Surfaces defined by pairs of polynomials
Extracted source text · arXiv 2305.08632
%uploaded to the arxive 15 May 2023
\documentclass[letterpaper,12pt]{article}
\usepackage{color}
\usepackage{amssymb}
\usepackage{amsmath}
\usepackage{amscd}
\usepackage{mathtools}
\usepackage[all]{xy}
\usepackage{tikz-cd}
\usepackage{hyperref}
\hypersetup{colorlinks=true,linkcolor=blue,anchorcolor=blue,citecolor=blue}
\usepackage{cleveref}
\usepackage{color}
\newcommand{\damian}[1]{{\color{red} \sf $\clubsuit\clubsuit\clubsuit$ Damián : [#1]}}
\newcommand{\alexei}[1]{{\color{blue} \sf $\clubsuit\clubsuit\clubsuit$ Alexei : [#1]}}
\newtheorem{theo}{Theorem}[section]
\newtheorem{prop}[theo]{Proposition}
\newtheorem{lem}[theo]{Lemma}
\newtheorem{cor}[theo]{Corollary}
\newtheorem{defi}[theo]{Definition}
\newtheorem{rem}[theo]{Remark}
\def \wh {\widehat}
\def \kbar {{\bar k}}
\def \qbar {{\overline{\bf Q}}}
\def \la {{\lambda}}
\def \bfx{{\bf x}}
\def \K{{\mathcal K}}
\def \deg {{\rm{deg}}}
\def \id {{\rm id}}
\def \Br {{\rm{Br}}}
\def \Si {{\Sigma}}
\def \si {{\sigma}}
\def \Ga {{\Gamma}}
\def \R {{\mathbb{R}}}
\def \Pic {{\rm {Pic}}}
\def \Div {{\rm {Div}}}
\def \div {{\rm{div}}}
\def \Gal {{\rm{Gal}}}
\def \Ker {{\rm{Ker}}}
\def \Im {{\rm {Im}}}
\def \Frob{{\rm{Frob}}}
\def \AA{{\mathbf A}}
\def \A{{\mathbb A}}
\def \P{{\mathbb P}}
\def \Spec {{\rm{Spec}}}
\def \dim {{\rm{dim}}}
\def \Hom {{\rm {Hom}}}
\def \End {{\rm {End}}}
\def \GL {{\rm {GL}}}
\def \SL {{\rm {SL}}}
\def \SO {{\rm {SO}}}
\def \Sp {{\rm {Sp}}}
\def \PGL {{\rm {PGL}}}
\def \Spin {{\rm {Spin}}}
\def \Aut{{\rm Aut}}
\def \Out{{\rm Out}}
\def \Inn{{\rm Inn}}
\def \Cl{{\rm Cl}}
%\def\la{\leftarrow}
\def\ov{\overline}
\def\un{\underline}
\def\k{\kappa}
\def \Z {{\mathbb Z}}
\def \Q {{\mathbb Q}}
\def \F {{\mathbb F}}
\def \Ext {{\rm Ext}}
\def \Id {{\rm Id}}
\def \AA {{\rm A}}
\def \BB {{\rm B}}
\def \DD {{\rm D}}
\def \EE {{\rm E}}
\def \RR {{\rm R}}
\def \PP {{\rm P}}
\def \CC {{\rm C}}
\def \FF {{\rm F}}
\def \ZZ {{\rm Z}}
\def \Tr {{\rm{Tr}}}
\def \Mat {{\rm{Mat}}}
\def \sym {{\rm{sym}}}
\def \U {{\cal U}}
\def \rk {{\rm{rk}}}
\def \val {{\rm{val}}}
%%%%%%%%%%%%
\def\G{{\mathbb G}}
\def\D{{\cal D}}
\def\GG{{\cal G}}
\def\HH{{\cal H}}
\def\C{{\mathbb C}}
\def\B{{\cal B}}
\def\calT{{\cal T}}
\def\T{{\cal T}}
\def\sZ{{\cal Z}}
\def\sY{{\cal Y}}
\def\sM{{\cal M}}
\def\lra{\longrightarrow}
\def\cl{{\rm cl}}
\def\N{{\rm N}}
\def\H{{\rm H}}
\def\ad{{\rm ad}}
\def\Tr{{\rm Tr}}
\def\Zar{{\rm Zar}}
\def\sA{{\cal A}}
\def\sB{{\cal B}}
\def\K{{\cal K}}
\def\X{{\cal X}}
\def\Y{{\cal Y}}
\def\L{{\cal L }}
\def\inv{{\rm inv}}
\def\Tor{{\rm Tor}}
\def\Stab{{\rm Stab}}
\def\NS{{\rm NS\,}}
\def\Sha{{\cyr X}}
\def\x{{\bf x}}
\def\O{{\cal O}}
\def\L{{\cal L}}
\def\sE{{\cal E}}
\def\sC{{\cal C}}
\def\sF{{\cal F}}
\def\sX{{\cal X}}
\def\sY{{\cal Y}}
\def\Si{\Sigma}
\def\si{\sigma}
\def\cl{{\rm cl}}
\def\ab{{\rm ab}}
\def\res{{\rm res}}
\def\val{{\rm val}}
\def\Lie{{\rm Lie}}
\def\codim{{\rm codim}}
\def\discr{{\rm discr\,}}
\def\Ga{\Gamma}
\def\ga{\gamma}
\def\e{\varepsilon}
\def\et{{\rm{\acute et}}}
\def\St{{\rm St}}
\def\const{{\rm const}}
\def\exp{{\rm exp}}
\def \g{{\mathfrak g}}
\def \h{{\mathfrak h}}
\def\p{{\mathfrak p}}
\def\n{{\mathfrak n}}
\def\m{{\mathfrak m}}
\def\fG{{\mathfrak G}}
\def\fH{{\mathfrak H}}
\def\fA{{\mathfrak A}}
\def\bS{{\bf S}}
\def\bA{{\bf A}}
\def\z{{\bf z}}
\newcommand{\bthe}{\begin{theo}}
\newcommand{\ble}{\begin{lem}}
\newcommand{\bpr}{\begin{prop}}
\newcommand{\bco}{\begin{cor}}
\newcommand{\bde}{\begin{defi}}
\newcommand{\brem}{\begin{rem}}
\newcommand{\ethe}{\end{theo}}
\newcommand{\ele}{\end{lem}}
\newcommand{\epr}{\end{prop}}
\newcommand{\eco}{\end{cor}}
\newcommand{\ede}{\end{defi}}
\newcommand{\erem}{\end{rem}}
\vsize=220mm %225mm
\textheight=220mm %225mm
\hsize=150mm %138mm
\textwidth=150mm %138mm
\parskip=2pt minus1pt
\parindent=12pt
\arraycolsep=2.2pt\def\arraystretch{1.2}
\setcounter{MaxMatrixCols}{20}
\title{Surfaces defined by pairs of polynomials}
\author{D.~Gvirtz-Chen and A.N.~Skorobogatov}
\date{\today}
\begin{document}
\maketitle
\begin{abstract} %AS I edited the abstract
We prove that the Brauer group of the generic diagonal surface of arbitrary degree is trivial.
The same method is applied to surfaces whose equation can be written as the sum of two bilinear forms.
This uses a general criterion for the triviality of the transcendental Brauer group of an
isotrivial variety.
We determine the field of definition of the
geometric Picard group of the Fermat surface of arbitrary degree,
thus finishing work of Shioda and Aoki.
\end{abstract}
\section{Introduction}
The starting point of the present investigation was our desire to compute the Brauer group
of the \emph{generic} diagonal surface. We have found that for a field $k_0$ of characteristic zero
and algebraically independent variables $t_1,t_2,t_3$,
the Brauer group of the surface given by
$$x_0^d+t_1x_1^d+t_2x_2^d+t_3x_3^d=0$$
in $\P^3_k$ where $k=k_0(t_1,t_2,t_3)$, is naturally isomorphic to the Brauer group of $k$,
see Theorem \ref{co1}. In contrast, a \emph{general} (in a suitable sense, see Corollary~\ref{co2}) diagonal surface over a number field $k$ containing a primitive $d$-th root of unity has non-trivial Brauer group when the degree $d$ is at least $3$. These statements were previously known for $d\leq 4$.
%the case of arbitrary degree
%was out of reach due to a lack of understanding of the structure of integral cohomology.
It turns out that our method can be applied to more general surfaces
$X\subset\P^3_k$ given by
$F(x_0,x_1)=G(x_2,x_3)$,
where $F$ and $G$ are homogeneous forms of degree $d$. Smooth surfaces of this form have been investigated extensively in the classical and modern geometric literature, e.g.~by Schur \cite{Schu},
Segre \cite{Seg1, Seg2, Seg3, Seg4}, Inose \cite{Ino}, Shioda \cite{Sh81},
Caporaso--Harris--Mazur \cite{CHM}, Boissi\`ere--Sarti \cite{BS, BS2},
Degtyarev \cite{Deg15}. When $d=4$, these are K3 surfaces and when $d\geq 5$, surfaces of general type. Unlike diagonal surfaces, surfaces defined by a pair of polynomials do not lie in a single geometric isomorphism class.
From a number theoretic perspective, such surfaces are of interest due to their partial separation of variables. Thus, Hooley was able to establish asymptotic upper bounds for the number of rational points of bounded height on $X$ over $\Q$ in his articles on binary forms \cite{Ho1, Ho2, Ho3} (see also \cite[p.~558]{HB}).
When $k$ is finitely generated over $\Q$, surfaces defined by a pair of polynomials are relevant
for the study of the Brauer--Manin obstruction to rational points because
their Brauer group $\Br(X)$
is known to be finite modulo the image of $\Br(k)$, see \cite[Cor.~1.8]{GS22}.
Without loss of generality, consider the surface
$X\subset\P^3_k$ given by the equation
\begin{equation}
f(x_1/x_0)x_0^d=ag(x_3/x_2)x_2^d, \label{X}
\end{equation}
where $a\in k^\times$, and $f(t)$, $g(t)$ are separable monic polynomials of degree $d$ with
coefficients in $k$. It is immediate to check that $X$ is smooth,
and hence geometrically integral. In this paper we develop a method which allows us to determine
$\Br(X)$ for given $f$ and $g$ assuming that $a$,
in an appropriate sense, is general enough.
Let $\bar k$ be an algebraic closure of $k$, let
$\Ga_k=\Gal(\bar k/k)$, and let $X_{\bar k}=X\times_k\bar k$.
Our main tool is the spectral sequence
\begin{equation}
E^{p,q}_2\colon \ \H^p(k,\H^q(X_{\bar k}, \G_m))\Rightarrow \H^{p+q}(X,\G_m), \label{ss}
\end{equation}
and an explicit description of the differential
$d_{1,1}\colon\H^1(k,\Pic(X_{\bar k})) \to \H^3(k,\bar k^\times)$.
Recall the standard notation
$$\Br_0(X)=\Im[\Br(k)\to\Br(X)], \quad \Br_1(X)=\Ker[\Br(X)\to \Br(X_{\bar k})].$$
The spectral sequence (\ref{ss}) implies that $\Br_1(X)/\Br_0(X)$ is canonically isomorphic to
the kernel of $d_{1,1}$; in particular, if $\H^1(k,\Pic(X_{\bar k}))=0$, then
$\Br_1(X)=\Br_0(X)$.
Define the field $L_{f,g}$ as the compositum of $k(\mu_d)$ and
the field of definition of $\Pic(X_{\bar k})$, where $X$ is given by
(\ref{X}) with $a=1$.
The method of this paper crucially depends on the assumption
$[L_{f,g}(\sqrt[d]{a}):L_{f,g}]=d$, which holds when $a$ is general enough for
given $f$ and $g$. A somewhat stronger condition is
$[K_{f,g}(\sqrt[d]{a}):K_{f,g}]=d$, where $K_{f,g}$ is a finite extension of $L_{f,g}$
explicitly described in terms of $f$ and $g$, see Proposition \ref{Lfg}.
Although our method works whenever $[L_{f,g}(\sqrt[d]{a}):L_{f,g}]=d$, to obtain a closed
formula for the Brauer group it is convenient to assume that
$f(t)$ and $g(t)$ are {\em irreducible} monic polynomials of degree $d$
whose splitting fields $k_f$ and $k_g$ are {\em linearly disjoint} over $k$. It is clear that,
in an appropriate sense, this holds for `almost all' pairs of polynomials $f$ and $g$.
For a finite group $G$ we write $G^*=\Hom(G,\Q/\Z)$.
For $a\in k^\times$ we denote by $[a]\in\H^1(k,\mu_d)$ the image of $a$ under the
connecting map $k^\times\to\H^1(k,\mu_d)$ of the Kummer exact sequence.
We shall use the following notation for an irreducible polynomial $f(t)\in k[t]$ with a root
$\xi\in\bar k$, $f(\xi)=0$. Let $K_f$ be the maximal Galois subextension of $k(\xi)$.
Define $G_f=\Gal(k_f/k)$, $S_f=\Gal(k_f/k(\xi))$, $N_f=\Gal(k_f/K_f)$.
Then $N_f$ is the normal closure of $S_f$ in $G_f$.
The abelianisation $(G_f/N_f)^{\rm ab}$ is the Galois group
of the maximal abelian subextension of $k\subset k(\xi)$ over $k$.
Let $G=G_f\times G_g$.
For a $G$-module $M$ we denote by $\H^i(G,M)_{\rm prim}$ the kernel of the restriction map
$\H^i(G,M)\to\H^i(G_f,M)\oplus \H^i(G_g,M)$.
The Künneth formula for group cohomology gives a canonical isomorphism
$\H^3(G,\Z)_{\rm prim}\cong(G_f^{\rm ab}\otimes G_g^{\rm ab})^*$.
Let $\Delta$ be the composition of the following maps
$$\Hom((G_f/N_f)^{\rm ab}\otimes (G_g/N_g)^{\rm ab},\Z/d)\hookrightarrow
(G_f^{\rm ab}\otimes G_g^{\rm ab})^*$$
$$\tilde\lra \H^3(G,\Z)_{\rm prim}\hookrightarrow\H^3(G,\Z)\stackrel{\rm inf}\lra
\H^3(k,\Z)\to\H^3(k,\bar k^\times),$$
where the last arrow is induced by the map sending $1\in\Z$ to $a\in \bar k^\times$.
Alternatively, one can describe $\Delta$ in terms of cup-products in Galois cohomology.
The K\"unneth formula for group homology and the universal coefficients formula
give rise to an isomorphism $\H^2(G,\Z/d)_{\rm prim}\cong
\Hom(G_f^{\rm ab}\otimes G_g^{\rm ab},\Z/d)$, see (5.36) on p.~154 of \cite{CTS21}.
Then $\Delta$ is the composition
$$\Hom((G_f/N_f)^{\rm ab}\otimes (G_g/N_g)^{\rm ab},\Z/d)\hookrightarrow
\Hom(G_f^{\rm ab}\otimes G_g^{\rm ab},\Z/d)$$
$$\tilde\lra\H^2(G,\Z/d)_{\rm prim}\hookrightarrow\H^2(G,\Z/d)\stackrel{\rm inf}\lra
\H^2(k,\Z/d)\stackrel{\cup[a]}\lra\H^3(k,\mu_d)\to\H^3(k,\bar k^\times).$$
The equivalence of the two descriptions of $\Delta$ can be checked by taking
quotients modulo $d$ in the bottom extension of (\ref{bruckner}), using
isomorphism (\ref{may}) and the commutativity of diagram (\ref{difficult}).
Let $\varphi\in\Hom((G_f/N_f)^{\rm ab}\otimes (G_g/N_g)^{\rm ab},\Z/d)$.
Using the canonical surjective function $G_f/S_f\to (G_f/N_f)^{\rm ab}$ and a similar map for $g$,
we obtain from $\varphi$ a function
$\widetilde\varphi\colon (G_f/S_f)\times (G_g/S_g)\to\Z/d$.
Define
$$\Pi_{f,g}\subset \Hom((G_f/N_f)^{\rm ab}\otimes (G_g/N_g)^{\rm ab},\Z/d)$$
as the subgroup of functions $\varphi$ such that
$$\sum_{y\in G_g/S_g}\widetilde\varphi(a,y)=
\sum_{x\in G_f/S_f} \widetilde\varphi(x,b)=0$$ for any $a\in G_f/S_f$ and any $b\in G_g/S_g$.
\bthe \label{f=g}
Let $f$ and $g$ be irreducible monic polynomials in $k[t]$ of degree $d$
whose splitting fields are linearly disjoint over $k$.
Let $X$ be the surface $(\ref{X})$, where $a\in k^\times$ is such that $[L_{f,g}(\sqrt[d]{a}):L_{f,g}]=d$.
Then the following statements hold.
{\rm (i)} We have $\H^1(k,\Pic(X_{\bar k}))\cong \Pi_{f,g}$.
If the orders of $(G_f/N_f)^{\rm ab}$ and $(G_g/N_g)^{\rm ab}$ are coprime, then
$\H^1(k,\Pic(X_{\bar k}))=0$.
{\rm (ii)} If at least one of $f$ and $g$ has primitive Galois group,
then $\H^1(k,\Pic(X_{\bar k}))=0$ except when $d$ is an odd prime
and both polynomials have cyclic Galois groups.
{\rm (iii)} The differential $d_{1,1}\colon \H^1(k,\Pic(X_{\bar k}))\to\H^3(k,\bar k^\times)$
can be identified with $\Delta$,
hence there is an isomorphism $\Br_1(X)/\Br_0(X)\cong \Pi_{f,g}\cap\Ker(\Delta)$.
\ethe
It is noteworthy that the proof of this theorem uses neither any knowledge of
the group $\Pic(X_{\bar k})$ (like an integral basis or the Galois action), nor of the field $L_{f,g}$.
Recall that a transitive subgroup $H\subset S_d$ is called {\em primitive}
if the stabiliser of a point is a maximal subgroup of $H$. Thus, under the assumptions
of Theorem \ref{f=g}, we have $\Br_1(X)=\Br_0(X)$ if at least one of
$G_f$ and $G_g$ is the symmetric group $S_d$ for $d\geq 3$, or the alternating group $A_d$ for
$d\geq 4$.
The same conclusion holds if $d$ is prime and at least one of $G_f$ and $G_g$ is not isomorphic to $\Z/d$, hence singling out the cyclic case as special.
For any $d$, if $G_f\simeq G_g\simeq\Z/d$, then (i) gives that $\H^1(k,\Pic(X_{\bar k}))$
is isomorphic to $\Z/d$ if $d$ is odd, and $\Z/(d/2)$ if $d$ is even. This applies to the exceptional
case in~(ii).
\subsubsection*{Diagonal surfaces}
As mentioned above, a case of particular interest is that of diagonal surfaces,
where we can take $f(t)=t^d+b$ and $g(t)=t^d+c$ for $b,c\in k^\times$.
These surfaces can be regarded as twisted forms of
the Fermat surface $F\subset\P^3_\Q$:
\begin{equation}
x_0^d+x_1^d+x_2^d+x_3^d=0.\label{Fermat}
\end{equation}
Let $L$ be the field of definition of $\Pic(F_{\ov\Q})$,
that is, the smallest extension of $\Q$
such that $\Gal(\ov\Q/L)$ acts trivially on $\Pic(F_{\ov\Q})$.
We have $L=L_{f,g}$, where $f(t)=g(t)=t^d+1$. To make our next result explicit
we determine the field $L$ for any $d$ via a Größencharacter computation, thereby completing previous work of Shioda and Aoki
\cite{Sh81, Aok, ASh}, see \S\ref{Pic}.
Consider the diagonal surface $X\subset\P^3_k$ given by
\begin{equation}
x_0^d+a_1x_1^d+a_2x_2^d+a_3x_3^d=0,\label{eqX}
\end{equation}
where $a_1,a_2,a_3\in k^\times$.
Let us denote by $k_L$ the compositum $kL$, and by $\langle a_1,a_2,a_3\rangle$
the subgroup of $k_L^\times/(k_L^\times)^d$ generated by the classes of $a_1,a_2,a_3$.
The following condition describes what we mean by `general' diagonal surfaces:
$$\langle a_1,a_2,a_3\rangle\simeq(\Z/d)^3. \eqno{(\star)}$$
Let $[a_1]\cup[a_2]\cup[a_3]\in \H^3(k,\mu_d^{\otimes 3})$ be the triple cup-product.
Fixing a primitive $d$-th root of unity we get a map
$$\H^3(k,\mu_d^{\otimes 3})\to \H^3(k(\mu_{2d}),\mu_d^{\otimes 3})
\to \H^3(k(\mu_{2d}),\bar k^*),$$
where the first arrow is restriction. Consider the following condition:
$$\text{the order of the image of $[a_1]\cup[a_2]\cup[a_3]$ in
$\H^3(k(\mu_{2d}),\bar k^*)$ is $d$.}\eqno{(\star\star)}$$
Let $\Q(\mu_{d})^{[2]}\subset \Q(\mu_{d})$ be the invariant subfield of
the 2-torsion subgroup $(\Z/d)^\times[2]$ of $(\Z/d)^\times\cong \Gal(\Q(\mu_{d})/\Q)$.
%In particular, $\Q(\mu_{d})^{[2]}=\Q$ if and only if $d$ is one of the following numbers: 1, 2, 3, 4, 6, 8, 12, 24.
%DG Feels a bit repetitive since it is also said after Cor. 1.3.
Write $\chi\colon\Ga_k\to (\Z/d)^\times$
for the cyclotomic character of degree $d$.
We have $2(\Z/d)\simeq\Z/d$ when $d$ is odd and
$2(\Z/d)\simeq\Z/(d/2)$ when $d$ is even.
\bthe\label{main}
Let $X\subset \P^3_k$ be the diagonal surface $(\ref{eqX})$
satisfying condition $(\star)$.
Then the following properties hold.
{\rm (i)}
The group $\H^1(k,\Pic(X_{\bar k}))$ is isomorphic to the largest subgroup of $2(\Z/d)$ on which
$\chi^2(\Ga_k)$ acts by $1$. In particular, if $\Q(\mu_{d})^{[2]}\subset k$, then
$\H^1(k,\Pic(X_{\bar k}))\simeq 2(\Z/d)$.
{\rm (ii)} If $\Q(\mu_{2d})\subset k$ and
$r|d$ is such that the order of the image of $[a_1]\cup[a_2]\cup[a_3]$ in
$\H^3(k,\bar k^*)$ is $d/r$, then $\Br_1(X)/\Br_0(X)\cong \H^1(k,\Pic(X_{\bar k}))[r]$.
For any $k$, if condition $(\star\star)$ holds, for example when $k=k_0(a_1,a_2,a_3)$
is a purely transcendental extension of transcendence degree $3$ of a subfield $k_0$,
then $\Br_1(X)=\Br_0(X)$.
{\rm (iii)} The natural map $\Br(k)\to\Br(X)$ is injective.
\ethe
The result of Theorem \ref{main} (i) for $d=3$ is due to J.-L.~Colliot-Th\'el\`ene,
D.~Kanevsky and J.-J.~Sansuc when $k$ contains $\Q(\mu_3)$
\cite[Prop.~1]{CTKS}. This was extended to arbitrary fields of characteristic zero in
\cite[Prop.~6.1]{PT}, see also \cite[Lemma 2.1]{BBL}. For $d=4$ it is due
to M.~Bright \cite{Bri02, Bri11}.
The condition $(\star\star)$ holds when $k=k_0(a_1,a_2,a_3)$
is a purely transcendental extension of $k_0$ of transcendence degree $3$.
In this case Theorem \ref{main} (ii)
was proved by T.~Uematsu for $d=3$ \cite[Thm.~1.2]{U} and by T.~Santens for $d=4$
\cite[Thm.~6.2]{San} using \cite{U2} and \cite{GS22}. Our method works for all degrees $d$.
Condition $(\star\star)$ does not hold when $k$ is a number field,
because in this case $\H^3(k,\bar k^\times)=0$. For number fields,
Theorem \ref{main} takes the following form.
\bco \label{co2}
Let $k$ be a number field, and let
$X\subset \P^3_k$ be the diagonal surface $(\ref{eqX})$
satisfying condition $(\star)$. The map $\Br(k)\to\Br(X)$ is injective.
The group $\Br_1(X)/\Br(k)$ is isomorphic to the largest subgroup of $2(\Z/d)$ on which
$\chi^2(\Ga_k)$ acts by $1$.
If $k$ contains $\Q(\mu_{2d})^{[2]}$, then $\Br_1(X)/\Br(k)\simeq 2(\Z/d)$.
\eco
Since $\Q(\mu_{d})^{[2]}=\Q$ precisely when
$d\in\{1, 2, 3, 4, 6, 8,12,24\}$,
for any degree $d$ from this list and for any number field $k$ we have
$\Br_1(X)/\Br(k)\simeq 2(\Z/d)$. It is easy to show that if $k=\Q$, then
$\Br_1(X)/\Br(\Q)$ is isomorphic to
$\Z/2^a\times \Z/3^b$, where $a=2$ if $8$ divides $d$, $a=1$ if
$4$ simply divides $d$, and $a=0$ if $d$ is not a multiple of $4$,
and $b=1$ if $3$ divides $d$ and $b=0$ otherwise. In particular,
if $d$ is not divisible by 3 or 4, then $\Br_1(X)=\Br(\Q)$.
Comparing Corollary \ref{co2} to Theorem \ref{main} (ii)
we see that in the family of diagonal surfaces (\ref{eqX}) over $\A^3_k$
`uniform generators' of the algebraic Brauer group do not exist.
\subsubsection*{Transcendental Brauer group}
For some `isotrivial' surfaces over a function field we can prove the triviality of the
transcendental Brauer group, see Theorem \ref{friday}. This implies
\bthe \label{n}
Let $k_0$ be a field of characteristic zero, and let $f$ and $g$ be separable
polynomials in $k_0[t]$ of degree $d$. Let $k=k_0(a)$
be a purely transcendental extension of $k_0$ of transcendence degree $1$, and
let $X\subset \P^3_k$ be the surface $(\ref{X})$.
Then the map $\Br(X)\to\Br(X_{\bar k})$ is zero. Moreover, if $f$ and $g$ are irreducible
with splitting fields linearly disjoint over $k_0$
and at least one of them has primitive Galois group,
then $\Br(k)\to\Br(X)$ is an isomorphism, except when $d$ is an odd prime
and both polynomials have cyclic Galois groups.
\ethe
Finally, we state the result mentioned in the beginning of the introduction.
\bthe \label{co1}
Let $k_0$ be a field of characteristic zero and let $k=k_0(a_1,a_2,a_3)$
be a purely transcendental extension of $k_0$ of transcendence degree $3$.
Let $X\subset \P^3_k$ be the diagonal surface $(\ref{eqX})$.
Then the map $\Br(k)\to\Br(X)$ is an isomorphism.
\ethe
In this case both conditions $(\star)$ and $(\star\star)$ hold.
This can be compared to `less generic' diagonal surfaces, where these conditions
fail, see the examples at the end of \S\ref{3}.
\subsubsection*{Outline of paper}
In \S\ref{1} we discuss the subgroup
$\Lambda\subset \Pic(X_{\bar k})$ generated by the classes of $d^2$ lines contained in
$f(x_1/x_0)x_0^d=g(x_3/x_2)x_2^d=0$. The main result of this section is
Theorem \ref{Lambda} which says that the natural map
$\H^1(k,\Lambda)\to\H^1(k,\Pic(X_{\bar k}))$ is an isomorphism if
$[L_{f,g}(\sqrt[d]{a}):L_{f,g}]=d$. The proof strategy is a topological deformation argument to the diagonal case, where we build on work of Shioda and Degtyarev.
The $\Ga_k$-module $\Lambda$ does not depend on $a$, and has a simple explicit description.
Moreover, the field of definition of $\Lambda$ is the compositum of
$k_f$ and $k_g$. This is why we do not need to know $\Pic(X_{\bar k})$ or $L_{f,g}$ explicitly.
We prove Theorems \ref{f=g} and \ref{main} in \S\ref{diff}.
In \S\ref{3} we prove the triviality of the transcendental Brauer group
for some `isotrivial' surfaces,
see Theorem \ref{friday}. Then we deduce Theorems \ref{n} and \ref{co1}.
Finally, \S\ref{Pic} contains results about the field of definition of
the geometric Picard group of the surfaces that we study: an `upper bound' for $L_{f,g}$
and a complete (though complicated) description of $L$.
\section{Reduction to $d^2$ lines} \label{1}
In this section $f(t)$ and $g(t)$ are separable monic polynomials of degree $d$ with coefficients
in a field $k$ of characteristic zero, and $X\subset\P^3_k$ is the surface
given by (\ref{X}), where $a\in k^\times$.
Define $Z\subset X$ as the union of closed subsets $x_0=0$ and
$f(x_1/x_0)x_0^d=0$. On $X_{\bar k}$, the zero set of $x_0$
is a smooth irreducible curve, while the zero set of $f(x_1/x_0)x_0^d$ is the union of $d^2$
lines. Let $U=X\setminus Z$.
Consider the subgroup $\Lambda\subset\Pic(X_{\bar k})$ generated by the
classes of $d^2$ lines contained in $Z_{\bar k}$.
The following lemma gives an explicit description of this Galois module.
In particular, it shows that the Galois module $\Lambda$
does not depend on $a$.
Write $V_f$ be the closed subscheme of the affine line $\A^1_k$ given by $f(t)=0$, and similarly
for $V_g\subset\A^1_k$.
For a 0-dimensional $k$-scheme $S$ of finite type we denote by $\Z[S]$ the free abelian group
whose generators bijectively correspond to the $\bar k$-points of $S$. Then $\Z[S]$
is a permutational $\Ga_k$-module. Let $\si_S\in \Z[S]$ be the sum of the generators.
We write $\si_f=\si_{V_f}$ and $\si_g=\si_{V_g}$.
\ble \label{otto}
We have a commutative diagram of $\Ga_k$-modules with exact rows
\begin{equation}
\begin{split}
\xymatrix{
0\ar[r]&\bar k^\times\ar[r]& \bar k(X)^\times\ar[r] &
\Div(X_{\bar k})\ar[r]&\Pic(X_{\bar k})\ar[r]& 0\\
0\ar[r]&\bar k^\times\ar[u]^{\rm id}_\cong\ar[r]& \bar k[U]^\times\ar[r] \ar@{_{(}->}[u]&
\Div_{Z_{\bar k}}(X_{\bar k})\ar@{_{(}->}[u]\ar[r]&\Lambda\ar@{_{(}->}[u]\ar[r]& 0\\
0\ar[r]& \Z\ar[r]^{p_1 \ \ \ \ \ \ \ \ }\ar[u]^a& \Z[V_f]\oplus\Z[V_g]\ar[r]^{p_2 \ \ \ \ \ }\ar[u]&
(\Z[V_f]\otimes \Z[V_g])\oplus\Z h\ar[r]\ar[u]_{\rm id}^\cong& \Lambda\ar[r]\ar[u]_{\rm id}^\cong& 0
}
\label{duo}
\end{split}
\end{equation}
The arrows from the middle row to the top row are the natural embeddings. The remaining
arrows are as follows:
\begin{itemize}
\item the vertical map with label $a$ sends $1\in\Z$ to $a\in\bar k^\times$;
\item $p_1(1)=\si_f-\si_g$;
\item $h\in\Z h$ is the hyperplane section $x_0=0$;
\item generators $\xi\in\Z[V_f]$, where $f(\xi)=0$, are the rational functions
$(x_1-\xi x_0)/x_0$;
\item generators $\rho\in\Z[V_g]$, where $g(\rho)=0$, are the rational functions
$(x_3-\rho x_2)/x_0$;
\item generators $\xi\otimes\rho$ of $\Z[V_f]\otimes \Z[V_g]$ are the lines
$x_1-\xi x_0=x_3-\rho x_2=0$;
\item $p_2$ sends each $\xi\in \Z[V_f]$ to $(\xi\otimes\si_g) -h$ and
each $\rho\in \Z[V_g]$ to $(\si_f\otimes\rho)-h$;
\item unlabeled horizontal arrows are the natural maps.
\end{itemize}
\ele
{\em Proof.} The commutativity of the diagram is easily checked. The exactness of the top and
middle sequences is clear. It remains to prove the exactness of the bottom sequence, which
is non-trivial only at $(\Z[V_f]\otimes \Z[V_g])\oplus\Z h$. We need to check
that every relation among the classes
of the $d^2$ lines in $\Lambda$ is a linear combination of the relations given by
the rational functions $(x_1-\xi x_0)/x_0$ and $(x_3-\rho x_2)/x_0$.
Write
$$f(t)=\prod_{i=0}^{d-1}(t-\xi_i), \quad\quad g(t)=\prod_{j=0}^{d-1}(t-\rho_j),$$
where all $\xi_j,\rho_j\in\bar k$.
Let $L_{ij}$ be the projective line given by $x_1-\xi_i x_0=x_3-\rho_j x_2=0$.
Let $D$ be the subgroup of $\Div_{Z_{\bar k}}(X_{\bar k})$
generated by the lines $L_{ij}$ with $i\neq 0$, $j\neq 0$. It is easy to see that
$\Div_{Z_{\bar k}}(X_{\bar k})={\rm Im}(p_2)+(D\oplus\Z h)$.
Hence we get a surjective map $\phi\colon D\oplus\Z h\to\Lambda$.
It is enough to prove that $\phi$ is an isomorphism. To prove the injectivity of $\phi$
suppose that $a[h]+\sum_{i,j=1}^{d-1}a_{ij}[L_{ij}]=0$ for some $a_{ij}\in\Z$.
The intersection pairing $\langle [L_{ij}], [L_{rs}]\rangle$
equals $-(d-2)$ if $i=r$ and $j=s$, $0$ if $i\not=r$ and $j\not=s$, and $1$ in all other cases.
Pairing our sum with $L_{0,0}$ we get $a=0$. Next, pairing it with each $L_{0,s}$
and $L_{r,0}$ we see that $\sum_{i=1}^{d-1}a_{is}=\sum_{j=1}^{d-1}a_{rj}=0$ for all $s$ and $r$. Finally, pairing with each $L_{rs}$, where $1\leq r,s\leq d-1$, we obtain $a_{rs}=0$. This
proves that $\phi$ is injective, hence an isomorphism. $\Box$
\bthe \label{Lambda}
Assume that $[L_{f,g}(\sqrt[d]{a}):L_{f,g}]=d$.
Then the natural inclusion of $\Ga_k$-modules
$\Lambda\to\Pic(X_{\bar k})$ gives rise to an isomorphism $\H^1(k,\Lambda)\tilde\lra
\H^1(k,\Pic(X_{\bar k}))$.
\ethe
\brem {\rm
The condition $[L_{f,g}(\sqrt[d]{a}):L_{f,g}]=d$ is a consequence of the stronger
condition $[K_{f,g}(\sqrt[d]{a}):K_{f,g}]=d$, where $K_{f,g}$ is a finite extension of
$L_{f,g}$ explicitly described in \S\ref{Pic}. }
\erem
{\em Proof of Theorem \ref{Lambda}.} The proof consists of several steps.
{\em Step} 1. Since $L_{f,g}$ contains $\Q(\mu_d)$, the extensions
$L_{f,g}(\sqrt[d]{a})/k$ and $L_{f,g}(\sqrt[d]{a})/L_{f,g}$ are Galois.
Write $G=\Gal(L_{f,g}(\sqrt[d]{a})/k)$ and
$H=\Gal(L_{f,g}(\sqrt[d]{a})/L_{f,g})$. By assumption, $H\simeq\Z/d$.
The Galois group $\Gal(\bar k/L_{f,g}(\sqrt[d]{a}))$ acts trivially on
$\Pic(X_{\bar k})$, thus $\Ga_k$ acts on $\Pic(X_{\bar k})$ via
its quotient $G$. Since $\Pic(X_{\bar k})$ is a free abelian group of finite rank,
the inflation map $\H^1(G,\Pic(X_{\bar k}))\to \H^1(k,\Pic(X_{\bar k}))$ is an isomorphism.
To finish the proof it is enough to prove the following statements:
(a$_X$) $\H^1(H,\Pic(X_{\bar k}))=0$;
(b$_X$) $\Pic(X_{\bar k})^H=\Lambda$.
\medskip
{\em Step} 2. Let $Y$ be the surface given by (\ref{X}) with $a=1$.
We note that $H$ acts on $Y_{L_{f,g}}$ by $L_{f,g}$-automorphisms which multiply the coordinates
$x_2$ and $x_3$ by the same root of unity of degree $d$, and leave $x_0$ and $x_1$
invariant.
Absorbing $\sqrt[d]{a}$
into $x_2$ and $x_3$ defines an isomorphism of base changes of
$X$ and $Y$ to $L_{f,g}(\sqrt[d]{a})$. This isomorphism translates the action of $H$
on $\Pic(X_{\bar k})$ induced by the action of $\Ga_k$ on $\bar k$
into the action of $H$
on $\Pic(Y_{\bar k})$ induced by the action of $H$ by automorphisms of $Y_{L_{f,g}}$.
Thus (a$_X$) and (b$_X$) are respectively equivalent to
(a$_Y$) $\H^1(H,\Pic(Y_{\bar k}))=0$;
(b$_Y$) $\Pic(Y_{\bar k})^H=\Lambda$.
\medskip
{\em Step} 3. The surface $Y_{L_{f,g}}$ together with the action of $H$ on $Y_{L_{f,g}}$ by automorphisms
can be defined over a subfield $L_0\subset L_{f,g}$ that is finitely generated over $\Q$.
Choose an embedding of $L_0$ into $\C$, and call $Y_\C$ the resulting complex surface.
This gives an isomorphism of $H$-modules $\Pic(Y_{\bar k})\cong\Pic(Y_\C)$.
The exponential sequence gives an exact sequence of $H$-modules
\begin{equation}
0\to \Pic(Y_\C)\to\H^2(Y_\C,\Z)\to T\to 0,
\label{exp}
\end{equation}
where all terms are free abelian groups of finite rank. The associated exact sequence
of cohomology groups of $H$, in view of semisimplicity of finite-dimensional
representations of $H$ over a field of characteristic zero, shows that (a$_Y$) and (b$_Y$) follow from
(a$_\C$) $\H^1(H,\H^2(Y_\C,\Z))=0$;
(b$_\C$) $\H^2(Y_\C,\Z)^H=\Lambda$.
\noindent These properties are purely topological, so they are invariant under
a continuous real deformation of complex polynomials $f(t)$ and $g(t)$ without multiple roots.
Therefore, it is enough to establish these properties for any chosen $f(t)$ and $g(t)$.
We shall check that (a$_\C$) and (b$_\C$) hold for
the complex Fermat surface $F\subset\P^3_\C$, defined in (\ref{Fermat}),
that is, in the case when $f(t)=g(t)=t^d+1$.
\medskip
{\em Step} 4. We claim that $\H^1(H,\H^2(F,\Z))=0$.
Fix a primitive $2d$-th root of unity $\zeta$ and write $\e=\zeta^2$.
We define $G\simeq(\Z/d)^3$ as the group of automorphisms of $F$
with generators $u_1,u_2,u_3$, where each $u_i$ multiplies $x_i$ by $\e$
and leaves the other variables invariant.
Thus $H$ is the subgroup of $G$ generated by $u_2u_3$.
It is enough to show that $\H^1(H,P)=0$, where
$P$ is the orthogonal complement to the hyperplane section class
$[h]$ in $\H^2(F,\Z)$. This is immediate from the
short exact sequence of $G$-modules
$$0\to P\to \H^2(F,\Z)\to \Z\to 0.$$
Let $\phi(x)\in\Z[x]$ be the polynomial $\sum_{i=0}^{d-1} x^i$. By \cite[Lemma 2.1]{GS22}
(based on work of Pham) there is an isomorphism of $G$-modules
$$P\cong\Z[u_1,u_2,u_3]/I, \ \text{where} \ I=(\phi(u_1), \phi(u_2), \phi(u_3), \phi(u_1u_2u_3)).$$
The group $\H^1(H,P)$ is the quotient of the subgroup of $P$ annihilated by $\phi(u_2u_3)$,
by $(1-u_2u_3)P$. Multiplication by $\phi(u_2u_3)$ on $P/(1-u_2u_3)P$ coincides with
multiplication by $d$. We note that
$$P/(1-u_2u_3)P\cong\Z[u_1,u_2]/(\phi(u_1),\phi(u_2))$$
is a torsion-free abelian group, so we conclude that $\H^1(H,P)=0$.
\medskip
{\em Step} 5. We claim that $\H^2(F_\C,\Z)^H=\Lambda$.
The group $H$ acts trivially on $\Lambda$, hence $\Lambda\subset\H^2(F_\C,\Z)^H$.
By work of Shioda \cite{Sh79,Sh81}, see also \cite[Lemma 2.6]{GS22},
the vector space $P\otimes\C$ is a direct sum of 1-dimensional
eigenspaces $V_\chi$ of $G$, for $\chi\in \widehat G=\Hom(G,\Q/\Z)$, so that
$g(v)=\chi(g)v$ for any $g\in G$ and $v\in V_\chi$. Here $\chi$ can be given by
$\chi(u_1)=l/d$, $\chi(u_2)=m/d$, $\chi(u_3)=n/d$,
where $1\leq l,m,n\leq d-1$ and $l+m+n$ is not divisible by $d$.
Thus $P^H\otimes\C$ is the direct sum of the eigenspaces $V_\chi$ such that $m+n=d$.
The $d^2$ lines in $\Lambda$ can be written as follows:
\[L_{ij}:\quad \zeta x_1-\e^i x_0=x_3-\e^j\zeta x_2=0, \quad\quad i,j=0,\dots,d-1.\]
Following Shioda \cite[p.~732]{Sh81},
for $\chi=(l,d-n,n)\in \widehat G$ we define
\[\omega_\chi=\sum_{i,j=0}^{d-1}\e^{il+jn}[L_{ij}]\in \Pic(F)\otimes\C.\]
Then $g\omega_\chi=\chi(g)\omega_g$ and
$\langle \omega_\chi,\ov{\omega_\chi}\rangle=-d^3$,
hence $\omega_\chi\neq 0$ spans $V_\chi$.
Thus $V_\chi\subset\Lambda\otimes\C$, so that
$P^H\otimes\C\subset\Lambda\otimes\C$. Since $[h]\in\Lambda$,
we deduce that $\H^2(F_\C,\C)^H=\Lambda\otimes\C$.
It follows that
$\Lambda$ is a subgroup of $\H^2(F_\C,\Z)^H$ of finite index.
In particular, $\Lambda$ has finite index in $\Pic(F)^H$.
By a result of Degtyarev \cite[Thm.~1.4]{Deg15}, the abelian group
$\Pic(F)^H/\Lambda$ is torsion-free, thus $\Lambda=\Pic(F)^H$.
Since the abelian group $T$ in (\ref{exp}) is torsion-free, we must have
$\Pic(F)^H=\H^2(F_\C,\Z)^H$, which proves our claim.
This finishes the proof of the theorem. $\Box$
\bco \label{2.4}
Assume that $[L_{f,g}(\sqrt[d]{a}):L_{f,g}]=d$. Then
the isomorphism of Theorem \ref{Lambda} identifies the differential $d_{1,1}$, up to sign,
with the composition of the connecting map
$\partial\colon\H^1(k,\Lambda)\to\H^3(k,\Z)$ attached to the bottom sequence of $(\ref{duo})$,
and $a_*\colon\H^3(k,\Z)\to \H^3(k,\bar k^\times)$ induced by the map sending $1\in \Z$ to
$a\in \bar k^\times$.
\eco
{\em Proof.} This follows from the
well known fact that, up to sign, $d_{1,1}$ is the connecting map attached to the top
sequence of (\ref{duo}), see \cite[Prop.~5.4.5]{CTS21}. $\Box$
\section{Differential $d_{1,1}$} \label{diff}
Let $k_f$ and $k_g$ be the splitting fields of $f$ and $g$, respectively.
Let $G_f=\Gal(k_f/k)$, $G_g=\Gal(k_g/k)$, and $G=\Gal(k_fk_g/k)$.
We denote the $G_f$-module $\Z[V_f]/\Z\sigma_f$ by $M_f$. Similarly, we have a $G_g$-module
$M_g=\Z[V_g]/\Z\sigma_g$.
It is clear that the structure of a $\Ga_k$-module on $\Lambda$ is obtained by inflation
from its structure of a $G$-module.
Passing to the quotient by $\Z$ in the bottom sequence of (\ref{duo})
we obtain a commutative diagram of $G$-modules with exact rows
\begin{equation}
\begin{split}
\xymatrix{
&&&0\ar[d]&0\ar[d]&\\
&&&\Z\ar[d]\ar[r]^\id&\Z\ar[d]&\\
0\ar[r]& \Z\ar[r]^{p_1 \ \ \ \ \ \ \ \ }\ar[d]^{\id}&
\Z[V_f]\oplus\Z[V_g]\ar[r]^{p_2 \ \ \ \ \ }\ar[d]^{\id}&
(\Z[V_f]\otimes \Z[V_g])\oplus\Z h\ar[r]\ar[d]& \Lambda\ar[r]\ar[d]& 0\\
0\ar[r]& \Z\ar[r]^{p_1 \ \ \ \ \ \ \ \ }&
\Z[V_f]\oplus\Z[V_g]\ar[r]&
\Z[V_f]\otimes \Z[V_g]\ar[r]\ar[d]& M_f\otimes M_g\ar[r]\ar[d]& 0\\
&&&0&0&
}
\label{bruckner}
\end{split}
\end{equation}
The commutativity of (\ref{bruckner}) implies that
$\partial\colon\H^1(G,\Lambda)\to \H^3(G,\Z)$ is the composition
\begin{equation}
\H^1(G,\Lambda)\hookrightarrow \H^1(G,M_f\otimes M_g)\stackrel{\partial_M}\lra \H^3(G,\Z),\label{new3}
\end{equation}
where $\partial_M$ is the connecting map attached to the bottom sequence of
(\ref{bruckner}). In the next proposition, under the assumptions of Theorem \ref{f=g}, we
calculate the group $\H^1(G,M_f\otimes M_g)$. Then we determine $\H^1(G,\Lambda)$
as a subgroup of this group. Finally, we compute the connecting map $\partial_M$.
\bpr \label{new1}
Assume that $f$ and $g$ are irreducible polynomials of the same degree
such that their splitting fields $k_f$ and $k_g$ are linearly disjoint over $k$, that is,
$k_f\cap k_g=k$. Let $S_f\subset G_f$ be the stabiliser of a $\bar k$-point of $V_f$, and let
$N_f\subset G_f$ be the normal closure of $S_f$ in $G_f$.
Define $S_g$ and $N_g$ similarly.
Then we have a natural isomorphism
\begin{equation}
\H^1(G, M_f\otimes M_g)\cong((G_f/N_f)^{\rm ab}\otimes (G_g/N_g)^{\rm ab})^*.\label{new5}
\end{equation}
If $\deg(f)=\deg(g)=d$, then
$\H^1(G, M_f\otimes M_g)\cong\Hom((G_f/N_f)^{\rm ab}\otimes (G_g/N_g)^{\rm ab},\Z/d)$.
\epr
{\em Proof}. The exact sequence of $G_f$-modules
$0\to\Z\to\Z[V_f]\to M_f\to 0 $
gives rise to the exact sequence
$$0\to\Z\tilde\lra \Z[V_f]^{G_f}\to M_f^{G_f}\to \H^1(G_f,\Z)=0,$$
where the second arrow is an isomorphism since $f$ is irreducible.
Thus $M_f^{G_f}=0$. Similarly, $M_g^{G_g}=0$.
We also see that $\H^1(G_f,M_f)$ is the kernel of the restriction map
$G_f^*\to S_f^*$. We deduce an isomorphism
\begin{equation}
\H^1(G_f,M_f)=((G_f/N_f)^{\rm ab})^*,\label{new2}
\end{equation}
and a similar isomorphism for $\H^1(G_g,M_g)$.
We have $G=G_f\times G_g$.
The K\"unneth formula for group cohomology in degree 1 gives the following (split) exact
sequence of abelian groups:
$$0\to\bigoplus_{i=0,1}\big( \H^i(G_f,M_f)\otimes \H^{1-i}(G_g,M_g)\big)
\to\H^1(G, M_f\otimes M_g) $$
$$\to\bigoplus_{i=0,1,2}\Tor^\Z_1(\H^i(G_f,M_f),\H^{2-i}(G_g,M_g))\to 0.$$
Since $M_f^{G_f}=0$ and $M_g^{G_g}=0$, this gives an isomorphism
$$\H^1(G, M_f\otimes M_g) \cong
\Tor^\Z_1(\H^1(G_f,M_f),\H^1(G_g,M_g)).$$
An easy calculation shows that for finite abelian groups $A$ and $B$
there is a canonical isomorphism $\Tor^\Z_1(A^*,B^*)\cong (A\otimes B)^*$. Combining these results proves (\ref{new5}).
If $\deg(f)=d$, then the index of $N_f$ in $G_f$ divides $d=[G_f:S_f]$.
This implies the last statement of the proposition.
$\Box$
\brem{\rm
Let $T$ be the torus given by $N_{k(\xi)/k}({\bf x})=N_{k(\rho)/k}({\bf y})\neq 0$.
The pointed affine cone over our surface (\ref{X}) is a closed subset
of the smooth locus $S$ of the affine hypersurface $N_{k(\xi)/k}({\bf x})=aN_{k(\rho)/k}({\bf y})$,
which is a partial compactification of a $k$-torsor for $T$. The resulting
isomorphism of Galois modules $\Pic(S_{\bar k})\cong M_f\otimes M_g$
explains the similarity between Proposition \ref{new1} and \cite[Cor.~3.2]{Wei}.
}
\erem
\bpr \label{new4}
The inclusion in $(\ref{new3})$ and the isomorphism $(\ref{new5})$ give rise to an exact sequence
\begin{equation}
0\to\H^1(G,\Lambda)\to
\Hom((G_f/N_f)^{\rm ab}\otimes (G_g/N_g)^{\rm ab},\Z/d)\stackrel{\chi}\lra
\Hom(G,\Z/d).
\label{new6}
\end{equation}
Here $\chi$ sends a bilinear function
$\varphi\colon (G_f/N_f)^{\rm ab}\otimes (G_g/N_g)^{\rm ab}\to \Z/d$
to the homomorphism $\chi_\varphi\colon G\to\Z/d$ whose value on $(a,b)$, with
$a\in G_f$ and $b\in G_g$, is
$$\chi_\varphi(a,b)=\sum_{y\in G_g/S_g}\widetilde\varphi(a,y)+\sum_{x\in G_f/S_f}\widetilde\varphi(x,b).$$
In particular, $\H^1(G,\Lambda)\cong\Pi_{f,g}$.
\epr
{\em Proof.} In view of (\ref{new3}), $d$ annihilates $\H^1(G,\Lambda)$ as well as
$\H^1(G, M_f\otimes M_g)$. Since $M_f^{G_f}=0$, the exact sequence
$$0\to M_f\otimes M_g\stackrel{[d]}\lra M_f\otimes M_g\to (M_f\otimes M_g)/d\to 0$$
gives rise to an isomorphism
\begin{equation}
\label{may}
\big((M_f\otimes M_g)/d\big)^{G}\cong \H^1(G, M_f\otimes M_g).
\end{equation}
Likewise, by the right hand column of (\ref{bruckner}) we have $\Lambda^{G}=\Z[h]$, so
the exact sequence
$$0\to\Lambda\stackrel{[d]}\lra\Lambda\to\Lambda/d\to 0$$
gives rise to an isomorphism
$(\Lambda/d)^{G}/(\Z/d)[h]\cong \H^1(G, \Lambda)$.
Thus we need to determine the $G$-invariant elements of $(M_f\otimes M_g)/d$ which
lift to $(\Lambda/d)^G$.
The elements of $(M_f\otimes M_g)/d$ bijectively correspond to
the functions $\varphi\colon G\to\Z/d$
that are constant on the left cosets by $S_f\times S_g$, considered modulo sums of functions $G_f/S_f\to\Z/d$ and $G_g/S_g\to\Z/d$. We can eliminate this indeterminacy by
normalising $\varphi$ so that $\varphi(e,y)=\varphi(x,e)=0$ for any $x\in G_f$ and $y\in G_g$.
The action of $h\in G_f$ sends the normalised $\varphi(x,y)$ to
$\varphi(h^{-1}x,y)-\varphi(h^{-1},y)$. Thus a normalised function $\varphi$ is
$G$-invariant if and only if it is bilinear. Equivalently, $\varphi$
is a bilinear function $(G_f/N_f)^{\rm ab}\otimes (G_g/N_g)^{\rm ab}\to\Z/d$.
(This gives another proof of the isomorphism (\ref{new5}).)
The elements of $\Lambda/d$ bijectively correspond to
the functions $\psi\colon G\to\Z/d$
that are constant on the left cosets by $S_f\times S_g$, considered modulo sums
$\sigma(x)+\tau(y)$, where
functions $\sigma\colon G_f/S_f\to\Z/d$ and $\tau\colon G_g/S_g\to\Z/d$ are
such that $$\sum_{x\in G_f/S_f}\si(x)+\sum_{y\in G_g/S_f}\tau(y)=0.$$
A normalised function $\varphi$ lifts to an element of $\Lambda/d$ uniquely up to an
element of $(\Z/d)[h]$, so the lift of $\varphi$ can be chosen to be $\varphi$ itself.
Let $a\in G_f$ and $b\in G_g$.
In $\Lambda/d$ we have $$\chi_\varphi\big((a,b)\big):=(a,b)\varphi-\varphi=
\sum_{y\in G_g/S_g}\varphi(a,y)+\sum_{x\in G_f/S_f}\varphi(x,b)\in(\Z/d)[h].$$
It is clear that $\chi_\varphi$ is a homomorphism $G\to\Z$, which vanishes precisely when
$\varphi$ is $G$-invariant as an element of $\Lambda/d$. $\Box$
\brem \label{bc}
{\rm
Suppose that $f$ and $g$ are cyclic polynomials whose splitting fields are linearly disjoint over $k$.
We have $G_f\simeq G_g\simeq\Z/d$, $N_f=S_f=\{e\}$,
$N_g=S_g=\{e\}$. Then (\ref{new6}) becomes the exact sequence
$$0\to\H^1((\Z/d)^2,\Lambda)\to\Z/d\to \Hom((\Z/d)^2,\Z/d).$$
The generator $1\in\Z/d$ goes to the character $\chi\colon(\Z/d)^2\to\Z/d$
such that $\chi((1,0))=\chi((0,1))$ is the sum of all elements of $\Z/d$.
This is zero if $d$ is odd, and $d/2\in\Z/d$ if $d$ is even.
Thus in this case we have $\H^1(k,\Lambda)\cong 2(\Z/d)$. }
\erem
The Künneth formula for group cohomology gives an exact sequence
$$0\to\H^3(G_f,\Z)\oplus \H^3(G_g,\Z)\to\H^3(G,\Z)\to
\Tor_1^\Z(\H^2(G_f,\Z),\H^2(G_g,\Z))\to 0,$$
see \cite[Exercise 6.1.8]{Weibel}. The restriction
map $\H^3(G,\Z)\to \H^3(G_f,\Z)\oplus \H^3(G_g,\Z)$ splits this sequence, so we get
an isomorphism $\H^3(G,\Z)_{\rm prim}\cong(G_f^{\rm ab}\otimes G_g^{\rm ab})^*$.
\bpr \label{zoe}
The connecting map $\partial_M\colon\H^1(G,M_f\otimes M_g)\to \H^3(G,\Z)$
is the composition
$$((G_f/N_f)^{\rm ab}\otimes (G_g/N_g)^{\rm ab})^*\hookrightarrow
(G_f^{\rm ab}\otimes G_g^{\rm ab})^*\tilde\lra \H^3(G,\Z)_{\rm prim}
\hookrightarrow \H^3(G,\Z).$$
\epr
{\em Proof.} A choice of a $\bar k$-point in $V_f$ defines a surjective map of sets
$G_f\to V_f$ compatible with the (left) action of $G_f$.
The induced map of $G_f$-modules $\Z[V_f]\to\Z[G_f]$ and a similar map for $g$
give rise to a commutative diagram of $G$-modules
\begin{equation}\label{sixterms}
\xymatrix{
0\ar[r]& \Z\ar[r]^{p_1 \ \ \ \ \ \ \ \ }\ar[d]^\id& \Z[V_f]\oplus\Z[V_g]\ar[r]\ar[d]&
\Z[V_f]\otimes \Z[V_g]\ar[r]\ar[d]& M_f\otimes M_g\ar[r]\ar[d]^\kappa& 0\\
0\ar[r]& \Z\ar[r]& \Z[G_f]\oplus\Z[G_g]\ar[r]&
\Z[G_f]\otimes \Z[G_g]\ar[r]&\widetilde M_f\otimes \widetilde M_g\ar[r]& 0}
\end{equation}
where $\widetilde M_f=\Z[G_f]/\Z$ and $\widetilde M_g=\Z[G_f]/\Z$.
We deduce that $\partial_M$ is the composition
$$\H^1(G,M_f\otimes M_g)\stackrel{\kappa_*}\lra \H^1(G,\widetilde M_f\otimes \widetilde M_g)
\stackrel{\partial_{\widetilde M}}\lra \H^3(G,\Z),$$
where $\partial_{\widetilde M}$ is the connecting map of the
bottom exact sequence. Proposition \ref{new1} gives an isomorphism
$\H^1(G,\widetilde M_f\otimes \widetilde M_g)\cong(G_f^{\rm ab}\otimes G_g^{\rm ab})^*$.
The natural map from
$G_f^{\rm ab}\otimes G_g^{\rm ab}$ to $(G_f/N_f)^{\rm ab}\otimes (G_g/N_g)^{\rm ab}$
is surjective, hence $\kappa_*$ is injective.
Let us write $M=(\Z[G_f]\oplus\Z[G_g])/\Z$. Since
$\Z[G_f]\otimes\Z[G_g]\cong\Z[G]$ is a free $\Z[G]$-module, we have a canonical
isomorphism $\H^1(G,\widetilde M_f\otimes \widetilde M_g)\cong\H^2(G,M)$.
The map $\H^2(G,\Z)\to \H^2(G,\Z[G_f]\oplus\Z[G_g])$ is the natural isomorphism
$G^*\tilde\lra G_g^*\oplus G_f^*$, thus the connecting map of the exact sequence of $G$-modules
$$0\to\Z\to\Z[G_f]\oplus\Z[G_g]\to M\to 0$$
sends $\H^2(G,M)$ isomorphically onto $\H^3(G,\Z)_{\rm prim}$.
It remains to check that the resulting isomorphism
$(G_f^{\rm ab}\otimes G_g^{\rm ab})^*\tilde\lra \H^3(G,\Z)_{\rm prim}$
agrees with the isomorphism provided by the Künneth formula.
The group $\H^i(G_f,\Z)$, $i\geq 0$, is the $i$-th cohomology group of the complex
$C_f^\bullet$ of abelian groups where $C_f^i$ is the group of functions $G_f^{i-1}\to \Z$.
The differential $C_f^0\to C_f^1$ sends $n\in\Z$ to $(g-1)n=0$
since $\Z$ is a trivial $G$-module. Thus $C_f^\bullet$ is the direct sum
of $\Z$ in degree 0 and the truncated complex $\tau_{\geq 1}C_f^\bullet$. From the definition of $\widetilde M_f$
it follows that $\H^i(G_g,\widetilde M_f)$ is the $i$-th cohomology group of
$(\tau_{\geq 1}C_f^\bullet)[1]$. For $i\geq 2$ we have
$$\H^i(G,\Z)=\H^i(C_f^\bullet\otimes C_g^\bullet)=\H^i(C_f^\bullet)\oplus
\H^i(C^\bullet_g)\oplus\H^{i-2}\big((\tau_{\geq 1}C_f^\bullet)[1]\otimes
(\tau_{\geq}C_g^\bullet)[1]\big),$$
so the K\"unneth formula isomorphism is given by
$$\H^1\big((\tau_{\geq 1}C_f^\bullet)[1]\otimes
(\tau_{\geq}C_g^\bullet)[1]\big)\to \H^3(C_f^\bullet\otimes C_g^\bullet).$$
On the other hand, we can notice that the bottom 2-extension of the previous diagram (\ref{sixterms})
is equivalent to the Yoneda product of $0\to\Z\to\Z[G_f]\to\widetilde M_f\to 0$ and
$$0\to \widetilde M_f\to \widetilde M_f\otimes\Z[G_g]\to
\widetilde M_f\otimes \widetilde M_g\to 0.$$
Thus the resulting connecting map is the composition
$$\H^1\big((\tau_{\geq 1}C_f^\bullet)[1]\otimes (\tau_{\geq}C_g^\bullet)[1]\big)\to
\H^2\big((\tau_{\geq 1}C_f^\bullet)[1]\otimes C_g^\bullet\big)
\to \H^3(C_f^\bullet\otimes C_g^\bullet),$$
and hence coincides with the K\"unneth formula isomorphism. $\Box$
\brem \label{bc1}
{\rm This continues Remark \ref{bc} and uses the same assumptions.
(1) The cup-product map
\begin{equation}
\H^1(G_f,\Z/d)\otimes\H^1(G_g,\Z/d)\to \H^2(G,\Z/d)_{\rm prim}\label{cupp}
\end{equation}
is surjective. Indeed,
it is clear by functoriality that the image of the cup-product is contained in
$\H^2(G,\Z/d)_{\rm prim}$. Arguing exactly as in the topological
case (see diagram (5.37) in \cite[\S 5.7.2]{CTS21}) we identify the map (\ref{cupp}) with the map
$$\Hom(G_f^{\rm ab},\Z/d)\otimes\Hom(G_g^{\rm ab},\Z/d)\to
\Hom(G_f^{\rm ab}\otimes G_g^{\rm ab},\Z/d),$$
given by multiplication in $\Z/d$. In our case $G_g^{\rm ab}\simeq\Z/d$, so
this is an isomorphism.
(2)
The composition of the embedding $\H^2(G,\Z/d)_{\rm prim}\hookrightarrow \H^2(G,\Z/d)$ with
the connecting map $\delta\colon\H^2(G,\Z/d)\to \H^3(G,\Z)$ is an isomorphism.
Indeed, we have $\H^3(\Z/d,\Z)=0$, hence
$\H^3(G,\Z)_{\rm prim}=\H^3(G,\Z)\simeq\Z/d$,
where the last isomorphism follows immediately from
the K\"unneth formula for group cohomology \cite[Exercise 6.1.8]{Weibel}.
Since $d$ annihilates $\H^2(G,\Z)\simeq(\Z/d)^2$, we have an exact sequence
$$0\to \H^2(G,\Z)\to \H^2(G,\Z/d)\to \H^3(G,\Z)\to 0.$$
Comparing it with similar sequences for $G_f$ and $G_g$ we see that the second arrow
can be identified with the canonical map $\H^2(G_f,\Z/d)\oplus\H^2(G_g,\Z/d)\to \H^2(G,\Z/d)$.
We conclude that $\delta\colon\H^2(G,\Z/d)_{\rm prim}\to \H^3(G,\Z)$ is an isomorphism.
(3)
Combining the results of (1) and (2), we see that the image of
the map $\H^3(G,\Z)\to \H^3(k,\Z)$ is generated by $\delta(\chi_f\cup\chi_g)$,
where $\chi_f$ is a generator of the cyclic group $\H^1(G_f,\Z/d)\simeq\Z/d$ and $\chi_g$ is a generator of
$\H^1(G_g,\Z/d)\simeq\Z/d$.
By Proposition \ref{zoe} the group $\H^1(G,M_f\otimes M_g)\cong \Z/d$ maps isomorphically
onto $\H^3(G,\Z)$. Thus the image of $\H^1(G,M_f\otimes M_g)$ in $\H^3(k,\Z)$
is generated by $\delta(\chi_f\cup\chi_g)$.
(4)
Now suppose that $\Q(\mu_{2d})\subset k$.
Fixing a primitive root, we identify $\Z/d$ with $\mu_d$. Write $f(t)=t^d-b$, $g(t)=t^d-c$,
where $b,c\in k^\times$. Recall that
for $a\in k^\times$ we denote by $[a]\in\H^1(k,\mu_d)$ the image of $a$ under the
connecting map $k^\times\to\H^1(k,\mu_d)$ of the Kummer exact sequence.
The images of $\chi_f$ and $\chi_g$ under inflation are $[b]$ and $[c]$,
respectively.
We claim that the image of
$$\H^1(G,M_f\otimes M_g)\to\H^3(k,\Z)\stackrel{\cup a}\lra\H^3(k,\bar k^\times),$$
where the map $\Z\to\bar k^\times$ sends $1$ to $a$,
is generated by the image of $[a]\cup[b]\cup[c]$ under the map
$\H^3(k,\mu_d^{\otimes 3})\to \H^3(k,\bar k^\times)$.
Indeed, there is a commutative diagram
\begin{equation}\begin{split}
\xymatrix{\H^2(k,\Z/d)\ar[r]^\delta\ar[d]_{\cup[a]}&\H^3(k,\Z)\ar[d]^{\cup a}\\
\H^3(k,\mu_d)\ar[r]&\H^3(k,\bar k^\times)}
\label{difficult}\end{split}
\end{equation}
This is proved exactly as in \cite[\S 1.4.4]{CTS21}.
This implies the claim.
We conclude that for the diagonal surface with coefficients $a_1,a_2,a_3$ the connecting map
$\partial$ sends
$\H^1(k,\Lambda)\simeq 2(\Z/d)$ to the subgroup of $\H^3(k,\bar k^\times)$ generated by
the image of $2[a_1]\cup[a_2]\cup[a_3]$. If condition $(\star)$ holds, then
$d_{1,1}$ sends $\H^1(k,\Pic(X_{\bar k}))\simeq 2(\Z/d)$ to
the subgroup generated by the image of $2[a_1]\cup[a_2]\cup[a_3]$. Furthermore,
if $(\star\star)$ also holds, then $d_{1,1}$ is injective.
}
\erem
\subsubsection*{Proof of Theorem \ref{f=g}}
(i) follows from Theorem \ref{Lambda} and Proposition \ref{new4}.
\noindent(ii)
By Theorem \ref{Lambda}, Proposition \ref{new1}, and the inclusion in (\ref{new3}) it is enough to
prove that $G_f=N_f$. If $d$ is prime, then any transitive group is primitive. Thus under the assumptions of (ii), the stabiliser $S_f$ is a maximal subgroup of $G_f$.
If $N_f$ is a proper subgroup of $G_f$, the inclusion $S_f\subset N_f$ is an equality, so
$S_f$ is normal in $G_f$.
In this case $S_f$ stabilises every root of $f$, hence $S_f=\{e\}$ and
$|G_f|=d$.
If $d$ is prime, we get $G_f\simeq \Z/d$.
If $d$ is not prime, then the trivial subgroup cannot be maximal in $G_f$, a contradiction.
\noindent(iii) follows from Corollary \ref{2.4}, formula (\ref{new3}) and Proposition \ref{zoe}.
$\Box$
\subsubsection*{Proof of Theorem \ref{main} (i)}
We consider the diagonal surface $X$ over $k$ given by (\ref{eqX}).
This is a particular case of the surface (\ref{X}) with $a=-a_2$,
$$f(t)=t^d+a_1=\prod_{i=0}^{d-1}(t-\e^i\xi), \quad
g(t)=t^d+a_3/a_2=\prod_{j=0}^{d-1}(t-\e^j\rho),$$
where $\xi,\rho\in\bar k$ are such that $\xi^d+a_1=\rho^d+a_3/a_2=0$.
We have $L_{f,g}\subset k_L(\xi,\rho)$, where
$L$ is the field of definition of the geometric Picard group of the Fermat surface, and $k_L=kL$.
Condition $(\star)$ implies that we can use Theorem \ref{Lambda}.
The action of $\Ga_k$ on $\Lambda$ factors through
$\mathfrak G=\Gal(k(\mu_d,\xi,\rho)/k)$, hence we have
$$\H^1(k,\Pic(X_{\bar k}))\cong\H^1(k,\Lambda)\cong\H^1(\mathfrak G,\Lambda).$$
Condition $(\star)$ implies that
$G=\Gal(k(\mu_d,\xi,\rho)/k(\mu_d))\cong (\Z/d)^2$, which acts on $\xi$ and $\rho$ by
multiplication by $d$-th roots of unity. Thus $\mathfrak G=G\rtimes \fH$,
where
$$\fH=\Gal(k(\mu_d)/k)=\Gal(k(\mu_d,\xi,\rho)/k(\xi,\rho))\subset (\Z/d)^\times.$$
Consider the semidirect product $G\rtimes(\Z/d)^\times$, where
$(\Z/d)^\times$ acts on $G=(\Z/d)^2$ by multiplication.
The lines in $\Lambda$ are naturally indexed by the elements of $G$.
This equips $\Lambda$ with an action of $G\rtimes(\Z/d)^\times$
which extends the action of $\mathfrak G$.
The right hand column of diagram (\ref{bruckner}) is an exact sequence
of $G\rtimes(\Z/d)^\times$-modules, hence also of $\mathfrak G$-modules
\begin{equation}
0\to\Z\to\Lambda\to M_f\otimes M_g\to 0.\label{monday}
\end{equation}
Let $\Lambda_0\subset \Lambda$ be the subgroup generated by the lines
indexed by the elements of $G\cong (\Z/d)^2$ with both coordinates non-zero.
(In other terms, these are the lines $x_1-\e^i\xi x_0=x_3-\e^j\rho x_2=0$ for $i\neq 0$, $j\neq 0$.)
As a $(\Z/d)^\times$-module, $\Lambda$ is isomorphic to the direct sum $\Z\oplus\Lambda_0$,
so that (\ref{monday}) splits as an exacts equence of $(\Z/d)^\times$-modules.
From (\ref{monday}) we get an exact sequence
\begin{equation}
0\to \Z\to\Lambda^G\to 0\to\H^1(G,\Lambda)\to \H^1(G, M_f\otimes M_g)\to \H^1(G,\Z/d), \label{rrr}
\end{equation}
where we used that $(M_f\otimes M_g)^G=0$. Thus we have $\Lambda^G=\Z$.
Taking $\fH$-invariants we obtain an exact sequence
\begin{equation}
0\to\H^1(G,\Lambda)^\fH\to \H^1(G, M_f\otimes M_g)^\fH\to \H^1(G,\Z/d), \label{grrr}
\end{equation}
If we denote the two factors of $G$ by $G_1\cong\Z/d$ and $G_2\cong\Z/d$, then
by Proposition \ref{new1} the group $\H^1(G,M_f\otimes M_g)$ is canonically isomorphic to
$\Hom(G_1\otimes G_2,\Z/d)\cong\Z/d$. Thus
$n\in (\Z/d)^\times$ acts on this group as multiplication by $-2n$.
We conclude that $\H^1(G,\Lambda)$,
which by Remark \ref{bc} is the subgroup $2(\Z/d)\subset\Z/d$,
inherits the same action of $(\Z/d)^\times$.
Since $d$ annihilates $\H^1(G,M_f\otimes M_g)$ and $\H^1(G,\Lambda)$,
it annihilates $\H^1(\mathfrak G,M_f\otimes M_g)\cong \H^1(G,M_f\otimes M_g)^\fH$
and $\H^1(\mathfrak G, \Lambda)\subset \H^1(G,\Lambda)$.
Thus we have isomorphisms
$$\big((M_f/d)\otimes (M_g/d)\big)^{\mathfrak G}\tilde\lra \H^1(\mathfrak G,M_f\otimes M_g)
\cong \H^1(G,M_f\otimes M_g)^\fH$$
and $(\Lambda/d)^{\mathfrak G}/(\Z/d)[h]\cong \H^1(\mathfrak G, \Lambda)$.
Finally, the reduction of (\ref{monday}) modulo $d$ gives an exact sequence
\begin{equation}
0\to\Z/d\to(\Lambda/d)^G\to \big((M_f/d)\otimes (M_g/d)\big)^G\to\H^1(G,\Z/d).\label{22}
\end{equation}
Since (\ref{monday}) splits as an exact sequence of $\fH$-modules, if $x\in\Lambda/d$
goes to an $\fH$-invariant element of $(M_f/d)\otimes (M_g/d)$, then $x\in(\Lambda/d)^\fH$.
Thus (\ref{22}) gives rise to an exact sequence
\begin{equation}
0\to\Z/d\to(\Lambda/d)^{\mathfrak G}\to \big((M_f/d)\otimes (M_g/d)\big)^{\mathfrak G}
\to\H^1(G,\Z/d).\label{23}
\end{equation}
The cokernel of the second arrow, which is $\H^1(\mathfrak G, \Lambda)$,
is isomorphic to the kernel of the last arrow, which is $\H^1(G,\Lambda)^H$, see
(\ref{grrr}). This proves that $\H^1(k,\Pic(X_{\bar k}))$ is the subgroup of
$2(\Z/d)$ on which the square of the cyclotomic character acts by $1$.
\subsubsection*{Proof of Theorem \ref{main} (ii)}
If $\H^1(k,\Pic(X_{\bar k}))=0$, the statement is clear from the spectral sequence (\ref{ss}).
The statement concerning the case $\Q(\mu_{2d})\subset k$ follows from Remark \ref{bc1}.
Now assume that condition $(\star\star)$ holds. We have a commutative diagram
$$\xymatrix{\H^1(k(\mu_{2d}),\Lambda)\ar[r]&\H^3(k(\mu_{2d}),\bar k^\times)\\
\H^1(k,\Lambda)\ar[r]\ar[u]&\H^3(k,\bar k^\times)\ar[u]}$$
The left hand restriction map is injective: the
$\Gal(\bar k/k(\mu_{2d}))$-invariant subgroup of $\Lambda$ is generated by $[h]$,
but $\H^1(\Gal(k(\mu_{2d})/k),\Z)=0$. By Remark \ref{bc1}
the top horizontal map sends $\H^1(k(\mu_{2d}),\Lambda)\simeq 2(\Z/d)$ to
the subgroup generated by the image of $2[a_1]\cup[a_2]\cup[a_3]$ in
$\H^3(k(\mu_{2d}),\bar k^\times)$, so by
condition $(\star\star)$ it is injective.
Thus the bottom horizontal map is also injective. By Theorem \ref{Lambda}
this implies that $d_{1,1}\colon\H^1(k,\Pic(X_{\bar k}))\to\H^3(k,\bar k^\times)$
is injective. The statement follows from the spectral sequence (\ref{ss}).
It remains to prove that condition $(\star\star)$ holds. For this we may assume that $k_0$
is algebraically closed.
The discrete valuation of $k$ given by the variable $t_3$ gives rise to the residue
maps $\H^3(k,\bar k^\times)\to \H^2(k_0(t_1,t_2),\Q/\Z)$, see \cite[\S 6.3]{GSz},
and $\H^3(k,\mu_d)\to\H^2(k_0(t_1,t_2),\Z/d)$, see \cite[\S 6.8]{GSz}.
By \cite[Prop.~6.8.9]{GSz} the following diagram {\em anticommutes}:
$$\xymatrix{\H^3(k,\bar k^\times)\ar[r]& \H^2(k_0(t_1,t_2),\Q/\Z)\\
\H^3(k,\mu_d)\ar[r]\ar[u]&\H^2(k_0(t_1,t_2),\Z/d)\ar[u]}$$
Since $k_0$ is algebraically closed of characteristic zero, we have an isomorphism
of $\Ga_k$-modules $\Q/\Z\cong (\Q/\Z)(1)$, which induces an isomorphism
$\H^2(k_0(t_1,t_2),\Q/\Z)\cong \H^2(k_0(t_1,t_2),\Q/\Z(1))$. The cokernel of
the natural inclusion $\Q/\Z(1)\to \bar k^\times$ is uniquely divisible,
hence is a $\Q$-vector space. This implies that the induced map
$\H^2(k_0(t_1,t_2),\Q/\Z(1))\to \H^2(k_0(t_1,t_2),\bar k^\times)$ is an isomorphism.
Combining the previous diagram with a similar (anticommuting)
diagram for the discrete valuation given by the variable $t_2$ we obtain
$$\xymatrix{\H^3(k,\bar k^\times)\ar[r]& \H^2(k_0(t_1,t_2),\Q/\Z)\ar[r]^\cong&
\H^2(k_0(t_1,t_2),\bar k^\times)\ar[r]&\H^1(k_0(t_1),\Q/\Z)\\
\H^3(k,\mu_d)\ar[r]\ar[u]&\H^2(k_0(t_1,t_2),\Z/d)\ar[u]\ar[r]^\cong&
\H^2(k_0(t_1,t_2),\mu_d)\ar[u]\ar[r]&\H^1(k_0(t_1),\Z/d)\ar[u]}$$
It is easy to see that the right hand vertical map is injective. Using \cite[Prop.~6.8.7]{GSz}
we obtain that, up to sign, the image of $[t_1]\cup[t_2]\cup[t_3]$ in $\H^2(k_0(t_1,t_2),\Z/d)$
is $[t_1]\cup[t_2]$, and that, again up to sign, $[t_1]\cup[t_2]$ goes to
$[t_1]\in \H^1(k_0(t_1),\Z/d)$, which has order $d$. This clearly implies
that the order of the image of $[t_1]\cup[t_2]\cup[t_3]$ in $\H^3(k,\bar k^\times)$
is $d$, and so finishes the proof. $\Box$
\subsubsection*{Proof of Theorem \ref{main} (iii)}
Recall that $G\simeq(\Z/d)^3$ is the group of automorphisms of the Fermat surface $F$
with generators $u_1,u_2,u_3$, where $u_i$ multiplies $x_i$ by $\e$
and leaves the other variables invariant.
Note that $(P\otimes\C)^G=0$ implies that $\H^2(F_\C,\Q)^G=\Q[h]$,
where $[h]$ is the hyperplane class. The intersection index of $[h]$ with the class of a
line on $F_\C$ is $1$, hence $\H^2(F_\C,\Z)^G=\Z[h]$. In particular, we obtain
$\Pic(F_\C)^G=\Z[h]$. This implies $\Pic(X_{\bar k})^{\Ga_k}=\Z[h]$.
But $[h]$ is contained in the image of the natural map
$\Pic(X)\to\Pic(X_{\bar k})^{\Ga_k}$, thus this map is an isomorphism.
The spectral sequence (\ref{ss}) now gives the injectivity of $\Br(k)\to\Br(X)$.
$\Box$
\brem \label{3.6}
{\rm
One can apply the same technique to the cases not covered by Theorems \ref{f=g}
and \ref{main}. For example, suppose that $f$ and $g$ are irreducible polynomials
such that $k_f=k_g$ and $G_f=G_g\simeq\Z/d$. Thus $G=\Gal(k_fk_g/k)\simeq\Z/d$.
Breaking the bottom row of (\ref{duo}) into two short exact sequences
\begin{equation}
0\to \Z\to \Z[V_f]\oplus\Z[V_g]\to M\to 0,\quad
0\to M\to (\Z[V_f]\otimes \Z[V_g])\oplus\Z \to \Lambda\to 0 \label{dve}
\end{equation}
and using
that $\Z[V_f]$ and $\Z[V_g]$ are free $\Z[G]$-modules we obtain embeddings
$$\H^1(G,\Lambda)\hookrightarrow\H^2(G,M)\hookrightarrow\H^3(G,\Z)\cong\H^1(\Z/d,\Z)=0.$$
Assuming $[L_{f,g}(\sqrt[d]{a}):L_{f,g}]=d$ we obtain
$\H^1(k,\Pic(X_{\bar k}))=0$ by Theorem \ref{Lambda}.
}
\erem
\brem \label{3.7}
{\rm
Another similar case is when $f$ is irreducible with $G_f\simeq\Z/d$, but $g$
is a product of linear factors. We have $G=\Gal(k_fk_g/k)\simeq\Z/d$.
The right hand column of (\ref{bruckner}) gives $\Lambda^G=\Z h$.
We also have $N\Lambda=\Z h$, where $N\in\Z[G]$ is the norm.
Thus the $0$-th Tate cohomology group $\widehat\H^0(G,\Lambda)$ is zero.
By periodicity of cohomology of cyclic groups we have $\H^2(G,\Lambda)=0$. Now
the long exact sequence attached to the second sequence in (\ref{dve})
gives $\H^1(G,M)=0$ and an exact sequence
$$0\to\H^1(G,\Lambda)\to\H^2(G,M)\to \Z/d\to 0.$$
The first exact equence in (\ref{dve}) gives
$$0\to\Z/d\to (\Z/d)^d\to \H^2(G,M)\to 0.$$
We deduce an isomorphism $\H^1(G,\Lambda)\simeq(\Z/d)^{d-2}$.
Assuming $[L_{f,g}(\sqrt[d]{a}):L_{f,g}]=d$ we obtain
$\H^1(k,\Pic(X_{\bar k}))\simeq(\Z/d)^{d-2}$ by Theorem \ref{Lambda}.
}
\erem
\section{Transcendental Brauer group} \label{3}
We start with a simple general lemma.
\ble \label{useful}
Let $X$ be a smooth, integral variety over a field $k$ of characteristic zero.
If for every prime $\ell$ and for every large integer $n$
the cycle class map induces an isomorphism
$(\NS(X_{\bar k})/\ell^n)^{\Ga_k}\tilde\lra \H^2_\et(X_{\bar k},\mu_{\ell^n})^{\Ga_k}$,
then the map $\Br(X)\to\Br(X_{\bar k})$ is zero.
\ele
{\em Proof.} For every prime power $\ell^n$
the Kummer exact sequences for $X$ and $X_{\bar k}$ give rise to
the following commutative diagram with exact rows
$$\xymatrix{0\ar[r]&(\NS(X_{\bar k})/\ell^n)^{\Ga_k}\ar[r]&
\H^2_\et(X_{\bar k},\mu_{\ell^n})^{\Ga_k}\ar[r]&
\Br(X_{\bar k})[\ell^n]^{\Ga_k}&\\
&&\H^2_\et(X,\mu_{\ell^n})\ar[r]\ar[u]&\Br(X)[\ell^n]\ar[u]\ar[r]&0}$$
Since $X$ is smooth, $\Br(X)$ is a torsion group \cite[Lemma 3.5.3]{CTS21}. Thus it is enough to
prove that the right hand vertical map is zero for all $\ell$ and all large
enough integers $n$. By the commutativity of the diagram, this holds if
the second map in the top exact sequence is an isomorphism. $\Box$
\bthe \label{friday}
Let $k_0$ be an algebraically closed field of characteristic zero, and let
$k$ be a field extension of $k_0$.
Let $k\subset K\subset\bar k$ be a finite Galois extension of $k$ with Galois group
$\g=\Gal(K/k)$. Let $c\colon\Ga_k\to\g$ be the $1$-cocycle for the trivial action of $\Ga_k$ on $\g$
given by the natural surjective homomorphism $\Ga_k\to\g$.
Let $Y$ be a smooth, projective, integral variety over $k_0$
such that $\H^3_\et(Y,\Z_\ell)$ is torsion-free for all primes $\ell$.
Suppose that $\g$ acts on $Y$ by automorphisms so that for every prime $\ell$ we have
{\rm (a)} $\H^1(\g,\H^2_\et(Y,\Z_\ell)/{\rm tors})=0$;
{\rm (b)} the cycle class map
$c_1\colon(\NS(Y)\otimes\Q_\ell)^\g\to\H^2_\et(Y,\Q_\ell)^\g$ is surjective.
\noindent Let $X$ be the $(K/k)$-twist of $Y_k$ by the $1$-cocycle $c\colon\Ga_k\to\g$.
Then $\Br(X)\to\Br(X_{\bar k})$ is the zero map.
\ethe
{\em Proof.} Since $\Br(X)$ is a torsion group, it is enough to prove the triviality of
the map of $\ell$-primary torsion subgroups $\Br(X)\{\ell\}\to\Br(X_{\bar k})\{\ell\}$ for all $\ell$.
From the definition of $X$ we see that there is an
isomorphism $X_{\bar k}\cong Y_{\bar k}$ under which
the action of $g\in\Ga_k$ on $X_{\bar k}$ is the same as the action of $g$
on $Y_{\bar k}$ followed by the action of $c(g)\in \g$. The same holds for the induced
action on cohomology: for any prime $\ell$ and any $n\geq 1$, $i\geq 0$ we have
isomorphisms
$\H^i_\et(X_{\bar k},\mu_{\ell^n})\cong\H^i_\et(Y_{\bar k},\mu_{\ell^n})$
translating the natural action of $g$ on $\H^i_\et(X_{\bar k},\mu_{\ell^n})$ into the natural
action of $g$ on $\H^i_\et(Y_{\bar k},\mu_{\ell^n})$ followed by the action of $c(g)$.
Since $k_0\subset \bar k$ is an extension of algebraically closed fields, and
$Y_{\bar k}$ is obtained from $Y$ by base change, we have a canonical
isomorphism $\H^i_\et(Y,\mu_{\ell^n})\to \H^i_\et(Y_{\bar k},\mu_{\ell^n})$, thus
$\H^i_\et(Y_{\bar k},\mu_{\ell^n})$ is a trivial $\Ga_k$-module. It follows that
the $\Ga_k$-module structure on $\H^i_\et(X_{\bar k},\mu_{\ell^n})$
is obtained from the $\g$-module $\H^i_\et(Y,\mu_{\ell^n})$ by inflation.
Using the Kummer sequence we identify the inclusion of
$\Ga_k$-modules $\NS(X_{\bar k})/\ell^n\hookrightarrow \H^2_\et(X_{\bar k},\mu_{\ell^n})$
with the inclusion of $\g$-modules
$\NS(Y)/\ell^n\hookrightarrow \H^2_\et(Y,\mu_{\ell^n})$.
In view of Lemma \ref{useful}, it is enough to prove that the induced map
of $\g$-invariant subgroups is an isomorphism.
Let us write $\NS(Y)_\ell:=\NS(Y)\otimes\Z_\ell$.
We can choose $n$ so that the torsion subgroup of $\H^2_\et(Y,\Z_\ell)$
is annihilated by $\ell^n$. Then the cycle class map and multiplication by $\ell^n$
give rise to a commutative diagram with exact rows
$$\xymatrix{0\ar[r]& \H^2_\et(Y,\Z_\ell)/{\rm tors}\ar[r]& \H^2_\et(Y,\Z_\ell)\ar[r]&
\H^2_\et(Y,\mu_{\ell^n})\ar[r]&0\\
0\ar[r]&\NS(Y)_\ell/{\rm tors}\ar[r]\ar[u]&\NS(Y)_\ell\ar[r]\ar[u]&\NS(Y)/\ell^n\ar[r]\ar[u]&0
}$$
where we used that $\H^3_\et(Y,\Z_\ell)_{\rm tors}=0$.
By condition (a) we obtain the following commutative diagram with exact rows
$$\xymatrix{
0\ar[r]& \big(\H^2_\et(Y,\Z_\ell)/{\rm tors}\big)^\g\ar[r]&\H^2_\et(Y,\Z_\ell)^\g\ar[r]&
\H^2_\et(Y,\mu_{\ell^n})^\g\ar[r]& 0\\
0\ar[r]&\big(\NS(Y)_\ell/{\rm tors}\big)^\g\ar[r]\ar[u]&\big(\NS(Y)_\ell\big)^\g\ar[r]\ar[u]&
\big(\NS(Y)/\ell^n\big)^\g\ar[u]&
}$$
The desired surjectivity of $(\NS(Y)/\ell^n)^\g\to \H^2_\et(Y,\mu_{\ell^n})^\g$
follows from the surjectivity of $(\NS(Y)_\ell)^\g \to \H^2_\et(Y,\Z_\ell)^\g$, so let us
establish this.
The cokernel of $\NS(Y)_\ell \to \H^2_\et(Y,\Z_\ell)$
is torsion-free, thus we have a commutative diagram of $\g$-modules with exact rows
$$\xymatrix{
0\ar[r]&\H^2_\et(Y,\Z_\ell)_{\rm tors}\ar[r]&\H^2_\et(Y,\Z_\ell)\ar[r]&
\H^2_\et(Y,\Z_\ell)/{\rm tors}\ar[r]&0\\
0\ar[r]&\NS(Y)\{\ell\}\ar[r]\ar[u]^{\cong}&\NS(Y)_\ell\ar[r]\ar[u]&
\NS(Y)_\ell/{\rm tors}\ar[r]\ar[u]&0
}$$
This gives rise to the following commutative diagram with exact rows
$$\xymatrix{0\ar[r]&\H^2_\et(Y,\Z_\ell)_{\rm tors}^\g\ar[r]&
\H^2_\et(Y,\Z_\ell)^\g\ar[r]&
\big(\H^2_\et(Y,\Z_\ell)/{\rm tors}\big)^\g\ar[r]&
\H^1(\g,\H^2_\et(Y,\Z_\ell)_{\rm tors})\\
0\ar[r]&\NS(Y)\{\ell\}^\g\ar[r]\ar[u]^{\cong}&
\big(\NS(Y)_\ell\big)^\g\ar[r]\ar[u]&
\big(\NS(Y)_\ell/{\rm tors}\big)^\g\ar[r]\ar[u]&
\H^1(\g, \NS(Y)\{\ell\})\ar[u]^{\cong}
}$$
Condition (b) implies the surjectivity of $\big(\NS(Y)_\ell/{\rm tors}\big)^\g\to
\big(\H^2_\et(Y,\Z_\ell)/{\rm tors}\big)^\g$. Now
the surjectivity of
$(\NS(Y)_\ell)^\g \to \H^2_\et(Y,\Z_\ell)^\g$
follows from the 5-Lemma. $\Box$
\subsubsection*{Proof of Theorem \ref{n}}
For any extension $k\subset k'\subset\bar k$ we have a commutative diagram
\begin{equation}
\begin{split}
\xymatrix{\Br(X_{k'})\ar[r]&\Br(X_{\bar k})\\
\Br(X)\ar[r]\ar[u]&\Br(X_{\bar k})\ar[u]_{\rm id}}\label{factor}
\end{split}
\end{equation}
Thus, in order to prove that the map $\Br(X)\to\Br(X_{\bar k})$ is zero,
we can replace $k_0$ by its algebraic closure in $\bar k$, and so
assume without loss of generality that $k_0$ is algebraically closed.
Let $Y$ be the surface (\ref{X}) with $a=1$ over $k_0$.
By \cite[XI.1.8]{SGA} we have $\Pic(Y)_{\rm tors}=0$, hence $\H^2_\et(Y,\Z_\ell)_{\rm tors}=0$
for all $\ell$. Poincar\'e duality implies that $\H^3_\et(Y,\Z_\ell)_{\rm tors}=0$
for all $\ell$.
%AS the next sentence has been corrected
This allows us to
apply Theorem \ref{friday} to $Y$, the extension $k=k_0(a)$ of $k_0$ of transcendence degree 1,
and the group $\g=H=\Gal(k(\sqrt[d]{a})/k)$ acting on $Y$ as in
the proof of Theorem \ref{Lambda}.
%
Indeed, conditions (a) and (b) hold, as follows from
comparison theorems between classical and
$\ell$-adic \'etale cohomology and statements $({\rm a}_\C)$ and $({\rm b}_\C)$
in the proof of Theorem \ref{Lambda}.
The last statement follows from Theorem \ref{f=g} (ii). $\Box$
\subsubsection*{Proof of Theorem \ref{co1}}
We are in a particular case of the situation considered in Theorem \ref{n}, so we can apply
this theorem to deduce that the map $\Br(X)\to\Br(X_{\bar k})$ is zero.
It remains to apply Theorem \ref{main} (ii) and (iii).
$\Box$
\subsubsection*{Further applications}
Our methods can be applied to other, `less generic' diagonal surfaces. Here we present several
examples, leaving it to the interested reader to explore other cases.
Let $k$ be a field containing $L$, the field of definition of
the geometric Picard group of the Fermat surface of degree $d$ over $\Q$.
In particular, $\Q(\mu_{2d})\subset k$.
\medskip
\noindent{\bf Example 1} Let $t\in k^\times$ be such that the class of $t$ in
$k^\times/k^{\times d}$ has order $d$.
Let $X\subset\P^3_k$ be the surface
$$x_0^d+x_1^d+tx_2^d+tx_3^d=0.$$
Theorem \ref{Lambda} implies that $\H^1(k,\Pic(X_{\bar k}))=0$.
If there is a subfield $k_0\subset k$ such that $k=k_0(t)$ has transcendence degree 1 over $k_0$,
then $\Br(X)=\Br_1(X)$ by
Theorem \ref{n}. Since $X$ has a $k$-point,
the map $\Br(k)\to\Br(X)$ is an isomorphism.
\medskip
\noindent{\bf Example 2} Let $s,t\in k^\times$ be such that the subgroup of
$k^\times/k^{\times d}$ generated by the classes of $s$ and $t$ is isomorphic to $(\Z/d)^2$.
Let $X\subset\P^3_k$ be the surface
$$x_0^d+sx_1^d+tx_2^d+s^rtx_3^d=0,$$
where $r$ is an integer coprime to $d$.
Remark \ref{3.6} gives $\H^1(k,\Pic(X_{\bar k}))=0$.
If there is a subfield $k_0\subset k$ such that $k=k_0(t)$ has transcendence degree 1 over $k_0$,
then $\Br(X)=\Br_1(X)$ by Theorem \ref{n}, thus $\Br(X)=\Br_0(X)$.
Now assume that $r=1$ and there is a subfield $k_1\subset k$ such that $k=k_1(s,t)$
has transcendence degree 2 over $k_1$. In this case,
by Proposition \ref{suppl} below, the kernel of
the map $\Br(k)\to\Br(X)$ is generated by
the class of the cyclic algebra $(s,t)_d$, so is isomorphic to $\Z/d$.
\medskip
\noindent{\bf Example 3} Let $s,t\in k^\times$ be such that the subgroup of
$k^\times/k^{\times d}$ generated by the classes of $s$ and $t$ is isomorphic to $(\Z/d)^2$.
Let $X\subset\P^3_k$ be the surface
$$x_0^d+x_1^d+tx_2^d+stx_3^d=0.$$
Remark \ref{3.7} gives $\H^1(k,\Pic(X_{\bar k}))\simeq(\Z/d)^{d-2}$.
Since $X$ has a $k$-point, the map $\Br(k)\to\Br(X)$ is injective and
$\Br_1(X)/\Br(k)\simeq(\Z/d)^{d-2}$. Moreover, this group is generated by
the unramified classes
$$\left(s,\frac{x_0-\zeta^{2n+1}x_1}{x_0-\zeta x_1}\right)_d,$$
%prove the last statement!
where $n=1,\ldots,d-2$. This generalises \cite[Thm.~1.1]{U}.
If there is a subfield $k_0\subset k$ such that $k=k_0(t)$ has transcendence degree 1 over $k_0$,
we have $\Br(X)=\Br_1(X)$ by Theorem \ref{n},
so in this case $\Br(X)/\Br(k)\simeq(\Z/d)^{d-2}$.
\brem
{\rm When $d=3$ and $4$, the above examples may be compared with those results obtained by Colliot-Thélène--Kanevsky--Sansuc and Bright. In the cubic case, Examples 1 and 2 agree with the first case and Example 3 agrees with the last case in \cite[Prop.~1]{CTKS}. In the quartic case,
Examples 1, 2, 3 agree with entries E10, E8, E20 in \cite[A.5]{Bri02}, respectively.}
\erem
\bpr \label{suppl}
Let $k$ be a field containing $\Q(\mu_{2d})$. Assume that $s,t\in k^\times$ satisfy the following conditions:
{\rm (i)} the subgroup of $k_L^\times/(k_L^\times)^d$
generated by the classes of $s$ and $t$ is isomorphic to $(\Z/d)^2$;
{\rm (ii)} the class of the cyclic algebra $(s,t)_d$ in $\Br(k)$ has order $d$.
\noindent Let $X\subset\P^3_k$ be the surface
$$x_0^d-sx_1^d-tx_2^d+stx_3^d=0.$$
Then the kernel of $\Br(k)\to\Br(X)$ is generated by the class of $(s,t)_d$.
For example, this holds when $k=k_0(s,t)$ is the purely transcendental extension
of degree $2$ of a subfield $k_0\subset k$.
\epr
{\em Proof.} Let us prove that the class of $(s,t)_d\in\Br(k)$ goes to zero in $\Br(X)$.
Note the kernel of $\Br(k)\to\Br(X)$ is equal to the kernel of $\Br(k)\to\Br(U)$,
where $U\subset X$ is the open subscheme given by $(x_0^d-sx_1^d)x_0\neq 0$.
Write $K=k(\sqrt[d]{s})$ and let $V$ be the $k$-torsor for the norm 1 torus
${\rm R}^1_{K/k}(\G_m)$ given by ${\rm N}_{K/k}({\bf x})=t$.
There is an obvious morphism $q\colon U\to V$, hence $\Br(k)\to\Br(U)$
is a composition $\Br(k)\to \Br(V)\to\Br(U)$. Thus it is enough to show that
the class of $(s,t)_d$ goes to zero in $\Br(V)$. By \cite[Prop.~7.1.12]{CTS21}
the variety $V$ is birationally equivalent to the Severi--Brauer variety attached to
$(s,t)_d$. By a theorem of Amitsur, see \cite[Thm.~5.4.1]{GSz},
the class of $(s,t)_d$ generates the kernel of $\Br(k)\to \Br(V)$.
In view of the spectral sequence (\ref{ss}) and condition (ii)
it remains to show that the index of $\Pic(X)$ in $\Pic(X_{\bar k})^{\Ga_k}$ divides $d$.
By condition (i) we have $\Pic(X_{\bar k})^{\Ga_k}=\Lambda^{\Ga_k}$.
The Galois group $\Ga_k$ acts on $\Div_{Z_{\bar k}}(X_{\bar k})$, and hence
also on $\Lambda$, via its quotient $\Z/d$, hence $\Lambda^{\Ga_k}=\Lambda^{\Z/d}$.
From the first exact sequence of (\ref{dve}) we obtain $\H^1(\Z/d,M)\simeq\Z/d$.
Now the second exact sequence of (\ref{dve}) gives an exact sequence
$$\Div_{Z_{\bar k}}(X_{\bar k})^{\Ga_k}\to \Lambda^{\Ga_k}\to \Z/d\to 0.$$
Recall that the first arrow sends a divisor to its class in $\Pic(X_{\bar k})$.
It is clear that $\Div_{Z_{\bar k}}(X_{\bar k})^{\Ga_k}=\Div_{Z}(X)$.
The map $\Div_{Z}(X)\to \Pic(X_{\bar k})^{\Ga_k}$ factors through
$\Pic(X)\to\Pic(X_{\bar k})^{\Ga_k}$, so we are done. $\Box$
\section{Fields of definition}\label{Pic}
We denote the Jacobian of a smooth projective curve $C$ by $J(C)$.
For a separable polynomial $f(t)\in k[t]$ of degree $d$ we write $C_f$ for the plane curve over $k$ defined by $y^d=f(x_1/x_0)x_0^d$.
Recall from the introduction that $L_{f,g}$ is the compositum of $k(\mu_d)$ and
the field of definition of $\Pic(X_{\bar k})$, where $X$ is given by (\ref{X}) with $a=1$.
We now give a computable `upper bound' for $L_{f,g}$. The number 3 in
the following proposition can be replaced by any larger integer.
\bpr\label{Lfg}
Let $K_{f,g}$ be the compositum of $k(\mu_d)$, $k_f$, $k_g$,
and the fields of definition of the
$3$-torsion subgroups $J(C_f)[3]$ and $J(C_g)[3]$. Then $L_{f,g}\subset K_{f,g}$.
\epr
{\em Proof.} Writing $C_g$ as the curve $v^d=g(u_1/u_0)u_0^d$, there is
a dominant rational map from $C_f\times C_g$ to $X$
sending $(x_0,x_1,y,u_0,u_1,v)$ to $(x_0 v,x_1 v,yu_0,yu_1)$. Unraveling its structure
(see \cite[Prop.~2.4, Rem.~2.7]{KSh} which develops \cite{Sas}),
one gets an embedding of $\Ga_k$-modules:
$$\H^2_\et(X_{\bar k},\Q_\ell(1))\hookrightarrow
\H^2_\et(C_{f,\bar k}\times C_{g,\bar k},\Q_\ell(1))\oplus(\Q_\ell[V_f]\otimes\Q_\ell[V_g]).$$
It is compatible with the cycle class map, so we obtain an embedding of $\Ga_k$-modules
$$\Pic(X_{\bar k})\otimes\Q_\ell\hookrightarrow
\big(\NS(C_{f,\bar k}\times C_{g,\bar k})\otimes\Q_\ell\big)
\oplus(\Q_\ell[V_f]\otimes\Q_\ell[V_g]).$$
It is well known (see, e.g.~\cite[Prop.~5.7.3]{CTS21}) that the $\Ga_k$-module
$\NS(C_{f,\bar k}\times C_{g,\bar k})$ is isomorphic to
$\Z^2\oplus\Hom(J(C_{f,\bar k}),J(C_{g,\bar k}))$.
Thus $L_{f,g}$ is contained in the compositum of $k_f$, $k_g$, and the field
of definition of $\Hom(J(C_{f,\bar k}),J(C_{g,\bar k}))$. The latter
is contained in the field of definition of
$\Hom(J(C_f)[n],J(C_g)[n])$, for any $n\geq 3$, see \cite[Prop.~2.3]{Sil}.
$\Box$
\medskip
We now focus on $L$, the field of definition of the geometric Picard group of
the Fermat surface $F$ of degree $d$ over $\Q$.
From \cite[\S2]{GS22}, we recall a few facts about the Galois action on $\Pic(F_{\ov\Q})$. Set $E=\Q(\mu_d)$ and fix a primitive $d$-th root of unity $\e\in E$.
The group $G\simeq(\Z/d)^3$ generated by $u_1,u_2,u_3$ acts via automorphisms on $F_{\ov\Q}$ where $u_i$ multiplies $x_i$ by $\e$.
Define $S_\flat$ as the set of quadruples $(a_0,a_1,a_2,a_3)\in\{1,\dots,d-1\}^4$ such that $\sum_{i=0}^3 (ta_i)\bmod d=2d$ for all $t\in(\Z/d)^\times$, where $(ta_i)\bmod d$
is understood as an integer in $[1,d-1]$.
An element $(a_i)\in S_\flat$ gives rise to a character $\chi\colon G\to \mu_d$ such that
$\chi(u_i)=\e^{a_i}$.
One then has a decomposition into 1-dimensional $G$-modules
$$\Pic(F_{\ov\Q})\otimes_\Z E=Eh\oplus\bigoplus_{\chi\in S_\flat} V_\chi.$$
Call $\chi\in S_\flat$ \emph{primitive} if $\gcd(a_i)=1$.
When a positive integer $d'$ divides $d$
we have a morphism $f\colon F\to F'$ where $F'$ is the Fermat surface of degree $d'$,
given by $f(x_0,x_1,x_2,x_3)=(x_0^m,x_1^m,x_2^m,x_3^m)$.
For $(a_i)\in S_\flat$ set
$m=\gcd(a_i)$, $d'=d/m$, and $\chi'=(a_0',a_1',a_2',a_3')$ where $a_i'=a_i/m$.
Then $\chi'$ is primitive and
$f^*\colon V_{\chi'}\tilde\lra V_{\chi}$ is an isomorphism of $\Ga_k$-modules.
Let $\mathfrak p$ be a prime of $E$ which does not divide $d$. Let $\psi\colon\F_{\mathfrak p}^\times\to\mu_d$ be the $d$-th power residue symbol. This is the multiplicative character which assigns to $\F_{\mathfrak p}^\times$ the unique $d$-th root of unity in $E$ whose reduction in $\F_{\mathfrak p}$ equals $x^{(N({\mathfrak p})-1)/d}$,
where $N({\mathfrak p})=|\F_{\mathfrak p}|$. For $\chi\in S$, define the \emph{Jacobi sum}
$$J_{\chi}(\mathfrak p)=
\sum_{x_1+x_2+x_3=0}\psi(x_1)^{a_1}\psi(x_2)^{a_2}\psi(x_3)^{a_3}.$$
Then $\Frob_\mathfrak p\in\Gal(\ov\Q/E)$ acts on $V_\chi$ via multiplication by\footnote{In
\cite[\S 2.5]{GS22} $\psi(-1)$ should be replaced by $\psi(-1)^r$ on p.~1320, lines 8, 14,
in Proposition 2.9, and on p.~1321, line $-7$.}
$$h_{\chi}(\mathfrak p)=\psi(-1)^{a_0}N(\mathfrak p)^{-1} J_\chi(\mathfrak p).$$ For an ideal
$\mathfrak m$ of $E$, let $I(\mathfrak m)$ be the group of ideals coprime to $\mathfrak m$.
One can multiplicatively extend the domain of $h_\chi$ to $I(d\mathfrak m)$.
By a \emph{Größencharacter} modulo $\mathfrak m$ in the sense of \cite[Ch.~VII, Def.~6.1]{N}, we understand a character $$h:I(\mathfrak m)\to S^1=\left\{z\in\C\middle|\,|z|=1\right\}$$ such that there exist characters
$$h^{\rm fin}:(O_E/\mathfrak m)^\times\to S^1, \quad\quad h^\infty:(E\otimes_\Q \R)^\times\to S^1$$
satisfying $h((a))=h^{\rm fin}(a)h^\infty(a)$ for all $a\in O_E$ coprime to $\mathfrak m$. (Such Größencharacters bijectively correspond to unitary characters of the id\`ele class group of $E$.) For a given Größencharacter, the smallest $\mathfrak m$ as above is called its \emph{conductor}.
By a \emph{generalised Dirichlet character} modulo $\mathfrak m$, we understand a character of the ray class group $\Cl(\mathfrak m)=I(\mathfrak m)/P(\mathfrak m)$, where $P(\mathfrak m)$
is the group of principal ideals coprime to $\mathfrak m$.
\bpr\label{field}
Let $F$ be the Fermat surface of degree $d$ over $\Q$. Let $L$ be the field of definition of
$\Pic(F_{\ov \Q})$.
{\rm (i)} If $d\leq 4$ or ${\rm gcd}(6,d)=1$, then $L=\Q(\mu_{2d})$.
{\rm (ii)} If $d$ is not divisible by any of the integers between $12$ and $180$ in the table of
{\rm \cite[p.~727]{Sh81}}, then $L=\Q(\mu_{2d},2^{2/d})$ if $d$ is $2$ or $4$ modulo $6$,
$L=\Q(\mu_{2d},3^{3/d})$ if $d$ is $3$ modulo $6$, and
$L=\Q(\mu_{2d},2^{2/d}, 3^{3/d})$ if $d$ is divisible by $6$.
%If $2$ divides $d$ but $3,5$ and $7$ do not divide $d$, then $L=\Q(\mu_{2d},2^{2/d})$.
%If $3$ divides $d$ but $2,5$ and $7$ do not divide $d$, then $L=\Q(\mu_{2d},3^{3/d})$.
%If $6$ divides $d$ but $4,5,7,9,11$ and $13$ do not divide $d$, then $L=\Q(\mu_{2d},2^{2/d}, 3^{3/d})$.
\epr
{\em Proof.}
{\rm (i)} Any line
in $F_{\ov\Q}$ is a unique effective divisor in its class because it has negative self-intersection by the Riemann--Roch theorem.
Thus from the equations of the $3d^2$ obvious lines in $F_{\ov\Q}$ we infer that
$\Q(\mu_{2d})\subset L$. By Shioda \cite[Thm.~7]{Sh81}, $\Pic(F_{\ov\Q})\otimes_\Z\Q$ is generated by the classes of these lines if $d\leq 4$ or ${\rm gcd}(6,d)=1$.
{\rm (ii)} Let $S_{\rm ind}\subset S_\flat$ be the subset of quadruples
$(a_0,a_1,a_2,a_3)$ such that $a_i+a_j\neq d$ for all $i\neq j$.
Let $\Lambda'\subseteq\Pic(F_{\ov \Q})$ be the subgroup generated by the $3d^2$ lines. Then $$\Pic(F_{\ov \Q})\otimes_\Z E=(\Lambda'\otimes_\Z E)\oplus\bigoplus_{\chi\in S_{\rm ind}}V_\chi.$$
Let $S_{\rm reg}\subseteq S_{\rm ind}$ be the subset of quadruples which, possibly after a permutation, can be written in one of the following forms
\begin{align*}
(i,d/2+i,d-2i,d/2),\ &i\neq d/4,\\
(i,d/2+i,d/2+2i,d-4i),\ &i\neq d/3,d/4,d/6,\\
(i,d/3+i,2d/3+i,d-3i),\ &i\neq d/6.
\end{align*}
In \cite[Thm.~1--3]{ASh}, Aoki and Shioda find for each $\chi\in S_{\rm reg}$ an explicit equation describing a divisor whose class generates $V_{\chi}$. These divisors are defined over $\Q(\mu_{2d},2^{2/d})$ (for $d$ divisible by $2$) or $\Q(\mu_{2d},3^{3/d})$ (for $d$ divisible by $3$) and have negative self-intersection, hence they are unique effective divisors in their classes.
By \cite[Thm.~C]{Aok}, if $d>180$ and $\chi\in S_\flat$ is primitive, then $\chi\in S_{\rm reg}$.
From the calculations in \cite{Sh81} it then follows that
if $d$ is not divisible by any of the integers between $12$ and $180$ in the table of
\cite[p.~727]{Sh81},
%(or equivalently not divisible by any of the integers $4,5,7,9,11,13$),
then $S_{\rm ind}=S_{\rm reg}$. The claim (ii) follows. $\Box$
\bpr \label{D}
Write $w=|(O_E^\times)_{\rm tors}|$, so that $w=d$ if $d$ is even, and $w=2d$ if $d$ is odd. Let $\Delta$ be the subgroup of $E^\times/E^{\times w}$
generated by $O_E^\times$ and the elements $1-\e^{d/p}$
for all prime divisors $p|d$. Then $L$ is contained in $K=E(\Delta^{1/w})$.
\epr
{\em Proof.} By Proposition \ref{field} (ii), it remains to determine the image of the action of $\Gal(\ov\Q/E)$ on $V_\chi$ for all $\chi=(a_0,a_1,a_2,a_3)\in S_{\rm ind}\setminus S_{\rm reg}$.
It was first shown by Weil \cite[Theorem]{W2} that $h_\chi$ is a Größencharacter of conductor dividing $d^2$, hence dividing $c\coloneq 2w^2$. Since $\Gal(\ov\Q/E)$ acts on $V_\chi\subset\Pic(F_{\ov\Q})\otimes_\Z E$ through a finite abelian quotient, it follows by \cite[Ch.~VII, Prop.~6.9]{N} that $h_\chi$ factors through a generalised Dirichlet character $h^\mathrm{fin}_\chi$ modulo~$c$.
By class field theory, the action of $\Gal(\ov\Q/E)$ on $V_\chi$ thus factors through the Galois group $\Gal(E(c)/E)\simeq\Cl(c)$ of the ray class field $E(c)$ of conductor $c$. By Lemma~\ref{kumext} below, we have $K\subset E(c)$ and hence the Artin map $a_{K/E}:I(c)\to \Gal(K/E)$ factors through a homomorphism $a^\mathrm{fin}_{K/E}:\Cl(c)\to\Gal(K/E)$. It remains to show that
\begin{equation}\label{cft}
\Ker(a^\mathrm{fin}_{K/E})\subset \Ker(h^\mathrm{fin}_\chi)
\end{equation}
in the finite ray class group $\Cl(c)$. For any given $\chi$, the determination of $\Ker(h^\mathrm{fin}_\chi)$ is a finite calculation with Jacobi sums which can be performed as in \cite[\S4.1]{W} with customary computer algebra packages (see Remark~\ref{magma}); and so is the determination of $\Ker(a^\mathrm{fin}_{K/E})$ using the characterisation of the Artin map of a Kummer extension from \cite[\S2]{Fie}.
It suffices to verify (\ref{cft}) for each primitive $\chi\in S_{\rm ind}\setminus S_{\rm reg}$ and all $d$.
By the aforementioned \cite[Thm.~C]{Aok}, such $\chi$ may only exist for those $12\leq d\leq 180$ that are listed in the table of \cite[p.~727]{Sh81}. Checking this finite list of $\chi$ completes the proof of (iii). $\Box$
\ble\label{kumext}
We have $K\subset E(c)$.
\ele
{\em Proof.} It is enough to show that if
$x\in O_E$ divides $d\geq 3$, then the conductor of the Kummer extension
$E(\sqrt[w]{x})$ of $E$ divides $c$.
The relative discriminant $\Delta_{E(\sqrt[{w}]{x})/E}$ divides the relative discriminant ideal of the polynomial $t^{w}-x$, which is $({w}^{w}x^{{w}-1})$.
A formula in Hasse's lectures on class field theory \cite[Sätze (163), (164)]{Has} says that the conductor of $E(\sqrt[{w}]{x})$ divides
$$\prod_{\mathfrak p\nmid {w}}\mathfrak p\prod_{\mathfrak p|{w}}\mathfrak p^{\left(v_p({w})+\frac{1}{p-1}\right)e(\mathfrak p/p)+1},$$
where both products are taken over all primes $\mathfrak p\subset O_E$ dividing $\Delta_{E(\sqrt[{w}]{x})/E}$, and $e(\mathfrak p/p)$ is the ramification index of $\mathfrak p$ over the underlying prime number $p$.
The first product is vacuous. Now if $v_p(w)>1$ or $p>2$, then $e(\mathfrak p/p)/(p-1)+1\leq v_p(w)e(\mathfrak p/p)$. If $p=2$ and $v_2(w)=1$, then $\left(v_2({w})+\frac{1}{2-1}\right)e(\mathfrak p/2)+1 = 2v_2(w)e(\mathfrak p/2)+1$. Thus the conductor of $E(\sqrt[{w}]{x})$ divides
\[\prod_{\mathfrak p|{w}}\mathfrak p^{2v_p(w)e(\mathfrak p/p)}\prod_{\mathfrak p|2}\mathfrak p,\]
hence divides $c$. $\Box$
\brem\label{magma} {\rm
The calculation in Proposition \ref{D}
leads to an explicit description of the exact field of definition $L$ of the geometric
Picard group of the Fermat surface of degree $d$, for all positive integers $d$.
%AS very minor editing of the next sentence
This is achieved by computing,
for each $12\leq d\leq 180$ in the table of \cite[p.~727]{Sh81} and for each primitive
$\chi\in S_\flat$ in degree $d$,
the exact kernel of $h^\mathrm{fin}_\chi$ %exactly (instead of establishing a lower bound)
and the corresponding abelian extension $L_\chi/E$. Let $M_d$ be the compositum of the fields
$L_\chi$, for all primitive $\chi\in S_\flat$ in degree $d$.
Then $L$ is the compositum of the fields $E=\Q(\mu_d)$, $\Q(2^{2/d})$ if $2\mid d$, $\Q(3^{3/d})$ if $3\mid d$, and the fields $M_{d'}$ for all $d'|d$, $12\leq d'\leq 180$. The table of the fields $M_d$ and the associated Magma code \cite{Magma}, which also verifies Proposition~\ref{D}, are available at
\href{https://github.com/dgvirtz/fermatdeffield}{github.com/dgvirtz/fermatdeffield}.
}
\erem
\begin{thebibliography}{99}
\bibitem[Aok83]{Aok} N.~Aoki.
On some arithmetic problems related to the Hodge cycles on the Fermat varieties.
{\em Math. Ann.} {\bf 266} (1983) % no.~1,
23--54. Erratum: {\em Math. Ann.} {\bf 267} (1984) %, no.~4,
572.
\bibitem[ASh83]{ASh} N.~Aoki and T.~Shioda.
Generators of the N\'eron--Severi group of a Fermat surface.
{\em Arithmetic and geometry}, vol.~I, 1--12,
Progress in Mathematics {\bf 35}, Birkh\"auser, 1983.
\bibitem[BS07]{BS} S.~Boissière and A.~Sarti.
Counting lines on surfaces.
{\em Ann. Sc. Norm. Super. Pisa Cl. Sci.} {\bf 6} (2007) %, no.~1,
39--52.
\bibitem[BS08]{BS2} S.~Boissière and A.~Sarti.
On the Néron-Severi group of surfaces with many lines.
{\em Proc. Amer. Math. Soc.} {\bf 136} (2008) %, no.~11,
3861--3867.
\bibitem[Bri02]{Bri02} M. Bright. {\em Computations on diagonal quartic surfaces}.
PhD thesis, University of Cambridge, 2002.
\bibitem[Bri11]{Bri11} M. Bright. The Brauer--Manin obstruction on a general diagonal quartic surface.
{\em Acta Arith.} {\bf 147} (2011) 291--302.
\bibitem[BBL16]{BBL} M.J. Bright, T.D. Browning and D. Loughran. Failures of weak approximation in families.
{\em Compositio Math.} {\bf 152} (2016) 1435--1475.
\bibitem[CHM95]{CHM} L.~Caporaso, J.~Harris and B.~Mazur.
How many rational points can a curve have? {\em The moduli space of curves}
(Texel Island, 1994) 13--31,
Progress in Mathematics {\bf 129}, Birkhäuser, 1995.
\bibitem[CKS87]{CTKS} J.-L.~Colliot-Th\'el\`ene, D.~Kanevsky and J.-J.~Sansuc.
Arithm\'etique des surfaces cubiques diagonales.
{\em Diophantine approximation and transcendence theory} (1985) 1--108,
Lecture Notes in Mathematics {\bf 1290}, Springer-Verlag, 1987.
\bibitem[CTS21]{CTS21} J.-L.~Colliot-Th\'el\`ene and A.N.~Skorobogatov.
{\em The Brauer--Grothendieck group.} Ergebnisse der Mathematik und ihrer Grenzgebiete,
3. Folge, Band 71, Springer, 2021.
\bibitem[Deg15]{Deg15} A.~Degtyarev. Lines generate the Picard groups of
certain Fermat surfaces. {\em J. Number Theory} {\bf 147} (2015) 454--477.
\bibitem[Fie01]{Fie} C.~Fieker.
Computing class fields via the Artin map.
{\em Math. Comp.} {\bf 70} (2001) 1293--1303.
\bibitem[GSz17]{GSz} P.~Gille and T.~Szamuely.
{\em Central simple algebras and Galois cohomology}. 2nd ed. Cambridge Studies in Advanced Mathematics {\bf 165}, Cambridge University Press, 2017.
\bibitem[GS22]{GS22} D.~Gvirtz and A.N.~Skorobogatov. Cohomology and the Brauer groups of diagonal surfaces. {\em Duke Math. J.} {\bf 171} (2022) 1299--1347.
\bibitem[Has67]{Has} H.~Hasse.
{\em Vorlesungen \"uber Klassenk\"orpertheorie}.
Thesaurus Mathematicae {\bf 6}, Physica-Verlag, 1967.
\bibitem[HB02]{HB} D.R.~Heath-Brown.
The density of rational points on curves and surfaces.
{\em Ann. of Math.} {\bf 155} (2002) 553--595.
\bibitem[Ho67]{Ho1} C.~Hooley.
On binary cubic forms.
{\em J.~reine angew. Math.} {\bf 226} (1967) 30--87.
\bibitem[Ho86]{Ho2} C.~Hooley.
On binary quartic forms.
{\em J.~reine angew. Math.} {\bf 366} (1986) 32--52.
\bibitem[Ho00]{Ho3} C.~Hooley. On binary cubic forms, II.
{\em J.~reine angew. Math.} {\bf 521} (2000) 185--240.
\bibitem[Ino76]{Ino} H.~Inose.
On certain Kummer surfaces which can be realized as non-singular quartic surfaces in $\P^3$.
{\em J. Fac. Sci. Univ. Tokyo Sect. IA Math.} {\bf 23} (1976) %, no. 3,
545--560.
\bibitem[KSh79]{KSh} T.~Katsura and T. Shioda. On Fermat varieties.
{\em T\^{o}hoku Math. J.} {\bf 31} (1979) 97--115.
\bibitem[Magma]{Magma} W. Bosma, J. Cannon, and C. Playoust.
The Magma algebra system. I The user language.
{\em J. Symbolic Comput.} {\bf 24} (1997), 235--265.
\bibitem[Mil80]{M} J.S.~Milne. {\em \'Etale cohomology.} Princeton University Press, 1980.
\bibitem[Neu99]{N} J.~Neukirch. {\em Algebraic number theory.} Grundlehren der mathematischen Wissenschaften {\bf 322}, Springer-Verlag, 1999.
\bibitem[PT01]{PT} E.~Peyre and Yu.~Tschinkel. Tamagawa numbers of diagonal cubic surfaces, numerical evidence. {\em Math. Comp.} {\bf 70} (2001) 367--387.
\bibitem[San]{San} T.~Santens. Diagonal quartic surfaces with a Brauer--Manin obstruction.
{\em Compositio Math.}, to appear. arXiv:2201.04573
\bibitem[Sas68]{Sas} N.~Sasakura. On some results on the Picard numbers of certain algebraic surfaces.
{\em J. Math. Soc. Japan} {\bf 20} (1968) 297--321.
\bibitem[Sch82]{Schu} F.~Schur.
Ueber eine besondre Classe von Fl\"achen vierter Ordnung.
{\em Math. Ann.} {\bf 20} (1882), no.~2, 254--296.
\bibitem[Seg45]{Seg1} B.~Segre. The biaxial surfaces, and the equivalence of binary forms.
{\em Proc. Cambridge Philos. Soc.} {\bf 41} (1945) 187--209.
\bibitem[Seg46]{Seg2} B.~Segre. Equivalenza ed automorfismi delle forme binarie in un dato anello o campo numerico.
{\em Univ. Nac. Tucumán. Revista A.} {\bf 5} (1946) 7--68.
\bibitem[Seg47]{Seg3} B.~Segre. On arithmetical properties of quartic surfaces.
{\em Proc. London Math. Soc.} {\bf 49} (1947) 353--395.
\bibitem[Seg64]{Seg4} B.~Segre.
Arithmetische Eigenschaften von Galois-R\"aumen. I.
{\em Math. Ann.} {\bf 154} (1964) 195--256.
\bibitem[Ser94]{Ser} J.-P.~Serre.
{\em Cohomologie galoisienne}. 5th ed.
Lecture Notes in Mathematics {\bf 5}, Springer-Verlag, 1994.
\bibitem[SGA7]{SGA} P.~Deligne et N.~Katz. {\em Groupes de monodromie en g\'eom\'etrie alg\'ebrique,~II}. S\'em. g\'eom\'etrie alg\'ebrique du Bois-Marie 1967--69 (SGA 7 II).
Lecture Notes in Mathematics {\bf 340}, Springer-Verlag, 1973.
\bibitem[Sh79]{Sh79} T.~Shioda.
The Hodge conjecture for Fermat varieties.
{\em Math. Ann.} {\bf 245} (1979) 175--184.
\bibitem[Sh81]{Sh81} T.~Shioda.
On the Picard number of a Fermat surface.
{\em J. Fac. Sci. Univ. Tokyo Sect. IA Math.} {\bf 28} (1981) 725--734.
\bibitem[Sil92]{Sil} A.~Silverberg.
Fields of definition for homomorphisms of abelian varieties.
{\em J. Pure Appl. Algebra} {\bf 77} (1992) 253--262.
\bibitem[Uem14]{U} T.~Uematsu. On the Brauer group of diagonal cubic surfaces.
{\em Quat. J. Math.} {\bf 65} (2014) 677--701.
\bibitem[Uem16]{U2} T.~Uematsu.
On the Brauer group of affine diagonal quadrics. {\em J. Number Theory} {\bf 163} (2016) 146--158.
\bibitem[Wat18]{W} M.~Watkins.
Jacobi sums and Grössencharacters. {\em Publ. math. Besan\c{c}on. Alg\`ebre et th\'eorie des nombres.} (2018) 111--122.
\bibitem[Wei21]{Wei} D. Wei. Strong approximation for a toric variety. {\em Acta Math. Sin.} (Engl. Ser.) {\bf 37} (2021) 95--103.
\bibitem[Wei52]{W2} A.~Weil. Jacobi sums as ``Gr\"ossencharaktere''.
{\em Trans. Amer. Math. Soc.} {\bf 73} (1952) 487--495.
\bibitem[Wei94]{Weibel} C. Weibel. {\em An introduction to homological algebra}.
Cambridge studies in advanced mathematics {\bf 38}, Cambridge University Press, 1994.
\end{thebibliography}
\noindent Department of Mathematics, University of Glasgow, University Place,
Glasgow, G12~8QQ United Kingdom
\medskip
\noindent \texttt{damian.gvirtz@glasgow.ac.uk}
\bigskip
\noindent Department of Mathematics, South Kensington Campus,
Imperial College London, SW7 2BZ England, U.K. -- and --
Institute for the Information Transmission Problems,
Russian Academy of Sciences, 19 Bolshoi Karetnyi, Moscow, 127994
Russia
\medskip
\noindent \texttt{a.skorobogatov@imperial.ac.uk}
\end{document}