\documentclass[reqno]{amsart} \usepackage{hyperref} \AtBeginDocument{{\noindent\small \emph{Electronic Journal of Differential Equations}, Vol. 2008(2008), No. 10, pp. 1--18.\newline ISSN: 1072-6691. URL: http://ejde.math.txstate.edu or http://ejde.math.unt.edu \newline ftp ejde.math.txstate.edu (login: ftp)} \thanks{\copyright 2008 Texas State University - San Marcos.} \vspace{8mm}} \begin{document} \title[\hfilneg EJDE-2008/10\hfil Blowup and life span bounds] {Blowup and life span bounds for a reaction-diffusion equation with a time-dependent generator} \author[E. T. Kolkovska, J. A. L\'opez-Mimbela, A. P\'erez\hfil EJDE-2008/10\hfilneg] {Ekaterina T. Kolkovska, Jos\'e Alfredo L\'opez-Mimbela, Aroldo P\'erez} % in alphabetical order \address{Ekaterina T. Kolkovska \newline Centro de Investigaci\'on en Matem\'{a}ticas\\ Apartado Postal 402, 36000 Guanajuato, Mexico} \email{todorova@cimat.mx} \address{Jos\'{e} Alfredo L\'opez-Mimbela \newline Centro de Investigaci\'{o}n en Matem\'{a}ticas \\ Apartado Postal 402, 36000 Guanajuato, Mexico} \email{jalfredo@cimat.mx} \address{Aroldo P\'erez P\'erez \newline Universidad Ju\'{a}rez Aut\'onoma de Tabasco\\ Divisi\'on Acad\'emica de Ciencias B\'asicas\\ Km. 1 Carretera Cunduac\'an-Jalpa de M\'endez\\ C.P. 86690 A.P. 24, Cunduac\'an, Tabasco, Mexico} \email{aroldo.perez@dacb.ujat.mx} \thanks{Submitted August 24, 2007. Published January 21, 2008.} \subjclass[2000]{60H30, 35K55, 35K57, 35B35} \keywords{Semilinear evolution equations; Feynman-Kac representation; \hfill\break\indent critical exponent; finite time blowup; nonglobal solution; life span} \begin{abstract} We consider the nonlinear equation $$ \frac{\partial}{\partial t} u (t) = k (t) \Delta _{\alpha }u (t) + u^{1+\beta } (t),\quad u(0,x)=\lambda \varphi (x),\; x\in \mathbb{R} ^{d}, $$ where $\Delta _{\alpha }:=-(-\Delta)^{\alpha /2}$ denotes the fractional power of the Laplacian; $0<\alpha \leq 2$, $\lambda$, $\beta >0$ are constants; $ \varphi$ is bounded, continuous, nonnegative function that does not vanish identically; and $k$ is a locally integrable function. We prove that any combination of positive parameters $d,\alpha,\rho,\beta$, obeying $0\frac{\alpha}{\rho\beta}$ implies existence of non-trivial global solutions of \eqref{cauchyec} for all sufficiently small initial values, and that, under the additional assumption $\beta\in\{1,2,\dots,\}$, the condition $d<\frac{\alpha}{\rho\beta}$ yields finite time blowup of any positive solution. Moreover, the case $ \rho=0$, which under condition (\ref{condK}) corresponds to an integrable $ k $, yields finite time blow up of \eqref{cauchyec} for any non-trivial initial value, regardless of the spatial dimension and the stability exponent $\alpha$. Here we consider the case $d<\frac{\alpha}{\rho\beta}$ with $\beta\in(0,\infty)$, and focus on the asymptotic behavior of the life span % $T_{\lambda\varphi}$ of \eqref{cauchyec} when the initial value is of the form $\lambda \varphi $, where $\lambda >0$ is a parameter. The life span asymptotics of semilinear parabolic Cauchy problems give insight on how the ``size" of the initial value affects the blowup time of their positive solutions; see \cite{GuiWa, Huang,Kobayashi,LiNi, MiYa, Pin2} and the references therein. Given two functions $f,g:[0,\infty)\to[0,\infty)$, let us say that $f\sim g$ near $c\in\{0,\infty\}$ if there exist two positive constants $C_1$, $C_2$ such that $C_1f(r)\ge g(r)\ge C_2 f(r)$ for all $r$ which are sufficiently close to $c$. In \cite{LiNi} it was proved, initially for $k(t)\equiv1$ and $\alpha=2$, that $T_{\lambda\varphi}\sim \lambda^{-\beta}$ near $\infty$ provided $\varphi\ge0$ is bounded, continuous and does not vanish identically. Later on, Gui and Wang \cite{GuiWa} showed that, in fact, $\lim_{\lambda\to\infty}T_{\lambda\varphi}\cdot\lambda^{\beta}= \beta^{-1}\|\varphi\|^{-\beta}_ {L^{\infty}(\mathbb{R}^d)}$. The behavior of $T_{\lambda \varphi}$ as $\lambda$ approaches $0$ was also investigated by Lee and Ni in \cite{LiNi}. One of their results addresses the case of initial values $\varphi\ge0$ obeying growth conditions of the form $0<\liminf_{|x|\to\infty}|x|^{a}\varphi(x)$ and $ \limsup_{|x|\to\infty}|x|^{a}\varphi(x)<\infty$, where $a>0$ is a given constant different from $d$. They proved that, in this case, \begin{equation}\label{growth} T_{\lambda\varphi} \sim (1/\lambda)^{(1/\beta\:-\:\frac{1}{2}\min\{a,d\})^{-1}} \quad\text{as } \lambda\to0. \end{equation} In the present paper we obtain upper and lower bounds for the life span $T_{\lambda \varphi }$ of \eqref{cauchyec}, and provide in this way a description of the behavior of $T_{\lambda \varphi }$ as $\lambda \to \infty $ and $\lambda \to 0$. Here is a brief outline. First we prove that any combination of positive parameters $d,\alpha,\rho,\beta$, obeying $00$, namely \begin{equation} \label{***} T_{\lambda\varphi} \leq \Big(C\lambda ^{-\beta }+[ (10 {\varepsilon _{2}}/{\varepsilon _{1}})^{1/\rho }\theta ] ^{\frac{\alpha -d\rho \beta }{\alpha }}\Big)^{\frac{\alpha }{ \alpha -d\rho \beta }}+\eta , \end{equation} where $C,\theta$ and $\eta $ are suitable positive constants. We remark that many of our arguments rely on the assumption $d\rho\beta/\alpha<1$. Thus, the blowup behavior and life span asymptotics of \eqref{cauchyec} in the ``critical'' case $d\rho\beta/\alpha=1$ remain to be investigated. As this paper is partly aimed at the multidisciplinary reader, in the next section we recall some basic facts regarding the Feynman-Kac formula. In Section \ref{bridgebounds} we obtain semigroup and bridge estimates that we shall need in the sequel. Section \ref{seccion_explosion} is devoted to prove that \eqref{cauchyec} does not admit nontrivial global solutions if $d<\frac{\alpha}{\rho\beta}$. In the remaining sections \ref{upper_estimate} and \ref{seccion6} we prove our bounds for the life span of \eqref{cauchyec}. \section{The Feynman-Kac representation and subsolutions} For any $T>0$ let us consider the initial-value problem \begin{equation} \label{1.4} \begin{gathered} \frac{\partial \varrho(t,x)}{\partial t} =k(t)\Delta _{\alpha}\varrho(t,x)+\zeta (t,x)\varrho(t,x),\quad 00$ is a constant, and $k$, $\varphi $ are as in (\ref {1.4}). From Theorem \ref{F-K} we know that \begin{equation*} w(t,y)=E_{y}\Big[ \varphi (W(t))\exp \Big(\int_{0}^{t}w^{\beta }( t-s,W(s))\,ds\Big)\Big] ,\quad (t,y)\in [ 0,T]\times \mathbb{R}^{d}, \end{equation*} for every positive $T< T_{\varphi}$. Hence, for every $y\in \mathbb{R}^{d}$, \begin{equation*} w(t,y)\,\geq \,E_{y}[ \varphi (W(t))] =:v_{0}(t,y),\quad t\geq 0, \end{equation*} so that $v_{0}$ is a subsolution of \eqref{ee}; i.e., $w( 0,\cdot)=v_{0}(0,\cdot )$ and $w(t,\cdot ) \geq v_{0}(t,\cdot )$ for every $t>0$. The next lemma, which we will need in the following section, is a direct consequence of the Feynman-Kac representation. \begin{lemma}\label{sub} Let $k$, $\varphi $ be as in \eqref{1.4}, and let $\zeta (\cdot ,\cdot )$ be a nonnegative, bounded and continuous subsolution of \eqref{ee}. Then, any solution of \begin{equation*} \frac{\partial \varrho(t,y)}{\partial t}=k(t)\Delta _{\alpha }\varrho(t,y)+\zeta ^{\beta }(t,y)\varrho(t,y),\qquad \varrho(0,\cdot ) =\varphi , \end{equation*} remains a subsolution of \eqref{ee}. \end{lemma} \section{Bridge and semigroup bounds\label{bridgebounds}} Let us denote by $p(t,x)$, $t\geq 0$, $x\in\mathbb{R}^{d}$, the transition densities of the $d$-dimensional symmetric $ \alpha $-stable process $\{Z(t)\}_{t\geq 0}$. Recall that $p(t,\cdot)$, $t>0$, are strictly positive, radially symmetric continuous functions that satisfy the following properties. \begin{lemma}\label{Lemma1} For any $s,t>0$, and $x,y\in \mathbb{R}^{d}$, $p(t,x)$ satisfies \begin{itemize} \item[(i)] $p(ts,x)=t^{-\frac{d}{\alpha}}p(s,t^{-\frac{1}{ \alpha }}x)$, \item[(ii)] $p(t,x)\leq p(t,y)$ when $| x| \geq |y| $, \item[(iii)] $p(t,x)\geq (\frac{s}{t})^{d/\alpha}p(s,x)$ for $t\geq s$, \item[(iv)] $p( t,\frac{1}{\tau }(x-y))\geq p(t,x) p(t,y)$ if $p(t,0)\leq 1$ and $\tau \geq 2$. \end{itemize} \end{lemma} For a proof of the above lemma, see \cite[page 493]{GuKi} or \cite[pages 46 and 47]{Sug}. Let $\varphi:\mathbb{R}^d\to[0,\infty)$ be bounded and measurable, and let $k:[0,\infty)\to[0,\infty)$ be locally integrable. Notice that the transition probabilities of the Markov process $\{W(t),\, t\ge0\}$ are given by \begin{equation}\label{MICHA} P(W(t)\in dy|W(s)=x)= p(\textstyle\int_s^tk(r)\,dr,y-x)\,dy,\quad 0\le s\le t, \ x\in\mathbb{R}^d. \end{equation} We define the function \begin{equation} \label{v0} v_{0}(t,x)=E_{x}\left[ \varphi (W(t)) \right] = E_{x}\left[ \varphi (Z(K( t,0)) )\right]= \int p(K(t,0),y-x)\varphi(y)\,dy, \end{equation} where $t\ge0$, $x\in\mathbb{R}^d$, $K(t,s):=\int_{s}^{t}k(r)dr$, $0\leq s\leq t$, and write $B(r)\equiv B_r\subset\mathbb{R}^d$ for the ball of radius $r$, centered at the origin. \begin{lemma} \label{lem4} There exists a constant $c_0>0$ satisfying \begin{equation} v_{0}(t,x)\geq c_{0}K^{-d/\alpha}( t,0)1_{B_{1}}\big(K^{-1/\alpha}(t,0) x\big)\label{firstest} \end{equation} for all $x\in\mathbb{R}^d$, and all $t>0$ such that $K(t,0)\geq1$. \end{lemma} \begin{proof} From Lemma \ref{Lemma1} (i), (ii) and the radial symmetry of $p(t,\cdot )$ we have, for $K^{1/\alpha}(t,0)\geq 1$, $x\in B_{K^{1/\alpha}( t,0)}$ and $z\in \partial B_{2}$, that \begin{align*} v_{0}(t,x) &= E_{0}[ \varphi (Z(K(t,0))+x )] \\ &= E_{0}\big[ \varphi (K^{1/\alpha}(t,0) ( Z(1)+ K^{-1/\alpha}(t,0)x))\big] \\ &\geq \int_{B_{1}}\varphi (K^{1/\alpha} ( t,0)y)P\left[ Z(1)\in dy-K^{-1/\alpha }(t,0)x\right] \\ &= \int_{B_{1}}\varphi (K^{1/\alpha}( t,0)y)p(1, y-K^{-1/\alpha}(t,0)x)dy \\ &\geq p(1,z)\int_{B_{1}}\varphi (K^{1/\alpha}(t,0)y)\, dy \\ &= p(1,z)K^{-d/\alpha}(t,0) \int_{B_{K^{1/\alpha}(t,0)}}\varphi (y)dy \\ &\geq p(1,z)K^{-d/\alpha}(t,0) 1_{B_{1}}(K^{-1/\alpha}(t,0)x)\int_{B_{1}}\varphi (y)dy. \end{align*} Letting $c_{0}=p(1,z)\int_{B_{1}}\varphi (y)dy $ yields (\ref{firstest}). \end{proof} Fix $\theta >0$ such that (\ref{condK}) holds for all $t \geq \theta$ and such that $K(\theta ,0)\geq 1$. Define $ \delta _{0}=\min \{(\frac{ \varepsilon _{1}}{2\varepsilon _{2}})^{1/\rho },1-(\frac{\varepsilon _{1}}{2\varepsilon _{2}}) ^{1/\rho }\}$. \begin{lemma}\label{Lemma2} There exists $c>0$ such that for all $x$, $y\in B_{1}$ and $t\ge \theta/\delta_0$, \begin{equation*} P_{x}\big[ W(s)\in B_{K^{1/\alpha}(t-s,0)}: W(t)=y \big] \geq c \end{equation*} for $s\in [ \theta ,\delta _{0}t] $. \end{lemma} \begin{proof} Using (\ref{condK}) and Lemma \ref{Lemma1} (i), we obtain \begin{equation}\label{ec1est} \begin{aligned} &P_{x}\big[ W(s)\in B_{K^{1/\alpha}(t-s,0)}:W(t)=y \big] \\ &=\int_{B_{K^{1/\alpha}(t-s,0)}}\frac{p(K(s,0),x-z)p(K(t,s) ,z-y)}{p(K(t,0),x-y)}\,dz \\ &\geq \int_{B_{rs^{\rho/\alpha}}}\frac{K^{-d/\alpha }(s,0) } {K^{- d/\alpha}(t,0) }\\ &\quad\times \frac{p(1,K^{-1/\alpha}(s,0)(x-z))K^{-d/\alpha}( t,s)p( 1,K^{-1/\alpha}(t,s)(z-y))} {p(1,K^{-1/\alpha}(t,0)(x-y))} \,dz, \end{aligned} \end{equation} with $r=\varepsilon _{1}^{1/\alpha}(\frac{1}{\delta _{0}}-1)^{\rho/\alpha}$. It is straightforward to verify that $K^{-1/\alpha}(s,0)(x-z)\in B_{r_{1}}$, where $r_{1}=2(\frac{ \varepsilon_{2}}{\varepsilon _{1}})^{1/\alpha} \delta ^{ \frac{\rho }{\alpha }}$ with $\delta =\max \{1,\frac{1}{\delta _{0}}-1\}$, hence Lemma \ref{Lemma1} (ii) and radial symmetry of $p(t,\cdot )$ imply \begin{equation*} p(1,K^{-1/\alpha}(s,0)(x-z))\geq p(1,\varsigma )\equiv c_{1} \end{equation*} for any $\varsigma \in \partial B_{r_{1}}$. Thus, the term in the right-hand side of (\ref{ec1est}) is bounded from below by \begin{equation*} \int_{B_{rs^{\rho/\alpha}}}\frac{c_{1}K^{-d/\alpha}(s,0)K^{-d/\alpha}(t,s)p(1,K^{- \frac{1}{\alpha }}(t,s)(z-y))} {K^{-d/\alpha}(t,0)p(1,K^{-1/\alpha}( t,0)(x-y))}\,dz. \end{equation*} Using (\ref{condK}), and the facts that $K(t,0)\geq K(t,s)$ and $p(t,x)\leq p( t,0)$ for all $t>0$ and $x\in \mathbb{R}^{d}$, it follows that \begin{equation} P_{x}\big[ W(s)\in B_{K^{1/\alpha}( t-s,0)}:W(t) =y \big] \geq \int_{B_{rs^{\rho/\alpha}}}c_{2}s^{-d\rho/\alpha}p(1,K^{-1/\alpha} (t,s)( z-y))dz, \label{ec2est} \end{equation} where $c_{2}=\frac{c_{1}\varepsilon _{2}^{-d/\alpha}}{p(1,0)}$. Since $\theta \leq s\leq \delta _{0}t$, we have from (\ref{condK}) and the definition of $\delta _{0}$ that \begin{equation*} K^{-1/\alpha}(t,s)=[ K(t,0) -K(s,0)] ^{-1/\alpha}\leq ( \varepsilon _{1}t^{\rho }-\varepsilon _{2}\delta _{0}^{\rho }t^{\rho })^{-1/\alpha} \leq c_{3}t^{-\rho/\alpha}, \end{equation*} where $c_{3}=(\frac{2}{\varepsilon _{1}}) ^{1/\alpha}$. Since $y\in B_{1}$, $z\in B_{rs^{\rho/\alpha}}$ and $\theta \leq s\leq \delta _{0}t$, we deduce that, for $t\geq 1$, $y\in B_{t^{\rho/\alpha}}$ and $z\in B_{r\delta _{0}^{\rho/\alpha}t^{\rho/\alpha}}$. Letting $\gamma =\max \{1,\;r\delta _{0}^{\rho/\alpha}\}$, it follows that $z-y\in B_{2\gamma t^{\rho/\alpha}}$, and thus $K^{-1/\alpha}(t,s)(z-y)\in B_{2\gamma c_{3}}$. Therefore, \begin{equation*} p(1,K^{-1/\alpha}(t,s)( z-y))\geq p(1,\varsigma )\equiv c_{4} \end{equation*} for any $\varsigma \in \partial B_{2\gamma c_{3}}$. From (\ref{ec2est}) we conclude that \begin{equation*} P_{x}\big[ W(s)\in B_{K^{1/\alpha}( t-s,0)\text{ }}:W(t) =y \big] \geq \int_{B_{rs^{\rho/\alpha}}}c_{5}s^{-d\rho /\alpha}dz\equiv c. \end{equation*} \end{proof} \section{Nonexistence of positive global solutions \label{seccion_explosion}} In this section we shall use the Feynman-Kac representation to construct a subsolution of \eqref{cauchyec} which grows to infinity uniformly on the unit ball. As we are going to prove afterward, this guarantees nonexistence of nontrivial positive solutions of \eqref{cauchyec}. Let $v$ solve the semilinear nonautonomous equation \begin{equation} \label{ec1iter} \begin{gathered} \frac{\partial v(t,x)}{\partial t} =k(t)\Delta _{\alpha }v(t,x)+v_{0}^{\beta }(t,x)v(t,x), \\ v(0,x)=\varphi (x),\quad x\in \mathbb{R}^{d}, \end{gathered} \end{equation} where $k$ and $\varphi$ are as in \eqref{cauchyec}, and $v_0$ is defined in (\ref{v0}). Since $v_{0}\leq u$, where $u$ is the solution of (\ref{cauchyec}), it follows from Lemma \ref{sub} that $v\leq u$ as well. Without loss of generality we shall assume that $\varphi$ does not a.e. vanish on the unit ball. \begin{proposition}\label{Prop3} There exist $c^{\prime }$, $c''>0$ such that, for all $x\in B_{1}$ and all $t>0$ large enough, \begin{equation*} v(t,x)\geq c^{\prime }t^{-d\rho/\alpha}\exp (c''t^{1-\frac{d\rho \beta }{\alpha }}). \end{equation*} \end{proposition} \begin{proof} In the sequel, $c_{0},c_{1},\dots,c_8 $\ denote suitable positive constants, $c_0,\dots,c_5$ being defined in Lemma \ref{Lemma2}. From Theorem \ref{F-K} we know that \begin{equation*} v(t,x)=\int_{\mathbb{R}^{d}}\varphi ( y)p(K(t,0),x-y) E_{x}\Big[\exp \int_{0}^{t}v_{0}^{\beta }(t-s,W(s))ds \:\Big|\:W(t)=y \Big] dy. \end{equation*} Let $\theta $ and $\delta _{0}$ be as in Lemma \ref{Lemma2}. For any $\theta \leq s\leq \delta _{0}t$, we have $t-s\geq t-\delta _{0}t=(1-\delta _{0})t\geq \delta _{0}t\geq \theta $, and therefore $K^{\frac{1}{ \alpha }}(t-s,0)\geq 1$. From here, using (\ref{firstest}) and Jensen's inequality, we get \begin{align*} v(t,x) &\geq \int_{\mathbb{R}^{d}}\varphi (y)p(K(t,0),x-y)\\ &\quad\times E_{x}\Big[ \exp \int_{\theta }^{\delta _{0}t}c_{0}^{\beta }K^{-d\beta/\alpha}(t-s,0)1_{B_{K^{ \frac{1}{\alpha }}(t-s,0)}}(W(s))\,ds\:\Big|\: W(t)=y \Big] dy \\ &\geq \int_{B_{1}}\varphi (y)p(K(t,0),x-y)\\ &\quad\times \exp \Big\{\int_{\theta }^{\delta _{0}t}c_{0}^{\beta }K^{-\frac{ d\beta }{\alpha }}(t-s,0)P_{x}\Big[ W(s)\in B_{K^{ \frac{1}{\alpha }}(t-s,0)\text{ }}\:\Big|\:W(t)=y\Big] \,ds\Big\}\,dy. \end{align*} It follows from Lemma \ref{Lemma1} and Lemma \ref{Lemma2} that \begin{align*} v(t,x)&\geq \int_{B_{1}}\varphi (y)p(K(t,0),x-y)\exp \int_{\theta }^{\delta _{0}t}c_{6}K^{-d\beta/\alpha}(t-s,0)\,ds\,dy \\ &=\int_{B_{1}}\varphi (y) K^{-d/\alpha}(t,0)p( 1,K^{-1/\alpha}(t,0)(x-y))dy\\ &\quad\times \exp \int_{\theta }^{\delta_{0}t}c_{6} K^{- d\beta/\alpha}(t-s,0)\,ds. \end{align*} Let $x$, $y\in B_{1}$. Then $K^{-1/\alpha}( t,0)(x-y)\in B_{2}$. Radial symmetry of $p(t,\cdot )$ implies \begin{equation*} p\big(1,K^{-1/\alpha}(t,0)(x-y)\big)\geq p(1,\varsigma )\equiv c_{7} \end{equation*} for any $\varsigma \in \partial B_{2}$. Therefore, \begin{equation} v(t,x)\geq \int_{B_{1}}c_{7}\varphi ( y)K^{-d/\alpha}(t,0)dy\exp \int_{\theta }^{\delta _{0}t}c_{6}K^{-d\beta/\alpha}(t-s,0)ds. \label{ec3est} \end{equation} Let $c_{8}=c_{7}\int_{B_{1}}\varphi (y)dy$. Using (\ref {condK}) and the fact that $K(t,0)\geq K( t-s,0)$, the term in the right of (\ref{ec3est}) is bounded below by \begin{align*} &c_{8}K^{-d/\alpha}(t,0)\exp \Big( c_{6}\int_{\theta }^{\delta _{0}t}K^{-\frac{d\beta }{\alpha } }(t,0)ds\Big)\\ &\geq c_{8}\varepsilon_{2}^{-d/\alpha} t^{-d\rho/\alpha} \exp \big[c_{6}\varepsilon _{2}^{-d\beta/\alpha}(\delta _{0}t^{1-\frac{d\beta \rho }{\alpha }}-\theta t^{- \frac{d\beta \rho }{\alpha }})\big] \end{align*} if $t>0$ is large. It follows that \begin{equation} \label{cprima} v(t,x)\geq c^{\prime }t^{- \frac{d\rho }{\alpha }}\exp (c''t^{1-\frac{d\beta \rho }{\alpha }}) \end{equation} for sufficiently large $t$, where $c'=c_{8}\varepsilon _{2}^{-d/\alpha}\exp \big( -c_{6}\theta \varepsilon _{2}^{-d\beta/\alpha}\big)$ and $c''=c_{6}\delta _{0}\varepsilon_{2}^{-d\beta/\alpha}$. \end{proof} As a consequence of Proposition \ref{Prop3}, if $0<\frac{d\beta \rho }{\alpha }<1$, then $\inf_{x\in B_{1}}v(t,x)\to \infty $ when $t\to \infty $. As $v$ is subsolution of Equation \eqref{cauchyec}, this implies that \begin{equation} \label{Mathias} C(t):=\inf_{x\in B_{1}}u(t,x) \to \infty \quad \text{as }t\to \infty . \end{equation} We need the following lemma. \begin{lemma}\label{lemma_nuevo} Let $$ \xi_1:=\min_{x\in B_1}\min_{0\le r\le1}P_x[Z(r)\in B_1], $$ where $\{Z(t),\, t\ge0\}$ denotes the symmetric $\alpha$-stable process. Then \begin{itemize} \item[(i)] $\xi_1>0$ \item[(ii)] For any $0\le s\le t$, $$ \xi:=\int_{B_1}p(K(t_0+t,t_0),y-x)\,dy\ge\int_{B_1}p(K(t+t_0,s+t_0),y-x)\,dv \ge \xi_1^{\lfloor K(t+t_0,s+t_0)\rfloor}, $$ where $\lfloor x\rfloor$ denotes the least integer no smaller than $x\in [0,\infty)$. \end{itemize} \end{lemma} \begin{proof} Let $\epsilon>0$, and let $f_{\epsilon}:\mathbb{R}^d\to\mathbb{R}_+$ be a continuous function bounded by 1 such that $f_{\epsilon}=1$ on $B_1$, and $f_{\epsilon}=0$ on the complement of the ball $B_{1+\epsilon}$. Because of strong continuity of the semigroup $\{S(t)\}_{t\ge0}$ corresponding to $\{Z(t)$, $t\ge0\}$, $\lim_{t\to0}S(t)f_{\epsilon}=f_{\epsilon}$ uniformly on $\mathbb{R}^d$. Moreover, uniformly on $B_1$, $$ 1_{B_1} = \lim_{\epsilon\downarrow0}f_{\epsilon} = \lim_{\epsilon\downarrow0}\lim_{t\to0} S(t)f_{\epsilon} = \lim_{t\to0}\lim_{\epsilon\downarrow0} S(t)f_{\epsilon} = \lim_{t\to0}S(t)1_{B_1}, $$ where the last equality follows from the bounded convergence theorem. Hence, there exists $\epsilon\equiv\epsilon_{1/2}>0$ such that $\sup_{x\in B_1}|S(t)1_{B_1}(x)-1_{B_1}(x)| < \frac{1}{2}$ for all $t<\epsilon $, which implies $$ P_x[Z(r)\in B_1] = S(r)1_{B_1}(x)\ge \frac{1}{2} $$ for all $r<\epsilon$ and $x\in B_1$. Therefore, in order to prove (i) it suffices to show that \begin{equation}\label{xi} \inf_{\epsilon \le r\le 1}P_x[Z(r)\in B_1]>\mathrm{A} \end{equation} for all $x\in B_1$, where the constant $\mathrm{A}>0$ does not depend on $x$. Using Lemma \ref{Lemma1} (i) we get \begin{align*} P_x[Z(r)\in B_1]&=\int_{B_1}p(r,z-x)\,dz\\ &= r^{-d/\alpha}\int_{B_1}p(1,r^{-1/\alpha}(z-x))\,dz\\ &\ge \inf_{y\in B_{2\epsilon^{-1/\alpha}}}p(1,y)\int_{B_1}dz, \end{align*} which yields (\ref{xi}). The assertion in part (ii) is deduced from the Chapman-Kolmogorov equation as follows. If $l:=K(t_0+t, t_0+s)\le1$, the statement follows directly from part (i). If $l>1$, then \begin{align*} P_x[Z(l)\in B_1] & = \int_{B_1}\int_{\mathbb{R}^d}p(l-1,y-x)p(1,z-y)\,dy\,dz \\ &\ge \int_{B_1}\int_{B_1}p(l-1,y-x)p(1,z-y)\,dy\,dz\\ &= S(l-1)((S(1)1_{B_1})1_{B_1})\\ &\ge \xi_1 P_x[Z(l-1)\in B_1]. \end{align*} Thus, applying the above procedure $\lfloor l-1\rfloor$ times we obtain the assertion. \end{proof} Now we are ready to prove that (\ref{Mathias}) is sufficient to guarantee finite-time blow up of \eqref{cauchyec}. \begin{theorem}\label{Theo4} If $0<\frac{d\rho \beta }{\alpha}<1$, then all nontrivial positive solutions of \eqref{cauchyec} are nonglobal. \end{theorem} \begin{proof} Let $u$ be the solution of \eqref{cauchyec}, and let $t_{0}>0$ be such that $\Vert u(t_{0},\cdot )\Vert _{\infty}<\infty $. Then \begin{align*} u(t+t_{0},x)&=\int_{\mathbb{R}^{d}}p(K(t+t_{0},t_{0}),y-x)u(t_{0},y)dy \\ &\quad+\int_{0}^{t}\int_{\mathbb{R}^{d}}p( K(t+t_{0},s+t_{0}),y-x)u^{1+\beta }(s+t_{0},y)\,dy\,ds \\ &\geq \int_{B_{1}}p(K(t+t_{0},t_{0}),y-x)u(t_{0},y)dy \\ &\quad +\int_{0}^{t}\int_{B_{1}}p(K(t+t_{0},s+t_{0}),y-x)u^{1+\beta }( s+t_{0},y)\,dy\,ds. \end{align*} Therefore, $w(t,\cdot ):=u(t_{0}+t,\cdot ) $ satisfies \begin{align*} w(t,x)&\geq C(t_{0}) \int_{B_{1}}p(K(t+t_{0},t_{0}),y-x)\, dy\\ &\quad +\int_{0}^{t}\int_{B_{1}}p(K( t+t_{0},s+t_{0}),y-x)\big(\min_{z\in B_{1}}w(s,z)\big)^{1+\beta }\,dy\,ds. \end{align*} It follows from Lemma \ref{lemma_nuevo} that for all $t\in [ 0,1] $, \begin{equation*} \min_{x\in B_{1}}w(t,x)\geq \xi C(t_{0}) +\xi \int_{0}^{t}\big(\min_{z\in B_{1}}w(s,z)\big)^{1+\beta }ds. \end{equation*} We put $\mathrm{w}(t)\equiv \min_{z\in B_{1}}w(t,z)$, $t\geq 0$, and consider the integral equation \begin{equation*} v(t)=\xi C(t_{0})+\xi \int_{0}^{t}v^{1+\beta }(s)\,ds, \end{equation*} whose solution satisfies \begin{equation}\label{chi(t)} v^{\beta }(t)=\frac{\left[ \xi C(t_{0}) \right] ^{\beta }}{1-\beta \xi ^{1+\beta }C^{\beta }(t_{0})t}. \end{equation} Choosing $t_{0}$ large enough that the blow up time of $v$ is smaller than one, renders \begin{equation*} \mathrm{w}(1)=\min_{x\in B_{1}}w(1,x)\geq v(1)=\infty , \end{equation*} which proves blow up of $u$ at time $t_0+1$. \end{proof} \section{Upper estimates for the life span \label{upper_estimate}} In this section we obtain two upper bounds for the life span of the solution to \eqref{cauchyec} with initial value $u(0,\cdot )=\lambda \varphi (\cdot )$, where $\lambda $ is a positive parameter. We first consider the case of small and positive $\lambda$. \begin{proposition} \label{prop9} If $00$ such that for all sufficiently small $\lambda>0$, $$ T_{\lambda\varphi} \leq C_n\lambda^{-\frac{\alpha\beta}{\alpha-d\beta\rho}}. $$ \end{proposition} \begin{proof} From \eqref{cprima} and (\ref{Mathias}) it follows that $$ C(t) \geq \lambda c^{\prime }t^{-d\rho/\alpha} \exp \big(c''t^{1-\frac{d\beta \rho }{\alpha }}\big) $$ for all $t\geq \frac{\theta }{ \delta _{0}}$; here we require $t\ge \theta/\delta_0$ in order to have an interval $[\theta,\delta_0t]$ so that $K^{-1/\alpha}(t,s)\le {\rm Const.} t^{-\rho/\alpha}$ for $s\in [\theta,\delta_0t]$, and then to use (\ref{ec2est}). Recall from \eqref{chi(t)} that $v(1)=\infty$ provided $\beta \xi ^{1+\beta }C^{\beta }(t_{0})=1$. Note that $ t_{0}\leq t_{1}$, where $t_{1}$ is such that $t_{1}\geq \frac{\theta }{ \delta _{0}}$ and \[ \beta\xi^{1+\beta}\lambda^{\beta}(c')^{\beta} t_{1}^{-d\beta\rho/\alpha}\exp \big(\beta c''t_{1}^{1-\frac{d\beta \rho }{\alpha }}\big)=1. \] Choosing $\theta\ge\delta_0$, from the inequality $e^{x}\geq \frac{x^{n+1}}{(n+1)!}$ and the fact that the condition $0<\frac{d\rho \beta }{\alpha }\leq \frac{n}{n+1}$ implies $ \frac{d\rho\beta}{n\alpha}\leq 1-\frac{d\rho\beta}{\alpha}$, we have that $t_{1}\leq t_{2}$, where $t_{2}$ is such that \[ t_2\ge\frac{\theta}{\delta_0} \mbox{ \ \ and \ \ } \frac{1}{(n+1)!}\beta^{n+2} \xi ^{1+\beta }(c')^{\beta}(c'')^{n+1}\lambda ^{\beta }t_{2}^{1-\frac{d\beta \rho }{\alpha }}=1, \] which is the same as \[ t_{2}=\big[\frac{(n+1)!}{\beta^{n+2}\xi^{1+\beta}(c')^{\beta} (c'')^{n+1}}\big]^{\frac{\alpha}{\alpha-d\beta\rho}} \lambda^{-\frac{\alpha\beta}{\alpha-d\beta\rho}}. \] Choosing \[ C_n=\big[ \frac{(n+1)!}{\beta^{n+2} \xi^{1+\beta}(c')^{\beta}(c'')^{n+1}}\big]^{\frac{\alpha}{\alpha-d\beta\rho}} \] renders $ t_{0}\leq t_{1}\leq t_{2}=C_n\lambda ^{-\frac{\alpha \beta }{\alpha -d\beta \rho }}. $ Hence $ T_{\lambda\varphi} \leq C_n\lambda ^{-\frac{\alpha \beta }{\alpha -d\beta \rho }} $ for all $\lambda>0$ such that $C_n\lambda ^{-\frac{\alpha \beta }{\alpha -d\beta \rho }} \ge \frac{\theta}{\delta_0}$. \end{proof} Let us define \begin{equation*} \upsilon (t)=\int_{\mathbb{R}^{d}}p( K(t,0),x)u(t,x)dx, \end{equation*} where $u$ is the solution of \eqref{cauchyec}, and let $\theta >0$ be such that (\ref{condK}) holds for all $t\geq \theta $. \begin{lemma}\label{Lemma5} If there exist $\tau_{0}\geq \theta $ such that $\upsilon (t)=\infty $ for $t\geq \tau_{0}$, then the solution to \eqref{cauchyec} blows up in finite time. \end{lemma} \begin{proof} Since $p(K(t,0),0)=K(t,0)^{-d/\alpha}p(1,0)$ and $K(t,0)\ge \varepsilon_1 t^{\rho}$ for all $t\ge\theta$, we can assume, by taking $\tau_0$ bigger if necessary, that $p(K(t,0),0)\leq 1$ for all $t\geq \tau_0$. If $\tau_{0}\leq \varepsilon _{1}^{1/\rho }t$ and $\varepsilon _{1}^{1/\rho }t\leq r\leq (2\varepsilon _{1})^{1/\rho }t$, we have, from the conditions on $k(t)$, that \begin{align*} \tau &\equiv \big[ \frac{K((10\varepsilon _{2})^{1/\rho}t,r)}{K(r,0)}\big] ^{1/\alpha} =\big[\frac{K((10\varepsilon _{2})^{1/\rho }t,0)-K(r,0)}{K(r,0)} \big] ^{1/\alpha} \\ &\geq \big[ \frac{K((10\varepsilon _{2})^{1/\rho }t,0)}{K((2\varepsilon _{1}) ^{1/\rho }t,0)}-1\big] ^{1/\alpha} \\ &\geq \big[ \frac{\varepsilon _{1}(10\varepsilon _{2}) t^{\rho}}{\varepsilon _{2}(2\varepsilon _{1})t^{\rho }}-1\big] ^{1/\alpha}=4^{1/\alpha}\geq 2. \end{align*} Using properties (i) and (iv) in Lemma \ref{Lemma1}, with $\tau=\big[ \frac{ K((10\varepsilon _{2})^{1/\rho }t,r)}{ K(r,0)}\big] ^{1/\alpha}$, yields \begin{align*} &p\Big(K\big((10\varepsilon _{2})^{1/\rho }t,r\big),x-y\Big) \\ &= p\Big(K(r,0)\big[ \frac{K(( 10\varepsilon _{2})^{1/\rho }t,r)}{K(r,0)}\big],x-y\Big)\\ &=\big[ \frac{K(r,0)}{K((10\varepsilon _{2})^{1/\rho }t,r)}\big] ^{d/\alpha}p(K(r,0),\big[ \frac{K(r,0) }{K((10\varepsilon _{2})^{1/\rho }t,r)}\big] ^{1/\alpha}(x-y))\\ &\geq \big[ \frac{K(r,0)}{K(( 10\varepsilon _{2})^{1/\rho }t,r)}\big] ^{d/\alpha}p(K(r,0),x)p(K(r,0) ,y). \end{align*} Since $\upsilon (t)=\infty $ for all $t\geq \tau_{0}$, it follows that \begin{align*} &\int_{\mathbb{R}^{d}}p\big(K((10\varepsilon _{2})^{1/\rho }t,r),x-y\big) u( r,y)dy\\ &\geq \big[ \frac{K(r,0)}{K(( 10\varepsilon _{2})^{1/\rho }t,r)}\big] ^{d/\alpha}p(K(r,0),x)\upsilon (r) =\infty . \end{align*} The solution $u(t,x)$ of \eqref{cauchyec} satisfies \begin{align*} u(t,x) &=\lambda \int_{\mathbb{R}^{d}}p(K(t,0),x-y)\varphi (y)dy +\int_{0}^{t}\Big(\int_{\mathbb{R}^{d}}p(K(t,r),x-y)u^{1+\beta }(r,y)dy\Big)dr \\ &\geq \int_{0}^{t}\Big(\int_{\mathbb{R}^{d}}p(K(t,r),x-y)u^{1+\beta }(r,y)dy \Big)dr. \end{align*} Thus, \begin{equation*} u\big((10\varepsilon _{2})^{1/\rho }t,x\big) \geq \int_{0}^{(10\varepsilon _{2})^{1/\rho}t} \Big(\int_{\mathbb{R}^{d}}p(K(( 10\varepsilon _{2})^{1/\rho }t,r),x-y) u^{1+\beta }(r,y)dy\Big)\, dr. \end{equation*} Jensen's inequality renders \begin{equation*} u\big((10\varepsilon _{2})^{1/\rho }t,x\big) \geq \int_{\varepsilon _{1}^{1/\rho }t}^{(2\varepsilon _{1})^{1/\rho }t} \Big(\int_{\mathbb{R}^{d}}p(K((10\varepsilon _{2})^{1/\rho }t,r),x-y)u(r,y)dy\Big)^{1+\beta }dr=\infty , \end{equation*} so that $u(t,x)=\infty $ for any $t\geq (10\frac{ \varepsilon_{2}}{\varepsilon _{1}})^{1/\rho }\tau_{0}$ and $ x\in \mathbb{R}^{d}$. \end{proof} \begin{proposition}\label{Prop6} Let $0<\frac{d\rho \beta }{\alpha}<1$. There exists a constant $C>0$ depending on $\alpha $, $\beta $, $d$, $\varepsilon _{1}$, $\varepsilon _{2}$, $\theta $, $\rho $ and $\varphi $, such that \begin{equation} T_{\lambda\varphi} \leq \big\{C\lambda ^{-\beta }+[ (10 \frac{\varepsilon _{2}}{\varepsilon _{1}})^{1/\rho }\theta ] ^{\frac{\alpha -d\rho \beta }{\alpha }}\big\} ^{\frac{\alpha }{ \alpha -d\rho \beta }}+\eta ,\quad \lambda >0, \label{upperest} \end{equation} where $\eta $ is any positive real number satisfying $p( K(\eta,0),0)\leq 1$. \end{proposition} \begin{proof} From Lemma \ref{Lemma1} we obtain \begin{align*} p(K(\eta ,0),x-y) &=p\big(K(\eta,0),\frac{1}{2}(2x-2y)\big)\\ &\geq p\big(K(\eta ,0),2x\big)p(K(\eta ,0),2y)\\ &= 2^{-d}p(2^{-\alpha }K(\eta ,0),x)p(K(\eta ,0),2y). \end{align*} Therefore \begin{align*} u(\eta ,x)&\geq \lambda\int_{\mathbb{R}^{d}}p( K(\eta ,0),x-y)\varphi (y)dy \\ &\geq 2^{-d}\lambda p(2^{-\alpha }K(\eta ,0) ,x)\int_{\mathbb{R}^{d}}p(K(\eta,0),2y) \varphi (y)dy \\ &=\lambda N_{0}p(2^{-\alpha }K(\eta ,0) ,x), \end{align*} where $N_{0}=2^{-d}\int_{\mathbb{R}^{d}}p(K( \eta ,0),2y)\varphi (y)dy$. Thus, for any $\lambda >0$ , $t\geq 0$ and $x\in \mathbb{R}^{d}$, \begin{align*} u(t+\eta ,x) &=\int_{\mathbb{R}^{d}}p(K(t+\eta,\eta),x-y)u(\eta ,y)\,dy \\ &\quad+\int_{0}^{t}\Big(\int_{\mathbb{R}^{d}}p(K(t +\eta,r +\eta),x-y)u^{1+\beta }(r+\eta ,y)\,dy\Big)\,dr \\ &\geq \lambda N_{0}\int_{\mathbb{R}^{d}}p(K( t+\eta,\eta) ,x-y)p(2^{-\alpha }K(\eta ,0),y)dy \\ &\quad +\int_{0}^{t}\Big(\int_{\mathbb{R}^{d}}p(K(t +\eta,r +\eta),x-y)u^{1+\beta }(r+\eta ,y)\,dy\Big)\,dr \\ &\geq \lambda N_{0}p(K(t+\eta,\eta)+2^{-\alpha}K(\eta,0),x)\\ &\quad +\int_{0}^{t}\Big(\int_{\mathbb{R}^{d}}p(K(t +\eta,r +\eta),x-y)u^{1+\beta }(r+\eta ,y)\,dy\Big)\,dr \\ &\geq w(t,x), \end{align*} where $w$ solves the equation \begin{equation} \label{intqw} \begin{aligned} w(t,x)&=\lambda N_{0}p(K( t+\eta,\eta)+2^{-\alpha}K(\eta ,0),x) \\ &\quad +\int_{0}^{t}\Big(\int_{\mathbb{R}^{d}}p(K(t +\eta,r +\eta),x-y)w^{1+\beta }(r,y)dy\Big)dr, \quad t\geq 0,\; x\in \mathbb{R}^{d}. \end{aligned} \end{equation} Hence, it is sufficient to prove that $w$ is non-global, and, because of Lemma \ref{Lemma5}, it suffices to show finite time blowup of \begin{equation*} \upsilon (t)=\int_{\mathbb{R}^{d}}p( K(t,0),x)w(t,x)\,dx,\quad t\geq 0. \end{equation*} Multiplying both sides of (\ref{intqw}) by $p(K( t,0),x)$ and integrating, we obtain \begin{align*} v(t) &=\int_{\mathbb{R}^{d}}p(K(t,0),x)w(t,x)\,dx \\ &=\lambda N_{0}\int_{\mathbb{R}^{d}}p(K(t+\eta,\eta) +2^{-\alpha }K(\eta ,0),x) p(K(t,0),x)\,dx \\ &\quad +\int_{\mathbb{R}^{d}}\int_{0}^{t} \int_{\mathbb{R}^{d}}p(K(t+\eta,r+\eta),x-y)p( K(t,0) ,x)w^{1+\beta }(r,y)\,dy\,dr\,dx \\ &=\lambda N_{0}p(K(t,0)+K(t+\eta,\eta)+2^{-\alpha }K(\eta,0),0)\\ &\quad +\int_{0}^{t}\int_{\mathbb{R}^{d}}p(K(t+\eta,r+\eta)+K(t,0),y) w^{1+\beta }(r,y)\,dy\,dr,\quad t\geq 0. \end{align*} From Lemma \ref{Lemma1} (i), we have $p(t,0)\leq p(s,0)$ for all $00$, depending on $\alpha $, $\beta $, $d$, $\varepsilon _{1}$, $\rho $ and $\varphi $, such that \begin{equation} T_{\lambda\varphi} \geq c\lambda ^{-\frac{\alpha \beta }{\alpha -d\rho \beta }},\quad\lambda >0. \label{lowest} \end{equation} \end{proposition} \begin{proof} The function \begin{equation*} \overline{u}(t,x): =\Big[ \lambda ^{-\beta }-\beta \int_{0}^{t}\| U(r,0)\varphi \| _{\infty }^{\beta }dr\Big] ^{-1/\beta}U(t,0)\varphi (x),\qquad t\geq 0,\quad x\in \mathbb{R}^{d}. \end{equation*} is a supersolution of \eqref{cauchyec}. Indeed, $\overline{u}(0,\cdot)=\lambda \varphi ( \cdot)$ and \begin{align*} \frac{\partial \overline{u}(t,x)}{\partial t} &=-\frac{1}{ \beta}\Big[ \lambda ^{-\beta }-\beta \int_{0}^{t}\| U(r,0)\varphi \| _{\infty }^{\beta }dr\Big] ^{-\frac{1}{\beta }-1}\Big[ -\beta \| U(t,0)\varphi \| _{\infty }^{\beta }\Big] U(t,0)\varphi (x)\\ &\quad +\Big[ \lambda ^{-\beta }-\beta \int_{0}^{t}\| U( r,0)\varphi \| _{\infty }^{\beta }dr\Big] ^{-1/\beta} k(t)\Delta _{\alpha }U(t,0) \varphi (x). \end{align*} Since $-\frac{1}{\beta }-1=-\frac{\beta +1}{\beta }$, we get \begin{align*} \frac{\partial \overline{u}(t,x)}{\partial t} &=\Big\{\Big[\lambda ^{-\beta }-\beta \int_{0}^{t}\| U(r,0) \varphi \| _{\infty }^{\beta }dr\Big] ^{-1/\beta}\Big\}^{\beta +1} \| U(t,0)\varphi \|_{\infty }^{\beta }U(t,0)\varphi (x)\\ &\quad +k(t)\Delta _{\alpha }\Big[ \lambda ^{-\beta }-\beta \int_{0}^{t}\| U(r,0)\varphi \| _{\infty }^{\beta }dr\Big] ^{-1/\beta}U(t,0)\varphi (x). \end{align*} Using the inequality \begin{equation*} \| U(t,0)\varphi \| _{\infty}^{\beta }U(t,0)\varphi (x)\geq [ U(t,0)\varphi (x)] ^{1+\beta } \end{equation*} it follows that \begin{equation*} \frac{\partial \overline{u}(t,x)}{\partial t}\geq k( t)\Delta _{\alpha }\overline{u}(t,x)+\overline{u} ^{1+\beta }(t,x), \end{equation*} showing that $\overline{u}$ is a supersolution of \eqref{cauchyec}. Writing $ \overline{L}(\lambda)$ for the life span of $\overline{u}$, it follows that \begin{equation*} \overline{L}(\lambda)\leq T_{\lambda\varphi},\quad\lambda\ge0. \end{equation*} Now, \begin{equation*} \overline{u}(t,x)=\Big[ \lambda ^{-\beta }-\beta \int_{0}^{t}\| U(r,0)\varphi \| _{\infty }^{\beta }dr\Big] ^{-1/\beta}U(t,0)\varphi (x)=\infty \end{equation*} when $\lambda ^{-\beta }=\beta \int_{0}^{t}\| U(r,0) \varphi \| _{\infty }^{\beta }\,dr$. By definition of $\overline{L}(\lambda)$, \begin{equation} \beta ^{-1}\lambda ^{-\beta }=\int_{0}^{\overline{L} (\lambda)}\| U(r,0)\varphi \| _{\infty }^{\beta}dr. \label{lowest2} \end{equation} Note that, by Lemma \ref{Lemma1} (i), (ii), \begin{align*} U(t,0)\varphi (x) &=S(K(t,0))\varphi (x)\\ &=\int_{\mathbb{R}^{d}}\varphi (y)p(K(t,0),x-y)dy \\ &\leq p(1,0)K^{-d/\alpha}(t,0)\| \varphi \| _{1},\quad t>0,\; x\in \mathbb{R}^{d}. \end{align*} Since, by assumption, (\ref{condK}) holds for any $t\geq 0$, we obtain \begin{equation*} \| U(t,0)\varphi \| _{\infty }\leq p( 1,0)(\varepsilon _{1}t^{\rho })^{-d/\alpha} \| \varphi \| _{1}. \end{equation*} Inserting this inequality in (\ref{lowest2}) and using that $0<\frac{d\rho \beta }{\alpha }<1$, we get \begin{align*} \beta ^{-1}\lambda ^{-\beta } &\leq (p(1,0)\|\varphi \| _{1})^{\beta } \varepsilon _{1}^{-d\beta/\alpha}\int_{0}^{\overline{L}(\lambda)} r^{-d\rho \beta/\alpha}\,dr \\ &= \frac{\alpha }{\alpha -d\rho \beta }(p(1,0) \| \varphi \| _{1})^{\beta }\varepsilon _{1}^{-d\beta/\alpha} \overline{L}(\lambda)^{\frac{\alpha -d\rho \beta }{\alpha}}, \end{align*} which gives \begin{equation*} \overline{L}(\lambda)^{\frac{\alpha -d\rho \beta }{\alpha }}\geq \frac{ \alpha -d\rho \beta }{\alpha \beta }(p(1,0) \| \varphi \| _{1})^{-\beta }\varepsilon _{1}^{\frac{d\beta }{ \alpha }}\lambda ^{-\beta }. \end{equation*} In this way we obtain the inequality \begin{equation*} T_{\lambda\varphi} \geq \Big[ \frac{\alpha -d\rho \beta }{\alpha \beta }\Big] ^{\frac{\alpha }{\alpha -d\rho \beta }}( p(1,0)\| \varphi \| _{1}) ^{-\frac{\alpha \beta }{\alpha -d\rho \beta }}\varepsilon _{1}^{\frac{d\beta }{\alpha -d\rho \beta } }\lambda ^{-\frac{\alpha \beta }{\alpha -d\rho \beta }}, \end{equation*} which proves the existence of a constant $c\equiv c(\alpha ,\beta,d,\varepsilon _{1},\rho ,\varphi )>0$ that satisfies (\ref{lowest}). \end{proof} Summarizing both, upper and lower bounds for the life span of ( \ref{cauchyec}), we get the following statement. \begin{theorem}\label{Theo8} Let $0<\frac{d\rho \beta }{\alpha}<1$, and let $T_{\lambda\varphi}$ be the life span of the nonautonomous semilinear equation \begin{gather*} \frac{\partial u(t,x)}{\partial t} =k(t)\Delta_{\alpha }u(t,x)+u^{1+\beta }(t,x)\\ u(0,x)= \lambda \varphi (x)\ge0 ,\quad x\in \mathbb{R}^{d}, \end{gather*} where $\lambda>0$. Then \begin{equation} \lim_{\lambda \to 0}T_{\lambda\varphi} =\infty ,\quad \lim_{\lambda \to \infty }T_{\lambda\varphi}\in \big[ 0,(10\frac{\varepsilon _{2}}{\varepsilon _{1}})^{1/\rho} \theta +\eta \big] , \label{asybeha} \end{equation} where $\theta $ and $\eta $ are any positive numbers such that $\varepsilon _{1}\theta ^{\rho }\leq K(\theta ,0)\leq \varepsilon _{2}\theta ^{\rho }$ and $p(K(\eta ,0),0)\leq 1$, respectively. \end{theorem} \begin{proof} Due to (\ref{upperest}) and (\ref{lowest}), \begin{equation*} c\lambda ^{-\frac{\alpha \beta }{\alpha -d\rho \beta }} \leq T_{\lambda\varphi} \leq \big\{C\lambda ^{-\beta } +\big[ (10\frac{\varepsilon _{2}}{\varepsilon _{1}})^{1/\rho }\theta \big] ^{\frac{ \alpha -d\rho \beta }{\alpha }}\big\} ^{\frac{\alpha }{\alpha -d\rho \beta }}+\eta , \end{equation*} from which (\ref{asybeha}) follows directly using the fact that $0<\frac{ d\rho \beta }{\alpha }<1$. \end{proof} \subsection*{Acknowledgements} The authors are grateful to the anonymous referees for their careful reading of the original manuscript, and for their useful suggestions. \begin{thebibliography}{00} \bibitem{Applebaum} Applebaum, D. L\'{e}vy; processes and stochastic calculus. \emph{Cambridge Studies in Advanced Mathematics} \textbf{93}. Cambridge University Press, Cambridge, 2004. \bibitem{Bardos} Bardos, C.; Penel, P.; Frisch, U.; Sulem, P.-L.; Modified Dissipativity for a Nonlinear Evolution Equation Arising in Turbulence, \emph{Arch. Rational Mech. Anal.} Vol. 71 (1979), no. 3, 237--256. \bibitem{BLMW} Birkner, M., L\'{o}pez-Mimbela, J. A. and Wakolbinger, A.; Blow-up of semilinear PDE's at the critical dimension. A probabilistic approach. \textit{Proc. Amer. Math. Soc.} \textbf{130 }(2002), 2431-2442. \bibitem{B-LM-W2} Birkner, M., L\'{o}pez-Mimbela, J. A. and Wakolbinger, A.; Comparison results and steady states for the Fujita equation with fractional Laplacian. \emph{Ann. Inst. Henri Poincar\'{e}, Anal. Non Lin\'{e}aire} \textbf{22} (2005), no.1, 83-97. \bibitem{Freidlin} Freidlin, M.; Functional integration and partial differential equations. \emph{Annals of Mathematics Studies} \textbf{109}. Princeton University Press, Princeton, NJ, 1985. \bibitem{GuKi} Guedda, M. and Kirane, M.; A note on nonexistence of global solutions to a nonlinear integral equation. \textit{Bull. Belg. Math. Soc. }\textbf{6 }(1999), 491-497. \bibitem{GuiWa} Gui, C. and Wang, X.; Life span of solutions of the Cauchy problem for a semilinear heat equation. \textit{J. Diff. Equations} \textbf{115} (1995), 166--172. \bibitem{Huang} Huang, Q., Mochizuki, K. and Mukai, K.; Life span and asymptotic behavior for a semilinear parabolic system with slowly decaying initial values. \emph{Hokkaido Math. J.\/} \textbf{27} (1998), no. 2, 393--407. \bibitem{KST} Kobayashi, K., Sirao, T. and Tanaka, H.; On the growing up problem for semilinear heat equations. \textit{J. Math. Soc. Japan} \textbf{29} (1977), 407--424. \bibitem{Kobayashi} Kobayashi, Y.; Behavior of the life span for solutions to the system of reaction-diffusion equations. \emph{Hiroshima Math. J.} \textbf{33} (2003), no. 2, 167--187. \bibitem{LiNi} Lee, Tzong-Yow and Ni, Wei-Ming; Global existence, large time behavior and life span of solutions of a semilinear parabolic Cauchy problem. \textit{Trans. Amer. Math. Soc. } \textbf{333} (1992), 365-378. \bibitem{JALM-APP} L\'{o}pez-Mimbela, J. A. and P\'{e}rez P\'{e}rez, A.; Finite time blow up and stability of a semilinear equation with a time dependent L\'{e}vy generator. \textit{Stoch. Models} Vol. 22 (2006), no. 4, pp. 735-–752. \bibitem{LM-P} L\'{o}pez-Mimbela, J. A. and Privault, N.; Blow-up and stability of semilinear PDEs with gamma generators. \emph{J. Math. Anal. Appl.} \textbf{307} (2005), no.1, 181-205. \bibitem{Mann} Mann Jr., J.A.; Woyczy\'{n}ski, W. A.; Growing Fractal Interfaces in the Presence of Self-similar Hopping Surface Diffusion, \emph{ Phys. A\/} Vol. 291(2001), 159-183. \bibitem{MiYa} Mizoguchi, N. and Yanagida, E.; Life span of\ solutions for a semilinear parabolic problem with small diffusion. \textit{J. Math. Anal. Appl. }\textbf{261}(2001), 350-368. \bibitem{Pazy} Pazy, A.; Semigroups of linear operators and applications to partial differential equations. Springer-Verlag, New York, 1983. \bibitem{Pin2} Pinsky, R. G.; The behavior of the life span for solutions to $u_{t}=\Delta u+a(x)u^{p}$ in $\mathbb{R}^{d}$. \textit{J. Diff. Equat.} \textbf{147 }(1998), 30-57. \bibitem{Sato} Sato, Ken-Iti.; L\'{e}vy processes and infinitely divisible distributions. Cambridge Studies in Advanced Mathematics, \textbf{68}. Cambridge University Press, Cambridge, 1999. \bibitem{Shlesinger} Shlesinger, M. F.; Zaslavsky, G. M.; Frisch, U. (eds); L\'{e}vy Flights and Related Topics in Physics, \emph{Lecture Notes in Physics\/} Vol. 450. Springer-Verlag, Berlin, 1995. \bibitem{Sug} Sugitani, S.; On nonexistence of global solutions for some nonlinear integral equations. \textit{Osaka J. Math. }\textbf{12} (1975), 45-51. \end{thebibliography} \end{document}