\documentclass[reqno]{amsart}
\usepackage{hyperref}
\usepackage{graphicx}
\AtBeginDocument{{\noindent\small
\emph{Electronic Journal of Differential Equations},
Vol. 2013 (2013), No. 92, pp. 1--25.\newline
ISSN: 1072-6691. URL: http://ejde.math.txstate.edu or http://ejde.math.unt.edu
\newline ftp ejde.math.txstate.edu}
\thanks{\copyright 2013 Texas State University - San Marcos.}
\vspace{9mm}}
\begin{document}
\title[\hfilneg EJDE-2013/92\hfil Nonautonomous ill-posed evolution problems]
{Nonautonomous ill-posed evolution problems \\
with strongly elliptic differential operators}
\author[M. A. Fury\hfil EJDE-2013/92\hfilneg]
{Matthew A. Fury} % in alphabetical order
\address{Matthew Fury \newline
Division of Science \& Engineering \\
Penn State Abington \\
1600 Woodland Road \\ Abington, PA 19001, USA\newline
Tel: 215-881-7553 \\ Fax: 215-881-7333}
\email{maf44@psu.edu}
\thanks{Submitted November 3, 2012. Published April 11, 2013.}
\subjclass[2000]{46B99, 47D06}
\keywords{Regularizing family of operators; ill-posed evolution equation;
\hfill\break\indent holomorphic semigroup; strongly elliptic operator}
\begin{abstract}
In this article, we consider the nonautonomous evolution problem
$du/dt=a(t)Au(t), 0\leq s\leq t< T$ with initial condition $u(s)=\chi$
where $-A$ generates a holomorphic semigroup of angle $\theta \in (0,\pi/2]$
on a Banach space $X$ and $a\in C([0,T]:\mathbb{R}^+)$.
The problem is generally ill-posed under such conditions, and so we employ
methods to approximate known solutions of the problem. In particular,
we prove the existence of a family of regularizing operators for the
problem which stems from the solution of an approximate well-posed problem.
In fact, depending on whether $\theta \in (0,\pi/4]$ or
$\theta \in (\pi/4,\pi/2]$, we provide two separate approximations each
yielding a regularizing family. The theory has applications to ill-posed
partial differential equations in $L^p(\Omega)$, $1
0, \; t \in [s,T]\}$ of bounded
linear operators on $X$ is called a \emph{family of regularizing operators
for the problem} \eqref{1} if for each solution $u(t)$ of \eqref{1}
with initial data $\chi \in X$, and for any $\delta>0$, there exists
$\beta(\delta)>0$ such that
\begin{itemize}
\item[(i)] $\beta(\delta)\to 0$ as $\delta \to 0$,
\item[(ii)] $\|u(t)-R_{\beta(\delta)}(t)\chi_\delta\|\to 0$ as $\delta \to 0$
for $s\leq t\leq T$ whenever $\|\chi - \chi_{\delta}\|\leq \delta$.
\end{itemize}
\end{definition}
As in the case of regularization for the autonomous problem \eqref{ACP},
we will show that a family of regularizing operators for \eqref{1} stems
from the solution of an approximate well-posed problem
\begin{equation} \label{2}
\begin{gathered}
\frac{dv}{dt} = f_{\beta}(t,A)v(t) \quad 0\leq s\leq t0$, the operators $f_{\beta}(t,A), 0\leq t\leq T$
are defined by two different approximations of the operators $a(t)A$
depending on where $\theta$ lies in the interval $(0,\pi/2]$:
\begin{equation} \label{f_beta}
f_{\beta}(t,A) = \begin{cases}
a(t)A-\beta A^{\sigma} & \text{if } \theta \in (0,\pi/4] \\
a(t)A(I+\beta A)^{-1} & \text{if } \theta \in (\pi/4,\pi/2]
\end{cases}
\end{equation}
where $\sigma>1$ when $\theta \in (0,\pi/4]$.
Each approximation in \eqref{f_beta} yields a well-posed problem \eqref{2},
and also continuous dependence on modeling for the ill-posed
problem \eqref{1} in the sense that as $\beta \to 0$, the operators
$f_{\beta}(t,A)$ approach the operators $a(t)A$, and given solutions
$u(t)$ and $v_{\beta}(t)$ of \eqref{1} and \eqref{2} respectively, we have
\begin{equation} \label{CDMsortof}
\|u(t)-v_{\beta}(t)\| \to 0 \quad \text{as } \beta \to 0
\end{equation}
for each $t\in [s,T]$. We use \eqref{CDMsortof} to establish the
main result of the paper, that the family $\{V_{\beta}(t,s) : \beta>0, \;
t \in [s,T]\}$ is a family of regularizing operators for the ill-posed
problem \eqref{1} where $V_{\beta}(t,s),0\leq s\leq t\leq T$
is an evolution system associated with the well-posed problem \eqref{2}
satisfying $V_{\beta}(t,s)\chi =v_{\beta}(t)$.
In other words, given a small change in the initial data
$\|\chi-\chi_{\delta}\|\leq \delta$ (which, since \eqref{1}
is ill-posed, could yield a very large difference in solutions),
there exists $\beta>0$ so that $\beta \to 0$ as $\delta\to 0$, and
$\|u(t)-V_{\beta}(t,s)\chi_{\delta}\|\to 0$ as $\delta \to 0$ for
$s\leq t\leq T$. Hence, although $u(t)$ may not be ``close"
to the solution of \eqref{1} with initial data $\chi_{\delta}$,
we can still approximate $u(t)$ by utilizing the well-posed
problem \eqref{2} with regularization parameter $\beta>0$.
The use of the two approximations in \eqref{f_beta} extends results
from previous works in which the approximations $A-\beta A^{\sigma}$
and $A(I+\beta A)^{-1}$ are used to obtain regularization for the
autonomous problem \eqref{ACP} where $-A$ generates a holomorphic
semigroup of angle $\theta$ on $X$
(cf. \cite{AmesandHughes,HuangZheng2, HuangZheng,Melnikova,MelnikovaandFilinkov}). For instance, in \cite{HuangZheng2}, Huang and Zheng obtain regularization for \eqref{ACP} using the quasi-reversibility method, first introduced by Lattes and Lions \cite{LandL}, which involves the approximation $A-\beta A^{\sigma}$ of the operator $A$. Here, the requirements that $\sigma>1$ and $\sigma (\pi/2-\theta)<\pi/2$ are crucial in order for $A-\beta A^{\sigma}$ to generate a semigroup (so as to yield an approximate well-posed problem). Hence, if $\theta \in (0,\pi/4]$, these requirements force $1<\sigma<2$ whence the use of the fractional power $A^{\sigma}$ is in order. In light of definition \eqref{f_beta}, we will adopt the same requirements in the current paper for the extension $a(t)A-\beta A^{\sigma}$. The second approximation $A(I+\beta A)^{-1}$, introduced by Showalter \cite{Showalter}, is applied by Ames and Hughes \cite{AmesandHughes} and Huang and Zheng \cite{HuangZheng} but only in the case where $\theta \in (\pi/4, \pi/2]$ because the perturbation methods used to establish regularization in these papers (and in the current paper) are not applicable when $\theta \in (0,\pi/4]$ (cf. \cite[pp. 3011--3012]{HuangZheng}).
Note, if $\theta \in (\pi/2,\pi/4]$, the approximation $a(t)A-\beta A^{\sigma}$
may still be used, but it is standard and easier in this case to let
$\sigma =2$ (cf. \cite{AmesandHughes,Fury,LandL,Melnikova,MelnikovaandFilinkov,
Miller1,Payne1,Payne2}). In this regard, the current paper also furthers
results from \cite{Fury} where the author uses the approximation
$\sum_{j=1}^ka_j(t)A^j-\beta A^{k+1}$ to obtain regularization for
the nonautonomous problem
\begin{gather*}
\frac{du}{dt} = \sum_{j=1}^ka_j(t)A^ju(t) \quad 0\leq s\leq t0$ on a Banach space $X$ is called a
\emph{bounded holomorphic semigroup of angle} $\theta$ if the following
conditions are satisfied:
\begin{itemize}
\item[(i)] $T(t)$ is the restriction to the positive real axis of an
analytic family of operators $T(z)$ in the open sector
$S_{\theta}=\{re^{i\theta'} : r> 0, \; |\theta'| <\theta\}$
satisfying $T(z+w)=T(z)T(w)$ for all $z,w\in S_{\theta}$.
\item[(ii)] For each $\theta_1< \theta$, $T(z)x\to x$ as $z\to 0$
in $S_{\theta_1}$ for all $x\in X$.
\item[(iii)] For each $\theta_1< \theta$, $T(z)$ is uniformly bounded
in the sector $S_{\theta_1}$.
\end{itemize}
More generally, a strongly continuous semigroup $T(t)$ on $X$ is
called a \emph{holomorphic semigroup of angle} $\theta$ if $T(t)$
satisfies all the properties of a bounded holomorphic semigroup of
angle $\theta$ with the exception of (iii).
\end{definition}
\begin{theorem}[{\cite[Theorem~X.52]{ReedandSimon}}]
\label{generator_holomorphic}
Let $A$ be a closed operator on a Banach space $X$.
Then $-A$ is the infinitesimal generator of a bounded holomorphic
semigroup of angle $\theta$ if and only if for each $\theta_1<\theta$
there exists a constant $M_1>0$ such that if
$w \not \in \bar{S}_{\pi/2-\theta_1}$, then $w \in \rho(A)$ and
\begin{equation}
\label{resolvent}
\| (w -A)^{-1}\|\leq \frac{M_1}{\operatorname{dist}(w,\bar{S}_{\pi/2-\theta_1})}.
\end{equation}
\end{theorem}
For this paper, we first assume that $-A$ generates a \emph{bounded}
holomorphic semigroup of angle $\theta$. In fact, for most of the paper,
we will make this assumption for convenience, but then generalize our
results at the end for holomorphic semigroups for which only conditions
(i) and (ii) of Definition~\ref{holomorphic_semigroup} hold.
Since $-A$ generates a bounded holomorphic semigroup of angle $\theta$,
by Theorem~\ref{generator_holomorphic}, it follows that the spectrum
$\sigma (A)$ of $A$ is contained in
$\bar{S}_{\pi/2-\theta}=\{re^{i\theta'} : r\geq 0, \;
|\theta'|\leq \pi/2-\theta\}$. Further, for $t>0$, $T(t)$
is given by the Cauchy integral formula
\begin{equation}
\label{T(t)}
T(t)=\frac{1}{2\pi i}\int_{\Gamma_{\phi}}e^{-tw}(w-A)^{-1}dw
\end{equation}
where $\pi/2>\phi>\pi/2-\theta$ and $\Gamma_{\phi}$ is a curve
in $\rho(A)$ consisting of three pieces:
$\Gamma_1=\{re^{i\phi} : r\geq 1\}$,
$\Gamma_2=\{e^{i\theta'} : \phi \leq \theta' \leq 2\pi -\phi\}$, and
$\Gamma_3=\{re^{-i\phi} : r\geq 1\}$; $\Gamma_{\phi}$ is oriented
so that it runs from $\infty e^{i\phi}$ to $\infty e^{-i\phi}$
(see Figure~\ref{fig:Gamma}). Similarly, for $z\in S_{\theta}$,
\[
T(z)=\frac{1}{2\pi i}\int_{\Gamma_{\phi}}e^{-zw}(w-A)^{-1}dw.
\]
\begin{figure}[ht]
\begin{center}
\includegraphics[height=9cm]{fig1} % newgamma.pdf}
\put(-60,242){\small$\pi/2-\theta$}
\put(-120,203){\small$\Gamma_1$}
\put(-170,95){\small$\Gamma_2$}
\put(-114,40){\small$\Gamma_3$}
\put(-18,201){\small$\theta$}
\put(-142,140){\small$1$}
\put(-93, 245){\small$\phi$}
\end{center}
\caption{$\Gamma_{\phi}$}
\label{fig:Gamma}
\end{figure}
We will first prove that the approximate problem \eqref{2} is well-posed
in the case that $\theta \in (0,\pi/4]$ and $f_{\beta}(t,A), 0\leq t\leq T$
is defined by $f_{\beta}(t,A)=a(t)A-\beta A^{\sigma}$
(Proposition~\ref{well-posed_approx1} below). The idea in this case
is to construct an evolution system $V_{\beta}(t,s)$ which will be defined
similarly as in \eqref{T(t)}. For this, we will need to choose an
appropriate value for $\phi$ in a contour similar to $\Gamma_{\phi}$.
In particular, we will require that $\sigma>1$ and $\sigma (\pi/2-\theta)<\pi/2$
in order to allow $\pi/2\sigma>\phi>\pi/2-\theta$. As noted in the introduction,
since $\theta \in (0,\pi/4]$, these requirements force $1<\sigma<2$ and so
we will need to make sense of the operator $A^{\sigma}$ which is defined by
the fractional power. To this end, we will require the assumption that
$0\in \rho(A)$ (see Definition~\ref{fractional_power} below).
\begin{definition}[{\cite[Definition~2.4]{HuangZheng2}}] \label{fractional_power}
\rm
Let $-A$ be the infinitesimal generator of a bounded holomorphic semigroup
of angle $\theta$, and let $0\in \rho (A)$. For $\sigma >0$,
the \emph{fractional power of} $A$ is defined as follows:
\begin{eqnarray}
\label{Asigma}
A^{-\sigma}=\frac{1}{2\pi i}\int_{\Gamma}w^{-\sigma}(w-A)^{-1}dw,
\end{eqnarray}
where $w^{-\sigma}$ is defined by the principal branch, and $\Gamma$
is a path running from $\infty e^{i\phi}$ to $\infty e^{-i\phi}$ with
$\pi>\phi>\pi/2-\theta$ while avoiding the negative real axis and the origin.
Define $A^{\sigma}=(A^{-\sigma})^{-1}$
(see Lemma \ref{fractional_props} (i) below) and $A^0=I$.
\end{definition}
Note, in Definition~\ref{fractional_power}, the definition of $A^{\sigma}$
relies on the fact that the operator in \eqref{Asigma} is invertible which
follows from the following properties of the fractional power.
\begin{lemma}[{\cite[Lemma~2.5]{HuangZheng2},
\cite[Lemma~2.6.6, Theorem~2.6.8]{Pazy}}]
\label{fractional_props}
Let $-A$ be the infinitesimal generator of a bounded holomorphic semigroup
of angle $\theta$, and let $0\in \rho (A)$. Then
\begin{itemize}
\item[(i)] $A^{-\sigma}$ is a bounded, injective operator for $\sigma>0$.
\item[(ii)] $A^{\sigma}$ is a closed operator, and
$\operatorname{Dom}(A^{\sigma})\subseteq \operatorname{Dom}(A^{\sigma'})$
for $\sigma > \sigma'>0$.
\item[(iii)] $\operatorname{Dom}(A^{\sigma})$ is dense in $X$ for every
$\sigma\geq 0$.
\item[(iv)] $A^{\sigma_1+\sigma_2}x=A^{\sigma_1}A^{\sigma_2}x$ for every
$\sigma_1$, $\sigma_2 \in \mathbb{R}$ and
$x\in \operatorname{Dom}(A^{\sigma})$ where
$\sigma = \max \{\sigma_1, \sigma_2, \sigma_1+\sigma_2\}$.
\end{itemize}
\end{lemma}
\begin{proposition}\label{well-posed_approx1}
Let $-A$ be the infinitesimal generator of a bounded holomorphic semigroup
of angle $\theta \in (0,\pi/4]$, and let $0\in \rho(A)$. Let $0<\beta<1$
and assume $\sigma$ satisfies $\sigma>1$ and $\sigma(\pi/2-\theta)<\pi/2$.
Define the family of operators $f_{\beta}(t,A), 0\leq t\leq T$ by
\[
f_{\beta}(t,A)=a(t)A-\beta A^{\sigma}.
\]
Then $\eqref{2}$ is well-posed with unique classical solution
$v_{\beta}(t)=V_{\beta}(t,s)\chi$ for each $\chi \in X$ where
\[
V_{\beta}(t,s) =
\begin{cases}
\frac{1}{2\pi i}\int_{\Gamma_{\phi}}
e^{\int_s^tf_{\beta}(\tau,w)d\tau}(w-A)^{-1}\,dw & 0\leq s\phi>\pi/2-\theta$
but avoids the negative real axis and the origin.
\end{proposition}
\begin{proof}
Notice our choice for $\phi$ is valid by the assumption
$\sigma(\pi/2-\theta)<\pi/2$. We first show that $V_{\beta}(t,s)$
is uniformly bounded for $0\leq s\leq t\leq T$.
Following \cite[Proof of Theorem~3.1]{HuangZheng2},
we will show this in two cases. Let $0\leq s\pi/2-\theta_1>\pi/2-\theta$.
We have $\operatorname{dist}(w,\bar{S}_{\pi/2-\theta_1})= |w|
\sin (\phi-(\pi/2-\theta_1))$ (cf. \cite[Figure~2]{Fury})
so that by Theorem~\ref{generator_holomorphic},
\begin{equation} \label{dist}
\|(w-A)^{-1}\| \leq \frac{M_1}{|w|\;\sin (\phi-(\pi/2-\theta_1))}.
\end{equation}
Set $M_1'=M_1/\sin (\phi-(\pi/2-\theta_1))$ and $B=\max_{t\in [0,T]}|a(t)|$.
Then
\begin{align*}
\big\|\int_{\Gamma^1\cup \Gamma^3}\big\|
&\leq M_1' \int_{\Gamma^1\cup \Gamma^3}\big|
e^{\int_s^t(a(\tau)w-\beta w^{\sigma})\; d\tau}\big|\; |w|^{-1}|dw| \\
&= 2M_1'\int_{(t-s)^{-1/\sigma}}^{\infty}e^{\int_s^t(a(\tau)
r\cos \phi -\beta r^{\sigma}\cos\sigma \phi)\; d\tau} r^{-1}dr \\
&\leq 2M_1'\int_{(t-s)^{-1/\sigma}}^{\infty}e^{B(t-s)r\cos \phi
-\beta (t-s)r^{\sigma}\cos\sigma \phi} r^{-1}dr \\
&= 2M_1'\int_1^{\infty}e^{B(t-s)^{1-1/\sigma}x\cos \phi
-\beta x^{\sigma}\cos\sigma \phi} x^{-1}dx \\
&\leq 2M_1'\int_1^{\infty}e^{BT^{1-1/\sigma}x\cos \phi
-\beta x^{\sigma}\cos\sigma \phi} dx
\leq K
\end{align*}
where $K$ is a constant independent of $t$ and $s$ since $\sigma>1$
and since $\pi/2\sigma>\phi>\pi/2-\theta$ implies $0< \phi< \sigma \phi <\pi/2$
so that $\cos \phi>0$ and $\cos (\sigma \phi)>0$.
Also, for $w \in \Gamma^2$, we have
\begin{align*}
\big\|\int_{\Gamma^2}\big\|
&\leq M_d\int_{\Gamma^2}
\big|e^{\int_s^t(a(\tau)w-\beta w^{\sigma})d\tau}\big| |dw| \\
&= M_d\int_{-\phi}^{\phi} e^{\int_s^t(a(\tau)(t-s)^{-1/\sigma}\cos\theta'
-\beta (t-s)^{-1}\cos\sigma \theta')d\tau}(t-s)^{-1/\sigma}d\theta' \\
&\leq dM_d\int_{-\phi}^{\phi} e^{B(t-s)^{1-1/\sigma}\cos\theta'
-\beta \cos\sigma \theta'}d\theta' \\
&\leq dM_d\int_{-\phi}^{\phi} e^{BT^{1-1/\sigma}\cos\theta'}d\theta' \\
&\leq dM_d\;e^{BT^{1-1/\sigma}}2\phi
\end{align*}
where we have set $M_d=\max_{|w|\leq d}\|(w-A)^{-1}\|$ since
$w\to (w-A)^{-1}$ is continuous on the interior of $\rho(A)$.
Hence, $V_{\beta}(t,s)$ is bounded uniformly for $0\leq s\leq t\leq T$
in the first case.
For the second case, if $(t-s)^{-1/\sigma}>d$, then we shift $\Gamma_{\phi}$
to the contour (see Figure~\ref{fig:Gammaavoid}) consisting of the seven pieces:
\begin{gather*}
\Gamma_1 = \{re^{i\phi} : r\geq (t-s)^{-1/\sigma}\}, \quad
\Gamma_2 = \{(t-s)^{-1/\sigma}e^{i\theta'} : \phi \leq \theta' \leq \pi\}, \\
\Gamma_3 = \{re^{i\pi} : d\leq r\leq (t-s)^{-1/\sigma} \} \quad
\Gamma_4 = \{de^{-i\theta'} : -\pi \leq \theta' \leq \pi\}, \\
\Gamma_5 = \{re^{-i\pi} : d\leq r\leq (t-s)^{-1/\sigma} \} \quad
\Gamma_6 = \{(t-s)^{-1/\sigma}e^{i\theta'} : -\pi \leq \theta' \leq -\phi\}, \\
\Gamma_7 = \{re^{-i\phi} : r\geq (t-s)^{-1/\sigma}\}.
\end{gather*}
\begin{figure}[ht]
\begin{center}
\includegraphics[height=9cm]{fig3} %gammaavoid.pdf
\put(-65,205){\small$\Gamma_1$}
\put(-180,185){\small$\Gamma_2$}
\put(-172,140){\small$\Gamma_3$}
\put(-113,147){\small$\Gamma_4$}
\put(-172,108){\small$\Gamma_5$}
\put(-178,67){\small$\Gamma_6$}
\put(-65,48){\small$\Gamma_7$}
\put(-53, 252){\small$\phi$}
\put(-70,115){\small$t'$}
\put(-140,134){\small$d$}
\end{center}
\caption{$t':=(t-s)^{-1/\sigma}> d$}
\label{fig:Gammaavoid}
\end{figure}
First, since $\Gamma_1=\Gamma^1$ and $\Gamma_7=\Gamma^3$, we have
$\|\int_{\Gamma_1\cup \Gamma_7}\|=\|\int_{\Gamma^1\cup \Gamma^3}\|\leq K$
as before. Next, note that \eqref{dist} holds for $w\in \Gamma_2$
since these $w$ satisfy the inequality
$\operatorname{dist}(w,\bar{S}_{\pi/2-\theta_1})\geq
\operatorname{dist}((t-s)^{-1/\sigma}e^{i\phi}, \bar{S}_{\pi/2-\theta_1})$.
Then
\begin{align*}
\big\|\int_{\Gamma_2}\big\|
&\leq M_1'\int_{\Gamma_2} \big|e^{\int_s^t(a(\tau)w-\beta w^{\sigma})d\tau}\big|
|w|^{-1}|dw| \\
&= M_1'\int_{\phi}^{\pi} e^{\int_s^t(a(\tau)(t-s)^{-1/\sigma}\cos\theta'
-\beta (t-s)^{-1}\cos\sigma \theta')d\tau}d\theta' \\
&\leq M_1'\int_{\phi}^{\pi} e^{BT^{1-1/\sigma}\cos\phi
-\beta \cos\sigma \theta'}d\theta' \\
&\leq M_1'\int_{\phi}^{\pi} e^{1+BT^{1-1/\sigma}\cos\phi}d\theta' \\
&= M_1'e^{1+BT^{1-1/\sigma}\cos\phi}(\pi-\phi)
\end{align*}
since $0<\beta<1$. The same estimate holds for $\|\int_{\Gamma_6}\|$.
Next, using \eqref{dist},
\begin{align*}
& \big\|\int_{\Gamma_3}+\int_{\Gamma_5}\big\| \\
&= \big\| \int_d^{(t-s)^{-1/\sigma}}
\Big(e^{\int_s^t(-a(\tau)r-\beta r^{\sigma}e^{-i\pi \sigma})d\tau}
-e^{\int_s^t(-a(\tau)r-\beta r^{\sigma}e^{i\pi \sigma})d\tau}\Big)
(-r-A)^{-1}dr\big\| \\
&\leq M_1'\int_d^{(t-s)^{-1/\sigma}}
\Big|e^{-(\int_s^ta(\tau)d\tau)r}
\Big(e^{-\beta (t-s)r^{\sigma}e^{-i\pi \sigma }}
-e^{-\beta (t-s)r^{\sigma}e^{i\pi \sigma}}\Big)\Big|r^{-1}dr \\
&= M_1'\int_d^{(t-s)^{-1/\sigma}}e^{-\left(\int_s^ta(\tau)d\tau\right)r}
\left|e^{-\beta (t-s)r^{\sigma}\cos\sigma \pi}2i
\sin (\beta (t-s)r^{\sigma}\sin \sigma\pi)\right|r^{-1}dr \\
&\leq M_1'\int_d^{(t-s)^{-1/\sigma}}e^{-\beta (t-s)r^{\sigma}\cos\sigma \pi}2|
\sin (\beta (t-s)r^{\sigma}\sin \sigma\pi)|r^{-1}dr \\
&= M_1'\int_{(t-s)^{1/\sigma}d}^1e^{-\beta x^{\sigma}\cos\sigma \pi}2|
\sin (\beta x^{\sigma}\sin \sigma\pi)|x^{-1}dx \\
&= M_1'\int_{(t-s)^{1/\sigma}d}^1x^{-1/2}e^{-\beta x^{\sigma}\cos\sigma \pi}
\left\{4x^{-1}\sin ^2(\beta x^{\sigma}\sin \sigma\pi)\right\}^{1/2}dx \\
&= M_1'\int_{(t-s)^{1/\sigma}d}^1x^{-1/2}e^{-\beta x^{\sigma}\cos\sigma \pi}
\left\{2x^{-1}(1-\cos (2\beta x^{\sigma}\sin \sigma\pi))\right\}^{1/2}dx.
\end{align*}
It is easily shown by L'Hospital's Rule that
\[
2x^{-1}(1-\cos (2\beta x^{\sigma}\sin \sigma \pi))\to 0 \quad \text{as }
x \to 0.
\]
Hence, we have for a possibly different constant $M_1'$ independent of $\beta$,
\[
\|\int_{\Gamma_3}+\int_{\Gamma_5}\|
\leq M_1'\int_0^1x^{-1/2}e^{-\beta x^{\sigma}\cos\sigma \pi}dx
\leq M_1'e\int_0^1x^{-1/2}dx
= M_1'2e
\]
since $0<\beta<1$. Finally,
\begin{align*}
\|\int_{\Gamma_4}\|
&\leq M_d\int_{\Gamma_4} \big|e^{\int_s^t(a(\tau)w-\beta w^{\sigma})d\tau}\big| |dw| \\
&= dM_d\int_{-\pi}^{\pi} e^{\int_s^t(a(\tau)d \cos\theta'-\beta d^{\sigma}\cos\sigma \theta')d\tau}d\theta' \\
&\leq dM_d\int_{-\pi}^{\pi} e^{BTd}e^{-\beta (t-s)d^{\sigma}\cos\sigma \theta'}d\theta' \\
&\leq dM_d e^{BTd}(1+e^{Td^{\sigma}})2\pi
\end{align*}
where $M_d=\max_{|w|\leq d}\|(w-A)^{-1}\|$ as before.
Thus we have shown that in both cases, each term may be bounded
independently of $t$ and $s$, and so $V_{\beta}(t,s)$ is uniformly
bounded on $0\leq s\leq t\leq T$.
Next, we show that $(t,s)\mapsto V_{\beta}(t,s)$ is strongly continuous
for $0\leq s\leq t\leq T$. It follows from \eqref{Asigma} and by a
standard argument using Cauchy's Integral Formula that
\[ %\label{strong_cont}
V_{\beta}(t,s)A^{-\sigma}
=\frac{1}{2\pi i}\int_{\Gamma_{\phi}}w^{-\sigma}
e^{\int_s^tf_{\beta}(\tau,w)d\tau}(w-A)^{-1}dw
\]
(cf. \cite[p. 46]{HuangZheng2}). Then since $t \mapsto f_{\beta}(t,w)$
is continuous, using the above calculations
for $\|V_{\beta}(t,s)\|$, it follows by a dominated convergence argument
that $\|V_{\beta}(t,s)A^{-\sigma}-V_{\beta}(t_0,s_0)A^{-\sigma}\|\to 0$
as $(t,s) \to (t_0,s_0)$. Then, for $x\in \operatorname{Dom}(A^{\sigma})$,
we have
\begin{align*}
\|V_{\beta}(t,s)x-V_{\beta}(t_0,s_0)x\|
&\leq \|V_{\beta}(t,s)A^{-\sigma}-V_{\beta}(t_0,s_0)A^{-\sigma}\|\|A^{\sigma}x\| \\
&\to 0 \quad \text{as } (t,s) \to (t_0,s_0).
\end{align*}
Strong continuity of $V_{\beta}(t,s)$ then follows since
$\operatorname{Dom}(A^{\sigma})$ is dense in $X$
(Lemma~\ref{fractional_props} (iii)) and $V_{\beta}(t,s)$
is uniformly bounded.
Now, we show that the mapping $[s,T]\to X$ given by
$t\mapsto V_{\beta}(t,s)\chi$ is a classical solution of \eqref{2}
for $\chi \in X$. We have already established that
$t\mapsto V_{\beta}(t,s)\chi$ is continuous on $[s,T]$.
Next, we show that
$\frac{\partial}{\partial t}V_{\beta}(t,s)\chi =f_{\beta}(t,A)V_{\beta}(t,s)\chi$
for $t \in (s,T)$. We have
\begin{align} \label{C1f}
\frac{\partial}{\partial t}V_{\beta}(t,s)\chi
&= \frac{1}{2\pi i}\int_{\Gamma_{\phi}}
\Big(\frac{\partial}{\partial t}e^{\int_s^tf_{\beta}(\tau,w)d\tau}\Big)
(w-A)^{-1}\chi\,dw \\
&= \frac{1}{2\pi i}\int_{\Gamma_{\phi}}e^{\int_s^tf_{\beta}
(\tau,w)d\tau}f_{\beta}(t,w)(w-A)^{-1}\chi\,dw \\
\label{eleven}
&= \frac{1}{2\pi i}\int_{\Gamma_{\phi}}e^{\int_s^tf_{\beta}(\tau,w)d\tau}
a(t)w(w-A)^{-1}\chi \,dw \\
\label{twelve}
& \quad+ \frac{1}{2\pi i}\int_{\Gamma_{\phi}}
e^{\int_s^tf_{\beta}(\tau,w)d\tau} (-\beta w^{\sigma})(w-A)^{-1}\chi \; dw.
\end{align}
Now,
\begin{align*}
\text{Expression \eqref{eleven}}
&= a(t)\frac{1}{2\pi i}\int_{\Gamma_{\phi}}e^{\int_s^tf_{\beta}
(\tau,w)d\tau}((w-A)+A)(w-A)^{-1}\chi\,dw \\
&= \Big(a(t)\frac{1}{2\pi i}\int_{\Gamma_{\phi}}e^{\int_s^tf_{\beta}
(\tau,w)d\tau}dw\Big)\chi \\
&\quad + a(t)\frac{1}{2\pi i}\int_{\Gamma_{\phi}}e^{\int_s^tf_{\beta}(\tau,w)d\tau}A(w-A)^{-1}\chi\,dw \\
&= a(t)AV_{\beta}(t,s)\chi
\end{align*}
where we have used Cauchy's Theorem since
$w\mapsto e^{\int_s^tf_{\beta}(\tau,w)d\tau}$ is analytic, and also the
fact that $A$ is a closed operator.
Next, fix $t\in (s,T)$ and set
$G=\frac{1}{2\pi i}\int_{\Gamma_{\phi}}w^{\sigma}
e^{\int_s^tf_{\beta}(\tau,w)d\tau}(w-A)^{-1} dw$.
It is clear that $G$ is a bounded operator on $X$ by calculations
similarly used to calculate $\|V_{\beta}(t,s)\|$.
Also, by \eqref{fractional_power} and a standard argument using Cauchy's
Integral Formula (cf. \cite[Equation IX.1.52]{Kato}), it follows that
$A^{-\sigma}G=V_{\beta}(t,s)$. Hence, by the fact that
$A^{\sigma}=(A^{-\sigma})^{-1}$, we have
$\operatorname{Ran}(V_{\beta}(t,s))\subseteq
\operatorname{Ran}(A^{-\sigma}) = \operatorname{Dom}(A^{\sigma})$ and
$G=A^{\sigma}V_{\beta}(t,s)$. Hence
$\eqref{twelve}=-\beta G\chi=-\beta A^{\sigma}V_{\beta}(t,s)\chi$,
and altogether we have shown
$\frac{\partial}{\partial t}V_{\beta}(t,s)
=a(t)AV_{\beta}(t,s)\chi-\beta A^{\sigma}V_{\beta}(t,s)\chi
=f_{\beta}(t,A)V_{\beta}(t,s)\chi$ for $t\in (s,T)$.
Also by definition, $V_{\beta}(s,s)\chi=\chi$. Thus,
$t\mapsto V_{\beta}(t,s)\chi$ satisfies \eqref{2}.
Finally, calculation \eqref{C1f}--\eqref{twelve} shows that
$t\mapsto f_{\beta}(t,A)V_{\beta}(t,s)\chi$ is continuous on
$(s,T)$ since $t\mapsto e^{\int_s^tf_{\beta}(\tau,w)d\tau}f_{\beta}(t,w)$
is continuous. Therefore, we have that $t\mapsto V_{\beta}(t,s)\chi$
is continuously differentiable on $(s,T)$, and so we have shown
altogether that $t\mapsto V_{\beta}(t,s)\chi$ is a classical solution
of \eqref{2}.
It follows that problem \eqref{2} is well-posed due to uniqueness
of the solution $t\mapsto V_{\beta}(t,s)\chi$ and continuous dependence
of solutions on initial data, both of which are proved by standard
arguments (see e.g. \cite[Proof of Proposition~2.3]{Fury}).
\end{proof}
\begin{corollary}\label{approx1bound}
Let $0<\beta<1$ and let the operators $f_{\beta}(t,A), 0\leq t\leq T$ and
$V_{\beta}(t,s),0\leq s\leq t\leq T$ be defined under the hypotheses
of Proposition~$\ref{well-posed_approx1}$. Then for small $\beta$,
\[
\|V_{\beta}(t,s)\|\leq K'e^{K\beta^{-1/(\sigma-1)}}
\]
for all $0\leq s\leq t\leq T$ where $K$ and $K'$ are constants
independent of $\beta$, $t$, and $s$.
\end{corollary}
\begin{proof}
Let $0\leq s0$ and let $\alpha>1$ satisfy $\alpha (\pi/2-\theta)<\pi/2$.
Then $e^{-\epsilon A^{\alpha}},\epsilon> 0$ defined by
\begin{equation} \label{C_epsilon}
e^{-\epsilon A^{\alpha}}
= \frac{1}{2\pi i}\int_{\Gamma_{\phi}}e^{-\epsilon w^{\alpha}}(w-A)^{-1}dw
\end{equation}
is a strongly continuous holomorphic semigroup generated by the fractional
power $-A^{\alpha}$ where $\Gamma_{\phi}$ is similar to the contour
described in Proposition~\ref{well-posed_approx1} but with
$\pi/2\alpha>\phi>\pi/2-\theta$ (cf. \cite[Definition~3.4]{deL1}).
For $\epsilon>0$, set $C_{\epsilon}=e^{-\epsilon A^{\alpha}}$.
It follows that $C_{\epsilon}$ is injective for $\epsilon>0$
(cf. \cite[Lemma 3.1]{deL1}). We construct $C_{\epsilon}$-regularized
evolution systems as follows.
\begin{proposition} \label{1_epsilon}
Let $\epsilon>0$ and let $\alpha >1$ satisfy $\alpha (\pi/2-\theta)<\pi/2$.
For every $\chi\in X$, the evolution problem
\begin{equation} \label{1epsilon}
\begin{gathered}
\frac{du}{dt} = a(t)Au(t) \quad 0\leq s\leq t< T \\
u(s) = C_{\epsilon}\chi
\end{gathered}
\end{equation}
has a unique classical solution $u(t)=U_{\epsilon}(t,s)\chi$ where
\[
U_{\epsilon}(t,s)=\frac{1}{2\pi i}\int_{\Gamma_{\phi}}
e^{-\epsilon w^{\alpha}}e^{\left(\int_s^ta(\tau)d\tau\right)w}(w-A)^{-1}dw
\]
for all $0\leq s\leq t\leq T$
and $\Gamma_{\phi}$ is similar to the contour described in
Proposition~$\ref{well-posed_approx1}$ with $\pi/2\alpha>\phi>\pi/2-\theta$.
\end{proposition}
\begin{proof}
The proof is similar to that of Proposition~\ref{well-posed_approx1}.
In particular, $U_{\epsilon}(t,s)$ is a uniformly bounded operator on
$X$ for $0\leq s\leq t\leq T$ by the assumptions on $\alpha$.
Also, the function $t\mapsto U_{\epsilon}(t,s)\chi$ is a unique
classical solution of \eqref{1epsilon} since
$\frac{\partial}{\partial t}U_{\epsilon}(t,s)\chi =a(t)AU_{\epsilon}(t,s)\chi$
for $t\in (s,T)$, and by equation \eqref{C_epsilon},
\[
U_{\epsilon}(s,s)\chi = \frac{1}{2\pi i}\int_{\Gamma_{\phi}}
e^{-\epsilon w^{\sigma}}(w-A)^{-1}\chi\,dw
= e^{-\epsilon A^{\sigma}}\chi = C_{\epsilon}\chi.
\]
\end{proof}
\begin{lemma} \label{C_epsilonu(t)}
Let $\chi \in X$. If $u(t)$ is a classical solution of problem $\eqref{1}$,
then
\[
C_{\epsilon}u(t)=U_{\epsilon}(t,s)\chi \quad \text{for all } t\in [s,T].
\]
\end{lemma}
\begin{proof}
Since $C_{\epsilon}\in B(X)$ and $C_{\epsilon}$ commutes with $A$,
it is easily shown that $C_{\epsilon}u(t)$ is a classical solution
of \eqref{1epsilon}. The uniqueness of solutions from
Proposition~\ref{1_epsilon} then yields the desired result.
\end{proof}
To establish regularization, we will make use of the nature in
which the operators $f_{\beta}(t,A)$ approximate the operators $a(t)A$.
Motivated by the approximation condition, Condition A of Ames and Hughes
(cf. \cite[Definition~1]{AmesandHughes}), we demonstrate the following property.
\begin{lemma} \label{Condition_Ap}
Let $-A$ be the infinitesimal generator of a bounded holomorphic
semigroup of angle $\theta$, and let $0 \in \rho (A)$. Let $0<\beta<1$
and let the family of operators $f_{\beta}(t,A), 0\leq t\leq T$
be defined by $\eqref{f_beta}$. Then there exist positive constants $R$
and $\kappa$ each independent of $\beta$ and $t$ such that
$\operatorname{Dom}(A^{1+\kappa})\subseteq \operatorname{Dom}(f_{\beta}(t,A))$
and
\begin{equation} \label{Ap}
\|(-a(t)A+f_{\beta}(t,A))\psi\|\leq R\beta\|A^{1+\kappa}\psi\|
\end{equation}
for all $t\in [0,T]$ and for all $\psi\in \operatorname{Dom}(A^{1+\kappa})$.
\end{lemma}
Note that in the statement of the lemma we use implicitly
that $\operatorname{Dom}(A^{1+\kappa})\subseteq \operatorname{Dom}(A)$
which follows from Lemma~\ref{fractional_props} (ii).
\begin{proof}
First, assume $\theta \in (0,\pi/4]$ so that $f_{\beta}(t,A)$ is defined
as in Proposition~\ref{well-posed_approx1} where $\sigma$ satisfies
$\sigma>1$ and $\sigma (\pi/2-\theta)<\pi/2$. Then for
$\psi \in \operatorname{Dom}(A^{\sigma})$ and $t\in [0,T]$, we have
$\psi \in \operatorname{Dom}(f_{\beta}(t,A))$ and
\[
\|(-a(t)A+f_{\beta}(t,A))\psi\|
= \|(-a(t)A+(a(t)A-\beta A^{\sigma}))\psi\|
= \beta \|A^{\sigma}\psi\|.
\]
Hence, \eqref{Ap} is satisfied with $R=1$ and $\kappa=\sigma-1$.
Next, we assume that $\theta \in (\pi/4,\pi/2]$ in which case $f_{\beta}(t,A)$
is defined as in Proposition~\ref{well-posed_approx2}.
Then $f_{\beta}(t,A)$ is a bounded, everywhere defined operator
and so $\operatorname{Dom}(f_{\beta}(t,A)) = X$ for each $t\in [0,T]$.
Further, for $\psi \in \operatorname{Dom}(A^2)$,
\begin{align*}
\|(-a(t)A+f_{\beta}(t,A))\psi\|
&= \|(-a(t)A+a(t)A(I+\beta A)^{-1})\psi\| \\
&= \|-a(t)A(I-(I+\beta A)^{-1})\psi\| \\
&= \|-a(t)A(\beta A(I+\beta A)^{-1})\psi\| \\
&= \|-a(t)\beta (I+\beta A)^{-1}A^2\psi\| \\
&\leq B \beta \|(I+\beta A)^{-1}\| \|A^2\psi\| \\
&\leq B C \beta \|A^2 \psi\|,
\end{align*}
where $B= \max_{t\in [0,T]}|a(t)|$ and $C$ is as in the proof of
Proposition~\ref{well-posed_approx2}. Hence, \eqref{Ap} is satisfied
with $R=BC$ and $\kappa =1$.
\end{proof}
In light of Lemma~\ref{Condition_Ap}, for each $t\in [0,T]$,
we define the operator $g_{\beta}(t,A)$ in $X$ by
\begin{equation} \label{g_beta}
g_{\beta}(t,A)x=-a(t)Ax+f_{\beta}(t,A)x
\end{equation}
for $x\in \operatorname{Dom}(A)\cap \operatorname{Dom}(f_{\beta}(t,A))$.
Properties of the operators $g_{\beta}(t,A), 0\leq t\leq T$ and
associated evolutions systems will be used heavily in proving
H\"{o}lder-continuous dependence on modeling, those of which we
provide now in the following proposition.
\begin{proposition} \label{W_beta}
Let $-A$ be the infinitesimal generator of a bounded holomorphic semigroup
of angle $\theta$, and let $0\in \rho (A)$. For $0<\beta<1$, let the
operators $f_{\beta}(t,A),0\leq t\leq T$ and $g_{\beta}(t,A)$, $0\leq t\leq T$
be defined by $\eqref{f_beta}$ and $\eqref{g_beta}$ respectively.
Then there exists an evolution system $W_{\beta}(t,s), 0\leq s\leq t\leq T$
associated with the family $g_{\beta}(t,A), 0\leq t\leq T$ satisfying the
following properties:
\begin{itemize}
\item[(i)] $\|W_{\beta}(t,s)\| \leq L$ for all $0\leq s\leq t\leq T$ where
$L$ is a constant independent of $t$, $s$, and $\beta$.
\item[(ii)] $\frac{\partial}{\partial t}W_{\beta}(t,s)\chi
=g_{\beta}(t,A)W_{\beta}(t,s)\chi$ for $0\leq s d$ where $d$ is the radius of the
disk contained in $\rho(A)$ as in Figure~\ref{fig:Gammadisk} and
Figure~\ref{fig:Gammaavoid}. For the pieces
\begin{gather*}
\Gamma^1 = \Gamma_1 = \{re^{i\phi} : r\geq
\beta^{-1/\sigma}(t-s)^{-1/\sigma}\}, \\
\Gamma^3 = \Gamma_7 = \{re^{-i\phi} : r\geq \beta^{-1/\sigma}(t-s)^{-1/\sigma}\},
\end{gather*}
we have the calculation
\begin{align*}
\|\int_{\Gamma_1\cup \Gamma_7}\|
= \|\int_{\Gamma^1\cup \Gamma^3}\|
&\leq M_1' \int_{\Gamma^1\cup \Gamma^3}\big|e^{-\beta (t-s)w^{\sigma}}\big|\;
|w|^{-1}|dw| \\
&= 2M_1'\int_{\beta^{-1/\sigma}(t-s)^{-1/\sigma}}^{\infty}
e^{-\beta (t-s)r^{\sigma}\cos\sigma \phi} r^{-1}dr \\
&= 2M_1'\int_1^{\infty}e^{-x^{\sigma}\cos\sigma \phi} x^{-1}dx \\
&\leq 2M_1'\int_1^{\infty}e^{-x^{\sigma}\cos\sigma \phi} dx
\leq K
\end{align*}
where $K$ is a constant independent of $t$, $s$, and $\beta$ since
$\sigma>1$ and $0< \sigma \phi <\pi/2$ because $\pi/2\sigma>\phi>\pi/2-\theta$.
Also, as in the proof of Proposition~\ref{well-posed_approx1}, in either
of the two cases, the remaining pieces of the contour may be bounded
independently of $t$, $s$, and $\beta$. Hence (i)--(iii) are satisfied
and the proposition is proved when $\theta \in (0,\pi/4]$.
If, on the other hand, $\theta \in (\pi/4,\pi/2]$ as in
Proposition~\ref{well-posed_approx2}, then
$g_{\beta}(t,A)=-a(t)A+a(t)A(I+\beta A)^{-1}$ and in this case,
we use perturbation theory to construct an evolution system
$W_{\beta}(t,s), 0\leq s\leq t\leq T$ satisfying (i)--(iii).
We've seen so far that $A(I+\beta A)^{-1}$ is a bounded operator on $X$.
Then since $-A$ generates a bounded holomorphic semigroup of angle $\theta$,
it follows that $-(A-A(I+\beta A)^{-1})$ is also the infinitesimal generator
of a holomorphic semigroup of the same angle (cf. \cite[Corollary~3.2.2]{Pazy}).
Set $G_{\beta}=A-A(I+\beta A)^{-1}$. It is shown in \cite{HuangZheng}
that $\mathbb{C}\backslash S_{\pi-2\theta} \subseteq \rho(G_{\beta})$
where $S_{\pi-2\theta}=\{re^{i\theta'} : r> 0, \; |\theta'| <\pi-2\theta\}$,
and
\[
\|(w-G_{\beta})^{-1}\| \leq \frac{M}{|w|} \quad \text{for }
w\in \mathbb{C}\backslash S_{\pi-2\theta}
\]
where $M$ is a constant independent of $\beta$
(cf. \cite[Theorem~2.1]{HuangZheng}). Hence for $0\leq s\leq t\leq T$,
the operator $W_{\beta}(t,s)$ defined by
\[
W_{\beta}(t,s) = \begin{cases}
\frac{1}{2\pi i}\int_{\Gamma_{\phi}}
e^{-(\int_s^ta(\tau)d\tau)w}(w-G_{\beta})^{-1}\,dw & 0\leq s\phi>\pi-2\theta$, is a well-defined uniformly bounded operator
satisfying $\|W_{\beta}(t,s)\|\leq L$ for $0\leq s\leq t\leq T$
where $L$ is a constant independent of $\beta$. Hence, (i) is satisfied.
Also, similar to calculation \eqref{C1f}--\eqref{twelve}, it is standard
to show that for every $\chi \in X$,
$\frac{\partial}{\partial t}W_{\beta}(t,s)\chi=-a(t)G_{\beta}W_{\beta}(t,s)\chi
=g_{\beta}(t,A)W_{\beta}(t,s)\chi$ for $0\leq s0$. Then
\[
U_{\epsilon}(t,s)W_{\beta}(t,s)=C_{\epsilon}V_{\beta}(t,s)
=W_{\beta}(t,s)U_{\epsilon}(t,s)
\]
for all $0\leq s\leq t\leq T$.
\end{corollary}
\begin{proof}
The result follows from uniqueness of solutions as each term applied
to $\chi\in X$ is a classical solution of the well-posed evolution
problem \eqref{2} with initial data $C_{\epsilon}\chi$.
\end{proof}
\section{H\"{o}lder-continuous dependence on modeling}
\label{HolderCDM}
We now use the results of Section~\ref{well-posed} and Section~\ref{lemmas} to prove H\"{o}lder-continuous dependence on modeling for the problems \eqref{1} and \eqref{2}, meaning a small change in the models from \eqref{1} to \eqref{2} implies a small change in the corresponding solutions. Again, as in Section~\ref{well-posed} and Section~\ref{lemmas}, we assume that $-A$ generates a bounded holomorphic semigroup $T(t)$ of angle $\theta$ on $X$ and $0 \in \rho(A)$. For $z\in S_{\theta}$, let us denote $T(z)$ by $T(z)=e^{-zA}$ and also define $e^{-zA}$ to be the identity operator when $z=0$.
Assume $u(t)$ and $v_{\beta}(t)$ are classical solutions of \eqref{1} and \eqref{2} respectively where $\chi \in X$ and let $\epsilon>0$ be arbitrary. Then since $C_{\epsilon}$ is bounded and since $e^{-zA}$ is uniformly bounded in each sector $S_{\theta_1}$, $\theta_1<\theta$ (Definition~\ref{holomorphic_semigroup} (iii)), we may define for $\theta_1 \in (0,\theta)$ and for $\zeta = t+re^{\pm i\theta_1}$ in the bent strip $S=\{\zeta = t+re^{\pm i\theta_1} : s\leq t\leq T,\; r \geq 0\}$, \[\phi_{\epsilon}(\zeta) = e^{-(re^{\pm i\theta_1})A}C_{\epsilon}(u(t)-v_{\beta}(t)).\] Ultimately, we will apply Carleman's Inequality (cf. \cite{Miller2}) to a function related to $\phi_{\epsilon}(\zeta)$ on the bent strip $S$. Our methods are motivated by Agmon and Nirenberg \cite{AN}.
\begin{lemma} \label{u(zeta)_v(zeta)}
Let $\epsilon>0$. Then
\[
\phi_{\epsilon}(\zeta) = e^{-(re^{\pm i\theta_1})A}
(U_{\epsilon}(t,s)\chi-C_{\epsilon}V_{\beta}(t,s)\chi)
\]
for all $\zeta = t+re^{\pm i\theta}\in S$.
\end{lemma}
The above lemma follows immediately from Lemma~\ref{C_epsilonu(t)}
and Corollary~\ref{well-posed_both}.
\begin{lemma}[{\cite[p. 148]{AN}}] \label{AN}
Let $\phi(z)$ be a continuous and bounded complex function on the bent
strip $S=\{z=x+\eta e^{\pm i\theta} : s\leq x \leq T, \; \eta \geq 0\}$.
For $\zeta = t+re^{\pm i\theta}\in S$, define
\[
\Phi(\zeta)=-\frac{1}{\pi}\int \int_S \phi(z)
\Big(\frac{1}{z-\zeta}+\frac{1}{\bar{z}+1+\zeta}\Big)dx d\eta.
\]
Then $\Phi(\zeta)$ is absolutely convergent,
$\bar{\partial}\Phi(\zeta)=\phi(\zeta)$ where $\bar{\partial}$ denotes
the Cauchy-Riemann operator, and there exists a constant $\tilde{K}$
such that
\[
\int_{-\infty}^{\infty}\big|\frac{1}{z-\zeta}+\frac{1}{\bar{z}+1+\zeta}
\big|d\eta\leq \tilde{K}\Big(1+{\rm{log}}\frac{1}{|x-t|}\Big)
\]
if $x \neq t$.
\end{lemma}
We prove now the following theorem establishing H\"{o}lder-continuous
dependence on modeling for problems \eqref{1} and \eqref{2}.
We will use the results of this theorem to aid us in proving regularization
in Section~\ref{reg_section}.
\begin{theorem} \label{approx_thm}
Let $-A$ be the infinitesimal generator of a bounded holomorphic
semigroup of angle $\theta$ on a Banach space $X$ and let $0 \in \rho(A)$.
For $0<\beta<1$, let the family of operators $f_{\beta}(t,A), 0\leq t\leq T$
be defined by $\eqref{f_beta}$. Let $u(t)$ and $v_{\beta}(t)$ be classical
solutions of $\eqref{1}$ and $\eqref{2}$ respectively with $\chi \in X$,
and assume that there exists a constant $M'\geq 0$ such that
$\|A^{2+\kappa}u(t)\|\leq M'$ for all $t\in [s,T]$ where $\kappa$ is
defined by Lemma~$\ref{Condition_Ap}$. Then there exist constants
$\tilde{C}$ and $M$ independent of $\beta$ such that for $0\leq s\leq t 0$, $\chi \in X$, and define
\[
\phi_{\epsilon}(\zeta) = e^{-(re^{\pm i\theta_1})A}C_{\epsilon}(u(t)
-v_{\beta}(t))
\]
for $\zeta=t+re^{\pm i\theta_1} \in S$ as in the discussion preceding
Lemma~\ref{u(zeta)_v(zeta)}. Intending to apply Lemma~\ref{AN},
we determine $\bar{\partial}\phi_{\epsilon}(\zeta)$.
Since $e^{-(re^{\pm i\theta_1})A}$ is bounded for every $r\geq 0$ and
since $C_{\epsilon}$ commutes with $A$, we have by
Lemma~\ref{u(zeta)_v(zeta)},
\begin{align*}
\frac{\partial}{\partial t}\phi_{\epsilon}(\zeta)
&= \frac{\partial}{\partial t}e^{-(re^{\pm i\theta_1})A}
(U_{\epsilon}(t,s)\chi - C_{\epsilon}V_{\beta}(t,s)\chi) \\
&= e^{-(re^{\pm i\theta_1})A}(\frac{\partial}{\partial t}
U_{\epsilon}(t,s)\chi - C_{\epsilon}\frac{\partial}{\partial t}
V_{\beta}(t,s)\chi) \\
&= e^{-(re^{\pm i\theta_1})A}(a(t)AU_{\epsilon}(t,s)\chi
- f_{\beta}(t,A)C_{\epsilon}V_{\beta}(t,s)\chi) .
\end{align*}
Also, since $-A$ generates $e^{-zA}$ and since both $U_{\epsilon}(t,s)\chi$
and $C_{\epsilon}V_{\beta}(t,s)$ are in $\operatorname{Dom}(A)$, we have
\begin{align*}
\frac{\partial}{\partial r}\phi_{\epsilon}(\zeta)
&= \frac{\partial}{\partial r}e^{-(re^{\pm i\theta_1})A}(U_{\epsilon}(t,s)\chi
- C_{\epsilon}V_{\beta}(t,s)\chi) \\
&= e^{-(re^{\pm i\theta_1})A}(-e^{\pm i\theta_1}A)(U_{\epsilon}(t,s)\chi
- C_{\epsilon}V_{\beta}(t,s)\chi).
\end{align*}
Therefore, by definition of the Cauchy-Riemann operator $\bar{\partial}$,
\begin{equation} \label{delbar}
\begin{aligned}
\bar{\partial}\phi_{\epsilon}(\zeta)
&= \frac{1}{2i\;\sin (\pm \theta_1)}
\Big(e^{\pm i\theta_1}\frac{\partial}{\partial t}\phi_{\epsilon}(\zeta)
-\frac{\partial}{\partial r}\phi_{\epsilon}(\zeta)\Big) \\
&= \frac{e^{\pm i\theta_1}}{2i\;\sin (\pm \theta_1)}
\Big[e^{-(re^{\pm i\theta_1})A}(a(t)AU_{\epsilon}(t,s)\chi
- f_{\beta}(t,A)C_{\epsilon}V_{\beta}(t,s)\chi) \\
&\quad + e^{-(re^{\pm i\theta_1})A}(AU_{\epsilon}(t,s)\chi
- AC_{\epsilon}V_{\beta}(t,s)\chi)\Big].
\end{aligned}
\end{equation}
Following \cite{AN}, define
\[
\Phi_{\epsilon}(\zeta)=-\frac{1}{\pi}\iint_S \bar{\partial}\phi_{\epsilon}(z)
\Big(\frac{1}{z-\zeta}+\frac{1}{\bar{z}+1+ \zeta}\Big)dx d\eta,
\]
where $z = x +\eta e^{\pm i \theta_1}$ and $\zeta =t+re^{\pm i\theta_1}$
are in $S$. In order to apply Lemma~\ref{AN}, we show that
$\bar{\partial}\phi_{\epsilon}(z)$ is continuous and bounded on $S$.
We first show that it is bounded on $S$. Let $z=x+\eta e^{\pm i \theta_1} \in S$
be arbitrary. We have from \eqref{delbar},
\begin{align*}
\|\bar{\partial}\phi_{\epsilon}(z)\| \
&\leq \frac{1}{2|\sin \theta_1|}\; \|e^{-(\eta e^{\pm i\theta_1})A}\|
\Big( \|a(x)AU_{\epsilon}(x,s)\chi - f_{\beta}(x,A)C_{\epsilon}V_{\beta}(x,s)
\chi\| \\
&\quad + \|AU_{\epsilon}(x,s)\chi - AC_{\epsilon}V_{\beta}(x,s)\chi\|
\Big) \\
&\leq \frac{\Theta}{2|\sin \theta_1|} \Big( \|a(x)AU_{\epsilon}(x,s)
\chi-a(x)AC_{\epsilon}V_{\beta}(x,s)\chi\| \\
&\quad + \|a(x)AC_{\epsilon}V_{\beta}(x,s)\chi-f_{\beta}(x,A)
C_{\epsilon}V_{\beta}(x,s)\chi\| \\
&\quad + \|AU_{\epsilon}(x,s)\chi-AC_{\epsilon}V_{\beta}(x,s)\chi\|\Big) \\
&\leq \frac{\Theta}{2|\sin \theta_1|}
\Big((B+1)\|AU_{\epsilon}(x,s)\chi-AC_{\epsilon}V_{\beta}(x,s)\chi\| \\
& \quad + \|a(x)AC_{\epsilon}V_{\beta}(x,s)\chi-f_{\beta}(x,A)C_{\epsilon}
V_{\beta}(x,s)\chi\|\Big)
\end{align*}
where we have set $\Theta=\max_{r \geq 0}\|e^{-(re^{\pm i\theta_1})A}\|$
and $B=\max_{t\in [0,T]}|a(t)|$. Since
$U_{\epsilon}(x,s)\chi \in C_{\epsilon}(X)\subseteq \operatorname{Dom}(A^j)$
for every $j\in \mathbb{N}$ (cf. \cite[Proposition~2.10]{deL1}), it follows
that $AU_{\epsilon}(x,s)\chi \in \operatorname{Dom}(A^j)$ for every $j$ as well.
Therefore, we have $AU_{\epsilon}(x,s)\chi \in \operatorname{Dom}(A^{1+\kappa})$
by Lemma~\ref{fractional_props} (ii). Hence, by Corollary~\ref{factor_evsys},
Proposition~\ref{W_beta}, and Lemma~\ref{Condition_Ap},
\begin{equation} \label{ev_sys_property}
\begin{aligned}
\|AU_{\epsilon}(x,s)\chi-AC_{\epsilon}V_{\beta}(x,s)\chi\|
&= \|AU_{\epsilon}(x,s)\chi-AW_{\beta}(x,s)U_{\epsilon}(x,s)\chi\| \\
&= \|(I-W_{\beta}(x,s))AU_{\epsilon}(x,s)\chi\| \\
&= \big\|\int_s^{x}\frac{\partial}{\partial \tau}(W_{\beta}(x,\tau)
AU_{\epsilon}(x,s)\chi)d\tau\big\| \\
&= \big\|\int_s^{x}-W_{\beta}(x,\tau)g_{\beta}(\tau,A)AU_{\epsilon}(x,s)
\chi d\tau\big\| \\
&\leq \int_s^{x}L\|g_{\beta}(\tau,A)AU_{\epsilon}(x,s)\chi\|d\tau \\
&\leq T LR\beta \|A^{1+\kappa}AU_{\epsilon}(x,s)\chi\|.
\end{aligned}
\end{equation}
Also, by Lemma~\ref{Condition_Ap},
\begin{align*}
\|a(x)AC_{\epsilon}V_{\beta}(x,s)\chi-f_{\beta}(x,A)C_{\epsilon}V_{\beta}(x,s)
\chi\|
&= \|(-a(x)A+f_{\beta}(x,A))C_{\epsilon}V_{\beta}(x,s)\chi\| \\
&\leq R\beta \|A^{1+\kappa}C_{\epsilon}V_{\beta}(x,s)\chi\| \\
&= R\beta \|A^{1+\kappa}W_{\beta}(x,s)U_{\epsilon}(x,s)\chi\| \\
&= R\beta \|W_{\beta}(x,s)A^{1+\kappa}U_{\epsilon}(x,s)\chi\| \\
&\leq LR\beta \|A^{1+\kappa}U_{\epsilon}(x,s)\chi\|.
\end{align*}
Thus we have shown that
\[
\|\bar{\partial}\phi_{\epsilon}(z)\|
\leq \frac{\Theta (T+1)LR\beta }{2 |\sin \theta_1|}
\Big((B+1)\|A^{1+\kappa}AU_{\epsilon}(x,s)\chi\|+\|A^{1+\kappa}U_{\epsilon}(x,s)
\chi\|\Big).
\]
Now, by the assumption that $\|A^{2+\kappa}u(t)\|\leq M'$ for all $t\in [s,T]$
and by Lemma~\ref{fractional_props} (iv), we have
$\|A^{1+\kappa}u(t)\|=\|A^{-1}A^{2+\kappa}u(t)\|\leq M''$ for all $t\in [s,T]$
for some constant $M''\geq 0$, where we have used the fact that $0 \in \rho(A)$.
By the fact that $C_{\epsilon}=e^{-\epsilon A^{\alpha}}, \epsilon>0$ is a
holomorphic semigroup, set $J= \sup_{0<\epsilon<1}\|C_{\epsilon}\|$.
Then for small $\epsilon>0$, since $C_{\epsilon}$ commutes with $A$,
we have from Lemma~\ref{C_epsilonu(t)},
\begin{equation} \label{p(D)u(t)}
\|A^{1+\kappa}U_{\epsilon}(x,s)\chi\|=\|A^{1+\kappa}C_{\epsilon}u(x)\|
=\|C_{\epsilon}A^{1+\kappa}u(x)\|\leq JM''
\end{equation}
and similarly
$\|A^{1+\kappa}AU_{\epsilon}(x,s)\chi\|=\|A^{2+\kappa}U_{\epsilon}(x,s)\chi\|
\leq JM'$. Therefore, we have shown that
\begin{equation} \label{delbar_bdd}
\|\bar{\partial}\phi_{\epsilon}(z)\| \leq \beta C',
\end{equation}
where $C'$ is a constant independent of $\epsilon$ and also of $\beta$
since $L$ is independent of $\beta$ (Proposition~\ref{W_beta} (i)).
We have shown that $\bar{\partial}\phi_{\epsilon}(z)$ is bounded on $S$.
It follows easily that $\bar{\partial}\phi_{\epsilon}(z)$ is also continuous
on $S$. Having satisfied the hypotheses of Lemma~\ref{AN}, it follows
that $\Phi_{\epsilon}(\zeta)$ is absolutely convergent,
$\bar{\partial}\Phi_{\epsilon}(\zeta)=\bar{\partial}\phi_{\epsilon}(\zeta)$,
and there exists a constant $\tilde{K}$ such that, for $x\neq t$,
\[
\int_{-\infty}^{\infty}\big|\frac{1}{z-\zeta}+\frac{1}{\bar{z}+1+ \zeta}
\big|d\eta \leq \tilde{K}\Big(1+\text{log}\frac{1}{|x-t|}\Big).
\]
We now construct a candidate to satisfy Carleman's Inequality.
Define $\Psi_{\epsilon}:S\to \mathbb{C}$ by
\[
\Psi_{\epsilon}(\zeta)=x^*(\phi_{\epsilon}(\zeta)-\Phi_{\epsilon}(\zeta))
\]
where $x^*\in X^*$, the dual space of $X$, is arbitrary.
For $\zeta $ in the interior of $S$, using the results from Lemma~\ref{AN},
\[
\bar{\partial}\Psi_{\epsilon}(\zeta)
= x^*(\bar{\partial}\phi_{\epsilon}(\zeta)-\bar{\partial}\Phi_{\epsilon}(\zeta))
= x^*(0)=0.
\]
Therefore, $\Psi_{\epsilon}$ is analytic on the interior of $S$
(cf. \cite[Theorem 11.2]{Rudin}). \\
\indent Next, we show that $\Psi_{\epsilon}$ is bounded on $S$.
Similar to the calculation in \eqref{ev_sys_property}, and
using \eqref{p(D)u(t)}, we have
\begin{equation} \label{phi_bdd}
\begin{aligned}
\|\phi_{\epsilon}(\zeta)\|
&= \|e^{-(re^{\pm i\theta_1})A}(U_{\epsilon}(t,s)\chi-C_{\epsilon}V_{\beta}(t,s)\chi)\| \\
&\leq \Theta \|U_{\epsilon}(t,s)\chi-C_{\epsilon}V_{\beta}(t,s)\chi\| \\
&\leq \Theta T LR\beta \|A^{1+\kappa}U_{\epsilon}(t,s)\chi\|
\leq \beta K'
\end{aligned}
\end{equation}
where $K'$ is a constant independent of $\beta$, $\epsilon$, and $\zeta$.
Next, from \eqref{delbar_bdd} and Lemma~\ref{AN},
\begin{equation} \label{Phi_bdd}
\begin{aligned}
\|\Phi_{\epsilon}(\zeta)\|
&= \Big\|-\frac{1}{\pi}\int \int_S \bar{\partial}\phi_{\epsilon}(z)
\Big(\frac{1}{z-\zeta}+\frac{1}{\bar{z}+1+\zeta}\Big)dxd\eta \Big\| \\
&\leq \frac{1}{\pi}\beta C'\int_s^T
\Big(\int_{-\infty}^{\infty} \big|\frac{1}{z-\zeta}+\frac{1}{\bar{z}+1+\zeta}\big|
d\eta \Big) dx \\
&\leq \beta \frac{\tilde{K}}{\pi}C'\int_s^T
\Big(1+\text{log}\frac{1}{|x - t|}\Big) dx
\leq \beta C'
\end{aligned}
\end{equation}
for a possibly different constant $C'$ independent of $\beta$, $\epsilon$,
and $\zeta$. Then from \eqref{phi_bdd} and \eqref{Phi_bdd}, we have for
$\zeta=t+re^{\pm i\theta_1} \in S$,
\begin{equation} \label{Psi_bdd}
\begin{aligned}
|\Psi_{\epsilon}(\zeta)|
&= |x^*(\phi_{\epsilon}(\zeta)-\Phi_{\epsilon}(\zeta))| \\
&\leq \|x^*\|\big(\|\phi_{\epsilon}(\zeta)\|+\|\Phi_{\epsilon}(\zeta)\|\big) \\
&\leq \beta M\|x^*\|
\end{aligned}
\end{equation}
where $M$ is a constant independent of $\beta$, $\epsilon$, and $\zeta$.
We have shown that $\Psi_{\epsilon}$ is bounded on $S$.
It is easy to show that $\Psi_{\epsilon}$ is also continuous on $S$,
and we have already seen that $\Psi_{\epsilon}$ is analytic on
the interior of $S$. By Carleman's Inequality (cf. \cite{Miller2}),
we then obtain
\begin{equation} \label{3LT}
|\Psi_{\epsilon}(t)| \leq M_{\epsilon}(s)^{1-h(t)}M_{\epsilon}(T)^{h(t)},
\end{equation}
for $s\leq t\leq T$, where
$M_{\epsilon}(t)= \sup_{r \geq 0}|\Psi_{\epsilon}(t+re^{\pm i\theta_1})|$ and
$h$ is a harmonic function which is bounded and continuous on $S$ and
assumes the values $0$ and $1$ respectively on the left and right hand
boundary curves of $S$. Note that
\begin{align*}
\|\phi_{\epsilon}(s+re^{\pm i\theta_1})\|
&= \|e^{-(re^{\pm i\theta_1})A}(U_{\epsilon}(s,s)\chi-C_{\epsilon}V_{\beta}(s,s)\chi)\| \\
&= \|e^{-(re^{\pm i\theta_1})A}(C_{\epsilon}\chi-C_{\epsilon}\chi)\| =0.
\end{align*}
Then from \eqref{Phi_bdd}, we have
\[
|\Psi_{\epsilon}(s+re^{\pm i\theta_1})| \leq \|x^*\|
\left(\|\phi_{\epsilon}(s+re^{\pm i\theta_1})\|
+\|\Phi_{\epsilon}(s+re^{\pm i\theta_1})\|\right)
\leq \|x^*\| \beta C',
\]
and so
\begin{equation} \label{M(0)}
M_{\epsilon}(s) = \sup_{r \geq 0} |\Psi_{\epsilon}(s+re^{\pm i\theta_1})|
\leq \beta C'\|x^*\|.
\end{equation}
Also, from \eqref{Psi_bdd} and the fact that $0<\beta<1$, we have
\begin{equation} \label{M(T)}
M_{\epsilon}(T) = \max_{r \geq 0} |\Psi_{\epsilon}(T+re^{\pm i\theta_1})|
\leq M\|x^*\|.
\end{equation}
From \eqref{3LT}, \eqref{M(0)}, and \eqref{M(T)}, it follows that
for $s\leq t0,\;
t\in [s,T]\}
\]
is a family of regularizing operators for the problem \eqref{1} where
$\tilde{V}_{\beta}(t,s), 0\leq s\leq t\leq T$ is the evolution system
of Corollary~$\ref{well-posed_both}$ corresponding to the operators
$f_{\beta}(t,A-\lambda), 0\leq t\leq T$ defined by
\begin{equation}
\label{f_beta_Atilde}
f_{\beta}(t,A-\lambda) = \begin{cases}
a(t)(A-\lambda)-\beta (A-\lambda)^{\sigma} & \text{if } \theta \in (0,\pi/4] \\
a(t)(A-\lambda)(I+\beta (A-\lambda))^{-1} & \text{if } \theta \in (\pi/4,\pi/2]
\end{cases}
\end{equation}
where $\sigma>1$ when $\theta \in (0,\pi/4]$. The regularization parameter
$\beta$ is chosen as follows: for a given perturbed initial data
$\chi_{\delta}$ where $\|\chi-\chi_{\delta}\|\leq \delta$,
\[
\beta = \begin{cases}
(-2K/\ln \delta)^{\sigma-1} & \text{if } \theta \in (0,\pi/4] \\
-2CT/\ln \delta & \text{if } \theta \in (\pi/4,\pi/2]
\end{cases}
\]
where $K$ and $C$ are constants independent of $\delta$.
\end{theorem}
\begin{proof}
First, in accordance with Theorem~\ref{approx_thm}, assume that $-A$
generates a bounded holomorphic semigroup and that $0\in \rho(A)$.
Let $u(t)$ be a classical solution of \eqref{1} with initial data
$\chi $ and assume $u(t)$ satisfies the stabilizing condition of
Theorem~\ref{approx_thm}, that is $\|A^{2+\kappa}u(t)\| \leq M'$
for all $t\in [s,T]$. Also, let $\|\chi-\chi_{\delta}\|\leq \delta$.
Let $v_{\beta}(t)$ be a solution of \eqref{2} and let
$V_{\beta}(t,s), 0\leq s\leq t\leq T$ be the evolution system given
in Corollary~\ref{well-posed_both}. Then for $0\leq s\leq t\leq T$,
we have $v_{\beta}(t)=V_{\beta}(t,s)\chi$ and
\begin{equation} \label{reg}
\begin{aligned}
\|u(t)-V_{\beta}(t,s)\chi_{\delta}\|
&\leq \|u(t)-v_{\beta}(t)\|+\|V_{\beta}(t,s)\chi-V_{\beta}(t,s)\chi_{\delta}\| \\
&\leq \|u(t)-v_{\beta}(t)\| + \delta \|V_{\beta}(t,s)\|.
\end{aligned}
\end{equation}
First consider $0\leq s\leq t0, \; t\in [s,T]\}$ is a family of regularizing
operators for problem \eqref{1}.
Now, for the general case, assume that $-A$ generates a holomorphic
semigroup of angle $\theta$ on $X$. It is known that for
$\theta' \in (0,\theta)$, then there exists $\lambda \in \mathbb{R}$
such that $-A+\lambda$ is the infinitesimal generator of a bounded
holomorphic semigroup of angle $\theta'$ on $X$ and
$0 \in \rho (A-\lambda)$ (cf. \cite[Section~X.8, p. 253]{ReedandSimon}).
Let $u(t)$ be a classical solution of \eqref{1} with initial data $\chi \in X$.
It is easily shown that $w(t)=e^{-(\int_s^ta(\tau)d\tau)\lambda}u(t)$
is then a classical solution of the evolution problem
\begin{equation} \label{1shifted}
\begin{gathered}
\frac{dw}{dt} = a(t)(A-\lambda)w(t) \quad 0\leq s\leq t0, \; t\in [s,T]\}$
is a family of regularizing operators for the problem \eqref{1shifted} where
$\tilde{V}_{\beta}(t,s), 0\leq s\leq t\leq T$ is the evolution system
of Corollary~\ref{well-posed_both} corresponding to the family of operators
$f_{\beta}(t,A-\lambda), 0\leq t\leq T$ defined by \eqref{f_beta_Atilde}.
Hence, given $\delta >0$ and $\|\chi-\chi_{\delta}\|\leq \delta$,
there exists $\beta>0$, such that $\beta \to 0$ as $\delta\to 0$ and
\begin{align*}
\|u(t)-e^{\left(\int_s^ta(\tau)d\tau\right)\lambda}
\tilde{V}_{\beta}(t,s)\chi_{\delta}\|
&= e^{\left(\int_s^ta(\tau)d\tau\right)\lambda}\|w(t)
-\tilde{V}_{\beta}(t,s)\chi_{\delta}\| \\
&\to 0 \quad \text{as } \delta \to 0
\end{align*}
for $0\leq s\leq t\leq T$, proving that
$\{e^{(\int_s^ta(\tau)d\tau)\lambda}\tilde{V}_{\beta}(t,s) :
\beta>0, \; t\in [s,T]\}$ is a family of regularizing operators
for the problem \eqref{1}.
\end{proof}
\section{Examples in $L^p$ spaces}
\label{ex_section}
In this final section, we apply the theory of regularization in
Section~\ref{reg_section} to ill-posed partial differential equations
in $L^p$ spaces where $A$ is a strongly elliptic differential operator.
We will use the following notation (cf. \cite[Chapter~7.1]{Pazy}).
For an $n$-tuple of nonnegative integers
$\alpha=(\alpha_1,\alpha_2,\dots,\alpha_n)$ (called a multi-index),
we define $|\alpha|=\sum_{i=1}^n\alpha_i$ and
$x^{\alpha}=x_1^{\alpha_1}x_2^{\alpha_2}\dots x_n^{\alpha_n}$ for
$x=(x_1,x_2,\dots,x_n) \in \mathbb{R}^n$. Also, denote
$D_k=\partial/\partial x_k$ and $D=(D_1,D_2,\dots,D_n)$.
Then $D^{\alpha}$ is defined by
\[
D^{\alpha}=D_1^{\alpha_1}D_2^{\alpha_2}\dots D_n^{\alpha_n}
=\frac{\partial^{\alpha_1}}{\partial x_1^{\alpha_1}}
\frac{\partial^{\alpha_2}}{\partial x_2^{\alpha_2}}\dots
\frac{\partial^{\alpha_n}}{\partial x_n^{\alpha_n}}.
\]
Finally, for a fixed domain $\Omega$ in $\mathbb{R}^n$, $W^{m,p}(\Omega)$
will denote the Sobolev space consisting of functions $u \in L^p(\Omega)$
whose derivatives $D^{\alpha}u$, in the sense of distributions,
of order $k\leq m$ are in $L^p(\Omega)$. Also, $W_0^{m,p}(\Omega)$
denotes the subspace of functions in $W^{m,p}(\Omega)$ with compact
support in $\Omega$.
Let $\Omega$ be a bounded domain in $\mathbb{R}^n$ with smooth boundary
$\partial \Omega$. Consider the differential operator of order $2m$,
\begin{equation} \label{diff_operator}
P(x,D)=\sum_{|\alpha|\leq 2m}h_{\alpha}(x)D^{\alpha}
\end{equation}
where the coefficients $h_{\alpha}(x)$ are sufficiently smooth
complex-valued functions of $x$ in $\overline{\Omega}$, the closure of $\Omega$.
\begin{definition}[{\cite[Definition~7.2.1]{Pazy}}] \rm
The operator $P(x,D)$ is called \emph{strongly elliptic} if there exists
a constant $c>0$ such that
\[
\text{Re}\{(-1)^mP_{2m}(x,\xi)\}\geq c|\xi|^{2m}
\]
for all $x \in \overline{\Omega}$ and $\xi \in \mathbb{R}^n$,
where $P_{2m}(x,\xi)=\sum_{|\alpha|=2m}h_{\alpha}(x)\xi^{\alpha}$.
\end{definition}
\begin{example} \label{example1} \rm
Following \cite[Example~5.2]{HuangZheng2}, consider the nonautonomous problem
\begin{equation} \label{1_elliptic}
\begin{gathered}
\frac{\partial}{\partial t} u(t,x)
= a(t)P(D)u(t,x), \quad (t,x)\in [s,T)\times \mathbb{R}^n \\
u(s,x) = \psi(x), \quad x\in \mathbb{R}^n
\end{gathered}
\end{equation}
where $a \in C([0,T]:\mathbb{R}^+)$ and $P:\mathbb{R}^n \to \mathbb{C}$
is a polynomial of order $2m$ such that $A=P(D)$ is strongly elliptic
with domain $W^{2m,p}(\mathbb{R}^n)$ . Set
\[
\mu_1= \sup_{|\xi|=1}|\text{Re}P_{2m}(\xi)|,\quad
\mu_2= \sup_{|\xi|=1}|\text{Im}P_{2m}(\xi)|.
\]
Then, as seen in \cite{ZhengLi}, $-A=-P(D)$ is the generator of a
holomorphic semigroup of angle $\theta$ on the Banach space
$X=L^p(\mathbb{R}^n)$, $10, \; t\in [s,T]\}$
is a family of regularizing operators for the ill-posed
problem \eqref{1_elliptic} where $\tilde{V}_{\beta}(t,s), 0\leq s\leq t\leq T$
is the evolution system of Corollary~\ref{well-posed_both} corresponding to
the operators
\[
f_{\beta}(t,P(D)-\lambda)=a(t)(P(D)-\lambda)-\beta (P(D)-\lambda)^{\sigma}.
\]
On the other hand, if $\mu_1> \mu_2$ or if $\mu_2=0$ so that
$\theta \in (\pi/4,\pi/2]$, then for some $\lambda \in \mathbb{R}$,
\eqref{2} becomes
\begin{gather*}
(1-\beta \lambda+\beta P(D))\frac{\partial}{\partial t} v(t,x)
= a(t)(P(D)-\lambda) v(t,x) \\
\text{for } (t,x)\in [s,T)\times \mathbb{R}^n, \\
v(s,x) = \psi(x) \quad \text{for } x\in \mathbb{R}^n.
\end{gather*}
Again, by Theorem~\ref{reg_thm},
$\{e^{(\int_s^ta(\tau)d\tau)\lambda}\tilde{V}_{\beta}(t,s) :
\beta>0, \; t\in [s,T]\}$ is a family of regularizing operators for
the ill-posed problem \eqref{1_elliptic} where
$\tilde{V}_{\beta}(t,s), 0\leq s\leq t\leq T$ is the evolution system
of Corollary~\ref{well-posed_both}, in this case corresponding to the
operators $f_{\beta}(t,P(D)-\lambda)=a(t)(P(D)-\lambda)(I+\beta (P(D)
-\lambda))^{-1}$. Note, as mentioned in the introduction, the model
\eqref{2_elliptic} may still be used with $\sigma =2$ if $\theta > \pi/4$.
\end{example}
\begin{example} \label{example2} \rm
Following \cite[Chapter~7.6]{Pazy}, consider the nonautonomous problem
\begin{equation} \label{1_elliptic2}
\begin{gathered}
\frac{\partial}{\partial t} u(t,x)
= a(t)P(x,D)u(t,x) \quad \text{for } (t,x)\in [s,T)\times \Omega \\
D^{\alpha}u(t,x) = 0, \quad |\alpha|0, \; t\in [s,T]\}$ is a family of regularizing operators
for the ill-posed problem \eqref{1_elliptic2}.
\end{example}
\subsection*{Acknowledgments}
The author would like to thank Rhonda J. Hughes for her enthusiasm,
encouragement, and willingness to offer what is always excellent advice.
\begin{thebibliography}{00}
\bibitem{AN} S. Agmon, L. Nirenberg;
\emph{Properties of solutions of ordinary differential equations in Banach space},
Comm. Pure Appl. Math. 16 (1963) 121--151.
\bibitem{AmesandHughes} K. A. Ames, R. J. Hughes;
\emph{Structural stability for ill-posed problems in Banach space},
Semigroup Forum 70 (2005) 127--145.
\bibitem{deL1} R. deLaubenfels;
\emph{Entire solutions of the abstract Cauchy problem},
Semigroup Forum 42 (1991) 83--105.
\bibitem{Fury} M. A. Fury;
\emph{Regularization for ill-posed parabolic evolution problems},
J. Inverse Ill-posed Probl. Volume 20, Issue 5-6, pp. 667-699,
DOI: 10.1515/jip-2012-0018, December 2012.
\bibitem{FuryandHughesSgF} M. A. Fury, R. J. Hughes;
\emph{Regularization for a class of ill-posed evolution problems in Banach space},
Semigroup Forum 85 (2012) 191-212, (DOI) 10.1007/s00233-011-9353-3.
\bibitem{Goldstein} J. A. Goldstein;
\emph{Semigroups of Linear Operators and Applications}, Oxford University Press,
New York, 1985.
\bibitem{HuangZheng2} Y. Huang, Q. Zheng;
\emph{Regularization for Ill-posed Cauchy problems associated with generators
of analytic semigroups}, J. Differential Equations 203 (2004) 38--54.
\bibitem{HuangZheng} Y. Huang, Q. Zheng;
\emph{Regularization for a class of ill-posed Cauchy problems},
Proc. Amer. Math. Soc. 133-10 (2005) 3005--3012.
\bibitem{Kato} T. Kato;
\emph{Perturbation Theory for Linear Operators}, Springer, New York, 1966.
\bibitem{LandL} R. Lattes, J. L. Lions;
\emph{The Method of Quasireversibility, Applications to Partial
Differential Equations}, Elsevier, New York, 1969.
\bibitem{Melnikova} I. V. Mel'nikova;
\emph{General theory of the ill-posed Cauchy problem},
J. Inverse Ill-posed Probl. 3 (1995) 149--171.
\bibitem{MelnikovaandFilinkov} I. V. Mel'nikova, A. I. Filinkov;
\emph{Abstract Cauchy Problems: Three Approaches},
Chapman \& Hall/CRC Monographs and Surveys in Pure and Applied Mathematics 120,
Chapman \& Hall, Boca Raton, 2001.
\bibitem{Miller1} K. Miller;
\emph{Stabilized quasi-reversibility and other nearly-best-possible
methods for non-well-posed problems}, in: Symposium on Non-Well-Posed
Problems and Logarithmic Convexity (Edinburgh, 1972),
Lecture Notes in Mathematics 316, Springer-Verlag, Berlin (1973), 161--176.
\bibitem{Miller2} K. Miller;
\emph{Logarithmic convexity results for holomorphic semigroups},
Pacific J. Math. 58 (1975) 549--551.
\bibitem{Payne1} L. E. Payne;
\emph{Improperly Posed Problems in Partial Differential Equations},
CBMS Regional Conference Series in Applied Mathematics 22,
Society for Industrial and Applied Mathematics, Philadelphia, 1975.
\bibitem{Payne2} L. E. Payne;
\emph{On stabilizing ill-posed problems against errors in geometry and modeling},
in: Inverse and Ill-Posed Problems, Academic Press, San Diego (1987), 399--416.
\bibitem{Pazy} A. Pazy;
\emph{Semigroups of Linear Operators and Applications to Partial Differential
Equations}, Springer-Verlag, New York, 1983.
\bibitem{ReedandSimon} M. Reed, B. Simon;
\emph{Methods of Modern Mathematical Physics, Vol. II: Fourier Analysis,
Self-Adjointness}, Academic Press, New York, 1975.
\bibitem{Rudin} W. Rudin;
\emph{Real and Complex Analysis}, Third edition, McGraw-Hill, New York, 1987.
\bibitem{Showalter} R. E. Showalter;
\emph{The final value problem for evolution equations},
J. Math. Anal. Appl. 47 (1974) 563--572.
\bibitem{Tanaka} N. Tanaka, Linear Evolution Equations In Banach Spaces, Proc. Lond. Math. Soc. 63 (1991) 657--672.
\bibitem{Tanaka2} N. Tanaka;
\emph{Approximation of regularized evolution operators},
Ach. Math. 81 (2003) 38--49.
\bibitem{Trong1} D. D. Trong, N. H. Tuan;
\emph{Regularization and error estimates for nonhomogeneous backward
heat problems}, Electron. J. Differential Equations 2006 (2006) No. 4 1--10.
\bibitem{Trong2} D. D. Trong, N. H. Tuan;
\emph{A nonhomogeneous backward heat problem: Regularization and error estimates},
Electron. J. Differential Equations 2008 (2008) No. 33 1--14.
\bibitem{ZhengLi} Q. Zheng, Y. Li;
\emph{Abstract parabolic systems and regularized semigroups},
Pacific J. Math. 182 (1998) 183--199.
\end{thebibliography}
\end{document}