Introduction
We consider the Cauchy problem for the semilinear heat equation with a source and a separable external forcing,
{ ∂ t u − Δ u = ∣ u ∣ p + ζ ( t ) w ( x ) , ( x , t ) ∈ R N × ( 0 , ∞ ) , u ( ⋅ , 0 ) = u 0 , x ∈ R N , (1) \begin{cases}
\partial_t u-\Delta u=|u|^p+\zeta(t)w(x), & (x,t)\in\mathbb{R}^N\times(0,\infty),\\
u(\cdot,0)=u_0, & x\in\mathbb{R}^N, \tag*{(1)}
\end{cases} { ∂ t u − Δ u = ∣ u ∣ p + ζ ( t ) w ( x ) , u ( ⋅ , 0 ) = u 0 , ( x , t ) ∈ R N × ( 0 , ∞ ) , x ∈ R N , (1)
where N ≥ 1 N\ge1 N ≥ 1 , p > 1 p>1 p > 1 , and u 0 u_0 u 0 , w w w and ζ \zeta ζ are real-valued and allowed to change sign. The distinguishing feature of the present work is that the temporal profile of the forcing is integrable ,
ζ ∈ L 1 ( 0 , ∞ ) , \zeta\in L^1(0,\infty), ζ ∈ L 1 ( 0 , ∞ ) ,
the model example being
ζ ( t ) = t σ e − t , σ > − 1 , ∫ 0 ∞ ζ d t = Γ ( σ + 1 ) . (2) \zeta(t)=t^\sigma e^{-t},\qquad\sigma>-1,\qquad\int_0^\infty\zeta\,dt=\Gamma(\sigma+1). \tag*{(2)} ζ ( t ) = t σ e − t , σ > − 1 , ∫ 0 ∞ ζ d t = Γ ( σ + 1 ) . (2)
Throughout the paper, we write Q : = R N × ( 0 , ∞ ) Q:=\mathbb{R}^N\times(0,\infty) Q := R N × ( 0 , ∞ ) , p ′ = p / ( p − 1 ) p'=p/(p-1) p ′ = p / ( p − 1 ) ,
p F : = 1 + 2 N (3) p_F:=1+\frac{2}{N} \tag*{(3)} p F := 1 + N 2 (3)
for the Fujita exponent, and
Z : = ∫ 0 ∞ ζ ( t ) d t , W : = ∫ R N w ( x ) d x , m 0 : = ∫ R N u 0 ( x ) d x . Z := \int_{0}^{\infty} \zeta(t)\,dt,\qquad W := \int_{\mathbb{R}^{N}} w(x)\,dx,\qquad m_{0} := \int_{\mathbb{R}^{N}} u_{0}(x)\,dx. Z := ∫ 0 ∞ ζ ( t ) d t , W := ∫ R N w ( x ) d x , m 0 := ∫ R N u 0 ( x ) d x .
The number
M : = m 0 + Z W = ∫ R N u 0 d x + ( ∫ 0 ∞ ζ d t ) ( ∫ R N w d x ) (4) M := m_{0}+ZW = \int_{\mathbb{R}^{N}} u_{0}\,dx+\left(\int_{0}^{\infty}\zeta\,dt\right)\left(\int_{\mathbb{R}^{N}}w\,dx\right) \tag*{(4)} M := m 0 + Z W = ∫ R N u 0 d x + ( ∫ 0 ∞ ζ d t ) ( ∫ R N w d x ) (4)
will be called the effective mass of the data: it is the bare mass of the initial datum augmented by the total impulse that the forcing injects over the whole time axis. In the model case (2) one has M = m 0 + Γ ( σ + 1 ) W M=m_{0}+\Gamma(\sigma+1)W M = m 0 + Γ ( σ + 1 ) W .
Background: the Fujita phenomenon
In his seminal paper [14 ] , Fujita showed that for ∂ t u = Δ u + u p \partial_{t}u=\Delta u+u^{p} ∂ t u = Δ u + u p with nonnegative data every nontrivial solution blows up in finite time when 1 < p < p F 1<p<p_{F} 1 < p < p F , whereas global positive solutions exist for small data when p > p F p>p_{F} p > p F . The critical case p = p F p=p_{F} p = p F belongs to the blow-up range; this was proved by Hayakawa [17 ] for N = 1 , 2 N=1,2 N = 1 , 2 , by Kobayashi, Sirao and Tanaka [24 ] in all dimensions, and by different methods by Aronson and Weinberger [2 ] and Weissler [36 ] . The local theory in Lebesgue spaces goes back to Weissler [35 ] ; the eigenfunction method of Kaplan [22 ] is one of the basic tools for blow-up. We refer to the surveys of Levine [27 ] and Deng and Levine [9 ] and to the monograph of Quittner and Souplet [34 ] for the extensive literature that followed.
For the equation ∂ t u − Δ u = ∣ u ∣ p \partial_{t}u-\Delta u=|u|^{p} ∂ t u − Δ u = ∣ u ∣ p with sign-changing data, the nonnegativity of the source makes the problem amenable to the nonlinear capacity (test-function) method of Mitidieri and Pohozaev [33 ] , which works directly with weak solutions and shows nonexistence of global solutions in the Fujita range under a sign condition on ∫ u 0 \int u_{0} ∫ u 0 . Sign-changing solutions exhibit many phenomena absent in the positive case; see for instance [8 ] and [34 ] , Chapter II. The size of the lifespan of small solutions is a closely related question: for u 0 = ε a u_{0}=\varepsilon a u 0 = ε a with a ≥ 0 a\geq0 a ≥ 0 , Lee and Ni [26 ] proved that T ε ≃ ε − ( 1 / ( p − 1 ) − N / 2 ) − 1 T_{\varepsilon}\simeq\varepsilon^{-(1/(p-1)-N/2)^{-1}} T ε ≃ ε − ( 1/ ( p − 1 ) − N /2 ) − 1 for p < p F p<p_{F} p < p F and log T ε ≃ ε − ( p − 1 ) \log T_{\varepsilon}\simeq\varepsilon^{-(p-1)} log T ε ≃ ε − ( p − 1 ) for p = p F p=p_{F} p = p F , and Ikeda and Sobajima [19 ] devised a test-function method yielding sharp upper bounds for weak solutions of critical problems.
A second classical theme that will play a role below is the behavior of the mass. For the heat equation with absorption ∂ t v = Δ v − v p \partial_{t}v=\Delta v-v^{p} ∂ t v = Δ v − v p , v 0 ≥ 0 v_{0}\geq0 v 0 ≥ 0 , the mass ∫ v ( t ) \int v(t) ∫ v ( t ) converges to a positive limit when p > p F p>p_{F} p > p F but tends to zero when p ≤ p F p\leq p_{F} p ≤ p F ; see Gmira and Véron [16 ] , Kamin and Peletier [21 ] , Brezis, Peletier and Terman [7 ] , Galaktionov, Kurdyumov and Samarskii [15 ] , Escobedo, Kavian and Matano [10 ] and Herraiz [18 ] . Analogous decay-of-mass results were obtained for the fractional Laplacian by Fino and Karch [12 ] and for mixed local–nonlocal operators by Kirane, Fino and Ayoub [23 ] . One of the outcomes of the present paper is that the same threshold governs the mass of every global solution of the source problem (1) , forced or not.
Forcing terms
The influence of an external forcing on the Fujita phenomenon was first understood for stationary forcings. Bandle, Levine and Zhang [3 ] studied ∂ t u = Δ u + u p + w ( x ) \partial_{t}u=\Delta u+u^{p}+w(x) ∂ t u = Δ u + u p + w ( x ) and proved that, when ∫ w > 0 \int w>0 ∫ w > 0 , the critical exponent jumps from p F p_{F} p F to N / ( N − 2 ) N/(N-2) N / ( N − 2 ) if N ≥ 3 N\geq3 N ≥ 3 , and that there is no global solution at all if N ∈ { 1 , 2 } N\in\{1,2\} N ∈ { 1 , 2 } ; see also Zhang [37 ] . For space–time forcings t σ w ( x ) t^{\sigma}w(x) t σ w ( x ) , Jleli, Kawakami and Samet [20 ] found the critical exponent
p ∗ ( σ ) = N − 2 σ N − 2 σ − 2 ( − 1 < σ < 0 ) , p ∗ ( σ ) = ∞ ( σ > 0 ) , p^{*}(\sigma)=\frac{N-2\sigma}{N-2\sigma-2}\quad(-1<\sigma<0),\qquad p^{*}(\sigma)=\infty\quad(\sigma>0), p ∗ ( σ ) = N − 2 σ − 2 N − 2 σ ( − 1 < σ < 0 ) , p ∗ ( σ ) = ∞ ( σ > 0 ) ,
which is discontinuous at σ = 0 \sigma=0 σ = 0 when N ≥ 3 N\geq3 N ≥ 3 . These results were extended to systems by Fino, Jleli and Samet [11 ] , to inhomogeneous and Hardy–Hénon nonlinearities in [28 , 29 ] , to general time profiles a ( t ) w ( x ) a(t)w(x) a ( t ) w ( x ) controlled by the averages t − 1 ∫ 0 t a t^{-1}\int_{0}^{t}a t − 1 ∫ 0 t a in [1 ] , to nonlocal nonlinearities and mixed local–nonlocal or space–time fractional diffusions by Ben Belgacem and Majdoub [4 , 5 , 6 ] , to degenerate and singular weights in [31 ] , and to spatial profiles with regularly varying mass in [25 ] . Related questions for time–space fractional problems were studied by Fino and Kirane [13 ] , and the robustness of p F p_{F} p F under singular perturbations of the diffusion was established in [30 , 32 ] .
In all these works the forcing is non-integrable in time (the power t σ t^{\sigma} t σ with σ > − 1 \sigma>-1 σ > − 1 has ∫ 0 T t σ d t → ∞ \int_0^T t^{\sigma}\,\mathrm{d}t\to\infty ∫ 0 T t σ d t → ∞ ), and its accumulated impulse ∫ 0 T ζ \int_0^T\zeta ∫ 0 T ζ grows without bound. This is precisely what shifts the critical exponent: in the test-function balance the forcing contributes a term of size T σ + 1 T^{\sigma+1} T σ + 1 , which competes with the diffusive cost R N + 2 − 2 p ′ R^{N+2-2p'} R N + 2 − 2 p ′ at the parabolic scale T = R 2 T=R^2 T = R 2 . As σ ↓ − 1 \sigma\downarrow-1 σ ↓ − 1 one has p ∗ ( σ ) ↓ p F p^*(\sigma)\downarrow p_F p ∗ ( σ ) ↓ p F , which suggests that integrable forcings sit at an endpoint of the theory. It is natural to ask what exactly happens there.
Questions and heuristics
When ζ ∈ L 1 ( 0 , ∞ ) \zeta\in L^1(0,\infty) ζ ∈ L 1 ( 0 , ∞ ) the forcing delivers a finite total impulse Z W ZW Z W . Integrating (1) in space, any solution with integrable data satisfies the mass balance
d d t ∫ R N u ( t ) d x = ∫ R N ∣ u ( t ) ∣ p d x + ζ ( t ) W , (5) \frac{\mathrm{d}}{\mathrm{d}t}\int_{\mathbb{R}^N}u(t)\,\mathrm{d}x=\int_{\mathbb{R}^N}|u(t)|^p\,\mathrm{d}x+\zeta(t)W, \tag*{(5)} d t d ∫ R N u ( t ) d x = ∫ R N ∣ u ( t ) ∣ p d x + ζ ( t ) W , (5)
so that, formally, ∫ u ( t ) \int u(t) ∫ u ( t ) tends to M + ∬ Q ∣ u ∣ p M+\iint_Q|u|^p M + ∬ Q ∣ u ∣ p as t → ∞ t\to\infty t → ∞ . This suggests that an integrable forcing acts as a mass shift and that the relevant quantity is the effective mass (1.3). The following questions arise naturally, and the purpose of this paper is to answer them.
Is p F p_F p F still the critical exponent? Since ζ \zeta ζ is integrable, one may hope that small forcings do not destroy global existence above p F p_F p F , while the extra mass Z W ZW Z W cannot prevent blow-up below p F p_F p F when it is positive.
Which data blow up in the Fujita range? For sign-changing data, the test-function method yields nonexistence when the relevant mass is positive . What happens in the borderline case M = 0 M=0 M = 0 , and when M < 0 M<0 M < 0 ?
What is the large-time behavior of global solutions, and what is the lifespan of small blowing-up solutions?
Main results
We now state our main results; the precise notions of weak and mild solutions are given in Definitions 2.1 and 2.2. For f ∈ L 1 ( R N ) ∩ L ∞ ( R N ) f\in L^1(\mathbb{R}^N)\cap L^\infty(\mathbb{R}^N) f ∈ L 1 ( R N ) ∩ L ∞ ( R N ) we write [ [ f ] ] : = max { ∥ f ∥ L 1 , ∥ f ∥ L ∞ } [[f]]:=\max\{\|f\|_{L^1},\|f\|_{L^\infty}\} [[ f ]] := max { ∥ f ∥ L 1 , ∥ f ∥ L ∞ } , G t ( x ) = ( 4 π t ) − N / 2 e − ∣ x ∣ 2 / 4 t G_t(x)=(4\pi t)^{-N/2}e^{-|x|^2/4t} G t ( x ) = ( 4 π t ) − N /2 e − ∣ x ∣ 2 /4 t denotes the heat kernel, S ( t ) f = G t ∗ f S(t)f=G_t*f S ( t ) f = G t ∗ f , and
h N ( t ) : = min { 1 , ( 4 π t ) − N / 2 } , ( ∣ ζ ∣ ⋆ h N ) ( t ) : = ∫ 0 t ∣ ζ ( s ) ∣ h N ( t − s ) d s , (6) h_N(t):=\min\{1,(4\pi t)^{-N/2}\},\qquad(|\zeta|\star h_N)(t):=\int_0^t|\zeta(s)|h_N(t-s)\,\mathrm{d}s, \tag*{(6)} h N ( t ) := min { 1 , ( 4 π t ) − N /2 } , ( ∣ ζ ∣ ⋆ h N ) ( t ) := ∫ 0 t ∣ ζ ( s ) ∣ h N ( t − s ) d s , (6)
so that ∥ S ( t ) f ∥ L ∞ ≤ h N ( t ) [ [ f ] ] \|S(t)f\|_{L^\infty}\le h_N(t)[[f]] ∥ S ( t ) f ∥ L ∞ ≤ h N ( t ) [[ f ]] for all t > 0 t>0 t > 0 .
The Fujita range: mass annihilation. Our central result is an exact identity satisfied by every global solution in the Fujita range.
Theorem 1.1. Let 1 < p ≤ p F 1<p\le p_F 1 < p ≤ p F , ζ ∈ L 1 ( 0 , ∞ ) \zeta\in L^1(0,\infty) ζ ∈ L 1 ( 0 , ∞ ) , w ∈ L 1 ( R N ) w\in L^1(\mathbb{R}^N) w ∈ L 1 ( R N ) and u 0 ∈ L 1 ( R N ) u_0\in L^1(\mathbb{R}^N) u 0 ∈ L 1 ( R N ) . If u u u is a global weak solution of (1) , then u ∈ L p ( Q ) u\in L^p(Q) u ∈ L p ( Q ) and
∬ Q ∣ u ∣ p d x d t = − M = − ( ∫ R N u 0 d x + Z W ) . (7) \iint_Q|u|^p\,\mathrm{d}x\,\mathrm{d}t=-M=-\left(\int_{\mathbb{R}^N}u_0\,\mathrm{d}x+ZW\right). \tag*{(7)} ∬ Q ∣ u ∣ p d x d t = − M = − ( ∫ R N u 0 d x + Z W ) . (7)
The identity (7) says that the total amount of mass produced by the source over the whole time axis exactly cancels the effective mass of the data. Its first consequence is a sharp nonexistence statement, which replaces the sign condition M > 0 M>0 M > 0 by M ≥ 0 M\ge0 M ≥ 0 .
Theorem 1.2. Assume 1 < p ≤ p F 1<p\le p_F 1 < p ≤ p F , ζ ∈ L 1 ( 0 , ∞ ) \zeta\in L^1(0,\infty) ζ ∈ L 1 ( 0 , ∞ ) , w ∈ L 1 ( R N ) w\in L^1(\mathbb{R}^N) w ∈ L 1 ( R N ) and u 0 ∈ L 1 ( R N ) u_0\in L^1(\mathbb{R}^N) u 0 ∈ L 1 ( R N ) .
(i) If
M = m 0 + Z W > 0 , (8) M=m_0+ZW>0, \tag*{(8)} M = m 0 + Z W > 0 , (8) then (1) admits no global weak solution.
(ii) If M = 0 M=0 M = 0 , then (1) admits a global weak solution if and only if u 0 = 0 u_0=0 u 0 = 0 a.e. and ζ ( t ) w ( x ) = 0 \zeta(t)w(x)=0 ζ ( t ) w ( x ) = 0 for a.e. ( x , t ) ∈ Q (x,t)\in Q ( x , t ) ∈ Q ; in that case u ≡ 0 u\equiv0 u ≡ 0 is the only global weak solution.
(iii) If M < 0 M<0 M < 0 , every global weak solution satisfies ∥ u ∥ L p ( Q ) p = ∣ M ∣ \|u\|_{L^p(Q)}^p=|M| ∥ u ∥ L p ( Q ) p = ∣ M ∣ .
For the unforced equation ∂ t u − Δ u = ∣ u ∣ p \partial_tu-\Delta u=|u|^p ∂ t u − Δ u = ∣ u ∣ p , statements (i)–(ii) are in line with the nonexistence results obtained for the corresponding differential inequality by the nonlinear capacity method [33 ] : nontrivial data with zero mass, such as u 0 = a − a ( ⋅ − x 1 ) u_0=a-a(\cdot-x_1) u 0 = a − a ( ⋅ − x 1 ) , never give rise to global solutions when p ≤ p F p\le p_F p ≤ p F . What is new is the exact identity (7) , its consequences for the dynamics of global solutions, and the fact that in the forced problem the relevant quantity is the effective mass M M M and not m 0 m_0 m 0 . Part (ii) shows, for instance, that the datum u 0 = − Z w u_0=-Zw u 0 = − Z w (zero effective mass) blows up whenever ζ w ≢ 0 \zeta w\not\equiv0 ζ w ≡ 0 . The next result shows that both alternatives genuinely occur when M < 0 M<0 M < 0 .
Theorem 1.3. Let p > 1 p>1 p > 1 , ζ ∈ L 1 ( 0 , ∞ ) \zeta\in L^1(0,\infty) ζ ∈ L 1 ( 0 , ∞ ) , and let u 0 , w ∈ L 1 ( R N ) ∩ L ∞ ( R N ) u_0,w\in L^1(\mathbb{R}^N)\cap L^\infty(\mathbb{R}^N) u 0 , w ∈ L 1 ( R N ) ∩ L ∞ ( R N ) satisfy u 0 ≤ 0 u_0\le0 u 0 ≤ 0 and ζ ( t ) w ( x ) ≤ 0 \zeta(t)w(x)\le0 ζ ( t ) w ( x ) ≤ 0 for a.e. ( x , t ) ∈ Q (x,t)\in Q ( x , t ) ∈ Q . Then (1.1) has a unique global mild solution, and
− ( S ( t ) ∣ u 0 ∣ + ∫ 0 t S ( t − s ) ∣ ζ ( s ) w ∣ d s ) ≤ u ( t ) ≤ 0 for all t > 0. -\left(S(t)|u_0|+\int_0^t S(t-s)|\zeta(s)w|\,\mathrm{d}s\right)\le u(t)\le0\qquad\text{for all }t>0. − ( S ( t ) ∣ u 0 ∣ + ∫ 0 t S ( t − s ) ∣ ζ ( s ) w ∣ d s ) ≤ u ( t ) ≤ 0 for all t > 0. If moreover 1 < p ≤ p F 1<p\le p_F 1 < p ≤ p F , then ∬ Q ∣ u ∣ p = ∥ u 0 ∥ L 1 + ∥ ζ w ∥ L 1 ( Q ) \iint_Q|u|^p=\|u_0\|_{L^1}+\|\zeta w\|_{L^1(Q)} ∬ Q ∣ u ∣ p = ∥ u 0 ∥ L 1 + ∥ ζ w ∥ L 1 ( Q ) and ∥ u ( t ) ∥ L 1 → 0 \|u(t)\|_{L^1}\to0 ∥ u ( t ) ∥ L 1 → 0 as t → ∞ t\to\infty t → ∞ .
On the other hand, data with M < 0 M<0 M < 0 containing a large positive bump blow up for every p > 1 p>1 p > 1 (Corollary 4.1(a) and Example 4.1). Thus, in the Fujita range, the sign of the effective mass decides the alternative completely when M ≥ 0 M\ge0 M ≥ 0 , while for M < 0 M<0 M < 0 both alternatives occur. This answers the question of whether an arbitrarily negative effective mass is still sufficient to force blow-up.
Large-time behavior and the asymptotic mass. The identity (7) has a striking dynamical counterpart.
Theorem 1.4. Let p > 1 p>1 p > 1 , ζ ∈ L 1 ( 0 , ∞ ) \zeta\in L^1(0,\infty) ζ ∈ L 1 ( 0 , ∞ ) and u 0 , w ∈ L 1 ( R N ) ∩ L ∞ ( R N ) u_0,w\in L^1(\mathbb{R}^N)\cap L^\infty(\mathbb{R}^N) u 0 , w ∈ L 1 ( R N ) ∩ L ∞ ( R N ) .
(i) If u u u is a global mild solution with ∣ u ∣ p ∈ L 1 ( Q ) |u|^p\in L^1(Q) ∣ u ∣ p ∈ L 1 ( Q ) , then, with m ∞ : = M + ∬ Q ∣ u ∣ p m_\infty:=M+\iint_Q|u|^p m ∞ := M + ∬ Q ∣ u ∣ p ,
lim t → ∞ ∥ u ( t ) − m ∞ G t ∥ L 1 = 0. \lim_{t\to\infty}\|u(t)-m_\infty G_t\|_{L^1}=0. t → ∞ lim ∥ u ( t ) − m ∞ G t ∥ L 1 = 0. (ii) If 1 < p ≤ p F 1<p\le p_F 1 < p ≤ p F , every global mild solution satisfies the hypothesis of (i) with m ∞ = 0 m_\infty=0 m ∞ = 0 ; in particular ∥ u ( t ) ∥ L 1 → 0 \|u(t)\|_{L^1}\to0 ∥ u ( t ) ∥ L 1 → 0 as t → ∞ t\to\infty t → ∞ .
(iii) If p > p F p>p_F p > p F and ζ \zeta ζ satisfies hypothesis ( H a ) (\mathrm{H}_a) ( H a ) below, then every global mild solution satisfying ∣ u ∣ ≤ κ V |u|\le\kappa V ∣ u ∣ ≤ κV (see Theorem 1.5 ) has ∣ u ∣ p ∈ L 1 ( Q ) |u|^p\in L^1(Q) ∣ u ∣ p ∈ L 1 ( Q ) and, for every q ∈ [ 1 , ∞ ] q\in[1,\infty] q ∈ [ 1 , ∞ ] ,
lim t → ∞ t N 2 ( 1 − 1 q ) ∥ u ( t ) − m ∞ G t ∥ L q = 0 , m ∞ ≥ M . \lim_{t\to\infty}t^{\frac N2(1-\frac1q)}\|u(t)-m_\infty G_t\|_{L^q}=0,\qquad m_\infty\ge M. t → ∞ lim t 2 N ( 1 − q 1 ) ∥ u ( t ) − m ∞ G t ∥ L q = 0 , m ∞ ≥ M .
Here and below we use the decay hypothesis
( H a ) (\mathrm{H}_a) ( H a ) ζ ∈ L 1 ( 0 , ∞ ) \zeta\in L^1(0,\infty) ζ ∈ L 1 ( 0 , ∞ ) and there exist C ζ > 0 C_\zeta>0 C ζ > 0 and a > max { 1 , N / 2 } a>\max\{1,N/2\} a > max { 1 , N /2 } such that ∣ ζ ( t ) ∣ ≤ C ζ t − a |\zeta(t)|\le C_\zeta t^{-a} ∣ ζ ( t ) ∣ ≤ C ζ t − a for a.e. t ≥ 1 t\ge1 t ≥ 1 ,
which holds for the model (1.2) with every a > 0 a>0 a > 0 . Theorem 1.4 yields a characterization of the Fujita exponent through the asymptotic mass (Corollary 7.1): p ≤ p F p\le p_F p ≤ p F if and only if every global solution loses all its mass.
Global existence above the Fujita exponent. Global existence is obtained through a comparison with the linear forced flow with absolute data
V ( x , t ) : = [ S ( t ) ∣ u 0 ∣ ] ( x ) + ∫ 0 t [ ζ ( s ) ∣ S ( t − s ) w ∣ ] ( x ) d s , V(x,t):=[S(t)|u_0|](x)+\int_0^t[\zeta(s)|S(t-s)w|](x)\,\mathrm{d}s, V ( x , t ) := [ S ( t ) ∣ u 0 ∣ ] ( x ) + ∫ 0 t [ ζ ( s ) ∣ S ( t − s ) w ∣ ] ( x ) d s ,
m ( t ) : = ∥ S ( t ) ∣ u 0 ∣ ∥ L ∞ + ∫ 0 t ∣ ζ ( s ) ∣ ∥ S ( t − s ) w ∥ L ∞ d s . (9) m(t):=\|S(t)|u_0|\|_{L^\infty}+\int_0^t|\zeta(s)|\|S(t-s)w\|_{L^\infty}\,\mathrm{d}s. \tag*{(9)} m ( t ) := ∥ S ( t ) ∣ u 0 ∣ ∥ L ∞ + ∫ 0 t ∣ ζ ( s ) ∣∥ S ( t − s ) w ∥ L ∞ d s . (9)
Theorem 1.5. Let p > 1 p>1 p > 1 , u 0 , w ∈ L 1 ( R N ) ∩ L ∞ ( R N ) u_0,w\in L^1(\mathbb{R}^N)\cap L^\infty(\mathbb{R}^N) u 0 , w ∈ L 1 ( R N ) ∩ L ∞ ( R N ) , ζ ∈ L l o c 1 ( [ 0 , ∞ ) ) \zeta\in L^1_{\mathrm{loc}}([0,\infty)) ζ ∈ L loc 1 ([ 0 , ∞ )) and 0 < T ≤ ∞ 0<T\le\infty 0 < T ≤ ∞ . Set κ : = p / ( p − 1 ) \kappa:=p/(p-1) κ := p / ( p − 1 ) and c p : = ( p − 1 ) p − 1 / p p c_p:=(p-1)^{p-1}/p^p c p := ( p − 1 ) p − 1 / p p . If
∫ 0 T m ( t ) p − 1 d t ≤ c p , (10) \int_0^T m(t)^{p-1}\,\mathrm{d}t\le c_p, \tag*{(10)} ∫ 0 T m ( t ) p − 1 d t ≤ c p , (10)
then (1.1) has a unique mild solution on ( 0 , T ) (0,T) ( 0 , T ) , and it satisfies the pointwise bound
∣ u ( x , t ) ∣ ≤ κ V ( x , t ) for a.e. ( x , t ) ∈ R N × ( 0 , T ) . (11) |u(x,t)| \le\kappa V(x,t) \qquad\text{for a.e. }(x,t) \in\mathbb{R}^{N} \times(0,T). \tag*{(11)} ∣ u ( x , t ) ∣ ≤ κV ( x , t ) for a.e. ( x , t ) ∈ R N × ( 0 , T ) . (11)
No sign condition on u 0 u_0 u 0 , w w w or ζ \zeta ζ is required.
When T = ∞ T=\infty T = ∞ , (10) can only hold if p > p F p>p_F p > p F (unless the data vanish), and it is then a smallness condition. Writing ω ζ : = ∫ 0 ∞ ( ∣ ζ ∣ ⋆ h N ) ( t ) p − 1 d t \omega_\zeta:= \int_0^\infty(|\zeta| \star h_N)(t)^{p-1}\,\mathrm{d}t ω ζ := ∫ 0 ∞ ( ∣ ζ ∣ ⋆ h N ) ( t ) p − 1 d t , we obtain the following.
Corollary 1.1. Let p > p F p>p_F p > p F and ζ ∈ L l o c 1 ( [ 0 , ∞ ) ) \zeta\in L^1_{\mathrm{loc}}([0,\infty)) ζ ∈ L loc 1 ([ 0 , ∞ )) with ω ζ < ∞ \omega_\zeta<\infty ω ζ < ∞ . There exists δ = δ ( N , p ) > 0 \delta=\delta(N,p)>0 δ = δ ( N , p ) > 0 such that, if u 0 , w ∈ L 1 ( R N ) ∩ L ∞ ( R N ) u_0,w \in L^1(\mathbb{R}^N) \cap L^\infty(\mathbb{R}^N) u 0 , w ∈ L 1 ( R N ) ∩ L ∞ ( R N ) satisfy
∥ u 0 ∥ 1 p − 1 + ω ζ ∥ w ∥ 1 p − 1 ≤ δ , \left\lVert u_0\right\rVert_1^{p-1}+\omega_\zeta\left\lVert w\right\rVert_1^{p-1}\le\delta, ∥ u 0 ∥ 1 p − 1 + ω ζ ∥ w ∥ 1 p − 1 ≤ δ , then (1) has a unique global mild solution, ∣ u ∣ ≤ κ V |u|\le\kappa V ∣ u ∣ ≤ κV , and for all t > 0 t>0 t > 0
∥ u ( t ) ∥ L ∞ ≤ κ ( ∥ u 0 ∥ h N ( t ) + ∥ w ∥ ( ∣ ζ ∣ ⋆ h N ) ( t ) ) , ∥ u ( t ) ∥ L 1 ≤ κ ( ∥ u 0 ∥ L 1 + ∥ w ∥ L 1 ∫ 0 t ∣ ζ ∣ ) . \left\lVert u(t)\right\rVert_{L^\infty}\le\kappa\left(\left\lVert u_0\right\rVert h_N(t)+\left\lVert w\right\rVert(|\zeta|\star h_N)(t)\right),\qquad
\left\lVert u(t)\right\rVert_{L^1}\le\kappa\left(\left\lVert u_0\right\rVert_{L^1}+\left\lVert w\right\rVert_{L^1}\int_0^t|\zeta|\right). ∥ u ( t ) ∥ L ∞ ≤ κ ( ∥ u 0 ∥ h N ( t ) + ∥ w ∥ ( ∣ ζ ∣ ⋆ h N ) ( t ) ) , ∥ u ( t ) ∥ L 1 ≤ κ ( ∥ u 0 ∥ L 1 + ∥ w ∥ L 1 ∫ 0 t ∣ ζ ∣ ) . Moreover ω ζ < ∞ \omega_\zeta<\infty ω ζ < ∞ in each of the following cases: (a) p ≥ 2 p\ge2 p ≥ 2 and ζ ∈ L 1 ( 0 , ∞ ) \zeta\in L^1(0,\infty) ζ ∈ L 1 ( 0 , ∞ ) ; (b) ζ \zeta ζ satisfies ( H a ) (\mathrm{H}_a) ( H a ) , in which case ∥ u ( t ) ∥ L ∞ ≤ C ( 1 + t ) − N / 2 \left\lVert u(t)\right\rVert_{L^\infty}\le C(1+t)^{-N/2} ∥ u ( t ) ∥ L ∞ ≤ C ( 1 + t ) − N /2 ; (c) N ≥ 3 N\ge3 N ≥ 3 , ζ ∈ L 1 ( 0 , ∞ ) \zeta\in L^1(0,\infty) ζ ∈ L 1 ( 0 , ∞ ) and ∣ ζ ( t ) ∣ ≤ C t − b |\zeta(t)|\le Ct^{-b} ∣ ζ ( t ) ∣ ≤ C t − b for t ≥ 1 t\ge1 t ≥ 1 with b > 1 b>1 b > 1 and b ( p − 1 ) > 1 b(p-1)>1 b ( p − 1 ) > 1 .
In particular, for the model (2) , small data of either sign give global solutions for every p > p F p>p_F p > p F , with ∥ u ( t ) ∥ L ∞ ≤ C ε ( 1 + t ) − N / 2 \left\lVert u(t)\right\rVert_{L^\infty}\le C\varepsilon(1+t)^{-N/2} ∥ u ( t ) ∥ L ∞ ≤ C ε ( 1 + t ) − N /2 . Hence the critical exponent for integrable forcings is exactly p F p_F p F ; nonexistence for every p > 1 p>1 p > 1 , which holds for growing forcings t σ w t^\sigma w t σ w , σ > 0 \sigma>0 σ > 0 [20 ] , fails here.
A localized obstruction valid for all $p>1. The test-function method also produces a necessary condition for global existence that does not use the integrability of the data.
Theorem 1.6. Let p > 1 p>1 p > 1 , u 0 ∈ L l o c 1 ( R N ) u_0\in L^1_{\mathrm{loc}}(\mathbb{R}^N) u 0 ∈ L loc 1 ( R N ) , w ∈ L l o c 1 ( R N ) w\in L^1_{\mathrm{loc}}(\mathbb{R}^N) w ∈ L loc 1 ( R N ) and ζ ∈ L l o c 1 ( [ 0 , ∞ ) ) \zeta\in L^1_{\mathrm{loc}}([0,\infty)) ζ ∈ L loc 1 ([ 0 , ∞ )) , and let η \eta η be the cutoff fixed in (19) . There is C 0 = C 0 ( N , p , η ) > 0 C_0=C_0(N,p,\eta)>0 C 0 = C 0 ( N , p , η ) > 0 such that, if (1) has a global weak solution, then for all x 0 ∈ R N x_0\in\mathbb{R}^N x 0 ∈ R N and R , T > 0 R,T>0 R , T > 0 ,
∫ R N u 0 η ( ∣ x − x 0 ∣ 2 R 2 ) 2 p ′ d x + ( ∫ 0 ∞ ζ ( t ) η ( t T ) 2 p ′ d t ) ( ∫ R N w η ( ∣ x − x 0 ∣ 2 R 2 ) 2 p ′ d x ) ≤ C 0 ( R N T 1 − p ′ + R N − 2 p ′ T ) . (12) \int_{\mathbb{R}^N}u_0\eta\left(\frac{|x-x_0|^2}{R^2}\right)^{2p'}\,\mathrm{d}x
+\left(\int_0^\infty\zeta(t)\eta\left(\frac{t}{T}\right)^{2p'}\,\mathrm{d}t\right)
\left(\int_{\mathbb{R}^N}w\eta\left(\frac{|x-x_0|^2}{R^2}\right)^{2p'}\,\mathrm{d}x\right)
\le C_0\left(R^N T^{1-p'}+R^{N-2p'}T\right). \tag*{(12)} ∫ R N u 0 η ( R 2 ∣ x − x 0 ∣ 2 ) 2 p ′ d x + ( ∫ 0 ∞ ζ ( t ) η ( T t ) 2 p ′ d t ) ( ∫ R N w η ( R 2 ∣ x − x 0 ∣ 2 ) 2 p ′ d x ) ≤ C 0 ( R N T 1 − p ′ + R N − 2 p ′ T ) . (12)
Corollary 4.1 derives from (12) that large data blow up for every p > 1 p>1 p > 1 whatever the forcing, that large forcings blow up for every p > 1 p>1 p > 1 , and that data decaying like ∣ x ∣ − γ |x|^{-\gamma} ∣ x ∣ − γ with γ < min { N , 2 / ( p − 1 ) } \gamma<\min\{N,2/(p-1)\} γ < min { N , 2/ ( p − 1 )} blow up (the forced analogue of a result of Lee and Ni [26 ] ). Corollary 4.2 treats non-integrable forcings and recovers the blow-up part of [20 ] .
Sharp lifespan. Finally, we determine the lifespan of small solutions with positive effective mass in the Fujita range. For a , b ∈ L 1 ( R N ) ∩ L ∞ ( R N ) a,b\in L^1(\mathbb{R}^N)\cap L^\infty(\mathbb{R}^N) a , b ∈ L 1 ( R N ) ∩ L ∞ ( R N ) and ε > 0 \varepsilon>0 ε > 0 , let T ε ∈ ( 0 , ∞ ] T_\varepsilon\in(0,\infty] T ε ∈ ( 0 , ∞ ] be the supremum of the T > 0 T>0 T > 0 for which (1) with u 0 = ε a u_0=\varepsilon a u 0 = ε a and w = ε b w=\varepsilon b w = ε b has a mild solution on ( 0 , T ) (0,T) ( 0 , T ) .
Theorem 1.7. Let 1 < p ≤ p F 1<p\le p_F 1 < p ≤ p F , ζ ∈ L 1 ( 0 , ∞ ) \zeta\in L^1(0,\infty) ζ ∈ L 1 ( 0 , ∞ ) , a , b ∈ L 1 ( R N ) ∩ L ∞ ( R N ) a,b\in L^1(\mathbb{R}^N)\cap L^\infty(\mathbb{R}^N) a , b ∈ L 1 ( R N ) ∩ L ∞ ( R N ) , and set M ∗ : = ∫ a + Z ∫ b M_*:=\int a+\mathcal{Z}\int b M ∗ := ∫ a + Z ∫ b .
(i) If M ∗ > 0 M_*>0 M ∗ > 0 , there is C > 0 C>0 C > 0 such that for all ε ∈ ( 0 , 1 ] \varepsilon\in(0,1] ε ∈ ( 0 , 1 ] ,
T ε ≤ C ε − 2 ( p − 1 ) 2 − N ( p − 1 ) if p < p F , T ε ≤ exp ( C ε − ( p − 1 ) ) if p = p F . T_\varepsilon\le C\varepsilon^{-\frac{2(p-1)}{2-N(p-1)}}\quad\text{if }p<p_F,\qquad
T_\varepsilon\le\exp\left(C\varepsilon^{-(p-1)}\right)\quad\text{if }p=p_F. T ε ≤ C ε − 2 − N ( p − 1 ) 2 ( p − 1 ) if p < p F , T ε ≤ exp ( C ε − ( p − 1 ) ) if p = p F . (ii) If ζ \zeta ζ satisfies ( H a ) (\mathrm{H}_a) ( H a ) (or, more generally, ( H K ) (\mathrm{H}_K) ( H K ) in (28) ), there are c , ε 0 > 0 c,\varepsilon_0>0 c , ε 0 > 0 such that for all ε ∈ ( 0 , ε 0 ] \varepsilon\in(0,\varepsilon_0] ε ∈ ( 0 , ε 0 ] ,
T ε ≥ c ε − 2 ( p − 1 ) 2 − N ( p − 1 ) if p < p F , T ε ≥ exp ( c ε − ( p − 1 ) ) if p = p F , T_\varepsilon\ge c\varepsilon^{-\frac{2(p-1)}{2-N(p-1)}}\quad\text{if }p<p_F,\qquad
T_\varepsilon\ge\exp\left(c\varepsilon^{-(p-1)}\right)\quad\text{if }p=p_F, T ε ≥ c ε − 2 − N ( p − 1 ) 2 ( p − 1 ) if p < p F , T ε ≥ exp ( c ε − ( p − 1 ) ) if p = p F , without any sign condition on a a a , b b b or M ∗ M_* M ∗ .
The upper bounds hold for weak solutions and use a test-function argument in the spirit of [19 ] ; the lower bounds follow from the local version of Theorem 1.5 . The exponents coincide with those of Lee and Ni [26 ] for the unforced problem with nonnegative data: integrable forcing does not alter the lifespan scale, only the constant through M ∗ . M_*. M ∗ .
Discussion and novelty
The picture that emerges is the following. An integrable forcing never changes the critical exponent, and its whole effect in the Fujita range is encoded in the single number M M M : a positive impulse Z W ZW Z W subsidizes the initial mass, so that even data with m 0 < 0 m_0<0 m 0 < 0 blow up when m 0 > − Z W m_0>-ZW m 0 > − Z W (Remark 3.1). This contrasts sharply with the non-integrable regime, where the accumulated impulse diverges and shifts or destroys the threshold [3 , 20 ] . The mass-annihilation identity is a quantitative form of the test-function argument: rather than estimating the right-hand side of the weak formulation from above, we show that it tends to zero for every global solution, which is what allows us to reach the borderline case M = 0 M=0 M = 0 and to describe the dynamics (vanishing of the L 1 L^1 L 1 norm) rather than merely nonexistence. In the unforced, nonpositive case it reduces to the classical extinction of mass for the absorption equation in the range p ≤ p F p\le p_F p ≤ p F , for which it provides a short proof (Remark 3.3).
On the existence side, the domination principle (1.10) is a simple but flexible device: it gives global existence for sign-changing data and forcing with a fully explicit smallness condition, it immediately provides the local existence time needed for sharp lifespan lower bounds, and it does not require integrability of ζ \zeta ζ (Remark 4.1).
The paper is organized as follows. § 2 collects notation, the definitions of weak and mild solutions, and the auxiliary lemmas (heat kernel, cutoff functions, damped Duhamel formula, kernel estimates for the forcing). § 3 proves Theorems 1.1 and 1.2 . § 4 is devoted to the localized obstruction and its consequences, including non-integrable forcings. § 5 proves Theorems 1.5 and 1.3 and Corollary 1.1. § 6 proves Theorem 1.7 , and § 7 proves Theorem 1.4 . Section 8 summarizes the complete picture for the model example (1.2), and § 9 lists open problems; all statements there are conjectural and are clearly separated from the results proved in the paper.
Preliminaries
Notation and the heat semigroup
For 1 ≤ q ≤ ∞ 1\le q\le\infty 1 ≤ q ≤ ∞ we write ∥ ⋅ ∥ q \|\cdot\|_q ∥ ⋅ ∥ q for the norm of L q ( R N ) L^q(\mathbb{R}^N) L q ( R N ) , and [ [ f ] ] = max { ∥ f ∥ 1 , ∥ f ∥ ∞ } [\![f]\!]=\max\{\|f\|_1,\|f\|_\infty\} [ [ f ] ] = max { ∥ f ∥ 1 , ∥ f ∥ ∞ } . For 0 < T ≤ ∞ 0<T\le\infty 0 < T ≤ ∞ we set Q T : = R N × ( 0 , T ) Q_T:=\mathbb{R}^N\times(0,T) Q T := R N × ( 0 , T ) , so that Q = Q ∞ Q=Q_\infty Q = Q ∞ . A function u u u on Q T Q_T Q T belongs to L l o c q ( R N × [ 0 , T ) ) L^q_{\mathrm{loc}}(\mathbb{R}^N\times[0,T)) L loc q ( R N × [ 0 , T )) if u ∈ L q ( K ∩ Q T ) u\in L^q(K\cap Q_T) u ∈ L q ( K ∩ Q T ) for every compact set K ⊂ R N × [ 0 , T ) K\subset\mathbb{R}^N\times[0,T) K ⊂ R N × [ 0 , T ) ; such compact sets may touch the initial time. For a measurable f f f on Q T Q_T Q T we use the Duhamel operator
D [ f ] ( x , t ) : = ∫ 0 t ∫ R N G t − s ( x − y ) f ( y , s ) d y d s , \mathcal{D}[f](x,t):=\int_0^t\int_{\mathbb{R}^N}G_{t-s}(x-y)f(y,s)\,dy\,ds, D [ f ] ( x , t ) := ∫ 0 t ∫ R N G t − s ( x − y ) f ( y , s ) d y d s ,
whenever the integral converges absolutely; we also write D [ f ] ( t ) = ∫ 0 t S ( t − s ) f ( s ) d s \mathcal{D}[f](t)=\int_0^t S(t-s)f(s)\,ds D [ f ] ( t ) = ∫ 0 t S ( t − s ) f ( s ) d s . The cutoff exponent
k : = 2 p ′ = 2 p p − 1 > 2 k:=2p'= \frac{2p}{p-1}>2 k := 2 p ′ = p − 1 2 p > 2
will be used throughout. The letter C C C denotes a positive constant whose dependence is indicated when relevant and which may change from line to line.
Lemma 2.1. The following properties hold.
(a) S ( t ) S(t) S ( t ) is positivity preserving, ∣ S ( t ) f ∣ ≤ S ( t ) ∣ f ∣ |S(t)f|\le S(t)|f| ∣ S ( t ) f ∣ ≤ S ( t ) ∣ f ∣ , S ( t ) 1 = 1 S(t)1=1 S ( t ) 1 = 1 , and S ( t ) S ( s ) = S ( t + s ) S(t)S(s)=S(t+s) S ( t ) S ( s ) = S ( t + s ) for t , s ≥ 0 t,s\ge0 t , s ≥ 0 .
(b) ∥ S ( t ) f ∥ q ≤ ∥ f ∥ q \|S(t)f\|_q\le\|f\|_q ∥ S ( t ) f ∥ q ≤ ∥ f ∥ q for 1 ≤ q ≤ ∞ 1\le q\le\infty 1 ≤ q ≤ ∞ , ∥ S ( t ) f ∥ ∞ ≤ ( 4 π t ) − N / 2 ∥ f ∥ 1 \|S(t)f\|_\infty\le(4\pi t)^{-N/2}\|f\|_1 ∥ S ( t ) f ∥ ∞ ≤ ( 4 π t ) − N /2 ∥ f ∥ 1 ; hence ∥ S ( t ) f ∥ ∞ ≤ h N ( t ) [ [ f ] ] \|S(t)f\|_\infty\le h_N(t)[\![f]\!] ∥ S ( t ) f ∥ ∞ ≤ h N ( t ) [ [ f ] ] .
(c) ∫ R N S ( t ) f d x = ∫ R N f d x \int_{\mathbb{R}^N}S(t)f\,dx=\int_{\mathbb{R}^N}f\,dx ∫ R N S ( t ) f d x = ∫ R N f d x for f ∈ L 1 ( R N ) f\in L^1(\mathbb{R}^N) f ∈ L 1 ( R N ) .
(d) For f ∈ L 1 ( R N ) f\in L^1(\mathbb{R}^N) f ∈ L 1 ( R N ) , ∥ S ( t ) f − ( ∫ f ) G t ∥ 1 → 0 \|S(t)f-(\int f)G_t\|_1\to0 ∥ S ( t ) f − ( ∫ f ) G t ∥ 1 → 0 as t → ∞ t\to\infty t → ∞ .
(e) For every fixed s ≥ 0 s \ge0 s ≥ 0 , ∥ G t − s − G t ∥ 1 → 0 \lVert G_{t-s}-G_t\rVert_1 \to0 ∥ G t − s − G t ∥ 1 → 0 as t → ∞ t \to\infty t → ∞ .
(f) h N h_N h N is nonincreasing, h N ( t / 2 ) ≤ 2 N / 2 h N ( t ) h_N(t/2) \le2^{N/2}h_N(t) h N ( t /2 ) ≤ 2 N /2 h N ( t ) for t > 0 t>0 t > 0 , and ∫ 0 ∞ h N ( t ) p − 1 d t < ∞ \int_0^\infty h_N(t)^{p-1}\,dt<\infty ∫ 0 ∞ h N ( t ) p − 1 d t < ∞ if and only if p > p F p>p_F p > p F .
Proof. Properties (a)–(c) are classical. For (d), assume first that f ∈ C c ( R N ) f\in C_c(\mathbb{R}^N) f ∈ C c ( R N ) with supp f ⊂ B ρ \operatorname{supp}f\subset B_\rho supp f ⊂ B ρ . Then
∥ S ( t ) f − ( ∫ f ) G t ∥ 1 ≤ ∫ B ρ ∣ f ( y ) ∣ ∥ G t ( ⋅ − y ) − G t ∥ 1 d y = ∫ B ρ ∣ f ( y ) ∣ ∥ G 1 ( ⋅ − y / t ) − G 1 ∥ 1 d y . \left\lVert S(t)f-\left(\int f\right)G_t\right\rVert_1\le\int_{B_\rho}|f(y)|\left\lVert G_t(\cdot-y)-G_t\right\rVert_1\,dy=\int_{B_\rho}|f(y)|\left\lVert G_1(\cdot-y/\sqrt{t})-G_1\right\rVert_1\,dy. S ( t ) f − ( ∫ f ) G t 1 ≤ ∫ B ρ ∣ f ( y ) ∣ ∥ G t ( ⋅ − y ) − G t ∥ 1 d y = ∫ B ρ ∣ f ( y ) ∣ G 1 ( ⋅ − y / t ) − G 1 1 d y .
which tends to zero by the continuity of translations in L 1 L^1 L 1 , uniformly for ∣ y ∣ ≤ ρ |y|\le\rho ∣ y ∣ ≤ ρ . For general f ∈ L 1 f\in L^1 f ∈ L 1 we approximate by f j ∈ C c f_j\in C_c f j ∈ C c and use ∥ S ( t ) ( f − f j ) ∥ 1 + ∥ ( f − f j ) ∥ 1 ∥ G t ∥ 1 ≤ 2 ∥ f − f j ∥ 1 \lVert S(t)(f-f_j)\rVert_1+\lVert(f-f_j)\rVert_1\lVert G_t\rVert_1\le2\lVert f-f_j\rVert_1 ∥ S ( t ) ( f − f j ) ∥ 1 + ∥( f − f j ) ∥ 1 ∥ G t ∥ 1 ≤ 2 ∥ f − f j ∥ 1 . For (e), the change of variables x = t y x=\sqrt{t}y x = t y gives ∥ G t − s − G t ∥ 1 = ∥ G ( t − s ) / t − G 1 ∥ 1 \lVert G_{t-s}-G_t\rVert_1=\lVert G_{(t-s)/t}-G_1\rVert_1 ∥ G t − s − G t ∥ 1 = ∥ G ( t − s ) / t − G 1 ∥ 1 , and λ ↦ G λ \lambda\mapsto G_\lambda λ ↦ G λ is continuous from ( 0 , ∞ ) (0,\infty) ( 0 , ∞ ) into L 1 L^1 L 1 by dominated convergence. In (f), the monotonicity is clear; if 2 π t ≥ 1 2\pi t\ge1 2 π t ≥ 1 then h N ( t / 2 ) = ( 2 π t ) − N / 2 = 2 N / 2 h N ( t ) h_N(t/2)=(2\pi t)^{-N/2}=2^{N/2}h_N(t) h N ( t /2 ) = ( 2 π t ) − N /2 = 2 N /2 h N ( t ) , while if 2 π t < 1 2\pi t<1 2 π t < 1 then h N ( t / 2 ) = 1 < 2 N / 2 h N ( t ) h_N(t/2)=1<2^{N/2}h_N(t) h N ( t /2 ) = 1 < 2 N /2 h N ( t ) because ( 4 π t ) − N / 2 > 2 − N / 2 (4\pi t)^{-N/2}>2^{-N/2} ( 4 π t ) − N /2 > 2 − N /2 . Finally h N p − 1 h_N^{p-1} h N p − 1 is bounded and behaves like t − N ( p − 1 ) / 2 t^{-N(p-1)/2} t − N ( p − 1 ) /2 at infinity, which is integrable exactly when N ( p − 1 ) / 2 > 1 N(p-1)/2>1 N ( p − 1 ) /2 > 1 , that is p > p F p>p_F p > p F . □
Weak and mild solutions
Definition 2.1. Let 0 < T ≤ ∞ 0<T\le\infty 0 < T ≤ ∞ , u 0 ∈ L l o c 1 ( R N ) u_0\in L^1_{\mathrm{loc}}(\mathbb{R}^N) u 0 ∈ L loc 1 ( R N ) , w ∈ L l o c 1 ( R N ) w\in L^1_{\mathrm{loc}}(\mathbb{R}^N) w ∈ L loc 1 ( R N ) and ζ ∈ L l o c 1 ( [ 0 , ∞ ) ) \zeta\in L^1_{\mathrm{loc}}([0,\infty)) ζ ∈ L loc 1 ([ 0 , ∞ )) . A function u ∈ L l o c p ( R N × [ 0 , T ) ) u\in L^p_{\mathrm{loc}}(\mathbb{R}^N\times[0,T)) u ∈ L loc p ( R N × [ 0 , T )) is called a weak solution of {(xref)}(1.1) on Q T Q_T Q T if, for every ψ ∈ C c ∞ ( R N × [ 0 , T ) ) \psi\in C_c^\infty(\mathbb{R}^N\times[0,T)) ψ ∈ C c ∞ ( R N × [ 0 , T )) ,
∬ Q T ∣ u ∣ p ψ d x d t + ∬ Q T ζ ( t ) w ( x ) ψ d x d t + ∫ R N u 0 ( x ) ψ ( x , 0 ) d x = − ∬ Q T u ( ∂ t ψ + Δ ψ ) d x d t . (13) \iint_{Q_T}|u|^p\psi\,dx\,dt+\iint_{Q_T}\zeta(t)w(x)\psi\,dx\,dt+\int_{\mathbb{R}^N}u_0(x)\psi(x,0)\,dx=-\iint_{Q_T}u(\partial_t\psi+\Delta\psi)\,dx\,dt. \tag*{(13)} ∬ Q T ∣ u ∣ p ψ d x d t + ∬ Q T ζ ( t ) w ( x ) ψ d x d t + ∫ R N u 0 ( x ) ψ ( x , 0 ) d x = − ∬ Q T u ( ∂ t ψ + Δ ψ ) d x d t . (13)
It is a global weak solution if T = ∞ T=\infty T = ∞ .
All integrals in {(xref)}(2.1) are finite, since ζ ⊗ w ∈ L l o c 1 ( R N × [ 0 , ∞ ) ) \zeta\otimes w\in L^1_{\mathrm{loc}}(\mathbb{R}^N\times[0,\infty)) ζ ⊗ w ∈ L loc 1 ( R N × [ 0 , ∞ )) by Fubini’s theorem. The test functions are compactly supported in space and time but need not vanish at t = 0 t=0 t = 0 ; consequently the initial datum is genuinely felt through the boundary term in {(xref)}(2.1). This is the point at which sign-changing data are correctly handled. We write L ψ : = ∂ t ψ + Δ ψ L\psi:=\partial_t\psi+\Delta\psi L ψ := ∂ t ψ + Δ ψ .
Remark 2.1. Identity {(xref)}(2.1) extends to every ψ ∈ C c 2 , 1 ( R N × [ 0 , T ) ) \psi\in C_c^{2,1}(\mathbb{R}^N\times[0,T)) ψ ∈ C c 2 , 1 ( R N × [ 0 , T )) , that is, to compactly supported functions such that ψ \psi ψ , ∇ ψ \nabla\psi ∇ ψ , D 2 ψ D^2\psi D 2 ψ and ∂ t ψ \partial_t\psi ∂ t ψ are continuous on R N × [ 0 , T ) \mathbb{R}^N\times[0,T) R N × [ 0 , T ) . Indeed, extend ψ \psi ψ to negative times by the reflection ψ ( x , − t ) : = 4 ψ ( x , t / 2 ) − 3 ψ ( x , t ) \psi(x,-t):=4\psi(x,t/2)-3\psi(x,t) ψ ( x , − t ) := 4 ψ ( x , t /2 ) − 3 ψ ( x , t ) , t > 0 t>0 t > 0 , which preserves the C 2 , 1 C^{2,1} C 2 , 1 regularity across t = 0 t=0 t = 0 , and mollify in ( x , t ) (x,t) ( x , t ) . The mollified functions ψ j \psi_j ψ j , restricted to t ≥ 0 t\ge0 t ≥ 0 , belong to C c ∞ ( R N × [ 0 , T ) ) C_c^\infty(\mathbb{R}^N\times[0,T)) C c ∞ ( R N × [ 0 , T )) , have supports in a fixed compact subset of R N × [ 0 , T ) \mathbb{R}^N\times[0,T) R N × [ 0 , T ) , and satisfy ψ j → ψ \psi_j\to\psi ψ j → ψ , L ψ j → L ψ L\psi_j\to L\psi L ψ j → L ψ uniformly; since u u u , ∣ u ∣ p |u|^p ∣ u ∣ p and ζ ⊗ w \zeta\otimes w ζ ⊗ w are integrable on that compact set and u 0 u_0 u 0 is integrable on its trace at t = 0 t=0 t = 0 , we can pass to the limit in each term of {(xref)}(2.1). All the cutoff functions used below are of class C 2 , 1 C^{2,1} C 2 , 1 , because they are powers η k \eta^k η k with k > 2 k>2 k > 2 of a smooth function.
Definition 2.2. Let u 0 , w ∈ L 1 ( R N ) ∩ L ∞ ( R N ) u_0,w\in L^1(\mathbb{R}^N)\cap L^\infty(\mathbb{R}^N) u 0 , w ∈ L 1 ( R N ) ∩ L ∞ ( R N ) , ζ ∈ L l o c 1 ( [ 0 , ∞ ) ) \zeta\in L^1_{\mathrm{loc}}([0,\infty)) ζ ∈ L loc 1 ([ 0 , ∞ )) and 0 < T ≤ ∞ 0<T\le\infty 0 < T ≤ ∞ . A measurable function u : Q T → R u:Q_T\to\mathbb{R} u : Q T → R is called a mild solution of {(xref)}(1.1) on ( 0 , T ) (0,T) ( 0 , T ) if, for every T ′ ∈ ( 0 , T ) T'\in(0,T) T ′ ∈ ( 0 , T ) , u ∈ L ∞ ( Q T ′ ) u\in L^\infty(Q_{T'}) u ∈ L ∞ ( Q T ′ ) , ess sup 0 < t < T ′ ∥ u ( t ) ∥ 1 < ∞ \operatorname*{ess\,sup}_{0<t<T'}\lVert u(t)\rVert_1<\infty ess sup 0 < t < T ′ ∥ u ( t ) ∥ 1 < ∞ , and
u ( t ) = S ( t ) u 0 + ∫ 0 t S ( t − s ) ∣ u ( s ) ∣ p d s + ∫ 0 t S ( t − s ) ( ζ ( s ) w ) d s (14) u(t)=S(t)u_0+\int_0^tS(t-s)|u(s)|^p\,ds+\int_0^tS(t-s)(\zeta(s)w)\,ds \tag*{(14)} u ( t ) = S ( t ) u 0 + ∫ 0 t S ( t − s ) ∣ u ( s ) ∣ p d s + ∫ 0 t S ( t − s ) ( ζ ( s ) w ) d s (14)
for a.e. ( x , t ) ∈ Q T (x,t)\in Q_T ( x , t ) ∈ Q T . It is a global mild solution if T = ∞ T=\infty T = ∞ .
The right-hand side of {(xref)}(2.2) does not depend on the representative of u u u and is defined at every point of Q T Q_T Q T ; we always use it as the representative of u u u . With this convention u ( t ) ∈ L 1 ∩ L ∞ u(t)\in L^1\cap L^\infty u ( t ) ∈ L 1 ∩ L ∞ for every t ∈ ( 0 , T ) t\in(0,T) t ∈ ( 0 , T ) , and the bounds in Definition 2.2 hold for every t t t . Note also that ∣ u ∣ p + ζ w ∈ L 1 ( Q T ) |u|^p+\zeta w\in L^1(Q_T) ∣ u ∣ p + ζ w ∈ L 1 ( Q T ) , since ∥ u ( s ) ∥ 1 ≤ ∥ u ∥ L ∞ ( Q T ′ ) p − 1 ∥ u ( s ) ∥ 1 \lVert u(s)\rVert_1\le\lVert u\rVert_{L^\infty(Q_{T'})}^{p-1}\lVert u(s)\rVert_1 ∥ u ( s ) ∥ 1 ≤ ∥ u ∥ L ∞ ( Q T ′ ) p − 1 ∥ u ( s ) ∥ 1 .
Lemma 2.2. Let u u u be a mild solution of (1) on ( 0 , T ) (0,T) ( 0 , T ) and set f : = ∣ u ∣ p + ζ w f := |u|^p+\zeta w f := ∣ u ∣ p + ζ w .
(a) u u u is a weak solution of (1) on Q T Q_T Q T .
(b) For every t ∈ ( 0 , T ) t\in(0,T) t ∈ ( 0 , T ) ,
∫ R N u ( x , t ) d x = m 0 + ∫ 0 t ∫ R N ∣ u ∣ p d x d s + W ∫ 0 t ζ ( s ) d s . (15) \int_{\mathbb{R}^N}u(x,t)\,dx=m_0+\int_0^t\int_{\mathbb{R}^N}|u|^p\,dx\,ds+W\int_0^t\zeta(s)\,ds. \tag*{(15)} ∫ R N u ( x , t ) d x = m 0 + ∫ 0 t ∫ R N ∣ u ∣ p d x d s + W ∫ 0 t ζ ( s ) d s . (15) (c) For 0 < τ < t < T 0<\tau<t<T 0 < τ < t < T , u ( t ) = S ( t − τ ) u ( τ ) + ∫ τ t S ( t − s ) f ( s ) d s u(t)=S(t-\tau)u(\tau)+\int_\tau^tS(t-s)f(s)\,ds u ( t ) = S ( t − τ ) u ( τ ) + ∫ τ t S ( t − s ) f ( s ) d s .
Proof. Fix ψ ∈ C c ∞ ( R N × [ 0 , T ) ) \psi\in C_c^\infty(\mathbb{R}^N\times[0,T)) ψ ∈ C c ∞ ( R N × [ 0 , T )) and T ′ < T T'<T T ′ < T with supp ψ ⊂ R N × [ 0 , T ′ ] \operatorname{supp}\psi\subset\mathbb{R}^N\times[0,T'] supp ψ ⊂ R N × [ 0 , T ′ ] . We first claim that for every g ∈ L 1 ( R N ) g\in L^1(\mathbb{R}^N) g ∈ L 1 ( R N ) and s ∈ [ 0 , T ) s\in[0,T) s ∈ [ 0 , T ) ,
∫ s T ∫ R N [ S ( t − s ) g ] ( x ) L ψ ( x , t ) d x d t = − ∫ R N g ( x ) ψ ( x , s ) d x . (16) \int_s^T\int_{\mathbb{R}^N}[S(t-s)g](x)L\psi(x,t)\,dx\,dt=-\int_{\mathbb{R}^N}g(x)\psi(x,s)\,dx. \tag*{(16)} ∫ s T ∫ R N [ S ( t − s ) g ] ( x ) L ψ ( x , t ) d x d t = − ∫ R N g ( x ) ψ ( x , s ) d x . (16)
By the symmetry of the heat kernel, ∫ [ S ( t − s ) g ] L ψ ( t ) d x = ∫ g S ( t − s ) [ L ψ ( t ) ] d x \int[S(t-s)g]L\psi(t)\,dx=\int gS(t-s)[L\psi(t)]\,dx ∫ [ S ( t − s ) g ] L ψ ( t ) d x = ∫ g S ( t − s ) [ L ψ ( t )] d x . For a function ϕ ∈ C 2 \phi\in C^2 ϕ ∈ C 2 with bounded derivatives one has d d τ S ( τ ) ϕ = S ( τ ) Δ ϕ \frac{d}{d\tau}S(\tau)\phi=S(\tau)\Delta\phi d τ d S ( τ ) ϕ = S ( τ ) Δ ϕ uniformly, hence Θ ( t ) : = ∫ g S ( t − s ) [ ψ ( t ) ] d x \Theta(t):=\int gS(t-s)[\psi(t)]\,dx Θ ( t ) := ∫ g S ( t − s ) [ ψ ( t )] d x is continuous on [ s , T ) [s,T) [ s , T ) , continuously differentiable on ( s , T ) (s,T) ( s , T ) with Θ ′ ( t ) = ∫ g S ( t − s ) [ L ψ ( t ) ] d x \Theta'(t)=\int gS(t-s)[L\psi(t)]\,dx Θ ′ ( t ) = ∫ g S ( t − s ) [ L ψ ( t )] d x (differentiation under the integral sign is justified by g ∈ L 1 g\in L^1 g ∈ L 1 and the uniform bounds on the derivatives of ψ \psi ψ ), Θ ( s ) = ∫ g ψ ( s ) d x \Theta(s)=\int g\psi(s)\,dx Θ ( s ) = ∫ g ψ ( s ) d x because S ( τ ) ψ ( s ) → ψ ( s ) S(\tau)\psi(s)\to\psi(s) S ( τ ) ψ ( s ) → ψ ( s ) uniformly as τ → 0 \tau\to0 τ → 0 , and Θ ( t ) = 0 \Theta(t)=0 Θ ( t ) = 0 for t ≥ T ′ t\ge T' t ≥ T ′ . Integrating Θ ′ \Theta' Θ ′ over ( s , T ′ ) (s,T') ( s , T ′ ) gives (16) .
We now integrate (14) against L ψ L\psi L ψ . Since ∥ S ( t − s ) f ( s ) ∥ 1 ≤ ∥ f ( s ) ∥ 1 \|S(t-s)f(s)\|_1\le\|f(s)\|_1 ∥ S ( t − s ) f ( s ) ∥ 1 ≤ ∥ f ( s ) ∥ 1 and ∫ 0 T ′ ∥ f ( s ) ∥ 1 d s < ∞ \int_0^{T'}\|f(s)\|_1\,ds<\infty ∫ 0 T ′ ∥ f ( s ) ∥ 1 d s < ∞ , Fubini’s theorem applies and (16) yields
∬ Q T u L ψ = ∫ 0 T ∫ R N S ( t ) u 0 L ψ d x d t + ∫ 0 T ( ∫ s T ∫ R N S ( t − s ) f ( s ) L ψ d x d t ) d s = − ∫ R N u 0 ψ ( 0 ) d x − ∬ Q T f ψ , \begin{aligned}
\iint_{Q_T}uL\psi=\int_0^T\int_{\mathbb{R}^N}S(t)u_0L\psi\,dx\,dt+\int_0^T\left(\int_s^T\int_{\mathbb{R}^N}S(t-s)f(s)L\psi\,dx\,dt\right)\,ds \\
&=-\int_{\mathbb{R}^N}u_0\psi(0)\,dx-\iint_{Q_T}f\psi,
\end{aligned} ∬ Q T uL ψ = ∫ 0 T ∫ R N S ( t ) u 0 L ψ d x d t + ∫ 0 T ( ∫ s T ∫ R N S ( t − s ) f ( s ) L ψ d x d t ) d s = − ∫ R N u 0 ψ ( 0 ) d x − ∬ Q T f ψ ,
which is (13) ; moreover u ∈ L l o c p ( R N × [ 0 , T ) ) u\in L^p_{\mathrm{loc}}(\mathbb{R}^N\times[0,T)) u ∈ L loc p ( R N × [ 0 , T )) since u u u is bounded on Q T ′ Q_{T'} Q T ′ . This proves (a). Integrating (14) over R N \mathbb{R}^N R N and using Lemma 2.1(c) and Fubini’s theorem gives (b). For (c), the semigroup property and Fubini’s theorem (all integrands are absolutely integrable) give S ( t − τ ) u ( τ ) = S ( t ) u 0 + ∫ 0 τ S ( t − s ) f ( s ) d s S(t-\tau)u(\tau)=S(t)u_0+\int_0^\tau S(t-s)f(s)\,ds S ( t − τ ) u ( τ ) = S ( t ) u 0 + ∫ 0 τ S ( t − s ) f ( s ) d s , and it remains to add ∫ τ t S ( t − s ) f ( s ) d s \int_\tau^tS(t-s)f(s)\,ds ∫ τ t S ( t − s ) f ( s ) d s .
Lemma 2.3. Two mild solutions of (1) on ( 0 , T ) (0,T) ( 0 , T ) coincide.
Proof. Let u 1 , u 2 u_1,u_2 u 1 , u 2 be mild solutions, fix T ′ < T T'<T T ′ < T , and let K K K bound ∣ u 1 ∣ |u_1| ∣ u 1 ∣ and ∣ u 2 ∣ |u_2| ∣ u 2 ∣ on Q T ′ Q_{T'} Q T ′ . Since ∣ ∣ a ∣ p − ∣ b ∣ p ∣ ≤ p K p − 1 ∣ a − b ∣ ||a|^p-|b|^p|\le pK^{p-1}|a-b| ∣∣ a ∣ p − ∣ b ∣ p ∣ ≤ p K p − 1 ∣ a − b ∣ for ∣ a ∣ , ∣ b ∣ ≤ K |a|,|b|\le K ∣ a ∣ , ∣ b ∣ ≤ K , Lemma 2.1(a) gives ∣ u 1 − u 2 ∣ ( t ) ≤ p K p − 1 ∫ 0 t S ( t − s ) ∣ u 1 − u 2 ∣ ( s ) d s |u_1-u_2|(t)\le pK^{p-1}\int_0^tS(t-s)|u_1-u_2|(s)\,ds ∣ u 1 − u 2 ∣ ( t ) ≤ p K p − 1 ∫ 0 t S ( t − s ) ∣ u 1 − u 2 ∣ ( s ) d s , hence ϕ ( t ) : = ∥ ( u 1 − u 2 ) ( t ) ∥ ∞ \phi(t):=\|(u_1-u_2)(t)\|_\infty ϕ ( t ) := ∥ ( u 1 − u 2 ) ( t ) ∥ ∞ satisfies ϕ ( t ) ≤ p K p − 1 ∫ 0 t ϕ ( s ) d s \phi(t)\le pK^{p-1}\int_0^t\phi(s)\,ds ϕ ( t ) ≤ p K p − 1 ∫ 0 t ϕ ( s ) d s for t < T ′ t<T' t < T ′ . Iterating this inequality n n n times yields ϕ ( t ) ≤ ( p K p − 1 t ) n ∥ ϕ ∥ L ∞ ( 0 , T ′ ) / n ! \phi(t)\le(pK^{p-1}t)^n\|\phi\|_{L^\infty(0,T')}/n! ϕ ( t ) ≤ ( p K p − 1 t ) n ∥ ϕ ∥ L ∞ ( 0 , T ′ ) / n ! , and letting n → ∞ n\to\infty n → ∞ gives ϕ ≡ 0 \phi\equiv0 ϕ ≡ 0 on ( 0 , T ′ ) (0,T') ( 0 , T ′ ) .
The next lemma is the variation-of-constants formula with an additional linear damping, which turns monotone nonlinearities into order-preserving maps.
Lemma 2.4. Let 0 < T ′ < ∞ 0<T'<\infty 0 < T ′ < ∞ , λ ≥ 0 \lambda\ge0 λ ≥ 0 , v 0 ∈ L ∞ ( R N ) v_0\in L^\infty(\mathbb{R}^N) v 0 ∈ L ∞ ( R N ) , and let F F F be measurable on Q T ′ Q_{T'} Q T ′ with ∫ 0 T ′ ∥ F ( s ) ∥ ∞ d s < ∞ \int_0^{T'}\|F(s)\|_\infty\,ds<\infty ∫ 0 T ′ ∥ F ( s ) ∥ ∞ d s < ∞ . For a bounded measurable function v v v on Q T ′ Q_{T'} Q T ′ , the identity
v ( t ) = S ( t ) v 0 + ∫ 0 t S ( t − s ) F ( s ) d s a.e. in Q T ′ (17) v(t)=S(t)v_0+\int_0^tS(t-s)F(s)\,ds\qquad\text{a.e. in }Q_{T'} \tag*{(17)} v ( t ) = S ( t ) v 0 + ∫ 0 t S ( t − s ) F ( s ) d s a.e. in Q T ′ (17) holds if and only if
v ( t ) = e − λ t S ( t ) v 0 + ∫ 0 t e − λ ( t − s ) S ( t − s ) ( F ( s ) + λ v ( s ) ) d s a.e. in Q T ′ . (18) v(t)=e^{-\lambda t}S(t)v_0+\int_0^t e^{-\lambda(t-s)}S(t-s)\bigl(F(s)+\lambda v(s)\bigr)\,ds\qquad\text{a.e. in }Q_{T'}. \tag*{(18)} v ( t ) = e − λ t S ( t ) v 0 + ∫ 0 t e − λ ( t − s ) S ( t − s ) ( F ( s ) + λ v ( s ) ) d s a.e. in Q T ′ . (18)
Proof. Assume (17) . Inserting it in the term containing λ v \lambda v λ v , using the semigroup property and Fubini’s theorem (justified by ∫ 0 T ′ ∥ F ( s ) ∥ ∞ d s < ∞ \int_0^{T'} \|F(s)\|_\infty\,ds<\infty ∫ 0 T ′ ∥ F ( s ) ∥ ∞ d s < ∞ ), we find
λ ∫ 0 t e − λ ( t − s ) S ( t − s ) v ( s ) d s = ( 1 − e − λ t ) S ( t ) v 0 + ∫ 0 t ( 1 − e − λ ( t − r ) ) S ( t − r ) F ( r ) d r , \lambda\int_0^t e^{-\lambda(t-s)}S(t-s)v(s)\,ds=(1-e^{-\lambda t})S(t)v_0+\int_0^t(1-e^{-\lambda(t-r)})S(t-r)F(r)\,dr, λ ∫ 0 t e − λ ( t − s ) S ( t − s ) v ( s ) d s = ( 1 − e − λ t ) S ( t ) v 0 + ∫ 0 t ( 1 − e − λ ( t − r ) ) S ( t − r ) F ( r ) d r ,
because λ ∫ r t e − λ ( t − s ) d s = 1 − e − λ ( t − r ) \lambda\int_r^t e^{-\lambda(t-s)}\,ds=1-e^{-\lambda(t-r)} λ ∫ r t e − λ ( t − s ) d s = 1 − e − λ ( t − r ) . Adding e − λ t S ( t ) v 0 + ∫ 0 t e − λ ( t − s ) S ( t − s ) F ( s ) d s e^{-\lambda t}S(t)v_0+\int_0^t e^{-\lambda(t-s)}S(t-s)F(s)\,ds e − λ t S ( t ) v 0 + ∫ 0 t e − λ ( t − s ) S ( t − s ) F ( s ) d s to both sides gives (18) . Conversely, assume (18) and let v ~ \widetilde v v be the right-hand side of (17) , which is bounded on Q T ′ Q_{T'} Q T ′ by ∥ v 0 ∥ ∞ + ∥ F ∥ ∞ \|v_0\|_\infty+\|F\|_\infty ∥ v 0 ∥ ∞ + ∥ F ∥ ∞ . By the first part, v ~ \widetilde v v satisfies (18) , so that z : = v − v ~ z:=v-\widetilde v z := v − v satisfies z ( t ) = λ ∫ 0 t e − λ ( t − s ) S ( t − s ) z ( s ) d s z(t)=\lambda\int_0^t e^{-\lambda(t-s)}S(t-s)z(s)\,ds z ( t ) = λ ∫ 0 t e − λ ( t − s ) S ( t − s ) z ( s ) d s and therefore ∥ z ( t ) ∥ ∞ ≤ λ ∫ 0 t ∥ z ( s ) ∥ ∞ d s \|z(t)\|_\infty\le\lambda\int_0^t\|z(s)\|_\infty\,ds ∥ z ( t ) ∥ ∞ ≤ λ ∫ 0 t ∥ z ( s ) ∥ ∞ d s . As in the proof of Lemma 2.3, iteration gives z = 0 z=0 z = 0 .
Cutoff functions
We fix once and for all a function
η ∈ C ∞ ( [ 0 , ∞ ) ) , 0 ≤ η ≤ 1 , η ≡ 1 on [ 0 , 1 ] , η ≡ 0 on [ 2 , ∞ ) , η ′ ≤ 0. (19) \eta\in C^\infty([0,\infty)),\qquad0\le\eta\le1,\qquad\eta\equiv1\text{ on }[0,1],\qquad\eta\equiv0\text{ on }[2,\infty),\qquad\eta'\le0. \tag*{(19)} η ∈ C ∞ ([ 0 , ∞ )) , 0 ≤ η ≤ 1 , η ≡ 1 on [ 0 , 1 ] , η ≡ 0 on [ 2 , ∞ ) , η ′ ≤ 0. (19)
We shall use two families of test functions built on η \eta η . The separated cutoffs , centred at x 0 ∈ R N x_0\in\mathbb{R}^N x 0 ∈ R N , are
Ψ R , T ( x , t ) : = η ( ∣ x − x 0 ∣ 2 R 2 ) k η ( t T ) k ∈ C c 2 , 1 ( R N × [ 0 , ∞ ) ) , Ψ R , T ( x , 0 ) = η ( ∣ x − x 0 ∣ 2 R 2 ) k , (20) \Psi_{R,T}(x,t):=\eta\left(\frac{|x-x_0|^2}{R^2}\right)^k\eta\left(\frac{t}{T}\right)^k\in C_c^{2,1}(\mathbb{R}^N\times[0,\infty)),\qquad\Psi_{R,T}(x,0)=\eta\left(\frac{|x-x_0|^2}{R^2}\right)^k, \tag*{(20)} Ψ R , T ( x , t ) := η ( R 2 ∣ x − x 0 ∣ 2 ) k η ( T t ) k ∈ C c 2 , 1 ( R N × [ 0 , ∞ )) , Ψ R , T ( x , 0 ) = η ( R 2 ∣ x − x 0 ∣ 2 ) k , (20)
and the parabolic cutoffs are
Ψ R ( x , t ) : = η ( s R ( x , t ) ) k , s R ( x , t ) : = ∣ x ∣ 2 + t R 2 , Ψ R ( x , 0 ) = η ( ∣ x ∣ 2 R 2 ) k . (21) \Psi_R(x,t):=\eta(s_R(x,t))^k,\qquad s_R(x,t):=\frac{|x|^2+t}{R^2},\qquad\Psi_R(x,0)=\eta\left(\frac{|x|^2}{R^2}\right)^k. \tag*{(21)} Ψ R ( x , t ) := η ( s R ( x , t ) ) k , s R ( x , t ) := R 2 ∣ x ∣ 2 + t , Ψ R ( x , 0 ) = η ( R 2 ∣ x ∣ 2 ) k . (21)
The choice η ( 0 ) = 1 \eta(0)=1 η ( 0 ) = 1 is essential: it makes the boundary term in (13) nontrivial and lets the forcing contribute its full impulse. In both lemmas below the integrands ∣ L Ψ ∣ p ′ Ψ 1 − p ′ |\mathcal{L}\Psi|^{p'}\Psi^{1-p'} ∣ L Ψ ∣ p ′ Ψ 1 − p ′ are understood to vanish where Ψ = 0 \Psi=0 Ψ = 0 ; this is consistent, since L Ψ \mathcal{L}\Psi L Ψ contains the factor η k − 2 \eta^{k-2} η k − 2 and therefore vanishes wherever Ψ \Psi Ψ does.
Lemma 2.5. There is a constant C = C ( N , p , η ) > 0 C=C(N,p,\eta)>0 C = C ( N , p , η ) > 0 such that for all x 0 ∈ R N x_0\in\mathbb{R}^N x 0 ∈ R N and R , T > 0 R,T>0 R , T > 0 ,
∬ Q ∣ ∂ t Ψ R , T ∣ p ′ Ψ R , T p ′ − 1 d x d t ≤ C R N T 1 − p ′ , (22) \iint_Q\frac{|\partial_t\Psi_{R,T}|^{p'}}{\Psi_{R,T}^{p'-1}}\,dx\,dt\le C R^N T^{1-p'}, \tag*{(22)} ∬ Q Ψ R , T p ′ − 1 ∣ ∂ t Ψ R , T ∣ p ′ d x d t ≤ C R N T 1 − p ′ , (22) ∬ Q ∣ Δ Ψ R , T ∣ p ′ Ψ R , T p ′ − 1 d x d t ≤ C R N − 2 p ′ T . (23) \iint_Q\frac{|\Delta\Psi_{R,T}|^{p'}}{\Psi_{R,T}^{p'-1}}\,dx\,dt\le C R^{N-2p'}T. \tag*{(23)} ∬ Q Ψ R , T p ′ − 1 ∣Δ Ψ R , T ∣ p ′ d x d t ≤ C R N − 2 p ′ T . (23) Consequently, with ∣ a + b ∣ p ′ ≤ 2 p ′ − 1 ( ∣ a ∣ p ′ + ∣ b ∣ p ′ ) |a+b|^{p'}\le2^{p'-1}(|a|^{p'}+|b|^{p'}) ∣ a + b ∣ p ′ ≤ 2 p ′ − 1 ( ∣ a ∣ p ′ + ∣ b ∣ p ′ ) ,
I R , T : = ∬ Q ∣ ∂ t Ψ R , T + Δ Ψ R , T ∣ p ′ Ψ R , T p ′ − 1 d x d t ≤ C ( R N T 1 − p ′ + R N − 2 p ′ T ) , (24) I_{R,T}:=\iint_Q\frac{|\partial_t\Psi_{R,T}+\Delta\Psi_{R,T}|^{p'}}{\Psi_{R,T}^{p'-1}}\,dx\,dt\le C\left(R^N T^{1-p'}+R^{N-2p'}T\right), \tag*{(24)} I R , T := ∬ Q Ψ R , T p ′ − 1 ∣ ∂ t Ψ R , T + Δ Ψ R , T ∣ p ′ d x d t ≤ C ( R N T 1 − p ′ + R N − 2 p ′ T ) , (24) and the parabolic balance T = R 2 T=R^2 T = R 2 yields I R , R 2 ≤ C R N + 2 − 2 p ′ = C R N − 2 p − 1 I_{R,R^2}\le C R^{N+2-2p'}=C R^{N-\frac{2}{p-1}} I R , R 2 ≤ C R N + 2 − 2 p ′ = C R N − p − 1 2 .
Proof. Write ϕ R ( x ) : = η ( ∣ x − x 0 ∣ 2 / R 2 ) \phi_R(x):=\eta(|x-x_0|^2/R^2) ϕ R ( x ) := η ( ∣ x − x 0 ∣ 2 / R 2 ) and η T ( t ) : = η ( t / T ) \eta_T(t):=\eta(t/T) η T ( t ) := η ( t / T ) , so that Ψ R , T = ϕ R k η T k \Psi_{R,T}=\phi_R^k\eta_T^k Ψ R , T = ϕ R k η T k . Since ∂ t Ψ R , T = k ϕ R k η T k − 1 η T ′ \partial_t\Psi_{R,T}=k\phi_R^k\eta_T^{k-1}\eta_T' ∂ t Ψ R , T = k ϕ R k η T k − 1 η T ′ and k = 2 p ′ k=2p' k = 2 p ′ ,
∣ ∂ t Ψ R , T ∣ p ′ Ψ R , T p ′ − 1 = k p ′ ϕ R k η T ( k − 1 ) p ′ − k ( p ′ − 1 ) ∣ η T ′ ∣ p ′ = k p ′ ϕ R k η T p ′ ∣ η T ′ ∣ p ′ ≤ k p ′ ∥ η ′ ∥ ∞ p ′ T − p ′ 1 { ∣ x − x 0 ∣ ≤ 2 R } 1 { T ≤ t ≤ 2 T } , \begin{aligned}
\frac{|\partial_t\Psi_{R,T}|^{p'}}{\Psi_{R,T}^{p'-1}}=k^{p'}\phi_R^k\eta_T^{(k-1)p'-k(p'-1)}|\eta_T'|^{p'}=k^{p'}\phi_R^k\eta_T^{p'}|\eta_T'|^{p'} \\
&\le k^{p'}\|\eta'\|_\infty^{p'}T^{-p'}\mathbf{1}_{\{|x-x_0|\le\sqrt{2}R\}}\mathbf{1}_{\{T\le t\le2T\}},
\end{aligned} Ψ R , T p ′ − 1 ∣ ∂ t Ψ R , T ∣ p ′ = k p ′ ϕ R k η T ( k − 1 ) p ′ − k ( p ′ − 1 ) ∣ η T ′ ∣ p ′ = k p ′ ϕ R k η T p ′ ∣ η T ′ ∣ p ′ ≤ k p ′ ∥ η ′ ∥ ∞ p ′ T − p ′ 1 { ∣ x − x 0 ∣ ≤ 2 R } 1 { T ≤ t ≤ 2 T } ,
because ( k − 1 ) p ′ − k ( p ′ − 1 ) = k − p ′ = p ′ (k-1)p'-k(p'-1)=k-p'=p' ( k − 1 ) p ′ − k ( p ′ − 1 ) = k − p ′ = p ′ and η T ′ ( t ) = T − 1 η ′ ( t / T ) \eta_T'(t)=T^{-1}\eta'(t/T) η T ′ ( t ) = T − 1 η ′ ( t / T ) vanishes outside [ T , 2 T ] [T,2T] [ T , 2 T ] . Integrating over a ball of radius 2 R \sqrt{2}R 2 R and an interval of length T T T proves (22) . For the Laplacian, Δ Ψ R , T = η T k Δ ( ϕ R k ) \Delta\Psi_{R,T}=\eta_T^k\Delta(\phi_R^k) Δ Ψ R , T = η T k Δ ( ϕ R k ) and
Δ ( ϕ R k ) = k ϕ R k − 1 Δ ϕ R + k ( k − 1 ) ϕ R k − 2 ∣ ∇ ϕ R ∣ 2 . \Delta(\phi_R^k)=k\phi_R^{k-1}\Delta\phi_R+k(k-1)\phi_R^{k-2}|\nabla\phi_R|^2. Δ ( ϕ R k ) = k ϕ R k − 1 Δ ϕ R + k ( k − 1 ) ϕ R k − 2 ∣∇ ϕ R ∣ 2 .
With r : = ∣ x − x 0 ∣ 2 / R 2 r:=|x-x_0|^2/R^2 r := ∣ x − x 0 ∣ 2 / R 2 we have ∇ ϕ R = 2 ( x − x 0 ) R − 2 η ′ ( r ) \nabla\phi_R=2(x-x_0)R^{-2}\eta'(r) ∇ ϕ R = 2 ( x − x 0 ) R − 2 η ′ ( r ) and Δ ϕ R = R − 2 ( 2 N η ′ ( r ) + 4 r η ′ ′ ( r ) ) \Delta\phi_R=R^{-2}(2N\eta'(r)+4r\eta''(r)) Δ ϕ R = R − 2 ( 2 N η ′ ( r ) + 4 r η ′′ ( r )) ; both vanish unless 1 ≤ r ≤ 2 1\le r\le2 1 ≤ r ≤ 2 , so that ∣ ∇ ϕ R ∣ 2 ≤ 8 ∥ η ′ ∥ ∞ 2 R − 2 |\nabla\phi_R|^2\le8\|\eta'\|_\infty^2R^{-2} ∣∇ ϕ R ∣ 2 ≤ 8∥ η ′ ∥ ∞ 2 R − 2 and ∣ Δ ϕ R ∣ ≤ ( 2 N ∥ η ′ ∥ ∞ + 8 ∥ η ′ ′ ∥ ∞ ) R − 2 |\Delta\phi_R|\le(2N\|\eta'\|_\infty+8\|\eta''\|_\infty)R^{-2} ∣Δ ϕ R ∣ ≤ ( 2 N ∥ η ′ ∥ ∞ + 8∥ η ′′ ∥ ∞ ) R − 2 . Using ϕ R ≤ 1 \phi_R\le1 ϕ R ≤ 1 we obtain ∣ Δ ( ϕ R k ) ∣ ≤ C R − 2 ϕ R k − 2 1 { R ≤ ∣ x − x 0 ∣ ≤ 2 R } |\Delta(\phi_R^k)|\le CR^{-2}\phi_R^{k-2}\mathbf{1}_{\{R\le|x-x_0|\le\sqrt{2}R\}} ∣Δ ( ϕ R k ) ∣ ≤ C R − 2 ϕ R k − 2 1 { R ≤ ∣ x − x 0 ∣ ≤ 2 R } , and after the exact cancellation ( k − 2 ) p ′ − k ( p ′ − 1 ) = k − 2 p ′ = 0 (k-2)p'-k(p'-1)=k-2p'=0 ( k − 2 ) p ′ − k ( p ′ − 1 ) = k − 2 p ′ = 0 ,
∣ Δ Ψ R , T ∣ p ′ Ψ R , T p ′ − 1 = η T k ∣ Δ ( ϕ R k ) ∣ p ′ ϕ R k ( p ′ − 1 ) ≤ C R − 2 p ′ 1 { R ≤ ∣ x − x 0 ∣ ≤ 2 R } 1 { 0 ≤ t ≤ 2 T } . \frac{|\Delta\Psi_{R,T}|^{p'}}{\Psi_{R,T}^{p'-1}}=\eta_T^k\frac{|\Delta(\phi_R^k)|^{p'}}{\phi_R^{k(p'-1)}}\le CR^{-2p'}\mathbf{1}_{\{R\le|x-x_0|\le\sqrt{2}R\}}\mathbf{1}_{\{0\le t\le2T\}}. Ψ R , T p ′ − 1 ∣Δ Ψ R , T ∣ p ′ = η T k ϕ R k ( p ′ − 1 ) ∣Δ ( ϕ R k ) ∣ p ′ ≤ C R − 2 p ′ 1 { R ≤ ∣ x − x 0 ∣ ≤ 2 R } 1 { 0 ≤ t ≤ 2 T } .
Integrating gives (23) , and (24) follows. Finally N + 2 − 2 p ′ = N − 2 / ( p − 1 ) N+2-2p'=N-2/(p-1) N + 2 − 2 p ′ = N − 2/ ( p − 1 ) because 2 p ′ − 2 = 2 / ( p − 1 ) 2p'-2=2/(p-1) 2 p ′ − 2 = 2/ ( p − 1 ) .
Lemma 2.6. For every R > 0 R>0 R > 0 the function Ψ R \Psi_R Ψ R belongs to C c 2 , 1 ( R N × [ 0 , ∞ ) ) C_c^{2,1}(\mathbb{R}^N\times[0,\infty)) C c 2 , 1 ( R N × [ 0 , ∞ )) , is supported in { ∣ x ∣ 2 + t ≤ 2 R 2 } \{|x|^2+t\le2R^2\} { ∣ x ∣ 2 + t ≤ 2 R 2 } , is nondecreasing in R R R , and Ψ R → 1 \Psi_R\to1 Ψ R → 1 pointwise as R → ∞ R\to\infty R → ∞ . Moreover there is C 1 = C 1 ( N , p , η ) C_1=C_1(N,p,\eta) C 1 = C 1 ( N , p , η ) such that
∣ L Ψ R ∣ ≤ C 1 R − 2 η ( s R ) k − 2 1 { 1 ≤ s R ≤ 2 } . (25) |\mathcal{L}\Psi_R|\le C_1R^{-2}\eta(s_R)^{k-2}\mathbf{1}_{\{1\le s_R\le2\}}. \tag*{(25)} ∣ L Ψ R ∣ ≤ C 1 R − 2 η ( s R ) k − 2 1 { 1 ≤ s R ≤ 2 } . (25)
Consequently, for every u ∈ L l o c p ( R N × [ 0 , ∞ ) ) u\in L_{\mathrm{loc}}^p(\mathbb{R}^N\times[0,\infty)) u ∈ L loc p ( R N × [ 0 , ∞ )) and R > 0 R>0 R > 0 ,
∬ Q ∣ u ∣ ∣ L Ψ R ∣ d x d t ≤ C 2 R N + 2 − 2 p ′ p ′ ( ∬ D R ∣ u ∣ p Ψ R d x d t ) 1 / p , (26) \iint_Q |u||\mathcal{L}\Psi_R|\,dx\,dt\le C_2R^{\frac{N+2-2p'}{p'}}\left(\iint_{D_R}|u|^p\Psi_R\,dx\,dt\right)^{1/p}, \tag*{(26)} ∬ Q ∣ u ∣∣ L Ψ R ∣ d x d t ≤ C 2 R p ′ N + 2 − 2 p ′ ( ∬ D R ∣ u ∣ p Ψ R d x d t ) 1/ p , (26)
where D R : = { ( x , t ) ∈ Q : R 2 ≤ ∣ x ∣ 2 + t ≤ 2 R 2 } D_R:=\{(x,t)\in Q:R^2\le|x|^2+t\le2R^2\} D R := {( x , t ) ∈ Q : R 2 ≤ ∣ x ∣ 2 + t ≤ 2 R 2 } , with C 2 = C 2 ( N , p , η ) C_2=C_2(N,p,\eta) C 2 = C 2 ( N , p , η ) , and
∬ Q ∣ L Ψ R ∣ p ′ Ψ R p ′ − 1 d x d t ≤ C 2 p ′ R N + 2 − 2 p ′ . (27) \iint_Q\frac{|\mathcal{L}\Psi_R|^{p'}}{\Psi_R^{p'-1}}\,dx\,dt\le C_2^{p'}R^{N+2-2p'}. \tag*{(27)} ∬ Q Ψ R p ′ − 1 ∣ L Ψ R ∣ p ′ d x d t ≤ C 2 p ′ R N + 2 − 2 p ′ . (27)
Proof. The monotonicity in R R R follows from η ′ ≤ 0 \eta'\le0 η ′ ≤ 0 , and Ψ R → η ( 0 ) k = 1 \Psi_R\to\eta(0)^k=1 Ψ R → η ( 0 ) k = 1 . Put g : = η k g:=\eta^k g := η k , so that Ψ R = g ( s R ) \Psi_R=g(s_R) Ψ R = g ( s R ) . Since ∂ t s R = R − 2 \partial_ts_R=R^{-2} ∂ t s R = R − 2 , ∇ s R = 2 x R − 2 \nabla s_R=2xR^{-2} ∇ s R = 2 x R − 2 and Δ s R = 2 N R − 2 \Delta s_R=2NR^{-2} Δ s R = 2 N R − 2 ,
L Ψ R = R − 2 [ ( 2 N + 1 ) g ′ ( s R ) + 4 ∣ x ∣ 2 R 2 g ′ ′ ( s R ) ] . \mathcal{L}\Psi_R=R^{-2}\left[(2N+1)g'(s_R)+4\frac{|x|^2}{R^2}g''(s_R)\right]. L Ψ R = R − 2 [ ( 2 N + 1 ) g ′ ( s R ) + 4 R 2 ∣ x ∣ 2 g ′′ ( s R ) ] .
Now g ′ = k η k − 1 η ′ g'=k\eta^{k-1}\eta' g ′ = k η k − 1 η ′ and g ′ ′ = k ( k − 1 ) η k − 2 ( η ′ ) 2 + k η k − 1 η ′ ′ g''=k(k-1)\eta^{k-2}(\eta')^2+k\eta^{k-1}\eta'' g ′′ = k ( k − 1 ) η k − 2 ( η ′ ) 2 + k η k − 1 η ′′ vanish outside [ 1 , 2 ] [1,2] [ 1 , 2 ] and are bounded by a constant times η k − 2 \eta^{k-2} η k − 2 there; since ∣ x ∣ 2 / R 2 ≤ s R ≤ 2 |x|^2/R^2\le s_R\le2 ∣ x ∣ 2 / R 2 ≤ s R ≤ 2 on the support, (25) follows. On { 1 ≤ s R ≤ 2 } \{1\le s_R\le2\} { 1 ≤ s R ≤ 2 } we write η k − 2 = ( η k ) 1 / p η k / p ′ − 2 = Ψ R 1 / p \eta^{k-2}=(\eta^k)^{1/p}\eta^{k/p'-2}=\Psi_R^{1/p} η k − 2 = ( η k ) 1/ p η k / p ′ − 2 = Ψ R 1/ p , since k / p ′ = 2 k/p'=2 k / p ′ = 2 . Hence, by Hölder’s inequality,
∬ Q ∣ u ∣ ∣ L Ψ R ∣ ≤ C 1 R − 2 ∬ D R ∣ u ∣ Ψ R 1 / p ≤ C 1 R − 2 ( ∬ D R ∣ u ∣ p Ψ R ) 1 / p ∣ D R ∣ 1 / p ′ , \iint_Q|u||\mathcal{L}\Psi_R|\le C_1R^{-2}\iint_{D_R}|u|\Psi_R^{1/p}\le C_1R^{-2}\left(\iint_{D_R}|u|^p\Psi_R\right)^{1/p}|D_R|^{1/p'}, ∬ Q ∣ u ∣∣ L Ψ R ∣ ≤ C 1 R − 2 ∬ D R ∣ u ∣ Ψ R 1/ p ≤ C 1 R − 2 ( ∬ D R ∣ u ∣ p Ψ R ) 1/ p ∣ D R ∣ 1/ p ′ ,
and ∣ D R ∣ ≤ ∣ B 1 ∣ ( 2 R ) N ⋅ 2 R 2 |D_R|\le|B_1|(\sqrt{2}R)^N\cdot2R^2 ∣ D R ∣ ≤ ∣ B 1 ∣ ( 2 R ) N ⋅ 2 R 2 . This gives (26) because − 2 + ( N + 2 ) / p ′ = ( N + 2 − 2 p ′ ) / p ′ -2+(N+2)/p'=(N+2-2p')/p' − 2 + ( N + 2 ) / p ′ = ( N + 2 − 2 p ′ ) / p ′ . Similarly,
∣ L Ψ R ∣ p ′ Ψ R − p ′ ≤ C 1 p ′ R − 2 p ′ η ( s R ) ( k − 2 ) p ′ − k ( p ′ − 1 ) 1 D R = C 1 p ′ R − 2 p ′ 1 D R , |\mathcal{L}\Psi_R|^{p'}\Psi_R^{-p'}\le C_1^{p'}R^{-2p'}\eta(s_R)^{(k-2)p'-k(p'-1)}\mathbf{1}_{D_R}=C_1^{p'}R^{-2p'}\mathbf{1}_{D_R}, ∣ L Ψ R ∣ p ′ Ψ R − p ′ ≤ C 1 p ′ R − 2 p ′ η ( s R ) ( k − 2 ) p ′ − k ( p ′ − 1 ) 1 D R = C 1 p ′ R − 2 p ′ 1 D R ,
which gives (27) .
The forcing kernel
The behavior of the linear forced flow is governed by the function ∣ ζ ∣ ⋆ h N |\zeta|\star h_N ∣ ζ ∣ ⋆ h N defined in (6) . Besides ( H a ) (\mathrm{H}_a) ( H a ) we shall use the following weaker hypothesis:
( H K ) ( ∣ ζ ∣ ⋆ h N ) ( t ) ≤ K h N ( t ) for all t > 0 , for some constant K ≥ 0. (28) (\mathrm{H}_K)\qquad(|\zeta|\star h_N)(t)\le Kh_N(t)\quad\text{for all }t>0,\ \text{for some constant }K\ge0. \tag*{(28)} ( H K ) ( ∣ ζ ∣ ⋆ h N ) ( t ) ≤ K h N ( t ) for all t > 0 , for some constant K ≥ 0. (28)
Lemma 2.7.
If ζ \zeta ζ satisfies ( H a ) (\mathrm{H}_a) ( H a ) , then ζ \zeta ζ satisfies ( H K ) (\mathrm{H}_K) ( H K ) and
lim t → ∞ t N / 2 ∫ t / 2 t ∣ ζ ( s ) ∣ h N ( t − s ) d s = 0. (29) \lim_{t\to\infty}t^{N/2}\int_{t/2}^{t}|\zeta(s)|h_N(t-s)\,ds=0. \tag*{(29)} t → ∞ lim t N /2 ∫ t /2 t ∣ ζ ( s ) ∣ h N ( t − s ) d s = 0. (29)
(ii) If N ≥ 3 N \ge3 N ≥ 3 , ζ ∈ L 1 ( 0 , ∞ ) \zeta\in L^1(0,\infty) ζ ∈ L 1 ( 0 , ∞ ) and ∣ ζ ( t ) ∣ ≤ C t − b |\zeta(t)| \le Ct^{-b} ∣ ζ ( t ) ∣ ≤ C t − b for a.e. t ≥ 1 t \ge1 t ≥ 1 with b > 1 b > 1 b > 1 , then ( ∣ ζ ∣ ⋆ h N ) ( t ) ≤ C ′ ( 1 + t ) − min { b , N / 2 } (|\zeta| \star h_N)(t) \le C'(1+t)^{-\min\{b,N/2\}} ( ∣ ζ ∣ ⋆ h N ) ( t ) ≤ C ′ ( 1 + t ) − m i n { b , N /2 } for all t > 0 t > 0 t > 0 .
(iii) If q ≥ 1 q \ge1 q ≥ 1 and ζ ∈ L 1 ( 0 , ∞ ) \zeta\in L^1(0,\infty) ζ ∈ L 1 ( 0 , ∞ ) , then ∥ ∣ ζ ∣ ⋆ h N ∥ L q ( 0 , ∞ ) ≤ ∥ ζ ∥ L 1 ∥ h N ∥ L q ( 0 , ∞ ) \lVert|\zeta| \star h_N\rVert_{L^q(0,\infty)} \le\lVert\zeta\rVert_{L^1}\lVert h_N\rVert_{L^q(0,\infty)} ∥ ∣ ζ ∣ ⋆ h N ∥ L q ( 0 , ∞ ) ≤ ∥ ζ ∥ L 1 ∥ h N ∥ L q ( 0 , ∞ ) .
Proof. (i) For t ≥ 2 t \ge2 t ≥ 2 we split the integral defining ( ∣ ζ ∣ ⋆ h N ) ( t ) (|\zeta| \star h_N)(t) ( ∣ ζ ∣ ⋆ h N ) ( t ) at s = t / 2 s=t/2 s = t /2 . On [ 0 , t / 2 ] [0,t/2] [ 0 , t /2 ] we have t − s ≥ t / 2 t-s \ge t/2 t − s ≥ t /2 , so that by Lemma 2.1 (f)
∫ 0 t / 2 ∣ ζ ( s ) ∣ h N ( t − s ) d s ≤ h N ( t / 2 ) ∥ ζ ∥ 1 ≤ 2 N / 2 ∥ ζ ∥ 1 h N ( t ) . \int_0^{t/2}|\zeta(s)|h_N(t-s)\,\mathrm{d}s \le h_N(t/2)\lVert\zeta\rVert_1 \le2^{N/2}\lVert\zeta\rVert_1h_N(t). ∫ 0 t /2 ∣ ζ ( s ) ∣ h N ( t − s ) d s ≤ h N ( t /2 ) ∥ ζ ∥ 1 ≤ 2 N /2 ∥ ζ ∥ 1 h N ( t ) .
On [ t / 2 , t ] [t/2,t] [ t /2 , t ] we have s ≥ 1 s \ge1 s ≥ 1 and ∣ ζ ( s ) ∣ ≤ C ζ 2 a t − a |\zeta(s)| \le C\zeta2^a t^{-a} ∣ ζ ( s ) ∣ ≤ C ζ 2 a t − a , hence
∫ t / 2 t ∣ ζ ( s ) ∣ h N ( t − s ) d s ≤ C ζ 2 a t − a ∫ 0 t / 2 h N ( r ) d r . \int_{t/2}^{t}|\zeta(s)|h_N(t-s)\,\mathrm{d}s \le C\zeta2^a t^{-a}\int_0^{t/2}h_N(r)\,\mathrm{d}r. ∫ t /2 t ∣ ζ ( s ) ∣ h N ( t − s ) d s ≤ C ζ 2 a t − a ∫ 0 t /2 h N ( r ) d r .
The last integral is bounded by 1 + ( t / 2 π ) 1 / 2 1+(t/2\pi)^{1/2} 1 + ( t /2 π ) 1/2 if N = 1 N=1 N = 1 , by C ( 1 + log ( 1 + t ) ) C(1+\log(1+t)) C ( 1 + log ( 1 + t )) if N = 2 N=2 N = 2 , and by a constant if N ≥ 3 N \ge3 N ≥ 3 . Since a > 1 a>1 a > 1 when N ≤ 2 N\le2 N ≤ 2 and a > N / 2 a>N/2 a > N /2 when N ≥ 3 N\ge3 N ≥ 3 , we obtain (29) . In particular ( ∣ ζ ∣ ⋆ h N ) ( t ) (|\zeta| \star h_N)(t) ( ∣ ζ ∣ ⋆ h N ) ( t ) is bounded on [ 2 , ∞ ) [2,\infty) [ 2 , ∞ ) , and since h N ( t ) = ( 4 π t ) − N / 2 h_N(t)=(4\pi t)^{-N/2} h N ( t ) = ( 4 π t ) − N /2 there, ( ∣ ζ ∣ ⋆ h N ) ≤ K 1 h N (|\zeta| \star h_N)\le K_1h_N ( ∣ ζ ∣ ⋆ h N ) ≤ K 1 h N on [ 2 , ∞ ) [2,\infty) [ 2 , ∞ ) . On ( 0 , 2 ] (0,2] ( 0 , 2 ] we simply use ( ∣ ζ ∣ ⋆ h N ) ( t ) ≤ ∥ ζ ∥ 1 ≤ ∥ ζ ∥ 1 h N ( 2 ) − 1 h N ( t ) (|\zeta| \star h_N)(t)\le\lVert\zeta\rVert_1\le\lVert\zeta\rVert_1h_N(2)^{-1}h_N(t) ( ∣ ζ ∣ ⋆ h N ) ( t ) ≤ ∥ ζ ∥ 1 ≤ ∥ ζ ∥ 1 h N ( 2 ) − 1 h N ( t ) .
(ii) The same splitting gives, for t ≥ 2 t\ge2 t ≥ 2 , ( ∣ ζ ∣ ⋆ h N ) ( t ) ≤ 2 N / 2 ∥ ζ ∥ 1 h N ( t ) + C 2 − b t − b ∥ h N ∥ L 1 ( 0 , ∞ ) (|\zeta| \star h_N)(t)\le2^{N/2}\lVert\zeta\rVert_1h_N(t)+C2^{-b}t^{-b}\lVert h_N\rVert_{L^1(0,\infty)} ( ∣ ζ ∣ ⋆ h N ) ( t ) ≤ 2 N /2 ∥ ζ ∥ 1 h N ( t ) + C 2 − b t − b ∥ h N ∥ L 1 ( 0 , ∞ ) , and h N ∈ L 1 ( 0 , ∞ ) h_N\in L^1(0,\infty) h N ∈ L 1 ( 0 , ∞ ) because N ≥ 3 N\ge3 N ≥ 3 ; for t ≤ 2 t\le2 t ≤ 2 the bound ∥ ζ ∥ 1 \lVert\zeta\rVert_1 ∥ ζ ∥ 1 suffices.
(iii) This is Young’s inequality on the real line, applied to ∣ ζ ∣ 1 ( 0 , ∞ ) |\zeta|\mathbf{1}_{(0,\infty)} ∣ ζ ∣ 1 ( 0 , ∞ ) and h N 1 ( 0 , ∞ ) h_N\mathbf{1}_{(0,\infty)} h N 1 ( 0 , ∞ ) , whose convolution coincides with ∣ ζ ∣ ⋆ h N |\zeta| \star h_N ∣ ζ ∣ ⋆ h N on ( 0 , ∞ ) (0,\infty) ( 0 , ∞ ) . □
Mass annihilation in the Fujita range
Proof of Theorem 1.1
Let u u u be a global weak solution and fix R ≥ 1 R\ge1 R ≥ 1 . By Remark 2.1 and Lemma 2.6 we may test (13) with the parabolic cutoff ψ = Ψ R \psi=\Psi_R ψ = Ψ R of (21) . Since Ψ R ( x , 0 ) = η ( ∣ x ∣ 2 / R 2 ) k \Psi_R(x,0)=\eta(|x|^2/R^2)^k Ψ R ( x , 0 ) = η ( ∣ x ∣ 2 / R 2 ) k , the identity reads
∬ Q ∣ u ∣ p Ψ R ⏟ = : J ( R ) + ∬ Q ζ w Ψ R ⏟ = : F R + ∫ R N u 0 η ( ∣ x ∣ 2 / R 2 ) k d x ⏟ = : I R = − ∬ Q u L Ψ R . (30) \underbrace{\iint_Q |u|^p\Psi_R}_{=:J(R)}
+\underbrace{\iint_Q \zeta w\Psi_R}_{=:F_R}
+\underbrace{\int_{\mathbb{R}^N}u_0\eta(|x|^2/R^2)^k\,\mathrm{d}x}_{=:I_R}
=-\iint_Q u\mathcal{L}\Psi_R. \tag*{(30)} =: J ( R ) ∬ Q ∣ u ∣ p Ψ R + =: F R ∬ Q ζ w Ψ R + =: I R ∫ R N u 0 η ( ∣ x ∣ 2 / R 2 ) k d x = − ∬ Q u L Ψ R . (30)
All four quantities are finite because Ψ R \Psi_R Ψ R has compact support.
The data terms. Since 0 ≤ Ψ R ≤ 1 0\le\Psi_R\le1 0 ≤ Ψ R ≤ 1 , Ψ R → 1 \Psi_R\to1 Ψ R → 1 pointwise, u 0 ∈ L 1 ( R N ) u_0\in L^1(\mathbb{R}^N) u 0 ∈ L 1 ( R N ) and ζ ⊗ w ∈ L 1 ( Q ) \zeta\otimes w\in L^1(Q) ζ ⊗ w ∈ L 1 ( Q ) (with ∥ ζ ⊗ w ∥ L 1 ( Q ) = ∥ ζ ∥ 1 ∥ w ∥ 1 \lVert\zeta\otimes w\rVert_{L^1(Q)}=\lVert\zeta\rVert_1\lVert w\rVert_1 ∥ ζ ⊗ w ∥ L 1 ( Q ) = ∥ ζ ∥ 1 ∥ w ∥ 1 ), dominated convergence gives
I R → R → ∞ m 0 , F R → R → ∞ ∬ Q ζ ( t ) w ( x ) d x d t = Z W , (31) I_R \xrightarrow[R\to\infty]{} m_0,\qquad
F_R \xrightarrow[R\to\infty]{} \iint_Q \zeta(t)w(x)\,\mathrm{d}x\,\mathrm{d}t=ZW, \tag*{(31)} I R R → ∞ m 0 , F R R → ∞ ∬ Q ζ ( t ) w ( x ) d x d t = Z W , (31)
and therefore
F R + I R → R → ∞ m 0 + Z W = M , ∣ F R + I R ∣ ≤ K 0 : = ∥ u 0 ∥ 1 + ∥ ζ ∥ 1 ∥ w ∥ 1 . (32) F_R+I_R \xrightarrow[R\to\infty]{}m_0+ZW=M,\qquad|F_R+I_R|\le K_0:=\lVert u_0\rVert_1+\lVert\zeta\rVert_1\lVert w\rVert_1. \tag*{(32)} F R + I R R → ∞ m 0 + Z W = M , ∣ F R + I R ∣ ≤ K 0 := ∥ u 0 ∥ 1 + ∥ ζ ∥ 1 ∥ w ∥ 1 . (32)
A quantitative inequality. By (26) and Young’s inequality, for every R > 0 R>0 R > 0 ,
∣ ∬ Q u L Ψ R ∣ ≤ C 2 R N + 2 − 2 p ′ p ′ J ( R ) 1 / p ≤ 1 p J ( R ) + C 2 p ′ p ′ R N + 2 − 2 p ′ . \left|\iint_Q u\mathcal{L}\Psi_R\right|\le C_2R^{\frac{N+2-2p'}{p'}}J(R)^{1/p}\le\frac{1}{p}J(R)+\frac{C_2^{p'}}{p'}R^{N+2-2p'}. ∬ Q u L Ψ R ≤ C 2 R p ′ N + 2 − 2 p ′ J ( R ) 1/ p ≤ p 1 J ( R ) + p ′ C 2 p ′ R N + 2 − 2 p ′ .
Inserting this into (30) and using 1 − 1 p = 1 p ′ 1-\frac{1}{p}=\frac{1}{p'} 1 − p 1 = p ′ 1 we obtain, for every p > 1 p>1 p > 1 and every R > 0 R>0 R > 0 ,
1 p ′ J ( R ) + F R + I R ≤ C 2 p ′ p ′ R N + 2 − 2 p ′ . (33) \frac{1}{p'}J(R)+F_R+I_R\le\frac{C_2^{p'}}{p'}R^{N+2-2p'}. \tag*{(33)} p ′ 1 J ( R ) + F R + I R ≤ p ′ C 2 p ′ R N + 2 − 2 p ′ . (33)
When 1 < p < p F 1<p<p_F 1 < p < p F we have N + 2 − 2 p ′ < 0 N+2-2p'<0 N + 2 − 2 p ′ < 0 , and letting R → ∞ R\to\infty R → ∞ in (33) already gives M ≤ 0 M\le0 M ≤ 0 ; to obtain the identity (7) , and to include p = p F p=p_F p = p F , we argue as follows. Integrability of ∣ u ∣ p |u|^{p} ∣ u ∣ p . Since p ≤ p F p \le p_F p ≤ p F means N + 2 − 2 p ′ ≤ 0 N+2-2p' \le0 N + 2 − 2 p ′ ≤ 0 , we have R N + 2 − 2 p ′ ≤ 1 R^{N+2-2p'} \le1 R N + 2 − 2 p ′ ≤ 1 for R ≥ 1 R \ge1 R ≥ 1 , and (33) together with (32) gives J ( R ) ≤ p ′ K 0 + C 2 p ′ = : K 1 J(R) \le p'K_0+C_2^{p'}=:K_1 J ( R ) ≤ p ′ K 0 + C 2 p ′ =: K 1 for all R ≥ 1 R \ge1 R ≥ 1 . As R ↗ ∞ R \nearrow\infty R ↗ ∞ the cutoffs Ψ R \Psi_R Ψ R increase to 1 (Lemma 2.6), so monotone convergence yields
∬ Q ∣ u ∣ p d x d t ≤ K 1 < ∞ , J ( R ) → R → ∞ ∬ Q ∣ u ∣ p d x d t . (34) \iint_Q |u|^p\,dx\,dt \le K_1<\infty,\qquad J(R)\xrightarrow[R\to\infty]{}\iint_Q |u|^p\,dx\,dt. \tag*{(34)} ∬ Q ∣ u ∣ p d x d t ≤ K 1 < ∞ , J ( R ) R → ∞ ∬ Q ∣ u ∣ p d x d t . (34)
Vanishing of the diffusive term. The function L Ψ R \mathcal{L}\Psi_R L Ψ R is supported in the annular region D R = { R 2 ≤ ∣ x ∣ 2 + t ≤ 2 R 2 } D_R=\{R^2\le|x|^2+t\le2R^2\} D R = { R 2 ≤ ∣ x ∣ 2 + t ≤ 2 R 2 } , which escapes to infinity as R → ∞ R\to\infty R → ∞ . By (34) and dominated convergence,
∬ D R ∣ u ∣ p Ψ R d x d t ≤ ∬ { ∣ x ∣ 2 + t ≥ R 2 } ∣ u ∣ p d x d t → R → ∞ 0. (35) \iint_{D_R}|u|^p\Psi_R\,dx\,dt\le\iint_{\{|x|^2+t\ge R^2\}}|u|^p\,dx\,dt\xrightarrow[R\to\infty]{}0. \tag*{(35)} ∬ D R ∣ u ∣ p Ψ R d x d t ≤ ∬ { ∣ x ∣ 2 + t ≥ R 2 } ∣ u ∣ p d x d t R → ∞ 0. (35)
Using (26) once more, with R ( N + 2 − 2 p ′ ) / p ′ ≤ 1 R^{(N+2-2p')/p'}\le1 R ( N + 2 − 2 p ′ ) / p ′ ≤ 1 ,
∣ ∬ Q u L Ψ R ∣ ≤ C 2 ( ∬ D R ∣ u ∣ p Ψ R ) 1 / p → R → ∞ 0. \left|\iint_Q u\,\mathcal{L}\Psi_R\right|\le C_2\left(\iint_{D_R}|u|^p\Psi_R\right)^{1/p}\xrightarrow[R\to\infty]{}0. ∬ Q u L Ψ R ≤ C 2 ( ∬ D R ∣ u ∣ p Ψ R ) 1/ p R → ∞ 0.
Conclusion. Letting R → ∞ R\to\infty R → ∞ in (30) and using (32) , (34) and the last display, we obtain ∬ Q ∣ u ∣ p + M = 0 \iint_Q |u|^p+M=0 ∬ Q ∣ u ∣ p + M = 0 , which is (7) .
Proof of Theorem 1.2
Statements (i) and (iii) are immediate consequences of (7) , whose left-hand side is nonnegative. Assume M = 0 M=0 M = 0 and let u u u be a global weak solution. Then (7) gives u = 0 u=0 u = 0 a.e. in Q Q Q , and (13) reduces to
∬ Q ζ ( t ) w ( x ) ψ d x d t + ∫ R N u 0 ψ ( x , 0 ) d x = 0 for all ψ ∈ C c ∞ ( R N × [ 0 , ∞ ) ) . \iint_Q \zeta(t)w(x)\psi\,dx\,dt+\int_{\mathbb{R}^N}u_0\psi(x,0)\,dx=0\qquad\text{for all }\psi\in C_c^\infty(\mathbb{R}^N\times[0,\infty)). ∬ Q ζ ( t ) w ( x ) ψ d x d t + ∫ R N u 0 ψ ( x , 0 ) d x = 0 for all ψ ∈ C c ∞ ( R N × [ 0 , ∞ )) .
Choosing first ψ ∈ C c ∞ ( Q ) \psi\in C_c^\infty(Q) ψ ∈ C c ∞ ( Q ) , which vanishes near t = 0 t=0 t = 0 , we see that the locally integrable function ζ ⊗ w \zeta\otimes w ζ ⊗ w vanishes as a distribution in Q Q Q , hence a.e. Choosing then ψ ( x , t ) = φ ( x ) χ ( t ) \psi(x,t)=\varphi(x)\chi(t) ψ ( x , t ) = φ ( x ) χ ( t ) with φ ∈ C c ∞ ( R N ) \varphi\in C_c^\infty(\mathbb{R}^N) φ ∈ C c ∞ ( R N ) and χ ∈ C ∞ ( [ 0 , ∞ ) ) \chi\in C^\infty([0,\infty)) χ ∈ C ∞ ([ 0 , ∞ )) , χ ( 0 ) = 1 \chi(0)=1 χ ( 0 ) = 1 , we get ∫ u 0 φ = 0 \int u_0\varphi=0 ∫ u 0 φ = 0 for all φ \varphi φ , that is u 0 = 0 u_0=0 u 0 = 0 a.e. Conversely, if u 0 = 0 u_0=0 u 0 = 0 and ζ ⊗ w = 0 \zeta\otimes w=0 ζ ⊗ w = 0 a.e., then u ≡ 0 u\equiv0 u ≡ 0 is a global weak solution, and any global weak solution satisfies ∬ Q ∣ u ∣ p = − M = 0 \iint_Q |u|^p=-M=0 ∬ Q ∣ u ∣ p = − M = 0 by (7) , so that u ≡ 0 u\equiv0 u ≡ 0 is the only one. By Fubini’s theorem, ζ ( t ) w ( x ) = 0 \zeta(t)w(x)=0 ζ ( t ) w ( x ) = 0 a.e. in Q Q Q if and only if ζ = 0 \zeta=0 ζ = 0 a.e. in ( 0 , ∞ ) (0,\infty) ( 0 , ∞ ) or w = 0 w=0 w = 0 a.e. in R N \mathbb{R}^N R N .
Comments
Remark 3.1. Theorems 1.1 and 1.2 isolate exactly what an integrable forcing does. The critical exponent is the Fujita exponent p F p_F p F of the unforced source equation ∂ t u − Δ u = ∣ u ∣ p \partial_tu-\Delta u=|u|^p ∂ t u − Δ u = ∣ u ∣ p ; the forcing does not change it. What the forcing changes is the admissibility threshold on the data: in place of the bare condition m 0 > 0 m_0>0 m 0 > 0 one obtains the effective mass condition (8) , m 0 + Z W > 0 m_0+ZW>0 m 0 + Z W > 0 , and its borderline version m 0 + Z W = 0 m_0+ZW=0 m 0 + Z W = 0 . A positive impulse Z W > 0 ZW>0 Z W > 0 subsidizes the initial mass, so that even sign-changing data with m 0 ≤ 0 m_0\le0 m 0 ≤ 0 must blow up provided m 0 ≥ − Z W m_0\ge-ZW m 0 ≥ − Z W and the data are nontrivial; conversely a negative impulse Z W < 0 ZW<0 Z W < 0 can be compensated by a positive initial mass. In the model case (2) the parameter σ \sigma σ enters only through the constant Z = Γ ( σ + 1 ) Z=\Gamma(\sigma+1) Z = Γ ( σ + 1 ) . Setting w ≡ 0 w\equiv0 w ≡ 0 recovers the Fujita statement m 0 ≥ 0 m_0\ge0 m 0 ≥ 0 , u 0 ≢ 0 ⇒ u_0\not\equiv0\Rightarrow u 0 ≡ 0 ⇒ no global solution for 1 < p ≤ p F 1<p\le p_F 1 < p ≤ p F .
Remark 3.2. The proof of Theorem 1.1 uses p ≤ p F p\le p_F p ≤ p F only through the boundedness of the diffusive cost R N + 2 − 2 p ′ R^{N+2-2p'} R N + 2 − 2 p ′ in (33) and (26) . For p > p F p>p_F p > p F the identity (7) is false in general: by Corollary 1.1 and Theorem 1.4 (iii), small global solutions with M > 0 M>0 M > 0 exist and have asymptotic mass m ∞ = M + ∬ Q ∣ u ∣ p > 0 m_\infty=M+\iint_Q|u|^p>0 m ∞ = M + ∬ Q ∣ u ∣ p > 0 .
Remark 3.3. If u 0 ≤ 0 u_0\le0 u 0 ≤ 0 and ζ w ≤ 0 \zeta w\le0 ζ w ≤ 0 , Theorem 1.3 provides a global solution u ≤ 0 u\le0 u ≤ 0 , and v : = − u ≥ 0 v:=-u\ge0 v := − u ≥ 0 solves the absorption problem
∂ t v − Δ v + v p = ∣ ζ ( t ) w ( x ) ∣ , v ( 0 ) = ∣ u 0 ∣ . \partial_t v-\Delta v+v^p=|\zeta(t)w(x)|,\qquad v(0)=|u_0|. ∂ t v − Δ v + v p = ∣ ζ ( t ) w ( x ) ∣ , v ( 0 ) = ∣ u 0 ∣.
For 1 < p ≤ p F 1 < p \le p_F 1 < p ≤ p F , Theorem 1.1 then states that ∬ Q v p = ∥ v 0 ∥ 1 + ∥ ζ w ∥ L 1 ( Q ) \iint_Q v^p = \|v_0\|_1 + \|\zeta w\|_{L^1(Q)} ∬ Q v p = ∥ v 0 ∥ 1 + ∥ ζ w ∥ L 1 ( Q ) : all the mass that is present initially or injected later is eventually absorbed, and by Theorem 1.4 (ii) ∥ v ( t ) ∥ 1 → 0 \|v(t)\|_1 \to0 ∥ v ( t ) ∥ 1 → 0 . For ζ w ≡ 0 \zeta w \equiv0 ζ w ≡ 0 this is the classical extinction of mass for the heat equation with absorption in the range p ≤ p F p \le p_F p ≤ p F [16 , 21 , 15 , 10 ] , for which the present argument gives a short proof based only on the weak formulation.
A localized obstruction for every p > 1 p>1 p > 1
Proof of Theorem 1.6
Let u u u be a global weak solution, and let x 0 ∈ R N x_0 \in\mathbb{R}^N x 0 ∈ R N , R , T > 0 R,T>0 R , T > 0 . We test (2.1) with the separated cutoff Ψ R , T \Psi_{R,T} Ψ R , T of (2.8), which is admissible by Remark 2.1. Writing ϕ R ( x ) = η ( ∣ x − x 0 ∣ 2 / R 2 ) \phi_R(x)=\eta(|x-x_0|^2/R^2) ϕ R ( x ) = η ( ∣ x − x 0 ∣ 2 / R 2 ) and η T ( t ) = η ( t / T ) \eta_T(t)=\eta(t/T) η T ( t ) = η ( t / T ) , we obtain
J R , T + F R , T + I R = − ∬ Q u L Ψ R , T , J R , T : = ∬ Q ∣ u ∣ p Ψ R , T , I R : = ∫ R N u 0 ϕ R k d x . (36) J_{R,T}+F_{R,T}+I_R=-\iint_Q u\mathcal{L}\Psi_{R,T},\qquad J_{R,T}:=\iint_Q |u|^p\Psi_{R,T},\qquad I_R:=\int_{\mathbb{R}^N}u_0\phi_R^k\,dx. \tag*{(36)} J R , T + F R , T + I R = − ∬ Q u L Ψ R , T , J R , T := ∬ Q ∣ u ∣ p Ψ R , T , I R := ∫ R N u 0 ϕ R k d x . (36)
and, because Ψ R , T \Psi_{R,T} Ψ R , T is a product, the forcing term separates:
F R , T : = ∬ Q ζ w Ψ R , T = ( ∫ 0 ∞ ζ ( t ) η T ( t ) k d t ) ( ∫ R N w ( x ) ϕ R ( x ) k d x ) . (37) F_{R,T}:=\iint_Q \zeta w\Psi_{R,T}=\left(\int_0^\infty\zeta(t)\eta_T(t)^k\,dt\right)\left(\int_{\mathbb{R}^N}w(x)\phi_R(x)^k\,dx\right). \tag*{(37)} F R , T := ∬ Q ζ w Ψ R , T = ( ∫ 0 ∞ ζ ( t ) η T ( t ) k d t ) ( ∫ R N w ( x ) ϕ R ( x ) k d x ) . (37)
Since L Ψ R , T \mathcal{L}\Psi_{R,T} L Ψ R , T vanishes where Ψ R , T \Psi_{R,T} Ψ R , T does, Hölder’s inequality (in the weight Ψ R , T \Psi_{R,T} Ψ R , T ) and Young’s inequality give
∣ ∬ Q u L Ψ R , T ∣ ≤ J R , T 1 / p I R , T 1 / p ′ ≤ 1 p J R , T + 1 p ′ I R , T , \left|\iint_Q u\mathcal{L}\Psi_{R,T}\right|\le J_{R,T}^{1/p}I_{R,T}^{1/p'}\le\frac{1}{p}J_{R,T}+\frac{1}{p'}I_{R,T}, ∬ Q u L Ψ R , T ≤ J R , T 1/ p I R , T 1/ p ′ ≤ p 1 J R , T + p ′ 1 I R , T ,
with I R , T I_{R,T} I R , T as in (2.12). Inserting this into (4.1) gives 1 p ′ J R , T + F R , T + I R ≤ 1 p ′ I R , T \frac{1}{p'}J_{R,T}+F_{R,T}+I_R\le\frac{1}{p'}I_{R,T} p ′ 1 J R , T + F R , T + I R ≤ p ′ 1 I R , T ; dropping the nonnegative first term and invoking Lemma 2.5 we obtain (1.11) with C 0 : = C / p ′ C_0:=C/p' C 0 := C / p ′ , where C C C is the constant of (2.12).
With the parabolic choice T = R 2 T=R^2 T = R 2 , (1.11) becomes
∫ R N u 0 ϕ R k d x + ( ∫ 0 ∞ ζ η ( t R 2 ) k d t ) ( ∫ R N w ϕ R k d x ) ≤ 2 C 0 R N − 2 p − 1 , (38) \int_{\mathbb{R}^N}u_0\phi_R^k\,dx+\left(\int_0^\infty\zeta\eta\left(\frac{t}{R^2}\right)^k\,dt\right)\left(\int_{\mathbb{R}^N}w\phi_R^k\,dx\right)\le2C_0R^{N-\frac{2}{p-1}}, \tag*{(38)} ∫ R N u 0 ϕ R k d x + ( ∫ 0 ∞ ζ η ( R 2 t ) k d t ) ( ∫ R N w ϕ R k d x ) ≤ 2 C 0 R N − p − 1 2 , (38)
and we note the trivial bounds ∣ ∫ 0 ∞ ζ η T k ∣ ≤ ∥ ζ ∥ L 1 ( 0 , 2 T ) \left|\int_0^\infty\zeta\eta_T^k\right|\le\|\zeta\|_{L^1(0,2T)} ∫ 0 ∞ ζ η T k ≤ ∥ ζ ∥ L 1 ( 0 , 2 T ) and ∣ ∫ w ϕ R k ∣ ≤ ∥ w ∥ L 1 ( B ( x 0 , 2 R ) ) \left|\int w\phi_R^k\right|\le\|w\|_{L^1(B(x_0,\sqrt{2}R))} ∫ w ϕ R k ≤ ∥ w ∥ L 1 ( B ( x 0 , 2 R )) .
Consequences
Corollary 4.1. Let p > 1 p>1 p > 1 .
(a) (Large data.) Let ζ ∈ L l o c 1 ( [ 0 , ∞ ) ) \zeta\in L^1_{\mathrm{loc}}([0,\infty)) ζ ∈ L loc 1 ([ 0 , ∞ )) and w ∈ L l o c 1 ( R N ) w\in L^1_{\mathrm{loc}}(\mathbb{R}^N) w ∈ L loc 1 ( R N ) be fixed, and let g ∈ L l o c 1 ( R N ) g\in L^1_{\mathrm{loc}}(\mathbb{R}^N) g ∈ L loc 1 ( R N ) satisfy ∣ { g > 0 } ∣ > 0 |\{g>0\}|>0 ∣ { g > 0 } ∣ > 0 . There is λ ∗ > 0 \lambda_*>0 λ ∗ > 0 such that (1.1) with u 0 = λ g u_0=\lambda g u 0 = λ g has no global weak solution for λ > λ ∗ \lambda>\lambda_* λ > λ ∗ .
(b) (Large forcing.) Let u 0 ∈ L l o c 1 ( R N ) u_0\in L^1_{\mathrm{loc}}(\mathbb{R}^N) u 0 ∈ L loc 1 ( R N ) be fixed, let ζ ∈ L l o c 1 ( [ 0 , ∞ ) ) \zeta\in L^1_{\mathrm{loc}}([0,\infty)) ζ ∈ L loc 1 ([ 0 , ∞ )) be nonnegative and not a.e. zero, and let w ∈ L l o c 1 ( R N ) w\in L^1_{\mathrm{loc}}(\mathbb{R}^N) w ∈ L loc 1 ( R N ) satisfy ∣ { w > 0 } ∣ > 0 |\{w>0\}|>0 ∣ { w > 0 } ∣ > 0 . There is μ ∗ > 0 \mu_*>0 μ ∗ > 0 such that (1.1) with forcing μ ζ ( t ) w ( x ) \mu\zeta(t)w(x) μ ζ ( t ) w ( x ) has no global weak solution for μ > μ ∗ \mu>\mu_* μ > μ ∗ .
(c) (Slowly decaying data.) Let ζ ∈ L 1 ( 0 , ∞ ) \zeta\in L^1(0,\infty) ζ ∈ L 1 ( 0 , ∞ ) , w ∈ L 1 ( R N ) w\in L^1(\mathbb{R}^N) w ∈ L 1 ( R N ) , and let u 0 ∈ L l o c 1 ( R N ) u_0\in L^1_{\mathrm{loc}}(\mathbb{R}^N) u 0 ∈ L loc 1 ( R N ) satisfy u 0 ( x ) ≥ A ∣ x ∣ − γ u_0(x)\ge A|x|^{-\gamma} u 0 ( x ) ≥ A ∣ x ∣ − γ for a.e. ∣ x ∣ ≥ R 0 |x|\ge R_0 ∣ x ∣ ≥ R 0 , for some A , R 0 > 0 A,R_0>0 A , R 0 > 0 and γ ≥ 0 \gamma\ge0 γ ≥ 0 . If γ < min { N , 2 / ( p − 1 ) } \gamma<\min\{N,2/(p-1)\} γ < min { N , 2/ ( p − 1 )} , there is no global weak solution. If γ = 2 / ( p − 1 ) < N \gamma=2/(p-1)<N γ = 2/ ( p − 1 ) < N , there is A ∗ = A ∗ ( N , p , η ) A_*=A_*(N,p,\eta) A ∗ = A ∗ ( N , p , η ) such that there is no global weak solution when A > A ∗ A>A_* A > A ∗ .
Proof. (a) Almost every point of { g > 0 } \{g>0\} { g > 0 } is a Lebesgue point of g g g ; fix such a point x 0 x_0 x 0 . Then
∣ R − N ∫ R N ( g ( x ) − g ( x 0 ) ) ϕ R ( x ) k d x ∣ ≤ R − N ∫ B ( x 0 , 2 R ) ∣ g − g ( x 0 ) ∣ d x → R → 0 0. \left|R^{-N}\int_{\mathbb{R}^N}(g(x)-g(x_0))\phi_R(x)^k\,dx\right|\le R^{-N}\int_{B(x_0,\sqrt{2}R)}|g-g(x_0)|\,dx\xrightarrow[R\to0]{}0. R − N ∫ R N ( g ( x ) − g ( x 0 )) ϕ R ( x ) k d x ≤ R − N ∫ B ( x 0 , 2 R ) ∣ g − g ( x 0 ) ∣ d x R → 0 0.
while R − N ∫ ϕ R k d x = c η : = ∫ η ( ∣ y ∣ 2 ) k d y > 0 R^{-N}\int\phi_R^k\,dx=c_\eta:=\int\eta(|y|^2)^k\,dy>0 R − N ∫ ϕ R k d x = c η := ∫ η ( ∣ y ∣ 2 ) k d y > 0 . Hence there is R 0 > 0 R_0>0 R 0 > 0 with G 0 : = ∫ g ϕ R 0 k d x ≥ 1 2 c η g ( x 0 ) R 0 N > 0 G_0:=\int g\phi_{R_0}^k\,dx\ge\frac{1}{2}c_\eta g(x_0)R_0^N>0 G 0 := ∫ g ϕ R 0 k d x ≥ 2 1 c η g ( x 0 ) R 0 N > 0 . If a global weak solution with u 0 = λ g u_0=\lambda g u 0 = λ g exists, (38) with R = R 0 R=R_0 R = R 0 gives
λ G 0 ≤ 2 C 0 R 0 N − 2 p − 1 + ∥ ζ ∥ L 1 ( 0 , 2 R 0 2 ) ∥ w ∥ L 1 ( B ( x 0 , 2 R 0 ) ) = : λ ∗ G 0 . \lambda G_0\le2C_0R_0^{N-\frac{2}{p-1}}+\|\zeta\|_{L^1(0,2R_0^2)}\|w\|_{L^1(B(x_0,\sqrt{2}R_0))}=:\lambda_*G_0. λ G 0 ≤ 2 C 0 R 0 N − p − 1 2 + ∥ ζ ∥ L 1 ( 0 , 2 R 0 2 ) ∥ w ∥ L 1 ( B ( x 0 , 2 R 0 )) =: λ ∗ G 0 .
(b) As in (a), choose x 0 , R 0 x_0,R_0 x 0 , R 0 with w 0 : = ∫ w ϕ R 0 k d x > 0 w_0:=\int w\phi_{R_0}^k\,dx>0 w 0 := ∫ w ϕ R 0 k d x > 0 , and choose T 0 > 0 T_0>0 T 0 > 0 with ∫ 0 T 0 ζ > 0 \int_0^{T_0}\zeta>0 ∫ 0 T 0 ζ > 0 , so that z 0 : = ∫ 0 ∞ ζ η T 0 k ≥ ∫ 0 T 0 ζ > 0 z_0:=\int_0^\infty\zeta\eta_{T_0}^k\ge\int_0^{T_0}\zeta>0 z 0 := ∫ 0 ∞ ζ η T 0 k ≥ ∫ 0 T 0 ζ > 0 . If a global weak solution exists with forcing μ ζ w \mu\zeta w μ ζ w , (12) gives μ z 0 w 0 ≤ C 0 ( R 0 N + R 0 N − 2 p ′ T 0 ) + ∥ u 0 ∥ L 1 ( B ( x 0 , 2 R 0 ) ) \mu z_0w_0\le C_0(R_0^N+R_0^{N-2p'}T_0)+\|u_0\|_{L^1(B(x_0,\sqrt{2}R_0))} μ z 0 w 0 ≤ C 0 ( R 0 N + R 0 N − 2 p ′ T 0 ) + ∥ u 0 ∥ L 1 ( B ( x 0 , 2 R 0 )) , which bounds μ \mu μ .
(c) Take x 0 = 0 x_0=0 x 0 = 0 and R ≥ R 0 R\ge R_0 R ≥ R 0 in (38) . Since ϕ R = 1 \phi_R=1 ϕ R = 1 on B R B_R B R and u 0 ≥ 0 u_0\ge0 u 0 ≥ 0 outside B R 0 B_{R_0} B R 0 ,
∫ u 0 ϕ R k d x ≥ − ∥ u 0 ∥ L 1 ( B R 0 ) + A ∫ R 0 ≤ ∣ x ∣ ≤ R ∣ x ∣ − γ d x = − ∥ u 0 ∥ L 1 ( B R 0 ) + A ∣ S N − 1 ∣ N − γ ( R N − γ − R 0 N − γ ) . \int u_0\phi_R^k\,dx\ge-\|u_0\|_{L^1(B_{R_0})}+A\int_{R_0\le|x|\le R}|x|^{-\gamma}\,dx=-\|u_0\|_{L^1(B_{R_0})}+\frac{A|\mathbb{S}^{N-1}|}{N-\gamma}\left(R^{N-\gamma}-R_0^{N-\gamma}\right). ∫ u 0 ϕ R k d x ≥ − ∥ u 0 ∥ L 1 ( B R 0 ) + A ∫ R 0 ≤ ∣ x ∣ ≤ R ∣ x ∣ − γ d x = − ∥ u 0 ∥ L 1 ( B R 0 ) + N − γ A ∣ S N − 1 ∣ ( R N − γ − R 0 N − γ ) .
while the forcing term is bounded below by − ∥ ζ ∥ 1 ∥ w ∥ 1 -\|\zeta\|_1\|w\|_1 − ∥ ζ ∥ 1 ∥ w ∥ 1 . Dividing (38) by R N − γ R^{N-\gamma} R N − γ and letting R → ∞ R\to\infty R → ∞ , we get A ∣ S N − 1 ∣ / ( N − γ ) ≤ lim sup R → ∞ 2 C 0 R γ − 2 / ( p − 1 ) A|\mathbb{S}^{N-1}|/(N-\gamma)\le\limsup_{R\to\infty}2C_0R^{\gamma-2/(p-1)} A ∣ S N − 1 ∣/ ( N − γ ) ≤ lim sup R → ∞ 2 C 0 R γ − 2/ ( p − 1 ) . The right-hand side vanishes if γ < 2 / ( p − 1 ) \gamma<2/(p-1) γ < 2/ ( p − 1 ) , a contradiction; if γ = 2 / ( p − 1 ) \gamma=2/(p-1) γ = 2/ ( p − 1 ) it equals 2 C 0 2C_0 2 C 0 , and we reach a contradiction when A > A ∗ : = 2 C 0 ( N − γ ) / ∣ S N − 1 ∣ A>A_*:=2C_0(N-\gamma)/|\mathbb{S}^{N-1}| A > A ∗ := 2 C 0 ( N − γ ) /∣ S N − 1 ∣ .
Part (c) is the forced counterpart of the result of Lee and Ni [26 ] for slowly decaying data; an integrable forcing does not affect it. In the Fujita range p ≤ p F p\le p_F p ≤ p F , the condition in (c) reads γ < N \gamma<N γ < N , so that u 0 ∉ L 1 u_0\notin L^1 u 0 ∈ / L 1 and, heuristically, M = + ∞ M=+\infty M = + ∞ , in accordance with Theorem 1.2 .
Example 4.1. Let 1 < p ≤ p F 1<p\le p_F 1 < p ≤ p F , let a ∈ C c ( B 1 ) a\in C_c(B_1) a ∈ C c ( B 1 ) be nonnegative and not identically zero, fix x 1 ∈ R N x_1\in\mathbb{R}^N x 1 ∈ R N with ∣ x 1 ∣ ≥ 4 |x_1|\ge4 ∣ x 1 ∣ ≥ 4 , and consider ζ ∈ L 1 ( 0 , ∞ ) \zeta\in L^1(0,\infty) ζ ∈ L 1 ( 0 , ∞ ) and w ∈ L 1 ( R N ) ∩ L ∞ ( R N ) w\in L^1(\mathbb{R}^N)\cap L^\infty(\mathbb{R}^N) w ∈ L 1 ( R N ) ∩ L ∞ ( R N ) with ζ ≥ 0 \zeta\ge0 ζ ≥ 0 , w ≤ 0 w\le0 w ≤ 0 . For λ > 0 \lambda>0 λ > 0 the data u 0 = − λ a u_0=-\lambda a u 0 = − λa have M = − λ ∫ a + Z W < 0 M=-\lambda\int a+ZW<0 M = − λ ∫ a + Z W < 0 , and Theorem 1.3 gives a global solution. The data u 0 = λ ( a − 2 a ( ⋅ − x 1 ) ) u_0=\lambda(a-2a(\cdot-x_1)) u 0 = λ ( a − 2 a ( ⋅ − x 1 )) also have M = − λ ∫ a + Z W < 0 M=-\lambda\int a+ZW<0 M = − λ ∫ a + Z W < 0 , but { u 0 > 0 } \{u_0>0\} { u 0 > 0 } has positive measure, and Corollary 4.1(a) shows that there is no global weak solution for λ \lambda λ large.
Non-integrable forcing
The obstruction (12) also explains why the integrability of ζ \zeta ζ is the dividing line between the present theory and the results for power forcings.
Corollary 4.2. Let u 0 ∈ L 1 ( R N ) u_0\in L^1(\mathbb{R}^N) u 0 ∈ L 1 ( R N ) , w ∈ L 1 ( R N ) w\in L^1(\mathbb{R}^N) w ∈ L 1 ( R N ) with W > 0 W>0 W > 0 , and ζ ∈ L l o c 1 ( [ 0 , ∞ ) ) \zeta\in L^1_{\mathrm{loc}}([0,\infty)) ζ ∈ L loc 1 ([ 0 , ∞ )) with negative part ζ − ∈ L 1 ( 0 , ∞ ) \zeta_-\in L^1(0,\infty) ζ − ∈ L 1 ( 0 , ∞ ) ; set Z ( T ) : = ∫ 0 T ζ d t Z(T):=\int_0^T\zeta\,dt Z ( T ) := ∫ 0 T ζ d t .
If 1 < p ≤ p F 1<p\le p_F 1 < p ≤ p F and Z ( T ) → + ∞ Z(T)\to+\infty Z ( T ) → + ∞ as T → ∞ T\to\infty T → ∞ , then (1) has no global weak solution, whatever the size of u 0 u_0 u 0 and w w w .
If p > p F p>p_F p > p F and lim sup T → ∞ T 1 p − 1 − N 2 Z ( T ) = + ∞ \limsup_{T\to\infty}T^{\frac{1}{p-1}-\frac{N}{2}}Z(T)=+\infty lim sup T → ∞ T p − 1 1 − 2 N Z ( T ) = + ∞ , then (1) has no global weak solution.
Proof. Take x 0 = 0 x_0=0 x 0 = 0 in (38) . Writing ζ = ζ + − ζ − \zeta=\zeta_+-\zeta_- ζ = ζ + − ζ − and using η T k = 1 \eta_T^k=1 η T k = 1 on [ 0 , T ] [0,T] [ 0 , T ] ,
∫ 0 ∞ ζ η T k d t ≥ ∫ 0 T ζ + d t − ∥ ζ − ∥ 1 ≥ Z ( T ) − ∥ ζ − ∥ 1 . \int_0^\infty\zeta\eta_T^k\,dt\ge\int_0^T\zeta_+\,dt-\|\zeta_-\|_1\ge Z(T)-\|\zeta_-\|_1. ∫ 0 ∞ ζ η T k d t ≥ ∫ 0 T ζ + d t − ∥ ζ − ∥ 1 ≥ Z ( T ) − ∥ ζ − ∥ 1 .
Since ∫ w ϕ R k → W > 0 \int w\phi_R^k\to W>0 ∫ w ϕ R k → W > 0 , there is R 2 ≥ 1 R_2\ge1 R 2 ≥ 1 with ∫ w ϕ R k ≥ W / 2 \int w\phi_R^k\ge W/2 ∫ w ϕ R k ≥ W /2 for R ≥ R 2 R\ge R_2 R ≥ R 2 . Hence, for R ≥ R 2 R\ge R_2 R ≥ R 2 such that Z ( R 2 ) ≥ ∥ ζ − ∥ 1 Z(R^2)\ge\|\zeta_-\|_1 Z ( R 2 ) ≥ ∥ ζ − ∥ 1 ,
W 2 ( Z ( R 2 ) − ∥ ζ − ∥ 1 ) ≤ ∥ u 0 ∥ 1 + 2 C 0 R N − 2 p − 1 . (39) \frac{W}{2}\left(Z(R^2)-\|\zeta_-\|_1\right)\le\|u_0\|_1+2C_0R^{N-\frac{2}{p-1}}. \tag*{(39)} 2 W ( Z ( R 2 ) − ∥ ζ − ∥ 1 ) ≤ ∥ u 0 ∥ 1 + 2 C 0 R N − p − 1 2 . (39)
In case (i), N − 2 / ( p − 1 ) ≤ 0 N-2/(p-1)\le0 N − 2/ ( p − 1 ) ≤ 0 , so the right-hand side of (39) is bounded while the left-hand side tends to + ∞ +\infty + ∞ . In case (ii), set β : = 1 p − 1 − N 2 < 0 \beta:=\frac{1}{p-1}-\frac{N}{2}<0 β := p − 1 1 − 2 N < 0 and choose T j = R j 2 → ∞ T_j=R_j^2\to\infty T j = R j 2 → ∞ with T j β Z ( T j ) → ∞ T_j^\beta Z(T_j)\to\infty T j β Z ( T j ) → ∞ ; in particular Z ( T j ) → ∞ Z(T_j)\to\infty Z ( T j ) → ∞ . Multiplying (39) by R j 2 β = R j 2 p − 1 − N R_j^{2\beta}=R_j^{\frac{2}{p-1}-N} R j 2 β = R j p − 1 2 − N gives W 2 T j β Z ( T j ) ≤ ( ∥ u 0 ∥ 1 + W 2 ∥ ζ − ∥ 1 ) T j β + 2 C 0 \frac{W}{2}T_j^\beta Z(T_j)\le(\|u_0\|_1+\frac{W}{2}\|\zeta_-\|_1)T_j^\beta+2C_0 2 W T j β Z ( T j ) ≤ ( ∥ u 0 ∥ 1 + 2 W ∥ ζ − ∥ 1 ) T j β + 2 C 0 , whose right-hand side is bounded, a contradiction.
Remark 4.1. For ζ ( t ) = t σ \zeta(t)=t^\sigma ζ ( t ) = t σ , σ > − 1 \sigma>-1 σ > − 1 , one has Z ( T ) = T σ + 1 / ( σ + 1 ) Z(T)=T^{\sigma+1}/(\sigma+1) Z ( T ) = T σ + 1 / ( σ + 1 ) , and Corollary 4.2 gives nonexistence whenever σ + 1 > N 2 − 1 p − 1 \sigma+1>\frac{N}{2}-\frac{1}{p-1} σ + 1 > 2 N − p − 1 1 , that is for p < N − 2 σ N − 2 σ − 2 p<\frac{N-2\sigma}{N-2\sigma-2} p < N − 2 σ − 2 N − 2 σ when N > 2 σ + 2 N>2\sigma+2 N > 2 σ + 2 , and for every p > 1 p>1 p > 1 when N ≤ 2 σ + 2 N\leq2\sigma+2 N ≤ 2 σ + 2 . This recovers the subcritical blow-up range of Jleli, Kawakami and Samet [20 ] , for weak solutions, in every dimension and for sign-changing u 0 ∈ L 1 u_0\in L^1 u 0 ∈ L 1 ; the critical case p = p ∗ ( σ ) p=p^*(\sigma) p = p ∗ ( σ ) requires a finer argument and is treated in [20 ] . As σ ↓ − 1 \sigma\downarrow-1 σ ↓ − 1 the exponent N − 2 σ N − 2 σ − 2 \frac{N-2\sigma}{N-2\sigma-2} N − 2 σ − 2 N − 2 σ decreases to N + 2 N = p F \frac{N+2}{N}=p_F N N + 2 = p F . An integrable forcing such as t σ e − t t^\sigma e^{-t} t σ e − t sits at this endpoint: its impulse is finite, the T T T -power in (39) is 0 0 0 , and the blow-up range collapses exactly to 1 < p ≤ p F 1<p\leq p_F 1 < p ≤ p F , now subject to the effective-mass condition.
The logarithmic profile ζ ( t ) = ( 1 + t ) − 1 \zeta(t)=(1+t)^{-1} ζ ( t ) = ( 1 + t ) − 1 lies exactly on the borderline between the two regimes. Here Z ( T ) = log ( 1 + T ) → ∞ Z(T)=\log(1+T)\to\infty Z ( T ) = log ( 1 + T ) → ∞ , so by Corollary 4.2(i) there is no global solution for 1 < p ≤ p F 1<p\leq p_F 1 < p ≤ p F as soon as W > 0 W>0 W > 0 , irrespective of the size and sign of u 0 u_0 u 0 : the effective mass is + ∞ +\infty + ∞ . On the other hand, splitting the convolution at t / 2 t/2 t /2 as in the proof of Lemma 2.7 one finds ( ∣ ζ ∣ ⋆ h N ) ( t ) ≤ C ( t − N / 2 log ( 1 + t ) + ( 1 + t ) − 1 ∫ 0 t / 2 h N ) (|\zeta|\star h_N)(t)\leq C(t^{-N/2}\log(1+t)+(1+t)^{-1}\int_0^{t/2}h_N) ( ∣ ζ ∣ ⋆ h N ) ( t ) ≤ C ( t − N /2 log ( 1 + t ) + ( 1 + t ) − 1 ∫ 0 t /2 h N ) for t ≥ 2 t\geq2 t ≥ 2 , which is O ( t − 1 / 2 log t ) O(t^{-1/2}\log t) O ( t − 1/2 log t ) if N = 1 N=1 N = 1 , O ( t − 1 log t ) O(t^{-1}\log t) O ( t − 1 log t ) if N = 2 N=2 N = 2 and O ( t − 1 ) O(t^{-1}) O ( t − 1 ) if N ≥ 3 N\geq3 N ≥ 3 . Hence ω ζ < ∞ \omega_\zeta<\infty ω ζ < ∞ whenever p > max { p F , 2 } p>\max\{p_F,2\} p > max { p F , 2 } , and Corollary 1.1 gives small-data global existence in that range. For N ∈ { 1 , 2 } N\in\{1,2\} N ∈ { 1 , 2 } , where p F ≥ 2 p_F\geq2 p F ≥ 2 , the critical exponent for ζ = ( 1 + t ) − 1 \zeta=(1+t)^{-1} ζ = ( 1 + t ) − 1 is therefore still p F p_F p F , but the dichotomy in the Fujita range no longer depends on the data. Note that Corollary 1.1 does not require ζ ∈ L 1 \zeta\in L^1 ζ ∈ L 1 .
Global existence
The linear forced flow
Let u 0 , w ∈ L 1 ( R N ) ∩ L ∞ ( R N ) u_0,w\in L^1(\mathbb{R}^N)\cap L^\infty(\mathbb{R}^N) u 0 , w ∈ L 1 ( R N ) ∩ L ∞ ( R N ) and ζ ∈ L l o c 1 ( [ 0 , ∞ ) ) \zeta\in L^1_{\mathrm{loc}}([0,\infty)) ζ ∈ L loc 1 ([ 0 , ∞ )) , and let V V V and m m m be defined by (9) . We record the elementary properties on which the proof of Theorem 1.5 rests. First, V = S ( t ) ∣ u 0 ∣ + D [ ∣ ζ ∣ ⊗ ∣ w ∣ ] V=S(t)|u_0|+D[|\zeta|\otimes|w|] V = S ( t ) ∣ u 0 ∣ + D [ ∣ ζ ∣ ⊗ ∣ w ∣ ] is the solution of the linear problem with absolute data, and by Lemma 2.1(a),(b),
0 ≤ V ( x , t ) ≤ m ( t ) for all ( x , t ) ∈ Q , S ( t − s ) V ( s ) ≤ V ( t ) for 0 ≤ s ≤ t . (40) 0\leq V(x,t)\leq m(t)\quad\text{for all }(x,t)\in Q,\qquad S(t-s)V(s)\leq V(t)\quad\text{for }0\leq s\leq t. \tag*{(40)} 0 ≤ V ( x , t ) ≤ m ( t ) for all ( x , t ) ∈ Q , S ( t − s ) V ( s ) ≤ V ( t ) for 0 ≤ s ≤ t . (40)
The second inequality follows from the semigroup property, Tonelli’s theorem and positivity: S ( t − s ) V ( s ) = S ( t ) ∣ u 0 ∣ + ∫ 0 s ∣ ζ ( r ) ∣ S ( t − r ) ∣ w ∣ d r ≤ V ( t ) S(t-s)V(s)=S(t)|u_0|+\int_0^s|\zeta(r)|S(t-r)|w|\,dr\leq V(t) S ( t − s ) V ( s ) = S ( t ) ∣ u 0 ∣ + ∫ 0 s ∣ ζ ( r ) ∣ S ( t − r ) ∣ w ∣ d r ≤ V ( t ) . Second, by Lemma 2.1(b),
m ( t ) ≤ [ [ u 0 ] ] h N ( t ) + [ [ w ] ] ( ∣ ζ ∣ ⋆ h N ) ( t ) ( t > 0 ) . (41) m(t)\leq[\![u_0]\!]h_N(t)+[\![w]\!](|\zeta|\star h_N)(t)\qquad(t>0). \tag*{(41)} m ( t ) ≤ [ [ u 0 ] ] h N ( t ) + [ [ w ] ] ( ∣ ζ ∣ ⋆ h N ) ( t ) ( t > 0 ) . (41)
which bounds the linear forced flow in terms of the forcing kernel. Third, the first inequality in (40) yields the pointwise interpolation inequality
V ( x , s ) p ≤ m ( s ) p − 1 V ( x , s ) , (42) V(x,s)^p\leq m(s)^{p-1}V(x,s), \tag*{(42)} V ( x , s ) p ≤ m ( s ) p − 1 V ( x , s ) , (42)
which replaces the usual L 1 L^1 L 1 –L ∞ L^\infty L ∞ interpolation ∥ f ∥ p p ≤ ∥ f ∥ 1 ∥ f ∥ ∞ p − 1 \|f\|_p^p\leq\|f\|_1\|f\|_\infty^{p-1} ∥ f ∥ p p ≤ ∥ f ∥ 1 ∥ f ∥ ∞ p − 1 by a pointwise statement.
Proof of Theorem 1.5
Let 0 < T ≤ ∞ 0<T\leq\infty 0 < T ≤ ∞ and assume (10) . Put
a ( t ) : = p κ p − 1 m ( t ) p − 1 , A ( t ) : = ∫ 0 t a ( s ) d s , a(t):=p\kappa^{p-1}m(t)^{p-1},\qquad A(t):=\int_0^t a(s)\,ds, a ( t ) := p κ p − 1 m ( t ) p − 1 , A ( t ) := ∫ 0 t a ( s ) d s ,
and note that A ( t ) ≤ p κ p − 1 c p = 1 A(t)\leq p\kappa^{p-1}c_p=1 A ( t ) ≤ p κ p − 1 c p = 1 on ( 0 , T ) (0,T) ( 0 , T ) , because p κ p − 1 c p = p ⋅ p p − 1 ( p − 1 ) 1 − p ( p − 1 ) p − 1 p − p = 1 p\kappa^{p-1}c_p=p\cdot p^{p-1}(p-1)^{1-p}(p-1)^{p-1}p^{-p}=1 p κ p − 1 c p = p ⋅ p p − 1 ( p − 1 ) 1 − p ( p − 1 ) p − 1 p − p = 1 . Let K \mathcal{K} K be the set of measurable functions u u u on Q T Q_T Q T with ∣ u ∣ ≤ κ V |u|\leq\kappa V ∣ u ∣ ≤ κV a.e., where functions that agree a.e. are identified, and define
d ( u , v ) : = ess sup { ( x , t ) ∈ Q T : V ( x , t ) > 0 } e − 2 A ( t ) ∣ u ( x , t ) − v ( x , t ) ∣ V ( x , t ) . d(u,v):=\underset{\{(x,t)\in Q_T:V(x,t)>0\}}{\operatorname{ess\,sup}}\ e^{-2A(t)}\frac{|u(x,t)-v(x,t)|}{V(x,t)}. d ( u , v ) := {( x , t ) ∈ Q T : V ( x , t ) > 0 } ess sup e − 2 A ( t ) V ( x , t ) ∣ u ( x , t ) − v ( x , t ) ∣ .
Every u ∈ K u\in\mathcal{K} u ∈ K vanishes a.e. on { V = 0 } \{V=0\} { V = 0 } , so d d d is a metric on K \mathcal{K} K with d ≤ 2 κ d\leq2\kappa d ≤ 2 κ . The space ( K , d ) (\mathcal{K},d) ( K , d ) is complete: if ( u n ) (u_n) ( u n ) is d d d -Cauchy then, since e − 2 A ≥ e − 2 e^{-2A}\geq e^{-2} e − 2 A ≥ e − 2 , the quotients u n / V u_n/V u n / V form a Cauchy sequence in L ∞ ( { V > 0 } ) L^\infty(\{V>0\}) L ∞ ({ V > 0 }) ; their limit ρ \rho ρ satisfies ∣ ρ ∣ ≤ κ |\rho|\leq\kappa ∣ ρ ∣ ≤ κ , and u : = ρ V u:=\rho V u := ρ V (extended by zero) belongs to K \mathcal{K} K with d ( u n , u ) → 0 d(u_n,u)\to0 d ( u n , u ) → 0 .
For u ∈ K u \in\mathcal{K} u ∈ K we set
Φ [ u ] ( t ) : = S ( t ) u 0 + ∫ 0 t S ( t − s ) ∣ u ( s ) ∣ p d s + ∫ 0 t S ( t − s ) ( ζ ( s ) w ) d s . \Phi[u](t) := S(t)u_0+\int_0^t S(t-s)|u(s)|^p\,ds+\int_0^t S(t-s)(\zeta(s)w)\,ds. Φ [ u ] ( t ) := S ( t ) u 0 + ∫ 0 t S ( t − s ) ∣ u ( s ) ∣ p d s + ∫ 0 t S ( t − s ) ( ζ ( s ) w ) d s .
By (42) , ∣ u ∣ p ≤ κ p V p ≤ κ p m p − 1 V |u|^p \le\kappa^p V^p \le\kappa^p m^{p-1}V ∣ u ∣ p ≤ κ p V p ≤ κ p m p − 1 V , and by (40) , ∫ 0 t S ( t − s ) ( m ( s ) p − 1 V ( s ) ) d s ≤ ( ∫ 0 t m p − 1 ) V ( t ) < ∞ \int_0^t S(t-s)(m(s)^{p-1}V(s))\,ds \le\left(\int_0^t m^{p-1}\right)V(t)<\infty ∫ 0 t S ( t − s ) ( m ( s ) p − 1 V ( s )) d s ≤ ( ∫ 0 t m p − 1 ) V ( t ) < ∞ , so Φ [ u ] \Phi[u] Φ [ u ] is well defined and measurable.
Φ \Phi Φ maps K \mathcal{K} K into itself. Using ∣ S ( τ ) f ∣ ≤ S ( τ ) ∣ f ∣ |S(\tau)f|\le S(\tau)|f| ∣ S ( τ ) f ∣ ≤ S ( τ ) ∣ f ∣ , the definition of V V V , (42) , (40) and (10) ,
∣ Φ [ u ] ( t ) ∣ ≤ V ( t ) + κ p ∫ 0 t S ( t − s ) V ( s ) p d s ≤ V ( t ) + κ p ∫ 0 t m ( s ) p − 1 S ( t − s ) V ( s ) d s ≤ ( 1 + κ p c p ) V ( t ) = κ V ( t ) . (43) \begin{aligned}
|\Phi[u](t)|\le V(t)+\kappa^p\int_0^t S(t-s)V(s)^p\,ds \\
&\le V(t)+\kappa^p\int_0^t m(s)^{p-1}S(t-s)V(s)\,ds\le(1+\kappa^pc_p)V(t)=\kappa V(t). \tag*{(43)}
\end{aligned} ∣Φ [ u ] ( t ) ∣ ≤ V ( t ) + κ p ∫ 0 t S ( t − s ) V ( s ) p d s ≤ V ( t ) + κ p ∫ 0 t m ( s ) p − 1 S ( t − s ) V ( s ) d s ≤ ( 1 + κ p c p ) V ( t ) = κV ( t ) . (43)
since κ p c p = p p ( p − 1 ) − p ( p − 1 ) p − 1 p − p = 1 p − 1 \kappa^pc_p=p^p(p-1)^{-p}(p-1)^{p-1}p^{-p}=\frac{1}{p-1} κ p c p = p p ( p − 1 ) − p ( p − 1 ) p − 1 p − p = p − 1 1 and 1 + 1 p − 1 = κ 1+\frac{1}{p-1}=\kappa 1 + p − 1 1 = κ .
Φ \Phi Φ is a contraction. For u , v ∈ K u,v\in\mathcal{K} u , v ∈ K , the mean value theorem gives
∣ ∣ u ∣ p − ∣ v ∣ p ∣ ≤ p max { ∣ u ∣ , ∣ v ∣ } p − 1 ∣ u − v ∣ ≤ a ∣ u − v ∣ , \bigl||u|^p-|v|^p\bigr|\le p\max\{|u|,|v|\}^{p-1}|u-v|\le a|u-v|, ∣ u ∣ p − ∣ v ∣ p ≤ p max { ∣ u ∣ , ∣ v ∣ } p − 1 ∣ u − v ∣ ≤ a ∣ u − v ∣ ,
and by definition of d d d , ∣ u − v ∣ ( s ) ≤ d ( u , v ) e 2 A ( s ) V ( s ) |u-v|(s)\le d(u,v)e^{2A(s)}V(s) ∣ u − v ∣ ( s ) ≤ d ( u , v ) e 2 A ( s ) V ( s ) a.e. Hence, by (40) ,
∣ Φ [ u ] − Φ [ v ] ∣ ( t ) ≤ ∫ 0 t a ( s ) S ( t − s ) ∣ u − v ∣ ( s ) d s ≤ d ( u , v ) V ( t ) ∫ 0 t a ( s ) e 2 A ( s ) d s = d ( u , v ) V ( t ) e 2 A ( t ) − 1 2 , (44) \begin{aligned}
|\Phi[u]-\Phi[v]|(t)\le\int_0^t a(s)S(t-s)|u-v|(s)\,ds \\
&\le d(u,v)V(t)\int_0^t a(s)e^{2A(s)}\,ds=d(u,v)V(t)\frac{e^{2A(t)}-1}{2}, \tag*{(44)}
\end{aligned} ∣Φ [ u ] − Φ [ v ] ∣ ( t ) ≤ ∫ 0 t a ( s ) S ( t − s ) ∣ u − v ∣ ( s ) d s ≤ d ( u , v ) V ( t ) ∫ 0 t a ( s ) e 2 A ( s ) d s = d ( u , v ) V ( t ) 2 e 2 A ( t ) − 1 , (44)
so that d ( Φ [ u ] , Φ [ v ] ) ≤ 1 2 d ( u , v ) d(\Phi[u],\Phi[v])\le\frac12d(u,v) d ( Φ [ u ] , Φ [ v ]) ≤ 2 1 d ( u , v ) .
Conclusion. By Banach’s fixed point theorem Φ \Phi Φ has a unique fixed point u ∈ K u\in\mathcal{K} u ∈ K . For every T ′ < T T'<T T ′ < T we have ∣ u ∣ ≤ κ V ≤ κ ( ∥ u 0 ∥ ∞ + ∥ w ∥ ∞ ∫ 0 T ′ ∣ ζ ∣ ) |u|\le\kappa V\le\kappa\left(\|u_0\|_\infty+\|w\|_\infty\int_0^{T'}|\zeta|\right) ∣ u ∣ ≤ κV ≤ κ ( ∥ u 0 ∥ ∞ + ∥ w ∥ ∞ ∫ 0 T ′ ∣ ζ ∣ ) on Q T ′ Q_{T'} Q T ′ and ∥ u ( t ) ∥ 1 ≤ κ ∥ V ( t ) ∥ 1 ≤ κ ( ∥ u 0 ∥ 1 + ∥ w ∥ 1 ∫ 0 T ′ ∣ ζ ∣ ) \|u(t)\|_1\le\kappa\|V(t)\|_1\le\kappa\left(\|u_0\|_1+\|w\|_1\int_0^{T'}|\zeta|\right) ∥ u ( t ) ∥ 1 ≤ κ ∥ V ( t ) ∥ 1 ≤ κ ( ∥ u 0 ∥ 1 + ∥ w ∥ 1 ∫ 0 T ′ ∣ ζ ∣ ) ; hence u u u is a mild solution on ( 0 , T ) (0,T) ( 0 , T ) satisfying (11) . Uniqueness among all mild solutions follows from Lemma 2.3. □
Remark 5.1. The constant c p c_p c p is the best one allowed by the argument: a supersolution of the form κ V \kappa V κV requires κ p ∫ 0 T m p − 1 ≤ κ − 1 \kappa^p\int_0^T m^{p-1}\le\kappa-1 κ p ∫ 0 T m p − 1 ≤ κ − 1 , and max κ > 1 ( κ − 1 ) κ − p = ( p − 1 ) p − 1 p − p \max_{\kappa>1}(\kappa-1)\kappa^{-p}=(p-1)^{p-1}p^{-p} max κ > 1 ( κ − 1 ) κ − p = ( p − 1 ) p − 1 p − p is attained at κ = p / ( p − 1 ) \kappa=p/(p-1) κ = p / ( p − 1 ) . If u 0 ≢ 0 u_0\not\equiv0 u 0 ≡ 0 , then m ( t ) ≥ ∥ S ( t ) ∣ u 0 ∣ ∥ ∞ ≥ c t − N / 2 m(t)\ge\|S(t)|u_0|\|_\infty\ge ct^{-N/2} m ( t ) ≥ ∥ S ( t ) ∣ u 0 ∣ ∥ ∞ ≥ c t − N /2 for large t t t , and similarly if ζ ⊗ w ≢ 0 \zeta\otimes w\not\equiv0 ζ ⊗ w ≡ 0 ; therefore (10) with T = ∞ T=\infty T = ∞ forces p > p F p>p_F p > p F , in agreement with Theorem 1.2.
Proof of Corollary 1.1
Let H N , p : = ∫ 0 ∞ h N p d t H_{N,p}:=\int_0^\infty h_N^p\,dt H N , p := ∫ 0 ∞ h N p d t , which is finite since p > p F p>p_F p > p F (Lemma 2.1(f)). By (41) and the elementary inequality ( α + β ) p − 1 ≤ 2 ( p − 2 ) + ( α p − 1 + β p − 1 ) (\alpha+\beta)^{p-1}\le2^{(p-2)_+}(\alpha^{p-1}+\beta^{p-1}) ( α + β ) p − 1 ≤ 2 ( p − 2 ) + ( α p − 1 + β p − 1 ) for α , β ≥ 0 \alpha,\beta\ge0 α , β ≥ 0 ,
∫ 0 ∞ m p − 1 d t ≤ 2 ( p − 2 ) + ( H N , p [ [ u 0 ] ] p − 1 + ω ζ [ [ w ] ] p − 1 ) ≤ 2 ( p − 2 ) + max { 1 , H N , p } ( [ [ u 0 ] ] p − 1 + ω ζ [ [ w ] ] p − 1 ) . \int_0^\infty m^{p-1}\,dt\le2^{(p-2)_+}\left(H_{N,p}[[u_0]]^{p-1}+\omega_\zeta[[w]]^{p-1}\right)
\le2^{(p-2)_+}\max\{1,H_{N,p}\}\left([[u_0]]^{p-1}+\omega_\zeta[[w]]^{p-1}\right). ∫ 0 ∞ m p − 1 d t ≤ 2 ( p − 2 ) + ( H N , p [[ u 0 ] ] p − 1 + ω ζ [[ w ] ] p − 1 ) ≤ 2 ( p − 2 ) + max { 1 , H N , p } ( [[ u 0 ] ] p − 1 + ω ζ [[ w ] ] p − 1 ) .
With δ : = c p 2 − ( p − 2 ) + / max { 1 , H N , p } \delta:=c_p2^{-(p-2)_+}/\max\{1,H_{N,p}\} δ := c p 2 − ( p − 2 ) + / max { 1 , H N , p } , the smallness assumption implies (10) with T = ∞ T=\infty T = ∞ , and Theorem 1.5 applies. The bounds on u ( t ) u(t) u ( t ) follow from ∣ u ∣ ≤ κ V |u|\le\kappa V ∣ u ∣ ≤ κV , (40) and (41) , and from ∥ V ( t ) ∥ 1 ≤ ∥ u 0 ∥ 1 + ∥ w ∥ 1 ∫ 0 t ∣ ζ ∣ \|V(t)\|_1\le\|u_0\|_1+\|w\|_1\int_0^t|\zeta| ∥ V ( t ) ∥ 1 ≤ ∥ u 0 ∥ 1 + ∥ w ∥ 1 ∫ 0 t ∣ ζ ∣ .
It remains to verify ω ζ < ∞ \omega_\zeta<\infty ω ζ < ∞ in cases (a)–(c). In case (a), Lemma 2.7(iii) with q = p − 1 ≥ 1 q=p-1\ge1 q = p − 1 ≥ 1 gives ω ζ 1 / ( p − 1 ) ≤ ∥ ζ ∥ 1 ∥ h N ∥ L p − 1 ( 0 , ∞ ) < ∞ \omega_\zeta^{1/(p-1)}\le\|\zeta\|_1\|h_N\|_{L^{p-1}(0,\infty)}<\infty ω ζ 1/ ( p − 1 ) ≤ ∥ ζ ∥ 1 ∥ h N ∥ L p − 1 ( 0 , ∞ ) < ∞ . In case (b), Lemma 2.7(i) gives ( H K ) (H_K) ( H K ) , so ω ζ ≤ K p − 1 H N , p \omega_\zeta\le K^{p-1}H_{N,p} ω ζ ≤ K p − 1 H N , p , and (41) yields ∥ u ( t ) ∥ ∞ ≤ κ ( [ [ u 0 ] ] + K [ [ w ] ] ) h N ( t ) ≤ C ( 1 + t ) − N / 2 \|u(t)\|_\infty\le\kappa([[u_0]]+K[[w]])h_N(t)\le C(1+t)^{-N/2} ∥ u ( t ) ∥ ∞ ≤ κ ([[ u 0 ]] + K [[ w ]]) h N ( t ) ≤ C ( 1 + t ) − N /2 , because h N ( t ) ≤ 2 N / 2 ( 1 + 4 π t ) − N / 2 h_N(t)\le2^{N/2}(1+4\pi t)^{-N/2} h N ( t ) ≤ 2 N /2 ( 1 + 4 π t ) − N /2 . In case (c), Lemma 2.7(ii) gives ( ∣ ζ ∣ ⋆ h N ) p − 1 ≤ C ( 1 + t ) − ( p − 1 ) min { b , N / 2 } (|\zeta|\star h_N)^{p-1}\le C(1+t)^{-(p-1)\min\{b,N/2\}} ( ∣ ζ ∣ ⋆ h N ) p − 1 ≤ C ( 1 + t ) − ( p − 1 ) m i n { b , N /2 } , which is integrable because ( p − 1 ) b > 1 (p-1)b>1 ( p − 1 ) b > 1 and ( p − 1 ) N / 2 > 1 (p-1)N/2>1 ( p − 1 ) N /2 > 1 . □
Remark 5.2. For N ≤ 2 N\le2 N ≤ 2 one has p F ≥ 2 p_F\ge2 p F ≥ 2 , so case (a) covers every p > p F p>p_F p > p F and every ζ ∈ L 1 ( 0 , ∞ ) \zeta\in L^1(0,\infty) ζ ∈ L 1 ( 0 , ∞ ) . For N ≥ 3 N\ge3 N ≥ 3 and p F < p < 2 p_F<p<2 p F < p < 2 the corollary requires some decay of ζ \zeta ζ , for instance ∣ ζ ( t ) ∣ ≲ t − b |\zeta(t)|\lesssim t^{-b} ∣ ζ ( t ) ∣ ≲ t − b with b > 1 / ( p − 1 ) b>1/(p-1) b > 1/ ( p − 1 ) . The reason is that the domination principle controls ∣ u ∣ p |u|^p ∣ u ∣ p through ∥ V ( t ) ∥ ∞ p \|V(t)\|_\infty^p ∥ V ( t ) ∥ ∞ p , and the near-field part of the forced flow, of size ∣ ζ ( t ) ∣ |\zeta(t)| ∣ ζ ( t ) ∣ near the support of w w w , decays only as fast as ζ \zeta ζ itself. Whether small-data global existence holds for every integrable ζ \zeta ζ when p F < p < 2 p_F < p < 2 p F < p < 2 is left open (Problem 9.3).
Proof of Theorem 1.3
Set v 0 : = ∣ u 0 ∣ = − u 0 v_0 := |u_0| = -u_0 v 0 := ∣ u 0 ∣ = − u 0 and g ( x , t ) : = ∣ ζ ( t ) w ( x ) ∣ = − ζ ( t ) w ( x ) ≥ 0 g(x,t) := |\zeta(t)w(x)| = -\zeta(t)w(x) \ge0 g ( x , t ) := ∣ ζ ( t ) w ( x ) ∣ = − ζ ( t ) w ( x ) ≥ 0 . Then V = S ( t ) v 0 + D [ g ] V = S(t)v_0 + D[g] V = S ( t ) v 0 + D [ g ] , and
0 ≤ V ≤ M V : = ∥ u 0 ∥ ∞ + ∥ ζ ∥ 1 ∥ w ∥ ∞ in Q . 0 \le V \le M_V := \|u_0\|_\infty+ \|\zeta\|_1\|w\|_\infty\qquad\text{in } Q. 0 ≤ V ≤ M V := ∥ u 0 ∥ ∞ + ∥ ζ ∥ 1 ∥ w ∥ ∞ in Q .
Let λ : = p M V p − 1 \lambda:= pM_V^{p-1} λ := p M V p − 1 and θ ( r ) : = λ r − r p \theta(r) := \lambda r - r^p θ ( r ) := λ r − r p . On [ 0 , M V ] [0,M_V] [ 0 , M V ] the function θ \theta θ is nonnegative and nondecreasing, since θ ( r ) = r ( λ − r p − 1 ) \theta(r) = r(\lambda-r^{p-1}) θ ( r ) = r ( λ − r p − 1 ) and θ ′ ( r ) = λ − p r p − 1 \theta'(r) = \lambda-pr^{p-1} θ ′ ( r ) = λ − p r p − 1 . For measurable v v v with 0 ≤ v ≤ V 0 \le v \le V 0 ≤ v ≤ V we define
Φ [ v ] ( t ) : = e − λ t S ( t ) v 0 + ∫ 0 t e − λ ( t − s ) S ( t − s ) ( g ( s ) + θ ( v ( s ) ) ) d s . \Phi[v](t) := e^{-\lambda t}S(t)v_0 + \int_0^t e^{-\lambda(t-s)}S(t-s)(g(s)+\theta(v(s)))\,ds. Φ [ v ] ( t ) := e − λ t S ( t ) v 0 + ∫ 0 t e − λ ( t − s ) S ( t − s ) ( g ( s ) + θ ( v ( s ))) d s .
Clearly Φ [ v ] ≥ 0 \Phi[v] \ge0 Φ [ v ] ≥ 0 , and Φ \Phi Φ is order preserving on the order interval { 0 ≤ v ≤ V } \{0 \le v \le V\} { 0 ≤ v ≤ V } because θ \theta θ is nondecreasing on [ 0 , M V ] [0,M_V] [ 0 , M V ] and S ( τ ) S(\tau) S ( τ ) is positivity preserving. Moreover, since θ ( V ) ≤ λ V \theta(V) \le\lambda V θ ( V ) ≤ λV , Lemma 2.4 applied on each Q T ′ Q_{T'} Q T ′ with F = g F=g F = g (note that ∫ 0 T ′ ∥ g ( s ) ∥ ∞ d s ≤ ∥ ζ ∥ 1 ∥ w ∥ ∞ \int_0^{T'}\|g(s)\|_\infty\,ds \le\|\zeta\|_1\|w\|_\infty ∫ 0 T ′ ∥ g ( s ) ∥ ∞ d s ≤ ∥ ζ ∥ 1 ∥ w ∥ ∞ ) gives
Φ [ V ] ( t ) ≤ e − λ t S ( t ) v 0 + ∫ 0 t e − λ ( t − s ) S ( t − s ) ( g ( s ) + λ V ( s ) ) d s = V ( t ) . \Phi[V](t) \le e^{-\lambda t}S(t)v_0 + \int_0^t e^{-\lambda(t-s)}S(t-s)(g(s)+\lambda V(s))\,ds = V(t). Φ [ V ] ( t ) ≤ e − λ t S ( t ) v 0 + ∫ 0 t e − λ ( t − s ) S ( t − s ) ( g ( s ) + λV ( s )) d s = V ( t ) .
Thus Φ \Phi Φ maps { 0 ≤ v ≤ V } \{0 \le v \le V\} { 0 ≤ v ≤ V } into itself. Define v 0 : = 0 v^0 := 0 v 0 := 0 and v n + 1 : = Φ [ v n ] v^{n+1} := \Phi[v^n] v n + 1 := Φ [ v n ] . By induction, 0 ≤ v n ≤ v n + 1 ≤ V 0 \le v^n \le v^{n+1} \le V 0 ≤ v n ≤ v n + 1 ≤ V for all n n n : indeed v 1 = Φ [ 0 ] ≥ 0 = v 0 v^1 = \Phi[0] \ge0 = v^0 v 1 = Φ [ 0 ] ≥ 0 = v 0 , and v n ≤ v n + 1 v^n \le v^{n+1} v n ≤ v n + 1 implies v n + 1 = Φ [ v n ] ≤ Φ [ v n + 1 ] = v n + 2 v^{n+1} = \Phi[v^n] \le\Phi[v^{n+1}] = v^{n+2} v n + 1 = Φ [ v n ] ≤ Φ [ v n + 1 ] = v n + 2 . Let v : = lim n v n v := \lim_n v^n v := lim n v n . Since θ \theta θ is continuous and nondecreasing on [ 0 , M V ] [0,M_V] [ 0 , M V ] , the integrands g + θ ( v n ) g+\theta(v^n) g + θ ( v n ) are nonnegative and increase to g + θ ( v ) g+\theta(v) g + θ ( v ) , so the monotone convergence theorem gives v = Φ [ v ] v=\Phi[v] v = Φ [ v ] .
As v v v is bounded and F : = g − v p F:=g-v^p F := g − v p satisfies ∫ 0 T ′ ∥ F ( s ) ∥ ∞ d s < ∞ \int_0^{T'}\|F(s)\|_\infty\,ds<\infty ∫ 0 T ′ ∥ F ( s ) ∥ ∞ d s < ∞ , and since g + θ ( v ) = F + λ v g+\theta(v)=F+\lambda v g + θ ( v ) = F + λ v , the converse part of Lemma 2.4 yields
v ( t ) = S ( t ) v 0 + ∫ 0 t S ( t − s ) ( g ( s ) − v ( s ) p ) d s a.e. in Q . v(t)=S(t)v_0+\int_0^t S(t-s)(g(s)-v(s)^p)\,ds \qquad\text{a.e. in } Q. v ( t ) = S ( t ) v 0 + ∫ 0 t S ( t − s ) ( g ( s ) − v ( s ) p ) d s a.e. in Q .
Setting u : = − v u:=-v u := − v and recalling − v 0 = u 0 -v_0=u_0 − v 0 = u 0 , − g = ζ w -g=\zeta w − g = ζ w and v p = ∣ u ∣ p v^p=|u|^p v p = ∣ u ∣ p , we obtain (14) . Since − V ≤ u ≤ 0 -V\le u\le0 − V ≤ u ≤ 0 , u u u is bounded on Q Q Q and ∥ u ( t ) ∥ 1 ≤ ∥ V ( t ) ∥ 1 ≤ ∥ u 0 ∥ 1 + ∥ ζ ∥ 1 ∥ w ∥ 1 \|u(t)\|_1\le\|V(t)\|_1\le\|u_0\|_1+\|\zeta\|_1\|w\|_1 ∥ u ( t ) ∥ 1 ≤ ∥ V ( t ) ∥ 1 ≤ ∥ u 0 ∥ 1 + ∥ ζ ∥ 1 ∥ w ∥ 1 ; hence u u u is a global mild solution, unique by Lemma 2.3.
Assume finally 1 < p ≤ p F 1<p\le p_F 1 < p ≤ p F . By Lemma 2.2(a), u u u is a global weak solution, and Theorem 1.1 gives ∫ Q ∣ u ∣ p = − M = ∥ u 0 ∥ 1 − Z W \int_Q|u|^p=-M=\|u_0\|_1-ZW ∫ Q ∣ u ∣ p = − M = ∥ u 0 ∥ 1 − Z W . Since ζ w ≤ 0 \zeta w\le0 ζ w ≤ 0 we have − Z W = − ∫ Q ζ w = ∥ ζ w ∥ L 1 ( Q ) -ZW=-\int_Q\zeta w=\|\zeta w\|_{L^1(Q)} − Z W = − ∫ Q ζ w = ∥ ζ w ∥ L 1 ( Q ) . The convergence ∥ u ( t ) ∥ 1 → 0 \|u(t)\|_1\to0 ∥ u ( t ) ∥ 1 → 0 is a special case of Theorem 1.4(ii), proved in Section 7 independently of the present theorem. □
Sharp lifespan in the Fujita range
Throughout this section 1 < p ≤ p F 1<p\le p_F 1 < p ≤ p F , ζ ∈ L 1 ( 0 , ∞ ) \zeta\in L^1(0,\infty) ζ ∈ L 1 ( 0 , ∞ ) , a , b ∈ L 1 ( R N ) ∩ L ∞ ( R N ) a,b\in L^1(\mathbb{R}^N)\cap L^\infty(\mathbb{R}^N) a , b ∈ L 1 ( R N ) ∩ L ∞ ( R N ) , and for ε > 0 \varepsilon>0 ε > 0 we consider (1.1) with u 0 = ε a u_0=\varepsilon a u 0 = ε a and w = ε b w=\varepsilon b w = ε b . By Lemma 2.3, mild solutions on different intervals coincide on their common interval, so T ε T_\varepsilon T ε is the maximal existence time of the mild solution.
Upper bounds
Assume M ∗ = ∫ a + Z ∫ b > 0 M_*=\int a+Z\int b>0 M ∗ = ∫ a + Z ∫ b > 0 , let ε ∈ ( 0 , 1 ] \varepsilon\in(0,1] ε ∈ ( 0 , 1 ] , and suppose that (1.1) has a mild solution u u u on ( 0 , T ) (0,T) ( 0 , T ) . By Lemma 2.2(a), u u u is a weak solution on Q T Q_T Q T . For 0 < R < R T : = T / 2 0<R<R_T:=\sqrt{T/2} 0 < R < R T := T /2 the parabolic cutoff Ψ R \Psi_R Ψ R is supported in { ∣ x ∣ 2 + t ≤ 2 R 2 } \{|x|^2+t\le2R^2\} { ∣ x ∣ 2 + t ≤ 2 R 2 } , a compact subset of R N × [ 0 , T ) \mathbb{R}^N\times[0,T) R N × [ 0 , T ) , and is therefore an admissible test function on Q T Q_T Q T . Exactly as in (3.1) we obtain
J ( R ) + ε c ( R ) = − ∬ Q T u L Ψ R , c ( R ) : = ∫ R N a η ( ∣ x ∣ 2 / R 2 ) k d x + ∬ Q ζ b Ψ R (45) J(R)+\varepsilon c(R)=-\iint_{Q_T}u\,\mathcal{L}\Psi_R,\qquad c(R):=\int_{\mathbb{R}^N}a\,\eta(|x|^2/R^2)^k\,dx+\iint_{Q}\zeta b\,\Psi_R \tag*{(45)} J ( R ) + ε c ( R ) = − ∬ Q T u L Ψ R , c ( R ) := ∫ R N a η ( ∣ x ∣ 2 / R 2 ) k d x + ∬ Q ζ b Ψ R (45)
where J ( R ) = ∬ Q T ∣ u ∣ p Ψ R J(R)=\iint_{Q_T}|u|^p\Psi_R J ( R ) = ∬ Q T ∣ u ∣ p Ψ R . By dominated convergence c ( R ) → M ∗ c(R)\to M_* c ( R ) → M ∗ as R → ∞ R\to\infty R → ∞ , so there is R 1 ≥ 1 R_1\ge1 R 1 ≥ 1 , depending only on a , b , ζ a,b,\zeta a , b , ζ and η \eta η , such that c ( R ) ≥ M ∗ / 2 c(R)\ge M_*/2 c ( R ) ≥ M ∗ /2 for all R ≥ R 1 R\ge R_1 R ≥ R 1 . If T ≤ 2 R 1 2 T\le2R_1^2 T ≤ 2 R 1 2 there is nothing to prove, so we assume R T > R 1 R_T>R_1 R T > R 1 .
The subcritical case p < p F p < p_F p < p F . The argument leading to (33) applies verbatim on Q T Q_T Q T and gives, for R 1 ≤ R < R T R_1 \le R < R_T R 1 ≤ R < R T ,
ε M ∗ 2 ≤ ε c ( R ) ≤ C 2 p ′ p ′ R − ( 2 p ′ − N − 2 ) . \frac{\varepsilon M_*}{2} \le\varepsilon c(R) \le\frac{C_2^{p'}}{p'} R^{-(2p'-N-2)}. 2 ε M ∗ ≤ ε c ( R ) ≤ p ′ C 2 p ′ R − ( 2 p ′ − N − 2 ) .
Letting R ↑ R T R \uparrow R_T R ↑ R T we get T ≤ 2 ( 2 C 2 p ′ / ( p ′ M ∗ ε ) ) 2 / ( 2 p ′ − N − 2 ) T \le2(2C_2^{p'}/(p'M_*\varepsilon))^{2/(2p'-N-2)} T ≤ 2 ( 2 C 2 p ′ / ( p ′ M ∗ ε ) ) 2/ ( 2 p ′ − N − 2 ) , and 2 2 p ′ − N − 2 = 2 ( p − 1 ) 2 − N ( p − 1 ) \frac{2}{2p'-N-2}=\frac{2(p-1)}{2-N(p-1)} 2 p ′ − N − 2 2 = 2 − N ( p − 1 ) 2 ( p − 1 ) because 2 p ′ − 2 = 2 p − 1 2p'-2=\frac{2}{p-1} 2 p ′ − 2 = p − 1 2 . Hence T ≤ max { 2 R 1 2 , C ε − 2 ( p − 1 ) / ( 2 − N ( p − 1 ) ) } ≤ C ′ ε − 2 ( p − 1 ) / ( 2 − N ( p − 1 ) ) T \le\max\{2R_1^2,C\varepsilon^{-2(p-1)/(2-N(p-1))}\} \le C'\varepsilon^{-2(p-1)/(2-N(p-1))} T ≤ max { 2 R 1 2 , C ε − 2 ( p − 1 ) / ( 2 − N ( p − 1 )) } ≤ C ′ ε − 2 ( p − 1 ) / ( 2 − N ( p − 1 )) for ε ∈ ( 0 , 1 ] \varepsilon\in(0,1] ε ∈ ( 0 , 1 ] , and the same bound holds for T ε T_\varepsilon T ε .
The critical case p = p F p=p_F p = p F . Now N + 2 − 2 p ′ = 0 N+2-2p'=0 N + 2 − 2 p ′ = 0 and (26) reads
J ( R ) + ε c ( R ) ≤ C 2 y ( R ) 1 / p , y ( ρ ) : = ∬ Q T ∣ u ∣ p η ∗ ( s ρ ) d x d t , η ∗ : = η k 1 [ 1 , 2 ] , (46) J(R)+\varepsilon c(R)\le C_2 y(R)^{1/p},\qquad y(\rho):=\iint_{Q_T}|u|^p\eta_*(s_\rho)\,dx\,dt,\qquad\eta_*:=\eta^k\mathbf{1}_{[1,2]}, \tag*{(46)} J ( R ) + ε c ( R ) ≤ C 2 y ( R ) 1/ p , y ( ρ ) := ∬ Q T ∣ u ∣ p η ∗ ( s ρ ) d x d t , η ∗ := η k 1 [ 1 , 2 ] , (46)
because Ψ R 1 1 D R = η ∗ ( s R ) \Psi_{R_1}\mathbf{1}_{D_R}=\eta_*(s_R) Ψ R 1 1 D R = η ∗ ( s R ) . Following the idea of Ikeda and Sobajima [19 ] , we introduce
Y ( R ) : = ∫ 0 R y ( ρ ) d ρ ρ , 0 < R < R T . Y(R):=\int_0^R y(\rho)\frac{d\rho}{\rho},\qquad0<R<R_T. Y ( R ) := ∫ 0 R y ( ρ ) ρ d ρ , 0 < R < R T .
By Tonelli’s theorem, Y ( R ) = ∬ Q T ∣ u ∣ p Θ R ( x , t ) d x d t Y(R)=\iint_{Q_T}|u|^p\Theta_R(x,t)\,dx\,dt Y ( R ) = ∬ Q T ∣ u ∣ p Θ R ( x , t ) d x d t with Θ R ( x , t ) : = ∫ 0 R η ∗ ( ( ∣ x ∣ 2 + t ) / ρ 2 ) d ρ ρ \Theta_R(x,t):=\int_0^R\eta_*( (|x|^2+t)/\rho^2 )\frac{d\rho}{\rho} Θ R ( x , t ) := ∫ 0 R η ∗ (( ∣ x ∣ 2 + t ) / ρ 2 ) ρ d ρ . For r : = ∣ x ∣ 2 + t > 0 r:=|x|^2+t>0 r := ∣ x ∣ 2 + t > 0 the substitution ς = r / ρ 2 \varsigma=r/\rho^2 ς = r / ρ 2 , d ρ ρ = − d ς 2 ς \frac{d\rho}{\rho}=-\frac{d\varsigma}{2\varsigma} ρ d ρ = − 2 ς d ς , gives
Θ R ( x , t ) = 1 2 ∫ r / R 2 ∞ η ∗ ( ς ) d ς ς ≤ 1 2 η ( r / R 2 ) k ∫ 1 2 d ς ς = log 2 2 Ψ R ( x , t ) , \Theta_R(x,t)=\frac{1}{2}\int_{r/R^2}^{\infty}\eta_*(\varsigma)\frac{d\varsigma}{\varsigma}\le\frac{1}{2}\eta(r/R^2)^k\int_1^2\frac{d\varsigma}{\varsigma}=\frac{\log2}{2}\Psi_R(x,t), Θ R ( x , t ) = 2 1 ∫ r / R 2 ∞ η ∗ ( ς ) ς d ς ≤ 2 1 η ( r / R 2 ) k ∫ 1 2 ς d ς = 2 log 2 Ψ R ( x , t ) ,
where we used that η ∗ \eta_* η ∗ vanishes outside [ 1 , 2 ] [1,2] [ 1 , 2 ] and that η ( ς ) ≤ η ( r / R 2 ) \eta(\varsigma)\le\eta(r/R^2) η ( ς ) ≤ η ( r / R 2 ) for ς ≥ r / R 2 \varsigma\ge r/R^2 ς ≥ r / R 2 . Consequently
Y ( R ) ≤ log 2 2 J ( R ) < ∞ . (47) Y(R)\le\frac{\log2}{2}J(R)<\infty. \tag*{(47)} Y ( R ) ≤ 2 log 2 J ( R ) < ∞. (47)
In particular y ( ρ ) / ρ y(\rho)/\rho y ( ρ ) / ρ is integrable on ( 0 , R ] (0,R] ( 0 , R ] for every R < R T R<R_T R < R T , so Y Y Y is absolutely continuous on [ R 1 , R ] [R_1,R] [ R 1 , R ] with Y ′ ( R ) = y ( R ) / R Y'(R)=y(R)/R Y ′ ( R ) = y ( R ) / R for a.e. R R R . Combining (46) , (47) and c ( R ) ≥ M ∗ / 2 c(R)\ge M_*/2 c ( R ) ≥ M ∗ /2 , we find for a.e. R ∈ [ R 1 , R T ) R\in[R_1,R_T) R ∈ [ R 1 , R T )
2 log 2 Y ~ ( R ) ≤ C 2 ( R Y ~ ′ ( R ) ) 1 / p , Y ~ ( R ) : = Y ( R ) + log 2 4 M ∗ ε > 0. \frac{2}{\log2}\widetilde{Y}(R)\le C_2(R\widetilde{Y}'(R))^{1/p},\qquad\widetilde{Y}(R):=Y(R)+\frac{\log2}{4}M_*\varepsilon>0. log 2 2 Y ( R ) ≤ C 2 ( R Y ′ ( R ) ) 1/ p , Y ( R ) := Y ( R ) + 4 log 2 M ∗ ε > 0.
Thus Y ~ ′ ( R ) ≥ c 3 Y ~ ( R ) p / R \widetilde{Y}'(R)\ge c_3\widetilde{Y}(R)^p/R Y ′ ( R ) ≥ c 3 Y ( R ) p / R with c 3 : = ( 2 / ( C 2 log 2 ) ) p c_3:=(2/(C_2\log2))^p c 3 := ( 2/ ( C 2 log 2 ) ) p . Since Y ~ \widetilde{Y} Y is absolutely continuous and bounded below by a positive constant, Y ~ 1 − p \widetilde{Y}^{1-p} Y 1 − p is absolutely continuous and ( Y ~ 1 − p ) ′ = ( 1 − p ) Y ~ − p Y ~ ′ ≤ − ( p − 1 ) c 3 / R (\widetilde{Y}^{1-p})'=(1-p)\widetilde{Y}^{-p}\widetilde{Y}'\le-(p-1)c_3/R ( Y 1 − p ) ′ = ( 1 − p ) Y − p Y ′ ≤ − ( p − 1 ) c 3 / R a.e. Integrating over [ R 1 , R ] [R_1,R] [ R 1 , R ] and using Y ~ ( R ) 1 − p > 0 \widetilde{Y}(R)^{1-p}>0 Y ( R ) 1 − p > 0 ,
( p − 1 ) c 3 log R R 1 < Y ~ ( R 1 ) 1 − p ≤ ( log 2 4 M ∗ ε ) 1 − p . (p-1)c_3\log\frac{R}{R_1}<\widetilde{Y}(R_1)^{1-p}\le\left(\frac{\log2}{4}M_*\varepsilon\right)^{1-p}. ( p − 1 ) c 3 log R 1 R < Y ( R 1 ) 1 − p ≤ ( 4 log 2 M ∗ ε ) 1 − p .
Letting R ↑ R T R\uparrow R_T R ↑ R T we obtain T ≤ 2 R 1 2 exp ( 2 C 4 ε − ( p − 1 ) ) T\le2R_1^2\exp(2C_4\varepsilon^{-(p-1)}) T ≤ 2 R 1 2 exp ( 2 C 4 ε − ( p − 1 ) ) with C 4 : = ( log 2 4 M ∗ ) 1 − p / ( ( p − 1 ) c 3 ) C_4:=\left(\frac{\log2}{4}M_*\right)^{1-p}/((p-1)c_3) C 4 := ( 4 l o g 2 M ∗ ) 1 − p / (( p − 1 ) c 3 ) , and therefore T ε ≤ exp ( C ε − ( p − 1 ) ) T_\varepsilon\le\exp(C\varepsilon^{-(p-1)}) T ε ≤ exp ( C ε − ( p − 1 ) ) for ε ∈ ( 0 , 1 ] \varepsilon\in(0,1] ε ∈ ( 0 , 1 ] , with C : = 2 C 4 + log + ( 2 R 1 2 ) C:=2C_4+\log_+(2R_1^2) C := 2 C 4 + log + ( 2 R 1 2 ) . This proves Theorem 1.7 (i).
Lower bounds
Assume (HK) (by Lemma 2.7(i) this is implied by (Ha _a a )), and set Λ : = [ [ a ] ] + K [ [ b ] ] \Lambda:=[[a]]+K[[b]] Λ := [[ a ]] + K [[ b ]] . If Λ = 0 \Lambda=0 Λ = 0 the data vanish and T ε = ∞ T_\varepsilon=\infty T ε = ∞ , so let Λ > 0 \Lambda>0 Λ > 0 . By (41) with u 0 = ε a u_0=\varepsilon a u 0 = ε a , w = ε b w=\varepsilon b w = ε b ,
m ( t ) ≤ ε ( [ [ a ] ] h N ( t ) + [ [ b ] ] ( ∣ ζ ∣ ⋆ h N ) ( t ) ) ≤ ε Λ h N ( t ) , m(t)\le\varepsilon\bigl([[a]]h_N(t)+[[b]](|\zeta|\star h_N)(t)\bigr)\le\varepsilon\Lambda h_N(t), m ( t ) ≤ ε ( [[ a ]] h N ( t ) + [[ b ]] ( ∣ ζ ∣ ⋆ h N ) ( t ) ) ≤ ε Λ h N ( t ) ,
hence
∫ 0 T m p − 1 d t ≤ ( ε Λ ) p − 1 H ( T ) , H ( T ) : = ∫ 0 T h N p − 1 d t . \int_0^T m^{p-1}\,dt\le(\varepsilon\Lambda)^{p-1}H(T),\qquad H(T):=\int_0^T h_N^{p-1}\,dt. ∫ 0 T m p − 1 d t ≤ ( ε Λ ) p − 1 H ( T ) , H ( T ) := ∫ 0 T h N p − 1 d t .
By Theorem 1.5 , T ε ≥ T T_\varepsilon\ge T T ε ≥ T whenever ( ε Λ ) p − 1 H ( T ) ≤ c p (\varepsilon\Lambda)^{p-1}H(T)\le c_p ( ε Λ ) p − 1 H ( T ) ≤ c p . Let β : = N ( p − 1 ) / 2 ≤ 1 \beta:=N(p-1)/2\le1 β := N ( p − 1 ) /2 ≤ 1 . If p < p F p<p_F p < p F , then β < 1 \beta<1 β < 1 and H ( T ) ≤ ∫ 0 T ( 4 π t ) − β d t = ( 4 π ) − β 1 − β T 1 − β H(T)\le\int_0^T(4\pi t)^{-\beta}\,dt=\frac{(4\pi)^{-\beta}}{1-\beta}T^{1-\beta} H ( T ) ≤ ∫ 0 T ( 4 π t ) − β d t = 1 − β ( 4 π ) − β T 1 − β . Choosing T = ( c p ( 1 − β ) ( 4 π ) β ) 1 / ( 1 − β ) ( ε Λ ) − ( p − 1 ) / ( 1 − β ) T=(c_p(1-\beta)(4\pi)^\beta)^{1/(1-\beta)}(\varepsilon\Lambda)^{-(p-1)/(1-\beta)} T = ( c p ( 1 − β ) ( 4 π ) β ) 1/ ( 1 − β ) ( ε Λ ) − ( p − 1 ) / ( 1 − β ) we obtain T ε ≥ c ε − ( p − 1 ) / ( 1 − β ) T_\varepsilon\ge c\varepsilon^{-(p-1)/(1-\beta)} T ε ≥ c ε − ( p − 1 ) / ( 1 − β ) , and p − 1 1 − β = 2 ( p − 1 ) 2 − N ( p − 1 ) \frac{p-1}{1-\beta}=\frac{2(p-1)}{2-N(p-1)} 1 − β p − 1 = 2 − N ( p − 1 ) 2 ( p − 1 ) .
If p = p F p=p_F p = p F , then β = 1 \beta=1 β = 1 and H ( T ) = 1 4 π ( 1 + log ( 4 π T ) ) H(T)=\frac{1}{4\pi}(1+\log(4\pi T)) H ( T ) = 4 π 1 ( 1 + log ( 4 π T )) for T ≥ 1 4 π T\ge\frac{1}{4\pi} T ≥ 4 π 1 . Choosing
T = 1 4 π exp ( 4 π c p ( ε Λ ) − ( p − 1 ) − 1 ) , T=\frac{1}{4\pi}\exp\left(4\pi c_p(\varepsilon\Lambda)^{-(p-1)}-1\right), T = 4 π 1 exp ( 4 π c p ( ε Λ ) − ( p − 1 ) − 1 ) ,
which is admissible as soon as 4 π c p ( ε Λ ) − ( p − 1 ) ≥ 1 4\pi c_p(\varepsilon\Lambda)^{-(p-1)}\ge1 4 π c p ( ε Λ ) − ( p − 1 ) ≥ 1 , we obtain
T ε ≥ 1 4 π exp ( 4 π c p Λ 1 − p ε − ( p − 1 ) − 1 ) ≥ exp ( 2 π c p Λ 1 − p ε − ( p − 1 ) ) T_\varepsilon\ge\frac{1}{4\pi}\exp\left(4\pi c_p\Lambda^{1-p}\varepsilon^{-(p-1)}-1\right)\ge\exp\left(2\pi c_p\Lambda^{1-p}\varepsilon^{-(p-1)}\right) T ε ≥ 4 π 1 exp ( 4 π c p Λ 1 − p ε − ( p − 1 ) − 1 ) ≥ exp ( 2 π c p Λ 1 − p ε − ( p − 1 ) )
for ε ≤ ε 0 \varepsilon\le\varepsilon_0 ε ≤ ε 0 small enough. This proves Theorem 1.7 (ii). □
Remark 6.1. The upper bounds use only ζ ∈ L 1 ( 0 , ∞ ) \zeta\in L^1(0,\infty) ζ ∈ L 1 ( 0 , ∞ ) and a , b ∈ L 1 a,b\in L^1 a , b ∈ L 1 , and hold for the lifespan of weak solutions; the lower bounds hold for arbitrary signs. Taking a = 0 a=0 a = 0 shows that a small forcing alone, with ∫ ζ b > 0 \int\zeta b>0 ∫ ζ b > 0 , produces blow-up with the same lifespan scale as a small initial datum of positive mass. The constants depend on the data through M ∗ M_* M ∗ (upper bounds) and Λ \Lambda Λ (lower bounds); for nonnegative data and ζ ≥ 0 \zeta\ge0 ζ ≥ 0 they are comparable. The lifespan of small data with M ∗ ≤ 0 M_*\le0 M ∗ ≤ 0 is not addressed here (Problem 9.2).
Large-time behavior and asymptotic mass
Lemma 7.1. Let φ ∈ L 1 ( R N ) \varphi\in L^1(\mathbb{R}^N) φ ∈ L 1 ( R N ) and f ∈ L 1 ( Q ) f\in L^1(Q) f ∈ L 1 ( Q ) . Then
lim t → ∞ ∥ S ( t ) φ + ∫ 0 t S ( t − s ) f ( s ) d s − ( ∫ R N φ d x + ∬ Q f d x d s ) G t ∥ 1 = 0. \lim_{t\to\infty}\left\|S(t)\varphi+\int_0^tS(t-s)f(s)\,ds-\left(\int_{\mathbb{R}^N}\varphi\,dx+\iint_Q f\,dx\,ds\right)G_t\right\|_1=0. t → ∞ lim S ( t ) φ + ∫ 0 t S ( t − s ) f ( s ) d s − ( ∫ R N φ d x + ∬ Q f d x d s ) G t 1 = 0.
Proof. The contribution of φ \varphi φ is handled by Lemma 2.1 (d). For the Duhamel term we write
∫ 0 t S ( t − s ) f ( s ) d s − ( ∬ Q f ) G t = ∫ 0 t [ S ( t − s ) f ( s ) − ( ∫ f ( s ) ) G t ] d s − ( ∫ t ∞ ∬ R N f d x d s ) G t . \int_0^tS(t-s)f(s)\,ds-\left(\iint_Qf\right)G_t=\int_0^t\left[S(t-s)f(s)-\left(\int f(s)\right)G_t\right]\,ds-\left(\int_t^\infty\iint_{\mathbb{R}^N}f\,dx\,ds\right)G_t. ∫ 0 t S ( t − s ) f ( s ) d s − ( ∬ Q f ) G t = ∫ 0 t [ S ( t − s ) f ( s ) − ( ∫ f ( s ) ) G t ] d s − ( ∫ t ∞ ∬ R N f d x d s ) G t .
The L 1 L^1 L 1 norm of the last term is at most ∫ t ∞ ∥ f ( s ) ∥ 1 d s → 0 \int_t^\infty\|f(s)\|_1\,ds\to0 ∫ t ∞ ∥ f ( s ) ∥ 1 d s → 0 . For the first term, set δ ( t , s ) : = 1 { s < t } ∥ S ( t − s ) f ( s ) − ( ∫ f ( s ) ) G t ∥ 1 \delta(t,s):=\mathbf{1}_{\{s<t\}}\left\|S(t-s)f(s)-\left(\int f(s)\right)G_t\right\|_1 δ ( t , s ) := 1 { s < t } S ( t − s ) f ( s ) − ( ∫ f ( s ) ) G t 1 . Then δ ( t , s ) ≤ 2 ∥ f ( s ) ∥ 1 \delta(t,s)\le2\|f(s)\|_1 δ ( t , s ) ≤ 2∥ f ( s ) ∥ 1 , which is integrable in s s s , and for a.e. fixed s s s ,
δ ( t , s ) ≤ ∥ S ( t − s ) f ( s ) − ( ∫ f ( s ) ) G t − s ∥ 1 + ∥ f ( s ) ∥ 1 ∥ G t − s − G t ∥ 1 → t → ∞ 0 \delta(t,s)\le\left\|S(t-s)f(s)-\left(\int f(s)\right)G_{t-s}\right\|_1+\|f(s)\|_1\|G_{t-s}-G_t\|_1\xrightarrow[t\to\infty]{}0 δ ( t , s ) ≤ S ( t − s ) f ( s ) − ( ∫ f ( s ) ) G t − s 1 + ∥ f ( s ) ∥ 1 ∥ G t − s − G t ∥ 1 t → ∞ 0
by Lemma 2.1 (d),(e). Dominated convergence gives ∫ 0 ∞ δ ( t , s ) d s → 0 \int_0^\infty\delta(t,s)\,ds\to0 ∫ 0 ∞ δ ( t , s ) d s → 0 . □
Proof of Theorem 1.4
(i) Let u u u be a global mild solution with ∣ u ∣ p ∈ L 1 ( Q ) |u|^p\in L^1(Q) ∣ u ∣ p ∈ L 1 ( Q ) . Then f : = ∣ u ∣ p + ζ w ∈ L 1 ( Q ) f:=|u|^p+\zeta w\in L^1(Q) f := ∣ u ∣ p + ζ w ∈ L 1 ( Q ) , since ∥ ζ ⊗ w ∥ L 1 ( Q ) = ∥ ζ ∥ 1 ∥ w ∥ 1 \|\zeta\otimes w\|_{L^1(Q)}=\|\zeta\|_1\|w\|_1 ∥ ζ ⊗ w ∥ L 1 ( Q ) = ∥ ζ ∥ 1 ∥ w ∥ 1 , and ∬ Q f = ∬ Q ∣ u ∣ p + Z W \iint_Qf=\iint_Q|u|^p+ZW ∬ Q f = ∬ Q ∣ u ∣ p + Z W . Applying Lemma 7.1 to (14) with φ = u 0 \varphi=u_0 φ = u 0 gives ∥ u ( t ) − m ∞ G t ∥ 1 → 0 \|u(t)-m_\infty G_t\|_1\to0 ∥ u ( t ) − m ∞ G t ∥ 1 → 0 with m ∞ = m 0 + Z W + ∬ Q ∣ u ∣ p = M + ∬ Q ∣ u ∣ p m_\infty=m_0+ZW+\iint_Q|u|^p=M+\iint_Q|u|^p m ∞ = m 0 + Z W + ∬ Q ∣ u ∣ p = M + ∬ Q ∣ u ∣ p .
(ii) Let 1 < p ≤ p F 1<p\le p_F 1 < p ≤ p F and let u u u be a global mild solution. By Lemma 2.2 (a) it is a global weak solution, and Theorem 1.1 gives ∬ Q ∣ u ∣ p = − M < ∞ \iint_Q|u|^p=-M<\infty ∬ Q ∣ u ∣ p = − M < ∞ . Part (i) applies with m ∞ = M + ∬ Q ∣ u ∣ p = 0 m_\infty=M+\iint_Q|u|^p=0 m ∞ = M + ∬ Q ∣ u ∣ p = 0 , so ∥ u ( t ) ∥ 1 = ∥ u ( t ) − 0 ⋅ G t ∥ 1 → 0 \|u(t)\|_1=\|u(t)-0\cdot G_t\|_1\to0 ∥ u ( t ) ∥ 1 = ∥ u ( t ) − 0 ⋅ G t ∥ 1 → 0 .
(iii) Let p > p F p>p_F p > p F , let ζ \zeta ζ satisfy ( H a ) (H_a) ( H a ) , and let u u u be a global mild solution with ∣ u ∣ ≤ κ V |u|\le\kappa V ∣ u ∣ ≤ κV . By Lemma 2.7(i), ( H K ) (H_K) ( H K ) holds, so that (40) and (41) give
∥ u ( s ) ∥ ∞ ≤ κ m ( s ) ≤ κ ( [ [ u 0 ] ] + K [ [ w ] ] h N ( s ) ) ≤ C 5 ( 1 + s ) − N / 2 , \|u(s)\|_\infty\le\kappa m(s)\le\kappa\bigl([ [u_0] ]+K[ [w] ]h_N(s)\bigr)\le C_5(1+s)^{-N/2}, ∥ u ( s ) ∥ ∞ ≤ κm ( s ) ≤ κ ( [[ u 0 ]] + K [[ w ]] h N ( s ) ) ≤ C 5 ( 1 + s ) − N /2 ,
∥ u ( s ) ∥ 1 ≤ κ ( ∥ u 0 ∥ 1 + ∥ ζ ∥ 1 ∥ w ∥ 1 ) = : C 6 . \|u(s)\|_1\le\kappa\left(\|u_0\|_1+\|\zeta\|_1\|w\|_1\right)=:C_6. ∥ u ( s ) ∥ 1 ≤ κ ( ∥ u 0 ∥ 1 + ∥ ζ ∥ 1 ∥ w ∥ 1 ) =: C 6 .
Hence ∬ Q ∣ u ∣ p ≤ ∫ 0 ∞ ∥ u ( s ) ∥ ∞ p − 1 ∥ u ( s ) ∥ 1 d s ≤ C 5 p − 1 C 6 ∫ 0 ∞ ( 1 + s ) − N ( p − 1 ) / 2 d s < ∞ \iint_{Q}|u|^{p}\leq\int_{0}^{\infty}\|u(s)\|_{\infty}^{p-1}\|u(s)\|_{1}\,ds\leq C_{5}^{p-1}C_{6}\int_{0}^{\infty}(1+s)^{-N(p-1)/2}\,ds<\infty ∬ Q ∣ u ∣ p ≤ ∫ 0 ∞ ∥ u ( s ) ∥ ∞ p − 1 ∥ u ( s ) ∥ 1 d s ≤ C 5 p − 1 C 6 ∫ 0 ∞ ( 1 + s ) − N ( p − 1 ) /2 d s < ∞ , because N ( p − 1 ) / 2 > 1 N(p-1)/2>1 N ( p − 1 ) /2 > 1 . Part (i) gives the L 1 L^{1} L 1 convergence. For the L ∞ L^{\infty} L ∞ estimate, let t ≥ 2 t\geq2 t ≥ 2 . By Lemma 2.2 (c) with τ = t / 2 \tau=t/2 τ = t /2 and the identity S ( t / 2 ) G t / 2 = G t S(t/2)G_{t/2}=G_{t} S ( t /2 ) G t /2 = G t ,
u ( t ) − m ∞ G t = S ( t / 2 ) [ u ( t / 2 ) − m ∞ G t / 2 ] + ∫ t / 2 t S ( t − s ) ∣ u ( s ) ∣ p d s + ∫ t / 2 t S ( t − s ) ( ζ ( s ) w ) d s u(t)-m_{\infty}G_{t}=S(t/2)\left[u(t/2)-m_{\infty}G_{t/2}\right]+\int_{t/2}^{t}S(t-s)|u(s)|^{p}\,ds+\int_{t/2}^{t}S(t-s)(\zeta(s)w)\,ds u ( t ) − m ∞ G t = S ( t /2 ) [ u ( t /2 ) − m ∞ G t /2 ] + ∫ t /2 t S ( t − s ) ∣ u ( s ) ∣ p d s + ∫ t /2 t S ( t − s ) ( ζ ( s ) w ) d s
By Lemma 2.1 (b) and part (i), the first term has L ∞ L^{\infty} L ∞ norm at most ( 2 π t ) − N / 2 ∥ u ( t / 2 ) − m ∞ G t / 2 ∥ 1 = o ( t − N / 2 ) (2\pi t)^{-N/2}\|u(t/2)-m_{\infty}G_{t/2}\|_{1}=o(t^{-N/2}) ( 2 π t ) − N /2 ∥ u ( t /2 ) − m ∞ G t /2 ∥ 1 = o ( t − N /2 ) . The second term is bounded in L ∞ L^{\infty} L ∞ by
∫ t / 2 t ∥ u ( s ) ∥ ∞ p d s ≤ t 2 C 5 p ( 1 + t / 2 ) − N p / 2 = O ( t 1 − N p / 2 ) = o ( t − N / 2 ) , \int_{t/2}^{t}\|u(s)\|_{\infty}^{p}\,ds\leq\frac{t}{2}C_{5}^{p}(1+t/2)^{-Np/2}=O(t^{1-Np/2})=o(t^{-N/2}), ∫ t /2 t ∥ u ( s ) ∥ ∞ p d s ≤ 2 t C 5 p ( 1 + t /2 ) − N p /2 = O ( t 1 − N p /2 ) = o ( t − N /2 ) ,
again because N ( p − 1 ) / 2 > 1 N(p-1)/2>1 N ( p − 1 ) /2 > 1 . The third term is bounded by ∥ w ∥ ∫ t / 2 t ∣ ζ ( s ) ∣ h N ( t − s ) d s = o ( t − N / 2 ) \|w\|\int_{t/2}^{t}|\zeta(s)|h_{N}(t-s)\,ds=o(t^{-N/2}) ∥ w ∥ ∫ t /2 t ∣ ζ ( s ) ∣ h N ( t − s ) d s = o ( t − N /2 ) by (29) . Hence t N / 2 ∥ u ( t ) − m ∞ G t ∥ ∞ → 0 t^{N/2}\|u(t)-m_{\infty}G_{t}\|_{\infty}\to0 t N /2 ∥ u ( t ) − m ∞ G t ∥ ∞ → 0 , and the case 1 < q < ∞ 1<q<\infty 1 < q < ∞ follows from the inequality ∥ g ∥ q ≤ ∥ g ∥ 1 1 / q ∥ g ∥ ∞ 1 − 1 / q \|g\|_{q}\leq\|g\|_{1}^{1/q}\|g\|_{\infty}^{1-1/q} ∥ g ∥ q ≤ ∥ g ∥ 1 1/ q ∥ g ∥ ∞ 1 − 1/ q :
t N 2 ( 1 − 1 q ) ∥ u ( t ) − m ∞ G t ∥ q ≤ ∥ u ( t ) − m ∞ G t ∥ 1 1 / q ( t N / 2 ∥ u ( t ) − m ∞ G t ∥ ∞ ) 1 − 1 / q ⟶ 0. t^{\frac{N}{2}(1-\frac{1}{q})}\|u(t)-m_{\infty}G_{t}\|_{q}\leq\|u(t)-m_{\infty}G_{t}\|_{1}^{1/q}\left(t^{N/2}\|u(t)-m_{\infty}G_{t}\|_{\infty}\right)^{1-1/q}\longrightarrow0. t 2 N ( 1 − q 1 ) ∥ u ( t ) − m ∞ G t ∥ q ≤ ∥ u ( t ) − m ∞ G t ∥ 1 1/ q ( t N /2 ∥ u ( t ) − m ∞ G t ∥ ∞ ) 1 − 1/ q ⟶ 0.
Finally m ∞ − M = ∬ Q ∣ u ∣ p ≥ 0 m_{\infty}-M=\iint_{Q}|u|^{p}\geq0 m ∞ − M = ∬ Q ∣ u ∣ p ≥ 0 .
Corollary 7.1. Let ζ \zeta ζ satisfy ( H a ) (\mathrm{H}_{a}) ( H a ) and let w ∈ L 1 ( R N ) ∩ L ∞ ( R N ) w\in L^{1}(\mathbb{R}^{N})\cap L^{\infty}(\mathbb{R}^{N}) w ∈ L 1 ( R N ) ∩ L ∞ ( R N ) . Consider (1) with forcing μ ζ ( t ) w ( x ) \mu\zeta(t)w(x) μ ζ ( t ) w ( x ) , μ ∈ R \mu\in\mathbb{R} μ ∈ R .
(i) If 1 < p ≤ p F 1<p\leq p_{F} 1 < p ≤ p F , then for every μ ∈ R \mu\in\mathbb{R} μ ∈ R and every u 0 ∈ L 1 ∩ L ∞ u_{0}\in L^{1}\cap L^{\infty} u 0 ∈ L 1 ∩ L ∞ , every global mild solution satisfies ∥ u ( t ) ∥ 1 → 0 \|u(t)\|_{1}\to0 ∥ u ( t ) ∥ 1 → 0 as t → ∞ t\to\infty t → ∞ .
(ii) If p > p F p>p_{F} p > p F , there exist μ ≠ 0 \mu\neq0 μ = 0 and u 0 ∈ L 1 ∩ L ∞ u_{0}\in L^{1}\cap L^{\infty} u 0 ∈ L 1 ∩ L ∞ for which (1) has a global mild solution with lim inf t → ∞ ∥ u ( t ) ∥ 1 > 0 \liminf_{t\to\infty}\|u(t)\|_{1}>0 lim inf t → ∞ ∥ u ( t ) ∥ 1 > 0 .
Thus p ≤ p F p\leq p_{F} p ≤ p F if and only if all global solutions lose their mass.
Proof. Part (i) is Theorem 1.4 (ii). For (ii), fix a ∈ L 1 ∩ L ∞ a\in L^{1}\cap L^{\infty} a ∈ L 1 ∩ L ∞ with a ≥ 0 a\geq0 a ≥ 0 and ∫ a > ∣ Z W ∣ \int a>|ZW| ∫ a > ∣ Z W ∣ , and take u 0 = ε a u_{0}=\varepsilon a u 0 = ε a , μ = ε \mu=\varepsilon μ = ε . Then M = ε ( ∫ a + Z W ) > 0 M=\varepsilon(\int a+ZW)>0 M = ε ( ∫ a + Z W ) > 0 . For ε \varepsilon ε small, Corollary 1.1(b) provides a global mild solution with ∣ u ∣ ≤ κ V |u|\leq\kappa V ∣ u ∣ ≤ κV , and by Theorem 1.4 (iii) and Lemma 2.1 (c), ∥ u ( t ) ∥ 1 ≥ ∫ ∣ u ( t ) ∣ → m ∞ ≥ M > 0 \|u(t)\|_{1}\geq\int|u(t)|\to m_{\infty}\geq M>0 ∥ u ( t ) ∥ 1 ≥ ∫ ∣ u ( t ) ∣ → m ∞ ≥ M > 0 .
Remark 7.1. For 1 < p ≤ p F 1<p\leq p_{F} 1 < p ≤ p F , the vanishing of ∥ u ( t ) ∥ 1 \|u(t)\|_{1} ∥ u ( t ) ∥ 1 holds for sign-changing solutions and in the presence of forcing, and it does not rely on any comparison argument: it is a consequence of the identity (7) and of the linear asymptotics of Lemma 7.1. For p > p F p>p_{F} p > p F the asymptotic mass m ∞ = M + ∬ Q ∣ u ∣ p m_{\infty}=M+\iint_{Q}|u|^{p} m ∞ = M + ∬ Q ∣ u ∣ p exceeds the effective mass by the total production of the source. For the small solutions of Corollary 1.1, ∬ Q ∣ u ∣ p ≤ κ p ( ∫ 0 ∞ m p − 1 ) sup t ∥ V ( t ) ∥ 1 \iint_{Q}|u|^{p}\leq\kappa^{p}(\int_{0}^{\infty}m^{p-1})\sup_{t}\|V(t)\|_{1} ∬ Q ∣ u ∣ p ≤ κ p ( ∫ 0 ∞ m p − 1 ) sup t ∥ V ( t ) ∥ 1 by (42) , so that m ∞ = M + O ( ε p ) m_{\infty}=M+O(\varepsilon^{p}) m ∞ = M + O ( ε p ) when the data are of size ε \varepsilon ε .
The model example ζ ( t ) = t σ e − t \zeta(t)=t^{\sigma}e^{-t} ζ ( t ) = t σ e − t
We now collect the results for the model forcing (2) , for which Z = Γ ( σ + 1 ) Z=\Gamma(\sigma+1) Z = Γ ( σ + 1 ) and ( H a ) (\mathrm{H}_{a}) ( H a ) holds with every a > 0 a>0 a > 0 .
Theorem 8.1. Let ζ ( t ) = t σ e − t \zeta(t)=t^{\sigma}e^{-t} ζ ( t ) = t σ e − t with σ > − 1 \sigma>-1 σ > − 1 , let u 0 , w ∈ L 1 ( R N ) ∩ L ∞ ( R N ) u_{0},w\in L^{1}(\mathbb{R}^{N})\cap L^{\infty}(\mathbb{R}^{N}) u 0 , w ∈ L 1 ( R N ) ∩ L ∞ ( R N ) , and let M = m 0 + Γ ( σ + 1 ) W M=m_{0}+\Gamma(\sigma+1)W M = m 0 + Γ ( σ + 1 ) W .
(i) (Fujita range.) If 1 < p ≤ p F 1<p\leq p_{F} 1 < p ≤ p F and (1) has a global weak solution, then either M < 0 M<0 M < 0 , or u 0 ≡ 0 u_{0}\equiv0 u 0 ≡ 0 and w ≡ 0 w\equiv0 w ≡ 0 . Every global weak solution satisfies ∬ Q ∣ u ∣ p = − M \iint_{Q}|u|^{p}=-M ∬ Q ∣ u ∣ p = − M , and every global mild solution satisfies ∥ u ( t ) ∥ 1 → 0 \|u(t)\|_{1}\to0 ∥ u ( t ) ∥ 1 → 0 . When M < 0 M<0 M < 0 , global solutions and blowing-up solutions both occur.
(ii) (Above the Fujita exponent.) If p > p F p > p_F p > p F , there exist ε > 0 \varepsilon> 0 ε > 0 and C ∗ > 0 C_* > 0 C ∗ > 0 , depending only on N N N , p p p and σ \sigma σ , such that if ⟦ u 0 ⟧ + ⟦ w ⟧ ≤ ε \llbracket u_0\rrbracket+ \llbracket w\rrbracket\le\varepsilon [ [ u 0 ] ] + [ [ w ] ] ≤ ε , then {(1.1)} possesses a unique global mild solution. It satisfies, for all t > 0 t > 0 t > 0 ,
∥ u ( t ) ∥ L 1 ≤ C ∗ ε , ∥ u ( t ) ∥ L ∞ ≤ C ∗ ε ( 1 + t ) − N / 2 , \|u(t)\|_{L^1} \le C_*\varepsilon,\qquad\|u(t)\|_{L^\infty} \le C_*\varepsilon(1+t)^{-N/2}, ∥ u ( t ) ∥ L 1 ≤ C ∗ ε , ∥ u ( t ) ∥ L ∞ ≤ C ∗ ε ( 1 + t ) − N /2 ,
and, with m ∞ : = M + ∫ 0 ∞ ∣ u ∣ p m_\infty:= M + \int_0^\infty|u|^p m ∞ := M + ∫ 0 ∞ ∣ u ∣ p , one has 0 ≤ m ∞ − M ≤ C ∗ ε p 0 \le m_\infty- M \le C_*\varepsilon^p 0 ≤ m ∞ − M ≤ C ∗ ε p and, for every q ∈ [ 1 , ∞ ] q \in[1,\infty] q ∈ [ 1 , ∞ ] ,
t N 2 ( 1 − 1 q ) ∥ u ( t ) − m ∞ G t ∥ q → 0. t^{\frac{N}{2}(1-\frac{1}{q})}\|u(t)-m_\infty G_t\|_q \to0. t 2 N ( 1 − q 1 ) ∥ u ( t ) − m ∞ G t ∥ q → 0.
No sign restriction on u 0 u_0 u 0 or w w w is required.
(iii) (Lifespan.) If 1 < p ≤ p F 1 < p \le p_F 1 < p ≤ p F , u 0 = ε a u_0 = \varepsilon a u 0 = ε a , w = ε b w = \varepsilon b w = ε b and ∫ a + Γ ( σ + 1 ) ∫ b > 0 \int a+\Gamma(\sigma+1)\int b>0 ∫ a + Γ ( σ + 1 ) ∫ b > 0 , then for ε \varepsilon ε small T ε ≃ ε − 2 ( p − 1 ) / ( 2 − N ( p − 1 ) ) T_\varepsilon\simeq\varepsilon^{-2(p-1)/(2-N(p-1))} T ε ≃ ε − 2 ( p − 1 ) / ( 2 − N ( p − 1 )) if p < p F p < p_F p < p F , and log T ε ≃ ε − ( p − 1 ) \log T_\varepsilon\simeq\varepsilon^{-(p-1)} log T ε ≃ ε − ( p − 1 ) if p = p F p=p_F p = p F .
(iv) (Large data.) For every p > 1 p > 1 p > 1 , data u 0 = λ g u_0 = \lambda g u 0 = λ g with ∣ { g > 0 } ∣ > 0 |\{g>0\}|>0 ∣ { g > 0 } ∣ > 0 give no global weak solution when λ \lambda λ is large, whatever w w w .
Proof. Since ζ ∈ L 1 ( 0 , ∞ ) \zeta\in L^1(0,\infty) ζ ∈ L 1 ( 0 , ∞ ) and ∣ ζ ( t ) ∣ ≤ C a t − a |\zeta(t)| \le C_at^{-a} ∣ ζ ( t ) ∣ ≤ C a t − a for t ≥ 1 t \ge1 t ≥ 1 and every a > 0 a > 0 a > 0 , ( H a ) (H_a) ( H a ) holds, and by Lemma 2.7(i) so does ( H K ) (H_K) ( H K ) with a constant K = K ( N , σ ) K=K(N,\sigma) K = K ( N , σ ) . Part (i) follows from Theorems 1.2 and 1.4 (ii) and Example 4.1. In (ii), ω ζ ≤ K p − 1 H N , p \omega_\zeta\le K^{p-1}H_{N,p} ω ζ ≤ K p − 1 H N , p depends only on N , p , σ N,p,\sigma N , p , σ , and ⟦ u 0 ⟧ p − 1 + ω ζ ⟦ w ⟧ p − 1 ≤ ( 1 + ω ζ ) ε p − 1 ≤ δ \llbracket u_0\rrbracket^{p-1}+\omega_\zeta\llbracket w\rrbracket^{p-1}\le(1+\omega_\zeta)\varepsilon^{p-1}\le\delta [ [ u 0 ] ] p − 1 + ω ζ [ [ w ] ] p − 1 ≤ ( 1 + ω ζ ) ε p − 1 ≤ δ for ε \varepsilon ε small; Corollary 1.1(b) gives the solution and the bounds, with ∥ u ( t ) ∥ 1 ≤ κ ( ∥ u 0 ∥ 1 + Γ ( σ + 1 ) ∥ w ∥ 1 ) \|u(t)\|_1\le\kappa(\|u_0\|_1+\Gamma(\sigma+1)\|w\|_1) ∥ u ( t ) ∥ 1 ≤ κ ( ∥ u 0 ∥ 1 + Γ ( σ + 1 ) ∥ w ∥ 1 ) . The bound on m ∞ − M m_\infty-M m ∞ − M follows from the last remark of § 7 , since ∫ 0 ∞ m p − 1 ≤ C ε p − 1 \int_0^\infty m^{p-1}\le C\varepsilon^{p-1} ∫ 0 ∞ m p − 1 ≤ C ε p − 1 and sup t ∥ V ( t ) ∥ 1 ≤ C ε \sup_t\|V(t)\|_1\le C\varepsilon sup t ∥ V ( t ) ∥ 1 ≤ C ε , and the asymptotics from Theorem 1.4 (iii). Parts (iii) and (iv) are Theorem 1.7 and Corollary 4.1(a). □
Remark 8.1. For growing forcings t σ w t^\sigma w t σ w , σ > 0 \sigma>0 σ > 0 , there is no global solution for any p > 1 p>1 p > 1 when W > 0 W>0 W > 0 [20 ] , and one might expect a similar conclusion for profiles such as t σ e − t t^\sigma e^{-t} t σ e − t which are large on bounded time intervals. This is not the case. The exponential factor makes ζ ∈ L 1 ( 0 , ∞ ) \zeta\in L^1(0,\infty) ζ ∈ L 1 ( 0 , ∞ ) , so the forcing delivers the finite total impulse Γ ( σ + 1 ) W \Gamma(\sigma+1)W Γ ( σ + 1 ) W and the lower bound {(3.3)} saturates at the constant M M M instead of growing with T T T . In the test-function balance this O ( 1 ) O(1) O ( 1 ) gain can defeat the diffusion cost R N + 2 − 2 p ′ R^{N+2-2p'} R N + 2 − 2 p ′ only when N + 2 − 2 p ′ ≤ 0 N+2-2p'\le0 N + 2 − 2 p ′ ≤ 0 , i.e. for p ≤ p F p\le p_F p ≤ p F . Theorem 8.1(ii) shows that the restriction is sharp: for p > p F p>p_F p > p F small data yield global solutions, so nonexistence cannot hold for every p > 1 p>1 p > 1 . The same conclusion applies to every forcing profile ζ \zeta ζ that is integrable and decays faster than any power, since such a ζ \zeta ζ satisfies ( H a ) (H_a) ( H a ) for every a a a ; the blow-up range for such forcings is therefore 1 < p ≤ p F 1<p\le p_F 1 < p ≤ p F (under the effective-mass condition), not p > 1 p>1 p > 1 . What remains true for every p > 1 p>1 p > 1 is blow-up for large data or forcing (Corollary 4.1).
Concluding remarks and open problems
The results of Sections 3–8 are proved in full. In this section we collect questions that we were not able to settle; everything stated here is conjectural or open and is clearly separated from the results above.
Remark 9.1. Theorem 1.5 and Corollary 1.1 are small-data statements. For p > p F p>p_F p > p F large data blow up (Corollary 4.1(a),(b)); because the nonlinearity is ∣ u ∣ p |u|^p ∣ u ∣ p , the source is nonnegative and only the positive part of the data drives blow-up, while negative data are damped (Theorem 1.3 ). In the preliminary version of this work it was asked whether the hypothesis {(1.7)} in Theorem 1.2 could be removed altogether for 1 < p ≤ p F 1<p\le p_F 1 < p ≤ p F , that is, whether a very negative effective mass still forces blow-up through the ∣ u ∣ p |u|^p ∣ u ∣ p source. Theorem 1.3 answers this in the negative, and Theorem 1.2 (ii) shows that the condition can be relaxed exactly to M ≥ 0 M\ge0 M ≥ 0 .
Open problem 9.1. Let 1 < p ≤ p F 1<p\le p_F 1 < p ≤ p F . Describe the set of data ( u 0 , ζ w ) (u_0,\zeta w) ( u 0 , ζ w ) with M < 0 M<0 M < 0 for which {(1.1)} has a global solution. In particular, is the nonpositive global solution of Theorem 1.3 stable under small perturbations of the data that keep M < 0 M<0 M < 0 , for instance in the norm ⟦ ⋅ ⟧ \llbracket\cdot\rrbracket [ [ ⋅ ] ] ? By {(1.6)} any such global solution produces exactly ∣ M ∣ |M| ∣ M ∣ units of mass through the source, which suggests that the answer depends on how the positive part of the perturbation is distributed relative to the negative mass.
Open problem 9.2. Let 1 < p ≤ p F 1 < p \le p_F 1 < p ≤ p F and u 0 = ε a u_0 = \varepsilon a u 0 = ε a , w = ε b w = \varepsilon b w = ε b with M ∗ = ∫ a + ∫ b ≤ 0 M_* = \int a + \int b \le0 M ∗ = ∫ a + ∫ b ≤ 0 and ( a , ζ b ) (a,\zeta b) ( a , ζ b ) not of the sign considered in Theorem 1.3 . Theorem 1.7 (ii) gives the lower bounds in all cases. When M ∗ = 0 M_* = 0 M ∗ = 0 and a a a has a nonzero first moment, one expects the lifespan to be governed by a higher-order term of the asymptotic expansion of the linear flow, as for the unforced equation with zero-mass data. Determine the sharp lifespan in this case.
Open problem 9.3. Let N ≥ 3 N \ge3 N ≥ 3 and p F < p < 2 p_F < p < 2 p F < p < 2 . Does small-data global existence hold for every ζ ∈ L 1 ( 0 , ∞ ) \zeta\in L^1(0,\infty) ζ ∈ L 1 ( 0 , ∞ ) ? Remark 5.2 explains why the domination principle requires some decay of ζ \zeta ζ . A heuristic computation of ∫ Q F p \int_Q F^p ∫ Q F p , where F F F is the linear forced flow, suggests that the answer is positive, but a proof would require a finer description of the near-field part of F F F than the L ∞ L^\infty L ∞ bound (5.2).
Open problem 9.4. For ζ \zeta ζ regularly varying at infinity with index − 1 -1 − 1 , so that Z ( T ) = ∫ 0 T ζ Z(T) = \int_0^T \zeta Z ( T ) = ∫ 0 T ζ is slowly varying in the sense of Karamata, Corollary 4.2 and Remark 4.1 show that the critical exponent is p F p_F p F (at least for N ≤ 2 N \le2 N ≤ 2 ) and that the Fujita range no longer depends on the data when Z ( T ) → ∞ Z(T) \to\infty Z ( T ) → ∞ and W > 0 W > 0 W > 0 . For profiles of index σ ∈ ( − 1 , 0 ) \sigma\in(-1,0) σ ∈ ( − 1 , 0 ) with slowly varying corrections, determine the behavior at the threshold p = p ∗ ( σ ) p = p^*(\sigma) p = p ∗ ( σ ) , where T 1 p − 1 − N 2 Z ( T ) T^{\frac{1}{p-1}-\frac{N}{2}} Z(T) T p − 1 1 − 2 N Z ( T ) is itself slowly varying; one expects blow-up if and only if this quantity is unbounded, in the spirit of [25 ] .
Open problem 9.5. For 1 < p ≤ p F 1 < p \le p_F 1 < p ≤ p F every global mild solution satisfies ∥ u ( t ) ∥ 1 → 0 \|u(t)\|_1 \to0 ∥ u ( t ) ∥ 1 → 0 . For the absorption equation with nonnegative data the rate is known: ∥ v ( t ) ∥ 1 ≃ t − ( 1 p − 1 − N 2 ) \|v(t)\|_1 \simeq t^{-(\frac{1}{p-1}-\frac{N}{2})} ∥ v ( t ) ∥ 1 ≃ t − ( p − 1 1 − 2 N ) for p < p F p < p_F p < p F [7 , 21 ] and ∥ v ( t ) ∥ 1 ≃ ( log t ) − N / 2 \|v(t)\|_1 \simeq(\log t)^{-N/2} ∥ v ( t ) ∥ 1 ≃ ( log t ) − N /2 for p = p F p = p_F p = p F [15 ] . Do sign-changing global solutions of (1.1) with integrable forcing decay at the same rates, possibly under a decay assumption on ζ \zeta ζ ?
Open problem 9.6. The proof of Theorem 1.1 uses the operator only through the scaling of the cutoff cost (2.15). We expect the identity (1.6), with p F p_F p F replaced by the corresponding critical exponent, to hold for the fractional Laplacian (with p F = 1 + 2 s / N p_F = 1 + 2s/N p F = 1 + 2 s / N ), for mixed local–nonlocal operators [23 , 5 ] , for time-fractional diffusions [6 , 13 ] , and for diffusions with interfaces [30 ] . For nonlocal operators the cutoffs must have polynomial tails, and the proof of the vanishing of the diffusive term (3.6) has to be adapted accordingly.
Declarations
Funding. This research received no external funding.
Conflict of interest. The authors declare that they have no conflicts of interest.
Data availability. No data were generated or analyzed in this study.
(Ahmad Z. Fino) Department of Mathematics, Faculty of Sciences, Lebanese University, P.O. Box 1352 Tripoli, Lebanon.
Email address: afino@ul.edu.lb
(M. Majdoub) Department of Mathematics, College of Science, Imam Abdulrahman Bin Faisal University, P. O. Box 1982, Dammam, Saudi Arabia.
(M. Majdoub) Basic and Applied Scientific Research Center, Imam Abdulrahman Bin Faisal University, P.O. Box 1982, 31441, Dammam, Saudi Arabia.
Email address: mmajdoub@iau.edu.sa
Email address: med.majdoub@gmail.com
Email address: mohamed.majdoub@fst.rnu.tn