Introduction
A special case of our main result is as follows. Let s ∈ ( 0 , 1 ) s \in(0,1) s ∈ ( 0 , 1 ) and u u u be a local weak solution to the fractional heat equation
∂ t u + ( − Δ ) s u = 0 in Q 4 = B 4 × ( − 4 2 s , 0 ] . \partial_t u + (-\Delta)^s u = 0 \quad\text{in } Q_4 = B_4 \times(-4^{2s},0]. ∂ t u + ( − Δ ) s u = 0 in Q 4 = B 4 × ( − 4 2 s , 0 ] .
Then, its spatial derivatives D i u D_i u D i u , i ∈ { 1 , … , N } i \in\{1,\ldots,N\} i ∈ { 1 , … , N } are locally continuous in Q 4 Q_4 Q 4 . Moreover, we have the local boundedness estimate:
sup Q 1 ∣ D u ∣ ≲ ∫ − 2 2 s 0 ∫ R N ∣ u ( x , t ) ∣ ( 1 + ∣ x ∣ ) N + 2 s d x d t , (1) \sup_{Q_1} |Du| \lesssim\int_{-2^{2s}}^0 \int_{\mathbb{R}^N} \frac{|u(x,t)|}{(1+|x|)^{N+2s}} \, dx\,dt, \tag*{(1)} Q 1 sup ∣ D u ∣ ≲ ∫ − 2 2 s 0 ∫ R N ( 1 + ∣ x ∣ ) N + 2 s ∣ u ( x , t ) ∣ d x d t , (1)
and the modulus of continuity at each point ( x o , t o ) ∈ Q 1 (x_o,t_o) \in Q_1 ( x o , t o ) ∈ Q 1 :
osc Q r ( x o , t o ) D i u ≲ ω r β + ∫ t o − r 2 s t o ∫ R N ∣ u ( x , t ) ∣ ( 1 + ∣ x − x o ∣ ) N + 2 s d x d t (2) \underset{Q_r(x_o,t_o)}{\operatorname{osc}} D_i u \lesssim\omega r^\beta+ \int_{t_o-r^{2s}}^{t_o}\int_{\mathbb{R}^N} \frac{|u(x,t)|}{(1+|x-x_o|)^{N+2s}}\,\mathrm{d}x\mathrm{d}t \tag*{(2)} Q r ( x o , t o ) osc D i u ≲ ω r β + ∫ t o − r 2 s t o ∫ R N ( 1 + ∣ x − x o ∣ ) N + 2 s ∣ u ( x , t ) ∣ d x d t (2)
for any r ∈ ( 0 , 1 ) r \in(0,1) r ∈ ( 0 , 1 ) , where β ∈ ( 0 , 1 ) \beta\in(0,1) β ∈ ( 0 , 1 ) depends on { s , N } \{s,N\} { s , N } and
ω = ∫ t o − 2 2 s t o ∫ R N ∣ u ( x , t ) ∣ ( 1 + ∣ x − x o ∣ ) N + 2 s d x d t . \omega= \int_{t_o-2^{2s}}^{t_o}\int_{\mathbb{R}^N} \frac{|u(x,t)|}{(1+|x-x_o|)^{N+2s}}\,\mathrm{d}x\mathrm{d}t. ω = ∫ t o − 2 2 s t o ∫ R N ( 1 + ∣ x − x o ∣ ) N + 2 s ∣ u ( x , t ) ∣ d x d t .
A distinctive feature of nonlocal equations is that the local behavior of solutions is influenced by their values at arbitrarily large distances. Consequently, any local regularity estimate must necessarily involve suitable global quantities measuring the long-range interactions encoded by the operator. For the fractional heat equation, the weighted space-time integral appearing on the right-hand side of (1) –(2) is precisely the natural quantity entering the definition of local weak solutions. The main point of the present paper is that this quantity completely controls their interior regularity up to C 1 C^1 C 1 -level.
Kassmann and Weidner [14 ] proved the local boundedness estimate of the form (1) for solutions to a general equation modeled on the fractional heat equation, having measurable coefficients; see also [6 ] in a different context. They also established Hölder continuity of solutions under the stronger assumption of L 1 + ε L^{1+\varepsilon} L 1 + ε -time-integrability and showed that Hölder continuity may fail under the natural L 1 L^1 L 1 -time-integrability assumption. This left open the question of whether continuity itself still follows from the minimal integrability required in the weak formulation. The question was answered affirmatively in [18 ] , where a quantitative modulus of continuity of the form (2) was established for solutions.
The present paper shows that the same natural global quantity governs the regularity of the gradient of solutions. More precisely, under the minimal L 1 L^1 L 1 -time-integrability assumption, we prove that the gradient exists, is continuous, and satisfies quantitative local estimates depending only on the corresponding tail of the solution. Thus, no additional global assumptions are required to obtain interior regularity up to C 1 C^1 C 1 -level. Beyond the linear fractional heat equation, the arguments developed here apply to a class of nonlinear nonlocal parabolic equations, for which they provide a unified gradient regularity theory.
Main results
Let Ω ⊂ R N \Omega\subset\mathbb{R}^N Ω ⊂ R N , N ≥ 2 N \ge2 N ≥ 2 be an open and bounded set; for T > 0 T>0 T > 0 define the space-time cylinder Ω T : = Ω × ( 0 , T ] \Omega_T := \Omega\times(0,T] Ω T := Ω × ( 0 , T ] . Consider the nonlinear nonlocal parabolic equation
∂ t u ( x , t ) + L u ( x , t ) = 0 in Ω T , (3) \partial_t u(x,t) + \mathcal{L}u(x,t) = 0 \quad\text{in } \Omega_T, \tag*{(3)} ∂ t u ( x , t ) + L u ( x , t ) = 0 in Ω T , (3)
where
L u ( x , t ) : = p.v. ∫ R N Φ ( u ( x , t ) − u ( y , t ) ∣ x − y ∣ s ) d y ∣ x − y ∣ N + s \mathcal{L}u(x,t) := \operatorname{p.v.}\int_{\mathbb{R}^N} \Phi\left(\frac{u(x,t)-u(y,t)}{|x-y|^s}\right)\frac{\mathrm{d}y}{|x-y|^{N+s}} L u ( x , t ) := p.v. ∫ R N Φ ( ∣ x − y ∣ s u ( x , t ) − u ( y , t ) ) ∣ x − y ∣ N + s d y
for some s ∈ ( 0 , 1 ) s \in(0,1) s ∈ ( 0 , 1 ) , and the nonlinearity Φ : R → R \Phi:\mathbb{R}\to\mathbb{R} Φ : R → R is an odd function such that for any ξ 1 , ξ 2 ∈ R \xi_1,\xi_2 \in\mathbb{R} ξ 1 , ξ 2 ∈ R and some Λ ≥ 1 \Lambda\ge1 Λ ≥ 1 , we have the growth and coercivity conditions:
∣ Φ ( ξ 1 ) − Φ ( ξ 2 ) ∣ ≤ Λ ∣ ξ 1 − ξ 2 ∣ |\Phi(\xi_1)-\Phi(\xi_2)| \le\Lambda|\xi_1-\xi_2| ∣Φ ( ξ 1 ) − Φ ( ξ 2 ) ∣ ≤ Λ∣ ξ 1 − ξ 2 ∣
( Φ ( ξ 1 ) − Φ ( ξ 2 ) ) ( ξ 1 − ξ 2 ) ≥ Λ − 1 ∣ ξ 1 − ξ 2 ∣ 2 . (4) (\Phi(\xi_1)-\Phi(\xi_2))(\xi_1-\xi_2) \ge\Lambda^{-1}|\xi_1-\xi_2|^2. \tag*{(4)} ( Φ ( ξ 1 ) − Φ ( ξ 2 )) ( ξ 1 − ξ 2 ) ≥ Λ − 1 ∣ ξ 1 − ξ 2 ∣ 2 . (4)
Our main result is as follows.
Theorem 1.1. Let s ∈ ( 0 , 1 ) s \in(0,1) s ∈ ( 0 , 1 ) and let u u u be a local weak solution to (3) –(4) in the sense of Definition 2.1. Then, its spatial derivatives D i u D_i u D i u , i ∈ { 1 , … , N } i \in\{1,\ldots,N\} i ∈ { 1 , … , N } are locally continuous functions in Ω T \Omega_T Ω T . Moreover, there exist C > 1 C>1 C > 1 and β ∈ ( 0 , 1 ) \beta\in(0,1) β ∈ ( 0 , 1 ) depending on { s , N , Λ } \{s,N,\Lambda\} { s , N , Λ } , such that for any Q 4 R ( y , τ ) ⊂ Ω T Q_{4R}(y,\tau) \subset\Omega_T Q 4 R ( y , τ ) ⊂ Ω T , we have the local boundedness estimate:
sup Q R ( y , τ ) ∣ D u ∣ ≤ C R ∫ τ − ( 2 R ) 2 s τ ∫ R N ∣ u ( x , t ) ∣ ( R + ∣ x ∣ ) N + 2 s d x d t . \sup_{Q_R(y,\tau)} |Du| \le\frac{C}{R}\int_{\tau-(2R)^{2s}}^\tau\int_{\mathbb{R}^N} \frac{|u(x,t)|}{(R+|x|)^{N+2s}}\,\mathrm{d}x\mathrm{d}t. Q R ( y , τ ) sup ∣ D u ∣ ≤ R C ∫ τ − ( 2 R ) 2 s τ ∫ R N ( R + ∣ x ∣ ) N + 2 s ∣ u ( x , t ) ∣ d x d t .
and the modulus of continuity at each point ( x o , t o ) ∈ Q R ( y , τ ) (x_o,t_o)\in Q_R(y,\tau) ( x o , t o ) ∈ Q R ( y , τ ) :
osc Q r ( x o , t o ) D i u ≤ C [ ω ( r R ) β + 1 R ∫ t o − ( r R ) 2 s t o ∫ R N ∣ u ( x , t ) ∣ ( R + ∣ x − x o ∣ ) N + 2 s d x d t ] \underset{Q_r(x_o,t_o)}{\operatorname{osc}}\,D_i u\le C\left[\omega\left(\frac{r}{R}\right)^\beta+\frac{1}{R}\int_{t_o-(rR)^{2s}}^{t_o}\int_{\mathbb{R}^N}\frac{|u(x,t)|}{(R+|x-x_o|)^{N+2s}}\,\mathrm{d}x\mathrm{d}t\right] Q r ( x o , t o ) osc D i u ≤ C [ ω ( R r ) β + R 1 ∫ t o − ( r R ) 2 s t o ∫ R N ( R + ∣ x − x o ∣ ) N + 2 s ∣ u ( x , t ) ∣ d x d t ]
for any r ∈ ( 0 , R ) r\in(0,R) r ∈ ( 0 , R ) , where
ω = 1 R ∫ t o − ( 4 R ) 2 s t o ∫ R N ∣ u ( x , t ) ∣ ( R + ∣ x − x o ∣ ) N + 2 s d x d t . (5) \omega=\frac{1}{R}\int_{t_o-(4R)^{2s}}^{t_o}\int_{\mathbb{R}^N}\frac{|u(x,t)|}{(R+|x-x_o|)^{N+2s}}\,\mathrm{d}x\mathrm{d}t. \tag*{(5)} ω = R 1 ∫ t o − ( 4 R ) 2 s t o ∫ R N ( R + ∣ x − x o ∣ ) N + 2 s ∣ u ( x , t ) ∣ d x d t . (5)
Remark 1.2. This result can be viewed as a nonlocal analogue of the well-known C 1 , α C^{1,\alpha} C 1 , α -regularity of local weak solutions to the quasilinear parabolic equation
∂ t u − div A ( D u ) = 0 \partial_tu-\operatorname{div} A(Du)=0 ∂ t u − div A ( D u ) = 0
where the C 1 C^1 C 1 -vector field A : R N → R N \mathbf{A}:\mathbb{R}^N\to\mathbb{R}^N A : R N → R N satisfies proper coercivity and growth conditions; cf. [15 Chapter V, § 3, Theorem 3.1] .
Remark 1.3. The obtained C 1 C^1 C 1 -regularity is sort of “end-point” regularity. Under a more stringent L 1 + ε L^{1+\varepsilon} L 1 + ε -time-integrability of the tail, Theorem 1.1 yields C 1 , α C^{1,\alpha} C 1 , α -regularity, and hence recovers [10 Theorem 1.6] . Our study also gives further improvement, which will appear elsewhere.
Remark 1.4. Nonlocal higher regularity in the time variable can be studied under additional conditions. However, under the minimal tail condition, one cannot expect further regularity in the time variable beyond absolute continuity.
The strategy
The method of difference quotient, a traditional barehanded technique in the analysis of partial differential equations, has found new applications in various nonlocal problems. See [1 , 2 , 3 , 4 , 5 , 12 , 10 , 11 ] for instance.
In a previous work [10 ] , the authors, within the same framework as this paper, assumed additional time-integrability in the tail condition and thereby established C 1 , α C^{1,\alpha} C 1 , α -regularity. Roughly speaking, they considered the difference quotient τ h u / ∣ h ∣ γ \tau_hu/|h|^\gamma τ h u /∣ h ∣ γ , γ ∈ ( 0 , 1 ] \gamma\in(0,1] γ ∈ ( 0 , 1 ] , and leveraged available Hölder estimates for linear equations with measurable kernels, which hold only when the additional tail condition is given, thereby inductively increasing the Hölder exponent of u u u . The method is built on finite iterations in Hölder spaces. Therefore, we cannot adapt their scheme to attack the “end-point” gradient regularity under the natural tail condition.
To reach our goal, we must insist on Sobolev spaces, examine the equation satisfied by τ h u / ∣ h ∣ \tau_hu/|h| τ h u /∣ h ∣ , and perform various more refined iterative arguments within a framework capable of handling the natural tail. In short, our regularity-upgrade-scheme can be illustrated by
L t 2 W x s , 2 ∩ L t 1 L x 2 s ⟶ L t 2 W x 1 , 2 ∩ L t 1 L x 2 s ⟶ L t ∞ W x 1 , ∞ ∩ L t 1 L x 2 s . L_t^{2}W_x^{s,2}\cap L_t^{1}L_x^{2s}\longrightarrow L_t^{2}W_x^{1,2}\cap L_t^{1}L_x^{2s}\longrightarrow L_t^{\infty}W_x^{1,\infty}\cap L_t^{1}L_x^{2s}. L t 2 W x s , 2 ∩ L t 1 L x 2 s ⟶ L t 2 W x 1 , 2 ∩ L t 1 L x 2 s ⟶ L t ∞ W x 1 , ∞ ∩ L t 1 L x 2 s .
In the first step, only the fractional differentiability is improved while the integrability in both time and space remains the same. This step adapts some arguments from [1 , 2 ] . In the second step, we use De Giorgi’s iteration to show that ∥ τ h u / ∣ h ∣ ∥ L t ∞ L x ∞ \|\tau_hu/|h|\|_{L_t^\infty L_x^\infty} ∥ τ h u /∣ h ∣ ∥ L t ∞ L x ∞ can be bounded by ∥ τ h u / ∣ h ∣ ∥ L t 2 L x 2 \|\tau_hu/|h|\|_{L_t^2L_x^2} ∥ τ h u /∣ h ∣ ∥ L t 2 L x 2 and the natural tail of u u u , thereby establishing the spatial Lipschitz regularity.
Continuity of the gradient entails further delicate analysis. Again, we examine the equation satisfied by τ h u / ∣ h ∣ \tau_hu/|h| τ h u /∣ h ∣ , use some techniques developed in [17 , 18 ] , and derive a modulus of continuity of τ h u / ∣ h ∣ \tau_hu/|h| τ h u /∣ h ∣ that is uniform in h h h . Passing to the limit, we eventually achieve the desired modulus of continuity for D u Du D u .
Last but not least, we would like to point out that even in the elliptic case, our route of upgrading the regularity from W s , 2 W^{s,2} W s , 2 to C 1 , α C^{1,\alpha} C 1 , α is new, comparing to [11 ] within the same framework. The latter again features finite iterations in Hölder spaces. Moreover, we do not use so-called “localization” techniques, but rather handle the tail terms directly.
Acknowledgments. This research was set out in May, 2025. No AI tools were used in the process. It was funded in whole or in part by the Austrian Science Fund (FWF) [10.55776/PAT1850524]. For open access purposes, the author has applied a CC BY public copyright license to any author accepted manuscript version arising from this submission.
Acknowledgments
Notation and definitions
Our notation is standard in the field. We denote the ball in R N \mathbb{R}^{N} R N with radius R R R and center x o x_o x o by B R ( x o ) B_R(x_o) B R ( x o ) ; a space-time cylinder at z o = ( x o , t o ) z_o=(x_o,t_o) z o = ( x o , t o ) is
Q R , S ( z o ) : = B R ( x o ) × ( t o − S , t o ] Q_{R,S}(z_o):=B_R(x_o)\times(t_o-S,t_o] Q R , S ( z o ) := B R ( x o ) × ( t o − S , t o ]
and from time to time, we also use
Q R , S ( z o ) : = B R ( x o ) × B R ( x o ) × ( t o − S , t o ] . \mathbb{Q}_{R,S}(z_o):=B_R(x_o)\times B_R(x_o)\times(t_o-S,t_o]. Q R , S ( z o ) := B R ( x o ) × B R ( x o ) × ( t o − S , t o ] .
The parabolic cylinders are Q R ( z o ) : = Q R , R 2 s ( z o ) Q_R(z_o):=Q_{R,R^{2s}}(z_o) Q R ( z o ) := Q R , R 2 s ( z o ) and Q R ( z o ) : = Q R , R 2 s ( z o ) \mathbb{Q}_R(z_o):=\mathbb{Q}_{R,R^{2s}}(z_o) Q R ( z o ) := Q R , R 2 s ( z o ) .
For s ∈ ( 0 , 1 ) s\in(0,1) s ∈ ( 0 , 1 ) define the weighted Lebesgue space
L 2 s 1 ( R N ) : = { u ∈ L l o c 1 ( R N ) : ∫ R N ∣ u ( x ) ∣ 1 + ∣ x ∣ N + 2 s d x < ∞ } L^1_{2s}(\mathbb{R}^{N}):=\left\{u\in L^1_{\mathrm{loc}}(\mathbb{R}^{N}):\int_{\mathbb{R}^{N}}\frac{|u(x)|}{1+|x|^{N+2s}}\,dx<\infty\right\} L 2 s 1 ( R N ) := { u ∈ L loc 1 ( R N ) : ∫ R N 1 + ∣ x ∣ N + 2 s ∣ u ( x ) ∣ d x < ∞ }
The global behavior of a function belonging to the space L 2 s 1 ( R N ) L^1_{2s}(\mathbb{R}^{N}) L 2 s 1 ( R N ) can be measured by the tail, which is defined by
Tail ( u ( t ) ; B R ( x o ) ) : = R 2 s ∫ R N ∖ B R ( x o ) ∣ u ( x , t ) ∣ ∣ x − x o ∣ N + 2 s d x (6) \operatorname{Tail}(u(t);B_R(x_o)):=R^{2s}\int_{\mathbb{R}^{N}\setminus B_R(x_o)}\frac{|u(x,t)|}{|x-x_o|^{N+2s}}\,dx \tag*{(6)} Tail ( u ( t ) ; B R ( x o )) := R 2 s ∫ R N ∖ B R ( x o ) ∣ x − x o ∣ N + 2 s ∣ u ( x , t ) ∣ d x (6)
When dealing with a time-dependent function u ∈ L 1 ( t o − S , t o ; L 2 s 1 ( R N ) ) u\in L^1(t_o-S,t_o;L^1_{2s}(\mathbb{R}^{N})) u ∈ L 1 ( t o − S , t o ; L 2 s 1 ( R N )) we also use
Tail ( u ; Q R , S ( z o ) ) : = ∫ t o − S t o ∫ R N ∖ B R ( x o ) ∣ u ( x , t ) ∣ ∣ x − x o ∣ N + 2 s d x d t . \operatorname{Tail}(u;Q_{R,S}(z_o)):=\int_{t_o-S}^{t_o}\int_{\mathbb{R}^{N}\setminus B_R(x_o)}\frac{|u(x,t)|}{|x-x_o|^{N+2s}}\,dx\,dt. Tail ( u ; Q R , S ( z o )) := ∫ t o − S t o ∫ R N ∖ B R ( x o ) ∣ x − x o ∣ N + 2 s ∣ u ( x , t ) ∣ d x d t .
Moreover, we introduce the abbreviations
T ( t ; B R ( x o ) ) : = ∥ u ( t ) ∥ L ∞ ( B R ( x o ) ) + Tail ( u ( t ) ; B R ( x o ) ) (7) \mathcal{T}(t;B_R(x_o)):=\|u(t)\|_{L^\infty(B_R(x_o))}+\operatorname{Tail}(u(t);B_R(x_o)) \tag*{(7)} T ( t ; B R ( x o )) := ∥ u ( t ) ∥ L ∞ ( B R ( x o )) + Tail ( u ( t ) ; B R ( x o )) (7)
and
T ( Q R , S ( z o ) ) : = ∥ u ∥ L ∞ ( Q R , S ( z o ) ) + Tail ( u ; Q R , S ( z o ) ) . (8) \mathcal{T}(Q_{R,S}(z_o)):=\|u\|_{L^\infty(Q_{R,S}(z_o))}+\operatorname{Tail}(u;Q_{R,S}(z_o)). \tag*{(8)} T ( Q R , S ( z o )) := ∥ u ∥ L ∞ ( Q R , S ( z o )) + Tail ( u ; Q R , S ( z o )) . (8)
In the case the center is the origin or is clear from the context, we omit it.
Throughout this paper, we let
d a t a : = { s , N , Λ } , \mathsf{data}:=\{s,N,\Lambda\}, data := { s , N , Λ } ,
and use C = C ( d a t a ) C=C(\mathsf{data}) C = C ( data ) as a generic positive constant, which can be determined by the data only and can vary from one estimate to another.
We end this section by defining local weak solutions to (1.3)–(1.4) as follows.
Definition 2.1. (Local weak solution) Let s ∈ ( 0 , 1 ) s\in(0,1) s ∈ ( 0 , 1 ) . A function
u ∈ C l o c ( 0 , T ; L l o c 2 ( Ω ) ) ∩ L l o c 2 ( 0 , T ; W l o c s , 2 ( Ω ) ) ∩ L l o c 1 ( 0 , T ; L 2 s 1 ( R N ) ) u\in C_{\mathrm{loc}}(0,T;L^2_{\mathrm{loc}}(\Omega))\cap L^2_{\mathrm{loc}}(0,T;W^{s,2}_{\mathrm{loc}}(\Omega))\cap L^1_{\mathrm{loc}}(0,T;L^1_{2s}(\mathbb{R}^{N})) u ∈ C loc ( 0 , T ; L loc 2 ( Ω )) ∩ L loc 2 ( 0 , T ; W loc s , 2 ( Ω )) ∩ L loc 1 ( 0 , T ; L 2 s 1 ( R N )) is a local, weak solution to (1.3)–(1.4), if for every sub-interval [ t 1 , t 2 ] ⊂ ( 0 , T ] [t_1,t_2]\subset(0,T] [ t 1 , t 2 ] ⊂ ( 0 , T ] , we have
0 = ∫ Ω ϕ u d x ∣ t 1 t 2 − ∫ t 1 t 2 ∫ Ω ∂ t ϕ u d x d t + ∫ t 1 t 2 ∫ R N ∫ R N Φ ( u ( x , t ) − u ( y , t ) ∣ x − y ∣ s ) ϕ ( x , t ) − ϕ ( y , t ) ∣ x − y ∣ N + s d y d x d t (9) 0=\left.\int_{\Omega}\phi u\,dx\right|_{t_1}^{t_2}-\int_{t_1}^{t_2}\int_{\Omega}\partial_t\phi u\,dx\,dt+\int_{t_1}^{t_2}\int_{\mathbb{R}^{N}}\int_{\mathbb{R}^{N}}\Phi\left(\frac{u(x,t)-u(y,t)}{|x-y|^s}\right)\frac{\phi(x,t)-\phi(y,t)}{|x-y|^{N+s}}\,dy\,dx\,dt \tag*{(9)} 0 = ∫ Ω ϕ u d x t 1 t 2 − ∫ t 1 t 2 ∫ Ω ∂ t ϕ u d x d t + ∫ t 1 t 2 ∫ R N ∫ R N Φ ( ∣ x − y ∣ s u ( x , t ) − u ( y , t ) ) ∣ x − y ∣ N + s ϕ ( x , t ) − ϕ ( y , t ) d y d x d t (9) for any ϕ ∈ L l o c 2 ( 0 , T ; W 0 s , 2 ( Ω ) ) ∩ W l o c 1 , 2 ( 0 , T ; L 2 ( Ω ) ) \phi\in L^2_{\mathrm{loc}}(0,T;W^{s,2}_0(\Omega))\cap W^{1,2}_{\mathrm{loc}}(0,T;L^2(\Omega)) ϕ ∈ L loc 2 ( 0 , T ; W 0 s , 2 ( Ω )) ∩ W loc 1 , 2 ( 0 , T ; L 2 ( Ω )) such that spt ϕ ( ⋅ , t ) ⊂ Ω \operatorname{spt}\phi(\cdot,t)\subset\Omega spt ϕ ( ⋅ , t ) ⊂ Ω .
Preliminaries
Algebraic inequalities and auxiliary lemmata
At several stages we will rely on algebraic inequalities.
Lemma 3.1. Let a , b , e , f ∈ R a,b,e,f \in\mathbb{R} a , b , e , f ∈ R . Then we have
( a − b ) ( a e 2 − b f 2 ) ≥ 1 4 ( a − b ) 2 ( e 2 + f 2 ) − 1 2 ( a + b ) 2 ∣ e − f ∣ 2 . (a-b)(ae^2-bf^2) \ge\frac{1}{4}(a-b)^2(e^2+f^2)-\frac{1}{2}(a+b)^2\lvert e-f\rvert^2. ( a − b ) ( a e 2 − b f 2 ) ≥ 4 1 ( a − b ) 2 ( e 2 + f 2 ) − 2 1 ( a + b ) 2 ∣ e − f ∣ 2 .
Proof. The left-hand side can be rewritten as
( a − b ) ( a e 2 − b f 2 ) = 1 2 ( a − b ) 2 ( e 2 + f 2 ) + 1 2 ( a − b ) ( a + b ) ( e 2 − f 2 ) . (a-b)(ae^2-bf^2)=\frac{1}{2}(a-b)^2(e^2+f^2)+\frac{1}{2}(a-b)(a+b)(e^2-f^2). ( a − b ) ( a e 2 − b f 2 ) = 2 1 ( a − b ) 2 ( e 2 + f 2 ) + 2 1 ( a − b ) ( a + b ) ( e 2 − f 2 ) .
The second term on the right-hand side can be bounded as
1 2 ∣ ( a − b ) ( a + b ) ( e 2 − f 2 ) ∣ ≤ 1 8 ( a − b ) 2 ( e + f ) 2 + 1 2 ( a + b ) 2 ( e − f ) 2 ≤ 1 4 ( a − b ) 2 ( e 2 + f 2 ) + 1 2 ( a + b ) 2 ( e − f ) 2 . \begin{aligned}
\frac{1}{2}\lvert(a-b)(a+b)(e^2-f^2)\rvert\le\frac{1}{8}(a-b)^2(e+f)^2+\frac{1}{2}(a+b)^2(e-f)^2 \\
&\le\frac{1}{4}(a-b)^2(e^2+f^2)+\frac{1}{2}(a+b)^2(e-f)^2.
\end{aligned} 2 1 ∣( a − b ) ( a + b ) ( e 2 − f 2 )∣ ≤ 8 1 ( a − b ) 2 ( e + f ) 2 + 2 1 ( a + b ) 2 ( e − f ) 2 ≤ 4 1 ( a − b ) 2 ( e 2 + f 2 ) + 2 1 ( a + b ) 2 ( e − f ) 2 .
Inserting it above yields the claimed inequality. □
An integral estimate
Next, we will introduce an estimate for the tail introduced as in (6) . This result can be found in [4 Lemma 2.3] and will be later applied on each time-slice of our space-time cylinder.
Lemma 3.2. For s ∈ ( 0 , 1 ) s \in(0,1) s ∈ ( 0 , 1 ) , B R ≡ B R ( x 0 ) ⋐ Ω B_R \equiv B_R(x_0) \Subset\Omega B R ≡ B R ( x 0 ) ⋐ Ω and r ∈ ( 0 , R ) r \in(0,R) r ∈ ( 0 , R ) , we have whenever u ∈ L l o c ∞ ( Ω ) ∩ L 2 s 1 ( R N ) u \in L^\infty_{\mathrm{loc}}(\Omega) \cap L^1_{2s}(\mathbb{R}^N) u ∈ L loc ∞ ( Ω ) ∩ L 2 s 1 ( R N ) that
Tail ( u ; B r ) ≤ C ( N ) ( R r ) N [ Tail ( u ; B R ) + ∥ u ∥ L ∞ ( B R ) ] . \operatorname{Tail}(u;B_r) \le C(N)\left(\frac{R}{r}\right)^N\left[\operatorname{Tail}(u;B_R)+\lVert u\rVert_{L^\infty(B_R)}\right]. Tail ( u ; B r ) ≤ C ( N ) ( r R ) N [ Tail ( u ; B R ) + ∥ u ∥ L ∞ ( B R ) ] .
Finite differences and fractional Sobolev spaces
Suppose that Ω ⊂ R N \Omega\subset\mathbb{R}^N Ω ⊂ R N is open and h ∈ R N h \in\mathbb{R}^N h ∈ R N is a direction vector. Then, let Ω h : = { x ∈ Ω : x + h ∈ Ω } \Omega_h:=\{x\in\Omega:x+h\in\Omega\} Ω h := { x ∈ Ω : x + h ∈ Ω } . This enables us to define for a measurable function v : Ω → R v:\Omega\to\mathbb{R} v : Ω → R the finite difference operator τ h : L 1 ( Ω ) → L 1 ( Ω h ) \tau_h:L^1(\Omega)\to L^1(\Omega_h) τ h : L 1 ( Ω ) → L 1 ( Ω h ) as
τ h v ( x , t ) : = v ( x + h , t ) − v ( x , t ) . \tau_hv(x,t):=v(x+h,t)-v(x,t). τ h v ( x , t ) := v ( x + h , t ) − v ( x , t ) .
Instead of v ( x + h , t ) v(x+h,t) v ( x + h , t ) we will often use the abbreviation v h ( x , t ) : = v ( x + h , t ) v_h(x,t):=v(x+h,t) v h ( x , t ) := v ( x + h , t ) . For a fixed unit vector e ∈ R N e\in\mathbb{R}^N e ∈ R N and h ∈ R h\in\mathbb{R} h ∈ R , we introduce the notation
τ h ( e ) v ( x , t ) : = v ( x + h e , t ) − v ( x , t ) \tau_h^{(e)}v(x,t):=v(x+he,t)-v(x,t) τ h ( e ) v ( x , t ) := v ( x + h e , t ) − v ( x , t )
and write τ h ( i ) v \tau_h^{(i)}v τ h ( i ) v in the case of a canonical basis vector e i ∈ R N e_i\in\mathbb{R}^N e i ∈ R N . The following property of finite differences are well known, cf. [9 Proposition 20.3] . For the rest of the section, we will omit the center of balls.
Lemma 3.3. Assume that M > 0 M>0 M > 0 . Then any v ∈ L 2 ( B R ) v\in L^2(B_R) v ∈ L 2 ( B R ) satisfying
( 3.1 ) ∫ B 1 2 R ∣ τ h ( i ) v ∣ 2 d x ≤ M 2 ∣ h ∣ 2 , (3.1)\qquad\int_{B_{\frac{1}{2}R}}\lvert\tau_h^{(i)}v\rvert^2\,dx\le M^2\lvert h\rvert^2, ( 3.1 ) ∫ B 2 1 R ∣ τ h ( i ) v ∣ 2 d x ≤ M 2 ∣ h ∣ 2 , for all 0 < ∣ h ∣ ≤ 1 2 R 0<\lvert h\rvert\le\frac{1}{2}R 0 < ∣ h ∣ ≤ 2 1 R , is weakly differentiable with respect to x i x_i x i on B 1 2 R B_{\frac{1}{2}R} B 2 1 R and
∫ B 1 2 R ∣ D i v ∣ 2 d x ≤ M 2 . \int_{B_{\frac{1}{2}R}}\lvert D_iv\rvert^2\,dx\le M^2. ∫ B 2 1 R ∣ D i v ∣ 2 d x ≤ M 2 . If (3.1) holds for any e i e_i e i , then v ∈ W 1 , 2 ( B 1 2 R ) v\in W^{1,2}(B_{\frac{1}{2}R}) v ∈ W 1 , 2 ( B 2 1 R ) .
The next result displays an estimate for finite differences of W γ , 2 W^{\gamma,2} W γ , 2 -functions, cf. [3 Proposition 2.6] .
Lemma 3.4. For γ ∈ ( 0 , 1 ) \gamma\in(0,1) γ ∈ ( 0 , 1 ) there exists a constant C = C ( N ) C=C(N) C = C ( N ) such that for v ∈ W γ , 2 ( B R ) v \in W^{\gamma,2}(B_R) v ∈ W γ , 2 ( B R ) and h ∈ R N ∖ { 0 } h \in\mathbb{R}^N \setminus\{0\} h ∈ R N ∖ { 0 } with ∣ h ∣ ≤ 1 2 R |h| \le\frac{1}{2}R ∣ h ∣ ≤ 2 1 R we have the estimate
∫ B 1 2 R ∣ τ h v ∣ 2 d x ≤ C ∣ h ∣ 2 γ ( [ v ] W γ , 2 ( B R ) 2 + R − 2 γ ∥ v ∥ L 2 ( B R ) 2 ) . \int_{B_{\frac{1}{2}R}} |\tau_hv|^2\,dx \le C|h|^{2\gamma}\left([v]_{W^{\gamma,2}(B_R)}^2+R^{-2\gamma}\|v\|_{L^2(B_R)}^2\right). ∫ B 2 1 R ∣ τ h v ∣ 2 d x ≤ C ∣ h ∣ 2 γ ( [ v ] W γ , 2 ( B R ) 2 + R − 2 γ ∥ v ∥ L 2 ( B R ) 2 ) .
The following lemma, important in handling second order differences, is essentially contained in [19 Chapter 5] ; see also [1 Lemma 2.17] , [3 Proposition 2.4] , as well as [13 Theorem 1.1] .
Lemma 3.5. Suppose that γ ∈ ( 0 , 1 ) \gamma\in(0,1) γ ∈ ( 0 , 1 ) , and M ≥ 0 M \ge0 M ≥ 0 . Assume that v ∈ L 2 ( B R ) v \in L^2(B_R) v ∈ L 2 ( B R ) satisfies
∫ B 1 2 R ∣ τ h 2 v ∣ 2 d x ≤ M 2 ∣ h ∣ 2 γ , \int_{B_{\frac{1}{2}R}} |\tau_h^2v|^2\,dx \le M^2|h|^{2\gamma}, ∫ B 2 1 R ∣ τ h 2 v ∣ 2 d x ≤ M 2 ∣ h ∣ 2 γ , for all h ∈ R N ∖ { 0 } h \in\mathbb{R}^N \setminus\{0\} h ∈ R N ∖ { 0 } satisfying ∣ h ∣ ≤ 1 4 R |h| \le\frac{1}{4}R ∣ h ∣ ≤ 4 1 R . Then, there exist an absolute constant C C C such that
∫ B 1 2 R ∣ τ h v ∣ 2 d x ≤ C ∣ h ∣ 2 γ [ M 2 ( 1 − γ ) 2 + R − 2 γ ∫ B R ∣ v ∣ 2 d x ] \int_{B_{\frac{1}{2}R}} |\tau_hv|^2\,dx \le C|h|^{2\gamma}\left[\frac{M^2}{(1-\gamma)^2}+R^{-2\gamma}\int_{B_R}|v|^2\,dx\right] ∫ B 2 1 R ∣ τ h v ∣ 2 d x ≤ C ∣ h ∣ 2 γ [ ( 1 − γ ) 2 M 2 + R − 2 γ ∫ B R ∣ v ∣ 2 d x ] for any h ∈ R N ∖ { 0 } h \in\mathbb{R}^N \setminus\{0\} h ∈ R N ∖ { 0 } satisfying 0 < ∣ h ∣ ≤ 1 8 R 0<|h| \le\frac{1}{8}R 0 < ∣ h ∣ ≤ 8 1 R .
At this point we will also introduce a tail estimate for finite differences, cf. [1 Lemma 3.1] . Moreover, note that due to the nature of the given problem we will apply it on each time-slice and with the choice p = 2 p=2 p = 2 .
Lemma 3.6. Let s ∈ ( 0 , 1 ) s \in(0,1) s ∈ ( 0 , 1 ) , R > 0 R>0 R > 0 , and r ∈ ( 0 , R ) r \in(0,R) r ∈ ( 0 , R ) . Consider the function
u ∈ L ∞ ( B 2 R ) ∩ L s 1 ( R N ) . u \in L^\infty(B_{2R}) \cap L_s^1(\mathbb{R}^N). u ∈ L ∞ ( B 2 R ) ∩ L s 1 ( R N ) . Then, there exists a constant C = C ( N , s ) C=C(N,s) C = C ( N , s ) such that for a.e. x ∈ B 1 2 ( R + r ) x \in B_{\frac{1}{2}(R+r)} x ∈ B 2 1 ( R + r ) and any h ∈ R N ∖ { 0 } h \in\mathbb{R}^N \setminus\{0\} h ∈ R N ∖ { 0 } satisfying 0 < ∣ h ∣ ≤ 1 4 ( R − r ) 0<|h| \le\frac{1}{4}(R-r) 0 < ∣ h ∣ ≤ 4 1 ( R − r ) , we have
∣ ∫ R N ∖ B R τ h u ( y ) ∣ x − y ∣ N + 2 s d y ∣ ≤ C ∣ h ∣ R 2 s + 1 ( R R − r ) N + 2 s + 1 T , \left|\int_{\mathbb{R}^N\setminus B_R}\frac{\tau_hu(y)}{|x-y|^{N+2s}}\,dy\right| \le\frac{C|h|}{R^{2s+1}}\left(\frac{R}{R-r}\right)^{N+2s+1}T, ∫ R N ∖ B R ∣ x − y ∣ N + 2 s τ h u ( y ) d y ≤ R 2 s + 1 C ∣ h ∣ ( R − r R ) N + 2 s + 1 T , where
T : = ∥ u ∥ L ∞ ( B 2 R ) + Tail ( u ; B 2 R ) . T:=\|u\|_{L^\infty(B_{2R})}+\operatorname{Tail}(u;B_{2R}). T := ∥ u ∥ L ∞ ( B 2 R ) + Tail ( u ; B 2 R ) .
In the following we state an embedding for fractional Sobolev spaces. For the proof we refer to [17 Proposition A.3] .
Proposition 3.7. Let s ∈ ( 0 , 1 ) s \in(0,1) s ∈ ( 0 , 1 ) and R > 0 R>0 R > 0 . Consider the function
u ∈ L 2 ( t 1 , t 2 ; W s , 2 ( B R ) ) ∩ L ∞ ( t 1 , t 2 ; L 2 ( B R ) ) , u \in L^2(t_1,t_2;W^{s,2}(B_R)) \cap L^\infty(t_1,t_2;L^2(B_R)), u ∈ L 2 ( t 1 , t 2 ; W s , 2 ( B R )) ∩ L ∞ ( t 1 , t 2 ; L 2 ( B R )) , which is supported in B 1 2 R B_{\frac{1}{2}R} B 2 1 R for a.e. t ∈ ( t 1 , t 2 ) t \in(t_1,t_2) t ∈ ( t 1 , t 2 ) . Then,
∫ t 1 t 2 ∫ B R ∣ u ∣ 2 ( N + 2 s ) N d x d t ≤ C [ R 2 s ∫ t 1 t 2 ∫ B R ∫ B R ∣ u ( x , t ) − u ( y , t ) ∣ 2 ∣ x − y ∣ N + 2 s d x d y d t + ∫ t 1 t 2 ∫ B R ∣ u ∣ 2 d x d t ] ⋅ [ ess sup t ∈ ( t 1 , t 2 ) ∫ B R × { t } ∣ u ∣ 2 d x ] 2 s N \int_{t_1}^{t_2}\int_{B_R}|u|^{\frac{2(N+2s)}{N}}\,dxdt
\le C\left[R^{2s}\int_{t_1}^{t_2}\int_{B_R}\int_{B_R}\frac{|u(x,t)-u(y,t)|^2}{|x-y|^{N+2s}}\,dxdydt+\int_{t_1}^{t_2}\int_{B_R}|u|^2\,dxdt\right]
\cdot\left[\operatorname*{ess\,sup}_{t\in(t_1,t_2)}\int_{B_R\times\{t\}}|u|^2\,dx\right]^{\frac{2s}{N}} ∫ t 1 t 2 ∫ B R ∣ u ∣ N 2 ( N + 2 s ) d x d t ≤ C [ R 2 s ∫ t 1 t 2 ∫ B R ∫ B R ∣ x − y ∣ N + 2 s ∣ u ( x , t ) − u ( y , t ) ∣ 2 d x d y d t + ∫ t 1 t 2 ∫ B R ∣ u ∣ 2 d x d t ] ⋅ [ t ∈ ( t 1 , t 2 ) ess sup ∫ B R × { t } ∣ u ∣ 2 d x ] N 2 s holds true for some C = C ( N , s ) C=C(N,s) C = C ( N , s ) .
Energy inequalities
In this section, we present two kinds of energy estimates regarding the finite difference τ h u \tau_h u τ h u . The first one will be used to upgrade the preset W s , 2 W^{s,2} W s , 2 -regularity of weak solutions to the W 1 , 2 W^{1,2} W 1 , 2 -regularity in the space variable.
Lemma 4.1. Let s ∈ ( 0 , 1 ) s \in(0,1) s ∈ ( 0 , 1 ) . There exists a constant C = C ( data ) > 0 C=C(\text{data})>0 C = C ( data ) > 0 such that, if u u u is a locally bounded, local weak solution to (3) in the sense of Definition 2.1, then for any 0 < r < R 0<r<R 0 < r < R and 0 < τ < S 0<\tau<S 0 < τ < S satisfying Q 2 R , S ( z o ) ⋐ Ω T Q_{2R,S}(z_o)\Subset\Omega_T Q 2 R , S ( z o ) ⋐ Ω T , and any h ∈ R N ∖ { 0 } h\in\mathbb{R}^N\setminus\{0\} h ∈ R N ∖ { 0 } with 0 < ∣ h ∣ ≤ 1 4 ( R − r ) 0<|h|\leq\frac{1}{4}(R-r) 0 < ∣ h ∣ ≤ 4 1 ( R − r ) , we have the energy inequality
ess sup t ∈ ( t o − τ , t o ) ∫ B r ∣ τ h u ∣ 2 d x + ∫ Q r , τ ∫ R N ∣ τ h u ( x , t ) − τ h u ( y , t ) ∣ 2 ∣ x − y ∣ N + 2 s d x d y d t ≤ C [ 1 R 2 s ( R R − r ) N + 2 s + 1 S − τ ] ∬ Q R , S ∣ τ h u ∣ 2 d x d t + C ∣ h ∣ 2 R 4 s + 2 ( R R − r ) 2 N + 4 s + 2 R N [ ∫ t o − S t o T ( t ; B 2 R ) d t ] 2 , \begin{aligned}
\operatorname*{ess\,sup}_{t\in(t_o-\tau,t_o)}\int_{B_r}|\tau_hu|^2\,dx+\int_{Q_{r,\tau}}\int_{\mathbb{R}^N}\frac{|\tau_hu(x,t)-\tau_hu(y,t)|^2}{|x-y|^{N+2s}}\,dx\,dy\,dt \\
&\leq C\left[\frac{1}{R^{2s}}\left(\frac{R}{R-r}\right)^{N+2s}+\frac{1}{S-\tau}\right]\iint_{Q_{R,S}}|\tau_hu|^2\,dx\,dt
+\frac{C|h|^2}{R^{4s+2}}\left(\frac{R}{R-r}\right)^{2N+4s+2}R^N\left[\int_{t_o-S}^{t_o}\mathcal{T}(t;B_{2R})\,dt\right]^2,
\end{aligned} t ∈ ( t o − τ , t o ) ess sup ∫ B r ∣ τ h u ∣ 2 d x + ∫ Q r , τ ∫ R N ∣ x − y ∣ N + 2 s ∣ τ h u ( x , t ) − τ h u ( y , t ) ∣ 2 d x d y d t ≤ C [ R 2 s 1 ( R − r R ) N + 2 s + S − τ 1 ] ∬ Q R , S ∣ τ h u ∣ 2 d x d t + R 4 s + 2 C ∣ h ∣ 2 ( R − r R ) 2 N + 4 s + 2 R N [ ∫ t o − S t o T ( t ; B 2 R ) d t ] 2 ,
where we omitted x o x_o x o , z o z_o z o and defined T ( t ; B 2 R ) \mathcal{T}(t;B_{2R}) T ( t ; B 2 R ) as in (7) .
Proof. First, consider the weak formulation (9) in [ t o − S , t 1 ] [t_o-S,t_1] [ t o − S , t 1 ] for some t 1 ∈ ( t o − S , t o ] t_1\in(t_o-S,t_o] t 1 ∈ ( t o − S , t o ] . By testing the weak formulation with τ − h ϕ ( x , t ) \tau_{-h}\phi(x,t) τ − h ϕ ( x , t ) and applying discrete integration by parts, we obtain
0 = ∫ Ω ϕ τ h u d x ∣ t o − S t 1 − ∫ t o − S t 1 ∫ Ω ∂ t ϕ τ h u d x d t + ∫ t o − S t 1 ∬ R N × R N ( Φ ( u h ( x , t ) − u h ( y , t ) ∣ x − y ∣ s ) − Φ ( u ( x , t ) − u ( y , t ) ∣ x − y ∣ s ) ) ϕ ( x , t ) − ϕ ( y , t ) ∣ x − y ∣ N + s d y d x d t . 0=\left.\int_{\Omega}\phi\tau_hu\,dx\right|_{t_o-S}^{t_1}-\int_{t_o-S}^{t_1}\int_{\Omega}\partial_t\phi\tau_hu\,dx\,dt
+\int_{t_o-S}^{t_1}\iint_{\mathbb{R}^N\times\mathbb{R}^N}\left(\Phi\left(\frac{u_h(x,t)-u_h(y,t)}{|x-y|^s}\right)-\Phi\left(\frac{u(x,t)-u(y,t)}{|x-y|^s}\right)\right)\frac{\phi(x,t)-\phi(y,t)}{|x-y|^{N+s}}\,dy\,dx\,dt. 0 = ∫ Ω ϕ τ h u d x t o − S t 1 − ∫ t o − S t 1 ∫ Ω ∂ t ϕ τ h u d x d t + ∫ t o − S t 1 ∬ R N × R N ( Φ ( ∣ x − y ∣ s u h ( x , t ) − u h ( y , t ) ) − Φ ( ∣ x − y ∣ s u ( x , t ) − u ( y , t ) ) ) ∣ x − y ∣ N + s ϕ ( x , t ) − ϕ ( y , t ) d y d x d t .
where u h ( x , t ) : = u ( x + h , t ) u_h(x,t):=u(x+h,t) u h ( x , t ) := u ( x + h , t ) . We rewrite the diffusion term by introducing
K ( x , y , t ) = Φ ( u h ( x , t ) − u h ( y , t ) ∣ x − y ∣ s ) − Φ ( u ( x , t ) − u ( y , t ) ∣ x − y ∣ s ) τ h u ( x , t ) − τ h u ( y , t ) ∣ x − y ∣ s 1 ∣ x − y ∣ N + 2 s . K(x,y,t)=\frac{\Phi\left(\frac{u_h(x,t)-u_h(y,t)}{|x-y|^s}\right)-\Phi\left(\frac{u(x,t)-u(y,t)}{|x-y|^s}\right)}{\frac{\tau_hu(x,t)-\tau_hu(y,t)}{|x-y|^s}}\frac{1}{|x-y|^{N+2s}}. K ( x , y , t ) = ∣ x − y ∣ s τ h u ( x , t ) − τ h u ( y , t ) Φ ( ∣ x − y ∣ s u h ( x , t ) − u h ( y , t ) ) − Φ ( ∣ x − y ∣ s u ( x , t ) − u ( y , t ) ) ∣ x − y ∣ N + 2 s 1 .
Observe that K : R N × R N × [ t o − S , t o ] → R K:\mathbb{R}^N\times\mathbb{R}^N\times[t_o-S,t_o]\to\mathbb{R} K : R N × R N × [ t o − S , t o ] → R is measurable and satisfies
Λ − 1 ∣ x − y ∣ N + 2 s ≤ K ( x , y , t ) = K ( y , x , t ) ≤ Λ ∣ x − y ∣ N + 2 s , (10) \frac{\Lambda^{-1}}{|x-y|^{N+2s}}\leq K(x,y,t)=K(y,x,t)\leq\frac{\Lambda}{|x-y|^{N+2s}}, \tag*{(10)} ∣ x − y ∣ N + 2 s Λ − 1 ≤ K ( x , y , t ) = K ( y , x , t ) ≤ ∣ x − y ∣ N + 2 s Λ , (10)
due to (1.4), for any x , y ∈ R N x,y\in\mathbb{R}^N x , y ∈ R N and t ∈ [ t o − S , t o ] t\in[t_o-S,t_o] t ∈ [ t o − S , t o ] . Thus,
0 = ∫ Ω ϕ τ h u d x ∣ t o − S t 1 − ∫ t o − S t 1 ∫ Ω ∂ t ϕ τ h u d x d t + ∫ t o − S t 1 ∬ R N × R N ( τ h u ( x , t ) − τ h u ( y , t ) ) ( ϕ ( x , t ) − ϕ ( y , t ) ) K ( x , y , t ) d y d x d t . (11) 0=\left.\int_{\Omega}\phi\tau_hu\,dx\right|_{t_o-S}^{t_1}-\int_{t_o-S}^{t_1}\int_{\Omega}\partial_t\phi\tau_hu\,dx\,dt
+\int_{t_o-S}^{t_1}\iint_{\mathbb{R}^N\times\mathbb{R}^N}(\tau_hu(x,t)-\tau_hu(y,t))(\phi(x,t)-\phi(y,t))K(x,y,t)\,dy\,dx\,dt. \tag*{(11)} 0 = ∫ Ω ϕ τ h u d x t o − S t 1 − ∫ t o − S t 1 ∫ Ω ∂ t ϕ τ h u d x d t + ∫ t o − S t 1 ∬ R N × R N ( τ h u ( x , t ) − τ h u ( y , t )) ( ϕ ( x , t ) − ϕ ( y , t )) K ( x , y , t ) d y d x d t . (11)
In (4.2) we choose ϕ ( x , t ) = η 2 τ h u ( x , t ) \phi(x,t)=\eta^2\tau_hu(x,t) ϕ ( x , t ) = η 2 τ h u ( x , t ) , where η ∈ C 1 ( Q R , S ( z o ) , [ 0 , 1 ] ) \eta\in C^1(Q_{R,S}(z_o),[0,1]) η ∈ C 1 ( Q R , S ( z o ) , [ 0 , 1 ]) is a cut-off function satisfying
{ spt η ⊂ Q 1 2 ( R + r ) , S ( z o ) , η ≡ 1 in Q r , τ ( z o ) , ∥ D η ∥ L ∞ ≤ 4 R − r , ∣ ∂ t η ∣ ≤ 2 S − τ . \begin{cases}
\operatorname{spt}\eta\subset Q_{\frac{1}{2}(R+r),S}(z_o),\\
\eta\equiv1\quad\text{in }Q_{r,\tau}(z_o),\\
\|D\eta\|_{L^\infty}\leq\frac{4}{R-r},\quad|\partial_t\eta|\leq\frac{2}{S-\tau}.
\end{cases} ⎩ ⎨ ⎧ spt η ⊂ Q 2 1 ( R + r ) , S ( z o ) , η ≡ 1 in Q r , τ ( z o ) , ∥ D η ∥ L ∞ ≤ R − r 4 , ∣ ∂ t η ∣ ≤ S − τ 2 .
Standard calculations modulo a time mollification, cf. [7 Proof of Proposition 2.6] and [17 Appendix B] , give
1 2 ∫ B R × { t 1 } η 2 ∣ τ h u ∣ 2 d x + 1 2 ∫ t 0 − S t 1 ∫ B R ∂ t η 2 ∣ τ h u ∣ 2 d x d t + ∫ t 0 − S t 1 ∬ R N × R N ( τ h u ( x , t ) − τ h u ( y , t ) ) ( ϕ ( x , t ) − ϕ ( y , t ) ) K ( x , y , t ) d y d x d t = 0. (12) \frac{1}{2}\int_{B_R\times\{t_1\}}\eta^2|\tau_hu|^2\,dx+\frac{1}{2}\int_{t_0-S}^{t_1}\int_{B_R}\partial_t\eta^2|\tau_hu|^2\,dxdt
+\int_{t_0-S}^{t_1}\iint_{\mathbb{R}^N\times\mathbb{R}^N}(\tau_hu(x,t)-\tau_hu(y,t))(\phi(x,t)-\phi(y,t))K(x,y,t)\,dydxdt=0. \tag*{(12)} 2 1 ∫ B R × { t 1 } η 2 ∣ τ h u ∣ 2 d x + 2 1 ∫ t 0 − S t 1 ∫ B R ∂ t η 2 ∣ τ h u ∣ 2 d x d t + ∫ t 0 − S t 1 ∬ R N × R N ( τ h u ( x , t ) − τ h u ( y , t )) ( ϕ ( x , t ) − ϕ ( y , t )) K ( x , y , t ) d y d x d t = 0. (12)
It suffices to deal with the third term on the left. Decomposing R N \mathbb{R}^N R N into B R B_R B R and its complement R N ∖ B R \mathbb{R}^N\setminus B_R R N ∖ B R gives
∫ t 0 − S t 1 ∫ B R ∫ B R ( τ h u ( x , t ) − τ h u ( y , t ) ) ( τ h u ( x , t ) η 2 ( x , t ) − τ h u ( y , t ) η 2 ( y , t ) ) K ( x , y , t ) d x d y d t + 2 ∫ t 0 − S t 1 ∫ B 1 2 ( R + r ) ∫ R N ∖ B R ( τ h u ( x , t ) − τ h u ( y , t ) ) τ h u ( x , t ) η 2 ( x , t ) K ( x , y , t ) d y d t = : D + 2 T . \begin{aligned}
\int_{t_0-S}^{t_1}\int_{B_R}\int_{B_R}(\tau_hu(x,t)-\tau_hu(y,t))(\tau_hu(x,t)\eta^2(x,t)-\tau_hu(y,t)\eta^2(y,t))K(x,y,t)\,dxdydt \\
&+2\int_{t_0-S}^{t_1}\int_{B_{\frac{1}{2}(R+r)}}\int_{\mathbb{R}^N\setminus B_R}(\tau_hu(x,t)-\tau_hu(y,t))\tau_hu(x,t)\eta^2(x,t)K(x,y,t)\,dydt
=: \mathcal{D}+2\mathcal{T}.
\end{aligned} ∫ t 0 − S t 1 ∫ B R ∫ B R ( τ h u ( x , t ) − τ h u ( y , t )) ( τ h u ( x , t ) η 2 ( x , t ) − τ h u ( y , t ) η 2 ( y , t )) K ( x , y , t ) d x d y d t + 2 ∫ t 0 − S t 1 ∫ B 2 1 ( R + r ) ∫ R N ∖ B R ( τ h u ( x , t ) − τ h u ( y , t )) τ h u ( x , t ) η 2 ( x , t ) K ( x , y , t ) d y d t =: D + 2 T .
We proceed by estimating the term D \mathcal{D} D . By applying (10) to the kernel K ( x , y , t ) K(x,y,t) K ( x , y , t ) and Lemma 3.1 for a = τ h u ( x , t ) a=\tau_hu(x,t) a = τ h u ( x , t ) , b = τ h u ( y , t ) b=\tau_hu(y,t) b = τ h u ( y , t ) , e = η ( x , t ) e=\eta(x,t) e = η ( x , t ) and f = η ( y , t ) f=\eta(y,t) f = η ( y , t ) to the integrand, we deduce
D ≥ 1 4 Λ − 1 ∫ t 0 − S t 1 ∫ B R ∫ B R ∣ τ h u ( x , t ) − τ h u ( y , t ) ∣ 2 ( η 2 ( x , t ) + η 2 ( y , t ) ) ∣ x − y ∣ N + 2 s d x d y d t \mathcal{D}\geq\frac{1}{4}\Lambda^{-1}\int_{t_0-S}^{t_1}\int_{B_R}\int_{B_R}\frac{|\tau_hu(x,t)-\tau_hu(y,t)|^2(\eta^2(x,t)+\eta^2(y,t))}{|x-y|^{N+2s}}\,dxdydt D ≥ 4 1 Λ − 1 ∫ t 0 − S t 1 ∫ B R ∫ B R ∣ x − y ∣ N + 2 s ∣ τ h u ( x , t ) − τ h u ( y , t ) ∣ 2 ( η 2 ( x , t ) + η 2 ( y , t )) d x d y d t
− 1 2 Λ − 1 ∫ t 0 − S t 1 ∫ B R ∫ B R ( ∣ τ h u ( x , t ) ∣ + ∣ τ h u ( y , t ) ∣ ) 2 ∣ x − y ∣ N + 2 s ∣ η ( x , t ) − η ( y , t ) ∣ 2 d x d y d t = : 1 4 D 1 − 1 2 D 2 . -\frac{1}{2}\Lambda^{-1}\int_{t_0-S}^{t_1}\int_{B_R}\int_{B_R}\frac{(|\tau_hu(x,t)|+|\tau_hu(y,t)|)^2}{|x-y|^{N+2s}}|\eta(x,t)-\eta(y,t)|^2\,dxdydt
=: \frac{1}{4}\mathcal{D}_1-\frac{1}{2}\mathcal{D}_2. − 2 1 Λ − 1 ∫ t 0 − S t 1 ∫ B R ∫ B R ∣ x − y ∣ N + 2 s ( ∣ τ h u ( x , t ) ∣ + ∣ τ h u ( y , t ) ∣ ) 2 ∣ η ( x , t ) − η ( y , t ) ∣ 2 d x d y d t =: 4 1 D 1 − 2 1 D 2 .
Next, we further estimate D 1 \mathcal{D}_1 D 1 from below by recalling that η ≡ 1 \eta\equiv1 η ≡ 1 in B r × ( t 0 − τ , t 0 ] B_r\times(t_0-\tau,t_0] B r × ( t 0 − τ , t 0 ] :
D 1 ≥ Λ − 1 ∫ t 0 − τ t 0 ∫ B r ∫ B r ∣ τ h u ( x , t ) − τ h u ( y , t ) ∣ 2 ∣ x − y ∣ N + 2 s d x d y d t . \mathcal{D}_1\geq\Lambda^{-1}\int_{t_0-\tau}^{t_0}\int_{B_r}\int_{B_r}\frac{|\tau_hu(x,t)-\tau_hu(y,t)|^2}{|x-y|^{N+2s}}\,dxdydt. D 1 ≥ Λ − 1 ∫ t 0 − τ t 0 ∫ B r ∫ B r ∣ x − y ∣ N + 2 s ∣ τ h u ( x , t ) − τ h u ( y , t ) ∣ 2 d x d y d t .
For D 2 \mathcal{D}_2 D 2 , note that by the Lipschitz continuity of η \eta η , we obtain
∣ η ( x , t ) − η ( y , t ) ∣ 2 ≤ ∥ D η ∥ L ∞ 2 ∣ x − y ∣ 2 ≤ 16 ( R − r ) 2 ∣ x − y ∣ 2 . |\eta(x,t)-\eta(y,t)|^2\leq\|D\eta\|_{L^\infty}^2|x-y|^2\leq\frac{16}{(R-r)^2}|x-y|^2. ∣ η ( x , t ) − η ( y , t ) ∣ 2 ≤ ∥ D η ∥ L ∞ 2 ∣ x − y ∣ 2 ≤ ( R − r ) 2 16 ∣ x − y ∣ 2 .
Hence, due to symmetry, we deduce
D 2 ≤ 16 Λ − 1 ( R − r ) 2 ∫ t 0 − S t 0 ∫ B R ∫ B R ( ∣ τ h u ( x , t ) ∣ + ∣ τ h u ( x , t ) ∣ ) 2 ∣ x − y ∣ N + 2 s − 2 d x d y d t ≤ 64 Λ − 1 ( R − r ) 2 ∫ t 0 − S t 0 ∫ B R [ ∫ B R 1 ∣ x − y ∣ N − ( 1 − s ) 2 d y ] ∣ τ h u ( x , t ) ∣ 2 d x d t ≤ C R 2 s ( R R − r ) 2 ∫ t 0 − S t 0 ∫ B R ∣ τ h u ∣ 2 d x d t , \begin{aligned}
\mathcal{D}_2\leq\frac{16\Lambda^{-1}}{(R-r)^2}\int_{t_0-S}^{t_0}\int_{B_R}\int_{B_R}\frac{(|\tau_hu(x,t)|+|\tau_hu(x,t)|)^2}{|x-y|^{N+2s-2}}\,dxdydt \\
&\leq\frac{64\Lambda^{-1}}{(R-r)^2}\int_{t_0-S}^{t_0}\int_{B_R}\left[\int_{B_R}\frac{1}{|x-y|^{N-(1-s)2}}\,dy\right]|\tau_hu(x,t)|^2\,dxdt \\
&\leq\frac{C}{R^{2s}}\left(\frac{R}{R-r}\right)^2\int_{t_0-S}^{t_0}\int_{B_R}|\tau_hu|^2\,dxdt,
\end{aligned} D 2 ≤ ( R − r ) 2 16 Λ − 1 ∫ t 0 − S t 0 ∫ B R ∫ B R ∣ x − y ∣ N + 2 s − 2 ( ∣ τ h u ( x , t ) ∣ + ∣ τ h u ( x , t ) ∣ ) 2 d x d y d t ≤ ( R − r ) 2 64 Λ − 1 ∫ t 0 − S t 0 ∫ B R [ ∫ B R ∣ x − y ∣ N − ( 1 − s ) 2 1 d y ] ∣ τ h u ( x , t ) ∣ 2 d x d t ≤ R 2 s C ( R − r R ) 2 ∫ t 0 − S t 0 ∫ B R ∣ τ h u ∣ 2 d x d t ,
where C = C ( d a t a ) C=C(\mathrm{data}) C = C ( data ) . Finally, we treat the nonlocal term T \mathcal{T} T . In particular, by applying (10) to the kernel K ( x , y , t ) K(x,y,t) K ( x , y , t ) , Lemma 3.6 and Lemma 3.2 slice-wise it follows that
∣ T ∣ ≤ Λ ∫ t 0 − S t 0 ∫ B 1 2 ( R + r ) ∣ ∫ R N ∖ B R τ h u ( x , t ) − τ h u ( y , t ) ∣ x − y ∣ N + 2 s d y ∣ η 2 ( x , t ) ∣ τ h u ( x , t ) ∣ d x d t ≤ C R 2 s ( R R − r ) N + 2 s ∫ t o − S t o ∫ B 1 2 ( R + r ) η 2 ∣ τ h u ∣ 2 d x d t + C ∣ h ∣ R 2 s + 1 ( R R − r ) N + 2 s + 1 ∫ t o − S t o T ( t ; B 2 R ) ∫ B 1 2 ( R + r ) η 2 ∣ τ h u ∣ d x d t = T 1 + T 2 , \begin{aligned}
|\mathcal{T}|\leq\Lambda\int_{t_0-S}^{t_0}\int_{B_{\frac{1}{2}(R+r)}}\left|\int_{\mathbb{R}^N\setminus B_R}\frac{\tau_hu(x,t)-\tau_hu(y,t)}{|x-y|^{N+2s}}\,dy\right|\eta^2(x,t)|\tau_hu(x,t)|\,dxdt \\
&\le\frac{C}{R^{2s}}\left(\frac{R}{R-r}\right)^{N+2s}\int_{t_o-S}^{t_o}\int_{B_{\frac{1}{2}(R+r)}}\eta^2|\tau_hu|^2\,dxdt
+\frac{C|h|}{R^{2s+1}}\left(\frac{R}{R-r}\right)^{N+2s+1}\int_{t_o-S}^{t_o}\mathcal{T}(t;B_{2R})\int_{B_{\frac{1}{2}(R+r)}}\eta^2|\tau_hu|\,dxdt \\
&=\mathcal{T}_1+\mathcal{T}_2,
\end{aligned} ∣ T ∣ ≤ Λ ∫ t 0 − S t 0 ∫ B 2 1 ( R + r ) ∫ R N ∖ B R ∣ x − y ∣ N + 2 s τ h u ( x , t ) − τ h u ( y , t ) d y η 2 ( x , t ) ∣ τ h u ( x , t ) ∣ d x d t ≤ R 2 s C ( R − r R ) N + 2 s ∫ t o − S t o ∫ B 2 1 ( R + r ) η 2 ∣ τ h u ∣ 2 d x d t + R 2 s + 1 C ∣ h ∣ ( R − r R ) N + 2 s + 1 ∫ t o − S t o T ( t ; B 2 R ) ∫ B 2 1 ( R + r ) η 2 ∣ τ h u ∣ d x d t = T 1 + T 2 ,
For the term T 1 \mathcal{T}_1 T 1 we just recall that 0 ≤ η ≤ 1 0\le\eta\le1 0 ≤ η ≤ 1 . We estimate T 2 \mathcal{T}_2 T 2 by again taking advantage of the fact that 0 ≤ η ≤ 1 0\le\eta\le1 0 ≤ η ≤ 1 and subsequently taking supremum in time of the spatial integral and then apply Hölder’s and Young’s inequalities. To shorten the notation, we omit B 2 R B_{2R} B 2 R from T \mathcal{T} T and set p = N + 2 s + 1 p=N+2s+1 p = N + 2 s + 1 . Then,
T 2 ≤ C ∣ h ∣ R 2 s + 1 ( R R − r ) p [ ∫ t o − S t o T ( t ) d t ] [ ess sup t ∈ ( t o − S , t o ) ∫ B R × { t } η ∣ τ h u ∣ d x ] ≤ C ∣ h ∣ R 2 s + 1 ( R R − r ) p [ ∫ t o − S t o T ( t ) d t ] ∣ B R ∣ 1 2 [ ess sup t ∈ ( t o − S , t o ) ∫ B R × { t } η 2 ∣ τ h u ∣ 2 d x ] 1 2 ≤ C ∣ h ∣ 2 R 4 s + 2 ( R R − r ) 2 p R N [ ∫ t o − S t o T ( t ) d t ] 2 + 1 4 ess sup t ∈ ( t o − S , t o ) ∫ B R × { t } η 2 ∣ τ h u ∣ 2 d x . \begin{aligned}
\mathcal{T}_2\le\frac{C|h|}{R^{2s+1}}\left(\frac{R}{R-r}\right)^p\left[\int_{t_o-S}^{t_o}\mathcal{T}(t)\,dt\right]\left[\mathop{\operatorname{ess\,sup}}_{t\in(t_o-S,t_o)}\int_{B_R\times\{t\}}\eta|\tau_hu|\,dx\right] \\
&\le\frac{C|h|}{R^{2s+1}}\left(\frac{R}{R-r}\right)^p\left[\int_{t_o-S}^{t_o}\mathcal{T}(t)\,dt\right]|B_R|^{\frac12}\left[\mathop{\operatorname{ess\,sup}}_{t\in(t_o-S,t_o)}\int_{B_R\times\{t\}}\eta^2|\tau_hu|^2\,dx\right]^{\frac12} \\
&\le\frac{C|h|^2}{R^{4s+2}}\left(\frac{R}{R-r}\right)^{2p}R^N\left[\int_{t_o-S}^{t_o}\mathcal{T}(t)\,dt\right]^2+\frac14\mathop{\operatorname{ess\,sup}}_{t\in(t_o-S,t_o)}\int_{B_R\times\{t\}}\eta^2|\tau_hu|^2\,dx.
\end{aligned} T 2 ≤ R 2 s + 1 C ∣ h ∣ ( R − r R ) p [ ∫ t o − S t o T ( t ) d t ] [ ess sup t ∈ ( t o − S , t o ) ∫ B R × { t } η ∣ τ h u ∣ d x ] ≤ R 2 s + 1 C ∣ h ∣ ( R − r R ) p [ ∫ t o − S t o T ( t ) d t ] ∣ B R ∣ 2 1 [ ess sup t ∈ ( t o − S , t o ) ∫ B R × { t } η 2 ∣ τ h u ∣ 2 d x ] 2 1 ≤ R 4 s + 2 C ∣ h ∣ 2 ( R − r R ) 2 p R N [ ∫ t o − S t o T ( t ) d t ] 2 + 4 1 ess sup t ∈ ( t o − S , t o ) ∫ B R × { t } η 2 ∣ τ h u ∣ 2 d x .
Now, we combine all estimates in (12) and use the arbitrariness of t 1 ∈ ( t o − τ , t o ) t_1\in(t_o-\tau,t_o) t 1 ∈ ( t o − τ , t o ) to conclude the proof. □
The second kind of energy estimates concern the truncations of τ h u \tau_hu τ h u ; it will be used to upgrade the obtained W 1 , 2 W^{1,2} W 1 , 2 -regularity to the Lipschitz regularity of weak solutions.
Lemma 4.2. Let s ∈ ( 0 , 1 ) s\in(0,1) s ∈ ( 0 , 1 ) . If u u u is a local weak solution to (3) in the sense of Definition 2.1, then for any absolutely continuous function k ( ⋅ ) k(\cdot) k ( ⋅ ) in ( 0 , T ) (0,T) ( 0 , T ) , any cylinder Q R , S ( z o ) ⋐ Ω T Q_{R,S}(z_o)\Subset\Omega_T Q R , S ( z o ) ⋐ Ω T , any h ∈ R N ∖ { 0 } h\in\mathbb{R}^N\setminus\{0\} h ∈ R N ∖ { 0 } with 0 < ∣ h ∣ ≤ 3 4 R 0<|h|\le\frac34R 0 < ∣ h ∣ ≤ 4 3 R and any non-negative, Lipschitz function ζ ( x , t ) \zeta(x,t) ζ ( x , t ) in Q R , S ( z o ) Q_{R,S}(z_o) Q R , S ( z o ) such that spt ζ ( ⋅ , t ) ⊂ B R ( x o ) \operatorname{spt}\zeta(\cdot,t)\subset B_R(x_o) spt ζ ( ⋅ , t ) ⊂ B R ( x o ) for any t ∈ ( t o − S , t o ) t\in(t_o-S,t_o) t ∈ ( t o − S , t o ) we have
ess sup t ∈ [ t o − S , t o ] ∫ B R × { t } ζ 2 w ± 2 d x + Λ − 1 ∭ Q R , S ∣ ζ w ± ( x , t ) − ζ w ± ( y , t ) ∣ 2 ∣ x − y ∣ N + 2 s d y d x d t \mathop{\operatorname{ess\,sup}}_{t\in[t_o-S,t_o]}\int_{B_R\times\{t\}}\zeta^2w_\pm^2\,dx+\Lambda^{-1}\iiint_{Q_{R,S}}\frac{|\zeta w_\pm(x,t)-\zeta w_\pm(y,t)|^2}{|x-y|^{N+2s}}\,dydxdt ess sup t ∈ [ t o − S , t o ] ∫ B R × { t } ζ 2 w ± 2 d x + Λ − 1 ∭ Q R , S ∣ x − y ∣ N + 2 s ∣ ζ w ± ( x , t ) − ζ w ± ( y , t ) ∣ 2 d y d x d t + Λ − 1 ∭ Q R , S ζ 2 ( x , t ) w ± ( x , t ) w ∓ ( y , t ) ∣ x − y ∣ N + 2 s d y d x d t +\Lambda^{-1}\iiint_{Q_{R,S}}\frac{\zeta^2(x,t)w_\pm(x,t)w_\mp(y,t)}{|x-y|^{N+2s}}\,dydxdt + Λ − 1 ∭ Q R , S ∣ x − y ∣ N + 2 s ζ 2 ( x , t ) w ± ( x , t ) w ∓ ( y , t ) d y d x d t ≤ 2 Λ ∭ Q R , S max { w ± 2 ( x , t ) , w ± 2 ( y , t ) } ∣ ζ ( y , t ) − ζ ( x , t ) ∣ 2 ∣ x − y ∣ N + 2 s d y d x d t \le2\Lambda\iiint_{Q_{R,S}}\max\{w_\pm^2(x,t),w_\pm^2(y,t)\}\frac{|\zeta(y,t)-\zeta(x,t)|^2}{|x-y|^{N+2s}}\,dydxdt ≤ 2Λ ∭ Q R , S max { w ± 2 ( x , t ) , w ± 2 ( y , t )} ∣ x − y ∣ N + 2 s ∣ ζ ( y , t ) − ζ ( x , t ) ∣ 2 d y d x d t + 4 Λ ∬ Q R , S [ ζ 2 w ± 2 ( x , t ) + ∣ k ( t ) ∣ ζ 2 w ± ( x , t ) ] ∫ R N ∖ B R d y ∣ x − y ∣ N + 2 s d x d t +4\Lambda\iint_{Q_{R,S}}\left[\zeta^2w_\pm^2(x,t)+|k(t)|\zeta^2w_\pm(x,t)\right]\int_{\mathbb{R}^N\setminus B_R}\frac{dy}{|x-y|^{N+2s}}\,dxdt + 4Λ ∬ Q R , S [ ζ 2 w ± 2 ( x , t ) + ∣ k ( t ) ∣ ζ 2 w ± ( x , t ) ] ∫ R N ∖ B R ∣ x − y ∣ N + 2 s d y d x d t + 4 Λ ∬ Q R , S ζ 2 w ± ( x , t ) ∣ ∫ R N ∖ B R τ h u ( y , t ) ∣ x − y ∣ N + 2 s d y ∣ d x d t +4\Lambda\iint_{Q_{R,S}}\zeta^2w_\pm(x,t)\left|\int_{\mathbb{R}^N\setminus B_R}\frac{\tau_hu(y,t)}{|x-y|^{N+2s}}\,dy\right|\,dxdt + 4Λ ∬ Q R , S ζ 2 w ± ( x , t ) ∫ R N ∖ B R ∣ x − y ∣ N + 2 s τ h u ( y , t ) d y d x d t ∓ 2 ∬ Q R , S k ′ ( t ) ζ 2 w ± d x d t + ∬ Q R , S w ± 2 ∣ ∂ t ζ 2 ∣ d x d t \mp2\iint_{Q_{R,S}}k'(t)\zeta^2w_\pm\,dxdt+\iint_{Q_{R,S}}w_\pm^2|\partial_t\zeta^2|\,dxdt ∓ 2 ∬ Q R , S k ′ ( t ) ζ 2 w ± d x d t + ∬ Q R , S w ± 2 ∣ ∂ t ζ 2 ∣ d x d t + ∫ B R × { t o − S } ζ 2 w ± 2 d x , +\int_{B_R\times\{t_o-S\}}\zeta^2w_\pm^2\,dx, + ∫ B R × { t o − S } ζ 2 w ± 2 d x , where we defined w : = τ h u − k w:=\tau_hu-k w := τ h u − k and omitted x o x_o x o and z o z_o z o .
Proof. We will only deal with the case of w + w_+ w + . Let t 1 ∈ ( t o − S , t o ] t_1\in(t_o-S,t_o] t 1 ∈ ( t o − S , t o ] and take identity (4.2) from the proof of Lemma 4.1 as starting point. Next, we choose the test function
ϕ ( x , t ) : = ζ 2 ( x , t ) w + ( x , t ) ≡ ζ 2 ( τ h u ( x , t ) − k ( t ) ) + , (13) \phi(x,t):=\zeta^2(x,t)w_+(x,t)\equiv\zeta^2(\tau_hu(x,t)-k(t))_+, \tag*{(13)} ϕ ( x , t ) := ζ 2 ( x , t ) w + ( x , t ) ≡ ζ 2 ( τ h u ( x , t ) − k ( t ) ) + , (13)
which can be justified by a time mollification. Integrating by parts the time part becomes
∫ B R ϕ τ h u d x ∣ t 0 − S t 1 − ∫ t 0 − S t 1 ∫ B R ∂ t ϕ τ h u d x d t = ∫ t 0 − S t 1 ∫ B R ϕ ∂ t ( τ h u ) d x d t = 1 2 ∫ B R × { t 1 } ζ 2 w + 2 d x − 1 2 ∫ B R × { t 0 − S } ζ 2 w + 2 d x − 1 2 ∫ t 0 − S t 1 ∫ B R ∂ t ζ 2 w + 2 d x d t + ∫ t 0 − S t 1 ∫ B R k ′ ζ 2 w + d x d t . \begin{aligned}
\left.\int_{B_R}\phi\tau_hu\,\mathrm{d}x\right|_{t_0-S}^{t_1}
&-\int_{t_0-S}^{t_1}\int_{B_R}\partial_t\phi\tau_hu\,\mathrm{d}x\mathrm{d}t\\
&=\int_{t_0-S}^{t_1}\int_{B_R}\phi\partial_t(\tau_hu)\,\mathrm{d}x\mathrm{d}t\\
&=\frac{1}{2}\int_{B_R\times\{t_1\}}\zeta^2w_+^2\,\mathrm{d}x-\frac{1}{2}\int_{B_R\times\{t_0-S\}}\zeta^2w_+^2\,\mathrm{d}x-\frac{1}{2}\int_{t_0-S}^{t_1}\int_{B_R}\partial_t\zeta^2w_+^2\,\mathrm{d}x\mathrm{d}t\\
&\quad+\int_{t_0-S}^{t_1}\int_{B_R}k'\zeta^2w_+\,\mathrm{d}x\mathrm{d}t.
\end{aligned} ∫ B R ϕ τ h u d x t 0 − S t 1 − ∫ t 0 − S t 1 ∫ B R ∂ t ϕ τ h u d x d t = ∫ t 0 − S t 1 ∫ B R ϕ ∂ t ( τ h u ) d x d t = 2 1 ∫ B R × { t 1 } ζ 2 w + 2 d x − 2 1 ∫ B R × { t 0 − S } ζ 2 w + 2 d x − 2 1 ∫ t 0 − S t 1 ∫ B R ∂ t ζ 2 w + 2 d x d t + ∫ t 0 − S t 1 ∫ B R k ′ ζ 2 w + d x d t .
As for the diffusion part D \mathcal{D} D , the last term in (4.2), we split the integral and take advantage of spt ζ ⊂ B R × ( t 0 − S , t 0 ) \operatorname{spt}\zeta\subset B_R\times(t_0-S,t_0) spt ζ ⊂ B R × ( t 0 − S , t 0 ) , as well as the integrand’s symmetry, to get
D = ∫ t 0 − S t 1 ∫ B R ∫ B R ( τ h u ( x , t ) − τ h u ( y , t ) ) ( ϕ ( x , t ) − ϕ ( y , t ) ) K ( x , y , t ) d y d x d t + 2 ∫ t 0 − S t 1 ∫ B R ∫ R N ∖ B R ( τ h u ( x , t ) − τ h u ( y , t ) ) ϕ ( x , t ) K ( x , y , t ) d y d x d t = : D 1 + D 2 . \begin{aligned}
\mathcal{D}
&=\int_{t_0-S}^{t_1}\int_{B_R}\int_{B_R}(\tau_hu(x,t)-\tau_hu(y,t))(\phi(x,t)-\phi(y,t))K(x,y,t)\,\mathrm{d}y\mathrm{d}x\mathrm{d}t\\
&\quad+2\int_{t_0-S}^{t_1}\int_{B_R}\int_{\mathbb{R}^N\setminus B_R}(\tau_hu(x,t)-\tau_hu(y,t))\phi(x,t)K(x,y,t)\,\mathrm{d}y\mathrm{d}x\mathrm{d}t\\
&=:\mathcal{D}_1+\mathcal{D}_2.
\end{aligned} D = ∫ t 0 − S t 1 ∫ B R ∫ B R ( τ h u ( x , t ) − τ h u ( y , t )) ( ϕ ( x , t ) − ϕ ( y , t )) K ( x , y , t ) d y d x d t + 2 ∫ t 0 − S t 1 ∫ B R ∫ R N ∖ B R ( τ h u ( x , t ) − τ h u ( y , t )) ϕ ( x , t ) K ( x , y , t ) d y d x d t =: D 1 + D 2 .
For D 1 \mathcal{D}_1 D 1 , we use the identity
( τ h u ( x , t ) − τ h u ( y , t ) ) ( ϕ ( x , t ) − ϕ ( y , t ) ) = [ ζ w + ( x , t ) − ζ w + ( y , t ) ] 2 + w − ( x , t ) w + ( y , t ) ζ 2 ( y , t ) + w − ( y , t ) w + ( x , t ) ζ 2 ( x , t ) − w + ( x , t ) w + ( y , t ) [ ζ ( x , t ) − ζ ( y , t ) ] 2 . \begin{aligned}
&(\tau_hu(x,t)-\tau_hu(y,t))(\phi(x,t)-\phi(y,t))\\
&=[\zeta w_+(x,t)-\zeta w_+(y,t)]^2+w_-(x,t)w_+(y,t)\zeta^2(y,t)\\
&\quad+w_-(y,t)w_+(x,t)\zeta^2(x,t)-w_+(x,t)w_+(y,t)[\zeta(x,t)-\zeta(y,t)]^2.
\end{aligned} ( τ h u ( x , t ) − τ h u ( y , t )) ( ϕ ( x , t ) − ϕ ( y , t )) = [ ζ w + ( x , t ) − ζ w + ( y , t ) ] 2 + w − ( x , t ) w + ( y , t ) ζ 2 ( y , t ) + w − ( y , t ) w + ( x , t ) ζ 2 ( x , t ) − w + ( x , t ) w + ( y , t ) [ ζ ( x , t ) − ζ ( y , t ) ] 2 .
Thus, using this identity and applying (10) to the kernel we deduce
D 1 ≥ ∫ t 0 − S t 1 ∫ B R ∫ B R ∣ ζ w + ( x , t ) − ζ w + ( y , t ) ∣ 2 K ( x , y , t ) d y d x d t + ∫ t 0 − S t 1 ∫ B R ∫ B R w − ( y , t ) w + ( x , t ) ζ 2 ( x , t ) K ( x , y , t ) d y d x d t − ∫ t 0 − S t 1 ∫ B R ∫ B R w + ( x , t ) w + ( y , t ) ∣ ζ ( y , t ) − ζ ( x , t ) ∣ 2 K ( x , y , t ) d y d x d t ≥ Λ − 1 ∫ t 0 − S t 1 ∫ B R ∫ B R ∣ ζ w + ( x , t ) − ζ w + ( y , t ) ∣ 2 ∣ x − y ∣ N + 2 s d y d x d t + Λ − 1 ∫ t 0 − S t 1 ∫ B R ∫ B R w − ( y , t ) w + ( x , t ) ζ 2 ( x , t ) ∣ x − y ∣ N + 2 s d y d x d t − Λ ∫ t 0 − S t 1 ∫ B R ∫ B R max { w + 2 ( x , t ) , w + 2 ( y , t ) } ∣ ζ ( y , t ) − ζ ( x , t ) ∣ 2 ∣ x − y ∣ N + 2 s d y d x d t . \begin{aligned}
\mathcal{D}_1
&\ge\int_{t_0-S}^{t_1}\int_{B_R}\int_{B_R}|\zeta w_+(x,t)-\zeta w_+(y,t)|^2K(x,y,t)\,\mathrm{d}y\mathrm{d}x\mathrm{d}t\\
&\quad+\int_{t_0-S}^{t_1}\int_{B_R}\int_{B_R}w_-(y,t)w_+(x,t)\zeta^2(x,t)K(x,y,t)\,\mathrm{d}y\mathrm{d}x\mathrm{d}t\\
&\quad-\int_{t_0-S}^{t_1}\int_{B_R}\int_{B_R}w_+(x,t)w_+(y,t)|\zeta(y,t)-\zeta(x,t)|^2K(x,y,t)\,\mathrm{d}y\mathrm{d}x\mathrm{d}t\\
&\ge\Lambda^{-1}\int_{t_0-S}^{t_1}\int_{B_R}\int_{B_R}\frac{|\zeta w_+(x,t)-\zeta w_+(y,t)|^2}{|x-y|^{N+2s}}\,\mathrm{d}y\mathrm{d}x\mathrm{d}t\\
&\quad+\Lambda^{-1}\int_{t_0-S}^{t_1}\int_{B_R}\int_{B_R}\frac{w_-(y,t)w_+(x,t)\zeta^2(x,t)}{|x-y|^{N+2s}}\,\mathrm{d}y\mathrm{d}x\mathrm{d}t\\
&\quad-\Lambda\int_{t_0-S}^{t_1}\int_{B_R}\int_{B_R}\max\{w_+^2(x,t),w_+^2(y,t)\}\frac{|\zeta(y,t)-\zeta(x,t)|^2}{|x-y|^{N+2s}}\,\mathrm{d}y\mathrm{d}x\mathrm{d}t.
\end{aligned} D 1 ≥ ∫ t 0 − S t 1 ∫ B R ∫ B R ∣ ζ w + ( x , t ) − ζ w + ( y , t ) ∣ 2 K ( x , y , t ) d y d x d t + ∫ t 0 − S t 1 ∫ B R ∫ B R w − ( y , t ) w + ( x , t ) ζ 2 ( x , t ) K ( x , y , t ) d y d x d t − ∫ t 0 − S t 1 ∫ B R ∫ B R w + ( x , t ) w + ( y , t ) ∣ ζ ( y , t ) − ζ ( x , t ) ∣ 2 K ( x , y , t ) d y d x d t ≥ Λ − 1 ∫ t 0 − S t 1 ∫ B R ∫ B R ∣ x − y ∣ N + 2 s ∣ ζ w + ( x , t ) − ζ w + ( y , t ) ∣ 2 d y d x d t + Λ − 1 ∫ t 0 − S t 1 ∫ B R ∫ B R ∣ x − y ∣ N + 2 s w − ( y , t ) w + ( x , t ) ζ 2 ( x , t ) d y d x d t − Λ ∫ t 0 − S t 1 ∫ B R ∫ B R max { w + 2 ( x , t ) , w + 2 ( y , t )} ∣ x − y ∣ N + 2 s ∣ ζ ( y , t ) − ζ ( x , t ) ∣ 2 d y d x d t .
Moreover, note that for the last term in the above display we used the estimate
w + ( x , t ) w + ( y , t ) ≤ max { w + 2 ( x , t ) , w + 2 ( y , t ) } . w_+(x,t)w_+(y,t)\le\max\{w_+^2(x,t),w_+^2(y,t)\}. w + ( x , t ) w + ( y , t ) ≤ max { w + 2 ( x , t ) , w + 2 ( y , t )} .
As for D 2 \mathcal{D}_2 D 2 , we apply (10) and split the term into
∣ D 2 ∣ ≤ 2 Λ ∣ ∫ t 0 − S t 1 ∫ B R ∫ R N ∖ B R τ h u ( x , t ) ∣ x − y ∣ N + 2 s ϕ ( x , t ) d y d x d t ∣ + 2 Λ ∣ ∫ t 0 − S t 1 ∫ B R ∫ R N ∖ B R τ h u ( y , t ) ∣ x − y ∣ N + 2 s ϕ ( x , t ) d y d x d t ∣ = : D 2 , 1 + D 2 , 2 . \begin{aligned}
|\mathcal{D}_2|
&\le2\Lambda\left|\int_{t_0-S}^{t_1}\int_{B_R}\int_{\mathbb{R}^N\setminus B_R}\frac{\tau_hu(x,t)}{|x-y|^{N+2s}}\phi(x,t)\,\mathrm{d}y\mathrm{d}x\mathrm{d}t\right|\\
&\quad+2\Lambda\left|\int_{t_0-S}^{t_1}\int_{B_R}\int_{\mathbb{R}^N\setminus B_R}\frac{\tau_hu(y,t)}{|x-y|^{N+2s}}\phi(x,t)\,\mathrm{d}y\mathrm{d}x\mathrm{d}t\right|\\
&=:\mathcal{D}_{2,1}+\mathcal{D}_{2,2}.
\end{aligned} ∣ D 2 ∣ ≤ 2Λ ∫ t 0 − S t 1 ∫ B R ∫ R N ∖ B R ∣ x − y ∣ N + 2 s τ h u ( x , t ) ϕ ( x , t ) d y d x d t + 2Λ ∫ t 0 − S t 1 ∫ B R ∫ R N ∖ B R ∣ x − y ∣ N + 2 s τ h u ( y , t ) ϕ ( x , t ) d y d x d t =: D 2 , 1 + D 2 , 2 .
The first term is estimated simply by
D 2 , 1 ≤ 2 Λ ∫ t 0 − S t 1 ∫ B R ∫ R N ∖ B R ζ 2 w + 2 ( x , t ) + ∣ k ( t ) ∣ ζ 2 w + ( x , t ) ∣ x − y ∣ N + 2 s d y d x d t . \mathcal{D}_{2,1} \le2\Lambda\int_{t_{0}-S}^{t_{1}} \int_{B_R} \int_{\mathbb{R}^N\setminus B_R} \frac{\zeta^2 w_+^2(x,t)+|k(t)|\zeta^2 w_+(x,t)}{|x-y|^{N+2s}}\,dy\,dx\,dt. D 2 , 1 ≤ 2Λ ∫ t 0 − S t 1 ∫ B R ∫ R N ∖ B R ∣ x − y ∣ N + 2 s ζ 2 w + 2 ( x , t ) + ∣ k ( t ) ∣ ζ 2 w + ( x , t ) d y d x d t .
and the second term by
D 2 , 2 ≤ 2 Λ ∫ t 0 − S t 1 ∫ B R ζ 2 w + ( x , t ) ∣ ∫ R N ∖ B R τ h u ( y , t ) ∣ x − y ∣ N + 2 s d y ∣ d x d t . \mathcal{D}_{2,2} \le2\Lambda\int_{t_{0}-S}^{t_{1}} \int_{B_R} \zeta^2 w_+(x,t)\left|\int_{\mathbb{R}^N\setminus B_R}\frac{\tau_hu(y,t)}{|x-y|^{N+2s}}\,dy\right|\,dx\,dt. D 2 , 2 ≤ 2Λ ∫ t 0 − S t 1 ∫ B R ζ 2 w + ( x , t ) ∫ R N ∖ B R ∣ x − y ∣ N + 2 s τ h u ( y , t ) d y d x d t .
Inserting these inequalities above and using the arbitrariness of t 1 ∈ ( t 0 − S , t 0 ] t_1 \in(t_0-S,t_0] t 1 ∈ ( t 0 − S , t 0 ] finishes the proof. □
Lipschitz regularity
W1,2-regularity
In this subsection we improve the preset W s , 2 W^{s,2} W s , 2 -regularity of local weak solutions to W 1 , 2 W^{1,2} W 1 , 2 . The approach consists of first exploiting the energy estimate for finite differences in Lemma 4.1 to obtain estimates for second order finite differences, and then performing a tailored finite-iteration argument to upgrade the spatial differentiability. The statement we aim to prove is as follows.
Proposition 5.1. Suppose that s ∈ ( 0 , 1 ) s \in(0,1) s ∈ ( 0 , 1 ) . Then, for any locally bounded, local weak solution u u u to (3) in the sense of Definition 2.1 , we have that
u ∈ L l o c 2 ( 0 , T ; W l o c 1 , 2 ( Ω ) ) . u \in L_{\mathrm{loc}}^2(0,T;W_{\mathrm{loc}}^{1,2}(\Omega)). u ∈ L loc 2 ( 0 , T ; W loc 1 , 2 ( Ω )) . Moreover, for Q 2 R ≡ Q 2 R ( z 0 ) ⋐ Ω T Q_{2R} \equiv Q_{2R}(z_0) \Subset\Omega_T Q 2 R ≡ Q 2 R ( z 0 ) ⋐ Ω T , there exists a constant C = C ( data ) C=C(\text{data}) C = C ( data ) such that the estimate
∬ Q R ∣ D u ∣ 2 d x d t ≤ C R 2 T 2 ( Q 2 R ) \iint_{Q_R}|Du|^2\,dx\,dt \le\frac{C}{R^2}\mathcal{T}^2(Q_{2R}) ∬ Q R ∣ D u ∣ 2 d x d t ≤ R 2 C T 2 ( Q 2 R ) holds, where T ( Q 2 R ) \mathcal{T}(Q_{2R}) T ( Q 2 R ) is defined as in (8) .
As mentioned above, before proceeding to the proof, we need a lemma concerning second-order finite differences.
Lemma 5.2. Suppose the assumptions of Proposition 5.1 hold true. Then, for any Q R ≡ Q R ( z 0 ) ⋐ Ω T Q_R \equiv Q_R(z_0) \Subset\Omega_T Q R ≡ Q R ( z 0 ) ⋐ Ω T we have
∬ Q 1 4 R ∣ τ h u ∣ 2 d x d t ≤ C ∣ h ∣ 2 s R 2 s [ ∬ Q R ∣ τ h u ∣ 2 d x d t + ∣ h ∣ 2 R 2 R N + 2 s T 2 ( Q 2 R ) ] \iint_{Q_{\frac14 R}}|\tau_hu|^2\,dx\,dt \le C\frac{|h|^{2s}}{R^{2s}}\left[\iint_{Q_R}|\tau_hu|^2\,dx\,dt+\frac{|h|^2}{R^2}R^{N+2s}\mathcal{T}^2(Q_{2R})\right] ∬ Q 4 1 R ∣ τ h u ∣ 2 d x d t ≤ C R 2 s ∣ h ∣ 2 s [ ∬ Q R ∣ τ h u ∣ 2 d x d t + R 2 ∣ h ∣ 2 R N + 2 s T 2 ( Q 2 R ) ] for any h ∈ R N ∖ { 0 } h \in\mathbb{R}^N\setminus\{0\} h ∈ R N ∖ { 0 } with ∣ h ∣ ≤ 1 8 R |h|\le\frac18 R ∣ h ∣ ≤ 8 1 R , where C = C ( data ) C=C(\text{data}) C = C ( data ) and T ( Q 2 R ) \mathcal{T}(Q_{2R}) T ( Q 2 R ) is as in (8) .
Proof. Upon rescaling, it suffices to prove it for R = 2 R=2 R = 2 and z 0 = ( 0 , 0 ) z_0=(0,0) z 0 = ( 0 , 0 ) . By applying Lemma 4.1 over Q 1 ⊂ Q 2 Q_1\subset Q_2 Q 1 ⊂ Q 2 we deduce that
∫ − 1 0 [ τ h u ( ⋅ , t ) ] W s , 2 ( B 1 ) 2 d t ≤ C ∬ Q 2 ∣ τ h u ∣ 2 d x d t + C ∣ h ∣ 2 T 2 ( Q 4 ) . \int_{-1}^{0}[\tau_hu(\cdot,t)]_{W^{s,2}(B_1)}^2\,dt \le C\iint_{Q_2}|\tau_hu|^2\,dx\,dt+C|h|^2\mathcal{T}^2(Q_4). ∫ − 1 0 [ τ h u ( ⋅ , t ) ] W s , 2 ( B 1 ) 2 d t ≤ C ∬ Q 2 ∣ τ h u ∣ 2 d x d t + C ∣ h ∣ 2 T 2 ( Q 4 ) .
Next, we apply Lemma 3.4 slice-wise with ( v , γ , R ) = ( τ h u , s , 1 ) (v,\gamma,R)=(\tau_hu,s,1) ( v , γ , R ) = ( τ h u , s , 1 ) , integrate in time and use the above display to get
∬ Q 1 2 ∣ τ h u ∣ 2 d x d t ≤ C ∣ h ∣ 2 s [ ∫ − 1 0 [ τ h u ( ⋅ , t ) ] W s , 2 ( B 1 ) 2 d t + ∬ Q 1 ∣ τ h u ∣ 2 d x d t ] ≤ C ∣ h ∣ 2 s [ ∬ Q 2 ∣ τ h u ∣ 2 d x d t + ∣ h ∣ 2 T 2 ( Q 4 ) ] \begin{aligned}
\iint_{Q_{\frac12}}|\tau_hu|^2\,dx\,dt \le C|h|^{2s}\left[\int_{-1}^{0}[\tau_hu(\cdot,t)]_{W^{s,2}(B_1)}^2\,dt+\iint_{Q_1}|\tau_hu|^2\,dx\,dt\right] \\
&\le C|h|^{2s}\left[\iint_{Q_2}|\tau_hu|^2\,dx\,dt+|h|^2\mathcal{T}^2(Q_4)\right]
\end{aligned} ∬ Q 2 1 ∣ τ h u ∣ 2 d x d t ≤ C ∣ h ∣ 2 s [ ∫ − 1 0 [ τ h u ( ⋅ , t ) ] W s , 2 ( B 1 ) 2 d t + ∬ Q 1 ∣ τ h u ∣ 2 d x d t ] ≤ C ∣ h ∣ 2 s [ ∬ Q 2 ∣ τ h u ∣ 2 d x d t + ∣ h ∣ 2 T 2 ( Q 4 ) ]
for 0 < ∣ h ∣ ≤ 1 2 0<|h|\le\frac12 0 < ∣ h ∣ ≤ 2 1 , where C = C ( data ) C=C(\text{data}) C = C ( data ) . □
Now, we perform finite iteration and establish the aim of this subsection.
Proof of Proposition 5.1. By a proper scaling argument, it suffices to show the desired estimate with R = 1 R=1 R = 1 . For i ∈ N 0 i\in\mathbb{N}_0 i ∈ N 0 set ϱ i : = 4 − i \varrho_i:=4^{-i} ϱ i := 4 − i and Q i : = Q ϱ i Q_i:=Q_{\varrho_i} Q i := Q ϱ i ; we apply Lemma 5.2 with R = ϱ i R=\varrho_i R = ϱ i to get
∬ Q i + 1 ∣ τ h 2 u ∣ 2 d x d t ≤ C ∣ h ∣ 2 s [ ∬ Q i ∣ τ h u ∣ 2 d x d t + ∣ h ∣ 2 T 2 ( Q 2 ϱ i ) ] . \iint_{Q_{i+1}}|\tau_h^2u|^2\,dxdt\leq C|h|^{2s}\left[\iint_{Q_i}|\tau_hu|^2\,dxdt+|h|^2\mathcal{T}^2(Q_{2\varrho_i})\right]. ∬ Q i + 1 ∣ τ h 2 u ∣ 2 d x d t ≤ C ∣ h ∣ 2 s [ ∬ Q i ∣ τ h u ∣ 2 d x d t + ∣ h ∣ 2 T 2 ( Q 2 ϱ i ) ] .
for some C = C ( i , d a t a ) C=C(i,\mathrm{data}) C = C ( i , data ) and for any h ∈ R N ∖ { 0 } h\in\mathbb{R}^N\setminus\{0\} h ∈ R N ∖ { 0 } with ∣ h ∣ ≤ 1 8 ϱ i |h|\leq\frac{1}{8}\varrho_i ∣ h ∣ ≤ 8 1 ϱ i , where
T ( Q 2 ϱ i ) : = ∥ u ∥ L ∞ ( Q 2 ϱ i ) + Tail ( u ; Q 2 ϱ i ) . \mathcal{T}(Q_{2\varrho_i}):=\|u\|_{L^\infty(Q_{2\varrho_i})}+\operatorname{Tail}(u;Q_{2\varrho_i}). T ( Q 2 ϱ i ) := ∥ u ∥ L ∞ ( Q 2 ϱ i ) + Tail ( u ; Q 2 ϱ i ) .
To simplify the above estimate, we note by Lemma 3.2 that
T ( Q 2 ϱ i ) ≤ C ( i , N ) T ( Q 2 ) . \mathcal{T}(Q_{2\varrho_i})\leq C(i,N)\mathcal{T}(Q_2). T ( Q 2 ϱ i ) ≤ C ( i , N ) T ( Q 2 ) .
Hence, there exists some C ^ i = C ^ i ( d a t a ) \widehat{C}_i=\widehat{C}_i(\mathrm{data}) C i = C i ( data ) such that
∬ Q i + 1 ∣ τ h 2 u ∣ 2 d x d t ≤ C ^ i + 1 ∣ h ∣ 2 s [ ∬ Q i ∣ τ h u ∣ 2 d x d t + ∣ h ∣ 2 T 2 ( Q 2 ) ] ≤ C ^ i + 1 ∣ h ∣ 2 s T 2 ( Q 2 ) . (14) \begin{aligned}
\iint_{Q_{i+1}}|\tau_h^2u|^2\,dxdt\leq\widehat{C}_{i+1}|h|^{2s}\left[\iint_{Q_i}|\tau_hu|^2\,dxdt+|h|^2\mathcal{T}^2(Q_2)\right] \\
&\leq\widehat{C}_{i+1}|h|^{2s}\mathcal{T}^2(Q_2). \tag*{(14)}
\end{aligned} ∬ Q i + 1 ∣ τ h 2 u ∣ 2 d x d t ≤ C i + 1 ∣ h ∣ 2 s [ ∬ Q i ∣ τ h u ∣ 2 d x d t + ∣ h ∣ 2 T 2 ( Q 2 ) ] ≤ C i + 1 ∣ h ∣ 2 s T 2 ( Q 2 ) . (14)
holds for all i ∈ N 0 i\in\mathbb{N}_0 i ∈ N 0 and for any h ∈ R N ∖ { 0 } h\in\mathbb{R}^N\setminus\{0\} h ∈ R N ∖ { 0 } with ∣ h ∣ ≤ 1 8 ϱ i + 1 |h|\leq\frac{1}{8}\varrho_{i+1} ∣ h ∣ ≤ 8 1 ϱ i + 1 . Next, we apply Lemma 3.5 with γ = s \gamma=s γ = s joining (14) i = 0 _{i=0} i = 0 , and conclude that
∬ Q 1 ∣ τ h u ∣ 2 d x d t ≤ C ∣ h ∣ 2 s [ ϱ 0 − 2 s ∬ Q 0 ∣ u ∣ 2 d x d t + C ^ 1 T 2 ( Q 2 ) ] ≤ C 1 ∣ h ∣ 2 s T 2 ( Q 2 ) , (15) \begin{aligned}
\iint_{Q_1}|\tau_hu|^2\,dxdt\leq\mathfrak{C}|h|^{2s}\left[\varrho_0^{-2s}\iint_{Q_0}|u|^2\,dxdt+\widehat{C}_1\mathcal{T}^2(Q_2)\right] \\
&\leq C_1|h|^{2s}\mathcal{T}^2(Q_2), \tag*{(15)}
\end{aligned} ∬ Q 1 ∣ τ h u ∣ 2 d x d t ≤ C ∣ h ∣ 2 s [ ϱ 0 − 2 s ∬ Q 0 ∣ u ∣ 2 d x d t + C 1 T 2 ( Q 2 ) ] ≤ C 1 ∣ h ∣ 2 s T 2 ( Q 2 ) , (15)
where C = C ( s ) \mathfrak{C}=\mathfrak{C}(s) C = C ( s ) and C 1 = C C ^ 1 ϱ 0 − 2 s C_1=\mathfrak{C}\widehat{C}_1\varrho_0^{-2s} C 1 = C C 1 ϱ 0 − 2 s , holds true for any h ∈ R N ∖ { 0 } h\in\mathbb{R}^N\setminus\{0\} h ∈ R N ∖ { 0 } with ∣ h ∣ ≤ 1 8 ϱ 1 |h|\leq\frac{1}{8}\varrho_1 ∣ h ∣ ≤ 8 1 ϱ 1 . If s ∈ [ 1 2 , 1 ) s\in\left[\frac{1}{2},1\right) s ∈ [ 2 1 , 1 ) , we are done. If s ∈ [ 1 4 , 1 2 ) s\in\left[\frac{1}{4},\frac{1}{2}\right) s ∈ [ 4 1 , 2 1 ) , we first plug (15) back in (14) i = 1 _{i=1} i = 1 to get
∬ Q 2 ∣ τ h 2 u ∣ 2 d x d t ≤ C ^ 2 ∣ h ∣ 2 s [ C 1 ∣ h ∣ 2 s T 2 ( Q 2 ) + ∣ h ∣ 2 T 2 ( Q 2 ) ] \iint_{Q_2}|\tau_h^2u|^2\,dxdt\leq\widehat{C}_2|h|^{2s}\left[C_1|h|^{2s}\mathcal{T}^2(Q_2)+|h|^2\mathcal{T}^2(Q_2)\right] ∬ Q 2 ∣ τ h 2 u ∣ 2 d x d t ≤ C 2 ∣ h ∣ 2 s [ C 1 ∣ h ∣ 2 s T 2 ( Q 2 ) + ∣ h ∣ 2 T 2 ( Q 2 ) ]
≤ 2 C ^ 2 C 1 ∣ h ∣ 2 ⋅ 2 s T 2 ( Q 2 ) ; (16) \leq2\widehat{C}_2C_1|h|^{2\cdot2s}\mathcal{T}^2(Q_2); \tag*{(16)} ≤ 2 C 2 C 1 ∣ h ∣ 2 ⋅ 2 s T 2 ( Q 2 ) ; (16)
then, we apply Lemma 3.5 with γ = 2 s < 1 \gamma=2s<1 γ = 2 s < 1 joining (16) and conclude that
∬ Q 2 ∣ τ h u ∣ 2 d x d t ≤ C ∣ h ∣ 2 ⋅ 2 s [ ϱ 1 − 2 s ∬ Q 1 ∣ u ∣ 2 d x d t + 2 C ^ 2 C 1 T 2 ( Q 2 ) ] ≤ C 2 ∣ h ∣ 2 ⋅ 2 s T 2 ( Q 2 ) \begin{aligned}
\iint_{Q_2}|\tau_hu|^2\,dxdt\leq\mathfrak{C}|h|^{2\cdot2s}\left[\varrho_1^{-2s}\iint_{Q_1}|u|^2\,dxdt+2\widehat{C}_2C_1\mathcal{T}^2(Q_2)\right] \\
&\leq C_2|h|^{2\cdot2s}\mathcal{T}^2(Q_2)
\end{aligned} ∬ Q 2 ∣ τ h u ∣ 2 d x d t ≤ C ∣ h ∣ 2 ⋅ 2 s [ ϱ 1 − 2 s ∬ Q 1 ∣ u ∣ 2 d x d t + 2 C 2 C 1 T 2 ( Q 2 ) ] ≤ C 2 ∣ h ∣ 2 ⋅ 2 s T 2 ( Q 2 )
for C 2 = 2 C C ^ 2 ϱ 1 − 2 s C 1 C_2=2\mathfrak{C}\widehat{C}_2\varrho_1^{-2s}C_1 C 2 = 2 C C 2 ϱ 1 − 2 s C 1 and for any h ∈ R N ∖ { 0 } h\in\mathbb{R}^N\setminus\{0\} h ∈ R N ∖ { 0 } with ∣ h ∣ ≤ 1 8 ϱ 2 |h|\leq\frac{1}{8}\varrho_2 ∣ h ∣ ≤ 8 1 ϱ 2 . Using induction, if s ∈ [ 1 2 i , 1 2 ( i − 1 ) ) s\in\left[\frac{1}{2i},\frac{1}{2(i-1)}\right) s ∈ [ 2 i 1 , 2 ( i − 1 ) 1 ) , i ∈ N i\in\mathbb{N} i ∈ N , then we will have
∬ Q i ∣ τ h u ∣ 2 d x d t ≤ C i ∣ h ∣ i ⋅ 2 s T 2 ( Q 2 ) \iint_{Q_i}|\tau_hu|^2\,dxdt\leq C_i|h|^{i\cdot2s}\mathcal{T}^2(Q_2) ∬ Q i ∣ τ h u ∣ 2 d x d t ≤ C i ∣ h ∣ i ⋅ 2 s T 2 ( Q 2 )
for C i = 2 C C ^ i ϱ i − 1 − 2 s C i − 1 C_i=2\mathfrak{C}\widehat{C}_i\varrho_{i-1}^{-2s}C_{i-1} C i = 2 C C i ϱ i − 1 − 2 s C i − 1 and for any h ∈ R N ∖ { 0 } h\in\mathbb{R}^N\setminus\{0\} h ∈ R N ∖ { 0 } with ∣ h ∣ ≤ 1 8 ϱ i |h|\leq\frac{1}{8}\varrho_i ∣ h ∣ ≤ 8 1 ϱ i . Therefore, for s ∈ ( 0 , 1 ) s\in(0,1) s ∈ ( 0 , 1 ) , there must be i o ∈ N i_o\in\mathbb{N} i o ∈ N such that s ∈ [ 1 2 i o , 1 2 ( i o − 1 ) ) s\in\left[\frac{1}{2i_o},\frac{1}{2(i_o-1)}\right) s ∈ [ 2 i o 1 , 2 ( i o − 1 ) 1 ) . The above estimate then gives that
∬ Q i o ∣ τ h u ∣ 2 d x d t ≤ C i o ∣ h ∣ i o ⋅ 2 s T 2 ( Q 2 ) ≤ C i o ∣ h ∣ 2 T 2 ( Q 2 ) \iint_{Q_{i_o}}|\tau_hu|^2\,dxdt\leq C_{i_o}|h|^{i_o\cdot2s}\mathcal{T}^2(Q_2)\leq C_{i_o}|h|^2\mathcal{T}^2(Q_2) ∬ Q i o ∣ τ h u ∣ 2 d x d t ≤ C i o ∣ h ∣ i o ⋅ 2 s T 2 ( Q 2 ) ≤ C i o ∣ h ∣ 2 T 2 ( Q 2 )
for C i o = ( 2 C ) i o 4 s i o ( i o − 1 ) ∏ j = 1 i o C ^ j C_{i_o}=(2\mathfrak{C})^{i_o}4^{s i_o(i_o-1)}\prod_{j=1}^{i_o}\widehat{C}_j C i o = ( 2 C ) i o 4 s i o ( i o − 1 ) ∏ j = 1 i o C j and for any h ∈ R N ∖ { 0 } h\in\mathbb{R}^N\setminus\{0\} h ∈ R N ∖ { 0 } with ∣ h ∣ ≤ 1 8 ϱ i o |h|\leq\frac{1}{8}\varrho_{i_o} ∣ h ∣ ≤ 8 1 ϱ i o . Note that i o i_o i o is determined by s s s , and hence C i o C_{i_o} C i o depends only on N N N , s s s and Λ \Lambda Λ . Using Lemma 3.3 slice-wise, we can now conclude that
∬ Q i o ∣ D u ∣ 2 d x d t ≤ C T 2 ( Q 2 ) \iint_{Q_{i_o}}|Du|^2\,dxdt\leq C\mathcal{T}^2(Q_2) ∬ Q i o ∣ D u ∣ 2 d x d t ≤ C T 2 ( Q 2 )
for some C = C ( data ) C=C(\text{data}) C = C ( data ) . A covering argument will complete the proof.
W1,∞ \infty ∞ -regularity
With the W 1 , 2 W^{1,2} W 1 , 2 -regularity at hand, we aim to further upgrade the regularity to W 1 , ∞ W^{1,\infty} W 1 , ∞ in this subsection. The approach consists in performing De Giorgi-type iterations based on the energy estimates from Lemma 4.2.
Proposition 5.3. Let s ∈ ( 0 , 1 ) s\in(0,1) s ∈ ( 0 , 1 ) and u u u be a locally bounded, weak solution to (3) in the sense of Definition 2.1 2.1 2.1 . Then,
u ∈ L l o c ∞ ( 0 , T ; W l o c 1 , ∞ ( Ω ) ) . u\in L^\infty_{\mathrm{loc}}(0,T;W^{1,\infty}_{\mathrm{loc}}(\Omega)). u ∈ L loc ∞ ( 0 , T ; W loc 1 , ∞ ( Ω )) . Moreover, if Q 32 R ≡ Q 32 R ( z o ) ⋐ Ω T Q_{32R}\equiv Q_{32R}(z_o)\Subset\Omega_T Q 32 R ≡ Q 32 R ( z o ) ⋐ Ω T , then
∥ D u ∥ L ∞ ( Q R ) ≤ C R [ ∥ u ∥ L ∞ ( Q 8 R ) + Tail ( u ; Q 8 R ) ] \lVert Du\rVert_{L^\infty(Q_R)}\leq\frac{C}{R}\left[\lVert u\rVert_{L^\infty(Q_{8R})}+\operatorname{Tail}(u;Q_{8R})\right] ∥ D u ∥ L ∞ ( Q R ) ≤ R C [ ∥ u ∥ L ∞ ( Q 8 R ) + Tail ( u ; Q 8 R ) ] holds true for some C = C ( data ) > 1 C=C(\text{data})>1 C = C ( data ) > 1 .
Proof. We closely follow the lines of [16 Theorem 1.5] . Suppose u u u is a bounded, weak solution to (3) in Q 32 Q_{32} Q 32 . For k > 0 k>0 k > 0 and n ∈ N 0 n\in\mathbb{N}_0 n ∈ N 0 we introduce
k n = k − 2 − n k , ϱ n = 2 − 1 + 2 − n − 1 , ϱ ~ n = 1 2 ( ϱ n + ϱ n + 1 ) , ϱ ^ n = 3 4 ϱ n + 1 4 ϱ n + 1 , ϱ ‾ n = 1 4 ϱ n + 3 4 ϱ n + 1 , B n = B ϱ n , B ~ n = B ϱ ~ n , B ^ n = B ϱ ^ n , B ‾ n = B ϱ ‾ n , Q n = Q ϱ n , Q ~ n = Q ϱ ~ n , Q ^ n = Q ϱ ^ n , Q ‾ n = Q ϱ ‾ n . \begin{aligned}
k_n&=k-2^{-n}k,\qquad\varrho_n=2^{-1}+2^{-n-1},\\
\widetilde{\varrho}_n&=\frac{1}{2}(\varrho_n+\varrho_{n+1}),\qquad
\widehat{\varrho}_n=\frac{3}{4}\varrho_n+\frac{1}{4}\varrho_{n+1},\qquad
\overline{\varrho}_n=\frac{1}{4}\varrho_n+\frac{3}{4}\varrho_{n+1},\\
B_n&=B_{\varrho_n},\qquad\widetilde{B}_n=B_{\widetilde{\varrho}_n},\qquad
\widehat{B}_n=B_{\widehat{\varrho}_n},\qquad\overline{B}_n=B_{\overline{\varrho}_n},\\
Q_n&=Q_{\varrho_n},\qquad\widetilde{Q}_n=Q_{\widetilde{\varrho}_n},\qquad
\widehat{Q}_n=Q_{\widehat{\varrho}_n},\qquad\overline{Q}_n=Q_{\overline{\varrho}_n}.
\end{aligned} k n ϱ n B n Q n = k − 2 − n k , ϱ n = 2 − 1 + 2 − n − 1 , = 2 1 ( ϱ n + ϱ n + 1 ) , ϱ n = 4 3 ϱ n + 4 1 ϱ n + 1 , ϱ n = 4 1 ϱ n + 4 3 ϱ n + 1 , = B ϱ n , B n = B ϱ n , B n = B ϱ n , B n = B ϱ n , = Q ϱ n , Q n = Q ϱ n , Q n = Q ϱ n , Q n = Q ϱ n .
Observe that Q n + 1 ⊂ Q ‾ n ⊂ Q ^ n ⊂ Q ~ n ⊂ Q n Q_{n+1}\subset\overline{Q}_n\subset\widehat{Q}_n\subset\widetilde{Q}_n\subset Q_n Q n + 1 ⊂ Q n ⊂ Q n ⊂ Q n ⊂ Q n . Choose a cutoff function ζ \zeta ζ in Q n Q_n Q n , vanishing outside Q ^ n \widehat{Q}_n Q n and equal to 1 1 1 in Q ~ n \widetilde{Q}_n Q n , such that
∣ D ζ ∣ ≤ 2 n + 5 and ∣ ∂ t ζ ∣ ≤ 2 n + 5 . |D\zeta|\leq2^{n+5}\qquad\text{and}\qquad|\partial_t\zeta|\leq2^{n+5}. ∣ D ζ ∣ ≤ 2 n + 5 and ∣ ∂ t ζ ∣ ≤ 2 n + 5 .
Next, we introduce
w ( x , t ) : = ( τ h u ( x , t ) − k n + 1 ( t ) ) + , w ~ ( x , t ) : = ( τ h u ( x , t ) − k n ( t ) ) + , k n ( t ) : = ℓ ( t ) + k n , ℓ ( t ) : = γ ∫ − 1 t T ( τ ; B 4 ) d τ \begin{aligned}
w(x,t)&:=(\tau_hu(x,t)-k_{n+1}(t))_+,\qquad
\widetilde{w}(x,t):=(\tau_hu(x,t)-k_n(t))_+,\\
k_n(t)&:=\ell(t)+k_n,\qquad
\ell(t):=\gamma\int_{-1}^{t}\mathcal{T}(\tau;B_4)\,d\tau
\end{aligned} w ( x , t ) k n ( t ) := ( τ h u ( x , t ) − k n + 1 ( t ) ) + , w ( x , t ) := ( τ h u ( x , t ) − k n ( t ) ) + , := ℓ ( t ) + k n , ℓ ( t ) := γ ∫ − 1 t T ( τ ; B 4 ) d τ
for some γ > 0 \gamma>0 γ > 0 to be specified, where we defined as in (7)
T ( t ; B 4 ) : = ∥ u ( t ) ∥ L ∞ ( B 4 ) + Tail ( u ( t ) ; B 4 ) . \mathcal{T}(t;B_4):=\lVert u(t)\rVert_{L^\infty(B_4)}+\operatorname{Tail}(u(t);B_4). T ( t ; B 4 ) := ∥ u ( t ) ∥ L ∞ ( B 4 ) + Tail ( u ( t ) ; B 4 ) .
The energy inequality in Lemma 4.2 on Q n Q_n Q n with the above choice of ζ \zeta ζ and w w w gives:
ess sup t ∈ [ − ϱ ~ n 2 s , 0 ] ∫ B ^ n × { t } w 2 d x + Λ − 1 ∭ Q ~ n ∣ w ( x , t ) − w ( y , t ) ∣ 2 ∣ x − y ∣ N + 2 s d x d y d t ≤ 2 Λ ∭ Q n max { w 2 ( x , t ) , w 2 ( y , t ) } ∣ ζ ( y , t ) − ζ ( x , t ) ∣ 2 ∣ x − y ∣ N + 2 s d y d x d t + 4 Λ ∬ Q n [ ζ 2 w 2 ( x , t ) + k n + 1 ( t ) ζ 2 w ( x , t ) ] ∫ R N ∖ B n d y ∣ x − y ∣ N + 2 s d x d t + 4 Λ ∬ Q n ζ 2 w ( x , t ) ∣ ∫ R N ∖ B n τ h u ( y , t ) ∣ x − y ∣ N + 2 s d y ∣ d x d t − 2 ∬ Q n ℓ ′ ζ 2 w d x d t + ∬ Q n w 2 ∣ ∂ t ζ 2 ∣ d x d t = : I + I I + I I I − I V + V . (17) \begin{aligned}
&\operatorname*{ess\,sup}_{t\in[-\widetilde{\varrho}_n^{2s},0]}
\int_{\widehat{B}_n\times\{t\}}w^2\,dx
+\Lambda^{-1}\iiint_{\widetilde{Q}_n}
\frac{|w(x,t)-w(y,t)|^2}{|x-y|^{N+2s}}\,dx\,dy\,dt\\
&\leq2\Lambda\iiint_{Q_n}
\max\{w^2(x,t),w^2(y,t)\}
\frac{|\zeta(y,t)-\zeta(x,t)|^2}{|x-y|^{N+2s}}\,dy\,dx\,dt\\
&\quad+4\Lambda\iint_{Q_n}
[\zeta^2w^2(x,t)+k_{n+1}(t)\zeta^2w(x,t)]
\int_{\mathbb{R}^N\setminus B_n}\frac{dy}{|x-y|^{N+2s}}\,dx\,dt\\
&\quad+4\Lambda\iint_{Q_n}\zeta^2w(x,t)
\left|\int_{\mathbb{R}^N\setminus B_n}
\frac{\tau_hu(y,t)}{|x-y|^{N+2s}}\,dy\right|\,dx\,dt\\
&\quad-2\iint_{Q_n}\ell'\zeta^2w\,dx\,dt
+\iint_{Q_n}w^2|\partial_t\zeta^2|\,dx\,dt\\
&=:I+II+III-IV+V. \tag*{(17)}
\end{aligned} t ∈ [ − ϱ n 2 s , 0 ] ess sup ∫ B n × { t } w 2 d x + Λ − 1 ∭ Q n ∣ x − y ∣ N + 2 s ∣ w ( x , t ) − w ( y , t ) ∣ 2 d x d y d t ≤ 2Λ ∭ Q n max { w 2 ( x , t ) , w 2 ( y , t )} ∣ x − y ∣ N + 2 s ∣ ζ ( y , t ) − ζ ( x , t ) ∣ 2 d y d x d t + 4Λ ∬ Q n [ ζ 2 w 2 ( x , t ) + k n + 1 ( t ) ζ 2 w ( x , t )] ∫ R N ∖ B n ∣ x − y ∣ N + 2 s d y d x d t + 4Λ ∬ Q n ζ 2 w ( x , t ) ∫ R N ∖ B n ∣ x − y ∣ N + 2 s τ h u ( y , t ) d y d x d t − 2 ∬ Q n ℓ ′ ζ 2 w d x d t + ∬ Q n w 2 ∣ ∂ t ζ 2 ∣ d x d t =: I + I I + I I I − I V + V . (17)
Here, the symbols Q ~ n \widetilde{Q}_n Q n and Q n Q_n Q n should be self-explanatory. Estimating the first and fifth terms is straightforward, cf. [16 ] . Namely, we have I I I and V V V bounded by
C 2 2 2 n ∬ Q n w ~ 2 d x d t . C^2 2^{2n}\iint_{Q_n}\widetilde{w}^2\,dx\,dt. C 2 2 2 n ∬ Q n w 2 d x d t .
For II we observe that ∣ x ∣ ≤ ϱ ^ n |x| \le\widehat{\varrho}_n ∣ x ∣ ≤ ϱ n due to spt ζ ( ⋅ , t ) ⊂ B n \operatorname{spt}\zeta(\cdot,t) \subset B_n spt ζ ( ⋅ , t ) ⊂ B n and ∣ y ∣ ≥ ϱ n |y| \ge\varrho_n ∣ y ∣ ≥ ϱ n . Then,
∣ y − x ∣ ∣ y ∣ ≥ 1 − ϱ ^ n ϱ n = 1 4 ( ϱ n − ϱ n + 1 ϱ n ) ≥ 1 2 n + 4 . (18) \frac{|y-x|}{|y|} \ge1-\frac{\widehat{\varrho}_n}{\varrho_n} = \frac{1}{4}\left(\frac{\varrho_n-\varrho_{n+1}}{\varrho_n}\right) \ge\frac{1}{2^{n+4}}. \tag*{(18)} ∣ y ∣ ∣ y − x ∣ ≥ 1 − ϱ n ϱ n = 4 1 ( ϱ n ϱ n − ϱ n + 1 ) ≥ 2 n + 4 1 . (18)
By (18) , we have
I I ≤ C 2 n ( N + 2 s ) ∬ Q n [ ζ 2 w 2 ( x , t ) + k n + 1 ( t ) ζ 2 w ( x , t ) ] d x d t . \mathrm{II} \le C2^{n(N+2s)}\iint_{Q_n}\left[\zeta^2w^2(x,t)+k_{n+1}(t)\zeta^2w(x,t)\right]\,dxdt. II ≤ C 2 n ( N + 2 s ) ∬ Q n [ ζ 2 w 2 ( x , t ) + k n + 1 ( t ) ζ 2 w ( x , t ) ] d x d t .
To continue, we recall w ~ = ( τ h u − k n ) + \widetilde{w}=(\tau_hu-k_n)_+ w = ( τ h u − k n ) + and estimate
∣ A n ∣ : = ∣ { τ h u ≥ k n + 1 } ∩ Q n ∣ ≤ 2 2 ( n + 1 ) k 2 ∬ Q n w ~ 2 d x d t . (19) |A_n|:=\left|\{\tau_hu\ge k_{n+1}\}\cap Q_n\right|\le\frac{2^{2(n+1)}}{k^2}\iint_{Q_n}\widetilde{w}^2\,dxdt. \tag*{(19)} ∣ A n ∣ := ∣ { τ h u ≥ k n + 1 } ∩ Q n ∣ ≤ k 2 2 2 ( n + 1 ) ∬ Q n w 2 d x d t . (19)
from which, applying Hölder’s inequality and using w ≤ w ~ w\le\widetilde{w} w ≤ w , we deduce
∬ Q n w d x d t ≤ ( ∬ Q n w 2 d x d t ) 1 2 ∣ A n ∣ 1 2 ≤ 2 n + 1 k ∬ Q n w ~ 2 d x d t . (20) \iint_{Q_n}w\,dxdt\le\left(\iint_{Q_n}w^2\,dxdt\right)^{\frac{1}{2}}|A_n|^{\frac{1}{2}}\le\frac{2^{n+1}}{k}\iint_{Q_n}\widetilde{w}^2\,dxdt. \tag*{(20)} ∬ Q n w d x d t ≤ ( ∬ Q n w 2 d x d t ) 2 1 ∣ A n ∣ 2 1 ≤ k 2 n + 1 ∬ Q n w 2 d x d t . (20)
Therefore, upon using (20) and k n + 1 ( t ) ≤ k + ℓ ( 0 ) k_{n+1}(t)\le k+\ell(0) k n + 1 ( t ) ≤ k + ℓ ( 0 ) , we have
I I ≤ C 2 ( N + 2 s + 1 ) n [ 1 + ℓ ( 0 ) k ] ∬ Q n w ~ 2 d x d t . \mathrm{II}\le C2^{(N+2s+1)n}\left[1+\frac{\ell(0)}{k}\right]\iint_{Q_n}\widetilde{w}^2\,dxdt. II ≤ C 2 ( N + 2 s + 1 ) n [ 1 + k ℓ ( 0 ) ] ∬ Q n w 2 d x d t .
Now, we will have a closer look at the third and forth terms, which need to be packed. To begin with, we decompose III as follows:
I I I ≤ 4 Λ ∬ Q n ζ 2 w ( x , t ) ∣ ∫ R N ∖ B 2 τ h u ( y , t ) ∣ x − y ∣ N + 2 s d y ∣ d x d t + 4 Λ ∬ Q n ζ 2 w ( x , t ) ∣ ∫ B 2 ∖ B n τ h u ( y , t ) ∣ x − y ∣ N + 2 s d y ∣ d x d t = : I I I 1 + I I I 2 . \begin{aligned}
\mathrm{III} &\le4\Lambda\iint_{Q_n}\zeta^2w(x,t)\left|\int_{\mathbb{R}^N\setminus B_2}\frac{\tau_hu(y,t)}{|x-y|^{N+2s}}\,dy\right|\,dxdt \\
&\quad+4\Lambda\iint_{Q_n}\zeta^2w(x,t)\left|\int_{B_2\setminus B_n}\frac{\tau_hu(y,t)}{|x-y|^{N+2s}}\,dy\right|\,dxdt \\
&=: \mathrm{III}_1+\mathrm{III}_2.
\end{aligned} III ≤ 4Λ ∬ Q n ζ 2 w ( x , t ) ∫ R N ∖ B 2 ∣ x − y ∣ N + 2 s τ h u ( y , t ) d y d x d t + 4Λ ∬ Q n ζ 2 w ( x , t ) ∫ B 2 ∖ B n ∣ x − y ∣ N + 2 s τ h u ( y , t ) d y d x d t =: III 1 + III 2 .
For I I I 1 \mathrm{III}_1 III 1 we apply Lemma 3.6 with ( r , R ) = ( 1 , 2 ) (r,R)=(1,2) ( r , R ) = ( 1 , 2 ) and find that for any 0 < ∣ h ∣ ≤ 1 4 0<|h|\le\frac{1}{4} 0 < ∣ h ∣ ≤ 4 1 ,
I I I 1 ≤ 4 Λ C ∣ h ∣ ∬ Q n T ( t ; B 4 ) ζ 2 w d x d t ≤ Λ C ∗ ∣ h ∣ ∬ Q n T ( t ; B 4 ) ζ 2 w d x d t . \mathrm{III}_1\le4\Lambda C|h|\iint_{Q_n}\mathcal{T}(t;B_4)\zeta^2w\,dxdt\le\Lambda C_*|h|\iint_{Q_n}\mathcal{T}(t;B_4)\zeta^2w\,dxdt. III 1 ≤ 4Λ C ∣ h ∣ ∬ Q n T ( t ; B 4 ) ζ 2 w d x d t ≤ Λ C ∗ ∣ h ∣ ∬ Q n T ( t ; B 4 ) ζ 2 w d x d t .
Then, by definition of ℓ ( t ) \ell(t) ℓ ( t ) we find that I I I 1 − I V ≤ 0 \mathrm{III}_1-\mathrm{IV}\le0 III 1 − IV ≤ 0 , if we choose
2 γ = Λ C ∗ ∣ h ∣ . 2\gamma=\Lambda C_*|h|. 2 γ = Λ C ∗ ∣ h ∣.
This choice of γ \gamma γ also fixes
ℓ ( t ) = 1 2 Λ C ∗ ∣ h ∣ ∫ − 1 t T ( τ ; B 4 ) d τ . \ell(t)=\frac{1}{2}\Lambda C_*|h|\int_{-1}^{t}\mathcal{T}(\tau;B_4)\,d\tau. ℓ ( t ) = 2 1 Λ C ∗ ∣ h ∣ ∫ − 1 t T ( τ ; B 4 ) d τ .
For I I I 2 \mathrm{III}_2 III 2 we use (18) and (20) to estimate
I I I 2 ≤ C 2 ( N + 2 s ) n ∬ Q n ζ 2 w ( x , t ) ∫ B 2 ∖ B n ∣ τ h u ( y , t ) ∣ ∣ y ∣ N + 2 s d y d x d t ≤ C 2 ( N + 2 s ) n ess sup t ∈ [ − 1 , 0 ] ∫ B 2 ∣ τ h u ( y , t ) ∣ d y ∬ Q n ζ 2 w d x d t ≤ C 2 ( N + 2 s + 1 ) n J k ∬ Q n w ~ 2 d x d t . \begin{aligned}
\mathrm{III}_2
&\le C2^{(N+2s)n}\iint_{Q_n}\zeta^2w(x,t)\int_{B_2\setminus B_n}\frac{|\tau_hu(y,t)|}{|y|^{N+2s}}\,dy\,dxdt \\
&\le C2^{(N+2s)n}\operatorname*{ess\,sup}_{t\in[-1,0]}\int_{B_2}|\tau_hu(y,t)|\,dy\iint_{Q_n}\zeta^2w\,dxdt \\
&\le C2^{(N+2s+1)n}\frac{\mathcal{J}}{k}\iint_{Q_n}\widetilde{w}^2\,dxdt.
\end{aligned} III 2 ≤ C 2 ( N + 2 s ) n ∬ Q n ζ 2 w ( x , t ) ∫ B 2 ∖ B n ∣ y ∣ N + 2 s ∣ τ h u ( y , t ) ∣ d y d x d t ≤ C 2 ( N + 2 s ) n t ∈ [ − 1 , 0 ] ess sup ∫ B 2 ∣ τ h u ( y , t ) ∣ d y ∬ Q n ζ 2 w d x d t ≤ C 2 ( N + 2 s + 1 ) n k J ∬ Q n w 2 d x d t .
Therefore, we have
I I I − I V ≤ I I I 1 + I I I 2 − I V ≤ I I I 2 ≤ C 2 ( N + 2 s + 1 ) n J k ∬ Q n w ~ 2 d x d t , \mathrm{III}-\mathrm{IV}\le\mathrm{III}_1+\mathrm{III}_2-\mathrm{IV}\le\mathrm{III}_2\le C2^{(N+2s+1)n}\frac{\mathcal{J}}{k}\iint_{Q_n}\widetilde{w}^2\,dxdt, III − IV ≤ III 1 + III 2 − IV ≤ III 2 ≤ C 2 ( N + 2 s + 1 ) n k J ∬ Q n w 2 d x d t ,
and, up to now we have obtained
I + I I + I I I − I V + V ≤ C 2 ( N + 2 s + 1 ) n [ 1 + ℓ ( 0 ) k + J k ] ∭ Q ~ n w ~ 2 d x d t . \mathrm{I}+\mathrm{II}+\mathrm{III}-\mathrm{IV}+\mathrm{V}\le C2^{(N+2s+1)n}\left[1+\frac{\ell(0)}{k}+\frac{\mathcal{J}}{k}\right]\iiint_{\widetilde{Q}_n}\widetilde{w}^{2}\,dx\,dt. I + II + III − IV + V ≤ C 2 ( N + 2 s + 1 ) n [ 1 + k ℓ ( 0 ) + k J ] ∭ Q n w 2 d x d t .
Let us stipulate that
ℓ ( 0 ) + J ≤ k (21) \ell(0)+\mathcal{J}\le k \tag*{(21)} ℓ ( 0 ) + J ≤ k (21)
such that the terms in the bracket are bounded by 2. Collecting all the estimates on the right-hand side of (17) gives us
ess sup t ∈ [ − ϱ e n 2 s , 0 ] ∫ B ~ n × { t } w 2 d x + Λ − 1 ∭ Q ~ n ∣ w ( x , t ) − w ( y , t ) ∣ 2 ∣ x − y ∣ N + 2 s d x d y d t ≤ C 2 ( N + 2 s + 1 ) n ∬ Q n w ~ 2 d x d t . (22) \begin{aligned}
\operatorname*{ess\,sup}_{t\in[-\varrho e_n^{2s},0]}\int_{\widetilde{B}_n\times\{t\}}w^{2}\,dx
+\Lambda^{-1}\iiint_{\widetilde{Q}_n}\frac{\lvert w(x,t)-w(y,t)\rvert^{2}}{\lvert x-y\rvert^{N+2s}}\,dx\,dy\,dt\\
\le C2^{(N+2s+1)n}\iint_{Q_n}\widetilde{w}^{2}\,dx\,dt. \tag*{(22)}
\end{aligned} t ∈ [ − ϱ e n 2 s , 0 ] ess sup ∫ B n × { t } w 2 d x + Λ − 1 ∭ Q n ∣ x − y ∣ N + 2 s ∣ w ( x , t ) − w ( y , t ) ∣ 2 d x d y d t ≤ C 2 ( N + 2 s + 1 ) n ∬ Q n w 2 d x d t . (22)
At this point consider a cutoff function ϕ ∈ C 1 ( Q ~ n , [ 0 , 1 ] ) \phi\in C^{1}(\widetilde{Q}_n,[0,1]) ϕ ∈ C 1 ( Q n , [ 0 , 1 ]) , which vanishes outside of Q ‾ n \overline{Q}_n Q n , is equal to 1 in Q n + 1 Q_{n+1} Q n + 1 and satisfies ∣ D ϕ ∣ ≤ 2 n + 4 \lvert D\phi\rvert\le2^{n+4} ∣ D ϕ ∣ ≤ 2 n + 4 . By applying Hölder’s inequality and the embedding from Proposition 3.7 , we obtain
∬ Q n + 1 w 2 d x d t ≤ ∬ Q ~ n ( ϕ w ) 2 d x d t ≤ [ ∭ Q ~ n ∣ ϕ w ( x , t ) − ϕ w ( y , t ) ∣ 2 ∣ x − y ∣ N + 2 s d x d y d t + ∬ Q ~ n ( w ϕ ) 2 d x d t ] N N + 2 s ∣ A n ∣ 2 s N + 2 s ⋅ [ ess sup t ∈ [ − ϱ e n 2 s , 0 ] ∫ B ~ n ( w ϕ ) 2 d x ] 2 s N + 2 s ∣ A n ∣ 2 s N + 2 s , (23) \begin{aligned}
\iint_{Q_{n+1}}w^{2}\,dx\,dt
&\le\iint_{\widetilde{Q}_n}(\phi w)^{2}\,dx\,dt\\
&\le\left[\iiint_{\widetilde{Q}_n}\frac{\lvert\phi w(x,t)-\phi w(y,t)\rvert^{2}}{\lvert x-y\rvert^{N+2s}}\,dx\,dy\,dt+\iint_{\widetilde{Q}_n}(w\phi)^{2}\,dx\,dt\right]^{\frac{N}{N+2s}}|A_n|^{\frac{2s}{N+2s}}\\
&\quad\cdot\left[\operatorname*{ess\,sup}_{t\in[-\varrho e_n^{2s},0]}\int_{\widetilde{B}_n}(w\phi)^{2}\,dx\right]^{\frac{2s}{N+2s}}|A_n|^{\frac{2s}{N+2s}}, \tag*{(23)}
\end{aligned} ∬ Q n + 1 w 2 d x d t ≤ ∬ Q n ( ϕw ) 2 d x d t ≤ [ ∭ Q n ∣ x − y ∣ N + 2 s ∣ ϕw ( x , t ) − ϕw ( y , t ) ∣ 2 d x d y d t + ∬ Q n ( w ϕ ) 2 d x d t ] N + 2 s N ∣ A n ∣ N + 2 s 2 s ⋅ [ t ∈ [ − ϱ e n 2 s , 0 ] ess sup ∫ B n ( w ϕ ) 2 d x ] N + 2 s 2 s ∣ A n ∣ N + 2 s 2 s , (23)
for A n A_n A n being defined according to (19) . The numerator of the integrand inside the fractional norm is estimated by
∣ ϕ w ( x , t ) − ϕ w ( y , t ) ∣ 2 ≤ 2 ϕ 2 ( x , t ) ∣ w ( x , t ) − w ( y , t ) ∣ 2 + 2 w 2 ( y , t ) ∣ ϕ ( x , t ) − ϕ ( y , t ) ∣ 2 . \lvert\phi w(x,t)-\phi w(y,t)\rvert^{2}
\le2\phi^{2}(x,t)\lvert w(x,t)-w(y,t)\rvert^{2}+2w^{2}(y,t)\lvert\phi(x,t)-\phi(y,t)\rvert^{2}. ∣ ϕw ( x , t ) − ϕw ( y , t ) ∣ 2 ≤ 2 ϕ 2 ( x , t ) ∣ w ( x , t ) − w ( y , t ) ∣ 2 + 2 w 2 ( y , t ) ∣ ϕ ( x , t ) − ϕ ( y , t ) ∣ 2 .
Therefore, we deduce that
∭ Q ~ n ∣ ϕ w ( x , t ) − ϕ w ( y , t ) ∣ 2 ∣ x − y ∣ N + 2 s d x d y d t ≤ 2 ∭ Q ~ n ∣ w ( x , t ) − w ( y , t ) ∣ 2 ∣ x − y ∣ N + 2 s d x d y d t + 2 ∭ Q ~ n w 2 ( y , t ) ∣ ϕ ( x , t ) − ϕ ( y , t ) ∣ 2 ∣ x − y ∣ N + 2 s d x d y d t ≤ 2 ∭ Q ~ n ∣ w ( x , t ) − w ( y , t ) ∣ 2 ∣ x − y ∣ N + 2 s d x d y d t + C 2 2 n ∬ Q ~ n w ~ 2 d y d t ≤ C 2 ( N + 2 s + 1 ) n ∬ Q n w ~ 2 d x d t . \begin{aligned}
\iiint_{\widetilde{Q}_n}\frac{\lvert\phi w(x,t)-\phi w(y,t)\rvert^{2}}{\lvert x-y\rvert^{N+2s}}\,dx\,dy\,dt
&\le2\iiint_{\widetilde{Q}_n}\frac{\lvert w(x,t)-w(y,t)\rvert^{2}}{\lvert x-y\rvert^{N+2s}}\,dx\,dy\,dt\\
&\quad+2\iiint_{\widetilde{Q}_n}\frac{w^{2}(y,t)\lvert\phi(x,t)-\phi(y,t)\rvert^{2}}{\lvert x-y\rvert^{N+2s}}\,dx\,dy\,dt\\
&\le2\iiint_{\widetilde{Q}_n}\frac{\lvert w(x,t)-w(y,t)\rvert^{2}}{\lvert x-y\rvert^{N+2s}}\,dx\,dy\,dt+C2^{2n}\iint_{\widetilde{Q}_n}\widetilde{w}^{2}\,dy\,dt\\
&\le C2^{(N+2s+1)n}\iint_{Q_n}\widetilde{w}^{2}\,dx\,dt.
\end{aligned} ∭ Q n ∣ x − y ∣ N + 2 s ∣ ϕw ( x , t ) − ϕw ( y , t ) ∣ 2 d x d y d t ≤ 2 ∭ Q n ∣ x − y ∣ N + 2 s ∣ w ( x , t ) − w ( y , t ) ∣ 2 d x d y d t + 2 ∭ Q n ∣ x − y ∣ N + 2 s w 2 ( y , t ) ∣ ϕ ( x , t ) − ϕ ( y , t ) ∣ 2 d x d y d t ≤ 2 ∭ Q n ∣ x − y ∣ N + 2 s ∣ w ( x , t ) − w ( y , t ) ∣ 2 d x d y d t + C 2 2 n ∬ Q n w 2 d y d t ≤ C 2 ( N + 2 s + 1 ) n ∬ Q n w 2 d x d t .
Note that we used (22) for the last estimate, where the constant Λ − 1 \Lambda^{-1} Λ − 1 was brought to the right-hand side and absorbed into C C C , as well as w ≤ w ~ w\le\widetilde{w} w ≤ w and Q ~ n ⊂ Q n \widetilde{Q}_n\subset Q_n Q n ⊂ Q n . Similarly, the sup-in-time term in (23) can be bounded by using (22) , whereas for ∣ A n ∣ \lvert A_n\rvert ∣ A n ∣ we use (19) . In this way, we deduce from (23) that
∬ Q n + 1 w 2 d x d t ≤ C [ 2 ( N + 2 s + 1 ) n ∬ Q n w ~ 2 d x d t + ∬ Q n w ~ 2 d x d t ] N N + 2 s ⋅ [ 2 ( N + 2 s + 2 ) n ∬ Q n w ~ 2 d x d t ] 2 s N + 2 s ⋅ [ 2 2 ( n + 1 ) k 2 ∬ Q n w ~ 2 d x d t ] 2 s N + 2 s . ≤ C b n k 2 s N + 2 s [ ∬ Q n w ~ 2 d x d t ] 1 + 2 s N + 2 s (24) \begin{aligned}
\iint_{Q_{n+1}}w^{2}\,dx\,dt
\le C\left[2^{(N+2s+1)n}\iint_{Q_n}\widetilde{w}^{2}\,dx\,dt+\iint_{Q_n}\widetilde{w}^{2}\,dx\,dt\right]^{\frac{N}{N+2s}}\\
\quad\cdot\left[2^{(N+2s+2)n}\iint_{Q_n}\widetilde{w}^{2}\,dx\,dt\right]^{\frac{2s}{N+2s}}
\cdot\left[\frac{2^{2(n+1)}}{k^{2}}\iint_{Q_n}\widetilde{w}^{2}\,dx\,dt\right]^{\frac{2s}{N+2s}}. \\
&\le\frac{Cb^n}{k^{\frac{2s}{N+2s}}}\left[\iint_{Q_n}\widetilde{w}^2\,\mathrm{d}x\mathrm{d}t\right]^{1+\frac{2s}{N+2s}} \tag*{(24)}
\end{aligned} ∬ Q n + 1 w 2 d x d t ≤ C [ 2 ( N + 2 s + 1 ) n ∬ Q n w 2 d x d t + ∬ Q n w 2 d x d t ] N + 2 s N ⋅ [ 2 ( N + 2 s + 2 ) n ∬ Q n w 2 d x d t ] N + 2 s 2 s ⋅ [ k 2 2 2 ( n + 1 ) ∬ Q n w 2 d x d t ] N + 2 s 2 s . ≤ k N + 2 s 2 s C b n [ ∬ Q n w 2 d x d t ] 1 + N + 2 s 2 s (24)
for some b = b ( N ) > 1 b=b(N)>1 b = b ( N ) > 1 . Defining
Y n : = 1 k 2 ∬ Q n ( τ h u − ℓ ( t ) − k n ) + 2 d x d t Y_n:=\frac{1}{k^2}\iint_{Q_n}\left(\tau_hu-\ell(t)-k_n\right)_+^2\,\mathrm{d}x\mathrm{d}t Y n := k 2 1 ∬ Q n ( τ h u − ℓ ( t ) − k n ) + 2 d x d t
and recalling the definitions of w w w and w ~ \widetilde{w} w will enable us as rewrite (24) as
Y n + 1 ≤ C b n Y n 1 + 2 s N + 2 s . Y_{n+1}\le Cb^nY_n^{1+\frac{2s}{N+2s}}. Y n + 1 ≤ C b n Y n 1 + N + 2 s 2 s .
This encodes a fast geometric convergence, cf. [8 Chapter I, Lemma 4.1] . In fact, if
Y 0 = 1 k 2 ∬ Q 0 ( τ h u − ℓ ( t ) ) + 2 d x d t ≤ C − N + 2 s 2 s b − ( N + 2 s 2 s ) 2 (25) Y_0=\frac{1}{k^2}\iint_{Q_0}\left(\tau_hu-\ell(t)\right)_+^2\,\mathrm{d}x\mathrm{d}t\le C^{-\frac{N+2s}{2s}}b^{-\left(\frac{N+2s}{2s}\right)^2} \tag*{(25)} Y 0 = k 2 1 ∬ Q 0 ( τ h u − ℓ ( t ) ) + 2 d x d t ≤ C − 2 s N + 2 s b − ( 2 s N + 2 s ) 2 (25)
then
lim n → ∞ ∬ Q n ( τ h u − ℓ ( t ) − k n ) + 2 d x d t = 0 , \lim_{n\to\infty}\iint_{Q_n}\left(\tau_hu-\ell(t)-k_n\right)_+^2\,\mathrm{d}x\mathrm{d}t=0, n → ∞ lim ∬ Q n ( τ h u − ℓ ( t ) − k n ) + 2 d x d t = 0 ,
which amounts to
τ h u ≤ ℓ + k a.e. in Q 1 2 . \tau_hu\le\ell+k\quad\text{a.e. in }Q_{\frac12}. τ h u ≤ ℓ + k a.e. in Q 2 1 .
We properly choose k k k to fulfill both (21) and (25) , and obtain from the last line that
ess sup Q 1 2 τ h u ≤ J + C ∣ h ∣ ∫ − 1 0 T ( t ; B 4 ) d t + C [ ∬ Q 1 ∣ τ h u ∣ 2 d x d t ] 1 2 = : J + L 1 + L 2 . (26) \begin{aligned}
\operatorname*{ess\,sup}_{Q_{\frac12}}\tau_hu\le\mathcal{J}+C|h|\int_{-1}^{0}\mathcal{T}(t;B_4)\,\mathrm{d}t+C\left[\iint_{Q_1}|\tau_hu|^2\,\mathrm{d}x\mathrm{d}t\right]^{\frac12} \\
&=: \mathcal{J}+\mathcal{L}_1+\mathcal{L}_2. \tag*{(26)}
\end{aligned} Q 2 1 ess sup τ h u ≤ J + C ∣ h ∣ ∫ − 1 0 T ( t ; B 4 ) d t + C [ ∬ Q 1 ∣ τ h u ∣ 2 d x d t ] 2 1 =: J + L 1 + L 2 . (26)
Our next aim is to estimate the right-hand side in the above display in terms of ∣ h ∣ |h| ∣ h ∣ . First, we will estimate J \mathcal{J} J by the energy inequality from Lemma 4.1. Indeed, we omit the second term on the left of the energy inequality and choose proper cut-off functions to get
J = ess sup t ∈ [ − 1 , 0 ] ∫ B 2 ∣ τ h u ∣ d x ≤ C [ ess sup t ∈ [ − 1 , 0 ] ∫ B 2 ∣ τ h u ∣ 2 d x ] 1 2 ≤ C [ ∬ Q 4 ∣ τ h u ∣ 2 d x d t ] 1 2 + C ∣ h ∣ [ ∫ − 2 0 T ( t ; B 4 ) d t ] = : J 1 + J 2 , \begin{aligned}
\mathcal{J}=\operatorname*{ess\,sup}_{t\in[-1,0]}\int_{B_2}|\tau_hu|\,\mathrm{d}x\le C\left[\operatorname*{ess\,sup}_{t\in[-1,0]}\int_{B_2}|\tau_hu|^2\,\mathrm{d}x\right]^{\frac12} \\
&\le C\left[\iint_{Q_4}|\tau_hu|^2\,\mathrm{d}x\mathrm{d}t\right]^{\frac12}+C|h|\left[\int_{-2}^{0}\mathcal{T}(t;B_4)\,\mathrm{d}t\right] \\
&=:\mathcal{J}_1+\mathcal{J}_2,
\end{aligned} J = t ∈ [ − 1 , 0 ] ess sup ∫ B 2 ∣ τ h u ∣ d x ≤ C [ t ∈ [ − 1 , 0 ] ess sup ∫ B 2 ∣ τ h u ∣ 2 d x ] 2 1 ≤ C [ ∬ Q 4 ∣ τ h u ∣ 2 d x d t ] 2 1 + C ∣ h ∣ [ ∫ − 2 0 T ( t ; B 4 ) d t ] =: J 1 + J 2 ,
where we defined as in (2.2)
T ( t ; B 4 ) : = ∥ u ( t ) ∥ L ∞ ( B 4 ) + Tail ( u ( t ) ; B 4 ) . \mathcal{T}(t;B_4):=\|u(t)\|_{L^\infty(B_4)}+\operatorname{Tail}(u(t);B_4). T ( t ; B 4 ) := ∥ u ( t ) ∥ L ∞ ( B 4 ) + Tail ( u ( t ) ; B 4 ) .
In order to estimate J 1 \mathcal{J}_1 J 1 , we apply Proposition 5.1 to get that
J 1 ≤ C ∣ h ∣ T , \mathcal{J}_1\le C|h|\mathcal{T}, J 1 ≤ C ∣ h ∣ T ,
where we defined as in (2.3)
T ≡ T ( Q 8 ) : = ∥ u ∥ L ∞ ( Q 8 ) + Tail ( u ; Q 8 ) . \mathcal{T}\equiv\mathcal{T}(Q_8):=\|u\|_{L^\infty(Q_8)}+\operatorname{Tail}(u;Q_8). T ≡ T ( Q 8 ) := ∥ u ∥ L ∞ ( Q 8 ) + Tail ( u ; Q 8 ) .
By Lemma 3.2 it is straightforward to see that T ( t ; B 4 ) ≤ C T ( t ; B 8 ) \mathcal{T}(t;B_4)\le C\mathcal{T}(t;B_8) T ( t ; B 4 ) ≤ C T ( t ; B 8 ) . Consequently,
J 2 ≤ C ∣ h ∣ T , \mathcal{J}_2\le C|h|\mathcal{T}, J 2 ≤ C ∣ h ∣ T ,
and, therefore, we get that
J ≤ J 1 + J 2 ≤ C ∣ h ∣ T . \mathcal{J}\le\mathcal{J}_1+\mathcal{J}_2\le C|h|\mathcal{T}. J ≤ J 1 + J 2 ≤ C ∣ h ∣ T .
The other two terms in (5.13) will be bounded by the same quantity. Indeed, by Lemma 3.2 it is not hard to see that
L 1 ≤ C ∣ h ∣ ∫ − 2 0 T ( t ; B 8 ) d t ≤ C ∣ h ∣ T . \mathcal{L}_1\le C|h|\int_{-2}^{0}\mathcal{T}(t;B_8)\,\mathrm{d}t\le C|h|\mathcal{T}. L 1 ≤ C ∣ h ∣ ∫ − 2 0 T ( t ; B 8 ) d t ≤ C ∣ h ∣ T .
Moreover, by applying Proposition 5.1 we obtain that
L 2 ≤ C ∣ h ∣ T . \mathcal{L}_2 \le C|h|T. L 2 ≤ C ∣ h ∣ T .
Combining all the estimates from above yields the desired inequality
ess sup Q 1 2 τ h u ≤ C ∣ h ∣ T , (27) \operatorname*{ess\,sup}_{Q_{\frac{1}{2}}}\tau_h u \le C|h|T, \tag*{(27)} Q 2 1 ess sup τ h u ≤ C ∣ h ∣ T , (27)
for any h ∈ R N ∖ { 0 } h \in\mathbb{R}^N \setminus\{0\} h ∈ R N ∖ { 0 } , as long as ∣ h ∣ ≤ 1 4 |h| \le\frac{1}{4} ∣ h ∣ ≤ 4 1 . To complete the proof, let us fix ( x , t ) ∈ Q 1 4 (x,t) \in Q_{\frac{1}{4}} ( x , t ) ∈ Q 4 1 and consider y y y satisfying ∣ x − y ∣ ≤ 1 4 |x-y| \le\frac{1}{4} ∣ x − y ∣ ≤ 4 1 . Apply the above estimate (27) with h = y − x h=y-x h = y − x to deduce
u ( y , t ) − u ( x , t ) ≤ C ∣ y − x ∣ T . u(y,t)-u(x,t) \le C|y-x|T. u ( y , t ) − u ( x , t ) ≤ C ∣ y − x ∣ T .
Switching the roles of x x x and y y y we may deduce that
∣ u ( y , t ) − u ( x , t ) ∣ ≤ C ∣ y − x ∣ T (28) |u(y,t)-u(x,t)| \le C|y-x|T \tag*{(28)} ∣ u ( y , t ) − u ( x , t ) ∣ ≤ C ∣ y − x ∣ T (28)
holds for any ( x , t ) ∈ Q 1 4 (x,t) \in Q_{\frac{1}{4}} ( x , t ) ∈ Q 4 1 and y y y satisfying ∣ x − y ∣ ≤ 1 4 |x-y| \le\frac{1}{4} ∣ x − y ∣ ≤ 4 1 . In other words, we have obtained
∥ D u ∥ L ∞ ( Q 1 4 ) ≤ C [ ∥ u ∥ L ∞ ( Q 8 ) + Tail ( u ; Q 8 ) ] . (29) \|Du\|_{L^\infty(Q_{\frac{1}{4}})} \le C\left[\|u\|_{L^\infty(Q_8)}+\operatorname{Tail}(u;Q_8)\right]. \tag*{(29)} ∥ D u ∥ L ∞ ( Q 4 1 ) ≤ C [ ∥ u ∥ L ∞ ( Q 8 ) + Tail ( u ; Q 8 ) ] . (29)
A proper scaling argument finishes the proof. □
Proof of Lipschitz estimates
First, by Proposition 5.3 for R = 1 R=1 R = 1 , [16 Theorem 1.6] with b = 0 b=0 b = 0 and the fact that u ∈ L l o c 1 ( 0 , T ; L 2 s 1 ( R N ) ) u \in L^1_{\mathrm{loc}}(0,T;L^1_{2s}(\mathbb{R}^N)) u ∈ L loc 1 ( 0 , T ; L 2 s 1 ( R N )) we obtain the estimate
∥ D u ∥ L ∞ ( Q 1 ) ≤ C [ ∥ u ∥ L ∞ ( Q 8 ) + Tail ( u ; Q 8 ) ] ≤ C [ ∫ − 32 2 s 0 ∫ R N u + ( x , t ) 1 + ∣ x ∣ N + 2 s d x d t + Tail ( u ; Q 8 ) ] ≤ C ∫ − 32 2 s 0 ∫ R N ∣ u ( x , t ) ∣ 1 + ∣ x ∣ N + 2 s d x d t . \begin{aligned}
\|Du\|_{L^\infty(Q_1)} \le C\left[\|u\|_{L^\infty(Q_8)}+\operatorname{Tail}(u;Q_8)\right] \\
&\le C\left[\int_{-32^{2s}}^0\int_{\mathbb{R}^N}\frac{u_+(x,t)}{1+|x|^{N+2s}}\,\mathrm{d}x\mathrm{d}t+\operatorname{Tail}(u;Q_8)\right] \\
&\le C\int_{-32^{2s}}^0\int_{\mathbb{R}^N}\frac{|u(x,t)|}{1+|x|^{N+2s}}\,\mathrm{d}x\mathrm{d}t.
\end{aligned} ∥ D u ∥ L ∞ ( Q 1 ) ≤ C [ ∥ u ∥ L ∞ ( Q 8 ) + Tail ( u ; Q 8 ) ] ≤ C [ ∫ − 3 2 2 s 0 ∫ R N 1 + ∣ x ∣ N + 2 s u + ( x , t ) d x d t + Tail ( u ; Q 8 ) ] ≤ C ∫ − 3 2 2 s 0 ∫ R N 1 + ∣ x ∣ N + 2 s ∣ u ( x , t ) ∣ d x d t .
We have proven the Lipschitz estimate in Theorem 1.1 after proper scaling and covering.
Expansion of positivity
Fix h ∈ R N ∖ { 0 } h \in\mathbb{R}^N \setminus\{0\} h ∈ R N ∖ { 0 } and consider
v ≡ v h : = τ h u ∣ h ∣ , (30) v \equiv v_h:=\frac{\tau_hu}{|h|}, \tag*{(30)} v ≡ v h := ∣ h ∣ τ h u , (30)
defined on Ω T \Omega_T Ω T . When no ambiguity arises, we omit h h h from v h v_h v h . Next, we state a reformulation of Lemma 4.2. In fact, dividing both sides of the energy inequality in Lemma 4.2 by ∣ h ∣ 2 |h|^2 ∣ h ∣ 2 , relabeling k / ∣ h ∣ k/|h| k /∣ h ∣ by k k k and taking the supremum over the term involving f f f , defined as in (31) , leads to the following energy inequality.
Lemma 6.1. Let s ∈ ( 0 , 1 ) s \in(0,1) s ∈ ( 0 , 1 ) . If u u u is a local weak solution to (3) in the sense of Definition 2.1, then for any absolutely continuous function k ( ⋅ ) k(\cdot) k ( ⋅ ) in ( 0 , T ) (0,T) ( 0 , T ) , any cylinder Q R , S ( z o ) Q_{R,S}(z_o) Q R , S ( z o ) in Ω T \Omega_T Ω T , any h ∈ R N ∖ { 0 } h \in\mathbb{R}^N \setminus\{0\} h ∈ R N ∖ { 0 } with 0 < ∣ h ∣ ≤ 3 4 R 0<|h|\le\frac{3}{4}R 0 < ∣ h ∣ ≤ 4 3 R and any non-negative, Lipschitz function ζ ( x , t ) \zeta(x,t) ζ ( x , t ) in Q R , S ( z o ) Q_{R,S}(z_o) Q R , S ( z o ) such that spt ζ ( ⋅ , t ) ⊂ B R ( x o ) \operatorname{spt}\zeta(\cdot,t)\subset B_R(x_o) spt ζ ( ⋅ , t ) ⊂ B R ( x o ) for any t ∈ ( t o − S , t o ) t\in(t_o-S,t_o) t ∈ ( t o − S , t o ) , we have
ess sup t ∈ [ t o − S , t o ] ∫ B R × { t } ζ 2 w ~ ± 2 d x + Λ − 1 ∭ Q R , S ∣ ζ w ~ ± ( x , t ) − ζ w ~ ± ( y , t ) ∣ 2 ∣ x − y ∣ N + 2 s d y d x d t \operatorname*{ess\,sup}_{t\in[t_o-S,t_o]}\int_{B_R\times\{t\}}\zeta^2\widetilde w_\pm^2\,\mathrm{d}x
+\Lambda^{-1}\iiint_{Q_{R,S}}\frac{|\zeta\widetilde w_\pm(x,t)-\zeta\widetilde w_\pm(y,t)|^2}{|x-y|^{N+2s}}\,\mathrm{d}y\mathrm{d}x\mathrm{d}t t ∈ [ t o − S , t o ] ess sup ∫ B R × { t } ζ 2 w ± 2 d x + Λ − 1 ∭ Q R , S ∣ x − y ∣ N + 2 s ∣ ζ w ± ( x , t ) − ζ w ± ( y , t ) ∣ 2 d y d x d t + Λ − 1 ∭ Q R , S ζ 2 ( x , t ) w ~ ± ( x , t ) w ~ ∓ ( y , t ) ∣ x − y ∣ N + 2 s d y d x d t ≤ 2 Λ ∭ Q R , S max { w ~ ± ( x , t ) , w ~ ± ( y , t ) } ∣ ζ ( y , t ) − ζ ( x , t ) ∣ 2 ∣ x − y ∣ N + 2 s d y d x d t +\Lambda^{-1}\iiint_{Q_{R,S}}\frac{\zeta^2(x,t)\widetilde w_\pm(x,t)\widetilde w_\mp(y,t)}{|x-y|^{N+2s}}\,\mathrm{d}y\mathrm{d}x\mathrm{d}t
\le2\Lambda\iiint_{Q_{R,S}}\frac{\max\{\widetilde w_\pm(x,t),\widetilde w_\pm(y,t)\}|\zeta(y,t)-\zeta(x,t)|^2}{|x-y|^{N+2s}}\,\mathrm{d}y\mathrm{d}x\mathrm{d}t + Λ − 1 ∭ Q R , S ∣ x − y ∣ N + 2 s ζ 2 ( x , t ) w ± ( x , t ) w ∓ ( y , t ) d y d x d t ≤ 2Λ ∭ Q R , S ∣ x − y ∣ N + 2 s max { w ± ( x , t ) , w ± ( y , t )} ∣ ζ ( y , t ) − ζ ( x , t ) ∣ 2 d y d x d t
+ 4 Λ ∬ Q R , S [ ζ 2 w ~ ± 2 ( x , t ) + ∣ k ( t ) ∣ ζ 2 w ~ ± ( x , t ) ] ∫ R N ∖ B R d y ∣ x − y ∣ N + 2 s d x d t + 4 Λ ∫ t 0 − S t 0 ∥ f ( ⋅ , t ) ∥ L ∞ ( spt ζ ( ⋅ , t ) ) ∫ B R ζ 2 w ~ ± 2 ( x , t ) d x d t = 2 ∬ Q R , S k 2 ζ 2 w ~ ± 2 d x d t + ∬ Q R , S w ~ ± 2 ∣ ∂ t ζ ∣ 2 d x d t + ∫ B R × { t 0 − S } ζ 2 w ~ ± 2 d x , +4\Lambda\iint_{Q_{R,S}}\left[\zeta^2\widetilde{w}_{\pm}^2(x,t)+|k(t)|\zeta^2\widetilde{w}_{\pm}(x,t)\right]\int_{\mathbb{R}^N\setminus B_R}\frac{dy}{|x-y|^{N+2s}}\,dxdt
+4\Lambda\int_{t_0-S}^{t_0}\|f(\cdot,t)\|_{L^\infty(\operatorname{spt}\zeta(\cdot,t))}\int_{B_R}\zeta^2\widetilde{w}_{\pm}^2(x,t)\,dxdt
=2\iint_{Q_{R,S}}k^2\zeta^2\widetilde{w}_{\pm}^2\,dxdt+\iint_{Q_{R,S}}\widetilde{w}_{\pm}^2|\partial_t\zeta|^2\,dxdt
+\int_{B_R\times\{t_0-S\}}\zeta^2\widetilde{w}_{\pm}^2\,dx, + 4Λ ∬ Q R , S [ ζ 2 w ± 2 ( x , t ) + ∣ k ( t ) ∣ ζ 2 w ± ( x , t ) ] ∫ R N ∖ B R ∣ x − y ∣ N + 2 s d y d x d t + 4Λ ∫ t 0 − S t 0 ∥ f ( ⋅ , t ) ∥ L ∞ ( spt ζ ( ⋅ , t )) ∫ B R ζ 2 w ± 2 ( x , t ) d x d t = 2 ∬ Q R , S k 2 ζ 2 w ± 2 d x d t + ∬ Q R , S w ± 2 ∣ ∂ t ζ ∣ 2 d x d t + ∫ B R × { t 0 − S } ζ 2 w ± 2 d x ,
where we omitted x o x_o x o , z o z_o z o and defined w ~ : = v − k \widetilde{w}:=v-k w := v − k as well as
f ( x , t ) : = ∫ R N ∖ B R v ( y , t ) ∣ x − y ∣ N + 2 s d y ∈ L l o c 1 ( 0 , T ; L l o c ∞ ( B R ) ) . (31) f(x,t):=\int_{\mathbb{R}^N\setminus B_R}\frac{v(y,t)}{|x-y|^{N+2s}}\,dy\in L^1_{\mathrm{loc}}(0,T;L^\infty_{\mathrm{loc}}(B_R)). \tag*{(31)} f ( x , t ) := ∫ R N ∖ B R ∣ x − y ∣ N + 2 s v ( y , t ) d y ∈ L loc 1 ( 0 , T ; L loc ∞ ( B R )) . (31)
With the energy estimates at hand, we aim to derive a modulus of continuity of v ≡ v h v\equiv v_h v ≡ v h that is uniform in h h h . The program is a tailored version of [17 , 18 ] . The gist is that proper energy estimates encode all information needed to perform regularity estimates.
Introduce a reference cylinder Q : = B R ( x o ) × ( t 1 , t 2 ] ⊂ Ω T Q:=B_R(x_o)\times(t_1,t_2]\subset\Omega_T Q := B R ( x o ) × ( t 1 , t 2 ] ⊂ Ω T . Assume that μ + \mu^+ μ + and ω \omega ω satisfy
μ + : = ess sup Q v , μ − : = ess inf Q v , ω ≥ μ + − μ − . (32) \mu^+:=\operatorname*{ess\,sup}_{Q}v,\qquad\mu^-:=\operatorname*{ess\,inf}_{Q}v,\qquad\omega\geq\mu^+-\mu^-. \tag*{(32)} μ + := Q ess sup v , μ − := Q ess inf v , ω ≥ μ + − μ − . (32)
Our goal in this section is to use the energy estimate from Lemma 6.1 to prove the following expansion of positivity for v v v .
Proposition 6.2 (Expansion of positivity). Let u u u be a locally bounded, local weak solution to (1.3) in Ω T \Omega_T Ω T and suppose that for some t o ∈ ( t 1 , t 2 ) t_o\in(t_1,t_2) t o ∈ ( t 1 , t 2 ) and some λ , ξ ∈ ( 0 , 1 ) \lambda,\xi\in(0,1) λ , ξ ∈ ( 0 , 1 ) we have
∣ { ± ( μ ± − v ( ⋅ , t o ) ) ≥ ξ ω } ∩ B ϱ ( x o ) ∣ ≥ λ ∣ B ϱ ∣ . |\{\pm(\mu^\pm-v(\cdot,t_o))\geq\xi\omega\}\cap B_\varrho(x_o)|\geq\lambda|B_\varrho|. ∣ { ± ( μ ± − v ( ⋅ , t o )) ≥ ξ ω } ∩ B ϱ ( x o ) ∣ ≥ λ ∣ B ϱ ∣. Then, there exist constants θ , η ∈ ( 0 , 1 ) \theta,\eta\in(0,1) θ , η ∈ ( 0 , 1 ) , γ ~ \widetilde{\gamma} γ depending only on the data and λ \lambda λ , such that either
γ ~ ∫ t 1 t 2 ∥ f ( ⋅ , t ) ∥ L ∞ ( B 8 ϱ ( x o ) ) d t > η ξ ω \widetilde{\gamma}\int_{t_1}^{t_2}\|f(\cdot,t)\|_{L^\infty(B_{8\varrho}(x_o))}\,dt>\eta\xi\omega γ ∫ t 1 t 2 ∥ f ( ⋅ , t ) ∥ L ∞ ( B 8 ϱ ( x o )) d t > η ξ ω or
± ( μ ± − v ) ≥ η ξ ω a.e. in B 2 ϱ ( x o ) × ( t o + 1 2 θ ϱ 2 s , t o + θ ϱ 2 s ] , \pm(\mu^\pm-v)\geq\eta\xi\omega\quad\text{a.e. in }B_{2\varrho}(x_o)\times\left(t_o+\frac{1}{2}\theta\varrho^{2s},t_o+\theta\varrho^{2s}\right], ± ( μ ± − v ) ≥ η ξ ω a.e. in B 2 ϱ ( x o ) × ( t o + 2 1 θ ϱ 2 s , t o + θ ϱ 2 s ] , provided
B 8 ϱ ( x o ) × ( t o , t o + θ ϱ 2 s ] ⊂ Q . B_{8\varrho}(x_o)\times(t_o,t_o+\theta\varrho^{2s}]\subset Q. B 8 ϱ ( x o ) × ( t o , t o + θ ϱ 2 s ] ⊂ Q . Moreover, we have the dependencies θ ≈ λ N + 3 \theta\approx\lambda^{N+3} θ ≈ λ N + 3 and η ≈ λ q \eta\approx\lambda^q η ≈ λ q for some q = q ( N ) > 1 q=q(N)>1 q = q ( N ) > 1 .
Preparatory tools
Three lemmata will be prepared before we prove the expansion of positivity. The first one is a De Giorgi-type lemma. Recall Q Q Q , μ ± \mu^\pm μ ± and ω \omega ω satisfying (6.3). Here and in what follows, we will use the cylinder Q ϱ ( θ ) : = B ϱ ( x o ) × ( t o − θ ϱ 2 s , t o ] Q_\varrho(\theta):=B_\varrho(x_o)\times(t_o-\theta\varrho^{2s},t_o] Q ϱ ( θ ) := B ϱ ( x o ) × ( t o − θ ϱ 2 s , t o ] and omit the vertex ( x o , t o ) (x_o,t_o) ( x o , t o ) .
Lemma 6.3 (De Giorgi-type lemma). Let u u u be a locally bounded, local weak solution to (1.3) in Ω T \Omega_T Ω T . For θ , ξ ∈ ( 0 , 1 ) \theta,\xi\in(0,1) θ , ξ ∈ ( 0 , 1 ) , ϱ ∈ ( 0 , 1 8 R ) \varrho\in(0,\frac{1}{8}R) ϱ ∈ ( 0 , 8 1 R ) and Q ϱ ( θ ) ⊂ Q Q_\varrho(\theta)\subset Q Q ϱ ( θ ) ⊂ Q , there exists a constant γ ~ > 1 \widetilde{\gamma}>1 γ > 1 depending only on the data and ν ∈ ( 0 , 1 ) \nu\in(0,1) ν ∈ ( 0 , 1 ) depending only on the data and θ \theta θ , such that if
∣ { ± ( μ ± − v ) ≤ ξ ω } ∩ Q ϱ ( θ ) ∣ ≤ ν ∣ Q ϱ ( θ ) ∣ , |\{\pm(\mu^\pm-v)\leq\xi\omega\}\cap Q_\varrho(\theta)|\leq\nu|Q_\varrho(\theta)|, ∣ { ± ( μ ± − v ) ≤ ξ ω } ∩ Q ϱ ( θ ) ∣ ≤ ν ∣ Q ϱ ( θ ) ∣ , then either
γ ~ ∫ t 1 t 2 ∥ f ( ⋅ , t ) ∥ L ∞ ( B 2 ϱ ) d t > ξ ω \widetilde{\gamma}\int_{t_1}^{t_2}\|f(\cdot,t)\|_{L^\infty(B_{2\varrho})}\,dt>\xi\omega γ ∫ t 1 t 2 ∥ f ( ⋅ , t ) ∥ L ∞ ( B 2 ϱ ) d t > ξ ω
or
± ( μ ± − v ) ≥ 1 4 ξ ω a.e. in Q 1 2 ϱ ( θ ) . \pm(\mu^{\pm}-v) \ge\frac{1}{4}\xi\omega\quad\text{a.e. in } Q_{\frac{1}{2}\varrho}(\theta). ± ( μ ± − v ) ≥ 4 1 ξ ω a.e. in Q 2 1 ϱ ( θ ) .
Moreover, we have the dependence ν ≈ θ q \nu\approx\theta^{q} ν ≈ θ q for some q = q ( N ) > 1 q=q(N)>1 q = q ( N ) > 1 .
Proof. We prove the statement for w − w_{-} w − with μ − = 0 \mu^{-}=0 μ − = 0 . Upon a translation, we may assume z o = ( 0 , 0 ) z_{o}=(0,0) z o = ( 0 , 0 ) and still use Q : = B R × ( t 1 , t 2 ] Q:=B_{R}\times(t_{1},t_{2}] Q := B R × ( t 1 , t 2 ] to denote the shifted reference cylinder. Next, we introduce
{ k n = ( 1 2 + 1 2 n + 1 ) ξ ω , ϱ n = ( 1 2 + 1 2 n + 1 ) ϱ , ϱ ~ n = 1 2 ( ϱ n + ϱ n + 1 ) , ϱ ^ n = 3 4 ϱ n + 1 4 ϱ n + 1 , ϱ ‾ n = 1 4 ϱ n + 3 4 ϱ n + 1 , B n = B ϱ n , B ~ n = B ϱ ~ n , B ^ n = B ϱ ^ n , B ‾ n = B ϱ ‾ n , Q n = B n × ( − θ ϱ n 2 s , 0 ) , Q ~ n = B ~ n × ( − θ ϱ ~ n 2 s , 0 ) , Q ^ n = B ^ n × ( − θ ϱ ^ n 2 s , 0 ) , Q ‾ n = B ‾ n × ( − θ ϱ ‾ n 2 s , 0 ) . \left\{
\begin{aligned}
k_{n}&=\left(\frac{1}{2}+\frac{1}{2^{n+1}}\right)\xi\omega,\qquad
\varrho_{n}=\left(\frac{1}{2}+\frac{1}{2^{n+1}}\right)\varrho,\\
\widetilde{\varrho}_{n}&=\frac{1}{2}(\varrho_{n}+\varrho_{n+1}),\qquad
\widehat{\varrho}_{n}=\frac{3}{4}\varrho_{n}+\frac{1}{4}\varrho_{n+1},\qquad
\overline{\varrho}_{n}=\frac{1}{4}\varrho_{n}+\frac{3}{4}\varrho_{n+1},\\
B_{n}&=B_{\varrho_{n}},\qquad
\widetilde{B}_{n}=B_{\widetilde{\varrho}_{n}},\qquad
\widehat{B}_{n}=B_{\widehat{\varrho}_{n}},\qquad
\overline{B}_{n}=B_{\overline{\varrho}_{n}},\\
Q_{n}&=B_{n}\times(-\theta\varrho_{n}^{2s},0),\qquad
\widetilde{Q}_{n}=\widetilde{B}_{n}\times(-\theta\widetilde{\varrho}_{n}^{2s},0),\\
\widehat{Q}_{n}&=\widehat{B}_{n}\times(-\theta\widehat{\varrho}_{n}^{2s},0),\qquad
\overline{Q}_{n}=\overline{B}_{n}\times(-\theta\overline{\varrho}_{n}^{2s},0).
\end{aligned}
\right. ⎩ ⎨ ⎧ k n ϱ n B n Q n Q n = ( 2 1 + 2 n + 1 1 ) ξ ω , ϱ n = ( 2 1 + 2 n + 1 1 ) ϱ , = 2 1 ( ϱ n + ϱ n + 1 ) , ϱ n = 4 3 ϱ n + 4 1 ϱ n + 1 , ϱ n = 4 1 ϱ n + 4 3 ϱ n + 1 , = B ϱ n , B n = B ϱ n , B n = B ϱ n , B n = B ϱ n , = B n × ( − θ ϱ n 2 s , 0 ) , Q n = B n × ( − θ ϱ n 2 s , 0 ) , = B n × ( − θ ϱ n 2 s , 0 ) , Q n = B n × ( − θ ϱ n 2 s , 0 ) .
Note that Q n + 1 ⊂ Q ‾ n ⊂ Q ^ n ⊂ Q ~ n ⊂ Q n Q_{n+1}\subset\overline{Q}_{n}\subset\widehat{Q}_{n}\subset\widetilde{Q}_{n}\subset Q_{n} Q n + 1 ⊂ Q n ⊂ Q n ⊂ Q n ⊂ Q n . Let ζ \zeta ζ in Q n Q_{n} Q n be a cutoff function, vanishing outside Q ^ n \widehat{Q}_{n} Q n and equal to 1 in Q ~ n \widetilde{Q}_{n} Q n such that
∣ D ζ ∣ ≤ 2 n + 4 ϱ and ∣ ∂ t ζ ∣ ≤ 2 2 s ( n + 4 ) θ ϱ 2 s . |D\zeta|\le\frac{2^{n+4}}{\varrho}
\qquad\text{and}\qquad
|\partial_{t}\zeta|\le\frac{2^{2s(n+4)}}{\theta\varrho^{2s}}. ∣ D ζ ∣ ≤ ϱ 2 n + 4 and ∣ ∂ t ζ ∣ ≤ θ ϱ 2 s 2 2 s ( n + 4 ) .
Applying the energy estimate from Lemma 6.1 in Q n Q_{n} Q n with ζ \zeta ζ and
{ w ~ ( x , t ) : = ( v ( x , t ) − k n ( t ) ) , k n ( t ) : = k n − ℓ ( t ) , ℓ ( t ) : = γ ~ ∫ t 1 t ∥ f ( ⋅ , τ ) ∥ L ∞ ( B 2 ϱ ) d τ , (33) \left\{
\begin{aligned}
\widetilde{w}(x,t)&:=(v(x,t)-k_{n}(t)),\\
k_{n}(t)&:=k_{n}-\ell(t),\qquad
\ell(t):=\widetilde{\gamma}\int_{t_{1}}^{t}\|f(\cdot,\tau)\|_{L^{\infty}(B_{2\varrho})}\,d\tau,
\end{aligned}
\right. \tag*{(33)} ⎩ ⎨ ⎧ w ( x , t ) k n ( t ) := ( v ( x , t ) − k n ( t )) , := k n − ℓ ( t ) , ℓ ( t ) := γ ∫ t 1 t ∥ f ( ⋅ , τ ) ∥ L ∞ ( B 2 ϱ ) d τ , (33)
for γ ~ > 0 \widetilde{\gamma}>0 γ > 0 to be specified later, gives us
ess sup t ∈ [ − θ ϱ ~ n 2 s , 0 ] ∫ B ~ n × { t } ζ 2 w ~ − 2 d x + Λ − 1 ∭ Q ~ n ∣ ζ w ~ − ( x , t ) − ζ w ~ − ( y , t ) ∣ 2 ∣ x − y ∣ N + 2 s d y d x d t ≤ 2 Λ ∭ Q n max { w ~ − 2 ( x , t ) , w ~ − 2 ( y , t ) } ∣ ζ ( y , t ) − ζ ( x , t ) ∣ 2 ∣ x − y ∣ N + 2 s d y d x d t + 4 Λ ∭ Q n [ ζ 2 w ~ − 2 ( x , t ) + ∣ k n + 1 ( t ) ∣ ζ 2 w ~ − ( x , t ) ] ∫ R N ∖ B n d y ∣ x − y ∣ N + 2 s d x d t + 4 Λ ∫ − θ ϱ n 2 s 0 ∥ f ( ⋅ , t ) ∥ L ∞ ( B ^ n ) ∫ B n ζ 2 w ~ − ( x , t ) d x d t − 2 ∬ Q n ℓ 2 ζ 2 w ~ − d x d t + ∬ Q n w ~ − 2 ∣ ∂ t ζ 2 ∣ d x d t = : I + I I + I I I − I V + V . \begin{aligned}
&\mathop{\operatorname{ess\,sup}}_{t\in[-\theta\widetilde{\varrho}_{n}^{2s},0]}
\int_{\widetilde{B}_{n}\times\{t\}}\zeta^{2}\widetilde{w}_{-}^{2}\,dx
+\Lambda^{-1}\iiint_{\widetilde{Q}_{n}}
\frac{|\zeta\widetilde{w}_{-}(x,t)-\zeta\widetilde{w}_{-}(y,t)|^{2}}{|x-y|^{N+2s}}\,dydxdt\\
&\le2\Lambda\iiint_{Q_{n}}
\frac{\max\{\widetilde{w}_{-}^{2}(x,t),\widetilde{w}_{-}^{2}(y,t)\}|\zeta(y,t)-\zeta(x,t)|^{2}}{|x-y|^{N+2s}}\,dydxdt\\
&\quad+4\Lambda\iiint_{Q_{n}}
\left[\zeta^{2}\widetilde{w}_{-}^{2}(x,t)+|k_{n+1}(t)|\zeta^{2}\widetilde{w}_{-}(x,t)\right]
\int_{\mathbb{R}^{N}\setminus B_{n}}\frac{dy}{|x-y|^{N+2s}}\,dxdt\\
&\quad+4\Lambda\int_{-\theta\varrho_{n}^{2s}}^{0}\|f(\cdot,t)\|_{L^{\infty}(\widehat{B}_{n})}
\int_{B_{n}}\zeta^{2}\widetilde{w}_{-}(x,t)\,dxdt\\
&\quad-2\iint_{Q_{n}}\ell^{2}\zeta^{2}\widetilde{w}_{-}\,dxdt
+\iint_{Q_{n}}\widetilde{w}_{-}^{2}|\partial_{t}\zeta^{2}|\,dxdt\\
&=:I+II+III-IV+V.
\end{aligned} ess sup t ∈ [ − θ ϱ n 2 s , 0 ] ∫ B n × { t } ζ 2 w − 2 d x + Λ − 1 ∭ Q n ∣ x − y ∣ N + 2 s ∣ ζ w − ( x , t ) − ζ w − ( y , t ) ∣ 2 d y d x d t ≤ 2Λ ∭ Q n ∣ x − y ∣ N + 2 s max { w − 2 ( x , t ) , w − 2 ( y , t )} ∣ ζ ( y , t ) − ζ ( x , t ) ∣ 2 d y d x d t + 4Λ ∭ Q n [ ζ 2 w − 2 ( x , t ) + ∣ k n + 1 ( t ) ∣ ζ 2 w − ( x , t ) ] ∫ R N ∖ B n ∣ x − y ∣ N + 2 s d y d x d t + 4Λ ∫ − θ ϱ n 2 s 0 ∥ f ( ⋅ , t ) ∥ L ∞ ( B n ) ∫ B n ζ 2 w − ( x , t ) d x d t − 2 ∬ Q n ℓ 2 ζ 2 w − d x d t + ∬ Q n w − 2 ∣ ∂ t ζ 2 ∣ d x d t =: I + I I + I I I − I V + V .
Note that the last term from Lemma 6.1 vanishes due to the support of ζ \zeta ζ . Estimating the terms I and V is standard, that is,
I ≤ C 2 2 n ( ξ ω ) 2 ϱ 2 s ∣ A n ∣ and V ≤ 2 2 s n θ ϱ 2 s ( ξ ω ) 2 ∣ A n ∣ , I\le C2^{2n}\frac{(\xi\omega)^{2}}{\varrho^{2s}}|A_{n}|
\qquad\text{and}\qquad
V\le\frac{2^{2sn}}{\theta\varrho^{2s}}(\xi\omega)^{2}|A_{n}|, I ≤ C 2 2 n ϱ 2 s ( ξ ω ) 2 ∣ A n ∣ and V ≤ θ ϱ 2 s 2 2 s n ( ξ ω ) 2 ∣ A n ∣ ,
where
A n : = { v ( x , t ) ≤ k n − ℓ ( t ) } ∩ Q n . A_{n}:=\{v(x,t)\le k_{n}-\ell(t)\}\cap Q_{n}. A n := { v ( x , t ) ≤ k n − ℓ ( t )} ∩ Q n .
Next, we treat II as follows. First, observe that if ∣ y ∣ ≥ ϱ n |y|\ge\varrho_{n} ∣ y ∣ ≥ ϱ n and ∣ x ∣ ≤ ϱ ^ n |x|\le\widehat{\varrho}_{n} ∣ x ∣ ≤ ϱ n , then
∣ y − x ∣ ∣ y ∣ ≥ 1 − ϱ ^ n ϱ n = 1 − 1 4 ( ϱ n − ϱ n + 1 ϱ n ) ≥ 1 2 n + 4 . \frac{|y-x|}{|y|}
\ge1-\frac{\widehat{\varrho}_{n}}{\varrho_{n}}
=1-\frac{1}{4}\left(\frac{\varrho_{n}-\varrho_{n+1}}{\varrho_{n}}\right)
\ge\frac{1}{2^{n+4}}. ∣ y ∣ ∣ y − x ∣ ≥ 1 − ϱ n ϱ n = 1 − 4 1 ( ϱ n ϱ n − ϱ n + 1 ) ≥ 2 n + 4 1 .
Therefore, we obtain assuming ℓ ( t 2 ) ≤ ξ ω \ell(t_2) \le\xi\omega ℓ ( t 2 ) ≤ ξ ω that
I I ≤ C 2 ( N + 2 s ) n ϱ 2 s ∬ Q n [ ζ 2 w ~ − 2 ( x , t ) + ∣ k n + 1 ( t ) ∣ ζ 2 w ~ − ( x , t ) ] d x d t ≤ C 2 ( N + 2 s ) n ϱ 2 s ( ξ ω ) 2 ∣ A n ∣ . \begin{aligned}
\mathrm{II} \le\frac{C^2(N+2s)n}{\varrho^{2s}} \iint_{Q_n} \left[\zeta^2\widetilde{w}_-^2(x,t)+|k_{n+1}(t)|\zeta^2\widetilde{w}_-(x,t)\right]\,dx\,dt \\
&\le\frac{C^2(N+2s)n}{\varrho^{2s}}(\xi\omega)^2|A_n|.
\end{aligned} II ≤ ϱ 2 s C 2 ( N + 2 s ) n ∬ Q n [ ζ 2 w − 2 ( x , t ) + ∣ k n + 1 ( t ) ∣ ζ 2 w − ( x , t ) ] d x d t ≤ ϱ 2 s C 2 ( N + 2 s ) n ( ξ ω ) 2 ∣ A n ∣.
since ∣ k n + 1 ( t ) ∣ ≤ ∣ k n + 1 ∣ + ∣ ℓ ( t ) ∣ ≤ 2 ξ ω |k_{n+1}(t)| \le|k_{n+1}|+|\ell(t)| \le2\xi\omega ∣ k n + 1 ( t ) ∣ ≤ ∣ k n + 1 ∣ + ∣ ℓ ( t ) ∣ ≤ 2 ξ ω . For the third term, we have
I I I ≤ 4 Λ ∫ − θ ϱ n 2 s 0 ∥ f ( ⋅ , t ) ∥ L ∞ ( B 2 ϱ ) ∫ B n ζ 2 w ~ − ( x , t ) d x d t \mathrm{III} \le4\Lambda\int_{-\theta\varrho_n^{2s}}^0 \|f(\cdot,t)\|_{L^\infty(B_{2\varrho})}\int_{B_n}\zeta^2\widetilde{w}_-(x,t)\,dx\,dt III ≤ 4Λ ∫ − θ ϱ n 2 s 0 ∥ f ( ⋅ , t ) ∥ L ∞ ( B 2 ϱ ) ∫ B n ζ 2 w − ( x , t ) d x d t
and then choose γ ~ = 2 Λ \widetilde{\gamma}=2\Lambda γ = 2Λ in the definition of ℓ ( t ) \ell(t) ℓ ( t ) , cf. (33) , such that
ℓ ′ ( t ) = 2 Λ ∥ f ( ⋅ , t ) ∥ L ∞ ( B 2 ϱ ) , \ell'(t)=2\Lambda\|f(\cdot,t)\|_{L^\infty(B_{2\varrho})}, ℓ ′ ( t ) = 2Λ∥ f ( ⋅ , t ) ∥ L ∞ ( B 2 ϱ ) ,
and hence I I I − I V ≤ 0 \mathrm{III}-\mathrm{IV}\le0 III − IV ≤ 0 . To summarize, we have deduced that
ess sup t ∈ [ − θ ϱ n 2 s , 0 ] ∫ B ~ n × { t } ζ 2 w ~ − 2 d x + Λ − 1 ∭ Q n ∣ ζ w ~ − ( x , t ) − ζ w ~ − ( y , t ) ∣ 2 ∣ x − y ∣ N + 2 s d y d x d t ≤ C 2 ( N + 2 s ) n θ ϱ 2 s ( ξ ω ) 2 ∣ A n ∣ . \begin{aligned}
\operatorname*{ess\,sup}_{t\in[-\theta\varrho_n^{2s},0]}\int_{\widetilde{B}_n\times\{t\}}\zeta^2\widetilde{w}_-^2\,dx+\Lambda^{-1}\iiint_{Q_n}\frac{|\zeta\widetilde{w}_-(x,t)-\zeta\widetilde{w}_-(y,t)|^2}{|x-y|^{N+2s}}\,dy\,dx\,dt \\
&\le\frac{C^2(N+2s)n}{\theta\varrho^{2s}}(\xi\omega)^2|A_n|.
\end{aligned} t ∈ [ − θ ϱ n 2 s , 0 ] ess sup ∫ B n × { t } ζ 2 w − 2 d x + Λ − 1 ∭ Q n ∣ x − y ∣ N + 2 s ∣ ζ w − ( x , t ) − ζ w − ( y , t ) ∣ 2 d y d x d t ≤ θ ϱ 2 s C 2 ( N + 2 s ) n ( ξ ω ) 2 ∣ A n ∣.
Proceeding as in (23) –(24) we obtain by defining Y n : = ∣ A n ∣ ∣ Q n ∣ Y_n:=\frac{|A_n|}{|Q_n|} Y n := ∣ Q n ∣ ∣ A n ∣ that
Y n + 1 ≤ C θ − N N + 2 s b 1 + 2 s N + 2 s Y n 1 + 2 s N + 2 s . Y_{n+1}\le C\theta^{-\frac{N}{N+2s}}b^{\frac{1+2s}{N+2s}}Y_n^{1+\frac{2s}{N+2s}}. Y n + 1 ≤ C θ − N + 2 s N b N + 2 s 1 + 2 s Y n 1 + N + 2 s 2 s .
Hence, by the fast geometric convergence, cf. [8 Chapter I, Lemma 4.1] , there exists a ν \nu ν depending only on the data and θ \theta θ , such that Y n → 0 Y_n\to0 Y n → 0 , if we require Y 0 ≤ ν Y_0\le\nu Y 0 ≤ ν , i.e.
∣ { v ( x , t ) + ℓ ( t ) ≤ ξ ω } ∩ Q ϱ ( θ ) ∣ ≤ ν ∣ Q ϱ ( θ ) ∣ . |\{v(x,t)+\ell(t)\le\xi\omega\}\cap Q_\varrho(\theta)|\le\nu|Q_\varrho(\theta)|. ∣ { v ( x , t ) + ℓ ( t ) ≤ ξ ω } ∩ Q ϱ ( θ ) ∣ ≤ ν ∣ Q ϱ ( θ ) ∣.
Then, Y n → 0 Y_n\to0 Y n → 0 implies that
v ( x , t ) + ℓ ( t ) ≥ 1 2 ξ ω a.e. in Q 1 2 ϱ ( θ ) . v(x,t)+\ell(t)\ge\frac{1}{2}\xi\omega\quad\text{a.e. in }Q_{\frac{1}{2}\varrho}(\theta). v ( x , t ) + ℓ ( t ) ≥ 2 1 ξ ω a.e. in Q 2 1 ϱ ( θ ) .
This implies that, if ℓ ( t 2 ) ≤ 1 4 ξ ω \ell(t_2)\le\frac{1}{4}\xi\omega ℓ ( t 2 ) ≤ 4 1 ξ ω , by adjusting γ ~ \widetilde{\gamma} γ appropriately, and
∣ { v ( x , t ) < ξ ω } ∩ Q ϱ ( θ ) ∣ ≤ ν ∣ Q ϱ ( θ ) ∣ , |\{v(x,t)<\xi\omega\}\cap Q_\varrho(\theta)|\le\nu|Q_\varrho(\theta)|, ∣ { v ( x , t ) < ξ ω } ∩ Q ϱ ( θ ) ∣ ≤ ν ∣ Q ϱ ( θ ) ∣ ,
then
v ≥ 1 4 ξ ω a.e. in Q 1 2 ϱ ( θ ) . v\ge\frac{1}{4}\xi\omega\quad\text{a.e. in }Q_{\frac{1}{2}\varrho}(\theta). v ≥ 4 1 ξ ω a.e. in Q 2 1 ϱ ( θ ) .
We conclude the proof by redefining 4 γ ~ 4\widetilde{\gamma} 4 γ to γ ~ \widetilde{\gamma} γ . □
Next, we formulate a lemma, which propagates measure theoretical information forward in time.
Lemma 6.4. Let u u u be a locally bounded, local weak solution to (3) in Ω T \Omega_T Ω T . Introduce the parameters ξ , λ ∈ ( 0 , 1 ) \xi,\lambda\in(0,1) ξ , λ ∈ ( 0 , 1 ) . There exist θ , ε ∈ ( 0 , 1 ) \theta,\varepsilon\in(0,1) θ , ε ∈ ( 0 , 1 ) depending only on the data and λ \lambda λ , such that if for some t o ∈ ( t 1 , t 2 ) t_o\in(t_1,t_2) t o ∈ ( t 1 , t 2 ) we have
∣ { ± ( μ ± − v ( ⋅ , t o ) ) ≥ ξ ω } ∩ B ϱ ( x o ) ∣ ≥ λ ∣ B ϱ ∣ , |\{\pm(\mu^\pm-v(\cdot,t_o))\ge\xi\omega\}\cap B_\varrho(x_o)|\ge\lambda|B_\varrho|, ∣ { ± ( μ ± − v ( ⋅ , t o )) ≥ ξ ω } ∩ B ϱ ( x o ) ∣ ≥ λ ∣ B ϱ ∣ , then either
1 θ ∫ t 1 t 2 ∥ f ( ⋅ , t ) ∥ L ∞ ( B 2 ϱ ) d t > ξ ω \frac{1}{\theta}\int_{t_1}^{t_2}\|f(\cdot,t)\|_{L^\infty(B_{2\varrho})}\,dt>\xi\omega θ 1 ∫ t 1 t 2 ∥ f ( ⋅ , t ) ∥ L ∞ ( B 2 ϱ ) d t > ξ ω or
∣ { ± ( μ ± − v ( ⋅ , t ) ) ≥ ε ξ ω } ∩ B ϱ ( x o ) ∣ ≥ 1 2 λ ∣ B ϱ ∣ for all t ∈ ( t o , t o + θ ϱ 2 s ] , |\{\pm(\mu^\pm-v(\cdot,t))\ge\varepsilon\xi\omega\}\cap B_\varrho(x_o)|\ge\frac{1}{2}\lambda|B_\varrho|\quad\text{for all }t\in(t_o,t_o+\theta\varrho^{2s}], ∣ { ± ( μ ± − v ( ⋅ , t )) ≥ ε ξ ω } ∩ B ϱ ( x o ) ∣ ≥ 2 1 λ ∣ B ϱ ∣ for all t ∈ ( t o , t o + θ ϱ 2 s ] , provided B ϱ ( x o ) × ( t o , t o + θ ϱ 2 s ] ⊂ Q B_\varrho(x_o)\times(t_o,t_o+\theta\varrho^{2s}]\subset Q B ϱ ( x o ) × ( t o , t o + θ ϱ 2 s ] ⊂ Q . Moreover, we have ε ≈ λ \varepsilon\approx\lambda ε ≈ λ and θ ≈ λ N + 3 \theta\approx\lambda^{N+3} θ ≈ λ N + 3 .
Proof. We will assume that μ − = 0 \mu^{-}=0 μ − = 0 , take ( x o , t o ) = ( 0 , 0 ) (x_o,t_o)=(0,0) ( x o , t o ) = ( 0 , 0 ) , and still denote the reference cylinder B R × ( t 1 , t 2 ] B_R\times(t_1,t_2] B R × ( t 1 , t 2 ] . Suppose that
1 θ ∫ t 1 t 2 ∥ f ( ⋅ , t ) ∥ L ∞ ( B 2 ϱ ) d t ≤ ξ ω . (34) \frac{1}{\theta}\int_{t_1}^{t_2}\|f(\cdot,t)\|_{L^\infty(B_{2\varrho})}\,dt\leq\xi\omega. \tag*{(34)} θ 1 ∫ t 1 t 2 ∥ f ( ⋅ , t ) ∥ L ∞ ( B 2 ϱ ) d t ≤ ξ ω . (34)
Let w ~ − ( x , t ) : = ( v ( x , t ) − ξ ω ) − \widetilde w_-(x,t):=(v(x,t)-\xi\omega)_- w − ( x , t ) := ( v ( x , t ) − ξ ω ) − , consider α ∈ ( 0 , 1 ) \alpha\in(0,1) α ∈ ( 0 , 1 ) , which will be determined later, and choose a cutoff function ζ ( x , t ) ≡ ζ ( x ) \zeta(x,t)\equiv\zeta(x) ζ ( x , t ) ≡ ζ ( x ) , such that ζ = 1 \zeta=1 ζ = 1 on B ( 1 − α ) ϱ B_{(1-\alpha)\varrho} B ( 1 − α ) ϱ and ζ = 0 \zeta=0 ζ = 0 outside B ( 1 − α 2 ) ϱ B_{(1-\frac{\alpha}{2})\varrho} B ( 1 − 2 α ) ϱ satisfying ∣ D ζ ∣ ≤ 4 α ϱ |D\zeta|\leq\frac{4}{\alpha\varrho} ∣ D ζ ∣ ≤ α ϱ 4 . Hence, the energy inequality from Lemma 6.1 in B ϱ × ( 0 , θ ϱ 2 s ] B_\varrho\times(0,\theta\varrho^{2s}] B ϱ × ( 0 , θ ϱ 2 s ] with w ~ − \widetilde w_- w − , ζ \zeta ζ and k = ξ ω k=\xi\omega k = ξ ω gives for all t ∈ ( 0 , θ ϱ 2 s ) t\in(0,\theta\varrho^{2s}) t ∈ ( 0 , θ ϱ 2 s ) that
∫ B ( 1 − α ) ϱ × { t } w ~ − 2 d x ≤ 2 Λ ∫ 0 θ ϱ 2 s ∫ B ϱ ∫ B ϱ max { w ~ − 2 ( x , t ) , w ~ − 2 ( y , t ) } ∣ ζ ( y , t ) − ζ ( x , t ) ∣ 2 ∣ x − y ∣ N + 2 s d y d x d t + 4 Λ ∫ 0 θ ϱ 2 s ∫ B ϱ [ ζ 2 w ~ − 2 ( x , t ) + ξ ω ζ 2 w ~ − ( x , t ) ] ∫ R N ∖ B ϱ d y ∣ x − y ∣ N + 2 s d x d t + 4 Λ ∫ 0 θ ϱ 2 s ∥ f ( ⋅ , t ) ∥ L ∞ ( spt ζ ) ∫ B ϱ ζ 2 w ~ − ( x , t ) d x d t + ∫ B ϱ × { 0 } w ~ − 2 d x = : I + I I + I I I + I V . \begin{aligned}
\int_{B_{(1-\alpha)\varrho}\times\{t\}}\widetilde w_-^2\,dx
&\leq2\Lambda\int_0^{\theta\varrho^{2s}}\int_{B_\varrho}\int_{B_\varrho}
\max\{\widetilde w_-^2(x,t),\widetilde w_-^2(y,t)\}
\frac{|\zeta(y,t)-\zeta(x,t)|^2}{|x-y|^{N+2s}}\,dydxdt\\
&\quad+4\Lambda\int_0^{\theta\varrho^{2s}}\int_{B_\varrho}
\left[\zeta^2\widetilde w_-^2(x,t)+\xi\omega\zeta^2\widetilde w_-(x,t)\right]
\int_{\mathbb R^N\setminus B_\varrho}\frac{dy}{|x-y|^{N+2s}}\,dxdt\\
&\quad+4\Lambda\int_0^{\theta\varrho^{2s}}\|f(\cdot,t)\|_{L^\infty(\operatorname{spt}\zeta)}
\int_{B_\varrho}\zeta^2\widetilde w_-(x,t)\,dxdt
+\int_{B_\varrho\times\{0\}}\widetilde w_-^2\,dx\\
&=: I+\mathrm{II}+\mathrm{III}+\mathrm{IV}.
\end{aligned} ∫ B ( 1 − α ) ϱ × { t } w − 2 d x ≤ 2Λ ∫ 0 θ ϱ 2 s ∫ B ϱ ∫ B ϱ max { w − 2 ( x , t ) , w − 2 ( y , t )} ∣ x − y ∣ N + 2 s ∣ ζ ( y , t ) − ζ ( x , t ) ∣ 2 d y d x d t + 4Λ ∫ 0 θ ϱ 2 s ∫ B ϱ [ ζ 2 w − 2 ( x , t ) + ξ ω ζ 2 w − ( x , t ) ] ∫ R N ∖ B ϱ ∣ x − y ∣ N + 2 s d y d x d t + 4Λ ∫ 0 θ ϱ 2 s ∥ f ( ⋅ , t ) ∥ L ∞ ( spt ζ ) ∫ B ϱ ζ 2 w − ( x , t ) d x d t + ∫ B ϱ × { 0 } w − 2 d x =: I + II + III + IV .
In the above application of Lemma 6.1 we omitted the second and third terms on the left-hand side, whereas the last two terms on the right-hand side vanished because k k k and ζ \zeta ζ are independent of time. We estimate the term I \mathrm{I} I by
I ≤ 2 Λ θ ϱ 2 s ( ξ ω ) 2 ∫ B ϱ ∫ B ϱ ∣ ζ ( x ) − ζ ( y ) ∣ 2 ∣ x − y ∣ N + 2 s d y d x ≤ 32 Λ θ ϱ 2 s ( ξ ω ) 2 ( α ϱ ) 2 ∫ B ϱ ∫ B ϱ 1 ∣ x − y ∣ N + 2 ( s − 1 ) d y d x ≤ C θ ( ξ ω ) 2 α 2 ∣ B ϱ ∣ . \begin{aligned}
\mathrm{I}
&\leq2\Lambda\theta\varrho^{2s}(\xi\omega)^2
\int_{B_\varrho}\int_{B_\varrho}
\frac{|\zeta(x)-\zeta(y)|^2}{|x-y|^{N+2s}}\,dydx\\
&\leq32\Lambda\theta\varrho^{2s}\frac{(\xi\omega)^2}{(\alpha\varrho)^2}
\int_{B_\varrho}\int_{B_\varrho}
\frac{1}{|x-y|^{N+2(s-1)}}\,dydx\\
&\leq C\frac{\theta(\xi\omega)^2}{\alpha^2}|B_\varrho|.
\end{aligned} I ≤ 2Λ θ ϱ 2 s ( ξ ω ) 2 ∫ B ϱ ∫ B ϱ ∣ x − y ∣ N + 2 s ∣ ζ ( x ) − ζ ( y ) ∣ 2 d y d x ≤ 32Λ θ ϱ 2 s ( α ϱ ) 2 ( ξ ω ) 2 ∫ B ϱ ∫ B ϱ ∣ x − y ∣ N + 2 ( s − 1 ) 1 d y d x ≤ C α 2 θ ( ξ ω ) 2 ∣ B ϱ ∣.
For the second term, note that since x ∈ spt ζ ⊂ B ( 1 − α 2 ) ϱ x\in\operatorname{spt}\zeta\subset B_{(1-\frac{\alpha}{2})\varrho} x ∈ spt ζ ⊂ B ( 1 − 2 α ) ϱ and y ∈ R N ∖ B ϱ y\in\mathbb R^N\setminus B_\varrho y ∈ R N ∖ B ϱ , we have
∣ x − y ∣ ∣ y ∣ ≥ 1 − ∣ x ∣ ∣ y ∣ ≥ α 2 . \frac{|x-y|}{|y|}\geq1-\frac{|x|}{|y|}\geq\frac{\alpha}{2}. ∣ y ∣ ∣ x − y ∣ ≥ 1 − ∣ y ∣ ∣ x ∣ ≥ 2 α .
Thus, it follows that
I I ≤ C α N + 2 s ϱ 2 s ∫ 0 θ ϱ 2 s ∫ B ϱ [ ζ 2 w ~ − 2 ( x , t ) + ξ ω ζ 2 w ~ + ( x , t ) ] d x d t ≤ C α N + 2 s ϱ 2 s ∣ Q ϱ ( θ ) ∣ ( ξ ω ) 2 = C θ ( ξ ω ) 2 α N + 2 s ∣ B ϱ ∣ . \begin{aligned}
\mathrm{II}
&\leq\frac{C}{\alpha^{N+2s}\varrho^{2s}}
\int_0^{\theta\varrho^{2s}}\int_{B_\varrho}
\left[\zeta^2\widetilde w_-^2(x,t)+\xi\omega\zeta^2\widetilde w_+(x,t)\right]\,dxdt\\
&\leq\frac{C}{\alpha^{N+2s}\varrho^{2s}}|Q_\varrho(\theta)|(\xi\omega)^2
= C\frac{\theta(\xi\omega)^2}{\alpha^{N+2s}}|B_\varrho|.
\end{aligned} II ≤ α N + 2 s ϱ 2 s C ∫ 0 θ ϱ 2 s ∫ B ϱ [ ζ 2 w − 2 ( x , t ) + ξ ω ζ 2 w + ( x , t ) ] d x d t ≤ α N + 2 s ϱ 2 s C ∣ Q ϱ ( θ ) ∣ ( ξ ω ) 2 = C α N + 2 s θ ( ξ ω ) 2 ∣ B ϱ ∣.
Next, due to (34) we obtain for I I I \mathrm{III} III the estimate
I I I ≤ C ξ ω ∣ B ϱ ∣ ∫ 0 θ ϱ 2 s ∥ f ( ⋅ , t ) ∥ L ∞ ( B 2 ϱ ) d t ≤ C θ ( ξ ω ) 2 ∣ B ϱ ∣ . \begin{aligned}
\mathrm{III}
&\leq C\xi\omega|B_\varrho|\int_0^{\theta\varrho^{2s}}
\|f(\cdot,t)\|_{L^\infty(B_{2\varrho})}\,dt\\
&\leq C\theta(\xi\omega)^2|B_\varrho|.
\end{aligned} III ≤ C ξ ω ∣ B ϱ ∣ ∫ 0 θ ϱ 2 s ∥ f ( ⋅ , t ) ∥ L ∞ ( B 2 ϱ ) d t ≤ C θ ( ξ ω ) 2 ∣ B ϱ ∣.
Furthermore, we obtain for the term I V \mathrm{IV} IV that
I V ≤ ( 1 − λ ) ( ξ ω ) 2 ∣ B ϱ ∣ . \mathrm{IV}\leq(1-\lambda)(\xi\omega)^2|B_\varrho|. IV ≤ ( 1 − λ ) ( ξ ω ) 2 ∣ B ϱ ∣.
By combining all the estimates from above we obtain for all t ∈ ( 0 , θ ϱ 2 s ) t \in(0,\theta\varrho^{2s}) t ∈ ( 0 , θ ϱ 2 s ) that
∫ B ( 1 − α ) ϱ × { t } w ~ 2 d x ≤ ( ξ ω ) 2 [ ( 1 − λ ) + C θ α N + 2 s ] ∣ B ϱ ∣ . (35) \int_{B_{(1-\alpha)\varrho\times\{t\}}} \widetilde{w}^{2}\,dx \leq(\xi\omega)^{2}\left[(1-\lambda)+C\frac{\theta}{\alpha^{N+2s}}\right]|B_{\varrho}|. \tag*{(35)} ∫ B ( 1 − α ) ϱ × { t } w 2 d x ≤ ( ξ ω ) 2 [ ( 1 − λ ) + C α N + 2 s θ ] ∣ B ϱ ∣. (35)
Let us estimate the left-hand side of the previous display from below by
∫ B ( 1 − α ) ϱ × { t } w ~ 2 d x ≥ ∫ B ( 1 − α ) ϱ × { t } w ~ 2 χ { v < ε ξ ω } d x ≥ ( 1 − ε ) 2 ( ξ ω ) 2 ∣ A ε , ( 1 − α ) ϱ ( t ) ∣ , \begin{aligned}
\int_{B_{(1-\alpha)\varrho\times\{t\}}}\widetilde{w}^{2}\,dx \geq\int_{B_{(1-\alpha)\varrho\times\{t\}}}\widetilde{w}^{2}\chi_{\{v<\varepsilon\xi\omega\}}\,dx \\
&\geq(1-\varepsilon)^{2}(\xi\omega)^{2}|A_{\varepsilon,(1-\alpha)\varrho}(t)|,
\end{aligned} ∫ B ( 1 − α ) ϱ × { t } w 2 d x ≥ ∫ B ( 1 − α ) ϱ × { t } w 2 χ { v < ε ξ ω } d x ≥ ( 1 − ε ) 2 ( ξ ω ) 2 ∣ A ε , ( 1 − α ) ϱ ( t ) ∣ ,
where
A ε , ( 1 − α ) ϱ ( t ) : = { v ( ⋅ , t ) < ε ξ ω } ∩ B ( 1 − α ) ϱ . A_{\varepsilon,(1-\alpha)\varrho}(t):=\{v(\cdot,t)<\varepsilon\xi\omega\}\cap B_{(1-\alpha)\varrho}. A ε , ( 1 − α ) ϱ ( t ) := { v ( ⋅ , t ) < ε ξ ω } ∩ B ( 1 − α ) ϱ .
Now, observe that
∣ A ε , ϱ ( t ) ∣ = ∣ A ε , ( 1 − α ) ϱ ( t ) ∪ ( A ε , ϱ ( t ) ∖ A ε , ( 1 − α ) ϱ ( t ) ) ∣ ≤ ∣ A ε , ( 1 − α ) ϱ ( t ) ∣ + ∣ B ϱ ∖ B ( 1 − α ) ϱ ∣ ≤ ∣ A ε , ( 1 − α ) ϱ ( t ) ∣ + N α ∣ B ϱ ∣ . (36) \begin{aligned}
|A_{\varepsilon,\varrho}(t)|=|A_{\varepsilon,(1-\alpha)\varrho}(t)\cup(A_{\varepsilon,\varrho}(t)\setminus A_{\varepsilon,(1-\alpha)\varrho}(t))| \\
&\leq|A_{\varepsilon,(1-\alpha)\varrho}(t)|+|B_{\varrho}\setminus B_{(1-\alpha)\varrho}| \\
&\leq|A_{\varepsilon,(1-\alpha)\varrho}(t)|+N\alpha|B_{\varrho}|. \tag*{(36)}
\end{aligned} ∣ A ε , ϱ ( t ) ∣ = ∣ A ε , ( 1 − α ) ϱ ( t ) ∪ ( A ε , ϱ ( t ) ∖ A ε , ( 1 − α ) ϱ ( t )) ∣ ≤ ∣ A ε , ( 1 − α ) ϱ ( t ) ∣ + ∣ B ϱ ∖ B ( 1 − α ) ϱ ∣ ≤ ∣ A ε , ( 1 − α ) ϱ ( t ) ∣ + N α ∣ B ϱ ∣. (36)
Combining (35) and (36) yields for all t ∈ ( 0 , θ ϱ 2 s ) t\in(0,\theta\varrho^{2s}) t ∈ ( 0 , θ ϱ 2 s ) that
∣ A ε , ϱ ( t ) ∣ ≤ 1 ( 1 − ε ) 2 [ ( 1 − λ ) + C θ α N + 2 s ] ∣ B ϱ ∣ + N α ∣ B ϱ ∣ . |A_{\varepsilon,\varrho}(t)|\leq\frac{1}{(1-\varepsilon)^{2}}\left[(1-\lambda)+C\frac{\theta}{\alpha^{N+2s}}\right]|B_{\varrho}|+N\alpha|B_{\varrho}|. ∣ A ε , ϱ ( t ) ∣ ≤ ( 1 − ε ) 2 1 [ ( 1 − λ ) + C α N + 2 s θ ] ∣ B ϱ ∣ + N α ∣ B ϱ ∣.
In order to finish the proof we choose the parameters α \alpha α , θ \theta θ and ε \varepsilon ε in such a way that
N α = 1 4 λ , C θ α N + 2 s = 1 8 λ and 1 − 7 8 λ ( 1 − ε ) 2 ≤ 1 − 3 4 λ N\alpha=\frac{1}{4}\lambda,\qquad C\frac{\theta}{\alpha^{N+2s}}=\frac{1}{8}\lambda\qquad\text{and}\qquad\frac{1-\frac{7}{8}\lambda}{(1-\varepsilon)^{2}}\leq1-\frac{3}{4}\lambda N α = 4 1 λ , C α N + 2 s θ = 8 1 λ and ( 1 − ε ) 2 1 − 8 7 λ ≤ 1 − 4 3 λ
are satisfied. Therefore, we deduce for all t ∈ ( 0 , θ ϱ 2 s ) t\in(0,\theta\varrho^{2s}) t ∈ ( 0 , θ ϱ 2 s ) that
∣ A ε , ϱ ( t ) ∣ ≤ ( 1 − 1 2 λ ) ∣ B ϱ ∣ . |A_{\varepsilon,\varrho}(t)|\leq\left(1-\frac{1}{2}\lambda\right)|B_{\varrho}|. ∣ A ε , ϱ ( t ) ∣ ≤ ( 1 − 2 1 λ ) ∣ B ϱ ∣.
This is just rephrasing the desired estimate. □
We conclude this section by deriving a measure shrinking lemma.
Lemma 6.5. Let u u u be a locally bounded, local weak solution to (3) in Ω T \Omega_T Ω T , as well as σ ∈ ( 0 , 1 2 ) \sigma\in(0,\frac{1}{2}) σ ∈ ( 0 , 2 1 ) and θ , ξ , λ ∈ ( 0 , 1 ) \theta,\xi,\lambda\in(0,1) θ , ξ , λ ∈ ( 0 , 1 ) . Suppose that for all t ∈ ( t o − θ ϱ 2 s , t o ] ⊂ ( t 1 , t 2 ) t\in(t_o-\theta\varrho^{2s},t_o]\subset(t_1,t_2) t ∈ ( t o − θ ϱ 2 s , t o ] ⊂ ( t 1 , t 2 ) we have
∣ { ± ( μ ± − v ( ⋅ , t ) ) ≥ ξ ω } ∩ B ϱ ( x o ) ∣ ≥ λ ∣ B ϱ ∣ . |\{\pm(\mu^\pm-v(\cdot,t))\geq\xi\omega\}\cap B_{\varrho}(x_o)|\geq\lambda|B_{\varrho}|. ∣ { ± ( μ ± − v ( ⋅ , t )) ≥ ξ ω } ∩ B ϱ ( x o ) ∣ ≥ λ ∣ B ϱ ∣. There exists C > 1 C>1 C > 1 depending only on the data, such that either
1 θ ∫ t 1 t 2 ∥ f ( ⋅ , t ) ∥ L ∞ ( B 2 ϱ ) d t > σ ξ ω \frac{1}{\theta}\int_{t_1}^{t_2}\|f(\cdot,t)\|_{L^\infty(B_{2\varrho})}\,dt>\sigma\xi\omega θ 1 ∫ t 1 t 2 ∥ f ( ⋅ , t ) ∥ L ∞ ( B 2 ϱ ) d t > σ ξ ω or
∣ { ± ( μ ± − v ) ≤ σ ξ ω } ∩ Q ϱ ( θ ) ∣ ≤ C σ θ λ ∣ Q ϱ ( θ ) ∣ |\{\pm(\mu^\pm-v)\leq\sigma\xi\omega\}\cap Q_{\varrho}(\theta)|\leq C\frac{\sigma}{\theta\lambda}|Q_{\varrho}(\theta)| ∣ { ± ( μ ± − v ) ≤ σ ξ ω } ∩ Q ϱ ( θ ) ∣ ≤ C θ λ σ ∣ Q ϱ ( θ ) ∣ provided Q 4 ϱ ( θ ) ⊂ Q Q_{4\varrho}(\theta)\subset Q Q 4 ϱ ( θ ) ⊂ Q .
Proof. We take ( x o , t o ) = ( 0 , 0 ) (x_o,t_o)=(0,0) ( x o , t o ) = ( 0 , 0 ) , still denote the reference cylinder B R × ( t 1 , t 2 ] B_R\times(t_1,t_2] B R × ( t 1 , t 2 ] , and prove the assertion for μ − = 0 \mu^-=0 μ − = 0 . Next, introduce the truncation
w ~ − ( x , t ) : = ( v ( x , t ) − σ ξ ω ) − \widetilde{w}_{-}(x,t):=(v(x,t)-\sigma\xi\omega)_{-} w − ( x , t ) := ( v ( x , t ) − σ ξ ω ) −
and a cutoff function ζ \zeta ζ in B 2 ϱ B_{2\varrho} B 2 ϱ , such that ζ = 1 \zeta=1 ζ = 1 in B ϱ B_{\varrho} B ϱ and ζ = 0 \zeta=0 ζ = 0 outside B 2 ϱ B_{2\varrho} B 2 ϱ satisfying ∣ D ζ ∣ ≤ 4 ϱ |D\zeta|\leq\frac{4}{\varrho} ∣ D ζ ∣ ≤ ϱ 4 . Then, from Lemma 6.1 , applied in B 4 ϱ × ( − θ ϱ 2 s , 0 ] B_{4\varrho}\times(-\theta\varrho^{2s},0] B 4 ϱ × ( − θ ϱ 2 s , 0 ] with w ~ − \widetilde{w}_{-} w − , ζ \zeta ζ and k = σ ξ ω k=\sigma\xi\omega k = σ ξ ω , we deduce
Λ − 1 ∬ Q ϱ ( θ ) w ~ − ( x , t ) ( ∫ B 4 ϱ w ~ + ( y , t ) ∣ x − y ∣ N + 2 s d y ) d x d t \Lambda^{-1}\iint_{Q_{\varrho}(\theta)}\widetilde{w}_{-}(x,t)\left(\int_{B_{4\varrho}}\frac{\widetilde{w}_{+}(y,t)}{|x-y|^{N+2s}}\,dy\right)dxdt Λ − 1 ∬ Q ϱ ( θ ) w − ( x , t ) ( ∫ B 4 ϱ ∣ x − y ∣ N + 2 s w + ( y , t ) d y ) d x d t
≤ 2 Λ ∫ − θ ϱ 2 s 0 ∫ B 4 ϱ ∫ B 4 ϱ max { w ~ − 2 ( x , t ) , w ~ − 2 ( y , t ) } ∣ ζ ( x , t ) − ζ ( y , t ) ∣ 2 ∣ x − y ∣ N + 2 s d y d x d t + 4 Λ ∫ − θ ϱ 2 s 0 ∫ B 4 ϱ [ ζ 2 w ~ − 2 ( x , t ) + σ ξ ω ζ 2 w ~ − ( x , t ) ] ∫ R N ∖ B 4 ϱ d y ∣ x − y ∣ N + 2 s d x d t + 4 Λ ∫ − θ ϱ 2 s 0 ∥ f ( ⋅ , t ) ∥ L ∞ ( spt ζ ) ∫ B 4 ϱ ζ 2 w ~ − ( x , t ) d x d t + ∫ B 4 ϱ × { x , − θ ϱ 2 s } ζ 2 w ~ − 2 d x = : I + I I + I I I + I V . \begin{aligned}
&\le2\Lambda\int_{-\theta\varrho^{2s}}^0\int_{B_{4\varrho}}\int_{B_{4\varrho}}
\max\{\widetilde w_-^2(x,t),\widetilde w_-^2(y,t)\}
\frac{|\zeta(x,t)-\zeta(y,t)|^2}{|x-y|^{N+2s}}\,dydxdt\\
&\quad+4\Lambda\int_{-\theta\varrho^{2s}}^0\int_{B_{4\varrho}}
\left[\zeta^2\widetilde w_-^2(x,t)+\sigma\xi\omega\zeta^2\widetilde w_-(x,t)\right]
\int_{\mathbb{R}^N\setminus B_{4\varrho}}\frac{dy}{|x-y|^{N+2s}}\,dxdt\\
&\quad+4\Lambda\int_{-\theta\varrho^{2s}}^0
\|f(\cdot,t)\|_{L^\infty(\operatorname{spt}\zeta)}
\int_{B_{4\varrho}}\zeta^2\widetilde w_-(x,t)\,dxdt\\
&\quad+\int_{B_{4\varrho}\times\{x,-\theta\varrho^{2s}\}}\zeta^2\widetilde w_-^2\,dx\\
&=: \mathrm{I}+\mathrm{II}+\mathrm{III}+\mathrm{IV}.
\end{aligned} ≤ 2Λ ∫ − θ ϱ 2 s 0 ∫ B 4 ϱ ∫ B 4 ϱ max { w − 2 ( x , t ) , w − 2 ( y , t )} ∣ x − y ∣ N + 2 s ∣ ζ ( x , t ) − ζ ( y , t ) ∣ 2 d y d x d t + 4Λ ∫ − θ ϱ 2 s 0 ∫ B 4 ϱ [ ζ 2 w − 2 ( x , t ) + σ ξ ω ζ 2 w − ( x , t ) ] ∫ R N ∖ B 4 ϱ ∣ x − y ∣ N + 2 s d y d x d t + 4Λ ∫ − θ ϱ 2 s 0 ∥ f ( ⋅ , t ) ∥ L ∞ ( spt ζ ) ∫ B 4 ϱ ζ 2 w − ( x , t ) d x d t + ∫ B 4 ϱ × { x , − θ ϱ 2 s } ζ 2 w − 2 d x =: I + II + III + IV .
Since, the terms I, II and III are treated similar as in the proof of Lemma 6.4 , we will only present the main difference. Starting with the first term we obtain
I ≤ C θ ϱ 2 s ( σ ξ ω ) 2 ϱ 2 ∣ B 4 ϱ ∣ ϱ 2 ( s − 1 ) ≤ C ( σ ξ ω ) 2 ϱ 2 s ∣ Q ϱ ( θ ) ∣ . \mathrm{I}\le C\theta\varrho^{2s}\frac{(\sigma\xi\omega)^2}{\varrho^2}\frac{|B_{4\varrho}|}{\varrho^{2(s-1)}}\le C\frac{(\sigma\xi\omega)^2}{\varrho^{2s}}|Q_\varrho(\theta)|. I ≤ C θ ϱ 2 s ϱ 2 ( σ ξ ω ) 2 ϱ 2 ( s − 1 ) ∣ B 4 ϱ ∣ ≤ C ϱ 2 s ( σ ξ ω ) 2 ∣ Q ϱ ( θ ) ∣.
Proceeding with II note that for ∣ x ∣ ≤ 3 4 ϱ |x|\le\frac{3}{4}\varrho ∣ x ∣ ≤ 4 3 ϱ and ∣ y ∣ ≥ 2 ϱ |y|\ge2\varrho ∣ y ∣ ≥ 2 ϱ we have
∣ x − y ∣ ∣ y ∣ ≥ 1 − ∣ x ∣ ∣ y ∣ ≥ 1 4 . \frac{|x-y|}{|y|}\ge1-\frac{|x|}{|y|}\ge\frac{1}{4}. ∣ y ∣ ∣ x − y ∣ ≥ 1 − ∣ y ∣ ∣ x ∣ ≥ 4 1 .
Thus, we obtain analogously as in the proof of Lemma 6.3 that
I I ≤ C θ ( σ ξ ω ) 2 ∣ B 4 ϱ ∣ ≤ C 4 N ( σ ξ ω ) 2 ϱ 2 s ∣ Q ϱ ( θ ) ∣ . \mathrm{II}\le C\theta(\sigma\xi\omega)^2|B_{4\varrho}|\le C4^N\frac{(\sigma\xi\omega)^2}{\varrho^{2s}}|Q_\varrho(\theta)|. II ≤ C θ ( σ ξ ω ) 2 ∣ B 4 ϱ ∣ ≤ C 4 N ϱ 2 s ( σ ξ ω ) 2 ∣ Q ϱ ( θ ) ∣.
For III, since
1 θ ∫ t 1 t 2 ∥ f ( ⋅ , t ) ∥ L ∞ ( B 2 ϱ ) d t ≤ σ ξ ω , \frac{1}{\theta}\int_{t_1}^{t_2}\|f(\cdot,t)\|_{L^\infty(B_{2\varrho})}\,dt\le\sigma\xi\omega, θ 1 ∫ t 1 t 2 ∥ f ( ⋅ , t ) ∥ L ∞ ( B 2 ϱ ) d t ≤ σ ξ ω ,
it follows that
I I I ≤ C θ ( σ ξ ω ) 2 ∣ B 4 ϱ ∣ ≤ C 4 N ( σ ξ ω ) 2 ϱ 2 s ∣ Q ϱ ( θ ) ∣ . \mathrm{III}\le C\theta(\sigma\xi\omega)^2|B_{4\varrho}|\le C4^N\frac{(\sigma\xi\omega)^2}{\varrho^{2s}}|Q_\varrho(\theta)|. III ≤ C θ ( σ ξ ω ) 2 ∣ B 4 ϱ ∣ ≤ C 4 N ϱ 2 s ( σ ξ ω ) 2 ∣ Q ϱ ( θ ) ∣.
Next, IV is estimated by
I V ≤ ( σ ξ ω ) 2 ∣ B 4 ϱ ∣ ≤ C ( σ ξ ω ) 2 θ ϱ 2 s ∣ Q ϱ ( θ ) ∣ . \mathrm{IV}\le(\sigma\xi\omega)^2|B_{4\varrho}|\le C\frac{(\sigma\xi\omega)^2}{\theta\varrho^{2s}}|Q_\varrho(\theta)|. IV ≤ ( σ ξ ω ) 2 ∣ B 4 ϱ ∣ ≤ C θ ϱ 2 s ( σ ξ ω ) 2 ∣ Q ϱ ( θ ) ∣.
Note that we used ∣ B 4 ϱ ∣ = 4 N ∣ B ϱ ∣ |B_{4\varrho}|=4^N|B_\varrho| ∣ B 4 ϱ ∣ = 4 N ∣ B ϱ ∣ . By combining all these estimate, while shifting the constant Λ − 1 \Lambda^{-1} Λ − 1 to the right-hand side and absorbing it into C C C , we can find the bound
∬ Q ϱ ( θ ) ∫ B 4 ϱ w ~ − ( x , t ) w ~ + ( y , t ) ∣ x − y ∣ N + 2 s d y d x d t ≤ C ( σ ξ ω ) 2 ϱ 2 s ∣ Q ϱ ( θ ) ∣ . (37) \iint_{Q_\varrho(\theta)}\int_{B_{4\varrho}}
\frac{\widetilde w_-(x,t)\widetilde w_+(y,t)}{|x-y|^{N+2s}}\,dydxdt
\le C\frac{(\sigma\xi\omega)^2}{\varrho^{2s}}|Q_\varrho(\theta)|. \tag*{(37)} ∬ Q ϱ ( θ ) ∫ B 4 ϱ ∣ x − y ∣ N + 2 s w − ( x , t ) w + ( y , t ) d y d x d t ≤ C ϱ 2 s ( σ ξ ω ) 2 ∣ Q ϱ ( θ ) ∣. (37)
Note that C C C only depends on the data. Next, by using the measure theoretical information the left-hand side can be estimated from below by
∬ Q ϱ ( θ ) w ~ − ( x , t ) χ { v ( x , t ) ≤ 1 2 σ ξ ω } ( ∫ B 4 ϱ w ~ + ( y , t ) χ { v ( y , t ) ≥ σ ξ ω } ∣ x − y ∣ N + 2 s d y ) d x d t ≥ 1 2 σ ξ ω ∣ { v ≤ 1 2 σ ξ ω } ∩ Q ϱ ( θ ) ∣ ( ξ ω λ ∣ B ϱ ∣ 2 ( 8 ϱ ) N + 2 s ) = 2 − ( 4 N + 2 + 4 s ) ( ξ ω ) 2 λ σ ϱ 2 s ∣ { v ≤ 1 2 σ ξ ω } ∩ Q ϱ ( θ ) ∣ . (38) \begin{aligned}
\iint_{Q_\varrho(\theta)}\widetilde w_-(x,t)\chi_{\{v(x,t)\le\frac{1}{2}\sigma\xi\omega\}}
\left(\int_{B_{4\varrho}}\frac{\widetilde w_+(y,t)\chi_{\{v(y,t)\ge\sigma\xi\omega\}}}{|x-y|^{N+2s}}\,dy\right)\,dxdt\\
\ge\frac{1}{2}\sigma\xi\omega|\{v\le\tfrac{1}{2}\sigma\xi\omega\}\cap Q_\varrho(\theta)|
\left(\frac{\xi\omega\lambda|B_\varrho|}{2(8\varrho)^{N+2s}}\right)\\
=2^{-(4N+2+4s)}\frac{(\xi\omega)^2\lambda\sigma}{\varrho^{2s}}
|\{v\le\tfrac{1}{2}\sigma\xi\omega\}\cap Q_\varrho(\theta)|. \tag*{(38)}
\end{aligned} ∬ Q ϱ ( θ ) w − ( x , t ) χ { v ( x , t ) ≤ 2 1 σ ξ ω } ( ∫ B 4 ϱ ∣ x − y ∣ N + 2 s w + ( y , t ) χ { v ( y , t ) ≥ σ ξ ω } d y ) d x d t ≥ 2 1 σ ξ ω ∣ { v ≤ 2 1 σ ξ ω } ∩ Q ϱ ( θ ) ∣ ( 2 ( 8 ϱ ) N + 2 s ξ ω λ ∣ B ϱ ∣ ) = 2 − ( 4 N + 2 + 4 s ) ϱ 2 s ( ξ ω ) 2 λσ ∣ { v ≤ 2 1 σ ξ ω } ∩ Q ϱ ( θ ) ∣. (38)
We combine (37) with (38) and adjust the constants appropriately to finish the proof. □ \square □
Proof of Proposition 6.2
We prove the assertion for μ − = 0 \mu^{-}=0 μ − = 0 and ( x o , t o ) = ( 0 , 0 ) (x_{o},t_{o})=(0,0) ( x o , t o ) = ( 0 , 0 ) . The reference cylinder B R × ( t 1 , t 2 ] B_{R}\times(t_{1},t_{2}] B R × ( t 1 , t 2 ] will be used. Let us rewrite the measure theoretical information at the initial time t o = 0 t_{o}=0 t o = 0 in the larger ball B 4 ϱ B_{4\varrho} B 4 ϱ , replace λ \lambda λ by 4 − N λ 4^{-N}\lambda 4 − N λ , and apply Lemma 6.4 in order to obtain θ \theta θ , ε ∈ ( 0 , 1 ) \varepsilon\in(0,1) ε ∈ ( 0 , 1 ) depending only on the data and λ \lambda λ , such that
∣ { v ( ⋅ , t ) ≥ ε ξ ω } ∩ B 4 ϱ ∣ ≥ 1 2 4 − N λ ∣ B 4 ϱ ∣ \left|\{v(\cdot,t)\geq\varepsilon\xi\omega\}\cap B_{4\varrho}\right|\geq\frac{1}{2}4^{-N}\lambda|B_{4\varrho}| ∣ { v ( ⋅ , t ) ≥ ε ξ ω } ∩ B 4 ϱ ∣ ≥ 2 1 4 − N λ ∣ B 4 ϱ ∣
for all t ∈ ( 0 , θ ( 4 ϱ ) 2 s ] t\in(0,\theta(4\varrho)^{2s}] t ∈ ( 0 , θ ( 4 ϱ ) 2 s ] , provided we enforce
1 θ ∫ t 1 t 2 ∥ f ( ⋅ , t ) ∥ L ∞ ( B 8 ϱ ) d t ≤ ξ ω . \frac{1}{\theta}\int_{t_{1}}^{t_{2}}\|f(\cdot,t)\|_{L^{\infty}(B_{8\varrho})}\,dt\leq\xi\omega. θ 1 ∫ t 1 t 2 ∥ f ( ⋅ , t ) ∥ L ∞ ( B 8 ϱ ) d t ≤ ξ ω .
This slice-wise measure theoretical estimate enables us to apply Lemma 6.5 in the cylinders
( 0 , t ˉ ) + Q 4 ϱ ( 1 2 θ ) : = B 4 ϱ × ( t ˉ − 1 2 θ ( 4 ϱ ) 2 s , t ˉ ] (0,\bar{t})+Q_{4\varrho}\left(\frac{1}{2}\theta\right):=B_{4\varrho}\times\left(\bar{t}-\frac{1}{2}\theta(4\varrho)^{2s},\bar{t}\right] ( 0 , t ˉ ) + Q 4 ϱ ( 2 1 θ ) := B 4 ϱ × ( t ˉ − 2 1 θ ( 4 ϱ ) 2 s , t ˉ ]
for any arbitrary t ˉ ∈ ( 1 2 θ ( 4 ϱ ) 2 s , θ ( 4 ϱ ) 2 s ] \bar{t}\in\left(\frac{1}{2}\theta(4\varrho)^{2s},\theta(4\varrho)^{2s}\right] t ˉ ∈ ( 2 1 θ ( 4 ϱ ) 2 s , θ ( 4 ϱ ) 2 s ] and with ξ , λ \xi,\lambda ξ , λ replaced by ε ξ , 1 2 4 − N λ \varepsilon\xi,\frac{1}{2}4^{-N}\lambda ε ξ , 2 1 4 − N λ respectively. Hence, the backward cylinders considered are contained in B 4 ϱ × ( 0 , θ ( 4 ϱ ) 2 s ] B_{4\varrho}\times(0,\theta(4\varrho)^{2s}] B 4 ϱ × ( 0 , θ ( 4 ϱ ) 2 s ] . Now let ν \nu ν be determined according to Lemma 6.3 depending only on the data and θ \theta θ . Moreover, choose σ \sigma σ according to Lemma 6.5 to satisfy
C σ θ λ < ν , i.e. σ ≤ ν θ λ C . C\frac{\sigma}{\theta\lambda}<\nu,\qquad\text{i.e.}\quad\sigma\leq\frac{\nu\theta\lambda}{C}. C θ λ σ < ν , i.e. σ ≤ C ν θ λ .
This choice is admissible, since C C C from Lemma 6.5 is independent of σ \sigma σ . Furthermore, letting γ ~ \tilde{\gamma} γ ~ be chosen in Lemma 6.3 , we enforce that
max { γ ~ , 1 / θ } ∫ t 1 t 2 ∥ f ( ⋅ , t ) ∥ L ∞ ( B 8 ϱ ) d t ≤ σ ε ξ ω , \max\{\tilde{\gamma},1/\theta\}\int_{t_{1}}^{t_{2}}\|f(\cdot,t)\|_{L^{\infty}(B_{8\varrho})}\,dt\leq\sigma\varepsilon\xi\omega, max { γ ~ , 1/ θ } ∫ t 1 t 2 ∥ f ( ⋅ , t ) ∥ L ∞ ( B 8 ϱ ) d t ≤ σ ε ξ ω ,
and apply Lemma 6.5 with the above chosen σ \sigma σ , followed by Lemma 6.3 with ξ \xi ξ replaced by σ ε ξ \sigma\varepsilon\xi σ ε ξ , in the cylinder
( 0 , t ˉ ) + Q 4 ϱ ( 1 2 θ ) = B 4 ϱ × ( t ˉ − 1 2 θ ( 4 ϱ ) 2 s , t ˉ ] (0,\bar{t})+Q_{4\varrho}\left(\frac{1}{2}\theta\right)=B_{4\varrho}\times\left(\bar{t}-\frac{1}{2}\theta(4\varrho)^{2s},\bar{t}\right] ( 0 , t ˉ ) + Q 4 ϱ ( 2 1 θ ) = B 4 ϱ × ( t ˉ − 2 1 θ ( 4 ϱ ) 2 s , t ˉ ]
for an arbitrary t ˉ ∈ ( 1 2 θ ( 4 ϱ ) 2 s , θ ( 4 ϱ ) 2 s ] \bar{t}\in\left(\frac{1}{2}\theta(4\varrho)^{2s},\theta(4\varrho)^{2s}\right] t ˉ ∈ ( 2 1 θ ( 4 ϱ ) 2 s , θ ( 4 ϱ ) 2 s ] . Due to the arbitrariness of t ˉ \bar{t} t ˉ we conclude that
v ≥ 1 4 σ ε ξ ω a.e. in B 2 ϱ × ( 1 2 θ ( 4 ϱ ) 2 s , θ ( 4 ϱ ) 2 s ] . v\geq\frac{1}{4}\sigma\varepsilon\xi\omega\quad\text{a.e. in }B_{2\varrho}\times\left(\frac{1}{2}\theta(4\varrho)^{2s},\theta(4\varrho)^{2s}\right]. v ≥ 4 1 σ ε ξ ω a.e. in B 2 ϱ × ( 2 1 θ ( 4 ϱ ) 2 s , θ ( 4 ϱ ) 2 s ] .
The proof is completed by setting η = 1 4 σ ε \eta=\frac{1}{4}\sigma\varepsilon η = 4 1 σ ε , properly adjusting the relevant constants in dependence of the data, as well as λ \lambda λ and relabeling max { γ ~ , 1 / θ } \max\{\tilde{\gamma},1/\theta\} max { γ ~ , 1/ θ } .
Continuity of the gradient
Recall that v ≡ v h v\equiv v_{h} v ≡ v h is defined in (30) . Before proving the continuity of D u Du D u , we exploit Proposition 6.2 to prove a modulus of continuity of v h v_{h} v h in the next result.
Proposition 7.1. Let u u u be a local weak solution to (3) in the sense of Definition 2.1. Then, there exist C > 1 C>1 C > 1 and β ∈ ( 0 , 1 ) \beta\in(0,1) β ∈ ( 0 , 1 ) depending only on the data, such that for 0 < r < R 0<r<R 0 < r < R , any h ∈ R N ∖ { 0 } h\in\mathbb{R}^{N}\setminus\{0\} h ∈ R N ∖ { 0 } with 0 < ∣ h ∣ ≤ 1 32 R 0<|h|\leq\frac{1}{32}R 0 < ∣ h ∣ ≤ 32 1 R , we have
ess osc Q r ( z o ) v h ≤ C [ ω ( r R ) β + ∫ t o − ( r R ) 2 s t o ∥ f ( ⋅ , t ) ∥ L ∞ ( B R / 8 ( x o ) ) d t ] . \operatorname*{ess\,osc}_{Q_{r}(z_{o})}v_{h}\leq C\left[\omega\left(\frac{r}{R}\right)^{\beta}+\int_{t_{o}-(rR)^{2s}}^{t_{o}}\|f(\cdot,t)\|_{L^{\infty}(B_{R/8}(x_{o}))}\,dt\right]. Q r ( z o ) ess osc v h ≤ C [ ω ( R r ) β + ∫ t o − ( r R ) 2 s t o ∥ f ( ⋅ , t ) ∥ L ∞ ( B R /8 ( x o )) d t ] . for some constant C = C ( d a t a ) C=C(\mathrm{data}) C = C ( data ) independent of h h h , where f f f is defined in (31) , i.e.
f ( x , t ) : = ∫ R N ∖ B R ( x o ) v h ( y , t ) ∣ x − y ∣ N + 2 s d y f(x,t):=\int_{\mathbb{R}^{N}\setminus B_{R}(x_{o})}\frac{v_{h}(y,t)}{|x-y|^{N+2s}}\,dy f ( x , t ) := ∫ R N ∖ B R ( x o ) ∣ x − y ∣ N + 2 s v h ( y , t ) d y and
ω : = 1 R ∫ t o − ( 64 R ) 2 s t o ∫ R N ∣ u ( x , t ) ∣ ( R + ∣ x − x o ∣ ) N + 2 s d x d t . (39) \omega:=\frac{1}{R}\int_{t_{o}-(64R)^{2s}}^{t_{o}}\int_{\mathbb{R}^{N}}\frac{|u(x,t)|}{(R+|x-x_{o}|)^{N+2s}}\,dx\,dt. \tag*{(39)} ω := R 1 ∫ t o − ( 64 R ) 2 s t o ∫ R N ( R + ∣ x − x o ∣ ) N + 2 s ∣ u ( x , t ) ∣ d x d t . (39)
Proof of Proposition 7.1
7.1.1. The first step. For R > 0 R>0 R > 0 let us introduce the cylinders at z o = ( x o , t o ) z_o=(x_o,t_o) z o = ( x o , t o ) :
Q o : = Q R ( z o ) ⊂ Q 2 R ( z o ) ⊂ Ω T . Q_o:=Q_R(z_o)\subset Q_{2R}(z_o)\subset\Omega_T. Q o := Q R ( z o ) ⊂ Q 2 R ( z o ) ⊂ Ω T .
For simplicity assume ( x o , t o ) = ( 0 , 0 ) (x_o,t_o)=(0,0) ( x o , t o ) = ( 0 , 0 ) . Consider a quantity ω \omega ω satisfying
ω ≥ 2 ess sup Q 2 R ∣ v ∣ . (40) \omega\ge2\operatorname*{ess\,sup}_{Q_{2R}}|v|. \tag*{(40)} ω ≥ 2 Q 2 R ess sup ∣ v ∣. (40)
In the end, we will find that this ω \omega ω can be selected as in (39) apart from a multiplicative constant. To continue, we define
μ + : = ess sup Q o v , μ − : = ess inf Q o v . \mu^+ := \operatorname*{ess\,sup}_{Q_o}v,\qquad\mu^- := \operatorname*{ess\,inf}_{Q_o}v. μ + := Q o ess sup v , μ − := Q o ess inf v .
We have from (40) the initial oscillation control
ess osc Q o v ≤ ω . (41) \operatorname*{ess\,osc}_{Q_o}v\le\omega. \tag*{(41)} Q o ess osc v ≤ ω . (41)
We will reduce the oscillation in a smaller cylinder contained in Q o Q_o Q o . To this end, fix θ = θ ( d a t a ) ∈ ( 0 , 1 ) \theta=\theta(\mathit{data})\in(0,1) θ = θ ( data ) ∈ ( 0 , 1 ) according to Proposition 6.2 with λ = 1 2 \lambda=\frac12 λ = 2 1 ; define
τ : = θ ( 1 9 R ) 2 s ; \tau:=\theta\left(\frac19R\right)^{2s}; τ := θ ( 9 1 R ) 2 s ;
consider the two alternatives:
{ ∣ { v ( ⋅ , − τ ) − μ − > 1 4 ω } ∩ B 1 9 R ∣ ≥ 1 2 ∣ B 1 9 R ∣ , ∣ { μ + − v ( ⋅ , − τ ) > 1 4 ω } ∩ B 1 9 R ∣ ≥ 1 2 ∣ B 1 9 R ∣ . (42) \left\{
\begin{aligned}
&\left|\{v(\cdot,-\tau)-\mu^->\tfrac14\omega\}\cap B_{\frac19R}\right|\ge\tfrac12|B_{\frac19R}|,\\
&\left|\{\mu^+-v(\cdot,-\tau)>\tfrac14\omega\}\cap B_{\frac19R}\right|\ge\tfrac12|B_{\frac19R}|.
\end{aligned}
\right. \tag*{(42)} ⎩ ⎨ ⎧ { v ( ⋅ , − τ ) − μ − > 4 1 ω } ∩ B 9 1 R ≥ 2 1 ∣ B 9 1 R ∣ , { μ + − v ( ⋅ , − τ ) > 4 1 ω } ∩ B 9 1 R ≥ 2 1 ∣ B 9 1 R ∣. (42)
We may assume μ + − μ − ≥ 1 2 ω \mu^+-\mu^-\ge\frac12\omega μ + − μ − ≥ 2 1 ω , then one of the two in (42) must hold, whereas the reverse case μ + − μ − < 1 2 ω \mu^+-\mu^-<\frac12\omega μ + − μ − < 2 1 ω is favorable and automatically yields the aimed oscillation estimate (43) . Now, suppose that the first alternative holds; apply Proposition 6.2 with λ = 1 2 \lambda=\frac12 λ = 2 1 , ξ = 1 4 \xi=\frac14 ξ = 4 1 and ϱ = 1 9 R \varrho=\frac19R ϱ = 9 1 R to determine η = η ( d a t a ) ∈ ( 0 , 1 ) \eta=\eta(\mathit{data})\in(0,1) η = η ( data ) ∈ ( 0 , 1 ) , as well as t 1 = − ( 1 9 R ) 2 s t_1=-\left(\frac19R\right)^{2s} t 1 = − ( 9 1 R ) 2 s and t 2 = 0 t_2=0 t 2 = 0 , such that either
γ ~ ∫ − ( 1 9 R ) 2 s 0 ∥ f ( ⋅ , t ) ∥ L ∞ ( B 8 9 R ) d t > 1 4 η ω \widetilde{\gamma}\int_{-\left(\frac19R\right)^{2s}}^0\|f(\cdot,t)\|_{L^\infty(B_{\frac89R})}\,dt>\tfrac14\eta\omega γ ∫ − ( 9 1 R ) 2 s 0 ∥ f ( ⋅ , t ) ∥ L ∞ ( B 9 8 R ) d t > 4 1 η ω
or
v − μ − ≥ 1 4 η ω a.e. in Q 1 9 R ( 1 2 θ ) . v-\mu^-\ge\tfrac14\eta\omega\quad\text{a.e. in }Q_{\frac19R}\left(\tfrac12\theta\right). v − μ − ≥ 4 1 η ω a.e. in Q 9 1 R ( 2 1 θ ) .
Denote γ ^ : = 4 γ ~ / η \widehat{\gamma}:=4\widetilde{\gamma}/\eta γ := 4 γ / η to simplify notation. Then, the above conclusion plus (41) yields that
ess osc Q 1 9 R ( 1 2 θ ) v ≤ max { ( 1 − 1 4 η ) ω , γ ^ ∫ − ( 1 9 R ) 2 s 0 ∥ f ( ⋅ , t ) ∥ L ∞ ( B 8 9 R ) d t } = : ω 1 . (43) \operatorname*{ess\,osc}_{Q_{\frac19R}(\frac12\theta)}v
\le\max\left\{\left(1-\tfrac14\eta\right)\omega,\widehat{\gamma}\int_{-\left(\frac19R\right)^{2s}}^0\|f(\cdot,t)\|_{L^\infty(B_{\frac89R})}\,dt\right\}=:\omega_1. \tag*{(43)} Q 9 1 R ( 2 1 θ ) ess osc v ≤ max { ( 1 − 4 1 η ) ω , γ ∫ − ( 9 1 R ) 2 s 0 ∥ f ( ⋅ , t ) ∥ L ∞ ( B 9 8 R ) d t } =: ω 1 . (43)
Next, we introduce R 1 = κ R R_1=\kappa R R 1 = κ R for some κ ≤ 1 9 \kappa\le\frac19 κ ≤ 9 1 in order to obtain the set inclusion
Q R 1 ⊂ Q 1 9 R ( 1 2 θ ) , i.e. κ : = 1 9 ( 1 2 θ ) 1 2 . Q_{R_1}\subset Q_{\frac19R}\left(\tfrac12\theta\right),\qquad\text{i.e. }\ \kappa:=\frac19\left(\tfrac12\theta\right)^{\frac12}. Q R 1 ⊂ Q 9 1 R ( 2 1 θ ) , i.e. κ := 9 1 ( 2 1 θ ) 2 1 .
As a result of this set inclusion and (43) we deduce
Q R 1 ⊂ Q o and ess osc Q R 1 v ≤ ω 1 , Q_{R_1}\subset Q_o\quad\text{and}\quad\operatorname*{ess\,osc}_{Q_{R_1}}v\le\omega_1, Q R 1 ⊂ Q o and Q R 1 ess osc v ≤ ω 1 ,
where the latter will play the role of (41) in the next step. Moreover, notice that the second alternative in (42) can be dealt with analogously.
The induction step
Assume for i = 1 , … , j i=1,\ldots,j i = 1 , … , j that we have built
R 0 = R , R i = κ R i − 1 , ω 0 = ω , B i = B R i , Q i = Q R i , Q ~ i = B i × ( − ( 1 9 R i ) 2 s , 0 ] , Q i ⊂ Q i − 1 , μ i + : = ess sup Q i v , μ i − : = ess inf Q i v , ess osc Q i v ≤ ω i , ω i = max { ( 1 − 1 4 η ) ω i − 1 , γ ^ ∫ − ( 1 9 R i − 1 ) 2 s 0 ∥ f ( ⋅ , t ) ∥ L ∞ ( B 8 9 R i − 1 ) d t } . \begin{gathered}
R_0=R,\quad R_i=\kappa R_{i-1},\quad\omega_0=\omega,\\
B_i=BR_i,\quad Q_i=Q_{R_i},\quad\widetilde{Q}_i=B_i\times\left(-\left(\frac{1}{9}R_i\right)^{2s},0\right],\quad Q_i\subset Q_{i-1},\\
\mu_i^+:=\operatorname{ess\,sup}_{Q_i}v,\quad\mu_i^-:=\operatorname{ess\,inf}_{Q_i}v,\quad\operatorname{ess\,osc}_{Q_i}v\leq\omega_i,\\
\omega_i=\max\left\{\left(1-\frac{1}{4}\eta\right)\omega_{i-1},\widehat{\gamma}\int_{-\left(\frac{1}{9}R_{i-1}\right)^{2s}}^0\|f(\cdot,t)\|_{L^\infty\left(B_{\frac{8}{9}R_{i-1}}\right)}\,dt\right\}.
\end{gathered} R 0 = R , R i = κ R i − 1 , ω 0 = ω , B i = B R i , Q i = Q R i , Q i = B i × ( − ( 9 1 R i ) 2 s , 0 ] , Q i ⊂ Q i − 1 , μ i + := ess sup Q i v , μ i − := ess inf Q i v , ess osc Q i v ≤ ω i , ω i = max ⎩ ⎨ ⎧ ( 1 − 4 1 η ) ω i − 1 , γ ∫ − ( 9 1 R i − 1 ) 2 s 0 ∥ f ( ⋅ , t ) ∥ L ∞ ( B 9 8 R i − 1 ) d t ⎭ ⎬ ⎫ .
Now we will show that the oscillation estimate continues to hold for the ( j + 1 ) (j+1) ( j + 1 ) -th step. Consider θ \theta θ fixed as above. We repeat the argument from Section 7.1.1 with μ j ± , ω j , Q j \mu_j^\pm,\omega_j,Q_j μ j ± , ω j , Q j and so on. Indeed, we define
τ j : = θ ( 1 9 R j ) 2 s \tau_j:=\theta\left(\frac{1}{9}R_j\right)^{2s} τ j := θ ( 9 1 R j ) 2 s
and consider the two alternatives:
{ ∣ { v ( ⋅ , − τ j ) − μ j − > 1 4 ω j } ∩ B 1 9 R j ∣ ≥ 1 2 ∣ B 1 9 R j ∣ , ∣ { μ j + − v ( ⋅ , − τ j ) > 1 4 ω j } ∩ B 1 9 R j ∣ ≥ 1 2 ∣ B 1 9 R j ∣ . (44) \left\{
\begin{aligned}
\left|\left\{v(\cdot,-\tau_j)-\mu_j^->\frac{1}{4}\omega_j\right\}\cap B_{\frac{1}{9}R_j}\right|&\geq\frac{1}{2}\left|B_{\frac{1}{9}R_j}\right|,\\
\left|\left\{\mu_j^+-v(\cdot,-\tau_j)>\frac{1}{4}\omega_j\right\}\cap B_{\frac{1}{9}R_j}\right|&\geq\frac{1}{2}\left|B_{\frac{1}{9}R_j}\right|.
\end{aligned}
\right. \tag*{(44)} ⎩ ⎨ ⎧ { v ( ⋅ , − τ j ) − μ j − > 4 1 ω j } ∩ B 9 1 R j { μ j + − v ( ⋅ , − τ j ) > 4 1 ω j } ∩ B 9 1 R j ≥ 2 1 B 9 1 R j , ≥ 2 1 B 9 1 R j . (44)
Like in the first step suppose that μ j + − μ j − ≥ 1 2 ω j \mu_j^+-\mu_j^-\geq\frac{1}{2}\omega_j μ j + − μ j − ≥ 2 1 ω j , such that one of the two alternatives is fulfilled. The case μ j + − μ j − ≤ 1 2 ω j \mu_j^+-\mu_j^-\leq\frac{1}{2}\omega_j μ j + − μ j − ≤ 2 1 ω j can be incorporated into the forthcoming oscillation estimate (45) . Without loss of generality assume that the first alternative in (44) holds. We apply Proposition 6.2 in Q j Q_j Q j , with λ = 1 2 , ξ = 1 4 \lambda=\frac{1}{2},\xi=\frac{1}{4} λ = 2 1 , ξ = 4 1 and ϱ = 1 9 R j \varrho=\frac{1}{9}R_j ϱ = 9 1 R j to get the same η \eta η as in the first step, as well as t 1 = − ( 1 9 R j ) 2 s t_1=-\left(\frac{1}{9}R_j\right)^{2s} t 1 = − ( 9 1 R j ) 2 s and t 2 = 0 t_2=0 t 2 = 0 , such that either
γ ^ ∫ − ( 1 9 R j ) 2 s 0 ∥ f ( ⋅ , t ) ∥ L ∞ ( B 8 9 R j ) d t > 1 4 η ω j \widehat{\gamma}\int_{-\left(\frac{1}{9}R_j\right)^{2s}}^0\|f(\cdot,t)\|_{L^\infty\left(B_{\frac{8}{9}R_j}\right)}\,dt>\frac{1}{4}\eta\omega_j γ ∫ − ( 9 1 R j ) 2 s 0 ∥ f ( ⋅ , t ) ∥ L ∞ ( B 9 8 R j ) d t > 4 1 η ω j
or
v − μ j − ≥ 1 4 η ω j a.e. in Q 1 9 R j ( 1 2 θ ) . v-\mu_j^-\geq\frac{1}{4}\eta\omega_j\quad\text{a.e. in }Q_{\frac{1}{9}R_j}\left(\frac{1}{2}\theta\right). v − μ j − ≥ 4 1 η ω j a.e. in Q 9 1 R j ( 2 1 θ ) .
Then, due to the j j j -th induction assumption we obtain that
ess osc Q 1 9 R j ( 1 2 θ ) v ≤ max { ( 1 − 1 4 η ) ω j , γ ^ ∫ − ( 1 9 R j ) 2 s 0 ∥ f ( ⋅ , t ) ∥ L ∞ ( B 8 9 R j ) d t } = : ω j + 1 . (45) \operatorname{ess\,osc}_{Q_{\frac{1}{9}R_j}\left(\frac{1}{2}\theta\right)}v\leq\max\left\{\left(1-\frac{1}{4}\eta\right)\omega_j,\widehat{\gamma}\int_{-\left(\frac{1}{9}R_j\right)^{2s}}^0\|f(\cdot,t)\|_{L^\infty\left(B_{\frac{8}{9}R_j}\right)}\,dt\right\}=:\omega_{j+1}. \tag*{(45)} ess osc Q 9 1 R j ( 2 1 θ ) v ≤ max ⎩ ⎨ ⎧ ( 1 − 4 1 η ) ω j , γ ∫ − ( 9 1 R j ) 2 s 0 ∥ f ( ⋅ , t ) ∥ L ∞ ( B 9 8 R j ) d t ⎭ ⎬ ⎫ =: ω j + 1 . (45)
where γ ^ = γ ^ ( d a t a ) \widehat{\gamma}=\widehat{\gamma}(\mathrm{data}) γ = γ ( data ) is as in (43) . Now, let R j + 1 = κ R j R_{j+1}=\kappa R_j R j + 1 = κ R j for the same κ = 1 9 ( 1 2 θ ) 1 2 s \kappa=\frac{1}{9}\left(\frac{1}{2}\theta\right)^{\frac{1}{2s}} κ = 9 1 ( 2 1 θ ) 2 s 1 chosen in the first step, in order to get that
Q R j + 1 ⊂ Q 1 9 R j ( 1 2 θ ) . (46) Q_{R_{j+1}}\subset Q_{\frac{1}{9}R_j}\left(\frac{1}{2}\theta\right). \tag*{(46)} Q R j + 1 ⊂ Q 9 1 R j ( 2 1 θ ) . (46)
As a consequence, (46) and (45) yield
Q R j + 1 ⊂ Q R j and ess osc Q R j + 1 v ≤ ω j + 1 , Q_{R_{j+1}}\subset Q_{R_j}\quad\text{and}\quad\operatorname{ess\,osc}_{Q_{R_{j+1}}}v\leq\omega_{j+1}, Q R j + 1 ⊂ Q R j and ess osc Q R j + 1 v ≤ ω j + 1 ,
which finishes the induction step for the first alternative in (44) . The same conclusion can be reached in case the second alternative of (44) holds true.
Modulus of continuity
From the previous section we have built R n = κ n R R_n=\kappa^nR R n = κ n R , Q n = Q R n Q_n=Q_{R_n} Q n = Q R n for n ∈ N n\in\mathbb{N} n ∈ N such that Q n ⊂ Q n − 1 Q_n\subset Q_{n-1} Q n ⊂ Q n − 1 and acquired the oscillation estimate
ess osc Q n v ≤ max { ( 1 − 1 4 η ) ω n − 1 , γ ^ ∫ − ( 1 9 R n − 1 ) 2 s 0 ∥ f ( ⋅ , t ) ∥ L ∞ ( B 8 9 R n − 1 ) d t } = : ω n . (47) \begin{aligned}
\operatorname{ess\,osc}_{Q_n}v\leq\max\left\{\left(1-\frac{1}{4}\eta\right)\omega_{n-1},\widehat{\gamma}\int_{-\left(\frac{1}{9}R_{n-1}\right)^{2s}}^0\|f(\cdot,t)\|_{L^\infty\left(B_{\frac{8}{9}R_{n-1}}\right)}\,dt\right\} \\
&=:\omega_n. \tag*{(47)}
\end{aligned} ess osc Q n v ≤ max ⎩ ⎨ ⎧ ( 1 − 4 1 η ) ω n − 1 , γ ∫ − ( 9 1 R n − 1 ) 2 s 0 ∥ f ( ⋅ , t ) ∥ L ∞ ( B 9 8 R n − 1 ) d t ⎭ ⎬ ⎫ =: ω n . (47)
Thus, by the definition of ω n \omega_n ω n we obtain the recursive estimate
ω n = max { ( 1 − 1 4 η ) ω n − 1 , C ∫ − ( 1 9 R n − 1 ) 2 s 0 ∥ f ( ⋅ , t ) ∥ L ∞ ( B 8 9 R n − 1 ) d t } ≤ max { ( 1 − 1 4 η ) ω n − 1 , C ∫ − ( 1 9 R ) 2 s 0 ∥ f ( ⋅ , t ) ∥ L ∞ ( B 8 9 R ) d t } . \begin{aligned}
\omega_n=\max\left\{\left(1-\frac{1}{4}\eta\right)\omega_{n-1},C\int_{-\left(\frac{1}{9}R_{n-1}\right)^{2s}}^0\|f(\cdot,t)\|_{L^\infty\left(B_{\frac{8}{9}R_{n-1}}\right)}\,dt\right\} \\
&\le\max\left\{\left(1-\frac{1}{4}\eta\right)\omega_{n-1}, C\int_{-\left(\frac{1}{9}R\right)^{2s}}^{0}\|f(\cdot,t)\|_{L^\infty(B_{\frac{8}{9}R})}\,\mathrm{d}t\right\}.
\end{aligned} ω n = max ⎩ ⎨ ⎧ ( 1 − 4 1 η ) ω n − 1 , C ∫ − ( 9 1 R n − 1 ) 2 s 0 ∥ f ( ⋅ , t ) ∥ L ∞ ( B 9 8 R n − 1 ) d t ⎭ ⎬ ⎫ ≤ max { ( 1 − 4 1 η ) ω n − 1 , C ∫ − ( 9 1 R ) 2 s 0 ∥ f ( ⋅ , t ) ∥ L ∞ ( B 9 8 R ) d t } .
Iterating the above inequality, while considering the estimate in (47) yields
ess osc Q r v ≤ ω n ≤ ( 1 − 1 4 η ) n ω + C ∫ − R 2 s 0 ∥ f ( ⋅ , t ) ∥ L ∞ ( B 8 9 R ) d t . (48) \underset{Q_r}{\operatorname{ess\,osc}} v \le\omega_n \le\left(1-\frac{1}{4}\eta\right)^n\omega+ C\int_{-R^{2s}}^{0}\|f(\cdot,t)\|_{L^\infty(B_{\frac{8}{9}R})}\,\mathrm{d}t. \tag*{(48)} Q r ess osc v ≤ ω n ≤ ( 1 − 4 1 η ) n ω + C ∫ − R 2 s 0 ∥ f ( ⋅ , t ) ∥ L ∞ ( B 9 8 R ) d t . (48)
Since { R n } n ∈ N 0 \{R_n\}_{n\in\mathbb{N}_0} { R n } n ∈ N 0 partitions ( 0 , R ) (0,R) ( 0 , R ) , for r ∈ ( 0 , R ) r\in(0,R) r ∈ ( 0 , R ) , there exists n ∈ N 0 n\in\mathbb{N}_0 n ∈ N 0 such that
R n + 1 ≤ r < R n . (49) R_{n+1}\le r<R_n. \tag*{(49)} R n + 1 ≤ r < R n . (49)
The second inequality in (49) implies that
Q r ⊂ Q n . Q_r\subset Q_n. Q r ⊂ Q n .
Using the definition of R n R_n R n and the first inequality in (49) we obtain that
r ≥ R n + 1 = κ n + 1 R , r\ge R_{n+1}=\kappa^{n+1}R, r ≥ R n + 1 = κ n + 1 R ,
which gives us that
r R ≥ κ n + 1 ⟹ ( 1 − 1 4 η ) n + 1 ≤ ( r R ) β , (50) \frac{r}{R}\ge\kappa^{n+1}\quad\Longrightarrow\quad\left(1-\frac{1}{4}\eta\right)^{n+1}\le\left(\frac{r}{R}\right)^\beta, \tag*{(50)} R r ≥ κ n + 1 ⟹ ( 1 − 4 1 η ) n + 1 ≤ ( R r ) β , (50)
where
β = ln ( 1 − 1 4 η ) ln κ . \beta=\frac{\ln\left(1-\frac{1}{4}\eta\right)}{\ln\kappa}. β = ln κ ln ( 1 − 4 1 η ) .
By inserting (50) into (48) we obtain for any r ∈ ( 0 , R ) r\in(0,R) r ∈ ( 0 , R ) that
ess osc Q r v ≤ 2 ω ( r R ) β + C ∫ − R 2 s 0 ∥ f ( ⋅ , t ) ∥ L ∞ ( B 8 9 R ) d t . (51) \underset{Q_r}{\operatorname{ess\,osc}} v\le2\omega\left(\frac{r}{R}\right)^\beta+C\int_{-R^{2s}}^{0}\|f(\cdot,t)\|_{L^\infty(B_{\frac{8}{9}R})}\,\mathrm{d}t. \tag*{(51)} Q r ess osc v ≤ 2 ω ( R r ) β + C ∫ − R 2 s 0 ∥ f ( ⋅ , t ) ∥ L ∞ ( B 9 8 R ) d t . (51)
Now, we perform an interpolation argument. Indeed, we may assume that (51) holds true, instead of R R R , for some R ‾ ∈ ( r , R ) \overline{R}\in(r,R) R ∈ ( r , R ) , and then setting R ‾ = ( r R ) 1 2 \overline{R}=(rR)^{\frac{1}{2}} R = ( r R ) 2 1 to get that
ess osc Q r v ≤ 2 ω ( r R ) β 2 + C ∫ − ( r R ) s 0 ∥ f ( ⋅ , t ) ∥ L ∞ ( B 8 9 R ) d t . (52) \underset{Q_r}{\operatorname{ess\,osc}} v\le2\omega\left(\frac{r}{R}\right)^{\frac{\beta}{2}}+C\int_{-(rR)^{s}}^{0}\|f(\cdot,t)\|_{L^\infty(B_{\frac{8}{9}R})}\,\mathrm{d}t. \tag*{(52)} Q r ess osc v ≤ 2 ω ( R r ) 2 β + C ∫ − ( r R ) s 0 ∥ f ( ⋅ , t ) ∥ L ∞ ( B 9 8 R ) d t . (52)
This conclusion is reached by assuming ω \omega ω satisfies (40) . Now, we choose ω \omega ω to satisfy (40) . In fact, using (28) with y = x + h y=x+h y = x + h , while dividing by ∣ h ∣ |h| ∣ h ∣ , as well as taking the supremum over Q 2 R Q_{2R} Q 2 R , and then [16 Theorem 1.6] with b = 0 b=0 b = 0 , we observe that
ess sup Q 2 R ∣ v ∣ ≤ C R [ ∥ u ∥ L ∞ ( Q 16 R ) + Tail ( u ; Q 16 R ) ] ≤ C R ∫ − ( 64 R ) 2 s 0 ∫ R N ∣ u ( x , t ) ∣ ( R + ∣ x ∣ ) N + 2 s d x d t . \begin{aligned}
\underset{Q_{2R}}{\operatorname{ess\,sup}} |v|\le\frac{C}{R}\left[\|u\|_{L^\infty(Q_{16R})}+\operatorname{Tail}(u;Q_{16R})\right] \\
&\le\frac{C}{R}\int_{-(64R)^{2s}}^{0}\int_{\mathbb{R}^N}\frac{|u(x,t)|}{(R+|x|)^{N+2s}}\,\mathrm{d}x\mathrm{d}t.
\end{aligned} Q 2 R ess sup ∣ v ∣ ≤ R C [ ∥ u ∥ L ∞ ( Q 16 R ) + Tail ( u ; Q 16 R ) ] ≤ R C ∫ − ( 64 R ) 2 s 0 ∫ R N ( R + ∣ x ∣ ) N + 2 s ∣ u ( x , t ) ∣ d x d t .
Hence, to fulfill (40) we only need to pick
ω = 2 C R ∫ − ( 64 R ) 2 s 0 ∫ R N ∣ u ( x , t ) ∣ ( R + ∣ x ∣ ) N + 2 s d x d t , \omega=\frac{2C}{R}\int_{-(64R)^{2s}}^{0}\int_{\mathbb{R}^N}\frac{|u(x,t)|}{(R+|x|)^{N+2s}}\,\mathrm{d}x\mathrm{d}t, ω = R 2 C ∫ − ( 64 R ) 2 s 0 ∫ R N ( R + ∣ x ∣ ) N + 2 s ∣ u ( x , t ) ∣ d x d t ,
and then (52) holds true. Finally, we substitute the above ω \omega ω in (52) , redefine β \beta β to be β 2 \frac{\beta}{2} 2 β , properly enlarge C C C , and conclude the proof of Proposition 7.1 with the ω \omega ω in (39) .
Proof of Theorem 1.1
We conclude similar as in [10 ] , Proof of Theorem 1.6. First, choose h = h e i h=h e_i h = h e i , where h ∈ ( 0 , 1 32 R ) h\in\left(0,\frac{1}{32}R\right) h ∈ ( 0 , 32 1 R ) and { e i : i = 1 , 2 , … , N } \{e_i:i=1,2,\ldots,N\} { e i : i = 1 , 2 , … , N } are canonical unit vectors in R N \mathbb{R}^N R N . By Proposition 7.1 we obtain for r ∈ ( 0 , R ) r\in(0,R) r ∈ ( 0 , R ) small enough that
ess osc Q r ( x o , t o ) v h ≤ C [ ω ( r R ) β + ∫ t o − ( r R ) 2 s t o ∥ f ( ⋅ , t ) ∥ L ∞ ( B 8 R ( x o ) ) d t ] . \operatorname*{ess\,osc}_{Q_r(x_o,t_o)} v_h \le C\left[\omega\left(\frac{r}{R}\right)^\beta+\int_{t_o-(rR)^{2s}}^{t_o}\|f(\cdot,t)\|_{L^\infty(B_{8R}(x_o))}\,\mathrm{d}t\right]. Q r ( x o , t o ) ess osc v h ≤ C [ ω ( R r ) β + ∫ t o − ( r R ) 2 s t o ∥ f ( ⋅ , t ) ∥ L ∞ ( B 8 R ( x o )) d t ] .
for some constant C = C ( d a t a ) C=C(\mathrm{data}) C = C ( data ) . Next, by applying Lemma 3.6 for the integrand of the last integral, followed by [16 ] , Theorem 1.6 with b = 0 b=0 b = 0 , we deduce
ess osc Q r ( x o , t o ) v h ≤ C [ ω ( r R ) β + 1 R 2 s + 1 ∫ t o − ( r R ) 2 s t o T ( t ; B 2 R ( x o ) ) d t ] ≤ C [ ω ( r R ) β + 1 R ∫ t o − ( r R ) 2 s t o ∫ R N ∣ u ( x , t ) ∣ ( R + ∣ x − x o ∣ ) N + 2 s d x d t ] . (53) \begin{aligned}
\operatorname*{ess\,osc}_{Q_r(x_o,t_o)}v_h
&\le C\left[\omega\left(\frac{r}{R}\right)^\beta+\frac{1}{R^{2s+1}}\int_{t_o-(rR)^{2s}}^{t_o}\mathcal{T}(t;B_{2R}(x_o))\,\mathrm{d}t\right] \\
&\le C\left[\omega\left(\frac{r}{R}\right)^\beta+\frac{1}{R}\int_{t_o-(rR)^{2s}}^{t_o}\int_{\mathbb{R}^N}\frac{|u(x,t)|}{(R+|x-x_o|)^{N+2s}}\,\mathrm{d}x\,\mathrm{d}t\right]. \tag*{(53)}
\end{aligned} Q r ( x o , t o ) ess osc v h ≤ C [ ω ( R r ) β + R 2 s + 1 1 ∫ t o − ( r R ) 2 s t o T ( t ; B 2 R ( x o )) d t ] ≤ C [ ω ( R r ) β + R 1 ∫ t o − ( r R ) 2 s t o ∫ R N ( R + ∣ x − x o ∣ ) N + 2 s ∣ u ( x , t ) ∣ d x d t ] . (53)
Note that the right-hand side does not depend on h h h anymore. Now, consider the sequence { v h n } n ∈ N \{v_{h_n}\}_{n\in\mathbb{N}} { v h n } n ∈ N , where h n → 0 h_n\to0 h n → 0 as n → ∞ n\to\infty n → ∞ , satisfying (53) . First of all, notice that this sequence is uniformly bounded, since for any z ∈ Q 1 2 r ( x o , t o ) : = B 1 2 r ( x o ) × [ t o − ( 1 2 r ) 2 s , t o ] z\in Q_{\frac12 r}(x_o,t_o):=B_{\frac12 r}(x_o)\times[t_o-(\frac12r)^{2s},t_o] z ∈ Q 2 1 r ( x o , t o ) := B 2 1 r ( x o ) × [ t o − ( 2 1 r ) 2 s , t o ] we have
∣ v h n ( z ) ∣ ≤ ∥ D u ∥ L ∞ ( Q 2 R ) ≤ C R [ ∥ u ∥ L ∞ ( Q 16 R ) + Tail ( u ; Q 16 R ) ] . (54) |v_{h_n}(z)|\le\|Du\|_{L^\infty(Q_{2R})}\le\frac{C}{R}\left[\|u\|_{L^\infty(Q_{16R})}+\operatorname{Tail}(u;Q_{16R})\right]. \tag*{(54)} ∣ v h n ( z ) ∣ ≤ ∥ D u ∥ L ∞ ( Q 2 R ) ≤ R C [ ∥ u ∥ L ∞ ( Q 16 R ) + Tail ( u ; Q 16 R ) ] . (54)
where for the last inequality we used Proposition 5.3 . Observe that the right-hand side is independent of z z z and h n h_n h n . Moreover, the sequence { v h n } n ∈ N \{v_{h_n}\}_{n\in\mathbb{N}} { v h n } n ∈ N is equicontinuous, since for any z 1 , z 2 ∈ Q 1 2 r ( x o , t o ) z_1,z_2\in Q_{\frac12 r}(x_o,t_o) z 1 , z 2 ∈ Q 2 1 r ( x o , t o ) we have
∣ v h n ( z 1 ) − v h n ( z 2 ) ∣ ≤ ess osc Q r ( x o , t o ) v h n ≤ C [ ω ( r R ) β + 1 R ∫ t o − ( r R ) 2 s t o ∫ R N ∣ u ( x , t ) ∣ ( R + ∣ x − x o ∣ ) N + 2 s d x d t ] . (55) \begin{aligned}
|v_{h_n}(z_1)-v_{h_n}(z_2)|\le\operatorname*{ess\,osc}_{Q_r(x_o,t_o)}v_{h_n} \\
&\le C\left[\omega\left(\frac{r}{R}\right)^\beta+\frac{1}{R}\int_{t_o-(rR)^{2s}}^{t_o}\int_{\mathbb{R}^N}\frac{|u(x,t)|}{(R+|x-x_o|)^{N+2s}}\,\mathrm{d}x\,\mathrm{d}t\right]. \tag*{(55)}
\end{aligned} ∣ v h n ( z 1 ) − v h n ( z 2 ) ∣ ≤ Q r ( x o , t o ) ess osc v h n ≤ C [ ω ( R r ) β + R 1 ∫ t o − ( r R ) 2 s t o ∫ R N ( R + ∣ x − x o ∣ ) N + 2 s ∣ u ( x , t ) ∣ d x d t ] . (55)
due to (53) . Combining (7.16) and (7.17) enables us to apply Arzelà–Ascoli’s theorem in order to obtain a subsequence { v h n k } k ∈ N \{v_{h_{n_k}}\}_{k\in\mathbb{N}} { v h n k } k ∈ N , which converges uniformly to D i u D_i u D i u . Then, taking the limit h n k → 0 h_{n_k}\to0 h n k → 0 in (7.15) on the smaller domain Q 1 2 r Q_{\frac12 r} Q 2 1 r gives us the estimate
ess osc Q 1 2 r ( x o , t o ) D i u ≤ C [ ω ( r R ) β + 1 R ∫ t o − ( r R ) 2 s t o ∫ R N ∣ u ( x , t ) ∣ ( R + ∣ x − x o ∣ ) N + 2 s d x d t ] . \operatorname*{ess\,osc}_{Q_{\frac12 r}(x_o,t_o)}D_i u\le C\left[\omega\left(\frac{r}{R}\right)^\beta+\frac{1}{R}\int_{t_o-(rR)^{2s}}^{t_o}\int_{\mathbb{R}^N}\frac{|u(x,t)|}{(R+|x-x_o|)^{N+2s}}\,\mathrm{d}x\,\mathrm{d}t\right]. Q 2 1 r ( x o , t o ) ess osc D i u ≤ C [ ω ( R r ) β + R 1 ∫ t o − ( r R ) 2 s t o ∫ R N ( R + ∣ x − x o ∣ ) N + 2 s ∣ u ( x , t ) ∣ d x d t ] .
for some constant C = C ( d a t a ) C=C(\mathrm{data}) C = C ( data ) . A scaling and covering argument completes the proof with ω \omega ω as in (5) .
Data Availability. Data sharing not applicable to this article as no datasets were generated or analyzed during the current study.
Conflict of Interest. The author declares no conflict of interest.
References
VERENA BÖGELEIN, FACHBEREICH MATHEMATIK, UNIVERSITÄT SALZBURG, HELLBRUNNER STR. 34, 5020 SALZBURG, AUSTRIA
Email address: verena.boegelein@plus.ac.at
NAIAN LIAO, FACHBEREICH MATHEMATIK, UNIVERSITÄT SALZBURG, HELLBRUNNER STR. 34, 5020 SALZBURG, AUSTRIA
Email address: naian.liao@plus.ac.at
CALVIN STANKO, FACHBEREICH MATHEMATIK, UNIVERSITÄT SALZBURG, HELLBRUNNER STR. 34, 5020 SALZBURG, AUSTRIA
Email address: calvin.stanko@plus.ac.at