\documentstyle[twoside]{article} \pagestyle{myheadings} \markboth{\hfil Quasireversibility Methods \hfil EJDE--1994/08}% {EJDE--1994/08\hfil G.W. Clark \& S. F. Oppenheimer\hfil} \newtheorem{theorem}{Theorem} \newtheorem{lemma}{Lemma} \newtheorem{corollary}{Corollary} \begin{document} \ifx\Box\undefined \newcommand{\Box}{\diamondsuit}\fi \title{\vspace{-1in}\parbox{\linewidth}{\footnotesize\noindent {\sc Electronic Journal of Differential Equations}\newline Vol. {\bf 1994}(1994), No. 08, pp. 1-9. Published November 29, 1994.\newline ISSN 1072-6691. URL: http://ejde.math.swt.edu or http://ejde.math.unt.edu \newline ftp (login: ftp) 147.26.103.110 or 129.120.3.113} \vspace{\bigskipamount} \\ Quasireversibility Methods for Non-Well-Posed Problems \thanks{ {\em 1991 Mathematics Subject Classifications:} 35A35, 35R25.\newline\indent {\em Key words and phrases:} Quasireversibility, Final Value Problems, Ill-Posed Problems. \newline\indent \copyright 1994 Southwest Texas State University and University of North Texas.\newline\indent Submitted: November 14, 1994.\newline\indent Partially supported by Army contract DACA 39-94-K-0018 (S. F. O.)} } \date{} \author{Gordon W. Clark \\ and \\ Seth F. Oppenheimer} \maketitle \begin{abstract} The final value problem, $$ \left\{ \begin{array}{ll} u_t+Au=0\,, & 00$, they use the {\it initial value\/} $u_0=v_a(0)$ in $$ \left\{ \begin{array}{cc} u_\alpha ^{\prime }(t)+Au_\alpha (t)=0\,, & 00$, uses the initial value $u_0=v_\alpha (0)$ in% $$ \left\{ \begin{array}{cc} u_\alpha ^{\prime }(t)+Au_\alpha (t)=0\,, & 00,$ and that the approximations $u_\alpha $ are stable. We show that $u_\alpha (T)$ converges to $f$ as $\alpha $ goes to zero and that the values $u_\alpha (t)$ converge on $[0,T]$ if and only if (FVP) has a solution. In the following, assume that $H$ is a separable Hilbert space and $A$ is as above and that $0$ is in the resolvent set of $A$. Let $S(t)$ be the compact contraction semi-group generated by $-A$. Since $A^{-1}$ is compact, there is an orthonormal eigenbasis ${\phi }_n$ for $H$ and eigenvalues $\frac 1{\lambda _n}$ of $A^{-1}$ such that $A^{-1}\phi _n=\frac 1{\lambda _n}\phi _n.$ Then the eigenvalues of $-A$ are $-\lambda _n$ and those for $S(t)$ are ${e}^{-t\lambda _n}${\ } (and possibly zero) [5]. In particular, for each positive $\alpha $, $\alpha I+S(T)$ is invertible. Also, if $% u=\sum_{i=1}^\infty a_i\phi _i,$ then $S(T)u=\sum_{i=1}^\infty {e}% ^{-T\lambda _i}a_i\phi _i$ and $$ \left( S(T)u,u\right) =\sum_{i=1}^\infty {e}^{-T\lambda i}a_i^2\geq 0\,. $$ From this accretive type condition we obtain% $$ \| \left( \alpha I+S(T)\right) ^{-1}\| \leq \frac 1\alpha\,. $$ It is useful to know exactly when (FVP) has a solution. The following lemma answers this question. \begin{lemma} If $f=\sum_{i=1}^\infty b_i\phi _i$, then (FVP) has a solution if and only if \newline $\sum_{i=1}^\infty b_i^2{e}^{2T\lambda _i}$ converges. \end{lemma} \paragraph{Proof.} If $\sum_{i=1}^\infty b_i^2{e}^{2T\lambda _i}$ converges, we merely define $u(t)=\sum_{i=1}^\infty {e}^{\left( T-t\right) \lambda _i}b_i\phi _i$. Let $u$ be a solution to (FVP). Then $u(0)$ has an eigenfunction expansion $u=\sum_{i=1}^\infty a_i\phi _i,$ and $$ S(T)u=\sum_{i=1}^\infty e^{-T\lambda _i}a_i\phi _i=f=\sum_{i=1}^\infty b_i\phi _i\,. $$ This implies that ${e}^{-T\lambda _i}a_i=b_i$ and thus $a_i=b_i{% e}^{T\lambda _i}.$ Since $u(0)$ is in $H$, we have $\left| \left| u\right| \right| ^2=\sum_{i=1}^\infty a_i^2<\infty $ and we are done. $\Box$ We wish to show that our approximate problem is well-posed and the following gives us what we need. \paragraph{Definition.} Define $u_\alpha (t)=S(t)(\alpha I+S(T))^{-1}f$, for $f$ in $H$, $\alpha >0$ and $t$ in $[0,T]$. \begin{theorem} The function $u_\alpha (t)$ is the unique solution of (QBVP) and it depends continuously on f. \end{theorem} \paragraph{Proof.} Since $(\alpha I+S(T))^{-1}f$ is in the domain of $% A$, it is clear that $u_\alpha $ is a classical solution of the differential equation. Furthermore, \begin{eqnarray*} \alpha u_\alpha (0)+u_\alpha (T)&=& \alpha (\alpha I+S(T))^{-1}f+S(T)(\alpha I+S(T))^{-1}f \\ &=&(\alpha I+S(T))(\alpha I+S(T))^{-1}f=f. \end{eqnarray*} To see the continuous dependence of $u_{\alpha}$ on $f$, compute \begin{eqnarray*} \lefteqn{\| S(t)(\alpha I+S(T))^{-1}f_1-S(t)(\alpha I+S(T))^{-1}f_2\|}\\ &=& \| S(t)(\alpha I+S(T))^{-1}(f_1-f_2)\| \\ &\leq&\frac 1\alpha \| f_1-f_2\|\,. \end{eqnarray*} Uniqueness follows from the fact that any solution $v$ must satisfy $% v(0)=(\alpha I+S(T))^{-1}f$ and the uniqueness of solutions to the forward problem. $\Box$ We make two observations at this point which will be useful later. First, from the above it is clear that $\| u_\alpha (t)\| \leq \frac 1\alpha \| f\|$. Secondly, if $u=\sum_{i=1}^\infty a_i\phi _i,$ then $(\alpha I+S(T))u=\sum_{i=1}^\infty (\alpha +{e}^{-T\lambda _i})a_i\phi _i$ and $$ (\alpha I+S(T))^{-1}u=\sum_{i=1}^\infty \frac{a_i}{\alpha +{e}^{-T\lambda _i}% }\phi _i\,. $$ \begin{theorem} For all $f$ in $H,$ $\alpha >0$, and $t$ in $[0,T]$ we have that $$ \| u_\alpha (t)\| \leq \alpha ^{\frac{t-T}T}\| f\|\, . $$ \end{theorem} \paragraph{Proof.} If $f=\sum_{i=1}^\infty b_i\phi _i,$ we have \begin{eqnarray*} \| u_\alpha (t)\|^2&=&\sum_{i=1}^\infty e^{-2t\lambda _i}b_i^2 \left( \alpha +e^{-T\lambda _i}\right)^{-2}\\ &\leq& \sum_{i=1}^\infty e^{-2t\lambda _i}b_i^2 \left[ \left( \alpha +{e}^{-T\lambda _i}\right) ^{\frac tT} \left( \alpha +e^{-T\lambda _i}\right) ^{1-\frac tT}\right] ^{-2} \\ &\leq& \sum_{i=1}^\infty b_i^2\left( \alpha ^{1-\frac tT}\right)^{-2} \\ &=& \left( \alpha ^{\frac{t-T}T}\right)^2\sum_{i=1}^\infty b_i^2 \end{eqnarray*} and we are done. $\Box$ \begin{theorem} For all $f$ in $H$, $\left| \left| u_\alpha \left( T\right) -f\right| \right| $ tends to zero as $\alpha $ tends to zero. That is $u_\alpha \left( T\right) $ converges to $f$ in $H.$ \end{theorem} \paragraph{Proof.} If $f=\sum_{i=1}^\infty b_i\phi _i$, then \begin{eqnarray*} \| u_\alpha (T)-f\| ^2&=&\|S(T)(\alpha I+S(T))^{-1}f-f\| ^2 \\ &=&\alpha ^2\|(\alpha I+S(T))^{-1}f\| ^2 \\ =\sum_{i=1}^\infty \alpha ^2b_i^2\left( \alpha +{e}^{-T\lambda _i}\right) ^{-2}\,. \end{eqnarray*} Fix $\epsilon >0.$ Choose $N$ so that $\sum_{i=N}^\infty b_i^2< \epsilon /2.$ Thus \begin{eqnarray*} \|u_\alpha (T)-f\| ^2&<&\sum_{i=1}^N \alpha ^2b_i^2\left( \alpha +{e}^{-T\lambda _i}\right) ^{-2}+\frac \epsilon 2 \\ &\leq& \alpha ^2\sum_{i=1}^Nb_i^2e^{2\lambda _iT}+\frac \epsilon 2\,. \end{eqnarray*} Now let $\alpha $ be such that $\alpha ^2< \epsilon \left(2\sum_{i=1}^Nb_i^2e^{2\lambda _iT}\right)^{-2}$ and we are done. $\Box$ \begin{theorem} For all $f$ in $H,$ (FVP) has a solution $u$ if and only if the sequence $% u_\alpha (0)$ converges in $H$. Furthermore, we then have that $u_\alpha (t)$ converges to $u(t)$ as $\alpha $ tends to zero uniformly in $t$. \end{theorem} \paragraph{Proof.} Assume that $\lim _{\alpha \downarrow 0}u_\alpha (0)=u_0$ exists. Let $u(t)=S(t)u_0$. Since $\lim _{\alpha \downarrow 0}u_\alpha (T)=f$, \begin{eqnarray*} \lim _{\alpha \downarrow 0}\|u(t)-u_\alpha (t)\| &=&\|S(t)u_0-u_\alpha (t)\| \\ &=&\lim _{\alpha \downarrow 0}\|S(t)\left( u_0-(\alpha I+S(T))^{-1}f\right) \| \\ &\leq& \lim _{\alpha \downarrow 0}\|u_0-(\alpha I+S(T))^{-1}f\| \\ &=&\lim _{\alpha \downarrow 0}\|u_0-u_\alpha (0)\| =0\,. \end{eqnarray*} Thus, $u(T)=f$ and $u(t)=S(t)u_0$ solves (FVP). We also see that $u_\alpha (t)$ converges to $u(t)$ uniformly in $t$. Now let us assume that $u(t)$ is the solution to (FVP). Let $\epsilon >0$ and $f=\sum_{i=1}^\infty b_i\phi _i$. From Lemma~1 we have that $\| u(0)\| ^2=\sum_{i=1}^\infty b_i^2e^{2T\lambda _i}$. Choose $N$ so that $\sum_{i=N}^\infty b_i^2{e}^{2T\lambda _i}<\frac \epsilon 2$. Let $\alpha ,\gamma >0$. Then% \begin{eqnarray*} \|u_\alpha (0)-u_\gamma (0)\| ^2 &=&\|(\alpha I+S(T))^{-1}f-(\gamma I+S(T))^{-1}f\| \\ &=&\|\sum_{i=1}^\infty \left( \frac 1{\alpha + e^{-T\lambda _i}} -\frac 1{\gamma +e^{-T\lambda _i}}\right) b_i\phi _i\| \\ &=&\sum_{i=1}^\infty (\gamma -\alpha )^2\left( \alpha \gamma +(\alpha +\gamma ) e^{-T\lambda _i}+{e}^{-2T\lambda _i}\right)^{-2} b_i^2 \\ &=&\sum_{i=1}^N (\gamma -\alpha )^2\left( \alpha \gamma +(\alpha +\gamma ) e^{-T\lambda _i}+e^{-2T\lambda _i}\right) ^{-2} b_i^2\\ &&+\sum_{i=N+1}^\infty (\gamma -\alpha )^2\left( \alpha \gamma +(\alpha +\gamma )e^{-T\lambda _i}+e^{-2T\lambda _i}\right)^{-2} b_i^2 \\ &\leq& \sum_{i=1}^N(\gamma -\alpha )^2e^{4T\lambda _i}b_i^2+\sum_{i=N+1}^\infty \left( \frac{\gamma -\alpha }{\alpha +\gamma }\right) ^2b_i^2e^{2T\lambda _i} \\ &\leq& \sum_{i=1}^N(\gamma -\alpha )^2e^{4T\lambda _i}b_i^2+ \frac \epsilon 2\,. \end{eqnarray*} Now if we choose $\delta >0$ so that $\delta ^2<\epsilon\left( \sum_{i=1}^Ne^{4T\lambda _i}b_i^2\right)^{-1}$ and require that $\alpha $ and $\gamma $ be less than $\delta $, we have that $$ \|u_\alpha (0)-u_\gamma (0)\| ^2<\epsilon\, . $$ We therefore have that $\{u_\alpha (0)\}$ is Cauchy and thus converges. From the first part of the theorem, we have that $u_\alpha (t)$ converges to $u(t) $ uniformly in $t$. $\Box$ We end this paper with a result that gives explicit convergence rates in the case that (FVP) is soluble for some positive final time. \begin{theorem} If $f=\sum_{i=1}^\infty b_i\phi _i$ is in $H$ and there exists an $\epsilon >0$ so that $\sum_{i=1}^\infty b_i^2e^{\epsilon \lambda _iT}$ converges, then $\|u_\alpha (T)-f\| $ converges to zero with order $\alpha ^\epsilon \epsilon ^{-2}$. \end{theorem} \paragraph{Proof.} Let $\epsilon $ be in $(0,2)$ such that $% \sum_{i=1}^\infty b_i^2e^{\epsilon \lambda _iT}$ is finite and let $k$ be in $(0,2)$. Fix a natural number $n$. Define $$ g_n(\alpha )=\frac{\alpha ^k}{(\alpha +e^{-\lambda _nT})^2}. $$ Differentiating with respect to $\alpha $ yields% $$ g_n'(\alpha )=\alpha ^{k-1}\frac{(k-2)\alpha +ke^{-T\lambda _n}}{% (\alpha +e^{-\lambda _nT})^3}. $$ Thus $g_n'(\alpha )=0$ when either $\alpha =0$ or $$ \alpha =\frac k{2-k}e^{-T\lambda _n}. $$ Since $g_n(\alpha )>0$, $g_n(0 )=0$, and $% \lim _{\alpha \rightarrow \infty }g_n(\alpha )=0$ we have that $\alpha _0=\frac k{2-k}e^{-T\lambda _n}$ is the critical value at which $g_n$ achieves its maximum. Thus we have the inequality% $$ g_n(\alpha )\leq \frac{\left( \frac k{2-k}\right) ^ke^{-kT\lambda _n}}{% (\alpha _0+e^{-\lambda _nT})^2}. $$ We now calculate \begin{eqnarray*} \|u_\alpha (T)-f\| ^2&=&\sum_{n=1}^\infty b_n^2 \alpha ^2(\alpha +e^{-\lambda _nT})^{-2} =\alpha^{2-k}\sum_{n=1}^\infty b_n^2g_n(\alpha ) \\ &\leq& \alpha ^{2-k}\sum_{n=1}^\infty b_n^2\left( \frac k{2-k}\right) ^ke^{-kT\lambda _n} (\alpha_0+e^{-\lambda _nT})^{-2} \\ &\leq& \alpha ^{2-k}\sum_{n=1}^\infty b_n^2\left( \frac k{2-k}\right) ^ke^{(2-k)T\lambda _n} (\alpha_0^2+2\alpha _0e^{\lambda _nT}+1)^{-1} \\ &\leq& \alpha ^{2-k}\sum_{n=1}^\infty b_n^2\left( \frac k{2-k}\right) ^ke^{(2-k)T\lambda _n} \\ &=&\alpha ^{2-k}\left( \frac k{2-k}\right) ^k\sum_{n=1}^\infty b_n^2e^{(2-k)T\lambda _n}. \end{eqnarray*} If we choose $k=2-\epsilon $ we arrive at $$ \|u_\alpha (T)-f\| ^2\leq \left( \frac 2\epsilon \right) ^2\alpha ^\epsilon \sum_{n=1}^\infty b_n^2e^{\epsilon T\lambda _n} =c\alpha ^\epsilon \epsilon ^{-2}\,. $$ \hfill $\Box$\newline If we assume that $\sum_{i=1}^\infty b_i^2e^{(2+\epsilon )\lambda _iT}$ converges, working as above, we have that \begin{eqnarray*} \|u_\alpha (0)-u(0)\| ^2&=&\alpha ^{2-k}\sum_{n=1}^\infty b_n^2g_n(\alpha )e^{2T\lambda _n} \\ &\leq& \alpha ^{2-k}\sum_{n=1}^\infty b_n^2\left( \frac k{2-k}\right) ^ke^{(4-k)T\lambda _n}\,. \end{eqnarray*} As above, letting $k=2-\epsilon $, we arrive at the following. \begin{corollary} If $f=\sum_{i=1}^\infty b_i\phi _i$ is in $H$ and there exists an $\epsilon >0$ so that $\sum_{i=1}^\infty b_i^2e^{(2+\epsilon )\lambda _iT}$ converges, then $\|u_\alpha (t)-u(t)\| $ converges to zero with order $\alpha ^\epsilon\epsilon ^{-2}$ uniformly in $t$. \end{corollary} \begin{thebibliography}{9} \bibitem{1} Conway, J.B., ``A Course in Functional Analysis, Springer-Verlag, New York, 1990 \bibitem{2} Lattes, R. and Lions, J.L., ``Methode de Quasi-Reversibility et Applications'', Dunod, Paris, 1967 (English translation R. Bellman, Elsevier, New York, 1969) \bibitem{3} Miller, K., {\it Stabilized quasireversibility and other nearly best possible methods for non-well-posed problems}, ``Symposium on Non-Well-Posed Problems and Logarithmic Convexity'', Lecture Notes in Mathematics, Vol. 316, Springer-Verlag, Berlin, 1973, pp 161-176 \bibitem{4} Payne, L.E., {\it Some general remarks on improperly posed problems for partial differential equations}, ``Symposium on Non-Well-Posed Problems and Logarithmic Convexity'', Lecture Notes in Mathematics, Vol. 316, Springer-Verlag, Berlin, 1973, pp 1-30 \bibitem{5} Pazy, A., ``Semigroups of Linear Operators and Applications to Partial Differential Equations'', Springer-Verlag, New York, 1983 \bibitem{6} Showalter, R.E., {\it The Final Value Problem for Evolution Equations}, J. Math. Anal. Appl. 47, 1974, pp 563-572 \bibitem{7} Showalter, R.E., {\it Cauchy Problem for Hyper-Parabolic Partial Differential Equations}, ``Trends in the Theory and Practice of Non-Linear Analysis'', Elsevier, 1983 \bibitem{8} Yosida, K.,``Functional Analysis'', Springer-Verlag, Berlin, 1980 \end{thebibliography} {\sc Gordon W. Clark \\ Department of Mathematics\\ Kennesaw State College\\ P O Box 444\\ Marietta, GA 30061}\\ E-mail address: clark@math.msstate.edu \smallskip {\sc Seth F. Oppenheimer\newline Department of Mathematics and Statistics\\ Mississippi State University\\ Drawer MA MSU, MS 39762}\\ E-mail address: seth@math.msstate.edu \end{document}