduction over $K$. Fix $q$ and a defining equation $F$
of $C$ as in formulas~\eqref{eq:conic} and \eqref{eq:Fq}. Then there exist
homogeneous polynomials $Q, \ G\in \cO[x_1,x_2,x_3]$ of degrees $2, 4$
respectively, and a positive integer $s$ such that
\begin{enumerate}[label=\emph{(\alph*)}]
\item $F=Q^2+\pi^{2s} G$ and $\ol{G}\ne 0$;
\item The smooth projective model $\cC$ of $C$ over $\cO$ is
$$\Proj \left(\cO[x_1, x_2, x_3, y]/(Q-\pi^s y, y^2+G) \right)$$
in the weighted projective space $\P^{(1,1,1,2)}_{\cO}$.
\item The conic $\ol{Q}=0$ and the quartic $\overline{G}=0$ in $\P^2_k$
intersect transversely at an effective divisor of degree $8$ over $k$;
\end{enumerate}
\end{theorem}
\begin{proof} Let $Q, G$ be as in equation~\eqref{eq:F-Q2} with the biggest
possible $r$. We claim that $\overline{G}$ is prime to $q$. Otherwise,
$G=QR+\pi G_1$ with $R, G_1$ homogeneous of degrees $2$ and $4$
respectively. Then the polynomial $F-(Q+\pi^rR/2)^2$ is divisible by
$\pi^{r+1}$. Contradiction with the assumption on $r$.
We now prove the theorem. Let $s=\lfloor r/2 \rfloor$ and
$\epsilon=r-2s\in \{0,1\}$. Let
$$\W=\Proj\left(\cO[x_1, x_2, x_3, y]/(Q-\pi^s y, y^2+\pi^{\epsilon} G)\right).$$
If $\epsilon=1$, as $G\notin (\pi, Q,y)$, Lemma~\ref{normality-proj} implies
that $\W$ is normal and isomorphic to $\cC$, hence smooth. This is absurd as
the special fiber of $\W$ is non-reduced. So $\epsilon=0$. The special
fiber $\mathcal W_k$ is reduced by Lemma~\ref{lm:reduced} (a). Again by
Lemma~\ref{normality-proj}(c), ${\W}\simeq \cC$. Part (c) is just
Lemma~\ref{lm:reduced}(b).
\end{proof}
\subsection{Good hyperelliptic reduction in $\caract(k)=2$}
When $\caract(k)=2$, \PHM s have singular close fibers. However, we still have
an analog result to Theorem~\ref{th:goodmodel} in this case.
\begin{theorem} \label{th:is2} Suppose that $\caract(k)=2$ and $C$ has good
hyperelliptic reduction over $K$. Fix $q$ and a definition equation $F$ of
$C$ as in formulas~\eqref{eq:conic} and \eqref{eq:Fq}. Then there exist
homogeneous polynomials
$$Q, H, G\in \cO[x_1,x_2,x_3]$$
of degrees $2, 2, 4$ respectively, $u\in 1+\pi \cO$ and a positive integer
$s$ such that
\begin{enumerate}[label=\emph{(\alph*)}]
\item $uF=Q^2+\pi^{s} QH+\pi^{2s}G$ and $\ol{H}$, $\ol{G}$ are prime to $q$;
\item The smooth projective model $\cC$ of $C$ over $\cO$ is
$$\Proj \left(\cO[x_1, x_2, x_3, y]/(Q-\pi^s y, y^2+Hy+G) \right)$$
in the weighted projective space $\P^{(1,1,1,2)}_{\cO}$.
\end{enumerate}
\end{theorem}
\begin{proof} In all the proof, capitalized italic letters denote homogeneous
polynomials in $\cO[x_1, x_2, x_3]$ of degree $2$ or $4$ (the exact degree
will be clear from the context). \medskip
{\bf Step 1.} {\sl Suppose we have a decomposition
$$F=Q^2+\pi^s QH+\pi^r P$$ with $1\le r< 2s$ (including $s=+\infty$, in which case $\pi^s$ stands for $0$). Then we have a new decomposition
$$F=Q_1^2+\pi^{s_1} Q_1H_1+ \pi^{r+1}P_1$$
with $Q_1\in Q+\pi^{[r/2]}\cO[x_1,x_2, x_3]$ and
$s_1\ge \min\{ s, r, v(2)+r/2\}$, where $v(2)$ is the normalized
$K$-valuation of $2$.}
Let $\ell=[r/2]$ and $\epsilon=r-2\ell$. Consider
$$ \W=\Proj\left(\cO[x_1, x_2, x_3,y]/(Q-\pi^{\ell}y, y^2+\pi^{s-\ell}Hy+\pi^{\epsilon}P)\right).$$
\underline{\bf Case 1}. Suppose that $\epsilon=1$ ($\ell\ge 0$). The
special fiber of $\W$ is not reduced (Lemma~\ref{lm:reduced2}(a) if
$\ell>0$), by Lemma~\ref{normality-proj} (b) and (c), we must have
$$P\in (\pi, Q_0, y)\cap \cO[x_1, x_2, x_3]=(\pi, Q)\cO[x_1,x_2, x_3].$$
Write $P=QH_0+\pi P_1$, then
$$F=Q^2+(\pi^s H+\pi^{r}H_0)Q+\pi^{r+1}P_1.$$
\underline{\bf Case 2}. \ Suppose that $\epsilon=0$. As $s>\ell$, $\W_k$ is
not semi-stable by Lemma~\ref{lm:reduced2}(b). Similarly to Case 1, this
implies that $\W_k$ is not reduced and $\ol{P}$ is a square mod $q$. Write
$P=R^2+QH_0+\pi P_0$. Let $Q_1=Q+\pi^{\ell}R$. Then
$$ F= Q_1^2+(\pi^sH+\pi^rH_0-2\pi^{\ell}R)Q_1+\pi^{r+1}P_1.$$
This achieves Step 1. \medskip
{\bf Step 2.} {\sl The polynomial $F$ admits a decomposition
$$uF=Q^2+\pi^sQH+\pi^{2s}G$$
with $\ol{H}$ prime to $q$ and $u\in 1+\pi\cO$. }
Let us first show that $F$ admits a decomposition
\begin{equation} \label{eq:qdh} F=Q^2+\pi^sQH+\pi^{2s}G.
\end{equation}
Suppose that this is not true. Starting with a relation~\eqref{eq:F-Q2} and
using repeatedly Step 1, we construct sequences
$Q_n, P_n, H_n\in \cO[x_1,x_2,x_3]$ and
$s_n\in {\mathbb N}^* \cup \{ +\infty \}$ such that
$$F=Q_n^2+\pi^{s_n}Q_nH_n+\pi^{r+n}P_n$$
and
$$Q_{n+1}-Q_n\in \pi^{[(r+n)/2]}\cO[x_1,x_2,x_3],\ s_n\ge \min\{s_1, r,
v(2)+r/2\}\text{ for all }n\ge 1\,.$$
The sequence $Q_n$ admits a limit $\hat{Q}\in \hat{\cO}[x_1, x_2,x_3]$. As
$\hat{Q}$ reduces mod $\pi$ to $q\ne 0$, this implies that
$Q_n+\pi^{s_n}H_n$ converges to some $\hat{R}\in\hat{\cO}[x_1,x_2, x_3]$
such that $F=\hat{Q}\hat{R}$. Contradiction because $C$ is geometrically
integral.
Now we prove the assertion of Step 2. Start with a
relation~\eqref{eq:qdh}. If $\ol{H}$ is prime to $q$ then we are
done. Otherwise, write $H=bQ+\pi H_0$ with $b\in \cO$. Then
$$F=(1+b\pi^s)Q^2+\pi^{s+1}QH_0+\pi^{2s}G.$$
Similarly to the above, applying repeatedly Step 1 to $(1+b\pi^{s})^{-1}F$,
we find
$$ (1+b\pi^{s})^{-1}F=Q_1^2+\pi^{s_1}Q_1H_1+\pi^{2s_1}G_1$$
with $s_1\ge s+1$, $Q_1-Q\in \pi^{s}\cO[x_1,x_2,x_3]$. If $\ol{H}_1$ is
prime to $q$, then we are done. Otherwise there exist $s_2\ge s_1+1$,
$b_1\in \cO$ such that
$$((1+b\pi^{s})(1+b_1\pi^{s_1}))^{-1}F=Q_2^2+\pi^{s_2}Q_2H_2+\pi^{2s_2}G_2.$$
Again by the geometric integrality of $C$, we see that this process must
stop after finitely many steps. \medskip
{\bf Step 3} Consider a relation given by Step 2. Let
$$ \W=\Proj\left(\cO[x_1, x_2, x_3,y]/(Q-\pi^{s}y, y^2+Hy+G)\right).$$
As $q$ is prime to $\ol{H}$, $\W_k$ is reduced by
Lemma~\ref{lm:reduced2}(a), hence $\W\simeq \cC$ by
Lemma~\ref{normality-proj}(b) and (c). This implies that $\ol{G}$ is prime
to $q$ because $\cC_k$ is irreducible.
\end{proof}
\begin{remark}\label{rk:singular2}Suppose that $C$ has stable irreducible
reduction over $K$. Then Theorems~\ref{th:goodmodel-f} and \ref{th:is2} hold
after replacing smooth by stable, and in \ref{th:goodmodel-f}(c), $V_+(q)$
and $V_+(\ol{G})$ intersect with multiplicities at most $2$ instead of $1$.
The proof is exactly the same by using Remark~\ref{rk:singular}.
\end{remark}
\begin{remark}
Let $\cO$ be a PID with field of fractions $K$ and $C : F=0$ be a smooth
projective plane quartic with good reduction everywhere over $K$. Gathering
the local information above, one can give a smooth model of $C$ over $\cO$
by
$$\cC = \Proj \left(\cO[x_1, x_2, x_3, y]/(Q-\delta y, uy^2+Hy+G) \right) $$
where $Q, H, G\in \cO[x_1, x_2, x_3]$ are of degrees $2, 2, 4$,
$\delta \in \Oc$ is divisible by a prime $\p$ if and only if $C$ has
hyperelliptic reduction at $\p$ and $u \in \Oc$ is congruent to $1$ modulo
the primes for which $C$ has hyperelliptic reduction.
\end{remark}
\section{Geometric Invariant Theory and characterization of potentially good quartic reduction}
\subsection{Some results on Geometry Invariant Theory (GIT)} \label{sec:appendix}
In this section we gather some definitions and results on the geometric
invariant theory over rings. The main
reference is \cite{Se}. \\
We let $S=\Spec R$ be a noetherian affine scheme, and $G$ be an affine smooth
group scheme over $S$ with connected and reductive geometric fibers (for
instance $\SL_{n,R}, \GL_{n,R}$). Let $V$ be a finite rank free $R$-module
endowed with an action of $G$:
$$ G\to \mathrm{Aut}_R(V) $$
(morphism of group schemes). Then $G$ acts ``linearly'' on the projective
space $\P(V^{\vee})$, where $V^{\vee}$ is the dual of $V$. Let $\Xb$ be a
$G$-stable closed subscheme of $\P(V^{\vee})$ of the form $\Xb=\Proj B$ where
$B$ is a $G$-stable homogeneous quotient of the symmetric algebra
$\mathrm{Sym}(V^{\vee})$ over $R$. The ring of $G$-invariant $B^G$ is
$$B^G=\{ b\in B \mid \sigma(b\otimes 1_A)=b\otimes 1_A \}$$
where $A$ runs through the $R$-algebras and $\sigma\in G(A)$. When $R$ is an
algebraically closed field $k$, the invariants of $B$ can be found using only
the $k$-rational points of $G$ and hence it corresponds to the usual
definition. Namely, denote by
$B^{G(k)}=\{ b\in B \mid g\cdot b=b, \ \forall g\in G(k)\}\supseteq B^G$, we
have the following result.
\begin{proposition} \label{punct-f} Let $k$ be an algebraically closed field
and let $B$ be a $k$-algebra endowed with the action of a smooth affine
algebraic group $G$. Then $B^{G}=B^{G(k)}$.
\end{proposition}
\begin{proof}
Let $b\in B^{G(k)}$. Let $A$ be a $k$-algebra. We have to show that
$b\otimes 1_A\in B\otimes_k A$ is invariant by all $\sigma\in G(A)$. Let
$h:=\sigma(b\otimes 1_A)-b\otimes 1_A \in B\otimes_k A$.
First suppose that $A$ is reduced and of finite type over $k$.
For any closed point $y\in \Spec A$
the image of $h$ by the canonical homomorphism
$$B\otimes_k A \to B\otimes_k k(y) = B\otimes_k k =B$$
is $\sigma(y)(b\otimes 1_{k(y)})-b\otimes 1_{k(y)}=\sigma(y)(b)-b=0$, where
$\sigma(y)\in G(k(y))=G(k)$ is the image of $\sigma$ by $G(A)\to G(k(y))$.
Let $\{ e_i\}_i$ be a basis of $B$ as $k$-vector space. Then $h=\sum_i
e_i\otimes a_i$ with $a_i\in A$ and we have $a_i(y)=0$ for all closed
points $y\in \Spec A$. Hence $a_i=0$ and $h=0$.
Now we consider the general case. Let $A$ be any $k$-algebra and let
$\sigma\in G(A)$. Let us denote by $\sigma^{\#}: A_0:=\cO_G(G)\to A$ the
corresponding $k$-algebra homomorphism. Let $\sigma_0\in G(A_0)$ be the
identity morphism. Then the map $G(A_0)\to G(A)$ induced by $\sigma^{\#}$
takes $\sigma_0$ to $\sigma$ and $h$ is the image of the element
$\sigma_0(b\otimes 1_{A_0})-b\otimes 1_{A_0}$ by
$\mathrm{Id}_B \otimes \sigma^{\#}$. Applying the previous case to the
reduced $k$-algebra of finite type $A_0$, we get $h=0$.
\end{proof}
\begin{corollary} \label{punct-f2} Let $R$ be an integral domain with field of
fractions $K$ and let $\bar{K}$ be an algebraic closure of $K$. Let $B$ be
a flat $R$-algebra endowed with the action of a smooth affine algebraic
group $G$ over $R$. Then
$$B^{G}=\{ b\in B \mid b\otimes 1_{\bar{K}}
\in (B\otimes_R \bar{K})^{G(\bar{K})} \}.$$
\end{corollary}
\begin{proof} Let $b\in B$. To check that $b\in B^G$, as in the proof of
Proposition~\ref{punct-f}, it is enough to check that
$\sigma(b\otimes 1_{A_0})=b\otimes 1_{A_0}$ in $B\otimes_R A_0$ where
$A_0=\cO_G(G)$. As $B$ and $A_0$ are flat over $R$, the morphism
$B\otimes_R A_0\to (B\otimes_R \bar{K})\otimes_{\bar{K}} (A_0\otimes_R
\bar{K})$
is injective. So it is enough to check the desired equality over $\bar{K}$.
This then follows from Proposition~\ref{punct-f}.
\end{proof}
Let $R=k$ be an algebraically closed field. A point $\xs \in \Xb(k)$ is
\emph{semi-stable} if there is a rational point $\hat{\xs}$ in the affine cone
$\hat{\Xb}$ of $\Xb$ lying over $\xs$ such that
$0\notin \overline{G(k)\cdot\hat{\xs}}$ (Zariski closure in $\hat{\Xb}$). It
is \emph{stable} if $G(k)\cdot \hat{\xs}$ is closed of dimension $\dim G$
(\ie~the stabilizer of $G$ at $\hat{\xs}$ is finite). When $\dim G>0$, stable
points are also semi-stable.
For arbitrary $R$, a geometric point $\xs\in \Xb(k)$, where $k$ an $R$-algebra
which is an algebraically closed field, is semi-stable (resp. stable) if it is
semi-stable (resp. stable) for the action of $G\times_S \Spec k$ on
$\Xb\times_S \Spec k$. Then $\xs$ is semi-stable if and only if there exists
a homogeneous $f\in B^{G}$ of positive degree such that $\xs\in D_+(f)$, see
\cite[Thm. 1]{Se}. As a consequence, the geometric points of the open subset
$\Xb^{\ds}:=\cup_f D_+(f)$ of $\Xb$, where $f$ runs through the homogeneous
elements of $B^G$ of positive degrees, are exactly the semi-stable geometric
points of $\Xb$.
\begin{theorem}[\cite{Se}, Theorem 4, p. 269 and Remarks 8, 9, p. 269,
271]\label{theorem4} Suppose that $R$ is excellent\footnote{See
[EGA], IV.7.8.2. Localizations of algebras of finite type over a field,
Dedekind domains of characteristic $0$, complete noetherian local rings
are excellent ([EGA], IV.7.8.3(ii)-(iii)). Excellent rings are universally
Japanese ([EGA], IV.7.8.3(vi)).}. Then we have the following properties.
\begin{enumerate}
\item \label{affine} The algebra $B^G$ is of finite type over $R$ and the
inclusion $B^G\subseteq B$ induces an affine surjective morphism
$$\phi : \Xb^{\ds}\to \Yb:=\Proj (B^{G}).$$
This morphism is a categorical quotient: any $G$-invariant morphism
$\Xb^{\ds}\to \Zb$ with trivial action on $\Zb$ factors uniquely through
$\Yb\to \Zb$.
\item \label{surj} If $\Zb$ is a $G$-stable closed subscheme of $\Xb$, then
$\phi(\Zb)$ is closed in $\Yb$.
\item\label{quot} For any pair of geometric points
$\xs_1, \xs_2\in \Xb^{\ds}(k)$, we have $\phi(\xs_1)=\phi(\xs_2)$ if and
only if
$$\overline{G(k)\cdot \xs_1}\cap\overline{G(k)\cdot \xs_2}\ne\emptyset.$$
\item There is an open subset $\Yb^s\subseteq \Yb$ such that the geometric
points of $\Xb^{\mathrm s}:=\phi^{-1}(\Yb^s)$ are exactly the stable
points. In particular, if $\xs_1, \xs_2\in \Xb^s(k)$ are two geometric
points, then $\phi(\xs_1)=\phi(\xs_2)$ if and only if
$\xs_2\in G(k)\cdot \xs_1$.
\end{enumerate}
\end{theorem}
\begin{remark}
In \cite[Appendice]{Bur}, some clarifications are added to the proof of
\cite[Prop.8]{Se}.
\end{remark}
The categorical quotient $\Yb$ is denoted by $\Xb^{\ds}\dq G$. Let
$\ys\in (\Xb^{\ds}\dq G)(k)$ be a geometric point. Then $\phi^{-1}(\ys)$ is a
union of closures of orbits. Any orbit of smallest dimension contained in
$\phi^{-1}(\ys)$ is closed, hence contained in (and equal to) the intersection
of all these closures. It is a \emph{minimal orbit}.
The following result appears in \cite[Prop.2.1]{shah} and
\cite[Lem.5.3]{mum-ens} in equal characteristics setting, in \cite[p.122]{Bur}
over the $p$-adics for $\Xb=\P(V)$ and in \cite{silverman-dyna} and
\cite{szpiro-semi} in a dynamical context.
\begin{corollary} \label{cor:shabur} Let $R$ be an excellent DVR with field of
fractions $K$. Let $\xs\in \Xb^s(K)$. Then there exists a generically finite
extension $R\subseteq R'$ of DVR's and an integral point
$\xfrak \in \Xb^{\ds}(R')$ such that $\xfrak_{K'}\in G(K')\cdot \xs$ (where
$K'=\mathrm{Frac}(R')$.) Moreover, we can ask the image of the closed point
of $\xfrak$ to belong to a minimal orbit.
\end{corollary}
\begin{proof}
Let $\ys=\phi(\xs)\in \Yb^s(K)$. Let
$\Gamma=\overline{\{ \ys \}}\subseteq \Yb$. As $\Yb/S$ is projective,
$\Gamma\to S$ is an isomorphism. Let $W_K$ be the closed subset
$\phi^{-1}(\Gamma)_K=\phi^{-1}(\ys)$ of $\Xb_K^{\ds}$ endowed with the
reduced subscheme structure. Let
$W=\overline{\phi^{-1}(\Gamma)_K}\subseteq \Xb^{\ds}$ be its
scheme-theoretical closure. It is integral hence flat over $S$. Therefore
it is a $G$-stable closed subscheme of $\Xb^{\ds}$. By Theorem
\ref{theorem4}(\ref{surj}), $\phi(W)\subseteq \Gamma$ is closed hence equal
to $\Gamma$. So $W\to S$ is surjective. Note that $W_K$ is a homogeneous
space under $G_K$, hence irreducible, then so is $W$. By \cite{liu-book},
Prop. 10.1.36, any closed point $w$ of the (non-empty) special fiber of
$W\to S$ belongs to an irreducible closed subscheme $\Gamma'\subseteq W$
quasi-finite and surjective over $S$.
Let $\Spec R'$ be the localization at a closed point of the normalization of
$\Gamma'$. Note that $R'$ is a DVR. The canonical morphism
$\xfrak: \Spec R'\to \Xb^{\ds}$ is an integral point as we are looking for:
its generic point belongs to the same geometric orbit as $\xs$, so
$\xfrak_{K'}\in G(K').\xs$ after a finite extension of $K'$ if
necessary. Moreover, the closure of the orbit of $w$ is contained in the
closed fiber $W_s$, so the minimal orbit is also contained in
$W_s$. Therefore we can chose $w$ in the minimal orbit.
\end{proof}
Now let us say a few words on the fibers of $\Xb^{\ds}\dq G\to S$. Let
$G, \Xb$ and $S$ (excellent) be as above. For any $T\to S$ with $T$ affine,
noetherian and excellent, we have $(\Xb\times_S T)^{\ds}=\Xb^{\ds}\times_S T$.
By the categorical quotient property, we have a canonical morphism
\begin{equation}
\label{eq:comp}
(\Xb^{\ds}\times_S T)\dq G_T \to (\Xb^{\ds}\dq G)\times_S T.
\end{equation}
As both sides are projective schemes over $T$ and the morphism induces a
bijection on the geometric points because Theorem~\ref{theorem4}(\ref{quot}),
it is a finite bijective morphism. This is an isomorphism if $T\to S$ is flat
\cite[p. 271]{Se}, or if $R$ is a field of characteristic $0$: \cite[Chap. 1,
Thm 1.1]{mumford-fogarty}. In particular, let $s\in S(k)$ be a geometric
point. Then we get a finite bijective, and hence homeomorphic, morphism of
projective schemes over $k$:
$$(\Xb_s)^{\ds}\dq G_s\to (\Xb^{\ds}\dq G)_s.$$
\begin{definition} \label{def:hsop} Let $C$ be a homogeneous algebra over a
ring $R$. Suppose the fibers of $\Proj C\to \Spec R$ all have the same
dimension $d\ge 0$. A \emph{homogeneous system of parameters (HSOP) over
$R$} of $C$ consists in $d+1$ homogeneous elements
$f_0, \dots, f_d\in C_+$ such that $V_+(f_0, \dots, f_d)=\emptyset$.
Equivalently, one has $C_+\subseteq \sqrt{(f_0, \dots, f_d)}$.
\end{definition}
\begin{remark}
Under the conditions of the definition, for any $R$-algebra $A$, the fibers
of the morphism $\Proj (C\otimes_R A)\to \Proj A$ have dimension $d$ and
$\{ f_0\otimes 1, \dots, f_d\otimes 1\}$ is a HSOP of $C\otimes_R A$.
Conversely, for $f_0, \dots, f_d$ to be a HSOP of $C$, it is enough that
their images in $C \otimes_R k(t)$ form a HSOP for all points
$t\in \Spec R$. If $R=\O$ is a DVR, by properness, a list of homogeneous
elements of $C$ is a HSOP over $\O$ if and only if their images in
$C \otimes_{\O} k$ form a HSOP over the residue field $k$.
\end{remark}
\begin{remark} Let $B$ a be homogeneous algebra over a field $k$, let $G$ be a
reductive algebraic group over $k$ acting on $B$. Let
$f_0, \dots, f_d\in B^G=C$ be homogeneous. Then they are a HSOP over $k$ of
$C$ if in $\Spec B$, we have $V(f_0, \dots, f_d)=V(\m B)$ (the null-cone)
where $\m=C_+$ is the irrelevant maximal ideal of $C$. Moreover $C$ will be
a finite algebra over $k[f_0, \dots, f_d]$, see \cite[Lemma 2.4.5.]{DerKem}.
\end{remark}
\begin{remark} \label{rem:hsop} Let $B$ be a homogeneous algebra over an
excellent DVR $\O$, let $G$ be an affine group scheme over $\Spec \O$ acting
on $\Proj B$ as at the beginning of the section. Using the two previous
remarks, Proposition~\ref{punct-f} and the bijection \eqref{eq:comp}, we see
that a list of homogeneous elements of $B^G$ is a HSOP over $\O$ of $B^G$ if
and only if their images in $B \otimes_{\O} \bar{k}$ define the zero set
$V((B \otimes_{\O} \bar{k})^{G(\bar{k})}_+ (B \otimes_{\O} \bar{k}))$ or
equivalently they form a HSOP of $(B\otimes_{\cO} \bar{k})^{G(\bar{k})}$
over $\bar{k}$. We can then use classical methods of GIT over a field to
prove that a given list of homogeneous elements of $B^G$ is a HSOP over $\O$
for $B^G$.
\end{remark}
A natural question is when a HSOP exists. A \emph{pictorsion ring}
(\cite{GLL}, Definition 0.3) is a commutative ring $R$ such that for any
finite homomorphism $R\to A$, the group $\mathrm{Pic}(A)$ is of torsion. The
ring of integers of a finite extension of $\mathbb Q$, the ring of regular
functions of an affine connected regular curve over a finite field, and
semi-local rings are pictorsion.
\begin{proposition} \label{hsop} Let $R$ be a noetherian pictorsion ring. Let
$C$ be a homogeneous $R$-algebra such that the fibers of
$\Proj C\to \Spec R$ all have the same dimension $d\ge 0$. Then $C$ has a
HSOP over $R$.
\end{proposition}
\begin{proof} This is done in the proof of Theorem 8.1 in \cite{GLL}.
\end{proof}
Now we return to the early situation with $R$ excellent and pictorsion, and
$\Xb=\Proj B$ has linear $G$-action.
\begin{corollary} \label{cor:hsop} Suppose that $S=\Spec R$ is connected,
$\Xb$ is flat over $S$ and that $\Xb_s$ is irreducible and has a stable
point for all $s\in S$. Then $B^G$ admits a HSOP over $R$.
\end{corollary}
\begin{proof} All fibers of $\Xb/S$ (resp. of $G/S$) have the same dimension
$n$ (resp. $m$). For all $s\in S$, as $\Xb_s$ is irreducible, its non-empty
open subsets $(\Xb_s)^{\ds}$ and $(\Xb_s)^{\mathrm{s}}$ are irreducible of
dimension $n$. Therefore $(\Xb_s)^{\mathrm{s}}\dq G_s$ has dimension
$\dim (\Xb_s)^{\mathrm{s}} -\dim G_s=n-m$. Let
$\Yb=\Xb^{\ds}\dq G=\Proj (B^G)$. For all $s\in S$, the fiber $\Yb_s$, which
is homeomorphic to $(\Xb_s)^{\ds}\dq G_s$, is irreducible and contains an
open subset of dimension $n-m$. So $\dim \Yb_s=n-m$ for all $s\in S$. Now
we can apply Proposition~\ref{hsop}.
\end{proof}
If $B^G$ admits a HSOP $I_0,\ldots,I_d$ over $R$, a geometric point
$\xs \in \Xb(k)$ is not semi-stable if and only if
$\xs \in V(I_0,\ldots,I_d)_k $. We then have the following result.
\begin{proposition} \label{prop:ssred} Let $\O$ be an excellent DVR with field
of fractions $K$ and $\Xb=\Proj B$ such that $B^G$ admits a HSOP over
$\O$. Let $\Iv$ be a list of homogeneous elements of $(B^G)_+$ of degree
$\dv$ containing the given HSOP and $\xs \in \Xb^s(K)$. Then after a
possible extension, there exists $\xfrak \in \Xb^{ss}(\O)$ such that
$\xfrak_K \in G(K) \cdot \xs$ and $\Iv(\xfrak)$ is a minimal representative
of $\Iv(\xs)$.
\end{proposition}
Let $R$ be a local ring. Let $d>2,\, n \geq 1$ be integers and
$N={n+d \choose d}$. Let $\Xb \simeq \P^{N}=\Proj R[\xv]$, the space of
hypersurfaces of degree $d$ in $\P^n$ under the classical action of
$G=\SL_{{n+1},R}$. An $X \in \Xb(R)$ is represented by an $(n\!+\!1)$-ary
form $F$, unique up to multiplication by an invertible element of $R$.
Moreover, there exists an invariant $D_{\ell} \in R[\xv]^G$ of degree
$\ell=(n+1)(d-1)^n$, called the \emph{discriminant}, such that the
hypersurface $F=0$ is smooth over $R$ if and only if $D_{\ell}(F)$ is
invertible in $R$ \cite[p.426]{gelfand}, \cite{demazure}.
\begin{corollary} \label{cor:disc0} Let $\O$ be an excellent DVR with field of
fractions $K$. Let $\Iv$ be a list of invariants over $\O$ containing a HSOP
over $\O$. If a hypersurface $X/K : F=0$ is such that
$v_{I}(D_{\ell}(F))=0$, then, after a possible finite extension of $K$,
there exists a hypersurface $\mathcal{X}/\O$ such that
$\mathcal{X}_K \simeq X$ and $\mathcal{X}_k$ is smooth.\end{corollary}
\subsection{Characterization of potentially good quartic
reduction} \label{sec:Shah}
We can now apply the general results of the previous section to get a proof of
Theorem~\ref{th:GNHR}.
\begin{theorem}[See Theorem~\ref{th:GNHR}] \label{th:gnhr} Let $\Iv$ be a list
of invariants containing a HSOP over $\O$ for ternary quartic forms under
the action of $\SL_{3,\O}$ (when $p \geq 7$, one can take $\Iv=\DOv$ by
Theorem~\ref{th:domodp}). A smooth plane quartic $C/K : F=0$ has potentially
good quartic reduction if and only if $v_{I}(D_{27}(F))=0$.
\end{theorem}
\begin{proof}
Since the reduction type of the stable model does not change by completion
\cite[Prop. 10.3.15]{liu-book} and neither does the assumption
$v_I(D_{27}(F))=0$, we can assume that $\O$ is complete, hence
excellent. Let $C/K : F=0$ be a smooth quartic such that
$v_{I}(D_{27}(F))=0$. Then, Corollary~\ref{cor:disc0} shows that
$\mathcal{C}$ is a quartic with good reduction so $C$ has potentially good
quartic reduction. Conversely, if $C$ has potentially good quartic
reduction, then there is a model $\mathcal{C}$ which is smooth over $\O$ and
the discussion at the beginning of Section~\ref{general-facts} shows that
$\mathcal{C}$ is a smooth quartic $F_0=0$. In particular
$v(D_{27}(F_0)) = 0$, and therefore $v_{I}(D_{27}(F)) = v(D_{27}(F_0))=0$.
\end{proof}
\begin{remark}
It is unclear how to effectively compute the stable model $\mathcal{C}$. In
some examples, the extension of $K$ to be considered can be huge. For
instance, $y^3=x^4-x$ has good quartic reduction over an extension of
minimal degree $108$ of $\Q_3$ (Bouw-Wewers, private communication).
\end{remark}
\begin{remark}\label{rmk:nhspecial} With the same assumptions as
Theorem~\ref{th:gnhr}, the invariants of the special fiber $\mathcal{C}_k$
are the reduction modulo $\pi$ of a minimal representative of
$\Iv(F)$. Hence, if an algorithm to reconstruct from the invariants is
available, one can get an equation for $ \mathcal{C}_k$. This is in
particular the case for generic plane quartics in characteristic $p>7$ when
$\DOv \subset \Iv$ using \cite{LRS16}.
\end{remark}
\begin{example} \label{ex:cm} Let us consider the curves $X_i$ in
\cite[Sec.5]{KLLRSS17} with the notation from this article. Using
Theorem~\ref{th:GNHR}, it is easy to see that for all primes $p>7$ dividing
$\Disc X_i$ and not dividing the corresponding $I_{27}^{\min}$, the curves
$X_i$ have potentially good quartic reduction. On the contrary, this is
never the case for $p \leq 7$.
\end{example}
We can also apply Corollary~\ref{cor:disc0} to binary octic forms when
$p \ne 2$. We have shown in \cite[Prop.1.9]{LR11} that the set of Shioda
invariants $j_2,\ldots,j_{10}$, contains a HSOP over $\O$ of the invariant
ring of binary octics when $p=0$ and $p>7$. Actually, taking into account that
the vanishing of $j_2,\ldots,j_7$ implies the vanishing of the discriminant
$D_{14}$, the proof of loc. cit. shows that
$\operatorname{\underline{Sh}}=(j_2,\ldots,j_7)$ is a HSOP over $\O$ when
$p>7$. Basson extends in his PhD~\cite{basson} these results to characteristic
$p=7$ and $3$ and exhibits several possible HSOP over $\O$, for instance one
of degree $(3,\, 4,\, 5,\, 6,\, 10,\, 14)$ for $p=7$ and one of degree
$(4,\, 5,\, 6,\, 7,\, 8,\, 9)$ for $p=3$. Similar techniques for $p=5$ lead
to a HSOP over $\O$ of degree $(4,\, 6,\, 6,\, 12,\, 14,\, 20)$. We will need
the following result only in characteristic $p>7$ or $0$.
\begin{corollary}\label{cor:Shiodas}
Let $C:\,y^2=f(x)$ over $K$ be a hyperelliptic curve of genus $3$ with
$f\in\mathcal{O}[x]$. Assume that $p \ne 2$ then $C$ has potentially good
reduction if and only if $v_{\operatorname{Sh}}(D_{14}(f))=0$, with
$\operatorname{\underline{Sh}}=(j_2,...,j_{7})$.
\end{corollary}
\subsection{The case of Picard curves}
As an illustration, let us look at the well-studied family of Picard
curves~\cite{BBW17, Hol, KLS18, kowe, LaSo}. We can easily know whether the
curve has potentially good reduction in this case. Indeed, since a Picard
curve $C/K$ cannot be hyperelliptic (see for instance \cite[p.13]{harris}),
the reduction of the stable model of $C$ is either a non-hyperelliptic
(Picard) curve or singular. If $C$ has potentially good (quartic) reduction,
the beginning of Section~\ref{general-facts} implies that its stable model
$\Cc_0/\O$ is again given by a plane quartic. Assuming that $\O$ contains a
third root of unity, we extend to $\O$ the cyclic covering to $\P^1/K$, and we
can even assume that $\Cc_0 : F_0=0$ where
$F_0=-x_2^3x_3+x_1^4+ax_1^2x_3^2+bx_1x_3^3+cx_3^4$, with $a,b,c \in \O$. This
implies the result conjectured in \cite[Remark~1.6]{WINE1-2} and generalizes
\cite[Lemma~7.13]{Hol}.
Actually, we can go further %and reprove this result
and give an easy characterization of the type, by working through the
necessary and sufficient condition $v_{\DO}(I_{27}(F_0))=0$. %
More precisely let
$$q_2={a}^{2}+12\,c \quad \textrm{and} \quad
q_3=2\,{a}^{3}+72\,ac-27\,{b}^{2}$$
be the the primary invariants of a quartic binary form
$x^4+a x^2 z^2+b x z^3 + c z^4$. Its discriminant $D_6$ satisfies the
relation $3^3\,D_6 = 4\,q_2^3 - q_3^2$. Working over the residue fields (see
Definition~\ref{def:hsop}), quick computations show that $(q_2,\ D_6)$ is a
HSOP over $\O$ for the ring of invariants of quartic binary forms under the
action of $\SL_{2,\O}$.
\begin{theorem}\label{ex:picard}
Let $\Cc_0 : F_0=0$ with
$F_0=- x_2^3x_3+x_1^4+ax_1^2x_3^2+bx_1x_3^3+cx_3^4= 0$ be a Picard curve
with $a,b,c\in\mathcal{O}$. The curve has potentially good (quartic)
reduction if and only if
\begin{equation}\label{eq:valuations}
6\,v(\,2^3 \cdot 3\ a\,) \geq v(\,D_6\,) \text{ and }
3\,v(\,q_2\,) \geq v(\,3^3\,D_6\,) \text{ and, when} \; p=3, \;
3\,v(\,a\,{q_3}^3\,) \geq 5\,v(\,3^3\,D_6\,).
\end{equation}
\end{theorem}
\begin{proof}
Let us assume that the characteristic $p$ is neither $2$ nor $3$. We have
the following expressions for the Dixmier-Ohno invariants of $F_0$\,:
\begin{displaymath}
\begin{array}{lcl}
I_9 & =& 2^{-12} \cdot 3^{-4}\ (8\,a\,{q_3}+81\,{{q_2}}^{2}),\\
J_9 & =& 2^{-12} \cdot 3^{-4}\ (16\,a\,{q_3}+27\,{{q_2}}^{2}),\\
I_{18} &=& 2^{-23} \cdot 3^{-6}\ (108\,{a}^{2}\,{{q_2}}^{3}+33\,a\,{q_3}\,{{q_2}}^{2}+8\,{{q_2}}^{4}-54\,{D_6}\,{q_2}),\\
J_{18} &=& 2^{-23} \cdot 3^{-7}\ (36\,{a}^{2}\,{{q_2}}^{3}+51\,a\,{q_3}\,{{q_2}}^{2}+16\,{{q_2}}^{4}-108\,{D_6}\,{q_2}),\\
I_{27} &=& 2^{-40} \cdot 3^9 \ \ {D_6}^2\,.
\end{array}
\end{displaymath}
Since $(q_2,\ D_6)$ is a HSOP, the condition
$3\,v(\,q_2\,) \geq v(\,3^3\,D_6\,)$ implies that
${x}^4+a\, {x}^2{z}^2+b\, {x}{z}^3 + c\,{z}^4$ has potentially good
reduction, and consequently we have $2\,v(\,q_3\,) \geq v(\,3^3\,D_6\,)$
too. We make use of these inequalities together with the
hypotheses~\eqref{eq:valuations} to establish lower bounds for the
valuations of the invariants in the HSOPs over $\O$ of ternary quartic forms
(Theorem~\ref{th:domodp}), and thus prove the reverse implication applying
Theorem~\ref{th:gnhr}.
For the direct implication, we first remark that ${q_2}^2$ is an invariant
of ternary forms, it is equal to $2^{12}\cdot3\cdot5^{-1}\,(2\,I_9-J_9)$.
Since $3^3\,D_6 = 4\,q_2^3 - q_3^2$\, comparing its valuation with $I_{27}$
yields $3\,v(\,q_2\,) \geq v(\,3^3\,D_6\,)$ and
$2\,v(\,q_3\,) \geq v(\,3^3\,D_6\,)$. Moreover, at least, one the these two
inequalities is an equality.
If $2\,v(\,q_3\,) = v(\,3^3\,D_6\,)$, then
$v(I_9-3\,J_9)/9 \geq v(I_{27})/27$ implies
$6\,v(a\,q_3) \geq 4\,v(3^3\,D_6)$, or equivalently
$6\,v(a) \geq 4\,v(3^3\,D_6) - 6v(q_3) = v(3^3\,D_6)$\,. If
$2\,v(\,q_3\,) \ne v(\,3^3\,D_6\,)$, we have
$3\,v(\,q_2\,) = v(\,3^3\,D_6\,)$. Considering the equality
\begin{displaymath}
2^{21} \cdot 3^4 \cdot 5^{-1} \cdot (2\,I_{18}-3\,J_{18} +
(I_{9}-3\,J_9)\,(2\,I_9-J_9)/20 ) = a^2\,q_2^3\,.
\end{displaymath}
We have $3\,v(a^2\,q_2^3) \geq 2\,v(I_{27})$, and thus
$6\,v(a) \geq v(3^3\,D_6)$ too.\\
In characteristic $2$ or $3$, we can make similar computations. We omit
these cumbersome details.
\end{proof}
A closer look at the equality $q_2 = {a}^{2}+12\,c$ yields the following
corollary.
\begin{corollary}
In characteristic $p \ne 2$ or $3$, the curve $\Cc_0$ has potentially good
reduction if and only if
\begin{displaymath}
\min(\,6\,v(a),\ 3\,v(c)\,) \geq v(D_6)\,.
\end{displaymath}
A stable model $\Cc$ is given by
\begin{math}
\mathcal{C}:\,-x_2^3x_3+x_1^4+a/\mathfrak{p}^6\,x_1^2x_3^2+b/\mathfrak{p}^9\,x_1x_3^3+c/\mathfrak{p}^{12}\,x_3^4=
0\,,
\end{math}
where $\mathfrak{p}=\pi^{v(D_6)/36}$\,, and the map
$\mathcal{C}\rightarrow\mathcal{C}_0:\,(x_1:x_2:x_3)\mapsto(\mathfrak{p}^3x_1:\mathfrak{p}^4x_2:x_3)$.
%
\end{corollary}
\begin{remark} In the on-going work \cite{BKKSW}, the authors describe the
reduction of the stable model of Picard curves in any characteristics,
including the more difficult $p=2,3$ cases.
\end{remark}
\section{Characterization of potentially good hyperelliptic reduction}
\subsection{Characterization of \PHM{} in terms of
invariants} \label{sec:comphyp}
From now on we restrict ourselves to a DVR $\O$ with residue field $k$ of
characteristic $p=0$ or $p>7$. We start with the following observation.
\begin{lemma} \label{lem:test-conic} Let $Q \in \O[x_1,x_2,x_3]$ be a
quadratic ternary. Then the Dixmier-Ohno invariants of $Q^2$ are
\begin{multline*}
\DOv(Q^2)= \left(I_3 , \frac{1}{2^2\cdot3^2\cdot5} I_3^2 ,
\frac{7^2}{2^2\cdot3^2} I_3^3 , \frac{7^2}{2^2\cdot3\cdot5} I_3^3 ,
\frac{7^3}{2^2\cdot3^4\cdot5} I_3^4 , \frac{7^2}{2^2\cdot3^2} I_3^4 ,
\frac{5\cdot7^3}{2^4\cdot3^5} I_3^5 , \right.
\\
\left. \frac{7^3}{2^4\cdot3^2\cdot5^2} I_3^5 , \frac{7^4}{2^4\cdot3^5}
I_3^6 , \frac{7^4}{2^4\cdot3^3\cdot5^2} I_3^6 ,
\frac{7^4}{2^2\cdot3^4\cdot5} I_3^7, \frac{7^4}{2^4\cdot3^2\cdot5} I_3^7
, 0\right)
\end{multline*}
where $I_3 = 5/(2^2\cdot3^2) \cdot D_3(Q)^2$ where $D_3$ is the discriminant
of ternary quadratic forms.
\end{lemma}
In particular, if $Q$ is non-degenerate, we have
\begin{small}
\begin{equation}\label{eq:1}
\DO(Q^2) = \left(1: \frac{1}{180}: \frac{49}{36}: \frac{49}{60}: \frac{343}{1620}: \frac{49}{36}: \frac{1715}{3888}:
\frac{343}{3600}: \frac{2401}{3888}: \frac{2401}{10800}: \frac{343}{1620}:
\frac{2401}{720}: 0\right).
\end{equation}
\end{small}
\begin{remark} \label{rem:kollar} As mentioned in the introduction, it appears
likely that the methods from \cite{kollar} and forthcoming
\cite{elsenhans-stoll} will allow to determine the reduction type over $K$
of $F=0$ (at least for any DVR $\O$ for which one can solve polynomial
equations over $k$). Indeed, their algorithms compute the finite list of
$\PGL_3(\O)$-equivalent classes of forms over $\O$ which are
$\PGL_3(K)$-equivalent to $F$ and with a discriminant of minimal
valuation. Hence, one has good quartic reduction if and only if there is a
form in the list with a discriminant of valuation $0$ (in this case there is
actually a unique class). One has good hyperelliptic reduction if and only
if at least one form $F_0$ in the list satisfies the conditions of
Proposition~\ref{2->1}. Indeed, in this case the discriminant is not of
valuation $0$ but $I_3(F_0)$ is, as we have just seen\footnote{Using the
HSOPs of Theorem~\ref{th:domodp}, the same property holds with the 6th
invariant in characteristic $2$, the first in characteristic $3$ and the
fifth in characteristic $5$.}. This means that the valuation of the
discriminant is minimal for a form having a \PHM{}. Unlike the case of good
quartic reduction, the list may contain several forms having either bad or
good hyperelliptic reduction. For instance $F/\Q_{11}=0$ where
$F=(x_2^2-4 x_1 x_3)^2 + \pi^4 (x_1^4+x_3^4)$ has good hyperelliptic
reduction whereas $F(\pi^2 x_1,\pi x_2,x_3)/\pi^4=0$ has bad reduction and
both have a discriminant with minimal valuation.
\end{remark}
\begin{proposition} \label{prop:MHR} The curve $C$ potentially admits a \PHM{}
if and only if any minimal representative of $\DO(F)$ reduces modulo $\pi$
to the invariants in equation~\eqref{eq:1}.
\end{proposition}
\begin{proof}
The direct implication is clear. For the converse, using
Proposition~\ref{prop:ssred}, and after a possible extension, we can assume
that we start with a model $C :F=0$ with $F \in \O[x_1,x_2,x_3]$. Moreover,
we can assume that $\bar{F}=0$ is a semi-stable (in the GIT sense) quartic
with minimal orbit. Since $\bar{F}=0$ has the same invariants as the square
of a non-degenerate quadric, which has an infinite automorphism group, it is
not stable in the GIT sense. The quartic $\bar{F}=0$ is therefore in
$X^{ss} \setminus X^s$. By \cite[Lem.1.4.iii]{artebani}, after possibly a
finite extension of $K$, $\bar{F}$ is either equivalent to the square of the
non-degenerate quadratic form $Q_0=(x_2^2-4x_1 x_3)$ or to the product of
$Q_0$ with a quadratic form $Q$ such that $Q_0=0$ is tangent to $Q=0$ (the
tangent line being, say, $x_1=0$). In the first case, this means exactly
that $F$ is a \PHM{}. In the second case, we have
$$\bar{F}=(x_2^2- 4 x_1 x_3) (a x_1^2+b x_1 x_2+c x_1 x_3+d x_2^2) \text{
with } (a,b,c,d) \in k^4.$$
Moreover, since $\DO(\bar{F})=\DO(Q_0^2)$, a computation shows that
$c \ne 0$ and $c=-4d$. Hence we can rewrite (up to a multiplicative
constant) $\bar{F}=Q_0^2 + Q_0 x_1 (ax_1+bx_2)$. We therefore see that
$$F=Q_0^2 + Q_0 x_1 (ax_1+bx_2) + \pi G$$ with $G \in \Oc[x_1,x_2,x_3]$.
Finally the change of variables
$(x_1,\,x_2,\,x_3) \mapsto (\pi^{1/4}\,x_1\,,\,\pi^{1/8}\,x_2\,,\, x_3),$
transforms $F$ into
\begin{displaymath}
\pi^{1/2} \cdot \left(Q_0^2 + \pi^{1/8} \underbrace{\left(Q_0 x_1 (a \pi^{1/8} x_1+b x_2) + \pi^{3/8} G(\pi^{1/4} x_1,\pi^{1/8} x_2,x_3)\right)}_{G_0}\right).
\end{displaymath}
Hence, over an extension of $K$, we find the required model
$F_0= Q_0^2 + \pi^{1/8} G_0$.
\end{proof}
Since having potentially good hyperelliptic reduction implies the existence of
a \PHM{}, Proposition~\ref{prop:MHR} is often strong enough to rule out this
scenario, as the following example shows.
\begin{example}
Let $K=\Q_{11}$ and consider the case $X_{12}/K : F=0$ from
\cite[Sec.5]{KLLRSS17}. The valuations of $\DOv(F)$ are
$(0, 0, 0, 1, 0, 0, 2, 0, 1, 0, 0, 0, 9 )$ and hence $\DOv(F)$ is a minimal
representative. Since $v_{\DO}(J_9(F))=v(J_9(F))>0$, the list $\DOv(F)$
cannot reduce modulo $\pi$ to equation~\eqref{eq:1}. Therefore $X_{12}$ has
geometrically bad reduction.
\end{example}
\subsection{The equivariant map $b_8$ and proof of Theorem~\ref{th:main}}
\label{sec:char-hyper-reduct}\label{sec:dixm-ohno-invar1}
The equivalence in Theorem~\ref{th:main} is more complicated to establish than
the one in Theorem~\ref{th:GNHR} as we must link the world of genus 3
non-hyperelliptic curves (as ternary quartic forms) and hyperelliptic curves
(as binary octics). The starting point is a beautiful result from
representation theory we will recall now.
Let $R$ be an integral domain and let $W$ (resp. $V$) be free $R$-modules of
rank $n=2$ (resp. $n=3$). We identify binary (resp. ternary) forms with
coefficients in $R$ of degree $d \geq 0$ with elements in $\Sym^d(W^{\vee})$
(resp. $\Sym^d(V^{\vee})$). The usual action of $\GL_n(R)$ on the left for $W$
(resp. $V$) induces a left action on $\Sym^d(W)$ (resp. $\Sym^d(V)$) and a
right action on the forms, which we denote in both cases with a dot. An
identification of $V$ with $\Sym^2(W)$ defines a morphism
$h : \SL(W) \to \SL(V)$ and a linear morphism
$b_{2d} : \Sym^d(V^{\vee}) \to \Sym^{2d}(W^{\vee})$ which is equivariant for
the previous action, \ie~$F.h(T) = b_{2d}(F).T$ for any $T \in \SL(W)$. The
identification of $V$ with $\Sym^2(W)$ also induces an embedding of $\P^1$
into $\P^2$ whose image is given by a conic $Q_0=0$ with $Q_0$ a quadratic
form with coefficients in $R$. The image of the morphism $h$ is contained in
the subgroup $\SO(Q_0)$ of elements $T \in \SL_3(R)$ such that $Q_0.T=Q_0$.
To make it concrete and keeping with the literature, when the characteristic
of $R$ is not equal to $2$, we choose the following particular basis for
$\Sym^2(W)$. Let $u,v$ (resp. $v_1,v_2,v_3$) be a basis of $W$ (resp. $V$)
and $x,z$ the dual basis of $W^{\vee}$ (resp. $x_1,x_2,x_3$ the dual basis of
$V^{\vee}$). We let $u^2,2uv,v^2$ be a basis for $\Sym^2(W)$. Then
$b_{2d}(F)=F(x^2,2xz,z^2)$, $Q_0=x_2^2-4 x_1 x_3$ and
\begin{equation}\label{eq:GL2ToGL3}
h
\begin{pmatrix}
a & b \\
c & d
\end{pmatrix}
=
\begin{pmatrix}
a^2 & 2 a b & b^2 \\
a c & a d + b c & b d \\
c^2 & 2 c d & d^2
\end{pmatrix}.
\end{equation}
Particularizing to the case of a quartic form $F$, a simple computation shows
that $b_8(F)$ contains valuable information on the reduction.
\begin{lemma} \label{lem:fond} Let $\Cc/\O : F=0$ with $F=Q_0^2 + \pi^s G$ be
a \PHM. The model $\Cc$ is a good \PHM{} if and only if
$\overline{b_8(G)}=\overline{G}(x^2,2xz,z^2)$ has 8 distinct roots. Moreover
an equation of $\bar{\Cc}$ is $y^2=\overline{b_8(G)}$.
\end{lemma}
Let $F = Q_0^2 + \pi^s G$ be a \PHM{} for a smooth plane quartic $C$. The
article \cite{LRS16} gives relations between the Shioda invariants of $b_8(F)$
and the Dixmier-Ohno invariants of $F$. These expressions can be shorten
considerably in our situation by looking at the first terms of their
$\pi$-expansions.
\begin{proposition} \label{prop:shortrel} Let $F = Q_0^2+ \pi^s G$ be a
\PHM. Let $f=b_8(G)=b_8(F)/\pi^s$. The Shioda invariants $j_i$ for
$2 \leq i \leq 7$ satisfy the congruences
$\pi^{s i} \cdot j_i(f) = \iota_{3 i}(F) + O(\pi^{s(i +1)})$ where
\begin{equation}\label{eq:DOshort}
\parbox{0.9\textwidth}{
\begin{dgroup*}[style={\footnotesize}]
%
\begin{dmath*}
\iota_6 = \frac{3^2}{2^5 \cdot 5 \cdot 7}\,%
({I_{3}}^{2}-180\,I_{6})\,,
\end{dmath*}
%
\begin{dmath*}
\iota_9 = \frac{3^5}{2^9 \cdot 5^2 \cdot 7^{3}}\,%
(14\,{I_{3}}^{3}-2520\,I_{3}\,I_{6}-81\,I_{9}+135\,J_{9})\,,
\end{dmath*}
%
\begin{dmath*}
\iota_{12} = \frac{ 3^3 }{ 2^{14} \cdot 5 \cdot 7^{3}}\,%
I_3\,(-32\,{I_{3}}^{3}+14580\,I_{3}\,I_{6}+261\,I_{9}-495\,J_{9})\,%
+ \frac{5^2}{2 \cdot 3 \cdot 7^{2}}\ \iota_6^2\,,
\end{dmath*}
%
\begin{dmath*}
\iota_{15} = \frac{ 3^4 }{ 2^{16} \cdot 5^2 \cdot 7^{5}}\,%
I_3\,(-592\,{I_{3}}^{4}+30780\,{I_{3}}^{2}I_{6}+2601\,I_{3}\,I_{9}-45\,I_{3}\,J_{9}+7290000\,{I_{6}}^{2}-2430\,J_{12})\,%
+ \frac{5^2}{3^2 \cdot 7}\ \iota_6\,\iota_9\,,
\end{dmath*}
%
\begin{dmath*}
\iota_{18} = \frac{ 3^8 }{ 2^{24} \cdot 5^2 \cdot 7^4 }\,%
I_3^2\,(-8\,{I_{3}}^{4}-14418\,{I_{3}}^{2}I_{6}-117\,I_{3}\,I_{9}+423\,I_{3}\,J_{9}+155520\,{I_{6}}^{2}-486\,I_{12})\,%
+ \frac{17^3}{2^6 \cdot 3^2 \cdot 7^3}\,{\iota_{6}}^{3}%
+ \frac{3\cdot 5}{2^5}\,\iota_{9}^{2}\,%
- \frac{17}{2^3 \cdot 7}\,\iota_{6}\,\iota_{12}\,,%
\end{dmath*}
%
\begin{dmath*}
\iota_{21} = \frac{ 3^7 }{ 2^{44} \cdot 5^7 \cdot 7^{5}}\,%
I_3^{2}\,(-128\,{I_{3}}^{5}+213912\,{I_{3}}^{3}I_{6}+2961\,{I_{3}}^{2}I_{9}-8541\,{I_{3}}^{2}J_{9}-18057600\,I_{3}\,{I_{6}}^{2}+12204\,I_{3}\,I_{12}+810\,I_{3}\,J_{12}-45360\,I_{6}\,I_{9}+285120\,I_{6}\,J_{9}-4860\,I_{15}-540\,J_{15})\,%
+{\frac {2\cdot 5^3}{3^3 \cdot 7^2}}\,\iota_{6}^{2}\iota_{9}-{\frac {13}{2\cdot 3^2}}\,\iota_{9}\,\iota_{12}-{\frac{17}{2^2\cdot 3 \cdot 7}}\,\iota_{6}\,\iota_{15}\,.
\end{dmath*}
%
\end{dgroup*}
}
\end{equation}
Finally $\pi^{14 s} \,D_{14}(f) = \iota_{42}(F) + O(\pi^{15s})$ with
\begin{equation} \label{eq:trompe}
\iota_{42} = \frac{3^{10}}{2^{18
} \cdot 5^5}\, I_3^5 \ I_{27}\,.
\end{equation}
\end{proposition}
\medskip
The previous proposition defines a list of $6$ invariants for ternary quartics
$\ivv=(\iota_6, \ldots, \iota_{21})$ whose evaluation at a generic ternary
quartic form $F$ induces a point in $\P^{(6,9,12,15,18,21)}.$
We can now prove the direct implication of Theorem~\ref{th:main}. Let
$F = Q_0^2 + \pi^s G$ be a good \PHM~corresponding to a stable model $\Cc/\O$
as in Proposition~\ref{2->1}. One has that $v_{\DO}(I_3(F)) = v(I_3(F))=0$
since $I_3(\bar{F})=320/9$. So $v_{\DO}(I_{27}(F))=v(I_{27}(F))>0$ since
$\bar{F}=0$ is singular. Moreover, the reduction of $f=b_8(G)$ gives an
equation for $\Cc_k : y^2=\bar{f}(x,z)$. This implies that $v(D_{14}(f))=0$
and therefore $v(\iota_{42}(F)) = 14s + v(D_{14}(f)) = 14 s.$ Since each of
the $\iota_{3i}(F)$ is divisible by $\pi^{si}$, we get that
$v_{\iv}(\iota_{42}(F))=v_{\iv}(I_3(F)^5 I_{27}(F)) = 0$.\\
The proof of the converse implication will keep us busy till the end of this
section. For a ternary quartic form $F \in \Sym^4(V^{\vee})$, let us denote
$\rho : \Sym^4(V^{\vee}) \to \Sym^2(V)$ the contravariant of degree $4$ and
order 2 defined in \cite{dixmier}. If $T \in\GL_3(R)$, then $\rho$ satisfies
the relation $\rho(F.T) = \det(T)^{6} \cdot T^{-1}.\rho(F)$. Let
$\rho_0=\rho(Q_0^2)$. Then
\begin{equation} \label{eq:rho0}
\rho_0= - 2^{12} \cdot 3^{-2} \cdot 5 \cdot 7 \cdot (v_2^2-v_1 v_3).
\end{equation}
\begin{lemma} Suppose that $\cO$ is complete. Let $\W$ be a smooth scheme over
$\cO$. Then for any $s\ge 1$, the canonical map
$$ \W(\cO)\to \W(\cO/(\pi^s))$$
is surjective.
\end{lemma}
\begin{proof}
Indeed, for any
$\sigma\in \W(\cO/(\pi^s))=\mathrm{Hom}_{\cO}(\Spec(\cO/(\pi^s)), \W)$ we
can replace $\W$ by an affine open neighborhood of the image of $\sigma$ and
it is enough to lift $\sigma$ there. So we can suppose $\W$ is affine. As
$\cO$ is complete, the canonical map
$$ \W(\cO)\to {\projlim}_{n\ge s} \W(\cO/(\pi^n))$$
is an isomorphism. By the smoothness of $\W$, for any pair of integers
$n>r>0$, the canonical map $\W(\cO/(\pi^n))\to \W(\cO/(\pi^r))$ is
surjective. So $\W(\cO)\to \W(\cO/(\pi^s))$ is surjective.
\end{proof}
\begin{proposition} \label{prop:rig} Suppose that $\O$ is complete. Let
$\rho \in \cO[v_1, v_2, v_3]$ be a quadratic (dual) form such that for some
$s\ge 1$ we have $\rho \equiv \rho_0 \mod \pi^s$. Then there exists
$T\in \mathrm{GL}_3(\cO)$ such that
$$T\equiv \mathrm{Id}_3 \mod \pi^s, \quad T.\rho=u \rho_0.$$
with $u \in \O$ such that $u \equiv 1 \mod \pi^s$.
\end{proposition}
\begin{proof}
Let $\W=\underline{\mathrm{Isom}}_{\cO}(V_+(\rho), V_+(\rho_0))$ be the
scheme of linear isomorphisms. For any $\cO$-algebra $A$, $\W(A)$ is the set
of matrices $U=(x_{ij})_{1\le i, j\le 3}$ such that
$${}^tU\rho U=y\rho_0, \quad y\in A^*$$ (here we identify the quadratic forms with their matrices
in the canonical basis). The scheme $\W$ is defined by $6$ equations in the
affine space of relative dimension $10$ ($9$ for the entries of $U$ plus $1$
for the ratio $y$). Its Jacobian matrix at the point
$I: U=\mathrm{Id}_{3}\in M_{3\times 3}(k), y=1$ has rank $4$\footnote{For
the explicit computations one can take for $\rho_0$ the sum of the squares
$\sum_i v_i^2$ as the residue characteristic is different from $2$.}.
This implies that $\W$ is smooth at this point.
By hypothesis, there exists a $\sigma\in \W(\cO/(\pi^s))$ whose image in
$\W(k)$ is equal to the point $I$ above. Applying the previous lemma to the
smooth locus of $\W$ (in fact $\W$ is smooth everywhere), we find a lifting
$T \in \W(\cO)$ of $\sigma$. Then $T \in \GL_3(\O)$ satisfies
$$T\equiv \mathrm{Id}_3 \mod \pi^s, \quad T.\rho= u \rho_0$$
with $u \equiv 1 \pmod{\pi^s}$. Thus $T$ satisfies the required properties.
\end{proof}
\begin{corollary}\label{cor:T1} Suppose that $\O$ is complete and let
$F=Q_0^2+\pi^s G$ be a \PHM. There exists a matrix
$T_1 \in\operatorname{GL}_3(\O)$ with
$T_1 \equiv \operatorname{Id}_3\, \bmod\, \pi^s$ and $s_1 \geq s$ such that
$F_1=F.T_1=Q_0^2+\pi^{s_1} G_1$ is a \PHM{} satisfying $\rho(F_1)
=\rho_0$.
\end{corollary}
\begin{proof}
Let $T$ be the matrix of Proposition~\ref{prop:rig} with $\rho=\rho(F) $ and
let $F'=F.T^{-1}$. Since one has
$T^{-1} \equiv \operatorname{Id}_3 \pmod{\pi^s}$, we see that
$F'=Q_0^2+ \pi^s G'$. Moreover
$\rho(F')=\det(T)^6 \cdot T.\rho(F) = \det(T)^6 \cdot u \cdot \rho_0.$ Let
now $\alpha = (\det(T)^6 \cdot u)^{-1/4}$ with
$\alpha \equiv 1 \pmod{\pi^s}$. Then the form
$F_1=\alpha F'= F.(\alpha T^{-1})$ satisfies the hypotheses.
\end{proof}
\begin{proposition}\label{newF2}
Assume $\O$ is complete and let $C/K: F=0$ with
$F=Q_0^2 + \pi^s G(x_1,x_2,x_3)$ be a \PHM{} of $C$ such that
$\rho(F) =\rho_0$. If $b_8(G)$ is stable over $K$ in the GIT sense, up to a
possible extension of $K$, we can assume that $\overline{b_8(G)}$ is
semi-stable in the GIT sense.
\end{proposition}
\begin{proof}
Let $f=b_8(G)$. By Corollary~\ref{cor:shabur}, after a possible extension of
$K$, there exists a matrix $T\in\SL_2(K)$ such that $f.T=\pi^r \cdot f_1$
with $r \in \Z$ and $f_1 \in \mathcal{O}[x,z]$ such that $\bar{f_1}=0$ is
semi-stable in the GIT sense. Comparing the discriminants on both sides, we
see that $r \geq 0$. Since the kernel of $b_8$ is $Q_0 Q$ where $Q$ is any
quadratic form, we can write $G.h(T)$ in a unique way as $Q_0 Q_1+G_1$ with
$Q_1 \in K[x_1,x_2,x_3]$ a quadratic form and
$$G_1=\beta_8 x_1^4 +\beta_7 x_1^3 x_2+ \beta_6 x_1^3 x_3 + \beta_5 x_1
x_2^3 + \beta_4 x_1 x_2^2 x_3 +\beta_3 x_2^3 x_3 +\beta_2 x_2^2 x_3^2
+\beta_1 x_2 x_3^3 + \beta_0 x_3^4 \in K[x_1,x_2,x_3].$$
Since
\begin{eqnarray*}
b_8(G_1) & =& \beta_8 x^8 + 2 \beta_7 x^7 z + \beta_6 x^6 z^2 + 8 \beta_5
x^5 z^3 + 4 \beta_4 x^4 z^4 + 8 \beta_3 x^3 z^5 + 4 \beta_2
x^2 z^6 + 2 \beta_1 x z^7 + \beta_0 z^8 \\
&=& \pi^{r} f_1,
\end{eqnarray*} we get that
$v(\beta_i) \geq r$ for all $i\in\{0,\ldots,8\}$, in particular there
exists $G_2 \in \O[x_1,x_2,x_3]$ such that $G_1=\pi^r G_2$. Let us now write
$Q_1=\pi^t Q_2$ with $t \in \Z$ and $Q_2 \in \O[x_1,x_2,x_3]$ primitive. We
have that
$$F.h(T) = Q_0^2 + \pi^s G.h(T) = Q_0^2 + \pi^{s+t} Q_0 Q_2 + \pi^{s+r} G_2.$$
Since $\rho(F.h(T))=\rho(F) =\rho_0$, Lemma~\ref{a>=b} implies that
$s+t \geq s+r$ so $t \geq r$. Hence, we get that
$$F.h(T) = Q_0^2 + \pi^{s+r} \underbrace{(G_2+ \pi^{t-r} Q_0 Q_2)}_{ G_3}.$$
We have that $b_8(G_3)=b_8(G_2)=b_8(G_1/\pi^r)=f_1$. This last equality also
implies that $G_3$ is primitive.
\end{proof}
\begin{lemma}\label{a>=b}
Let $F=Q_0^2+\pi^a Q_0Q + \pi^b G \in K[x_1,x_2,x_3]$ be a ternary quartic
form such that $a,b \in\mathbb{Z}$, $b>0$, $Q$ is an integral primitive
quadric and $G$ is an integral quartic form. If $\rho(F) \equiv \rho_0$
$\pmod{\pi^{a+1}}$, then $a\geq b$.
\end{lemma}
\begin{proof}
Assume $a<b$ and write
$Q=a_1 x_1^2 + a_2 x_1x_2 + a_3 x_1x_3 + a_4 x_2^2 + a_5 x_2x_3 + a_6
x_3^2$.
We see that modulo $\pi^{a+1}$ , the coefficients of $\rho(F)$ are equal to
\begin{equation*}
\begin{split}
\left( \frac{17920}{9}\pi^a a_6, -\frac{35840}{9} \pi^a a_5,
-\frac{89600}{9}\pi^a a_4 + \frac{71680}{3}\pi^a a_3 + \frac{143360}{9},\right.\\
\left.\frac{35840}{3}\pi^a a_4 - \frac{143360}{9}\pi^a a_3 - \frac{143360}{9},
-\frac{35840}{9}\pi^a a_2, \frac{17920}{9}\pi^a a_1\right).
\end{split}
\end{equation*}
The equality $\rho(F) \equiv \rho_0$ $\pmod{\pi^{a+1}}$ imposes that they
should be congruent to
$$(0,0,\frac{143360}{9}, - \frac{143360}{9}, 0,0)$$ modulo $\pi^{a+1}$, from where we easily check that
$v(a_i)\geq 1$ for all $i\in\{1,\ldots,6\}$ and we get a contradiction with
$Q$ being primitive. \end{proof} We can now finish the proof of
Theorem~\ref{th:main}. Let us assume that
$$v_{\DO}(I_3(F))=0, \; v_{\DO}(I_{27}(F))>0 \text{ and } v_{\iv}(\iota_{42}(F))=0.$$
As for the case of potentially good quartic reduction, we can moreover assume
that $\O$ is complete, hence excellent. By Proposition~\ref{prop:ssred},
after a possible extension of $K$, there exists a form $F_0$ that is
$\GL_3(K)$-equivalent to $F$, integral and primitive, and such that
$v(I_3(F_0))=0$ and $v(I_{27}(F_0))>0$. Since
$v_{\iv}(\iota_{42}(F)) = v_{\iv}(\iota_{42}(F_0)) = 0$ and
$v(\iota_{42}(F_0)) = 5 v(I_3(F_0)) + v(I_{27}(F_0))>0$, the condition
$v_{\iv}(\iota_{42}(F))=0$ imposes that $v(\iota_{3i}(F_0)) > 0$ for
$i=2,\ldots,7$. This yields these 6 equations that must be satisfied by
$\bar{F_0}$:
\begin{displaymath}
I_{6} = \frac{1}{180}\,I_3^2,\ %
I_{9} = \frac{49}{36}\,I_3^3,\ %
J_{9} = \frac{49}{60}\,I_3^3,\ %
I_{12} = \frac{343}{1620}\,I_3^4,\ %
J_{12} = \frac{49}{36}\,I_3^4,\ %
J_{15} + 9\,I_{15} = \frac{2744}{675}\,I_3^5.
\end{displaymath}
Substituting these expressions in the 23 relations satisfied by the
Dixmier-Ohno invariants \cite{ohno}, we obtain a polynomial system that
decomposes into 5 distinct radical components. The three first ones are such
that $I_3(\bar{F_0})=0$, and the last one is such that
$I_{27}(\bar{F_0})\neq0$, which in both cases contradicts our
hypotheses. There thus only remains one case for which $\DO(\bar{F}_0)$ are
equal to the ones in \eqref{eq:1}. We therefore get by
Proposition~\ref{prop:MHR} that $F$ admits a \PHM{} $Q_0^2+\pi^r G=0$.
Using Corollary~\ref{cor:T1}, we can assume that $\rho(F) = \rho_0$. Moreover
since $v_{\iota}(\iota_{42}(F))=0$, equation~\eqref{eq:trompe} shows that
$v(D_{14}(b_8(G)) \ne \infty$, \ie~$D_{14}(b_8(G)) \ne 0$ and so $b_8(G)=0$ is
stable over $K$ in the GIT sense. We are therefore in the situation of
Proposition~\ref{newF2}. In particular, we know that there exists a \PHM{}
$F_1=Q_0^2 + \pi^s G_1$ such that $f_1=b_8(G_1)$ has semi-stable reduction in
the GIT sense. Hence, by Corollary~\ref{cor:Shiodas}, there exists
$2 \leq i_0 \leq 7$ such that $v(j_{i_0}(f_1))=0$. From
equation~\eqref{eq:DOshort}, we get that $v(\iota_{3i_0}(F_1)) = \pi^{i_0s}$.
Hence
\begin{displaymath}
\frac{\iota_6(F_1)}{\pi^{2s}},\ \ldots,\ \frac{\iota_{21}(F_1)}{\pi^{7s}},\ %
\frac{\iota_{42}(F_1)}{\pi^{14s}}
\end{displaymath}
is a minimal representative of the invariants $\iota$ of $F$. Since we have
assume that $v_{\iv}(\iota_{42}(F))=0$, this implies that
$v(\iota_{42}(F_1)) = 14 s$, so $v(D_{14}(f_1))=0$ and $F_1$ is a good \PHM{},
which concludes the proof.
Actually, one can get expressions as in Proposition~\ref{prop:shortrel} for
all the Shioda invariants $j_2,\ldots,j_{10}$. In the case of potentially good
hyperelliptic reduction, we can therefore reconstruct an equation of the
special fiber over $\bar{k}$ using the results of \cite{LR11}.
\begin{example} \label{ex:bass} We resume with the example of the
introduction. Let $C/\Q : F=0$ where
$$F=(x_2 +x_3) x_1^3-(2 x_2^2 + x_3 x_2) x_1^2+(x_2^3-x_3 x_2^2+2 x_3^2 x_2-x_3^3) x_1-2 x_3^2 x_2^2+3 x_3^3 x_2.$$
One can check that $D_{27}(F)=-13^6$ so the curve has good quartic reduction
away from $13$. At $13$ the valuation of Dixmier-Ohno invariants are all $0$
except for the one of $D_{27}(F)$ which is $6$. Moreover using the formulas
of Proposition~\ref{prop:shortrel}, we get that the valuations at $13$ of
the invariants $\iota$ are $(1, 2, 2, 3, 3, 3)$. Therefore
$$v_{\DO}(I_3(F))=0, \; v_{\DO}(D_{27}(F))=\frac{6}{27}>0, \; v_{\iota}(I_3(F)^5 D_{27}(F)) =\frac{6}{42}-\min\left\{\frac{1}{6},\frac{2}{9},\frac{2}{12},\frac{3}{15},\frac{3}{18},\frac{3}{21},\frac{6}{42}\right\}=0.$$
The curve has hence potentially good hyperelliptic reduction at $13$.
\end{example}
\begin{remark} \label{rem:p>7} The assumption $p>7$ was used in various
occasions. If in Lemma~\ref{lem:test-conic} or in
Proposition~\ref{prop:shortrel}, a better choice of invariants or
normalization of the expressions may overcome this restriction, the
appearance of these primes in equation \eqref{eq:rho0} is more
symptomatic. Let us for instance consider $p=7$. It is known that the Klein
quartic $C/\Q : F=0$ where $F=x^3 y+y^3z+z^3 x$ has potentially good
hyperelliptic reduction modulo $7$. However, since
$\Aut_{\bar{\Q}}(C) \simeq \PSL_2(\F_7)$ and the latter is not a subgroup of
$\SO_3(\bar{\Q}) \simeq \PSL_2(\bar{\Q})$, it cannot be the automorphism
group of any non-degenerate conic. Hence, $\rho_0(F)=0$ and this is the case
for any other possible covariant or contravariant of order 2. Therefore, we
cannot go further with our current strategy in this case. However, in this
particular case it is easy to find a \PHM, see \cite[p.56 and p.82]{Elkies}.
\end{remark}
\section{Homogeneous system of parameters over $\O$}
\label{sec:dixm-invar-char}
In this article, we make a significant use of HSOP over $\O$ for ternary
quartic forms under the action of $\SL_{3,\O}$ where $\O$ is an excellent DVR
with residue field $k$ of characteristic $p$. Dixmier proves~\cite{dixmier}
that the $7$ invariants $I_{3}$, $I_{6}$, $I_{9}$, $I_{12}$, $I_{15}$,
$I_{18}$ and $I_{27}$ form a HSOP over a field of characteristic $0$. The
result does not carry in characteristic $p>0$ but the strategy he followed is
valid in positive characteristic, although it needs to be handled with care.
We summarize our results in the following theorem (we refer to~\cite{HSOPList}
for the large expressions of the invariants when $p=2$ or $3$).
\begin{theorem}\label{th:domodp}
Let $\O$ be an excellent DVR with residue field of characteristic $p$. A
HSOP over $\O$ for ternary quartic forms under the action of $\SL_{3,\O}$ is
formed by
\begin{itemize}
\item explicit invariants of degree
\begin{math}
(3, 9, 12, 15, 18, 24, 27)
\end{math} when $p=2$;
\item explicit invariants of degree
\begin{math}
(3, 9, 15, 18, 27, 27, 36)
\end{math} when $p=3$;
\item \begin{math} ( I_{3},\ I_{6},\ I_{9}^{(5)},\ I_{12},\ J_{15}^{(5)},\
I_{18},\ I_{27} )
\end{math}
where $I_{9}^{(5)} = (J_9 + 3\,I_9)/5$ and
%
\begin{small}
\begin{multline*}
J_{15}^{(5)} = (-64\,{{I_{3}}}^{3}{I_{6}}-24\,{I_{3}}\,{{I_{6}}}^{2}+39\,{I_{3}}\,{I_{12}}-11\,{I_{3}}\,{J_{12}}+42\,{I_{6}}\,{I_{9}}-21\,{I_{6}}\,{J_{9}}-1143\,{I_{15}}+{J_{15}})/5^3\\
+({{253}}\,{{I_{3}}}^{2}{I_{9}}-{{79}}\,{{I_{3}}}^{2}{J_{9}})/5^4\,\,\text{when
}p=5;
\end{multline*}
\end{small}
%
\item \begin{math} ( I_{3},\, I_{6},\, I_{9} - J_{9},\, I_{12},\, I_{15},\,
I_{18},\, I_{27} )
\end{math}
when $p = 7$, $19$, $47$, $277$ and $523$;
\item \begin{math} ( I_{3},\, I_{6},\, I_{9},\, I_{12},\, I_{15},\,
I_{18},\, I_{27} )
\end{math}
otherwise.
%
\end{itemize}
\end{theorem}
\begin{proof}
By Remark~\ref{rem:hsop}, it is enough to prove that the previous sets are
HSOP over their residue field. To do this, let us consider ternary quartic
forms $F$ in characteristic $p$ such that $I(F) = 0$ for all $I \in \Iv$,
where $\Iv$ is the HSOP given in the theorem. We write $F$ as
\begin{multline*}
a_{04}\,{y}^{4}+a_{13}\,x{y}^{3}+a_{22}\,{x}^{2}{y}^{2}+a_{31}\,{x}^{3}y+a_{40}\,{x}^{4}\,
+ ( a_{03}\,{y}^{3}+a_{12}\,x{y}^{2}+a_{21}\,{x}^{2}y+a_{30}\,{x}^{3}
)\,z\, \\+ ( a_{02}\,{y}^{2}+a_{11}\,xy+a_{20}\,{x}^{2} )\,{z}^{2}\,+ (
a_{01}\,y+a_{10}\,x )\,{z}^{3}\,+ a_{00}\,{z}^{4}\,.
\end{multline*}
%
The set $\Iv$ is a HSOP in characteristic $p$ if and only if $F$ is unstable
in the GIT sense, or equivalently $F$ has a triple point or consists of a
cubic and an inflectional tangent line~\cite[§1.12]{mum-ens}.
%
Since $I_{27}(F) =D_{27}(F)= 0$, the quartic $F=0$ has a singular point,
that we can, up to a linear change of variables, set to
$\Pi=(0:0:1)$. Following~\cite{dixmier}, we use the remaining degrees of
freedom to finally restrict to the six different cases that we sum up in
Table~\ref{tab:singmodels} and that we will consider one after the
other. The linear transformations used to get these normalizations are valid
over fields of characteristic $p \geqslant 5$ and $0$.
\begin{table}[htbp]
\centering
\setlength{\tabcolsep}{2pt}
\renewcommand{\arraystretch}{1.2}
\begin{tabular}{c|l|l|l|}
Case & \multicolumn{2}{l|}{\ Condition}& \ $F(x,y,z)$\\\hline\hline
0& \multicolumn{2}{l|}{$\Pi$ of order $\geqslant 3$} &
$
\begin{array}{c}
a_{04}\,{y}^{4}+a_{13}\,x{y}^{3}+a_{22}\,{x}^{2}{y}^{2}+a_{31}\,{x}^{3}y+a_{40}\,{x}^{4}+\\
( a_{03}\,{y}^{3}+a_{12}\,x{y}^{2}+a_{21}\,{x}^{2}y+a_{30}\,{x}^{3} )\,z
\end{array}
$\\\hline\hline
1.1& \multirow{2}%
{0.2\textwidth}%
{$\Pi$ of order 2 \& $x=0$ is the only tangent at $\Pi$}%
& $\begin{array}{c}\ a_{03} = 0\end{array}$ &
$\begin{array}{c}
a_{04}\,{y}^{4}+a_{13}\,{y}^{3}x+a_{22}\,{y}^{2}{x}^{2}+a_{31}\,y{x}^{3}+a_{40}\,{x}^{4}+\\
( a_{12}\,{y}^{2}x+a_{21}\,y{x}^{2}+a_{30}\,{x}^{3} )\, z+{x}^{2}\,{z}^{2}
\end{array}$%
\\\cline{1-1}\cline{3-4}
1.2 & & $\begin{array}{c}\ a_{03} \neq 0\end{array}$ &
$\begin{array}{c}
a_{04}\,{y}^{4}+a_{13}\,{y}^{3}x+a_{22}\,{y}^{2}{x}^{2}+a_{31}\,y{x}^{3}+a_{40}\,{x}^{4}+\\
{y}^{3}z+{x}^{2}{z}^{2}
\end{array}$
\\\hline\hline
2.1& \multirow{5}%
{0.2\textwidth}%
{$\Pi$ of order 2 \& $x=0$, $y=0$ are both tangent at $\Pi$}%
&
$\begin{array}{c}
a_{03} \neq 0\\
\&\,a_{30} \neq 0
\end{array}$
&
$\begin{array}{c}
a_{04}\,{y}^{4}+a_{13}\,{y}^{3}x+a_{22}\,{y}^{2}{x}^{2}+a_{31}\,y{x}^{3}+a_{40}\,{x}^{4}+\\
( {y}^{3}+{x}^{3} ) z
+yx{z}^{2}
\end{array}$
\\\cline{1-1}\cline{3-4}
2.2 & &
$\begin{array}{c}
a_{03} = 0\\
\&\,a_{30} \neq 0
\end{array}$ &
$\begin{array}{c}
a_{04}\,{y}^{4}+a_{13}\,{y}^{3}x+a_{22}\,{y}^{2}{x}^{2}+a_{31}\,y{x}^{3}+a_{40}\,{x}^{4}+\\
{x}^{3}z+yx{z}^{2}
\end{array}$
\\\cline{1-1}\cline{3-4}
2.2 & &
$\begin{array}{c}
a_{03} = 0\\
\&\,a_{30} = 0
\end{array}$ &
$\begin{array}{c}
a_{04}\,{y}^{4}+a_{13}\,{y}^{3}x+a_{22}\,{y}^{2}{x}^{2}+a_{31}\,y{x}^{3}+a_{40}\,{x}^{4}+\\
yx{z}^{2}
\end{array}$\\\hline\hline
\end{tabular}
\smallskip
\caption{Singular quartics in characteristic $p \geqslant 5$ and $0$}
\label{tab:singmodels}
\end{table}
\noindent$\bullet$ Case $0$ yields quartics with $\Pi$ being a triple point
and these curves are unstable \cite[Lem.1.4]{artebani}.\smallskip
\noindent$\bullet$ Case $1.1$ is also easily manageable; we use the computer
algebra software \Magma{} to shortcut some cumbersome computations (that
Dixmier made by hand over the rationals!). In particular, we compute a
Gr\"obner basis \emph{over $\Z$} (based on the computation of normal Hermite
forms of integral Macaulay matrices) of the algebraic system
$\{\,I(F) = 0 \,\}_{I\in\Iv}$ for the following lexicographic order:
$$a_{30}>a_{22}>a_{21}>a_{31}>a_{13}>a_{40}>a_{04}.$$ We get
\begin{displaymath}
\{{a_{12}}^{8}+144\,{a_{04}}^{4},\ %
{a_{12}}^{4}a_{04}+144\,{a_{04}}^{3},\ %
2\,{a_{12}}^{2}+12\,a_{04},\ %
360\,{a_{04}}^{3},\ %
720\,{a_{04}}^{2}\}\,.
\end{displaymath}
In characteristic $p>5$ and $p=0$, we thus have $a_{04} = a_{12} = 0$\,.
This yields quartics $F=0$ which are the union of a cubic curve and an
inflectional tangent at $\Pi$, which we can assume to be $x=0$, thus
unstable quartics \cite[Lem.1.4]{artebani}.\smallskip
\noindent$\bullet$ Case $1.2$ is done in a similar manner. Gr\"obner basis
can also be computed over $\Z$. When $p>5$ and $p=0$, we derive from it
that $a_{04}=a_{31}=a_{22}=0$, and
\begin{displaymath}
\{{a_{13}}^{2}+9953670985\,a_{40},\ 2^{4}\cdot 3^{13}\cdot 5^{2} \cdot 7\cdot 19\cdot 47 \,a_{40}\}\,.
\end{displaymath}
Thus, for $p\neq 2$, $3$, $5$, $7$, $19$ and $47$, we have
$a_{13}=a_{40}=0$. This yields quartics of the same type as in Case $1.1$,
and thus, unstable.\smallskip
\noindent$\bullet$ Case $2.1$ is the most difficult case, at least
computationally speaking. From $I_3(F) = I_{6}(F) = 0$, we easily deduce
that $a_{22}=-9/2$ and $a_{04}a_{40}=9/4$. Now, the equations $I(F)=0$ for
the other invariants $I$ yield more algebraic relations between the
remaining variables, but of higher degrees, and we were not able any more to
compute an integral Gr\"obner basis. Instead, we find sufficient conditions
on the remaining $a_{ij}$ by solving with resultants. We eliminate, first
$a_{04}$, then $a_{40}$ and finally $a_{31}$ (resp. $a_{13}$, $a_{31}$ and
$a_{04}$), and obtain a polynomial of degree 48 in $a_{13}$ (resp. of degree
28 in $a_{40}$),
\begin{displaymath}
N_{13}\, (4\,a_{13} + 15)^{16}\, (16\,a_{13}^2 - 60\,a_{13} + 225)^{16}\ \ \ \ %
(\text{resp. }N_{40}\, a_{40}^4\, (2\,a_{40} + 3)^8\, (4a_{40}^2 - 6\,a_{40} + 9)^8\,)\,,
\end{displaymath}
where $N_{13}$ (resp. $N_{40}$) is an integer, too large to be completely
factorized. We can have $N_{13}$ and $N_{40}$ both equal to zero modulo $p$,
but this can only happen for the few prime divisors $p$ of
$\gcd(N_{13}, N_{40})$. These cases can be easily studied independently (see
the end of the proof). Let us assume from now that it is not the case, i.e
$p \ne 2$, $3$, $5$, $7$, $13$, $43$, $47$, $89$, $277$ and $523$. Either
$N_{13}\ne0$ or $N_{40}\ne0$.
\begin{itemize}[label=\raisebox{0.25ex}{\tiny$\bullet$}]
\item If $N_{13} \not \equiv 0 \bmod p$, then we have two possibilities\,:
\begin{itemize}
%
\item $4\,a_{13}+15=0$\,, and with a couple of additional resultants, we
show that $2\,a_{40} + 3 = 0$ when $p \ne 2$, $3$, $13$, $1613$, $3469$
and 6 other 5-24 digit primes;
%
\item $16\,a_{13}^2-60\,a_{13}+225=0$\,, and this implies that
$4a_{40}^2 - 6\,a_{40} + 9 = 0$ when additionally $p \ne 37$, $61$ and
$210037$.
\end{itemize}
\item If $N_{40}\not \equiv 0 \bmod p$, then we have the two symmetric
possibilities\,:
\begin{itemize}
%
\item $2\,a_{40} + 3 = 0$\,, which implies that $4\,a_{13}+15=0$ when
additionally $p \ne 53$, $4969$ and 3 new 7-19 digit primes;
%
\item $4a_{40}^2 - 6\,a_{40} + 9 = 0$\,, which implies
$16\,a_{13}^2-60\,a_{13}+225=0$ when $p$ is not one of the primes above.
\end{itemize}
\end{itemize}
In other words, except modulo the few primes $p$ that we have listed above,
we have either
\begin{multline*}
a_{31} = -15/4,\ a_{13} = -15/4,\ a_{40} = -3/2 \text{ and }a_{04} = -3/2\text{ or }\\
a_{04}^2 - 3/2\,a_{04} + 9/4=0,\ a_{40} = -a_{04} + 3/2,\ a_{13} =
-5/2\,a_{04} + 15/4\text{ and }a_{31} = 5/2\,a_{04}\,.
\end{multline*}
In both cases, this yields quartics $F=0$ which are the union of a cubic
curve and an inflectional tangent, thus unstable.\smallskip
\noindent$\bullet$ Cases~2.2 and~2.3 are much easier and can be handled
again with integral Gr\"obner basis computations. They do not yield
additional exceptional primes. We omit the details. \medskip
To end the proof, we have to consider each exceptional prime $p$ one by one,
and check over $\F_{p}$ the stability of the quartics $F$ that are in the
algebraic set defined by $I_3(F)=I_6(F)=\ldots=I_{27}(F)=0$. Since we have
to deal now with Gr\"obner basis computations in finite fields, and no
longer over $\Z$, these computations can be easily done with \Magma. In this
way, we refine the list to only 6 exceptional primes $p$\,: $5$, $7$, $19$,
$47$, $277$ and $523$. Then, to conclude, it is easy to check with similar
computations that replacing $I_9$ by $I_{9} - J_{9}$ for $p \geqslant 7$, or
$I_{9}$ by $I_{9}^{(5)}$ and $I_{15}$ by $J_{15}^{(5)}$ for $p=5$, yields a
HSOP.
For $p=2$ or $p=3$, a similar work can be done to obtain the same six cases
of Table~\ref{tab:singmodels}, with slightly distinct models $F$. We omit
the details.
\end{proof}
\begin{remark} \label{rem:nonlift} Finding a set of generators in small
characteristics $p$ to replace or complete the Dixmier-Ohno invariants is,
as far as we know, an open problem. There are invariants which do not come
from reduction of an invariant in characteristic $0$. For $p=3$, for
instance, the degree-6 homogeneous component is not as in characteristic $0$
of dimension $2$, but of dimension $3$, generated by $i_3^2$, $i_6$ and
$i_6'$ where
\begin{displaymath}
i_3 = 2^{4} \cdot 3^3\ I_3,\quad i_6 = 2^{11} \cdot 3^2\ (18\,{I_{6}}-{{I_{3}}}^{2})\ \text{ and }\ i'_6 = 2^{12} \cdot 3^5\ (9\,{{{I_{9}}}-{{I_{3}}}^{3})\,/\,{{i_{3}}}}.
\end{displaymath}
Notice that these expressions must be understood as the reduction of the
polynomial expressions over $\Z$.
\end{remark}
\newcommand{\etalchar}[1]{$^{#1}$}
\begin{thebibliography}{DDMM17}
\bibitem[ACGH85]{harris}
E.~Arbarello, M.~Cornalba, P.~Griffiths, and J.~Harris.
\newblock {\em Geometry of algebraic curves, Vol. {I}}, volume 267.
\newblock Grundlehren der Mathematischen Wissenschaften, Springer-Verlag,
New-York, 1985.
\bibitem[Art09]{artebani}
M.~Artebani.
\newblock A compactification of {$M_3$} via {$K3$} surfaces.
\newblock {\em Nagoya Math. J.}, 196:1--26, 2009.
\bibitem[Bas15]{basson}
R.~Basson.
\newblock {\em Arithm{\'e}tique des espaces de modules des courbes
hyperelliptiques de genre $3$ en caract\'eristique positive}.
\newblock PhD thesis, Universit\'e de Rennes 1, Rennes, 2015.
\bibitem[BBW17]{BBW17}
M.~Borner, I.~{Bouw}, and S.~Wewers.
\newblock Picard curves with small conductor.
\newblock Preprint, 2017.
\bibitem[BCL{\etalchar{+}}15]{WINE1}
I.~Bouw, J.~Cooley, K.~Lauter, E.~{Lorenzo Garc\'\i a}, M.~Manes, R.~Newton,
and E.~Ozman.
\newblock Bad reduction of genus three curves with complex multiplication.
\newblock In {\em Women in numbers {E}urope}, volume~2 of {\em Assoc. Women
Math. Ser.}, pages 109--151. Springer, Cham, 2015.
\bibitem[BCP97]{Magma}
W.~Bosma, J.~Cannon, and C.~Playoust.
\newblock The {M}agma algebra system. {I}. {T}he user language.
\newblock {\em J. Symbolic Comput.}, 24(3-4):235--265, 1997.
\newblock Computational algebra and number theory (London, 1993).
\bibitem[BDS{\etalchar{+}}17]{BDMTV17}
J.~S. {Balakrishnan}, N.~{Dogra}, J.~{Steffen M{\"u}ller}, J.~{Tuitman}, and
J.~{Vonk}.
\newblock {Explicit Chabauty-Kim for the Split Cartan Modular Curve of Level
13}.
\newblock {\em ArXiv e-prints}, 2017.
\bibitem[BKK{\etalchar{+}}18]{BKKSW}
I.~I. Bouw, R.~Kohls, A.~Koutsianas, J.~Sijsling, and S.~Wewers.
\newblock Conductor and discrimant of picard curves.
\newblock {\em In progress}, 2018.
\bibitem[Bur92]{Bur}
J.-F. Burnol.
\newblock Remarques sur la stabilit\'{e} en arithm\'{e}tique.
\newblock {\em Internat. Math. Res. Notices}, (6):117--127, 1992.
\bibitem[BW17]{BW}
I.~I. Bouw and S.~Wewers.
\newblock Computing {$L$}-functions and semistable reduction of superelliptic
curves.
\newblock {\em Glasg. Math. J.}, 59(1):77--108, 2017.
\bibitem[Cle80]{clemens}
C.~H. Clemens.
\newblock {\em A scrapbook of complex curve theory}.
\newblock Plenum Press, New York-London, 1980.
\newblock The University Series in Mathematics.
\bibitem[DDMM17]{maistret}
T.~{Dokchitser}, V.~{Dokchitser}, C.~{Maistret}, and A.~{Morgan}.
\newblock {Semistable types of hyperelliptic curves}.
\newblock {\em ArXiv e-prints}, 2017.
\bibitem[Dem12]{demazure}
M.~Demazure.
\newblock R\'esultant, discriminant.
\newblock {\em Enseign. Math. (2)}, 58(3-4):333--373, 2012.
\bibitem[Dix87]{dixmier}
J.~Dixmier.
\newblock On the projective invariants of quartic plane curves.
\newblock {\em Adv. in Math.}, 64:279--304, 1987.
\bibitem[DK02]{DerKem}
H.~Derksen and G.~Kemper.
\newblock {\em Computational invariant theory}.
\newblock Invariant Theory and Algebraic Transformation Groups, I.
Springer-Verlag, Berlin, 2002.
\newblock Encyclopaedia of Mathematical Sciences, 130.
\bibitem[Edi90]{edixhoven90}
B.~Edixhoven.
\newblock {Minimal resolution and stable reduction of {$X_0(N)$}.}
\newblock {\em {Ann. Inst. Fourier}}, 40(1):31--67, 1990.
\bibitem[Elk99]{Elkies}
N.~D. Elkies.
\newblock The {K}lein quartic in number theory.
\newblock {\em Math. Sci. Res. Inst. Publ.}, 35:51--101, 1999.
\bibitem[Els15]{elsenhans}
A.-S. Elsenhans.
\newblock Explicit computations of invariants of plane quartic curves.
\newblock {\em J. Symbolic Comput.}, 68(part 2):109--115, 2015.
\bibitem[ES18]{elsenhans-stoll}
A.-S. Elsenhans and M.~Stoll.
\newblock Good models for cubic surfaces.
\newblock {\em In progress}, 2018.
\bibitem[GKZ94]{gelfand}
I.~M. Gelfand, M.~M. Kapranov, and A.~V. Zelevinsky.
\newblock {\em Discriminants, resultants, and multidimensional determinants}.
\newblock Mathematics: Theory \& Applications. Birkh\"auser Boston Inc.,
Boston, MA, 1994.
\bibitem[Gla79]{glass}
J.~P. Glass.
\newblock Bitangents of plane quartics.
\newblock {\em Bull. Austral. Math. Soc.}, 20(2):207--210, 1979.
\bibitem[GLL15]{GLL}
O.~Gabber, Q.~Liu, and D.~Lorenzini.
\newblock Hypersurfaces in projective schemes and a moving lemma.
\newblock {\em Duke Math. J.}, 164(7):1187--1270, 2015.
\bibitem[Har87]{harriscm}
J.~Harris.
\newblock Curves and their moduli.
\newblock In {\em Algebraic geometry, {B}owdoin, 1985 ({B}runswick, {M}aine,
1985)}, volume~46 of {\em Proc. Sympos. Pure Math.}, pages 99--143. Amer.
Math. Soc., Providence, RI, 1987.
\bibitem[Has99]{hassett}
B.~Hassett.
\newblock Stable log surfaces and limits of quartic plane curves.
\newblock {\em Manuscripta Math.}, 100(4):469--487, 1999.
\bibitem[HL10]{hyeon}
D.~Hyeon and Y.~Lee.
\newblock Log minimal model program for the moduli space of stable curves of
genus three.
\newblock {\em Math. Res. Lett.}, 17(4):625--636, 2010.
\bibitem[HLP00]{howe00}
E.~W. {Howe}, F.~{Lepr\'evost}, and B.~{Poonen}.
\newblock {Large torsion subgroups of split Jacobians of curves of genus two or
three.}
\newblock {\em {Forum Math.}}, 12(3):315--364, 2000.
\bibitem[HM98]{harris-moduli}
J.~Harris and I.~Morrison.
\newblock {\em Moduli of curves}, volume 187 of {\em Graduate Texts in
Mathematics}.
\newblock Springer-Verlag, New York, 1998.
\bibitem[Hol95]{Hol}
R.-P. Holzapfel.
\newblock {\em The ball and some {H}ilbert problems}.
\newblock Lectures in Mathematics ETH Z\"urich. Birkh\"auser Verlag, Basel,
1995.
\newblock Appendix I by J. Estrada Sarlabous.
\bibitem[Kan00]{kang}
P.-L. Kang.
\newblock On singular plane quartics as limits of smooth curves of genus three.
\newblock {\em J. Korean Math. Soc.}, 37(3):411--436, 2000.
\bibitem[KLL{\etalchar{+}}16]{WINE1-2}
P.~K{\i}l{\i}\c{c}er, K.~Lauter, E.~{Lorenzo Garc{\'i}a}, R.~Newton, E.~Ozman,
and M.~Streng.
\newblock A bound on the primes of bad reduction of {CM} curves of genus 3.
\newblock 2016.
\bibitem[KLL{\etalchar{+}}18]{KLLRSS17}
P.~K{\i}l\i\c{c}er, H.~Labrande, R.~Lercier, C.~Ritzenthaler, J.~Sijsling, and
M.~Streng.
\newblock Plane quartics over {$\Bbb {Q}$} with complex multiplication.
\newblock {\em Acta Arith.}, 185(2):127--156, 2018.
\bibitem[KLS18]{KLS18}
P.~K{\i}l{\i}\c{c}er, E.~{Lorenzo Garc{\'i}a}, and M.~Streng.
\newblock Primes dividing invariants of {CM} picard curves.
\newblock Preprint, 2018.
\bibitem[Kol97]{kollar}
J.~Koll\'{a}r.
\newblock Polynomials with integral coefficients, equivalent to a given
polynomial.
\newblock {\em Electron. Res. Announc. Amer. Math. Soc.}, 3:17--27, 1997.
\bibitem[Kon00]{kondo}
S.~Kond\=o.
\newblock A complex hyperbolic structure for the moduli space of curves of
genus three.
\newblock {\em J. Reine Angew. Math.}, 525:219--232, 2000.
\bibitem[KW05]{kowe}
K.~Koike and A.~Weng.
\newblock Construction of {CM} {P}icard curves.
\newblock {\em Math. Comp.}, 74(249):499--518 (electronic), 2005.
\bibitem[Liu93]{liug2}
Q.~Liu.
\newblock Courbes stables de genre {$2$} et leur sch\'ema de modules.
\newblock {\em Math. Ann.}, 295(2):201--222, 1993.
\bibitem[Liu02]{liu-book}
Q.~Liu.
\newblock {\em Algebraic geometry and arithmetic curves}, volume~6 of {\em
Oxford Graduate Texts in Mathematics}.
\newblock Oxford University Press, Oxford, 2002.
\newblock Translated from the French by Reinie Ern\'e, Oxford Science
Publications.
\bibitem[LL99]{LiLo}
Q.~Liu and D.~Lorenzini.
\newblock Models of curves and finite covers.
\newblock {\em Compositio Math.}, 118:61--102, 1999.
\bibitem[LLLR18]{HSOPList}
R.~{Lercier}, Q.~{Liu}, E.~{Lorenzo Garc{\'i}a}, and C.~{Ritzenthaler}.
\newblock Homogenous system of parameters ({HSOP}) for the algebra of ternary
quadratic forms under the action of $\mathrm{SL}_3$.
\newblock Available at
\href{https://arxiv.org/src/1803.05816/anc/DixmierHSOP.txt}{\tt
https://arxiv.org/src/1803.05816/anc/DixmierHSOP.txt}, 2018.
\bibitem[LR12]{LR11}
R.~Lercier and C.~Ritzenthaler.
\newblock Hyperelliptic curves and their invariants: geometric, arithmetic and
algorithmic aspects.
\newblock {\em J. Algebra}, 372:595--636, 2012.
\bibitem[LRS18]{LRS16}
R.~{Lercier}, C.~{Ritzenthaler}, and J.~{Sijsling}.
\newblock {Reconstructing plane quartics from their invariants}.
\newblock {\em Discrete \& Computational Geometry}, pages 1--41, 2018.
\bibitem[LS16]{LaSo}
J.-C. Lario and A.~{Somoza}.
\newblock A note on picard curves of {CM}-type.
\newblock Preprint, 2016.
\bibitem[MF82]{mumford-fogarty}
D.~Mumford and J.~Fogarty.
\newblock {\em Geometric invariant theory}, volume~34 of {\em Ergebnisse der
Mathematik und ihrer Grenzgebiete [Results in Mathematics and Related
Areas]}.
\newblock Springer-Verlag, Berlin, second edition, 1982.
\bibitem[Mum77]{mum-ens}
D.~Mumford.
\newblock Stability of projective varieties.
\newblock {\em Enseignement Math. (2)}, 23(1-2):39--110, 1977.
\bibitem[Ohn07]{ohno}
T.~Ohno.
\newblock The graded ring of invariants of ternary quartics {I}, 2007.
\newblock unpublished.
\bibitem[Ses77]{Se}
C.~S. Seshadri.
\newblock Geometric reductivity over arbitrary base.
\newblock {\em Advances in Math.}, 26(3):225--274, 1977.
\bibitem[Sha80]{shah}
J.~Shah.
\newblock A complete moduli space for {$K3$}\ surfaces of degree {$2$}.
\newblock {\em Ann. of Math. (2)}, 112(3):485--510, 1980.
\bibitem[Shi67]{shioda67}
T.~Shioda.
\newblock On the graded ring of invariants of binary octavics.
\newblock {\em American J. of Math.}, 89(4):1022--1046, 1967.
\bibitem[Sil98]{silverman-dyna}
J.~H. Silverman.
\newblock The space of rational maps on {$\bold P^1$}.
\newblock {\em Duke Math. J.}, 94(1):41--77, 1998.
\bibitem[STW14]{szpiro-semi}
L.~Szpiro, M.~Tepper, and P.~Williams.
\newblock Semi-stable reduction implies minimality of the resultant.
\newblock {\em J. Algebra}, 397:489--498, 2014.
\bibitem[Tsu86]{tsuyumine-g3}
S.~Tsuyumine.
\newblock On {S}iegel modular forms of degree three.
\newblock {\em Amer. J. Math.}, 108(4):755--862, 1986.
\bibitem[Yuk86]{yukie}
A.~Yukie.
\newblock {\em A{pplications} {of} {Equivariant} {Morse} {Stratifications}
({Geometric} {Invariant} {Theory}, {Eisenstein} {Series})}.
\newblock ProQuest LLC, Ann Arbor, MI, 1986.
\newblock Thesis (Ph.D.)--Harvard University.
\end{thebibliography}
\end{document}