Odd values of the Ramanujan tau function

Extracted source text · arXiv 2101.02933

\documentclass{amsart}
\usepackage{tabu}
\usepackage[margin=95pt]{geometry}
%\usepackage[round,authoryear]{natbib}
%\usepackage[hidelinks]{hyperref}
\usepackage{enumitem}
\allowdisplaybreaks

\usepackage{amssymb}
\usepackage{amsmath}
\usepackage{amscd}
\usepackage{amsbsy}
\usepackage{comment}
\usepackage[matrix,arrow]{xy}
\usepackage{hyperref}
\setlength{\columnsep}{1cm}
%\usepackage{showlabels}

\DeclareMathOperator{\Rad}{Rad}
\DeclareMathOperator{\LCM}{LCM}

\DeclareMathOperator{\Norm}{Norm}
\DeclareMathOperator{\Disc}{Disc}
\DeclareMathOperator{\GL}{GL}
\DeclareMathOperator{\SL}{SL}
\DeclareMathOperator{\Sel}{Sel}
\DeclareMathOperator{\Ker}{Ker}
\DeclareMathOperator{\Img}{Im}
\DeclareMathOperator{\Res}{Res}
\DeclareMathOperator{\rank}{rank}
\DeclareMathOperator{\Pic}{Pic}
\DeclareMathOperator{\red}{red}
\newcommand{\vv}{{\upsilon}}
\DeclareMathOperator{\Div}{Div}
\DeclareMathOperator{\divisor}{div}
\DeclareMathOperator{\Alb}{Alb}
\DeclareMathOperator{\Gal}{Gal}
\DeclareMathOperator{\norm}{Norm}
\DeclareMathOperator{\trace}{Tr}
\DeclareMathOperator{\ord}{ord}
\DeclareMathOperator{\image}{im}
\DeclareMathOperator{\lcm}{lcm}
\DeclareMathOperator{\Reg}{Reg}
\DeclareMathOperator{\Aut}{Aut}
\newcommand{\crt}{\sqrt[3]{2}}
\newcommand{\bxi}{\boldsymbol{\xi}}
%\newcommand{\SHA}[1]{{\cyr X}(#1)}
%\newcommand{\p}{{\mathfrak P}}
%\newcommand{\A}{{\mathbb A}}
\newcommand{\mS}{\mathcal{S}}
\newcommand{\Q}{{\mathbb Q}}
\newcommand{\Z}{{\mathbb Z}}
%\newcommand{\N}{{\mathbb N}}
\newcommand{\C}{{\mathbb C}}
\newcommand{\R}{{\mathbb R}}
\newcommand{\F}{{\mathbb F}}
%\newcommand{\D}{{\mathbb D}}
\newcommand{\PP}{{\mathbb P}}
%\newcommand{\LL}{{\mathbb L}}
%\newcommand{\HH}{{G_{\LL}}}
\newcommand{\cA}{\mathcal{A}}
\newcommand{\cB}{\mathcal{B}}
\newcommand{\cK}{\mathcal{K}}
\newcommand{\cL}{\mathcal{L}}
\newcommand{\cI}{\mathcal{I}}
\newcommand{\cM}{\mathcal{M}}
\newcommand{\cY}{\mathcal{Y}}
\newcommand{\cE}{\mathcal{E}}
\newcommand{\cF}{\mathcal{F}}
\newcommand{\cG}{\mathcal{G}}
\newcommand{\cH}{\mathcal{H}}
\newcommand{\cN}{\mathcal{N}}
\newcommand{\cP}{\mathcal{P}}
\newcommand{\cQ}{\mathcal{Q}}
\newcommand{\cR}{\mathcal{R}}
\newcommand{\cT}{\mathcal{T}}
\newcommand{\cS}{\mathcal{S}}
\newcommand{\cZ}{\mathcal{Z}}
\newcommand{\cJ}{{\mathcal J}}
\newcommand{\cC}{\mathcal C}
\newcommand{\cD}{\mathcal D}
\newcommand{\cV}{\mathcal V}
\newcommand{\cW}{\mathcal W}
\newcommand{\OO}{{\mathcal O}}
\newcommand{\ga}{{\mathfrak{a}}}
\newcommand{\gb}{{\mathfrak{b}}}
\newcommand{\gc}{{\mathfrak{c}}}
\newcommand{\pic}{{\rm Pic}^0(Y)}
\newcommand{\fY}{\mathfrak{Y}}
\newcommand{\ff}{\mathfrak{f}}
\newcommand{\fg}{\mathfrak{g}}
\newcommand{\fp}{\mathfrak{p}}
\newcommand{\fq}{\mathfrak{q}}
\newcommand{\fa}{\mathfrak{a}}
\newcommand{\fb}{\mathfrak{b}}
\newcommand{\mP}{\mathcal{P}}
\newcommand{\xx}{\mathbf{x}}
\newcommand{\cc}{\mathbf{c}}
\newcommand{\bs}{\mathbf{s}}
\DeclareMathOperator{\Jac}{Jac}

\newcommand\margnote[1]{\marginpar{\tiny\begin{minipage}{20mm}\begin{flushleft} #1\end{flushleft}\end{minipage}}}

%\renewcommand{\baselinestretch}{1.5}

\def\mod#1{{\ifmmode\text{\rm\ (mod~$#1$)}
\else\discretionary{}{}{\hbox{ }}\rm(mod~$#1$)\fi}}

\begin {document}

\newtheorem{thm}{Theorem}
\newtheorem{lem}{Lemma}[section]
\newtheorem{prop}[lem]{Proposition}
\newtheorem{alg}[lem]{Algorithm}
\newtheorem{cor}[lem]{Corollary}
\newtheorem*{conj}{Conjecture}

\theoremstyle{definition}

\newtheorem{ex}[thm]{Example}

\theoremstyle{remark}
\newtheorem{rem}[thm]{Remark}

\newtheorem{ack}{Acknowledgement}

\title[Ramanujan tau function]
{Odd Values of the Ramanujan tau function}
\author[Michael Bennett]{Michael A. Bennett}
\address{Department of Mathematics, University of British Columbia, Vancouver, B.C., V6T 1Z2 Canada}
\email{bennett@math.ubc.ca}

\author{Adela Gherga}
\address{Mathematics Institute, University of Warwick, Coventry CV4 7AL, United Kingdom}
\email{Adela.Gherga@warwick.ac.uk}

\author{Vandita Patel}
\address{Department of Mathematics, University of  Manchester, Manchester M13 9PL, United Kingdom}
\email{vandita.patel@manchester.ac.uk}

\author{Samir Siksek}
\address{Mathematics Institute, University of Warwick, Coventry CV4 7AL, United Kingdom}
\email{S.Siksek@warwick.ac.uk}
\thanks{ The first-named author is supported by NSERC.
The second and fourth-named authors are
supported by an EPSRC Grant EP/S031537/1 \lq\lq Moduli of
elliptic curves and classical Diophantine problems\rq\rq.}

\date{\today}

\keywords{Ramanujan tau-function, Exponential equation,
Lucas sequence,  Galois representation,
Frey curve,
modularity, level lowering, Baker's bounds,
Thue-Mahler equations}
\subjclass[2010]{Primary 11D61, Secondary 11D41, 11F80, 11F41}

\begin {abstract}
We prove a number of results regarding odd values of the 
Ramanujan $\tau$-function.
For example, we prove the existence of an effectively computable positive
constant $\kappa$ such that if $\tau(n)$ is odd and $n \ge 25$
then either
\[
	P(\tau(n)) \; > \; \kappa \cdot \frac{\log\log\log{n}}{\log\log\log\log{n}}
\]
or there exists a prime $p \mid n$ with $\tau(p)=0$. Here $P(m)$ denotes
the largest prime factor of $m$.
We also solve the equation $\tau(n)=\pm 3^{b_1} 5^{b_2} 7^{b_3} 11^{b_4}$
and the equations $\tau(n)=\pm q^b$ where $3\le q < 100$ is prime and the exponents are arbitrary nonnegative integers.
We make use of a variety of methods, including the Primitive Divisor
Theorem of Bilu, Hanrot and Voutier, bounds for solutions
to Thue--Mahler equations due to Bugeaud and Gy\H{o}ry,
and the modular approach via Galois representations of Frey-Hellegouarch
elliptic curves.
\end {abstract}
\maketitle

%------------------------------
\section{Introduction}
%------------------------------

The {\it Ramanujan $\tau$-function} $\tau (n)$ is defined via the expansion
\begin{equation} \label{taudef}
\Delta (z) = q \prod_{n=1}^\infty (1-q^n)^{24} = \sum_{n=1}^\infty \tau (n) q^n, \; \; \; \; q=e^{2\pi i z}.
\end{equation}
It was conjectured by Ramanujan \cite{Ram} and proved by Mordell \cite{Mor} that $\tau (n)$ is a multiplicative function, i.e. that
$$
\tau (n_1n_2)=\tau (n_1) \tau (n_2),
$$
for all coprime pairs of positive  integers $n_1$ and $n_2$. Further, we have
$$
 \sum_{n=1}^\infty \tau (n) q^n \equiv q \prod_{n=1}^\infty (1+q^{8n})^{3} \equiv q \prod_{n=1}^\infty   (1-q^{8n}) (1+q^{8n})^{2}   \equiv \sum_{n=0}^\infty q^{(2n+1)^2} \mod{2},
 $$
 via Jacobi's triple product formula,
whence $\tau(n)$ is odd precisely when $n$ is an odd square and, in particular, $\tau (p)$ is even for every prime $p$.

 Amongst the many open questions about the possible values of $\tau (n)$, the most notorious is a conjecture of Lehmer \cite{Leh} to the effect that $\tau (n)$ never vanishes.
  In terms of the size of values of $\tau$, one has  the upper bound of Deligne \cite{Deligne} (originally conjectured by Ramanujan) :
 \begin{equation} \label{Del:bound}
 \left| \tau (p) \right| \leq 2 \cdot p^{11/2},
 \end{equation}
 valid for prime $p$.
In the other direction, Atkin and Serre \cite{Ser}  conjectured (as a strengthening of Lehmer's conjecture) that, for $\epsilon > 0$,
$$
 \left| \tau (p) \right| \gg_\epsilon p^{9/2-\epsilon},
 $$
so that, in particular, given a fixed integer $a$, there are at most finitely many primes $p$ for which $\tau (p)=a$.
While this problem remains open, in the special case where the integer $a$ is odd, Murty, Murty and Shorey \cite{MuMuSh} proved that the equation
\begin{equation} \label{equal}
\tau (n)=a,
\end{equation}
has at most finitely many solutions in integers $n$ (note that, in this case, $n$ is necessarily an odd square). More precisely, they demonstrated the existence of an effectively computable positive constant  $c$ such that if $\tau (n)$ is odd, then
$$
|\tau (n)| > (\log (n))^c.
$$
A number of recent papers have treated the problem of explicitly demonstrating that equation (\ref{equal}) has, in fact, no solutions, for various odd values of $a$, including $a=\pm 1$ (Lygeros and Rozier \cite{LyRo}), $|a| < 100$ an odd prime (Balakrishnan, Craig and Ono \cite{BaCrOn}, Balakrishnan, Craig, Ono and Tsai \cite{BaCrOnTs},  Dembner and Jain \cite{DeJa}), and $|a| < 100$ an odd integer (Hanada and Madhukara \cite{HaMa}).

In this paper, we derive what might be considered a non-Archimedean analogue of the work of Murty, Murty and Shorey.
Let us define $P(m)$ to be the greatest prime factor of an integer $|m|>1$. We prove the following.

\begin{thm} \label{thm1}
There exists an effectively computable constant $\kappa > 0$ such that if $\tau (n)$ is odd, with $n \geq 25$, then either
\begin{equation} \label{guppy}
P(\tau (n)) \; > \; \kappa  \cdot \frac{ \log \log  \log n}{\log \log \log \log n},
\end{equation}
or there exists a prime $p \mid n$ for which $\tau (p)=0$.
\end{thm}

Recall that a {\it powerful number} (also known as {\it squarefull} or {\it $2$-full}) is defined to be an integer $n$ with the property that if a prime $p \mid n$, then necessarily $p^2 \mid n$. Equivalently, we can write such an integer as $n=a^2b^3$, where $a$ and $b$ are integers. Our techniques actually show the following (from which Theorem \ref{thm1} is an immediate consequence).
\begin{thm} \label{thm1a}
We have
$$
\lim_{n \rightarrow \infty} P(\tau (n)) = \infty,
$$
where the limit is taken over powerful numbers $n$ for which $\tau (p) \neq 0$ for each $p \mid n$.
More precisely, there exists an effectively computable constant $\kappa > 0$ such that if $n \geq 25$ is powerful,  either
\begin{equation}\label{eqn:lb}
	P(\tau (n)) \; > \; \kappa \cdot \frac{ \log \log  \log n}{\log \log \log \log n}.
\end{equation}
or there exists a prime $p \mid n$ for which $\tau (p)=0$.
\end{thm}

The restriction that $n$ has no prime divisors $p$ for which $\tau (p)=0$ is, in fact, necessary if one wishes to obtain a lower bound upon  $P(\tau (n))$ that tends to $\infty$ with $n$.
Indeed, one may observe that, if $\tau (p) =0$, then
(see \eqref{eqn:recursion} below)
$$
P \left( \tau (p^{2k}) \right) = P \left( (-1)^k p^{11k} \right) =p
$$
is bounded independently of $k$. While Lehmer's conjecture remains unproven, we do know that if there is a prime $p$ for which $\tau (p)=0$, then
\begin{equation}\label{eqn:Derickx}
p > 816212624008487344127999,
\end{equation}
by work of Derickx, van Hoeij and Zeng \cite{DeHoZe}.

Theorem~\ref{thm1a} is an easy consequence of the following
result.
\begin{thm}\label{thm:main}
There is a computable positive constant $\eta$ such that
for any prime $p$ with $\tau(p) \ne 0$ and any $m \ge 2$, 
\begin{equation}\label{eqn:required}
	P(\tau(p^{m})) \; > \; \eta \cdot \frac{ \log \log  (p^{m})}{\log \log \log (p^{m})}.
\end{equation}
%Moreover,
%\begin{equation}\label{eqn:precise}
%	P(\tau(p^{m})) \; > \; \eta_2 \cdot \log\log(p), \qquad \text{for $m \in \{2, 3, 4, 5, 7, 9, 11\}$}.
%\end{equation}
\end{thm}
We note that Theorem~\ref{thm:main} implies Theorem~\ref{thm1a}. Indeed, let $n$
be a powerful number and $p^{m}$ be the largest prime power divisor of
$n$.  Then $m \ge 2$, and $p^m \gg \log{n}$, whence \eqref{eqn:lb}  follows  immediately from \eqref{eqn:precise}. Our arguments show the following.
\begin{thm}\label{thm:main2}
Let $m \ge 2$.  There is a  computable positive constant $\delta(m)$, depending
only on $m$, such that
for any prime $p$ with $\tau(p) \ne 0$, 
\begin{equation}\label{eqn:precise}
	P(\tau(p^{m})) \; > \; \delta(m) \cdot \log\log(p). %\qquad \text{for $m \in \{2, 3, 4, 5, 7, 9, 11\}$}.
\end{equation}
\end{thm}


%We will actually show that, given $m \geq 2$ and prime $p$ for which $\tau (p) \neq 0$, 
%$$
%P(\tau (p^m)) \gg \log \log  (p^m), \; \mbox{ for }m \in \{ 2, 3, 4, 5 \},
%$$
%and
%$$
%P(\tau (p^m)) \gg \frac{ \log \log  (p^m)}{\log \log \log (p^m)},
%$$
%more generally, where the implied constants are absolute. 
We note that our bounds neither imply nor are implied by work of Luca and Shparlinski \cite{LuSh} who proved that
$$
P \left( \tau (p) \tau(p^2) \tau(p^3) \right) \gg \frac{\log \log  (p) \log \log \log (p)}{\log \log \log \log (p)}.
$$

\medskip

To demonstrate that these results and the techniques underlying them are somewhat practical, we prove the
 following computational ``coda'', solving equation (\ref{equal}) where the prime divisors of $a$, rather than $a$ itself, are fixed.

\begin{thm} \label{thm2a}
If $n$ is a powerful positive integer, then either $n=8$, where we have
$$
\tau (8) = 2^9 \cdot 3 \cdot 5 \cdot 11,
$$
or
$$
P(\tau (n)) \geq 13.
$$
\end{thm}

\begin{cor} \label{thm2}
If $n$ is a positive integer for which $\tau (n)$ is odd, then
\begin{equation} \label{boundary}
P(\tau (n)) \geq 13.
\end{equation}
In other words, the equation
\begin{equation} \label{fred}
\tau (n) = \pm 3^\alpha 5^\beta 7^\gamma 11^\delta
\end{equation}
has no solutions in integers $n \geq 2$ and $\alpha, \beta, \gamma, \delta \geq 0$.
\end{cor}

It is conjectured that $|\tau (n)|$ takes on infinitely many prime values, the smallest of which corresponds to
$$
\tau (251^2) = -80561663527802406257321747.
$$
Our arguments enable us to eliminate the possibility of powers of small primes arising as values of $\tau$. By way of example, we have the following.
\begin{thm} \label{thm3}
The equation
$$
\tau (n) = \pm q^\alpha
$$
has no solutions in prime $q$ with $3 \leq q < 100$, and $\alpha \ge 0$, $n \ge 2$ integers.
\end{thm}

It is worth observing that the techniques we employ here are readily extended to treat more generally coefficients $\lambda_f(n)$ of cuspidal newforms of (even) weight $k \geq 4$ for $\Gamma_0(N)$, with trivial character and $\lambda_f(p)$ even for suitably large prime $p$; our results correspond to the case of $\Delta (z)$ in (\ref{taudef}), where $k=12$ and $N=1$. For simplicity, we will restrict our attention to $\tau (n)$ and $\Delta (z)$; readers interested in the more general situation should consult the paper of Murty and Murty \cite{MuMu} (see also \cite{BaCrOnTs}).

We should also comment on the particular choice of the constant $``13"$ on the right hand side of inequality (\ref{boundary}) in Corollary \ref{thm2} (and analogously in Theorem \ref{thm2a}). As we shall observe, the weaker result with $13$ replaced by $11$ (corresponding to equation (\ref{fred}) with $\delta=0$) reduces via local arguments to the resolution of a single Thue equation; this is the content of Proposition 6 of  Luca, Maboso, Stanica \cite{LuMaSt}. Corollary \ref{thm2}  as stated requires (apparently at least) the full use of our various techniques, including the Primitive Divisor
Theorem,  solution of a variety of Thue--Mahler equations,
and resolution of hyperelliptic equations through appeal to the modularity of Galois representations attached to Frey-Hellegouarch
elliptic curves. A stronger version of Corollary \ref{thm2}  with $13$ replaced by $17$ in  (\ref{boundary}) is possibly within range of this approach, though computationally significantly more involved. An analogous result with $13$ replaced by $19$  would likely require fundamentally new ideas.


\bigskip

This paper is organized as follows. 
In Section~\ref{sec:congruences}, we recall some standard congruences
for the Ramanujan-tau function that we  use. later in the paper.
In Section~\ref{sec:lucas}, we connect the sequence
$m \mapsto \tau(p^{m-1})$, for a fixed prime $p$, to a Lucas sequence $\{u_m\}$, allowing
us to appeal to  the Primitive Divisor Theorem
of Bilu, Hanrot and Voutier. 
In Section~\ref{sec:Fn}, we introduce a sequence of homogenous
polynomials $\Psi_m(X,Y) \in \Z[X,Y]$ that are intimately connected to the $\{u_m\}$.
We will use these polynomials in Section~\ref{sec:Bugeaud}, together  with  a theorem of Bugeaud on prime divisors of $ax^u+by^v$, to prove Theorem~\ref{thm:main2}.
In Section~\ref{sec:thm1a}, we relate the 
equation $\tau(p^m)=\pm p_1^{\alpha_1} \cdots p_r^{\alpha_1}$
to a Thue--Mahler equation, whence a theorem of Bugeaud and Gy\H{o}ry
enables us to deduce Theorem~\ref{thm:main}.
In Sections~\ref{sec:taup2}, \ref{sec:taup4} and \ref{sec:taup3},
we solve the equations $\tau(p^k)= \pm q^b$ where $k \in \{2, 4 \}$,
$p$ and $q$ are prime, and $3 \le b <100$, and also the equations
$\tau(p^k)= \pm 3^{b_1} 5^{b_2} 7^{b_3} 11^{b_4}$, where $2 \le k \le 4$
and $p$ is prime. Our method in Section~\ref{sec:taup2}, \ref{sec:taup4} and \ref{sec:taup3}
is to associate to a hypothetical solution a Frey-Hellegouarch curve and relate 
this to a weight $2$ modular form
of small level, using recipes of the first author and Skinner
which in turn builds on the modularity of elliptic curves
due to Wiles, Breuil, Conrad Diamond and Taylor, and on
Ribet's Level-Lowering Theorem. In Section~\ref{sec:thm2a},
we prove Theorem~\ref{thm2a}, by combining the results of
Sections~\ref{sec:taup2}, \ref{sec:taup4} and \ref{sec:taup3},
and using the Primitive Divisor Theorem.
Finally, in Section~\ref{sec:thm3}, we prove Theorem~\ref{thm3};
the results of previous sections allow us to
reduce the equation $\tau(n)=\pm q^a$ with $3\le q < 100$
prime to Thue--Mahler equations of high degree, which are then
solved using an algorithm of the second and fourth author, von K\"{a}nel and Matschke.

%-----------------------------------------------------------------------------------------------------
\section{Congruences for the $\tau$ function} \label{sec:congruences}
%------------------------------------------------------------------------------------------------------

For future use, it will be of value for us to record some basic arithmetic facts about $\tau (n)$;
these are taken from Swinnerton--Dyer's article \cite{SwD}.
Here $\sigma_v(n)$ denotes the sum of the $v$-th powers of the divisors
of $n$. %Thus $\sigma_v(p)=p^v+1$ for a prime $p$.

\begin{equation}\label{mod2}
\begin{cases}
\tau(n) \equiv \sigma_{11}(n) \mod{2^{11}} & \text{if $n \equiv 1 \mod{8}$}\\
\tau(n) \equiv 1217 \cdot \sigma_{11}(n) \mod{2^{13}} & \text{if $n \equiv 3 \mod{8}$}\\
\tau(n) \equiv 1537 \cdot \sigma_{11}(n) \mod{2^{12}} & \text{if $n \equiv 5 \mod{8}$}\\
\tau(n) \equiv 705 \cdot \sigma_{11}(n) \mod{2^{14}} & \text{if $n \equiv 7 \mod{8}$}\\
\end{cases}
\end{equation}

\begin{equation}\label{mod3}
\tau(n) \equiv n^{-610} \cdot \sigma_{1231}(n)
\begin{cases}
\mod{3^6} & \text{if $n \equiv 1 \mod{3}$}\\
\mod{3^7} & \text{if $n \equiv 2 \mod{3}$}\\
\end{cases}
\end{equation}

\begin{equation}\label{mod5}
\tau(n) \equiv n^{-30} \sigma_{71}(n) \mod{5^3} \qquad
\text{if $5 \nmid n$}
\end{equation}

\begin{equation}\label{mod7}
\tau(n) \equiv n \cdot \sigma_9(n)
\begin{cases}
\mod{7} & \text{if $n \equiv 0$, $1$, $2$ or $4 \mod{7}$}\\
\mod{7^2} & \text{if $n \equiv 3$, $5$ or $6 \mod{7}$}\\
\end{cases}
\end{equation}

\begin{equation}\label{mod23}
\begin{cases}
\tau(p) \equiv 0 \mod{23} & \text{if $p$ is a quadratic non-residue
mod $23$}\\
\tau(p) \equiv 2 \mod{23} & \text{if $p=u^2+23v^2$ with $u \ne 0$}\\
\tau(p) \equiv -1 \mod{23} & \text{for other $p \ne 23$}
\end{cases}
\end{equation}

\begin{equation}\label{mod691}
\tau(n) \equiv \sigma_{11}(n) \mod{691}.
\end{equation}

\begin{lem}\label{lem:mod7}
Let $p \ne 7$ be a prime. Then $7 \nmid \tau(p^2)$.
\end{lem}
\begin{proof}
Suppose $7 \mid \tau(p^2)$. Then by \eqref{mod7}
\[
p^{18}+p^9+1 \equiv 0 \mod{7}.
\]
But $p^{18}=(p^6)^3 \equiv 1$ and $p^9 \equiv p^3 \equiv \pm 1 \mod{7}$ giving
a contradiction.
\end{proof}
\begin{lem}\label{lem:mod5}
Let $p \ne 5$ be a prime. Then $5 \nmid \tau(p^2)$.
\end{lem}
\begin{proof}
Suppose $5 \mid \tau(p^2)$. Then by \eqref{mod5}
\[
(p^{71})^2+p^{71}+1 \equiv 0 \mod{5}.
\]
However, this contradicts the fact that  the congruence $x^2+x+1 \equiv 0 \mod{5}$ has no solutions.
\end{proof}
\begin{lem}\label{lem:mod9}
Let $p \ne 3$ be a prime. Then $9 \nmid \tau(p^2)$.
\end{lem}
\begin{proof}
Suppose $9 \mid \tau(p^2)$. Then by \eqref{mod3}
\[
(p^{1231})^2+p^{1231}+1 \equiv 0 \mod{9}.
\]
Since the congruence $x^2+x+1 \equiv 0 \mod{9}$ has no solutions, we obtain the desired contradiction.
\end{proof}

\begin{comment}
Suppose we have
$$
\tau^2(p)-3 \cdot 11^\beta = p^{11}.
$$
In all cases, we have $\tau (p) \equiv 1+p^{11} \equiv 1+p \mod{8}$, whence
$$
p+2 \equiv 3 \cdot 11^\beta \mod{8}.
$$
It follows that $p \equiv \pm 1 \mod{8}$. If $p \equiv 1 \mod{8}$, then $\beta$ is even, while $p \equiv -1 \mod{8}$ implies that $\beta$ is odd. In the first case, we consider
$$
E \; \; : \; \; Y^2=X^3 + 2 \tau (p) X^2 + 3 \cdot 11^\beta X.
$$
This level lowers to $N=2^5 \cdot 3 \cdot 11 = 1056$. If we have $\tau (p) \equiv 0 \mod{23}$ (so that $p^{11} \equiv -1 \mod{23}$), then
$$
Y^2 \equiv X^3 +X \mod{23},
$$
whence $a_{23}(E)=0$ (and we hit, in Cremona's notation, one of 1056 b, d, f or g). If $\tau (p) \equiv 2 \mod{23}$ (so that $p^{11} \equiv 1 \mod{23}$), then
$$
Y^2 \equiv X^3 +4X^2+3X \mod{23},
$$
with, again, $a_{23}(E)=0$.  If $\tau (p) \equiv -1 \mod{23}$ (so that $p^{11} \equiv 1 \mod{23}$), then
$23 \mid 3 \cdot 11^\beta$, a contradiction.

Now
$$
\tau(p) \equiv p^{-30} (1+p^{71}) \mod{5^3}
$$
implies that
$$
\tau (p) \equiv p^2+p \mod{5},
$$
whence
$$
3 \equiv 3 \cdot 11^\beta = \tau^2(p)-p^{11} \equiv 1+ p^3+p^2 \mod{5}.
$$
It follows that $p \equiv 1, 2 \mod{5}$. If $p \equiv 2 \mod{5}$, we find that $a_5(E) = 0$, while $p \equiv 1 \mod{5}$ implies that $a_5(E)=2$. We are thus in the latter case (and are at 1056d or 1056g).

We also have
$$
\tau (p) \equiv p (1+p^9) \mod{7},
$$
whence
$$
\tau^2(p)-p^{11} \equiv p^2+2 \cdot p^4 +p-p^5 \mod{7}.
$$
\end{comment}



%-------------------------------------------------------------------
\section{Lucas Sequences}\label{sec:lucas}
%-------------------------------------------------------------------

In this section, for a fixed prime $p$ with $\tau(p) \ne 0$,
we show that the sequence $m \mapsto \tau(p^{m-1})$ can be appropriately
scaled to yield a Lucas sequence.
We begin by introducing Lucas sequences and recalling some of their
properties, 
mostly following the article of Bilu, Hanrot and Voutier \cite{BHV}.
A \textbf{Lucas pair} is a pair $(\alpha,\beta)$ of algebraic
numbers such that $\alpha+\beta$ and $\alpha\beta$ are
non-zero coprime rational integers, and $\alpha/\beta$
is not a root of unity. In particular, associated
to the Lucas pair $(\alpha,\beta)$ is a
\textbf{characteristic polynomial}
\[
X^2-(\alpha+\beta) X + \alpha \beta \; \in \; \Z[X].
\]
This polynomial has discriminant $D=(\alpha-\beta)^2 \in \Z \setminus \{0\}$.
Given a Lucas pair $(\alpha,\beta)$,
the corresponding \textbf{Lucas sequence} is given by
\[
u_n=\frac{\alpha^n-\beta^n}{\alpha-\beta}, \qquad n=0,1,2,\dotsc.
\]
Let $(\alpha,\beta)$ be a Lucas pair. A prime $\ell$ is a
\textbf{primitive divisor} of the $n$-th term of the corresponding
Lucas sequence
if $\ell$ divides $u_n$ but $\ell$ fails to divide $(\alpha-\beta)^2 \cdot u_1 u_2 \dotsc u_{n-1}$.
%A Lucas pair $(\alpha,\beta)$ is called \textbf{$n$-defective}
%if $u_n$ does not have a primitive divisor. 
We shall make
essential use of the celebrated Primitive Divisor Theorem
of Bilu, Hanrot and Voutier \cite{BHV}.
\begin{thm}[Bilu, Hanrot and Voutier]\label{thm:BHV}
Let $(\alpha,\beta)$ be a Lucas pair.
If $n \ge 5$ and $n \ne 6$ then $u_n$ has a primitive divisor.
\end{thm}
%We shall also make use of refinements of this theorem
%due to the same authors, and also due to
%Abouzaid \cite{Ab}, which provide a complete classification
%of $n$-defective Lucas pairs for $n=2$, $3$, $4$ and $6$.

Let $\ell$ be a prime. The smallest positive integer $m$
such that $\ell \mid u_m$ is called the \textbf{rank of
apparition of $\ell$}; we denote this
by $m_\ell$.
We shall also have need of  the following classical theorem of Carmichael \cite{Car}.
\begin{thm}[Carmichael]\label{thm:Carmichael}
Let $(\alpha,\beta)$ be a Lucas pair and $\ell$ be a prime.
\begin{enumerate}[label=(\roman*)]
\item If $\ell \mid \alpha \beta$ then $\ell \nmid u_m$
for all positive integers $m$.
\item Suppose $\ell \nmid \alpha \beta$.
Write $D=(\alpha-\beta)^2 \in \Z$.
\begin{enumerate}[label=(\alph*)]
\item If $\ell \ne 2$ and $\ell \mid D$, then
$m_\ell=\ell$.
\item If $\ell \ne 2$ and $\left(\frac{D}{\ell}\right)=1$, then $m_\ell \mid (\ell-1)$.
\item If $\ell \ne 2$ and $\left(\frac{D}{\ell} \right)=-1$, then $m_\ell \mid (\ell+1)$.
\item If $\ell=2$, then $m_\ell=2$ or $3$.
\end{enumerate}
\item If $\ell \nmid \alpha \beta$ then
\[
\ell \mid u_m \iff m_\ell \mid m.
\]
\end{enumerate}
\end{thm}
\begin{proof}
Note that the sequence $\{u_n\}$ satisfies the recurrence
\[
u_{n+2}-(\alpha+\beta) u_{n+1}+\alpha \beta u_n=0, \qquad u_0=0, \quad
u_1=1.
\]
	If $\ell \mid \alpha \beta$ then $u_n \equiv (\alpha+\beta)^{n-1} \pmod{\ell}$
for all $n \ge 1$. Since $\alpha+\beta$ and $\alpha \beta$
are coprime, $\ell \nmid (\alpha+\beta)$ and so $\ell \nmid u_n$
for all $n \ge 1$.

Suppose now that $\ell \nmid \alpha \beta$. Let $K=\Q(\alpha)=\Q(\beta)
=\Q(\sqrt{D})$
and $\lambda$ be a prime of $\OO_K$ above $\ell$. We first consider
(a). Here $\alpha=\beta+\gamma$ where $\lambda \mid \gamma$.
Thus
\[
u_n=\frac{\alpha^n-\beta^n}{\alpha-\beta}=
\frac{(\beta+\gamma)^n-\beta^n}{\gamma} \equiv n \beta^{n-1}  \pmod{\lambda}.
\]
Thus $\ell \mid u_n$ if and only if $\ell \mid n$.

Next we consider cases (b) and (c) together. Note that
\[
\ell \mid u_n \iff (\alpha/\beta)^n \equiv 1 \pmod{\lambda}.
\]
%Here $\lambda$ is a degree $1$
%prime ideal, and so $\OO_K/\lambda \cong \F_\ell$. 
Thus
$m_\ell$ is equal to the multiplicative order
of the image of $\alpha/\beta$ in $(\OO_K/\lambda)^*$.
If $D$ is a quadratic residue modulo $\ell$,
then $\ell$ splits as a product of two degree $1$
primes $\lambda$, $\lambda^\prime$ of $\OO_K$.
Thus $(\OO_K/\lambda)^* \cong \F_\ell^*$ and so
$m_\ell \mid (\ell-1)$.
Finally we suppose $D$ is a quadratic non-residue modulo $\ell$.
Then $\lambda=\ell \OO_K$, and so the natural map
$\Gal(K/\Q) \rightarrow \Gal(\F_\lambda/\F_\ell)$ is an isomorphism.
Note that $\alpha$ and $\beta$ are conjugates, and so
$(\alpha/\beta)^\ell \equiv \beta/\alpha \pmod{\lambda}$.
Thus $(\alpha/\beta)^{\ell+1} \equiv 1 \pmod{\lambda}$, whence
$m_\ell \mid (\ell+1)$. The final part of the theorem
is now also clear.
\end{proof}

\bigskip

Let us fix a prime $p$ and consider
the sequence 
\begin{equation} \label{Seq}
\left\{ 1, \tau (p), \tau (p^2), \tau (p^3), \ldots \right\}.
\end{equation}
We will associate to  this a Lucas pair and a corresponding Lucas sequence.
Our starting point is the identity
\begin{equation}\label{eqn:recursion}
\tau (p^m) = \tau (p) \tau (p^{m-1}) - p^{11} \tau (p^{m-2}),
\end{equation}
valid for all integer $m \geq 2$. Once again, this was conjectured by Ramanujan
\cite{Ram} and proved by Mordell \cite{Mor}.

Let $\gamma$ and
$\delta$ be the
roots of the quadratic equation
\[
X^2- \tau (p) X + p^{11}=0,
\]
so that
\[
\gamma+\delta=\tau (p) \qquad \text{ and } \qquad
\gamma \delta = p^{11}.
\]
Then
\[
(\gamma-\delta)^2 = (\gamma+\delta)^2 - 4
\gamma \delta = \tau^2 (p) - 4 p^{11}.
\]
It follows from
Deligne's bounds (\ref{Del:bound})  that  $\gamma$
and $\delta$ are non-real Galois conjugates.
From (\ref{eqn:recursion}), we have
\begin{equation}\label{eqn:taulucas}
\tau (p^m)
=\frac{{\gamma}^{m+1}-{\delta}^{m+1}}{\gamma-\delta}.
\end{equation}
\begin{lem}\label{lem:tauval}
Suppose $\tau(p) \ne 0$. Then $\ord_p(\tau(p)) \le 5$.
\end{lem}
\begin{proof}
From (\ref{Del:bound}), 
if $p^6 \mid \tau(p)$ and $\tau(p) \ne 0$, then necessarily 
$p \le 3$. However,
\[
\tau(2)=-2^3 \times 3 \; \; \mbox{ and } \; \;  \tau(3)=2^2 \times 3^2 \times 7,
\]
providing a contradiction and completing the proof.
\end{proof}
\begin{lem}\label{lem:notrootofunity}
Suppose $\tau(p) \ne 0$. Then $\gamma/\delta$
is not a root of unity.
\end{lem}
\begin{proof}
Observe that
\[
\frac{\gamma}{\delta}+
\frac{\delta}{\gamma}+2
\; = \; \frac{\tau(p)^2}{p^{11}}.
\]
By the previous lemma, the rational number $\tau(p)^2/p^{11}$
is not an integer and therefore not an algebraic integer.
It follows that $\gamma/\delta$ is not a root
of unity.
\end{proof}
The following is an immediate
consequence of \eqref{eqn:taulucas} and Lemma~\ref{lem:notrootofunity}.
\begin{lem}\label{taup=0}
If $\tau(p)=0$ then
\[
\tau(p^{m})=\begin{cases}
0 & \text{if $m$ is odd},\\
(-p^{11})^{m/2} & \text{if $m$ is even}.
\end{cases}
\]
If $\tau(p) \ne 0$ then $\tau(p^m) \ne 0$ for all $m \ge 1$.
\end{lem}

Note that $\gcd(\gamma+ \delta,\gamma \delta)
=\gcd(\tau(p),p^{11})=1$
if and only if $p \nmid \tau(p)$. Thus
the sequence $m \mapsto \tau(p^{m-1})$ is a Lucas
sequence precisely when $p \nmid \tau(p)$. We note that $p \mid \tau(p)$
for
$$
p =  2, 3, 5, 7, 2411,  7758337633, \ldots
$$
We expect that $p \mid \tau(p)$
for infinitely many primes $p$; see Lygeros and Rozier \cite{LyRo0} for a
discussion of this problem and related computations. We will scale the pair
$(\gamma,\delta)$ to obtain a Lucas pair, 
provided $\tau(p) \ne
0$.
\begin{lem}\label{lem:associated}
Suppose $\tau(p) \ne 0$. Write $r=\ord_p(\tau(p))$ and
let
\[
\alpha=\frac{\gamma}{p^r}, \qquad \beta=\frac{\delta}{p^r}.
\]
Then $(\alpha,\beta)$ is a Lucas pair. Denoting the corresponding
Lucas sequence by $u_n$, we have
\begin{equation}\label{eqn:unform}
u_n=\frac{\tau(p^{n-1})}{p^{r(n-1)}}, \qquad n \ge 1.
\end{equation}
Moreover, $p \nmid u_n$ for all $n \ge 1$.
\end{lem}
\begin{proof}
Note that $\alpha+\beta=\tau(p)/p^{r}$ and $\alpha \beta=p^{11-2r}$
are coprime rational integers thanks to Lemma~\ref{lem:tauval}.
The identity \eqref{eqn:unform} follows immediately
from \eqref{eqn:taulucas}.
The last part is a consequence of part (i) of Theorem~\ref{thm:Carmichael}
since $p \mid \alpha \beta$.

For future reference, we note that, for $\{ u_n \}$, we have
\begin{equation} \label{disc}
D = \left( \alpha-\beta \right)^2 = p^{-2r} \left(  \tau^2 (p) - 4 p^{11} \right).
\end{equation}
\end{proof}

%\begin{prop} \label{prop:Lucas}
%%Suppose that $p$ is prime and that $\tau(p) \ne 0$. 
%\begin{enumerate}[label=(\roman*),align=left]
%\item If $\ord_p(\tau(p)) \le 4$, then $u_n$ has a primitive
%		divisor for all $n \geq 3$.
%\item If $\ord_p(\tau(p))=5$, then $u_n$ has a primitive divisor
%		for $n=5$ and $n \ge 7$.
%%\end{enumerate}
%	\margnote{Mike: shall we delete this propostion? We do not use it later.}
%Then either
%the sequence $\{ u_n \}$ defined in (\ref{eqn:unform})
%is a non-defective Lucas sequence, in that $u_n$ has a primitive divisor for every $n \geq 3$, or $\ord_p(\tau(p)) =5$ and $\{ u_n \}$ is non-defective for $n=5$ and $n > 6$.
%\end{prop}
%\begin{proof}
%From Theorem \ref{thm:BHV}, Lemma \ref{lem:tauval} and Lemma
%\ref{lem:associated}, it remains only to show that $u_n$  has, for each
%$n \in \{ 3, 4, 6 \}$, a primitive divisor, provided we assume that
%$\ord_p(\tau(p)) \leq 4$. Since $\tau (2)=-24$ and $\tau (3)=252$, we
%find that, for $p=2$,
%$$
%D=-7 \cdot 17,\quad u_2=-3,\quad u_3=-23,\quad u_4=3\cdot 5 \cdot 11,\quad %u_5=241,\quad 
%	u_6=-3^2 \cdot 23 \cdot 29
%	,u_4,u_6)=(-23,165,-6003) = (-23, 3 \cdot 5 \cdot 11, - 3^2 \cdot 23 \cdot 29), \; \mbox{ if } p=2
%$$
%and, for $p=3$,
%%$$
%\begin{gathered}
%D=-2^2 \cdot 11 \cdot 181, \quad u_2=2^2 \cdot 7, \quad u_3=-23 \cdot 61, \\
%	u_4=-2^3  \cdot 5 \cdot 7 \cdot 359, 
%u_5=253801, 
%\quad u_6=2^2 \cdot  7 \cdot 23 \cdot 53 \cdot 61 \cdot 109.
%(u_3,u_4,u_6)=(-1403,-100520,226943668) = (-23 \cdot 61, -2^3  \cdot 5 \cdot 7 \cdot 359, 2^2 \cdot  7 \cdot 23 \cdot 53 \cdot 61 \cdot 109), \; \mbox{ if } p=3.
%\end{gathered}
%$$
%We may therefore suppose that $p \geq 5$ and treat the cases $n=3$, $n=4$ and $n=6$ separately.

%Note that
%\[
%\begin{gathered}
%	D=\frac{\tau(p)^2-4 p^{11}}{p^{2r}}, \qquad
%	u_2=\frac{\tau(p)}{p^{2r}}, \qquad
%	u_3=\frac{\tau(p)^2-p^{11}}{p^{2r}}, \\
%	u_4=u_2 \cdot \left(\frac{\tau(p)^2-2p^{11}}{p^{2r}}\right),\qquad
%	u_6=u_2 \cdot u_3 \cdot \left(\frac{\tau(p)^2-3 p^{11}}{p^{2r}}\right).
%\end{gathered}
%\]
%Suppose $u_3$ does not have a primitive divisor.
%By the definition of primitive divisor, and Theorem~\ref{thm:Carmichael},
%every prime divisor of $u_3$ divides $D$. But $D-u_3=3p^{11-2r}$.
%Moreover, $p \nmid u_3$ by Lemma~\ref{lem:associated}.
%Thus the only possible prime divisor of $u_3$ is $3$.
%By \eqref{eqn:unform} we have $u_3 \mid \tau(p^2)$,
%and so $9 \nmid \tau(p^2)$ by Lemma \ref{lem:mod9}.
%It follows that
%$$
%	\left( \frac{\tau (p)}{p^r} \right)^2 -p^{11-2r}=u_3=\pm 3^\kappa, \qquad \kappa \in \{0,1\}.
%$$
%Further, since $r=\ord_p(\tau(p)) \leq 4$, we have that $11-2r \geq 3$ and hence Mihailescu's theorem (in case $\kappa=0$), a classical result of Nagell \cite{Nag} (if $\pm 3^\kappa=-3$) and Theorem 1.1 of Barros \cite{Barr} (if $\pm 3^\kappa=3$) lead us to a contradiction.
%Thus $u_3$ does have a primitive divisor.

%Next suppose that $u_4$ does not have a primitive divisor. It follows that every prime divisor of $(\tau(p)^2-2p^{11})/p^{2r}$
%	divides either $u_2$ or $D$. Moreover, $\tau(p)$ is even as $p>2$. It follows that
%Appealing further to \cite{Ab}, in case $n=4$ we have that
%$$
%u_4 =\frac{\tau(p^{3})}{p^{3r}}= \frac{\tau (p)}{p^r} \times \frac{\left(\tau^2(p)-2 p^{11}\right)}{p^{2r}}
%$$
%has a primitive divisor unless 
%$$
%\left( \frac{\tau (p)}{p^r} \right)^2 -2 p^{11-2r}=\pm 2. % \in \{ \pm 1, \pm 2 \}.
%$$
%Since $p > 2$ and $\tau (p)$ is even, we must have 
%Thus
%$$
%	p^{11-2r} = 2x^2 \pm 1, \qquad x=\frac{\tau(p)}{2p^{2r}}. %2 \left( \frac{\tau (p)}{2p^r} \right)^2 \pm 1.
%$$
%For $r \le 3$ this contradicts Theorem 1.1  of \cite{BenS}. 
%Suppose $r=4$. Then $p^3\mp 1=2x^2$. 
%We see that $(X,Y)=(2p,4x)$ is an integral point on $Y^2=X^3+8$ or $Y^2=X^3-8$.
%We determined these integral points using \texttt{Magma} \cite{magma},
%and found that the only possible solution to $p^3 \mp 1=2x^2$ with $p$ prime is
%$(p,x)=(23,78)$.
%However $23 \nmid \tau(23)$, giving a contradiction for $r=4$ as well.
%Thus $u_4$ has a primitive divisor.

%Finally, suppose $u_6$ does not have a primitive divisor.
%$$
%u_6 =\frac{\tau(p^{5})}{p^{5r}}= \frac{\tau (p)}{p^r} \times \frac{\left(\tau^2(p)-3 p^{11}\right)}{p^{2r}}
%\times \frac{\left(\tau^2(p)- p^{11}\right)}{p^{2r}}.
%$$
%Then, every prime divisor of $(\tau(p)^2-3p^{11})/p^{2r}$ divides either $D$, or $u_2$ or $u_3$.
%It follows that the only possible prime divisors of $(\tau(p)^2-3p^{11})/p^{2r}$ are $2$ and $3$.
%Since $\tau(p)$ is even, it follows that
%From \cite{Ab}, this term has a primitive divisor unless
%$$
%\left( \frac{\tau (p)}{p^r} \right)^2 -3 p^{11-2r} =\pm 3^\delta, %2^\kappa,
%$$
%where $\delta \in \{ 0, 1 \}$. After considering the reduction modulo $3$, We have that either
%and $\kappa \geq 0$. Again, the fact that $p \geq 5$ and $\tau (p)$ is even gives that $\kappa=0$, and hence, modulo $3$, that either
%$$
%\left( \frac{\tau (p)}{p^r} \right)^2 -1 =3 p^{11-2r},
%$$
%or that
%$$
%p^{11-2r} \pm 1 = 3 \left( \frac{\tau (p)}{3p^r} \right)^2.
%$$
%The latter equation again has no solutions for $r \le 3$ via Theorem 1.1  of \cite{BenS}, while routine factorization arguments,
%allow us show that the former equation has no solutions, and likewise the latter with $r=4$.
%\end{proof}


%----------------------------------------------------------------
\section{On three sequences of polynomials}\label{sec:Fn}
%---------------------------------------------------------------
%\margnote{Mike: 
%I've shifted the indices of the $F_n$ so that they're multiplicative:
%$m \mid n$ implies $F_m \mid F_n$. 
%It was getting too confusing before this this.}
We begin by defining, for $m \ge 0$, a sequence of polynomials $H_m(Z,W) \in \Z[Z,W]$
\begin{equation}\label{eqn:Hndef}
	H_m(Z,W)=\begin{cases}
		(Z^m-W^m)/(Z-W) \qquad \text{if $m$ is odd}\\
		(Z^m-W^m)/(Z^2-W^2) \qquad \text{if $m$ is even}.
		\end{cases}
\end{equation}
Let $G$ be the group generated the involutions $\kappa_1$ and $\kappa_2$ on $\Z[Z,W]$ given by
\[
\kappa_1 \; : \; 
\begin{cases}
Z \mapsto -Z, \\
W \mapsto -W,
\end{cases}
\qquad 
\kappa_2 \; : \; 
\begin{cases}
Z \mapsto W, \\ 
W \mapsto Z.
\end{cases}
\]
We compute the subring of invariants $\Z[Z,W]^G$.
\begin{lem}
$\Z[Z,W]^G=\Z[ZW,(Z+W)^2]$.
\end{lem}
\begin{proof}
It is clear that $\Z[ZW,(Z+W)^2]$ belongs to the ring of invariants.
Let $F \in \Z[Z,W]$ belong to the ring of invariants. We would
like to show that $F \in \Z[ZW,(Z+W)^2]$. Observe that $\kappa_1$ and $\kappa_2$
send monomials to monomials and preserve the degree.
Thus every homogenous component of $F$ belongs to the ring of invariants,
and we may suppose that $F$ is homogeneous. As $F$ is invariant
under $\kappa_1$ it has even degree, $2n$ say, and we may write
\[
	F=\sum_{k=0}^{2n} a_k Z^{2n} W^{2n-k}.
\]
As $F$ is invariant under $\kappa_2$ we have $a_k=a_{2n-k}$ for $k=0,\dotsc,n$.
Thus
\[
	F=a_0(Z^{2n}+W^{2n})+a_1 (ZW) (Z^{2n-2}+W^{2n-2})+a_2 (ZW)^2 (Z^{2n-4}+W^{2n-4})+\cdots+a_n (ZW)^n.
\]
	To complete the proof all we need to show is that $Z^{2n}+W^{2n} \in \Z[ZW,(Z+W)^2]$ for all $n$.
This follows from an easy induction using the identity
\[
	Z^{2n}+W^{2n} \; =\; ((Z+W)^2-2ZW) \cdot (Z^{2n-2}+W^{2n-2}) \, - \, (ZW)^2 \cdot (Z^{2n-4}+W^{2n-4}).
\]
\end{proof}
Note that the $H_m(Z,W)$ are invariant under $\kappa_1$, $\kappa_2$ and so belongs to the invariant ring $\Z[ZW,(Z+W)^2]$.
It follows that there is a sequence of polynomials $F_m(X,Y) \in \Z[X,Y]$ such that
\begin{equation}\label{eqn:FnHn}
	F_m(ZW,(Z+W)^2)=H_m(Z,W).
\end{equation}
The following lemma aids in the computation of the $F_m$.
\begin{lem}
The sequence $F_m(X,Y) \in \Z[X,Y]$ satisfies	
\[
F_0=0, \quad F_1=F_2=1, \quad F_3=-X+Y, 
\]
and the recurrence
\begin{equation} \label{eqn:Fnrecurrence}
	F_{m+2}(X,Y)=(-2X+Y)\cdot F_m(X,Y)-X^2\cdot F_{m-2}(X,Y), \qquad \text{for $m \ge 2$}. 
\end{equation}
\end{lem}
\begin{proof}
	Since $H_0=0$ and $H_1=H_2=1$ we have $F_0=0$ and $F_1=F_2=1$. Moreover, $H_3=Z^2+ZW+W^2=-ZW+(Z+W)^2$
	so $F_3=-X+Y$.
The map 
\[
	\Z[X,Y] \rightarrow \Z[ZW,(Z+W)^2], \qquad X \mapsto ZW, \quad Y \mapsto (Z+W)^2
\]
	is an isomorphism of rings that sends $F_m(X,Y)$ to $H_m(Z,W)$. Applying this isomorphism to  \eqref{eqn:Fnrecurrence}
	gives
\[
	H_{m+2}(Z,W)=(Z^2+W^2) H_m (Z,W)-(ZW)^2 H_{m-2}(Z,W)
\]
and so it is enough to prove this identity. However this identity easily follows from the definition of $H_m$
	in \eqref{eqn:Hndef}.
\end{proof}

\begin{lem}
If $m$ and $n$ are positive integers, then
\begin{enumerate}[label=(\roman*)]
	\item $F_m$ is homogeneous of degree $\lfloor (m-1)/2 \rfloor$.
	\hskip12ex \item $F_n \mid F_m$ whenever $n \mid m$.
\end{enumerate}		
\end{lem}
\begin{proof}
These follow immediately from the corresponding properties for the $H_m$.
\end{proof}


\begin{lem}\label{lem:split}
Let $m \ge 3$. Then
\begin{equation}\label{eqn:FmFactor}
	F_m(X,Y) = \prod_{j=1}^{\lfloor (m-1)/2 \rfloor} (Y-4\cos^2(\pi j/m) X)
\end{equation}
\end{lem}
\begin{proof}
Fix $m \ge 3$ and write
$\zeta=\exp(2\pi i/m)$.
Note that
\[
\begin{split}
	H_m(Z,W) &=\prod_{j=1}^{\lfloor (m-1)/2 \rfloor} (Z-\zeta^j W)(Z-\zeta^{-j} W)\\
		&=\prod_{j=1}^{\lfloor (m-1)/2 \rfloor} (Z^2+W^2-(\zeta^j+\zeta^{-j}) ZW)\\
		&=\prod_{j=1}^{\lfloor (m-1)/2 \rfloor} ((Z+W)^2-(\zeta^j+\zeta^{-j}+2) ZW)\\
		&=\prod_{j=1}^{\lfloor (m-1)/2 \rfloor} ((Z+W)^2-(2+2\cos{2\pi j/m}) ZW)\\
		&=\prod_{j=1}^{\lfloor (m-1)/2 \rfloor} ((Z+W)^2- 4 \cos^2(\pi j/m) ZW).\\
\end{split}
\]
The lemma follows.
\end{proof}

Next we define
\begin{equation}\label{eqn:PsimFactor}
	\Psi_m(X,Y)=\prod_{\stackrel{j=1}{(j,m)=1}}^{\lfloor (m-1)/2 \rfloor} (Y-4 \cos^2(\pi j/m) X).
\end{equation}
Note that $\Psi_m(X,Y) \in \Z[X,Y]$. Indeed,
\[
	\Psi_m(X,Y)=\frac{F_m(X,Y)}{\LCM\{F_n(X,Y) \: : \; n \mid m,~ n<m\}}.
\]
It follows that $\Psi_m(X,Y) \mid \Psi_n(X,Y)$ (with the divisibility being valid
in $\Z[X,Y]$) whenever $m \mid n$.
From \eqref{eqn:FmFactor} and \eqref{eqn:PsimFactor}, we see that
\[
	F_m(X,Y)=\prod_{d \mid m} \Psi_d(X,Y).
\]
We deduce that
\begin{equation}\label{eqn:inversion}
	\Psi_m(X,Y)=\prod_{d \mid m} F_d(X,Y)^{\mu(m/d)}
\end{equation}
where $\mu$ denotes the M\"{o}bius function.
\begin{lem}\label{lem:abelian}
Let $m \ge 3$ and write $\zeta_m=\exp(2\pi i/m)$.
	The polynomial $\Psi_m(1,Y)$ is monic and irreducible
	of degree $\phi(m)/2$. It is a defining polynomial
	for the abelian extension $\Q(\zeta_m)^+/\Q$.
\end{lem}
\begin{proof}
Recall that $\Q(\zeta_m)^+=\Q(\zeta_m+\zeta_m^{-1})$. 
The elements of the Galois group for $\Q(\zeta_m)/\Q$ 
are the automorphisms $\sigma_j : \zeta_m \mapsto \zeta_m^j$
with $\gcd(j,m)=1$.
Therefore Galois conjugates
of $\zeta_m+\zeta_m^{-1}+2$ are precisely $\zeta_m^j+\zeta_m^{-j}+2$ with $\gcd (j,m)=1$,
and these are the roots of $\Psi_m(1,Y)$.
The lemma follows.
%The roots of 
%Let 
%\[
%	\Phi_m(Z,W)=\prod_{\stackrel{j=1}{(j,m)=1}}^{m-1} (Z-\zeta^j W).
%\]
%Then $\Phi_m(Z,1)$ is the $m$-th cyclotomic polynomial, which is a defining
%polynomial for $\Q(\zeta_m)$. 
\end{proof}
We shall need the following weak bound for the coefficients of $\Psi_m$.
\begin{lem}\label{lem:Hbound}
	The coefficients of $\Psi_m$ are bounded in absolute value by $5^{\phi(m)/2}$.
\end{lem}
\begin{proof}
The roots of the monic polynomial $\Psi_m(1,Y)$ are bounded in
absolute value by $4$. Writing $\Psi_m(1,Y)=\sum a_i Y^i$
	and $(4+Y)^{\phi(m)/2}=\sum b_i Y^i$
	we have $\lvert a_i \rvert \le b_i$.
Thus 
$$
\sum \lvert a_i \rvert \le \sum b_i =5^{\phi(m)/2}.
$$
\end{proof}
\begin{lem}\label{lem:hR}
Let $m \ge 3$ and write $\zeta_m=\exp(2\pi i/m)$.
%The homogenous polynomials $\Psi_m(X,Y)$ splits completely
%into linear factors over $K_m=\Q(\zeta_m)^+$.
Write $h_m$ and $R_m$ for the class
number and regulator of $K_m$. Then as $m \rightarrow \infty$,
\[
	\log(h_m)=O(m\log{m}), \qquad \log(h_m R_m)=O(m \log{m})
\]
where the implicit constants are absolute and effective.
\end{lem}
\begin{proof}
%The first part of is clear from the definition of $\Psi_m$.
Write $d=\phi(m)/2$ for the degree of $K_m$.
By \cite[Propostion 2.7]{Wa} and \cite[Lemma 4.19]{Wa},
\[
	\log(\lvert \Disc(K_m)\rvert) \; \le\;
	\frac{1}{2} \log (\Disc(\Q(\zeta_m))) 
	\; \le \;
	\frac{1}{2}\phi(m) \log(m) .
\]
A theorem of Lenstra \cite[Theorem 6.5]{Lenstra} asserts that
for a number field $K$ of degree $d \ge 2$, signature $(r,s)$,
absolute discriminant $D$, class number $h$ and regulator $R$,
\[
	h \; \le \; \frac{1}{(d-1)!} \cdot \Delta \cdot 
        (d-1+\log{\Delta})^s,
        \qquad \Delta=(2/\pi)^s \cdot D^{1/2}.
\]
and
\[
	hR \; \le \; \frac{1}{(d-1)!} \cdot \Delta \cdot 
	(d-1+\log{\Delta})^s \cdot
	(\log{\Delta})^{d-1-s} . 
\]
We take $K=K_m$, so $d=\phi(m)/2$, $s=0$,
and $\Delta=\lvert \Disc(K_m) \rvert^{1/2}$.
The lemma follows.
\end{proof}
We can also deduce the bound $\log(h_m R_m)=O(m \log{m})$
from the Brauer--Siegel theorem, at the cost of introducing ineffectivity.

\begin{lem}\label{lem:tauFm}
Let $p$ be a prime. Then, for $m \ge 1$,
\[
	\tau(p^{m-1})=\tau(p)^\varepsilon \cdot F_m(p^{11},\tau(p)^2), \qquad
	\varepsilon=\begin{cases}
		0 & \text{if $m$ is odd}\\
		1 & \text{if $m$ is even}.
	\end{cases}
\]
In particular, $\Psi_m(p^{11},\tau(p)^2) \mid \tau(p^{m-1})$.
\end{lem}
\begin{proof}
From \eqref{eqn:taulucas}
\[
\tau(p^{m-1})=
\frac{{\gamma}^{m}-{\delta}^m}{\gamma-\delta}=\begin{cases}
	H_m(\gamma,\delta) & \text{if $m$ is odd}\\
	(\gamma+\delta) H_m(\gamma,\delta) & \text{if $m$ is even}, 
\end{cases}
\]
where $\gamma+\delta=\tau(p)$ and $\gamma\delta= p^{11}$.
The lemma follows from \eqref{eqn:FnHn}.
\begin{comment}
Note that
\[
\begin{split}
	(\tau(p)^2-2p^{11}) \tau(p^{m-1}) & =({\alpha^\prime}^2+{\beta^\prime}^2) \cdot \left(\frac{{\alpha^\prime}^{m}-{\beta^\prime}^m}{\alpha^\prime-\beta^\prime}\right)\\
	&=\frac{{\alpha^\prime}^{m+2}-{\beta^\prime}^{m+2} \;+\; (\alpha\beta)^2 ({\alpha^\prime}^{m-2}-{\beta^\prime}^{m-2})}{\alpha^\prime-\beta^\prime}\\
	&=\tau(p^{m+1})+p^{22} \tau(p^{m-3}).
\end{split}
\]
Thus
\[
	\tau(p^{m+1})=(-2 p^{11}+\tau(p)^2) \tau(p^{m-1})-p^{22} \tau(p^{m-3}).
\]
The lemma follows from this identity by induction.
\end{comment}
\end{proof}



\medskip


\begin{lem}\label{lem:ramification}
Let $m=5$ or $m \ge 7$. Then precisely the same primes
ramify in $L_m=\Q(\zeta_m)$
as in $K_m=\Q(\zeta_m)^+$.
\end{lem}
\begin{proof}
Note that for $m \in \{2, 3, 4, 6\}$, we have $K_m=\Q$
so the conclusion of the lemma is false in those cases.

By the proof of Proposition 2.15 of \cite{Wa}, we know that
$L_m/K_m$ is unramified if $m$ is divisible by at least two distinct
odd primes, or divisible by $4$ and an odd prime. 
We may therefore suppose that $m \in \{ 2^a, p^a, 2 p^a\}$ where
$p$ is an odd prime and $a$ is a positive integer. If $m=2^a$ with $a \ge 3$, then
$K_m$ has degree $2^{a-2}>1$, and the set of ramified
primes for both $L_m$ and $K_m$ is $\{2\}$. 
Let $p$ be an odd prime and $a \ge 1$.
Then $L_{2p^{a}}=L_{p^{a}}$ and $K_{2 p^a}=K_{p^a}$.
Now the set of ramified primes for $L_{p^{a}}$
and $K_{p^a}$ is just $\{p\}$ as long
as the degree $\phi(p^a)/2$ of $K_{p^{a}}$
is $>1$. The lemma follows.
\end{proof}


\begin{lem}\label{lem:preThue}
Let $m =5$ or $m \ge 7$. Let $x$ and $y$ be coprime integers, and $q$
be a prime. Suppose $q^a \mid\mid \Psi_m(x,y)$ with $a \ge 1$ an integer. Then
either $q \equiv \pm 1 \pmod{m}$ or $q^a \mid m$. 
\end{lem}
\begin{proof}
Write $L_m=\Q(\zeta_m)$. Recall the isomorphism
\[
	(\Z/m\Z)^* \cong \Gal(L_m/\Q), \qquad j \; \mapsto\;  
	(\sigma_j \; : \; \zeta_m \mapsto \zeta_m^{j}).
\]
The subfield $K_m=\Q(\zeta_m+\zeta_m^{-1})$ is the fixed
	field for $\langle \sigma_{-1} \rangle=\{\sigma_1,\sigma_{-1}\}$.
Let $q$ be a rational prime that does not ramify in $K_m$
(and hence in $L_m$ by Lemma~\ref{lem:ramification}). 
The Frobenius automorphism
for $q$ is simply $\sigma_q$. The prime $q$ splits completely
in $K_m$ if and only if the restriction of $\sigma_q$ to $K_m$
is trivial. This is equivalent to 
	$\sigma_q \in \{\sigma_1,\sigma_{-1}\}$
and therefore equivalent to
$q \equiv \pm 1 \pmod{m}$.

Let $\lambda=\zeta_m+\zeta_m^{-1}+2$. This is a root
for $\Psi_m(1,Y)$ and also a generator for $K_m$. 
Note \cite[Proposition 2.16]{Wa} that $\OO_{K_m}=\Z[\lambda]$.
We are given that $q \mid \Psi_m(x,y)$.
Since $\Psi_m(1,Y)$ is monic, if $q \mid x$ then $q \mid y$
giving a contradiction. Hence $q \nmid x$ and so
$\Psi_m(1,y/x) \equiv 0 \pmod{q}$.
By the Dedekind-Kummer Theorem, there is a degree $1$ prime
ideal $\fq$ above $q$. As $K_m/\Q$ is Galois, all primes
above $q$ must therefore have degree $1$. Thus $q$ is either totally
split or ramified in $K_m$. 
If $q$ is totally split, then $q \equiv \pm 1 \pmod{m}$
and we are finished.

We shall therefore suppose that $q$ is ramified in $K_m$.
Let $\fq_1,\dotsc,\fq_r$ be the prime ideals of $\OO_{K_m}$
above $q$. Write $G$ for the Galois group of $\Gal(K_m/Q)$,
and let $I$ be the inertia subgroup for $\fq_1$. 
As $G$ is abelian, $I$ is also the inertia subgroup
for all $\fq_i$. Thus $\fq_i^\sigma=\fq_i$
for all $\sigma \in I$ and for $i=1,\dots,r$.
	Since $q$ is ramified, $I \ne 1$. Fix $\sigma_j \in \Gal(L_m/\Q)$
whose restriction to $K_m$ is a non-trivial element of $I$.
Thus $\gcd (j,m)=1$ and $j \not \equiv \pm 1 \pmod{m}$.
Write $\lambda_j=\zeta_m^{j}+\zeta_m^{-j}+2=\sigma_j(\lambda)$.

We factor the ideal $(y-\lambda x)\OO_{K_m}$
as
\begin{equation}\label{eqn:fafb}
	(y-\lambda x)\OO_{K_m}=\fa \fb
\end{equation}
where $\fa$, $\fb$ are ideals with $\fa$ supported
on $\fq_1,\dotsc,\fq_r$, and $\fb$ not divisible
by $\fq_1,\dotsc,\fq_r$. 
By assumption $q^a \mid\mid \Psi_m(x,y)$.
However $\Psi_m(x,y)=\Norm_{K_m/\Q}(y-\lambda x)$
and thus $\Norm_{K_m/\Q}(\fa)=q^a$.
Note that any ideal dividing both $x$ and $\fa$
must also divide $y$ by \eqref{eqn:fafb}.
As $x$, $y$ are coprime, we deduce that $x$ and $\fa$
are coprime.


Since $\fq^{\sigma_j}=\fq$ for all $\fq \mid \fa$, 
we have $\fa^{\sigma_j}=\fa$.  
Hence $\fa$ divides
\[
	(y-\lambda^{\sigma_j} x) - (y - \lambda x)= (\lambda-\lambda^{\sigma_j})x. 
\]
Thus $\fa$ divides
\[
	\lambda-\lambda^{\sigma_j}=
	(\zeta_m+\zeta_m^{-1})-(\zeta_m^{j}+\zeta_m^{-j})
	=\zeta_m^{-1} (\zeta_m^{j+1}-1) (\zeta_m^{-j+1}-1)
\]
and it follows that $q^{2a}=\Norm_{K_m/K}(\fa)^2$ divides
\[
	\begin{split}
		(\Norm_{K_m/\Q}(\lambda-\lambda^{\sigma_j}))^2 &=
	\Norm_{L_m/\Q}(\lambda-\lambda^{\sigma_j})\\
		&=
		\Norm_{L_m/\Q}(\zeta_m^{j+1}-1) \cdot
		\Norm_{L_m/\Q}(\zeta_m^{-j+1}-1). \\
	\end{split}
\]
This divides $\prod_{i=1}^{m-1} (\zeta_m^i-1)^2=m^2$.
Hence $q^a \mid m$ as required.
\end{proof}

\begin{lem}\label{lem:psiDiv}
Let $p$ be a prime and suppose $\tau(p)=0$.
Let $r=\ord_p(\tau(p))$ and write
\[
	x=p^{11-2r} \; \mbox{ and } \;  y=\frac{\tau(p)^2}{p^{2r}}.
\]
Let $\{u_m\}$ be the Lucas sequence defined in Lemma~\ref{lem:associated}. Then,
for $m \ge 3$,
\begin{equation}\label{eqn:psiu}
	\Psi_m(x,y)= \prod_{d \mid m} u_d^{\mu(m/d)}.
\end{equation}
Moreover, if $m =5$ or $m \ge 7$, then $\Psi_m(x,y)$
is divisible by some prime $\ell \nmid m$.
\end{lem}
\begin{proof}
Note that $x$ and $y$ are in fact integers by Lemma~\ref{lem:tauval},
and are coprime by the definition of $r$.
Let $\{u_m\}$ be the Lucas sequence defined in Lemma~\ref{lem:associated}.
Thus
\[
u_m=\frac{\tau(p^{m-1})}{p^{r(m-1)}}=\frac{\alpha^m-\beta^m}{\alpha-\beta}, 
\qquad \alpha \beta=p^{11-2r}, \qquad \alpha+\beta=\tau(p)/p^r.
\]
Write
\[
\varepsilon(m)=\begin{cases}
	0 & \text{if $m$ is odd}\\
	1 & \text{if $m$ is even}.
\end{cases}
\]
Then
\[
\begin{split}
	F_m(x,y) &=\frac{1}{p^{2 r \deg(F_m)}} \cdot F_m(p^{11}, \tau(p)^2)\\
		& =\frac{\tau(p^{m-1})}{p^{2 r \deg(F_m) } \cdot \tau(p)^{\varepsilon(m)}}
	\qquad \text{(by Lemma~\ref{lem:tauFm})}
		\\ 
		&=\frac{p^{r(m-1)} \cdot u_m}{p^{2 r \deg(F_m) } \cdot \tau(p)^{\varepsilon(m)}} .
	\end{split}
\]
However, since $\deg(F_m)=\lfloor (m-1)/2 \rfloor$, it follows that
\[
	F_m(x,y)=\left( \frac{p^r}{\tau(p)} \right)^{\varepsilon(m)} \cdot u_m.
\]
By \eqref{eqn:inversion},
\[
	\Psi_m(x,y)=\left( \frac{p^r}{\tau(p)} \right)^{\sum_{d \mid m} \varepsilon(d) \mu(m/d)} \cdot \prod_{d \mid m} u_d^{\mu(m/d)}.
\]
It is easy to see that 
\[
	\sum_{d \mid m} \varepsilon(d) \mu(m/d)=\begin{cases}
		0 & \text{if $m \ne 2$}\\
		1 & \text{if $m=2$}.
	\end{cases}
\]
This completes the proof of \eqref{eqn:psiu}.

Now let $m=5$ or $m \ge 7$. By Theorem~\ref{thm:BHV}, the term $u_m$
has a prime divisor $\ell$ that does not divide $(\alpha-\beta)^2$
nor $u_1 u_2\cdots u_{m-1}$. 
%In particular $\ell \nmid \tau(p)$ as $\tau(p)=p^r \cdot u_2$.
By Theorem~\ref{thm:Carmichael},
we know that $\ell \ne p$, that $m=m_\ell$,
and that $m\mid (\ell-1)$
or $m \mid (\ell+1)$. In particular, $\ell \nmid m$.
From \eqref{eqn:psiu}, we have $\ell \mid \Psi_m(x,y)$
as required.
\end{proof}

\section{Proof of Theorem~\ref{thm:main2}}\label{sec:Bugeaud}
We shall need the following theorem \cite[Theorem 1]{Bug-BLMS}.
\begin{thm}[Bugeaud]
Let $K$ be a number field. Let $u \ge 2$ and $v \ge 3$ be integers, and 
let $a$, $b\in \OO_K\setminus \{0\}$.
%Let $a$, $b$ be non-zero integers, and let $u\ge 3$ and $v \ge 2$ be integers.
There exist effectively computable positive constants $\varepsilon_1$, $\varepsilon_2$ depending only on $a$, $b$, $u$, $v$ and $K$
such that every pair of coprime $x$, $y \in \OO_K$ with 
\[
	\max\{\lvert \Norm_{K/\Q}(x) \rvert,~\lvert \Norm_{K/\Q}(y) \rvert\} \; > \; \varepsilon_1
\]
satisfy
\[
	P(a x^u+by^v) \; > \; \varepsilon_2 \cdot \log\log \max\{ \lvert \Norm_{K/\Q}(x)\rvert,~\lvert \Norm_{K/\Q}(y) \rvert\}.
\]
\end{thm}
In the above theorem, $P(\delta)$ for $\delta \in \OO_K$ denotes the largest rational prime
that is below a prime ideal dividing $\delta$.

\bigskip

We now prove Theorem~\ref{thm:main2}.
Let $p$ be a prime and suppose $\tau(p) \ne 0$.
Let $m \ge 3$. We want to show that
\[
	P(\tau(p^{m-1})) \; \gg_{m} \; \log\log{p}.
\]
%We shall rewrite \eqref{eqn:precise} as follows:
%\[
%	P(\tau(p^{m-1})) \; > \; \eta_2 \log\log(p) \qquad \text{for $m \in \{3,4,5,6,8,10,12\}$}.
%\]
%The values  $m\in \{3,4,5,6,8,10,12\}$ are precisely those values for which $\Psi_m$
%is linear or quadratic.
Let $r=\ord_p(\tau(p))$ and recall that $\ord_p(\tau(p^{m-1}))=r(m-1)$ by Lemma~\ref{lem:tauval}.
If $r \ge 1$ then
\[
	P(\tau(p^{m-1})) \; \ge \; p,
\]
whereby we may suppose that $r=0$. 
Recall that $\Psi_m(p^{11},\tau(p)^2) \mid \tau(p^{m-1})$ by Lemma~\ref{lem:tauFm}.
Let $K=K_m=\Q(\zeta_m)^+$ and let $\lambda=\zeta_m+\zeta_m^{-1}+2$
which is a root of the monic polynomial $\Psi_m(1,Y)$. 
Then 
\[
	\Psi_m(p^{11},\tau(p)^2) \; =\; \Norm_{K/\Q}(\tau(p)^2- \lambda \cdot p^{11})
\]
and therefore 
\[
	P(\tau(p^{m-1})) \; \ge \; P(\Psi_m(p^{11},\tau(p)^2) ) \; = \; 
	P(\tau(p)^2-\lambda \cdot p^{11}).
\]
We now apply Bugeaud's theorem with $u=2$, $v=11$, $a=1$, $b=-\lambda$,
$x=\tau(p)$, $y=p$ to deduce that $P(\tau(p)^2- \lambda \cdot p^{11}) \; \gg_m \; \log\log{p}$. 
This completes the proof.
\begin{comment}
Thus it will be sufficient to show that $P(\Psi(p^{11},\tau(p)^2) \gg \log\log(p^m)$.
Consider first $m \in \{3,4,6\}$. Then $\Psi_m(p^{11},\tau(p)^2)$ is respectively
\[
	\tau(p)^2-p^{11}, \qquad 2 \tau(p)^2-p^{11}, \qquad 3\tau(p)^2-p^{11}.
\]
The desired inequality follows from Bugeaud's theorem on letting $x=\tau(p)$, $y=p$, $u=2$, $v=11$,
$a=1$ and $b=-1$.

Next we consider $m \in \{5,8,10,12\}$. Then $\Psi_m(X,Y)$ is respectively,
\[
	 Y^2 - 3XY + X^2, \qquad
 Y^2 - 4XY + 2X^2, \qquad
 Y^2 - 5XY + 5X^2, \qquad
 Y^2 - 4XY + X^2.
\]
Completing the square in each instance and applying Bugeaud's theorem
gives the desired inequality. For example,
\[
	4 \Psi_5(p^{11},\tau(p)^2)=(2\tau(p)^2-3p^{11})^2-5p^{22},
\]
and we simply apply Bugeaud's theorem with $x=2\tau(p)^2-3p^{11}$, $y=p$,
$u=2$, $v=22$, $a=1$, $b=-4$.
\end{comment}



\begin{comment}
%--------------------------------------------------------------------
\section{$S$-Unit Equations}\label{sec:sunit}
%--------------------------------------------------------------------
We shall make use of a theorem of Gy\H{o}ry and Yu \cite{GY}
giving bounds for solutions to an $S$-unit equation. 
Let $K$ be a number field of degree $d$, class number $h$, regulator $R$.
Let $S_\infty$ be the set of infinite places
of $K$, and let $S=S_\infty \cup \{\fp_1,\fp_2,\dotsc,\fp_t\}$
where the $\fp_i$ are distinct prime ideals of $\OO_K$.
Let $s=\#S=\#S_\infty + t$. Let 
\[
	P_S=\begin{cases}
		\max \Norm(\fp_i) & \text{if $t \ge 1$}\\
		1 & \text{if $t=0$}.
	\end{cases}
\]
Write $\OO_S^*$ for the group of $S$-units of $K$. 
Let $R_S$ be the regulator of $\OO_S^*$; we note the following
standard bound \cite[Proof of Lemma 3]{BG-sunit} for $R_S$
\begin{equation}\label{eqn:RSbound}
	R_S \; \le  h \cdot R \cdot (\log^*{P_S})^t. 
\end{equation}

For $\alpha \in K$, we denote by $H(\alpha)$ the absolute height of $\alpha$:
\[
	H(\alpha)=\left( \prod_{\upsilon \in M_K} \max\{1,\lvert \alpha \rvert_\upsilon\}^{d_\upsilon} \right)^1/d
\]
where $M_K$ denotes the set of places of $K$ and $d_\upsilon$ the degree of $K_{\upsilon}$.
We write $h(\alpha)=\log(H(\alpha))$ for the logarithmic height of $\alpha$.

Let $\alpha$, $\beta \in K^*$ and consider the $S$-unit equation
\begin{equation}\label{eqn:sunit}
	\alpha u + \beta v=1, \qquad u,~v \in \OO_S^*.
\end{equation}
The following is \cite[Theorem 1]{GY}.
\begin{thm}[Gy\H{o}ry and Yu]
All solutions to \eqref{eqn:sunit} satisfy
\[
	\max\{h(u),h(v)\} \; \le 
	c(s,d) \cdot P_S \cdot R_S \cdot  \left( 1+ \frac{\log^*{R_S}}{\log^*{P_S}} \right)  \cdot \max\{h(\alpha),h(\beta),1\},
\]
where 
\[
	c(d,s)\; = \;
	s^{2s+3.5} \cdot 2^{7s+27} \cdot \log(2s) \cdot d^{2(s+1)} \cdot
	(\log^*(2d))^3.
\]
\end{thm}
\end{comment}


%--------------------------------------------------------------------
\section{Proof of Theorem \ref{thm:main}}\label{sec:thm1a}
%--------------------------------------------------------------------
In this section, we prove Theorem~\ref{thm:main}.
For this we appeal to 
a result of Bugeaud and Gy\H{o}ry \cite{BuGy}
which provides bounds for solutions to Thue--Mahler
equations.
Let $F(X,Y) \in \Z[X,Y]$ be an irreducible binary form of degree $n \ge 3$,
and let $b$ a non-zero rational integer with absolute value
at most $B\ge e$. 
Let $H \ge 3$ be an upper bound for the absolute values
of the coefficients of $F$. 

Let $\alpha_1$, $\alpha_2$ and $\alpha_3$ be three distinct roots of $F(1,Y)$.
Define
\[
	\mathbb{M}=\Q(\alpha_1), \; \;  \mathbb{M}_{123}=\Q(\alpha_1,\alpha_2,\alpha_3) \; \;
	\mbox{ and } \; \; N=[\mathbb{M}_{123} : \Q].
\]
Write $h_{\mathbb{M}}$ for the class number of $\mathbb{M}$ and $R_{\mathbb{M}}$ for its regulator.
Let $p_1,p_2,\dotsc,p_s$ ($s>0$) be distinct primes not exceeding $P$. 
Consider the \textbf{Thue--Mahler equation}
\begin{equation}\label{eqn:TM}
	F(x,y) \; =\; b \cdot p_1^{z_1} p_2^{z_2} \cdots p_s^{z_s}, \qquad x,~y,~z_i \in \Z, \qquad \gcd(x,y,p_1p_2\cdots p_s)=1.
\end{equation}
For a positive real number $a$, we write $\log^*{a}=\max\{1,\log{a}\}$.
\begin{thm}[Bugeaud and Gy\H{o}ry]\label{thm:BG}
All solutions to \eqref{eqn:TM} satisfy
\begin{multline*}
	\log \max\{ \lvert x \rvert,~\lvert y \rvert,~ p_1^{z_1} \cdots p_k^{z_k}\} \; \le  
	\\
	c(n,s) \cdot P^N \cdot (\log{P})^{ns+2} \cdot R_{\mathbb{M}} h_{\mathbb{M}} \cdot (\log^*(R_{\mathbb{M}} h_{\mathbb{M}}))^2
	\cdot (R_\mathbb{M}+s h_\mathbb{M}+\log(HB)),
\end{multline*}
where
\[
	c(n,s)=3^{n(2s+1)+27} \cdot n^{2n (7s+13)+13} \cdot (s+1)^{5n(s+1)+15}.
\]
\end{thm}
%\margnote{Is this bit about $N$ correct?}
The theorem as stated is the first part of Theorem 4 in \cite{BuGy}, with only one
minor difference. In \cite{BuGy} the authors take $N=n(n-1)(n-2)$.
However, in their proof $N$ is simply taken as an upper bound for the 
degree $[\mathbb{M}_{123} : \Q]$, and so we can take $N=[\mathbb{M}_{123}:\Q]$.

\bigskip

We now embark on the proof of Theorem~\ref{thm:main}.
In what follows $\eta_2,\eta_3,\dots$ will denote absolute
effectively computable positive constants.
Let us fix a prime $p$ and suppose $\tau(p) \neq 0$.
We will in fact show that
\begin{equation}\label{eqn:show}
	P(\tau(p^{m-1})) \; \ge \; \eta_2 \cdot \frac{\log\log(p^m)}{\log\log\log(p^m)},
\end{equation}
for $m \ge 3$ which implies \eqref{eqn:required}.
In view of Theorem~\ref{thm:main2} (which was proved in Section~\ref{sec:Bugeaud}),
we shall suppose that $m=7$, $9$, $11$ or $m \ge 13$.
In particular, $\Psi_m(X,Y)$ is irreducible
of degree $\phi(m)/2 \ge 3$. 
Write $r=\ord_p(\tau(p))$. 
By Lemma~\ref{lem:associated}, we have $\ord_p(\tau(p^{m-1}))=r(m-1)$.
Recall that $r=\ord_p(\tau(p)) \le 5$ by Lemma~\ref{lem:tauval}.
Let 
\[
	x=p^{11-2r}, \qquad y=\tau(p)^2/p^{2r},
\]
and observe that $\gcd(x,y)=1$.
By Lemma~\ref{lem:tauFm}, we know that $\Psi_m(x,y)$
is a divisor of $\tau(p^{m-1})$ and therefore
\[
	P(\tau(p^{m-1})) \; \ge \; P(\Psi_m(x,y)) 
\]
To prove \eqref{eqn:show}, we shall show that
\begin{equation}\label{eqn:desired}
	P(\Psi_m(x,y)) \; > \; \eta_3 \cdot \frac{\log\log(p^m)}{\log\log\log(p^m)}.
\end{equation}
By Lemma~\ref{lem:preThue}, we can write
\begin{equation}\label{eqn:PsiTM}
	\Psi_m(x,y)\; =\; b \cdot p_1^{z_1} p_2^{z_2} \cdots p_s^{z_s}, 
\end{equation}
where the $p_i$ are primes, and
\[
	b \mid m, \; \;  p_i \equiv \pm 1 \pmod{m} \; \; \mbox{ and } \; \; p_1<p_2<\cdots<p_s.
\]
From Lemma~\ref{lem:psiDiv}, we have $s \ge 1$.
It is clear that 
\[
	s \; < \; \eta_4 \cdot \frac{p_s}{m}.
\]
In what follows we make use of the following inequalities
\[
n \; < \; m \; \; \mbox{ and } \; \; 	ns \; < \; ms \; < \; \eta_4 \cdot p_s.
\]
Moreover, since $p_s \equiv \pm 1 \pmod{m}$, we have
\[
	p_s \; \ge \;  m-1.
\]
We will apply Theorem~\ref{thm:BG} to \eqref{eqn:PsiTM}. We take
\[
	F=\Psi_m, \; \;  B=m, \; \;  P=p_s, \; \;  n=N=\phi(m)/2 \; \; \mbox{ and } \; \; H=5^{n/2}.
\]
where the choice of $H$ is justified by Lemma~\ref{lem:Hbound}.
%Since $b \mid m$ we can take $B=m$.
%Let $\alpha_1=\zeta_m+\zeta_m^{-1}+2$ which is a root
%of $F(1,Y)$, and $\mathbb{M}=K_m=\Q(\alpha)$. This is a Galois extension,
%so for any choices of $\alpha_2$, $\alpha_3$ we have $\mathbb{M}_{123}=\mathbb{M}$.
%Thus in the notation of Theorem~\ref{thm:BG}, we have
%$n=N=\phi(m)/2$.
By Lemma~\ref{lem:hR},
we have 
\[
	\log(h_{\mathbb{M}}) \; < \; \eta_5 \cdot m \log{m} \; \; \mbox{ and } \; \; 	\log(h_{\mathbb{M}} R_{\mathbb{M}}) \; < \; \eta_6 \cdot m \log{m}.
\]
Since $x=p^{11-2r}$ with $r \le 5$, 
we have $\log\log{p} \le \log\log{x}$.
Taking logarithms in Theorem~\ref{thm:BG}, and making repeated use of the above
inequalities and bounds, yields
\[
	\log\log{p}  \; < \; \eta_7 \cdot p_s \cdot \log{p_s}. 
\]
But
\[
	\log\log(p^m) \; =
	\;
	\log\log{p}+\log{m}
	\; <\; \eta_7 \cdot p_s \cdot \log{p_s}+\log(p_s+1) \; < \; \eta_8 \cdot p_s \cdot \log{p_s}.
\]
The desired inequality \eqref{eqn:desired} follows, completing the proof 
of Theorem~\ref{thm:main}.
%To proceed, we will argue as in the proof of Theorem 4 of Bugeaud and Gy\H{o}ry \cite{BuGy}, specializing to use the additional information available to us (principally, that the number field defined by $F(X,1)=0$ is Galois). In the notation of \cite{BuGy}, this implies that
%$$
%\left[ \mathbb{M}_{123} : \mathbb{Q} \right] = \left[ \mathbb{M} (\tau_1,\tau_2,\tau_3)  : \mathbb{Q} \right] = \frac{1}{2}  \,\phi (q^\alpha)
%$$
%and hence we obtain a version of Theorem 4 of \cite{BuGy} with $N$ replaced by $\frac{1}{2}  \,\phi (q^\alpha)$. In  the terminology of that result, we have (appealing to (\ref{buster})), 
%\begin{equation} \label{uppers}
%n=\frac{1}{2}  \, \phi (q^\alpha), \; \; P \leq p_k, \; \; s \ll \max \left\{ 1, \frac{p_k}{q^\alpha \log p_k}  \right\} \, \mbox{ and } \; B=e.
%\end{equation}
%Since the roots of $F(1,Y)=0$ are bounded in modulus by $4$, arguing crudely, the largest coefficient of $F$ is thus at most $4^{\phi (q^\alpha)/2}$. Since $F(X,Y) \mid F_m(X,Y)$, from (\ref{discy}), it follows that the discriminant of $F$, $D(F(X,1))$, satisfies
%$$
%\log |D(F(X,1))| \ll m \log m
%$$
%and so appealing to Bugeaud and Gy\H{o}ry's displayed equation (6.27)  and the argument following it, we find that 
%$$
%\log \left( R_{\mathbb{M}} h_{\mathbb{M}}  \right) \ll m \log m.
%$$
%$$
%, \, H = \max \left\{ \binom{m-j}{j} \; : \; 0 \leq j \leq [m/2] \right\}.
%$$
%We may show, via an explicit version of Stirling's formula, that
%$$
%H < \left( (27/4)^{1/4} \right)^m.
%$$


%Note that Bugeaud and Gy\H{o}ry's displayed equation (6.27) becomes, with (\ref{discy}) and, say, $m$ even,
%$$
%R_{\mathbb{M}} h_{\mathbb{M}} < \left( \left( m/2-1 \right)! \right)^{-1} (m+1)^{(m-2)/4} \left( \log ((m+1)^{(m-2)/2}) \right)^{m/2-1}
%$$
%whence, very roughly,
%$$
%R_{\mathbb{M}} h_{\mathbb{M}} \ll (e^2m)^{m/4} \log^{m/2} (m).
%$$


%Taking logarithms of the first inequality in Theorem 4 of Bugeaud and Gy\H{o}ry, with $N=\phi (q^\alpha)/2$, multiplying by $m$ and again taking logarithms leads to the conclusion that
%$$
%\log \log (p^m) \ll m \log (m)  + m \log p_k.
%$$
%Applying inequality (\ref{upper}), we thus have
%$$
%\log \log (p^m) \ll  p_k \log p_k,
%$$
%from which \eqref{eqn:required} follows. This completes the proof of Lemma~\ref{lem:main}
%and therefore the proof of Theorem~\ref{thm1a}.

\begin{comment}
Thus from \eqref{eqn:RSbound},
\[
	\log(R_S) \; < \; c_3 \cdot m \log{m}+t \cdot \log\log{P_S}
\]


Next, suppose that $m \geq 6$. If 
\begin{equation} \label{special}
m \in \{ 7, 9, 11, 14,  19, 23, 29, 39,  55, 59, 119, 167, 839  \},
\end{equation}
then, in each case, we have that $\tau (p^m)$ is divisible by at least one of $\tau (p^2)$, $\tau (p^3)$ or $\tau (p^4)$. It follows that there 
exists an effectively computable absolute positive constant  $c_2$ such that
\begin{equation} \label{small2}
P(\tau (p^m))  \geq c_2 \log \log (p^m), 
\end{equation}
for $m$ as in (\ref{special}).

\medskip

In general for  $m \geq 6$, we observe that, via induction on (\ref{eqn:recursion}), we can write
\begin{equation} \label{formy}
\tau (p^m) = \tau^\delta (p) \,  F_m (p^{11},\tau^2 (p)),
\end{equation}
where  $\delta \in \{ 0, 1 \}$, $\delta \equiv m \mod{2}$ and $F_m(X,Y)$ is a binary form with integer coefficients and degree $[m/2]$. Explicitly, expanding
$$
\frac{1}{1-\sqrt{Y}T + XT^2} = \sum_{m=0}^\infty Y^{\delta/2} \, F_m(X,Y)\, T^m,
$$
we find, for example, that
$$
F_0(X,Y)=F_1(X,Y)=1, \; F_2(X,Y)=-X+Y, \; F_3(X,Y)=-2X+Y, \; F_4(X,Y)=X^2-3XY+Y^2,
$$
$$
F_5(X,Y)=(3X-Y)(X-Y) \; \mbox{ and } \; F_6(X,Y)=-X^3+6X^2Y-5XY^2+Y^3.
$$
Note that from \eqref{eqn:taulucas} we have the following divisibility property
%the fact that the Lucas sequences $\{ u_n \}$ defined in (\ref{eqn:unform}) are divisibility sequences,
we have
$$
F_{m}(X,Y) \mid F_{n}(X,Y) \mbox{ if } n \equiv -1 \mod{m+1},
$$
whence many of these forms are, in fact, reducible (and, in particular, this is true of $F_{2n+1}(X,Y)$, for all $n \geq 2$).
Explicitly, we can write
\begin{equation} \label{cool}
F_m(X,Y) = \sum_{j=1}^{[m/2]} (-1)^j \binom{m-j}{j} Y^{[m/2]-j}X^j = \prod_{j=1}^{[m/2]} \left( Y-4X \cos^2 \left(\frac{\pi j}{m+1} \right) \right).
\end{equation}
 We observe that
\begin{equation}\label{cool1.5}
F_m(X,Y)=G_m(X,Y-2X),
\end{equation}
where
\begin{equation} \label{cool2}
G_m(X,Y) =  \prod_{j=1}^{[m/2]} \left( Y-2X \cos \left(\frac{2\pi j}{m+1} \right) \right).
\end{equation}
We also have that the discriminant of $F_m$ satisfies
\begin{equation} \label{discy}
D(F_m(X,1))=
\left\{
\begin{array}{cc}
(m+1)^{(m-2)/2} & \mbox{ if $m$ is even,} \\
4 \cdot (m+1)^{(m-5)/2} & \mbox{ if $m$ is odd.} \\
\end{array}
\right.
\end{equation}

Suppose now that $m \geq 6$ and that $m$ is not in the set defined by (\ref{special}). 
From Proposition \ref{Lucas}, (\ref{start}) and (\ref{scale}),  it follows that there exists an integer $j \in \{ 1, 2, \ldots, k \}$ such that $p_j$ is a primitive divisor for $u_{m+1}$;  via Lemma \ref{lem:associated}, $p_j \neq p$. We thus have that $m_{p_j} = m+1$, whereby, from Theorem \ref{thm:Carmichael} and the fact that $p_j \le p_k$,
\begin{equation} \label{upper}
m \leq p_k.
\end{equation}

If  we now define $m_0$ such that $m_0+1$ is the largest prime power divisor of $m+1$, say $m_0+1=q^\alpha$, then
\begin{equation} \label{toby}
q^\alpha \gg \log m
\end{equation}
and, since $m$ is outside the set (\ref{special}), $q^\alpha=7$ or $q^\alpha \geq 9$. From (\ref{eqn:unform}), we may write
\begin{equation} \label{tool}
\tau (p^m) = p^{rm} u_{m+1} =p^{rm} \cdot \frac{u_{m+1}}{u_{q^\alpha}} \cdot \frac{u_{q^\alpha}}{u_{q^{\alpha-1}}}
\cdot u_{q^{\alpha-1}}.
\end{equation}
Here, the term $\frac{u_{q^\alpha}}{u_{q^{\alpha-1}}}$ is an irreducible binary form $F(X,Y)$ in $X=p^{11-2r}$ and $Y=p^{-2r} \tau^2 (p)$, of degree
$$
\frac{1}{2}  \,\phi (q^\alpha) \geq 3,
$$
where $\phi$ denotes the Euler phi-function. Explicitly, we have
\begin{equation} \label{cooler}
F(X,Y) = \prod_{\stackrel{j=1}{\gcd(j,q)=1}}^{\phi (q^\alpha)/2} \left( Y-4X \cos^2 \left(\frac{\pi j}{q^\alpha} \right) \right).
\end{equation}

Suppose that $p_i \mid F(X,Y)$, $p_i \neq p$. From Theorem \ref{thm:Carmichael}, it follows that $m_{p_i} \mid q^\alpha$. 
Since the resultant
$$
\mbox{Res} \left( \frac{u_{q^\alpha}}{u_{q^{\alpha-1}}}, u_{q^{\alpha-1}} \right)
$$
is a power of $q$ and since $\{ u_n \}$ is a divisibility sequence, we may conclude that $p_i=q$ or that $m_{p_i} = q^\alpha$. 
Assuming that $p_i \neq q$, it follows from Theorem \ref{thm:Carmichael} that
$$
p_i \equiv \pm 1 \mod{q^\alpha}.
$$
From (\ref{start}), (\ref{tool}) and Dirichlet's theorem on primes in arithmetic progressions, we may conclude  that
\begin{equation} \label{buster}
\omega (F(X,Y)) \ll \max \left\{ 1, \frac{p_k}{q^\alpha \log p_k} \right\}.
\end{equation}

We now consider the {\it Thue-Mahler} equation
\begin{equation} \label{start-new}
F(X,Y)=  \pm p_1^{\beta_1} p_2^{\beta_2} \cdots p_k^{\beta_k},
\end{equation}
where the $\beta_i$ are nonnegative integers with precisely $\omega (F(X,Y))$ of them nonzero, where this latter quantity is bounded as in (\ref{buster}).
Here, again, $X=p^{11-2r}$ and $Y=p^{-2r} \tau^2 (p)$, and $F$ is as in (\ref{cooler}).


%Now (\ref{start}) and (\ref{scale}), imply that the number of distinct prime divisors of $u_{m+1}$ satisfies 
%$$
%\omega (u_{m+1}) \in \{ k-1, k \}.
%$$ 
%Since the Lucas sequence $\{ u_n \}$ is a {\it divisibility sequence} (in the sense that $u_{n_0} \mid u_n$ whenever $n_0 \mid n$), Proposition \ref{Lucas} thus implies that the number of distinct divisors of $m+1$ itself, $\sigma_0(m+1)$, satisfies 
%$$
%\sigma_0(m+1) \ll k.
%$$
%whereby ?????
%\begin{equation} \label{upper2}
%k \ll \log m.
%\end{equation}




To proceed, we will argue as in the proof of Theorem 4 of Bugeaud and Gy\H{o}ry \cite{BuGy}, specializing to use the additional information available to us (principally, that the number field defined by $F(X,1)=0$ is Galois). In the notation of \cite{BuGy}, this implies that
$$
\left[ \mathbb{M}_{123} : \mathbb{Q} \right] = \left[ \mathbb{M} (\tau_1,\tau_2,\tau_3)  : \mathbb{Q} \right] = \frac{1}{2}  \,\phi (q^\alpha)
$$
and hence we obtain a version of Theorem 4 of \cite{BuGy} with $N$ replaced by $\frac{1}{2}  \,\phi (q^\alpha)$. In  the terminology of that result, we have (appealing to (\ref{buster})), 
\begin{equation} \label{uppers}
n=\frac{1}{2}  \, \phi (q^\alpha), \; \; P \leq p_k, \; \; s \ll \max \left\{ 1, \frac{p_k}{q^\alpha \log p_k}  \right\} \, \mbox{ and } \; B=e.
\end{equation}
Since the roots of $F(1,Y)=0$ are bounded in modulus by $4$, arguing crudely, the largest coefficient of $F$ is thus at most $4^{\phi (q^\alpha)/2}$. Since $F(X,Y) \mid F_m(X,Y)$, from (\ref{discy}), it follows that the discriminant of $F$, $D(F(X,1))$, satisfies
$$
\log |D(F(X,1))| \ll m \log m
$$
and so appealing to Bugeaud and Gy\H{o}ry's displayed equation (6.27)  and the argument following it, we find that 
$$
\log \left( R_{\mathbb{M}} h_{\mathbb{M}}  \right) \ll m \log m.
$$
%$$
%, \, H = \max \left\{ \binom{m-j}{j} \; : \; 0 \leq j \leq [m/2] \right\}.
%$$
%We may show, via an explicit version of Stirling's formula, that
%$$
%H < \left( (27/4)^{1/4} \right)^m.
%$$


%Note that Bugeaud and Gy\H{o}ry's displayed equation (6.27) becomes, with (\ref{discy}) and, say, $m$ even,
%$$
%R_{\mathbb{M}} h_{\mathbb{M}} < \left( \left( m/2-1 \right)! \right)^{-1} (m+1)^{(m-2)/4} \left( \log ((m+1)^{(m-2)/2}) \right)^{m/2-1}
%$$
%whence, very roughly,
%$$
%R_{\mathbb{M}} h_{\mathbb{M}} \ll (e^2m)^{m/4} \log^{m/2} (m).
%$$


Taking logarithms of the first inequality in Theorem 4 of Bugeaud and Gy\H{o}ry, with $N=\phi (q^\alpha)/2$, multiplying by $m$ and again taking logarithms leads to the conclusion that
$$
\log \log (p^m) \ll m \log (m)  + m \log p_k.
$$
Applying inequality (\ref{upper}), we thus have
$$
\log \log (p^m) \ll  p_k \log p_k,
$$
from which \eqref{eqn:required} follows. This completes the proof of Lemma~\ref{lem:main}
and therefore the proof of Theorem~\ref{thm1a}.
\end{comment}


%------------------------------
\section{The equation $\tau(p^2)=\kappa \cdot q^b$}\label{sec:taup2}
%-----------------------------

In this section we establish the following two propositions.
\begin{prop}\label{prop:pmqb}
Let $3 \le q <100$ be a prime.  The equation
\begin{equation}\label{eqn:pmqb}
\tau(p^2) \; = \; \pm q^b,
\qquad
\text{$p$ prime,}
\qquad \text{$b \ge 0$}
\end{equation}
has no solutions.
\end{prop}
\begin{prop}\label{prop:tpsmooth}
The equation
\begin{equation}\label{eqn:tpsmooth}
\tau(p^2) \; = \; \pm 3^{b_1} 5^{b_2} 7^{b_3} 11^{b_4},
\qquad
\text{$p$ prime,}
\qquad \text{$b_1,~b_2,~b_3,~b_4 \ge 0$}
\end{equation}
has no solutions.
\end{prop}

We consider first the following general equation.
\begin{equation}\label{eqn:quad}
\tau(p^2) \; = \; \kappa \cdot q^b,
\qquad
\text{$p \nmid 2\kappa q$ prime,}
\qquad \text{$b \ge 0$}.
\end{equation}
Here $\kappa$ is an odd integer, $q$ is an odd prime, and we assume for convenience
that $q \nmid \kappa$.
Recall that $\tau(p^2)=\tau(p)^2-p^{11}$.
Equation \eqref{eqn:quad} can be written as
\[
p^{11}+ (\kappa \cdot q^b) \cdot 1^{11}\; =\; \tau(p)^2
\]
and so is an equation of signature $(11,11,2)$.
Following the first author and Skinner \cite{BenS}, we associate to a solution of
\eqref{eqn:quad} the Frey-Hellegouarch curve
\[
\begin{cases}
E_{p}\; : \;
Y^2=X(X^2+2\tau(p) X + \tau(p)^2-p^{11}) & \text{if $p \equiv 1 \mod{4}$},\\
E_{p} \; : \;
Y^2=X(X^2+2 \tau(p) X +p^{11}) & \text{if $p \equiv 3 \mod{4}$}.\\
\end{cases}
\]
Let
\begin{equation}\label{eqn:cond}
N=\begin{cases}
2^5 \cdot \Rad(\kappa) \cdot q \cdot p & \text{if $b >0$}\\
2^5 \cdot \Rad(\kappa) \cdot p & \text{if $b=0$},
\end{cases}
\qquad
N^\prime=\begin{cases}
2^5 \cdot \Rad(\kappa) \cdot q & \text{if $11 \nmid b$}\\
2^5 \cdot \Rad(\kappa) & \text{if $11 \mid b$}.
\end{cases}
\end{equation}
Here $\Rad(\kappa)$ denotes the product of the prime divisors of $\kappa$.
The Frey-Hellegouarch curve $E_p$ has conductor $N$.
Moreover, it follows from the recipes of the first author and Skinner \cite{BenS}
(based on the modularity theorem and Ribet's level lowering theorem)
that there is a normalized newform
\begin{equation}\label{eqn:newform}
f=q+\sum_{n=1}^\infty c_n q^n
\end{equation}
of weight $2$ and level $N^\prime$
and a prime $\varpi \mid 11$ in the integers of $K=\Q(c_1,c_2,\dotsc)$
so that
\begin{equation}\label{eqn:mod11}
\overline{\rho}_{E_p,11} \sim \overline{\rho}_{f,\varpi}.
\end{equation}
The restrictions on $\kappa$ and $q$ being coprime
odd integers merely reduce the number of possibilities for $N$, $N^\prime$,
yet cover all the cases we are interested in solving.
The restriction $p \nmid 2 \kappa q$ is needed
so that the minimal discriminant $\Delta$ of the Frey-Hellegouarch curve $E_p$
satisfies $\ord_p(\Delta) \equiv 0 \mod{11}$ which is necessary for
 application of  Ribet's level lowering theorem
in order to obtain a weight $2$ newform $f$ of level $N^\prime$ not divisible
by $p$.

Throughout what follows, $\ell$ will be a prime satisfying
\begin{equation}\label{eqn:lassump}
\ell \nmid 2 \cdot 11 \cdot \kappa q  p.
\end{equation}
Then, taking traces of the images of the Frobenius element at $\ell$
in \eqref{eqn:mod11} we obtain $a_\ell(E_p) \equiv c_\ell \mod{\varpi}$ and
so
\begin{equation}\label{eqn:mod11norm}
\Norm_{K/\Q}(a_\ell(E_p)-c_\ell) \equiv 0 \mod{11}.
\end{equation}

We shall use both the congruences for the $\tau$-function
\eqref{mod2}--\eqref{mod23} and also \eqref{eqn:mod11norm}
to derive congruences for $b$.

\begin{lem}\label{lem:presieve}
Let $(p,b)$ be a solution to \eqref{eqn:quad} and suppose $p \ne 3$, $23$.
Let $\ell$ be a prime satisfying \eqref{eqn:lassump}. Let
\[
\cA_{\ell}=\{(s,t) \; : \; s,~t \in \F_\ell, \quad s \not \equiv 0 \mod{\ell},
\qquad t^2-s^{11} \not \equiv 0 \mod{\ell}
\},
\]
and
\begin{equation}\label{eqn:cB}
\cB_{\ell}=
\begin{cases}
\cA_{\ell} & \ell \ne 3,~5,~7,~23;\\
\{(s,t) \in \cA_3 \; : \; t \equiv s+1 \mod{3}\} & \ell=3;\\
\{(s,t) \in \cA_{5} \; : \; t \equiv s^2(s^{3}+1) \mod{5} \} & \ell=5;\\
\{(s,t) \in \cA_{7} \; : \; t \equiv s(s^{3}+1) \mod{7} \} & \ell=7;\\
\{(s,0) \; : \; s \in \F_{23}^*\setminus (\F_{23}^*)^2\} \cup
\{(s,t) \; : \; s \in (\F_{23}^*)^2,~t=2,~-1\}  & \ell=23.
\end{cases}
\end{equation}
For $(s,t) \in \cB_\ell$ let
\[
%\begin{cases}
E_{s,t,1}/\F_\ell \; : \; Y^2=X(X^2+2 t X +t^2-s^{11}), %\qquad \text{if $p \equiv
%1 \mod{4}$}\\
\qquad
E_{s,t,3}/\F_\ell \; : \; Y^2=X(X^2+2 t X +s^{11}).
%\qquad \text{if $p \equiv 3 \mod{4}$}.
%\end{cases}
\]
Let $f$ be a newform of weight $2$ and level $N^\prime$
so that \eqref{eqn:mod11} is satisfied, and $c_\ell$
be its $\ell$-th coefficient.
For $j=1$, $3$, let
\begin{equation}\label{eqn:cC}
\cC_{\ell,j}(f)=
\{(s,t) \in \cB_\ell \; : \; \Norm(a_\ell(E_{s,t,j})-c_\ell) \equiv 0
\mod{11}\},
\end{equation}
and
\[
\cD_{\ell,j}(f)=\{ t^2-s^{11} \; : \; (s,t) \in \cC_{\ell,j}(f)\} \subseteq \F_\ell.
\]
If $p \equiv 1 \mod{4}$ then
$(\kappa \cdot q^b \mod \ell) \in \cD_{\ell,1}(f)$.
If $p \equiv 3 \mod{4}$ then
$(\kappa \cdot q^b \mod \ell) \in \cD_{\ell,3}(f)$.
\end{lem}
\begin{proof}
Since $\ell \nmid 2 \kappa q p$, and
$\tau(p)^2-p^{11}=\tau(p^2)=\kappa \cdot q^b$
we see that there is some $(s,t) \in \cA_\ell$
so that $(p,\tau(p)) \equiv (s,t) \mod{\ell}$.
Moreover, from the congruences
for $\tau$ in \eqref{mod3}--\eqref{mod23}
there is some $(s,t) \in \cB_\ell$
so that $(p,\tau(p)) \equiv (s,t) \mod{\ell}$;
it is here that we make use of the assumption $p \ne 3$, $23$.
For such a pair $(s,t)$, the
reduction modulo $\ell$ of the Frey-Hellegouarch curve $E_p$
is $E_{s,t,j}/\F_\ell$,
where $j=1$ or $3$ according to whether $p \equiv 1$ or $3 \mod{4}$.
Thus $a_\ell(E_{s,t,j}) =a_\ell(E_p)$. Hence,
$\Norm(a_{\ell}(E_{s,t,j})-c_\ell) \equiv 0 \mod{11}$
by \eqref{eqn:mod11norm}, and so $(s,t) \in \cC_{\ell,j}(f)$.
Since $t^2-s^{11} \equiv \tau(p)^2-p^{11} \equiv \kappa \cdot q^{b} \mod{\ell}$
we see that
$(\kappa \cdot q^b \bmod \ell) \in \cD_{\ell,j}(f)$.
\end{proof}

For any prime $\ell$ satisfying \eqref{eqn:lassump},
the lemma gives congruences
for $q^b$ modulo $\ell$,
and hence leads to congruences for $b$ modulo $O_{\ell}(q)$,
where $O_{\ell}(q)$ will be our notation for the multiplicative order
of $q$ modulo $\ell$. This idea is formalized
in the following lemma.
\begin{lem}\label{lem:sieve}
Let $(p,b)$ be a solution to \eqref{eqn:quad} with $p \ne 3$, $23$,
and let $M$ be a positive integer satisfying $22 \mid M$.
Define $\cE_1$ and $\cE_3$ via
\[
\cE_1=
\{ 0 \le \beta \le M-1 \; : \; \kappa \cdot q^\beta \equiv 3 \mod{4}\} 
\; \; \mbox{ and } \; \; 
\cE_3=\{ 0 \le \beta \le M-1 \; : \; \kappa \cdot q^\beta \equiv 1 \mod{4}\}.
\]
Let $f$ be a newform of weight $2$ and level $N^\prime$
so that \eqref{eqn:mod11} is satisfied.
For $j=1$, $3$, define
\[
\cF_{j}(f)=
\begin{cases}
\{ \beta \in \cE_j \; : \; 11 \nmid \beta\}
& \text{if $N^\prime=2^5 \cdot
\Rad(\kappa) \cdot q$}\\
\{ \beta \in \cE_j \; : \; 11 \mid \beta\}
& \text{if $N^\prime=2^5 \cdot
\Rad(\kappa)$}.\\
\end{cases}
\]
Suppose now that $\cL$ is a set of
primes satisfying
\begin{equation}\label{eqn:cL}
\ell \nmid 2 \cdot 11 \cdot \kappa q p, \qquad
O_{\ell}(q) \mid M.
\end{equation}
For $\ell \in \cL$ and $j=1$, $3$, let
\[
\cG_{\ell,j}(f) \; = \; \{ \beta \in \cF_{j}(f) \; : \;
(\kappa \cdot q^\beta \bmod \ell) \in \cD_{\ell,j}(f)
\}.
\]
Let
\[
\cH_j(f) \; = \; \bigcap_{\ell \in \cL} \cG_{\ell,j}(f).
\]
If $p \equiv 1 \mod{4}$ then
there is some $\beta \in \cH_1(f)$ such that
$b \equiv \beta \mod{M}$.
If $p \equiv 3 \mod{4}$ then
there is some $\beta \in \cH_3(f)$ such that
$b \equiv \beta \mod{M}$.
\end{lem}
\begin{proof}
Let $0 \le \beta \le M-1$ be the unique integer
such that $\beta \equiv b \mod{M}$.
Let $j=1$, $3$ according to whether $p \equiv 1$ or $3 \mod{4}$
respectively.
As $2 \mid M$ and $q$ is odd we have $\kappa \cdot q^\beta \equiv \kappa
\cdot q^b \mod{4}$.
Note from \eqref{mod2} that
\[
\kappa \cdot q^\beta \equiv
\kappa \cdot q^b=\tau(p^2) = \tau(p)^2-p^{11} \equiv  p^2+p+1 \equiv
\begin{cases}
3 \mod{4} & \text{if $p \equiv 1 \mod{4}$}\\
1 \mod{4} & \text{if $p \equiv 3 \mod{4}$}.\\
\end{cases}
\]
Thus $\beta \in \cE_j$.

Also $11 \mid M$. Hence $11 \mid b$
if and only if $11 \mid \beta$.
From the definition of $N^\prime$
in \eqref{eqn:cond} we see that $\beta \in \cF_j(f)$.
%Thus $\beta \in \cD_1$.

Now let $\ell \in \cL$. By Lemma~\ref{lem:presieve},
we know that $(\kappa \cdot q^b \bmod{\ell})
\in \cD_{\ell,j}(f)$. However $O_\ell(q) \mid M$ and $M \mid (\beta-b)$.
Thus $\kappa \cdot q^{\beta} \equiv \kappa \cdot q^{b} \mod{\ell}$,
and so $(\kappa \cdot q^\beta \bmod{\ell}) \in \cD_{\ell,j}(f)$.
We deduce that $\beta \in \cG_{\ell,j}(f)$ for all $\ell \in \cL$.
Therefore $\beta \in \cH_j(f)$ completing the proof.
\end{proof}

\begin{proof}[Proof of Proposition~\ref{prop:pmqb}]
We checked that \eqref{eqn:pmqb} has no solutions
with $p<200$ for primes $3 \le q <100$. We shall henceforth suppose
that $p>200$. In particular, $p \ne q$.
Moreover, any solution to \eqref{eqn:pmqb}
is a solution to \eqref{eqn:quad} with $\kappa=1$ or $-1$.
For a given $3 \le q < 100$ we shall let
\[
M=396=2^2 \cdot 3^2 \cdot 11, \qquad
\cL=\{\text{$3 \le \ell <200$ prime, $\ell \ne 11$, $q$} \; : \;  O_\ell(q) \mid M\}.
\]
Observe that since $p>200$ that every $\ell \in \cL$ satisfies
 \eqref{eqn:cL}.

Suppose first that $11 \mid b$ 
and write $b=11c$. Then $(x,y,z)=(p,\pm q^{c},\tau(p))$
is a solution to the equation $x^{11}+y^{11}=z^2$ satisfying $\gcd(x,y,z)=1$.
Darmon and Merel \cite{DM} showed that the equation $x^n+y^n=z^2$
has no solutions $(x,y,z) \in \Z^3$ with $n \ge 4$, $\gcd(x,y,z)=1$.
This contradiction completes the proof for $q \ne 3$. 

Thus $11 \nmid b$, and so
 in \eqref{eqn:cond} the level is $N^\prime=2^5 q$.
We will consider the case $q=3$ a little later. Suppose $5 \le q < 100$.
We wrote a \texttt{Magma} script which for each prime
$5 \le q <100$, computes the weight $2$ newforms $f$ of level
$N^\prime=2^5 q$,
and the sets $\cH_1(f)$ and $\cH_3(f)$ both for $\kappa=1$, $\kappa=-1$.
We found all of these
to be empty. By Lemma~\ref{lem:sieve}, we conclude that
\eqref{eqn:pmqb} has no solutions with $5 \le q< 100$.


\bigskip

It remains to consider the case $q=3$. By Lemma~\ref{lem:mod9}, we see that
	$b=0$ or $1$. But $11 \nmid b$, therefore $b=1$.
Thus
\begin{equation}\label{eqn:pm3}
\tau(p)^2-p^{11}=\pm 3.
\end{equation}
We consider this modulo $23$ using \eqref{mod23}.
If $p$ is a quadratic non-residue
modulo $23$, then $p^{11} \equiv -1 \mod{23}$ and $\tau(p) \equiv 0
\mod{23}$ giving a contradiction.
If $p$ is a quadratic residue
modulo $23$, then $p^{11} \equiv 1 \mod{23}$ and $\tau(p) \equiv 2$,
$-1
\mod{23}$. We conclude that $\tau(p) \equiv 2 \mod{23}$ and
$\tau(p)^2-p^{11}=3$.
Thus
\[
(\tau(p)+\sqrt{3})(\tau(p)-\sqrt{3})=p^{11}.
\]
The two factors on the left-hand side are coprime integers in
$\Z[\sqrt{3}]$. We see that
\[
\tau(p)+\sqrt{3}=(2+\sqrt{3})^a \gamma^{11}, \qquad \gamma \in \Z[\sqrt{3}],
\qquad \Norm(\gamma)=p, \qquad 0 \le a \le 10.
\]
Let $\fq=(2+3\sqrt{3})\Z[\sqrt{3}]$. Then $23 \Z[\sqrt{3}]=\fq \overline{\fq}$.
Since $\fq$ has residue field $\F_{23}$, we see that $\gamma^{11} \equiv \pm 1
\mod{\fq}$. Moreover, as $\tau(p) \equiv 2 \mod{23}$ we have
\[
2+\sqrt{3} \equiv \pm (2+\sqrt{3})^a \mod{\fq}.
\]
However, $2+\sqrt{3}$ has multiplicative order $11$ in $\Z[\sqrt{3}]/\fq=\F_{23}$.
As $0 \le a \le 10,$ we conclude that $a=1$. Thus
\[
\tau(p)+\sqrt{3}=(2+\sqrt{3}) (U+V\sqrt{3})^{11}, \qquad U,~V \in \Z.
\]
Comparing coefficients of $\sqrt{3}$ we obtain the Thue equation
\begin{multline*}
U^{11} + 22 U^{10} V + 165 U^9 V^2 + 990 U^8 V^3 + 2970 U^7 V^4 + 8316 U^6 V^5 +\\
        12474 U^5 V^6 + 17820 U^4 V^7 + 13365 U^3 V^8 + 8910 U^2 V^9 +
        2673 U V^{10} + 486 V^{11} \; =\; 1.
\end{multline*}
The \texttt{Magma} Thue equation solver (based
on algorithms in \cite{Smart}) gives that
the only solution is $(U,V)=(1,0)$.
Thus  $p=U^2-3V^2=1$ which is a contradiction.
\end{proof}

\noindent \textbf{Remark.}
The reader might be wondering if the case $11 \mid b$ can
also be tackled using Lemma~\ref{lem:sieve} instead of
appealing to Darmon and Merel. In that case, $N^\prime=32$,
and there is precisely one weight $2$ newform $f$ of level $32$.
This has rational eigenvalues and corresponds to the elliptic curve
\[
E \; : \; Y^2=X^3-X.
\]
Let $\ell \ne 2$ be a prime.
By inspection of the definition of $\cB_\ell$ in Lemma~\ref{lem:presieve},
we note that $(-1,0) \in \cB_\ell$ and that
$E_{-1,0,3}$ is the reduction modulo $\ell$
of the elliptic curve $E$. Thus $a_\ell(E_{-1,0,3})=a_\ell(E)=c_\ell$
where $c_\ell$ is the $\ell$-th coefficient of $f$. Thus
$(-1,0) \in \cC_{\ell,3}(f)$, and therefore $1 \in \cD_{\ell,3}(f)$.
Let $\kappa=1$. Going through the definitions in Lemma~\ref{lem:sieve},
it it easy to verify that $0 \in \cH_3(f)$ regardless of the choice
of $M$ and $\cL$. Hence we cannot use Lemma~\ref{lem:sieve}
to rule out the case $\kappa=1$ and $11 \mid b$.

\bigskip

There is a similar explanation for why we are unable to
use Lemma~\ref{lem:sieve} on its own to rule out the case
$q=3$, $\kappa=1$ and $11 \nmid b$. Here $N^\prime=96$.
There are two weight $2$ newforms of level $96$ and we
take $f$ to be the one corresponding to the elliptic curve
\[
E \; : \; Y^2=X^3+4X^2+3X.
\]
Let $\ell \nmid 6$ be  a prime.
We note that $(1,2) \in \cB_\ell$. Moreover, $E_{1,2,1}$ is
the reduction modulo $\ell$ of $E$. Hence $a_\ell(E_{1,2,1})=a_\ell(E)
=c_\ell$ which is as before the $\ell$-th coefficient of $f$.
We therefore have $(1,2) \in \cC_\ell(f)$ and so $3 \in \cD_{\ell,1}(f)$.
It follows, for $\kappa=1$, that $1 \in \cH_1(f)$
regardless of the choice of $M$ and $\cL$.

\begin{proof}[Proof of Proposition~\ref{prop:tpsmooth}]
Again we checked that equation \eqref{eqn:tpsmooth} has no
solutions with $p<200$ so we may suppose that $p>200$.
Moreover, by Lemmas \ref{lem:mod7}, \ref{lem:mod5} and \ref{lem:mod9},
we have $b_1=0$ or $b_1=1$, and $b_2=b_3=0$ in \eqref{eqn:tpsmooth}.
If $b_1=0$ then equation \eqref{eqn:tpsmooth} becomes
$\tau(p)^2-p^{11}=\pm 11^{b_4}$ which does not have any solutions
by Proposition~\ref{prop:pmqb}. Hence $b_1=1$. For convenience
we write $b$ for $b_4$, so equation \eqref{eqn:tpsmooth} becomes
\begin{equation}\label{eqn:BS2}
\tau(p)^2-p^{11}=\pm 3 \cdot 11^b.
\end{equation}
We apply Lemma~\ref{lem:sieve} with $q=11$
and $\kappa=\pm 3$. Here $N^\prime=96$ if $11 \mid b$
and $N^\prime=96 \times 11=1056$ if $11 \nmid b$.
For the newforms $f$ at both these levels and for
$\kappa=3$ and $\kappa=-3$, we computed $\cH_1(f)$ and $\cH_3(f)$.
We found that all these are empty with precisely one exception.
For that exception $\kappa=3$, and $f$ is the newform of level $96$
corresponding to the elliptic curve $E$ with Cremona label
\texttt{96a1}, when we find
\[
\cH_1(f)=\{0,\, 22,\, 44,\, 66,\, 88,\, 110,\, 132,\, 154,\, 176,\,
198,\, 220,\, 242,\, 264,\, 286,\, 308,\, 330,\,
    352,\, 374\},
\]
and so Lemma~\ref{lem:sieve} does not provide a contradiction.
However, we know that if $(p,b)$ is a solution to \eqref{eqn:BS2}
then $\overline{\rho}_{E_p,11} \sim \overline{\rho}_{f,\varpi} \sim
\overline{\rho}_{E,11}$. Suppose $b \ne 0$. Then the Frey-Hellegouarch curve
$E_p$ has conductor $96 \cdot 11$ and so multiplicative reduction
at $11$. The curve $E$ has conductor $96$ and hence good reduction
at $11$. Comparing the traces of Frobenius at $11$
in the two representations $\overline{\rho}_{E_p,11} \sim
\overline{\rho}_{E,11}$ (see \cite{KO}) we obtain
$\pm (11+1) \equiv a_{11}(E) \mod{11}$. However, $a_{11}(E)=4$
giving a contradiction. Thus $b=0$. Equation~\eqref{eqn:BS2}
now becomes equation~\eqref{eqn:pm3}, which we showed, in the proof of
Proposition~\ref{prop:pmqb}, to have no solutions.
This completes the proof.
\end{proof}

%-----------------------------
\section{The equation $\tau(p^4)= \kappa \cdot q^b$}\label{sec:taup4}
%-----------------------------

In this section, we establish the following two propositions.
\begin{prop}\label{prop:4pmqb}
Let $3 \le q <100$ be a prime.  The equation
\begin{equation}\label{eqn:4pmqb}
\tau(p^4) \; = \; \pm q^b,
\qquad
\text{$p$ prime,}
\qquad \text{$b \ge 0$}
\end{equation}
has no solutions.
\end{prop}
\begin{prop}\label{prop:4tpsmooth}
The equation
\begin{equation}\label{eqn:4tpsmooth}
\tau(p^4) \; = \; \pm 3^{b_1} 5^{b_2} 7^{b_3} 11^{b_4},
\qquad
\text{$p$ prime,}
\qquad \text{$b_1,~b_2,~b_3,~b_4 \ge 0$}
\end{equation}
has no solutions.
\end{prop}

We consider first the following general equation.
\begin{equation}\label{eqn:4quad}
\tau(p^4) \; = \; \kappa \cdot q^b,
\qquad
\text{$p \nmid 2\kappa q$ prime,}
\qquad \text{$b \ge 0$}.
\end{equation}
Here $\kappa$ is an odd integer, $q$ is an odd prime, and we assume for convenience
that
\[
q \nmid 5 \kappa, \qquad \text{$\ord_5(\kappa)=0$ or $1$}.
\]
Using the recursion \eqref{eqn:recursion} we find that
\[
\tau(p^4)=\tau(p)^4-3p^{11} \tau(p)^2 +p^{22}.
\]
which can be written as
\begin{equation}\label{eqn:rewrite}
4 \tau(p^4) = (2 \tau(p)^2-3 p^{11})^2 - 5 p^{22}.
\end{equation}
We may therefore rewrite \eqref{eqn:4quad} as
\[
5 (p^2)^{11}+ (4 \cdot \kappa \cdot q^b) \cdot 1^{11}\; =\; (2\tau(p)^2-3 p^{11})^2,
\]
which is an equation of signature $(11,11,2)$.
As before we follow the first author and Skinner \cite{BenS},
and associate to a solution of
\eqref{eqn:4quad} the Frey-Hellegouarch curve
\[
\begin{cases}
E_{p}\; : \;
Y^2=X(X^2+(3p^{11}-2 \tau(p)^2) X + \tau(p)^4-3p^{11} \tau(p)^2 +p^{22}) & \text{if $p \equiv 1 \mod{4}$},\\
E_{p} \; : \;
Y^2=X(X^2+(2 \tau(p)^2-3p^{11}) X +\tau(p)^4-3p^{11} \tau(p)^2 +p^{22}) & \text{if $p \equiv 3 \mod{4}$}.\\
\end{cases}
\]
Let
\begin{equation}\label{eqn:4N}
N=\begin{cases}
2^3 \cdot 5 \cdot \Rad(\kappa) \cdot q \cdot p & \text{if $b >0$, $\ord_5(\kappa)=0$}\\
2^3 \cdot 5 \cdot \Rad(\kappa) \cdot p & \text{if $b=0$, $\ord_5(\kappa)=0$}\\
2^3 \cdot 5^2 \cdot \Rad(\kappa/5) \cdot q \cdot p & \text{if $b >0$, $\ord_5(\kappa)=1$}\\
2^3 \cdot 5^2 \cdot \Rad(\kappa/5) \cdot p & \text{if $b=0$,
$\ord_5(\kappa)=1$},
\end{cases}
\end{equation}
and
\begin{equation}\label{eqn:4Nd}
N^\prime=\begin{cases}
2^3 \cdot 5 \cdot \Rad(\kappa) \cdot q  & \text{if $11 \nmid b$, $\ord_5(\kappa)=0$}\\
2^3 \cdot 5 \cdot \Rad(\kappa)  & \text{if $11 \mid b$, $\ord_5(\kappa)=0$}\\
2^3 \cdot 5^2 \cdot \Rad(\kappa/5) \cdot q  & \text{if $11 \nmid b$, $\ord_5(\kappa)=1$}\\
2^3 \cdot 5^2 \cdot \Rad(\kappa/5)  & \text{if $11 \mid b$,
$\ord_5(\kappa)=1$}.
\end{cases}
\end{equation}
The Frey curve $E_p$ has conductor $N$, and again it follows from
the recipes of the first author and Skinner \cite{BenS}
that there is a normalized newform $f$ as in \eqref{eqn:newform}
of weight $2$ and level $N^\prime$
and a prime $\varpi \mid 11$ in the integers of $K=\Q(c_1,c_2,\dotsc)$
so that \eqref{eqn:mod11} holds.

Throughout what follows, $\ell$ will be a prime satisfying
\begin{equation}\label{eqn:lassump4}
\ell \nmid 2 \cdot 5 \cdot 11 \cdot \kappa \cdot q \cdot p.
\end{equation}
As before \eqref{eqn:mod11norm} holds.

\begin{lem}\label{lem:presieve4}
Let $(p,b)$ be a solution to \eqref{eqn:4quad} and suppose $p \ne 3$, $23$.
Let $\ell$ be a prime satisfying \eqref{eqn:lassump4}. Let
\[
\cA_{\ell}=\{(s,t) \; : \; s,~t \in \F_\ell, \quad s \not \equiv 0 \mod{\ell},
\qquad
t^4-3s^{11} t^2 +s^{22} \not \equiv 0 \mod{\ell}
\},
\]
and let $\cB_{\ell}$ be as in \eqref{eqn:cB}.
%\[
%\cB_{\ell}=
%\begin{cases}
%\cA_{\ell} & \ell \ne 3,~5,~7,~23;\\
%\{(s,t) \in \cA_3 \; : \; t \equiv s+1 \mod{3}\} & \ell=3;\\
%\{(s,t) \in \cA_{5} \; : \; t \equiv s^2(s^{3}+1) \mod{5} \} & \ell=5;\\
%\{(s,t) \in \cA_{7} \; : \; t \equiv s(s^{3}+1) \mod{7} \} & \ell=7;\\
%\{(s,0) \; : \; s \in \F_{23}^*\setminus (\F_{23}^*)^2\} \cup
%\{(s,t) \; : \; s \in (\F_{23}^*)^2,~t=2,~-1\}  & \ell=23.
%\end{cases}
%\]
For $(s,t) \in \cB_\ell$ let
\[
E_{s,t,1}/\F_\ell \; : \; Y^2=X(X^2+(3s^{11}-2t^2) X +t^4-3s^{11}t+s^{22}),
\]
\[
E_{s,t,3}/\F_\ell \; : \; Y^2=X(X^2+(2t^2-3s^{11})X +t^4-3s^{11}t+s^{22}).
\]
Let $f$ be a newform of weight $2$ and level $N^\prime$
so that \eqref{eqn:mod11} is satisfied, and $c_\ell$
be its $\ell$-th coefficient.
For $j=1$, $3$, let $\cC_{\ell,j}(f)$
be as in \eqref{eqn:cC}, and let
%\[
%\cC_{\ell,j}(f)=
%\{(s,t) \in \cB_\ell \; : \; \Norm(a_\ell(E_{s,t,j})-c_\ell) \equiv 0
%\mod{11}\},
%\]
\[
\cD_{\ell,j}(f)=\{ t^4-3 s^{11} t^2+s^{22} \; : \; (s,t) \in \cC_{\ell,j}(f)\} \subseteq \F_\ell.
\]
If $p \equiv 1 \mod{4}$ then
$(\kappa \cdot q^b \mod \ell) \in \cD_{\ell,1}(f)$.
If $p \equiv 3 \mod{4}$ then
$(\kappa \cdot q^b \mod \ell) \in \cD_{\ell,3}(f)$.
\end{lem}
\begin{proof}
The proof is practically identical to that of Lemma~\ref{lem:presieve}.
\begin{comment}
Since $\ell \nmid 2 \kappa q p$, and
$\tau(p)^2-p^{11}=\tau(p^2)=\kappa \cdot q^b$
we see that there is some $(s,t) \in \cA_\ell$
so that $(p,\tau(p)) \equiv (s,t) \mod{\ell}$.
Moreover, from the congruences
for $\tau$ in \eqref{mod3}--\eqref{mod23}
there is some $(s,t) \in \cB_\ell$
so that $(p,\tau(p)) \equiv (s,t) \mod{\ell}$.
For such a pair $(s,t)$, the
reduction modulo $\ell$ of the Frey-Hellegouarch curve $E_p$
is $E_{s,t,j}/\F_\ell$,
where $j=1$ or $3$ according to whether $p \equiv 1$ or $3 \mod{4}$.
Thus $a_\ell(E_{s,t,j}) =a_\ell(E_p)$. Hence,
$\Norm(a_{\ell}(E_{s,t})-c_\ell) \equiv 0 \mod{11}$
by \eqref{eqn:mod11norm}, and so $(s,t) \in \cC_{\ell,j}(f)$.
Since $t^2-s^{11} \equiv \tau(p)^2-p^{11} \equiv \kappa \cdot q^{b} \mod{\ell}$
we see that
$(\kappa \cdot q^b \bmod \ell) \in \cD_{\ell,j}(f)$.
\end{comment}
\end{proof}

\begin{lem}\label{lem:sieve4}
Let $(p,b)$ be a solution to \eqref{eqn:4quad} with $p \ne 3$, $23$.
Let $M$ be a positive integer satisfying $22 \mid M$.
Let
\[
\cE=
\{ 0 \le \beta \le M-1 \; : \; \kappa \cdot q^\beta \equiv 1 \mod{4}\}.
%\qquad
%\cE_3=\{ 0 \le \beta \le M-1 \; : \; \kappa \cdot q^\beta \equiv 1 \mod{4}\}.
\]
Let $f$ be a newform of weight $2$ and level $N^\prime$
so that \eqref{eqn:mod11} is satisfied.
%For $j=1$, $3$ let
Let
\[
\cF(f)=
\begin{cases}
\{ \beta \in \cE \; : \; 11 \nmid \beta\}
& \text{if $N^\prime=
2^3 \cdot 5 \cdot \Rad(\kappa) \cdot q$ or $2^3 \cdot 5^2 \cdot \Rad(\kappa/5) \cdot q$}\\
\{ \beta \in \cE \; : \; 11 \mid \beta\}
& \text{if $N^\prime=
2^3 \cdot 5 \cdot \Rad(\kappa)$ or $2^3 \cdot 5^2 \cdot \Rad(\kappa/5)$}\\
\end{cases}
\]
Let $\cL$ be a set of
primes satisfying
\begin{equation}\label{eqn:cL4}
\ell \nmid 2 \cdot 5 \cdot 11 \cdot \kappa q p, \qquad
O_{\ell}(q) \mid M.
\end{equation}
For $\ell \in \cL$ and $j=1$, $3$ %and $i=0$, $1$ let
\[
\cG_{\ell,j}(f) \; = \; \{ \beta \in \cF(f) \; : \;
(\kappa \cdot q^\beta \bmod \ell) \in \cD_{\ell,j}(f)
\}.
\]
Let
\[
\cH_j(f) \; = \; \bigcap_{\ell \in \cL} \cG_{\ell,j}(f).
\]
If $p \equiv 1 \mod{4}$ then
there is some $\beta \in \cH_1(f)$ such that
$b \equiv \beta \mod{M}$.
If $p \equiv 3 \mod{4}$ then
there is some $\beta \in \cH_3(f)$ such that
$b \equiv \beta \mod{M}$.
\end{lem}
\begin{proof}
This is almost identical to the proof of Lemma~\ref{lem:sieve}.
The main difference is that the sets $\cE$, $\cF(f)$ do not depend on the class
of $p$ modulo $4$, and we explain this now. Observe from \eqref{mod2}
that
\[
\kappa \cdot q^b = \tau(p^4) \equiv p^{44}+p^{33}+p^{22}+p^{11}+1 \equiv 3+2p \equiv 1 \pmod{4}
\]
regardless of the residue class of $p$ modulo $4$.
\end{proof}

\begin{lem}\label{lem:bne0}
The equations $\tau(p^4)=\pm 1$ and $\tau(p^4)=\pm 5$ have no solutions with $p$ prime.
\end{lem}
\begin{proof}
From the proof of Lemma~\ref{lem:sieve4}, we know that $\tau(p^4) \equiv 1
\pmod{4}$. Thus we need only consider the equations $\tau(p^4)=1$ and
$\tau(p^4)=5$.
Suppose $\tau(p^4)=1$ and write $z=2 \tau(p)^2-3p^{11}$. From \eqref{eqn:rewrite} we have
\[
z^2-5 p^{22}= 4.
\]
Write $\varepsilon=(1+\sqrt{5})/2$.
Then $(\lvert z\rvert +p^{11} \sqrt{5})/2$ is
a positive unit in $\Z[\varepsilon]$ with norm $+1$.
Hence
\[
\frac{z+p^{11} \sqrt{5}}{2} \; = \; \varepsilon^{2n}, \qquad
\varepsilon=(1+\sqrt{5})/2.
\]
for some $n \in \Z$. Thus
\[
p^{11}= \frac{\varepsilon^{2n}-\overline{\varepsilon}^{2n}}{\sqrt{5}}
=F_{2n}
\]
where $F_n$ denotes
the $n$-th Fibonacci number.
By \cite{BMS}
the only perfect powers in the Fibonacci sequence are
$0$, $1$, $8$ and $144$, giving a contradiction. Alternatively,
$F_{2n}=F_n L_n$ where $L_n$ is the $n$-th Lucas number.
From the identity $L_n^2-5 F_n^2=4 \cdot (-1)^n$ we
see that $\gcd(F_n,L_n)=1$ or $2$. Thus $F_n=1$ or $L_n=1$
quickly leading to a contradiction.

Next we suppose  that  $\tau(p^4)=5$ and 
write $z=5w$. Hence
\[
5w^2-p^{22}=4
\]
and it follows that there is an integer $n$ such that
\[
p^{11}=\varepsilon^n+\overline{\varepsilon}^n=L_n,
\]
where $L_n$ denotes the $n$-th Lucas number.
By \cite{BMS},
the only perfect powers in the Lucas sequence are $1$ and $4$,
again giving a contradiction.
\end{proof}

\begin{proof}[Proof of Proposition~\ref{prop:4pmqb}]
We checked that \eqref{eqn:4pmqb} has no solutions
with $p<200$ for primes $3 \le q <100$. We shall henceforth suppose
that $p>200$. In particular, $p \ne q$.
Moreover, any solution to \eqref{eqn:4pmqb}
is a solution to \eqref{eqn:4quad} with $\kappa=1$ or $-1$.

We consider $q=5$ first. By \eqref{eqn:rewrite},
$\ord_5(\tau(p^4))=0$ or $1$. Thus we reduce to the equations
$\tau(p^4)=\pm 1$ and $\tau(p^4)=\pm 5$. These do not have solutions
by Lemma~\ref{lem:bne0} and hence we may assume that $q \ne 5$. From Lemma~\ref{lem:bne0} again
we have $b>0$. By \eqref{eqn:rewrite},
$5$ is a quadratic residue modulo $q$. The possible
values of $q$ are
\begin{equation}\label{eqn:q}
11,\; 19,\; 29,\; 31,\; 41,\; 59,\; 61,\; 71,\; 79,\; 89.
\end{equation}
For each of these values we take
\begin{equation}\label{eqn:McL}
M=396=2^2 \cdot 3^2 \cdot 11, \qquad
\cL=\{\text{$3 \le \ell <200$ prime, $\ell \ne 5$, $11$, $q$} \; : \;  O_\ell(q) \mid M\}.
\end{equation}
Observe that since $p>200$ that $\ell \ne p$, and thus
satisfies \eqref{eqn:cL4}.

We consider first the case $11 \nmid b$.
Thus, in \eqref{eqn:4Nd}, the level $N^\prime=2^3 \cdot 5 \cdot q$.
We computed for each newform $f$ of level $N^\prime$
the sets $\cH_1(f)$ and $\cH_3(f)$, both for $\kappa=1$, $\kappa=-1$.
We found all of these
to be empty. By Lemma~\ref{lem:sieve}, we conclude that
\eqref{eqn:4pmqb} has no solutions with $11 \nmid b$.

Next we consider $11 \mid b$. Thus $N^\prime=2^3 \cdot 5$.
There is a unique newform $f$ of level $N^\prime$ which corresponds
to the elliptic curve $E$ with Cremona label \texttt{40a1}.
Thus, from \eqref{eqn:mod11}
we obtain $\overline{\rho}_{E_p} \sim \overline{\rho}_{E}$.
Note, by \eqref{eqn:4N}
that $E_p$ has multiplicative reduction at $q$. However, $E$ has good reduction
	at $q$. Thus, by \cite{KO}, 
	we have $\pm (q+1) \equiv a_q(E) \pmod{11}$. We checked that this
does not hold for all the values of $q$ in \eqref{eqn:q}.
This completes the proof.
\end{proof}

\begin{proof}[Proof of Proposition~\ref{prop:4tpsmooth}]
Again we checked that \eqref{eqn:4tpsmooth} has no solutions with $p<200$, whence
we may suppose $p>200$. Moreover, as $5$ is a quadratic non-residue modulo
$3$ and $7$, we see from \eqref{eqn:rewrite} that $b_1=b_3=0$ in
\eqref{eqn:4tpsmooth}. Also $5^2 \nmid \tau(p^4)$ from \eqref{eqn:rewrite},
so $b_2=0$ or $1$.
But from Proposition~\ref{prop:4pmqb} we have $b_2 \ne 0$, and so $b_2=1$.
We have thus reduced to consideration of the equation
\[
\tau(p^4)=\pm 5 \cdot 11^b,
\]
whereby we have $\kappa=\pm 5$ and $q=11$.
Observe that $b>0$ by Lemma~\ref{lem:bne0}. Suppose $11 \nmid b$.
Thus $N^\prime=8 \cdot 25 \cdot 11=2200$.
We take $M$ and $\cL$ as in \eqref{eqn:McL}.
There are $25$ conjugacy
classes of newforms $f$ of weight $2$ and level $2200$. For each, we
found $\cH_1(f)$ and $\cH_3(f)$ to be empty, both for $\kappa=5$ and $\kappa=-5$.
By Lemma~\ref{lem:sieve4}, there are no solutions with $11 \nmid b$.
Thus $11 \mid b$, and so $N^\prime=2^3 \cdot 25=200$. There are
five weight $2$ newforms of level $200$. We computed $\cH_1(f)$
and $\cH_3(f)$ for these, both for $\kappa=5$ and $\kappa=-5$.
The only non-empty one we found was $\cH_3(f)$ for $\kappa=5$
where $f$ is the rational newform corresponding to the elliptic curve $E$
with Cremona label \texttt{200b1}.
Then $\overline{\rho}_{E_p,11} \sim \overline{\rho}_{E,11}$.
Here $E_p$ has multiplicative
reduction at $11$, though $E$ has good reduction at $11$.
As before,
$\pm (11+1) \equiv a_{11}(E) \mod{11}$.
However, $a_{11}(E)=-4$, giving a contradiction and completing the proof.
\end{proof}

%------------------------------------------------------------------
%\section{$P(\tau(p^3)) \le 11$}
\section{On the largest prime divisor of $\tau(p^3)$}\label{sec:taup3}
%------------------------------------------------------------------
\begin{prop}\label{prop:tp3smooth}
Let $p$ be a prime for which $\tau(p) \ne 0$. Then
$P(\tau(p^3)) \ge 13$, unless $p=2$, in which case we have
$\tau(8)=2^9 \cdot 3 \cdot 5 \cdot 11$.
\end{prop}
We consider
\begin{equation}\label{eqn:tp3smooth}
P(\tau(p^3)) \le 11.
\end{equation}
We checked that the only $p< 200$ satisfying \eqref{eqn:tp3smooth}
is $p=2$.
We shall therefore suppose $p>200$.
Recall that $\tau(p^3)=\tau(p) \cdot (\tau(p)^2-2 p^{11})$.
From \eqref{mod3} and \eqref{mod7}, we easily see that $3$ and $7$
do not divide $\tau(p)^2-2p^{11}$. Moreover, we recall that $\tau(p)$
is even, so $\ord_{2}(\tau(p)^2-2p^{11})=1$. Thus
\begin{equation}\label{eqn:factored}
\tau(p)^2-2 p^{11}= \pm 2 \cdot 5^a \cdot 11^b \; \; \mbox{ and } \; \;  \tau(p)=\pm 2^r \cdot 3^s
\cdot 5^t \cdot 7^u \cdot 11^v.
\end{equation}
As before, we associate to this a Frey-Hellegouarch curve
\[
E_p \; : \; Y^2=X(X^2+2 \tau(p) X + 2 p^{11}).
\]
By the recipes of the first author and Skinner, 
the conductor of $E_p$ is one of
\[
N=2^8 \cdot p, \qquad 2^8\cdot 5 \cdot p, \qquad 2^8 \cdot 11 \cdot p, \qquad 2^8 \cdot 5 \cdot
11 \cdot p,
\]
and \eqref{eqn:mod11}
holds for some weight $2$ newform $f$ whose level $N^\prime$
is one of the following
\begin{equation}\label{eqn:Ndhere}
N^\prime=2^8, \qquad 2^8\cdot 5, \qquad 2^8 \cdot 11, \qquad 2^8 \cdot 5 \cdot
11.
\end{equation}
There are a total of $123$ conjugacy classes of newforms $f$ at these levels.
Let $f$ be any of these such that \eqref{eqn:mod11} holds. Let $\ell \ne 2$,
$5$, $11$, $p$ be a prime. Then $11 \mid \Norm(a_\ell(E_p)-c_\ell(f))$ where $c_\ell(f)$
is the $\ell$-th coefficient of $f$.
\begin{lem}
Let $\ell \ne 2$, $5$, $11$ be a prime $<200$.
Let $p$ be an odd prime with $\tau(p) \ne 0$ and $P(\tau(p^3)) \leq 11$.
Let $f$ be a newform of weight $2$ and one of the levels $N^\prime$
in \eqref{eqn:Ndhere} so that \eqref{eqn:mod11} is satisfied.
Write
\[
\cA_{\ell}=
\begin{cases}
\{(s,t) \; : \; s,~t \in \F_\ell, \quad s (t^2-2s^{11})\not \equiv 0
\mod{\ell}\}, \qquad \ell=3,~7\\
\{(s,t) \; : \; s,~t \in \F_\ell, \quad s t (t^2-2s^{11})\not \equiv 0
\mod{\ell}\}, \qquad \ell \ge 13.\\
\end{cases}
\]
Let $\cB_\ell$ be as in \eqref{eqn:cB}. Let
\[
E_{s,t}/\F_{\ell} \; : \; Y^2=X(X^2+2tX+s^{11}),
\]
and
\[
\cC_{\ell}(f)=\{(s,t) \in \cB_\ell \; : \; \Norm(a_\ell(E_{s,t})-c_\ell(f))
\equiv 0 \pmod{11}\}.
\]
Then there is some $(s,t) \in \cC_\ell(f)$ so that
$(p,\tau(p)) \equiv (s,t) \pmod{\ell}$.
\end{lem}
\begin{proof}
This is is similar to the proof of Lemma~\ref{lem:presieve}.
\end{proof}
\begin{proof}[Proof of Proposition~\ref{prop:tp3smooth}]
For each of the $123$ conjugacy classes of newforms $f$ we computed
$\cC_{\ell}(f)$ for $\ell=3$, $7$, $13$ and $23$. We found that
at least one of these four empty, except for the three rational
newforms which correspond to the elliptic curves \texttt{256a1},
\texttt{256b1} and \texttt{256c1}. All three elliptic curves have
CM, respectively by $\Q(\sqrt{-2})$, $\Q(\sqrt{-1})$, $\Q(\sqrt{-1})$.
Note that $11$ splits in $\Q(\sqrt{-2})$ and is inert in $\Q(\sqrt{-1})$.
Hence the image of $\rho_{E_p,11} \sim \rho_{f,11}$ belongs
to the normalizer of split Cartan subgroup in the first case,
and the normalizer of a non-split Cartan subgroup in the second
and third case. Thanks to the work of Momose \cite{Momose},
and Darmon and Merel \cite[Theorem 8.1]{DM}, the $j(E_p) \in \Z[1/11]$.
However, $E_p$ has multiplicative reduction at $p$ giving a contradiction.
\end{proof}

 %---------------------------------------------------
\section{Proof of Theorem \ref{thm2a}}\label{sec:thm2a}
%----------------------------------------------------

\begin{lem}\label{lem:ge13}
Let $p \le 11$ be a prime. Suppose
$P(\tau(p^{m-1})) \le 11$ with $m \ge 3$.
Then $p=2$ and $m=4$.
\end{lem}
\begin{proof}
First let $p=2$.
Let $m \ge 3$ be such that $P(\tau(2^{m-1})) \le 11$.
Note that $\tau(2)=-2^3 \times 3$.
Let $\{u_n\}$ be the Lucas sequence defined in Lemma~\ref{lem:associated},
with characteristic polynomial $X^2-3X+2^5$.
Then $P(u_m) \le 11$. Moreover, by part (i) of Theorem~\ref{thm:Carmichael},
we have $2 \nmid u_n$ for all $n \ge 1$.
We note that
\[
u_2=-3, \quad u_3=-23, \quad
u_4=3 \times 5 \times 11, \quad u_5=241,
\]
\[ u_6=-3^2 \times 23 \times 29,
\quad
u_7=7 \times 1471, \quad u_8=3 \times 5 \times 11 \times 977.
\]
By the Primitive Divisor Theorem (Theorem~\ref{thm:BHV}),
every term $u_n$ with $n \ge 9$ is divisible
by some prime $\ell \ge 13$. Thus the only terms with $P(u_n) \le 11$
are $u_2$ and $u_4$. Since $m \ge 3$ we have $m=4$.

By a similar strategy we checked that $P(\tau(p^{m-1})) \ge 13$
for $3 \le p \le 11$ and $m\ge 3$.
\end{proof}

\begin{lem}\label{lem:taupm}
Let $p$ be a prime. Let $m \ge 3$ be an integer such
that $\tau(p^{m-1}) \ne 0$ and
\begin{equation}\label{taupm}
P(\tau (p^{m-1})) \leq 11.
\end{equation}
Then $p=2$ and $m=4$.
\end{lem}
\begin{proof}
By Lemma~\ref{lem:ge13},  we may suppose $p \ge 13$.
If $\tau(p)=0$, by Lemma~\ref{taup=0}, we have $\tau(p^{m-1})=0$
or a power of $p$ contradicting the hypotheses of the lemma.
We may therefore suppose $\tau(p) \ne 0$.
Fix $p$ and let $m$ be the least value $ \ge 3$ such that \eqref{taupm}
is satisfied.
By Propositions~\ref{prop:tpsmooth}, \ref{prop:4tpsmooth}
and \ref{prop:tp3smooth}, we know that $m \ge 6$.

Suppose first that $p \mid \tau(p)$. By induction from
\eqref{eqn:recursion} we have $p \mid \tau(p^n)$ for all $n \ge 1$.
Hence $p \mid \tau(p^{m-1})$ and so $p \le 11$ giving a contradiction.
Thus $p \nmid \tau(p)$. Let $u_n=\tau(p^{n-1})$ for $n \ge 1$.
Then $\{u_n\}$ is a Lucas sequence by Lemma~\ref{lem:associated}.
Now $u_k \mid u_n$ if $k \mid n$. As $m \ge 6$, it is divisible
either by $4$ or an odd prime. However $u_4=\tau(p^3)$,
and so $P(u_4) \ge 13$ by Proposition~\ref{prop:tp3smooth}.
Hence $m$ is divisible by an odd prime, and from the minimality
of $m$ it follows that $m \ge 7$ is a prime. 
By the Primitive Divisor Theorem, $u_m=\tau(p^{m-1})$ has a prime divisor $q$
that does not divide
$u_1 u_2 \cdots u_{m-1}$ nor $D=(\alpha-\beta)^2$
(where $\alpha$, $\beta$ are as in Lemma~\ref{lem:associated}).
Here $q=2$, $3$, $5$, $7$ or $11$. 
But $m_q$, the rank of apparition of $q$, divides $m$ by
Theorem~\ref{thm:Carmichael}, and so $m_q=m$. However
$m_q \mid (q-1)(q+1)$, again from Theorem~\ref{thm:Carmichael}.
But $(q-1)(q+1)$ is not divisible
by a prime $\ge 7$ for $q=2$, $3$, $5$, $7$ or $11$.
This contradiction completes the proof.
\begin{comment}
Let $q$ be a prime
divisor of $u_m=\tau(p^{m-1})$. Then $q=2$, $3$, $5$, $7$ or $11$.
But $m_q$, the rank of apparition of $q$, divides $m$ by
Theorem~\ref{thm:Carmichael}, and so $m_q=m$. However
$m_q \mid q (q-1)(q+1)$, again from Theorem~\ref{thm:Carmichael}.
Since $m \ge 7$ we conclude that $q \ne 2$, $3$, $5$.
Thus the only possible prime divisors of $u_m$ are $q=7$ and $q=11$.
Moreover, $7(7-1)(7+1)$ and $11(11-1)(11+1)$ are not divisible
by a common prime $\ge 7$. Thus 
	\begin{enumerate}[label=(\roman*),align=left]
	\item either $q=m=7$ and $u_7=\pm 7^a$, for some $a \ge 0$;
	\item or $q=m=11$ and $u_{11}=\pm 11^a$, for some $a \ge 0$;
	\end{enumerate}
However, by the Primitive Divisor Theorem,
the term $u_m$
must have a primitive divisor that divides $u_m$
but not $u_1 u_2 \cdots u_{m-1}$ nor $(\alpha-\beta)^2$
(where $\alpha$, $\beta$ are as in Lemma~\ref{lem:associated}).
	This must be $q=7$ in case (i), and $q=11$ in case (ii).
Moreover, as $q \nmid (\alpha-\beta)^2$, we have by Theorem~\ref{thm:Carmichael}
that $q=m=m_q \mid (q-1)(q+1)$,
giving a contradiction, and completing the proof. 
\end{comment}
\begin{comment}
If $u_m=\pm 7^a$ then $m \ge 7$
is a prime divisor of $7(7-1)(7+1)$ and so $m=7$.
If $u_m= \pm 11^{a}$ then $m \ge 7$
is a prime divisor of $11(11-1)(11+1)$ and so $m=11$.
%Note that $2$ is not a primitive divisor of $u_m$ since
%$u_2=\tau(p)$ is even.
%By the Primitive Divisor Theorem, $\tau(p^{m-1})$
%has a primitive divisor $\ell$ which satisfies $\ell \nmid \tau(p^{n-1})$
%for $n<m$ and $\ell \nmid (\tau(p)^2 - 4 p^{11})$
%Thus this primitive divisor $\ell$
%of $u_m$ must equal one of $3$, $5$, $7$,
%or $11$.
%Observe, by the minimality of $m$,
%that $m=m_\ell(p)$, where $m_\ell(p)$ is the rank of
%apparition of $\ell$.
%But, by Theorem~\ref{thm:Carmichael},
%$m_\ell(p)$ divides $\ell-1$
%or divides $\ell+1$. Thus $m=6$, $7$, $10$, $11$ or $12$.
%However, $u_k \mid u_m$ if $k \mid m$.
%Thus again by the minimality of $m$, we see that $m=7$ or $11$.
%Now, as $m-1$ is even, we deduce that $2 \nmid \tau(p^{m-1})=u_m$.
%Moreover,
%by \cite[Corollaire 2.2]{Ab}, if $q \mid \tau(p^{m-1})$ is a prime
%then either $p=q$ or $m_q(p) \mid m$. However, since $p \ne q$,
%since $p \ge 13$ and $q \le 11$. Thus $m_q(p) \mid m$. But $m=7$ or $11$
%is prime and $m_q(p)>1$. Hence $m_q(p)=m$.
%It follows that the only possible $q \mid \tau(p^{m-1})$
%is $7$ if $m=7$, and is $11$ if $m=11$.
By \eqref{formy}, we see that $(X,Y)=(p^{11},\tau(p))$
is a solution to either
\begin{equation}\label{eqn:p6}
F_6(X,Y)=
-X^3 + 6 X^2 Y - 5 X Y^2 + Y^3
=\pm 7^{a}
\end{equation}
or
\begin{equation}\label{eqn:p10}
F_{10}(X,Y)=
-X^5 + 15 X^4 Y - 35 X^3 Y^2 + 28 X^2 Y^3 - 9 X Y^4 + Y^5
=\pm 11^a
\end{equation}
We solved these two Thue--Mahler equations
using the \texttt{Magma} implementation
of the Thue--Mahler solver described in \cite{GKMS}.
The solutions to \eqref{eqn:p6} are
\[
(X,Y,a)=( -3, -5, 1 ), \quad
    ( -1, -3, 0 ), \quad
    ( -9, -14, 0 ), \quad
    ( -1, -2, 0 ), \quad
    ( 1, 1, 0 ),
\]
\[
    ( 2, 1, 1 ),\quad
    ( 2, 3, 0 ),\quad
    ( 4, 13, 0 ),\quad
    ( 1, 4, 1 ),\quad
    ( -1, 0, 0 ),\quad
    ( 5, 1, 0 ),\quad
    ( 0, 1, 0 ),
\]
\[
    ( -1, -1, 0 ), \quad
    ( -2, -3, 0 ),\quad
    ( 1, 2, 0 ),\quad
    ( 1, 0, 0 ),\quad
    ( 9, 14, 0 ),\quad
    ( -5, -1, 0 ),\quad
    ( -4, -13, 0 ),
\]
\[
    ( -2, -1, 1 ),\quad
    ( 3, 5, 1 ),\quad
    ( -1, -4, 1 ),\quad
    ( 1, 3, 0 ),\quad
    ( 0, -1, 0 ).
\]
The solutions to \eqref{eqn:p10} are
\[
(X,Y,a)=( -1, -1, 0 ), \quad
    ( -1, -3, 0 ),\quad
    ( 0, 1, 0 ),\quad
    ( 1, 2, 0 ),\quad
    ( -1, 0, 0 ),\quad
    ( 1, 4, 1 ),
\]
\[
 ( -1, -4, 1 ),\quad
    ( 1, 1, 0 ),\quad
    ( 1, 3, 0 ),\quad
    ( 1, 0, 0 ),\quad
    ( 0, -1, 0 ),\quad
    ( -1, -2, 0 ).
\]
For none of these is $X$ of the form $p^{11}$.
This completes the proof.
\end{comment}
\end{proof}

\begin{proof}[Proof of Theorem~\ref{thm2a}]
Suppose $n$
is a powerful number such that $\tau(n) \ne 0$ and $P(\tau(n)) \le 11$.
Let $p$ be a prime divisor of $n$.
Thus $p^{m-1} \mid\mid n$, where, as $n$ is powerful,
$m \ge 3$. Now, as $\tau$ is multiplicative,
$\tau(p^{m-1}) \ne 0$ and
$\tau(p^{m-1}) \mid \tau(n)$.
In particular, $P(\tau(p^{m-1})) \le 11$.
By Lemma~\ref{lem:taupm} we have $p=2$ and $m=4$.
Thus $n=8$ as required.
\end{proof}

%---------------------------------------
\section{Proof of Theorem~\ref{thm3}}\label{sec:thm3}
%--------------------------------------

\begin{proof}[Proof of Theorem~\ref{thm3}]
Suppose $\tau(n)=\pm q^\alpha$ where $3 \le q <100$
is a prime and $n \ge 2$. Then $\tau(n)$
is odd, and so $n$ must be an odd square. Thus there is
a prime $p$ and an integer $m \ge 3$
such that $p^{m-1} \mid\mid n$. Hence $\tau(p^{m-1})=\pm q^a$
for some $a \ge 0$. The following lemma completes the proof.
\end{proof}

\begin{lem}
Let $3 \le q <100$ be a prime and $a$ be a nonnegative integer. Let $p$ be a prime and $m \ge 3$
an odd integer. Then $\tau(p^{m-1}) \ne \pm q^a$.
\end{lem}
\begin{proof}
	We argue by contradiction. 
Suppose $m \ge 3$ is the smallest
odd integer such that
\begin{equation}
\tau(p^{m-1})=\pm q^a.
\end{equation}
By Propositions~\ref{prop:pmqb} and ~\ref{prop:4pmqb},
we have $m \ge 7$.
We treat first the case $\tau(p)=0$.
By Lemma~\ref{taup=0},
we see that $\pm q^a=\tau(p^{m-1})$ is either $0$ or a power of $p$.
Thus $p=q<100$, which gives a contradiction
since any $p$ for which $\tau(p)=0$ satisfies
 \eqref{eqn:Derickx}.
Thus $\tau(p) \ne 0$.

Let $\{u_n\}$ be the 
Lucas sequence given in Lemma~\ref{lem:associated}.
It follows from that lemma that $u_n \mid \tau(p^{n-1})$
and $p \nmid u_n$ for all $n \ge 1$.
If $p =q$, then $u_{m}= \pm 1$
which contradicts the Primitive Divisor
Theorem (Theorem~\ref{thm:BHV}), as $m \ge 7$. We conclude that
$p \ne q$.

Next we consider the case $p \mid \tau(p)$.
Then $p \mid \tau(p^n)$ for all $n \ge 1$
by \eqref{eqn:recursion}, and so $p=q$, giving a contradiction.
Thus $p \nmid \tau(p)$.
It follows that $u_n=\tau(p^{n-1})$ for all $n \ge 1$.
Recall that if $k \mid n$ then $u_k \mid u_n$.
By the minimality of $m$ we see that $m \ge 7$ is a prime.
We invoke the Primitive Divisor Theorem again to conclude
that $q \nmid (\alpha-\beta)^2 u_1 u_2 \cdots u_{m-1}$
(in the notation of Lemma~\ref{lem:associated}).
From Theorem~\ref{thm:Carmichael}, $m=m_q \mid (q-1)(q+1)$.
The possible pairs of primes
$(q,m)$ with $3 \le q < 100$ and $m \mid (q-1)(q+1)$ are
$$
\begin{array}{l}
(13, 7), \;
(23, 11), \;
(29, 7), \;
(37, 19), \;
(41, 7),\;
(43, 7),\;
(43, 11), \;
(47, 23), \;
(53, 13),\;
(59, 29),\\
(61, 31), \;
(67, 11),\;
(67, 17),\;
(71, 7), \;
(73, 37),\;
(79, 13),\;
(83, 7),\;
(83, 41),\;
(89, 11),\;
(97, 7).
\end{array}
$$
Fixing any of these pairs $(q,m)$, it remains to solve
$\tau(p^{m-1})=\pm q^a$.
By Lemma~\ref{lem:tauFm}, and the fact that $m$ is prime, 
we see that $(X,Y,a)=(p^{11},\tau(p),a)$
is a solution to the Thue--Mahler equation
\[
	\Psi_m(X,Y)=\pm q^{a}.
\]
We solved these Thue--Mahler equations
using the \texttt{Magma} implementation
of the Thue--Mahler solver described in \cite{GKMS}.
None of the solutions are of the form $(p^{11},\tau(p),a)$.
This completes the proof of Theorem~\ref{thm3}. We illustrate this by providing
more details for  the case $q=83$. Here $m$ is a prime $\ge 7$ dividing $83^2-1=2^3 \times 3
\times 7 \times 41$, and thus the possible pairs $(q,m)$ are $(83,7)$ and
$(83,41)$. For the first pair, the Thue--Mahler equation is
\[
-X^3 + 6 X^2 Y - 5 X Y^2 + Y^3=83^a,
\]
and the solutions are
\[
(X,Y,a)= ( 5, 1, 0 ),\;
    ( -9, -14, 0 ),\;
    ( 2, 3, 0 ),\;
    ( -7, -1, 1 ),\;
    ( 5, 2, 1 ),\;
    ( 0, 1, 0 ),\;
    ( -1, -2, 0 ),
    \]
    \[
    ( -17, -38, 2 ),\;
    ( -8, -13, 1 ),\;
    ( 13, 20, 1 ),\;
    ( 1, 1, 0 ),\;
    ( 4, 13, 0 ),\;
    ( -6, -19, 1 ),\;
    ( -1, 0, 0 ),\;
    ( 21, 25, 2 ),
    \]
    \[
    ( 3, 11, 1 ),\;
    ( -4, 13, 2 ),\;
    ( -1, -3, 0 ),\;
    ( -5, -2, 1 ),\;
    ( 0, -1, 0 ),\;
    ( 17, 38, 2 ),\;
    ( 6, 19, 1 ),\;
    ( 7, 1, 1 ),
    \]
    \[
    ( 1, 0, 0 ),\;
    ( -4, -13, 0 ),\;
    ( 4, -13, 2 ),\;
    ( 9, 14, 0 ),\;
    ( -3, -11, 1 ),\;
    ( 1, 3, 0 ),\;
    ( -1, -1, 0 ),
    \]
    \[
    ( -13, -20, 1 ),\;
    ( -5, -1, 0 ),\;
    ( -21, -25, 2 ),\;
    ( 8, 13, 1 ),\;
    ( 1, 2, 0 ),\;
    ( -2, -3, 0 ).
\]
For the pair
$(q,m)=(83,41)$ the Thue--Mahler equation
is
\begin{multline*}
X^{20} - 210 X^{19} Y + 7315 X^{18} Y^2 - 100947 X^{17} Y^3 + 735471 X^{16} Y^4 -
    3268760 X^{15} Y^5\\ + 9657700 X^{14} Y^6 - 20058300 X^{13} Y^7 + 30421755
    X^{12} Y^8
    - 34597290 X^{11} Y^9 + 30045015 X^{10} Y^{10}\\ - 20160075 X^9 Y^{11} +
    10518300 X^8 Y^{12} - 4272048 X^7 Y^{13} + 1344904 X^6 Y^{14} - 324632 X^5
    Y^{15} \\
    +
    58905 X^4 Y^{16} - 7770 X^3 Y^{17} + 703 X^2 Y^{18} - 39 X Y^{19} + Y^{20}
= \pm 83^{a},
\end{multline*}
and the solutions are
\[
    ( -1, -3, 0 ),\;
    ( -1, -2, 0 ),\;
    ( 1, 2, 0 ),\;
    ( 1, 0, 0 ),\;
    ( -1, 0, 0 ),
\]
\[
    ( 1, 3, 0 ),\;
    ( 0, 1, 0 ),\;
    ( 0, -1, 0 ),\;
    ( 1, 1, 0 ),\;
    ( -1, -1, 0 ).
\]
\end{proof}


\noindent \textbf{Remark.}
The aforementioned Thue--Mahler solver requires
knowledge of the class group and unit group of
the number field defined by the equation $\Psi_{m}(1,Y)=0$;
this number field has degree $\phi(m)/2=(m-1)/2$. Ordinarily, if the
degree is too large, this might not be practical,
or might require assuming the Generalized Riemann Hypothesis. However,
from Lemma~\ref{lem:abelian}, this number field is 
$\Q(\zeta_m)^+$.
For the values of $m$ under consideration (and in fact
for all prime $m \le 67$), the class number $h_m^+$ of
$\Q(\zeta_m)^+$ is known to be $1$; see for example
\cite[Theorem 1]{Linden}. Moreover, if we denote
the unit group of $\Q(\zeta_m)^+$ by $E_m^+$ and the subgroup
of cyclotomic units by $C_m^+$ then $[E_m^+:C_m^+]=h_m^+$; see
\cite[Theorem 8.2]{Wa}. Hence in all our cases, $E_m^+=C_m^+$, and
is generated \cite[Lemma 8.1]{Wa} by $-1$ and $(1-\zeta_m^a)/(1-\zeta_m)$
with $1<a<m/2$. Thus for all values of $m$ under consideration
we know the class group and unit group.



%----------------------------
\section{Concluding remarks}
%-----------------------------

As noted in the introduction, it would likely be extremely challenging computationally to extend, for example, Corollary \ref{thm2} to explicitly find all $n$ with $\tau (n)$ odd and, say,
$$
P(\tau (n)) \leq 17.
$$
The bottleneck in our approach is related to the difficulty involved in classifying the primes $p$ for which $P(\tau (p^2))$ and $P(\tau (p^4))$ are ``small''. 
For larger exponents $m$, finding the $p$ with $P(\tau (p^m))$ bounded appears to be somewhat more tractable.
By way of example,
we may show, by direct application of the Thue-Mahler solver described in  \cite{GKMS}, the following result.
\begin{prop}
The equation
\begin{equation}\label{eqn:6psmooth}
\tau(p^6) \; = \; \pm 3^{b_1} 5^{b_2} 7^{b_3} 11^{b_4} 13^{b_5} 17^{b_6} 19^{b_7} 23^{b_8 } 29^{b_9} 31^{b_{10}} 37^{b_{11}},
\qquad
\text{$p$ prime,}
\qquad \text{$b_i \in \mathbb{Z}$}
\end{equation}
has no solutions.
\end{prop}

This amounts to solving the Thue-Mahler equation
\[
-X^3 + 6 X^2 Y - 5 X Y^2 + Y^3= \pm 3^{b_1} 5^{b_2} 7^{b_3} 11^{b_4} 13^{b_5} 17^{b_6} 19^{b_7} 23^{b_8 } 29^{b_9} 31^{b_{10}} 37^{b_{11}}
\]
and checking to see if any solutions have $X=p^{11}$ for some prime $p$.
Appealing to  \cite{GKMS}, we find that all solutions in coprime integers $X$ and $Y$ have either $\max \{ |X|,|Y| \} < 100$, or satisfy
$$
\begin{array}{r} 
\pm (X,Y) \in  \left\{ (31,105),  (33,107),  (33,109), (41,124), (67,219), (74,115),  (74,117), (76,119),  \right. \\
 \left (83,125), (152,237), (207,-152), (251, 815), (313,62), (359,925), (564,877),  (566,773) \right\}. \\
\end{array}
$$





%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{comment}
%----------------------------
\section{Prime power values of $\tau (n)$}
%-----------------------------

We must treat

$$
F_6(x,y)=-x^3 + 6x^2y - 5xy^2 + y^3 = \pm q^\beta, \; \; q \in \{ 13, 29, 41, 43, 71, 83, 97 \},
$$
We note here that our congruences show that
$$
F_6(p^{11},\tau^2(p)) \equiv 1, 7 \mod{9}.
$$

$$
 -x^5 + 15x^4y - 35x^3y^2 + 28x^2y^3 - 9xy^4 + y^5 =  \pm q^\beta, \; \; q \in \{ 23,  43,  67, 89 \},
$$
$$
x^6 - 21x^5y + 70x^4y^2 - 84x^3y^3 + 45x^2y^4 - 11xy^5 + y^6 =  \pm q^\beta, \; \; q \in \{ 13,  53,  79 \},
$$
$$
x^8 - 36x^7y + 210x^6y^2 - 462x^5y^3 + 495x^4y^4 - 286x^3y^5 + 91x^2y^6 - 15xy^7 + y^8 = \pm q^\beta, \; \; q \in \{ 17,  67 \},
$$
$$
-x^9 + 45x^8y - 330x^7y^2 + 924x^6y^3 - 1287x^5y^4 + 1001x^4y^5 - 455x^3y^6 + 120x^2y^7 - 17xy^8 + y^9 = \pm q^\beta, \; \; q \in \{ 19,  37 \},
$$

We also must treat
$$
F_{2m_0}(x,y)= \pm q^\beta,
$$
for
$$
m_0=11, \; q \in \{ 23, 43 \},
$$
$$
m_0=14, \; q \in \{ 29, 59 \},
$$
$$
m_0=15, \; q \in \{ 31, 61 \},
$$
$$
m_0=18, \; q \in \{ 37, 73 \},
$$
$$
m_0=20, \; q \in \{ 41, 83 \},
$$
$$
m_0 \in \{ 21,23, 29, 30, 33, 35, 36, 39, 41, 44, 48 \},  \; q =m+1=2m_0+1.
$$
Note that every case where $q=m+1$ reduces to (already solved by Bilu-Hanrot-Voutier) Thue equations
$$
F_m(x,y)= \pm q^\delta, \; \; \delta \in \{ 0, 1 \}.
$$
The highest degree form we need to work with is thus degree $20$.

%----------------------------------------
\section{Low hanging fruit}
%----------------------------------------

Note that there are no $E$ over $\mathbb{Q}$ with rational $2$-torsion and conductor $N=2^5 \cdot 691$.

%-------------------------------------------------------------
\section{ Proof of Theorem \ref{thm2a}}
%-------------------------------------------------------------

Let us suppose that $n$ is a powerful number, so that, for each prime $p$ with $p \mod n$, we have that $p^2 \mid n$.

Note that
$$
\tau (2) = -2^3 \cdot 3, \; \tau (2^2) = -2^6 \cdot 23, \; \tau (2^3) = 2^9 \cdot 3 \cdot 5 \cdot 11,
\; \tau (2^4) = 2^{12} \cdot 241, \; \tau(2^5) = -2^{15} \cdot 3^2 \cdot 23 \cdot 29,
$$
and
$$
\tau(2^6) = 2^{18} \cdot 7 \cdot 1471.
$$
\end{comment}

%---------------------------------------------------
\begin{thebibliography}{}
%---------------------------------------------------

%\bibitem{Ab}
%M. Abouzaid,
%\newblock Les nombres de Lucas et Lehmer sans diviseur primitif,
%\newblock {\em J. Th\'eor. Nombres Bordeaux} 18 (2006), 299--313.

\bibitem{BaCrOn}
J. S. Balakrishnan, W. Craig and K. Ono,
\newblock Variations of Lehmer's conjecture for Ramanujan's tau-function,
\newblock {\em J. Number Theory}, to appear.

\bibitem{BaCrOnTs}
J. S. Balakrishnan, W. Craig, K. Ono and W.-L. Tsai,
\newblock Variants of Lehmer's speculation for newforms,
\newblock submitted for publication.

\bibitem{BenHyp}
M. A. Bennett,
\newblock Rational approximation to algebraic numbers of small height : the Diophantine equation $|ax^n-by^n|=1$,
\newblock {\em J.\ Reine Angew.\ Math.} 535 (2001), 1--49.

\bibitem{BenS}
M.\ A.\ Bennett and C.\ Skinner,
\newblock Ternary Diophantine equations via Galois representations and modular forms,
\newblock {\em Canad.\ J.\ Math.} 56 (2004), no.\ 1, 23--54.

\bibitem{Barr}
C. F. Barros,
\newblock On the Lebesgue-Nagell equation and related subjects,
\newblock PhD thesis, University of Warwick, 2010.

\bibitem{BDMS}
M. A. Bennett, S. Dahmen, M. Mignotte and S. Siksek,
\newblock Shifted powers in binary recurrence sequences,
\newblock {\em Math. Proc. Cambridge Philos. Soc.}  158 (2015), 305--329.

\bibitem{BHV}
Y. Bilu, G. Hanrot and P. Voutier,
\newblock Existence of primitive divisors of Luca and Lehmer numbers,
\newblock {\em J.\ Reine Angew.\ Math.} 539 (2001), 75--122.

\bibitem{magma}
W.\ Bosma, J.\ Cannon and C.\ Playoust,
\newblock The Magma Algebra System I: The User Language,
\newblock {\em J.\ Symb.\ Comp.} {\bf 24} (1997), 235--265. (See also \url{http://magma.maths.usyd.edu.au/magma/})
%W. Bosma, J. Cannon, et al.,
%\emph{Magma} computer algebra system, available from the website
%from \url{http://magma.maths.usyd.edu.au/}, 2005.

\bibitem{Bug-BLMS}
Y. Bugeaud,
On the greatest prime factor of $ax^m-by^n$ II,
		\emph{Bull. London Math. Soc.}  32 (2000), no. 6, 673–678.

\bibitem{BuGy}
Y. Bugeaud and K. Gy\H{o}ry,
\newblock Bounds for the solutions of Thue-Mahler equations and norm form equations,
\newblock {\em Acta Arith.} 74 (1996), 273--292.

%\bibitem{BG-sunit}
%Y. Bugeaud and K. Gy\H{o}ry,
%\newblock Bounds for the solutions of the unit equation,
%\newblock {\em Acta Arith.} 74 (1996), 1--14.

%\bibitem{BG-tm}
%Y. Bugeaud and K. Gy\H{o}ry,
%\newblock Bounds for the solutions of Thue-Mahler equations and norm form equations,
%\newblock {\em Acta Arith.} 74 (1996), 273--292.

%\bibitem{BLMS}
%Y. Bugeaud, F. Luca, M. Mignotte and S. Siksek,
%\newblock Fibonacci numbers at most one away from a perfect power,
%\newblock  {\em  Elem. Math.} 63 (2008), no. 2, 65--75.

%\bibitem{BLMS2}
%Y. Bugeaud, F. Luca, M. Mignotte and S. Siksek,
%\newblock Almost powers in the Lucas sequences,
%\newblock {\em J. Th\'eor. Nombres Bordeaux} 20 (2008), 555--600.


\bibitem{BMS}
Y. Bugeaud, M. Mignotte and S. Siksek,
\newblock Classical and modular approaches to exponential Diophantine equations I. Fibonacci and Lucas perfect powers,
\newblock {\em Ann. of Math.} 163 (2006), 969--1018.

%\bibitem{BMS2}
%Y. Bugeaud, M. Mignotte and S. Siksek,
%\newblock Classical and modular approaches to exponential Diophantine equations II. the Lebesgue-Nagell equation,
%\newblock {\em Compositio Math.} 142 (2006), 31--62.

%\bibitem{integral}
%Y.\  Bugeaud, M.\ Mignotte, S.\ Siksek, M.\ Stoll and Sz.\ Tengely,
%\newblock Integral Points on Hyperelliptic Curves,
%\newblock {\em Algebra \& Number Theory} 2 (2008), 859--885.

\bibitem{Car}
R. D. Carmichael,
\newblock On the numerical factors of the arithmetic forms $\alpha^n \pm \beta^n$,
\newblock {\em Ann. Math.} 15 (1913), 30--70.

%\bibitem{Cohen}
%H.\ Cohen,
%\emph{Advanced Topics in Computational Algebraic Number Theory},
%GTM \textbf{193}, Springer--Verlag, 2000.

\bibitem{DM}
H.\ Darmon and L.\ Merel,
\newblock Winding quotients and some variants of Fermat's Last Theorem,
\newblock {\em J. Reine Angew.\ Math.} 490 (1997), 81--100.

%\bibitem{DavidI} A.\ David,
%\newblock Caract\`ere d'isog\'enie et critt\'{e}res d'irr\'{e}ductibilit\'e,
%\newblock {\it arXiv:1103.3892v2 [math.NT]}.

\bibitem{Deligne}
P. Deligne,
\newblock La conjecture de Weil I,
\newblock {\em Inst. Hautes \'Etudes Sci. Publ. Math.} 43 (1974), 273--307.

%\bibitem{DV}
%L.\ Demb{\'e}l{\'e} and J.\ Voight,
%\newblock Explicit methods for {H}ilbert modular forms,
%\newblock In \emph{Elliptic curves, {H}ilbert modular forms and {G}alois
%  deformations}, Adv. Courses Math. CRM Barcelona, pages 135--198.
%  Birkh\"auser/Springer, Basel, 2013.

\bibitem{DeJa}
S. Dembner and V. Jain,
\newblock Hyperelliptic curves and newform coefficients,
\newblock submitted for publication.

\bibitem{DeHoZe}
M. Derickx, M. Van Hoeij and Jinxiang Zeng,
\newblock Computing Galois representations and equations for modular curves $X_H(\ell)$,
\newblock arXiv:1312.6819v2.


%\bibitem{Freitas} N.\ Freitas,
%{\em Some Generalized Fermat-type Equations via $\Q$-curves and
%Modularity}, PhD thesis, University of Barcelona, 2012.

%\bibitem{FreitasRecipes}
%N.\ Freitas,
%{\em Recipes for Fermat type equations of the form $x^r+y^r=Cz^p$},
%preprint, 2013.

%\bibitem{FLS}
%N.\ Freitas, B.\ Le Hung and  S.\ Siksek,
%\newblock Elliptic curves over real quadratic fields are modular,
% \newblock {\em Invent. Math.} 201 (2015), 159--206.

%\bibitem{FS}
%N.\ Freitas and S.\ Siksek,
%\newblock The Asymptotic Fermat's Last Theorem for Five-Sixths of Real Quadratic Fields,
%\newblock {\em Compositio Math.} 151 (2015), 1395--1415.

%\bibitem{FS2}
%N.\ Freitas and S.\ Siksek,
%\newblock Criteria for the irreducibility of mod $p$ representations of Frey curves,
%\newblock {\em J.\ Th\'{e}or.\ Nombres Bordeaux},  27 (2015), 67--76.

\bibitem{GKMS}
A. Gherga, R. von K\"anel, B. Matschke and S. Siksek,
\newblock Efficient resolution of Thue--Mahler equations,
to appear.

%\bibitem{GY}
%K. Gy\H{o}ry and K. Yu,
%\newblock Bounds for the solutions of $S$-unit equations and decomposable form equations,
%\newblock {\em Acta Arith.} 123 (2006), 9--41.

\bibitem{HaMa}
M. Hanada and R. Madhukara,
\newblock Fourier coefficients of level $1$ Hecke eigenforms,
\newblock submitted for publication.

%\bibitem{Iv}
%W. Ivorra,
%\newblock Courbes elliptiques sur $\mathbb{Q}$, ayant un point d'ordre $2$ rationnel sur $\mathbb{Q}$, de conducteur $2^Np$,
%\newblock {Dissertationes Math.} 429 (2004), 1--55.

\bibitem{KO}
A. Kraus and J. Oesterl\'e,
\newblock Sur une question de B.\ Mazur,
\newblock {\em Math. Ann.} 293 (2002), 259--275.

%\bibitem{LaTr}
%S. Lang and H. Trotter,
%\newblock {\it Frobenius distributions in $\mbox{GL}_2$ Extensions},
%\newblock Lecture Notes in Mathematics, Vol. 504 (Springer-Verlag, 1976), 274pp.

\bibitem{Leh}
D. H. Lehmer,
\newblock The vanishing of Ramaujan's function $\tau (n)$,
\newblock {\em Duke Math. J.} 14 (1947), 429--433.

\bibitem{Lenstra}
H. W. Lenstra,
\newblock Algorithms in algebraic number theory,
\emph{Bull. Amer. Math. Soc.} \textbf{26} (1992), 211--244.

\bibitem{Linden}
F.\ J.\ van der Linden,
\newblock Class number computations of real abelian number fields,
\newblock {\em Mathematics of Computation} \textbf{39} (1982), 693--707.

\bibitem{LuMaSt}
F. Luca, S. Mabaso and P. Stanica,
\newblock On the prime factors of the iterates of the Ramanujan $\tau$-function,
\newblock {\em Proc. Edinburgh Math. Soc.}, to appear.


\bibitem{LuSh}
F. Luca and I. E. Shparlinski,
\newblock Arithmetic properties of the Ramanujan function,
\newblock {\em Proc. Indian Acad. Sci. (Math. Sci.)} 116 (2006), 1--8.


\bibitem{LyRo0}
N. Lygeros and O. Rozier,
\newblock  A new solution to the equation $\tau (p) \equiv 0 \mod{p}$,
\newblock {\em J. Integer Sequences} 13 (2010),  Article 10.7.4.

\bibitem{LyRo}
N. Lygeros and O. Rozier,
\newblock  Odd prime values of the Ramanujan tau function,
\newblock {\em Ramanujan Journal} 32 (2013),  269--280.

\bibitem{Momose}
F.\ Momose,
\newblock Isogenies of prime degree over number fields,
\newblock {\em Compositio Math.} 97 (1995), 329--348.

\bibitem{Mor}
L. J. Mordell,
\newblock On Mr. Ramanujan's empirical expansions of modular functions,
\newblock {\em Proc. Cambridge Phil. Soc.} 19 (1917), 117--124.

\bibitem{MuMu}
M. Ram Murty and V. Kumar Murty,
\newblock Odd values of Fourier coefficients of certain modular forms,
\newblock {\em Int. J. Number Theory} 3 (2007), 455-470.

\bibitem{MuMuSh}
M. Ram Murty, V. Kumar Murty and T. N. Shorey,
\newblock Odd values of the Ramanujan $\tau$-function,
\newblock {\em Bulletin Soc. Math. France} 115 (1987), 391--395.

%\bibitem{PARI2}
%The PARI~Group,
%\newblock PARI/GP version {\tt 2.7.0},
%\newblock Bordeaux, 2014, \url{http://pari.math.u-bordeaux.fr/}.

%\bibitem{Nag}
%T. Nagell,
%\newblock Sur l'impossibilit\'e de quelques \'equations \`a deux ind\'etermin\'ees,
%\newblock {\em Norsk. Mat. Forensings Skrifter} 13 (1923), 65--82.


\bibitem{OnTa}
K. Ono and Y. Taguchi,
\newblock $2$-adic properties of certain modular forms and their applications to arithmetic functions,
\newblock {\em Int. J. Number Theory} 1 (2005), 75--101.

%\bibitem{Voutier}
%P. M. Voutier, 
%\newblock An effective lower bound for the height of algebraic numbers, 
%\newblock \emph{Acta Arith.} 74 (1996), 81--95.

\bibitem{Ram}
S. Ramanujan,
\newblock On certain arithmetical functions,
\newblock {\em Trans. Camb. Philos. Soc.} 22 (1916), 159--184.

\bibitem{Ser}
J.-P.\ Serre,
\newblock Divisibilit\'e de certaines fonction arithmetiques,
\newblock {\em L'Ens. Math.} 22 (176), 227--260.

%\bibitem{Serre}
%J.-P.\ Serre,
%\emph{Abelian $\ell$-adic representations of elliptic curves},
%Addison-Wesley, 1989.

%\bibitem{ShSt}
%T. Shorey and C. L. Stewart,
%\newblock Pure powers in recurrence sequences and some related Diophantine equations,
%\newblock {\em J. Number Theory} 27 (1987), 324--352.

\bibitem{Siksek}
S. Siksek,
\newblock The Modular Approach to Diophantine Equations,
in \emph{Explicit Methods in Number Theory: Rational Points and Diophantine
Equations}, ed. Belabas et al.,
Panoramas et synth\`{e}ses \textbf{36}, 2012.

%\bibitem{SilvermanII}
%J.\ H.\ Silverman,
%\emph{Advanced topics in the arithmetic of elliptic curves},
%    {Graduate Texts in Mathematics}
%   \textbf{151},
% {Springer-Verlag, New York}, {1994}.

\bibitem{Smart}
N.\ P.\ Smart,
\newblock The Algorithmic Resolution of Diophantine Equations,
\newblock Cambridge University Press, 1998.

\bibitem{St13}
C. L. Stewart,
\newblock On divisors of Lucas and Lehmer numbers,
\newblock {\em Acta Math.} 211 (2013), 291--314.

%\bibitem{St}
%C. L. Stewart,
%\newblock On some Diophantine equations and related linear recurrence sequences,
%\newblock {\em S\'em. de Th\'eorie des Nombres}, Paris 1980--1981, Birkhauser, Boston, 1982.

\bibitem{SwD}
H.\ P.\ F.\ Swinnerton-Dyer, 
\newblock On $\ell$-adic representations and congruences for coefficients of modular forms, 
page 1--55 of  W.\ Kuyk and J.-P.\ Serre (eds.), Modular functions of one variable, III, Lecture Notes in Mathematics 350, 1973. 

\bibitem{Wa}
L.\ C.\ Washington,
Introduction to Cyclotomic Fields, Springer Verlag,
1982.
\end{thebibliography}

\end{document}


In this section we prove the following theorem.
\begin{thm}\label{thm:case4}
Equation \eqref{case4} has no solutions.
\end{thm}

Clearly, in \eqref{case4}, the exponent $\alpha$ can only
take the values $0$ or $1$.
It will be convenient to rewrite \eqref{case4} as
\begin{equation}\label{eqn:convenient}
z^2-5p^{22}= \kappa \cdot 4 \cdot 5^a \cdot 11^b, \qquad z=2 \tau(p)^2-3p^{11},
\qquad
a=0,~1, \qquad \kappa =-1,~1,
\end{equation}
where $p$ is a prime.
We checked that \eqref{eqn:convenient} has no solutions with $p<200$,
so we may suppose $p>200$.

Equation \eqref{eqn:convenient}
can be viewed as $5 (p^{2})^{11}+ (\kappa \cdot 4 \cdot 5^a \cdot 11^b) \cdot
1^{11} =
z^2$
and so is an equation with signature $(11,11,2)$.
To this we may attach a Frey-Hellegouarch curve following Bennett and Skinner \cite{BenS}.
We shall attach
to a solution of \eqref{eqn:convenient} the Frey-Hellegouarch curve
\[
E_p \; : \; Y^2= X \left( X^2 + z X + \frac{1}{4} (z^2-5 p^{22}) \right).
\]
If $p \equiv 3 \mod{4}$ then this is precisely
the Frey curve in \cite{BenS}, and if $p \equiv 1 \mod{4}$
then this is  a quadratic twist by $-1$ of the Frey curve in
\cite{BenS}.
We apply
the recipes in \cite{BenS}, making almost trivial modifications to account
for the possible twisting. We find that
the model $E_p$ is minimal with discriminant
\[
\Delta=2^4 \cdot 5^{2a+1} \cdot 11^{2b} \cdot p^{22},
\]
and conductor
\[
N=2^u \cdot 5^{a+1} \cdot 11^v \cdot p
\]
where
\[
u=\begin{cases}
3 & \text{if $p \equiv 3 \mod{4}$}\\
4 & \text{if $p \equiv 1 \mod{4}$}
\end{cases},
\qquad \qquad
v=\begin{cases}
0 & \text{if $b=0$}\\
1 & \text{if $b \ne 0$}.
\end{cases}
\]
Here the congruences \eqref{mod2} were needed to compute
$u$.
Applying Ribet's level-lowering theorem (c.f. \cite{BenS}
or \cite{Siksek})
we find that
\begin{equation}\label{eqn:mod11new}
\overline{\rho}_{E_p,11} \sim
\overline{\rho}_{f,\varpi}
\end{equation}
where $f$ is a cuspidal newform
of weight $2$ and level
\[
N^\prime=2^u \cdot 5^{a+1} \cdot 11^w, \qquad \qquad
w=\begin{cases}
0 & \text{if $11 \mid b$}\\
1 & \text{if $11 \nmid b$}.
\end{cases}
\]
It follows that
\begin{equation}\label{eqn:levels}
\begin{cases}
N^\prime \in \{ 40,~80,~440,~880\} & \text{if $a=0$}\\
N^\prime \in \{200,~400,~2200,~ 4400\} & \text{if $a=1$}.
\end{cases}
\end{equation}
There are in total $120$ weight $2$ newforms belonging to these levels.
We let $f$ denote one of them.
We will adapt the approach in Section~\ref{sec:case2} to
our present Frey curve. Let $3 \le \ell < 200$ with $\ell \ne 5$.
Unlike in Section~\ref{sec:case2}, we now allow $\ell=3$ but not $\ell=5$, as the Frey curve $E_p$
has good reduction at $3$, but bad reduction at $5$.
\begin{equation}\label{eqn:cA}
\cA_{\ell}=\{(s,t) \; : \; s,~t \in \F_\ell, \quad s \not \equiv 0 \mod{\ell},
\qquad (2t^2-3s^{11})^2-5s^{22} \not \equiv 0 \mod{\ell}
\}.
\end{equation}
Note that $(p,\sigma(p)) \equiv (s,t) \mod{\ell}$ for some $(s,t) \in \cA_{\ell}$.
Let
\[
\cB_{\ell}=
\begin{cases}
\cA_{\ell} & \ell \ne 3,~7,~23;\\
\{(s,t) \in \cA_{3} \; : \; t \equiv s+1 \mod{3} \} & \ell=3;\\
\{(s,t) \in \cA_{7} \; : \; t \equiv s(s^{3}+1) \mod{7} \} & \ell=7;\\
\{(s,0) \; : \; s \in \F_{23}^*\setminus (\F_{23}^*)^2\} \cup
\{(s,t) \; : \; s \in (\F_{23}^*)^2,~t=2,~-1\}  & \ell=23.
\end{cases}
\]
It follows from \eqref{mod3}, \eqref{mod7}, \eqref{mod23}
that $(p,\sigma(p)) \equiv (s,t) \mod{\ell}$ for
some $(s,t) \in \cB_{\ell}$. Thus
$a_\ell(E_p)=a_\ell(E_{s,t})$ where
\[
E_{s,t}/\F_\ell \; : \; Y^2=X(X^2+(2 t^2-3s^{11}) X
+((2t^2-3s^{11})^2-5s^{22})/4).
\]
Now let $f$ be the newform for which \eqref{eqn:mod11new}
holds, and let its $q$-expansion be given by \eqref{eqn:newform}.
As before, write $K=\Q(c_0,c_1,\dotsc)$. From \eqref{eqn:mod11new}
we obtain $a_\ell(E_p) \equiv c_\ell \mod{\varpi}$
and thus $\Norm(a_\ell(E_p)-c_\ell) \equiv 0 \mod{11}$.
Let
\[
\cC_\ell(f) \; = \; \{ (s,t) \in \cB_\ell \; : \;
\Norm(a_\ell(E_{s,t})-c_\ell(f)) \equiv 0 \mod{11}\}.
\]
It follows that $(p,\tau(p)) \equiv (s,t) \mod{\ell}$
for some $(s,t) \in \cC_\ell(f)$.
\begin{lem}\label{lem:intersect2}
Let
\[
\cL=\{3,\; 7,\; 13,\; 19,\; 23,\; 37,\; 61,\; 67,\; 89,\; 199\}.
\]
For $\ell \in \cL$,  $\kappa \in \{-1,1\}$  and $a \in \{0,1\}$
let
\begin{multline*}
\mathcal{D}_{\ell}(f,\kappa,a)=
\{ \beta \in \{0,1,\dotsc,131\} \quad : \quad
11^\beta
\equiv  \frac{(2t^2-3s^{11})^2-5 s^{22}}{\kappa \cdot 4 \cdot 5^a} \mod{\ell} \\
\text{for some $(s,t) \in \cC_{\ell}(f)$}\},
\end{multline*}
and
\[
\mathcal{D}(f,\kappa,a)=\bigcap_{\ell \in \cL} \mathcal{D}_{\ell}(f,\kappa,a).
\]
Let $(p,b)$ be a solution to \eqref{eqn:convenient} and let $f$
be a newform such that \eqref{eqn:mod11new} is satisfied. Then
there is some $\beta \in \mathcal{D}(f,\kappa,a)$ such that
$b \equiv \beta \mod{132}$.
\end{lem}
\begin{proof}
From the above we know that $(p,\tau(p)) \equiv (s,t) \mod{\ell}$
for some $(s,t) \in \cC_{\ell}(f)$. However,
\[
\frac{(2t^2-3s^{11})^2-5 s^{22}}{\kappa \cdot 4 \cdot 5^a}
\equiv
\frac{z^2-5 p^{22}}{\kappa \cdot 4 \cdot 5^a} \equiv 11^b \mod{\ell}
\]
by \eqref{eqn:convenient}.
The primes in $\cL$ are precisely the primes $3 \le \ell <200$
with $\ell \ne 5$, $11$, and such that the multiplicative
order of $11$ modulo $\ell$ divides $132$.
%The multiplicative order of $11$ modulo $3$, $7$, $23$ and $88$
%is respectively $2$, $3$, $22$, $22$, and their least
%common multiple is $66$.
It follows that there is some $\beta \in \mathcal{D}(f,\kappa,a)$ such that $b \equiv \beta \mod{132}$.
\end{proof}

\begin{lem}\label{lem:b=0}
Equation~(\ref{eqn:convenient}) has no solutions with
$\kappa=-1$ or $b>0$.
%If $(p,b,a)$ is a solution with $\kappa=1$ then
%$44 \mid b$.
\end{lem}
\begin{proof}
We computed $\mathcal{D}(f,\kappa,a)$ for all possible choices
for $\kappa$, $a$ and for $f$ of weight $2$ and level
given by one of the possibilities in \eqref{eqn:levels}.
We found that $\mathcal{D}(f,\kappa,a)=\emptyset$
for all possible choices of $\kappa$, $a$, $f$
except the following two possibilities,
\begin{enumerate}
\item[(i)] $\kappa=1$, $a=0$,
and $f$ is the rational newform
corresponding to the elliptic curve \texttt{40a1};
\item[(ii)] $\kappa=1$, $a=1$,
and $f$ is the rational newform
corresponding to the elliptic curve \texttt{400d1}.
\end{enumerate}
For these two possiblities $\mathcal{D}(f,\kappa,a)=\{0,44,88\}$.
By Lemma~\ref{lem:intersect2},
$\kappa=1$ and $b \equiv 0$, $44$ or $88 \mod{132}$. In particular,
$11 \mid b$. Let $E$ be the elliptic curve \texttt{40a1}
in case (i) and \texttt{400d1} in case (ii).
Then $\overline{\rho}_{E_p,11} \sim \overline{\rho}_{E,11}$.
Suppose $b \ne 0$, so that $E_p$ has multiplicative
reduction at $11$, though $E$ has good reduction at $11$.
As before (by a result of Kraus and Oesterl\'{e} \cite{KO}),
$\pm (11+1) \equiv a_{11}(E) \mod{11}$. However,
for both possibilities for the elliptic curve $E$,
we have $a_{11}(E)=4$ giving a contradiction.
We conclude that $b=0$.
\end{proof}

\noindent \textbf{Remark.}
We cannot eliminate the possiblities (i), (ii) appearing
in the proof of the lemma by adding more primes to $\cL$.
To see this, consider (i), whence $\kappa=1$, $a=0$
and $f$ the rational newform corresponding to the
elliptic curve \texttt{40a1}, which has the model
\[
E \; : \; Y^2=X(X^2+3X+1).
\]
We observe directly from the definition of $\cB_\ell$
that the three pairs
$(-1,0)$, $(-11^2,0)$, $(-11^4,0)$ belong to $\cB_\ell$
for any prime $\ell \ne 2$, $5$, $11$. The elliptic curve
$E_{-1,0}/\F_\ell$ is simply the reduction modulo $\ell$ of $E$,
and so $a_\ell(E_{-1,0})=a_\ell(E)=c_\ell(f)$. Thus $(-1,0) \in \cC_\ell(f)$.
We note that $E_{-11^2,0}$ and $E_{-11^4,0}$ are isomorphic
to $E_{-1,0}$. Thus $(-11^2,0)$, $(-11^4,0)$ also belong to $\cC_\ell(f)$.
From the three pairs $(s,t)=(-1,0)$, $(-11^2,0)$, $(-11^4,0)$
we see that $\beta=0$, $44$, $88$ belongs to $\mathcal{D}_\ell(f,1,0)$
regardless of $\ell$.
The difficult is essentially
caused by the equality
$(2\times 0 - 3 \times (-1)^{11})^2-5 \times (-1)^{22}=4 \cdot 5^0 \cdot
11^{0}$. A similar explanation holds for (ii).

\begin{proof}[Proof of Theorem~\ref{thm:case4}]
By Lemma~\ref{lem:b=0}, we have $\kappa=1$ and $b=0$
in \eqref{eqn:convenient}.
We consider the cases $a=0$, $a=1$
separately. Suppose first $a=0$. Then
\[
z^2-5 p^{22}= 4.
\]
Write $\varepsilon=(1+\sqrt{5})/2$.
Then $(\lvert z\rvert +p^{11} \sqrt{5})/2$ is
a positive unit in $\Z[\varepsilon]$ with norm $+1$.
Hence
\[
\frac{z+p^{11} \sqrt{5}}{2} \; = \; \pm \varepsilon^{2n}, \qquad
\varepsilon=(1+\sqrt{5})/2.
\]
for some $n \in \Z$. Thus
\[
p^{11}= \frac{\varepsilon^{2n}-\overline{\varepsilon}^{2n}}{\sqrt{5}}
=F_{2n}
\]
where $F_n$ denotes
the $n$-th Fibonacci number.
By \cite{BMS}
the only perfect powers in the Fibonacci sequence are
$0$, $1$, $8$ and $144$, giving a contradiction. Alternatively,
$F_{2n}=F_n L_n$ where $L_n$ is the $n$-th Lucas number.
From the identity $L_n^2-5 F_n^2=4 \cdot (-1)^n$ we
see that $\gcd(F_n,L_n)=1$ or $2$. Thus $F_n=1$ or $L_n=1$
quickly leading to a contradiction.

Next we suppose that $a=1$. Then $5 \mid\mid z$ and we
write $z=5w$. Hence
\[
5w^2-p^{22}=4.
\]
It follows that there is an integer $n$ such that
\[
p^{11}=\varepsilon^n+\overline{\varepsilon}^n=L_n
\]
where $L_n$ denotes the $n$-th Lucas number.
By \cite{BMS},
the only perfect powers in the Lucas sequence are $1$ and $4$,
again giving a contradiction.
\begin{comment}
By Lemma~\ref{lem:44midb} we need only consider
equation \eqref{eqn:convenient} with $b=44c$
and $\kappa=1$. Suppose first that $a=0$.
Thus we have
\[
z^2-5 p^{22}=4 \cdot 11^{44 c}.
\]
Hence
\[
(\lvert z \rvert +2 \cdot 11^{22c})(\lvert z \rvert - 2 \cdot 11^{22c})=5
p^{22}.
\]
The factors are on the left are coprime and positive, and the first
is larger than the second. Thus
\begin{equation}\label{eqn:fact1}
\lvert z \rvert +2 \cdot 11^{22c}=p^{22}, \qquad
\lvert z \rvert - 2 \cdot 11^{22c}=5
\end{equation}
or
\begin{equation}\label{eqn:fact2}
\lvert z \rvert +2 \cdot 11^{22c}=5 p^{22}, \qquad
\lvert z \rvert - 2 \cdot 11^{22c}=1
\end{equation}
Suppose first that \eqref{eqn:fact1} holds. Then
\[
4 \cdot 11^{22c}= p^{22}-5,
\]
which impossible modulo $3$. Hence we consider \eqref{eqn:fact2},
which leads to
\[
5 p^{22}-4 \cdot 11^{22c}=1.
\]
Bennett \cite{BenHyp} showed that an equation of the form
\[
\lvert A x^n-B y^n \rvert=1
\]
with $A$, $B$ non-zero integers and $n \ge 3$ has at most
one solution $(x,y)$ in positive integers. The equation
$\lvert 5 x^{11}-4 y^{11} \rvert=1$ has the solutions
$(1,1)$ and $(p^2,11^{2c})$, giving a contradiction.

Next we consider $a=1$. Thus our equation becomes
\[
(2 \tau(p)^2-3 p^{11})^2-5p^{22} = 4 \cdot 5 \cdot 11^{44c}.
\]
Note that $11^{44c}\equiv p^{22} \equiv 1 \mod{23}$. Thus
\[
(2 \tau(p)^2 -3 p^{11})^2 \equiv 2 \mod{23}.
\]
However, from \eqref{mod23}
we see $(\tau(p),p^{11}) \equiv (0,-1)$, $(2,1)$ or $(-1,1)
\mod{23}$.
We conclude that $(\tau(p),p^{11}) \equiv (2,1) \mod{23}$
as it is the only possibility consistent with the above.
Write $w=(2 \tau(p)^2-3 p^{11})/5 \in \Z$. Then
\[
4 \cdot 11^{44c} - 5 w^2=-p^{22}.
\]
Let $K=\Q(\sqrt{5})$ and $\varepsilon=(1+\sqrt{5})/2$.
Then
\[
2 \cdot 11^{22c}+w \sqrt{5}=\varepsilon^{r} \cdot \gamma^{11}
\]
with $\gamma \in \OO_K$.
\end{comment}
\end{proof}
\begin{comment}
We may now suppose that $11 \nmid b$ and so $N^\prime=2^5 \times 3 \times 11$.
There are $14$ conjugacy class of newforms of weight $2$ and
level $N^\prime$, and $10$ of these are rational.
We will denote the $10$ rational
newforms by $f_1,f_2,\dotsc,f_{10}$,
and $f_{11},f_{12},f_{13}$, $f_{14}$ will denote fixed
representatives of the other four conjugacy classes of newforms.
In what follows, $f$ will denote
one of these $14$ newforms so that \eqref{eqn:mod11} holds,
and the $c_j$ will denote its coefficients as in \eqref{eqn:newform}.
Moreover, $K=\Q(c_1,c_2,\dotsc)$.

From \eqref{eqn:mod11norm} we have that $(p,\sigma(p)) \equiv (s,t)
\mod{\ell}$ for some $(s,t) \in \cC_{\ell}(f)$.
\begin{lem}\label{lem:intersect}
Let
\[
\cL=\{5,~7,~23,~89\}.
\]
For $\ell \in \cL$ and $\kappa \in \{-3,3\}$ let
\[
\mathcal{D}_{\ell}(f,\kappa)=\{ \beta \in \{0,1,\dotsc,65\} \; : \;
\text{$\kappa \cdot 11^\beta
\equiv t^2-s^{11} \mod{\ell}$ for some $(s,t) \in \cC_{\ell}(f)$}\},
\]
and
\[
\mathcal{D}(f,\kappa)=\bigcap_{\ell \in \cL} \mathcal{D}_{\ell}(f,\kappa).
\]
Let $(p,b)$ be a solution to \eqref{eqn:BS3} and let $f$
be a newform such that \eqref{eqn:mod11} is satisfied. Then
there is some $\beta \in \mathcal{D}(f,\kappa)$ such that
$b \equiv \beta \mod{66}$.
\end{lem}
\begin{proof}
From the above we know that $(p,\tau(p)) \equiv (s,t) \mod{\ell}$
for some $(s,t) \in \cC_{\ell}(f)$. Now,
$\kappa \times 11^b=\tau(p)^2-p^{11} \equiv t^2 -s^{11} \mod{\ell}$.
The multiplicative order of $11$ modulo $5$, $7$, $23$ and $89$
is respectively $1$, $3$, $22$, $22$, and their least
common multiple is $66$. It follows that there is some
$\beta \in \mathcal{D}(f,\kappa)$ such that $b \equiv \beta \mod{66}$.
\end{proof}
\begin{proof}[Proof of Theorem~\ref{case2}]
We wrote a short \texttt{Magma}
script which computed $\mathcal{D}(f,\kappa)$ for $f=f_1,\dotsc,f_{14}$
and $\kappa=-3$, $3$. In all cases we found
$\mathcal{D}(f,\kappa)=\emptyset$, except for $\kappa=-3$
and the rational newform $f$ corresponding to the
elliptic curve with Cremona label \texttt{1056d1},
when $\mathcal{D}(f,-3)=\{0,22,44\}$. It follows
from Lemma~\ref{lem:intersect}
that $11 \mid b$.
This contradicts Lemma~\ref{lem:not11}.
\end{proof}

%------------------------------
\section{Equation \eqref{case2}}\label{sec:case2}
%-----------------------------
\begin{thm}\label{thm:case2}
Equation \eqref{case2} has no solutions.
\end{thm}
We checked that \eqref{case2} has no solutions with $p<100$,
so we may suppose $p>100$.
By the lemmas in Section~\ref{sec:congruences} we reduce to the two
equations
\begin{equation}\label{eqn:BS1}
\tau(p)^2-p^{11}=\pm 11^b
\end{equation}
and
\begin{equation}\label{eqn:BS2}
\tau(p)^2-p^{11}=\pm 3 \cdot 11^b.
\end{equation}
Here $p$ is a prime $>100$ and $b$ is a non-negative integer.
The equation
\[
x^n+11^\beta y^n=z^2, \qquad \beta \ge 0, \qquad  \gcd(x,y,z)=1, \qquad \text{$n \ge 11$
prime}
\]
has no solutions with $xy \ne 0$, $\pm 1$, by the work
of Bennett and Skinner \cite[Theorem 1.3]{BenS} for $n \nmid \beta$ and
Darmon
and Merel \cite{DM}
for $n \mid \beta$. Thus \eqref{eqn:BS1} has no solutions,
and we focus on \eqref{eqn:BS2}.

Following Bennett and Skinner \cite{BenS}, we associate to a solution to
\eqref{eqn:BS2} the Frey curve
\[
E_p \; : \;
\begin{cases}
Y^2=X(X^2+2 \tau(p) X +p^{11}) & \text{if $p \equiv 3 \mod{4}$},\\
Y^2=X(X^2+2\tau(p) X + \tau(p)^2-p^{11}) & \text{if $p \equiv 1 \mod{4}$}.\\
\end{cases}
\]
Let
\[
N=\begin{cases}
2^5 \times 3 \times 11 \times p & \text{if $b >0$}\\
2^5 \times 3 \times p & \text{if $b=0$},
\end{cases}
\qquad
N^\prime=\begin{cases}
2^5 \times 3 \times 11 & \text{if $11 \nmid b$}\\
2^5 \times 3 & \text{if $11 \mid b$},
\end{cases}
\]
It follows from the recipes of Bennett and Skinner \cite{BenS}
that $E_p$ has conductor $N$, and that there is a
normalized newform
\begin{equation}\label{eqn:newform}
f=q+\sum_{n=1}^\infty c_n q^n
\end{equation}
of weight $2$ and level $N^\prime$
and a prime $\varpi \mid 11$ in the integers of $K=\Q(c_1,c_2,\dotsc)$
so that
\begin{equation}\label{eqn:mod11}
\overline{\rho}_{E_p,11} \sim \overline{\rho}_{f,\varpi}.
\end{equation}
\begin{lem}\label{lem:not11}
Equation \eqref{eqn:BS2} has no solutions with $11 \mid b$.
\end{lem}
\begin{proof}
Suppose first that $11 \mid b$ but $b \ne 0$.
Then $E_p$ has multiplicative reduction at $11$.
Here $N^\prime=96$. There are two newforms of weight $2$ and level $96$.
These have rational eigenvalues and
correspond to elliptic curves $F_1=96A1$ and $F_2=96B1$ in Cremona's notation.
Thus $\overline{\rho}_{E_p,11} \sim \overline{\rho}_{F_j,11}$ where
$j=1$, $2$.
Comparing the trace of Frobeninus at $11$ in the two representations
we have (c.f.\ Kraus and Oesterl\'{e} \cite{KO}) that
$\pm (11+1) \equiv a_{11}(F_j) \mod {11}$.
However, $a_{11}(F_j)=4$, $-4$ according to whether
$j=1$, $2$, giving a contradiction.

Thus we may suppose $b=0$. Thus $\tau(p)^2-p^{11}=\pm 3$.
We consider this modulo $23$ using \eqref{mod23}.
If $p$ is a quadratic non-residue
modulo $23$, then $p^{11} \equiv -1 \mod{23}$ and $\tau(p) \equiv 0
\mod{23}$ giving a contradiction.
If $p$ is a quadratic residue
modulo $23$, then $p^{11} \equiv 1 \mod{23}$ and $\tau(p) \equiv 2$,
$-1
\mod{23}$. We conclude that $\tau(p) \equiv 2 \mod{23}$ and
$\tau(p)^2-p^{11}=3$.
Thus
\[
(\tau(p)+\sqrt{3})(\tau(p)-\sqrt{3})=p^{11}.
\]
The two factors on the left-hand side are coprime integers in
$\Z[\sqrt{3}]$. We see that
\[
\tau(p)+\sqrt{3}=(2+\sqrt{3})^a \gamma^{11}, \qquad \gamma \in \Z[\sqrt{3}],
\qquad \Norm(\gamma)=p, \qquad 0 \le a \le 10.
\]
Let $\fq=(2+3\sqrt{3})\Z[\sqrt{3}]$. Then $23 \Z[\sqrt{3}]=\fq \overline{\fq}$.
Since $\fq$ has residue field $\F_{23}$, we see that $\gamma^{11} \equiv \pm 1
\mod{\fq}$. Moreover, as $\tau(p) \equiv 2 \mod{23}$ we have
\[
2+\sqrt{3} \equiv \pm (2+\sqrt{3})^a \mod{\fq}.
\]
However, $2+\sqrt{3}$ has multiplicative order $11$ in $\Z[\sqrt{3}]/\fq=\F_{23}$.
As $0 \le a \le 10$ we conclude that $a=1$. Thus
\[
\tau(p)+\sqrt{3}=(2+\sqrt{3}) (U+V\sqrt{3})^{11}, \qquad U,~V \in \Z.
\]
Comparing coefficients of $\sqrt{3}$ we obtain the Thue equation
\begin{multline*}
U^{11} + 22 U^{10} V + 165 U^9 V^2 + 990 U^8 V^3 + 2970 U^7 V^4 + 8316 U^6 V^5 +\\
        12474 U^5 V^6 + 17820 U^4 V^7 + 13365 U^3 V^8 + 8910 U^2 V^9 +
        2673 U V^{10} + 486 V^{11} \; =\; 1.
\end{multline*}
The \texttt{Magma} Thue equation solver (based
on algorithms in \cite{Smart}) gives that
the only solution is $(U,V)=(1,0)$.
Thus  $p=U^2-3V^2=1$ which is a contradiction.
\end{proof}
We may now suppose that $11 \nmid b$ and so $N^\prime=2^5 \times 3 \times 11$.
There are $14$ conjugacy class of newforms of weight $2$ and
level $N^\prime$, and $10$ of these are rational.
We will denote the $10$ rational
newforms by $f_1,f_2,\dotsc,f_{10}$,
and $f_{11},f_{12},f_{13}$, $f_{14}$ will denote fixed
representatives of the other four conjugacy classes of newforms.
In what follows, $f$ will denote
one of these $14$ newforms so that \eqref{eqn:mod11} holds,
and the $c_j$ will denote its coefficients as in \eqref{eqn:newform}.
Moreover, $K=\Q(c_1,c_2,\dotsc)$.

For ease, we shall rewrite \eqref{eqn:BS2} as
\begin{equation}\label{eqn:BS3}
\tau(p)^2-p^{11}=\kappa \cdot 11^b, \qquad \kappa \in \{3,-3\}.
\end{equation}
To simplify, let
\[
E^\prime_p \; : \; Y^2=X(X^2+2\tau(p)X+p^{11}).
\]
If $p \equiv 3 \mod{4}$ then $E^\prime_p=E_p$.
%If $p \equiv 1 \mod{4}$
Moreover, $E^\prime_p$ is $2$-isogenous to the curve
\[
Y^2=X(X^2+4\tau(p)+4(\tau(p)^2-p^{11})),
\]
which, for $p \equiv 1 \mod{4}$, is the quadratic twist of $E_p$ by $2$. Thus, for any prime $\ell \ne 2$,
$3$, $11$, $p$
\[
a_\ell(E_p^\prime)=
\begin{cases}
a_\ell(E_p) &  \text{if $\ell \equiv \pm 1 \mod{8}$}\\
\pm a_\ell(E_p) & \text{if $\ell \equiv \pm 3 \mod{8}$}.\\
\end{cases}
\]
Let $5 \le \ell<100$ be a prime with $\ell \ne 11$.
But $p>100$, whence $\ell \ne p$.
Then, taking traces of the images of the Frobenius element at $\ell$
in \eqref{eqn:mod11} we obtain $a_\ell(E_p) \equiv c_\ell \mod{p}$ and
so $\Norm_{K/\Q}(a_\ell(E_p)-c_\ell) \equiv 0 \mod{11}$.
We deduce that
\begin{equation}\label{eqn:mod11norm}
\begin{cases}
\Norm_{K/\Q} (a_\ell(E_p^\prime)-c_\ell) \equiv 0 \mod{11} & \text{if $\ell \equiv \pm 1 \mod{8}$}\\
\Norm_{K/\Q} (a_\ell(E_p^\prime)^2-c_\ell^2) \equiv 0 \mod{11} & \text{if $\ell \equiv \pm 3 \mod{8}$}.\\
\end{cases}
\end{equation}
Let
\[
\cA_{\ell}=\{(s,t) \; : \; s,~t \in \F_\ell, \quad s \not \equiv 0 \mod{\ell},
\qquad t^2-s^{11} \not \equiv 0 \mod{\ell}
\}.
\]
Note that $(p,\sigma(p)) \equiv (s,t) \mod{\ell}$ for some $(s,t) \in \cA_{\ell}$.
Let
\[
\cB_{\ell}=
\begin{cases}
\cA_{\ell} & \ell \ne 5,~7,~23;\\
\{(s,t) \in \cA_{5} \; : \; t \equiv s^2(s^{3}+1) \mod{5} \} & \ell=5;\\
\{(s,t) \in \cA_{7} \; : \; t \equiv s(s^{3}+1) \mod{7} \} & \ell=7;\\
\{(s,0) \; : \; s \in \F_{23}^*\setminus (\F_{23}^*)^2\} \cup
\{(s,2) \; : \; s \in (\F_{23}^*)^2\}  & \ell=23.
\end{cases}
\]
It follows from \eqref{mod5}, \eqref{mod7}, \eqref{mod23}
that $(p,\sigma(p)) \equiv (s,t) \mod{\ell}$ for
some $(s,t) \in \cB_{\ell}$. Thus
$a_\ell(E_p^\prime)=a_\ell(E_{s,t})$ where
\[
E_{s,t}/\F_\ell \; : \; Y^2=X(X^2+2 t X +s).
\]
Let
\[
\cC_{\ell}(f)=
\begin{cases}
\{(s,t) \in \cB_\ell \; : \; \Norm(a_\ell(E_{s,t})-c_\ell(f)) \equiv 0
\mod{11}\} & \ell \equiv \pm 1 \mod{8}\\
\{(s,t) \in \cB_\ell \; : \; \Norm(a_\ell(E_{s,t})^2-c_\ell(f)^2) \equiv 0
\mod{11}\} & \ell \equiv \pm 3 \mod{8}\\
\end{cases}
\]
From \eqref{eqn:mod11norm} we have that $(p,\sigma(p)) \equiv (s,t)
\mod{\ell}$ for some $(s,t) \in \cC_{\ell}(f)$.
\begin{lem}\label{lem:intersect}
Let
\[
\cL=\{5,~7,~23,~89\}.
\]
For $\ell \in \cL$ and $\kappa \in \{-3,3\}$ let
\[
\mathcal{D}_{\ell}(f,\kappa)=\{ \beta \in \{0,1,\dotsc,65\} \; : \;
\text{$\kappa \cdot 11^\beta
\equiv t^2-s^{11} \mod{\ell}$ for some $(s,t) \in \cC_{\ell}(f)$}\},
\]
and
\[
\mathcal{D}(f,\kappa)=\bigcap_{\ell \in \cL} \mathcal{D}_{\ell}(f,\kappa).
\]
Let $(p,b)$ be a solution to \eqref{eqn:BS3} and let $f$
be a newform such that \eqref{eqn:mod11} is satisfied. Then
there is some $\beta \in \mathcal{D}(f,\kappa)$ such that
$b \equiv \beta \mod{66}$.
\end{lem}
\begin{proof}
From the above we know that $(p,\tau(p)) \equiv (s,t) \mod{\ell}$
for some $(s,t) \in \cC_{\ell}(f)$. Now,
$\kappa \times 11^b=\tau(p)^2-p^{11} \equiv t^2 -s^{11} \mod{\ell}$.
The multiplicative order of $11$ modulo $5$, $7$, $23$ and $89$
is respectively $1$, $3$, $22$, $22$, and their least
common multiple is $66$. It follows that there is some
$\beta \in \mathcal{D}(f,\kappa)$ such that $b \equiv \beta \mod{66}$.
\end{proof}

\begin{proof}[Proof of Theorem~\ref{case2}]
We wrote a short \texttt{Magma}
script which computed $\mathcal{D}(f,\kappa)$ for $f=f_1,\dotsc,f_{14}$
and $\kappa=-3$, $3$. In all cases we found
$\mathcal{D}(f,\kappa)=\emptyset$, except for $\kappa=-3$
and the rational newform $f$ corresponding to the
elliptic curve with Cremona label \texttt{1056d1},
when $\mathcal{D}(f,-3)=\{0,22,44\}$. It follows
from Lemma~\ref{lem:intersect}
that $11 \mid b$.
This contradicts Lemma~\ref{lem:not11}.
\end{proof}
\end{comment}

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

Let us suppose that we have a solution to equation (\ref{covid}) in nonzero integers $x,y,k,n$ and $q$, with $\gcd(x,q)=1$, where, to begin with, we suppose that $q$ and $n$ are primes with $n$ odd, say $n=p$.

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

  %---------------------------
\section{Primitive divisors} \label{PrimDiv}
%---------------------------

Here is a theorem of Cohn; we should probably redo this from scratch, assuming that $y$ is odd :

\begin{thm} \label{thm-cohn}
Suppose that $D \not\equiv 7 \mod{8}$ is a positive integer and write $D = d D_0^2$ where $d$ is square free. Let $h$ denote the class number of the imaginary quadratic field $\mathbb{Q} (\sqrt{-d})$ and suppose that there exist positive integers $x, y$ and an odd prime $n$ with $\gcd (x,D)=1$ and
$$
x^2+D = y^n.
$$
If $n$ fails to divide $h$, then we have one of
\begin{itemize}
\item there exist $u, v \in \mathbb{Z}$ such that $v\mid D_0$, $v \neq \pm D_0$, $x+D_0 \sqrt{-d} = (u+v \sqrt{-d})^n$ and $y = u^2 + d v^2$, or
\item $n=3, d \equiv 3 \mod{8}$ and there exist odd integers $U$ and $V$ with $V \mid D_0$, $x+D_0 \sqrt{-d} = \left(\frac{U+V \sqrt{-d}}{2}\right)^3$ and $y = \frac{1}{4} (U^2 + d V^2)$, or
\item $n=3$ and $D = 3A^2 \pm 8$,  $x = A^3 \pm 3 A$, or $D=3A^2 \pm 1$, $x=8A^2 \pm 3A$, for some integer $A$, or
\item $(n,D,x) = (5,19,22434)$ or $(5,341,2759646)$.
\end{itemize}
\end{thm}

Probably, we have

\begin{thm} \label{thm-start}
Suppose that $D$ is a positive integer and write $D = d D_0^2$ where $d$ is square free. Let $h$ denote the class number of the imaginary quadratic field $\mathbb{Q} (\sqrt{-d})$ and suppose that there exist positive integers $x, y$ and a prime $n \geq 5$ with $\gcd (x,D)=1$, $y$ odd and
$$
x^2+D = y^n.
$$
If $n$ fails to divide $h$ and $(n,D,x) \neq (5,250,401)$ then there exists a prime $p \mid D_0$ such that $p$ fails to divide $d$ and
$$
n \mid p - \left( \frac{-d}{p} \right).
$$
Here, $\left( \frac{\cdot}{p} \right)$ denotes the standard Legendre symbol.
\end{thm}

\begin{comment}
\subsection{The Minus Cases}
Write
\begin{equation}\label{eqn:x}
x=2 \tau(p)^2-3p^{11}.
\end{equation}
We consider first
\[
x^2-5p^{22}=-4 \cdot 5^{\alpha} \cdot 11^\beta.
\]

We split our considerations into four cases
as to whether $(\alpha,\beta)=(0,2\gamma)$, $(0,2\gamma+1)$,
$(1,2\gamma)$, $(1,2\gamma+1)$. We reduce each of these
cases to a Thue--Mahler equation which we solve
using the \texttt{Magma} implementation
of the Thue--Mahler solver described in \cite{GKMS}.

\medskip

Suppose first that $(\alpha,\beta)=(0,2\gamma)$. Then, after
possibly replacing $x$ by $-x$,
\[
x+2 \cdot 11^{\gamma} i =(2+i)(u+iv)^{11}, \qquad u,~v \in \Z, \qquad
p^2=u^2+v^2.
\]
Comparing the coefficients of $i$ we obtain
the Thue--Mahler equation
\begin{multline*}
u^{11} + 22 u^{10} v - 55 u^9 v^2 - 330 u^8 v^3 + 330 u^7 v^4 + 924 u^6 v^5 -
\\
    462 u^5 v^6 - 660 u^4 v^7 + 165 u^3 v^8 + 110 u^2 v^9 - 11 u v^{10} - 2
    v^{11}
    =2 \cdot 11^{\gamma}.
\end{multline*}
The only solution is $(u,v)=(0,-1)$. This leads to $p=1$
giving a contradiction.

\medskip

Next we suppose $(\alpha,\beta)=(0,2\gamma+1)$. We factor
$x^2+4 \cdot 11^{2\gamma+1}$ in $K=\Q(\sqrt{-11})$. This has
class number $1$. The prime $5$ splits as
\[
5=\frac{(3+\sqrt{-11})}{2} \cdot
\frac{(3-\sqrt{-11})}{2}.
\]
Hence, after possible changing the sign of $x$
\[
x+2 \cdot 11^\gamma \cdot \sqrt{-11} = \frac{(3+\sqrt{-11})}{2}
\cdot \left(u+v \cdot \frac{(1+\sqrt{-11})}{2} \right)^{11}.
\]
From this we obtain the Thue--Mahler equation
\begin{multline*}
u^{11} + 22 u^{10} v - 55 u^9 v^2 - 1155 u^8 v^3 - 1320 u^7 v^4 + 7854 u^6 v^5
+\\
    13398 u^5 v^6 - 7260 u^4 v^7 - 17985 u^3 v^8 - 2365 u^2 v^9 + 3124 u v^{10} +
    413 v^{11} \; = \; 4 \cdot 11^\gamma.
\end{multline*}
We find that this does not have any solutions.

\medskip

Next we suppose $(\alpha,\beta)=(1,2\gamma)$. Thus $5 \mid x$
and we let $y=x/5$. Thus
\[
4 \cdot 11^{2\gamma}+5 y^2=p^{22}.
\]
We factor in $\Z[\sqrt{-5}]$ Then
\[
\left( 2 \cdot 11^{\gamma}+y \sqrt{-5} \right) \cdot \Z[\sqrt{-5}] \; = \;
\fp^{22}
\]
where $\fp$ is a prime above $22$. As $\Z[\sqrt{-5}]$
has class number $2$, the ideal $\fp^2$ is principal, and thus
\[
2\cdot 11^{\gamma}+y \sqrt{-5} \; = \; (u+v\sqrt{-5})^{11}, \qquad u^2+5v^2=p^2.
\]
Comparing coefficients of $\sqrt{-5}$ we obtain
\[
u \cdot G(u,v) =
        2 \cdot 11^{\gamma}, \qquad
G(u,v)=u^{10} - 275 u^8 v^2 + 8250 u^6 v^4 - 57750 u^4 v^6 + 103125 u^2 v^8 -
        34375 v^{10}.
\]
Hence $G(u,v)=r \cdot 11^\delta$ where $r=1$, $-1$, $2$, $-2$.
We solved these four Thue--Mahler equation, and found that the
only solution is $(u,v,\delta)=(1,0,0)$ for $r=1$ and no
solutions in the other cases. As $p^2=u^2+5v^2$ we obtain a
contradiction.

\medskip

Finally
 we suppose $(\alpha,\beta)=(1,2\gamma+1)$.
\[
x^2+55 \cdot (2 \cdot 11^\gamma)^2=5p^{22}.
\]
Let $K=\Q(\sqrt{-55})$.
It follows that
\[
(x \, + \, 2 \cdot 11^\gamma \cdot \sqrt{-55}) \cdot \OO_K \; = \; \fq \cdot \fp^{22}
\]
where $5 \OO_K=\fq^2$ and $\fp$ is a prime ideal of norm $p$.
However, the class group of $\OO_K$ is cyclic of order $4$,
and the class of $\fq$ has order $2$. Since $\fq \fp^{22}$ and $\fp^4$
are both principal, we see that $\fq \fp^2$ is principal.
Observe that
\[
(x \, + \, 2 \cdot 11^\gamma \cdot \sqrt{-55}) \cdot \OO_K \; = \; \fq^{-10} \cdot
(\fq \fp^2)^{11}.
\]
However, $\fq^{-10}=5^{-5} \cdot \OO_K$. We deduce that
\[
5^5 \cdot (x \, + \, 2 \cdot 11^\gamma \cdot \sqrt{-55}) \; = \;
\left(u+v \cdot \frac{(1+\sqrt{-55})}{2} \right)^{11},
\qquad u^2+uv+14v^2=5p^2.
\]
Comparing the coefficients of $\sqrt{-55}$ we obtain
\begin{multline*}
11 v\cdot (u^{10} + 5 u^9 v - 195 u^8 v^2 - 810 u^7 v^3 + 6510 u^6 v^4 + 22386
u^5 v^5 -\\
    49110 u^4 v^6 - 136485 u^3 v^7 + 69095 u^2 v^8 + 141205 u v^9 - 4751 v^{10})
    \; = \; 4 \cdot 5^5 \cdot 11^{\gamma}.
\end{multline*}
Thus
\begin{multline*}
u^{10} + 5 u^9 v - 195 u^8 v^2 - 810 u^7 v^3 + 6510 u^6 v^4 + 22386
u^5 v^5 -\\
    49110 u^4 v^6 - 136485 u^3 v^7 + 69095 u^2 v^8 + 141205 u v^9 - 4751 v^{10}
    \; = \; \pm 2^r \cdot 5^s \cdot 11^t.
\end{multline*}
The only solutions to this Thue--Mahler equation are
$(u,v,r,s,t)=(\pm 1, 0,0,0,0)$, $(-1,2,0,5,4)$, $(1,-2,0,5,4)$.
Thus $5p^2=u^2+uv+14v^2=1$ or $55$ giving a contradiction.

\subsection{Plus Cases}
\end{proof}

Let
\[
y=\begin{cases}
2 \tau(p)^2-3p^{11} & \text{if $p \equiv 3 \mod{4}$}\\
3 p^{11}-2 \tau(p)^2 & \text{if $p \equiv 1 \mod{4}$}.
\end{cases}
\]
\end{comment}

$$
\alpha_p=\frac{a+\sqrt{b}}{2}, \; \; \beta_p=\frac{a-\sqrt{b}}{2},
$$
with  $\lambda_f (p)=a$ even, $a^2-b=4 \cdot p^{k-1}$,
where one of the following occurs.

\subsection{$n=3$}

$$
a=m \mbox{ and } b = -4-3m^2
$$
implies that
$$
a^2-b= 4m^2-4=4 p^{k-1}
$$
and so $m^2-1=p^{k-1}$, a contradiction since $m=a$ is even. Without this restriction, we have
$$
p=2, \; k=4, \; \lambda_f(2)=\pm 3.
$$

$$
a=m \mbox{ and } b = 4-3m^2
$$
implies that
$$
a^2-b= 4m^2+4=4 p^{k-1}
$$
and so $m^2+1=p^{k-1}$, again a contradiction.

$$
a=m \mbox{ and } b = 4 \cdot 3^j \epsilon - 3m^2, \; j \geq 1
$$
implies that
$$
a^2-b= 4m^2 - 4 \cdot 3^j \epsilon=4 p^{k-1}
$$
and so $m^2- 3^j \epsilon=p^{k-1}$. The case $\epsilon=-1$ is completely solved (see e.g. Luca), while the case $\epsilon = +1$ is done for suitably large (or small) exponent.

\begin{equation} \label{special1}
m^2-3^j=p^{k-1}.
\end{equation}

Note that, in this case, we have
$$
\lambda_f(p^2) = \lambda_f^2(p)-p^{k-1} =3^j.
$$
It is likely that the only solutions to (\ref{special1}) are given by
$$
m =\pm 2, \; j=1.
$$

\subsection{$n=4$}

$$
a=m \mbox{ and } b = 2 \epsilon - m^2,
$$
whence
$$
a^2-b= 2m^2-2 \epsilon=4 p^{k-1},
$$
a contradiction modulo $4$ (since $m=a$ is even). If
$$
a=m \mbox{ and } b = 4 \epsilon - m^2,
$$
then
$$
a^2-b= 2m^2-4 \epsilon=4 p^{k-1},
$$
and so, writing $m=2m_0$,
$$
2m_0^2- \epsilon= p^{k-1},
$$
handled in \cite{BenS}.

\subsection{$n=6$}

We could have
$$
a=m \mbox{ and } b = (4-m^2)/3,
$$
whence
$$
m^2-1=3 p^{k-1},
$$
an easy contradiction, via factoring. Also,
we could have
$$
a=m \mbox{ and } b = ((-2)^{j+2}-m^2)/3,
$$
and so
$$
m^2-(-2)^j =3 p^{k-1},
$$
contradicting $m$ even.
If
$$
a=m \mbox{ and } b = 4 \epsilon-m^2/3,
$$
with $m=3m_0$,
then
$$
3 m_0^2- \epsilon= p^{k-1},
$$
again treated in \cite{BenS}.
Finally, if
$$
a=m \mbox{ and } b = 2^{j+2}  \epsilon-m^2/3,
$$
then
$$
3m_0^2- 2^j \epsilon= p^{k-1},
$$
contradicting $m$ even.
============================================================

for some prime $p$ and integer $m \geq 2$.  We begin by noting that
$$
\left\{ 1, \tau (p), \tau (p^2), \tau (p^3), \ldots \right\}
$$
is a non-defective Lucas sequence (so that each term, from $\tau (p^2)$ on, has a  primitive divisor).
Further, we have that the rank of apparition $m_3(p)$ of $3$ divides $3$ or $4$, while the rank of apparition of $5$, $m_5(p)$,  divides $4$, $5$ or $6$, $m_7(p)$ divides one of $6, 7, 8$ and $m_{11}(p)$ divides one of $10, 11, 12$.

 We will show that $p=2$ and $m=3$. If $m=2, 3$ or $4$, then we must treat equations of the shape
\begin{equation} \label{case2}
\tau^2 (p) - p^{11} =  \pm 3^{\beta_1} 5^{\beta_2} 7^{\beta_3} 11^{\beta_4},
\end{equation}
\begin{equation} \label{case3}
\tau (p) \left( \tau^2 (p) -2  p^{11} \right) =  \pm 2^{\beta_0} 3^{\beta_1} 5^{\beta_2} 7^{\beta_3} 11^{\beta_4},
\end{equation}
and
\begin{equation} \label{case4}
(2 \tau^2(p) - 3 p^{11})^2-5p^{22} = \pm 4 \cdot 5^\alpha 11^\beta,
\end{equation}
respectively, where the $\beta_i$ are nonnegative integers. For these equations, we prove the following.
\begin{prop} \label{prop2}
Equation (\ref{case2}) has no solutions.
\end{prop}
\begin{prop} \label{prop3}
Equation (\ref{case3}) has only the  solution $p=2$.
\end{prop}
\begin{prop} \label{prop4}
Equation (\ref{case4}) has no solutions.
\end{prop}
\noindent Assuming these results for the time being, we may thus conclude that $m \geq 5$ and, since
$$
\tau (p^m) \mid \tau (p^{(m+1)k-1}),
$$
for all positive integer $k$, necessarily $m \not \in \{ 5, 7, 8, 9, 11 \}$.
It follows that either $m=6$ or $m=10$, with corresponding equations to consider
\begin{equation} \label{case6}
F_6(p^{11},\tau^2(p)) = \pm 7^\alpha
\end{equation}
and
\begin{equation} \label{case10}
F_{10}(p^{11},\tau^2(p)) = \pm 11^\alpha.
\end{equation}

Since $\tau (5)=4830$, we have that $(2 \tau^2(5) - 3 \cdot 5^{11})^2-5 \cdot 5^{22}$ is divisible by $179$. We may thus suppose that $p \neq 5$ in (\ref{case4}). If $\alpha > 1$ in this equation, we thus reach an immediate conclusion. We therefore have $\alpha \in \{ 0, 1 \}$. The case $\alpha=0$ leads to
$$
Y^2-5p^{22} = \pm 4 \cdot 11^\beta
$$
while $\alpha=1$ yields
$$
5Y^2-p^{22} = \pm 4 \cdot 11^\beta.
$$

Suppose our recurrence sequence is non-defective and that, say,
$$
\tau (p^m)= \pm 3^{\beta_1} 5^{\beta_2} 7^{\beta_3} 11^{\beta_4},
$$
for some even $m$, with each $\beta_i > 0$.

It's worth observing that the $F_m(X,Y)$ have interesting local properties. For example, if $\gcd (X,Y)=1$ and $2m+1$ is prime, then
$$
\nu_{2m+1} (F_{2m}(X,Y)) \leq 1.
$$
This makes solving (\ref{case6}) and (\ref{case10}) very easy.

%-------------------------------
\section{Equations \eqref{case6} and \eqref{case10}}
%---------------------------------

The polynomials $F_6(X,Y)$ and $F_{10}(X,Y)$ are explicitly
given by
\[
F_6(X,Y)=-64 X^3 + 96 X^2 Y - 20 X Y^2 + Y^3,
\]
and
\[
F_{10}(X,Y)=-1024 X^5 + 3840 X^4 Y - 2240 X^3 Y^2 + 448 X^2 Y^3 - 36 X Y^4 +
Y^5.
\]
CHECK THESE!! We thus consider the Thue--Mahler equations
\begin{equation}\label{eqn:TM1}
-64 X^3 + 96 X^2 Y - 20 X Y^2 + Y^3 \; =\; \pm 7^\alpha
\end{equation}
and
\begin{equation}\label{eqn:TM2}
-1024 X^5 + 3840 X^4 Y - 2240 X^3 Y^2 + 448 X^2 Y^3 - 36 X Y^4 +
Y^5 \; = \; \pm 11^\alpha.
\end{equation}
We solved these using the \texttt{Magma} implementation
of the Thue--Mahler solver described in \cite{GKMS}.
The solutions to \eqref{eqn:TM1} are
\[
(X,Y,\alpha) \; =\; (0,1,0), \quad (0,-1,0), \quad (1,13,0), \quad (-1,-13,0)
\]
and the solutions to \eqref{eqn:TM2} are
\[
(X,Y,\alpha) \; = \; (0,1,0), \quad (0,-1,0).
\]
It follows that \eqref{case6} and \eqref{case10} have solutions.

%%%%%%%%%%%%%

Let us suppose first that $p \mid F_m (p^{11},\tau^2 (p))$, so that $p \mid \tau (p)$. Since $\tau (p) \neq 0$ and $|\tau (p)| < 2 \cdot p^{11/2}$,  it follows that
$$
\nu_p (\tau (p)) \in \{ 1, 2, 3, 4, 5 \},
$$
whence
$$
\nu_p (  F_m (p^{11},\tau^2 (p))) = 2 \, m \,  \nu_p(\tau (p)).
$$
Writing
\begin{equation} \label{XY}
X=p^{11-2 \nu_p (\tau (p))} \; \mbox{ and } \; Y=\tau^2 (p)/p^{2 \nu_p(\tau (p))},
\end{equation}
it follows that $F_m(X,Y)$ is coprime to $p$ and divides $F_m (p^{11},\tau^2 (p))$. Let us write
\begin{equation} \label{for-real}
F_m(X,Y) = \pm 2^\alpha p_{0,1}^{\delta_{1}}  p_{0,2}^{\delta_{2}} \cdots  p_{0,k_0}^{\delta_{k_0}}
\cdot  p_{1,1}^{\gamma_{1}}  p_{1,2}^{\gamma_{2}} \cdots  p_{1,k_1}^{\gamma_{k_1}}
\cdot  p_{2,1}^{\omega_{1}}  p_{2,2}^{\omega_{2}} \cdots  p_{2,k_2}^{\omega_{k_2}},
\end{equation}
where we now assume that $\alpha$ is a nonnegative integer, that each $\delta_j, \gamma_{j}$ and $\omega_{j}$ is a positive integer and that the $p_{i,j}$ are distinct odd primes with
\begin{itemize}
\item $p_{0,j} \mid  \tau^2 (p) - 4 p^{11}, \; \; j = 1, 2, \ldots, k_0$,
\item $\left( \frac{ \tau^2 (p) - 4 p^{11}}{p_{1,j} }\right)=1,\; \; j = 1, 2, \ldots, k_1$, and
\item $\left( \frac{ \tau^2 (p) - 4 p^{11}}{p_{2,j} }\right)=-1, \; \; j = 1, 2, \ldots, k_2$.
\end{itemize}

Note that the assumption $p_{0,j}\mid F_m (p^{11},\tau^2 (p))$ implies, since $p_{0,j}\mid \tau^2 (p) - 4 p^{11}$ and  $p_{0,j} \neq p$, that
$$
F_m (1, 4) \equiv 0 \mod{p_{0,j}}.
$$
We have that
\begin{equation} \label{eff}
F_m(1,4)=
\left\{
\begin{array}{cc}
m+1 & \mbox{ if $m$ is even,} \\
\frac{m+1}{2} & \mbox{ if $m$ is odd,} \\
\end{array}
\right.
\end{equation}
whence $p_{0,j} \mid m+1$ for each $j=1, 2, \ldots, k_0$. We claim (maybe we'd better prove this!) that $\delta_j=1$ for all $j$.
For the $p_{1,j}$ and $p_{2,j}$, from our previous results about primitive divisors and ranks of apparition, we have
$$
p_{1,j} \equiv 1 \mod{m} \; \mbox{ and } \; p_{2,j} \equiv -1 \mod{m},
$$
for each $j$.

%%%%%%%%%%%%%%%%%%%%%

%---------------------------
\section{Modular forms}
%-----------------------------

The results we have described here are part of a more general phenomenon. We have that $\Delta(z)$ is a weight $12$ cuspidal newform for the whole modular group, with the property that $\tau (p)$ is even for all primes $p$.
Generally, suppose that $f$ is a normalized holomorphic cuspidal eigenform of (even) weight $k \geq 4$ and level $N$, with trivial Nebentypus and Fourier expansion
$$
f(z) = \sum_{n=1}^\infty \lambda_f (n) e^{2 \pi i nz}.
$$
Here, $\lambda_f(1)=1$. If we assume that
$$
K_f = \mathbb{Q}(\lambda_f(2), \lambda_f(3), \ldots) = \mathbb{Q},
$$
so that $\lambda_f(n) \in \mathbb{Z}$ for all $n$, and  that
$$
\lambda_f(p) \equiv 0 \mod{2}
$$
for all suitably large primes $p$ (we need to quantify this and establish the dependence of $c$ upon it), we may show that if $\lambda_f(n)$ is odd, and coprime to $N$, then
$$
P(\lambda_f(n))  > c \log \log n.
$$
There will be a dependence upon the weight $k$ in this\ldots

Note that if a prime $p\,  \| \, N$, then
$$
\lambda_f(p^m) = \pm p^{(k/2-1)m}
$$
and hence some dependence upon $N$ is necessary.

Via Deligne \cite{Deligne}, we have
\begin{equation} \label{good}
\lambda_f(p^m) = \frac{\alpha_p^{m+1} - \beta_p^{m+1}}{\alpha_p-\beta_p},
\end{equation}
where
$$
|\alpha_p|=|\beta_p|=p^{(k-1)/2} \mbox{ and } \alpha_p \beta_p = p^{k-1}.
$$
From this,
$$
\lambda_f(p^m) = \lambda_f(p) \lambda_f(p^{m-1}) -p^{k-1} \lambda_f(p^{m-2}).
$$
We note that to apply the primitive divisor theorem, we need $\lambda_f(p)$ to be an integer. The argument of Murty-Murty has no such restriction.

For the time being, let's suppose that $K_f =\mathbb{Q}$.

Note that
$$
\tau (3^2) = -3^4 \cdot 23 \cdot 61,
$$
$$
\tau (5^2) = -5^2 \cdot 79 \cdot 12911,
$$
$$
\tau(7^2)=-3 \cdot 7^2 \cdot 11543981,
$$
$$
\tau(9^2)=3^8 \cdot 253801.
$$

%-----------------------------------------
\section{Small weights}
%------------------------------------------

Generally, if we have
$$
\lambda_f(p)=0,
$$
then
$$
\lambda_f(p^2)=-p^{k-1},
$$
assuming trivial character.

If we have weight $k=4$, then, supposing that $P(\lambda_f(p^m)) \leq 13$, say, we are left to handle
$$
y^2=x^3 \pm 3^a5^b7^c11^d13^e,
$$
$$
F_6(X,Y)=\pm 7^a 13^b,
$$
$$
F_{10}(X,Y)=\pm11^a,
$$
$$
F_{12}(X,Y)=\pm  13^a.
$$