Unidirectional Entropic Solutions of the Pressureless Euler Alignment System
Extracted source text · arXiv 2606.11159
\documentclass{amsart}
\usepackage{amsfonts,amsmath,latexsym,amssymb,verbatim,amsbsy,times, esint}
\usepackage{amsthm}%, pdfsync}
\usepackage{pstricks}
\usepackage{mathtools}
\usepackage{graphicx}
\usepackage{caption}
\usepackage{mathrsfs}
\captionsetup{justification=centering}
%\usepackage[notcite]{showkeys}
\usepackage{bm}
\usepackage{tikz-cd}
\usepackage{tikz}
\usetikzlibrary{patterns, arrows.meta, decorations.pathreplacing}
\usepackage{mathrsfs}
\usepackage[top=0.8 in, bottom=0.8in, left=0.8in, right=0.8in]{geometry}
%\usepackage[dvipsnames]{xcolor}
\usepackage[colorlinks=true, pdfstartview=FitV, linkcolor=blue,citecolor=green, urlcolor=blue]{hyperref}
\usepackage{enumitem}
%Citation keys in blue, small on the side
%\providecommand*\showkeyslabelformat[1]{{\normalfont \tiny#1}}
%\usepackage[notref,notcite,color]{showkeys}
%\definecolor{labelkey}{rgb}{0,0,1}
\theoremstyle{plain}
\newtheorem{THEOREM}{Theorem}[section]
\newtheorem{COROL}[THEOREM]{Corollary}
\newtheorem{LEMMA}[THEOREM]{Lemma}
\newtheorem{PROP}[THEOREM]{Proposition}
\newtheorem{QUESTION}[THEOREM]{Question}
\newtheorem{theorem}[THEOREM]{Theorem}
\newtheorem{corollary}[THEOREM]{Corollary}
\newtheorem{lemma}[THEOREM]{Lemma}
\newtheorem{proposition}[THEOREM]{Proposition}
\newtheorem{remark}[THEOREM]{Remark}
\theoremstyle{definition}
\newtheorem{DEF}[THEOREM]{Definition}
\newtheorem{definition}[THEOREM]{Definition}
\theoremstyle{remark}
\newtheorem{REMARK}[THEOREM]{Remark}
\newtheorem{CLAIM}[THEOREM]{Claim}
\newtheorem{PROBLEM}[THEOREM]{Problem}
\newtheorem{EXAMPLE}[THEOREM]{Example}
%\newtheorem{remark}[THEOREM]{Remark}
\newtheorem{claim}[THEOREM]{Claim}
\newtheorem{problem}[THEOREM]{Problem}
\newtheorem{example}[THEOREM]{Example}
\newtheorem{EXERCISE}{Exercise}[section]
%\newtheorem{EXERCISE}[THEOREM]{Exercise}
%%%%%%%%%%%%%%%%%%%%%%%%%
\newcommand{\thm}[1]{Theorem~\ref{#1}}
\newcommand{\lem}[1]{Lemma~\ref{#1}}
\newcommand{\defin}[1]{Definition~\ref{#1}}
\newcommand{\rem}[1]{Remark~\ref{#1}}
\newcommand{\cor}[1]{Corollary~\ref{#1}}
\newcommand{\prop}[1]{Proposition~\ref{#1}}
\newcommand{\prob}[1]{Problem~\ref{#1}}
\newcommand{\exam}[1]{Example~\ref{#1}}
\newcommand{\sect}[1]{Section~\ref{#1}}
%%%%%%%%%%%%%%%%%%%%%%%%%
\newcommand{\N}{\ensuremath{\mathbb{N}}} %%% naturals
\newcommand{\Z}{\ensuremath{\mathbb{Z}}} %%% integers
\newcommand{\R}{\ensuremath{\mathbb{R}}} %%% reals
\newcommand{\C}{\ensuremath{\mathbb{C}}} %%% complex
\newcommand{\T}{\ensuremath{\mathbb{T}}} %%% torus
\newcommand{\K}{\ensuremath{\mathbb{K}}}
\renewcommand{\S}{\ensuremath{\mathbb{S}}}
\newcommand{\Q}{\ensuremath{\mathbb{Q}}}
\newcommand{\RP}{\ensuremath{\mathbb{RP}}}
\newcommand{\myr}[1]{{\color{red}{#1}}}
%%%%%%%%%%%%%%% GREEKS %%%%%%%%%%%%%%%5
\def \RR {\mathbb{R}}
\def \a {\alpha}
\def \b {\beta}
\def \d {\delta}
\def \g {\gamma}
\def \e {\varepsilon}
\def \f {\varphi}
\def \F {{\bf \Phi}}
\def \k {\kappa}
\def \l {\lambda}
\def \L {\Lambda}
\def \i {\iota}
\def \n {\nabla}
\def \s {\sigma}
\def \th {\theta}
\def \Th {\Theta}
\def \t {\tau}
\def \w {\omega}
\def \W {\mathcal{W}}
\def \O {\Omega}
%\def \Psi {{\bf \Psi}}
\def \fm {\mathfrak{m}}
\def \ba {{\bf a}}
\def \bA {{\bf A}}
\def \bbb {{\bf b}}
\def \bB {{\bf B}}
\def \bc {{\bf c}}
\def \bC {{\bf C}}
\def \bF {{\bf F}}
\def \bP {{\bf P}}
\def \bPi {\boldsymbol{\Pi}}
\def \bpi {\boldsymbol{\pi}}
\def \bs {\boldsymbol{\sigma}}
\def \bI {{\bf I}}
\def \id {{\bf id}}
\def \br {{\bf r}}
\def \bp {\mathrm{{\bf p}}}
\def \bE {{\bf E}}
\def \boldf {{\bf f}}
\def \bG {{\bf G}}
\def \bH {{\bf H}}
\def \bK {{\bf K}}
\def \bL {{\bf L}}
%\def \bm {{\bf m}}
\def \bn {{\bf n}}
\def \bR {{\bf R}}
\def \bS {{\bf S}}
\def \bT {{\bf T}}
\def \bx {{\bf x}}
\def \bX {{\bf X}}
\def \bu {{\bf u}}
\def \bU {{\bf U}}
\def \bv {{\bf v}}
\def \bV {{\bf V}}
\def \bw {{\bf w}}
\def \cA {\mathcal{A}}
\def\cB {\mathcal {B}}
\def \cC {\mathcal{C}}
\def \cD {\mathcal{D}}
\def \cE {\mathcal{E}}
\def\cF {\mathcal {F}}
\def\cG {\mathcal {G}}
\def \cH {\mathcal{H}}
\def \cJ {\mathcal{J}}
\def \cK {\mathcal{K}}
\def \cL {\mathcal{L}}
\def \cM {\mathcal{M}}
\def \cN {\mathcal{N}}
\def\cO{ \mathcal {O}}
\def \cP {\mathcal{P}}
\def \cQ {\mathcal{Q}}
\def \cR {\mathcal{R}}
\def \cS {\mathcal{S}}
\def\cT {\mathcal {T}}
\def \cU {\mathcal{U}}
\def \cV {\mathcal{V}}
\def \cW {\mathcal{W}}
\def \tf {\widetilde{\f}}
\def \tphi {\widetilde{\phi}}
\def \tg {\widetilde{g}}
\def \tr {\widetilde{r}}
\def \tR {\widetilde{R}}
\def \tX {\widetilde{X}}
\def \tx {\widetilde{x}}
%\def \tu {\widetilde{u}}
\def \tv {\widetilde{v}}
\def \tz {\widetilde{z}}
\def \tS {\widetilde{\mathsf{S}}}
\def \bFi {{\bf \Phi}}
\def \bPsi {{\bf \Psi}}
\def \bs {\boldsymbol{\sigma}}
\def \bt {\boldsymbol{\tau}}
\def \bl {\boldsymbol{\lambda}}
\def \M {\mathcal{M}}
\def \torus {{\T^n}}
\def \rn {\R^n \backslash\{0\}}
\def \zn {{\dot{\Z}^n}}
\def \schwartz {\mathcal{S}}
\def\cprime{$'$}
\def \ex {\vec{e}_1}
\def \ey {\vec{e}_2}
\def \besov {B^{1}_{\infty,\infty}}
\def \loc {\mathrm{loc}}
\def \mmaxm {\mu^m_{\mathrm{max}}}
\def \mmaxk {\mu^k_{\mathrm{max}}}
\def \mminm {\mu^m_{\mathrm{min}}}
\def \mmax {\mu_{\mathrm{max}}}
\def \mmin {\mu_{\mathrm{min}}}
\def \sobcon {\tilde{H}^m(\T^n)}
\def \encon {{L}_\mathcal{F}^2}
\def \enmz {{L}^2_0(\T^n)}
\def \endiv {{L}^2_{\mathrm{div}}(\T^n)}
\def \engrad {{L}^2_{\n}(\T^n)}
\def \new {\ba_{\text{new}}}
\def \red {\bF^{\mathrm{red}}}
\newcommand{\tnu}{\widetilde{\nu}}
%%%%%%%%%%%%%%%%%%%%%% Shortcuts %%%%%%%%%%%%%%%%%%%%%%
\def \lan {\langle}
\def \ran {\rangle}
\def \p {\partial}
\def \ra {\rightarrow}
\def \ss {\subset}
\newcommand{\comb}[2]{\left(\begin{matrix}
#1 \\
#2
\end{matrix} \right)}
\newcommand{\der}[2]{(#1 \cdot \nabla) #2}
\newcommand{\ess}[1]{\s_{\mathrm{ess}}(#1)}
\newcommand{\disc}[1]{\s_{\mathrm{disc}}(#1)}
\newcommand{\ress}[1]{r_{\mathrm{ess}}(#1)}
\newcommand{\sym}[1]{\mathcal{S}^{#1}}
\newcommand{\class}[1]{\mathcal{L}^{#1}}
\newcommand{\sob}[1]{H^{#1}(\torus)}
\newcommand{\sobmz}[1]{H^{#1}_0(\torus)}
\newcommand{\sobdiv}[1]{H^{#1}_{\mathrm{div}}(\torus)}
\newcommand{\sobgrad}[1]{H^{#1}_{\n}(\torus)}
%\newcommand{\sqint}[1]{L^{2}(\T^{#1})}
\newcommand{\sqintmz}[1]{L^{2}_0(\T^{#1})}
\newcommand{\sqdiv}[1]{L^{2}_{\mathrm{div}}(\T^{#1})}
\newcommand{\tA}[2]{\tilde{A}_{#1,\, #2}}
\newcommand{\tB}[2]{\tilde{B}_{#1,\, #2}}
\newcommand{\sgn}{\mathrm{sgn}}
\newcommand{\esssup}{\mathop{\mathrm{ess\,sup}}}
%%%%%%%%%%%% MY FUNCTIONS %%%%%%%%%%%%55
\DeclareMathOperator{\col}{col} %
\DeclareMathOperator{\Int}{Int} %
\DeclareMathOperator{\Lim}{Lim} %
\DeclareMathOperator{\per}{per} %
\DeclareMathOperator{\Span}{span} %
\DeclareMathOperator{\proj}{proj} %
\DeclareMathOperator{\epi}{epi} %
\DeclareMathOperator{\Cl}{Cl} %
\DeclareMathOperator{\diam}{diam} %
\DeclareMathOperator{\supp}{supp} %
\DeclareMathOperator{\conv}{conv} %
\DeclareMathOperator{\diver}{div} %
\DeclareMathOperator{\curl}{curl} %
\DeclareMathOperator{\op}{Op} %
\DeclareMathOperator{\re}{Re} %
\DeclareMathOperator{\im}{Im} %
\DeclareMathOperator{\Rg}{Rg} %
%\DeclareMathOperator{\tr}{Tr} %
\DeclareMathOperator{\Ker}{Ker} %
\newcommand{\rest}[2]{#1\raisebox{-0.3ex}{\mbox{$\mid_{#2}$}}}
\newcommand{\dist}[2]{\mathrm{dist}(#1,#2)}
\newcommand{\tri}[3]{#1 \otimes #2 : \n #3}
\def \cP {\mathcal{P}}
\def \dom {\T^3}
\def \Lip {\mathrm{Lip}}
\def \dd {\mathrm{d}}
\def \da {\mathrm{d}a}
\def \db {\mathrm{d}b}
\def \dc {\mathrm{d}c}
\def \dh {\mathrm{d}h}
\def \ds {\mathrm{d}s}
\def \dt {\mathrm{d}t}
\def \dtau {\mathrm{d}\tau}
\def \du {\mathrm{d}u}
\def \dv {\mathrm{d}v}
\def \dx {\mathrm{d}x}
\def \dy {\mathrm{d}y}
\def \dz {\mathrm{d}z}
\def \dm {\mathrm{d}m}
\def \dr {\mathrm{d}r}
\def \dl {\mbox{d}\l}
\def \dmu {\mathrm{d}\mu}
\def \dr {\mathrm{d}r}
\def \dR {\mathrm{d}R}
\def \ds {\mathrm{d}s}
\def \dw {\mathrm{d}w}
\def \dth {\mathrm{d}\th}
\def \dE {\mathrm{d}\cE}
\def \OR {\mathcal{OR}}
\def \Div {\,\mathrm{Div}}
\def \rad {\mathrm{rad}}
\def \Rf {\mathrm{ref}}
\def \sfS { \mathsf{S}}
\def \sfT { \mathsf{T}}
\def \wtrho { \widetilde{\rho}}
\def \wtu { \widetilde{u}}
\def \brho {\bar{\rho}}
\def \barf {\bar{f}}
\def \barx {\bar{x}}
\def \barv {\bar{v}}
\def \barbv {\bar{\mathbf{v}}}
\def \barS {\overline{\sfS}}
\def \barW {\overline{W}}
\def \barPsi {\overline{\Psi}}
\def \barA {\overline{A}}
\def \barxi {\bar{\xi}}
\def \barphi {\bar{\phi}}
\def \barL {\overline{\cL}}
\def \bareta {\bar{\eta}}
\def \eff {\mathrm{eff}}
\usetikzlibrary{cd}
\usetikzlibrary{snakes}
\newcommand{\mres}{\mathbin{\vrule height 1.6ex depth 0pt width
0.13ex\vrule height 0.13ex depth 0pt width 1.3ex}}
\setlength{\parskip}{0.1em}
\renewcommand{\baselinestretch}{1.15}
\setcounter{tocdepth}{2}
\title{Unidirectional Entropic Solutions of the Pressureless Euler Alignment System}
\author{Joshua O. Adeleke}
\address{Department of Applied Mathematics,
Illinois Institute of Technology}
\email{jadeleke@hawk.illinoistech.edu}
\author{Trevor M. Leslie}
\address{Department of Applied Mathematics,
Illinois Institute of Technology}
\email{tleslie@illinoistech.edu}
\date{}
\thanks{\textit{Acknowledgment.} JA and TL are supported in part by NSF grant DMS-2408585 (PI: Leslie). Both authors thank Changhui Tan for helpful comments on an earlier draft of the manuscript.}
\makeatletter
\@namedef{subjclassname@2020}{\textup{2020} Mathematics Subject Classification}
\makeatother
\subjclass[2020]{35Q35, 35L65, 35D30, 35B30, 76N10, 82C22}
\keywords{Euler Alignment, Cucker--Smale, collective dynamics, scalar balance law, weak solutions, entropy conditions, sticky particle dynamics, flocking}
\begin{document}
\maketitle
\vspace{-10 mm}
\begin{abstract}
We study the pressureless Euler Alignment system with unidirectional velocity $\bu = (u,0, \ldots, 0)$. By re-casting the system as a family of coupled scalar balance laws---one for each horizontal slice of $\mathbb{R}^d$---we are able to prove existence, uniqueness, and stability within the class of unidirectional solutions, under the assumption of a bounded Lipschitz communication protocol. The nonlocal coupling between horizontal slices provides the system with a rich structure that is absent from the 1D setting and also constitutes the main technical difficulty of the present work. We construct our solutions as limits of sticky particle Cucker–Smale dynamics, discretizing first transverse to the flow and then along it. The transverse discretization depends crucially on the more subtle of our two complementary stability estimates, which relies only on $L^1$-$L^\infty$ control of the flux difference. This low-regularity estimate is essential since our discretized fluxes cannot in general be expected to converge in (for instance) Lipschitz seminorm. The estimate itself is inspired by work of Bouchut and Perthame~\cite{BouchutPerthame1998} and adapted here through a careful additional analysis of the nonlocality. In order to meaningfully compare the dynamics along different horizontal slices, both stability estimates are most naturally formulated in terms of quantities that involve the optimal coupling between the projections of two density profiles onto~$\mathbb{R}^{d-1}$.
We also investigate the long-time behavior of unidirectional solutions. In addition to treating the standard heavy-tailed regime, we make the simple observation that the unidirectional geometry allows for flocking (with a rate independent of the number of agents) even when the communication protocol vanishes in a cylindrical neighborhood of the axis parallel to the flow. This demonstrates that direct communication along the direction of motion is not necessary for flocking to occur.
\end{abstract}
\vspace{-2 mm}
\begin{small}
\tableofcontents
\end{small}
\section{Introduction}
We are concerned with the following Euler Alignment system of collective behavior:
\begin{equation}
\label{e:EA}
\left\{
\begin{split}
\partial_t \rho + \nabla \cdot (\rho \mathbf{u}) & = 0 \\
\partial_t (\rho \bu) + \nabla \cdot (\rho \bu \otimes \bu)
& = \int_{\R^d} \phi(\bm{x} - \bm{\xi}) (\mathbf{u}(\bm{\xi}) - \mathbf{u}(\bm{x})) \rho(\bm{x})\rho(\bm{\xi}) \,\dd \bm{\xi}
\end{split}
\right.
\qquad \qquad
\bm{x}\in \mathbb{R}^d, \,t>0.
\end{equation}
In this system, $\rho = \rho(\bm{x},t)$ and $\bu = \bu(\bm{x},t)$ denote the macroscopic density and velocity of a continuum of agents, which interact through the nonnegative \textit{communication protocol} $\phi:\mathbb{R}^d\to \mathbb{R}$. This paper will focus on the case of \textit{unidirectional} dynamics, where all velocities point in the same direction. Without loss of generality, we write $\bu(\bm{x},t) = (u(\bm{x},t), 0, \ldots, 0)$ in this case, where $u(\bm{x},t)$ is real-valued. The unidirectional dynamics are then governed by the following system. (Here and below, we use $\partial_1$ to denote differentiation with respect to the first spatial variable, and we write $\bm{x} = (x,y)\in \mathbb{R}\times \R^{d-1}$.)
\begin{equation}
\label{e:EAuni}
\left\{
\begin{split}
\partial_t \rho + \partial_1 (\rho u) & = 0 \\
\partial_t (\rho u) + \partial_1 (\rho u^2) & = \int_{\mathbb{R}\times \mathbb{R}^{d-1}} \phi(x-\xi, y - \eta) (u(\xi, \eta) - u(x,y)) \rho(x,y) \rho(\xi, \eta) \,\dd\xi\,\dd\eta,
\end{split}
\quad (x,y)\in \mathbb{R}\times \mathbb{R}^{d-1},\,t>0.
\right.
\end{equation}
We are concerned with the existence, uniqueness, and stability of weak solutions of \eqref{e:EAuni}, as well as the long-time behavior of these weak solutions (and of solutions of the related Cucker--Smale system of ODEs, introduced below). We will make two key assumptions on our communication protocol for the entire manuscript. First, we will assume that $\phi=\phi(x,y)$ is even in both its arguments:
\begin{equation}
\label{e:phieven}
\phi(x,y) = \phi(-x,y) = \phi(x,-y),
\qquad \text{ for all } x\in \mathbb{R}, \, y\in \mathbb{R}^{d-1}.
\end{equation}
And second, we will assume that $\phi$ is bounded and Lipschitz in its second argument:
\begin{equation}
\label{e:phiLip2}
\sup_{(x,y)\in \mathbb{R}\times \mathbb{R}^{d-1}} \phi(x,y) + \sup_{\substack{x\in \mathbb{R},\, y_1, y_2\in \mathbb{R}^{d-1}\\y_1 \ne y_2}} \frac{|\phi(x,y_1) - \phi(x,y_2)|}{|y_1 - y_2|} < +\infty.
\end{equation}
In particular, we will \textit{not} analyze Euler Alignment dynamics equipped with a strongly singular communication protocol---this requires different tools. We will remark later on about the extent to which assumption \eqref{e:phiLip2} might be able to be weakened.
Finally, we will always normalize $\rho$ to have mass~$1$:
\begin{equation}
\int_{\mathbb{R}^d} \rho = 1.
\end{equation}
\subsection{Cucker--Smale-type systems and flocking dynamics}
The Euler Alignment system \eqref{e:EA} arises as a hydrodynamic limit of the celebrated Cucker--Smale system of ODE's \cite{CS2007a, CS2007b}:
\begin{equation}
\label{e:CS}
\left\{
\begin{split}
\dot{\bm{x}}_i & = \bm{v}_i,\\
\dot{\bm{v}}_i & = \sum_{\substack{j=1 \\ j\ne i}}^N m_j\phi(\bm{x}_i - \bm{x}_j)(\bm{v}_j - \bm{v}_i).
\end{split}
\right.
\end{equation}
The Cucker--Smale system as written above consists of $N$ agents, each associated to a triple $(m_i, \bm{x}_i, \bm{v}_i)$ indicating its mass $m_i\ge 0$ (fixed in time), position $\bm{x}_i = \bm{x}_i(t)\in \mathbb{R}^d$, and velocity $\bm{v}_i = \bm{v}_i(t)\in \mathbb{R}^d$. When the number of agents is very large, the use of \eqref{e:EA} becomes appropriate, and $\rho = \rho(\bm{x},t)$ and $\bu = \bu(\bm{x},t)$ denote the corresponding macroscopic quantities. Often one passes through a kinetic equation (not discussed here), see for example \cite{HT2008, HaLiu2008}, and then passes from the kinetic equation to \eqref{e:EA}, see \cite{ShvydkoyBook, KangVasseur2015, FigalliKang2019}, and also \cite{KMT2013, KMT2014, KMT2015} for the kinetic-to-hydrodynamic limit that includes a pressure term in the limiting system. In the framework developed below, we will circumvent the kinetic level and pass directly from~\eqref{e:CS} (equipped with collision rules) to \eqref{e:EA}.
One of the key features of Cucker--Smale-type systems is their compatibility with so-called `flocking dynamics,' where the velocities of all agents align in the time-asymptotic limit, and the diameter of the positions of the agents remains uniformly bounded for all time. More specifically, we say that \textit{flocking} occurs when
\begin{equation}
\label{e:flockingdef}
\sup_{t\ge 0} \mathscr{D}(t) \le \overline{\mathscr{D}}<+\infty,
\qquad
\mathscr{D}(t) = \max_{i,j}|\bm{x}_i(t) - \bm{x}_j(t)|,
\end{equation}
while \textit{velocity alignment} refers to the situation where
\begin{equation}
\label{e:velocityalignmentdef}
\lim_{t\to \infty} \mathscr{A}(t) = 0,
\qquad \mathscr{A}(t) = \max_{i,j}|\bm{v}_i(t) - \bm{v}_j(t)|.
\end{equation}
These definitions have natural counterparts in the hydrodynamic setting, which we will discuss later on. The behavior captured by \eqref{e:flockingdef} and \eqref{e:velocityalignmentdef} mimics---in a simplistic but analytically tractable way---the alignment dynamics of naturally occurring many-agent systems in the real-world, the prototypical example being a flock of birds. At the level of the equations, the driving mechanism appears in the `alignment forcing' on the right side of equations \eqref{e:EA}$_2$ and \eqref{e:CS}$_2$. In \eqref{e:CS}, for example, each term in the sum works to bring the velocity of agent~$i$ closer to that of agent~$j$, and the contribution of each term to the sum depends on the difference $\bm{v}_j(t) - \bm{v}_i(t)$ in velocities, the mass $m_j$ of agent~$j$, and the strength $\phi(\bm{x}_i(t)- \bm{x}_j(t))$ of the communication between agents $i$ and $j$. The emergence of a global structure (in this case, velocity alignment and flocking) from these possibly small-scale, pairwise interactions, is a complex phenomenon that provides a rich field of study. In particular, understanding exactly which initial configurations and which communication protocols will lead to flocking behavior is a subject of much active research and recent progress. We make no attempt to give a comprehensive overview of the by now expansive literature on the subject, but we refer the reader to the survey papers \cite{MT2014, CCP2017, Tadmor2021review, ShvydkoySurvey}, the book \cite{ShvydkoyBook}, and references therein, for a more thorough discussion of the state of the art.
\subsection{The main focus: wellposedness for unidirectional solutions}
Naturally, a prerequisite for the emergence of flocking in a solution of~\eqref{e:EA} or~\eqref{e:CS} is the global-in-time existence of that solution. For the ODE system~\eqref{e:CS}, existence and uniqueness of solutions is either trivial or well-understood under mild assumptions on~$\phi$. But wellposedness of the Euler Alignment system~\eqref{e:EA} remains open except in some important special cases. The most complete theory is available when~$d=1$ and additionally~$\phi$ is even and locally integrable.
In this case, one can predict from the initial data whether the solution will remain regular for all time or lose regularity in finite time \cite{TT2014,CCTT2016,tan2020euler,L2019CTC}. Furthermore, if the solution loses regularity in finite time, one can still continue it as a weak solution \cite{LeslieTan2023,HaHuangWang2014wk, Galtung2025}, and it is even possible to say a fair amount about the large-time structure of the density profile \cite{LeslieTan2024, Galtung2025}. In what follows, we will be concerned exclusively with the above-mentioned case where $\phi$ is locally integrable (and in fact satisfies \eqref{e:phiLip2}); however, a relevant parallel line of research concerns the case where $\phi$ is instead strongly singular near the origin; see for example \cite{KT,DKRT,ShvydkoyTadmorI,ShvydkoyTadmorII,ShvydkoyTadmorIII, Leslie2019, ArnaizCastro2019, DanchinMuchaPeszekWroblewski2018}.
Most of the literature on the 1D case leans heavily on the presence of an additional conserved quantity and also on the ordering property of the real line---properties that both disappear in higher dimensions. Largely for this reason, wellposedness remains much less well-understood in general than it is on~$\mathbb{R}$ (but see for example \cite{Shvydkoy2018NearlyAligned, tan2021eulerian, HT2016} for some partial results). There is, however, a class of dynamics for \eqref{e:EA} that is considerably richer than the pure 1D case but still comes with many of the technical advantages of the latter. We are referring to the so-called \textit{unidirectional} setting already mentioned above, in which $\mathbf{u}(\bm{x},t) = (u(\bm{x},t),0,\ldots, 0)$ for some scalar-valued function~$u$. In this setting, described by the system \eqref{e:EAuni}, particle trajectories are constrained to move along straight lines (preserving the ordering property from the 1D setting), and a special conserved quantity is still available. Therefore, for some purposes, the unidirectional initial value problem can be treated like a family of 1D problems. On the other hand, these 1D problems are all coupled together through the nonlocal communication, which causes considerable difficulty in understanding the regularity in the directions transverse to the flow and also allows for a fundamentally higher-dimensional richness in the system's emergent structures.
For bounded, Lipschitz, radial $\phi$, the existence of classical unidirectional solutions for `sub-critical' initial data was first proven in \cite{LearShvydkoy2019}. (See also \cite{Lear2021unising,LearShvydkoy2021, LiMiaoTanXue2024} for results in the strongly singular protocol regime.) As in the 1D case, solutions starting from `supercritical' initial data may lose regularity in finite time. However, unlike the 1D case, there is at present no theory for continuing such a solution beyond the blowup time. The main objective of this paper is to develop such a theory, and more generally, to establish the existence, uniqueness, and stability of unidirectional weak solutions of \eqref{e:EA}.
\subsection{The scalar balance law formulation}
The key observation that allowed for the development of the 1D weak solution theory in \cite{LeslieTan2023} was the reduction of the two-equation Euler Alignment system to a single scalar balance law, in the spirit of Brenier and Grenier's reduction of the pressureless Euler equations to a single scalar conservation law \cite{brenier1998sticky}. A similar reduction is available in the unidirectional setting; however, the result in this case is instead a \textit{family} of coupled scalar balance laws parametrized by $y$, and one must decompose the density profile properly in order to obtain it. We give a quick, formal derivation below (which mostly parallels the one in \cite{LeslieTan2023}); a rigorous justification will be provided in Section \ref{s:Recovery}.
We start by noting that \eqref{e:EAuni} can be rewritten in the form
\begin{equation}
\label{e:EAunipsi}
\left\{
\begin{split}
\partial_t \rho + \partial_1 (\rho u) & = 0 \\
\partial_t (\rho\psi) + \partial_1 (\rho\psi u) & = 0,
\end{split}
\right.
\qquad (x,y)\in \mathbb{R}\times \mathbb{R}^{d-1},\,t>0,
\end{equation}
where
\begin{equation}
\psi = u + \Phi*\rho,
\end{equation}
\begin{equation}
\Phi(x,y) = \int_0^x \phi(\xi, y)\, \dd\xi,
\end{equation}
and $*$ represents convolution in the spatial variables:
\begin{equation}
(\Phi*\rho)(x,y,t) = \int_{\mathbb{R}^{d-1}} \int_{\mathbb{R}} \Phi(x - \xi, y - \eta) \rho(\xi, \eta,t) \,\dd\xi\,\dd\eta.
\end{equation}
Next, we consider the density profile as a probability measure on $\mathbb{R}^d$ and disintegrate it with respect to horizontal slices. Letting $p_2:\mathbb{R}\times \mathbb{R}^{d-1}\to \mathbb{R}^{d-1}$ denote projection onto the second argument, i.e., $p_2(x,y) = y$, we define
\[
\nu = (p_2)_{\sharp} \rho(t)
\]
(note that unidirectionality guarantees that $\nu$ is time-independent), and then
\[
\rho(\dx,\dy,t) = \rho_y(\dx,t) \,\nu(\dy),
\]
where $\rho_y(t)$ is a probability measure on $\mathbb{R}$ for each $t$ and $\nu$-a.e. $y$. More specifically, the disintegration means that
\[
\int_{\mathbb{R}^{d-1}\times \mathbb{R}} f(x,y) \rho(\dx, \dy,t) =
\int_{\mathbb{R}^{d-1}} \bigg( \int_{\mathbb{R}} f(x,y) \rho_y(\dx,t)\bigg) \nu(\dy),
\]
for any bounded Borel measurable function $f:\mathbb{R}\times \mathbb{R}^{d-1}\to \mathbb{R}$. We will refer to $\nu$ as the `transverse measure' of the system.
We may therefore define
\[
M(x,y,t) = \rho_y((-\infty, x],t),
\qquad
Q(x,y,t) = \int_{(-\infty,x]} \psi(x,y,t) \rho_y(\dx,t).
\]
Integrating the system \eqref{e:EAunipsi} yields
\begin{equation}
\label{e:primitive}
\left\{
\begin{split}
\partial_t M + u \partial_1 M & = 0 \\
\partial_t Q + u \partial_1 Q & = 0.
\end{split}
\right.
\end{equation}
Given initial data $(M^0, Q^0)$ for this system (with $M^0$ nondecreasing), we may choose a function $A:[0,1]\times \mathbb{R}^{d-1}\to \mathbb{R}$ such that $A(M^0(x,y),y) = Q^0(x,y)$. (This choice is unique if $M^0$ is continuous.) For smooth solutions of \eqref{e:primitive}, this relationship propagates in time: $A(M(x,y,t),y) = Q(x,y,t)$, so that
\[
\partial_t M + \partial_x (A(M,y)) = -u\partial_1 M + \partial_1 Q = (\psi - u)\partial_1 M = B[\nu,M] \partial_1 M,
\]
where $B[\nu,M]$ is equal to $\Phi*\rho$ but is expressed in terms of $\nu$ and $M$. Altogether, our system is:
\begin{equation}
\label{e:ScalarBalance}
\left\{
\begin{split}
& \partial_t M(x,y,t) + \partial_x(A(M(x,y,t),y)) = B[\nu, M] \partial_x M(x,y,t), \\
& B[\nu, M](x,y,t) = (\Phi*(\nu \partial_1 M))(x,y,t) = \int_{\mathbb{R}^{d-1}} \int_\mathbb{R} \Phi(x-\xi, y-\eta) \,\partial_x M(\dd\xi, \eta,t)\,\nu(\dd\eta).
\end{split}
\right.
\end{equation}
\subsection{Entropy solutions}
The system \eqref{e:ScalarBalance} admits a natural entropy condition in the sense of Kruzkov. We require
\begin{equation}
\label{e:entropydistributional}
\partial_t |M-\alpha| \nu + \partial_x \big( \sgn(M-\alpha)(A(M,y) - A(\alpha, y))\big) \nu \le B[\nu, M]\,\partial_x|M-\alpha| \nu,
\qquad \text{ for all } \alpha\in \mathbb{R},
\end{equation}
in the sense of distributions on $\mathbb{R}\times \mathbb{R}^{d-1}\times (0,T)$. More explicitly, we require that for every $\alpha\in \mathbb{R}$ and every nonnegative $\zeta\in C^\infty_c(\mathbb{R}\times \mathbb{R}^{d-1}\times (0,T))$, the function $M$ should satisfy
\begin{equation}\label{eq:entropy-M}
\int_0^T\int_{\mathbb{R}^{d-1}}\int_{\mathbb{R}}
\Big[
|M-\alpha|\,\p_t\zeta
+ \sgn(M-\alpha)\big(A(M,y)-A(\alpha,y)\big)\,\p_x\zeta
+\zeta\,B[\nu, M]\,\p_x|M-\alpha|
\Big]\dx\,\nu(dy)\,\dt\ge0.
\end{equation}
Note that by taking $\zeta(x,y,t) = \widetilde{\zeta}(x,t)b_\e(y)$, where $b_\e$ is a standard mollifier, and taking $\e\to 0+$, we can reduce \eqref{eq:entropy-M} to the condition that for $\nu$-a.e. $y$, we have
\begin{equation}\label{eq:entropy-M-alt}
\int_0^T\int_\RR
\Big[
|M-\alpha|\,\p_t\widetilde{\zeta}
+ \sgn(M-\alpha)\big(A(M,y)-A(\alpha,y)\big)\,\p_x\widetilde{\zeta}
+\widetilde{\zeta}\,B[\nu, M]\,\p_x|M-\alpha|
\Big]\dx\,\dt\ge0,
\end{equation}
for every nonnegative $\widetilde{\zeta}\in C_c^\infty(\RR\times(0,T))$. In what follows, it will be useful to have access to both \eqref{eq:entropy-M} and \eqref{eq:entropy-M-alt}.
\medskip
We formalize the above in the following definitions.
\begin{DEF}
\label{def:SBIC}
We say that the triple $(M^0, A, \nu)$ of initial data, flux, and transverse measure is \textit{admissible} for the system \eqref{e:ScalarBalance} if the following conditions are satisfied.
\begin{itemize}
\item $\nu$ is a compactly supported probability measure on $\mathbb{R}^{d-1}$.
\item The function $M^0:\mathbb{R}\times \mathbb{R}^{d-1}\to \R$ is Borel measurable. Furthermore, there exists $R^0>0$ such that for $\nu$-a.e. $y$, we have that $M^0(x,y)=0$ for $x<-R^0$ and $M^0(x,y)=1$ for $x>R^0$. Finally, the function $x\mapsto M^0(x,y)$ is nondecreasing on $\mathbb{R}$ for $\nu$-a.e. $y$.
\item The function $A:[0,1]\times \mathbb{R}^{d-1}\to \mathbb{R}$ is Borel measurable, and normalized so that $A(0,y) = 0$ for $\nu$-a.e. $y\in \mathbb{R}^{d-1}$. Furthermore, the Lipschitz seminorm of the map $m\mapsto A(m,y)$ is uniformly bounded in $y$.
\end{itemize}
\end{DEF}
\begin{DEF}
\label{def:entropysoln}
Fix $T>0$. We say that a function $M\in C([0,T); L^1_{\loc}(\mathbb{R}\times \mathbb{R}^{d-1}, \dx \otimes \nu))$ is an \textit{entropy solution} of~\eqref{e:ScalarBalance} associated to the admissible triple $(M^0, A, \nu)$ if the following conditions are satisfied.
\begin{enumerate}[label = (\alph*)]
\item There exists $R(T)>0$ such that for all $t\in [0,T)$ and $\nu$-a.e. $y\in \mathbb{R}^{d-1}$, we have that $M(x,y,t)=0$ for $x<-R(T)$ and $M(x,y,t) = 1$ for $x>R(T)$. Furthermore, the function $x\mapsto M(x,y,t)$ is nondecreasing on $\mathbb{R}$ for $\nu$-a.e. $y\in \mathbb{R}^{d-1}$ and for all $t\in [0,T)$.
\item $M$ satisfies \eqref{e:entropydistributional} in the sense of distributions on $\mathbb{R}\times \mathbb{R}^{d-1} \times (0,T)$.
\item We have
\begin{equation}
\label{e:MIC}
\lim_{t\to 0+} \|M(\cdot, y,t) - M^0(\cdot, y)\|_{L^1(\mathbb{R}\times \mathbb{R}^{d-1}, \dx \otimes \nu)} = 0.
\end{equation}
\end{enumerate}
\end{DEF}
\begin{REMARK}
In light of the requirement \eqref{e:MIC}, we will sometimes use $M(x,y,0)$ (implicitly or explicitly) to mean $M^0(x,y)$. This will occasionally save us from writing down separate notation for quantities defined in terms of the initial conditions.
\end{REMARK}
\subsection{Analytical framework and preliminary statement of results}
The analysis of the present work begins essentially where the 1D theory of \cite{LeslieTan2023} leaves off. Most of the core results of the latter can be thought of as proving the existence, uniqueness, and stability of entropy solutions of \eqref{e:ScalarBalance} in the special case where $\nu$ is the Dirac mass at a single point in $\mathbb{R}^{d-1}$. In fact, as discussed in Section \ref{ss:Existdisc} below, this theory can be extended without too much trouble to the case where $\nu$ is any fixed, atomic measure.
It is natural, then, to try to construct entropy solutions of \eqref{e:ScalarBalance} as the limit of a sequence of solutions $M^k(x,y,t)$ associated to admissible triples $(M^{0,k}, A^k, \nu_k)_{k=1}^\infty$, where $\nu_k$ is atomic---indeed, this is more or less what we ultimately do in the present work. However, meaningfully measuring the difference between two entropy solutions $M$, $\widetilde{M}$ associated to distinct transverse measures $\nu$, $\widetilde{\nu}$ (which may not even have the same support in $\mathbb{R}^{d-1}$) requires us to be able to compare values of $M$ and $\widetilde{M}$ associated to different horizontal slices. To resolve this difficulty, we embed the optimal coupling between two such measures (with respect to the cost $c(y,\widetilde{y}) = |y-\widetilde{y}|$), which we denote by $\bpi(\dy, \dd\widetilde{y})$, across our analytical framework. Our stability results, for instance, are statements about the quantity $\int_{\mathbb{R}^{d-1}} \|M(\cdot, y, t) - \widetilde{M}(\cdot, \widetilde{y}, t)\|_{L^1(\mathbb{R})} \bpi(\dy, \dd\widetilde{y})$. We believe this is the most natural way to develop the theory. Note that this quantity is related to the usual Wasserstein-1 distance between the underlying density profiles via
\begin{equation}
\label{e:Wassersteinrelationship}
W_1(\rho(t), \widetilde{\rho}(t)) \le \int_{\mathbb{R}^{d-1}} \|M(\cdot, y, t) - \widetilde{M}(\cdot, z, t)\|_{L^1(\mathbb{R})} \bpi(\dy, \dz) + W_1(\nu, \widetilde{\nu}),
\end{equation}
where the notation `$W_1$' is used to denote the Wasserstein-1 distance in both $d$ and $d-1$ dimensions (and will also be used below to denote the same in $1$ dimension). One can see that this inequality holds by realizing that the integral on the right side can be written as
\[ \int_{\mathbb{R}^{d-1}} \!\!\|M(\cdot, y, t) - \widetilde{M}(\cdot, \widetilde{y}, t)\|_{L^1(\mathbb{R})} \bpi(\dy, \dd\widetilde{y})
= \!\!\int_{\mathbb{R}^{d-1}} \!\!\!W_1(\rho_y(t), \widetilde{\rho}_{\widetilde{y}}(t)) \,\bpi(\dy, \dd\widetilde{y})
=\! \int_{\mathbb{R}^{d-1}}\!\int_\mathbb{R} |x-\widetilde{x}| \,\gamma_{y, \widetilde{y}}(\dx, \dd\widetilde{x},t) \,\bpi(\dy, \dd\widetilde{y}),
\]
where $\gamma_{y, \widetilde{y}}(t)$ is an optimal coupling between $\rho_y(t)$ and $\widetilde{\rho}_{\widetilde{y}}(t)$. Since $\gamma_{y, \widetilde{y}}(\dx, \dd\widetilde{x},t) \bpi(\dy,\dd\widetilde{y})$ is a coupling between $\rho(t)$ and $\widetilde{\rho}(t)$, we have
\begin{align*}
W_1(\rho(t), \widetilde{\rho}(t))
& \le \int_{\mathbb{R}^{d-1}} \int_{\mathbb{R}} (|x-\widetilde{x}| + |y - \widetilde{y}|)\, \gamma_{y, \widetilde{y}}(\dx, \dd\widetilde{x},t) \,\bpi(\dy, \dd\widetilde{y}) \\
& = \int_{\mathbb{R}^{d-1}} \|M(\cdot, y, t) - \widetilde{M}(\cdot, z, t)\|_{L^1(\mathbb{R})} \bpi(\dy, \dz) + W_1(\nu, \widetilde{\nu}),
\end{align*}
as claimed.
We are now in a position to summarize our results and approach.
\begin{itemize}[itemsep = 0.56em, leftmargin = 2em]
\item \textbf{Uniqueness and Stability of Entropy Solutions.} In Section \ref{s:Stability}, we use the basic Kruzkov doubling-of-the-variables technique (c.f. \cite{Kruzkov1970, Lucier1986}), adapted to the unidirectional framework described above, to establish two different stability estimates on entropy solutions of \eqref{e:ScalarBalance} generated by admissible triples $(M^0, A, \nu)$ and $(\widetilde{M}^0, \widetilde{A}, \widetilde{\nu})$. The first bound, which generalizes a 1-dimensional estimate from \cite{LeslieTan2023}, grows linearly in time and depends on the Lipschitz seminorm (in the first component) of the flux differences. The second bound grows exponentially in time but relies only on the $L^\infty$ difference (in the first component) between fluxes, making it a crucial tool in our existence argument later on. In general, the discretized fluxes from our approximation scheme converge to $A$ in an $L^1$-$L^\infty$ sense (made precise in Section \ref{ss:Existgen}) but might not converge in the stronger sense where $\|\cdot\|_{L^\infty}$ is replaced by the Lipschitz seminorm. This renders the linear-in-time stability estimate insufficient for passing to the limit for general initial data. Our approach to the second stability bound is inspired by the work \cite{BouchutPerthame1998} of Bouchut and Perthame. The full statement of our uniqueness and stability result is contained in Theorem \ref{th:uniqueness-main}.
\item \textbf{Existence of Entropy Solutions.} In Section \ref{s:Existence}, we prove existence of solutions associated to an admissible triple $(M^0, A, \nu)$. Our argument proceeds in two steps. First, we consider the special case where $\nu$ is atomic and construct the unique entropy solution associated to $(M^0, A, \nu)$ as the limit of a sequence of `sticky particle Cucker--Smale' (SPCS) solutions. We describe the SPCS dynamics in Section \ref{ss:Existdisc}. Since this first part of the construction closely parallels that of \cite{LeslieTan2023}, we omit arguments that are essentially unchanged. Next, we discretize an arbitrary admissible triple in the $y$-variable to obtain a sequence $(M^{0,k}, A^k, \nu_k)_{k=1}^\infty$ of admissible approximants. We generate an associated sequence of entropy solutions $(M^k)_{k=1}^\infty$ and pass to the limit, using our previously established stability estimates. Our limiting procedure is shown to preserve the entropy conditions, thus finishing the existence proof.
\item \textbf{Recovery and Approximation of the Euler Alignment dynamics.} In Section \ref{s:Recovery}, we rigorously justify our use of the system \eqref{e:ScalarBalance} to describe the unidirectional Euler Alignment system \eqref{e:EAuni}. We show how to generate an admissible triple $(M^0, A, \nu)$ from initial conditions $(\rho^0, u^0)$, and conversely how to recover a solution $(\rho, u)$ of \eqref{e:EAuni} from an entropy solution $M$ of \eqref{e:ScalarBalance}. Theorem \ref{t:entropicselection} gives the precise statement. We also demonstrate how our weak solutions can be approximated by sticky particle Cucker--Smale dynamics in Theorem \ref{t:SPCSconvgen}, Theorem \ref{t:SPCSconvrateweak}, and Corollary \ref{c:SPCSratestrong}. The last two of these require additional assumptions on the initial data $(\rho^0, u^0)$ but provide Wasserstein-$1$ rates of convergence for the density profile.
\item \textbf{Flocking.} In Section \ref{s:Flocking}, we turn to the question of long-time behavior. We show that, under natural assumptions on the communication protocol $\phi$ and the initial density profile, solutions of the Cucker--Smale ODE's associated to unidirectional initial data must experience flocking and velocity alignment, with a rate that is independent of the number of agents. The novelty here is that our assumptions allow for $\supp \phi$ to have a cylinder of degeneracy around the $x$-axis. This has significant implications for our understanding of the
fundamental mechanism underlying the emergent behavior: Communication in the direction of motion is not a necessary condition for flocking to occur. We state flocking theorems for both the Cucker--Smale and Euler Alignment systems, but we write down proofs only for the discrete setting. A blueprint for how to extend the results to the continuum level is contained in \cite{LeslieTan2023}.
\end{itemize}
\subsection{Additional commentary and related literature}
\subsubsection{Degenerate protocols}
A few words are in order about our flocking result (see Section \ref{s:Flocking} for additional context). First of all, despite the implications mentioned above, our proof of flocking does not require any special tools. On the contrary, it is a rather simple adaptation of an existing argument (found, for instance, in \cite{ShvydkoyBook}) that is used to establish flocking for any initial data under the assumption of a radial, heavy-tailed protocol (i.e., $\phi\in L^1_{\loc}(\mathbb{R}^d)\backslash L^1(\mathbb{R}^d)$). We speculate that the reason it has gone unnoticed until now is because our argument as written really only applies in the context of unidirectional dynamics in two or more spatial dimensions. It would be very interesting to try to extend the paradigm suggested by our theorem to more general `almost unidirectional' dynamics, c.f. \cite{LearShvydkoy2019}. We leave the exploration of this problem for future work.
Our flocking result also shares some similarity to work of Dietert and Shvydkoy \cite{DietertShvydkoy2019}, which used a sophisticated corrector method to prove flocking for arbitrary initial conditions when the communication protocol is radial and degenerate on a ball near the origin. We emphasize that despite the similarity in spirit, the two results cover essentially disjoint regimes: Ours requires a very special geometry in the initial conditions but is more flexible in the allowed degeneracy of communication; on the other hand, the work of Dietert and Shvydkoy can handle arbitrary initial conditions but requires the region of communication degeneracy to be bounded.
\subsubsection{Limiting configurations}
Similar to many previous results, our flocking theorem allows us to deduce the existence of a limiting density profile $\rho_\infty$ to which $\rho(t)$ converges (in an appropriate reference frame, and in the weak-$*$ sense of measures). The understanding the structure of this limiting configuration is an interesting but subtle question, which has been studied for 1D weak solutions of the Euler Alignment system \cite{LeslieTan2024, Galtung2025}, 1D and unidirectional classical solutions of the Euler Alignment system \cite{LS2019, LLST2020geometric,Lear2021unising}, and 1D Cucker--Smale systems \cite{HaKimParkZhang2019, HaParkZhang2018}. Much remains unknown even in the already-studied regimes, but the present work opens the door to the consideration of an additional setting, namely that of unidirectional weak solutions.
\subsubsection{Comparison with \cite{CarrilloChoiTadmor2026}}
A recent paper by Carrillo, Choi, and Tadmor \cite{CarrilloChoiTadmor2026} (c.f. also \cite{Tadmor2021review, Tadmor2026}) considers the following version of the Euler Alignment system, with the Reynolds stress $\tau$ determined from the Lagrangian dynamics of the underlying kinetic equation (not discussed here):
\begin{equation}
\label{e:EAtau}
\left\{
\begin{split}
\partial_t \rho + \nabla \cdot (\rho \mathbf{u}) & = 0 \\
\partial_t (\rho \bu) + \nabla \cdot (\rho \bu \otimes \bu + \tau)
& = \int_{\R^d} \phi(\bm{x} - \bm{\xi}) (\mathbf{u}(\bm{\xi}) - \mathbf{u}(\bm{x})) \rho(\bm{x})\rho(\bm{\xi}) \,\dd \bm{\xi},
\end{split}
\right.
\qquad \qquad
\bm{x}\in \mathbb{R}^d, \,t>0.
\end{equation}
The authors of \cite{CarrilloChoiTadmor2026} develop a wellposedness theory for \eqref{e:EAtau} and show that, under natural assumptions on $\phi$, the Reynolds stress vanishes in the time-asymptotic limit (though it may be nontrivial for all finite time). Of course, if $\tau \equiv 0$ for all time, then \eqref{e:EA} and \eqref{e:EAtau} are the same. The wellposedness theory provided by \cite{CarrilloChoiTadmor2026} therefore reduces the question of global-in-time regularity of solutions to \eqref{e:EA} to a question of whether or not $\tau$ is ever nontrivial for the dynamics of the underlying kinetic system. In the regime covered by our work, on the other hand (which is limited to unidirectional dynamics), the primary interest is in the case where the Lagrangian trajectories associated with the kinetic system would intersect, generating a nonzero Reynolds stress. Therefore, our results are essentially disjoint from theirs. However, it would be interesting to compare the solutions of the system \eqref{e:EAtau} provided by \cite{CarrilloChoiTadmor2026} generated from unidirectional data to those of \eqref{e:EAuni} provided by the present work.
\subsubsection{Weakly singular protocols} The analysis of the second author and Tan in \cite{LeslieTan2023} develops the wellposedness theory for 1D entropic solutions under the assumption of an arbitrary symmetric communication protocol $\phi\in L^1_{\loc}(\mathbb{R})$. It could be expected that the unidirectional theory might be available for a class of protocols as broad as those satisfying \eqref{e:phieven} and
\begin{equation}
\label{e:phiL1locx}
\sup_{y\in \mathbb{R}^{d-1}} \int_{[-R,R]} \phi(x,y)\,\dx <+\infty,
\qquad \text{ for any fixed } R>0.
\end{equation}
Certain aspects of our analysis below (e.g., uniqueness of solutions, flocking for Cucker--Smale dynamics) do indeed still work for such protocols, with no essential changes. Other aspects (e.g., our Bouchut--Perthame-type stability estimate) have little hope of being adapted successfully to cover the case of \eqref{e:phiL1locx}. The authors are cautiously optimistic that the framework established in this paper can serve as a basis for the development of a weak solution theory that includes protocols satisfying~\eqref{e:phiL1locx}, and we plan to explore this direction in future work. Relaxing the assumption \eqref{e:phiLip2}, however, appears to require sufficiently different ideas and techniques that the resulting theory would be more naturally developed in separate work.
\subsubsection{Sticky particle dynamics and the pressureless Euler equations} As has already been mentioned above, the reduction of the 1D Euler Alignment system to a single scalar balance law was a key feature of the analysis of \cite{LeslieTan2023}. This reduction allowed the authors to adapt the framework of Brenier and Grenier's work \cite{brenier1998sticky} on the pressureless Euler equations. Both \cite{brenier1998sticky} and \cite{LeslieTan2023} approximate the solutions to the scalar equation using discrete sticky particle dynamics. Generally speaking, the structural similarities (to a point) between the well-studied pressureless Euler equations and the Euler Alignment system has been a productive direction in recent years. We highlight in particular the paper of Galtung \cite{Galtung2025}, who exhibited a gradient flow structure for the 1D Euler Alignment system using techniques developed by Natile and Savar\'e \cite{NatileSavare2009} for the pressureless Euler system, and demonstrated that the solutions generated this way were equivalent to the ones from \cite{LeslieTan2023}. It is beyond the scope of this paper to give a complete overview of the literature on the pressureless Euler system, but a partial list of references is as follows:
\cite{Zeldovich1969sb,Grenier1995,ERykovSinai1996,WangDing1997,WangHuangDing1997,BouchutJames1998, BouchutJames1999, HuangWang2001stickyunique,nguyen2008pressureless,NatileSavare2009,BrenierGangboSavareWestdickenberg2013,NguyenTudorascu2015,CavallettiSedjroWestdickenberg2015, Hynd2019LagrangianSP, Hynd2020trajectory, Hynd2020probmeasureSP, BianchiniDaneri2023, carrillogaltung2025}. These references cover many different viewpoints, and it is a subject of continuing interest to explore which of the techniques are applicable for the Euler Alignment equations and related systems.
\subsection{Notation and Conventions}
\label{ss:notation}
Most of the notation we use is standard, but let us collect a few conventions here for the convenience of the reader. Unless otherwise specified, `$C$' will denote a constant that may change from line to line but is independent of any quantities that may later be taken to zero or infinity. All functions and measures we consider will be defined on some Euclidean space or a subset thereof, unless explicitly specified otherwise. We will use $[f]_{\Lip}$ to denote the Lipschitz seminorm of a function $f$, i.e., $[f]_{\Lip} = \sup_{x_1\ne x_2} \frac{|f(x_1) - f(x_2)|}{|x_1 - x_2|}$. We use the word `measurable' to mean `Borel measurable' unless explicitly noted otherwise.
We will use $\cM$ to denote the space of signed Radon measures, and $\cP$ will denote the space of probability measures (with $\cP_c$ denoting probability measures with compact support). We will always equip $\cM$ and $\cP$ with the weak-$*$ topology, defined via convergence with respect to test functions $\zeta\in C_c$ which are continuous and compactly supported. (We will use the standard notation $\stackrel{*}{\rightharpoonup}$ to denote weak-$*$ convergence.) However, all the probability measures we will consider will belong to a family whose support is contained in some compact set. For such a family, weak-$*$ convergence is equivalent to weak convergence (i.e., convergence with respect to test functions which are continuous and bounded), and the weak topology is metrizable via the Wasserstein-1 metric, or equivalently the bounded Lipschitz distance. For probability measures $\mu$ and $\widetilde{\mu}$ on $\mathbb{R}^k$, this is defined as follows.
\begin{equation}
W_1(\mu, \widetilde{\mu}) = \inf_{\bpi} \int\!\!\!\int |y-\widetilde{y}|\, \bpi(\dy, \dd\widetilde{y})
= \sup\bigg\{ \int f(y) \,\mu(\dy) - \int f(y) \,\widetilde{\mu}(\dy) : \sup |f|, [f]_{\Lip} \le 1\bigg\},
\end{equation}
where the infimum in the first expression is taken over all couplings $\bpi$ whose first marginal is $\mu$ and whose second marginal is $\widetilde{\mu}$, i.e., all $\bpi\in \cP(\mathbb{R}^k\times \mathbb{R}^k)$ such that $\bpi(E_1\times \mathbb{R}^k) = \mu(E_1)$ and $\bpi(\mathbb{R}^k\times E_2) = \widetilde{\mu}(E_2)$ for all Borel sets $E_1, E_2$ of $\mathbb{R}^k$. We will use $W_1$ to denote the Wasserstein-$1$ distance in any spatial dimension $k$ (though $k=1$, $k=d-1$, and $k=d$ will be the only cases we use). The value of $k$ will be clear from context.
Given $\mu\in \cM$ and a measurable function $f$, we denote by $f_{\sharp} \mu$ the \textit{pushforward} measure of $\mu$ by $f$, defined via $f_{\sharp} \mu(E) = \mu(f^{-1}(E))$. This implies that $\int g(x) (f_\sharp \mu)(\dx) = \int g\circ f(x)\, \mu(\dx)$ for any measurable function $g$ such that $g\circ f\in L^1(\mu)$.
For a right-continuous, nondecreasing function $f:\mathbb{R}\to [0,1]$, we define its \textit{generalized inverse} via
\begin{equation}
\label{e:geninvdef}
f^{-1}(m) = \inf\{ x\in \mathbb{R} : f(x)\ge m\}.
\end{equation}
If $f:\mathbb{R}\to [0,1]$ is the cumulative distribution function of a probability measure $\mu$ on $\mathbb{R}$, i.e., $f(x) = \mu((-\infty, x])$, then $\mu$ is the pushforward of Lebesgue measure on $\mathbb{R}$ under $f^{-1}$. If $\widetilde{\mu}$ is another probability measure on $\mathbb{R}$ with cumulative distribution function $\widetilde{f}$, then
\begin{equation}
\label{e:W1id}
W_1(\mu, \widetilde{\mu}) = \|f - \widetilde{f}\|_{L^1(\mathbb{R})} = \|f^{-1} - \widetilde{f}^{-1}\|_{L^1([0,1])}.
\end{equation}
Furthermore, if $g$ is bounded and measurable, then
\begin{equation}
\label{e:geninvid}
\int_{(-\infty,x]} g(\xi) \,\mu(\dd\xi) = \int_0^{f(x)} (g\circ f^{-1})(m)\,\dm,
\qquad \text{ for all } x\in \mathbb{R}.
\end{equation}
\section{Uniqueness and stability}
\label{s:Stability}
This section is dedicated to the proof of uniqueness and stability of entropy solutions of \eqref{e:ScalarBalance}. The precise statement we prove is as follows. Here and below, we denote the gradient of $\phi(x,y)$ with respect to its second argument by $\n_2 \phi$.
\begin{THEOREM}\label{th:uniqueness-main}
Assume $\phi$ satisfies \eqref{e:phieven} and \eqref{e:phiLip2}. Let $(M^0, A, \nu)$ and $(\widetilde{M}^0, \widetilde{A}, \widetilde{\nu})$ be two admissible triples. Assume $M$ and $\widetilde{M}$ are entropy solutions of \eqref{e:ScalarBalance} on $[0,T)$ associated to these triples, and that $R(T)$ satisfies the requirements of Definition \ref{def:entropysoln} for both $M$ and $\widetilde{M}$. Let $\bpi$ be an optimal coupling between $\nu$ and $\widetilde{\nu}$, associated to the cost $c(y,\widetilde{y}) = |y-\widetilde{y}|$. Then we have the following estimates, for $0\le s \le t < T$.
First, we have
\begin{equation}\label{eq:stab-pi}
\begin{aligned}
& \int\!\|M(\cdot,y,t)-\widetilde{M}(\cdot,\widetilde{y},t)\|_{L^1}\,\bpi(\dd y,\dd \widetilde{y}) \\
& \le
\int\!\bigg[\|M(\cdot,y,s)-\widetilde M(\cdot,\widetilde{y},s)\|_{L^1} + (t-s) [A(\cdot,y)-\widetilde{A}(\cdot,\widetilde{y})]_{\Lip} \bigg] \,\bpi(\dd y,\dd \widetilde{y})
+(t-s)\,8 R(T)\,
\|\n_2\phi\|_{L^\infty}
\,W_1(\nu,\tnu).
\end{aligned}
\end{equation}
In the special case where $\nu = \widetilde{\nu}$, this estimate (without the last term on the right) can be obtained under the weaker assumption \eqref{e:phiL1locx} instead of the Lipschitz assumption on the second argument.
Second, denoting
\begin{equation}\label{eq:BP-E}
\mathcal{D}(\tau):=
\int\!\|M(\cdot,y,\tau)-\widetilde M(\cdot,z,\tau)\|_{L^1}\,\bpi(\dd y,\dd z),\qquad \tau\in [0,T).
\end{equation}
\begin{equation}
\label{e:Ainftydef}
\cA_\infty:=
\int\!\|A(\cdot,y)-\widetilde A(\cdot,\widetilde{y})\|_{L^\infty([0,1])}\,\bpi(\dd y,\dd \widetilde{y}),
\end{equation}
we also have the following, for some absolute constant $C>0$.
\begin{equation}\label{eq:main-unidir-optimal}
\begin{aligned}
\cD(t)
& \le
\bigg( \cD(s)
+ C \sqrt{ 2R(T) \cA_\infty (t-s)}
+
C(t-s)
\, \big( \cA_\infty + R(T)\|\n_2\phi\|_{L^\infty} W_1(\nu,\widetilde\nu) \big) \bigg)e^{C\|\phi\|_{L^\infty}(t-s)}.
\end{aligned}
\end{equation}
\end{THEOREM}
The remainder of this section is dedicated to the proof of Theorem \ref{th:uniqueness-main}, which we split across three subsections. The proofs of both \eqref{eq:stab-pi} and \eqref{eq:main-unidir-optimal} rely on a doubling-of-the-variables argument, which we set up in Section \ref{ss:stabsetup} before specializing to \eqref{eq:stab-pi} in Section \ref{ss:stab1} and \eqref{eq:main-unidir-optimal} in Section \ref{ss:stab2}. Before we begin the proof in earnest, we pause to recall the BV chain rule (c.f. \cite{AmbrosioDalMaso1990}), which we will use repeatedly in our stability estimates (mostly without comment) and also later in Section \ref{s:Recovery}.
\begin{LEMMA}
\label{l:BVChain}
Suppose $W\in BV_{\loc}(\mathcal{U})$, where $\mathcal{U}$ is an open subset of $\mathbb{R}^k$. Assume also that $f$ is a real-valued Lipschitz function whose domain of definition includes the image of $W$. Then $f\circ W\in BV_{\loc}(\mathcal{U})$, and $|\partial_i (f\circ W)| \le [f]_{\Lip} |\partial_{i} W|$ for each $i=1,\ldots, k$ in the sense of measures. Furthermore, there exists a single bounded Borel measurable function $g$ on $\mathcal{U}$, with $|g|\le [f]_{\Lip}$, such that $\partial_i (f\circ W) = g \partial_i W$ for each $i=1,\ldots, k$ on $\mathcal{U}$.
\end{LEMMA}
\subsection{Doubling of the variables and preliminary estimates}
\label{ss:stabsetup}
We begin by following the usual Kruzkov doubling-of-the-variables argument. We substitute $\alpha=\widetilde M(\widetilde x,\widetilde{y},\widetilde\tau)$ and $\widetilde{\zeta}(x,\tau)=w(x,\tau,\widetilde x,\widetilde\tau)$ into \eqref{eq:entropy-M-alt}:
\begin{equation}
\label{e:Mdif1}
0 \le \int_0^T\int_\RR \Big[
|M-\widetilde M|\,\p_\tau w + \sgn(M-\widetilde M)\big(A(M,y)-A(\widetilde M,y)\big)\,\p_x w + w\,B[\nu,M]\,\p_x |M-\widetilde M|
\Big] \, \dx\, \dtau .
\end{equation}
Similarly,
\begin{equation}
\label{e:Mdif2}
0 \le \int_0^T\int_\RR \Big[
|M-\widetilde M|\,\p_{\widetilde\tau} w + \sgn(\widetilde M-M)\big(\widetilde A(\widetilde M,\widetilde{y})-\widetilde A(M,\widetilde{y})\big)\,\p_{\widetilde x} w + w\,B[\widetilde\nu,\widetilde M]\,\p_{\widetilde x} |M-\widetilde M|
\Big] \, \dd \widetilde x\, \dd \widetilde\tau .
\end{equation}
We integrate \eqref{e:Mdif1} over $\widetilde{x}$ and $\widetilde{\tau}$ and \eqref{e:Mdif2} over $x$ and $\tau$, then add the results together. We also add and subtract the quantity $\sgn(M-\widetilde{M})(A(M,y) - A(\widetilde{M},y))\partial_{\widetilde{x}} w$, for reasons that will become clear below. To keep the notation as light as possible, we omit the domain of integration for all spatial integrals. The domain of integration will always be all of $\mathbb{R}$ for integrals in $x$ and $\widetilde{x}$. The result of these manipulations is as follows.
\begin{equation}
\begin{aligned}\label{eq:prior-commonest}
0 & \le \int_0^T \!\!\int_0^T\!\! \int\!\!\!\int
|M-\widetilde M|(\p_\tau w+\p_{\widetilde\tau}w)
\,\dx\,\dd \widetilde x\,\dd \tau\,\dd \widetilde\tau
\\
& \quad +\int_0^T \!\!\int_0^T\!\! \int\!\!\!\int
\sgn(M-\widetilde M)\big(A(M,y)-A(\widetilde M,y)\big)
(\p_x w+\p_{\widetilde x}w)
\,\dd x\,\dd \widetilde x\,\dd \tau\,\dd \widetilde\tau
\\
& \quad -\int_0^T \!\!\int_0^T\!\!\int\!\!\!\int
\sgn(M-\widetilde M)
\Big(
(A(M,y)-\widetilde A(M,\widetilde{y}))
-
(A(\widetilde M,y)-\widetilde A(\widetilde M,\widetilde{y}))
\Big)
\p_{\widetilde x}w
\,\dd x\,\dd \widetilde x\,\dd \tau\,\dd \widetilde\tau
\\
& \quad +\int_0^T \!\!\int_0^T\!\!\int\!\!\!\int
w\Big(
B[\nu,M]\p_x|M-\widetilde M|
+
B[\widetilde\nu,\widetilde M]\p_{\widetilde x}|M-\widetilde M|
\Big)\,\dd x\,\dd \widetilde x\,\dd \tau\,\dd \widetilde\tau .
\end{aligned}
\end{equation}
Now, we fix a nonnegative even function $b\in C^\infty_c(\mathbb{R})$ with $\supp(b)\subset (-1,1)$, $\int b = 1$. Define $b_\e(x) = \frac{1}{\e} b\big( \frac{x}{\e} \big)$ for each $\e>0$, so that $b_\e\stackrel{*}{\rightharpoonup} \delta$ as $\e\to 0+$. Next, fix $s,t\in [0,T)$ with $s<t$. For $0<\lambda<\min\{t-s, T-t\}$, let $h_\lambda:[0,\infty)\to \mathbb{R}$ be a function which is identically zero outside $[s,t+\lambda]$, identically $1$ on $[s+\lambda, t]$, and linear on $[s,s+\lambda]$ and $[t, t+\lambda]$. As $\lambda\to 0+$, we have $h_\lambda\stackrel{*}{\rightharpoonup} 1_{[s,t]}$ and $h_\lambda'\stackrel{*}{\rightharpoonup}\delta(\cdot - s) - \delta(\cdot - t)$. Define
\[
w_{\Delta, \e, \lambda}(x,\tau,\widetilde{x},\widetilde{\tau}) = b_{\Delta}(x-\widetilde{x}) b_\e(\tau-\widetilde{\tau}) h_\lambda(\tau).
\]
Then
\[
(\partial_\tau + \partial_{\widetilde{\tau}})w_{\Delta, \e, \lambda} = b_\Delta(x-\widetilde{x}) b_\e(\tau - \widetilde{\tau}) h_\lambda'(\tau),
\]
\[
(\partial_x + \partial_{\widetilde{x}})w_{\Delta, \e, \lambda} = 0.
\]
We substitute $w = w_{\Delta, \e, \lambda}$ in \eqref{eq:prior-commonest} (note that the second integral on the right vanishes completely) and take $\e\to 0+$, then $\lambda\to 0+$. After an additional integration, \eqref{eq:prior-commonest} becomes
\begin{equation}\label{e:commonest}
\mathcal{D}_{\Delta}(t)\le \mathcal{D}_{\Delta}(s)+\int_s^t \mathcal F_{\Delta}(\tau)\,\dd \tau+\int_s^t \mathcal N_{\Delta}(\tau)\,\dd \tau,
\end{equation}
where
\begin{equation}\label{eq:BP-E-EDelta}
\mathcal{D}_{\Delta}(t):=
\int\!\!\!\int\!\!\!\int\!|M(x,y,t)-\widetilde M(\widetilde{x},\widetilde{y},t)| b_{\Delta}(x-\widetilde x)\,\dd x\,\dd \widetilde{x}\,\bpi(\dd y,\dd z),
\end{equation}
\begin{equation}\label{eq:BP-FDelta-def}
\begin{aligned}[t]
\mathcal F_{\Delta}(\tau)
&:=\!\!
\int\!\!\!\int\!\!\!\int
\operatorname{sgn}\!\big(M(x,y,\tau)-\widetilde M(\widetilde x,\widetilde{y},\tau)\big)
\Big(
(A(M(x,y,\tau),y)-\widetilde A(M(x,y,\tau),\widetilde{y}))
\\
&\hspace{53 mm}
-
(A(\widetilde M(\widetilde x,\widetilde{y},\tau),y)-\widetilde A(\widetilde M(\widetilde x,\widetilde{y},\tau),\widetilde{y}))
\Big)
\,b_{\Delta}'(x-\widetilde x)\,\mathrm{d}x\,\mathrm{d}\widetilde x\;\bpi(\mathrm{d}y,\mathrm{d}\widetilde{y}),
\end{aligned}
\end{equation}
and
\begin{equation}
\label{e:NL}
\begin{aligned}
\mathcal{N}_\Delta(\tau) & = \int\!\!\!\int\!\!\!\int
\Big(
B[\nu,M](x,y,\tau)\p_x|M-\widetilde M|
+
B[\widetilde\nu,\widetilde M](\widetilde x,\widetilde{y},\tau)\p_{\widetilde x}|M-\widetilde M|
\Big)
b_\Delta(x-\widetilde{x})
\,\dd x\,\dd \widetilde x\;\bpi(\dd y,\dd \widetilde{y}).
\end{aligned}
\end{equation}
We refer to $\mathcal{F}_\Delta$ (or $\int_s^t \mathcal{F}_\Delta(\tau)\,\dtau$) as the `flux term' and $\mathcal{N}_\Delta$ (or $\int_s^t \mathcal{N}_\Delta(\tau)\,\dtau$) as the `nonlocal term' in what follows.
\medskip
The inequality \eqref{e:commonest} serves as a starting point for the proofs of both \eqref{eq:stab-pi} and \eqref{eq:main-unidir-optimal}. We now prove each of these in turn.
\subsection{First stability estimate}
\label{ss:stab1}
In order to deduce \eqref{eq:stab-pi} from \eqref{e:commonest}, we would like to take $\Delta\to 0+$. However, before we can do so, we need to move the derivative that appears on $b_\Delta$ in the definition of $\cF_\Delta$. We integrate by parts and use the BV chain rule to estimate the result. We begin by noting that $\widetilde{M}(\widetilde{x},z,t)$ does not depend on $x$, and furthermore that the function
\[
m\mapsto \sgn(m - \widetilde{m})\big[(A(m,y) - \widetilde{A}(m,\widetilde{y})) - (A(\widetilde{m},y) - \widetilde{A}(\widetilde{m},\widetilde{y})) \big]
\]
is Lipschitz in $m$, uniformly in $\widetilde{m}$, with Lipschitz constant not exceeding $[A(\cdot, y) - \widetilde{A}(\cdot, \widetilde{y})]_{\Lip}$. Therefore,
\begin{equation}\label{eq:BP-flux-by-parts}
\begin{aligned}
&\bigg| \int_s^t \int\!\!\!\int
\sgn(M-\widetilde M)
\Big(
(A(M,y)-\widetilde A(M,\widetilde{y}))
-
(A(\widetilde M,y)-\widetilde A(\widetilde M,\widetilde{y}))
\Big)
b_\Delta'(x-\widetilde{x})
\,\dd x\,\dd \widetilde x\,\dd \tau \bigg|
\\
&=
\bigg| \int_s^t \int\!\!\!\int
\p_x
\Bigg[
\sgn(M-\widetilde M)
\Big(
(A(M,y)-\widetilde A(M,\widetilde{y}))
-
(A(\widetilde M,y)-\widetilde A(\widetilde M,\widetilde{y}))
\Big)
\Bigg]
b_\Delta(x-\widetilde x)
\,\dd x\,\dd \widetilde x\,\dd \tau \bigg| \\
& \le \int_s^t \int\!\!\!\int
[A(\cdot,y)-\widetilde A(\cdot,\widetilde{y})]_{\Lip}\,
\p_x M(x,y,\tau)\,
b_\Delta(x-\widetilde x)
\,\dd x\,\dd \widetilde x\,\dd \tau \\
& = (t-s)\,[A(\cdot,y)-\widetilde A(\cdot,\widetilde{y})]_{\Lip}.
\end{aligned}
\end{equation}
We substitute this bound into \eqref{e:commonest} and take $\Delta\to 0+$, to obtain
\begin{equation}
\label{e:Kruzkovstep1}
\begin{aligned}
& \int \|M(\cdot,y,t)-\widetilde M(\cdot, \widetilde{y},t)\|_{L^1(\mathbb{R})} \, \bpi(\dy, \dd\widetilde{y}) \\
& \quad \le \int \big( \|M(\cdot,y,s)-\widetilde M(\cdot, \widetilde{y},s)\|_{L^1(\mathbb{R})}
+ (t-s) [A(\cdot,y)-\widetilde A(\cdot,\widetilde{y})]_{\Lip} \big) \,\bpi(\dy, \dd\widetilde{y})
+ \limsup_{\Delta\to 0+} \int_s^t \mathcal{N}_\Delta(\tau) \, \dtau.
\end{aligned}
\end{equation}
It remains to estimate the nonlocal term. We turn to this presently. From now on we will suppress dependence on the time $\tau$, since the latter plays no role in the subsequent computations. We introduce the change of variables
\[
x_\pm:=\frac{x\pm \widetilde x}{2},
\qquad
\p_x=\frac12(\p_{x_+}+\p_{x_-}),
\qquad
\p_{\widetilde x}=\frac12(\p_{x_+}-\p_{x_-}),
\]
so that $\mathcal{N}_\Delta$ becomes
\begin{equation}\label{eq:BP-nonlocal-xpm}
\begin{aligned}
\mathcal{N}_\Delta & = \int\!\!\!\int\!\!\!\int
\Big[
\big(
B[\nu,M](x_+ + x_-, y)
+
B[\widetilde\nu,\widetilde M](x_+ - x_-,\widetilde{y})
\big)\p_{x_+}|M-\widetilde M|
\\
& \quad \qquad \quad +
\big(
B[\nu,M](x_+ + x_-,y)
-
B[\widetilde\nu,\widetilde M](x_+ - x_-,\widetilde{y})
\big)\p_{x_-}|M-\widetilde M|
\Big]
b_\Delta(2x_-)\,\dd x_+\,\dd x_-\,\;\bpi(\dd y,\dd \widetilde{y}).
\end{aligned}
\end{equation}
Letting $\Delta\to0+$ yields
\begin{equation}\label{eq:BP-nonlocal-diagonal-xpm}
\limsup_{\Delta\to0^+}
\mathcal{N}_\Delta \le
\frac12 ( I + II ),
\end{equation}
where
\begin{equation}\label{eq:BP-I-II-def}
\begin{aligned}
I
&:=
\int\!\!\!\int
\Big(
B[\nu,M](x,y)+B[\widetilde\nu,\widetilde M](x,\widetilde{y})
\Big)\,
\p_1|M(x,y)-\widetilde M(x,\widetilde{y})|
\,\dd x\,\bpi(\dd y,\dd \widetilde{y}),
\\
II
&:=
\int\!\!\!\int
\Big(
\p_1M(x,y)+\p_1\widetilde M(x,\widetilde{y})
\Big)
\Big|
B[\nu,M](x,y)-B[\widetilde\nu,\widetilde M](x,\widetilde{y})
\Big|
\,\dd x\,\bpi(\dd y,\dd \widetilde{y}).
\end{aligned}
\end{equation}
\medskip
We estimate $I$, starting by manipulating the part of the integrand in parentheses. We obtain
\[
\begin{aligned}
B[\nu,M](x,y)+B[\widetilde\nu,\widetilde M](x,\widetilde{y})
& =
\int\!\!\!\int
\Big[
\Phi(x-\xi,y-\eta)\,\p_1 M(\xi,\eta)
+
\Phi(x-\xi,\widetilde{y}-\widetilde{\eta})\,\p_1\widetilde M(\xi,\widetilde{\eta})
\Big]
\,\dd \xi\,\bpi(\dd \eta,\dd \widetilde{\eta})\\
& =
\frac12
\int\!\!\!\int
\Big(
\Phi(x-\xi,y-\eta)-\Phi(x-\xi,\widetilde{y}-\widetilde{\eta})
\Big)
\Big(
\p_1 M(\xi,\eta)-\p_1\widetilde M(\xi,\widetilde{\eta})
\Big)
\,\dd \xi\,\bpi(\dd \eta,\dd \widetilde{\eta})
\\
& \quad
+
\frac12
\int\!\!\!\int
\Big(
\Phi(x-\xi,y-\eta)+\Phi(x-\xi,\widetilde{y}-\widetilde{\eta})
\Big)
\Big(
\p_1M(\xi,\eta)+\p_1\widetilde M(\xi,\widetilde{\eta})
\Big)
\,\dd \xi\,\bpi(\dd \eta,\dd \widetilde{\eta}),
\end{aligned}
\]
where we have used the elementary identity $a_1a_2+b_1b_2=\frac12\big[(a_1-b_1)(a_2-b_2)+(a_1+b_1)(a_2+b_2)\big]$.
\medskip
Inserting this decomposition into the expression for $I$, we may write
\[
I = I^{a}+I^{b},
\]
where
\begin{align*}
I^a
& =
\frac12
\int\!\!\!\int\!\!\!\int\!\!\!\int \bigg[
\Big(
\Phi(x-\xi,y-\eta)-\Phi(x-\xi,\widetilde{y}-\widetilde{\eta})
\Big) \\
& \qquad \qquad \qquad \qquad \cdot
\Big(
\p_1M(\xi,\eta)-\p_1\widetilde M(\xi,\widetilde{\eta})
\Big) \p_1|M(x,y)-\widetilde M(x,\widetilde{y})| \bigg]
\,\dd x\,\dd \xi\,\bpi(\dd y,\dd \widetilde{y})
\,\bpi(\dd \eta,\dd \widetilde{\eta})
\end{align*}
and
\begin{align*}
I^{b}
& =
\frac12
\int\!\!\!\int\!\!\!
\int\!\!\!\int
\bigg[ \Big(
\Phi(x-\xi,y-\eta)+\Phi(x-\xi,\widetilde{y}-\widetilde{\eta})
\Big) \\
& \qquad \qquad \qquad \qquad \cdot
\big(
\p_1 M(\xi,\eta)+\p_1\widetilde M(\xi,\widetilde{\eta})
\big) \p_1|M(x,y)-\widetilde M(x,\widetilde{y})| \bigg] \,\dx \,\dd \xi\,\bpi(\dd \eta,\dd \widetilde{\eta})
\,\bpi(\dd y,\dd \widetilde{y}).
\end{align*}
The quantity $I^b$ will ultimately cancel with a corresponding term from $II$; on the other hand, $I^a$ admits a very natural estimate:
\[
\begin{aligned}
|I^{a}|
& \le \int\!\!\! \int \bigg[ R(T)
\|\n_2 \phi \|_{L^\infty} \big| (y - \eta) - (\widetilde{y} - \widetilde{\eta}) \big| \\
& \qquad\qquad \cdot \bigg( \int
\big(\p_1 M(\xi,\eta)+\p_1\widetilde M(\xi,\widetilde{\eta})\big) \,\dd \xi \bigg) \bigg( \int \big( \partial_1 M(x,y) + \partial_1\widetilde M(x,\widetilde{y}) \big) \,\dx \bigg) \bigg] \,\bpi(\dd \eta,\dd \widetilde{\eta})\,\bpi(\dd y,\dd \widetilde{y}) \\
& \le 8 R(T) \|\n_2 \phi\|_{L^\infty} W_1(\nu, \widetilde{\nu})
\end{aligned}
\]
We next estimate $II$. Reasoning as we did for $I$, we begin by writing
\[
\begin{aligned}
B[\nu,M](x,y)-B[\widetilde\nu,\widetilde M](x,\widetilde{y})
& =
\int\!\!\!\int
\Big[
\Phi(x-\xi,y-\eta)\,\p_1 M(\xi,\eta)
-
\Phi(x-\xi,\widetilde{y}-\widetilde{\eta})\,\p_1\widetilde M(\xi,\widetilde{\eta})
\Big]
\,\dd \xi\,\bpi(\dd \eta,\dd \widetilde{\eta})
\\
& = \frac12
\int\!\!\!\int
\big(
\Phi(x-\xi,y-\eta)-\Phi(x-\xi,\widetilde{y}-\widetilde{\eta})
\big)
\big(
\p_1 M(\xi,\eta)+\p_1\widetilde M(\xi,\widetilde{\eta})
\big)
\,\dd \xi\,\bpi(\dd \eta,\dd \widetilde{\eta})
\\
&\quad
+\frac12
\int\!\!\!\int
\big(
\Phi(x-\xi,y-\eta)+\Phi(x-\xi,\widetilde{y}-\widetilde{\eta})
\big)
\big(
\p_1M(\xi,\eta)-\p_1\widetilde M(\xi,\widetilde{\eta})
\big)
\,\dd \xi\,\bpi(\dd \eta,\dd \widetilde{\eta}),
\end{aligned}
\]
We may thus write
\[
II = II^{a}+II^{b},
\]
where $|II^a|$ obeys exactly the same estimate that $|I^a|$ does, and $II^b \le -I^b$. To establish this second fact, we note that since $\Phi$ is nondecreasing in its first argument, we have
\[
\int
\Phi(x-\xi,\cdot)
\big(
\p_1 M(\xi,\eta)-\p_1 \widetilde M(\xi,\widetilde{\eta})
\big)
\,\dd \xi
\le
\int
\Phi(x-\xi,\cdot)
\p_1\big|M(\xi,\eta)-\widetilde M(\xi,\widetilde{\eta})\big|
\,\dd \xi.
\]
Using this estimate in the second line below, we see that
\begin{align*}
II^{b} & =
\frac12
\int\!\!\!\int\!\!\!\int\!\!\!\int \bigg[
\big(
\Phi(x-\xi,y-\eta)+\Phi(x-\xi,\widetilde{y}-\widetilde{\eta})
\big)\\
& \qquad \qquad \qquad \qquad \cdot \big(
\p_1M(x,y)+\p_1\widetilde M(x,\widetilde{y})
\big)
\p_1\big(M(\xi,\eta)-\widetilde M(\xi,\widetilde{\eta})\big) \bigg]
\,\dd x\,\dd \xi\,\bpi(\dd \eta,\dd \widetilde{\eta})\,\bpi(\dd y,\dd \widetilde{y}) \\
& \le
\frac12
\int\!\!\!\int\!\!\!\int\!\!\!\int \bigg[
\big(
\Phi(x-\xi,y-\eta)+\Phi(x-\xi,\widetilde{y}-\widetilde{\eta})
\big)\\
& \qquad \qquad \qquad \qquad \cdot \big(
\p_1M(x,y)+\p_1\widetilde M(x,\widetilde{y})
\big)
\p_1\big|M(\xi,\eta)-\widetilde M(\xi,\widetilde{\eta})\big| \bigg]
\,\dd x\,\dd \xi\,\bpi(\dd \eta,\dd \widetilde{\eta})\,\bpi(\dd y,\dd \widetilde{y}).
\end{align*}
After interchanging $(x,y,\widetilde{y})\leftrightarrow(\xi,\eta,\widetilde{\eta})$, and recalling that $\Phi$ is odd in its first argument and even in its second, we see that the quantity in the last line above is exactly $-I^b$, justifying our claim.
Putting together all the estimates above, we see that
\[
\limsup_{\Delta\to0^+}
\mathcal{N}_\Delta \le
\frac12 ( |I^a| + |II^a| + I^b + II^b )
\le
8 R(T)\,\|\n_2\phi\|_{L^\infty}\,W_1(\nu,\widetilde\nu).
\]
Together with \eqref{e:Kruzkovstep1}, this estimate completes the proof of \eqref{eq:stab-pi}.
\medskip
\subsection{Second stability estimate}
\label{ss:stab2}
As we prove our second stability estimate, we will perform our analysis with fixed $\Delta>0$ rather than taking $\Delta\to 0+$. We will ultimately be estimating the same quantity as before, namely $\cD(t)$ as defined in \eqref{eq:BP-E}. However, we will first work with the related quantity $\mathcal{D}_\Delta(t)$, which is defined in \eqref{eq:BP-E-EDelta} and appears on the left side of \eqref{e:commonest}.
The discrepancy between $\mathcal{D}(t)$ and $\mathcal{D}_\Delta(t)$ is of order~$\Delta$ (see Lemma \ref{l:E-EDelta} below), so we may use our estimates on $\mathcal{D}_\Delta$ to write down an inequality for $\mathcal{D}(t)$ in terms of quantities involving $\Delta$ and $\frac{1}{\Delta}$. We will ultimately make a choice of $\Delta$ to balance these two contributions and obtain \eqref{eq:main-unidir-optimal}; however, we will write our argument with arbitrary $\Delta>0$ in order to keep its role transparent. The main idea of this strategy is inspired by work of Bouchut and Perthame \cite{BouchutPerthame1998}.
We collect a few basic estimates in the following Lemma before turning to the main thrust of our proof. (These can also essentially be found in \cite{BouchutPerthame1998}.)
\begin{LEMMA}
\label{l:E-EDelta}
The following estimates hold:
\begin{equation}\label{eq:BP-E-EDelta-bound}
|\mathcal{D}(\tau)-\mathcal{D}_\Delta(\tau)| \le \Delta
;
\end{equation}
\begin{equation}\label{eq:BP-bDelta-prime-bound}
\Bigg|
\int\!\!\!\int
|M(x,y,\tau)-\widetilde M(\widetilde x,\widetilde{y},\tau)|\,b_{\Delta}'(x-\widetilde x)\,\dx\,\dd \widetilde x
\Bigg|
\le C.
\end{equation}
\end{LEMMA}
\begin{proof}
We use repeatedly the fact that if $F:\R\to \R$ is the cumulative distribution function of a probability measure, then
\begin{equation}
\label{e:Mdhest}
\int_\R \big( F(x+h) - F(x)\big)\, \dx \le |h|.
\end{equation}
To prove \eqref{eq:BP-E-EDelta-bound}, we estimate as follows:
\begin{align*}
|\mathcal{D}(\tau)-\mathcal{D}_\Delta(\tau)|
& = \bigg| \int\!\!\!\int\!\!\!\int \big( |M(x,y,\tau) - \widetilde{M}(x,y,\tau)| - |M(x,y,\tau) - \widetilde{M}(\widetilde{x},\widetilde{y},\tau)|\big) b_\Delta(x-\widetilde{x}) \,\dx \, \dd \widetilde{x} \,\bpi(\dy, \dd\widetilde{y}) \bigg| \\
& \le
\int\!\!\!\int\!\!\!\int
|\widetilde M(x,y,\tau)-\widetilde M(\widetilde x,\widetilde{y},\tau)|
\,b_\Delta(x-\widetilde x)\,\dx\,\dd \widetilde x\;\bpi(\dd y,\dd \widetilde{y}) \\
& = \int \!\!\!\int \bigg( \int \big| \widetilde{M}(\widetilde{x}+h,y,\tau) - \widetilde{M}(\widetilde{x},\widetilde{y},\tau)\big| \,\dd \widetilde{x} \bigg) b_\Delta(h)\,\dh \,\bpi(\dy, \dd\widetilde{y}) \\
& \le
\int_\RR |h|\,b_\Delta(h)\,\dd h \le \Delta.
\end{align*}
We turn to the proof of \eqref{eq:BP-bDelta-prime-bound}. We use the fact that $\int b_\Delta'(x-\widetilde{x})\,\dx = 0$ in order to pass to the second line below.
\begin{align*}
& \Bigg|
\int\!\!\!\int
|M(x,y,\tau)-\widetilde M(\widetilde x,\widetilde{y},\tau)|\,b_{\Delta}'(x-\widetilde x)\,\dx\,\dd \widetilde x
\Bigg|\\
& \qquad = \Bigg|
\int\!\!\!\int
\big[ |M(x,y,\tau)-\widetilde M(\widetilde x,\widetilde{y},\tau)|- |M(\widetilde x,y,\tau)-\widetilde M(\widetilde x,\widetilde{y},\tau)|\big] \,b_{\Delta}'(x-\widetilde x)\,\dx\,\dd \widetilde x
\Bigg|\\
& \qquad \le
\int\!\!\!\int
|M(x,y,\tau)-M(\widetilde x,y,\tau)|\,|b'_\Delta(x-\widetilde x)|
\,\dx\,\dd \widetilde x \\
& \qquad \le \int |h| |b_\Delta'(h)| \,\dh = C.
\end{align*}
\end{proof}
We now begin the proof of \eqref{eq:main-unidir-optimal} in earnest, starting with the following simple estimate of the flux term:
\begin{equation}
\label{e:FDeltaest}
\begin{split}
|\cF_\Delta(\tau)|
& \le
\bigg( \int 2 \|A(\cdot, y) - \widetilde{A}(\cdot, \widetilde{y})\|_{L^\infty} \bpi(\dy, \dd\widetilde{y}) \bigg) \bigg( \int_{-R(T)-\Delta}^{R(T)+\Delta} \int |b_\Delta'(x-\widetilde{x})| \, \dx \, \dd \widetilde{x} \bigg) \le C\bigg( \frac{ R(T)}{\Delta} + 1\bigg) \cA_\infty.
\end{split}
\end{equation}
Almost all of the remainder of our proof is dedicated to estimating the nonlocal term $\cN_\Delta$. Integrating by parts in \eqref{e:NL}, we may write
\begin{equation}\label{eq:BP-NDelta-split}
\mathcal N_{\Delta}
=
\mathcal N_{\Delta}^{(a)}+\mathcal N_{\Delta}^{(b)}
+ N_{\Delta}^{(c)},
\end{equation}
where (noting that $\partial_x B[\nu,M] = \phi*(\nu \partial_1 M)$)
\begin{align*}
\mathcal N_{\Delta}^{(a)}
& =
-\int\!\!\!\int\!\!\!\int
\Big(
(\phi*(\nu\,\p_1M))(x,y)
+
(\phi*(\widetilde\nu\,\p_1\widetilde M))(\widetilde x,\widetilde{y})
\Big)
|M(x,y)-\widetilde M(\widetilde x,\widetilde{y})|\,b_{\Delta}(x-\widetilde{x})
\,\dx\,\dd \widetilde x\;\bpi(\dd y,\dd \widetilde{y}),
\\
\mathcal N_{\Delta}^{(b)}
& =-\int\!\!\!\int\!\!\!\int
\big(
B[\nu,M](x,y)-B[\nu,M](\widetilde x,y)
\big)
|M(x,y)-\widetilde M(\widetilde x,\widetilde{y})|\,b'_{\Delta}(x-\widetilde{x})
\,\dx\,\dd \widetilde x\;\bpi(\dd y,\dd \widetilde{y}), \\
\mathcal N_{\Delta}^{(c)}
& =-\int\!\!\!\int
\big(
B[\nu,M](\widetilde{x},y)-B[\widetilde\nu,\widetilde M](\widetilde x,\widetilde{y})
\big) \bigg( \int
|M(x,y)-\widetilde M(\widetilde x,\widetilde{y})| \,b'_{\Delta}(x-\widetilde{x})
\,\dx\bigg) \,\dd \widetilde x\;\bpi(\dd y,\dd \widetilde{y}).
\end{align*}
The first of these can be estimated easily:
\begin{equation}\label{eq:BP-NDelta-a-est}
|\mathcal N_{\Delta}^{(a)}(\tau)|
\le
2\|\phi\|_{L^\infty}\,\mathcal{D}_{\Delta}(\tau).
\end{equation}
The second requires a bit of manipulation but is still straightforward:
\begin{align*}
|\cN_\Delta^{(b)}(\tau)| & \le \int\!\!\!\int \big( \|\phi\|_{L^\infty} |h|\big) \bigg( \int \big[ |M(x+h,y,\tau) - M(x,y,\tau)| + |M(x,y,\tau) - \widetilde{M}(x,\widetilde{y},\tau)|\, \dx \bigg) |b_\Delta'(h)| \, \dh \, \bpi(\dy, \dd\widetilde{y}) \\
& \le \|\phi\|_{L^\infty}\int\!\!\!\int \bigg( |h| + \|M(\cdot, y,\tau) - \widetilde{M}(\cdot, \widetilde{y},\tau)\|_{L^1} \bigg) |h b_\Delta'(h)| \,\dh \,\bpi(\dy, \dd\widetilde{y}) \\
& \le C \|\phi\|_{L^\infty} \big( \Delta + \mathcal{D}(\tau) \big)
\end{align*}
To estimate $\cN_\Delta^{(c)}$,we note first of all that
\begin{align*}
\big|B[\nu, M](\widetilde{x},y,\tau) - B[\widetilde{\nu}, \widetilde{M}](\widetilde{x},\widetilde{y},\tau) \big|
& \le \bigg| \int\!\!\!\int \big[ \Phi(\widetilde{x}-\xi, y-\eta) \partial_1 M(\xi, \eta,\tau) - \Phi(\widetilde{x}-\xi, \widetilde{y}-\widetilde{\eta}) \partial_1 \widetilde{M}(\xi, \widetilde{\eta},\tau)\big] \,\dd \xi\,\bpi(\dd \eta, \dd \widetilde{\eta})\bigg| \\
& \le 2 R(T) \|\n_2 \phi\|_{L^\infty} \big( |y-\widetilde{y}| + W_1(\nu, \widetilde{\nu}) \big)
+ \|\phi\|_{L^\infty} \mathcal{D}(\tau).
\end{align*}
We also note that the integral over $x$ in $\cN_\Delta^{(c)}$ can be treated using the same trick as in the proof of \eqref{e:Mdhest}:
\begin{align*}
\bigg| \int
|M(x,y)-\widetilde M(\widetilde x,\widetilde{y})| \,b'_{\Delta}(x-\widetilde{x})
\,\dx\bigg|
& = \bigg| \int
\big[ |M(x,y)-\widetilde M(\widetilde x,\widetilde{y})| - |M(\widetilde{x},y) - \widetilde{M}(\widetilde{x},\widetilde{y})|\big] \,b'_{\Delta}(x-\widetilde{x})
\,\dx\bigg| \\
& \le \int |M(x,y) - M(\widetilde{x},y)||b_\Delta'(x-\widetilde{x})| \,\dx.
\end{align*}
The integral over $\widetilde{x}$ of this last quantity is uniformly bounded. Therefore, we may write
\begin{align*}
|\cN_\Delta^{(c)}(\tau)|
& \le C R(T) \|\n_2 \phi\|_{L^\infty} W_1(\nu, \widetilde{\nu}) + C \|\phi\|_{L^\infty} \mathcal{D}(\tau).
\end{align*}
Combining the estimates for $\cN_\Delta^{(a)}$, $\cN_\Delta^{(b)}$, and $\cN_\Delta^{(c)}$, and using \eqref{eq:BP-E-EDelta-bound}, we get
\begin{equation}\label{eq:BP-NDelta-final-est}
|\mathcal N_{\Delta}(\tau)|
\le
C\|\phi\|_{L^\infty}\big(\mathcal{D}_\Delta(\tau) + \Delta \big)
+ C R(T)\|\n_2\phi\|_{L^\infty} W_1(\nu,\widetilde\nu).
\end{equation}
We are now in a position to finish the argument. Substituting the estimates \eqref{e:FDeltaest} and \eqref{eq:BP-NDelta-final-est} into \eqref{e:commonest} yields
\begin{equation}\label{eq:BP-EDelta-integral}
\begin{aligned}
\mathcal{D}_{\Delta}(t)
&\le
\mathcal{D}_{\Delta}(s)
+
C (t-s) \bigg[
\bigg( \!\frac{R(T)}{\Delta}\!+ 1\!\bigg) \cA_\infty
+ \|\phi\|_{L^\infty} \Delta + R(T) \|\n_2 \phi\|_{L^\infty} W_1(\nu, \widetilde{\nu}) \bigg]
\!+ C\|\phi\|_{L^\infty}\! \int_s^t \!\mathcal{D}_{\Delta}(\tau) \,\dtau,
\end{aligned}
\end{equation}
Applying Gr\"onwall's inequality gives
\[
\begin{aligned}
\mathcal{D}_{\Delta}(t)
&\le
\mathcal{D}_{\Delta}(s)e^{C\|\phi\|_{L^\infty}(t-s)}
+
\bigg[ \bigg( \frac{R(T)}{\Delta} +1 \bigg) \cA_\infty
+\|\phi\|_{L^\infty}\,\Delta
+R(T)\|\n_2\phi\|_{L^\infty}\,W_1(\nu,\widetilde\nu)
\bigg]
\frac{e^{C\|\phi\|_{L^\infty}(t-s)}-1}{\|\phi\|_{L^\infty}}\\
& \le e^{C\|\phi\|_{L^\infty}(t-s)}\bigg( \mathcal{D}_{\Delta}(s)
+
C(t-s)\bigg[ \bigg( \frac{R(T)}{\Delta} +1 \bigg) \cA_\infty
+R(T)\|\n_2\phi\|_{L^\infty}\,W_1(\nu,\widetilde\nu)
\bigg]\bigg) + ( e^{C\|\phi\|_{L^\infty}(t-s)}-1)\Delta
\end{aligned}
\]
Using \eqref{eq:BP-E-EDelta-bound}, we obtain
\begin{equation}\label{eq:BP-E-before-opt}
\begin{aligned}
\mathcal{D}(t)
&\le e^{C\|\phi\|_{L^\infty}(t-s)} \bigg(
\mathcal{D}(s)
+ 2 \Delta
+ C(t-s) \bigg[ \bigg( \frac{R(T)}{\Delta} + 1\bigg) \,\mathcal A_\infty + R(T)\|\n_2\phi\|_{L^\infty} W_1(\nu,\widetilde\nu) \big)\bigg] \bigg).
\end{aligned}
\end{equation}
We obtain \eqref{eq:main-unidir-optimal} after choosing $\Delta = \sqrt{R(T)\cA_\infty (t-s)}$ and adjusting the constant $C$ appropriately.
\section{Existence of entropy solutions}
\label{s:Existence}
We now turn to the matter of existence. Our main result toward this end can be stated quite simply:
\begin{THEOREM}
\label{t:existence}
Assume $\phi$ satisfies \eqref{e:phieven} and \eqref{e:phiLip2}. Let $(M^0, A, \nu)$ be an admissible triple and fix $T>0$. There exists a unique entropy solution $M\in C([0,T); L^1_{\loc}(\mathbb{R}\times \mathbb{R}^{d-1}, \dx\otimes \nu))$ of \eqref{e:ScalarBalance} on $[0,T)$ associated to $(M^0, A, \nu)$.
\end{THEOREM}
The rest of this section is dedicated to the proof of Theorem \ref{t:existence}. Our analysis, however, gives more than just existence---it furnishes an approximation scheme. We construct our entropy solution $M$ as the limit (in a sense to be described below) of entropy solutions $M^k$ that are discrete in $y$, in the sense that $M^k$ is associated to an admissible triple $(M^{0,k}, A^k, \nu_k)$ such that $\nu_k$ is atomic. These solutions $M^k$ are constructed in turn as limits of entropy solutions $M^k_N$ which are discrete in both $x$ and~$y$, and which are completely determined by the evolution of what we refer to as the \textit{sticky particle Cucker--Smale} dynamics (SPCS). The SPCS dynamics require agents to evolve according to the Cucker--Smale ODEs on any time interval during which no collisions occur. Collisions are required to be completely inelastic: the agents involved stick together for all later times, with a post-collisional velocity determined by conservation of momentum.
In Section \ref{ss:Existdisc} below, we construct the solutions $M_N^k$ from SPCS dynamics and obtain $M^k$ as a limit as $N\to \infty$. (We drop the superscript $k$ in this section to avoid a notational overload.) We do not need the full strength of the assumption on $\phi$ for our analysis of the discrete-in-$y$ case to go through; only \eqref{e:phiL1locx} is needed at this point. Then, in Section \ref{ss:Existgen}, we take the limit as $k\to \infty$ to obtain~$M$. Theorem~\ref{thm:existence-main} and the discussion preceding it provide a more precise description of the approximation scheme used to construct the solutions $M^k$, as well as the sense in which they approximate $M$. We detail rates of convergence in Section \ref{ss:rate} under additional assumptions on $M^0$ and $A$; see also Sections \ref{ss:approximation} and \ref{ss:rate2} for details at the level of the Euler Alignment system.
In summary, our solutions are constructed with two levels of discretization: one in the directions transverse to the flow, to reduce the dynamics to finitely many slices, and one in the direction of the flow. The `finite-slice' analysis follows \cite{LeslieTan2023} rather closely (though we provide enough details to make the current work self-contained); the discretization transverse to the flow is new and is our main focus for this section. With a bit of care, we can approximate our general solutions directly by SPCS dynamics by taking both the number of slices and the number of agents per slice to infinity simultaneously; see sections \ref{ss:rate}, \ref{ss:approximation}, and \ref{ss:rate2} below.
\subsection{Entropy solutions associated to atomic $\nu$}
\label{ss:Existdisc}
We now construct the unique entropy solution associated to any admissible triple $(M^0, A, \nu)$ such that $\nu$ is atomic. Our analysis closely follows that of \cite{LeslieTan2023} due to the second author and Tan; the approach of the latter was in turn inspired by work of Brenier and Grenier \cite{brenier1998sticky} on the pressureless Euler equations.
\subsubsection{Entropy conditions for piecewise constant solutions}
We begin by recasting the Kruzkov entropy condition \eqref{e:entropydistributional} in a form that will be easier to verify for the piecewise constant solutions we generate using SPCS dynamics. Equation \eqref{e:RH} and inequality \eqref{e:Oleinik} below together constitute the simplified criteria and will be referred to as `Rankine--Hugoniot' and `Oleinik' conditions, respectively, on our piecewise constant solutions.
Suppose the measure $\nu$ is atomic; that is, assume it takes the form $\nu = \sum_{\ell=1}^L \mu_\ell \delta(\cdot - y_\ell)$. Suppose additionally that for each $\ell\in \{1, \ldots, L\}$, the function $M(x,y_\ell,t)$ (assumed to be nondecreasing in $x$, as usual) is piecewise constant, with discontinuity curves that are piecewise $C^1$. Such a function must take the form
\begin{equation}
M(x,y_\ell, t) = \sum_{i=1}^{I_\ell} m_{i,\ell} H(x - x_{i,\ell}(t)),
\end{equation}
where $H:\mathbb{R}\to \mathbb{R}$ is the Heaviside function ($H(x) = 1_{[0,\infty)}(x)$) and each $x_{i,\ell}:[0,\infty)\to \mathbb{R}$ is a continuous, piecewise $C^1$ function of $t$.
Write $\eta_\alpha(m) = |m-\alpha|$ and $q_\alpha(m, y_\ell) = \sgn(m-\alpha)(A(m,y_\ell) - A(\alpha, y_\ell))$. One can show that the Kruzkov entropy condition \eqref{e:entropydistributional} reduces to the requirement that for each $\ell\in \{1, \ldots, L\}$ and each $i\in \{1, \ldots, I_\ell\}$, we have
\begin{equation}
\label{e:entropypiecewise}
\big( \dot{x}_{i,\ell}(t) + B[\nu,M](x_{i,\ell}(t), y_\ell, t) \big) [[\eta_\alpha(M(\cdot, y_\ell, t))]] \ge [[q_\alpha(M(\cdot, y_\ell, t), y_\ell)]],
\end{equation}
for any time $t$ at which $x_{i,\ell}$ is differentiable, and any $\alpha\in \mathbb{R}$. Here $[[f]]$ denotes the jump in $f$ across $x_{i,\ell}$, i.e., $[[f]] = f(x_{i,\ell}(t)+) - f(x_{i,\ell}(t)-)$.
Denote $m_\pm = M(x_{i,\ell}(t)\pm, y_\ell, t)$. If we take $\alpha \le m_-$ then \eqref{e:entropypiecewise} takes the form
\begin{equation}
\label{e:entropypiecewise1}
\big( \dot{x}_{i,\ell}(t) + B[\nu,M](x_{i,\ell}(t), y_\ell, t) \big) [[M(\cdot, y_\ell, t))]] \ge [[A(M(\cdot, y_\ell, t), y_\ell)]],
\end{equation}
whereas taking $\alpha\ge m_+$ gives the opposite inequality. Combining both cases gives us
\begin{equation}
\label{e:RH}
\dot{x}_{i,\ell}(t) + B[\nu,M](x_{i,\ell}(t), y_\ell, t) = \frac{[[A(M(\cdot, y_\ell, t), y_\ell)]]}{[[M(\cdot, y_\ell, t)]]}.
\end{equation}
The final case, namely $m_-<\alpha<m_+$, yields \begin{equation}
\label{e:entropypiecewise2}
\big( \dot{x}_{i,\ell}(t) + B[\nu,M](x_{i,\ell}(t), y_\ell, t) \big) \big( [[M(\cdot, y_\ell, t))]] - 2(\alpha - m_-)\big) \ge [[A(M(\cdot, y_\ell, t), y_\ell)]] - 2(A(\alpha, y_\ell) - A(m_-, y_\ell)).
\end{equation}
Substituting \eqref{e:RH} into \eqref{e:entropypiecewise2} and rearranging gives
\begin{equation}
\label{e:Oleinik}
\frac{[[A(M(\cdot, y_\ell, t), y_\ell)]]}{[[M(\cdot, y_\ell, t)]]} \le \frac{A(\alpha, y_\ell) - A(m_-, y_\ell)}{\alpha - m_-},
\qquad m_-<\alpha<m_+.
\end{equation}
We will refer to \eqref{e:RH} as the `Rankine--Hugoniot' condition and \eqref{e:Oleinik} as the `Oleinik entropy condition' for these kinds of solutions. Together, \eqref{e:RH} and \eqref{e:Oleinik} are equivalent to \eqref{e:entropydistributional} for piecewise constant solutions of \eqref{e:ScalarBalance}.
The piecewise constant solutions that we have in mind are the ones where the trajectories $x_{i,\ell}$ satisfy the \textit{sticky particle Cucker--Smale dynamics}, which we introduce presently (in the unidirectional case only).
\subsubsection{Sticky particle Cucker--Smale (SPCS) dynamics}
In this section, we work with a collection of agents with positions $(x_{i,\ell}(t), y_\ell)\in \mathbb{R}\times \mathbb{R}^{d-1}$, horizontal velocities $\dot{x}_{i,\ell}(t) = v_{i,\ell}(t)\in \mathbb{R}$, and masses $\mu_\ell m_{i,\ell}$, where $\ell\in \{1, \ldots, L\}$ indexes the `horizontal slice' on which each agent lives and $i\in \{1, \ldots, I_\ell\}$ indexes its position relative to other agents on the same slice. We assume that
\[
x_{1,\ell}(t)\le x_{2,\ell}(t) \le \cdots \le x_{I_\ell, \ell}(t),
\qquad t\ge 0,\, \ell\in \{1, \ldots, L\}.
\]
We also make the normalizing assumption that $\sum_{\ell=1}^L \mu_\ell = 1$, and $\sum_{i=1}^{I_\ell} m_{i,\ell} = 1$ for each $\ell$. At the level of the scalar balance law, these choices correspond to working with the transverse measure $\nu = \sum_{\ell=1}^L \mu_\ell \delta(\cdot - y_\ell)$ and piecewise constant function $M(x,y_\ell, t) = \sum_{i=1}^{I_\ell} m_{i,\ell} H(x - x_{i,\ell}(t))$ in the previous subsection. Note also that, under this notation, we have
\begin{equation}
\label{e:BnuMdiscrete}
B[\nu,M](x_{i,\ell}(t), y_\ell, t) = \sum_{p=1}^L \sum_{j=1}^{I_p} \mu_p m_{j,p} \Phi(x_{i,\ell}(t) - x_{j,p}(t), y_\ell - y_p).
\end{equation}
We also make the convention that velocities are right-continuous:
\[
v_{i,\ell}(t+) = v_{i,\ell}(t),
\qquad \ell\in \{1, \ldots, L\}, \, i\in \{1, \ldots, I_\ell\}.
\]
For each $\ell\in \{1, \ldots, L\}$ and each $i\in \{1, \ldots, I_\ell\}$, we denote
\[
\mathcal{J}_{i,\ell}(t) = \{j: x_{j,\ell}(t) = x_{i,\ell}(t)\},
\qquad
i^*(t,\ell) = \max J_{i,\ell}(t),
\qquad
i_*(t,\ell) = \min J_{i,\ell}(t),
\]
and we say that time $t$ is a \textit{collision time} if the cardinality of some $\cJ_{i,\ell}$ increases at time $t$. Otherwise it is a \textit{noncollisional time}. A time interval is \textit{noncollisional} if it contains no collision times.
Finally, when we specify initial conditions $(x^0_{i,\ell},v^0_{i,\ell})$, we set $x_{i,\ell}(0) = x^0_{i,\ell}$ and $v_{i,\ell}(0-) = v_{i,\ell}^0$. The quantity $v_{i,\ell}(0)$ will be determined from the rules below.
With the above notation and conventions, we are ready to specify the characteristics of the sticky particle Cucker--Smale dynamics, which are determined from the following three rules.
\begin{enumerate}[label = (\alph*)]
\item (Stickiness of collisions) For any time $t\ge 0$ (but in particular at collision times) we have
\begin{equation}
\cJ_{i,\ell}(s)\subseteq \cJ_{i,\ell}(t),
\qquad \text{ whenever } 0\le s \le t.
\end{equation}
\item (Conservation of momentum across collisions) We have
\begin{equation}
\label{e:collisioncons}
\sum_{j\in \cJ_{i,\ell}(t)} m_{j,\ell} v_{j,\ell}(t) = \sum_{j\in \cJ_{i,\ell}(t)} m_{j,\ell} v_{j,\ell}(t-),
\qquad t\ge 0.
\end{equation}
\item (Cucker--Smale dynamics at noncollisional times) On any noncollisional time interval, $(x_{i,\ell},v_{i,\ell})$ evolve according to the Cucker--Smale ODEs:
\begin{equation}
\label{e:CSuni}
\left\{
\begin{split}
\dot{x}_{i,\ell} & = v_{i,\ell} \\
\dot{v}_{i,\ell} & = \sum_{p = 1}^L \sum_{j=1}^{I_p} \mu_p m_{j,p} \phi(x_{i,\ell} - x_{j,p}, y_\ell - y_p)(v_{j,p} - v_{i,\ell})
\end{split}
\right.
\end{equation}
\end{enumerate}
Note that rule (a) guarantees that only finitely many collisions may occur, so the dynamics follow \eqref{e:CSuni} except for at these finitely many collision times. We will also use the following relationship, which results from combining (a) and (b).
\begin{equation}
\label{e:collisioncons2}
v_{i,\ell}(t) = \frac{\sum_{j\in \cJ_{i,\ell}(t)} m_{j,\ell} v_{j,\ell}(t)}{\sum_{j\in \cJ_{i,\ell}(t)} m_{j,\ell}} = \frac{\sum_{j\in \cJ_{i,\ell}(t)} m_{j,\ell} v_{j,\ell}(t-)}{\sum_{j\in \cJ_{i,\ell}(t)} m_{j,\ell}},
\qquad t\ge 0.
\end{equation}
\subsubsection{Basic properties of SPCS}
Let us briefly review a few properties of the SPCS dynamics. We state these facts without proof; the details are all analogous to the arguments in \cite{LeslieTan2023}.
\begin{PROP}[Maximum Principle]
With the notation above, if the unidirectional sticky particle Cucker--Smale system is equipped with the masses $\mu_\ell m_{i,\ell}$ and initial conditions $(x_{i,\ell}^0,y_\ell, v_{i,\ell}^0)$, $1\le \ell \le L$, $1\le i \le I_\ell$, then
\begin{equation}
\min_{j,p} v_{j,p}(s) \le v_{i,\ell}(t)\le \max_{j,p} v_{j,p}(s),
\qquad 0\le s \le t,
\end{equation}
and
\begin{equation}
|x_{i,\ell}(t)| \le \max_{j,p} |x_{j,p}^0| + t \max_{j', p'} |v_{j', p'}^0|,
\qquad t\ge 0.
\end{equation}
\end{PROP}
The remaining properties concern the following quantity.
\begin{equation}
\psi_{i,\ell}(t) = v_{i,\ell}(t) + \sum_{p=1}^L \sum_{j=1}^{I_p} \mu_p m_{j,p} \Phi(x_{i,\ell}(t) - x_{j,p}(t), y_\ell - y_p),
\qquad 1\le i\le I_\ell, \, 1\le \ell \le L, \, t\ge 0.
\end{equation}
We also define
\[
\psi_{i,\ell}^0 = v_{i,\ell}^0 + \sum_{p=1}^L \sum_{j=1}^{I_p} \mu_p m_{j,p} \Phi(x_{i,\ell}^0 - x_{j,p}^0, y_\ell - y_p).
\]
\begin{PROP}[Properties of $\psi$] The quantities $\psi_{i,\ell}$ have the following conservation property:
\begin{equation}
\label{e:collisionconspsi}
\psi_{i,\ell}(t) = \frac{\sum_{j\in \cJ_{i,\ell}(t)} m_{j,\ell} \psi_{j,\ell}(t)}{\sum_{j\in \cJ_{i,\ell}(t)} m_{j,\ell}} = \frac{\sum_{j\in \cJ_{i,\ell}(t)} m_{j,\ell} \psi_{j,\ell}(s)}{\sum_{j\in \cJ_{i,\ell}(t)} m_{j,\ell}},
\qquad 0\le s\le t.
\end{equation}
Furthermore, we have the following ordering property:
\begin{equation}
\psi_{i_*(s,\ell),\ell}(s-)\ge \psi_{i_*(s,\ell)+1,\ell}(s-) \ge \cdots \ge \psi_{i^*(s,\ell),\ell}(s-),
\qquad s\ge 0.
\end{equation}
Consequently, for $k\in \cJ_{i,\ell}(t)$, the functions
\[
s\mapsto \frac{\sum_{j=i_*(t,\ell)}^k m_{j,\ell} \psi_{j,\ell}(s)}{\sum_{j=i_*(t,\ell)}^k m_{j,\ell}},
\qquad
s\mapsto \frac{\sum_{j=k}^{i^*(t,\ell)} m_{j,\ell} \psi_{j,\ell}(s)}{\sum_{j=k}^{i_*(t,\ell)} m_{j,\ell}},
\]
are nonincreasing and nondecreasing, respectively, on the interval $s\in [0,t]$. Therefore, we have
\begin{equation}
\label{e:psibary}
\frac{\sum_{j=i_*(t,\ell)}^k m_{j,\ell} \psi_{j,\ell}(s)}{\sum_{j=i_*(t,\ell)}^k m_{j,\ell}}
\ge \psi_{i,\ell}(t)
\ge \frac{\sum_{j=k}^{i^*(t,\ell)} m_{j,\ell} \psi_{j,\ell}(s)}{\sum_{j=k}^{i_*(t,\ell)} m_{j,\ell}},
\qquad k\in \cJ_{i,\ell}(t),\; 0\le s \le t.
\end{equation}
\end{PROP}
\subsubsection{Entropy solutions generated by SPCS dynamics}
Equipped with the tools developed above, we can now complete our verification that unidirectional SPCS dynamics generate entropy solutions of \eqref{e:ScalarBalance} and make the relationship completely precise.
\begin{THEOREM}
\label{t:SPCSentropy}
Assume $\phi$ satisfies \eqref{e:phieven} and \eqref{e:phiL1locx}. Fix a collection of initial positions $(x^0_{i,\ell}, y_\ell)\in \mathbb{R}\times \mathbb{R}^{d-1}$, initial horizontal velocities $v^0_{i,\ell}$, and masses $\mu_\ell m_{i,\ell}$, with $\sum_{\ell = 1}^L \mu_\ell = 1$, $\sum_{i=1}^{I_\ell} m_{i,\ell} = 1$, $i\in \{1, \ldots, I_\ell\}$, $\ell\in \{1, \ldots, L\}$. Assume $(x_{i,\ell}(t), v_{i,\ell}(t))$ satisfy the sticky particle Cucker--Smale dynamics associated to this initial data. Define $\psi_{i,\ell}(t)$ and $\psi_{i,\ell}^0$ as in the previous subsection and define $\theta_{i,\ell} = \sum_{j=1}^i m_{j,\ell}$, $\theta_{0,\ell} = 0$. Then
\begin{equation}
\label{e:MSPCS}
M(x,y_\ell,t) = \sum_{i=1}^{I_\ell} m_{i,\ell} H(x - x_{i,\ell}(t))
\end{equation}
defines the unique entropy solution of \eqref{e:ScalarBalance} associated to the admissible triple $(M^0, A, \nu)$, where
\begin{equation}
\label{e:nuSPCS}
\nu = \sum_{\ell=1}^L \mu_\ell \delta(\cdot - y_\ell),
\end{equation}
\begin{equation}
\label{e:M0SPCS}
M^0(x,y_\ell) = \sum_{i=1}^{I_\ell} m_{i,\ell} H(x - x^0_{i,\ell}),
\end{equation}
and $A(\cdot, y_\ell)$ is the piecewise linear function determined by
\begin{equation}
\label{e:ASPCS}
A(\theta_{i,\ell},y_\ell) = \sum_{j=1}^i m_{j,\ell} \psi^0_{j,\ell},
\qquad \qquad
\partial_m A(\theta, y_\ell) = \psi^0_{i,\ell} \quad \text{ for } \quad \theta\in (\theta_{i-1, \ell}, \theta_{i,\ell}).
\end{equation}
\end{THEOREM}
\begin{proof}
It suffices to check \eqref{e:RH} and \eqref{e:Oleinik}, with
\[
m_- = M(x_{i,\ell}(t)-, y_\ell,t) = \theta_{i_*(t,\ell)-1,\ell},
\qquad
m_+ = M(x_{i,\ell}(t)+, y_\ell,t) = \theta_{i^*(t,\ell),\ell}
\]
For the first verification, we note that
\begin{align*}
[[A(M(\cdot, y_\ell, t), y_\ell)]] & = A(\theta_{i^*(t,\ell),\ell}, y_\ell) - A(\theta_{i_*(t,\ell)-1,\ell}, y_\ell)
= \sum_{j\in \cJ_{i,\ell}(t)} m_{j,\ell} \psi^0_{j,\ell} \\
& = \bigg( \sum_{j\in \cJ_{i,\ell}(t)} m_{j,\ell}\bigg) \psi_{i,\ell}(t) \\
& = [[M(\cdot, y_\ell, t)]]\cdot \big( \dot{x}_{i,\ell}(t) + B[\nu,M](x_{i,\ell}(t), y_\ell, t) \big).
\end{align*}
In passing to the second line, we have used \eqref{e:collisionconspsi}; in passing to the third, we used \eqref{e:BnuMdiscrete}. To check the Oleinik condition, we argue similarly, using \eqref{e:psibary}.
Note that piecewise linearity of $A(\cdot, y_\ell)$ means that we only need to check \eqref{e:Oleinik} at the breakpoints $\theta_{k,\ell}$, with $k\in \cJ_{i,\ell}(t)\backslash \{i_*(t,\ell), i^*(t,\ell)\}$. To this end, we have
\[
\frac{A(\theta_{k,\ell}, y_\ell)- A(\theta_{i_*(t,\ell)-1,\ell}, y_\ell)}{\theta_{k,\ell}- \theta_{i_*(t,\ell)-1,\ell}}
= \frac{\sum_{j=i_*(t,\ell)}^k m_{j,\ell} \psi^0_{j,\ell}}{\sum_{j=i_*(t,\ell)}^k m_{j,\ell}}
\ge \psi_{i,\ell}(t) = \dot{x}_{i,\ell}(t) + B[\nu,M](x_{i,\ell}(t), y_\ell, t).
\]
\end{proof}
With Theorem \ref{t:SPCSentropy} in hand, we set the following terminology:
\begin{DEF}
\label{def:SPCSentropysoln}
An \textit{SPCS entropy solution} of \eqref{e:ScalarBalance} is an entropy solution $M$ associated to an admissible triple $(M^0, A, \nu)$, such that $M$ takes the form \eqref{e:MSPCS}, $\nu$ takes the form \eqref{e:nuSPCS}, $M^0$ takes the form \eqref{e:M0SPCS}, and $A$ takes the form \eqref{e:ASPCS}.
\end{DEF}
\subsubsection{Entropy solutions supported on finitely many slices}
Finally, we may state the main result of this subsection, which establishes existence of entropy solutions when the associated transverse measure $\nu$ is atomic. It also clarifies the sense in which such a solution can be approximated by SPCS dynamics.
\begin{THEOREM}
\label{t:sliceexist}
Fix $T>0$. Let $(M^0, A, \nu)$ be an admissible triple such that that $\nu$ is atomic: $\nu = \sum_{\ell=1}^L \mu_\ell \delta(\cdot - y_\ell)$. Then there exists a unique entropy solution $M$ on $[0,T)$ associated to this triple, and $M(\cdot, y_\ell, \cdot)\in BV(\mathbb{R}\times (0,T))$ for each $\ell\in \{1, \ldots, L\}$. The number $R(T)$ in the definition of entropy solution can be taken as
\begin{equation}
R(T) = R^0 + T \big(\max_\ell [A(\cdot, y_\ell)]_{\Lip} + \max_\ell \Phi(2R^0, y_\ell)\big),
\end{equation} where $R^0>0$ is chosen large enough so that $\{x:M^0(x, y_\ell)\notin \{0,1\}\}\subset [-R^0,R^0]$ for each $\ell\in \{1, \ldots, L\}$. Moreover, there exists a sequence $(M_N)_{N=1}^\infty$ of SPCS entropy solutions (with associated admissible triples $(M^0_N, A_N, \nu)$) such that
\begin{equation}
\label{e:MNtoMdisc}
(M_N - M)(\cdot, y_\ell, \cdot)\to 0 \quad \text{ in } \; C([0,T);L^1(\mathbb{R})),\qquad \text{ for all } \ell\in \{1, \ldots, L\}.
\end{equation}
and
\begin{equation}
\label{e:ptMNtoptMdisc}
\partial_t M_N(\cdot, y_\ell, t) \stackrel{*}{\rightharpoonup} \partial_t M(\cdot, y_\ell, t) \qquad \text{ in } \cM(\mathbb{R}),
\qquad \text{ for all } \ell\in \{1, \ldots, L\},\;t\in (0,T).
\end{equation}
Finally, if $\partial_m A(\cdot, y_\ell)$ is H\"older continuous of order $\epsilon\in (0,1]$, for each $\ell$, then $(M_N^0, A_N)$ can be chosen so that \eqref{e:MNtoMdisc} holds with a rate:
\begin{equation}
\label{e:MNtoMdiscrate}
\sup_{t\in [0,T)} \sum_{\ell=1}^{L} \mu_\ell \|M_N(\cdot, y_\ell, t) - M(\cdot, y_\ell, t)\|_{L^1(\mathbb{R})} \le C (1+T) N^{-\epsilon},
\end{equation}
\end{THEOREM}
The full proof of this theorem involves three steps: (1) Discretize the initial condition $M^0$ in $x$ and $A$ in $m$, then generate SPCS entropy solutions $M_N$ associated to the discretized initial data; (2) Extract a limit $M$ as $N\to \infty$ that belongs to the required regularity class and satisfies the required convergences \eqref{e:MNtoMdisc} and \eqref{e:ptMNtoptMdisc}; and (3) Show that the limiting process preserves the Kruzkov entropy condition \eqref{e:entropydistributional}. Now that we have set up the SPCS framework, all of these steps in the $L$-slice configuration under consideration follow exactly as in the 1D case. Rather than duplicating the argument here, we refer the reader to Section 5.1 of \cite{LeslieTan2023} for more details. In any case, the reader will see an argument of almost the same structure below when we prove the existence of a unique entropy solution generated from an arbitrary admissible triple. We will construct these solutions via approximation by a sequence of solutions which are discrete in $y$.
Before moving on to the general case, we state one more Lemma (adapted from Proposition 6.6 of \cite{LeslieTan2023}) that will be useful in Section \ref{ss:rate2} below.
\begin{LEMMA}
\label{l:ICspecialrate}
Let $\nu = \sum_{\ell = 1}^L \mu_\ell \delta(\cdot - y_\ell)$ be an atomic measure on $\mathbb{R}^{d-1}$, and assume that $M^0$ satisfies the requirements in Definition \ref{def:SBIC} for an admissible triple, with some associated $R^0$ such that $\{x:M^0(x,y_\ell)\notin \{0,1\}\} \subset [-R^0, R^0]$ for each $\ell\in \{1, \ldots, L\}$. Let $X^0(\cdot, y_\ell)$ denote the generalized inverse of $M^0(\cdot, y_\ell)$ for each $\ell$:
\[
X^0(m,y_\ell) = (M^0(\cdot, y_\ell))^{-1}(m) = \inf\{x\in \mathbb{R} : M^0(x,y_\ell)\ge m\}.
\]
Then given $N\in \mathbb{N}$, there exists $(m_{i,\ell,N}, x^0_{i,\ell,N})_{i=1}^N$ such that for
\begin{equation}
M^0_N(x,y_\ell) = \sum_{i=1}^N m_{i,\ell, N} H(x - x^0_{i,\ell,N}),
\qquad
X^0_N(\cdot, y_\ell) = (M^0_N(\cdot, y_\ell))^{-1},
\end{equation}
we have for each $\ell\in \{1, \ldots, L\}$ that
\begin{equation}
\label{e:ICspecialrate}
\|M^0(\cdot, y_\ell) - M_N^0(\cdot, y_\ell)\|_{L^1(\mathbb{R})} \le \frac{2R^0}{N},
\qquad
\|X^0(\cdot, y_\ell) - X_N^0(\cdot, y_\ell)\|_{L^\infty(0,1]} \le \frac{2R^0}{N}.
\end{equation}
\end{LEMMA}
\subsection{The general case}
\label{ss:Existgen}
We are ready to tackle the proof of existence of entropy solutions associated to an arbitrary admissible triple $(M^0, A, \nu)$, which we fix for the remainder of this section. We proceed in the following three steps.
\begin{enumerate}[label = \arabic*.]
\item We discretize the triple $(M^0, A, \nu)$ to obtain a sequence $(M^{0,k}, A^k, \nu_k)$, where each $\nu_k$ is atomic. More precisely, $\nu_k = (r_k)_{\sharp} \nu$, where $r_k:\mathbb{R}^{d-1}\to \mathbb{R}^{d-1}$ projects $\supp \nu$ onto a finite subset of the standard dyadic grid.
\item Letting $M^k$ denote the unique entropy solution on $[0,T)$ associated to the triple $(M^{0,k}, A^k, \nu_k)$,
we use \eqref{eq:main-unidir-optimal} to show that the sequence $(M^k(\cdot, r_k(\cdot), \cdot))_{k=1}^\infty$ is Cauchy in the Banach space $C([0,T); L^1(\mathbb{R}\times \mathbb{R}^{d-1};\, \dx \otimes \nu))$ and thus converges to a limit $M$ in this space.
\item We take limits as $k\to \infty$ and show that $M$ satisfies the Kruzkov entropy conditions \eqref{e:entropydistributional}, finishing the proof.
\end{enumerate}
We begin by describing our discretization scheme, which will allow us to state Theorem \ref{thm:existence-main} below. Then we carry out the rest of the agenda detailed above.
\subsubsection{Discretization scheme and statement of the theorem}
Let $\cQ_k$ denote the collection of all cubes of the form $2^{-k}\mathbf{z} + (0, 2^{-k}]^{d-1}$, where $\mathbf{z}\in \Z^{d-1}$. Each cube in $\cQ_k$ is half-open, has edges parallel to the coordinate axes in $\mathbb{R}^{d-1}$, and has side length $2^{-k}$. For fixed $k$, all the cubes in $\cQ_k$ are disjoint and have union all of $\mathbb{R}^{d-1}$. Let $\{Q_{k,\ell}\}_{\ell=1}^{L_k}$ be an enumeration of those cubes in $\cQ_k$ which have strictly positive $\nu$-measure, and let $y_{k,\ell}$ denote the center of $Q_{k,\ell}$. Define $r_k:\mathbb{R}^{d-1}\to \mathbb{R}^{d-1}$ by setting $r_k(y) = y_{k,\ell}$ whenever $y\in Q_{k,\ell}$.
Define the admissible triple $(M^{0,k}, A^k, \nu_k)$ via
\begin{equation}
\label{e:nuk}
\nu_k:=(r_k)_\#\nu=\sum_{\ell=1}^{L_k}\nu(Q_{k,\ell})\,\delta_{y_{k,\ell}},
\end{equation}
\begin{equation}\label{eq:M0k-def-existence}
M^{0,k}(x,y_{k,\ell})
:=
\frac{1}{\nu(Q_{k,\ell})}
\int_{Q_{k,\ell}} M^0(x,\eta)\,\nu(\dd \eta),
\end{equation}
\begin{equation}\label{eq:Ak-def-existence}
A^k(m,y_{k,\ell})
:=
\frac{1}{\nu(Q_{k,\ell})}
\int_{Q_{k,\ell}} A(m,\eta)\,\nu(\dd \eta).
\end{equation}
\begin{THEOREM}
\label{thm:existence-main}
Assume $\phi$ satisfies \eqref{e:phieven} and \eqref{e:phiLip2}; let $(M^0, A, \nu)$ be an admissible triple. For each $k\in~\mathbb{N}$, define $(M^{0,k},A^k,\nu_k)$ as in \eqref{e:nuk}, \eqref{eq:M0k-def-existence}, \eqref{eq:Ak-def-existence}, and let $M^k$ denote the unique entropy solution associated to $(M^{0,k}, A^k, \nu_k)$. Then $(M^k(\cdot, r_k(\cdot), \cdot))_{k=1}^\infty$ converges in $C([0,T);L^1(\mathbb{R}\times \mathbb{R}^{d-1},\dx\otimes \nu))$ to a function $M$,
\begin{equation}
\label{e:MTkconv}
\sup_{t\in [0,T)}\int_{\mathbb{R}^{d-1}} \bigl\| M^{k}(\cdot,r_k(y),t) - M(\cdot,y,t) \bigr\|_{L^1(\mathbb{R})} \,\nu(\dd y) \to 0,
\end{equation}
and $M$ is the unique entropy solution associated to the admissible triple $(M^0, A, \nu)$. Furthermore,
for all $t\in [0,T)$, we have
\begin{equation}
\label{e:partialtMkconv}
\nu_k \partial_t M^k(\cdot, \cdot, t) \stackrel{*}{\rightharpoonup} \nu \partial_t M(\cdot, \cdot, t),\qquad \text{ in } \cM(\mathbb{R}\times \mathbb{R}^{d-1}).
\end{equation}
\end{THEOREM}
\begin{proof}
We first show that the sequence $(M^k(\cdot, r_k(\cdot), \cdot))_{k=1}^\infty$ is Cauchy in $C([0,T), L^1(\mathbb{R}\times \mathbb{R}^{d-1}, \dx \otimes \nu))$ and thus that \eqref{e:MTkconv} holds for some function $M$. We then verify that the entropy conditions enjoyed by $M^k$ are inherited by $M$ in the limit. The convergence \eqref{e:partialtMkconv} falls out as a consequence of these calculations, as we show at the end of the proof.
\subsubsection{Step 1: Existence of a limit}
We start by noting that, by construction, $r_k$ is an optimal transport map (with respect to the cost $c(y,z) = |y-z|$) from $\nu$ to $\nu_k$ (and also from $\nu_{\widetilde{k}}$ to $\nu_k$ for any $\widetilde{k}\ge k$). Furthermore, since we have $|r_k(y) - y|\le C 2^{-k}$ for $\nu$-a.e. $y\in \mathbb{R}^{d-1}$, it follows that
\begin{equation}
\label{e:nukconv}
W_1(\nu_k, \nu) \le C 2^{-k},
\qquad k\in \mathbb{N}.
\end{equation}
Next, the discretizations \eqref{eq:M0k-def-existence} and \eqref{eq:Ak-def-existence} are designed exactly so that
\begin{equation}
\label{e:M0kconv}
\int_{\mathbb{R}^{d-1}} \bigl\| M^{0,k}(\cdot,r_k(y)) - M^0(\cdot,y) \bigr\|_{L^1(\mathbb{R})} \,\nu(\dd y) \to 0,
\end{equation}
\begin{equation}
\label{e:Akconv}
\int_{\mathbb{R}^{d-1}} \bigl\| A^k(\cdot,r_k(y)) - A(\cdot,y) \bigr\|_{L^\infty([0,1])} \,\nu(\dd y) \to 0.
\end{equation}
Therefore, we may apply \eqref{eq:main-unidir-optimal} to $M^k$ and $M^{\widetilde{k}}$ to obtain
\begin{equation}
\label{e:DklCauchy}
\sup_{0\le t<T} \cD^{k,\widetilde{k}}(t)
\le
\bigg( \cD^{k,\widetilde{k}}(0)
+ C \sqrt{ 2R(T) \cA^{k,\widetilde{k}}_\infty T}
+
CT
\, \big( \cA^{k,\widetilde{k}}_\infty + R(T)\|\n_2\phi\|_{L^\infty} W_1(\nu_k,\nu_{\widetilde{k}}) \big) \bigg)e^{C\|\phi\|_{L^\infty}T},
\end{equation}
where
\begin{align*}
\cD^{k,\widetilde{k}}(\tau) & = \int_{\mathbb{R}^{d-1}} \bigl\| M^{k}(\cdot,r_k(y),\tau) - M^{\widetilde{k}}(\cdot,r_{\widetilde{k}}(y),\tau) \bigr\|_{L^1(\mathbb{R})} \,\nu(\dd y),
\qquad \tau\in [0,T), \\
\cA^{k,\widetilde{k}} & = \int_{\mathbb{R}^{d-1}} \bigl\| A^k(\cdot,r_k(y)) - A^{\widetilde{k}}(\cdot,r_{\widetilde{k}}(y)) \bigr\|_{L^\infty([0,1])} \,\nu(\dd y), \\
R(T) & = R^0 + T \big( \sup_y [A(\cdot, y)]_{\Lip} + \sup_y \Phi(2R^0, y)\big),
\end{align*} and $R^0>0$ is chosen large enough so that $\{x:M^0(x,y)\notin \{0,1\}\} \subset [-R^0,R^0]$ for $\nu$-a.e. $y\in \mathbb{R}^{d-1}$. The inequality \eqref{e:DklCauchy}, together with the convergences \eqref{e:M0kconv}, \eqref{e:Akconv}, guarantees that the sequence $(M^k(\cdot, r^k(\cdot), \cdot))_{k=1}^\infty$ is Cauchy (and thus converges to some function $M$) in the space $C([0,T);L^1_{\loc}(\mathbb{R}\times \mathbb{R}^{d-1}, \dx \otimes \nu))$.
\begin{REMARK}
Use of the $L^\infty$ norm in $m$ rather than the Lipschitz seminorm is crucial for \eqref{e:Akconv}, since $A(m,y)$ is only assumed to be Lipschitz in $m$ (with a uniform-in-$y$ bound on the Lipschitz seminorm). One cannot expect in general to generate a nicer $A$ than what we have assumed from arbitrary $\rho^0\in \cP_c(\mathbb{R}\times \mathbb{R}^{d-1})$ and $u^0\in L^\infty(\rho^0)$. This is why the stability estimate \eqref{eq:main-unidir-optimal} is crucial for us. However, if we somehow knew that $A$ were better behaved (for example, if $\partial_m A(m,y)$ were continuous in~$m$), then we would instead be able to replace the $L^\infty$ norm in \eqref{e:Akconv} with a Lipschitz seminorm, and we could use \eqref{eq:stab-pi} instead of \eqref{eq:main-unidir-optimal} in our existence proof. This issue also has implications for the rate of convergence; see Sections \ref{ss:rate} and \ref{ss:rate2} for more details.
\end{REMARK}
\subsubsection{Step 2: Verification of the entropy conditions}
The main task that remains is to show that the limit $M$ constructed above satisfies the entropy condition \eqref{e:entropydistributional}. We use the fact that each $M^k$ is an entropy solution associated to $(M^{0,k}, A^k, \nu_k)$, and therefore satisfies a version of \eqref{eq:entropy-M}, which we record after a few manipulations using the definition $\nu_k = (r_k)_{\sharp}\nu$. Fix an arbitrary nonnegative $\zeta\in C^\infty_c(\mathbb{R}\times \mathbb{R}^{d-1}\times (0,T))$, and set
\[
\widehat{M}^k(x,y,t) = M^k(x,r_k(y),t),
\qquad
\widehat{A}^k(m,y) = A^k(m,r_k(y)),
\qquad
\widehat{\zeta}^k(x,y,t) = \zeta(x,r_k(y),t),
\]
\[
q_\alpha(m,y) = \sgn(m-\alpha) \big( A(m,y) - A(\alpha,y) \big),
\qquad
q_\alpha^k(m,y) = \sgn(m-\alpha) \big( \widehat{A}^k(m,y) - \widehat{A}^k(\alpha,y) \big).
\]
Then the resulting version of \eqref{eq:entropy-M} reads as follows.
\begin{equation}
\label{e:Mkentropy}
0 \le \int_0^T\!\!\int_{\mathbb{R}^{d-1}}\int_{\mathbb{R}} \bigl[ |\widehat{M}^k - \alpha| \partial_t \widehat{\zeta}^k +
q_\alpha^k(\widehat{M}^k, y)
\partial_x \widehat{\zeta}^k
+ \widehat{\zeta}^k B[\nu_k, M^k](x,r_k(y),t) \partial_x |\widehat{M}^k - \alpha| \bigl]\,\dx\,\nu(\dy)\,\dt.
\end{equation}
We show that each of the terms on the left side above converges to its natural limit, and thus that \eqref{eq:entropy-M} holds for $M$ as well. We note first that the contributions from $\widehat{\zeta}^k$, $\partial_t \widehat{\zeta}^k$, and $\partial_x \widehat{\zeta}^k$ pose no difficulty. Indeed, for $\nu$-a.e. $y\in \mathbb{R}^{d-1}$ and all $x\in \mathbb{R}$, $t\in [0,T)$, we have
\[
|\widehat{\zeta}^k(x,y,t) - \zeta(x,y,t)| = |\zeta(x,r_k(y),t) - \zeta(x,y,t)| \le (\sup |\n_2 \zeta|)\cdot |r_k(y) - y|\le C_\zeta 2^{-k}.
\]
Similar bounds hold for $\partial_t \widehat{\zeta}^k$ and $\partial_x \widehat{\zeta}^k$. Here and in what follows, $C_\zeta$ will denote a constant that may depend on $\diam\supp \zeta$ and the supremum of $\zeta$ and its partial derivatives of order up to $2$.
The convergence \eqref{e:MTkconv} says that $\widehat{M}^k\to M$ in $C([0,T);L^1_{\loc}(\mathbb{R}\times \mathbb{R}^{d-1}, \dx \otimes \nu))$, and the convergence of the first term in \eqref{e:Mkentropy} follows immediately:
\[
\int_0^T\!\!\int_{\mathbb{R}^{d-1}}\int_{\mathbb{R}} |\widehat{M}^k - \alpha| \partial_t \widehat{\zeta}^k \,\dx\,\nu(\dy)\,\dt
\to
\int_0^T\!\!\int_{\mathbb{R}^{d-1}}\int_{\mathbb{R}} |M - \alpha| \partial_t \zeta \,\dx\,\nu(\dy)\,\dt, \quad \text{ as } k\to \infty.
\]
Next, we write
\begin{align*}
& \int_0^T\!\!\int_{\mathbb{R}^{d-1}} |q_\alpha^k(\widehat{M}^k,y)\widehat{\zeta}^k - q_\alpha(M,y)\zeta| \, \,\dx\, \nu(\dy)\,\dt \\
& \;\; \le
\int\!\!\!\int\!\!\!\int_{\supp \widehat{\zeta}^k} \bigl[ |q_\alpha^k(\widehat{M}^k,y) - q_\alpha(\widehat{M}^k,y)| + |q_\alpha(\widehat{M}^k,y) - q_\alpha(M,y)| \bigl] \,\dx\, \nu(\dy)\,\dt
+ C_\zeta 2^{-k} \|A\|_{L^\infty(\dx \otimes \nu)} \\
& \;\; \le C_\zeta \int_{\mathbb{R}^{d-1}} \|A^k(\cdot, r_k(y)) - A(\cdot, y)\|_{L^\infty} \nu(\dy) + C_\zeta \|\partial_m A\|_{L^\infty} \int_{\mathbb{R}^{d-1}} \|M^k(\cdot, r_k(y),t) - M(\cdot,y)\|_{L^1} \nu(\dy) + C_\zeta 2^{-k} \|A\|_{L^\infty}.
\end{align*}
The right side converges to zero by \eqref{e:Akconv} and \eqref{e:MTkconv}.
It remains to deal with the nonlocal term. We first show that the quantity $B[\nu_k,M^k](\cdot, r_k(\cdot), \cdot)$ converges to $B[\nu, M]$ in $L^\infty(\dx\otimes \nu\otimes \dt)$. Indeed, for $\nu$-a.e. $y$, we have
\begin{align*}
& |B[\nu_k, M^k](x, r_k(y), t) - B[\nu,M](x,y,t)|\\
& \quad \le \int \!\!\!\int |\Phi(x-\xi, r_k(y) - r_k(\eta)) - \Phi(x - \xi, y - \eta)| \partial_1 \widehat{M}^k(\dd\xi, \eta, t) \, \nu(\dd\eta)
+ \int\!\!\!\int \phi(x - \xi, y - \eta)|(\widehat{M}^k - M)(\xi, \eta, t)| \,\dd \xi \, \nu(\dd\eta) \\
& \quad \le C 2^{-k} R(T) \|\n_2 \phi\|_{L^\infty} + \|\phi\|_{L^\infty} \sup_{t\in [0,T)} \int \|M^k(\cdot, r_k(y),t) - M(\cdot, y, t)\|_{L^1(\mathbb{R})}\,\nu(\dy),
\end{align*}
and the right side converges to zero by \eqref{e:MTkconv}. This allows us to treat the full nonlocal term in \eqref{e:Mkentropy} as follows:
\begin{align*}
& \bigg| \int_0^T\!\!\int_{\mathbb{R}^{d-1}}\int_{\mathbb{R}}
\widehat{\zeta}^k B[\nu_k, M^k](x,r_k(y),t) \partial_x |\widehat{M}^k - \alpha| \bigl]\,\dx\,\nu(\dy)\,\dt
-
\int_0^T\!\!\int_{\mathbb{R}^{d-1}}\int_{\mathbb{R}}
\zeta B[\nu, M] \partial_x |M^k - \alpha| \bigl]\,\dx\,\nu(\dy)\,\dt \bigg| \\
& \quad \le \int_0^T\!\!\int_{\mathbb{R}^{d-1}}\int_{\mathbb{R}} \big| \widehat{\zeta}^k B[\nu_k, M^k](x, r_k(y), t) - \zeta B[\nu,M](x,y,t) \big| \partial_x |\widehat{M}^k - \alpha| \,\dx\,\nu(\dy)\,\dt \\
& \qquad + \int_0^T\!\!\int_{\mathbb{R}^{d-1}}\int_{\mathbb{R}} \big| \partial_x\big( \zeta B[\nu,M] \big) \big| |\widehat{M}^k - M| \,\dx\,\nu(\dy)\,\dt \\
& \quad \le \|\widehat{\zeta}^k B[\nu_k, M^k](\cdot, r_k(\cdot), \cdot) - \zeta B[\nu,M] \|_{L^\infty(\dx \otimes \nu \otimes \dt)} + C_\zeta \|\phi\|_{L^\infty} \sup_{t\in [0,T)} \int \|M^k(\cdot, r_k(y), t) - M(\cdot, y, t)\|_{L^1} \,\nu(\dy).
\end{align*}
We have already shown that both terms on the right converge to zero.
Combining the three convergences above, we may take $k\to \infty$ in all terms in \eqref{e:Mkentropy} to conclude that
\begin{equation}
0 \le \int_0^T\!\!\int_{\mathbb{R}^{d-1}}\int_{\mathbb{R}} \bigl[ |M - \alpha| \partial_t \zeta +
q_\alpha(M, y)
\partial_x \zeta
+ \zeta B[\nu, M] \partial_x |M - \alpha| \bigl]\,\dx\,\nu(\dy)\,\dt,
\end{equation}
which is the same as the required \eqref{eq:entropy-M}.
For the final claim \eqref{e:partialtMkconv}, we note that, since the entropy inequality implies that the scalar balance law \eqref{e:ScalarBalance} holds in the sense of distributions, it suffices to take the weak-$*$ limit on both sides of the equation
\[
\partial_t M^k \nu_k = - \partial_x(A^k(M^k,y)) \nu_k + B[\nu_k,M^k] \partial_x M^k \nu_k.
\]
Our previous computations involving the limits \eqref{e:Mkentropy} show that the right side of this identity converges weak-$*$ to the measure $-\partial_x(A(M,y))\nu + B[\nu,M]\partial_x M \nu = \partial_t M \nu$. This finishes the proof of Theorem \ref{thm:existence-main}.
\end{proof}
\subsection{Rates of convergence}
\label{ss:rate}
In our analysis so far, we have been careful to make sure our arguments work for general admissible triples $(M^0, A, \nu)$, without assuming additional regularity. However, if such regularity is available, we can use it to obtain rates of convergence for the approximation schemes we have already discussed above.
\begin{COROL}
\label{c:MktoMrate}
Assume the hypotheses of Theorem \ref{thm:existence-main}, and let $M$ denote the unique entropy solution generated from $(M^0, A, \nu)$. Assume that $M^0$ satisfies the following H\"older-type estimate, for some $\beta\in (0,1]$ and some constant $C$:
\begin{equation}
\label{e:M0Holdery}
\|M^0(\cdot,y) - M^0(\cdot,\widetilde{y})\|_{L^1(\mathbb{R})}
\le C |y-\widetilde{y}|^\beta,
\qquad \nu\text{-a.e. } y,\widetilde{y}\in \mathbb{R}^{d-1},
\end{equation}
Assume additionally that $A$ satisfies either (for some $\gamma\in (0,1]$)
\begin{equation}
\label{e:AHoldery}
|A(m,y) - A(m,\widetilde{y})| \le C|y - \widetilde{y}|^\gamma,
\qquad m\in [0,1],\;\nu\text{-a.e. } y,\widetilde{y}\in \mathbb{R}^{d-1},
\end{equation}
or
\begin{equation}
\label{e:partialmAHoldery}
|\partial_m A(m,y) - \partial_m A(m,\widetilde{y})|
\le C |y-\widetilde{y}|^\beta,
\qquad m\in [0,1], \;\nu\text{-a.e. } y,\widetilde{y}\in \mathbb{R}^{d-1}.
\end{equation}
\begin{enumerate}[label = (\alph*)]
\item If $A$ satisfies \eqref{e:AHoldery}, then
\begin{equation}
\label{e:MTkconvrate1}
\sup_{t\in [0,T)}\int_{\mathbb{R}^{d-1}} \bigl\| M^{k}(\cdot,r_k(y),t) - M(\cdot,y,t) \bigr\|_{L^1(\mathbb{R})} \,\nu(\dd y) \le C(1+T) 2^{-\min\{\beta, \frac{\gamma}{2}\} k}.
\end{equation}
\item If $A$ satisfies \eqref{e:partialmAHoldery}, then
\begin{equation}
\label{e:MTkconvrate2}
\sup_{t\in [0,T)}\int_{\mathbb{R}^{d-1}} \bigl\| M^{k}(\cdot,r_k(y),t) - M(\cdot,y,t) \bigr\|_{L^1(\mathbb{R})} \,\nu(\dd y) \le C(1+T) 2^{-\beta k}.
\end{equation}
\end{enumerate}
\end{COROL}
\begin{proof}
Under the stated hypotheses, a few elementary estimates yield the following. Whenever \eqref{e:M0Holdery} holds, we have
\begin{equation}
\label{e:M0kconvrate}
\int_{\mathbb{R}^{d-1}} \bigl\| M^{0,k}(\cdot,r_k(y)) - M^0(\cdot,y) \bigr\|_{L^1(\mathbb{R})} \,\nu(\dd y) \le C 2^{-\beta k}.
\end{equation}
Whenever \eqref{e:AHoldery} holds, we have
\begin{equation}
\label{e:Akconvrate}
\int_{\mathbb{R}^{d-1}} \sup_{m\in [0,1]} \big| A^k(m,r_k(y)) - A(m,y) \big| \,\nu(\dd y) \le C 2^{-\gamma k}
\end{equation}
And finally, whenever \eqref{e:partialmAHoldery} holds,
\begin{equation}
\label{e:AkLipconvrate}
\int_{\mathbb{R}^{d-1}} \bigl[ A^k(\cdot,r_k(y)) - A(\cdot,y) \bigr]_{\Lip([0,1])} \,\nu(\dd y) \le C 2^{-\beta k}
\end{equation}
Substituting these bounds (together with \eqref{e:nukconv}) into either \eqref{eq:main-unidir-optimal} or \eqref{eq:stab-pi} yields \eqref{e:MTkconvrate1} and \eqref{e:MTkconvrate2}.
\end{proof}
Combining Corollary \ref{c:MktoMrate} with the last part of Theorem \ref{t:SPCSentropy}, we may obtain a rate of convergence for a sequence of SPCS solutions to a general entropy solution of \eqref{e:ScalarBalance}.
\begin{COROL}\label{c:SPCSrate}
Let $(M^0, A, \nu)$ be an admissible triple. Assume $M^0$ satisfies \eqref{e:M0Holdery}
and that $A$ satisfies
\begin{equation}
\label{e:AHolderm}
|\partial_m A(m,y) - \partial_m A(\widetilde{m},y)|\le C|m-\widetilde{m}|^\epsilon, \qquad m,\widetilde{m}\in [0,1],\;\nu\text{-a.e. } y\in \mathbb{R}^{d-1}.
\end{equation}
We will assume additionally that $A$ satisfies either \eqref{e:AHoldery} or \eqref{e:partialmAHoldery}. Define $(M^{0,k}, A^k, \nu_k)$ as in Theorem \ref{t:existence} (so that each $A^k$ also satisfies \eqref{e:AHolderm}, with $\nu$ replaced by $\nu_k$ but with the same constant $C$); then, for each $k$, let $(M^{0,k}_N, A^k_N)$ be any SPCS initial data and flux that guarantee the convergence \eqref{e:MNtoMdiscrate} for $M^k$. Then the sequence $(M^k_{N(k)})_{k=1}^\infty$ (with $N(k)$ specified below) satisfies the following.
\begin{enumerate}[label = (\alph*)]
\item If $A$ satisfies \eqref{e:AHoldery} and $N(k)\sim 2^{\min\{\beta, \frac{\gamma}{2}\} k/\epsilon}$, then
\begin{equation}
\sup_{t\in [0,T)} \int_{\mathbb{R}^{d-1}} \|M^k_{N(k)}(\cdot, r_k(y), t) - M(\cdot, y, t)\|_{L^1(\mathbb{R})} \nu(\dy) \le C (1+T) 2^{-\min\{\beta, \frac{\gamma}{2}\} k}.
\end{equation}
\item If $A$ satisfies \eqref{e:partialmAHoldery} and $N(k)\sim 2^{\beta k/\epsilon}$, then
\begin{equation}
\sup_{t\in [0,T)} \int_{\mathbb{R}^{d-1}} \|M^k_{N(k)}(\cdot, r_k(y), t) - M(\cdot, y, t)\|_{L^1(\mathbb{R})} \nu(\dy) \le C (1+T) 2^{-\beta k}.
\end{equation}
\end{enumerate}
\end{COROL}
\begin{REMARK}
We note that the total number of agents involved in the discretization in Statement (a) of the Corollary is $O(2^{k\big( \min\{\beta, \frac{\gamma}{2}\}/\epsilon + (d-1)\big)})$, while the total number of agents involved in Statement (b) is $O(2^{k(d-1 + \frac{\beta}{\epsilon})})$. Thus, in (b), the rate of convergence `per agent' is $O(n^{-\beta/(d-1 + \frac{\beta}{\epsilon})})$, where $n$ is the number of agents. This of course reduces to $O(n^{-1/d})$ when $\beta = \epsilon = 1$, whereas the `best-case' scenario for the estimate in statement (a) is a rate of $O(n^{-1/(2d-1)})$.
\end{REMARK}
\section{Weak solutions of the unidirectional Euler Alignment system}
\label{s:Recovery}
In this section we demonstrate how to use the theory developed in the previous sections to generate uniquely determined weak solutions of the unidirectional Euler Alignment system (Section \ref{ss:entropicselection}), and how to approximate them via sticky particle dynamics (Sections \ref{ss:approximation} and \ref{ss:rate2}). We start by making precise what we mean by `weak solution.'
\medskip
\begin{DEF}\label{def:weak_EA}
Let \(\rho^0 \in \mathcal{P}_c(\mathbb{R}\times\mathbb{R}^{d-1})\) be a compactly supported probability measure, and let \(u^0 \in L^\infty(\rho^0)\) be a measurable velocity field. We say that a pair \((\rho, u)\) is a \textit{weak solution} to the unidirectional Euler Alignment system \eqref{e:EAuni} on \([0,T)\) associated to the initial data \((\rho^0, u^0)\) if the following conditions hold:
\begin{enumerate}[label=(\alph*)]
\item \(\rho \in C([0,T); \mathcal{P}_c(\mathbb{R}\times\mathbb{R}^{d-1}))\).
\item $u(t)\in L^\infty(\rho(t))$ for all $t\in [0,T)$.
\item The pair \((\rho, u)\) satisfies \eqref{e:EAuni} in the sense of distributions on \(\mathbb{R}\times\mathbb{R}^{d-1}\times(0,T)\).
\item As $t\to 0+$, we have $\rho(t)\stackrel{*}{\rightharpoonup}\rho^0$ and $\rho u(t)\stackrel{*}{\rightharpoonup}\rho^0 u^0$ in $\cM(\mathbb{R}^d)$.
\end{enumerate}
\end{DEF}
\subsection{The entropic selection principle}
\label{ss:entropicselection}
We now state two fundamental results. The first of these, Theorem \ref{t:entropicselection}, provides a procedure, which we refer to as the \textit{entropic selection principle}, for generating a uniquely determined weak solution of the unidirectional Euler Alignment system \eqref{e:EAuni}. Proposition \ref{p:entropicdiscrete} asserts that the procedure associates atomic initial data with atomic solutions.
\begin{THEOREM}
\label{t:entropicselection} Assume $\phi$ satisfies \eqref{e:phieven} and \eqref{e:phiLip2}.
Suppose $\rho^0\in \cP_c(\mathbb{R}\times \mathbb{R}^{d-1})$ and $u^0\in L^\infty(\rho^0)$. Define $(\rho, u)$ through the following procedure:
\begin{enumerate}[label = \arabic*.]
\item (Construct the admissible triple) Let $p_2:\mathbb{R}\times \mathbb{R}^{d-1}\to \mathbb{R}$ denote projection onto the second component: $p_2(x,y) = y$, and denote $\nu = (p_2)_{\sharp} \rho^0$. Disintegrate $\rho^0$ via $\rho^0(\dx, \dy) = \rho^0_y(\dx) \nu(\dy)$ and define $M^0:\mathbb{R}\times \mathbb{R}^{d-1}\to [0,1]$ via $M^0(x,y) = \rho^0_y((-\infty,x])$. Define $\psi^0 = u^0 + \Phi*\rho^0$ and $A:[0,1]\times \mathbb{R}^{d-1}\to \mathbb{R}$ via
\begin{equation}
\label{e:defA}
A(m,y) = \int_0^m \psi^0(X^0(\widetilde{m},y),y)\,\dd\widetilde{m},
\qquad X^0(m,y) = \inf\{x\in \mathbb{R}: M^0(x,y)\ge m\} = (M^0(\cdot, y))^{-1}(m).
\end{equation}
\item (Generate the solution to the scalar balance law) Let $M$ denote the unique entropy solution of \eqref{e:ScalarBalance} associated to the admissible triple $(M^0,A,\nu)$.
\item (Generate the solution of the Euler Alignment system) Define $\rho(t)$ via $\rho(\dx, \dy, t) = \partial_x M(\dx, y, t)\nu(\dy)$ and $P(t)$ via $P(\dx, \dy, t) = -\partial_t M(\dx, y, t)\nu(\dy)$, and let $u(t)\in L^\infty(\rho(t))$ denote the Radon-Nikodym derivative of $P(t)$ with respect to $\rho(t)$; that is, $u(t) = \frac{\dd P(t)}{\dd\rho(t)}$.
\end{enumerate}
Then $(\rho, u)$ is a weak solution of \eqref{e:EAuni} in the sense of Definition \ref{def:weak_EA}. If $\nu$ is atomic, then the assumption \eqref{e:phiLip2} can be relaxed to~\eqref{e:phiL1locx}.
\end{THEOREM}
\begin{proof}
The continuity equation $\partial_t \rho + \partial_x (\rho u) = 0$ follows simply from the definitions of $P$ and $u$:
\[
\partial_t \rho + \partial_x (\rho u) = \partial_t \partial_x M \nu + \partial_x P = 0.
\]
The BV chain rule (c.f. Lemma \ref{l:BVChain}) guarantees that we can define $\psi(t)\in L^\infty(\rho(t))$ via
\begin{equation}
\partial_x (A(M,\cdot)) = \psi \partial_x M,
\qquad
\partial_t (A(M,\cdot)) = \psi\partial_t M.
\end{equation}
Therefore,
\[
\partial_x(A(M,y)) \nu = \rho \psi,
\qquad \partial_t (A(M,y)) \nu = -P\psi = -\rho u \psi.
\]
This immediately implies that $\partial_t(\rho \psi) + \partial_x (\rho u \psi) = 0$ (which is \eqref{e:EAunipsi}$_2$). In order to upgrade to the original formulation \eqref{e:EAuni}$_2$, we need to establish the formula $\psi = u + \Phi*\rho$ directly from the procedure outlined in the statement of Theorem \ref{t:entropicselection}. The distributional form of the scalar balance law provides the link:
\[
\rho u = P = -\partial_t M \nu = \partial_x (A(M,y))\nu - (\Phi*\nu\partial_x M) \partial_x M \nu = (\psi - \Phi*\rho) \rho.
\]
Substituting this relationship into \eqref{e:EAunipsi} immediately yields \eqref{e:EAuni}$_2$.
We now consider the initial conditions and take $t\to 0+$. The convergence $\rho(t)\stackrel{*}{\rightharpoonup} \rho^0$ follows from the corresponding convergence of $M(t)$ to $M^0$. Continuity of $\Phi*\rho$ then implies that $(\Phi*\rho(t))\rho(t)\stackrel{*}{\rightharpoonup} (\Phi*\rho^0)\rho^0$. Next, we have
\[
\rho \psi(t) = \partial_x (A(M(t),\cdot))\nu \stackrel{*}{\rightharpoonup} \partial_x (A(M^0, \cdot)) \nu = \partial_x \bigg( \int_{(-\infty, x]} \psi^0 \partial_x M^0 \bigg)\nu = \rho^0 \psi^0,
\]
where the penultimate identity follows from the definition \eqref{e:defA} of $A$ and the identity \eqref{e:geninvid}. Combining the two convergences $\rho \psi(t)\stackrel{*}{\rightharpoonup} \rho^0\psi^0$ and $(\Phi*\rho)\rho(t)\stackrel{*}{\rightharpoonup} (\Phi*\rho^0)\rho^0$ with the relationships $\psi = u + \Phi*\rho$ and $\psi^0 = u^0 + \Phi*\rho^0$ gives us the desired convergence $\rho u(t)\stackrel{*}{\rightharpoonup} \rho^0 u^0$.
\end{proof}
\begin{PROP}
\label{p:entropicdiscrete}
Assume $\phi$ satisfies \eqref{e:phieven} and \eqref{e:phiL1locx}. Assume $\rho^0\in \cP_c(\mathbb{R}^d)$ is atomic: $\rho^0 = \sum_{i=1}^N m_i \delta(\cdot - \bm{x}_i^0)$ and fix $u^0\in L^\infty(\rho^0)$ (real-valued), so that $\rho^0 u^0$ is of the form $\rho^0 u^0 = \sum_{i=1}^N m_i v_i^0 \delta(\cdot - \bm{x}_i^0)$. Let $(m_i, \bm{x}_i(t), v_i(t))_{i=1}^N$ denote the unidirectional sticky particle Cucker--Smale dynamics associated to $(m_i, \bm{x}_i^0, v_i^0)_{i=1}^N$. Then the solution $(\rho, u)$ of \eqref{e:EAuni} generated by the entropic selection principle in Theorem \ref{t:entropicselection} is given by
\begin{equation}
\label{e:EASPCS}
\rho(t) = \sum_{i=1}^N m_i \delta(\cdot - \bm{x}_i(t)),
\qquad
\rho u(t) = \sum_{i=1}^N m_i v_i(t) \delta(\cdot - \bm{x}_i(t)),
\qquad t\ge 0.
\end{equation}
\end{PROP}
We omit the proof of this proposition, which is analogous to that of Proposition 6.4 in \cite{LeslieTan2023}. We refer to solutions of \eqref{e:EAuni} of the form \eqref{e:EASPCS} as `sticky particle' weak solutions of the unidirectional Euler Alignment system.
\subsection{Approximation by atomic solutions}
\label{ss:approximation}
We finally spell out explicitly the sense in which the SPCS weak solutions generated in Proposition \ref{p:entropicdiscrete} can be used to approximate solutions arising from general unidirectional initial data $(\rho^0, u^0)$. Theorem~\ref{t:SPCSconvgen} provides weak-$*$ convergence of the density and momentum profiles for arbitrary weak solutions. We explore additional assumptions in the next subsection that imply a rate of convergence of the density profile.
\begin{THEOREM}
\label{t:SPCSconvgen}
Assume $\phi$ satisfies \eqref{e:phieven} and \eqref{e:phiLip2}.
Suppose $\rho^0\in \cP_c(\mathbb{R}^{d})$ and $u^0\in L^\infty(\rho^0)$, and let $(\rho, u)$ be the solution generated from $(\rho^0, u^0)$ by the entropic selection principle in Theorem \ref{t:entropicselection}. There exists a sequence $(\rho_n, u_n)_{n=1}^\infty$ of sticky particle weak solutions such that $\rho_n(t)\stackrel{*}{\rightharpoonup}\rho(t)$ and $\rho_n u_n(t)\stackrel{*}{\rightharpoonup}\rho u(t)$ for all $t\ge 0$.
\end{THEOREM}
\begin{proof}
Let $(M^0, A, \nu)$ be the admissible triple from Step 1 of the procedure in Theorem \ref{t:entropicselection}. For each $k\in \mathbb{N}$, define $(M^{0,k}, A^k, \nu_k)$ as in \eqref{e:nuk}, \eqref{eq:M0k-def-existence}, \eqref{eq:Ak-def-existence}. Define $\rho^{0,k}$ and $u^{0,k}$ via
\[
\rho^{0,k}(\dx, \dy) = \partial_1 M^{0,k}(\dx, y)\nu_k(\dy),
\qquad \qquad
\rho^{0,k} u^{0,k} = \partial_x(A^k(M^{0,k}, \cdot))\nu_k - (\Phi*\rho^{0,k})\rho^{0,k}.
\]
Let $M^k$ denote the entropy solution of \eqref{e:ScalarBalance} generated from $(M^{0,k}, A^k, \nu_k)$ by Theorem \ref{t:existence}, and let $(\rho^k, u^k)$ denote the solution generated from $(\rho^{0,k}, u^{0,k})$ from Theorem \ref{t:entropicselection}. Combining \eqref{e:Wassersteinrelationship} and \eqref{e:nukconv} yields
\begin{equation}
\label{e:W1vsL1M}
W_1(\rho^k(t), \rho(t)) \le C2^{-k} + \int_{\mathbb{R}^{d-1}} \|M^k(\cdot, r_k(y), t) - M(\cdot, y, t)\|_{L^1(\mathbb{R})} \,\nu(\dy),
\end{equation}
so that by \eqref{e:MTkconv}, we have $\rho^k(t)\stackrel{*}{\rightharpoonup} \rho(t)$. The fact that $\rho^k u^k(t)\stackrel{*}{\rightharpoonup} \rho u(t)$ is exactly \eqref{e:partialtMkconv}.
Similarly, Theorem \ref{t:sliceexist} provides for each $k$ the existence of a sequence $(\rho^k_N, u^k_N)_{N=1}^\infty$ of sticky particle weak solutions such that $(\rho^k_N(t))_{N=1}^\infty$ and $(\rho^k_N u^k_N(t))_{N=1}^\infty$ converge weak-$*$ to $\rho^k(t)$ and $\rho^k u^k(t)$, respectively, for each $t\in [0,T)$.
Choose $R>0$ large enough so that $\supp \rho^0$ is contained in the open ball $B(0,R)$ of radius $R$ in $\mathbb{R}^d$, centered at the origin. Let $(\zeta_i)_{i=1}^\infty$ be an enumeration of a countable dense subset of the space of continuous functions on $\overline{B(0,R)}$ under the uniform norm. We can find a sequence of integers $(N_{k_1})_{k_1=1}^\infty$ such that $\int \zeta_1 \rho_{N_{k_1}}^{k_1}(t)\to \int \zeta_1 \rho(t)$ and $\int \zeta_1 \rho_{N_{k_1}}^{k_1}u_{N_{k_1}}^{k_1}(t)\to \int \zeta_1 \rho u(t)$, respectively, as $k_1\to \infty$, and for all rational $t\in [0,T)$. Having chosen $(N_{k_{j-1}})_{k_{j-1}=1}^\infty$, choose a further subsequence $(N_{k_j})_{k_j=1}^\infty$ such that $\int \zeta_j \rho_{N_{k_j}}^{k_j}(t)\to \int \zeta_j \rho(t)$ and $\int \zeta_j \rho_{N_{k_j}}^{k_j}u_{N_{k_j}}^{k_j}(t)\to \int \zeta_j \rho u(t)$, for all rational $t\in [0,T)$. Relabeling $(\rho_n, u_n) := (\rho^{k_n}_{N_{k_n}}, u^{k_n}_{N_{k_n}})$, a density argument shows that we have the required convergence against all $\zeta\in C(\overline{B(0,R)})$ (and thus all $\zeta\in C_c(\mathbb{R}^d)$) for rational times. Finally, continuity in time of $\rho$, $\rho u$, and each $\rho_n$ and $\rho_n u_n$, allows us to extend the convergence from rational $t$ to all $t\in [0,T)$.
\end{proof}
\subsection{Rates of Convergence}
\label{ss:rate2}
Under additional assumptions on our initial conditions $(\rho^0, u^0)$, we can guarantee a rate of convergence of a sequence of sticky particle weak solution density profiles $(\rho^k_{N(k)})_{k=1}^\infty$ (with $N(k)$ explicitly specified) to the density profile $\rho$ of the weak solution generated by Theorem \ref{t:entropicselection}. We formulate two results toward this end; let us provide some context before presenting their full statements and proofs.
The simpler of our two statements is Corollary \ref{c:SPCSratestrong}; its hypotheses are designed so that Corollary \ref{c:SPCSrate} can be applied, and its proof is essentially a matter of translating between the scalar balance law formulation and the density/momentum formulation. However, the assumptions on $(\rho^0, u^0)$ needed to guarantee H\"older regularity of $\partial_m A(m,y) = \psi^0(X^0(m,y),y)$ in $m$ (c.f. \eqref{e:AHolderm}) and $y$ (c.f. \eqref{e:partialmAHoldery}) are rather stringent, mostly because $m\mapsto X^0(m,y)$ contains jumps across intervals of vacuum, no matter how nice $\rho^0$ and $u^0$ are otherwise.
The more subtle (and arguably more interesting) of these two results is Theorem \ref{t:SPCSconvrateweak}, which is valid under much milder hypotheses than Corollary \ref{c:SPCSratestrong}. Its proof leverages Corollary \ref{c:MktoMrate}(a) to discretize in $y$, but the subsequent horizontal discretization avoids making additional assumptions on the flux by leveraging information directly from (the discrete-in-$y$ versions of) $\rho^0$ and $u^0$ via Lemma \ref{l:ICspecialrate}. Since the first step of the discretization is based on the estimate \eqref{eq:main-unidir-optimal} rather than \eqref{eq:stab-pi}, the end result suffers from a bottleneck in the convergence rate related to the square root on the flux term in \eqref{eq:main-unidir-optimal}; however, its more flexible hypotheses give it a distinct advantage over Corollary \ref{c:SPCSratestrong} in terms of applicability. Note in particular that Corollary~\ref{c:SPCSratestrong} requires a uniform lower bound on the density within its support, while no such condition is needed in Theorem \ref{t:SPCSconvrateweak}.
\begin{COROL}
\label{c:SPCSratestrong}
Assume $\phi$ satisfies \eqref{e:phieven} and \eqref{e:phiLip2}.
Assume that $\rho^0\in \cP_c(\mathbb{R}^d)$ and $u^0\in L^\infty(\rho^0)$. Assume that $u^0$ has a representative that satisfies a H\"older estimate of the form
\begin{equation}
\label{e:u0Holder}
|u^0(x, y) - u^0(\widetilde{x}, \widetilde{y})|\le C\big(|x - \widetilde{x}|^\delta + |y - \widetilde{y}|^\gamma\big).
\end{equation} Assume additionally that $\rho^0$ can be identified with a Lebesgue integrable function, which we continue to denote by $\rho^0$, i.e., $\rho^0\in L^1(\mathbb{R}^d)$, and denote $\nu(y) = \int_{\mathbb{R}} \rho^0(x,y)\,\dx$.
Assume that $\supp \rho^0$ is of the form
\[
\supp \rho^0 = \{(x,y)\in \mathbb{R}\times \mathbb{R}^{d-1}: a(y)\le x\le b(y)\},
\qquad a,b\in C^1(\supp \nu),
\qquad a(y)<b(y),
\]
and that $\rho^0\ge c>0$ on $\supp \rho^0$. Finally, assume that $\rho^0$ satisfies the following H\"older estimate:
\begin{equation}
\label{e:rho0Holder}
|\rho^0(x,y) - \rho^0(x,\widetilde{y})| \le C|y - \widetilde{y}|^\beta,
\qquad (x,y),(x,\widetilde{y})\in \supp \rho^0.
\end{equation}
Then there exists a sequence of explicitly constructed sticky particle solutions $(\rho_k, u_k)_{k=1}^\infty$, consisting of $O(2^{k(d-1 + \beta)})$ agents, such that
\begin{equation}
\sup_{t\in [0,T)} W_1(\rho_k(t), \rho(t)) \le C(1+T) 2^{-\beta \delta k}.
\end{equation}
\end{COROL}
\begin{proof}
The stated assumptions guarantee that
\begin{equation}
|\nu(y) - \nu(\widetilde{y})|\le C|y-\widetilde{y}|^\beta,
\quad y,\widetilde{y}\in \supp \nu,
\qquad \nu\ge c\min_{y\in \supp \nu} (b(y) - a(y))>0.
\end{equation}
This and the assumption \eqref{e:rho0Holder} imply that
\begin{equation}
|M^0(x,y) - M^0(x,\widetilde{y})|\le C|y - \widetilde{y}|^\beta,
\qquad (x,y), (x,\widetilde{y})\in \supp \rho^0.
\end{equation}
This implies in particular that \eqref{e:M0Holdery} holds. Next, the regularity of $M^0$ and the lower bound on $\rho^0$ imply that
\[
|X^0(m,y) - X^0(\widetilde{m}, \widetilde{y})|\le C\big( |m- \widetilde{m}| + |y - \widetilde{y}|^\beta\big),
\qquad y,\widetilde{y}\in \supp \nu.
\]
The assumption \eqref{e:u0Holder} guarantees that $\psi^0 = u^0 + \Phi*\rho^0$ also satisfies a H\"older estimate of the same form that $u^0$ does. Therefore,
\begin{align*}
|\partial_m A(m,y) - \partial_m A(\widetilde{m}, \widetilde{y})|
\le C\big(|X^0(m,y) - X^0(\widetilde{m}, \widetilde{y})|^\delta + |y - \widetilde{y}|^\beta\big) \le C \big( |m-\widetilde{m}|^\delta + |y - \widetilde{y}|^{\beta\delta}\big),
\qquad y,\widetilde{y} \in \supp \nu.
\end{align*}
Corollary \ref{c:SPCSrate} is thus applicable and allows us to finish the proof.
\end{proof}
\begin{THEOREM}
\label{t:SPCSconvrateweak}
Assume $\phi$ satisfies \eqref{e:phieven} and \eqref{e:phiLip2}.
Assume that $\rho^0\in \cP_c(\mathbb{R}^d)$; define $\nu = (p_2)_{\sharp} \rho^0$ and $\rho^0_y$ via $\rho^0(\dx, \dy) = \rho^0_y(\dx) \nu(\dy)$ as usual. Assume that $u^0$ satisfies \eqref{e:u0Holder} and that $y\mapsto \rho^0_y$ satisfies the following H\"older-type estimate:
\begin{equation}
\label{e:rho0Holdery}
W_1(\rho^0_y, \rho^0_{\widetilde{y}}) \le C |y - \widetilde{y}|^{\beta},
\qquad y,\widetilde{y}\in \supp \nu.
\end{equation}
Then there exists a sequence of sticky particle weak solutions $(\rho_k, u_k)_{k=1}^\infty$, consisting of $O(2^{k( (d-1) + \frac{1}{2} \min\{\beta, \frac{\gamma}{\delta}\})})$ agents, such that
\begin{equation}
\sup_{t\in [0,T)} W_1(\rho(t), \rho_k(t)) \le C 2^{-\frac{k}{2} \min\{\beta \delta, \gamma \}}.
\end{equation}
\end{THEOREM}
\begin{proof}
Define $M^0$ and $A$ as in Theorem \ref{t:entropicselection}. Our assumption \eqref{e:rho0Holdery} is equivalent to \eqref{e:M0Holdery}. Our first aim is therefore to show that \eqref{e:AHoldery} holds with the appropriate exponent $\min\{\beta\delta, \gamma\}$, so that we can apply Corollary \ref{c:MktoMrate}(a). We estimate as follows, noting that assumption \eqref{e:u0Holder} implies that $\psi^0 = u^0 + \Phi*\rho^0$ satisfies an estimate of the same form and using \eqref{e:W1id} to pass to the third line below.
\begin{align*}
|A(m,y) - A(m,\widetilde{y})|
& \le \int_0^1 \big|\psi^0(X^0(m,y),y) - \psi^0(X^0(m,\widetilde{y}), \widetilde{y})\big| \dm \\
& \le C \int_0^1 |X^0(m,y) - X^0(m,\widetilde{y})|^\delta \dm + C|y-\widetilde{y}|^\gamma\\
& \le C W_1(\rho^0_y, \rho^0_{\widetilde{y}})^\delta + C|y - \widetilde{y}|^\gamma \\
& \le C |y-\widetilde{y}|^{\min\{\beta\delta, \gamma\}}
\end{align*}
We may thus define (for each $k\in \N$) the triple $(M^{0,k}, A^k, \nu_k)$ as in \eqref{e:nuk}, \eqref{eq:M0k-def-existence} and \eqref{eq:Ak-def-existence}, and the entropy solution $M^k$ generated from this admissible triple will satisfy \eqref{e:MTkconvrate1}, per Corollary \ref{c:MktoMrate}(a). For use below, we also denote by $(\rho^{0,k}, u^{0,k})$ the corresponding Euler Alignment initial data and $(\rho^k, u^k)$ the corresponding weak solution. Finally, we define $\psi^{0,k} = u^{0,k} + \Phi*\rho^{0,k}$.
Next, we discretize in the direction of the flow. Our current assumptions do not grant us access to \eqref{e:AHolderm} (which would have in turn allowed us to use the rate in the last statement of Theorem \ref{t:sliceexist}); we instead use the discretization provided by Lemma~\ref{l:ICspecialrate}. Write $\nu_k = \sum_{\ell=1}^{L_k} \mu_{k,\ell} \delta(\cdot - y_{k,\ell})$, and let $(m^{k}_{i,\ell, N}, x^{0,k}_{i,\ell,N})_{i=1}^N$, $M^{0,k}_N$, and $X^{0,k}_N$ be as in Lemma \ref{l:ICspecialrate}, so that for each $k$, $\ell$, and $N$, we have
\[
\|M^{0,k}(\cdot, y_{k,\ell}) - M_N^{0,k}(\cdot, y_{k,\ell})\|_{L^1(\mathbb{R})}\le \frac{2R^0}{N},
\qquad
\|X^{0,k}(\cdot, y_\ell) - X^{0,k}_N(\cdot, y_\ell)\|_{L^\infty(0,1]} \le \frac{2R^0}{N}.
\]
Define
\[
\psi^{0,k}_{i,\ell, N}
= \psi^{0,k}(x^{0,k}_{i,\ell, N}, y_{k,\ell}),
\qquad
v^{0,k}_{i,\ell, N}
= \psi^{0,k}_{i,\ell, N} - \sum_{p = 1}^{L_k} \sum_{j=1}^N \mu_{k,p} m^k_{i,p, N} \Phi( x^{0,k}_{i,\ell, N} - x^{0,k}_{j,p, N}, y_{k,\ell} - y_{k,p}),
\]
\[
\theta^k_{i,\ell, N} = \sum_{j=1}^i m^k_{j,\ell, N},\;\;i=1,\ldots, N,\qquad \theta^k_{0,\ell, N} = 0.
\]
Finally, let $(M^{0,k}_N, A^k_N, \nu_k)$ denote the admissible triple associated to the SPCS initial data defined above, and let $M^k_N$ denote the entropy solution generated from it. Let $(\rho^{0,k}_N, u^{0,k}_N)$ and $(\rho^k_N, u^k_N)$ denote the associated Euler Alignment initial data and weak solution. For $m\in (\theta^k_{i-1,\ell, N}, \theta^k_{i,\ell, N})$, we have
\begin{align*}
|\partial_m A^k (m,y_{k,\ell}) - \partial_m A^k_N(m,y_{k,\ell})|
& = |\psi^{0,k}(X^0(m,y_{k,\ell}), y_{k,\ell}) - \psi^{0,k}_{i,\ell, N}| \\
& = |\psi^{0,k}(X^0(m,y_{k,\ell}), y_{k,\ell}) - \psi^{0,k}(X^{0,k}_N(\theta^k_{i,\ell, N}, y_{k,\ell}), y_{k,\ell})| \\
& \le C|X^0(m,y_{k,\ell}) - X^{0,k}_N(m,y_{k,\ell})|^\delta \\
& \le CN^{-\delta}
\end{align*}
This allows us to use \eqref{eq:main-unidir-optimal} to conclude that
\begin{equation} \sup_{t\in [0,T)} \|M^k_N(t) - M^k(t)\|_{L^1(\dx\otimes \nu_k)} \le CN^{-1} + CN^{-\delta} \le CN^{-\delta}.
\end{equation}
Therefore,
\begin{equation}
\sup_{t\in [0,T)} W_1(\rho(t),\rho^k_N(t)) \le \sup_{t\in [0,T)} W_1(\rho(t), \rho^k(t)) + \sup_{t\in [0,T)} W_1(\rho^k(t), \rho^k_N(t)) \le C 2^{-\frac{k}{2} \min\{\beta \delta, \gamma\}} + C N^{-\delta}
\end{equation}
Choosing $N\sim 2^{\frac{k}{2} \min\{\beta, \frac{\gamma}{\delta}\}}$ finishes the proof.
\end{proof}
\section{Flocking}
\label{s:Flocking}
We now consider the long-time behavior associated with unidirectional Cucker--Smale and Euler Alignment dynamics. We start with the former, which already contains the main novelty for our purposes.
\subsection{Flocking for unidirectional Cucker--Smale dynamics}
It has been known for some time (c.f. \cite{HaLiu2008}) that solutions of the Cucker--Smale ODEs experience flocking and exponentially fast velocity alignment for any initial configuration, if the communication protocol $\phi:\mathbb{R}^d\to \mathbb{R}$ is even and `heavy-tailed'. When we say that $\phi$ is `heavy-tailed', we mean that there exists a radially nonincreasing function $\underline{\phi}:[0,\infty)\to \mathbb{R}$ such that $\phi(\bm{x})\ge \underline{\phi}(|\bm{x}|)$ and $\underline{\phi}$ is nonintegrable at infinity:
\begin{equation}
\label{e:heavytail}
\int_1^\infty \underline{\phi} = +\infty.
\end{equation}
In fact, the authors of \cite{HaLiu2008} already obtain a more quantitative version of the heavy-tail condition that depends on the initial configuration. Whether \eqref{e:heavytail} or a more refined version is used, however, one can think of the underlying mechanism as follows: Initially, every agent interacts with every other agent. The heavy tail condition ensures that the spatial decay of these interactions is just slow enough that it is impossible for one agent to completely escape the influence of any other; it becomes trapped within some fixed diameter of the rest of the flock, so that the minimal strength of all pairwise interactions remains bounded below, which drives the exponentially fast velocity alignment.
Actually, even in Cucker and Smale's paper \cite{CS2007b}, it was shown that flocking can be guaranteed even if some agents never communicate with each other. What is important is the connectivity of the underlying graph structure of the communication network. In particular, flocking and exponentially fast velocity alignment is guaranteed as long as the time dependent `weighted Fiedler number' $\kappa_2$ is nonintegrable: $\int_0^\infty \kappa_2(t)\,\dt = +\infty$. For more on this direction (and a precise definition of the weighted Fiedler number), the reader may consult \cite{CS2007b, MT2014, Tadmor2021review, ShvydkoyBook}, among many other references. Using the Fiedler number to establish flocking, however, typically results in bounds on the flock diameter (and the rate of alignment) that depend on the number of agents $N$; as such, these techniques often do not extend well in the hydrodynamic setting.
In the theorem below, we prove that the Cucker--Smale system experiences flocking (at a rate independent of the number of agents) even if the communication degenerates in a cylinder around the agents, provided that there is enough communication with agents outside each cylinder. The cylinders are invariant under the unidirectional dynamics, so the communication network is effectively frozen if one imposes an analog of the heavy-tail condition outside the cylinder of degeneracy. If we take the radius $r$ of the cylinders to be zero, we recover a more standard type of flocking estimate, exactly in line with the quantitative version of Ha and Liu's estimate.
\begin{THEOREM}
\label{t:flockingCS}
Assume $\phi$ satisfies \eqref{e:phieven} and \eqref{e:phiL1locx}. Let $(m_i, \bm{x}_i(t), \bm{v}_i(t))_{i=1}^N$ be a solution of the Cucker--Smale system \eqref{e:CS} associated to the initial data $(\bm{x}_i^0,\bm{v}_i^0)_{i=1}^N$. Assume that $\sum_{i=1}^N m_i = 1$ and that the initial data (and therefore the solution) is unidirectional; denote $\bm{x}_i(t) = (x_i(t), y_i)$, $\bm{v}_i(t) = (v_i(t), 0)$. Define furthermore the quantities
\begin{equation}
\label{e:DAdef}
\mathscr{D}_1(t) = \max_{i,j} (x_i(t) - x_j(t)),
\qquad
\mathscr{D}_2 = \max_{i,j} |y_i - y_j|,
\qquad
\mathscr{A}(t) = \max_{i,j} (v_i(t) - v_j(t)),
\end{equation}
and define $\mathscr{D}_1^0$ and $\mathscr{A}^0$ analogously in terms of the initial data. Assume there exists $r>0$ with the following properties (here $B(y,r)$ denotes the open ball of radius $r$ in $\mathbb{R}^{d-1}$ centered at $y$):
\begin{itemize}
\item The proportion of the total mass contained in any two balls of radius $r$ in $\mathbb{R}^{d-1}$ is bounded away from $1$:
\begin{equation}
\label{e:flockingreq1}
\sum_{\{i:y_i\in B(y,r)\cup B(z,r)\}} m_i < 1-c,
\qquad \text{ for all } y,z\in \mathbb{R}^{d-1}
\end{equation}
\item There exists a continuous function $\underline{\phi}:[0,\infty)\times [0,\infty)\to \mathbb{R}$, nonincreasing in both its arguments on $[0,\infty) \times [r,\mathscr{D}_2]$, such that $\phi(x,y)\ge \underline{\phi}(|x|,|y|)$, and $\underline{\phi}$ additionally satisfies (with the same constant $c$ as in \eqref{e:flockingreq1}):
\begin{equation}
\label{e:heavytailsub}
c\int_{\mathscr{D}^0_1}^\infty \underline{\phi}(x,\mathscr{D}_2)\,\dx > \mathscr{A}^0,
\end{equation}
\end{itemize}
Then the solution experiences flocking and exponentially fast alignment: There exists $\overline{\mathscr{D}}_1>0$ such that
\begin{equation}
\label{e:uniflocking}
\sup_{t\ge 0} \mathscr{D}_1(t)\le \overline{\mathscr{D}}_1,
\qquad
\mathscr{A}(t)
\le \mathscr{A}^0 e^{-ct \underline{\phi}(\overline{\mathscr{D}}_1,\mathscr{D}_2)}.
\end{equation}
\end{THEOREM}
\begin{REMARK}
Our purpose in presenting this theorem is to point out that unidirectional dynamics provide a powerful mechanism for driving flocking and velocity alignment, even when the communication protocol involves degeneracies. Our hypotheses are almost certainly not optimal, and we fully expect that the statement could be sharpened by using chain connectivity arguments in the spirit of, e.g., \cite{MPT2019} or \cite{STtopo}.
\end{REMARK}
\begin{proof}
Fix a time $t$ such that $\cA(t)$ is differentiable, then choose a pair $(i,j)$ such that $\mathscr{A}(t) = v_i(t) - v_j(t)$. Then $v_k(t)-v_i(t)$ and $v_j(t) - v_k(t)$ are nonpositive for any $k$, so
\begin{align*}
\dot{\mathscr{A}}(t)
& = \dot{v_i}(t) - \dot{v}_j(t) \\
& = -\sum_{k=1}^N m_k \phi(\bm{x}_i(t) - \bm{x}_k(t)) (v_i(t) - v_k(t))
+ \sum_{k=1}^N m_k \phi(\bm{x}_j(t) - \bm{x}_k(t)) (v_j(t) - v_k(t)) \\
& \le \sum_{k:y_k\notin B(y_i,r)\cup B(y_j,r)} m_k \underbrace{\phi(\bm{x}_i(t) - \bm{x}_k(t))}_{\ge \underline{\phi}(\mathscr{D}_1(t), \mathscr{D}_2)} (v_k(t) - v_i(t))
+ m_k \underbrace{\phi(\bm{x}_j(t) - \bm{x}_k(t))}_{\ge \underline{\phi}(\mathscr{D}_1(t), \mathscr{D}_2)} (v_j(t) - v_k(t)) \\
& \le - \Bigg( \sum_{k:y_k\notin B(y_i,r)\cup B(y_j,r)} m_k \Bigg) \underline{\phi}(\mathscr{D}_1(t), \mathscr{D}_2) \,(v_i(t) - v_j(t)) \\
& \le - c \underline{\phi}(\mathscr{D}_1(t), \mathscr{D}_2) \mathscr{A}(t).
\end{align*}
Rademacher's Theorem implies that the Lipschitz function $\mathscr{A}$ is differentiable for a.e. $t$, so the bound established above is available globally. From here, it is easy to see that
\begin{equation}
\mathscr{L}(t) := \mathscr{A}(t) + c \int_0^{\mathscr{D}_1(t)} \underline{\phi}(x,\mathscr{D}_2)\,\dx
\end{equation}
is a Lyapunov function for the dynamics. It follows that if we choose $\overline{\mathscr{D}}_1$ such that
\[
c \int_0^{\overline{\mathscr{D}}_1} \underline{\phi}(x,\mathscr{D}_2)\,\dx = \mathscr{L}(0),
\]
which is possible by our assumption \eqref{e:heavytailsub}, then \eqref{e:uniflocking} holds.
\end{proof}
\begin{figure}[htbp]
\begin{minipage}{0.44\linewidth}
\resizebox{\textwidth}{!}{%
\begin{tikzpicture}[
>=Latex,
agent/.style={circle, fill=black, inner sep=1.5pt}, %
hatch/.style={pattern=north east lines, pattern color=black!50},
note/.style={align=left, font=\small}
]
\begin{scope}[xshift=0cm, scale=0.9, every node/.append style={scale=0.9}]
\node[font=\bfseries] at (0, 4.4) {Support of the Communication Kernel $\phi$};
\draw[->, thick] (-5.4, 0) -- (4.0, 0) node[right] {$x$};
\draw[->, thick] (0, -3.8) -- (0, 3.9) node[left] {$y$};
\fill[hatch] (-3.5, 0.6) rectangle (3.5, 3.5);
\fill[hatch] (-3.5, -3.5) rectangle (3.5, -0.6);
\draw[thick] (-3.5, 3.5) -- (3.5, 3.5);
\draw[thick] (-3.5, 0.6) -- (3.5, 0.6);
\draw[thick] (-3.5, -0.6) -- (3.5, -0.6);
\draw[thick] (-3.5, -3.5) -- (3.5, -3.5);
\node[right, fill=white, inner sep=1.5pt, font=\small] at (0.1, 3.5) {$\mathscr{D}_2$};
\node[right, fill=white, inner sep=1.5pt, font=\small] at (0.1, 0.6) {$r$};
\node[right, fill=white, inner sep=1.5pt, font=\small] at (0.1, -0.6) {$-r$};
\node[right, fill=white, inner sep=1.5pt, font=\small] at (0.1, -3.5) {$-\mathscr{D}_2$};
\node[fill=white, inner sep=2pt, font=\small] at (2, 2.05) {$\operatorname{supp}(\phi)$};
\node[fill=white, inner sep=2pt, font=\small] at (2, -2.05) {$\operatorname{supp}(\phi)$};
\draw[dotted, thick, black] (-3.5, 3.5) -- (-5.2, 3.5);
\draw[dotted, thick, black] (-3.5, 0.6) -- (-4.2, 0.6);
\draw[dotted, thick, black] (0, 0) -- (-5.2, 0);
\draw[dotted, thick, black] (-3.5, -0.6) -- (-4.2, -0.6);
\draw[dotted, thick, black] (-3.5, -3.5) -- (-5.2, -3.5);
\draw[<->, thick, black] (-4.9, 0) -- (-4.9, 3.5) node[midway, fill=white, inner sep=1pt] {$\mathscr{D}_2$};
\draw[<->, thick, black] (-4.9, 0) -- (-4.9, -3.5) node[midway, fill=white, inner sep=1pt] {$\mathscr{D}_2$};
\draw[<->, thick, black] (-3.9, 0) -- (-3.9, 0.6) node[midway, left] {$r$};
\draw[<->, thick, black] (-3.9, 0) -- (-3.9, -0.6) node[midway, left] {$r$};
\end{scope}
\end{tikzpicture}
}
\end{minipage}\hfill
\begin{minipage}{0.54\linewidth}
\resizebox{\textwidth}{!}{%
\begin{tikzpicture}[
>=Latex,
agent/.style={circle, fill=black, inner sep=1.5pt}, %
hatch/.style={pattern=north east lines, pattern color=black!50},
note/.style={align=left, font=\small}
]
\begin{scope}[xshift=12.2cm]
\node[font=\bfseries] at (0, 4.5) {Communication Domain of Agent $i$};
\draw[->, thick] (-6.7, -0.4) -- (4.6, -0.4) node[right] {$x$};
\draw[->, thick] (0, -3.3) -- (0, 4.2) node[left] {$y$};
\fill[hatch] (-3.5, 1.0) rectangle (4.2, 3.9);
\fill[hatch] (-3.5, -3.1) rectangle (4.2, -0.2);
\draw[thick] (-3.5, 3.9) -- (4.2, 3.9);
\draw[thick] (-3.5, 1.0) -- (4.2, 1.0);
\draw[thick] (-3.5, -0.2) -- (4.2, -0.2);
\draw[thick] (-3.5, -3.1) -- (4.2, -3.1);
\foreach \y in {-2.0, -1.3, -0.6, 0.0, 0.4, 0.8, 1.5} {
\draw[gray!30, thin] (-3.5, \y) -- (4.2, \y);
}
\node[agent] (a1) at (-1.0, -2.0) {}; \draw[->, thick] (a1) -- ++(0.9, 0);
\node[agent] (a2) at (0.2, -2.0) {}; \draw[->, thick] (a2) -- ++(0.3, 0);
\node[agent] (a3) at (1.5, -2.0) {}; \draw[->, thick] (a3) -- ++(0.7, 0);
\node[agent] (b1) at (-1.8, -1.3) {}; \draw[->, thick] (b1) -- ++(0.4, 0);
\node[agent] (b2) at (-0.5, -1.3) {}; \draw[->, thick] (b2) -- ++(0.8, 0);
\node[agent] (b3) at (1.0, -1.3) {}; \draw[->, thick] (b3) -- ++(0.2, 0);
\node[agent] (b4) at (2.2, -1.3) {}; \draw[->, thick] (b4) -- ++(1.0, 0);
\node[agent] (c1) at (-2.2, -0.6) {}; \draw[->, thick] (c1) -- ++(0.6, 0);
\node[agent] (c2) at (-0.8, -0.6) {}; \draw[->, thick] (c2) -- ++(0.3, 0);
\node[agent] (c3) at (0.5, -0.6) {}; \draw[->, thick] (c3) -- ++(1.1, 0);
\node[agent] (c4) at (2.0, -0.6) {}; \draw[->, thick] (c4) -- ++(0.5, 0);
\node[agent] (d1) at (-2.0, 0.0) {}; \draw[->, thick] (d1) -- ++(0.8, 0);
\node[agent] (d2) at (-0.2, 0.0) {}; \draw[->, thick] (d2) -- ++(0.4, 0);
\node[agent] (d3) at (1.2, 0.0) {}; \draw[->, thick] (d3) -- ++(0.9, 0);
\node[agent] (d4) at (2.8, 0.0) {}; \draw[->, thick] (d4) -- ++(0.2, 0);
\node[agent] (e1) at (-1.5, 0.4) {}; \draw[->, thick] (e1) -- ++(0.3, 0);
\node[agent] (ci) at (0.8, 0.4) {}; \draw[->, thick] (ci) -- ++(0.7, 0);
\node[agent] (e3) at (3.2, 0.4) {}; \draw[->, thick] (e3) -- ++(0.5, 0);
\draw[thick] (ci) circle (3.5pt);
\node[above right=3pt, font=\small, fill=white, inner sep=1pt] at (ci) {agent $i$};
\node[agent] (f1) at (-0.5, 0.8) {}; \draw[->, thick] (f1) -- ++(0.9, 0);
\node[agent] (f2) at (2.4, 0.8) {}; \draw[->, thick] (f2) -- ++(0.4, 0);
\node[agent] (f3) at (3.5, 0.8) {}; \draw[->, thick] (f3) -- ++(0.6, 0);
\node[agent] (g1) at (0.5, 1.5) {}; \draw[->, thick] (g1) -- ++(0.5, 0);
\node[agent] (g2) at (2.0, 1.5) {}; \draw[->, thick] (g2) -- ++(0.8, 0);
\draw[dotted, thick, black] (-3.5, 3.9) -- (-6.5, 3.9);
\draw[dotted, thick, black] (-3.5, -3.1) -- (-6.5, -3.1);
\draw[<->, thick, black] (-6.2, 0.4) -- (-6.2, 3.9) node[midway, fill=white, inner sep=1pt] {$\mathscr{D}_2$};
\draw[<->, thick, black] (-6.2, 0.4) -- (-6.2, -3.1) node[midway, fill=white, inner sep=1pt] {$\mathscr{D}_2$};
\draw[dotted, thick, black] (-6.5, 0.4) -- (0.8, 0.4);
\draw[dotted, thick, black] (-5.5, 1.5) -- (-3.5, 1.5);
\draw[dotted, thick, black] (-5.5, -2.0) -- (-3.5, -2.0);
\draw[<->, thick, black] (-5.2, -2.0) -- (-5.2, 1.5) node[midway, fill=white, inner sep=1pt] {$\mathscr{D}_2$};
\draw[dotted, thick, black] (-3.5, 1.0) -- (-4.5, 1.0);
\draw[dotted, thick, black] (-3.5, -0.2) -- (-4.5, -0.2);
\draw[<->, thick, black] (-4.2, 0.4) -- (-4.2, 1.0) node[midway, left] {$r$};
\draw[<->, thick, black] (-4.2, 0.4) -- (-4.2, -0.2) node[midway, left] {$r$};
\end{scope}
\end{tikzpicture}%
}
\end{minipage}
\caption{A schematic depiction of the flocking paradigm implicated by Theorem \ref{t:flockingCS}. Each agent may have an associated cylinder of degeneracy around it, inside which it does not communicate with other agents. Flocking will still occur in this situation if the communication is strong enough outside this cylinder, as long as no two such cylinders contain all the agents. }
\label{fig:degenerate-slice-communication}
\end{figure}
\vspace{-3 mm}
\subsection{Flocking for the unidirectional Euler Alignment system}
\begin{THEOREM}
\label{t:flockingEA}
Assume $\phi$ satisfies \eqref{e:phieven} and \eqref{e:phiLip2}. Let $\rho^0$ be a compactly supported probability measure on $\mathbb{R}^d$, i.e., $\rho^0\in \cP_c(\mathbb{R}^d)$. Assume $u^0\in L^\infty(\rho^0)$. Let $(\rho, u)$ be the associated solution of \eqref{e:EAuni} generated by the entropic selection principle. Let $\rho(\dx, \dy, t) = \rho_y(\dx,t) \nu(\dy)$ be the usual disintegration with respect to the projection $(x,y)\mapsto y$. Define the quantities
\[
x_+(t) = \max_{y\in \supp\nu} \max \supp \rho_y(t),
\qquad
x_-(t) = \min_{y\in \supp\nu} \min \supp \rho_y(t),
\]
\[
u_+(t) = \inf\{v:\rho(t)(\{(x,y): u(x,y,t)>v\}) = 0\},
\qquad
u_-(t) = \sup\{v:\rho(t)(\{(x,y): u(x,y,t)<v\}) = 0\},
\]
\[
\mathscr{D}_1(t) = x_+(t) - x_-(t),
\qquad
\mathscr{D}_2 = \diam\supp \nu,
\qquad
\mathscr{A}(t) = u_+(t) - u_-(t).
\]
Define $\mathscr{D}_1^0$ and $\mathscr{A}^0$ analogously. Assume there exists $r>0$ with the following properties:
\begin{itemize}
\item The proportion of the total mass contained in any two balls of radius $r$ in $\mathbb{R}^{d-1}$ is bounded away from $1$:
\begin{equation}
\label{e:flockingreq1cont}
\nu(B(y,r)\cup B(z,r)) < 1-c,
\qquad \text{ for all } y,z\in \mathbb{R}^{d-1}
\end{equation}
\item There exists a continuous function $\underline{\phi}:[0,\infty)\times [0,\infty)\to \mathbb{R}$, nonincreasing in both its arguments on $[0,\infty) \times [r,\mathscr{D}_2]$, such that $\phi(x,y)\ge \underline{\phi}(|x|,|y|)$, and $\underline{\phi}$ additionally satisfies (with the same constant $c$ as in \eqref{e:flockingreq1cont}):
\begin{equation}
\label{e:heavytailsubcont}
c\int_{\mathscr{D}^0_1}^\infty \underline{\phi}(x,\mathscr{D}_2)\,\dx > \mathscr{A}^0,
\end{equation}
\end{itemize}
Then the solution experiences flocking and exponentially fast alignment: There exists $\overline{\mathscr{D}}_1>0$ such that
\begin{equation}
\label{e:uniflockingcont}
\sup_{t\ge 0} \mathscr{D}_1(t)\le \overline{\mathscr{D}}_1,
\qquad
\mathscr{A}(t)
\le \mathscr{A}^0 e^{-ct \underline{\phi}(\overline{\mathscr{D}}_1,\mathscr{D}_2)}.
\end{equation}
Finally, denoting the average velocity of the system by $\overline{u} = \int u^0 \rho^0$, there exists $\rho_\infty\in \cP_c(\mathbb{R}^d)$ such that
\begin{equation}
\label{e:strongflocking}
W_1(\rho(\cdot + \overline{u}t, \cdot,t), \rho_\infty) \to 0,
\qquad \text{ as } t\to +\infty.
\end{equation}
\end{THEOREM}
We omit the proof of this theorem, which follows from Theorem \ref{t:flockingCS} using exactly the same procedures as was used in \cite{LeslieTan2023} to extend from the discrete setting to the continuous one. We approximate the solution $(\rho, u)$ by SPCS dynamics and then use Theorem \ref{t:flockingCS} to control the latter. Note that collisions cannot inhibit flocking, but rather can only enhance it (as noted in \cite{LeslieTan2023}, we have $\mathscr{A}(t)\le \mathscr{A}(t-)$ at collision times). Therefore, Theorem \ref{t:flockingCS} applies equally well to the classical and sticky particle Cucker--Smale dynamics.
%\bibliographystyle{plain}
%\bibliography{unirefs}
\def\cprime{$'$}
\begin{thebibliography}{10}
\bibitem{AmbrosioDalMaso1990}
Luigi Ambrosio and Gianni Dal~Maso.
\newblock A general chain rule for distributional derivatives.
\newblock {\em Proc. Amer. Math. Soc.}, 108(3):691--702, 1990.
\bibitem{ArnaizCastro2019}
Victor Arnaiz and \'{A}ngel Castro.
\newblock Singularity formation for the fractional {E}uler-alignment system in
1{D}.
\newblock {\em Trans. Amer. Math. Soc.}, 374(1):487--514, 2021.
\bibitem{BianchiniDaneri2023}
Stefano Bianchini and Sara Daneri.
\newblock On the sticky particle solutions to the multi-dimensional
pressureless {E}uler equations.
\newblock {\em J. Differential Equations}, 368:173--202, 2023.
\bibitem{BouchutJames1998}
Fran\c{c}ois Bouchut and Fran\c{c}ois James.
\newblock One-dimensional transport equations with discontinuous coefficients.
\newblock {\em Nonlinear Anal.}, 32(7):891--933, 1998.
\bibitem{BouchutJames1999}
Fran\c{c}ois Bouchut and Fran\c{c}ois James.
\newblock Duality solutions for pressureless gases, monotone scalar
conservation laws, and uniqueness.
\newblock {\em Comm. Partial Differential Equations}, 24(11-12):2173--2189,
1999.
\bibitem{BouchutPerthame1998}
Fran\c{c}ois Bouchut and Benoit Perthame.
\newblock Kru\v{z}kov's estimates for scalar conservation laws revisited.
\newblock {\em Trans. Amer. Math. Soc.}, 350(7):2847--2870, 1998.
\bibitem{BrenierGangboSavareWestdickenberg2013}
Yann Brenier, Wilfrid Gangbo, Giuseppe Savar\'{e}, and Michael Westdickenberg.
\newblock Sticky particle dynamics with interactions.
\newblock {\em J. Math. Pures Appl.}, 99(5):577--617, 2013.
\bibitem{brenier1998sticky}
Yann Brenier and Emmanuel Grenier.
\newblock Sticky particles and scalar conservation laws.
\newblock {\em SIAM J. Numer. Anal}, 35(6):2317--2328, 1998.
\bibitem{CCP2017}
Jos\'{e}~A. Carrillo, Young-Pil Choi, and Sergio~P. Perez.
\newblock A review on attractive-repulsive hydrodynamics for consensus in
collective behavior.
\newblock In {\em Active particles. {V}ol. 1. {A}dvances in theory, models, and
applications}, Model. Simul. Sci. Eng. Technol., pages 259--298.
Birkh\"auser/Springer, Cham, 2017.
\bibitem{CCTT2016}
Jos\'{e}~A. Carrillo, Young-Pil Choi, Eitan Tadmor, and Changhui Tan.
\newblock Critical thresholds in 1{D} {E}uler equations with non-local forces.
\newblock {\em Math. Models Methods Appl. Sci.}, 26(1):185--206, 2016.
\bibitem{carrillogaltung2025}
Jos{\'e}~A Carrillo and Sondre Galtung.
\newblock Equivalence of entropy solutions and gradient flows for pressureless
{1D} {E}uler systems.
\newblock {\em arXiv preprint arXiv:2312.04932}, 2023.
\bibitem{CarrilloChoiTadmor2026}
José~A. Carrillo, Young-Pil Choi, and Eitan Tadmor.
\newblock Lagrangian formulation and eulerian closure in alignment dynamics.
\newblock {\em arXiv preprint arXiv:2604.10253}, 2026.
\bibitem{CavallettiSedjroWestdickenberg2015}
Fabio Cavalletti, Marc Sedjro, and Michael Westdickenberg.
\newblock A simple proof of global existence for the 1{D} pressureless gas
dynamics equations.
\newblock {\em SIAM J. Math. Anal.}, 47(1):66--79, 2015.
\bibitem{CS2007a}
Felipe Cucker and Steve Smale.
\newblock Emergent behavior in flocks.
\newblock {\em IEEE Trans. Automat. Control}, 52(5):852--862, 2007.
\bibitem{CS2007b}
Felipe Cucker and Steve Smale.
\newblock On the mathematics of emergence.
\newblock {\em Jpn. J. Math.}, 2(1):197--227, 2007.
\bibitem{DanchinMuchaPeszekWroblewski2018}
Rapha\"{e}l Danchin, Piotr~B. Mucha, Jan Peszek, and Bartosz Wr\'{o}blewski.
\newblock Regular solutions to the fractional {E}uler alignment system in the
{B}esov spaces framework.
\newblock {\em Math. Models Methods Appl. Sci.}, 29(1):89--119, 2019.
\bibitem{DietertShvydkoy2019}
Helge Dietert and Roman Shvydkoy.
\newblock On {C}ucker-{S}male dynamical systems with degenerate communication.
\newblock {\em Anal. Appl. (Singap.)}, 19(4):551--573, 2021.
\bibitem{DKRT}
Tam Do, Alexander Kiselev, Lenya Ryzhik, and Changhui Tan.
\newblock Global regularity for the fractional {E}uler {A}lignment system.
\newblock {\em Arch. Ration. Mech. Anal.}, 228(1):1--37, 2018.
\bibitem{ERykovSinai1996}
Weinan E, Yu~G. Rykov, and Ya~G. Sinai.
\newblock Generalized variational principles, global weak solutions and
behavior with random initial data for systems of conservation laws arising in
adhesion particle dynamics.
\newblock {\em Comm. Math. Phys.}, 177(2):349--380, 1996.
\bibitem{FigalliKang2019}
Alessio Figalli and Moon-Jin Kang.
\newblock A rigorous derivation from the kinetic {C}ucker-{S}male model to the
pressureless {E}uler system with nonlocal alignment.
\newblock {\em Anal. PDE}, 12(3):843--866, 2019.
\bibitem{Galtung2025}
Sondre~Tesdal Galtung.
\newblock The sticky particle dynamics of the 1{D} pressureless
{E}uler-alignment system as a gradient flow.
\newblock {\em Appl. Math. Optim.}, 91(2):Paper No. 27, 49, 2025.
\bibitem{Grenier1995}
Emmanuel Grenier.
\newblock Existence globale pour le syst\`eme des gaz sans pression.
\newblock {\em C. R. Acad. Sci. Paris S\'{e}r. I Math.}, 321(2):171--174, 1995.
\bibitem{HaHuangWang2014wk}
Seung-Yeal Ha, Feimin Huang, and Yi~Wang.
\newblock A global unique solvability of entropic weak solution to the
one-dimensional pressureless {E}uler system with a flocking dissipation.
\newblock {\em J. Differential Equations}, 257(5):1333--1371, 2014.
\bibitem{HaKimParkZhang2019}
Seung-Yeal Ha, Jeongho Kim, Jinyeong Park, and Xiongtao Zhang.
\newblock Complete cluster predictability of the {C}ucker-{S}male flocking
model on the real line.
\newblock {\em Arch. Ration. Mech. Anal.}, 231(1):319--365, 2019.
\bibitem{HaLiu2008}
Seung-Yeal Ha and Jian-Guo Liu.
\newblock A simple proof of the {C}ucker-{S}male flocking dynamics and
mean-field limit.
\newblock {\em Commun. Math. Sci.}, 7(2):297--325, 2009.
\bibitem{HaParkZhang2018}
Seung-Yeal Ha, Jinyeong Park, and Xiongtao Zhang.
\newblock A first-order reduction of the {C}ucker-{S}male model on the real
line and its clustering dynamics.
\newblock {\em Commun. Math. Sci.}, 16(7):1907--1931, 2018.
\bibitem{HT2008}
Seung-Yeal Ha and Eitan Tadmor.
\newblock From particle to kinetic and hydrodynamic descriptions of flocking.
\newblock {\em Kinet. Relat. Models}, 1(3):415--435, 2008.
\bibitem{HT2016}
Siming He and Eitan Tadmor.
\newblock Global regularity of two-dimensional flocking hydrodynamics.
\newblock {\em C. R. Math. Acad. Sci. Paris}, 355(7):795--805, 2017.
\bibitem{HuangWang2001stickyunique}
Feimin Huang and Zhen Wang.
\newblock Well posedness for pressureless flow.
\newblock {\em Comm. Math. Phys.}, 222(1):117--146, 2001.
\bibitem{Hynd2019LagrangianSP}
Ryan Hynd.
\newblock Lagrangian coordinates for the sticky particle system.
\newblock {\em SIAM J. Math. Anal.}, 51(5):3769--3795, 2019.
\bibitem{Hynd2020probmeasureSP}
Ryan Hynd.
\newblock Probability measures on the path space and the sticky particle
system.
\newblock {\em Annali della Scuola Normale Superiore di Pisa, Classe di
Scienze}, XXI(5):1333--1357, 2020.
\bibitem{Hynd2020trajectory}
Ryan Hynd.
\newblock A trajectory map for the pressureless {E}uler equations.
\newblock {\em Trans. Amer. Math. Soc.}, 373(10):6777--6815, 2020.
\bibitem{KangVasseur2015}
Moon-Jin Kang and Alexis~F. Vasseur.
\newblock Asymptotic analysis of {V}lasov-type equations under strong local
alignment regime.
\newblock {\em Math. Models Methods Appl. Sci.}, 25(11):2153--2173, 2015.
\bibitem{KMT2013}
Trygve~K. Karper, Antoine Mellet, and Konstantina Trivisa.
\newblock Existence of weak solutions to kinetic flocking models.
\newblock {\em SIAM J. Math. Anal.}, 45(1):215--243, 2013.
\bibitem{KMT2014}
Trygve~K. Karper, Antoine Mellet, and Konstantina Trivisa.
\newblock On strong local alignment in the kinetic {C}ucker-{S}male model.
\newblock In {\em Hyperbolic conservation laws and related analysis with
applications}, volume~49 of {\em Springer Proc. Math. Stat.}, pages 227--242.
Springer, Heidelberg, 2014.
\bibitem{KMT2015}
Trygve~K. Karper, Antoine Mellet, and Konstantina Trivisa.
\newblock Hydrodynamic limit of the kinetic {C}ucker-{S}male flocking model.
\newblock {\em Math. Models Methods Appl. Sci.}, 25(1):131--163, 2015.
\bibitem{KT}
Alexander Kiselev and Changhui Tan.
\newblock Global regularity for 1{D} {E}ulerian dynamics with singular
interaction forces.
\newblock {\em SIAM J. Math. Anal.}, 50(6):6208--6229, 2018.
\bibitem{Kruzkov1970}
S.~N. Kru\v{z}kov.
\newblock First order quasilinear equations with several independent variables.
\newblock {\em Mat. Sb. (N.S.)}, 81(123):228--255, 1970.
\bibitem{Lear2021unising}
Daniel Lear.
\newblock Global existence and limiting behavior of unidirectional flocks for
the fractional {E}uler alignment system.
\newblock {\em SIAM J. Math. Anal.}, 55(4):3731--3754, 2023.
\bibitem{LLST2020geometric}
Daniel Lear, Trevor~M. Leslie, Roman Shvydkoy, and Eitan Tadmor.
\newblock Geometric structure of mass concentration sets for pressureless
{E}uler alignment systems.
\newblock {\em Adv. Math.}, 401:Paper No. 108290, 30, 2022.
\bibitem{LearShvydkoy2021}
Daniel Lear and Roman Shvydkoy.
\newblock Unidirectional flocks in hydrodynamic {E}uler alignment system {II}:
{S}ingular models.
\newblock {\em Commun. Math. Sci.}, 19(3):807--828, 2021.
\bibitem{LearShvydkoy2019}
Daniel Lear and Roman Shvydkoy.
\newblock Existence and stability of unidirectional flocks in hydrodynamic
{E}uler alignment systems.
\newblock {\em Anal. PDE}, 15(1):175--196, 2022.
\bibitem{Leslie2019}
Trevor~M. Leslie.
\newblock Weak and strong solutions to the forced fractional {E}uler alignment
system.
\newblock {\em Nonlinearity}, 32(1):46--87, 2019.
\bibitem{L2019CTC}
Trevor~M. Leslie.
\newblock On the {L}agrangian trajectories for the one-dimensional {E}uler
alignment model without vacuum velocity.
\newblock {\em Comptes Rendus. Math\'ematique}, 358(4):421--433, 2020.
\bibitem{LS2019}
Trevor~M. Leslie and Roman Shvydkoy.
\newblock On the structure of limiting flocks in hydrodynamic {E}uler
{A}lignment models.
\newblock {\em Math. Models Methods Appl. Sci.}, 29(13):2419--2431, 2019.
\bibitem{LeslieTan2023}
Trevor~M. Leslie and Changhui Tan.
\newblock Sticky particle cucker–smale dynamics and the entropic selection
principle for the 1d euler-alignment system.
\newblock {\em Communications in Partial Differential Equations},
48(5):753--791, 2023.
\bibitem{LeslieTan2024}
Trevor~M. Leslie and Changhui Tan.
\newblock Finite- and infinite-time cluster formation for alignment dynamics on
the real line.
\newblock {\em J. Evol. Equ.}, 24(1):Paper No. 8, 45, 2024.
\bibitem{LiMiaoTanXue2024}
Yatao Li, Qianyun Miao, Changhui Tan, and Liutang Xue.
\newblock Global well-posedness and refined regularity criterion for the
uni-directional {E}uler-alignment system.
\newblock {\em Int. Math. Res. Not. IMRN}, (23):14393--14422, 2024.
\bibitem{Lucier1986}
Bradley~J. Lucier.
\newblock A moving mesh numerical method for hyperbolic conservation laws.
\newblock {\em Math. Comp.}, 46(173):59--69, 1986.
\bibitem{MPT2019}
Javier Morales, Jan Peszek, and Eitan Tadmor.
\newblock Flocking {W}ith {S}hort-{R}ange {I}nteractions.
\newblock {\em J. Stat. Phys.}, 176(2):382--397, 2019.
\bibitem{MT2014}
Sebastien Motsch and Eitan Tadmor.
\newblock Heterophilious dynamics enhances consensus.
\newblock {\em SIAM Rev.}, 56(4):577--621, 2014.
\bibitem{NatileSavare2009}
Luca Natile and Giuseppe Savar\'{e}.
\newblock A {W}asserstein approach to the one-dimensional sticky particle
system.
\newblock {\em SIAM J. Math. Anal.}, 41(4):1340--1365, 2009.
\bibitem{nguyen2008pressureless}
Truyen Nguyen and Adrian Tudorascu.
\newblock Pressureless {E}uler/{E}uler--{P}oisson systems via adhesion dynamics
and scalar conservation laws.
\newblock {\em SIAM J. Math. Anal.}, 40(2):754--775, 2008.
\bibitem{NguyenTudorascu2015}
Truyen Nguyen and Adrian Tudorascu.
\newblock One-dimensional pressureless gas systems with/without viscosity.
\newblock {\em Comm. Partial Differential Equations}, 40(9):1619--1665, 2015.
\bibitem{Shvydkoy2018NearlyAligned}
Roman Shvydkoy.
\newblock Global existence and stability of nearly aligned flocks.
\newblock {\em Journal of Dynamics and Differential Equations}, Aug 2018.
\bibitem{ShvydkoyBook}
Roman Shvydkoy.
\newblock {\em {D}ynamics and {A}nalysis of {A}lignment {M}odels of
{C}ollective {B}ehavior}, volume~4 of {\em Ne\v{c}as Center Series}.
\newblock Birkh\"{a}user Basel, 2021.
\bibitem{ShvydkoySurvey}
Roman Shvydkoy.
\newblock Environmental averaging.
\newblock {\em EMS Surv. Math. Sci.}, 11(2):277--413, 2024.
\bibitem{ShvydkoyTadmorI}
Roman Shvydkoy and Eitan Tadmor.
\newblock Eulerian dynamics with a commutator forcing.
\newblock {\em Transactions of Mathematics and Its Applications}, 1(1), 2017.
\bibitem{ShvydkoyTadmorII}
Roman {Shvydkoy} and Eitan {Tadmor}.
\newblock Eulerian dynamics with a commutator forcing {II}: {F}locking.
\newblock {\em Discrete Contin. Dyn. Syst.}, 37(11):5503--5520, 2017.
\bibitem{ShvydkoyTadmorIII}
Roman Shvydkoy and Eitan Tadmor.
\newblock Eulerian dynamics with a commutator forcing {III}. {F}ractional
diffusion of order {$0<\alpha<1$}.
\newblock {\em Phys. D}, 376/377:131--137, 2018.
\bibitem{STtopo}
Roman Shvydkoy and Eitan Tadmor.
\newblock Topologically based fractional diffusion and emergent dynamics with
short-range interactions.
\newblock {\em SIAM Journal on Mathematical Analysis}, 52(6):5792--5839, 2020.
\bibitem{Tadmor2021review}
Eitan Tadmor.
\newblock On the mathematics of swarming: Emergent behavior in alignment
dynamics.
\newblock {\em Notices of the AMS}, 68(4):493--503, 2021.
\bibitem{Tadmor2026}
Eitan Tadmor.
\newblock Decrease of entropy and emergence of order in collective dynamics.
\newblock {\em Commun. Contemp. Math.}, 28(5):Paper No. 2540006, 2026.
\bibitem{TT2014}
Eitan Tadmor and Changhui Tan.
\newblock Critical thresholds in flocking hydrodynamics with non-local
alignment.
\newblock {\em Philos. Trans. R. Soc. Lond. Ser. A Math. Phys. Eng. Sci.},
372:20130401, 2014.
\bibitem{tan2020euler}
Changhui Tan.
\newblock On the {E}uler-alignment system with weakly singular communication
weights.
\newblock {\em Nonlinearity}, 33(4):1907--1924, 2020.
\bibitem{tan2021eulerian}
Changhui Tan.
\newblock {E}ulerian dynamics in multi-dimensions with radial symmetry.
\newblock {\em SIAM J. Math. Anal.}, 53(3):3040--3071, 2021.
\bibitem{WangDing1997}
Zhen Wang and Xiaqi Ding.
\newblock Uniqueness of generalized solution for the {C}auchy problem of
transportation equations.
\newblock {\em Acta Math. Sci. (English Ed.)}, 17(3):341--352, 1997.
\bibitem{WangHuangDing1997}
Zhen Wang, Feimin Huang, and Xiaqi Ding.
\newblock On the {C}auchy problem of transportation equations.
\newblock {\em Acta Math. Appl. Sinica (English Ser.)}, 13(2):113--122, 1997.
\bibitem{Zeldovich1969sb}
Ya.~B. Zeldovich.
\newblock Gravitational instability: An approximate theory for large density
perturbations.
\newblock {\em Astron. Astrophys.}, 5:84--89, 1970.
\end{thebibliography}
\end{document}