Variational inequalities for the Ornstein--Uhlenbeck semigroup: the higher--dimensional case
Extracted source text · arXiv 2405.00497
%%rewieved version for JGA (November 9 2024)
%%Versione rivista caricata su ArXiv: abbiamo modificato la definizione di F_\varrho a pag.11.
\documentclass[reqno,12pt,a4paper]{amsart}
\usepackage{pgfplots}
\usepackage[english]{babel}
\pdfoutput=1
\usepackage{amsmath,amsthm,amssymb,amsfonts, mathdots}
\usepackage{scrtime, cancel}
\usepackage{enumitem, color}
%%\usepackage[notref, notcite]{showkeys}
%\usepackage[notref, notcite]{showkeys}
\definecolor{pinegreen}{rgb}{0.0, 0.47, 0.44}
\definecolor{brilliantrose}{rgb}{1.0, 0.33, 0.64}
\usepackage{mathtools}
\usepackage{ifpdf}
\ifpdf
\usepackage[backref]{hyperref}
\else
\usepackage[hypertex]{hyperref}
\fi
\usepackage{pdfsync,verbatim}
\usepackage{color}
\mathtoolsset{showonlyrefs}
\numberwithin{equation}{section} %%numera le equazioni sezione per sezione
\allowdisplaybreaks
\newtheorem{theorem}{Theorem}[section]
\newtheorem{lemma}[theorem]{Lemma}
\newtheorem{proposition}[theorem]{Proposition}
\newtheorem{corollary}[theorem]{Corollary}
\newtheorem{definition}[theorem]{Definition}
\newtheorem{claim}[theorem]{Claim}
\newtheorem{observation}[theorem]{Observation}
\theoremstyle{definition}
\newtheorem{remark}[theorem]{Remark}
\usepackage{graphicx}
\makeatletter
\newcommand*\bigcdot{\mathpalette\bigcdot@{.5}}
\newcommand*\bigcdot@[2]{\mathbin{\vcenter{\hbox{\scalebox{#2}{$\m@th#1\bullet$}}}}}
\makeatother
\addtolength{\headheight}{3.2pt} %% leave room for symbol in header
\textwidth=15,2cm
\textheight=22cm
\setlength{\oddsidemargin}{15.5pt}
\setlength{\evensidemargin}{15.5pt}
\newcommand{\R}{\mathbb{R}}
\newcommand{\C}{\mathbb{C}}
\newcommand{\N}{\mathbb{N}}
\newcommand{\Z}{\mathbb{Z}}
\newcommand{\ellipses}{E}
\def\misgausskd{d \gamma_\infty}
\def\misgaussk{\gamma_\infty}
\DeclareMathOperator{\tr}{tr} % trace
\newcount\dotcnt\newdimen\deltay
\def\Ddot#1#2(#3,#4,#5,#6){\deltay=#6\setbox1=\hbox to0pt{\smash{\dotcnt=1
\kern#3\loop\raise\dotcnt\deltay\hbox to0pt{\hss#2}\kern#5\ifnum\dotcnt<#1
\advance\dotcnt 1\repeat}\hss}\setbox2=\vtop{\box1}\ht2=#4\box2}
\def\succvarepsilon{{\underline{\varepsilon}}}
\newcommand\nucleozero{ \mathcal{M}_{0} }
\newcommand\nucleoinf{ \mathcal{M}_{\infty} }
\newcommand{\dbtilde}[1]{\accentset{\approx}{#1}}
\newcommand{\vardbtilde}[1]{\widetilde{\raisebox{0pt}[0.85\height]{$\widetilde{#1}$}}}
\newcommand\nucleorieszfirstorder{\mathcal{R}_j }
\newcommand\nucleorieszsecondorder{\mathcal{R}_{ij}}
\newcommand\nucleorieszsecondorderloc{\mathcal{R}_{ij,0}}
\newcommand\nucleorieszsecondorderinf{\mathcal{R}_{ij,\infty}}
\newcommand\nucleorieszfirstorderloc{\mathcal{R}_{j,0} }
\newcommand\nucleorieszfirstorderinf{\mathcal{R}_{j,\infty} }
\newcommand{\Red}{\color{red}}
\def\Blue{\color{blue}}
\def\Black{\color{black}}
\def\Yellow{\color{yellow}}
\def\Green{\color{pinegreen}}
\def\Rose{\color{brilliantrose}}
\pgfplotsset{compat = newest}
% per omettere virgola prima dell'ultimo autore in amsart
\makeatletter
\def\author@andify{%
\nxandlist {\unskip ,\penalty-1 \space\ignorespaces}%
{\unskip {} \@@and~}%
{\unskip \penalty-2 \space \@@and~}%
}
\makeatother
\title[Variation for Ornstein--Uhlenbeck: the higher--dimensional case]{
Variational inequalities \\ for the Ornstein--Uhlenbeck semigroup: \\the higher--dimensional case}
\author{Valentina Casarino}
\address{DTG, Universit\`a degli Studi di Padova\\ Stradella san Nicola 3 \\I-36100 Vicenza \\ Italy}
\email{valentina.casarino@unipd.it}
\author{Paolo Ciatti}
\address{Dipartimento di Matematica "Tullio Levi Civita", Universit\`a degli Studi di Padova\\Via Trieste, 63, 35131 Padova, \\ Italy}
\email{paolo.ciatti@unipd.it}
\author{Peter Sj\"ogren}
\address{Mathematical Sciences, University of Gothenburg and Mathematical Sciences,
Chalmers University of Technology \\ SE - 412 96 G\"oteborg, Sweden}
\email{peters@chalmers.se}
\keywords{Variation seminorm,
Ornstein--Uhlenbeck semigroup,
Mehler kernel, vector-valued Calder\'on--Zygmund theory}
\subjclass[2000]{
47D03, %%% - Groups and semigroups of linear operators
42B20, %% - Singular and oscillatory integrals (Calderón-Zygmund, etc.)
42B35, %% Function spaces arising in harmonic analysis
42B99%% - Harmonic analysis in several variables
}
\thanks{The first and second authors are members of the Gruppo Nazionale per l'Analisi Matematica, la Probabilità e le loro Applicazioni (GNAMPA)
of the Istituto Nazionale di Alta Matematica (INdAM) and were partially supported by Progetto GNAMPA 2024: ``$L^p$ estimates for
singular integrals in nondoubling settings". The third author would like to thank the University of Padova for its generosity during his visits.}
%This research was carried out while the third author was GNAMPA Professore Visitatore at the University of Padova, Italy. The third author also profited from a grant from the Adlerbert Research Foundation in Sweden.}
%\address{Universit\`a degli Studi di Padova\\Stradella san Nicola 3 \\I-36100 Vicenza \\ Italy}\email{valentina.casarino@unipd.it}
\date{\today}
\begin{document}
\maketitle
\begin{abstract}
We study the $\varrho$-th order variation seminorm of a general Ornstein--Uhlenbeck semigroup $\left(\mathcal H_t\right)_{t>0}$ in $\R^n$, taken with respect to $t$.
We prove that this seminorm defines an operator of weak type $(1,1)$ with respect to the invariant measure when $\varrho> 2$. For large $t$,
one has
an enhanced version of the standard weak-type $(1,1)$ bound.
For small $t$, the proof hinges on vector-valued Calder\'on--Zygmund techniques in the local region, and
on the fact that
the $t$ derivative of the integral kernel of $\mathcal H_t$ in the global region has a bounded number of zeros in $(0,1]$.
A counterexample is given for $\varrho= 2$;
in fact, we
prove that the second order variation seminorm of $\left(\mathcal H_t\right)_{t>0}$,
and therefore also the $\varrho$-th order variation seminorm for any $\varrho\in [1,2)$,
is not of strong nor weak type $(p,p)$ for any $p \in [1,\infty)$ with respect to the invariant measure.
\end{abstract}
\section{Introduction}
In this paper we prove
the weak type $(1,1)$ for the variation operator of a general Ornstein--Uhlenbeck semigroup $\big( \mathcal H_t\big)_{t> 0}$ in $\R^n$ for any $n\ge 1$.
Recently the authors proved
this result in dimension one \cite{CCS6}, answering a question asked by Almeida et al.\ in \cite[p.\,31]{Almeida}.
In this article we provide an answer to the question in \cite{Almeida}
covering the higher--dimensional case.
The methods we use are different from those of \cite{CCS6}, which seem hard to adapt to the case $n>1$.
The proof relies on extensive application of properties of the Ornstein--Uhlenbeck semigroup and its integral kernel $K_t(x,u)$ (known as the Mehler kernel),
recently proved by the authors in a series of papers concerning various issues arising from harmonic analysis
(see \cite{CCS1, CCS2, CCS3, CCS4, CCS5, CCS6} and also the brief introductory summary \cite{CCS7}). We also apply vector--valued Calder\'on--Zygmund theory.
As far as we know, the first application of this theory in this context was in \cite{Harboure2, Harboure} (see also \cite{Rubio},
where \cite{Harboure2, Harboure} have their roots, and \cite{Crescimbeni}).
Let $\big(T_t\big)_{t> 0}$
be a family of bounded operators between spaces of functions defined on a measure space $(X,\mu)$. %% defined on $\R^n$.
The $\varrho$-th order variation operator of $\big(T_t\big)_{t> 0}$ on an interval $I\subset \R_+$ is defined,
for $1 \le \varrho < \infty$ and suitable $f$, by
\begin{equation}\label{def:intro_rho2}
\|T_t\,f(x)\|_{v(\varrho), I} = \sup\, \left( \sum_{i=1}^N |T_{t_i} f(x)- T_{t_{i-1}}f(x)|^\varrho \right)^{1/\varrho},\qquad x\in X,
\end{equation}
where the supremum is taken over all finite, increasing
sequences $\left(t_i \right)_0^N$ of points in $I$.
In the last fifty years,
variational inequalities, stating that the $L^p (\mu)$ norm of \eqref{def:intro_rho2}
is bounded by a constant times the $L^p(\mu)$ norm of $f$,
have been %%vastly
widely investigated. The first bounds, due to D. Lépingle \cite{Lepingle} for bounded martingales and to
V. F.~Gaposhkin \cite{Gaposhkin1, Gaposhkin2} and J. Bourgain \cite{Bourgain} for ergodic averages,
have been generalized in various directions.
We refer to \cite{CCS6} for a brief description of recent results
with a focus on harmonic analysis; see \cite{Seeger, Campbell, Jones, Jones1, Jones1-higher, Jones2, Jones4, Harboure, Betancor1, Betancor2, Betancor3, {AlmeidaMedit}, Ma1, Ma2}. %%focused
More recently,
the study of oscillatory and jump inequalities,
in addition
to variational inequalities,
began to develop from both an analytic and a number--theoretic perspective; we refer in particular to
\cite{Bourgain_etal, Mirek1, Mirek2, Mirek3, Mirek4, Mirek5, Mirek6}.
For an overview of the connections
with ergodic theory, analytic number theory, and harmonic analysis, see in particular \cite{Krause}.
In this paper
the family $\big(T_t\big)_{t> 0}$ in \eqref{def:intro_rho2} will always be the Ornstein-Uhlenbeck semigroup $\big( \mathcal H_t\big)_{t> 0}$ in $\R^n$.
This is the semigroup generated by the elliptic operator
\begin{equation}\label{def:opL}
\mathcal L=
\frac12
\mathrm{tr}
\big( Q\nabla^2 \big)+\langle Bx, \nabla \rangle,
\end{equation}
called the Ornstein-Uhlenbeck operator.
Here $\nabla$ is the gradient and $\nabla^2$ the Hessian matrix. Moreover,
$Q$ is a real, symmetric and positive definite $n\times n$
matrix, called the covariance of $\mathcal L$, and
$B $ is a real $n\times n$ matrix whose eigenvalues have negative real parts;
$B $ indicates the drift of $\mathcal L$.
In Section \ref{preliminaries}
we will provide explicit expressions for $\mathcal H_t$, seen as an integral operator with a kernel $K_t(x,u)$.
It is well known that there exists an invariant measure under the action of $\mathcal H_t$, unique up to a positive factor. This measure, denoted by $\gamma_\infty$, is basic in Ornstein-Uhlenbeck theory; it is described explicitly in Section \ref{preliminaries}.
In 2001 Jones and Reinhold \cite{Jones1} proved that for $\varrho > 2$ the variation operator
of any symmetric
diffusion semigroup is $L^p$ bounded for $1<p<\infty$. Ten years later Le Merdy and Xu \cite{Le Merdy}
extended this result to a nonsymmetric context. In fact, Corollary 4.5 in \cite{Le Merdy}
applies to
$\mathcal H_t$ (see \cite[p. 31]{Almeida} for a discussion). It says that the operator given by
$f \mapsto \| \mathcal H_t f(x)\|_{v(\varrho),\Bbb R_+}$,
is bounded on $L^p (\gamma_\infty)$ for $1<p<\infty$ and $\varrho > 2$.
Our result gives the corresponding weak type $(1,1)$, as follows.
\begin{theorem}\label{thm}
For each $\varrho > 2$ the operator that maps $f \in L^1(\gamma_\infty)$ to the function
\begin{equation*} %\label{norm-op}
\| \mathcal H_t f(x)\|_{v(\varrho),\Bbb R_+}, \quad x \in \mathbb R^n,
\end{equation*}
where the $v(\varrho)$ seminorm is taken in the variable $t$, is of weak type $(1,1)$ with respect to the measure $\gamma_\infty$.
\end{theorem}
In other words, the inequality
\begin{equation} \label{main-ineq}
\gamma_\infty
\{x\in\R^n : \| \mathcal H_t f(x)\|_{v(\varrho),\Bbb R_+}
> \alpha\} \le \frac{C}\alpha\,\|f\|_{L^1( \gamma_\infty)}, \qquad \alpha>0,
\end{equation}
holds for some $C > 0$ and all functions $f\in L^1 (\gamma_\infty)$.
The pointwise estimates of the Mehler kernel and its derivative needed to prove Theorem \ref{thm}
are quite different for small and large $t$. Thus we are led to
distinguish between the variation in the interval $0 < t \le 1$ and that in $1 \le t < \infty$.
The variation in $[1,\infty)$ is treated following a geometric approach
recently developed by the authors in \cite{CCS1, CCS2, CCS3, CCS5}, which relies on a system of polar-like coordinates. Proposition \ref{t>1} is actually an enhanced version of \eqref{main-ineq} with $\Bbb R_+$ replaced by
$[1,\infty)$.
The study of the variation in $(0,1]$ is more delicate and
requires a further distinction between local and global parts of the Ornstein-Uhlenbeck semigroup operator.
In the Ornstein--Uhlenbeck setting,
the local part is usually defined by the relation $|x-u| \lesssim 1/(1+|x|)$ between the two variables of the Mehler kernel.
This splitting was first introduced by
Muckenhoupt \cite{Mu} in one dimension, and by the third author \cite{Sj} in higher dimension. It has since been widely used in the literature, in particular in \cite{Perez}. In this paper, P\'erez shows that the local parts of many operators related to the Ornstein--Uhlenbeck semigroup behave precisely like the corresponding classical operators in Euclidean space.
The reason for this definition of the local part is that it makes the density of the measure $\gamma_\infty$ have the same order of magnitude at the two points $x$ and $u$. But this remains true if $x$ and $u$ belong to a suitable elliptic ring of width roughly $1/(1+|x|)$. In this paper, we will define the local part by splitting $\R^n$
into rings of this type. This makes the arguments more explicit; for details see Section \ref{local1}. The expression ``local part" is not quite adequate here, since
$x$ and $u$ may be far apart, but we prefer to keep it, since this part plays the same role in the proofs as it does in earlier work.
To prove \eqref{main-ineq} for the global part and $0 < t \le 1$, we estimate its kernel using a method from
\cite{CCS5}. It is
based on the observation that the
number of zeros in $(0,1]$ of the function $t \mapsto \partial_t K_t(x,u)$
is bounded, uniformly in $(x,u)\in \mathbb R^n\times\mathbb R^n$.
The argument for
the local part and $0 < t \le 1$ is based on vector--valued Calder\'on--Zygmund techniques for singular integrals
from \cite{Harboure2, Harboure, Crescimbeni}. This requires a transition to Lebesgue measure.
The parameter $\varrho$ deserves more attention.
Some of our results hold for $\varrho\ge 1$ as well. This is the case of the variational bounds in $[1,\infty)$
and also in $(0,1]$ for
the global part; we refer to Theorem \ref{t>1} and Theorem \ref{t<1,global}, respectively.
What forces the restriction $\varrho>2$ in Theorem
\ref{thm} is the variational bound
for
the local part of $\mathcal H_t$ when $t\in (0,1]$;
see Theorem \ref{locsmallt_v2}, where we apply results from \cite{Le Merdy} holding only for $\varrho> 2$.
In the last section, we provide a counterexample showing that the condition $\varrho>2$ is necessary in Theorem \ref{thm}.
\medskip
\subsection{Structure of the paper}
We gather some known facts about the Ornstein--Uhlenbeck semigroup and its integral kernel $K_t$ in Section~\ref{preliminaries}; in particular,
we give pointwise bounds for $K_t$ and its time derivative {$\dot K_t = \partial_t K_t$}.
In Section~\ref{variation} some basic properties of the variation seminorm are discussed; here
we also introduce a few reductions which will simplify the proof of Theorem~\ref{thm}.
Section \ref{t large}
is devoted to the proof of the weak type inequality \eqref{main-ineq} with $\R_+$ replaced by $[1,+\infty)$.
The proof of \eqref{main-ineq}
{\ for $(0,1]$} is carried out in Sections~\ref{local1}, \ref{The variation of the global part} and \ref{tsmall}. More precisely, in Section \ref{local1} we describe our localization procedure and the local and global parts, and Section~\ref{The variation of the global part} deals with the global part. The local part is treated in Section
\ref{tsmall}. The discussione of a technical issue, arising in the local context and involving Bochner integral, is postponed to the Appendix (Section \ref{Appendix}).
In Section \ref{counterexample} we conclude by showing
that
Theorem \ref{thm} does not hold for $\varrho=2$.
\subsection{Notation} We will use the symbols $0<c,\,C<\infty$ to denote constants, not necessarily equal at different occurrences. These constants will depend
only on $n$, $Q$ and $B$, unless otherwise explicitly stated.
If $a$ and $b$ are positive quantities, $a \lesssim b$ or
equivalently $b \gtrsim a$ means $a \le C b$. If both $a \lesssim b$ and $b \lesssim a$ hold,
we shall write $a \simeq b$.
By $\N$ we denote the set of all nonnegative integers.
We write \[\dot K_t = \partial_t K_t,\] that is, we adopt the dot notation for differentiation with respect to the variable $t$.
\section{The Mehler kernel}\label{preliminaries}
{{The Ornstein--Uhlenbeck semigroup may be formally written as
$
\mathcal H_t=e^{t\mathcal L}$,
${t> 0}$, with $\mathcal L$ given by \eqref{def:opL}. In order to give more explicit expressions for this semigroup,
we introduce the
positive definite, symmetric matrices
\begin{equation} \label{defQt}
Q_t=\int_0^t e^{sB}Qe^{sB^*}ds, \qquad \text{ $0<t\leq +\infty$},
\end{equation}
and the
normalized Gaussian measures $\gamma_t $ in $\R^n$, with $t\in (0,+\infty]$, whose densities
with respect to Lebesgue measure $dx$ are given by
\[
x\mapsto (2\pi)^{-\frac{n}{2}}
(\text{det} \, Q_t)^{-\frac{1}{2} }
\exp\left({-\frac12 \langle Q_t^{-1}x,x\rangle}\right).
\]
Here the case $t=+\infty$ gives the invariant measure $\gamma_\infty$ that appears
already in the introduction.
%%(we point out that, with a slight abuse of notation, the symbol $\gamma_\infty$ will denote both the function $x\mapsto (2\pi)^{-\frac{n}{2}}(\text{det} Q_\infty)^{-\frac{1}{2} }
%%\exp\big({-\frac12 \langle Q_\infty^{-1}x,x\rangle}\big)$ on $\R^n$ and the measure with this density %%$\gamma_\infty$ with respect to$dx$). DA CANCELLARE, SUGGERISCO
Kolmogorov's formula states that
for all bounded and continuous functions in $\R^n$ one has
\begin{equation}\label{Kolmo}
\mathcal H_t
f(x)=
\int
f(e^{tB}x-y)\,d\gamma_t (y)\,, \qquad x\in\R^n,\;\; t>0.
\end{equation}
Starting from \eqref{Kolmo}, one may write $\mathcal H_t $ as an integral operator
with a (Mehler) kernel $K_t(x,u)$.
We point out that the term kernel in this paper refers to integration with respect to our basic measure $ \gamma_\infty$, with an exception in Section \ref{tsmall}.
In fact, we saw in \cite{CCS2} that for each $f\in L^1(\gamma_\infty)$ and all $t>0$ one has
\begin{align} \label{def-int-ker}
\mathcal H_t
f(x) &=
\int
K_t
(x,u)\,
f(u)\,
d\gamma_\infty(u)
\,, \qquad x\in\R^n,
\end{align}
where for $x,u\in\R^n$ and $t>0$ the Mehler kernel $K_t$ is given by
\begin{align} \label{mehler} %\label{defKRt}
K_t (x,u)\!
=\!
\Big(
\frac{\det \, Q_\infty}{\det \, Q_t}
\Big)^{{1}/{2} }
e^{R(x)}\,
\exp \Big[
{-\frac12
\left\langle \left(
Q_t^{-1}-Q_\infty^{-1}\right) (u-D_t \,x) \,, u-D_t\, x\right\rangle}\Big]\!.\;\;\;\;
\end{align}
Here
$R(x)$ is
the quadratic form
\begin{equation*}%%%%%%\label{quadratic-form}
R(x) ={\frac12 \left\langle Q_\infty^{-1}x ,x \right\rangle}, \qquad\text{$x\in\R^n$},
\end{equation*}
and
\begin{equation}\label{def:Dt}
D_t =
Q_\infty \, e^{-tB^*} Q_\infty^{-1} ,
\end{equation}
which is a one-parameter group, defined for all $t \in\R$.
It will be convenient to introduce another norm on $\R^n$ by
\[
|x|_Q := | Q_\infty^{-1/2}\,x|, \qquad x \in \R^n.
\]
Then
$|x|_Q \simeq |x|$,
and $R(x) = |x|_Q^2/2 , \;\,\text{$x\in\R^n$}$.
Observe that the density of $\gamma_\infty$ is proportional to $ e^{-R(x)}$.
\subsection{Pointwise estimates for the Mehler kernel}
We shall repeatedly use the following pointwise estimate of the Mehler kernel. For $0<t\leq 1$ and
all $(x,u) \in \mathbb R^n\times \mathbb R^n$ one has
\begin{equation}\label{litet}
%%%%% \frac{ e^{R( x)}}{t^{n/2}}\exp\left(-C\,\frac{|u-D_t \,x |^2}t\right)\,\lesssim\,
K_t(x,u)
\,\lesssim \, \frac{ e^{R( x)}}{t^{n/2}} \exp\left(-c\,\frac{|u-D_t\, x |^2}t\right).
\end{equation}
This is proved in \cite[(3.4)]{CCS2}.
%%and
%%%\begin{equation} \label{tstort}
%%e^{R(x)}\exp \Big[ -C\left| D_{-t}\,u- x\right|_Q^2 \Big]\,\lesssim\,
%%%K_t (x,u)\,\lesssim\,e^{R(x)}\exp \Big[-\frac12 \left| D_{-t}\,u- x\right|_Q^2\Big],\end{equation}holding for $t\geq 1$
%%%(see \cite[(2.11)]{CCS2}).
We shall need estimates for the time derivative of the Mehler kernel as well.
For an explicit expression of $\dot K_t$, the reader is referred to \cite[Lemma 4.2]{CCS5}; we only recall here the following pointwise bounds for $\dot K_t$
(see \cite[(5.5) and (5.4)]{CCS5}):
\begin{equation}\label{dotKeps}
|\dot K_t(x,u)|\lesssim
\, e^{R(x)}\,t^{-n/2}\,
\exp{\left(-c\,\frac{|u- D_{t}\,x|^2}t \right)}\,
\left(t^{-1}+|x| t^{-1/2}\right),%%% \lesssim e^{R(x)}\,(1+|x|)\,t^{-n/2-1}
\end{equation}
for $0<t \le 1$, and
\begin{equation}\label{dotK1} |\dot K_t (x,u)| \lesssim \, e^{R(x)}\,\exp{\left(-c\,| D_{-t}\,u- x|^2\right)}\,\left(|D_{-t}\,u| + e^{-ct}\right) \end{equation}for $t\geq1$.
Both formulae hold
for all $(x,u) \in \mathbb R^n\times \mathbb R^n$.
\section{The variation seminorm}\label{variation}
To state some basic properties of the variation seminorm, we define it for a generic continuous
real-valued function $\phi$ defined in an interval $I$, by setting for any $1 \le \varrho < \infty$
%Let $\phi$ be a real- or complex-valued function defined in an interval $I$. Then for $1 \le \varrho < \infty$ we introduce the variation seminorm $v(\varrho)$, given by
%To be more precise, if $1 \le \varrho < \infty$ and the real-valued function $\phi(t)$ is defined in an interval $I$, its variation norm is defined as
\begin{equation} \label{def_var_gen}
\|\phi\|_{v(\varrho), I} = \sup \left( \sum_{i=1}^{N} |\phi(t_i) - \phi(t_{i-1})|^\varrho \right)^{1/\varrho}.
\end{equation}
As in \eqref{def:intro_rho2}, the supremum is taken over all finite, increasing sequences $\left(t_i \right)_0^N$ of points in $I$.
This seminorm vanishes only for constant functions.
Observe that the seminorm
$\|.\|_{v(\varrho), I}$ is decreasing in $\varrho$ for $1 \le \varrho < \infty$.
It is also subadditive in $I$, in the following sense. Take an inner point $\tau$ of $I$ and set $I_+ = I \cap [\tau, +\infty)$ and $I_- = I \cap (-\infty, \tau]$. Then for $1 \le \varrho < \infty$ and any $\phi$
\begin{equation*}
\|\phi \|_{v(\varrho), I} \le \|\phi \|_{v(\varrho), I_+} + \|\phi \|_{v(\varrho), I_-}.
\end{equation*}
If $\phi \in C^1(I)$ and $\phi' \in L^1(I)$, then for $1 \le \varrho < \infty$
%$\phi \in V(\varrho,I)$ and
\begin{equation}\label{var}
\|\phi \|_{v(\varrho), I} \le \int_{I} |\phi'(t)|\,dt,
\end{equation}
see \cite[Lemma 2.1]{CCS6}.
In the last section, we will consider a discrete version of the variation, obtained by replacing $I$ in the definition \eqref{def_var_gen}
by a set $\mathcal I$ which is the intersection of $\N_+$ and an interval, and where $\phi$ is defined.
%Then it is natural to replace $t$ by $$.
For $\varrho = 2$ one has the simple estimate
\begin{equation}\label{discrete_trivial}
\|\phi\|_{v(2), \mathcal I} \lesssim \left( \sum_{\ell\in \mathcal I} \phi(\ell)^2 \right)^{1/2}.
\end{equation}
For future convenience, we note that
%, if $I$ denotes either $(0,1]$ or $[1,\infty)$, then
a combination of
\eqref{var}
and \cite[Lemma 5.3]{CCS5} yields that for any $\varrho \in [1,\infty)$, any interval $I \subset \R_+$ and all $x\in\R^n$
\begin{align} \label{important}
\|\mathcal H_tf(x)\|_{v(\varrho), I}&\le %\int_I \left| \frac{\partial}{\partial t} H_tf(x)\right| dt \notag
\int_I \left|\frac{\partial}{\partial t} \int K_t(x,u) f(u)\,d\gamma_\infty (u) \right|\,dt\notag
=\int_I \left| \int \dot K_t(x,u) f(u) \,d\gamma_\infty (u) \right|\,dt \notag \\
&\le \int \int_I \big| \dot K_t(x,u)\big| \,dt \, |f(u)|\, d\gamma_\infty (u),\qquad f\in L^1(\gamma_\infty).
\end{align}
\subsection{Simplifications}\label{Simplificationsss}
By means of a few
observations, it is possible to simplify the proof of Theorem~\ref{thm}.
First of all, when proving the inequality \eqref{main-ineq} one may take
$f $ such that
$\|f\|_{L^1( \misgaussk)}=1$. As a consequence, $\alpha$ in the same estimate may be assumed large, for instance $\alpha >2$, since $\gamma_\infty$ is finite.
\smallskip
Moreover, as already observed in the study of the weak-type $(1,1)$ of operators related to the Ornstein--Uhlenbeck semigroup
(see \cite[Section 5]{CCS2}),
when proving \eqref{main-ineq}
we mostly need to take into account only points $x$ belonging to
the ellipsoidal annulus
\begin{equation*} %\label{crown}
{\mathcal C_\alpha}=\left\{
x \in\R^n:\, \frac12 \log \alpha\le
R(x)
\le 2 \log \alpha
\right\}.
\end{equation*}
Indeed, the unbounded component of the complement of $\mathcal C_\alpha$ can always be neglected, since
%%%5\begin{equation}\label{est_tless1}
$\gamma_\infty\left( \left\{
x \in\R^n:\,
R(x)
> 2 \log \alpha
\right\}\right)\lesssim 1/\alpha$.
%\end{equation}
For the variation of $\mathcal H_t$ in $[1,+\infty)$,
the bounded component of the complement of $\mathcal C_\alpha$
is negligible as well. Indeed, we see from \cite[(5.4) and Lemma 5.1]{CCS5}
that \begin{equation}\label{ineq100}
\int_1^\infty
|\dot K_t (x,u)|\,
dt
\lesssim e^{R(x)}.
\end{equation}
Combined with \eqref{important}, where $I=[1,\infty)$ and $f$ is normalized in $L^1( \gamma_\infty)$,
this leads to
\begin{align*} % \label{important_two}
\|\mathcal H_tf(x)\|_{v(\varrho), [1,+\infty)}&\lesssim e^{R(x)} \lesssim \sqrt{\alpha}
\end{align*}
for $R(x) < \frac12 \log \alpha$.
Taking $\alpha$ suitable large, we will have
%\begin{equation}\label{est_tgreater1}
$\|\mathcal H_tf(x)\|_{v(\varrho), [1,+\infty)}
% \lesssim \sqrt{\alpha}
< \alpha$ in the bounded component of the complement of $\mathcal C_\alpha$.
\medskip
\section{The variation of $\mathcal H_tf(x)$ in $[1,+\infty)$}\label{t large}
We first introduce the adapted polar coordinates from \cite{CCS2}.
Fix $\beta>0$ and consider the ellipsoid
\begin{equation*}
\ellipses_\beta
=\{x\in\R^n:\, R(x)=
\beta\}
\,.
\end{equation*}
As a consequence of \cite[formula (4.3)]{CCS2},
the map
$s\mapsto R(D_s z)$ is strictly increasing for each $0 \ne z\in\R^n$.
Thus we may write any $x\in\R^n,\, x\neq 0$, uniquely as
\begin{equation*} %\label{def-coord}
x=D_s \tilde x
%= \big(e^{\lambda_j s}\tilde x_j, x_{\text{loc}}\big)\,, \,\, j=1,\ldots,k
\,,
\end{equation*}
for some $\tilde x\in \ellipses_\beta$
and $s\in\R$, and $s$ and $\tilde x$ are
our polar coordinates of $x$.
We recall two results, previously proved by the authors, which will be essential in our arguments.
The following lemma is \cite[Lemma 5.1]{CCS3} with $\sigma = 1$; the factor $1/4$ occurring in \cite{CCS3} is replaced by a generic $\delta > 0$, which causes no problem.
\begin{lemma}\label{preliminary-t-large-not-kernel}
\cite{CCS3}
% \cite[Lemma 5.1 with $\sigma = 1$]{CCS3} %%%and Remark 5.5]{CCS3}
Let $\delta > 0$. For $x,u\in\R^n$, one has
\begin{equation*}
\int_1^{+\infty}\exp \Big({-\delta\left| D_{-t}\,u- x
\right|^2 }\Big)\big| D_{-t} \, u\big|\, dt \lesssim
1, %\label{stima-5-inf}
\end{equation*}
where the implicit constant may depend on $\delta$, $n$, $Q$ and $B$.
\end{lemma}
%%For a proof of the second lemma we refer to \cite[Lemma 6.2]{CCS5}
%%%(see also \cite[Lemma 7.2]{CCS2} for an earlier version).
\begin{lemma}\label{bounding-GMAa} \cite[Lemma 7.2]{CCS3}, \cite[Proposition 6.1]{CCS2}
Let $\delta > 0$ and $\alpha > 2$, and assume $f$ normalized in $L^1 (\gamma_\infty)$. Then
%the $d\gamma_\infty$ measure of the level set
\[
\gamma_\infty \Big\{ x = D_s\,\tilde x \in\mathcal C_\alpha:\:
e^{R(x)}
\int
\exp
\big(
- \delta\, \big|
\tilde x-\tilde u\big|^2\big)
\,|f(u)|\,\misgausskd
(u)
>
\alpha
\Big\}
\lesssim \frac{1}{\alpha\,\sqrt{\log \alpha}}.
\]
%is no larger than $C/\left(\alpha\,\sqrt{\log \alpha}\right)$.
\end{lemma}
Here the polar coordinates are defined with $\beta = \log \alpha$, the implicit constant is as in Lemma \ref{preliminary-t-large-not-kernel}, and $\sigma = 1$ in \cite[Lemma~7.2]{CCS3}.
We can now deduce the bound for the variation operator in $[1,+\infty)$.
\begin{theorem} \label{t>1}
Let $1\le \varrho <\infty$. The operator that maps $f \in L^1(\gamma_\infty)$ to the function
\begin{equation*}
\| \mathcal H_t f(x)\|_{v(\varrho), [1,+\infty)}, \quad x \in \mathbb R^n,
\end{equation*}
is of weak type $(1,1)$ with respect to the measure $\gamma_\infty$.
In fact, the following stronger result holds: if $\|f\|_{L^1( \gamma_\infty)} = 1$, then
\begin{equation} \label{stronger}
\gamma_\infty
\left\{x\in\R^n: \| H_t f(x)\|_{v(\varrho),[1,\infty)}
> \alpha \right\} \lesssim \frac{1}{\alpha \sqrt{\log \alpha}}, \qquad \alpha > 2.
\end{equation}
\end{theorem}
\begin{proof}
This proof is similar to that of Proposition 6.1 in \cite{CCS5}; we sketch it for the sake of completeness.
Let $f$ be normalized in $L^1(\gamma_\infty)$.
In the light of the considerations made in Subsection \ref{Simplificationsss},
it suffices to consider $\alpha$ large and $x \in\mathcal C_\alpha$.
Using polar coordinates with $\beta = \log\alpha$
for $x\in\mathcal C_\alpha$ and $u\ne 0$,
we write $x=D_s\,\tilde x$ and
$u=D_\sigma\,\tilde u$, with $s,\sigma\in\R$.
Since $R(x)\ge \beta/2$, \, \cite[Lemma 4.3{\it{(i)}}]{CCS2}
yields
\begin{align*}
|D_{-t}\,u-x| &= |D_{\sigma-t}\,\tilde u-D_s\, \tilde x|
\gtrsim |\tilde x-\tilde u|, \qquad
x \in\mathcal C_\alpha.
\end{align*}
By reducing the value of $c$ in the exponential factor in \eqref{dotK1},
we arrive at
\begin{equation*} %\label{dotK11}
|\dot K_t (x,u)| \lesssim
e^{R(x)}
\,
\exp
\Big(
-c\,
|\tilde x-\tilde u|^2\Big)
\,\exp{\left(-c\,| D_{-t}\,u- x|^2\right)}\,(|D_{-t}\,u| + e^{-ct}), \qquad t>1.
\end{equation*}
%%%Changing the value of $c$ here, we may neglect the factors $|D_{-t}\,u|$.
%%(compare with \cite[formula (5.4)]{CCS5}).
An application of Lemma \ref{preliminary-t-large-not-kernel} leads to
\begin{equation*} %\label{vv}
\int_1^\infty
|\dot K_t (x,u)|
dt
\lesssim e^{R( x)}
\,
\exp\big(- c\,\big|\tilde x-\tilde u\big|^2\big), \qquad
x \in\mathcal C_\alpha,
\end{equation*}
and then \eqref{important} yields
\begin{align*} %\label{importantvv}
\|\mathcal H_tf(x)\|_{v(\varrho), [1,\infty)} \lesssim e^{R( x)}
\,\int
\exp\big(- c\,\big|\tilde x-\tilde u\big|^2\big) \,|f(u)|\,\misgausskd.
\end{align*}
Finally, Lemma \ref{bounding-GMAa} implies \eqref{stronger}, and the weak type $(1,1)$ also follows.
\end{proof}
\begin{remark}
Theorem \ref{t>1} says that for the variation in $[1,\infty)$
the standard weak type $(1,1)$ estimate is enhanced by a logarithmic factor.
This phenomenon was observed in the one-dimensional case in \cite{CCS6}.
\end{remark}
\section{Local versus global region}
\label{local1}
From now on, we shall focus on the variation of $\mathcal H_t$ in the interval $(0,1]$.
%study of weak type $(1,1)$ for small $t$.
This case requires a further distinction between the local and the global parts of $\mathcal H_t$.
We start describing the localization procedure.
\subsection{\color{black}{Splitting of $\R^n$ into rings}}\label{subs1}
Let
\begin{align}\label{def:rings}
R_j =
\{x\in \R^n : j\le R(x)\le j+1\}, \qquad j =0,1,\ldots.
\end{align}
These rings cover $\R^n$ and are pairwise disjoint except for boundaries.
We also set $R_j =
\emptyset$ if $j<0$.
The width of $R_j$, defined as the $|\cdot|_Q$ distance between the two components of its boundary, is for
$j \ge 1$
\begin{equation} \label{width}
\sqrt 2\left(\sqrt{j+1}-\sqrt j\right) =\frac{\sqrt 2}{\sqrt{j+1}+\sqrt j}
\in \left(\frac1{2\sqrt{j}}, \frac{\sqrt 2}{\sqrt{j}}\right),
\end{equation}
as easily verified.
We take a sequence of smooth non-negative functions $(r_j)_{j\in\N}$
satisfying $\sum_{j=0}^\infty r_j (x)=1$ for all $x$, with $\text{supp}\,r_j\subseteq R_j\cup R_{j+1}$.
%% (with $R_j$ defined by \eqref{def:rings}).
Further, we may choose them such that
\begin{equation}\label{nabla-r}
\big|\nabla r_j (x)\big|\lesssim {1+|x|}.
\end{equation}
We also introduce slightly larger functions $ \widetilde{r}_j \in\mathcal C_0^\infty (\mathbb R^n)$
taking values in $[0,1]$ such that
$\widetilde{r_j} = 1 $ in $\cup_{\nu=j-1}^{j+2} R_{\nu}$ and
$\text{supp}\,\widetilde{r_j}\subset \cup_{\nu=j-2}^{j+3} R_{\nu}$.}
They also satisfy
\begin{equation}\label{nabla2}
\big|\nabla \widetilde{r}_j (x)\big|\lesssim {1+|x|}.
\end{equation}
Then the supports of these $\widetilde{r_j}$ have bounded overlap,
and
\begin{equation}\label{densityinring}
e^{-R(x)}\simeq e^{-j} \qquad \text{for} \qquad
x\in \text{supp}\,\widetilde{r_j}.
\end{equation}
\subsection{The splitting of $\mathcal H_t $}
We can now split the operator $\mathcal H_t f$ in a local and a global part, in a way adapted to the rings. % and $\gamma_\infty$.
The local part is defined by
\begin{align*} %\label{loc_def}
\mathcal H_t^{\mathrm{loc}} f(x)
:=& \, \sum_{j=0}^\infty \widetilde r_j (x)
\mathcal H_t\big(f r_j\big)(x)\\
= & \, \sum_{j=0}^\infty \widetilde r_j (x)
\int_{\R^n} K_t(x,u)\, f(u)\, r_j (u)\,
d\gamma_\infty(u).
\notag
\end{align*}
This sum is locally finite and thus well defined for any $f \in L^1(\gamma_\infty)$,
because of the bounded overlap of the sets $\text{supp}\:\widetilde{r_j}$. %, \;j=0,1,\dots$.
Setting
\begin{equation*} %\label{def:eta}
\eta(x,u)=\sum_{j=0}^\infty
\widetilde{r_j} (x) \,r_j (u),
\end{equation*}
we get
\begin{align}\label{kernelHloc}
\mathcal H_t^{\mathrm{loc}} f(x)& = \int_{\R^n} K_t(x,u)\, \eta (x,u)\, f(u)\,
d\gamma_\infty(u),
\end{align}
and $0 \le \eta(x,u)\le 1$ for all $x,u\in\R^n$.
Moreover, if $\mathrm{supp}\, f \subset R_j\cup R_{j+1}$, then the support of $\mathcal H_t^{\mathrm{loc}} f $ is contained in $\cup_{\nu=j-2}^{j+4} R_{\nu}$.
The global part of $\mathcal H_t$ is defined by $ \mathcal H_t^{\mathrm{glob}} = \mathcal H_t - \mathcal H_t^{\mathrm{loc}}$, or equivalently
\begin{equation*} %\label{def_7sett_nucleo_gl}
\mathcal H_t^{\mathrm{glob}} f(x)= \int_{\R^n} K_t(x,u)\, (1- \eta (x,u))\, f(u)\,
d\gamma_\infty(u), \qquad x \in \R^n.
\end{equation*}
\smallskip
The next two lemmas give sufficient conditions for the kernel $K_t(x,u)\, (1- \eta (x,u))$ of the global part to vanish.
\begin{lemma} \label{uno}
Let $j \in \{0,1,\dots\}$.
If $x,\,u\in R_j \cup R_{j+1}$, then $\eta(x,u) = 1$.
\end{lemma}
\begin{proof}
When $u\in R_j \cup R_{j+1}$ the function $r_i(u)$ can be nonzero only for $i\in\{j-1,j,j+1\}$, and then
$r_{j-1}(u)+r_j(u)+r_{j+1}(u)=1$.
Also, for $x\in R_j \cup R_{j+1}$ one has ${\widetilde r}_{j-1} (x)={\widetilde r}_{j} (x)={\widetilde r}_{j+1} (x) = 1$.
This implies that %for $x,u\in R_j \cup R_{j+1}$
%\begin{equation*} %\label{rj}
$\eta(x,u)=\sum_{\nu =j-1}^{j+1} \widetilde{r_\nu} (x)\, r_\nu(u) =1$.
%\end{equation*}
\end{proof}
\begin{lemma} \label{due-molt}
If
\begin{equation*}
|x-u|_Q\le \frac1{2(1+|x|_Q)},
\end{equation*}
then $\eta(x,u) = 1$.
\end{lemma}
\begin{proof}
Suppose $x\in R_j$. Then $|x|_Q \ge \sqrt{2j}$ and we can write
\begin{equation*}
|x-u|_Q\le \frac1{2(1+|x|_Q)} \le \frac1{2(1+\sqrt{j})} \le \frac1{2\sqrt{j+1}}.
\end{equation*}
From this and \eqref{width}, we see that $|x-u|_Q$ is less than the widths of $R_{j-1}$ and $R_{j+1}$ (only that of $R_{j+1}$ if $j=0$ or $j=1$).
Since $x \in R_j$, this means that both $x$ and $u$ must be in $ R_{j-1} \cup R_j$
or $ R_j \cup R_{j+1}$. From Lemma \ref{uno} we then get the assertion.
\end{proof}
We will need an estimate for the gradient of $\eta(x,u)$. If a point $(x,u)$ is in the support of $\eta$, then $x$ and $u$ are both in the support of some
$\widetilde{r}_j$. It follows that $1+|u| \simeq 1+|x|$, and \eqref{nabla-r} and \eqref{nabla2} then
imply that
\begin{equation}\label{gradeta}
|\nabla_x\, \eta(x,u)| + |\nabla_u\, \eta(x,u)| \lesssim 1+|x|.
\end{equation}
\medskip
\section{The variation of the global part}\label{The variation of the global part}
The result of this section is the following.
\begin{theorem} \label{t<1,global}
For each $\varrho \ge 1$ the operator that maps $f \in L^1(\gamma_\infty)$ to the function
\begin{equation*}
\|{ \mathcal H}_t^{\mathrm{glob}} f(x)\|_{v(\varrho), (0,1]}, \quad x \in \mathbb R^n,
\end{equation*}
is of weak type $(1,1)$ with respect to the measure $\gamma_\infty$.
\end{theorem}
\begin{proof}
%In order to prove the weak type $(1,1)$ of $ \mathcal H_t^{\mathrm{glob}} $ for $t\in (0,1]$ w
This proof follows the pattern from that of \cite[Proposition 10.2]{CCS5}. In particular, we will
need the following result % from \cite{CCS5}
concerning a maximal operator.
\begin{theorem}\label{stima per M con t piccolo}
The maximal operator defined by
\begin{align*}
S_{0}^{{\rm{glob}}}
f(x) = \int \sup_{0<t\leq 1}K_t(x,u)\,
\big(1-\eta(x,u)\big) \,|f(u)|\, d\gamma_\infty(u)
\end{align*}
is
of weak type $(1,1)$ with respect to the invariant measure $\gamma_\infty$.
%Here $\eta$ is the function defined in \eqref{def:eta}.
\end{theorem}
This statement seems to coincide with that of \cite[Theorem 10.1]{CCS5}, but
our $\eta(x,u)$ is not the same as that in \cite{CCS5}. However, by tracing the proof in \cite{CCS5},
one sees that what matters is only the implication
\begin{equation*} %\label{ineq_eta_1}
\eta(x,u)<1 \qquad \Rightarrow \qquad
|x-u|_Q > \frac1{2(1+|x|_Q)},
\end{equation*}
which is a consequence of our Lemma \ref{due-molt}. In this way, Theorem \ref{stima per M con t piccolo} follows.
\vskip5pt
We will also need
%We need an estimate of the number of zeros of the function $t \mapsto \dot K_t(x,u)$ for $0 < t < 1$. From
\cite[Proposition 9.1]{CCS5}. It says
that for any $(x,u)\in \mathbb R^n\times\mathbb R^n$,
the number of zeros $N(x,u)$ in $(0,1]$ of the map $t \mapsto \dot K_t(x,u)$ is bounded
by a positive integer $\bar N$ that depends only on $n$ and $B$.
As in \cite[Proposition 10.2]{CCS5},
we denote these zeros by $t_1(x,u)< \cdots< t_{N(x,u)}(x,u)$,
and set $t_0(x,u) =0$, \hskip4pt
$t_{N(x,u)+1}(x,u) =1$.
Since $K_t(x,u)$ vanishes at $t=0$ if $x\ne u$,
it follows from the fundamental theorem of calculus that
\begin{multline*}
\int_0^1 \left|\dot K_t(x,u)\right| dt
%&= \sum_{i=0}^{N(x,u)-1} \int_{t_i(x,u)}^{t_{i+1}(x,u)} |\dot K_t(x,u)|\, dt
= \sum_{i=0}^{N(x,u)} \left|\int_{t_i(x,u)}^{t_{i+1}(x,u)} \dot K_t(x,u) dt\right|
\\
= \sum_{i=0}^{N(x,u)} \left|K_{t_{i+1}(x,u)}(x,u) - K_{t_i(x,u)}(x,u)\right|
\\
\leq 2
\sum_{i=0}^{N(x,u)+1}K_{t_i(x,u)}(x,u)
%\leq 2(\bar N+1) \sup_{0<t\leq 1}K_{t}(x,u)
\;\lesssim \;\bar N
\sup_{0<t\leq 1}K_{t}(x,u).
\end{multline*}
By \eqref{important}, which remains valid with the extra factor $1-\eta(x,u)$, we obtain
\begin{equation*} %\label{33}
\| \mathcal H_t^{\mathrm{glob}} f(x)\|_{v(\varrho), (0,1]} \lesssim
\int \sup_{0<t\leq 1}K_t(x,u)\,(1-\eta(x,u)\, |f(u)|\,d\gamma_\infty(u).
\end{equation*}
Now Theorem \ref{stima per M con t piccolo} implies the assertion of Theorem \ref{t<1,global}.
\end{proof}
\vskip15pt
\section{The local part for small $t$}\label{tsmall}
In this section, we prove the following result. %, concerning $\mathcal H_t^{\rm{loc}}$.
\begin{theorem} \label{locsmallt}
For each $\varrho > 2$ the operator that maps $f \in L^1(\gamma_\infty)$ to the function
\begin{equation*}
\|\mathcal H_t^{\mathrm{loc}} f(x)\|_{v(\varrho),(0,1]}, \quad x \in \mathbb R^n,
\end{equation*}
is of weak type $(1,1)$ with respect to the measure $\gamma_\infty$.
\end{theorem}
The proof
relies on vector-valued Calder\'on--Zygmund theory.
\subsection{Preparations}
Part of our notation in this section follows that of \cite{Crescimbeni}.
%In fact, as in \cite{Crescimbeni},
Let $\Theta$ be the set of all finite, increasing sequences in $(0,1]$ written
\begin{equation*}
\succvarepsilon=(\varepsilon_i)_0^N
\end{equation*}
for some $ N = N(\succvarepsilon)\in \N_+.${\Blue{
%Set $\N_+ = \{1,2,\dots\}$.
%Let $F$ be the vector space of all functions $g: \N_+\times \Theta \to \R$ such that
%$g(i,\succvarepsilon) = 0$ for $i > N(\succvarepsilon)$.
}}
Then we let $F$ be the vector space of all functions
\begin{equation*}
g: \{(i,\succvarepsilon) \in \N \times \Theta: 1 \le i \le N(\succvarepsilon)\} \to \R.
\end{equation*}
For $\varrho \in [1,\infty)$ we define $ F_\varrho$ as that subspace of $ F$ consisting of those $g $ for which the mixed norm
\begin{equation}
\|g\|_{ F_\varrho}:=
\sup_{\succvarepsilon \in\Theta}
\left(\sum_{i=1}^{N(\succvarepsilon)}
|g(i,\succvarepsilon)|^\varrho\right)^{1/\varrho}
\end{equation}
is finite. Then $ F_\varrho$
is a normed space and a Banach space.
We shall work with the operator $V$ that maps real-valued functions $f\in L^1(\gamma_\infty)$ to $F$-valued functions defined in $\R^n$
and is given by
\begin{align*}
Vf(x)(i, \succvarepsilon) = \mathcal H_{\varepsilon_{i}}f(x)-
\mathcal H_{\varepsilon_{i-1}}f(x), \qquad i = 1,\dots,N(\succvarepsilon ), \;\; \succvarepsilon \in \Theta.
\end{align*}
Similarly, we set
\begin{align} \label{Vloc}
V^{\text{loc}}f(x)(i, \succvarepsilon) = \mathcal H^{\text{loc}}_{\varepsilon_{i}}f(x)-
\mathcal H^{\text{loc}}_{\varepsilon_{i-1}}f(x), \qquad i = 1,\dots,N(\succvarepsilon),
\end{align}
and analogously for $V^{\text{glob}}f$.
Then for $\varrho \in [1,\infty)$
\begin{equation} \label{equivalence}
\|Vf(x)\|_{F_\varrho}=\| \mathcal H_t f(x)\|_{v(\varrho), (0,1]},
\end{equation}
and similar equalities hold with superscripts $\text{loc}$ or $\text{glob}$.
Theorem \ref{locsmallt} can now be rewritten in the following equivalent way.
\begin{theorem} \label{locsmallt_v2}
For each $\varrho > 2$ the operator that maps $f \in L^1(\gamma_\infty)$ to the function
\begin{equation*}
x\mapsto\|V^{\rm{loc}}f(x)\|_{F_\varrho}\,, \qquad x\in\R^n,
\end{equation*}
is of weak type $(1,1)$ with respect to the measure $\gamma_\infty$.
\end{theorem}
The advantage of Theorem
\ref{locsmallt_v2} is that the proof can be based on vector-valued Calder\'on--Zygmund theory.
To apply this machinery, we will pass to Lebesgue measure,
and set
\begin{equation}\label{deftildeV:oper}
\widetilde V^{\text{loc}}\,g(x)=
e^{-R(x)}\,
V^{\text{loc}}\big ( g(\cdot)\, e^{R(\cdot)}\,\big)(x).
\end{equation}
Lebesgue measure will be written either $dx$ or $du$.
The following proposition %% \ref{prop81}
clarifies the connection between
$\widetilde V^{\text{loc}}$ and $V^{\text{loc}}$.
\begin{proposition}\label{prop81}
Let $\varrho>2$. If
the operator that maps $g \in L^1(du)$ to the function
\begin{equation*}
x\mapsto \left\|\widetilde V^{\rm{loc}}\,g(x)\right\|_{F_\varrho}, \quad x \in \mathbb R^n,
\end{equation*}
is of weak type $(1,1)$ with respect to Lebesgue measure,
then the operator that maps $f \in L^1(\gamma_\infty)$ to the function
\begin{equation*}
x\mapsto \|V^{\rm{loc}}f(x)\|_{F_\varrho}\,, \quad x \in \mathbb R^n,
\end{equation*}
is of weak type $(1,1)$ with respect to $\gamma_\infty$.
\end{proposition}
\begin{proof}
We have
\begin{align*} %\label{split}
\|V^{\text{loc}}f(x)\|_{F_\varrho}
= &\, \left\| \sum_{j=0}^\infty \widetilde {r_j} (x)\, V (fr_j)(x)\right\|_{F_\varrho}
\\ \le & \,\left\|\sum_{j=0}^\infty \widetilde r_j(x)\,V^{\text{glob}} (fr_j)(x)\right\|_{F_\varrho} +
\left\| \sum_{j=0}^\infty \widetilde r_j(x)\,V^{\text{loc}} (fr_j)(x)\right\|_{F_\varrho} = I_{\text{glob}} + I_{\text{loc}}.
\end{align*}
Since the $\widetilde {r_j}$ have supports with bounded overlap,
the sums here are uniformly locally finite. It follows that
\begin{align*}
\| I_{\text{glob}} \|_{L^{1,\infty}(\gamma_\infty)} \lesssim \sum_{j=0}^\infty \left\|\widetilde r_j(x)\,\|V^{\text{glob}} (fr_j)(x)\|_{F_\varrho}\, \right\|_{L^{1,\infty}(\gamma_\infty)}.
\end{align*}
The $F_\varrho$ norm here equals $\| \mathcal H_t^{\text{glob}} (fr_j)(x)\|_{v(\varrho), (0,1]}$. After estimating the factor $\widetilde r_j(x)$ by 1, we can apply Theorem
\ref{t<1,global} and get
\begin{align*}
\| I_{\text{glob}} \|_{L^{1,\infty}(\gamma_\infty)} \lesssim \sum_{j=0}^\infty \left\|fr_j \right\|_{L^{1}(\gamma_\infty)} \simeq \|f\|_{L^{1}(\gamma_\infty)}.
\end{align*}
We estimate the $L^{1,\infty}(\gamma_\infty)$ quasinorm of $II$ similarly, but then apply the definition of
$\widetilde V^{\text{loc}}$. This gives
\begin{align*}
\| I_{\text{loc}} \|_{L^{1,\infty}(\gamma_\infty)} \lesssim & \sum_{j=0}^\infty \left\| \widetilde r_j(x)\,\|V^{\text{loc}} (fr_j)(x)\|_{F_\varrho} \right\|_{L^{1,\infty}(\gamma_\infty)} \\
\simeq &
\sum_{j=0}^\infty \left\| \widetilde r_j(x)\,e^{R(x)}\left\|\widetilde V^{\text{loc}} (fr_je^{-R(\cdot)})(x)\right\|_{F_\varrho} \right\|_{L^{1,\infty}(\gamma_\infty)} \\
\simeq & \sum_{j=0}^\infty \left\| \widetilde r_j(x)\,\left\|\widetilde V^{\text{loc}} (fr_je^{-R(\cdot)})(x)\right\|_{F_\varrho} \right\|_{L^{1,\infty}(dx)};
\end{align*}
%% %%%E' con \simeq perché nel supporto di \widetilde r_j la densità di \gamma_infty è \simeq e^-j \simeq e^-R(x) dunque essenzialmente costante
in the last step here, we switched
from $\gamma_\infty$ to Lebesgue measure. This was possible because of \eqref{densityinring}. Next, we estimate $\widetilde r_j(x)$ by 1 and use the hypothesis of the proposition, getting
\begin{align*}
\| I_{\text{loc}} \|_{L^{1,\infty}(\gamma_\infty)} \lesssim \sum_{j=0}^\infty \left\|fr_j e^{-R(\cdot)}\right\|_{L^{1}(du)} \simeq \|f\|_{L^{1}(\gamma_\infty)}.
\end{align*}
The proposition is proved.
\end{proof}
%
\subsection{Vector-valued Calder\'on--Zygmund operators}\label{CalderonZygmund}
This theory will be applied to % $\widetilde V^{\text{loc}}$.
the operator $\widetilde V^{\text{loc}}$. The following proposition gives the strong $(p,p)$ bound needed.
%Then we shall verify that the kernel of the operator fulfills standard
\begin{proposition}\label{prp2}
For any $\varrho>2$ and any $1<p<\infty$, the operator that maps $g \in L^p(du)$ to the function
\begin{equation*}
x\mapsto \left\|\widetilde V^{\text{loc}}\,g(x)\right\|_{F_\varrho}, \qquad x \in \mathbb R^n,
\end{equation*}
is bounded from $L^p(du)$ to $L^{p} (dx )$.
\end{proposition}
\begin{proof}
As a consequence of the definitions of $\widetilde V^{\text{loc}}$ and $ V^{\text{loc}}$ and the bounded overlap of the $\widetilde{r_j}$,
one has
\begin{align*} % \label{important_local}
\left\|\, \left\|\widetilde V^{\text{loc}}\,g(x)\right\|_{F_\varrho}\,\right\|_{L^p(dx)}^p
& =
\int \left\|
\sum_{j=0}^\infty
\widetilde r_j (x) \,e^{-R(x)}\,
V\left(
g \,r_j\, e^{R(\cdot)} \right)(x)\right\|_{F_\varrho}^p \,dx\\
&\lesssim
\sum_{j=0}^\infty
\int \Big\| \widetilde r_j (x) e^{-R(x)}\,
V\left ( g\, r_j \,e^{R(\cdot) }\right) (x)
\Big\|_{F_\varrho}^p\, dx\\
&\simeq \sum_{j=0}^\infty e^{-jp} \int
\Big\|
\widetilde{r}_j (x)\,
V\left ( g\, r_j \,e^{R(\cdot) }\right) (x)
\Big\|_{F_\varrho}^p \,dx\\
&\simeq
\sum_{j=0}^\infty e^{j(1-p)} \int
\Big\|
\widetilde{r}_j (x)\,
V\left ( g\, r_j \,e^{R(\cdot) }\right) (x)
\Big\|_{F_\varrho}^p \,d\gamma_\infty(x).
\end{align*}
Notice that we also used \eqref{densityinring} here.
In the last expression, we delete the factor $\widetilde r_j (x)$ and observe that the resulting $F_\varrho$ seminorm equals
$ \left\|\mathcal H_t \left ( g\, r_j \,e^{R(\cdot) }\right) (x)\right\|_{v(\varrho), (0,1]}$.
As mentioned in the introduction, the variation operator for $\mathcal H_t$ is bounded on $L^p(\gamma_\infty)$ with $1<p<\infty$, for $\varrho > 2$ \cite{Le Merdy}. We therefore obtain
\begin{align*}
\left\|\, \left\|\widetilde V^{\text{loc}}\,g(x)\right\|_{F_\varrho}\,\right\|_{L^p(dx)}^p
& \lesssim \sum_{j=0}^\infty e^{j(1-p)} \int
\left\|\mathcal H_t \left( g\, r_j \,e^{R(\cdot) }\right) (x)
\right\|_{v(\varrho), (0,1]}^p d\gamma_\infty(x)
\\ & \lesssim
\sum_{j=0}^\infty e^{j(1-p)}\left\| g\, r_j \,e^{R(\cdot)} \right\|_{L^p(\gamma_\infty)}^p
\simeq \sum_{j=0}^\infty \| g\, r_j \, \|_{L^p(du)}^p \le \big\|g\big\|^p_{L^p(du)},
\end{align*}
proving the assertion.
\end{proof}
\vskip5pt
We shall now prove the appropriate standard estimates for the vector-valued kernel of
$\widetilde V^{\text{loc}}$.
From \eqref{Vloc} and \eqref{kernelHloc}, we see that
$V^{\text{loc}}$ is an integral operator with an $F$-valued
kernel $ M^{\text{loc}} (x,u)$ given by
\begin{equation*} %\label{def:nucleo_CZ}
M^{\text{loc}}(x,u) (i,\succvarepsilon) =\big(
K_{\varepsilon_{i}}(x,u)-K_{\varepsilon_{i-1}} (x,u)\big)\, \eta(x,u), \qquad
i = 1,\dots,N(\succvarepsilon).
\end{equation*}
This means that for $f \in L^1(\gamma_\infty)$
\begin{equation}\label{formula1_Bochner}
V^{\text{loc}} f(x)=\int M^{\text{loc}} (x,u) f(u)\,d\gamma_\infty (u).
\end{equation}
We will need the
kernel $\widetilde M^{\rm{loc}} (x,u)$ of $\widetilde V^{\text{loc}}$, for integration against Lebesgue measure in the sense that
\begin{equation}\label{formula2_Bochner}
\widetilde V^{\text{loc}}\,f(x)=
\int \widetilde M^{\rm{loc}}(x,u)\,f(u)\,du
\end{equation}
for suitable $f$.
\begin{remark}
Notice that formulae \eqref{formula1_Bochner} and \eqref{formula2_Bochner} make sense in $F$ for all $x\in\R^n$, since the integrals may be computed
coordinatewise, that is, at one $(i,\epsilon) $ at a time.
In the Appendix (see Section \ref{Appendix}) it will be shown that
for $x\notin {\text{supp}}\,f$
the integral in \eqref{formula2_Bochner} is an $F_\varrho$-Bochner integral, and thus
$\widetilde V^{\text{loc}}\,f(x) \in F_\varrho$.
\end{remark}
From \eqref{deftildeV:oper} we get
\begin{multline}\label{def:Pcalli}
\widetilde M^{\rm{loc}} (x,u)(i,\succvarepsilon)=
e^{-R(x)}\,
M^{{\rm{loc}}} (x,u)(i,\succvarepsilon)
= e^{-R(x)}\,
\big(K_{\varepsilon_{i+1}}(x,u)-K_{\varepsilon_{i}}(x,u)\big)\,\eta(x,u), \\ \qquad\qquad
i = 1,\dots,N(\succvarepsilon).
\end{multline}
In analogy with \eqref{equivalence}, this implies
\begin{align} \label{equiv}
\|\widetilde M^{\rm{loc}}(x,y) \|_{F_\varrho}
&=\,e^{-R(x)}\,
\|K_t(x,u) \,\eta(x,u)\|_{v(\varrho), (0,1]}\notag
\\&= e^{-R(x)} \,\eta(x,u)\,
\|K_t(x,u)\|_{v(\varrho), (0,1]},
\end{align}
where the variation is meant with respect to $t$.
We will need an auxiliary result.
\begin{proposition} % \label{fromCCS5} %
\label{lemma-integral_dt}
Let $p,\: r\ge 0$ with $p+ r/2 > 1 $. Assume that $\eta(x,u)\neq 0$ and $x\neq u$. Then for $\delta>0$
\begin{equation*} %\label{propclaim}
\int_0^{1}
t^{-p} \exp\left(-\delta\,\frac{|u- D_t \,x |^2}t\right) |x|^{r} \, dt\lesssim C\,{|u-x|^{-2p-r+2}}.
\end{equation*}
Here the constant $C$ may depend on $\delta,\: p$ and $r$, in addition to $n$, $Q$ and $B$.
\end{proposition}
\begin{proof}
The statement of this proposition is similar to that of \cite[Proposition 8.3]{CCS5}, which is based on
\cite[Lemma 8.1]{CCS3}. But our function $\eta$ is not the same as there, and we must verify that
for $\eta(x,u)\neq 0$ and $0<t<1$
\begin{equation}\label{occorre}
\frac{|D_t\,x - u|^2}{t} > c \,\frac{|x - u|^2}{t} +c t |x|^2-C.
\end{equation}
Once this is established, we can follow the argument for \cite[Lemma 8.1]{CCS3} and finish the proof.
To verify \eqref{occorre}, assume first that
$t < c_0 |x-u|/|x|$ for a small $c_0 >0$ to be chosen. Then \cite[Lemma~2.3]{CCS3} implies
$|D_t\,x - x| \le C t |x| < Cc_0 |x-u| $ and thus
\begin{equation*}
|D_t\,x - u| \ge |x - u| - |D_t\,x - x| > |x - u| - Cc_0 |x-u|.
\end{equation*}
So with $c_0 >0$ small enough we get
\begin{equation*}
|D_t\,x - u| \, \ge \frac12
\, |x - u| \,
\gtrsim\, |x - u| +t |x|.
\end{equation*}
To obtain \eqref{occorre} in this case, take squares and divide by $t$.
In the opposite case $c_0 |x-u|/|x| \le t < 1$, we use \cite[Lemma 4.1]{CCS2} to conclude that
\begin{equation*}
\frac \partial{\partial t}\, | D_t\,x|_Q = \frac \partial{\partial t}\, \sqrt{2R( D_t\,x)}
\simeq | D_t\,x|_Q \simeq | x|_Q.
\end{equation*}
Integration yields
\begin{equation*}
| D_t\,x|_Q - | x|_Q \simeq t | x|_Q\,
\gtrsim\, t | x|_Q + |x-u|.
\end{equation*}
Thus
\begin{align}
|D_t\,x - u|_Q &\ge | D_t\,x|_Q - | u|_Q = | D_t\,x|_Q - | x|_Q + | x|_Q - | u|_Q\notag
\\
&\ge ct | x|_Q + c|x-u| - \big|| x|_Q - | u|_Q\big|.\label{nuovo}
\end{align}
Since $\eta(x,u) > 0$, there exists a $j \in \mathbb N $ such that such that $\widetilde r_j (x) > 0$ and $ r_j (u) > 0$.
Thus $\big|| x|_Q - | u|_Q\big|
\lesssim 1/\left(1+\sqrt j\right) \simeq \frac 1{1 +|x|}$,
%%Prima era: {\Rose{$\big|| x|_Q - | u|_Q\big| \lesssim 1/(1+|x|)$ since $\eta(x,u)\neq 0$}}
and by squaring \eqref{nuovo} we get
\begin{align*}
|D_t\,x - u|_Q^2 \ge &\, ct^2 | x|_Q^2 + c|x-u|^2 - C(t | x|_Q +|x-u|) /(1+|x|)\\
\ge &\, ct^2 | x|_Q^2 +c|x-u|^2 - Ct. %\ge c|x-u|^2 + c t^2 |x|^2- Ct.
\end{align*}
This implies \eqref{occorre}.
\end{proof}
\vskip7pt
\begin{proposition}\label{lemma-Calderon-no}
(\rm a)
For all $(x,u)$ such that $x\neq u$ %and $\eta(x,u)\neq 0$,the following estimates hold:\medskip
$$
\|\widetilde M^{\rm{loc}}(x,u) \|_{F_\varrho}\lesssim \frac 1{|x-u|^{n}} .
$$
%% $\big|\mathcal{R}_{j} (x,u)\%\eta (x,u)\big| \lesssim e^{R( x)}\,{|u-x|^{-n}}$;
\medskip
(\rm b)
If the points $x,\:u$ and $u'$ satisfy $|x-u| > 2 |u-u'|$, then
$$
\| \widetilde M^{\rm{loc}}(x,u) - \widetilde M^{\rm{loc}}(x,u')\|_{F_\varrho}
\lesssim \frac{|u-u'|}{ |x-u|^{n+1}}.
$$
%%$\big|\nabla_x \big(\mathcal{R}_{j} (x,u)\\eta (x,u)\big)\big| \lesssim { e^{R( x)}\,}{|u-x|^{-(n+1)}}.$$
%\medskip
%(c)
% $ \|\nabla_u \widetilde M^{\rm{loc}}(x,u) \|_{F_\varrho} \lesssim |x-u|^{-n-1}$, \medskip
The implicit constants in this proposition depend only on $n$, $B$ and $Q$.
\end{proposition}
\begin{proof}
\hskip4pt (a) \hskip4pt
Starting from \eqref{equiv}, we use \eqref{var} and then \eqref{dotKeps} to get
\begin{align*} %\label{varMzero1}
\|\widetilde M^{\rm{loc}}(x,u) \|_{F_\varrho}
& = e^{-R(x)}\, \eta(x,u)\, \| \ K_t(x,u)\|_{v(\varrho), (0,1]}
\le e^{-R(x)}\, \int_0^1 \big| \dot K_t(x,u)\big|\,dt \\
& \lesssim
\int_0^1 t^{-\frac n2} \exp\left(- c\, \frac{|u-D_t\,x |^2}t\right)\,
\left(\frac{1}{t}
+\frac{|x|}{\sqrt t}
\right)\, dt.
\end{align*}
Since $\eta(x,u)\neq 0$,
Proposition
\ref{lemma-integral_dt}
tells us that the last integral is bounded
by $ C |u-x|^{-n}$,
and (a) follows.
\medskip
\noindent (b) \hskip4pt
We start almost as in (a), and find
\begin{align*}
\| \widetilde M^{\rm{loc}}(x,u) - \widetilde M^{\rm{loc}}(x,u')\|_{F_\varrho} & =
e^{-R(x)}\, \| \eta(x,u)\, K_t(x,u) - \eta(x,u')\, K_t(x,u')\|_{v(\varrho), (0,1]} \\
& \le e^{-R(x)}\, |\eta(x,u) - \eta(x,u')| \,\|\ K_t(x,u)\|_{v(\varrho), (0,1]} \\& +
e^{-R(x)}\, \eta(x,u') \,\| \ K_t(x,u) - K_t(x,u')\|_{v(\varrho), (0,1]} =: \mathrm{I} + \mathrm{II}.
\end{align*}
To estimate the term I we use \eqref{gradeta} to get $|\eta(x,u) - \eta(x,u')| \lesssim (1+|x|)\,|u-u'| $ and
estimate $\| K_t(x,u)\|_{v(\varrho), (0,1]}$ precisely as in (a). If $|x| > 1$, Proposition
\ref{lemma-integral_dt} then leads to
\begin{align} \label{termI}
|I| \lesssim |u-u'|\,|u-x|^{-n-1}.
\end{align}
In the opposite case $|x| \le 1$, we get
$|\mathrm{I}| \lesssim |u-u'|\,|u-x|^{-n}$ in the same way. But in that case $|u-x| \lesssim 1$ since $\eta(x,u)$ or $\eta(x,u')$ is nonzero, so \eqref{termI} follows again.
For II we estimate $\eta(x,u')$ by 1 and let $w = u-u'$. We can apply \eqref{var} to write
\begin{align}\label{termII}
\| \ K_t(x,u) - K_t(x,u')\|_{v(\varrho), (0,1]} &\le \int_0^1 \big| \dot K_t(x,u) - \dot K_t(x,u')\big|\,dt \\
& \le \int_0^1 \left|\int_0^1 \langle \nabla_2\dot K_t(x,u'+sw),w \rangle\,ds\right|\,dt,
\end{align}
where $\nabla_2$ denotes the gradient with respect to the second variable.
We need an estimate of the second derivative $\nabla_2\dot K_t$\,, to begin with at the point $(x,u)$.
Let for simplicity $\ell \in \{1,\dots,n\}$.
% Writing $u'' =u'+sw$, we use
From \cite[Lemma 4.1]{CCS3} we have
\begin{align*}
\partial_{u_\ell} K_t(x,u) =
- K_t(x,u) \,R_\ell(t,x,u),\qquad t>0,
\end{align*}
where
\begin{align*}
R_\ell(t,x,u) = \left\langle Q_t^{-1} e^{tB} \, ( D_{-t}\, u - x), \,e_\ell\right\rangle.
\end{align*}
Further
\begin{equation} \label{Rell}
|R_\ell(t,x,u) |\lesssim
\frac{|D_{-t}\,u-x|}t \simeq
\frac{|u-D_t\, x|}t,
\end{equation}
since $t\in (0,1)$,
and %Moreover, by \eqref{8} and \eqref{derd-t} we obtain
\begin{align*}
& \dot R_\ell(t,x,u) =
{{-\left\langle Q_t^{-1}\, e^{tB}\, Q\, e^{tB^*}\, Q_t^{-1} e^{tB} \, ( D_{-t}\, u - x), \,e_\ell\right\rangle}}\\
&\qquad\qquad+{{\left\langle Q_t^{-1} Be^{tB} \, ( D_{-t}\, u - x), \,e_\ell\right\rangle}}+{{\left\langle Q_t^{-1} e^{tB} \, Q_\infty\, B^* \, Q_\infty^{-1}\, D_{-t}\, u , \,e_\ell\right\rangle}}.
\end{align*}
In the last term here, we write $D_{-t}\, u = (D_{-t}\, u - x) +x$, and it follows that
\begin{align} \label{dotRell}
| &\dot R_\ell(t,x,u)| \lesssim
\frac{| u-D_t\,x|}{t^2}+\frac{| D_{-t}\, u -x|+|x|}{t}\lesssim \frac{| u-D_t\,x|}{t^2}+\frac{|x|}t.
\end{align}
Clearly,
\begin{align*}
\partial_t\,\partial_{u_\ell} K_t(x,u) & = \dot K_t(x,u)\,R_\ell(t,x,u) +
K_t(x,u) \, \dot R_\ell(t,x,u),
\end{align*}
and to estimate the two terms here, we combine \eqref{dotKeps} with \eqref{Rell} and
\eqref{litet} with \eqref{dotRell}. As a result,
\begin{align*}
|\partial_t\,\partial_{u_\ell} K_t(x,u)| & \lesssim e^{-R(x)}\, t^{-\frac n2} \exp\left(- c\, \frac{|u-D_t\,x|^2}t\right)\,
\left(\frac{1}{t}+\frac{|x|}{\sqrt t}\right)\, \frac{|u-D_t\, x|}t \\ &+
e^{-R(x)}\, t^{-\frac n2} \exp\left(- c\, \frac{|u-D_t\,x|^2}t\right)\,
\left( \frac{| u-D_t\,x|}{t^2}+\frac{|x|}t\right)
\end{align*}
By reducing the constants $c$ in the factors
$\exp\left(- c\, {|u - D_t\,x|^2}/t\right)$ ,
we can eliminate factors ${|D_t\,x -u|}/\sqrt t$ and get
\begin{align*}
|\partial_t\,\partial_{u_\ell} K_t(x,u)| & \lesssim e^{-R(x)}\, t^{-\frac n2}
\exp\left(- c\, \frac{|u-D_t\,x|^2}t\right)\,
\left(\frac{1}{t^{3/2}}+\frac{|x|}{ t}\right).
\end{align*}
Proposition \ref{lemma-integral_dt} now implies
\begin{align}\label{dtdu}
\int_{0}^{1} |\partial_t\,\partial_{u_\ell} K_t(x,u)| \,dt \lesssim |u-x|^{-n-1}.
\end{align}
The next step is to observe that the arguments leading to this estimate remain valid if $u$ is replaced by
$u'+sw$ for any $s \in [0,1]$, in the left-hand side. For the right-hand side this replacement does not change the order of magnitude, so we can keep $|u-x|$ there.
Swapping the order of integration in \eqref{termII} and observing that $|w| = |u-u'|$, we conclude from this that
\begin{equation*}
|\mathrm{II}| \lesssim \int_{0}^{1} \frac{|u-u'|}{ |x-u|^{n+1}}\,ds = \frac{|u-u'|}{ |x-u|^{n+1}}.
\end{equation*}
Together with \eqref{termI}, this estimate completes the proof of (b) and that of Proposition~\ref{lemma-Calderon-no}.
% The proof of Proposition \ref{lemma-Calderon-no} is complete.
\end{proof}
\vskip30pt
%\end{proof}
We now combine Propositions \ref{prp2} and \ref{lemma-Calderon-no} with the Bochner integral representation in
\eqref{formula2_Bochner} for $x\notin \mathrm{supp}\, f$. This is all we need to apply the standard Calderón-Zygmund argument with values in $F_{\varrho}$, to obtain the weak type (1,1) estimate in Theorem \ref{locsmallt_v2}.
Theorem \ref{locsmallt} also follows.
Finally, Theorem \ref{thm} follows from
Theorems \ref{t>1}, \ref{t<1,global} and \ref{locsmallt}.
\section{A counterexample}\label{counterexample}
\begin{theorem}\label{thm_counterex}
The variation operator
\begin{equation*}
f \mapsto \|\mathcal{H}_t f (x)\|_{v(2),\R_+}, \qquad x\in \R^n,
\end{equation*}
is not of strong nor weak type $(p,p)$ with respect to $\gamma_\infty$, for any $p \in [1,\infty)$.
\end{theorem}
\smallskip
This theorem shows that the condition $\varrho>2$ is necessary in Theorem \ref{locsmallt} and therefore also in Theorem \ref{thm}. The proof of Theorem \ref{thm_counterex}
falls naturally into several parts and will occupy the entire section.
\medskip
We first describe the plan of the proof.
It is based on a probabilistic result for the one-dimensional torus due to Qian \cite{Qian}.
To relate our setting to that of Qian, we first approximate $\mathcal{H}_t$ by a convolution operator $\mathcal{H}_t^c$. Staying in a compact set in $\R^n$, we consider only $t \in (0,1]$ and verify that this approximation is admissible. The kernel $\mathcal{H}_t^c$ is then diagonalized, which gives a tensor product of one-dimensional convolution operators.
Our focus will be on the first coordinate, and there we apply the convolution operator to a function defined as a sum of Rademacher functions. We also move from a compact interval to the one-dimensional torus, where we will have a convolution operator acting on a similar Rademacher sum. The parameter $t$ is now restricted to a dyadic sequence.
Next, we verify that we can replace the operator by a simpler mean value operator on the torus, which is then replaced by a dyadic conditional expectation operator, to which Qian's result applies. This gives the counterexample that proves Theorem \ref{thm_counterex}.
%{\em{Proof of Theorem \ref{thm_counterex}}.}
\subsection{Approximation of the kernel}
We consider only $t \in (0,1]$, and $f$ will be supported in a compact set. Moreover, $\mathcal{H}_t f$ will only be considered at points $x$ in
another compact set.
For the $L^p$ norms of $f$ and those of the variation of $\mathcal{H}_t f$, we can therefore use Lebesgue measure instead of $\gamma_\infty$. In the integral \eqref{def-int-ker} defining $\mathcal{H}_t f(x)$, we can write $du$ instead of $d\gamma_\infty(u)$.
This also allows us to delete the factors ${(\det Q_\infty)}^{{1}/{2}}$ and $e^{R(x)}$ in the expression
\eqref{mehler} for the Mehler kernel. What remains is the kernel
\begin{align} \label{mehlersimpl} %\label{defKRt}
\widetilde K_t (x,u)
=
{(\det \, Q_t)}^{-{1}/{2} }\,
\exp \Big[
{-\frac12
\left\langle \left(
Q_t^{-1}-Q_\infty^{-1}\right) (u-D_t \,x) \,, u- D_t\, x\right\rangle}\Big]\!,
\end{align}
and we will approximate this kernel.
The $\mathcal O(\cdot)$ symbol will be used for
scalars, vectors and matrices and is for $t \to 0$. The implicit constants involved may depend on the compact sets mentioned, as well as on $Q$ and $B$.
Since $t \le 1$, the definition \eqref{defQt} implies $Q_t = Q t+ \mathcal O(t^2)$, and then
\begin{equation*}
Q_t^{-1}-Q_\infty^{-1} = Q^{-1} t^{-1} \big(1+ \mathcal O(t)\big).
\end{equation*}
Further, $D_t\, x = x + \mathcal O(t)$ because of \eqref{def:Dt}. Since $x$ and $u$ stay bounded, it follows that
\begin{align}\label{diff:section9}
&\left\langle \left(
Q_t^{-1}-Q_\infty^{-1}\right) (u-D_t \,x) \,, u-D_t\, x\right\rangle
%%=\left\langle\big( Q^{-1} t^{-1} + \mathcal O(1)\big)(u-x + \mathcal O(t)x), u-x + \mathcal O(t)x\right\rangle\\
\notag\\
&=
\left\langle
Q^{-1} t^{-1} (u-x + \mathcal O(t)),
u-x + \mathcal O(t)
\right\rangle
+\mathcal O\big(|x-u|+t\big)
\notag\\
&=
\left\langle
Q^{-1} t^{-1} (u-x),
u-x
\right\rangle+
\mathcal O\big(|x-u|+t\big)
.
\end{align}
We also observe that $\det \, Q_t =t^n\, \det Q\, (1+ \mathcal O(t))$, so that
\begin{equation}\label{det}
{(\det \, Q_t)}^{-{1}/{2} } = {(\det \, Q)}^{-{1}/{2} }\,t^{-{n}/{2}} (1+ \mathcal O(t)).
\end{equation}
This makes it natural to approximate $\widetilde K_t (x,u)$ by the simple convolution kernel
$K^{\mathrm c}_t (x-u)$, where
\begin{equation}\label{ktc}
K^{\mathrm c}_t (y) = {(\det \, Q)}^{-{1}/{2} }\,t^{-{n}/{2}} \,\exp \Big(
{-\frac12\, t^{-1}
\left| Q^{-1/2} y \right|^2}\Big),
\end{equation}
and we let
\begin{align*}
\mathcal H^{\mathrm c}_tf(x) = K^{\mathrm c}_t * f(x).
% \int K^{\mathrm c}_t(x,u)\,f(u)\, d\gamma_\infty(u) \,, \qquad x\in\R^n,
\end{align*}
%when $x$ and $u$ stay in compact sets.
\subsection{The difference operator}
Let $\Delta_t$ be the operator defined by the kernel $\widetilde K_t(x,u)- K^{\mathrm c}_t (x-u)$.
\begin{proposition}\label{lemmaIIt}
Let $C_1$ and $C_2$ be compact subsets of $\R^n$. If $f \in L^2(\R^n)$ has support contained in
$C_1$, then
\begin{equation*}
\left\| \left\|\Delta_{2^{-2\ell}} f \right\|_{v(2),\N_+} \right\|_{L^2(C_2)} \lesssim \|f\|_{L^2(C_1)}.
\end{equation*}
Here the implicit constants may depend on $C_1$ and $C_2$, in addition to $Q$ and $B$.
\end{proposition}
\begin{proof}
By means of \eqref{diff:section9} and \eqref{det}, we get for $x \in C_2$ and $u \in C_1$
\begin{align*}
&\widetilde K_t(x,u)\\
=&\,
(\det \, Q)^{-{1}/{2}}\,t^{-{n}/{2} }(1+\mathcal O(t))\,
\exp\Big( -\frac12 t^{-1}
\left\langle Q^{-1} (u-x), u-x \right\rangle
+ \mathcal O\big(|x-u|+t\big)
\Big)\\
= &\, K^{\mathrm c}_t (x-u)\,(1+\mathcal O(t))\,\exp\Big(\mathcal O\big(|x-u|+t\big)
\Big)\\
%%=&\, K^{\mathrm c}_t (x-u) + K^{\mathrm c}_t (x-u)\,\left[ \exp\Big(\mathcal O\big(|x-u|+t\big)
%%%\Big) -1+\mathcal O(t)\right]\\
=&\, K^{\mathrm c}_t (x-u) + K^{\mathrm c}_t (x-u)\,\mathcal O\big(|x-u|+t\big).
\end{align*}
In the last term here, we may replace $|x-u|$ by $t^{1/2}$, if we reduce the factor $1/2$
in the exponential factor in the expression \eqref{ktc} for $K^{\mathrm c}_t $. Thus
\begin{align*}
&|\widetilde K_t(x,u) - K^{\mathrm c}_t (x-u)| \lesssim
t^{1/2}\, t^{-n/2}\,\exp\Big( -c t^{-1}
|Q^{-1/2} (x-u)|^2
\Big).
\end{align*}
With $f $ supported in $C_1$ and $x \in C_2$, we thus have
\begin{equation*}
|\Delta_t f(x)| \lesssim t^{1/2}\, \int f(u)\, t^{-n/2}\,\exp\Big( -c t^{-1}
|
Q^{-1/2} (x-u)
|^2\Big)\,du.
\end{equation*}
The integral here is the convolution of $f$ and an integrable kernel, and hence
\begin{equation*}
\|\Delta_t f\|_{L^2(C_2)} \lesssim t^{1/2}\, \|f\|_{L^2(C_1)}.
\end{equation*}
Letting $t = 2^{-2\ell}$, squaring and summing, we obtain
\begin{equation*}
\left\|\left(\sum_{\ell =1}^{\infty}(\Delta_{2^{-2\ell}} f)^2\right)^{1/2}\right\|_{L^2(C_2)} \lesssim \|f\|_{L^2(C_1)}.
\end{equation*}
Because of \eqref{discrete_trivial}, this completes the proof.
\end{proof}
\subsection{A counterexample for $ K^{\mathrm c}_t$}
For convenience we normalize $ K^{\mathrm c}_t$ by multiplying by $(2\pi)^{-n/2}$.
Then we make an orthogonal change of variables in $x$ and in $u$ which diagonalizes the matrix $Q$.
By also scaling the coordinates, we can then replace $Q$ by the identity matrix. The new variables will be called $x'$ and $u'$, and we replace $f$ by $g$ defined by $f(u) = g(u')$, with equivalent $L^p$ norms. In the new coordinates, the relevant kernel is a tensor product
\begin{equation*}
K_t^*(y') = \prod_{i=1}^{n} J_t(y'_i)
\end{equation*}
where $J_t$ is the standard one-dimensional gaussian kernel
\begin{equation*}
J_t(y'_i) = \frac{1}{\sqrt{2\pi t}}\,\exp \Big(
{-\frac12\, t^{-1}
\left| y_i' \right|^2}\Big)
\end{equation*}
and the operator is given by $\mathcal H_t^*\, g = K_t^* * g$.
We will choose $g$ as a tensor product
$g(u') = \prod_{i=1}^{n} g_i(u'_i)$, so that
\begin{equation*}
\mathcal H_t^*\, g(x') = \prod_{i=1}^{n} J_t * g_j(x'_i),
\end{equation*}
with the convolutions taken in $\mathbb R$. In the sequel, we will consider the values of $\mathcal H_t^* g$ only at points $x'$ in the cube $[0,1]^{n}$, and each $g_i$ will have support in $[-1,2]$.
This determines the compact sets $C_1$ and $C_2$, via the change of variables.
For
$i=2,\dots, n$, we simply choose $g_i = \chi_{[-1,2]}$, and observe that for these $i$
\begin{equation*}
J_t * g_i(x'_i) = 1+ \mathcal O(t), \qquad x'_i \in [0,1].
\end{equation*}
It follows that
\begin{equation}\label{i>1}
1 - \prod_{i=2}^{n} J_t * g_i =\mathcal O(t) \qquad \mathrm{in} \quad [0,1]^{n-1}.
\end{equation}
The construction of $g_1$ requires more care. In the rest of this subsection, we will write simply
$x,\;u,\; y$ instead of $x'_1,\;u'_1,\;y'_1$ in one dimension. As already mentioned, we restrict $t$ to the set $\{2^{-2\ell},\:\ell = 1,2, \dots \}$, and define operators
$$
A_\ell\, g_1 = J_{2^{-2\ell}} * g_1,\qquad \ell=1,2,\ldots.
$$
Notice that $J_{2^{-2\ell}}(y) = ({2\pi })^{-1/2}\,2^\ell\exp \left(
{-\frac12\,
\left| 2^\ell y \right|^2}\right)$.
%%%\begin{proposition}\label{1-dim} The operator
%% \begin{equation*} g \mapsto \|A_\ell\, g(x)\|_{v(2),\N_+}, \qquad x\in [0,1], \end{equation*} is not bounded nor weakly bounded from $L^p[-1,2]$ to $L^p[0,1]$
%%% for any $p \in [1,\infty)$ with respect to $\gamma_\infty$. \end{proposition}
We will consider the variation in $\ell$ with $\ell$ ranging over the set
$$
\mathcal I_N :=\{\ell \in \mathbb Z: 2N < \ell \le 3N \},
$$
for large $N \in \N$.
To choose our function $g_1$, we use the
Rademacher functions, supported in $[0,1]$ and given by
\begin{equation*}
r_k = \sum_{j=1}^{2^{k-1}} \left( \chi_{((2j-2)2^{-k },\;(2j-1)2^{-k })} - \chi_{((2j-1)2^{-k },\; 2j2^{-k })} \right), \qquad k=1,2,\dots.
\end{equation*}
We identify $[0,1)$ with the torus $\mathbb T = \R/\Z$, considered as a probability space with Lebesgue measure. Then the $r_k$ are independent Bernoulli random variables, taking the values $\pm 1$ with equal probability $1/2$.
In $\R$ we define functions coinciding with the ${r}_k$ in $(0,1)$ and supported in $[-1,2]$, by setting for $k = 1,2,\dots$
\begin{equation*}
q_k(u) = r_k(u+1) + r_k(u) +r_k(u-1), \qquad u \in \R.
\end{equation*}
The function $g_1$ will be
\begin{equation}\label{eq:def_gN}
g_N := \sum_{k \in \mathcal I_N} {q}_k.
\end{equation}
Notice that the notation $g_N$ is not in conflict with $g_i$, $i=1,\ldots,n$, since $N$ is large.
Then for $1<p<\infty$
%% \begin{equation}\label{khinchine}
%%% \|\beta_{\ell} g_N\|_{L^p(\R)} \simeq \| g_N\|_{L^p(\R)} \simeq \|\beta_{\ell} g_N\|_{L^p(\mathbb T)} \simeq
%%% \| g_N\|_{L^p(\mathbb T)} \simeq \sqrt N,
%%%% \end{equation}
\begin{equation}\label{khinchine}
\| g_N\|_{L^p(\R)} \simeq
\| g_N\|_{L^p(\mathbb T)} \simeq \sqrt N,
\end{equation}
due to Khinchine's inequality. We also observe the trivial estimate
\begin{equation}\label{gNtriv}
|g_N(x)| \le N, \qquad x \in \mathbb R.
\end{equation}
%\end{proof}
%%%Proposition \ref{1-dim} is a consequence of the following result.
The following is the main result of the present subsection.
\begin{proposition}\label{controes2}
For some constant $c>0$, the Lebesgue measure of the set
\begin{equation}
\left\{x \in (0,1): \:\| A_\ell \,g_N(x) \|_{v(2),\mathcal I_N} > c \,\sqrt{N \,\mathrm{\log \log} N}\right\}
%\left\{x \in (0,1): \:\|A_{\ell} (\beta_{\ell} g_N)(x) \|_{v(2),\mathcal I_N} > c \,\sqrt{N \mathrm{LL} N}\right\}
\end{equation}
tends to 1 as $N \to \infty$. Here the variation is taken in $\ell$.
% Here $\mathrm{LL}$ is short for $\log\log$.
\end{proposition}
\begin{proof}
Starting with $A_\ell$, we will change operator in several steps, until we arrive at an operator to which Qian's result in \cite{Qian} applies.
\vskip5pt
{\textbf{Step 1}} will take us from $A_\ell$ to the
mean value operator $D_\ell,\: \ell= 1, 2,\dots $, given by
\begin{equation*}
D_\ell f(x) = 2^{\ell-1}\,\int_{x-2^{-\ell}}^{x+2^{-\ell}} f(y)\,dy.
\end{equation*}
This definition works equally well in $\R$ and in $\mathbb T$, and we will write $D_\ell^\R$ and $D_\ell^{\mathbb T}$ to distinguish the two.
Jones and Wang \cite[Section 2]{Jones4} have introduced an equivalence relation $ \sim_p$ between sequences of operators acting on functions defined on the torus.
%, for $n \in \N_+$.
We define a similar relation for operators $P_\ell$ and $Q_\ell$, $\ell \in \mathbb N_+$, acting on functions in $\R$, with $p=2$.
By $(P_\ell) \sim_2 (Q_\ell)$ we mean
that for any sequence $\big(\nu_\ell\big)_1^\infty $, with $|\nu_\ell| \le 1$, and any $f \in L^2(\R)$
\begin{equation*}
\|\sum_{\ell\ge 1}(P_\ell f - Q_\ell f)\, \nu_\ell \|_2 \lesssim \|f\|_2.
\end{equation*}
Here the norm is that in $L^2(\R)$.
As shown in \cite[Theorem 2.5]{Jones4}, this implies that
\begin{equation*}
\left\|\left(\sum_{\ell\ge 1}|P_\ell f - Q_\ell f|^2\right)^{1/2} \right\|_2 \lesssim \|f\|_2,
\end{equation*}
which by \eqref{discrete_trivial} yields
\begin{equation} \label{paragon}
\left\| \left\|P_\ell f - Q_\ell f \right\|_{v(2),\N_+} \right\|_2 \lesssim \|f\|_2.
\end{equation}
%The variation is taken in the variable $n$. % here and in the rest of this section.
\begin{lemma}\label{AD}
We have
%%The operators $P_n f = A_n(\beta_{\ell} f)$ and $Q_n f = D_{\ell}^\R(\beta_{\ell} f)$ sattisfy
$$
(A_\ell) \sim_2 (D_\ell^\R).
$$
\end{lemma}
\begin{proof}
We adapt the proof of \cite[Lemma 2.8]{Jones4} to $\R$.
Observe first that $A_\ell$ and $D_\ell^\R$ are convolution operators with kernels
$J_{2^{-2\ell}} = (2\pi)^{-1/2}\, 2^\ell\, \exp\left(-\frac12\,\left(2^\ell\,y\right)^2\right)$ and
$\chi_\ell(y) = 2^{\ell-1}\,\chi\left(2^\ell\,y\right)$, respectively, $\chi$ denoting the characteristic function of the interval $(-1,1)$.
We use Fourier transforms, defined by
$\mathcal F\,f (\xi)=\hat f(\xi)=
\int f(x) e^{-2\pi i x\,\xi} \,dx$,
and the Plancherel theorem, to deduce
\begin{align*}
\|\sum_{\ell\ge 1} (A_\ell f - D_\ell^\R f)\, \nu_\ell \|_2^2 % \lesssim \|f\|_2.
& = \int_{\mathbb R}
\Big|
\sum_{\ell\ge 1}
\mathcal F(A_\ell f - D_\ell^\R f)(\xi)\, \nu_\ell\Big|^2 d\xi \\
& \le \int_{\mathbb R}|\hat f(\xi)|^2
\Big(
\sum_{\ell\ge 1}\big|
{\widehat {J_{2^{-2\ell}}}(\xi)-\widehat {\chi_\ell}(\xi)\big|}\Big)^2 d\xi.
\end{align*}
It is thus enough to prove that
\begin{equation}\label{haer}
\sum_{\ell\ge 1}\big|
{\widehat {J_{2^{-2\ell}}}(\xi)-\widehat {\chi_\ell}(\xi)\big|}\lesssim 1.
\end{equation}
The Fourier transform of the kernel $J_{2^{-2\ell}}$ is
$$
\widehat {J_{2^{-2\ell}}}(\xi)= \exp\left(-2\pi^2({2^{-\ell}\xi})^2\right),
$$
so that $|1-\widehat {J_{2^{-2\ell}}}(\xi)| \lesssim |{2^{-\ell}}\xi|^2$ for $|2^{-2\ell}\xi| \le 1$ and
$|\widehat {J_{2^{-\ell}}}(\xi)| \lesssim |2^{-\ell}\xi|^{-2}$ for $|2^{-\ell}\xi| \ge 1$.
The Fourier transform
$$
\widehat {\chi_\ell}(\xi) = \frac{\sin (2\pi2^{-\ell} \xi)}{2\pi 2^{-\ell}\xi}
$$
similarly satisfies $|1-\widehat {\chi_{{\ell}}}(\xi)| \lesssim |{2^{-\ell}}\xi|^2$ for
$|2^{-\ell}\xi| \le 1$ but only $|\widehat {\chi_{{\ell}}}(\xi)| \lesssim |2^{-\ell}\xi|^{-1}$ for $|2^{-\ell}\xi| \ge 1$. From this \eqref{haer} and the lemma follow.
\end{proof}
To $A_{\ell} (g_N) (x)$ and $D_{\ell}^\R (g_N) (x)$ we now apply the triangle inequality
for the variation seminorm and write for $x \in \R$
\begin{align} \label{ADR}
\left\| A_{\ell} (g_N) (x) \right\|_{v(2),\mathcal I_N} \ge
% \left\| D_{\ell}^{\R} (\beta_{\ell} g_N) (x) \right\|_{v(2),\mathcal I}
% - \left\|A_{\ell} (\beta_{\ell} g_N) (x) - D_{\ell}^{\R} (\beta_{\ell} g_N) \right\|_{v(2),\mathcal I} \\
\left\| D_{\ell}^{\R} ( g_N) (x) \right\|_{v(2),\mathcal I_N} - h_1 (x),
\end{align}
where
\begin{equation*}
h_1 (x) = \left\|A_{\ell} ( g_N) (x) - D_{\ell}^{\R} ( g_N) (x) \right\|_{v(2),\mathcal I_N}.
\end{equation*}
From Lemma \ref{AD}, \eqref{paragon} and \eqref{khinchine}, we conclude that
$\|h_1\|_{L^2(\R)} \lesssim \sqrt N$.
\vskip17pt
{\textbf{Step 2}} consists in passing from $\R$ to $\mathbb T$. Observe that for any point $x \in (0,1)$, also considered as a point in $\mathbb T$, one has
$D_{\ell}^{\R} (q_k)(x) = D_\ell^{\mathbb T} (r_k)(x)$.
Summing in $k$, we get for these $x$
$$
D_{\ell}^{\R} ( g_N) (x) = D_{\ell}^{\mathbb T} (T_N)(x),
$$
where the function
$$
T_N = \sum_{k \in \mathcal I_N} r_k
$$
is defined in $\mathbb T$.
%Notice that \eqref{khinchine} remains true with $T_N$ instead of $g_N$.
From \eqref{khinchine} we conclude
\begin{equation}\label{khinch}
\| T_N \|_{L^2(\mathbb T)} \simeq \sqrt N.
\end{equation}
\vskip17pt
For {\textbf{Step 3}}, we
let $E_{\ell}$ for $\ell= 1,2,\dots$ denote the conditional expectation operator
which replaces an integrable function defined in $\mathbb T$ by its mean value in each dyadic subinterval $((j-1)2^{-\ell },\; j2^{-\ell }) \subset [0,1]$. Then
\begin{equation}\label{Tn}
E_{\ell} (T_N)= \sum_{k \in \mathcal I_N, \: k \le \ell} r_k
\end{equation}
for $ \ell\in \mathcal I_N$, so that the $E_{\ell} (T_N)$
form a finite martingale for the sequence of $\sigma$-algebras generated by the dyadic intervals of length $2^{-\ell}$,
$\ell \in \mathcal I_N$, contained in $[0,1]$.
Step 3 aims at
replacing $D_{\ell}^{\mathbb T}$ by $E_{\ell}$.
Jones and Rosenblatt
prove in \cite[Remark 4]{Jones_Rosenblatt} that for functions $f \in L^2(\mathbb T)$
\begin{equation*}
\left\|\left(\sum_{\ell=1}^\infty |D_{\ell}^{\mathbb T} f - E_{\ell} f|^2\right)^{1/2} \right\|_2 \lesssim \|f\|_2.
\end{equation*}
Here we refer to the norm in $L^2(\mathbb T)$. These authors use a one-sided, right mean value operator (see \cite[p.528]{Jones_Rosenblatt}),
but since our $D_{\ell}^{\mathbb T}$ is the mean of the right and left operators, this is no problem.
Letting here $f = T_N$, we can apply \eqref{discrete_trivial} and then \eqref{khinch}, to get
\begin{equation*}
\left\| \left\|D_{\ell}^{\mathbb T} ( T_N) - E_{\ell} ( T_N)\right\|_{v(2),\mathcal I_N}\right\|_2
\le \left\|\left(\sum_{\ell \in \mathcal I_N} |D_{\ell}^{\mathbb T} ( T_N) - E_{\ell} ( T_N)|^2\right)^{1/2} \right\|_2 \lesssim \|T_N\|_2 \simeq \sqrt N.
\end{equation*}
The triangle inequality for the seminorm $\|\cdot\|_{v(2),\mathcal I_N}$ now implies that for $x \in \mathbb T$
\begin{equation} \label{DT}
\left\| D_{\ell}^{\mathbb T} ( T_N)(x) \right\|_{v(2),\mathcal I_N} \ge
\left\| E_{\ell} ( T_N) (x) \right\|_{v(2),\mathcal I_N}
- h_2(x),
\end{equation}
with
\begin{equation}\label{h2}
\|h_2\|_{L^2(\mathbb T)} \lesssim \sqrt N.
\end{equation}
\vskip17pt
In {\textbf{Step 4}} we observe that the $r_k$ with $k \in \mathcal I_N$ are $N$ independent random variables fulfilling the hypotheses of
Qian \cite[Theorem 2.2]{Qian}. Because of \eqref{Tn}, this result
tells us that the probability, i.e., the Lebesgue measure, of the set
%for large enough $N$
\begin{equation*}
\left\{x \in \mathbb T: \:\|E_{\ell} (T_N)(x) \|_{v(2),\mathcal J_N} > c \,\sqrt{N \,\mathrm{\log \log} N}\right\}
\end{equation*}
tends to 1 as $N \to \infty$, for some $c>0$.
\vskip17pt
Finally, {\textbf{Step 5}} combines the result of Step 4 with the preceding steps, in reverse order. Considering Step 3, we see that \eqref{h2} and Chebyshev's inequality imply that
$|h_2(x)|$ is much smaller than $\sqrt{N\, \mathrm{log \log} N}$ except on a subset of $\mathbb T$ whose measure tends to 0 as $N \to \infty$.
Then \eqref{DT} implies that the measure of the set
\begin{equation*}
\left\{x \in \mathbb T: \:\|D_{\ell}^{\mathbb T} (T_N)(x) \|_{v(2),\mathcal I_N} > c \,\sqrt{N\, \mathrm{\log \log} N}\right\}
\end{equation*}
tends to 1 as $N \to \infty$, for some $c>0$.
Now Step 2 says that this set is the same as
\begin{equation*}
\left\{x \in [0,1): \:\|D_{\ell}^{\R} (g_N)(x) \|_{v(2),\mathcal I_N} > c \,\sqrt{N \,\mathrm{\log \log} N}\right\}.
\end{equation*}
As for Step 1, we can apply \eqref{ADR} with $x \in [0,1)$ and argue as we did for Step 3. Thus the set
\begin{equation} \label{A>}
\left\{x \in [0,1): \:\|A_{\ell} (g_N)(x) \|_{v(2),\mathcal I_N} > c \,\sqrt{N \mathrm{\log \log} N}\right\}.
\end{equation}
also has measure tending to 1, for some $c>0$.
%%Turning to Step 1, it is clear from \eqref{frak} and Lemma \ref{step1} that in the set \eqref{A>} one has
%%$\|\beta_{\ell} \,A_{\ell} ( g_N(\beta_{\ell} \cdot))(x) \|_{v(2),\mathcal I_N} > c \,\sqrt{N \mathrm{LL} N}$, possibly with another value of $c$.
The proof of Proposition \ref{controes2} is complete.
\end{proof}
\subsection{End of proof of Theorem \ref{thm_counterex}}
The function $g=\otimes_{i=1}^n g_i$, where $g_1=g_N$,
satisfies
for $x' \in [0,1]^{n}$
\begin{align*}
\mathcal H_t^* \,g(x')
& = J_t * g_N(x'_1)\, \prod_{i=2}^{n} J_t * g_i(x'_1)\\
& = J_t * g_N(x'_1) - J_t * g_1(x'_1)\, \left(1-\prod_{i=2}^{n} J_t * g_i(x'_i)\right)
\end{align*}
To estimate the second term here, we apply \eqref{gNtriv} to its first factor and \eqref{i>1} to the second factor. The second term is thus no larger than constant times $Nt$, which is dominated by
$2^{-N}$ if $t = 2^{-2\ell}$ with $\ell\in \mathcal I_N$. From \eqref{discrete_trivial} we see that the
seminorm
$\|\cdot\|_{v(2),\mathcal I_N}$
of the second term is no larger than $N^{1/2}2^{-N}$, since the cardinality of $\mathcal I_N$ is $N$.
Then we apply Proposition \ref{controes2} to
the first term, and conclude that the statement of this proposition also holds for $\mathcal H_t^* \,g(x')$.
The same is then true for $\mathcal H_t^c f(x)$,
where $f(u)=g(u')$ as before, and $x$ is now in some compact set.
Finally, Proposition \ref{lemmaIIt}
allows us to replace
$\mathcal H_t^c $
by the operator defined by the kernel
$\widetilde K_t (x,u)$,
and Theorem \ref{thm_counterex} follows.
\vskip15pt
\section{Appendix}\label{Appendix}
We consider %for $f \in L^1(du)$ and a point $x\notin \mathrm{supp}\, f$
the integral in \eqref{formula2_Bochner}.
\begin{lemma}
If $f \in L^1(du)$ and $x$ is a point such that $x\notin \mathrm{supp}\, f$, then
\begin{equation} \label{integral} % {eq:Bochner_integral}
% \widetilde V^{\text{loc}}\,f(x) =
\int \widetilde M^{\rm{loc}}(x,u)\,f(u)\,du
\end{equation}
is a Bochner integral with respect to the Banach space $F_\varrho$.
\end{lemma}
\begin{proof}
The function $f$ and the point $x\notin \mathrm{supp}\, f$ will be fixed.
Proposition \ref{lemma-Calderon-no}($a$) shows that %$\widetilde V^{\text{loc}}\,f(x) \in F_\varrho$
the integrand in \eqref{integral} %in \eqref{formula2_Bochner}
is in $F_\varrho$ for each $u$.
We will construct
% the definition of the Bochner integral (see stated in \cite[Appendix C]{Engel}) says that we must find
a family $S_r(u)$, with $0<r<r_0$ for some $r_0>0$, of $F_\varrho$-valued simple functions such that for a.a. $u$
\begin{equation} \label{measurability}
\left\| \widetilde M^{\rm{loc}}(x,u)\,f(u) - S_r(u) \right\|_{F_\varrho} \to 0 \qquad \mathrm{as}\quad r \to 0
\end{equation}
(the integrand is then called strongly measurable), and also
\begin{equation}\label{integrability}
\int \left\| \widetilde M^{\rm{loc}}(x,u)\,f(u) - S_r(u) \right\|_{F_\varrho}\,du \to 0 \qquad \mathrm{as}\quad r \to 0.
\end{equation}
Then \eqref{integral} fulfills the definition of an $F_\varrho$-Bochner integral, see \cite[Appendix~C]{Engel}.
%In order to do this, we must find a sequence $(S_m(u))_m$ of $F_\varrho$-valued simple functions such that \begin{equation} \label{measurability} \left\| \widetilde M^{\rm{loc}}(x,u)\,f(u) - S_m(u) \right\|_{F_\varrho}\to 0\end{equation}as $m \to \infty.$
To find these simple functions, we start by covering $\mathrm{supp} \,f$ with
small cubes $Q_\nu\,,$ pairwise disjoint and all of side $r$. Here $r>0$ is so small that the distance from $x$ to the union of these cubes is larger than some $\delta_0 > 0$ independent of $r$. The center of $Q_\nu$ is denoted $d_\nu$.
In this proof, we allow all implicit constants to depend on $x,\;\delta_0$ and $f$.
The integrand $ \widetilde M^{\rm{loc}}(x,u)\,f(u)$ will be approximated by %the simple $F_\varrho$-valued function
\begin{equation*}
S_r(u) = \sum_\nu \widetilde M^{\rm{loc}}(x,d_\nu)\, \frac 1 {|Q_\nu|}\int_{Q_\nu} f(y)\,dy\: \chi_{Q_\nu}(u).
\end{equation*}
We then have
\begin{align*}
\widetilde M^{\rm{loc}}(x,u)&\,f(u) - S_r(u)\\ &=
\sum_\nu \frac 1 {|Q_\nu|}\int_{Q_\nu}
\left( \widetilde M^{\rm{loc}}(x,u) \,f(u)- \widetilde M^{\rm{loc}}(x,d_\nu)\, f(y) \right)\,dy\: \chi_{Q_\nu}(u)\\
&=
\sum_\nu \widetilde M^{\rm{loc}}(x,d_\nu)\, \frac 1 {|Q_\nu|}\int_{Q_\nu} (f(u)-
f(y))\,dy
\,\chi_{Q_\nu}(u) \\ &+
\sum_\nu \left(\widetilde M^{\rm{loc}}(x,u)
-\widetilde M^{\rm{loc}}(x,d_\nu)\right)\, \frac 1 {|Q_\nu|}\int_{Q_\nu} \,dy\,f(u)\: \chi_{Q_\nu}(u)
=: \mathcal I + \mathcal J .
\end{align*}
Starting with $ \mathcal J$, we claim that
%Prop.\ 7.6(c) that
$\|\widetilde M^{\rm{loc}}(x,u) -\widetilde M^{\rm{loc}}(x,d_\nu) \|_{F_\varrho} = \mathcal O(r)$ as $r \to 0$.
As in the proof of Proposition \ref{lemma-Calderon-no}($a$) we have
\begin{align*}
\|\widetilde M^{\rm{loc}}(x,u) -\widetilde M^{\rm{loc}}(x,d_\nu) \|_{F_\varrho}&=
e^{-R(x)}\, \|\eta(x,u)\,K_t(x,u) - \eta(x,d_\nu)\,K_t(x,d_\nu)\|_{v(\varrho),(0,1]} \\& \le
e^{-R(x)}\, |\eta(x,u) -\eta(x,d_\nu)|\,\|K_t(x,u)\|_{v(\varrho),(0,1]} \,\\ &+ \,
e^{-R(x)}\, \eta(x,d_\nu)\, \|K_t(x,u) - K_t(x,d_\nu)\|_{v(\varrho),(0,1]}\,.
\end{align*}
In the sum here, the first term is no larger than constant times $e^{-R(x)}\, r\,\int_{0}^{1}| \dot K_t(x,u)|\,dt \lesssim r\,|x-u|^{-n} = \mathcal O(r)$, again because of
Proposition \ref{lemma-Calderon-no}($a$) and its proof. The second term does not exceed
\begin{equation*}
e^{-R(x)}\, \int_{0}^{1}| \partial_t(K_t(x,u) - K_t(x,d_\nu))|\,dt\,.
\end{equation*}
If $w$ denotes the vector $u - d_\nu$ and $\nabla_2$
%%%$\mathrm{grad}_2$
the gradient in the second variable, we can write
\begin{align*}
\partial_t(K_t(x,u) - K_t(x,d_\nu)) &
= \partial_t \, \int_{0}^{1} \left\langle \nabla_2
%%%\mathrm{grad}_2
\,K_t(x,d_\nu + sw), w \right\rangle \,ds\\
& = \int_{0}^{1} \left\langle
\nabla_2
%%%%\mathrm{grad}_2
\,\dot K_t(x,d_\nu + sw), w \right\rangle \,ds.
\end{align*}
The estimate \eqref{dtdu} shows that
\begin{align*}
e^{-R(x)}\, \int_{0}^{1} |
\nabla_2
%%%\mathrm{grad}_2
\,\dot K_t(x,d_\nu + sw)|\,ds \lesssim 1.
\end{align*}
Since $|w| \lesssim r$, the claim follows, and we see that
\begin{equation*}
\left\| \mathcal J \right\|_{F_\varrho} \lesssim C r \,|f(u)| \to 0 \qquad \mathrm{a.e.} % \mathcal O\left(r \right) .
% \left\| \frac 1 {|Q_\nu|}\int_{Q_\nu} \left(\widetilde M^{\rm{loc}}(x,u)
% -\widetilde M^{\rm{loc}}(x,y)\right)\,dy \:\chi_{Q_\nu}(u)\right\|_{F_\varrho} = \mathcal O(r).
\end{equation*}
%% \begin{equation*} \int \left\| II \right\|_{F_\varrho} \,du = \mathcal O\left(\| f\|_{L^1}\,r \right). \end{equation*}
%Integrating this against $f(u)\,du$, we see that the $F_\varrho$ norm of the last sum in the formula above is $ \mathcal O\left(\| f\|_{L^1}\,r \right)$ and tends to 0 with $r$.
As for the term $ \mathcal I $, Proposition \ref{lemma-Calderon-no}($a$) implies that
\begin{align*}
\| \mathcal I \|_{F_\varrho} \lesssim \sum_\nu \frac 1 {|Q_\nu|}\int_{Q_\nu} \,|f(u)- f(y)|\,dy \: \chi_{Q_\nu}(u),
\end{align*}
whence
$ \| \mathcal I \|_{F_\varrho}\to 0$ as $r\to 0$ at any Lebesgue point.
Thus the measurability condition \eqref{measurability} is satisfied.
It remains to verify \eqref{integrability}.
Reasoning as above, one has since $|w| \lesssim r$
\begin{equation*}
\int \left\| \mathcal J \right\|_{F_\varrho} \,du = \mathcal O\left(\| f\|_{L^1}\,r \right).
\end{equation*}
For the term $ \mathcal I $, we write
\begin{align*}
\int \left\| \mathcal I \right\|_{F_\varrho}\,du & \lesssim
\sum_\nu \int_{Q_\nu} \frac 1 {|Q_\nu|}\int_{Q_\nu} \,|f(u)- f(y)|\,dy \,du \\
& \le \sum_\nu \int_{Q_\nu} \frac 1 {|Q_\nu|}\int_{|v|< r\sqrt n} \,|f(u)- f(u+v)|\,dv \,du \\
& = r^{-n} \,\int_{|v|< r\sqrt n} \,dv \, \sum_\nu \, \int_{Q_\nu} |f(u)- f(u+v)|\,du,
\end{align*}
since $|Q_\nu| = r^n$. Denoting by $\tau_v$ the translation operator, we see that the last expression in this formula equals
\begin{align*}
r^{-n} \, \int_{|v|< r\sqrt n} \,dw \, \|f- \tau_v \,f \|_{L^1} \lesssim \sup_{|v|< r\sqrt n} \|f- \tau_v \,f \|_{L^1}.
\end{align*}
Since translation is continuous in the $L^1$ norm, this tends to 0 with $r$.
Altogether, we have proved \eqref{integrability},
%% that $\int \left\| \widetilde M^{\rm{loc}}(x,u)\,f(u) - S_r(u) \right\|_{F_\varrho}\,du \to 0$
and so the integral \eqref{integral} is ${F_\varrho}$-Bochner.
\end{proof}
\subsection*{Conflict of interest }
The authors
declare that they do not have any conflict of interest for this work.
\begin{thebibliography}{M}
\bibitem{AlmeidaMedit}
V. Almeida, J. J. Betancor, P. Quijano and L. Rodr\'iguez-Mesa,
Maximal operator, Littlewood--Paley functions and variation operators associated with nonsymmetric Ornstein-Uhlenbeck operators,
{\em Mediterr. J. Math.} \textbf{20}, 196 (2023).
\bibitem{Almeida}
V. Almeida, J. J. Betancor, J. C. Farina, P. Quijano and L. Rodriguez-Mesa,
Littlewood--Paley functions associated with general Ornstein--Uhlenbeck semigroups,
{\tt arXiv:2202.06136.}
\bibitem{Seeger}
D. Beltran, R. Oberlin, L. Roncal, A. Seeger and B. Stovall,
Variation bounds for spherical averages,
{\em Mathematische Annalen}, \textbf{382} (2022), 459--512.
\bibitem{Betancor1}
J. Betancor, R. Crescimbeni and J.L. Torrea,
Oscillation and variation of the Laguerre heat and Poisson semigroups and Riesz transforms,
{\em Acta Mathematica Scientia} \textbf{32} (2012), 907--928.
\bibitem{Betancor2}
J. Betancor, A. Castro and M. De León-Contreras,
Variation operators for semigroups associated with Fourier-Bessel expansions,
{\em Comm. Pure and Applied Anal.}, \textbf{21} (2022), 239--273.
\bibitem{Betancor3}
J. Betancor and M. De León-Contreras,
Variation and oscillation for semigroups associated with discrete Jacobi operators,
{\em Anal. Math. Phys.} \textbf{13} (2023), 92. %%% https://doi.org/10.1007/s13324-023-00852-4.
\bibitem{Bourgain}
J. Bourgain, Pointwise ergodic theorems for arithmetic sets,
{\em Publications mathématiques de l’I.H.\'E.S.}, \textbf{69} (1989), 5--45. With an appendix by the author, Harry Furstenberg, Yitzhak Katznelson and Donald S. Ornstein.
\bibitem{Bourgain_etal}
J. Bourgain, M. Mirek, E. M. Stein and J. Wright,
On a multi-parameter variant of the Bellow-Furstenberg problem,
{\em Forum of Mathematics, Pi}, (2023); 11:e23. doi:10.1017/fmp.2023.21
\bibitem{Campbell}
J.T. Campbell, R.L. Jones and K. Reinhold, M. Wierdl,
Oscillation and variation for the Hilbert transform,
{\em Duke Math. J.} \textbf{105} (2000), 59--83.
\bibitem{CCS1}
V. Casarino, P. Ciatti and P. Sj\"ogren,
The maximal operator of a
normal
Ornstein--Uhlenbeck semigroup is of weak type $(1,1)$,
{\em Ann. Sc. Norm. Super. Pisa Cl. Sci.} (5) Vol. XXI (2020), 385-410, %doi : $10.2422/2036-2145.201805\_012$,
{\tt arXiv:1705.00833}.
\bibitem{CCS2}
V. Casarino, P. Ciatti and P. Sj\"ogren, { On the maximal operator of a general Ornstein--Uhlenbeck semigroup}, {\em Math. Z.} (2022) \textbf{301}, 2393--2413.
\bibitem{CCS3}
V. Casarino, P. Ciatti and P. Sj\"ogren,
{Riesz transforms of a general Ornstein--Uhlenbeck semigroup},
{\em Calculus of Variations and Partial Differential Equations},
\textbf{60}, 135 (2021).% https://doi.org$/10.1007/s00526-021-01988-6.$
\bibitem{CCS4} V. Casarino, P. Ciatti and P. Sj\"ogren,
{On the orthogonality of generalized eigenspaces for the Ornstein--Uhlenbeck operator}, {\em Archiv der Matematik},
117 (2021), 547--556.
\bibitem{CCS5} V.\ Casarino, P.\ Ciatti and P.\ Sj\"ogren, Spectral multipliers in a general Gaussian setting,
%%{\tt arXiv:2202.01547},
to appear in {\em Ann.\ Sc.\ Norm.\ Super.\ Pisa Cl.\ Sci.},
\href{https://doi.org/10.2422/2036-2145.202311_014}{https://doi.org/10.2422/2036-2145.202311\_014}.
\bibitem{CCS6}
V. Casarino, P. Ciatti and P. Sj\"ogren, {
On the variation operator for the Ornstein--Uhlenbeck semigroup
in dimension one}, {\em Annali di Matematica} \textbf{203} (2024), 205--219.
\bibitem{CCS7}
V. Casarino, P. Ciatti and P. Sj\"ogren, Some Harmonic Analysis in a General Gaussian Setting,
in ``Modern Problems in PDEs and Applications",
Trends in Mathematics, vol. 4, Birkhäuser Cham, 13--18, (2024).
\bibitem{Crescimbeni}
R. Crescimbeni, R. A. Macías, T. Menárguez, J. L. Torrea and B. Viviani,
The $\varrho$-variation as an operator between maximal operators and singular integrals, {\em J. Evol. Equ.}
\textbf{9} (2009), 81--102.
\bibitem{Engel}
K. J. Engel and R. Nagel, {\em One parameter semigroups for linear evolution equations},
194
Springer Graduate Texts in Mathematics, 2000.
\bibitem{Gaposhkin1}
V. F. Gaposhkin,
A theorem on the convergence almost everywhere of a sequence of measurable functions, and its applications to sequences of stochastic integrals,
{\em Mathematics of the USSR-Sbornik} \textbf{33} (1977), 1--17.
\bibitem{Gaposhkin2}
V. F. Gaposhkin,
Individual ergodic theorems for normal operators on $L^2$,
{\em Functional Analysis and its Applications} \textbf{15} (1981), 14--18.
%%\bibitem{Grafakos} L. Grafakos,{\em Classical Fourier Analysis}, Graduate Texts in Mathematics, 249, Springer New York, NY, (2014).
\bibitem{Harboure}
E. Harboure, R. A. Mac\'ias, M. T. Menàrguez, and J. L. Torrea, Oscillation and variation for the Gaussian Riesz
transforms and Poisson integral, {\em Proc. Roy. Soc. Edinburgh Sect. A} \textbf{135}, 1 (2005), 85--104.
\bibitem{Harboure2}
E. Harboure, J. L. Torrea and B. Viviani, Vector-valued extensions of operators related to the Ornstein-Uhlenbeck
semigroup, {\em J. Anal. Math.} \textbf{91} (2003), 1--29.
\bibitem{Jones}
R.L. Jones, R. Kaufman, J. M. Rosenblatt and M. Wierdl,
Oscillation in ergodic theory,
{\em Ergodic Theory Dynam. Systems} \textbf{18} (1998), 889--935.
\bibitem{Jones1}
R. L. Jones and K. Reinhold, Oscillation and variation inequalities for convolution powers, {\em Ergodic Theory Dyn. Syst.}, \textbf{6} 21 (2001), 1809--1829.
\bibitem{Jones_Rosenblatt}
R. L. Jones, J. M. Rosenblatt, Differential and ergodic transforms, {\em Mathematische Annalen}, \textbf{323} (2002) 525--546.
\bibitem{Jones1-higher}
R. L. Jones, J. M. Rosenblatt and M. Wierdl,
Oscillation in erodic theory: higher dimensional results, {\em Israel J. Math.}, \textbf{135} (2003), 1--27.
\bibitem{Jones2}
R. L. Jones, A. Seeger and J. Wright. Strong variational and jump inequalities in harmonic analysis, {\em Trans. Amer. Math. Soc.}, \textbf{12} 360 (2008),
6711--6742.
\bibitem{Jones4}
R. L. Jones and G. Wang,
Variation inequalities for the Fejér and Poisson kernels,
{\em Trans. Amer. Math. Soc.}
\textbf{356} (2004), 4493--4518.
\bibitem{Krause}
B. Krause, {\em Discrete Analogues in Harmonic Analysis: Bourgain, Stein, and Beyond},
Graduate Studies in Mathematics, 224, American Mathematical Society, Providence, RI, 2023.
\bibitem{Le Merdy}
C. Le Merdy and Q. Xu,
Strong q-variation inequalities for analytic semigroups, {\em Ann. Inst Fourier}, \textbf{62},
6 (2012), 2069--2097.
\bibitem{Lepingle}
D. Lépingle La variation d’ordre p des semi-martingales, {\em Z. Wahrscheinlichkeitstheorie und Verw. Gebiete} \textbf{36}, 4 (1976),
295--316.
\bibitem{Ma1}
T. Ma, J. L. Torrea, and Q. Xu, Weighted variation inequalities for differential operators and singular integrals. {\em J. Funct.
Anal.} \textbf{268}, 2 (2015), 376--416.
\bibitem{Ma2}
T. Ma, J. L. Torrea, and Q. Xu, Weighted variation inequalities for differential operators and singular integrals in higher
dimensions, {\em Sci. China Math.} \textbf{60}, 8 (2017), 1419--1442.
\bibitem{Mirek1}
M. Mirek, E. M. Stein and P. Zorin-Kranich, Jump inequalities via real interpolation,
{\em Math. Ann.}, \textbf{376} (2020), 797--819.
\bibitem{Mirek2}
M. Mirek, E. M. Stein and P. Zorin-Kranich,
A bootstrapping approach to jump inequalities and their applications,
{\em Analysis and PDE}, \textbf{13} (2020), 527--558.
\bibitem{Mirek3}
M. Mirek, E.M. Stein and P. Zorin-Kranich, Jump inequalities for translation-invariant operators of Radon type on
$\mathbb Z^d$, {\em Adv. Math.} \textbf{365} (2020), 107065.
\bibitem{Mirek4}
M. Mirek, T. Z. Szarek and J. Wright, Oscillation inequalities in ergodic theory and analysis: one-parameter and multi-parameter perspectives,
{\em Rev. Mat. Iberoam.} \textbf{38} (2022), 2249--2284
\bibitem{Mirek5}
M. Mirek, W. S\l\,\!omian and T.Z. Szarek,
Some remarks on oscillation inequalities,
{\em Ergodic Theory and Dynamical
Systems} (2022), 1--30.
\bibitem{Mirek6}
M. Mirek, T.Z. Szarek and J. Wright, Oscillation inequalities in ergodic theory and analysis: one-parameter and
multi-parameter perspective, {\em Rev. Mat. Iberoam.}
\textbf{38} (2022), 2249--2284.
\bibitem{Mu}
B. Muckenhoupt,
Hermite conjugate expansions, {\em Trans. Amer. Math. Soc.} \textbf{139} (1969), 243--260.
\bibitem{Perez}
S. Pérez, The local part and the strong type for operators related to the Gaussian measure,
{\em J.~Geom.
Anal.} \textbf{11}, 491--507 (2001).
\bibitem{Qian}
J. Qian, The $p$ variation of partial sum processes and the empirical process, {\em Ann. of Prob.}, \textbf{26} (1998) 1370--1383.
\bibitem{Rubio}
J. L. Rubio de Francia, F. J. Ruiz and J. L. Torrea, Calder\'on--Zygmund theory for operator-valued kernels,
{\em Adv. in Math.},
\textbf{62}, 1 (1986), 7--48.
\bibitem{Sj}
P. Sj\"ogren, On the maximal function for the Mehler kernel. In {\em Harmonic Analysis,
Proceedings, Cortona 1982}. {\em Lecture Notes in Mathematics (Springer)} \textbf{992} (1983),
73--82.
%%\bibitem{Stein} E. M. Stein,{\em Topics in Harmonic Analysis
%%%Related to the Littlewood--PaleyTheory}, Annals in Mathematics Studies. Princeton University Press, Princeton (1970).}}
%%\bibitem{Urbina}W. Urbina-Romero,{\emph {Gaussian Harmonic Analysis}}, Springer Monographs in Mathematics, 2019.
\end{thebibliography}
\end{document}