Introduction and Main Result
Let E ⊂ R N E \subset\mathbb{R}^{N} E ⊂ R N be open, T > 0 T > 0 T > 0 , q > 0 q > 0 q > 0 , and s ∈ ( 0 , 1 ) s \in(0,1) s ∈ ( 0 , 1 ) . We consider locally bounded, local weak solutions to
∂ t u q + L u = 0 in E T : = E × ( 0 , T ) , (1) \partial_t u^q + \mathcal{L}u = 0 \qquad\text{in } E_T := E \times(0,T), \tag*{(1)} ∂ t u q + L u = 0 in E T := E × ( 0 , T ) , (1)
where u q = ∣ u ∣ q − 1 u u^q = |u|^{q-1}u u q = ∣ u ∣ q − 1 u for u ≠ 0 u \ne0 u = 0 and 0 q = 0 0^q = 0 0 q = 0 . The nonlocal operator is formally given by
L u ( x , t ) = P.V. ∫ R N K ( x , y , t ) ( u ( x , t ) − u ( y , t ) ) d y . \mathcal{L}u(x,t) = \operatorname{P.V.} \int_{\mathbb{R}^{N}} K(x,y,t)(u(x,t)-u(y,t))\,dy. L u ( x , t ) = P.V. ∫ R N K ( x , y , t ) ( u ( x , t ) − u ( y , t )) d y .
The kernel is measurable, symmetric, and satisfies
C 0 ∣ x − y ∣ N + 2 s ≤ K ( x , y , t ) = K ( y , x , t ) ≤ C 1 ∣ x − y ∣ N + 2 s (2) \frac{C_0}{|x-y|^{N+2s}} \le K(x,y,t) = K(y,x,t) \le\frac{C_1}{|x-y|^{N+2s}} \tag*{(2)} ∣ x − y ∣ N + 2 s C 0 ≤ K ( x , y , t ) = K ( y , x , t ) ≤ ∣ x − y ∣ N + 2 s C 1 (2)
for almost every ( x , y , t ) (x,y,t) ( x , y , t ) , where C 0 , C 1 > 0 C_0,C_1 > 0 C 0 , C 1 > 0 . The parameters { N , s , q , C 0 , C 1 } \{N,s,q,C_0,C_1\} { N , s , q , C 0 , C 1 } are referred to as the data.
The purpose of this paper is to give a quantitative local continuity estimate when the nonlocal tail is merely integrable in time. The dependence on its integrals over shrinking time intervals is made explicit in the estimate below.
Theorem 1.1 (Local modulus of continuity). Let u u u be a locally bounded, local weak solution to (1) –(2) in E T E_T E T , in the sense of Definition 1.1. There exist C > 1 C > 1 C > 1 and β ∈ ( 0 , s ) \beta\in(0,s) β ∈ ( 0 , s ) , depending only on the data, such that the following holds. Let Q = K R ( x 0 ) × ( T 1 , T 2 ) ⊂ E T Q = K_R(x_0) \times(T_1,T_2) \subset E_T Q = K R ( x 0 ) × ( T 1 , T 2 ) ⊂ E T , choose M M M so that
2 M ≥ ess sup Q ∣ u ∣ + C Tail ( u ; Q ) 1 / q , (3) 2M \ge\underset{Q}{\operatorname{ess\,sup}}\,|u| + C\operatorname{Tail}(u;Q)^{1/q}, \tag*{(3)} 2 M ≥ Q ess sup ∣ u ∣ + C Tail ( u ; Q ) 1/ q , (3)
2020 Mathematics Subject Classification. 47G20, 35R11, 35K65, 35B65.
Key words and phrases. Nonlocal diffusion, fractional porous medium, continuity, intrinsic scaling. and suppose ( x o , t o ) + Q 2 ϱ ( θ ) ⊂ Q (x_o,t_o)+Q_{2\varrho}(\theta)\subset Q ( x o , t o ) + Q 2 ϱ ( θ ) ⊂ Q with θ = M q − 1 \theta=M^{q-1} θ = M q − 1 . Then, for every 0 < r < ϱ 0<r<\varrho 0 < r < ϱ ,
ess osc ( x o , t o ) + Q r ( θ ) u ≤ C M ( r ϱ ) β / 2 + C M [ 1 M q ∫ t o − θ ϱ 2 s ( r / ϱ ) β t o ∫ R N ∖ K R ( x o ) ∣ u ( y , t ) ∣ ∣ y − x o ∣ N + 2 s d y d t ] min { 1 , 1 / q } . (4) \underset{(x_o,t_o)+Q_r(\theta)}{\operatorname{ess\,osc}} u
\leq CM\left(\frac{r}{\varrho}\right)^{\beta/2}
+CM\left[\frac{1}{M^q}\int_{t_o-\theta\varrho^{2s}(r/\varrho)^\beta}^{t_o}\int_{\mathbb{R}^N\setminus K_R(x_o)}
\frac{\lvert u(y,t)\rvert}{\lvert y-x_o\rvert^{N+2s}}\,dy\,dt\right]^{\min\{1,1/q\}}. \tag*{(4)} ( x o , t o ) + Q r ( θ ) ess osc u ≤ C M ( ϱ r ) β /2 + C M [ M q 1 ∫ t o − θ ϱ 2 s ( r / ϱ ) β t o ∫ R N ∖ K R ( x o ) ∣ y − x o ∣ N + 2 s ∣ u ( y , t )∣ d y d t ] m i n { 1 , 1/ q } . (4)
Remark 1.1. The nonlocal porous medium or fast diffusion equation is often written as ∂ t v + L v m = 0 \partial_t v+\mathcal{L}v^m=0 ∂ t v + L v m = 0 . Thus 0 < q < 1 0<q<1 0 < q < 1 corresponds to porous medium case, whereas q > 1 q>1 q > 1 corresponds to fast diffusion. We allow solutions of either sign and require the equation to hold only locally in E T E_T E T .
Remark 1.2. Our result concerns from–L l o c ∞ L^\infty_{\mathrm{loc}} L loc ∞ –to–C l o c 0 C^0_{\mathrm{loc}} C loc 0 –type regularity. Whether solutions are locally bounded is a separate issue; see recent work [12 ] in this connection.
Novelty and Significance
Nonlocal nonlinear diffusion equations arise, among other settings, in the boundary heat control problems formulated in the book [15 ] of Duvaut and Lions. The continuity theory of their solutions was initiated by Athanasopoulos and Caffarelli [1 ] . Over the past 15 years, they have been subject to intensive studies in various setups; see [3 , 4 , 5 , 8 , 11 ] for instance. In particular, the regularity theory of these equations was further developed in [2 , 7 , 9 , 10 , 16 ] ; they proved continuity of globally bounded weak solutions. More recently, by combining techniques from [20 , 21 ] , Hölder regularity of locally bounded solutions with the L ∞ L^\infty L ∞ -in-time tail is proven in [18 ] , whereas Q. Li [19 ] established it under the weaker L 1 + ε L^{1+\varepsilon} L 1 + ε -in-time tail.
The present work establishes local continuity at the natural L 1 L^1 L 1 -in-time tail endpoint. More specifically, we consider locally bounded, possibly sign-changing weak solutions, require the equation to hold locally, and allow measurable symmetric kernels depending on both space and time. The result applies to every q > 0 q>0 q > 0 , covering both porous medium and fast diffusion cases. No global boundedness or additional time integrability of the tail is imposed.
The general modulus of continuity (4) is a natural continuation of the one derived in [22 ] for the parabolic fractional p p p -Laplacian. It separates a power decay arising from the local oscillation argument from the contribution of the exterior values over a shrinking time interval. The latter vanishes by absolute continuity of the tail integral, without any prescribed rate. Thus the estimate describes how interior continuity depends on the temporal concentration of the solution outside the reference domain. When the tail belongs to L 1 + ε L^{1+\varepsilon} L 1 + ε in time, the same estimate yields a Hölder modulus. This kind of Hölder regularity generalizes the linear result in [17 Theorem 1.5] ; see also [6 , 22 ] for the fractional parabolic p p p -Laplacian.
A central issue is the interaction between the control of the exterior values and the degeneracy or singularity of the time nonlinearity at zero. The intrinsic time scale depends on the amplitude of the solution, and hence changes during the oscillation reduction. The tail must therefore be controlled on time intervals whose lengths are determined by the same iteration. Theorem 1.1 shows that these two effects can be treated together, with constants depending only on the data and with the exterior contribution displayed explicitly. We will come back to the technical aspect later.
The continuity conclusion also supports the pointwise theory of these equations. In particular, it provides the continuous representative and the topological information required in the intrinsic Harnack argument of [23 Theorem 7.1] ; see also [23 Remark 1.3] . This connection makes endpoint continuity relevant to the development of intrinsic Harnack estimates under natural tail assumptions.
Method of proof
As in the local theory of porous medium type equations, we distinguish cylinders where the solution approaches or crosses zero from cylinders where it is quantitatively separated from zero; cf. [13 , 14 , 20 ] . The distinction is necessary because adding constants to a solution does not yield another one due to the power nonlinearity in the time derivative. To distinguish these cylinders, the measure concentration of u u u near zero is at stake. If it always has a sizable portion near zero along nested, intrinsic cylinders, then we keep reducing its amplitude. Otherwise, we show that u u u must be away from zero in the next nested cylinder, where then the non-degenerate regularity theory can be applied after normalization.
Apart from this main idea, the treatment of the nonlocal tail in this argument is quite delicate. It enters into positivity estimates in the following pattern: either the tail is large relative to the current amplitude, or the solution satisfies an improved pointwise bound in a smaller cylinder.
With the positivity tools at hand, we then construct nested cylinders and an accompanying sequence of amplitude bounds. The amplitudes are allowed to decrease only as far as permitted by the tail. The exterior integral at each step is split into the region outside the original reference domain and the intervening spatial annuli. On the annuli, bounds obtained at earlier steps produce a geometric sum that can be absorbed into the amplitude reduction. The remaining contribution is the original exterior tail integrated over the current shrinking time interval. Preserving this interval throughout the iteration is essential: replacing it simply by the initial time interval would leave an error that need not tend to zero, cf. (63) .
If the solution becomes separated from zero at some stage, we stop the near-zero iteration and normalize at that scale. The normalized solution lies in a fixed positive range. The endpoint oscillation argument of [22 ] can then be adapted to this normalized equation.
The final step is to express both alternatives in the original intrinsic geometry and obtain an estimate independent of the stopping index. For a prescribed reference radius, we compare the stopping scale with an intermediate iteration scale. If the iteration reaches that intermediate scale, its amplitude estimate controls the oscillation. If it stops earlier, the normalized nondegenerate estimate applies. Tracking the amplitude factors, the annular contributions, and the time interval in the rescaled tail places both estimates on a common cylinder. This quantitative matching yields the explicit modulus in (4) .
Definitions and notation
For x o ∈ R N x_o \in\mathbb{R}^{N} x o ∈ R N , let K ϱ ( x o ) K_{\varrho}(x_o) K ϱ ( x o ) denote the ball of radius ϱ \varrho ϱ centered at x o x_o x o , and write K ϱ = K ϱ ( 0 ) K_{\varrho}=K_{\varrho}(0) K ϱ = K ϱ ( 0 ) . We use the cylinders
( x o , t o ) + Q ϱ ( θ ) = K ϱ ( x o ) × ( t o − θ ϱ 2 s , t o ) , ( x o , t o ) + Q ( R , S ) = K R ( x o ) × ( t o − S , t o ) . (x_o,t_o)+Q_{\varrho}(\theta)=K_{\varrho}(x_o)\times(t_o-\theta\varrho^{2s},t_o),
\\
(x_o,t_o)+Q(R,S)=K_R(x_o)\times(t_o-S,t_o). ( x o , t o ) + Q ϱ ( θ ) = K ϱ ( x o ) × ( t o − θ ϱ 2 s , t o ) , ( x o , t o ) + Q ( R , S ) = K R ( x o ) × ( t o − S , t o ) .
If the context is clear, we omit ( x 0 , t 0 ) (x_0,t_0) ( x 0 , t 0 ) . For Q = K R ( x 0 ) × I Q=K_R(x_0)\times I Q = K R ( x 0 ) × I , set
Tail ( f ; Q ) = ∫ I ∫ R N ∖ K R ( x 0 ) ∣ f ( y , t ) ∣ ∣ y − x 0 ∣ N + 2 s d y d t . \operatorname{Tail}(f;Q)=\int_I\int_{\mathbb{R}^N\setminus K_R(x_0)}\frac{\lvert f(y,t)\rvert}{\lvert y-x_0\rvert^{N+2s}}\,dy\,dt. Tail ( f ; Q ) = ∫ I ∫ R N ∖ K R ( x 0 ) ∣ y − x 0 ∣ N + 2 s ∣ f ( y , t )∣ d y d t .
Unless otherwise specified, C ≥ 1 C\ge1 C ≥ 1 depends only on the data and may change from line to line. The constants c q , C q > 0 c_q,C_q>0 c q , C q > 0 depend only on q q q . For a measurable set A A A , ∣ A ∣ \lvert A\rvert ∣ A ∣ denotes its Lebesgue measure.
For simplicity, with a ∈ R a\in\mathbb{R} a ∈ R and α > 0 \alpha>0 α > 0 , we define
a α = ∣ a ∣ α − 1 a for a ≠ 0 and 0 α = 0. a^\alpha=\lvert a\rvert^{\alpha-1}a\text{ for }a\ne0\text{ and }0^\alpha=0. a α = ∣ a ∣ α − 1 a for a = 0 and 0 α = 0.
We use the following notion of weak solution.
Definition 1.1. A measurable function u : R N × ( 0 , T ) → R u:\mathbb{R}^N\times(0,T)\to\mathbb{R} u : R N × ( 0 , T ) → R is a local weak subsolution to (1) if
u ∈ C l o c ( ( 0 , T ) ; L l o c q + 1 ( E ) ) ∩ L l o c 2 ( ( 0 , T ) ; W l o c s , 2 ( E ) ) , u\in C_{\mathrm{loc}}\left((0,T);L_{\mathrm{loc}}^{q+1}(E)\right)\cap L_{\mathrm{loc}}^2\left((0,T);W_{\mathrm{loc}}^{s,2}(E)\right), u ∈ C loc ( ( 0 , T ) ; L loc q + 1 ( E ) ) ∩ L loc 2 ( ( 0 , T ) ; W loc s , 2 ( E ) ) , if, for every compact interval [ t 1 , t 2 ] ⊂ ( 0 , T ) [t_1,t_2]\subset(0,T) [ t 1 , t 2 ] ⊂ ( 0 , T ) ,
∫ t 1 t 2 ∫ R N ∣ u ( x , t ) ∣ ( 1 + ∣ x ∣ ) N + 2 s d x d t < ∞ , \int_{t_1}^{t_2}\int_{\mathbb{R}^N}\frac{\lvert u(x,t)\rvert}{(1+\lvert x\rvert)^{N+2s}}\,dx\,dt<\infty, ∫ t 1 t 2 ∫ R N ( 1 + ∣ x ∣ ) N + 2 s ∣ u ( x , t )∣ d x d t < ∞ , and if
∫ D u q ϕ d x ∣ t 1 t 2 − ∫ t 1 t 2 ∫ D u q ∂ t ϕ d x d t + 1 2 ∫ t 1 t 2 ∫ R N ∫ R N ( u ( x , t ) − u ( y , t ) ) ( ϕ ( x , t ) − ϕ ( y , t ) ) K ( x , y , t ) d y d x d t ≤ 0 (5) \int_D u^q\phi\,dx\bigg|_{t_1}^{t_2}-\int_{t_1}^{t_2}\int_D u^q\partial_t\phi\,dx\,dt
+\frac{1}{2}\int_{t_1}^{t_2}\int_{\mathbb{R}^N}\int_{\mathbb{R}^N}(u(x,t)-u(y,t))(\phi(x,t)-\phi(y,t))K(x,y,t)\,dy\,dx\,dt\le0 \tag*{(5)} ∫ D u q ϕ d x t 1 t 2 − ∫ t 1 t 2 ∫ D u q ∂ t ϕ d x d t + 2 1 ∫ t 1 t 2 ∫ R N ∫ R N ( u ( x , t ) − u ( y , t )) ( ϕ ( x , t ) − ϕ ( y , t )) K ( x , y , t ) d y d x d t ≤ 0 (5) for nonnegative admissible test functions with compact spatial support in D ⊆ E D\subseteq E D ⊆ E . Here the test functions belong to
L 2 ( t 1 , t 2 ; W 0 s , 2 ( D ) ) ∩ W 1 , q + 1 ( t 1 , t 2 ; L q + 1 ( D ) ) , L^2(t_1,t_2;W_0^{s,2}(D))\cap W^{1,q+1}(t_1,t_2;L^{q+1}(D)), L 2 ( t 1 , t 2 ; W 0 s , 2 ( D )) ∩ W 1 , q + 1 ( t 1 , t 2 ; L q + 1 ( D )) , and are extended by zero outside D D D . A supersolution satisfies the reverse inequality. A weak solution is both a subsolution and a supersolution.
Structure of the Paper
Section 2 collects the preliminary inequalities. The energy estimate with time-dependent truncations is proved in Section 3. Sections 4 and 5 contain the measure estimates used in the intrinsic iteration. The proof of Theorem 1.1 is given in Section 6.
Acknowledgements. The author would like to express his sincere gratitude to Professor Naian Liao, for suggesting the problem and for his generous guidance throughout this work.
Preliminaries
We collect the inequalities used below; see [21 , 24 ] .
Algebraic inequalities
The signed-power estimate below is [24 Lemma 2.9] . We also record the corresponding estimates for the time-energy primitive.
Lemma 2.1. For q > 0 q>0 q > 0 and a , b ∈ R a,b\in\mathbb{R} a , b ∈ R ,
c q ( ∣ a ∣ + ∣ b ∣ ) q − 1 ∣ a − b ∣ ≤ ∣ a q − b q ∣ ≤ C q ( ∣ a ∣ + ∣ b ∣ ) q − 1 ∣ a − b ∣ . (6) c_q(|a|+|b|)^{q-1}|a-b|\leq|a^q-b^q|\leq C_q(|a|+|b|)^{q-1}|a-b|. \tag*{(6)} c q ( ∣ a ∣ + ∣ b ∣ ) q − 1 ∣ a − b ∣ ≤ ∣ a q − b q ∣ ≤ C q ( ∣ a ∣ + ∣ b ∣ ) q − 1 ∣ a − b ∣. (6) At a = b = 0 a=b=0 a = b = 0 , all expressions are assigned the value zero. For
g ± ( a , k ) = ± ∫ k a ∣ r ∣ q − 1 ( r − k ) ± d r , g_{\pm}(a,k)=\pm\int_k^a |r|^{q-1}(r-k)_{\pm}\,\mathrm{d}r, g ± ( a , k ) = ± ∫ k a ∣ r ∣ q − 1 ( r − k ) ± d r , one has g ± ≥ 0 g_{\pm}\geq0 g ± ≥ 0 and
c q ( ∣ a ∣ + ∣ k ∣ ) q − 1 ( a − k ) ± 2 ≤ g ± ( a , k ) ≤ C q ( ∣ a ∣ + ∣ k ∣ ) q − 1 ( a − k ) ± 2 . (7) c_q(|a|+|k|)^{q-1}(a-k)_{\pm}^{2}\leq g_{\pm}(a,k)\leq C_q(|a|+|k|)^{q-1}(a-k)_{\pm}^{2}. \tag*{(7)} c q ( ∣ a ∣ + ∣ k ∣ ) q − 1 ( a − k ) ± 2 ≤ g ± ( a , k ) ≤ C q ( ∣ a ∣ + ∣ k ∣ ) q − 1 ( a − k ) ± 2 . (7) In particular, if ∣ a ∣ ≤ 2 M |a|\leq2M ∣ a ∣ ≤ 2 M and M / 4 ≤ k ≤ M M/4\leq k\leq M M /4 ≤ k ≤ M , then
c q M q − 1 ( a − k ) − 2 ≤ g − ( a , k ) ≤ C q M q − 1 ( a − k ) − 2 , (8) c_qM^{q-1}(a-k)_{-}^{2}\leq g_{-}(a,k)\leq C_qM^{q-1}(a-k)_{-}^{2}, \tag*{(8)} c q M q − 1 ( a − k ) − 2 ≤ g − ( a , k ) ≤ C q M q − 1 ( a − k ) − 2 , (8) ( a q − k q ) − ≥ c q M q − 1 ( a − k ) − . (9) (a^q-k^q)_{-}\geq c_qM^{q-1}(a-k)_{-}. \tag*{(9)} ( a q − k q ) − ≥ c q M q − 1 ( a − k ) − . (9) For M ≤ k ≤ 2 M M\leq k\leq2M M ≤ k ≤ 2 M , the analogous estimates hold for g + ( a , k ) g_{+}(a,k) g + ( a , k ) and ( a q − k q ) + (a^q-k^q)_{+} ( a q − k q ) + . We shall also use
q g − ( z 1 / q , k ) = ∫ z k q ( k − r 1 / q ) + d r qg_{-}(z^{1/q},k)=\int_z^{k^q}(k-r^{1/q})_{+}\,\mathrm{d}r q g − ( z 1/ q , k ) = ∫ z k q ( k − r 1/ q ) + d r and
q ∣ g ± ( z 1 1 / q , k ) − g ± ( z 2 1 / q , k ) ∣ ≤ ( ∣ z 1 ∣ 1 / q + ∣ z 2 ∣ 1 / q + ∣ k ∣ ) ∣ z 1 − z 2 ∣ . (10) q\left|g_{\pm}(z_1^{1/q},k)-g_{\pm}(z_2^{1/q},k)\right|\leq(|z_1|^{1/q}+|z_2|^{1/q}+|k|)|z_1-z_2|. \tag*{(10)} q g ± ( z 1 1/ q , k ) − g ± ( z 2 1/ q , k ) ≤ ( ∣ z 1 ∣ 1/ q + ∣ z 2 ∣ 1/ q + ∣ k ∣ ) ∣ z 1 − z 2 ∣. (10)
Fractional embedding and auxiliary estimates
Write
[ v ] W s , 2 ( K ϱ ) 2 = ∫ K ϱ ∫ K ϱ ∣ v ( x ) − v ( y ) ∣ 2 ∣ x − y ∣ N + 2 s d y d x . [v]_{W^{s,2}(K_\varrho)}^2=\int_{K_\varrho}\int_{K_\varrho}\frac{|v(x)-v(y)|^2}{|x-y|^{N+2s}}\,\mathrm{d}y\,\mathrm{d}x. [ v ] W s , 2 ( K ϱ ) 2 = ∫ K ϱ ∫ K ϱ ∣ x − y ∣ N + 2 s ∣ v ( x ) − v ( y ) ∣ 2 d y d x .
The next statement is the case p = 2 p=2 p = 2 of [21 Proposition A.3] ; see also [24 Lemma 2.2] .
Lemma 2.2. Let I I I be a bounded interval and v ∈ L ∞ ( I ; L 2 ( K ϱ ) ) ∩ L 2 ( I ; W s , 2 ( K ϱ ) ) v\in L^\infty(I;L^2(K_\varrho))\cap L^2(I;W^{s,2}(K_\varrho)) v ∈ L ∞ ( I ; L 2 ( K ϱ )) ∩ L 2 ( I ; W s , 2 ( K ϱ )) , which is compactly supported in K ( 1 − σ ) ϱ K_{(1-\sigma)\varrho} K ( 1 − σ ) ϱ for σ ∈ ( 0 , 1 ) \sigma\in(0,1) σ ∈ ( 0 , 1 ) and for a.e. t ∈ I t\in I t ∈ I . Then
∫ I ∫ K ϱ ∣ v ∣ m ^ d x d t ≤ C ( N , s ) ( ess sup t ∈ I ∫ K ϱ ∣ v ( x , t ) ∣ 2 d x ) 2 s / N × ∫ I ( ϱ 2 s [ v ( ⋅ , t ) ] W s , 2 ( K ϱ ) 2 + σ − ( N + 2 s ) ∫ K ϱ ∣ v ∣ 2 d x ) d t , \int_I\int_{K_\varrho}|v|^{\widehat m}\,\mathrm{d}x\,\mathrm{d}t\leq C(N,s)\left(\operatorname*{ess\,sup}_{t\in I}\int_{K_\varrho}|v(x,t)|^2\,\mathrm{d}x\right)^{2s/N}
\times\int_I\left(\varrho^{2s}[v(\cdot,t)]_{W^{s,2}(K_\varrho)}^2+\sigma^{-(N+2s)}\int_{K_\varrho}|v|^2\,\mathrm{d}x\right)\,\mathrm{d}t, ∫ I ∫ K ϱ ∣ v ∣ m d x d t ≤ C ( N , s ) ( t ∈ I ess sup ∫ K ϱ ∣ v ( x , t ) ∣ 2 d x ) 2 s / N × ∫ I ( ϱ 2 s [ v ( ⋅ , t ) ] W s , 2 ( K ϱ ) 2 + σ − ( N + 2 s ) ∫ K ϱ ∣ v ∣ 2 d x ) d t , where m ^ = 2 ( 1 + 2 s / N ) \widehat m=2(1+2s/N) m = 2 ( 1 + 2 s / N ) .
Now, we collect some useful auxiliary estimates to be used. Let ζ ∈ C 1 ( K ϱ , [ 0 , 1 ] ) \zeta\in C^1(K_\varrho,[0,1]) ζ ∈ C 1 ( K ϱ , [ 0 , 1 ]) , and ∣ D ζ ∣ ≤ L |D\zeta|\leq L ∣ D ζ ∣ ≤ L , then for x ∈ K ϱ x\in K_\varrho x ∈ K ϱ ,
∫ K ϱ ∣ ζ ( x ) − ζ ( y ) ∣ 2 ∣ x − y ∣ N + 2 s d y ≤ C ( N , s ) L 2 ϱ 2 − 2 s . (11) \int_{K_\varrho}\frac{|\zeta(x)-\zeta(y)|^2}{|x-y|^{N+2s}}\,\mathrm{d}y\leq C(N,s)L^2\varrho^{2-2s}. \tag*{(11)} ∫ K ϱ ∣ x − y ∣ N + 2 s ∣ ζ ( x ) − ζ ( y ) ∣ 2 d y ≤ C ( N , s ) L 2 ϱ 2 − 2 s . (11)
Moreover,
∣ v ( x ) ζ ( x ) − v ( y ) ζ ( y ) ∣ 2 ≤ 2 ζ ( x ) 2 ∣ v ( x ) − v ( y ) ∣ 2 + 2 v ( y ) 2 ∣ ζ ( x ) − ζ ( y ) ∣ 2 . (12) |v(x)\zeta(x)-v(y)\zeta(y)|^2 \le2\zeta(x)^2|v(x)-v(y)|^2+2v(y)^2|\zeta(x)-\zeta(y)|^2. \tag*{(12)} ∣ v ( x ) ζ ( x ) − v ( y ) ζ ( y ) ∣ 2 ≤ 2 ζ ( x ) 2 ∣ v ( x ) − v ( y ) ∣ 2 + 2 v ( y ) 2 ∣ ζ ( x ) − ζ ( y ) ∣ 2 . (12)
For x ∈ K ϱ ( x o ) x\in K_{\varrho}(x_o) x ∈ K ϱ ( x o ) , y ∉ K R ( x o ) y\notin K_R(x_o) y ∈ / K R ( x o ) , and R ≥ 2 ϱ R\ge2\varrho R ≥ 2 ϱ ,
∣ x − y ∣ ≥ 1 2 ∣ y − x o ∣ , ∫ R N ∖ K ϱ ( x o ) d y ∣ y − x o ∣ N + 2 s = C ( N , s ) ϱ 2 s . (13) |x-y|\ge\frac{1}{2}|y-x_o|,\qquad\int_{\mathbb{R}^N\setminus K_{\varrho}(x_o)}\frac{dy}{|y-x_o|^{N+2s}}=\frac{C(N,s)}{\varrho^{2s}}. \tag*{(13)} ∣ x − y ∣ ≥ 2 1 ∣ y − x o ∣ , ∫ R N ∖ K ϱ ( x o ) ∣ y − x o ∣ N + 2 s d y = ϱ 2 s C ( N , s ) . (13)
More generally, if x ∈ K ϱ ^ ( x o ) x\in K_{\hat{\varrho}}(x_o) x ∈ K ϱ ^ ( x o ) , y ∉ K ϱ ( x o ) y\notin K_{\varrho}(x_o) y ∈ / K ϱ ( x o ) , and ϱ ^ < ϱ \hat{\varrho}<\varrho ϱ ^ < ϱ , then
∣ y − x o ∣ ∣ x − y ∣ ≤ ϱ ϱ − ϱ ^ . (14) \frac{|y-x_o|}{|x-y|}\le\frac{\varrho}{\varrho-\hat{\varrho}}. \tag*{(14)} ∣ x − y ∣ ∣ y − x o ∣ ≤ ϱ − ϱ ^ ϱ . (14)
Time regularization and geometric convergence
We use the forward and backward exponential regularizations as in [21 Appendix B] and [24 Lemma B.1] . For a Banach space X X X and a < b a<b a < b , set
( f ) h ( t ) = 1 h ∫ a t e − ( t − τ ) / h f ( τ ) d τ , [ f ] h ( t ) = 1 h ∫ t b e ( t − τ ) / h f ( τ ) d τ . (15) (f)_h(t)=\frac{1}{h}\int_a^t e^{-(t-\tau)/h}f(\tau)\,d\tau,\qquad[f]_h(t)=\frac{1}{h}\int_t^b e^{(t-\tau)/h}f(\tau)\,d\tau. \tag*{(15)} ( f ) h ( t ) = h 1 ∫ a t e − ( t − τ ) / h f ( τ ) d τ , [ f ] h ( t ) = h 1 ∫ t b e ( t − τ ) / h f ( τ ) d τ . (15)
Both regularizations converge strongly to f f f in L p ( a , b ; X ) L^p(a,b;X) L p ( a , b ; X ) for 1 ≤ p < ∞ 1\le p<\infty 1 ≤ p < ∞ , and
∂ t ( f ) h = f − ( f ) h h , ∂ t [ f ] h = [ f ] h − f h . (16) \partial_t(f)_h=\frac{f-(f)_h}{h},\qquad\partial_t[f]_h=\frac{[f]_h-f}{h}. \tag*{(16)} ∂ t ( f ) h = h f − ( f ) h , ∂ t [ f ] h = h [ f ] h − f . (16)
If f ∈ C ( [ a , b ] ; X ) f\in C([a,b];X) f ∈ C ([ a , b ] ; X ) and f ( b ) = 0 f(b)=0 f ( b ) = 0 , then by [24 Lemma B.1, (vi)]
[ f ] h ⟶ f in C ( [ a , b ] ; X ) . (17) [f]_h\longrightarrow f\quad\text{in }C([a,b];X). \tag*{(17)} [ f ] h ⟶ f in C ([ a , b ] ; X ) . (17)
The condition f ( b ) = 0 f(b)=0 f ( b ) = 0 is essential here, since [ f ] h ( b ) = 0 [f]_h(b)=0 [ f ] h ( b ) = 0 .
We finally recall the fast geometric convergence lemma; see [24 Lemma 2.10] .
Lemma 2.3. Let Y n ≥ 0 Y_n\ge0 Y n ≥ 0 satisfy Y n + 1 ≤ A B n Y n 1 + α Y_{n+1}\le AB^nY_n^{1+\alpha} Y n + 1 ≤ A B n Y n 1 + α , where A > 0 A>0 A > 0 , B > 1 B>1 B > 1 , and α > 0 \alpha>0 α > 0 . If
Y 0 ≤ A − 1 / α B − 1 / α 2 , Y_0\le A^{-1/\alpha}B^{-1/\alpha^2}, Y 0 ≤ A − 1/ α B − 1/ α 2 , then Y n → 0 Y_n\to0 Y n → 0 as n → ∞ n\to\infty n → ∞ .
The energy estimate
Proposition 3.1. Let u u u be a local weak subsolution (respectively, supersolution) of (1.1), and let Q ( R , S ) = K R × ( t o − S , t o ) ⊂ E T Q(R,S)=K_R\times(t_o-S,t_o)\subset E_T Q ( R , S ) = K R × ( t o − S , t o ) ⊂ E T . Let k ∈ W 1 , 1 ( t o − S , t o ) k\in W^{1,1}(t_o-S,t_o) k ∈ W 1 , 1 ( t o − S , t o ) , and let 0 ≤ ζ ≤ 1 0\le\zeta\le1 0 ≤ ζ ≤ 1 be a piecewise smooth cutoff whose spatial support is contained in a fixed compact subset of K R K_R K R . Write w ± = ( u − k ( t ) ) ± w_{\pm}=(u-k(t))_{\pm} w ± = ( u − k ( t ) ) ± . With the upper sign for a subsolution
and the lower sign for a supersolution, the following estimate holds:
q ∫ K R g ± ( u , k ) ζ 2 d x ∣ t 0 − S t 0 ± ∫ t 0 − S t 0 ∫ K R k ′ ( t ) ζ 2 ( u q − k q ) d x d t + 1 2 ∫ t 0 − S t 0 ∫ K R ∫ K R ∣ ζ ( x , t ) w ± ( x , t ) − ζ ( y , t ) w ± ( y , t ) ∣ 2 K ( x , y , t ) d y d x d t + ∫ t 0 − S t 0 ∫ K R ∫ K R w ∓ ( y , t ) w ± ( x , t ) ζ ( x , t ) 2 K ( x , y , t ) d y d x d t ≤ q ∫ t 0 − S t 0 ∫ K R g ± ( u , k ) ∂ t ( ζ 2 ) d x d t + ∫ t 0 − S t 0 ∫ K R ∫ K R ∣ ζ ( x , t ) − ζ ( y , t ) ∣ 2 w ± ( x , t ) w ± ( y , t ) K ( x , y , t ) d y d x d t + ∫ t 0 − S t 0 ∫ K R ∫ R N ∖ K R w ∓ ( y , t ) ζ ( x , t ) 2 w ± ( x , t ) K ( x , y , t ) d y d x d t . (18) \begin{aligned}
q\int_{K_R} g_{\pm}(u,k)\zeta^2\,\mathrm{d}x\Big|_{t_0-S}^{t_0}
&\pm\int_{t_0-S}^{t_0}\int_{K_R} k'(t)\zeta^2(u^q-k^q)\,\mathrm{d}x\,\mathrm{d}t\\
&+\frac{1}{2}\int_{t_0-S}^{t_0}\int_{K_R}\int_{K_R}|\zeta(x,t)w_{\pm}(x,t)-\zeta(y,t)w_{\pm}(y,t)|^2K(x,y,t)\,\mathrm{d}y\,\mathrm{d}x\,\mathrm{d}t\\
&+\int_{t_0-S}^{t_0}\int_{K_R}\int_{K_R}w_{\mp}(y,t)w_{\pm}(x,t)\zeta(x,t)^2K(x,y,t)\,\mathrm{d}y\,\mathrm{d}x\,\mathrm{d}t\\
&\le q\int_{t_0-S}^{t_0}\int_{K_R}g_{\pm}(u,k)\partial_t(\zeta^2)\,\mathrm{d}x\,\mathrm{d}t\\
&+\int_{t_0-S}^{t_0}\int_{K_R}\int_{K_R}|\zeta(x,t)-\zeta(y,t)|^2w_{\pm}(x,t)w_{\pm}(y,t)K(x,y,t)\,\mathrm{d}y\,\mathrm{d}x\,\mathrm{d}t\\
&+\int_{t_0-S}^{t_0}\int_{K_R}\int_{\mathbb{R}^N\setminus K_R}w_{\mp}(y,t)\zeta(x,t)^2w_{\pm}(x,t)K(x,y,t)\,\mathrm{d}y\,\mathrm{d}x\,\mathrm{d}t. \tag*{(18)}
\end{aligned} q ∫ K R g ± ( u , k ) ζ 2 d x t 0 − S t 0 ± ∫ t 0 − S t 0 ∫ K R k ′ ( t ) ζ 2 ( u q − k q ) d x d t + 2 1 ∫ t 0 − S t 0 ∫ K R ∫ K R ∣ ζ ( x , t ) w ± ( x , t ) − ζ ( y , t ) w ± ( y , t ) ∣ 2 K ( x , y , t ) d y d x d t + ∫ t 0 − S t 0 ∫ K R ∫ K R w ∓ ( y , t ) w ± ( x , t ) ζ ( x , t ) 2 K ( x , y , t ) d y d x d t ≤ q ∫ t 0 − S t 0 ∫ K R g ± ( u , k ) ∂ t ( ζ 2 ) d x d t + ∫ t 0 − S t 0 ∫ K R ∫ K R ∣ ζ ( x , t ) − ζ ( y , t ) ∣ 2 w ± ( x , t ) w ± ( y , t ) K ( x , y , t ) d y d x d t + ∫ t 0 − S t 0 ∫ K R ∫ R N ∖ K R w ∓ ( y , t ) ζ ( x , t ) 2 w ± ( x , t ) K ( x , y , t ) d y d x d t . (18)
The same statement holds with any intermediate terminal time in place of t o t_o t o .
Proof. We prove the supersolution case.
Step 1: the test function and the time term. Choose 0 < ε < S / 4 0<\varepsilon<S/4 0 < ε < S /4 and set
ψ ε ( t ) = { ( t − t o + S ) / ε , t o − S ≤ t ≤ t o − S + ε , 1 , t o − S + ε ≤ t ≤ t o − ε , 2 ( t o − ε / 2 − t ) / ε , t o − ε < t < t o − ε / 2 , 0 , t o − ε / 2 ≤ t ≤ t o . \psi_{\varepsilon}(t)=
\begin{cases}
(t-t_o+S)/\varepsilon, & t_o-S\le t\le t_o-S+\varepsilon,\\
1, & t_o-S+\varepsilon\le t\le t_o-\varepsilon,\\
2(t_o-\varepsilon/2-t)/\varepsilon, & t_o-\varepsilon<t<t_o-\varepsilon/2,\\
0, & t_o-\varepsilon/2\le t\le t_o.
\end{cases} ψ ε ( t ) = ⎩ ⎨ ⎧ ( t − t o + S ) / ε , 1 , 2 ( t o − ε /2 − t ) / ε , 0 , t o − S ≤ t ≤ t o − S + ε , t o − S + ε ≤ t ≤ t o − ε , t o − ε < t < t o − ε /2 , t o − ε /2 ≤ t ≤ t o .
Set φ = ζ 2 ψ w − \varphi=\zeta^2\psi w_- φ = ζ 2 ψ w − and use ϕ = [ φ ] h \phi=[\varphi]_h ϕ = [ φ ] h in (5) , with the regularizations in (15) taken over ( t o − S , t o ) (t_o-S,t_o) ( t o − S , t o ) . Fubini’s theorem gives
− ∫ t o − S t o ∫ K R u q ∂ t [ φ ] h d x d t = ∫ t o − S t o ∫ K R φ ∂ t ( u q ) h d x d t . -\int_{t_o-S}^{t_o}\int_{K_R}u^q\partial_t[\varphi]_h\,\mathrm{d}x\,\mathrm{d}t
=\int_{t_o-S}^{t_o}\int_{K_R}\varphi\,\partial_t(u^q)_h\,\mathrm{d}x\,\mathrm{d}t. − ∫ t o − S t o ∫ K R u q ∂ t [ φ ] h d x d t = ∫ t o − S t o ∫ K R φ ∂ t ( u q ) h d x d t .
Here [ φ ] h ( t o ) = 0 [\varphi]_h(t_o)=0 [ φ ] h ( t o ) = 0 . The initial endpoint contribution is
− ∫ K R u q ( x , t o − S ) [ φ ] h ( x , t o − S ) d x . -\int_{K_R}u^q(x,t_o-S)[\varphi]_h(x,t_o-S)\,\mathrm{d}x. − ∫ K R u q ( x , t o − S ) [ φ ] h ( x , t o − S ) d x .
It tends to zero as h ↓ 0 h\downarrow0 h ↓ 0 , by (17) , Hölder’s inequality, and φ ( t o − S ) = 0 \varphi(t_o-S)=0 φ ( t o − S ) = 0 .
The map z ↦ ( k − z 1 / q ) + z\mapsto(k-z^{1/q})_+ z ↦ ( k − z 1/ q ) + is nonincreasing. Thus
[ ( u − k ) − − ( ( u q ) h 1 / q − k ) − ] [ u q − ( u q ) h ] ≤ 0. \left[(u-k)_--\left((u^q)_h^{1/q}-k\right)_-\right]\left[u^q-(u^q)_h\right]\le0. [ ( u − k ) − − ( ( u q ) h 1/ q − k ) − ] [ u q − ( u q ) h ] ≤ 0.
Using (16) , we infer
∫ t o − S t o ∫ K R φ ∂ t ( u q ) h d x d t ≤ ∫ t o − S t o ∫ K R ζ 2 ψ ( k − ( u q ) h 1 / q ) + ∂ t ( u q ) h d x d t . (19) \int_{t_o-S}^{t_o}\int_{K_R}\varphi\partial_t(u^q)_h\,\mathrm{d}x\,\mathrm{d}t
\le\int_{t_o-S}^{t_o}\int_{K_R}\zeta^2\psi\left(k-(u^q)_h^{1/q}\right)_+\partial_t(u^q)_h\,\mathrm{d}x\,\mathrm{d}t. \tag*{(19)} ∫ t o − S t o ∫ K R φ ∂ t ( u q ) h d x d t ≤ ∫ t o − S t o ∫ K R ζ 2 ψ ( k − ( u q ) h 1/ q ) + ∂ t ( u q ) h d x d t . (19)
By the formal differentiation, we have
( k − ( u q ) h 1 / q ) + ∂ t ( u q ) h = − q ∂ t g − ( ( u q ) h 1 / q , k ) + k ′ ( t ) ( k q − ( u q ) h ) + . \left(k-(u^q)_h^{1/q}\right)_+\partial_t(u^q)_h
=-q\partial_t g_-\left((u^q)_h^{1/q},k\right)+k'(t)\left(k^q-(u^q)_h\right)_+. ( k − ( u q ) h 1/ q ) + ∂ t ( u q ) h = − q ∂ t g − ( ( u q ) h 1/ q , k ) + k ′ ( t ) ( k q − ( u q ) h ) + .
Substituting it into (19) , integrating by parts, using ψ ε ( t o − S ) = ψ ε ( t o ) = 0 \psi_{\varepsilon}(t_o-S)=\psi_{\varepsilon}(t_o)=0 ψ ε ( t o − S ) = ψ ε ( t o ) = 0 , turns the right-hand side of (19) into
q ∫ t o − S t o ∫ K R g − ( ( u q ) h 1 / q , k ) ζ 2 ψ ε ′ d x d t + q ∫ t o − S t o ∫ K R g − ( ( u q ) h 1 / q , k ) ψ ε ∂ t ( ζ 2 ) d x d t + ∫ t o − S t o ∫ K R ζ 2 ψ ε ′ ( t ) ( k q − ( u q ) h ) + d x d t . (20) q\int_{t_o-S}^{t_o}\int_{K_R}g_{-}\left((\mathbf{u}^{q})_h^{1/q},k\right)\zeta^2\psi_{\varepsilon}'\,\mathrm{d}x\,\mathrm{d}t
+q\int_{t_o-S}^{t_o}\int_{K_R}g_{-}\left((\mathbf{u}^{q})_h^{1/q},k\right)\psi_{\varepsilon}\partial_t(\zeta^2)\,\mathrm{d}x\,\mathrm{d}t
+\int_{t_o-S}^{t_o}\int_{K_R}\zeta^2\psi_{\varepsilon}'(t)\left(k^q-(\mathbf{u}^q)_h\right)_+\,\mathrm{d}x\,\mathrm{d}t. \tag*{(20)} q ∫ t o − S t o ∫ K R g − ( ( u q ) h 1/ q , k ) ζ 2 ψ ε ′ d x d t + q ∫ t o − S t o ∫ K R g − ( ( u q ) h 1/ q , k ) ψ ε ∂ t ( ζ 2 ) d x d t + ∫ t o − S t o ∫ K R ζ 2 ψ ε ′ ( t ) ( k q − ( u q ) h ) + d x d t . (20)
Step 2: passage to the limit in time. We first send h ↓ 0 h\downarrow0 h ↓ 0 in (20) . Since u ∈ C ( [ t o − S , t o ] ; L q + 1 ( K R ) ) u\in C([t_o-S,t_o];L^{q+1}(K_R)) u ∈ C ([ t o − S , t o ] ; L q + 1 ( K R )) , the signed power u q \mathbf{u}^q u q belongs to C ( [ t o − S , t o ] ; L ( q + 1 ) / q ( K R ) ) C([t_o-S,t_o];L^{(q+1)/q}(K_R)) C ([ t o − S , t o ] ; L ( q + 1 ) / q ( K R )) . Its forward averages converge strongly in L ( q + 1 ) / q ( K R × ( t o − S , t o ) ) L^{(q+1)/q}(K_R\times(t_o-S,t_o)) L ( q + 1 ) / q ( K R × ( t o − S , t o )) and in L ( q + 1 ) / q ( K R ) L^{(q+1)/q}(K_R) L ( q + 1 ) / q ( K R ) at each t > t o − S t>t_o-S t > t o − S . They are uniformly bounded in L ∞ ( t o − S , t o ; L ( q + 1 ) / q ( K R ) ) L^\infty(t_o-S,t_o;L^{(q+1)/q}(K_R)) L ∞ ( t o − S , t o ; L ( q + 1 ) / q ( K R )) . By (10) and Hölder’s inequality,
g − ( ( u q ) h 1 / q , k ) ⟶ g − ( u , k ) in L 1 ( K R × ( t o − S , t o ) ) . (21) g_{-}\left((\mathbf{u}^q)_h^{1/q},k\right)\longrightarrow g_{-}(u,k)\quad\text{in }L^1(K_R\times(t_o-S,t_o)). \tag*{(21)} g − ( ( u q ) h 1/ q , k ) ⟶ g − ( u , k ) in L 1 ( K R × ( t o − S , t o )) . (21)
Here k k k is bounded and ( u q ) h 1 / q (\mathbf{u}^q)_h^{1/q} ( u q ) h 1/ q is uniformly bounded in L q + 1 L^{q+1} L q + 1 on this cylinder. For fixed ε \varepsilon ε , (21) gives the limits of the first two terms in (20) .
For the last term, the absolute value of the error is at most
∫ t o − S t o ∣ k ′ ( t ) ∣ ∥ ( u q ) h ( t ) − u q ( t ) ∥ L 1 ( K R ) d t . \int_{t_o-S}^{t_o}|k'(t)|\left\|(\mathbf{u}^q)_h(t)-\mathbf{u}^q(t)\right\|_{L^1(K_R)}\,\mathrm{d}t. ∫ t o − S t o ∣ k ′ ( t ) ∣ ∥ ( u q ) h ( t ) − u q ( t ) ∥ L 1 ( K R ) d t .
The integrand converges to zero for t > t o − S t>t_o-S t > t o − S and is bounded by a constant times ∣ k ′ ( t ) ∣ |k'(t)| ∣ k ′ ( t ) ∣ . The dominated convergence theorem forces the above integral to become zero. Hence, after h ↓ 0 h\downarrow0 h ↓ 0 , we get from (20) that
q ∫ t o − S t o ∫ K R g − ( u , k ) ζ 2 ψ ε ′ d x d t + q ∫ t o − S t o ∫ K R g − ( u , k ) ψ ε ∂ t ( ζ 2 ) d x d t + ∫ t o − S t o ∫ K R ζ 2 ψ ε ′ ( t ) ( k q − u q ) + d x d t . q\int_{t_o-S}^{t_o}\int_{K_R}g_{-}(u,k)\zeta^2\psi_{\varepsilon}'\,\mathrm{d}x\,\mathrm{d}t
+q\int_{t_o-S}^{t_o}\int_{K_R}g_{-}(u,k)\psi_{\varepsilon}\partial_t(\zeta^2)\,\mathrm{d}x\,\mathrm{d}t
+\int_{t_o-S}^{t_o}\int_{K_R}\zeta^2\psi_{\varepsilon}'(t)(k^q-u^q)_+\,\mathrm{d}x\,\mathrm{d}t. q ∫ t o − S t o ∫ K R g − ( u , k ) ζ 2 ψ ε ′ d x d t + q ∫ t o − S t o ∫ K R g − ( u , k ) ψ ε ∂ t ( ζ 2 ) d x d t + ∫ t o − S t o ∫ K R ζ 2 ψ ε ′ ( t ) ( k q − u q ) + d x d t .
Next let ε ↓ 0 \varepsilon\downarrow0 ε ↓ 0 . By (10) , the function t ↦ ∫ K R g − ( u , k ) ζ 2 d x t\mapsto\int_{K_R}g_{-}(u,k)\zeta^2\,\mathrm{d}x t ↦ ∫ K R g − ( u , k ) ζ 2 d x is continuous. The limit of (20) is therefore
− q ∫ K R g − ( u , k ) ζ 2 d x ∣ t o − S t o + q ∫ t o − S t o ∫ K R g − ( u , k ) ∂ t ( ζ 2 ) d x d t + ∫ t o − S t o ∫ K R k ′ ( t ) ζ 2 ( k q − u q ) + d x d t . (22) -q\left.\int_{K_R}g_{-}(u,k)\zeta^2\,\mathrm{d}x\right|_{t_o-S}^{t_o}
+q\int_{t_o-S}^{t_o}\int_{K_R}g_{-}(u,k)\partial_t(\zeta^2)\,\mathrm{d}x\,\mathrm{d}t
+\int_{t_o-S}^{t_o}\int_{K_R}k'(t)\zeta^2(k^q-u^q)_+\,\mathrm{d}x\,\mathrm{d}t. \tag*{(22)} − q ∫ K R g − ( u , k ) ζ 2 d x t o − S t o + q ∫ t o − S t o ∫ K R g − ( u , k ) ∂ t ( ζ 2 ) d x d t + ∫ t o − S t o ∫ K R k ′ ( t ) ζ 2 ( k q − u q ) + d x d t . (22)
Step 3: convergence of the spatial term. We first split the fractional term into
∫ t 0 − S t 0 ∫ K R ∫ K R ( u ( x , t ) − u ( y , t ) ) ( [ φ ] h ( x , t ) − [ φ ] h ( y , t ) ) K ( x , y , t ) d x d y d t + 2 ∫ t 0 − S t 0 ∫ R N ∖ K R ∫ K R u ( x , t ) [ φ ] h ( x , t ) K ( x , y , t ) d x d y d t − 2 ∫ t 0 − S t 0 ∫ R N ∖ K R ∫ K R u ( y , t ) [ φ ] h ( x , t ) K ( x , y , t ) d x d y d t . (23) \begin{aligned}
&\int_{t_0-S}^{t_0}\int_{K_R}\int_{K_R}(u(x,t)-u(y,t))([\varphi]_h(x,t)-[\varphi]_h(y,t))K(x,y,t)\,\mathrm{d}x\,\mathrm{d}y\,\mathrm{d}t\\
&\quad+2\int_{t_0-S}^{t_0}\int_{\mathbb{R}^N\setminus K_R}\int_{K_R}u(x,t)[\varphi]_h(x,t)K(x,y,t)\,\mathrm{d}x\,\mathrm{d}y\,\mathrm{d}t\\
&\quad-2\int_{t_0-S}^{t_0}\int_{\mathbb{R}^N\setminus K_R}\int_{K_R}u(y,t)[\varphi]_h(x,t)K(x,y,t)\,\mathrm{d}x\,\mathrm{d}y\,\mathrm{d}t. \tag*{(23)}
\end{aligned} ∫ t 0 − S t 0 ∫ K R ∫ K R ( u ( x , t ) − u ( y , t )) ([ φ ] h ( x , t ) − [ φ ] h ( y , t )) K ( x , y , t ) d x d y d t + 2 ∫ t 0 − S t 0 ∫ R N ∖ K R ∫ K R u ( x , t ) [ φ ] h ( x , t ) K ( x , y , t ) d x d y d t − 2 ∫ t 0 − S t 0 ∫ R N ∖ K R ∫ K R u ( y , t ) [ φ ] h ( x , t ) K ( x , y , t ) d x d y d t . (23)
On K R × K R K_R\times K_R K R × K R , the Cauchy–Schwarz inequality bounds the error by
C 1 ( ∫ t 0 − S t 0 [ u ( ⋅ , t ) ] W s , 2 ( K R ) 2 d t ) 1 / 2 ( ∫ t 0 − S t 0 ∥ [ φ ] h ( ⋅ , t ) − φ ( ⋅ , t ) ∥ W s , 2 ( K R ) 2 d t ) 1 / 2 → h → 0 0. C_1\left(\int_{t_0-S}^{t_0}[u(\cdot,t)]_{W^{s,2}(K_R)}^2\,\mathrm{d}t\right)^{1/2}\left(\int_{t_0-S}^{t_0}\|[\varphi]_h(\cdot,t)-\varphi(\cdot,t)\|_{W^{s,2}(K_R)}^2\,\mathrm{d}t\right)^{1/2}\xrightarrow{h\to0}0. C 1 ( ∫ t 0 − S t 0 [ u ( ⋅ , t ) ] W s , 2 ( K R ) 2 d t ) 1/2 ( ∫ t 0 − S t 0 ∥ [ φ ] h ( ⋅ , t ) − φ ( ⋅ , t ) ∥ W s , 2 ( K R ) 2 d t ) 1/2 h → 0 0.
Thus the first term of (23) goes to
∫ t 0 − S t 0 ∫ K R ∫ K R ( u ( x , t ) − u ( y , t ) ) ( φ ( x , t ) − φ ( y , t ) ) K ( x , y , t ) d x d y d t . \int_{t_0-S}^{t_0}\int_{K_R}\int_{K_R}(u(x,t)-u(y,t))(\varphi(x,t)-\varphi(y,t))K(x,y,t)\,\mathrm{d}x\,\mathrm{d}y\,\mathrm{d}t. ∫ t 0 − S t 0 ∫ K R ∫ K R ( u ( x , t ) − u ( y , t )) ( φ ( x , t ) − φ ( y , t )) K ( x , y , t ) d x d y d t .
For the interactions with R N ∖ K R \mathbb{R}^N\setminus K_R R N ∖ K R , use the positive distance between the fixed spatial support of ζ \zeta ζ and ∂ K R \partial K_R ∂ K R . The error of the second term of (23) tends to zero by local L 2 L^2 L 2 convergence. The error of the final term is bounded by
C ( ζ , R ) sup t ∈ [ t 0 − S , t 0 ] ∥ [ φ ] h ( t ) − φ ( t ) ∥ L 1 ( K R ) ∫ t 0 − S t 0 ∫ R N ∣ u ( y , t ) ∣ ( 1 + ∣ y ∣ ) N + 2 s d y d t ⟶ 0. C(\zeta,R)\sup_{t\in[t_0-S,t_0]}\|[\varphi]_h(t)-\varphi(t)\|_{L^1(K_R)}\int_{t_0-S}^{t_0}\int_{\mathbb{R}^N}\frac{|u(y,t)|}{(1+|y|)^{N+2s}}\,\mathrm{d}y\,\mathrm{d}t\longrightarrow0. C ( ζ , R ) t ∈ [ t 0 − S , t 0 ] sup ∥ [ φ ] h ( t ) − φ ( t ) ∥ L 1 ( K R ) ∫ t 0 − S t 0 ∫ R N ( 1 + ∣ y ∣ ) N + 2 s ∣ u ( y , t ) ∣ d y d t ⟶ 0.
The supremum tends to zero by (17) ; φ \varphi φ vanishes near t 0 t_0 t 0 . Finally ψ ε → 1 \psi_\varepsilon\to1 ψ ε → 1 almost everywhere. Thus, (23) tends to
∫ t 0 − S t 0 ∫ R N ∫ R N ( u ( x , t ) − u ( y , t ) ) ( ζ 2 w − ( x , t ) − ζ 2 w − ( y , t ) ) K ( x , y , t ) d x d y d t , \int_{t_0-S}^{t_0}\int_{\mathbb{R}^N}\int_{\mathbb{R}^N}(u(x,t)-u(y,t))\left(\zeta^2w_-(x,t)-\zeta^2w_-(y,t)\right)K(x,y,t)\,\mathrm{d}x\,\mathrm{d}y\,\mathrm{d}t, ∫ t 0 − S t 0 ∫ R N ∫ R N ( u ( x , t ) − u ( y , t )) ( ζ 2 w − ( x , t ) − ζ 2 w − ( y , t ) ) K ( x , y , t ) d x d y d t ,
which for simplicity we denote as I \mathrm{I} I . The next step consists of using an algebraic identity to decompose the above integral.
Step 4: the truncation algebra. Fix t t t and use w ± = ( u − k ) ± w_\pm=(u-k)_\pm w ± = ( u − k ) ± . The exact identity is
( u ( x ) − u ( y ) ) ( ζ ( x ) 2 w − ( x ) − ζ ( y ) 2 w − ( y ) ) = − ∣ ζ ( x ) w − ( x ) − ζ ( y ) w − ( y ) ∣ 2 + w − ( x ) w − ( y ) ∣ ζ ( x ) − ζ ( y ) ∣ 2 − w + ( x ) w − ( y ) ζ ( y ) 2 − w + ( y ) w − ( x ) ζ ( x ) 2 . \begin{aligned}
(u(x)-u(y))\left(\zeta(x)^2w_-(x)-\zeta(y)^2w_-(y)\right)
={}&-\left|\zeta(x)w_-(x)-\zeta(y)w_-(y)\right|^2\\
&+w_-(x)w_-(y)|\zeta(x)-\zeta(y)|^2\\
&-w_+(x)w_-(y)\zeta(y)^2-w_+(y)w_-(x)\zeta(x)^2.
\end{aligned} ( u ( x ) − u ( y )) ( ζ ( x ) 2 w − ( x ) − ζ ( y ) 2 w − ( y ) ) = − ∣ ζ ( x ) w − ( x ) − ζ ( y ) w − ( y ) ∣ 2 + w − ( x ) w − ( y ) ∣ ζ ( x ) − ζ ( y ) ∣ 2 − w + ( x ) w − ( y ) ζ ( y ) 2 − w + ( y ) w − ( x ) ζ ( x ) 2 .
Then, multiplying above identity by 1 2 K ( x , y , t ) \frac{1}{2}K(x,y,t) 2 1 K ( x , y , t ) , integrating it over ( t o − S , t o ) × R N × R N (t_o-S,t_o)\times\mathbb{R}^N\times\mathbb{R}^N ( t o − S , t o ) × R N × R N , and then decomposing the integrals yield
I 2 = − 1 2 ∫ t o − S t o ∫ K R ∫ K R ∣ ζ ( x , t ) w − ( x , t ) − ζ ( y , t ) w − ( y , t ) ∣ 2 K ( x , y , t ) d x d y d t − ∫ t o − S t o ∫ R N ∖ K R ∫ R N ∣ ζ ( x , t ) w − ( x , t ) ∣ 2 K ( x , y , t ) d x d y d t + 1 2 ∫ t o − S t o ∫ K R ∫ K R w − ( x , t ) w − ( y , t ) ∣ ζ ( x , t ) − ζ ( y , t ) ∣ 2 K ( x , y , t ) d x d y d t + ∫ t o − S t o ∫ R N ∖ K R ∫ R N w − ( x , t ) w − ( y , t ) ζ ( x , t ) 2 K ( x , y , t ) d x d y d t − ∫ t o − S t o ∫ K R ∫ K R w + ( x , t ) w − ( y , t ) ζ ( y , t ) 2 K ( x , y , t ) d x d y d t − ∫ t o − S t o ∫ K R ∫ R N ∖ K R w + ( x , t ) w − ( y , t ) ζ ( y , t ) 2 K ( x , y , t ) d x d y d t . (24) \begin{aligned}
\frac{I}{2}={}&-\frac{1}{2}\int_{t_o-S}^{t_o}\int_{K_R}\int_{K_R}|\zeta(x,t)w_-(x,t)-\zeta(y,t)w_-(y,t)|^2K(x,y,t)\,dx\,dy\,dt\\
&-\int_{t_o-S}^{t_o}\int_{\mathbb{R}^N\setminus K_R}\int_{\mathbb{R}^N}|\zeta(x,t)w_-(x,t)|^2K(x,y,t)\,dx\,dy\,dt\\
&+\frac{1}{2}\int_{t_o-S}^{t_o}\int_{K_R}\int_{K_R}w_-(x,t)w_-(y,t)|\zeta(x,t)-\zeta(y,t)|^2K(x,y,t)\,dx\,dy\,dt\\
&+\int_{t_o-S}^{t_o}\int_{\mathbb{R}^N\setminus K_R}\int_{\mathbb{R}^N}w_-(x,t)w_-(y,t)\zeta(x,t)^2K(x,y,t)\,dx\,dy\,dt\\
&-\int_{t_o-S}^{t_o}\int_{K_R}\int_{K_R}w_+(x,t)w_-(y,t)\zeta(y,t)^2K(x,y,t)\,dx\,dy\,dt\\
&-\int_{t_o-S}^{t_o}\int_{K_R}\int_{\mathbb{R}^N\setminus K_R}w_+(x,t)w_-(y,t)\zeta(y,t)^2K(x,y,t)\,dx\,dy\,dt. \tag*{(24)}
\end{aligned} 2 I = − 2 1 ∫ t o − S t o ∫ K R ∫ K R ∣ ζ ( x , t ) w − ( x , t ) − ζ ( y , t ) w − ( y , t ) ∣ 2 K ( x , y , t ) d x d y d t − ∫ t o − S t o ∫ R N ∖ K R ∫ R N ∣ ζ ( x , t ) w − ( x , t ) ∣ 2 K ( x , y , t ) d x d y d t + 2 1 ∫ t o − S t o ∫ K R ∫ K R w − ( x , t ) w − ( y , t ) ∣ ζ ( x , t ) − ζ ( y , t ) ∣ 2 K ( x , y , t ) d x d y d t + ∫ t o − S t o ∫ R N ∖ K R ∫ R N w − ( x , t ) w − ( y , t ) ζ ( x , t ) 2 K ( x , y , t ) d x d y d t − ∫ t o − S t o ∫ K R ∫ K R w + ( x , t ) w − ( y , t ) ζ ( y , t ) 2 K ( x , y , t ) d x d y d t − ∫ t o − S t o ∫ K R ∫ R N ∖ K R w + ( x , t ) w − ( y , t ) ζ ( y , t ) 2 K ( x , y , t ) d x d y d t . (24)
Dropping the second and the last terms of (24) and combining the remaining terms with (22) , we prove (18) . The subsolution case is obtained by testing with ( u − k ) + (u-k)_+ ( u − k ) + , or by applying the supersolution result to − u -u − u . □ \square □
Remark 3.1. If ζ \zeta ζ vanishes at the initial time, apply (18) up to an arbitrary terminal time, and then take the essential supremum. Adding the resulting bound to the bound for the full time interval controls both the time supremum and the spatial energy, at the cost of a fixed constant. When a moving level is used to absorb the tail, this absorption is performed before taking the supremum. The coefficient 1 / 2 1/2 1/2 is harmless in all subsequent iterations, but it must agree with the normalization of (5) .
A De Giorgi lemma separating the solution from zero
In this and the next sections, fix a reference cylinder
Q = K R ( x o ) × ( T 1 , T 2 ) ⊆ E T . Q=K_R(x_o)\times(T_1,T_2)\subseteq E_T. Q = K R ( x o ) × ( T 1 , T 2 ) ⊆ E T .
Throughout the sections, we assume for some M > 0 M>0 M > 0 ,
ess sup Q ∣ u ∣ ≤ 2 M \operatorname*{ess\,sup}_{Q}|u|\leq2M Q ess sup ∣ u ∣ ≤ 2 M
and denote θ = M q − 1 \theta=M^{q-1} θ = M q − 1 . Every time interval used in a lemma is assumed to lie in ( T 1 , T 2 ) (T_1,T_2) ( T 1 , T 2 ) . This convention is necessary also for the forward time intervals in Section 5.
Lemma 4.1. There exist ν ∈ ( 0 , 1 ) \nu\in(0,1) ν ∈ ( 0 , 1 ) and γ ∗ ∗ ≥ 1 \gamma_{**}\geq1 γ ∗∗ ≥ 1 , depending only on data, with the following property. Let u u u be a locally bounded, local weak supersolution, let 2 ϱ ≤ R 2\varrho\leq R 2 ϱ ≤ R , and assume ( x o , t o ) + Q ϱ ( θ ) ⊆ Q (x_o,t_o)+Q_\varrho(\theta)\subseteq Q ( x o , t o ) + Q ϱ ( θ ) ⊆ Q . If
∣ { u ≤ M } ∩ ( ( x o , t o ) + Q ϱ ( θ ) ) ∣ ≤ ν ∣ Q ϱ ( θ ) ∣ , \left|\{u\leq M\}\cap\big((x_o,t_o)+Q_\varrho(\theta)\big)\right|\leq\nu\left|Q_\varrho(\theta)\right|, { u ≤ M } ∩ ( ( x o , t o ) + Q ϱ ( θ ) ) ≤ ν ∣ Q ϱ ( θ ) ∣ ,
then either
γ ∗ ∗ Tail ( u − ; Q ) > M q (25) \gamma_{**}\operatorname{Tail}(u_-;Q)>M^q \tag*{(25)} γ ∗∗ Tail ( u − ; Q ) > M q (25)
or
u ≥ M / 4 a.e. in ( x o , t o ) + Q ϱ / 2 ( θ ) . (26) u \ge M/4 \qquad\text{a.e. in } (x_o,t_o)+Q_{\varrho/2}(\theta). \tag*{(26)} u ≥ M /4 a.e. in ( x o , t o ) + Q ϱ /2 ( θ ) . (26)
For a subsolution, the corresponding statement applies to − u -u − u , with u + u_+ u + in (25) .
Proof. Translate ( x o , t o ) (x_o,t_o) ( x o , t o ) to ( 0 , 0 ) (0,0) ( 0 , 0 ) . We assume that (25) fails and choose γ ∗ ∗ \gamma_{**} γ ∗∗ shortly. Introduce the levels and radii
k n = M 2 + M 2 n + 1 , ϱ n = ϱ 2 + ϱ 2 n + 1 , k_n=\frac{M}{2}+\frac{M}{2^{n+1}}, \qquad\varrho_n=\frac{\varrho}{2}+\frac{\varrho}{2^{n+1}}, k n = 2 M + 2 n + 1 M , ϱ n = 2 ϱ + 2 n + 1 ϱ ,
ϱ ^ n = 3 ϱ n + ϱ n + 1 4 , ϱ ~ n = ϱ n + ϱ n + 1 2 , ϱ ‾ n = ϱ n + 3 ϱ n + 1 4 . \widehat{\varrho}_n=\frac{3\varrho_n+\varrho_{n+1}}{4}, \qquad\widetilde{\varrho}_n=\frac{\varrho_n+\varrho_{n+1}}{2}, \qquad\overline{\varrho}_n=\frac{\varrho_n+3\varrho_{n+1}}{4}. ϱ n = 4 3 ϱ n + ϱ n + 1 , ϱ n = 2 ϱ n + ϱ n + 1 , ϱ n = 4 ϱ n + 3 ϱ n + 1 .
Set
K n = K ϱ n , K ~ n = K ϱ ~ n , K ^ n = K ϱ ^ n , K ‾ n = K ϱ ‾ n , Q n = K n × ( − θ ϱ n 2 s , 0 ) , Q ~ n = K ~ n × ( − θ ϱ n 2 s , 0 ) , Q ^ n = K ^ n × ( − θ ϱ n 2 s , 0 ) , Q ‾ n = K ‾ n × ( − θ ϱ n 2 s , 0 ) . \begin{gathered}
K_n=K_{\varrho_n}, \qquad\widetilde{K}_n=K_{\widetilde{\varrho}_n}, \qquad\widehat{K}_n=K_{\widehat{\varrho}_n}, \qquad\overline{K}_n=K_{\overline{\varrho}_n},\\
Q_n=K_n\times(-\theta\varrho_n^{2s},0), \qquad\widetilde{Q}_n=\widetilde{K}_n\times(-\theta\varrho_n^{2s},0),\\
\widehat{Q}_n=\widehat{K}_n\times(-\theta\varrho_n^{2s},0), \qquad\overline{Q}_n=\overline{K}_n\times(-\theta\varrho_n^{2s},0).
\end{gathered} K n = K ϱ n , K n = K ϱ n , K n = K ϱ n , K n = K ϱ n , Q n = K n × ( − θ ϱ n 2 s , 0 ) , Q n = K n × ( − θ ϱ n 2 s , 0 ) , Q n = K n × ( − θ ϱ n 2 s , 0 ) , Q n = K n × ( − θ ϱ n 2 s , 0 ) .
Thus
Q n + 1 ⊂ Q ‾ n ⊂ Q ~ n ⊂ Q ^ n ⊂ Q n . Q_{n+1}\subset\overline{Q}_n\subset\widetilde{Q}_n\subset\widehat{Q}_n\subset Q_n. Q n + 1 ⊂ Q n ⊂ Q n ⊂ Q n ⊂ Q n .
Choose ζ \zeta ζ equal to one in Q ~ n \widetilde{Q}_n Q n , supported spatially in K ~ n \widetilde{K}_n K n , vanishing at the bottom of Q n Q_n Q n , and zero at times below the bottom of Q ^ n \widehat{Q}_n Q n . It can be chosen so that
∣ D ζ ∣ ≤ C 2 n ϱ , ∣ ∂ t ζ ∣ ≤ C ( s ) 2 n θ ϱ 2 s . |D\zeta|\le\frac{C2^n}{\varrho}, \qquad|\partial_t\zeta|\le\frac{C(s)2^n}{\theta\varrho^{2s}}. ∣ D ζ ∣ ≤ ϱ C 2 n , ∣ ∂ t ζ ∣ ≤ θ ϱ 2 s C ( s ) 2 n .
Step 1: moving levels and tail cancellation. Define
ℓ ( t ) = γ ~ M 1 − q ∫ T 1 t ∫ R N ∖ K R u − ( y , τ ) ∣ y ∣ N + 2 s d y d τ , k n ( t ) = k n − ℓ ( t ) . \ell(t)=\widetilde{\gamma}M^{1-q}\int_{T_1}^{t}\int_{\mathbb{R}^N\setminus K_R}\frac{u_-(y,\tau)}{|y|^{N+2s}}\,dy\,d\tau, \qquad k_n(t)=k_n-\ell(t). ℓ ( t ) = γ M 1 − q ∫ T 1 t ∫ R N ∖ K R ∣ y ∣ N + 2 s u − ( y , τ ) d y d τ , k n ( t ) = k n − ℓ ( t ) .
Here γ ~ \widetilde{\gamma} γ depends only on data and will be chosen later.
If γ ∗ ∗ ≥ 4 γ ~ \gamma_{**}\ge4\widetilde{\gamma} γ ∗∗ ≥ 4 γ , failure of (25) implies
0 ≤ ℓ ( t ) ≤ M / 4 , M / 4 ≤ k n ( t ) ≤ M . 0\le\ell(t)\le M/4, \qquad M/4\le k_n(t)\le M. 0 ≤ ℓ ( t ) ≤ M /4 , M /4 ≤ k n ( t ) ≤ M .
For each n n n , set
w − = ( u − k n ( t ) ) − , A n = { u + ℓ ( t ) < k n } ∩ Q n . w_-=(u-k_n(t))_-, \qquad A_n=\{u+\ell(t)<k_n\}\cap Q_n. w − = ( u − k n ( t ) ) − , A n = { u + ℓ ( t ) < k n } ∩ Q n .
On A n A_n A n , we have 0 < w − ≤ 3 M 0<w_-\le3M 0 < w − ≤ 3 M , and by (8) –(9) , we have
c q M q − 1 w − 2 ≤ g − ( u , k n ( t ) ) ≤ C q M q − 1 w − 2 , c_qM^{q-1}w_-^2\le g_-(u,k_n(t))\le C_qM^{q-1}w_-^2, c q M q − 1 w − 2 ≤ g − ( u , k n ( t )) ≤ C q M q − 1 w − 2 ,
( k n ( t ) q − u q ) + ≥ c q M q − 1 w − . (k_n(t)^q-u^q)_+\ge c_qM^{q-1}w_-. ( k n ( t ) q − u q ) + ≥ c q M q − 1 w − .
Thus, the time-cutoff term in (18) is bounded by
q ∬ Q n g − ( u , k ) ∂ t ( ζ 2 ) d x d t ≤ C 2 n M q + 1 θ ϱ 2 s ∣ A n ∣ . q\iint_{Q_n}g_-(u,k)\partial_t(\zeta^2)\,dx\,dt\le\frac{C2^nMq+1}{\theta\varrho^{2s}}|A_n|. q ∬ Q n g − ( u , k ) ∂ t ( ζ 2 ) d x d t ≤ θ ϱ 2 s C 2 n M q + 1 ∣ A n ∣.
By (11) , the spatial cutoff term is estimated by
∫ − θ ϱ n 2 s 0 ∫ K n ∫ K n ∣ ζ ( x , t ) − ζ ( y , t ) ∣ 2 w − ( x , t ) w − ( y , t ) K ( x , y , t ) d y d x d t ≤ C 4 n M 2 ϱ 2 s ∣ A n ∣ . \begin{aligned}
\int_{-\theta\varrho_n^{2s}}^0\int_{K_n}\int_{K_n}|\zeta(x,t)-\zeta(y,t)|^2w_-(x,t)w_-(y,t)K(x,y,t)\,dy\,dx\,dt \\
&\le C4^n\frac{M^2}{\varrho^{2s}}|A_n|.
\end{aligned} ∫ − θ ϱ n 2 s 0 ∫ K n ∫ K n ∣ ζ ( x , t ) − ζ ( y , t ) ∣ 2 w − ( x , t ) w − ( y , t ) K ( x , y , t ) d y d x d t ≤ C 4 n ϱ 2 s M 2 ∣ A n ∣.
Split R N \ K n \mathbb{R}^{N}\backslash K_n R N \ K n into K R \ K n K_R\backslash K_n K R \ K n and R N \ K R \mathbb{R}^{N}\backslash K_R R N \ K R . For the first part, the bound w − ≤ 3 M w_- \le3M w − ≤ 3 M and (14) applied on the support of ζ \zeta ζ give
∫ − θ ϱ n 2 s 0 ∫ K n ∫ K R \ K n w − ( y , t ) ζ ( x , t ) 2 w − ( x , t ) K ( x , y , t ) d y d x d t ≤ B n C M 2 ϱ 2 s ∣ A n ∣ . \int_{-\theta\varrho_n^{2s}}^{0}\int_{K_n}\int_{K_R\backslash K_n} w_-(y,t)\zeta(x,t)^2w_-(x,t)K(x,y,t)\,dy\,dx\,dt \le B^n C\frac{M^2}{\varrho^{2s}}|A_n|. ∫ − θ ϱ n 2 s 0 ∫ K n ∫ K R \ K n w − ( y , t ) ζ ( x , t ) 2 w − ( x , t ) K ( x , y , t ) d y d x d t ≤ B n C ϱ 2 s M 2 ∣ A n ∣.
For the second part, using w − ( y , t ) ≤ M + u − ( y , t ) w_-(y,t)\le M+u_-(y,t) w − ( y , t ) ≤ M + u − ( y , t ) and (13) yield,
∫ − θ ϱ n 2 s 0 ∫ K n ∫ R N \ K R w − ( y , t ) ζ ( x , t ) 2 w − ( x , t ) K ( x , y , t ) d y d x d t ≤ C M 2 ϱ 2 s ∣ A n ∣ + C ∫ − θ ϱ n 2 s 0 ∫ K n ζ 2 w − ∫ R N \ K R u − ( y , t ) ∣ y ∣ N + 2 s d y d x d t . \begin{aligned}
\int_{-\theta\varrho_n^{2s}}^{0}\int_{K_n}\int_{\mathbb{R}^{N}\backslash K_R} w_-(y,t)\zeta(x,t)^2w_-(x,t)K(x,y,t)\,dy\,dx\,dt\\
\le\frac{CM^2}{\varrho^{2s}}|A_n|+C\int_{-\theta\varrho_n^{2s}}^{0}\int_{K_n}\zeta^2w_-\int_{\mathbb{R}^{N}\backslash K_R}\frac{u_-(y,t)}{|y|^{N+2s}}\,dy\,dx\,dt.
\end{aligned} ∫ − θ ϱ n 2 s 0 ∫ K n ∫ R N \ K R w − ( y , t ) ζ ( x , t ) 2 w − ( x , t ) K ( x , y , t ) d y d x d t ≤ ϱ 2 s C M 2 ∣ A n ∣ + C ∫ − θ ϱ n 2 s 0 ∫ K n ζ 2 w − ∫ R N \ K R ∣ y ∣ N + 2 s u − ( y , t ) d y d x d t .
Thus, we combine them to get
∫ − θ ϱ n 2 s 0 ∫ K n ∫ R N \ K n w − ( y , t ) ζ ( x , t ) 2 w − ( x , t ) K ( x , y , t ) d y d x d t ≤ C B n M 2 ϱ 2 s ∣ A n ∣ + C ∫ − θ ϱ n 2 s 0 ∫ K n ζ 2 w − ∫ R N \ K R u − ( y , t ) ∣ y ∣ N + 2 s d y d x d t . (27) \begin{aligned}
\int_{-\theta\varrho_n^{2s}}^{0}\int_{K_n}\int_{\mathbb{R}^{N}\backslash K_n}w_-(y,t)\zeta(x,t)^2w_-(x,t)K(x,y,t)\,dy\,dx\,dt\\
\le\frac{CB^nM^2}{\varrho^{2s}}|A_n|+C\int_{-\theta\varrho_n^{2s}}^{0}\int_{K_n}\zeta^2w_-\int_{\mathbb{R}^{N}\backslash K_R}\frac{u_-(y,t)}{|y|^{N+2s}}\,dy\,dx\,dt. \tag*{(27)}
\end{aligned} ∫ − θ ϱ n 2 s 0 ∫ K n ∫ R N \ K n w − ( y , t ) ζ ( x , t ) 2 w − ( x , t ) K ( x , y , t ) d y d x d t ≤ ϱ 2 s C B n M 2 ∣ A n ∣ + C ∫ − θ ϱ n 2 s 0 ∫ K n ζ 2 w − ∫ R N \ K R ∣ y ∣ N + 2 s u − ( y , t ) d y d x d t . (27)
Here and below B > 1 B>1 B > 1 depends only on data and may be enlarged once to dominate all geometric factors. Since k n ′ ( t ) = − ℓ ′ ( t ) k_n'(t)=-\ell'(t) k n ′ ( t ) = − ℓ ′ ( t ) and M / 4 ≤ k n ( t ) ≤ M M/4\le k_n(t)\le M M /4 ≤ k n ( t ) ≤ M , by (9) with a = u ( x , t ) a=u(x,t) a = u ( x , t ) , k = k n ( t ) k=k_n(t) k = k n ( t ) , we derive
∬ Q n ζ ( x , t ) 2 k n ( t ) q ( u q − k n ( t ) q ) − d x d t ≤ − c q M q − 1 ∫ − θ ϱ n 2 s 0 ∫ K n ℓ ζ 2 w − d x d t . \iint_{Q_n}\zeta(x,t)^2k_n(t)^q\bigl(u^q-k_n(t)^q\bigr)_-\,dx\,dt\le-c_qM^{q-1}\int_{-\theta\varrho_n^{2s}}^{0}\int_{K_n}\ell\,\zeta^2w_-\,dx\,dt. ∬ Q n ζ ( x , t ) 2 k n ( t ) q ( u q − k n ( t ) q ) − d x d t ≤ − c q M q − 1 ∫ − θ ϱ n 2 s 0 ∫ K n ℓ ζ 2 w − d x d t .
Choose γ ~ : = C c q \tilde{\gamma}:=\frac{C}{c_q} γ ~ := c q C to cancel the last term of (27) . The same choice works for every n n n .
Step 2: the energy bound. Discarding the nonnegative mixed term in (18) and using Remark 3.1, we obtain
M q − 1 ess sup − θ ϱ n 2 s < t < 0 ∫ K n w 2 ζ 2 d x + ∫ − θ ϱ n 2 s 0 ∫ K ~ n ∫ K ~ n ∣ w ( x , t ) − w ( y , t ) ∣ 2 ∣ x − y ∣ N + 2 s d y d x d t ≤ C B n ϱ 2 s ( M q + 1 θ + M 2 ) ∣ A n ∣ ≤ C B n M 2 ϱ 2 s ∣ A n ∣ . (28) M^{q-1}\operatorname*{ess\,sup}_{-\theta\varrho_n^{2s}<t<0}\int_{K_n}w^2\zeta^2\,dx+\int_{-\theta\varrho_n^{2s}}^{0}\int_{\widetilde K_n}\int_{\widetilde K_n}\frac{|w(x,t)-w(y,t)|^2}{|x-y|^{N+2s}}\,dy\,dx\,dt
\le\frac{CB^n}{\varrho^{2s}}\left(\frac{M^{q+1}}{\theta}+M^2\right)|A_n|\le\frac{CB^nM^2}{\varrho^{2s}}|A_n|. \tag*{(28)} M q − 1 − θ ϱ n 2 s < t < 0 ess sup ∫ K n w 2 ζ 2 d x + ∫ − θ ϱ n 2 s 0 ∫ K n ∫ K n ∣ x − y ∣ N + 2 s ∣ w ( x , t ) − w ( y , t ) ∣ 2 d y d x d t ≤ ϱ 2 s C B n ( θ M q + 1 + M 2 ) ∣ A n ∣ ≤ ϱ 2 s C B n M 2 ∣ A n ∣. (28)
The last inequality uses θ = M q − 1 \theta=M^{q-1} θ = M q − 1 .
Step 3: embedding and iteration. Choose a second cutoff φ \varphi φ supported in Q ~ n \widetilde Q_n Q n , equal to one in Q ‾ n \overline Q_n Q n , with ∣ D φ ∣ ≤ C 2 n / ϱ |D\varphi|\le C2^n/\varrho ∣ D φ ∣ ≤ C 2 n / ϱ . An application of Lemma 2.2 with v = w − φ v=w-\varphi v = w − φ , I = ( − θ ϑ n 2 s , 0 ) I=(-\theta\vartheta_n^{2s},0) I = ( − θ ϑ n 2 s , 0 ) , and (12) with ζ = φ \zeta=\varphi ζ = φ gives that
∬ Q n w m ^ d x d t ≤ ∫ − θ ϑ n 2 s 0 ∫ K ~ n ∣ w − φ ( x , t ) ∣ m ^ d x d t ≤ C ( ϑ n − 2 s ∫ − θ ϑ n 2 s 0 ∫ K ~ n ∫ K ~ n ∣ w − φ ( x , t ) − w − φ ( y , t ) ∣ 2 ∣ x − y ∣ N + 2 s d y d x + 2 n ( N + 2 s ) ∫ − θ ϑ n 2 s 0 ∫ K ~ n ∣ w − φ ( x , t ) ∣ 2 d x d t ) × ( e s s s u p t ∈ ( − θ ϑ n 2 s , 0 ) ∫ K ~ n ∣ w − φ ( x , t ) ∣ 2 d x d t ) 2 s N \begin{aligned}
\iint_{Q_n} w^{\widehat m}\,dx\,dt \le\int_{-\theta\vartheta_n^{2s}}^0\int_{\widetilde K_n}|w-\varphi(x,t)|^{\widehat m}\,dx\,dt \\
&\le C\left(\vartheta_n^{-2s}\int_{-\theta\vartheta_n^{2s}}^0\int_{\widetilde K_n}\int_{\widetilde K_n}\frac{|w-\varphi(x,t)-w-\varphi(y,t)|^2}{|x-y|^{N+2s}}\,dy\,dx
+2^n(N+2s)\int_{-\theta\vartheta_n^{2s}}^0\int_{\widetilde K_n}|w-\varphi(x,t)|^2\,dx\,dt\right) \\
&\times\left(\mathop{\mathrm{ess\,sup}}_{t\in(-\theta\vartheta_n^{2s},0)}\int_{\widetilde K_n}|w-\varphi(x,t)|^2\,dx\,dt\right)^{\frac{2s}{N}}
\end{aligned} ∬ Q n w m d x d t ≤ ∫ − θ ϑ n 2 s 0 ∫ K n ∣ w − φ ( x , t ) ∣ m d x d t ≤ C ( ϑ n − 2 s ∫ − θ ϑ n 2 s 0 ∫ K n ∫ K n ∣ x − y ∣ N + 2 s ∣ w − φ ( x , t ) − w − φ ( y , t ) ∣ 2 d y d x + 2 n ( N + 2 s ) ∫ − θ ϑ n 2 s 0 ∫ K n ∣ w − φ ( x , t ) ∣ 2 d x d t ) × ( ess sup t ∈ ( − θ ϑ n 2 s , 0 ) ∫ K n ∣ w − φ ( x , t ) ∣ 2 d x d t ) N 2 s
(2.7)
≤ C ( ϑ n 2 s ∫ − θ ϑ n 2 s 0 ∫ K ~ n ∫ K ~ n ∣ w − ( x , t ) − w − ( y , t ) ∣ 2 ∣ x − y ∣ N + 2 s d y d x d t + B n ∫ − θ ϑ n 2 s 0 ∫ K ~ n w 2 ( y , t ) d y d t ) × ( e s s s u p t ∈ ( − θ ϑ n 2 s , 0 ) ∫ K n ∣ w − ( x , t ) ∣ 2 ζ ( x , t ) 2 d x ) 2 s N ≤ C B n M m ^ M ( 1 − q ) 2 s N ϱ − 2 s ( 1 + 2 s N ) ∣ A n ∣ 1 + 2 s N . (29) \begin{aligned}
\le C\left(\vartheta_n^{2s}\int_{-\theta\vartheta_n^{2s}}^0\int_{\widetilde K_n}\int_{\widetilde K_n}\frac{|w_-(x,t)-w_-(y,t)|^2}{|x-y|^{N+2s}}\,dy\,dx\,dt
+B^n\int_{-\theta\vartheta_n^{2s}}^0\int_{\widetilde K_n}w^2(y,t)\,dy\,dt\right) \\
&\times\left(\mathop{\mathrm{ess\,sup}}_{t\in(-\theta\vartheta_n^{2s},0)}\int_{K_n}|w_-(x,t)|^2\zeta(x,t)^2\,dx\right)^{\frac{2s}{N}} \\
&\le CB^nM^{\widehat m}M^{(1-q)\frac{2s}{N}}\varrho^{-2s(1+\frac{2s}{N})}|A_n|^{1+\frac{2s}{N}}. \tag*{(29)}
\end{aligned} ≤ C ( ϑ n 2 s ∫ − θ ϑ n 2 s 0 ∫ K n ∫ K n ∣ x − y ∣ N + 2 s ∣ w − ( x , t ) − w − ( y , t ) ∣ 2 d y d x d t + B n ∫ − θ ϑ n 2 s 0 ∫ K n w 2 ( y , t ) d y d t ) × ( ess sup t ∈ ( − θ ϑ n 2 s , 0 ) ∫ K n ∣ w − ( x , t ) ∣ 2 ζ ( x , t ) 2 d x ) N 2 s ≤ C B n M m M ( 1 − q ) N 2 s ϱ − 2 s ( 1 + N 2 s ) ∣ A n ∣ 1 + N 2 s . (29)
The last inequality is derived by (28) .
On A n + 1 A_{n+1} A n + 1 , w − ≥ k n − k n + 1 = M / 2 n + 2 w_-\ge k_n-k_{n+1}=M/2^{n+2} w − ≥ k n − k n + 1 = M / 2 n + 2 . Using Q n + 1 ⊂ Q n ‾ Q_{n+1}\subset\overline{Q_n} Q n + 1 ⊂ Q n , divide (29) by ( M / 2 n + 2 ) m ^ (M/2^{n+2})^{\widehat m} ( M / 2 n + 2 ) m to get
∣ A n + 1 ∣ ≤ C B n M ( 1 − q ) 2 s N ϱ − 2 s ( 1 + 2 s N ) ∣ A n ∣ 1 + 2 s N . |A_{n+1}|\le CB^nM^{(1-q)\frac{2s}{N}}\varrho^{-2s(1+\frac{2s}{N})}|A_n|^{1+\frac{2s}{N}}. ∣ A n + 1 ∣ ≤ C B n M ( 1 − q ) N 2 s ϱ − 2 s ( 1 + N 2 s ) ∣ A n ∣ 1 + N 2 s .
Set Y n = ∣ A n ∣ / ∣ Q n ∣ Y_n=|A_n|/|Q_n| Y n = ∣ A n ∣/∣ Q n ∣ . The ratios ∣ Q n ∣ / ∣ Q n + 1 ∣ |Q_n|/|Q_{n+1}| ∣ Q n ∣/∣ Q n + 1 ∣ are bounded, and ∣ Q n ∣ ≃ θ ϱ N + 2 s |Q_n|\simeq\theta\varrho^{N+2s} ∣ Q n ∣ ≃ θ ϱ N + 2 s . Since 2 s N ( N + 2 s ) = 2 s ( 1 + 2 s N ) \frac{2s}{N}(N+2s)=2s(1+\frac{2s}{N}) N 2 s ( N + 2 s ) = 2 s ( 1 + N 2 s ) and M ( 1 − q ) 2 s N θ 2 s N M − 2 s N = 1 M^{(1-q)\frac{2s}{N}}\theta^{\frac{2s}{N}}M^{-\frac{2s}{N}}=1 M ( 1 − q ) N 2 s θ N 2 s M − N 2 s = 1 , the powers cancel:
Y n + 1 ≤ C B n Y n 1 + 2 s N . Y_{n+1}\le CB^nY_n^{1+\frac{2s}{N}}. Y n + 1 ≤ C B n Y n 1 + N 2 s .
Lemma 2.3 gives Y n → 0 Y_n\to0 Y n → 0 if Y 0 ≤ ν Y_0\le\nu Y 0 ≤ ν , with ν = min { 1 / 2 , C − N / ( 2 s ) B − N 2 / ( 4 s 2 ) } \nu=\min\{1/2,C^{-N/(2s)}B^{-N^2/(4s^2)}\} ν = min { 1/2 , C − N / ( 2 s ) B − N 2 / ( 4 s 2 ) } . The initial assumption ensures this because A 0 ⊂ { u ≤ M } ∩ Q ϱ ( θ ) A_0\subset\{u\le M\}\cap Q_\varrho(\theta) A 0 ⊂ { u ≤ M } ∩ Q ϱ ( θ ) . It follows that u + ℓ ( t ) ≥ M / 2 u+\ell(t)\ge M/2 u + ℓ ( t ) ≥ M /2 almost everywhere in Q ϱ / 2 ( θ ) Q_{\varrho/2}(\theta) Q ϱ /2 ( θ ) . Since ℓ ≤ M / 4 \ell\le M/4 ℓ ≤ M /4 , this is (26) . □
Measure estimates near zero
We recall that the amplitude assumption made at the beginning of Section 4 is still assumed here.
A time slice with positive measure
Lemma 5.1. Let ν ∈ ( 0 , 1 ) \nu\in(0,1) ν ∈ ( 0 , 1 ) . If
∣ { u ≤ M } ∩ Q ϱ ( θ ) ∣ > ν ∣ Q ϱ ( θ ) ∣ , |\{u\le M\}\cap Q_\varrho(\theta)|>\nu|Q_\varrho(\theta)|, ∣ { u ≤ M } ∩ Q ϱ ( θ ) ∣ > ν ∣ Q ϱ ( θ ) ∣ , then there is a time τ ∈ ( − θ ϱ 2 s , − ν 2 θ ϱ 2 s ) \tau\in(-\theta\varrho^{2s},-\frac{\nu}{2}\theta\varrho^{2s}) τ ∈ ( − θ ϱ 2 s , − 2 ν θ ϱ 2 s ) such that
∣ { u ( ⋅ , τ ) ≤ M } ∩ K ϱ ∣ > ν 2 ∣ K ϱ ∣ . |\{u(\cdot,\tau)\le M\}\cap K_\varrho|>\frac{\nu}{2}|K_\varrho|. ∣ { u ( ⋅ , τ ) ≤ M } ∩ K ϱ ∣ > 2 ν ∣ K ϱ ∣.
The time τ \tau τ may be chosen to be an admissible time slice for the energy estimates. The same statement holds for − u -u − u .
Proof. If the slice estimate failed almost everywhere on ( − θ ϱ 2 s , − ν 2 θ ϱ 2 s ) \left(-\theta\varrho^{2s},-\frac{\nu}{2}\theta\varrho^{2s}\right) ( − θ ϱ 2 s , − 2 ν θ ϱ 2 s ) , Fubini’s theorem would give
∣ { u ≤ M } ∩ Q ϱ ( θ ) ∣ ≤ ν 2 ∣ K ϱ ∣ ( 1 − ν 2 ) θ 2 s + ∣ K ϱ ∣ ν 2 θ 2 s = ( ν − ν 2 4 ) ∣ Q ϱ ( θ ) ∣ < ν ∣ Q ϱ ( θ ) ∣ . \left|\{u\le M\}\cap Q_{\varrho}(\theta)\right|\le\frac{\nu}{2}|K_{\varrho}|\left(1-\frac{\nu}{2}\right)\theta^{2s}+|K_{\varrho}|\frac{\nu}{2}\theta^{2s}
= \left(\nu-\frac{\nu^{2}}{4}\right)|Q_{\varrho}(\theta)|<\nu|Q_{\varrho}(\theta)|. ∣ { u ≤ M } ∩ Q ϱ ( θ ) ∣ ≤ 2 ν ∣ K ϱ ∣ ( 1 − 2 ν ) θ 2 s + ∣ K ϱ ∣ 2 ν θ 2 s = ( ν − 4 ν 2 ) ∣ Q ϱ ( θ ) ∣ < ν ∣ Q ϱ ( θ ) ∣.
This is a contradiction. The set of qualifying slices has positive measure, so it intersects the full-measure set of admissible slices. □ \square □
Lemma 5.2. Let u u u be a local weak subsolution, let m ∈ ( 0 , 1 ] m\in(0,1] m ∈ ( 0 , 1 ] , and suppose 2 ϱ ≤ R 2\varrho\le R 2 ϱ ≤ R . If
∣ { u ( ⋅ , τ ) ≤ 2 M − m M } ∩ K ϱ ∣ ≥ ν 2 ∣ K ϱ ∣ , \left|\{u(\cdot,\tau)\le2M-mM\}\cap K_{\varrho}\right|\ge\frac{\nu}{2}|K_{\varrho}|, ∣ { u ( ⋅ , τ ) ≤ 2 M − m M } ∩ K ϱ ∣ ≥ 2 ν ∣ K ϱ ∣ ,
then there exist δ ∈ ( 0 , 1 ) \delta\in(0,1) δ ∈ ( 0 , 1 ) and ϵ ∈ ( 0 , 1 / 2 ) \epsilon\in(0,1/2) ϵ ∈ ( 0 , 1/2 ) , depending only on data and ν \nu ν , such that
∣ { u ( ⋅ , t ) ≤ 2 M − ϵ m M } ∩ K ϱ ∣ ≥ ν 4 ∣ K ϱ ∣ (30) \left|\{u(\cdot,t)\le2M-\epsilon mM\}\cap K_{\varrho}\right|\ge\frac{\nu}{4}|K_{\varrho}| \tag*{(30)} ∣ { u ( ⋅ , t ) ≤ 2 M − ϵ m M } ∩ K ϱ ∣ ≥ 4 ν ∣ K ϱ ∣ (30)
for almost every t ∈ ( τ , τ + δ θ ϱ 2 s ) t\in(\tau,\tau+\delta\theta\varrho^{2s}) t ∈ ( τ , τ + δ θ ϱ 2 s ) , provided
Tail ( u + ; Q ) ≤ δ m M q . (31) \operatorname{Tail}(u_+;Q)\le\delta mM^q. \tag*{(31)} Tail ( u + ; Q ) ≤ δ m M q . (31)
The interval must be contained in the reference time interval. The constants δ , ϵ \delta,\epsilon δ , ϵ are independent of m m m . Both the interval and the tail bound may be shortened by replacing δ \delta δ with a smaller positive number. For supersolutions, apply the statement to − u -u − u .
Proof. Put k = 2 M − m M k=2M-mM k = 2 M − m M and w + = ( u − k ) + w_+=(u-k)_+ w + = ( u − k ) + . Choose a nonnegative spatial cutoff ζ \zeta ζ which equals one in K ( 1 − σ ) ϱ K_{(1-\sigma)\varrho} K ( 1 − σ ) ϱ , is supported in K ( 1 − σ / 2 ) ϱ K_{(1-\sigma/2)\varrho} K ( 1 − σ /2 ) ϱ , and satisfies ∣ D ζ ∣ ≤ C / ( σ ϱ ) |D\zeta|\le C/(\sigma\varrho) ∣ D ζ ∣ ≤ C / ( σ ϱ ) . The parameter σ ∈ ( 0 , 1 / 2 ) \sigma\in(0,1/2) σ ∈ ( 0 , 1/2 ) will be fixed below. Since k ≥ M k\ge M k ≥ M and u ≤ 2 M u\le2M u ≤ 2 M , the active values of u u u lie in [ M , 2 M ] [M,2M] [ M , 2 M ] .
For a terminal time t ≤ τ + δ θ ϱ 2 s t\le\tau+\delta\theta\varrho^{2s} t ≤ τ + δ θ ϱ 2 s , Proposition 3.1 gives
q ∫ K ϱ g + ( u ( t ) , k ) ζ 2 d x − q ∫ K ϱ g + ( u ( τ ) , k ) ζ 2 d x ≤ ∫ τ τ + δ θ ϱ 2 s ∫ K ϱ ∫ K ϱ ∣ ζ ( x ) − ζ ( y ) ∣ 2 w + ( x , t ) w + ( y , t ) K ( x , y , t ) d y d x d t + ∫ τ τ + δ θ ϱ 2 s ∫ K ϱ ∫ R N ∖ K ϱ w + ( y , t ) ζ ( x ) 2 w + ( x , t ) K ( x , y , t ) d y d x d t . q\int_{K_{\varrho}}g_+(u(t),k)\zeta^2\,dx-q\int_{K_{\varrho}}g_+(u(\tau),k)\zeta^2\,dx
\le\int_{\tau}^{\tau+\delta\theta\varrho^{2s}}\int_{K_{\varrho}}\int_{K_{\varrho}}|\zeta(x)-\zeta(y)|^2w_+(x,t)w_+(y,t)K(x,y,t)\,dy\,dx\,dt
+\int_{\tau}^{\tau+\delta\theta\varrho^{2s}}\int_{K_{\varrho}}\int_{\mathbb{R}^N\setminus K_{\varrho}}w_+(y,t)\zeta(x)^2w_+(x,t)K(x,y,t)\,dy\,dx\,dt. q ∫ K ϱ g + ( u ( t ) , k ) ζ 2 d x − q ∫ K ϱ g + ( u ( τ ) , k ) ζ 2 d x ≤ ∫ τ τ + δ θ ϱ 2 s ∫ K ϱ ∫ K ϱ ∣ ζ ( x ) − ζ ( y ) ∣ 2 w + ( x , t ) w + ( y , t ) K ( x , y , t ) d y d x d t + ∫ τ τ + δ θ ϱ 2 s ∫ K ϱ ∫ R N ∖ K ϱ w + ( y , t ) ζ ( x ) 2 w + ( x , t ) K ( x , y , t ) d y d x d t .
Since w + ( x , t ) ≤ m M w_+(x,t)\le mM w + ( x , t ) ≤ m M for x ∈ K R x\in K_R x ∈ K R , and (11) , the first term on the right-hand side is estimated by
∫ τ τ + δ θ ϱ 2 s ∫ K ϱ ∫ K ϱ ( ζ ( x ) − ζ ( y ) ) 2 w + ( y , t ) w + ( x , t ) K ( x , y , t ) d x d y d t ≤ C ( m M ) 2 θ δ σ 2 ∣ K ϱ ∣ . (32) \int_{\tau}^{\tau+\delta\theta\varrho^{2s}}\int_{K_{\varrho}}\int_{K_{\varrho}}(\zeta(x)-\zeta(y))^2w_+(y,t)w_+(x,t)K(x,y,t)\,dx\,dy\,dt
\le C\frac{(mM)^2\theta\delta}{\sigma^2}|K_{\varrho}|. \tag*{(32)} ∫ τ τ + δ θ ϱ 2 s ∫ K ϱ ∫ K ϱ ( ζ ( x ) − ζ ( y ) ) 2 w + ( y , t ) w + ( x , t ) K ( x , y , t ) d x d y d t ≤ C σ 2 ( m M ) 2 θ δ ∣ K ϱ ∣. (32)
Regarding the second term, for y ∉ K R y \notin K_R y ∈ / K R , w + ( y , t ) ≤ u + ( y , t ) w_+(y,t) \le u_+(y,t) w + ( y , t ) ≤ u + ( y , t ) because k > 0 k>0 k > 0 ; for y ∈ K R ∖ K ϱ y \in K_R\setminus K_\varrho y ∈ K R ∖ K ϱ , x ∈ K ( 1 − σ / 2 ) ϱ x \in K_{(1-\sigma/2)\varrho} x ∈ K ( 1 − σ /2 ) ϱ , we use w + ( ⋅ ; t ) ≤ m M w_+(\cdot;t) \le mM w + ( ⋅ ; t ) ≤ m M , thus
∫ τ τ + δ θ 2 s ∫ R N ∖ K ϱ ∫ K ϱ w + ( y , t ) ζ ( x ) 2 w + ( x , t ) K ( x , y , t ) d x d y d t ≤ m M ∫ τ τ + δ θ 2 s ∫ K ϱ ζ ( x ) 2 ∫ R N ∖ K R u + ( y , t ) ∣ x − y ∣ N + 2 s d y d x d t + ( m M ) 2 ∫ τ τ + δ θ 2 s ∫ K ϱ ζ ( x ) 2 ∫ K R ∖ K ϱ 1 ∣ x − y ∣ N + 2 s d y d x d t ≤ C m M ∣ K ϱ ∣ Tail ( u + ; Q ) + C ( m M ) 2 ∣ K ϱ ∣ δ θ σ N + 2 s . (33) \begin{aligned}
&\int_{\tau}^{\tau+\delta\theta^{2s}}\int_{\mathbb{R}^N\setminus K_\varrho}\int_{K_\varrho} w_+(y,t)\,\zeta(x)^2w_+(x,t)K(x,y,t)\,dx\,dy\,dt\\
&\le mM\int_{\tau}^{\tau+\delta\theta^{2s}}\int_{K_\varrho}\zeta(x)^2\int_{\mathbb{R}^N\setminus K_R}\frac{u_+(y,t)}{|x-y|^{N+2s}}\,dy\,dx\,dt\\
&\quad+(mM)^2\int_{\tau}^{\tau+\delta\theta^{2s}}\int_{K_\varrho}\zeta(x)^2\int_{K_R\setminus K_\varrho}\frac{1}{|x-y|^{N+2s}}\,dy\,dx\,dt\\
&\le C\,mM\left|K_\varrho\right|\operatorname{Tail}(u_+;Q)+C\frac{(mM)^2|K_\varrho|\delta\theta}{\sigma^{N+2s}}. \tag*{(33)}
\end{aligned} ∫ τ τ + δ θ 2 s ∫ R N ∖ K ϱ ∫ K ϱ w + ( y , t ) ζ ( x ) 2 w + ( x , t ) K ( x , y , t ) d x d y d t ≤ m M ∫ τ τ + δ θ 2 s ∫ K ϱ ζ ( x ) 2 ∫ R N ∖ K R ∣ x − y ∣ N + 2 s u + ( y , t ) d y d x d t + ( m M ) 2 ∫ τ τ + δ θ 2 s ∫ K ϱ ζ ( x ) 2 ∫ K R ∖ K ϱ ∣ x − y ∣ N + 2 s 1 d y d x d t ≤ C m M ∣ K ϱ ∣ Tail ( u + ; Q ) + C σ N + 2 s ( m M ) 2 ∣ K ϱ ∣ δ θ . (33)
Thus, combining above estimates with θ = M q − 1 \theta=M^{q-1} θ = M q − 1 and (31) , we have for all τ < t ≤ τ + δ θ 2 s \tau<t\le\tau+\delta\theta^{2s} τ < t ≤ τ + δ θ 2 s , that
q ∫ K ϱ g + ( u ( t ) , k ) ζ 2 d x − q ∫ K ϱ g + ( u ( τ ) , k ) ζ 2 d x ≤ C δ M q − 1 ( m M ) 2 ( 1 + σ − 2 + σ − N − 2 s ) ∣ K ϱ ∣ . (34) q\int_{K_\varrho}g_+(u(t),k)\zeta^2\,dx-q\int_{K_\varrho}g_+(u(\tau),k)\zeta^2\,dx
\le C\delta M^{q-1}(mM)^2(1+\sigma^{-2}+\sigma^{-N-2s})|K_\varrho|. \tag*{(34)} q ∫ K ϱ g + ( u ( t ) , k ) ζ 2 d x − q ∫ K ϱ g + ( u ( τ ) , k ) ζ 2 d x ≤ C δ M q − 1 ( m M ) 2 ( 1 + σ − 2 + σ − N − 2 s ) ∣ K ϱ ∣. (34)
By initial measure assumption, the second term on left-hand side of (34) is bounded by
q ∫ K ϱ g + ( u ( τ ) , k ) ζ 2 d x ≤ q g + ( 2 M , k ) ( 1 − ν / 2 ) ∣ K ϱ ∣ . (35) q\int_{K_\varrho}g_+(u(\tau),k)\zeta^2\,dx\le qg_+(2M,k)(1-\nu/2)|K_\varrho|. \tag*{(35)} q ∫ K ϱ g + ( u ( τ ) , k ) ζ 2 d x ≤ q g + ( 2 M , k ) ( 1 − ν /2 ) ∣ K ϱ ∣. (35)
For the first term on the left-hand side of (34) , use ζ = 1 \zeta=1 ζ = 1 in K ( 1 − σ ) ϱ K_{(1-\sigma)\varrho} K ( 1 − σ ) ϱ and the monotonicity of a ↦ g + ( a , k ) a\mapsto g_+(a,k) a ↦ g + ( a , k ) for a ≥ k a\ge k a ≥ k :
q ∫ K ϱ g + ( u ( t ) , k ) ζ 2 d x ≥ q ∫ K ( 1 − σ ) ϱ ∩ { u ( t ) > 2 M − ε m M } g + ( u ( t ) , k ) d x ≥ q g + ( 2 M − ε m M , k ) ∣ { u ( t ) > 2 M − ε m M } ∩ K ( 1 − σ ) ϱ ∣ . \begin{aligned}
q\int_{K_\varrho}g_+(u(t),k)\zeta^2\,dx\ge q\int_{K_{(1-\sigma)\varrho}\cap\{u(t)>2M-\varepsilon mM\}}g_+(u(t),k)\,dx \\
&\ge qg_+(2M-\varepsilon mM,k)\left|\{u(t)>2M-\varepsilon mM\}\cap K_{(1-\sigma)\varrho}\right|.
\end{aligned} q ∫ K ϱ g + ( u ( t ) , k ) ζ 2 d x ≥ q ∫ K ( 1 − σ ) ϱ ∩ { u ( t ) > 2 M − ε m M } g + ( u ( t ) , k ) d x ≥ q g + ( 2 M − ε m M , k ) { u ( t ) > 2 M − ε m M } ∩ K ( 1 − σ ) ϱ .
Since k = 2 M − m M ≥ M k=2M-mM\ge M k = 2 M − m M ≥ M and ε ≤ 1 / 2 \varepsilon\le1/2 ε ≤ 1/2 , we estimate
g + ( 2 M − ε m M , k ) = ∫ k 2 M − ε m M r q − 1 ( r − k ) d r ≥ c q M q − 1 ( m M ) 2 , g_+(2M-\varepsilon mM,k)=\int_k^{2M-\varepsilon mM}r^{q-1}(r-k)\,dr\ge c_qM^{q-1}(mM)^2, g + ( 2 M − ε m M , k ) = ∫ k 2 M − ε m M r q − 1 ( r − k ) d r ≥ c q M q − 1 ( m M ) 2 ,
g + ( 2 M , k ) − g + ( 2 M − ε m M , k ) = ∫ 2 M − ε m M 2 M r q − 1 ( r − k ) d r ≤ C q M q − 1 ( m M ) ( ε m M ) . g_+(2M,k)-g_+(2M-\varepsilon mM,k)=\int_{2M-\varepsilon mM}^{2M}r^{q-1}(r-k)\,dr\le C_qM^{q-1}(mM)(\varepsilon mM). g + ( 2 M , k ) − g + ( 2 M − ε m M , k ) = ∫ 2 M − ε m M 2 M r q − 1 ( r − k ) d r ≤ C q M q − 1 ( m M ) ( ε m M ) .
Consequently, we get
g + ( 2 M , k ) g + ( 2 M − ε m M , k ) ≤ 1 + C q ε . (36) \frac{g_+(2M,k)}{g_+(2M-\varepsilon mM,k)}\le1+C_q\varepsilon. \tag*{(36)} g + ( 2 M − ε m M , k ) g + ( 2 M , k ) ≤ 1 + C q ε . (36)
uniformly in m m m . Combining these estimates with (34) and ∣ K ϱ \ K ( 1 − σ ) ϱ ∣ ≤ N σ ∣ K ϱ ∣ |K_\varrho\backslash K_{(1-\sigma)\varrho}|\le N\sigma|K_\varrho| ∣ K ϱ \ K ( 1 − σ ) ϱ ∣ ≤ N σ ∣ K ϱ ∣ yields
∣ { u ( t ) > 2 M − ε M } ∩ K ϱ ∣ ∣ K ϱ ∣ ≤ ( 1 + C q ε ) ( 1 − ν / 2 ) + C δ ( 1 + σ − 2 + σ − N − 2 s ) + N σ . (37) \frac{\left|\{u(t)>2M-\varepsilon M\}\cap K_\varrho\right|}{|K_\varrho|}
\le(1+C_q\varepsilon)(1-\nu/2)+C\delta(1+\sigma^{-2}+\sigma^{-N-2s})+N\sigma. \tag*{(37)} ∣ K ϱ ∣ ∣ { u ( t ) > 2 M − εM } ∩ K ϱ ∣ ≤ ( 1 + C q ε ) ( 1 − ν /2 ) + C δ ( 1 + σ − 2 + σ − N − 2 s ) + N σ . (37)
Choose first σ = ν / ( 16 N ) \sigma=\nu/(16N) σ = ν / ( 16 N ) , then ε \varepsilon ε so that C q ε ( 1 − ν / 2 ) ≤ ν / 8 C_q\varepsilon(1-\nu/2)\le\nu/8 C q ε ( 1 − ν /2 ) ≤ ν /8 , and finally δ \delta δ so that C δ ( 1 + σ − 2 + σ − N − 2 s ) ≤ ν / 16 C\delta(1+\sigma^{-2}+\sigma^{-N-2s})\le\nu/16 C δ ( 1 + σ − 2 + σ − N − 2 s ) ≤ ν /16 . The right-hand side of (37) is at most 1 − ν / 4 1-\nu/4 1 − ν /4 . Taking the complement proves (30) . The same estimates hold on every shorter time interval. □
A shrinking lemma
Lemma 5.3. Let u u u be a locally bounded local weak subsolution in Q Q Q . Let θ = M q − 1 \theta=M^{q-1} θ = M q − 1 , σ , ε , δ ∈ ( 0 , 1 / 2 ) \sigma,\varepsilon,\delta\in(0,1/2) σ , ε , δ ∈ ( 0 , 1/2 ) , and 2 ϱ ≤ R 2\varrho\le R 2 ϱ ≤ R . Write I = ( τ , τ + δ θ 2 s ) I=(\tau,\tau+\delta\theta^{2s}) I = ( τ , τ + δ θ 2 s ) and assume K 2 ϱ × I ⊂ Q K_{2\varrho}\times I\subset Q K 2 ϱ × I ⊂ Q . If
∣ { u ( ⋅ , t ) ≤ 2 M − ε M } ∩ K ϱ ∣ ∣ K ϱ ∣ ≥ ν 4 \frac{\left|\{u(\cdot,t)\le2M-\varepsilon M\}\cap K_\varrho\right|}{|K_\varrho|}\ge\frac{\nu}{4} ∣ K ϱ ∣ ∣ { u ( ⋅ , t ) ≤ 2 M − εM } ∩ K ϱ ∣ ≥ 4 ν for a.e. t ∈ I t\in I t ∈ I , then either
Tail ( u + ; Q ) > σ ε 2 M q (38) \operatorname{Tail}(u_+;Q)>\frac{\sigma\varepsilon}{2}M^q \tag*{(38)} Tail ( u + ; Q ) > 2 σ ε M q (38) or
∣ { u ≥ 2 M − σ ε M / 2 } ∩ ( K ϱ × I ) ∣ ∣ K ϱ × I ∣ ≤ C σ ν ( 1 + δ − 1 ) , (39) \frac{\left|\{u\ge2M-\sigma\varepsilon M/2\}\cap(K_\varrho\times I)\right|}{|K_\varrho\times I|}
\le\frac{C\sigma}{\nu}(1+\delta^{-1}), \tag*{(39)} ∣ K ϱ × I ∣ ∣ { u ≥ 2 M − σ εM /2 } ∩ ( K ϱ × I ) ∣ ≤ ν C σ ( 1 + δ − 1 ) , (39) where C C C depends only on data. For supersolutions, replace u u u with − u -u − u .
Proof. Assume that (38) fails. Take k = 2 M − σ ε M k=2M-\sigma\varepsilon M k = 2 M − σ εM , set w ± = ( u − k ) ± w_\pm=(u-k)_\pm w ± = ( u − k ) ± , and choose a spatial cutoff ζ = 1 \zeta=1 ζ = 1 in K ϱ K_\varrho K ϱ , supported in K 3 ϱ / 2 K_{3\varrho/2} K 3 ϱ /2 , with ∣ D ζ ∣ ≤ C / ϱ |D\zeta|\le C/\varrho ∣ D ζ ∣ ≤ C / ϱ . Apply (18) in K 2 ϱ × I K_{2\varrho}\times I K 2 ϱ × I and keep the term involving mixed truncations:
∫ I ∫ K 2 ϱ ∫ K 2 ϱ w − ( y , t ) w + ( x , t ) ζ ( x ) 2 K ( x , y , t ) d y d x d t ≤ q ∫ I ∫ K 2 ϱ g + ( u ( τ ) , k ) ζ ( x ) 2 d x + ∫ I ∫ K 2 ϱ ∫ K 2 ϱ ∣ ζ ( x ) − ζ ( y ) ∣ 2 w + ( x , t ) w + ( y , t ) K ( x , y , t ) d y d x d t + ∫ I ∫ K 2 ϱ ∫ R N \ K 2 ϱ w + ( y , t ) ζ ( x ) 2 w + ( x , t ) K ( x , y , t ) d y d x d t . \begin{aligned}
&\int_I\int_{K_{2\varrho}}\int_{K_{2\varrho}}w_-(y,t)w_+(x,t)\zeta(x)^2K(x,y,t)\,dy\,dx\,dt \\
&\quad\le q\int_I\int_{K_{2\varrho}}g_+(u(\tau),k)\zeta(x)^2\,dx \\
&\qquad+\int_I\int_{K_{2\varrho}}\int_{K_{2\varrho}}|\zeta(x)-\zeta(y)|^2w_+(x,t)w_+(y,t)K(x,y,t)\,dy\,dx\,dt \\
&\qquad+\int_I\int_{K_{2\varrho}}\int_{\mathbb{R}^N\backslash K_{2\varrho}}w_+(y,t)\zeta(x)^2w_+(x,t)K(x,y,t)\,dy\,dx\,dt.
\end{aligned} ∫ I ∫ K 2 ϱ ∫ K 2 ϱ w − ( y , t ) w + ( x , t ) ζ ( x ) 2 K ( x , y , t ) d y d x d t ≤ q ∫ I ∫ K 2 ϱ g + ( u ( τ ) , k ) ζ ( x ) 2 d x + ∫ I ∫ K 2 ϱ ∫ K 2 ϱ ∣ ζ ( x ) − ζ ( y ) ∣ 2 w + ( x , t ) w + ( y , t ) K ( x , y , t ) d y d x d t + ∫ I ∫ K 2 ϱ ∫ R N \ K 2 ϱ w + ( y , t ) ζ ( x ) 2 w + ( x , t ) K ( x , y , t ) d y d x d t .
Observe that w + ≤ σ ε M w_+\le\sigma\varepsilon M w + ≤ σ εM in Q Q Q . By (8) , the first term on the right-hand side is estimated by
q ∫ K 2 ϱ g + ( u ( τ ) , k ) ζ ( x ) 2 d x ≤ C M q − 1 ( σ ε M ) 2 ∣ K ϱ ∣ . q\int_{K_{2\varrho}}g_+(u(\tau),k)\zeta(x)^2\,dx\le CM^{q-1}(\sigma\varepsilon M)^2|K_\varrho|. q ∫ K 2 ϱ g + ( u ( τ ) , k ) ζ ( x ) 2 d x ≤ C M q − 1 ( σ εM ) 2 ∣ K ϱ ∣.
As for the second term, (11) gives
∫ I ∫ K 2 ϱ ∫ K 2 ϱ ∣ ζ ( x ) − ζ ( y ) ∣ 2 w + ( x , t ) w + ( y , t ) K ( x , y , t ) d y d x d t ≤ C ( σ ε M ) 2 ϱ − 2 s ∣ K ϱ × I ∣ . \int_I\int_{K_{2\varrho}}\int_{K_{2\varrho}}|\zeta(x)-\zeta(y)|^2w_+(x,t)w_+(y,t)K(x,y,t)\,dy\,dx\,dt
\le C(\sigma\varepsilon M)^2\varrho^{-2s}|K_\varrho\times I|. ∫ I ∫ K 2 ϱ ∫ K 2 ϱ ∣ ζ ( x ) − ζ ( y ) ∣ 2 w + ( x , t ) w + ( y , t ) K ( x , y , t ) d y d x d t ≤ C ( σ εM ) 2 ϱ − 2 s ∣ K ϱ × I ∣.
The third term can be estimated by computations similar to those in (5.3)
∫ I ∫ K 2 ϱ ∫ R N \ K 2 ϱ w + ( y , t ) ζ ( x ) 2 w + ( x , t ) K ( x , y , t ) d y d x d t ≤ C σ ε M ∣ K ϱ ∣ Tail ( u + ; Q ) + C ( σ ε M ) 2 ϱ − 2 s ∣ K ϱ × I ∣ . \begin{aligned}
&\int_I\int_{K_{2\varrho}}\int_{\mathbb{R}^N\backslash K_{2\varrho}}w_+(y,t)\zeta(x)^2w_+(x,t)K(x,y,t)\,dy\,dx\,dt \\
&\quad\le C\sigma\varepsilon M|K_\varrho|\operatorname{Tail}(u_+;Q)+C(\sigma\varepsilon M)^2\varrho^{-2s}|K_\varrho\times I|.
\end{aligned} ∫ I ∫ K 2 ϱ ∫ R N \ K 2 ϱ w + ( y , t ) ζ ( x ) 2 w + ( x , t ) K ( x , y , t ) d y d x d t ≤ C σ εM ∣ K ϱ ∣ Tail ( u + ; Q ) + C ( σ εM ) 2 ϱ − 2 s ∣ K ϱ × I ∣.
Recalling Tail ( u + ; Q ) ≤ σ ε 2 M q \operatorname{Tail}(u_{+};Q)\leq\frac{\sigma\varepsilon}{2}M^{q} Tail ( u + ; Q ) ≤ 2 σ ε M q , then combining all these estimates, we get
∫ I ∫ K 2 ϱ ∫ K 2 ϱ w − ( y , t ) w + ( x , t ) ζ ( x ) 2 K ( x , y , t ) d y d x d t ≤ C ( σ ε M ) 2 ϱ 2 s ( 1 + δ − 1 ) ∣ K ϱ × I ∣ . (40) \int_I\int_{K_{2\varrho}}\int_{K_{2\varrho}} w_{-}(y,t)w_{+}(x,t)\zeta(x)^2K(x,y,t)\,dy\,dx\,dt
\leq\frac{C(\sigma\varepsilon M)^2}{\varrho^{2s}}(1+\delta^{-1})|K_{\varrho}\times I|. \tag*{(40)} ∫ I ∫ K 2 ϱ ∫ K 2 ϱ w − ( y , t ) w + ( x , t ) ζ ( x ) 2 K ( x , y , t ) d y d x d t ≤ ϱ 2 s C ( σ εM ) 2 ( 1 + δ − 1 ) ∣ K ϱ × I ∣. (40)
For the left-hand side of (40) , if u ( y , t ) ≤ 2 M − ε M u(y,t)\leq2M-\varepsilon M u ( y , t ) ≤ 2 M − εM and u ( x , t ) ≥ 2 M − σ ε M / 2 u(x,t)\geq2M-\sigma\varepsilon M/2 u ( x , t ) ≥ 2 M − σ εM /2 , then
w − ( y , t ) ≥ ( 1 − σ ) ε M , w + ( x , t ) ≥ σ ε M / 2. w_{-}(y,t)\geq(1-\sigma)\varepsilon M,\qquad w_{+}(x,t)\geq\sigma\varepsilon M/2. w − ( y , t ) ≥ ( 1 − σ ) εM , w + ( x , t ) ≥ σ εM /2.
For x , y ∈ K ϱ x,y\in K_{\varrho} x , y ∈ K ϱ , we also have ∣ x − y ∣ ≤ 2 ϱ |x-y|\leq2\varrho ∣ x − y ∣ ≤ 2 ϱ . Using the lower bound on the measure of { u ( ⋅ , t ) ≤ 2 M − ε M } ∩ K ϱ \{u(\cdot,t)\leq2M-\varepsilon M\}\cap K_{\varrho} { u ( ⋅ , t ) ≤ 2 M − εM } ∩ K ϱ , we obtain
∫ I ∫ K 2 ϱ ∫ K 2 ϱ w − ( y , t ) w + ( x , t ) ζ ( x ) 2 K ( x , y , t ) d y d x d t ≥ C ν σ ( 1 − σ ) ( ε M ) 2 ϱ 2 s ∫ I ∣ { u ( ⋅ , t ) ≥ 2 M − σ ε M / 2 } ∩ K ϱ ∣ d t . (41) \int_I\int_{K_{2\varrho}}\int_{K_{2\varrho}}w_{-}(y,t)w_{+}(x,t)\zeta(x)^2K(x,y,t)\,dy\,dx\,dt
\geq\frac{C\nu\sigma(1-\sigma)(\varepsilon M)^2}{\varrho^{2s}}\int_I\left|\{u(\cdot,t)\geq2M-\sigma\varepsilon M/2\}\cap K_{\varrho}\right|\,dt. \tag*{(41)} ∫ I ∫ K 2 ϱ ∫ K 2 ϱ w − ( y , t ) w + ( x , t ) ζ ( x ) 2 K ( x , y , t ) d y d x d t ≥ ϱ 2 s C ν σ ( 1 − σ ) ( εM ) 2 ∫ I ∣ { u ( ⋅ , t ) ≥ 2 M − σ εM /2 } ∩ K ϱ ∣ d t . (41)
Combine (40) and (41) , and use 1 − σ ≥ 1 / 2 1-\sigma\geq1/2 1 − σ ≥ 1/2 . This proves (39) .
A De Giorgi lemma near the upper bound
Lemma 5.4. Let u u u be a locally bounded local weak subsolution in Q Q Q . Let δ , ξ ∈ ( 0 , 1 ) \delta,\xi\in(0,1) δ , ξ ∈ ( 0 , 1 ) and θ = M q − 1 \theta=M^{q-1} θ = M q − 1 . Assume 2 ϱ ≤ R 2\varrho\leq R 2 ϱ ≤ R and K ϱ × ( τ , τ + δ θ ϱ 2 s ) ⊂ Q K_{\varrho}\times(\tau,\tau+\delta\theta\varrho^{2s})\subset Q K ϱ × ( τ , τ + δ θ ϱ 2 s ) ⊂ Q . After enlarging γ ∗ ∗ \gamma_{**} γ ∗∗ in Lemma 4.1 if necessary, there is ν 0 = ν 0 ( δ , d a t a ) ∈ ( 0 , 1 ) \nu_{0}=\nu_{0}(\delta,\mathrm{data})\in(0,1) ν 0 = ν 0 ( δ , data ) ∈ ( 0 , 1 ) , independent of ξ \xi ξ , such that
∣ { u ≥ 2 M − ξ M } ∩ ( K ϱ × ( τ , τ + δ θ ϱ 2 s ) ) ∣ ≤ ν 0 ∣ K ϱ ∣ δ θ ϱ 2 s , (42) \left|\{u\geq2M-\xi M\}\cap(K_{\varrho}\times(\tau,\tau+\delta\theta\varrho^{2s}))\right|\leq\nu_{0}|K_{\varrho}|\delta\theta\varrho^{2s}, \tag*{(42)} { u ≥ 2 M − ξ M } ∩ ( K ϱ × ( τ , τ + δ θ ϱ 2 s )) ≤ ν 0 ∣ K ϱ ∣ δ θ ϱ 2 s , (42) γ ∗ ∗ Tail ( u + ; Q ) ≤ ξ M q (43) \gamma_{**}\operatorname{Tail}(u_{+};Q)\leq\xi M^{q} \tag*{(43)} γ ∗∗ Tail ( u + ; Q ) ≤ ξ M q (43) imply
u ≤ 2 M − ξ M 4 a.e. in K ϱ / 2 × ( τ + 3 δ θ ϱ 2 s / 4 , τ + δ θ ϱ 2 s ) . u\leq2M-\frac{\xi M}{4}\quad\text{a.e. in }K_{\varrho/2}\times(\tau+3\delta\theta\varrho^{2s}/4,\tau+\delta\theta\varrho^{2s}). u ≤ 2 M − 4 ξ M a.e. in K ϱ /2 × ( τ + 3 δ θ ϱ 2 s /4 , τ + δ θ ϱ 2 s ) . The analogous assertion holds for − u -u − u .
Proof. Translate the final time to zero and keep the radii ϱ n , ϱ ^ n , ϱ ~ n , ϱ ‾ n \varrho_n,\widehat{\varrho}_n,\widetilde{\varrho}_n,\overline{\varrho}_n ϱ n , ϱ n , ϱ n , ϱ n and the balls K n , K ^ n , K ~ n , K ‾ n K_n,\widehat K_n,\widetilde K_n,\overline K_n K n , K n , K n , K n used in Lemma 4.1. To obtain exactly the final quarter of the time interval, define the heights independently of the radii:
d n = δ θ ϱ 2 s ( 1 4 + 3 4 ⋅ 2 n ) , d_n=\delta\theta\varrho^{2s}\left(\frac{1}{4}+\frac{3}{4\cdot2^n}\right), d n = δ θ ϱ 2 s ( 4 1 + 4 ⋅ 2 n 3 ) , d ^ n = 3 d n + d n + 1 4 , d ~ n = d n + d n + 1 2 , d ‾ n = d n + 3 d n + 1 4 . \widehat d_n=\frac{3d_n+d_{n+1}}{4},\qquad\widetilde d_n=\frac{d_n+d_{n+1}}{2},\qquad\overline d_n=\frac{d_n+3d_{n+1}}{4}. d n = 4 3 d n + d n + 1 , d n = 2 d n + d n + 1 , d n = 4 d n + 3 d n + 1 . Set
{ Q n = K n × ( − d n , 0 ) , Q ~ n = K ~ n × ( − d ~ n , 0 ) , Q ^ n = K ^ n × ( − d ^ n , 0 ) , Q ‾ n = K ‾ n × ( − d ‾ n , 0 ) . \left\{
\begin{aligned}
Q_n&=K_n\times(-d_n,0),&\qquad\widetilde Q_n&=\widetilde K_n\times(-\widetilde d_n,0),\\
\widehat Q_n&=\widehat K_n\times(-\widehat d_n,0),& \qquad\overline Q_n&=\overline K_n\times(-\overline d_n,0).
\end{aligned}
\right. { Q n Q n = K n × ( − d n , 0 ) , = K n × ( − d n , 0 ) , Q n Q n = K n × ( − d n , 0 ) , = K n × ( − d n , 0 ) . Choose ζ \zeta ζ supported in Q ^ n \widehat Q_n Q n , equal to one in Q ~ n \widetilde Q_n Q n , and satisfying
∣ D ζ ∣ ≤ C 2 n / ϱ , ∣ ∂ t ζ ∣ ≤ C 2 n / δ θ ϱ 2 s . |D\zeta|\leq C2^n/\varrho,\qquad|\partial_t\zeta|\leq C2^n/\delta\theta\varrho^{2s}. ∣ D ζ ∣ ≤ C 2 n / ϱ , ∣ ∂ t ζ ∣ ≤ C 2 n / δ θ ϱ 2 s .
Support here concerns the lateral boundary and the initial time; the cutoff need not vanish at the final time.
Take
k n = 2 M − ξ M 2 − ξ M 2 n + 1 , k n ( t ) = k n + ℓ ( t ) , k_n = 2M - \frac{\xi M}{2} - \frac{\xi M}{2^{n+1}}, \qquad k_n(t) = k_n + \ell(t), k n = 2 M − 2 ξ M − 2 n + 1 ξ M , k n ( t ) = k n + ℓ ( t ) ,
where
ℓ ( t ) = γ ~ M 1 − q ∫ T 1 t ∫ R N ∖ K R u + ( y , τ ) ∣ y ∣ N + 2 s d y d τ . \ell(t) = \widetilde{\gamma}M^{1-q}\int_{T_1}^{t}\int_{\mathbb{R}^N\setminus K_R}\frac{u_+(y,\tau)}{|y|^{N+2s}}\,dy\,d\tau. ℓ ( t ) = γ M 1 − q ∫ T 1 t ∫ R N ∖ K R ∣ y ∣ N + 2 s u + ( y , τ ) d y d τ .
Choose γ ∗ ∗ ≥ 4 γ ~ \gamma_{**} \ge4\widetilde{\gamma} γ ∗∗ ≥ 4 γ . Assumption (43) then gives 0 ≤ ℓ ≤ ξ M / 4 0 \le\ell\le\xi M/4 0 ≤ ℓ ≤ ξ M /4 . In particular k n ( t ) ∈ [ M , 2 M ] k_n(t) \in[M,2M] k n ( t ) ∈ [ M , 2 M ] . Let w + = ( u − k n ( t ) ) + w_+ = (u-k_n(t))_+ w + = ( u − k n ( t ) ) + and A n = { u > k n ( t ) } ∩ Q n A_n = \{u>k_n(t)\}\cap Q_n A n = { u > k n ( t )} ∩ Q n . Then w + ≤ ξ M w_+ \le\xi M w + ≤ ξ M in Q n Q_n Q n and
c q M q − 1 w + 2 ≤ g + ( u , k n ( t ) ) ≤ C q M q − 1 w + 2 , c_qM^{q-1}w_+^2 \le g_+(u,k_n(t)) \le C_qM^{q-1}w_+^2, c q M q − 1 w + 2 ≤ g + ( u , k n ( t )) ≤ C q M q − 1 w + 2 ,
( u q − k n ( t ) q ) + ≥ c q M q − 1 w + . (u^q-k_n(t)^q)_+ \ge c_qM^{q-1}w_+. ( u q − k n ( t ) q ) + ≥ c q M q − 1 w + .
We spell out the energy estimates to keep track of the dependence on δ \delta δ and ξ \xi ξ . As in Lemma 4.1, B > 1 B>1 B > 1 is a fixed constant depending only on the data, chosen to dominate the powers of 2 from the cutoff bounds and the level differences. In particular, B B B is independent of n n n , δ \delta δ , and ξ \xi ξ . Then, the time-cutoff term on the left-hand side of (18) is bounded by
q ∬ Q n g + ( u , k n ) ∂ t ( ζ 2 ) d x d t ≤ C B n M q − 1 ( ξ M ) 2 ( δ θ 2 s ) − 1 ∣ A n ∣ . q\iint_{Q_n} g_+(u,k_n)\partial_t(\zeta^2)\,dx\,dt \le CB^nM^{q-1}(\xi M)^2(\delta\theta^{2s})^{-1}|A_n|. q ∬ Q n g + ( u , k n ) ∂ t ( ζ 2 ) d x d t ≤ C B n M q − 1 ( ξ M ) 2 ( δ θ 2 s ) − 1 ∣ A n ∣.
The spatial cutoff term is estimated by
∫ − d n 0 ∫ K n ∫ K n ∣ ζ ( x , t ) − ζ ( y , t ) ∣ 2 w + ( x , t ) w + ( y , t ) K ( x , y , t ) d y d x d t ≤ C B n ( ξ M ) 2 ϱ 2 s ∣ A n ∣ . \int_{-d_n}^{0}\int_{K_n}\int_{K_n}| \zeta(x,t)-\zeta(y,t)|^2w_+(x,t)w_+(y,t)K(x,y,t)\,dy\,dx\,dt \le CB^n\frac{(\xi M)^2}{\varrho^{2s}}|A_n|. ∫ − d n 0 ∫ K n ∫ K n ∣ ζ ( x , t ) − ζ ( y , t ) ∣ 2 w + ( x , t ) w + ( y , t ) K ( x , y , t ) d y d x d t ≤ C B n ϱ 2 s ( ξ M ) 2 ∣ A n ∣.
For the exterior interaction, w + ( y , t ) ≤ ξ M w_+(y,t)\le\xi M w + ( y , t ) ≤ ξ M in K R ∖ K n K_R\setminus K_n K R ∖ K n and w + ( y , t ) ≤ u + ( y , t ) w_+(y,t)\le u_+(y,t) w + ( y , t ) ≤ u + ( y , t ) outside K R K_R K R . Thus it is bounded by
∫ − d n 0 ∫ K n ∫ R N ∖ K n w + ( y , t ) ζ ( x , t ) 2 w + ( x , t ) K ( x , y , t ) d y d x d t ≤ C B n ( ξ M ) 2 ϱ 2 s ∣ A n ∣ + C ∫ − d n 0 ∫ K n ζ 2 w + ( x , t ) ∫ R N ∖ K R u + ( y , t ) ∣ y ∣ N + 2 s d y d x d t . \begin{aligned}
\int_{-d_n}^{0}\int_{K_n}\int_{\mathbb{R}^N\setminus K_n}w_+(y,t)\zeta(x,t)^2w_+(x,t)K(x,y,t)\,dy\,dx\,dt \\
&\le\frac{CB^n(\xi M)^2}{\varrho^{2s}}|A_n|+C\int_{-d_n}^{0}\int_{K_n}\zeta^2w_+(x,t)\int_{\mathbb{R}^N\setminus K_R}\frac{u_+(y,t)}{|y|^{N+2s}}\,dy\,dx\,dt.
\end{aligned} ∫ − d n 0 ∫ K n ∫ R N ∖ K n w + ( y , t ) ζ ( x , t ) 2 w + ( x , t ) K ( x , y , t ) d y d x d t ≤ ϱ 2 s C B n ( ξ M ) 2 ∣ A n ∣ + C ∫ − d n 0 ∫ K n ζ 2 w + ( x , t ) ∫ R N ∖ K R ∣ y ∣ N + 2 s u + ( y , t ) d y d x d t .
The last term is absorbed by the term − c q M q − 1 ∫ ℓ ′ ζ 2 w + -c_qM^{q-1}\int\ell'\zeta^2w_+ − c q M q − 1 ∫ ℓ ′ ζ 2 w + , derived by the second term on the left-hand side of (18) , after choosing γ ~ = C c q \widetilde{\gamma}=\frac{C}{c_q} γ = c q C . Consequently,
M q − 1 ess sup − d n < t < 0 ∫ K n w + 2 ζ 2 d x + ∫ − d n 0 ∫ K ~ n ∫ K ~ n ∣ w + ( x , t ) − w + ( y , t ) ∣ 2 ∣ x − y ∣ N + 2 s d y d x d t ≤ C B n ( ξ M ) 2 ϱ 2 s ( 1 + δ − 1 ) ∣ A n ∣ . (44) \begin{aligned}
M^{q-1}\mathop{\operatorname{ess\,sup}}_{-d_n<t<0}\int_{K_n}w_+^2\zeta^2\,dx+\int_{-d_n}^{0}\int_{\widetilde K_n}\int_{\widetilde K_n}\frac{|w_+(x,t)-w_+(y,t)|^2}{|x-y|^{N+2s}}\,dy\,dx\,dt \\
&\le\frac{CB^n(\xi M)^2}{\varrho^{2s}}(1+\delta^{-1})|A_n|. \tag*{(44)}
\end{aligned} M q − 1 ess sup − d n < t < 0 ∫ K n w + 2 ζ 2 d x + ∫ − d n 0 ∫ K n ∫ K n ∣ x − y ∣ N + 2 s ∣ w + ( x , t ) − w + ( y , t ) ∣ 2 d y d x d t ≤ ϱ 2 s C B n ( ξ M ) 2 ( 1 + δ − 1 ) ∣ A n ∣. (44)
Apply Lemma 2.2 with the second cutoff supported in Q ‾ n \overline{Q}_n Q n and equal to one in Q ‾ n \overline{Q}_n Q n , just as in Lemma 4.1. By arguments similar to those in Step 3 of Lemma 4.1, we get
∬ Q ‾ n w + m ^ d x d t ≤ C ( ϱ 2 s ∫ − d ‾ n 0 ∫ K ~ n ∫ K ~ n ∣ w + ( x , t ) − w + ( y , t ) ∣ 2 ∣ x − y ∣ N + 2 s d y d x d t + B n ∫ − d ‾ n 0 ∫ K ~ n w + 2 ( y , t ) d y d t ) × ( e s s s u p ( − d ‾ n , 0 ) ∫ K n ∣ w + ( x , t ) ∣ 2 ζ ( x , t ) 2 d x ) 2 s N ≤ C B n ( ξ M ) m ~ M ( 1 − q ) 2 s N ϱ − 2 s ( 1 + 2 s N ) ( 1 + δ − 1 ) 1 + 2 s N ∣ A n ∣ 1 + 2 s N . \begin{aligned}
\iint_{\overline{Q}_n} w_+^{\hat m}\,\mathrm{d}x\,\mathrm{d}t
&\le C\left(\varrho^{2s}\int_{-\underline d_n}^{0}\int_{\widetilde K_n}\int_{\widetilde K_n}\frac{\lvert w_+(x,t)-w_+(y,t)\rvert^2}{\lvert x-y\rvert^{N+2s}}\,\mathrm{d}y\,\mathrm{d}x\,\mathrm{d}t \right.\\
&\qquad\left.{}+B^n\int_{-\underline d_n}^{0}\int_{\widetilde K_n}w_+^2(y,t)\,\mathrm{d}y\,\mathrm{d}t\right)\times\left(\mathop{\mathrm{ess\,sup}}_{(-\underline d_n,0)}\int_{K_n}\lvert w_+(x,t)\rvert^2\zeta(x,t)^2\,\mathrm{d}x\right)^{\frac{2s}{N}}\\
&\le CB^n(\xi M)^{\tilde m}M^{(1-q)\frac{2s}{N}}\varrho^{-2s(1+\frac{2s}{N})}(1+\delta^{-1})^{1+\frac{2s}{N}}\lvert A_n\rvert^{1+\frac{2s}{N}}.
\end{aligned} ∬ Q n w + m ^ d x d t ≤ C ( ϱ 2 s ∫ − d n 0 ∫ K n ∫ K n ∣ x − y ∣ N + 2 s ∣ w + ( x , t ) − w + ( y , t ) ∣ 2 d y d x d t + B n ∫ − d n 0 ∫ K n w + 2 ( y , t ) d y d t ) × ( ess sup ( − d n , 0 ) ∫ K n ∣ w + ( x , t ) ∣ 2 ζ ( x , t ) 2 d x ) N 2 s ≤ C B n ( ξ M ) m ~ M ( 1 − q ) N 2 s ϱ − 2 s ( 1 + N 2 s ) ( 1 + δ − 1 ) 1 + N 2 s ∣ A n ∣ 1 + N 2 s .
The last inequality is derived by (44) .
Since w + ≥ k n + 1 − k n = ξ M / 2 n + 2 w_+ \ge k_{n+1}-k_n=\xi M/2^{n+2} w + ≥ k n + 1 − k n = ξ M / 2 n + 2 on A n + 1 A_{n+1} A n + 1 , one obtains
∣ A n + 1 ∣ ≤ C B n M ( 1 − q ) 2 s N ϱ − 2 s ( 1 + 2 s N ) ( 1 + δ − 1 ) 1 + 2 s N ∣ A n ∣ 1 + 2 s N . \lvert A_{n+1}\rvert\le CB^nM^{(1-q)\frac{2s}{N}}\varrho^{-2s(1+\frac{2s}{N})}(1+\delta^{-1})^{1+\frac{2s}{N}}\lvert A_n\rvert^{1+\frac{2s}{N}}. ∣ A n + 1 ∣ ≤ C B n M ( 1 − q ) N 2 s ϱ − 2 s ( 1 + N 2 s ) ( 1 + δ − 1 ) 1 + N 2 s ∣ A n ∣ 1 + N 2 s .
Normalizing by ∣ Q n ∣ = δ M q − 1 ϱ N + 2 s \lvert Q_n\rvert=\delta M^{q-1}\varrho^{N+2s} ∣ Q n ∣ = δ M q − 1 ϱ N + 2 s yields
Y n + 1 ≤ C B n δ 2 s N ( 1 + δ − 1 ) 1 + 2 s N Y n 1 + 2 s N , Y n = ∣ A n ∣ ∣ Q n ∣ . Y_{n+1}\le CB^n\delta^{\frac{2s}{N}}(1+\delta^{-1})^{1+\frac{2s}{N}}Y_n^{1+\frac{2s}{N}},\qquad Y_n=\frac{\lvert A_n\rvert}{\lvert Q_n\rvert}. Y n + 1 ≤ C B n δ N 2 s ( 1 + δ − 1 ) 1 + N 2 s Y n 1 + N 2 s , Y n = ∣ Q n ∣ ∣ A n ∣ .
It suffices to set
ν 0 ( δ ) = min { 1 2 , [ C δ 2 s N ( 1 + δ − 1 ) 1 + 2 s N ] − N 2 s B − N 2 4 s 2 } . (45) \nu_0(\delta)=\min\left\{\frac{1}{2},\left[C\delta^{\frac{2s}{N}}(1+\delta^{-1})^{1+\frac{2s}{N}}\right]^{-\frac{N}{2s}}B^{-\frac{N^2}{4s^2}}\right\}. \tag*{(45)} ν 0 ( δ ) = min { 2 1 , [ C δ N 2 s ( 1 + δ − 1 ) 1 + N 2 s ] − 2 s N B − 4 s 2 N 2 } . (45)
Since k 0 = 2 M − ξ M k_0=2M-\xi M k 0 = 2 M − ξ M and ℓ ≥ 0 \ell\ge0 ℓ ≥ 0 , (42) ensures Y 0 ≤ ν 0 ( δ ) Y_0\le\nu_0(\delta) Y 0 ≤ ν 0 ( δ ) . The iteration gives u ≤ 2 M − ξ M / 2 + ℓ ( t ) ≤ 2 M − ξ M / 4 u\le2M-\xi M/2+\ell(t)\le2M-\xi M/4 u ≤ 2 M − ξ M /2 + ℓ ( t ) ≤ 2 M − ξ M /4 in the limiting cylinder, as required. □ \square □
Reaching the final time by finitely many steps
Lemma 5.5. Let u u u be a locally bounded local weak subsolution in Q Q Q . Let ν \nu ν be the constant in Lemma 4.1, and assume K 2 ϱ × ( − θ ϱ 2 s , 0 ) ⊂ Q K_{2\varrho}\times(-\theta\varrho^{2s},0)\subset Q K 2 ϱ × ( − θ ϱ 2 s , 0 ) ⊂ Q . There exist η ∈ ( 0 , 1 / 2 ) \eta\in(0,1/2) η ∈ ( 0 , 1/2 ) and γ ∈ ( 0 , 1 / 4 ) \gamma\in(0,1/4) γ ∈ ( 0 , 1/4 ) , depending only on data, such that
∣ { u ≤ M } ∩ ( K ϱ × ( − θ ϱ 2 s , 0 ) ) ∣ > ν ∣ K ϱ ∣ θ ϱ 2 s , Tail ( u + ; Q ) ≤ γ M q \lvert\{u\le M\}\cap(K_\varrho\times(-\theta\varrho^{2s},0))\rvert>\nu\lvert K_\varrho\rvert\theta\varrho^{2s},\qquad\operatorname{Tail}(u_+;Q)\le\gamma M^q ∣{ u ≤ M } ∩ ( K ϱ × ( − θ ϱ 2 s , 0 ))∣ > ν ∣ K ϱ ∣ θ ϱ 2 s , Tail ( u + ; Q ) ≤ γ M q
imply
u ≤ 2 M ( 1 − η ) a.e. in K ϱ / 2 × ( − 7 16 θ ϱ 2 s , 0 ) . (46) u\le2M(1-\eta)\quad\text{a.e. in }K_{\varrho/2}\times\left(-\frac{7}{16}\theta\varrho^{2s},0\right). \tag*{(46)} u ≤ 2 M ( 1 − η ) a.e. in K ϱ /2 × ( − 16 7 θ ϱ 2 s , 0 ) . (46)
For a supersolution, apply the statement to − u -u − u .
Proof. We combine the time-slice, propagation, shrinking, and De Giorgi estimates in a finite sequence. The reference amplitude M M M and the scale θ = M q − 1 \theta=M^{q-1} θ = M q − 1 remain fixed throughout the proof.
Step 1: the first interval. Lemma 5.1 gives an admissible τ ∈ ( − θ ϱ 2 s , − ν θ ϱ 2 s / 2 ) \tau\in(-\theta\varrho^{2s},-\nu\theta\varrho^{2s}/2) τ ∈ ( − θ ϱ 2 s , − ν θ ϱ 2 s /2 ) with ∣ { u ( τ ) ≤ M } ∩ K ϱ ∣ > ν ∣ K ϱ ∣ / 2 \lvert\{u(\tau)\le M\}\cap K_\varrho\rvert>\nu\lvert K_\varrho\rvert/2 ∣{ u ( τ ) ≤ M } ∩ K ϱ ∣ > ν ∣ K ϱ ∣ /2 . Use Lemma 5.2 with m = 1 m=1 m = 1 and δ , ε \delta,\varepsilon δ , ε , depending on ν \nu ν and data. Note that δ ≤ ν / 16 \delta\le\nu/16 δ ≤ ν /16 . Let ε \varepsilon ε be its propagation parameter.
By Lemma 5.3, we have
∣ { u ≥ 2 M − σ ε 2 M } ∩ K ϱ × ( τ , τ + δ θ ϱ 2 s ) ∣ ≤ C σ ( 1 + 1 δ ) ν ∣ K ϱ × ( τ , τ + δ θ ϱ 2 s ) ∣ \left\lvert\left\{u\ge2M-\frac{\sigma\varepsilon}{2}M\right\}\cap K_\varrho\times(\tau,\tau+\delta\theta\varrho^{2s})\right\rvert\le\frac{C\sigma(1+\frac{1}{\delta})}{\nu}\left\lvert K_\varrho\times(\tau,\tau+\delta\theta\varrho^{2s})\right\rvert { u ≥ 2 M − 2 σ ε M } ∩ K ϱ × ( τ , τ + δ θ ϱ 2 s ) ≤ ν C σ ( 1 + δ 1 ) K ϱ × ( τ , τ + δ θ ϱ 2 s )
where C C C is the structural constant in that lemma. Choose σ ∈ ( 0 , 1 / 2 ) \sigma\in(0,1/2) σ ∈ ( 0 , 1/2 ) so small that
C σ ( 1 + δ − 1 ) ν − 1 ≤ ν 0 ( δ ) . C\sigma(1+\delta^{-1})\nu^{-1} \le\nu_0(\delta). C σ ( 1 + δ − 1 ) ν − 1 ≤ ν 0 ( δ ) .
This is a noncircular choice because (45) does not depend on ξ = σ ε / 2 \xi=\sigma\varepsilon/2 ξ = σ ε /2 . Lemma 5.4 gives
u ≤ 2 M − m 1 M in K ϱ / 2 × ( a 0 , b 0 ) , m 1 = σ ε 8 . (47) u \le2M-m_1M \quad\text{in } K_{\varrho/2}\times(a_0,b_0), \qquad m_1=\frac{\sigma\varepsilon}{8}. \tag*{(47)} u ≤ 2 M − m 1 M in K ϱ /2 × ( a 0 , b 0 ) , m 1 = 8 σ ε . (47)
Thus m 1 M m_1M m 1 M is the first gap below the upper bound 2 M 2M 2 M . Here
a 0 = τ + 3 4 δ θ 2 s , b 0 = τ + δ θ 2 s . a_0=\tau+\frac{3}{4}\delta\theta^{2s}, \qquad b_0=\tau+\delta\theta^{2s}. a 0 = τ + 4 3 δ θ 2 s , b 0 = τ + δ θ 2 s .
The required tail conditions are all fulfilled if
Tail ( u + ; Q ) ≤ min { δ , σ ε / ( 2 γ ∗ ∗ ) } M q . \operatorname{Tail}(u_+;Q)\le\min\{\delta,\sigma\varepsilon/(2\gamma_{**})\}M^q. Tail ( u + ; Q ) ≤ min { δ , σ ε / ( 2 γ ∗∗ )} M q .
Here γ ∗ ∗ ≥ 1 \gamma_{**}\ge1 γ ∗∗ ≥ 1 also covers the shrinking condition. Moreover,
a 0 ≤ − ν 2 θ 2 s + 3 ν 64 θ 2 s = − 29 ν 64 θ 2 s < − 7 ν 16 θ 2 s . a_0\le-\frac{\nu}{2}\theta^{2s}+\frac{3\nu}{64}\theta^{2s}=-\frac{29\nu}{64}\theta^{2s}<-\frac{7\nu}{16}\theta^{2s}. a 0 ≤ − 2 ν θ 2 s + 64 3 ν θ 2 s = − 64 29 ν θ 2 s < − 16 7 ν θ 2 s .
If b 0 ≥ 0 b_0\ge0 b 0 ≥ 0 , the proof is finished; otherwise, we proceed the proof.
Step 2: a step with uniform parameters. Fix ν ˉ = 2 − N \bar{\nu}=2^{-N} ν ˉ = 2 − N . Let m ∈ ( 0 , 1 ] m\in(0,1] m ∈ ( 0 , 1 ] and suppose that, at an admissible time s s s , u ( ⋅ , s ) ≤ 2 M − m M u(\cdot,s)\le2M-mM u ( ⋅ , s ) ≤ 2 M − m M in K ϱ / 2 K_{\varrho/2} K ϱ /2 . The number m M mM m M is the current gap below 2 M 2M 2 M ; on the first interval we have m = m 1 m=m_1 m = m 1 from (47) . This bound gives
∣ { u ( x , s ) ≤ 2 M − m M } ∩ K ϱ ∣ > ∣ { u ( x , s ) ≤ 2 M − m M } ∩ K 1 2 ϱ ∣ ≥ 1 2 ∣ K ϱ 2 ∣ = ν ˉ 2 ∣ K ϱ ∣ . \left|\{u(x,s)\le2M-mM\}\cap K_{\varrho}\right|>\left|\{u(x,s)\le2M-mM\}\cap K_{\frac{1}{2}\varrho}\right|\ge\frac{1}{2}|K_{\frac{\varrho}{2}}|=\frac{\bar{\nu}}{2}|K_{\varrho}|. ∣ { u ( x , s ) ≤ 2 M − m M } ∩ K ϱ ∣ > { u ( x , s ) ≤ 2 M − m M } ∩ K 2 1 ϱ ≥ 2 1 ∣ K 2 ϱ ∣ = 2 ν ˉ ∣ K ϱ ∣.
Apply Lemma 5.2 with ν ˉ \bar{\nu} ν ˉ in place of ν \nu ν . Write its constants as ε ˉ , δ ˉ \bar{\varepsilon},\bar{\delta} ε ˉ , δ ˉ and decrease δ ˉ \bar{\delta} δ ˉ so that δ ˉ ≤ δ / 4 \bar{\delta}\le\delta/4 δ ˉ ≤ δ /4 . These constants are independent of m m m .
For later use we permit any duration d θ 2 s d\theta^{2s} d θ 2 s with d ∈ [ δ ˉ / 2 , δ ˉ ] d\in[\bar{\delta}/2,\bar{\delta}] d ∈ [ δ ˉ /2 , δ ˉ ] . The expression in (45) has a positive minimum on this interval. Choose σ ˉ ∈ ( 0 , 1 / 2 ) \bar{\sigma}\in(0,1/2) σ ˉ ∈ ( 0 , 1/2 ) so that
C σ ˉ ( 1 + 2 / δ ˉ ) ν ˉ ≤ inf δ ˉ / 2 ≤ d ≤ δ ˉ ν 0 ( d ) . (48) \frac{C\bar{\sigma}(1+2/\bar{\delta})}{\bar{\nu}}\le\inf_{\bar{\delta}/2\le d\le\bar{\delta}}\nu_0(d). \tag*{(48)} ν ˉ C σ ˉ ( 1 + 2/ δ ˉ ) ≤ δ ˉ /2 ≤ d ≤ δ ˉ inf ν 0 ( d ) . (48)
Starting with depth m m m , the three lemmas now give depth a m am am on the final quarter of the new interval, where
a = σ ˉ ε ˉ 8 ∈ ( 0 , 1 ) . a=\frac{\bar{\sigma}\bar{\varepsilon}}{8}\in(0,1). a = 8 σ ˉ ε ˉ ∈ ( 0 , 1 ) .
It is sufficient that
Tail ( u + ; Q ) ≤ c ∗ m M q , c ∗ = min { δ ˉ / 2 , σ ˉ ε ˉ / ( 2 γ ∗ ∗ ) } . \operatorname{Tail}(u_+;Q)\le c_*mM^q,\qquad c_*=\min\{\bar{\delta}/2,\bar{\sigma}\bar{\varepsilon}/(2\gamma_{**})\}. Tail ( u + ; Q ) ≤ c ∗ m M q , c ∗ = min { δ ˉ /2 , σ ˉ ε ˉ / ( 2 γ ∗∗ )} .
Indeed, the propagated depth is ε ˉ m \bar{\varepsilon}m ε ˉ m , and the level parameter for Lemma 5.4 is ξ = σ ˉ ε ˉ m / 2 \xi=\bar{\sigma}\bar{\varepsilon}m/2 ξ = σ ˉ ε ˉ m /2 .
Step 3: covering the remaining time interval. Set m ( 0 ) = m 1 = σ ε / 8 m^{(0)}=m_1=\sigma\varepsilon/8 m ( 0 ) = m 1 = σ ε /8 , as in (47) . Suppose that u ≤ 2 M − m ( j ) M u\le2M-m^{(j)}M u ≤ 2 M − m ( j ) M on K ϱ / 2 × ( a 0 , b j ) K_{\varrho/2}\times(a_0,b_j) K ϱ /2 × ( a 0 , b j ) , where b j < 0 b_j<0 b j < 0 . Put h = δ ˉ θ 2 s h=\bar{\delta}\theta^{2s} h = δ ˉ θ 2 s and choose an admissible starting time
s j ∈ ( b j − 3 h / 4 , b j − 11 h / 16 ) . s_j\in(b_j-3h/4,b_j-11h/16). s j ∈ ( b j − 3 h /4 , b j − 11 h /16 ) .
This interval lies in ( a 0 , b j ) (a_0,b_j) ( a 0 , b j ) : initially b 0 − a 0 = δ θ 2 s / 4 ≥ h b_0-a_0=\delta\theta^{2s}/4\ge h b 0 − a 0 = δ θ 2 s /4 ≥ h , and the covered interval only grows. First consider duration d j θ 2 s = 4 3 ( b j − s j ) d_j\theta^{2s}=\frac{4}{3}(b_j-s_j) d j θ 2 s = 3 4 ( b j − s j ) , which lies in [ 11 h / 12 , h ] [11h/12,h] [ 11 h /12 , h ] . Then its final quarter starts exactly at b j b_j b j and ends at
b j + 1 = b j + 1 3 ( b j − s j ) ≥ b j + 11 h 48 . b_{j+1}=b_j+\frac{1}{3}(b_j-s_j)\ge b_j+\frac{11h}{48}. b j + 1 = b j + 3 1 ( b j − s j ) ≥ b j + 48 11 h .
If this end would exceed zero, the proof is finished. All these durations belong to [ δ / 2 , δ ] θ 2 s [\delta/2,\delta]\theta^{2s} [ δ /2 , δ ] θ 2 s , as required in (48) .
At each step set m ( j + 1 ) = a m ( j ) m^{(j+1)}=a m^{(j)} m ( j + 1 ) = a m ( j ) . The previous estimate also holds with this smaller depth, so the union remains bounded by 2 M − m ( j + 1 ) M 2M-m^{(j+1)}M 2 M − m ( j + 1 ) M . Because b 0 > − θ ϱ 2 s b_0>-\theta\varrho^{2s} b 0 > − θ ϱ 2 s , the construction reaches zero after at most
J = ⌈ 48 11 δ ⌉ + 1 J=\left\lceil\frac{48}{11\delta}\right\rceil+1 J = ⌈ 11 δ 48 ⌉ + 1
additional steps. Define m min = m 1 a J > 0 m_{\min}=m_1a^J>0 m m i n = m 1 a J > 0 . All the tail conditions are simultaneously satisfied if
γ ≤ min { 1 8 , δ , σ ε 2 γ ∗ ∗ , c ∗ m min } . \gamma\le\min\left\{\frac18,\delta,\frac{\sigma\varepsilon}{2\gamma_{**}},c_*m_{\min}\right\}. γ ≤ min { 8 1 , δ , 2 γ ∗∗ σ ε , c ∗ m m i n } .
Finally take η = m min / 2 \eta=m_{\min}/2 η = m m i n /2 . The estimate holds on K ϱ / 2 × ( a 0 , 0 ) K_{\varrho/2}\times(a_0,0) K ϱ /2 × ( a 0 , 0 ) , which contains the cylinder in (46) . □
Proof of Theorem 1.1
We first carry out the intrinsic iteration near zero. If the solution separates from zero at a finite step, we normalize it and apply the oscillation argument of [22 ] .
Translate ( x o , t o ) (x_o,t_o) ( x o , t o ) to ( 0 , 0 ) (0,0) ( 0 , 0 ) and fix Q = K ϱ × ( T 1 , T 2 ) ⊂ E T Q=K_\varrho\times(T_1,T_2)\subset E_T Q = K ϱ × ( T 1 , T 2 ) ⊂ E T containing a neighbourhood of that point. Choose M 0 > 0 M_0>0 M 0 > 0 so that
2 M 0 ≥ ess sup Q ∣ u ∣ + 2 ( 4 γ ∗ ∗ ) 1 / q Tail ( u ; Q ) 1 / q . (49) 2M_0\ge\operatorname*{ess\,sup}_{Q}|u|+2(4\gamma_{**})^{1/q}\operatorname{Tail}(u;Q)^{1/q}. \tag*{(49)} 2 M 0 ≥ Q ess sup ∣ u ∣ + 2 ( 4 γ ∗∗ ) 1/ q Tail ( u ; Q ) 1/ q . (49)
where γ ∗ ∗ \gamma_{**} γ ∗∗ is the constant in Lemma 4.1. This gives both ess sup Q ∣ u ∣ ≤ 2 M 0 \operatorname*{ess\,sup}_{Q}|u|\le2M_0 ess sup Q ∣ u ∣ ≤ 2 M 0 and Tail ( u ; Q ) ≤ M 0 q / ( 4 γ ∗ ∗ ) \operatorname{Tail}(u;Q)\le M_0^q/(4\gamma_{**}) Tail ( u ; Q ) ≤ M 0 q / ( 4 γ ∗∗ ) . The factor in front of the tail is absorbed into the structural constant of (3) . Set θ 0 = M 0 q / ( q − 1 ) \theta_0=M_0^{q/(q-1)} θ 0 = M 0 q / ( q − 1 ) and choose ϱ 0 > 0 \varrho_0>0 ϱ 0 > 0 such that Q 0 : = Q 2 ϱ 0 ( θ 0 ) ⊂ Q Q_0:=Q_{2\varrho_0}(\theta_0)\subset Q Q 0 := Q 2 ϱ 0 ( θ 0 ) ⊂ Q . For j ≥ 0 j\ge0 j ≥ 0 , we shall use the notation
ϱ j = λ j ϱ 0 , θ j = M j q − 1 , Q j = Q 2 ϱ j ( θ j ) , Q ^ j = K 2 θ j × ( − θ j ( c j ) 2 s , 0 ) . \varrho_j=\lambda^j\varrho_0,\qquad\theta_j=M_j^{q-1},\qquad Q_j=Q_{2\varrho_j}(\theta_j),\qquad\widehat Q_j=K_{2\theta_j}\times(-\theta_j(c_j)^{2s},0). ϱ j = λ j ϱ 0 , θ j = M j q − 1 , Q j = Q 2 ϱ j ( θ j ) , Q j = K 2 θ j × ( − θ j ( c j ) 2 s , 0 ) .
The constants c , λ ∈ ( 0 , 1 ) c,\lambda\in(0,1) c , λ ∈ ( 0 , 1 ) are chosen below. Let the ν \nu ν be the constant in Lemma 4.1, the iteration continues as long as
∣ { u ≤ M j } ∩ Q c ϱ j ( θ j ) ∣ > ν ∣ Q c ϱ j ( θ j ) ∣ and \left|\{u\le M_j\}\cap Q_{c\varrho_j}(\theta_j)\right|>\nu|Q_{c\varrho_j}(\theta_j)|\quad\text{and} { u ≤ M j } ∩ Q c ϱ j ( θ j ) > ν ∣ Q c ϱ j ( θ j ) ∣ and
∣ { − u ≤ M j } ∩ Q c ϱ j ( θ j ) ∣ > ν ∣ Q c ϱ j ( θ j ) ∣ . (50) \left|\{-u\le M_j\}\cap Q_{c\varrho_j}(\theta_j)\right|>\nu|Q_{c\varrho_j}(\theta_j)|. \tag*{(50)} { − u ≤ M j } ∩ Q c ϱ j ( θ j ) > ν ∣ Q c ϱ j ( θ j ) ∣. (50)
The first index at which one of these conditions fails is treated in § 6.2 .
Reduction of oscillation near zero
The first step
Assume (6.2) for j = 0 j=0 j = 0 . Apply Lemma 5.5 to u u u and − u -u − u , with local radius c ϱ 0 c\varrho_0 c ϱ 0 and reference cylinder Q ^ 0 \widehat{Q}_0 Q 0 . The required inclusion is
K 2 c ϱ 0 × ( − θ 0 ( c ϱ 0 ) 2 s , 0 ) ⊂ Q ^ 0 . K_{2c\varrho_0}\times\left(-\theta_0(c\varrho_0)^{2s},0\right)\subset\widehat{Q}_0. K 2 c ϱ 0 × ( − θ 0 ( c ϱ 0 ) 2 s , 0 ) ⊂ Q 0 .
If Tail ( u ; Q ^ 0 ) ≤ γ M 0 q \operatorname{Tail}(u;\widehat{Q}_0)\leq\gamma M_0^q Tail ( u ; Q 0 ) ≤ γ M 0 q , the lemma gives ∣ u ∣ ≤ 2 M 0 ( 1 − η ) |u|\leq2M_0(1-\eta) ∣ u ∣ ≤ 2 M 0 ( 1 − η ) in K c ϱ 0 / 2 × ( − 7 16 ϑ 0 ( c ϱ 0 ) 2 s , 0 ) K_{c\varrho_0/2}\times\left(-\frac{7}{16}\vartheta_0(c\varrho_0)^{2s},0\right) K c ϱ 0 /2 × ( − 16 7 ϑ 0 ( c ϱ 0 ) 2 s , 0 ) . By (6.1), and taking
γ ‾ ≥ max { 2 γ − 1 / q , 2 ( 4 γ ∗ ∗ ) 1 / q , 2 } , (51) \overline{\gamma}\geq\max\{2\gamma^{-1/q},2(4\gamma_{**})^{1/q},2\}, \tag*{(51)} γ ≥ max { 2 γ − 1/ q , 2 ( 4 γ ∗∗ ) 1/ q , 2 } , (51)
we obtain
ess sup K c ϱ 0 / 2 × ( − 7 16 ϑ 0 ( c ϱ 0 ) 2 s , 0 ) ∣ u ∣ ≤ 2 M 1 , \underset{K_{c\varrho_0/2}\times\left(-\frac{7}{16}\vartheta_0(c\varrho_0)^{2s},0\right)}{\operatorname{ess\,sup}} |u|\leq2M_1, K c ϱ 0 /2 × ( − 16 7 ϑ 0 ( c ϱ 0 ) 2 s , 0 ) ess sup ∣ u ∣ ≤ 2 M 1 ,
2 M 1 = max { 2 M 0 ( 1 − η ) , γ ‾ Tail ( u ; Q ^ 0 ) 1 / q } . 2M_1=\max\{2M_0(1-\eta),\overline{\gamma}\operatorname{Tail}(u;\widehat{Q}_0)^{1/q}\}. 2 M 1 = max { 2 M 0 ( 1 − η ) , γ Tail ( u ; Q 0 ) 1/ q } .
Splitting the tail at radius R R R and using ∣ u ∣ ≤ 2 M 0 |u|\leq2M_0 ∣ u ∣ ≤ 2 M 0 in Q Q Q ,
Tail ( u ; Q ^ 0 ) ≤ ∫ T 1 T 2 ∫ R N ∖ K R ∣ u ( y , t ) ∣ ∣ y ∣ N + 2 s d y d t + ( 2 M 0 ) ∫ − ( c ϱ 0 ) 2 s ϑ 0 0 ∫ K R ∖ K 2 ϱ 0 1 ∣ y ∣ N + 2 s d y d t ≤ Tail ( u ; Q ) + C c 2 s M 0 q . (52) \begin{aligned}
\operatorname{Tail}(u;\widehat{Q}_0)\leq\int_{T_1}^{T_2}\int_{\mathbb{R}^N\setminus K_R}\frac{|u(y,t)|}{|y|^{N+2s}}\,dy\,dt+(2M_0)\int_{-(c\varrho_0)^{2s}\vartheta_0}^{0}\int_{K_R\setminus K_{2\varrho_0}}\frac{1}{|y|^{N+2s}}\,dy\,dt \\
&\leq\operatorname{Tail}(u;Q)+Cc^{2s}M_0^q. \tag*{(52)}
\end{aligned} Tail ( u ; Q 0 ) ≤ ∫ T 1 T 2 ∫ R N ∖ K R ∣ y ∣ N + 2 s ∣ u ( y , t ) ∣ d y d t + ( 2 M 0 ) ∫ − ( c ϱ 0 ) 2 s ϑ 0 0 ∫ K R ∖ K 2 ϱ 0 ∣ y ∣ N + 2 s 1 d y d t ≤ Tail ( u ; Q ) + C c 2 s M 0 q . (52)
Consequently, there is a γ ∗ ≥ 1 \gamma_*\geq1 γ ∗ ≥ 1 , depending only on the data, such that
( 1 − η ) M 0 ≤ M 1 ≤ γ ∗ M 0 . (1-\eta)M_0\leq M_1\leq\gamma_*M_0. ( 1 − η ) M 0 ≤ M 1 ≤ γ ∗ M 0 .
We fix
γ ∗ = max { γ ‾ ( 1 ( 4 γ ∗ ∗ ) 1 / q + C ) , 2 + C ( 1 − η ) 1 / q } \gamma_*=\max\left\{\overline{\gamma}\left(\frac{1}{(4\gamma_{**})^{1/q}}+C\right),2+\frac{C}{(1-\eta)^{1/q}}\right\} γ ∗ = max { γ ( ( 4 γ ∗∗ ) 1/ q 1 + C ) , 2 + ( 1 − η ) 1/ q C }
and put
γ ^ = max { γ ∗ q − 1 , ( 1 − η ) q − 1 } . \widehat{\gamma}=\max\{\gamma_*^{q-1},(1-\eta)^{q-1}\}. γ = max { γ ∗ q − 1 , ( 1 − η ) q − 1 } .
Then θ 1 ≤ γ ^ θ 0 \theta_1\leq\widehat{\gamma}\theta_0 θ 1 ≤ γ θ 0 . Choose
λ ≤ min { c 4 , c 2 ( 7 ϑ 32 γ ^ ) 1 / ( 2 s ) } . (53) \lambda\leq\min\left\{\frac{c}{4},\frac{c}{2}\left(\frac{7\vartheta}{32\widehat{\gamma}}\right)^{1/(2s)}\right\}. \tag*{(53)} λ ≤ min { 4 c , 2 c ( 32 γ 7 ϑ ) 1/ ( 2 s ) } . (53)
This ensures
Q 1 ⊂ K c ϱ 0 / 2 × ( − 7 16 ϑ 0 ( c ϱ 0 ) 2 s , 0 ) ⊂ Q 0 , Q_1\subset K_{c\varrho_0/2}\times\left(-\frac{7}{16}\vartheta_0(c\varrho_0)^{2s},0\right)\subset Q_0, Q 1 ⊂ K c ϱ 0 /2 × ( − 16 7 ϑ 0 ( c ϱ 0 ) 2 s , 0 ) ⊂ Q 0 ,
and hence
ess sup Q 1 ∣ u ∣ ≤ 2 M 1 . \underset{Q_1}{\operatorname{ess\,sup}}|u|\leq2M_1. Q 1 ess sup ∣ u ∣ ≤ 2 M 1 .
The second step and the induction
If (6.2) holds at j = 1 j=1 j = 1 , repeat the argument in Q ^ 1 \widehat{Q}_1 Q 1 and set
2 M 2 = max { 2 M 1 ( 1 − η ) , γ ‾ Tail ( u ; Q ^ 1 ) 1 / q } . 2M_2=\max\{2M_1(1-\eta),\overline{\gamma}\operatorname{Tail}(u;\widehat{Q}_1)^{1/q}\}. 2 M 2 = max { 2 M 1 ( 1 − η ) , γ Tail ( u ; Q 1 ) 1/ q } .
The nesting of the time intervals and ∣ u ∣ ≤ 2 M 0 |u|\leq2M_0 ∣ u ∣ ≤ 2 M 0 in Q 0 Q_0 Q 0 give
Tail ( u ; Q ^ 1 ) = ∫ − θ 1 ( c ϱ 1 ) 2 s 0 ∫ R N ∖ K 2 ϱ 0 + ∫ − θ 1 ( c ϱ 1 ) 2 s 0 ∫ K 2 ϱ 0 ∖ K 2 ϱ 1 ∣ u ( y , t ) ∣ ∣ y ∣ N + 2 s d y d t ≤ Tail ( u ; Q ^ 0 ) + C c 2 s M 1 q − 1 M 0 . \begin{aligned}
\operatorname{Tail}(u;\widehat{Q}_1)=\int_{-\theta_1(c\varrho_1)^{2s}}^{0}\int_{\mathbb{R}^N\setminus K_{2\varrho_0}}+\int_{-\theta_1(c\varrho_1)^{2s}}^{0}\int_{K_{2\varrho_0}\setminus K_{2\varrho_1}}\frac{|u(y,t)|}{|y|^{N+2s}}\,dy\,dt \\
&\leq\operatorname{Tail}(u;\widehat{Q}_0)+Cc^{2s}M_1^{q-1}M_0.
\end{aligned} Tail ( u ; Q 1 ) = ∫ − θ 1 ( c ϱ 1 ) 2 s 0 ∫ R N ∖ K 2 ϱ 0 + ∫ − θ 1 ( c ϱ 1 ) 2 s 0 ∫ K 2 ϱ 0 ∖ K 2 ϱ 1 ∣ y ∣ N + 2 s ∣ u ( y , t ) ∣ d y d t ≤ Tail ( u ; Q 0 ) + C c 2 s M 1 q − 1 M 0 .
Since γ q Tail ( u ; Q ^ 0 ) ≤ ( 2 M 1 ) q \gamma^q\operatorname{Tail}(u;\widehat{Q}_0) \le(2M_1)^q γ q Tail ( u ; Q 0 ) ≤ ( 2 M 1 ) q and M 0 ≤ M 1 / ( 1 − η ) M_0 \le M_1/(1-\eta) M 0 ≤ M 1 / ( 1 − η ) , thus
( 1 − η ) M 1 ≤ M 2 ≤ γ ∗ M 1 . (1-\eta)M_1 \le M_2 \le\gamma_*M_1. ( 1 − η ) M 1 ≤ M 2 ≤ γ ∗ M 1 .
At a general step, define
2 M j + 1 = max { 2 M j ( 1 − η ) , γ ˉ Tail ( u ; Q ^ j ) 1 / q } . (54) 2M_{j+1}=\max\{2M_j(1-\eta),\bar{\gamma}\operatorname{Tail}(u;\widehat{Q}_j)^{1/q}\}. \tag*{(54)} 2 M j + 1 = max { 2 M j ( 1 − η ) , γ ˉ Tail ( u ; Q j ) 1/ q } . (54)
Lemma 5.5 bounds ∣ u ∣ |u| ∣ u ∣ by 2 M j + 1 2M_{j+1} 2 M j + 1 on the corresponding smaller cylinder. For j ≥ 1 j\ge1 j ≥ 1 , the same splitting gives
Tail ( u ; Q ^ j ) ≤ Tail ( u ; Q ^ j − 1 ) + C 2 s M j q − 1 M j − 1 ≤ ( 2 M j γ ˉ ) q + C 2 s 1 − η M j q . (55) \begin{aligned}
\operatorname{Tail}(u;\widehat{Q}_j)\le\operatorname{Tail}(u;\widehat{Q}_{j-1})+C^{2s}M_j^{q-1}M_{j-1} \\
&\le\left(\frac{2M_j}{\bar{\gamma}}\right)^q+\frac{C^{2s}}{1-\eta}M_j^q. \tag*{(55)}
\end{aligned} Tail ( u ; Q j ) ≤ Tail ( u ; Q j − 1 ) + C 2 s M j q − 1 M j − 1 ≤ ( γ ˉ 2 M j ) q + 1 − η C 2 s M j q . (55)
Inductively, we have
( 1 − η ) M j ≤ M j + 1 ≤ γ ∗ M j , θ j + 1 ≤ θ ^ j , (1-\eta)M_j\le M_{j+1}\le\gamma_*M_j,\qquad\theta_{j+1}\le\widehat{\theta}_j, ( 1 − η ) M j ≤ M j + 1 ≤ γ ∗ M j , θ j + 1 ≤ θ j ,
Q j + 1 ⊂ K c ϑ j / 2 × ( − 7 16 ϑ j ( c ϱ j ) 2 s , 0 ) ⊂ Q j , ess sup Q j + 1 ∣ u ∣ ≤ 2 M j + 1 . (56) Q_{j+1}\subset K_{c\vartheta_j/2}\times\left(-\frac{7}{16}\vartheta_j(c\varrho_j)^{2s},0\right)\subset Q_j,\qquad\underset{Q_{j+1}}{\operatorname{ess\,sup}}|u|\le2M_{j+1}. \tag*{(56)} Q j + 1 ⊂ K c ϑ j /2 × ( − 16 7 ϑ j ( c ϱ j ) 2 s , 0 ) ⊂ Q j , Q j + 1 ess sup ∣ u ∣ ≤ 2 M j + 1 . (56)
In particular, (53) gives
θ j + 1 ( c ϱ j + 1 ) 2 s ≤ γ ^ 2 s θ j ( c ϱ j ) 2 s , γ ^ 2 s < 1. (57) \theta_{j+1}(c\varrho_{j+1})^{2s}\le\widehat{\gamma}^{2s}\theta_j(c\varrho_j)^{2s},\qquad\widehat{\gamma}^{2s}<1. \tag*{(57)} θ j + 1 ( c ϱ j + 1 ) 2 s ≤ γ 2 s θ j ( c ϱ j ) 2 s , γ 2 s < 1. (57)
6.1.3. The tail estimate. Split R N \ K 2 ϱ j \mathbb{R}^N\backslash K_{2\varrho_j} R N \ K 2 ϱ j into R N \ K R \mathbb{R}^N\backslash K_R R N \ K R , K R \ K 2 ϱ 0 K_R\backslash K_{2\varrho_0} K R \ K 2 ϱ 0 , and K 2 ϱ i − 1 \ K 2 ϱ i K_{2\varrho_{i-1}}\backslash K_{2\varrho_i} K 2 ϱ i − 1 \ K 2 ϱ i , 1 ≤ i ≤ j 1\le i\le j 1 ≤ i ≤ j . The bound ∣ u ∣ ≤ 2 M i − 1 |u|\le2M_{i-1} ∣ u ∣ ≤ 2 M i − 1 holds on each such annulus throughout the time interval of Q ^ j \widehat{Q}_j Q j . Therefore
Tail ( u ; Q ^ j ) = ∫ − θ j ( c ϱ j ) 2 s 0 ∫ R N \ K 2 ϱ j ∣ u ( y , t ) ∣ ∣ y ∣ N + 2 s d y d t ≤ ∫ − θ j ( c ϱ j ) 2 s 0 ∫ R N \ K R ∣ u ( y , t ) ∣ ∣ y ∣ N + 2 s d y d t + ∫ − θ j ( c ϱ j ) 2 s 0 ∫ K R \ K 2 ϱ 0 ∣ u ( y , t ) ∣ ∣ y ∣ N + 2 s d y d t + ∑ i = 1 j ∫ − θ j ( c ϱ j ) 2 s 0 ∫ K 2 ϱ i − 1 \ K 2 ϱ i ∣ u ( y , t ) ∣ ∣ y ∣ N + 2 s d y d t ≤ ∫ − θ j ( c ϱ j ) 2 s 0 ∫ R N \ K R ∣ u ( y , t ) ∣ ∣ y ∣ N + 2 s d y d t + C θ j ( c ϱ j ) 2 s ( M 0 ϱ 0 − 2 s + ∑ i = 1 j M i − 1 ϱ i − 2 s ) . (58) \begin{aligned}
\operatorname{Tail}(u;\widehat{Q}_j)
&=\int_{-\theta_j(c\varrho_j)^{2s}}^0\int_{\mathbb{R}^N\backslash K_{2\varrho_j}}\frac{|u(y,t)|}{|y|^{N+2s}}\,dy\,dt\\
&\le\int_{-\theta_j(c\varrho_j)^{2s}}^0\int_{\mathbb{R}^N\backslash K_R}\frac{|u(y,t)|}{|y|^{N+2s}}\,dy\,dt
+\int_{-\theta_j(c\varrho_j)^{2s}}^0\int_{K_R\backslash K_{2\varrho_0}}\frac{|u(y,t)|}{|y|^{N+2s}}\,dy\,dt\\
&\qquad+\sum_{i=1}^j\int_{-\theta_j(c\varrho_j)^{2s}}^0\int_{K_{2\varrho_{i-1}}\backslash K_{2\varrho_i}}\frac{|u(y,t)|}{|y|^{N+2s}}\,dy\,dt\\
&\le\int_{-\theta_j(c\varrho_j)^{2s}}^0\int_{\mathbb{R}^N\backslash K_R}\frac{|u(y,t)|}{|y|^{N+2s}}\,dy\,dt+C\theta_j(c\varrho_j)^{2s}\left(M_0\varrho_0^{-2s}+\sum_{i=1}^jM_{i-1}\varrho_i^{-2s}\right). \tag*{(58)}
\end{aligned} Tail ( u ; Q j ) = ∫ − θ j ( c ϱ j ) 2 s 0 ∫ R N \ K 2 ϱ j ∣ y ∣ N + 2 s ∣ u ( y , t ) ∣ d y d t ≤ ∫ − θ j ( c ϱ j ) 2 s 0 ∫ R N \ K R ∣ y ∣ N + 2 s ∣ u ( y , t ) ∣ d y d t + ∫ − θ j ( c ϱ j ) 2 s 0 ∫ K R \ K 2 ϱ 0 ∣ y ∣ N + 2 s ∣ u ( y , t ) ∣ d y d t + i = 1 ∑ j ∫ − θ j ( c ϱ j ) 2 s 0 ∫ K 2 ϱ i − 1 \ K 2 ϱ i ∣ y ∣ N + 2 s ∣ u ( y , t ) ∣ d y d t ≤ ∫ − θ j ( c ϱ j ) 2 s 0 ∫ R N \ K R ∣ y ∣ N + 2 s ∣ u ( y , t ) ∣ d y d t + C θ j ( c ϱ j ) 2 s ( M 0 ϱ 0 − 2 s + i = 1 ∑ j M i − 1 ϱ i − 2 s ) . (58)
Using M i − 1 ≤ ( 1 − η ) − ( j − i + 1 ) M j M_{i-1}\le(1-\eta)^{-(j-i+1)}M_j M i − 1 ≤ ( 1 − η ) − ( j − i + 1 ) M j and ϱ j / ϱ i = λ j − i \varrho_j/\varrho_i=\lambda^{j-i} ϱ j / ϱ i = λ j − i , we have
θ j ( c ϱ j ) 2 s M 0 ϱ 0 − 2 s ≤ c 2 s M j q ( λ 2 s 1 − η ) j , (59) \theta_j(c\varrho_j)^{2s}M_0\varrho_0^{-2s}\le c^{2s}M_j^q\left(\frac{\lambda^{2s}}{1-\eta}\right)^j, \tag*{(59)} θ j ( c ϱ j ) 2 s M 0 ϱ 0 − 2 s ≤ c 2 s M j q ( 1 − η λ 2 s ) j , (59)
θ j ( c ϱ j ) 2 s ∑ i = 1 j M i − 1 ϱ i − 2 s ≤ c 2 s M j q 1 − η ∑ i = 1 j ( λ 2 s 1 − η ) j − i . (60) \theta_j(c\varrho_j)^{2s}\sum_{i=1}^jM_{i-1}\varrho_i^{-2s}\le\frac{c^{2s}M_j^q}{1-\eta}\sum_{i=1}^j\left(\frac{\lambda^{2s}}{1-\eta}\right)^{j-i}. \tag*{(60)} θ j ( c ϱ j ) 2 s i = 1 ∑ j M i − 1 ϱ i − 2 s ≤ 1 − η c 2 s M j q i = 1 ∑ j ( 1 − η λ 2 s ) j − i . (60)
Impose λ 2 s / ( 1 − η ) ≤ 1 / 2 \lambda^{2s}/(1-\eta) \le1/2 λ 2 s / ( 1 − η ) ≤ 1/2 and sum the series. Putting (59) ,(60) back in (58) ,
Tail ( u ; Q ^ j ) ≤ ∫ − θ j ( c ϱ j ) 2 s 0 ∫ R N ∖ K R ∣ u ( y , t ) ∣ ∣ y ∣ N + 2 s d y d t + C c 2 s M j q . \operatorname{Tail}(u;\widehat{Q}_j) \le\int_{-\theta_j(c\varrho_j)^{2s}}^0 \int_{\mathbb{R}^N\setminus K_R} \frac{|u(y,t)|}{|y|^{N+2s}} \,dy\,dt + Cc^{2s}M_j^q. Tail ( u ; Q j ) ≤ ∫ − θ j ( c ϱ j ) 2 s 0 ∫ R N ∖ K R ∣ y ∣ N + 2 s ∣ u ( y , t ) ∣ d y d t + C c 2 s M j q .
Using ( a + b ) 1 / q ≤ C ( q ) ( a 1 / q + b 1 / q ) (a+b)^{1/q} \le C(q)(a^{1/q}+b^{1/q}) ( a + b ) 1/ q ≤ C ( q ) ( a 1/ q + b 1/ q ) and choosing
c ≤ ( η 4 γ C ) q / ( 2 s ) , c \le\left(\frac{\eta}{4\gamma C}\right)^{q/(2s)}, c ≤ ( 4 γ C η ) q / ( 2 s ) ,
we obtain from (54) that
2 M j + 1 ≤ ( 1 − η 2 ) 2 M j + C ( ∫ − θ j ( c ϱ j ) 2 s 0 ∫ R N ∖ K R ∣ u ( y , t ) ∣ ∣ y ∣ N + 2 s d y d t ) 1 / q . (61) 2M_{j+1} \le\left(1-\frac{\eta}{2}\right)2M_j+C\left(\int_{-\theta_j(c\varrho_j)^{2s}}^0\int_{\mathbb{R}^N\setminus K_R}\frac{|u(y,t)|}{|y|^{N+2s}}\,dy\,dt\right)^{1/q}. \tag*{(61)} 2 M j + 1 ≤ ( 1 − 2 η ) 2 M j + C ( ∫ − θ j ( c ϱ j ) 2 s 0 ∫ R N ∖ K R ∣ y ∣ N + 2 s ∣ u ( y , t ) ∣ d y d t ) 1/ q . (61)
The discrete estimate
The time intervals are nested by (57) . Summing (61) and using (49) , we first obtain
2 M j ≤ 2 M 0 + C Tail ( u ; Q ( R , θ 0 ( c ϱ 0 ) 2 s ) ) 1 / q ≤ 2 γ ~ M 0 , (62) 2M_j\le2M_0+C\operatorname{Tail}(u;Q(R,\theta_0(c\varrho_0)^{2s}))^{1/q}\le2\widetilde{\gamma}M_0, \tag*{(62)} 2 M j ≤ 2 M 0 + C Tail ( u ; Q ( R , θ 0 ( c ϱ 0 ) 2 s ) ) 1/ q ≤ 2 γ M 0 , (62)
where γ ~ ≥ 1 \widetilde{\gamma}\ge1 γ ≥ 1 depends only on the data.
For j ≥ 2 j\ge2 j ≥ 2 , we now iterate (61) only from ⌊ j / 2 ⌋ \lfloor j/2\rfloor ⌊ j /2 ⌋ to j j j . By (57) , for ⌊ j / 2 ⌋ ≤ i ≤ j \lfloor j/2\rfloor\le i\le j ⌊ j /2 ⌋ ≤ i ≤ j ,
θ i ( c ϱ i ) 2 s ≤ θ 0 ( c ϱ 0 ) 2 s ( γ ~ λ 2 s ) ⌊ j / 2 ⌋ . \theta_i(c\varrho_i)^{2s}\le\theta_0(c\varrho_0)^{2s}(\widetilde{\gamma}\lambda^{2s})^{\lfloor j/2\rfloor}. θ i ( c ϱ i ) 2 s ≤ θ 0 ( c ϱ 0 ) 2 s ( γ λ 2 s ) ⌊ j /2 ⌋ .
The geometric sum of the coefficients of the tail terms is bounded by
∑ i = ⌊ j / 2 ⌋ j − 1 ( 1 − η 2 ) j − 1 − i ≤ 2 η . \sum_{i=\lfloor j/2\rfloor}^{j-1}\left(1-\frac{\eta}{2}\right)^{j-1-i}\le\frac{2}{\eta}. i = ⌊ j /2 ⌋ ∑ j − 1 ( 1 − 2 η ) j − 1 − i ≤ η 2 .
Consequently, (62) gives
2 M j ≤ 2 γ ~ M 0 ( 1 − η 2 ) j − ⌊ j / 2 ⌋ + C ( ∫ − θ 0 ( c ϱ 0 ) 2 s ( γ ~ λ 2 s ) ⌊ j / 2 ⌋ 0 ∫ R N ∖ K R ∣ u ( y , t ) ∣ ∣ y ∣ N + 2 s d y d t ) 1 / q . (63) 2M_j\le2\widetilde{\gamma}M_0\left(1-\frac{\eta}{2}\right)^{j-\lfloor j/2\rfloor}
+C\left(\int_{-\theta_0(c\varrho_0)^{2s}(\widetilde{\gamma}\lambda^{2s})^{\lfloor j/2\rfloor}}^0\int_{\mathbb{R}^N\setminus K_R}\frac{|u(y,t)|}{|y|^{N+2s}}\,dy\,dt\right)^{1/q}. \tag*{(63)} 2 M j ≤ 2 γ M 0 ( 1 − 2 η ) j − ⌊ j /2 ⌋ + C ( ∫ − θ 0 ( c ϱ 0 ) 2 s ( γ λ 2 s ) ⌊ j /2 ⌋ 0 ∫ R N ∖ K R ∣ y ∣ N + 2 s ∣ u ( y , t ) ∣ d y d t ) 1/ q . (63)
This estimate holds at every completed step, including the first stopping index. If the near-zero alternative persists, both terms tend to zero as j → ∞ j\to\infty j → ∞ . The initial radius ϱ 0 \varrho_0 ϱ 0 remains fixed throughout the argument.
Reduction of oscillation away from zero
Let ℓ \ell ℓ be the first index at which (50) fails. Replacing u u u with − u -u − u if necessary, we may assume
∣ { u ≤ M ℓ } ∩ Q c ϱ ℓ ( θ ℓ ) ∣ ≤ ν ∣ Q c ϱ ℓ ( θ ℓ ) ∣ . |\{u\le M_\ell\}\cap Q_{c\varrho_\ell}(\theta_\ell)|\le\nu|Q_{c\varrho_\ell}(\theta_\ell)|. ∣ { u ≤ M ℓ } ∩ Q c ϱ ℓ ( θ ℓ ) ∣ ≤ ν ∣ Q c ϱ ℓ ( θ ℓ ) ∣.
For ℓ ≥ 1 \ell\ge1 ℓ ≥ 1 , (55) gives
Tail ( u ; Q ^ ℓ ) ≤ ( 2 γ ~ ) q M ℓ q + C c 2 s 1 − η M ℓ q . (64) \operatorname{Tail}(u;\widehat{Q}_\ell)\le\left(\frac{2}{\widetilde{\gamma}}\right)^qM_\ell^q+\frac{Cc^{2s}}{1-\eta}M_\ell^q. \tag*{(64)} Tail ( u ; Q ℓ ) ≤ ( γ 2 ) q M ℓ q + 1 − η C c 2 s M ℓ q . (64)
By (51) , the first coefficient is at most 1 / ( 4 γ ∗ ∗ ) 1/(4\gamma_{**}) 1/ ( 4 γ ∗∗ ) . Select
c ≤ min { ( η 4 γ C ) q 2 s , ( 1 8 C γ ∗ ∗ ) 1 2 s } . c \le\min\left\{\left(\frac{\eta}{4\gamma C}\right)^{\frac{q}{2s}},\left(\frac{1}{8C\gamma_{**}}\right)^{\frac{1}{2s}}\right\}. c ≤ min { ( 4 γ C η ) 2 s q , ( 8 C γ ∗∗ 1 ) 2 s 1 } .
so that the second coefficient has the same bound. For ℓ = 0 \ell=0 ℓ = 0 , use (49) and (52) . Thus, in either case, γ ∗ ∗ Tail ( w ; Q ~ ℓ ) ≤ M ℓ q / 2 \gamma_{**}\operatorname{Tail}(w;\widetilde{Q}_{\ell})\le M_{\ell}^{q}/2 γ ∗∗ Tail ( w ; Q ℓ ) ≤ M ℓ q /2 . Lemma 4.1, with radius c Q ℓ cQ_{\ell} c Q ℓ and reference cylinder Q ~ ℓ \widetilde{Q}_{\ell} Q ℓ , now yields
1 4 M ℓ ≤ u ≤ 2 M ℓ a.e. in Q c Q ℓ / 2 ( θ ℓ ) . \frac{1}{4}M_{\ell}\le u\le2M_{\ell}\quad\text{a.e. in }Q_{cQ_{\ell}/2}(\theta_{\ell}). 4 1 M ℓ ≤ u ≤ 2 M ℓ a.e. in Q c Q ℓ /2 ( θ ℓ ) .
The upper bound follows from (56) .
Introduce
x ~ = x , t ~ = t ( M ℓ / 4 ) q − 1 , v ~ ( x ~ , t ~ ) = 4 M ℓ u ( x ~ , ( M ℓ / 4 ) q − 1 t ~ ) . \widetilde{x}=x,\qquad\widetilde{t}=\frac{t}{(M_{\ell}/4)^{q-1}},\qquad\widetilde{v}(\widetilde{x},\widetilde{t})=\frac{4}{M_{\ell}}u\left(\widetilde{x},(M_{\ell}/4)^{q-1}\widetilde{t}\right). x = x , t = ( M ℓ /4 ) q − 1 t , v ( x , t ) = M ℓ 4 u ( x , ( M ℓ /4 ) q − 1 t ) .
Then
1 ≤ v ~ ≤ 8 in Q c Q ℓ / 2 ( 4 q − 1 ) , 1\le\widetilde{v}\le8\quad\text{in }Q_{cQ_{\ell}/2}(4^{q-1}), 1 ≤ v ≤ 8 in Q c Q ℓ /2 ( 4 q − 1 ) ,
and v ~ \widetilde{v} v is a weak solution to
∂ t ~ v ~ q + L ~ v ~ = 0. \partial_{\widetilde{t}}\widetilde{v}^{q}+\widetilde{\mathcal{L}}\widetilde{v}=0. ∂ t v q + L v = 0.
Here K ~ ( x ~ , y ~ , t ~ ) = K ( x ~ , y ~ , ( M ℓ / 4 ) q − 1 t ~ ) \widetilde{K}(\widetilde{x},\widetilde{y},\widetilde{t})=K(\widetilde{x},\widetilde{y},(M_{\ell}/4)^{q-1}\widetilde{t}) K ( x , y , t ) = K ( x , y , ( M ℓ /4 ) q − 1 t ) satisfies (2) with the same constants.
The oscillation estimate after normalization
Set
c q = q min { 1 , 8 q − 1 } , C q = q max { 1 , 8 q − 1 } . c_q=q\min\{1,8^{q-1}\},\qquad C_q=q\max\{1,8^{q-1}\}. c q = q min { 1 , 8 q − 1 } , C q = q max { 1 , 8 q − 1 } .
For two numbers a , k ∈ [ 1 , 8 ] a,k\in[1,8] a , k ∈ [ 1 , 8 ] , by (7) and (6) , we have
c q 2 ( a − k ) ± 2 ≤ q ± ( a , k ) ≤ C q 2 ( a − k ) ± 2 , c q ( a − k ) ± ≤ ( a q − k q ) ± ≤ C q ( a − k ) ± . (65) \begin{aligned}
\frac{c_q}{2}(a-k)_{\pm}^{2}&\le q_{\pm}(a,k)\le\frac{C_q}{2}(a-k)_{\pm}^{2},\\
c_q(a-k)_{\pm}&\le(a^q-k^q)_{\pm}\le C_q(a-k)_{\pm}. \tag*{(65)}
\end{aligned} 2 c q ( a − k ) ± 2 c q ( a − k ) ± ≤ q ± ( a , k ) ≤ 2 C q ( a − k ) ± 2 , ≤ ( a q − k q ) ± ≤ C q ( a − k ) ± . (65)
Since v ~ \widetilde{v} v satisfies a similar equation as (1) in Q c Q ℓ / 2 ( 4 q − 1 ) Q_{cQ_{\ell}/2}(4^{q-1}) Q c Q ℓ /2 ( 4 q − 1 ) , we may derive an analogous energy estimate in K r × ( τ 1 , τ 2 ) ⊂ Q c Q ℓ / 2 ( 4 q − 1 ) K_r\times(\tau_1,\tau_2)\subset Q_{cQ_{\ell}/2}(4^{q-1}) K r × ( τ 1 , τ 2 ) ⊂ Q c Q ℓ /2 ( 4 q − 1 ) as that of Proposition 3.1 by using (65) and assuming that the time derivative of the moving level satisfies k ′ ≥ 0 k'\ge0 k ′ ≥ 0 in the subsolution case, and k ′ ≤ 0 k'\le0 k ′ ≤ 0 in the supersolution case.
c q 2 ∫ K r w ± ( x , τ 2 ) 2 ζ ( x , τ 2 ) 2 d x + c q ∫ τ 1 τ 2 ∫ K r ∣ k ′ ∣ 2 w ± d x d t ~ + 1 2 ∫ τ 1 τ 2 ∫ K r ∫ K r ∣ w ± ( x ) − w ± ( y ) ∣ 2 K ~ d y d x d t ~ + ∫ τ 1 τ 2 ∫ K r ∫ K r w ∓ ( y ) w ± ( x ) ζ ( x ) 2 K ~ d y d x d t ~ ≤ c q 2 ∫ K r w ± ( x , τ 1 ) 2 ζ ( x , τ 1 ) 2 d x + c q 2 ∫ τ 1 τ 2 ∫ K r w ± 2 ∣ ∂ t ζ ∣ 2 d x d t ~ + ∫ τ 1 τ 2 ∫ K r ∫ K r ∣ ζ ( x ) − ζ ( y ) ∣ 2 w ± ( x ) w ± ( y ) K ~ d y d x d t ~ + ∫ τ 1 τ 2 ∫ K r ∫ R N ∖ K r w ± ( y ) ζ ( x ) 2 w ± ( x ) K ~ d y d x d t ~ . (66) \begin{aligned}
&\frac{c_q}{2}\int_{K_r}w_{\pm}(x,\tau_2)^2\zeta(x,\tau_2)^2\,dx+c_q\int_{\tau_1}^{\tau_2}\int_{K_r}|k'|^2w_{\pm}\,dx\,d\tilde t\\
&\quad+\frac{1}{2}\int_{\tau_1}^{\tau_2}\int_{K_r}\int_{K_r}|w_{\pm}(x)-w_{\pm}(y)|^2\widetilde K\,dy\,dx\,d\tilde t\\
&\quad+\int_{\tau_1}^{\tau_2}\int_{K_r}\int_{K_r}w_{\mp}(y)w_{\pm}(x)\zeta(x)^2\widetilde K\,dy\,dx\,d\tilde t\\
&\leq\frac{c_q}{2}\int_{K_r}w_{\pm}(x,\tau_1)^2\zeta(x,\tau_1)^2\,dx+\frac{c_q}{2}\int_{\tau_1}^{\tau_2}\int_{K_r}w_{\pm}^2|\partial_t\zeta|^2\,dx\,d\tilde t\\
&\quad+\int_{\tau_1}^{\tau_2}\int_{K_r}\int_{K_r}|\zeta(x)-\zeta(y)|^2w_{\pm}(x)w_{\pm}(y)\widetilde K\,dy\,dx\,d\tilde t\\
&\quad+\int_{\tau_1}^{\tau_2}\int_{K_r}\int_{\mathbb{R}^N\setminus K_r}w_{\pm}(y)\zeta(x)^2w_{\pm}(x)\widetilde K\,dy\,dx\,d\tilde t. \tag*{(66)}
\end{aligned} 2 c q ∫ K r w ± ( x , τ 2 ) 2 ζ ( x , τ 2 ) 2 d x + c q ∫ τ 1 τ 2 ∫ K r ∣ k ′ ∣ 2 w ± d x d t ~ + 2 1 ∫ τ 1 τ 2 ∫ K r ∫ K r ∣ w ± ( x ) − w ± ( y ) ∣ 2 K d y d x d t ~ + ∫ τ 1 τ 2 ∫ K r ∫ K r w ∓ ( y ) w ± ( x ) ζ ( x ) 2 K d y d x d t ~ ≤ 2 c q ∫ K r w ± ( x , τ 1 ) 2 ζ ( x , τ 1 ) 2 d x + 2 c q ∫ τ 1 τ 2 ∫ K r w ± 2 ∣ ∂ t ζ ∣ 2 d x d t ~ + ∫ τ 1 τ 2 ∫ K r ∫ K r ∣ ζ ( x ) − ζ ( y ) ∣ 2 w ± ( x ) w ± ( y ) K d y d x d t ~ + ∫ τ 1 τ 2 ∫ K r ∫ R N ∖ K r w ± ( y ) ζ ( x ) 2 w ± ( x ) K d y d x d t ~ . (66)
Time arguments in the integrands are t ~ \tilde t t ~ when suppressed. This is the counterpart of [22 Proposition 2.1] for p = 2 p=2 p = 2 , from which Lemmas 3.1, 3.2 and 3.4 in [22 ] can be reproduced. We also note that the reduction μ = 0 \mu=0 μ = 0 made there for convenience is not needed. On the other hand, unlike in [22 ] , the coefficients c q / 2 c_q/2 c q /2 , c q / 2 c_q/2 c q /2 of the first terms on the left and right sides are generally different. This will affect reproducing Lemma 3.3 of that paper. To get it, we need to keep the exact time energy terms ∫ K r g ± ( u , k ) ζ ( x , τ ) 2 d x ∣ τ 1 τ 2 \left.\int_{K_r}g_{\pm}(u,k)\zeta(x,\tau)^2\,dx\right|_{\tau_1}^{\tau_2} ∫ K r g ± ( u , k ) ζ ( x , τ ) 2 d x τ 1 τ 2 , before using (65) , and estimate them as in (35) –(36) . Then the remaining terms in (66) can be handled by following the estimates in [22 Lemma 3.3] .
With these preparation, we may therefore follow the oscillation and tail iteration in [22 Sections 4.2–4.4] .
To this end, set
R ^ = 2 − [ ( q − 1 ) − s + 2 ] c Q ℓ . (67) \widehat R=2^{-\left[\frac{(q-1)_{-}}{s}+2\right]}c_Q\ell. \tag*{(67)} R = 2 − [ s ( q − 1 ) − + 2 ] c Q ℓ . (67)
Then, (64) and ∣ u ∣ ≤ 2 M ℓ |u|\leq2M_\ell ∣ u ∣ ≤ 2 M ℓ in Q ℓ Q_\ell Q ℓ give
∫ − R ^ 2 s 0 ∫ R N ∖ K c ℓ / 2 ∣ v ˉ ( y ~ , t ~ ) ∣ ∣ y ~ ∣ N + 2 s d y ~ d t ~ ≤ C . \int_{-\widehat R^{2s}}^0\int_{\mathbb{R}^N\setminus K_{c_\ell/2}}\frac{|\bar v(\tilde y,\tilde t)|}{|\tilde y|^{N+2s}}\,d\tilde y\,d\tilde t\leq C. ∫ − R 2 s 0 ∫ R N ∖ K c ℓ /2 ∣ y ~ ∣ N + 2 s ∣ v ˉ ( y ~ , t ~ ) ∣ d y ~ d t ~ ≤ C .
Here C C C depends only on the data, including q q q , and is independent of ℓ \ell ℓ . Indeed,
∫ − R ^ 2 s 0 ∫ R N ∖ K c ℓ / 2 ∣ v ˉ ( y ~ , t ~ ) ∣ ∣ y ~ ∣ N + 2 s d y ~ d t ~ = ( M ℓ 4 ) − q ∫ − R ^ 2 s ( M ℓ / 4 ) q − 1 0 ∫ R N ∖ K c ℓ / 2 ∣ u ( y , t ) ∣ ∣ y ∣ N + 2 s d y d t = ( M ℓ 4 ) − q ∫ − R ^ 2 s ( M ℓ / 4 ) q − 1 0 ∫ R N ∖ K 2 ℓ ∣ u ( y , t ) ∣ ∣ y ∣ N + 2 s d y d t + ( M ℓ 4 ) − q ∫ − R ^ 2 s ( M ℓ / 4 ) q − 1 0 ∫ K 2 ℓ ∖ K c ℓ / 2 ∣ u ( y , t ) ∣ ∣ y ∣ N + 2 s d y d t ≤ ( M ℓ 4 ) − q Tail ( u ; Q ^ ℓ ) + C R ^ ℓ 2 s ( c ϱ ℓ / 2 ) − 2 s ≤ C . \begin{aligned}
\int_{-\widehat R^{2s}}^0\int_{\mathbb{R}^N\setminus K_{c_\ell/2}}\frac{|\bar v(\tilde y,\tilde t)|}{|\tilde y|^{N+2s}}\,d\tilde y\,d\tilde t
&=\left(\frac{M_\ell}{4}\right)^{-q}\int_{-\widehat R^{2s}(M_\ell/4)^{q-1}}^0\int_{\mathbb{R}^N\setminus K_{c_\ell/2}}\frac{|u(y,t)|}{|y|^{N+2s}}\,dy\,dt\\
&=\left(\frac{M_\ell}{4}\right)^{-q}\int_{-\widehat R^{2s}(M_\ell/4)^{q-1}}^0\int_{\mathbb{R}^N\setminus K_{2\ell}}\frac{|u(y,t)|}{|y|^{N+2s}}\,dy\,dt\\
&\quad+\left(\frac{M_\ell}{4}\right)^{-q}\int_{-\widehat R^{2s}(M_\ell/4)^{q-1}}^0\int_{K_{2\ell}\setminus K_{c_\ell/2}}\frac{|u(y,t)|}{|y|^{N+2s}}\,dy\,dt \\
&\le\left(\frac{M_{\ell}}{4}\right)^{-q}\operatorname{Tail}(u;\widehat{Q}_{\ell})+C\widehat{R}_{\ell}^{2s}(c\varrho_{\ell}/2)^{-2s}\le C.
\end{aligned} ∫ − R 2 s 0 ∫ R N ∖ K c ℓ /2 ∣ y ~ ∣ N + 2 s ∣ v ˉ ( y ~ , t ~ ) ∣ d y ~ d t ~ = ( 4 M ℓ ) − q ∫ − R 2 s ( M ℓ /4 ) q − 1 0 ∫ R N ∖ K c ℓ /2 ∣ y ∣ N + 2 s ∣ u ( y , t ) ∣ d y d t = ( 4 M ℓ ) − q ∫ − R 2 s ( M ℓ /4 ) q − 1 0 ∫ R N ∖ K 2 ℓ ∣ y ∣ N + 2 s ∣ u ( y , t ) ∣ d y d t + ( 4 M ℓ ) − q ∫ − R 2 s ( M ℓ /4 ) q − 1 0 ∫ K 2 ℓ ∖ K c ℓ /2 ∣ y ∣ N + 2 s ∣ u ( y , t ) ∣ d y d t ≤ ( 4 M ℓ ) − q Tail ( u ; Q ℓ ) + C R ℓ 2 s ( c ϱ ℓ /2 ) − 2 s ≤ C .
Here in the second-to-last inequality we used R ^ ℓ 2 s ( M ℓ / 4 ) q − 1 ≤ ℓ ℓ ( c ϱ ℓ ) 2 s \widehat{R}_{\ell}^{2s}(M_{\ell}/4)^{q-1}\le\ell_{\ell}(c\varrho_{\ell})^{2s} R ℓ 2 s ( M ℓ /4 ) q − 1 ≤ ℓ ℓ ( c ϱ ℓ ) 2 s to enlarge the time interval in the first integral and control it by the corresponding tail, and the fact that ∣ u ∣ ≤ 2 M ℓ |u|\le2M_{\ell} ∣ u ∣ ≤ 2 M ℓ in Q ℓ Q_{\ell} Q ℓ to estimate the second integral. To get the last bound, we used (64) and (67) .
Thus, [22 Section 4.4] gives, for some β ∈ ( 0 , s ) \beta\in(0,s) β ∈ ( 0 , s ) and sufficiently small r / R ^ r/\widehat{R} r / R ,
ess osc Q r ( 1 ) v ‾ ≤ C ( r R ^ ) β + C ∫ − C ( R ^ ) s 0 ∫ R N ∖ K c ϱ ℓ / 2 ∣ v ‾ ( y ~ , t ~ ) ∣ ∣ y ~ ∣ N + 2 s d y ~ d t ~ . (68) \underset{Q_{r}(1)}{\operatorname{ess\,osc}}\overline{v}\le C\left(\frac{r}{\widehat{R}}\right)^{\beta}+C\int_{-C(\widehat{R})^{s}}^{0}\int_{\mathbb{R}^{N}\setminus K_{c\varrho_{\ell}/2}}\frac{|\overline{v}(\widetilde{y},\widetilde{t})|}{|\widetilde{y}|^{N+2s}}\,d\widetilde{y}\,d\widetilde{t}. \tag*{(68)} Q r ( 1 ) ess osc v ≤ C ( R r ) β + C ∫ − C ( R ) s 0 ∫ R N ∖ K c ϱ ℓ /2 ∣ y ∣ N + 2 s ∣ v ( y , t ) ∣ d y d t . (68)
The constants depend only on the data. The range of r r r is chosen so that the time interval in the tail is contained in ( − R ^ 2 s , 0 ) (-\widehat{R}^{2s},0) ( − R 2 s , 0 ) .
Rescaling to u u u .
For any measurable spatial region A A A and admissible H > 0 H>0 H > 0 , the change of variables gives
∫ − H 0 ∫ A ∣ v ‾ ( y ~ , t ~ ) ∣ ∣ y ~ ∣ N + 2 s d y ~ d t ~ = ( M ℓ 4 ) − q ∫ − ( M ℓ / 4 ) q − 1 H 0 ∫ A ∣ u ( y , t ) ∣ ∣ y ∣ N + 2 s d y d t . (69) \int_{-H}^{0}\int_{A}\frac{|\overline{v}(\widetilde{y},\widetilde{t})|}{|\widetilde{y}|^{N+2s}}\,d\widetilde{y}\,d\widetilde{t}=\left(\frac{M_{\ell}}{4}\right)^{-q}\int_{-(M_{\ell}/4)^{q-1}H}^{0}\int_{A}\frac{|u(y,t)|}{|y|^{N+2s}}\,dy\,dt. \tag*{(69)} ∫ − H 0 ∫ A ∣ y ∣ N + 2 s ∣ v ( y , t ) ∣ d y d t = ( 4 M ℓ ) − q ∫ − ( M ℓ /4 ) q − 1 H 0 ∫ A ∣ y ∣ N + 2 s ∣ u ( y , t ) ∣ d y d t . (69)
For almost every time in the interval of Q ^ ℓ \widehat{Q}_{\ell} Q ℓ , the annular decomposition used in (58) gives
∫ K R ∖ K c ϱ ℓ / 2 ∣ u ( y , t ) ∣ ∣ y ∣ N + 2 s d y ≤ C ( M ℓ ϱ ℓ − 2 s + M 0 ϱ 0 − 2 s + ∑ i = 1 ℓ M ℓ − i ϱ ℓ − i − 2 s ) ≤ C M ℓ ϱ ℓ − 2 s . \int_{K_{R}\setminus K_{c\varrho_{\ell}/2}}\frac{|u(y,t)|}{|y|^{N+2s}}\,dy\le C\left(M_{\ell}\varrho_{\ell}^{-2s}+M_{0}\varrho_{0}^{-2s}+\sum_{i=1}^{\ell}M_{\ell-i}\varrho_{\ell-i}^{-2s}\right)\le CM_{\ell}\varrho_{\ell}^{-2s}. ∫ K R ∖ K c ϱ ℓ /2 ∣ y ∣ N + 2 s ∣ u ( y , t ) ∣ d y ≤ C ( M ℓ ϱ ℓ − 2 s + M 0 ϱ 0 − 2 s + i = 1 ∑ ℓ M ℓ − i ϱ ℓ − i − 2 s ) ≤ C M ℓ ϱ ℓ − 2 s .
Indeed, after division by M ℓ ϱ ℓ − 2 s M_{\ell}\varrho_{\ell}^{-2s} M ℓ ϱ ℓ − 2 s , the last two terms are bounded by
( λ 2 s 1 − η ) ℓ + 1 1 − η ∑ i = 1 ℓ ( λ 2 s 1 − η ) ℓ − i , \left(\frac{\lambda^{2s}}{1-\eta}\right)^{\ell}+\frac{1}{1-\eta}\sum_{i=1}^{\ell}\left(\frac{\lambda^{2s}}{1-\eta}\right)^{\ell-i}, ( 1 − η λ 2 s ) ℓ + 1 − η 1 i = 1 ∑ ℓ ( 1 − η λ 2 s ) ℓ − i ,
which is uniformly bounded. For ℓ = 0 \ell=0 ℓ = 0 the sum is empty. Consequently, whenever the corresponding time interval lies in that of Q ^ ℓ \widehat{Q}_{\ell} Q ℓ , (69) yields
M ℓ 4 ∫ − H 0 ∫ K R ∖ K c ϱ ℓ / 2 ∣ v ‾ ( y ~ , t ~ ) ∣ ∣ y ~ ∣ N + 2 s d y ~ d t ~ ≤ C M ℓ H ϱ ℓ − 2 s . (70) \frac{M_{\ell}}{4}\int_{-H}^{0}\int_{K_{R}\setminus K_{c\varrho_{\ell}/2}}\frac{|\overline{v}(\widetilde{y},\widetilde{t})|}{|\widetilde{y}|^{N+2s}}\,d\widetilde{y}\,d\widetilde{t}\le CM_{\ell}H\varrho_{\ell}^{-2s}. \tag*{(70)} 4 M ℓ ∫ − H 0 ∫ K R ∖ K c ϱ ℓ /2 ∣ y ∣ N + 2 s ∣ v ( y , t ) ∣ d y d t ≤ C M ℓ H ϱ ℓ − 2 s . (70)
Consequently,
∫ − C ( R ^ r ) s 0 ∫ R N ∖ K ϵ ℓ / 2 ∣ v ‾ ( y ~ , t ~ ) ∣ ∣ y ~ ∣ N + 2 s d y ~ d t ~ = ( M ℓ 4 ) − q ∫ − ( M ℓ / 4 ) q − 1 C ( R ^ r ) s 0 ∫ R N ∖ K ϵ ℓ / 2 ∣ u ( y , t ) ∣ ∣ y ∣ N + 2 s d y d t = ( M ℓ 4 ) − q ∫ − ( M ℓ / 4 ) q − 1 C ( R ^ r ) s 0 ∫ R N ∖ K R ∣ u ( y , t ) ∣ ∣ y ∣ N + 2 s d y d t + ( M ℓ 4 ) − q ∫ − ( M ℓ / 4 ) q − 1 C ( R ^ r ) s 0 ∫ K R ∖ K ϵ ℓ / 2 ∣ u ( y , t ) ∣ ∣ y ∣ N + 2 s d y d t ≤ ( M ℓ 4 ) − q ∫ − ( M ℓ / 4 ) q − 1 C ( R ^ r ) s 0 ∫ R N ∖ K R ∣ u ( y , t ) ∣ ∣ y ∣ N + 2 s d y d t + C ( r / ϱ ℓ ) s (71) \begin{aligned}
\int_{-C(\widehat{R}r)^s}^{0}\int_{\mathbb{R}^{N}\setminus K_{\epsilon\ell/2}}\frac{|\overline{v}(\widetilde{y},\widetilde{t})|}{|\widetilde{y}|^{N+2s}}\,d\widetilde{y}\,d\widetilde{t}
&=\left(\frac{M_{\ell}}{4}\right)^{-q}\int_{-(M_{\ell}/4)^{q-1}C(\widehat{R}r)^s}^{0}\int_{\mathbb{R}^{N}\setminus K_{\epsilon\ell/2}}\frac{|u(y,t)|}{|y|^{N+2s}}\,dy\,dt\\
&=\left(\frac{M_{\ell}}{4}\right)^{-q}\int_{-(M_{\ell}/4)^{q-1}C(\widehat{R}r)^s}^{0}\int_{\mathbb{R}^{N}\setminus K_{R}}\frac{|u(y,t)|}{|y|^{N+2s}}\,dy\,dt\\
&\quad+\left(\frac{M_{\ell}}{4}\right)^{-q}\int_{-(M_{\ell}/4)^{q-1}C(\widehat{R}r)^s}^{0}\int_{K_{R}\setminus K_{\epsilon\ell/2}}\frac{|u(y,t)|}{|y|^{N+2s}}\,dy\,dt\\
&\leq\left(\frac{M_{\ell}}{4}\right)^{-q}\int_{-(M_{\ell}/4)^{q-1}C(\widehat{R}r)^s}^{0}\int_{\mathbb{R}^{N}\setminus K_{R}}\frac{|u(y,t)|}{|y|^{N+2s}}\,dy\,dt+C(r/\varrho_{\ell})^s \tag*{(71)}
\end{aligned} ∫ − C ( R r ) s 0 ∫ R N ∖ K ϵ ℓ /2 ∣ y ∣ N + 2 s ∣ v ( y , t ) ∣ d y d t = ( 4 M ℓ ) − q ∫ − ( M ℓ /4 ) q − 1 C ( R r ) s 0 ∫ R N ∖ K ϵ ℓ /2 ∣ y ∣ N + 2 s ∣ u ( y , t ) ∣ d y d t = ( 4 M ℓ ) − q ∫ − ( M ℓ /4 ) q − 1 C ( R r ) s 0 ∫ R N ∖ K R ∣ y ∣ N + 2 s ∣ u ( y , t ) ∣ d y d t + ( 4 M ℓ ) − q ∫ − ( M ℓ /4 ) q − 1 C ( R r ) s 0 ∫ K R ∖ K ϵ ℓ /2 ∣ y ∣ N + 2 s ∣ u ( y , t ) ∣ d y d t ≤ ( 4 M ℓ ) − q ∫ − ( M ℓ /4 ) q − 1 C ( R r ) s 0 ∫ R N ∖ K R ∣ y ∣ N + 2 s ∣ u ( y , t ) ∣ d y d t + C ( r / ϱ ℓ ) s (71)
Substituting above inequality into (68) , and rescaling v ‾ \overline{v} v back to u u u yield
κ r × ( − ( M ℓ / 4 ) q − 1 r 2 s , 0 ) u ≤ C M ℓ ( r ϱ ℓ ) β + C ( M ℓ 4 ) 1 − q ∫ − C ( M ℓ / 4 ) q − 1 ( ϱ ℓ r ) s 0 ∫ R N ∖ K R ∣ u ( y , t ) ∣ ∣ y ∣ N + 2 s d y d t . (72) \begin{aligned}
\kappa_{r}\times(- (M_{\ell}/4)^{q-1}r^{2s},0)\quad u \\
&\leq CM_{\ell}\left(\frac{r}{\varrho_{\ell}}\right)^{\beta}+C\left(\frac{M_{\ell}}{4}\right)^{1-q}\int_{-C(M_{\ell}/4)^{q-1}(\varrho_{\ell}r)^s}^{0}\int_{\mathbb{R}^{N}\setminus K_{R}}\frac{|u(y,t)|}{|y|^{N+2s}}\,dy\,dt. \tag*{(72)}
\end{aligned} κ r × ( − ( M ℓ /4 ) q − 1 r 2 s , 0 ) u ≤ C M ℓ ( ϱ ℓ r ) β + C ( 4 M ℓ ) 1 − q ∫ − C ( M ℓ /4 ) q − 1 ( ϱ ℓ r ) s 0 ∫ R N ∖ K R ∣ y ∣ N + 2 s ∣ u ( y , t ) ∣ d y d t . (72)
Here r / ϱ ℓ r/\varrho_{\ell} r / ϱ ℓ is sufficiently small in terms of the data; in particular, the time interval in the integral is contained in that of Q ^ ℓ \widehat{Q}_{\ell} Q ℓ . All constants are independent of the stopping index ℓ \ell ℓ .
An explicit modulus of continuity
Decrease the exponent β \beta β in (72) , if necessary, so that
0 < β < s 2 , 0<\beta<\frac{s}{2}, 0 < β < 2 s ,
2 β log [ λ ( 1 − η ) ( q − 1 ) + / ( 2 s ) ] > log ( 1 − η / 2 ) , 2\beta\log\left[\lambda(1-\eta)^{(q-1)+/(2s)}\right]>\log(1-\eta/2), 2 β log [ λ ( 1 − η ) ( q − 1 ) + / ( 2 s ) ] > log ( 1 − η /2 ) ,
4 β log [ λ ( 1 − η ) ( q − 1 ) + / ( 2 s ) ] > log ( γ ~ λ 2 s ) . (73) 4\beta\log\left[\lambda(1-\eta)^{(q-1)+/(2s)}\right]>\log(\widetilde{\gamma}\lambda^{2s}). \tag*{(73)} 4 β log [ λ ( 1 − η ) ( q − 1 ) + / ( 2 s ) ] > log ( γ λ 2 s ) . (73)
Such a choice is possible since both expressions on the right are negative. These inequalities also imply 1 − β ( q − 1 ) / ( 4 s ) > 0 1-\beta(q-1)/(4s)>0 1 − β ( q − 1 ) / ( 4 s ) > 0 when q ≥ 1 q\geq1 q ≥ 1 .
If the iteration stops at ℓ \ell ℓ , we set M k = M ℓ M_k=M_{\ell} M k = M ℓ for k > ℓ k>\ell k > ℓ when defining the remaining cylinders Q k Q_k Q k . Then
( 1 − η ) k M 0 ≤ M k ≤ γ ~ M 0 , θ k + 1 ϱ k + 1 2 s ≤ γ ~ λ 2 s θ k ϱ k 2 s < θ k ϱ k 2 s , (1-\eta)^kM_0\leq M_k\leq\widetilde{\gamma}M_0,\qquad\theta_{k+1}\varrho_{k+1}^{2s}\leq\widetilde{\gamma}\lambda^{2s}\theta_k\varrho_k^{2s}<\theta_k\varrho_k^{2s}, ( 1 − η ) k M 0 ≤ M k ≤ γ M 0 , θ k + 1 ϱ k + 1 2 s ≤ γ λ 2 s θ k ϱ k 2 s < θ k ϱ k 2 s ,
and the cylinders remain nested. This continuation is used only to select the index below; (63) will be used only at indices not exceeding ℓ \ell ℓ .
Fix r ∈ ( 0 , ϱ 0 ) r\in(0,\varrho_0) r ∈ ( 0 , ϱ 0 ) . In what follows, we will choose j j j by partitioning the radii when 0 < q < 1 0<q<1 0 < q < 1 , and the cylinder heights when q ≥ 1 q\geq1 q ≥ 1 . The observation cylinder will be Q σ ϱ ( ϱ 0 ) Q_{\sigma\varrho}(\varrho_0) Q σ ϱ ( ϱ 0 ) , where σ ∈ ( 0 , 1 ) \sigma\in(0,1) σ ∈ ( 0 , 1 ) is a structural constant. Now, we take on the task.
Case 1: 0 < q < 1 0<q<1 0 < q < 1 . Choose j ≥ 0 j\geq0 j ≥ 0 so that
ϱ j + 1 < r ≤ ϱ j . (74) \varrho_{j+1}<r\leq\varrho_j. \tag*{(74)} ϱ j + 1 < r ≤ ϱ j . (74)
Since M j ≤ γ ~ M 0 M_j \leq\widetilde{\gamma}M_0 M j ≤ γ M 0 and q − 1 < 0 q-1<0 q − 1 < 0 ,
θ j = M j q − 1 ≥ γ ~ q − 1 θ 0 . \theta_j=M_j^{q-1}\geq\widetilde{\gamma}^{q-1}\theta_0. θ j = M j q − 1 ≥ γ q − 1 θ 0 .
Choose σ 2 s ≤ γ ~ q − 1 \sigma^{2s}\leq\widetilde{\gamma}^{q-1} σ 2 s ≤ γ q − 1 . The spatial and temporal inclusions then follow from
σ r ≤ ϱ j , θ 0 ( σ r ) 2 s ≤ θ j ϱ j 2 s . \sigma r\leq\varrho_j,\qquad\theta_0(\sigma r)^{2s}\leq\theta_j\varrho_j^{2s}. σ r ≤ ϱ j , θ 0 ( σ r ) 2 s ≤ θ j ϱ j 2 s .
Thus Q σ r ( θ 0 ) ⊂ Q j Q_{\sigma r}(\theta_0)\subset Q_j Q σ r ( θ 0 ) ⊂ Q j . The other inequality in (74) gives r / ϱ 0 > λ j + 1 r/\varrho_0>\lambda^{j+1} r / ϱ 0 > λ j + 1 .
Case 2: q ≥ 1 q\geq1 q ≥ 1 . The heights θ j ϱ j 2 s \theta_j\varrho_j^{2s} θ j ϱ j 2 s decrease to zero. Choose j ≥ 0 j\geq0 j ≥ 0 so that
θ j + 1 ϱ j + 1 2 s < θ 0 r 2 s ≤ θ j ϱ j 2 s . (75) \theta_{j+1}\varrho_{j+1}^{2s}<\theta_0r^{2s}\leq\theta_j\varrho_j^{2s}. \tag*{(75)} θ j + 1 ϱ j + 1 2 s < θ 0 r 2 s ≤ θ j ϱ j 2 s . (75)
The right-hand inequality and M j ≤ γ ~ M 0 M_j\leq\widetilde{\gamma}M_0 M j ≤ γ M 0 give
r ≤ γ ~ ( q − 1 ) / ( 2 s ) ϱ j . r\leq\widetilde{\gamma}^{(q-1)/(2s)}\varrho_j. r ≤ γ ( q − 1 ) / ( 2 s ) ϱ j .
Choose σ ≤ γ ~ ( 1 − q ) / ( 2 s ) \sigma\leq\widetilde{\gamma}^{(1-q)/(2s)} σ ≤ γ ( 1 − q ) / ( 2 s ) . Then σ r ≤ ϱ j \sigma r\leq\varrho_j σ r ≤ ϱ j and θ 0 ( σ r ) 2 s ≤ θ j ϱ j 2 s \theta_0(\sigma r)^{2s}\leq\theta_j\varrho_j^{2s} θ 0 ( σ r ) 2 s ≤ θ j ϱ j 2 s , so again Q σ r ( θ 0 ) ⊂ Q j Q_{\sigma r}(\theta_0)\subset Q_j Q σ r ( θ 0 ) ⊂ Q j . The left-hand inequality, together with M j + 1 ≥ ( 1 − η ) j + 1 M 0 M_{j+1}\geq(1-\eta)^{j+1}M_0 M j + 1 ≥ ( 1 − η ) j + 1 M 0 , yields
θ 0 r 2 s > θ 0 ϱ 0 2 s [ ( 1 − η ) q − 1 λ 2 s ] j + 1 . \theta_0r^{2s}>\theta_0\varrho_0^{2s}\left[(1-\eta)^{q-1}\lambda^{2s}\right]^{j+1}. θ 0 r 2 s > θ 0 ϱ 0 2 s [ ( 1 − η ) q − 1 λ 2 s ] j + 1 .
For q = 1 q=1 q = 1 , all θ j = 1 \theta_j=1 θ j = 1 , and (75) is exactly the radius partition (74) .
In either case, we have
Q σ r ( θ 0 ) ⊂ Q j , r ϱ 0 > [ λ ( 1 − η ) ( q − 1 ) / ( 2 s ) ] j + 1 . (76) Q_{\sigma r}(\theta_0)\subset Q_j,\qquad\frac{r}{\varrho_0}>\left[\lambda(1-\eta)^{(q-1)/(2s)}\right]^{j+1}. \tag*{(76)} Q σ r ( θ 0 ) ⊂ Q j , ϱ 0 r > [ λ ( 1 − η ) ( q − 1 ) / ( 2 s ) ] j + 1 . (76)
We may decrease σ \sigma σ further below. Put i = ⌊ j / 2 ⌋ i=\lfloor j/2\rfloor i = ⌊ j /2 ⌋ , and first take r / ϱ 0 r/\varrho_0 r / ϱ 0 sufficiently small that i ≥ 2 i\geq2 i ≥ 2 . Now we consider two possibilities.
The iteration reaches i i i . Suppose that there is no stopping index before i i i . By nesting, Q σ r ( θ 0 ) ⊂ Q i Q_{\sigma r}(\theta_0)\subset Q_i Q σ r ( θ 0 ) ⊂ Q i . Using (73) , (76) , and i = ⌊ j / 2 ⌋ i=\lfloor j/2\rfloor i = ⌊ j /2 ⌋ , we obtain
( 1 − η 2 ) i − ⌊ i / 2 ⌋ ≤ C ( 1 − η 2 ) j / 4 ≤ C ( σ r ϱ 0 ) β / 2 . \left(1-\frac{\eta}{2}\right)^{i-\lfloor i/2\rfloor}\leq C\left(1-\frac{\eta}{2}\right)^{j/4}\leq C\left(\frac{\sigma r}{\varrho_0}\right)^{\beta/2}. ( 1 − 2 η ) i − ⌊ i /2 ⌋ ≤ C ( 1 − 2 η ) j /4 ≤ C ( ϱ 0 σ r ) β /2 .
Likewise, for sufficiently small r / ϱ 0 r/\varrho_0 r / ϱ 0 ,
( c ϱ 0 ) 2 s ( γ ~ λ 2 s ) ⌊ i / 2 ⌋ ≤ C ϱ 0 2 s ( γ ~ λ 2 s ) j / 4 ≤ ϱ 0 2 s ( σ r ϱ 0 ) β . (c\varrho_0)^{2s}(\widetilde{\gamma}\lambda^{2s})^{\lfloor i/2\rfloor}\leq C\varrho_0^{2s}(\widetilde{\gamma}\lambda^{2s})^{j/4}\leq\varrho_0^{2s}\left(\frac{\sigma r}{\varrho_0}\right)^\beta. ( c ϱ 0 ) 2 s ( γ λ 2 s ) ⌊ i /2 ⌋ ≤ C ϱ 0 2 s ( γ λ 2 s ) j /4 ≤ ϱ 0 2 s ( ϱ 0 σ r ) β .
The last inequality uses the strict choice of β \beta β in (73) ; σ \sigma σ is a fixed structural constant. Applying (63) at index i i i gives
ess osc Q σ r ( θ 0 ) u ≤ C M 0 ( σ r ϱ 0 ) β / 2 + C ( ∫ − θ 0 ϱ 0 2 s ( σ r / ϱ 0 ) β 0 ∫ R N ∖ K r ∣ u ( y , t ) ∣ ∣ y ∣ N + 2 s d y d t ) 1 / q . \underset{Q_{\sigma r}(\theta_0)}{\operatorname{ess\,osc}}u\leq CM_0\left(\frac{\sigma r}{\varrho_0}\right)^{\beta/2}+C\left(\int_{-\theta_0\varrho_0^{2s}(\sigma r/\varrho_0)^\beta}^{0}\int_{\mathbb{R}^N\setminus K_r}\frac{|u(y,t)|}{|y|^{N+2s}}\,\mathrm{d}y\,\mathrm{d}t\right)^{1/q}. Q σ r ( θ 0 ) ess osc u ≤ C M 0 ( ϱ 0 σ r ) β /2 + C ( ∫ − θ 0 ϱ 0 2 s ( σ r / ϱ 0 ) β 0 ∫ R N ∖ K r ∣ y ∣ N + 2 s ∣ u ( y , t ) ∣ d y d t ) 1/ q .
The iteration stops before i i i . Let ℓ < i \ell<i ℓ < i be the first stopping index, and set
r ~ = σ r max { 1 , ( 4 M 0 M ℓ ) ( q − 1 ) / ( 2 s ) } . \widetilde r=\sigma r\max\left\{1,\left(\frac{4M_0}{M_\ell}\right)^{(q-1)/(2s)}\right\}. r = σ r max { 1 , ( M ℓ 4 M 0 ) ( q − 1 ) / ( 2 s ) } .
Then
Q σ r ( θ 0 ) ⊂ K r ~ × ( − ( M ℓ / 4 ) q − 1 r ~ 2 s , 0 ) , r ~ ≤ C σ r ( M 0 / M ℓ ) ( q − 1 ) / ( 2 s ) . (77) Q_{\sigma r}(\theta_0)\subset K_{\widetilde r}\times\left(-(M_\ell/4)^{q-1}\widetilde r^{2s},0\right),\qquad\widetilde r\leq C\sigma r(M_0/M_\ell)^{(q-1)/(2s)}. \tag*{(77)} Q σ r ( θ 0 ) ⊂ K r × ( − ( M ℓ /4 ) q − 1 r 2 s , 0 ) , r ≤ C σ r ( M 0 / M ℓ ) ( q − 1 ) / ( 2 s ) . (77)
We use the two partitions once more to estimate r ~ / ϱ ℓ \widetilde r/\varrho_\ell r / ϱ ℓ . For 0 < q < 1 0<q<1 0 < q < 1 , (74) implies
r ~ ϱ ℓ ≤ C σ λ j − ℓ ≤ C σ λ j / 2 ≤ C ( r ϱ 0 ) 1 / 2 . (78) \frac{\widetilde r}{\varrho_{\ell}} \le C\sigma\lambda^{j-\ell} \le C\sigma\lambda^{j/2} \le C\left(\frac{r}{\varrho_0}\right)^{1/2}. \tag*{(78)} ϱ ℓ r ≤ C σ λ j − ℓ ≤ C σ λ j /2 ≤ C ( ϱ 0 r ) 1/2 . (78)
Here we used ℓ < i ≤ j / 2 \ell<i\le j/2 ℓ < i ≤ j /2 and r > ϱ j + 1 r>\varrho_{j+1} r > ϱ j + 1 .
For q ≥ 1 q\ge1 q ≥ 1 , the continuation gives M j = M j + 1 = M ℓ M_j=M_{j+1}=M_\ell M j = M j + 1 = M ℓ . Thus (75) reads
( M ℓ M 0 ) ( q − 1 ) / ( 2 s ) λ j + 1 < r ϱ 0 ≤ ( M ℓ M 0 ) ( q − 1 ) / ( 2 s ) λ j . (79) \left(\frac{M_\ell}{M_0}\right)^{(q-1)/(2s)}\lambda^{j+1}<\frac{r}{\varrho_0}\le\left(\frac{M_\ell}{M_0}\right)^{(q-1)/(2s)}\lambda^j. \tag*{(79)} ( M 0 M ℓ ) ( q − 1 ) / ( 2 s ) λ j + 1 < ϱ 0 r ≤ ( M 0 M ℓ ) ( q − 1 ) / ( 2 s ) λ j . (79)
Its upper bound and (77) yield
r ~ ϱ ℓ ≤ C σ λ j − ℓ . \frac{\widetilde r}{\varrho_\ell}\le C\sigma\lambda^{j-\ell}. ϱ ℓ r ≤ C σ λ j − ℓ .
In both cases, we may fix σ ≤ c 2 − ( q − 1 ) / s + 2 \sigma\le c2^{-(q-1)/s+2} σ ≤ c 2 − ( q − 1 ) / s + 2 for (72) to apply at radius r ~ \widetilde r r .
For 0 < q < 1 0<q<1 0 < q < 1 , (78) and M ℓ ≤ γ ~ M 0 M_\ell\le\widetilde\gamma M_0 M ℓ ≤ γ M 0 directly give
M ℓ ( r ~ ϱ ℓ ) β ≤ C M 0 ( r ϱ 0 ) β / 2 . M_\ell\left(\frac{\widetilde r}{\varrho_\ell}\right)^\beta\le CM_0\left(\frac{r}{\varrho_0}\right)^{\beta/2}. M ℓ ( ϱ ℓ r ) β ≤ C M 0 ( ϱ 0 r ) β /2 .
For q ≥ 1 q\ge1 q ≥ 1 , the lower bound in (79) gives
M ℓ ( r ~ ϱ ℓ ) β ≤ C M ℓ λ β ( j − ℓ ) ≤ C M 0 ( r ϱ 0 ) β / 2 ( M ℓ M 0 ) 1 − β ( q − 1 ) / ( 4 s ) λ β ( j / 2 − ℓ ) ≤ C M 0 ( r ϱ 0 ) β / 2 . \begin{aligned}
M_\ell\left(\frac{\widetilde r}{\varrho_\ell}\right)^\beta\le CM_\ell\lambda^{\beta(j-\ell)} \\
&\le CM_0\left(\frac{r}{\varrho_0}\right)^{\beta/2}\left(\frac{M_\ell}{M_0}\right)^{1-\beta(q-1)/(4s)}\lambda^{\beta(j/2-\ell)} \\
&\le CM_0\left(\frac{r}{\varrho_0}\right)^{\beta/2}.
\end{aligned} M ℓ ( ϱ ℓ r ) β ≤ C M ℓ λ β ( j − ℓ ) ≤ C M 0 ( ϱ 0 r ) β /2 ( M 0 M ℓ ) 1 − β ( q − 1 ) / ( 4 s ) λ β ( j /2 − ℓ ) ≤ C M 0 ( ϱ 0 r ) β /2 .
The last step uses ℓ < i ≤ j / 2 \ell<i\le j/2 ℓ < i ≤ j /2 , M ℓ ≤ γ ~ M 0 M_\ell\le\widetilde\gamma M_0 M ℓ ≤ γ M 0 , and 1 − β ( q − 1 ) / ( 4 s ) > 0 1-\beta(q-1)/(4s)>0 1 − β ( q − 1 ) / ( 4 s ) > 0 .
The tail interval in (72) is controlled as follows. For 0 < q < 1 0<q<1 0 < q < 1 , (57) and (78) give
C ( M ℓ / 4 ) q − 1 ( ϱ ℓ r ~ ) s ≤ C θ ϱ 0 2 s ( γ ~ λ 2 s ) ℓ ( r ϱ 0 ) s / 2 . C(M_\ell/4)^{q-1}(\varrho_\ell\widetilde r)^s\le C\theta\varrho_0^{2s}(\widetilde\gamma\lambda^{2s})^\ell\left(\frac{r}{\varrho_0}\right)^{s/2}. C ( M ℓ /4 ) q − 1 ( ϱ ℓ r ) s ≤ C θ ϱ 0 2 s ( γ λ 2 s ) ℓ ( ϱ 0 r ) s /2 .
For q ≥ 1 q\ge1 q ≥ 1 , (77) gives instead
C ( M ℓ / 4 ) q − 1 ( ϱ ℓ r ~ ) s ≤ C θ ϱ 0 2 s ( M ℓ M 0 ) ( q − 1 ) / 2 ( ϱ ℓ ϱ 0 ) s ( r ϱ 0 ) s . C(M_\ell/4)^{q-1}(\varrho_\ell\widetilde r)^s\le C\theta\varrho_0^{2s}\left(\frac{M_\ell}{M_0}\right)^{(q-1)/2}\left(\frac{\varrho_\ell}{\varrho_0}\right)^s\left(\frac{r}{\varrho_0}\right)^s. C ( M ℓ /4 ) q − 1 ( ϱ ℓ r ) s ≤ C θ ϱ 0 2 s ( M 0 M ℓ ) ( q − 1 ) /2 ( ϱ 0 ϱ ℓ ) s ( ϱ 0 r ) s .
Since β < s / 2 \beta<s/2 β < s /2 , both bounds imply
C ( M ℓ / 4 ) q − 1 ( ϱ ℓ r ~ ) s ≤ θ ϱ 0 2 s ( σ r ϱ 0 ) β C(M_\ell/4)^{q-1}(\varrho_\ell\widetilde r)^s\le\theta\varrho_0^{2s}\left(\frac{\sigma r}{\varrho_0}\right)^\beta C ( M ℓ /4 ) q − 1 ( ϱ ℓ r ) s ≤ θ ϱ 0 2 s ( ϱ 0 σ r ) β
for all sufficiently small r / ϱ 0 r/\varrho_0 r / ϱ 0 .
For q ≥ 1 q\ge1 q ≥ 1 , compare M ℓ M_\ell M ℓ with
Tail ( w ; Q ( R , θ ϱ 0 2 s ( σ r ϱ 0 ) β ) ) 1 / q . \operatorname{Tail}\left(w;Q\left(R,\theta\varrho_0^{2s}\left(\frac{\sigma r}{\varrho_0}\right)^\beta\right)\right)^{1/q}. Tail ( w ; Q ( R , θ ϱ 0 2 s ( ϱ 0 σ r ) β ) ) 1/ q .
If M ℓ M_\ell M ℓ does not exceed this quantity, use ess osc Q σ r ( θ 0 ) u ≤ 4 M ℓ \operatorname{ess\,osc}_{Q_{\sigma r}(\theta_0)} u \le4M_\ell ess osc Q σ r ( θ 0 ) u ≤ 4 M ℓ . Otherwise,
M ℓ 1 − q Tail ( u ; Q ( R , θ 0 ρ 0 2 s ( σ r ρ 0 ) β ) ) ≤ Tail ( u ; Q ( R , θ 0 ρ 0 2 s ( σ r ρ 0 ) β ) ) 1 / q . M_\ell^{1-q}\operatorname{Tail}\left(u;Q\left(R,\theta_0\rho_0^{2s}\left(\frac{\sigma r}{\rho_0}\right)^\beta\right)\right) \le\operatorname{Tail}\left(u;Q\left(R,\theta_0\rho_0^{2s}\left(\frac{\sigma r}{\rho_0}\right)^\beta\right)\right)^{1/q}. M ℓ 1 − q Tail ( u ; Q ( R , θ 0 ρ 0 2 s ( ρ 0 σ r ) β ) ) ≤ Tail ( u ; Q ( R , θ 0 ρ 0 2 s ( ρ 0 σ r ) β ) ) 1/ q .
For 0 < q < 1 0<q<1 0 < q < 1 , the tail term in (72) is bounded using M ℓ 1 − q ≤ C M 0 1 − q M_\ell^{1-q}\le CM_0^{1-q} M ℓ 1 − q ≤ C M 0 1 − q . Together with the spatial estimates, this treats every ℓ < i \ell<i ℓ < i , including ℓ = 0 \ell=0 ℓ = 0 .
The normalized tail is at most one by (49) . Consequently, both possibilities give
ess osc Q σ r ( θ 0 ) u ≤ C M 0 ( σ r ρ 0 ) β / 2 + C M 0 [ 1 M 0 q ∫ − θ 0 ρ 0 2 s ( σ r / ρ 0 ) β 0 ∫ R N ∖ K R ∣ u ( y , t ) ∣ ∣ y ∣ N + 2 s d y d t ] min { 1 , 1 / q } . \operatorname{ess\,osc}_{Q_{\sigma r}(\theta_0)}u \le CM_0\left(\frac{\sigma r}{\rho_0}\right)^{\beta/2}
+CM_0\left[\frac{1}{M_0^q}\int_{-\theta_0\rho_0^{2s}(\sigma r/\rho_0)^\beta}^{0}\int_{\mathbb{R}^N\setminus K_R}\frac{|u(y,t)|}{|y|^{N+2s}}\,dy\,dt\right]^{\min\{1,1/q\}}. ess osc Q σ r ( θ 0 ) u ≤ C M 0 ( ρ 0 σ r ) β /2 + C M 0 [ M 0 q 1 ∫ − θ 0 ρ 0 2 s ( σ r / ρ 0 ) β 0 ∫ R N ∖ K R ∣ y ∣ N + 2 s ∣ u ( y , t ) ∣ d y d t ] m i n { 1 , 1/ q } .
Here the fixed spatial factors involving σ \sigma σ have been absorbed into C C C . Relabeling the observation radius σ r \sigma r σ r as r r r proves (4) for r < σ ρ 0 r<\sigma\rho_0 r < σ ρ 0 , after translation to ( x 0 , t 0 ) (x_0,t_0) ( x 0 , t 0 ) . For the remaining σ ρ 0 ≤ r < ρ 0 \sigma\rho_0\le r<\rho_0 σ ρ 0 ≤ r < ρ 0 , the estimate follows from ess osc Q u ≤ 4 M 0 \operatorname{ess\,osc}_{Q}u\le4M_0 ess osc Q u ≤ 4 M 0 after increasing C C C . The initial radius ρ 0 \rho_0 ρ 0 has remained fixed. This completes the proof. □
Declaration of the use of AI: The author used ChatGPT (OpenAI) to assist with language polishing, manuscript formatting and readability improvements. All AI-assisted content was independently verified by the author, and the use of AI did not affect the mathematical content or conclusions. The author takes full responsibility for the manuscript. School of Mathematical Sciences, University of Electronic Science and Technology of China, 2006 Xinyuan Avenue, 611731, Chengdu, P.R. China,
Email address: zhengli@uestc.edu.cn