%
\section{Introduction}\label{sec:1}

\subsection{The model and the main result}\label{ssec:1.1}

Three-dimensional U(1) lattice gauge theory is the simplest gauge theory in
which static charges are expected to be confined at every coupling. By
Polyakov's mechanism \cite{Po77,BMK77}, magnetic monopoles form a plasma that
screens the magnetic field at the Debye length and squeezes the electric flux
between two static charges into a tube of finite tension. G\"opfert and Mack
proved this for the Villain action \cite{GM82}. In this paper we prove it for the
Wilson action.

\begin{definition}[Wilson and Villain measures and the Wilson loop]\label{def:1.1}
A finite box $\Lam\subset\Zthree$ is a cell complex with vertex set $\Lam$,
bonds (oriented edges) $e$ and plaquettes $p$ (free boundary conditions
throughout); for $\theta\in(\RR/2\pi\ZZ)^{E(\Lam)}$ and an oriented plaquette
$p$, $\theta_p$ is the sum of $\theta_e$ around $\partial p$.
\begin{enumerate}
\item The \emph{Wilson measure} at inverse coupling $\beta>0$ and the
\emph{Villain weight} at coupling $\tau>0$ are
\begin{equation}\label{eq:1.wilson}
\mu^{W}_{\Lam,\beta}(d\theta)=\frac{1}{Z^{W}_{\Lam,\beta}}\prod_{p}e^{\beta\cos\theta_p}\prod_{e}d\theta_e ,\qquad
\vt_{\tau}(\theta)\propto\sum_{m\in\ZZ}e^{-\tau(\theta-2\pi m)^2/2};
\end{equation}
the Villain measure $\mu^{V}_{\Lam,\tau}$ is $\mu^{W}_{\Lam,\beta}$ with
$e^{\beta\cos\theta_p}$ replaced by $\vt_\tau(\theta_p)$.
\item For a rectangle $C=C_{R,T}$ of sides $R,T\in\NN$ in a coordinate plane,
an oriented closed lattice path, $\theta_C:=\sum_{e\in C}\theta_e$, and the
Wilson loop expectation is $\ex{\cos\theta_C}^{W}_{\Lam,\beta}$.
\item The model has an \emph{area law at coupling $\beta$} if there are
$\sigma>0$, $c<\infty$ and $R_1<\infty$ such that
$\ex{\cos\theta_C}^{W}_{\Lam,\beta}\le\exp(-\sigma RT+c(R+T))$ for every
rectangle $C$ with $\min(R,T)\ge R_1$ and every finite box $\Lam\supseteq C$.
\end{enumerate}
\end{definition}

This area law, uniform in the box and with an explicit perimeter term, is
stronger than positivity of the iterated-limit string tension $\alpha$ of
\cite[(1.7)]{GM82}, whose finite-loop estimate carries the loss term $P_F$ of
\cite[(7.7)]{GM82}. Theorem~\ref{thm:2.1}(c) shows that the infinite-volume limit exists
and that the area law passes to it.

\begin{theorem}[Main theorem]\label{thm:1.2}
For every $\beta>0$ there are $\sigma(\beta)>0$ and $\Rone(\beta)<\infty$ such
that for every rectangle $C$ of sides $R,T$ with $\min(R,T)\ge\Rone(\beta)$
and every finite box $\Lam\supseteq C$,
\begin{equation}\label{eq:1.main}
\ex{\cos\theta_C}^{W}_{\Lam,\beta}\le\exp\bigl(-\sigma(\beta)RT+c(R+T)\bigr),
\end{equation}
where $c$ is an absolute constant. Moreover, for every $K>\Kst:=377.38$ there is
$\beta_0(K)<\infty$ such that, for every $\beta\ge\beta_0(K)$, $\sigma(\beta)$
and $\Rone(\beta)$ can be chosen, depending on $K$, so that \eqref{eq:1.main}
holds and
\begin{equation}\label{eq:1.sigma}
\sigma(\beta)\ge1.35\cdot10^{-4}\,\cg\cL\,(K\beta)^{-1/2}e^{-2.49436\,K\beta},
\end{equation}
where $\cg,\cL$ are the fixed regularization constants of
Section~\ref{app:D}; for $\beta<\beta_0(K)$, \eqref{eq:1.main} then holds with
$\sigma(\beta)$ and $\Rone(\beta)$ equal to the values so chosen at
$\beta_0(K)$.
\end{theorem}

The first assertion follows from the second with one fixed $K>\Kst$. In the first assertion the
absoluteness of $c$ is no restriction: if \eqref{eq:1.main} holds with a constant $c'=c'(\beta)$, then it
holds with $c=0$ and with $\sigma(\beta)/2$ and $\max(\Rone(\beta),4c'/\sigma(\beta))$ in place of
$\sigma(\beta)$ and $\Rone(\beta)$, since $c'(R+T)\le2c'\max(R,T)\le\frac12\sigma(\beta)RT$ when
$\min(R,T)\ge4c'/\sigma(\beta)$. Its content is the uniformity in $\Lam$; the absolute $c$ matters
together with \eqref{eq:1.sigma}. With
$\cg=1/4$ and $\cL=1.6\cdot10^{-19}$ (Section~\ref{ssec:D.1}),
\eqref{eq:1.sigma} reads
$\sigma(\beta)\ge5.4\cdot10^{-24}(K\beta)^{-1/2}e^{-2.49436K\beta}$
(Section~\ref{ssec:D.4}).

\begin{corollary}[Groups containing the scalars, with or without Higgs fields blind to the centre]\label{cor:1.5}\label{cor:1.3}
Let $G\subseteq U(n)$ be a closed subgroup with $\{zI:\abs{z}=1\}\subseteq G$,
in $d=3$ with the Wilson action $\beta\ReP\Tr U_p$ (free boundary conditions),
with or without a Higgs field coupled through any action that depends on the
links only through $\rho(U_e)$, $\rho$ a representation of $G$ trivial on the
scalars. At every $\beta>0$ and all Higgs couplings, for every rectangle $C$ with
$\min(R,T)\ge\Rone(2n\beta)$ and every finite box $\Lam\supseteq C$,
$\bigl|\ex{\Tr\textstyle\prod_{e\in C}U_e}_{G,\beta}\bigr|\le2n\exp\bigl(-\sigma(2n\beta)RT+c(R+T)\bigr)$;
in particular for U(2) with an adjoint Higgs field.
\end{corollary}

\begin{corollary}[Georgi\textendash Glashow model at infinite Higgs coupling]\label{cor:1.4}
The SU(2) lattice gauge theory with the Wilson action $\beta\cdot\frac12\Tr U_p$
and a unit adjoint Higgs field at hopping coupling $\kappa=\infty$, defined in
each finite box as the limit $\kappa\to\infty$, is the U(1) Wilson model at
coupling $\beta$, and it obeys the area law \eqref{eq:1.main} at every
$\beta>0$.
\end{corollary}

\subsection{Background}\label{ssec:1.2}
\emph{The Villain action.} G\"opfert and Mack \cite{GM82} made Polyakov's
mechanism rigorous for the Villain action. By duality the model is a Coulomb
gas of monopoles, which they control by an iterated Mayer expansion down to
the Debye scale \cite{GM81} and a single-scale expansion built on the analysis
of Debye screening of Brydges and Federbush \cite{BF80,B78}. Their Theorem~1
is the large-$\beta$ bound $\alpha\ge c\,m_D\beta^{-1}$ \cite[(1.9)]{GM82},
and all couplings follow \cite[Corollary~2]{GM82} by the 3-dimensional version
of Guth's inequality \cite[Lemma~4.1]{Gu80}. The expansions use the Gaussian
form of the dual weight: in \cite[Appendix~A]{GM82} the Villain action is chosen
explicitly, in place of their Wilson action (A.4).

\emph{The Wilson action.} In three dimensions the rigorous results are the area
law at strong coupling \cite{OS78}, for all compact groups, and logarithmic
bounds at all couplings, for U(1) \cite{GJ77} and for closed $G\subseteq U(n)$
containing the scalars \cite{Ch26}. Fr\"ohlich's comparison \cite{Fr79} bounds
the Wilson loop of such a group by a U(1) loop with the Wilson action
$A'=-\beta'\sum_p\cos(\theta_{\partial p})$ of \cite[(6)]{Fr79}, at coupling
$2n\beta$ \cite[(23),~(24)]{Fr79}. Thus an area law at every coupling for groups
with a central U(1) requires the Wilson case of U(1), not the Villain case.
This case is stated as known in \cite{CNS,Fo}, and so is a mass gap for all $\beta$
in \cite{Ni}. For the Wilson action, Cao, Park and Sheffield
\cite[open problem~(b)]{CPS25} state that when $d=3$ and $N=1$ the area law
holds for all $\beta$, citing \cite{GM82}, and that, since U(1) is the centre of
U($N$) for general $N$, Fr\"ohlich's comparison then gives the U($N$) area law for
all $\beta$ and all $N\ge1$. Chatterjee \cite[v2]{Ch26} leaves open whether
Fr\"ohlich's comparison can be combined with the Villain result of \cite{GM82}
to give an area law for U($n$) with the Wilson action, on the grounds that a
careful argument would need additional technology and a careful comparison of
the formulations and limits involved.

\emph{Related work.} The mixture of Lemma~\ref{lem:2.2} is used to prove lower bounds
for the XY model \cite[Lemma~9.6]{AHPS} and the annealed Villain model
\cite{DvEG}. Here we use it for an upper bound. The same Bessel dual weights and their
Gaussian mixture enter recent work on the critical planar XY model
\cite[\S2.1]{OAI216}. In four dimensions a Coulomb phase is
proved for the Villain \cite{Gu80,FS82} and the Wilson action \cite{Ki86}.

\subsection{Strategy}\label{ssec:1.3}
The proof has four steps.

\emph{(i) Comparison} (Section~\ref{sec:2}). Lemma~\ref{lem:2.2} expresses
$e^{\beta\cos\theta}/I_0(\beta)$ as a mixture of Villain
weights $\vt_{1/u}$, by a theorem of Hartman and
Watson \cite{HW74}, and Lemma~\ref{lem:2.3} shows that Villain loops are nondecreasing
in each coupling. We raise the couplings below
$\tau=K\beta$ to $\tau$ and the others to $\infty$, and Lemmas~\ref{lem:2.4} and~\ref{lem:2.6}
allow us to dominate the set of
the latter by a Bernoulli set of density $\delta=e^{-\cdel(K)\beta}$, where
$\cdel(K)/K\to\infty$. In this way Theorem~\ref{thm:2.5} bounds the
Wilson loop at $\beta$ by the average over a Bernoulli set $D$ of the Villain
loop at $\tau$ with $D$ frozen. Proposition~\ref{prop:2.9} places this Villain loop
in one box $\Lamo$ of
the size of the loop, where frozen plaquettes are
insulating walls (Proposition~\ref{prop:2.10}).

\emph{(ii) Regularization} (Sections~\ref{sec:3} and~\ref{sec:4}). A
deterministic construction maps each $D$ to a hole set and frozen sheets. It
freezes more plaquettes, which only increases the loop. The hole set of its first
phase is the smallest minimizer of a cut functional that penalizes blocks
crowded with marks. Thus each cost bound is a comparison with an explicit
competitor. A local seal removes thin pieces of blocks, and a second phase
enforces non-fragmentation, which is not local. The output, a slit domain $\Gp(D)$, has
uniform isoperimetry, cells of controlled size and shape with a Poincar\'e
inequality, and non-fragmentation, with constants independent of $\beta$, $K$,
$C$, $D$ (Theorems~\ref{thm:3.20}, \ref{thm:3.21}).

\emph{(iii) Deterministic area law} (Sections~\ref{sec:stage1}
and~\ref{sec:stage2}, Appendices~\ref{app:A} to~\ref{app:C}
and~\ref{app:F}). On each output $\Gp(D)$ we run the analysis of
\cite{GM82,GM81,BF80} with uniform constants, grading the outer partition
functions by the periods of their walls around the handles of
the insulating set. We adapt the steps we need from these sources (Section~\ref{ssec:1.4}),
and Appendix~\ref{app:A} proves a tree-graph counting
step of \cite{B78} for which the deduction given there is incomplete.
Theorem~\ref{thm:7.61} states that on each output the loop decays like the
exponential of the min-cut area
$\AG(C)$, in which frozen plaquettes are free.

\emph{(iv) Min-cut lower tail} (Section~\ref{sec:mincut}). A Peierls estimate
shows that $\AG(C)\ge(1-\etaT)RT$ outside an event of probability at most
$2^{4\abs{C}+5}e^{-RT}$, with $\etaT\to0$ as $K\to\infty$
(Theorem~\ref{thm:5.1}). We impose the condition $K>\Kst$ for this estimate alone
(Lemma~\ref{lem:2.6} and the remark after it).

In Section~\ref{sec:8} we combine the four steps into the following chain: for
$\beta\ge\beta_0(K)$, with $D\sim\Ber(\delta)$ on the interior plaquettes of
$\Lamo$,
\begin{align}
\ex{\cos\theta_C}^{W}_{\Lam,\beta}
&\le\EE_D\,\gfix_{\Lamo,\tau}(C;D)\label{eq:8.1}\\
&\le\EE_D\,W_{\Gp(D)}(C)\label{eq:8.2}\\
&\le\EE_D\exp\bigl(-\sigV A_{\Gp(D)}(C)+\sigV\etaS RT+c_N(R+T)\bigr)\label{eq:8.3}\\
&=\exp\bigl(-\sigV RT+\sigV\etaS RT+c_N(R+T)\bigr)\,\EE_D\exp\bigl(\sigV(RT-A_{\Gp(D)}(C))\bigr)\label{eq:8.4}\\
&\le\exp\bigl(-\sigV(1-\etaT-\etaS)RT+c_N(R+T)+4\abs{C}+4\bigr),\label{eq:8.5}
\end{align}
and Theorem~\ref{thm:2.1} extends it to all $\beta$.

Two natural routes fail. First, the Wilson dual weight $I_n(\beta)/I_0(\beta)$,
of order $e^{-\abs{n}\log\abs{n}(1+o(1))}$, is dominated by no Villain dual
weight $e^{-n^2/(2t)}$. Second, on the carpet graphs of \cite{CG25}, where the
Villain action is a limit of Wilson actions, the Wilson model is the limit
with the auxiliary plaquettes frozen, and hence Ginibre's inequality bounds the
Wilson loop from below, not from above.
In Appendix~\ref{app:G} we show that the first obstruction persists on every
finite box: for the Wilson action the charge-$k$ loop around a plaquette is at
least $I_k(\beta)/I_0(\beta)$, while every Villain mixture with couplings at
most $\tau$ gives at most $e^{-\kappa k^2/(2\tau)}$, where $\kappa>0$ depends
only on the box and the plaquette (Theorem~\ref{thm:G.2}). Hence no
comparison that holds for all charges at once, such as those obtained from
correlation inequalities, bounds Wilson loops by Villain loops with bounded
couplings, even with a perimeter factor (Corollary~\ref{cor:G.4}), and this is
why Theorem~\ref{thm:2.5} keeps a random set of frozen plaquettes
(Remark~\ref{rem:G.7}).

\subsection{Conventions}\label{ssec:1.4}
\emph{Dependence between sections.} Sections~\ref{sec:2} and~\ref{sec:8} need one fact about the
construction, stated in Theorem~\ref{thm:3.20}: the output is a deterministic function of
$D$ with $D\subseteq\FF$ of \eqref{eq:2.FF}. Sections~\ref{sec:stage1}
and~\ref{sec:stage2} can be read independently of the construction, given $\AG(C)$ and the
items listed after Theorem~\ref{thm:7.61},
namely (R2), (R5) to (R10), (N3), Lemma~\ref{lem:4.5}, the frame of Section~\ref{ssec:3.1}
and the coarse alignment of the output. Section~\ref{sec:mincut} relies on the construction of
Section~\ref{sec:3} and on the history bounds of Theorem~\ref{thm:3.21}. The properties (G1) to (G5) contain all
that we use about the loop-adapted grid after its definition, and (R5) to (R8) all that later sections use
about the cells and interfaces of the output. Appendix~\ref{app:G} uses only Section~\ref{sec:2}, and no
other part of the paper uses it.

\emph{Couplings.} We write $\beta$ for the Wilson coupling, $\tau=K\beta$ for the Villain
coupling of the comparison, and $\bGM=4\pi^2\tau$ for the coupling in the
normalization of \cite{GM82}, which we use in Sections~\ref{sec:stage1} and~\ref{sec:stage2} and in
Appendices~\ref{app:A} to~\ref{app:C}.

\emph{Constants and units.} We fix the constants once, in the order of
Section~\ref{app:D}. Those fixed before $K$ do not depend on $\beta$, $\tau$,
$K$, $C$ or $D$, and $c$, $C$ with or without indices denote such constants,
except where a dependence is shown or stated. This is the case for $\cdel(K)$, which also
depends on $\beta$, for $c_g'$ and $c_u$, which depend on $\beta$ through the size and the rounding
of $L_c$, and for $\cdep$. The letter $C$ also denotes the loop.
One site is one coarse cell of the loop-adapted grid in Sections~\ref{sec:3}
to~\ref{sec:mincut}, unless ``fine'' is written, and the lattice spacing is one
in Sections~\ref{sec:stage1} and~\ref{sec:stage2} (Section~\ref{ssec:2.5}).

\emph{Supplement and editions.} Sections S1, S2 and S3 of the supplementary
material give the values of the constants with their arithmetic, full proofs
of steps sketched here, and numerical checks (Section~\ref{app:E}). Part~S4, a separate file of the
supplementary material, uses the sources only as a guide and gives complete proofs of the steps that the main
text takes from the literature with changes or presents in compressed form. Sections of the main text
that contain such proofs name the relevant sections of Part~S4 at their start, and the tables of changes,
such as Table~\ref{tab:6.subst}, are meant for readers who know the sources. Equation, section and page numbers of
\cite{GM82} and \cite{GM81} refer to the preprints \cite{GM82pre} (DESY
81-036) and \cite{GM81pre} (DESY 81-014), equation numbers of \cite{Fr79} to
its preprint IHES/P/79/02, and lemma numbers of \cite{Gu80} to its preprint
CLNS 79/439.