Preprint
Article

This version is not peer-reviewed.

Optimal Liouville Rigidity for Asuperlinear Fractional Parabolic Equation

Submitted:

20 October 2026

Posted:

21 October 2026

You are already at the latest version

Abstract
Let \( n\in\mathbb N \), 0 < \( s \) < 1, \( n>2s \) and 1 < \( p \) < \( \frac{n+2s}{n-2s} \). We prove that every nonnegative classical entire solution of \( u_t+(-\Delta)^su=u^p\qquad\text{in }\mathbb{R}^n\times\mathbb{R} \) is identically zero. Our result extends the optimal Liouville theorem in [42] for the classical heat equation (\( s=1 \)) to the fractional cases 0 < \( s \) < 1 for the full range 1 < \( p \) < \( \frac{n+2s}{n-2s} \) without global boundedness, higher regularity or pointwise decay assumptions.
Keywords: 
;  ;  

MSC:  Primary: 35K58; 35B53; Secondary; 35B44; 35R11

1. Introduction

1.1. Background and Setting of Our Problem

In this paper, we consider the following fractional parabolic equation
u t + ( − Δ ) s u = u p in R n × R ,
where n ∈ N , 0 < s < 1 , n > 2 s and p > 1 . For each fixed t ∈ R , the fractional Laplacian with respect to x is defined by
( − Δ ) s u ( x , t ) : = c n , s P.V. ∫ R n u ( x , t ) − u ( y , t ) | x − y | n + 2 s d y = c n , s lim ε ↓ 0 ∫ R n ∖ B ε ( x ) u ( x , t ) − u ( y , t ) | x − y | n + 2 s d y ,
where P.V. denotes the Cauchy principal value, and c n , s : = 2 2 s Γ n + 2 s 2 π n 2 | Γ ( − s ) | . Define
L 2 s ( R n ) : = f ∈ L loc 1 ( R n ) : ∫ R n | f ( x ) | 1 + | x | n + 2 s d x < ∞ .
For each fixed t ∈ R , ( − Δ ) s u ( x , t ) is well-defined (see, e.g., [35]), if
u ( · , t ) ∈ C l o c 1 , 1 ( R n ) ∩ L 2 s ( R n ) for every t ∈ R .
Definition 1.1. 
A function u : R n × R → [ 0 , ∞ ) is called a nonnegative classical entire solution of (1) if it satisfies (2), has the regularity
u , u t ∈ C ( R n × R ) , sup | t | ≤ T ∥ u ( · , t ) ∥ C 1 , 1 ( B R ¯ ) < ∞ for all R , T > 0 ,
and solves (1) pointwise on R n × R .
Equation (1) is invariant under the parabolic scaling
u λ ( x , t ) : = λ q u ( λ x , λ 2 s t ) , λ > 0 ,
where
α : = 1 p − 1 , q : = 2 s p − 1 = 2 s α .
The relevant Fujita and Sobolev exponents are
p F ( n , s ) : = 1 + 2 s n , p S ( n , s ) : = n + 2 s n − 2 s .
Let μ : = 2 α + 1 − n 2 s , then
p < p S ( n , s ) ⟺ μ > 0 .
Our aim is to prove that every nonnegative classical entire solution u in the sense of Definition 1.1 satisfies u ≡ 0 throughout the range 1 < p < p S ( n , s ) .
If u is independent of time, (1) reduces to the fractional Lane–Emden equation ( − Δ ) s u = u p in R n . Its Liouville theory builds on classical symmetry and classification results [2,28,29,30,36,37,48], and on extension, integral-equation and direct methods in the fractional setting [3,9,10,11]. In particular, the direct moving-sphere theorem of Li, Xu, Yang, and Zhuo [35] excludes nontrivial nonnegative solutions for 1 < p < p S ( n , s ) and provides the elliptic rigidity used in our proof. Related results on finite Morse index, higher-order and Hénon–Hardy problems, Hartree equations and elliptic systems appear in [4,5,13,14,15,16,17,18,19,20]. See also [7,33,40,46] for regularity, compactness and a priori estimates. [6,7,8]
For time-dependent solutions, the classical theory provides testing, rescaling and energy methods [27,31,32,34,39,44], with extensions beyond scale-invariant nonlinearities in [45,47]. Poláčik, Quittner, and Souplet [41] developed rescaling arguments linking Liouville theorems to universal estimates and removing the boundedness assumption. Quittner [42] proved nonexistence of positive classical entire solutions throughout the Sobolev-subcritical range without global boundedness or spatial decay. His refined energy method is the starting point for our argument. For s = 1 2 , under some assumptions, Deng, Sire, Wei and Wu [21] derived a monotonicity formula and classified self-similar solutions for n ≤ 4 and a restricted subcritical exponent range. Further Liouville, symmetry and monotonicity results for fractional parabolic equations can be found in [8,24,49], and results on blow-up, regularity and a priori estimates in [6,7,25,26,43].

1.2. Our Main Result

Our main result is the following Liouville theorem.
Theorem 1.2. 
Let n ∈ N , 0 < s < 1 , n > 2 s , and 1 < p < p S ( n , s ) . Suppose that u is a nonnegative classical entire solution of (1) in the sense of Definition 1.1. Then
u ( x , t ) = 0 for every ( x , t ) ∈ R n × R .
Remark 1.3. 
Theorem 1.2 extends the optimal Liouville theorem in [42] for the classical heat equation ( s = 1 ) to the fractional case 0 < s < 1 .
The restriction p < p S ( n , s ) in Theorem 1.2 is sharp. Actually, if p = p S ( n , s ) , the fractional Sobolev bubbles
U λ , x 0 ( x ) = κ n , s λ λ 2 + | x − x 0 | 2 n − 2 s 2 , λ > 0 , x 0 ∈ R n , κ n , s > 0
satisfy ( − Δ ) s U λ , x 0 = U λ , x 0 p S ( n , s ) , see [11]. Since U λ , x 0 ∈ L 2 s ( R n ) , then U λ , x 0 is a nonzero stationary entire solution of (1) for p = p S ( n , s ) .

1.3. Main Difficulties and Innovations in Our Proof

Sketch of our proof. First, we construct the localized energy and corresponding commutator terms for bounded solutions. Then, we establish the estimates of energy and commutator terms through local Sobolev estimates and the Morrey bound, without requiring an explicit formula for the fractional heat kernel. Next, by establishing the level-set estimates, we adapt the energy iteration approach of [42] to our nonlocal setting, to prove that bounded solution is independent of time. Thus elliptic Liouville theorem implies that every bounded solution is identically zero. Finally, by a doubling and rescaling argument, we remove the boundedness assumption.
We would like to mention some of the main difficulties and innovations in our proof.

1.3.1. Main Difficulties in Our Proof

We have three main difficulties as follows.
(i) The lack of global boundedness or asymptotic conditions as | x | → ∞ . Definition 1.1 and Theorem 1.2 do not assume global boundedness of solution or asymptotic conditions as | x | → ∞ . Consequently, the powerful tools such as the method of moving planes and the sliding method are not applicable in our setting.
(ii) The lack of explicit heat-kernel formulae. The rescaling and energy arguments in [42] for s = 1 and in [21] for s = 1 2 rely on explicit Gaussian or Poisson kernel formulae. Unfortunately, we do not have explicit formulae for heat-kernel for general s ∈ ( 0 , 1 ) and s ≠ 1 2 . Moreover, the space-time testing argument in [24] is applicable only to the Fujita range 1 < p ≤ p F ( n , s ) . Thus the difficulty is to refine and apply the rescaling and energy arguments for the full range 1 < p < p S ( n , s ) , without relying on explicit heat-kernel formulae.
(iii) The nonlocality of the fractional Laplacian. The nonlocality of the fractional Laplacian ( − Δ ) s brings essential difficulties to the cutoff method and compactness argument throughout our proof.

1.3.2. Main Novelty and Key Ingredients in Our Proof

In order to overcome the above difficulties, we refine the rescaling and energy strategy of [21,42] and introduce the following three key innovations.
(i) Localized energy estimates for Z [ w ] . Under the change of variables (52), we have Z [ w ] : = w σ + D w = r α + 1 u t . Our aim is to prove Z [ w ] = 0 , which yields u t = 0 and allows us to apply the elliptic Liouville theorem in [9], see also [4,11,18]. To estimate Z [ w ] , we first construct the localized energy and the associated commutator (see (65)):
F η ( σ ) [ w ] : = 1 2 E s ( η w ) − 1 p + 1 ∫ R n η 2 w p + 1 d y , C η ( σ ) [ w ] : = ( − Δ ) s ( η w ) − η ( − Δ ) s w .
Then we establish an identity containing Z [ w ] , F η ( σ ) [ w ] and C η ( σ ) [ w ] , see (66) in Proposition 4.2. Thanks to this, we obtain that the local integrals of | Z [ w ] | 2 can be controlled by F η ( σ ) [ w ] and C η ( σ ) [ w ] , without requiring an explicit formula for the heat kernel. The estimates for Z [ w ] , F η ( σ ) [ w ] and C η ( σ ) [ w ] are established in Lemmas 4.3, 4.5 and 4.6, which provide the energy bounds needed for the subsequent energy iteration.
(ii) Energy improvement by iteration. The key point of our argument is the quantitative relation between large values of w and the local integral of | Z [ w ] | 2 , established in Lemmas 5.2-5.4. This relation provides the weak- L 2 p estimate (131) in Proposition 5.6. These estimates yield the following strict improvement in energy exponent:
γ ⟼ κ * γ , where κ * : = p + 1 2 p < 1 ,
see Proposition 5.7. These estimates ensure that the energy exponent decreases at each iteration. For every p F < p < p S , we make finitely many iterations until the energy exponent is less than μ . Then, rescaling back to the original variables implies that u t = 0 .
(iii) Morrey estimates and control of nonlocal remainder terms. Our treatment of nonlocality uses a Morrey estimate to control both localization errors and rescaling limits. Testing against the first Dirichlet eigenfunction yields the following Morrey estimate:
sup t ∈ R M q [ u ( · , t ) ] ≤ K ( n , s , p ) ,
see Proposition 3.5.
For the localized energy estimates, we split w into a localized part and a part vanishing near supp η . The Morrey bound (4) controls the commutator of the part vanishing near supp η through Lemma 2.3, while local Sobolev estimates control the commutator of the localized part, see (75) and (78). For the rescaling limits, the invariance of (4) provides uniform control on the nonlocal remainder terms, even though the rescaled solutions are uniformly bounded only on expanding cylinders. Together with local uniform convergence, this justifies passage to the limit in the integral in (6) by Lemma 2.1. Consequently, we can derive the compactness results in Lemmas 3.7 and 5.1, which enable the removal of the boundedness assumption in Section 6.
The rest of this paper is organized as follows. Section 2 collects preliminaries, which are useful in subsequent sections. Section 3 treats the Fujita range and establishes the Morrey estimate and compactness result. Section 4 derives the localized energy estimates and an initial bound for Z [ w ] . Section 5 proves defect lower bounds and a weak- L 2 p estimate, then improves the local bound for Z [ w ] . Finally, Section 6 proves the bounded Liouville theorem by iteration and then removes the boundedness assumption to complete the proof of Theorem 1.2.

2. Preliminaries

In this section, we collect several auxiliary results needed in the sequel. These include a convergence result for nonlocal integrals, a pointwise representation and decay estimates for the commutator, and Liouville-type theorems for bounded nonnegative weak solutions of fractional parabolic and elliptic equations.

2.1. Convergence Result and Decay Estimates for Commutator

For 0 < q ≤ n and f ≥ 0 , set
M q [ f ] : = sup a ∈ R n , R > 0 R q − n ∫ B R ( a ) f ( x ) d x .
The following lemma shows that local uniform convergence and a uniform Morrey bound allow passage to the limit in nonlocal integrals, which is useful in subsequent compactness arguments.
Lemma 2.1. 
Fix 0 < q ≤ n and a bounded open interval I ⊂ R . Assume that f , f j ≥ 0 satisfy
sup j ≥ 1 sup t ∈ I M q [ f j ( · , t ) ] ≤ K , f j → f locally uniformly on R n × I .
Then for every ψ ∈ C c ∞ ( R n × I ) , there exists a positive constant C = C ( ψ , | I | , n , s , q ) , such that
sup t ∈ I M q [ f ( · , t ) ] ≤ K , ∫ I ∫ R n f ( x , t ) | ( − Δ ) s ψ ( x , t ) | d x d t ≤ C K ,
and
∫ I ∫ R n f j ( x , t ) ( − Δ ) s ψ ( x , t ) d x d t ⟶ ∫ I ∫ R n f ( x , t ) ( − Δ ) s ψ ( x , t ) d x d t .
Proof. 
For every a ∈ R n , R > 0 , and t ∈ I , by the locally uniform convergence of f j , we have
R q − n ∫ B R ( a ) f ( x , t ) d x = lim j → ∞ R q − n ∫ B R ( a ) f j ( x , t ) d x ≤ K .
This yields that sup t ∈ I M q [ f ( · , t ) ] ≤ K . Fix ψ ∈ C c ∞ ( R n × I ) , choose R ψ ≥ 1 and a compact interval J ⋐ I such that supp ψ ⊂ B R ψ × J . Then ( − Δ ) s ψ ( · , t ) = 0 for t ∉ J . For | x | > 2 R ψ and t ∈ J , we have ψ ( x , t ) = 0 and | x − y | ≥ | x | 2 whenever y ∈ B R ψ . Thus for any | x | > 2 R ψ and t ∈ J , we have
( − Δ ) s ψ ( x , t ) = − c n , s ∫ B R ψ ψ ( y , t ) | x − y | n + 2 s d y , | ( − Δ ) s ψ ( x , t ) | ≤ C ψ | x | − n − 2 s ,
where C ψ is independent of t. On the annuli A m : = B 2 m + 1 R ψ ∖ B 2 m R ψ , m ≥ 1 , by (7), we have
∫ A m f ( x , t ) | ( − Δ ) s ψ ( x , t ) | d x ≤ C ψ ( 2 m R ψ ) − n − 2 s ∫ B 2 m + 1 R ψ f ( x , t ) d x ≤ C ψ K R ψ − q − 2 s 2 − m ( q + 2 s ) .
Since ψ is smooth, we may assume | ( − Δ ) s ψ | ≤ C ψ on B 2 R ψ × J . Then by (8), we obtain, uniformly for t ∈ J ,
∫ R n f ( x , t ) | ( − Δ ) s ψ ( x , t ) | d x ≤ C ψ K ( 2 R ψ ) n − q + C ψ K R ψ − q − 2 s ∑ m = 1 ∞ 2 − m ( q + 2 s ) ≤ C ψ , n , s , q K .
Since ( − Δ ) s ψ ( · , t ) = 0 for t ∉ J , integrating over J and using | J | ≤ | I | , we obtain (5).
To prove (6), fix A > 2 R ψ . By the locally uniform convergence of f j on B A ¯ × J , we have
∫ J ∫ B A ( f j − f ) ( x , t ) ( − Δ ) s ψ ( x , t ) d x d t ≤ sup B A ¯ × J | f j − f | ∫ J ∫ B A | ( − Δ ) s ψ ( x , t ) | d x d t ⟶ 0 .
Since sup j ≥ 1 sup t ∈ I M q [ f j ( · , t ) ] ≤ K , repeating the estimate (8) for f j on B 2 m + 1 A ∖ B 2 m A , m ≥ 0 , we have
∫ J ∫ { | x | > A } | ( f j − f ) ( x , t ) | | ( − Δ ) s ψ ( x , t ) | d x d t ≤ C K A − q − 2 s ,
where C is independent of j and A. Therefore, by using ( − Δ ) s ψ = 0 for t ∉ J , we have
lim sup j → ∞ ∫ I ∫ R n ( f j − f ) ( x , t ) ( − Δ ) s ψ ( x , t ) d x d t ≤ C K A − q − 2 s .
Letting A → ∞ proves (6). □
For h , k ∈ H s ( R n ) , define
E s ( h , k ) : = c n , s 2 ∫ ∫ R n × R n ( h ( x ) − h ( y ) ) ( k ( x ) − k ( y ) ) | x − y | n + 2 s d x d y , E s ( h ) : = E s ( h , h ) = ∥ ( − Δ ) s 2 h ∥ 2 2 .
For ε > 0 and measurable functions h , k , define the truncated form of E s as follows
E s , ε ( h , k ) : = c n , s 2 ∫ ∫ { | x − y | > ε } ( h ( x ) − h ( y ) ) ( k ( x ) − k ( y ) ) | x − y | n + 2 s d x d y , E s , ε ( h ) : = E s , ε ( h , h ) ,
whenever the integral is absolutely convergent. For f ∈ L 2 s ( R n ) and χ ∈ C c ∞ ( R n ) , define the commutator [ ( − Δ ) s , χ ] f : = ( − Δ ) s ( χ f ) − χ ( − Δ ) s f in the sense of distributions by
[ ( − Δ ) s , χ ] f , φ : = ∫ R n f ( x ) χ ( x ) ( − Δ ) s φ ( x ) − ( − Δ ) s ( χ φ ) ( x ) d x , φ ∈ C c ∞ ( R n ) .
Actually, similar to (7), for every φ ∈ C c ∞ ( R n ) , | χ ( − Δ ) s φ − ( − Δ ) s ( χ φ ) | ≤ C χ , φ ( 1 + | x | ) − n − 2 s . Then it follows from f ∈ L 2 s ( R n ) that [ ( − Δ ) s , χ ] f is well defined.
Lemma 2.2 provides a pointwise formula for the commutator [ ( − Δ ) s , χ ] f , which is used to prove the decay estimates for commutator in Lemmas 2.3 and 2.4.
Lemma 2.2. 
Let χ ∈ C c ∞ ( R n ) and 0 ≤ f ∈ L 2 s ( R n ) . 
If f ∈ L loc 2 ( R n ) such that χ f ∈ H s ( R n ) , then E s ( f , χ 2 f ) : = lim ε ↓ 0 E s , ε ( f , χ 2 f ) exists and satisfies
E s ( f , χ 2 f ) = E s ( χ f ) − R χ [ f ] ,
where
R χ [ f ] : = c n , s 2 ∫ ∫ R n × R n f ( x ) f ( y ) ( χ ( x ) − χ ( y ) ) 2 | x − y | n + 2 s d x d y .
If f is C 1 , 1 in a neighborhood of x, then [ ( − Δ ) s , χ ] f has the pointwise representation
[ ( − Δ ) s , χ ] f ( x ) : = ( − Δ ) s ( χ f ) ( x ) − χ ( x ) ( − Δ ) s f ( x ) = c n , s P.V. ∫ R n ( χ ( x ) − χ ( y ) ) f ( y ) | x − y | n + 2 s d y .
Proof. 
First assume that 0 ≤ f ∈ L 2 s ( R n ) ∩ L loc 2 ( R n ) such that χ f ∈ H s ( R n ) . Choose a ball B such that dist ( supp χ , R n ∖ B ) ≥ 1 . Since
( χ ( x ) − χ ( y ) ) 2 = 0 for ( x , y ) ∈ ( supp χ ) c × ( supp χ ) c ,
combining with the symmetry integral in (10), we have
0 ≤ R χ [ f ] ≤ c n , s ∫ ∫ supp χ × R n f ( x ) f ( y ) ( χ ( x ) − χ ( y ) ) 2 | x − y | n + 2 s d x d y .
For x , y ∈ B , by | χ ( x ) − χ ( y ) | 2 ≤ C χ min { 1 , | x − y | 2 } , the integrability of min { 1 , | z | 2 } | z | − n − 2 s and Young’s inequality, we deduce
∫ ∫ B × B f ( x ) f ( y ) ( χ ( x ) − χ ( y ) ) 2 | x − y | n + 2 s d x d y ≤ C χ ∥ f ∥ L 2 ( B ) 2 .
For x ∈ supp χ and y ∉ B , since | x − y | ≥ 1 , we have
sup x ∈ supp χ ∫ R n ∖ B f ( y ) | x − y | n + 2 s d y ≤ C B ∥ f ∥ L 2 s .
Thus
∫ ∫ supp χ × ( R n ∖ B ) f ( x ) f ( y ) ( χ ( x ) − χ ( y ) ) 2 | x − y | n + 2 s d x d y ≤ C χ , B ∥ f ∥ L 1 ( B ) ∥ f ∥ L 2 s .
By (12) and (13) we obtain 0 ≤ R χ [ f ] < ∞ . For ε > 0 , define
R χ , ε [ f ] : = c n , s 2 ∫ ∫ | x − y | > ε f ( x ) f ( y ) ( χ ( x ) − χ ( y ) ) 2 | x − y | n + 2 s d x d y .
By the identity
( f ( x ) − f ( y ) ) ( χ ( x ) 2 f ( x ) − χ ( y ) 2 f ( y ) ) = ( χ ( x ) f ( x ) − χ ( y ) f ( y ) ) 2 − f ( x ) f ( y ) ( χ ( x ) − χ ( y ) ) 2 ,
together with χ f ∈ H s ( R n ) and R χ [ f ] < ∞ , we have
E s , ε ( f , χ 2 f ) = E s , ε ( χ f ) − R χ , ε [ f ] .
Then by the monotone convergence theorem, we have
E s , ε ( χ f ) → E s ( χ f ) , R χ , ε [ f ] → R χ [ f ] as ε ↓ 0 .
This yields the existence of E s ( f , χ 2 f ) and thus proves (9). Next we assume that f ∈ L 2 s ( R n ) and f is C 1 , 1 in a neighborhood U of x. For h ∈ L 2 s ( R n ) and ε > 0 , define
A ε h ( x ) : = c n , s ∫ { | x − y | > ε } h ( x ) − h ( y ) | x − y | n + 2 s d y , [ A ε , χ ] h : = A ε ( χ h ) − χ A ε h .
For ψ ∈ C c ∞ ( R n ) , since h ∈ L 2 s ( R n ) , symmetry of the integrand and Fubini’s theorem yield
∫ R n ( A ε h ) ψ d x = ∫ R n h A ε ψ d x .
Applying this identity to ( h , ψ ) = ( f , χ φ ) and ( h , ψ ) = ( χ f , φ ) , respectively, we get
[ A ε , χ ] f , φ = ∫ R n f ( x ) χ ( x ) A ε φ ( x ) − A ε ( χ φ ) ( x ) d x .
For every ψ ∈ C c ∞ ( R n ) , similar to the estimate in (7), we have
A ε ψ ( x ) ⟶ ( − Δ ) s ψ ( x ) , sup 0 < ε ≤ 1 | A ε ψ ( x ) | ≤ C ψ ( 1 + | x | ) − n − 2 s .
Applying this estimate with ψ = φ and ψ = χ φ respectively, then using f ∈ L 2 s ( R n ) and the dominated convergence theorem, we obtain
[ A ε , χ ] f , φ ⟶ ∫ R n f ( x ) χ ( x ) ( − Δ ) s φ ( x ) − ( − Δ ) s ( χ φ ) ( x ) d x = [ ( − Δ ) s , χ ] f , φ .
Hence [ A ε , χ ] f → [ A , χ ] f in the sense of distributions. Since f is C 1 , 1 on a neighborhood U of x and f , χ f ∈ L 2 s ( R n ) , we have
[ A ε , χ ] f → ( − Δ ) s ( χ f ) − χ ( − Δ ) s f locally uniformly on U .
Therefore, we have
[ ( − Δ ) s , χ ] f ( z ) = ( − Δ ) s ( χ f ) ( z ) − χ ( z ) ( − Δ ) s f ( z ) , z ∈ U .
Finally, direct calculation yields
[ A ε , χ ] f ( x ) = c n , s ∫ { | x − y | > ε } ( χ ( x ) − χ ( y ) ) f ( y ) | x − y | n + 2 s d y .
Letting ε → 0 proves (11). □
Lemma 2.3. 
Let f : R n → [ 0 , ∞ ) be measurable and satisfy M q [ f ] ≤ K , where 0 < q ≤ n . Fix χ 0 , χ 1 ∈ C c ∞ ( R n ) with χ 0 ¬ ≡ 0 , 0 ≤ χ 0 , χ 1 ≤ 1 and χ 1 = 1 on a neighborhood of supp χ 0 . For λ > 0 , set
χ i , λ ( x ) : = χ i x λ , i = 0 , 1 , f λ ♯ : = ( 1 − χ 1 , λ ) f .
Then there exists a positive constant C depending only on n , s , q , χ 0 , χ 1 , such that
∥ [ ( − Δ ) s , χ 0 , λ ] f λ ♯ ∥ L ∞ ( supp χ 0 , λ ) ≤ C K λ − q − 2 s .
Proof. 
For x ∈ R n and ρ > 0 , let A m : = B 2 m + 1 ρ ( x ) ∖ B 2 m ρ ( x ) , m ≥ 0 . Since R n ∖ B ρ ( x ) = ⋃ m ≥ 0 A m , by M q [ f ] ≤ K we have
∫ R n ∖ B ρ ( x ) f ( y ) | x − y | n + 2 s d y ≤ ∑ m = 0 ∞ ( 2 m ρ ) − n − 2 s ∫ B 2 m + 1 ρ ( x ) f ( y ) d y ≤ 2 n − q K ρ − q − 2 s ∑ m = 0 ∞ 2 − m ( q + 2 s ) ≤ C K ρ − q − 2 s .
Taking x = 0 and ρ = 1 , and using M q [ f ] ≤ K , we get f ∈ L 2 s ( R n ) . Thus f λ ♯ ∈ L 2 s ( R n ) because 0 ≤ f λ ♯ ≤ f . Let
d * : = dist ( supp χ 0 , supp ( 1 − χ 1 ) ) > 0 .
Then for every λ > 0 , we have
dist supp χ 0 , λ , supp ( 1 − χ 1 , λ ) = λ d * .
Hence f λ ♯ = 0 on a neighborhood of supp χ 0 , λ , and χ 0 , λ f λ ♯ ≡ 0 . Therefore by (11), for x ∈ supp χ 0 , λ , we have
[ ( − Δ ) s , χ 0 , λ ] f λ ♯ ( x ) = c n , s χ 0 , λ ( x ) ∫ R n ∖ B λ d * ( x ) f λ ♯ ( y ) | x − y | n + 2 s d y .
By letting ρ = λ d * in (15), and using 0 ≤ χ 0 , λ ≤ 1 and 0 ≤ f λ ♯ ≤ f , we obtain
[ ( − Δ ) s , χ 0 , λ ] f λ ♯ ( x ) ≤ c n , s ∫ R n ∖ B λ d * ( x ) f ( y ) | x − y | n + 2 s d y ≤ C K ( λ d * ) − q − 2 s ≤ C K λ − q − 2 s ,
where the constant C depends only on n , s , q , χ 0 , χ 1 . Taking the supremum over x ∈ supp χ 0 , λ proves (14). Thus we complete the proof of Lemma 2.3. □
Lemma 2.4. 
Let χ ∈ C c ∞ ( R n ) satisfy χ = 1 on B 1 , and set χ R ( x ) = χ ( x R ) , where R ≥ 2 . Then for every Ψ ∈ W 1 , 1 ( R n ) , there exists a positive constant C = C ( χ , n , s ) , such that
∥ [ ( − Δ ) s , χ R ] Ψ ∥ L 1 ≤ C ∥ Ψ ∥ W 1 , 1 ( 1 + log R ) R − min { 2 s , 1 } .
Proof. 
We first prove the estimate for Ψ ∈ C c ∞ ( R n ) . By (11), we have
[ ( − Δ ) s , χ R ] Ψ ( x ) = Ψ ( x ) ( − Δ ) s χ R ( x ) + c n , s ∫ R n ( χ R ( x ) − χ R ( y ) ) ( Ψ ( y ) − Ψ ( x ) ) | x − y | n + 2 s d y .
By scaling we have ∥ ( − Δ ) s χ R ∥ ∞ = R − 2 s ∥ ( − Δ ) s χ ∥ ∞ , and hence
∥ Ψ ( x ) ( − Δ ) s χ R ( x ) ∥ L 1 ≤ C R − 2 s ∥ Ψ ∥ L 1 .
Let H ( x ) : = c n , s ∫ R n ( χ R ( x ) − χ R ( y ) ) ( Ψ ( y ) − Ψ ( x ) ) | x − y | n + 2 s d y . By | χ R ( x ) − χ R ( x + h ) | ≤ C min { 1 , | h | R } , using translation invariance, the triangle inequality, and Fubini’s theorem, we have
∥ H ∥ L 1 ≤ C ∫ R n min { 1 , | h | R } | h | n + 2 s ∥ Ψ ( · + h ) − Ψ ( · ) ∥ L 1 d h ≤ C ∫ R n min { 1 , | h | R } | h | n + 2 s min { 2 ∥ Ψ ∥ L 1 , | h | ∥ ∇ Ψ ∥ L 1 } d h = C ∫ 0 ∞ min { 1 , r R } r 1 + 2 s min { 2 ∥ Ψ ∥ L 1 , r ∥ ∇ Ψ ∥ L 1 } d r ,
where we have used ∥ Ψ ( · + h ) − Ψ ( · ) ∥ L 1 ≤ min { 2 ∥ Ψ ∥ L 1 , | h | ∥ ∇ Ψ ∥ L 1 } .
For 0 ≤ r ≤ 1 , since R ≥ 2 , we have
∫ 0 1 min { 1 , r R } r 1 + 2 s min { 2 ∥ Ψ ∥ L 1 , r ∥ ∇ Ψ ∥ L 1 } d r ≤ C R − 1 ∥ ∇ Ψ ∥ L 1 ∫ 0 1 r 1 − 2 s d r ≤ C R − 1 ∥ ∇ Ψ ∥ L 1 .
For 1 < r < R , using ∥ Ψ ( · + h ) − Ψ ( · ) ∥ L 1 ≤ 2 ∥ Ψ ∥ L 1 , we have
∫ 1 R min { 1 , r R } r 1 + 2 s min { 2 ∥ Ψ ∥ L 1 , r ∥ ∇ Ψ ∥ L 1 } d r ≤ C ∥ Ψ ∥ L 1 R − 2 s , 0 < s < 1 2 , R − 1 log R , s = 1 2 , R − 1 , 1 2 < s < 1 .
Finally, for r > R , we have
∫ R ∞ min { 1 , r R } r 1 + 2 s min { 2 ∥ Ψ ∥ L 1 , r ∥ ∇ Ψ ∥ L 1 } d r ≤ C ∥ Ψ ∥ 1 ∫ R ∞ r − 1 − 2 s d r ≤ C R − 2 s ∥ Ψ ∥ 1 .
The bounds (18), (19), and (20), together with (17), imply (16). If Ψ ∈ W 1 , 1 ( R n ) , choose Ψ k → Ψ in W 1 , 1 ( R n ) . Applying (16) to Ψ k − Ψ ℓ shows that { [ ( − Δ ) s , χ R ] Ψ k } is a Cauchy sequence in L 1 ( R n ) . Assume that [ ( − Δ ) s , χ R ] Ψ k → H R in L 1 ( R n ) . Then (16) holds for H R . Moreover, for every ϕ ∈ C c ∞ ( R n ) , we have
[ ( − Δ ) s , χ R ] ( Ψ k − Ψ ) , ϕ = ∫ R n ( Ψ k − Ψ ) χ R ( − Δ ) s ϕ − ( − Δ ) s ( χ R ϕ ) d x .
Together with Ψ k → Ψ in W 1 , 1 ( R n ) , we have H R = [ ( − Δ ) s , χ R ] Ψ . Thus (16) holds for every Ψ ∈ W 1 , 1 ( R n ) . □

2.2. The Integral Formula of Bounded Solutions

With the Fourier convention g ^ ( ξ ) : = ∫ R n e − i x · ξ g ( x ) d x , define the heat kernel of v t + ( − Δ ) s v = 0 by
P t ( x ) : = 1 ( 2 π ) n ∫ R n e i x · ξ − t | ξ | 2 s d ξ , t > 0 .
Then P ^ t ( ξ ) = e − t | ξ | 2 s . Set P 0 : = δ 0 , where δ 0 is the Dirac measure at the origin, so P 0 * g = g . For bounded initial data g, the linear heat flow is
e − t ( − Δ ) s g ( x ) = ( P t * g ) ( x ) = ∫ R n P t ( x − y ) g ( y ) d y .
For t > 0 , P t ( x ) > 0 and ∫ R n P t ( x ) d x = 1 .
We have the following integral formula and comparison principle for bounded solutions.
Lemma 2.5. 
Let u ∈ C ( R n × R ) ∩ L ∞ ( R n × R ) be a nonnegative weak solution of (1). Then for every t 0 < t 1 , there holds
u ( x , t 1 ) = ( P t 1 − t 0 * u ( · , t 0 ) ) ( x ) + ∫ t 0 t 1 ( P t 1 − τ * u ( · , τ ) p ) ( x ) d τ , x ∈ R n .
The following comparison principle also holds. Assume that u 1 , u 2 are bounded nonnegative continuous functions on R n × [ t 0 , t 1 ] satisfying (21) on ( t 0 , t 1 ] and u 1 ( · , t 0 ) ≤ u 2 ( · , t 0 ) , then u 1 ( x , t ) ≤ u 2 ( x , t ) for all ( x , t ) ∈ R n × [ t 0 , t 1 ] . In particular, if u ¬ ≡ 0 , then u ( x , t ) > 0 for every ( x , t ) ∈ R n × R . 
Proof. 
Fix t 0 < t 1 and φ ∈ C c ∞ ( R n ) , and set Ψ ( x , t ) : = ( P t 1 − t * φ ) ( x ) , t 0 ≤ t ≤ t 1 . Then
Ψ ( · , t 1 ) = φ , − ∂ t Ψ + ( − Δ ) s Ψ = 0 .
Since, for t 0 ≤ t < t 1 , ∫ R n P t 1 − t ( x ) d x = 1 and ∇ Ψ ( · , t ) = P t 1 − t * ∇ φ , by Young’s inequality, we have
∥ Ψ ( · , t ) ∥ W 1 , 1 ≤ ∥ P t 1 − t ∥ L 1 ∥ φ ∥ W 1 , 1 = ∥ φ ∥ W 1 , 1 .
Together with Ψ ( · , t 1 ) = φ , we deduce
sup t 0 ≤ t ≤ t 1 ∥ Ψ ( · , t ) ∥ W 1 , 1 ≤ ∥ φ ∥ W 1 , 1 .
Choose χ ∈ C c ∞ ( B 2 ) with 0 ≤ χ ≤ 1 and χ = 1 on B 1 , and set χ R ( x ) = χ ( x R ) for R ≥ 2 . Multiplying (1) by χ R Ψ , integrating by parts over R n × [ t 0 , t 1 ] , and using the equation for Ψ , we obtain
∫ R n u ( x , t 1 ) χ R ( x ) φ ( x ) d x − ∫ R n u ( x , t 0 ) χ R ( x ) Ψ ( x , t 0 ) d x = ∫ t 0 t 1 ∫ R n u ( x , t ) p χ R ( x ) Ψ ( x , t ) d x d t − ∫ t 0 t 1 ∫ R n u ( x , t ) [ ( − Δ ) s , χ R ] Ψ ( x , t ) d x d t .
By Lemma 2.4 and (22), we have
∫ t 0 t 1 ∫ R n u [ ( − Δ ) s , χ R ] Ψ d x d t ≤ C ( t 1 − t 0 ) ∥ u ∥ ∞ ∥ φ ∥ W 1 , 1 ( 1 + log R ) R − min { 2 s , 1 } ⟶ 0 .
The other terms in (23) converge by dominated convergence theorem. Actually, the integrals on the left-hand side of (23) are bounded by ∥ u ∥ ∞ ∥ φ ∥ 1 , and
∫ t 0 t 1 ∫ R n | u p Ψ | d x d t ≤ ( t 1 − t 0 ) ∥ u ∥ ∞ p ∥ φ ∥ 1 .
Letting R → ∞ , and then using symmetry of P r , Fubini’s theorem and the definition of Ψ , we obtain
∫ R n u ( x , t 1 ) φ ( x ) d x = ∫ R n ( P t 1 − t 0 * u ( · , t 0 ) ) ( x ) φ ( x ) d x + ∫ t 0 t 1 ∫ R n ( P t 1 − τ * u ( · , τ ) p ) ( x ) φ ( x ) d x d τ .
Thus (21) holds in the sense of distributions in x. The right-hand side of (21) is continuous in x. Indeed, for each r > 0 , by the translation continuity of P r in L 1 and u ∈ L ∞ ( R n × R ) , we deduce that ( P t 1 − t 0 * u ( · , t 0 ) ) ( x ) is continuous. Moreover, note that
∥ P t 1 − τ * u ( · , τ ) p ∥ ∞ ≤ ∥ u ∥ ∞ p , t 0 ≤ τ < t 1 .
By the dominated convergence theorem in τ , we conclude that ∫ t 0 t 1 ( P t 1 − τ * u ( · , τ ) p ) ( x ) d τ is also continuous. Since u ( · , t 1 ) is continuous, (21) holds pointwise.
Next, assume that u 1 , u 2 are bounded nonnegative continuous functions on R n × [ t 0 , t 1 ] satisfying (21) on ( t 0 , t 1 ] and u 1 ( · , t 0 ) ≤ u 2 ( · , t 0 ) , choose M > 0 such that 0 ≤ u 1 , u 2 ≤ M . Since
( u 1 p − u 2 p ) + ≤ p M p − 1 ( u 1 − u 2 ) + ,
we have
( u 1 − u 2 ) + ( x , t ) ≤ ∫ t 0 t P t − τ * ( u 1 p − u 2 p ) + ( · , τ ) ( x ) d τ ≤ p M p − 1 ∫ t 0 t P t − τ * ( u 1 − u 2 ) + ( · , τ ) ( x ) d τ .
Here we have used 0 ≤ u 1 , u 2 ≤ M in the last inequality. Moreover, since P r ≥ 0 and ∫ R n P r d y = 1 ,
P r * ( u 1 − u 2 ) + ( · , τ ) ( x ) = ∫ R n P r ( x − y ) ( u 1 − u 2 ) + ( y , τ ) d y ≤ ∥ ( u 1 − u 2 ) + ( · , τ ) ∥ ∞ ∫ R n P r ( x − y ) d y = ∥ ( u 1 − u 2 ) + ( · , τ ) ∥ ∞ .
This yields
∥ ( u 1 − u 2 ) + ( · , t ) ∥ ∞ ≤ p M p − 1 ∫ t 0 t ∥ ( u 1 − u 2 ) + ( · , τ ) ∥ ∞ d τ , t 0 ≤ t ≤ t 1 .
Then Gronwall’s inequality yields ( u 1 − u 2 ) + ≡ 0 , proving the comparison principle.
Finally, suppose that u ¬ ≡ 0 . If u ( · , τ ) ≡ 0 at some time τ , by applying the comparison principle with u 1 = u and u 2 ≡ 0 , we obtain u ( · , t ) ≡ 0 for every t ≥ τ . For any t 0 < τ , (21) shows that
0 = ( P τ − t 0 * u ( · , t 0 ) ) ( x ) + ∫ t 0 τ ( P τ − t * u ( · , t ) p ) ( x ) d t , x ∈ R n .
The strict positivity of P τ − t 0 implies u ( · , t 0 ) ≡ 0 . Hence u ≡ 0 , a contradiction. Thus u ( · , t ) is nonzero at every time. Fix ( x , t ) ∈ R n × R and choose t 0 < t . Since u ( · , t 0 ) is nonnegative, continuous, and nonzero, (21) and strict positivity of the heat kernel imply
u ( x , t ) ≥ ∫ R n P t − t 0 ( x − y ) u ( y , t 0 ) d y > 0 .
Thus we complete the proof of Lemma 2.5. □

2.3. Liouville Theorems for Bounded Solutions

For Ω ⊂ R n , I ⊂ R and α , β ∈ ( 0 , 1 ) , define the parabolic Hölder seminorm by
[ u ] C x , t β , α ( Ω × I ) : = sup t , t ′ ∈ I , x , x ′ ∈ Ω ( x , t ) ≠ ( x ′ , t ′ ) | u ( x , t ) − u ( x ′ , t ′ ) | | t − t ′ | β + | x − x ′ | α .
We also denote
[ v ] C x δ ( Ω × I ) : = sup t ∈ I [ v ( · , t ) ] C δ ( Ω ) , [ v ] C t δ 2 s ( Ω × I ) : = sup x ∈ Ω [ v ( · , t ) ] C δ 2 s ( I ) .
The following Hölder estimate follows from [25][Theorem 1.3] after a change of time variable.
Proposition 2.6. 
Fix 0 < s < 1 , x 0 ∈ R n , b ∈ R , and R , T > 0 . Let Q = B 2 R ( x 0 ) × ( b − 2 T , b ) and Q ′ = B R ( x 0 ) × ( b − T , b ) . Assume that a ∈ C ( ( b − 2 T , b ) ) satisfies 0 < a * ≤ a ( t ) ≤ a * < ∞ , and f ∈ L ∞ ( Q ) . Let v ∈ L ∞ ( R n × ( b − 2 T , b ) ) be a weak solution of
v t + a ( t ) ( − Δ ) s v = f in Q .
Then for every 0 < δ < min { 1 , 2 s } , there exists a positive constant C, depending only on n , s , δ , a * , a * , R , T , such that
[ v ] C t δ 2 s ( Q ′ ) + [ v ] C x δ ( Q ′ ) ≤ C ∥ v ∥ L ∞ ( R n × ( b − 2 T , b ) ) + ∥ f ∥ L ∞ ( Q ) .
Moreover, if v extends continuously to B R ( x 0 ) × { b } , then (25) holds with Q ′ = B R ( x 0 ) × ( b − T , b ] . 
Proof. 
Let θ ( t ) : = − ∫ t b a ( r ) d r , b − 2 T ≤ t ≤ b . Since a ∈ C ( ( b − 2 T , b ) ) and satisfies 0 < a * ≤ a ( t ) ≤ a * < ∞ , J ^ : = θ ( ( b − 2 T , b ) ) = ( θ ( b − 2 T ) , 0 ) is an open interval. Define
w ( x , θ ( t ) ) : = v ( x , t ) , g ( x , θ ( t ) ) : = f ( x , t ) a ( t ) .
Then
a * | t − r | ≤ | θ ( t ) − θ ( r ) | ≤ a * | t − r | , ∥ g ∥ L ∞ ( B 2 R ( x 0 ) × J ^ ) ≤ a * − 1 ∥ f ∥ L ∞ ( Q ) ,
and
∥ w ∥ L ∞ ( R n × J ^ ) = ∥ v ∥ L ∞ ( R n × ( b − 2 T , b ) ) , [ w ] C x δ ( Q ′ ) = [ v ] C x δ ( Q ′ )
By (24), w ( x , θ ) is a weak solution of
w θ + ( − Δ ) s w = g in B 2 R ( x 0 ) × J ^ .
Let Q ^ ′ : = B R ( x 0 ) × ( θ ( b − T ) , 0 ) . By (26), there holds
min { 1 , ( a * ) − δ 2 s } [ v ] C t δ 2 s ( Q ′ ) ≤ [ w ] C θ δ 2 s ( Q ^ ′ ) ≤ max { 1 , a * − δ 2 s } [ v ] C t δ 2 s ( Q ′ ) .
Applying [25][Theorem 1.3] to w on B 2 R ( x 0 ) × J ^ with ε = 2 s − δ , using scaling and a finite covering of Q ^ ′ , we have
[ w ] C θ δ 2 s ( Q ^ ′ ) + [ w ] C x δ ( Q ^ ′ ) ≤ C ∥ w ∥ L ∞ ( R n × J ^ ) + ∥ g ∥ L ∞ ( B 2 R ( x 0 ) × J ^ ) ,
where C depends only on n , s , δ , a * , a * , R , T . Then by (26), (27) and (28), we obtain (25).
Finally, if v extends continuously to t = b , by (25), for any ( x , t ) , ( y , r ) ∈ B R ( x 0 ) × ( b − T , b ) ,
| v ( x , t ) − v ( y , r ) | ≤ C ∥ v ∥ L ∞ ( R n × ( b − 2 T , b ) ) + ∥ f ∥ L ∞ ( Q ) | x − y | δ + | t − r | δ 2 s .
Letting r → b with t < b fixed, and taking t = r → b respectively, extends (29) to all ( x , t ) , ( y , r ) ∈ B R ( x 0 ) × ( b − T , b ] .
Thus we complete the proof of Proposition 2.6. □
By Proposition 2.6, we have the following Liouville type theorems for nonnegative bounded weak solutions of fractional parabolic and elliptic equations.
Proposition 2.7. 
Let 0 < s < 1 and p > 1 . Suppose that 0 ≤ V ∈ L ∞ ( R n × R ) is a weak solution of
V t + ( − Δ ) s V = V p in R n × R .
Then V is a bounded nonnegative classical entire solution in the sense of Definition 1.1. Moreover, if V ¬ ≡ 0 , then
V ( x , t ) > 0 for every ( x , t ) ∈ R n × R .
Proof. 
Set M : = ∥ V ∥ L ∞ ( R n × R ) . If M = 0 , the conclusion is trivially valid. Next we assume that M > 0 . Choose 0 < β < min { 2 s , 1 } such that 2 s + β ∉ N . For every x 0 ∈ R n and compact interval J ⋐ R , applying Proposition 2.6 with a ≡ 1 and f = V p on Q ′ = B 1 ( x 0 ) × J , we have
[ V ] C x β ( B 1 ( x 0 ) × J ) + [ V ] C t β 2 s ( B 1 ( x 0 ) × J ) ≤ C J ( M + M p ) ,
where C J is independent of x 0 . If | x 0 − y | ≥ 1 , we have | V ( x 0 , t ) − V ( y , t ) | ≤ 2 M . Thus, the fact that x 0 ∈ R n is arbitrary, after increasing C J , we obtain
[ V ] C t , x β 2 s , β ( R n × J ) ≤ C J ( M + M p ) < ∞ .
Since p > 1 , the mean value theorem yields | a p − b p | ≤ p M p − 1 | a − b | for 0 ≤ a , b ≤ M . Thus we have
[ V p ] C t , x β 2 s , β ( R n × J ) ≤ p M p − 1 [ V ] C t , x β 2 s , β ( R n × J ) < ∞ .
Hence by [25][Corollary 3.6], we get
∥ V ∥ C t 1 + β 2 s ( R n × J ) + sup t ∈ J ∥ V ( · , t ) ∥ C 2 s + β ( R n ) ≤ C J
for every compact time interval J ⊂ R . Thus V t and ( − Δ ) s V are continuous, and (30) holds pointwise. By Lemma 2.5, we have
V ( x , t ) > 0 for every ( x , t ) ∈ R n × R .
We first improve the spatial regularity beyond order 1. Since 0 ≤ V ≤ M , by the mean value theorem, we have
∥ V ( · , t ) p ∥ C β ( R n ) ≤ p M p − 1 ∥ V ( · , t ) ∥ C β ( R n ) .
Fix a compact time interval J ⊂ R . Choose 0 < δ < min { 2 s , 1 } such that β + j δ ≠ 1 for every integer j ≥ 1 , and let N be the smallest integer such that β + N δ > 1 . Choose bounded open intervals J ⋐ J N ⋐ J N − 1 ⋐ … ⋐ J 0 . Then
sup t ∈ J 0 ∥ V ( · , t ) ∥ C β ( R n ) ≤ C J 0 < ∞ .
Starting with this bound, for any fixed x 0 ∈ R n , by applying the spatial estimate in [25][Corollary 3.4] successively on B 1 ( x 0 ) × J i with ε = 2 s − δ , for j = 0 , … , N − 1 , then using V ∈ L ∞ ( R n × R ) and the arbitrary of x 0 , we deduce
sup t ∈ J j + 1 ∥ V ( · , t ) ∥ C β + ( j + 1 ) δ ( R n ) ≤ C j sup t ∈ J j ∥ V ( · , t ) ∥ C β + j δ ( R n ) + sup t ∈ J j ∥ V ( · , t ) p ∥ C β + j δ ( R n ) ≤ C j ( 1 + p M p − 1 ) sup t ∈ J j ∥ V ( · , t ) ∥ C β + j δ ( R n ) .
Here, the second inequality applies because β + j δ < 1 . By the choice of N, we have 1 < β + N δ < 1 + δ < 2 . Set η : = β + N δ − 1 ∈ ( 0 , 1 ) , the finite iteration therefore yields
sup t ∈ J ∥ V ( · , t ) ∥ C 1 , η ( R n ) < ∞ .
By applying [25][Corollary 3.8] and decreasing η if necessary, we obtain
sup | t | ≤ T ∥ V ( · , t ) ∥ C 2 , η ( B R ¯ ) < ∞ for R , T > 0 .
In particular, the uniform local C 1 , 1 bounds in Definition 1.1 hold. Finally, boundedness yields V ( · , t ) ∈ L 2 s ( R n ) for every t ∈ R . This proves that V is a bounded nonnegative classical entire solution in the sense of Definition 1.1. □

3. The Fujita Range, Morrey Estimates and Compactness

3.1. Liouville Theorem for Bounded Solutions in Fujita Range

Set ρ ( x ) = ( 1 + | x | 2 ) − n + 2 s 2 and ρ R ( x ) = ρ ( x R ) . We have the following decay estimate for ( − Δ ) s ρ .
Lemma 3.1. 
There is C = C ( n , s ) such that
| ( − Δ ) s ρ ( x ) | ≤ C ρ ( x ) , | ( − Δ ) s ρ R ( x ) | ≤ C R − 2 s ρ R ( x ) .
Proof. 
The second estimate follows from scaling, so it is enough to prove the first inequality. For x ∈ B 4 and { | x − y | < 1 } , by Taylor’s formula we have | ρ ( x ) − ρ ( y ) − ∇ ρ ( x ) · ( x − y ) | ≤ C | x − y | 2 . Thus, by symmetry of integral, we have
( − Δ ) s ρ ( x ) = c n , s ∫ | x − y | < 1 ρ ( x ) − ρ ( y ) | x − y | n + 2 s d y + ∫ | x − y | ≥ 1 ρ ( x ) − ρ ( y ) | x − y | n + 2 s d y = c n , s ∫ | x − y | < 1 ∇ ρ ( x ) · ( x − y ) + O ( | x − y | 2 ) | x − y | n + 2 s d y + ∫ | x − y | ≥ 1 ρ ( x ) − ρ ( y ) | x − y | n + 2 s d y = c n , s ∫ | x − y | < 1 O ( | x − y | 2 ) | x − y | n + 2 s d y + ∫ | x − y | ≥ 1 ρ ( x ) − ρ ( y ) | x − y | n + 2 s d y .
Since ρ has a positive lower bound on B 4 , it follows that | ( − Δ ) s ρ ( x ) | ≤ C ρ ( x ) for x ∈ B 4 .
For x ∈ R n ∖ B 4 , let r = | x | > 4 and Ω 1 = { | x − y | < r 2 } , Ω 2 = { | x − y | ≥ r 2 , | y | < r 2 } , Ω 3 = { | x − y | ≥ r 2 , | y | ≥ r 2 } . For y ∈ Ω 1 , every point on the line segment joining x and y has the form z = x + θ ( y − x ) , with 0 ≤ θ ≤ 1 . Hence | z | ≥ | x | − | z − x | = r − θ | y − x | > r 2 . Differentiating ρ ( z ) along this segment, we deduce
| D 2 ρ ( z ) | ≤ C ( 1 + | z | 2 ) − n + 2 s + 2 2 ≤ C r − n − 2 s − 2
Taylor’s formula therefore yields
| ρ ( x ) − ρ ( y ) − ∇ ρ ( x ) · ( x − y ) | ≤ C r − n − 2 s − 2 | x − y | 2 , y ∈ Ω 1 .
Then by symmetry we get
∫ Ω 1 ρ ( x ) − ρ ( y ) | x − y | n + 2 s d y = ∫ Ω 1 ρ ( x ) − ρ ( y ) − ∇ ρ ( x ) · ( x − y ) | x − y | n + 2 s d y ≤ C r − n − 2 s − 2 ∫ | h | < r 2 | h | 2 − n − 2 s d h = C r − n − 2 s − 2 ∫ 0 r 2 τ 1 − 2 s d τ = C r − n − 4 s .
On Ω 2 , | x − y | ≥ r 2 , and therefore
∫ Ω 2 | ρ ( x ) − ρ ( y ) | | x − y | n + 2 s d y ≤ C r − n − 2 s ρ ( x ) | B r 2 | + ∫ R n ρ ( y ) d y ≤ C r − n − 2 s .
On Ω 3 , since | x − y | ≥ r 2 , one has
ρ ( x ) ∫ Ω 3 | x − y | − n − 2 s d y ≤ C ρ ( x ) ∫ { | h | ≥ r 2 } | h | − n − 2 s d h ≤ C r − n − 4 s .
Since | y | ≥ r 2 on Ω 3 , we have ρ ( y ) ≤ C r − n − 2 s on Ω 3 , this yields
∫ Ω 3 ρ ( y ) | x − y | n + 2 s d y ≤ C r − n − 2 s ∫ { | h | ≥ r 2 } | h | − n − 2 s d h ≤ C r − n − 4 s .
Thus | ( − Δ ) s ρ ( x ) | ≤ C r − n − 2 s ≤ C ρ ( x ) for x ∈ R n ∖ B 4 . Together with | ( − Δ ) s ρ ( x ) | ≤ C ρ ( x ) for x ∈ B 4 , we prove the first inequality in (31). Finally, ( − Δ ) s ρ R ( x ) = R − 2 s ( ( − Δ ) s ρ ) ( x R ) , which proves the second estimate. □
Choose 0 ≤ η ≤ 1 , η ∈ C c ∞ ( ( − 2 , 2 ) ) , with η = 1 on [ − 1 , 1 ] . Let p ′ = p p − 1 and fix an integer m ≥ p ′ . For R > 0 , define
η R ( t ) = η ( t R 2 s ) , Φ R ( x , t ) = η R ( t ) m ρ R ( x ) , A R = | ∂ t Φ R | + | ( − Δ ) s Φ R | .
Since m ≥ p ′ > 1 is an integer and ρ R > 0 , the equality Φ R = 0 implies ∂ t Φ R = ( − Δ ) s Φ R = 0 . Thus A R = 0 on { Φ R = 0 } . In the following lemma, we define Φ R − 1 p − 1 A R p ′ = 0 on { Φ R = 0 } .
Lemma 3.2. 
The functions Φ R and A R satisfy
∫ ∫ R n + 1 Φ R − 1 p − 1 A R p ′ d x d t ≤ C R n + 2 s − 2 s p ′ = C R n − q .
If p = p F ( n , s ) , then { A R } is bounded in L p ′ ( R n + 1 ) , and A R → 0 in L loc ∞ ( R n + 1 ) as R → ∞ . 
Proof. 
By Lemma 3.1, we have
| ∂ t Φ R | ≤ m R − 2 s | η ′ ( t R 2 s ) | η R m − 1 ρ R ≤ C m R − 2 s η R m − 1 ρ R , | ( − Δ ) s Φ R | ≤ C R − 2 s η R m ρ R .
On { Φ R > 0 } , there holds
Φ R − 1 p − 1 ( R − 2 s ρ R η R m − 1 ) p ′ = R − 2 s p ′ ρ R η R m − p ′ , Φ R − 1 p − 1 ( R − 2 s ρ R η R m ) p ′ = R − 2 s p ′ ρ R η R m .
Here p ′ − 1 p − 1 = 1 and ( m − 1 ) p ′ − m p − 1 = m − p ′ . Since m − p ′ ≥ 0 , the inequality ( a + b ) p ′ ≤ 2 p ′ − 1 ( a p ′ + b p ′ ) yields
∫ ∫ R n + 1 Φ R − 1 p − 1 A R p ′ d x d t ≤ C R − 2 s p ′ ∫ { η R > 0 } ∫ R n ρ R ( x ) η R ( t ) m − p ′ + η R ( t ) m d x d t ≤ C R − 2 s p ′ ∫ − 2 R 2 s 2 R 2 s ∫ R n ρ R ( x ) d x d t ≤ C R n + 2 s − 2 s p ′ .
Since 2 s p ′ = 2 s + 2 s p − 1 = 2 s + q , this proves (32).
Suppose that p = p F ( n , s ) . Then q = n and 2 s p ′ = n + 2 s . By (33) we have
∫ ∫ R n + 1 A R p ′ d x d t ≤ C R − 2 s p ′ ∫ − 2 R 2 s 2 R 2 s ∫ R n ρ R ( x ) p ′ d x d t ≤ C R n + 2 s − 2 s p ′ = C .
If K ⋐ R n + 1 , then η R = 1 and ∂ t η R = 0 on K for all sufficiently large R. Consequently,
∥ A R ∥ L ∞ ( K ) = ∥ ( − Δ ) s ρ R ∥ L ∞ ( K ) ≤ C R − 2 s → 0 .
This completes the proof of Lemma 3.2. □
Theorem 3.3. 
Assume 1 < p ≤ p F ( n , s ) . Let u be a nonnegative bounded classical entire solution of (1), in the sense of Definition 1.1. Then
u ( x , t ) = 0 for every ( x , t ) ∈ R n × R .
Proof. 
Fix R ≥ 1 . Choose χ ∈ C c ∞ ( B 2 ) with 0 ≤ χ ≤ 1 and χ = 1 on B 1 , and set χ L ( x ) : = χ ( x L ) for L ≥ 2 . Testing (1) against χ L Φ R and integrating by parts in space and time, and using η ∈ C c ∞ ( ( − 2 , 2 ) ) , yield
∫ ∫ R n + 1 u p χ L Φ R d x d t = ∫ ∫ R n + 1 u − χ L ∂ t Φ R + ( − Δ ) s ( χ L Φ R ) d x d t = ∫ ∫ R n + 1 u χ L − ∂ t Φ R + ( − Δ ) s Φ R d x d t + E L , R ,
where E L , R : = ∫ ∫ R n + 1 u [ ( − Δ ) s , χ L ] Φ R d x d t . Since Φ R ( · , t ) = η R ( t ) m ρ R , we have
sup t ∈ R ∥ Φ R ( · , t ) ∥ W 1 , 1 ( R n ) ≤ ∥ ρ R ∥ W 1 , 1 ( R n ) < ∞ .
Thus Lemma 2.4 yields
| E L , R | ≤ ∥ u ∥ L ∞ ( R n + 1 ) ∫ − 2 R 2 s 2 R 2 s ∥ [ ( − Δ ) s , χ L ] Φ R ( · , t ) ∥ 1 d t ≤ C R ∥ u ∥ L ∞ ( R n + 1 ) ( 1 + log L ) L − min { 2 s , 1 } ⟶ 0 ( L → ∞ ) .
For fixed R, then Φ R and A R vanish for | t | ≥ 2 R 2 s and satisfy Φ R ≤ ρ R and A R ≤ C R − 2 s ρ R . Since ρ R ∈ L 1 ( R n ) , we have
∫ ∫ R n + 1 Φ R d x d t ≤ 4 R 2 s ∫ R n ρ R ( x ) d x < ∞ , ∫ ∫ R n + 1 A R d x d t ≤ C R − 2 s 4 R 2 s ∫ R n ρ R ( x ) d x < ∞ .
Thus Φ R , A R ∈ L 1 ( R n + 1 ) . By the boundedness of u and 0 ≤ χ L ≤ 1 ,
0 ≤ u p χ L Φ R ≤ ∥ u ∥ L ∞ ( R n + 1 ) p Φ R , u χ L − ∂ t Φ R + ( − Δ ) s Φ R ≤ ∥ u ∥ L ∞ ( R n + 1 ) A R .
Since χ L ( x ) → 1 for every x ∈ R n , by the dominated convergence theorem, we have
∫ ∫ R n + 1 u p χ L Φ R d x d t ⟶ ∫ ∫ R n + 1 u p Φ R d x d t , ∫ ∫ R n + 1 u χ L − ∂ t Φ R + ( − Δ ) s Φ R d x d t ⟶ ∫ ∫ R n + 1 u − ∂ t Φ R + ( − Δ ) s Φ R d x d t .
Combining these limits with E L , R → 0 , we obtain
I R : = ∫ ∫ R n + 1 u p Φ R d x d t = ∫ ∫ R n + 1 u − ∂ t Φ R + ( − Δ ) s Φ R d x d t ≤ ∫ ∫ R n + 1 u A R d x d t .
Since A R = 0 on { Φ R = 0 } , Hölder’s inequality yields
∫ ∫ R n + 1 u A R d x d t = ∫ ∫ { Φ R > 0 } ( u Φ R 1 p ) ( A R Φ R − 1 p ) d x d t ≤ I R 1 p ∫ ∫ R n + 1 A R p ′ Φ R − 1 p − 1 d x d t 1 p ′ .
Young’s inequality and (32) from Lemma 3.2 therefore imply I R ≤ 1 2 I R + C R n − q , and hence
I R ≤ C R n − q .
As R → ∞ , η R ( t ) → 1 and ρ R ( x ) → 1 , so Φ R ( x , t ) → 1 at every ( x , t ) ∈ R n + 1 . If p < p F , then q > n , and Fatou’s lemma yields
0 ≤ ∫ ∫ R n + 1 u p d x d t ≤ lim inf R → ∞ I R ≤ lim R → ∞ C R n − q = 0 .
Thus u ≡ 0 by continuity.
Suppose now that p = p F ( n , s ) . Then q = n , so (35) and Fatou’s lemma imply
u ∈ L p ( R n + 1 ) , ∥ u ∥ L p ( R n + 1 ) p ≤ lim inf R → ∞ I R ≤ C .
Lemma 3.2 also implies
sup R ≥ 1 ∥ A R ∥ L p ′ ( R n + 1 ) ≤ C , ∥ A R ∥ L ∞ ( K ) ⟶ 0
for every compact set K ⊂ R n + 1 . For any ε > 0 , by (36), we can choose a compact set K with ∥ u ∥ L p ( K c ) < ε . By (34) and applying Hölder’s inequality on K c , we obtain
I R ≤ ∫ ∫ K u A R d x d t + ∫ ∫ K c u A R d x d t ≤ ∥ u ∥ L 1 ( K ) ∥ A R ∥ L ∞ ( K ) + ∥ u ∥ L p ( K c ) ∥ A R ∥ L p ′ ( R n + 1 ) ≤ ∥ u ∥ L 1 ( K ) ∥ A R ∥ L ∞ ( K ) + C ε .
This shows lim sup R → ∞ I R ≤ C ε . Since ε > 0 is arbitrary, I R → 0 . Fatou’s lemma then yields
∫ ∫ R n + 1 u p d x d t = 0 ,
and continuity again implies u ≡ 0 . This completes the proof of Theorem 3.3. □

3.2. Differential Inequality, Morrey Bounds, and Compactness

Let φ be the positive first Dirichlet eigenfunction of ( − Δ ) s on B 1 , normalized by ∫ B 1 φ ( x ) d x = 1 , i.e.,
( − Δ ) s φ = λ 1 φ in B 1 , φ = 0 in B 1 c .
For a ∈ R n and R > 0 , define
φ R , a ( x ) : = R − n φ ( x − a R ) , λ R : = λ 1 R − 2 s , F R , a ( t ) : = ∫ R n u ( x , t ) φ R , a ( x ) d x .
Lemma 3.4. 
Let u : R n × R → [ 0 , ∞ ) be a classical entire solution of (1) in the sense of Definition 1.1. Then for every a ∈ R n and R > 0 , F R , a ∈ C 1 ( R ) and satisfies
F R , a ′ ( t ) ≥ F R , a ( t ) p − λ R F R , a ( t ) , t ∈ R .
Proof. 
Fix a ∈ R n and R > 0 , and let D : = B R ( a ) and d ( x ) : = dist ( x , ∂ D ) . Then φ R , a = 0 on D c and ∫ D φ R , a ( x ) d x = 1 . By scaling the first-eigenfunction estimate in [12][(3.10)] to B 1 and using the normalization ∫ B 1 φ ( x ) d x = 1 , we obtain
c dist ( x , ∂ B 1 ) s ≤ φ ( x ) ≤ C dist ( x , ∂ B 1 ) s , x ∈ B 1 .
In particular, φ is bounded, so [46][Proposition 1.1], applied to ( − Δ ) s φ = λ 1 φ , implies φ ∈ C s ( R n ) . Translation and scaling therefore yield
c R d ( x ) s ≤ φ R , a ( x ) ≤ C R d ( x ) s ( x ∈ D ) , ∥ φ R , a ∥ C s ( R n ) ≤ C R ,
where c R , C R > 0 depend only on n , s , R . For x ∈ R n ∖ D ¯ , since φ R , a ( x ) = 0 we have
( − Δ ) s φ R , a ( x ) = − c n , s ∫ D φ R , a ( y ) | x − y | n + 2 s d y < 0 .
The global C s bound in (38) implies φ R , a ( y ) ≤ C R | x − y | s . Since | x − y | ≥ d ( x ) for y ∈ D , we obtain
| ( − Δ ) s φ R , a ( x ) | ≤ C R ∫ { | h | ≥ d ( x ) } | h | − n − s d h ≤ C R d ( x ) − s , x ∈ R n ∖ D ¯ .
For | x − a | > 2 R , the inequality | x − y | ≥ | x − a | 2 and the normalization of φ R , a also imply
| ( − Δ ) s φ R , a ( x ) | ≤ C | x − a | − n − 2 s .
Inside D, we have ( − Δ ) s φ R , a = λ R φ R , a . Moreover, d − s is integrable near ∂ D because s < 1 . Thus by (40), (41), local boundedness of u and the assumption u ( · , t ) ∈ L 2 s ( R n ) , we obtain
∫ R n u ( x , t ) | ( − Δ ) s φ R , a ( x ) | d x < ∞ ( t ∈ R ) .
Fix t ∈ R , (42) yields
∫ D φ R , a ( x ) ( − Δ ) s u ( x , t ) d x = ∫ R n u ( x , t ) ( − Δ ) s φ R , a ( x ) d x .
Since u t is continuous and φ R , a has fixed compact support, differentiation under the integral implies F R , a ∈ C 1 ( R ) and F R , a ′ ( t ) = ∫ D u t ( x , t ) φ R , a ( x ) d x . Using (1), (43), and ( − Δ ) s φ R , a = λ R φ R , a in D, we obtain
F R , a ′ ( t ) = ∫ D u ( x , t ) p φ R , a ( x ) d x − ∫ R n u ( x , t ) ( − Δ ) s φ R , a ( x ) d x = ∫ D u ( x , t ) p φ R , a ( x ) d x − λ R F R , a ( t ) − ∫ R n ∖ D ¯ u ( x , t ) ( − Δ ) s φ R , a ( x ) d x ≥ ∫ D u ( x , t ) p φ R , a ( x ) d x − λ R F R , a ( t ) ,
where the last inequality follows from u ≥ 0 and (39). Finally, Jensen’s inequality for the probability measure φ R , a ( x ) d x on D yields
∫ D u ( x , t ) p φ R , a ( x ) d x ≥ ∫ D u ( x , t ) φ R , a ( x ) d x p = F R , a ( t ) p .
This proves (37). □
Proposition 3.5. 
Let u : R n × R → [ 0 , ∞ ) be a classical entire solution of (1) in the sense of Definition 1.1. Then, for every a ∈ R n , R > 0 , and t ∈ R ,
F R , a ( t ) ≤ λ 1 1 p − 1 R − q .
Consequently, there exists a constant K = K ( n , s , p ) > 0 such that
sup a ∈ R n , R > 0 , t ∈ R R q − n ∫ B R ( a ) u ( x , t ) d x ≤ K .
Proof. 
Fix a ∈ R n and R > 0 , and let F : = F R , a and λ : = λ R . By Lemma 3.4, F ∈ C 1 ( R ) and satisfies (37). Suppose that F ( t 0 ) > λ 1 p − 1 for some t 0 ∈ R , and set δ : = 1 − λ F ( t 0 ) 1 − p > 0 . Since F ′ ( t 0 ) > 0 , we have F ( t ) > F ( t 0 ) for t > t 0 sufficiently close to t 0 . Suppose that F ( t ) ≤ F ( t 0 ) at some t > t 0 . By continuity, there exists a first time t 1 > t 0 such that F ( t 1 ) = F ( t 0 ) , with F ( t ) > F ( t 0 ) for t 0 < t < t 1 . Hence
F ′ ( t 1 ) = lim h ↓ 0 F ( t 1 ) − F ( t 1 − h ) h ≤ 0 .
On the other hand, since F ( t 1 ) = F ( t 0 ) , (37) implies F ′ ( t 1 ) ≥ δ F ( t 0 ) p > 0 , a contradiction. Hence F ( t ) ≥ F ( t 0 ) > 0 for every t ≥ t 0 , and
d d t F ( t ) 1 − p = ( 1 − p ) F ( t ) − p F ′ ( t ) ≤ − ( p − 1 ) δ .
Integration from t 0 to t yields
F ( t ) 1 − p ≤ F ( t 0 ) 1 − p − ( p − 1 ) δ ( t − t 0 ) , t ≥ t 0 .
Set t * : = t 0 + F ( t 0 ) 1 − p ( p − 1 ) δ . Since F ∈ C 1 ( R ) and F ( t * ) ≥ F ( t 0 ) > 0 , we have F ( t * ) 1 − p > 0 , whereas (46) at t = t * shows F ( t * ) 1 − p ≤ 0 . This contradiction proves
F R , a ( t ) ≤ λ R 1 p − 1 = λ 1 1 p − 1 R − q , t ∈ R ,
which is (44).
To prove (45), set m φ : = min | z | ≤ 1 2 φ ( z ) > 0 . For x ∈ B R ( a ) , the definition of φ 2 R , a implies
φ 2 R , a ( x ) = ( 2 R ) − n φ x − a 2 R ≥ ( 2 R ) − n m φ .
Consequently, applying (44) with radius 2 R , we obtain
∫ B R ( a ) u ( x , t ) d x ≤ ( 2 R ) n m φ F 2 R , a ( t ) ≤ 2 n − q λ 1 1 p − 1 m φ − 1 R n − q .
Since λ 1 and the normalized eigenfunction φ depend only on n , s , this proves (45) with K : = 2 n − q λ 1 1 p − 1 m φ − 1 . □
Lemma 3.6. 
Let 0 < q ≤ n and suppose that f : R n → [ 0 , ∞ ) is measurable with M q [ f ] ≤ K . Then there exists a constant C = C ( n , s ) > 0 such that, for every a ∈ R n and R ≥ 1 ,
∫ R n f ( x ) 1 + | x − a | n + 2 s d x ≤ C K , ∫ R n ∖ B R ( a ) f ( x ) | x − a | n + 2 s d x ≤ C K R − q − 2 s .
In particular, f ∈ L 2 s ( R n ) . 
Proof. 
Fix a ∈ R n and R ≥ 1 . For m ≥ 0 , set A m : = B 2 m + 1 R ( a ) ∖ B 2 m R ( a ) . Then M q [ f ] ≤ K yields
∫ A m f ( x ) | x − a | n + 2 s d x ≤ ( 2 m R ) − n − 2 s ∫ B 2 m + 1 R ( a ) f ( x ) d x ≤ 2 n − q K R − q − 2 s 2 − m ( q + 2 s ) .
Summing over m ≥ 0 and using 0 < q ≤ n , we obtain
∫ R n ∖ B R ( a ) f ( x ) | x − a | n + 2 s d x ≤ 2 n − q 1 − 2 − ( q + 2 s ) K R − q − 2 s ≤ 2 n 1 − 2 − 2 s K R − q − 2 s ,
which proves the second inequality in (47) with a constant depending only on n , s .
By M q [ f ] ≤ K we have ∫ B 1 ( a ) f ( x ) d x ≤ K , then we have
∫ R n f ( x ) 1 + | x − a | n + 2 s d x ≤ ∫ B 1 ( a ) f ( x ) d x + ∫ R n ∖ B 1 ( a ) f ( x ) | x − a | n + 2 s d x ≤ K + 2 n 1 − 2 − 2 s K ≤ C K .
This proves the first inequality in (47). Taking a = 0 also proves f ∈ L 2 s ( R n ) . □
Lemma 3.7. 
Fix 0 < q ≤ n and 0 < a * ≤ a * < ∞ . Let J , J j ⊂ R be open intervals such that J j ⊆ J j + 1 for every j ≥ 1 and ⋃ j = 1 ∞ J j = J . Suppose that 0 ≤ v j ∈ L l o c ∞ ( R n × J j ) are weak solutions of
( v j ) t + a j ( t ) ( − Δ ) s v j = f j in R n × J j ,
where a j ∈ C ( J j ) , a * ≤ a j ≤ a * , and f j ∈ L loc ∞ ( R n × J j ) . Assume that sup j sup t ∈ J j M q [ v j ( · , t ) ] ≤ K , and for every cylinder Q ⋐ R n × J , there is a constant C Q > 0 , independent of j, such that
∥ v j ∥ L ∞ ( Q ) + ∥ f j ∥ L ∞ ( Q ) ≤ C Q for all sufficiently large j .
Then there exist a subsequence of { v j } , still denoted by { v j } , and a continuous function V : R n × J → [ 0 , ∞ ) such that
v j ⟶ V locally uniformly on R n × J .
Moreover, for every ψ ∈ C c ∞ ( R n × J ) ,
∫ J j ∫ R n v j ( x , t ) ( − Δ ) s ψ ( x , t ) d x d t ⟶ ∫ J ∫ R n V ( x , t ) ( − Δ ) s ψ ( x , t ) d x d t .
Proof. 
Fix 0 < β < min { 1 , 2 s } and a cylinder Q ⋐ R n × J . Choose R , δ 0 > 0 and t 1 < t 2 such that
Q ¯ ⊂ B R × [ t 1 , t 2 ] , [ t 1 − δ 0 , t 2 + δ 0 ] ⋐ J .
By (48), there exist j 0 ∈ N and C 0 > 0 such that, for every j ≥ j 0 , [ t 1 − δ 0 , t 2 + δ 0 ] ⋐ J j and
∥ v j ∥ L ∞ ( B 4 R × [ t 1 − δ 0 , t 2 + δ 0 ] ) + ∥ f j ∥ L ∞ ( B 4 R × [ t 1 − δ 0 , t 2 + δ 0 ] ) ≤ C 0 .
Choose θ 1 ∈ C c ∞ ( B 4 ) with 0 ≤ θ 1 ≤ 1 and θ 1 = 1 on B 3 , and set
χ 1 , R ( x ) : = θ 1 ( x R ) , v ¯ j : = χ 1 , R v j .
Then
∥ v ¯ j ∥ L ∞ ( R n × [ t 1 − δ 0 , t 2 + δ 0 ] ) ≤ C 0 .
Let Q 1 : = B 2 R × ( t 1 − δ 0 , t 2 + δ 0 ) , since χ 1 , R = 1 on B 2 R , the equation for v j yields
( v ¯ j ) t + a j ( t ) ( − Δ ) s v ¯ j = f j + a j ( t ) ( − Δ ) s ( ( χ 1 , R − 1 ) v j ) = : f ¯ j in Q 1
Since χ 1 , R = 1 on B 3 R , we have ( χ 1 , R − 1 ) v j ( x , t ) = 0 for any x ∈ B 3 R . Thus, for x ∈ B 2 R ,
( − Δ ) s ( ( χ 1 , R − 1 ) v j ) ( x , t ) = c n , s ∫ R n ∖ B 3 R ( 1 − χ 1 , R ( y ) ) v j ( y , t ) | x − y | n + 2 s d y .
Since | x − y | ≥ R in (50), by (15) with ρ = R and sup j sup t ∈ J j M q [ v j ( · , t ) ] ≤ K , we get
( − Δ ) s ( ( χ 1 , R − 1 ) v j ) ( x , t ) ≤ c n , s ∫ { | x − y | > R } v j ( y , t ) | x − y | n + 2 s d y ≤ C K R − q − 2 s .
Consequently,
∥ f ¯ j ∥ L ∞ ( Q 1 ) ≤ C 0 + C a * K R − q − 2 s .
Apply Proposition 2.6 to v ¯ j with exponent β and T = δ 0 2 on the cylinders
B 2 R × ( b − δ 0 , b ) ⊂ Q 1 , b ∈ [ t 1 + δ 0 4 , t 2 + δ 0 4 ] .
Since v ¯ j = v j on B R , we obtain
∥ v j ∥ C t β 2 s ( B R × ( b − δ 0 2 , b ) ) + ∥ v j ∥ C x β ( B R × ( b − δ 0 2 , b ) ) ≤ C , j ≥ j 0 ,
Note that
Q ¯ ⊂ B R × [ t 1 , t 2 ] ⊂ ⋃ b ∈ [ t 1 + δ 0 4 , t 2 + δ 0 4 ] B R × ( b − δ 0 2 , b ) .
By a finite covering and (48), we have
∥ v j ∥ C t β 2 s ( Q ) + ∥ v j ∥ C x β ( Q ) ≤ C , j ≥ j 0 ,
where C is independent of j.
Since J j ⊆ J j + 1 for every j ≥ 1 and ⋃ j = 1 ∞ J j = J , by the Arzelà-Ascoli theorem and a diagonal argument, there exists a continuous nonnegative function V such that, after passing to a subsequence,
v j ⟶ V locally uniformly on R n × J .
Finally, fix a bounded open interval I ⋐ J and ψ ∈ C c ∞ ( R n × I ) . For all sufficiently large j, I ⊂ J j , then by applying Lemma 2.1 on I with f j = v j and f = V and using (51), we obtain (49). □

4. Localized Energy Estimates

Let u be a nonnegative classical solution of (1). Fix k > 0 and a ∈ R n , for t < k , let
r = k − t = e − σ , y = x − a r 1 2 s , w a , k ( y , σ ) = r α u ( x , t ) .
For notational simplicity, let w = w a , k . Since u = r − α w , we have
u t = r − α − 1 w σ + 1 2 s y · ∇ w + α w .
Moreover, it follows from α p = α + 1 that ( − Δ ) x s u = r − α − 1 ( − Δ ) y s w , u p = r − α − 1 w p . Hence w satisfies
w σ + ( − Δ ) s w + D w = w p , D w : = 1 2 s y · ∇ w + α w .
Define
Z [ w ] : = w σ + D w = w p − ( − Δ ) s w = r α + 1 u t .
Then Z [ w ] = 0 is equivalent to u t = 0 . Under the similarity transformation (52), we have
M q [ w ( · , σ ) ] = M q [ u ( · , k − e − σ ) ] .
Indeed, q = 2 s α yields
L q − n ∫ B L ( b ) w ( y , σ ) d y = r 1 2 s L q − n ∫ B r 1 2 s L ( a + r 1 2 s b ) u ( x , t ) d x .
Then taking the supremum over b ∈ R n and L > 0 , we deduce (55). For η ( y , σ ) ∈ C c ∞ ( R n × R ) , let
ϕ ( x , t ) : = η x − a ( k − t ) 1 2 s , − log ( k − t ) = η ( ( x − a ) r − 1 2 s , σ ) , t < k .
Then we have η ( y , σ ) w ( y , σ ) = r α ϕ ( x , t ) u ( x , t ) . Consequently, by the definition of E s , there holds
E s ( η w ) = r 2 α − n s + n + 2 s 2 s E s ( ϕ u ) = r μ E s ( ϕ u ) ,
and
∫ R n η 2 w p + 1 d y = r α ( p + 1 ) − n 2 s ∫ R n ϕ 2 u p + 1 d x = r μ ∫ R n ϕ 2 u p + 1 d x .
Here we have used α ( p + 1 ) = 2 α + 1 and μ = 2 α + 1 − n 2 s .

4.1. Weighted Means and Localized Energy Estimates

Let L s : = − ( − Δ ) s − 1 2 s y · ∇ . Then L s has the formal adjoint L s * ρ = − ( − Δ ) s ρ + 1 2 s div ( y ρ ) with respect to Lebesgue measure. Let ρ s = P 1 , then ρ s ^ ( ξ ) = e − | ξ | 2 s . By the estimates in [1,23], we have
0 < ρ s ( y ) ≤ C ( 1 + | y | ) − n − 2 s , | ∇ ρ s ( y ) | ≤ C ( 1 + | y | ) − n − 2 s − 1 .
Moreover, by Fourier transformation, we have
L s * ρ s ^ ( ξ ) = − | ξ | 2 s e − | ξ | 2 s − 1 2 s ξ · ∇ ξ e − | ξ | 2 s = 0 .
Thus L s * ρ s = 0 in S ′ ( R n ) . Since ρ s ≥ 0 and ∫ R n ρ s d y = 1 , we deduce that ρ s is an invariant probability density for L s , i.e., for bounded classical solutions of v τ = L s v , the weighted mean ∫ R n v ( y , τ ) ρ s ( y ) d y is independent of τ .
Lemma 4.1. 
Let 0 < q ≤ n . Suppose that w : R n × R → [ 0 , ∞ ) satisfies (53) pointwise, with
w , w σ ∈ C ( R n × R ) , w ( · , σ ) ∈ C l o c 1 , 1 ( R n ) ∩ L 2 s ( R n ) for every σ ∈ R .
Assume that, for every L > 0 and every compact interval I ⋐ R ,
sup σ ∈ I ∥ w ( · , σ ) ∥ C 1 , 1 ( B L ¯ ) < ∞ , sup σ ∈ R M q [ w ( · , σ ) ] ≤ K ,
and
B I : = sup σ ∈ I ∥ w ( · , σ ) ∥ ∞ < ∞ for every I ⋐ R .
Define the weighted means
M 1 ( σ ) = ∫ R n w ( y , σ ) ρ s ( y ) d y , M p ( σ ) = ∫ R n w ( y , σ ) p ρ s ( y ) d y .
Then the function M 1 defined above is locally absolutely continuous, hence belongs to W loc 1 , 1 ( R ) . It satisfies
M 1 ′ + α M 1 = M p for a . e . σ ∈ R ,
and obeys the uniform bounds
M 1 ( σ ) ≤ α 1 p − 1 ( σ ∈ R ) , ∫ σ 1 σ 2 M p ( σ ) d σ ≤ C ( 1 + σ 2 − σ 1 ) ( σ 1 < σ 2 ) .
Proof. 
The identity L s * ρ s = 0 established above implies
∫ R n ( L s h ) ρ s d y = 0 , h ∈ S ( R n ) .
We first justify its use in the weighted mean equation by approximating the noncompact weight ρ s .
Choose χ ∈ C c ∞ ( B 2 ) with 0 ≤ χ ≤ 1 and χ = 1 on B 1 , and set χ R ( y ) = χ ( y R ) , ρ s , R = χ R ρ s . It follows from L s * ρ s = 0 that
L s * ( χ R ρ s ) = − [ ( − Δ ) s , χ R ] ρ s + 1 2 s R y · ∇ χ ( y R ) ρ s .
Moreover, since ρ s , ∇ ρ s ∈ L 1 by (58), it follows from Lemma 2.4 that, for R ≥ 2 ,
∥ [ ( − Δ ) s , χ R ] ρ s ∥ 1 ≤ C ∥ ρ s ∥ W 1 , 1 ( 1 + log R ) R − min { 2 s , 1 } .
Note that y · ∇ χ ( y R ) is supported in R ≤ | y | ≤ 2 R , where | y | R ≤ 2 . Hence (58) yields
1 R y · ∇ χ ( y R ) ρ s 1 ≤ C ∫ R ≤ | y | ≤ 2 R ρ s ( y ) d y ≤ C R − 2 s .
By (61) and (62), ∥ L s * ρ s , R ∥ 1 → 0 as R → ∞ . Next fix I ⋐ R . Apply the estimate for the nonlocal remainder term in (47) of Lemma 3.6 to f = w ( · , σ ) with a = 0 . Using (58) and w p ≤ B I p − 1 w , we obtain
∫ | y | > R w ( y , σ ) + w ( y , σ ) p ρ s ( y ) d y ≤ C I K R − q − 2 s ( R ≥ 1 , σ ∈ I ) .
To show continuity of M 1 , fix σ 0 ∈ I . Splitting the integral into B R and its complement, and using ∫ R n ρ s d y = 1 , yields
| M 1 ( σ ) − M 1 ( σ 0 ) | ≤ sup y ∈ B R ¯ | w ( y , σ ) − w ( y , σ 0 ) | + 2 C I K R − q − 2 s .
For fixed R, the first term tends to zero as σ → σ 0 by local continuity of w. Choosing R large and then σ sufficiently close to σ 0 proves that M 1 is continuous on I.
Applying the first inequality in (47) with f = w ( · , σ ) , a = 0 and R = 1 , and using (58), yields
M 1 ( σ ) ≤ C ∫ R n w ( y , σ ) 1 + | y | n + 2 s d y ≤ C K .
Moreover, w p ≤ B I p − 1 w implies
M p ( σ ) ≤ B I p − 1 M 1 ( σ ) ≤ C K B I p − 1 ( σ ∈ I ) .
Both weighted means are therefore bounded on the finite interval I, hence belong to L 1 ( I ) . Since I ⋐ R is arbitrary, M 1 , M p ∈ L loc 1 ( R ) . Test (53) against θ ( σ ) ρ s , R ( y ) , where θ ∈ C c ∞ ( I ) . After integration by parts in time and in the drift term, the tested equation becomes
∫ I ∫ R n − w θ ′ + ( α w − w p ) θ ρ s , R d y d σ = ∫ I θ ( σ ) ∫ R n w L s * ρ s , R d y d σ .
The right-hand side is bounded in absolute value by B I ∥ θ ∥ L 1 ( I ) ∥ L s * ρ s , R ∥ 1 , which tends to zero as R → ∞ . On the left, the uniform bound on the nonlocal remainder terms allows ρ s , R to be replaced by ρ s in all three terms as R → ∞ . Thus
− ∫ I M 1 θ ′ d σ + α ∫ I M 1 θ d σ = ∫ I M p θ d σ ,
which is the weak form of (59). Thus M 1 ′ = M p − α M 1 weakly on I, and M 1 ∈ W 1 , 1 ( I ) . Its absolutely continuous representative agrees almost everywhere with the continuous integral defining M 1 , hence everywhere. Since I is arbitrary, this proves (59).
It remains to bound the weighted means. Jensen’s inequality yields M p ≥ M 1 p , since ρ s d y is a probability measure. Set m * : = α 1 p − 1 . If M 1 ( σ 0 ) > m * for some σ 0 ∈ R , the comparison principle for M 1 ′ ≥ M 1 p − α M 1 implies M 1 ( σ ) ≥ M 1 ( σ 0 ) for σ ≥ σ 0 . With δ : = 1 − α M 1 ( σ 0 ) 1 − p > 0 , it follows that M 1 ′ ≥ δ M 1 p , or equivalently, M 1 1 − p ′ ≤ − δ ( p − 1 ) . Integration yields
M 1 ( σ ) 1 − p ≤ M 1 ( σ 0 ) 1 − p − δ ( p − 1 ) ( σ − σ 0 ) .
This inequality forces M 1 ( σ ) to blow up no later than the finite time
σ * : = σ 0 + M 1 ( σ 0 ) 1 − p δ ( p − 1 ) .
However, we have 0 < M 1 ( σ * ) ≤ ∥ w ( · , σ * ) ∥ ∞ ≤ B I < ∞ on any compact interval I containing [ σ 0 , σ * ] , a contradiction. Hence M 1 ( σ ) ≤ m * for all σ ∈ R . Finally, integrating (59) yields
∫ σ 1 σ 2 M p ( σ ) d σ = M 1 ( σ 2 ) − M 1 ( σ 1 ) + α ∫ σ 1 σ 2 M 1 ( σ ) d σ ≤ m * + α m * ( σ 2 − σ 1 ) ,
which proves (60). □
Fix a reference time σ * ∈ R and a smooth spatial cutoff η * ∈ C c ∞ ( R n ) with 0 ≤ η * ≤ 1 . Define
η ( y , σ ) : = η * e − σ − σ * 2 s y .
We call η a material cutoff. By construction, η ∈ C ∞ ( R n × R ) , 0 ≤ η ≤ 1 , and η ( · , σ ) is compactly supported for every σ . Differentiating (63) yields
η σ + 1 2 s y · ∇ η = 0 .
For any material cutoff η , define the localized energy and the associated commutator by
F η ( σ ) [ w ] : = 1 2 E s ( η w ) − 1 p + 1 ∫ R n η 2 w p + 1 d y , C η ( σ ) [ w ] : = ( − Δ ) s ( η w ) − η ( − Δ ) s w .
Proposition 4.2. 
Let u be a bounded nonnegative classical entire solution of (1). Fix a ∈ R n and k > 0 , set w = w a , k and let η be a material cutoff as defined above. Then the map σ ↦ F η ( σ ) [ w ] has a locally absolutely continuous representative and, for almost every σ,
d d σ F η + μ F η + ∫ R n η 2 | Z [ w ] | 2 d y = 〈 C η [ w ] , η Z [ w ] 〉 .
Moreover, for σ 1 < σ 2 ,
∫ σ 1 σ 2 e μ ( σ − σ 2 ) ∫ R n η 2 | Z [ w ] | 2 d y d σ ≤ 2 e μ ( σ 1 − σ 2 ) F η ( σ 1 ) − 2 F η ( σ 2 ) + ∫ σ 1 σ 2 e μ ( σ − σ 2 ) ∥ C η [ w ] ∥ L 2 ( supp η ) 2 d σ .
Proof. 
Fix σ * and set λ * : = e − σ * 2 s . By (52) and (63),
ϕ ( x , t ) = η * e − σ − σ * 2 s e σ 2 s ( x − a ) = η * x − a λ * .
Then ϕ t = 0 . Let E ϕ [ u ] ( t ) : = 1 2 E s ( ϕ u ( t ) ) − 1 p + 1 ∫ R n ϕ 2 u ( t ) p + 1 d x , A = ( − Δ ) s and f ( t ) = ϕ u ( t ) . Since supp f ( t ) ⊂ supp ϕ , the local regularity in Definition 1.1 implies that, for every compact time interval J, f ∈ L ∞ ( J , H 2 ( R n ) ) , f t = ϕ u t ∈ L ∞ ( J , L 2 ( R n ) ) . The Fourier characterization [22][Section 3] implies D ( A ) = H 2 s ( R n ) and D ( A 1 2 ) = H s ( R n ) for the positive self-adjoint realization of A on L 2 . Since 2 s < 2 and J is finite,
f ∈ L 2 ( J , D ( A ) ) ∩ W 1 , 2 ( J , L 2 ) .
Moreover, ϕ A u = ϕ ( u p − u t ) ∈ L ∞ ( J , L 2 ) , hence [ A , ϕ ] u : = A f − ϕ A u ∈ L 2 ( J , L 2 ) .
Local continuity of u and the compact support of ϕ imply f ∈ C ( J , L 2 ) . We next prove f ∈ C ( J , D ( A 1 2 ) ) , so that the Dirichlet energy E s ( f ( t ) ) = ∥ A 1 2 f ( t ) ∥ 2 2 is continuous in time. To this end, we first show that its spectral approximations form a Cauchy sequence in C ( J , L 2 ) .
Let P N be the spectral projection of A on [ 0 , N ] , defined by P N g ^ ( ξ ) = 1 { | ξ | 2 s ≤ N } g ^ ( ξ ) . Since A 1 2 P N is independent of t and ∥ A 1 2 P N ∥ L 2 → L 2 ≤ N 1 2 , it preserves W 1 , 2 ( J , L 2 ) and commutes with weak time differentiation. Hence
A 1 2 P N f ∈ W 1 , 2 ( J , L 2 ) ⊂ C ( J , L 2 ) , ∂ t ( A 1 2 P N f ) = A 1 2 P N f t .
To estimate A 1 2 P M f − A 1 2 P N f uniformly in time, set Q N , M = P M − P N for 0 < N < M . Then A 1 2 P M f − A 1 2 P N f = A 1 2 Q N , M f . Note that, for every t , r ∈ J , we have
∥ A 1 2 Q N , M f ( t ) ∥ 2 2 − ∥ A 1 2 Q N , M f ( r ) ∥ 2 2 = 2 ∫ r t 〈 A Q N , M f , Q N , M f t 〉 d τ .
For every r , t ∈ J , the preceding identity implies
∥ A 1 2 Q N , M f ( t ) ∥ 2 2 ≤ ∥ A 1 2 Q N , M f ( r ) ∥ 2 2 + 2 ∫ J 〈 A Q N , M f , Q N , M f t 〉 d τ .
Averaging this inequality in r ∈ J and applying Cauchy–Schwarz yield a bound independent of t as follows
sup t ∈ J ∥ A 1 2 Q N , M f ( t ) ∥ 2 2 ≤ 1 | J | ∥ A 1 2 Q N , M f ∥ L 2 ( J , L 2 ) 2 + 2 ∥ A Q N , M f ∥ L 2 ( J , L 2 ) ∥ Q N , M f t ∥ L 2 ( J , L 2 ) .
Thus it remains to show that the right-hand side tends to zero uniformly for M > N as N → ∞ .
For every g ∈ L 2 ( J , L 2 ) , the convergence of the spectral remainder term to zero implies
∥ Q N , M g ∥ L 2 ( J , L 2 ) ≤ ∥ ( I − P N ) g ∥ L 2 ( J , L 2 ) ⟶ 0 as N → ∞ ,
uniformly for M > N . Since A 1 2 Q N , M f = Q N , M ( A 1 2 f ) and A Q N , M f = Q N , M ( A f ) , the preceding estimate for Q N , M g , applied to g = A 1 2 f , A f , f t , makes the right-hand side of (68) tend to zero as N → ∞ , uniformly for all M > N . Hence A 1 2 P N f is Cauchy in C ( J , L 2 ) and, by completeness, converges to some h ∈ C ( J , L 2 ) .
For each t ∈ J , we have P N f ( t ) → f ( t ) and A 1 2 P N f ( t ) → h ( t ) in L 2 . Closedness of A 1 2 therefore implies f ( t ) ∈ D ( A 1 2 ) and A 1 2 f ( t ) = h ( t ) . Since both f and h are continuous in L 2 , it follows that f ∈ C ( J , D ( A 1 2 ) ) .
Since A P N f → A f and P N f t → f t in L 2 ( J , L 2 ) , the pairings converge in L 1 ( J ) . Note that for every t , r ∈ J , we also have
1 2 ∥ A 1 2 P N f ( t ) ∥ 2 2 − 1 2 ∥ A 1 2 P N f ( r ) ∥ 2 2 = ∫ r t 〈 A P N f , P N f τ 〉 d τ ( t , r ∈ J ) .
Since A 1 2 P N f → A 1 2 f in C ( J , L 2 ) and 〈 A P N f , P N f τ 〉 → 〈 A f , f τ 〉 in L 1 ( J ) , letting N → ∞ yields
1 2 ∥ A 1 2 f ( t ) ∥ 2 2 − 1 2 ∥ A 1 2 f ( r ) ∥ 2 2 = ∫ r t 〈 A f , f τ 〉 d τ ( t , r ∈ J ) .
Thus 1 2 E s ( f ) = 1 2 ∥ A 1 2 f ∥ 2 2 is absolutely continuous, with derivative 〈 A f , f t 〉 almost everywhere. Since ϕ is independent of t and compactly supported, and u , u t are locally bounded, differentiation under the spatial integral yields
d d t 1 p + 1 ∫ R n ϕ ( x ) 2 u ( x , t ) p + 1 d x = ∫ R n ϕ ( x ) 2 u ( x , t ) p u t ( x , t ) d x .
Then using u p = A u + u t , we obtain
d d t E ϕ [ u ] = 〈 A ( ϕ u ) , ϕ u t 〉 − ∫ R n ϕ 2 u p u t d x = 〈 [ A , ϕ ] u , ϕ u t 〉 − ∫ R n ϕ 2 u t 2 d x .
Both energy terms are absolutely continuous, so this identity holds almost everywhere and in integrated form for every pair of times. Return to the variables in (52), with r = k − t = e − σ . By (56) and (57), E ϕ [ u ] = r − μ F η [ w ] , and thus we have
d d t E ϕ [ u ] = r − μ − 1 ∂ σ F η + μ F η .
For corresponding points ( x , t ) and ( y , σ ) , fractional scaling and the definition of Z [ w ] yield
C η ( σ ) [ w ] ( y ) = r α + 1 [ A , ϕ ] u ( x , t ) and η ( y , σ ) Z [ w ] ( y , σ ) = r α + 1 ϕ ( x ) u t ( x , t ) .
Since − 2 α − 2 + n 2 s = − μ − 1 , we have
〈 [ A , ϕ ] u , ϕ u t 〉 = r − μ − 1 〈 C η [ w ] , η Z [ w ] 〉 and ∫ R n ϕ 2 u t 2 d x = r − μ − 1 ∫ R n η 2 | Z [ w ] | 2 d y .
Cancelling r − μ − 1 proves (66). The smooth change of variables also implies local absolute continuity of F η . Finally, Young’s inequality on supp η in (66) yields
d d σ F η + μ F η + 1 2 ∫ R n η 2 | Z [ w ] | 2 d y ≤ 1 2 ∥ C η [ w ] ∥ L 2 ( supp η ) 2 .
Multiplying by e μ ( σ − σ 2 ) and using
e μ ( σ − σ 2 ) ∂ σ F η + μ F η = d d σ e μ ( σ − σ 2 ) F η ,
we integrate from σ 1 to σ 2 to obtain (67). □
Lemma 4.3. 
Let u be a bounded nonnegative classical entire solution of (1). For every t 0 ∈ R and ϕ ∈ C c ∞ ( R n ) with 0 ≤ ϕ ≤ 1 , there holds
1 2 E s ( ϕ u ( t 0 ) ) − 1 p + 1 ∫ R n ( ϕ u ( t 0 ) ) p + 1 d x ≥ 0 .
Proof. 
We first construct a global solution with initial value v 0 = ϕ u ( t 0 ) . The local regularity of u and the smoothness and compact support of ϕ imply v 0 ∈ X ∩ H s , where X = C 0 ( R n ) ∩ L 2 ( R n ) has norm ∥ z ∥ X = ∥ z ∥ ∞ + ∥ z ∥ 2 . The fractional heat semigroup is a strongly continuous contraction semigroup on X, and N ( z ) = ( z + ) p is locally Lipschitz on X. Thus Theorem 1.4 in [38][Chapter 6] ensures the existence of a unique maximal mild solution of
v t + A v = N ( v ) , v ( 0 ) = v 0 ,
with the continuation criterion
T max < ∞ ⟹ lim sup t ↑ T max ∥ v ( t ) ∥ X = ∞ .
Lemma 2.5 yields that v ≥ 0 , therefore N ( v ) = v p and v ( t ) = e − t A v 0 + ∫ 0 t e − ( t − τ ) A v ( τ ) p d τ . Moreover, since 0 ≤ v 0 ≤ u ( t 0 ) , we also have
0 ≤ v ( x , t ) ≤ u ( x , t 0 + t ) ≤ M , M : = ∥ u ∥ ∞ , 0 ≤ t < T max .
Then the L 2 -contraction property of the fractional heat semigroup e − t A and v p ≤ M p − 1 v imply
∥ v ( t ) ∥ 2 ≤ ∥ v 0 ∥ 2 + M p − 1 ∫ 0 t ∥ v ( τ ) ∥ 2 d τ .
Hence Gronwall’s inequality yields ∥ v ( t ) ∥ 2 ≤ e M p − 1 t ∥ v 0 ∥ 2 on [ 0 , T max ) . If T max < ∞ , these L ∞ and L 2 estimates keep ∥ v ( t ) ∥ X bounded for 0 ≤ t < T max , contradicting the continuation criterion. Hence T max = ∞ , and the Gronwall estimate yields
sup 0 ≤ t ≤ T ∥ v ( t ) ∥ 2 ≤ e M p − 1 T ∥ v 0 ∥ 2 < ∞ ( 0 < T < ∞ ) .
We next establish the regularity needed for the energy identity. Fix 0 < T < ∞ . Since 0 ≤ v ≤ M , (70) implies v p ∈ L 2 ( 0 , T , L 2 ) . For λ ≥ 0 , let P λ be the spectral projection of A onto [ 0 , λ ] . For the term e − t A v 0 , the Fourier multiplier formula yields A e − t A v 0 ^ ( ξ ) = | ξ | 2 s e − t | ξ | 2 s v ^ 0 ( ξ ) . Since
∫ 0 T | ξ | 4 s e − 2 t | ξ | 2 s d t = 1 2 | ξ | 2 s ( 1 − e − 2 T | ξ | 2 s ) ,
Plancherel’s identity and Tonelli’s theorem yield
∫ 0 T ∥ A e − t A v 0 ∥ 2 2 d t = 1 2 ∥ A 1 2 v 0 ∥ 2 2 − 1 2 ∥ A 1 2 e − T A v 0 ∥ 2 2 ≤ 1 2 ∥ A 1 2 v 0 ∥ 2 2 < ∞ ,
where finiteness follows from v 0 ∈ H s = D ( A 1 2 ) . Moreover, the weak time derivative satisfies ∂ t ( e − t A v 0 ) = − A e − t A v 0 ∈ L 2 ( 0 , T , L 2 ) . Fix τ ∈ [ 0 , T ] . As t → τ within [ 0 , T ] , e − t | ξ | 2 s → e − τ | ξ | 2 s for every ξ , and
( 1 + | ξ | 2 ) s e − t | ξ | 2 s − e − τ | ξ | 2 s 2 | v ^ 0 ( ξ ) | 2 ≤ 4 ( 1 + | ξ | 2 ) s | v ^ 0 ( ξ ) | 2 .
The right-hand side is integrable because v 0 ∈ H s . Dominated convergence and the Fourier characterization of H s therefore imply ∥ e − t A v 0 − e − τ A v 0 ∥ H s ⟶ 0 . Since τ ∈ [ 0 , T ] is arbitrary, e − t A v 0 ∈ C ( [ 0 , T ] , H s ) .
Let z ( t ) = ∫ 0 t e − ( t − r ) A v p ( x , r ) d r and extend the restriction of v p ( x , r ) to ( 0 , T ) by zero outside ( 0 , T ) . Taking the Fourier transform in x and multiplying by λ = | ξ | 2 s , we obtain
λ z ^ ( ξ , t ) = ∫ − ∞ t λ e − λ ( t − r ) v p ^ ( ξ , r ) d r ( 0 < t < T ) ,
where extending the integral to r = − ∞ does not change its value because v p ( x , r ) = 0 for r < 0 .
The time-convolution kernel λ e − λ t 1 { t > 0 } has L 1 ( R ) norm at most one and vanishes when λ = 0 . Young’s inequality in time and Plancherel in space therefore yield
∥ A z ∥ L 2 ( 0 , T , L 2 ) ≤ ∥ v p ( x , t ) ∥ L 2 ( 0 , T , L 2 ) .
Since z t = v p − A z in distributions, we also have z t ∈ L 2 ( 0 , T , L 2 ) . Repeating for z the argument used above to prove f ∈ C ( J , D ( A 1 2 ) ) for f = ϕ u , with J = [ 0 , T ] , yields z ∈ C ( [ 0 , T ] , D ( A 1 2 ) ) . Since v ( t ) = e − t A v 0 + z ( t ) , the preceding estimates for A e − t A v 0 , A z , and the time derivatives, together with the H s continuity of both summands, yield
v ∈ L 2 ( 0 , T , D ( A ) ) ∩ W 1 , 2 ( 0 , T , L 2 ) ∩ C ( [ 0 , T ] , H s ) .
Define
E ( v ) : = 1 2 E s ( v ) − 1 p + 1 ∥ v ∥ p + 1 p + 1 .
Since 0 ≤ v ≤ M , (70) yields
∫ R n v ( x , t ) p + 1 d x ≤ M p − 1 ∥ v ( t ) ∥ 2 2 < ∞ ( t ≥ 0 ) .
Then E ( v ) is well defined. The regularity estimates for v , A v and v t ensure the absolute continuity of E ( v ( t ) ) . Then by using v t + A v = v p , we obtain
d d t E ( v ( t ) ) = 〈 A v , v t 〉 − ∫ R n v p v t d x = 〈 A v − v p , v t 〉 = − ∥ v t ∥ 2 2 for a . e . t ∈ ( 0 , T ) .
Consequently,
E ( v ( t ) ) + ∫ 0 t ∥ v τ ( τ ) ∥ 2 2 d τ = E ( v 0 ) ( 0 ≤ t ≤ T ) .
Assume for contradiction that E ( v 0 ) = − e 0 < 0 , and set
Y ( t ) = 1 2 ∥ v ( t ) ∥ 2 2 , H ( t ) = ∫ 0 t Y ( τ ) d τ , D ( t ) = ∫ 0 t ∥ v τ ( τ ) ∥ 2 2 d τ .
Using v t = − A v + v p and E ( v ( t ) ) = − e 0 − D ( t ) , we obtain for almost every t,
Y ′ ( t ) = 〈 v , v t 〉 = − E s ( v ) + ∥ v ∥ p + 1 p + 1 = p − 1 2 E s ( v ) + ( p + 1 ) ( e 0 + D ( t ) ) ≥ ( p + 1 ) ( e 0 + D ( t ) ) .
Thus Y ( t ) ≥ Y ( 0 ) + ( p + 1 ) e 0 t , so Y ( t ) → ∞ . Note that Cauchy–Schwarz also yields
( Y ( t ) − Y ( 0 ) ) 2 = ∫ 0 t 〈 v , v τ 〉 d τ 2 ≤ ∫ 0 t ∥ v ∥ 2 2 d τ ∫ 0 t ∥ v τ ∥ 2 2 d τ = 2 H ( t ) D ( t ) .
Since H ′ = Y and H ′ ′ = Y ′ almost everywhere, (71)–(72) imply
H H ′ ′ ≥ ( p + 1 ) e 0 H + ( p + 1 ) H D ≥ ( p + 1 ) e 0 H + p + 1 2 ( H ′ − Y ( 0 ) ) 2 .
Choose 0 < δ < p − 1 2 . Because H ′ ( t ) = Y ( t ) → ∞ , there is t 1 > 0 such that
p + 1 2 1 − Y ( 0 ) H ′ ( t ) 2 ≥ 1 + δ ( t ≥ t 1 ) .
Hence
H H ′ ′ ≥ ( 1 + δ ) ( H ′ ) 2 for a . e . t ≥ t 1 .
Since H , H ′ > 0 for t > 0 , the function G = H − δ satisfies
G ′ = − δ H − δ − 1 H ′ < 0 , G ′ ′ = δ H − δ − 2 ( 1 + δ ) ( H ′ ) 2 − H H ′ ′ ≤ 0
on [ t 1 , ∞ ) , with the second inequality understood almost everywhere. Thus G is concave and
G ( t ) ≤ G ( t 1 ) + G ′ ( t 1 ) ( t − t 1 ) ( t ≥ t 1 ) .
The right-hand side becomes nonpositive at a finite time. However, (70) yields
0 < H ( t ) ≤ t 2 sup 0 ≤ τ ≤ t ∥ v ( τ ) ∥ 2 2 < ∞ ( 0 < t < ∞ ) ,
so G ( t ) = H ( t ) − δ > 0 at every finite positive time. This contradiction proves E ( v 0 ) ≥ 0 , which is (69). □
Remark 4.4. 
To control the term − F η ( σ 2 ) [ w ] in (67), define the transition potential
T η [ w ] : = 1 p + 1 ∫ R n ( η 2 − η p + 1 ) w p + 1 d y ≥ 0 .
Its integrand is nonnegative and vanishes where η = 0 or 1. At each similarity time, (56)–(57) and Lemma 4.3 yield
r − μ F η [ w ] + T η [ w ] = 1 2 E s ( ϕ u ( t ) ) − 1 p + 1 ∫ R n ( ϕ u ( t ) ) p + 1 d x ≥ 0 .
This yields the required lower bound
F η [ w ] ≥ − T η [ w ] .

4.2. An Initial Local Estimate for Z [ w ]

Fix a finite interval I and a reference time σ * . Choose fixed cutoffs 0 ≤ η 0 , * , η 1 , * ≤ 1 in C c ∞ ( R n ) and define the material cutoffs
η i ( y , σ ) : = η i , * e − σ − σ * 2 s y , i = 0 , 1 .
Assume that η 1 = 1 on a uniform neighborhood of supp η 0 for σ ∈ I . For f = f ( y , σ ) , let f σ : = f ( · , σ ) and η i , σ : = η i ( · , σ ) .
Lemma 4.5. 
With the above notation, let f : R n × I → [ 0 , ∞ ) be a measurable function and satisfy sup σ ∈ I M q [ f σ ] ≤ K . At every σ ∈ I for which η 1 , σ f σ ∈ H s ( R n ) , the commutator C η 0 , σ [ f σ ] = [ ( − Δ ) s , η 0 , σ ] f σ has an L 2 -representative on supp η 0 , σ and satisfies
∥ C η 0 , σ [ f σ ] ∥ L 2 ( supp η 0 , σ ) 2 ≤ C I K 2 + ∥ η 1 , σ f σ ∥ H s 2 ϑ s ∥ η 1 , σ f σ ∥ 2 2 ( 1 − ϑ s ) ,
where ϑ s : = ( 2 s − 1 ) + s ∈ [ 0 , 1 ) and the constant C I depends only on n , s , p , the fixed cutoffs η 0 , * , η 1 , * , and sup σ ∈ I | σ − σ * | . 
Proof. 
Fix σ ∈ I such that η 1 , σ f σ ∈ H s ( R n ) , and suppress σ in f and η i . Set λ = e σ − σ * 2 s , so that η i ( y ) = η i , * ( y λ ) . Consider the decomposition f = η 1 f + ( 1 − η 1 ) f = : g + f ♯ . Since η 1 = 1 on a uniform neighborhood of supp η 0 for σ ∈ I , we have η 0 f ♯ ≡ 0 on R n . Therefore, C η 0 [ f ♯ ] = − η 0 ( − Δ ) s f ♯ . Applying Lemma 2.3 with scale λ , we deduce
∥ C η 0 [ f ♯ ] ∥ L ∞ ( supp η 0 ) ≤ C K λ − q − 2 s .
Let L I : = sup σ ∈ I | σ − σ * | , since λ = e σ − σ * 2 s , we have e − L I 2 s ≤ λ ≤ e L I 2 s . Since | supp η 0 | = λ n | supp η 0 , * | , it follows that
∥ C η 0 [ f ♯ ] ∥ L 2 ( supp η 0 ) 2 ≤ C K 2 λ n − 2 q − 4 s ≤ C K 2 exp | n − 2 q − 4 s | L I 2 s .
We next estimate C η 0 [ g ] , where g = η 1 f ∈ H s ( R n ) . First assume g ∈ C c ∞ ( R n ) . Applying the Fourier transform to ( − Δ ) s ( η 0 g ) − η 0 ( − Δ ) s g , we have
C η 0 [ g ] ^ ( ξ ) = 1 ( 2 π ) n ∫ R n η 0 ^ ( ξ − ζ ) | ξ | 2 s − | ζ | 2 s g ^ ( ζ ) d ζ .
For ℓ ∈ { 1 , 2 s } , since e − L I 2 s ≤ λ ≤ e L I 2 s and η 0 , * ∈ C c ∞ ( R n ) , we have
∫ R n | h | ℓ | η 0 ^ ( h ) | d h = λ − ℓ ∫ R n | z | ℓ | η 0 , * ^ ( z ) | d z ≤ C I .
If 0 < 2 s ≤ 1 , then | ξ | 2 s − | ζ | 2 s ≤ | ξ − ζ | 2 s . Thus (76) implies | C η 0 [ g ] ^ | ≤ C | · | 2 s | η 0 ^ | * | g ^ | . Young’s inequality and Plancherel’s theorem yield
∥ C η 0 [ g ] ∥ 2 ≤ C ∥ | · | 2 s η 0 ^ ∥ 1 ∥ g ^ ∥ 2 ≤ C I ∥ g ∥ 2 .
If 1 < 2 s < 2 , by the mean value theorem, we have
| ξ | 2 s − | ζ | 2 s ≤ C | ξ − ζ | | ζ | 2 s − 1 + | ξ − ζ | 2 s .
Consequently,
| C η 0 [ g ] ^ | ≤ C | · | | η 0 ^ | * | · | 2 s − 1 | g ^ | + C | · | 2 s | η 0 ^ | * | g ^ | .
Applying Young’s inequality and Plancherel’s theorem again, we obtain
∥ C η 0 [ g ] ∥ 2 ≤ C I ∥ | · | 2 s − 1 g ^ ∥ 2 + ∥ g ^ ∥ 2 ≤ C I ∥ g ∥ H 2 s − 1 ≤ C I ∥ g ∥ H s .
The last inequality follows from H s ( R n ) ↪ H 2 s − 1 ( R n ) , since 2 s − 1 < s .
We now extend these estimates to g ∈ H s ( R n ) . Choose g j ∈ C c ∞ ( R n ) with g j → g in H s . Applying (77) or (78) to g j − g k shows
∥ C η 0 [ g j ] − C η 0 [ g k ] ∥ 2 ≤ C I ∥ g j − g k ∥ H s .
Hence C η 0 [ g j ] → G in L 2 for some G. For every φ ∈ C c ∞ ( R n ) ,
〈 C η 0 [ g j − g ] , φ 〉 = ∫ R n ( g j − g ) η 0 ( − Δ ) s φ − ( − Δ ) s ( η 0 φ ) d y ⟶ 0 ,
because η 0 ( − Δ ) s φ − ( − Δ ) s ( η 0 φ ) ∈ L 2 and g j → g in L 2 . Thus G represents C η 0 [ g ] . Passing to the limit yields (77)–(78) for g. For s > 1 2 , the identity 2 s − 1 = ϑ s s and Hölder’s inequality yield
∥ g ∥ H 2 s − 1 2 ≤ C ∫ R n ( 1 + | ξ | 2 ) s | g ^ ( ξ ) | 2 ϑ s | g ^ ( ξ ) | 2 1 − ϑ s d ξ ≤ C ∥ g ∥ H s 2 ϑ s ∥ g ∥ 2 2 ( 1 − ϑ s ) .
For s ≤ 1 2 , (77) supplies the corresponding L 2 bound. Combining these local estimates with (75), we obtain
∥ C η 0 [ f ] ∥ L 2 ( supp η 0 ) 2 ≤ 2 ∥ C η 0 [ g ] ∥ 2 2 + 2 ∥ C η 0 [ f ♯ ] ∥ L 2 ( supp η 0 ) 2 ≤ C I K 2 + ∥ g ∥ H s 2 ϑ s ∥ g ∥ 2 2 ( 1 − ϑ s ) .
Here the product of norms is understood as ∥ g ∥ 2 2 when s ≤ 1 2 . Substituting g = η 1 f proves (74). Hence we complete the proof of Lemma 4.5. □
Lemma 4.6. 
Let 0 < q ≤ n and I ⊂ R be an open interval containing the compact interval [ a , b ] . Suppose that 0 ≤ ζ ≤ 1 and ζ ∈ C c ∞ ( B R ) . Let w : R n × I → [ 0 , ∞ ) satisfy (53) pointwise, with
w , w σ ∈ C ( R n × I ) , w ( · , σ ) ∈ C l o c 1 , 1 ( R n ) ∩ L 2 s ( R n ) for every σ ∈ I .
Suppose that, for every L > 0 and every compact interval J ⋐ I ,
sup σ ∈ J ∥ w ( · , σ ) ∥ C 1 , 1 ( B L ¯ ) < ∞ and sup σ ∈ [ a , b ] M q [ w ( · , σ ) ] ≤ K .
Then there exists a positive constant C, depending only on n , s , p , K , ζ and the positive distance dist ( supp ζ , ∂ B R ) , such that
∫ a b E s ( ζ w ) d σ ≤ C 1 + ( b − a ) + ∫ R n ζ ( y ) 2 w ( y , a ) 2 d y + ∫ a b ∫ B R ( w 2 + w p + 1 ) d y d σ .
Proof. 
Since ζ ∈ C c ∞ ( B R ) and sup σ ∈ J ∥ w ( · , σ ) ∥ C 1 , 1 ( B L ¯ ) < ∞ for every L > 0 and every compact interval J ⋐ I , we have sup σ ∈ [ a , b ] ∥ ζ w ( · , σ ) ∥ H 2 ( R n ) < ∞ . In particular, ζ w ( · , σ ) ∈ H s ( R n ) for every σ ∈ [ a , b ] . Moreover, the uniform H 2 bound and Plancherel’s identity yield that
E s ( ζ w ) = ∥ ( − Δ ) s 2 ( ζ w ) ^ ∥ 2 2 = C ∫ R n | ξ | 2 s | ζ w ^ ( ξ ) | 2 d ξ ≤ C sup σ ∈ [ a , b ] ∥ ζ w ( · , σ ) ∥ H 2 2 < ∞ .
For notational simplicity, fix σ ∈ [ a , b ] and let w ( y ) denote w ( y , σ ) in the spatial calculations below. For fixed ε > 0 , let A ε , E s , ε and R ζ , ε be the truncated operators used in Lemma 2.2. By calculations similar to those in the proof of Lemma 2.2, we obtain
∫ R n ζ 2 w A ε w d y = E s , ε ( w , ζ 2 w ) = E s , ε ( ζ w ) − R ζ , ε [ w ] .
Since sup σ ∈ J ∥ w ( · , σ ) ∥ C 1 , 1 ( B L ¯ ) < ∞ for every L > 0 and every compact interval J ⋐ I , we have
2 w ( y , σ ) − w ( y + h , σ ) − w ( y − h , σ ) = O ( | h | 2 )
uniformly for y ∈ supp ζ , σ ∈ [ a , b ] , and small h. For each σ ∈ [ a , b ] , the hypotheses of Lemma 2.2 hold with f = w ( · , σ ) and χ = ζ . By (82) we deduce A ε w → ( − Δ ) s w uniformly on supp ζ × [ a , b ] . Thus, passing to the limit in (81) and using Lemma 2.2, we obtain
∫ R n ζ 2 w ( − Δ ) s w d y = lim ε ↓ 0 E s , ε ( w , ζ 2 w ) = E s ( ζ w ) − R ζ [ w ] , σ ∈ [ a , b ] .
Since ζ ( y ) − ζ ( z ) = 0 on ( supp ζ ) c × ( supp ζ ) c , by symmetry we only need to estimate
∫ ∫ supp ζ × R n w ( y ) w ( z ) ( ζ ( y ) − ζ ( z ) ) 2 | y − z | n + 2 s d y d z .
For z ∈ B R , we have
| ζ ( y ) − ζ ( z ) | 2 ≤ C min { 1 , | y − z | 2 } .
Note that min { 1 , | h | 2 } | h | − n − 2 s belongs to L 1 ( R n ) because 0 < s < 1 . Hence 2 w ( y ) w ( z ) ≤ w ( y ) 2 + w ( z ) 2 yields
∫ ∫ supp ζ × B R w ( y ) w ( z ) ( ζ ( y ) − ζ ( z ) ) 2 | y − z | n + 2 s d y d z ≤ C ∫ B R w 2 d y .
If y ∈ supp ζ and z ∉ B R , set d : = dist ( supp ζ , ∂ B R ) > 0 , then | y − z | ≥ d . Since sup σ ∈ [ a , b ] M q [ w ( · , σ ) ] ≤ K , we have
∫ R n ∖ B R w ( z ) | y − z | n + 2 s d z ≤ ∑ j = 0 ∞ ( 2 j d ) − n − 2 s ∫ B 2 j + 1 d ( y ) w ( z ) d z ≤ C K d − q − 2 s ∑ j = 0 ∞ 2 − j ( q + 2 s ) ≤ C K ,
uniformly for y ∈ supp ζ and σ ∈ [ a , b ] . For y ∈ supp ζ and z ∉ B R , we have ( ζ ( y ) − ζ ( z ) ) 2 = ζ ( y ) 2 ≤ 1 . Thus,
∫ ∫ supp ζ × ( R n ∖ B R ) w ( y ) w ( z ) ( ζ ( y ) − ζ ( z ) ) 2 | y − z | n + 2 s d y d z ≤ C K ∫ supp ζ w d y ≤ C K | supp ζ | 1 2 ∫ B R w 2 d y 1 2 ≤ C 1 + ∫ B R w 2 d y .
Combining (84) with (85) and using w ≥ 0 , we obtain
0 ≤ R ζ [ w ] ≤ C 1 + ∫ B R w 2 d y .
Set H ( σ ) : = ∫ R n ζ ( y ) 2 w ( y , σ ) 2 d y . Since ζ is independent of time and w , w σ are continuous on supp ζ × [ a , b ] , we have H ∈ C 1 ( [ a , b ] ) and
∫ R n ζ 2 w w σ d y = 1 2 H ′ ( σ ) .
Recall that D w = 1 2 s y · ∇ w + α w , integrating by parts we have
∫ R n ζ 2 w D w d y = α H ( σ ) + 1 4 s ∫ R n ζ 2 y · ∇ ( w 2 ) d y = α − n 4 s H ( σ ) − 1 2 s ∫ R n ζ ( y ) ( y · ∇ ζ ) w 2 d y ≥ − α − n 4 s ∫ B R w 2 d y − 1 2 s ∥ ζ ( y ) ( y · ∇ ζ ) ∥ ∞ ∫ B R w 2 d y ≥ − C ∫ B R w 2 d y ,
where we have used the fact that ζ ∈ C c ∞ ( B R ) to eliminate the boundary term.
Multiplying (53) by ζ 2 w and integrating by parts, then using (83), (87) and (86), we obtain
1 2 H ′ ( σ ) + E s ( ζ w ) = ∫ R n ζ 2 w p + 1 d y + R ζ [ w ] − ∫ R n ζ 2 w D w d y ≤ ∫ B R w p + 1 d y + C ∫ B R w 2 d y + C .
Since H ∈ C 1 ( [ a , b ] ) , E s ( ζ w ) ∈ L 1 ( a , b ) by (80), and w is bounded on B R ¯ × [ a , b ] , we may integrate (88) over [ a , b ] to obtain
1 2 H ( b ) + ∫ a b E s ( ζ w ) d σ ≤ 1 2 H ( a ) + C ( b − a ) + C ∫ a b ∫ B R ( w 2 + w p + 1 ) d y d σ .
Since H ( b ) 2 ≥ 0 and H ( a ) = ∫ R n ζ ( y ) 2 w ( y , a ) 2 d y , we obtain (79). □
Proposition 4.7. 
Let u be a classical entire solution of (1) satisfying
0 ≤ u ≤ 1 , sup t ∈ R M q [ u ( · , t ) ] ≤ K .
For k ≥ 2 , set s k : = − log k . Then, for every fixed T − , T + , R > 0 , there exists a constant C = C T − , T + , R , K > 0 such that
∫ s k − T − s k + T + ∫ B R | Z [ w a , k ] | 2 d y d σ ≤ C k α for all a ∈ R n and k ≥ 2 .
Proof. 
For every fixed k ≥ 2 and T − , T + , R > 0 , let A k = s k − T − , B k = s k + T + , J k = [ A k − 3 , B k + 3 ] . Then for σ ∈ J k , we have e − T + − 3 k ≤ r = k e − ( σ − s k ) ≤ e T − + 3 k , 0 ≤ w = r α u ≤ C k α . Fix R * > 0 large enough. Choose material cutoff functions η 0 , η 1 and a time-independent function ζ ∈ C c ∞ ( B R * ) with 0 ≤ ζ ≤ 1 , such that for every σ ∈ [ A k − 2 , B k + 2 ] ,
B R ⋐ { y : η 0 ( y , σ ) = 1 } ⋐ supp η 0 ( · , σ ) , supp η 0 ( · , σ ) ⋐ { y : η 1 ( y , σ ) = 1 } ⋐ supp η 1 ( · , σ ) , supp η 1 ( · , σ ) ⋐ { y : ζ ( y ) = 1 } ⋐ supp ζ ⋐ B R * .
Since ρ s is continuous and strictly positive, c R * : = min B R * ¯ ρ s > 0 . Thus (60) implies
∫ J k ∫ B R * w p d y d σ ≤ c R * − 1 ∫ J k M p ( σ ) d σ ≤ C ,
where C is independent of a and k. Since w ≤ C k α , we have
∫ J k ∫ B R * w p + 1 ≤ C k α .
If p ≥ 2 , use w 2 ≤ 1 + w p and (90), we have ∫ J k ∫ B R * w 2 d y d σ ≤ ∫ J k ∫ B R * ( 1 + w p ) d y d σ ≤ C . Since k ≥ 2 and α > 0 , we have k α ≥ 1 , therefore ∫ J k ∫ B R * w 2 ≤ C k α . If 1 < p < 2 , by w ≤ C k α we have w 2 ≤ ( C k α ) 2 − p w p . Then it follows from (90), 0 < 2 − p < 1 and k ≥ 2 that ∫ J k ∫ B R * w 2 d y d σ ≤ C k α ( 2 − p ) ∫ J k ∫ B R * w p d y d σ ≤ C k α . Thus, in both cases, we have
∫ J k ∫ B R * ( w 2 + w p + 1 ) d y d σ ≤ C k α .
By (91), we have ∫ A k − 3 A k − 2 ∫ B R * ( w 2 + w p + 1 ) d y d σ ≤ C k α . Then by the mean value theorem for integrals, there exists σ 0 ∈ [ A k − 3 , A k − 2 ] such that
∫ B R * w ( y , σ 0 ) 2 d y ≤ C k α .
By applying Lemma 4.6 with [ a , b ] = [ σ 0 , B k + 2 ] , we have
∫ σ 0 B k + 2 E s ( ζ w ) d σ ≤ C 1 + ( B k + 2 − σ 0 ) + ∫ R n ζ 2 w ( σ 0 ) 2 d y + ∫ σ 0 B k + 2 ∫ B R * ( w 2 + w p + 1 ) d y d σ .
Since supp ζ ⊂ B R * and 0 ≤ ζ ≤ 1 , we have ∫ R n ζ 2 w ( σ 0 ) 2 d y ≤ ∫ B R * w ( σ 0 ) 2 d y ≤ C k α . Note that B k + 2 − σ 0 ≤ | J k | ≤ T − + T + + 6 , we deduce 1 + ( B k + 2 − σ 0 ) ≤ C k α since k ≥ 2 and α > 0 . Therefore, by (91) we deduce ∫ σ 0 B k + 2 E s ( ζ w ) d σ ≤ C k α . Adding the L 2 bound from (91) and using σ 0 ≤ A k − 2 , we obtain
∫ A k − 2 B k + 2 ∥ ζ w ∥ H s 2 d σ ≤ C k α .
By (89) we have η i w = η i ( ζ w ) for i = 0 , 1 . Since 0 ≤ η i ≤ 1 and ∥ ∇ y η i ( · , σ ) ∥ ∞ ≤ M , with M independent of σ and k, then we have
∥ η i w ∥ H s = ∥ η i ( ζ w ) ∥ H s ≤ C ∥ ζ w ∥ H s , i = 0 , 1 , σ ∈ [ A k − 2 , B k + 2 ] .
Consequently, (92) yields
∫ A k − 2 B k + 2 ∥ η 1 w ∥ H s 2 + ∥ η 1 w ∥ 2 2 d σ ≤ C k α .
For 0 < ϑ s < 1 , Hölder’s inequality and (93) yield
∫ A k − 2 B k + 2 ∥ η 1 w ∥ H s 2 ϑ s ∥ η 1 w ∥ 2 2 ( 1 − ϑ s ) d σ ≤ ∫ A k − 2 B k + 2 ∥ η 1 w ∥ H s 2 d σ ϑ s ∫ A k − 2 B k + 2 ∥ η 1 w ∥ 2 2 d σ 1 − ϑ s ≤ C k α .
When ϑ s = 0 , ∥ η 1 w ∥ H s 2 ϑ s ∥ η 1 w ∥ 2 2 ( 1 − ϑ s ) is understood as ∥ η 1 w ∥ 2 2 , by (93), the same bound holds. Applying Lemma 4.5 with f σ = w , then integrating (74), we deduce that
∫ A k − 2 B k + 2 ∥ C η 0 [ w ] ∥ L 2 ( supp η 0 ) 2 d σ ≤ C k α ,
where we have used K 2 ( T − + T + + 4 ) ≤ C k α .
By (92) and the mean value theorem for integrals, there exists σ − ∈ [ A k − 2 , A k − 1 ] such that ∥ ζ w ( · , σ − ) ∥ H s 2 ≤ C k α . Then by the definition of F η 0 [ w ] , we get
F η 0 ( σ − ) [ w ] ≤ 1 2 E s η 0 ( · , σ − ) w ( · , σ − ) ≤ C ∥ ζ w ( · , σ − ) ∥ H s 2 ≤ C k α .
Similarly, by (91) and the mean value theorem for integrals, there exists σ + ∈ [ B k + 1 , B k + 2 ] such that
∫ B R * w ( y , σ + ) p + 1 d y ≤ C k α .
Since 0 ≤ η 0 2 − η 0 p + 1 ≤ 1 , the definition of T η 0 and (73) yield
T η 0 ( σ + ) [ w ] ≤ C k α , F η 0 ( σ + ) [ w ] ≥ − T η 0 ( σ + ) [ w ] ≥ − C k α .
By the choice of σ − and σ + ,
[ A k , B k ] ⊂ [ σ − , σ + ] ⊂ [ A k − 2 , B k + 2 ] , σ + − σ − ≤ L : = T − + T + + 4 .
Hence
e − | μ | L ≤ e μ ( σ − σ + ) ≤ e | μ | L ( σ ∈ [ σ − , σ + ] ) .
Applying Proposition 4.2 with η = η 0 , using (94), (95) and (96) in (67), we conclude
∫ σ − σ + e μ ( σ − σ + ) ∫ R n η 0 2 | Z [ w ] | 2 d y d σ ≤ C k α .
Finally, η 0 ( y , σ ) = 1 on B R × [ A k , B k ] . By (97) we have
∫ A k B k ∫ B R | Z [ w ] | 2 d y d σ ≤ e | μ | L ∫ σ − σ + e μ ( σ − σ + ) ∫ R n η 0 2 | Z [ w ] | 2 d y d σ ≤ C k α .
All constants are independent of a and k, which proves Proposition 4.7. □

5. Weak- L 2 p Bounds and Energy Improvement

For b ∈ R n , σ ∈ R , and r , d , L > 0 , define
Q ( b , σ , r , d ) : = ( y , θ ) ∈ R n × R : σ − d ≤ θ ≤ σ , y − e θ − σ 2 s b ≤ r ,
and Q L : = B L × [ − L 2 s , 0 ] . Throughout this section, assume p F ( n , s ) < p < p S ( n , s ) , so that q = 2 s α ∈ ( 0 , n ) and μ > 0 .

5.1. A Defect Lower Bound for Normalized Solutions

The following lemma establishes local uniform convergence up to τ = 0 , so that the normalization v j ( 0 , 0 ) = 1 used in the proof of Lemma 5.2 passes to the limit.
Lemma 5.1. 
Fix 0 < q ≤ n and 0 < a * ≤ a * < ∞ , and let T j > 0 satisfy T j → ∞ . Suppose that 0 ≤ v j ∈ C ( R n × ( − T j , 0 ] ) are weak solutions of
( v j ) τ + a j ( τ ) ( − Δ ) s v j = f j in R n × ( − T j , 0 ) ,
where a j ∈ C ( ( − T j , 0 ] ) and a * ≤ a j ≤ a * . 
Assume that sup j sup τ ∈ ( − T j , 0 ] M q [ v j ( · , τ ) ] ≤ K , and for every R , T > 0 , there exist j 0 = j 0 ( R , T ) ∈ N and C R , T > 0 such that, for all j ≥ j 0 , T j > 2 T and
∥ v j ∥ L ∞ ( B 2 R × [ − 2 T , 0 ] ) + ∥ f j ∥ L ∞ ( B 2 R × [ − 2 T , 0 ] ) ≤ C R , T .
Then, after passing to a subsequence, there exists a continuous function V : R n × ( − ∞ , 0 ] → [ 0 , ∞ ) such that
v j ⟶ V locally uniformly on R n × ( − ∞ , 0 ] .
Moreover, for every ψ ∈ C c ∞ ( R n × ( − ∞ , 0 ) ) ,
∫ − T j 0 ∫ R n v j ( − Δ ) s ψ d x d τ ⟶ ∫ − ∞ 0 ∫ R n V ( − Δ ) s ψ d x d τ .
Proof. 
Fix R , T > 0 , by (99), there exist j 0 = j 0 ( R , T ) and C R , T such that, for every j ≥ j 0 , T j > 2 T and
∥ v j ∥ L ∞ ( B 4 R × [ − 2 T , 0 ] ) + ∥ f j ∥ L ∞ ( B 4 R × [ − 2 T , 0 ] ) ≤ C R , T .
Choose θ 0 , θ 1 ∈ C c ∞ ( R n ) with 0 ≤ θ 0 , θ 1 ≤ 1 and
θ 0 = 1 on B 2 , supp θ 0 ⋐ B 3 , θ 1 = 1 on B 3 , supp θ 1 ⋐ B 4 .
Set
χ i , R ( x ) : = θ i ( x R ) , h j : = χ 1 , R v j , v j , R ♯ : = ( 1 − χ 1 , R ) v j .
Since supp χ 1 , R ⋐ B 4 R , we have ∥ h j ∥ L ∞ ( R n × [ − 2 T , 0 ] ) ≤ C R , T . Moreover, χ 1 , R = 1 on B 3 R , so h j is a weak solution of
( h j ) τ + a j ( τ ) ( − Δ ) s h j = f j + a j ( τ ) ( − Δ ) s ( ( χ 1 , R − 1 ) v j ) = : f ˜ j in B 2 R × ( − 2 T , 0 ) .
Since f j is uniformly bounded on B 2 R × [ − 2 T , 0 ] and 0 < a j ≤ a * , it remains to bound ( − Δ ) s ( ( χ 1 , R − 1 ) v j ) uniformly there to obtain a uniform bound for f ˜ j . Since χ 1 , R = 1 on B 3 R ⊃ supp χ 0 , R , we have χ 0 , R v j , R ♯ ≡ 0 on R n . Since χ 0 , R = 1 on B 2 R , we have
v j , R ♯ = ( − Δ ) s χ 0 , R v j , R ♯ − χ 0 , R ( − Δ ) s v j , R ♯ = − χ 0 , R ( − Δ ) s v j , R ♯ = − ( − Δ ) s v j , R ♯ = ( − Δ ) s ( ( χ 1 , R − 1 ) v j ) on B 2 R .
Applying (14) to f = v j ( · , τ ) with λ = R , and using sup j sup τ ∈ ( − T j , 0 ] M q [ v j ( · , τ ) ] ≤ K , yields
sup j ≥ j 0 sup τ ∈ [ − 2 T , 0 ] sup x ∈ B 2 R ( − Δ ) s ( ( χ 1 , R − 1 ) v j ) ( x , τ ) ≤ C K R − q − 2 s .
Together with a j ≤ a * and (99), this yields
∥ f ˜ j ∥ L ∞ ( B 2 R × ( − 2 T , 0 ) ) ≤ C R , T , j ≥ j 0 .
Fix 0 < β < min { 1 , 2 s } . Apply (25) in Proposition 2.6 to (101), with b = 0 . Since h j = v j on B R × [ − T , 0 ] , we obtain
| v j ( x , τ ) − v j ( x ′ , τ ′ ) | ≤ C R , T | x − x ′ | + | τ − τ ′ | 1 2 s β
for x , x ′ ∈ B R and − T < τ , τ ′ < 0 . The constant is independent of j and of the distances of τ , τ ′ from zero. By continuity of v j , (102) holds on B R ¯ × [ − T , 0 ] . Thus { v j } is uniformly bounded and equicontinuous on every such compact set B R ¯ × [ − T , 0 ] . By the Arzelà-Ascoli theorem and a diagonal argument, there exists a nonnegative function V ∈ C ( R n × ( − ∞ , 0 ] ) and a subsequence, still denoted by v j , such that
v j ⟶ V locally uniformly on R n × ( − ∞ , 0 ] .
Finally, let ψ ∈ C c ∞ ( R n × ( − ∞ , 0 ) ) . Choose a bounded open interval I 0 such that I 0 ⋐ ( − ∞ , 0 ) and error ψ ⋐ R n × I 0 . For all sufficiently large j, we have I 0 ⊂ ( − T j , 0 ) . By applying Lemma 2.1 with I = I 0 , we have
lim j → ∞ ∫ − T j 0 ∫ R n v j ( − Δ ) s ψ d x d τ = lim j → ∞ ∫ I 0 ∫ R n v j ( − Δ ) s ψ d x d τ = ∫ I 0 ∫ R n V ( − Δ ) s ψ d x d τ = ∫ − ∞ 0 ∫ R n V ( − Δ ) s ψ d x d τ .
This proves (100). □
Lemma 5.2. 
For any fixed K > 0 , there exist L 0 > 4 , c 0 > 0 , and ε 0 > 0 , depending only on n , s , p , K , with the following property. Suppose 0 < ε ≤ ε 0 and v : R n × ( − ( 2 L 0 ) 2 s , 0 ] → [ 0 , ∞ ) satisfies the following equation pointwise
v τ + ( − Δ ) s v + ε 2 s 1 2 s z · ∇ v + α v = v p in R n × ( − ( 2 L 0 ) 2 s , 0 ] .
Assume further that v satisfies
v , v τ ∈ C ( R n × ( − ( 2 L 0 ) 2 s , 0 ] ) , v ( · , τ ) ∈ C l o c 1 , 1 ( R n ) ∩ L 2 s ( R n ) for every τ ∈ ( − ( 2 L 0 ) 2 s , 0 ] ,
and
sup τ ∈ J ′ ∥ v ( · , τ ) ∥ C 1 , 1 ( B L ¯ ) < ∞ , ∥ v ∥ L ∞ ( R n × J ′ ) < ∞ for every L > 0 , J ′ ⋐ ( − ( 2 L 0 ) 2 s , 0 ] .
If, in addition, v ( 0 , 0 ) = 1 , 0 ≤ v ≤ 2 on Q 2 L 0 , sup τ ∈ ( − ( 2 L 0 ) 2 s , 0 ] M q [ v ( · , τ ) ] ≤ K , then
∫ Q L 0 v p − ( − Δ ) s v 2 d z d τ ≥ c 0 .
Proof. 
We prove by contradiction. Suppose that for every L 0 > 4 , c 0 > 0 , and ε 0 > 0 , there exist 0 < ε ≤ ε 0 and a solution v satisfying the hypotheses such that
∫ Q L 0 | v p − ( − Δ ) s v | 2 d z d τ < c 0 .
For each integer j > 4 , taking ( L 0 , c 0 , ε 0 ) = ( j , j − 1 , j − 1 ) yields 0 < ε j ≤ j − 1 and a bounded nonnegative classical solution v j of (103) on R n × ( − ( 2 j ) 2 s , 0 ] satisfying
v j ( 0 , 0 ) = 1 , 0 ≤ v j ≤ 2 on Q 2 j , sup τ ∈ ( − ( 2 j ) 2 s , 0 ] M q [ v j ( · , τ ) ] ≤ K , ∫ Q j | v j p − ( − Δ ) s v j | 2 < j − 1 .
Let A j = ε j 2 s and define
v ˜ j ( ζ , τ ) : = e α A j τ v j ( e A j τ 2 s ζ , τ ) .
Then we have
v j ( z , τ ) = e − α A j τ v ˜ j ( e − A j τ 2 s z , τ ) .
Let z = e A j τ 2 s ζ , then we have
∂ τ v ˜ j = e α A j τ ( v j ) τ + A j 1 2 s z · ∇ v j + α v j , ( − Δ ) ζ s v ˜ j = e ( α + 1 ) A j τ ( − Δ ) z s v j .
Since α p = α + 1 , by (103), v ˜ j satisfies ∂ τ v ˜ j + e − A j τ ( − Δ ) s v ˜ j = e − A j τ v ˜ j p . Let a j ( τ ) : = e − A j τ , since A j ≤ j − 2 s , for any τ ∈ ( − ( 2 j ) 2 s , 0 ] , there holds
1 ≤ a j ( τ ) ≤ e A j ( 2 j ) 2 s ≤ e 2 2 s .
Moreover, q = 2 s α implies that
M q [ v ˜ j ( · , τ ) ] = e ( α − q 2 s ) A j τ M q [ v j ( · , τ ) ] = M q [ v j ( · , τ ) ] ≤ K .
Fix R , T > 0 , for ζ ∈ B R and − T ≤ τ ≤ 0 , we have | e A j τ 2 s ζ | ≤ | ζ | < R . Thus, once 2 j > R and ( 2 j ) 2 s > T , we have ( e A j τ 2 s ζ , τ ) ∈ Q 2 j . Since 0 < e α A j τ ≤ 1 , (104) and (105) yield 0 ≤ v ˜ j ≤ 2 on B R × [ − T , 0 ] . Let f j : = e − A j τ v ˜ j p , we also have 0 ≤ f j ≤ e 2 2 s 2 p on B R × [ − T , 0 ] for all sufficiently large j. Lemma 5.1, applied with T j = ( 2 j ) 2 s , therefore implies that there exists a nonnegative function V ∈ C ( R n × ( − ∞ , 0 ] ) and a subsequence, still denoted by { v ˜ j } , such that
v ˜ j ⟶ V locally uniformly on R n × ( − ∞ , 0 ] .
Since A j → 0 , we have e − α A j τ → 1 and e − A j τ 2 s z → z uniformly on compact sets. Using v ˜ j → V locally uniformly and the continuity of V in (106), we obtain v j → V locally uniformly. In particular,
V ( 0 , 0 ) = 1 , 0 ≤ V ≤ 2 .
We next prove that V is independent of time. By (103) we have
( v j ) τ + A j 1 2 s z · ∇ v j + α v j = v j p − ( − Δ ) s v j .
Fix a bounded open interval I 0 ⋐ ( − ∞ , 0 ) and ψ ∈ C c ∞ ( R n × I 0 ) . For all large j, I 0 ⊂ ( − ( 2 j ) 2 s , 0 ) and supp ψ ⊂ Q j . Choosing ψ as a test function in (109), we obtain
− ∫ I 0 ∫ R n v j ψ τ d z d τ + A j ∫ I 0 ∫ R n v j α ψ − 1 2 s div ( z ψ ) d z d τ = ∫ I 0 ∫ R n v j p − ( − Δ ) s v j ψ d z d τ .
Since v j ≤ 2 on supp ψ and ψ ∈ C c ∞ ( R n × I 0 ) , there exists a positive constant C ψ , such that
A j ∫ I 0 ∫ R n v j α ψ − 1 2 s div ( z ψ ) d z d τ ≤ C ψ A j .
By (104), we have
∫ I 0 ∫ R n v j p − ( − Δ ) s v j ψ d z d τ ≤ j − 1 2 ∥ ψ ∥ 2 .
Thus by (111), (112) and letting j → ∞ , we arrive at
− ∫ I 0 ∫ R n V ψ τ d z d τ = 0 .
Since I 0 and ψ are arbitrary, ∂ τ V = 0 in distributions on R n × ( − ∞ , 0 ) . Continuity of V and (108) yield
V ( z , τ ) = V * ( z ) ( τ ≤ 0 ) , V * ( 0 ) = 1 .
To identify the equation for V * , choose ψ = φ θ , where φ ∈ C c ∞ ( R n ) and θ ∈ C c ∞ ( I 0 ) with ∫ I 0 θ d τ = 1 . For all sufficiently large j, supp ( φ θ ) ⊂ Q j . By (110), we have
− ∫ I 0 ∫ R n v j φ θ ′ d z d τ + A j ∫ I 0 ∫ R n v j α φ − 1 2 s div ( z φ ) θ d z d τ = ∫ I 0 ∫ R n v j p − ( − Δ ) s v j φ θ d z d τ .
By (104), the local classical regularity of v j and a similar argument in the proof of Lemma 2.2, we have
∫ I 0 ∫ R n v j p − ( − Δ ) s v j φ θ d z d τ = ∫ I 0 ∫ R n v j p φ − v j ( − Δ ) s φ θ d z d τ .
Thus by (111), (112) and ∫ I 0 θ = 1 , and letting j → ∞ , we deduce
∫ R n V * p φ d z = ∫ R n V * ( − Δ ) s φ d z
for any φ ∈ C c ∞ ( R n ) . Thus V * satisfies ( − Δ ) s V * = V * p in the sense of distributions in R n . Fatou’s lemma yields M q [ V * ] ≤ K , and Lemma 3.6 implies V * ∈ L 2 s ( R n ) . Define W ( z , t ) : = V * ( z ) for ( z , t ) ∈ R n × R . Since 0 ≤ V * ≤ 2 , W is a bounded nonnegative weak solution of (30). Proposition 2.7, together with V * ( 0 ) = 1 , shows that V * is a positive classical solution of the elliptic equation. Since 1 < p < p S ( n , s ) , the elliptic Liouville theorem in [9] (see also [4,11,18]) excludes such a solution. This contradiction completes the proof of Lemma 5.2. □

5.2. A Defect Lower Bound near Large Values of w

Lemma 5.3. 
Fix C 0 , K > 0 , and let L 0 > 4 , c 0 > 0 , and ε 0 > 0 be the constants provided by Lemma 5.2 for this K. Then there exist H 0 > 1 and C 1 > 0 , depending only on n , s , p , C 0 , K , with the following property. 
Let J ⊂ R be an open interval. Suppose that w : R n × J → [ 0 , ∞ ) satisfies (53) pointwise, with
w , w σ ∈ C ( R n × J ) , w ( · , σ ) ∈ C l o c 1 , 1 ( R n ) ∩ L 2 s ( R n ) for every σ ∈ J .
Assume that, for every L > 0 and every compact interval J ′ ⋐ J ,
sup σ ∈ J ′ ∥ w ( · , σ ) ∥ C 1 , 1 ( B L ¯ ) < ∞ , sup σ ∈ J M q [ w ( · , σ ) ] ≤ K ,
and
∥ w ∥ L ∞ ( R n × J ′ ) < ∞ for every compact interval J ′ ⋐ J .
For any k ≥ 2 , b ∈ R n , σ ∈ J , and H ≥ H 0 , suppose that
w ( b , σ ) ≥ H , [ σ − C 1 H − 1 α , σ ] ⋐ J ,
and
w ≤ C 0 k α on Q H : = Q b , σ , C 1 H − 1 q , C 1 H − 1 α .
Then there exists ( b ^ , σ ^ ) ∈ Q H with W : = w ( b ^ , σ ^ ) ≥ H such that
w ≤ 2 W on Q b ^ , σ ^ , 2 L 0 W − 1 q , ( 2 L 0 ) 2 s W − 1 α ⊂ Q H .
Proof. 
Set S t : = ∑ m = 0 ∞ 2 − m α , S x : = ∑ m = 0 ∞ 2 − m q , and define C t : = ( 2 L 0 ) 2 s ( S t + 1 ) , C x : = 2 L 0 e 1 4 s S x + 2 L 0 e 1 2 s . Choose C 1 ≥ max { C t , C x } and H 0 > 1 sufficiently large that
( 2 L 0 ) 2 s S t H 0 − 1 α ≤ 1 2 , C 1 H 0 − 1 α ≤ 1 .
We select points X j = ( b j , σ j ) inductively, starting from X 0 = ( b 0 , σ 0 ) : = ( b , σ ) . For each selected point X j , set A j : = w ( X j ) and T j : = Q b j , σ j , 2 L 0 A j − 1 q , ( 2 L 0 ) 2 s A j − 1 α . The construction will ensure that
A j ≥ 2 j H .
Suppose that X 0 , … , X j have been chosen, with X m + 1 ∈ T m and A m + 1 > 2 A m for 0 ≤ m < j . We first show that T j ⊂ Q H . By the definition of T m , (117) and (118), we have
0 ≤ σ − σ j = ∑ m = 0 j − 1 ( σ m − σ m + 1 ) ≤ ( 2 L 0 ) 2 s ∑ m = 0 j − 1 A m − 1 α ≤ ( 2 L 0 ) 2 s S t H − 1 α ≤ 1 2 .
For 0 ≤ m ≤ j , let ξ m : = e σ − σ m 2 s b m , then ξ 0 = b . Define δ m : = b m + 1 − e σ m + 1 − σ m 2 s b m . Since X m + 1 ∈ T m , we have
b m + 1 = e σ m + 1 − σ m 2 s b m + δ m , | δ m | ≤ 2 L 0 A m − 1 q .
Multiplying both sides of (120) by e σ − σ m + 1 2 s , by the definition of ξ m and using (119), we obtain
| ξ m + 1 − ξ m | = e σ − σ m + 1 2 s | δ m | ≤ 2 L 0 e 1 4 s A m − 1 q .
Consequently, (121) yields
| ξ j − b | ≤ ∑ m = 0 j − 1 | ξ m + 1 − ξ m | ≤ 2 L 0 e 1 4 s S x H − 1 q .
Now for any ( y , θ ) ∈ T j , by (119) we have
0 ≤ σ − θ ≤ ( σ − σ j ) + ( 2 L 0 ) 2 s A j − 1 α ≤ C 1 H − 1 α ≤ 1 ,
and consequently
e σ − θ 2 s y − ξ j = e σ − θ 2 s y − e θ − σ j 2 s b j ≤ 2 L 0 e 1 2 s A j − 1 q ≤ 2 L 0 e 1 2 s H − 1 q .
Combining this estimate with (122), we deduce
y − e θ − σ 2 s b = e θ − σ 2 s e σ − θ 2 s y − b ≤ C 1 H − 1 q .
Here we have used θ ≤ σ and C 1 ≥ C x . Together with σ − C 1 H − 1 α ≤ θ ≤ σ , this proves T j ⊂ Q H .
Since Q H ⊂ R n × J and T j ⊂ Q H , we know that w is defined throughout T j . If w ≤ 2 A j on T j , set ( b ^ , σ ^ ) : = ( b j , σ j ) , W : = w ( b ^ , σ ^ ) = A j . Then ( b ^ , σ ^ ) ∈ Q H , W ≥ H , and
T j = Q b ^ , σ ^ , 2 L 0 W − 1 q , ( 2 L 0 ) 2 s W − 1 α ⊂ Q H .
Thus (116) holds, and we stop. Otherwise, choose X j + 1 = ( b j + 1 , σ j + 1 ) ∈ T j such that
A j + 1 : = w ( X j + 1 ) > 2 A j .
Since T j ⊂ Q H and A j ≥ 2 j H , X j + 1 ∈ Q H , A j + 1 > 2 A j ≥ 2 j + 1 H . We then repeat the construction with X j + 1 and A j + 1 . The construction must stop after finitely many steps. Indeed, every selected point X j ∈ Q H , and hence
2 j H ≤ A j = w ( X j ) ≤ C 0 k α ,
which cannot hold for arbitrarily large j. Thus, there exists a finite index N, such that w ≤ 2 A N on T N . The point ( b ^ , σ ^ ) = X N and the value W = A N , selected at this step, satisfy (116). □
Lemma 5.4. 
Under the assumptions of Lemma 5.3, after increasing H 0 , still depending only on n , s , p , C 0 , K , if necessary, we have
∫ Q H | Z [ w ] | 2 d y d θ ≥ c 0 H μ α ,
where c 0 > 0 is the constant provided by Lemma 5.2 and depends only on n , s , p , K . 
Proof. 
Let ( b ^ , σ ^ ) and W be as in Lemma 5.3, and set ε : = W − 1 q . Since q = 2 s α and α ( p − 1 ) = 1 , ε q W = 1 , ε 2 s = W − 1 α , q p = q + 2 s . Define y = e ε 2 s τ 2 s b ^ + ε z , θ = σ ^ + ε 2 s τ , v ( z , τ ) : = ε q w ( y , θ ) .
By (116) and (115), we have σ ^ − ( 2 L 0 ) 2 s ε 2 s , σ ^ ⋐ J . Moreover, it follows from (114) that
M q [ v ( · , τ ) ] = M q [ w ( · , θ ) ] ≤ K , ∥ v ∥ L ∞ ( R n × [ − ( 2 L 0 ) 2 s , 0 ] ) < ∞
for every bounded interval J ′ ⋐ J . By direct calculations, we have
( − Δ ) z s v ( z , τ ) = ε q + 2 s ( − Δ ) y s w ( y , θ ) , v τ = ε q + 2 s w θ + 1 2 s ( y − ε z ) · ∇ y w ,
and
ε 2 s 1 2 s z · ∇ z v + α v = ε q + 2 s 1 2 s ε z · ∇ y w + α w .
Then using (53) and q p = q + 2 s , we obtain
v τ + ( − Δ ) s v + ε 2 s 1 2 s z · ∇ v + α v = v p .
Moreover,
v p − ( − Δ ) s v = ε q + 2 s w p − ( − Δ ) s w ( y , θ ) = ε q + 2 s Z [ w ] ( y , θ ) .
For ( z , τ ) ∈ Q 2 L 0 , the corresponding point satisfies
y − e θ − σ ^ 2 s b ^ = ε | z | ≤ 2 L 0 ε , σ ^ − ( 2 L 0 ) 2 s ε 2 s ≤ θ ≤ σ ^ .
Thus (116), together with ε q W = 1 , implies v ( 0 , 0 ) = 1 , 0 ≤ v ≤ 2 on Q 2 L 0 .
Now, increase H 0 , if necessary, so that H 0 − 1 q ≤ ε 0 . Since W ≥ H ≥ H 0 , we have ε ≤ H 0 − 1 q ≤ ε 0 . All hypotheses of Lemma 5.2 are therefore satisfied, and it yields
c 0 ≤ ∫ Q L 0 | v p − ( − Δ ) s v | 2 d z d τ .
Let Q ^ be the image of Q L 0 under ( z , τ ) ↦ ( y , θ ) . Since Q L 0 ⊂ Q 2 L 0 , (116) implies Q ^ ⊂ Q H . Hence (123) and (124) yield
c 0 ≤ ε 2 q + 4 s ε − n − 2 s ∫ Q ^ | Z [ w ] | 2 d y d θ = ε 2 q + 2 s − n ∫ Q ^ | Z [ w ] | 2 d y d θ .
Finally, since 2 q + 2 s − n q = μ α > 0 , combining with Q ^ ⊂ Q H and W ≥ H , we deduce
∫ Q H | Z [ w ] | 2 d y d θ ≥ c 0 ε − ( 2 q + 2 s − n ) = c 0 W μ α ≥ c 0 H μ α .
Thus we complete the proof of Lemma 5.4. □

5.3. A Covering Lemma with Bounded Overlap

Lemma 5.5. 
Fix R > 0 , let J ⊂ R be a nonempty bounded interval, and set L J : = diam J . Let E ⊂ B R × J and 0 < r , d ≤ 1 . Then there exist constants N , C > 0 , depending only on n , s , L J , and a finite family of points ( b j , σ j ) ∈ E with associated measurable sets B j ⊂ R n × R such that
E ⊂ ⋃ j B j , Q ( b j , σ j , r , d ) ⊂ B j for every j ,
and
∑ j 1 B j ≤ N , | B j | ≤ C r n d for every j .
Proof. 
he conclusion holds trivially if E = ⌀ . Next we assume that E ≠ ⌀ . Let σ c : = inf J + sup J 2 , and define
Ψ c ( y , θ ) : = e − θ − σ c 2 s y , θ − σ c ≜ ( ζ , ϑ ) .
For ( b , σ ) ∈ E , let ( ζ b , ϑ b ) : = Ψ c ( b , σ ) . If ( y , θ ) ∈ Q ( b , σ , r , d ) , then ϑ ∈ [ ϑ b − d , ϑ b ] and ζ − ζ b = e − θ − σ c 2 s y − e θ − σ 2 s b . Let A 0 : = e L J + 4 2 s , since | θ − σ c | ≤ L J + 1 , we obtain
Ψ c Q ( b , σ , r , d ) ⊂ B A 0 r ( ζ b ) × [ ϑ b − d , ϑ b ] .
Since E ≠ ⌀ , choose ( ζ 1 , ϑ 1 ) ∈ Ψ c ( E ) . For each j ≥ 1 , suppose that ( ζ i , ϑ i ) have been selected for 1 ≤ i ≤ j . If there exists a point ( ζ , ϑ ) ∈ Ψ c ( E ) such that
| ζ − ζ i | ≥ 4 A 0 r or | ϑ − ϑ i | ≥ 4 d ( 1 ≤ i ≤ j ) ,
select one such point and denote it by ( ζ j + 1 , ϑ j + 1 ) . For each selected point ( ζ j , ϑ j ) , define D j : = B 2 A 0 r ( ζ j ) × ( ϑ j − 2 d , ϑ j + 2 d ) , then { D j } are pairwise disjoint. Indeed, if ( ζ , ϑ ) ∈ D i ∩ D j for i ≠ j , then the triangle inequality yields
| ζ i − ζ j | ≤ | ζ i − ζ | + | ζ − ζ j | < 4 A 0 r , | ϑ i − ϑ j | ≤ | ϑ i − ϑ | + | ϑ − ϑ j | < 4 d ,
contradicting the selection condition (128). Since Ψ c ( E ) is bounded, we can choose R 1 , T 1 > 0 such that Ψ c ( E ) ⊂ B R 1 × [ − T 1 , T 1 ] . By the definition of D j , for any ( ζ , ϑ ) ∈ D j , there holds
| ζ | ≤ | ζ − ζ j | + | ζ j | < 2 A 0 r + R 1 , | ϑ | ≤ | ϑ − ϑ j | + | ϑ j | < 2 d + T 1 .
Therefore, we obtain D j ⊂ B R 1 + 2 A 0 r × ( − T 1 − 2 d , T 1 + 2 d ) . Let V : = ( 2 T 1 + 4 d ) | B R 1 + 2 A 0 r | . Since { D j } are pairwise disjoint, and | D j | = 4 d | B 2 A 0 r | for each j, after selecting m points, there holds
∑ j = 1 m | D j | = ⋃ j = 1 m D j = m 4 d | B 2 A 0 r | ≤ V .
Thus the number of selected points is bounded by V 4 d | B 2 A 0 r | , so the selection stops after finitely many steps. Denote the total number of selected points by M. Then we have
Ψ c ( E ) ⊂ ⋃ j = 1 M B 4 A 0 r ( ζ j ) × ( ϑ j − 4 d , ϑ j + 4 d ) .
Hence the sets B ˜ j : = B 8 A 0 r ( ζ j ) × [ ϑ j − 4 d , ϑ j + 4 d ] , 1 ≤ j ≤ M , cover Ψ c ( E ) . By (127), they also satisfy
Ψ c Q ( b j , σ j , r , d ) ⊂ B ˜ j , ( b j , σ j ) : = Ψ c − 1 ( ζ j , ϑ j ) ∈ E .
Define B j : = Ψ c − 1 ( B ˜ j ) , then B j are measurable and satisfy
E ⊂ ⋃ j = 1 M B j , Q ( b j , σ j , r , d ) ⊂ B j ,
which proves (125). Finally, we prove (126). For any fixed ( ζ , ϑ ) ∈ R n × R , let Z : = { j : ( ζ , ϑ ) ∈ B ˜ j } and m : = # Z . For any j ∈ Z , the definition of B ˜ j yields
| ζ − ζ j | < 8 A 0 r , | ϑ − ϑ j | ≤ 4 d .
For each such index j and any ( ζ ′ , ϑ ′ ) ∈ D j , we have
| ζ ′ − ζ | ≤ | ζ ′ − ζ j | + | ζ j − ζ | < 2 A 0 r + 8 A 0 r = 10 A 0 r , | ϑ ′ − ϑ | ≤ | ϑ ′ − ϑ j | + | ϑ j − ϑ | < 2 d + 4 d = 6 d .
Thus we have D j ⊂ B 10 A 0 r ( ζ ) × ( ϑ − 6 d , ϑ + 6 d ) for any j ∈ Z . Since { D j } are pairwise disjoint, and | D j | = 4 d | B 2 A 0 r | for each j, we obtain
4 d m | B 2 A 0 r | ≤ 12 d | B 10 A 0 r | , and m ≤ | B 10 A 0 r | | B 2 A 0 r | 12 d 4 d = 3 · 5 n = : N .
Since ( ζ , ϑ ) was arbitrary, we obtain ∑ j = 1 M 1 B ˜ j ≤ N . Since B j = Ψ c − 1 ( B ˜ j ) , for every ( y , θ ) ∈ R n × R we have
∑ j = 1 M 1 B j ( y , θ ) = ∑ j = 1 M 1 B ˜ j ( ζ , ϑ ) ≤ N .
Since ( b j , σ j ) ∈ E ⊂ B R × J , we have | ϑ j | = | σ j − σ c | ≤ L J 2 . Thus, using d ≤ 1 , every ( ζ , ϑ ) ∈ B ˜ j satisfies
| ϑ | ≤ | ϑ j | + 4 d ≤ L J 2 + 4 ≤ L J + 4 .
Hence e n ϑ 2 s ≤ A 0 n . Since B j = Ψ c − 1 ( B ˜ j ) , we obtain
| B j | = ∫ B ˜ j e n ϑ 2 s d ζ d ϑ ≤ A 0 n | B ˜ j | = A 0 n | B 8 A 0 r | 8 d = C r n d ,
where C : = 8 n + 1 | B 1 | A 0 2 n = 8 n + 1 | B 1 | e n ( L J + 4 ) s depends only on n , s , L J . By (129) and (130), we prove (126). □

5.4. Improvement of the Local Estimate for Z [ w ]

Proposition 5.6. 
Let u be a bounded nonnegative classical entire solution of (1), and set w = w a , k as in (52), where a ∈ R n and k ≥ 2 . Let I , I ˜ be bounded open intervals with I ¯ ⋐ I ˜ , and let B R ⋐ B R ˜ . Set
δ t : = dist ( I ¯ , R ∖ I ˜ ) , δ x : = R ˜ − R , L I : = diam I .
Assume that, for some C 0 , K > 0 and 0 < γ ≤ α ,
0 ≤ w ≤ C 0 k α on B R ˜ × I ˜ , sup σ ∈ I ˜ M q [ w ( · , σ ) ] ≤ K , ∫ I ˜ ∫ B R ˜ | Z [ w ] | 2 d y d σ ≤ C 0 k γ .
Then there exist H 0 , C > 0 such that, for every H 0 ≤ H ≤ C 0 k α ,
{ ( y , σ ) ∈ B R × I : w ( y , σ ) > H } ≤ C k γ H − 2 p .
Consequently, for every 0 < d < 2 p ,
∫ I ∫ B R ( 1 + w ) d d y d σ ≤ C d 1 + k d γ 2 p .
Here H 0 , C depend only on n , s , p , C 0 , K , R , L I , δ t , δ x , and C d additionally depends on d. 
Proof. 
Let C 1 be the constant provided by Lemma 5.3. For H > 0 , set
E H : = { ( y , σ ) ∈ B R × I : w ( y , σ ) > H } , r H : = C 1 H − 1 q , d H : = C 1 H − 1 α .
We will apply Lemma 5.5 to E H with J = I , r = r H , and d = d H . With I , I ˜ , R , R ˜ fixed as above, we first show that H 0 can be chosen sufficiently large so that, for every H ≥ H 0 ,
r H , d H ≤ 1 , B j ⊂ B R ˜ × I ˜ for every j ,
where B j are the sets constructed in the proof of Lemma 5.5 at these scales. Set
σ c : = inf I + sup I 2 , A 0 : = e L I + 4 2 s ,
and let Ψ c be the map defined in the proof of Lemma 5.5 with J = I . For r H , d H ≤ 1 , by construction we have
B j = Ψ c − 1 ( B ˜ j ) , B ˜ j : = B 8 A 0 r H ( ζ j ) × [ ϑ j − 4 d H , ϑ j + 4 d H ] ,
where ( ζ j , ϑ j ) : = Ψ c ( b j , σ j ) , ( b j , σ j ) ∈ E H ⊂ B R × I . Fix ( y , θ ) ∈ B j and let ( ζ , ϑ ) : = Ψ c ( y , θ ) . Then ( ζ , ϑ ) ∈ B ˜ j and thus
| ζ − ζ j | < 8 A 0 r H , | ϑ − ϑ j | ≤ 4 d H .
Since σ j ∈ I and d H ≤ 1 ,
| ϑ | ≤ | ϑ j | + 4 d H = | σ j − σ c | + 4 d H ≤ L I 2 + 4 ≤ L I + 4 .
Thus e ϑ 2 s ≤ A 0 . Using the definitions of Ψ c and ( ζ j , ϑ j ) , we obtain
θ − σ j = ϑ − ϑ j , y − e θ − σ j 2 s b j = e ϑ 2 s ( ζ − ζ j ) .
Consequently, we have
| θ − σ j | ≤ 4 d H , y − e θ − σ j 2 s b j ≤ 8 A 0 2 r H .
Since | b j | ≤ R and | θ − σ j | ≤ 4 d H ≤ 4 , by the mean value theorem, we have
e θ − σ j 2 s − 1 | b j | ≤ | θ − σ j | 2 s e | θ − σ j | 2 s R ≤ C s R d H , C s : = 2 s e 2 s .
Since ( b j , σ j ) ∈ B R × I , it follows that dist ( θ , I ) ≤ | θ − σ j | ≤ 4 d H and
dist ( y , B R ) ≤ | y − b j | ≤ y − e θ − σ j 2 s b j + e θ − σ j 2 s − 1 | b j | ≤ 8 A 0 2 r H + C s R d H ≤ C ( r H + R d H ) ,
where C : = max { 8 A 0 2 , C s } depends only on n , s , L I .
Choose H 0 large enough to satisfy the conditions in Lemmas 5.3 and 5.4, and also
r H 0 ≤ 1 , d H 0 ≤ 1 , 4 d H 0 < δ t 2 , C fl ( r H 0 + R d H 0 ) < δ x 2 .
Then for every H ≥ H 0 , we have r H , d H ≤ 1 . Moreover, for every ( y , θ ) ∈ B j , there holds
| θ − σ j | ≤ 4 d H < δ t 2 , | y − b j | ≤ C fl ( r H + R d H ) < δ x 2 .
Since ( b j , σ j ) ∈ E H ⊂ B R × I , the definition of δ t implies that θ ∈ I ˜ . Also,
| y | ≤ | b j | + | y − b j | < R + δ x 2 < R ˜ .
Thus B j ⊂ B R ˜ × I ˜ for every j. With H 0 chosen as above, fix H with H 0 ≤ H ≤ C 0 k α . Since r H , d H ≤ 1 , apply Lemma 5.5 to E H with J = I and scales r H , d H . Denote the selected points by ( b j , σ j ) ∈ E H , 1 ≤ j ≤ M , and the corresponding sets by B j . For each j, define Q j : = Q ( b j , σ j , r H , d H ) . Then Lemma 5.5 yields
E H ⊂ ⋃ j = 1 M B j , Q j ⊂ B j ⊂ B R ˜ × I ˜ , ∑ j = 1 M 1 B j ≤ N , | B j | ≤ C r H n d H .
Since u is bounded, w is bounded on R n × J ′ for every compact interval J ′ ⋐ I ˜ . Since ( b j , σ j ) ∈ E H , we have w ( b j , σ j ) > H ≥ H 0 . The choice of H 0 ensures that [ σ j − d H , σ j ] ⋐ I ˜ . Moreover, we have
0 ≤ w ≤ C 0 k α on Q j , sup θ ∈ I ˜ M q [ w ( · , θ ) ] ≤ K .
By Lemma 5.4 with ( b , σ ) = ( b j , σ j ) and Q H = Q j , we have
∫ Q j | Z [ w ] | 2 d y d σ ≥ c 0 H μ α ,
where c 0 is the constant in Lemma 5.4. Therefore, we obtain
c 0 M H μ α ≤ ∑ j = 1 M ∫ Q j | Z [ w ] | 2 d y d σ ≤ ∑ j = 1 M ∫ B j | Z [ w ] | 2 d y d σ = ∫ I ˜ ∫ B R ˜ ∑ j = 1 M 1 B j | Z [ w ] | 2 d y d σ ≤ N ∫ I ˜ ∫ B R ˜ | Z [ w ] | 2 d y d σ ≤ N C 0 k γ .
Hence M ≤ C k γ H − μ α . Consequently, we obtain
| E H | ≤ ∑ j = 1 M | B j | ≤ C M r H n d H ≤ C k γ H − μ α − n q − 1 α .
Since q = 2 s α , μ = 2 α + 1 − n 2 s , and α = 1 p − 1 , we have μ α + n q + 1 α = 2 + 2 α = 2 p . This proves (131).
We next derive (132). Set V 0 : = | B R | | I | and choose a fixed constant C large enough that
A k : = C ( 1 + k γ ) ≥ V 0 H 0 2 p .
If 0 < H < H 0 , we have A k H − 2 p ≥ V 0 ( H 0 H ) 2 p ≥ V 0 , so | E H | ≤ V 0 . If H 0 ≤ H ≤ C 0 k α , by (131) we have | E H | ≤ A k H − 2 p . If H > C 0 k α , E H is empty. Thus, for every H > 0 , we have
| E H | ≤ min { V 0 , A k H − 2 p } .
Define H * : = A k V 0 1 2 p , V 0 = A k H * − 2 p . Then | E H | ≤ V 0 . For 0 < d < 2 p , the layer-cake formula therefore yields
∫ I ∫ B R w d d y d σ = d ∫ 0 ∞ H d − 1 | E H | d H ≤ d V 0 ∫ 0 H * H d − 1 d H + d A k ∫ H * ∞ H d − 1 − 2 p d H = V 0 H * d + d 2 p − d A k H * d − 2 p = 2 p 2 p − d V 0 1 − d 2 p A k d 2 p ≤ C d 1 + k d γ 2 p .
In the last step we used A k = C ( 1 + k γ ) and ( 1 + k γ ) d 2 p ≤ 1 + k d γ 2 p . Finally, by ( 1 + w ) d ≤ c d ( 1 + w d ) with c d = max { 1 , 2 d − 1 } , we prove (132). □
Proposition 5.6 controls the local integrals of w 2 and w p + 1 . Using Lemmas 4.5 and 4.6 and the localized energy identity in Proposition 4.2, we then improve the bound for ∫ | Z [ w ] | 2 on a smaller region. The time intervals and cutoff reference times are shifted together, while the interval lengths, separations, and spatial radii remain fixed. Hence the constants are independent of s k = − log k .
Proposition 5.7. 
Assume p F ( n , s ) < p < p S ( n , s ) . Fix nonempty bounded open intervals I 3 ⋐ I 2 ⋐ I 1 ⋐ I 0 and R 1 > 0 . Then there exists R 2 > R 1 , depending only on s , R 1 and diam I 2 , with the following property. 
Let u be a bounded nonnegative classical entire solution of (1). For a ∈ R n and k ≥ 2 , set w = w a , k as in (52). Suppose that, for some C 0 , K > 0 and 0 < γ ≤ α , the following bounds hold:
0 ≤ w ( y , σ ) ≤ C 0 k α for all ( y , σ ) ∈ R n × I 0 ,
and
sup σ ∈ I 0 M q [ w ( · , σ ) ] ≤ K , ∫ I 0 ∫ B R 2 | Z [ w ] | 2 d y d σ ≤ C 0 k γ .
Then
∫ I 3 ∫ B R 1 | Z [ w ] | 2 d y d σ ≤ C k κ * γ , κ * : = p + 1 2 p < 1 .
The constant C depends only on n , s , p , K , C 0 , R 1 , R 2 , the lengths of the four intervals, and the positive distances dist ( I ¯ j + 1 , R ∖ I j ) , j = 0 , 1 , 2 . 
Proof. 
Let σ c be the midpoint of I 2 and set
L : = 1 2 diam I 2 , Λ : = e L 2 s , R ˜ : = 32 Λ 4 R 1 , R 2 : = 2 R ˜ = 64 Λ 4 R 1 .
Thus R 2 depends only on s , R 1 and diam I 2 . Fix a radial function χ ∈ C c ∞ ( B 2 ) with 0 ≤ χ ≤ 1 and χ = 1 on B 1 . Define λ ( σ ) : = e σ − σ c 2 s and
η 0 ( y , σ ) : = χ y 2 Λ R 1 λ ( σ ) , η 1 ( y , σ ) : = χ y 5 Λ 3 R 1 λ ( σ ) , ζ ( y ) : = χ y 12 Λ 4 R 1 .
For every σ ∈ I 2 , since Λ − 1 ≤ λ ( σ ) ≤ Λ , we have
η 0 ( · , σ ) = 1 on B 2 R 1 , supp η 0 ( · , σ ) ⊂ B 4 Λ 2 R 1 , η 1 ( · , σ ) = 1 on B 5 Λ 2 R 1 , supp η 1 ( · , σ ) ⊂ B 10 Λ 4 R 1 ,
and
ζ = 1 on B 12 Λ 4 R 1 , supp ζ ⊂ B 24 Λ 4 R 1 ⋐ B R ˜ .
We apply Proposition 5.6 to B R ˜ × I 1 ⋐ B R 2 × I 0 . Since 2 , p + 1 < 2 p , by (132) with d = 2 and d = p + 1 respectively, we have
∫ I 1 ∫ B R ˜ w 2 d y d σ ≤ C k γ p , ∫ I 1 ∫ B R ˜ w p + 1 d y d σ ≤ C k κ * γ ,
where the last inequalities use k ≥ 2 and γ > 0 . Let J − be the left component of I 1 ∖ I ¯ 2 , | J − | ≥ dist ( I ¯ 2 , R ∖ I 1 ) > 0 . By (140) and the mean value theorem for integrals, there exists σ 0 ∈ J − such that
∫ B R ˜ w ( y , σ 0 ) 2 d y ≤ C | J − | k γ p ≤ C k γ p .
Applying Lemma 4.6 on [ σ 0 , sup I 2 ] ⋐ I 1 with cutoff ζ , together with (139)-(141) and 1 p < p + 1 2 p = κ * , we obtain
∫ I 2 ∥ ζ w ∥ H s 2 d σ ≤ C 1 + k γ p + k κ * γ ≤ C k κ * γ .
For every σ ∈ I 2 , by (138) and (139), we have η i w ( · , σ ) = η i ( ζ w ) ( · , σ ) for i = 0 , 1 . Then combining the definitions of η i with Λ − 1 ≤ λ ( σ ) ≤ Λ , we obtain
∥ η i ( · , σ ) w ( · , σ ) ∥ H s ≤ C ∥ ζ ( · ) w ( · , σ ) ∥ H s , i = 0 , 1 , σ ∈ I 2 ,
where C depends only on n , s , R 1 , L . Using (142) and (140), we therefore get
∫ I 2 ∥ η 1 w ∥ H s 2 + ∥ η 1 w ∥ 2 2 d σ ≤ C k κ * γ .
By Lemma 4.5, (143) and an argument similar to the one used to prove (94) in the proof of Proposition 4.7, we obtain
∫ I 2 ∥ C η 0 [ w ] ∥ L 2 ( supp η 0 ( · , σ ) ) 2 d σ ≤ C K 2 | I 2 | + k κ * γ ≤ C k κ * γ .
Let J L , J R be the left and right components of I 2 ∖ I ¯ 3 respectively. Then | J L | , | J R | ≥ dist ( I ¯ 3 , R ∖ I 2 ) > 0 . By (140), (142) and arguments similar to those used to derive (95) and (96), together with (73), there exist σ − ∈ J L and σ + ∈ J R , such that
F η 0 ( σ − ) [ w ] ≤ C k κ * γ , F η 0 ( σ + ) [ w ] ≥ − C k κ * γ .
Since I 3 ⊂ [ σ − , σ + ] , η 0 ( y , σ ) = 1 on B R 1 × [ σ − , σ + ] and e − μ | I 2 | ≤ e μ ( σ − σ + ) ≤ 1 on [ σ − , σ + ] , by (144), (145) and a similar argument in the proof of (98), we obtain
∫ I 3 ∫ B R 1 | Z [ w ] | 2 d y d σ ≤ C k κ * γ .
This proves (136). Thus we complete the proof of Proposition 5.7. □

6. Proof of the Liouville Theorem

6.1. Proof for Bounded Solutions

Theorem 6.1. 
If 1 < p < p S ( n , s ) , let u be a bounded nonnegative classical entire solution of (1). Then u ≡ 0 . 
Proof. 
The range 1 < p ≤ p F ( n , s ) follows from Theorem 3.3. We therefore assume p F ( n , s ) < p < p S ( n , s ) , so 0 < q < n and μ > 0 . Suppose that u ¬ ≡ 0 and set M = ∥ u ∥ L ∞ ( R n × R ) > 0 . Choose ( x 0 , t 0 ) ∈ R n × R such that u ( x 0 , t 0 ) > M 2 . Set λ = M − 1 q and define v ( x , t ) : = λ q u ( x 0 + λ x , t 0 + λ 2 s t ) . By translation invariance and (3), v is also a bounded nonnegative classical entire solution of (1). Moreover, we have 0 ≤ v ≤ 1 and v ( 0 , 0 ) > 1 2 . Therefore, without loss of generality, we may assume
0 ≤ u ≤ 1 , u ( 0 , 0 ) = c * ≥ 1 2 .
For each integer k ≥ 2 , define w k = w 0 , k by (52) with a = 0 , namely,
w k ( y , σ ) : = e − α σ u e − σ 2 s y , k − e − σ , ( y , σ ) ∈ R n × R .
Set s k = − log k , then we have w k ( 0 , s k ) = k α u ( 0 , 0 ) = c * k α . Since 0 < κ * = p + 1 2 p < 1 and μ > 0 , set N : = min { m ∈ N : m ≥ 1 , κ * m α < μ } , A : = 4 N + 1 , γ j : = κ * j α ( 0 ≤ j ≤ N ) .
For y ∈ R n and | θ | ≤ A , the definition of w k and 0 ≤ u ≤ 1 yield
0 ≤ w k ( y , s k + θ ) ≤ e α A k α .
Moreover, by Proposition 3.5 and (55), we have sup σ ∈ R M q [ w k ( · , σ ) ] ≤ K ( n , s , p ) .
Set ρ N = 1 and define ρ j : = 64 e 2 ( A − 4 j − 2 ) s ρ j + 1 for j = N − 1 , … , 0 . For each 0 ≤ j < N , define the four intervals I ℓ ( j ) : = ( − A + 4 j + ℓ , A − 4 j − ℓ ) , ℓ = 0 , 1 , 2 , 3 . Let Q j ( k ) : = B ρ j × [ s k − A + 4 j , s k + A − 4 j ] , 0 ≤ j ≤ N . Then, for 0 ≤ j < N , Q j + 1 ( k ) ⊂ B ρ j + 1 × ( s k + I 3 ( j ) ) , B ρ j × ( s k + I 0 ( j ) ) ⊂ Q j ( k ) , and Q N ( k ) = B 1 × [ s k − 1 , s k + 1 ] . We prove by induction on j that, for each 0 ≤ j ≤ N , there exists a constant D j > 0 , independent of k, such that
∫ Q j ( k ) | Z [ w k ] | 2 d y d σ ≤ D j k γ j .
For j = 0 , set γ 0 : = α . By applying Proposition 4.7 on Q 0 ( k ) , we obtain a constant D 0 > 0 , independent of k, such that
∫ Q 0 ( k ) | Z [ w k ] | 2 d y d σ ≤ D 0 k α = D 0 k γ 0 .
Suppose that, for some 0 ≤ j < N ,
∫ Q j ( k ) | Z [ w k ] | 2 d y d σ ≤ D j k γ j ,
where D j is independent of k. Using (147), (150) and applying Proposition 5.7 to the intervals s k + I ℓ ( j ) and radii ρ j + 1 , ρ j , with γ = γ j and C 0 = max { e α A , D j } , we obtain that there exists a constant D j + 1 > 0 , independent of k, such that
∫ Q j + 1 ( k ) | Z [ w k ] | 2 d y d σ ≤ D j + 1 k κ * γ j = D j + 1 k γ j + 1 .
This, together with (149), implies (148). In particular, we obtain
∫ s k − 1 s k + 1 ∫ B 1 | Z [ w k ] | 2 d y d σ ≤ D N k γ N , γ N < μ .
Define v k ( z , τ ) : = k − α w k ( k − 1 2 s z , s k + τ k ) , D z v : = 1 2 s z · ∇ z v + α v . Then we have
( v k ) τ + ( − Δ ) s v k + k − 1 D z v k = v k p , v k ( 0 , 0 ) = c * ,
and
v k ( z , τ ) = e − α τ k u e − τ 2 s k z , k ( 1 − e − τ k ) .
Since 0 ≤ u ≤ 1 , (153) implies
0 ≤ v k ≤ e α on R n × [ − k , k ] .
Moreover, we have
v k ⟶ u locally uniformly on R n × R .
Let y = k − 1 2 s z and σ = s k + τ k , define R k ( z , τ ) : = ( v k ) τ ( z , τ ) + k − 1 D z v k ( z , τ ) = k − α − 1 Z [ w k ] ( y , σ ) . Since μ = 2 α + 1 − n 2 s > 0 , then (151) yields
∫ − k k ∫ B k 1 2 s | R k | 2 d z d τ = k − μ ∫ s k − 1 s k + 1 ∫ B 1 | Z [ w k ] | 2 d y d σ ≤ D N k γ N − μ ⟶ 0 .
Let ψ ∈ C c ∞ ( R n × R ) be a test function in (152). For all sufficiently large k, we have supp ψ ⊂ B k 1 2 s × [ − k , k ] . Using the definition of R k and integrating by parts, we obtain
− ∫ ∫ R n × R v k ∂ τ ψ d z d τ + 1 k ∫ ∫ R n × R v k α ψ − 1 2 s div ( z ψ ) d z d τ = ∫ ∫ R n × R R k ψ d z d τ .
By (154), we have
1 k ∫ ∫ R n × R v k α ψ − 1 2 s div ( z ψ ) d z d τ ≤ e α k ∫ ∫ R n × R α ψ − 1 2 s div ( z ψ ) d z d τ ⟶ 0 .
By the Cauchy-Schwarz inequality and (156), we conclude that
∫ ∫ R n × R R k ψ d z d τ ≤ ∥ R k ∥ L 2 ( supp ψ ) ∥ ψ ∥ 2 ⟶ 0 .
Then by (155), (157)-(159), we have
∫ ∫ R n × R u ∂ τ ψ d z d τ = 0 .
Thus u t = 0 in distributions. Since u t is continuous, then u t = 0 holds pointwise. Hence u ( x , t ) = u ( x , 0 ) , and (1) reduces to
( − Δ ) s u ( · , 0 ) = u ( · , 0 ) p in R n .
The elliptic Liouville theorem in [9] (see also [4,11,18]) then implies u ( · , 0 ) ≡ 0 , contradicting (146). □

6.2. Proof for General Solutions

The following lemma adapts the doubling argument of [40][Lemma 5.1] to the time scale r 2 s . We include the proof here for completeness.
Lemma 6.2. 
Let M : R n × R → [ 0 , ∞ ) be a continuous function and satisfy M ¬ ≡ 0 . Then for every ℓ ≥ 1 , there exists X ℓ = ( x ℓ , t ℓ ) with M ( X ℓ ) > 0 such that
M ( x , t ) ≤ 2 M ( X ℓ )
whenever
| x − x ℓ | ≤ ℓ M ( X ℓ ) , | t − t ℓ | ≤ ℓ M ( X ℓ ) 2 s .
Proof. 
Fix ℓ ≥ 1 and choose X ( 0 ) = ( x ( 0 ) , t ( 0 ) ) with M 0 : = M ( X ( 0 ) ) > 0 . We define the sequence of points X ( j ) = ( x ( j ) , t ( j ) ) inductively. Suppose that X ( j ) = ( x ( j ) , t ( j ) ) has been chosen with M j : = M ( X ( j ) ) > 0 . If
M ( x , t ) ≤ 2 M j for ( x , t ) ∈ B ℓ M j ( x ( j ) ) × t ( j ) − ℓ M j 2 s , t ( j ) + ℓ M j 2 s ,
then take X ℓ : = X ( j ) and the proof is complete. Otherwise, there exists a point X ( j + 1 ) = ( x ( j + 1 ) , t ( j + 1 ) ) that satisfies
X ( j + 1 ) ∈ B ℓ M j ( x ( j ) ) × t ( j ) − ℓ M j 2 s , t ( j ) + ℓ M j 2 s , M j + 1 : = M ( X ( j + 1 ) ) > 2 M j > 0 .
Choose this point as X ( j + 1 ) and repeat the construction.
If this construction continued for all j ≥ 0 , then M j ≥ 2 j M 0 . Since s > 0 , we have
∑ j = 0 ∞ | x ( j + 1 ) − x ( j ) | ≤ ℓ M 0 ∑ j = 0 ∞ 2 − j < ∞ , ∑ j = 0 ∞ | t ( j + 1 ) − t ( j ) | ≤ ℓ M 0 2 s ∑ j = 0 ∞ 2 − 2 s j < ∞ .
Thus ( x ( j ) ) and ( t ( j ) ) are Cauchy sequences. Hence X ( j ) → X ∞ ∈ R n × R . By continuity,
M j = M ( X ( j ) ) ⟶ M ( X ∞ ) < ∞ ,
contradicting M j ≥ 2 j M 0 → ∞ . Therefore, there exists an index N, such that
M ( x , t ) ≤ 2 M N for ( x , t ) ∈ B ℓ M N ( x ( N ) ) × t ( N ) − ℓ M N 2 s , t ( N ) + ℓ M N 2 s .
With X ℓ : = X ( N ) , we obtain (160) and (161). This completes the proof of Lemma 6.2. □
Proof 
(Proof of Theorem 1.2). Let K = K ( n , s , p ) be the constant in Proposition 3.5. If 1 < p < p F ( n , s ) , then q > n . For fixed a ∈ R n , t ∈ R , and ρ > 0 , it follows from (45) that
0 ≤ ∫ B ρ ( a ) u ( x , t ) d x ≤ ∫ B R ( a ) u ( x , t ) d x ≤ K R n − q ⟶ 0 as R → ∞ .
Since ρ > 0 and t ∈ R is arbitrary, together with the nonnegativity and continuity of u, we obtain u ≡ 0 . We now assume p F ( n , s ) ≤ p < p S ( n , s ) , so that 0 < q ≤ n . Then Proposition 3.5 yields
sup t ∈ R M q [ u ( · , t ) ] ≤ K .
Suppose for contradiction that u ¬ ≡ 0 , and set M ( x , t ) : = u ( x , t ) 1 q . For each integer k ≥ 1 , by applying Lemma 6.2 with ℓ = k , we obtain X k = ( x k , t k ) , m k : = M ( X k ) > 0 , λ k : = m k − 1 . Define v k ( z , τ ) : = λ k q u ( x k + λ k z , t k λ k 2 s τ ) . Then v k is a nonnegative classical entire solution of
( v k ) τ + ( − Δ ) s v k = v k p , v k ( 0 , 0 ) = λ k q m k q = 1 .
For ( z , τ ) ∈ B k × [ − k 2 s , k 2 s ] , we have | λ k z | ≤ k m k , | λ k 2 s τ | ≤ k m k 2 s . Hence (160), applied to M = u 1 q , yields
0 ≤ v k ≤ 2 q on B k × [ − k 2 s , k 2 s ] , v k ( 0 , 0 ) = 1 .
Moreover, note that for every b ∈ R n , R > 0 , and τ ∈ R ,
R q − n ∫ B R ( b ) v k ( z , τ ) d z = ( λ k R ) q − n ∫ B λ k R ( x k + λ k b ) u ( x , t k + λ k 2 s τ ) d x ≤ K .
Consequently, we have
sup k ≥ 1 sup τ ∈ R M q [ v k ( · , τ ) ] ≤ K .
Therefore, by applying Lemma 3.6 with f = v k , we deduce that there exists a positive constant C = C ( n , s ) , uniformly in k , τ , and a ∈ R n , such that
∫ R n v k ( z , τ ) 1 + | z − a | n + 2 s d z ≤ C K , ∫ | z − a | > R v k ( z , τ ) | z − a | n + 2 s d z ≤ C K R − q − 2 s , R ≥ 1 .
For every φ ∈ C c ∞ ( R n × R ) , testing (162) for v k against φ and integrating by parts, we deduce
∫ R ∫ R n v k − ∂ τ φ + ( − Δ ) s φ d z d τ = ∫ R ∫ R n v k p φ d z d τ .
Fix R , T > 0 , for all sufficiently large k, B R × [ − T , T ] ⊂ B k × [ − k 2 s , k 2 s ] . Thus v k and v k p are uniformly bounded on B R × [ − T , T ] . Together with (164), by applying Lemma 3.7 with a k ≡ 1 and f k = v k p , we conclude that there exist a subsequence of { v k } , still denoted by { v k } , and a continuous function V, such that
v k ⟶ V locally uniformly on R n × R .
Passing to the limit in (163) at each fixed point, we have
0 ≤ V ≤ 2 q on R n × R , V ( 0 , 0 ) = 1 .
At each fixed time τ , Fatou’s lemma shows that the estimates in (165) also hold, with v k replaced by V. In particular, V ( · , τ ) ∈ L 2 s ( R n ) for every τ ∈ R . Moreover, by local uniform convergence of v k , and (49) in Lemma 3.7, we have
∫ R ∫ R n v k ∂ τ φ d z d τ ⟶ ∫ R ∫ R n V ∂ τ φ d z d τ , ∫ R ∫ R n v k p φ d z d τ ⟶ ∫ R ∫ R n V p φ d z d τ ,
and
∫ R ∫ R n v k ( − Δ ) s φ d z d τ ⟶ ∫ R ∫ R n V ( − Δ ) s φ d z d τ .
Thus
∫ R ∫ R n V − ∂ τ φ + ( − Δ ) s φ d z d τ = ∫ R ∫ R n V p φ d z d τ .
This yields
V τ + ( − Δ ) s V = V p in distributions on R n × R .
Proposition 2.7 now shows that V is a bounded nonnegative classical entire solution in the sense of Definition 1.1. Since 1 < p < p S ( n , s ) , it follows from Theorem 6.1 that V ≡ 0 , contradicting V ( 0 , 0 ) = 1 . Therefore u ≡ 0 , proving Theorem 1.2. □

References

  1. Bogdan, K.; Grzywny, T.; Ryznar, M. Heat kernel estimates for the fractional Laplacian with Dirichlet conditions. Ann. Probab. 2010, 38, 1901–1923. [Google Scholar] [CrossRef]
  2. Caffarelli, L.; Gidas, B.; Spruck, J. Asymptotic symmetry and local behavior of semilinear elliptic equations with critical Sobolev growth. Comm. Pure Appl. Math. 1989, 42(no. 3), 271–297. [Google Scholar] [CrossRef]
  3. Caffarelli, L.; Silvestre, L. An extension problem related to the fractional Laplacian. Comm. Partial Differ. Equ. 2007, 32(no. 8), 1245–1260. [Google Scholar] [CrossRef]
  4. Cao, D.; Dai, W.; Qin, G. Super poly-harmonic properties, Liouville theorems and classification of nonnegative solutions to equations involving higher-order fractional Laplacians. Trans. Amer. Math. Soc. 2021, 374(no. 7), 4781–4813. [Google Scholar] [CrossRef]
  5. Chen, W.; Dai, W.; Qin, G. Liouville type theorems, a priori estimates and existence of solutions for critical and super-critical order Hardy–Hénon type equations in Rn. Math. Z. Paper No. 104. 2023, 303(no. 4). [Google Scholar] [CrossRef]
  6. Chen, W.; Dai, W.; Guo, Y. Some recent developments on fractional parabolic equations. Discret. Contin. Dyn. Syst. 2024, 44(no. 9), 2713–2791. [Google Scholar] [CrossRef]
  7. Chen, W.; Guo, Y. Fractional elliptic and parabolic equations—Liouville theorems, regularities, and a priori estimates. Commun. Pure Appl. Anal. 2025, 24(no. 11), 2038–2077. [Google Scholar] [CrossRef]
  8. Chen, W.; Hu, Y.; Ma, L. Moving planes and sliding methods for fractional elliptic and parabolic equations. Adv. Nonlinear Stud. 2024, 24(no. 2), 359–398. [Google Scholar] [CrossRef]
  9. Chen, W.; Li, C.; Li, Y. A direct method of moving planes for the fractional Laplacian. Adv. Math. 2017, 308, 404–437. [Google Scholar] [CrossRef]
  10. Chen, W.; Li, C.; Ou, B. Classification of solutions for an integral equation. Comm. Pure Appl. Math. 2006, 59(no. 3), 330–343. [Google Scholar] [CrossRef]
  11. Chen, W.; Li, Y.; Zhang, R. A direct method of moving spheres on fractional order equations. J. Funct. Anal. 2017, 272, 4131–4157. [Google Scholar] [CrossRef]
  12. Chen, Z.-Q.; Kim, P.; Song, R. Heat kernel estimates for the Dirichlet fractional Laplacian. J. Eur. Math. Soc. (JEMS) 2010, 12, 1307–1329. [Google Scholar] [CrossRef]
  13. Dai, W.; Fang, Y.; Qin, G. Classification of positive solutions to fractional order Hartree equations via a direct method of moving planes. J. Differ. Equ. 2018, 265(no. 5), 2044–2063. [Google Scholar] [CrossRef]
  14. Dai, W.; Liu, Z. Classification of nonnegative solutions to static Schrödinger–Hartree and Schrödinger–Maxwell equations with combined nonlinearities. Calc. Var. Partial Differ. Equ. Paper No. 156. 2019, 58(no. 4). [Google Scholar] [CrossRef]
  15. Dai, W.; Liu, Z.; Qin, G. Classification of nonnegative solutions to static Schrödinger–Hartree–Maxwell type equations. SIAM J. Math. Anal. 2021, 53(no. 2), 1379–1410. [Google Scholar] [CrossRef]
  16. Dai, W.; Qin, G. Liouville type theorems for elliptic equations with Dirichlet conditions in exterior domains. J. Differ. Equ. 2020, 269(no. 9), 7231–7252. [Google Scholar] [CrossRef]
  17. Dai, W.; Qin, G. Liouville type theorem for critical order Hénon–Lane–Emden type equations on a half space and its applications. J. Funct. Anal. Paper No. 109227. 2021, 281(no. 10). [Google Scholar] [CrossRef]
  18. Dai, W.; Qin, G. Liouville-type theorems for fractional and higher-order Hénon–Hardy type equations via the method of scaling spheres. Int. Math. Res. Not. IMRN 2023, no. 11, 9001–9070. [Google Scholar] [CrossRef]
  19. Dai, W.; Qin, G. Classification of solutions to conformally invariant systems with mixed order and exponentially increasing or nonlocal nonlinearity. SIAM J. Math. Anal. 2023, 55(no. 3), 2111–2149. [Google Scholar] [CrossRef]
  20. Dávila, J.; Dupaigne, L.; Wei, J. On the fractional Lane–Emden equation. Trans. Amer. Math. Soc. 2017, 369(no. 9), 6087–6104. [Google Scholar] [CrossRef]
  21. Deng, B.; Sire, Y.; Wei, J.; Wu, K. Classification of blow-ups and monotonicity formula for half-Laplacian nonlinear heat equation. Calc. Var. Partial Differ. Equ. Paper No. 130. 2021, 60. [Google Scholar] [CrossRef]
  22. Di Nezza, E.; Palatucci, G.; Valdinoci, E. Hitchhiker’s guide to the fractional Sobolev spaces. Bull. Sci. Math. 2012, 136, 521–573. [Google Scholar] [CrossRef]
  23. Du, K.; Zhang, X. Optimal gradient estimates of heat kernels of stable-like operators. Proc. Amer. Math. Soc. 2019, 147, 3559–3565. [Google Scholar] [CrossRef]
  24. Duong, A. T.; Nguyen, V. H. Liouville type theorems for fractional parabolic problems. In Differential Equations; Dynam, J., Ed.; 2023; Volume 35, pp. 3187–3200. [Google Scholar]
  25. Fernández-Real, X.; Ros-Oton, X. Regularity theory for general stable operators: parabolic equations. J. Funct. Anal. 2017, 272, 4165–4221. [Google Scholar] [CrossRef]
  26. Ferreira, R.; de Pablo, A. Blow-up rates for a fractional heat equation. Proc. Amer. Math. Soc. 2021, 149, 2011–2018. [Google Scholar] [CrossRef]
  27. Fujita, H. On the blowing up of solutions of the Cauchy problem for ut=Δu+u1+α. J. Fac. Sci. Univ. Tokyo Sect. I 1966, 13, 109–124. [Google Scholar]
  28. Gidas, B.; Ni, W.-M.; Nirenberg, L. Symmetry and related properties via the maximum principle. Comm. Math. Phys. 1979, 68, 209–243. [Google Scholar] [CrossRef]
  29. Gidas, B.; Spruck, J. Global and local behavior of positive solutions of nonlinear elliptic equations. Comm. Pure Appl. Math. 1981, 34(no. 4), 525–598. [Google Scholar] [CrossRef]
  30. Gidas, B.; Spruck, J. A priori bounds for positive solutions of nonlinear elliptic equations. Comm. Partial Differ. Equ. 1981, 6(no. 8), 883–901. [Google Scholar] [CrossRef]
  31. Giga, Y. A bound for global solutions of semilinear heat equations. Comm. Math. Phys. 1986, 103, 415–421. [Google Scholar] [CrossRef]
  32. Giga, Y.; Kohn, R. V. Asymptotically self-similar blow-up of semilinear heat equations. Comm. Pure Appl. Math. 1985, 38, 297–319. [Google Scholar] [CrossRef]
  33. Jin, T.; Li, Y. Y.; Xiong, J. On a fractional Nirenberg problem, part I: blow up analysis and compactness of solutions. J. Eur. Math. Soc. (JEMS) 2014, 16(no. 6), 1111–1171. [Google Scholar] [CrossRef]
  34. Kaplan, S. On the growth of solutions of quasi-linear parabolic equations. Comm. Pure Appl. Math. 1963, 16, 305–330. [Google Scholar] [CrossRef]
  35. Li, C.; Xu, M.; Yang, H.; Zhuo, R. The direct moving sphere for fractional Laplace equation. J. Funct. Anal. Paper No. 111010. 2025, 289(no. 8). [Google Scholar] [CrossRef]
  36. Li, Y. Y. Remark on some conformally invariant integral equations: the method of moving spheres. J. Eur. Math. Soc. (JEMS) 2004, 6(no. 2), 153–180. [Google Scholar] [CrossRef]
  37. Li, Y. Y.; Zhu, M. Uniqueness theorems through the method of moving spheres. Duke Math. J. 1995, 80(no. 2), 383–417. [Google Scholar] [CrossRef]
  38. Pazy, A. Semigroups of Linear Operators and Applications to Partial Differential Equations. In Applied Mathematical Sciences; Springer-Verlag: New York, 1983; vol. 44. [Google Scholar]
  39. Poláčik, P.; Quittner, P. A Liouville-type theorem and the decay of radial solutions of a semilinear heat equation. Nonlinear Anal. 2006, 64(no. 8), 1679–1689. [Google Scholar] [CrossRef]
  40. Poláčik, P.; Quittner, P.; Souplet, P. Singularity and decay estimates in superlinear problems via Liouville-type theorems. Part I: Elliptic equations and systems. Duke Math. J. 2007, 139, 555–579. [Google Scholar] [CrossRef]
  41. Poláčik, P.; Quittner, P.; Souplet, P. Singularity and decay estimates in superlinear problems via Liouville-type theorems. Part II: Parabolic equations. Ind. Univ. Math. J. 2007, 56, 879–908. [Google Scholar] [CrossRef]
  42. Quittner, P. Optimal Liouville theorems for superlinear parabolic problems. Duke Math. J. 2021, 170, 1113–1136. [Google Scholar] [CrossRef]
  43. Quittner, P. A priori estimates of solutions to local and nonlocal superlinear parabolic problems. J. Evol. Equ. 2025, 25, Paper No. 67. [Google Scholar] [CrossRef]
  44. Quittner, P.; Souplet, P. Superlinear Parabolic Problems: Blow-up, Global Existence and Steady States. In Birkhäuser Advanced Texts: Basler Lehrbücher, 2nd ed.; Birkhäuser/Springer: Cham, 2019. [Google Scholar]
  45. Quittner, P.; Souplet, P. Liouville theorems and universal estimates for superlinear parabolic problems without scale invariance. J. Differ. Equ. 2025, 427, 404–441. [Google Scholar] [CrossRef]
  46. Ros-Oton, X.; Serra, J. The Dirichlet problem for the fractional Laplacian: regularity up to the boundary. J. Math. Pures Appl. 2014, 101, 275–302. [Google Scholar] [CrossRef]
  47. Souplet, P. Universal estimates and Liouville theorems for superlinear problems without scale invariance. Discret. Contin. Dyn. Syst. 2023, 43(no. 3-4), 1702–1734. [Google Scholar] [CrossRef]
  48. Wei, J.; Xu, X. Classification of solutions of higher order conformally invariant equations. Math. Ann. 1999, 313, 207–228. [Google Scholar] [CrossRef]
  49. Wu, L.; Chen, W. Ancient solutions to nonlocal parabolic equations. Adv. Math. 2022, 408. [Google Scholar] [CrossRef]
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.
Copyright: This open access article is published under a Creative Commons CC BY 4.0 license, which permit the free download, distribution, and reuse, provided that the author and preprint are cited in any reuse.