\documentclass[reqno]{amsart}
\usepackage{hyperref}
\usepackage{amssymb}

\AtBeginDocument{{\noindent\small
\emph{Electronic Journal of Differential Equations},
Vol. 2016 (2016), No. 325, pp. 1--18.\newline
ISSN: 1072-6691. URL: http://ejde.math.txstate.edu or http://ejde.math.unt.edu}
\thanks{\copyright 2016 Texas State University.}
\vspace{8mm}}

\begin{document}
\title[\hfilneg EJDE-2016/325\hfil Conservation laws with discontinuous flux]
{Existence of solutions for a scalar conservation law with a flux of low regularity}

\author[M. Lazar, D. Mitrovi\'c \hfil EJDE-2016/325\hfilneg]
{Martin Lazar, Darko Mitrovi\'c}

\address{Martin Lazar\newline
University of Dubrovnik, Croatia}
\email{mlazar@unidu.hr}

\address{Darko Mitrovi\'c \newline
Faculty of Mathematics, University of
Montenegro, Montenegro}
\email{darkom@ac.me}

\thanks{Submitted May 18, 2016. Published December 22, 2016.}
\subjclass[2010]{35L65}
\keywords{Scalar conservation law; discontinuous coefficient; existence; 
\hfill\break\indent velocity averaging; H-measure}

\begin{abstract}
 We prove existence of solutions to Cauchy problem for scalar conservation
 laws with non-degenerate discontinuous flux
 $$
 \partial_t u+ \operatorname{div}f(t,\mathbf{x},u)=s(t,\mathbf{x},u), \quad
 t\geq 0, \mathbf{x}\in \mathbb{R}^d,
 $$
 where for every $(t,\mathbf{x})\in \mathbb{R}^+\times \mathbb{R}$,
 the flux $f(t,\mathbf{x},\cdot) \in \mathrm{Lip}(\mathbb{R};\mathbb{R}^d)$ and
 $\partial_\lambda f \in L^r(\mathbb{R}^+\times \mathbb{R}^d\times \mathbb{R})$,
 additionally satisfying $\max_{|\lambda| \leq M} f(\cdot,\cdot,\lambda)
 \in L^r(\mathbb{R}^+\times \mathbb{R}^d)$, for some $r>1$ and every $M>0$,
 and, for every $\lambda \in \mathbb{R}$,
 $\operatorname{div}_{(t,\mathbf{x})} f(\cdot,\cdot,\lambda) \in
 \mathcal{M}(\mathbb{R}^+\times \mathbb{R}^d)$ where
 $\mathcal{M}(\mathbb{R}^+\times \mathbb{R}^d)$ is the space of Radon measures.
 Moreover, the function $s$ is measurable and both $f$ and $s$ satisfy certain
 growth rate assumptions with respect to $\lambda$.  The result is obtained
 by means of the H-measures.
\end{abstract}

\maketitle
\numberwithin{equation}{section}
\newtheorem{theorem}{Theorem}[section]
\newtheorem{lemma}[theorem]{Lemma}
\newtheorem{remark}[theorem]{Remark}
\newtheorem{definition}[theorem]{Definition}
\newtheorem{corollary}[theorem]{Corollary}
\allowdisplaybreaks

%\newcommand{\norm[1]}{\|#1\|}

\section{Introduction}

In this article, we consider the Cauchy problem for the non-linear transport 
equation
\begin{gather}\label{d-p}
  \partial_t u+\operatorname{div}_{\mathbf{x}} f(t,\mathbf{x},u)
 =s(t,\mathbf{x},u)\\
\label{ic}
u|_{t=0}=u_0(\mathbf{x}) \in L^1(\mathbb{R}^d)\cap L^p(\mathbb{R}^d), \quad p>1.
\end{gather} 
The given equation describes many natural phenomena.  
We mention some of them: flow in porous media, sedimentation processes, 
traffic flow, radar shape-from-shading problems, blood flow, etc.

While the local Lipschitz assumption on $\lambda\mapsto f(t,\mathbf{x},\lambda)$ is
natural in many applications, the assumption of regular dependence
on the spatial variable $\mathbf{x}$ is very restrictive. Indeed, even in
the simplified situation in which the diffusion is neglected such as
road traffic with variable number of lanes
\cite{BKT2009-multilanes}, Buckley-Leverett equation in a layered
porous medium \cite{AndrCances, Kaas}, sedimentation processes
\cite{BurgerEtAl, Die1, Die2}, etc. It may appear in models with discontinuous 
(in $\mathbf{x}$) flux functions.

Scalar conservation laws with discontinuous flux attracted significant 
amount of attention in recent years. It appears that neither existence 
nor uniqueness for \eqref{d-p}, \eqref{ic} can be resolved applying the 
methods used  when the flux is regular. More precisely, the well-posedness 
of the Cauchy problem for \eqref{d-p} with a regular flux is completely settled 
in \cite{Kru} by means of the vanishing viscosity method and the shifting 
of variables (existence) and the doubling of variables (uniqueness). 
The mentioned techniques cannot be extended on \eqref{d-p} without substantial 
improvements. This fact generated development of various methods and concepts 
during attempts to solve and understand equation \eqref{d-p}. 
Different uniqueness concepts are thoroughly described in \cite{AKR} and 
they are further developed in \cite{AnM, crasta1, crasta2}. 
As for the existence of solution to \eqref{d-p}, \eqref{ic}, which is the 
main objective of this article, the first general result is obtained in 
one-dimensional case using the compensated compactness method \cite{KRT}. 
This result is later extended on two-dimensional case  \cite{AlM, KT} 
under a non-degeneracy conditions (see a similar requirement \eqref{non-deg} below).
 The first multidimensional existence result is obtained in \cite{pan_arma} 
through an extension of  H-measures \cite{Ger, Tar}.

The strategy of the proof based on H-measures requires rewriting \eqref{d-p} 
in the kinetic formulation \cite{D, LPT} (see Theorem \ref{kf}) and then 
applying the velocity averaging results (see Theorem \ref{va}).
 We remark that Theorem \ref{va} is the most general velocity averaging result 
from the viewpoint of regularity of coefficients of the underlying transport 
equation. We dedicated Section 3 to this issue where we also provided a short 
overview of the related results.

The aim of this article is to generalize results from \cite{pan_arma} 
when the flux $f$ and the source $s$ satisfy:
\begin{itemize}
\item[(i)] For each fixed $(t,\mathbf{x})\in \mathbb{R}^+\times \mathbb{R}^d$ 
the flux $f(t,\mathbf{x},\cdot) \in \mathrm{Lip}(\mathbb{R};\mathbb{R}^d)$ and
$\partial_\lambda f \in L^r(\mathbb{R}^+\times \mathbb{R}^d \times \mathbb{R})$, 
$r>1$;

\item[(ii)]  For every $M>0$, it holds 
$\sup_{|\lambda|<M} |f(t,\mathbf{x},\lambda)|\in L^r(\mathbb{R}^+\times\mathbb{R}^d)$,
 $r>1$.

\item[(iii)] For every $\lambda \in \mathbb{R}$, the divergence of the flux is a 
Radon measure,
\[ 
\operatorname{div}_{(t,\mathbf{x})} f(\cdot,\cdot,\lambda)\in 
\mathcal{M}(\mathbb{R}^+\times \mathbb{R}^d);
\]

\item[(iv)] There exists a non-negative convex function $S:\mathbb{R}\to \mathbb{R}$ 
such that $S(0)=0$ and there exists $M, C, c>0$ such that for every $|\lambda|<M$,
\begin{gather}\label{2}
c\leq \frac{S(\lambda)}{|\lambda|^p} \leq C,\\
\label{1}
\int_{\mathbb{R}^+\times \mathbb{R}^{d}}
\sup_{\lambda \in \mathbb{R}}|S''(\lambda) \operatorname{div}_{(t,\mathbf{x})} 
f(t,\mathbf{x},\lambda)| d\lambda dt d\mathbf{x}<C_1<\infty
\\
\label{3}
\max \Big\{  \int_{\mathbb{R}^+\times \mathbb{R}^d} 
\sup_{\lambda \in \mathbb{R}} |s(t,\mathbf{x},\lambda)| d\lambda, 
\int_{\mathbb{R}^+\times \mathbb{R}^d}
 \sup_{\lambda \in \mathbb{R}} |s(t,\mathbf{x},\lambda) S'(\lambda)| d\lambda \Big\} 
<C_2<\infty.
\end{gather}
\end{itemize} 
In \cite{pan_arma}, it was assumed that the flux $f(t,\mathbf{x},\lambda)$ 
satisfies (ii), and (iii), but for $r>2$.  The reason for such an assumption 
is in the tools used -- the H-measures. Original H-measures were adapted 
for continuous coefficients (see \cite{Ger, Tar}). 
In \cite{pan_arma}, the notion of  H-measures is generalized for application 
in the framework of $L^q$, $q>2$, coefficients. After that, we made 
generalizations in this direction \cite{LM2} but we were not able to decrease 
regularity of the coefficients under $L^2$.

To achieve such a low regularity, one can apply  H-distributions which were 
introduced in \cite{AM} and further generalized in \cite{LM_crass, MM}.  
H-distributions are an extension of  H-measures to the $L^p$, $p>1$, setting. 
Another approach, followed in this paper, relies on an appropriate 
generalization of H-measures which takes into account additional regularity 
of the defining sequence that compensates for a low regularity of test 
functions (coefficients).

In the last section, we shall introduce the notion of the quasi-solution
 and prove strong precompactness of the sequence of quasi-solutions to 
\eqref{d-p} under non-degeneracy conditions \eqref{non-deg}. 
As a corollary of the precompactness, we prove existence of weak solutions 
to \eqref{d-p}.

As for the assumption (i), in \cite{pan_arma} it was assumed that for each 
fixed $(t,\mathbf{x})\in \mathbb{R}^+\times \mathbb{R}^d$ the flux 
$f(t,\mathbf{x},\cdot) \in C^1(\mathbb{R};\mathbb{R}^d)$, but there was no
assumption $\partial_\lambda f \in L^r(\mathbb{R}^+\times \mathbb{R}^d 
\times \mathbb{R})$, $r>1$.

Let us also remark that a usual assumption for the scalar conservation 
laws with discontinuous flux (in particular, such an assumption is used 
in \cite{pan_arma} also) is that there exists $a,b \in \mathbb{R}$ such that
$$
f(t,\mathbf{x},a)=f(t,\mathbf{x},b)=0 \quad\text{and}\quad a \leq u_0 \leq b.
$$ 
This provides boundedness of the sequence of approximate solutions obtained 
with the vanishing viscosity perturbation of \eqref{d-p}. 
We replace such assumptions by iv) given above which provides only 
$L^q$-control of the approximate solutions. Details are given in the last section.

Let us shortly remind on the contents of the paper. 
Section 2 provides results on H-measures that we are going to use. 
In Section 3 we prove a velocity averaging result which will be in 
the essence of the existence proof given in Section 4.

\section{H-measures}

H-measures were introduced by  Tartar \cite{Tar} and,
independently, by  Gerard \cite{Ger} as an object which measures
defect of strong  $L^2$-precompactness for bounded sequences in 
$L^2(\mathbb{R}^d)$. More precisely,  a sequence converging weakly
in $L_{\rm loc}^2({\mathbb{R}^d})$ converges strongly as well, if and only if the corresponding 
H-measure is zero.  H-measures appeared to be
very powerful tool in the analysis of PDEs and have been successfully 
applied in many mathematical fields - let us here  mention the generalization 
of compensated compactness results to equations with variable coefficients 
\cite{Ger,Tar}, applications in the control theory \cite{DLR, LZ}, 
the velocity averaging results  \cite{Ger, LM2}, as well as
explicit formulae and bounds in homogenisation  \cite{ALjmaa, ALrwa, Tar}.
\smallskip

\noindent\textbf{Notation.}
Throughout this article, by  $C_0({\mathbb{R}^d})$ we denote the closure of 
$C_{\rm c}({\mathbb{R}^d})$ in $L^\infty({\mathbb{R}^d})$  topology, while $\mathcal{M}_{b}(\Omega)$ 
stands for the space of bounded Radon measures on a set $\Omega\subseteq{\mathbb{R}^d}$.

For a fixed $p\geq 1$, we shall denote by $p'$ its conjugate i.e. 
the number such that ${1}/{p}+{1}/{p'}=1$.

${\rm S}^{d-1}$ stands for the unit sphere in ${\mathbb{R}^d}$ centered 
at the origin.

By $L_{ w^\ast}^2(\mathbb{R}^{2m}; \mathcal{M}_{b}({\mathbb{R}^d}\times {\rm S}^{d-1}))$
 we  denote the dual of $L^2(\mathbb{R}^{2m}; {\rm C}_0({\mathbb{R}^d}\times {\rm S}^{d-1}))$, 
which is a Banach space of weakly $\ast$ measurable functions 
$\mu: \mathbb{R}^{2m} \to \mathcal{M}_{b}({\mathbb{R}^d}\times {\rm S}^{d-1})$  
such that $\int_{\mathbb{R}^{2m}} \|\mu(\mathbf{y}, \tilde{\mathbf{y}})\|^2  
d\mathbf{y} d\tilde{\mathbf{y}} 
<\infty$.

The Fourier transform is defined as
$\hat {\sf u}(\boldsymbol{\xi}):=\int_{\mathbb{R}^d} e^{-2\pi i \boldsymbol{\xi}\cdot \mathbf{x}}
 {\sf u}(\mathbf{x})\,d\mathbf{x}$, and its inverse as 
$({\sf u})^\vee(\boldsymbol{\xi}):=\int_{\mathbb{R}^d} e^{2\pi i \boldsymbol{\xi}\cdot
\mathbf{x}}{\sf u}(\mathbf{x})\,d\mathbf{x}$.

 A (Fourier) multiplier operator $\mathcal{A}_\psi$ on
$\mathbb{R}^d$ with a symbol $\psi$ is defined as
$$
\left(\mathcal{A}_\psi u\right)(x) 
= \left(\psi({\boldsymbol{\xi}}) \hat u(\boldsymbol{\xi})\right)^\vee (x).
$$
In this article,  where there is no fear of ambiguity, we shall not 
distinguish the symbols for a function $\psi$ defined on the unit sphere 
${\rm S}^{d-1}$ and its extension to ${\mathbb{R}^d}$ given by  
$\psi(\boldsymbol{\xi}/|\boldsymbol{\xi})$.

Throughout this paper $\langle \cdot,\cdot\rangle$ stands for a sesquilinear dual product, 
taken to be antilinear in the first, while linear in the second variable. 
By  $\otimes$ we denote the tensor product of functions in different variables.

A sequence $(u_n)$ of functions from $L^1(\Omega)$ is called equi-integrable 
if for every $\varepsilon>0$ there exists $\delta>0$ such that for every set 
$E\subset \Omega$ satisfying $\operatorname{meas}(E)<\delta$ it holds
$\sup_n \int_E |u_n(\mathbf{x})| d\mathbf{x}<\varepsilon$.


Let us now recall the extension of H-measures
introduced in \cite{LM2} whose existence and properties are restated
in the next theorem.

\begin{theorem} \label{lfeb518-r} 
Assume that a sequence $(u_n)$  converges weakly to zero in 
$L^2(\mathbb{R}^{d+m})\cap L^2({\mathbb{R}^m}; L^p({\mathbb{R}^d}))$, $p\geq2$. 
Then, after passing to a subsequence (not relabeled),  
there exists a  measure $\mu \in L_{ w^\ast}^2(\mathbb{R}^{2m};
\mathcal{M}_{b}({\mathbb{R}^d}\times {\rm S}^{d-1}))$ such that for all 
$\phi_1\in L^2({\mathbb{R}^m};  L^{\tilde p'}({\mathbb{R}^d}))$,
 $\frac{1}{\tilde{p}'}+\frac{2}{p}=1$ (with
$L^\infty({\mathbb{R}^d})$ being replaced by $C_0({\mathbb{R}^d})$ if $p=2$),
$\phi_2\in L_c^2(\mathbb{R}^{m};C_0({\mathbb{R}^d}))$,  and
$\psi\in C^{d}({\rm S}^{d-1})$ it holds
\begin{align*}
&\lim_{n}\int_{\mathbb{R}^{2m}}\int_{{\mathbb{R}^d}} (\phi_1
u_{n})(\mathbf{x}, \mathbf{y}) \,\Bigl(\overline{\mathcal{A}_{\psi} \,\phi_2
u_{n}(\cdot, \tilde{\mathbf{y}})}\Bigr)(\mathbf{x}) d\mathbf{x} d\mathbf{y} d\tilde{\mathbf{y}}\\
&=\int_{\mathbb{R}^{2m}} \langle
\mu(\mathbf{y},\tilde{\mathbf{y}},\cdot,\cdot),{\phi}_1(\cdot,\mathbf{y})\overline{\phi_2}
(\cdot,\tilde{\mathbf{y}})\otimes\overline\psi\rangle d\mathbf{y} d\tilde{\mathbf{y}}\,,
\end{align*}
where $\mathcal{A}_{\psi}$ is the (Fourier) multiplier operator on
$\mathbb{R}^d$ associated to $\psi(\boldsymbol{\xi}/|\boldsymbol{\xi})$.

Furthermore, the operator $\mu$ has the form
\begin{equation}\label{repr_1} 
\mu(\mathbf{y}, \tilde{\mathbf{y}},\mathbf{x},\boldsymbol{\xi})=f(\mathbf{y},
\tilde{\mathbf{y}},\mathbf{x},\boldsymbol{\xi})\nu(\mathbf{x},\boldsymbol{\xi}),
\end{equation}
where $\nu\in \mathcal{M}_b(\mathbb{R}^d\times {\rm S}^{d-1})$ is a non-negative scalar
Radon measure whose ${\mathbb{R}^d}$ projection 
$\int_{{\rm S}^{d-1}} d\nu(\mathbf{x},\boldsymbol{\xi})$  can
be extended to a bounded functional on $L^{\tilde p'}({\mathbb{R}^d})$ in the
case $p>2$, while $f$ is a function from $L^2(\mathbb{R}^{2m}; 
L^1({\mathbb{R}^d}\times {\rm S}^{d-1}:\nu))$.
\end{theorem}

An H-measure defined above is an object associated to a single 
$L^2$ sequence. However, there are no obstacles to adjoin a similar
object to two different sequences. This can be done by forming a vector
sequence, and consider non-diagonal elements of corresponding
(matrix) H-measure (e.g. \cite {Tar}). Another way is to joint two sequences in a
single one by means of a dummy variable, as it is done in the next
theorem.

\begin{theorem}\label{th10} 
Let $(u_n)$ be a bounded sequence in 
$L^2(\mathbb{R}^{d+m})\cap L^2({\mathbb{R}^d}; L^p(\mathbb{R}^{m})) \cap 
L^p(\mathbb{R}^{d+m})$ and
let $(v_n)$ be a sequence weakly converging to zero in 
$L^2(\mathbb{R}^{d}) \cap L^q (\mathbb{R}^{d})$ for some $p,q\geq 2$ 
such that there exists $r>1$ satisfying $1/r+1/p+1/q=1$. Then, after
passing to a subsequences (not relabeled), there exists a  measure
$\mu \in L_{ w^\ast}^2(\mathbb{R}^{m}; \mathcal{M}_{b}({\mathbb{R}^d}\times {\rm S}^{d-1}))$ such that for
all $\phi_1 \in { L}^{p'}(\mathbb{R}^{m}; { L}^r (\mathbb{R}^{d}))$ (with
$L^\infty({\mathbb{R}^d})$ being replaced by $C_0({\mathbb{R}^d})$ if $p=q=2$), $\phi_2 \in C_0({\mathbb{R}^d})$,
$\psi\in C^d(S^{d-1})$, we have
\begin{equation} \label{rev11}
\begin{aligned}
&\int_{\mathbb{R}^{m}}   \langle \mu(\mathbf{y},\cdot,\cdot), \phi_1(\cdot,\mathbf{y}) 
\overline{\phi_2} \otimes \overline\psi \rangle dy \\
&=\lim_{n\to \infty}\int_{\mathbb{R}^{m+d}} \phi_1(\mathbf{x}, \mathbf{y})u_n(\mathbf{x},\mathbf{y})
\overline{\mathcal{A}_{\psi} \big(\phi_2 v_n\big)(\mathbf{x})} d\mathbf{x}
d\mathbf{y}\,.
\end{aligned}
\end{equation}
Furthermore, the measure $\mu$ is of the form
\begin{equation}
\label{repr} \mu(\mathbf{y},\mathbf{x},\boldsymbol{\xi})
=f(\mathbf{y},\mathbf{x},\boldsymbol{\xi}) d\nu(\mathbf{x},\boldsymbol{\xi}),
\end{equation} 
where  $ f$ belongs to $L^2(\mathbb{R}^{m};
L^1({\mathbb{R}^d}\times {\rm S}^{d-1}:\nu))$, while $\nu\in \mathcal{M}_b(\mathbb{R}^d\times {\rm S}^{d-1})$ 
is a non-negative, bounded, scalar Radon measure.
In the case $\min\{p,q\}>2$ its  ${\mathbb{R}^d}$ projection 
$\int_{{\rm S}^{d-1}} d\nu(\mathbf{x},\boldsymbol{\xi})$ can be extended to a bounded 
functional on $L^s({\mathbb{R}^d})$, where $s$ is the dual index of $\min\{p,q\}/2$, i.e. 
there exists an $h\in L^{s'}({\mathbb{R}^d})$ such that 
$\int_{{\rm S}^{d-1}}d\nu(\mathbf{x},\boldsymbol{\xi}) =h(\mathbf{x}) d\mathbf{x}$.

We call $\mu$ the generalized H-measure
corresponding to (sub)se\-quen\-ces (of) $(u_n)$ and $(v_n)$.
\end{theorem}

\begin{proof}
  Let us first prove that the relation \eqref{rev11} holds for smooth 
(with respect to $\mathbf{x}$) test functions $\phi_1$ and $\phi_2$.

To this effect denote by $u$ an $L^2$ weak limit of the sequence
$(u_n)$ along a (non-relabeled) subsequence. Fix an arbitrary
non-negative compactly supported $\rho\in C_{\rm c}(\mathbb{R}^m)$ with the total
mass equal to one. Let
\begin{equation*}
W_n(\mathbf{x},\mathbf{y},\lambda)=
\begin{cases}
(u_n-u)(\mathbf{x},\mathbf{y}), & \lambda\in \langle 0,1\rangle\\
\rho(\mathbf{y})v_n(\mathbf{x}), & \lambda\in \langle-1,0\rangle\\
0, & \text{else}.
\end{cases}
\end{equation*}
Clearly, we have that $W_n \rightharpoonup 0$ in 
$L^2(\mathbb{R}^{d+m+1})\cap L^2(\mathbb{R}^{m+1};L^{\min\{p,q\}}({\mathbb{R}^d}))$, and by 
Theorem \ref{lfeb518-r} it admits a measure
$\tilde{\mu}\in L_{ w^\ast}^2(\mathbb{R}^{2(m+1)}; 
\mathcal{M}_b(\mathbb{R}^d\times {\rm S}^{d-1}))$ 
such that for any $\tilde{\phi}_1\!\in \!{L}^{2}(\mathbb{R}^{m+1}; C_0({\mathbb{R}^d}))$,
$\tilde{\phi}_2\in L_c^2(\mathbb{R}^{m+1}; C_0({\mathbb{R}^d}))$, and $\psi \in {\rm C}({\rm S}^{d-1})$,
it holds
\begin{equation} \label{rev111}
\begin{aligned}
 &\int_{\mathbb{R}^{2m}} \langle
\tilde\mu(\mathbf{y},\tilde{\mathbf{y}},\cdot,\cdot),{\tilde\phi}_1(\cdot,\mathbf{y})\overline{\tilde\phi_2}(\cdot,\tilde{\mathbf{y}})\otimes\overline\psi\rangle d\mathbf{y} d\tilde{\mathbf{y}}
\\&=\lim_{n\to
\infty}\int_{\mathbb{R}^{d+2(m+1)}} (\tilde\phi_1 W_n)(\mathbf{x},\mathbf{y},\lambda)
\overline{\mathcal{A}_{\psi} \big(\tilde\phi_2 W_n
(\cdot,\tilde{\mathbf{y}},\tilde\lambda)\big)(\mathbf{x})} d\mathbf{x} d\textbf{w},
\end{aligned}
\end{equation}
where $\textbf{w}=(\mathbf{y},\tilde{\mathbf{y}},\lambda,\tilde{\lambda})\in \mathbb{R}^{2m+2}$.

According to the representation \eqref{repr_1}, the measure
$\tilde{\mu}$ is of the form
\begin{equation*}
\tilde{\mu}=\tilde
f(\mathbf{y},\tilde{\mathbf{y}},\lambda,\tilde{\lambda},\mathbf{x},\boldsymbol{\xi})
d\nu(\mathbf{x},\boldsymbol{\xi}), \quad
 \mathbf{y}, \tilde{\mathbf{y}} \in \mathbb{R}^m, \; \lambda,\tilde{\lambda} \in \mathbb{R},
\end{equation*}
where $\tilde f$ is a function from $L^2(\mathbb{R}^{2(m+1)};
  L^1({\mathbb{R}^d}\times {\rm S}^{d-1}:\nu))$, while
$\nu\in \mathcal{M}_b(\mathbb{R}^d\times {\rm S}^{d-1})$ is a non-negative scalar
Radon measure,   whose  ${\mathbb{R}^d}$ projection 
$\int_{{\rm S}^{d-1}} d\nu(\mathbf{x},\boldsymbol{\xi})$can
be extended to a bounded functional on $L^s({\mathbb{R}^d})$, where $s$ 
is the dual index of $\min\{p,q\}/2$  in the case $\min\{p,q\}>2$.

By taking in \eqref{rev111}
$\tilde{\phi}_1(\mathbf{x},\mathbf{y},\lambda)=\phi_1(\mathbf{x},\mathbf{y}) \otimes
\theta_1(\lambda)$, and
$\tilde{\phi}_2(\mathbf{x},\tilde{\mathbf{y}},\tilde{\lambda})= \phi_2(\mathbf{x})
\otimes \rho_2(\tilde{\mathbf{y}})\otimes\theta_2(\tilde{\lambda})$, where
$\phi_1\in {L}^{2}(\mathbb{R}^{m}; C_0({\mathbb{R}^d}))$ and $\phi_2\in C_0({\mathbb{R}^d})$ are
arbitrary test functions, while $\theta_{1}=\chi_{[0,1]},  
\theta_{2}=\chi_{[-1,0]}$, and $\rho_2(\tilde{\mathbf{y}})=1$ for
$\tilde{\mathbf{y}}\in {\rm supp}\rho$, we see that the measure
$$
d\mu(\mathbf{y}, \mathbf{x},\boldsymbol{\xi})
=\Big(\int_{-1}^0\int_{0}^1\int_{\mathbb{R}^{m}} \tilde
f(\mathbf{y},\tilde{\mathbf{y}},\lambda,\tilde{\lambda},\mathbf{x},\boldsymbol{\xi})\, d\tilde{\mathbf{y}}
d\lambda d\tilde{\lambda} \Big) d\nu(\mathbf{x},\boldsymbol{\xi})d\mathbf{y},
$$
satisfies \eqref{rev11} and \eqref{repr} for
all $\phi_1 \in {L}^{2}(\mathbb{R}^{m}; C_0({\mathbb{R}^d}))$, $\phi_2 \in C_0({\mathbb{R}^d})$,
and $\psi\in C^d(S^{d-1})$.

In the second part of the proof we show that the relation \eqref{rev11} 
extends to test functions $\phi_1$ taken from the space 
${ L}^{p'}(\mathbb{R}^{m}; { L}^r (\mathbb{R}^{d}))$. 
To this effect, take an arbitrary such function  and denote by $(\phi_1^k)$ 
a sequence  of compactly supported
continuous functions such that $\phi_1^k \to \phi_1$ strongly in 
${L}^{p'}(\mathbb{R}^{m}; { L}^r (\mathbb{R}^{d}))$. 
For $\phi_2$ and  $\psi$ as in \eqref{rev11},
we define
\begin{equation} \label{cshy}
\begin{split}
&\int_{\mathbb{R}^{2m}} \langle \mu(\mathbf{y},\cdot,\cdot),
 \varphi_1(\cdot,\mathbf{y}) \bar{\varphi}_2\otimes\bar\psi\rangle d\mathbf{p} d\mathbf{q} \\
&:=\lim_{k\to \infty}\int_{\mathbb{R}^{2m}} \langle
\mu(\mathbf{y},\cdot,\cdot),\varphi^k_1\bar{\varphi}_2
\otimes\bar\psi\rangle d\mathbf{p} d\mathbf{q}\\
&\,=\lim_{k\to \infty}\lim_{n\to \infty}\int_{\mathbb{R}^{2m}}\int_{{\mathbb{R}^d}}
(\varphi_1^k u_{n})(\mathbf{x},\mathbf{p})
\Bigl(\overline{\mathcal{A}_{\psi} \,\varphi_2 u_{n}}\Bigr)(\mathbf{x})
d\mathbf{x} d\mathbf{p} d\mathbf{q}.
\end{split}
\end{equation}
The above limit (with respect to $k$) exists, since by means of the H\" older
inequality and the Marcinkiewicz multiplier theorem
(see Remark \ref{bound} below) for $k_1, k_2 \in \mathbb{N}$ it holds
\begin{align*}
&\Big|\,\int_{\mathbb{R}^{2m}} \langle
\mu(\mathbf{y},\cdot,\cdot),(\varphi_1^{k_1}-\varphi_1^{k_2})(\cdot,\mathbf{y})\bar{\varphi}_2
\otimes\bar\psi\rangle d\mathbf{y} \Big|\\
&\leq\limsup_{n} \int_{\mathbb{R}^{2m}}\int_{{\mathbb{R}^d}}
\Big|  (\varphi_1^{k_1}-\varphi_1^{k_2} ) u_{n}(\mathbf{x},\mathbf{y})
\Bigl(\overline{\mathcal{A}_{\psi} \,\varphi_2 v_{n}}\Bigr)(\mathbf{x}) \Big| d\mathbf{x} d\mathbf{y}
\\
&\leq \limsup_{n} C\, \|\psi\|_{C^d(S^{d-1})}  \|\varphi_2 v_{n}\|_{L^q({\mathbb{R}^d})}
\int_{\mathbb{R}^{m}}
\|(\varphi_1^{k_1}-\varphi_1^{k_2})(\cdot,\mathbf{y})\|_{L^r({\mathbb{R}^d})} \\
&\quad\times \|( u_{n}(\cdot,\mathbf{y})\|_{L^p({\mathbb{R}^d})} d\mathbf{y}
\\
& \leq \limsup_{n} C \|\psi\|_{C^d(S^{d-1})}
  \|(\varphi_1^{k_1}-\varphi_1^{k_2})\|_{L^{p'}({\mathbb{R}^m}; L^ r({\mathbb{R}^d}))}
\|\varphi_2 \|_{L^\infty({\mathbb{R}^d})} \\
&\quad\times \| u_{n}\|_{L^p(\mathbb{R}^{d+m})}\| v_{n}\|_{L^q({\mathbb{R}^d})} \,,
\end{align*}
where $C$ depends on $p$ and $d$ only.
Since $(\varphi_1^k)$ is a convergent sequence, hence the Cauchy one,
the limit in \eqref{cshy} exists, and it does not depend on the
approximating sequence $(\varphi_1^k)$.

The same analysis as above implies
\begin{equation*}
\lim_{k\to \infty }\int_{\mathbb{R}^{2m}}\int_{{\mathbb{R}^d}}
\Big|  (\varphi_1^{k}-\varphi_1 ) u_{n}(\mathbf{x},\mathbf{p})
\,\Bigl(\overline{\mathcal{A}_{\psi} \,\varphi_2 u_{n}(\cdot,
\mathbf{q})}\Bigr)(\mathbf{x}) \Big|
d\mathbf{x} d\mathbf{p} d\mathbf{q} = 0
\end{equation*}
and the convergence is uniform with respect to $n$. 
Thus we can exchange limits in the second line of \eqref{cshy}, 
which proves \eqref{rev11}.
\end{proof}

The last theorem takes into account additional regularity of sequences 
$(u_n)$ and $(v_n)$. If the first one is bounded in $L^p$ and the 
latter one in $L^{\infty}$, test functions can be taken from $L^{p'}$ almost 
(more precisely, from $L^{p'+\varepsilon}$, $\varepsilon>0$ arbitrary),
 which is  maximal regularity one can expect in order for the right hand 
side of \eqref{rev11} to make sense.

The following statement on the measure $\nu$ now follows from  
results on slicing measures \cite[Theorem 1.10]{Evans}.

\begin{lemma} \label{lemma-slicing}
Under the assumptions of the above theorem, in the case $s>1$ for 
 a.e.  $\mathbf{x}\in {\mathbb{R}^d}$ there exists a Radon probability measure 
$\nu_\mathbf{x}$ such that 
$d\nu(\mathbf{x}, \boldsymbol{\xi})= d \nu_\mathbf{x}  (\boldsymbol{\xi})
 h(\mathbf{x}) d\mathbf{x}$, where $h$ is a $L^{s'}$ 
function from Theorem \ref{th10}. More precisely, for each 
$\phi\in{\rm C}_0({\mathbb{R}^d}\times {\rm S}^{d-1})$,
$$
\int_{{\mathbb{R}^d}\times {\rm S}^{d-1}} \phi(\mathbf{x}, \boldsymbol{\xi})d\nu(\mathbf{x}, 
\boldsymbol{\xi}) 
= \int_{\mathbb{R}^d} \Big( \int_{{\rm S}^{d-1}} \phi(\mathbf{x}, \boldsymbol{\xi}) d 
\nu_\mathbf{x}  (\boldsymbol{\xi}) \Big)h(\mathbf{x}) d\mathbf{x}\,.
$$
The above result is also valid if we take a test function 
$\phi \in L^s({\mathbb{R}^d}; {\rm C}({\rm S}^{d-1}))$.
\end{lemma}

We end this section with the analysis  of the  Fourier multipliers 
which  form the basis in construction and application of H-measures.

\begin{definition} \label{multiplier} \rm
For a given $p\in[ 1\infty\rangle$, let the multiplier operator  $\mathcal{A}_\psi$
satisfy  
$$
\|\mathcal{A}_\psi (u)\|_{L^p} \leq C \| u\|_{L^p}, \quad u\in \mathcal{S},
$$
where $C$ is a positive constant, while $\mathcal{S}$ stands for a Schwartz space. 
Then its symbol $\psi$ is called an $L^p$ (Fourier) multiplier.
\end{definition}

There are many criteria on a symbol $\psi$ providing it to be an
$L^p$ multiplier.  In the paper, we shall need the Marcinkiewicz
multiplier theorem \cite[Theorem 5.2.4.]{Gra}, more precisely its
corollary which we provide here:

\begin{corollary}\label{m1} 
Suppose that  $\psi\in C^d(\mathbb{R}^d\backslash \cup_{j=1}^d\{\xi_j= 0\})$ 
is a bounded function such that for some constant $C>0$ it holds
\begin{equation}\label{c-mar} 
|\boldsymbol{\xi}^{\tilde{\boldsymbol{\alpha}}} \partial^{\tilde{\boldsymbol{\alpha}}} 
\psi(\boldsymbol{\xi})|\leq C,\quad
 \boldsymbol{\xi}\in \mathbb{R}^d\backslash \cup_{j=1}^d\{\xi_j= 0\}
\end{equation}
for  every multi-index
$\tilde{\boldsymbol{\alpha}}=(\tilde\alpha_1,\dots,\tilde\alpha_d) \in \mathbf{N}_0^d$ such that
$|\tilde{\boldsymbol{\alpha}}|=\tilde\alpha_1+\tilde\alpha_2+\dots+\tilde\alpha_d \leq d$. 
Then, the function $\psi$ is an ${ L}^p$-multiplier for $p\in \langle1,+\infty\rangle$, and the
operator norm of $\mathcal{A}_\psi$ depends only on $C, p$ and $d$.
\end{corollary}

\begin{remark} \label{bound} \rm
Using this corollary, it is proved in \cite{Gra} that for a bounded function 
$\psi$ defined on the unit sphere ${\rm S}^{d-1}$ and smooth outside coordinate 
hyperplanes, its extension (not relabelled) 
$\psi(\boldsymbol{\xi}/|\boldsymbol{\xi}|)$ is an $L^p$ multiplier
(see also \cite[Lemma 5]{LM2}).
If in addition we assume that $\psi$ is smooth on the whole manifold, 
i.e. $\psi \in C^d({\rm S}^{d-1})$, then the corresponding operator satisfies
\[ %\label{marz_bnd} 
\|\mathcal{A}_{\psi} \|_{L^p\to L^p}
\leq C \|\psi\|_{C^d({\rm S}^{d-1})},
\]
with a constant $C$ depending only on $p\in \langle 1,\infty\rangle$ and $d$.
\end{remark}

Here, we shall need a similar statement.

\begin{lemma} \label{l1IIv}
Let   $\theta:\mathbb{R}^d\to \mathbb{R}$ be a smooth compactly
supported function equal to one on the unit ball centered at the origin.
Then  for any $\gamma>0$ the multiplier operator $\mathcal{T}^{\gamma}$ with
the symbol
$$
T^\gamma(\boldsymbol{\xi})(1-\theta(\boldsymbol{\xi}))
=\frac{1}{|\boldsymbol{\xi}|^{\gamma}}(1-\theta(\boldsymbol{\xi}))
$$
is a continuous $L^p({\mathbb{R}^d})\to { W}^{\gamma, p}(\mathbb{R}^d)$ operator for 
any $p\in \langle 1,+\infty\rangle$.
Specially, due to the Rellich theorem it is a compact 
$L^p({\mathbb{R}^d})\to L^p_{\rm loc}({\mathbb{R}^d})$ 
operator.
\end{lemma}

\begin{proof}
We shall first prove that the operator $\mathcal{T}^\gamma$ is a continuous 
operator on $L^p({\mathbb{R}^d})$. To this effect, remark that
it is sufficient to prove that $T^\gamma$ satisfies condition
of Corollary \ref{m1} away from the origin. Around the origin, the
operator $\mathcal{T}^\gamma$ is controlled by the term $(1-\theta)$ (which is
equal to zero on $B(0,1)$ and obviously satisfies conditions of
Corollary \ref{m1}).
We use the induction argument with respect to the order of derivative 
in \eqref{c-mar}.

$\bullet$  $n=1$:
In this case, we compute
\begin{equation*}
\partial_{k}T^\gamma(\boldsymbol{\xi})=
C_{k}\frac{1}{\xi_k} T^\gamma(\boldsymbol{\xi}) 
\big(\frac{\xi_k}{|\boldsymbol{\xi}|})\big)^{2}
\end{equation*}
for some constant $C_{k}$. From here, it obviously
follows $|\xi_k \partial_{k} T^\gamma(\boldsymbol{\xi})| \leq C$ for
$\boldsymbol{\xi}\in \mathbb{R}^d$ away from the origin.


$\bullet$  $n=m$:
Our inductive hypothesis is that a $\boldsymbol{\alpha}$-order derivatives of 
$T^\gamma(\boldsymbol{\xi})$
can be represented as
\begin{equation}
\label{ih}
\partial^{\boldsymbol{\alpha}} T^\gamma(\boldsymbol{\xi})
=\frac{1}{\boldsymbol{\xi}^{\boldsymbol{\alpha}}} T^\gamma(\boldsymbol{\xi})
 P_{\boldsymbol{\alpha}}(\boldsymbol{\xi}),
\end{equation}
where $P_{\boldsymbol{\alpha}}$ is a bounded function satisfying \eqref{c-mar}
for $|\tilde{\boldsymbol{\alpha}}| \leq d - |\boldsymbol{\alpha}|$.

$\bullet$ $n=m+1$:
To prove that \eqref{ih} holds for $|\boldsymbol{\alpha}|=m+1$ it is enough to
notice that $\boldsymbol{\alpha}={\sf e}_k+\boldsymbol{\alpha}'$, where $|\boldsymbol{\alpha}'|=m$, and that 
according to the induction hypothesis we have
\[
\partial^{\boldsymbol{\alpha}} T^\gamma =\partial_k \partial^{\boldsymbol{\alpha}'}
T^\gamma
=\partial_k\Big(\frac{1}{\boldsymbol{\xi}^{\boldsymbol{\alpha}'}} 
 T^\gamma(\boldsymbol{\xi}) P_{\boldsymbol{\alpha}'}(\boldsymbol{\xi})\Big)
=\frac{1}{\boldsymbol{\xi}^{\boldsymbol{\alpha}}} T^\gamma(\boldsymbol{\xi}) P_{\boldsymbol{\alpha}}(\boldsymbol{\xi}),
\]
where
$$
P_{\boldsymbol{\alpha}}(\boldsymbol{\xi})= (P_{{\sf e}_k} P_{\boldsymbol{\alpha}'}+ \xi_k \partial_k P_{\boldsymbol{\alpha}'}
-\alpha_k P_{\boldsymbol{\alpha}'}) (\boldsymbol{\xi}),
$$
thus satisfying conditions \eqref{c-mar} as well.


From these, \eqref{c-mar} immediately follows for
$T^\gamma$ away from the origin, thus proving that the operator 
$\mathcal{T}^\gamma$ is a continuous operator on $L^p({\mathbb{R}^d})$.

It remains to prove that for any $j$  the multiplier operator
$\partial^{\gamma}_{x_j} \mathcal{T}^\gamma$ is a continuous 
${ L}^p(\mathbb{R}^d)\to {L}^p(\mathbb{R}^d)$ operator. 
To accomplish this, notice that its symbol is
$(1-\theta(\boldsymbol{\xi})) (2\pi i \xi_j/|\boldsymbol{\xi}|)^{\gamma}$.
Thus, away from the origin, it is a composition of a  smooth function 
and the projection $\boldsymbol{\xi}\to \boldsymbol{\xi}/|\boldsymbol{\xi}|$, 
and by Remark \ref{bound}  it satisfies conditions of Corollary \ref{m1}.
\end{proof}

\section{Velocity averaging}

In this section, we provide  a velocity averaging result for a linear transport 
equation with a low  regularity assumptions on the  coefficients.

More precisely,  we consider a sequence of functions
$(u_n)$ weakly converging to zero in ${ L}^{p}(\mathbb{R}^{d+m})$ for $p\geq 2$,
and satisfying the following sequence of  equations
\begin{equation}\label{lin-eq}
\mathcal{P}u_n(\mathbf{x},\mathbf{y})=\sum_{k=1}^d\partial_{x_k}
\left(a_k(\mathbf{x},\mathbf{y}) u_n(\mathbf{x},\mathbf{y})\right)=\partial^{\boldsymbol{\kappa}}_\mathbf{y}
G_n(\mathbf{x},\mathbf{y})\,,
\end{equation}
where
$\partial_\mathbf{y}^{\boldsymbol{\kappa}}=\partial^{\kappa_1}_{y_1}\dots \partial^{\kappa_m}_{y_m}$
for a multi-index $\boldsymbol{\kappa}=(\kappa_1,\dots,\kappa_m)\in \mathbb{N}^m$.

By $A$ we denote the principal symbol of the differential
operator $\mathcal{P}$, which is of the form
\begin{equation}
\label{glavni}
 A(\mathbf{y},\mathbf{x},\boldsymbol{\xi})=\sum_{k=1}^d 2\pi
i\boldsymbol{\xi}_k a_k(\mathbf{x},\mathbf{y}) .
\end{equation}
We assume it satisfies the classical non-degeneracy
conditions
\begin{equation}\label{kingnl} 
\forall (\mathbf{x}, \boldsymbol{\xi}) \in D\times{\rm S}^{d-1} 
\quad A(\mathbf{y},\mathbf{x},\boldsymbol{\xi})\not= 0
\quad \text{(a.e. $\mathbf{y} \in \mathbb{R}^{m}$)}\,,
\end{equation}
where $D\subseteq{\mathbb{R}^d}$ is a full measure set, while ${\rm S}^{d-1}$ stands 
for the unit sphere in ${\mathbb{R}^d}$.

As for the coefficients from \eqref{lin-eq}, we assume that
\begin{itemize}
\item[(a)]
$a_k\in { L}^{\bar p'}(\mathbb{R}^{m+d})$ for some 
$\bar{p}\in \langle 1,p\rangle$ $k=1,\dots,d$;

\item[(b)] The sequence $(G_n)$ is strongly precompact in the
space 
${ W}_{\rm loc}^{-1,q}(\mathbb{R}^{m+d})$, where $q>1$ 
is determined by the relation
$1+\frac{1}{p}=\frac{1}{\bar{p}}+\frac{1}{q}$ .
\end{itemize}

The idea behind the extension of the velocity averaging result to coefficients 
of lower regularity (as given in (a)) is to consider the H-measure corresponding 
to the sequence $(u_n)$ and the truncated sequence $(T_l(u_n))$ for the 
truncation operator $T_l$ given as follows: for $l\in \mathbb{N}$,
\begin{equation} \label{trunc}
T_l(u)=\begin{cases} \operatorname{sign}(u) l, & |u|>l\\
u, & u\in [-l,l]\,.
\end{cases} 
\end{equation} 
The operator $T_l$ and its variants have been widely used \cite{BK, DHM} 
where it was noticed that convergence of $(T_l (u_n))$, for every $l\in \mathbb{N}$, in
$L^1_{\rm loc}(\mathbb{R}^d)$ implies the strong convergence of $(u_n)$ in
$L^1_{\rm loc}(\mathbb{R}^d)$.
This property, beside being explored in the proof of Theorem \ref{va},  
will be also used for proving an existence of a solution to non-linear 
transport equation \eqref{d-p} in the next section.

Let us now prove necessary properties of the truncation operator.

\begin{lemma}\label{trunc-l}
 Let $(u_n)$ be an equi-integrable sequence, bounded in $L^1(\Omega)$, 
where $\Omega$ is an open set in $\mathbb{R}^d$.
Then for   the sequence of truncated functions it holds
\begin{equation} \label{uni_n}
\lim_l \sup_n \|T_l(u_{n}) -  u_{n}\|_{L^1(\Omega)} \to 0\,.
\end{equation}
\end{lemma}

\begin{proof}
Denote by
$\Omega_n^l=\{ \mathbf{x}\in\Omega:\, u_{n}(\mathbf{x}) > l \}$.
Since $(u_{n})$ is bounded in $L^1(\Omega)$ we have
\[
\sup_{k\in \mathbb{N}} \int_{\Omega }|u_{n}(\mathbf{x})| d\mathbf{x} 
\geq \sup_{k\in \mathbb{N}} \int_{\Omega_n^l} l d\mathbf{x}
 \implies \frac{1}{l}\sup_{k\in \mathbb{N}} \int_{\Omega }|u_{n}(\mathbf{x})| d\mathbf{x} \geq
\sup_{k\in \mathbb{N}} \operatorname{meas}(\Omega_n^l),
\]
implying that
\begin{equation} \label{conv1}
\lim_{l\to \infty} \sup_{n\in \mathbb{N}} \operatorname{meas}(\Omega_n^l) =0.
\end{equation}
Now
$$
\int_\Omega |u_{n}-T_l(u_{n})|dx
\leq \int_{\Omega_n^l} |u_{n}|dx \underset{l\to \infty}\to 0
$$
uniformly with respect to $n$ according to \eqref{conv1} and equi-integrability 
of $(u_n)$. Thus, \eqref{uni_n} is proved.
\end{proof}


\begin{lemma} \label{DHM} 
Let $(u_n)$ be an equi-integrable sequence, bounded in $L^1(\Omega)$,  
where $\Omega$ is an open set in $\mathbb{R}^d$. Suppose
that for each $l\in\mathbb{N}$ the sequence of truncated functions
$(T_l(u_n))$ is precompact in ${L}^1(\Omega)$. Then there
exists a subsequence $(u_{n_k})$ and function $u\in L^1(\Omega)$ such that
$$
u_{n_k} \to u  \quad \text{in }L^1(\Omega).
$$
\end{lemma}

\begin{proof}
By the strong precompactness assumptions on truncated sequences, 
there exists a subsequence  $(u_{n_k})$  such that for every $l\in \mathbb{N}$
the sequence $(T_l(u_{n_k}))$ is convergent in $L^1(\Omega)$, with  a limit 
denoted by $u^l$.
We  prove that the obtained sequence $(u^l)$  converges strongly in
$L^1(\Omega)$ as well.
To this end, note that
\begin{align*} %\label{ul} 
\|u^{l_1}-u^{l_2}\|_{L^1(\Omega)}
&\leq \|u^{l_1}-T_{l_1}(u_{n_k})\|_{L^1(\Omega)}+\|T_{l_1}(u_{n_k})-u_{n_k}
 \|_{L^1(\Omega)}\\
&\quad +\|T_{l_2}(u_{n_k})-u_{n_k}\|_{L^1(\Omega)}
 +\|T_{l_2}(u_{n_k})-u^{l_2}\|_{L^1(\Omega)}\,,
\end{align*}
which together with  Lemma \ref{trunc-l} implies that $(u^l)$ is a Cauchy sequence. 
Thus, there exists $u\in L^1(\Omega)$ such that
\begin{equation} \label{conv2}
u^l \to u \quad\text{in }L^1(\Omega).
\end{equation}

Now it is not difficult to see that entire $(u_{n_k})$ converges toward $u$
in $L^1(\Omega)$ as well. Namely, it holds
\[
  \|u_{n_k}-u\|_{L^1(\Omega)} \leq
\|u_{n_k}-T_l(u_{n_k})\|_{L^1(\Omega)}+\|T_l(u_{n_k})-u^l\|_{L^1(\Omega)}
+\|u^l-u\|_{L^1(\Omega)},
\]
which by the definition of functions $u^l$, and convergences \eqref{uni_n} 
and \eqref{conv2} imply the statement.
\end{proof}

An elementary corollary of the previous lemma is:

\begin{corollary}\label{DHM-1}
Let $(u_n)$ be a bounded sequence in $L^p(\Omega) \cap L^1(\Omega)$ for some $p>1$, 
 where $\Omega$ is an open set in $\mathbb{R}^d$. Suppose
that for each $l\in\mathbb{N}$ the sequence of truncated functions
$(T_l(u_n))$ is precompact in ${L}^1(\Omega)$. Then there
exists a subsequence $(u_{n_k})$ and function $u\in L^p(\Omega)$ such that
$$
u_{n_k} \to u  \quad \text{in } L^1(\Omega).
$$
\end{corollary}

\begin{proof}
It is sufficinet to notice that every sequence $(u_n)$ which is bounded 
in $L^p(\Omega)$, for $p>1$, is equi-integrable.
\end{proof}

The result  given in Theorem \ref{va} below is usually called a velocity
averaging lemma. Its importance is demonstrated in many works, but we shall mention
only very famous \cite{LPT} (kinetic formulation of conservation laws) and 
\cite{8} (existence of weak solution to the Boltzman equation). 
Concerning the averaging lemma itself, one can consult e.g. 
\cite{13, 18,32, 36}. We remark that almost all the velocity averaging 
results were given for homogeneous equations
(i.e.~the ones where coefficients do not depend on $\mathbf{x}\in \mathbb{R}^d$). 
The reason for this one
can search in the fact that, in the homogeneous situation, one can
separate the solutions $u_n$ from the coefficients (e.g. by applying
the Fourier transform with respect to $\mathbf{x}$), and this is basis of
most of the methods (see e.g. \cite{36} and references therein). We
remark that more detailed observations on this issue one can find in
the introduction of \cite{LM2}. In order to deal with the heterogeneous situation that we have here, we need microlocal defect tools, in particular H-measures. This brings obstacles of their own which forced us to use the restrictive non-degeneracy
condition in \cite{LM_crass}:
\[ % \label{ggnc}
\frac{|A|^{2}}{|A|^{2}+\delta} \to 1 \quad\text{in } 
  L_{\rm loc}^{\bar p'}(\mathbb{R}^{d+m}; C^d({\rm S}^{d-1}))
\]
strongly as $\delta\to 0$.
By using the characterization of H-measures provided by Theorem \ref{th10} 
we are able to  generalize that  result by assuming merely the classical
 non-degeneracy condition \eqref{kingnl}.

 In addition,   by allowing for lower regularity assumptions on
the  coefficients $a_k$ the following theorem also generalizes the
velocity averaging results provided in \cite{LM2}.

\begin{theorem} \label{va}
 Assume that $u_n\rightharpoonup 0$  weakly in
${L}_{\rm loc}^{p}(\mathbb{R}^{d+m})$, $p\geq 2$, where
$u_n$ represent weak solutions to \eqref{lin-eq} with conditions (a) and (b)
being fulfilled. Furthermore, we assume that the classical non-degeneracy
conditions \eqref{kingnl}  are satisfied.

Then, for any $\rho\in L^2_c(\mathbb{R}^m)$,
\begin{equation} \label{va_result}
\int_{\mathbb{R}^m}\rho(\mathbf{y}) u_n(\mathbf{x},\mathbf{y})d\mathbf{y} \to 0 \quad 
\text{strongly in } L_{\rm loc}^1({\mathbb{R}^d}).
\end{equation}
\end{theorem}


\begin{proof}
Fix $\rho\in C_{\rm c}(\mathbb{R}^m)$, $\varphi\in L_c^\infty({\mathbb{R}^d})$,  and $l\in \mathbb{N}$. 
Denote by $V_l$ a weak $\ast$ ${ L}^\infty(\mathbb{R}^{d})$ limit 
along a subsequence of truncated averages defined by 
\[
V^l_n=\varphi T_l ( \int_{\mathbb{R}^m} \rho(\tilde{\mathbf{y}})
u_n(\cdot,\tilde{\mathbf{y}}) d\tilde{\mathbf{y}}),
\]
 where $T_l$ is the truncation operator 
introduced in \eqref{trunc} below. Denote $v^l_n=V^l_n-V_l$ and remark
that $v^l_n  \stackrel{\ast}{\rightharpoonup} 0$
in ${L}^\infty(\mathbb{R}^{d})$ with respect to $n$.

Next, let $\mu_l$ be the generalized H-measure defined in Theorem
\ref{th10} corresponding to  (sub)se\-quen\-ces (of) $\varphi u_n$ and $v^l_n$.

Take a dual product of \eqref{lin-eq} with the  test functions
\begin{equation*}
g_n(\mathbf{x},\mathbf{y})=\varphi_1(\mathbf{x}) \rho_1(\mathbf{y}) 
(\mathcal{T}^1\circ \mathcal{A}_{\psi_{{\rm S}^{d-1}}}) (v_n)(\mathbf{x}),
\end{equation*}
where  $\psi\in C^d(S^{d-1})$, $\varphi_1\in {\rm C}^{\infty}_c({\mathbb{R}^d})$, and 
$\rho_1 \in {\rm C}_c^{|\boldsymbol{\kappa}|}(\mathbb{R}^m)$ 
are arbitrary test functions, while $\boldsymbol{\kappa}$ is the multi-index appearing
in \eqref{lin-eq}, while $\mathcal{T}^1$ is the multiplier operator  
defined in Lemma \ref{l1IIv} with
the symbol equal to  $\frac{1}{|\boldsymbol{\xi}|}$ outside the unit ball 
centered at the origin.

Letting $n\to \infty$ in such obtained expression, and
taking into account Theorem \ref{lfeb518-r}, we obtain
\[
\int_{\mathbb{R}^{2m}}\int_{\mathbb{R}^d\times \rm P}
A(\mathbf{x},\boldsymbol{\xi},\mathbf{p})\overline{\rho_1(\mathbf{p})
\rho_2(\mathbf{q}){\varphi}(x)\psi(\boldsymbol{\xi})}
d\mu_l(\mathbf{p},\mathbf{q},\mathbf{x},\boldsymbol{\xi})d\mathbf{p} d\mathbf{q}=0.
\]
The right-hand side of the last expression equals zero as, by assumption (b), 
the sequence  $(G_n)$ of functions on the right hand side of \eqref{lin-eq}
converges strongly to zero in $L^1(\mathbb{R}^m;W^{-1,q}(\mathbb{R}^d))$,
 while, according to Lemma \ref{l1IIv}, the
multiplier operator $\mathcal{T}^1 \circ \mathcal{A}_{\psi_{\rm P}}:
L^q({\mathbb{R}^d}) \to W^{1,q}(\mathbb{R}^d)$ is bounded.

As  the test functions $\rho_i$, $\varphi$, and $\psi$ are taken from dense 
subsets in appropriate spaces,   one gets  that $\mu_l$
satisfies localization principle
\begin{equation} \label{vv2} 
A \mu_l=0,
\end{equation} 
where $A$ is the principal symbol given by \eqref{glavni}.

 We aim to prove that from here, under condition \eqref{kingnl}, it follows
that $\mu_l \equiv 0$.
To this effect, we take an arbitrary $\delta>0$, and for 
$\rho\in L_c^2(\mathbb{R}^m)$
and $\phi \in C_{\rm c}(\mathbb{R}^d);C^d(S^{d-1})$ we consider the test function
\[
\frac{\rho(\mathbf{y})\phi(\mathbf{x},\boldsymbol{\xi})\overline{A(\mathbf{x},\boldsymbol{\xi},\mathbf{y})}}{|A(\mathbf{x},\boldsymbol{\xi},\mathbf{y})|^2
+\delta}.
\]
The localization principle \eqref{vv2} implies
\[
\big\langle{\mu_l},{\rho(\mathbf{y})\phi(\mathbf{x},\boldsymbol{\xi})
\frac{|{A(\mathbf{x},\mathbf{y},\boldsymbol{\xi})}|^2}{|A(\mathbf{x},
\mathbf{y},\boldsymbol{\xi})|^2+\delta}}\big\rangle=0\,,
\]
which by means of representation \eqref{repr} and Fubini's theorem
takes the form
\begin{equation}\label{fin_4} 
\int_{\mathbb{R}^d\times {\rm S}^{d-1}}\int_{\mathbb{R}^{m}}
\frac{\rho(\mathbf{y})\phi(\mathbf{x},\boldsymbol{\xi})|A(\mathbf{x},
\boldsymbol{\xi},\mathbf{y})|^2}{
|A(\mathbf{x},\boldsymbol{\xi},\mathbf{y})|^2 
+\delta}f_l(\mathbf{y},\mathbf{x},\boldsymbol{\xi}) d\mathbf{y}
d\nu_l(\mathbf{x},\boldsymbol{\xi})=0.
\end{equation}
Let us denote
$$
I_\delta(\mathbf{x}, \boldsymbol{\xi})=  \int_{\mathbb{R}^{m}}
\rho(\mathbf{y})\frac{|A(\mathbf{x},\boldsymbol{\xi},\mathbf{y})|^2}{ |A(\mathbf{x},
\boldsymbol{\xi},\mathbf{y})|^2
+\delta}f_l(\mathbf{y},\mathbf{x},\boldsymbol{\xi})d\mathbf{y}\,.
$$
According to the non-degeneracy condition \eqref{kingnl} and the representation
of the measure $\nu_l$ given in Lemma \ref{lemma-slicing},  we have
$$
I_\delta(\mathbf{x}, \boldsymbol{\xi})\to   \int_{\mathbb{R}^{m}}
\rho(\mathbf{y})f_l(\mathbf{y},\mathbf{x},\boldsymbol{\xi})d\mathbf{y},
$$
as $\delta\to 0$ for $\nu_l$ -  a.e. $(\mathbf{x}, \boldsymbol{\xi}) \in \mathbb{R}^d\times
{\rm S}^{d-1}$. By using the Lebesgue dominated convergence theorem, it
follows from \eqref{fin_4} after letting $\delta\to 0$:
\[
\langle{\mu_l},{\rho\otimes\phi}\rangle
= \int_{\mathbb{R}^d\times
{\rm S}^{d-1}}\int_{\mathbb{R}^{m}}\rho(\mathbf{y})\phi(\mathbf{x},
\boldsymbol{\xi})f_l(\mathbf{y},\mathbf{x},\boldsymbol{\xi}) d\mathbf{y}
d\nu_l(\mathbf{x},\boldsymbol{\xi}) =0\,,
\]
i.e. $\mu_l=0$ for every $l$.

From the definitions of the generalized H-measures and the truncation 
operator $T_l$,  we conclude by taking in \eqref{rev11} test functions 
$\psi=1$ and $\phi_1\phi_2=\chi_{{\rm supp} \varphi}\otimes\rho$ for the previously 
chosen $\varphi$ and $\rho$  (see the beginning of the proof):
\begin{equation} \label{bd103}
0=\lim_{n\to \infty} \int_{{\mathbb{R}^d} } \varphi^2(x)
\Big|T_l \int_{\mathbb{R}^m}\rho(\mathbf{y})u_n(\mathbf{x},\mathbf{y})d\mathbf{y}\Big|^2 d\mathbf{x},
\quad l\in \mathbb{N}.
\end{equation}
Now, using Corollary \ref{DHM-1}, we obtain the desired convergence
\eqref{va_result}.
\end{proof}

\section{Quasi-solutions and kinetic formulation}

In this section, we shall introduce the notion of quasi-solution to
\eqref{d-p}. A similar notion was first introduced in \cite{Pan}. 
In a special situation, e.g. in the case when the flux is independent of 
$(t,\mathbf{x})$ and when the measure $\zeta$ from below is non-negative,  
then the quasi-solution is an
entropy admissible solution that singles out a physically relevant
solutions to the equation \eqref{d-p} (see \cite{Kru}). 
The notion of quasi-solution
will lead to an appropriate kinetic formulation of the equation
under consideration which will enable us to use  H-measures.


\begin{definition} \rm
A measurable function $u$ defined on $\mathbb{R}^+\times \mathbb{R}$ is called a
quasi-solution to \eqref{d-p}, accompanied by assumptions (i) and (ii),  
if the following Kruzhkov type entropy equality holds
\begin{equation} \label{e-c}
\partial_t |u-\lambda|+\operatorname{div} [ \operatorname{sgn}
(u-\lambda)(f(t,\mathbf{x},u)-f(t,\mathbf{x},\lambda) )]
= -\zeta(t,\mathbf{x},\lambda),
\end{equation}
for some $\zeta \in \mathcal{M}(\mathbb{R}^+\times\mathbb{R}^d\times \mathbb{R})$
that we call the quasi-entropy defect measure.
\end{definition}

From the above entropy condition, the following kinetic
formulation can be derived.

\begin{theorem} \label{kf}
Denote $F=\partial_\lambda f$.
If the function $u$ is a quasi-solution to \eqref{d-p} then
the function
\begin{equation}\label{equil}
 h(t,\mathbf{x},\lambda)=\operatorname{sgn}(u(t,\mathbf{x})-\lambda)=
 -\partial_\lambda |u(t,x)-\lambda|
\end{equation} 
is a weak solution to the  linear equation:
\begin{equation} \label{k-1}
\partial_t h + \operatorname{div}\left( F(t,\mathbf{x},\lambda)
h\right)= \partial_\lambda
\zeta(t,\mathbf{x},\lambda)
\end{equation}
\end{theorem}

\begin{proof}
It is sufficient to find the derivative of \eqref{e-c} with respect to $\lambda\in
\mathbb{R}$ to obtain \eqref{k-1}.
\end{proof}


To prove existence of a solution to the  Cauchy problem \eqref{d-p}, 
\eqref{ic}, we need the following general statement.

\begin{theorem}\label{main-tvr}
 Assume that the function $F=\partial_\lambda f$, where $f$ is the flux appearing in
\eqref{d-p}, is such that for almost every $(t,\mathbf{x})\in \mathbb{R}_+^{d}$ and
every $\boldsymbol{\xi}\in S^{d}$ the mapping
\begin{equation} \label{non-deg}
 \lambda \mapsto  \Big( \xi_0+
\sum_{k=1}^d F_k(t,\mathbf{x},\lambda) \xi_k \Big),
\end{equation} 
is not zero on any set of positive measure.

Let $(u_n)$ be a sequence  of quasi-solutions
to \eqref{d-p}  bounded in $L^p(\mathbb{R}^+\times \mathbb{R}^d), p>1$. 
Then it is strongly precompact in $L^1_{\rm loc}(\mathbb{R}^+\times \mathbb{R}^d)$ 
if the sequence of corresponding entropy defect measures $(\zeta_n)$ is
strongly precompact in 
$W_{\rm loc}^{-1,q}(\mathbb{R}^+\times \mathbb{R}^d\times\mathbb{R})$ 
for some $q>1$.
\end{theorem}

\begin{proof}
Denote by $h_\infty\in L^\infty(\mathbb{R}^+\times \mathbb{R}^d\times \mathbb{R})$ 
a weak-$\star$ limit along a subsequence of the sequence 
$h_n(t, x,\lambda):=sign(u_n(t,x)-\lambda)$. By the boundedness assumption, 
the sequence $(\zeta_n)$ of quasi-entropy defect
measures corresponding to $(u_n)$ weakly converges to 
$\zeta \in L^1(\mathbb{R}; W^{-1,q}(\mathbb{R}^+\times \mathbb{R}^d))$ 
(possibly passing to a subsequence).
Put $v_n(t,\mathbf{x},\lambda)=h(t,\mathbf{x},\lambda)-h_n(t,\mathbf{x},\lambda)$ 
and $\sigma_n=\zeta-\zeta_n$. The sequence $(v_n)$ satisfies
\begin{equation} \label{k-11}
\partial_t v_n + \operatorname{div}\left( F(t,\mathbf{x},\lambda)
v_n\right)= \partial_\lambda
\sigma_n(t,\mathbf{x},\lambda)
\end{equation}
Applying Theorem \ref{va} we obtain
$$
\int\rho(\lambda) v_n(t,\mathbf{x},\lambda) d\lambda \to 0 \quad\text{in }
L^1_{\rm loc}(\mathbb{R}^+\times \mathbb{R}^d),
$$
for any test function $\rho\in L^2_c(\mathbb{R}^m)$.
Since $v_n=h_n-h$, and $h_n(t,\mathbf{x},\lambda)= \operatorname{sgn}(u_n-\lambda)$,
we conclude that for every $l>0$,
$$
\int_{-l}^{l} h_n(t,\mathbf{x},\lambda)d\lambda=2
T_l(u_n)(t,\mathbf{x}),
$$
is strongly $L^1_{\rm loc}(\mathbb{R}^+\times \mathbb{R}^d)$ precompact itself.
Applying Corollary \ref{DHM-1}, the statement follows directly.
\end{proof}

\begin{theorem} \label{existence}
Assume that the coefficients of \eqref{d-p} satisfy the non-degeneracy condition 
\eqref{non-deg}.
Then there exists a weak solution to \eqref{d-p} augmented with
the initial condition $u|_{t=0}=u_0 \in L^1 \cap L^p(\mathbb{R}^d)$, $p>1$. 
The weak solution is at the same time the quasi-solution to \eqref{d-p}.
\end{theorem}

\begin{proof}
It is sufficient to consider the regularization of problem \eqref{d-p}:
\begin{gather} \label{reg-eq} 
\partial_t u_n+\operatorname{div}_x f^n(t,\mathbf{x},u_n)
=s^n(t,\mathbf{x},u_n)\!+\!\frac{1}{n}\Delta u_n\\
\label{reg-ic} 
u_n|_{t=0}=u_0(x)\in L^1\cap L^p(\mathbb{R}^d),
\end{gather} 
where $f^n$ and $s^n$ are smooth
regularization in variables $t, x$ of the functions $f$ and $s$ respectively 
given by the convolution with a smooth, compactly supported, non-negative 
kernel $\omega$ with total mass one.
$$
f^n(t,\mathbf{x},\lambda)=f(\cdot,\cdot,\lambda)\star_{t,\mathbf{x}} n^{d+1} 
\omega_n(t,\mathbf{x}), \quad
 s^n(t,\mathbf{x},\lambda)=s(\cdot,\cdot,\lambda)\star_{t,\mathbf{x}} n^{d+1}
 \omega_n(t,\mathbf{x}),
$$ 
where $\omega_n(t,\mathbf{x})=\omega(n t, n\mathbf{x})$.

The existence of smooth $L^1$-solution to \eqref{reg-eq}, \eqref{reg-ic} 
can be found  in \cite{LSU}. Let us prove that the sequence of such solutions 
$(u_n)$ is bounded in $L_{\rm loc}^p(\mathbb{R}^+\times \mathbb{R}^d)$ 
(locally with respect to $t\in \mathbb{R}^+$).

Take the function $S$ given in (iv) and multiply \eqref{reg-eq} by $S'(u)$.
 After standard manipulations, we reach 
\begin{align*}
&\partial_t S(u_n) +\operatorname{div}\Big(\int_0^{u_n} S'(\lambda) 
 \partial_\lambda f^n(t,\mathbf{x},\lambda) d\lambda \Big) 
 + \int_0^{u_n} S''(\lambda)\operatorname{div}f^n(t,\mathbf{x},\lambda) d\lambda\\
&=s^n(t,\mathbf{x},u_n) S'(u_n) +\frac{1}{n}\Delta S(u_n)
 - \frac{1}{n} S''(u_n) |\nabla u_n|^2.
\end{align*} 
Integrating this over $[0,t]\times \mathbb{R}^d$ and using \eqref{1} and \eqref{3}, 
we see that
\begin{equation}\label{lp-bound}
\begin{aligned}
&\int_{\mathbb{R}^d} S(u_n(t,\mathbf{x})) d\mathbf{x} 
 +\frac{1}{n} \int_0^t \int_{\mathbb{R}^d} S''(u_n) |\nabla u_n|^2 d\mathbf{x} dt' \\
&\leq \int_{\mathbb{R}^d} S(u_0(\mathbf{x})) d\mathbf{x}+ C_1+C_2,
\end{aligned}
\end{equation} and the $L^p$-bound follows from the condition \eqref{2}.
We remark that  we also have
 $\frac{1}{n} \int_0^t \int_{\mathbb{R}^d}  |\nabla u_n|^2 d\mathbf{x} dt' \leq  C<\infty$ 
since $S$ is convex.

Next, remark that  for any $M>0$ and 
$K\subset\subset \mathbb{R}^+\times \mathbb{R}^d$, and any $q<r$, we have
\begin{equation}\label{app} 
\|\sup_{\lambda\in [-M,M]}
|f^n(\cdot,\lambda)-f(\cdot,\lambda)| \|_{L^q(K)}\to 0,
\end{equation} 
as $n\to \infty$. Indeed, since $f$ is continuous with respect to $\lambda$, 
for any fixed $(t,\mathbf{x}) \in K$ there exists $\lambda(t,\mathbf{x}) \in [-M,M]$ 
such that
$$
\sup_{\lambda\in [-M,M]}
|f^n(t,\mathbf{x},\lambda)-f(t,\mathbf{x},\lambda)|
=|f^n(t,\mathbf{x},\lambda(t,\mathbf{x}))-f(t,\mathbf{x},\lambda(t,\mathbf{x}))|.
$$ 
Then,  according to the definition of the convolution, we have
\begin{equation} \label{app1}
\begin{aligned}
&\|\sup_{\lambda\in [-M,M]}
|f^n(\cdot,\lambda)-f(\cdot,\lambda)| \|^q_{L^q(K)}\\
&=\int_{K} \Big|\int_{\mathbb{R}^{d+1}}f(\tau,\mathbf{y},\lambda(t,\mathbf{x}))
 \omega_n(t-\tau,\mathbf{x}-\mathbf{y}) d\tau d\mathbf{y}- f(t,\mathbf{x},\lambda(t,\mathbf{x}))
 \Big|^q dt d\mathbf{x}  \\
&=\int_{K} \Big|\int_{\mathbb{R}^{d+1}}(f(t+\frac{1}{n} \eta,\mathbf{x}
 +\frac{1}{n} \mathbf{z},\lambda(t,\mathbf{x}))- f(t,\mathbf{x},\lambda(t,\mathbf{x})) )
 \omega(\eta,\mathbf{z}) d\tau d\mathbf{z} \Big|^p dt d\mathbf{x} \\
& \leq \int_{K} \Big|\int_{\mathbb{R}^{d+1}}(f(t+\frac{1}{n} \eta,\mathbf{x}
 +\frac{1}{n} \mathbf{z},\lambda(t+\frac{1}{n} \eta,\mathbf{x}+\frac{1}{n} \mathbf{z})) \\
& \quad- f(t,\mathbf{x},\lambda(t,\mathbf{x})) )\omega(\eta,\mathbf{z}) d\tau d\mathbf{z}
 \Big|^q dt d\mathbf{x}  \\
&\quad +\int_{K} \Big|\int_{\mathbb{R}^{d+1}}(f(t+\frac{1}{n} \eta,\mathbf{x}
 +\frac{1}{n} \mathbf{z},\lambda(t+\frac{1}{n} \eta,\mathbf{x}
 +\frac{1}{n} \mathbf{z})) \\
& \quad - f(t+\frac{1}{n} \eta,\mathbf{x}+\frac{1}{n} \mathbf{z},\lambda(t,\mathbf{x})) )
 \omega(\eta,\mathbf{z}) d\tau d\mathbf{z} \Big|^q dt d\mathbf{x}.
\end{aligned}
\end{equation}
Now, since $f(t,\mathbf{x},\lambda(t,\mathbf{x})) \in L^r(K)$,
it follows that for almost every $(t,\mathbf{x}) \in K$ and
$(\tau,\mathbf{z}) \in {\rm supp}\, \omega$  it holds
$f(t+\frac{1}{n} \eta,\mathbf{x}+\frac{1}{n} \mathbf{z},\lambda(t+\frac{1}{n} \eta,\mathbf{x}
+\frac{1}{n} \mathbf{z})) \to f(t,\mathbf{x},\lambda(t,\mathbf{x}))$ pointwise
(since almost every point of $L^1_{\rm loc}$ function is the Lebesgue one).
The same holds for the bounded function $\lambda(t,\mathbf{x})$,
such implying the following almost everywhere convergence as  $n \to \infty$:
\begin{align*}
&|f(t+\frac{1}{n} \eta,\mathbf{x}+\frac{1}{n} \mathbf{z},\lambda(t+\frac{1}{n} \eta,\mathbf{x}
+\frac{1}{n} \mathbf{z}))- f(t+\frac{1}{n} \eta,\mathbf{x}
 +\frac{1}{n} \mathbf{z},\lambda(t,\mathbf{x}))| \\
& \leq \sup_{|\lambda| <M} |\partial_\lambda f(t+\frac{1}{n} \eta,\mathbf{x}
+\frac{1}{n} \mathbf{z},\lambda)| \, |\lambda(t+\frac{1}{n} \eta,\mathbf{x}
+\frac{1}{n} \mathbf{z})-\lambda(t,\mathbf{x})| \to 0.
\end{align*}
Therefore, the Lebesgue dominated convergence theorem applied in \eqref{app1}
provides \eqref{app}.

In the rest of the proof, we shall show that conditions of Theorem \ref{main-tvr} 
are satisfied for $(u_n)$.
By multiplying \eqref{reg-eq} by $\operatorname{sgn}(u_n-\lambda)$, 
it is easy to see that the sequence $(u_n)$ satisfies the  entropy
inequality
\begin{align*}
&\partial_t |u_n-\lambda|+\sum_{k=1}^d \partial_{x_k}
\Big(\operatorname{sgn}(u_n-\lambda)(f^n_k(t,\mathbf{x},u_n)
 -f^n_k(t,\mathbf{x},\lambda))\Big)\\
&+\frac{1}{n}\sum_{k}^d\partial^2_{x_k}|u_n-\lambda|
  -\operatorname{sgn}(u_n-\lambda)s^n(t,\mathbf{x},u_n)
+\sum_{k=1}^d \operatorname{sgn}(u_n-\lambda)f^n_{k,x_k}(t,\mathbf{x},u_n)  \leq 0.
\end{align*}
By the Schwartz lemma on non-negative distributions,
\begin{equation} \label{rel-for-prec}
\begin{aligned}
&\partial_t |u_n-\lambda|+\sum_{k=1}^d
\partial_{x_k}\Big(\operatorname{sgn}(u_n-\lambda)(f_k(t,\mathbf{x},u_n)
-f_k(t,\mathbf{x},\lambda))\Big) \\
&= \gamma_n(t,\mathbf{x},\lambda)
 - \sum_{k=1}^d \partial_{x_k}
\Big(\operatorname{sgn}(u-\lambda)(f^n_k(t,\mathbf{x},u_n)-f_k(t,\mathbf{x},u_n)\\
&\quad +f_k(t,\mathbf{x},\lambda)-f^n_k(t,\mathbf{x},\lambda))\Big)
-\frac{1}{n}\sum_{k}^d\partial^2_{x_k}|u_n-\lambda| \\
&\quad -\sum_{k=1}^d \operatorname{sgn}(u_n-\lambda)f^n_{k,x_k}(t,\mathbf{x},u_n)
+\operatorname{sgn}(u_n-\lambda)s^n(t,\mathbf{x},u_n),
\end{aligned}
\end{equation}
where $(\gamma_n)$ is a bounded sequence in
$\mathcal{M}(\mathbb{R}^+\times\mathbb{R}^d\times\mathbb{R})$.
In the sequel we shall prove  that every term of the right-hand side of
\eqref{rel-for-prec} is strongly precompact in
$W_{\rm loc}^{-1,q}(\mathbb{R}^+\times \mathbb{R}^d\times\mathbb{R})$ for every
$q\in [1,\frac{d+1}{d})$.
Let us fix here such a $q$ and use it throughout this proof.

Indeed,  $(\gamma_n)$ is bounded in 
$\mathcal{M}(\mathbb{R}^+\times\mathbb{R}^d\times\mathbb{R}))$, and same holds 
for the penultimate term  due to conditions from (iii) on $\operatorname{div}f$.
The compactness for measures result (e.g. \cite[Theorem 1.6]{Evans}) 
provides that both terms are strongly precompact in 
$W_{\rm loc}^{-1,q}(\mathbb{R}^+\times \mathbb{R}^d\times\mathbb{R})$.

To proceed, fix relatively compact $K\Subset \mathbb{R}^+\times \mathbb{R}^d$. 
Then, for an arbitrary fixed $\varepsilon>0$ find $M>0$ such that
\begin{equation}
\label{M-assumpt}
\operatorname{meas}(\{|u_n|>M \} \cap K) \leq \varepsilon \quad \text{for any } n\in \mathbb{N}.
\end{equation} 
Such a choice is possible because of Lemma \ref{trunc-l}, more precisely 
relation \eqref{conv1} and the fact that every $L^p$, $p>1$, bounded sequence 
is equi-integrable.

Now, we can prove that  for every $k=1,\dots,d$,
\begin{equation} \label{f-conv1}
\|f_k^n(t,\mathbf{x},u_n)-f_k(t,\mathbf{x},u_n)\|_{L^q(K)}   \to 0 \quad
\text{as } n\to \infty.
\end{equation}
First notice that
\begin{equation} \label{f-conv2}
\begin{aligned}
&\|f_k^n(t,\mathbf{x},u_n)-f_k(t,\mathbf{x},u_n)\|_{L^q(K)} \\
& \leq \|(f_k^n(t,\mathbf{x},u_n)-f_k(t,\mathbf{x},u_n))
 \|_{L^q(K \cap \{|u_n|>M \})}\\
&\quad  + \|\sup_{\lambda \in[-M,M]}(f_k^n(t,\mathbf{x},\lambda)
 -f_k(t,\mathbf{x},\lambda)) \|_{L^q(K)}.
\end{aligned}
\end{equation}
 Then, according to \eqref{M-assumpt},   by the H\"older inequality we have
\begin{align*}
&\|(f_k^n(t,\mathbf{x},u_n)-f_k(t,\mathbf{x},u_n)) \|_{L^q(K \cap \{|u_n|>M \})} \\
& \leq  \|(f_k^n(t,\mathbf{x},u_n)-f_k(t,\mathbf{x},u_n))
 \|_{L^r(K \cap \{|u_n|>M \})} \, \operatorname{meas}(\{|u_n|>M \}
 \cap K)^{\frac{r-q}{r q}} \\
& \leq 2 \varepsilon^{\frac{r-q}{r q}} \|\sup_{\lambda \in \mathbb{R}}
 f(\cdot,\cdot,\lambda)\|_{L^r(\mathbb{R}^+\times \mathbb{R}^d)}.
\end{align*}
Now, according to \eqref{app}
$$
\|\sup_{\lambda \in
[-M,M]}(f_k^n(t,\mathbf{x},\lambda)-f_k(t,\mathbf{x},\lambda)) \|_{L^p(K)}
\to 0 \quad \text{as }  n\to \infty.
$$
From the last two estimates, \eqref{f-conv2}, and arbitrariness of $\varepsilon$,
 we conclude that \eqref{f-conv1} holds.
This implies
$$
\sum_{k=1}^d \partial_{x_k}\left(\operatorname{sgn}(u-\lambda)(f^n_k(t,\mathbf{x},u_n)
-f_k(t,\mathbf{x},u_n)+f_k(t,\mathbf{x},\lambda)-f^n_k(t,\mathbf{x},\lambda))\right)
 \to 0
$$
as $n\to\infty$ in $W^{-1,q}_{\rm loc}(\mathbb{R}^+\times \mathbb{R}^d \times
\mathbb{R})$.

Also, according to \eqref{2},  for any relatively compact
$K\subset \subset \mathbb{R}^+\times \mathbb{R}^d$, the sequence of functions
$(\operatorname{sgn}(u_n-\lambda)s^n(t,x,u_n))$ satisfies
$$
\sup_n \|s^n(t,\mathbf{x},u_n)\|_{L^1(K)} <\infty,
$$
i.e. $s^n(t,\mathbf{x},u_n) \in \mathcal{M}(\mathbb{R}^+\times \mathbb{R}^d
\times \mathbb{R})$, and thus it is strongly precompact in
$W_{\rm loc}^{-1,q}(\mathbb{R}^+\times \mathbb{R}^d\times \mathbb{R})$.

Next, according to (iii)  the term
$$
\sum_{k=1}^d \operatorname{sgn}(u_n-\lambda)f^n_{k,x_k}(t,\mathbf{x},u_n)
=\operatorname{sgn}(u-\lambda) \operatorname{div}
f^n(t,\mathbf{x},\lambda)|_{\lambda=u_n}
$$ 
s bounded in $\mathcal{M}(\mathbb{R}^+\times \mathbb{R}^d \times \mathbb{R})$ 
and thus precompact in 
$W_{\rm loc}^{-1,q}(\mathbb{R}^+\times \mathbb{R}^d\times\mathbb{R})$.

Finally, the sequence $(\frac{1}{n}\Delta |u_n-\lambda|)$ strongly converges
to zero as $n\to \infty$ in 
$W^{-1,2}_{\rm loc}(\mathbb{R}^+\times \mathbb{R}^d\times \mathbb{R})$ since, 
By the energy inequality $\sup_n \|\frac{1}{\sqrt{n}}\nabla
u_n\|_{L^2(\mathbb{R}^+\times \mathbb{R}^d)}<\infty$ (see \eqref{lp-bound}) 
implying that
$\frac{1}{\sqrt{n}} \big( \operatorname{sgn}(u_n-\lambda) 
\frac{1}{\sqrt{n}} \nabla u_n \big) \to 0$ in
$L^2_{\rm loc}(\mathbb{R}^+\times \mathbb{R}^d \times \mathbb{R})$.
Therefore
\[
\frac{1}{n}\Delta |u_n-\lambda|= \operatorname{div}\Big(
\frac{1}{\sqrt{n}}
\big( \operatorname{sgn}(u_n-\lambda) \frac{1}{\sqrt{n}} \nabla u_n \big)\Big)
\]
 converges strongly to zero in 
$W^{-1,2}_{\rm loc}(\mathbb{R}^+\times \mathbb{R}^d \times \mathbb{R})$ 
and thus it is precompact in 
$W^{-1,q}_{\rm loc}(\mathbb{R}^+\times \mathbb{R}^d \times \mathbb{R})$ 
as well (recall that $q<2$).

Consequently, $(u_n)$ is a sequence of quasi-solutions to \eqref{d-p} 
satisfying assumptions of Theorem \ref{main-tvr}, which implies its  
strong precompactness in $L^1_{\rm loc}(\mathbb{R}^+\times \mathbb{R}^d)$. 
The strong $L^1_{\rm loc}$-limit of $(u_n)$ along a
subsequence represents a weak solution to \eqref{d-p}. 
\end{proof}


\subsection*{Acknowledgements}
This research was partially developed while Martin Lazar was a
Postdoctoral Fellow on the Basque Center for Applied Mathematics
(Bilbao, Spain) within the  NUMERIWAVES FP7-246776 project. 
The work is  partly supported by Croatian Science Foundation under Project 
WeConMApp/HRZZ-9780, by the bilateral Croatian-Montenegro project {\it
Multiscale methods and calculus of variations}, as well as by the DAAD
project {\it Center of Excellence for Applications of Mathematics}.

\begin{thebibliography}{99}

\bibitem{AlM} J.~Aleksic, D.~Mitrovic; 
\emph{On the compactness for two dimensional scalar conservation law with 
discontinuous flux}, Commun. Math. Sci., \textbf{7} (2009), 963--971.

\bibitem{AndrCances} B. Andreianov, C. Canc\`es; 
\emph{Vanishing capillarity solutions of Buckley-Leverett equation with gravity
in two-rocks medium}, Comput. Geosci., \textbf{17}(3) (2013),
  551--572.

\bibitem{AKR} B. Andreianov, K. H. Karlsen, N. H. Risebro; 
\emph{A theory of $L^1$-dissipative solvers for scalar conservation laws
 with discontinuous flux},
Arch. Ration. Mech. Anal., \textbf{201} (2011), 27--86.

\bibitem{AnM} B.~Andreianov, D.~Mitrovic; 
\emph{Entropy conditions for scalar conservation laws with discontinuous 
flux revisited}, Annales Institut Henri Poincare C,
  Analyse Non Lineaire Vol. \textbf{6} (2015), 1307--1335

\bibitem{AM} {N. Antoni\'c, D. Mitrovi\'c}; 
\emph{H-distributions - an extension of the H-measures in $L^p-L^q$ setting},
 Abstr. Appl. Anal., \textbf{2011} (2011), 12 pp.

\bibitem{ALjmaa} N.~Antoni\'c, M.~Lazar; 
\emph{H-measures and variants applied to parbolic equations},
  J. Math.~Anal. Appl., \textbf{343} (2008), 207--225.

\bibitem{ALrwa} N. Antoni\'{c}, M. Lazar;
\emph{Parabolic variant of H-measures in homogenisation of a model 
problem based on Navier-Stokes equation},
{\it Nonlinear Anal.~B: Real World Appl.}, \textbf{11} (2010), 4500--4512.

\bibitem{BK} M. Bendahmane, K.~H.~Karlsen; 
\emph{Renormalized entropy solutions for quasi-linear anisotropic 
degenerate parabolic equations}, Siam J. Math. Anal., \textbf{36} (2004), 405-422.

\bibitem{BKT2009-multilanes} R. Burger, K. H. Karlsen, J. Towers; 
\emph{A conservation law with discontinuous flux modelling traffic
flow with abruptly changing road surface conditions.} in Hyperbolic
problems: theory, numerics and applications, 455-464, Proc. Sympos.
Appl. Math. 67, Part 2, Amer. Math. Soc. Providence, 2009.

\bibitem{BurgerEtAl} R. Burger, K. H. Karlsen, J. Towers; 
\emph{A model of continuous sedimentation of flocculated
suspensions in clarifier-thickener units},
 SIAM J. Appl. Math. ,\textbf{65}(3) (2005),  882--940.

\bibitem{crasta1} G. Crasta, V. De Cicco, G. De Philippis;
\emph{Kinetic formulation and uniqueness for scalar conservation laws
 with discontinuous flux},
Comm. in PDEs, \textbf{40} (2015), 694--726,

\bibitem{crasta2} G. Crasta, V. De Cicco, G. De Philippis, F. Ghiraldin;
\emph{Structure of solutions of multidimensional conservation laws with 
discontinuous flux and applications to uniqueness},
to appear in Arch. Ration. Mech. Anal.

\bibitem{D} A.~L.~D'Alibard; 
\emph{Kinetic formulation for heterogeneous scalar conservation laws}, 
Annales Institut Henri Poincare C,  Analyse Non Lineaire Vol. \textbf{23} (2006), 
475--498.

\bibitem{DLR}  B. Dehman, M. L\'eautaud, J. Le Rousseau;
\emph{Controllability of two coupled wave equations on a compact manifold}.
 Arch. Rational Mech. Anal., \textbf{211(1)} (2014) 113--187.

\bibitem{Die1}  S. Diehl; 
\emph{On scalar conservation law with point source and discontinuous flux function 
modelling continuous sedimentation},
 SIAM J. Math. Anal., \textbf{6} (1995), 1425--1451.

\bibitem{Die2} S. Diehl; 
\emph{A conservation law with point source and
discontinuous flux function modelling continuous sedimentation},
SIAM J. Appl. Anal., \textbf{2} (1996), 388--419.

\bibitem{8} R. J. Diperna, P. L. Lions; 
\emph{Global Solutions of Boltzmann Equations and the Entropy
Inequality}, Arch. Rat. Mech. Anal., \textbf{114} (1991), 47--55.

\bibitem{13}  R. J. DiPerna, P. L. Lions, Y. Meyer; 
\emph{$L^p$ regularity of velocity averages}, Ann. Inst. H. Poincar\' e Anal. 
Non Lin\'eaire, \textbf{8} (1991), 271--287.


\bibitem{DHM} G.~Dolzmann, N.~Hungerbuhler, S.~M\"uller;
\emph{Nonlinear elliptic systems with measure valued right-hand side}, 
Math. Zeitschrift, \textbf{226}, (1997) 545--574.

\bibitem{18} F. Golse, L. Saint-Raymond; 
\emph{Velocity averaging in $L^1$ for the transport equation},
C. R. Acad. Sci. Paris S\'er. I Math., \textbf{334} (2002), 557--562.

\bibitem{Evans} L. C. Evans; 
\emph{Weak Convergence Methods for Nonlinear Partial
Differential Equations},  Regional Conference Series in Mathematics,
No. 74. Conference Board of the Mathematical Sciences, 1990.

\bibitem{Ger} P. G\'erard; 
\emph{Microlocal Defect Measures}, Comm. Partial Differential
Equations \textbf{16} (1991), 1761--1794.

\bibitem{Gra} L.~Grafakos; 
\emph{Classical Fourier Analysis},
Graduate Text in Mathematics 249,  Springer Science and Business
Media, 2008

\bibitem{Kaas} E. Kaasschieter; 
\emph{Solving the Buckley-Leverret equation with
gravity in a heterogeneous porous media}, Comput. Geosci. \textbf{3}
(1999), 23--48.

\bibitem{KT} K.~H.~ Karlsen, M.~ Rascle,  E.~ Tadmor; 
\emph{On the existence and compactness of a two-
dimensional resonant system of conservation laws}, 
Commun. Math. Sci. \textbf{5} (2007), 253--265.

\bibitem{KRT} K. H. Karlsen, N. H. Risebro, J. Towers; 
\emph{$L^1$-stability for entropy solutions of nonlinear degenerate 
parabolic connection-diffusion equations with discontinuous coefficients}, 
Skr. K. Nor. Vid. Selsk, \textbf{3} (2003), 1--49.

\bibitem{Kru} S. N. Kruzhkov; 
\emph{First order quasilinear equations in several independent variables},
Mat. Sb. \textbf{81} (1970), 217-243.

\bibitem{LSU} O.~A.~Ladyzhenskaja, V.~A.~Solonnikov, N.~N.~Ural'ceva, (1968); 
\emph{Linear and quasi-linear equations of parabolic type}, Translations 
of Mathematical Monographs 23, Providence, RI: American Mathematical Society, 
pp. XI+648

\bibitem{LM2} M.~Lazar, D.~Mitrovic; 
\emph{Velocity averaging -- a general framework}, Dynamics of PDEs, 
\textbf{9} (2012), 239-260.

\bibitem{LM_crass} M. Lazar, D. Mitrovi\'c; 
\emph{On an extension of a bilinear functional on $L^p(\mathbb{R}^d)\times E$
to Bochner spaces with an application on velocity averaging}, 
C. R. Acad. Sci. Paris S\'er. I Math.  \textbf{351} (2013), 261--264.

\bibitem{LZ}  M. Lazar, E. Zuazua;
\emph{Averaged control and observation of parameter-depending wave equations},
  C. R. Acad. Sci. Paris, Ser. I, \textbf{352} (2014) 497--502.

\bibitem{LPT}  P.~L.~Lions, B.~Perthame, E.~Tadmor; 
\emph{A kinetic formulation of multidimensional scalar conservation law 
and related equations}, J. Amer. Math. Soc., \textbf{7}, (1994), 169--191.

\bibitem{MM} M.~Mi\v{s}ur, D.~Mitrovi\'c; 
\emph{On a generalization of compensated compactness in the $L^p-L^q$ setting}, 
Journal of Functional Analysis, \textbf{268} (2015), 1904--1927.


\bibitem{Pan}  E. Yu. Panov; 
\emph{Existence of strong traces for generalized solutions of multidimensional 
scalar conservation laws.},
J. Hyperb. Differ. Equ., \textbf{2} (2005), 885--908.

\bibitem{pan_arma} E.~Yu.~Panov; 
\emph{Existence and strong pre-compactness properties for entropy solutions
of a first-order quasilinear equation with discontinuous flux},
Arch. Rational Mech. Anal. \textbf{195} (2010) 643-673.

\bibitem{32}  B.\,Perthame, P.\,Souganidis; 
\emph{A limiting case for velocity averaging}, Ann.
Sci. Ec. Norm. Sup. \textbf{4} (1998), 591--598.

\bibitem{36} T.\,Tao, E.\,Tadmor; 
\emph{Velocity Averaging, Kinetic Formulations,
and Regularizing Effects in Quasi-Linear Partial Differential
Equations}, Comm. Pure Appl. Math. \textbf{60} (2007), 1488--1521.

\bibitem{Tar} L.~Tartar;
 \emph{H-measures, a new approach for studying homogenisation,
oscillation and concentration effects in PDEs},
Proc.~Roy.~Soc.~Edinburgh Sect. A, \textbf{115} (1990) 193--230.

\end{thebibliography}

\end{document}
