\documentclass[reqno]{amsart}
\usepackage{hyperref}
\usepackage{amssymb}

\AtBeginDocument{{\noindent\small
\emph{Electronic Journal of Differential Equations},
Vol. 2011 (2011), No. 10, pp. 1--10.\newline
ISSN: 1072-6691. URL: http://ejde.math.txstate.edu or http://ejde.math.unt.edu
\newline ftp ejde.math.txstate.edu}
\thanks{\copyright 2011 Texas State University - San Marcos.}
\vspace{8mm}}

\begin{document}
\title[\hfilneg EJDE-2011/10\hfil Vector-valued Morrey theorem]
{Vector-valued Morrey's embedding theorem and H\"older continuity
in parabolic problems}

\author[P. J. Rabier\hfil EJDE-2011/10\hfilneg]
{Patrick J. Rabier}

\address{Patrick J. Rabier \newline
Department of Mathematics,
University of Pittsburgh, Pittsburgh, PA 15260, USA}
\email{rabier@imap.pitt.edu}

\thanks{Submitted January 11, 2011. Published January 19, 2011.}
\subjclass[2000]{46E35, 46E40, 35K90, 35K55}
\keywords{Morrey's theorem; embedding; vector-valued
Sobolev space; \hfill\break\indent mixed norm}

\begin{abstract}
 If $I\subset \mathbb{R}$ is an open interval and 
 $\Omega \subset \mathbb{R}^N$ an open subset with $\partial \Omega $ 
 Lipschitz continuous,  we show that the space 
 $W^{1,p}(I,L^q (\Omega ))\cap L^p(I,W^{1,q}(\Omega ))$ is continuously 
 embedded in $C^{0,\frac{1}{p'}-\frac{N}{q}}(\overline{\Omega \times
  I})\cap L^{\infty }(\Omega \times I)$ if $p,q\in (1,\infty )$ and
 $q>Np'$. When $p=q$, this coincides with Morrey's
 embedding theorem for $W^{1,p}(\Omega \times I)$. While weaker
 results have been obtained by various methods, including very
 technical ones, the proof given here follows that of Morrey's
 theorem in the scalar case and relies only on widely known
 properties of the classical Sobolev spaces and of the Bochner
 integral.

 This embedding is useful to formulate nonlinear evolution problems
 as functional equations, but it has other applications. As an
 example, we derive apparently new space-time H\"older continuity
 properties for $u_t=Au+f,u(\cdot ,0)=u_0$ when $A$ generates a
 holomorphic semigroup on $L^q (\Omega)$.
\end{abstract}

\maketitle
\numberwithin{equation}{section}
\newtheorem{theorem}{Theorem}[section]
\newtheorem{lemma}[theorem]{Lemma}
\newtheorem{remark}[theorem]{Remark}

\section{Introduction}

Let $I\subset \mathbb{R}$ be an open interval and
$\Omega \subset \mathbb{R}^N$
an open subset satisfying the strong local Lipschitz condition
\cite[p. 83]{AdFo03}. For $p,q\in (1,\infty )$, we set
\begin{equation*}
\mathcal{V}^{p,q}(I,\Omega ):=W^{1,p}(I,L^q (\Omega ))\cap
L^p(I,W^{1,q}( \Omega )),
\end{equation*}
a Banach space for the natural norm. Subspaces of
$\mathcal{V}^{p,q}(I, \Omega )$ arise naturally in evolution
problems and have become especially important due to recent
progress in the so-called ``$L^p$ maximal regularity'' issue and
related topics. In such problems, the space of interest is
$W^{1,p}(I,L^q (\Omega ))\cap L^p(I,D(A))$, where $A$ is an
unbounded linear operator on $L^q (\Omega )$ with domain $D(A)$
usually contained in $W^{1,q}(\Omega )$ or even $W^{k,q}(\Omega )$
with $k\geq 2$. See for instance Denk \textit{et al.}
\cite{DeHiPr03} and the references therein, or Arendt and Bu
\cite{ArBu02} and Arendt and Rabier \cite{ArRa08} for the
time-periodic case.

Every $u\in \mathcal{V}^{p,q}(I,\Omega )$ can be identified with a
measurable real-valued function $u(x,t)$ (further details below). The
primary goal of this note is to give a simple proof of the following
embedding theorem:

\begin{theorem} \label{thm1}
If $p,q\in (1,\infty )$ and $\frac{1}{p}+\frac{N}{q}<1$
(i.e., $ q>Np'$ with $p':=\frac{p}{p-1}$), then
$\mathcal{V} ^{p,q}(I,\Omega )\hookrightarrow
C^{0,\frac{1}{p'}-\frac{N}{q}}( \overline{\Omega \times
I})\cap L^{\infty }(\Omega \times I)$. In particular, if $I$ and
$\Omega $ are bounded, the embedding $\mathcal{V} ^{p,q}(I,\Omega
)\hookrightarrow C^{0,\alpha }(\overline{\Omega \times I})$ is
compact for every $0\leq \alpha <\frac{1}{p'}-\frac{N}{q}$.
\end{theorem}

When $p=q$, then $\mathcal{V}^{p,p}(I,\Omega )=W^{1,p}(\Omega
\times I)$, so that Morrey's theorem is recovered. The latter is
of notorious importance in elliptic PDEs, especially for the
formulation of nonlinear problems as functional equations. In
particular, it is often used to prove the definiteness of
Nemytskii operators. Theorem \ref{thm1} can be used in the same way
in problems of parabolic type, now that the relevance of
$\mathcal{V }^{p,q}(I,\Omega )$ and its subspaces is becoming more
widely known (see Morris \cite{Mo06} when $p=q$). In
time-dependent problems, it is not uncommon to require different
integrability properties in space and time, so that the case
$p\neq q$ has concrete value. In fact, a completely different
application to a regularity question is discussed in Section
\ref{example}, in which $p=q$ would be far too restrictive.

There are many connections between Theorem \ref{thm1} and the
existing literature, starting with the so-called anisotropic
Sobolev spaces with mixed norm. First, recall that
$L^p(I,L^q (\Omega ))$ is isometrically isomorphic to
$L^{\mathbf{r}}(\Omega \times I)$ where $\mathbf{r} :=(q,\dots
,q,p)\in (1,\infty )^{N+1}$ (see Besov \textit{et al}. \cite[p.
7]{BeIlNi78} for the definition of $L^{\mathbf{r}}(G)$; the
definition includes measurability on $G$). A proof can be found in
Benedek and Panzone \cite[pp. 318-319]{BePa61}.

It follows readily from the identification of
$L^p(I,L^q (\Omega ))$ and $ L^{\mathbf{r}}(\Omega \times I)$
and from the definitions of the derivatives of scalar and
vector-valued distributions (\cite{Sc57}) that if $u\in
W^{1,p}(I,L^q (\Omega ))$, the function corresponding to
$\frac{du}{dt}\in L^p(I,L^q (\Omega ))$ in
$L^{\mathbf{r}}(\Omega \times I)$ is just $u_t$ (the partial
derivative of $u$ as a scalar distribution on $\Omega \times I$ ).
Likewise, if $u\in L^p(I,W^{1,q}(\Omega ))$, the ``spatial''
partial derivative $u_{x_j}$ of $u$ as a scalar distribution on
$\Omega \times I$ corresponds to the vector-valued derivative
$u_{x_j}\in L^p(I,L^q (\Omega )),1\leq j\leq N$.

Thus, $u\in \mathcal{V}^{p,q}(I,\Omega )$ if and only if
$u,u_t,u_{x_j}\in L^{\mathbf{r}}(\Omega \times I),1\leq j\leq N$,
when $u$ is viewed as a function of $(x,t)$. In turn, this
characterizes the elements of $W^{\mathbf{1},\mathbf{r}}(\Omega
\times I))$ where $\mathbf{1} :=(1,\dots ,1)\in \mathbb{N}^{N+1}$
(\cite[p. 165]{BeIlNi78}). That $W^{\mathbf{1}, \mathbf{r}}(\Omega
\times I))\hookrightarrow C^0(\overline{\Omega \times I} )\cap
L^{\infty }(\Omega \times I)$ then follows from \cite[Theorem
10.4] {BeIlNi78} for $G=\mathbb{R}^{N+1}$ and the extension
theorem \cite[Theorem 9.6] {BeIlNi78} (the condition $\kappa
<1$ in \cite[Theorem 10.4]{BeIlNi78} is exactly $q>Np'$). If $I$
and $\Omega $ are bounded, the embedding is compact by
\cite[Theorem 26.3.5]{BeIlNi79} (not based on Ascoli's theorem)
and the estimate in \cite[Theorem 10.4]{BeIlNi78}. However, the
only H\"older continuity result, \cite[Theorem 27.4.2, p.
248]{BeIlNi79}, does not allow for $p\neq q$. Therefore, this
quite lengthy and technical approach does no prove Theorem
\ref{thm1} when $p\neq q$ (and when $p=q$, the much simpler
classical theorem of Morrey suffices).

When $I=\mathbb{R}$ and $\Omega =\mathbb{R}^N$, an embedding into
$C^0(\mathbb{R} ^{N+1})$ was proved by Rao \cite{Ra78} for a
related but different space with mixed norm involving a
``half-derivative'' operator instead of $t$ -differentiation,
under the stronger requirement $\frac{2}{p}+\frac{N}{q}<1$. The
proof is by convolution arguments (parabolic Riesz potentials).
Still under the same condition $\frac{2}{p}+\frac{N}{q}<1$ and yet
again by other methods, Pr\"uss \cite{Pr91} and Engler \cite[Lemma
A3]{En96} have obtained other embeddings for the smaller space
$W^{1,p}(I,L^q (\Omega ))\cap L^p(I,W^{2,q}(\Omega )\cap
W_0 ^{1,q}(\Omega ))$.

If both $I$ and $\Omega $ are bounded, a more recent abstract
theorem of Amann \cite[Theorem 1.1]{Am00} yields the compactness
of the embedding of $ \mathcal{V}^{p,q}(I,\Omega )$ into
$C^{0,\alpha }(I,B^{\frac{N}{q} ;q,1}(\Omega ))$ for every $0\leq
\alpha <\frac{1}{p'}-\frac{N}{q}$ (but not for $\alpha
=\frac{1}{p'}-\frac{N}{q}$), where $B^{\frac{N }{q};q,1}(\Omega
):=[L^q (\Omega ),W^{1,q}(\Omega )]_{\frac{N}{q},1}$ is the usual
Besov space. (Specifically, let $E_0 =L^q (\Omega
),E_{1}=W^{1,q}(\Omega ),\theta =\frac{N}{q}$ and
$p_0 =p_{1}=p,s_0 =1$ in Amann's theorem.) Since
$\frac{N}{q}q=N$, it is only true that $B^{\frac{N}{q
};q,1}(\Omega )\hookrightarrow C_{B}^0(\Omega )$ (bounded
continuous functions on $\Omega $; see \cite[p. 231]{AdFo03}), so
that this proves the compactness of the embedding of
$\mathcal{V}^{p,q}(I,\Omega )$ into
$ C^{0,\alpha }(I,C_{B}^0(\Omega ))$, but not even its embedding
into $C^0(\overline{\Omega \times I})$, let alone into any
H\"older space on $ \overline{\Omega \times I}$.

In retrospect, Theorem \ref{thm1} is certainly not surprising.
Yet, it is apparently not so intuitive, if only to judge by the
foregoing review of some of the literature that addresses similar
issues but fails to deliver the same property. In addition, the
proof given here is elementary, insofar as being based only on
widely known results about Sobolev spaces and the Bochner
integral. It follows the standard proof of Morrey's theorem in the
scalar case (\cite{AdFo03}, \cite{Br83}) with of course extra
technicalities.

The case when $I=\mathbb{R}$ and $\Omega =\mathbb{R}^N$ is
discussed first, in the next section. The general case (Section
\ref{proof2}) follows from the existence of a linear extension
operator $\mathcal{V}^{p,q}(I,\Omega )\to
\mathcal{V}^{p,q}(\mathbb{R},\mathbb{R}^N)$. This is well known
when $p=q$ since $\mathcal{V}^{p,p}(I,\Omega )=W^{1,p}(\Omega
\times I)$, but must be proved in general.

As pointed out earlier, Theorem \ref{thm1} is useful to establish
the well-posedness of nonlinear problems of parabolic type in
suitable function spaces, but it also has a direct application to
the H\"older continuity, jointly in space and time, of their
solutions, even in the linear case. This short application is
discussed in Section \ref{example}, first for the autonomous
equation $u_t=Au$, $u(\cdot ,0)=u_0 $, where $A$ is the generator
of a $C_0 $ holomorphic semigroup on $L^q (\Omega )$ (Theorem
\ref{thm2}) and next in greater generality. In spite of a multitude
of related results in the literature, we have found no evidence
that the same H\"older continuity feature has previously been
proved by other arguments.

\section{Proof of Theorem \ref{thm1} for
$\mathcal{V}^{p,q}(\mathbb{R},\mathbb{R}^N)$\label{proof1}}

We begin with the following denseness lemma.

\begin{lemma} \label{lem2}
If $p,q\in [1,\infty )$,  then
$C_0 ^{\infty }( \mathbb{R}^{N+1})$ is dense in
$\mathcal{V}^{p,q}(\mathbb{R},\mathbb{R} ^N)$.
\end{lemma}

\begin{proof}
First, we show that $u\in \mathcal{V}^{p,q}(\mathbb{R},\mathbb{R}^N)$
can be approximated by a sequence
$(u_n)\subset \mathcal{V}^{p,q}(\mathbb{R},\mathbb{R}^N)$ such that
$u_n(x,t)=0$ for $(x,t)$ outside a cube in $\mathbb{R}^{N+1}$.

Indeed, let $\zeta \in C_0 ^{\infty }(\mathbb{R})$ be such that
$\zeta =1$ on $ [-1,1]$ and let $\zeta _n(t):=\zeta (t/n)$.
Then, $\zeta _nu\to u $ in
$\mathcal{V}^{p,q}(\mathbb{R},\mathbb{R}^N)$, so that we may
assume with no loss of generality that $u$ has compact support in
$t$ in the first place.

If $\xi _n(x):=\zeta _n(|x|)$, then $\xi _n\in C_0 ^{\infty
}(\mathbb{R} ^N)$ and $\xi _nu(\cdot ,t)\to u(\cdot ,t)$
in $W^{1,q}(\mathbb{R} ^N)$ for a.e. $t\in \mathbb{R}$. It is
readily checked that there is a constant $C>0$ independent of $t$
and $n$ such that
$\|\xi _nu(\cdot,t)\|_{W^{1,q}(\mathbb{R}^N)}\leq C\|u(\cdot
,t)\|_{W^{1,q}(\mathbb{R}^N)}$. Thus, by dominated convergence,
$\xi _nu\to u$ in $L^p(\mathbb{R},W^{1,q}(\mathbb{R}^N))$.
Similar arguments and the remark that
$\frac{d}{dt} (\xi _nu)=\xi _n\frac{du}{dt}$ show that
$\xi_nu\to u$ in $ W^{1,p}(\mathbb{R},L^q (\mathbb{R}^N))$,
so that $\xi _nu\to u$ in $\mathcal{V}^{p,q}(\mathbb{R},\mathbb{R}^N)$.
This proves the claim since $\xi _nu$ has compact support in $t$
and $x$.

From the above, it suffices to show that if $u\in
\mathcal{V}^{p,q}(\mathbb{R}, \mathbb{R}^N)$ and $u(x,t)=0$
outside a cube in $\mathbb{R}^{N+1}$, then $u$ can be approximated
in $\mathcal{V}^{p,q}(\mathbb{R},\mathbb{R}^N)$ by a sequence from
$C_0 ^{\infty }(\mathbb{R}^{N+1})$. Let $\Theta \in C_0 ^{\infty
}(\mathbb{R }^N)$ be such that $\Theta \geq 0$ and
$\int_{\mathbb{R}^N}\Theta =1$ and let $\Theta _n(x):=n^N\Theta
(nx)$. Set $u_n:=\Theta _n*_{x}u$, where $*_{x}$ denotes
convolution with respect to the $x$ variable.

For a.e. $t\in \mathbb{R}$, we have $u(\cdot ,t)\in
W^{1,q}(\mathbb{R}^N)$ and so $u_n(\cdot ,t)\to u(\cdot
,t)$ in $W^{1,q}(\mathbb{R}^N)$, as is well known. In addition, by
Young's inequality, there is a constant $C>0$ independent of $t$
and $n$ such that $\|u_n(\cdot ,t)\|_{W^{1,q}(\mathbb{R}
^N)}\leq C\|u(\cdot ,t)\|_{W^{1,q}(\mathbb{R}^N)}$. Thus, by
dominated convergence, $u_n\to u$ in
$L^p(\mathbb{R};W^{1,q}(\mathbb{R}^N))$. Similar arguments and
the remark that $\frac{du_n}{dt}=\Theta _n*_{x} \frac{du}{dt}$
show that $u_n\to u$ in
$W^{1,p}(\mathbb{R},L^q (\mathbb{R}^N))$, so that
$u_n\to u$ in $\mathcal{V}^{p,q}(\mathbb{R}, \mathbb{R}^N)$.

Now, let $\theta \in C_0 ^{\infty }(\mathbb{R})$ be such that
$\theta \geq 0$ and $\int_{\mathbb{R}}\theta =1$ and let $\theta
_{m}(t):=m\theta (mt)$. With $ n\in \mathbb{N}$ being fixed, set
$u_{nm}(x,t):=\theta _{m}*_tu_n(x,t)$, where $*_t$ denotes
convolution with respect to the $t$ variable. The standard
properties of convolution imply that $u_{nm}\to u_n$ in
$ W^{1,p}(\mathbb{R};L^q (\mathbb{R}^N))$ and in
$L^p(\mathbb{R};W^{1,q}(\mathbb{R} ^N))$ as $m\to \infty
$, so that $u_{nm}\to u_n$ in $
\mathcal{V}^{p,q}(\mathbb{R},\mathbb{R}^N)$ as $m\to
\infty $. Thus, $ u_{nm}$ is arbitrarily close to $u$ in
$\mathcal{V}^{p,q}(\mathbb{R},\mathbb{R} ^N)$ if $n$ and $m$ are
large enough. On the other hand, $u_{nm}=\theta _{m}*_t(\Theta
_n*_{x}u)=(\theta _{m}\otimes \Theta _n)*u$, where $*$ denotes
convolution with respect to the $(x,t)$ variable and $u$ is
identified with a function of $L^{\mathbf{r}}(\mathbb{R}^N\times
\mathbb{R})$ (see the Introduction). Since $u$ has compact support
(so that $u\in L^1( \mathbb{R}^{N+1})$), the same thing is true
of $u_{nm}$. Furthermore, $u_{nm}$ is $C^{\infty }$ since $\theta
_{m}\otimes \Theta _n$ is $C^{\infty }$.
\end{proof}

The proof of Theorem \ref{thm1} for
$\mathcal{V}^{p,q}(\mathbb{R},\mathbb{R}^N)$
follows at once from Lemma \ref{lem2} and the following result.

\begin{lemma}\label{lem3}
There is a constant $C>0$ such that
\begin{equation}
\|\varphi \|_{L^{\infty }(\mathbb{R}^{N+1})}\leq C\|\varphi
\|_{\mathcal{V} ^{p,q}(\mathbb{R},\mathbb{R}^N)}  \label{1}
\end{equation}
and that
\begin{equation}
|\varphi (x,t)-\varphi (y,s)|\leq
C|(x,t)-(y,s)|^{\frac{1}{p'}- \frac{N}{q}}\|\varphi
\|_{\mathcal{V}^{p,q}(\mathbb{R},\mathbb{R}^N)},  \label{2}
\end{equation}
for every $\varphi \in C_0 ^{\infty }(\mathbb{R}^{N+1})$ and
every $ (x,t),(y,s)\in \mathbb{R}^{N+1}$.
\end{lemma}

\begin{proof}
Let $J_{\rho }\subset \mathbb{R}$ be an interval of length $\rho
>0$ containing $0$ and $Q_{\rho }\subset \mathbb{R}^N$ a cube with
side $\rho $ containing $ 0 $ and parallel to the coordinate axes.
For $(x,t)\in Q_{\rho }\times J_{\rho }$ and $\varphi \in
C_0 ^{\infty }(\mathbb{R}^{N+1})$ and since $ |t|,|x_j|\leq \rho
$,
\begin{equation*}
|\varphi (x,t)-\varphi (0)|\leq \rho \Big( \int_0 ^1|\varphi
_t(\tau x,\tau t)|d\tau +\sum_{j=1}^N\int_0 ^1|\varphi
_{x_j}(\tau x,\tau t)|d\tau \Big) .
\end{equation*}
Upon integrating this inequality over $Q_{\rho }\times J_{\rho }$
and with $ \overline{\varphi }$ denoting the average of $\varphi $
on $Q_{\rho }\times J_{\rho }$, we obtain
\begin{equation} \label{3}
\begin{aligned}
|\overline{\varphi }-\varphi (0)|
&\leq  \rho ^{-N}\int_0 ^1\tau ^{-N-1}d\tau
\Big( \int_{\tau Q_{\rho }\times \tau J_{\rho }}|\varphi _t(y,s)|dy\,ds\\
&\quad +\sum_{j=1}^N\int_{\tau Q_{\rho }\times \tau J_{\rho
}}|\varphi _{x_j}(y,s)|dy\,ds\Big) .
\end{aligned}
\end{equation}
Let $\tau \in [0,1]$ and $s\in \tau J_{\rho }$ be fixed. By
H\"older's inequality,
\begin{align*}
\int_{\tau Q_{\rho }}|\varphi _t(y,s)|dy
&\leq |\tau Q_{\rho }|^{1/q'} \Big( \int_{\tau Q_{\rho }}|\varphi
_t(y,s)|^q dy\Big) ^{1/q}\\
&\leq |\tau Q_{\rho}|^{1/q'}\Big( \int_{Q_{\rho }}|\varphi
_t(y,s)|^q dy\Big) ^{1/q},
\end{align*}
where $|\tau Q_{\rho }|=\tau ^N|Q_{\rho }|=\tau ^N\rho ^N$ is the
measure of $\tau Q_{\rho }$. Thus, by using once more H\"older's
inequality and since $|\tau J_{\rho }|=\tau \rho $,
\begin{align*}
\int_{\tau Q_{\rho }\times \tau J_{\rho }}|\varphi
_t(y,s)|dy\,ds
&\leq \tau ^{ \frac{N}{q'}}\rho
^{\frac{N}{q'}}\int_{\tau J_{\rho
}}ds\Big( \int_{Q_{\rho }}|\varphi _t(y,s)|^q dy\Big) ^{1/q} \\
&\leq \tau ^{\frac{N}{q'}+\frac{1}{p'}}\rho ^{\frac{N}{
q'}+\frac{1}{p'}}\Big( \int_{J_{\rho }}ds\Big( \int_{Q_{\rho
}}|\varphi _t(y,s)|^q dy\Big) ^{p/q}\Big) ^{1/p} \\
&=\tau ^{\frac{N}{q'}+\frac{1}{p'}}\rho ^{\frac{N}{
q'}+\frac{1}{p'}}\|\varphi _t\|_{L^p(J_{\rho },L^q (Q_{\rho
}))}.
\end{align*}

Altogether and since the norm of
$L^p(J_{\rho },L^q (Q_{\rho}))$ is majorized by the norm of
$\mathcal{V}^{p,q}(\mathbb{R},\mathbb{R}^N)$, this yields
$\int_{\tau Q_{\rho }\times \tau J_{\rho }}|\varphi
_t(y,s)|dyds\leq \tau ^{\frac{N}{q'}+\frac{1}{p'}}\rho ^{
\frac{N}{q'}+\frac{1}{p'}}\|\varphi \|_{\mathcal{V}^{p,q}(
\mathbb{R},\mathbb{R}^N)}$.

By the same procedure,
$$
\int_{\tau Q_{\rho }\times \tau J_{\rho
}}|\varphi _{x_j}(y,s)|dyds\leq \tau ^{\frac{N}{q'}+\frac{1}{p'}
}\rho ^{\frac{N}{q'}+\frac{1}{p'}}\|\varphi \|_{\mathcal{V}
^{p,q}(\mathbb{R},\mathbb{R}^N)}
$$
for $1\leq j\leq N$. Thus, by
substitution into \eqref{3} and a straightforward calculation,
\begin{equation}
|\overline{\varphi }-\varphi (0)|\leq C\rho ^{\frac{1}{p'}-\frac{N}{
q}}\|\varphi \|_{\mathcal{V}^{p,q}(\mathbb{R},\mathbb{R}^N)},  \label{4}
\end{equation}
where $C:=\frac{p'q(N+1)}{q-Np'}$.

By translation invariance, \eqref{4} remains valid -with the same
constant $ C $- when $\varphi (0)$ is replaced by any $\varphi
(x,t)$, provided that $ \overline{\varphi }$ now denotes the
average of $\varphi $ over $Q_{\rho }\times J_{\rho }$ and
$J_{\rho }\subset \mathbb{R}$ is any interval of length $ \rho $
containing $t$ while $Q_{\rho }\subset \mathbb{R}^N$ is any cube
with side $\rho $ containing $x$. In particular, by choosing $\rho
=1$ and noticing that, in this case (from H\"older's inequality)
$\overline{ \varphi }\leq \|\varphi \|_{L^1(Q_{1}\times
J_{1})}\leq \|\varphi \|_{L^p(J_{1},L^q (Q_{1}))}\leq \|\varphi
\|_{L^p(\mathbb{R},L^q (\mathbb{R} ^N))}\leq \|\varphi
\|_{\mathcal{V}^{p,q}(\mathbb{R},\mathbb{R}^N)}$, we obtain
$|\varphi (x,t)|\leq C\|\varphi
\|_{\mathcal{V}^{p,q}(\mathbb{R},\mathbb{R} ^N)}$ after changing
$C$ in \eqref{4} into $C+1$. This proves \eqref{1}.

Next, arbitrary pairs $(x,t)$ and $(y,s)$ in $\mathbb{R}^{N+1}$
are contained in the same cube $Q_{\rho }\times J_{\rho }$ with
side $\rho =|(x,t)-(y,s)|$. Thus, \eqref{4} holds with $\varphi
(0)$ replaced by $\varphi (x,t)$ or $ \varphi (y,s)$ (with the
same $\overline{\varphi }$), which implies
$$
 |\varphi (x,t)-\varphi (y,s)|\leq 2C|(x,t)-(y,s)|^{\frac{1}{p'}-
\frac{N}{q}}\|\varphi
\|_{\mathcal{V}^{p,q}(\mathbb{R},\mathbb{R}^N)}.
$$
 This proves \eqref{2} after changing $C$ in \eqref{4} into $2C$.
\end{proof}

It should be obvious that Lemmas \ref{lem2} and \ref{lem3} imply
Theorem \ref {thm1} for
$\mathcal{V}^{p,q}(\mathbb{R},\mathbb{R}^N)$.

\section{Proof of Theorem \ref{thm1}\label{proof2}}

We shall rely on the following extension property:

\begin{lemma}\label{lem4}
Let $p,q\in [1,\infty )$. There is a bounded linear
operator $ \mathcal{E}:\mathcal{V}^{p,q}(I,\Omega )\to
\mathcal{V}^{p,q}(I,\mathbb{R}^N)$ such that
$(\mathcal{E}u)(x,t)=u(x,t)$ for a.e. $(x,t)\in \Omega \times I$.
\end{lemma}

\begin{proof}
From the assumptions about $\Omega $ and by the Stein extension
theorem (\cite{AdFo03}, \cite{St70}), there is an extension
operator $E\in \mathcal{L} (L^q (\Omega ),L^q (\mathbb{R}^N))$
mapping $W^{1,q}(\Omega )$ into $ W^{1,q}(\mathbb{R}^N)$ and such
that $E\in \mathcal{L}(W^{1,q}(\Omega ),W^{1,q}(\mathbb{R}^N))$.
Given $u\in L^p(I,L^q (\Omega ))$, set
\begin{equation*}
(\mathcal{E}u)(\cdot ,t):=Eu(\cdot ,t),\quad \text{a.e. }t\in I,
\end{equation*}
so that $\mathcal{E}u:I\to L^q (\mathbb{R}^N)$ is
strongly measurable and that $\|(\mathcal{E}u)(\cdot
,t)\|_{L^q (\mathbb{R}^N)}\leq C_0 \|u(\cdot ,t)\|_{L^q (\Omega
)}$ for a.e. $t\in I$, where $
C_0 :=\|E\|_{\mathcal{L}(L^q (\Omega ),L^q (\mathbb{R}^N))}$ is
independent of $u$ and $t$. Thus, $\mathcal{E}u\in
L^p(I,L^q (\mathbb{R} ^N))$ and
$\|\mathcal{E}u\|_{L^p(I,L^q (\mathbb{R}^N))}\leq
C_0 \|u\|_{L^p(I,L^q (\Omega ))}$.

As noted in the Introduction, $u$ is measurable on $\Omega \times
I$ and $ \mathcal{E}u$ is measurable on $\mathbb{R}^N\times I$. As
a result, $v(x,t):=( \mathcal{E}u)(x,t)-u(x,t)$ is measurable on
$\Omega \times I$. Since also $ v(x,t)=0$ for $t\notin S$ and
$x\notin \Sigma _t$ where both $S\subset I$ and $\Sigma _t\subset
\Omega $ have measure $0$, it follows from Tonelli's theorem that
$v=0$ a.e. in $\Omega \times I$, so that $(\mathcal{E}
u)(x,t)=u(x,t)$ for a.e. $(x,t)\in \Omega \times I$.

If now $u\in L^p(I,W^{1,q}(\Omega ))$, then $Eu(\cdot ,t)\in
W^{1,q}(\mathbb{R }^N)$, i.e., $(\mathcal{E}u)(\cdot ,t)\in
W^{1,q}(\mathbb{R}^N)$ and $\|( \mathcal{E}u)(\cdot
,t)\|_{W^{1,q}(\mathbb{R}^N)}\leq C_{1}\|u(\cdot
,t)\|_{W^{1,q}(\Omega )}$ where
$C_{1}:=\|E\|_{\mathcal{L}(W^{1,q}(\Omega
),W^{1,q}(\mathbb{R}^N))}$, so that $\mathcal{E}u\in
L^p(I,W^{1,q}(\mathbb{R} ^N))$ with
$\|\mathcal{E}u\|_{L^p(I,W^{1,q}(\mathbb{R}^N))}\leq
C_{1}\|u\|_{L^p(I,W^{1,q}(\Omega ))}$. This shows that
$\mathcal{E}\in \mathcal{L}(L^p(I,L^q (\Omega
)),L^p(I,L^q (\mathbb{R}^N)))$ is an extension operator
continuously mapping $L^p(I,W^{1,q}(\Omega ))$ into $
L^p(I,W^{1,q}(\mathbb{R}^N))$.

In particular, if $u\in \mathcal{V}^{p,q}(I,\Omega )$, then
$\mathcal{E}u\in L^p(I,W^{1,q}(\mathbb{R}^N))$ and
\begin{equation}
\|\mathcal{E}u\|_{L^p(I,W^{1,q}(\mathbb{R}^N))}\leq C_{1}\|u\|_{\mathcal{V}
^{p,q}(I,\Omega )}.  \label{5}
\end{equation}
We now show that $\mathcal{E}u\in W^{1,p}(I,L^q (\mathbb{R}^N))$.
First, we claim that
\begin{equation}
\frac{d}{dt}(\mathcal{E}u)=\mathcal{E}\frac{du}{dt},  \label{6}
\end{equation}
where the left-hand side is the derivative of $\mathcal{E}u$ in
the sense of distributions on $I$ with values in
$L^q (\mathbb{R}^N)$. The right-hand side is defined since
$\frac{du}{dt}\in L^p(I,L^q (\Omega ))$. For the proof of
\eqref{6}, let $\psi \in C_0 ^{\infty }(I)$. Then,
\begin{equation*}
\langle \frac{d}{dt}(\mathcal{E}u),\psi \rangle
=-\langle \mathcal{E}u,\psi '\rangle
=-\int_{I}\psi '(t)(\mathcal{E}u)(\cdot ,t)dt\in
L^q (\mathbb{R}^N).
\end{equation*}
Now, $\psi '(t)(\mathcal{E}u)(\cdot ,t)=\psi '(t)Eu(\cdot ,t)=E\left( \psi '(t)u(\cdot ,t)\right) $ by
the linearity of $E$. Thus,
\begin{equation*}
\langle \frac{d}{dt}(\mathcal{E}u),\psi \rangle
=-\int_{I}E\left( \psi
'(t)u(\cdot ,t)dt\right)
=-E\Big( \int_{I}\psi '(t)u(\cdot ,t)dt\Big) ,
\end{equation*}
where the second equality follows from the Bochner integral
commuting with bounded linear operators (\cite[p. 153]{DuSc88}).
Next, $\int_{I}\psi '(t)u(\cdot ,t)dt=-\int_{I}\psi
(t)\frac{du}{dt}(\cdot ,t)dt$ since $u\in W^{1,p}(I,L^q (\Omega
))$, so that
$$
\langle \frac{d}{dt}(\mathcal{E} u),\psi \rangle
=E\Big( \int_{I}\psi (t)\frac{du}{dt}(\cdot ,t)dt\Big).
$$
 The same linearity and commutativity properties yield
$$
\langle \frac{d}{dt}( \mathcal{E}u),\psi \rangle
=\int_{I}\psi (t)E\Big(\frac{du}{dt}(\cdot ,t)\Big) dt
=\int_{I}\psi (t)\Big(\mathcal{E}\frac{du}{dt}\Big) (\cdot ,t)dt.
$$
This proves \eqref{6}.

By \eqref{6} and the continuity of
$\mathcal{E}:L^p(I,L^q (\Omega ))\to
L^p(I,L^q (\mathbb{R}^N))$, it follows that $\frac{d}{dt}(
\mathcal{E}u)\in L^p(I,L^q (\mathbb{R}^N))$ and that
$\|\frac{d}{dt} \mathcal{E}u\|_{L^p(I,L^q (\mathbb{R}^N))}\leq
C_0 \|\frac{du}{dt} \|_{L^p(I,L^q (\Omega ))}$. Thus,
$\|\mathcal{E}u\|_{W^{1,p}(I,L^q (\mathbb{R}^N))}\leq
C_0 \|u\|_{W^{1,p}(I,L^q (\Omega ))}\leq C_0 \|u\|_{
\mathcal{V}^{p,q}(I,\Omega )}$. Together with \eqref{5}, we obtain
that $ \mathcal{E}u\in \mathcal{V}^{p,q}(I,\mathbb{R}^N)$ with
$\|\mathcal{E}u\|_{ \mathcal{V}^{p,q}(I,\mathbb{R}^N)}\leq
(C_{1}+C_0 )\|u\|_{\mathcal{V} ^{p,q}(I,\Omega )}$. This
completes the proof.
\end{proof}

\begin{proof}[End of the proof of Theorem \ref{thm1}]
For $u\in \mathcal{V}^{p,q}(I,\Omega )$, let $\mathcal{E}u$ be the
extension to $\mathcal{V} ^{p,q}(I,\mathbb{R}^N)$ obtained in
Lemma \ref{lem4}. If $I\neq \mathbb{R}$ is infinite, a reflection
of $\mathcal{E}u$ about the endpoint of $I$ yields a (bounded,
linear) extension of $u$ to
$\mathcal{V}^{p,q}(\mathbb{R},\mathbb{R} ^N)$. If $I=(a,b)$ is
bounded, a first reflection about $a$ followed by a reflection
about $b$ and multiplication by a smooth function with compact
support and equal to $1$ on $[a,b]$ produces the same result.
Thus, in all cases, we obtain a bounded linear extension
$\widetilde{\mathcal{E}}: \mathcal{V}^{p,q}(I,\Omega )\to
\mathcal{V}^{p,q}(\mathbb{R},\mathbb{R} ^N)$. It is then obvious
that the first part of Theorem \ref{thm1} is implied by the same
result for $\mathcal{V}^{p,q}(\mathbb{R},\mathbb{R}^N)$ proved in
Section \ref{proof1}.

 If $I$ and $\Omega $ are bounded, the compactness of the
embedding $ \mathcal{V}^{p,q}\hookrightarrow C^{0,\alpha
}(\overline{\Omega \times I})$ when $0\leq \alpha
<\frac{1}{p'}-\frac{N}{q}$ follows from the well-known
properties of H\"older spaces on bounded domains; see e.g.
\cite[p.12]{AdFo03}.
\end{proof}

Various other embedding theorems can be deduced from Theorem
\ref{thm1}, for example that $W^{2,p}(I,L^q (\Omega ))\cap
W^{1,p}(I,W^{2,q}(\Omega ))\hookrightarrow
C^{1,\frac{1}{p'}-\frac{N}{q}}(\overline{\Omega \times I})$ under
the same condition $q>Np'$. This follows from $ u,u_t,u_{x_j}\in
\mathcal{V}^{p,q}(I,\Omega ),1\leq j\leq N$ and Theorem \ref{thm1}
together with elementary properties of distributions.

For full disclosure, it should be mentioned that some embedding
theorems for anisotropic Sobolev spaces with mixed norm in
\cite{BeIlNi78}, \cite {BeIlNi79}, closely related to Theorem
\ref{thm1}, cannot be obtained by the same simple arguments. This
happens when $I\times \Omega $ satisfies an ``$l$ -horn
condition'' with $l=(l_{1},\dots ,l_{N+1})$ and not all the
$l_{k}$ are the same, which however places quite stringent
restrictions about the geometry of $\Omega $. For example, if
$\Omega =J$ is an interval, $ W^{1,p}(I,L^q (J))\cap
L^p(I,W^{2,q}(J))\hookrightarrow C^0(\overline{ J\times I})$
(and the embedding is compact if $I$ and $J$ are bounded) if $
q\geq 1$ and $q>\frac{p'}{2}$, because rectangles satisfy a strong
$ (2,1)$-horn condition \footnote{ But most open subsets,
including disks, \emph{do not}.} (\cite[p. 155] {BeIlNi78}).
Theorem \ref{thm1} cannot take advantage of the fact that $
W^{1,q}(J)$ is replaced by $W^{2,q}(J)$ and so yields the same
result (plus H\"older continuity) only when $q>p'$.

Finally, Theorem \ref{thm1} can be generalized when $I$ is replaced
by an open subset $\omega $ of $\mathbb{R}^{M}$ satisfying a
strong local Lipschitz condition, provided that
$\frac{M}{p}+\frac{N}{q}<1$. The H\"older exponent becomes
$1-\frac{M}{p}-\frac{N}{q}$. The only significant difference
occurs in the last step, to extend elements of $\mathcal{V}
^{p,q}(\omega ,\mathbb{R}^N)$ to
$\mathcal{V}^{p,q}(\mathbb{R}^{M},\mathbb{R}^N)$. This amounts to
finding an extension from $L^p(\omega ,W^{1,q}(\mathbb{R} ^N))$
to $L^p(\mathbb{R}^{M},W^{1,q}(\mathbb{R}^N))$, which is also an
extension from $W^{1,p}(\omega ,L^q (\mathbb{R}^N))$ to
$W^{1,p}(\mathbb{R} ^{M},L^q (\mathbb{R}^N))$. When $\omega
=\mathbb{R}_{+}^{M}$, this can be done by reflection. The general
case reduces to $\mathbb{R}_{+}^{M}$ by localization and
bi-Lipschitz change of coordinates.

\section{H\"older continuity of the solutions of parabolic
equations\label{example}}

Let $\Omega \subset \mathbb{R}^N$ denote once again an open subset
satisfying the strong local Lipschitz condition and let $A$ be the
generator of a $C_0  $ holomorphic semigroup $S(t)$ on
$L^q (\Omega ),q\in (1,\infty )$, whose domain $D(A)$
(equipped with the graph norm) is continuously embedded in
$ W^{1,q}(\Omega)$. This class includes many of the ``classical'' elliptic
operators with various boundary conditions; see the comments after
Theorem \ref{thm2}. If $u_0 \in L^q (\Omega )$, then
\begin{equation*}
u(t):=S(t)u_0 
\end{equation*}
is the unique solution of $u_t=Au$ such that $u(\cdot ,0)=u_0 $.
Basic semigroup theory yields only rather weak continuity
properties of $u$ near $ t=0$, that is, $u\in
C^1([0,T],L^q (\Omega ))\cap C^0([0,T],D(A))$ if $ u_0 \in
D(A)$ and only $u\in C^0([0,T],L^q (\Omega ))$ if $u_0 \notin
D(A)$.

On the other hand, optimal time or space regularity in Sobolev
spaces was proved by Di Blasio \cite{Di84} when $L^q (\Omega )$
is replaced by a general Banach space $E$. In that setting,
``space'' regularity is accounted for by the real interpolation
spaces between $E$ and $D(A)$ while time regularity is measured by
the $E$-valued Sobolev-Slobodeckii spaces on $ (0,T)$.

By combining Di Blasio's results with Theorem \ref{thm1}, we shall
obtain a more refined and stronger space-time H\"older regularity
of $u$, provided that $q$ is large enough. If $X$ and $Y$ are
Banach spaces and $\theta \in (0,1),p\in [1,\infty ]$, we denote
by $(X,Y)_{\theta ,p}$ and $[X,Y]_{\theta }$ the real and complex
interpolation spaces between $X$ and $Y$, respectively.

\begin{theorem}\label{thm2}
(i) Suppose that $q>N$ and $u_0 \in D(A)$. Then, $u\in
C^{0,\alpha -\frac{N}{q}}([0,T]\times \overline{\Omega })$ for
every $T>0$ and every $0\leq \alpha <1$.

(ii) Suppose
that $p\in (1,\infty )$, $q\in (Np',\infty )$ and $
u_0 \in (L^q (\Omega ),D(A))_{1/p',p}$. Then,
$u\in C^{0, \frac{1}{p'}-\frac{N}{q}}([0,T]\times
\overline{\Omega })$ for every $T>0$.

 (iii) Suppose that
$\theta \in (0,1),q\in \left( \frac{N}{\theta },\infty \right) $
and $u_0 \in [L^q (\Omega ),D(A)]_{\theta }$. Then, $u\in
C^{0,\alpha -\frac{N}{q}}([0,T]\times \overline{\Omega })$ for
every $T>0$ and every $0\leq \alpha <\theta $.
\end{theorem}

\begin{proof}
(i) First, $u\in C^1([0,T],L^q (\Omega ))\cap
C^0([0,T],D(A))\subset \mathcal{V}^{p,q}((0,T), \Omega )$
for every $p\in (1,\infty )$ (because $D(A)\hookrightarrow
W^{1,q}(\Omega )$). Next, since $q>N$ and $ \alpha <1$, choose $p$
large enough that $q>Np'$ and $\alpha \leq
\frac{1}{p'}$ and use Theorem \ref{thm1}.

(ii) From \cite[Theorem 4 and Theorem 9]{Di84}
(see also \cite[Theorem 3.4.2]{BuBe67}) the hypothesis
$u_0 \in (L^q (\Omega ),D(A))_{\frac{1}{p'},p}$
implies $u\in W^{1,p}((0,T),L^q (\Omega ))\cap L^p((0,T),D(A))$,
so that $u\in \mathcal{V}^{p,q}((0,T),\Omega )$ and the
conclusion follows from Theorem \ref{thm1}.

(iii) It is shown in \cite[Theorem 4.7.1]{BeLo76} that
$[L^q (\Omega ),D(A)]_{\theta }\subset (L^q (\Omega
),D(A))_{\theta ,\infty }$. Furthermore, $(L^q (\Omega
),D(A))_{\theta ,\infty }\subset (L^q (\Omega ),D(A))_{\theta
-\varepsilon ,p}$ for every $\varepsilon \in (0,\theta )$ and
every $p\in [1,\infty ]$ since $D(A)\subset L^q (\Omega )$. Thus,
if $ \varepsilon \in (0,\theta )$ is chosen small enough that
$q>\frac{N}{\theta -\varepsilon }$ and that $\alpha \leq \theta
-\varepsilon $, it suffices to use (ii) with $p=\frac{1}{1-\theta
+\varepsilon }$.
\end{proof}

Evidently, (i) is not directly implied by $u\in C^1([0,T],L^q (\Omega
))\cap C^0([0,T],D(A))$. Under the assumptions of (ii), Di
Blasio proves the ``mixed'' regularity \cite[Theorem 14]{Di84}\
$u\in W^{\varepsilon ,p}((0,T),(L^q (\Omega
),D(A))_{1-\varepsilon ,p})$ for every $\varepsilon \in (0,1)$.
Since $D(A)\hookrightarrow W^{1,q}(\Omega )$, it follows that $
u\in W^{\varepsilon ,p}((0,T),B^{1-\varepsilon ;q,p}(\Omega ))$,
but this does not yield (ii) or (iii). Indeed, by the classical
embedding theorems, $ u\in C^{0,\varepsilon
-\frac{1}{p}}([0,T],C^{0,1-\varepsilon -\frac{N}{q}}(
\overline{\Omega }))$ if $\frac{1}{p}<\varepsilon <1-\frac{N}{q}$
(hence, $ q>Np'$). Thus, (ii) could be recovered if 
$\min \{\varepsilon -\frac{1}{p},1-\varepsilon-\frac{N}{q}\} 
\geq \frac{1}{p'}- \frac{N}{q}$;
i.e., $1-\frac{N}{q}\leq \varepsilon \leq \frac{1}{p}$, but
this requires $q\leq Np'$ and therefore never happens.

A similar argument shows that no choice of $\varepsilon \in (0,1)$
yields $ u\in C^{0,\alpha }(\overline{\Omega }\times [0,T])$ if
$\alpha <\frac{1}{ p'}-\frac{N}{q}$ is close enough to
$\frac{1}{p'}-\frac{N}{ q}$, whence (iii) cannot be
proved that way either. Actually, even if $ D(A)\hookrightarrow
W^{k,q}(\Omega )$ with $k>1$, (ii) and (iii) can be deduced from
Di Blasio's mixed regularity result only when $q<kNp'$
(but then $q>\frac{Np'}{k}$ suffices, instead of
$q>Np'$).

In connection with the above discussion, it is instructive to
notice that if $\alpha \in (0,1)$, then $C^{0,\alpha
}([0,T],C^{0,\alpha }(\overline{\Omega }))\varsubsetneq
C^{0,\alpha }(\overline{\Omega }\times [0,T])$ (example: $
(0,T)=\Omega,u(x,t):=(x+t)^{\alpha }$), so that membership to the
latter space does not require membership to the former.

In the proof of (ii), we saw that the assumption $u_0 \in
(L^q (\Omega ),D(A))_{\frac{1}{p'},p}$ implies $u\in
W^{1,p}((0,T),L^q (\Omega ))\cap L^p((0,T),D(A))$. The converse
is true, from the very definition of real interpolation spaces by
the trace method. Thus, the assumption $ u_0 \in (L^q (\Omega
),D(A))_{\frac{1}{p'},p}$ is also necessary for $u\in
W^{1,p}((0,T),L^q (\Omega ))\cap L^p((0,T),D(A))$ to solve the
nonautonomous problem
\[ \left\{
\begin{gathered}
u_t=Au+f, \\
u(\cdot ,0)=u_0 ,
\end{gathered}\right.
\]
when $f\in L^p((0,T),L^q (\Omega ))$. In turn, the existence
and uniqueness of such a solution $u$ is known under additional
assumptions about $A$. See \cite[Theorem 4.4]{DeHiPr03} when
$u_0 =0$ (of course, the general case reduces to this case after
adding $S(t)u_0 $). It may be worth pointing out that the
``$L^p$ maximal regularity'' in that theorem refers to the
aforementioned existence and uniqueness (see Remark \ref{rmk1}
below) and that, when $q\in (1,\infty )$, the space $L^q (\Omega
)$ is a Banach space of class $\mathcal{HT}$ according to the
terminology of \cite{DeHiPr03} (more commonly called a UMD Banach
space). If so, Theorem \ref{thm2} remains true since the H\"older
regularity is solely based on the fact that $u\in
\mathcal{V}^{p,q}((0,T),\Omega )$.

\begin{remark} \label{rmk1} \rm
In \cite{DeHiPr03}, $L^p$ maximal regularity is
defined (when $ u_0 =0$) by the condition that $u_t\in
L^p((0,T),L^q (\Omega ))$, which may seem weaker than stated
above. However, since $T<\infty $, this implies at once that both
$u(t)=\int_0 ^{t}u_t(s)ds$ and $Au=u_t-f$ are in
$L^p((0,T),L^q (\Omega ))$, so that $u\in L^p((0,T),D(A))$
when $D(A) $ is equipped with the graph norm and that $u\in
W^{1,p}((0,T),L^q (\Omega ))$. If $(0,T)$ is replaced by
$(0,\infty )$, this remains true only if the invertibility of $A$
is added to the assumptions of \cite[Theorem 4.4]{DeHiPr03}.
\end{remark}

If $-A$ is sectorial with bounded imaginary powers, part (iii) of
Theorem \ref{thm2} holds with $u_0 \in D((-A)^{\theta })$ since it
is known that $ [L^q (\Omega ),D(A)]_{\theta }=D((-A)^{\theta })$
in this case \cite[p. 103]{Tr78}.

Everything can be extended to systems, that is, when
$L^q (\Omega)$ is replaced by
$(L^q (\Omega ))^{r},r\in \mathbb{N}$, provided
that $ D(A)\hookrightarrow (W^{1,q}(\Omega ))^{r}$. Indeed, it
suffices to use Theorem \ref{thm1} componentwise.

Of course, the case when $A$ is the realization of a linear
elliptic differential operator with homogeneous boundary
conditions is of special importance. If so,
$D(A)=W_{\mathcal{B}}^{2m,q}(\Omega ):=\{v\in W^{2m,q}:
\mathcal{B}v=0\}$ where $m\in \mathbb{N}$ and $\mathcal{B}$ is a
system of boundary operators of order less than $2m$. The
condition $D(A)\subset W^{1,q}(\Omega )$ always holds, but some
regularity assumptions are needed for $A$ to have some or all of
the properties listed above. Recall also that $(L^q (\Omega
),D(A))_{\theta ,p}$ is often explicitly known in this case
(Grisvard\footnote{ Unfortunately, details are given only when
$q=2$, the (important) case when $ 2m\theta $ is an integer is
left out when $q\neq 2$ and the result is mostly limited to scalar
problems. Also, $\theta $ is $1-\theta $ in Grisvard's paper due
to the reverse ordering of the interpolation pair.} \cite[p. 63]
{Gr67}), as is $[L^q (\Omega ),D(A)]_{\theta }$ (Seeley
\cite{Se72}).

Hypotheses about $\Omega $ and about the coefficients that  ensure
$L^p$ maximal regularity are spelled out in \cite[Theorem
8.2]{DeHiPr03}. A little more must be assumed for $-A$ to have
bounded imaginary powers and even a bounded $\mathcal{H}^{\infty
}$ calculus, which is stronger. Until fairly recently, this was
mostly known for operators with constant coefficients, but it has
now been proved for a broader class by Denk \textit{et al}.
\cite{DeDoHiPrVe04}.

\begin{thebibliography}{00}

\bibitem{AdFo03}  Adams, R. A. and Fournier, J. J. F.;
 \textit{Sobolev spaces }, 2nd edition, Academic Press, Boston 2003.

\bibitem{Am00}  Amann, H.;
 Compact embeddings of vector-valued Sobolev and
Besov spaces, \textit{Glas. Mat.} \textbf{35} (2000) 161-177.

\bibitem{ArBu02}  Arendt, W. and Bu, S.;
 The operator-valued Marcinkiewicz
multiplier theorem and maximal regularity, \textit{Math. Z.} \textbf{240}
(2002) 311-343.

\bibitem{ArRa08}  Arendt, W. and Rabier, P. J.;
 Linear Evolution operators
on spaces of periodic functions, \textit{Comm. Pure Appl. Anal.} \textbf{8}
(2009) 5-36.

\bibitem{BePa61}  Benedek, A. and Panzone, R.;
 The spaces $L^{P}$, with
mixed norm, \textit{Duke Math J.} \textbf{28} (1961) 301-324.

\bibitem{BeIlNi78}  Besov, O. V.; Il'in, V. P. and Nikol'skii, S. M.;
\textit{Integral representation of functions and imbedding theorems,}
Nauka Moscow 1975, English translation, Vol. I, Wiley, New York 1978.

\bibitem{BeIlNi79}  Besov, O. V.; Il'in, V. P. and Nikol'skii, S. M.;
\textit{Integral representation of functions and imbedding theorems,}
Nauka Moscow 1975, English translation, Vol. II, Wiley, New York 1979.

\bibitem{BeLo76}  Bergh, J. and L\"ofstr\"om, J.;
\textit{Interpolation spaces,} Grund. math. Wiss. \textbf{223},
Springer-Verlag, Berlin 1976.

\bibitem{Br83}  Br\'{e}zis, H.;
 \textit{Analyse fonctionnelle,} Masson,
Paris 1983.

\bibitem{BuBe67}  Butzer, P. L. and Berens, H.;
\textit{Semi-groups of operators and approximation},
Grund. math. Wiss. \textbf{145},
Springer-Verlag, New York 1967.

\bibitem{DeHiPr03}  Denk, R., Hieber, M. and Pr\"{u}ss, J.;
$\mathcal{R}$-\textit{boundedness, Fourier multipliers and
problems of elliptic and parabolic type,}
Mem. Amer. Math. Soc. no. \textbf{788},
Providence, 2003.

\bibitem{DeDoHiPrVe04}  Denk, R.; Dore, G.; Hieber, M.;
 J Pr\"{u}ss, J. and Venni, A.;
New thoughts on old results of R T Seeley, \textit{Math. Annalen}
\textbf{328} (2004) 545-583.

\bibitem{Di84} Di Blasio, G.;
Linear parabolic evolution equations in $L^p$-spaces,
\textit{Ann. Mat. Pura Appl.} \textbf{138} (1984) 55-104.

\bibitem{DuSc88}  Dunford, N. and Schwartz, J. T.;
\textit{Linear operators, part I,} Wiley, New York 1988.

\bibitem{En96}  Engler, H.;
Global smooth solutions for a class of parabolic
integrodifferential equations, \textit{Trans. Amer. Math. Soc.}
\textbf{348} (1996) 267-290.

\bibitem{Gr67}  Grisvard, P.;
Caract\'erisation de quelques espaces
d'interpolation, \textit{Arch. Rational Mech. Anal.} \textbf{25} (1967)
40-63.

\bibitem{Mo06}  Morris, J. R.;
Nonautonomous semilinear parabolic equations
on an infinite interval, \textit{Dynamics of PDEs}
\textbf{3} (2006) 209-233.

\bibitem{Pr91}  Pr\"uss, J.;
 Quasilinear parabolic Volterra equations in
spaces of integrable functions. In: \textit{Semigroup Theory and Evolution
Equations,} Proceedings, Delft 1989, Cl\'ement, Mitidieri,
de Pagter, eds, Lect. Notes Pure Appl. Math. \textbf{135},
pp. 401-420, Marcel Dekker, New York 1991.

\bibitem{Ra78}  Rao, V. R. G.;
 Parabolic function spaces with mixed norm,
\textit{Trans. Amer. Math. Soc.} \textbf{246} (1978) 451-461.

\bibitem{Sc57}  Schwartz, L.;
 Distributions \`a valeurs vectorielles I,
\textit{Ann. Instit. Fourier} \textbf{7} (1957) 1-141.

\bibitem{Se72} Seeley, R.;
 Interpolation in $L^p$ with boundary
conditions, \textit{Studia Math.} \textbf{44} (1972) 47-60.

\bibitem{St70}  Stein, E. M.;
\textit{Singular integrals and differentiability properties
of functions,} Princeton University Press, Princeton 1970.

\bibitem{Tr78} Triebel, H.;
\textit{Interpolation theory, function spaces,
differential operators,} North-Holland Mathematical Library \textbf{18},
North-Holland, Amsterdam, 1978.

\end{thebibliography}

\end{document}