\begin{filecontents}{CNZ1.eps} %!PS-Adobe-2.0 EPSF-2.0 %%Title: CNZ1.eps %%Creator: fig2dev Version 3.2 Patchlevel 1 %%CreationDate: Mon Mar 6 17:15:27 2000 %%For: cachia@cptpc26 (Vincent Cachia thesard Zagrebnov) %%Orientation: Portrait %%BoundingBox: 0 0 362 362 %%Pages: 0 %%BeginSetup %%EndSetup %%Magnification: 1.0000 %%EndComments /$F2psDict 200 dict def $F2psDict begin $F2psDict /mtrx matrix put /col-1 {0 setgray} bind def /col0 {0.000 0.000 0.000 srgb} bind def /col1 {0.000 0.000 1.000 srgb} bind def /col2 {0.000 1.000 0.000 srgb} bind def /col3 {0.000 1.000 1.000 srgb} bind def /col4 {1.000 0.000 0.000 srgb} bind def /col5 {1.000 0.000 1.000 srgb} bind def /col6 {1.000 1.000 0.000 srgb} bind def /col7 {1.000 1.000 1.000 srgb} bind def /col8 {0.000 0.000 0.560 srgb} bind def /col9 {0.000 0.000 0.690 srgb} bind def /col10 {0.000 0.000 0.820 srgb} bind def /col11 {0.530 0.810 1.000 srgb} bind def /col12 {0.000 0.560 0.000 srgb} bind def /col13 {0.000 0.690 0.000 srgb} bind def /col14 {0.000 0.820 0.000 srgb} bind def /col15 {0.000 0.560 0.560 srgb} bind def /col16 {0.000 0.690 0.690 srgb} bind def /col17 {0.000 0.820 0.820 srgb} bind def /col18 {0.560 0.000 0.000 srgb} bind def /col19 {0.690 0.000 0.000 srgb} bind def /col20 {0.820 0.000 0.000 srgb} bind def /col21 {0.560 0.000 0.560 srgb} bind def /col22 {0.690 0.000 0.690 srgb} bind def /col23 {0.820 0.000 0.820 srgb} bind def /col24 {0.500 0.190 0.000 srgb} bind def /col25 {0.630 0.250 0.000 srgb} bind def /col26 {0.750 0.380 0.000 srgb} bind def /col27 {1.000 0.500 0.500 srgb} bind def /col28 {1.000 0.630 0.630 srgb} bind def /col29 {1.000 0.750 0.750 srgb} bind def /col30 {1.000 0.880 0.880 srgb} bind def /col31 {1.000 0.840 0.000 srgb} bind def /col32 {0.733 0.741 0.800 srgb} bind def end save -179.0 469.0 translate 1 -1 scale /cp {closepath} bind def /ef {eofill} bind def /gr {grestore} bind def /gs {gsave} bind def /sa {save} bind def /rs {restore} bind def /l {lineto} bind def /m {moveto} bind def /rm {rmoveto} bind def /n {newpath} bind def /s {stroke} bind def /sh {show} bind def /slc {setlinecap} bind def /slj {setlinejoin} bind def /slw {setlinewidth} bind def /srgb {setrgbcolor} bind def /rot {rotate} bind def /sc {scale} bind def /sd {setdash} bind def /ff {findfont} bind def /sf {setfont} bind def /scf {scalefont} bind def /sw {stringwidth} bind def /tr {translate} bind def /tnt {dup dup currentrgbcolor 4 -2 roll dup 1 exch sub 3 -1 roll mul add 4 -2 roll dup 1 exch sub 3 -1 roll mul add 4 -2 roll dup 1 exch sub 3 -1 roll mul add srgb} bind def /shd {dup dup currentrgbcolor 4 -2 roll mul 4 -2 roll mul 4 -2 roll mul srgb} bind def /reencdict 12 dict def /ReEncode { reencdict begin /newcodesandnames exch def /newfontname exch def /basefontname exch def /basefontdict basefontname findfont def /newfont basefontdict maxlength dict def basefontdict { exch dup /FID ne { dup /Encoding eq { exch dup length array copy newfont 3 1 roll put } { exch newfont 3 1 roll put } ifelse } { pop pop } ifelse } forall newfont /FontName newfontname put newcodesandnames aload pop 128 1 255 { newfont /Encoding get exch /.notdef put } for newcodesandnames length 2 idiv { newfont /Encoding get 3 1 roll put } repeat newfontname newfont definefont pop end } def /isovec [ 8#200 /grave 8#201 /acute 8#202 /circumflex 8#203 /tilde 8#204 /macron 8#205 /breve 8#206 /dotaccent 8#207 /dieresis 8#210 /ring 8#211 /cedilla 8#212 /hungarumlaut 8#213 /ogonek 8#214 /caron 8#220 /dotlessi 8#230 /oe 8#231 /OE 8#240 /space 8#241 /exclamdown 8#242 /cent 8#243 /sterling 8#244 /currency 8#245 /yen 8#246 /brokenbar 8#247 /section 8#250 /dieresis 8#251 /copyright 8#252 /ordfeminine 8#253 /guillemotleft 8#254 /logicalnot 8#255 /endash 8#256 /registered 8#257 /macron 8#260 /degree 8#261 /plusminus 8#262 /twosuperior 8#263 /threesuperior 8#264 /acute 8#265 /mu 8#266 /paragraph 8#267 /periodcentered 8#270 /cedilla 8#271 /onesuperior 8#272 /ordmasculine 8#273 /guillemotright 8#274 /onequarter 8#275 /onehalf 8#276 /threequarters 8#277 /questiondown 8#300 /Agrave 8#301 /Aacute 8#302 /Acircumflex 8#303 /Atilde 8#304 /Adieresis 8#305 /Aring 8#306 /AE 8#307 /Ccedilla 8#310 /Egrave 8#311 /Eacute 8#312 /Ecircumflex 8#313 /Edieresis 8#314 /Igrave 8#315 /Iacute 8#316 /Icircumflex 8#317 /Idieresis 8#320 /Eth 8#321 /Ntilde 8#322 /Ograve 8#323 /Oacute 8#324 /Ocircumflex 8#325 /Otilde 8#326 /Odieresis 8#327 /multiply 8#330 /Oslash 8#331 /Ugrave 8#332 /Uacute 8#333 /Ucircumflex 8#334 /Udieresis 8#335 /Yacute 8#336 /Thorn 8#337 /germandbls 8#340 /agrave 8#341 /aacute 8#342 /acircumflex 8#343 /atilde 8#344 /adieresis 8#345 /aring 8#346 /ae 8#347 /ccedilla 8#350 /egrave 8#351 /eacute 8#352 /ecircumflex 8#353 /edieresis 8#354 /igrave 8#355 /iacute 8#356 /icircumflex 8#357 /idieresis 8#360 /eth 8#361 /ntilde 8#362 /ograve 8#363 /oacute 8#364 /ocircumflex 8#365 /otilde 8#366 /odieresis 8#367 /divide 8#370 /oslash 8#371 /ugrave 8#372 /uacute 8#373 /ucircumflex 8#374 /udieresis 8#375 /yacute 8#376 /thorn 8#377 /ydieresis] def /Times-Roman /Times-Roman-iso isovec ReEncode /Times-Italic /Times-Italic-iso isovec ReEncode /DrawEllipse { /endangle exch def /startangle exch def /yrad exch def /xrad exch def /y exch def /x exch def /savematrix mtrx currentmatrix def x y tr xrad yrad sc 0 0 1 startangle endangle arc closepath savematrix setmatrix } def /$F2psBegin {$F2psDict begin /$F2psEnteredState save def} def /$F2psEnd {$F2psEnteredState restore end} def %%EndProlog $F2psBegin 10 setmiterlimit n -1000 8812 m -1000 -1000 l 10012 -1000 l 10012 8812 l cp clip 0.06000 0.06000 sc % Arc 15.000 slw gs n 6001.3 4800.0 1351.3 -56.4 56.4 arcn gs col7 0.90 shd ef gr gs col-1 s gr gr % Polyline n 6750 3675 m 8400 4800 l 6750 5925 l gs col7 0.90 shd ef gr gs col-1 s gr % Arc 7.500 slw gs n 8107.5 4657.5 551.2 -123.9 165.0 arcn gs col-1 s gr gr % Arc gs n 5983.7 4800.0 1708.7 -44.6 44.6 arcn gs col-1 s gr gr % Arc 15.000 slw gs n 8175.0 4725.0 167.7 -116.6 153.4 arcn gs col-1 s gr gr 7.500 slw % Ellipse n 6000 4800 2400 2400 0 360 DrawEllipse gs col-1 s gr [60] 0 sd % Ellipse n 7200 4800 1200 1200 0 360 DrawEllipse gs col-1 s gr [] 0 sd % Polyline gs clippath 5970 1920 m 6000 1800 l 6030 1920 l 6030 1785 l 5970 1785 l cp clip n 6000 7800 m 6000 1800 l gs col-1 s gr gr % arrowhead n 5970 1920 m 6000 1800 l 6030 1920 l col-1 s % Polyline n 7200 3600 m 8400 4800 l 7200 6000 l gs col-1 s gr % Polyline [30] 0 sd n 6000 4800 m 6750 3675 l gs col-1 s gr [] 0 sd % Polyline [30] 0 sd n 6600 3900 m 6825 4050 l 6975 3825 l gs col-1 s gr [] 0 sd % Polyline [30] 0 sd n 6000 4800 m 7200 6000 l gs col-1 s gr [] 0 sd % Polyline [30] 0 sd n 7050 5850 m 7200 5700 l 7350 5850 l gs col-1 s gr [] 0 sd % Polyline gs clippath 5448 3216 m 5325 3225 l 5429 3159 l 5301 3201 l 5320 3258 l cp clip n 5550 3150 m 5325 3225 l gs col-1 s gr gr % arrowhead n 5448 3216 m 5325 3225 l 5429 3159 l col-1 s % Polyline [30] 0 sd gs clippath 7489 3931 m 7425 3825 l 7531 3889 l 7436 3793 l 7393 3836 l cp clip n 8400 4800 m 7425 3825 l gs col-1 s gr gr [] 0 sd % arrowhead n 7489 3931 m 7425 3825 l 7531 3889 l col-1 s % Polyline gs clippath 6704 6327 m 6825 6300 l 6731 6380 l 6852 6320 l 6825 6266 l cp clip n 6675 6375 m 6825 6300 l gs col-1 s gr gr % arrowhead n 6704 6327 m 6825 6300 l 6731 6380 l col-1 s % Polyline gs clippath 8880 4770 m 9000 4800 l 8880 4830 l 9015 4830 l 9015 4770 l cp clip n 3000 4800 m 9000 4800 l gs col-1 s gr gr % arrowhead n 8880 4770 m 9000 4800 l 8880 4830 l col-1 s /Times-Roman-iso ff 180.00 scf sf 5850 5025 m gs 1 -1 sc (0) col-1 sh gr /Times-Roman-iso ff 210.00 scf sf 8475 5025 m gs 1 -1 sc (1) col-1 sh gr /Times-Roman-iso ff 180.00 scf sf 4725 5025 m gs 1 -1 sc (-sin) col-1 sh gr /Times-Roman-iso ff 210.00 scf sf 6150 2250 m gs 1 -1 sc (i) col-1 sh gr /Symbol ff 210.00 scf sf 3300 5025 m gs 1 -1 sc (-1) col-1 sh gr /Symbol ff 210.00 scf sf 5325 3000 m gs 1 -1 sc (G) col-1 sh gr /Symbol ff 210.00 scf sf 5250 3450 m gs 1 -1 sc (G) col-1 sh gr /Times-Italic-iso ff 150.00 scf sf 5325 3525 m gs 1 -1 sc (a) col-1 sh gr /Symbol ff 210.00 scf sf 7425 4500 m gs 1 -1 sc (b) col-1 sh gr /Times-Italic-iso ff 210.00 scf sf 7200 3525 m gs 1 -1 sc (A) col-1 sh gr /Symbol ff 180.00 scf sf 4125 5025 m gs 1 -1 sc (b) col-1 sh gr /Times-Italic-iso ff 210.00 scf sf 6600 5325 m gs 1 -1 sc (r) col-1 sh gr /Times-Roman-iso ff 180.00 scf sf 3825 5025 m gs 1 -1 sc (-sin) col-1 sh gr /Symbol ff 180.00 scf sf 5025 5025 m gs 1 -1 sc (a) col-1 sh gr /Symbol ff 210.00 scf sf 7800 4650 m gs 1 -1 sc (a) col-1 sh gr /Times-Roman-iso ff 150.00 scf sf 7275 5400 m gs 1 -1 sc (1/2) col-1 sh gr /Times-Roman-iso ff 210.00 scf sf 7125 5325 m gs 1 -1 sc (D) col-1 sh gr /Times-Italic-iso ff 210.00 scf sf 7200 6225 m gs 1 -1 sc (B) col-1 sh gr /Symbol ff 210.00 scf sf 5175 4275 m gs 1 -1 sc (S) col-1 sh gr /Times-Italic-iso ff 150.00 scf sf 5400 3075 m gs 1 -1 sc (r) col-1 sh gr /Times-Italic-iso ff 150.00 scf sf 5325 4350 m gs 1 -1 sc (a) col-1 sh gr /Times-Roman-iso ff 150.00 scf sf 5400 4425 m gs 1 -1 sc (*) col-1 sh gr /Times-Roman-iso ff 150.00 scf sf 5400 3600 m gs 1 -1 sc (*) col-1 sh gr /Times-Italic-iso ff 210.00 scf sf 6375 4500 m gs 1 -1 sc (a) col-1 sh gr /Times-Roman-iso ff 180.00 scf sf 6525 4575 m gs 1 -1 sc (*) col-1 sh gr $F2psEnd rs \end{filecontents}{CNZ1.eps} \documentstyle[12pt,amsfonts,epsf,float]{article} \oddsidemargin 0.5cm \topmargin -2.5cm \textheight 24.0cm \textwidth 16.0cm \headheight 0cm \headsep 0cm \footskip 1cm \topmargin 0cm \linespread{1} \newcommand{\bR}{{\mathbb R}^1} \newcommand{\bT}{{\bf T}^1} \newcommand{\bD}{{\mathbb D}} \newcommand{\bC}{{\mathbb C}} \newcommand{\bZ}{{\mathbb Z}} \newcommand{\bQ}{{\mathbb Q}} \newcommand{\bN}{{\mathbb N}} \newcommand{\kH}{{\frak H}} \newcommand{\real}{{\Re{\mathrm e\,}}} \newcommand{\imag}{{\Im{\mathrm m\,}}} \newcommand{\dom}{{\mathrm{dom}}} \newcommand{\ran}{{\mathrm{ran}}} \newcommand{\sign}{{\mathrm{sign}}} \newcommand{\supp}{{\mathrm{supp}}} \newtheorem{theo}{Theorem} \newtheorem{prop}[theo]{Proposition} \newtheorem{lem}[theo]{Lemma} \newtheorem{cor}[theo]{Corollary} \newtheorem{exam}[theo]{Example} \newtheorem{defi}[theo]{Definition} \def\pitchfork{\hbox{$\cap \hskip -0.18cm \vert$}} \makeatletter \def\@ypreuve[#1]{\@preuve{ #1}} \def\@preuve#1{\begin{trivlist}\item[]{\em Proof#1. }} \newenvironment{preuve}{\@ifnextchar[{\@ypreuve}{\@preuve{}}}{\hfill$\Box$\end{t rivlist}} \makeatother \begin{document} \def\Label{\label} \def\Ref{\ref} \font\fifteen=cmbx10 at 15pt \font\twelve=cmbx10 at 12pt \begin{titlepage} \begin{center} \renewcommand{\thefootnote}{\fnsymbol{footnote}} {\twelve Centre de Physique Th\'eorique\footnote{Unit\'e Propre de Recherche 7061 }, CNRS Luminy, Case 907} {\twelve F-13288 Marseille -- Cedex 9} \vspace{2 cm} {\fifteen ACCRETIVE PERTURBATIONS AND ERROR ESTIMATES FOR THE TROTTER PRODUCT FORMULA\footnote{\bf To the memory of Tosio Kato}} \vspace{1.5 cm} \setcounter{footnote}{0} \renewcommand{\thefootnote}{\arabic{footnote}} {\bf V. CACHIA\footnote{Allocataire-moniteur at Universit\'e de la M\'editerran\'ee (Aix-Marseille II)\\ Email : cachia@cpt.univ-mrs.fr}, H. NEIDHARDT\footnote{Weierstra\ss\ Institut f\"ur angewandte Analysis und Stochastik, Mohrenstr. 39, D-10117 Berlin, Germany - Email : neidhard@wias-berlin.de} and V.A. ZAGREBNOV\footnote{Universit\'e de la M\'editerran\'ee (Aix-Marseille II) - Email : zagrebnov@cpt.univ-mrs.fr} } \vspace{1.5 cm} {\bf Abstract} \end{center} We study the error bound estimate in the operator-norm topology for the exponential Trotter product formula in the case of accretive perturbations. Let $A$ be a semibounded from below self-adjoint operator on a separable Hilbert space. Let $B$ be a closed maximal accretive operator which is, together with $B^*$, Kato-small with respect to $A$ with relative bounds less than one. We show that in this case the operator-norm error bound estimate for the exponential Trotter product formula is the same as for the self-adjoint $B$ \cite{NZ1}: $$\left\|\left(e^{-tA/n}e^{-tB/n}\right)^n - e^{-t(A+B)}\right\| \leq L {\ln n\over n},\ n = 2,3,\ldots\,.$$ We verify that the operator $-(A+B)$ generates a holomorphic contraction semigroup. One gets a similar result when $B$ is substituted by $B^*$. \vspace{1 cm} \bigskip \noindent Keywords : Semigroups, Trotter product formula, accretive operators, operator-norm estimates. \noindent Mathematics Subject Classification : 47D03, 47B25, 35K22, 41A80 \bigskip \bigskip \noindent anonymous ftp : ftp.cpt.univ-mrs.fr\\ web : www.cpt.univ-mrs.fr\\ February 2000\\ CPT-2000/P.3961 \renewcommand{\thefootnote}{\fnsymbol{footnote}} \end{titlepage} \setcounter{footnote}{0} \renewcommand{\thefootnote}{\arabic{footnote}} \newpage \section{Introduction and Setup of the Problem} \setcounter{equation}{0} \renewcommand{\theequation}{\arabic{section}.\arabic{equation}} \setcounter{theo}{0} \renewcommand{\thetheo}{\arabic{section}.\arabic{theo}} Let $-A$ and $-B$ be generators of strong continuous semigroups on some separable Hilbert space ${\frak H}$. For those operators the convergence of the Trotter product formula % \begin{equation}\Label{1} \tau -\lim_{n\rightarrow +\infty}(e^{-tA/n}e^{-tB/n})^{n} = e^{-tH}, \end{equation} % where $H$ is the sum of the operators $A$ and $B$ defined in a suitable sense, was of interest for many years only with respect to the strong topology $\tau =s$, see \cite{Ch1}, \cite{Ch2}, \cite{Tr1}. A final result in this direction was obtained by Kato \cite{Ka2}, \cite{Ka1} who has shown that the convergence in the strong topology always takes place if the operators $A$ and $B$ are self-adjoint and non-negative, or even more generally they may be m-sectorial. Here and below we follow the terminology of the Kato's book \cite{Kato}. Moreover, he identified $H$ as the form-sum of $A$ and $B$ and generalized the product formula (\ref{1}) to the form in which the exponential functions are replaced by functions of a certain class which is often called the Kato-class. However, certain applications of the Trotter or the more general Trotter-Kato product formul\ae\ require that % % \begin{enumerate} % % \item[(a)] the convergence in (\Ref{1}) is in the operator-norm topology or in stronger norms, for instance, in the trace- or the Hilbert-Schmidt-norm, % % \item[(b)] error estimates in these norms, in particular, in operator-norm, are available, % % \item[(c)] the convergence and the error estimates are valid not only for self-adjoint operators $A$ and $B$. % % \end{enumerate} Until now the problems (a) and (b) were solved in a satisfactory manner only for non-negative self-adjoint operators $A$ and $B$. In papers \cite{NZ0}-\cite{NZ4} the conditions (partially necessary and sufficient) are found for convergence of the Trotter-Kato product formula in the operator-norm or in norms of symmetrically-normed ideals of compact operators, see also \cite{Hi1}. Error bound estimates in the operator-norm for non-negative self-adjoint operators were given in \cite{IT1}, \cite{NZ1}, \cite{NZ2} and \cite{Ro1} and for the trace-norm in \cite{IT2} and \cite{NZ4}. On the other hand there are only few results concerning the problem (c). The first result in this direction was that of Lie who has shown the convergence of the exponential product formula (\ref{1}) for arbitrary matrices $A$ and $B$. In fact, his proof provides the operator-norm error bound estimate for a couple of bounded operators $A$ and $B$ in a complete normed space: % % \begin{equation}\Label{1.4-1} \left\|(e^{-tA/n}e^{-tB/n})^{n} - e^{-tH}\right\| \le \frac{C}{n}, \quad n = 1,2,\ldots, \end{equation} % % where $H = A + B$, see \cite{RS1}. So the Lie's result solves problems (a) - (c) for bounded operators in a Banach space. In \cite{CZ2} the problem (a) in a Hilbert space is solved in the operator-norm topology $\tau = \|\cdot\|$ for a class of unbounded m-sectorial operators $A$ and $B$ with semi-angles $\theta_A, \theta_B\in (0,\pi/2)$. A solution of the problem (b) in a Banach space was presented in \cite{CZ1} for particular domain conditions. Namely, let $-A$ be generator of a holomorphic contraction semigroup and let $-B$ be generator of a contraction semigroup satisfying the conditions: $\dom(B)\supseteq \dom(A^\alpha)$ for some $\alpha\in [0,1)$ and $\dom(B^*)\supseteq \dom(A^*)$. then there is a constant $C_\alpha$ such that % % \begin{equation}\Label{1.5} \left\|\left(e^{-tB/n}e^{-tA/n}\right)^{n}-e^{-tH}\right\| \leq C_\alpha\frac{(\ln(n))^{s(\alpha)}}{n^{1-\alpha}}, \quad n = 2,3,\ldots, \end{equation} % % uniformly in $t \in [0,T]$, $0 < T < +\infty$, where $H = A +B$ on $\dom(H) = \dom(A)$, and $s(\alpha=0)=2$, $s(\alpha\neq 0)=1$. The condition $\dom(B)\supseteq \dom(A^\alpha)$ implies that $B$ is relatively bounded with respect to $A$ with the relative bound $a=0$. Thus, comparing with the corresponding results in the self-adjoint case on a Hilbert space, see \cite{NZ1} and \cite{NZ2}, the result (\ref{1.5}) is not optimal. For instance in \cite{NZ1} it is proven that if $A\geq I$, $B\geq 0$ and $B$ is relatively bounded with respect to $A$ with $a<1$, then there is a constant $C$ such that \begin{equation}\Label{1.5-2} \left\|\left(e^{-tA/n}e^{-tB/n}\right)^{n} - e^{-tH}\right\| \leq C\frac{\ln(n)}{n}, \quad n = 2,3,\ldots, \end{equation} % % uniformly in $t \ge 0$, where $H=A+B$ on $\dom(H) = \dom(A)$. In what follows we suppose that $A=A^{\ast }$ be a semibounded from below densely defined self-adjoint operator on the Hilbert space ${\kH}$ which generates a semigroup $\{U_{A}(t)= e^{-tA}\}_{t\geq 0}$. Without loss of generality one can suppose that $A\geq I$. Then it is known (see, e.g. the Kato's book \cite{Kato}) that $U_{A}(t)$ is a contraction holomorphic semigroup of angle $\omega =\pi /2$, i.e. $\Vert U_{A}(t\geq 0)\Vert \leq 1$ and $U_{A}(t)$ has a holomorphic extension from $t\geq 0$ into the open sector of the complex plane: $S_{\omega =\pi /2}=\{z\in {\bC}:z\neq 0$ and $|{\arg }(z)|<\pi /2\}.$ About the generator $B$ we suppose that : \begin{enumerate} \item[$(i)$] domain $\dom(B)\supseteq \dom(A)$ and there exists $a>0$ such that \begin{equation} \|Bu\| \leq a\|Au\|,\ u\in\dom(A);\label{1.1} \end{equation} \item[$(ii)$] domain $\dom(B^*)\supseteq \dom(A)$ and there exists $a_*> 0$ such that \begin{equation} \|B^*u\| \leq a_*\|Au\|,\ u\in\dom(A);\label{1.2} \end{equation} \item[$(iii)$] $B$ is a m-accretive (i.e. closed and maximal accretive) operator on ${\kH}$, i.e. for $\lambda>0$ one has \begin{equation} \Vert (B+\lambda)^{-1}\Vert \leq \lambda^{-1}. \label{1.3} \end{equation} \end{enumerate} The latter implies that $\{U_{B}(t)= e^{-tB}\}_{t\geq 0}$ is a (non self-adjoint) contraction semigroup, but $B$ is not necessarily sectorial: $\theta_B$ may be $\pi/2$. Notice that under conditions $(i)$-$(iii)$ the convergence of the Trotter product formula (\ref{1}) is known, due to \cite{Ch1}, \cite{Ch2}, \cite{Tr1}, only in the strong topology $\tau =s$. In our main theorem (see Section 3) we prove that under conditions $(i)$-$(iii)$ and $a,a_*<1$ there is a constant $L>0$ such that uniformly in $t\geq 0$ one gets (cf (\ref{1.5-2}) \begin{equation} \left\| \left(e^{-tA/n}e^{-tB/n}\right)^{n}-e^{-tH}\right\| \leq L{\ln (n)\over n}\ ,\ n=1,2,3,\dots\quad . \label{1.6} \end{equation} Here $H=A+B$ is defined on $\dom(H)=\dom(A)$. Similar result holds for $B^{\ast }$ and $H^{\ast }$. The paper is organized as follows. In Section 2 we prove some technical results, which have an interest in itself, and which are crucial for the proof of the main theorem, which is presented in Section 3. The proof of this main theorem (while inspired by) needs a considerable deviation from the line of reasoning known for the self-adjoint case. Finally, in Section 4 we make some remarks and comments. \section{Preliminaries} \setcounter{equation}{0} \renewcommand{\theequation}{\arabic{section}.\arabic{equation}} \setcounter{theo}{0} \renewcommand{\thetheo}{\arabic{section}.\arabic{theo}} Let $-H$ be the generator of a semigroup $\{U_H(t)=e^{-tH}\}_{t \ge 0}$. The semigroup $U_H(t)$ is called a bounded holomorphic semigroup with angle $\omega \in (0,\pi/2]$ if % % \begin{enumerate} % \item[($\alpha$)] $\{U(t)\}_{t \ge 0}$ is the restriction to the positive real axis of an analytic family of bounded operators $\{U(z)\}_{z \in S_\omega}$, $S_\omega = \{z \in {\mathbb C}: z \not= 0, \; |\arg z| < \omega\}$, obeying $U(z + z') = U(z)U(z')$, $z,z' \in S_\omega$, and \item[($\beta$)] for every $\theta \in (0,\omega)$ the family $\{U(z)\}_{z \in S_\omega}$ is uniformly bounded in the sector $S_\theta$ and $U(z)f \longrightarrow f$ as $z \longrightarrow 0$ in $z \in S_\theta$ for each $f \in {\frak H}$. \end{enumerate} It turns out that this is the case if and only if the spectrum $\sigma(H)$ of $H$ is contained in the closed sector $\overline{S_{\pi/2-\omega}}$, i.e. $\sigma(H) \subseteq \overline{S_{\pi/2-\omega}}$, and for each $\theta\in [0,\omega)$ there is a constant $M_\theta > 0$ such that % % \begin{equation}\Label{2.1} \|(H + z)^{-1}\| \le \frac{M_\theta}{|z|} \end{equation} % % for $z \in S_{\pi/2+\theta}$, see e.g. \cite[Ch.IX]{Kato}. % % \begin{theo}\Label{II.1} % % Let $A \ge I$ and let $B$ be a closed accretive operator satisfying condition (i) with $a < 1$. Then the operator $-H=-(A+B)$ is the boundedly invertible generator of a contraction holomorphic semigroup of angle $\omega = \arccos a$, in the sense that $\|e^{-tH}\|\leq 1$ for $t\geq 0$. % % \end{theo} % % \begin{preuve} Our goal is to verify (\Ref{2.1}) for $z\in S_{\pi/2+\theta}$, $\theta\in (0,\omega)$, where $\omega = \arccos a < \pi/2$. To this end we use the representation % % \begin{equation}\Label{2.2} (H + z)f = \{I + BA^{-1}A(A + z)^{-1}\}(A + z)f, \qquad f \in \dom(H) = \dom(A). \end{equation} % % Since % % \begin{equation}\Label{2.3} \|A(A + z)^{-1}\| \le \sup_{\lambda \ge 1}\left|\frac{\lambda}{\lambda + z}\right| \leq \sup_{\lambda \ge 0}\left|\frac{\lambda}{\lambda + z}\right| \end{equation} % % a straightforward computation shows that % % \begin{equation}\Label{2.4} \|A(A + z)^{-1}\| \le \left\{\begin{array}{l} 1 \mbox{ for } \real z\geq 0 \\ |z|/|\imag z| \mbox{ for } \real z<0,\ \imag z\neq 0 \end{array}\right. \end{equation} % % For $z \in S_{\pi/2+\theta}$ and $\real z < 0$, then % % \begin{equation}\Label{2.5} \frac{|z|}{|\imag z|} \leq {1\over\cos\theta}. \end{equation} % % Hence (\ref{2.4}) leads to the estimate % % \begin{equation}\Label{2.6} \|A(A + z)^{-1}\| \le \frac{1}{\cos\theta} \end{equation} % % for $z \in S_{\pi/2+\theta}$ and $\real z<0$. If $\real z\geq 0$ then (\ref{2.4}) yields the estimate % \begin{equation}\Label{2.6-1} \|A(A + z)^{-1}\| \le 1 \le \frac{1}{\cos\theta}. \end{equation} Therefore the estimate (\ref{2.6}) holds for any $z\in S_{\pi/2+ \theta}$. Using the condition $(i)$ we find the estimate % % \begin{equation}\Label{2.8} \|BA^{-1}A(A + z)^{-1}\| \le \|BA^{-1}\|\|A(A + z)^{-1}\| \le \frac{a}{\cos\theta} < 1 \end{equation} % % for $z \in S_{\pi/2+\theta}$. Hence the operator $I + B(A + z)^{-1}$ is invertible for $z \in S_{\pi/2+\theta}$ and we have the estimate % % \begin{equation}\Label{2.9} \|(I + B(A + z)^{-1})^{-1}\| \le \frac{\cos\theta}{\cos\theta - a} \end{equation} % % for $z \in S_{\pi/2+\theta}$. Then from (\Ref{2.2}) we obtain that $z \in S_{\pi/2+\theta} \subseteq \rho(-H)$ the resolvent set of the operator $-H$ and the representation % % \begin{equation}\Label{2.10} (H + z)^{-1} = (A + z)^{-1}(I + B(A + z)^{-1})^{-1}, \end{equation} % and the estimate % \begin{equation}\Label{2.11} \|(H + z)^{-1}\| \le \frac{\cos\theta}{\cos\theta - a}\|(A + z)^{-1}\| \end{equation} for $z \in S_{\pi/2+\theta}$. % Since $A$ is self-adjoint, we have \begin{equation} \|(z+A)^{-1}\| \leq {1\over\cos\theta}{1\over|z|} \end{equation} for any $z\in S_{\pi/2+\theta}$. Thus we conclude \begin{equation}\Label{2.20} \|(H + z)^{-1}\| \le \frac{M_\theta}{|z|} \end{equation} for all $z\in S_{\pi/2+\theta}$ with \begin{equation}\Label{2.19} M_\theta := \frac{1}{\cos\theta - a} \end{equation} which shows that $-H$ is the generator of a holomorphic semigroup with angle $\omega = \arccos a$. Since $B$ is accretive and $A$ is self-adjoint positive, $H=A+B$ is accretive and since $\lambda+H$ is boundedly invertible for $\lambda>0$, $H$ is m-accretive, i.e. $-H$ generates a contraction semigroup. Finally by the representation $H=(I+BA^{-1})A$ with $\|BA^{-1}\|\leq a<1$, the operator $H$ is boundedly invertible. \end{preuve} {\noindent\bf Remark:} Since $\sigma(H)\subseteq S_{\pi/2-\omega}$ implies $\sigma(H^*)\subseteq S_{\pi/2 - \omega}$, the characterisation (\ref{2.1}) entails that $-H^*$ is also the generator of a holomorphic semigroup. Below we denote by $ \bD = \{z \in {\mathbb C}: |z| \le 1\}$ and $\mbox{D}_{1/2} = \{z \in {\mathbb C}: \left|z - 1/2\right| \le 1/2\}$, two discs in the complex plane $\bC$. % \begin{lem}\Label{II.2} % % Let $A \ge I$. Then for any $\tau > 0$ % % \begin{equation}\Label{2.21} \|(I - e^{-\tau A})(z - e^{-\tau A})^{-1}\| \le \left\{\begin{array}{l} |z|^{-1} \qquad z \in \mathbb{D} \setminus \mathrm{D}_{1/2} \\ |1 - z|/|\Im{\mathrm m}\, z| \qquad z \in \mathrm{D}_{1/2} \setminus [0,1] \end{array}\right. \end{equation} % % %\begin{equation}\Label{2.22} %\mbox{D}_{1/2} := \left\{z \in {\mathbb C}: \left|z - \frac{1}{2}\right| \le %\frac{1}{2}\right\}. %\end{equation} % % \end{lem} % % {\em Proof:} We use the functional calculus for self-adjoint operators. Obviously, we have % % \begin{equation}\Label{2.23} \|(I - e^{-\tau A})(z - e^{-\tau A})^{-1}\| \le \sup_{\lambda \ge 1} \left|{1 - e^{-\tau\lambda}\over z - e^{-\tau\lambda}}\right| \le \sup_{\lambda \ge 0} \left|{1 - e^{-\lambda}\over z - e^{-\lambda}}\right|. \end{equation} % % Let $z = x + iy$ and % % \begin{equation}\Label{2.24} m(\lambda) := \frac{1 - e^{-\lambda}}{|z - e^{-\lambda}|} = \frac{1 - e^{-\lambda}}{\sqrt{(x - e^{-\lambda})^2 + y^2}}, \quad \lambda \ge 0. \end{equation} % % If $z \in \bD \setminus \mbox{D}_{1/2}$, then $m'(\lambda) \ge 0$ for $\lambda \ge 0$. Therefore $m(\lambda)$ is a non-decreasing function such that % % \begin{equation}\Label{2.26} \sup_{\lambda\geq 0}m(\lambda) = \lim_{\lambda \to +\infty} m(\lambda) = \frac{1}{|z|}, \end{equation} % % which proves the first estimate (\Ref{2.21}). If $z \in \mbox{D}_{1/2} \setminus [0,1]$, then $m(\lambda)$ attains its maximum at $\lambda_0$ such that % % \begin{equation}\Label{2.27} e^{-\lambda_0} = \frac{\frac{1}{4} - (x - \frac{1}{2})^2 - y^2}{1 - x}. \end{equation} % % Inserting (\Ref{2.27}) into (\Ref{2.24}) we obtain % % \begin{equation}\Label{2.28} m(\lambda_0) = \frac{|1 - z|}{|\imag z|}, \end{equation} % % which yields % % \begin{equation}\Label{2.29} 0 \le m(\lambda) \le \frac{|1 - z|}{|\imag z|}, \end{equation} % % The inequality (\ref{2.29}) verifies the second estimate in (\Ref{2.21}). \hfill$\Box$ Since for $A\geq I$ and an m-accretive $B$ the operator % % \begin{equation}\Label{2.30} T(\tau) = e^{-\tau A}e^{-\tau B}, \qquad \tau > 0, \end{equation} % % is a contraction, its spectrum $\sigma(T(\tau)) \subseteq \bD$, $\tau > 0$. However, for $B$ small relative to $A$ we can show more. To this end we introduce the family of closed convex sets in $\bC$, for $r\in [0,1]$ \begin{equation}\Label{2.33} \Sigma_{r} := \Sigma^0_{r} \cup \Sigma^1_{r}, \end{equation} % where % \begin{equation}\Label{2.31} \Sigma^0_{r} := \{z \in \mbox{D}_{1/2}:\ |\imag z| \le r|1 - z|\} \end{equation} % and % \begin{equation}\Label{2.32} \Sigma^1_r := \{z\in\bD\setminus\mbox{D}_{1/2}:\ |z|\leq r\}, \end{equation} % see Figure 1. We remark that the family $\Sigma_r$ is increasing with $r$: for $r 0$. Moreover, for any $r\in(a_*,1)$, one obtains the estimate \begin{equation}\Label{2.61} \|(z - T(\tau))^{-1}\| \le \frac{r}{r - a_*}\|(z - e^{-\tau A})^{-1}\|,\ z\in R_r=\bD\setminus\Sigma_r. \end{equation} % % \end{theo} % % {\em Proof:} One has to prove that each point $z \in \bD \setminus \Sigma_{a_*}=R_{a_*}$ belongs to the resolvent set $\rho(T(\tau))$ of $T(\tau)$ for any $\tau > 0$. To this end we use the representation % % \begin{equation}\Label{2.34} z - T(\tau) = z - e^{-\tau A} + e^{-\tau A}(I - e^{-\tau B}). \end{equation} % % If $\tau > 0$, then the operator $(I - e^{-\tau A})$ is invertible. Let % % \begin{equation}\Label{2.35} X(\tau,z) := (I - e^{-\tau A})(z - e^{-\tau A})^{-1} \end{equation} % % and % % \begin{equation}\Label{2.36} Y(\tau) := (I - e^{-\tau A})^{-1}e^{-\tau A}(I - e^{-\tau B}). \end{equation} % % Then one obviously gets the representation % % \begin{equation}\Label{2.37} z - T(\tau) = (z - e^{-\tau A})\{I + X(\tau,z)Y(\tau)\}. \end{equation} % % For each $f \in \dom(B)$ one has % % \begin{equation}\Label{2.38} (I - e^{-\tau B})f = \int^\tau_0 ds Be^{-s B}f. \end{equation} % % Hence we get % % \begin{equation}\Label{2.39} A^{-1}(I - e^{-\tau B})f = \int^\tau_0 ds A^{-1}Be^{-s B}f. \end{equation} % % This leads to the estimate % % \begin{equation}\Label{2.40} \|A^{-1}(I - e^{-\tau B})\| \le \tau a_* \end{equation} % % Using for (\ref{2.36}) the representation % % \begin{equation}\Label{2.42} Y(\tau) = (I - e^{-\tau A})^{-1}Ae^{-\tau A}A^{-1}(I - e^{-\tau B}) \end{equation} % % we obtain % % \begin{equation}\Label{2.43} \|Y(\tau)\| \le \tau a_* \|(I - e^{-\tau A})^{-1}Ae^{-\tau A}\|. \end{equation} % % By the functional calculus for self-adjoint operators one verifies that % % \begin{equation}\Label{2.44} \tau\|(I - e^{-\tau A})^{-1}Ae^{-\tau A}\| \le 1 \end{equation} % % for each $\tau > 0$ and by consequence % % \begin{equation}\Label{2.45} \|Y(\tau)\| \le a_*\ . \end{equation} % % Let $r>a_*$. To estimate $X(\tau,z)$ we have to consider two cases: 1. Let $z \in \mbox{D}_{1/2}\cap R_{r}$. Then $|\imag z| > r|1 - z|$. % % Since $z \in \mbox{D}_{1/2}$ by Lemma \Ref{II.2} we find % % \begin{equation}\Label{2.47} \|X(\tau,z)\| \le \frac{|1 - z|}{|\imag z|}. \end{equation} % % Hence we have % \begin{equation}\Label{2.50} \|X(\tau,z)\| < \frac{1}{r} \end{equation} % % for $z \in R_{r}\cap\mbox{D}_{1/2}$ and $\tau > 0$. 2. Let $z \in R_{r} \setminus \mbox{D}_{1/2}$. Since $|z|>r$ by Lemma \Ref{II.2} we find % % \begin{equation}\Label{2.52} \|X(\tau,z)\| \le \frac{1}{|z|} < \frac{1}{r}. \end{equation} % Finally we see that the estimate % % \begin{equation}\Label{2.55} \|X(\tau,z)\| < \frac{1}{r} \end{equation} % % is valid for any $z \in R_{r}$ and $\tau > 0$. Moreover, by (\Ref{2.45}) and (\Ref{2.55}) we have % % \begin{equation}\Label{2.56} \| X(\tau,z)Y(\tau)\| \le \|X(\tau,z)\|\|Y(\tau)\| < {a_*\over r} \leq 1 \end{equation} % % for $z \in R_{r}$, $\tau > 0$. This yields that the operator $\{I + Y(\tau) X(\tau,z) \}$ is invertible for $z \in R_{a_*}=\bigcup_{r>a_*}R_r$ and $\tau > 0$, with the following estimate for $z\in R_r$, $r>a_*$: \begin{equation} \|(I+X(\tau,z)Y(\tau,z))^{-1}\| \leq {1\over 1-a_*/r},\ \tau>0. \end{equation} Since the operator $(z - e^{-\tau A})$ is invertible for $z \in R_{a_*}$ and $\tau > 0$, we obtain from the representation (\Ref{2.37}) that the operator $T(\tau)$ is invertible for $z \in R_{a_*}$, $\tau > 0$, and the announced estimate for $z\in R_r$, $r\in(a_*,1)$. \hfill $\Box$ Let us define the closed curve $\Gamma_r=\partial\Sigma_r$. Outside the disc $\mbox{D}_{1/2}$, it coincides with the arc of centre $0$ and of radius $r$, whereas inside $\mbox{D}_{1/2}$, this curve consists of two segments of tangents to the preceding arc passing the point $1$ (see Figure 1). \begin{figure}[H] \epsfysize=10cm $$\epsfbox{CNZ1.eps}$$ \caption{Illustration of the set $\Sigma_{a_*}$ (shaded domain) with boundary $\Gamma_{a_*}=\partial\Sigma_{a_*}$, where $a_* = \sin \alpha$, as well as of our choice of the contour $\Gamma_r$ in the resolvent set $\rho(F(\tau))$, where $r=\sin \beta >a_*$. The contour $\Gamma_r$ consists of two segments of tangent straight lines $(1,A)$ and $(1,B)$ and the arc $(A,B)$ of radius $r$. The dotted circle $\partial \mbox{D}_{1/2}$ corresponds to the set of tangent points for the different values of $r\in[0,1]$.} \end{figure} \begin{theo}\Label{II.5} % % Let $A \ge I$ and let $B$ be an m-accretive operator obeying condition (ii) with $a_*<1$. Then there is a constant $C$ such that % % \begin{equation}\Label{2.68} \|T(\tau)^k(I - T(\tau))\| \le \frac{C}{k + 1}, \qquad k = 0,1,2,\ldots, \end{equation} % % holds for any $\tau > 0$. % % \end{theo} % % {\em Proof:} By Theorem \ref{II.3}, one gets that for any $r>a_*$ and $\tau > 0$, $\Gamma_r\in\rho(T(\tau))$, the resolvent set of $T(\tau)$. Using the Dunford-Taylor calculus we get the representation % % \begin{equation}\Label{2.69} T(\tau)^k(I - T(\tau)) = \frac{1}{2\pi i}\int_{\Gamma_r} z^k(1 - z)(z - T(\tau))^{-1} dz. \end{equation} % % We estimate the integral (\Ref{2.69}) with help of the estimate (\ref{2.61}): \begin{equation}\label{esinteg1} \left\|\frac{1}{2\pi i}\int_{\Gamma_r} z^k(1 - z)(z - T(\tau))^{-1} dz\right\| \leq {1\over 2\pi}\int_{\Gamma_r} |z^k(1-z)|{r\over r-a_*}\|(z-e^{-\tau A})^{-1}\||dz|. \end{equation} Since the integral in the right hand side of (\ref{esinteg1}) is invariant with respect to conjugation $z\rightarrow\overline{z}$, it is sufficient to estimate the integral on the path $\Gamma^+_r$: the branch of $\Gamma_r$ with $\imag z>0$. This branch consists of segment $(1,A)$ and arc $(\pi/2-\beta,\pi)$ of radius $r$, see Figure 1. Let us parametrize $(1,A)$ by \begin{equation}\label{para1} z = 1 - s e^{-i\beta},\ 0\leq s\leq \cos\beta,\mbox{ with } r=\sin\beta. \end{equation} Since for (\ref{para1}) and the self-adjoint $e^{-\tau A}$ one has $\|(z-e^{-\tau A})^{-1}\| \leq (\imag z)^{-1} \leq (s \sin\beta)^{-1}$, we get \begin{equation}\label{esinteg2} \int_{(1,A)}|z^k(1-z)|\|(z-e^{-\tau A})^{-1}\||dz| \leq \int_0^{\cos\beta} ds |1-s e^{-i\beta}|^k{1\over\sin\beta} \end{equation} Notice that for $0\leq s\leq \cos\beta$ one gets by convexity that \begin{equation} |1-s e^{-i\beta}| \leq 1 - s{1-\sin\beta\over\cos\beta} \end{equation} Therefore, we obtain for (\ref{esinteg2}) the estimate \begin{equation}\label{esinteg3} \int_{(1,A)}|z^k(1-z)|\|(z-e^{-\tau A})^{-1}\||dz| \leq {\cos\beta\over\sin\beta (1-\sin\beta)}{1-(\sin\beta)^k\over k+1} \end{equation} % For the arc $(\pi/2-\beta,\pi)$ of radius $r$, one has the estimate $\|(z-e^{-\tau A})^{-1}\| \leq (r\cos\beta)^{-1}$. Then we find: \begin{equation}\label{esinteg4} \int_{\Gamma^+_r\setminus (1,A)}|z^k(1-z)|\|(z-e^{-\tau A})^{-1}\||dz| \leq (\pi/2 + \beta)r{2r^k\over r\cos\beta} \leq {2\pi\over\cos\beta}{1/r\over e \ln (1/r)} {1\over k+1}. \end{equation} In the last inequality we use that: $\sup_{k\geq 0} (k+1) e^{-k\ln x} = x/(e\ln x)$. Finally, combining (\ref{esinteg1}), (\ref{esinteg3}) and (\ref{esinteg4}) we obtain: \begin{eqnarray} & & \hspace{-2cm} \left\|\frac{1}{2\pi i}\int_{\Gamma_r} z^k(1 - z)(z - T(\tau))^{-1} dz\right\| \leq \\ & & 2\left({\cos\beta\over 2\pi\sin\beta(1-\sin\beta)} + {1/r\over e\cos\beta\ln (1/r)}\right){r\over r-a_*}{1\over k+1} \nonumber \end{eqnarray} which gives the announced estimate (\ref{2.68}) with \begin{equation} C=\left({r\cos\beta\over 2\pi\sin\beta(1-\sin\beta)} + {1\over e\cos\beta\ln (1/r)}\right){2\over r-a_*}. \end{equation} \hfill $\Box$ \noindent{\bf Remark:} If $T(\tau) = e^{-\tau A}e^{-\tau B}$ is replaced by $F(\tau) = e^{-\tau B}e^{-\tau A}$, then the above Theorems \ref{II.3} and \ref{II.5} remain valid if $B$ satisfies, instead of $(ii)$, the condition $(i)$ with $a<1$. For the proof one has to consider the adjoint family $F(\tau)^* = e^{-\tau A}e^{-\tau B^*}$. In this case we have simply to replace $a_*$ by $a$ in the course of proofs. \section{Proof of the main Theorem} \setcounter{equation}{0} \renewcommand{\theequation}{\arabic{section}.\arabic{equation}} \setcounter{theo}{0} \renewcommand{\thetheo}{\arabic{section}.\arabic{theo}} First we recall the following property of holomorphic semigroups (see e.g. \cite[Ch.IX]{Kato}): \begin{prop}\Label{III.2} % % Let $-H$ be the generator of a bounded holomorphic semigroup. Then $\ran (e^{-tH}) \subseteq \dom(H)$, and we have the estimate: % % \begin{equation}\Label{CH} \left\|He^{-tH}\right\| \le \frac{C_H}{t}, \qquad t > 0. \end{equation} % % \end{prop} \begin{lem}\label{III.4} % % Let $A$ be a self-adjoint operator, $A\geq I$. Let operator $B$ be m-accretive, satisfying (ii) with $a_*<1$. Then operator $I - T(\tau)$ is boundedly invertible for each $\tau > 0$ and % % \begin{equation}\Label{3.16} \left\|(I - T(\tau))^{-1}u\right\| \le \frac{1}{1 - a_*}\left(\|u\| + \frac{1}{\tau}\|A^{-1}u\|\right), \qquad u \in \kH, \end{equation} % % \end{lem} % % {\em Proof:} We use representation % % \begin{equation}\Label{3.17} I - T(\tau) = I - e^{-\tau A} + e^{-\tau A}(I - e^{-\tau B}). \end{equation} % % Since $\|e^{-\tau A}\|\leq e^{-\tau}<1$, the operator $I - e^{-\tau A}$ is boundedly invertible provided $\tau > 0$. Hence, one gets representation % % \begin{equation}\Label{3.18} I - T(\tau) = (I - e^{-\tau A})\left(I + Y(\tau)\right). \end{equation} % where we have used the notation (\Ref{2.36}). By (\Ref{2.45}) and $a_* < 1$ we obtain that $I - T(\tau)$ is invertible. Moreover, we have the estimate : \begin{equation}\label{es3.5} \|(I - T(\tau))^{-1}u\| \leq {1\over 1-a_*} \left\|\left(I - e^{-\tau A}\right)^{-1}u\right\|, \quad u\in \kH \end{equation} By Lemma 2.3 of \cite{NZ1}, we have for any $u\in{\kH}$ and $\tau>0$ that \begin{equation}\label{es3.6} \left\|\left(I - e^{-\tau A}\right)^{-1}u\right\| \leq \|u\| + {1\over \tau}\|A^{-1}u\|. \end{equation} The estimates (\ref{es3.5}), (\ref{es3.6}) lead to the desired inequality (\Ref{3.16}) \hfill$\Box$ \begin{lem}\label{III.5} Let $-A$, boundedly invertible, and $-B$ be generators of bounded semigroups, with $B$ satisfying the conditions (i), (ii) for some (not necessarily bounded by 1) $a$ and $a_*$. Assume that operator $-H=-(A+B)$ is the boundedly invertible generator of a bounded semigroup. Then there exists a constant $L_1$ such that for all $\tau>0$: \begin{eqnarray} \left\|A^{-1}\left(e^{-\tau A}e^{-\tau B} - e^{-\tau(A+B)}\right)\right\| & \leq & L_1\tau \label{L11}\\ \left\|\left(e^{-\tau A}e^{-\tau B} - e^{-\tau(A+B)}\right)A^{-1}\right\| & \leq & L_1\tau \label{L12} \end{eqnarray} \end{lem} \begin{preuve} See Lemma 5 in \cite{CZ1}. \end{preuve} \begin{lem}\label{III.6} Let $A$, $B$ and $H=A+B$ be the same as in Lemma \ref{III.5}. Then there exists a constant $L_2$ such that for all $\tau\geq 0$ : \begin{equation} \left\|A^{-1}\left(e^{-\tau A}e^{-\tau B} - e^{-\tau(A+B)}\right)A^{-1}\right\| \leq L_2\tau^2 \label{L21} \end{equation} \end{lem} \begin{preuve} See Lemma 6 in \cite{CZ1}. \end{preuve} Now we are in position to prove our main Theorem. % \begin{theo}\Label{III.7} % % Let $A$ be a self-adjoint operator and $A\geq I$. Let $B$ satisfy conditions (i), (ii), (iii), and $a,a_*<1$. Then there is a constant $L > 0$ such that the estimate % % \begin{equation}\Label{3.24} \left\|\left(e^{-tA/n}e^{-tB/n}\right)^n - e^{-tH}\right\| \le L\;\frac{\ln(n)}{n}, \qquad n = 3,4,\ldots, \end{equation} % % holds uniformly in $t \ge 0$, where $H=A+B$ with $\dom(H) = \dom(A)$. % % \end{theo} % % {\em Proof:} Let $\tau = t/n$ and $U(\tau) = e^{-\tau H}$, which is a holomorphic contraction semigroup by Theorem \ref{II.1}. We start with the representation % % \begin{eqnarray}\Label{3.25} \lefteqn{T(\tau)^n - U(\tau)^n} \\ & = & T(\tau)^{n-1}(T(\tau) - U(\tau)) + \sum^{n-1}_{m=1}T(\tau)^{n-m-1}(T(\tau) - U(\tau))U(\tau)^m. \nonumber \end{eqnarray} % % Since by Lemma \Ref{III.4} the operator $(I - T(\tau))^{-1}$ exists for any $\tau>0$, we find % % \begin{eqnarray}\Label{3.26} \lefteqn{T(\tau)^n - U(\tau)^n = T(\tau)^{n-1}(T(\tau) - U(\tau))}\\ & & + \sum^{n-1}_{m=1}T(\tau)^{n-m-1}(I - T(\tau))(I - T(\tau))^{-1}(T(\tau) - U(\tau))U(\tau)^m. \nonumber \end{eqnarray} % % This leads to the estimate % % \begin{eqnarray}\Label{3.27} \lefteqn{\|T(\tau)^n - U(\tau)^n\| \le \|T(\tau)^{n-1}(T(\tau) - U(\tau))\|}\\ & & + \sum^{n-1}_{m=1}\|T(\tau)^{n-m-1}(I - T(\tau))\|\|(I - T(\tau))^{-1}(T(\tau) - U(\tau))U(\tau)^m\|. \nonumber \end{eqnarray} % % Then by (\Ref{3.16}) we obtain % % \begin{eqnarray}\Label{3.28} \lefteqn{\|(I - T(\tau))^{-1}(T(\tau) - U(\tau))U(\tau)^m\|}\\ & \le & \frac{1}{1 - a_*}\left\{\|(T(\tau) - U(\tau))U(\tau)^m\| + \frac{1}{\tau}\left\|A^{-1}(T(\tau) - U(\tau))U(\tau)^m\right\|\right\}. \nonumber \end{eqnarray} % % Since % % \begin{equation}\Label{3.29} (T(\tau) - U(\tau))U(\tau)^m = (T(\tau) - U(\tau))A^{-1}AH^{-1}HU(\tau)^m, \end{equation} % % one gets % % \begin{equation}\Label{3.30} \|(T(\tau) - U(\tau))U(\tau)^m\| \le \left\|(T(\tau) - U(\tau))A^{-1}\right\|\|AH^{-1}\|\|HU(\tau)^m\|. \end{equation} % % Applying now Proposition \Ref{III.2} we obtain % % \begin{equation}\Label{3.31} \|(T(\tau) - U(\tau))U(\tau)^m\| \le \left\|A^{-1}(T(\tau) - U(\tau))\right\|\|AH^{-1}\|\frac{C_H}{\tau}\frac{1}{m}. \end{equation} % % By $AH^{-1} = (I + BA^{-1})^{-1}$ and $\|BA^{-1}\|\leq a<1$ one gets % % \begin{equation}\Label{3.32} \|AH^{-1}\| \le \frac{1}{1 - a}. \end{equation} % % Hence % % \begin{equation}\Label{3.34} \|(T(\tau) - U(\tau))U(\tau)^m\| \le \left\|A^{-1}(T(\tau) - U(\tau))\right\|\frac{C_H}{(1 - a)\tau}\frac{1}{m}. \end{equation} % % Similarly we get % % \begin{equation}\Label{3.35} \|A^{-1}(T(\tau) - U(\tau))U(\tau)^m\| \le \left\|A^{-1}(T(\tau) - U(\tau))A^{-1}\right\|\frac{C_H}{(1 - a)\tau}\frac{1}{m}. \end{equation} % % Inserting (\Ref{3.34}) and (\Ref{3.35}) into (\Ref{3.28}) we find % % \begin{eqnarray}\Label{3.40} \lefteqn{\|(I - T(\tau))^{-1}(T(\tau) - U(\tau))U(\tau)^m\|}\\ & \le & \left\{\frac{1}{\tau}\|(A^{-1}(T(\tau) - U(\tau))\| + \frac{1}{\tau^2}\left\|A^{-1}(T(\tau) - U(\tau))A^{-1}\right\|\right\}\frac{C_H}{(1 - a_*)(1-a)}\frac{1}{m}. \nonumber \end{eqnarray} % % Applying to (\Ref{3.40}) Lemma \Ref{III.5} and Lemma \Ref{III.6} we find % % \begin{equation}\Label{3.41} \|(I - T(\tau))^{-1}(T(\tau) - U(\tau))U(\tau)^m\| \le C_H\frac{L_1 + L_2}{(1 - a_*)(1-a)}\frac{1}{m} = \frac{L_3}{m}, \end{equation} % % where we put $L_3 := C_H(L_1 + L_2)/(1-a_*)(1-a)$. Further, by Theorem \Ref{II.5} we have % % \begin{equation}\Label{3.42} \|T(\tau)^{n-m-1}(I - T(\tau))\| \le \frac{C}{n - m}, \qquad n = 2,3,\ldots, \qquad m = 0,1,2,\ldots,n-1. \end{equation} % % Inserting (\Ref{3.41}) and (\Ref{3.42}) into (\Ref{3.27}) we get % % \begin{equation}\Label{3.43} \|T(\tau)^n - U(\tau)^n\| \le \|T(\tau)^{n-1}(T(\tau) - U(\tau))\| + C L_3\sum^{n-1}_{m=1}\frac{1}{(n-m)m}. \end{equation} % % Since % % \begin{eqnarray}\Label{3.44} \lefteqn{\sum^{n-1}_{m=1}\frac{1}{(n-m)m}}\\ & = & \frac{2}{n}\sum^{n-1}_{m=1}\frac{1}{m} \le \frac{2}{n}\left(1 + \ln(n-1)\right) \le 4\frac{\ln(n)}{n}, \qquad n = 2,3,\dots, \nonumber \end{eqnarray} % % one has % % \begin{equation}\Label{3.45} \|T(\tau)^n - U(\tau)^n\| \le \|T(\tau)^{n-1}(F(\tau) - U(\tau))\| + 4C L_3\frac{\ln(n)}{n}. \end{equation} % % It remains to estimate $\|T(\tau)^{n-1}(T(\tau) - U(\tau))\|$. Again by invertibility of $I - T(\tau)$ (see Lemma 3.2) we find % % \begin{equation}\Label{3.46} \|T(\tau)^{n-1}(T(\tau) - U(\tau))\| \le \|T(\tau)^{n-1}(I - T(\tau)\|\|(I - T(\tau))^{-1}(T(\tau) - U(\tau))\|. \end{equation} % % By (\Ref{3.16}) we get % % \begin{eqnarray}\Label{3.47} \lefteqn{\|(I - T(\tau))^{-1}(T(\tau) - U(\tau))\|}\\ & \le & \frac{1}{1 - a_*}\left\{\|T(\tau) - U(\tau)\| + \frac{1}{\tau}\left\|A^{-1}(T(\tau) - U(\tau))\right\|\right\} \nonumber \\ & \le &\frac{1}{1 - a_*}\left\{2 + \frac{1}{\tau}\left\|A^{-1}(T(\tau) - U(\tau))\right\|\right\}. \nonumber \end{eqnarray} % % Applying to the right-hand side of (\Ref{3.47}) Lemma \Ref{III.5} we find % % \begin{equation}\Label{3.63} \|(I - T(\tau))^{-1}(T(\tau) - U(\tau))\| \le \frac{2 + L_1}{1 - a_*}. \end{equation} % % Inserting (\Ref{3.63}) into (\Ref{3.46}) one gets % % \begin{equation}\Label{3.64} \|T(\tau)^{n-1}(T(\tau) - U(\tau))\| \le \frac{2 + L_1}{1 - a_*}\|T(\tau)^{n-1}(I - F(\tau)\|. \end{equation} % % Then applying Theorem \Ref{II.5} to the right-hand side of (\Ref{3.64}) we obtain % % \begin{equation}\Label{3.65} \|T(\tau)^{n-1}(T(\tau) - U(\tau))\| \le \frac{2 + L_1}{1 - a_*}\;\frac{C}{n},\ n=1,2,\dots\ . \end{equation} % % Inserting (\Ref{3.65}) into (\Ref{3.45}) we find the estimate (\Ref{3.24}): % % \begin{equation}\Label{3.66} \|T(\tau)^n - U(\tau)^n\| \le L\;\frac{\ln(n)}{n}, \qquad n = 3,4,\dots\ . \end{equation} % % Here $L := C(2 + L_1)(1 - a_*)^{-1} + 4C L_3$ and we used $1/n<\ln(n)/n$ to estimate (\ref{3.65}), for $n=3,4,\dots$ \hfill$\Box$ % \begin{cor}\label{cor1} Let $A$ and $B$ satisfy the conditions of Theorem \ref{III.7}. Then there are constants $L', L''$ such that \begin{eqnarray} \left\|\left(e^{-tB/n}e^{-tA/n}\right)^n - e^{-tH}\right\| & \leq & L'\;\frac{\ln(n)}{n}, \qquad n = 3,4,\ldots,\label{escor11}\\ \left\|\left(e^{-tA/2n}e^{-tB/n}e^{-tA/2n}\right)^n - e^{-tH}\right\| & \leq & L''\;\frac{\ln(n)}{n}, \qquad n = 3,4,\ldots\ .\label{escor12} \end{eqnarray} holds uniformly in $t\geq 0$. \end{cor} \begin{preuve} Since $B^*$ satisfies conditions of Theorem \ref{III.7}, we have \begin{equation} \left\|\left(e^{-tA/n}e^{-tB^*/n}\right)^n - e^{-tH^*}\right\| \leq L'\,\frac{\ln (n)}{n}. \end{equation} Notice that $H^* = A+B^*$ is well defined and m-accretive. By taking the adjoint sequence of operators, we find (\ref{escor11}). In order to obtain (\ref{escor12}), one has to observe that the line of reasoning in the proof of Theorem \ref{III.7} follows thought verbatim for the symmetrized formula. \end{preuve} \section{Conclusion} \setcounter{equation}{0} \renewcommand{\theequation}{\arabic{section}.\arabic{equation}} \setcounter{theo}{0} \renewcommand{\thetheo}{\arabic{section}.\arabic{theo}} The aim of the present paper was to show that the operator-norm error bound estimates for the exponential Trotter product formula persist if for a couple {$A = A^*, B$} we pass from a self-adjoint $A$-small perturbation $B$ to an m-accretive $A$-small operator $B$. This generalizes the previous result \cite{NZ1} to the case of a non-self-adjoint $B$ under conditions that $\|Bu\| \leq a\|Au\|$, $u\in\dom(A)$ with $a\in(0,1)$ and $\|B^*u\|\leq a_*\|Au\|$, $u\in\dom(A)$ with $a_*\in(0,1)$, see Theorem \ref{III.7}. Let us consider the following special case: $B(z) = zV$ where $V$ is a self-adjoint operator which is relatively small with respect to $A \ge I$, i.e. $\dom(A) \subseteq \dom(V)$ and % % \begin{equation}\Label{4.0} \|Vf\| \le a\|Af\|, \quad f \in \dom(A), \quad \mbox{for some} \quad a \in (0,1). \end{equation} % % Obviously, the operators $B(z) = zV$ and $B(z)^* = \overline{z}V$ are m-accretive and satisfy the conditions $(i)$-$(ii)$ for $\real z\geq 0$. If, in addition, $|z|\leq 1$, then the constants $a,a_*$ for $B(z)$ remain smaller than $1$. Applying Theorem \Ref{III.7} we find, for instance, that the error estimate % % \begin{equation}\Label{4.0-1} \|(e^{-tA/n}e^{-tzV/n})^n - e^{-tH(z)}\| \le L\frac{\ln(n)}{n}, \quad n = 2,3\ldots, \end{equation} % % holds uniformly in $t \ge 0$ and in $z$ such that $\real z\geq 0$, $|z|\leq 1$, where $H(z) = A+zV$ on $\dom(H(z)) = \dom(A)$. In the case $z=i$, we consider the perturbation of a self-adjoint contraction semigroup by a unitary group. % % Since $a,a^*$ may be strictly positive, the present paper is also a generalisation of \cite{CZ1}, where the estimate (\ref{1.5}) was found for $B$ which is {\it infinitesimally} Kato-small with respect to $A$. Notice that in \cite{CZ1} we did not suppose that the operator $A$ is self-adjoint but only that it generates a contraction holomorphic semigroup. It is the case in particular when $A$ is an m-sectorial operator with vertex in $0$, see \cite[Ch.IX]{Kato}.\\ Therefore, it is relevant to ask: why do we need $A = A^*$ in the main Theorem of the present paper ? First, this condition is important for estimate $\|(z-e^{-\tau A})^{-1}\|$, see (\ref{esinteg2}) and (\ref{esinteg4}) in the proof of Theorem 2.4. Second, it makes possible to use the spectral theorem to obtain the estimates (\ref{2.21}), see Lemma 2.2, and (\ref{3.16}), see Lemma 3.2.\\ Suppose, for example, that instead to be self-adjoint the operator $A$ is m-sectorial. Then operator-norm continuity of the holomorphic semigroup $U_{A}(t)= e^{-tA}$ for $ t > 0$, allows to localize the {\it spectrum} of this semigroup $\sigma(e^{-tA})$ but {\it not} the numerical range $\Theta(e^{-tA})$ ! However to estimate the norm of the resolvent $(z-e^{-\tau A})^{-1}$, the spectrum is not sufficient in general, except for normal (in particular self-adjoint) $A$, for which $\overline{\Theta(e^{-tA})}$ is the convex hull of $\sigma(e^{-tA})$. Thus we solve the problem here only for the couple of self-adjoint $A$ and m-accretive $B$, i.e. the limiting case of angles $\theta_A = 0$ and $\theta_B = \pi/2$.\\ Although the above mentioned result of \cite{CZ1} allows to pass to an m-sectorial operator $A$ with $0 < \theta_A < \pi/2$, the price is that $B$ and $B^*$ should be infinitesimally $A-$small if one insists on the {\it error bound estimate}. If {\it not}, then \cite{CZ2} for two m-sectorial operators $A$ and $B$ ( $\theta_A,\theta_B\in(0,\pi/2)$ ) such that: -- either $(I+A)^{-1}$ or $(I+\real A)^{-1/2}(I+\real B)^{1/2}$ is compact, -- or $A$ and $B$ satisfy conditions % \begin{enumerate} \item[$(i)$] ${\Re \mbox{\rm e }}A , {\Re \mbox{\rm e }}B \geq \gamma > 0$ ; % \item[$(ii)$] $\dom({\Re \mbox{\rm e }}A) \subseteq \dom({\Im \mbox{\rm m }}A)$ ; % \item[$(iii)$] $\dom({\Re \mbox{\rm e }}B) \subseteq \dom({\Im \mbox{\rm m }}B)$ ; % \item[$(iv)$] $\dom(({\Re \mbox{\rm e }}A)^\alpha) \subseteq \dom(({\Im \mbox{\rm m }}B)^\alpha)$ for some $\alpha\in(1/2,1]$ ; % \item[$(v)$] $\|({\Re \mbox{\rm e }}B)^\alpha u\| \leq c \|({\Re \mbox{\rm e }}A)^\alpha u\|$ , $u\in \dom (({\Re \mbox{\rm e }}A)^\alpha)$, $0 < c < 1$ and some $\alpha\in(1/2,1]$, % \end{enumerate} one gets \begin{equation} \|.\|-\lim_{n\rightarrow +\infty }(e^{-tA/n}e^{-tB/n})^{n}=e^{-tH} . \label{4.1} \end{equation} The operator-norm convergence in (\ref{4.1}) is uniform in $t$ for any compact set of the sector $S_{\omega}=\{z\in {\bC}:z\neq 0$ and $|{\arg }(z)|<\omega\}$, where ${\omega}= {\pi/2}- \max\{\theta_A,\theta_B\}$ and $H$ is the form-sum of $A$ and $B$.\\ We conclude by remark that our main Theorem 3.5 is valid if, instead of $e^{-tA/n}$, one considers operator $ f(tA/n)$. Here $f : [0,+\infty) {\rightarrow} [0,1]$ is a Borel function such that, cf.\cite{NZ1} :\\ \begin{eqnarray} C_0 & := &\sup_{x > 0} \left\{\frac{x\sqrt{f(x)}}{1 - f(x)}\right\} < + \infty \Label{4.2} \\[2mm] C_1 & := & \sup_{x > 0} \left\{ \frac{1 - f(x)}{x}\right\} < + \infty \Label{4.3} \\[2mm] C_2 & := & \sup_{x > 0} \left|\left(f(x) - \frac{1}{1 + x}\right)\frac{1}{x^2}\right| < + \infty \Label{4.4} \\[2mm] \end{eqnarray} % % as well as % % \begin{equation} 0 < C_0a < 1. \Label{4.5} \end{equation} Then we get \begin{equation} \left\|\left(f(tA/n)e^{-tB/n}\right)^n - e^{-t(A+B)}\right\| \leq L_0 {\ln n\over n},\ n\geq 3 \Label{4.6} \end{equation} uniformly in $t\geq 0$. The proof needs only minor modifications of Lemmata 3.2-3.4 as it is outlined in \cite{NZ1}. \section*{Acknowledgements} H.N. would like to thank Deutsche Forschungsgemeinschaft for financial support and Centre de Physique Th\'{e}orique for hospitality. \begin{thebibliography}{99} \bibitem{CZ2} Cachia, V.; Zagrebnov, V.A.: Operator-norm convergence of the Trotter product formula for sectorial generators, to appear in {\em Lett. Math. Phys.} (2000). \bibitem{CZ1} Cachia, V.; Zagrebnov, V. A.: Operator-norm convergence of the Trotter product formula for holomorphic semigroups, to appear in {\em J. Operator Theory} (2000). \bibitem{Ch1} Chernoff, P.R.: Note on product formulas for operator semigroups. {\em J. Funct. Anal.} {\bf 2} (1968), 238-242. \bibitem{Ch2} Chernoff, P.R.: Product formulas, nonlinear semigroups and addition of unbounded operators. {\em Mem. Am. Math. Soc. } {\bf 140} (1974), 1-121. \bibitem{Hi1} Hiai, F.: Trace norm convergence of exponential product formula {\em Lett. Math. Phys.} {\bf 33} (1995), 147-158. \bibitem{IT1} Ichinose, T.; Tamura, H.: Error estimate in operator norm for Trotter-Kato product formula. {\em Integr. Equ. Oper. Theory} {\bf 27} (1997), 195-207. \bibitem{IT2} Ichinose, T.; Tamura, H.: Error bound in Trace norm for Trotter-Kato product formula of Gibbs semigroups. {\em Asymptotic Analysis} {\bf 17} (1998), 239-266. \bibitem{Kato} Kato T.: Perturbation theory for linear operators, Springer Verlag, Berlin 1966. \bibitem{Ka2} Kato, T: On the Trotter-Lie product formula. {\em Proc. Japan Acad.} {\bf 50} (1974), 694-698. \bibitem{Ka1} Kato, T.: Trotter's product formula for an arbitrary pair of self-adjoint contraction semigroups. {\em Topics in Funct. Anal., Ad. Math. Suppl. Studies} Vol. {\bf 3}, 185-195 (I.Gohberg and M.Kac eds.). Acad. Press, New York 1978. \bibitem{NZ0} Neidhardt, H.; Zagrebnov V.A.: The Trotter product formula for Gibbs semigroups. {\it Comm. Math. Phys.} {\bf 131} (1990), 333-346. \bibitem{NZ1} Neidhardt, H.; Zagrebnov V.A.: On error estimates for the Trotter-Kato product formula. {\it Lett. Math. Phys.} {\bf 44} (1998), 169-186. \bibitem{NZ2} Neidhardt, H.; Zagrebnov V.A.: Fractional powers of self-adjoint operators and Trotter-Kato product formula. {\it Integr. Equ. Oper. Theory,} {\bf 35} (1999), 209-231. \bibitem{NZ3} Neidhardt, H.; Zagrebnov V.A.: Trotter-Kato product formula and operator-norm convergence. {\it Comm. Math. Phys.} {\bf 205} (1999), 129-159. \bibitem{NZ4} Neidhardt, H.; Zagrebnov V.A.: Trotter-Kato product formula and symmetrically normed ideals. {\it J. Funct. Anal.} {\bf 167} (1999), 113-167. \bibitem{RS1} Reed, M.; Simon, B.: Methods of Modern Mathematical Physics, Vol.I: Functional Analysis, Acad. Press, New York 1972. \bibitem{Ro1} Rogava, D.L.: Error bounds for Trotter-type formulas for self-adjoint operators.{\em Funct. Anal. Application} {\bf 27}, No. 3 (1993), 217-219. \bibitem{Tr1} Trotter, H.F.: On the products of semigroups of operators. {\em Proc. Am. Math. Soc.} {\bf 10} (1959), 545-551. \end{thebibliography} \end{document}