Introduction
Diffusions on fractals often exhibit scaling behavior different from that of Brownian motion in Euclidean spaces. This behavior is described by the walk dimension β 2 \beta_2 β 2 appearing in sub-Gaussian heat kernel estimates. For example, on the standard Sierpiński gasket, β 2 = log 5 / log 2 > 2 \beta_2=\log5/\log2>2 β 2 = log 5/ log 2 > 2 (see [8 , 29 ] ), whereas on the standard Sierpiński carpet, β 2 ≈ 2.09697 \beta_2\approx2.09697 β 2 ≈ 2.09697 (see [5 , 7 , 31 , 6 ] ). The corresponding local regular Dirichlet forms provide analogues of the classical Dirichlet integral ∫ R d ∣ ∇ f ( x ) ∣ 2 d x \int_{\mathbb{R}^d}|\nabla f(x)|^2\mathrm{d}x ∫ R d ∣∇ f ( x ) ∣ 2 d x ; see [21 ] .
For p ∈ ( 1 , + ∞ ) ∖ { 2 } p\in(1,+\infty)\setminus\{2\} p ∈ ( 1 , + ∞ ) ∖ { 2 } , the non-linear counterpart is a p p p -energy, which generalizes
∫ R d ∣ ∇ f ( x ) ∣ p d x . \int_{\mathbb{R}^d}|\nabla f(x)|^p\mathrm{d}x. ∫ R d ∣∇ f ( x ) ∣ p d x .
Following the early work in [24 ] , the study of p p p -energies on fractals and general metric measure spaces has advanced considerably. Recent developments include constructions and analysis on p.c.f. self-similar sets [12 ] , Sierpiński carpets [39 , 34 ] , Vicsek-type fractals [10 , 14 ] , conductively homogeneous compact metric spaces [30 ] , Cheeger spaces [1 ] , and Laakso-type fractal spaces [2 ] . Further developments in non-linear potential theory and a survey of the subject can be found in [27 , 28 ] . In this setting, the p p p -walk dimension β p \beta_p β p describes the spatial scaling in the associated functional inequalities, including the Poincaré inequality, and in the associated capacity estimates. For p = 2 p=2 p = 2 , β 2 \beta_2 β 2 agrees with the walk dimension of the corresponding diffusion.
Continuing the study initiated in [43 ] , we investigate the dependence of β p \beta_p β p on p p p . On the Vicsek set, the p p p -walk dimension is explicitly given by β p = p + d h − 1 \beta_p=p+d_h-1 β p = p + d h − 1 , where d h = log 5 / log 3 d_h=\log5/\log3 d h = log 5/ log 3 is the Hausdorff dimension; see [10 ] . On the Sierpiński gasket and the Sierpiński carpet, β p > p \beta_p>p β p > p for any p ∈ ( 1 , + ∞ ) p\in(1,+\infty) p ∈ ( 1 , + ∞ ) ; see [27 Theorems 9.8 and 9.13] . These examples motivate the study of the regularity and monotonicity of β p \beta_p β p as a function of p p p on general metric measure spaces.
The Besov characterization of p p p -energies in [40 ] , together with the results on critical Besov exponents in [9 ] , shows, under volume doubling and suitable functional inequalities,
Date: November 3, 2026
MSC2020: 31E05, 28A80
Keywords: walk dimensions, Poincaré inequalities, capacity upper bounds, cutoff Sobolev inequalities.
The author is grateful to Joseph Feneuil for sharing, in February 2025, a preprint version of [20 ] on Riesz transforms on fractal-like spaces, which motivated [43 ] and the study of the monotonicity of p ↦ β p p p\mapsto\frac{\beta_p}{p} p ↦ p β p ; see Remark 2.4 for a discussion of this connection. The author is also grateful to Naotaka Kajino and Ryosuke Shimizu for helpful discussions on the monotonicity of a related exponent arising in their work [26 ] on nested fractals. This exponent corresponds to β p − d h p − 1 \frac{\beta_p-d_h}{p-1} p − 1 β p − d h in the notation of this paper. that p ↦ β p p \mapsto\beta_p p ↦ β p is continuous and non-decreasing, while p ↦ β p p p \mapsto\frac{\beta_p}{p} p ↦ p β p is non-increasing. In [43 ] , we established a dichotomy: under volume doubling and suitable functional inequalities for a family of p p p -energies indexed by p p p in an interval I ⊆ ( 1 , + ∞ ) I \subseteq(1,+\infty) I ⊆ ( 1 , + ∞ ) , either β p = p \beta_p=p β p = p for any p ∈ I p \in I p ∈ I , or β p > p \beta_p>p β p > p for any p ∈ I p \in I p ∈ I . The present paper supersedes that preprint. We strengthen the second alternative by proving that p ↦ β p p p \mapsto\frac{\beta_p}{p} p ↦ p β p is strictly decreasing on I I I . This strict monotonicity is indeed related to the unboundedness of Riesz transforms on fractal-like spaces studied in [20 ] ; see Remark 2.4.
On a d h d_h d h -Ahlfors regular space satisfying the chain condition, the Poincaré inequality and the capacity upper bound with exponent β p \beta_p β p imply
p ≤ β p ≤ d h + p − 1 ; p \leq\beta_p \leq d_h+p-1; p ≤ β p ≤ d h + p − 1 ;
see [44 Proposition 2.1] and Figure 1 . This leads to the normalized exponent
Figure 1. The admissible region of ( d h , β p ) (d_h,\beta_p) ( d h , β p )
γ p : = β p − d h p − 1 ≤ 1. \gamma_p := \frac{\beta_p-d_h}{p-1} \leq1. γ p := p − 1 β p − d h ≤ 1.
The regularity and monotonicity of p ↦ γ p p \mapsto\gamma_p p ↦ γ p can provide information about the analytic structure of the space. Roughly speaking, when β p > d h \beta_p>d_h β p > d h , this quantity arises as a Hölder exponent for p p p -harmonic functions in the resistance form setting; see [27 Theorem 6.31] . The same exponent also describes the relation between the resistance metric and the original metric. Indeed, under the above conditions and when β p > d h \beta_p>d_h β p > d h , the function
ρ ( p ) ( x , y ) : = R ( p ) ( x , y ) 1 p − 1 ≍ d ( x , y ) γ p \rho^{(p)}(x,y):=R^{(p)}(x,y)^{\frac{1}{p-1}}\asymp d(x,y)^{\gamma_p} ρ ( p ) ( x , y ) := R ( p ) ( x , y ) p − 1 1 ≍ d ( x , y ) γ p
is a metric, called the p p p -resistance metric; see [27 Corollary 6.32 and Definition 6.33] .
Strict monotonicity of γ p \gamma_p γ p is closely connected with the singularity of energy measures associated with distinct values of p p p . On the standard two-dimensional Sierpiński gasket, it is proved in [28 Theorems 4.8 and 4.9] that p ↦ γ p p \mapsto\gamma_p p ↦ γ p is strictly increasing, and the resulting strict inequality implies mutual singularity of p p p - and q q q -energy measures for p ≠ q p \ne q p = q . In [26 ] , they obtain corresponding strict monotonicity and singularity results for a class of affine nested fractals, including higher-dimensional and higher-level Sierpiński gaskets; see also the announcement in [27 p. 14] .
For the golden ratio Sierpiński gasket, non-linear renormalization was studied and p p p -resistance forms were constructed in [15 ] . It was proved that the standard renormalization factor r p ∈ ( 0 , 1 ) r_p \in(0,1) r p ∈ ( 0 , 1 ) is continuous and strictly decreasing in p ∈ ( 1 , + ∞ ) p \in(1,+\infty) p ∈ ( 1 , + ∞ ) , while p ↦ r p 1 p − 1 p \mapsto r_p^{\frac{1}{p-1}} p ↦ r p p − 1 1 is continuous and non-increasing; see [15 Theorem 3] . Their strict monotonicity result concerns the renormalization factor itself.
These results, together with the identity γ p = 1 \gamma_p=1 γ p = 1 on the Vicsek set, suggest a second dichotomy on general metric measure spaces: either γ p = 1 \gamma_p=1 γ p = 1 throughout the interval I I I , or γ p < 1 \gamma_p<1 γ p < 1 throughout the interval I I I and p ↦ γ p p \mapsto\gamma_p p ↦ γ p is strictly increasing on I I I . We prove this dichotomy. Both dichotomies in this paper follow from geometric and analytic assumptions on the metric measure space and its p p p -energies, without requiring a self-similar structure.
Statement of main results
Before stating our main results, we fix some notation. Throughout this paper, C C C , C 1 C_1 C 1 , C 2 C_2 C 2 , C A C_A C A , C B C_B C B denote positive constants whose values may change at each occurrence. The sign ≍ \asymp ≍ means that the ratio of the two sides is bounded from above and below by positive constants. The sign ≲ \lesssim ≲ (≳ \gtrsim ≳ ) means that the LHS is bounded above (below) by a positive constant times the RHS. We use x + x_+ x + to denote the positive part of x ∈ R x \in\mathbb{R} x ∈ R , that is, x + : = max { x , 0 } x_+ := \max\{x,0\} x + := max { x , 0 } . For two Borel measures μ , ν \mu,\nu μ , ν , we write μ ≤ ν \mu\leq\nu μ ≤ ν if μ ( A ) ≤ ν ( A ) \mu(A) \leq\nu(A) μ ( A ) ≤ ν ( A ) for any Borel set A A A , and write μ ≪ ν \mu\ll\nu μ ≪ ν if μ ≤ C ν \mu\leq C\nu μ ≤ C ν for some constant C > 0 C>0 C > 0 . When μ , ν \mu,\nu μ , ν are σ \sigma σ -finite, μ ≤ ν \mu\leq\nu μ ≤ ν is equivalent to μ ≪ ν \mu\ll\nu μ ≪ ν and d μ d ν ≤ 1 \frac{d\mu}{d\nu}\leq1 d ν d μ ≤ 1 ν \nu ν -a.e. We use # A \#A # A to denote the cardinality of a set A A A .
Let ( X , d , m ) (X,d,m) ( X , d , m ) be a complete unbounded metric measure space, that is, ( X , d ) (X,d) ( X , d ) is a complete unbounded locally compact separable metric space and m m m is a positive Radon measure on X X X with full support. Throughout this paper, we always assume that all metric balls are relatively compact. For any x ∈ X x \in X x ∈ X and any r > 0 r>0 r > 0 , denote B ( x , r ) = { y ∈ X : d ( x , y ) < r } B(x,r)=\{y\in X:d(x,y)<r\} B ( x , r ) = { y ∈ X : d ( x , y ) < r } and V ( x , r ) = m ( B ( x , r ) ) V(x,r)=m(B(x,r)) V ( x , r ) = m ( B ( x , r )) . If B = B ( x , r ) B=B(x,r) B = B ( x , r ) , then denote ∂ B = B ( x , δ r ) \partial B=B(x,\delta r) ∂ B = B ( x , δ r ) for any δ > 0 \delta>0 δ > 0 . Let B ( X ) \mathcal{B}(X) B ( X ) be the family of all Borel measurable subsets of X X X . Let C ( X ) C(X) C ( X ) be the family of all continuous functions on X X X . Let C c ( X ) C_c(X) C c ( X ) be the family of all continuous functions on X X X with compact support. Denote f A = 1 m ( A ) ∫ A f f_A = \frac{1}{m(A)}\int_A f f A = m ( A ) 1 ∫ A f and u A = ∫ A u d m u_A = \int_A u\,\mathrm{d}m u A = ∫ A u d m for any measurable set A A A with m ( A ) ∈ ( 0 , + ∞ ) m(A) \in(0,+\infty) m ( A ) ∈ ( 0 , + ∞ ) and any function u u u such that the integral ∫ A u d m \int_A u\,\mathrm{d}m ∫ A u d m is well-defined.
Let ε > 0 \varepsilon> 0 ε > 0 . We say that W ⊆ X W \subseteq X W ⊆ X is an ε \varepsilon ε -net if for any distinct x , y ∈ W x,y \in W x , y ∈ W , we have d ( x , y ) ≥ ε d(x,y) \ge\varepsilon d ( x , y ) ≥ ε , and for any z ∈ X z \in X z ∈ X , there exists x ∈ W x \in W x ∈ W such that d ( x , z ) < ε d(x,z) < \varepsilon d ( x , z ) < ε . Since ( X , d ) (X,d) ( X , d ) is separable, all ε \varepsilon ε -nets are countable.
We say that the chain condition CC holds if there exists a constant C c c > 0 C_{\mathrm{cc}} > 0 C cc > 0 such that for any x , y ∈ X x,y \in X x , y ∈ X and any integer n ≥ 1 n \ge1 n ≥ 1 , there exists a sequence { x k : 0 ≤ k ≤ n } \{x_k : 0 \le k \le n\} { x k : 0 ≤ k ≤ n } of points in X X X with x 0 = x x_0=x x 0 = x and x n = y x_n=y x n = y such that
d ( x k , x k − 1 ) ≤ C c c d ( x , y ) n for any k = 1 , … , n . (1) d(x_k,x_{k-1}) \le C_{\mathrm{cc}}\frac{d(x,y)}{n}\quad\text{for any }k=1,\ldots,n. \tag*{(1)} d ( x k , x k − 1 ) ≤ C cc n d ( x , y ) for any k = 1 , … , n . (1)
Throughout this paper, we always assume CC.
We say that the volume doubling condition VD holds if there exists a constant C V D > 0 C_{\mathrm{VD}} > 0 C VD > 0 such that
V ( x , 2 r ) ≤ C V D V ( x , r ) for any x ∈ X , r > 0. (2) V(x,2r) \le C_{\mathrm{VD}}V(x,r)\quad\text{for any }x\in X,\ r>0. \tag*{(2)} V ( x , 2 r ) ≤ C VD V ( x , r ) for any x ∈ X , r > 0. (2)
Under VD, the metric space ( X , d ) (X,d) ( X , d ) is doubling, that is, there exists an integer N ≥ 1 N \ge1 N ≥ 1 , depending only on C V D C_{\mathrm{VD}} C VD , such that for any r > 0 r > 0 r > 0 , any ball in ( X , d ) (X,d) ( X , d ) with radius 2 r 2r 2 r can be covered by at most N N N balls with radius r r r ; see [16 Remarque on p. 67] or [23 pp. 81–82] .
For d h > 0 d_h > 0 d h > 0 , we say that the Ahlfors regular condition V ( d h ) V(d_h) V ( d h ) holds if there exists a constant C R > 0 C_R > 0 C R > 0 such that
1 C R r d h ≤ V ( x , r ) ≤ C R r d h for any x ∈ X , r > 0. (3) \frac{1}{C_R}r^{d_h} \le V(x,r) \le C_Rr^{d_h}\quad\text{for any }x\in X,\ r>0. \tag*{(3)} C R 1 r d h ≤ V ( x , r ) ≤ C R r d h for any x ∈ X , r > 0. (3)
We say that ( E , F ) (\mathcal{E},\mathcal{F}) ( E , F ) is a p p p -energy on ( X , d , m ) (X,d,m) ( X , d , m ) if F \mathcal{F} F is a dense subspace of L p ( X ; m ) L^p(X;m) L p ( X ; m ) and E : F → [ 0 , + ∞ ) \mathcal{E}:\mathcal{F}\to[0,+\infty) E : F → [ 0 , + ∞ ) satisfies the following conditions.
( E ) 1 / p (\mathcal{E})^{1/p} ( E ) 1/ p is a seminorm on F \mathcal{F} F , that is, for any f , g ∈ F f,g\in\mathcal{F} f , g ∈ F , c ∈ R c\in\mathbb{R} c ∈ R , we have E ( f ) ≥ 0 \mathcal{E}(f)\ge0 E ( f ) ≥ 0 , E ( c f ) 1 / p = ∣ c ∣ E ( f ) 1 / p \mathcal{E}(cf)^{1/p}=|c|\mathcal{E}(f)^{1/p} E ( c f ) 1/ p = ∣ c ∣ E ( f ) 1/ p , and E ( f + g ) 1 / p ≤ E ( f ) 1 / p + E ( g ) 1 / p \mathcal{E}(f+g)^{1/p}\le\mathcal{E}(f)^{1/p}+\mathcal{E}(g)^{1/p} E ( f + g ) 1/ p ≤ E ( f ) 1/ p + E ( g ) 1/ p
(Closed property) ( F , ( E ( ⋅ ) + ∥ ⋅ ∥ L p ( X ; m ) p ) 1 / p ) (\mathcal{F},(\mathcal{E}(\cdot)+\|\cdot\|_{L^p(X;m)}^p)^{1/p}) ( F , ( E ( ⋅ ) + ∥ ⋅ ∥ L p ( X ; m ) p ) 1/ p ) is a Banach space.
(Markovian property) For any φ ∈ C ( R ) \varphi\in C(\mathbb{R}) φ ∈ C ( R ) with φ ( 0 ) = 0 \varphi(0)=0 φ ( 0 ) = 0 and ∣ φ ( t ) − φ ( s ) ∣ ≤ ∣ t − s ∣ |\varphi(t)-\varphi(s)|\le|t-s| ∣ φ ( t ) − φ ( s ) ∣ ≤ ∣ t − s ∣ for any t , s ∈ R t,s\in\mathbb{R} t , s ∈ R , for any f ∈ F f\in\mathcal{F} f ∈ F , we have φ ( f ) ∈ F \varphi(f)\in\mathcal{F} φ ( f ) ∈ F and E ( φ ( f ) ) ≤ E ( f ) \mathcal{E}(\varphi(f))\le\mathcal{E}(f) E ( φ ( f )) ≤ E ( f ) .
(Regular property) F ∩ C c ( X ) \mathcal{F}\cap C_c(X) F ∩ C c ( X ) is uniformly dense in C c ( X ) C_c(X) C c ( X ) and ( E ( ⋅ ) + ∥ ⋅ ∥ L p ( X ; m ) p ) 1 / p (\mathcal{E}(\cdot)+\|\cdot\|_{L^p(X;m)}^p)^{1/p} ( E ( ⋅ ) + ∥ ⋅ ∥ L p ( X ; m ) p ) 1/ p -dense in F \mathcal{F} F .
(Strongly local property) For any f , g ∈ F f,g\in\mathcal{F} f , g ∈ F with compact support and g g g constant in an open neighborhood of supp ( f ) \operatorname{supp}(f) supp ( f ) , we have E ( f + g ) = E ( f ) + E ( g ) \mathcal{E}(f+g)=\mathcal{E}(f)+\mathcal{E}(g) E ( f + g ) = E ( f ) + E ( g ) .
(p p p -Clarkson’s inequality) For any f , g ∈ F f,g\in\mathcal{F} f , g ∈ F , we have
E ( f + g ) + E ( f − g ) ≥ 2 ( E ( f ) 1 p − 1 + E ( g ) 1 p − 1 ) p − 1 if p ∈ ( 1 , 2 ] , E ( f + g ) + E ( f − g ) ≤ 2 ( E ( f ) 1 p − 1 + E ( g ) 1 p − 1 ) p − 1 if p ∈ [ 2 , + ∞ ) . (4) \begin{aligned}
\mathcal{E}(f+g)+\mathcal{E}(f-g)&\ge2\left(\mathcal{E}(f)^{\frac{1}{p-1}}+\mathcal{E}(g)^{\frac{1}{p-1}}\right)^{p-1} &&\text{if }p\in(1,2],\\
\mathcal{E}(f+g)+\mathcal{E}(f-g)&\le2\left(\mathcal{E}(f)^{\frac{1}{p-1}}+\mathcal{E}(g)^{\frac{1}{p-1}}\right)^{p-1} &&\text{if }p\in[2,+\infty). \tag*{(4)}
\end{aligned} E ( f + g ) + E ( f − g ) E ( f + g ) + E ( f − g ) ≥ 2 ( E ( f ) p − 1 1 + E ( g ) p − 1 1 ) p − 1 ≤ 2 ( E ( f ) p − 1 1 + E ( g ) p − 1 1 ) p − 1 if p ∈ ( 1 , 2 ] , if p ∈ [ 2 , + ∞ ) . (4)
We also assume the following condition.
( F ∩ L ∞ ( X ; m ) (\mathcal{F}\cap L^\infty(X;m) ( F ∩ L ∞ ( X ; m ) is an algebra) For any f , g ∈ F ∩ L ∞ ( X ; m ) f,g\in\mathcal{F}\cap L^\infty(X;m) f , g ∈ F ∩ L ∞ ( X ; m ) , we have f g ∈ F fg\in\mathcal{F} f g ∈ F and
E ( f g ) 1 / p ≤ ∥ f ∥ L ∞ ( X ; m ) E ( g ) 1 / p + ∥ g ∥ L ∞ ( X ; m ) E ( f ) 1 / p . (5) \mathcal{E}(fg)^{1/p}\le\|f\|_{L^\infty(X;m)}\mathcal{E}(g)^{1/p}+\|g\|_{L^\infty(X;m)}\mathcal{E}(f)^{1/p}. \tag*{(5)} E ( f g ) 1/ p ≤ ∥ f ∥ L ∞ ( X ; m ) E ( g ) 1/ p + ∥ g ∥ L ∞ ( X ; m ) E ( f ) 1/ p . (5)
For any λ > 0 \lambda>0 λ > 0 , denote
E λ ( ⋅ ) : = E ( ⋅ ) + λ ∥ ⋅ ∥ L p ( X ; m ) p . \mathcal{E}_\lambda(\cdot):=\mathcal{E}(\cdot)+\lambda\|\cdot\|_{L^p(X;m)}^p. E λ ( ⋅ ) := E ( ⋅ ) + λ ∥ ⋅ ∥ L p ( X ; m ) p .
The generalized p p p -contraction property introduced in [27 ] implies Cla and Alg and holds on a large family of metric measure spaces.
By [37 Theorem 2.4] , a p p p -energy ( E , F ) (\mathcal{E},\mathcal{F}) ( E , F ) corresponds to a (canonical) p p p -energy measure Γ : F × B ( X ) → [ 0 , + ∞ ) \Gamma:\mathcal{F}\times\mathcal{B}(X)\to[0,+\infty) Γ : F × B ( X ) → [ 0 , + ∞ ) , ( f , A ) ↦ Γ ( f ) ( A ) (f,A)\mapsto\Gamma(f)(A) ( f , A ) ↦ Γ ( f ) ( A ) satisfying the following conditions. (a) For any f ∈ F f \in\mathcal{F} f ∈ F , Γ ( f ) ( ⋅ ) \Gamma(f)(\cdot) Γ ( f ) ( ⋅ ) is a positive Radon measure on X X X with Γ ( f ) ( X ) = E ( f ) \Gamma(f)(X)=\mathcal{E}(f) Γ ( f ) ( X ) = E ( f ) .
(b) For any A ∈ B ( X ) A \in\mathcal{B}(X) A ∈ B ( X ) , Γ ( ⋅ ) ( A ) 1 / p \Gamma(\cdot)(A)^{1/p} Γ ( ⋅ ) ( A ) 1/ p is a seminorm on F \mathcal{F} F .
(c) For any f , g ∈ F ∩ C c ( X ) f,g \in\mathcal{F}\cap C_c(X) f , g ∈ F ∩ C c ( X ) , A ∈ B ( X ) A \in\mathcal{B}(X) A ∈ B ( X ) , if f − g f-g f − g is constant on A A A , then Γ ( f ) ( A ) = Γ ( g ) ( A ) \Gamma(f)(A)=\Gamma(g)(A) Γ ( f ) ( A ) = Γ ( g ) ( A ) .
(d) (p p p -Clarkson’s inequality) For any f , g ∈ F f,g \in\mathcal{F} f , g ∈ F , for any A ∈ B ( X ) A \in\mathcal{B}(X) A ∈ B ( X ) , we have
{ Γ ( f + g ) ( A ) + Γ ( f − g ) ( A ) ≥ 2 ( Γ ( f ) ( A ) 1 p − 1 + Γ ( g ) ( A ) 1 p − 1 ) p − 1 if p ∈ ( 1 , 2 ] , Γ ( f + g ) ( A ) + Γ ( f − g ) ( A ) ≤ 2 ( Γ ( f ) ( A ) 1 p − 1 + Γ ( g ) ( A ) 1 p − 1 ) p − 1 if p ∈ [ 2 , + ∞ ) . \begin{cases}
\Gamma(f+g)(A)+\Gamma(f-g)(A)\ge2\left(\Gamma(f)(A)^{\frac{1}{p-1}}+\Gamma(g)(A)^{\frac{1}{p-1}}\right)^{p-1} & \text{if }p\in(1,2],\\
\Gamma(f+g)(A)+\Gamma(f-g)(A)\le2\left(\Gamma(f)(A)^{\frac{1}{p-1}}+\Gamma(g)(A)^{\frac{1}{p-1}}\right)^{p-1} & \text{if }p\in[2,+\infty).
\end{cases} ⎩ ⎨ ⎧ Γ ( f + g ) ( A ) + Γ ( f − g ) ( A ) ≥ 2 ( Γ ( f ) ( A ) p − 1 1 + Γ ( g ) ( A ) p − 1 1 ) p − 1 Γ ( f + g ) ( A ) + Γ ( f − g ) ( A ) ≤ 2 ( Γ ( f ) ( A ) p − 1 1 + Γ ( g ) ( A ) p − 1 1 ) p − 1 if p ∈ ( 1 , 2 ] , if p ∈ [ 2 , + ∞ ) .
(e) (Chain rule) For any f , g ∈ F ∩ C c ( X ) f,g \in\mathcal{F}\cap C_c(X) f , g ∈ F ∩ C c ( X ) , for any piecewise C 1 C^1 C 1 functions φ , ψ : R → R \varphi,\psi:\mathbb{R}\to\mathbb{R} φ , ψ : R → R with φ ( 0 ) = ψ ( 0 ) = 0 \varphi(0)=\psi(0)=0 φ ( 0 ) = ψ ( 0 ) = 0 , we have
d Γ ( φ ( f ) ; ψ ( g ) ) = sgn ( φ ′ ( f ) ) ∣ φ ′ ( f ) ∣ p − 1 ψ ′ ( g ) d Γ ( f ; g ) , \mathrm{d}\Gamma(\varphi(f);\psi(g))=\operatorname{sgn}(\varphi'(f))\lvert\varphi'(f)\rvert^{p-1}\psi'(g)\mathrm{d}\Gamma(f;g), d Γ ( φ ( f ) ; ψ ( g )) = sgn ( φ ′ ( f )) ∣ φ ′ ( f ) ∣ p − 1 ψ ′ ( g ) d Γ ( f ; g ) ,
where Γ ( f ; g ) \Gamma(f;g) Γ ( f ; g ) is a signed measure given by Γ ( f ; g ) = 1 p d d t Γ ( f + t g ) ∣ t = 0 \Gamma(f;g)=\left.\frac{1}{p}\frac{\mathrm{d}}{\mathrm{d}t}\Gamma(f+tg)\right|_{t=0} Γ ( f ; g ) = p 1 d t d Γ ( f + t g ) t = 0 .
(f) (Leibniz rule) For any f , g , h ∈ F ∩ C c ( X ) f,g,h \in\mathcal{F}\cap C_c(X) f , g , h ∈ F ∩ C c ( X ) , we have d Γ ( f ; g h ) = g d Γ ( f ; h ) + h d Γ ( f ; g ) \mathrm{d}\Gamma(f;gh)=g\mathrm{d}\Gamma(f;h)+h\mathrm{d}\Gamma(f;g) d Γ ( f ; g h ) = g d Γ ( f ; h ) + h d Γ ( f ; g ) .
Using the chain rule, we have the following condition.
(Strong sub-additivity) For any f , g ∈ F f,g \in\mathcal{F} f , g ∈ F , we have f ∨ g , f ∧ g ∈ F f\vee g,f\wedge g\in\mathcal{F} f ∨ g , f ∧ g ∈ F and
E ( f ∨ g ) + E ( f ∧ g ) ≤ E ( f ) + E ( g ) . \mathcal{E}(f\vee g)+\mathcal{E}(f\wedge g)\le\mathcal{E}(f)+\mathcal{E}(g). E ( f ∨ g ) + E ( f ∧ g ) ≤ E ( f ) + E ( g ) .
SubAdd
By Cla and [27 Theorem 3.7] , for any u , v ∈ F u,v\in\mathcal{F} u , v ∈ F , the derivative
E ( u ; v ) : = 1 p d d t E ( u + t v ) ∣ t = 0 ∈ R \mathcal{E}(u;v):=\left.\frac{1}{p}\frac{\mathrm{d}}{\mathrm{d}t}\mathcal{E}(u+tv)\right|_{t=0}\in\mathbb{R} E ( u ; v ) := p 1 d t d E ( u + t v ) t = 0 ∈ R
exists, the map E ( u ; ⋅ ) : F → R \mathcal{E}(u;\cdot):\mathcal{F}\to\mathbb{R} E ( u ; ⋅ ) : F → R is linear, and E ( u ; u ) = E ( u ) \mathcal{E}(u;u)=\mathcal{E}(u) E ( u ; u ) = E ( u ) .
For any bounded open subset U ⊆ X U\subseteq X U ⊆ X , let F ( U ) \mathcal{F}(U) F ( U ) be the closure of F ∩ C c ( U ) \mathcal{F}\cap C_c(U) F ∩ C c ( U ) with respect to E 1 / p \mathcal{E}^{1/p} E 1/ p . By [42 Lemma 6.2] ,
F ( U ) = { u ∈ F : u ~ = 0 q.e. on X ∖ U } , \mathcal{F}(U)=\{u\in\mathcal{F}:\widetilde{u}=0\ \text{q.e. on }X\setminus U\}, F ( U ) = { u ∈ F : u = 0 q.e. on X ∖ U } ,
where u ~ \widetilde{u} u is a quasi-continuous modification of u u u ; see [44 Section 8] for more details.
We say that u ∈ F u\in\mathcal{F} u ∈ F is harmonic in U U U if E ( u ; v ) = 0 \mathcal{E}(u;v)=0 E ( u ; v ) = 0 for any v ∈ F ∩ C c ( U ) v\in\mathcal{F}\cap C_c(U) v ∈ F ∩ C c ( U ) . We say that u u u is superharmonic in U U U (resp. subharmonic in U U U ) if E ( u ; v ) ≥ 0 \mathcal{E}(u;v)\ge0 E ( u ; v ) ≥ 0 (resp. E ( u ; v ) ≤ 0 \mathcal{E}(u;v)\le0 E ( u ; v ) ≤ 0 ) for any non-negative v ∈ F ∩ C c ( U ) v\in\mathcal{F}\cap C_c(U) v ∈ F ∩ C c ( U ) . Since U U U is bounded, these definitions are unchanged when F ∩ C c ( U ) \mathcal{F}\cap C_c(U) F ∩ C c ( U ) is replaced by F ( U ) \mathcal{F}(U) F ( U ) , retaining the non-negativity requirement in the superharmonic and subharmonic cases.
Let β p > 0 \beta_p>0 β p > 0 . We say that the Poincaré inequality PI ( β p ) \operatorname{PI}(\beta_p) PI ( β p ) holds if there exist constants C P I > 0 C_{PI}>0 C P I > 0 and A P I ≥ 1 A_{PI}\ge1 A P I ≥ 1 such that for any ball B ( x , r ) B(x,r) B ( x , r ) and any f ∈ F f\in\mathcal{F} f ∈ F ,
∫ B ( x , r ) ∣ f − f B ( x , r ) ∣ p d m ≤ C P I r β p ∫ B ( x , A P I r ) d Γ ( f ) . \int_{B(x,r)}\lvert f-f_{B(x,r)}\rvert^p\mathrm{d}m\le C_{PI}r^{\beta_p}\int_{B(x,A_{PI}r)}\mathrm{d}\Gamma(f). ∫ B ( x , r ) ∣ f − f B ( x , r ) ∣ p d m ≤ C P I r β p ∫ B ( x , A P I r ) d Γ ( f ) .
PI ( β p ) \operatorname{PI}(\beta_p) PI ( β p )
Let A 1 , A 2 ∈ B ( X ) A_1,A_2\in\mathcal{B}(X) A 1 , A 2 ∈ B ( X ) . We define the E \mathcal{E} E -capacity between A 1 A_1 A 1 and A 2 A_2 A 2 as
cap ( A 1 , A 2 ) : = inf { E ( ϕ ) : ϕ ∈ F , ϕ = 1 in an open neighborhood of A 1 , ϕ = 0 in an open neighborhood of A 2 } , \operatorname{cap}(A_1,A_2):=\inf\left\{\mathcal{E}(\phi):\phi\in\mathcal{F},\quad\phi=1\text{ in an open neighborhood of }A_1,\quad\phi=0\text{ in an open neighborhood of }A_2\right\}, cap ( A 1 , A 2 ) := inf { E ( ϕ ) : ϕ ∈ F , ϕ = 1 in an open neighborhood of A 1 , ϕ = 0 in an open neighborhood of A 2 } ,
where we use the convention that inf ∅ = + ∞ \inf\varnothing=+\infty inf ∅ = + ∞ .
We say that the capacity upper bound cap ( β p ) ≤ \operatorname{cap}(\beta_p)_{\le} cap ( β p ) ≤ holds if there exist constants C cap > 0 C_{\operatorname{cap}}>0 C cap > 0 and A cap > 1 A_{\operatorname{cap}}>1 A cap > 1 such that for any ball B ( x , r ) B(x,r) B ( x , r ) ,
cap ( B ( x , r ) , X ∖ B ( x , A cap r ) ) ≤ C cap V ( x , r ) r β p . \operatorname{cap}(B(x,r),X\setminus B(x,A_{\operatorname{cap}}r))\le C_{\operatorname{cap}}\frac{V(x,r)}{r^{\beta_p}}. cap ( B ( x , r ) , X ∖ B ( x , A cap r )) ≤ C cap r β p V ( x , r ) .
cap ( β p ) ≤ \operatorname{cap}(\beta_p)_{\le} cap ( β p ) ≤
Fix p > 1 p>1 p > 1 , assume VD, and let ( E , F ) (\mathcal{E},\mathcal{F}) ( E , F ) be a p p p -energy satisfying PI ( β p ) \operatorname{PI}(\beta_p) PI ( β p ) and cap ( β p ) ≤ \operatorname{cap}(\beta_p)_{\le} cap ( β p ) ≤ . By [44 Proposition 2.1 and Remark 2.2] ,
β p ≥ p . (6) \beta_p\ge p. \tag*{(6)} β p ≥ p . (6)
By [18 ] , the substantially stronger cutoff Sobolev inequality CS ( β p ) \operatorname{CS}(\beta_p) CS ( β p ) holds. Consequently, when applying results from the literature that assume VD \operatorname{VD} VD , PI ( β p ) \operatorname{PI}(\beta_p) PI ( β p ) , and CS ( β p ) \operatorname{CS}(\beta_p) CS ( β p ) , it suffices to assume VD \operatorname{VD} VD , PI ( β p ) \operatorname{PI}(\beta_p) PI ( β p ) , and cap ( β p ) ≤ \operatorname{cap}(\beta_p)_{\le} cap ( β p ) ≤ . If V ( d h ) \operatorname{V}(d_h) V ( d h ) also holds, then [44 Proposition 2.1] gives
p ≤ β p ≤ d h + p − 1 , (7) p \le\beta_p \le d_h+p-1, \tag*{(7)} p ≤ β p ≤ d h + p − 1 , (7)
see Figure 1 .
In what follows, we use the superscript ( p ) (p) ( p ) to indicate explicit dependence on p p p . For example, we write ( E ( p ) , F ( p ) ) (\mathcal{E}^{(p)},\mathcal{F}^{(p)}) ( E ( p ) , F ( p ) ) for a p p p -energy, PI ( p ) ( ⋅ ) \operatorname{PI}^{(p)}(\cdot) PI ( p ) ( ⋅ ) for the Poincaré inequality, and C ( p ) \mathcal{C}^{(p)} C ( p ) and A ( p ) \mathcal{A}^{(p)} A ( p ) for constants depending on p p p .
The first main result of this paper is the following dichotomy.
Theorem 2.1. Assume VD \operatorname{VD} VD . Let I ⊂ ( 1 , + ∞ ) I \subset(1,+\infty) I ⊂ ( 1 , + ∞ ) be an open interval. Assume that, for any p ∈ I p \in I p ∈ I , there exists a p p p -energy ( E ( p ) , F ( p ) ) (\mathcal{E}^{(p)},\mathcal{F}^{(p)}) ( E ( p ) , F ( p ) ) satisfying PI ( p ) ( β p ) \operatorname{PI}^{(p)}(\beta_p) PI ( p ) ( β p ) and cap ( p ) ( β p ) ≤ \operatorname{cap}^{(p)}(\beta_p)_{\le} cap ( p ) ( β p ) ≤ , where β p ≥ p \beta_p \ge p β p ≥ p by (2.1). Then exactly one of the following alternatives holds:
(i) β p = p \beta_p=p β p = p for any p ∈ I p\in I p ∈ I ;
(ii) the function p ↦ β p p p\mapsto\frac{\beta_p}{p} p ↦ p β p takes values in ( 1 , + ∞ ) (1,+\infty) ( 1 , + ∞ ) and is strictly decreasing on I I I .
Remark 2.2. Roughly speaking, β p \beta_p β p can be interpreted as the scaling order of the p p p -Laplacian. In Euclidean space and other settings with a classical gradient structure, we typically have β p = p \beta_p=p β p = p , so that β p p = 1 \frac{\beta_p}{p}=1 p β p = 1 corresponds to the order of a first derivative. On fractals, we typically have β p > p \beta_p>p β p > p , and the ratio β p p \frac{\beta_p}{p} p β p may instead be viewed heuristically as the “effective” order of an analogue of the gradient. Its deviation from 1 thus reflects a departure from classical first-order scaling. In this sense, the above result means that the scaling of the p p p -energy becomes “progressively” closer to classical first-order scaling as p p p increases.
Remark 2.3. By [43 Theorem 2.1, Lemmas 5.1 and 5.2] , the dichotomy holds with (ii) replaced by the following weaker alternative:
(ii′ ' ′ ) the function p ↦ β p p p\mapsto\frac{\beta_p}{p} p ↦ p β p takes values in ( 1 , + ∞ ) (1,+\infty) ( 1 , + ∞ ) and is decreasing on I I I .
It therefore remains only to prove that the monotonicity in (ii′ ' ′ ) is strict.
Remark 2.4. We sketch how the strict monotonicity result relates to the unboundedness of the Riesz transform for any p > 2 p>2 p > 2 on the fractal-like spaces considered in [20 ] . The following discussion is heuristic and does not constitute a formal proof.
Assume that on the corresponding fractals, for any p ∈ ( 1 , 2 ] p\in(1,2] p ∈ ( 1 , 2 ] , there exists a p p p -energy satisfying PI ( p ) ( β p ) \operatorname{PI}^{(p)}(\beta_p) PI ( p ) ( β p ) and cap ( p ) ( β p ) ≤ \operatorname{cap}^{(p)}(\beta_p)_{\le} cap ( p ) ( β p ) ≤ . Then suitable variants of these two inequalities also hold on the fractal-like graphs and manifolds. In the graph or manifold setting considered there, a suitable gradient operator ∇ \nabla ∇ is well-defined. We use the notation of [20 ] and refer the reader to that paper for further details. By duality, to establish the unboundedness of the Riesz transform for any p > 2 p>2 p > 2 , it suffices to prove the failure of the reverse Riesz inequality ( R R p , 1 2 ) (\mathrm{RR}_{p,\frac{1}{2}}) ( RR p , 2 1 ) for any p ∈ ( 1 , 2 ) p\in(1,2) p ∈ ( 1 , 2 ) . Under the sub-Gaussian upper heat kernel estimate ( U E β ) (\mathrm{UE}_{\beta}) ( UE β ) with β > 2 \beta>2 β > 2 , [20 Proposition 4.6 and Remark 4.7] show that ( R R p , 1 2 ) (\mathrm{RR}_{p,\frac{1}{2}}) ( RR p , 2 1 ) implies ( P p , β 2 ) (\mathrm{P}_{p,\frac{\beta}{2}}) ( P p , 2 β ) , that is, for any ball B = B ( x , r ) B=B(x,r) B = B ( x , r ) and any f f f supported in B B B ,
∥ f ∥ L p ( B ) ≲ r β 2 ∥ ∇ f ∥ L p ( B ) . (8) \lVert f\rVert_{L^{p}(B)}\lesssim r^{\frac{\beta}{2}}\lVert\nabla f\rVert_{L^{p}(B)}. \tag*{(8)} ∥ f ∥ L p ( B ) ≲ r 2 β ∥ ∇ f ∥ L p ( B ) . (8)
Here, β \beta β in ( U E β ) (\mathrm{UE}_{\beta}) ( UE β ) corresponds to the 2-walk dimension β 2 \beta_{2} β 2 in our setting. The transferred version of cap ( p ) ( β p ) ≤ \operatorname{cap}^{(p)}(\beta_{p})_{\leq} cap ( p ) ( β p ) ≤ ensures that for any ball B = B ( x , r ) B=B(x,r) B = B ( x , r ) with r > 1 r>1 r > 1 , there exists a function f f f supported in B B B such that 0 ≤ f ≤ 1 0\leq f\leq1 0 ≤ f ≤ 1 in B B B , f = 1 f=1 f = 1 in 1 2 B \frac{1}{2}B 2 1 B ,
∥ ∇ f ∥ L p ( B ) p ≲ m ( B ) r β p . (9) \lVert\nabla f\rVert_{L^{p}(B)}^{p}\lesssim\frac{m(B)}{r^{\beta_{p}}}. \tag*{(9)} ∥ ∇ f ∥ L p ( B ) p ≲ r β p m ( B ) . (9)
Moreover,
∥ f ∥ L p ( B ) p ≥ m ( 1 2 B ) ≳ m ( B ) , \lVert f\rVert_{L^{p}(B)}^{p}\geq m\left(\frac{1}{2}B\right)\gtrsim m(B), ∥ f ∥ L p ( B ) p ≥ m ( 2 1 B ) ≳ m ( B ) ,
where the comparability follows from the volume doubling condition VD on the fractal-like space. Combining this with (8) and (9) , we obtain
r β p p ≲ r β 2 2 for any r > 1. (10) r^{\frac{\beta_{p}}{p}}\lesssim r^{\frac{\beta_{2}}{2}}\quad\text{for any }r>1. \tag*{(10)} r p β p ≲ r 2 β 2 for any r > 1. (10)
Since β 2 > 2 \beta_{2}>2 β 2 > 2 , Theorem 2.1 gives β p p > β 2 2 \frac{\beta_{p}}{p}>\frac{\beta_{2}}{2} p β p > 2 β 2 for any p ∈ ( 1 , 2 ) p\in(1,2) p ∈ ( 1 , 2 ) . Letting r → + ∞ r\to+\infty r → + ∞ in (10) , we obtain a contradiction.
The second main result of this paper is as follows.
Theorem 2.5. Assume V ( d h ) V(d_{h}) V ( d h ) . Let I ⊆ ( 1 , + ∞ ) I\subseteq(1,+\infty) I ⊆ ( 1 , + ∞ ) be an open interval. Assume that, for any p ∈ I p\in I p ∈ I , there exists a p p p -energy ( E ( p ) , F ( p ) ) (\mathcal{E}^{(p)},\mathcal{F}^{(p)}) ( E ( p ) , F ( p ) ) satisfying P I ( p ) ( β p ) \mathrm{PI}^{(p)}(\beta_{p}) PI ( p ) ( β p ) and cap ( p ) ( β p ) ≤ \operatorname{cap}^{(p)}(\beta_{p})_{\leq} cap ( p ) ( β p ) ≤ , where β p ≤ d h + p − 1 \beta_{p}\leq d_{h}+p-1 β p ≤ d h + p − 1 by (7) . Then:
(1) The function p ↦ γ p : = β p − d h p − 1 p\mapsto\gamma_{p}:=\frac{\beta_{p}-d_{h}}{p-1} p ↦ γ p := p − 1 β p − d h is increasing on I I I .
(2) If γ p = 1 \gamma_{p}=1 γ p = 1 for some p ∈ I p\in I p ∈ I , then γ q = 1 \gamma_{q}=1 γ q = 1 for any q ∈ I q\in I q ∈ I .
(3) Let p , q ∈ I p,q\in I p , q ∈ I satisfy p < q p<q p < q . If γ p < 1 \gamma_{p}<1 γ p < 1 , or equivalently, γ q < 1 \gamma_{q}<1 γ q < 1 , then γ p < γ q \gamma_{p}<\gamma_{q} γ p < γ q .
Consequently, exactly one of the following alternatives holds:
(a) β p = d h + p − 1 \beta_{p}=d_{h}+p-1 β p = d h + p − 1 for any p ∈ I p\in I p ∈ I ;
(b) the function p ↦ β p − d h p − 1 p\mapsto\frac{\beta_{p}-d_{h}}{p-1} p ↦ p − 1 β p − d h takes values in ( − ∞ , 1 ) (-\infty,1) ( − ∞ , 1 ) and is strictly increasing on I I I .
Remark 2.6. Roughly speaking, for p p p sufficiently large that β p > d h \beta_{p}>d_{h} β p > d h and γ p ∈ ( 0 , 1 ] \gamma_{p}\in(0,1] γ p ∈ ( 0 , 1 ] , the exponent γ p \gamma_{p} γ p describes the Hölder regularity of p p p -harmonic functions (see [27 Theorem 6.31] ), which may be viewed heuristically as among the “most” regular functions in the domain of the p p p -energy. The case γ p = 1 \gamma_{p}=1 γ p = 1 occurs in spaces with some kind of tree structure, such as R \mathbb{R} R and the Vicsek set (see [10 ] ), where p p p -harmonic functions are indeed Lipschitz. If γ p < 1 \gamma_{p}<1 γ p < 1 for some, or equivalently any, admissible p p p , the above result and (7) imply that γ p \gamma_{p} γ p is strictly increasing and converges to 1 1 1 as p → + ∞ p\to+\infty p → + ∞ . Hence, p p p -harmonic functions become “strictly smoother” as p p p increases, in the sense that their Hölder regularity exponents strictly increase towards the Lipschitz exponent 1 1 1 .
As a direct corollary, we obtain the following strict monotonicity result, which covers both the concrete fractals discussed in the Introduction and the Sierpiński carpet.
Corollary 2.7. Let K K K be the Sierpiński gasket, the Sierpiński carpet, or the golden ratio Sierpiński gasket, and let d h d_{h} d h be its Hausdorff dimension. For any p ∈ ( 1 , + ∞ ) p\in(1,+\infty) p ∈ ( 1 , + ∞ ) , let ( E ( p ) , F ( p ) ) (\mathcal{E}^{(p)},\mathcal{F}^{(p)}) ( E ( p ) , F ( p ) ) be the p p p -energy constructed in [24 ] , [34 ] , or [15 ] , respectively, with p p p -walk dimension β p \beta_{p} β p . Then
the function p ↦ β p p p\mapsto\frac{\beta_{p}}{p} p ↦ p β p takes values in ( 1 , + ∞ ) (1,+\infty) ( 1 , + ∞ ) and is strictly decreasing on ( 1 , + ∞ ) (1,+\infty) ( 1 , + ∞ ) ;
the function p ↦ β p − d h p − 1 p\mapsto\frac{\beta_{p}-d_{h}}{p-1} p ↦ p − 1 β p − d h takes values in ( − ∞ , 1 ) (-\infty,1) ( − ∞ , 1 ) and is strictly increasing on ( 1 , + ∞ ) (1,+\infty) ( 1 , + ∞ ) .
Proof. On the Sierpiński gasket, the resistance estimate R ( p ) ( β p ) R^{(p)}(\beta_p) R ( p ) ( β p ) , defined below, holds by [30 Theorem 4.6] and [27 Theorem 8.30] . On the golden ratio Sierpiński gasket, the same estimate holds by [15 Theorem 2] . By [42 Theorem 2.3] , this resistance estimate is equivalent to the conjunction of PI ( p ) ( β p ) \operatorname{PI}^{(p)}(\beta_p) PI ( p ) ( β p ) and cap ( p ) ( β p ) ≤ \operatorname{cap}^{(p)}(\beta_p)\leq cap ( p ) ( β p ) ≤ . Hence, both inequalities hold on these two fractals. On the Sierpiński carpet, these two inequalities hold by [34 Theorem 8.21 and Proposition 6.21] . By the well-known fact that 2 < β 2 < d h + 1 2<\beta_2<d_h+1 2 < β 2 < d h + 1 on these fractals, the results follow directly from Theorems 2.1 and 2.5 .
This paper is organized as follows. In Section 3, we collect some preliminary results for later use. In Section 4, we give the proof of Theorem 2.1. In Section 5, we give the proof of Theorem 2.5(1). In Section 6, we give the proof of Theorem 2.5(2). In Section 7, we give the proof of Theorem 2.5(3).
Preliminaries
In this section, we collect some preliminary results for later use.
For any p > 1 p>1 p > 1 , β > 0 \beta>0 β > 0 , and r > 0 r>0 r > 0 , let U ⊆ X U\subseteq X U ⊆ X be open and let u u u be a measurable function. We define
E β p , U ( p ) ( u , r ) : = 1 r β ∫ U ( 1 V ( x , r ) ∫ B ( x , r ) ∣ u ( x ) − u ( y ) ∣ p m ( d y ) ) m ( d x ) . \mathcal{E}^{(p)}_{\beta_p,U}(u,r):=\frac{1}{r^\beta}\int_U\left(\frac{1}{V(x,r)}\int_{B(x,r)}|u(x)-u(y)|^p m(dy)\right)m(dx). E β p , U ( p ) ( u , r ) := r β 1 ∫ U ( V ( x , r ) 1 ∫ B ( x , r ) ∣ u ( x ) − u ( y ) ∣ p m ( d y ) ) m ( d x ) .
and
U ( r ) : = ⋃ x ∈ U B ( x , r ) . U(r):=\bigcup_{x\in U}B(x,r). U ( r ) := x ∈ U ⋃ B ( x , r ) .
For any α > 0 \alpha>0 α > 0 , define the Besov space
B p , α ( X ) : = { u ∈ L p ( X ; m ) : sup r > 0 E α , X ( p ) ( u , r ) < + ∞ } . \mathcal{B}^{p,\alpha}(X):=\left\{u\in L^p(X;m):\sup_{r>0}\mathcal{E}^{(p)}_{\alpha,X}(u,r)<+\infty\right\}. B p , α ( X ) := { u ∈ L p ( X ; m ) : r > 0 sup E α , X ( p ) ( u , r ) < + ∞ } .
Obviously, B p , α ( X ) \mathcal{B}^{p,\alpha}(X) B p , α ( X ) is decreasing in α \alpha α and may become trivial if α \alpha α is too large. We define the following critical exponent
α p ( X ) : = sup { α > 0 : B p , α ( X ) contains non-constant functions } ≤ + ∞ . \alpha_p(X):=\sup\{\alpha>0:\mathcal{B}^{p,\alpha}(X)\text{ contains non-constant functions}\}\leq+\infty. α p ( X ) := sup { α > 0 : B p , α ( X ) contains non-constant functions } ≤ + ∞.
The value of α p ( X ) \alpha_p(X) α p ( X ) depends only on the metric measure space ( X , d , m ) (X,d,m) ( X , d , m ) . The following lemma records some basic properties of α p ( X ) \alpha_p(X) α p ( X ) .
Lemma 3.1 ([9 Theorem 4.1] ). Assume VD.
(i) For any p ∈ ( 1 , + ∞ ) p\in(1,+\infty) p ∈ ( 1 , + ∞ ) , we have α p ( X ) ≥ 1 \alpha_p(X)\geq1 α p ( X ) ≥ 1 .
(ii) The function p ↦ p α p ( X ) p\mapsto p\alpha_p(X) p ↦ p α p ( X ) is monotone increasing for p ∈ ( 1 , + ∞ ) p\in(1,+\infty) p ∈ ( 1 , + ∞ ) .
(iii) The function p ↦ α p ( X ) p\mapsto\alpha_p(X) p ↦ α p ( X ) is monotone decreasing for p ∈ ( 1 , + ∞ ) p\in(1,+\infty) p ∈ ( 1 , + ∞ ) .
Hence
(a) For p ∈ ( 1 , + ∞ ) p\in(1,+\infty) p ∈ ( 1 , + ∞ ) , the functions p ↦ p α p ( X ) p\mapsto p\alpha_p(X) p ↦ p α p ( X ) and p ↦ α p ( X ) p\mapsto\alpha_p(X) p ↦ α p ( X ) are continuous.
(b) If α p ( X ) < + ∞ \alpha_p(X)<+\infty α p ( X ) < + ∞ for some p ∈ ( 1 , + ∞ ) p\in(1,+\infty) p ∈ ( 1 , + ∞ ) , then α p ( X ) < + ∞ \alpha_p(X)<+\infty α p ( X ) < + ∞ for any p ∈ ( 1 , + ∞ ) p\in(1,+\infty) p ∈ ( 1 , + ∞ ) .
The value of α p ( X ) \alpha_p(X) α p ( X ) can be determined provided that there exists a p p p -energy satisfying certain functional inequalities.
Lemma 3.2 ([40 Theorem 4.6] ). Assume VD. Let ( E ( p ) , F ( p ) ) (\mathcal{E}^{(p)},\mathcal{F}^{(p)}) ( E ( p ) , F ( p ) ) be a p p p -energy satisfying PI ( p ) ( β p ) \operatorname{PI}^{(p)}(\beta_p) PI ( p ) ( β p ) and cap ( p ) ( β p ) ≤ \operatorname{cap}^{(p)}(\beta_p)\leq cap ( p ) ( β p ) ≤ . Then α p ( X ) = β p p \alpha_p(X)=\frac{\beta_p}{p} α p ( X ) = p β p .
We have the following results concerning the quantity E β p , U ( p ) \mathcal{E}^{(p)}_{\beta_p,U} E β p , U ( p ) .
Lemma 3.3 ([40 Proposition 3.5 and Equation (1.18)] ). Assume VD. Let ( E ( p ) , F ( p ) ) (\mathcal{E}^{(p)},\mathcal{F}^{(p)}) ( E ( p ) , F ( p ) ) be a p p p -energy satisfying PI ( p ) ( β p ) \operatorname{PI}^{(p)}(\beta_p) PI ( p ) ( β p ) . Then there exists a constant C ( p ) > 0 C^{(p)}>0 C ( p ) > 0 such that for any open subset U ⊆ X U\subseteq X U ⊆ X , u ∈ F ( p ) u\in\mathcal{F}^{(p)} u ∈ F ( p ) , and r > 0 r>0 r > 0 ,
E β p , U ( p ) ( u , r ) ≤ C ( p ) Γ ( p ) ( u ) ( U ( 2 A PI ( p ) r ) ) , \mathcal{E}^{(p)}_{\beta_p,U}(u,r)\leq C^{(p)}\Gamma^{(p)}(u)\left(U(2A_{\operatorname{PI}}^{(p)}r)\right), E β p , U ( p ) ( u , r ) ≤ C ( p ) Γ ( p ) ( u ) ( U ( 2 A PI ( p ) r ) ) , where A PI ( p ) ≥ 1 A_{\operatorname{PI}}^{(p)}\geq1 A PI ( p ) ≥ 1 is the constant in PI ( p ) ( β p ) \operatorname{PI}^{(p)}(\beta_p) PI ( p ) ( β p ) .
Lemma 3.4 ([40 Corollaries 1.14 and 1.15, and Equation (1.18)] ). Assume VD. Let ( E ( p ) , F ( p ) ) (\mathcal{E}^{(p)},\mathcal{F}^{(p)}) ( E ( p ) , F ( p ) ) be a p p p -energy satisfying P I ( p ) ( β p ) PI^{(p)}(\beta_p) P I ( p ) ( β p ) and c a p ( p ) ( β p ) ≤ ∞ cap^{(p)}(\beta_p)\le\infty c a p ( p ) ( β p ) ≤ ∞ . Then
F ( p ) = { u ∈ L p ( X ; m ) : lim inf r ↓ 0 E β p , X ( p ) ( u , r ) < + ∞ } \mathcal{F}^{(p)}=\left\{u\in L^p(X;m):\liminf_{r\downarrow0}\mathcal{E}^{(p)}_{\beta_p,X}(u,r)<+\infty\right\} F ( p ) = { u ∈ L p ( X ; m ) : r ↓ 0 lim inf E β p , X ( p ) ( u , r ) < + ∞ } = { u ∈ L p ( X ; m ) : lim sup r ↓ 0 E β p , X ( p ) ( u , r ) < + ∞ } = B p , β p p ( X ) . =\left\{u\in L^p(X;m):\limsup_{r\downarrow0}\mathcal{E}^{(p)}_{\beta_p,X}(u,r)<+\infty\right\}=B^{p,\frac{\beta_p}{p}}(X). = { u ∈ L p ( X ; m ) : r ↓ 0 lim sup E β p , X ( p ) ( u , r ) < + ∞ } = B p , p β p ( X ) . Moreover, there exists a constant C ( p ) ≥ 1 C^{(p)}\ge1 C ( p ) ≥ 1 such that for any u ∈ F ( p ) u\in\mathcal{F}^{(p)} u ∈ F ( p ) ,
1 C ( p ) E ( p ) ( u ) ≤ lim inf r ↓ 0 E β p , X ( p ) ( u , r ) ≤ sup r > 0 E β p , X ( p ) ( u , r ) ≤ C ( p ) E ( p ) ( u ) . \frac{1}{C^{(p)}}\mathcal{E}^{(p)}(u)\le\liminf_{r\downarrow0}\mathcal{E}^{(p)}_{\beta_p,X}(u,r)\le\sup_{r>0}\mathcal{E}^{(p)}_{\beta_p,X}(u,r)\le C^{(p)}\mathcal{E}^{(p)}(u). C ( p ) 1 E ( p ) ( u ) ≤ r ↓ 0 lim inf E β p , X ( p ) ( u , r ) ≤ r > 0 sup E β p , X ( p ) ( u , r ) ≤ C ( p ) E ( p ) ( u ) .
Lemma 3.5 ([40 Theorem 3.12 and Equation (1.18)] ). Assume VD. Let ( E ( p ) , F ( p ) ) (\mathcal{E}^{(p)},\mathcal{F}^{(p)}) ( E ( p ) , F ( p ) ) be a p p p -energy satisfying P I ( p ) ( β p ) PI^{(p)}(\beta_p) P I ( p ) ( β p ) and c a p ( p ) ( β p ) ≤ ∞ cap^{(p)}(\beta_p)\le\infty c a p ( p ) ( β p ) ≤ ∞ . Then there exists a constant C ( p ) > 0 C^{(p)}>0 C ( p ) > 0 such that for any open subset U ⊆ X U\subseteq X U ⊆ X and any u ∈ F ( p ) u\in\mathcal{F}^{(p)} u ∈ F ( p ) ,
Γ ( p ) ( u ) ( U ) ≤ C ( p ) lim inf r ↓ 0 E β p , U ( p ) ( u , r ) . \Gamma^{(p)}(u)(U)\le C^{(p)}\liminf_{r\downarrow0}\mathcal{E}^{(p)}_{\beta_p,U}(u,r). Γ ( p ) ( u ) ( U ) ≤ C ( p ) r ↓ 0 lim inf E β p , U ( p ) ( u , r ) .
Let A 1 , A 2 ∈ B ( X ) A_1,A_2\in\mathcal{B}(X) A 1 , A 2 ∈ B ( X ) . We define the E ( p ) \mathcal{E}^{(p)} E ( p ) -effective resistance between A 1 A_1 A 1 and A 2 A_2 A 2 as
R ( p ) ( A 1 , A 2 ) : = 1 cap ( p ) ( A 1 , A 2 ) . R^{(p)}(A_1,A_2):=\frac{1}{\operatorname{cap}^{(p)}(A_1,A_2)}. R ( p ) ( A 1 , A 2 ) := cap ( p ) ( A 1 , A 2 ) 1 .
where we use the convention that 0 − 1 = + ∞ 0^{-1}=+\infty 0 − 1 = + ∞ and ( + ∞ ) − 1 = 0 (+\infty)^{-1}=0 ( + ∞ ) − 1 = 0 . If A 1 ⊆ A 2 A_1\subseteq A_2 A 1 ⊆ A 2 and B 1 ⊆ B 2 B_1\subseteq B_2 B 1 ⊆ B 2 , then
R ( p ) ( A 1 , B 1 ) ≥ R ( p ) ( A 2 , B 2 ) . R^{(p)}(A_1,B_1)\ge R^{(p)}(A_2,B_2). R ( p ) ( A 1 , B 1 ) ≥ R ( p ) ( A 2 , B 2 ) .
For any x , y ∈ X x,y\in X x , y ∈ X with x ≠ y x\ne y x = y , let R ( p ) ( x , x ) = R ( p ) ( { x } , { x } ) = 0 R^{(p)}(x,x)=R^{(p)}(\{x\},\{x\})=0 R ( p ) ( x , x ) = R ( p ) ({ x } , { x }) = 0 and
R ( p ) ( x , y ) = R ( p ) ( { x } , { y } ) : = lim r ↓ 0 R ( p ) ( B ( x , r ) , B ( y , r ) ) . R^{(p)}(x,y)=R^{(p)}(\{x\},\{y\}):=\lim_{r\downarrow0}R^{(p)}(B(x,r),B(y,r)). R ( p ) ( x , y ) = R ( p ) ({ x } , { y }) := r ↓ 0 lim R ( p ) ( B ( x , r ) , B ( y , r )) .
Let β p > 0 \beta_p>0 β p > 0 . We say that the resistance estimate R ( p ) ( β p ) R^{(p)}(\beta_p) R ( p ) ( β p ) holds if there exists a constant C R ≥ 1 C_R\ge1 C R ≥ 1 such that for any x , y ∈ X x,y\in X x , y ∈ X with x ≠ y x\ne y x = y ,
1 C R d ( x , y ) β p m ( B ( x , d ( x , y ) ) ) ≤ R ( p ) ( x , y ) ≤ C R d ( x , y ) β p m ( B ( x , d ( x , y ) ) ) . \frac{1}{C_R}\frac{d(x,y)^{\beta_p}}{m(B(x,d(x,y)))}\le R^{(p)}(x,y)\le C_R\frac{d(x,y)^{\beta_p}}{m(B(x,d(x,y)))}. C R 1 m ( B ( x , d ( x , y ))) d ( x , y ) β p ≤ R ( p ) ( x , y ) ≤ C R m ( B ( x , d ( x , y ))) d ( x , y ) β p .
R ( p ) ( β p ) R^{(p)}(\beta_p) R ( p ) ( β p )
Lemma 3.6. Assume V ( d h ) V(d_h) V ( d h ) and β p > d h \beta_p>d_h β p > d h . Let ( E ( p ) , F ( p ) ) (\mathcal{E}^{(p)},\mathcal{F}^{(p)}) ( E ( p ) , F ( p ) ) be a p p p -energy. Then
P I ( p ) ( β p ) + c a p ( p ) ( β p ) ≤ ∞ ⟺ R ( p ) ( β p ) . (11) PI^{(p)}(\beta_p)+cap^{(p)}(\beta_p)\le\infty\Longleftrightarrow R^{(p)}(\beta_p). \tag*{(11)} P I ( p ) ( β p ) + c a p ( p ) ( β p ) ≤ ∞ ⟺ R ( p ) ( β p ) . (11) Moreover, whenever these equivalent conditions hold, the following assertions hold:
(1) F ( p ) ⊆ C ( X ) \mathcal{F}^{(p)}\subseteq C(X) F ( p ) ⊆ C ( X ) . For any x , y ∈ X x,y\in X x , y ∈ X with x ≠ y x\ne y x = y ,
R ( p ) ( x , y ) = sup { ∣ u ( x ) − u ( y ) ∣ p E ( p ) ( u ) : u ∈ F ( p ) ∩ C c ( X ) , E ( p ) ( u ) > 0 } , R^{(p)}(x,y)=\sup\left\{\frac{|u(x)-u(y)|^p}{\mathcal{E}^{(p)}(u)}:u\in\mathcal{F}^{(p)}\cap C_c(X),\ \mathcal{E}^{(p)}(u)>0\right\}, R ( p ) ( x , y ) = sup { E ( p ) ( u ) ∣ u ( x ) − u ( y ) ∣ p : u ∈ F ( p ) ∩ C c ( X ) , E ( p ) ( u ) > 0 } , which implies
∣ u ( x ) − u ( y ) ∣ p ≤ R ( p ) ( x , y ) E ( p ) ( u ) for any u ∈ F ( p ) , x , y ∈ X . |u(x)-u(y)|^p\le R^{(p)}(x,y)\mathcal{E}^{(p)}(u)\quad\text{for any }u\in\mathcal{F}^{(p)},\ x,y\in X. ∣ u ( x ) − u ( y ) ∣ p ≤ R ( p ) ( x , y ) E ( p ) ( u ) for any u ∈ F ( p ) , x , y ∈ X . Moreover,
ρ ( p ) : = ( R ( p ) ) 1 p − 1 : X × X → [ 0 , + ∞ ) , ( x , y ) ↦ ( R ( p ) ( x , y ) ) 1 p − 1 (12) \rho^{(p)}:=(R^{(p)})^{\frac{1}{p-1}}:X\times X\to[0,+\infty),\qquad(x,y)\mapsto(R^{(p)}(x,y))^{\frac{1}{p-1}} \tag*{(12)} ρ ( p ) := ( R ( p ) ) p − 1 1 : X × X → [ 0 , + ∞ ) , ( x , y ) ↦ ( R ( p ) ( x , y ) ) p − 1 1 (12) is a metric on X X X , and there exists a constant L ≥ 1 L\ge1 L ≥ 1 such that
1 L d ( x , y ) β p − d h p − 1 ≤ ρ ( p ) ( x , y ) ≤ L d ( x , y ) β p − d h p − 1 for any x , y ∈ X . (13) \frac{1}{L}d(x,y)^{\frac{\beta_p-d_h}{p-1}}\le\rho^{(p)}(x,y)\le Ld(x,y)^{\frac{\beta_p-d_h}{p-1}}\quad\text{for any }x,y\in X. \tag*{(13)} L 1 d ( x , y ) p − 1 β p − d h ≤ ρ ( p ) ( x , y ) ≤ L d ( x , y ) p − 1 β p − d h for any x , y ∈ X . (13) We write B ρ ( p ) ( x , r ) B_{\rho^{(p)}}(x,r) B ρ ( p ) ( x , r ) for the ball centered at x x x with radius r r r with respect to ρ ( p ) \rho^{(p)} ρ ( p ) , and denote by H ρ ( p ) 1 H^1_{\rho^{(p)}} H ρ ( p ) 1 the 1 1 1 -dimensional Hausdorff measure on ( X , ρ ( p ) ) (X,\rho^{(p)}) ( X , ρ ( p ) ) .
(2) There exist constants C 1 , C 2 > 0 C_1,C_2>0 C 1 , C 2 > 0 such that the following holds. For any compact set K K K and any bounded open set U U U satisfying K ⊆ U K\subseteq U K ⊆ U , let u ∈ F ( p ) ⊆ C ( X ) u\in\mathcal{F}^{(p)}\subseteq C(X) u ∈ F ( p ) ⊆ C ( X ) be the E ( p ) \mathcal{E}^{(p)} E ( p ) -capacitary potential for ( K , X ∖ U ) (K,X\setminus U) ( K , X ∖ U ) ; that is, 0 ≤ u ≤ 1 0\leq u\leq1 0 ≤ u ≤ 1 on X X X , u = 1 u=1 u = 1 on K K K , u = 0 u=0 u = 0 on X ∖ U X\setminus U X ∖ U , u u u is harmonic in U ∖ K U\setminus K U ∖ K , and E ( p ) ( u ) = cap ( p ) ( K , X ∖ U ) \mathcal{E}^{(p)}(u)=\operatorname{cap}^{(p)}(K,X\setminus U) E ( p ) ( u ) = cap ( p ) ( K , X ∖ U ) . Then
∣ u ( x ) − u ( y ) ∣ ≤ C 1 E ( p ) ( u ) 1 p − 1 d ( x , y ) β p − d h p − 1 (14) |u(x)-u(y)|\leq C_1\mathcal{E}^{(p)}(u)^{\frac{1}{p-1}}d(x,y)^{\frac{\beta_p-d_h}{p-1}} \tag*{(14)} ∣ u ( x ) − u ( y ) ∣ ≤ C 1 E ( p ) ( u ) p − 1 1 d ( x , y ) p − 1 β p − d h (14)
≤ C 2 E ( p ) ( u ) 1 p − 1 ρ ( p ) ( x , y ) (15) \leq C_2\mathcal{E}^{(p)}(u)^{\frac{1}{p-1}}\rho^{(p)}(x,y) \tag*{(15)} ≤ C 2 E ( p ) ( u ) p − 1 1 ρ ( p ) ( x , y ) (15)
Moreover, the pushforward u ∗ Γ ( p ) ( u ) u_*\Gamma^{(p)}(u) u ∗ Γ ( p ) ( u ) of Γ ( p ) ( u ) \Gamma^{(p)}(u) Γ ( p ) ( u ) under u u u is a constant multiple of the one-dimensional Lebesgue measure L 1 \mathcal{L}^1 L 1 on [ 0 , 1 ] [0,1] [ 0 , 1 ] . More precisely,
u ∗ Γ ( p ) ( u ) = E ( p ) ( u ) L 1 ∣ [ 0 , 1 ] . (16) u_*\Gamma^{(p)}(u)=\mathcal{E}^{(p)}(u)\mathcal{L}^1\big|_{[0,1]}. \tag*{(16)} u ∗ Γ ( p ) ( u ) = E ( p ) ( u ) L 1 [ 0 , 1 ] . (16)
Consequently, for any x ∈ X x\in X x ∈ X and any r > 0 r>0 r > 0 ,
Γ ( p ) ( u ) ( B ρ ( p ) ( x , r ) ) ≤ 2 C 2 E ( p ) ( u ) p p − 1 r , (17) \Gamma^{(p)}(u)(B_{\rho^{(p)}}(x,r))\leq2C_2\mathcal{E}^{(p)}(u)^{\frac{p}{p-1}}r, \tag*{(17)} Γ ( p ) ( u ) ( B ρ ( p ) ( x , r )) ≤ 2 C 2 E ( p ) ( u ) p − 1 p r , (17)
and
Γ ( p ) ( u ) ≤ 2 C 2 E ( p ) ( u ) p p − 1 H ρ ( p ) 1 . (18) \Gamma^{(p)}(u)\leq2C_2\mathcal{E}^{(p)}(u)^{\frac{p}{p-1}}\mathcal{H}^1_{\rho^{(p)}}. \tag*{(18)} Γ ( p ) ( u ) ≤ 2 C 2 E ( p ) ( u ) p − 1 p H ρ ( p ) 1 . (18)
Proof. The equivalence (3.1) follows from [42 Theorem 2.3] .
(1) The function ρ ( p ) = ( R ( p ) ) 1 p − 1 \rho^{(p)}=(R^{(p)})^{\frac{1}{p-1}} ρ ( p ) = ( R ( p ) ) p − 1 1 is a metric by the argument of [27 Corollary 6.32] . Indeed, that argument remains valid under (RF1)p _p p –(RF4)p _p p , the Markovian property, p p p -Clarkson’s inequality, and strong sub-additivity; the full generalized p p p -contraction property is not required. These properties also hold for ( E ( p ) , F e ( p ) ) (\mathcal{E}^{(p)},\mathcal{F}^{(p)}_e) ( E ( p ) , F e ( p ) ) , where F e ( p ) \mathcal{F}^{(p)}_e F e ( p ) will be given in (6.4). The remaining results follow from [42 Lemma 5.2] , R ( p ) ( β p ) R^{(p)}(\beta_p) R ( p ) ( β p ) , and the same arguments as those used in [22 Lemma 8.6 and Corollary 8.9] .
(2) The existence of such an E ( p ) \mathcal{E}^{(p)} E ( p ) -capacitary potential u u u follows from [41 Lemma 3.8] . We claim that for any a , b ∈ [ 0 , 1 ] a,b\in[0,1] a , b ∈ [ 0 , 1 ] with a < b a<b a < b ,
Γ ( p ) ( u ) ( { x ∈ X : a < u ( x ) < b } ) = ( b − a ) E ( p ) ( u ) . (19) \Gamma^{(p)}(u)(\{x\in X:a<u(x)<b\})=(b-a)\mathcal{E}^{(p)}(u). \tag*{(19)} Γ ( p ) ( u ) ({ x ∈ X : a < u ( x ) < b }) = ( b − a ) E ( p ) ( u ) . (19)
Indeed, since ( u − a ) + ∧ ( b − a ) − ( b − a ) u ∈ F ( p ) ( U ∖ K ) (u-a)_+\wedge(b-a)-(b-a)u\in\mathcal{F}^{(p)}(U\setminus K) ( u − a ) + ∧ ( b − a ) − ( b − a ) u ∈ F ( p ) ( U ∖ K ) and u u u is harmonic in U ∖ K U\setminus K U ∖ K , by the strongly local property,
0 = E ( p ) ( u ; ( u − a ) + ∧ ( b − a ) ) − ( b − a ) E ( p ) ( u ) = Γ ( p ) ( u ) ( { x ∈ X : a < u ( x ) < b } ) − ( b − a ) E ( p ) ( u ) . \begin{aligned}
0=\mathcal{E}^{(p)}(u;(u-a)_+\wedge(b-a))-(b-a)\mathcal{E}^{(p)}(u) \\
&=\Gamma^{(p)}(u)(\{x\in X:a<u(x)<b\})-(b-a)\mathcal{E}^{(p)}(u).
\end{aligned} 0 = E ( p ) ( u ; ( u − a ) + ∧ ( b − a )) − ( b − a ) E ( p ) ( u ) = Γ ( p ) ( u ) ({ x ∈ X : a < u ( x ) < b }) − ( b − a ) E ( p ) ( u ) .
For any x , y ∈ X x,y\in X x , y ∈ X with x ≠ y x\neq y x = y , without loss of generality, we may assume that u ( x ) > u ( y ) u(x)>u(y) u ( x ) > u ( y ) , let v = ( u − u ( y ) ) + ∧ ( u ( x ) − u ( y ) ) ∈ F ( p ) ⊆ C ( X ) v=(u-u(y))_+\wedge(u(x)-u(y))\in\mathcal{F}^{(p)}\subseteq C(X) v = ( u − u ( y ) ) + ∧ ( u ( x ) − u ( y )) ∈ F ( p ) ⊆ C ( X ) . By (3.9),
R ( p ) ( x , y ) ≥ ∣ v ( x ) − v ( y ) ∣ p E ( p ) ( v ) = ∣ u ( x ) − u ( y ) ∣ p Γ ( p ) ( u ) ( { z ∈ X : u ( y ) < u ( z ) < u ( x ) } ) = ∣ u ( x ) − u ( y ) ∣ p − 1 E ( p ) ( u ) . R^{(p)}(x,y)\geq\frac{|v(x)-v(y)|^p}{\mathcal{E}^{(p)}(v)}
=\frac{|u(x)-u(y)|^p}{\Gamma^{(p)}(u)(\{z\in X:u(y)<u(z)<u(x)\})}
=\frac{|u(x)-u(y)|^{p-1}}{\mathcal{E}^{(p)}(u)}. R ( p ) ( x , y ) ≥ E ( p ) ( v ) ∣ v ( x ) − v ( y ) ∣ p = Γ ( p ) ( u ) ({ z ∈ X : u ( y ) < u ( z ) < u ( x )}) ∣ u ( x ) − u ( y ) ∣ p = E ( p ) ( u ) ∣ u ( x ) − u ( y ) ∣ p − 1 .
Combining this with R ( p ) ( β p ) R^{(p)}(\beta_p) R ( p ) ( β p ) ,
∣ u ( x ) − u ( y ) ∣ p − 1 ≲ E ( p ) ( u ) d ( x , y ) β p − d h , |u(x)-u(y)|^{p-1}\lesssim\mathcal{E}^{(p)}(u)d(x,y)^{\beta_p-d_h}, ∣ u ( x ) − u ( y ) ∣ p − 1 ≲ E ( p ) ( u ) d ( x , y ) β p − d h ,
which gives (3.4). Together with (3.3), we obtain (3.5).
To prove (3.6), take any ϕ ∈ C 1 ( [ 0 , 1 ] ) \phi\in C^1([0,1]) ϕ ∈ C 1 ([ 0 , 1 ]) satisfying ϕ ( 0 ) = ϕ ( 1 ) = 0 \phi(0)=\phi(1)=0 ϕ ( 0 ) = ϕ ( 1 ) = 0 . For any ϵ ∈ R \epsilon\in\mathbb{R} ϵ ∈ R , let ψ ( t ) = t + ϵ ϕ ( t ) \psi(t)=t+\epsilon\phi(t) ψ ( t ) = t + ϵ ϕ ( t ) . Then ψ ∘ u ∈ F ( p ) \psi\circ u\in\mathcal{F}^{(p)} ψ ∘ u ∈ F ( p ) , with ψ ∘ u = 1 \psi\circ u=1 ψ ∘ u = 1 on K K K and ψ ∘ u = 0 \psi\circ u=0 ψ ∘ u = 0 on X ∖ U X\setminus U X ∖ U . By the chain rule,
∫ [ 0 , 1 ] d ( u ∗ Γ ( p ) ( u ) ) = E ( p ) ( u ) = cap ( p ) ( K , X ∖ U ) ≤ E ( p ) ( ψ ∘ u ) = ∫ X ∣ ψ ′ ( u ) ∣ p d Γ ( p ) ( u ) = ∫ [ 0 , 1 ] ∣ 1 + ϵ ϕ ′ ( t ) ∣ p d ( u ∗ Γ ( p ) ( u ) ) . \begin{aligned}
\int_{[0,1]} \mathrm{d}(u_*\Gamma^{(p)}(u))=\mathcal{E}^{(p)}(u)=\operatorname{cap}^{(p)}(K,X\setminus U)\leq\mathcal{E}^{(p)}(\psi\circ u) \\
&=\int_X|\psi'(u)|^p\mathrm{d}\Gamma^{(p)}(u)=\int_{[0,1]}|1+\epsilon\phi'(t)|^p\mathrm{d}(u_*\Gamma^{(p)}(u)).
\end{aligned} ∫ [ 0 , 1 ] d ( u ∗ Γ ( p ) ( u )) = E ( p ) ( u ) = cap ( p ) ( K , X ∖ U ) ≤ E ( p ) ( ψ ∘ u ) = ∫ X ∣ ψ ′ ( u ) ∣ p d Γ ( p ) ( u ) = ∫ [ 0 , 1 ] ∣1 + ϵ ϕ ′ ( t ) ∣ p d ( u ∗ Γ ( p ) ( u )) .
Since ϵ ∈ R \epsilon\in\mathbb{R} ϵ ∈ R is arbitrary and ϕ ′ \phi' ϕ ′ is bounded on [ 0 , 1 ] [0,1] [ 0 , 1 ] , differentiating with respect to ϵ \epsilon ϵ and then letting ϵ = 0 \epsilon=0 ϵ = 0 , we have
∫ [ 0 , 1 ] ϕ ′ d ( u ∗ Γ ( p ) ( u ) ) = 0. \int_{[0,1]}\phi'\mathrm{d}(u_*\Gamma^{(p)}(u))=0. ∫ [ 0 , 1 ] ϕ ′ d ( u ∗ Γ ( p ) ( u )) = 0.
By Lemma 3.7 below,
u ∗ Γ ( p ) ( u ) = ( ( u ∗ Γ ( p ) ( u ) ) ( [ 0 , 1 ] ) ) L 1 ∣ [ 0 , 1 ] = E ( p ) ( u ) L 1 ∣ [ 0 , 1 ] . u_*\Gamma^{(p)}(u)=\left(\left(u_*\Gamma^{(p)}(u)\right)([0,1])\right)\mathcal{L}^1\big|_{[0,1]}=\mathcal{E}^{(p)}(u)\mathcal{L}^1\big|_{[0,1]}. u ∗ Γ ( p ) ( u ) = ( ( u ∗ Γ ( p ) ( u ) ) ([ 0 , 1 ]) ) L 1 [ 0 , 1 ] = E ( p ) ( u ) L 1 [ 0 , 1 ] .
For any ball B ρ ( p ) ( x , r ) B_{\rho^{(p)}(x,r)} B ρ ( p ) ( x , r ) , by (15) and (16) ,
Γ ( p ) ( u ) ( B ρ ( p ) ( x , r ) ) ≤ Γ ( p ) ( u ) ( { y ∈ X : ∣ u ( x ) − u ( y ) ∣ < C 2 E ( p ) ( u ) 1 p − 1 r } ) = E ( p ) ( u ) L 1 ( { t ∈ [ 0 , 1 ] : ∣ t − u ( x ) ∣ < C 2 E ( p ) ( u ) 1 p − 1 r } ) ≤ 2 C 2 E ( p ) ( u ) p p − 1 r , \begin{aligned}
\Gamma^{(p)}(u)\left(B_{\rho^{(p)}(x,r)}\right)\leq\Gamma^{(p)}(u)\left(\left\{y\in X:|u(x)-u(y)|<C_2\mathcal{E}^{(p)}(u)^{\frac{1}{p-1}}r\right\}\right) \\
&=\mathcal{E}^{(p)}(u)\mathcal{L}^1\left(\left\{t\in[0,1]:|t-u(x)|<C_2\mathcal{E}^{(p)}(u)^{\frac{1}{p-1}}r\right\}\right)\leq2C_2\mathcal{E}^{(p)}(u)^{\frac{p}{p-1}}r,
\end{aligned} Γ ( p ) ( u ) ( B ρ ( p ) ( x , r ) ) ≤ Γ ( p ) ( u ) ( { y ∈ X : ∣ u ( x ) − u ( y ) ∣ < C 2 E ( p ) ( u ) p − 1 1 r } ) = E ( p ) ( u ) L 1 ( { t ∈ [ 0 , 1 ] : ∣ t − u ( x ) ∣ < C 2 E ( p ) ( u ) p − 1 1 r } ) ≤ 2 C 2 E ( p ) ( u ) p − 1 p r ,
which is (17) . By Frostman’s lemma or the mass distribution principle (see [33 Theorem 8.17] , [19 p. 67] , or [4 Theorem 2.4] ), we obtain (18) .
Lemma 3.7. Let λ \lambda λ be a non-negative finite Borel measure on [ 0 , 1 ] [0,1] [ 0 , 1 ] . Assume that for any ϕ ∈ C 1 ( [ 0 , 1 ] ) \phi\in C^1([0,1]) ϕ ∈ C 1 ([ 0 , 1 ]) with ϕ ( 0 ) = ϕ ( 1 ) = 0 \phi(0)=\phi(1)=0 ϕ ( 0 ) = ϕ ( 1 ) = 0 , we have ∫ [ 0 , 1 ] ϕ ′ d λ = 0 \int_{[0,1]}\phi'\,d\lambda=0 ∫ [ 0 , 1 ] ϕ ′ d λ = 0 . Then λ = λ ( [ 0 , 1 ] ) L 1 ∣ [ 0 , 1 ] \lambda=\lambda([0,1])\mathcal{L}^1\big|_{[0,1]} λ = λ ([ 0 , 1 ]) L 1 [ 0 , 1 ] .
Proof. We only need to show that for any f ∈ C ( [ 0 , 1 ] ) f\in C([0,1]) f ∈ C ([ 0 , 1 ]) ,
∫ [ 0 , 1 ] f d λ = λ ( [ 0 , 1 ] ) ∫ [ 0 , 1 ] f d L 1 . \int_{[0,1]}f\,d\lambda=\lambda([0,1])\int_{[0,1]}f\,d\mathcal{L}^1. ∫ [ 0 , 1 ] f d λ = λ ([ 0 , 1 ]) ∫ [ 0 , 1 ] f d L 1 .
Indeed, let g ( t ) = f ( t ) − ∫ 0 1 f ( s ) d s g(t)=f(t)-\int_0^1f(s)\,ds g ( t ) = f ( t ) − ∫ 0 1 f ( s ) d s , then g ∈ C ( [ 0 , 1 ] ) g\in C([0,1]) g ∈ C ([ 0 , 1 ]) and ∫ 0 1 g ( t ) d t = 0 \int_0^1g(t)\,dt=0 ∫ 0 1 g ( t ) d t = 0 . Let ϕ ( t ) = ∫ 0 t g ( s ) d s \phi(t)=\int_0^t g(s)\,ds ϕ ( t ) = ∫ 0 t g ( s ) d s , then ϕ ∈ C 1 ( [ 0 , 1 ] ) \phi\in C^1([0,1]) ϕ ∈ C 1 ([ 0 , 1 ]) , ϕ ( 0 ) = ϕ ( 1 ) = 0 \phi(0)=\phi(1)=0 ϕ ( 0 ) = ϕ ( 1 ) = 0 , and ϕ ′ = g \phi'=g ϕ ′ = g . By assumption,
0 = ∫ [ 0 , 1 ] ϕ ′ d λ = ∫ [ 0 , 1 ] g d λ = ∫ [ 0 , 1 ] f d λ − λ ( [ 0 , 1 ] ) ∫ [ 0 , 1 ] f d L 1 . 0=\int_{[0,1]}\phi'\,d\lambda=\int_{[0,1]}g\,d\lambda=\int_{[0,1]}f\,d\lambda-\lambda([0,1])\int_{[0,1]}f\,d\mathcal{L}^1. 0 = ∫ [ 0 , 1 ] ϕ ′ d λ = ∫ [ 0 , 1 ] g d λ = ∫ [ 0 , 1 ] f d λ − λ ([ 0 , 1 ]) ∫ [ 0 , 1 ] f d L 1 .
Proof of Theorem 2.1
Suppose that there exist p , q ∈ I p,q\in I p , q ∈ I with p < q p<q p < q such that β p p = β q q > 1 \frac{\beta_p}{p}=\frac{\beta_q}{q}>1 p β p = q β q > 1 . Let U ⊆ X U\subseteq X U ⊆ X be open and let u u u be a measurable function. For any r > 0 r>0 r > 0 , by Hölder’s inequality,
E β p , U ( p ) ( u , r ) ≤ 1 r β p ∫ U ( 1 V ( x , r ) ∫ B ( x , r ) ∣ u ( x ) − u ( y ) ∣ q m ( d y ) ) p q m ( d x ) ≤ 1 r β p [ ∫ U ( 1 V ( x , r ) ∫ B ( x , r ) ∣ u ( x ) − u ( y ) ∣ q m ( d y ) ) m ( d x ) ] p q m ( U ) 1 − p q = E β q , U ( q ) ( u , r ) p q m ( U ) 1 − p q . (20) \begin{aligned}
E_{\beta_p,U}^{(p)}(u,r)\leq\frac{1}{r^{\beta_p}}\int_U\left(\frac{1}{V(x,r)}\int_{B(x,r)}|u(x)-u(y)|^q\,m(dy)\right)^{\frac{p}{q}}m(dx) \\
&\leq\frac{1}{r^{\beta_p}}\left[\int_U\left(\frac{1}{V(x,r)}\int_{B(x,r)}|u(x)-u(y)|^q\,m(dy)\right)m(dx)\right]^{\frac{p}{q}}m(U)^{1-\frac{p}{q}} \\
&=E_{\beta_q,U}^{(q)}(u,r)^{\frac{p}{q}}m(U)^{1-\frac{p}{q}}. \tag*{(20)}
\end{aligned} E β p , U ( p ) ( u , r ) ≤ r β p 1 ∫ U ( V ( x , r ) 1 ∫ B ( x , r ) ∣ u ( x ) − u ( y ) ∣ q m ( d y ) ) q p m ( d x ) ≤ r β p 1 [ ∫ U ( V ( x , r ) 1 ∫ B ( x , r ) ∣ u ( x ) − u ( y ) ∣ q m ( d y ) ) m ( d x ) ] q p m ( U ) 1 − q p = E β q , U ( q ) ( u , r ) q p m ( U ) 1 − q p . (20)
where the last equality follows from the fact that β p p = β q q \frac{\beta_p}{p}=\frac{\beta_q}{q} p β p = q β q . By VD and P I ( q ) ( β q ) \mathrm{PI}^{(q)}(\beta_q) PI ( q ) ( β q ) , we may apply Lemma 3.3 to obtain, for any u ∈ F ( q ) u\in\mathcal{F}^{(q)} u ∈ F ( q ) ,
E β q , U ( q ) ( u , r ) ≤ C ( q ) Γ ( q ) ( u ) ( U ( 2 A r ) ) , (21) E_{\beta_q,U}^{(q)}(u,r)\leq C^{(q)}\Gamma^{(q)}(u)(U(2Ar)), \tag*{(21)} E β q , U ( q ) ( u , r ) ≤ C ( q ) Γ ( q ) ( u ) ( U ( 2 A r )) , (21)
where A = A P I ( q ) A=A_{\mathrm{PI}}^{(q)} A = A PI ( q ) .
Since ( E ( q ) , F ( q ) ) (\mathcal{E}^{(q)},\mathcal{F}^{(q)}) ( E ( q ) , F ( q ) ) is regular, we may choose a non-zero function u ∈ F ( q ) ∩ C c ( X ) u\in\mathcal{F}^{(q)}\cap C_c(X) u ∈ F ( q ) ∩ C c ( X ) . Take B = B ( x 0 , R ) ⊇ supp ( u ) B=B(x_0,R)\supseteq\operatorname{supp}(u) B = B ( x 0 , R ) ⊇ supp ( u ) . For any r ∈ ( 0 , R ) r\in(0,R) r ∈ ( 0 , R ) , by (20) ,
E β p , X ( p ) ( u , r ) = E β p , 2 B ( p ) ( u , r ) ≤ E β q , 2 B ( q ) ( u , r ) p q m ( 2 B ) 1 − p q = E β q , X ( q ) ( u , r ) p q m ( 2 B ) 1 − p q . E_{\beta_p,X}^{(p)}(u,r)=E_{\beta_p,2B}^{(p)}(u,r)\leq E_{\beta_q,2B}^{(q)}(u,r)^{\frac{p}{q}}m(2B)^{1-\frac{p}{q}}=E_{\beta_q,X}^{(q)}(u,r)^{\frac{p}{q}}m(2B)^{1-\frac{p}{q}}. E β p , X ( p ) ( u , r ) = E β p , 2 B ( p ) ( u , r ) ≤ E β q , 2 B ( q ) ( u , r ) q p m ( 2 B ) 1 − q p = E β q , X ( q ) ( u , r ) q p m ( 2 B ) 1 − q p .
Since u ∈ F ( q ) u\in\mathcal{F}^{(q)} u ∈ F ( q ) , by VD, P I ( q ) ( β q ) \mathrm{PI}^{(q)}(\beta_q) PI ( q ) ( β q ) , and cap ( q ) ( β q ) ≤ \operatorname{cap}^{(q)}(\beta_q)_{\leq} cap ( q ) ( β q ) ≤ , Lemma 3.4 gives lim inf r ↓ 0 E β q , X ( q ) ( u , r ) < + ∞ \liminf_{r\downarrow0}E_{\beta_q,X}^{(q)}(u,r)<+\infty lim inf r ↓ 0 E β q , X ( q ) ( u , r ) < + ∞ , which implies lim inf r ↓ 0 E β p , X ( p ) ( u , r ) < + ∞ \liminf_{r\downarrow0}E_{\beta_p,X}^{(p)}(u,r)<+\infty lim inf r ↓ 0 E β p , X ( p ) ( u , r ) < + ∞ . Since u ∈ C c ( X ) ⊆ L p ( X ; m ) u\in C_c(X)\subseteq L^p(X;m) u ∈ C c ( X ) ⊆ L p ( X ; m ) , by VD, P I ( p ) ( β p ) \mathrm{PI}^{(p)}(\beta_p) PI ( p ) ( β p ) , and cap ( p ) ( β p ) ≤ \operatorname{cap}^{(p)}(\beta_p)_{\leq} cap ( p ) ( β p ) ≤ , Lemma 3.4 gives u ∈ F ( p ) u\in\mathcal{F}^{(p)} u ∈ F ( p ) . Let
μ : = Γ ( q ) ( u ) , \mu:=\Gamma^{(q)}(u), μ := Γ ( q ) ( u ) ,
then μ \mu μ is a finite positive Radon measure. By the strongly local property, supp ( μ ) ⊆ supp ( u ) \operatorname{supp}(\mu) \subseteq\operatorname{supp}(u) supp ( μ ) ⊆ supp ( u ) . For any r ∈ ( 0 , 1 8 A R ) r \in\left(0,\frac{1}{8A}R\right) r ∈ ( 0 , 8 A 1 R ) , let W W W be an r r r -net. By VD, there exists an integer N ≥ 1 N \ge1 N ≥ 1 depending only on C V D C_{VD} C V D and A A A such that
∑ z ∈ W 1 B ( z , 4 A r ) ≤ N . (22) \sum_{z \in W} \mathbf{1}_{B(z,4Ar)} \le N. \tag*{(22)} z ∈ W ∑ 1 B ( z , 4 A r ) ≤ N . (22)
By (4.1) and (4.2), for any z ∈ W z \in W z ∈ W ,
E β p , B ( z , r ) ( p ) ( u , r ) ≤ C μ ( B ( z , 4 A r ) ) p q m ( B ( z , r ) ) 1 − p q , E_{\beta_p,B(z,r)}^{(p)}(u,r) \le C\mu(B(z,4Ar))^{\frac{p}{q}}m(B(z,r))^{1-\frac{p}{q}}, E β p , B ( z , r ) ( p ) ( u , r ) ≤ C μ ( B ( z , 4 A r ) ) q p m ( B ( z , r ) ) 1 − q p ,
where C = ( C ( q ) ) p q C=(C^{(q)})^{\frac{p}{q}} C = ( C ( q ) ) q p , hence
E β p , X ( p ) ( u , r ) ≤ ∑ z ∈ W ∩ ( 2 B ) E β p , B ( z , r ) ( p ) ( u , r ) ≤ C ∑ z ∈ W ∩ ( 2 B ) μ ( B ( z , 4 A r ) ) p q m ( B ( z , r ) ) 1 − p q . (23) E_{\beta_p,X}^{(p)}(u,r) \le\sum_{z\in W\cap(2B)} E_{\beta_p,B(z,r)}^{(p)}(u,r) \le C\sum_{z\in W\cap(2B)}\mu(B(z,4Ar))^{\frac{p}{q}}m(B(z,r))^{1-\frac{p}{q}}. \tag*{(23)} E β p , X ( p ) ( u , r ) ≤ z ∈ W ∩ ( 2 B ) ∑ E β p , B ( z , r ) ( p ) ( u , r ) ≤ C z ∈ W ∩ ( 2 B ) ∑ μ ( B ( z , 4 A r ) ) q p m ( B ( z , r ) ) 1 − q p . (23)
Since β q > q \beta_q>q β q > q , by [42 Theorem 2.1] , μ ⊥ m \mu\perp m μ ⊥ m . For any ε > 0 \varepsilon>0 ε > 0 , by the regularity of μ \mu μ and m m m , there exist a compact set K K K and an open set G G G with K ⊆ G ⊆ B K\subseteq G\subseteq B K ⊆ G ⊆ B such that
μ ( X ∖ K ) < ε , m ( G ) < ε . (24) \mu(X\setminus K)<\varepsilon,\qquad m(G)<\varepsilon. \tag*{(24)} μ ( X ∖ K ) < ε , m ( G ) < ε . (24)
Let r 0 = dist ( K , X ∖ G ) ∈ ( 0 , + ∞ ) r_0=\operatorname{dist}(K,X\setminus G)\in(0,+\infty) r 0 = dist ( K , X ∖ G ) ∈ ( 0 , + ∞ ) . For any r < 1 8 A r 0 r<\frac{1}{8A}r_0 r < 8 A 1 r 0 , if B ( z , 4 A r ) ∩ K ≠ ∅ B(z,4Ar)\cap K\ne\varnothing B ( z , 4 A r ) ∩ K = ∅ , then B ( z , 4 A r ) ⊆ G B(z,4Ar)\subseteq G B ( z , 4 A r ) ⊆ G . Write W ∩ ( 2 B ) = W 1 ⊔ W 2 W\cap(2B)=W_1\sqcup W_2 W ∩ ( 2 B ) = W 1 ⊔ W 2 , where
W 1 : = { z ∈ W ∩ ( 2 B ) : B ( z , 4 A r ) ∩ K ≠ ∅ } , W_1:=\{z\in W\cap(2B):B(z,4Ar)\cap K\ne\varnothing\}, W 1 := { z ∈ W ∩ ( 2 B ) : B ( z , 4 A r ) ∩ K = ∅ } ,
W 2 : = { z ∈ W ∩ ( 2 B ) : B ( z , 4 A r ) ∩ K = ∅ } , W_2:=\{z\in W\cap(2B):B(z,4Ar)\cap K=\varnothing\}, W 2 := { z ∈ W ∩ ( 2 B ) : B ( z , 4 A r ) ∩ K = ∅ } ,
then
⋃ z ∈ W 1 B ( z , 4 A r ) ⊆ G , ⋃ z ∈ W 2 B ( z , 4 A r ) ⊆ X ∖ K . \bigcup_{z\in W_1}B(z,4Ar)\subseteq G,\qquad\bigcup_{z\in W_2}B(z,4Ar)\subseteq X\setminus K. z ∈ W 1 ⋃ B ( z , 4 A r ) ⊆ G , z ∈ W 2 ⋃ B ( z , 4 A r ) ⊆ X ∖ K .
By (4.4), Hölder’s inequality, (4.3), and (4.5),
E β p , X ( p ) ( u , r ) ≤ C ∑ k = 1 2 ∑ z ∈ W k μ ( B ( z , 4 A r ) ) p q m ( B ( z , r ) ) 1 − p q ≤ C N ∑ k = 1 2 ( ∑ z ∈ W k μ ( B ( z , 4 A r ) ) ) p q ( ∑ z ∈ W k m ( B ( z , r ) ) ) 1 − p q ≤ C N ∑ k = 1 2 μ ( ⋃ z ∈ W k B ( z , 4 A r ) ) p q m ( ⋃ z ∈ W k B ( z , r ) ) 1 − p q ≤ C N ( μ ( X ) p q m ( G ) 1 − p q + μ ( X ∖ K ) p q m ( 4 B ) 1 − p q ) < C N ( μ ( X ) p q ε 1 − p q + ε p q m ( 4 B ) 1 − p q ) . \begin{aligned}
E_{\beta_p,X}^{(p)}(u,r)
&\le C\sum_{k=1}^{2}\sum_{z\in W_k}\mu(B(z,4Ar))^{\frac{p}{q}}m(B(z,r))^{1-\frac{p}{q}}\\
&\le CN\sum_{k=1}^{2}\left(\sum_{z\in W_k}\mu(B(z,4Ar))\right)^{\frac{p}{q}}\left(\sum_{z\in W_k}m(B(z,r))\right)^{1-\frac{p}{q}}\\
&\le CN\sum_{k=1}^{2}\mu\left(\bigcup_{z\in W_k}B(z,4Ar)\right)^{\frac{p}{q}}m\left(\bigcup_{z\in W_k}B(z,r)\right)^{1-\frac{p}{q}}\\
&\le CN\left(\mu(X)^{\frac{p}{q}}m(G)^{1-\frac{p}{q}}+\mu(X\setminus K)^{\frac{p}{q}}m(4B)^{1-\frac{p}{q}}\right)\\
&< CN\left(\mu(X)^{\frac{p}{q}}\varepsilon^{1-\frac{p}{q}}+\varepsilon^{\frac{p}{q}}m(4B)^{1-\frac{p}{q}}\right).
\end{aligned} E β p , X ( p ) ( u , r ) ≤ C k = 1 ∑ 2 z ∈ W k ∑ μ ( B ( z , 4 A r ) ) q p m ( B ( z , r ) ) 1 − q p ≤ C N k = 1 ∑ 2 ( z ∈ W k ∑ μ ( B ( z , 4 A r )) ) q p ( z ∈ W k ∑ m ( B ( z , r )) ) 1 − q p ≤ C N k = 1 ∑ 2 μ ( z ∈ W k ⋃ B ( z , 4 A r ) ) q p m ( z ∈ W k ⋃ B ( z , r ) ) 1 − q p ≤ C N ( μ ( X ) q p m ( G ) 1 − q p + μ ( X ∖ K ) q p m ( 4 B ) 1 − q p ) < C N ( μ ( X ) q p ε 1 − q p + ε q p m ( 4 B ) 1 − q p ) .
Hence lim r ↓ 0 E β p , X ( p ) ( u , r ) = 0 \lim_{r\downarrow0}E_{\beta_p,X}^{(p)}(u,r)=0 lim r ↓ 0 E β p , X ( p ) ( u , r ) = 0 . By VD, PI ( p ) ( β p ) \operatorname{PI}^{(p)}(\beta_p) PI ( p ) ( β p ) , and cap ( p ) ( β p ) ≤ \operatorname{cap}^{(p)}(\beta_p)\le cap ( p ) ( β p ) ≤ , Lemma 3.4 again gives E ( p ) ( u ) = 0 \mathcal{E}^{(p)}(u)=0 E ( p ) ( u ) = 0 . By PI ( p ) ( β p ) \operatorname{PI}^{(p)}(\beta_p) PI ( p ) ( β p ) , u u u must be constant. Since u ∈ C c ( X ) u\in C_c(X) u ∈ C c ( X ) and X X X is unbounded, it follows that u = 0 u=0 u = 0 , contradicting the assumption that u u u is non-zero.
Proof of Theorem 2.5(1)
Lemma 5.1. Assume VD. Let I ⊆ ( 1 , + ∞ ) I\subseteq(1,+\infty) I ⊆ ( 1 , + ∞ ) be an open interval. Assume that, for any p ∈ I p\in I p ∈ I , there exists a p p p -energy ( E ( p ) , F ( p ) ) (\mathcal{E}^{(p)},\mathcal{F}^{(p)}) ( E ( p ) , F ( p ) ) satisfying PI ( p ) ( β p ) \operatorname{PI}^{(p)}(\beta_p) PI ( p ) ( β p ) and cap ( p ) ( β p ) ≤ \operatorname{cap}^{(p)}(\beta_p)\le cap ( p ) ( β p ) ≤ . Then the function p ↦ β p p\mapsto\beta_p p ↦ β p is strictly increasing on I I I .
Proof. By Lemmas 3.1 and 3.2, the function p ↦ β p p\mapsto\beta_p p ↦ β p is continuous and increasing on I I I . It remains to show that the monotonicity is strict. Indeed, suppose that there exist p , q ∈ I p,q\in I p , q ∈ I with p < q p<q p < q such that β p = β q \beta_p=\beta_q β p = β q . By regularity, there exists a non-zero u ∈ F ( p ) ∩ C c ( X ) u\in\mathcal{F}^{(p)}\cap C_c(X) u ∈ F ( p ) ∩ C c ( X ) . By the uniform continuity of u ∈ C c ( X ) u \in C_c(X) u ∈ C c ( X ) , for any ε > 0 \varepsilon> 0 ε > 0 , there exists δ > 0 \delta> 0 δ > 0 such that for any r ∈ ( 0 , δ ) r \in(0,\delta) r ∈ ( 0 , δ ) and any x , y ∈ X x,y \in X x , y ∈ X with d ( x , y ) < r d(x,y) < r d ( x , y ) < r , ∣ u ( x ) − u ( y ) ∣ < ε |u(x)-u(y)| < \varepsilon ∣ u ( x ) − u ( y ) ∣ < ε , then
E β q , X ( q ) ( u , r ) = 1 r β q ∫ X ( 1 V ( x , r ) ∫ B ( x , r ) ∣ u ( x ) − u ( y ) ∣ q m ( d y ) ) m ( d x ) ≤ ε q − p r β p ∫ X ( 1 V ( x , r ) ∫ B ( x , r ) ∣ u ( x ) − u ( y ) ∣ p m ( d y ) ) m ( d x ) = ε q − p E β p , X ( p ) ( u , r ) . \begin{aligned}
\mathcal{E}^{(q)}_{\beta_q,X}(u,r)=\frac{1}{r^{\beta_q}}\int_X\left(\frac{1}{V(x,r)}\int_{B(x,r)}|u(x)-u(y)|^q m(dy)\right)m(dx) \\
&\leq\frac{\varepsilon^{q-p}}{r^{\beta_p}}\int_X\left(\frac{1}{V(x,r)}\int_{B(x,r)}|u(x)-u(y)|^p m(dy)\right)m(dx)=\varepsilon^{q-p}\mathcal{E}^{(p)}_{\beta_p,X}(u,r).
\end{aligned} E β q , X ( q ) ( u , r ) = r β q 1 ∫ X ( V ( x , r ) 1 ∫ B ( x , r ) ∣ u ( x ) − u ( y ) ∣ q m ( d y ) ) m ( d x ) ≤ r β p ε q − p ∫ X ( V ( x , r ) 1 ∫ B ( x , r ) ∣ u ( x ) − u ( y ) ∣ p m ( d y ) ) m ( d x ) = ε q − p E β p , X ( p ) ( u , r ) .
By Lemma 3.4,
lim inf r ↓ 0 E β q , X ( q ) ( u , r ) ≤ ε q − p lim inf r ↓ 0 E β p , X ( p ) ( u , r ) ≲ ε q − p E ( p ) ( u ) < + ∞ , \liminf_{r\downarrow0}\mathcal{E}^{(q)}_{\beta_q,X}(u,r)\leq\varepsilon^{q-p}\liminf_{r\downarrow0}\mathcal{E}^{(p)}_{\beta_p,X}(u,r)\lesssim\varepsilon^{q-p}\mathcal{E}^{(p)}(u)<+\infty, r ↓ 0 lim inf E β q , X ( q ) ( u , r ) ≤ ε q − p r ↓ 0 lim inf E β p , X ( p ) ( u , r ) ≲ ε q − p E ( p ) ( u ) < + ∞ ,
which gives u ∈ F ( q ) u \in\mathcal{F}^{(q)} u ∈ F ( q ) and E ( q ) ( u ) ≲ ε q − p E ( p ) ( u ) \mathcal{E}^{(q)}(u)\lesssim\varepsilon^{q-p}\mathcal{E}^{(p)}(u) E ( q ) ( u ) ≲ ε q − p E ( p ) ( u ) . Since ε > 0 \varepsilon>0 ε > 0 is arbitrary, E ( q ) ( u ) = 0 \mathcal{E}^{(q)}(u)=0 E ( q ) ( u ) = 0 . By PI ( q ) ( β q ) \operatorname{PI}^{(q)}(\beta_q) PI ( q ) ( β q ) , u u u must be constant. Since u ∈ C c ( X ) u\in C_c(X) u ∈ C c ( X ) and X X X is unbounded, it follows that u = 0 u=0 u = 0 , contradicting the assumption that u u u is non-zero.
Fix p , q ∈ I p,q\in I p , q ∈ I with p < q p<q p < q . If β p ≤ d h \beta_p\leq d_h β p ≤ d h , then by Lemma 5.1,
γ p = β p − d h p − 1 ≤ β p − d h q − 1 < β q − d h q − 1 = γ q . (25) \gamma_p=\frac{\beta_p-d_h}{p-1}\leq\frac{\beta_p-d_h}{q-1}<\frac{\beta_q-d_h}{q-1}=\gamma_q. \tag*{(25)} γ p = p − 1 β p − d h ≤ q − 1 β p − d h < q − 1 β q − d h = γ q . (25)
We may assume that β q > β p > d h \beta_q>\beta_p>d_h β q > β p > d h . It suffices to prove the following result.
Lemma 5.2. Assume V ( d h ) V(d_h) V ( d h ) and β p > d h \beta_p>d_h β p > d h . Let ( E ( p ) , F ( p ) ) (\mathcal{E}^{(p)},\mathcal{F}^{(p)}) ( E ( p ) , F ( p ) ) be a p p p -energy satisfying PI ( p ) ( β p ) \operatorname{PI}^{(p)}(\beta_p) PI ( p ) ( β p ) and cap ( p ) ( β p ) ≤ \operatorname{cap}^{(p)}(\beta_p)\leq cap ( p ) ( β p ) ≤ . For any ball B = B ( x 0 , R ) B=B(x_0,R) B = B ( x 0 , R ) , let u ∈ F ( p ) ⊂ C ( X ) u\in\mathcal{F}^{(p)}\subset C(X) u ∈ F ( p ) ⊂ C ( X ) be the E ( p ) \mathcal{E}^{(p)} E ( p ) -capacitary potential for ( B ‾ , X \ 2 B ) (\overline{B},X\backslash2B) ( B , X \2 B ) as in Lemma 3.6(2). Then for any q > p q>p q > p and any β ∈ ( 0 , ( q − 1 ) γ p + d h ] \beta\in(0,(q-1)\gamma_p+d_h] β ∈ ( 0 , ( q − 1 ) γ p + d h ] , we have u ∈ B q , β ( X ) u\in\mathcal{B}^{q,\beta}(X) u ∈ B q , β ( X ) .
Proof of Theorem 2.5(1) using Lemma 5.2. Assume β q > β p > d h \beta_q>\beta_p>d_h β q > β p > d h . Under V ( d h ) V(d_h) V ( d h ) , PI ( p ) ( β p ) \operatorname{PI}^{(p)}(\beta_p) PI ( p ) ( β p ) , and cap ( p ) ( β p ) ≤ \operatorname{cap}^{(p)}(\beta_p)\leq cap ( p ) ( β p ) ≤ , let u ∈ F ( p ) ⊂ C ( X ) u\in\mathcal{F}^{(p)}\subset C(X) u ∈ F ( p ) ⊂ C ( X ) be given as in Lemma 5.2, then u u u is non-zero and u ∈ B q , β ( X ) u\in\mathcal{B}^{q,\beta}(X) u ∈ B q , β ( X ) for any β ∈ ( 0 , ( q − 1 ) γ p + d h ] \beta\in(0,(q-1)\gamma_p+d_h] β ∈ ( 0 , ( q − 1 ) γ p + d h ] . Under V ( d h ) V(d_h) V ( d h ) , PI ( q ) ( β q ) \operatorname{PI}^{(q)}(\beta_q) PI ( q ) ( β q ) , and cap ( q ) ( β q ) ≤ \operatorname{cap}^{(q)}(\beta_q)\leq cap ( q ) ( β q ) ≤ , by Lemma 3.2, the existence of such a function u u u would imply
γ q = q α q ( X ) − d h q − 1 ≥ ( q − 1 ) γ p + d h − d h q − 1 = γ p . \gamma_q=\frac{q\alpha_q(X)-d_h}{q-1}\geq\frac{(q-1)\gamma_p+d_h-d_h}{q-1}=\gamma_p. γ q = q − 1 q α q ( X ) − d h ≥ q − 1 ( q − 1 ) γ p + d h − d h = γ p .
Proof of Lemma 5.2. By Lemma 3.6, u ∈ F ( p ) ⊂ C ( X ) u\in\mathcal{F}^{(p)}\subset C(X) u ∈ F ( p ) ⊂ C ( X ) satisfies 0 ≤ u ≤ 1 0\leq u\leq1 0 ≤ u ≤ 1 on X X X , u = 1 u=1 u = 1 on B ‾ \overline{B} B , u = 0 u=0 u = 0 on X \ 2 B X\backslash2B X \2 B , u u u is harmonic in U : = 2 B \ B ‾ U:=2B\backslash\overline{B} U := 2 B \ B , and E ( p ) ( u ) = cap ( p ) ( B ‾ , X \ 2 B ) ≲ 1 R β p − d h \mathcal{E}^{(p)}(u)=\operatorname{cap}^{(p)}(\overline{B},X\backslash2B)\lesssim\frac{1}{R^{\beta_p-d_h}} E ( p ) ( u ) = cap ( p ) ( B , X \2 B ) ≲ R β p − d h 1 , where the last inequality follows from cap ( p ) ( β p ) ≤ \operatorname{cap}^{(p)}(\beta_p)\leq cap ( p ) ( β p ) ≤ . Moreover,
∣ u ( x ) − u ( y ) ∣ ≤ C 1 ( d ( x , y ) R ) γ p = C 2 d ( x , y ) γ p for any x , y ∈ X , (26) |u(x)-u(y)|\leq C_1\left(\frac{d(x,y)}{R}\right)^{\gamma_p}=C_2d(x,y)^{\gamma_p}\quad\text{for any }x,y\in X, \tag*{(26)} ∣ u ( x ) − u ( y ) ∣ ≤ C 1 ( R d ( x , y ) ) γ p = C 2 d ( x , y ) γ p for any x , y ∈ X , (26)
where C 1 > 0 C_1>0 C 1 > 0 is a constant and C 2 = C 1 R γ p C_2=\frac{C_1}{R^{\gamma_p}} C 2 = R γ p C 1 . For any β > 0 \beta>0 β > 0 and any r > 0 r>0 r > 0 , by V ( d h ) V(d_h) V ( d h ) , [9 Lemma 3.1] gives
E β , X ( q ) ( u , r ) ≤ 2 q − 1 ( 2 d h C R 2 + 1 ) r β ∥ u ∥ L q ( X ; m ) q = C 3 r β ∥ u ∥ L q ( X ; m ) q , \mathcal{E}^{(q)}_{\beta,X}(u,r)\leq\frac{2^{q-1}(2^{d_h}C_R^2+1)}{r^\beta}\|u\|_{L^q(X;m)}^q=\frac{C_3}{r^\beta}\|u\|_{L^q(X;m)}^q, E β , X ( q ) ( u , r ) ≤ r β 2 q − 1 ( 2 d h C R 2 + 1 ) ∥ u ∥ L q ( X ; m ) q = r β C 3 ∥ u ∥ L q ( X ; m ) q ,
where C 3 = 2 q − 1 ( 2 d h C R 2 + 1 ) C_3=2^{q-1}(2^{d_h}C_R^2+1) C 3 = 2 q − 1 ( 2 d h C R 2 + 1 ) , hence
sup r ∈ [ R , + ∞ ) E β , X ( q ) ( u , r ) ≤ C 3 R β ∥ u ∥ L q ( X ; m ) q < + ∞ . (27) \sup_{r\in[R,+\infty)}\mathcal{E}^{(q)}_{\beta,X}(u,r)\leq\frac{C_3}{R^\beta}\|u\|_{L^q(X;m)}^q<+\infty. \tag*{(27)} r ∈ [ R , + ∞ ) sup E β , X ( q ) ( u , r ) ≤ R β C 3 ∥ u ∥ L q ( X ; m ) q < + ∞. (27)
For any r ∈ ( 0 , R ) r\in(0,R) r ∈ ( 0 , R ) , by (5.2),
E β , X ( q ) ( u , r ) = E β , 4 B ( q ) ( u , r ) ≤ 1 r β ∫ 4 B ( 1 V ( x , r ) ∫ B ( x , r ) ( C 2 d ( x , y ) ) γ p ∣ u ( x ) − u ( y ) ∣ p m ( d y ) ) m ( d x ) ≤ C 2 q − p r ( q − p ) γ p + β p − β E E β p , 4 B ( p ) ( u , r ) = C 2 q − p r ( q − 1 ) γ p + d h − β E E β p , X ( p ) ( u , r ) . \begin{aligned}
\mathcal{E}^{(q)}_{\beta,X}(u,r)=\mathcal{E}^{(q)}_{\beta,4B}(u,r) \\
&\le\frac{1}{r^{\beta}} \int_{4B} \left(\frac{1}{V(x,r)} \int_{B(x,r)} \left(C2d(x,y)\right)^{\gamma_p} |u(x)-u(y)|^p m(dy)\right) m(dx) \\
&\le C_{2}^{q-p}r^{(q-p)\gamma_p+\beta_p-\beta_E} E_{\beta_p,4B}^{(p)}(u,r)=C_{2}^{q-p}r^{(q-1)\gamma_p+d_h-\beta_E} E_{\beta_p,X}^{(p)}(u,r).
\end{aligned} E β , X ( q ) ( u , r ) = E β , 4 B ( q ) ( u , r ) ≤ r β 1 ∫ 4 B ( V ( x , r ) 1 ∫ B ( x , r ) ( C 2 d ( x , y ) ) γ p ∣ u ( x ) − u ( y ) ∣ p m ( d y ) ) m ( d x ) ≤ C 2 q − p r ( q − p ) γ p + β p − β E E β p , 4 B ( p ) ( u , r ) = C 2 q − p r ( q − 1 ) γ p + d h − β E E β p , X ( p ) ( u , r ) .
Since u ∈ F ( p ) u\in\mathcal{F}^{(p)} u ∈ F ( p ) , by V ( d h ) V(d_h) V ( d h ) , PI ( p ) ( β p ) \operatorname{PI}^{(p)}(\beta_p) PI ( p ) ( β p ) , and cap ( p ) ( β p ) ≤ \operatorname{cap}^{(p)}(\beta_p)_{\leq} cap ( p ) ( β p ) ≤ , Lemma 3.4 gives
sup r ∈ ( 0 , R ) E β p , X ( p ) ( u , r ) < + ∞ . \sup_{r\in(0,R)} E_{\beta_p,X}^{(p)}(u,r)<+\infty. r ∈ ( 0 , R ) sup E β p , X ( p ) ( u , r ) < + ∞.
Hence for any β ∈ ( 0 , ( q − 1 ) γ p + d h ] \beta\in(0,(q-1)\gamma_p+d_h] β ∈ ( 0 , ( q − 1 ) γ p + d h ] ,
sup r ∈ ( 0 , R ) E β , X ( q ) ( u , r ) < + ∞ . \sup_{r\in(0,R)} E_{\beta,X}^{(q)}(u,r)<+\infty. r ∈ ( 0 , R ) sup E β , X ( q ) ( u , r ) < + ∞.
Combining this with (5.3), u ∈ B q β q ( X ) u\in B^{\frac{\beta}{q}}_q(X) u ∈ B q q β ( X ) for any β ∈ ( 0 , ( q − 1 ) γ p + d h ] \beta\in(0,(q-1)\gamma_p+d_h] β ∈ ( 0 , ( q − 1 ) γ p + d h ] .
Proof of Theorem 2.5(2)
By CC and [25 PROPOSITION A.1] , d d d is Lipschitz equivalent to a geodesic metric. Throughout this section, we may assume without loss of generality that d d d is a geodesic metric.
By Lemmas 3.1 and 3.2, together with Theorem 2.5(1) and (2.2), the function p ↦ γ p = β p − d h p − 1 p\mapsto\gamma_p=\frac{\beta_p-d_h}{p-1} p ↦ γ p = p − 1 β p − d h takes values in ( − ∞ , 1 ] (-\infty,1] ( − ∞ , 1 ] and is continuous and increasing on I I I . Hence,
J : = { p ∈ I : γ p = 1 } . J:=\{p\in I:\gamma_p=1\}. J := { p ∈ I : γ p = 1 } .
is relatively closed in I I I . Moreover, if p ∈ J p\in J p ∈ J , then [ p , + ∞ ) ∩ I ⊆ J [p,+\infty)\cap I\subseteq J [ p , + ∞ ) ∩ I ⊆ J . It therefore suffices to show that, for any p ∈ J p\in J p ∈ J , there exists ε > 0 \varepsilon>0 ε > 0 such that ( p − ε , p ] ⊆ J (p-\varepsilon,p]\subseteq J ( p − ε , p ] ⊆ J . This would imply that J J J is relatively open in I I I . Since I I I is an open interval, either J = ∅ J=\varnothing J = ∅ or J = I J=I J = I . In what follows, we assume that
I 0 : = { p ∈ I : β p > d h } ≠ ∅ I_0:=\{p\in I:\beta_p>d_h\}\ne\varnothing I 0 := { p ∈ I : β p > d h } = ∅
and restrict our attention to parameters in I 0 I_0 I 0 , where Lemma 3.6 applies.
The proof is motivated by the argument for the upper bound in [44 Proposition 2.1] , particularly by the conditions under which equality can hold. That argument suggests that, if there exists p p p with γ p = 1 \gamma_p=1 γ p = 1 , that is, β p = d h + p − 1 \beta_p=d_h+p-1 β p = d h + p − 1 , then, roughly speaking, there exists an integer N N N such that for any pair of concentric balls, at most N N N “disjoint” chains of small balls can connect their boundaries, as on the Vicsek set; see Figure 2 . This suggests that the space should have some form of tree structure. However, connecting each pair of adjacent integers in Z \mathbb{Z} Z by two parallel edges gives a space containing cycles, as illustrated in Figure 3 , but this space has the same Hausdorff dimension d h = 1 d_h=1 d h = 1 and the same p p p -walk dimension β p = p \beta_p=p β p = p as R \mathbb{R} R . Thus, β p = d h + p − 1 \beta_p=d_h+p-1 β p = d h + p − 1 . Nevertheless, this example has a local tree structure, suggesting that such a structure may still be expected at small scales. This motivates the following definition of a balanced triple, which is intended to capture this structure quantitatively.
Figure 2. Disjoint chains across an annulus
Figure 3. The integer lattice with doubled edges
For r > 0 r>0 r > 0 and x , y , z ∈ X x,y,z\in X x , y , z ∈ X , we say that ( x , y , z ) (x,y,z) ( x , y , z ) is a balanced triple at scale r r r if
1 2 r ≤ min { d ( x , y ) , d ( x , z ) , d ( y , z ) } ≤ max { d ( x , y ) , d ( x , z ) , d ( y , z ) } ≤ 2 r . \frac{1}{2}r\leq\min\{d(x,y),d(x,z),d(y,z)\}\leq\max\{d(x,y),d(x,z),d(y,z)\}\leq2r. 2 1 r ≤ min { d ( x , y ) , d ( x , z ) , d ( y , z )} ≤ max { d ( x , y ) , d ( x , z ) , d ( y , z )} ≤ 2 r .
We define its deficit by
Δ ( p ) ( x , y , z ; r ) : = 1 r γ p ( ρ ( p ) ( x , y ) + ρ ( p ) ( y , z ) − ρ ( p ) ( x , z ) ) . \Delta^{(p)}(x,y,z;r) := \frac{1}{r^{\gamma_p}}\left(\rho^{(p)}(x,y)+\rho^{(p)}(y,z)-\rho^{(p)}(x,z)\right). Δ ( p ) ( x , y , z ; r ) := r γ p 1 ( ρ ( p ) ( x , y ) + ρ ( p ) ( y , z ) − ρ ( p ) ( x , z ) ) .
Then by the triangle inequality for ρ ( p ) \rho^{(p)} ρ ( p ) and (3.3), 0 ≤ Δ ( p ) ( x , y , z ; r ) ≲ 1 0 \leq\Delta^{(p)}(x,y,z;r) \lesssim1 0 ≤ Δ ( p ) ( x , y , z ; r ) ≲ 1 . The quantity Δ ( p ) \Delta^{(p)} Δ ( p ) is a normalized ordered triangle inequality deficit for the resistance metric ρ ( p ) \rho^{(p)} ρ ( p ) . Such deficits and triples with comparable side lengths have been considered previously; see, for example, [38 Section 1.1, Remark 1.2] and [17 Eqs. (2.2)–(2.4)] .
The key ingredient in the proof is the following result.
Proposition 6.1. Let p , q ∈ I 0 p,q \in I_0 p , q ∈ I 0 . Assume γ p = 1 \gamma_p=1 γ p = 1 . Then Δ ( p ) → 0 \Delta^{(p)} \to0 Δ ( p ) → 0 implies Δ ( q ) → 0 \Delta^{(q)} \to0 Δ ( q ) → 0 . More precisely, for any ξ > 0 \xi>0 ξ > 0 , there exists κ > 0 \kappa>0 κ > 0 such that, for any r > 0 r>0 r > 0 and any balanced triple ( x , y , z ) (x,y,z) ( x , y , z ) at scale r r r satisfying Δ ( p ) ( x , y , z ; r ) < κ \Delta^{(p)}(x,y,z;r)<\kappa Δ ( p ) ( x , y , z ; r ) < κ , we have Δ ( q ) ( x , y , z ; r ) < ξ \Delta^{(q)}(x,y,z;r)<\xi Δ ( q ) ( x , y , z ; r ) < ξ .
Proof of Theorem 2.5(2) using Proposition 6.1. For simplicity, take x , y ∈ X x,y\in X x , y ∈ X with d ( x , y ) = 1 d(x,y)=1 d ( x , y ) = 1 , then there exists a family of points
{ x n , k : k = 0 , 1 , … , 2 n , n = 0 , 1 , … } , \{x_{n,k}: k=0,1,\ldots,2^n,\ n=0,1,\ldots\}, { x n , k : k = 0 , 1 , … , 2 n , n = 0 , 1 , … } ,
such that, for any n ≥ 0 n\geq0 n ≥ 0 , x n , 0 = x x_{n,0}=x x n , 0 = x and x n , 2 n = y x_{n,2^n}=y x n , 2 n = y , for any k = 0 , 1 , … , 2 n − 1 k=0,1,\ldots,2^n-1 k = 0 , 1 , … , 2 n − 1 , d ( x n , k , x n , k + 1 ) = 1 2 n d(x_{n,k},x_{n,k+1})=\frac{1}{2^n} d ( x n , k , x n , k + 1 ) = 2 n 1 , and x n , k = x n + 1 , 2 k x_{n,k}=x_{n+1,2k} x n , k = x n + 1 , 2 k . For any s ∈ I 0 s\in I_0 s ∈ I 0 and any n ≥ 0 n\geq0 n ≥ 0 , let
L n ( s ) : = ∑ k = 0 2 n − 1 ρ ( s ) ( x n , k , x n , k + 1 ) . L_n^{(s)}:=\sum_{k=0}^{2^n-1}\rho^{(s)}(x_{n,k},x_{n,k+1}). L n ( s ) := k = 0 ∑ 2 n − 1 ρ ( s ) ( x n , k , x n , k + 1 ) .
Since ρ ( s ) ( x , y ) ≍ d ( x , y ) γ s \rho^{(s)}(x,y)\asymp d(x,y)^{\gamma_s} ρ ( s ) ( x , y ) ≍ d ( x , y ) γ s ,
L n ( s ) ≍ ∑ k = 0 2 n − 1 d ( x n , k , x n , k + 1 ) γ s = 2 ( 1 − γ s ) n . (28) L_n^{(s)}\asymp\sum_{k=0}^{2^n-1}d(x_{n,k},x_{n,k+1})^{\gamma_s}=2^{(1-\gamma_s)n}. \tag*{(28)} L n ( s ) ≍ k = 0 ∑ 2 n − 1 d ( x n , k , x n , k + 1 ) γ s = 2 ( 1 − γ s ) n . (28)
By the triangle inequality for ρ ( s ) \rho^{(s)} ρ ( s ) , { L n ( s ) } n \{L_n^{(s)}\}_n { L n ( s ) } n is increasing. Moreover,
L n + 1 ( s ) − L n ( s ) = ∑ k = 0 2 n − 1 ( ρ ( s ) ( x n + 1 , 2 k , x n + 1 , 2 k + 1 ) + ρ ( s ) ( x n + 1 , 2 k + 1 , x n + 1 , 2 k + 2 ) − ρ ( s ) ( x n , k , x n , k + 1 ) ) = ∑ k = 0 2 n − 1 ( ρ ( s ) ( x n , k , x n + 1 , 2 k + 1 ) + ρ ( s ) ( x n + 1 , 2 k + 1 , x n , k + 1 ) − ρ ( s ) ( x n , k , x n , k + 1 ) ) = 1 2 γ s n ∑ k = 0 2 n − 1 Δ ( s ) ( x n , k , x n + 1 , 2 k + 1 , x n , k + 1 ; 1 2 n ) = 1 2 γ s n ∑ k = 0 2 n − 1 Δ n , k ( s ) . \begin{aligned}
L_{n+1}^{(s)}-L_n^{(s)}
&=\sum_{k=0}^{2^n-1}\left(\rho^{(s)}(x_{n+1,2k},x_{n+1,2k+1})+\rho^{(s)}(x_{n+1,2k+1},x_{n+1,2k+2})-\rho^{(s)}(x_{n,k},x_{n,k+1})\right)\\
&=\sum_{k=0}^{2^n-1}\left(\rho^{(s)}(x_{n,k},x_{n+1,2k+1})+\rho^{(s)}(x_{n+1,2k+1},x_{n,k+1})-\rho^{(s)}(x_{n,k},x_{n,k+1})\right)\\
&=\frac{1}{2^{\gamma_s n}}\sum_{k=0}^{2^n-1}\Delta^{(s)}\left(x_{n,k},x_{n+1,2k+1},x_{n,k+1};\frac{1}{2^n}\right)
=\frac{1}{2^{\gamma_s n}}\sum_{k=0}^{2^n-1}\Delta_{n,k}^{(s)}.
\end{aligned} L n + 1 ( s ) − L n ( s ) = k = 0 ∑ 2 n − 1 ( ρ ( s ) ( x n + 1 , 2 k , x n + 1 , 2 k + 1 ) + ρ ( s ) ( x n + 1 , 2 k + 1 , x n + 1 , 2 k + 2 ) − ρ ( s ) ( x n , k , x n , k + 1 ) ) = k = 0 ∑ 2 n − 1 ( ρ ( s ) ( x n , k , x n + 1 , 2 k + 1 ) + ρ ( s ) ( x n + 1 , 2 k + 1 , x n , k + 1 ) − ρ ( s ) ( x n , k , x n , k + 1 ) ) = 2 γ s n 1 k = 0 ∑ 2 n − 1 Δ ( s ) ( x n , k , x n + 1 , 2 k + 1 , x n , k + 1 ; 2 n 1 ) = 2 γ s n 1 k = 0 ∑ 2 n − 1 Δ n , k ( s ) .
where Δ n , k ( s ) : = Δ ( s ) ( x n , k , x n + 1 , 2 k + 1 , x n , k + 1 ; 1 2 n ) \Delta_{n,k}^{(s)}:=\Delta^{(s)}\left(x_{n,k},x_{n+1,2k+1},x_{n,k+1};\frac{1}{2^n}\right) Δ n , k ( s ) := Δ ( s ) ( x n , k , x n + 1 , 2 k + 1 , x n , k + 1 ; 2 n 1 ) is bounded by a constant C ( s ) > 0 C^{(s)}>0 C ( s ) > 0 independent of n n n and k k k . Let
M n ( s ) : = 1 2 n ∑ k = 0 2 n − 1 Δ n , k ( s ) , M_n^{(s)}:=\frac{1}{2^n}\sum_{k=0}^{2^n-1}\Delta_{n,k}^{(s)}, M n ( s ) := 2 n 1 k = 0 ∑ 2 n − 1 Δ n , k ( s ) ,
then
L n + 1 ( s ) − L n ( s ) = 2 ( 1 − γ s ) n M n ( s ) . (29) L_{n+1}^{(s)}-L_n^{(s)}=2^{(1-\gamma_s)n}M_n^{(s)}. \tag*{(29)} L n + 1 ( s ) − L n ( s ) = 2 ( 1 − γ s ) n M n ( s ) . (29)
Since γ p = 1 \gamma_p=1 γ p = 1 , by (28) , { L n ( p ) } n \{L_n^{(p)}\}_n { L n ( p ) } n is bounded and increasing. By (29) , ∑ n = 0 + ∞ M n ( p ) < + ∞ \sum_{n=0}^{+\infty}M_n^{(p)}<+\infty ∑ n = 0 + ∞ M n ( p ) < + ∞ , which gives lim n → + ∞ M n ( p ) = 0 \lim_{n\to+\infty}M_n^{(p)}=0 lim n → + ∞ M n ( p ) = 0 .
Let q ∈ I 0 q \in I_0 q ∈ I 0 . For any ξ > 0 \xi> 0 ξ > 0 , by Proposition 6.1 , there exists κ > 0 \kappa> 0 κ > 0 such that Δ n , k ( p ) < κ \Delta_{n,k}^{(p)} < \kappa Δ n , k ( p ) < κ implies Δ n , k ( q ) < ξ \Delta_{n,k}^{(q)} < \xi Δ n , k ( q ) < ξ . Noting that
M n ( p ) ≥ 1 2 n ∑ k : Δ n , k ( p ) ≥ κ Δ n , k ( p ) ≥ κ 2 n # { k : Δ n , k ( p ) ≥ κ } , M_n^{(p)} \ge\frac{1}{2^n}\sum_{k:\Delta_{n,k}^{(p)}\ge\kappa}\Delta_{n,k}^{(p)} \ge\frac{\kappa}{2^n}\#\{k:\Delta_{n,k}^{(p)}\ge\kappa\}, M n ( p ) ≥ 2 n 1 k : Δ n , k ( p ) ≥ κ ∑ Δ n , k ( p ) ≥ 2 n κ # { k : Δ n , k ( p ) ≥ κ } ,
we have
M n ( q ) = 1 2 n ( ∑ k : Δ n , k ( p ) < κ + ∑ k : Δ n , k ( p ) ≥ κ ) Δ n , k ( q ) ≤ 1 2 n ∑ k : Δ n , k ( p ) < κ ξ + 1 2 n ∑ k : Δ n , k ( p ) ≥ κ C ( q ) ≤ ξ + C ( q ) 2 n # { k : Δ n , k ( p ) ≥ κ } ≤ ξ + C ( q ) κ M n ( p ) . \begin{aligned}
M_n^{(q)} &= \frac{1}{2^n}\left(\sum_{k:\Delta_{n,k}^{(p)}<\kappa}+\sum_{k:\Delta_{n,k}^{(p)}\ge\kappa}\right)\Delta_{n,k}^{(q)}
\le\frac{1}{2^n}\sum_{k:\Delta_{n,k}^{(p)}<\kappa}\xi+\frac{1}{2^n}\sum_{k:\Delta_{n,k}^{(p)}\ge\kappa}C^{(q)}\\
&\le\xi+\frac{C^{(q)}}{2^n}\#\{k:\Delta_{n,k}^{(p)}\ge\kappa\}\le\xi+\frac{C^{(q)}}{\kappa}M_n^{(p)}.
\end{aligned} M n ( q ) = 2 n 1 k : Δ n , k ( p ) < κ ∑ + k : Δ n , k ( p ) ≥ κ ∑ Δ n , k ( q ) ≤ 2 n 1 k : Δ n , k ( p ) < κ ∑ ξ + 2 n 1 k : Δ n , k ( p ) ≥ κ ∑ C ( q ) ≤ ξ + 2 n C ( q ) # { k : Δ n , k ( p ) ≥ κ } ≤ ξ + κ C ( q ) M n ( p ) .
Letting n → + ∞ n\to+\infty n → + ∞ and then ξ ↓ 0 \xi\downarrow0 ξ ↓ 0 , we obtain lim n → + ∞ M n ( q ) = 0 \lim_{n\to+\infty}M_n^{(q)}=0 lim n → + ∞ M n ( q ) = 0 . By (28) and (29) ,
0 ≤ L n + 1 ( q ) L n ( q ) − 1 = 2 ( 1 − γ q ) n L n ( q ) M n ( q ) ≲ M n ( q ) → 0. 0\le\frac{L_{n+1}^{(q)}}{L_n^{(q)}}-1=\frac{2^{(1-\gamma_q)n}}{L_n^{(q)}}M_n^{(q)}\lesssim M_n^{(q)}\to0. 0 ≤ L n ( q ) L n + 1 ( q ) − 1 = L n ( q ) 2 ( 1 − γ q ) n M n ( q ) ≲ M n ( q ) → 0.
Thus lim n → + ∞ log L n + 1 ( q ) L n ( q ) = 0 \lim_{n\to+\infty}\log\frac{L_{n+1}^{(q)}}{L_n^{(q)}}=0 lim n → + ∞ log L n ( q ) L n + 1 ( q ) = 0 , and lim N → + ∞ 1 N ∑ n = 0 N − 1 log L n + 1 ( q ) L n ( q ) = 0 \lim_{N\to+\infty}\frac{1}{N}\sum_{n=0}^{N-1}\log\frac{L_{n+1}^{(q)}}{L_n^{(q)}}=0 lim N → + ∞ N 1 ∑ n = 0 N − 1 log L n ( q ) L n + 1 ( q ) = 0 , which gives
lim N → + ∞ 1 N log L N ( q ) = 0. \lim_{N\to+\infty}\frac{1}{N}\log L_N^{(q)}=0. N → + ∞ lim N 1 log L N ( q ) = 0.
On the other hand, by (28) ,
lim N → + ∞ 1 N log L N ( q ) = ( 1 − γ q ) log 2. \lim_{N\to+\infty}\frac{1}{N}\log L_N^{(q)}=(1-\gamma_q)\log2. N → + ∞ lim N 1 log L N ( q ) = ( 1 − γ q ) log 2.
Therefore, γ q = 1 \gamma_q=1 γ q = 1 .
A map θ : [ 0 , 1 ] → X \theta:[0,1]\to X θ : [ 0 , 1 ] → X is called a curve if it is continuous. For A ⊆ X A\subseteq X A ⊆ X , write θ ∩ A = θ ( [ 0 , 1 ] ) ∩ A \theta\cap A=\theta([0,1])\cap A θ ∩ A = θ ([ 0 , 1 ]) ∩ A . Let K 0 , K 1 , U ⊆ X K_0,K_1,U\subseteq X K 0 , K 1 , U ⊆ X be such that K 0 K_0 K 0 and K 1 K_1 K 1 are compact sets, U U U is open, and K 0 ∪ K 1 ⊆ U K_0\cup K_1\subseteq U K 0 ∪ K 1 ⊆ U . Define
Path ( K 0 , K 1 ; U ) : = { θ : [ 0 , 1 ] → X : θ is a curve , θ ( 0 ) ∈ K 0 , θ ( 1 ) ∈ K 1 , θ ( [ 0 , 1 ] ) ⊆ U } . \operatorname{Path}(K_0,K_1;U):=\{\theta:[0,1]\to X:\theta\text{ is a curve},\ \theta(0)\in K_0,\ \theta(1)\in K_1,\ \theta([0,1])\subseteq U\}. Path ( K 0 , K 1 ; U ) := { θ : [ 0 , 1 ] → X : θ is a curve , θ ( 0 ) ∈ K 0 , θ ( 1 ) ∈ K 1 , θ ([ 0 , 1 ]) ⊆ U } .
For x , y ∈ U x,y\in U x , y ∈ U , write
Path ( x , y ; U ) = Path ( { x } , { y } ; U ) . \operatorname{Path}(x,y;U)=\operatorname{Path}(\{x\},\{y\};U). Path ( x , y ; U ) = Path ({ x } , { y } ; U ) .
The proof of Proposition 6.1 reduces to the following two results.
Lemma 6.2. Let p ∈ I 0 p\in I_0 p ∈ I 0 . For any ξ > 0 \xi>0 ξ > 0 , there exist δ 0 , L 0 > 0 \delta_0,L_0>0 δ 0 , L 0 > 0 such that, for any δ ∈ ( 0 , δ 0 ] \delta\in(0,\delta_0] δ ∈ ( 0 , δ 0 ] , L ∈ [ L 0 , + ∞ ) L\in[L_0,+\infty) L ∈ [ L 0 , + ∞ ) , and r > 0 r>0 r > 0 , and any balanced triple ( x , y , z ) (x,y,z) ( x , y , z ) at scale r r r satisfying Δ ( p ) ( x , y , z ; r ) ≥ ξ \Delta^{(p)}(x,y,z;r)\ge\xi Δ ( p ) ( x , y , z ; r ) ≥ ξ , there exists θ ∈ Path ( x , z ; B ( y , L r ) ) \theta\in\operatorname{Path}(x,z;B(y,Lr)) θ ∈ Path ( x , z ; B ( y , L r )) such that θ ∩ B ( y , δ r ) = ∅ \theta\cap B(y,\delta r)=\varnothing θ ∩ B ( y , δ r ) = ∅ .
Lemma 6.3. Let p ∈ I 0 p\in I_0 p ∈ I 0 . Assume γ p = 1 \gamma_p=1 γ p = 1 . For any δ , L > 0 \delta,L>0 δ , L > 0 , there exists κ > 0 \kappa>0 κ > 0 such that, for any r > 0 r>0 r > 0 and any balanced triple ( x , y , z ) (x,y,z) ( x , y , z ) at scale r r r , the following holds: if there exists θ ∈ Path ( x , z ; B ( y , L r ) ) \theta\in\operatorname{Path}(x,z;B(y,Lr)) θ ∈ Path ( x , z ; B ( y , L r )) such that θ ∩ B ( y , δ r ) = ∅ \theta\cap B(y,\delta r)=\varnothing θ ∩ B ( y , δ r ) = ∅ , then Δ ( p ) ( x , y , z ; r ) ≥ κ \Delta^{(p)}(x,y,z;r)\ge\kappa Δ ( p ) ( x , y , z ; r ) ≥ κ .
Proof of Proposition 6.1 using Lemmas 6.2 and 6.3. Let p , q ∈ I 0 p,q\in I_0 p , q ∈ I 0 . Assume γ p = 1 \gamma_p=1 γ p = 1 . Let ( x , y , z ) (x,y,z) ( x , y , z ) be a balanced triple at scale r r r . For any ξ > 0 \xi>0 ξ > 0 , choose δ 0 , L 0 > 0 \delta_0,L_0>0 δ 0 , L 0 > 0 as in Lemma 6.2 with q q q in place of p p p , and set δ = δ 0 \delta=\delta_0 δ = δ 0 and L = L 0 L=L_0 L = L 0 . For these choices, let κ > 0 \kappa>0 κ > 0 be as in Lemma 6.3 for the exponent p p p .
We claim that if Δ ( p ) ( x , y , z ; r ) < κ \Delta^{(p)}(x,y,z;r)<\kappa Δ ( p ) ( x , y , z ; r ) < κ , then Δ ( q ) ( x , y , z ; r ) < ξ \Delta^{(q)}(x,y,z;r)<\xi Δ ( q ) ( x , y , z ; r ) < ξ . Indeed, suppose that Δ ( q ) ( x , y , z ; r ) ≥ ξ \Delta^{(q)}(x,y,z;r)\ge\xi Δ ( q ) ( x , y , z ; r ) ≥ ξ . By Lemma 6.2, there exists θ ∈ Path ( x , z ; B ( y , L r ) ) \theta\in\operatorname{Path}(x,z;B(y,Lr)) θ ∈ Path ( x , z ; B ( y , L r )) such that θ ∩ B ( y , δ r ) = ∅ \theta\cap B(y,\delta r)=\varnothing θ ∩ B ( y , δ r ) = ∅ . It follows from Lemma 6.3 that Δ ( p ) ( x , y , z ; r ) ≥ κ \Delta^{(p)}(x,y,z;r)\ge\kappa Δ ( p ) ( x , y , z ; r ) ≥ κ , a contradiction.
We will frequently use the following elementary identity: for any A , B > 0 A,B>0 A , B > 0 ,
min { A ( 1 − λ ) p + B λ p : λ ∈ [ 0 , 1 ] } = ( A ( 1 − λ ) p + B λ p ) ∣ λ = A 1 p − 1 A 1 p − 1 + B 1 p − 1 = 1 ( 1 A 1 p − 1 + 1 B 1 p − 1 ) p − 1 . (30) \min\{A(1-\lambda)^p+B\lambda^p:\lambda\in[0,1]\}
=\left.(A(1-\lambda)^p+B\lambda^p)\right|_{\lambda=\frac{A^{\frac{1}{p-1}}}{A^{\frac{1}{p-1}}+B^{\frac{1}{p-1}}}}
=\frac{1}{\left(\frac{1}{A^{\frac{1}{p-1}}}+\frac{1}{B^{\frac{1}{p-1}}}\right)^{p-1}}. \tag*{(30)} min { A ( 1 − λ ) p + B λ p : λ ∈ [ 0 , 1 ]} = ( A ( 1 − λ ) p + B λ p ) ∣ λ = A p − 1 1 + B p − 1 1 A p − 1 1 = ( A p − 1 1 1 + B p − 1 1 1 ) p − 1 1 . (30)
The value of λ \lambda λ displayed above is the unique minimizer.
For p ∈ I 0 p\in I_0 p ∈ I 0 , define
F e ( p ) : = { u : u is an m -measurable function that is finite m -a.e., and there exists an E ( p ) -Cauchy sequence { u n } ⊆ F ( p ) , such that u n → u m -a.e. } . (31) \mathcal{F}^{(p)}_e:=\left\{u:\begin{array}{l}
u\text{ is an }m\text{-measurable function that is finite }m\text{-a.e.,}\\
\text{and there exists an }\mathcal{E}^{(p)}\text{-Cauchy sequence }\{u_n\}\subseteq\mathcal{F}^{(p)},\\
\text{such that }u_n\to u\ m\text{-a.e.}
\end{array}\right\}. \tag*{(31)} F e ( p ) := ⎩ ⎨ ⎧ u : u is an m -measurable function that is finite m -a.e., and there exists an E ( p ) -Cauchy sequence { u n } ⊆ F ( p ) , such that u n → u m -a.e. ⎭ ⎬ ⎫ . (31)
Arguing as in [22 Lemma 8.6] , we deduce from the Morrey–Sobolev inequality and an approximation argument that any f ∈ F e ( p ) f\in\mathcal{F}^{(p)}_e f ∈ F e ( p ) admits a continuous version, which we still denote by f f f . The same arguments show that ( E ( p ) , F e ( p ) ) (\mathcal{E}^{(p)},\mathcal{F}^{(p)}_e) ( E ( p ) , F e ( p ) ) satisfies (RF1)p _p p –(RF4)p _p p of [27 Definition 6.1] . The Markovian property, p p p -Clarkson’s inequality, and strong sub-additivity extend to F e ( p ) \mathcal{F}^{(p)}_e F e ( p ) by approximation. In particular, 1 ∈ F e ( p ) 1\in\mathcal{F}^{(p)}_e 1 ∈ F e ( p ) . Moreover, for any u ∈ F e ( p ) u\in\mathcal{F}^{(p)}_e u ∈ F e ( p ) and any x , y ∈ X x,y\in X x , y ∈ X ,
∣ u ( x ) − u ( y ) ∣ p ≤ R ( p ) ( x , y ) E ( p ) ( u ) . |u(x)-u(y)|^p\le R^{(p)}(x,y)\mathcal{E}^{(p)}(u). ∣ u ( x ) − u ( y ) ∣ p ≤ R ( p ) ( x , y ) E ( p ) ( u ) .
By approximation, closedness, and Alg, we have η f ∈ F ( p ) ∩ C c ( X ) \eta f\in\mathcal{F}^{(p)}\cap C_c(X) η f ∈ F ( p ) ∩ C c ( X ) for any f ∈ F e ( p ) f\in\mathcal{F}^{(p)}_e f ∈ F e ( p ) and any η ∈ F ( p ) ∩ C c ( X ) \eta\in\mathcal{F}^{(p)}\cap C_c(X) η ∈ F ( p ) ∩ C c ( X ) ; see also the localization argument in the proof of [22 Lemma 8.11] . We can also extend the energy measures to F e ( p ) \mathcal{F}^{(p)}_e F e ( p ) , preserving their locality property.
For any non-empty disjoint compact subsets A , B ⊆ X A,B\subseteq X A , B ⊆ X , there exists a unique u ∈ F e ( p ) u\in\mathcal{F}^{(p)}_e u ∈ F e ( p ) with u = 0 u=0 u = 0 on A A A and u = 1 u=1 u = 1 on B B B , such that
E ( p ) ( u ) = cap ( p ) ( A , B ) = inf { E ( p ) ( u ) : u ∈ F e ( p ) , u = 0 on A , u = 1 on B } . \mathcal{E}^{(p)}(u)=\operatorname{cap}^{(p)}(A,B)=\inf\{\mathcal{E}^{(p)}(u):u\in\mathcal{F}^{(p)}_e,\ u=0\text{ on }A,\ u=1\text{ on }B\}. E ( p ) ( u ) = cap ( p ) ( A , B ) = inf { E ( p ) ( u ) : u ∈ F e ( p ) , u = 0 on A , u = 1 on B } .
Such a function u u u is called the equilibrium potential between A A A and B B B . It satisfies 0 ≤ u ≤ 1 0\le u\le1 0 ≤ u ≤ 1 on X X X and is harmonic in X ∖ ( A ∪ B ) X\setminus(A\cup B) X ∖ ( A ∪ B ) .
Finally, by [41 Theorem 2.3] , ( E ( p ) , F ( p ) ) (\mathcal{E}^{(p)},\mathcal{F}^{(p)}) ( E ( p ) , F ( p ) ) satisfies the elliptic Harnack inequality, which also holds for bounded non-negative harmonic functions in F e ( p ) \mathcal{F}^{(p)}_e F e ( p ) by localization.
First, we prove Lemma 6.2, which follows directly from the following result.
Lemma 6.4. Let p ∈ I 0 p\in I_0 p ∈ I 0 . There exist δ 0 , L 0 , C > 0 \delta_0,L_0,C>0 δ 0 , L 0 , C > 0 such that, for any δ ∈ ( 0 , δ 0 ] \delta\in(0,\delta_0] δ ∈ ( 0 , δ 0 ] , L ∈ [ L 0 , + ∞ ) L\in[L_0,+\infty) L ∈ [ L 0 , + ∞ ) , and r > 0 r>0 r > 0 , and any balanced triple ( x 0 , y 0 , z 0 ) (x_0,y_0,z_0) ( x 0 , y 0 , z 0 ) at scale r r r , if for any θ ∈ Path ( x 0 , z 0 ; B ( y 0 , L r ) ) \theta\in\operatorname{Path}(x_0,z_0;B(y_0,Lr)) θ ∈ Path ( x 0 , z 0 ; B ( y 0 , L r )) , we have θ ∩ B ( y 0 , 2 δ r ) ≠ ∅ \theta\cap B(y_0,2\delta r)\ne\varnothing θ ∩ B ( y 0 , 2 δ r ) = ∅ , then
Δ ( p ) ( x 0 , y 0 , z 0 ; r ) ≤ C ( δ β p − d h p + 1 L β p − d h ) . \Delta^{(p)}(x_0,y_0,z_0;r)\le C\left(\delta^{\frac{\beta_p-d_h}{p}}+\frac{1}{L^{\beta_p-d_h}}\right). Δ ( p ) ( x 0 , y 0 , z 0 ; r ) ≤ C ( δ p β p − d h + L β p − d h 1 ) .
Proof of Lemma 6.2 using Lemma 6.4. Let δ 0 , L 0 , C > 0 \delta_0,L_0,C>0 δ 0 , L 0 , C > 0 be as in Lemma 6.4. For any ξ > 0 \xi>0 ξ > 0 , take δ 1 < δ 0 \delta_1<\delta_0 δ 1 < δ 0 and L 1 > L 0 L_1>L_0 L 1 > L 0 such that
C ( δ 1 β p − d h p + 1 L 1 β p − d h ) < ξ . (32) C\left(\delta_1^{\frac{\beta_p-d_h}{p}}+\frac{1}{L_1^{\beta_p-d_h}}\right)<\xi. \tag*{(32)} C ( δ 1 p β p − d h + L 1 β p − d h 1 ) < ξ . (32)
For any δ ∈ ( 0 , δ 1 ] \delta\in(0,\delta_1] δ ∈ ( 0 , δ 1 ] , L ∈ [ L 1 , + ∞ ) L\in[L_1,+\infty) L ∈ [ L 1 , + ∞ ) , any r > 0 r>0 r > 0 , and any balanced triple ( x , y , z ) (x,y,z) ( x , y , z ) at scale r r r satisfying Δ ( p ) ( x , y , z ; r ) ≥ ξ \Delta^{(p)}(x,y,z;r)\ge\xi Δ ( p ) ( x , y , z ; r ) ≥ ξ , we claim that there exists θ ∈ Path ( x , z ; B ( y , L r ) ) \theta\in\operatorname{Path}(x,z;B(y,Lr)) θ ∈ Path ( x , z ; B ( y , L r )) such that θ ∩ B ( y , δ r ) = ∅ \theta\cap B(y,\delta r)=\varnothing θ ∩ B ( y , δ r ) = ∅ . Indeed, suppose that no such path exists, then any path θ ∈ Path ( x , z ; B ( y , L r ) ) \theta\in\operatorname{Path}(x,z;B(y,Lr)) θ ∈ Path ( x , z ; B ( y , L r )) satisfies θ ∩ B ( y , 2 δ r ) ≠ ∅ \theta\cap B(y,2\delta r)\ne\varnothing θ ∩ B ( y , 2 δ r ) = ∅ . By Lemma 6.4 and (6.5), Δ ( p ) ( x , y , z ; r ) < ξ \Delta^{(p)}(x,y,z;r)<\xi Δ ( p ) ( x , y , z ; r ) < ξ , a contradiction. □
Proof of Lemma 6.4. For notational convenience, we omit the superscript ( p ) (p) ( p ) when no ambiguity can arise. Let δ ≪ 1 \delta\ll1 δ ≪ 1 , L ≫ 1 L\gg1 L ≫ 1 be chosen later, and let U = B ( y 0 , L r ) U=B(y_0,Lr) U = B ( y 0 , L r ) , K = B ( y 0 , 2 δ r ) ‾ K=\overline{B(y_0,2\delta r)} K = B ( y 0 , 2 δ r ) . First, we show that there exists a constant C 1 > 0 C_1 > 0 C 1 > 0 such that
ρ ( x 0 , y 0 ) ≥ R ( x 0 , B ( y 0 , 3 δ r ) ‾ ) 1 p − 1 ≥ ( 1 − C 1 δ β p − d h p ) ρ ( x 0 , y 0 ) , (33) \rho(x_0,y_0) \ge R\left(x_0,\overline{B(y_0,3\delta r)}\right)^{\frac{1}{p-1}} \ge\left(1-C_1\delta^{\frac{\beta_p-d_h}{p}}\right)\rho(x_0,y_0), \tag*{(33)} ρ ( x 0 , y 0 ) ≥ R ( x 0 , B ( y 0 , 3 δ r ) ) p − 1 1 ≥ ( 1 − C 1 δ p β p − d h ) ρ ( x 0 , y 0 ) , (33)
ρ ( z 0 , y 0 ) ≥ R ( z 0 , B ( y 0 , 3 δ r ) ‾ ) 1 p − 1 ≥ ( 1 − C 1 δ β p − d h p ) ρ ( z 0 , y 0 ) . (34) \rho(z_0,y_0) \ge R\left(z_0,\overline{B(y_0,3\delta r)}\right)^{\frac{1}{p-1}} \ge\left(1-C_1\delta^{\frac{\beta_p-d_h}{p}}\right)\rho(z_0,y_0). \tag*{(34)} ρ ( z 0 , y 0 ) ≥ R ( z 0 , B ( y 0 , 3 δ r ) ) p − 1 1 ≥ ( 1 − C 1 δ p β p − d h ) ρ ( z 0 , y 0 ) . (34)
Take h ∈ F e h\in\mathcal{F}_e h ∈ F e with h ( x 0 ) = 1 h(x_0)=1 h ( x 0 ) = 1 and h ( y 0 ) = 0 h(y_0)=0 h ( y 0 ) = 0 , such that E ( h ) = cap ( x 0 , y 0 ) ≍ r d h − β p \mathcal{E}(h)=\operatorname{cap}(x_0,y_0)\asymp r^{d_h-\beta_p} E ( h ) = cap ( x 0 , y 0 ) ≍ r d h − β p . For any y ∈ B ( y 0 , 3 δ r ) y\in B(y_0,3\delta r) y ∈ B ( y 0 , 3 δ r ) ,
0 ≤ h ( y ) = ∣ h ( y ) − h ( y 0 ) ∣ ≤ ( R ( y , y 0 ) E ( h ) ) 1 p ≲ ( ( δ r ) β p − d h ( r d h − β p ) ) 1 p = δ β p − d h p , 0\le h(y)=\lvert h(y)-h(y_0)\rvert\le\left(R(y,y_0)\mathcal{E}(h)\right)^{\frac{1}{p}}\lesssim\left((\delta r)^{\beta_p-d_h}\left(r^{d_h-\beta_p}\right)\right)^{\frac{1}{p}}=\delta^{\frac{\beta_p-d_h}{p}}, 0 ≤ h ( y ) = ∣ h ( y ) − h ( y 0 )∣ ≤ ( R ( y , y 0 ) E ( h ) ) p 1 ≲ ( ( δ r ) β p − d h ( r d h − β p ) ) p 1 = δ p β p − d h ,
hence
M : = sup y ∈ B ( y 0 , 3 δ r ) h ( y ) ≲ δ β p − d h p ∈ ( 0 , 1 ) . (35) M:=\sup_{y\in B(y_0,3\delta r)}h(y)\lesssim\delta^{\frac{\beta_p-d_h}{p}}\in(0,1). \tag*{(35)} M := y ∈ B ( y 0 , 3 δ r ) sup h ( y ) ≲ δ p β p − d h ∈ ( 0 , 1 ) . (35)
Then the function ( h − M ) + 1 − M ∈ F e \frac{(h-M)_+}{1-M}\in\mathcal{F}_e 1 − M ( h − M ) + ∈ F e takes the value 1 1 1 at x 0 x_0 x 0 and the value 0 0 0 on B ( y 0 , 3 δ r ) ‾ \overline{B(y_0,3\delta r)} B ( y 0 , 3 δ r ) , hence
R ( x 0 , B ( y 0 , 3 δ r ) ‾ ) − 1 = cap ( x 0 , B ( y 0 , 3 δ r ) ‾ ) ≤ E ( ( h − M ) + 1 − M ) ≤ 1 ( 1 − M ) p E ( h ) = 1 ( 1 − M ) p 1 ρ ( x 0 , y 0 ) p − 1 , \begin{aligned}
R\left(x_0,\overline{B(y_0,3\delta r)}\right)^{-1}=\operatorname{cap}\left(x_0,\overline{B(y_0,3\delta r)}\right) \\
&\le\mathcal{E}\left(\frac{(h-M)_+}{1-M}\right)\le\frac{1}{(1-M)^p}\mathcal{E}(h)=\frac{1}{(1-M)^p}\frac{1}{\rho(x_0,y_0)^{p-1}},
\end{aligned} R ( x 0 , B ( y 0 , 3 δ r ) ) − 1 = cap ( x 0 , B ( y 0 , 3 δ r ) ) ≤ E ( 1 − M ( h − M ) + ) ≤ ( 1 − M ) p 1 E ( h ) = ( 1 − M ) p 1 ρ ( x 0 , y 0 ) p − 1 1 ,
which gives
R ( x 0 , B ( y 0 , 3 δ r ) ‾ ) 1 p − 1 ≥ ( 1 − M ) p p − 1 ρ ( x 0 , y 0 ) ≥ ( 1 − p p − 1 M ) ρ ( x 0 , y 0 ) . R\left(x_0,\overline{B(y_0,3\delta r)}\right)^{\frac{1}{p-1}}\ge(1-M)^{\frac{p}{p-1}}\rho(x_0,y_0)\ge\left(1-\frac{p}{p-1}M\right)\rho(x_0,y_0). R ( x 0 , B ( y 0 , 3 δ r ) ) p − 1 1 ≥ ( 1 − M ) p − 1 p ρ ( x 0 , y 0 ) ≥ ( 1 − p − 1 p M ) ρ ( x 0 , y 0 ) .
Combining this with (6.8) and the fact that R ( x 0 , B ( y 0 , 3 δ r ) ‾ ) ≤ R ( x 0 , y 0 ) = ρ ( x 0 , y 0 ) p − 1 R\left(x_0,\overline{B(y_0,3\delta r)}\right)\le R(x_0,y_0)=\rho(x_0,y_0)^{p-1} R ( x 0 , B ( y 0 , 3 δ r ) ) ≤ R ( x 0 , y 0 ) = ρ ( x 0 , y 0 ) p − 1 , we obtain (6.6). Similarly, we have (6.7).
Second, we give an upper bound for cap ( x 0 , z 0 ) \operatorname{cap}(x_0,z_0) cap ( x 0 , z 0 ) . Take f , g ∈ F e f,g\in\mathcal{F}_e f , g ∈ F e with 0 ≤ f , g ≤ 1 0\le f,g\le1 0 ≤ f , g ≤ 1 on X X X , f ( x 0 ) = g ( z 0 ) = 1 f(x_0)=g(z_0)=1 f ( x 0 ) = g ( z 0 ) = 1 and f = g = 0 f=g=0 f = g = 0 on B ( y 0 , 3 δ r ) ‾ \overline{B(y_0,3\delta r)} B ( y 0 , 3 δ r ) , such that
E ( f ) = cap ( x 0 , B ( y 0 , 3 δ r ) ‾ ) , E ( g ) = cap ( z 0 , B ( y 0 , 3 δ r ) ‾ ) . \mathcal{E}(f)=\operatorname{cap}\left(x_0,\overline{B(y_0,3\delta r)}\right),\qquad
\mathcal{E}(g)=\operatorname{cap}\left(z_0,\overline{B(y_0,3\delta r)}\right). E ( f ) = cap ( x 0 , B ( y 0 , 3 δ r ) ) , E ( g ) = cap ( z 0 , B ( y 0 , 3 δ r ) ) .
By [42 Theorem 2.3] , there exists φ ∈ F ∩ C c ( X ) \varphi\in\mathcal{F}\cap C_c(X) φ ∈ F ∩ C c ( X ) satisfying 0 ≤ φ ≤ 1 0\le\varphi\le1 0 ≤ φ ≤ 1 on X X X , φ = 1 \varphi=1 φ = 1 on B ( y 0 , 1 2 r ) ‾ \overline{B(y_0,\frac{1}{2}r)} B ( y 0 , 2 1 r ) , supp ( φ ) ⊆ B ( y 0 , L r ) \operatorname{supp}(\varphi)\subseteq B(y_0,Lr) supp ( φ ) ⊆ B ( y 0 , L r ) , and E ( φ ) ≤ C 2 ( L r ) β p − d h \mathcal{E}(\varphi)\le\frac{C_2}{(Lr)^{\beta_p-d_h}} E ( φ ) ≤ ( L r ) β p − d h C 2 , where C 2 > 0 C_2>0 C 2 > 0 is a constant. Then f ∧ φ , g ∧ φ ∈ F ∩ C c ( X ) f\wedge\varphi,g\wedge\varphi\in\mathcal{F}\cap C_c(X) f ∧ φ , g ∧ φ ∈ F ∩ C c ( X ) , supp ( f ∧ φ ) , supp ( g ∧ φ ) ⊆ B ( y 0 , L r ) ∖ B ( y 0 , 3 δ r ) ‾ \operatorname{supp}(f\wedge\varphi),\operatorname{supp}(g\wedge\varphi)\subseteq B(y_0,Lr)\setminus\overline{B(y_0,3\delta r)} supp ( f ∧ φ ) , supp ( g ∧ φ ) ⊆ B ( y 0 , L r ) ∖ B ( y 0 , 3 δ r ) , ( f ∧ φ ) ( x 0 ) = ( g ∧ φ ) ( z 0 ) = 1 (f\wedge\varphi)(x_0)=(g\wedge\varphi)(z_0)=1 ( f ∧ φ ) ( x 0 ) = ( g ∧ φ ) ( z 0 ) = 1 , f ∧ φ = g ∧ φ = 0 f\wedge\varphi=g\wedge\varphi=0 f ∧ φ = g ∧ φ = 0 on B ( y 0 , 3 δ r ) ‾ \overline{B(y_0,3\delta r)} B ( y 0 , 3 δ r ) . By SubAdd,
E ( f ∧ φ ) ≤ E ( f ) + E ( φ ) ≤ cap ( x 0 , B ( y 0 , 3 δ r ) ‾ ) + C 2 ( L r ) β p − d h , \mathcal{E}(f\wedge\varphi)\le\mathcal{E}(f)+\mathcal{E}(\varphi)\le\operatorname{cap}\left(x_0,\overline{B(y_0,3\delta r)}\right)+\frac{C_2}{(Lr)^{\beta_p-d_h}}, E ( f ∧ φ ) ≤ E ( f ) + E ( φ ) ≤ cap ( x 0 , B ( y 0 , 3 δ r ) ) + ( L r ) β p − d h C 2 ,
E ( g ∧ φ ) ≤ E ( g ) + E ( φ ) ≤ cap ( z 0 , B ( y 0 , 3 δ r ) ‾ ) + C 2 ( L r ) β p − d h . (36) \mathcal{E}(g\wedge\varphi)\le\mathcal{E}(g)+\mathcal{E}(\varphi)\le\operatorname{cap}\left(z_0,\overline{B(y_0,3\delta r)}\right)+\frac{C_2}{(Lr)^{\beta_p-d_h}}. \tag*{(36)} E ( g ∧ φ ) ≤ E ( g ) + E ( φ ) ≤ cap ( z 0 , B ( y 0 , 3 δ r ) ) + ( L r ) β p − d h C 2 . (36)
Let Ω 1 \Omega_1 Ω 1 and Ω 2 \Omega_2 Ω 2 be the connected components of B ( y 0 , L r ) ∖ B ( y 0 , 2 δ r ) ‾ B(y_0,Lr)\setminus\overline{B(y_0,2\delta r)} B ( y 0 , L r ) ∖ B ( y 0 , 2 δ r ) containing x 0 x_0 x 0 and z 0 z_0 z 0 , respectively. Since d d d is geodesic, and any θ ∈ Path ( x 0 , z 0 ; B ( y 0 , L r ) ) \theta\in\operatorname{Path}(x_0,z_0;B(y_0,Lr)) θ ∈ Path ( x 0 , z 0 ; B ( y 0 , L r )) satisfies θ ∩ B ( y 0 , 2 δ r ) ≠ ∅ \theta\cap B(y_0,2\delta r)\ne\varnothing θ ∩ B ( y 0 , 2 δ r ) = ∅ , we have Ω 1 ∩ Ω 2 = ∅ \Omega_1\cap\Omega_2=\varnothing Ω 1 ∩ Ω 2 = ∅ . Let F = 1 Ω 1 ( f ∧ φ ) F=1_{\Omega_1}(f\wedge\varphi) F = 1 Ω 1 ( f ∧ φ ) and G = 1 Ω 2 ( g ∧ φ ) G=1_{\Omega_2}(g\wedge\varphi) G = 1 Ω 2 ( g ∧ φ ) , then by the following Lemma 6.5, F , G ∈ F ∩ C c ( X ) F,G\in\mathcal{F}\cap C_c(X) F , G ∈ F ∩ C c ( X ) , d Γ ( F ) = 1 Ω 1 d Γ ( f ∧ φ ) d\Gamma(F)=1_{\Omega_1}d\Gamma(f\wedge\varphi) d Γ ( F ) = 1 Ω 1 d Γ ( f ∧ φ ) , and d Γ ( G ) = 1 Ω 2 d Γ ( g ∧ φ ) d\Gamma(G)=1_{\Omega_2}d\Gamma(g\wedge\varphi) d Γ ( G ) = 1 Ω 2 d Γ ( g ∧ φ ) .
For any t ∈ [ 0 , 1 ] t\in[0,1] t ∈ [ 0 , 1 ] , the function ( 1 − t ) F + t ( 1 − G ) ∈ F e (1-t)F+t(1-G)\in\mathcal{F}_e ( 1 − t ) F + t ( 1 − G ) ∈ F e takes the value 1 1 1 at x 0 x_0 x 0 and the value 0 0 0 at z 0 z_0 z 0 , by the strongly local property,
cap ( x 0 , z 0 ) ≤ E ( ( 1 − t ) F + t ( 1 − G ) ) = ( 1 − t ) p E ( F ) + t p E ( G ) ≤ ( 1 − t ) p E ( f ∧ φ ) + t p E ( g ∧ φ ) . \operatorname{cap}(x_0,z_0)\le\mathcal{E}((1-t)F+t(1-G))=(1-t)^p\mathcal{E}(F)+t^p\mathcal{E}(G)\le(1-t)^p\mathcal{E}(f\wedge\varphi)+t^p\mathcal{E}(g\wedge\varphi). cap ( x 0 , z 0 ) ≤ E (( 1 − t ) F + t ( 1 − G )) = ( 1 − t ) p E ( F ) + t p E ( G ) ≤ ( 1 − t ) p E ( f ∧ φ ) + t p E ( g ∧ φ ) .
Taking the infimum over t ∈ [ 0 , 1 ] t\in[0,1] t ∈ [ 0 , 1 ] and using (6.3), we obtain
cap ( x 0 , z 0 ) ≤ 1 ( 1 E ( f ∧ φ ) 1 p − 1 + 1 E ( g ∧ φ ) 1 p − 1 ) p − 1 . \operatorname{cap}(x_0,z_0)\le\frac{1}{\left(\frac{1}{\mathcal{E}(f\wedge\varphi)^{\frac{1}{p-1}}}+\frac{1}{\mathcal{E}(g\wedge\varphi)^{\frac{1}{p-1}}}\right)^{p-1}}. cap ( x 0 , z 0 ) ≤ ( E ( f ∧ φ ) p − 1 1 1 + E ( g ∧ φ ) p − 1 1 1 ) p − 1 1 .
Combining this with (6.6), (6.7), and (6.9),
ρ ( x 0 , z 0 ) = 1 cap ( x 0 , z 0 ) 1 p − 1 ≥ 1 ( cap ( x 0 , B ( y 0 , 3 δ r ) ) + C 2 ( L r ) β p − d h ) 1 / p + 1 ( cap ( z 0 , B ( y 0 , 3 δ r ) ) + C 2 ( L r ) β p − d h ) 1 / p ≥ ( 1 − C 1 δ β p − d h p ) ( ρ ( x 0 , y 0 ) + ρ ( z 0 , y 0 ) ) − C 3 r γ p L β p − d h , \begin{aligned}
\rho(x_0,z_0)=\frac{1}{\operatorname{cap}(x_0,z_0)^{\frac{1}{p-1}}} \\
&\ge\frac{1}{\left(\operatorname{cap}(x_0,B(y_0,3\delta r))+\frac{C_2}{(Lr)^{\beta_p-d_h}}\right)^{1/p}}+\frac{1}{\left(\operatorname{cap}(z_0,B(y_0,3\delta r))+\frac{C_2}{(Lr)^{\beta_p-d_h}}\right)^{1/p}} \\
&\ge\left(1-C_1\delta^{\frac{\beta_p-d_h}{p}}\right)\left(\rho(x_0,y_0)+\rho(z_0,y_0)\right)-C_3\frac{r^{\gamma_p}}{L^{\beta_p-d_h}},
\end{aligned} ρ ( x 0 , z 0 ) = cap ( x 0 , z 0 ) p − 1 1 1 ≥ ( cap ( x 0 , B ( y 0 , 3 δ r )) + ( L r ) β p − d h C 2 ) 1/ p 1 + ( cap ( z 0 , B ( y 0 , 3 δ r )) + ( L r ) β p − d h C 2 ) 1/ p 1 ≥ ( 1 − C 1 δ p β p − d h ) ( ρ ( x 0 , y 0 ) + ρ ( z 0 , y 0 ) ) − C 3 L β p − d h r γ p ,
where, in the last inequality, we used the fact that ρ ( x 0 , y 0 ) , ρ ( z 0 , y 0 ) ≍ r γ p \rho(x_0,y_0),\rho(z_0,y_0)\asymp r^{\gamma_p} ρ ( x 0 , y 0 ) , ρ ( z 0 , y 0 ) ≍ r γ p , and C 3 > 0 C_3>0 C 3 > 0 is a constant. Hence
Δ ( x 0 , y 0 , z 0 ; r ) = 1 r γ p ( ρ ( x 0 , y 0 ) + ρ ( z 0 , y 0 ) − ρ ( x 0 , z 0 ) ) ≲ δ β p − d h p + 1 L β p − d h . \Delta(x_0,y_0,z_0;r)=\frac{1}{r^{\gamma_p}}\left(\rho(x_0,y_0)+\rho(z_0,y_0)-\rho(x_0,z_0)\right)\lesssim\delta^{\frac{\beta_p-d_h}{p}}+\frac{1}{L^{\beta_p-d_h}}. Δ ( x 0 , y 0 , z 0 ; r ) = r γ p 1 ( ρ ( x 0 , y 0 ) + ρ ( z 0 , y 0 ) − ρ ( x 0 , z 0 ) ) ≲ δ p β p − d h + L β p − d h 1 .
Lemma 6.5. Let U U U be an open set and K ⊆ U K\subseteq U K ⊆ U a compact set. Let Ω \Omega Ω be a connected component of U ∖ K U\setminus K U ∖ K . Let u ∈ F ( p ) ∩ C c ( X ) u\in\mathcal{F}^{(p)}\cap C_c(X) u ∈ F ( p ) ∩ C c ( X ) satisfy supp ( u ) ⊆ U \operatorname{supp}(u)\subseteq U supp ( u ) ⊆ U and u = 0 u=0 u = 0 in an open neighborhood of K K K . Then 1 Ω u ∈ F ( p ) ∩ C c ( X ) \mathbb{1}_{\Omega}u\in\mathcal{F}^{(p)}\cap C_c(X) 1 Ω u ∈ F ( p ) ∩ C c ( X ) and d Γ ( p ) ( 1 Ω u ) = 1 Ω d Γ ( p ) ( u ) \mathrm{d}\Gamma^{(p)}(\mathbb{1}_{\Omega}u)=\mathbb{1}_{\Omega}\mathrm{d}\Gamma^{(p)}(u) d Γ ( p ) ( 1 Ω u ) = 1 Ω d Γ ( p ) ( u ) .
Proof. Since u ∈ C c ( X ) u\in C_c(X) u ∈ C c ( X ) , supp ( u ) ⊆ U \operatorname{supp}(u)\subseteq U supp ( u ) ⊆ U , and u = 0 u=0 u = 0 in an open neighborhood of K K K , supp ( u ) \operatorname{supp}(u) supp ( u ) is a compact subset of U ∖ K U\setminus K U ∖ K . Since Ω \Omega Ω is a connected component of U ∖ K U\setminus K U ∖ K , Ω \Omega Ω is both relatively open and relatively closed in U ∖ K U\setminus K U ∖ K . Hence there exists a closed set F F F such that Ω = F ∩ ( U ∖ K ) \Omega=F\cap(U\setminus K) Ω = F ∩ ( U ∖ K ) , which gives
supp ( u ) ∩ Ω = supp ( u ) ∩ F ∩ ( U ∖ K ) = supp ( u ) ∩ F , \operatorname{supp}(u)\cap\Omega=\operatorname{supp}(u)\cap F\cap(U\setminus K)=\operatorname{supp}(u)\cap F, supp ( u ) ∩ Ω = supp ( u ) ∩ F ∩ ( U ∖ K ) = supp ( u ) ∩ F ,
where we used supp ( u ) ⊆ U ∖ K \operatorname{supp}(u)\subseteq U\setminus K supp ( u ) ⊆ U ∖ K in the second equality. Hence supp ( u ) ∩ Ω \operatorname{supp}(u)\cap\Omega supp ( u ) ∩ Ω is compact. Since U ∖ K U\setminus K U ∖ K is open, Ω \Omega Ω is open. Therefore, supp ( u ) ∩ Ω \operatorname{supp}(u)\cap\Omega supp ( u ) ∩ Ω is a compact subset of the open set Ω \Omega Ω . By the regular property of ( E ( p ) , F ( p ) ) (\mathcal{E}^{(p)},\mathcal{F}^{(p)}) ( E ( p ) , F ( p ) ) , there exists ϕ ∈ F ( p ) ∩ C c ( X ) \phi\in\mathcal{F}^{(p)}\cap C_c(X) ϕ ∈ F ( p ) ∩ C c ( X ) such that 0 ≤ ϕ ≤ 1 0\leq\phi\leq1 0 ≤ ϕ ≤ 1 on X X X , ϕ = 1 \phi=1 ϕ = 1 on supp ( u ) ∩ Ω \operatorname{supp}(u)\cap\Omega supp ( u ) ∩ Ω , and supp ( ϕ ) ⊆ Ω \operatorname{supp}(\phi)\subseteq\Omega supp ( ϕ ) ⊆ Ω . By Alg, 1 Ω u = ϕ u ∈ F ( p ) ∩ C c ( X ) \mathbb{1}_{\Omega}u=\phi u\in\mathcal{F}^{(p)}\cap C_c(X) 1 Ω u = ϕ u ∈ F ( p ) ∩ C c ( X ) . By the strongly local property, d Γ ( p ) ( 1 Ω u ) = 1 Ω d Γ ( p ) ( u ) \mathrm{d}\Gamma^{(p)}(\mathbb{1}_{\Omega}u)=\mathbb{1}_{\Omega}\mathrm{d}\Gamma^{(p)}(u) d Γ ( p ) ( 1 Ω u ) = 1 Ω d Γ ( p ) ( u ) .
Second, we prove Lemma 6.3 by contradiction.
Proof of Lemma 6.3. For notational convenience, we omit the superscript ( p ) (p) ( p ) when no ambiguity can arise. Since γ p = 1 \gamma_p=1 γ p = 1 , R ( x , y ) ≍ d ( x , y ) β p − d h = d ( x , y ) p − 1 R(x,y)\asymp d(x,y)^{\beta_p-d_h}=d(x,y)^{p-1} R ( x , y ) ≍ d ( x , y ) β p − d h = d ( x , y ) p − 1 and ρ ( x , y ) ≍ d ( x , y ) \rho(x,y)\asymp d(x,y) ρ ( x , y ) ≍ d ( x , y ) . Suppose that this result does not hold, then there exist δ , L > 0 \delta,L>0 δ , L > 0 , a sequence { r n } ⊆ ( 0 , + ∞ ) \{r_n\}\subseteq(0,+\infty) { r n } ⊆ ( 0 , + ∞ ) , and a sequence of triples { ( x n , y n , z n ) } \{(x_n,y_n,z_n)\} {( x n , y n , z n )} such that, for any n n n , ( x n , y n , z n ) (x_n,y_n,z_n) ( x n , y n , z n ) is balanced at scale r n r_n r n and there exists θ n ∈ Path ( x n , z n ; B ( y n , L r n ) ) \theta_n\in\operatorname{Path}(x_n,z_n;B(y_n,Lr_n)) θ n ∈ Path ( x n , z n ; B ( y n , L r n )) satisfying θ n ∩ B ( y n , δ r n ) = ∅ \theta_n\cap B(y_n,\delta r_n)=\varnothing θ n ∩ B ( y n , δ r n ) = ∅ , whereas Δ n : = Δ ( x n , y n , z n ; r n ) → 0 \Delta_n:=\Delta(x_n,y_n,z_n;r_n)\to0 Δ n := Δ ( x n , y n , z n ; r n ) → 0 .
For any n n n , let u n ∈ F e u_n\in\mathcal{F}_e u n ∈ F e be the equilibrium potential between x n x_n x n and z n z_n z n with u n ( x n ) = 1 u_n(x_n)=1 u n ( x n ) = 1 and u n ( z n ) = 0 u_n(z_n)=0 u n ( z n ) = 0 , then cap ( x n , z n ) = E ( u n ) \operatorname{cap}(x_n,z_n)=\mathcal{E}(u_n) cap ( x n , z n ) = E ( u n ) . Let λ n : = u n ( y n ) ∈ [ 0 , 1 ] \lambda_n:=u_n(y_n)\in[0,1] λ n := u n ( y n ) ∈ [ 0 , 1 ] . If λ n ∈ ( 0 , 1 ) \lambda_n\in(0,1) λ n ∈ ( 0 , 1 ) , let
v n : = ( u n − λ n ) + 1 − λ n , w n : = ( λ n − u n ) + λ n , v_n:=\frac{(u_n-\lambda_n)_+}{1-\lambda_n},\qquad w_n:=\frac{(\lambda_n-u_n)_+}{\lambda_n}, v n := 1 − λ n ( u n − λ n ) + , w n := λ n ( λ n − u n ) + ,
then v n , w n ∈ F e v_n,w_n\in\mathcal{F}_e v n , w n ∈ F e satisfy v n ( x n ) = w n ( z n ) = 1 v_n(x_n)=w_n(z_n)=1 v n ( x n ) = w n ( z n ) = 1 and v n ( y n ) = w n ( y n ) = 0 v_n(y_n)=w_n(y_n)=0 v n ( y n ) = w n ( y n ) = 0 , hence
cap ( x n , y n ) ≤ E ( v n ) = 1 ( 1 − λ n ) p E ( ( u n − λ n ) + ) , \operatorname{cap}(x_n,y_n)\leq\mathcal{E}(v_n)=\frac{1}{(1-\lambda_n)^p}\mathcal{E}((u_n-\lambda_n)_+), cap ( x n , y n ) ≤ E ( v n ) = ( 1 − λ n ) p 1 E (( u n − λ n ) + ) ,
cap ( z n , y n ) ≤ E ( w n ) = 1 λ n p E ( ( λ n − u n ) + ) . (37) \operatorname{cap}(z_n,y_n)\leq\mathcal{E}(w_n)=\frac{1}{\lambda_n^p}\mathcal{E}((\lambda_n-u_n)_+). \tag*{(37)} cap ( z n , y n ) ≤ E ( w n ) = λ n p 1 E (( λ n − u n ) + ) . (37)
By the strongly local property,
cap ( x n , z n ) = E ( u n ) = E ( ( u n − λ n ) + ) + E ( ( λ n − u n ) + ) ≥ ( 1 − λ n ) p cap ( x n , y n ) + λ n p cap ( z n , y n ) , \operatorname{cap}(x_n,z_n)=\mathcal{E}(u_n)=\mathcal{E}((u_n-\lambda_n)_+)+\mathcal{E}((\lambda_n-u_n)_+)\geq(1-\lambda_n)^p\operatorname{cap}(x_n,y_n)+\lambda_n^p\operatorname{cap}(z_n,y_n), cap ( x n , z n ) = E ( u n ) = E (( u n − λ n ) + ) + E (( λ n − u n ) + ) ≥ ( 1 − λ n ) p cap ( x n , y n ) + λ n p cap ( z n , y n ) ,
which gives
1 ρ ( x n , z n ) p − 1 ≥ ( 1 − λ n ) p ρ ( x n , y n ) p − 1 + λ n p ρ ( z n , y n ) p − 1 , \frac{1}{\rho(x_n,z_n)^{p-1}}\geq\frac{(1-\lambda_n)^p}{\rho(x_n,y_n)^{p-1}}+\frac{\lambda_n^p}{\rho(z_n,y_n)^{p-1}}, ρ ( x n , z n ) p − 1 1 ≥ ρ ( x n , y n ) p − 1 ( 1 − λ n ) p + ρ ( z n , y n ) p − 1 λ n p ,
that is,
1 ≥ ( 1 − λ n ) p ( ρ ( x n , z n ) ρ ( x n , y n ) ) p − 1 + λ n p ( ρ ( x n , z n ) ρ ( z n , y n ) ) p − 1 . (38) 1\geq(1-\lambda_n)^p\left(\frac{\rho(x_n,z_n)}{\rho(x_n,y_n)}\right)^{p-1}+\lambda_n^p\left(\frac{\rho(x_n,z_n)}{\rho(z_n,y_n)}\right)^{p-1}. \tag*{(38)} 1 ≥ ( 1 − λ n ) p ( ρ ( x n , y n ) ρ ( x n , z n ) ) p − 1 + λ n p ( ρ ( z n , y n ) ρ ( x n , z n ) ) p − 1 . (38)
If λ n = u n ( y n ) = 0 \lambda_n=u_n(y_n)=0 λ n = u n ( y n ) = 0 , then u n ( x n ) = 1 u_n(x_n)=1 u n ( x n ) = 1 gives cap ( x n , y n ) ≤ E ( u n ) = cap ( x n , z n ) \operatorname{cap}(x_n,y_n)\leq\mathcal{E}(u_n)=\operatorname{cap}(x_n,z_n) cap ( x n , y n ) ≤ E ( u n ) = cap ( x n , z n ) , that is, ρ ( x n , z n ) ρ ( x n , y n ) ≤ 1 \frac{\rho(x_n,z_n)}{\rho(x_n,y_n)}\leq1 ρ ( x n , y n ) ρ ( x n , z n ) ≤ 1 . Hence (6.11) still holds. Similarly, if λ n = u n ( y n ) = 1 \lambda_n=u_n(y_n)=1 λ n = u n ( y n ) = 1 , then (6.11) also holds. By (30) ,
( 1 − λ n ) p ( ρ ( x n , z n ) ρ ( x n , y n ) ) p − 1 + λ n p ( ρ ( x n , z n ) ρ ( z n , y n ) ) p − 1 ≥ min { ( 1 − λ ) p ( ρ ( x n , z n ) ρ ( x n , y n ) ) p − 1 + λ p ( ρ ( x n , z n ) ρ ( z n , y n ) ) p − 1 : λ ∈ [ 0 , 1 ] } = 1 ( ρ ( x n , y n ) ρ ( x n , z n ) + ρ ( z n , y n ) ρ ( x n , z n ) ) p − 1 = ( ρ ( x n , z n ) ρ ( x n , y n ) + ρ ( z n , y n ) ) p − 1 , (39) \begin{aligned}
(1-\lambda_n)^p\left(\frac{\rho(x_n,z_n)}{\rho(x_n,y_n)}\right)^{p-1}
+\lambda_n^p\left(\frac{\rho(x_n,z_n)}{\rho(z_n,y_n)}\right)^{p-1}
&\ge\min\left\{(1-\lambda)^p\left(\frac{\rho(x_n,z_n)}{\rho(x_n,y_n)}\right)^{p-1}
+\lambda^p\left(\frac{\rho(x_n,z_n)}{\rho(z_n,y_n)}\right)^{p-1}:\lambda\in[0,1]\right\}\\
&=\frac{1}{\left(\frac{\rho(x_n,y_n)}{\rho(x_n,z_n)}+\frac{\rho(z_n,y_n)}{\rho(x_n,z_n)}\right)^{p-1}}
=\left(\frac{\rho(x_n,z_n)}{\rho(x_n,y_n)+\rho(z_n,y_n)}\right)^{p-1}, \tag*{(39)}
\end{aligned} ( 1 − λ n ) p ( ρ ( x n , y n ) ρ ( x n , z n ) ) p − 1 + λ n p ( ρ ( z n , y n ) ρ ( x n , z n ) ) p − 1 ≥ min { ( 1 − λ ) p ( ρ ( x n , y n ) ρ ( x n , z n ) ) p − 1 + λ p ( ρ ( z n , y n ) ρ ( x n , z n ) ) p − 1 : λ ∈ [ 0 , 1 ] } = ( ρ ( x n , z n ) ρ ( x n , y n ) + ρ ( x n , z n ) ρ ( z n , y n ) ) p − 1 1 = ( ρ ( x n , y n ) + ρ ( z n , y n ) ρ ( x n , z n ) ) p − 1 , (39)
where the minimum is attained uniquely at λ = 1 ρ ( x n , y n ) 1 ρ ( x n , y n ) + 1 ρ ( z n , y n ) \lambda=\frac{\frac{1}{\rho(x_n,y_n)}}{\frac{1}{\rho(x_n,y_n)}+\frac{1}{\rho(z_n,y_n)}} λ = ρ ( x n , y n ) 1 + ρ ( z n , y n ) 1 ρ ( x n , y n ) 1 . Recall that
0 ≤ Δ n = 1 r n ( ρ ( x n , y n ) + ρ ( z n , y n ) − ρ ( x n , z n ) ) → 0. 0\le\Delta_n=\frac{1}{r_n}\bigl(\rho(x_n,y_n)+\rho(z_n,y_n)-\rho(x_n,z_n)\bigr)\to0. 0 ≤ Δ n = r n 1 ( ρ ( x n , y n ) + ρ ( z n , y n ) − ρ ( x n , z n ) ) → 0.
Combining this with (38) and (39) ,
1 ≥ ( ρ ( x n , z n ) ρ ( x n , y n ) + ρ ( z n , y n ) ) p − 1 = ( 1 − r n ρ ( x n , y n ) + ρ ( z n , y n ) Δ n ) p − 1 . (40) 1\ge\left(\frac{\rho(x_n,z_n)}{\rho(x_n,y_n)+\rho(z_n,y_n)}\right)^{p-1}
=\left(1-\frac{r_n}{\rho(x_n,y_n)+\rho(z_n,y_n)}\Delta_n\right)^{p-1}. \tag*{(40)} 1 ≥ ( ρ ( x n , y n ) + ρ ( z n , y n ) ρ ( x n , z n ) ) p − 1 = ( 1 − ρ ( x n , y n ) + ρ ( z n , y n ) r n Δ n ) p − 1 . (40)
Since ρ ( x n , y n ) ≍ d ( x n , y n ) ≍ r n \rho(x_n,y_n)\asymp d(x_n,y_n)\asymp r_n ρ ( x n , y n ) ≍ d ( x n , y n ) ≍ r n and ρ ( z n , y n ) ≍ d ( z n , y n ) ≍ r n \rho(z_n,y_n)\asymp d(z_n,y_n)\asymp r_n ρ ( z n , y n ) ≍ d ( z n , y n ) ≍ r n , r n ρ ( x n , y n ) + ρ ( z n , y n ) ≍ 1 \frac{r_n}{\rho(x_n,y_n)+\rho(z_n,y_n)}\asymp1 ρ ( x n , y n ) + ρ ( z n , y n ) r n ≍ 1 . Since Δ n → 0 \Delta_n\to0 Δ n → 0 , the RHS of (40) converges to 1. This has the following important consequences: First, by the continuity with respect to λ \lambda λ of the function appearing in (39) ,
λ n − 1 ρ ( x n , y n ) 1 ρ ( x n , y n ) + 1 ρ ( z n , y n ) → 0 , \lambda_n-\frac{\frac{1}{\rho(x_n,y_n)}}{\frac{1}{\rho(x_n,y_n)}+\frac{1}{\rho(z_n,y_n)}}\to0, λ n − ρ ( x n , y n ) 1 + ρ ( z n , y n ) 1 ρ ( x n , y n ) 1 → 0 ,
where the sequence { 1 ρ ( x n , y n ) 1 ρ ( x n , y n ) + 1 ρ ( z n , y n ) } ⊂ [ 0 , 1 ] \left\{\frac{\frac{1}{\rho(x_n,y_n)}}{\frac{1}{\rho(x_n,y_n)}+\frac{1}{\rho(z_n,y_n)}}\right\}\subset[0,1] { ρ ( x n , y n ) 1 + ρ ( z n , y n ) 1 ρ ( x n , y n ) 1 } ⊂ [ 0 , 1 ] is uniformly bounded away from 0 and 1, hence { λ n } ⊂ [ 0 , 1 ] \{\lambda_n\}\subset[0,1] { λ n } ⊂ [ 0 , 1 ] is also uniformly bounded away from 0 and 1. Second, the RHS of (38) converges to 1. Combining this with (37) and the fact that both { 1 − λ n } \{1-\lambda_n\} { 1 − λ n } and { λ n } \{\lambda_n\} { λ n } are uniformly bounded away from 0, we obtain
1 ≤ E ( v n ) cap ( x n , y n ) → 1 , 1 ≤ E ( w n ) cap ( z n , y n ) → 1. 1\le\frac{\mathcal{E}(v_n)}{\operatorname{cap}(x_n,y_n)}\to1,\qquad
1\le\frac{\mathcal{E}(w_n)}{\operatorname{cap}(z_n,y_n)}\to1. 1 ≤ cap ( x n , y n ) E ( v n ) → 1 , 1 ≤ cap ( z n , y n ) E ( w n ) → 1.
Since cap ( x n , y n ) ≍ d ( x n , y n ) 1 − p ≍ r n 1 − p \operatorname{cap}(x_n,y_n)\asymp d(x_n,y_n)^{1-p}\asymp r_n^{1-p} cap ( x n , y n ) ≍ d ( x n , y n ) 1 − p ≍ r n 1 − p and cap ( z n , y n ) ≍ d ( z n , y n ) 1 − p ≍ r n 1 − p \operatorname{cap}(z_n,y_n)\asymp d(z_n,y_n)^{1-p}\asymp r_n^{1-p} cap ( z n , y n ) ≍ d ( z n , y n ) 1 − p ≍ r n 1 − p ,
E ( v n ) = cap ( x n , y n ) + o ( r n 1 − p ) , E ( w n ) = cap ( z n , y n ) + o ( r n 1 − p ) . (41) \mathcal{E}(v_n)=\operatorname{cap}(x_n,y_n)+o(r_n^{1-p}),\qquad
\mathcal{E}(w_n)=\operatorname{cap}(z_n,y_n)+o(r_n^{1-p}). \tag*{(41)} E ( v n ) = cap ( x n , y n ) + o ( r n 1 − p ) , E ( w n ) = cap ( z n , y n ) + o ( r n 1 − p ) . (41)
Let V n ∈ F e V_n\in\mathcal{F}_e V n ∈ F e be the equilibrium potential between x n x_n x n and y n y_n y n with V n ( x n ) = 1 V_n(x_n)=1 V n ( x n ) = 1 and V n ( y n ) = 0 V_n(y_n)=0 V n ( y n ) = 0 , then cap ( x n , y n ) = E ( V n ) \operatorname{cap}(x_n,y_n)=\mathcal{E}(V_n) cap ( x n , y n ) = E ( V n ) . We claim that
E ( v n − V n ) = o ( r n 1 − p ) . (42) \mathcal{E}(v_n-V_n)=o(r_n^{1-p}). \tag*{(42)} E ( v n − V n ) = o ( r n 1 − p ) . (42)
Indeed, if p ∈ ( 1 , 2 ] p\in(1,2] p ∈ ( 1 , 2 ] , then by Cla,
E ( v n ) + cap ( x n , y n ) = E ( v n ) + E ( V n ) \mathcal{E}(v_n)+\operatorname{cap}(x_n,y_n)=\mathcal{E}(v_n)+\mathcal{E}(V_n) E ( v n ) + cap ( x n , y n ) = E ( v n ) + E ( V n )
≥ 2 ( E ( v n + V n 2 ) 1 p − 1 + E ( v n − V n 2 ) 1 p − 1 ) p − 1 \ge2\left(\mathcal{E}\left(\frac{v_n+V_n}{2}\right)^{\frac{1}{p-1}}+\mathcal{E}\left(\frac{v_n-V_n}{2}\right)^{\frac{1}{p-1}}\right)^{p-1} ≥ 2 ( E ( 2 v n + V n ) p − 1 1 + E ( 2 v n − V n ) p − 1 1 ) p − 1
≥ 2 ( cap ( x n , y n ) 1 p − 1 + E ( v n − V n 2 ) 1 p − 1 ) p − 1 , \ge2\left(\operatorname{cap}(x_n,y_n)^{\frac{1}{p-1}}+\mathcal{E}\left(\frac{v_n-V_n}{2}\right)^{\frac{1}{p-1}}\right)^{p-1}, ≥ 2 ( cap ( x n , y n ) p − 1 1 + E ( 2 v n − V n ) p − 1 1 ) p − 1 ,
where the last inequality follows from the fact that v n + V n 2 ∈ F e \frac{v_n+V_n}{2}\in\mathcal{F}_e 2 v n + V n ∈ F e takes the value 1 at x n x_n x n and the value 0 at y n y_n y n . By (41) , we obtain (42) . Similarly, if p ∈ [ 2 , + ∞ ) p\in[2,+\infty) p ∈ [ 2 , + ∞ ) , then Cla and (41) also give (42) .
Finally, we consider V n ( z n ) V_n(z_n) V n ( z n ) and derive a contradiction. On the one hand, by (42) ,
∣ V n ( z n ) ∣ = ∣ ( v n − V n ) ( z n ) ∣ = ∣ ( v n − V n ) ( z n ) − ( v n − V n ) ( y n ) ∣ ≤ ( R ( z n , y n ) E ( v n − V n ) ) 1 p ≲ ( d ( z n , y n ) p − 1 E ( v n − V n ) ) 1 p ≲ ( r n p − 1 E ( v n − V n ) ) 1 p = ( r n p − 1 o ( r n 1 − p ) ) 1 p = o ( 1 ) , \begin{aligned}
|V_n(z_n)| &= |(v_n-V_n)(z_n)| = |(v_n-V_n)(z_n)-(v_n-V_n)(y_n)| \le\left(R(z_n,y_n)\mathcal{E}(v_n-V_n)\right)^{\frac{1}{p}} \\
&\lesssim\left(d(z_n,y_n)^{p-1}\mathcal{E}(v_n-V_n)\right)^{\frac{1}{p}} \lesssim\left(r_n^{p-1}\mathcal{E}(v_n-V_n)\right)^{\frac{1}{p}} = \left(r_n^{p-1}o(r_n^{1-p})\right)^{\frac{1}{p}} = o(1),
\end{aligned} ∣ V n ( z n ) ∣ = ∣ ( v n − V n ) ( z n ) ∣ = ∣ ( v n − V n ) ( z n ) − ( v n − V n ) ( y n ) ∣ ≤ ( R ( z n , y n ) E ( v n − V n ) ) p 1 ≲ ( d ( z n , y n ) p − 1 E ( v n − V n ) ) p 1 ≲ ( r n p − 1 E ( v n − V n ) ) p 1 = ( r n p − 1 o ( r n 1 − p ) ) p 1 = o ( 1 ) ,
hence V n ( z n ) → 0 V_n(z_n)\to0 V n ( z n ) → 0 . On the other hand, recall that θ n ∈ Path ( x n , z n ; B ( y n , L r n ) ) \theta_n\in\operatorname{Path}(x_n,z_n;B(y_n,Lr_n)) θ n ∈ Path ( x n , z n ; B ( y n , L r n )) satisfies θ n ∩ B ( y n , δ r n ) = ∅ \theta_n\cap B(y_n,\delta r_n)=\varnothing θ n ∩ B ( y n , δ r n ) = ∅ . Let ε ∈ ( 0 , δ 2 ∧ 1 4 ) \varepsilon\in(0,\frac{\delta}{2}\wedge\frac{1}{4}) ε ∈ ( 0 , 2 δ ∧ 4 1 ) be chosen later, then for any x ∈ B ( x n , ε r n ) x\in B(x_n,\varepsilon r_n) x ∈ B ( x n , ε r n ) ,
∣ V n ( x ) − 1 ∣ = ∣ V n ( x ) − V n ( x n ) ∣ ≤ ( R ( x , x n ) E ( V n ) ) 1 p ≲ ( d ( x , x n ) p − 1 cap ( x n , y n ) 1 − p ) 1 p ≲ ( ( ε r n ) p − 1 r n 1 − p ) 1 p = ε p − 1 p . \begin{aligned}
|V_n(x)-1| &= |V_n(x)-V_n(x_n)| \le\left(R(x,x_n)\mathcal{E}(V_n)\right)^{\frac{1}{p}} \\
&\lesssim\left(d(x,x_n)^{p-1}\operatorname{cap}(x_n,y_n)^{1-p}\right)^{\frac{1}{p}} \lesssim\left((\varepsilon r_n)^{p-1}r_n^{1-p}\right)^{\frac{1}{p}} = \varepsilon^{\frac{p-1}{p}}.
\end{aligned} ∣ V n ( x ) − 1∣ = ∣ V n ( x ) − V n ( x n ) ∣ ≤ ( R ( x , x n ) E ( V n ) ) p 1 ≲ ( d ( x , x n ) p − 1 cap ( x n , y n ) 1 − p ) p 1 ≲ ( ( ε r n ) p − 1 r n 1 − p ) p 1 = ε p p − 1 .
Hence there exists a constant ε ≪ 1 \varepsilon\ll1 ε ≪ 1 , such that ∣ V n ( x ) − 1 ∣ ≤ 1 2 |V_n(x)-1|\le\frac{1}{2} ∣ V n ( x ) − 1∣ ≤ 2 1 for any x ∈ B ( x n , ε r n ) x\in B(x_n,\varepsilon r_n) x ∈ B ( x n , ε r n ) , which gives V n ≥ 1 2 V_n\ge\frac{1}{2} V n ≥ 2 1 on B ( x n , ε r n ) B(x_n,\varepsilon r_n) B ( x n , ε r n ) ; in particular, let t n : = sup { t ∈ [ 0 , 1 ] : θ n ( t ) ∈ ∂ B ( x n , ε r n ) } t_n:=\sup\{t\in[0,1]:\theta_n(t)\in\partial B(x_n,\varepsilon r_n)\} t n := sup { t ∈ [ 0 , 1 ] : θ n ( t ) ∈ ∂ B ( x n , ε r n )} and x ~ n : = θ n ( t n ) \widetilde{x}_n:=\theta_n(t_n) x n := θ n ( t n ) , then t n < 1 t_n<1 t n < 1 , { x ~ n } = ( θ n ∣ [ t n , 1 ] ) ∩ ∂ B ( x n , ε r n ) \{\widetilde{x}_n\}=(\theta_n|_{[t_n,1]})\cap\partial B(x_n,\varepsilon r_n) { x n } = ( θ n ∣ [ t n , 1 ] ) ∩ ∂ B ( x n , ε r n ) , and V n ( x ~ n ) ≥ 1 2 V_n(\widetilde{x}_n)\ge\frac{1}{2} V n ( x n ) ≥ 2 1 ; see Figure 4 .
Figure 4. The path θ n \theta_n θ n avoiding B ( y n , δ r n ) B(y_n,\delta r_n) B ( y n , δ r n )
Since V n V_n V n is harmonic in X ∖ { x n , y n } X\setminus\{x_n,y_n\} X ∖ { x n , y n } , we may apply the elliptic Harnack inequality along a suitable Harnack chain following θ n ∣ [ t n , 1 ] \theta_n|_{[t_n,1]} θ n ∣ [ t n , 1 ] ; see [13 Definition 2.10] . Indeed, since θ n ∣ [ t n , 1 ] \theta_n|_{[t_n,1]} θ n ∣ [ t n , 1 ] has distance at least ε r n \varepsilon r_n ε r n from x n x_n x n and at least δ r n \delta r_n δ r n from y n y_n y n , there exists a > 0 a>0 a > 0 , independent of n n n , sufficiently small such that the Harnack dilates of balls of radius a r n ar_n a r n centered on θ n ∣ [ t n , 1 ] \theta_n|_{[t_n,1]} θ n ∣ [ t n , 1 ] avoid { x n , y n } \{x_n,y_n\} { x n , y n } . Choose a maximal ( 1 2 a r n ) (\frac{1}{2}ar_n) ( 2 1 a r n ) -separated subset of θ n ∣ [ t n , 1 ] \theta_n|_{[t_n,1]} θ n ∣ [ t n , 1 ] . The balls of radius a r n ar_n a r n centered at these points cover θ n ∣ [ t n , 1 ] \theta_n|_{[t_n,1]} θ n ∣ [ t n , 1 ] . Since θ n ∣ [ t n , 1 ] \theta_n|_{[t_n,1]} θ n ∣ [ t n , 1 ] is connected, there exists a chain of overlapping balls in this cover joining x ~ n \widetilde{x}_n x n to z n z_n z n without repeating balls. By VD and the fact that θ n ∣ [ t n , 1 ] ⊂ B ( y n , L r n ) \theta_n|_{[t_n,1]}\subset B(y_n,Lr_n) θ n ∣ [ t n , 1 ] ⊂ B ( y n , L r n ) , the number of balls in the covering, and hence the length of the chain, is bounded independently of n n n . Applying the elliptic Harnack inequality along this chain and using V n ( x ~ n ) ≥ 1 2 V_n(\widetilde{x}_n)\ge\frac{1}{2} V n ( x n ) ≥ 2 1 , we obtain V n ( z n ) ≥ c > 0 V_n(z_n)\ge c>0 V n ( z n ) ≥ c > 0 , where c > 0 c>0 c > 0 is a constant independent of n n n . This contradicts V n ( z n ) → 0 V_n(z_n)\to0 V n ( z n ) → 0 .
Proof of Theorem 2.5(3)
By (25) , we only need to prove the following result.
Proposition 7.1. Let p , q ∈ I p,q\in I p , q ∈ I satisfy p < q p<q p < q and γ p ∈ ( 0 , 1 ) \gamma_p\in(0,1) γ p ∈ ( 0 , 1 ) . Then γ p < γ q \gamma_p<\gamma_q γ p < γ q .
By Theorem 2.5 (1), γ p ≤ γ q \gamma_p\le\gamma_q γ p ≤ γ q . Suppose that
c : = γ p = γ q ∈ ( 0 , 1 ) . (43) c:=\gamma_p=\gamma_q\in(0,1). \tag*{(43)} c := γ p = γ q ∈ ( 0 , 1 ) . (43)
We will show that this assumption leads to a contradiction. For notational convenience, let
ρ : = ρ ( p ) = ( R ( p ) ) 1 p − 1 \rho:= \rho^{(p)} = \left(R^{(p)}\right)^{\frac{1}{p-1}} ρ := ρ ( p ) = ( R ( p ) ) p − 1 1
be the metric given by (12) associated with ( E ( p ) , F ( p ) ) \left(\mathcal{E}^{(p)},\mathcal{F}^{(p)}\right) ( E ( p ) , F ( p ) ) , and write B ρ ( x , r ) B_{\rho}(x,r) B ρ ( x , r ) and H ρ 1 \mathcal{H}_{\rho}^{1} H ρ 1 for the ball and the 1-dimensional Hausdorff measure on ( X , ρ ) (X,\rho) ( X , ρ ) .
The intuition behind the proof is as follows. First, we construct a suitable function u u u , which is in fact an E ( p ) \mathcal{E}^{(p)} E ( p ) -capacitary potential, such that its energy measure Γ ( p ) ( u ) \Gamma^{(p)}(u) Γ ( p ) ( u ) admits “good” upper and lower estimates; see Definition 7.2 and Lemma 7.3 . In particular, the equality (43) plays a key role in obtaining these lower estimates. Second, using the cutoff functions and the partition of unity from Lemma 7.4 , we employ the measure Γ ( p ) ( u ) \Gamma^{(p)}(u) Γ ( p ) ( u ) to construct a finite graph. We then apply the discrete potential theory from Proposition 7.7, in particular, some estimates for flows, potentials, and transportation costs, to obtain the existence of suitable paths in this graph. This enables us to show that, for any sufficiently small τ ≪ 1 \tau\ll1 τ ≪ 1 , there exist r ≪ 1 r \ll1 r ≪ 1 and { x i k : k = 0 , … , m } ⊆ X \{x_{i_k}: k=0,\ldots,m\}\subseteq X { x i k : k = 0 , … , m } ⊆ X such that
ρ ( x i 0 , x i m ) ≳ r , ∑ k = 1 m ρ ( x i k − 1 , x i k ) ≲ r , \rho(x_{i_0},x_{i_m}) \gtrsim r,\qquad\sum_{k=1}^{m}\rho(x_{i_{k-1}},x_{i_k}) \lesssim r, ρ ( x i 0 , x i m ) ≳ r , k = 1 ∑ m ρ ( x i k − 1 , x i k ) ≲ r ,
and
ρ ( x i k − 1 , x i k ) ≲ τ r for any k = 1 , … , m . \rho(x_{i_{k-1}},x_{i_k}) \lesssim\tau r\quad\text{for any }k=1,\ldots,m. ρ ( x i k − 1 , x i k ) ≲ τ r for any k = 1 , … , m .
Then applying (13) and the triangle inequality for the metric d d d ,
r ≲ ρ ( x i 0 , x i m ) ≃ d ( x i 0 , x i m ) c ≤ ( ∑ k = 1 m d ( x i k − 1 , x i k ) ) c ≃ ( ∑ k = 1 m ρ ( x i k − 1 , x i k ) 1 c ) c ≤ ( max k = 1 , … , m ρ ( x i k − 1 , x i k ) 1 c − 1 ∑ k = 1 m ρ ( x i k − 1 , x i k ) ) c ≲ ( ( τ r ) 1 c − 1 r ) c = τ 1 − c r , \begin{aligned}
r &\lesssim\rho(x_{i_0},x_{i_m}) \simeq d(x_{i_0},x_{i_m})^{c}
\leq\left(\sum_{k=1}^{m}d(x_{i_{k-1}},x_{i_k})\right)^{c}
\simeq\left(\sum_{k=1}^{m}\rho(x_{i_{k-1}},x_{i_k})^{\frac{1}{c}}\right)^{c}\\
&\leq\left(\max_{k=1,\ldots,m}\rho(x_{i_{k-1}},x_{i_k})^{\frac{1}{c}-1}\sum_{k=1}^{m}\rho(x_{i_{k-1}},x_{i_k})\right)^{c}
\lesssim\left((\tau r)^{\frac{1}{c}-1}r\right)^{c}
=\tau^{1-c}r,
\end{aligned} r ≲ ρ ( x i 0 , x i m ) ≃ d ( x i 0 , x i m ) c ≤ ( k = 1 ∑ m d ( x i k − 1 , x i k ) ) c ≃ ( k = 1 ∑ m ρ ( x i k − 1 , x i k ) c 1 ) c ≤ ( k = 1 , … , m max ρ ( x i k − 1 , x i k ) c 1 − 1 k = 1 ∑ m ρ ( x i k − 1 , x i k ) ) c ≲ ( ( τ r ) c 1 − 1 r ) c = τ 1 − c r ,
hence 1 ≲ τ 1 − c 1\lesssim\tau^{1-c} 1 ≲ τ 1 − c . Since c ∈ ( 0 , 1 ) c\in(0,1) c ∈ ( 0 , 1 ) , letting τ ↓ 0 \tau\downarrow0 τ ↓ 0 , we obtain a contradiction.
Definition 7.2. Fix a ball B = B ( x 0 , R ) B=B(x_{0},R) B = B ( x 0 , R ) , let u ∈ F ( p ) ⊆ C ( X ) u\in\mathcal{F}^{(p)}\subseteq C(X) u ∈ F ( p ) ⊆ C ( X ) be the E ( p ) \mathcal{E}^{(p)} E ( p ) -capacitary potential for ( B ‾ , X ∖ 2 B ) (\overline{B},X\setminus2B) ( B , X ∖ 2 B ) , and define
Ω : = 2 B ∖ B ‾ , μ : = Γ ( p ) ( u ) . \Omega:=2B\setminus\overline{B},\qquad\mu:=\Gamma^{(p)}(u). Ω := 2 B ∖ B , μ := Γ ( p ) ( u ) . By Lemma 3.6(2), 0 ≤ u ≤ 1 0\leq u\leq1 0 ≤ u ≤ 1 on X X X , u = 1 u=1 u = 1 on B ‾ \overline{B} B , u = 0 u=0 u = 0 on X ∖ 2 B X\setminus2B X ∖ 2 B , u u u is harmonic in Ω \Omega Ω , E ( p ) ( u ) = cap ( p ) ( B ‾ , X ∖ 2 B ) \mathcal{E}^{(p)}(u)=\operatorname{cap}^{(p)}(\overline{B},X\setminus2B) E ( p ) ( u ) = cap ( p ) ( B , X ∖ 2 B ) , and μ ( X ∖ Ω ) = 0 \mu(X\setminus\Omega)=0 μ ( X ∖ Ω ) = 0 . Moreover, by (43) , Lemmas 5.2 and 3.4 give that u ∈ B q , q β q ( X ) = F ( q ) u\in B^{\beta_q}_{q,q}(X)=\mathcal{F}^{(q)} u ∈ B q , q β q ( X ) = F ( q ) and E ( q ) ( u ) ∈ ( 0 , + ∞ ) \mathcal{E}^{(q)}(u)\in(0,+\infty) E ( q ) ( u ) ∈ ( 0 , + ∞ ) .
Lemma 7.3. Let u u u and μ \mu μ be as in Definition 7.2 . Then there exist constants η , ϵ > 0 \eta,\epsilon>0 η , ϵ > 0 , a Borel subset W ⊆ Ω W\subseteq\Omega W ⊆ Ω , x ∈ W x\in W x ∈ W , and { r n } ⊆ ( 0 , + ∞ ) \{r_n\}\subseteq(0,+\infty) { r n } ⊆ ( 0 , + ∞ ) with r n ↓ 0 r_n\downarrow0 r n ↓ 0 , such that μ ( W ) > 0 \mu(W)>0 μ ( W ) > 0 , μ ∣ W ≥ η H ρ 1 ∣ W \mu|_{W}\geq\eta\mathcal{H}_{\rho}^{1}|_{W} μ ∣ W ≥ η H ρ 1 ∣ W , that is, for any Borel subset A ⊆ W A\subseteq W A ⊆ W ,
μ ( A ) ≥ η H ρ 1 ( A ) , (44) \mu(A)\geq\eta\mathcal{H}_{\rho}^{1}(A), \tag*{(44)} μ ( A ) ≥ η H ρ 1 ( A ) , (44) and, moreover,
lim r ↓ 0 μ ( B ρ ( x , r ) ∖ W ) r = 0 , (45) \lim_{r\downarrow0}\frac{\mu(B_{\rho}(x,r)\setminus W)}{r}=0, \tag*{(45)} r ↓ 0 lim r μ ( B ρ ( x , r ) ∖ W ) = 0 , (45) and
μ ( B ρ ( x , r n ) ∩ W ) ≥ ϵ r n for any n . (46) \mu(B_{\rho}(x,r_n)\cap W)\geq\epsilon r_n\quad\text{for any }n. \tag*{(46)} μ ( B ρ ( x , r n ) ∩ W ) ≥ ϵ r n for any n . (46) Roughly speaking, W W W consists of points x x x where the function u u u has “large” oscillations in small balls around x x x . A suitable Morrey–Sobolev inequality then yields lower estimates for the energy measure μ = Γ ( p ) ( u ) \mu=\Gamma^{(p)}(u) μ = Γ ( p ) ( u ) . The equality (43) will play a key role in ensuring that W W W has positive μ \mu μ -measure.
Proof. By (43) , for any open subset U ⊆ X U\subseteq X U ⊆ X and any θ , r > 0 \theta,r>0 θ , r > 0 ,
E β q , U ( q ) ( u , r ) = 1 r β p ∫ U ( 1 V ( x , r ) ∫ B ( x , r ) ( ∣ u ( x ) − u ( y ) ∣ r c ) q − p ∣ u ( x ) − u ( y ) ∣ p m ( d y ) ) m ( d x ) = 1 r β p ( ∫ { x ∈ U : osc B ( x , r ) u ≤ θ r c } + ∫ { x ∈ U : osc B ( x , r ) u > θ r c } ) ⋯ m ( d x ) = : I 1 + I 2 , (47) \begin{aligned}
E_{\beta_q,U}^{(q)}(u,r)=\frac{1}{r^{\beta_p}}\int_U\left(\frac{1}{V(x,r)}\int_{B(x,r)}\left(\frac{|u(x)-u(y)|}{r^c}\right)^{q-p}|u(x)-u(y)|^p\,m(dy)\right)m(dx) \\
&= \frac{1}{r^{\beta_p}}\left(\int_{\{x\in U:\operatorname{osc}_{B(x,r)}u\le\theta r^c\}}+\int_{\{x\in U:\operatorname{osc}_{B(x,r)}u>\theta r^c\}}\right)\cdots m(dx)=:I_1+I_2, \tag*{(47)}
\end{aligned} E β q , U ( q ) ( u , r ) = r β p 1 ∫ U ( V ( x , r ) 1 ∫ B ( x , r ) ( r c ∣ u ( x ) − u ( y ) ∣ ) q − p ∣ u ( x ) − u ( y ) ∣ p m ( d y ) ) m ( d x ) = r β p 1 ( ∫ { x ∈ U : osc B ( x , r ) u ≤ θ r c } + ∫ { x ∈ U : osc B ( x , r ) u > θ r c } ) ⋯ m ( d x ) =: I 1 + I 2 , (47)
where
I 1 ≤ θ q − p E β p , U ( p ) ( u , r ) ≤ θ q − p sup r > 0 E β p , X ( p ) ( u , r ) = θ q − p M , (48) I_1\le\theta^{q-p}E_{\beta_p,U}^{(p)}(u,r)\le\theta^{q-p}\sup_{r>0}E_{\beta_p,X}^{(p)}(u,r)=\theta^{q-p}M, \tag*{(48)} I 1 ≤ θ q − p E β p , U ( p ) ( u , r ) ≤ θ q − p r > 0 sup E β p , X ( p ) ( u , r ) = θ q − p M , (48)
where M : = sup r > 0 E β p , X ( p ) ( u , r ) ∈ ( 0 , + ∞ ) M:=\sup_{r>0}E_{\beta_p,X}^{(p)}(u,r)\in(0,+\infty) M := sup r > 0 E β p , X ( p ) ( u , r ) ∈ ( 0 , + ∞ ) , which follows from u ∈ F ( p ) u\in\mathcal{F}^{(p)} u ∈ F ( p ) and E ( p ) ( u ) ∈ ( 0 , + ∞ ) \mathcal{E}^{(p)}(u)\in(0,+\infty) E ( p ) ( u ) ∈ ( 0 , + ∞ ) , and Lemma 3.4. By (3.5) and (7.1), there exists C > 0 C>0 C > 0 such that for any x ∈ X x\in X x ∈ X and any r > 0 r>0 r > 0 ,
osc B ( x , r ) u ≤ C E ( p ) ( u ) 1 p − 1 r c = K r c , \operatorname{osc}_{B(x,r)}u\le C\mathcal{E}^{(p)}(u)^{\frac{1}{p-1}}r^c=Kr^c, osc B ( x , r ) u ≤ C E ( p ) ( u ) p − 1 1 r c = K r c ,
where K : = C E ( p ) ( u ) 1 p − 1 ∈ ( 0 , + ∞ ) K:=C\mathcal{E}^{(p)}(u)^{\frac{1}{p-1}}\in(0,+\infty) K := C E ( p ) ( u ) p − 1 1 ∈ ( 0 , + ∞ ) , hence
I 2 ≤ K q − p E β p , { x ∈ U : osc B ( x , r ) u > θ r c } ( p ) ( u , r ) = K q − p E β p , W r ( p ) ( u , r ) , (49) I_2\le K^{q-p}E_{\beta_p,\{x\in U:\operatorname{osc}_{B(x,r)}u>\theta r^c\}}^{(p)}(u,r)=K^{q-p}E_{\beta_p,W_r}^{(p)}(u,r), \tag*{(49)} I 2 ≤ K q − p E β p , { x ∈ U : osc B ( x , r ) u > θ r c } ( p ) ( u , r ) = K q − p E β p , W r ( p ) ( u , r ) , (49)
where
W r : = { x ∈ U : osc B ( x , r ) u > θ r c } . W_r:=\{x\in U:\operatorname{osc}_{B(x,r)}u>\theta r^c\}. W r := { x ∈ U : osc B ( x , r ) u > θ r c } .
Since u ∈ C ( X ) u\in C(X) u ∈ C ( X ) , the set { x ∈ X : osc B ( x , r ) u > θ r c } \{x\in X:\operatorname{osc}_{B(x,r)}u>\theta r^c\} { x ∈ X : osc B ( x , r ) u > θ r c } is an open subset of X X X , and so is W r W_r W r . By Lemma 3.3, there exists C 1 > 0 C_1>0 C 1 > 0 such that
E β p , W r ( p ) ( u , r ) ≤ C 1 Γ ( p ) ( u ) ( W r ( 2 A P I ( p ) r ) ) = C 1 μ ( W r ( 2 A P I ( p ) r ) ) . (50) E_{\beta_p,W_r}^{(p)}(u,r)\le C_1\Gamma^{(p)}(u)(W_r(2A_{PI}^{(p)}r))=C_1\mu(W_r(2A_{PI}^{(p)}r)). \tag*{(50)} E β p , W r ( p ) ( u , r ) ≤ C 1 Γ ( p ) ( u ) ( W r ( 2 A P I ( p ) r )) = C 1 μ ( W r ( 2 A P I ( p ) r )) . (50)
Combining (7.5)–(7.8),
E β q , U ( q ) ( u , r ) ≤ θ q − p M + C 1 K q − p μ ( W r ( 2 A P I ( p ) r ) ) . (51) E_{\beta_q,U}^{(q)}(u,r)\le\theta^{q-p}M+C_1K^{q-p}\mu(W_r(2A_{PI}^{(p)}r)). \tag*{(51)} E β q , U ( q ) ( u , r ) ≤ θ q − p M + C 1 K q − p μ ( W r ( 2 A P I ( p ) r )) . (51)
Since u ∈ F ( q ) u\in\mathcal{F}^{(q)} u ∈ F ( q ) and E ( q ) ( u ) ∈ ( 0 , + ∞ ) \mathcal{E}^{(q)}(u)\in(0,+\infty) E ( q ) ( u ) ∈ ( 0 , + ∞ ) , there exists an open subset U ⊆ X U\subseteq X U ⊆ X such that Γ ( q ) ( u ) ( U ) ∈ ( 0 , + ∞ ) \Gamma^{(q)}(u)(U)\in(0,+\infty) Γ ( q ) ( u ) ( U ) ∈ ( 0 , + ∞ ) . By Lemma 3.5, there exist C 2 > 0 C_2>0 C 2 > 0 and { r n } ⊆ ( 0 , + ∞ ) \{r_n\}\subseteq(0,+\infty) { r n } ⊆ ( 0 , + ∞ ) with r n ↓ 0 r_n\downarrow0 r n ↓ 0 such that
Γ ( q ) ( u ) ( U ) ≤ C 2 lim inf r ↓ 0 E β q , U ( q ) ( u , r ) = C 2 lim n → + ∞ E β q , U ( q ) ( u , r n ) . \Gamma^{(q)}(u)(U)\le C_2\liminf_{r\downarrow0}E_{\beta_q,U}^{(q)}(u,r)=C_2\lim_{n\to+\infty}E_{\beta_q,U}^{(q)}(u,r_n). Γ ( q ) ( u ) ( U ) ≤ C 2 r ↓ 0 lim inf E β q , U ( q ) ( u , r ) = C 2 n → + ∞ lim E β q , U ( q ) ( u , r n ) .
Combining this with (7.9) and taking θ > 0 \theta>0 θ > 0 small enough that θ q − p M ≤ 1 2 C 2 Γ ( q ) ( u ) ( U ) \theta^{q-p}M\le\frac{1}{2C_2}\Gamma^{(q)}(u)(U) θ q − p M ≤ 2 C 2 1 Γ ( q ) ( u ) ( U ) ,
lim inf n → + ∞ μ ( W r n ( 2 A P I ( p ) r n ) ) ≥ Γ ( q ) ( u ) ( U ) 2 C 1 C 2 K q − p > 0. \liminf_{n\to+\infty}\mu(W_{r_n}(2A_{PI}^{(p)}r_n))\ge\frac{\Gamma^{(q)}(u)(U)}{2C_1C_2K^{q-p}}>0. n → + ∞ lim inf μ ( W r n ( 2 A P I ( p ) r n )) ≥ 2 C 1 C 2 K q − p Γ ( q ) ( u ) ( U ) > 0.
Let
W : = lim sup n → + ∞ W r n ( 2 A P I ( p ) r n ) = ⋂ N ≥ 1 ⋃ n ≥ N W r n ( 2 A P I ( p ) r n ) , W:=\limsup_{n\to+\infty}W_{r_n}(2A_{PI}^{(p)}r_n)=\bigcap_{N\ge1}\bigcup_{n\ge N}W_{r_n}(2A_{PI}^{(p)}r_n), W := n → + ∞ lim sup W r n ( 2 A P I ( p ) r n ) = N ≥ 1 ⋂ n ≥ N ⋃ W r n ( 2 A P I ( p ) r n ) ,
then W W W is a Borel subset of X X X , and since μ \mu μ is a finite measure on X X X , by Fatou’s lemma,
μ ( W ) ≥ lim sup n → + ∞ μ ( W r n ( 2 A P I ( p ) r n ) ) ≥ lim inf n → + ∞ μ ( W r n ( 2 A P I ( p ) r n ) ) ≥ Γ ( q ) ( u ) ( U ) 2 C 1 C 2 K q − p > 0. \mu(W)\ge\limsup_{n\to+\infty}\mu(W_{r_n}(2A_{PI}^{(p)}r_n))\ge\liminf_{n\to+\infty}\mu(W_{r_n}(2A_{PI}^{(p)}r_n))\ge\frac{\Gamma^{(q)}(u)(U)}{2C_1C_2K^{q-p}}>0. μ ( W ) ≥ n → + ∞ lim sup μ ( W r n ( 2 A P I ( p ) r n )) ≥ n → + ∞ lim inf μ ( W r n ( 2 A P I ( p ) r n )) ≥ 2 C 1 C 2 K q − p Γ ( q ) ( u ) ( U ) > 0.
Since μ ( X ∖ Ω ) = 0 \mu(X\setminus\Omega)=0 μ ( X ∖ Ω ) = 0 , replacing W W W by W ∩ Ω W\cap\Omega W ∩ Ω , we may assume that W W W is a Borel subset of Ω \Omega Ω . For any x ∈ W x\in W x ∈ W , by definition, for any N ≥ 1 N\ge1 N ≥ 1 , there exist n ≥ N n\ge N n ≥ N and y ∈ W r n y\in W_{r_n} y ∈ W r n such that d ( x , y ) < 2 A P I ( p ) r n d(x,y)<2A_{PI}^{(p)}r_n d ( x , y ) < 2 A P I ( p ) r n , which gives B ( x , 4 A P I ( p ) r n ) ⊇ B ( y , r n ) B(x,4A_{PI}^{(p)}r_n)\supseteq B(y,r_n) B ( x , 4 A P I ( p ) r n ) ⊇ B ( y , r n ) and
osc B ( x , 4 A P I ( p ) r n ) u ≥ osc B ( y , r n ) u > θ r n c , \operatorname{osc}_{B(x,4A_{PI}^{(p)}r_n)}u\ge\operatorname{osc}_{B(y,r_n)}u>\theta r_n^c, osc B ( x , 4 A P I ( p ) r n ) u ≥ osc B ( y , r n ) u > θ r n c ,
hence
lim sup r ↓ 0 osc B ( x , r ) u r c ≥ θ ( 4 A P I ( p ) ) c > 0. (52) \limsup_{r\downarrow0}\frac{\operatorname{osc}_{B(x,r)}u}{r^c}\ge\frac{\theta}{(4A_{PI}^{(p)})^c}>0. \tag*{(52)} r ↓ 0 lim sup r c osc B ( x , r ) u ≥ ( 4 A P I ( p ) ) c θ > 0. (52)
By the Morrey–Sobolev inequality implied by PI ( p ) ( β p ) \operatorname{PI}^{(p)}(\beta_p) PI ( p ) ( β p ) (see, for example, [42 Lemma 5.2] ) and (7.1), there exists C 3 > 0 C_3>0 C 3 > 0 such that
( osc B ( x , r ) u ) p ≤ C 3 r β p − d h Γ ( p ) ( u ) ( B ( x , 8 A P I ( p ) r ) ) = C 3 r ( p − 1 ) c μ ( B ( x , 8 A P I ( p ) r ) ) . (\operatorname{osc}_{B(x,r)}u)^p\le C_3r^{\beta_p-d_h}\Gamma^{(p)}(u)(B(x,8A_{PI}^{(p)}r))=C_3r^{(p-1)c}\mu(B(x,8A_{PI}^{(p)}r)). ( osc B ( x , r ) u ) p ≤ C 3 r β p − d h Γ ( p ) ( u ) ( B ( x , 8 A P I ( p ) r )) = C 3 r ( p − 1 ) c μ ( B ( x , 8 A P I ( p ) r )) .
Combining this with (52) ,
lim sup r ↓ 0 μ ( B ( x , r ) ) r c ≥ θ ρ C 3 ( 4 A P ( p ) ) p ( 8 A P ( p ) ) c > 0. \limsup_{r\downarrow0}\frac{\mu(B(x,r))}{r^{c}}\geq\frac{\theta^{\rho}}{C_{3}(4A_{P}^{(p)})^{p}(8A_{P}^{(p)})^{c}}>0. r ↓ 0 lim sup r c μ ( B ( x , r )) ≥ C 3 ( 4 A P ( p ) ) p ( 8 A P ( p ) ) c θ ρ > 0.
By (3.3), there exists ε > 0 \varepsilon>0 ε > 0 , depending only on the constant in (3.3) and the constant on the RHS of the preceding inequality, such that
lim sup r ↓ 0 μ ( B ρ ( x , r ) ) r ≥ 4 ε > 0. (53) \limsup_{r\downarrow0}\frac{\mu(B_{\rho}(x,r))}{r}\geq4\varepsilon>0. \tag*{(53)} r ↓ 0 lim sup r μ ( B ρ ( x , r )) ≥ 4 ε > 0. (53)
For any Borel subset A ⊆ W A\subseteq W A ⊆ W , recall that H ρ 1 ( A ) = lim a ↓ 0 H ρ , a 1 ( A ) \mathcal{H}_{\rho}^{1}(A)=\lim_{a\downarrow0}\mathcal{H}_{\rho,a}^{1}(A) H ρ 1 ( A ) = lim a ↓ 0 H ρ , a 1 ( A ) and
H ρ , a 1 ( A ) : = inf { ∑ i ≥ 1 diam ρ ( E i ) : A ⊆ ⋃ i ≥ 1 E i , diam ρ ( E i ) ≤ a } , \mathcal{H}_{\rho,a}^{1}(A):=\inf\left\{\sum_{i\geq1}\operatorname{diam}_{\rho}(E_i):A\subseteq\bigcup_{i\geq1}E_i,\ \operatorname{diam}_{\rho}(E_i)\leq a\right\}, H ρ , a 1 ( A ) := inf { i ≥ 1 ∑ diam ρ ( E i ) : A ⊆ i ≥ 1 ⋃ E i , diam ρ ( E i ) ≤ a } ,
where diam ρ ( E ) : = sup { ρ ( x , y ) : x , y ∈ E } \operatorname{diam}_{\rho}(E):=\sup\{\rho(x,y):x,y\in E\} diam ρ ( E ) := sup { ρ ( x , y ) : x , y ∈ E } for E ⊆ X E\subseteq X E ⊆ X . For any a > 0 a>0 a > 0 and any open set G ⊇ A G\supseteq A G ⊇ A , for any x ∈ A x\in A x ∈ A , by (53) , there exists r x ∈ ( 0 , a ) r_x\in(0,a) r x ∈ ( 0 , a ) such that B ρ ( x , r x ) ⊆ G B_{\rho}(x,r_x)\subseteq G B ρ ( x , r x ) ⊆ G and μ ( B ρ ( x , r x ) ) ≥ 2 ε r x \mu(B_{\rho}(x,r_x))\geq2\varepsilon r_x μ ( B ρ ( x , r x )) ≥ 2 ε r x . By the 5 B 5B 5 B -covering lemma (see [23 Theorem 1.2] ), there exists a family of pairwise disjoint balls { B ρ ( x i , r i ) } i \{B_{\rho}(x_i,r_i)\}_i { B ρ ( x i , r i ) } i such that A ⊆ ⋃ i B ρ ( x i , 5 r i ) A\subseteq\bigcup_iB_{\rho}(x_i,5r_i) A ⊆ ⋃ i B ρ ( x i , 5 r i ) , then
H ρ , 10 a 1 ( A ) ≤ ∑ i ( 10 r i ) ≤ 5 ε ∑ i μ ( B ρ ( x i , r i ) ) ≤ 5 ε μ ( G ) . \mathcal{H}_{\rho,10a}^{1}(A)\leq\sum_i(10r_i)\leq\frac{5}{\varepsilon}\sum_i\mu(B_{\rho}(x_i,r_i))\leq\frac{5}{\varepsilon}\mu(G). H ρ , 10 a 1 ( A ) ≤ i ∑ ( 10 r i ) ≤ ε 5 i ∑ μ ( B ρ ( x i , r i )) ≤ ε 5 μ ( G ) .
Letting a ↓ 0 a\downarrow0 a ↓ 0 , H ρ 1 ( A ) ≤ 5 ε μ ( G ) \mathcal{H}_{\rho}^{1}(A)\leq\frac{5}{\varepsilon}\mu(G) H ρ 1 ( A ) ≤ ε 5 μ ( G ) . By the outer regular property of μ \mu μ , H ρ 1 ( A ) ≤ 5 ε μ ( A ) \mathcal{H}_{\rho}^{1}(A)\leq\frac{5}{\varepsilon}\mu(A) H ρ 1 ( A ) ≤ ε 5 μ ( A ) . Therefore, we obtain (7.2) with η = 5 ε \eta=\frac{5}{\varepsilon} η = ε 5 .
Since ( X , d ) (X,d) ( X , d ) is a doubling metric space, by Assouad’s embedding theorem (see [3 , 23 , 35 ] ) and (3.3), there exist a positive integer N N N and a constant L ≥ 1 L\geq1 L ≥ 1 such that ( X , ρ ) (X,\rho) ( X , ρ ) admits an L L L -bi-Lipschitz embedding into ( R N , ∣ ⋅ ∣ ) (\mathbb{R}^{N},|\cdot|) ( R N , ∣ ⋅ ∣ ) , where ∣ ⋅ ∣ |\cdot| ∣ ⋅ ∣ is the standard Euclidean metric. Equivalently, there exists ι : X → R N \iota:X\to\mathbb{R}^{N} ι : X → R N such that
1 L ρ ( x , y ) ≤ ∣ ι ( x ) − ι ( y ) ∣ ≤ L ρ ( x , y ) \frac{1}{L}\rho(x,y)\leq|\iota(x)-\iota(y)|\leq L\rho(x,y) L 1 ρ ( x , y ) ≤ ∣ ι ( x ) − ι ( y ) ∣ ≤ L ρ ( x , y )
for any x , y ∈ X x,y\in X x , y ∈ X . For any x ∈ X x\in X x ∈ X and r > 0 r>0 r > 0 , let
U ( x , r ) : = ι − 1 ( B R N ( ι ( x ) , r ) ) , U(x,r):=\iota^{-1}(B_{\mathbb{R}^{N}}(\iota(x),r)), U ( x , r ) := ι − 1 ( B R N ( ι ( x ) , r )) ,
then
B ρ ( x , r ) ⊆ U ( x , L r ) ⊆ B ρ ( x , L 2 r ) . (54) B_{\rho}(x,r)\subseteq U(x,Lr)\subseteq B_{\rho}(x,L^{2}r). \tag*{(54)} B ρ ( x , r ) ⊆ U ( x , L r ) ⊆ B ρ ( x , L 2 r ) . (54)
By the density theorem applied to the pushforward measure ι ∗ μ \iota_{*}\mu ι ∗ μ on R N \mathbb{R}^{N} R N , for μ \mu μ -a.e. x ∈ W x\in W x ∈ W ,
lim r ↓ 0 ( ι ∗ μ ) ( B R N ( ι ( x ) , r ) ∩ ι ( W ) ) ( ι ∗ μ ) ( B R N ( ι ( x ) , r ) ) = 1 , \lim_{r\downarrow0}\frac{(\iota_{*}\mu)(B_{\mathbb{R}^{N}}(\iota(x),r)\cap\iota(W))}{(\iota_{*}\mu)(B_{\mathbb{R}^{N}}(\iota(x),r))}=1, r ↓ 0 lim ( ι ∗ μ ) ( B R N ( ι ( x ) , r )) ( ι ∗ μ ) ( B R N ( ι ( x ) , r ) ∩ ι ( W )) = 1 ,
that is,
lim r ↓ 0 μ ( U ( x , r ) ∩ W ) μ ( U ( x , r ) ) = 1. \lim_{r\downarrow0}\frac{\mu(U(x,r)\cap W)}{\mu(U(x,r))}=1. r ↓ 0 lim μ ( U ( x , r )) μ ( U ( x , r ) ∩ W ) = 1.
Fix such an x ∈ W x\in W x ∈ W , and define f ( r ) : = μ ( U ( x , r ) ∖ W ) μ ( U ( x , r ) ) f(r):=\frac{\mu(U(x,r)\setminus W)}{\mu(U(x,r))} f ( r ) := μ ( U ( x , r )) μ ( U ( x , r ) ∖ W ) , then lim r ↓ 0 f ( r ) = 0 \lim_{r\downarrow0}f(r)=0 lim r ↓ 0 f ( r ) = 0 . By (54) and (3.7),
μ ( B ρ ( x , r ) ∖ W ) ≤ μ ( U ( x , L r ) ∖ W ) = f ( L r ) μ ( U ( x , L r ) ) ≤ f ( L r ) μ ( B ρ ( x , L 2 r ) ) ≲ f ( L r ) L 2 r . \mu(B_{\rho}(x,r)\setminus W)\leq\mu(U(x,Lr)\setminus W)=f(Lr)\mu(U(x,Lr))\leq f(Lr)\mu(B_{\rho}(x,L^{2}r))\lesssim f(Lr)L^{2}r. μ ( B ρ ( x , r ) ∖ W ) ≤ μ ( U ( x , L r ) ∖ W ) = f ( L r ) μ ( U ( x , L r )) ≤ f ( L r ) μ ( B ρ ( x , L 2 r )) ≲ f ( L r ) L 2 r .
On the one hand, this gives μ ( B ρ ( x , r ) ∖ W ) r ≲ f ( L r ) → 0 \frac{\mu(B_{\rho}(x,r)\setminus W)}{r}\lesssim f(Lr)\to0 r μ ( B ρ ( x , r ) ∖ W ) ≲ f ( L r ) → 0 as r ↓ 0 r\downarrow0 r ↓ 0 , hence we obtain (7.3). On the other hand, by (53) , there exists { r n } ⊆ ( 0 , + ∞ ) \{r_n\}\subseteq(0,+\infty) { r n } ⊆ ( 0 , + ∞ ) with r n ↓ 0 r_n\downarrow0 r n ↓ 0 such that μ ( B ρ ( x , r n ) ) ≥ 2 ε r n \mu(B_{\rho}(x,r_n))\geq2\varepsilon r_n μ ( B ρ ( x , r n )) ≥ 2 ε r n . The same estimate yields that for any sufficiently large n n n , μ ( B ρ ( x , r n ) ∖ W ) ≤ ε r n \mu(B_{\rho}(x,r_n)\setminus W)\leq\varepsilon r_n μ ( B ρ ( x , r n ) ∖ W ) ≤ ε r n , hence we obtain (7.4).
We have the following cutoff functions and partition of unity on ( X , ρ ) (X,\rho) ( X , ρ ) .
Lemma 7.4. There exist constants C > 0 C > 0 C > 0 and A > 1 A > 1 A > 1 such that the following hold:
(1) For any ball B ρ ( x , r ) B_{\rho}(x,r) B ρ ( x , r ) in ( X , ρ ) (X,\rho) ( X , ρ ) , there exists ϕ ∈ F ( p ) \phi\in\mathcal{F}^{(p)} ϕ ∈ F ( p ) satisfying 0 ≤ ϕ ≤ 1 0 \le\phi\le1 0 ≤ ϕ ≤ 1 on X X X , ϕ = 1 \phi= 1 ϕ = 1 in B ρ ( x , r ) B_{\rho}(x,r) B ρ ( x , r ) , supp ( ϕ ) ⊆ B ρ ( x , A r ) \operatorname{supp}(\phi) \subseteq B_{\rho}(x,Ar) supp ( ϕ ) ⊆ B ρ ( x , A r ) , and
E ( p ) ( ϕ ) ≤ C r p − 1 , Lip ρ ( ϕ ) ≤ C r , Γ ( p ) ( ϕ ) ≤ C r p H ρ 1 . (55) \mathcal{E}^{(p)}(\phi) \le\frac{C}{r^{p-1}}, \qquad\operatorname{Lip}_{\rho}(\phi) \le\frac{C}{r}, \qquad\Gamma^{(p)}(\phi) \le\frac{C}{r^{p}}\mathcal{H}^{1}_{\rho}. \tag*{(55)} E ( p ) ( ϕ ) ≤ r p − 1 C , Lip ρ ( ϕ ) ≤ r C , Γ ( p ) ( ϕ ) ≤ r p C H ρ 1 . (55) where Lip ρ ( ϕ ) \operatorname{Lip}_{\rho}(\phi) Lip ρ ( ϕ ) denotes the Lipschitz constant of ϕ \phi ϕ with respect to ρ \rho ρ .
(2) For any r > 0 r > 0 r > 0 and any r r r -net V V V of ( X , ρ ) (X,\rho) ( X , ρ ) , there exists a family of functions { ϕ v ∈ F ( p ) : v ∈ V } \{\phi_v \in\mathcal{F}^{(p)} : v \in V\} { ϕ v ∈ F ( p ) : v ∈ V } satisfying the following conditions:
For any v ∈ V v \in V v ∈ V , 0 ≤ ϕ v ≤ 1 0 \le\phi_v \le1 0 ≤ ϕ v ≤ 1 on X X X and supp ( ϕ v ) ⊆ B ρ ( v , 2 r ) \operatorname{supp}(\phi_v) \subseteq B_{\rho}(v,2r) supp ( ϕ v ) ⊆ B ρ ( v , 2 r ) .
We have ∑ v ∈ V ϕ v = 1 \sum_{v \in V}\phi_v = 1 ∑ v ∈ V ϕ v = 1 on X X X .
For any v ∈ V v \in V v ∈ V ,
E ( p ) ( ϕ v ) ≤ C r p − 1 , Lip ρ ( ϕ v ) ≤ C r , Γ ( p ) ( ϕ v ) ≤ C r p H ρ 1 . (56) \mathcal{E}^{(p)}(\phi_v) \le\frac{C}{r^{p-1}}, \qquad\operatorname{Lip}_{\rho}(\phi_v) \le\frac{C}{r}, \qquad\Gamma^{(p)}(\phi_v) \le\frac{C}{r^{p}}\mathcal{H}^{1}_{\rho}. \tag*{(56)} E ( p ) ( ϕ v ) ≤ r p − 1 C , Lip ρ ( ϕ v ) ≤ r C , Γ ( p ) ( ϕ v ) ≤ r p C H ρ 1 . (56)
Proof. Let L ≥ 1 L \ge1 L ≥ 1 be the constant in (13) . For any ball B ρ ( x , r ) B_{\rho}(x,r) B ρ ( x , r ) in ( X , ρ ) (X,\rho) ( X , ρ ) ,
B ρ ( x , r ) ⊆ B ( x , ( L r ) 1 c ) ⊆ B ( x , ( 2 L r ) 1 c ) ⊆ B ρ ( x , 2 L 1 c r ) . B_{\rho}(x,r) \subseteq B\left(x,(Lr)^{\frac{1}{c}}\right) \subseteq B\left(x,(2Lr)^{\frac{1}{c}}\right) \subseteq B_{\rho}(x,2L^{\frac{1}{c}}r). B ρ ( x , r ) ⊆ B ( x , ( L r ) c 1 ) ⊆ B ( x , ( 2 L r ) c 1 ) ⊆ B ρ ( x , 2 L c 1 r ) .
Let ϕ ∈ F ( p ) \phi\in\mathcal{F}^{(p)} ϕ ∈ F ( p ) be the E ( p ) \mathcal{E}^{(p)} E ( p ) -capacitary potential for ( B ( x , ( L r ) 1 c ) ‾ , X ∖ B ( x , ( 2 L r ) 1 c ) ) \left(\overline{B\left(x,(Lr)^{\frac{1}{c}}\right)},X\setminus B\left(x,(2Lr)^{\frac{1}{c}}\right)\right) ( B ( x , ( L r ) c 1 ) , X ∖ B ( x , ( 2 L r ) c 1 ) ) , then 0 ≤ ϕ ≤ 1 0 \le\phi\le1 0 ≤ ϕ ≤ 1 on X X X , ϕ = 1 \phi= 1 ϕ = 1 in B ρ ( x , r ) B_{\rho}(x,r) B ρ ( x , r ) , supp ( ϕ ) ⊆ B ρ ( x , 3 L 1 c r ) \operatorname{supp}(\phi) \subseteq B_{\rho}(x,3L^{\frac{1}{c}}r) supp ( ϕ ) ⊆ B ρ ( x , 3 L c 1 r ) , and
E ( p ) ( ϕ ) ≲ 1 ( ( L r ) 1 c ) β p − d h ≤ 1 r p − 1 . \mathcal{E}^{(p)}(\phi) \lesssim\frac{1}{\left((Lr)^{\frac{1}{c}}\right)^{\beta_p-d_h}} \le\frac{1}{r^{p-1}}. E ( p ) ( ϕ ) ≲ ( ( L r ) c 1 ) β p − d h 1 ≤ r p − 1 1 .
By (15) and (18) , Lip ρ ( ϕ ) ≲ 1 r \operatorname{Lip}_{\rho}(\phi) \lesssim\frac{1}{r} Lip ρ ( ϕ ) ≲ r 1 and Γ ( p ) ( ϕ ) ≲ 1 r p H ρ 1 \Gamma^{(p)}(\phi) \lesssim\frac{1}{r^p}\mathcal{H}^{1}_{\rho} Γ ( p ) ( ϕ ) ≲ r p 1 H ρ 1 , hence we obtain (55) . Then for any r > 0 r > 0 r > 0 and any r r r -net V V V , the family { ϕ v : v ∈ V } \{\phi_v:v \in V\} { ϕ v : v ∈ V } can be constructed by the same argument as in the proof of [44 Lemma 3.2] . □ \square □
Remark 7.5. The same argument as in the proof of [44 Lemma 3.5] shows that, for any A > 1 A > 1 A > 1 , there exists a constant C > 0 C > 0 C > 0 such that, for any x ∈ X x \in X x ∈ X and any r > 0 r > 0 r > 0 , there exists a function ϕ ∈ F ( p ) \phi\in\mathcal{F}^{(p)} ϕ ∈ F ( p ) satisfying the conditions in (1).
We need some results from discrete potential theory. Let G = ( V , E ) G = (V,E) G = ( V , E ) be a finite directed graph, where V V V is the vertex set and E E E is the directed edge set. We write e = ( u , v ) e = (u,v) e = ( u , v ) , or equivalently u → v u \to v u → v , for a directed edge from u u u to v v v , where u , v ∈ V u,v \in V u , v ∈ V are vertices.
A path P P P in ( V , E ) (V,E) ( V , E ) is a finite sequence of vertices P = ( v 0 , v 1 , … , v n ) P = (v_0,v_1,\ldots,v_n) P = ( v 0 , v 1 , … , v n ) such that ( v i − 1 , v i ) ∈ E (v_{i-1},v_i) \in E ( v i − 1 , v i ) ∈ E for any i = 1 , … , n i = 1,\ldots,n i = 1 , … , n . The path P P P is called simple if the vertices v 0 , … , v n v_0,\ldots,v_n v 0 , … , v n are pairwise distinct. Given A , B ⊆ V A,B \subseteq V A , B ⊆ V , we say that P P P starts in A A A if v 0 ∈ A v_0 \in A v 0 ∈ A and ends in B B B if v n ∈ B v_n \in B v n ∈ B . A cycle C C C is a finite sequence C = ( v 0 , v 1 , … , v n ) C = (v_0,v_1,\ldots,v_n) C = ( v 0 , v 1 , … , v n ) such that n ≥ 1 n \ge1 n ≥ 1 , v n = v 0 v_n = v_0 v n = v 0 , ( v i − 1 , v i ) ∈ E (v_{i-1},v_i) \in E ( v i − 1 , v i ) ∈ E for any i = 1 , … , n i = 1,\ldots,n i = 1 , … , n , and the vertices v 0 , … , v n − 1 v_0,\ldots,v_{n-1} v 0 , … , v n − 1 are pairwise distinct. Given a path P = ( v 0 , … , v n ) P = (v_0,\ldots,v_n) P = ( v 0 , … , v n ) , we say that an edge e ∈ E e \in E e ∈ E belongs to P P P , and write e ∈ P e \in P e ∈ P , if e = ( v i − 1 , v i ) e = (v_{i-1},v_i) e = ( v i − 1 , v i ) for some i = 1 , … , n i = 1,\ldots,n i = 1 , … , n . For a cycle C C C , the notation e ∈ C e \in C e ∈ C is defined analogously.
Let f : E → [ 0 , + ∞ ) f:E \to[0,+\infty) f : E → [ 0 , + ∞ ) be a flow on G G G . Let ∂ f : V → R \partial f:V \to\mathbb{R} ∂ f : V → R be given by
( ∂ f ) ( u ) : = ∑ v ∈ V : v → u f ( ( v , u ) ) − ∑ v ∈ V : u → v f ( ( u , v ) ) , u ∈ V . (\partial f)(u) := \sum_{v \in V:v \to u} f((v,u)) - \sum_{v \in V:u \to v} f((u,v)), \qquad u \in V. ( ∂ f ) ( u ) := v ∈ V : v → u ∑ f (( v , u )) − v ∈ V : u → v ∑ f (( u , v )) , u ∈ V .
Define
S : = { u ∈ V : ( ∂ f ) ( u ) < 0 } , T : = { u ∈ V : ( ∂ f ) ( u ) > 0 } . S := \{u \in V:(\partial f)(u) < 0\}, \qquad T := \{u \in V:(\partial f)(u) > 0\}. S := { u ∈ V : ( ∂ f ) ( u ) < 0 } , T := { u ∈ V : ( ∂ f ) ( u ) > 0 } .
Thus, S S S and T T T are the source set and the target set of the flow f f f , respectively.
Let l : E → ( 0 , + ∞ ) l:E \to(0,+\infty) l : E → ( 0 , + ∞ ) be a length function on E E E . Define the transportation cost M l ( f ) M_l(f) M l ( f ) of the flow f f f with respect to l l l by
M l ( f ) : = ∑ e ∈ E f ( e ) l ( e ) . M_l(f) := \sum_{e \in E} f(e)l(e). M l ( f ) := e ∈ E ∑ f ( e ) l ( e ) .
For a path P = ( v 0 , … , v n ) P=(v_0,\ldots,v_n) P = ( v 0 , … , v n ) , define its length by
l ( P ) : = ∑ i = 1 n l ( ( v i − 1 , v i ) ) = ∑ e ∈ P l ( e ) , l(P):=\sum_{i=1}^{n}l((v_{i-1},v_i))=\sum_{e\in P}l(e), l ( P ) := i = 1 ∑ n l (( v i − 1 , v i )) = e ∈ P ∑ l ( e ) ,
where the sum ∑ e ∈ P \sum_{e\in P} ∑ e ∈ P counts occurrences with multiplicity. The length l ( C ) l(C) l ( C ) of a cycle C C C is defined analogously.
Let H : V → R H:V\to\mathbb{R} H : V → R be a potential on V V V . Define d H : E → R dH:E\to\mathbb{R} d H : E → R by
( d H ) ( e ) : = H ( v ) − H ( u ) , e = ( u , v ) ∈ E . (dH)(e):=H(v)-H(u),\qquad e=(u,v)\in E. ( d H ) ( e ) := H ( v ) − H ( u ) , e = ( u , v ) ∈ E .
For a path P = ( v 0 , … , v n ) P=(v_0,\ldots,v_n) P = ( v 0 , … , v n ) , define the potential increment of H H H along P P P by
Δ P H : = H ( v n ) − H ( v 0 ) = ∑ i = 1 n ( d H ) ( ( v i − 1 , v i ) ) = ∑ e ∈ P ( d H ) ( e ) , \Delta_P H:=H(v_n)-H(v_0)=\sum_{i=1}^{n}(dH)((v_{i-1},v_i))=\sum_{e\in P}(dH)(e), Δ P H := H ( v n ) − H ( v 0 ) = i = 1 ∑ n ( d H ) (( v i − 1 , v i )) = e ∈ P ∑ ( d H ) ( e ) ,
where the sum ∑ e ∈ P \sum_{e\in P} ∑ e ∈ P also counts occurrences with multiplicity. The discrete Green formula then reads
( f , d H ) E : = ∑ e ∈ E f ( e ) ( d H ) ( e ) = ∑ u ∈ V ( ∂ f ) ( u ) H ( u ) = : ( ∂ f , H ) V . (f,dH)_E:=\sum_{e\in E}f(e)(dH)(e)=\sum_{u\in V}(\partial f)(u)H(u)=:(\partial f,H)_V. ( f , d H ) E := e ∈ E ∑ f ( e ) ( d H ) ( e ) = u ∈ V ∑ ( ∂ f ) ( u ) H ( u ) =: ( ∂ f , H ) V .
This identity follows from a direct calculation by changing the order of summation. See also, for example, [32 Section 2.4, Exercise 2.11] .
Any flow can be decomposed into a linear combination of indicator functions of simple paths and cycles as follows.
Lemma 7.6 ([11 Proposition 1.1] ). Let f : E → [ 0 , + ∞ ) f:E\to[0,+\infty) f : E → [ 0 , + ∞ ) be a flow on G G G . Then there exist a finite family P \mathcal{P} P of simple paths starting in S S S and ending in T T T , a finite family C \mathcal{C} C of cycles, and positive coefficients q P q_P q P , P ∈ P P\in\mathcal{P} P ∈ P , and r C r_C r C , C ∈ C C\in\mathcal{C} C ∈ C , such that
f = ∑ P ∈ P q P 1 P + ∑ C ∈ C r C 1 C . f=\sum_{P\in\mathcal{P}}q_P\mathbf{1}_P+\sum_{C\in\mathcal{C}}r_C\mathbf{1}_C. f = P ∈ P ∑ q P 1 P + C ∈ C ∑ r C 1 C . Moreover,
∑ P ∈ P q P = − ∑ u ∈ S ( ∂ f ) ( u ) = ∑ u ∈ T ( ∂ f ) ( u ) = 1 2 ∥ ∂ f ∥ ℓ 1 ( V ) , (57) \sum_{P\in\mathcal{P}}q_P=-\sum_{u\in S}(\partial f)(u)=\sum_{u\in T}(\partial f)(u)=\frac{1}{2}\|\partial f\|_{\ell^1(V)}, \tag*{(57)} P ∈ P ∑ q P = − u ∈ S ∑ ( ∂ f ) ( u ) = u ∈ T ∑ ( ∂ f ) ( u ) = 2 1 ∥ ∂ f ∥ ℓ 1 ( V ) , (57) where
∥ ∂ f ∥ ℓ 1 ( V ) : = ∑ u ∈ V ∣ ( ∂ f ) ( u ) ∣ \|\partial f\|_{\ell^1(V)}:=\sum_{u\in V}|(\partial f)(u)| ∥ ∂ f ∥ ℓ 1 ( V ) := u ∈ V ∑ ∣ ( ∂ f ) ( u ) ∣ is the total variation of ∂ f \partial f ∂ f , equivalently, the total boundary mass of f f f . For any length function l : E → ( 0 , + ∞ ) l:E\to(0,+\infty) l : E → ( 0 , + ∞ ) ,
M l ( f ) = ∑ P ∈ P q P l ( P ) + ∑ C ∈ C r C l ( C ) . (58) M_l(f)=\sum_{P\in\mathcal{P}}q_Pl(P)+\sum_{C\in\mathcal{C}}r_Cl(C). \tag*{(58)} M l ( f ) = P ∈ P ∑ q P l ( P ) + C ∈ C ∑ r C l ( C ) . (58) For any potential H : V → R H:V\to\mathbb{R} H : V → R ,
( f , d H ) E = ( ∂ f , H ) V = ∑ P ∈ P q P Δ P H . (59) (f,dH)_E=(\partial f,H)_V=\sum_{P\in\mathcal{P}}q_P\Delta_P H. \tag*{(59)} ( f , d H ) E = ( ∂ f , H ) V = P ∈ P ∑ q P Δ P H . (59)
The following result will play a key role in our argument.
Proposition 7.7. Let f : E → [ 0 , + ∞ ) f:E\to[0,+\infty) f : E → [ 0 , + ∞ ) be a flow on G G G , let l : E → ( 0 , + ∞ ) l:E\to(0,+\infty) l : E → ( 0 , + ∞ ) be a length function on E E E , and let H : V → R H:V\to\mathbb{R} H : V → R be a potential on V V V . Assume that there exist constants C 1 , C 2 , C 3 , C 4 > 0 C_1,C_2,C_3,C_4>0 C 1 , C 2 , C 3 , C 4 > 0 such that
∥ ∂ f ∥ ℓ 1 ( V ) ≤ C 1 , M l ( f ) ≤ C 2 , ( ∂ f , H ) V ≥ C 3 , \|\partial f\|_{\ell^1(V)}\le C_1,\qquad M_l(f)\le C_2,\qquad(\partial f,H)_V\ge C_3, ∥ ∂ f ∥ ℓ 1 ( V ) ≤ C 1 , M l ( f ) ≤ C 2 , ( ∂ f , H ) V ≥ C 3 , and
∣ H ( s ) − H ( t ) ∣ ≤ C 4 for any s ∈ S and t ∈ T . |H(s)-H(t)|\le C_4\quad\text{for any }s\in S\text{ and }t\in T. ∣ H ( s ) − H ( t ) ∣ ≤ C 4 for any s ∈ S and t ∈ T . Then there exists a simple path P ∗ P_* P ∗ starting in S S S and ending in T T T such that
l ( P ∗ ) ≤ 2 C 2 C 4 C 3 , Δ P ∗ H ≥ C 3 C 1 . l(P_*)\le\frac{2C_2C_4}{C_3},\qquad\Delta_{P_*}H\ge\frac{C_3}{C_1}. l ( P ∗ ) ≤ C 3 2 C 2 C 4 , Δ P ∗ H ≥ C 1 C 3 .
Roughly speaking, consider a flow f f f through a network equipped with a length function l l l and a potential H H H . Assume that the total boundary mass ∥ ∂ f ∥ ℓ 1 ( V ) \|\partial f\|_{\ell^1(V)} ∥ ∂ f ∥ ℓ 1 ( V ) and the transportation cost M l ( f ) M_l(f) M l ( f ) are bounded, that the pairing ( ∂ f , H ) V (\partial f,H)_V ( ∂ f , H ) V has a positive lower bound, meaning intuitively that H H H detects a definite amount of net transport from the source set to the target set, and that the oscillation of H H H between the source and target sets is controlled. Then one can find a source-to-target path which is simultaneously “short” and along which H H H has a definite positive “increment”.
Proof. For any λ > 0 \lambda>0 λ > 0 , by assumption, (59) , and (57) ,
C 3 ≤ ( ∂ f , H ) V = ( ∑ P ∈ P : Δ P H < λ + ∑ P ∈ P : Δ P H ≥ λ ) q P Δ P H ≤ λ ∑ P ∈ P : Δ P H < λ q P + C 4 ∑ P ∈ P : Δ P H ≥ λ q P ≤ 1 2 λ C 1 + C 4 ∑ P ∈ P : Δ P H ≥ λ q P . C_3 \leq(\partial f,H)_V =
\left(\sum_{P\in\mathcal{P}:\Delta_P H<\lambda}
+
\sum_{P\in\mathcal{P}:\Delta_P H\geq\lambda}\right)q_P\Delta_P H
\leq\lambda\sum_{P\in\mathcal{P}:\Delta_P H<\lambda}q_P+C_4\sum_{P\in\mathcal{P}:\Delta_P H\geq\lambda}q_P
\leq\frac{1}{2}\lambda C_1+C_4\sum_{P\in\mathcal{P}:\Delta_P H\geq\lambda}q_P. C 3 ≤ ( ∂ f , H ) V = ( P ∈ P : Δ P H < λ ∑ + P ∈ P : Δ P H ≥ λ ∑ ) q P Δ P H ≤ λ P ∈ P : Δ P H < λ ∑ q P + C 4 P ∈ P : Δ P H ≥ λ ∑ q P ≤ 2 1 λ C 1 + C 4 P ∈ P : Δ P H ≥ λ ∑ q P .
Taking λ = C 3 C 1 \lambda=\frac{C_3}{C_1} λ = C 1 C 3 , we obtain
∑ P ∈ P : Δ P H ≥ C 3 C 1 q P ≥ C 3 2 C 4 > 0 , \sum_{P\in\mathcal{P}:\Delta_P H\geq\frac{C_3}{C_1}}q_P\geq\frac{C_3}{2C_4}>0, P ∈ P : Δ P H ≥ C 1 C 3 ∑ q P ≥ 2 C 4 C 3 > 0 ,
which implies { P ∈ P : Δ P H ≥ C 3 C 1 } ≠ ∅ \{P\in\mathcal{P}:\Delta_P H\geq\frac{C_3}{C_1}\}\neq\varnothing { P ∈ P : Δ P H ≥ C 1 C 3 } = ∅ . By assumption and (58) ,
C 2 ≥ M l ( f ) ≥ ∑ P ∈ P : Δ P H ≥ C 3 C 1 q P l ( P ) , C_2\geq M_l(f)\geq\sum_{P\in\mathcal{P}:\Delta_P H\geq\frac{C_3}{C_1}}q_P l(P), C 2 ≥ M l ( f ) ≥ P ∈ P : Δ P H ≥ C 1 C 3 ∑ q P l ( P ) ,
hence
∑ P ∈ P : Δ P H ≥ C 3 C 1 q P l ( P ) ∑ P ∈ P : Δ P H ≥ C 3 C 1 q P ≤ C 2 C 3 / ( 2 C 4 ) = 2 C 2 C 4 C 3 , \frac{\sum_{P\in\mathcal{P}:\Delta_P H\geq\frac{C_3}{C_1}}q_P l(P)}
{\sum_{P\in\mathcal{P}:\Delta_P H\geq\frac{C_3}{C_1}}q_P}
\leq\frac{C_2}{C_3/(2C_4)}
=\frac{2C_2C_4}{C_3}, ∑ P ∈ P : Δ P H ≥ C 1 C 3 q P ∑ P ∈ P : Δ P H ≥ C 1 C 3 q P l ( P ) ≤ C 3 / ( 2 C 4 ) C 2 = C 3 2 C 2 C 4 ,
which implies there exists P ∗ ∈ P P_*\in\mathcal{P} P ∗ ∈ P with Δ P ∗ H ≥ C 3 C 1 \Delta_{P_*}H\geq\frac{C_3}{C_1} Δ P ∗ H ≥ C 1 C 3 such that l ( P ∗ ) ≤ 2 C 2 C 4 C 3 l(P_*)\leq\frac{2C_2C_4}{C_3} l ( P ∗ ) ≤ C 3 2 C 2 C 4 .
We now prove Proposition 7.1 , which implies Theorem 2.5 (3).
Proof of Proposition 7.1. Let u u u and μ \mu μ be as in Definition 7.2 , and let W , x , { r n } W,x,\{r_n\} W , x , { r n } be as in Lemma 7.3 . Since x ∈ W ⊆ Ω = 2 B ∖ B ‾ x\in W\subseteq\Omega=2B\setminus\overline{B} x ∈ W ⊆ Ω = 2 B ∖ B , we have B ρ ( x , 2 r n ) ⊆ Ω B_\rho(x,2r_n)\subseteq\Omega B ρ ( x , 2 r n ) ⊆ Ω for any sufficiently large n n n . Let r = r n r=r_n r = r n . By Lemma 7.4 (1) and Remark 7.5, there exists χ ∈ F ( p ) \chi\in\mathcal{F}^{(p)} χ ∈ F ( p ) satisfying 0 ≤ χ ≤ 1 0\leq\chi\leq1 0 ≤ χ ≤ 1 on X X X , χ = 1 \chi=1 χ = 1 in B ρ ( x , r ) B_\rho(x,r) B ρ ( x , r ) , supp ( χ ) ⊆ B ρ ( x , 2 r ) \operatorname{supp}(\chi)\subseteq B_\rho(x,2r) supp ( χ ) ⊆ B ρ ( x , 2 r ) , and (55) . Let τ ≪ 1 \tau\ll1 τ ≪ 1 , let { x i : i ∈ I } \{x_i:i\in I\} { x i : i ∈ I } be a ( τ r ) (\tau r) ( τ r ) -net of ( X , ρ ) (X,\rho) ( X , ρ ) , and let { φ i : i ∈ I } \{\varphi_i:i\in I\} { φ i : i ∈ I } be the family of functions given in Lemma 7.4 (2). The constants implicit in the notation ≲ \lesssim ≲ , ≳ \gtrsim ≳ , and ≍ \asymp ≍ below are independent of τ \tau τ and r r r .
For any f , g ∈ F ( p ) f,g\in\mathcal{F}^{(p)} f , g ∈ F ( p ) , define
Υ ( f ; g ) : = ∫ W f d Γ ( p ) ( u ; g ) , \Upsilon(f;g):=\int_W f\,\mathrm{d}\Gamma^{(p)}(u;g), Υ ( f ; g ) := ∫ W f d Γ ( p ) ( u ; g ) ,
then Υ : F ( p ) × F ( p ) → R \Upsilon:\mathcal{F}^{(p)}\times\mathcal{F}^{(p)}\to\mathbb{R} Υ : F ( p ) × F ( p ) → R is bilinear. For any i , j ∈ I i,j\in I i , j ∈ I , let
a i j : = Υ ( χ φ i ; φ j ) − Υ ( χ φ j ; φ i ) , a_{ij}:=\Upsilon(\chi\varphi_i;\varphi_j)-\Upsilon(\chi\varphi_j;\varphi_i), a ij := Υ ( χ φ i ; φ j ) − Υ ( χ φ j ; φ i ) ,
then the strongly local property yields that a i j ≠ 0 a_{ij}\neq0 a ij = 0 only if ρ ( x i , x j ) < 4 τ r \rho(x_i,x_j)<4\tau r ρ ( x i , x j ) < 4 τ r .
For j ∈ I j\in I j ∈ I , consider the total variation ∣ Γ ( p ) ( u ; φ j ) ∣ |\Gamma^{(p)}(u;\varphi_j)| ∣ Γ ( p ) ( u ; φ j ) ∣ of the signed measure Γ ( p ) ( u ; φ j ) \Gamma^{(p)}(u;\varphi_j) Γ ( p ) ( u ; φ j ) on W W W , for any Borel subset A ⊆ W A\subseteq W A ⊆ W , by [27 Proposition 4.8] , (56) , and (44) ,
∣ Γ ( p ) ( u ; φ j ) ∣ ( A ) ≤ μ ( A ) Γ ( p ) ( φ j ) ( A ) p − 1 p ≲ μ ( A ) p − 1 p ( 1 ( τ r ) p H ρ 1 ( A ) ) 1 p ≲ 1 τ r μ ( A ) , |\Gamma^{(p)}(u;\varphi_j)|(A)\leq\mu(A)\Gamma^{(p)}(\varphi_j)(A)^{\frac{p-1}{p}}\lesssim\mu(A)^{\frac{p-1}{p}}\left(\frac{1}{(\tau r)^p}\mathcal{H}^1_\rho(A)\right)^{\frac{1}{p}}\lesssim\frac{1}{\tau r}\mu(A), ∣ Γ ( p ) ( u ; φ j ) ∣ ( A ) ≤ μ ( A ) Γ ( p ) ( φ j ) ( A ) p p − 1 ≲ μ ( A ) p p − 1 ( ( τ r ) p 1 H ρ 1 ( A ) ) p 1 ≲ τ r 1 μ ( A ) ,
hence
∣ Γ ( p ) ( u ; φ j ) ∣ W ≲ 1 τ r μ ∣ W . (60) |\Gamma^{(p)}(u;\varphi_j)|_W\lesssim\frac{1}{\tau r}\mu|_W. \tag*{(60)} ∣ Γ ( p ) ( u ; φ j ) ∣ W ≲ τ r 1 μ ∣ W . (60)
Similarly,
∣ Γ ( p ) ( u ; χ ) ∣ W ≲ 1 r μ ∣ W . (61) |\Gamma^{(p)}(u;\chi)|_W\lesssim\frac{1}{r}\mu|_W. \tag*{(61)} ∣ Γ ( p ) ( u ; χ ) ∣ W ≲ r 1 μ ∣ W . (61)
For any i , j ∈ I i,j \in I i , j ∈ I , by the strongly local property and (60) ,
∣ Υ ( χ φ i ; φ j ) ∣ ≤ ∫ W ∩ B ρ ( x i , 2 τ r ) ∩ B ρ ( x j , 2 τ r ) χ φ i d ∣ Γ ( p ) ( u ; φ j ) ∣ ≲ 1 τ r ∫ W ∩ B ρ ( x i , 2 τ r ) ∩ B ρ ( x j , 2 τ r ) χ φ i d μ . (62) |\Upsilon(\chi\varphi_i;\varphi_j)| \leq\int_{W\cap B_{\rho}(x_i,2\tau r)\cap B_{\rho}(x_j,2\tau r)} \chi\varphi_i\,\mathrm{d}|\Gamma^{(p)}(u;\varphi_j)|
\lesssim\frac{1}{\tau r}\int_{W\cap B_{\rho}(x_i,2\tau r)\cap B_{\rho}(x_j,2\tau r)}\chi\varphi_i\,\mathrm{d}\mu. \tag*{(62)} ∣Υ ( χ φ i ; φ j ) ∣ ≤ ∫ W ∩ B ρ ( x i , 2 τ r ) ∩ B ρ ( x j , 2 τ r ) χ φ i d ∣ Γ ( p ) ( u ; φ j ) ∣ ≲ τ r 1 ∫ W ∩ B ρ ( x i , 2 τ r ) ∩ B ρ ( x j , 2 τ r ) χ φ i d μ . (62)
Let ( V , E ) (V,E) ( V , E ) be a finite directed graph given by
V : = { i ∈ I : B ρ ( x i , 2 τ r ) ∩ B ρ ( x , 2 r ) ≠ ∅ } , E : = { ( i , j ) : i , j ∈ V , a i j > 0 } . V:=\{i\in I:B_{\rho}(x_i,2\tau r)\cap B_{\rho}(x,2r)\neq\varnothing\},\qquad E:=\{(i,j):i,j\in V,a_{ij}>0\}. V := { i ∈ I : B ρ ( x i , 2 τ r ) ∩ B ρ ( x , 2 r ) = ∅ } , E := {( i , j ) : i , j ∈ V , a ij > 0 } .
Since ( X , d ) (X,d) ( X , d ) is doubling, (13) implies that ( X , ρ ) (X,\rho) ( X , ρ ) is also doubling. Hence there exists a positive integer N ( τ ) N(\tau) N ( τ ) , depending only on τ \tau τ and the doubling constants of ( X , ρ ) (X,\rho) ( X , ρ ) , such that # V ≤ N ( τ ) \#V\leq N(\tau) # V ≤ N ( τ ) . We have
∑ j ∈ V φ j = ∑ j ∈ I φ j = 1 in B ρ ( x , 2 r ) ⊇ supp ( χ ) . (63) \sum_{j\in V}\varphi_j=\sum_{j\in I}\varphi_j=1\text{ in }B_{\rho}(x,2r)\supseteq\operatorname{supp}(\chi). \tag*{(63)} j ∈ V ∑ φ j = j ∈ I ∑ φ j = 1 in B ρ ( x , 2 r ) ⊇ supp ( χ ) . (63)
Let f : E → [ 0 , + ∞ ) f:E\to[0,+\infty) f : E → [ 0 , + ∞ ) be a flow given by f ( ( i , j ) ) = a i j f((i,j))=a_{ij} f (( i , j )) = a ij for ( i , j ) ∈ E (i,j)\in E ( i , j ) ∈ E . Then for any j ∈ V j\in V j ∈ V ,
( ∂ f ) ( j ) : = ∑ i ∈ V a i j = ∑ i ∈ V Υ ( χ φ i ; φ j ) − ∑ i ∈ V Υ ( χ φ j ; φ i ) = Υ ( χ ; φ j ) = ∫ W χ d Γ ( p ) ( u ; φ j ) . (64) (\partial f)(j):=\sum_{i\in V}a_{ij}=\sum_{i\in V}\Upsilon(\chi\varphi_i;\varphi_j)-\sum_{i\in V}\Upsilon(\chi\varphi_j;\varphi_i)=\Upsilon(\chi;\varphi_j)=\int_W\chi\,\mathrm{d}\Gamma^{(p)}(u;\varphi_j). \tag*{(64)} ( ∂ f ) ( j ) := i ∈ V ∑ a ij = i ∈ V ∑ Υ ( χ φ i ; φ j ) − i ∈ V ∑ Υ ( χ φ j ; φ i ) = Υ ( χ ; φ j ) = ∫ W χ d Γ ( p ) ( u ; φ j ) . (64)
By the Leibniz rule, since u u u is harmonic in Ω \Omega Ω and χ φ j ∈ F ( p ) ( Ω ) \chi\varphi_j\in\mathcal{F}^{(p)}(\Omega) χ φ j ∈ F ( p ) ( Ω ) ,
( ∂ f ) ( j ) = ∫ W d Γ ( p ) ( u ; χ φ j ) − ∫ W φ j d Γ ( p ) ( u ; χ ) = − ∫ Ω ∖ W d Γ ( p ) ( u ; χ φ j ) − ∫ W φ j d Γ ( p ) ( u ; χ ) . (65) (\partial f)(j)=\int_W\mathrm{d}\Gamma^{(p)}(u;\chi\varphi_j)-\int_W\varphi_j\,\mathrm{d}\Gamma^{(p)}(u;\chi)=-\int_{\Omega\setminus W}\mathrm{d}\Gamma^{(p)}(u;\chi\varphi_j)-\int_W\varphi_j\,\mathrm{d}\Gamma^{(p)}(u;\chi). \tag*{(65)} ( ∂ f ) ( j ) = ∫ W d Γ ( p ) ( u ; χ φ j ) − ∫ W φ j d Γ ( p ) ( u ; χ ) = − ∫ Ω ∖ W d Γ ( p ) ( u ; χ φ j ) − ∫ W φ j d Γ ( p ) ( u ; χ ) . (65)
where
∣ ∫ Ω ∖ W d Γ ( p ) ( u ; χ φ j ) ∣ = ∣ ∫ B ρ ( x , 2 r ) ∖ W d Γ ( p ) ( u ; χ φ j ) ∣ ≤ μ ( B ρ ( x , 2 r ) ∖ W ) p − 1 p E ( p ) ( χ φ j ) 1 p . \left|\int_{\Omega\setminus W}\mathrm{d}\Gamma^{(p)}(u;\chi\varphi_j)\right|=\left|\int_{B_{\rho}(x,2r)\setminus W}\mathrm{d}\Gamma^{(p)}(u;\chi\varphi_j)\right|\leq\mu(B_{\rho}(x,2r)\setminus W)^{\frac{p-1}{p}}\mathcal{E}^{(p)}(\chi\varphi_j)^{\frac{1}{p}}. ∫ Ω ∖ W d Γ ( p ) ( u ; χ φ j ) = ∫ B ρ ( x , 2 r ) ∖ W d Γ ( p ) ( u ; χ φ j ) ≤ μ ( B ρ ( x , 2 r ) ∖ W ) p p − 1 E ( p ) ( χ φ j ) p 1 .
Since 0 ≤ χ , φ j ≤ 1 0\leq\chi,\varphi_j\leq1 0 ≤ χ , φ j ≤ 1 on X X X , by Alg, (55) , and (56) ,
E ( p ) ( χ φ j ) 1 p ≤ E ( p ) ( φ j ) 1 p + E ( p ) ( χ ) 1 p ≲ 1 r p − 1 p + 1 ( τ r ) p − 1 p ≲ 1 ( τ r ) p − 1 p , \mathcal{E}^{(p)}(\chi\varphi_j)^{\frac{1}{p}}\leq\mathcal{E}^{(p)}(\varphi_j)^{\frac{1}{p}}+\mathcal{E}^{(p)}(\chi)^{\frac{1}{p}}\lesssim\frac{1}{r^{\frac{p-1}{p}}}+\frac{1}{(\tau r)^{\frac{p-1}{p}}}\lesssim\frac{1}{(\tau r)^{\frac{p-1}{p}}}, E ( p ) ( χ φ j ) p 1 ≤ E ( p ) ( φ j ) p 1 + E ( p ) ( χ ) p 1 ≲ r p p − 1 1 + ( τ r ) p p − 1 1 ≲ ( τ r ) p p − 1 1 ,
which gives
∑ j ∈ V ∣ ∫ Ω ∖ W d Γ ( p ) ( u ; χ φ j ) ∣ ≲ N ( τ ) τ p − 1 p ( μ ( B ρ ( x , 2 r ) ∖ W ) r ) p − 1 p . (66) \sum_{j\in V}\left|\int_{\Omega\setminus W}\mathrm{d}\Gamma^{(p)}(u;\chi\varphi_j)\right|\lesssim\frac{N(\tau)}{\tau^{\frac{p-1}{p}}}\left(\frac{\mu(B_{\rho}(x,2r)\setminus W)}{r}\right)^{\frac{p-1}{p}}. \tag*{(66)} j ∈ V ∑ ∫ Ω ∖ W d Γ ( p ) ( u ; χ φ j ) ≲ τ p p − 1 N ( τ ) ( r μ ( B ρ ( x , 2 r ) ∖ W ) ) p p − 1 . (66)
By (63) , (61) , and (17) ,
∑ j ∈ V ∣ ∫ W φ j d Γ ( p ) ( u ; χ ) ∣ ≤ ∑ j ∈ V ∫ W φ j d ∣ Γ ( p ) ( u ; χ ) ∣ = ∫ B ρ ( x , 2 r ) ∩ W d ∣ Γ ( p ) ( u ; χ ) ∣ ≲ 1 r μ ( B ρ ( x , 2 r ) ∩ W ) ≲ 1. \sum_{j\in V}\left|\int_W\varphi_j\,\mathrm{d}\Gamma^{(p)}(u;\chi)\right|\leq\sum_{j\in V}\int_W\varphi_j\,\mathrm{d}|\Gamma^{(p)}(u;\chi)|
=\int_{B_{\rho}(x,2r)\cap W}\mathrm{d}|\Gamma^{(p)}(u;\chi)|\lesssim\frac{1}{r}\mu(B_{\rho}(x,2r)\cap W)\lesssim1. j ∈ V ∑ ∫ W φ j d Γ ( p ) ( u ; χ ) ≤ j ∈ V ∑ ∫ W φ j d ∣ Γ ( p ) ( u ; χ ) ∣ = ∫ B ρ ( x , 2 r ) ∩ W d ∣ Γ ( p ) ( u ; χ ) ∣ ≲ r 1 μ ( B ρ ( x , 2 r ) ∩ W ) ≲ 1.
Combining this with (65) and (66) ,
∑ j ∈ V ∣ ( ∂ f ) ( j ) ∣ ≲ 1 + N ( τ ) τ p − 1 p ( μ ( B ρ ( x , 2 r ) ∖ W ) r ) p − 1 p . (67) \sum_{j\in V}|(\partial f)(j)|\lesssim1+\frac{N(\tau)}{\tau^{\frac{p-1}{p}}}\left(\frac{\mu(B_{\rho}(x,2r)\setminus W)}{r}\right)^{\frac{p-1}{p}}. \tag*{(67)} j ∈ V ∑ ∣ ( ∂ f ) ( j ) ∣ ≲ 1 + τ p p − 1 N ( τ ) ( r μ ( B ρ ( x , 2 r ) ∖ W ) ) p p − 1 . (67)
Let l : E → ( 0 , + ∞ ) l:E\to(0,+\infty) l : E → ( 0 , + ∞ ) be a length function given by l ( ( i , j ) ) = ρ ( x i , x j ) τ r ∈ ( 0 , 4 ) l((i,j))=\frac{\rho(x_i,x_j)}{\tau r}\in(0,4) l (( i , j )) = τ r ρ ( x i , x j ) ∈ ( 0 , 4 ) for ( i , j ) ∈ E (i,j)\in E ( i , j ) ∈ E . Then by (62) , the bounded overlap property, (63) , and (17) , the transportation cost
M l ( f ) : = ∑ ( i , j ) ∈ E f ( ( i , j ) ) l ( ( i , j ) ) ≲ ∑ i , j ∈ V ∣ Υ ( χ φ i ; φ j ) ∣ M_l(f):=\sum_{(i,j)\in E}f((i,j))l((i,j))\lesssim\sum_{i,j\in V}|\Upsilon(\chi\varphi_i;\varphi_j)| M l ( f ) := ( i , j ) ∈ E ∑ f (( i , j )) l (( i , j )) ≲ i , j ∈ V ∑ ∣Υ ( χ φ i ; φ j ) ∣
≲ 1 τ r ∑ i , j ∈ V ∫ W ∩ B ρ ( x i , 2 τ r ) ∩ B ρ ( x j , 2 τ r ) χ ϕ i d μ ≲ 1 τ r ∑ i ∈ V ∫ W ∩ B ρ ( x i , 2 τ r ) χ ϕ i d μ \lesssim\frac{1}{\tau r}\sum_{i,j\in V}\int_{W\cap B_{\rho}(x_i,2\tau r)\cap B_{\rho}(x_j,2\tau r)}\chi\phi_i\,d\mu
\lesssim\frac{1}{\tau r}\sum_{i\in V}\int_{W\cap B_{\rho}(x_i,2\tau r)}\chi\phi_i\,d\mu ≲ τ r 1 i , j ∈ V ∑ ∫ W ∩ B ρ ( x i , 2 τ r ) ∩ B ρ ( x j , 2 τ r ) χ ϕ i d μ ≲ τ r 1 i ∈ V ∑ ∫ W ∩ B ρ ( x i , 2 τ r ) χ ϕ i d μ
≲ 1 τ r μ ( B ρ ( x , 2 r ) ) ≲ 1 τ . (68) \lesssim\frac{1}{\tau r}\mu(B_{\rho}(x,2r))\lesssim\frac{1}{\tau}. \tag*{(68)} ≲ τ r 1 μ ( B ρ ( x , 2 r )) ≲ τ 1 . (68)
Let v = u − u ( x ) v=u-u(x) v = u − u ( x ) and let H : V → R H:V\to\mathbb{R} H : V → R be given by H ( i ) = v ( x i ) H(i)=v(x_i) H ( i ) = v ( x i ) . Then by (3.5), for any i , j ∈ V i,j\in V i , j ∈ V ,
∣ H ( i ) − H ( j ) ∣ = ∣ u ( x i ) − u ( x j ) ∣ ≲ ρ ( x i , x j ) ≲ r . (69) |H(i)-H(j)|=|u(x_i)-u(x_j)|\lesssim\rho(x_i,x_j)\lesssim r. \tag*{(69)} ∣ H ( i ) − H ( j ) ∣ = ∣ u ( x i ) − u ( x j ) ∣ ≲ ρ ( x i , x j ) ≲ r . (69)
By the strongly local property and (7.22),
( ∂ f , H ) V : = ∑ j ∈ V ( ∂ f ) ( j ) H ( j ) = ∫ W χ d Γ ( p ) ( v ; ∑ j ∈ V v ( x j ) ϕ j ) = ∫ W χ d Γ ( p ) ( v ; w ) , (70) (\partial f,H)_V:=\sum_{j\in V}(\partial f)(j)H(j)=\int_W\chi\,d\Gamma^{(p)}\left(v;\sum_{j\in V}v(x_j)\phi_j\right)=\int_W\chi\,d\Gamma^{(p)}(v;w), \tag*{(70)} ( ∂ f , H ) V := j ∈ V ∑ ( ∂ f ) ( j ) H ( j ) = ∫ W χ d Γ ( p ) v ; j ∈ V ∑ v ( x j ) ϕ j = ∫ W χ d Γ ( p ) ( v ; w ) , (70)
where
w : = ∑ j ∈ V v ( x j ) ϕ j . w:=\sum_{j\in V}v(x_j)\phi_j. w := j ∈ V ∑ v ( x j ) ϕ j .
By (7.4), there exists a constant C 1 > 0 C_1>0 C 1 > 0 such that
∫ W χ d Γ ( p ) ( v ; v ) ≥ μ ( B ρ ( x , r ) ∩ W ) ≥ 1 C 1 r . (71) \int_W\chi\,d\Gamma^{(p)}(v;v)\geq\mu(B_{\rho}(x,r)\cap W)\geq\frac{1}{C_1}r. \tag*{(71)} ∫ W χ d Γ ( p ) ( v ; v ) ≥ μ ( B ρ ( x , r ) ∩ W ) ≥ C 1 1 r . (71)
By the Leibniz rule, since u u u is harmonic in Ω \Omega Ω and χ ( v − w ) ∈ F ( p ) ( Ω ) \chi(v-w)\in\mathcal{F}^{(p)}(\Omega) χ ( v − w ) ∈ F ( p ) ( Ω ) ,
∫ W χ d Γ ( p ) ( v ; v − w ) = ∫ W d Γ ( p ) ( v ; χ ( v − w ) ) − ∫ W ( v − w ) d Γ ( p ) ( v ; χ ) = − ∫ Ω ∖ W d Γ ( p ) ( v ; χ ( v − w ) ) − ∫ W ( v − w ) d Γ ( p ) ( v ; χ ) . (72) \begin{aligned}
\int_W\chi\,d\Gamma^{(p)}(v;v-w)
&=\int_Wd\Gamma^{(p)}(v;\chi(v-w))-\int_W(v-w)d\Gamma^{(p)}(v;\chi)\\
&=-\int_{\Omega\setminus W}d\Gamma^{(p)}(v;\chi(v-w))-\int_W(v-w)d\Gamma^{(p)}(v;\chi). \tag*{(72)}
\end{aligned} ∫ W χ d Γ ( p ) ( v ; v − w ) = ∫ W d Γ ( p ) ( v ; χ ( v − w )) − ∫ W ( v − w ) d Γ ( p ) ( v ; χ ) = − ∫ Ω ∖ W d Γ ( p ) ( v ; χ ( v − w )) − ∫ W ( v − w ) d Γ ( p ) ( v ; χ ) . (72)
By (7.21) and (3.5), in B ρ ( x , 2 r ) B_{\rho}(x,2r) B ρ ( x , 2 r ) ,
∣ v − w ∣ = ∣ ∑ j ∈ V ( v − v ( x j ) ) ϕ j ∣ ≤ ∑ j ∈ V ∣ u − u ( x j ) ∣ ϕ j ≲ τ r , (73) |v-w|=\left|\sum_{j\in V}(v-v(x_j))\phi_j\right|\leq\sum_{j\in V}|u-u(x_j)|\phi_j\lesssim\tau r, \tag*{(73)} ∣ v − w ∣ = j ∈ V ∑ ( v − v ( x j )) ϕ j ≤ j ∈ V ∑ ∣ u − u ( x j ) ∣ ϕ j ≲ τ r , (73)
hence by (7.19) and (3.7),
∣ ∫ W ( v − w ) d Γ ( p ) ( v ; χ ) ∣ ≤ ∫ B ρ ( x , 2 r ) ∩ W ∣ v − w ∣ d ∣ Γ ( p ) ( v ; χ ) ∣ ≲ ( τ r ) ( 1 r μ ( B ρ ( x , 2 r ) ∩ W ) ) ≲ τ r . \left|\int_W(v-w)d\Gamma^{(p)}(v;\chi)\right|\leq\int_{B_{\rho}(x,2r)\cap W}|v-w|d|\Gamma^{(p)}(v;\chi)|\lesssim(\tau r)\left(\frac{1}{r}\mu(B_{\rho}(x,2r)\cap W)\right)\lesssim\tau r. ∫ W ( v − w ) d Γ ( p ) ( v ; χ ) ≤ ∫ B ρ ( x , 2 r ) ∩ W ∣ v − w ∣ d ∣ Γ ( p ) ( v ; χ ) ∣ ≲ ( τ r ) ( r 1 μ ( B ρ ( x , 2 r ) ∩ W ) ) ≲ τ r .
That is, there exists a constant C 2 > 0 C_2>0 C 2 > 0 such that
∣ ∫ W ( v − w ) d Γ ( p ) ( v ; χ ) ∣ ≤ C 2 τ r . (74) \left|\int_W(v-w)d\Gamma^{(p)}(v;\chi)\right|\leq C_2\tau r. \tag*{(74)} ∫ W ( v − w ) d Γ ( p ) ( v ; χ ) ≤ C 2 τ r . (74)
Moreover,
∣ ∫ Ω ∖ W d Γ ( p ) ( v ; χ ( v − w ) ) ∣ = ∣ ∫ B ρ ( x , 2 r ) ∖ W d Γ ( p ) ( v ; χ ( v − w ) ) ∣ ≤ μ ( B ρ ( x , 2 r ) ∖ W ) p − 1 p Γ ( p ) ( χ ( v − w ) ) ( B ρ ( x , 2 r ) ) 1 p . (75) \begin{aligned}
\left|\int_{\Omega\setminus W}d\Gamma^{(p)}(v;\chi(v-w))\right|
&=\left|\int_{B_{\rho}(x,2r)\setminus W}d\Gamma^{(p)}(v;\chi(v-w))\right|\\
&\leq\mu(B_{\rho}(x,2r)\setminus W)^{\frac{p-1}{p}}\Gamma^{(p)}(\chi(v-w))(B_{\rho}(x,2r))^{\frac{1}{p}}. \tag*{(75)}
\end{aligned} ∫ Ω ∖ W d Γ ( p ) ( v ; χ ( v − w )) = ∫ B ρ ( x , 2 r ) ∖ W d Γ ( p ) ( v ; χ ( v − w )) ≤ μ ( B ρ ( x , 2 r ) ∖ W ) p p − 1 Γ ( p ) ( χ ( v − w )) ( B ρ ( x , 2 r ) ) p 1 . (75)
By Alg, (7.31), and (7.13),
Γ ( p ) ( χ ( v − w ) ) ( B ρ ( x , 2 r ) ) 1 p ≤ ∥ χ ∥ L ∞ ( B ρ ( x , 2 r ) ) Γ ( p ) ( v − w ) ( B ρ ( x , 2 r ) ) 1 p + ∥ v − w ∥ L ∞ ( B ρ ( x , 2 r ) ) Γ ( p ) ( χ ) ( B ρ ( x , 2 r ) ) 1 p ≲ Γ ( p ) ( v − w ) ( B ρ ( x , 2 r ) ) 1 p + τ r 1 p . (76) \begin{aligned}
\Gamma^{(p)}(\chi(v-w))(B_{\rho}(x,2r))^{\frac{1}{p}}
&\leq\|\chi\|_{L^\infty(B_{\rho}(x,2r))}\Gamma^{(p)}(v-w)(B_{\rho}(x,2r))^{\frac{1}{p}}\\
&\quad+\|v-w\|_{L^\infty(B_{\rho}(x,2r))}\Gamma^{(p)}(\chi)(B_{\rho}(x,2r))^{\frac{1}{p}}\\
&\lesssim\Gamma^{(p)}(v-w)(B_{\rho}(x,2r))^{\frac{1}{p}}+\tau r^{\frac{1}{p}}. \tag*{(76)}
\end{aligned} Γ ( p ) ( χ ( v − w )) ( B ρ ( x , 2 r ) ) p 1 ≤ ∥ χ ∥ L ∞ ( B ρ ( x , 2 r )) Γ ( p ) ( v − w ) ( B ρ ( x , 2 r ) ) p 1 + ∥ v − w ∥ L ∞ ( B ρ ( x , 2 r )) Γ ( p ) ( χ ) ( B ρ ( x , 2 r ) ) p 1 ≲ Γ ( p ) ( v − w ) ( B ρ ( x , 2 r ) ) p 1 + τ r p 1 . (76)
By (7.31) and the bounded overlap property,
Γ ( p ) ( v − w ) ( B ρ ( x , 2 r ) ) = Γ ( p ) ( ∑ j ∈ V ( v − v ( x j ) ) ϕ j ) ( B ρ ( x , 2 r ) ) ≤ ∑ i ∈ V Γ ( p ) ( ∑ j : ρ ( x i , x j ) < 4 τ r ( v − v ( x j ) ) φ j ) ( B ρ ( x i , 2 τ r ) ) \begin{aligned}
\Gamma^{(p)}(v-w)(B_{\rho}(x,2r))=\Gamma^{(p)}\left(\sum_{j\in V}(v-v(x_j))\phi_j\right)(B_{\rho}(x,2r)) \\
&\le\sum_{i\in V}\Gamma^{(p)}\left(\sum_{j:\rho(x_i,x_j)<4\tau r}(v-v(x_j))\varphi_j\right)(B_\rho(x_i,2\tau r))
\end{aligned} Γ ( p ) ( v − w ) ( B ρ ( x , 2 r )) = Γ ( p ) j ∈ V ∑ ( v − v ( x j )) ϕ j ( B ρ ( x , 2 r )) ≤ i ∈ V ∑ Γ ( p ) j : ρ ( x i , x j ) < 4 τ r ∑ ( v − v ( x j )) φ j ( B ρ ( x i , 2 τ r ))
≲ ∑ i ∈ V ∑ j : ρ ( x i , x j ) < 4 τ r Γ ( p ) ( ( v − v ( x j ) ) φ j ) ( B ρ ( x i , 2 τ r ) ) , \lesssim\sum_{i\in V}\sum_{j:\rho(x_i,x_j)<4\tau r}\Gamma^{(p)}\left((v-v(x_j))\varphi_j\right)(B_\rho(x_i,2\tau r)), ≲ i ∈ V ∑ j : ρ ( x i , x j ) < 4 τ r ∑ Γ ( p ) ( ( v − v ( x j )) φ j ) ( B ρ ( x i , 2 τ r )) ,
where for any such i , j i,j i , j , by Alg, (15) , (56) , and (17) ,
Γ ( p ) ( ( v − v ( x j ) ) φ j ) ( B ρ ( x i , 2 τ r ) ) 1 p ≤ ∥ v − v ( x j ) ∥ L ∞ ( B ρ ( x i , 2 τ r ) ) Γ ( p ) ( φ j ) ( B ρ ( x i , 2 τ r ) ) 1 p + ∥ φ j ∥ L ∞ ( B ρ ( x i , 2 τ r ) ) Γ ( p ) ( v − v ( x j ) ) ( B ρ ( x i , 2 τ r ) ) 1 p \Gamma^{(p)}\left((v-v(x_j))\varphi_j\right)(B_\rho(x_i,2\tau r))^{\frac{1}{p}}
\le\|v-v(x_j)\|_{L^\infty(B_\rho(x_i,2\tau r))}\Gamma^{(p)}(\varphi_j)(B_\rho(x_i,2\tau r))^{\frac{1}{p}}
+\|\varphi_j\|_{L^\infty(B_\rho(x_i,2\tau r))}\Gamma^{(p)}(v-v(x_j))(B_\rho(x_i,2\tau r))^{\frac{1}{p}} Γ ( p ) ( ( v − v ( x j )) φ j ) ( B ρ ( x i , 2 τ r ) ) p 1 ≤ ∥ v − v ( x j ) ∥ L ∞ ( B ρ ( x i , 2 τ r )) Γ ( p ) ( φ j ) ( B ρ ( x i , 2 τ r ) ) p 1 + ∥ φ j ∥ L ∞ ( B ρ ( x i , 2 τ r )) Γ ( p ) ( v − v ( x j )) ( B ρ ( x i , 2 τ r ) ) p 1
≲ ( τ r ) 1 ( τ r ) p − 1 p + ( τ r ) 1 p = 2 ( τ r ) 1 p , \lesssim(\tau r)\frac{1}{(\tau r)^{\frac{p-1}{p}}}+(\tau r)^{\frac{1}{p}}=2(\tau r)^{\frac{1}{p}}, ≲ ( τ r ) ( τ r ) p p − 1 1 + ( τ r ) p 1 = 2 ( τ r ) p 1 ,
hence
Γ ( p ) ( v − w ) ( B ρ ( x , 2 r ) ) ≲ ∑ i ∈ V ∑ j : ρ ( x i , x j ) < 4 τ r τ r ≲ ∑ i ∈ V τ r ≤ N ( τ ) τ r . \Gamma^{(p)}(v-w)(B_\rho(x,2r))\lesssim\sum_{i\in V}\sum_{j:\rho(x_i,x_j)<4\tau r}\tau r\lesssim\sum_{i\in V}\tau r\le N(\tau)\tau r. Γ ( p ) ( v − w ) ( B ρ ( x , 2 r )) ≲ i ∈ V ∑ j : ρ ( x i , x j ) < 4 τ r ∑ τ r ≲ i ∈ V ∑ τ r ≤ N ( τ ) τ r .
Combining this with (75) and (76) , we obtain a constant C 3 > 0 C_3>0 C 3 > 0 such that
∣ ∫ Ω ∖ W d Γ ( p ) ( v ; χ ( v − w ) ) ∣ ≤ C 3 ( μ ( B ρ ( x , 2 r ) ∖ W ) r ) p − 1 p ( N ( τ ) τ ) 1 p r . \left|\int_{\Omega\setminus W} \mathrm{d}\Gamma^{(p)}(v;\chi(v-w))\right|
\le C_3\left(\frac{\mu(B_\rho(x,2r)\setminus W)}{r}\right)^{\frac{p-1}{p}}(N(\tau)\tau)^{\frac{1}{p}}r. ∫ Ω ∖ W d Γ ( p ) ( v ; χ ( v − w )) ≤ C 3 ( r μ ( B ρ ( x , 2 r ) ∖ W ) ) p p − 1 ( N ( τ ) τ ) p 1 r .
Together with (70) , (71) , (72) , and (74) , this yields
( ∂ f , H ) V = ∫ W χ d Γ ( p ) ( v ; v ) − ∫ W χ d Γ ( p ) ( v ; v − w ) ≥ ( 1 C 1 − C 2 τ − C 3 ( μ ( B ρ ( x , 2 r ) ∖ W ) r ) p − 1 p ( N ( τ ) τ ) 1 p ) r . (77) \begin{aligned}
(\partial f,H)_V=\int_W\chi\mathrm{d}\Gamma^{(p)}(v;v)-\int_W\chi\mathrm{d}\Gamma^{(p)}(v;v-w) \\
&\ge\left(\frac{1}{C_1}-C_2\tau-C_3\left(\frac{\mu(B_\rho(x,2r)\setminus W)}{r}\right)^{\frac{p-1}{p}}(N(\tau)\tau)^{\frac{1}{p}}\right)r. \tag*{(77)}
\end{aligned} ( ∂ f , H ) V = ∫ W χ d Γ ( p ) ( v ; v ) − ∫ W χ d Γ ( p ) ( v ; v − w ) ≥ ( C 1 1 − C 2 τ − C 3 ( r μ ( B ρ ( x , 2 r ) ∖ W ) ) p p − 1 ( N ( τ ) τ ) p 1 ) r . (77)
Fix τ ∈ ( 0 , 1 2 C 1 C 2 ) \tau\in\left(0,\frac{1}{2C_1C_2}\right) τ ∈ ( 0 , 2 C 1 C 2 1 ) . Since r = r n ↓ 0 r=r_n\downarrow0 r = r n ↓ 0 , (45) yields an integer N N N such that, for any n > N n>N n > N ,
N ( τ ) τ p − 1 ( μ ( B ρ ( x , 2 r ) ∖ W ) r ) p − 1 p ≤ 1 , C 3 ( μ ( B ρ ( x , 2 r ) ∖ W ) r ) p − 1 p ( N ( τ ) τ ) 1 p ≤ 1 4 C 1 . \frac{N(\tau)}{\tau^{p-1}}\left(\frac{\mu(B_\rho(x,2r)\setminus W)}{r}\right)^{\frac{p-1}{p}}\le1,\qquad
C_3\left(\frac{\mu(B_\rho(x,2r)\setminus W)}{r}\right)^{\frac{p-1}{p}}(N(\tau)\tau)^{\frac{1}{p}}\le\frac{1}{4C_1}. τ p − 1 N ( τ ) ( r μ ( B ρ ( x , 2 r ) ∖ W ) ) p p − 1 ≤ 1 , C 3 ( r μ ( B ρ ( x , 2 r ) ∖ W ) ) p p − 1 ( N ( τ ) τ ) p 1 ≤ 4 C 1 1 .
Then (67) and (77) give
∑ j ∈ V ∣ ( ∂ f ) ( j ) ∣ ≲ 1 , ( ∂ f , H ) V ≥ 1 4 C 1 r . \sum_{j\in V}|(\partial f)(j)|\lesssim1,\qquad(\partial f,H)_V\ge\frac{1}{4C_1}r. j ∈ V ∑ ∣ ( ∂ f ) ( j ) ∣ ≲ 1 , ( ∂ f , H ) V ≥ 4 C 1 1 r .
Combining these two inequalities with (68) and (69) and applying Proposition 7.7, we obtain i 0 , … , i m ∈ V i_0,\ldots,i_m\in V i 0 , … , i m ∈ V with ( i k − 1 , i k ) ∈ E (i_{k-1},i_k)\in E ( i k − 1 , i k ) ∈ E for any k = 1 , … , m k=1,\ldots,m k = 1 , … , m such that
∑ k = 1 m l ( i k − 1 , i k ) ≲ 1 τ , H ( i m ) − H ( i 0 ) ≳ r , \sum_{k=1}^{m}l(i_{k-1},i_k)\lesssim\frac{1}{\tau},\qquad H(i_m)-H(i_0)\gtrsim r, k = 1 ∑ m l ( i k − 1 , i k ) ≲ τ 1 , H ( i m ) − H ( i 0 ) ≳ r ,
that is,
∑ k = 1 m ρ ( x i k − 1 , x i k ) ≲ r , u ( x i m ) − u ( x i 0 ) ≳ r . \sum_{k=1}^{m}\rho(x_{i_{k-1}},x_{i_k})\lesssim r,\qquad u(x_{i_m})-u(x_{i_0})\gtrsim r. k = 1 ∑ m ρ ( x i k − 1 , x i k ) ≲ r , u ( x i m ) − u ( x i 0 ) ≳ r .
It follows from (15) and the second inequality above that
ρ ( x i 0 , x i m ) ≳ r . \rho(x_{i_0},x_{i_m})\gtrsim r. ρ ( x i 0 , x i m ) ≳ r .
For any k = 1 , … , m k=1,\ldots,m k = 1 , … , m , since ( i k − 1 , i k ) ∈ E (i_{k-1},i_k)\in E ( i k − 1 , i k ) ∈ E , ρ ( x i k − 1 , x i k ) ∈ ( 0 , 4 τ r ) \rho(x_{i_{k-1}},x_{i_k})\in(0,4\tau r) ρ ( x i k − 1 , x i k ) ∈ ( 0 , 4 τ r ) . Then by (13) and the triangle inequality for the metric d d d ,
r ≲ ρ ( x i 0 , x i m ) ≃ d ( x i 0 , x i m ) c ≤ ( ∑ k = 1 m d ( x i k − 1 , x i k ) ) c ≲ ( ∑ k = 1 m ρ ( x i k − 1 , x i k ) 1 c ) c ≤ ( max k = 1 , … , m ρ ( x i k − 1 , x i k ) 1 c − 1 ∑ k = 1 m ρ ( x i k − 1 , x i k ) ) c ≲ ( ( τ r ) 1 c − 1 r ) c = τ 1 − c r , \begin{aligned}
r\lesssim\rho(x_{i_0},x_{i_m})\simeq d(x_{i_0},x_{i_m})^c\le\left(\sum_{k=1}^{m}d(x_{i_{k-1}},x_{i_k})\right)^c\lesssim\left(\sum_{k=1}^{m}\rho(x_{i_{k-1}},x_{i_k})^{\frac{1}{c}}\right)^c \\
&\le\left(\max_{k=1,\ldots,m} \rho(x_{i_{k-1}},x_{i_k})^{\frac{1}{c}-1}\sum_{k=1}^{m}\rho(x_{i_{k-1}},x_{i_k})\right)^{c}\lesssim\left((\tau r)^{\frac{1}{c}-1}r\right)^{c}=\tau^{1-c}r,
\end{aligned} r ≲ ρ ( x i 0 , x i m ) ≃ d ( x i 0 , x i m ) c ≤ ( k = 1 ∑ m d ( x i k − 1 , x i k ) ) c ≲ ( k = 1 ∑ m ρ ( x i k − 1 , x i k ) c 1 ) c ≤ ( k = 1 , … , m max ρ ( x i k − 1 , x i k ) c 1 − 1 k = 1 ∑ m ρ ( x i k − 1 , x i k ) ) c ≲ ( ( τ r ) c 1 − 1 r ) c = τ 1 − c r ,
hence 1 ≲ τ 1 − c 1\lesssim\tau^{1-c} 1 ≲ τ 1 − c for any τ ∈ ( 0 , 1 2 C 1 C 2 ) \tau\in\left(0,\frac{1}{2C_{1}C_{2}}\right) τ ∈ ( 0 , 2 C 1 C 2 1 ) . Since c ∈ ( 0 , 1 ) c\in(0,1) c ∈ ( 0 , 1 ) , letting τ ↓ 0 \tau\downarrow0 τ ↓ 0 , we obtain a contradiction.
AI statement
The author’s study of the regularity of the p p p -walk dimension began in February 2025, motivated by a preprint version of [20 ] . In § 7 , the idea of “tangent current” was suggested by OpenAI’s ChatGPT (GPT-5.6 Sol) in mid-August 2026, with reference to [36 ] . The author subsequently simplified the idea to obtain the elementary formulation presented in Proposition 7.7. During the final preparation of this paper, ChatGPT (GPT-6 Astra) was also used to assist with proofreading, polishing the English, and generating TikZ code for figures. The author takes full responsibility for the content of this paper.
Department of Mathematics, Aarhus University, 8000 Aarhus C, Denmark
E-mail address: yangmengqh@gmail.com
E-mail address: yang@math.au.dk