Recurrence for pretentious systems along generalized Pythagorean triples
Extracted source text · arXiv 2508.09778
\documentclass[11pt]{amsart}
%%\usepackage{hyperref}
%%\usepackage{amssymb,amsfonts}
%%\usepackage{amssymb,textcomp}
%%\usepackage{graphicx}
\usepackage[svgnames,x11names]{xcolor}
\usepackage{amssymb,amsfonts}
\usepackage[pagebackref,colorlinks,urlcolor={DarkOliveGreen4},
linkcolor={DarkSlateBlue},
citecolor={Chocolate4}]{hyperref}
\usepackage[capitalize]{cleveref}
\usepackage{amssymb,textcomp}
\usepackage{graphicx}
\numberwithin{equation}{section}
%%\usepackage{showkeys}
\newcommand{\1}{\mathbbm{1}}
\makeatletter
\def\ps@myplain{
\let\@oddhead\@empty
\let\@evenhead\@empty
\let\@oddfoot\@empty
\let\@evenfoot\@empty
\def\@oddhead{{\tiny\hfill\thepage}} % Tiny font, right-aligned
\def\@evenhead{{\tiny\hfill\thepage}} % Same for even pages
}
\makeatother
\pagestyle{myplain}
\usepackage{comment}
\usepackage[official]{eurosym}
\usepackage[greek,english]{babel}
\usepackage{bbm}
\usepackage[utf8]{inputenc}
\usepackage[T1]{fontenc}
\usepackage{pdfpages}
\usepackage{tensor}
\usepackage{enumerate}
\usepackage{relsize}
\usepackage[normalem]{ulem}
\usepackage[stable]{footmisc}
\usepackage{mathtools}
\usepackage[mathscr]{eucal}
\usepackage[a4paper, twoside=false, vmargin={2cm,3cm}, includehead]{geometry}
\usepackage{pgfgantt}
\usepackage{graphicx}
\usepackage{xcolor}
\ganttset{group/.append style={orange},
milestone/.append style={red},
progress label node anchor/.append style={text=red}}
\usepackage{cancel}
\usepackage{amsmath}
%
\usepackage{enumerate}
\usepackage[utf8]{inputenc}
\usepackage[T1]{fontenc}
\usepackage[mathscr]{eucal}
\subjclass{Primary: 37A44; Secondary: 05D10, 11N37.}
\keywords{Multiplicative actions, partition regularity, density regularity, multiple recurrence, generalized Pythagorean triples,
multiplicative functions, concentration inequalities}
%%\usepackage[svgnames,x11names]{xcolor}
%%\hypersetup{ -
%% colorlinks, linkcolor={DarkSlateBlue},%{MidnightBlue},
%% %RoyalBlue4},
%%citecolor={Chocolate4}, urlcolor={DarkOliveGreen4}
%%}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% A4 Paper
\setcounter{tocdepth}{1}
% Theorems
\newtheorem{lemma}{Lemma}[section]
\newtheorem{theorem}[lemma]{Theorem}
\newtheorem*{theorem*}{Theorem}
\newtheorem{corollary}[lemma]{Corollary}
\newtheorem{question}{Question}
\newtheorem*{question*}{Question}
\newtheorem{proposition}[lemma]{Proposition}
\newtheorem*{proposition*}{Proposition}
\newtheorem*{MCP}{Multiplicative correspondence principle}
\newtheorem{conjecture}{Conjecture}
\newtheorem{problem}{Problem}
\newtheorem*{problem*}{Problem}
\theoremstyle{definition}
\newtheorem{definition}{Definition}[section]
\newtheorem*{claim*}{Claim}
\newtheorem*{convention}{Convention}
\newtheorem*{notation}{Notation}
\newtheorem{rmk}{Remark}
\newtheorem*{equivdefinition}{Equivalent Definition}
\newtheorem*{example}{Example}
\newtheorem*{examples}{Examples}
\newtheorem*{remark}{Remark}
\newtheorem*{remarks}{Remarks}
%NumberTheory
\newcommand{\legendre}[2]{\ensuremath{\left( \frac{#1}{#2} \right) }}
% Enumerations
\renewcommand{\theenumi}{\roman{enumi}}
\renewcommand{\labelenumi}{{\rm (\roman{enumi})}}
\renewcommand{\theenumii}{{\rm \alph{enumii}}}
\newcommand{\rem}[1]{{\bf $\clubsuit$ #1 $\clubsuit$}}
% Blackboard letters
\DeclareMathOperator{\Dlim}{D-lim}
\newcommand{\lE}{\mathbb{E}^{\log}}
\newcommand{\llE}{\mathbb{E}^{\log\log}}
\newcommand{\C}{{\mathbb C}}
\newcommand{\E}{{\mathbb E}}
\newcommand{\D}{{\mathbb D}}
\newcommand{\F}{{\mathbb F}}
\newcommand{\N}{{\mathbb N}}
\renewcommand{\P}{{\mathbb P}}
\newcommand{\Q}{{\mathbb Q}}
\newcommand{\R}{{\mathbb R}}
\renewcommand{\S}{\mathbb{S}}
\newcommand{\T}{{\mathbb T}}
\newcommand{\Z}{{\mathbb Z}}
\newcommand{\U}{{\mathbb U}}
\newcommand{\X}{{\mathcal{X}}}
\newcommand{\cesE}{\mathlarger{\mathbb{E}}}
\newcommand{\inflim}{\underline{\lim}}
\newcommand{\suplim}{\overline{\lim}}
% Calligraphic letters
\newcommand{\CA}{{\mathcal A}}
\newcommand{\CC}{{\mathcal C}}
\newcommand{\CE}{{\mathcal E}}
\newcommand{\CF}{{\mathcal F}}
\newcommand{\CI}{{\mathcal I}}
\newcommand{\CK}{{\mathcal K}}
\newcommand{\CM}{{\mathcal M}}
\newcommand{\CN}{{\mathcal N}}
\newcommand{\CO}{{\mathcal O}}
\newcommand{\CP}{{\mathcal P}}
\newcommand{\CQ}{{\mathcal Q}}
\newcommand{\CX}{{\mathcal X}}
\newcommand{\CY}{{\mathcal Y}}
\newcommand{\CW}{{\mathcal W}}
\newcommand{\CZ}{{\mathcal Z}}
\newcommand{\M}{{\mathcal{M}}}
\newcommand{\A}{{\mathcal{A}}}
\newcommand{\Oh}{{\rm O}}
\newcommand{\oh}{{\rm o}}
% Bold letters
\newcommand{\bm}{{\mathbf{m}}}
\newcommand{\bn}{{\mathbf{n}}}
\newcommand{\be}{{\mathbf{e}}}
\newcommand{\bg}{{\mathbf{g}}}
\newcommand{\bh}{{\mathbf{h}}}
\newcommand{\bk}{{\mathbf{k}}}
\newcommand{\bt}{{\mathbf{t}}}
\newcommand{\bu}{{\mathbf{u}}}
\newcommand{\bv}{{\mathbf{v}}}
\newcommand{\bx}{{\mathbf{x}}}
\newcommand{\by}{{\mathbf{y}}}
\newcommand{\bz}{{\mathbf{z}}}
\newcommand{\bI}{{\mathbf{I}}}
\newcommand{\bM}{{\mathbf{M}}}
\newcommand{\bN}{{\mathbf{N}}}
\newcommand{\balpha}{{\boldsymbol{\alpha}}}
\newcommand{\bepsilon}{{\boldsymbol{\epsilon}}}
\newcommand{\bgamma}{{\boldsymbol{\gamma}}}
\newcommand{\bdelta}{{\boldsymbol{\delta}}}
\newcommand{\bzero}{{\boldsymbol{0}}}
\newcommand{\btheta}{{\boldsymbol{\theta}}}
\newcommand{\beeta}{{\boldsymbol{\eta}}}
\newcommand{\bbeta}{{\boldsymbol{\beta}}}
\newcommand{\uh}{{\underline{h}}}
\newcommand{\ua}{{\underline{a}}}
\newcommand{\ub}{{\underline{b}}}
\newcommand{\up}{{\underline{p}}}
\newcommand{\uq}{{\underline{q}}}
\newcommand{\ur}{{\underline{r}}}
% Various abbreviations
\newcommand{\wh}{\widehat}
\newcommand{\wt}{\widetilde}
\newcommand{\e}{\varepsilon}% for the function exp
\newcommand{\one}{\mathbf{1}}
\newcommand{\Var}{\operatorname{Var}}
\newcommand{\Cov}{\operatorname{Cov}}
\newcommand{\Folner}{F\o{}lner}
\newcommand{\ZN}{\Z_{\widetilde N}}
\newcommand{\tN}{{\widetilde N}}
\newcommand{\tM}{{\widetilde M}}
\newcommand{\CCM}{{\CM_1^c}}
\newcommand{\tZN}{\Z_\tN}
\newcommand{\norm}[1]{\left\Vert #1\right\Vert}
\newcommand{\nnorm}[1]{\lvert\!|\!| #1|\!|\!\rvert}
\newcommand{\lip}{{\text{\rm Lip}}}
\newcommand{\inv}{^{-1}}
\newcommand{\st}{{\text{\rm st}}}
\newcommand{\un}{{\text{\rm un}}}
\newcommand{\er}{{\text{\rm er}}}
\DeclareMathOperator{\spec}{Spec}
\DeclareMathOperator{\id}{id}
\DeclareMathOperator{\Span}{Span}
\newcommand{\dmult}{d_\textrm{mult}}
\DeclareMathOperator{\reel}{Re}
%%\renewcommand{\Re}{\reel}
\DeclareMathOperator{\Imag}{Im}
\renewcommand{\Im}{\Imag}
\DeclareMathOperator{\poly}{poly}
\newcommand{\Krat}{{\CK_{\text{\rm rat}}}}
%\newcommand{\rem}[1]{{\bf $\clubsuit$ #1 $\clubsuit$}}
\title[Recurrence for pretentious systems along generalized Pythagorean triple]{Recurrence for pretentious systems along generalized Pythagorean triples}
\date{\today}
\author{Nikos Frantzikinakis and Andreas Mountakis}
\address[Nikos Frantzikinakis]{University of Crete, Department of mathematics and applied mathematics, Voutes University Campus, Heraklion 70013, Greece} \email{frantzikinakis@gmail.com}
\address[Andreas Mountakis]{University of Crete, Department of mathematics and applied mathematics, Voutes University Campus, Heraklion 70013, Greece} \email{a.mountakis@uoc.gr}
\begin{document}
\vspace*{-0.015cm}
\begin{abstract}
We establish multiple recurrence results for pretentious measure-preserving multiplicative
actions along generalized Pythagorean triples, that is, solutions to the equation $ax^2 + b y^2 = c z^2$.
This confirms the ergodic-theoretic form of the generalized Pythagorean partition regularity
conjecture in this critical case of structured measure-preserving actions. As a consequence of our main theorem, any finite coloring
of $\mathbb{N}$ generated by the level sets of finitely many pretentious completely
multiplicative functions, must contain a monochromatic generalized Pythagorean triple.
\end{abstract}
\thanks{The authors were supported by the Hellenic Foundation for Research
and Innovation ELIDEK HFRI-NextGenerationEU-15689.}
\maketitle
\tableofcontents
\section{Introduction and main results}
\subsection{Introduction}
A central principle in arithmetic Ramsey theory asserts that any finite
partition of the natural numbers must contain at least one cell exhibiting
significant internal structure. Here, by structure we mean the
existence of nontrivial solutions to algebraic equations or systems of
equations, with all variables lying in the same cell of the partition.
\begin{definition}
An equation $P(x, y, z) = 0$ is \emph{partition regular} if, for every finite
partition of $\mathbb{N}$, there exist distinct $x, y, z$ in the same cell that
satisfy the equation.
\end{definition}
The partition regularity of linear equations is classically illustrated by the
results of Schur~\cite{schur_thm} and van der Waerden~\cite{van_der_Waerden_thm},
and was substantially
extended by Rado~\cite{Rado-original}, who gave a complete classification of
systems of linear equations that are partition regular.
\begin{definition}
We say that $(a,b,c) \in \mathbb{N}^3$ is a \emph{Rado triple} if $a = c$,
$b = c$, or $a + b = c$.
\end{definition}
In the case of a single linear equation, Rado’s criterion asserts that
\begin{equation}\label{E:RadoLinear}
a x + b y = c z
\end{equation}
is partition regular if and only if $(a,b,c)$ is a Rado triple.
Polynomial equations pose far greater challenges in the study of partition
regularity. The most prominent example is the long-standing open problem of
Erd\H{o}s and Graham~\cite{Graham-prob-1, Graham-prob-2}, which asks whether the
Pythagorean equation $x^2 + y^2 = z^2$ is partition regular.
More generally, one may ask about the partition regularity of broader classes
of homogeneous quadratic equations in three variables. This problem captures
the essential structural difficulties of the nonlinear setting and has become
a focus of considerable recent research. In this direction we state the
following conjecture:
\begin{conjecture}\label{Rado}
If $a, b, c \in \mathbb{N}$, then
\begin{equation}\label{gen-pyth-eq}
a x^2 + b y^2 = c z^2
\end{equation}
is partition regular if and only if $(a,b,c)$ is a Rado triple.
\end{conjecture}
The necessity of the Rado triple condition follows from the fact that
partition regularity of \eqref{gen-pyth-eq} implies partition regularity of its
linear counterpart \eqref{E:RadoLinear}. The sufficiency, however, is far less
understood: no instance of a Rado triple $(a,b,c)$ is currently known for which
partition regularity of \eqref{gen-pyth-eq} has been established. Depending on
the coefficients, the problem exhibits a range of distinct and often
deeply challenging behaviors.
In \cite{Fra-Host-structure-multiplicative}, among other results, the authors
prove partition regularity with respect to the pair $(x,y)$ for the equation
$16x^2 + 9y^2 = z^2$; that is, for every finite partition of $\mathbb{N}$ there
exist distinct $x, y$ in the same cell and $z \in \mathbb{N}$ such that the
equation is satisfied. In \cite{Fra-Klu-Mor}, partition regularity with respect
to $(x,y)$ and with respect to $(y,z)$ is established for the Pythagorean equation
$x^2 + y^2 = z^2$. This result was later extended in \cite{fra-klu-mor-2} to
partition regularity of \eqref{gen-pyth-eq} with respect to $(x,y)$ and $(y,z)$
whenever $a = c$ or $b=c$. For further results on the partition regularity of quadratic
equations, see
\cite{bergelson-ergodic-ramsey, browning-prendiville-roth, chapman-part-reg,
chow-lindqvist-prendiville, csikvari-gyarmati-sarkozy, dinasso-luperi,
green-lindqvist, heule-kullmann-marek, khalfalah-szemeredi, moreira-monochromatic}.
The techniques developed in
\cite{Fra-Host-structure-multiplicative, fra-klu-mor-2, Fra-Klu-Mor} are
inherently limited to pair configurations and do not extend to partition
regularity problems involving triples. To overcome this obstacle, we employ the
framework of ergodic theory to formulate sufficient conditions for the partition
regularity of \eqref{gen-pyth-eq}.
\begin{definition}
A \emph{multiplicative action} is a quadruple $(X, \mathcal{X}, \mu, T_n)$,
where $(X, \mathcal{X}, \mu)$ is a probability space and, for
$n \in \mathbb{N}$, the transformations $T_n \colon X \to X$ are invertible,
measure-preserving, and satisfy $T_1 = \mathrm{id}$ and
$T_m \circ T_n = T_{mn}$ for all $m, n \in \mathbb{N}$. This uniquely
determines a $(\mathbb{Q}_{>0}, \times)$-action on $(X, \mathcal{X}, \mu)$
defined by $T_{m/n} := T_m \circ T_n^{-1}$. The action is said to be
\emph{finitely generated} if the set $\{T_p : p \in \mathbb{P}\}$ is finite.
\end{definition}
Using a variant of Furstenberg’s correspondence
principle~\cite{furstenberg_szemeredi_thm} for
dilation-invariant densities from \cite{Bergelson_multiplicative},
together with the fact that
partition regularity of homogeneous equations follows from the existence of
solutions in every multiplicatively syndetic subset of the integers,
Conjecture~\ref{Rado} reduces to the following multiple recurrence statement:
\begin{conjecture}[Ergodic variant of Conjecture~\ref{Rado}]\label{RadoErgodic}
Let $(a,b,c)$ be a Rado triple. Then for every multiplicative action
$(X,\mathcal{X},\mu,T_n)$ and every set $A \in \mathcal{X}$ with
$X = \bigcup_{j=1}^k T_j^{-1} A$ for some $k \in \mathbb{N}$, there exist
distinct $x, y, z \in \mathbb{N}$ such that
$$
\mu\bigl(T_x^{-1} A \cap T_y^{-1} A \cap T_z^{-1} A\bigr) > 0
\quad\text{and}\quad
a x^2 + b y^2 = c z^2.
$$
\end{conjecture}
If $a + b \neq c$, simple examples show this multiple recurrence can fail for certain positive measure sets $A$. For instance, if $a=b=c=1$, take the multiplicative action by dilations on $\mathbb{T}$ with the Haar measure $m_{\mathbb{T}}$, given by $T_n x = n^2 x \pmod{1}$, and $A = [1/3, 2/3]$.
\subsection{Main results}
In this paper we establish Conjecture~\ref{RadoErgodic} for a
class of structured systems the \emph{pretentious multiplicative actions}. Although not the
most technically demanding case, they are the most plausible setting for
counterexamples to nonlinear partition regularity of homogeneous equations,
making them a critical test case. Verifying the
conjecture in this setting provides strong evidence for both
Conjectures~\ref{Rado} and~\ref{RadoErgodic}.
Pretentious systems in multiplicative dynamics play a role analogous to that of
rational Kronecker systems in the additive setting. The analogy is useful but
imperfect: in Kronecker systems, multiple recurrence is typically straightforward,
whereas pretentious systems possess a richer structure, making their analysis
substantially more delicate. The notation for the next definition is given in
Sections~\ref{SS:mf} and~\ref{SS:factors}.
\begin{definition}
A multiplicative action $(X,\CX,\mu,T_n)$ is \emph{pretentious} if the spectral
measure of every $F \in L^2(\mu)$ is supported on the set of pretentious
multiplicative functions.\footnote{Given a measure space $(X, \X, \mu)$, we say
that $\mu$ is supported on $E \in \X$ if $\mu(X \setminus E) = 0$.}
\end{definition}
The simplest example of a pretentious multiplicative action is a rotation by a
pretentious multiplicative function: if $f \colon \N \to \S^1$ is pretentious
and $\S^1$ carries the Haar measure, define $T_n \colon \S^1 \to \S^1$ by
$T_n z := f(n) \cdot z$. If $(X,\CX,\mu,T_n)$ is finitely generated, Charamaras
showed in \cite{Charamaras-multiplicative} that pretentious systems are spanned by
``pretentious
rational eigenfunctions''. This fails for general pretentious multiplicative
actions, as shown in \cite{Fra-Hom_Triples}.
Our main result, stated below, remains nearly as difficult to prove even in the
simpler setting of multiplicative rotations by several pretentious multiplicative
functions. We choose to work in the broader framework of pretentious multiplicative actions,
as this is likely to be a necessary intermediate step toward establishing
Conjecture~\ref{RadoErgodic}. Further discussion of this perspective is given
in Section~\ref{SS:Problem}.
\begin{theorem}\label{T:Main}
Let $(a,b,c)$ be a Rado triple. For every pretentious multiplicative action
$(X,\CX,\mu,T_n)$ and every set $A \in \CX$ with $\mu(A) > 0$, the following
holds: for any $\varepsilon > 0$ there exist distinct $x, y, z \in \N$ such that
$$
\mu\bigl(A \cap T^{-1}_x A \cap T^{-1}_y A \cap T^{-1}_z A\bigr)
\geq (\mu(A))^4 - \varepsilon
\quad \text{and} \quad a x^2 + b y^2 = c z^2.
$$
\end{theorem}
\begin{rmk}
Our argument extends, without substantive changes, to the setting of three
pretentious multiplicative actions $T_n, S_n, R_n$ acting on the same
probability space $(X,\CX,\mu)$, without assuming commutativity. Moreover, we
show that the conclusion holds for many triples $x, y, z \in \N$; precise
formulations appear in Theorems~\ref{T: main a=c} and \ref{T: main a+b=c}.
For a general multiplicative action $(X,\CX,\mu,T_n)$, the functions
$F \in L^2(\mu)$ whose spectral measure is supported on pretentious
multiplicative functions form the \emph{pretentious factor} of the system (see
\cite[Theorem~2.4]{Fra-Hom_Triples}). Theorem~\ref{T:Main} therefore establishes
multiple recurrence along generalized Pythagorean triples within this factor.
\end{rmk}
We deduce the following corollary, which confirms
Conjecture~\ref{Rado} in the case where the partition is defined by the level
sets of finitely many pretentious multiplicative functions.
\begin{corollary}
Let $(a,b,c)$ be a Rado triple, and let $f_1, \ldots, f_s \colon \N \to \S^1$
be pretentious completely multiplicative functions. Then for every open arc $I$
of $\S^1$ containing $1$ there exist distinct $x, y, z \in \N$ such that
$$
a x^2 + b y^2 = c z^2 \quad \text{and} \quad
f_j(x), f_j(y), f_j(z) \in I \quad \text{for } j = 1, \ldots, s.
$$
\end{corollary}
\begin{proof}
Let
\begin{equation}\label{E:Id}
I_\delta := \{ e^{2\pi i t} \colon t \in (-\delta, \delta) \}
\end{equation}
for some
$\delta \in (0,1/4)$ chosen small enough so that $I_{2\delta} \subset I$.
Let $X:= (\S^1)^s$, let $\CX$ be the Borel $\sigma$-algebra of $X$, and
let $\mu$ denote the Haar measure on $X$. For $n \in \N$ and $z_1,\ldots, z_s\in \S^1$, define
$$
T_n(z_1, \ldots, z_s) := (f_1(n) z_1, \ldots, f_s(n) z_s).
$$
We claim that $(X,\CX,\mu,T_n)$ is a pretentious multiplicative action.
Indeed, the spectral measure of each character of $X$ is supported on the set
$$
\{ f_1^{k_1} \cdots f_s^{k_s} \colon k_1,\ldots,k_s \in \Z \},
$$
which consists entirely of pretentious, completely multiplicative functions.
Since linear combinations of characters are dense in $L^2(\mu)$, and the class of
functions in $L^2(\mu)$ whose spectral measures are supported entirely on
pretentious multiplicative functions is closed, the claim follows.
Let $A := I_\delta \times \cdots \times I_\delta$. Then $\mu(A) > 0$, and
Theorem~\ref{T:Main} provides distinct $x,y,z \in \N$ such that
$$
\mu(A \cap T_x^{-1}A \cap T_y^{-1}A \cap T_z^{-1}A) > 0.
$$
Consequently, for each $j=1,\ldots,s$ we have
$$
I_\delta \cap (f_j(x)^{-1} I_\delta) \cap (f_j(y)^{-1} I_\delta) \cap
(f_j(z)^{-1} I_\delta) \neq \emptyset,
$$
and hence
$$
f_j(x), \ f_j(y), \ f_j(z) \in I_{2\delta} \subset I
\quad \text{for } j=1,\ldots,s.
$$
This completes the proof.
\end{proof}
\subsection{Proof strategy}
We begin with a rough outline of our general proof strategy; for the
Pythagorean equation, a more detailed sketch appears in
Section~\ref{Sec_proof_plan}.
To prove Theorem~\ref{T:Main}, we work with solutions of
$ a x^2 + b y^2 = c z^2 $ in the parametric form
$$
x = k \, P_1(m,n), \quad y = k\, P_2(m,n), \quad z = k \, P_3(m,n),
\quad k, m, n \in \N,
$$
where $P_j$, $j \in \{1,2,3\}$, are binary quadratic forms that will be explicitly described later.
The key averages to analyze are
(averaging notation is given in Section~\ref{SS:Notation}):
\begin{equation}\label{proof_strategy_eq_1}
\cesE_{(m,n) \in [N]^2}
\int F \cdot T_{P_1(m,n)}F \cdot T_{P_2(m,n)}F \cdot T_{P_3(m,n)}F \, d\mu.
\end{equation}
When the action is ergodic and finitely generated, we restrict the averaging
to a two-dimensional grid $(Qm+1, Qn)$, where $Q \in \N$ is taken to be highly
divisible.
Applying the spectral theorem together with the concentration estimates for
multiplicative functions given in Propositions~\ref{linear_conc}
and~\ref{quad_conc}, we find that for at least two indices
$j \in \{1,2,3\}$, the term $T_{P_j(Qm+1,Qn)}F$ in
\eqref{proof_strategy_eq_1} can be replaced by $F$, while the third term can
be replaced either by $F$ itself or by $T_{aQn}F$ for some $a \in \N$.
Taking a multiplicative average over $Q$ then yields a grid for which the following lower bound is achieved
$$
\int F\, d\mu \cdot \int F^3 \, d\mu - \varepsilon
\ \geq\ \Big(\int F\, d\mu\Big)^4 - \varepsilon,
$$
completing the proof.
This outline describes the argument in the finitely generated case.
For general actions, two significant new difficulties arise:
\begin{enumerate}[(i)]
\item \emph{Archimedean characters.}
Non-concentration phenomena appear for functions whose spectral measures
are supported on Archimedean characters. As in earlier work, we address
this by introducing a cutoff $S_{\delta}$ (see \eqref{S_delta_def}) that,
for sufficiently small $\delta$, localizes the
Archimedean characters, so the corresponding component of the functions becomes concentrated around itself.
\item \emph{$1$-pretentious oscillating multiplicative functions.}
A greater challenge comes from functions whose spectral measures are
supported on $1$-pretentious oscillating multiplicative functions (see the example after \cref{linear_conc}).
Our new crucial idea here is to average over a grid $(Qm+1, Qn+v)$, carefully
choosing $Q$ and $v=v(Q_1,Q_2,Q_3)$ to satisfy congruences such as those in
\cref{lemma_choice_of_v_1}. This allows us to factor out highly divisible,
pairwise coprime terms $Q_j$ from each $P_j(Qm+1, Qn+v)$ for $j \in \{1,2,3\}$.
We then perform a triple multiplicative averaging over the $Q_j$,
thereby eliminating the contribution from the oscillating component of $F$.
The resulting bound is
$$
\int F \cdot F_1 \cdot F_2 \cdot F_3 \, d\mu - \varepsilon,
$$
where for $j \in \{1,2,3\}$ the functions $F_j$ are projections of $F$ onto suitable
factors. The proof is completed by applying the estimate in
\cref{chu-lemma}.
\end{enumerate}
The fact that $(a,b,c)$ is a Rado triple plays a crucial role in the
preceding arguments. It ensures that we can parametrize the solutions to
\eqref{gen-pyth-eq} so that at least two of the quadratic forms
$P_j$, $j \in \{1,2,3\}$, have leading coefficient $1$ with respect to the variable $m$.
This property is essential even in the finitely generated case, as it allows
us to replace $T_{P_j(Qm+1,Qn+v)}F$ by $F$ for at least two values of $j$.
In the general case, it is also used to guarantee that the sets $S_{\delta}$
have positive upper density.
To illustrate with a concrete example why our method fails
when $(a,b,c)$ is not a Rado triple, consider the equation
$x^2 + y^2 = 4 z^2$, which admits the parametrization
$$
x = k \, 2(m^2-n^2), \quad y = k \, 4 m n, \quad
z = k \, (m^2 + n^2), \quad k,m,n \in \N.
$$
This equation is partition regular with respect to each of the pairs
$(x,y)$, $(y,z)$, and $(x,z)$ (see \cite[Theorem~1.2]{fra-klu-mor-2}), but it
is not partition regular with respect to $(x,y,z)$ since $(1,1,4)$ is not a
Rado triple.
If the action is finitely generated and ergodic, our argument yields a lower
bound of the form
$$
\int F \, d\mu \cdot \int F^2 \cdot T_2 F \, d\mu,
$$
and the presence of the term $T_2F$ in place of $F$ prevents positivity, as
expected.
\subsection{An open problem}\label{SS:Problem}
For finitely generated actions, it appears plausible that the multiple
recurrence property in Conjecture~\ref{RadoErgodic} holds for every
set $A \in \mathcal{X}$ of positive measure.
In the special case of the Pythagorean equation, this would follow directly
from Theorem~\ref{T:Main} together with a positive solution to the problem
below, which is a special case of \cite[Problem~4]{Fra-Hom_Triples}.
\begin{problem*}
Let $(X,\mathcal{X}, \mu, T_n)$ be a finitely generated multiplicative action,
and let $F, G, H \in L^\infty(\mu)$. If either $F$ or $G$ is
orthogonal to the pretentious factor of the system,\footnote{Equivalently,
as shown in \cite[Proposition~5.6]{Fra-Hom_Triples}, the spectral measure of
this function is entirely supported on aperiodic completely multiplicative
functions.} then
$$
\lim_{N\to\infty} \cesE_{(m,n) \in [N]^2}
\int
T_{mn}F \cdot T_{m^2-n^2}G \cdot T_{m^2+n^2}H \, d\mu= 0.
$$
\end{problem*}
If one of the functions is constant, it follows from
\cite{Fra-Host-structure-multiplicative} that the problem admits a positive
answer for general actions, without assuming finite generation.
However, the action by dilations described after
Conjecture~\ref{RadoErgodic} shows that, in general, finite generation cannot
be omitted unless the orthogonality assumption is strengthened.
\subsection{Notational conventions}\label{SS:Notation}
We let $\N:=\{1, 2, \ldots\}$, $\N_0:=\{0, 1, 2, \ldots\}$, $\Q_{>0}:=\Q\cap (0,+\infty)$, $\mathbb{S}^1$
be the unit circle in $\C$, $\mathbb{U}$ the closed complex unit disk,
and $\T:=\R/\Z$ the torus. For $x\in \R$, we set $e(x):=e^{2\pi i x}$. For
$z\in \C$, we denote by $\Re(z)$ the real part of $z$. For $N\in \N$, we let $[N]:=\{1, \dots, N\}$.
We use $\P$ for the set of prime numbers, and throughout, $p$ denotes a
prime. For $p \in \P$, $\ell \in \N_0$, and $a\in \Z$, we say that
$p^{\ell}$ fully divides $a$, written $p^{\ell} \parallel a$, if
$p^{\ell} \mid a$ and $p^{\ell+1} \nmid a$. We denote this exponent $\ell$
by $\theta_p(a)$. The notation $(a_1,\ldots, a_k)$ stands for the greatest
common divisor of $a_1,\ldots, a_k$. The inverse of an element $a$ in
$\Z_{m}^{\times}$ is denoted by $a^{-1}$.
If $A$ is a finite non-empty set and $a\colon A\to \C$, we
define
$$
\E_{n\in A}\, a(n) := \frac{1}{|A|}\sum_{n\in A} a(n).
$$
For two functions $f$ and $g$, if there exists a constant $C>0$ such
that $|f(X)| \leq C |g(X)|$ for all $X$ sufficiently large, we write
$f(X) = \Oh(g(X))$ or $f(X) \ll g(X)$. When the constant $C$ depends on
a quantity $K$, we write $f(X) = \Oh_{K}(g(X))$ or
$f(X) \ll_{K} g(X)$. Given a collection of variables $X_1, \dots, X_r$ and $l_1, \dots, l_r \in
\R \cup \{\infty, -\infty\}$, we write
$\oh_{X_1, \dots, X_s, X_{s+1}\to l_{s+1}, \dots, X_r \to l_r}(1)$ for an
error term $\e(X_1, \dots, X_r)$ such that for each $X_1, \dots, X_s$
we have
$$
\lim_{X_{s+1}\to l_{s+1}} \limsup_{X_{s+2}\to l_{s+2}} \dots
\limsup_{X_{r}\to l_{r}} |\e(X_1, \dots, X_r)| = 0.
$$
Finally, we abbreviate
$$
\lim_{X_{1}\to l_1} \limsup_{X_{2}\to l_2} \dots \limsup_{X_{r}\to l_r}
\quad \text{by} \quad \lim_{X_1, \dots, X_r}
$$
when the value on the left is zero; and
$$
\limsup_{X_{1}\to l_1} \limsup_{X_{2}\to l_2} \dots \limsup_{X_{r}\to l_r}
\quad \text{by} \quad \suplim_{X_1, \dots, X_r}.
$$
\section{Background and preliminary results}\label{S:Background}
In this section we collect notions and results from number theory and ergodic
theory that are going to be useful for us later on.
\subsection{Multiplicative averaging}
In our setting, a binary quadratic form is a homogeneous polynomial $P(m,n)$
of degree $2$ with integer coefficients.
Recall that a binary quadratic form
$P(m,n)=\alpha m^2 + \beta mn + \gamma n^2\in \Z[m,n]$, where $\alpha,\beta, \gamma\in \Z$, is
irreducible if and only if its discriminant
$\Delta=\beta^2 -4 \alpha \gamma $ is not a perfect square.
Following \cite{fra-klu-mor-2}, given
a binary quadratic form $P$ and $r \in \N$, we define
\begin{equation}\label{omega_def}
\omega_{P}(r):=|\{ n \in \Z_r: P(1,n)\equiv 0 \pmod{r} \}|.
\end{equation}
In \cite{fra-klu-mor-2}, $\omega_P$ is defined using
$P(n,1)$ instead of $P(1,n)$, but the same results hold because of symmetry.
From \cite[Lemma 4.3]{fra-klu-mor-2} we know that if $P$ is irreducible,
then for all but finitely many $p\in \P$, $\omega_P(p)=0$ or $2$, and
$\sum_{p\in \P} \frac{\omega_P(p)}{p} =+ \infty.$
We will also need the notion of multiplicative \Folner{} sequences.
Given $r,K\in \N$ with $K>r$, let
\begin{equation}\label{eq_foln_def_1}
\Phi_{r}:=\Big\{ \prod_{p\leq r} p^{\theta_p} \colon
r< \theta_p \leq 3r/2\Big \} \:\:
\text{ and } \:\:
\Phi_{r,K}:=\Big\{ \prod_{r<p\leq K} p^{\theta_p} \colon
K< \theta_p \leq 3K/2\Big \}.
\end{equation}
In addition, given a binary quadratic form $P$ and $r,K \in \N$ with
$K>r$, we denote by $\Phi_{r,K,P}$ the set
\begin{equation}\label{eq_foln_def_2}
\Phi_{r,K, P}:=
\Bigg\{ \prod_{\substack{r< p\leq K\\ \omega_P (p)>0}}
p^{\theta_p} : K<\theta_p \leq {3K}/{2} \Bigg\}.
\end{equation}
The sets $\Phi_r$ are asymptotically invariant as $r\to \infty$ under dilation,
while the sets
$\Phi_{r,K}$ (respectively $\Phi_{r,K, P}$) are asymptotically invariant as
$K\to \infty$ under dilation by $p\in \P$ with $p>r$ (respectively with $p>r$
such that $\omega_P (p)>0$).
For each $L\in \N$, let
\begin{equation}\label{Q_l_def}
Q_L:= \prod_{p\leq L} p^{2L}.
\end{equation}
Since $\log2$ and $\pi$ are known to be rationally independent (this is a consequence of the Lindemann-Weierstrass theorem), $(n\log 2)_{n\in \N}$ is equidistributed $\mod{2\pi}$,
which implies
that given $\delta>0, L\in \N$, there is $n_{\delta, L} \in \N$ such that
$|2^{in_{\delta, L}} Q_{L}^i -1| \leq \delta$. Given $\delta, L$ we let
\begin{equation}\label{Q_delta_l_def}
Q_{\delta , L}:=2^{n_{\delta,L}} Q_{L}
\end{equation}
and from the previous we have
\begin{equation}\label{Q delta L def}
|Q_{\delta , L}^i -1|\leq \delta.
\end{equation}
\subsection{Background on multiplicative functions}\label{SS:mf}
Recall that a function $f \colon \N \to \C$ is called
\emph{multiplicative} if $f(mn) = f(m) \cdot f(n)$ whenever
$(m,n) = 1$, and \emph{completely multiplicative} if the same identity
holds for all $m,n \in \N$.
The periodic completely multiplicative functions are precisely the
{\em Dirichlet characters}, which we typically denote by $\chi$. For our context, we define the {\em conductor} of $\chi$ to be its smallest period.
We let
\begin{equation*}
\M := \{ f \colon \N \to \mathbb{S}^1 \; : \; f \text{ is completely multiplicative} \}
\end{equation*}
and endow $\M$ with the topology of pointwise
convergence, under which $\M$ is a compact metric space.
Following the terminology of Granville and Soundararajan
\cite{Granville-Sound-book}, given two multiplicative functions
$f, g \colon \N \to \U$, we define their \emph{pretentious distance}
between two real numbers $A,B$ by
\begin{equation}\label{E: definition of pretentious distance}
\D(f,g;A,B)
:= \Big( \sum_{A < p \leq B}
\frac{1}{p}\big(1 - \Re \big( f(p)\, \overline{g(p)} \big)
\Big)^{1/2}.
\end{equation}
We set $\D(f,g) := \lim\limits_{X \to \infty} \D(f,g; 1, X)$, and say
that $f$ \emph{pretends} to be $g$, denoted $f \sim g$, if
$\D(f,g) < +\infty$.
\begin{definition} A multiplicative function $f$ is called \emph{pretentious} if there
exist a Dirichlet character $\chi$ and a real number $t$ such that
$f \sim \chi \cdot n^{it}$.
We denote by $\M_p$ the set of pretentious multiplicative functions in $\M$.
\end{definition}
By \cite[Lemma~3.6]{Fra-Klu-Mor}, the set $\M_p$ is a Borel subset of $\M$.
Given $K, N \in \N$ with $L < N$, a binary quadratic form $P$, and a
multiplicative function $f$, we define
\begin{equation}\label{def_F_N}
F_N(f,L) := \sum_{L < p \leq N}
\frac{1}{p} \, \big(f(p)\, \overline{\chi(p)} \, p^{-it} - 1\big)
\end{equation}
and
\begin{equation}\label{def_G_P,N}
G_{P,N}(f,L) := \sum_{L < p \leq N}
\frac{\omega_P(p)}{p}\, \big( f(p) \, \overline{\chi(p)}\, p^{-it} - 1 \big).
\end{equation}
We will need the following concentration estimates for pretentious
multiplicative functions, first established in \cite{Klu-Man-Po-Ter}
and subsequently adapted and applied in numerous works
(see, for example, \cite{cha-mou-tsin,fra-klu-mor-2,Fra-Klu-Mor})
\begin{proposition}{\cite[Lemma 2.5]{Klu-Man-Po-Ter}}\label{linear_conc}
Let $f \colon \N \to \U$ be a multiplicative function such that
$f \sim \chi \cdot n^{it}$ for some $t \in \R$ and some Dirichlet
character $\chi$ of period $q$.
Let $Q, K \geq 1$ be integers with
$\prod_{p \leq K} p \mid Q$, $q \mid Q$, and all prime factors of $Q$
lying in the interval $[1, K]$.
Then, for any $a \in \N$ with $(a, Q) = 1$, we have
\begin{multline}\label{lin_conc_eq_1}
\limsup_{N \to \infty}
\cesE_{n \in [N]}
\big| f(Qn+a) - \chi(a) \cdot (Qn)^{it}
\cdot \exp(F_N(f,K)) \big| \\
\ll \D(f, \chi \cdot n^{it}, K, \infty) + K^{-1/2},
\end{multline}
where $\D$ and $F_N(f,L)$ are as in
\eqref{E: definition of pretentious distance} and \eqref{def_F_N},
and the implicit constant is absolute.
\end{proposition}
\begin{example}[$1$-pretentious oscillatory multiplicative function]
If $f(p) := e(1 / \log\log p)$ for all
$p \in \P$, then $f \sim 1$, but the imaginary part of the
series $\sum_{p \in \P} \frac{1}{p} \big( 1 - f(p) \big)$ diverges.
Consequently, the term $\exp\!\big(F_N(f,K)\big)$ in~\eqref{lin_conc_eq_1} is
oscillatory.
\end{example}
The next is a variant that deals with binary quadratic forms.
\begin{proposition}{\cite[Proposition 4.9]{fra-klu-mor-2}}\label{quad_conc}
Let $P \in \Z[m,n]$ be an irreducible binary quadratic form and
let $f \colon \Z \to \U$ be an even multiplicative function such that
$f \sim \chi \cdot n^{it}$ for some $t \in \R$ and some
Dirichlet character $\chi$ of period $q$.
Let $Q, K \geq 1$ be integers, suppose that all the prime factors of $Q$
belong to $[1,K]$, and let $K$ be sufficiently large, depending only
on $P$, and so that
$\D(f,\chi \cdot n^{it}, K, \infty) \leq 1$. Then
\begin{multline}\label{quad_conc_eq_1}
\limsup_{N\to \infty} \cesE _{m,n\in [N]} |f(P_c (Qm+a, Qn+b)) - \\
\chi(P_c(a,b)) \cdot (P_c(Qm, Qn))^{it} \cdot \exp(G_{P,N}(f,K)) |
\ll_{P} \D(f, \chi \cdot n^{it},K,\infty) + K^{-1/2},
\end{multline}
where in the previous
$c:=(P(a,b),Q)$, $P_c:=P/c$, and $a,b$ are such that $q$ and $\prod_{p\leq K}p$
divide $Q/c$. Finally, $\D$ and $G_{P,N}$ are as in
\eqref{E: definition of pretentious distance} and \eqref{def_G_P,N}, and the implicit constant only
depends on $P$.
\end{proposition}
\begin{rmk}\label{uniform_in_Q}
Note that in \eqref{lin_conc_eq_1} and \eqref{quad_conc_eq_1}, the bounds
on the right-hand side depend only on $K$ and not on $Q$. Therefore,
if $\mathcal{C}_K$ is a finite collection of $Q$'s and, for each
$Q \in \mathcal{C}_K$, $a_Q$ is such that \eqref{lin_conc_eq_1} holds
(respectively $a_Q, b_Q$ are such that \eqref{quad_conc_eq_1} holds),
then in \eqref{lin_conc_eq_1} (respectively
\eqref{quad_conc_eq_1}) one may replace
``$\limsup_{N\to \infty} \cesE_{m,n\in [N]}$'' by
``$\limsup_{N\to \infty} \max_{Q \in \mathcal{C}_K} \cesE_{m,n\in [N]}$''.
\end{rmk}
\subsection{Spectral measures}
Note that $\mathbb{S}^1$-valued completely multiplicative functions on
$\Q_{>0}$ are uniquely determined by their values on $\N$, so we may
identify $\M$ with the Pontryagin dual of the discrete group
$(\Q_{>0}, \times)$.
Let $(X, \CX, \mu, T_n)$ be a multiplicative action and let $F \in L^2(\mu)$.
The map
$$
(\Q_{>0}, \times) \to \C, \quad r/s \mapsto \int_X T_r F \cdot T_s \overline{F} \, d\mu
$$
is well-defined and positive definite. By the Bochner-Herglotz theorem,
there exists a finite Borel measure $\sigma_F$ on $\M$
(called the \emph{spectral measure} of $F$) such that
$$
\int_X T_r F \cdot T_s \overline{F} \, d\mu
= \int_{\M} f(r) \cdot \overline{f(s)} \, d\sigma_F(f)
\quad \text{for all } r,s \in \Q_{>0}.
$$
From the above, we can easily deduce the following identity, which we will use frequently:
\begin{equation}\label{spec_eq_1}
\Bigg\lVert \sum_{k=1} ^{l} c_k T_{r_k} F \Bigg\rVert_{L^2(\mu)}=
\Bigg\lVert \sum_{k=1} ^{l} c_k f({r_k}) \Bigg\rVert_{L^2(\sigma_F(f))}
\end{equation}
for all $l\in \N$, $c_1, \ldots, c_l \in \C$,
$r_1, \ldots, r_l \in \Q_{>0}$, where, for $f\in \M$ and $r=m/n \in \Q_{>0}$,
$f(r)=f(m)/f(n)=f(m)\overline{f(n)}$.
From \eqref{spec_eq_1}, we immediately obtain that for all $r\in \Q_{>0}$,
\begin{equation}\label{spec_eq_2}
\lVert T_{r} F -F \rVert_{L^2(\mu)}=
\lVert f({r}) - 1 \rVert_{L^2(\sigma_F(f))}.
\end{equation}
\subsection{Factors of multiplicative actions}\label{SS:factors}
In the proof of our main result, we will require
that certain subsets of $L^{\infty}(\mu)$ form factors
of the original system. Recall that $\mathcal{M}_p$ is the set
of pretentious multiplicative functions. We denote by $\mathcal{A}$
the set of Archimedean characters, namely
\begin{equation}\label{archim_def}
\A:=\{n^{it}: t\in \R\},
\end{equation}
and by $\M_{p,\textup{fin.supp.}}$ the set
\begin{multline}\label{def_pretent_conc}
\M_{p,\textup{fin.supp.}}:= \{ f\in \M_p \colon \text{ there are } \chi, t \text{ such that } f(p)=\chi(p) \, p ^{it} \\
\text{ for all but finitely many primes }p\}.
\end{multline}
Finally, given an irreducible binary quadratic form $P$, let
\begin{multline}\label{def_pretent_conc_1}
\M_{p,\textup{fin.supp.}, P}:= \{ f\in \M_p \colon \text{ there are } \chi, t \text{ such that } f(p)=\chi(p) \, p ^{it} \\
\text{ for all but finitely many
primes }p \text{ with } \omega_P(p)>0 \}.
\end{multline}
\begin{rmk}\label{useful_rmk_1}
If $f\in \M_{p,\textup{fin.supp.}, P}$, then there exist
$\chi_1, \chi_2, t_1, t_2$ such that $f\sim \chi_1 \cdot n^{it_1}$ and
$f(p)=\chi_2 (p)\, p^{it_2}$ for all but finitely many primes $p$ with
$\omega_P(p)>0$. By \cite[Lemma 4.5]{fra-klu-mor-2}, this implies that
$t_1 = t_2$ and $\chi_1(p) = \chi_2(p)$ for all but finitely many such $p$.
Consequently, we have $f\sim \chi_1\cdot n^{it_1}$ and
$f(p) = \chi_1 (p)\, p^{it_1}$ for all but finitely many primes $p$ with
$\omega_P(p)>0$.
\end{rmk}
The set $\A$ is a Borel subset of $\M$, since it is a countable union of compact
sets. We will now prove that $\M_{p,\textup{fin.supp.}}$ and
$\M_{p,\textup{fin.supp.}, P}$ are also Borel subsets of $\M$.
\begin{lemma}
The sets $\M_{p,\textup{fin.supp.}}$ and
$\M_{p,\textup{fin.supp.}, P}$ are Borel subsets of $\M$.
\end{lemma}
\begin{proof}
We first prove that $\M_{p,\textup{fin. supp.}}$ is Borel.
Given $n, k \in \N$ and a Dirichlet character $\chi$,
consider the map
$$
H_{n,k,\chi}: \M \to [0,+\infty), \quad
H_{n,k,\chi}(f):=\inf_{t\in \Q \cap [-k,k]}
\sum_{p\geq n} \frac{1}{p^2} \, |f(p)-\chi(p) \, p^{it}|.
$$
The function $H_{n,k,\chi}$ is Borel measurable, since it is the
countable infimum of measurable functions.\footnote{Each series in this
expression is a Borel measurable function with respect to $f$, as it is the pointwise
limit of continuous functions.}
It follows that
$$
E_{n,k,\chi}:=\{f\in \M: H_{n,k,\chi}(f)=0\}
$$
is Borel measurable.
For each $f\in \M$, the map $\R \to [0,+\infty)$ given by
$$
t\mapsto \sum_{p\geq n} \frac{1}{p^2}\, |f(p)-\chi(p)
\, p^{it}|
$$
is continuous. Using this observation, it is straightforward to verify
that
$$
\M_{p,\textup{fin.supp.}}=\M _p \cap \Big( \bigcup_{n,k\in \N}
\bigcup_{\chi} E_{n,k,\chi}\Big),
$$
where the innermost union is taken over all Dirichlet characters $\chi$.
Since the set of Dirichlet characters is countable, we deduce that
$\M_{p,\textup{fin.supp.}}$ is Borel measurable.
If, in the previous argument, we instead define $H_{n,k,\chi}(f)$ by
$$
H_{n,k,\chi}(f):=\inf_{t\in \Q \cap [-k,k]}
\sum_{p\geq n} \frac{\omega_P(p)}{p^2} |f(p)-\chi(p) p^{it}|,
$$
then the same reasoning shows that $\M_{p,\textup{fin. supp.},P}$ is
also Borel measurable.
This completes the proof.
\end{proof}
Given a pretentious multiplicative action $(X, \X, \mu, T_n)$, consider the
subsets $X_{\A}$, \\
$X_{\textup{fin.supp.}}$ and $X_{\textup{fin.supp.},P}$ of $L^{\infty}(\mu)$ defined by
\begin{equation}\label{def_X_A}
X_{\A}:=\{ F \in L^{\infty}(\mu): \sigma_F \text{ is supported on } \A\},
\end{equation}
\begin{equation}\label{def_X_p_fin_supp}
X_{p,\textup{fin.supp.}}:=\{ F \in L^{\infty}(\mu): \sigma_F \text{ is supported on } \M_{p,\textup{fin.supp.}}\},
\end{equation}
and
\begin{equation}\label{def_X_p_fin_supp_P}
X_{p,\textup{fin.supp.},P}:=\{ F \in L^{\infty}(\mu): \sigma_F \text{ is supported on } \M_{p,\textup{fin.supp.},P}\}.
\end{equation}
We will prove that $X_{\A}, X_{p,\textup{fin.supp.}}$, and
$X_{p,\textup{fin.supp.},P}$ are conjugation closed algebras, so they
define factors $\X_{\A}, \X_{p,\textup{fin.supp.}},$ and $\X_{p,\textup{fin.supp.},P}$ of the original system.
In view of \cite[Lemma 4.6 (i)-(iii)]{Fra-Hom_Triples} we have that
$X_{\A}, X_{p,\textup{fin.supp.}}$ and $X_{p,\textup{fin.supp.},P}$
are closed $T_n$-invariant subspaces of $L^2(\mu)$, and one easily sees that
they are conjugation closed.
As a result, to prove that they define factors, it suffices to prove that for
$\mathcal{H}\in \{ X_{\A}, X_{p,\textup{fin.supp.}}, X_{p,\textup{fin.supp.},P} \}$, the following implication holds
\begin{equation}\label{E:F12}
F_1, F_2 \in \mathcal{H} \implies F_1 \cdot F_2 \in \mathcal{H}.
\end{equation}
For $r,K \in \N$ with $r<K$ and $Q \in \Phi_{r,K}$, we use the Chinese
remainder theorem to pick $v=v(r, K, Q)\in \N$ such that
\begin{itemize}
\item $v\equiv Q^{-1} \pmod {p^r}$ for $p\leq r$,
\item $v\equiv 1 \pmod {p}$ for $r<p\leq K$.
\end{itemize}
Then it is easy to see that
\begin{equation}\label{eq_cong_lw_1}
\big(v, \prod_{p\leq K} p \big)=1 \: \text{ and } \:
v\equiv Q^{-1} \! \!\!\! \pmod{\prod_{p\leq r} p^r}.
\end{equation}
On the other hand, given an irreducible binary quadratic form $P$,
$r, K\in \N$ with $r<K$ and $Q\in \Phi_{r, K, P}$,
use the Chinese remainder theorem to pick $u=u(r, K, Q,P)\in \N$ with
\begin{itemize}
\item $u\equiv Q^{-1} \pmod {p^r}$ for $p\leq r$,
\item $u\equiv 1 \pmod {p}$ for $r<p\leq K$.
\end{itemize}
Similarly, as before, we have
\begin{equation}\label{eq_cong_lw_2}
\big(u, \prod_{p\leq K} p \big)=1 \: \text{ and } \:
u\equiv Q^{-1} \! \! \! \! \pmod{\prod_{p\leq r} p^r}.
\end{equation}
To deduce the implication \eqref{E:F12}, we will use the following concentration
estimates for functions in $L^2(\mu)$.
\begin{proposition}\label{P: conc_L_2_factors}
Let $F\in L^{\infty}(\mu)$ and $P$ be an irreducible binary
quadratic form, and let $\Phi_r$, $\Phi_{r,K}$, $Q_K$, be as in
\eqref{eq_foln_def_1}, \eqref{Q_l_def}.
Then
\begin{enumerate}
\item \label{factors_i} $ F \in X_{\A}$ if and only if
\begin{equation}\label{eq_arch}
\lim_{r\to \infty} \limsup_{N\to \infty} \max_{Q \in \Phi_r} \cesE_{n\in [N]}
\Big \lVert T_{\frac{Qn+1}{Qn}}F -F
\Big \rVert _{L^2(\mu)}=0.
\end{equation}
\item \label{factors_ii} $ F \in X_{p,\textup{fin.supp.}}$ if and only if
\begin{equation}\label{eq_fin_sup_factor}
\lim_{r\to \infty} \limsup_{K\to \infty} \limsup_{N\to \infty}
\max_{Q \in \Phi_{r,K}} \cesE_{n\in [N]}
\Big \lVert T_{\frac{Q(Q_K Q n+v(r,K,Q))}{Q_K Q^2 n+1}}F -F
\Big \rVert _{L^2(\mu)}=0.
\end{equation}
\item \label{factors_iii} $ F \in X_{p,\textup{fin.supp.},P}$ if and only if
\begin{equation}\label{eq_fin_sup_P_factor}
\lim_{r\to \infty} \limsup_{K\to \infty} \limsup_{N\to \infty}
\max_{Q \in \Phi_{r,K,P}} \cesE_{n\in [N]}
\Big \lVert T_{\frac{Q(Q_K Q n+u(r,K,Q,P))}{Q_K Q^2 n+1}}F -F
\Big \rVert _{L^2(\mu)}=0.
\end{equation}
\end{enumerate}
\end{proposition}
\begin{proof}
In what follows, $\D$ and $F_N(f,K)$ are always as in
\eqref{E: definition of pretentious distance} and \eqref{def_F_N} respectively.
Since the proofs of \eqref{factors_i}, \eqref{factors_ii},
\eqref{factors_iii}
are similar, we give a detailed proof of
\eqref{factors_iii}, and then we sketch the proofs of
\eqref{factors_i}, \eqref{factors_ii}.
Let $F\in X_p$. In the following, we simply write $u$ instead of
$u(r,K,Q,P)$.
Given $n,r,K\in \N$ and $Q\in \Phi_{r,K,P}$, using \eqref{spec_eq_2} we have
\begin{equation}\label{factors_eq_1}
\Big \lVert T_{\frac{Q(Q_K Q n+u)}{Q_K Q^2 n+1}}F -F
\Big \rVert _{L^2(\mu)} ^2
=
\int_{\M _p} \big | {f(Q)}f(Q_K Qn+u)\overline{f(Q_K Q^2 n+1)} - 1
\big | ^2 d \sigma_F(f).
\end{equation}
If $F\in X_{p, \textup{fin.supp.},P}$,
then $\sigma_F$ is supported on $\M_{p, \textup{fin.supp.},P}$, so using
\eqref{factors_eq_1}, Cauchy-Schwarz and Fatou's lemma one sees that in
order to prove
\eqref{eq_fin_sup_P_factor}, it suffices to prove that for each
$f\in \M_{p, \textup{fin.supp.},P}$ we have
\begin{equation}\label{factors_eq_3}
\lim_{r\to \infty}
\limsup_{K\to \infty}
\limsup_{N\to \infty}
\max_{Q \in \Phi_{r,K,P}} \cesE_{n\in [N]}
| {f(Q)}f(Q_K Qn+u)\overline{f(Q_K Q^2 n+1)} - 1
\big |=0.
\end{equation}
Let $f\in \M_{p, \textup{fin.supp.},P}$. From \eqref{def_pretent_conc_1}
and \cref{useful_rmk_1} there exist a Dirichlet character $\chi$,
$t\in \R$, and $r_0\in \N$ such that
$f\sim \chi \cdot n^{it}$ and $f(p)= \chi(p) \cdot p^{it}$ for all primes
$p>r_0$ with $\omega_P(p)>0$.
Let $q_{\chi}$ be the conductor of $\chi$, and suppose that $r>r_0$ and
$r$ is large enough so that $q_{\chi} \mid \prod_{p\leq r} p^r$.
From \eqref{eq_cong_lw_2} we deduce that $(u,Q_KQ)=1$ and
$u\equiv Q^{-1}\pmod{q_{\chi}}$, so using \cref{linear_conc}
and \cref{uniform_in_Q} we infer that
\begin{multline}\label{factors_eq_2}
\limsup_{N\to \infty} \max_{Q \in \Phi_{r,K,P}} \cesE_{n\in [N]}
| {f(Q)}f(Q_K Qn+u)\overline{f(Q_K Q^2 n+1)} - 1
\big | \ll
\\
\limsup_{N\to \infty} \max_{Q \in \Phi_{r,K,P}} \cesE_{n\in [N]}
\Big | {f(Q)}\, \overline{\chi(Q)} \, (Q_K Qn)^{it} \,
\exp(F_N(f,K))
{(Q_K Q^2 n)^{-it}}\cdot\\ \overline{\exp(F_N(f,K))} - 1 \Big |
+ \oh_{K\to \infty}(1) =
|\exp(-2\D^2(f, \chi \cdot n^{it}; K, \infty))-1| +
\oh_{K\to \infty}(1),
\end{multline}
where in the last equality we used that every prime dividing $Q$ is
greater than $r>r_0$ and has $\omega_P(p)>0$, so
$f(Q)=\chi(Q)Q^{it}$, and that $2\Re(F_N(f,K))=-2\D^2(f, \chi \cdot n^{it};K, N)$, which converges to
$-2\D^2(f, \chi \cdot n^{it}; K, \infty)$ as $N\to \infty$. Since
$\D^2(f, \chi \cdot n^{it};1, \infty)< \infty$, taking
$K\to \infty$ in \eqref{factors_eq_2} we see that \eqref{factors_eq_3}
holds\footnote{One can also deduce \eqref{factors_eq_3} as follows: for
$K>r>r_0$ and $r$ large enough so that $q_{\chi} \mid \prod_{p\leq r} p^r$,
since $(u, Q_KQ)=1$, every prime dividing $Q_KQn+u$ is greater than $r_0$,
so $f(Q_K Qn +u)=\chi(u)\, (Q_K Qn+u)^{it}=$ $\overline{\chi(Q)} \, (Q_K Qn)^{it}
+ \oh_{t, n\to \infty}(1)$. Analogously, one has $f(Q_K Q^2 n +1)=
(Q_K Q^2n)^{it} + \oh_{t, n\to \infty}(1)$.}, and therefore $F$ satisfies \eqref{eq_fin_sup_P_factor}.
For the opposite direction, suppose that $F$ satisfies
\eqref{eq_fin_sup_P_factor}. Using \cite[Lemma 4.6 (v)]{Fra-Hom_Triples}
we can write $F=F_1 + F_2$, with $\sigma_{F_1}$ supported on
$\M_{p, \textup{fin.supp.},P}$ and $\sigma_{F_2}$ supported on
$\M_p \setminus \M_{p, \textup{fin.supp.},P}$.
Then using the reverse triangle inequality we have
\begin{multline*}
\Big \lVert \cesE_{Q \in \Phi_{r,K,P}} \cesE_{n\in [N]}
T_{\frac{Q(Q_K Q n+u)}{Q_K Q^2 n+1}}F -F
\Big \rVert _{L^2(\mu)}
\geq
\lVert F_2 \rVert_{L^2(\mu)} \\ -
\Big \lVert \cesE_{Q \in \Phi_{r,K,P}} \cesE_{n\in [N]}
T_{\frac{Q(Q_K Q n+u)}{Q_K Q^2 n+1}}F_1
-F_1 \Big \rVert _{L^2(\mu)} - \Big \lVert
\cesE_{Q \in \Phi_{r,K,P}} \cesE_{n\in [N]}
T_{\frac{Q(Q_K Q n+u)}{Q_K Q^2 n+1}}F_2
\Big \rVert _{L^2(\mu)},
\end{multline*}
which in turn implies that
\begin{multline}\label{factors_eq_4}
\lVert F_2 \rVert_{L^2(\mu)} \leq \max_{Q \in \Phi_{r,K,P}}
\cesE_{n\in [N]}
\Big \lVert T_{\frac{Q(Q_K Q n+u)}{Q_K Q^2 n+1}}F -F
\Big \rVert _{L^2(\mu)} \\ +
\max_{Q \in \Phi_{r,K,P}} \cesE_{n\in [N]}
\Big \lVert T_{\frac{Q(Q_K Q n+u)}{Q_K Q^2 n+1}}F_1 -F_1
\Big \rVert _{L^2(\mu)}
+
\Big \lVert \cesE_{Q \in \Phi_{r,K,P}} \cesE_{n\in [N]}
T_{\frac{Q(Q_K Q n+u)}{Q_K Q^2 n+1}}F_2
\Big \rVert _{L^2(\mu)}.
\end{multline}
Since $F_1 \in X_{p, \textup{fin.supp.},P}$, from the already established direction we get that
\begin{equation}\label{factors_eq_5}
\lim_{r\to \infty} \limsup_{K\to \infty} \limsup_{N\to \infty}
\max_{Q \in \Phi_{r,K,P}} \cesE_{n\in [N]}
\Big \lVert T_{\frac{Q(Q_K Q n+u)}{Q_K Q^2 n+1}}F_1 -F_1
\Big \rVert _{L^2(\mu)}=0.
\end{equation}
For $F_2$, using \eqref{spec_eq_1} we infer that
\begin{multline}\label{factors_eq_8}
\Big \lVert \cesE_{Q \in \Phi_{r,K,P}} \cesE_{n\in [N]}
T_{\frac{Q(Q_K Q n+u)}{Q_K Q^2 n+1}}F_2
\Big \rVert _{L^2(\mu)} ^2 \\ =
\int_{\M_p \setminus \M_{p, \textup{fin.supp.},P}}
\Big | \cesE_{Q \in \Phi_{r,K,P}} \cesE_{n\in [N]}
f(Q) f(Q_K Q n+u) \overline{f(Q_K Q^2 n+1)}
\Big | ^2 d \sigma_{F_2}(f).
\end{multline}
Let $f\in \M_{p}\setminus \M_{p, \textup{fin.supp.},P}$ with
$f\sim \chi \cdot n^{it}$. Let
$q_{\chi}$ be the conductor of $\chi$ and suppose that
$r$ is large enough so that
$q_{\chi} \mid \prod_{p\leq r} p^r$.
From \eqref{eq_cong_lw_2} we deduce that $(u,Q_KQ)=1$ and
$u\equiv Q^{-1}\pmod{q_{\chi}}$, so using \cref{linear_conc}
and \cref{uniform_in_Q} we infer that
\begin{multline}\label{factors_eq_6}
\cesE_{Q \in \Phi_{r,K,P}} \cesE_{n\in [N]}
f(Q) \, f(Q_K Q n+u) \, \overline{f(Q_K Q^2 n+1)}
\\ =
\exp(2\Re(F_N(f,K)))\cdot
\cesE_{Q \in \Phi_{r,K,P}}
f(Q) \, Q ^{-it} \, \overline{\chi(Q)} +
\oh_{K\to \infty, N\to \infty}(1).
\end{multline}
Since $f \notin \M_{p, \textup{fin.supp.},P}$,
there is $p_0 \in \P$ with
$p_0>r$ so that $\omega_P(p_0)>0$ and
$f(p_0)\neq p_0 ^{it}\cdot \chi({p_0})$. Note that since
$p_0>r$ and $q_{\chi} \mid \prod_{p\leq r}p^r$, we have $|\chi(p_0)|=1$,
and using that $\Phi_{r,K,P}$ is asymptotically invariant under
dilation by $p_0$ as $K\to \infty$ we get that
\begin{align*}
f(p_0) \, p_0 ^{-it} \, \overline{\chi({p_0})}\,
\cesE_{Q \in \Phi_{r,K,P}}
f(Q) \, Q ^{-it} \, \overline{\chi(Q)}&=
\cesE_{Q \in p_0 \Phi_{r,K,P}}
f(Q)\, Q ^{-it} \, \overline{\chi(Q)} \\& =
\cesE_{Q \in \Phi_{r,K,P}}
f(Q)\, Q ^{-it}\, \overline{\chi(Q)} + \oh_{K \to \infty}(1),
\end{align*}
which implies that
\begin{equation}\label{factors_eq_7}
\lim_{K\to \infty}
\cesE_{Q \in \Phi_{r,K,P}}
f(Q)\, Q ^{-it} \, \overline{\chi(Q)}=0.
\end{equation}
From \eqref{factors_eq_6} and \eqref{factors_eq_7} we get that for
$f\in \M_p \setminus \M_{p, \textup{fin.supp.},P}$ we have
\begin{equation*}
\lim_{r\to \infty} \limsup_{K\to \infty}
\limsup_{N\to \infty} \Big|
\cesE_{Q \in \Phi_{r,K,P}} \cesE_{n\in [N]}
f(Q)\, f(Q_K Q n+u)\, \overline{f(Q_K Q^2 n+1)}
\Big|,
\end{equation*}
which in view of \eqref{factors_eq_8} implies
\begin{equation}\label{factors_eq_9}
\lim_{r\to \infty}
\limsup_{K\to \infty}
\limsup_{N\to \infty}
\Big \lVert \cesE_{Q \in \Phi_{r,K,P}} \cesE_{n\in [N]}
T_{\frac{Q(Q_K Q n+u)}{Q_K Q ^2 n+1}}F_2
\Big \rVert _{L^2(\mu)} =0.
\end{equation}
Taking $N\to \infty$, then $K\to \infty$ and finally $r\to \infty$ in
\eqref{factors_eq_4} and using \eqref{factors_eq_5},
\eqref{factors_eq_9} and the fact that
$F$ satisfies \eqref{eq_fin_sup_P_factor},
we get that $\lVert F_2 \rVert _{L^2(\mu)}=0$, so $F=F_1 \in X_{p, \textup{fin.supp.},P}$.
This concludes the proof \eqref{factors_iii}.
If in the previous one replaces $u(r,K,Q,P)$ by $v(r,K,Q)$ and
$\Phi_{r,K,P}$ by $\Phi_{r,K}$, then \eqref{factors_ii} follows with
the same proof.
Finally, we sketch the proof of \eqref{factors_i}.
If $F\in X_{\A}$,
then $\sigma_F$ is supported on $\A$, so in order to establish
\eqref{eq_arch} it suffices to prove that for each $f\in \A$ we have
\begin{equation}\label{factors_eq_10}
\lim_{r\to \infty} \limsup_{N\to \infty} \max_{Q \in \Phi_r} \cesE_{n\in [N]}
| f(Qn+1)\, \overline{f(Qn)} -1 |=0.
\end{equation}
For $f\in \A$ with $f=n^{it}$ we have
\begin{equation*}
\max_{Q \in \Phi_r} \cesE_{n\in [N]}
| \overline{f(Qn)} \, f(Qn+1)-1 |=
\max_{Q \in \Phi_r} \cesE_{n\in [N]}
| (Qn)^{-it}\, (Qn+1)^{it}-1 |=\oh_{t, r, N\to \infty}(1),
\end{equation*}
and the result follows.
For the opposite direction, suppose that $F$ satisfies
\eqref{eq_arch}. Again, write $F=F_1 + F_2$, with $\sigma_{F_1}$
supported on $\A$ and $\sigma_{F_2}$ supported on $\M_p \setminus \A$.
From reverse triangle inequality we have
\begin{multline}\label{factors_eq_11}
\lVert F_2 \rVert_{L^2(\mu)} \leq \max_{Q \in \Phi_{r}}
\cesE_{n\in [N]}
\Big \lVert T_{\frac{Qn+1}{Qn}}F -F
\Big \rVert _{L^2(\mu)} + \\
\max_{Q \in \Phi_{r}} \cesE_{n\in [N]}
\Big \lVert T_{\frac{Qn+1}{Qn}}F_1 -F_1
\Big \rVert _{L^2(\mu)}
+
\Big \lVert \cesE_{Q \in \Phi_{r}} \cesE_{n\in [N]}
T_{\frac{Qn+1}{Qn}}F_2
\Big \rVert _{L^2(\mu)}.
\end{multline}
Since $F_1 \in X_{\A}$, from the already established direction we get
that
\begin{equation}\label{factors_eq_12}
\lim_{r\to \infty} \limsup_{N\to \infty}
\max_{Q \in \Phi_{r}} \cesE_{n\in [N]}
\Big \lVert T_{\frac{Qn+1}{Qn}}F_1 -F_1
\Big \rVert _{L^2(\mu)}=0.
\end{equation}
For $f\in \M_{p}\setminus \A$ with $f\sim \chi \cdot n^{it}$,
using \cref{linear_conc}
and \cref{uniform_in_Q} we infer that
\begin{multline*}
\cesE_{Q \in \Phi_{r}} \cesE_{n\in [N]}
f(Qn+1) \, \overline{f(Qn)}
\\ =
\exp(F_N(f,r)) \Big(\cesE_{n\in [N]} \overline{f(n)}\, n^{it} \Big)
\Big( \cesE_{Q \in \Phi_{r}} \overline{f(Q)}\, Q ^{it} \Big) +
\oh_{r\to \infty, N\to \infty}(1).
\end{multline*}
Since $f \notin \A$ there is $p_0 \in \P$ so that
$f(p_0)\neq p_0 ^{it}$. Again, using that $\Phi_{r}$ is asymptotically
invariant under dilation by $p_0$ as $r\to \infty$ we get that
\begin{equation*}
\lim_{r\to \infty}
\cesE_{Q \in \Phi_{r}}
\overline{f(Q)}\, Q ^{it} =0.
\end{equation*}
Note that $|\exp(F_N(f,r))|=\exp(-\D^2(f,\chi \cdot n^{it};r, N))$, which
converges to $1$ as $N\to \infty$ and then $r\to \infty$, and combining
with the previous we get that for $f\in \M_p \setminus \A$ we have
\begin{equation*}
\lim_{r\to \infty} \limsup_{N \to \infty}
\Big| \cesE_{Q \in \Phi_{r}} \cesE_{n\in [N]}
f(Qn+1)\, \overline{f(Qn)} \Big| =0.
\end{equation*}
Since $\sigma_{F_2}$ is supported on $\M_p \setminus \A$, we
infer that
\begin{equation}\label{factors_eq_14}
\lim_{r\to \infty} \limsup_{N \to \infty}
\Big \lVert \cesE_{Q \in \Phi_{r}} \cesE_{n\in [N]}
T_{\frac{Qn+1}{Qn}}F_2
\Big \rVert _{L^2(\mu)}=0 .
\end{equation}
Using \eqref{factors_eq_11}, \eqref{factors_eq_12},
\eqref{factors_eq_14}, and the fact that $F$ satisfies \eqref{eq_arch},
we get that $\lVert F_2 \rVert _{L^2(\mu)}=0$, so $F=F_1 \in X_{A}$.
This concludes the proof \eqref{factors_i}.
\end{proof}
Finally, we have the following lemma, which concludes the proof that
$X_{\A}, X_{p,\textup{fin.supp.}}$, $X_{p,\textup{fin.supp.},P}$
are factors.
\begin{lemma}
For $\mathcal{H}\in \{ X_{\A}, X_{p,\textup{fin.supp.}},
X_{p,\textup{fin.supp.},P} \}$, if $F_1, F_2 \in \mathcal{H}$,
then $F_1 \cdot F_2 \in \mathcal{H}$.
\end{lemma}
\begin{proof}
Since the proofs are identical, we will only prove the case when
$\mathcal{H}=X_{\A}$. For $F_1, F_2 \in X_{\A}$, $Q\in \Phi_r$
and $n\in \N$ we have
\begin{align*}
\Big \lVert T_{\frac{Qn+1}{Qn}}(F_1 \cdot F_2) - F_1 \cdot F_2
\Big \rVert _{L^2(\mu)}
\leq
&\lVert F_1
\rVert _{L^{\infty}(\mu)}
\Big \lVert T_{\frac{Qn+1}{Qn}} F_2 - F_2
\Big \rVert _{L^2(\mu)}
+ \\
&\lVert F_2
\rVert _{L^{\infty}(\mu)}
\Big \lVert T_{\frac{Qn+1}{Qn}} F_1- F_1
\Big \rVert _{L^2(\mu)}.
\end{align*}
In the previous, taking the maximum with respect to $Q\in \Phi_r$,
sending $N\to \infty$ and then $r\to \infty$, and using that
$F_1, F_2$ satisfy \eqref{eq_arch}, we see that $F_1 \cdot F_2$ also
satisfies \eqref{eq_arch}, which in view of \cref{P: conc_L_2_factors}
\eqref{factors_i} implies that $F_1 \cdot F_2 \in X_{\A}$. This
concludes the proof.
\end{proof}
\subsection{Polynomial visit times of Archimedean characters}
In the proof of our main result, we will average over sets of the form
\begin{equation}\label{S_delta_def}
S_{\delta}:=\{(m,n)\in \N^2 : m> \alpha n, |(P_j(m,n))^i -1|\leq \delta\,
\text{ for } j=1,2,3\},
\end{equation}
where $P_j$, $j=1,2,3,$ are homogeneous binary quadratic forms, $\alpha >0$
and $\delta \in(0,1/2)$. In this section we prove that the sets $S_{\delta}$ of
interest have positive upper density, where for a set $E\subset \N^2$, its
upper density $\overline{\textup{d}}(E)$ is defined as
$$\overline{\textup{d}}(E):=\limsup_{N\to \infty} \frac{|E\cap [N]^2|}{N^2}$$
and $[N]^2: = [N]\times [N]$.
\begin{lemma}\label{L:positive1}
Let $a,b\in \N$. Suppose that $P_1,P_2,P_3$ are either
\begin{enumerate}
\item \label{I:pos1} $m^2-2bmn-abn^2$, $m^2+2amn-abn^2$, $m^2+abn^2$; or
\item \label{I:pos2} $2amn$, $m^2-abn^2$, $m^2+abn^2$.
\end{enumerate}
% Let $\alpha>0$ and $P_1,P_2,P_3 \in \Z[m,n]$ be %binary quadratic forms such that the leading %coefficients of the polynomials $P_j(m,1)$, $j=1,2,3,$ %are positive. Suppose also that either all forms are %irreducible and the leading coefficients of $P_j(m,1)$, %$j=1,2,3,$ coincide, or $P_1$ is reducible, the leading %coefficients of $P_2(m,1),P_3(m,1)$ coincide, and %$P_1,P_2$ have no common linear factors.
Let $\alpha>0$ so that for $m> \alpha n$,
$P_j(m,n) >0$, $j\in \{1,2,3\}$.
Then for every $\delta>0$ we have
$$
\underline{\textup{d}}(\{(m,n)\in \N^2 \colon m> \alpha n,
\, \, |(P_j(m,n)/P_{j+1}(m,n))^i-1|\leq \delta,\, j=1,2\} )>0.
$$
\end{lemma}
The proof of \cref{L:positive1} is similar to the proofs of
{\cite[Lemma 6.4]{fra-klu-mor-2}} and {\cite[Lemma 3.3]{Fra-Klu-Mor}},
so we only sketch it here.
\begin{proof}[Sketch of proof]
Let $I_\delta$ be as in \eqref{E:Id}
and
$F_\delta$ be the trapezoidal function that is equal to $1$ on the arc $I_{\delta/2}$ and $0$ outside $I_\delta$. Arguing as in \cite[Lemma 6.4]{fra-klu-mor-2}, it suffices to show that for every $\delta>0$ we have
\begin{equation}\label{E:positive}
\int_0^1\int_0^1 \prod_{j=1,2} F_\delta((P_j(x,y))^i\cdot (P_{j+1}(x,y))^{-i}) \cdot {\bf 1}_{x> \alpha y}
%{\bf 1}_{P_1(x,y)>0,\ldots, P_s(x,y)>0}
\, dxdy>0.
\end{equation}
Suppose that we are in case \eqref{I:pos1}.
% first that all forms are irreducible and the leading %coefficients of $P_j(m,1)$, $j=1,2,3,$ coincide.
Then $\lim_{\beta \to\infty} P_j(\beta,1)/P_{j+1}(\beta,1)=1$ for $j=1,2$, and since the polynomials are homogeneous, we deduce that for all
sufficiently large $\beta\in \R$ we have $(P_j(x,y))^i\cdot (P_{j+1}(x,y))^{-i}\in I_{\delta/2}$, $j=1,2$, for all $(x,y)$ in a neighborhood of the line segment $(\beta t,t)$ that lies within the square $[0,1]\times [0,1]$. Taking also $\beta>\alpha$ establishes the needed positivity in \eqref{E:positive}.
Suppose now that we are in case \eqref{I:pos2}.
%$P_1$ is reducible and the leading coefficients of %$P_2(m,1),P_3(m,1)$ coincide.
As before, we get that there exists $\beta_0>0$ such that if $\beta\geq \beta_0$ we have
$(P_2(x,y))^i\cdot (P_3(x,y))^{-i}\in I_{\delta/2}$ for all $(x,y)$ in a neighborhood of the line segment $(\beta t,t)$ that lies within the square $[0,1]\times [0,1]$.
We also get that there exist $\beta_k\to \infty$ such that $(P_1(x,y))^i\cdot (P_2(x,y))^{-i}\in I_{\delta/2}$ for all $(x,y)$ in a neighborhood of the line segment $(\beta_k t,t)$ that lies within the square $[0,1]\times [0,1]$. Indeed, a quick computation shows that $\beta_k:=e^{2k\pi}+\sqrt{e^{4k\pi}+b/a}$ works since $P_2(\beta_k,1)=e^{2k\pi}P_1(\beta_k,1)$.
Choosing $k_0$ such that $\beta_{k_0}>\beta_0$ and $\beta_{k_0}>\alpha$ establishes the needed positivity in \eqref{E:positive}.
\end{proof}
We also need the following lemma:
\begin{lemma}\label{L:positive2}
Let $\alpha>0$ and $P_1,\ldots, P_{k}\in \Z[m,n]$ be homogeneous
polynomials such that for $m> \alpha n$,
$P_j(m,n) >0$, $j\in \{1,\dots, k\}$, and suppose that
for every $\delta>0$ we have
$$
\overline{\textup{d}}(\{(m,n)\in \N^2 \colon m> \alpha n, \, \, |(P_j(m,n)/P_{j+1}(m,n))^i-1|\leq \delta,\, j=1,\ldots, k-1 \})>0.
$$
Then for every $\delta>0$ we have
$$
\overline{\textup{d}}(\{(m,n)\in \N ^2 \colon m> \alpha n, \, \, |(P_j(m,n))^i-1|\leq \delta, j=1,\ldots, k \})>0.
$$
\end{lemma}
\begin{proof}
Let $\delta>0$. Our assumption gives that
$$
\overline{\textup{d}}(\{(m,n)\in \N^2 \colon m> \alpha n, \, \, |(P_j(m,n))^i-(P_{j+1}(m,n))^i|\leq \delta',\, j=1,\ldots, k-1\})>0,
$$
where $\delta'$ depends on $\delta$ and will be determined later.
%Let $F_{\delta}=\{e( l \delta), l=0,\ldots, [1/\delta]\}$.
Note that if for some $m,n\in\N$ we have
$|(P_j(m,n))^i-(P_{j+1}(m,n))^i|\leq \delta'$ for $j=1,\ldots, k-1$, then there exists $l_{m,n}\in \{0,\ldots, [1/\delta']\}$ such that
$$
|(P_j(m,n))^i-e(l_{m,n} \delta')|\leq k\delta',\quad j=1,\ldots, k.
$$
So if $$
A:=\{(m,n)\in \N^2 \colon m> \alpha n, \, \,
|(P_j(m,n))^i-(P_{j+1}(m,n))^i|\leq \delta', j=1,\ldots, k-1\},
$$ and for $l\in \Z_+$ we let
\begin{equation}\label{E:Al}
A_l:=\{(m,n)\in \N^2: m> \alpha n, \,\, |(P_j(m,n))^i-e(l \delta')|\leq k\delta', j=1,\ldots, k\},
\end{equation}
we have
$A\subset \bigcup_{l=0}^{[1/\delta']}A_l$.
Since, by assumption, we have $\overline{\textup{d}}(A)>0$, we deduce using the subadditivity of $\overline{\textup{d}}$ that
there exists $l_0\in \{0,\ldots, [1/\delta']\}$ such that
$\overline{\textup{d}}(A_{l_0})>0$.
Moreover, since the set $\{k^{2i},k\in \N\}$ is dense in $\S^1$, there exists
$k_0\in \N$, that is bounded by some $K_0=K_0(\delta)$ (in particular, although $K_0$ depends on $l_0$ and $\delta'$, $K_0$ does not), such that
\begin{equation}\label{E:k0}
|(k_0^2)^i-e(-l_0\delta')|\leq \delta/2.
\end{equation}
Combining \eqref{E:Al} and \eqref{E:k0}, and since
$P_j(k_0m,k_0n)=k_0^2P_j(m,n)$,
we get that if $(m,n)\in A_{l_0}$, then
$$
|(P_j(k_0m,k_0n))^i-1|\leq K_0\delta'+\delta/2,\quad j=1,\ldots, k.
$$
Hence, if we choose $\delta'$ so that $K_0\delta'\leq \delta/2$, we get that for $(m,n)\in k_0\cdot A_{l_0}$ we have
$$
|(P_j(m,n))^i-1|\leq \delta,\quad j=1,\ldots, k.
$$
It follows that the set $\{(m,n)\in \N^2 \colon m> \alpha n, \,\,|(P_j(m,n))^i-1|\leq \delta, j=1,\ldots, k\}$ contains the set $k_0\cdot A_{l_0}$ and hence has upper density greater than
$\overline{\textup{d}}(A_{l_0})/k_0^2>0$.
\end{proof}
Combining \cref{L:positive1} and \cref{L:positive2} we deduce the following:
\begin{corollary}\label{Cor S_d dense}
Let $\alpha>0$ and $P_1,P_2,P_3 \in \Z[m,n]$ be as in \cref{L:positive1}.
Then for every $\delta>0$ the set $S_{\delta}$ defined in
\eqref{S_delta_def} has positive upper density.
\end{corollary}
\subsection{Two useful estimates}
In order to obtain not only positiveness but also the asserted lower
bound in \eqref{T:Main}, we will use the following estimate.
\begin{proposition}[{\cite[Lemma 1.6]{chu-2-commuting}}]\label{chu-lemma}
Let $(X, \X, \mu)$ be a probability space, let $\X_1, \dots, \X_{\ell} $ be sub-$\sigma$-algebras of $\X$ and $F\in L^{\infty}(\mu)$ be non-negative. Then
\begin{equation*}
\int_X F\cdot \E(F| \X_{1}) \cdots \E(F| \X_{\ell}) \, d \mu \geq \Big( \int_X F \, d \mu\Big) ^{\ell+1}.
\end{equation*}
\end{proposition}
We will also frequently use the following elementary result.
\begin{lemma}[{\cite[Lemma 4.11]{Fra-Hom_Triples}}]\label{bilinear_lemma}
Let $V$ be a normed space, let $v: \N \to V$ be a $1$-bounded even
sequence, and $l_1, l_2\in \Z$, not both of them $0$. Let $l=|l_1|+|l_2|$
and suppose
that for some $\e>0$ and for some $1$-bounded sequence $(v_{N})_{N\in \N}$ of elements in $V$ we have
\begin{equation*}
\limsup_{N\to \infty} \cesE_{n\in [N]} \lVert v(n) - v_N \rVert
\leq \e.
\end{equation*}
Then
\begin{equation*}
\limsup_{N\to \infty} \cesE_{m,n \in [N]} \lVert v({l_1m + l_2 n})
-v_N \rVert \leq 4 l \e.
\end{equation*}
\end{lemma}
\section{Proof sketch for the Pythagorean equation}\label{Sec_proof_plan}
In this section, we provide a sketch of the proof of \cref{T:Main} in
case $a=b=c=1$, which corresponds to the classical Pythagorean equation
\begin{equation}\label{proof plan eq 1}
x^2 + y^2 =z^2.
\end{equation}
We use solutions of \eqref{proof plan eq 1} given in parametric form by
\begin{equation}\label{proof plan eq 2}
x=k\, 2mn , \quad y=k\, (m^2 -n^2), \quad z=k\, (m^2 + n^2), \quad k,m,n\in \N.
\end{equation}
If we let
\begin{equation}\label{proof plan eq 4}
P_1(m,n):=2mn, \quad P_2(m,n):=m^2 -n^2, \quad P_3(m,n):=m^2 + n^2,
\end{equation}
we see that the case $a=b=c=1$ of \cref{T:Main} follows from the next result.
\begin{theorem}\label{proof plan thm 1}
Let $(X,\CX,\mu,T_n)$ be a pretentious multiplicative action, let
$F\in L^{\infty}(\mu)$ with $F\geq 0$,
and let $P_1,P_2,P_3$ be as in
\eqref{proof plan eq 4}. Then for every $\e>0$ we have
\begin{equation}\label{proof plan eq 3}
\int_X F \cdot T_{P_1(m,n)} F \cdot T_{P_2(m,n)} F \cdot T_{P_3(m,n)} F d\mu
\geq \Big(\int_X F d \mu \Big)^4 -\e
\end{equation}
for a set of $(m,n)\in \N^2$, $m>n$, with positive upper density.
\end{theorem}
All notions used in this proof sketch are defined in Section~\ref{S:Background}.
{\em Finitely generated case.} For simplicity, first suppose that the action $T_n$ is ergodic and finitely
generated. We will average over the grid $(Qm+1, Qn)$, where $Q$ is highly
divisible, and apply the $Q$-trick (as in \cite{Fra-Klu-Mor}) to obtain
\cref{proof plan thm 1}. More precisely, by \cite{Charamaras-multiplicative}
for each $F \in L^2(\mu)$, its
spectral measure is supported on
$$
\M_{p,\mathrm{f.g.}} := \{ f \colon \N \to \mathbb{S}^1, \; \text{$f$ is
pretentious and finitely generated} \},
$$
where finitely generated means that $\{ f(p) \colon p \in \P \}$ is finite.
For every $f \in \M_{p,\mathrm{f.g.}}$, if $f \sim \chi \cdot n^{it}$, then
$t = 0$, while for the oscillating factors $F_N(f,L)$ and $G_{P,N}(f,L)$
from \eqref{def_F_N} and \eqref{def_G_P,N}, we have
$\exp(F_N(f,L)), \exp(G_{P,N}(f,L)) \to 1$ as $N\to \infty$ and then $L \to \infty$; finite
generation is essential here and the claimed properties are established in \cite[Lemma~B.3 and B.4]{Charamaras-multiplicative}. These facts make it easier to handle the
terms $T_{P_j(m,n)}F$ for $j \in \{1,2,3\}$.
Using the spectral theorem together with
Propositions~\ref{linear_conc} and \ref{quad_conc}, we obtain
$$
\lim_{K \to \infty} \limsup_{N \to \infty} \max_{Q \in \Phi_K}
\cesE_{(m,n) \in [N]^2,\, m>n}
\lVert T_{P_1(Qm+1, Qn)} F -T_{2Qn} F \rVert_{L^2(\mu)} = 0,
$$
while
$$
\lim_{K \to \infty} \limsup_{N \to \infty} \max_{Q \in \Phi_K}
\cesE_{(m,n) \in [N]^2,\, m>n}
\lVert T_{P_j(Qm+1, Qn)} F - F \rVert_{L^2(\mu)} = 0, \quad j=2,3,
$$
where $(\Phi_K)_{K \in \N}$ is a multiplicative \Folner{} sequence as in
\eqref{eq_foln_def_1}.
For $j=1$, let $F_0 := F - \int_X F\, d\mu$. Using the spectral theorem again and the fact that multiplicative averages of non-trivial
multiplicative functions vanish, we deduce
$$
\lim_{K \to \infty} \limsup_{N \to \infty}
\lVert \cesE_{Q \in \Phi_K} \cesE_{(m,n) \in [N]^2,\, m>n}
T_{P_1(Qm+1, Qn)} F_0 \rVert_{L^2(\mu)} = 0.
$$
Combining these results we obtain
$$
\liminf_{K \to \infty} \liminf _{N \to \infty}
\cesE_{Q \in \Phi_K} \cesE_{(m,n) \in [N]^2,\, m>n}
\int_X F \cdot \prod_{j=1}^3 T_{P_j(Qm+1, Qn)} F\, d\mu
$$
is equal to
$$
\int_X F\, d\mu \cdot \int_X F^3\, d\mu
\geq \Big( \int_X F\, d\mu \Big)^4.
$$
{\em General case.}
We now return to the general case. Since the action $T_n$ is not
necessarily finitely generated, it is no longer true that the oscillating
factors $\exp(F_N(f,L))$ and $\exp(G_{P,N}(f,L))$ converge to $1$, and the
parameter $t$ need not be zero.
To control the $n^{it}$ terms, we restrict
our $(m,n)$-averaging to a suitably chosen set $S_{\delta}$ (see for
instance \eqref{S_delta_def}), and then take $\delta\to 0^+$.
To treat the oscillating factors, we take a triple multiplicative average
over parameters $Q_1 \in \Phi_r$, $Q_2 \in \Phi_{r,K}$, and
$Q_3 \in \Phi_{K,L,P_3}$, where the sets $\Phi_r$, $\Phi_{r,K}$, and
$\Phi_{K,L,P_3}$ are defined in \eqref{eq_foln_def_1} and
\eqref{eq_foln_def_2}. The primes dividing $Q_1$, $Q_2$, and $Q_3$ lie in
disjoint ranges $[1,r]$, $(r,K]$, and $(K,L]$ respectively. Using the
Chinese Remainder Theorem, we choose $v = v(Q_1,Q_2,Q_3) \in \N$ as in
\cref{lemma_choice_of_v_1}, and average over the grid
$(Q_{\delta,L} m + 1,\, Q_{\delta,L} n + v)$, where $Q_{\delta,L}$ is the
highly divisible integer in \eqref{Q_delta_l_def}. The disjointness of the
prime factors of $Q_1$, $Q_2$, and $Q_3$ is essential for the
congruences in \cref{lemma_choice_of_v_1} to be solvable.
With this choice of $v$, averaging over the grid
$(Q_{\delta,L} m + 1,\, Q_{\delta,L} n + v)$ allows us to factor out $Q_j$
from $P_j(Q_{\delta,L} m + 1,\, Q_{\delta,L} n + v)$ for each
$j \in \{1,2,3\}$, and ensures that
$T_{P_j(Q_{\delta,L} m + 1,\, Q_{\delta,L} n + v)} F$ is asymptotically
independent of the specific value of $Q_l$ for $l \neq j$
(see \cref{L: main a=c P_2 reducible}).
Then for every $j \in \{1,2,3\}$, we show that there exists a factor
$\X_j$ of the system $(X, \X, \mu, T_n)$ such that if we write
$F = F_{1,j} + F_{2,j}$, where $F_{1,j} := \E(F \mid \X_j)$ and
$F_{2,j} \perp F_{1,j}$, then
\begin{equation}\label{proof plan eq 8}
\cesE_{(m,n) \in S_{\delta} \cap [N]^2}
\lVert \cesE_{Q_j} T_{P_j(Q_{\delta,L} m + 1, Q_{\delta,L} n + v)}
F_{2,j} \rVert_{L^2(\mu)}
\end{equation}
goes to $0$ when all relevant parameters go to infinity
(see for example \eqref{aux_eq_3}).
Moreover, the term $T_{P_j(Q_{\delta,L} m + 1, Q_{\delta,L} n + v)}
F_{1,j}$ concentrates around $F_{1,j}$, namely
\begin{equation}\label{proof plan eq 9}
\max_{Q_j} \cesE_{(m,n) \in S_{\delta} \cap [N]^2}
\lVert T_{P_j(Q_{\delta,L} m + 1, Q_{\delta,L} n + v)} F_{1,j}
- F_{1,j} \rVert_{L^2(\mu)}
\end{equation}
is asymptotically $0$ when all relevant parameters go to infinity
(see for example \eqref{eq_useful_2}).
The factors $\X_j$ are chosen so that whenever $F \in \X_j$ and $f$ lies
in the support of the spectral measure $\sigma_F$, the oscillating factor $\exp(F_N(f,L))$
(or $\exp(G_{P,N}(f,L))$) arising from the concentration estimates
(Proposition \ref{linear_conc} or \ref{quad_conc}) equals $1$ for
$N > L$ and $L$ sufficiently large. This property is crucial in proving
\eqref{proof plan eq 9}.
The specific factors are
$\X_1 := \X_{\A}$,
$\X_2 := \X_{p,\textup{fin.supp.}}$,
and $\X_3 := \X_{p,\textup{fin.supp.},P_3}$,
defined in
\eqref{def_X_A}, \eqref{def_X_p_fin_supp}, and
\eqref{def_X_p_fin_supp_P} respectively. It is also crucial that the factor $\X_1$ does not contain
modified Dirichlet characters (unlike the other two factors), since otherwise we could not
force concentration at $F_{1,1}$, as $P_1(1,0) \neq 1$ (while $P_j(1,0)=1 $ for $j=2,3$).
Using \eqref{proof plan eq 8} and \eqref{proof plan eq 9}, and averaging
over $Q_1 \in \Phi_r$, $Q_2 \in \Phi_{r,K}$, and
$Q_3 \in \Phi_{K,L,P_3}$, together with $(m,n) \in S_{\delta}$ in
\eqref{proof plan eq 3}, and then letting
$N \to \infty$, $L \to \infty$, $K \to \infty$, $r \to \infty$, and
finally $\delta \to 0^{+}$, in this order, yields
\begin{multline*}
\suplim_{\delta,r,k,L,N} \, \cesE_{Q_1\in \Phi_r, Q_2\in \Phi_{r,K}, Q_3\in \Phi_{K,L,P_3}}
\cesE_{(m,n) \in S_{\delta} \cap [N]^2}
\int_X F \cdot \prod_{j=1}^3
T_{P_j(Q_{\delta,L} m + 1, Q_{\delta,L} n + v)} F \, d\mu \\
= \int_X F \cdot \prod_{j=1}^3 \E(F \mid \X_j) \, d\mu
\geq \Big( \int_X F \, d\mu \Big)^4,
\end{multline*}
where the last estimate follows from \cref{chu-lemma}. Note that we use $\limsup$ so that we can take advantage of the fact
that the set $S_\delta$ has positive upper density.
This completes the proof of \cref{proof plan thm 1}.
\section{Proof of the main theorem when $a=c$ or $b=c$}\label{Sec a=c or b=c}
In this section we prove \cref{T:Main} when $a=c$ or $b=c$.
The cases are symmetric, so we can assume that $a=c$.
One easily verifies that
\begin{equation}\label{eq_param_a=c_2}
x=k \,(m^2 - ab n^2),\:\: y=k \, 2 a mn, \:\:
z=k \, ( m^2 + ab n^2),\:\:\: k, m,n\in \N,
\end{equation}
are solutions of $a x^2 + b y^2=c z^2$ when $a=c$. Let
\begin{equation}\label{quadratic_forms_a=c}
P_1(m,n):=2a mn, \:\:\:
P_2(m,n):=m^2 - ab n^2, \:\:\:
P_3(m,n):= m^2 + ab n^2.
\end{equation}
It is easy to see that, since $a, b \neq 0$,
the polynomials $P_1, P_2, P_3$ are pairwise independent,
a property that is implicitly used in our arguments when we try
to solve some simultaneous congruences. Also, for all $(m,n)\in \N^2$,
$P_1(m,n), P_3(m,n) >0$, and if $m> \sqrt{ab}n$,
we also have that $P_2(m,n) >0$.
The case $a=c$ of \cref{T:Main} follows from the following result.
\begin{theorem}\label{T: main a=c}
Let $(X,\CX,\mu,T_n)$ be a pretentious multiplicative action, let
$F\in L^{\infty}(\mu)$ with $F\geq 0$,
and let $P_1,P_2,P_3$ be as in \eqref{quadratic_forms_a=c}. Then for every $\e>0$ we have
\begin{equation}\label{T: main a=c eq}
\int_X F \cdot T_{P_1(m,n)} F \cdot T_{P_2(m,n)} F \cdot T_{P_3(m,n)} F \, d\mu
\geq \Big(\int_X F \, d \mu \Big)^4 -\e
\end{equation}
for a set of $(m,n)\in \N^2$, $m>\sqrt{ab} n$, with positive upper
density.
\end{theorem}
Note that $P_1$ is reducible and
$P_3$ is irreducible. We split the proof of \cref{T: main a=c} into two cases,
according to whether $P_2$ is irreducible or not.
Throughout \cref{Sec a=c or b=c}, $\Phi_r$, $\Phi_{r,K}$,
$\Phi_{r,K,P}$ are as in \eqref{eq_foln_def_1}, \eqref{eq_foln_def_2},
$Q_{\delta,L}$ is as in \eqref{Q_delta_l_def}. Furthermore,
$\omega_P$ is as in \eqref{omega_def}, and $\D$, $F_N(f,L)$, $G_{P,N}(f,L)$
are as in \eqref{E: definition of pretentious distance},
\eqref{def_F_N}, \eqref{def_G_P,N}.
Finally, $X_{\A}$, $X_{p,\textup{fin.supp.}}$, and $X_{p,\textup{fin.supp.},P}$
are the factors defined respectively in \eqref{def_X_A},
\eqref{def_X_p_fin_supp}, and \eqref{def_X_p_fin_supp_P}.
\subsection{$P_2$ reducible}\label{Sec : a=c_P_2_reducible}
Here we handle the case when $P_2$ is reducible.
A representative example is when $a=b=c=1$, which
corresponds to the classical Pythagorean equation
$x^2 + y^2 = z^2,$
in which case
\eqref{quadratic_forms_a=c} gives
$$
P_1(m,n)=2mn, \quad P_2(m,n) = m^2-n^2, \quad
P_3(m,n)=m^2+n^2.
$$
Since $P_2$ is reducible, we have that $ab$ is a perfect square, i.e.,
$ab=\gamma ^2$ for some $\gamma \in \N$, and we have that
$P_2(m,n)=(m - \gamma n)(m + \gamma n)$.
Let $\kappa \in \N$ so that $1-\kappa \gamma \neq 0$.
To prove the result, given $Q_1 \in \Phi_r$, $Q_2 \in \Phi_{r,K}$, and
$Q_3 \in \Phi_{K,L, P_3}$,
we will choose a suitable integer $v$, depending on $\delta, r,K,L,Q_1,Q_2, Q_3$, and
we will average over the grid
$(Q_{\delta,L} m+1, Q_{\delta,L} n +v)$ in \eqref{T: main a=c eq} where $Q_{\delta,L}$
is as in \eqref{Q_delta_l_def}.
From now on, we assume that $r$ is large enough so that for $p>r$ we have
$\omega_{P_3}(p)\in \{0,2\}$ and
\begin{equation}\label{E1: p not dividing}
p\nmid 2\kappa \gamma ab(1+ab)(1-\kappa \gamma)
(1+\kappa \gamma)(1+\kappa ^2 ab).
\end{equation}
\begin{lemma}\label{lemma_choice_of_v_1}
Let $\delta>0$, $r,K,L\in \N$ with $r<K<L/2$,
$P_1,P_2,P_3$ be as in \eqref{quadratic_forms_a=c}, and let $Q_1 \in \Phi_r$,
$Q_2 \in \Phi_{r,K}$, $Q_3 \in \Phi_{K,L,P_3}$. Then there is an integer
$v$ with $0\leq v \leq Q_{\delta, L}-1$ such that
\begin{enumerate}[(i)]
\item \label{E1: lemma_choice_of_v_1} $(Q_{\delta, L},v)=Q_1$,
$\frac{v}{Q_1}\equiv 1 \pmod {Q_1}$,
\item \label{E2: lemma_choice_of_v_1}
$(Q_{\delta, L},1-\gamma v)= Q_2$,
$\frac{1-\gamma v}{Q_2} \equiv Q_2 ^{-1}
\pmod{{Q_1}}$,
\item \label{E3: lemma_choice_of_v_1} $(Q_{\delta, L},1+\gamma v)=1$,
${1+\gamma v}\equiv 1 \pmod {{Q_1}}$,
\item \label{E4: lemma_choice_of_v_1}
$(Q_{\delta, L},P_3(1,v))=Q_3$,
$\frac{P_3(1,v)}{Q_3} \equiv Q_3^{-1} \pmod{{Q_1}}$.
\end{enumerate}
\end{lemma}
\begin{proof}
Let $p\in \P$ with $K<p \leq L$ and $\omega_{P_3}(p)=2$, and consider the
polynomial $g(x):=P_3(1,x)-p^{\theta_p(Q_3)}$.
Then $\textup{mod}\: p$ the polynomial
becomes $g(x)=P_3(1,x)$, and it has two simple roots
$z_1, z_2$ with $z_1 \not \equiv z_2 \pmod p$. Using
Hensel's lemma, we may lift $z_1$ to a unique root
$\zeta = \zeta (p, Q_3)$ of $g \mod {p^{\theta_p(Q_3)+1}}$.
Using the Chinese Remainder theorem we may then choose $v$ with
$$0\leq v \leq \prod_{p\leq K} p^{4K} \prod_{K<p\leq L}p^{1+3L/2}-1\leq
Q_{\delta,L}-1$$
such that
\begin{itemize}
\item $v\equiv Q_1 \pmod {p^{2 \theta_p(Q_1)}}$ for
$p\leq r$,
\item $v\equiv \gamma ^{-1} - \gamma ^{-1}p^{\theta_p(Q_2)}$
$\pmod {p^{\theta_p(Q_2)+1}}$ for $r< p\leq K$,
\item $v\equiv \kappa \pmod {p} $ for $K<p\leq L$ with $\omega_{P_3}(p)=0$,
\item $v\equiv \zeta(p,Q_3) \pmod {p^{\theta_p(Q_3)+1}}$ for
$K<p\leq L$ with $\omega_{P_3}(p)=2$.
\end{itemize}
For $p\leq r$, $v\equiv Q_1 \pmod{p^{2\theta_p(Q_1)}}$, so
$p^{\theta_p(Q_1)}\parallel v$. For $r<p\leq K$, $v\equiv \gamma^{-1}
\pmod {p}$, so $(p,v)=1$. For $K<p \leq L$ with $\omega_{P_3}(p)=0$,
$v\equiv \kappa \pmod {p}$, so in view of \eqref{E1: p not dividing} we
get $(p,v)=1$. Finally, for
$K<p \leq L$ with $\omega_{P_3}(p)=2$, we have $v\equiv \zeta(p,Q_3)
\pmod {p^{\theta_p(Q_3)+1}}$, so
$v\equiv z_1 \pmod {p}$. If $p\mid v$, then $p\mid z_1$, which
implies that $P_3(1,z_1) \equiv 1 \pmod {p}$. On the other hand,
$P_3(1,z_1)\equiv 0 \pmod {p}$, so we reach a contradiction. Therefore,
$(p,v)=1$. Combining all the above, we infer that
$(Q_{\delta, L},v)=Q_1$. For each $p\leq r$,
$\frac{v}{Q_1} \equiv 1 \pmod {p^{\theta_p(Q_1)}}$, so
$\frac{v}{Q_1} \equiv 1 \pmod {Q_1}$. Therefore, $v$ satisfies
\eqref{E1: lemma_choice_of_v_1}.
For $p\leq r$, $1-\gamma v\equiv 1 \pmod{p}$, so
$(p,1-\gamma v)=1$. For $r<p\leq K$, one sees that
$1-\gamma v \equiv p^{\theta_p(Q_2)}
\pmod {p^{\theta_p(Q_2)+1}}$, so $p^{\theta_p(Q_2)} \parallel v$.
For $K<p \leq L$ with $\omega_{P_3}(p)=0$,
$1-\gamma v \equiv 1-\kappa \gamma \pmod {p}$, which in view of
\eqref{E1: p not dividing} implies $(p,-\gamma v)=1$.
Finally, for
$K<p \leq L$ with $\omega_{P_3}(p)=2$, we have
$1-\gamma v\equiv 1-\gamma \zeta(p,Q_3) \pmod{p^{\theta_p(Q_3)+1}}$, so
$1-\gamma v\equiv1-\gamma z_1 \pmod {p}$.
If $p\mid 1-\gamma v$, then $p\mid 1-\gamma z_1$, which
implies that $P_2(1,z_1) \equiv 0 \pmod {p}$. Recall that
$P_3(1,z_1)\equiv 0 \pmod {p}$, so adding we obtain that
$2\equiv 0 \pmod p$, which contradicts
\eqref{E1: p not dividing}. Therefore, $(p,1-\gamma v)=1$.
Combining all the above, we infer that
$(Q_{\delta, L},1-\gamma v)=Q_2$. From the choice of $v$ we have
$1-\gamma v\equiv 1 \pmod {Q_1}$, so
$\frac{1-\gamma v}{Q_2} \equiv Q_2 ^{-1} \pmod {Q_1}$, and we
conclude that $v$ satisfies
\eqref{E2: lemma_choice_of_v_1}.
For $p\leq r$, $1+\gamma v\equiv 1 \pmod{p}$, so
$(p,1+\gamma v)=1$. For $r<p\leq K$, if $p\mid 1+\gamma v$, then,
since $p\mid 1-\gamma v$, we get that $p\mid 2$, which contradicts
\eqref{E1: p not dividing}. Therefore, $(p,1+\gamma v)=1$.
For $K<p \leq L$ with $\omega_{P_3}(p)=0$,
$1+\gamma v \equiv 1+\kappa \gamma \pmod {p}$, which in view of
\eqref{E1: p not dividing} implies that $(p,1+\gamma v)=1$.
Finally, for
$K<p \leq L$ with $\omega_{P_3}(p)=2$,
if $p\mid 1+ \gamma v$, then as before we have
$P_2(1,z_1) \equiv 0 \pmod {p}$, and since $P_3(1,z_1)\equiv 0 \pmod {p}$,
we get that $2\equiv 0 \pmod p$, contradicting
\eqref{E1: p not dividing}. Therefore, $(p,1+\gamma v)=1$.
Combining all the above, we obtain that
$(Q_{\delta, L},1-\gamma v)=1$ and
$1+\gamma v\equiv 1 \pmod {Q_1}$, so $v$ satisfies
\eqref{E3: lemma_choice_of_v_1}.
It remains to prove \eqref{E4: lemma_choice_of_v_1}.
For $p\leq r$, $P_3(1,v)\equiv 1 \pmod{p}$, so
$(p,P_3(1,v))=1$. For $r<p\leq K$, if $p\mid P_3(1,v)$, then
since $p\mid P_2(1,v)$, we get that $p\mid P_2(1,v)+P_3(1,v)=2$,
which contradicts
\eqref{E1: p not dividing}. Therefore, $(p, P_3(1,v))=1$.
For $K<p \leq L$ with $\omega_{P_3}(p)=0$,
$P_3(1,v) \equiv 1+\kappa^2 ab \pmod {p}$, which in view of
\eqref{E1: p not dividing} implies that $(p, P_3(1,v))=1$.
Finally, for
$K<p \leq L$ with $\omega_{P_3}(p)=2$, we have
$v\equiv \zeta(p,Q_3) \pmod {p^{\theta_p(Q_3)+1}}$, which implies that
$P_3(1,v)-p^{\theta_p(Q_3)}\equiv 0 \pmod{p^{\theta_p(Q_3) +1}}$.
We deduce that $p^{\theta_p(Q_3)}\parallel P_3(1,v)$.
Combining all the above, we obtain that
$(Q_{\delta, L},P_3(1,v))=Q_3$. In addition, since $v\equiv 0 \pmod {Q_1}$,
we have $P_3(1,v) \equiv 1 \pmod {Q_1}$, so
$\frac{P_3(1,v)}{Q_3}\equiv Q_3 ^{-1} \pmod {Q_1}$, so $v$ satisfies
\eqref{E4: lemma_choice_of_v_1}. This concludes the proof
of the lemma.
\end{proof}
For the remainder of \cref{Sec : a=c_P_2_reducible}, we suppress the
dependence of $v$ on the parameters $\delta, r,K,L,Q_1,Q_2,Q_3$, and simply
write $v$ in place of $v(\delta,r,K,L,Q_1,Q_2,Q_3)$. In several other cases
we leave the dependence on the parameters explicit, at the cost of
making some formulas less aesthetically pleasing, as omitting it would
obscure the underlying arguments.
Given $\delta>0$, $r, K, L, N\in \N$ with
$r<K<L<N $, $Q_1 \in \Phi_r$, $Q_2\in \Phi_{r,K}$,
$Q_3\in \Phi_{K,L,P_3}$, and $m,n\in \N$, let
\begin{enumerate}
\item \label{R_1 def} $R_1 (\delta, r, K, L, N, Q_1, Q_2, Q_3,m,n):= 2a
({Q_{\delta,L}} m+1)
(\frac{Q_{\delta,L}}{Q_1}n+ \frac{v}{Q_1} )$,
\item \label{R_2 def} $R_2 (\delta, r, K, L, N, Q_1, Q_2, Q_3,m,n):=
(\frac{Q_{\delta,L}}{Q_2} (m -\gamma n)+
\frac{1-\gamma v}{Q_2})({Q_{\delta,L}}
(m+\gamma n)+ 1+\gamma v )$,
\item \label{R_3 def} $R_3 (\delta, r, K, L, N, Q_1, Q_2, Q_3,m,n):=
\frac{P_3(Q_{\delta,L} m +1, Q_{\delta,L} n +v )}{Q_3}$,
\end{enumerate}
where $v$ is as in \cref{lemma_choice_of_v_1}.
Then for $j\in \{1,2,3\}$ we have
\begin{equation}\label{eq_extr_1}
P_j(Q_{\delta,L} m+1, Q_{\delta,L} n +v)= Q_j \,
R_j (\delta, r, K, L, N, Q_1, Q_2, Q_3,m,n).
\end{equation}
Sometimes in the notation we suppress the dependence of $R_j$ on
the parameters
$\delta, r, K$, $L$, $N,$ $ Q_1,$ $Q_2, Q_3$, and we only write $R_j(m,n)$.
Observe that if $m>2\sqrt{ab}n=2 \gamma n$ and $n$ is sufficiently large, then since
$0\leq v \leq Q_{\delta,L}-1$, we have that $Qm+1 > \sqrt{ab} (Qn+v)$,
so $P_2(Qm+1, Qn+v)>0$, while for $j=1,3$ we have $P_j(Qm+1, Qn+v)>0$ for all
$m,n \in \N$. For each $\delta\in(0,1/2)$, let
\begin{equation}\label{S_delta_a=c_P_2_reduc}
S_{\delta}:=\{(m,n)\in \N^2 : m> 2 \sqrt{ab}n, \:
|(P_j(m,n))^i -1|\leq \delta \,
\text{ for } j=1,2,3\}.
\end{equation}
From \cref{Cor S_d dense} we have that $S_{\delta}$ has
positive upper density.
When $P_2$ is reducible, \cref{T: main a=c} follows from the following
result.
\begin{proposition}\label{P: main a=c P_2 reducible}
Let $(X,\CX,\mu,T_n)$ be a pretentious multiplicative action, let
$F\in L^{\infty}(\mu)$ with $F\geq 0$,
let $P_1,P_2,P_3$ be as in \eqref{quadratic_forms_a=c}, and suppose that $P_2$ is reducible.
Also, let $S_{\delta}$ be as in \eqref{S_delta_a=c_P_2_reduc} and $v=v(\delta, r,K,L,Q_1,Q_2,Q_3)$ be as in
\cref{lemma_choice_of_v_1}.
Then
\begin{multline}\label{main_term_1}
\suplim_{\delta, r,K,L,N}\:
\cesE_{Q_{1}\in \Phi_{r}} \cesE_{Q_{2}\in \Phi_{r,K}}
\cesE_{Q_{3}\in \Phi_{K,L,P_3}}
\frac{1}{\overline{\textup{d}}(S_{\delta})} \cesE_{(m,n) \in [N]^2}
\1_{S_{\delta}}(m,n)\cdot \\
\int_{X} F\cdot \prod_{j=1} ^3
T_{P_j(Q_{\delta,L} m+1, Q_{\delta,L} n +v)} F
\, d \mu \geq \Big(\int_X F \, d \mu\Big)^4.
\end{multline}
\end{proposition}
\begin{proof}[Proof that \cref{P: main a=c P_2 reducible} implies
\cref{T: main a=c} when $P_2$ is reducible]
Let $\e>0$. Using \eqref{main_term_1} we infer that there are
$\delta\in (0,1/2)$, $r,K,L \in \N$, and $Q_1 \in \Phi_r$,
$Q_2 \in \Phi_{r,K},$ $Q_3 \in \Phi_{r,K,P_3}$ such that
\begin{multline}\label{E: prop implies thm 1}
\limsup_{N\to \infty} \frac{1}{\overline{\textup{d}}(S_{\delta})} \cesE_{(m,n) \in [N]^2}
\1_{S_{\delta}}(m,n)\cdot \int_{X} F\cdot \prod_{j=1} ^3
T_{P_j(Q_{\delta,L} m+1, Q_{\delta,L} n +v)} F
\, d \mu \\ \geq \Big(\int_X F \, d \mu\Big)^4 -\frac{\e}{2}.
\end{multline}
From \eqref{E: prop implies thm 1} and the fact that $S_{\delta}$
has positive upper density, we get that the set
\begin{equation}\label{E: prop implies thm 2}
\Big\{ (m,n)\in S_{\delta} : \int_{X} F\cdot \prod_{j=1} ^3
T_{P_j(Q_{\delta,L} m+1, Q_{\delta,L} n +v)} F
\, d \mu \geq \Big(\int_X F \, d \mu\Big)^4 -\e\Big\}
\end{equation}
has positive upper density. Now, if a set $A\subset \N^2$ has positive
upper density, and $Q,w_1, w_2 \in \N$, then $B=\{(Qm+w_1, Qn + w_2):
(m,n)\in A\}$ also has positive upper density. In addition, as we
commented earlier, if $(m,n)\in S_{\delta}$, then $m>2\sqrt{ab}n$, which
implies $Qm+1 > \sqrt{ab}(Qn+v)$. Combining the above with
\eqref{E: prop implies thm 2} we deduce that the set
\begin{equation*}
\{ (m,n)\in \N^2 : m>\sqrt{ab}n, \:
\int_X F \cdot \prod_{j=1} ^3 T_{P_j(m,n)} F \, d\mu
\geq \Big(\int_X F \, d \mu \Big)^4 - \e\}
\end{equation*}
has positive upper density. This concludes the proof.
\end{proof}
It remains to prove \cref{P: main a=c P_2 reducible}. For that we need the
following lemma.
\begin{lemma}\label{L: main a=c P_2 reducible}
Let $(X, \X, \mu, T_n)$, $F$, and
$R_1, R_2, R_3$ be as in \eqref{R_1 def}-\eqref{R_3 def}.
Then for every $\delta>0$, all sufficiently large $r\in \N$, and all $K\in \N$ with $K>r$, we have
\begin{align}
& \lim_{L, N} \,
\max_{Q_1 \in \Phi_r, \; Q_2, Q_{2}' \in \Phi_{r,K}, \:Q_3, Q_{3}' \in \Phi_{K,L,P_3} }
\cesE_{(m,n) \in [N]^2}
\label{L: main a=c P_2 reducible eq 1}\\
& \hspace{10mm} \lVert
T_{R_1(\delta, r, K, L, N, Q_1, Q_2, Q_3,m,n)} F -
T_{R_1(\delta, r, K, L, N, Q_1, Q_2 ', Q_3 ',m,n)} F
\rVert _{L^2(\mu)}=0, \nonumber \\
& \lim_{L, N} \,
\max_{Q_1, Q_1 ' \in \Phi_r, \; Q_2 \in \Phi_{r,K}, \:Q_3, Q_{3}' \in \Phi_{K,L,P_3} }
\cesE_{(m,n) \in [N]^2}
\label{L: main a=c P_2 reducible eq 2} \\
& \hspace{10mm} \lVert
T_{R_2(\delta, r, K, L, N, Q_1, Q_2, Q_3,m,n)} F -
T_{R_2(\delta, r, K, L, N, Q_1 ', Q_2, Q_3 ',m,n)} F
\rVert _{L^2(\mu)}=0, \nonumber \\
& \lim_{ L, N}\,
\max_{Q_1, Q_1 ' \in \Phi_r, \; Q_2, Q_2 ' \in \Phi_{r,K}, \:Q_3
\in \Phi_{K,L, P_3} }
\cesE_{(m,n) \in [N]^2}
\label{L: main a=c P_2 reducible eq 3} \\
& \hspace{10mm} \lVert
T_{R_3(\delta, r, K, L, N, Q_1, Q_2, Q_3,m,n)} F -
T_{R_3(\delta, r, K, L, N, Q_1 ', Q_2 ', Q_3,m,n)} F
\rVert _{L^2(\mu)}=0. \nonumber
\end{align}
\end{lemma}
\begin{remark}
\cref{L: main a=c P_2 reducible} essentially says that
in the integral in \eqref{main_term_1},
the first term is independent of $Q_2, Q_3$, the second term
is independent of $Q_1, Q_3$ and the third term
is independent of $Q_1, Q_2$.
\end{remark}
\begin{proof}
Let $\delta>0$ be fixed. We can assume that
$\lVert F \rVert_{L^{\infty}(\mu)}\leq 1$.
%In the proof, it is always implicit that $r<K<L<N$, and $L$ is
%sufficiently large depending on $r,K$.
We start by proving \eqref{L: main a=c P_2 reducible eq 1}.
Note by \eqref{spec_eq_2} we have
\begin{multline*}
\lVert
T_{R_1(\delta, r, K, L, N, Q_1, Q_2, Q_3,m,n)} F -
T_{R_1(\delta, r, K, L, N, Q_1, Q_2 ', Q_3 ',m,n)} F
\rVert _{L^2(\mu)}= \\
\lVert
f(R_1(\delta, r, K, L, N, Q_1, Q_2, Q_3,m,n)) -
f(R_1(\delta, r, K, L, N, Q_1, Q_2 ', Q_3 ',m,n))
\rVert _{L^2(\sigma_{F}(f))},
\end{multline*}
so after applying the Cauchy-Schwarz inequality and Fatou's lemma
multiple times, one
sees that in order to establish \eqref{L: main a=c P_2 reducible eq 1}, it
suffices to prove that for each $f\in \M_p $ we have
\begin{multline}\label{eq_conc_1}
\lim_{L, N}\,
\max_{Q_1 \in \Phi_r, \; Q_2, Q_{2}' \in \Phi_{r,K}, \:Q_3, Q_{3}' \in \Phi_{K,L,P_3}} \cesE_{(m,n) \in [N]^2}
\\
|f(R_1(\delta, r, K, L, N, Q_1, Q_2, Q_3,m,n))
\overline{f(R_1(\delta, r, K, L, N, Q_1, Q_2 ', Q_3 ',m,n))}-1 |
=0.
\end{multline}
Let $f\in \M_p$ and let $\chi,t$ so that $f\sim \chi\cdot n^{it}$. Let
$q_{\chi}$ be the conductor of $\chi$, and suppose that
$r,L$ are large enough so that $q_{\chi}$ divides
$ \frac{Q_{\delta, L}}{Q_1}, Q_1$ for every $Q_1 \in \Phi_r$. Using
\cref{lemma_choice_of_v_1} \eqref{E1: lemma_choice_of_v_1} and
applying \cref{linear_conc} and \cref{uniform_in_Q} we get that
\begin{align*}
&\limsup_{N\to \infty} \max_{Q_1 \in \Phi_r, \; Q_2, Q_{2}' \in \Phi_{r,K}, \:Q_3, Q_{3}' \in \Phi_{K,L,P_3}}
\cesE_{(m,n) \in [N]^2} \\
&|f(R_1(\delta, r, K, L, N, Q_1, Q_2, Q_3,m,n)) \cdot
\overline{f(R_1(\delta, r, K, L, N, Q_1, Q_2 ', Q_3 ',m,n))}-1 | \\
&\ll \limsup_{N\to \infty}
|\exp(2\Re (F_N(f,L))) -1| +
\D(f, \chi \cdot n^{it}; L, \infty)
+L^{-1/2}.
\end{align*}
Since $f\sim \chi \cdot n^{it}$, we have that $2\Re(F_N(f,L))$ converges to
$2\sum_{p>L} \frac{1}{p}(\Re(f(p) \overline{\chi(p)}p^{-it}) -1)$ as $N\to\infty$. Then
\eqref{eq_conc_1} follows by
taking $L\to \infty$ in the last relation. This concludes the proof
of \eqref{L: main a=c P_2 reducible eq 1}.
For the proof of \eqref{L: main a=c P_2 reducible eq 2}, arguing as
before, it suffices to prove that for every
$f\in \M_p $
\begin{multline}\label{eq_conc_2}
\lim_{ L, N}\,
\max_{Q_1, Q_1 ' \in \Phi_r, \; Q_2, \in \Phi_{r,K}, \:Q_3, Q_{3}' \in \Phi_{K,L,P_3} }
\cesE_{(m,n) \in [N]^2}
\\
| f({R_2(\delta, r, K, L, N, Q_1, Q_2, Q_3,m,n)}) \cdot
\overline{f(R_2(\delta, r, K, L, N, Q_1 ', Q_2, Q_3 ',m,n))}
-1| =0.
\end{multline}
Again, let $f\in \M_p$, let $\chi,t$ so that $f\sim \chi\cdot n^{it}$,
and let $q_{\chi}$ be the conductor of $\chi$.
Suppose that $r,L$ are sufficiently large so that $q_{\chi}$ divides
$\frac{Q_{\delta, L}}{Q_{2}}, {Q_1}$ for every $Q_1 \in \Phi_r$ and
$Q_{2} \in \Phi_{r,K}$.
Using \cref{lemma_choice_of_v_1} \eqref{E2: lemma_choice_of_v_1},
\eqref{E3: lemma_choice_of_v_1} and
applying \cref{linear_conc}, \cref{uniform_in_Q} and
\cref{bilinear_lemma} we get that
\begin{align*}
&\limsup_{N\to \infty}
\max_{Q_1, Q_1 ' \in \Phi_r, \; Q_2 \in \Phi_{r,K}, \:Q_3, Q_{3}' \in \Phi_{K,L,P_3} }
\cesE_{(m,n) \in [N]^2} \\
&| f({R_2(\delta, r, K, L, N, Q_1, Q_2, Q_3,m,n)})\cdot
\overline{f(R_2(\delta, r, K, L, N, Q_1 ', Q_2, Q_3 ',m,n))}
-1| \\ & \ll_{a,b}
\limsup_{N\to \infty} |\exp(4\Re (F_N(f,L))) -1| +
\D(f, \chi \cdot n^{it}; L, \infty) +L^{-1/2},
\end{align*}
so again \eqref{eq_conc_2} follows by taking $L\to \infty$ in the last
relation and using that $f\sim \chi \cdot n^{it}$.
Finally, to prove \eqref{L: main a=c P_2 reducible eq 3},
it suffices to prove that for every
$f\in \M_p $ we have
\begin{multline}\label{eq_conc_3}
\lim _{ L, N}\,
\max_{Q_1, Q_1 ' \in \Phi_r, \; Q_2, Q_2' \in \Phi_{r,K}, \:Q_3 \in \Phi_{K,L,P_3} }
\cesE_{(m,n) \in [N]^2}
\\
| f({R_3(\delta, r, K, L, N, Q_1, Q_2, Q_3,m,n)}) \cdot
\overline{f(R_3(\delta, r, K, L, N, Q_1 ', Q_2 ', Q_3,m,n))}
-1| =0.
\end{multline}
Let $f\in \M_p$, let $\chi,t$ so that $f\sim \chi\cdot n^{it}$, and let
$q_{\chi}$ be the conductor of $\chi$.
Suppose that $r,L$ are sufficiently large so that $q_{\chi}$ divides
$\frac{Q_{\delta, L}}{{Q}_{3}}, {Q_1}$ for every
$Q_1 \in \Phi_r$ and $Q_{3} \in \Phi_{K,L,P_3}$.
Then, using \cref{lemma_choice_of_v_1} \eqref{E4: lemma_choice_of_v_1},
and applying \cref{quad_conc} and \cref{uniform_in_Q} we get that
\begin{align*}
&\limsup_{N\to \infty} \max_{Q_1, Q_1 ' \in \Phi_r, \; Q_2, Q_2' \in \Phi_{r,K}, \:Q_3 \in \Phi_{K,L,P_3} }
\cesE_{(m,n)\in[N]^2} \\
&| f({R_3(\delta, r, K, L, N, Q_1, Q_2, Q_3,m,n)})\cdot
\overline{f(R_3(\delta, r, K, L, N, Q_1 ', Q_2 ', Q_3,m,n))}
-1|\\ & \ll_{P_3}
\limsup_{N\to \infty} |\exp(2\Re(G_{P_3,N}(f,L))) -1| +
\D(f, \chi \cdot n^{it}; L, \infty)+L^{-1/2}.
\end{align*}
Again, using $f\sim \chi \cdot n^{it}$, we have that
$2\Re(G_{P_3,N}(f,L))$ converges as $N\to\infty$ to
$$2\sum_{p>L} \frac{\omega_{P_3}(p)}{p}\big(\Re(f(p)
\overline{\chi(p)}p^{-it}) -1\big),$$ and
\eqref{eq_conc_3} follows by
taking $L\to \infty$ in the last relation. This concludes the proof
of the lemma.
\end{proof}
We are now ready to prove \cref{P: main a=c P_2 reducible}. The argument is somewhat lengthy, so we divide it into several steps.
\begin{proof}[Proof of \cref{P: main a=c P_2 reducible}]
We can assume that
$\lVert F\rVert_{L^{\infty}(\mu)}\leq 1$. In the proof, it is always
implicit that $r<K<L<N$, and that each of the previous variables is
sufficiently large depending on the smaller ones.
In view of \eqref{eq_extr_1}, the quantity in
$\suplim_{\delta,r,K,L,N}$ on the left-hand side in
\eqref{main_term_1} is equal to
\begin{multline*}
\cesE_{Q_{1}\in \Phi_{r}} \cesE_{Q_{2}\in \Phi_{r,K}}
\cesE_{Q_{3}\in \Phi_{K,L,P_3}} \frac{1}{\overline{\textup{d}}(S_{\delta})} \cesE_{(m,n) \in [N]^2}
\1_{S_{\delta}}(m,n) \cdot \\
\int_{X} F \cdot T_{Q_1 R_1(m,n)} F \cdot
T_{Q_2 R_2(m,n)} F \cdot
T_{{Q}_3 R_3(m,n)} F
\, d \mu.
\end{multline*}
For each $r,K,L\in \N$, choose any $Q_{1,r}\in \Phi_r$,
$Q_{2, r, K} \in \Phi_{r,K}$ and $Q_{3,K,L}\in \Phi_{K,L,P_3}$.
Using \cref{L: main a=c P_2 reducible}, one sees that in order to prove
\eqref{main_term_1}, it suffices to prove that
\begin{multline}\label{main_term_3}
\suplim_{\delta,r, K, L,N}
\int_{X} F \cdot \frac{1}{\overline{\textup{d}}(S_{\delta})}
\cesE_{(m,n) \in [N]^2}
\1_{S_{\delta}}(m,n) \cdot \\
\Big( \cesE_{Q_{1}\in \Phi_{r}} T_{Q_1 R_1(\delta, r, K, L, N, Q_1, Q_{2,r,K} , Q_{3,K,L} ,m,n)} F \Big)
\cdot \Big( \cesE_{Q_{2}\in \Phi_{r,K}}
T_{Q_2 R_2(\delta, r, K, L, N, Q_{1,r},
Q_2 , Q_{3,K,L},m,n)} F \Big) \cdot \\ \Big( \cesE_{Q_{3}\in \Phi_{K,L,P_3}}
T_{{Q}_3 R_3(\delta, r, K, L, N, Q_{1,r}, Q_{2,r,K} , Q_3 ,m,n)} F \Big)
d \mu \geq \Big(\int_X F \, d \mu\Big)^4 .
\end{multline}
Our plan is to handle each of the three averages separately,
showing in each case that it can be replaced by the conditional
expectation of $F$ with respect to a suitable factor of the action.
While these steps share many similarities, there are also key
technical differences. For the sake of completeness, we present the
full details, even at the expense of relatively longer argument.
\smallskip
{\bf Step 1 (Dealing with the average over $Q_1$).} Write $F=F_1 + F_2$, where $F_1:=\E(F | {\X}_{\A})$ and $F_2 \perp F_1$.
Then the integral on the left-hand side of \eqref{main_term_3} is equal to
\begin{multline}\label{main_term_4}
\sum_{j\in \{1,2\}}\int_{X} F \cdot \frac{1}{\overline{\textup{d}}(S_{\delta})}
\cesE_{(m,n) \in [N]^2}
\1_{S_{\delta}}(m,n) \cdot \\
\Big( \cesE_{Q_{1}\in \Phi_{r}} T_{Q_1 R_1(\delta, r, K, L, N, Q_1, Q_{2,r,K} , Q_{3,K,L} ,m,n)} F_j \Big)
\cdot \Big( \cesE_{Q_{2}\in \Phi_{r,K}}
T_{Q_2 R_2(\delta, r, K, L, N, Q_{1,r},
Q_2 , Q_{3,K,L},m,n)} F \Big) \cdot \\ \Big( \cesE_{Q_{3}\in \Phi_{K,L,P_3}}
T_{{Q}_3 R_3(\delta, r, K, L, N, Q_{1,r}, Q_{2,r,K} , Q_3 ,m,n)} F \Big)
\, d \mu.
\end{multline}
The term corresponding to $j=2$ in \eqref{main_term_4} is bounded by
\begin{equation}\label{aux_quant_1}
\frac{1}{\overline{\textup{d}}(S_{\delta})}
\cesE_{(m,n) \in [N]^2} \big \lVert
\cesE_{Q_{1}\in \Phi_{r}}
T_{Q_1 R_1(\delta, r, K, L, N, Q_1, Q_{2,r,K} , Q_{3,K,L} ,m,n)} F_2
\big \rVert_{L^2(\mu)}.
\end{equation}
We will show that the quantity in \eqref{aux_quant_1} goes to $0$ as
$N\to \infty, L \to \infty, K \to \infty, r \to \infty$.
Since $\sigma_{F_2}$ is supported on
$\M_p \setminus \A$, using \eqref{spec_eq_1} and applying Fatou's lemma
multiple times, one sees
that it suffices to prove that for every
$f\in \M_p \setminus \A$, we have
\begin{equation*}
\lim _{r,K,L,N} \,
\cesE_{(m,n) \in [N]^2} \big |
\cesE_{Q_{1}\in \Phi_{r}} f({Q_1
R_1(\delta, r, K, L, N, Q_1, Q_{2,r,K} , Q_{3,K,L} ,m,n)}) \big |=0.
\end{equation*}
Let $f\in \M_p \setminus \A$, let $\chi, t$ so that $f\sim \chi \cdot n^{it}$,
and suppose that $r,L$ are large enough so that the conductor $q_{\chi}$ of
$\chi$ divides $\frac{Q_{\delta, L}}{Q_1}$, $Q_1$ for all
$Q_1 \in \Phi_r$.
Using \cref{lemma_choice_of_v_1} \eqref{E1: lemma_choice_of_v_1} and
applying \cref{linear_conc} and \cref{uniform_in_Q}, we deduce that
\begin{align}\label{eq_useful_1}
&\limsup_{N\to \infty} \cesE_{(m,n) \in [N]^2} \big |
\cesE_{Q_{1}\in \Phi_{r}} f({Q_1
R_1(\delta, r, K, L, N, Q_1, Q_{2,r,K} , Q_{3,K,L} ,m,n)})
\big | \ll \\
&\limsup_{N\to \infty} \cesE_{(m,n) \in [N]^2} \big |
\cesE_{Q_{1}\in \Phi_{r}} f(Q_1) \, Q_1 ^{-it} \, f(2a)
\, Q_{\delta,L}^{2it} \, m^{it} \, n^{it}
\exp(2F_N(f,L)) \big | \nonumber \\
& \hspace{20mm} + \D(f, \chi \cdot n^{it}; L, \infty)
+L^{-1/2}
= \nonumber \\
& \limsup_{N\to \infty} |\exp(2F_N(f,L))|\cdot
\big| \cesE_{Q_1 \in \Phi_r} f(Q_1)\, Q_1 ^{-it}\,
\big|+ \D(f, \chi \cdot n^{it}; L, \infty)
+L^{-1/2}. \nonumber
\end{align}
Note that $|\exp(2F_N(f,L))|=\exp(2\Re(F_N(f,L))) \to |\exp(-2\D^2(f, \chi \cdot n^{it};L, \infty))|$ as $N\to \infty$.
Since $f \in \M_p \setminus \A$, we have ${f(n)} \, n^{-it}\neq 1$, and one easily deduces that
\begin{equation*}
\lim_{r\to\infty} \cesE_{Q_1 \in \Phi_r} {f(Q_1)} \, Q_1^{-it} = 0.
\end{equation*}
Combining this with \eqref{eq_useful_1} we infer that for every $\delta>0$ we have
\begin{equation}\label{aux_eq_3}
\lim_{r,K,L,N}\,
\cesE_{(m,n) \in [N]^2} \big \lVert
\cesE_{Q_{1}\in \Phi_{r}}
T_{Q_1 R_1(\delta, r, K, L, N, Q_1, Q_{2,r,K} , Q_{3,K,L} ,m,n)} F_2
\big \rVert_{L^2(\mu)}=0.
\end{equation}
On the other hand, for the term corresponding to $j=1$ in
\eqref{main_term_4} we claim that
\begin{multline}\label{eq_useful_2}
\lim_{\delta, r, K, L, N}\,
\max_{Q_1 \in \Phi_r}
\frac{1}{\overline{\textup{d}}(S_{\delta})}\cdot
\cesE_{(m,n) \in [N]^2} \1_{S_{\delta}}(m,n)\cdot \\
\lVert
T_{Q_1 R_1(\delta, r, K, L, N, Q_1, Q_{2,r,K} , Q_{3,K,L} ,m,n)}
F_1 -F_1 \rVert_{L^2(\mu)} =0.
\end{multline}
To see this, observe that for $f\in \A$ with $f=n^{it}$, using Taylor
expansion we have
\begin{multline}\label{eq_useful_5}
\frac{1}{\overline{\textup{d}}(S_{\delta})}
\cesE_{(m,n) \in [N]^2} \1_{S_{\delta}}(m,n) |f(Q_1)
f(R_1(\delta, r, K, L, N, Q_1, Q_{2,r,K} , Q_{3,K,L} ,m,n)) - 1 | \\
= \frac{1}{\overline{\textup{d}}(S_{\delta})}
\cesE_{(m,n) \in [N]^2} \1_{S_{\delta}}(m,n)\, | Q_{\delta, L}^{2it}\, P_1(m,n)^{it} -1|
+\oh_{t,Q_1, Q_2, Q_3, N\to \infty}(1).
\end{multline}
Using \eqref{Q delta L def} and \eqref{S_delta_a=c_P_2_reduc} we deduce
that
\begin{equation*}
\frac{1}{\overline{\textup{d}}(S_{\delta})}
\cesE_{(m,n) \in [N]^2} \1_{S_{\delta}}(m,n) | Q_{\delta, L}^{2it} \, P_1(m,n)^{it} -1| \ll_{t} \delta
\frac{1}{\overline{\textup{d}}(S_{\delta})}
\cesE_{(m,n) \in [N]^2} \1_{S_{\delta}}(m,n),
\end{equation*}
so the quantity in \eqref{eq_useful_5} goes to $0$ when we take $\lim_{\delta, r, K, L, N}$.
Equation \eqref{eq_useful_2} then follows from \eqref{spec_eq_2} and
the fact that the spectral measure of
$F_1$ is supported on $\A$.
Using \eqref{main_term_4}, \eqref{aux_eq_3}, \eqref{eq_useful_2} we see
that in order to prove \eqref{main_term_3}, it suffices to show that
\begin{multline}\label{main_term_5}
\suplim_{\delta,r, K, L,N}
\int_{X} F \cdot F_1 \cdot \frac{1}{\overline{\textup{d}}(S_{\delta})}
\cesE_{(m,n) \in [N]^2}
\1_{S_{\delta}}(m,n) \cdot \\
\Big( \cesE_{Q_{2}\in \Phi_{r,K}}
T_{Q_2 R_2(\delta, r, K, L, N, Q_{1,r},
Q_2 , Q_{3,K,L},m,n)} F \Big) \cdot \\ \Big( \cesE_{Q_{3}\in \Phi_{K,L,P_3}}
T_{{Q}_3 R_3(\delta, r, K, L, N, Q_{1,r}, Q_{2,r,K} , Q_3 ,m,n)} F \Big)
\, d \mu \geq \Big(\int_X F \, d \mu\Big)^4 .
\end{multline}
{\bf Step 2 (Dealing with the average over $Q_2$).} Writing $F=F_3 + F_4$ where $F_3:=\E(F | \X_{p, \textup{fin.supp.}})$
and $F_4 \perp F_3$, one sees that the main term
\eqref{main_term_5} is equal to
\begin{multline}\label{main_term_6}
\sum_{j\in \{3,4\}}\int_{X} F \cdot F_1 \cdot \frac{1}{\overline{\textup{d}}(S_{\delta})}
\cesE_{(m,n) \in [N]^2}
\1_{S_{\delta}}(m,n) \cdot \\
\Big( \cesE_{Q_{2}\in \Phi_{r,K}}
T_{Q_2 R_2(\delta, r, K, L, N, Q_{1,r},
Q_2 , Q_{3,K,L},m,n)} F_j \Big) \cdot \\
\Big( \cesE_{Q_{3}\in \Phi_{K,L,P_3}}
T_{{Q}_3 R_3(\delta, r, K, L, N, Q_{1,r}, Q_{2,r,K} , Q_3 ,m,n)} F \Big)
\, d \mu.
\end{multline}
The term corresponding to $j=4$ in \eqref{main_term_6} is bounded by
\begin{equation}\label{aux_quant_4}
\frac{1}{\overline{\textup{d}}(S_{\delta})}
\cesE_{(m,n) \in [N]^2} \big \lVert
\cesE_{Q_{2}\in \Phi_{r,K}}
T_{Q_2 R_2(\delta, r, K, L, N, Q_{1,r}, Q_2 , Q_{3,K,L},m,n)} F_4
\big \rVert_{L^2(\mu)}.
\end{equation}
We will show that the quantity in \eqref{aux_quant_4} goes to $0$ as
$N\to \infty, L \to \infty, K \to \infty, r \to \infty$.
Since $\sigma_{F_4}$ is supported on
$\M_p \setminus \M_{p, \textup{fin.supp.}}$, using \eqref{spec_eq_1}
and applying Fatou's lemma multiple times, we see
that it suffices to show that if
$f\in \M_p \setminus \M_{p, \textup{fin.supp.}}$, then
\begin{equation*}
\lim _{r,K,L,N} \,
\cesE_{(m,n) \in [N]^2} \big |
\cesE_{Q_{2}\in \Phi_{r,K}} f(Q_2
R_2(\delta, r, K, L, N, Q_{1,r}, Q_2 , Q_{3,K,L},m,n)) \big |=0.
\end{equation*}
Let $f\in \M_p \setminus \M_{p, \textup{fin.supp.}}$,
let $\chi, t$ so that $f\sim \chi \cdot n^{it}$,
and suppose that $r,L$ are large enough so that the conductor $q_{\chi}$ of
$\chi$ divides $Q_{1,r}$ $\frac{Q_{\delta, L}}{Q_2}$ for all $Q_2 \in \Phi_{r,K}$.
Using \cref{lemma_choice_of_v_1} \eqref{E2: lemma_choice_of_v_1},
\eqref{E3: lemma_choice_of_v_1}, and
applying \cref{linear_conc}, \cref{uniform_in_Q},
and \cref{bilinear_lemma}, we deduce that
\begin{align}\label{eq_useful_3}
&\limsup_{N\to \infty} \cesE_{(m,n) \in [N]^2} \big |
\cesE_{Q_{2}\in \Phi_{r,K}} f({Q_2
R_2(\delta, r, K, L, N, Q_{1,r}, Q_{2} , Q_{3,K,L} ,m,n)})
\big | \ll_{a,b} \\
&\limsup_{N\to \infty} \cesE_{(m,n) \in [N]^2} \big |
\cesE_{Q_{2}\in \Phi_{r,K}} f({Q_2})\, \overline{\chi(Q_2)} \, Q_2 ^{-it}
\, P_2(m,n)^{it} \, Q_{\delta, L}^{2it}
\exp(2F_N(f,L)) \big | \nonumber \\
& \hspace{20mm} + \D(f, \chi \cdot n^{it}; L, \infty)
+L^{-1/2}
= \nonumber \\
& \limsup_{N\to \infty} |\exp(2F_N(f,L))|\cdot
\big| \cesE_{Q_2 \in \Phi_{r,K}} f(Q_2)\, \overline{\chi(Q_2)} \, Q_2 ^{-it}
\big|+ \D(f, \chi \cdot n^{it}; L, \infty)
+L^{-1/2}. \nonumber
\end{align}
Using that $f\sim \chi \cdot n^{it}$, as before we infer that
$\lim_{L\to \infty} \lim_{N\to \infty} |\exp(2F_N(f,L))|=1$.
Since $f \in \M_p \setminus \M_{p, \textup{fin.supp.}}$, for each $r$ there is
$p>r$ so that $f(p) \neq \chi(p)\, p^{it}$, and since $\Phi_{r,K}$ is asymptotically
invariant under dilation by $p$ as $K\to \infty$ we have
\begin{multline*}
{f(p)}\, \overline{\chi(p)}\, p^{-it} \,
\cesE_{Q_{2} \in \Phi_{r,K}} f(Q_{2})\, \overline{\chi(Q_{2})}\, Q_{2} ^{-it}
=
\cesE_{Q_{2} \in p \Phi_{r,K}} f(Q_{2}) \, \overline{\chi(Q_{2})} \,Q_{2} ^{-it}
\\ =
\cesE_{Q_{2} \in \Phi_{r,K}} f(Q_{2})\, \overline{\chi(Q_{2})} \, Q_{2} ^{-it}
+ \oh_{K\to \infty}(1).
\end{multline*}
Using $f(p) \neq \chi(p)\, p^{it}$, the above implies that
\begin{equation}\label{eq_aux_5}
\lim_{K\to \infty} \cesE_{Q_{2} \in \Phi_{r,K}} \, f(Q_{2})
\, \overline{\chi(Q_{2})} \, Q_{2} ^{-it} = 0.
\end{equation}
Combining this with \eqref{eq_useful_3} we infer that
\begin{equation}\label{aux_eq_4}
\lim_{r,K,L,N}\,
\cesE_{(m,n) \in [N]^2} \big \lVert
\cesE_{Q_{2}\in \Phi_{r,K}}
T_{Q_2 R_2(\delta, r, K, L, N, Q_{1,r}, Q_{2} , Q_{3,K,L} ,m,n)} F_4
\big \rVert_{L^2(\mu)}=0.
\end{equation}
On the other hand, for the term corresponding to $j=3$ in
\eqref{main_term_6}, we claim that
\begin{multline}\label{eq_useful_4}
\lim_{\delta, r, K, L, N}\,
\max_{Q_2 \in \Phi_{r,K}}
\frac{1}{\overline{\textup{d}}(S_{\delta})}\cdot
\cesE_{(m,n) \in [N]^2} \1_{S_{\delta}}(m,n)\cdot \\
\lVert
T_{Q_2 R_2(\delta, r, K, L, N, Q_{1,r}, Q_{2} , Q_{3,K,L} ,m,n)} F_3
-F_3 \rVert_{L^2(\mu)} =0.
\end{multline}
To see this, let $f\in M_{p, \textup{fin.supp.}}$, and
let $\chi, t$ and $r_0 \in \N$ so that
$f(p)=\chi(p)\, p^{it}$ for all primes $p>r_0$. Also,
suppose that $r,L$ are sufficiently
large so that $r>r_0$ and the conductor $q_{\chi}$ of $\chi$
divides $Q_{1,r}$ and $\frac{Q_{\delta, L}}{Q_2}$ for every
$Q_{2} \in \Phi_{r,K}$.
Using \cref{lemma_choice_of_v_1} \eqref{E2: lemma_choice_of_v_1},
\eqref{E3: lemma_choice_of_v_1} and
applying \cref{linear_conc}, \cref{uniform_in_Q}, and
\cref{bilinear_lemma}, we get that
\begin{align}\label{eq_useful_6}
&\limsup_{N\to \infty} \max_{Q_2 \in \Phi_{r,K}}
\frac{1}{\overline{\textup{d}}(S_{\delta})}
\cesE_{(m,n) \in [N]^2} \1_{S_{\delta}}(m,n) \cdot \\
& \hspace{20mm}\big|f(Q_2)
f(R_2(\delta, r, K, L, N, Q_{1,r}, Q_{2} , Q_{3,K,L} ,m,n)) -1\big|
\ll_{a,b} \nonumber \\
& \limsup_{N\to \infty} \max_{Q_2 \in \Phi_{r,K}}
\frac{1}{\overline{\textup{d}}(S_{\delta})}
\cesE_{(m,n) \in [N]^2} \1_{S_{\delta}}(m,n) \cdot
|f(Q_{2}) \, \overline{\chi(Q_{2})} \, Q_{2}^{-it} \cdot \nonumber \\
& \hspace{20mm} P_2(m,n)^{it} \, Q_{\delta, L}^{2it} \,
\exp(2F_N(f,L)) -1| + \D(f, \chi \cdot n^{it}, L, \infty) +
L^{-1/2}. \nonumber
\end{align}
For $p>r$, $f(p)=\chi(p)\, p^{it}$, which implies that
$f(Q_{2})= \chi(Q_{2}) \, Q_{2}^{it}$, and since $L>r_0$, we have
$F_N(f,L)=0$. Therefore, the last quantity in \eqref{eq_useful_6}
is equal to
\begin{equation*}
\limsup_{N\to \infty}
\frac{1}{\overline{\textup{d}}(S_{\delta})}
\cesE_{(m,n) \in [N]^2} \1_{S_{\delta}}(m,n) \cdot
| P_2(m,n)^{it} \, Q_{\delta, L}^{2it}
-1| + \oh_{L\to \infty}(1) \ll_{t} \delta + \oh_{L\to \infty}(1),
\end{equation*}
where for the last estimate we used
\eqref{Q delta L def} and \eqref{S_delta_a=c_P_2_reduc}. We infer that
for $f\in M_{p, \textup{fin.supp.}}$, we have
\begin{multline*}
\lim_{\delta, r, K, L, N}\,
\max_{Q_2 \in \Phi_{r,K}}
\frac{1}{\overline{\textup{d}}(S_{\delta})}
\cesE_{(m,n) \in [N]^2} \1_{S_{\delta}}(m,n) \cdot \\
\big|f(Q_2)
f(R_2(\delta, r, K, L, N, Q_{1,r}, Q_{2} , Q_{3,K,L} ,m,n)) -1\big|=0,
\end{multline*}
and since the spectral measure of $F_3$ is supported on
$M_{p, \textup{fin.supp.}}$, \eqref{eq_useful_4} follows.
Using \eqref{main_term_6}, \eqref{aux_eq_4}, \eqref{eq_useful_4} we see
that in order to prove \eqref{main_term_5}, it suffices to show that
\begin{multline}\label{main_term_7}
\suplim_{\delta,r, K, L,N}
\int_{X} F \cdot F_1 \cdot F_3 \cdot \frac{1}{\overline{\textup{d}}(S_{\delta})}
\cesE_{(m,n) \in [N]^2}
\1_{S_{\delta}}(m,n) \cdot \\
\Big( \cesE_{Q_{3}\in \Phi_{K,L,P_3}}
T_{{Q}_3 R_3(\delta, r, K, L, N, Q_{1,r}, Q_{2,r,K} , Q_3 ,m,n)} F \Big)
d \mu \geq \Big(\int_X F \, d \mu\Big)^4 .
\end{multline}
{\bf Step 3 (Dealing with the average over $Q_3$).} Finally, write $F=F_5 + F_6$ where
$F_5:=\E(F | \X_{p, \textup{fin.supp.}, P_3})$
and $F_6 \perp F_5$.
Then the main term in \eqref{main_term_7} is equal to
\begin{multline}\label{main_term_8}
\sum_{j\in \{5,6\}}\int_{X} F \cdot F_1 \cdot F_3 \cdot \frac{1}{\overline{\textup{d}}(S_{\delta})}
\cesE_{(m,n) \in [N]^2}
\1_{S_{\delta}}(m,n) \cdot \\
\Big( \cesE_{Q_{3}\in \Phi_{K,L,P_3}}
T_{{Q}_3 R_3(\delta, r, K, L, N, Q_{1,r}, Q_{2,r,K} , Q_3 ,m,n)} F_j \Big)
d \mu.
\end{multline}
The term corresponding to $j=6$ in \eqref{main_term_6} is bounded by
\begin{equation}\label{aux_quant_6}
\frac{1}{\overline{\textup{d}}(S_{\delta})}
\cesE_{(m,n) \in [N]^2} \big \lVert
\cesE_{Q_{3}\in \Phi_{K,L,P_3}}
T_{{Q}_3 R_3(\delta, r, K, L, N, Q_{1,r}, Q_{2,r,K} , Q_3 ,m,n)} F_6
\big \rVert_{L^2(\mu)}.
\end{equation}
We will show that the quantity in \eqref{aux_quant_6} goes to $0$ as
$N\to \infty, L \to \infty, K \to \infty, r \to \infty$.
Using that $\sigma_{F_6}$ is supported on
$\M_p \setminus \M_{p, \textup{fin.supp.},P_3}$,
it suffices to show that for every
$f\in \M_p \setminus \M_{p, \textup{fin.supp.},P_3}$, we have
\begin{equation*}
\lim _{r,K,L,N} \,
\cesE_{(m,n) \in [N]^2} \big |
\cesE_{Q_{3}\in \Phi_{K,L,P_3}} f(Q_3
R_3(\delta, r, K, L, N, Q_{1,r}, Q_{2,r,K} , Q_3 ,m,n)) \big |=0.
\end{equation*}
Let $f\in \M_p \setminus \M_{p, \textup{fin.supp.},P_3}$,
let $\chi, t$ so that $f\sim \chi \cdot n^{it}$,
and suppose that $r,L$ are large enough so that the conductor $q_{\chi}$ of
$\chi$ divides $Q_{1,r}$ $\frac{Q_{\delta, L}}{Q_3}$ for all $Q_3 \in
\Phi_{K,L,P_3}$.
Using \cref{lemma_choice_of_v_1} \eqref{E4: lemma_choice_of_v_1} and
applying \cref{quad_conc} and \cref{uniform_in_Q}, we deduce that
\begin{align}\label{eq_useful_7}
&\limsup_{N\to \infty} \cesE_{(m,n) \in [N]^2} \big |
\cesE_{Q_{3}\in \Phi_{K,L,P_3}} f({Q_3
R_3(\delta, r, K, L, N, Q_{1,r}, Q_{2,r,K} , Q_{3} ,m,n)})
\big | \ll_{P_3} \\
&\limsup_{N\to \infty} \cesE_{(m,n) \in [N]^2} \big |
\cesE_{Q_{3}\in \Phi_{K,L,P_3}} f({{Q}_3})\, \overline{\chi({Q}_3)} \,
{Q}_3 ^{-it} \,Q_{\delta, L}^{2it} \, P_3(m,n)^{it}\,
\exp(G_{P_3,N}(f,L)) \big | \nonumber \\
& \hspace{20mm} + \D(f, \chi \cdot n^{it}; L, \infty)
+L^{-1/2}
= \nonumber \\
& \limsup_{N\to \infty} |\exp(G_{P_3,N}(f,L))|\cdot
\big| \cesE_{Q_3 \in \Phi_{K,L,P_3}} f(Q_3)\, \overline{\chi(Q_3)} \,Q_3 ^{-it}
\big|+ \D(f, \chi \cdot n^{it}; L, \infty)
+L^{-1/2}. \nonumber
\end{align}
Since $f\sim \chi \cdot n^{it}$, we have
$\lim_{L\to \infty} \lim_{N\to \infty} |\exp(G_{P_3,N}(f,L))|=1$. Moreover,
since $f \in \M_p \setminus \M_{p, \textup{fin.supp.},P_3}$, for each
$K$ there is
$p>K$ such that $\omega_{P_3}(p)=2$ and $f(p) \neq \chi(p)\, p^{it}$. Using
that $\Phi_{K,L,P_3}$ is asymptotically
invariant under dilation by $p$ as $L\to \infty$ we have
\begin{align*}
f(p)\, \overline{\chi(p)}\, p^{-it} \,
\cesE_{Q_{3} \in \Phi_{K,L,P_3}} f(Q_{3})\,
\overline{\chi(Q_{3})}\,Q_{3} ^{-it}
&=
\cesE_{Q_{3} \in p \Phi_{K,L,P_3}} f(Q_{3})\, \overline{\chi(Q_{3})}\,
Q_{3} ^{-it}
\\ &=
\cesE_{Q_{3} \in \Phi_{K,L,P_3}} f(Q_{3})\, \overline{\chi(Q_{3})}\,
Q_{3} ^{-it}
+ \oh_{L\to \infty}(1).
\end{align*}
Using $f(p) \neq \chi(p)\, p^{it}$ for some $p>K$, the above implies that
\begin{equation}\label{eq_aux_7}
\lim_{L\to \infty} \cesE_{Q_{3} \in \Phi_{K,L,P_3}} f(Q_{3}) \,
\overline{\chi(Q_{3})}\, Q_{3} ^{-it} = 0.
\end{equation}
Combining this with \eqref{eq_useful_7} we infer that
\begin{equation}\label{aux_eq_5}
\lim_{r,K,L,N}\,
\cesE_{(m,n) \in [N]^2} \big \lVert
\cesE_{Q_{3}\in \Phi_{K,L,P_3}}
T_{Q_3 R_3(\delta, r, K, L, N, Q_{1,r}, Q_{2,r,K} , Q_{3} ,m,n)} F_6
\big \rVert_{L^2(\mu)}=0.
\end{equation}
On the other hand, for the term corresponding to $j=5$ in
\eqref{main_term_8}, we claim that
\begin{multline}\label{eq_useful_8}
\lim_{\delta, r, K, L, N}\,
\max_{Q_3 \in \Phi_{K,L,P_3}}
\frac{1}{\overline{\textup{d}}(S_{\delta})}\cdot
\cesE_{(m,n) \in [N]^2} \1_{S_{\delta}}(m,n)\\
\lVert
T_{Q_3 R_3(\delta, r, K, L, N, Q_{1,r}, Q_{2,r,K} , Q_{3} ,m,n)} F_5
-F_5 \rVert_{L^2(\mu)} =0.
\end{multline}
Let $f\in M_{p, \textup{fin.supp.},P_3}$, and
let $\chi, t$ and $r_0 \in \N$ so that $f\sim\chi \cdot n^{it}$ and
$f(p)=\chi(p)\, p^{it}$ for all primes $p>r_0$ with $\omega_{P_3}(p)>0$.
Also, suppose that $r,L$ are sufficiently
large so that $r>r_0$ and the conductor $q_{\chi}$ of $\chi$
divides $Q_{1,r}$ and $\frac{Q_{\delta, L}}{Q_3}$ for every
$Q_{3} \in \Phi_{K,L,P_3}$.
Using \cref{lemma_choice_of_v_1} \eqref{E4: lemma_choice_of_v_1} and
applying \cref{quad_conc} and \cref{uniform_in_Q}, we deduce that
\begin{align}\label{eq_useful_9}
&\limsup_{N\to \infty} \max_{Q_3 \in \Phi_{K,L,P_3}}
\frac{1}{\overline{\textup{d}}(S_{\delta})}
\cesE_{(m,n) \in [N]^2} \1_{S_{\delta}}(m,n) \cdot \\
& \hspace{20mm}\big|f(Q_3) \,
f(R_3(\delta, r, K, L, N, Q_{1,r}, Q_{2,r,K} , Q_{3} ,m,n)) -1\big|
\ll_{P_3} \nonumber \\
& \limsup_{N\to \infty} \max_{Q_3 \in \Phi_{K,L,P_3}}
\frac{1}{\overline{\textup{d}}(S_{\delta})}
\cesE_{(m,n) \in [N]^2} \1_{S_{\delta}}(m,n) \cdot
|f(Q_{3})\, \overline{\chi(Q_{3})} \, Q_{3}^{-it} \cdot \nonumber \\
& \hspace{20mm} P_3(m,n)^{it} \, Q_{\delta, L}^{2it} \,
\exp(G_{P_3,N}(f,L)) -1| + \D(f, \chi \cdot n^{it}, L, \infty) +
L^{-1/2}. \nonumber
\end{align}
For $p>r$ with $\omega_{P_3}(p)>0$, we have
$f(p)=\chi(p)\,p^{it}$, which implies that
$f(Q_{3})= \chi(Q_{3}) \,Q_{3}^{it}$, and since $L>r_0$, we have
$G_{P_3,N}(f,L)=0$. Therefore, the last quantity in \eqref{eq_useful_7}
is equal to
\begin{equation*}
\limsup_{N\to \infty}
\frac{1}{\overline{\textup{d}}(S_{\delta})}
\cesE_{(m,n) \in [N]^2} \1_{S_{\delta}}(m,n) \cdot
| P_3(m,n)^{it} \, Q_{\delta, L}^{2it}
-1| + \oh_{L\to \infty}(1) \ll_{t} \delta + \oh_{L\to \infty}(1),
\end{equation*}
where for the last estimate we used
\eqref{Q delta L def} and \eqref{S_delta_a=c_P_2_reduc}.
From the above we obtain that
for $f\in M_{p, \textup{fin.supp.},P_3}$, we have
\begin{multline*}
\suplim_{\delta, r, K, L, N}
\max_{Q_3 \in \Phi_{K,L,P_3}}
\frac{1}{\overline{\textup{d}}(S_{\delta})}
\cesE_{(m,n) \in [N]^2} \1_{S_{\delta}}(m,n) \cdot \\
\big|f(Q_3)
f(R_3(\delta, r, K, L, N, Q_{1,r}, Q_{2,r,K} , Q_{3} ,m,n)) -1\big|=0,
\end{multline*}
and since the spectral measure of $F_3$ is supported on
$M_{p, \textup{fin.supp.}}$, \eqref{eq_useful_8} follows.
\smallskip
{\bf Step 4 (Finishing the argument).} Using \eqref{main_term_8}, \eqref{aux_eq_5}, \eqref{eq_useful_8}, we see
that in order to prove \eqref{main_term_7} it suffices to show that
\begin{equation}\label{main_term_10}
\int_{X} F \cdot F_1 \cdot F_3 \cdot F_5\, d\mu \cdot \suplim_{\delta,r, K, L,N}\, \frac{1}{\overline{\textup{d}}(S_{\delta})}
\cesE_{(m,n) \in [N]^2}
\1_{S_{\delta}}(m,n) \geq \Big(\int_X F \, d \mu\Big)^4 .
\end{equation}
Since
$$
\int_X F\cdot F_1 \cdot F_3 \cdot F_5 \, d \mu =
\int_X F\cdot \E(F | {\X}_{\A}) \cdot
\E(F | \X_{p, \textup{fin.supp.}})
\cdot \E(F | \X_{p, \textup{fin.supp.},P_3}) \, d \mu,
$$
it follows from \cref{chu-lemma} that the last quantity is greater or equal to $(\int F\, d\mu)^4$. This establishes \eqref{main_term_10}
and concludes the proof of the proposition.
\end{proof}
\subsection{$P_2$ irreducible}\label{Sec : a=c_P_2_irreducible}
Here we handle the case where $P_2$ is irreducible.
A representative example is the equation $x^2+2y^2=z^2$, in which case
\eqref{quadratic_forms_a=c} gives
$$
P_1(m,n)=2mn, \quad P_2(m,n) = m^2-2n^2, \quad
P_3(m,n)=m^2+2n^2.
$$
The proof of this case is very similar to the case
when $P_2$ is reducible, so we only point out the
necessary changes.
Let $\kappa \in \N$ so that $P_2(1, \kappa)= 1- ab \kappa^2 \neq 0$, and
from now on, suppose that $r$ is large enough so that
for $p>r$ we have $\omega_{P_2}(p), \omega_{P_3}(p)\in \{0,2\}$ and
\begin{equation}\label{E2: p not dividing}
p\nmid 2\kappa ab(1+ab)(1- ab \kappa ^2)(1+ ab \kappa ^2).
\end{equation}
To prove the result, given $Q_1 \in \Phi_r$, $Q_2 \in \Phi_{r,K,P_2}$,
$Q_3 \in \Phi_{K,L, P_3}$,
we will choose an integer $v$ depending on $\delta, r,K,L,Q_1,Q_2, Q_3$, and
we will average over the grid
$(Q_{\delta,L} m+1, Q_{\delta,L} n +v)$ in \eqref{T: main a=c eq}.
\begin{lemma}\label{lemma_choice_of_v_1_a=c_P_2_irred}
Let $\delta>0$, $r,K,L\in \N$ with $r<K<L/2$, $P_1,P_2,P_3$ be as in \eqref{quadratic_forms_a=c}, and let $Q_1 \in \Phi_r$,
$Q_2 \in \Phi_{r,K,P_2}$, $Q_3 \in \Phi_{K,L,P_3}$. Then there is an integer
$v$ with $0\leq v \leq Q_{\delta, L}-1$ such that
\begin{enumerate}[(i)]
\item \label{E1: lemma_choice_of_v_1_a=c_P_2_irred} $(Q_{\delta, L},v)=Q_1$,
$\frac{v}{Q_1}\equiv 1 \pmod {Q_1}$,
\item \label{E2: lemma_choice_of_v_1_a=c_P_2_irred}
$(Q_{\delta, L},P_2(1,v))= Q_2$,
$\frac{P_2(1,v)}{Q_2} \equiv Q_2 ^{-1}
\pmod{{Q_1}}$,
\item \label{E3: lemma_choice_of_v_1_a=c_P_2_irred}
$(Q_{\delta, L},P_3(1,v))=Q_3$,
$\frac{P_3(1,v)}{Q_3} \equiv Q_3^{-1} \pmod{{Q_1}}$.
\end{enumerate}
\end{lemma}
\begin{proof}
Let $p\in \P$ with $r<p \leq K$ and $\omega_{P_2}(p)=2$, and consider the
polynomial $g_2(x):=P_2(1,x)-p^{\theta_p(Q_2)}$.
Then $\textup{mod}\: p$ the polynomial
becomes $g_2(x)=P_2(1,x)$, and it has two simple roots
$z_2, z_2'$ with $z_2 \not \equiv z_2' \pmod p$. Using
Hensel's lemma, we may lift $z_2$ to a unique root
$\zeta_2 = \zeta_2 (p, Q_2)$ of $g_2 \mod {p^{\theta_p(Q_2)+1}}$.
Analogously, for $p\in \P$ with $K<p \leq L$ and $\omega_{P_3}(p)=2$,
we may find $\zeta_3 = \zeta_3 (p, Q_3)$ so that if
$g_3(x):=P_3(1,x)-p^{\theta_p(Q_3)}$, then $g_3(\zeta_3)\equiv 0
\pmod {p^{\theta_p(Q_3)+1}}$.
Using the Chinese Remainder theorem we may then choose $v$ with
$0\leq v \leq Q_{\delta,L}-1$
such that
\begin{itemize}
\item $v\equiv Q_1 \pmod {p^{2 \theta_p(Q_1)}}$ for
$p\leq r$,
\item $v\equiv \kappa$
$\pmod {p}$ for $r< p\leq K$ with $\omega_{P_2}(p)=0$,
\item $v\equiv \zeta_2(p, Q_2)$
$\pmod {p^{\theta_p(Q_2)+1}}$ for $r< p\leq K$ with
$\omega_{P_2}(p)=2$,
\item $v\equiv \kappa \pmod {p} $ for $K<p\leq L$ with $\omega_{P_3}(p)=0$,
\item $v\equiv \zeta_3(p,Q_3) \pmod {p^{\theta_p(Q_3)+1}}$ for
$K<p\leq L$ with $\omega_{P_3}(p)=2$.
\end{itemize}
The rest of the proof is very similar to the proof of
\cref{lemma_choice_of_v_1}, so it is omitted.
\end{proof}
As in \cref{Sec : a=c_P_2_reducible}, for each $\delta\in(0,1/2)$, let
\begin{equation}\label{S_delta_a=c_P_2_irreduc}
S_{\delta}:=\{(m,n)\in \N^2 : m> 2 \sqrt{ab}n, \:
|(P_j(m,n))^i -1|\leq \delta \, \text{ for }
j=1,2,3\}.
\end{equation}
and recall that in view of \cref{Cor S_d dense}, $S_{\delta}$ has
positive upper density. When $P_2$ is irreducible, \cref{T: main a=c} follows from the following
result.
\begin{proposition}\label{P: main a=c P_2 irreducible}
Let $(X,\CX,\mu,T_n)$ be a pretentious multiplicative action, let
$F\in L^{\infty}(\mu)$ with $F\geq 0$,
let $P_1,P_2,P_3$ be as in \eqref{quadratic_forms_a=c}, and suppose that $P_2$ is irreducible.
Also, let $S_{\delta}$ be as in \eqref{S_delta_a=c_P_2_irreduc} and $v=v(\delta, r,K,L,Q_1,Q_2,Q_3)$ be as in
\cref{lemma_choice_of_v_1_a=c_P_2_irred}.
Then
\begin{multline}\label{E: P: main a=c P_2 irreducible}
\suplim_{\delta, r,K,L,N}\:
\cesE_{Q_{1}\in \Phi_{r}} \cesE_{Q_{2}\in \Phi_{r,K,P_2}}
\cesE_{Q_{3}\in \Phi_{K,L,P_3}}
\frac{1}{\overline{\textup{d}}(S_{\delta})} \cesE_{(m,n) \in [N]^2}
\1_{S_{\delta}}(m,n)\cdot \\
\int_{X} F\cdot \prod_{j=1} ^3
T_{P_j(Q_{\delta,L} m+1, Q_{\delta,L} n +v)} F
\, d \mu \geq \Big(\int_X F \, d \mu\Big)^4.
\end{multline}
\end{proposition}
The proof that \cref{P: main a=c P_2 irreducible} implies
\cref{T: main a=c} when $P_2$ is irreducible is identical to the case
when $P_2$ is reducible, so it is omitted.
To prove \cref{P: main a=c P_2 irreducible}, one first has to prove an
analog of \cref{L: main a=c P_2 reducible}, which asserts that in the
integral in \eqref{E: P: main a=c P_2 irreducible},
the first term is independent of $Q_2, Q_3$, the second term
is independent of $Q_1, Q_3$ and the third term
is independent of $Q_1, Q_2$, and then using that prove
\cref{P: main a=c P_2 irreducible}. Both proofs are very similar to the
case when $P_2$ is reducible, and the only essential difference is that
in order to handle the term
$T_{P_2(Q_{\delta,L} m+1, Q_{\delta,L} n +v)} F$, one has to use
\cref{quad_conc} instead of \cref{linear_conc}. We omit the details.
\section{Proof of the main theorem when $a+b=c$}\label{Sec a+b=c}
In this section we prove \cref{T:Main} when $a+b=c$.
One easily verifies that
\begin{equation}\label{eq_param_a+b=c_2}
x=k (m^2 -2bmn - ab n^2),\:\: y=k (m^2 +2amn - ab n^2), \:\:
z=k ( m^2 + ab n^2),\:\:\: k, m,n\in \N,
\end{equation}
are solutions of $a x^2 + b y^2=c z^2$ when $a+b=c$. Let
\begin{equation}\label{quadratic_forms_a+b=c}
P_1(m,n):=m^2 + ab n^2, \:\:\:
P_2(m,n):=m^2 -2bmn - ab n^2, \:\:\:
P_3(m,n):= m^2 +2amn - ab n^2.
\end{equation}
It is easy to see that, since $a, b \neq 0$,
the polynomials $P_1, P_2, P_3$ are pairwise independent.
This property will be implicitly used in order to
solve certain simultaneous congruences. Let
\begin{equation}\label{gamma_defn}
\gamma:=\max\{4b, \sqrt{2ab}\}>0.
\end{equation}
For all $(m,n)\in \N^2$, $P_1(m,n)>0$, and if
$m> \gamma n$, then we have that also $P_2(m,n)$, $P_3(m,n)>0$.
The case $a+b=c$ of \cref{T:Main} follows from the following result.
\begin{theorem}\label{T: main a+b=c}
Let $(X,\CX,\mu,T_n)$ be a pretentious multiplicative action, let
$F\in L^{\infty}(\mu)$ with $F\geq 0$,
and let $P_1,P_2,P_3$, $\gamma$ be as in
\eqref{quadratic_forms_a+b=c}, \eqref{gamma_defn}.
Then for every $\e>0$ we have
\begin{equation}\label{T: main a+b=c eq}
\int_X F \cdot T_{P_1(m,n)} F \cdot T_{P_2(m,n)} F \cdot T_{P_3(m,n)} F \, d\mu
> \Big(\int_X F \, d \mu \Big)^4 -\e
\end{equation}
for a set of $(m,n)\in \N^2$, $m>\gamma n$, with positive upper
density.
\end{theorem}
The polynomial $P_1$ is clearly irreducible. We split the proof of \cref{T: main a+b=c}
into cases according to whether
$P_2$, $P_3$ are irreducible or not.
Throughout \cref{Sec a+b=c}, $\Phi_{r,K}$ and $\Phi_{r,K,P}$ are
as in \eqref{eq_foln_def_1} and \eqref{eq_foln_def_2} (in some instances, $s,r$
take the place of $r,K$),
while $Q_s$ and $Q_{\delta,L}$ are as in \eqref{Q_l_def}
and \eqref{Q_delta_l_def}.
Furthermore,
$\omega_P$ is as in \eqref{omega_def}, and $\D$, $F_N(f,L)$, $G_{P,N}(f,L)$
are as in \eqref{E: definition of pretentious distance},
\eqref{def_F_N}, \eqref{def_G_P,N}.
Finally, $X_{p,\textup{fin.supp.}}$ and $X_{p,\textup{fin.supp.},P}$
are the factors defined respectively in
\eqref{def_X_p_fin_supp} and \eqref{def_X_p_fin_supp_P}.
\subsection{$P_2,P_3$ irreducible}\label{Sec P2 P3 irreducible}
Here we handle the case when $P_2,P_3$ are both
irreducible. A representative example is the equation $x^2+y^2=2z^2$, in which
case \eqref{quadratic_forms_a+b=c} gives
$$
P_1(m,n)=m^2+n^2, \quad P_2(m,n) = m^2-2mn-n^2, \quad
P_3(m,n)=m^2+2mn-n^2.
$$
To prove the result, given $Q_1 \in \Phi_{s,r,P_1}$,
$Q_2 \in \Phi_{r,K,P_2}$ and
$Q_3 \in \Phi_{K,L, P_3}$,
we will choose an integer $v$ depending on $s,r,K,L,Q_1,Q_2, Q_3$, and
we will average over the grid
$(Q_{\delta,L} m+1, Q_{\delta,L} n +v)$ in \eqref{T: main a+b=c eq}.
From now on, we assume that $s$ is large enough so that for $p>s$ we have
$\omega_{P_j}(p)\in \{0,2\}$ for $j\in\{1,2,3\}$ and
\begin{equation}\label{a+b=c_new_E1: p not dividing}
p\nmid 2ab(a+b)=2abc.
\end{equation}
\begin{lemma}\label{a+b=c_new_lemma_choice_of_v_1}
Let $\delta>0$, $s,r,K,L\in \N$ with $s<r<K<L/2$, $P_1,P_2,P_3$ be as in \eqref{quadratic_forms_a+b=c}, and let
$Q_1 \in \Phi_{s,r,P_1}$,
$Q_2 \in \Phi_{r,K,P_2}$ and $Q_3 \in \Phi_{K,L,P_3}$.
Then there is an integer
$v$ with $0\leq v \leq Q_{\delta, L}-1$ such that
\begin{enumerate}[(i)]
\item \label{a+b=c_new_E1: lemma_choice_of_v_1}
$(Q_{\delta, L},P_1(1,v))=Q_1$,
$\frac{P_1(1,v)}{Q_1} \equiv Q_1^{-1} \pmod{{Q_s}}$,
\item \label{a+b=c_new_E2: lemma_choice_of_v_1}
$(Q_{\delta, L},P_2(1,v))=Q_2$,
$\frac{P_2(1,v)}{Q_2} \equiv Q_2^{-1} \pmod{{Q_s}}$,
\item \label{a+b=c_new_E3: lemma_choice_of_v_1}
$(Q_{\delta, L},P_3(1,v))=Q_3$,
$\frac{P_3(1,v)}{Q_3} \equiv Q_3^{-1} \pmod{{Q_s}}$.
\end{enumerate}
\end{lemma}
\begin{proof}
Let $p\in \P$ with $s<p \leq r$ and $\omega_{P_1}(p)=2$, and consider the
polynomial $g_1(x):=P_1(1,x)-p^{\theta_p(Q_1)}$.
Then $\textup{mod}\: p$ the polynomial
becomes $g_1(x)=P_1(1,x)$, and it has two simple roots
$z_1, z_1'$ with $z_1 \not \equiv z_1' \pmod p$. Using
Hensel's lemma, we may lift $z_1$ to a unique root
$\zeta_1 = \zeta_1 (p, Q_1)$ of $g_1 \mod {p^{\theta_p(Q_1)+1}}$.
Analogously, for $p\in \P$ with $r<p \leq K$ and $\omega_{P_2}(p)=2$,
we may find $\zeta_2 = \zeta_2 (p, Q_2)$ so that if
$g_2(x):=P_2(1,x)-p^{\theta_p(Q_2)}$, then $g_2(\zeta_2)\equiv 0
\pmod {p^{\theta_p(Q_2)+1}}$, and for
$p\in \P$ with $K<p \leq L$ and $\omega_{P_3}(p)=2$,
we may find $\zeta_3 = \zeta_3 (p, Q_3)$ so that if
$g_3(x)=P_3(1,x)-p^{\theta_p(Q_3)}$, then $g_3(\zeta_3)\equiv 0
\pmod {p^{\theta_p(Q_3)+1}}$.
Using the Chinese Remainder theorem we may then choose $v$ with
$0\leq v \leq Q_{\delta,L}-1$
such that
\begin{itemize}
\item $v\equiv 0 \pmod {p^{2s}}$ for $p\leq s$,
\item $v\equiv 0$ $\pmod {p}$ for $s< p\leq r$ with
$\omega_{P_1}(p)=0$,
\item $v\equiv \zeta_1(p,Q_1) \pmod {p^{\theta_p(Q_1)+1}}$
for $s< p\leq r$ with $\omega_{P_1}(p)=2$,
\item $v\equiv 0$ $\pmod {p}$ for $r< p\leq K$ with
$\omega_{P_2}(p)=0$,
\item $v\equiv \zeta_2(p,Q_2) \pmod {p^{\theta_p(Q_2)+1}}$ for
$r<p\leq K$ with $\omega_{P_2}(p)=2$,
\item $v\equiv 0 \pmod {p}$ for
$K<p\leq L$ with $\omega_{P_3}(p)=0$,
\item $v\equiv \zeta_3(p,Q_3) \pmod {p^{\theta_p(Q_3)+1}}$ for
$K<p\leq L$ with $\omega_{P_3}(p)=2$.
\end{itemize}
Recall that $P_1(1,v)=1+abv^2$. For $p\leq s$, $P_1(1,v)\equiv 1 \pmod p$,
so $(p, P_1(1,v))=1$. For $s<p\leq r$ with $\omega_{P_1}(p)=0$,
$P_1(1,v)\equiv 1 \pmod p$, so $(p, P_1(1,v))=1$.
For $s<p\leq r$ with $\omega_{P_1}(p)=2$,
$P_1(1,v)\equiv P_1(1,\zeta_1)\equiv p^{\theta_p(Q_1)}
\pmod {p^{\theta_p(Q_1)+1}}$, so $p^{\theta_p(Q_1)}\parallel P_1(1,v)$.
For $r<p\leq K$ with $\omega_{P_2}(p)=0$,
$P_1(1,v)\equiv 1 \pmod p$, so $(p, P_1(1,v))=1$.
For $r<p\leq K$ with $\omega_{P_2}(p)=2$, suppose that $P_1(1,v)\equiv 0
\pmod p$.
We know that $P_2(1,v)\equiv 0 \pmod p$, so adding we get that
$p\mid P_1(1,v)+P_2(1,v)=2(1-bv)$. Since $p>r$ we have $p\neq 2$, and
we infer
that $p\mid (1-bv)$, which implies that $v\equiv b^{-1} \pmod p$. This
implies that $P_2(1,v)\equiv -1 -ab^{-1} \pmod p$, and since
$P_2(1,v) \equiv 0 \pmod p$, we obtain that $a+b \equiv 0 \pmod p$, which
contradicts \eqref{a+b=c_new_E1: p not dividing}. Therefore,
again we have $(p, P_1(1,v))=1$.
For $K<p\leq L$ with $\omega_{P_3}(p)=0$,
$P_1(1,v)\equiv 1 \pmod p$, so $(p, P_1(1,v))=1$.
For $K<p\leq L$ with $\omega_{P_3}(p)=2$, suppose that
$P_1(1,v)\equiv 0 \pmod p$.
We know that $P_3(1,v)\equiv 0 \pmod p$, so adding we get that
$p\mid P_1(1,v)+P_3(1,v)=2(1+av)$. Since $p>r$ we have $p\neq 2$, and
we infer
that $p\mid (1+av)$, which implies that $v\equiv -a^{-1} \pmod p$. This
implies that $P_3(1,v)\equiv -1 -a^{-1}b \pmod p$, and since
$P_3(1,v) \equiv 0 \pmod p$, we obtain that $a+b \equiv 0 \pmod p$, again
contradicting \eqref{a+b=c_new_E1: p not dividing}. Therefore,
we have $(p, P_1(1,v))=1$. Combining all of the above we get that
$(Q_{\delta, L}, P_1(1,v))=Q_1$. For $p\leq s$,
$P_1(1,v) \equiv 1 \pmod {p^{2s}}$, so $P_1(1,v) \equiv 1 \pmod {Q_s}$,
and since $(Q_s, Q_1)=1$, we have that
$\frac{P_1(1,v)}{Q_1} \equiv Q_1^{-1} \pmod{{Q_s}}$. This proves
\eqref{a+b=c_new_E1: lemma_choice_of_v_1}. The proofs of
\eqref{a+b=c_new_E2: lemma_choice_of_v_1},
\eqref{a+b=c_new_E3: lemma_choice_of_v_1} are identical, so
they are omitted.
\end{proof}
For the rest of \cref{Sec P2 P3 irreducible}, we suppress
the dependence of $v$ on the parameters
$s,r,K,L,Q_1,Q_2, Q_3$, and we simply write $v$
instead of $v(s,r,K,L,Q_1,Q_2, Q_3)$.
Given $\delta>0$, $s,r, K, L, N\in \N$ with
$s<r<K<L<N $, $Q_1 \in \Phi_{s,r,P_1}$, $Q_2\in \Phi_{r,K,P_2}$,
$Q_3\in \Phi_{K,L,P_3}$ and $m,n\in \N$, for $j\in \{1,2,3\}$, let
\begin{equation}\label{a+b=c_new_R_i def}
R_j (\delta,s, r, K, L, N, Q_1, Q_2, Q_3,m,n):=
\frac{P_j(Q_{\delta,L} m +1, Q_{\delta,L} n +v )}{Q_j},
\end{equation}
where $v$ is as in \cref{a+b=c_new_lemma_choice_of_v_1}.
Then for $j\in \{1,2,3\}$ we have
\begin{equation}\label{a+b=c_new_eq_extr_1}
P_j(Q_{\delta,L} m+1, Q_{\delta,L} n +v)= Q_j
R_j (\delta,s, r, K, L, N, Q_1, Q_2, Q_3,m,n).
\end{equation}
Sometimes in the notation we suppress the dependence of $R_j$ on
the parameters
$\delta, s, r$, $K, L, N, Q_1, Q_2, Q_3$, and we only write $R_j(m,n)$.
If $m>2\gamma n$, then since
$0\leq v \leq Q_{\delta,L}-1$, we have that $Qm+1 > \gamma (Qn+v)$,
so $P_j(Qm+1, Qn+v)>0$ for $j=2,3$, while $P_1(Qm+1, Qn+v)>0$ for all
$m,n \in \N$. For each $\delta\in(0,1/2)$, let
\begin{equation}\label{a+b=c_new_S_delta_a=c_P_2_reduc}
S_{\delta}:=\{(m,n)\in \N^2 : m> 2 \gamma n, \:
|(P_j(m,n))^i -1|\leq \delta \, \text{ for }
j=1,2,3\}.
\end{equation}
It follows from \cref{Cor S_d dense} that $S_{\delta}$ has
positive upper density.
%Using the exact same proof as in page $14$, one sees that
%when $P_2, P_3$ are irreducible,
\cref{T: main a+b=c} follows from
the next result.
\begin{proposition}\label{a+b=c_new_P: main a=c P_2 reducible}
Let $(X,\CX,\mu,T_n)$ be a pretentious multiplicative action, let
$F\in L^{\infty}(\mu)$ with $F\geq 0$,
let $P_1,P_2,P_3$ be as in
\eqref{quadratic_forms_a+b=c}, and suppose that $P_2,P_3$ are irreducible.
Also, let $S_{\delta}$ be as in \eqref{a+b=c_new_S_delta_a=c_P_2_reduc} and $v=v(\delta,s, r,K,L,Q_1,Q_2,Q_3)$ be as in
\cref{a+b=c_new_lemma_choice_of_v_1}.
Then
\begin{multline}\label{a+b=c_new_main_term_1}
\suplim_{\delta, s,r,K,L,N}\:
\cesE_{Q_{1}\in \Phi_{s,r,P_1}} \cesE_{Q_{2}\in \Phi_{r,K,P_2}}
\cesE_{Q_{3}\in \Phi_{K,L,P_3}}
\frac{1}{\overline{\textup{d}}(S_{\delta})} \cesE_{(m,n) \in [N]^2}
\1_{S_{\delta}}(m,n)\cdot \\
\int_{X} F\cdot \prod_{j=1} ^3
T_{P_j(Q_{\delta,L} m+1, Q_{\delta,L} n +v)} F
\, d \mu \geq \Big(\int_X F \, d \mu\Big)^4.
\end{multline}
\end{proposition}
It remains to prove \cref{a+b=c_new_P: main a=c P_2 reducible}.
For that we need the following lemma.
\begin{lemma}\label{a+b=c_new_L: main a=c P_2 reducible}
Let $(X, \X, \mu, T_n)$, $F$ and
$R_j$, $j\in \{1,2,3\}$, be as in \eqref{a+b=c_new_R_i def}.
Then for every $\delta>0$, all sufficiently large $s\in \N$, and all $r,K,\in \N$ with $s<r<K$, we have
\begin{align}
& \lim_{L, N} \,
\max_{Q_1 \in \Phi_{s,r,P_1}, \: Q_2, Q_{2}' \in \Phi_{r,K,P_2}, \:Q_3, Q_{3}' \in \Phi_{K,L,P_3} }
\cesE_{(m,n) \in [N]^2}
\label{a+b=c_new_L: main a=c P_2 reducible eq 1}\\
& \hspace{10mm} \lVert
T_{R_1(\delta,s, r, K, L, N, Q_1, Q_2, Q_3,m,n)} F -
T_{R_1(\delta,s, r, K, L, N, Q_1, Q_2 ', Q_3 ',m,n)} F
\rVert _{L^2(\mu)}=0, \nonumber \\
& \lim_{L, N} \,
\max_{Q_1, Q_1 ' \in \Phi_{s,r,P_1}, \: Q_2 \in \Phi_{r,K,P_2}, \:Q_3, Q_{3}' \in \Phi_{K,L,P_3} }
\cesE_{(m,n) \in [N]^2}
\label{a+b=c_new_L: main a=c P_2 reducible eq 2} \\
& \hspace{10mm} \lVert
T_{R_2(\delta,s, r, K, L, N, Q_1, Q_2, Q_3,m,n)} F -
T_{R_2(\delta,s, r, K, L, N, Q_1 ', Q_2, Q_3 ',m,n)} F
\rVert _{L^2(\mu)}=0, \nonumber \\
& \lim_{L, N}\,
\max_{Q_1, Q_1 ' \in \Phi_{s,r,P_1}, \: Q_2, Q_2 ' \in \Phi_{r,K,P_2},
\:Q_3
\in \Phi_{K,L, P_3} }
\cesE_{(m,n) \in [N]^2}
\label{a+b=c_new_L: main a=c P_2 reducible eq 3} \\
& \hspace{10mm} \lVert
T_{R_3(\delta,s, r, K, L, N, Q_1, Q_2, Q_3,m,n)} F -
T_{R_3(\delta,s, r, K, L, N, Q_1 ', Q_2 ', Q_3,m,n)} F
\rVert _{L^2(\mu)}=0. \nonumber
\end{align}
\end{lemma}
\begin{remark}
Similarly to \cref{L: main a=c P_2 reducible},
\cref{a+b=c_new_L: main a=c P_2 reducible}
essentially says that
in the integral in \eqref{a+b=c_new_main_term_1},
the first term is independent of $Q_2, Q_3$, the second term
is independent of $Q_1, Q_3$ and the third term
is independent of $Q_1, Q_2$.
\end{remark}
\begin{proof}
We omit the proof, as it closely follows the proof of \cref{L: main
a=c P_2 reducible}. The essential step is to apply the
concentration estimates \cref{linear_conc} and \cref{quad_conc}, with
the only change being the use of \cref{a+b=c_new_lemma_choice_of_v_1}
in place of \cref{lemma_choice_of_v_1}.
%We can assume that
%$\lVert F \rVert_{L^{\infty}(\mu)}\leq 1$.
%In the proof, it is always implicit that $s<r<K<L<N$, and $L$ is
%sufficiently large depending on $s,r,K$.
%We start by proving \eqref{a+b=c_new_L: main a=c P_2 reducible eq 1}.
%Note that from \eqref{spec_eq_2} we have
%\begin{multline*}
%\lVert
%T_{R_1(\delta,s, r, K, L, N, Q_1, Q_2, Q_3,m,n)} F -
%T_{R_1(\delta,s, r, K, L, N, Q_1, Q_2 ', Q_3 ',m,n)} F
%\rVert _{L^2(\mu)}= \\
%\lVert
%f(R_1(\delta,s, r, K, L, N, Q_1, Q_2, Q_3,m,n)) -
%f(R_1(\delta,s, r, K, L, N, Q_1, Q_2 ', Q_3 ',m,n))
%\rVert _{L^2(\sigma_{F}(f))},
%\end{multline*}
%so after applying the Cauchy-Schwarz inequality and Fatou's lemma
%multiple times, one
%sees that in order to establish \eqref{L: main a=c P_2 reducible eq 1}, it
%suffices to prove that for each $f\in \M_p $
%\begin{multline}\label{a+b=c_new_eq_conc_1}
%\lim_{\delta, s,r, K, L, N}
%\max_{Q_1 \in \Phi_{s,r,P_1}, \: Q_2, Q_{2}' \in \Phi_{r,K,P_2},
%\:Q_3, Q_{3}' \in \Phi_{K,L,P_3} }
%\frac{1}{\overline{\textup{d}}(S_{\delta})} \cesE_{(m,n) \in [N]^2}
% \1_{S_{\delta}}(m,n)\cdot \\
%|f(R_1(\delta,s, r, K, L, N, Q_1, Q_2, Q_3,m,n))
%\overline{f(R_1(\delta,s, r, K, L, N, Q_1, Q_2 ', Q_3 ',m,n))}-1 |
%=0.
%\end{multline}
%Let $f\in \M_p$ and let $\chi,t$ so that $f\sim \chi\cdot n^{it}$. Let
%$q_{\chi}$ be the conductor of $\chi$, and suppose that
%$s,L$ are large enough so that $q_{\chi}\mid Q_s$ and
%$q_{\chi}\mid \frac{Q_{\delta, L}}{Q_1}$ for every $Q_1 \in
%\Phi_{s,r,P_1}$. Using
%\cref{a+b=c_new_lemma_choice_of_v_1}
%\eqref{a+b=c_new_E1: lemma_choice_of_v_1} and
%applying \cref{quad_conc} and \cref{uniform_in_Q} we get that
%\begin{align*}
%&\limsup_{N\to \infty} \max_{Q_1 \in \Phi_{s,rP_1}, \; Q_2, Q_{2}' \in \Phi_{r,K,P_2}, %\:Q_3, Q_{3}' \in \Phi_{K,L,P_3}}
%\cesE_{(m,n) \in [N]^2} \1_{S_{\delta}}(m,n) \cdot \\
%&|f(R_1(\delta, r, K, L, N, Q_1, Q_2, Q_3,m,n)) \cdot
% \overline{f(R_1(\delta, r, K, L, N, Q_1, Q_2 ', Q_3 ',m,n))}-1 | \\
%&\ll_{P_1} \limsup_{N\to \infty}
%|\exp(2\Re (G_{P_1,N}(f,L))) -1| +
%\D(f, \chi \cdot n^{it}; L, \infty)
%+L^{-1/2}.
%\end{align*}
%Since $f\sim \chi \cdot n^{it}$, we have that $2\Re(G_{P_1,N}(f,L))$ converges to
%$$2\sum_{p>L}\frac{ \omega_{P_1}(p)}{p} \big({\Re(f(p) \overline{\chi(p)}p^{-it}) %-1}\big).$$ Then
%\eqref{a+b=c_new_eq_conc_1} follows by
%taking $L\to \infty$ in the last relation. This concludes the proof
%of \eqref{a+b=c_new_L: main a=c P_2 reducible eq 1}.
%The proofs of \eqref{a+b=c_new_L: main a=c P_2 reducible eq 2},
%\eqref{a+b=c_new_L: main a=c P_2 reducible eq 3} are
%identical, so they are omitted.
\end{proof}
We are now ready to prove \cref{a+b=c_new_P: main a=c P_2 reducible}.
\begin{proof}[Proof of \cref{a+b=c_new_P: main a=c P_2 reducible}]
The argument is similar to that in \cref{P: main a=c P_2 reducible},
but there are some non-trivial differences. For completeness, we present it
here with some details omitted.
We can assume that
$\lVert F\rVert_{L^{\infty}(\mu)}\leq 1$. In the proof, it is always
implicit that $s<r<K<L<N$, and that each of the previous variables is
sufficiently large depending on the smaller ones.
In view of \eqref{a+b=c_new_eq_extr_1}, the quantity in
$\suplim_{\delta, s,r,K,L,N}$ on the left-hand side in
\eqref{a+b=c_new_main_term_1} equals
\begin{multline*}
\cesE_{Q_{1}\in \Phi_{s,r,P_1}} \cesE_{Q_{2}\in \Phi_{r,K,P_2}}
\cesE_{Q_{3}\in \Phi_{K,L,P_3}} \frac{1}{\overline{\textup{d}}(S_{\delta})} \cesE_{(m,n) \in [N]^2}
\1_{S_{\delta}}(m,n) \cdot \\
\int_{X} F \cdot T_{Q_1 R_1(m,n)} F \cdot
T_{Q_2 R_2(m,n)} F \cdot
T_{{Q}_3 R_3(m,n)} F
\, d \mu.
\end{multline*}
For each $s,r,K,L\in \N$, choose any $Q_{1,s,r}\in \Phi_{s,r,P_1}$,
$Q_{2, r, K} \in \Phi_{r,K,P_2}$ and $Q_{3,K,L}\in \Phi_{K,L,P_3}$.
Using \cref{a+b=c_new_L: main a=c P_2 reducible}, one sees that in order to prove
\eqref{a+b=c_new_main_term_1}, it suffices to prove that
\begin{multline}\label{a+b=c_new_main_term_3}
\suplim_{\delta,s,r, K, L,N}
\int_{X} F \cdot \frac{1}{\overline{\textup{d}}(S_{\delta})}
\cesE_{(m,n) \in [N]^2}
\1_{S_{\delta}}(m,n) \cdot
\Big( \cesE_{Q_{1}\in \Phi_{s,r,P_1}} \\ T_{Q_1 R_1(\delta,s, r, K, L, N, Q_1, Q_{2,r,K} , Q_{3,K,L} ,m,n)} F \Big)
\cdot \Big( \cesE_{Q_{2}\in \Phi_{r,K,P_2}}
T_{Q_2 R_2(\delta,s, r, K, L, N, Q_{1,s,r},
Q_2 , Q_{3,K,L},m,n)} F \Big) \cdot \\ \Big( \cesE_{Q_{3}\in \Phi_{K,L,P_3}}
T_{{Q}_3 R_3(\delta,s, r, K, L, N, Q_{1,s,r}, Q_{2,r,K} , Q_3 ,m,n)} F \Big)
d \mu \geq \Big(\int_X F \, d \mu\Big)^4 .
\end{multline}
As in the proof of \cref{P: main a=c P_2 reducible} we handle each of the three averages separately.
\smallskip
{\bf Step 1 (Dealing with the average over $Q_1$).}
Write $F=F_1 + F_2$, where $F_1:=\E(F | {\X}_{p, \textup{fin.supp.},P_1})$
and $F_2 \perp F_1$.
Then the main term in \eqref{a+b=c_new_main_term_3} is equal to
\begin{multline}\label{a+b=c_new_main_term_4}
\sum_{j\in \{1,2\}}\int_{X} F \cdot \frac{1}{\overline{\textup{d}}(S_{\delta})}
\cesE_{(m,n) \in [N]^2}
\1_{S_{\delta}}(m,n) \cdot
\Big( \cesE_{Q_{1}\in \Phi_{s,r,P_1}} \\ T_{Q_1 R_1(\delta,s, r, K, L, N, Q_1, Q_{2,r,K} , Q_{3,K,L} ,m,n)} F_j \Big)
\cdot \Big( \cesE_{Q_{2}\in \Phi_{r,K,P_2}}
T_{Q_2 R_2(\delta,s, r, K, L, N, Q_{1,s,r},
Q_2 , Q_{3,K,L},m,n)} F \Big) \cdot \\ \Big( \cesE_{Q_{3}\in \Phi_{K,L,P_3}}
T_{{Q}_3 R_3(\delta,s, r, K, L, N, Q_{1,s,r}, Q_{2,r,K} , Q_3 ,m,n)} F \Big)
d \mu.
\end{multline}
The term corresponding to $j=2$ in \eqref{a+b=c_new_main_term_4}
is bounded by
\begin{equation}\label{a+b=c_new_aux_quant_1}
\frac{1}{\overline{\textup{d}}(S_{\delta})}
\cesE_{(m,n) \in [N]^2} \big \lVert
\cesE_{Q_{1}\in \Phi_{s,r,P_1}}
T_{Q_1 R_1(\delta,s, r, K, L, N, Q_1, Q_{2,r,K} , Q_{3,K,L} ,m,n)} F_2
\big \rVert_{L^2(\mu)}.
\end{equation}
We will show that the quantity in \eqref{a+b=c_new_aux_quant_1} goes to
$0$ as $N\to \infty, L \to \infty, K \to \infty, r \to \infty,
s \to \infty$.
Since $\sigma_{F_2}$ is supported on
$\M_p \setminus \M_{p, \textup{fin.supp.},P_1}$, using \eqref{spec_eq_1} and
applying Fatou's lemma multiple times, one sees
that it suffices to prove that for every
$f\in \M_p \setminus \M_{p, \textup{fin. supp.},P_1}$, we have
\begin{equation*}
\lim _{s,r,K,L,N} \:
\cesE_{(m,n) \in [N]^2} \big |
\cesE_{Q_{1}\in \Phi_{s,r,P_1}} f({Q_1
R_1(\delta,s, r, K, L, N, Q_1, Q_{2,r,K} , Q_{3,K,L} ,m,n)}) \big |=0.
\end{equation*}
Let $f\in \M_p \setminus \M_{p, \textup{fin.supp.},P_1}$,
let $\chi, t$ so that $f\sim \chi \cdot n^{it}$,
and suppose that $s,L$ are large enough so that the conductor $q_{\chi}$ of
$\chi$ divides $Q_s$ and $\frac{Q_{\delta, L}}{Q_1}$ for all
$Q_1 \in \Phi_{s,r,P_1}$.
Using \cref{a+b=c_new_lemma_choice_of_v_1} \eqref{a+b=c_new_E1: lemma_choice_of_v_1} and
applying \cref{quad_conc} and \cref{uniform_in_Q} we deduce that
\begin{align}\label{a+b=c_new_eq_useful_1}
&\limsup_{N\to \infty} \cesE_{(m,n) \in [N]^2} \big |
\cesE_{Q_{1}\in \Phi_{s,r,P_1}} f({Q_1
R_1(\delta,s, r, K, L, N, Q_1, Q_{2,r,K} , Q_{3,K,L} ,m,n)})
\big | \ll_{P_1} \\
&\limsup_{N\to \infty} \cesE_{(m,n) \in [N]^2} \big |
\cesE_{Q_{1}\in \Phi_{s,r,P_1}} f(Q_1)\, \overline{\chi(Q_1)} \,
Q_1 ^{-it} \, Q_{\delta,L}^{2it} \, (P_{1}(m,n))^{it} \,
\exp(G_{P_1,N}(f,L)) \big | \nonumber \\
& \hspace{20mm} + \D(f, \chi \cdot n^{it}; L, \infty)
+L^{-1/2}
= \nonumber \\
& \limsup_{N\to \infty} |\exp(G_{P_1,N}(f,L))|\cdot
\big| \cesE_{Q_1 \in \Phi_{s,r,P_1}}
f(Q_1) \, \overline{\chi(Q_1)} \, Q_1 ^{-it}
\big|+ \D(f, \chi \cdot n^{it}; L, \infty)
+L^{-1/2}. \nonumber
\end{align}
From $f\sim \chi \cdot n^{it}$ we deduce that
$\lim_{L\to \infty} \lim_{N\to \infty} |\exp(G_{P_1,N}(f,L))|=1$.
Since $f \in \M_p \setminus \M_{p, \textup{fin.supp.},P_1}$, for each
$s$ there is
$p>s$ such that $\omega_{P_1}(p)>0$ and $f(p) \neq \chi(p) p^{it}$. Using
that $\Phi_{s,r,P_1}$ is asymptotically
invariant under dilation by $p$ as $r\to \infty$ we have
\begin{multline*}
{f(p)}\, \overline{\chi(p)}\, p^{-it}
\cesE_{Q_{1} \in \Phi_{s,r,P_1}} f(Q_{1})\,
\overline{\chi(Q_{1})}\, Q_{1} ^{-it}
=
\cesE_{Q_{1} \in p \Phi_{s,r,P_1}} f(Q_{1})\, \overline{\chi(Q_{1})}\,
Q_{1} ^{-it}
\\ =
\cesE_{Q_{1} \in \Phi_{s,r,P_1}} f(Q_{1}) \, \overline{\chi(Q_{1})}\,
Q_{1} ^{-it}
+ \oh_{r\to \infty}(1).
\end{multline*}
Using $f(p) \neq \chi(p)\, p^{it}$, the above implies that
\begin{equation*}
\lim_{r\to \infty} \cesE_{Q_{1} \in \Phi_{s,r,P_1}} f(Q_{1})
\cdot \overline{\chi(Q_{1})} \cdot Q_{1} ^{-it} = 0.
\end{equation*}
Combining this with \eqref{a+b=c_new_eq_useful_1} we infer that
\begin{equation}\label{a+b=c_new_aux_eq_3}
\lim_{s,r,K,L,N}
\cesE_{(m,n) \in [N]^2} \big \lVert
\cesE_{Q_{1}\in \Phi_{s,r,P_1}}
T_{Q_1 R_1(\delta,s, r, K, L, N, Q_1, Q_{2,r,K} , Q_{3,K,L} ,m,n)} F_2
\big \rVert_{L^2(\mu)}=0.
\end{equation}
On the other hand, for the term corresponding to $j=1$ in
\eqref{a+b=c_new_main_term_4}, we claim that
\begin{multline}\label{a+b=c_new_eq_useful_2}
\lim_{\delta,s, r, K, L, N}
\max_{Q_1 \in \Phi_{s,r,P_1}}
\frac{1}{\overline{\textup{d}}(S_{\delta})}\cdot
\cesE_{(m,n) \in [N]^2} \1_{S_{\delta}}(m,n)\cdot \\
\lVert
T_{Q_1 R_1(\delta,s, r, K, L, N, Q_1, Q_{2,r,K} , Q_{3,K,L} ,m,n)}
F_1 -F_1 \rVert_{L^2(\mu)} =0.
\end{multline}
To see this, let $f\in M_{p, \textup{fin.supp.},P_1}$, and
let $\chi, t$ and $s_0 \in \N$ so that $f\sim \chi \cdot n^{it}$ and
$f(p)=\chi(p)\, p^{it}$ for all primes $p>s_0$ with $\omega_{P_1}(p)>0$.
Also, suppose that $s,L$ are sufficiently
large so that $s>s_0$ and the conductor $q_{\chi}$ of $\chi$
divides $Q_s$ and $\frac{Q_{\delta, L}}{Q_1}$ for every
$Q_{1} \in \Phi_{s,r,P_1}$.
Using \cref{a+b=c_new_lemma_choice_of_v_1} \eqref{a+b=c_new_E1: lemma_choice_of_v_1} and
applying \cref{quad_conc} and \cref{uniform_in_Q}, we get
\begin{align}\label{last day in the office eq 2}
&\limsup_{N\to \infty} \max_{Q_1 \in \Phi_{s,r,P_1}}
\frac{1}{\overline{\textup{d}}(S_{\delta})}
\cesE_{(m,n) \in [N]^2} \1_{S_{\delta}}(m,n) \cdot \\
& \hspace{20mm}\big|f(Q_1)
f(R_1(\delta,s, r, K, L, N, Q_{1}, Q_{2,r,K} , Q_{3,K,L} ,m,n)) -1\big|
\ll_{P_1} \nonumber \\
& \limsup_{N\to \infty} \max_{Q_1 \in \Phi_{s,r,P_1}}
\frac{1}{\overline{\textup{d}}(S_{\delta})}
\cesE_{(m,n) \in [N]^2} \1_{S_{\delta}}(m,n) \cdot
|f(Q_{1})\, \overline{\chi(Q_{1})} \, Q_{1}^{-it} \cdot \nonumber \\
& \hspace{20mm} Q_{\delta, L}^{2it} \, (P_1(m,n))^{it} \,
\exp(G_{P_1,N}(f,L)) -1| + \D(f, \chi \cdot n^{it}, L, \infty) +
L^{-1/2}. \nonumber
\end{align}
For $p>s_0$ with $\omega_{P_1}(p)>0$, we have
$f(p)=\chi(p)\, p^{it}$, which implies that
$f(Q_{1})= \chi(Q_{1})\, Q_{1}^{it}$, and since $L>s_0$, we have
$G_{P_1,N}(f,L)=0$. Therefore, the last quantity in
\eqref{last day in the office eq 2}
is equal to
\begin{equation*}
\limsup_{N\to \infty}
\frac{1}{\overline{\textup{d}}(S_{\delta})}
\cesE_{(m,n) \in [N]^2} \1_{S_{\delta}}(m,n) \cdot
| Q_{\delta, L}^{2it} P_1(m,n)^{it}
-1| + \oh_{L\to \infty}(1) \ll_{t} \delta + \oh_{L\to \infty}(1),
\end{equation*}
where for the last estimate we used
\eqref{Q delta L def} and \eqref{a+b=c_new_S_delta_a=c_P_2_reduc}.
We infer that
for $f\in M_{p, \textup{fin.supp.},P_1}$,
\begin{multline*}
\lim_{\delta,s, r, K, L, N}
\max_{Q_1 \in \Phi_{s,r,P_1}}
\frac{1}{\overline{\textup{d}}(S_{\delta})}
\cesE_{(m,n) \in [N]^2} \1_{S_{\delta}}(m,n) \cdot \\
\big|f(Q_1)
f(R_1(\delta, s, r, K, L, N, Q_{1}, Q_{2,r,K} , Q_{3,K,L} ,m,n))
-1\big|=0,
\end{multline*}
and since the spectral measure of $F_1$ is supported on
$M_{p, \textup{fin.supp.},P_1}$, \eqref{a+b=c_new_eq_useful_2} follows.
Using \eqref{a+b=c_new_main_term_4}, \eqref{a+b=c_new_aux_eq_3}, and \eqref{a+b=c_new_eq_useful_2}, we see
that in order to prove \eqref{a+b=c_new_main_term_3},
it suffices to prove that
\begin{multline}\label{a+b=c_new_main_term_5}
\suplim_{\delta,s,r, K, L,N}
\int_{X} F \cdot F_1 \cdot \frac{1}{\overline{\textup{d}}(S_{\delta})}
\cesE_{(m,n) \in [N]^2}
\1_{S_{\delta}}(m,n) \cdot
\\ \Big( \cesE_{Q_{2}\in \Phi_{r,K,P_2}}
T_{Q_2 R_2(\delta,s, r, K, L, N, Q_{1,s,r},
Q_2 , Q_{3,K,L},m,n)} F \Big) \cdot \\ \Big( \cesE_{Q_{3}\in \Phi_{K,L,P_3}}
T_{{Q}_3 R_3(\delta,s, r, K, L, N, Q_{1,s,r}, Q_{2,r,K} , Q_3 ,m,n)} F \Big)
d \mu \geq \Big(\int_X F \, d \mu\Big)^4.
\end{multline}
{\bf Steps 2 and 3 (Dealing with the averages over $Q_2$ and $Q_3$).}
Writing $F=F_3 + F_4$, where $F_3:=\E(F | \X_{p, \textup{fin.supp.},P_2})$
and $F_4 \perp F_3$, and utilizing a similar argument we get
\begin{equation}\label{a+b=c_new_aux_eq_4}
\lim_{s,r,K,L,N}
\cesE_{(m,n) \in [N]^2} \big \lVert
\cesE_{Q_{2}\in \Phi_{r,K,P_2}}
T_{Q_2 R_2(\delta,s, r, K, L, N, Q_{1,s,r}, Q_{2} , Q_{3,K,L} ,m,n)} F_4
\big \rVert_{L^2(\mu)}=0
\end{equation}
and
\begin{multline}\label{a+b=c_new_eq_useful_4}
\lim_{\delta,s, r, K, L, N}
\max_{Q_2 \in \Phi_{r,K,P_2}}
\frac{1}{\overline{\textup{d}}(S_{\delta})}\cdot
\cesE_{(m,n) \in [N]^2} \1_{S_{\delta}}(m,n)\cdot\\
\lVert
T_{Q_2 R_2(\delta,s, r, K, L, N, Q_{1,s,r}, Q_{2} , Q_{3,K,L} ,m,n)} F_3
-F_3 \rVert_{L^2(\mu)} =0.
\end{multline}
Similarly, if we write $F=F_5 + F_6$, where
$F_5:=\E(F | \X_{p, \textup{fin.supp.}, P_3})$
and $F_6 \perp F_5$, then
\begin{equation}\label{a+b=c_new_aux_eq_5}
\lim_{s,r,K,L,N}
\cesE_{(m,n) \in [N]^2} \big \lVert
\cesE_{Q_{3}\in \Phi_{K,L,P_3}}
T_{Q_3 R_3(\delta,s, r, K, L, N, Q_{1,s,r}, Q_{2,r,K} , Q_{3} ,m,n)} F_6
\big \rVert_{L^2(\mu)}=0
\end{equation}
for every $\delta>0$, and
\begin{multline}\label{a+b=c_new_eq_useful_8}
\lim_{\delta,s, r, K, L, N}
\max_{Q_3 \in \Phi_{K,L,P_3}}
\frac{1}{\overline{\textup{d}}(S_{\delta})}
\cesE_{(m,n) \in [N]^2} \1_{S_{\delta}}(m,n)\cdot \\
\lVert
T_{Q_3 R_3(\delta,s, r, K, L, N, Q_{1,s,r}, Q_{2,r,K} , Q_{3} ,m,n)} F_5
-F_5 \rVert_{L^2(\mu)} =0.
\end{multline}
Using \eqref{a+b=c_new_aux_eq_4}-\eqref{a+b=c_new_eq_useful_8} and arguing as in Step 1, we see
that in order to prove \eqref{a+b=c_new_main_term_5},
it suffices to prove that
\begin{equation}\label{a+b=c_new_main_term_10}
\suplim_{\delta,s,r, K, L,N}
\int_{X} F \cdot F_1 \cdot F_3 \cdot F_5 \, d\mu \cdot \frac{1}{\overline{\textup{d}}(S_{\delta})}
\cesE_{(m,n) \in [N]^2}
\1_{S_{\delta}}(m,n)
\geq \Big(\int_X F \, d \mu\Big)^4 .
\end{equation}
{\bf Step 4 (Finishing the argument).}
Note that
$$
\int_X F\cdot F_1 \cdot F_3 \cdot F_5 \, d \mu =
\int_X F\cdot \E(F | {\X}_{p, \textup{fin.supp.},P_1}) \cdot
\E(F | \X_{p, \textup{fin.supp.},P_2})
\cdot \E(F | \X_{p, \textup{fin.supp.},P_3}) \, d \mu,
$$
and \cref{chu-lemma} implies that the last quantity is greater or equal to $(\int F\, d\mu)^4$. This establishes \eqref{a+b=c_new_main_term_10} and finishes the proof of the proposition.
\end{proof}
\subsection{$P_2$ reducible, $P_3$ irreducible or $P_2$ irreducible, $P_3$ reducible}\label{Sec : a+b=c_P_2_reducible P_3 irreducible}
Those cases are symmetric, so we may assume $P_2$ is reducible and
$P_3$ is irreducible. A representative example is $3x^2+y^2=4z^2$, in
which case \eqref{quadratic_forms_a+b=c} gives
$$
P_1(m,n)=m^2+3n^2, \quad P_2(m,n)=(m-3n)(m+n), \quad P_3(m,n)=m^2+6mn-3n^2.
$$
Since the proof is very similar to when both $P_2$ and $P_3$ are
irreducible, we only highlight the changes needed in the proof from
\cref{Sec P2 P3 irreducible} to obtain the result in this case.nted in \cref{Sec P2 P3 irreducible} to obtain the result in this
case.
Since $P_2$ is reducible, it is easy to see that there are
$\lambda_1, \lambda_2 \in \Z\setminus \{0\}$ such that
$\lambda_1+ \lambda_2 =-2b$, $\lambda_1\lambda_2 =-ab$, and
\begin{equation}\label{E:P2reducible}
P_2(m,n)=(m+\lambda_1 n)(m+\lambda_2 n).
\end{equation}
Since $\lambda_1 \lambda_2<0$,
we have $\lambda_1 \neq \lambda_2$.
From now on we assume that $s$ is large enough so that
for $p>s$ we have $\omega_{P_1}(p), \omega_{P_3}(p) \in \{0,2\}$, and
\begin{equation}\label{a+b=c_new_E2: p not dividing}
p\nmid 2 ab(a+b)\lambda_1 \lambda_2 (\lambda_1 - \lambda_2)
(ab + \lambda_1 ^2).
\end{equation}
Given $Q_1 \in \Phi_{s,r,P_1}$, $Q_2 \in \Phi_{r,K}$ and
$Q_3 \in \Phi_{K,L, P_3}$,
we will choose an integer $v$ depending on $r,K,L,Q_1,Q_2, Q_3$, and
we will average over the grid
$(Q_{\delta,L} m+1, Q_{\delta,L} n +v)$ in \eqref{T: main a+b=c eq}.
\begin{lemma}\label{a+b=c_new_lemma_choice_of_v_1_a=c_P_3_irred}
Let $\delta>0$, $s,r,K,L\in \N$ with $s<r<K<L/2$,
$P_1,P_2,P_3$ be as in
\eqref{quadratic_forms_a+b=c}, with $P_2$ as in \eqref{E:P2reducible}, and let $Q_1 \in
\Phi_{s,r,P_1}$, $Q_2 \in \Phi_{r,K}$ and $Q_3 \in \Phi_{K,L,P_3}$.
Then there is an integer
$v$ with $0\leq v \leq Q_{\delta, L}-1$ such that
\begin{enumerate}[(i)]
\item $(Q_{\delta, L}, P_1(1,v))=Q_1$,
$\frac{P_1(1,v)}{Q_1}\equiv Q_1 ^{-1} \pmod {Q_s}$,
\item $(Q_{\delta, L},1+\lambda_1 v)= Q_2$,
$\frac{1+\lambda_1 v}{Q_2} \equiv Q_2 ^{-1}
\pmod{{Q_s}}$,
\item $(Q_{\delta, L},1+\lambda_2 v)= 1$,
$1+\lambda_2 v \equiv 1 \pmod{{Q_s}}$,
\item $(Q_{\delta, L},P_3(1,v))=Q_3$,
$\frac{P_3(1,v)}{Q_3} \equiv Q_3^{-1} \pmod{{Q_s}}$.
\end{enumerate}
\end{lemma}
\begin{proof}
Let $p\in \P$ with $s<p \leq r$ and $\omega_{P_2}(p)=2$, and consider the
polynomial $g_1(x):=P_1(1,x)-p^{\theta_p(Q_1)}$.
Then $\textup{mod}\: p$ the polynomial
becomes $g_1(x)=P_1(1,x)$, and it has two simple roots
$z_1, z_1'$ with $z_1 \not \equiv z_1' \pmod p$. Using
Hensel's lemma, we may lift $z_1$ to a unique root
$\zeta_1 = \zeta_1 (p, Q_1)$ of $g_1 \mod {p^{\theta_p(Q_1)+1}}$.
Analogously, for $p\in \P$ with $K<p \leq L$ and $\omega_{P_3}(p)=2$,
we may find $\zeta_3 = \zeta_3 (p, Q_3)$ so that if
$g_3(x)=P_3(1,x)-p^{\theta_p(Q_3)}$, then $g_3(\zeta_3)\equiv 0
\pmod {p^{\theta_p(Q_3)+1}}$.
Using the Chinese Remainder theorem we may then choose $v$ with
$0\leq v \leq Q_{\delta,L}-1$
such that
\begin{itemize}
\item $v\equiv 0 \pmod {p^{2s}}$ for $p\leq s$,
\item $v\equiv 0 \pmod {p}$ for $s<p \leq r$ with $\omega_{P_1}(p)=0$,
\item $v\equiv \zeta_1(p, Q_1)$,
$\pmod {p^{\theta_p(Q_1)+1}}$ for $s< p\leq r$ with
$\omega_{P_1}(p)=2$,
\item $v\equiv \lambda_1 ^{-1} (p^{\theta_p(Q_2)} -1 )\pmod{p^{\theta_p(Q_2)+1}}$ for $r<p\leq K$,
\item $v\equiv 0 \pmod {p} $ for $K<p\leq L$ with $\omega_{P_3}(p)=0$,
\item $v\equiv \zeta_3(p,Q_3) \pmod {p^{\theta_p(Q_3)+1}}$ for
$K<p\leq L$ with $\omega_{P_3}(p)=2$.
\end{itemize}
The rest of the proof is very similar to the proof of
\cref{a+b=c_new_lemma_choice_of_v_1}, so it is omitted.
\end{proof}
As in \cref{Sec P2 P3 irreducible}, for each $\delta\in(0,1/2)$, let
\begin{equation}\label{a+b=c_new_S_delta_a=c_P_3_irreduc}
S_{\delta}:=\{(m,n)\in \N^2 : m> 2\gamma n, \:
|(P_j(m,n))^i -1|\leq \delta\, \text{ for }
j=1,2,3\},
\end{equation}
and recall that in view of \cref{Cor S_d dense}, $S_{\delta}$ has
positive upper density. Similarly to \cref{Sec P2 P3 irreducible},
for $P_2$ reducible and $P_3$ irreducible,
\cref{T: main a+b=c} follows from the following
result.
\begin{proposition}\label{a+b=c_new_P: main a=c P_3 irreducible}
Let $(X,\CX,\mu,T_n)$ be a pretentious multiplicative action, let
$F\in L^{\infty}(\mu)$ with $F\geq 0$,
let $P_1,P_2,P_3$ be as in
\eqref{quadratic_forms_a+b=c}, and suppose that
$P_2$ is reducible and $P_3$ is irreducible.
Also, let $S_{\delta}$ be as in \eqref{a+b=c_new_S_delta_a=c_P_3_irreduc} and $v=v(\delta,s, r, K,L,Q_1,Q_2,Q_3)$ be as in
\cref{a+b=c_new_lemma_choice_of_v_1_a=c_P_3_irred}.
Then
\begin{multline}\label{a+b=c_new_E: P: main a=c P_3 irreducible}
\suplim_{\delta,s, r,K,L,N}\:
\cesE_{Q_{1}\in \Phi_{s,r,P_1}} \cesE_{Q_{2}\in \Phi_{r,K}}
\cesE_{Q_{3}\in \Phi_{K,L,P_3}}
\frac{1}{\overline{\textup{d}}(S_{\delta})} \cesE_{(m,n) \in [N]^2}
\1_{S_{\delta}}(m,n)\cdot \\
\int_{X} F\cdot \prod_{j=1} ^3
T_{P_j(Q_{\delta,L} m+1, Q_{\delta,L} n +v)} F
\, d \mu \geq \Big(\int_X F \, d \mu\Big)^4.
\end{multline}
\end{proposition}
To prove \cref{a+b=c_new_P: main a=c P_3 irreducible}, one must first
establish an analog of \cref{a+b=c_new_L: main a=c P_2 reducible},
which states that in the integral \eqref{a+b=c_new_E: P: main a=c P_3
irreducible}, the first term is independent of $Q_2, Q_3$, the second
term is independent of $Q_1, Q_3$, and the third term is independent
of $Q_1, Q_2$. This result is then used to prove \cref{a+b=c_new_P:
main a=c P_3 irreducible}. Both proofs are very similar to those in
\cref{Sec P2 P3 irreducible}, with the only essential change being
that, to handle the term $T_{P_2(Q_{\delta,L} m+1, Q_{\delta,L} n+v)}
F$, one must use \cref{linear_conc} and \cref{bilinear_lemma} instead
of \cref{quad_conc}.
\subsection{$P_2$, $P_3$ reducible}\label{Sec : a+b=c P_2 P_3 reducible}
Finally, we handle the case when both $P_2$ and $P_3$ are reducible.
A representative example is $9x^2+16y^2=25z^2$, in
which case \eqref{quadratic_forms_a+b=c} gives
$$
P_1(m,n)=m^2+144n^2, \quad P_2(m,n)=(m-36n)(m+4n), \quad P_3(m,n)=(m-6n)(m+24n).
$$
Again, since the proof is very similar to the case where both $P_2$
and $P_3$ are irreducible, we only note the modifications needed in
the proof from \cref{Sec P2 P3 irreducible} to obtain the result in
this case.
Using that $P_2$, $P_3$ are reducible, we see that there are
$\lambda_1, \lambda_2, \lambda_3, \lambda_4 \in \Z\setminus \{0\}$
such that
$\lambda_1+ \lambda_2 =-2b$, $\lambda_1\lambda_2 =-ab$,
$\lambda_3+ \lambda_4 =2a$, $\lambda_1\lambda_2 =-ab$
and
\begin{equation}\label{E:P23reducible}
P_2(m,n)=(m+\lambda_1 n)(m+\lambda_2 n),\quad
P_3(m,n)=(m+\lambda_3 n)(m+\lambda_4 n).
\end{equation}
Since $\lambda_1 \lambda_2<0$ and $\lambda_3 \lambda_4<0$,
we have $\lambda_1 \neq \lambda_2$ and $\lambda_3 \neq \lambda_4$.
From now on we assume that $s$ is large enough so that
for $p>s$ we have $\omega_{P_1}(p) \in \{0,2\}$ and
\begin{equation}\label{new_a+b=c_new_E2: p not dividing}
p\nmid 2 ab(a+b)\lambda_1 \lambda_2 (\lambda_1 - \lambda_2)
(ab + \lambda_1 ^2)
\lambda_3 \lambda_4 (\lambda_3 - \lambda_4)
(ab + \lambda_3 ^2).
\end{equation}
Given $Q_1 \in \Phi_{s,r,P_1}$, $Q_2 \in \Phi_{r,K}$ and
$Q_3 \in \Phi_{K,L}$,
we will choose an integer $v$ depending on $r,K,L,Q_1,Q_2, Q_3$, and
we will average over the grid
$(Q_{\delta,L} m+1, Q_{\delta,L} n +v)$ in \eqref{T: main a+b=c eq}.
\begin{lemma}\label{a+b=c_new_lemma_choice_of_v_1_a=c_P_3_red}
Let $\delta>0$, $s,r,K,L\in \N$ with $s<r<K<L/2$,
$P_1,P_2,P_3$ be as in
\eqref{quadratic_forms_a+b=c}, with $P_2,P_3$ as in \eqref{E:P23reducible},
and let $Q_1 \in
\Phi_{s,r,P_1}$, $Q_2 \in \Phi_{r,K}$ and $Q_3 \in \Phi_{K,L}$.
Then there is an integer
$v$ with $0\leq v \leq Q_{\delta, L}-1$ such that
\begin{enumerate}[(i)]
\item $(Q_{\delta, L}, P_1(1,v))=Q_1$,
$\frac{P_1(1,v)}{Q_1}\equiv Q_1 ^{-1} \pmod {Q_s}$,
\item $(Q_{\delta, L},1+\lambda_1 v)= Q_2$,
$\frac{1+\lambda_1 v}{Q_2} \equiv Q_2 ^{-1} \pmod{{Q_s}}$,
\item $(Q_{\delta, L},1+\lambda_2 v)=1$ and
$1+\lambda_2 v \equiv 1 \pmod{{Q_s}}$,
\item $(Q_{\delta, L}, 1+\lambda_3 v)= Q_3$,
$\frac{1+\lambda_3 v}{Q_3} \equiv Q_3 ^{-1} \pmod{{Q_s}}$,
\item $(Q_{\delta, L},1+\lambda_4 v)=1$ and
$1+\lambda_4 v \equiv 1 \pmod{{Q_s}}$.
\end{enumerate}
\end{lemma}
\begin{proof}
For $p\in \P$ with $s<p \leq r$ and $\omega_{P_1}(p)=2$,
we may use Hensel's lemma to find $\zeta_1 = \zeta_1 (p, Q_1)$ so that if
$g_1(x):=P_1(1,x)-p^{\theta_p(Q_1)}$, then $g_1(\zeta_1)\equiv 0
\pmod {p^{\theta_p(Q_1)+1}}$.
Using the Chinese Remainder theorem ,we may then choose $v$ with
$0\leq v \leq Q_{\delta,L}-1$
such that
\begin{itemize}
\item $v\equiv 0 \pmod {p^{2s}}$ for $p\leq s$,
\item $v\equiv 0 \pmod {p}$ for $s<p \leq r$ with $\omega_{P_1}(p)=0$,
\item $v\equiv \zeta_1(p, Q_1)$
$\pmod {p^{\theta_p(Q_1)+1}}$ for $s< p\leq r$ with
$\omega_{P_1}(p)=2$,
\item $v\equiv \lambda_1 ^{-1} (p^{\theta_p(Q_2)} -1 )\pmod{p^{\theta_p(Q_2)+1}}$ for $r<p\leq K$,
\item $v\equiv \lambda_3 ^{-1} (p^{\theta_p(Q_3)} -1 )\pmod{p^{\theta_p(Q_3)+1}}$ for $K<p\leq L$.
\end{itemize}
The rest of the proof is very similar to the proof of
\cref{a+b=c_new_lemma_choice_of_v_1}, so it is omitted.
\end{proof}
As in \cref{Sec P2 P3 irreducible}, for each $\delta\in(0,1/2)$, let
\begin{equation}\label{a+b=c_new_S_delta_a=c_P_3_reduc}
S_{\delta}:=\{(m,n)\in \N^2 : m> 2\gamma n, \:
|(P_j(m,n))^i -1|\leq \delta\, \text{ for }
j=1,2,3\}.
\end{equation}
The case when $P_2$, $P_3$ are reducible in
\cref{T: main a+b=c} follows from the following
result.
\begin{proposition}\label{a+b=c_new_P: main a=c P_3 reducible}
Let $(X,\CX,\mu,T_n)$ be a pretentious multiplicative action, let
$F\in L^{\infty}(\mu)$ with $F\geq 0$,
let $P_1,P_2,P_3$ be as in
\eqref{quadratic_forms_a+b=c}, and suppose that
$P_2,P_3$ are reducible.
Also, let $S_{\delta}$ be as in \eqref{a+b=c_new_S_delta_a=c_P_3_reduc} and $v=v(\delta, s, r,K,L,Q_1,Q_2,Q_3)$ be as in
\cref{a+b=c_new_lemma_choice_of_v_1_a=c_P_3_red}.
Then
\begin{multline}\label{a+b=c_new_E: P: main a=c P_3 reducible}
\suplim_{\delta,s, r,K,L,N}\:
\cesE_{Q_{1}\in \Phi_{s,r,P_1}} \cesE_{Q_{2}\in \Phi_{r,K}}
\cesE_{Q_{3}\in \Phi_{K,L}}
\frac{1}{\overline{\textup{d}}(S_{\delta})} \cesE_{(m,n) \in [N]^2}
\1_{S_{\delta}}(m,n)\cdot \\
\int_{X} F\cdot \prod_{j=1} ^3
T_{P_j(Q_{\delta,L} m+1, Q_{\delta,L} n +v)} F
\, d \mu \geq \Big(\int_X F \, d \mu\Big)^4.
\end{multline}
\end{proposition}
Again, to prove \cref{a+b=c_new_P: main a=c P_3 reducible}, one must
first establish an analog of \cref{a+b=c_new_L: main a=c P_2 reducible},
and then use that result to prove \cref{a+b=c_new_P: main a=c P_3 reducible}.
In both proofs, the only essential difference from those in \cref{Sec
P2 P3 irreducible} is that, to handle the terms
$T_{P_2(Q_{\delta,L} m+1, Q_{\delta,L} n+v)} F$ and
$T_{P_3(Q_{\delta,L} m+1, Q_{\delta,L} n+v)} F$, one must use
\cref{linear_conc} and \cref{bilinear_lemma} instead of \cref{quad_conc}.
\bibliographystyle{abbrv}
\bibliography{refs}
\end{document}
\begin{thebibliography}{10}
\bibitem{bergelson-ergodic-ramsey}
V.~Bergelson.
\newblock Ergodic {R}amsey theory.
\newblock In {\em Logic and combinatorics ({A}rcata, {C}alif., 1985)}, volume \textbf{65} of {\em Contemp. Math.}, pages 63--87. Amer. Math. Soc., Providence, RI, 1987.
\bibitem{Bergelson_multiplicative}
V.~Bergelson.
\newblock Multiplicatively large sets and ergodic {R}amsey theory.
\newblock {\em Israel Journal of Mathematics}, \textbf{148}:23--40, 2005.
\bibitem{browning-prendiville-roth}
T.~D. Browning and S.~M. Prendiville.
\newblock A transference approach to a {R}oth-type theorem in the squares.
\newblock {\em International Mathematics Research Notices}, (7):2219--2248, 2017.
\bibitem{chapman-part-reg}
J.~Chapman.
\newblock Partition regularity for systems of diagonal equations.
\newblock {\em International Mathematics Research Notices}, (17):13272--13316, 2022.
\bibitem{Charamaras-multiplicative}
D.~Charamaras.
\newblock Mean value theorems in multiplicative systems and joint ergodicity of additive and multiplicative actions.
\newblock {\em Transactions of the American Mathematical Society}, \textbf{378}(3):1883--1937, 2025.
\bibitem{cha-mou-tsin}
D.~Charamaras, A.~Mountakis, and K.~Tsinas.
\newblock On multiplicative recurrence along linear patterns.
\newblock Preprint 2024, arXiv:2412.03504.
\bibitem{chow-lindqvist-prendiville}
S.~Chow, S.~Lindqvist, and S.~Prendiville.
\newblock Rado's criterion over squares and higher powers.
\newblock {\em Journal of the European Mathematical Society}, \textbf{23}(6):1925--1997, 2021.
\bibitem{chu-2-commuting}
Q.~Chu.
\newblock Multiple recurrence for two commuting transformations.
\newblock {\em Ergodic Theory and Dynamical Systems}, \textbf{31}(3):771--792, 2011.
\bibitem{csikvari-gyarmati-sarkozy}
P.~Csikv\'ari, K.~Gyarmati, and A.~S\'ark\"ozy.
\newblock Density and {R}amsey type results on algebraic equations with restricted solution sets.
\newblock {\em Combinatorica}, \textbf{32}(4):425--449, 2012.
\bibitem{dinasso-luperi}
M.~Di~Nasso and L.~Luperi~Baglini.
\newblock Ramsey properties of nonlinear {D}iophantine equations.
\newblock {\em Advances in Mathematics}, \textbf{324}:84--117, 2018.
\bibitem{Fra-Hom_Triples}
N.~Frantzikinakis.
\newblock Decomposition results for multiplicative actions and applications.
\newblock Preprint 2025, arXiv:2503.15175.
\bibitem{Fra-Host-structure-multiplicative}
N.~Frantzikinakis and B.~Host.
\newblock Higher order {F}ourier analysis of multiplicative functions and applications.
\newblock {\em Journal of the American Mathematical Society}, \textbf{30}(1):67--157, 2017.
\bibitem{fra-klu-mor-2}
N.~Frantzikinakis, O.~Klurman, and J.~Moreira.
\newblock Partition regularity of generalized {P}ythagorean pairs.
\newblock Preprint 2024, arXiv:2407.08360.
\bibitem{Fra-Klu-Mor}
N.~Frantzikinakis, O.~Klurman, and J.~Moreira.
\newblock Partition regularity of {P}ythagorean pairs.
\newblock {\em Forum of Mathematics, Pi}, \textbf{13}:Paper No. e5, 52, 2025.
\bibitem{furstenberg_szemeredi_thm}
H.~Furstenberg.
\newblock Ergodic behavior of diagonal measures and a theorem of {S}zemer\'edi on arithmetic progressions.
\newblock {\em Journal d'Analyse Math\'ematique}, \textbf{31}:204--256, 1977.
\bibitem{Graham-prob-1}
R.~Graham.
\newblock Some of my favorite problems in {R}amsey theory.
\newblock In {\em Combinatorial number theory}, pages 229--236. de Gruyter, Berlin, 2007.
\bibitem{Graham-prob-2}
R.~Graham.
\newblock Old and new problems and results in {R}amsey theory.
\newblock In {\em Horizons of combinatorics}, volume \textbf{17} of {\em Bolyai Soc. Math. Stud.}, pages 105--118. Springer, Berlin, 2008.
\bibitem{Granville-Sound-book}
A.~Granville and K.~Soundararajan.
\newblock Multiplicative number theory: The pretentious approach. {B}ook manuscript in preparation.
\bibitem{green-lindqvist}
B.~J. Green and S.~Lindqvist.
\newblock Monochromatic solutions to {$x+y=z^2$}.
\newblock {\em Canadian Journal of Mathematics}, \textbf{71}(3):579--605, 2019.
\bibitem{heule-kullmann-marek}
M.~J.~H. Heule, O.~Kullmann, and V.~W. Marek.
\newblock Solving and verifying the {B}oolean {P}ythagorean triples problem via cube-and-conquer.
\newblock In {\em Theory and applications of satisfiability testing---{SAT} 2016}, volume \textbf{9710} of {\em Lecture Notes in Comput. Sci.}, pages 228--245. Springer, [Cham], 2016.
\bibitem{khalfalah-szemeredi}
A.~Khalfalah and E.~Szemer\'edi.
\newblock On the number of monochromatic solutions of {$x+y=z^2$}.
\newblock {\em Combinatorics, Probability and Computing}, \textbf{15}(1-2):213--227, 2006.
\bibitem{Klu-Man-Po-Ter}
O.~Klurman, A.~P. Mangerel, C.~Pohoata, and J.~Ter\"av\"ainen.
\newblock Multiplicative functions that are close to their mean.
\newblock {\em Transactions of the American Mathematical Society}, \textbf{374}(11):7967--7990, 2021.
\bibitem{moreira-monochromatic}
J.~Moreira.
\newblock Monochromatic sums and products in {$\Bbb N$}.
\newblock {\em Annals of Mathematics. (2)}, \textbf{185}(3):1069--1090, 2017.
\bibitem{Rado-original}
R.~Rado.
\newblock Studien zur {K}ombinatorik.
\newblock {\em Mathematische Zeitschrift}, \textbf{36}(1):424--470, 1933.
\bibitem{schur_thm}
I.~Schur.
\newblock \"{U}ber die {K}ongruenz $x^m + y^m \equiv z^m$ (\textup{mod} $p$).
\newblock {\em Jahresbericht der Deutschen Mathematiker-Vereinigung}, \textbf{25}:114--117, 1916.
\bibitem{van_der_Waerden_thm}
B.~L. van~der Waerden.
\newblock Beweis einer baudetschen vermutung.
\newblock {\em Nieuw Archief voor Wiskunde}, \textbf{15}:212--216, 1927.
\end{thebibliography}