Specialization of N\'eron-Severi groups in positive characteristic

Extracted source text · arXiv 1810.06481

\documentclass[a4paper, 11pt]{amsart}
\usepackage{amsfonts, amsthm, amssymb, amsmath, stmaryrd}
\usepackage{mathrsfs,array}
\usepackage{eucal,color,times,enumerate,accents}
\usepackage{tikz-cd}
\usepackage{url}
\usepackage{bbm}
\usepackage[utf8]{inputenc}
\usetikzlibrary{decorations.text}
\usepackage{pstricks}
\usetikzlibrary{decorations.pathmorphing, patterns,shapes}
\usetikzlibrary{arrows,shapes,positioning}
\usetikzlibrary{decorations.markings}
\tikzstyle arrowstyle=[scale=1]
\tikzstyle directed=[postaction={decorate,decoration={markings,
    mark=at position .65 with {\arrow[arrowstyle]{stealth}}}}]
\tikzstyle reverse directed=[postaction={decorate,decoration={markings,
    mark=at position .65 with {\arrowreversed[arrowstyle]{stealth};}}}]
%\usepackage{xcolor}
\usepackage{pgfplots}
\usepackage{lipsum}
\usepackage{graphicx}
\usepackage{rotating}
\usepackage{enumitem}
\usepackage{enumerate}
\usepackage{textcomp}
\usepackage[pdftex,bookmarksnumbered,bookmarksopen]{hyperref}
\hypersetup{colorlinks, linkcolor=blue}
\usepackage[colorinlistoftodos]{todonotes}
\usepackage[left=4cm,right=4cm,top=3cm,bottom=3cm]{geometry}
\usepackage{mathtools}
%\usepackage[capitalize]{cleveref}
\input xy
\xyoption{all}
\usepackage{yfonts}
\usepackage{hyphenat}
\hyphenation{neu-tra-li-za-tion neu-tra-li-za-tions}
\setcounter{secnumdepth}{3}
\setcounter{tocdepth}{1}

\setlength{\hoffset}{-1.6cm}
\setlength{\voffset}{-0.5cm}
\addtolength{\textwidth}{3.2cm}
\addtolength{\textheight}{1.5cm}




\newcommand{\PP}{\mathbb{P}}
%\P doesn't work

%Rings and N
\newcommand{\C}{\mathbb{C}}
\newcommand{\pri}{\mathbb{P}}

\newcommand{\F}{\mathbb{F}}
\newcommand{\Ga}{\mathbb{G}_a}
\newcommand{\LL}{\mathbb{L}}
\newcommand{\KK}{\mathbb{K}}

\newcommand{\Q}{\mathbb{Q}}
\newcommand{\Z}{\mathbb{Z}}
\newcommand{\N}{\mathbb{N}}
\newcommand{\FCrys}{{\mathbf{F\textrm{-}Crys}}}
\newcommand{\Crys}{{\mathbf{Crys}}}

\newcommand{\LS}{\mathbf{LS}}
\newcommand{\Coef}{\mathbf{Coef}}
\newcommand{\Foi}{\Fisoc^\textrm{\dag}}
\newcommand{\oi}{\Isoc^\textrm{\dag}}
\newcommand{\pioi}{\pi_1^{\mathrm{Isoc}^\dag}}
\newcommand{\pibo}{\pi_1^{?}}

\newcommand{\iso}{\xrightarrow{\sim}}
\newcommand{\hZ}{\widehat{\Z}}
\newcommand{\Za}{\Z^{\mathrm{alg}}}
\newcommand{\Zl}{\mathbb{Z}_{\ell}}
\newcommand{\Ql}{\mathbb{Q}_{\ell}}
\newcommand{\Kbar}{\overline{K}}
\newcommand{\Qp}{\mathbb{Q}_p}
\newcommand{\Qlbar}{\overline{\mathbb{Q}_{\ell}}}
\newcommand{\Qpbar}{\overline{\mathbb{Q}_{p}}}
\newcommand{\Qlp}{\mathbb{Q}_{\ell'}}
\newcommand{\Qlpbar}{\overline{\mathbb{Q}_{\ell'}}}
\newcommand{\Qbar}{\overline{\mathbb{Q}}}
\newcommand{\kbar}{\overline{\mathbb{k}}}
\newcommand{\Ff}{\mathbb{F}}
\newcommand{\Gm}{\mathbb{G}_m}
\newcommand{\dcirc}{\circ\circ}
\newcommand{\et}{\textrm{ét}}
\newenvironment{forcedcentertikzcd}
 {\begin{lrbox}{\forcedcentertikzcdbox}\begin{tikzcd}}
 {\end{tikzcd}\end{lrbox}\makebox[0pt]{\usebox{\forcedcentertikzcdbox}}}
\newsavebox{\forcedcentertikzcdbox}
%mathcals
\newcommand{\calA}{\mathcal{A}}
\newcommand{\calB}{\mathcal{B}}
\newcommand{\calC}{\mathcal{C}}
\newcommand{\calE}{\mathcal{E}}
\newcommand{\calF}{\mathcal{F}}
\newcommand{\calG}{\mathcal{G}}
\newcommand{\calH}{\mathcal{H}}
\newcommand{\calO}{\mathcal{O}}
\newcommand{\calP}{\mathcal{P}}
\newcommand{\calM}{\mathcal{M}}
\newcommand{\calN}{\mathcal{N}}
\newcommand{\calX}{\mathcal{X}}
\newcommand{\calY}{\mathcal{Y}}
\newcommand{\calU}{\mathcal{U}}

\newcommand{\calV}{\mathcal{V}}
\newcommand{\ucalV}{\underline{\mathcal{V}}}
\newcommand{\ucalVz}{\underline{\mathcal{V}_0}}
\newcommand{\calW}{\mathcal{W}}
\newcommand{\calS}{\mathcal{S}}
\newcommand{\calK}{\mathcal{K}}
\newcommand{\calZ}{\mathcal{Z}}


%categories
\newcommand{\Weil}{\mathbf{Weil}}
\newcommand{\VVec}{\mathbf{Vec}}



%mathrm
\newcommand{\Spec}{\mathrm{Spec}}
\newcommand{\Gal}{\mathrm{Gal}}
\newcommand{\Hom}{\mathrm{Hom}}
\newcommand{\HHom}{\mathcal{H}\mathrm{om}}
\newcommand{\Ker}{\mathrm{Ker}}
\newcommand{\Isom}{\mathrm{Isom}}
\newcommand{\End}{\mathrm{End}}
\newcommand{\Aut}{\mathrm{Aut}}
\newcommand{\Mod}{\mathrm{Mod}}
\newcommand{\GL}{\mathrm{GL}}
\newcommand{\id}{\mathrm{id}}
\newcommand{\Image}{\mathrm{Im}}
\newcommand{\Char}{\mathrm{char}}
\newcommand{\ord}{\mathrm{ord}}
\newcommand{\Tr}{\mathrm{Tr}}
\newcommand{\Sym}{\mathrm{Sym}}
\newcommand{\DGal}{\mathrm{DGal}}
\newcommand{\Rep}{\mathbf{Rep}}
\newcommand{\Fisoc}{\mathbf{F\textrm{-}Isoc}}
\newcommand{\Isoc}{\mathbf{Isoc}}
\newcommand{\Strat}{\mathbf{Strat}}

\newcommand{\OKbar}{O_{\overline{\mathbb{K}}}}
\newcommand{\etale}{\'etale }
\newcommand{\neron}{N\'eron }
%others
\newcommand{\hX}{\widehat{X}}
\newcommand{\Xbar}{X_{\overline k}}



\begin{document}
\bibliographystyle{alpha}
\newtheorem{theorem}{Theorem}[subsubsection]
%\numberwithin{theorem}{subsection}
\newtheorem{hope}[theorem]{Hope}
\newtheorem{proposition}[theorem]{Proposition}
\newtheorem{lemma}[theorem]{Lemma}
\newtheorem{claim}[theorem]{Claim}
\newtheorem{corollary}[theorem]{Corollary}
\newtheorem{fact}[theorem]{Fact}
\theoremstyle{definition}
\newtheorem{definition}[theorem]{Definition}
\newtheorem{question}[theorem]{Question}
\newtheorem{conjecture}[theorem]{Conjecture}
\newtheorem{answer}[theorem]{Answer}
\newtheorem{remark}[theorem]{Remark}
\newtheorem{example}[theorem]{Example}
\newtheorem{warning}[theorem]{Warning}
\newtheorem{notation}[theorem]{Notation}
\newtheorem{construction}[theorem]{Construction}
\newtheorem*{corollarynoname}{Corollary}

\numberwithin{equation}{theorem}

\title{Specialization of N\'eron-Severi groups in positive characteristic}
\author{Emiliano Ambrosi}

\begin{abstract}
Let $k$ be an infinite finitely generated field of characteristic $p>0$. Fix a reduced scheme $X$, smooth, geometrically connected, separated and of finite type over $k$ and a smooth proper morphism $f:Y\rightarrow X$. The main result of this paper is that there are ``lots" of closed points $x\in X$ such that the fibre of $f$ at $x$ has the same geometric Picard rank as the generic fibre. If $X$ is a curve we show that this is true for all but finitely many $k$-rational points. In characteristic zero, these results have been proved by Andr\'e (existence) and Cadoret-Tamagawa (finiteness) using Hodge theoretic methods. To extend the argument in positive characteristic we use the variational Tate conjecture in crystalline cohomology, the comparison between different $p$-adic cohomology theories and independence techniques. The result has applications to the Tate conjecture for divisors, uniform boundedness of Brauer groups, proper families of projective varieties and to the study of families of hyperplane sections of smooth projective varieties. 
\end{abstract}
\maketitle

\tableofcontents
\newpage
\section{Introduction}
\numberwithin{theorem}{subsection} 
\subsection{Conventions}
For a field $k$, a $k$-variety is a reduced scheme separated and of finite type over $k$. For a $k$-variety $X$, write $|X|$ for the set of closed points and if $x\in X$, write $k(x)$ for its residue field. If $Y\rightarrow X$ is a morphism and $x\in X$ write $i_{x}:Y_x\rightarrow Y$ for the natural inclusion of the fibre $Y_x$ at $x$ in $Y$. We use $\twoheadrightarrow$ and $\hookrightarrow$ to denote surjective and injective maps respectively. If $\F_q$ is a finite field, write $\F$ for its algebraic closure. If $\mathcal C$ is an abelian category write $\mathcal C\otimes \Q$ for its isogeny category and $\otimes \Q:\mathcal C\rightarrow \mathcal C\otimes \Q$ for the canonical functor.
\numberwithin{theorem}{subsection} 
\subsection{Galois generic points}\label{introduction}
Let $k$ be a field of characteristic $p>0$ with algebraic closure $\overline k$, $X$ a smooth and geometrically connected $k$-variety with generic point $\eta$ and $f:Y\rightarrow X$ a smooth proper morphism of $k$-varieties. For $x\in X$, fix a geometric point $\overline x$ over $x$ and an \etale path from $\overline x$ to $\overline \eta$. For every $\ell\neq p$, by smooth proper base change $R^2f_*\Ql(1)$ is a lisse sheaf on $X$ and the choice of the \etale path gives equivariant isomorphisms

\begin{center}

\begin{tikzcd}[
 row sep=small,
  ar symbol/.style = {draw=none,"\textstyle#1" description,sloped},
  isomorphic/.style = {ar symbol={\cong}},
  ]
    H^2(Y_{\overline \eta},\Ql(1)) \arrow[loop below]\arrow[phantom]{r}{\simeq}& R^2f_*\Ql(1)_{\overline \eta} \arrow[phantom]{r}{\simeq} & R^2f_*\Ql(1)_{\overline x}\arrow[phantom]{r}{\simeq} &  H^2(Y_{\overline x},\Ql(1)) \arrow[loop below]\\
    \pi_1(X,\overline \eta)\arrow[phantom]{r}{\simeq} &  \pi_1(X,\overline x)& & \pi_1(x,\overline x)\arrow{ll}.
\end{tikzcd}
\end{center}
\begin{definition}
A point $x\in X$ is $\ell$-Galois generic (resp. strictly $\ell$-Galois generic) for $f:Y\rightarrow X$ if the image of $\pi_1(x,\overline x)\rightarrow \pi_1(X,\overline \eta)\rightarrow GL(H^2(Y_{\overline \eta},\Ql(1)))$ is open (resp. coincides with) in the image of $\pi_1(X,\overline \eta)\rightarrow GL(H^2(Y_{\overline \eta},\Ql(1)))$.
\end{definition}
By \cite[Theorem 1.1]{uniformadelic}, $x$ is $\ell$-Galois generic for one $\ell\neq p$ if and only if $x$ is $\ell$-Galois generic for every $\ell\neq p$. So one simply says that $x$ is Galois generic for $f$. This is not true for strictly Galois generic points, and one says that $x$ is strictly Galois generic if there exists an $\ell\neq p$ such that $x$ is strictly $\ell$-Galois generic.
\numberwithin{theorem}{subsection}
 \subsection{Neron-severi generic points}\label{nerongeneric}
 \numberwithin{theorem}{subsubsection}

 \subsubsection{Tate conjecture for divisors}
The geometric N\'eron-Severi group $NS(Z_{\overline k})$ of a smooth proper $k$-variety $Z$ is a finitely generated abelian group such that $NS(Z_{\overline k})\otimes \Q$ identifies with the image of the cycle class map for $\ell$-adic cohomology
$$c_{Z_{\overline k}}:Pic(Z_{\overline k})\otimes \Q\rightarrow H^2(Z_{\overline k},\Ql(1)).$$
Since $NS(Z_{\overline k})$ is a finitely generated abelian group, $\pi_1(k)$ acts on it trough a finite quotient and hence $NS(Z_{\overline k})\subseteq H^2(Z_{\overline k},\Ql(1))$ is fixed under the action of the connected component $G^0_{\ell}$ of the Zariski closure of the image $G_{\ell}$ of $\pi_1(k)$ acting on $H^2(Z_{\overline k},\Ql(1))$. Recall that the $\ell$-adic Tate conjecture for divisors (\cite{tateconj}) predicts the following:
\begin{conjecture}[$T(Z,\ell)$]\label{tateconjectureconj}
Let $k$ be a finitely generated field and $Z$ a smooth proper $k$-variety. Then the map
$c_{Z_{\overline k}}:NS(Z_{\overline k})\otimes \Ql\rightarrow H^2(Z_{\overline k},\Ql(1))^{G^0_{\ell}}$
is an isomorphism
\end{conjecture}
\subsubsection{Specialization morphisms}
Retain the notation and the assumptions of Section \ref{introduction}.
For every $x\in X$, there is an injective specialization homomorphism (see e.g. \cite[Proposition 3.6.]{poonen})
$$sp_{\eta,x}:NS(Y_{\overline \eta})\otimes \Q\rightarrow NS(Y_{\overline x})\otimes \Q$$
compatible with the cycle class map, in the sense that the following diagram commutes:
\begin{center}
\begin{tikzcd}
Pic(Y_{\eta})\otimes \Q\arrow{d}{c_{Y_{\eta}}} & Pic(Y)\otimes \Q\arrow[swap]{l}{i_{\eta}^*}\arrow{r}{i_{x}^*} &  Pic(Y_{x})\otimes \Q \arrow{d}{c_{Y_x}}\\
NS(Y_{\overline{\eta}})\otimes \Q\arrow[hook]{rr}{sp_{\eta,x}}\arrow[hook]{d} && NS(Y_{\overline x})\otimes \Q\arrow[hook]{d}\\
 H^2(Y_{\overline{\eta}},\Ql(1))\arrow[phantom]{rr}{\simeq} && H^2(Y_{\overline{x}},\Ql(1))
\end{tikzcd}
\end{center}
Since the \neron-Severi group is invariant under extensions of algebraically closed fields (see e.g. \cite[Proposition 3.1]{poonen}), the map $sp_{\eta,x}$ is well defined, independently of the choice of the geometric points $\overline \eta$ over $\eta$ and $\overline x $ over $x$. 

The abelian group $NS(Y_{\overline{\eta}})\otimes \Q$ is a $\pi_1(X,\overline \eta)$-module and hence the group $\pi_1(x,\overline x)$ acts on $NS(Y_{\overline{\eta}})\otimes \Q$ by restriction trough the morphism $\pi_1(x,\overline x)\rightarrow \pi_1(X,\overline x)\simeq \pi_1(X,\overline \eta)$. Since the map $sp_{\eta,x}$ is $\pi_1(x,\overline x)$-equivariant with respect to the natural action of $\pi_1(x,\overline x)$ on $NS(Y_{\overline x})\otimes \Q$, one constructs an injective specialization map
$$sp^{ar}_{\eta,x}:NS(Y_{\eta})\otimes \Q\subseteq (NS(Y_{\overline \eta})\otimes \Q)^{\pi_1(x,\overline x)}\xrightarrow{sp_{\eta,x}} NS(Y_{x})\otimes \Q,$$
where for a smooth proper $k$-variety $Z$ one writes $NS(Z)\otimes \Q:=(NS(Z_{\overline k})\otimes \Q)^{\pi_1(k)}$.
\begin{definition}
One says that $x$ is NS-generic (resp. strictly NS-generic) for $f:Y\rightarrow X$ if $sp_{\eta,x}$ (resp. $sp^{ar}_{\eta,x}$) is an isomorphism.
\end{definition}
Conjecture \ref{tateconjectureconj} predicts that every (strictly) Galois generic point is (strictly) NS-generic.
Our main result is that this holds (without assuming Conjecture \ref{tateconjectureconj}), at least when $f:Y\rightarrow X$ is projective.
\begin{theorem}\label{main}
Let $k$ be a finitely generated field and $f:Y\rightarrow X$ a smooth projective morphism. If $x\in X$ is Galois-generic (resp. strictly Galois generic) for $f:Y\rightarrow X$ then it is $NS$-generic (resp. strictly NS-generic) for $f:Y\rightarrow X$. If $f:Y\rightarrow X$ is smooth and proper, the same is true for all $x$ in a dense open subset of $X$.
\end{theorem}
\numberwithin{theorem}{subsection} 
\subsection{Proof in characteristic zero}\label{char0}
When $k$ is a field of characteristic zero Theorem \ref{main} is due to Andr\'e (\cite{AndreIHES}; see also \cite[Corollary 5.4]{MC} and \cite[Proposition 3.2.1]{brauer}) and it holds for $f:Y\rightarrow X$ proper. Since it is the starting point for our proof we briefly recall the argument when $k\subseteq \mathbb C$ and $x$ is a closed point. 
Fix a smooth compactification $Y\subseteq \overline Y$ of $Y$. The commutative diagram of $k$-varieties
\begin{center}
\begin{tikzcd}
Y_{x}\arrow{r}\arrow{d}\arrow[phantom]{dr}{\Box} & Y\arrow[hook]{r}\arrow{d}& \overline Y\\
k(x)\arrow{r}{x} & X
\end{tikzcd}
\end{center}
induces a commutative diagram:
\begin{center}
\begin{tikzcd}
H^0(X_{\mathbb C},R^2f_*\Q(1))\arrow{d}{\simeq} & H^{2}(Y_{\mathbb C},\Q(1))\arrow{l}{Leray} & NS(Y_{\overline \eta})\otimes \Q\arrow[bend left=100]{dd}{sp_{\eta,x}}\\
H^{2}(Y_{\overline x},\Q(1))^{\pi^{top}_1(X_{\mathbb C})}\arrow[hook]{d} & H^2(\overline Y_{\C}, \Q(1))\arrow{l}\arrow{u} & NS(\overline Y_{\C})\otimes \Q\arrow{d}\arrow{u}\arrow[hook]{l}\\
H^{2}(Y_{\overline x},\Q(1))& & NS(Y_{\overline x})\otimes \Q\arrow[hook]{ll}{c_{Y_{\overline x}}}
\end{tikzcd}
\end{center}
where $Leray$ is the edge map in the Leray spectral sequence attached to $f:Y\rightarrow X$. 
Take any $z_x\in NS(Y_{\overline x})\otimes \Q$. Since $z$ is fixed by an open subgroup of $\pi_1(x)$ and $x$ is Galois-generic, up to replacing $X$ with a finite \etale cover one can assume that $z_x$ is fixed by $\pi_1(X_{\C})$. By the comparison between the \etale and the singular sites, $z_x$ is fixed by $\pi^{top}_1(X_{\C})$. 
By Deligne's fixed part theorem (\cite[Theoreme 4.1.1]{HodgeII}) the map
$$H^2(\overline Y_{\C}, \Q(1))\rightarrow H^{2}(Y_{\overline x},\Q(1))^{\pi^{top}_1(X_{\mathbb C})}$$
is surjective. By semisemplicity, the map $H^2(\overline Y_{\C}, \Q(1))\rightarrow H^{2}(Y_{\overline x},\Q(1))^{\pi^{top}_1(X_{\mathbb C})}$ splits in the category of polarized $\Q$-Hodge structures, so that $z_x$ is the image of a $z\in H^{0,0}(\overline Y_{\C},\Q(1))$. By the Lefschetz (1,1) theorem, $z$ lies in $NS(\overline Y_{\mathbb C})\otimes \Q$. One concludes the proof observing that, by construction, the restriction $z_{\eta}$ of $z$ to $NS(Y_{\overline \eta})\otimes \Q$ is such that $sp_{\eta,x}(z_{\eta})=z_x$.
\numberwithin{theorem}{subsection} 
\subsection{Strategy in positive characteristic}
In characteristic zero the main ingredients are the combination of Deligne's fixed part theorem and the Lefschetz-(1,1) theorem (what is called the variational Hodge conjecture for divisors; see e.g. \cite[Conjecture 9.6, Remark 9.7]{poonen}) and the comparison between the \etale and the singular sites. To try and make the argument of Section \ref{char0} works in positive characteristic the idea is to replace Betti cohomology with crystalline cohomology. The main reason for this is that the variational Tate conjecture for projective morphisms (Fact \ref{thmvariationalTate}), that we now recall, is known in this setting.
\numberwithin{theorem}{subsubsection}
\subsubsection{Crystalline variational Tate conjecture}
Let $\F_q$ be the finite field with $q=p^s$ elements, $\mathcal X$ a connected smooth $\F_q$-variety and $\mathfrak f:\mathcal Y\rightarrow \mathcal X$ a smooth proper morphism of $\F_q$-varieties (in our application $\mathfrak f:\mathcal Y\rightarrow \mathcal X$ is a model for $f:Y\rightarrow X$). Write respectively $Mod(\mathcal X|W)$, $Mod(\mathcal Y|W)$ for the categories of $\mathcal O_{\mathcal X|W}$, $\mathcal O_{\mathcal Y|W}$ modules in the crystalline site of $\mathcal X$, $\mathcal Y$ over $W:=W(\F_q)$ (\cite[Section 2]{Notmor}). Then there is a higher direct image functor  
$$R^{i}\mathfrak f_{crys,*}: Mod(\mathcal Y|W)\rightarrow Mod(\mathcal X|W)$$
and, for every $\mathfrak t\in \mathcal X(\F_q)$, a commutative diagram 
\begin{center}
\begin{tikzcd}
H^2_{crys}(\mathcal Y)\arrow{d}{Leray}\arrow{rd}{i^*_{\mathfrak t}}& Pic(\mathcal Y)\otimes \Q \arrow[swap]{l}{c_{\mathcal Y}}\arrow{rd}{i^*_{\mathfrak t}}\\
H^0(\mathcal X,R^2\mathfrak f_{crys, *}\mathcal O_{\mathcal Y/W})\otimes \Q\arrow{r} & H^{2}_{crys}(\mathcal Y_{\mathfrak t}) & Pic(\mathcal Y_{\mathfrak t})\otimes \Q\arrow{l}{c_{\mathcal Y_{\mathfrak t}}}
\end{tikzcd}
\end{center}
where $H^2_{crys}(\mathcal Y)$ and $H^{2}_{crys}(\mathcal Y_{\mathfrak t})$ are the (rational) crystalline cohomology of $\mathcal Y$ and $\mathcal Y_{\mathfrak t}$ respectively, Leray is the edge map in the Leray spectral sequence attached to $\mathfrak f:\mathcal Y\rightarrow \mathcal X$ and $c_{\mathcal Y},c_{\mathcal Y_{\mathfrak t}}$ are the crystalline cycle class maps. Write $F$ for the $s$-power of the absolute Frobenius of $\mathcal X$ and recall that the images of $c_{\mathcal Y}$ and $c_{\mathcal Y_{\mathfrak t}}$ lie in $H^2_{crys}(\mathcal Y)^{F=q}$ and $H^2_{crys}(\mathcal Y_{\mathfrak t})^{F=q}$, respectively. Then we have the variational Tate conjecture in crystalline cohomology:
\begin{fact}\label{thmvariationalTate}\cite[Theorem 1.4]{Varmor}
If $\mathfrak f:\mathcal Y\rightarrow \mathcal X$ is projective, for every $z_{\mathfrak t}\in Pic(\mathcal Y_{\mathfrak t})\otimes \Q$ the following are equivalent:
\begin{enumerate}
\item There exists $z\in Pic(\mathcal Y)\otimes \Q$ such that $c_{\mathcal Y_{\mathfrak t}}(z_{\mathfrak t})=i^*_{\mathfrak t}(c_{\mathcal Y}(z))$;
\item $c_{\mathcal Y_{\mathfrak t}}(z_{\mathfrak t})$ lies in $H^0(\mathcal X,R^{2}\mathfrak f_{crys, *}\mathcal O_{\mathcal Y/W})\otimes \Q$;
\item $c_{\mathcal Y_{\mathfrak t}}(z_{\mathfrak t})$ lies in $H^0(\mathcal X,R^{2}\mathfrak f_{crys, *}\mathcal O_{\mathcal Y/W})^{F=q}\otimes \Q$.
\end{enumerate}
\end{fact}
However, to apply Fact \ref{thmvariationalTate} in our setting, there are two difficulties to overcome:
\begin{enumerate}
\item Crystalline cohomology works well only over a perfect field, while our base field $k$ is not perfect;
\item There is no direct way to compare the $\ell$-adic and the crystalline sites, so that one has to find a different way to transfer the Galois generic assumption to the crystalline setting. 
\end{enumerate}
\numberwithin{theorem}{subsubsection}
\subsubsection{Spreading out}\label{introspreading}
To overcome (1) one uses a spreading out argument, so that our morphism $f:Y\rightarrow X$ will appear as the generic fibre of a smooth projective morphism $\mathfrak f:\mathcal Y\rightarrow \mathcal X$, where $\mathcal X$ is a smooth geometrically connected $\F_q$-variety. The idea is then to lift an element $\epsilon_x\in NS(Y_{\overline x})\otimes \Q$ to $NS(Y_{\overline \eta})\otimes \Q$ by specializing it first to an element $\epsilon_{\mathfrak t}\in NS(\mathcal Y_{\overline{\mathfrak t}})\otimes \Q $ of a closed fibre of $\mathcal Y\rightarrow \mathcal X$ and then to try and lift $\epsilon_{\mathfrak t}$ to an element $\epsilon\in Pic(\mathcal Y)\otimes \Q$, via the crystalline variational Tate conjecture over $\F_q$.
\numberwithin{theorem}{subsubsection}
\subsubsection{From \texorpdfstring{$\ell$}- to p}\label{seccrysttol}
In order to show that $\epsilon_{\mathfrak t}\in NS(Y_{\overline{\mathfrak t}})\otimes \Q $ satisfies the assumption of Fact \ref{thmvariationalTate}, one has to transfer the $\ell$-adic information that $x$ is Galois generic to crystalline cohomology. For this the key ingredient is Theorem \ref{ltocrys} below.
Assume that $\mathcal Z$ is a smooth geometrically connected  $\F_q$-variety admitting an $\F_q$-rational point $\mathfrak t$ and that there is a map $\mathfrak g:\mathcal Z\rightarrow \mathcal X$ (in our application $\mathfrak g:\mathcal Z\rightarrow \mathcal X$ is a model for $x:k(x)\rightarrow X$). The cartesian square
\begin{center}
\begin{tikzcd}
\mathcal Y_{\mathcal Z}\arrow{r}\arrow{d}{\mathfrak f_{\mathcal Z}}\arrow[phantom]{rd}{\Box} & \mathcal Y\arrow{d}{\mathfrak f}\\
\mathcal Z\arrow{r}{\mathfrak g} & \mathcal X
\end{tikzcd}
\end{center}
induces representations
$$\pi_1(\mathcal Z,\overline{\mathfrak t})\rightarrow \pi_1(\mathcal X,\overline{\mathfrak t})\rightarrow GL(H^{i}(\mathcal Y_{\overline{\mathfrak t}},\Ql(j)).$$
\begin{theorem}\label{ltocrys}
Assume that the image of  $\pi_1(\mathcal Z,\overline{\mathfrak t})\rightarrow \pi_1(\mathcal X,\overline{\mathfrak t})\rightarrow GL(H^{i}(\mathcal Y_{\overline{\mathfrak t}},\Ql(j))$ is open in the image of $\pi_1(\mathcal X,\overline{\mathfrak t})\rightarrow GL(H^{i}(\mathcal Y_{\overline{\mathfrak t}},\Ql(j))$ and that the Zariski closures of the images of $\pi_1(\mathcal X,\overline{\mathfrak t})$ and $\pi_1(\mathcal X_{\mathbb F},\overline{\mathfrak t})$ acting on $H^{i}(\mathcal Y_{\overline{\mathfrak t}},\Ql(j))$ are connected.
Then the base change map
$$H^{0}(\mathcal X,R^{i}\mathfrak f_{crys,*}\mathcal O_{\mathcal Y/W})^{F=q^j}\otimes \Q\rightarrow H^{0}(\mathcal Z,R^{i}\mathfrak f_{\mathcal Z, crys,*}\mathcal O_{\mathcal Y_{\mathcal Z}/W})^{F=q^j}\otimes \Q$$
is an isomorphism.
\end{theorem}
The subtle point in the proof of Theorem \ref{ltocrys} is to compare the category of $F$-isocrystals, where the crystalline variational Tate conjecture is stated, with the category of $\ell$-adic sheaves. These categories behaves differently. For example, if $\mathfrak f:\mathcal Y\rightarrow \mathcal X$ is a non-isotrivial family of ordinary elliptic curves, $R^1\mathfrak f_{crys,*} \mathcal O_{\mathcal Y/W}\otimes \Q$ carries a two steps filtrations, reflecting the decomposition of the $p$-divisible groups of the generic fibre of $\mathfrak f:\mathcal Y\rightarrow \mathcal X$ into \etale and connected parts, while $R^1\mathfrak f_{*} \Ql$ is irreducible. This leads to consider the smaller category of F-overconvergent isocrystals, whose behaviour is closer to the one of $\ell$-adic sheaves. Then the proof of Theorem \ref{ltocrys} decomposes as follows:
\begin{enumerate}
\item $R^{i}\mathfrak f_{crys,*}\mathcal O_{\mathcal Y/W}\otimes \Q$ and $R^{i}\mathfrak f_{\mathcal Z, crys,*}\mathcal O_{\mathcal Y_{\mathcal Z}/W}\otimes \Q$ are $F$-overconvergent isocrystals (Fact \ref{Shiho}, which uses a technical result of Shiho proved in the Part 3) and by Fact \ref{kedlaya} one doesn't loose information passing from crystalline cohomology to $F$-overconvergent isocrystals;
\item Let $G_p$ and $G_{\mathcal Z,p}$ be the Tannakian groups of $R^{i}\mathfrak f_{crys,*}\mathcal O_{\mathcal Y/W}\otimes \Q$ and $R^{i}\mathfrak f_{\mathcal Z, crys,*}\mathcal O_{\mathcal Y_{\mathcal Z}/W}\otimes \Q$ as F-overconvergent isocrystals. Theorem \ref{ltocrys} amounts to showing that $G_p=G_{\mathcal Z,p}$. 
\item The assumption implies that the Zariski closures $G_{\ell}$ and $G_{\mathcal Z,\ell}$ of the image of $\pi_1(\mathcal X,\overline{\mathfrak t})$ and $\pi_1(\mathcal Z,\overline{\mathfrak t})$ acting on $H^{i}(\mathcal Y_{\overline{\mathfrak t}},\Ql(j))$ are equal.
\item To show that $(3)$ implies $(2)$, one uses the theory of Frobenius weights and that semisimple algebraic groups are characterized by their tensor invariants. 
\end{enumerate}

\begin{remark}
In Theorem \ref{ltocrys}, the assumptions that $\mathcal Z$ has a $\F_q$ rational point and that the Zariski closure of the image of $\pi_1(\mathcal X_{\mathbb F},\overline{\mathfrak t})$ is connected are not necessary, but the proof without these assumptions requires the more elaborated formalism of $\overline \Q_p$-overconvergent isocrystals. In our application to Theorem \ref{main} one can reduce to the case where these assumptions are satisfied, so that we did not include the proof of the general form of Theorem \ref{ltocrys}.
\end{remark}
\begin{remark}
In characteristic 0, the proof sketched in \ref{char0} shows that the variational Hodge conjecture implies the $\ell$-adic variational Tate conjecture over fields of characteristic zero. 
In positive characteristic, our method does not show that the crystalline variational Tate conjecture implies the $\ell$-adic one. The issue comes from the fact that one does not know how to compare the $\ell$-adic and the crystalline cycle class maps.
\end{remark}
\numberwithin{theorem}{subsection} 
\subsection{Applications}
\numberwithin{theorem}{subsubsection}
\subsubsection{Existence of NS-generic points}
Let $k$ be a field of transcendence degree $\geq 1$ over $\F_p$ and $X$ a smooth geometrically connected $k$-variety with generic point $\eta$. Let $f:Y\rightarrow X$ be a smooth proper morphism of $k$-varieties. Recall the following:
\begin{fact}\label{sisagiaspec}
Assume that $k$ is finitely generated. Then:
\begin{itemize}
\item  The subset of non strictly $\ell$-Galois-generic points for $f:Y\rightarrow X$ is sparse. In particular there exists an integer  $d\geq 1$ such that there are infinitely many $x\in |X|$ with $[k(x):k]\leq d$ that are strictly $\ell$-Galois-generic for $f:Y\rightarrow X$. (\cite[Section 10.6]{Serreweil}). 
\item If $X$ is a curve, all but finitely many $x\in X(k)$ are Galois-generic for $f:Y\rightarrow X$. (\cite{UOIpositive}).
\end{itemize}
\end{fact}
Theorem \ref{main}, together with Fact \ref{sisagiaspec} and the fact that if $S\subseteq |X|$ is a subset and $U\subseteq X$ is a dense open subscheme such that $U\cap S\subseteq U$ is sparse then  $S\subseteq |X|$ is again sparse (\cite[Proposition 8.5 (a)]{poonen}), implies:
\begin{corollary}\label{existence}
Assume that $k$ is finitely generated. Then:
\begin{itemize}
\item The subset of closed non strictly NS-generic points for $f:Y\rightarrow X$ is sparse. In particular there exists an integer $d\geq 1$ such that there are infinitely many $x\in |X|$ with $[k(x):k]\leq d$ that are strictly NS-generic for $f:Y\rightarrow X$.
\item  If $X$ is a curve, all but finitely many $x\in X(k)$ are NS-generic for $f:Y\rightarrow X$.
\end{itemize}
\end{corollary}
Via a spreading out argument one has the following extension of the main result of \cite{poonen} to positive characteristic:
\begin{corollary}\label{miciomicio miao miao}
If $k$ is a field of transcendence degree $\geq 1$ over $\F_p$, then $X$ has a closed NS-generic point.
\end{corollary}
\begin{remark}
Atticus Christensen (\cite[Theorem 1.0.1]{Chr}) has independently proved Corollary \ref{miciomicio miao miao}. His proof is very different from ours, since his approach is inspired from the analytic approach in \cite{poonen}, while ours is inspired from the Hodge theoretic approach in \cite{AndreIHES}. On the other hand, it seems that Corollary \ref{existence} (that will be used to prove Corollaries \ref{exinstencehyperplane},\ref{corollaryTate} and \ref{corollarybrauer}) can not be obtained via his method, that gives different information on the set of NS generic points (\cite[Theorems 1.0.3, 1.0.4.]{Chr}).
\end{remark}
From Corollary \ref{miciomicio miao miao} one easily deduces the following results on the behaviour of the Tate conjecture in families:
\begin{corollary}\label{Tate family}
If $T(Y_x,\ell)$ holds for all $x\in |X|$, then $T(Y_{\eta},\ell)$ holds.
\end{corollary}
\begin{remark}
Corollary \ref{Tate family} together with a spreading out argument can be used to reduce the Tate conjecture for smooth proper varieties over arbitrary finitely generated fields of characteristic $p$, to fields of transcendence degree one over $\F_p$, extending results from \cite{Mori}, specific to abelian schemes, to arbitrary families of varieties. 
\end{remark}
The argument in \cite[Theorem 7.1.]{poonen} shows that Corollary \ref{miciomicio miao miao} is enough to prove the following:
\begin{corollary}\label{properprojective}
Assume furthermore that $Y_{x}$ is projective for every $x\in |X|$. Then there exists a dense open subscheme $U\subseteq X$ such that the base change  $f_U:U\times_X Y\rightarrow U$ of $f:Y\rightarrow X$ trough $U\subseteq X$ is projective.
\end{corollary} 
\numberwithin{theorem}{subsubsection}
\begin{remark}
Whether the analogue of Corollary \ref{properprojective} holds over fields algebraic over $\F_p$ is not known.
The problem over this kind of fields is that it is not true in general that there exists a NS-generic closed point (as the example of a family of abelian surfaces such that the generic fibre has not complex multiplication shows). 
\end{remark}
\numberwithin{theorem}{subsubsection}
\subsubsection{Hyperplane sections}
From now on, assume that $k$ is finitely generated.
Assume that $Z$ is a smooth projective $k$-variety of dimension $\geq 3$ and let $Z\subseteq \mathbb P_k^n$ be a projective embedding. One can ask whether there exists a smooth hyperplane section $D$ of $Z$ such that the canonical map 
$$NS(Z_{\overline k})\otimes \Q\rightarrow NS(D_{\overline k})\otimes \Q$$
is an isomorphism. This is not true in general (see Example \ref{counter}), but one can apply Theorem \ref{main} to obtain the following arithmetic variant:
\numberwithin{theorem}{subsubsection}

\begin{corollary}\label{exinstencehyperplane}
If $dim(Z)\geq 3$ there are infinitely many smooth $k$-rational hyperplane sections $D\subseteq Z$ such that
the canonical map 
$$NS(Z)\otimes \Q\rightarrow NS(D)\otimes \Q$$ is an isomorphism.
\end{corollary}
\numberwithin{theorem}{subsubsection}
As already mentioned in Section \ref{nerongeneric}, Conjecture \ref{tateconjectureconj} implies Theorem \ref{main}.
The $\ell$-adic Tate conjecture for divisors conjecture is still widely open except for some special classes of varieties like Abelian varieties and K3 surfaces. Using Corollary \ref{exinstencehyperplane}, one can enlarge the class of varieties for which it holds:
\begin{corollary}\label{corollaryTate}
Let $Z$ be a smooth projective $k$-variety of dimension $\geq 3$ and choose a projective embedding $Z\subseteq \mathbb P^n_k$. If $T(D,\ell)$ holds for the smooth hyperplane sections $D\subseteq Z$, then $T(Z,\ell)$ holds.
\end{corollary}
\begin{remark}
Corollary \ref{corollaryTate} can be used to reduce the $\ell$-adic Tate conjecture for divisors on smooth proper $k$-varieties to smooth projective $k$-surfaces, extending an unpublished result (\cite{djtate}) of De Jong (whose proof has been simplified in \cite[Theorem 4.3]{Varmor}) to infinite finitely generated fields.
\end{remark}
\numberwithin{theorem}{subsubsection}
\subsubsection{Uniform boundedness of Brauer groups}
Combining Theorem \ref{main} with the main result of \cite{UOIpositive} and the arguments of \cite{brauer}, one gets the following application to uniform boundedness for the $\ell$-primary torsion of the cohomological Brauer group in smooth proper families of $k$-varieties. 
\begin{corollary}\label{corollarybrauer}
Let $X$ be a smooth geometrically connected $k$-curve and let $f:Y\rightarrow X$ be a smooth proper morphism of $k$-varieties. If $T(Y_{\overline x},\ell)$ holds for all $x\in |X|$, then there exists a constant $C:=C(Y\rightarrow X, \ell)$ such that $$|Br(Y_{\overline x})[\ell^{\infty}]^{\pi_1(x,\overline{x})}|\leq C $$ for all $x\in X(k)$.
\end{corollary}
Corollary \ref{corollarybrauer} extends to positive characteristic the main result of \cite{brauer} and gives some evidence for a positive charateristic version of the conjectures on the uniform boundedness of Brauer group in \cite{VAV}. Elaborating the argument in the proof of Corollary \ref{corollarybrauer}, one  gets also an unconditional variant of Corollary \ref{corollarybrauer} (Corollary \ref{uncondintionalbrauer}) and a result on the specialization of the $p$-adic Tate module of the Brauer group (Corollary \ref{corptate}).
\numberwithin{theorem}{subsection} 
\subsection{Organization of the Paper}
The paper is divided in three parts. Part \ref{part1}, is devoted to the proof of Theorem \ref{main}. In Section \ref{implication} we show that Theorem \ref{ltocrys} implies Theorem \ref{main} and in Section \ref{strategy} we show Theorem \ref{ltocrys}. Part \ref{secapp} is devoted to applications of Theorem \ref{main}: in Section \ref{sechyperplane} we prove Corollary \ref{exinstencehyperplane}, and in Section \ref{secbrauer} we give the proof of Corollary \ref{corollarybrauer}. Finally, in Part \ref{appendix}, we prove the overconvergence of the higher direct image in crystalline cohomology (Theorem \ref{berthelotconjecture}), which is used in the proof of Theorem \ref{ltocrys}.
\subsection{Acknowledgements}
This paper is part of the author’s Ph.D. thesis under the supervision of Anna Cadoret.
He thanks her for suggesting the problem and several sessions of discussion. He is also very grateful for her careful
re-readings of this paper and her constructive suggestions. Several hours of discussion with Marco D'addezio on overconvergent isocrystals and their monodromy groups has been of great help. The author would like to thanks Tomoyuki Abe for having pointed out that there might be a gap in \cite{berthelotconjecture} and for suggesting the use of Shiho's work on relative rigid cohomology to get the overconvergence of the higher direct image in crystalline cohomology. He thanks also Atushi Shiho for kindly answering some technical questions on his work and for some comments on a preliminary draft of this paper. This circle of ideas has been presented in a small course at RIMS, a Joint Usage/Research Center located in Kyoto University. The author would like to thanks the participants for their interest and Akio Tamagawa for the invitation. A discussion around Example \ref{example} with Matilde Manzaroli has been of great help. Finally the author thanks Vincent de Daruvar and Gregorio Baldi for their interest and for some help to improve the exposition.
\part{Proof of Theorem \texorpdfstring{\ref{main}}-}\label{part1}
\section{Proof of Theorem \texorpdfstring{\ref{main}}- assuming Theorem \texorpdfstring{\ref{ltocrys}}-}\label{implication}
\numberwithin{theorem}{subsection} 
In Section \ref{prelim}, we collect some preliminary remarks. The proof when $f:Y\rightarrow X$ is proper is a technical elaboration (involving alteration and the trace formalism) of the proof when $f:Y\rightarrow X$ is projective. To clarify the exposition we carry out the proof when $f:Y\rightarrow X$ is projective in Section \ref{problectionsection} and turn to the general case in Section \ref{seccrys}.  
\subsection{Preliminary remarks}\label{prelim}
\subsubsection{Strictly generic vs generic}
Observe that the assertion for Galois generic points implies the assertion for strictly Galois generic points. Indeed, strictly Galois generic implies Galois generic, hence for a strictly Galois generic point $x\in X$ the specialization morphism

$$sp_{\eta,x}:NS(Y_{\overline \eta})\otimes \Q\rightarrow  NS(Y_{\overline x})\otimes \Q$$
is an isomorphism. Recall that, as explained in \ref{nerongeneric}, the map $sp_{\eta,x}$ is $\pi_1(x,\overline x)$-equivariant. Since $\pi_1(x,\overline x)$ and $\pi_1(X,\overline \eta)\simeq \pi_1(X,\overline x)$ acting on $H^2(Y_{\overline \eta},\Ql(1))\simeq H^2(Y_{\overline x},\Ql(1))$ have the same image $\Pi_{\ell}$ (since $x$ is strictly Galois generic), taking $\Pi_{\ell}$-invariants in $sp_{\eta,x}$, one deduces the statement for strictly Galois generic points. So, from on, we focus on the assertion for Galois generic points. To simplify, in this section, we omit base points in our notation for the \etale fundamental group. 
\subsubsection{Finite cover}\label{etale cover}
If $X'\rightarrow X$ is a surjective finite morphism of smooth connected $k$-varieties, the map $\pi_1(X')\rightarrow \pi_1(X)$ has open image. So $x\in X$ is Galois generic (resp. NS generic) for $f:Y\rightarrow X$ if and only if any lifting $x'\in X'$ of $x$ if Galois generic (resp. NS generic) for the base change $f_{X'}:Y'\times_XX'\rightarrow X'$ of $f:Y\rightarrow X$ along $X'\rightarrow X$. As a consequence we can freely replace $X$ with $X'$ during the proof.
\subsection{Proof when \texorpdfstring{$f$}- is projective}\label{problectionsection}
Let $f:Y\rightarrow X$ be smooth projective. For the general strategy of the proof see Section \ref{introspreading}.
\numberwithin{theorem}{subsubsection}
\subsubsection{Step 1: Spreading out}\label{spreading}
Replacing $k$ with a finite field extension (\ref{etale cover}), one can assume that there exists a finite field $\F_q$, smooth and geometrically connected $\F_q$-varieties $\mathcal K$, $\mathcal Z$ with generic points $\zeta:k\rightarrow \mathcal K$, $\beta:k(x)\rightarrow \mathcal Z$ and a commutative cartesian diagram
 \begin{center}
\begin{tikzcd}
\mathcal Y_{\mathfrak t} \arrow{r}\arrow{d}\arrow[phantom]{rd}{\Box}&\mathcal Y_{\mathcal Z}\arrow{r}\arrow{d}{\mathfrak f_{\mathcal Z}}\arrow[phantom]{rd}{\Box}& \mathcal Y\arrow{d}{\mathfrak f}\arrow[phantom]{rd}{\Box} & Y\arrow{l}\arrow{d}{f}\arrow[phantom]{rd}{\Box}&Y_x\arrow{l}\arrow{d}{f_x}\arrow[bend right]{lll}\\

\F_q \arrow{r}{\mathfrak t}\arrow[equal]{dr}&\mathcal Z\arrow{d}\arrow{r}\arrow{rd}&\mathcal X\arrow[phantom]{rd}{\Box}\arrow{d}& X\arrow{l}\arrow{d}&k(x)\arrow[swap]{l}{x}\arrow[bend left,near start]{lll}{\beta}\\
              & \F_q & \arrow{l}\mathcal K & k\arrow{l}{\zeta}


\end{tikzcd}
\end{center}
where $\mathfrak f:\mathcal Y\rightarrow \mathcal X$ is a smooth projective morphism and the base change of $f_{\mathcal Z}:\mathcal Y_{\mathcal Z}\rightarrow \mathcal Z$ along $\beta:k(x)\rightarrow \mathcal Z$ identifies with $f_x:Y_x\rightarrow k(x)$. 
Replacing $X$ with a finite \etale cover (\ref{etale cover}) one can also assume that
\begin{enumerate}
\item $NS(Y_{\overline x})\otimes \Q=NS(Y_{x})\otimes \Q$ and $NS(\mathcal Y_{\overline{\mathfrak t}})\otimes \Q=NS(\mathcal Y_{\mathfrak t})\otimes \Q$;
\item the Zariski closures of the images of $\pi_1(\mathcal X)\rightarrow GL(H^{2}(Y_{\overline \eta},\Ql(1)))$ and $\pi_1(\mathcal X_{\F})\rightarrow  GL(H^{2}(Y_{\overline \eta},\Ql(1)))$ are connected.
\end{enumerate}
Note that, by smooth proper base change, one has the following factorization
\begin{center}
\begin{tikzcd}
\pi_1(x)\arrow{r}\arrow[two heads]{d} &\pi_1(X)\arrow[two heads]{d}\arrow{r} &GL(H^{2}(Y_{\overline \eta},\Ql(1)))\simeq GL(H^{2}(\mathcal Y_{\overline t},\Ql(1)))\\
\pi_1(\mathcal Z)\arrow{r}&\pi_1(\mathcal X)\arrow{ru}.
\end{tikzcd}
\end{center}
In particular, since $x$ is Galois generic with respect to $f:Y\rightarrow X$, the image of $\pi_1(\mathcal Z)\rightarrow \pi_1(\mathcal X)\rightarrow GL(H^{2}(\mathcal Y_{\overline t},\Ql(1)))$ is open in the image of $\pi_1(\mathcal X)\rightarrow GL(H^{2}(\mathcal Y_{\overline t},\Ql(1)))$.
Hence by \ref{spreading}(2) and Theorem \ref{ltocrys} the base change map
$$H^0(\mathcal X,R^2\mathfrak f_{crys,*}\mathcal O_{\mathcal Y/W})^{F=q}\otimes \Q\rightarrow H^0(\mathcal Z,R^2\mathfrak f_{\mathcal Z,crys,*}\mathcal O_{\mathcal Y_{\mathcal Z}/W})^{F=q}\otimes \Q$$
is an isomorphism.
\numberwithin{theorem}{subsubsection}
\subsubsection{Step 2: Using the variational Tate conjecture}\label{diagram}
Since $\mathfrak t$ is a specialization of $x$ (in $\mathcal Z$) and $x$ is a specialization of $\eta$ (in $\mathcal X$), there is a canonical commutative diagram
\begin{center}
\begin{tikzcd}[column sep=tiny, row sep=small]
z\in Pic(\mathcal Y)\otimes \Q \arrow[bend right=65,swap]{dddd}{c_{\mathcal Y}}\arrow{rr}\arrow{d}\arrow{dr}{i_{\mathfrak t}^*}& & Pic(\calY_{\mathcal Z})\otimes \Q\ni z_x\arrow[two heads]{d}{(i)}\arrow[bend left=65]{dddd}{c_{\mathcal Y_{\mathcal Z}}} \arrow[swap]{dl}{i_{\mathfrak t}^*}\\
Pic(Y_{\eta})\otimes \Q\arrow{d}& Pic(\mathcal Y_{\mathfrak t})\otimes \Q\arrow[near start,two heads]{dd}{(ii)}\arrow[bend left=100]{dddd}{c_{\mathcal Y_{\mathfrak t}}}\ni z_t & Pic(Y_x)\otimes \Q\arrow[two heads]{d}{(ii)}\\
\epsilon_{\eta} \in NS(Y_{\overline{\eta}})\otimes \Q\arrow[hook, near start]{rr}{sp_{\eta,x}}\arrow[hook]{dr}{sp_{\eta,\mathfrak t}}& & NS(Y_{\overline x})\otimes \Q \ni \epsilon_x \arrow[hook,swap]{dl}{sp_{x,\mathfrak t}}\\
& NS(\mathcal Y_{\mathfrak t})\otimes \Q=NS(\mathcal Y_{\overline{\mathfrak t}})\arrow[hook]{dd}\otimes \Q \ni \epsilon_{\mathfrak t} &\\


H^2_{crys}(\mathcal Y)\arrow{rd}{i_{\mathfrak t}^*}\arrow[swap]{dd}{Leray} & & H^2_{crys}(\mathcal Y_{\mathcal Z})\arrow{dd}{Leray}\arrow[near end, swap]{ld}{i_{\mathfrak t}^*}\\
& H^2_{crys}(\mathcal Y_{\mathfrak t}) &\\
H^0(\mathcal X,R^2\mathfrak f_{crys,*}\mathcal O_{\mathcal Y/W})\otimes \Q\arrow[hook]{ur}\arrow[hook]{rr}& &H^0(\mathcal Z,R^2\mathfrak f_{\mathcal Z,crys,*}\mathcal O_{\mathcal Y_{\mathcal Z}/W})\otimes \Q\arrow[hook]{ul}\\
H^0(\mathcal X,R^2\mathfrak f_{crys,*}\mathcal O_{\mathcal Y/W})^{F=q}\otimes \Q\arrow{rr}{(iii)}\arrow[hook]{u}& &H^0(\mathcal Z,R^2\mathfrak f_{\mathcal Z,crys,*}\mathcal O_{\mathcal Y_{\mathcal Z}/W})^{F=q}\otimes \Q\arrow[hook]{u}
\end{tikzcd}
\end{center}
where the arrow $(i)$ is surjective, since an open immersion of smooth varieties induces a surjection on the Picard groups, the arrows $(ii)$ are surjective by \ref{spreading}$(1)$ and the arrow $(iii)$ is an isomorphism by Theorem \ref{ltocrys}. 

The images of 
$$Pic(\mathcal Y)\otimes \Q\rightarrow H^0(\mathcal X,R^2\mathfrak f_{crys,*}\mathcal O_{\mathcal Y/W})\otimes \Q\quad \text{and}\quad Pic(\calY_{\mathcal Z})\otimes \Q\rightarrow H^0(\mathcal Z,R^2\mathfrak f_{\mathcal Z,crys,*}\mathcal O_{\mathcal Y_{\mathcal Z}/W})\otimes \Q$$ lie in $H^0(\mathcal X,R^2\mathfrak f_{crys,*}\mathcal O_{\mathcal Y/W})^{F=q}\otimes \Q$ and $H^0(\mathcal Z,R^2\mathfrak f_{\mathcal Z,crys,*}\mathcal O_{\mathcal Y_{\mathcal W}/K})^{F=q}\otimes \Q$ respectively. Take an $\epsilon_x$ in $NS(Y_x)\otimes \Q$ with lifting $z_x\in Pic(\mathcal Y_{\mathcal Z})\otimes \Q$ and write 
$$z_{\mathfrak t}:=i_{\mathfrak t}^*(z_x)\in Pic(\mathcal Y_{\mathfrak t})\otimes \Q \quad \text{and} \quad \epsilon_{\mathfrak t}=sp_{x,\mathfrak t}(\epsilon_x)=c_{\mathcal Y_{\mathfrak t}}i_{\mathfrak t}^*(z_x)\in NS(\mathcal Y_{\mathfrak t})\otimes \Q.$$
By construction $\epsilon_{\mathfrak t}$ is in the image of $$(iii):H^0(\mathcal X,R^2\mathfrak f_{crys,*}\mathcal O_{\mathcal Y/W})^{F=q}\otimes \Q\xrightarrow{\sim} H^0(\mathcal Z,R^2\mathfrak f_{\mathcal Z,crys,*}\mathcal O_{\mathcal Y_{\mathcal Z}/W})^{F=q}\otimes \Q . $$ Moreover $\epsilon_{\mathfrak t}=c_{\mathcal Y_{\mathfrak t}}i_{\mathfrak t}^*(z_x)$ and so, by Fact \ref{thmvariationalTate} applied to $\mathfrak f:\mathcal Y\rightarrow \mathcal Z$ and $\mathfrak t$, there exists $z\in Pic(\mathcal Y)\otimes \Q$ such that $i_{\mathfrak t}^*c_{\mathcal Y}(z)=\epsilon_{\mathfrak t}$. Let $\epsilon_{\eta}$ be the image of $z$ in $NS(Y_{\overline{\eta}})\otimes \Q$. By construction and the commutativity of the diagram one has:
$$sp_{x,\mathfrak t}(sp_{\eta,x}(\epsilon_{\eta}))=sp_{\eta,\mathfrak t}(\epsilon_{\eta})=\epsilon_{\mathfrak t}=sp_{x,\mathfrak t}(\epsilon_x)$$
Since $sp_{x,\mathfrak t}$ is injective, this concludes the proof of Theorem \ref{main} when $f$ is projective.
\endproof
\numberwithin{theorem}{subsection} 
\subsection{Proof when \texorpdfstring{$f$}- is proper}\label{seccrys}
Assume now that $f:Y\rightarrow X $ is only proper. Since Fact \ref{thmvariationalTate} is only available when $f$ is projective, we cannot longer apply it directly to $\mathfrak f:\mathcal Y\rightarrow \mathcal X$. To overcome this difficulty we proceed as follows. Using De Jong's alteration theorem and replacing $X$ with a dense open subset, one first constructs a commutative diagram
\begin{center}
\begin{tikzcd}
\widetilde Y\arrow{rr}{g}\arrow[swap]{dr}{\widetilde f}&& Y\arrow{dl}{f}\\
& X
\end{tikzcd}
\end{center}
with $\widetilde f$ smooth projective and $g$ dominant and generally finite. While every $x\in X$ which is NS-generic for $\widetilde f$ is NS-generic for $f$ (as the argument in \ref{trace} shows), the hypothesis of being Galois generic for $f$ does not transfer to $\widetilde f$ in general, so that one cannot reduce directly the assertion for the (proper) morphism $f:Y\rightarrow X$ to the assertion for the (projective) morphism $\widetilde f:\widetilde Y\rightarrow X$. However, the trace formalism is functorial enough to allow us to transfer information from $f:Y\rightarrow X$ to $\widetilde f:\widetilde Y\rightarrow X$ for cohomology classes coming from $Y$.
\subsubsection{Step 1: De Jong's alterations theorem}
First one reduces to the situation where$f$ has geometrically connected fibres (this hypothesis is used in \ref{trace} to apply Poincare duality). By \cite[X, Proposition 1.2]{EGA} and replacing $X$ with a finite \etale cover (\ref{etale cover}), one can assume that $f:Y\rightarrow X$ decomposes in a disjoint union of morphisms $f_i:Y_i\rightarrow X$ with geometrically connected fibres.
Since for every (not necessarily closed) point $x\in X$ there are natural decompositions
\begin{center}
\begin{tikzcd}
NS(Y_{\overline x})\otimes\Q\arrow{r}{c_{Y_{\overline x}}}\arrow{d}{\simeq}&H^2(Y_{\overline x},\Ql(i))\arrow{d}{\simeq}\\
\oplus_i NS(Y_{i,\overline x})\otimes\Q\arrow{r}{c_{Y_{\overline x}}}&\oplus_i H^2(Y_{i,\overline x},\Ql(i))
\end{tikzcd}
\end{center}
one may work with each $f_i:Y_i\rightarrow X$ separately and hence assume that $f:Y\rightarrow X$ has geometrically connected fibres.

By De Jong's alterations theorem (\cite{alteration}) for $Y_{\overline \eta}$ over $\overline{k(\eta)}$, there exists a proper, surjective and generically finite morphism $\widetilde Y_{\overline \eta}\rightarrow Y_{\overline \eta}$, where $\widetilde Y_{\overline \eta}$ is a connected, smooth and projective $\overline{k(\eta)}$-variety. By descent and spreading out, there exists a commutative diagram of connected smooth $k$-varieties:
\begin{center}
\begin{tikzcd}
\widetilde Y_{\overline \eta}\arrow{r}\arrow{d}\arrow[phantom]{rd}{\Box} & \widetilde Y_{\eta'}\arrow[phantom]{rd}{\Box}\arrow{r}\arrow{d} &\widetilde Y\arrow{d}{g}\arrow[bend right, swap,near start]{dd}{\widetilde f}\\
Y_{\overline \eta}\arrow{r}\arrow{d}\arrow[phantom]{rd}{\Box} & Y_{\eta'}\arrow{r}\arrow{d}\arrow[phantom]{rd}{\Box}&Y_{U'}\arrow{r}\arrow{d}{f_{U'}}\arrow{d}\arrow[phantom]{rd}{\Box} & Y_U\arrow{r}\arrow{d}{f_U}\arrow[phantom]{rd}{\Box} & Y\arrow{d}{f}\\
\overline{k(\eta)} \arrow{r} & k(\eta')\arrow{r}{\eta'} & U'\arrow{r}{j}&U\arrow{r}{i}& X
\end{tikzcd}
\end{center}
where $\eta':k(\eta')\rightarrow U'$ is the generic point of $U'$, $i:U\rightarrow X$ is a open immersion with dense image, $j:U'\rightarrow U$ is a finite surjective morphism, $\widetilde f:\widetilde Y\rightarrow U'$ is smooth, projective with geometrically connected fibres and $g:\widetilde Y\rightarrow Y_{U'}$ is proper, surjective and generically finite. In conclusion, replacing $X$ with $U'$ (\ref{etale cover}), one can assume that there exists a diagram
\begin{center}
\begin{tikzcd}
\widetilde Y\arrow{rr}{g}\arrow[swap]{dr}{\widetilde f}&& Y\arrow{dl}{f}\\
& X
\end{tikzcd}
\end{center}
where $\widetilde f:\widetilde Y\rightarrow X$ is smooth projective with geometrically connected fibres, $f:Y\rightarrow X$ is smooth proper with geometrically connected  fibres and $g:\widetilde Y\rightarrow Y$ is generically finite and dominant.
\numberwithin{theorem}{subsubsection}
\subsubsection{Step 2: Spreading out}
Now one spreads out to finite fields. Up to replacing $k$ with a finite field extension (\ref{etale cover}), there exists a finite field $\F_q$, smooth and geometrically connected $\F_q$-varieties $\mathcal K$,$\mathcal Z$ with generic points $\zeta:k\rightarrow \mathcal K$, $\beta:k(x)\rightarrow \mathcal Z$ and a commutative cartesian diagram:
 \begin{center}
\begin{tikzcd}
\widetilde{\mathcal Y}_{\mathfrak t}\arrow{r}\arrow{d}{\mathfrak g_{\mathfrak t}}\arrow[phantom]{rd}{\Box}\arrow[bend right,swap,near start]{dd}{\widetilde{\mathfrak f}_{\mathfrak t}} & \widetilde{\mathcal Y}_{\mathcal Z}\arrow{r}\arrow{d}{\mathfrak g_{\mathcal Z}}\arrow[phantom]{rd}{\Box} \arrow[bend right,swap,near start]{dd}{\widetilde{\mathfrak f}_{\mathcal Z}}& \widetilde{\mathcal Y}\arrow{d}{\mathfrak g} \arrow[phantom]{rd}{\Box}\arrow[bend right,swap,near start]{dd}{\widetilde{\mathfrak f}}& \arrow{l} \widetilde Y\arrow[bend right, swap, near start]{dd}{\widetilde f}\arrow{d}{g}\arrow[phantom]{rd}{\Box}& \arrow{l}\widetilde Y_x\arrow{d}{g_x}\\
\mathcal Y_{\mathfrak t} \arrow{r}\arrow{d}\arrow[phantom]{rd}{\Box}&\mathcal Y_{\mathcal Z}\arrow{r}\arrow{d}{\mathfrak f_{\mathcal Z}}\arrow[phantom]{rd}{\Box}& \mathcal Y\arrow{d}{\mathfrak f}\arrow[phantom]{rd}{\Box} & Y\arrow{l}\arrow{d}{f}\arrow[phantom]{rd}{\Box}&Y_x\arrow{l}\arrow{d}{f_x}\\

\F_q \arrow{r}{\mathfrak t}\arrow[equal]{dr}&\mathcal Z\arrow{d}\arrow{r}&\mathcal X\arrow[phantom]{rd}{\Box}\arrow{d}& X\arrow{l}\arrow{d}&k(x)\arrow[swap]{l}{x}\arrow[bend left]{lll}{\beta}\\
              & \F_q & \arrow{l}\mathcal K & k\arrow{l}{\zeta}


\end{tikzcd}
\end{center}
where $\mathfrak f:\mathcal Y\rightarrow \mathcal X$ is smooth proper with geometrically connected fibres, $\widetilde{\mathfrak f}:\widetilde{\mathcal Y}\rightarrow \mathcal X$ is smooth projective with geometrically connected fibres, $\mathfrak{g}:\widetilde{\mathcal Y}\rightarrow \mathcal Y$ is a dominant generically finite morphism and the base change of $\widetilde{\mathcal Y}_{\mathcal Z}\rightarrow \mathcal Y_{\mathcal Z}\rightarrow \mathcal Z$ along $k(x)\rightarrow \mathcal Z$ identifies with $\widetilde Y_x\rightarrow Y_x\rightarrow k(x)$. 
Replacing $X$ with a finite \etale cover (\ref{etale cover}) one can also assume that
\begin{enumerate}
\item $NS(Y_{\overline x})\otimes \Q=NS(Y_{x})\otimes \Q$, $NS(\mathcal Y_{\overline{\mathfrak t}})\otimes \Q=NS(\mathcal Y_{\mathfrak t})\otimes \Q$, $NS(\widetilde{\mathcal Y}_{\mathfrak t})\otimes \Q=NS(\widetilde{\mathcal Y}_{\overline{\mathfrak t}})\otimes \Q$;
\item the Zariski closures of the images of $\pi_1(\mathcal X)\rightarrow GL(H^{2}(Y_{\overline \eta},\Ql(1)))$ and $\pi_1(\mathcal X_{\F})\rightarrow  GL(H^{2}(Y_{\overline \eta},\Ql(1)))$ are connected.
\end{enumerate}
Note that, by smooth proper base change, one has the following factorization
\begin{center}
\begin{tikzcd}
\pi_1(x)\arrow{r}\arrow[two heads]{d} &\pi_1(X)\arrow[two heads]{d}\arrow{r} &GL(H^{2}(Y_{\overline \eta},\Ql(1)))\simeq GL(H^{2}(\mathcal Y_{\overline t},\Ql(1)))\\
\pi_1(\mathcal Z)\arrow{r}&\pi_1(\mathcal X)\arrow{ru}
\end{tikzcd}
\end{center}
In particular, since $x$ is Galois generic for $f:Y\rightarrow X$, the image of $\pi_1(\mathcal Z)\rightarrow \pi_1(\mathcal X)\rightarrow GL(H^{2}(\mathcal Y_{\overline t},\Ql(1)))$ is open in the image of $\pi_1(\mathcal X)\rightarrow GL(H^{2}(\mathcal Y_{\overline t},\Ql(1)))$.
Hence by (2) and Theorem \ref{ltocrys} the base change map
$$H^0(\mathcal X,R^2\mathfrak f_{crys,*}\mathcal O_{\mathcal Y/W})^{F=q}\otimes \Q\rightarrow H^0(\mathcal Z,R^2\mathfrak f_{\mathcal Z,crys,*}\mathcal O_{\mathcal Y_{\mathcal Z}/W})^{F=q}\otimes \Q$$
is an isomorphism.

\numberwithin{theorem}{subsubsection}
\subsubsection{Step 3: Using the Variational Tate conjecture}
Take an $\epsilon_x$ in $NS(Y_x)\otimes \Q$. The goal of this subsection is to prove that there exists a $\widetilde{\epsilon}_{\eta}\in NS(\widetilde{Y}_{\overline \eta})\otimes \Q$ such that $\widetilde{sp}_{\eta,x}(\widetilde{\epsilon}_{\eta})=g^*(\epsilon_x)$, where 
$$\widetilde{sp}_{\eta,x}:NS(\widetilde{Y}_{\overline \eta})\otimes \Q\rightarrow NS(\widetilde{Y}_{\overline x})\otimes \Q$$
is the specialization map for $\widetilde f:\widetilde Y\rightarrow X$. Consider the commutative diagram in \ref{diagram}.
Let $z_x\in Pic(\mathcal Y_{\mathcal Z})\otimes \Q$ be a lift of $\epsilon_x$ and write 
$$z_{\mathfrak t}:=i_{\mathfrak t}^*(z_x)\in Pic(\mathcal Y_{\mathfrak t})\otimes \Q \quad \text{and} \quad \epsilon_{\mathfrak t}=sp_{x,\mathfrak t}(\epsilon_x)=c_{\mathcal Y_{\mathfrak t}}i_{\mathfrak t}^*(z_x)\in NS(\mathcal Y_{\mathfrak t})\otimes \Q.$$
By construction $\epsilon_{\mathfrak t}$ is in the image of 
$$(iii):H^0(\mathcal X,R^2\mathfrak f_{crys,*}\mathcal O_{\mathcal Y/W})^{F=q}\otimes \Q \xrightarrow{\sim} H^0(\mathcal Z,R^2\mathfrak f_{\mathcal Z,crys,*}\mathcal O_{\mathcal Y_{\mathcal Z}/W})^{F=q}\otimes \Q.$$
Moreover $\epsilon_{\mathfrak t}=c_{\mathcal Y_{\mathfrak t}}i_{\mathfrak t}^*(z_x)$. 
Since $\mathfrak f:\mathcal Y\rightarrow \mathcal X$ is only assumed to be proper, one cannot apply directly Fact \ref{thmvariationalTate} to it.
However the previous reasoning shows that 
$$H^0(\mathcal X, R^2\widetilde{\mathfrak f}_{*,crys}\mathcal O_{\widetilde{\mathcal Y}/W})\otimes \Q \supseteq \mathfrak g^*(H^0(\mathcal X,R^2\mathfrak f_{crys,*}\mathcal O_{\mathcal Y/W})\otimes \Q)\ni \mathfrak g^*_{\mathfrak t}(\epsilon_{\mathfrak t})=\mathfrak g^*_{\mathfrak t}(c_{\mathcal Y_{\mathfrak t}}i_{\mathfrak t}^*(z_x))=c_{\widetilde{\mathcal Y}_{\mathfrak t}}\widetilde i_{\mathfrak t}^*\mathfrak g^*_{\mathcal Z}(z_x),$$
where the notation is as in the canonical commutative diagram:
\begin{center}
\begin{tikzcd}[column sep=tiny, row sep=small]
\widetilde z\in Pic(\widetilde{\mathcal Y})\otimes \Q \arrow[bend right=65,swap]{dddd}{c_{\widetilde{\mathcal Y}}}\arrow{rr}\arrow{d}{\widetilde i_{\eta}}\arrow{dr}{\widetilde i_{\mathfrak t}^*}& & Pic(\widetilde{\calY}_{\mathcal Z})\otimes \Q\ni \mathfrak g_{\mathcal Z}^*(z_x)\arrow[two heads]{d}{(i)}\arrow[bend left=65]{dddd}{c_{\widetilde{\mathcal Y}_{\mathcal Z}}} \arrow[swap]{dl}{\widetilde i_{\mathfrak t}^*}\\


Pic(\widetilde{Y}_{\eta})\otimes \Q\arrow{d}& Pic(\widetilde{\mathcal Y}_{\mathfrak t})\otimes \Q\arrow[near start,two heads]{dd}\arrow[bend left=100]{dddd}{c_{\widetilde{\mathcal Y}_{\mathfrak t}}}\ni z_t & Pic(\widetilde{Y}_x)\otimes \Q\arrow[two heads]{d}{(ii)}\\

\widetilde{\epsilon}_{\eta} \in NS(\widetilde{Y}_{\overline{\eta}})\otimes \Q\arrow[hook, near start]{rr}{\widetilde{sp}_{\eta,x}}\arrow[hook]{dr}{\widetilde{sp}_{\eta,\mathfrak t}}& & NS(\widetilde{Y}_{\overline x})\otimes \Q \ni g^*(\epsilon_x) \arrow[hook,swap]{dl}{\widetilde{sp}_{x,\mathfrak t}}\\

&NS(\widetilde{\mathcal Y}_{\overline{\mathfrak t}})\arrow[hook]{dd}\otimes \Q \ni \mathfrak g_{\mathfrak t}(\epsilon_{\mathfrak t}) &\\


H^2_{crys}(\widetilde{\mathcal Y})\arrow{rd}{\widetilde i_{\mathfrak t}^*}\arrow[swap]{dd}{Leray} & & H^2_{crys}(\widetilde{\mathcal Y}_{\mathcal Z})\arrow{dd}{Leray}\arrow[near end, swap]{ld}{\widetilde i_{\mathfrak t}^*}\\
& H^2_{crys}(\widetilde{\mathcal Y}_{\mathfrak t}) &\\

\mathfrak g^*(\epsilon_{\mathfrak t})\in H^0(\mathcal X,R^2\widetilde{\mathfrak f}_{crys,*}\mathcal O_{\widetilde{\mathcal Y}/W})\otimes \Q\arrow[hook]{ur}\arrow[hook]{rr}& &H^0(\mathcal Z,R^2\widetilde{\mathfrak f}_{\mathcal Z,crys,*}\mathcal O_{\widetilde{\mathcal Y}_{\mathcal Z}/W})\otimes \Q\arrow[hook]{ul}\ni \mathfrak g_{\mathcal Z}^*(\epsilon_{\mathfrak t})\\
\epsilon_{\mathfrak t}\in H^0(\mathcal X,R^2\mathfrak f_{crys,*}\mathcal O_{\mathcal Y/W})\otimes \Q\arrow{u}{\mathfrak g^*}\arrow{rr}& &H^0(\mathcal Z,R^2\mathfrak f_{\mathcal Z,crys,*}\mathcal O_{\mathcal Y_{\mathcal Z}/W})\otimes \Q\arrow{u}{\mathfrak g_{\mathcal Z}^*}\ni \epsilon_{\mathfrak t}
\end{tikzcd}
\end{center}
So, by Fact \ref{thmvariationalTate} applied to $\widetilde{\mathfrak f}:\widetilde{\mathcal Y}\rightarrow \mathcal X$ and $\mathfrak t$, there exists $\widetilde z\in Pic(\widetilde{\mathcal Y})\otimes \Q$ such that $\mathfrak g^*_{\mathfrak t}(\epsilon_{\mathfrak t})=\widetilde i_{\mathfrak t}^*c_{\widetilde{\mathcal Y}}(\widetilde z)$.
Write $\widetilde{\epsilon}_{\eta}:=\widetilde i_{\eta}^*c_{\widetilde{\mathcal Y}}(\widetilde z)$.
From the commutative diagram
\begin{center}
\begin{tikzcd}[row sep=tiny]
Pic(\widetilde{\mathcal Y})\otimes \Q\arrow{dd} \arrow{r}{\widetilde i^*_{\eta}} &Pic(\widetilde Y_{\eta})\otimes \Q\arrow{r}{c_{\widetilde Y_{\eta}}}& NS(\widetilde Y_{\overline \eta})\otimes \Q\arrow{dd}{\widetilde sp_{\eta,\mathfrak t}}\arrow{rd}{\widetilde sp_{\eta,x}}\\
&&&NS(\widetilde Y_{\overline{x}})\arrow{ld}{\widetilde sp_{x,\mathfrak t}}\\
Pic(\widetilde{\mathcal Y}_{\mathfrak t})\arrow{rr}{c_{\widetilde{\mathcal Y}_{\mathfrak t}}} && NS(\widetilde{\mathcal Y}_{\overline{\mathfrak t}})\otimes \Q
\end{tikzcd}
\end{center}
one deduces $$\widetilde{sp}_{x,\mathfrak t}(\widetilde{sp}_{\eta,x}(\widetilde{\epsilon}_{\eta}))=\widetilde{sp}_{x,\mathfrak t}(g^*\epsilon_{x}).$$
Since $\widetilde{sp}_{x,\mathfrak t}$ is injective this implies
$$\widetilde{sp}_{\eta,x}(\widetilde{\epsilon}_{\eta})=g^*(\epsilon_{x}).$$
\numberwithin{theorem}{subsubsection}
\subsubsection{Step 4: Trace argument}\label{trace}
To conclude the proof one has to descend from $\widetilde{Y}$ to $Y$. For this we use the trace formalism.
Since $f:Y\rightarrow X$ and $\widetilde f:\widetilde Y\rightarrow X$ are smooth proper morphisms with geometrically connected fibres, by the relative Poincar\'e duality (\cite[Expos\'e XVIII]{SGA4}), there are canonical isomorphisms
$$R^2f_*\Q_{\ell}\simeq (R^{2d-2}f_*\Ql(d))^{\vee}\quad \text{and} \quad R^2\widetilde f_*\Q_{\ell}\simeq (R^{2d-2}\widetilde f_*\Ql(d))^{\vee},$$
where $d=Dim(Y_x)=Dim(\widetilde Y_x)$.
Dualizing and twisting the base change map
$$R^{2d-2}f_*\Ql(d)\rightarrow R^{2d-2}\widetilde f_*\Ql(d),$$
one gets a morphism
$$g_*:R^2\widetilde f_*\Ql(1)\simeq (R^{2d-2}\widetilde f_*\Ql(d))^{\vee}(1)\rightarrow (R^{2d-2}f_*\Ql(d))^{\vee}(1)\simeq R^2f_*\Ql(1).$$
By the compatibility of Poincar\'e duality with base change, for every (not necessarily closed) $x\in X$, the fibre of $g_*$ at $\overline x$ is the usual push forward map $g_{x,*}:H^2(\widetilde Y_{\overline x},\Ql(1))\rightarrow H^2(Y_{\overline x},\Ql(1))$
in \etale cohomology. In particular it is compatible with the push forward of algebraic cycles $g_{x,*}:Pic(\widetilde Y_{\overline x})\otimes \Q\rightarrow Pic(Y_{\overline x})\otimes \Q$. Since $g^*$ and $g_*$ are maps of sheaves, they are compatible with the specialization isomorphisms and hence the following canonical diagram commutes:
\begin{center}
\begin{tikzcd}
Pic(Y_{\overline{\eta}})\otimes \Q\arrow{r}{g^*}\arrow{d}{c_{Y_{\overline \eta}}} & Pic(\widetilde Y_{\overline{\eta}})\otimes \Q\arrow{r}{g_*}\arrow{d}{c_{\widetilde Y_{\overline \eta}}} & Pic(Y_{\overline{\eta}})\otimes \Q\arrow{d}{c_{Y_{\overline \eta}}}\\
H^2(Y_{\overline{\eta}},\Ql(1))\arrow{r}{g^*}\arrow{d}{sp_{\eta,x}} & H^2(\widetilde Y_{\overline{\eta}},\Ql(1))\arrow{r}{g_*}\arrow{d}{sp'_{\eta,x}} &H ^2(Y_{\overline{\eta}},\Ql(1))\arrow{d}{sp_{\eta,x}}\\
H^2(Y_{\overline x},\Ql(1))\arrow{r}{g^*} & H^2(\widetilde Y_{\overline x},\Ql(1))\arrow{r}{g_*} &H ^2(Y_{\overline x},\Ql(1))\\
Pic(Y_{\overline x})\otimes \Q\arrow{r}{g^*}\arrow[swap]{u}{c_{Y_{\overline x}}} & Pic(\widetilde Y_{\overline x})\otimes \Q\arrow{r}{g_*}\arrow[swap]{u}{c_{\widetilde Y_{\overline x}}} & Pic(Y_{\overline x})\otimes \Q\arrow[swap]{u}{c_{Y_{\overline x}}}
\end{tikzcd}
\end{center}
Since the the horizontal arrows are the multiplication by $n:=deg(g)$, one concludes observing that $g_*c_{\widetilde Y_{\overline \eta}}(\widetilde{\epsilon}_\eta)\in NS(Y_{\overline \eta})\otimes \Q$ is such that $sp_{\eta,x}(\frac{1}{n}g_*c_{\widetilde Y_{\overline \eta}}(\widetilde{\epsilon}_\eta))=\epsilon_x$. 
\endproof
\section{Proof of Theorem \texorpdfstring{\ref{ltocrys}}-}\label{strategy}
This section is devoted to the proof of Theorem \ref{ltocrys}. In Section \ref{tannkianref}, after recalling the various categories of isocrystals needed in our argument, we reformulate Theorem \ref{ltocrys} in terms of monodromy groups of F-overconvergent isocrystals. In Section \ref{conclusion}, we use independence techniques to prove Theorem \ref{ltocrys}.
\numberwithin{theorem}{subsection} 
\subsection{Tannakian reformulation of Theorem \texorpdfstring{\ref{ltocrys}}-}\label{tannkianref}
\numberwithin{theorem}{subsubsection}
\subsubsection{Overconvergent isocrystals}
Let $\mathcal X$ be a smooth geometrically connected $\F_q$-variety with $q=p^s$ and write $F$ for $s$-power of the absolute Frobenius on $\mathcal X$. Write $W:=W(\F_q)$ for the Witt ring of $\F_q$, $K$ for its fraction field and  $Mod(\mathcal X|W)$ for the category of $\mathcal O_{\mathcal X|W}$-modules in the crystalline site of $\mathcal X$.
Consider the following categories:
\begin{center}
\begin{table}[h]
\begin{tabular}{|l|l|l|l|}
\hline
Category & Notation & Name       & Reference   \\
\hline

$\Isoc(\mathcal X|W)$     & $\Isoc(\mathcal X)$  & Isocrystals  & \cite[Section 2]{Notmor}        \\
\hline

$\oi(\mathcal X|K)$   & $\oi(\mathcal X)$  & Overconvergent Isocrystals & \cite[Definition 2.3.6]{berthelot}       \\
\hline

$\oi(\mathcal X,\mathcal X|K)$   & $\oi(\mathcal X,\mathcal X)$ & Convergent isocrystals & \cite[Definition 2.3.2]{berthelot}      \\
\hline 
\end{tabular}
\end{table}
\end{center}
and their enriched version with Frobenius structure: $\Fisoc(X)$, $\Foi(\mathcal X)$ and $\Foi(\mathcal X,\mathcal X)$.
They fit into the following commutative diagram:
\begin{center}
\begin{tikzcd}
& \Fisoc(\mathcal X)\arrow[r,"(1)","\simeq "']\arrow[swap]{d}{(-)^{geo}}& \Foi(\mathcal X,\mathcal X)& \arrow[bend right,swap]{ll}{(3)}\arrow[swap]{l}{(2)}\arrow{ddl}{(4)} \Foi(\mathcal X)\arrow{d}{(-)^{geo}}\\
& \Isoc(\mathcal X)\arrow[swap]{rd}{(5)} & & \oi(\mathcal X)\\
& & Mod(\mathcal X|W)\otimes \Q 
\end{tikzcd}
\end{center}
where $(-)^{geo}$ are the forgetful functors, $(1)$ is the equivalence of categories constructed in \cite[Theoreme 2.4.2]{berthelot} and $(2)$ is the obvious functor.
Write $$R^i\mathfrak f_{crys,*}\mathcal O_{\mathcal Y/K}:=R^i\mathfrak f_{crys,*}\mathcal O_{\mathcal Y/W}\otimes \Q\in Mod(\mathcal X|W)\otimes \Q $$
and recall the following facts:
\begin{fact}\cite[Theorem 1.1]{Kedfull}\label{kedlaya}
The functor $(3)$ is fully faithful.
\end{fact}
\begin{fact}\label{Shiho}
Let $\mathfrak f:\mathcal Y\rightarrow X$ be a smooth proper morphism. Then $R^i\mathfrak f_{crys,*}\mathcal O_{\mathcal Y/K}\in Mod(\mathcal X|W)\otimes \Q$ lies in the essential image of $(4)$.
\end{fact}
\proof
By \cite[Proposition 3.2]{Notmor}, $R^i\mathfrak f_{crys,*}\mathcal O_{\mathcal Y/K}$ is in the essential image of $(5)\circ (-)^{geo}$. Under the equivalence $(1)$, $R^i\mathfrak f_{crys,*}\mathcal O_{\mathcal Y/K}$ is sent to the Ogus higher direct image $R^if_{Ogus,*}\mathcal O_{Y/K}$, see \cite[Section 3, Theorem 3.1]{Ogus} and \cite[Corollary 6.2]{Notmor}.
One concludes by Theorem \ref{berthelotconjecture}, which says that $R^if_{Ogus,*}\mathcal O_{Y/K}$ is in the image of an F-overconvergent isocrystal.
\endproof
Write $R^i\mathfrak f_{*}\mathcal O_{\mathcal Y|K}^{\dagger}$ for the (unique up to isomorphism) object of $\Foi(\mathcal X)$ lifting $R^i\mathfrak f_{crys,*}\mathcal O_{\mathcal Y/K}$.
\numberwithin{theorem}{subsubsection} 
\subsubsection{Tannakian formalism}

Since $\mathcal X$ is a geometrically connected $\F_q$ variety, $\Foi(\mathcal X)$ and $\oi(\mathcal X)$ are $K$-linear Tannakian categories over $K$\footnote{Recall that $F$ is the $s$-power Frobenius, so that its action on $\oi(\mathcal X)$ is $K$-linear.}, see \cite[Section 3.2]{marcuzzo} for more details. 
If one furthermore assumes that $\calX$ has a $\F_q$-rational point $\mathfrak t$, the categories $\Foi(\mathcal X)$, $\oi(\mathcal X)$ are neutralized by the fibre functors
\begin{center}
\begin{tikzcd}
\Foi(\mathcal X)\arrow{r}{(-)^{geo}}\arrow[swap]{dr}{\mathfrak t^*}& \oi(\mathcal X)\arrow{d}{\mathfrak t^*}\\
& \oi(\F_q)\simeq Vect_{K}.
\end{tikzcd}
\end{center}
Write $\pi_1^{\dagger}(\mathcal X,\mathfrak t)$ and $\pi_1^{\dagger,geo}(\mathcal X,\mathfrak t)$ for the Tannakian groups of $\Foi(\mathcal X)$ and $\oi(\mathcal X)$ respectively. For $\mathcal F\in \Foi(\mathcal X)$, let $G(\mathcal F,\mathfrak t)$, $G^{geo}(\mathcal F,\mathfrak t)$ denote the Tannakian groups of $\langle\mathcal F\rangle^{\otimes}\numberwithin{theorem}{subsection} \subseteq \Foi(\mathcal X)$ and $\langle\mathcal F^{geo}\rangle^{\otimes}\subseteq \oi(\mathcal X)$ respectively. By the general Tannakian formalism the forgetful functor $(-)^{geo}:\langle\mathcal F\rangle^{\otimes}\rightarrow \langle\mathcal F^{geo}\rangle^{\otimes}$ corresponds to a closed immersion 
$$G^{geo}(\mathcal F,\mathfrak t)\numberwithin{theorem}{subsection} \subseteq G(\mathcal F,\mathfrak t).$$
Alternatively $G^{geo}(\mathcal F,\mathfrak t)$, $G(\mathcal F,\mathfrak t)$ can be described as the images of 
$$\pi_1^{\dagger,geo}(\mathcal X,\mathfrak t)\rightarrow \pi_1^{\dagger}(\mathcal X,\mathfrak t)\rightarrow GL(\mathcal F_{\mathfrak t})$$
where $\mathcal F_{\mathfrak t}:=\mathfrak t^*\mathcal F$.
If $\mathfrak g: \mathcal Z\rightarrow \mathcal X$ is a morphism of geometrically connected $\F_q$-varieties and $\mathfrak t\in \mathcal Z(\mathbb F_q)$ the canonical functors
\begin{center}
\begin{tikzcd}
\langle\mathcal F\rangle^{\otimes}\arrow{r}{g^*}\arrow[swap]{d}{(-)^{geo}}& \langle\mathfrak g^*\mathcal F\rangle^{\otimes}\arrow{d}{(-)^{geo}}\\
\langle\mathcal F^{geo}\rangle^{\otimes}\arrow{r}{g^*} & \langle\mathfrak g^*\mathcal F^{geo}\rangle
\end{tikzcd}
\end{center}
correspond to a commutative diagram of closed immersions
\begin{center}
\begin{tikzcd}
G^{geo}(\mathfrak g^*\mathcal F,\mathfrak t)\arrow[hook]{r}\arrow[hook]{d}& G^{geo}(\mathcal F,\mathfrak t)\arrow[hook]{d}\\
G(\mathfrak g^*\mathcal F,\mathfrak t)\arrow[hook]{r}& G(\mathcal F,\mathfrak t).
\end{tikzcd}
\end{center}
Let $\mathfrak f:\mathcal Y\rightarrow \mathcal X$ be a smooth proper morphism of $\F_q$-varieties and write $\mathcal F_p:=R^i\mathfrak f_{*}\mathcal O_{\mathcal Y|K}^{\dagger}(j)$ for the $j^{th}$-twist of the F-overconvergent isocrystals provided by Fact \ref{Shiho}.
\numberwithin{theorem}{subsubsection} 
\begin{proposition}\label{globalsectionareinvariants}
There is a natural isomorphism
$$H^0(\mathcal X, R^{i}\mathfrak f_{crys,*}\mathcal O_{\mathcal Y/K})^{F=q^j}\simeq \mathcal F_{p,\mathfrak t}^{G(\mathcal F,\mathfrak t)}.$$
\end{proposition}
\proof 
Write $\mathcal O_{\mathcal X/K}$ and $\mathcal O^{\dagger}_{\mathcal X/K}$ for the structural sheaves in $\Fisoc(\mathcal X)$ and $\Foi(\mathcal X)$ respectively. 
One has $$H^0(\mathcal X, R^{i}\mathfrak f_{crys,*}\mathcal O_{\mathcal Y/K})\simeq Hom_{\Isoc(\mathcal X)}(\mathcal O_{\mathcal X/K}, R^{i}\mathfrak f_{crys,*}\mathcal O_{\mathcal Y/K})$$ and so an isomorphism 
$$H^0(\mathcal X, R^{i}\mathfrak f_{crys,*}\mathcal O_{\mathcal Y/K})^{F=q^j}\simeq Hom_{\Fisoc(\mathcal X)}(\mathcal O_{\mathcal X/K}(-j), R^{i}\mathfrak f_{crys,*}\mathcal O_{\mathcal Y/K}).$$
By Fact \ref{kedlaya} 
$$Hom_{\Fisoc(\mathcal X)}(\mathcal O_{\mathcal X/K}(-j), R^{i}\mathfrak f_{crys,*}\mathcal O_{\mathcal Y/K})\simeq Hom_{\Foi(\mathcal X)}(\mathcal O^{\dagger}_{\mathcal X/K}(-j), R^{i}\mathfrak f_{*}\mathcal O^{\dagger}_{\mathcal Y/K})\simeq$$ 
$$\simeq Hom_{\Foi(\mathcal X)}(\mathcal O^{\dagger}_{\mathcal X/K}, R^{i}\mathfrak f_{*}\mathcal O^{\dagger}_{\mathcal Y/K}(j))\simeq Hom_{\Foi(\mathcal X)}(\mathcal O^{\dagger}_{\mathcal X/K}, \calF_{p}).$$
Since $\Foi(\mathcal X)$ is a Tannakian category with fibre functor $\mathfrak t^*$ and $\mathfrak t^*\mathcal O^{\dagger}_{\mathcal X/K}$ corresponds to the trivial one dimensional representation $K$ of $\pi^{\dagger}_{1}(\mathcal X,\mathfrak t)$,  one deduces
$$Hom_{\Foi(\mathcal X)}(\mathcal O^{\dagger}_{\mathcal X/K}, \calF_{p})\simeq Hom_{Rep_K(\pi^{\dagger}_1(\mathcal X,\mathfrak t))}(\mathfrak t^*\mathcal O^{\dagger}_{\mathcal X/K}, \mathfrak t^*\calF_{p})\simeq Hom_{Rep_K(\pi^{\dagger}_1(\mathcal X,\mathfrak t))}(K, \calF_{p,\mathfrak t})\simeq\calF_{p,\mathfrak t}^{\pi^{\dagger}_1(\mathcal X,\mathfrak t)}.$$
Since the image of the action of $\pi^{\dagger}_1(\mathcal X,\mathfrak t)$ on $\calF_{p,\mathfrak t}$ is $G_{p}(\mathcal F,\mathfrak t)$, one sees that 
$$\calF_{p,\mathfrak t}^{\pi^{\dagger}_1(\mathcal X,\mathfrak t)}\simeq\calF_{p,\mathfrak t}^{G_{p}(\mathcal F,\mathfrak t)}$$
and this concludes the proof.
\endproof
\numberwithin{theorem}{subsubsection}
\subsubsection{Tannakian reinterpretation of Theorem \texorpdfstring{\ref{ltocrys}}-}
We now retain the notation and assumption of Theorem \ref{ltocrys}.
With the notation of Fact \ref{Shiho}, write
$$\mathcal F_p:=R^i\mathfrak f_*\mathcal O^{\dagger}_{\mathcal Y/K}(j)$$
$$\calF_{\mathcal Z, p}:=R^i\mathfrak f_{\mathcal Z,*}\mathcal O^{\dagger}_{\mathcal Y_{\mathcal Z}|K}(j)\simeq \mathfrak g^*\calF_{p}$$
where the isomorphism comes from smooth proper base change in crystalline cohomology (e.g. \cite[Proposition 3.2]{Notmor}) and Fact \ref{kedlaya}.
By Proposition \ref{globalsectionareinvariants} one has a commutative diagram
\begin{center}
\begin{tikzcd}
H^0(\mathcal X,R^i\mathfrak f_{crys,*}\mathcal O_{\mathcal Y/K})^{F=q}\arrow{d}\arrow{r}{\simeq}\arrow{r}{\simeq} & \calF_{p,\mathfrak t}^{G(\calF_{p},\mathfrak t)}\arrow[hook]{d}\\

H^0(\mathcal Z,R^i\mathfrak f_{\mathcal Z,crys,*}\mathcal O_{\mathcal Y_{\mathcal Z}/K})^{F=q}\arrow{r}{\simeq} & \calF_{p,\mathfrak t}^{G(\calF_{\mathcal Z, p},\mathfrak t)}
\end{tikzcd}
\end{center}
Hence it is enough to show that the natural inclusion $G(\calF_{\mathcal Z, p},\mathfrak t)\numberwithin{theorem}{subsection} \subseteq G(\mathcal F_p,\mathfrak t)$
is an isomorphism. 

\numberwithin{theorem}{subsection} 
\subsection{End of the proof}\label{conclusion}
\subsubsection{Compatibility}
For $\ell\neq p$ write 
$$\mathcal F_{\ell}:=R^i\mathfrak f_*\mathcal \Ql(j) \quad \calF_{\mathcal Z, \ell}:=R^i\mathfrak f_{\mathcal Z,*}\Ql(j)$$
and let $G(\calF_{\mathcal Z,\ell},\mathfrak t)$ (resp. $G(\mathcal F_{\ell},\mathfrak t)$) denotes the Zariski closure of the image of $\pi_1(\mathcal Z,\mathfrak t)$ (resp. $\pi_1(\mathcal X,\mathfrak t)$) acting on $H^{i}(\mathcal Y_{\overline t},\Ql(j))$. Since $\pi_1(\mathcal Z,\overline{\mathfrak t})\rightarrow \pi_1(\mathcal X,\overline{\mathfrak t})\rightarrow GL(H^{i}(\mathcal Y_{\overline{\mathfrak t}},\Ql(j))$ is open in the image of $\pi_1(\mathcal X,\overline{\mathfrak t})\rightarrow GL(H^{i}(\mathcal Y_{\overline{\mathfrak t}},\Ql(j))$ and $G(\mathcal F_{\ell},\mathfrak t)$ is connected, one has   $G(\mathcal F_{\ell},\mathfrak t)=G(\calF_{\mathcal Z,\ell},\mathfrak t)$.
To prove that the natural inclusion 
$G(\calF_{\mathcal Z, p},\mathfrak t)\numberwithin{theorem}{subsection} \subseteq G(\mathcal F_p,\mathfrak t)$ is an isomorphism, 
the idea is to compare $G(\calF_{\mathcal Z, p},\mathfrak t)$ and $G(\mathcal F_p,\mathfrak t)$ with $G(\calF_{\mathcal Z,\ell},\mathfrak t)$ and $G(\mathcal F_{\ell},\mathfrak t)$. To do this, the main ingredient is the following. 
\begin{fact}\label{compatible}\cite{Weil2}, \cite{KatzMessing}
$\calF_{p}$,$\calF_{\ell}$ (resp. $\calF_{\mathcal Z,p}$,$\calF_{\mathcal Z,\ell}$) is a $\Q$-rational compatible system on $\mathcal X$ (resp. $\mathcal Z$)  pure of weight $i+2j$.
\end{fact}
\numberwithin{theorem}{subsubsection}
\subsubsection{Geometric monodromy}

Write $\calF_{\ell}^{geo}$ (resp. $\calF_{\mathcal Z,\ell}^{geo}$) for the restriction of $\calF_{\ell}$ (resp. $\calF_{\mathcal Z,\ell}$) to $\mathcal X_{\F}$ (resp. $\mathcal Z_{\F}$) and $G^{geo}(\calF_{\ell},\mathfrak t)$ (resp. $G^{geo}(\mathcal F_{\mathcal Z,\ell},\mathfrak t)$) for the Zariski closure of the image of $\pi_1(\mathcal X_{\mathbb F},\mathfrak t)$ (resp. $\pi_1(\mathcal Z_{\mathbb F},\mathfrak t)$) acting on $H^{i}(\mathcal Y_{\overline t},\Ql(j))$. 
Recall the following:
\begin{fact}\label{reductiontogeometric}
For $?\in \{\ell,p\}$ one has:
\begin{enumerate}
\item The groups $G^{geo}(\calF_{?},\mathfrak t)$ and $G^{geo}(\calF_{\mathcal Z,?},\mathfrak t)$ are semisimple algebraic groups.
\item $G(\calF_{?},\mathfrak t)^0=G(\calF_{\mathcal Z,?},\mathfrak t)^0$ if and only if $G^{geo}(\calF_{?},\mathfrak t)^0=G^{geo}(\calF_{\mathcal Z,?},\mathfrak t)^0$
\end{enumerate}
\end{fact}
\proof
\begin{enumerate}
\item By \cite[Theorem 3.4.1]{Weil2} if $?=\ell$ and \cite[Remark 10.6]{notesked} if $?=p$, $\mathcal F^{geo}_{?}$ and $\mathcal F_{\mathcal Z,?}^{geo}$ are semisimple. Then the assertion follows from the global monodromy theorem (\cite[Corollarie 1.3.9]{Weil2} if $?=\ell$ and the proof of \cite[Corollary 4.10]{Crew}) if $?=p$. See \cite[Theorem 3.5.3, Corollary 3.5.8]{marcuzzo} for more details. 
\item One implication follows from the fact that $G^{geo}(\calF_{?},\mathfrak t)^0$ and $G^{geo}(\calF_{\mathcal Z,?},\mathfrak t)^0$ are the derived subgroup of $G(\calF_{?},\mathfrak t)^0$ and $G(\calF_{\mathcal Z,?},\mathfrak t)^0$ (see e.g. \cite[Corollary 3.5.9]{marcuzzo}). The other one is \cite[Lemma 4.2.14]{marcuzzo}.
\end{enumerate}
\endproof
By assumption $G(\mathcal F_{\ell},\mathfrak t)=G(\mathcal F_{\mathcal Z,\ell},\mathfrak t)$ and $G^{geo}(\mathcal F_{\ell},\mathfrak t)$ are connected, so by Fact \ref{reductiontogeometric} for $\ell\neq p$, one gets that $G^{geo}(\mathcal F_{\ell},\mathfrak t)=G^{geo}(\mathcal F_{\mathcal Z,\ell},\mathfrak t)$ is connected. By \cite[Theorem 4.1.1.]{marcuzzo}, being connected is independent from $?\in\{\ell,p\}$, so that $G(\mathcal F_{p},\mathfrak t),G(\mathcal F_{\mathcal Z,p},\mathfrak t),G^{geo}(\mathcal F_{p},\mathfrak t)$ and $G^{geo}(\mathcal F_{\mathcal Z,p},\mathfrak t)$ are connected. 
So, by Fact \ref{reductiontogeometric} for $\ell=p$, it is enough to show that $G^{geo}(\mathcal F_{\mathcal Z,p},\mathfrak t)$ and $G^{geo}(\mathcal F_{p},\mathfrak t)$ have the same dimension.
\numberwithin{theorem}{subsubsection}
\subsubsection{Purity}
Since $G^{geo}(\mathcal F_{\ell},\mathfrak t)=G^{geo}(\mathcal F_{\mathcal Z,\ell},\mathfrak t)$, it is enough to prove the following:
\begin{proposition}
$Dim(G^{geo}(\mathcal F_{\ell},\mathfrak t))=Dim(G^{geo}(\mathcal F_{p},\mathfrak t))$ and $Dim(G^{geo}(\mathcal F_{\mathcal Z,\ell},\mathfrak t))=Dim(G^{geo}(\mathcal F_{\mathcal Z,p},\mathfrak t))$
\end{proposition}
\proof
We prove the first equality. The second is analogous.
Write $r$ for the dimension of $\calF_{p,\mathfrak t}$ and $K_{?}=K$ if $?=p$ and $K_{?}=\Ql$ if $?=\ell$. Chose embeddings of $K_{?}$ in $\mathbb C$. Via these embeddings there are isomorphisms $\calF_{?,\mathfrak t}\otimes \mathbb C\simeq \mathbb C^{r}$ hence closed immersions of algebraic groups
$$i_{?}:G^{geo}(\mathcal F_{\mathcal Z,?},\mathfrak t)_{\mathbb C}:=G^{geo}(\mathcal F_{\mathcal Z,?},\mathfrak t)\times \mathbb C\rightarrow GL_{r,\mathbb C}.$$
Consider the functions
$$\phi_{?}:Rep_{\mathbb C}(GL_{r,\mathbb C})\to \N$$
$$ V\mapsto Dim_{\C}(V^{G^{geo}(\mathcal F_{\mathcal Z,?},\mathfrak t)_{\mathbb C}})$$
By Fact \ref{reductiontogeometric} and \cite{larsenepink} it is enough to show that $\phi_{?}$ does not depend on $?$. Since $GL_{n}$ is reductive, it is enough to check that the restriction of $\phi_{?}$ to irreducible representations of $GL_{r,\mathbb C}$ does not depend on $?$. 
Recall that every $GL_{r,\mathbb C}$-representation $V$ is defined over $\Q$ and that $V$ gives $\calF^{V}_{\ell}\in Rep_{\Ql}(\pi_1(\mathcal X))$ and $\calF^{V}_{p}\in \Foi(\mathcal X)$ via $i_{?}$. So let $V$ be an irreducible representation of $GL_{r,\C}$ and observe that $\calF^V_{\ell}$ and $\calF^V_{p}$ are again pure (of some weight $w$) and compatible (see e.g. \cite[Remark 4.1.3]{marcuzzo}). 
Then, as in the proof of Proposition \ref{globalsectionareinvariants}, one has
$$(\calF^V_{\ell,\mathfrak t})^{G^{geo}(\mathcal F_{\ell},\mathfrak t)
}=H^0(\mathcal X_{\F},(\calF^V_{\ell})^{geo})\quad \text{and}\quad (\calF^V_{p,\mathfrak t})^{G^{geo}(\mathcal F_{p},\mathfrak t)
}=H^0(\mathcal X,(\calF^V_{p})^{geo}).$$
So it is enough to show that
$$Dim(H^0(\mathcal X_{\F},(\calF^V_{\ell})^{geo}))=Dim(H^0(\mathcal X,(\calF^V_{p})^{geo})).$$
By Grothendieck-Lefschetz fixed point formula (\cite[Theorem 10.5.1, page 603]{Leifu} if $?=\ell$ and \cite[Theorem 6.3]{Trace} if $?=p$) and the purity of $\calF^V_{?},$ the left and the right hand sides are the number of poles, counted with multiplicity, with absolute value $q^{w/2}$ in the L-function of $(\calF^V_{\ell})^\vee(d)$ and $(\calF^V_{p})^\vee(d)$  (see e.g. \cite[Proposition 3.5.7]{marcuzzo} for more details), where $d$ is the dimension of $\mathcal X$. Since $\calF^V_{\ell}$ and $\calF^V_{p}$ are pure and compatible, the same is true for $(\calF^V_{\ell})^\vee(d)$ and $(\calF^V_{p})^\vee(d)$ and hence the $L$-function of $(\calF^V_{?})^\vee(d)$ does not depend on $?\in \{\ell,p\}$. This concludes the proof.
\endproof
\part{Applications}\label{secapp}
In this part $k$ is an infinite field of characteristic $p>0$ assumed to be finitely generated except in Subsection \ref{ptate}.
\section{Hyperplane sections}\label{sechyperplane}
In this section we apply Theorem \ref{main} to Lefschetz pencils of hyperplane sections. The main result is Corollary \ref{exinstencehyperplane}.
\numberwithin{theorem}{subsection}
 \subsection{Geometric versus arithmetic hyperplane sections.}
Let $Z$ be a smooth projective $k$-variety and fix a closed embedding $Z \subseteq \mathbb P_k^n$. One can ask whether there exists a smooth hyperplane section $D$ of $Z$ such that the canonical map 
$$i_{D_{\overline k}}:=NS(Z_{\overline k})\otimes \Q\rightarrow NS(D_{\overline k})\otimes \Q$$
is an isomorphism. If $Dim(Z)=2$, then $D$ is a curve so that $NS(D_{\overline k})\otimes \Q=\Q$ hence $i_{D_{\overline k}}$ is not  injective as soon as $NS(Z_{\overline k})\otimes \Q$ has rank $\geq 2$. Weak Lefschetz (\cite[Thm. 7.1, p. 318]{milne}) and  Grothendieck–Lefschetz (\cite[Exp. XI]{SGA2}) theorems ensure that $i_{D_{\overline k}}$ is injective if $Dim(Z)\geq 3$, and an isomorphism if $Dim(Z)\geq 4$. There are smooth projective varieties of dimension $3$ such that the surjectivity of $i_{D_{\overline k}}$ fails for all smooth hyperplane sections. 
\begin{example}\label{counter}
Take $Z=\mathbb P_k^3$ embedded in $\mathbb P_k^9$ via the Veronese embedding:

\begin{align*}
 \mathbb P_k^3&\rightarrow \mathbb P_k^9 \\
 [x:y:z:w]&\mapsto [x^2:y^2:z^2:w^2:xy:xz:xw:yz:yw:zw].
\end{align*}
Then a smooth hyperplane section $D\subseteq Z$ in $\mathbb P_k^9$ is a smooth quadric surface in $\mathbb P_k^3$, so that $D_{\overline k}\simeq \mathbb P^1_{\overline k}\times \mathbb P^1_{\overline k}$. Hence $NS(D_{\overline k})\simeq \Z\times \Z$, while $NS(Z_{\overline k})=\Z$.
\end{example}
But things change if one replaces the geometric N\'eron-Severi groups with their arithmetic counterparts.
\begin{example}\label{example}
Assume $p\geq 17$ and consider the pencil of hyperplane sections of $Z$ in $\mathbb P_k^9$ given by the hyperplanes $a(x_1+x_2+x_3+x_4)+b(x_1+4x_2+9x_3+16x_4)$, where $[a:b]\in \mathbb P^1(k)$ and $x_1,...,x_{10}$ are the coordinates in $\mathbb P_k^9$. This corresponds in $\mathbb P^3$ to the pencil of quadric surfaces
$$Q_{[a,b]}:a(x^2+y^2+z^2+w^2)+b(x^2+4y^2+9z^2+16w^2)=0.$$
When $Q_{[a:b]}$ is smooth, it a is quadric surface and, for a $[a:b]$ in an open subset of $\mathbb P_k^1$ $$NS(D)\otimes \Q\simeq (NS(D_{\overline k})\otimes \Q)^{\pi_1(k)}=(\Q^2)^{\pi_1(k)}=\Q$$
is generated by the sum of the two families of $\mathbb P^1_{\overline k}$. So there are ``lots" of $[a:b]\in \mathbb P^1(k)$ such that the canonical map $$NS(\mathbb P_k^3)\otimes \Q\rightarrow NS(Q_{[a:b]})\otimes \Q$$
is an isomorphism. 
\end{example}
The main result is of this subsection is Corollary \ref{exinstencehyperplane} that we now recall:
\begin{corollarynoname}
\textit{If} $dim(Z)>2$ \textit{there are infinitely many} $k$\textit{-rational hyperplane sections} $D$ \textit{such that
the canonical map}
$$NS(Z)\otimes \Q\rightarrow NS(D)\otimes \Q$$ \textit{is an isomorphism.}
\end{corollarynoname}
\numberwithin{theorem}{subsection} 
\subsection{Proof of Corollary \texorpdfstring{\ref{exinstencehyperplane}}- }\label{ammazzachemazza}
By (\cite[Exp. XVII]{SGA7}) there exists a pencil of hyperplanes $L:=\{H_x\}_{x\in \check{\mathbb P}_k^1}$ such that:
\begin{itemize}
\item For all $x$ in an dense open subscheme $U\subseteq \check{\mathbb P}_k^1$, the intersection $H_x\cap Z$ is smooth;
\item The base locus $B:=\cap_{x\in L} Z\cap H_{x}\subseteq Z$ is smooth.
\end{itemize}
Then one gets a diagram
\begin{center}
\begin{tikzcd}
Z & \arrow[swap]{l}{\pi}\arrow{r}{f}\widetilde Z & \check{\mathbb P}_k^1
\end{tikzcd}
\end{center}
where $\pi:\widetilde Z\rightarrow Z$ is the blow up of $Z$ along $B$, $f:\widetilde Z\rightarrow \check{\mathbb P}_k^1$ is a projective flat morphism, smooth over $U$ and for each $x\in \check{\mathbb P}_k^1$ the fibre $\widetilde Z_x$ of $f:\widetilde Z\rightarrow \check{\mathbb P}_k^1$ at $x$ identifies via $\pi:\widetilde Z\rightarrow Z$ with the hyperplane section $Z\cap H_x\subseteq Z$. Write $E:=\pi^{-1}(B)$ for the exceptional divisor.
Explicitly $\widetilde Z$ is the closed subscheme of $Z\times \check{\mathbb P}_k^1$ defined by
$$\widetilde Z:=\{(z,x)\in Z\times \check{\mathbb P}^1_k \text{ with } z\in Z\cap H_x\}\hookrightarrow Z\times \check{\mathbb P}^1_k$$
$\pi:\widetilde Z\rightarrow Z$, $f:\widetilde Z\rightarrow \check{\mathbb P}^1_k$ identify with the canonical projections onto $Z$, $\check{\mathbb P}^1_k$ respectively and $E$ with $B\times \check{\mathbb P}_k^1$. Write $\eta$ for the generic point of $\check{\mathbb P}^1_k$.
Combing Lemma \ref{hyperplane} below with Corollary \ref{existence} one gets Corollary \ref{exinstencehyperplane}.
\begin{lemma}\label{hyperplane}
If $Dim(Z)>2$, the canonical map 
$$i^*_{\widetilde Z_{\eta}}:NS(Z)\otimes \Q\rightarrow NS(\widetilde Z_{\eta})\otimes \Q$$ is an isomorphism.
\end{lemma}
\proof
This is inspired from \cite[Corollary 1.5]{Varmor}. 
Fix $x\in U$. The natural commutative diagram 
\begin{center}
\begin{tikzcd}
& \widetilde Z_{\eta}\arrow{r}\arrow{d}\arrow[swap]{dl}{i_{\widetilde Z_{\eta}}}\arrow[phantom]{rd}{\Box}\arrow[near start, bend left]{dd}{i_{\eta}} & k(\eta)\arrow{d}{\eta}\arrow[bend left]{dd}\\
Z& \widetilde Z\arrow[swap]{l}{\pi}\arrow{r}{f}\arrow[phantom]{rd} & \check{\mathbb P}_k^1\\
&  \widetilde Z_U\arrow{u}{i_U}\arrow{r}{f_U}\arrow[phantom]{rd}{\Box} & U\arrow{u}\\
& \arrow{uul}{i_{\widetilde Z_x}}\arrow{u}{i_x}\widetilde Z_{x}=Z\cap H_x\arrow{r} &k(x)\arrow{u}{x}
\end{tikzcd}
\end{center}
induces a commutative diagram:
\begin{center}
\begin{tikzcd}
Pic(Z)\otimes \Q\arrow{r}{\pi^*}\arrow[two heads]{d}{c_Z} & Pic(\widetilde Z)\otimes \Q\arrow[two heads]{r}{i^*_U}\arrow[two heads]{d}{c_{\widetilde Z}} & Pic(\widetilde Z_{U})\otimes \Q\arrow[two heads, bend left=15]{rr}{i^*_{\eta}}\arrow{r}{i^*_x} & Pic(\widetilde Z_x)\otimes \Q\arrow[two heads]{d}{c_{\widetilde Z_x}}&Pic(\widetilde Z_{\eta})\otimes \Q\arrow[two heads]{d}{c_{\widetilde Z_\eta}} \\
NS(Z)\otimes \Q\arrow{r}{\pi^*}\arrow[bend right=10,swap]{rrr}{i^*_{\widetilde Z_x}} & NS(\widetilde Z)\otimes \Q\arrow{rr} & & NS(\widetilde Z_{x})\otimes \Q & NS(\widetilde Z_{\eta})\otimes \Q \arrow[hook]{l}{sp^{ar}_{\eta,x}} 
\end{tikzcd}
\end{center}
Since $Dim(Z)>2$, the map 
$$i^*_{\widetilde Z_{\eta}}:NS(Z)\otimes \Q\rightarrow  NS(\widetilde Z_{\eta})\otimes \Q$$
is injective.
To prove the surjectivity, let $\epsilon\in NS(\widetilde Z_{\eta})\otimes \Q$ with lift $z\in Pic(\widetilde Z_{\eta})\otimes \Q$. Since $sp^{ar}_{\eta,x}$ is injective, it is enough to show that $\epsilon_x:=sp^{ar}_{\eta,x}(\epsilon)$ is in the image of $i^*_{\widetilde Z_x}$. 
Since the maps $i^*_U$ and $i^*_{\eta}$ are surjective, $z\in Pic(\widetilde Z_{\eta})$ lifts to a $\widetilde z\in Pic(\widetilde Z)\otimes \Q$ and, by the commutativity of the diagram, $\widetilde z$ maps to $\epsilon_x$ in $NS(\widetilde Z_{x})\otimes \Q$. Now, since $\pi$ is the blow up of $Z$ along $B$, $c_{\widetilde Z}(\widetilde z)$ can be written as $\pi^*c_{Z}(z')+bc_{\widetilde Z}(E)$, where $z'\in Pic(Z)\otimes \Q$ and $b\in \Q$. The conclusion follows from the following claim, since it implies that $\epsilon_x$ is the image of $c_Z(z')+bc_Z(Z\cap H_x)\in NS(Z)\otimes \Q$.\\
\textbf{Claim:} \textit{The restrictions of} $c_{\widetilde Z}(E)$ \textit{and} $\pi^*c_{Z}(Z\cap H_x)$ \textit{to} $\widetilde Z_x$ \textit{coincide.}\proof[Proof of the claim]
By direct computations, one sees that $E=B\times \check{\mathbb P}_k^1$ intersects transversally $\widetilde Z_{x}$ and that $E\cap \widetilde Z_x=B$, so that the restriction of $c_{\widetilde Z}(E)$ to $\widetilde Z_{x}$ is given by $c_{\widetilde Z_x}(B)\in Pic(\widetilde Z_x)\otimes \Q$. To compute the restriction $\pi^*c_{Z}(Z\cap H_x)$ to $\widetilde Z_x$ observe that it is equal to $i_{\widetilde Z_x}^*(c_{Z}(Z\cap H_x))$. Then take any $y\neq x\in L$ and compute $i_{\widetilde Z_x}^*(c_Z(Z\cap H_x))$ as $c_{\widetilde Z_x}(Z\cap H_x\cap H_y)=c_{\widetilde Z_x}(B)$, since $B=Z\cap H_x\cap H_y$ for any $x\neq y\in L$.
\endproof


\begin{remark}
The key fact that $Pic(\widetilde Z)\otimes \Q\rightarrow Pic(\widetilde Z_{\eta})\otimes \Q$ is surjective, does not hold for $Pic(\widetilde Z_{\overline k})\otimes \Q\rightarrow Pic(\widetilde Z_{\overline \eta})\otimes \Q$ (see Example \ref{example}).
This is why it is not true in general that for a point $x\in |U|$, which is Galois-generic for $\widetilde Z_{U}\rightarrow U$ the canonical map
$$i_{\widetilde Z_x}:NS(\widetilde Z_x)\otimes \Q\rightarrow NS(Z)\otimes \Q$$ is an isomorphism and one really needs to restrict to strictly Galois generic points: during the proof one cannot replace $U$ with a finite \etale cover, since any base change destroys the geometry of the pencil. 
\end{remark}
\subsection{Proof of Corollary \texorpdfstring{\ref{corollaryTate}}-}

Replacing $k$ with a finite extension, it is enough to show that the map
$$NS(Z)\otimes \Ql\rightarrow H^2(Z_{\overline k},\Ql(1))^{\pi_1(k)}$$ is an isomorphism. 
By Corollary \ref{exinstencehyperplane}, there exists $k$-rational hyperplane section $D\rightarrow Z$ such that the canonical map $NS(Z)\otimes \Ql\xrightarrow{\sim} NS(D)\otimes \Ql$ is an isomorphism. The conclusion follows from the commutative diagram
\begin{center}
\begin{tikzcd}
NS(Z)\otimes \Ql\arrow[hook]{r}\arrow{d}{i_{D}^*}& H^2(Z_{\overline k},\Ql(1))^{\pi_1(k)}\arrow[hook]{d}{i_{D}^*}\\
NS(D)\otimes \Ql\arrow[hook]{r}{(2)}& H^2(D_{\overline k},\Ql(1))^{\pi_1(k)}
\end{tikzcd}
\end{center}
since $i_{D}^*$ is an isomorphism by the choice of $D$ and $(2)$ is an isomorphism by $T(D,\ell)$.
\endproof
\section{Brauer groups in families}\label{secbrauer}
\numberwithin{theorem}{subsection} 
\subsection{Specialization of Brauer groups}\label{brauergen}
\subsubsection{Brauer group}
For a smooth proper $k$-variety $Z$ write $H^2(Z_{\overline k},\mathbb G_m):=Br(Z_{\overline k})$ for the (cohomological) Brauer group of $Z_{\overline k}$, $Br(Z_{\overline k})[n]$ for its $n$-torsion subgroup and
$$T_{\ell}(Br(Z_{\overline k})):=\varprojlim_{n}Br(Z_{\overline k})[\ell^n],\quad Br(Z_{\overline k})[\ell^{\infty}]:=\varinjlim_{n}Br(Z_{\overline k})[\ell^n],\quad Br(Z_{\overline k})[p']:=\varinjlim_{n\nmid p}Br(Z_{\overline k})[n].$$

Recall that $Br(Z_{\overline k})$ is a torsion group and that Kummer theory induces, for every $p\nmid n\in \N$, an exact sequence:
$$0\rightarrow NS(Z_{\overline k})/n\rightarrow H^2(Z_{\overline k},\mu_{n})\rightarrow Br(Z_{\overline k})[n]\rightarrow 0.$$
It is classically known that if $T(Z,\ell)$ holds, then $Br(Z_{\overline k})[\ell^{\infty}]^{\pi_1(k)}$ is finite (see e.g. \cite[Proposition 2.1.1]{brauer}). 
\subsubsection{Brauer generic points}\label{brauergeneric}

Let $X$ be a smooth geometrically connected $k$-variety with generic point $\eta$ and $f:Y\rightarrow X$ a smooth proper morphism of $k$-varieties. Taking the direct limit over $p\nmid n$ on the Kummer exact sequence, one gets a commutative specialization exact diagram
\begin{center}
\begin{tikzcd}
0\arrow{r} & \varinjlim_{n\nmid p} NS(Y_{\overline \eta})/n\arrow[hook]{d}{sp_{\eta,x}}\arrow{r} & \varinjlim_{n\nmid p} H^2(Y_{\overline \eta},\mu_n)\arrow{r}\arrow{d}{\simeq}& Br(Y_{\overline \eta})[p']\arrow[two heads]{d}{{sp^{Br}_{\eta,x}}}\arrow{r} & 0\\
0\arrow{r} & \varinjlim_{n\nmid p} NS(Y_{\overline x})/n\arrow{r} & \varinjlim_{n\nmid p} H^2(Y_{\overline x},\mu_n)\arrow{r}& Br(Y_{\overline x})[p']\arrow{r} & 0
\end{tikzcd}
\end{center}
Since the group $Ker(sp_{\eta,x})$ is of $p$-torsion and $Coker(sp_{\eta,x})_{tors}=Coker(sp_{\eta,x})[p^{\infty}]$ (see \cite[Proposition 3.6]{poonen}), one sees that a $x\in |X|$ is $NS$-generic if and only if the map 
$$sp^{Br}_{\eta,x}:Br(Y_{\overline \eta})[p']\rightarrow Br(Y_{\overline x})[p']$$
is an isomorphism. In particular (Corollary \ref{existence}) the set of $x\in |X|$ such that $sp^{Br}_{\eta,x}$ is not an isomorphism is sparse and if $X$ is a curve it contains at most finitely many of $k$-rational points. 
\subsection{Uniform boundedness}
Retain the notation and the assumptions as in the previous Section \ref{brauergeneric}.
Taking inverse limit in the Kummer exact sequence, one gets a commutative exact diagram
\begin{center}
\begin{tikzcd}
0\arrow{r} & NS(Y_{\overline \eta})\otimes \Z_{\ell} \arrow{r}\arrow[hook]{d}{sp_{\eta,x}} & H^2(Y_{\overline \eta},\Zl(1))\arrow{r}\arrow{d}{\simeq} &T_{\ell}(Br(Y_{\overline \eta})) \arrow{r}\arrow[two heads]{d}{sp^{Br}_{\eta,x}} & 0\\
0\arrow{r} & NS(Y_{\overline x})\otimes \Z_{\ell} \arrow{r} & H^2(Y_{\overline x},\Zl(1))\arrow{r} &T_{\ell}(Br(Y_{\overline x})) \arrow{r} & 0
\end{tikzcd}
\end{center}
The group $\pi_1(x,\overline x)$ acts on $T_{\ell}(Br(Y_{\overline \eta}))$ by restriction trough the map $\pi_1(x,\overline x)\rightarrow \pi_1(X,\overline x)\simeq \pi_1(X,\overline \eta)$ and $sp^{Br}_{\eta,x}$ is $\pi_1(x,\overline x)$-equivariant with respect to the natural action of $\pi_1(x,\overline x)$ on $T_{\ell}(Br(Y_{\overline x}))$. Hence the arguments in Section \ref{brauergen} combined with Theorem \ref{main} show the following
\begin{lemma}\label{lemmaidentification}
Up to replacing $X$ with an open subset, for every Galois generic $x\in |X|$ and every $\ell\neq p$, the $\pi_1(x,\overline x)$-equivariant specialization morphism
$$sp^{Br}_{\eta,x}:T_{\ell}(Br(Y_{\overline \eta}))\rightarrow T_{\ell}(Br(Y_{\overline x}))$$
is an isomorphism
\end{lemma}
Replacing \cite[Proposition 3.2.1]{brauer} with Lemma \ref{lemmaidentification} and \cite[Fact 3.4.1]{brauer} with the main result of \cite{UOIpositive}, one can make the arguments in the proof of \cite[Theorem 1.2.1]{brauer} work in positive characteristic and prove Corollary \ref{corollarybrauer}. 
In the same way, using the arguments in the proof of \cite[Theorem 3.5.1]{brauer}, one gets the following unconditional variant:
\begin{corollary}\label{uncondintionalbrauer}
Let $X$ be a curve and assume that the Zariski closure of the image of $\pi_1(X)$ acting on $H^2(Y_{\overline \eta},\Ql(1))$ is connected. Then there exists an integer $C:=C(Y\rightarrow X,\ell)$ 
such that $$[Br(Y_{\overline{x}})^{\pi_1(x,\overline x)}[\ell^\infty]:Br(Y_{\overline{\eta}})^{\pi_1(X,\overline \eta)}[\ell^\infty]]\leq C $$ 
for all but finitely many $x\in X(k)$.
\end{corollary}
\numberwithin{theorem}{subsection} 
\subsection{p-adic Tate module}\label{ptate}
Assume that $X$ is a smooth connected $k$-variety with generic point $\eta$, where $k$ is an algebraically closed field of characteristic $p$ and that $Y\rightarrow X$ is a smooth projective morphism. 
\begin{corollary}\label{corptate}
There exists an $x\in |X|$ such that 
$Rank(T_p(Br(Y_{\overline x})))=Rank(T_p(Br(Y_{\overline \eta})))$
\end{corollary}
\proof
For every geometric point $\overline t\in X$, one has (\cite[Proposition 5.12]{derhamwitt}):
$$Dim(NS(Y_{\overline t})\otimes \Q_p)=Dim(H^2_{crys}(Y_{\overline t}))-2Dim(H^2_{crys}(Y_{\overline t}))_{[1]})-Rank(T_p(Br(Y_{\overline t})))$$
where $H^2_{crys}(Y_{\overline t})_{[1]}$ is the slope one part of the crystalline cohomology of $Y_{\overline t}$ (see e.g. \cite[Section 3]{notesked} for the definition). By \cite[Theorem 3.12, Corollary 4.2]{notesked} there exists a dense open subset $U$ of $X$ such that for all $x\in |U|$ one has
$$Dim(H^2_{crys}(Y_{\overline x})_{[1]})=Dim(H^2_{crys}(Y_{\overline{\eta}})_{[1]})$$
Since $Dim(H^2_{crys}(Y_{\overline x}))$ is independent of $x\in X$ (smooth proper base change in crystalline cohomology), one concludes applying Corollary \ref{existence} to $Y_U\rightarrow U$.
\endproof
\newpage
\part{Overconvergence of \texorpdfstring{$R^if_{Ogus,*}\mathcal O_{Y/K}$}- (after A.Shiho)}\label{appendix}
In this part we use the work of Shiho on relative rigid cohomology to prove Theorem \ref{berthelotconjecture}, which is a key ingredient in the proof of Fact \ref{Shiho}. In Section \ref{secprel}, we recall the definitions of various categories of isocrystals, the relations between them and we state Theorem \ref{berthelotconjecture}. In Section \ref{secproof}, we prove Theorem \ref{berthelotconjecture}.
\section{Preliminaries}\label{secprel}
\numberwithin{theorem}{subsection} 
\subsection{Notation}
Let $k$ be a perfect field of characteristic $p>0$. Write $K$ for the fraction field of the Witt ring $W:=W(k)$ of $k$ and $|-|:K\rightarrow \mathbb R$ for the norm induced by the ideal $pW\numberwithin{theorem}{subsection} \subseteq W$. For any $k$-variety $X$, write $F_X$ for a power of the absolute Frobenius on $X$ and, if there is no danger of confusion, one often drops the lower index and writes just $F$.

Gothic letters ($\mathfrak T,\mathfrak X,\mathfrak U...$) denote separated, $p$-adic formal schemes over $W$.
Write $\mathfrak X_1$ for the special fibre of $\mathfrak X$, $\mathfrak X_K$ for its rigid analytic generic fibre and $sp:\mathfrak X_{K}\rightarrow \mathfrak X_1$ for the specialization map. There is an equivalence between the isogeny category $Coh(\mathfrak X)\otimes \Q$ of the category $Coh(\mathfrak X)$ of coherent sheaves on $\mathfrak X$ and the category $Coh(\mathfrak X_K)$ of coherent sheaves on $\mathfrak X_K$ (see \cite[Remark 1.5]{Ogus}). 


If $f:X\rightarrow \mathfrak X_1$ is a closed immersion, one can consider the open tube $]X[_{\mathfrak X}:=sp^{-1}(X)$ and the closed tube of radius $|p|$,  $[X]_{\mathfrak X,|p|}$ of $X$ in $\mathfrak X$ (see \cite[Definition 1.1.2, Section 1.1.8]{berthelot}). They are admissible open subsets of $\mathfrak X_K$ and there is an inclusion $[X]_{\mathfrak X,|p|}\subseteq ]X[_{\mathfrak X}$.


A couple $(X,\overline X)$ is an open immersion of $k$-varieties $X\rightarrow \overline X$ and a frame $(X,\overline X,\mathfrak X)$ is a couple $(X,\overline X)$ together with a closed immersion of $\overline X$ into a $p$-adic formal schemes $\mathfrak X$. Morphisms of couples and frames are defined in the obvious way. A couple $(Y,\overline Y)$ over a frame $(X,\overline X,\mathfrak X)$ is a morphism of couples $(Y,\overline Y)\rightarrow (X,\overline X)$ and a frame $(X,\overline X,\mathfrak X)$ over a couple $(Y,\overline Y)$ is a morphism of couples $(X,\overline X)\rightarrow (Y,\overline Y)$. If $(X,\overline X,\mathfrak X)$ is a frame, for any sheaf $\mathcal F$ over $]\overline X[_{\mathfrak X}$ one writes $$j^{\dagger}_{X}\calF:=\varinjlim_{V}j_{V*}j_{V}^{*}\mathcal F$$
where the limit runs over all the strict neighbourhoods $V$ of $X$ in $\overline X$ (see \cite[Definition 1.2.1]{berthelot}) and $j_V:V\rightarrow ]\overline X[_{\mathfrak X}$ in the inclusion map.




If $f:Y\rightarrow X$ is a morphism of $k$-varieties, for every morphism $Z\rightarrow X$ write:
\begin{center}
\begin{tikzcd}
Y_{Z}\arrow{r}\arrow{d}{f_Z}\arrow[phantom]{dr}{\Box}& Y\arrow{d}{f}\\
Z\arrow{r}& X
\end{tikzcd}
\end{center}
\numberwithin{theorem}{subsection} 
\subsection{Categories of isocrystals}
To a $k$-variety $X$ one can associate the following categories of isocrystals:
\begin{itemize}
\item $\Isoc^{(p)}(X)$, the $p$-adically convergent isocrystals (see \cite[Definition 2.1]{Ogus});
\item $\Isoc^{(1)}(X)$, the convergent isocrystals (see \cite[Definition 2.1]{Ogus});
\end{itemize}
If $(X,\overline X)$ is a couple there is a category $\Isoc^{\dagger}(X,\overline X)$ of isocrystal on $X$ overconvergent along $\overline X-X$ see \cite[Definition 2.3.2]{berthelot}. If $\overline X$ is a compactification of $X$, one writes $\Isoc^{\dagger}(X,\overline X):=\Isoc^{\dagger}(X)$ and calls the object in there overconvergent isocrystals on $X$. It is known that $\Isoc^{\dagger}(X)$ does not depend on the choice of the compactification, so that $\Isoc^{\dagger}(X)$ is well defined (see \cite[2.3.6]{berthelot}).
\begin{itemize}

\item $\Isoc^{(p)}(X)$ (resp. $\Isoc^{(1)}(X)$). Write $I^{(p)}_X$ (resp. $I^{(1)}_X$) for the category of $p$-adic enlargements (resp. enlargements). This is the category of pairs $(\mathfrak T,z_{\mathfrak T})$ such that $\mathfrak T$ is a flat $p$-adic formal $W$-scheme and $z_{\mathfrak T}$ is a morphism $\mathfrak T_1\rightarrow X$ (resp. $(\mathfrak T_1)_{red}\rightarrow X$). A morphism $g:(\mathfrak Z,z_{\mathfrak Z})\rightarrow (\mathfrak T,z_{\mathfrak T})$ between $p$-adic enlargements (resp. enlargements) is a morphism $g:\mathfrak Z\rightarrow \mathfrak T$ such that $z_{\mathfrak Z}\circ g_1= z_{\mathfrak T}$ (resp. $z_{\mathfrak Z}\circ (g_1)_{red}=z_{\mathfrak T}$), where $g_1:\mathfrak Z_1\rightarrow \mathfrak T_1$ (resp. $(g_1)_{red}:(\mathfrak Z_1)_{red}\rightarrow (\mathfrak T_1)_{red}$) are the natural morphisms induced by $g$.
A $p$-adically convergent isocrystals (resp. a convergent isocrystal) is the following set of data:
\begin{itemize}
\item For every $(\mathfrak T,z_{\mathfrak T})\in Ob(I^{(p)}_X)$ (resp. $\in Ob(I^{(1)}_X)$), a $\mathcal M_{(\mathfrak T,z_{\mathfrak T})}\in Coh(\mathfrak T_K)$;
\item For every morphism $g:(\mathfrak Z,z_{\mathfrak Z})\rightarrow (\mathfrak T,z_{\mathfrak T})$ in $I^{(p)}_X$ (resp. $I^{(1)}_X$) an isomorphism
$$\phi_g:g^*\mathcal M_{(\mathfrak T,z_{\mathfrak T})}\rightarrow \mathcal M_{(\mathfrak Z,z_{\mathfrak Z})}$$
in $Coh(\mathfrak Z_K)$ such that $\phi_{Id}=Id$ and for every other morphism $h:(\mathfrak T,z_{\mathfrak T})\rightarrow (\mathfrak U,z_{\mathfrak U})$ one has $g^*\phi_h=\phi_{h\circ g}$.
\end{itemize}
A morphism of $p$-adically convergent isocrystals (resp. convergent isocrystals) $\mathcal M\rightarrow \mathcal N$ is a collection of morphisms $\{\mathcal M_{(\mathfrak T,z_{\mathfrak T})}\rightarrow \mathcal N_{(\mathfrak T,z_{\mathfrak T})}\}_{(\mathfrak T,z_{\mathfrak T})\in Ob(I^{(p)}_X)}$ (resp. $_{(\mathfrak T,z_{\mathfrak T})\in Ob(I^{(1)}_X)}$) compatible with the isomorphism $\phi_g$ for all morphisms $g$.
\item $\Isoc^{\dagger}(X,\overline X)$. Write $I_{(X,\overline X)}$ for the category of frames over $(X,\overline X)$.
Then an isocrystals on $X$ overconvergent along $\overline X-X$ is the following set of data:
\begin{itemize}
\item For every $(T,\overline T,\mathfrak T)\in Ob(I_{(X,\overline X)})$ a coherent $j^{\dagger}_T\mathcal O_{]\overline T[_{\mathfrak T}}$ module $\mathcal M_{(T,\overline T,\mathfrak T)}$;
\item For every morphism $g:(Z,\overline Z,\mathfrak Z)\rightarrow (T,\overline T,\mathfrak T)$ in $I_{(X,\overline X)}$ an isomorphism
$$\phi_g:g^*\mathcal M_{(T,\overline T,\mathfrak T)}\rightarrow \mathcal M_{(Z,\overline Z,\mathfrak Z)}$$
of coherent $j^{\dagger}_{Z}\mathcal O_{]\overline Z[_{\mathfrak Z}}$ modules such that $\phi_{Id}=Id$ and for every other morphism $h:(T,\overline T,\mathfrak T)\rightarrow (U,\overline U,\mathfrak U)$ one has $g^*\phi_h=\phi_{h\circ g}$.
\end{itemize}
A morphism $\mathcal M\rightarrow \mathcal N$ in $\Isoc^{\dagger}(X,\overline X)$ is a collection of morphisms $\{\mathcal M_{(T,\overline T,\mathfrak T)}\rightarrow \mathcal N_{(T,\overline T,\mathfrak T)}\}_{(T,\overline T,\mathfrak T)\in Ob(I_{(X,\overline X)})}$ compatible with the isomorphism $\phi_g$ for all morphisms $g$.
\end{itemize}
There are also enriched versions of the previous categories with Frobenius structure, which we denote $\Fisoc^{(p)}(X)$, $\Fisoc^{(1)}(X)$ and $\Fisoc^{\dagger}(X,\overline X)$. For example, the absolute Frobenius $F_X$ induces an endofunctor $$F_X^*:\Isoc^{(p)}(X)\rightarrow \Isoc^{(p)}(X)$$ and $\Fisoc^{(p)}(X)$ is the category of pairs $(\mathcal M,\Phi)$, where $\mathcal M\in \Isoc^{(p)}(X)$ and $\Phi$ is a Frobenius structure  on $\mathcal M$, i.e. an isomorphism $F_X^*\mathcal M\rightarrow \mathcal M$. A morphism in $\Fisoc^{(p)}(X)$ is a morphism in $\Isoc^{(p)}(X)$ compatible with the Frobenius structures. The constructions of $\Fisoc^{(1)}(X)$ and $\Fisoc^{\dagger}(X,\overline X)$ are similar.
\numberwithin{theorem}{subsection} 
\subsection{Functors between the categories}\label{functors}
For every couple $(X,\overline X)$ there is a canonical commutative diagram of functors:
\begin{center}
\begin{tikzcd}
\Fisoc^{(p)}(X)\arrow{d}& \arrow{l}{F1-Fp} \Fisoc^{(1)}(X)\arrow{d} & \Fisoc^{\dagger}(X,X)\arrow{l}{Fconv-F1}\arrow{d}& \Fisoc^{\dagger}(X,\overline X)\arrow{d} \arrow{l}{Fov-Fconv}\\
\Isoc^{(p)}(X)& \arrow{l}{1-p} \Isoc^{(1)}(X)& \Isoc^{\dagger}(X,X)\arrow{l}{conv-1}& \Isoc^{\dagger}(X,\overline X) \arrow{l}{ov-conv}
\end{tikzcd}
\end{center}
and 
\begin{itemize}
\item $F1-Fp, conv-1, Fconv-F1$ are equivalences of categories (\cite[Proposition 2.18]{Ogus}, \cite[2.3.4]{berthelot}) ;
\item $Fov-Fconv$ is fully faithful if $X$ is smooth (\cite[Theorem 1.1]{Kedfull}).
\end{itemize}

All the functors are easy to construct from the definitions.
For example, to construct $conv-1$, to an enlargement $(\mathfrak T,z_{\mathfrak T})$ one associates the frame
$((\mathfrak T_1)_{red},(\mathfrak T_1)_{red},\mathfrak T)$ over $(X,X)$ and so, for every $\mathcal M\in \Foi(X,X)$, one defines
$$conv-1(\mathcal M)_{(\mathfrak T,z_{\mathfrak T})}:=\mathcal M_{((\mathfrak T_1)_{red},(\mathfrak T_1)_{red},\mathfrak T)}.$$
The constructions of $1-p,ov-conv$ are similar. In view of these functors, if $\mathcal M$ is in $\Fisoc^{\dagger}(X,\overline X)$ and $(\mathfrak T,z_{\mathfrak T})$ is a $p$-adic enlargement of $X$, write 
$$\mathcal M_{(\mathfrak T,z_{\mathfrak T})}:=\mathcal M_{(\mathfrak T_1,\mathfrak T_1,\mathfrak T)}$$
\numberwithin{theorem}{subsection}
\subsection{Stratification}
Assume that $X$ admits a closed immersion into a $p$-adic formal scheme $\mathfrak X$ formally smooth over $W$. Then the categories of isocrystals on $X$ admit a more concrete description in term of modules with a stratification. We now recall the notion of universal $p$-adic enlargement and we use it to define modules with a stratification.
\subsubsection{Universal p-adic enlargements}
By \cite[Proposition 2.3]{Ogus}, there exists a universal $p$-adic enlargement $(\mathfrak T(X), z_{\mathfrak T(X)})$ of $X$ in $\mathfrak X$. The $p$-adic enlargement $(\mathfrak T(X), z_{\mathfrak T(X)})$ of $X$ is endowed with a map $g:\mathfrak T(X)\rightarrow \mathfrak X$ making the following diagram commutative:
\begin{center}
\begin{tikzcd}
& \mathfrak T(X)_1\arrow[hook]{r}\arrow[swap]{dl}{z_{\mathfrak T(X)}}\arrow{d}{g_1} &\mathfrak T(X)\arrow{d}{g}\\
X\arrow[hook]{r} & \mathfrak X_1\arrow{r}& \mathfrak X
\end{tikzcd}
\end{center}
which is universal for all the $p$-adic enlargements $(\mathfrak Y, z_{\mathfrak Y})$ of $X$ in $\mathfrak X$, i.e for all the $p$-adic enlargements $(\mathfrak Y,z_{\mathfrak Y})$ admitting a map $g:\mathfrak Y\rightarrow \mathfrak X$ making the previous diagram commutative.


Write $\mathfrak T(X)(1)$ for the universal $p$-adic enlargements of $X$ in $\mathfrak X\times \mathfrak X$, where one considers $X$ embedded in $\mathfrak X_1\times \mathfrak X_1$ via the diagonal immersion. 
The $p$-adic formal schemes $\mathfrak T(X)$ and $\mathfrak T(X)(1)$ are such that such that $\mathfrak T(X)_K=[X]_{\mathfrak X,|p|}$ and $\mathfrak T(X)(1)_K=[X]_{\mathfrak X\times \mathfrak X,|p|}$ (see \cite[1.1.10]{berthelot}).
\subsubsection{Stratifications}
Let  $\Strat^{(p)}(X,\mathfrak X)$ be the category of modules with a $p$-adically convergent stratifications (\cite[Proposition 2.11]{Ogus}). An object $(\mathcal M,\epsilon)$ in $\Strat^{(p)}(X,\mathfrak X)$ is a coherent module $\mathcal M$ over $[X]_{\mathfrak X,|p|}=\mathfrak T(X)_K$ together with an isomorphism $\epsilon:p_1^*\calM\rightarrow p_2^*\calM$ satisfying a natural cocycle condition, where the $p_i's$ are the two projections $\mathfrak T(X)(1)_K=[X]_{\mathfrak X\times \mathfrak X,|p|}\rightarrow \mathfrak T(X)_K=[X]_{\mathfrak X,|p|}$.
The projections $p_1,p_2:\mathfrak X\times \mathfrak X\rightarrow  \mathfrak X$
give morphisms of enlargements $p_1,p_2:(\mathfrak T(X)(1),z_{\mathfrak T(X)(1)})\rightarrow (\mathfrak T(X),z_{\mathfrak T(X)})$.
Hence, if $\mathcal M$ is in $\Isoc^{(p)}(X)$, there is an isomorphism
$$\epsilon_{\mathcal M,\mathfrak X}:p_1^*(\mathcal M_{(\mathfrak T(X),z_{\mathfrak T(X)})})\simeq \mathcal M_{(\mathfrak T(X)(1),z_{\mathfrak T(X)(1)})} \simeq p_2^*(\mathcal M_{(\mathfrak T(X),z_{\mathfrak T(X)})}). $$
This gives a functor
$$(-_{(\mathfrak T(X),z_{\mathfrak T(X)})},\epsilon_{-,\mathfrak X}):\Isoc^{(p)}(X)\rightarrow \Strat^{(p)}(X,\mathfrak X)$$ that sends $\mathcal M$ to $(\mathcal M_{(\mathfrak T(X),z_{\mathfrak T(X)})},\epsilon_{\mathcal M,\mathfrak X})$. By the universal property of $\mathfrak T(X)$, this functor is an equivalence of categories (\cite[Proposition 2.11]{Ogus}).

Given a frame $(X,\overline X,\mathfrak X)$, one can define the category $\Strat(X,\overline X,\mathfrak X)$ of modules with a stratification on $X$ overconvergent along $\overline X-X$, see \cite[P. 50]{ShihoI} where it is denoted by $I^{\dagger}((X,\overline X)/W|\mathfrak X)$). An object ($\calM,\epsilon$) in $\Strat(X,\overline X,\mathfrak X)$ is a coherent $j^{\dagger}_{X}\mathcal O_{]\overline X[_{\mathfrak X}}$ module together with a $j^{\dagger}_X\mathcal O_{]\overline X[_{\mathfrak X\times \mathfrak X}}$-linear isomorphism $\epsilon:p_1^*\calM\simeq p_2^*\mathcal M$ satisfying a natural cocycle condition, where the $p_i's$ are the two projection maps 
$]\overline X[_{\mathfrak X\times \mathfrak X}\rightarrow ]\overline X[_{\mathfrak X}$. 
As in the $p$-adically convergent situation, one constructs a functor 
$$(-_{(X,\overline X,\mathfrak X)},\epsilon_{-,\mathfrak X}):\Isoc^{\dagger}(X,\overline X)\rightarrow \Strat(X,\overline X,\mathfrak X)$$
which is an equivalence of categories (see \cite[Propositions 7.2.2 and 7.3.11]{Lestum}).
\subsection{Relative p-adic cohomology theories} 
\numberwithin{theorem}{subsubsection}
\subsubsection{Relative p-adic cohomology theories}\label{relativepadic}
Fix a smooth proper morphism of $k$-varieties $f:Y\rightarrow X$ and a closed immersion $i:X\rightarrow \mathfrak X$, where $\mathfrak X$ is a flat $p$-adic formal scheme. Assume that $f:Y\rightarrow X$ has (log-) smooth parameter in the sense of \cite[Definition 3.4]{ShihoI}.
\begin{remark}\label{logsmooth}
If $f:Y\rightarrow X$ has (log-) smooth parameter, for every morphism of $k$-varieties $Z\rightarrow X$, the base change $Y_Z\rightarrow Z$ has (log-) smooth parameter (\cite[Remark 3.5]{ShihoI}). Moreover, if $X$ is smooth, every smooth proper morphism $f:Y\rightarrow X$ of $k$-varieties has (log-) smooth proper parameter.
\end{remark}
Depending on the nature of $i:X\rightarrow \mathfrak X$ one defines different $p$-adic cohomology theories:
\begin{itemize}
\item If $X=\mathfrak X_1$ and   $i:X\rightarrow \mathfrak X$ is the canonical inclusion, then one can define the crystalline higher direct image $R^if_{\mathfrak X,crys,*}\mathcal O_{Y/\mathfrak X}$, that is the higher direct image in the relative crystalline site of $X$ in $\mathfrak X$, well defined since $X\numberwithin{theorem}{subsection} \subseteq \mathfrak X$ is defined by the ideal $(p)$. It lives in $Coh(\mathfrak X_K)$, see e.g \cite[Section 1]{ShihoI}.
\item If $i:X\rightarrow \mathfrak X$ is an homeomorphism, then one can define the convergent higher direct image $R^if_{\mathfrak X,conv,*}\mathcal O_{Y/\mathfrak X}$, that is the higher direct image in the relative convergent site of $X$ in $\mathfrak X$. It lives in $Coh(\mathfrak X_K)$. See e.g \cite[Sections 2-3]{ShihoI}.
\item If $i:X\rightarrow \mathfrak X$ is an arbitrary  closed immersion, one can define the analytic higher direct image $R^if_{\mathfrak X,an,*}\mathcal O_{Y/\mathfrak X}$. It is defined via descent using De Rham cohomology, and it lives in $Coh(]X[_{\mathfrak X})$. For details see \cite[Section 4]{ShihoI}.
\end{itemize}
We complete the picture discussing higher direct images for couples and frames, in the context of overconvergent isocrystals.
If $(Y,\overline Y)$ is a couple, write $\mathcal O^{\dagger}_{(Y,\overline Y)}\in \oi(Y,\overline Y)$ for the unique overconvergent isocrystal such that, for every frame $(Z,\overline Z,\mathfrak Z)$ over $(Y,\overline Y)$, the restriction of $\mathcal O^{\dagger}_{(Y,\overline Y)}$ to $(Z,\overline Z,\mathfrak Z)$ is given by $j_Z^{\dagger}\mathcal O_{]\overline Z[_{\mathfrak Z}}$.
If $(Y,\overline Y)$ is a couple over a frame $(X,\overline X,\mathfrak X)$ and the first arrow $f:Y\rightarrow X$ is smooth and proper, one can define the overconvergent higher direct image $R^if_{(Y,\overline Y)/\mathfrak X, rig,*}\mathcal O^{\dagger}_{(Y,\overline Y)}$. It is again defined using De Rham cohomology and descent. See \cite[Section 5]{Shiho} for the definition, it is a  $j^{\dagger}_X\mathcal O_{]\overline X[_{\mathfrak X}}$ module. It is still an open question whether $R^if_{(Y,\overline Y)/\mathfrak X, rig,*}\mathcal O^{\dagger}_{(Y,\overline Y)}$ is a coherent $j^{\dagger}_X\mathcal O_{]\overline X[_{\mathfrak X}}$- module.
\numberwithin{theorem}{subsubsection}
\subsubsection{Comparison}\label{comparisonhigher}
In some particular situation one can compare the various higher direct images defined in Section \ref{relativepadic}.
Assume that $i:X\rightarrow \mathfrak X$ is a closed immersion, $\mathfrak X$ is formally smooth over $W$ and $f:Y\rightarrow X$ is smooth proper with (log-) smooth parameter. 
Using $f$ one considers $(Y,Y)$ as a couple over the frame $(X,X,\mathfrak X)$. The universal $p$-adic enlargement $\mathfrak T(X)$ of $X$ in $\mathfrak X$ induces a commutative diagram:
\begin{center}
\begin{tikzcd}
Y_{\mathfrak T(X)_1}\arrow{r}{f_{\mathfrak T(X)_1}}\arrow{dd}\arrow[phantom]{rdd}{\Box}& \mathfrak T(X)_1\arrow{rr}\arrow{dd}{u}\arrow{rd} && \mathfrak T(X)\arrow{dd}{u}\\
&& \mathfrak X_1\arrow{rd}\\
Y\arrow{r}{f}& X\arrow{rr}\arrow{ru} && \mathfrak X
\end{tikzcd}
\end{center}
By remark \ref{logsmooth}, the morphism $f_{\mathfrak T(X)_1}:Y_{\mathfrak T(X)_1}\rightarrow \mathfrak T(X)_1$ has (log-) smooth parameter.
In this situation one has $j^{\dagger}_X\mathcal O_{]X[_{\mathfrak X}}=\mathcal O_{]X[_{\mathfrak X}}$, so that
$R^if_{(Y,Y)/\mathfrak X,rig,*}\mathcal O^{\dagger}_{(Y,Y)}$ and $R^if_{\mathfrak X,an,*}\mathcal O_{Y/\mathfrak X}$ are coherent $\mathcal O_{]X[_{\mathfrak X}}$ modules (see \cite[Theorem 5.13]{ShihoI}), while $R^if_{\mathfrak T(X)_1,\mathfrak T(X),an,*}\mathcal O_{Y_{\mathfrak T(X)_1}/\mathfrak T(X)}$, $R^if_{\mathfrak T(X)_1,\mathfrak T(X),conv,*}\mathcal O_{Y_{\mathfrak T(X)_1}/\mathfrak T(X)}$ and $R^if_{\mathfrak T(X)_1,\mathfrak T(X),crys,*}\mathcal O_{Y_{\mathfrak T(X)_1}/\mathfrak T(X)}$ are coherent modules over $[X]_{\mathfrak X, |p|}=\mathfrak T(X)_K$. 
Write
$$u:[X]_{\mathfrak X, |p|}\rightarrow ]X[_{\mathfrak X}$$
for the natural inclusion.
By \cite[Theorem 5.13]{ShihoI}, one has a canonical isomorphism of $\mathcal O_{]X[_{\mathfrak X}}$-modules
$$R^if_{(Y,Y)/\mathfrak X,rig,*}\mathcal O^{\dagger}_{(Y,Y)}\simeq R^if_{\mathfrak X,an,*}\mathcal O_{Y/\mathfrak X}.$$
Pulling back along $u$, one finds canonical isomorphisms of coherent $[X]_{\mathfrak X,|p|}$-modules 
 \numberwithin{equation}{subsubsection}

\begin{equation}\label{equation}
\begin{split}
u^*R^if_{(Y,Y)/\mathfrak X,rig,*}\mathcal O^{\dagger}_{(Y,Y)}\simeq u^*R^if_{\mathfrak X,an,*}\mathcal O_{Y/\mathfrak X}\simeq R^if_{\mathfrak T(X)_1,\mathfrak T(X),an,*}\mathcal O_{Y_{\mathfrak T(X)_1}/\mathfrak T(X)}\\
\simeq R^if_{\mathfrak T(X)_1,\mathfrak T(X),conv,*}\mathcal O_{Y_{\mathfrak T(X)_1}/\mathfrak T(X)}\simeq R^if_{\mathfrak T(X)_1,\mathfrak T(X),crys,*}\mathcal O_{Y_{\mathfrak T(X)_1}/\mathfrak T(X)}.
\end{split}
\end{equation}
where the second isomorphism comes from \cite[Remark 4.2]{ShihoI}, the third from \cite[Theorem 4.6]{ShihoI} and the last one from \cite[Theorem 2.3.6]{ShihoI}.


These isomorphisms are functorial in the following sense. Assume that there is a $k$-variety $Z$, a closed embedding $Z\rightarrow \mathfrak Z$ into a  $p$-adic formal scheme $\mathfrak Z$ formally smooth over $W$ and a commutative diagram
\begin{center}
\begin{tikzcd}
Z\arrow{r}\arrow{d}{g}& \mathfrak Z\arrow{d}{g}\\
X\arrow{r} & \mathfrak X
\end{tikzcd}
\end{center}
By the universal property of $\mathfrak T(X)$, there is an induced map $\mathfrak T(Z)\rightarrow \mathfrak T(X)$ that fits into a commutative diagram
\begin{center}
\begin{tikzcd}
&&& Y_{Z}\arrow{r}\arrow{d}{f_Z}\arrow[phantom]{rd}{\Box} & Y\arrow{d}{f}\\
Y_{\mathfrak T(Z)_1}\arrow{r}\arrow{d}{f_{\mathfrak T(Z)_1}}\arrow{rrru}\arrow[phantom]{rd}{\Box} &Y_{\mathfrak T(X)_1}\arrow[swap]{d}{f_{\mathfrak T(X)_1}}\arrow{rrru}&& Z\arrow{r}{g}\arrow{d} & X\arrow{d}\\
\mathfrak T(Z)_1\arrow{r}\arrow{d}\arrow{rrru} \arrow[phantom,near end]{rrruu}{\Box}&\mathfrak T(X)_1\arrow{rrru}\arrow{d}\arrow[phantom,near start]{rrruu}{\Box}&& \mathfrak Z\arrow{r}{g} & \mathfrak X\\
\mathfrak T(Z)\arrow{r}{g}\arrow{rrru}{u} &\mathfrak T(X)\arrow{rrru}{u}
\end{tikzcd}
\end{center}
Then the following diagram is commutative
\begin{equation}\label{gigadiagram}
{\begin{tikzcd}
u^*g^*R^if_{(Y,Y)/\mathfrak X,rig,*}\mathcal O^{\dagger}_{(Y,Y)}\arrow{r}\arrow{d}{\simeq}& u^*R^if_{(Y_{Z},Y_{Z})/\mathfrak Z,rig,*}\mathcal O^{\dagger}_{(Y_{Z},Y_{Z})}\arrow{d}{\simeq}\\
g^*u^*R^if_{(Y,Y)/\mathfrak X,rig,*}\mathcal O^{\dagger}_{(Y,Y)}\arrow{r}\arrow{d}{\simeq}& u^*R^if_{(Y_{Z},Y_{Z})/\mathfrak Z,rig,*}\mathcal O^{\dagger}_{(Y_{Z},Y_{Z})}\arrow{d}{\simeq}\\
g^*u^*R^if_{\mathfrak X,an,*}\mathcal O_{Y/\mathfrak X}\arrow{r}\arrow{d}{\simeq} & u^*R^if_{\mathfrak Z,an,*}\mathcal O_{Y_{Z}/\mathfrak Z}\arrow{d}{\simeq}\\
g^*R^if_{\mathfrak T(X)_1,\mathfrak T(X),an,*}\mathcal O_{Y_{\mathfrak T(X)_1}/\mathfrak T(X)}\arrow{d}{\simeq} \arrow{r}& R^if_{\mathfrak T(Z)_1,\mathfrak T(Z),an,*}\mathcal O_{Y_{\mathfrak T(Z)_1}/\mathfrak T(Z)}\arrow{d}{\simeq}\\
g^*R^if_{\mathfrak T(X)_1,\mathfrak T(X),conv,*}\mathcal O_{Y_{\mathfrak T(X)_1}/\mathfrak T(X)}\arrow{r}\arrow{d}{\simeq} & R^if_{\mathfrak T(Z)_1,\mathfrak T(Z),conv,*}\mathcal O_{Y_{\mathfrak T(Z)_1}/\mathfrak T(Z)}\arrow{d}{\simeq}\\
g^*R^if_{\mathfrak T(X)_1,\mathfrak T(X),crys,*}\mathcal O_{Y_{\mathfrak T(X)_1}/\mathfrak T(X)}\arrow{r} & R^if_{\mathfrak T(Z)_1,\mathfrak T(Z),crys,*}\mathcal O_{Y_{\mathfrak T(Z)_1}/\mathfrak T(Z)}
\end{tikzcd}}
\end{equation}
where the vertical arrows are the isomorphisms in (\ref{equation}) and the horizontal arrows are the base change maps.
\numberwithin{theorem}{subsection} 
\subsubsection{Ogus higher direct image}\label{Ogushigher}
Fix a smooth proper morphism $f:Y\rightarrow X$ of $k$-varieties. Write $R^if_{Ogus,*}\mathcal O_{Y/K}$ in $\Fisoc^{(p)}(X)$ for the Ogus higher direct image (\cite[Section 3, Theorem 3.1]{Ogus}) and recall that its formation is compatible with base change (\cite[Proposition 3.5]{Ogus}).

As object in $\Isoc^{(p)}(X)$, $R^if_{Ogus,*}\mathcal O_{Y/K}$ is characterized by the property that for every $p$-adic enlargement $(\mathfrak T,z_{\mathfrak T})$ one has 
$$(R^if_{Ogus,*}\mathcal O_{Y/K})_{(\mathfrak T,z_{\mathfrak T})}=R^if_{\mathfrak T_1,\mathfrak T,crys,*}\mathcal O_{Y_{\mathfrak T_1}/\mathfrak T}$$
and if $g:(\mathfrak T,z_{\mathfrak T})\rightarrow (\mathfrak Z,z_{\mathfrak Z})$ if a morphism of $p$-adic enlargements, the map
\begin{center}
\begin{tikzcd}
g^*(R^if_{Ogus,*}\mathcal O_{Y/K})_{(\mathfrak Z,z_{\mathfrak Z})}\arrow{r}{\phi_g}\arrow[equal]{d}&(R^if_{Ogus,*}\mathcal O_{Y/K})_{(\mathfrak T,z_{\mathfrak T})}\arrow[equal]{d}\\
g^*R^if_{\mathfrak Z_1,\mathfrak Z,crys,*}\mathcal O_{Y_{\mathfrak Z_1}/\mathfrak Z}\arrow{r}{\phi_g} & R^if_{\mathfrak T_1,\mathfrak T,crys,*}\mathcal O_{Y_{\mathfrak T_1}/\mathfrak T}
\end{tikzcd}
\end{center}
is the base change morphism induced by $g$ (see the proof of \cite[Theorem 3.1]{Ogus}).
In particular if $X$ admits a closed immersion into a $p$-adic formal scheme $\mathfrak X$ formally smooth over $W$, the image of $R^if_{Ogus,*}\mathcal O_{Y/K}$ in $\Strat^{(p)}(X,\mathfrak X)$ is given by the couple $$(R^if_{\mathfrak T(X)_1,\mathfrak T(X),crys,*}\mathcal O_{Y_{\mathfrak T(X)_1}/\mathfrak T(X)},\epsilon_{R^if_{Ogus,*}\mathcal O_{Y/K},\mathfrak X}),$$ where $\epsilon_{R^if_{Ogus,*}\mathcal O_{Y/K},\mathfrak X}$ is induced by the base change morphisms
$$
\begin{forcedcentertikzcd}[column sep=tiny]
p_1^*R^if_{\mathfrak T(X)_1,\mathfrak T(X),crys,*}\mathcal O_{Y_{\mathfrak T(X)_1}/\mathfrak T(X)}\arrow{r}{\simeq}&
R^if_{\mathfrak T(X)(1)_1,\mathfrak T(X)(1),crys,*}\mathcal O_{Y_{\mathfrak T(X)(1)_1}/\mathfrak T(X)(1)}&
p_2^*R^if_{\mathfrak T(X)_1,\mathfrak T(X),crys,*}\mathcal O_{Y_{\mathfrak T(X)_1}/\mathfrak T(X)}\arrow[swap]{l}{\simeq}.
\end{forcedcentertikzcd}
$$
%\begin{center}
%\begin{tikzcd}
%p_1^*R^if_{\mathfrak T(X)_1,\mathfrak T(X),crys,*}\mathcal O_{Y_{\mathfrak T(X)_1}/\mathfrak T(X)}\arrow{d}{\simeq}\\
%R^if_{\mathfrak T(X)(1)_1,\mathfrak T(X)(1),crys,*}\mathcal O_{Y_{\mathfrak T(X)(1)_1}/\mathfrak T(X)(1)}\\
%p_2^*R^if_{\mathfrak T(X)_1,\mathfrak T(X),crys,*}\mathcal O_{Y_{\mathfrak T(X)_1}/\mathfrak T(X)}\arrow[swap]{u}{\simeq}
%\end{tikzcd}
%\end{center}
The Frobenius structure 
$$F_X^*R^if_{Ogus,*}\mathcal O_{Y/K}\rightarrow R^if_{Ogus,*}\mathcal O_{Y/K}$$ 
is constructed in the following way (see the proof of \cite[Theorem 3.7]{Ogus}). 
Consider the commutative cartesian diagram
\begin{center}
\begin{tikzcd}
Y'\arrow{d}{f'}\arrow{r}\arrow[phantom]{rd}{\Box}& Y\arrow{d}{f}\\

X\arrow{r}{F_X}& X
\end{tikzcd}
\end{center}
and for every $p$-adic enlargement $(\mathfrak T,z_{\mathfrak T})$ of $X$, consider the following diagram
\begin{center}
\begin{tikzcd}
Y_{\mathfrak T_1} \arrow[bend left]{rrr}{F_{Y_{\mathfrak T_1} }}\arrow[swap]{rr}{Fr_{Y_{\mathfrak T_1}/\mathfrak T_1}}\arrow{d}{f_{\mathfrak T_1}}&&Y'_{\mathfrak T_1}\arrow{r}\arrow{d}{f'_{\mathfrak T_1}}\arrow[phantom]{rd}{\Box} & Y_{\mathfrak T_1}\arrow{d}{f_{\mathfrak T_1}}\\
\mathfrak T_{1}\arrow{rr}\arrow{d}\arrow[equal]{rr}&&\mathfrak T_1\arrow{r}{F_{\mathfrak T_1}}\arrow{d} & \mathfrak T_1\\
\mathfrak T\arrow[equal]{rr}&& \mathfrak T
\end{tikzcd}
\end{center}
where $Fr_{Y_{\mathfrak T_1}/\mathfrak T_1}$ is the relative Frobenius morphism.
By the compatibility of $R^if_{Ogus,*}\mathcal O_{Y/K}$ with base change, there is a canonical isomorphism
$$F_X^*R^if_{Ogus,*}\mathcal O_{Y/K}\simeq R^if'_{Ogus,*}\mathcal O_{Y'/K}$$
and hence
$$(F_X^*R^if_{Ogus,*}\mathcal O_{Y/K})_{(\mathfrak T,z_{\mathfrak T})}\simeq (R^if'_{Ogus,*}\mathcal O_{Y'/K})_{(\mathfrak T,z_{\mathfrak T})}=R^if'_{\mathfrak T_1,\mathfrak T,crys,*}\mathcal O_{Y'_{\mathfrak T_1}/\mathfrak T}.$$
Then the Frobenius structure is constructed as the base change map
\begin{center}
\begin{tikzcd}
(F_X^*R^if_{Ogus,*}\mathcal O_{Y/K})_{(\mathfrak T,z_{\mathfrak T})}\arrow{r}\arrow{d}{\simeq}&(R^if_{Ogus,*}\mathcal O_{Y/K})_{(\mathfrak T,z_{\mathfrak T})}\arrow{d}{\simeq}\\
R^if'_{\mathfrak T_1,\mathfrak T,crys,*}\mathcal O_{Y'_{\mathfrak T_1}/\mathfrak T}\arrow{r} & R^if_{\mathfrak T_1,\mathfrak T,crys,*}\mathcal O_{Y_{\mathfrak T_1}/\mathfrak T}
\end{tikzcd}
\end{center}
induced by $Fr_{Y_{\mathfrak T_1}/\mathfrak T_1}$.

\subsubsection{Statement of Theorem \texorpdfstring{\ref{berthelotconjecture}}-}
The aim of this part is to prove the following theorem. 
\begin{theorem}\label{berthelotconjecture}
Assume that $X$ is a smooth $k$-variety and $f:Y\rightarrow X$ a smooth proper morphism. Then 
$R^if_{Ogus,*}\mathcal O_{Y/K}$ is in the essential image of 
$\Fisoc^{\dagger}(X)\rightarrow \Fisoc^{\dagger}(X,X)\simeq \Fisoc^{(p)}(X)$. 
\end{theorem}
\begin{remark}
Theorem \ref{berthelotconjecture} already appears in the literature as \cite[Corollary 6.2]{berthelotconjecture}, but, as pointed out to us by T.Abe, there might be a gap in the proof. The problem is in the gluing process in \cite[Corollary 6.1]{berthelotconjecture}. The author uses the theory of arithmetic $D$-modules and he tries to compare the higher direct image in that world with $R^i\mathfrak f_{crys,*}\mathcal O_{\mathcal Y/K}(j)$. Locally they coincide, but it is not so clear that the gluing data are compatible, since the isomorphism is defined not at level of complex but only on the level of the derived category. 
So, following a suggestion of T.Abe, we give another proof of Theorem \ref{berthelotconjecture}, using the work of Shiho on the relative log crystalline cohomology (\cite{Shiho}).
\end{remark}
\begin{remark}
The proof actually works more generally for every $E\in \Fisoc^{(p)}(Y)$. The construction of $R^if_{Ogus,*}E$ does not appear in the literature, so we decided to restrict ourself to $R^if_{Ogus,*}\mathcal O_{Y/K}$.
\end{remark}

\section{Proof of Theorem \texorpdfstring{\ref{berthelotconjecture}}-} \label{secproof}
\numberwithin{theorem}{subsection} 
\subsection{Construction of an overconvergent F-isocrystal}\label{shihosssss}
Fix compactifications $Y\subseteq \overline Y$ and $X\subseteq \overline X$ such that the morphism $f:Y\rightarrow X$ extends to a map of couples $(Y,\overline Y)\rightarrow (X,\overline X)$ and $X$ (resp. $Y$) is dense in $\overline X$ (resp. $\overline Y$). We start recalling the main result of \cite{Shiho}. This gives a $\mathcal M$ in $\Fisoc^{\dagger}(X,\overline X)$ which, after a base change and on appropriate frames, looks like $R^if_{Ogus,*}\mathcal O_{Y/K}$. 
To recall the statement, it is helpful to give the following definition.
\begin{definition}
If $(Z,\overline Z, \mathfrak Z)\rightarrow  (X',\overline X',\mathfrak X')$ is a morphism of frames over $(X,\overline X)$ we say that $(Z,\overline Z, \mathfrak Z)$ has $(P_{(X',\overline X',\mathfrak X')})$ if $Z=X\times_{\overline X}\overline Z$ and $\mathfrak Z\rightarrow \mathfrak X'$ is formally smooth.
\end{definition}

By \cite[Theorem 7.9]{Shiho} (and its proof) there exists a frame $(X',\overline X',\mathfrak X')$ over $(X,\overline X)$ such that 
\begin{itemize}
\item $X':=X\times_{\overline X} \overline X'$;
\item $\mathfrak X'$ is formally smooth over $W$;
\item the map $\overline X'\rightarrow \overline X$ is a composition of a surjective proper map followed by a surjective \etale map;
\end{itemize}
and an $\mathcal M$ in $\Foi(X,\overline X)$ with the following properties:
\begin{enumerate}
\item Let $(Z,\overline Z, \mathfrak Z)$ be a frame over $(X',\overline X',\mathfrak X')$ that has $(P_{(X',\overline X',\mathfrak X')})$, so that there is a commutative diagram
 \begin{center}
\begin{tikzcd}
(Y,\overline Y)\arrow{d}\arrow[phantom]{rd}{\Box}&(Y_{X'},\overline Y_{\overline X'})\arrow[phantom]{rd}{\Box}\arrow{d}\arrow{l}& (Y_{Z},\overline Y_{\overline Z})\arrow{l}\arrow{d}\\
(X,\overline X)& (X',\overline X',\mathfrak X')\arrow{l}& (Z,\overline Z,\mathfrak Z)\arrow{l}.
\end{tikzcd}
\end{center}
Then, the image of $\mathcal M$ in $\Strat(Z,\overline Z, \mathfrak Z)$ is given by $$(R^if_{(Y_{Z},\overline Y_{\overline Z})/\mathfrak Z, rig,*}\mathcal O^{\dagger}_{(Y_Z,\overline Y_{\overline Z})},\epsilon)$$
where $\epsilon$ is an isomorphism:
 $$p_1^*R^if_{(Y_{Z},\overline Y_{\overline Z})/\mathfrak Z, rig,*}\mathcal O^{\dagger}_{(Y_Z,\overline Y_{\overline Z})}\rightarrow R^if_{(Y_{Z},\overline Y_{\overline Z})/\mathfrak Z\times \mathfrak Z, rig,*}\mathcal O^{\dagger}_{(Y_Z,\overline Y_{\overline Z})} \leftarrow p_2^*R^if_{(Y_{Z},\overline Y_{\overline Z})/\mathfrak Z, rig,*}\mathcal O^{\dagger}_{(Y_Z,\overline Y_{\overline Z})}$$
 and $p_1,p_2:]\overline Z[_{\mathfrak Z\times_W\mathfrak Z}\rightarrow ]\overline Z[_{\mathfrak Z} $ are the projection maps. 
 If moreover $Z=\overline Z$, then $\epsilon$ is induced by the base change morphisms (\cite[Last paragraph of page 74]{Shiho} and \cite[Theorem 5.14]{ShihoI});
\item Let $h:(Z,\overline Z, \mathfrak Z)\rightarrow (T,\overline T, \mathfrak T)$ be a morphism of frames over $(X',\overline X',\mathfrak X')$ that have $(P_{(X',\overline X',\mathfrak X')})$, so that there is a commutative diagram
 \begin{center}
\begin{tikzcd}
(Y_{Z},\overline Y_{\overline Z})\arrow{d}\arrow{r}& (Y_{T},\overline Y_{\overline T})\arrow{d}\\
(Z,\overline Z,\mathfrak Z)\arrow{r}& (T,\overline T,\mathfrak T).
\end{tikzcd}
\end{center}
 Then, the isomorphism 
\begin{center}
\begin{tikzcd}
h^*\mathcal M_{(Z,\overline Z,\mathfrak Z)}\arrow{r}{\phi_h}\arrow{d}{\simeq} & \mathcal M_{(T,\overline T,\mathfrak T)}\arrow{d}{\simeq}\\
h^*R^if_{(Y_Z,\overline Y_{\overline Z})/\mathfrak Z,rig,*}\mathcal O^{\dagger}_{(Y_Z,\overline Y_{\overline Z})}\arrow{r} & R^if_{(Y_T,\overline Y_{\overline T})/\mathfrak T,rig,*}\mathcal O^{\dagger}_{(Y_T,\overline Y_{\overline T})}
\end{tikzcd}
\end{center}
 given by the isocrystals structure is the base change morphism (This is the functoriality in the statement of \cite[Theorem 7.9]{Shiho}, see \cite[Proof of Theorem 4.8]{ShihoI});
\item Let $(Z,\overline Z, \mathfrak Z)$ be a frame over $(X,\overline X)$ that has $(P_{(X',\overline X',\mathfrak X')})$ and assume that $\mathfrak Z$ admits a lifting $\sigma_{\mathfrak Z}$ of $F_{\mathfrak Z_1}$, so that there is a commutative diagram
 \begin{center}
\begin{tikzcd}
(Y_{Z},\overline Y_{\overline Z})\arrow{d}\arrow{r}{(F_{Y_Z},F_{\overline Y_{\overline Z}})}& (Y_{Z},\overline Y_{\overline Z})\arrow{d}\\
(Z,\overline Z,\mathfrak Z)\arrow{r}{(F_Z,F_{\overline Z},\sigma_{\mathfrak Z})}& (Z,\overline Z,\mathfrak Z).
\end{tikzcd}
\end{center}
Then, the isomorphism induced by the Frobenius structure 
 \begin{center}
\begin{tikzcd}
\sigma_{\mathfrak Z}^*\mathcal M_{(Z,\overline Z,\mathfrak Z)}\simeq (F_X^*\mathcal M)_{(Z,\overline Z,\mathfrak Z)}\arrow{r}\arrow{d}{\simeq} & \mathcal M_{(Z,\overline Z,\mathfrak Z)}\arrow{d}{\simeq}\\
\sigma^*_{\mathfrak Z}R^if_{(Y_Z,\overline Y_{\overline Z})/\mathfrak Z,rig,*}\mathcal O^{\dagger}_{(Y_Z,\overline Y_{\overline Z})}\arrow{r} &R^if_{(Y_Z,\overline Y_{\overline Z})/\mathfrak Z,rig,*}\mathcal O^{\dagger}_{(Y_Z,\overline Y_{\overline Z})}
\end{tikzcd}
\end{center}
is given by the base change morphism induced by $\sigma_{\mathfrak Z}$ and $F_{Y_{Z}}$ (\cite[Proof of Theorem 7.9]{Shiho}).
\end{enumerate}

\numberwithin{theorem}{subsection} 
\subsection{Strategy}\label{strategyp}
To prove Theorem \ref{berthelotconjecture}, it is enough to show that the image of $\mathcal M$ in $\Fisoc^{(p)}(X)$ is isomorphic to $R^{i}f_{Ogus,*}\mathcal O_{Y/K}$. Since $X'$ does not admit a closed immersion directly in $\mathfrak X'$ and $\mathfrak X'$ does not admits a lifting of the absolute Frobenius of $\mathfrak X'_1$, one can't use directly the description of $\mathcal M$ given in previous Section \ref{shihosssss}. But there exists\footnote{To construct it, consider a finite covering $\{Spf(A_i)\}$  of $\mathfrak X'$ by formal affine open sub schemes. Then $\{Spec(A_{i,1})\}$ is a covering of $\mathfrak X'_1$ by affine open sub schemes and $\{V_i:=Spec(A_{i,1})\times_{\mathfrak X'_1} X'\}$ is a Zariski open covering of $X'$. Consider a finite covering $\{U_{i,j}\}$ of $V_i$ by affine open sub schemes. Then the maps $U_{i,j}\rightarrow Spec(A_{i,1})$ are affine and of finite type, so that there are closed immersions $U_{i,j}\rightarrow \mathbb A^{n_{i,j}}_{\Spec(A_{i,1})  }$. Write $\mathfrak U_{i,j}$ for the formal affine space of dimension $n_{i,j}$ over $Spf(A_i)$. Then $U:=\coprod_{i,j}U_{i,j}$ admits a closed immersion into  $\mathfrak U:=\coprod_{i,j}\mathfrak U_{i,j}$ and $\mathfrak U$ is formally smooth over $\mathfrak X'$. To show that $\mathfrak U$ admits a lifting of $F_{\mathfrak U_1}$ it is enough to show that each $\mathfrak U_{i,j}$ admits a lifting of $F_{\mathfrak U_{i,j,1}}$. This follows from the fact that $\mathfrak U_{i,j}$ is formally affine and formally smooth over $W$.} an \etale surjective morphism $U\rightarrow X'$ such that $U$ admits a closed immersion into a $p$-adic formal scheme $\mathfrak U$ which is formally smooth over $\mathfrak X'$ and it is endowed with a lifting $\sigma_{\mathfrak U}$ of $F_{\mathfrak U_1}$.
Write $g$ for the composition $U\rightarrow X'\rightarrow X$. 

To prove Theorem \ref{berthelotconjecture}, first one constructs an isomorphism 
$$\psi:g^*\mathcal M\simeq g^*R^{i}f_{Ogus,*}\mathcal O_{Y/K}\simeq R^{i}f_{U,Ogus,*}\mathcal O_{Y_U/K}\text{ in } \Fisoc^{(1)}(U)$$ where the isomorphism on the right comes from the fact that the formation of $R^{i}f_{Ogus,*}\mathcal O_{Y/K}$ is compatible with base change, see Section \ref{Ogushigher}. Then one uses \etale and proper descent for convergent isocrystals to deduce that $\psi$ descent to $\Fisoc^{(p)}(X)$.
More precisely the proof decomposes as follows:
\begin{enumerate}
\item One constructs an isomorphism $$\psi:g^*\mathcal M\simeq R^{i}f_{U,Ogus,*}\mathcal O_{Y_U/K}$$ in $\Isoc^{(p)}(U)\simeq \Strat^{(p)}(U,\mathfrak U)$. This is done in Section \ref{comparisonisocrystal},  using that $(U,U,\mathfrak U)$ has $(P_{(X',\overline X',\mathfrak X')})$ (so that one can apply the property $(1)$ of $\mathcal M$) and the comparison isomorphisms in \ref{comparisonhigher};
\item One verifies that the $\psi$ commutes with the Frobenius structures i.e. that $\psi$ makes the following diagram commutative
\begin{center}
\begin{tikzcd}
F_U^*g^*\mathcal M\arrow{r}\arrow{d}{F_U^*\psi} & g^*\mathcal M\arrow{d}{\psi}\\
F_U^*R^if_{U,Ogus,*}\mathcal O_{Y_U/K}\arrow{r} & R^if_{U,Ogus,*}\mathcal O_{Y_U/K}
\end{tikzcd}
\end{center}
in $\Isoc^{(p)}(U)\simeq \Strat^{(p)}(U,\mathfrak U)$.  This is done in Section \ref{comparisonfrobenius}, using that $\mathfrak U$ has a lifting of $F_{\mathfrak U_1}$ (so that one can apply the property $(3)$ of $\mathcal M$) and the comparison isomorphisms in \ref{comparisonhigher};
\item By the equivalence $F1-Fp$ in Section \ref{functors}, the first two steps imply that there is an isomorphism 
$$\psi:g^*\mathcal M\simeq R^{i}f_{U,Ogus,*}\mathcal O_{Y_U/K}$$ in $\Fisoc^{(1)}(U)$;
\item To apply descent for convergent isocrystals, one has to check that $\psi$ makes the following diagram in $\Fisoc^{(1)}(U\times_X U)$ commutative:
\begin{center}
\begin{tikzcd}
q_1^*g^*\mathcal M\arrow{r}\arrow{d}{q_1^*\psi} & q_2^*g^*\mathcal M\arrow{d}{q^*_2\psi}\\
q_1^*g^*R^if_{Ogus,*}\mathcal O_{Y/K}\arrow{r} & q_2^*g^*R^if_{Ogus,*}\mathcal O_{Y/K}
\end{tikzcd}
\end{center}
where $q_1,q_2:U\times_XU\rightarrow U$ are the projections. To check this, by the equivalence $F1-Fp$, it is enough to show that it is commutative in $\Fisoc^{(p)}(U\times_X U)$ or equivalently in $\Isoc^{(p)}(U\times_X U)\simeq \Strat^{(p)}(U\times_X U,\mathfrak U\times_W \mathfrak U)$. This is done in Section \ref{descent}, using that $q_1,q_2:(U\times_X U,U\times_X U,\mathfrak U\times_W \mathfrak U)\rightarrow (U,U,\mathfrak U)$ are morphisms of frames that have $(P_{(X',\overline X',\mathfrak X')})$ (so that one can apply the property $(2)$ of $\mathcal M$) and the comparison isomorphisms in \ref{comparisonhigher}. 
\end{enumerate}  
\begin{remark}
The reason why one needs to pass back and forth between $\Fisoc^{(p)}(U)$ and $\Fisoc^{(1)}(U)$ is that proper descent is not known for the category $\Isoc^{(p)}(U)$, while proper descent for the category $\Isoc^{(1)}(U)$ (and hence for $\Fisoc^{(1)}(U)$) is proved in \cite{Ogus}. On the other hand one knows the value of $R^if_{U,Ogus,*}\mathcal O_{Y_U/K}$ only on $p$-adic enlargements. The equivalences of categories in Section \ref{functors} allow to combine these informations. 
\end{remark}
\subsection{Comparison of isocrystals}\label{comparisonisocrystal}
In this section we construct an isomorphism
$$\psi:g^*\mathcal M\simeq R^if_{U,Ogus,*}\mathcal O_{Y_{U}/K} \text{ in } \Isoc^{(p)}(U).$$
Consider the universal $p$-adic enlargements $\mathfrak T(U)$ and $\mathfrak T(U)(1)$ of $U$ in $\mathfrak U$ and $\mathfrak U\times \mathfrak U$ and write $u,p_{1},p_2$ for the natural morphisms
\begin{center}
\begin{tikzcd}
(\mathfrak T(U)(1)_{1},\mathfrak T(U)(1)_{1},\mathfrak T(U)(1))\arrow{r}{u}\arrow[d, shift left=1.5ex,"p_1"]\arrow[swap]{d}{p_2} & (U,U,\mathfrak U\times \mathfrak U)\arrow[d, shift left=1.5ex,"p_1"]\arrow[swap]{d}{p_2}\\
(\mathfrak T(U)_{1},\mathfrak T(U)_{1},\mathfrak T(U))\arrow{r}{u} & (U,U,\mathfrak U)
\end{tikzcd}
\end{center}
Since $(U,U,\mathfrak U)$ has $(P_{(X',\overline X',\mathfrak X')})$, by the property $(1)$ of $\mathcal M$ in \ref{shihosssss}, one has:
$$\mathcal M_{(U,U,\mathfrak U)}=R^if_{(Y_{U},Y_{U})/\mathfrak U, rig,*}\mathcal O^{\dagger}_{(Y_{U},Y_{U})} \text{ in } Coh(]U[_{\mathfrak U}).$$
Since $\mathcal M$ is an isocrystal, one gets $\mathcal M_{(\mathfrak T(U)_1,\mathfrak T(U)_1,\mathfrak T(U))}\simeq u^*\mathcal M_{(U,U,\mathfrak U)} \text{ in } Coh(\mathfrak T(U)_K).$
Then as in (\ref{equation}): 
$$\mathcal M_{(\mathfrak T(U),z_{\mathfrak T(U)})}\simeq u^*R^if_{(Y_{U},Y_{U})/\mathfrak U, rig,*}\mathcal O^{\dagger}_{(Y_{U},Y_{U})}\simeq u^*R^if_{U,\mathfrak U,an,*}\mathcal O_{Y_{U}/\mathfrak U}\simeq R^if_{\mathfrak T(U)_1,\mathfrak T(U),an,*}\mathcal O_{Y_{\mathfrak T(U)_{1}}/\mathfrak T(U)}$$
$$\simeq R^if_{\mathfrak T(U)_{1},\mathfrak T(U),conv,*}\mathcal O_{Y_{\mathfrak T(U)_{1}}/\mathfrak T(U)}\simeq R^if_{\mathfrak T(U)_{1},\mathfrak T(U),crys,*}\mathcal O_{Y_{\mathfrak T(U)_{1}}/\mathfrak T(U)} \text{ in } Coh(\mathfrak T(U)_K).$$

Since, by construction (\ref{Ogushigher}), $$R^if_{\mathfrak T(U)_{1},\mathfrak T(U),crys,*}\mathcal O_{Y_{\mathfrak T(U)_{1}}/\mathfrak T(U)}=(R^if_{U,Ogus,*}\mathcal O_{Y_{U}/K})_{(\mathfrak T(U),z_{\mathfrak T(U)})}$$
one has an isomorphism 
$$\psi:\mathcal M_{(\mathfrak T(U),z_{\mathfrak T(U)})}\simeq (R^if_{U,Ogus,*}\mathcal O_{Y_{U}/K})_{(\mathfrak T(U),z_{\mathfrak T(U)})} \text{ in } Coh(\mathfrak T(U)_K).$$
To promote $\psi$ to an isomorphism in $\Strat^{(p)}(U,\mathfrak U)\simeq \Isoc^{(p)}(U)$ one has to check that $\psi$ is compatible with the stratifications $\epsilon_{g^*\mathcal M,\mathfrak U}$ on $g^*\mathcal M$ and $\epsilon_{R^if_{U,Ogus,*}\mathcal O_{Y_{U}/K},\mathfrak U}$ on $(R^if_{U,Ogus,*}\mathcal O_{Y_{U}/K})_{(\mathfrak T(U),z_{\mathfrak T(U)})}$.

Since $(U,U,\mathfrak U)$ has $(P_{(X',\overline X',\mathfrak X')})$, by the property (1) in \ref{shihosssss}, the stratification $\epsilon_{g^*\mathcal M,\mathfrak U}$ is given by the base change morphisms:
$$p_1^*R^if_{(Y_{U},Y_{U})/\mathfrak U, rig,*}\mathcal O^{\dagger}_{(Y_{U},Y_{U})}\rightarrow R^if_{(Y_{U},Y_{U})/\mathfrak U\times \mathfrak U, rig,*}\mathcal O^{\dagger}_{(Y_{U},Y_{U})} \leftarrow p_2^*R^if_{(Y_{U},Y_{U})/\mathfrak U, rig,*}\mathcal O^{\dagger}_{(Y_{U},Y_{U})}.$$
As in (\ref{gigadiagram}) pulling back to $u^*$, one has a commutative diagram
\begin{center}
\begin{tikzcd}[column sep=tiny]
u^*p_1^*R^if_{(Y_{U},Y_{U})/\mathfrak U, rig,*}\mathcal O^{\dagger}_{(Y_{U},Y_{U})}\arrow{r}{\simeq}\arrow{d}{\simeq} &\arrow{d}{\simeq} u^*R^if_{(Y_{U},Y_{U})/\mathfrak U\times \mathfrak U, rig,*}\mathcal O^{\dagger}_{(Y_{U},Y_{U})} & \arrow{d}{\simeq}u^*p_2^*R^if_{(Y_{U},Y_{U})/\mathfrak U, rig,*}\mathcal O^{\dagger}_{(Y_{U},Y_{U})} \arrow{l}{\simeq}\arrow{d}{\simeq}\\
p_1^*u^*R^if_{(Y_{U},Y_{U})/\mathfrak U, rig,*}\mathcal O^{\dagger}_{(Y_{U},Y_{U})}\arrow{r}{\simeq}\arrow{d}{\simeq} &\arrow{d}{\simeq} u^*R^if_{(Y_{U},Y_{U})/\mathfrak U\times \mathfrak U, rig,*}\mathcal O^{\dagger}_{(Y_{U},Y_{U})} & \arrow{d}{\simeq}p_2^*u^*R^if_{(Y_{U},Y_{U})/\mathfrak U, rig,*}\mathcal O^{\dagger}_{(Y_{U},Y_{U})} \arrow{l}{\simeq}\arrow{d}{\simeq}\\
p_1^*u^*R^if_{U,\mathfrak U,an,*}\mathcal O_{Y_{U}/\mathfrak U}\arrow{r}{\simeq}\arrow{d}{\simeq} & u^*R^if_{U,\mathfrak U\times \mathfrak U,an,*}\mathcal O_{Y_{U}/\mathfrak U\times \mathfrak U}\arrow{d}{\simeq}& p_2^*u^*R^if_{U,\mathfrak U,an,*}\mathcal O_{Y_{U}/\mathfrak U} \arrow{l}{\simeq}\arrow{d}{\simeq}\\
p_1^*R^if_{\mathfrak T(U)_{1},\mathfrak T(U),crys,*}\mathcal O_{Y_{\mathfrak T(U)_1}/\mathfrak T(U)}\arrow{r}{\simeq}&R^if_{\mathfrak T(U)(1)_{1},\mathfrak T(U)(1),crys,*}\mathcal O_{Y_{\mathfrak T(U)(1)_1}/\mathfrak T(U)(1)} &\arrow{l}{\simeq}p_2^*R^if_{\mathfrak T(U)_{1},\mathfrak T(U),crys,*}\mathcal O_{Y_{\mathfrak T(U)_1}/\mathfrak T(U)} \arrow{l}{\simeq}
\end{tikzcd}
\end{center}
where the horizontal maps are the natural base change maps. So, the stratification $\epsilon_{g^*\mathfrak M,\mathfrak U}$ on $g^*\mathcal M$
\begin{center}
\begin{tikzcd}[column sep=tiny]
p_1^*\mathcal M_{(\mathfrak T(U),z_{\mathfrak T(U)})}\arrow{r}\arrow{d}{\simeq}&\mathcal M_{(\mathfrak T(U)(1),z_{\mathfrak T(U)(1)})}\arrow{d}{\simeq}& \arrow{l}p_2^*\mathcal M_{(\mathfrak T(U),z_{\mathfrak T(U)})}\arrow{d}{\simeq}\\
p_1^*R^if_{\mathfrak T(U)_{1},\mathfrak T(U),crys,*}\mathcal O_{Y_{\mathfrak T(U)_1}/\mathfrak T(U)}\arrow{r}{\simeq}&R^if_{\mathfrak T(U)(1)_{1},\mathfrak T(U)(1),crys,*}\mathcal O_{Y_{\mathfrak T(U)(1)_1}/\mathfrak T(U)(1)} &\arrow{l}{\simeq}p_2^*R^if_{\mathfrak T(U)_{1},\mathfrak T(U),crys,*}\mathcal O_{Y_{\mathfrak T(U)_1}/\mathfrak T(U)} \arrow{l}{\simeq}
\end{tikzcd}
\end{center}
is induced by the base change morphisms.
Since $\epsilon_{R^if_{U,Ogus,*}\mathcal O_{Y_{U}/K},\mathfrak U}$ is induced by the base change morphisms by construction (\ref{Ogushigher}), one concludes that 
$$\psi:\mathcal M_{(\mathfrak T(U),z_{\mathfrak T(U)})}\simeq (R^if_{U,Ogus,*}\mathcal O_{Y_{U}/K})_{(\mathfrak T(U),z_{\mathfrak T(U)})}\text{ in } Coh(\mathfrak T(U)_K)$$ is compatible with the stratifications and hence induces an isomorphism 
$$\psi: g^*\mathcal M\simeq R^if_{U,Ogus,*}\mathcal O_{Y_{U}/K}  \text{ in } \Strat^{(p)}(U,\mathfrak U)\simeq \Isoc^{(p)}(U).$$
\subsection{Comparison of Frobenius structures}\label{comparisonfrobenius}
We now check that $\psi$ is compatible with the Frobenius structures, i.e. that the following diagram in $\Isoc^{(p)}(U)$
is commutative:
\begin{center}
\begin{tikzcd}
F_U^*g^*\mathcal M\arrow{r}\arrow{d}{F_U^*\psi} & g^*\mathcal M\arrow{d}{\psi}\\
F_U^*R^if_{U,Ogus,*}\mathcal O_{Y_U/K}\arrow{r} & R^if_{U,Ogus,*}\mathcal O_{Y_U/K}
\end{tikzcd}
\end{center}
Since 
$$(-_{(\mathfrak T(U),z_{\mathfrak T(U)})},\epsilon_{-,\mathfrak U}):\Isoc^{(p)}(U)\rightarrow \Strat^{(p)}(U,\mathfrak U)$$
is an equivalence of categories, it is enough to show that 
\begin{center}
\begin{tikzcd}
(F_U^*g^*\mathcal M)_{(\mathfrak T(U),z_{\mathfrak T(U)})}\arrow{r}\arrow{d} & g^*\mathcal M_{(\mathfrak T(U),z_{\mathfrak T(U)})}\arrow{d}\\
(F_U^*R^if_{U,Ogus,*}\mathcal O_{Y_U/K})_{(\mathfrak T(U),z_{\mathfrak T(U)})}\arrow{r} & (R^if_{U,Ogus,*}\mathcal O_{Y_U/K})_{(\mathfrak T(U),z_{\mathfrak T(U)})}
\end{tikzcd}
\end{center}
is commutative.
Since ($U,U,\mathfrak U)$ has $(P_{(X',\overline X',\mathfrak X')})$ and it is endowed with a morphism $\sigma_{\mathfrak U}$ lifting $F_{\mathfrak U_1}$, by the property $(3)$ of $\mathcal M$ in \ref{shihosssss}, the Frobenius structure on $\mathcal M_{(U,U,\mathfrak U)}$ is given by the base change map induced by $\sigma_{\mathfrak U}$ and $F_{Y_{U}}:$
$$\sigma_{\mathfrak U}^*R^if_{(Y_{U},Y_{U})/\mathfrak U, rig,*}\mathcal O^{\dagger}_{(Y_{U},Y_{U})}\rightarrow R^if_{(Y_{U},Y_{U})/\mathfrak U, rig,*}\mathcal O^{\dagger}_{(Y_{U},Y_{U})}.$$
By the universal property of the universal $p$-adic enlargement one gets a commutative diagram:
\begin{center}
\begin{tikzcd}
U\arrow{r}\arrow{d}{F_U} & \mathfrak T(U)\arrow{r}{u}\arrow{d}{\sigma_{\mathfrak T(U)}} & \mathfrak U\arrow{d}{\sigma_{\mathfrak U}}\\
U\arrow{r} & \mathfrak T(U)\arrow{r}{u}& \mathfrak U
\end{tikzcd}
\end{center}
Pulling back via $u$, there is a commutative diagram (\ref{gigadiagram}):
\begin{center}
\begin{tikzcd}
u^*\sigma_{\mathfrak U}^*R^if_{(Y_{U},Y_{U})/\mathfrak U, rig,*}\mathcal O^{\dagger}_{(Y_{U},Y_{U})}\arrow{r}\arrow{d}{\simeq} & u^*R^if_{(Y_{U},Y_{U})/\mathfrak U, rig,*}\mathcal O^{\dagger}_{(Y_{U},Y_{U})}\arrow{d}{\simeq}\\
\sigma_{\mathfrak T(U)}^*u^*R^if_{(Y_{U},Y_{U})/\mathfrak U, rig,*}\mathcal O^{\dagger}_{(Y_{U},Y_{U})}\arrow{r}\arrow{d}{\simeq} & u^*R^if_{(Y_{U},Y_{U})/\mathfrak U, rig,*}\mathcal O^{\dagger}_{(Y_{U},Y_{U})}\arrow{d}{\simeq}\\
\sigma_{\mathfrak T(U)}^*R^if_{\mathfrak T(U)_1,\mathfrak T(U),an,*}\mathcal O_{Y_{\mathfrak T(U)_1}/\mathfrak T(U)}\arrow{r}\arrow{d}{\simeq} & R^if_{\mathfrak T(U)_1\mathfrak T(U),an,*}\mathcal O_{Y_{\mathfrak T(U)_1}/\mathfrak T(U)}\arrow{d}{\simeq}\\
\sigma_{\mathfrak T(U)}^* R^if_{\mathfrak T(U)_1,\mathfrak T(U),crys,*}\mathcal O_{Y_{\mathfrak T(U)_1}/\mathfrak T(U)}\arrow{r} & R^if_{\mathfrak T(U)_1,\mathfrak T(U),crys,*}\mathcal O_{Y_{\mathfrak T(U)_1}/\mathfrak T(U)}
\end{tikzcd}
\end{center}
where the horizontal morphisms are the base change morphisms. 
So the morphism
\begin{center}
\begin{tikzcd}
\sigma_{\mathfrak T}^*\mathcal M_{(\mathfrak T(U),z_{\mathfrak T(U)})}\simeq (F_U^*g^*\mathcal M)_{(\mathfrak T(U),z_{\mathfrak T(U)})}\arrow{r}\arrow{d}{\simeq} & g^*\mathcal M_{(\mathfrak T(U),z_{\mathfrak T(U)})}\arrow{d}{\simeq}\\
\sigma_{\mathfrak T(U)}^* R^if_{\mathfrak T(U)_1,\mathfrak T(U),crys,*}\mathcal O_{Y_{\mathfrak T(U)_1}/\mathfrak T(U)}\arrow{r} & R^if_{\mathfrak T(U)_1,\mathfrak T(U),crys,*}\mathcal O_{Y_{\mathfrak T(U)_1}/\mathfrak T(U)}
\end{tikzcd}
\end{center}
is induced by the base change morphism for $F_{Y_{\mathfrak T(U)_1}}$ and $\sigma_{\mathfrak T(U)}$. 


We check that the same is true for $R^if_{U,Ogus,*}\mathcal O_{Y_U/K}$. Consider the commutative diagram
\begin{center}
\begin{tikzcd}
Y_{\mathfrak T(U)_1} \arrow[bend left]{rrr}{F_{Y_{\mathfrak T(U)_1} }}\arrow[swap]{rr}{Fr_{Y_{\mathfrak T(U)_1}/\mathfrak T(U)_1}}\arrow{d}{f_{\mathfrak T(U)_1}}&&Y'_{\mathfrak T(U)_1}\arrow{r}\arrow{d}{f'_{\mathfrak T(U)_1}}\arrow[phantom]{rd}{\Box} & Y_{\mathfrak T(U)_1}\arrow{d}{f_{\mathfrak T(U)_1}}\\
\mathfrak T(U)_{1}\arrow{rr}\arrow{d}\arrow[equal]{rr}&&\mathfrak T(U)_1\arrow{r}{F_{\mathfrak T(U)_1}}\arrow{d} & \mathfrak T(U)_1\arrow{d}\\
\mathfrak T(U)\arrow[equal]{rr}&& \mathfrak T(U)\arrow{r}{\sigma_{\mathfrak T(U)}} &\mathfrak T(U)
\end{tikzcd}
\end{center}
and recall (\ref{Ogushigher}) that the Frobenius structure is defined by the base change map induced by $Fr_{Y_{\mathfrak T(U)_1}/\mathfrak T(U)_1}$:
$$(F_U^*R^if_{U,Ogus,*}\mathcal O_{Y_U/K})_{(\mathfrak T(U)_1,\mathfrak T(U)_1,\mathfrak T(U))}\simeq R^if'_{\mathfrak T(U)_1,\mathfrak T(U),crys,*}\mathcal O_{Y'_{\mathfrak T(U)_1}/\mathfrak T(U)}\rightarrow  R^if_{\mathfrak T(U)_1,\mathfrak T(U),crys,*}\mathcal O_{Y_{\mathfrak T(U)_1}/\mathfrak T(U)}$$
Since there is a lifting $\sigma_{\mathfrak T(U)}$ of $F_{\mathfrak T(U)_1}$, there is a morphism of enlargements
$$\sigma_{\mathfrak T(U)}:(\mathfrak T(U),z_{\mathfrak T(U)})\rightarrow (\mathfrak T(U),z_{\mathfrak T(U)}\circ F_{\mathfrak T(U)_1}).$$ Since $F^*_{U}R^{i}f_{Ogus,*}\mathcal O_{\mathcal Y_U/K} $ is a crystal, there is an isomorphism
$$\sigma_{\mathfrak T(U)}^*R^if_{\mathfrak T(U)_1,\mathfrak T(U),crys,*}\mathcal O_{Y_{\mathfrak T(U)_1}/\mathfrak T(U)}\rightarrow R^if'_{\mathfrak T(U)_1,\mathfrak T(U),crys,*}\mathcal O_{Y'_{\mathfrak T(U)_1}/\mathfrak T(U)}.$$
that, as recalled in section \ref{Ogushigher}, identifies with the base change map induced by $\sigma_{\mathfrak T(U)}$ and $F_{\mathfrak T(U)_1}$.
So the Frobenius structure
\begin{center}
\begin{tikzcd}
\sigma_{\mathfrak T(U)}^*(R^if_{U,Ogus,*}\mathcal O_{Y_U/K})_{(\mathfrak T(U),z_{\mathfrak T(U)})}\simeq (F_U^*R^if_{U,Ogus,*}\mathcal O_{Y_U/K})_{(\mathfrak T(U),z_{\mathfrak T(U)})}\arrow{r}\arrow{d}{\simeq} & (R^if_{U,Ogus,*}\mathcal O_{Y_U/K})_{(\mathfrak T(U),z_{\mathfrak T(U)})}\arrow{d}{\simeq}\\
\sigma_{\mathfrak T(U)}^* R^if_{\mathfrak T(U)_1,\mathfrak T(U),crys,*}\mathcal O_{Y_{\mathfrak T(U)_1}/\mathfrak T(U)}\arrow{r} & R^if_{\mathfrak T(U)_1,\mathfrak T(U),crys,*}\mathcal O_{Y_{\mathfrak T(U)_1}/\mathfrak T(U)}
\end{tikzcd}
\end{center}
is given by the composition of the base change maps induced by $Fr_{Y_{\mathfrak T(U)_1}/\mathfrak T(U)_1}$ followed by the base change map induced by $\sigma_{\mathfrak T(U)}$ and $F_{\mathfrak T(U)_1}$, hence it is given by the base change map induced by $F_{Y_{\mathfrak T(U)_1}}$ and $\sigma_{\mathfrak T(U)}$.


In conclusion, $\psi$ is compatible with the Frobenius structures of $\mathcal M$ and $R^if_{U,Ogus,*}\mathcal O_{Y_U/K}$, so that $\psi$ is gives an isomorphism 
$$\psi: R^if_{U,Ogus,*}\mathcal O_{Y_U/K}\simeq g^*\mathcal M \text{ in } \Fisoc^{(p)}(U)$$ and hence an isomorphism 
$$\psi: R^if_{U,Ogus,*}\mathcal O_{Y_U/K}\simeq g^*\mathcal M \text{ in } \Fisoc^{(1)}(U).$$
\subsection{Descent}\label{descent}
 \numberwithin{equation}{subsection}

Now one has to descend from $U$ to $X$.
To do this, consider the closed immersion
$$U\times_XU\rightarrow U\times_{k} U\rightarrow \mathfrak U_1\times_{k}\mathfrak U_1\rightarrow \mathfrak U\times_{W}\mathfrak U$$
where the first map is a closed immersion by \cite[Tag 01KR]{stacks-project} since $X$ is separated.
Write $\mathfrak T(U\times_X U)$ for the universal $p$-adic enlargement of $U\times_XU$ in $\mathfrak U\times_{W}\mathfrak U$ and $q_1,q_2$ for the projections $$U\times_X U\rightarrow U \quad \text{and}\quad \mathfrak U\times_{W}\mathfrak U\rightarrow \mathfrak U.$$ Finally write $u_{\mathfrak T(U\times_X U)},q_{\mathfrak T(U\times_X U),1}$ and $q_{\mathfrak T(U\times_X U),2}$ for the natural morphisms:
\begin{center}
\begin{tikzcd}
(\mathfrak T(U\times_X U)_{1},\mathfrak T(U\times_X U)_{1},\mathfrak T(U\times_X U))\arrow{r}{u_{\mathfrak T(U\times_X U)}}\arrow[d, shift left=1.5ex,"q_{\mathfrak T(U\times_X U),1}"]\arrow[swap]{d}{q_{\mathfrak T(U\times_X U),2}} & (U\times_X U,U\times_X U, \mathfrak U\times_{W} \mathfrak U)\arrow[d, shift left=1.5ex,"q_1"]\arrow[swap]{d}{q_2}\\
(\mathfrak T(U)_{1},\mathfrak T(U)_{1},\mathfrak T(U))\arrow{r}{u} & (U,U,\mathfrak U)
\end{tikzcd}
\end{center} 
and $g'$ for $U\times_X U\rightarrow X$.
By the equivalence $F1-Fp$ in Section \ref{functors}, to show that the descent diagram in $\Fisoc^{(1)}(U\times_X U)$
\begin{equation}\label{descentdiagram}
{
\begin{tikzcd}
q_1^*g^*\mathcal M\arrow{r}\arrow{d} & q_2^*g^*\mathcal M\arrow{d}\\
q_1^*g^*R^if_{Ogus,*}\mathcal O_{Y/K}\arrow{r} & q_2^*g^*R^if_{Ogus,*}\mathcal O_{Y/K}
\end{tikzcd}}
\end{equation}
is commutative, it is enough to show that it is commutative in $\Fisoc^{(p)}(U\times_XU)$. Then one decomposes \ref{descentdiagram} as follows:
\begin{center}
\begin{tikzcd}
q_1^*g^*\mathcal M\arrow{r}\arrow{d} &g^{',*}\mathcal M\arrow{d} & q_2^*g^*\mathcal M\arrow{d}\arrow{l}\\
q_1^*R^if_{Ogus,*}\mathcal O_{Y/K}\arrow{r} &g^{',*}R^if_{Ogus,*}\mathcal O_{Y/K}\arrow{r} & q_2^*R^if_{Ogus,*}\mathcal O_{Y/K}\arrow{l}
\end{tikzcd}
\end{center}
So it is enough to show that, for $?\in \{1,2\}$, the following diagram is commutative
\begin{center}
\begin{tikzcd}
q_{?}^*g^{*}\mathcal M\arrow{r}\arrow{d} & g^{',*}\mathcal M\arrow{d}\\
q_{?}^*g^{*}R^if_{Ogus,*}\mathcal O_{Y/K}\arrow{r} & g^{',*}R^if_{Ogus,*}\mathcal O_{Y/K}
\end{tikzcd}
\end{center}
in $\Fisoc^{(p)}(U\times_X U)$ or, equivalently, in $\Isoc^{(p)}(U\times_X U)$.
Since $$(-_{(\mathfrak T(U\times_X U),z_{\mathfrak T(U\times_X U)})},\epsilon_{-,\mathfrak U\times_W \mathfrak U}):\Isoc^{(p)}(U\times_X U)\rightarrow \Strat^{(p)}(U\times_X U,\mathfrak U\times_W\mathfrak U)$$
is an equivalence of categories, it is enough to show that  
\begin{center}
\begin{tikzcd}
(q_{?}^*g^{*}\mathcal M)_{(\mathfrak T(U\times_X U),z_{\mathfrak T(U\times_X U)})}\arrow{r}\arrow{d} & (g^{',*}\mathcal M)_{(\mathfrak T(U\times_X U),z_{\mathfrak T(U\times_X U)})}\arrow{d}\\
(q_{?}^*g^{*}R^if_{Ogus,*}\mathcal O_{Y/K})_{(\mathfrak T(U\times_X U),z_{\mathfrak T(U\times_X U)})}\arrow{r} & (g^{',*}R^if_{Ogus,*}\mathcal O_{Y/K})_{(\mathfrak T(U\times_X U),z_{\mathfrak T(U\times_X U)})}
\end{tikzcd}
\end{center}
commutes.
Since $q_?:(U\times_X U,U\times_X U,\mathfrak U\times_{W} \mathfrak U)\rightarrow (U,U,\mathfrak U)$ is a morphism of frame over $(X',\overline X', \mathfrak X')$ that have ($P_{(X',\overline X', \mathfrak X')}$), by the property $(2)$ of $\mathcal M$ in \ref{shihosssss}, the morphism given by the isocrystals structure 
$$q_{?}^*R^if_{(Y_{U},Y_{U})/\mathfrak U, rig,*}\mathcal O^{\dagger}_{(Y_{U},Y_{U})}\rightarrow R^if_{(Y_{U\times_{X}U},Y_{U\times_{X}U})/\mathfrak U\times \mathfrak U, rig,*}\mathcal O^{\dagger}_{(Y_{U\times_{X}U},Y_{U\times_{X}U})}$$
is the natural base change map. Pulling back via $u_{\mathfrak T(U\times_X U)}$ one finds a commutative diagram (\ref{gigadiagram})
\begin{center}
\begin{tikzcd}
u_{\mathfrak T(U\times_X U)}^*q_{?}^*R^if_{(Y_{U},Y_{U})/\mathfrak U, rig,*}\mathcal O^{\dagger}_{(Y_{U},Y_{U})}\arrow{r}\arrow{d}{\simeq} & u_{\mathfrak T(U\times_X U)}^* R^if_{(Y_{U\times_{X}U},Y_{U\times_{X}U})/\mathfrak U\times_{W} \mathfrak U, rig,*}\mathcal O^{\dagger}_{(Y_{U\times_{X}U},Y_{U\times_{X}U})}\arrow{d}{\simeq}\\


q_{\mathfrak T(U\times_X U),?}^*u^*R^if_{(Y_{U},Y_{U})/\mathfrak U, rig,*}\mathcal O^{\dagger}_{(Y_{U},Y_{U})}\arrow{r}\arrow{d}{\simeq} & u_{\mathfrak T(U\times_X U)}^* R^if_{(Y_{U\times_{X}U},Y_{U\times_{X}U})/\mathfrak U\times_{W} \mathfrak U, rig,*}\mathcal O^{\dagger}_{(Y_{U\times_{X}U},Y_{U\times_{X}U})}\arrow{d}{\simeq}\\


q_{\mathfrak T(U\times_X U),?}^*R^if_{\mathfrak T(U)_1,\mathfrak T(U),crys,*}\mathcal O_{Y_{\mathfrak T(U)_1}/\mathfrak T(U)}\arrow{r} & R^if_{\mathfrak T(U\times_X U)_1,\mathfrak T(U\times_X U),crys,*}\mathcal O_{Y_{\mathfrak T(U\times_X U)_1}/\mathfrak T(U\times_X U)}
\end{tikzcd}
\end{center}
such that the horizontal morphism are the base change morphism. So the morphism
\begin{center}
\begin{tikzcd}
(q_{?}^*g^{*}\mathcal M)_{(\mathfrak T(U\times_X U),z_{\mathfrak T(U\times_X U)})}\arrow{r}\arrow{d}{\simeq} & (g^{',*}\mathcal M)_{(\mathfrak T(U\times_X U),z_{\mathfrak T(U\times_X U)})}\arrow{d}{\simeq}\\
q_{\mathfrak T(U\times_X U),?}^*R^if_{\mathfrak T(U)_1,\mathfrak T(U),crys,*}\mathcal O_{Y_{\mathfrak T(U)_1}/\mathfrak T(U)}\arrow{r} & R^if_{\mathfrak T(U\times_X U)_1,\mathfrak T(U\times_X U),crys,*}\mathcal O_{Y_{\mathfrak T(U\times_X U)_1}/\mathfrak T(U\times_X U)}
\end{tikzcd}
\end{center}
is induced by the base change morphism.
Since the same is true for $R^if_{Ogus,*}\mathcal O_{Y/K}$ by construction (\ref{Ogushigher}), this shows that the descent diagram \ref{descentdiagram} is commutative, hence $$\psi:g^*\mathcal M\simeq g^*R^if_{Ogus,*}\mathcal O_{Y/K}$$ gives an isomorphism in the category of descent data for the category $\Fisoc^{(1)}(U)$ of $U$ over $X$. 

By \etale and proper descent for convergent isocrystals (\cite[Theorems 4.5 and 4.6]{Ogus}), this implies that $\psi$ descends to an isomorphism 
$$\mathcal M\simeq R^if_{Ogus,*}\mathcal O_{Y/K} \text{ in } \Fisoc^{(1)}(X)$$
and concludes the proof of Theorem \ref{berthelotconjecture}.
\endproof

\begin{thebibliography}{99}
\bibitem[Amb17]{UOIpositive}
E. Ambrosi, A uniform open image theorem in positive characteristic, preprint available
\href{https://arxiv.org/pdf/1711.06132.pdf}{here}.
\bibitem[And96]{AndreIHES} Y. Andr\'e, 
Pour une th\'eorie inconditionnelle des motifs, Publ. Math. IHES 83, p. 5-49, 1996. 
\bibitem[Ber96]{berthelot} 
P. Berthelot, Cohomologie rigide et cohomologie ridige \'a supports propres, premi\'ere partie, preprint available \href{https://perso.univ-rennes1.fr/pierre.berthelot/publis/Cohomologie_Rigide_I.pdf}{here}.
\bibitem[Cad12]{MC} A. Cadoret, 
Motivated cycles under specialization, In "Groupes de Galois Géométriques et differentiels", P. Boalch and J.-M. Couveignes eds., Séminaires et Congrès, S.M.F., p. 25-55, 2012.
\bibitem[Cad17]{uniformadelic}
A. Cadoret, An open adelic image theorem for motivic representations over function fields, to appear in Math. Res. Lett.
\bibitem[CC18]{brauer} A. Cadoret and F. Charles,
A remark on uniform boundedness of Brauer, preprint available 
\href{https://arxiv.org/pdf/1801.07322.pdf}{here}.
\bibitem[CK16]{cadoretkret}
A. Cadoret, A.Kret, Galois generic points on Shimura varieties, Algebra and Number Theory 10, p. 1893-1934, 2016.
\bibitem[Chr18]{Chr} A. Christensen, Specialization of N\'eron-Severi groups in characteristic p, preprint available \href{http://math.mit.edu/~atticusc}{here}
\bibitem[Cre92]{Crew} R. Crew, F-isocrystals and their monodromy groups, Ann. Sci. Ecole Norm. Sup. 25, p.429–464, 1992.
\bibitem[D'Ad17]{marcuzzo}
M. D'Addezio, The monodromy groups of lisse sheaves and overconvergent F-isocrystals, preprint available
\href{https://arxiv.org/abs/1711.06669v3}{here}.
\bibitem[dJ]{djtate} A.J. de Jong, Tate conjecture for divisors, unpublished note.
\bibitem[dJ96]{alteration} A.J. de Jong, Smoothness, semi-stability and alterations. Publ. Math. I.H.E.S. 83, p. 51–93, 1996.
\bibitem[Del71]{HodgeII}
P. Deligne, Th\'eorie de Hodge, II, Publ. Math. IHES 40, p. 5–57, 1971.
\bibitem[Del80]{Weil2}
 P. Deligne,  La Conjecture de Weil, II, Publ. Math. IHES 52, p. 137-252, 1980.
\bibitem[ES93]{Trace}
Etesse, J.-Y., Le Stum, B.: Fonctions $L$ associèes aux $F$-isocristaux surconvergent I. Interprétation cohomologique, Math. Ann. 296, p. 557–576, 1993.
\bibitem[Il79]{derhamwitt} L.Illusie . Complexe de de Rham-Witt et cohomologie cristalline, Ann. Sci. Ecole 
Norm. Sup. 12, 501–661, 1979.
\bibitem[K79]{Katz} N. M. Katz,, Slope filtrations of F-crystals, Asterisque 63, p. 113–164, 1979
\bibitem[KM74]{KatzMessing}  N. M.Katz, W. Messing, Some consequences of the Riemann hypothesis
for varieties over finite fields, Invent. Math. 23, p. 73–77, 1974.
\bibitem[Ked04]{Kedfull}
K. S. Kedlaya, Full faithfulness for overconvergent F-isocrystals, in (refereed proceedings) Adolphson et al (eds.), Geometric Aspects of Dwork Theory (Volume II), de Gruyter (Berlin), p. 819-83, 2004.
\bibitem[Ked17]{notesked}
K. S. Kedlaya, Notes on isocrystals, preprint available \href{https://arxiv.org/abs/1606.01321v5}{here}, 2017.
\bibitem[LP90]{larsenepink} M. Larsen and R. Pink, Determining representations from invariant dimensions, Invent.
Math. 102, p. 377-398, 1990.
\bibitem[Laz16]{berthelotconjecture}
C. Lazda, Incarnations of Berthelot's conjecture, J. Number Theory 166, p. 137-157, 2016.
\bibitem[LeS07]{Lestum}
 B. Le Stum, Rigid cohomology, Cambridge Tracts in Mathematics, Cambridge University Press, 2007.
\bibitem[LF15]{Leifu} L.Fu, Etale Cohomology Theory, Nankai Tracts in Mathematics, 2015.
\bibitem[MP12]{poonen}
D. Maulik, B. Poonen, Neron-Severi groups under specialization, Duke Mathematical Journal 161, 2012.
\bibitem[Mil80]{milne}
J.S. Milne, \'Etale cohomology, Princeton Mathematical Series, Princeton University Press, 1980.
\bibitem[Mo77]{Mori}
S. Mori, On Tate conjecture concerning endomorphisms of Abelian varieties, International symposium of Algebraic Geometry, Kyoto, p. 219-230, 1977.
\bibitem[Mor13]{Notmor}
M. Morrow,  A note on higher direct images in crystalline cohomology, appendix to A Variational Tate Conjecture in crystalline cohomology, to appear in Journal of the European Mathematical Society.
\bibitem[Mor15]{Varmor}
M. Morrow,  A Variational Tate Conjecture in crystalline cohomology, to appear in Journal of the European Mathematical Society.
\bibitem[Og84]{Ogus}
A. Ogus, $F$-isocrystals and de Rham cohomology II - Convergent isocrystals, Duke Math. J. 51, p. 765–850, 1984.
\bibitem[Ser89]{Serreweil}
J. P. Serre, Lectures on the Mordell-Weil theorem, Aspects of Mathematics, Friedr. Vieweg and Sohn, 1989.
\bibitem[Sh08a]{ShihoI}
A.Shiho, Relative log convergent cohomology and relative rigid cohomology I,  preprint available
\href{https://arxiv.org/pdf/0707.1742.pdf}{here}.
\bibitem[Sh08b]{Shiho}
A.Shiho, Relative log convergent cohomology and relative rigid cohomology II,  preprint available
\href{https://arxiv.org/pdf/0707.1743.pdf}{here}.
\bibitem[SGA1]{EGA} Revetements \'etales et groupe fondamental, Lecture Notes in Mathematics, Springer-
Verlag, Berlin-New York, 1971, Seminaire de Geometrie Algebrique du Bois Marie 1960–1961, Dirige par Alexandre Grothendieck. Augmente de deux exposes de M. Raynaud. 
\bibitem[SGA2]{SGA2} A.Grothendieck, Cohomologie locale des faisceaux coherents et theoremes de
Lefschetz locaux et globaux. North-Holland Publishing Co., Amsterdam, Masson and Cie, 1968.
\bibitem[SGA4]{SGA4} Theorie des topos et cohomologie \etale des schemas. Tome 3, Seminaire de Geometrie Algebrique du Bois-Marie 1963–1964. Dirige par M. Artin, A. Grothendieck et J. L. Verdier. Avec la collaboration de P. Deligne et B. Saint-Donat. Lecture Notes in Mathematics, Springer, 1973.
\bibitem[SGA7]{SGA7} P. Deligne and N. Katz, Groupe de monodromie en g\'eom\'etrie alg\'ebrique (SGA7) - 2\'eme partie, Lecture Notes in Math., Springer-Verlag, Berlin-New York, 1973.
\bibitem[SP]{stacks-project}
The Stacks Project Authors, Stacks Project, available at \href{http://stacks.math.columbia.edu}{http://stacks.math.columbia.edu}
\bibitem[Ta65]{tateconj} J. Tate, Algebraic cohomology classes and poles of zeta functions, in Schilling, O. F. G., Arithmetical Algebraic Geometry (Proc. Conf. Purdue Univ., 1963), New York: Harper and Row, p. 93–110, 1965.
\bibitem[Ta94]{Tate} J. Tate, Conjectures on algebraic cycles in $\ell$-adic cohomology, in Motives (Seattle, WA, 1991), Proc. Sympos. Pure Math., 55, Part 1, Amer. Math. Soc., p. 71–83, 1994.
\bibitem[VAV16]{VAV} A. Varilly-Alvarado, B. Viray, Abelian $n$-division fields of elliptic curves and Brauer groups of product of Kummer and abelian surfaces, Forum of Mathematics, Sigma, to appear.
\end{thebibliography}
\vskip 1\baselineskip
 \textit{emilianomaria.ambrosi@gmail.com}\\
Centre de Math\'ematiques Laurent Schwartz - Ecole Polytechnique,\\
91128 PALAISEAU, FRANCE.
\end{document}