Introduction
Subwavelength resonator systems provide a mathematical framework for the confinement and manipulation of waves on scales smaller than the wavelength. One particular way of realising such subwavelength resonances is through high contrast inclusions embedded in a uniform background medium. In this setting, one can show that, at leading order in the contrast parameter δ \delta δ , the resonance frequencies and eigenmodes are governed by the eigenvalues and eigenmodes of a generalised capacitance matrix [4 , 6 ] . Minnaert resonances [6 , 24 ] of an air bubble in water are a classical example of such systems. Subwavelength physics also provide a promising research direction for metamaterials design [4 , 18 ] .
This connection with metamaterial design motivates the study of large resonator systems. A particularly important question then concerns the off-diagonal decay of the capacitance matrix. Good estimates on the asymptotic decay can be leveraged to obtain results about spectral convergence for finite systems to their infinite counterpart [5 ] , localisation properties of defect modes [3 ] and Anderson localisation [2 , 13 ] , where the decay plays a particularly important role [1 ] .
Beyond wave physics, the capacitance matrix is a central object in electrostatics, where it is computed for integrated circuits by floating random walk solvers [23 , 27 ] . Here, decay estimates indicate how much computational effort is needed to accurately resolve distant resonator interactions, and when far-away resonators can be neglected.
Results on this decay, however, are scarce. While it is known that the decay is exponential for crystals[11 ] , only recently an upper bound was establish for a periodic chain whose fundamental cell contain one resonator [2 ] . The proof of this revolves around analysing the symbol of the associated Laurent operator, hence the approach is restricted to periodic systems only. To the best of our knowledge, no decay estimates are available for disordered configurations or screens.
In this paper, we establish upper and lower bounds for the off-diagonal entries of the capacitance matrix for three classes of infinite, not necessarily periodic, configurations of spherical resonators: chains contained in a cylinder, screens contained in a slab, and relatively dense three-dimensional configurations. For two distinct resonators with centres x i x_i x i and x j x_j x j , write d i j = ∣ x i − x j ∣ d_{ij} = |x_i-x_j| d ij = ∣ x i − x j ∣ . For chains and screens, respectively, we prove
∣ C i j ∣ ≍ 1 d i j log 2 ( 2 + d i j ) , ∣ C i j ∣ ≍ d i j − 3 |C_{ij}| \asymp\frac{1}{d_{ij}\log^{2}(2+d_{ij})}, \qquad|C_{ij}| \asymp d_{ij}^{-3} ∣ C ij ∣ ≍ d ij log 2 ( 2 + d ij ) 1 , ∣ C ij ∣ ≍ d ij − 3
where ≍ \asymp ≍ denotes upper and lower bounds by positive constant multiples. For crystals, we obtain
C 4 e − c 4 d i j ≤ ∣ C i j ∣ ≤ C 3 e − c 3 d i j C_{4}e^{-c_{4}d_{ij}} \le|C_{ij}| \le C_{3}e^{-c_{3}d_{ij}} C 4 e − c 4 d ij ≤ ∣ C ij ∣ ≤ C 3 e − c 3 d ij
with positive constants C 3 , c 3 , C 4 , c 4 C_{3},c_{3},C_{4},c_{4} C 3 , c 3 , C 4 , c 4 . The upper and lower exponential rates are allowed to differ. All constants depend only on the geometric bounds specified in Theorems 2.5 to 2.7. In particular, they are uniform over all distinct index pairs and all configurations satisfying those fixed bounds.
Our approach uses the classical connection between the Dirichlet problem and Brownian motion [15 , 25 ] . The unit potential associated with a resonator is the probability that Brownian motion first reaches that resonator before any other. The upper bounds follow from survival estimates and bounds on the killed transition density. The lower bounds combine Green’s function estimates in suitable subdomains with path constructions that avoid the other resonators.
The paper is organised as follows. In Section 2, we introduce the setting and state our main results. Section 3 collects the necessary facts on Brownian motion and proves the probabilistic representation of the capacitance matrix. In Section 4, we study Brownian motion killed on the resonators and relate hitting probabilities to occupation times. Sections 5 and 6 contain the proofs of the upper and lower bounds, respectively, and we conclude in Section 7.
Setting and Main Result
Let I \mathcal{I} I be a countable index set and { D i } i ∈ I \{D_i\}_{i\in\mathcal{I}} { D i } i ∈ I a family of bounded Lipschitz domains in R 3 \mathbb{R}^{3} R 3 with pairwise disjoint closures and set D = ⋃ i ∈ I D i D=\bigcup_{i\in\mathcal{I}}D_i D = ⋃ i ∈ I D i . We further write Ω \Omega Ω for R 3 ∖ D ‾ \mathbb{R}^{3}\setminus\overline{D} R 3 ∖ D . Inside each resonator D i D_i D i , let the material density and bulk modulus be given by ρ i \rho_i ρ i and κ i \kappa_i κ i respectively. Outside the resonators, we denote these quantities by ρ \rho ρ and κ \kappa κ . The interior and exterior wavespeeds are then given by
v i = κ i ρ i and v = κ ρ . (1) v_i=\sqrt{\frac{\kappa_i}{\rho_i}}\qquad\text{and}\qquad v=\sqrt{\frac{\kappa}{\rho}}. \tag*{(1)} v i = ρ i κ i and v = ρ κ . (1)
We assume that there exist v max ≥ v min > 0 v_{\max}\geq v_{\min}>0 v m a x ≥ v m i n > 0 such that v max ≥ v i , v ≥ v min v_{\max}\geq v_i,v\geq v_{\min} v m a x ≥ v i , v ≥ v m i n . If we define the contrast parameter δ i = ρ i ρ \delta_i=\frac{\rho_i}{\rho} δ i = ρ ρ i the resonant frequencies ω ( δ ) ∈ C \omega(\delta)\in\mathbb{C} ω ( δ ) ∈ C are values such that the system below has a nontrivial solution u u u .
Δ u + ω 2 v 2 u = 0 in Ω , Δ u + ω 2 v i 2 u = 0 in D i , i ∈ I , u ∣ + = u ∣ − on ∂ D i , i ∈ I , δ i ∂ u ∂ ν ∣ + = ∂ u ∂ ν ∣ − on ∂ D i , i ∈ I , u is outgoing. (2) \begin{aligned}
\Delta u+\frac{\omega^2}{v^2}u&=0 &&\text{in }\Omega,\\
\Delta u+\frac{\omega^2}{v_i^2}u&=0 &&\text{in }D_i,\quad i\in\mathcal{I},\\
u\big|_{+}&=u\big|_{-} &&\text{on }\partial D_i,\quad i\in\mathcal{I},\\
\delta_i\frac{\partial u}{\partial\nu}\bigg|_{+}&=\frac{\partial u}{\partial\nu}\bigg|_{-} &&\text{on }\partial D_i,\quad i\in\mathcal{I},\\
u&\text{ is outgoing.} \tag*{(2)}
\end{aligned} Δ u + v 2 ω 2 u Δ u + v i 2 ω 2 u u + δ i ∂ ν ∂ u + u = 0 = 0 = u − = ∂ ν ∂ u − is outgoing. in Ω , in D i , i ∈ I , on ∂ D i , i ∈ I , on ∂ D i , i ∈ I , (2)
Here, ν \nu ν denotes the unit normal on ∂ D i \partial D_i ∂ D i pointing out of D i D_i D i and the signs + + + and − - − denote exterior and interior traces, respectively. In the interest of readability, we will restrict ourselves to the case δ i ≡ δ \delta_i\equiv\delta δ i ≡ δ , however all the main results should be adaptable to the more general case without much difficulty. In this manuscript, we will be concerned with the subwavelength resonance frequencies setting where δ → 0 \delta\to0 δ → 0 . To this end, we define subwavelength resonance frequencies as follows:
Definition 2.1. Given δ > 0 \delta>0 δ > 0 , a subwavelength resonance frequency ω ( δ ) ∈ C \omega(\delta)\in\mathbb{C} ω ( δ ) ∈ C is such that
(2.2) has a nontrivial solution u u u ,
ω ( δ ) \omega(\delta) ω ( δ ) is continuous and satisfies ω ( δ ) = O ( δ ) \omega(\delta)=O(\sqrt{\delta}) ω ( δ ) = O ( δ ) .
This regime is physically interesting, because the wavelength of the associated eigenmodes can be magnitudes larger than the resonators themselves. The perhaps most classical example of this are so-called Minnaert Resonances, which are the resonant frequencies of air bubbles in water [6 , 24 ] . This peculiar property has made the subwavelength regime an attractive platform for metamaterial design [14 , 17 ] .
In studying the subwavelength setting, the capacitance matrix approach has proven to be very fruitful, for a thorough introduction we refer the reader to [4 ] . To introduce the capacitance matrix, we define V i V_i V i to be the solution to
Δ V i = 0 , in Ω , V i = δ i j on ∂ D j , V i ( x ) = O ( ∣ x ∣ − 1 ) . (3) \begin{aligned}
\Delta V_i&=0, &&\text{in }\Omega,\\
V_i&=\delta_{ij} &&\text{on }\partial D_j,\\
V_i(x)&=O(|x|^{-1}). \tag*{(3)}
\end{aligned} Δ V i V i V i ( x ) = 0 , = δ ij = O ( ∣ x ∣ − 1 ) . in Ω , on ∂ D j , (3)
The Maximum Principle implies that V i V_i V i is unique if it exists.
Remark 2.2. In previous literature, the existence of V i V_i V i has at times been a delicate issue. In the seminal work [6 ] , layer potential techniques were used. This approach however only works when ∣ I ∣ < ∞ |\mathcal{I}|<\infty ∣ I ∣ < ∞ , as otherwise the single layer potential may fail to be invertible [4 ] . In more recent literature [11 , 12 ] , it was defined through the Dirichlet-to-Neumann operator T ω \mathcal{T}^{\omega} T ω .
One now defines the capacitance matrix as
C i j = − ∫ ∂ D j ∂ ν V i d σ , i , j ∈ I . (4) C_{ij}=-\int_{\partial D_j}\partial_{\nu}V_i\,\mathrm{d}\sigma,\qquad i,j\in\mathcal{I}. \tag*{(4)} C ij = − ∫ ∂ D j ∂ ν V i d σ , i , j ∈ I . (4)
We will further need the coefficient matrix V = diag ( v i 2 ∣ D i ∣ ) \mathcal{V}=\operatorname{diag}\left(\frac{v_i^2}{|D_i|}\right) V = diag ( ∣ D i ∣ v i 2 ) , where ∣ ⋅ ∣ |\cdot| ∣ ⋅ ∣ denotes the usual Lebesgue measure.
Theorem 2.3 (Fundamental Theorem of Subwavelength Physics). Assume that ∣ I ∣ < ∞ |\mathcal{I}|<\infty ∣ I ∣ < ∞ . Then the subwavelength resonance problem (2) admits ∣ I ∣ |\mathcal{I}| ∣ I ∣ resonant frequencies (with positive real parts) that satisfy
ω n = δ λ n + O ( δ ) , (5) \omega_n=\sqrt{\delta\lambda_n}+O(\delta), \tag*{(5)} ω n = δ λ n + O ( δ ) , (5) with associated eigenmode
u n ( x ) = ∑ i ∈ I ( u n ) i V i ( x ) + O ( δ ) , (6) u_n(x)=\sum_{i\in\mathcal{I}}(\mathbf{u}_n)_iV_i(x)+O(\sqrt{\delta}), \tag*{(6)} u n ( x ) = i ∈ I ∑ ( u n ) i V i ( x ) + O ( δ ) , (6) where ( λ n , u n ) (\lambda_n,\mathbf{u}_n) ( λ n , u n ) is the n n n th eigenvalue-eigenvector pair of V C \mathcal{V}\mathcal{C} V C .
Remark 2.4. If ∣ I ∣ = ∞ |\mathcal{I}|=\infty ∣ I ∣ = ∞ , the situation is more complicated. If the underlying resonator configuration is periodic, one approach is to Floquet-Bloch transform Equation (2) directly and study the resulting problem using quasiperiodic Green’s functions. This strategy is of course limited to the periodic setting. While one can define the capacitance matrix in the disordered setting (in the crystal case by inverting the Laplacian [11 , 12 ] or alternatively by using the coercivity of the Laplacian in Beppo-Levi spaces), a rigorous reduction will be proven in a future paper.
Our main results will concern the off-diagonal decay of the capacitance matrix. Estimates on this decay have direct implications for the decay of defect modes, as well as spectral convergence of finite systems.
In what follows we will assume D i = B r i ( x i ) D_i=B_{r_i}(x_i) D i = B r i ( x i ) , D = ⋃ i ∈ I D i \mathcal{D}=\bigcup_{i\in\mathcal{I}}D_i D = ⋃ i ∈ I D i , where { r i } i ∈ I ⊂ R + \{r_i\}_{i\in\mathcal{I}}\subset\mathbb{R}_+ { r i } i ∈ I ⊂ R + and { x i } i ∈ I ⊂ R 3 \{x_i\}_{i\in\mathcal{I}}\subset\mathbb{R}^3 { x i } i ∈ I ⊂ R 3 satisfy
0 < r min ≤ r i ≤ r max < ∞ , (7) 0<r_{\min}\le r_i\le r_{\max}<\infty, \tag*{(7)} 0 < r m i n ≤ r i ≤ r m a x < ∞ , (7)
for some r max ≥ r min > 0 r_{\max}\ge r_{\min}>0 r m a x ≥ r m i n > 0 and, for some constant d min d_{\min} d m i n ,
0 < d min ≤ inf i ≠ j dist ( D i , D j ) . (8) 0<d_{\min}\le\inf_{i\ne j}\operatorname{dist}(D_i,D_j). \tag*{(8)} 0 < d m i n ≤ i = j inf dist ( D i , D j ) . (8)
Throughout, we write z = ( z ⊥ , z ∥ ) ∈ R 2 × R z=(z_\perp,z_\parallel)\in\mathbb{R}^2\times\mathbb{R} z = ( z ⊥ , z ∥ ) ∈ R 2 × R for points of R 3 \mathbb{R}^3 R 3 and, likewise, B = ( B ⊥ , B ∥ ) B=(B_\perp,B_\parallel) B = ( B ⊥ , B ∥ ) for the Brownian motion of § 3.1 .
Our first main result concerns chains of spherical resonators. A first result in this direction has recently been proven in [2 ] for a periodic chain whose fundamental cell contains a single resonator. We extend this result to a much larger class of chain-like geometries.
Theorem 2.5 (Chain decay). Let I = Z \mathcal{I}=\mathbb{Z} I = Z and write x i = ( a i , s i ) ∈ R 2 × R x_i=(a_i,s_i)\in\mathbb{R}^2\times\mathbb{R} x i = ( a i , s i ) ∈ R 2 × R for the centres of the resonators. Assume that additionally to assumptions (7) and (8) there exist L , ρ > 0 L,\rho>0 L , ρ > 0 such that
∣ a i ∣ + r i ≤ ρ and 0 ≤ s i + 1 − s i ≤ L . (9) |a_i|+r_i\le\rho\quad\text{and}\quad0\le s_{i+1}-s_i\le L. \tag*{(9)} ∣ a i ∣ + r i ≤ ρ and 0 ≤ s i + 1 − s i ≤ L . (9) Then there exist constants C 1 ≥ c 1 > 0 C_1\ge c_1>0 C 1 ≥ c 1 > 0 , depending only on r min , r max , d min , ρ , L r_{\min},r_{\max},d_{\min},\rho,L r m i n , r m a x , d m i n , ρ , L , such that the following bounds hold
c 1 d i j log 2 ( 2 + d i j ) ≤ ∣ C i j ∣ ≤ C 1 d i j log 2 ( 2 + d i j ) , (10) \frac{c_1}{d_{ij}\log^2(2+d_{ij})}\le|C_{ij}|\le\frac{C_1}{d_{ij}\log^2(2+d_{ij})}, \tag*{(10)} d ij log 2 ( 2 + d ij ) c 1 ≤ ∣ C ij ∣ ≤ d ij log 2 ( 2 + d ij ) C 1 , (10) for all distinct i , j ∈ I i,j\in\mathcal{I} i , j ∈ I , where d i j = ∣ x i − x j ∣ d_{ij}=|x_i-x_j| d ij = ∣ x i − x j ∣ .
Equation (9) implies that all resonators are contained in an infinite cylinder of radius ρ \rho ρ along the third coordinate axis. Moreover, as the balls D i D_i D i are disjoint with radii at least r min r_{\min} r m i n , only finitely many of them meet a bounded part of this cylinder, so that s i → + ∞ s_i\to+\infty s i → + ∞ as i → + ∞ i\to+\infty i → + ∞ ; together with 0 ≤ s i + 1 − s i ≤ L 0\le s_{i+1}-s_i\le L 0 ≤ s i + 1 − s i ≤ L this shows that every point of the axis lies within distance L L L of some s i s_i s i . For a visualisation of this setup, see Figure 1 , where Z k = { z : ∣ z ∥ ∣ < 2 k } Z_k=\{z:|z_\parallel|<2^k\} Z k = { z : ∣ z ∥ ∣ < 2 k } denotes a cylinder whose boundary is given by ∂ Z k = { z : ∣ z ∥ ∣ = 2 k } \partial Z_k=\{z:|z_\parallel|=2^k\} ∂ Z k = { z : ∣ z ∥ ∣ = 2 k } .
Figure 1. A schematic axial view after scaling to ρ ≤ 1 / 2 \rho\le1/2 ρ ≤ 1/2 (Theorem 2.8).
Our second main result concerns screens of resonators. To the best of our knowledge, no prior results exist for this setup, which is illustrated in Figure 2 .
Figure 2. Schematic geometry of a screen of finite thickness. Left: a portion of the projected centre set. Right: representative resonators lie in ∣ z ∥ ∣ ≤ ρ \lvert z_{\parallel}\rvert\le\rho ∣ z ∥ ∣ ≤ ρ , with protective shells about D i , D j D_i,D_j D i , D j . The points z k = x k + 3 n z_k = x_k + 3n z k = x k + 3 n , n = ( 0 , 0 , 1 ) n = (0,0,1) n = ( 0 , 0 , 1 ) , are the centres of the balls Q k Q_k Q k of Theorem 6.1, which lie in the halfspace H = { z ∥ > 2 } \mathcal{H} = \{z_{\parallel} > 2\} H = { z ∥ > 2 } used for the lower bound (after scaling to ρ ≤ 1 / 2 \rho\le1/2 ρ ≤ 1/2 ; not to scale).
Theorem 2.6 (Screen decay). Let I = Z \mathcal{I} = \mathbb{Z} I = Z and again write x i = ( a i , s i ) ∈ R 2 × R \boldsymbol{x}_i = (a_i,s_i) \in\mathbb{R}^2 \times\mathbb{R} x i = ( a i , s i ) ∈ R 2 × R for the centres of the resonators. Assume that additionally to assumptions (2.7) and (2.8) there exist L , ρ > 0 L,\rho> 0 L , ρ > 0 such that
∣ s i ∣ + r i ≤ ρ and sup w ∈ R 2 inf i ∈ I ∣ a i − w ∣ ≤ L . (11) \lvert s_i\rvert+ r_i \le\rho\quad\text{and}\quad\sup_{w\in\mathbb{R}^2}\inf_{i\in\mathcal{I}}\lvert a_i-w\rvert\le L. \tag*{(11)} ∣ s i ∣ + r i ≤ ρ and w ∈ R 2 sup i ∈ I inf ∣ a i − w ∣ ≤ L . (11) Then there exist constants C 2 ≥ c 2 > 0 C_2 \ge c_2 > 0 C 2 ≥ c 2 > 0 , depending only on r min , r max , d min , ρ , L r_{\min},r_{\max},d_{\min},\rho,L r m i n , r m a x , d m i n , ρ , L , such that the following bounds hold
c 2 d i j − 3 ≤ ∣ C i j ∣ ≤ C 2 d i j − 3 , (12) c_2d_{ij}^{-3} \le\lvert C_{ij}\rvert\le C_2d_{ij}^{-3}, \tag*{(12)} c 2 d ij − 3 ≤ ∣ C ij ∣ ≤ C 2 d ij − 3 , (12) for all distinct i , j ∈ I i,j \in\mathcal{I} i , j ∈ I , where again d i j = ∣ x i − x j ∣ d_{ij} = \lvert\boldsymbol{x}_i-\boldsymbol{x}_j\rvert d ij = ∣ x i − x j ∣ .
Equation (2.11) implies that all resonators are contained in a slab of width 2 ρ 2\rho 2 ρ around the plane P = { z : z ∥ = 0 } P = \{z : z_{\parallel} = 0\} P = { z : z ∥ = 0 } and that every point of P P P lies within distance L L L of some projected centre a i a_i a i , see figure 1 for illustration.
Finally, we are going to provide a novel proof of exponential decay for crystals of resonators, based on the same strategies employed in the proofs of Theorem 2.5 and Theorem 2.6 , therefore showing that our new probabilistic approach presents a unified framework, not limited to certain geometries.
Theorem 2.7 (Crystal decay). Let I = Z \mathcal{I} = \mathbb{Z} I = Z . Assume that additionally to assumptions (2.7) and (2.8) there exists L > 0 L > 0 L > 0 such that
sup z ∈ R 3 inf i ∈ I ∣ x i − z ∣ ≤ L . (13) \sup_{z\in\mathbb{R}^3}\inf_{i\in\mathcal{I}}\lvert x_i-z\rvert\le L. \tag*{(13)} z ∈ R 3 sup i ∈ I inf ∣ x i − z ∣ ≤ L . (13)
Then there exist constants C 3 , c 3 , C 4 , c 4 > 0 C_3,c_3,C_4,c_4>0 C 3 , c 3 , C 4 , c 4 > 0 , depending only on r min , r max , d min , L r_{\min},r_{\max},d_{\min},L r m i n , r m a x , d m i n , L , such that the following bound holds
C 4 exp ( − c 4 d i j ) ≤ ∣ C i j ∣ ≤ C 3 exp ( − c 3 d i j ) , (14) C_4\exp(-c_4d_{ij})\leq|C_{ij}|\leq C_3\exp(-c_3d_{ij}), \tag*{(14)} C 4 exp ( − c 4 d ij ) ≤ ∣ C ij ∣ ≤ C 3 exp ( − c 3 d ij ) , (14)
where d i j = ∣ x i − x j ∣ d_{ij}=|x_i-x_j| d ij = ∣ x i − x j ∣ .
In the proofs of Theorems 2.5 and 2.6 , we will make use of the following scaling property, which allows us to restrict ourselves to the case ρ ≤ 1 / 2 \rho\leq1/2 ρ ≤ 1/2 , that is, to configurations contained in the cylinder Z 0 Z_0 Z 0 , respectively the slab { z : ∣ z ∥ ∣ < 1 } \{z:|z_{\parallel}|<1\} { z : ∣ z ∥ ∣ < 1 } .
Lemma 2.8 (Scaling). Let λ > 0 \lambda>0 λ > 0 and denote by V i λ V_i^\lambda V i λ and C λ C^\lambda C λ the potentials and the capacitance matrix of the rescaled configuration D i λ = B λ r i ( λ x i ) D_i^\lambda=B_{\lambda r_i}(\lambda x_i) D i λ = B λ r i ( λ x i ) , i ∈ I i\in\mathcal{I} i ∈ I . Then
V i λ ( λ z ) = V i ( z ) for z ∈ Ω and C i j λ = λ C i j for i , j ∈ I . (15) V_i^\lambda(\lambda z)=V_i(z)\quad\text{for }z\in\Omega\qquad\text{and}\qquad C_{ij}^\lambda=\lambda C_{ij}\quad\text{for }i,j\in\mathcal{I}. \tag*{(15)} V i λ ( λ z ) = V i ( z ) for z ∈ Ω and C ij λ = λ C ij for i , j ∈ I . (15) In particular, it suffices to prove Theorems 2.5 and 2.6 under the additional assumption ρ ≤ 1 / 2 \rho\leq1/2 ρ ≤ 1/2 .
Proof. The dilated configuration satisfies (7) and (8) with λ r min , λ r max , λ d min \lambda r_{\min},\lambda r_{\max},\lambda d_{\min} λ r m i n , λ r m a x , λ d m i n in place of r min , r max , d min r_{\min},r_{\max},d_{\min} r m i n , r m a x , d m i n , and its exterior domain is λ Ω \lambda\Omega λ Ω . The function z ↦ V i λ ( λ z ) z\mapsto V_i^\lambda(\lambda z) z ↦ V i λ ( λ z ) is harmonic in Ω \Omega Ω , equals δ i j \delta_{ij} δ ij on ∂ D j \partial D_j ∂ D j and is O ( ∣ z ∣ − 1 ) O(|z|^{-1}) O ( ∣ z ∣ − 1 ) , hence it coincides with V i V_i V i by the uniqueness of the solution to (3) . Differentiating, we obtain ∂ ν V i λ ( λ y ) = λ − 1 ∂ ν V i ( y ) \partial_\nu V_i^\lambda(\lambda y)=\lambda^{-1}\partial_\nu V_i(y) ∂ ν V i λ ( λ y ) = λ − 1 ∂ ν V i ( y ) for y ∈ ∂ D j y\in\partial D_j y ∈ ∂ D j . As the surface measure of ∂ D i λ \partial D_i^\lambda ∂ D i λ is λ 2 \lambda^2 λ 2 times that of ∂ D j \partial D_j ∂ D j , Equation (4) yields C i j λ = λ C i j C_{ij}^\lambda=\lambda C_{ij} C ij λ = λ C ij .
Preliminaries
Brownian motion and stopping times
Throughout, B B B denotes a standard Brownian motion in R n \mathbb{R}^n R n . We write P z \mathbb{P}_z P z and E z \mathbb{E}_z E z for probability and expectation when B ( 0 ) = z B(0)=z B ( 0 ) = z . We use the canonical path space
Ω 0 = C ( [ 0 , ∞ ) ; R n ) \Omega_0=C([0,\infty);\mathbb{R}^n) Ω 0 = C ([ 0 , ∞ ) ; R n )
and the right-continuous filtration ( F + ( t ) ) t ≥ 0 (\mathcal{F}^+(t))_{t\geq0} ( F + ( t ) ) t ≥ 0 of [25 Sections 1.1 and 2.1–2.2] . All stopping times are taken with respect to this filtration.
For a closed set H H H , we define the stopping time
τ ( H ) = inf { t ≥ 0 : B ( t ) ∈ H } , inf ∅ = ∞ . \tau(H)=\inf\{t\geq0:B(t)\in H\},\qquad\inf\varnothing=\infty. τ ( H ) = inf { t ≥ 0 : B ( t ) ∈ H } , inf ∅ = ∞.
We will often write τ : = τ ( ∂ D ) \tau:=\tau(\partial D) τ := τ ( ∂ D ) , where the intention is clear. Hitting times of closed sets are stopping times, see [25 Remark 2.14] or [15 Proposition 1.15] .
We use the strong Markov property in following form [15 Theorem 1.2] :
Theorem 3.1. Let T T T be a stopping time, E ∈ F + ( T ) E\in\mathcal{F}^+(T) E ∈ F + ( T ) and f : C ( [ 0 , ∞ ) ; R n ) → [ 0 , ∞ ] f:C([0,\infty);\mathbb{R}^n)\to[0,\infty] f : C ([ 0 , ∞ ) ; R n ) → [ 0 , ∞ ] be measurable. Then it holds that
E z [ 1 E ∩ { T < ∞ } f ( B ( T + ⋅ ) ) ] = E z [ 1 E ∩ { T < ∞ } E B ( T ) [ f ( B ) ] ] . (16) \mathbb{E}_z\left[\mathbf{1}_{E\cap\{T<\infty\}}f(B(T+\cdot))\right]=\mathbb{E}_z\left[\mathbf{1}_{E\cap\{T<\infty\}}\mathbb{E}_{B(T)}[f(B)]\right]. \tag*{(16)} E z [ 1 E ∩ { T < ∞ } f ( B ( T + ⋅ )) ] = E z [ 1 E ∩ { T < ∞ } E B ( T ) [ f ( B )] ] . (16)
For a measurable functional F F F of paths we write F ∘ θ T = F ( B ( T + ⋅ ) ) F\circ\theta_T=F(B(T+\cdot)) F ∘ θ T = F ( B ( T + ⋅ )) on { T < ∞ } \{T<\infty\} { T < ∞ } ; for instance, if T ′ = inf { t ≥ 0 : B ( t ) ∈ H } T'=\inf\{t\geq0:B(t)\in H\} T ′ = inf { t ≥ 0 : B ( t ) ∈ H } , then T ′ ∘ θ T = inf { t ≥ 0 : B ( T + t ) ∈ H } T'\circ\theta_T=\inf\{t\geq0:B(T+t)\in H\} T ′ ∘ θ T = inf { t ≥ 0 : B ( T + t ) ∈ H } .
Elementary estimates for Brownian motion
We first collect three estimates: a Gaussian displacement bound, an exit-time bound for a disc, and the logarithmic tail of a planar return time.
Let Φ ( x ) = 1 2 π ∫ x ∞ e − v 2 / 2 d v \Phi(x)=\frac{1}{\sqrt{2\pi}}\int_x^\infty e^{-v^2/2}\,\mathrm{d}v Φ ( x ) = 2 π 1 ∫ x ∞ e − v 2 /2 d v denote the standard Gaussian tail.
Lemma 3.2 (Gaussian estimates). ( a ) (a) ( a ) For x ≥ 0 x\geq0 x ≥ 0 , Φ ‾ ( x ) ≤ 1 2 e − x 2 / 2 \overline{\Phi}(x)\leq\frac{1}{2}e^{-x^2/2} Φ ( x ) ≤ 2 1 e − x 2 /2 .
( b ) (b) ( b ) Let W W W be a standard Brownian motion in R n \mathbb{R}^n R n , n ∈ N n\in\mathbb{N} n ∈ N . For s > 0 s>0 s > 0 and a > 0 a>0 a > 0 ,
P ( sup 0 ≤ u ≤ s ∣ W ( u ) ∣ ≥ a ) ≤ 2 n exp ( − a 2 2 n s ) . \mathbb{P}\left(\sup_{0\leq u\leq s}|W(u)|\geq a\right)\leq2n\exp\left(-\frac{a^2}{2ns}\right). P ( 0 ≤ u ≤ s sup ∣ W ( u ) ∣ ≥ a ) ≤ 2 n exp ( − 2 n s a 2 ) .
Proof. (a) As ( x + v ) 2 ≥ x 2 + v 2 (x+v)^2 \ge x^2+v^2 ( x + v ) 2 ≥ x 2 + v 2 for x , v ≥ 0 x,v \ge0 x , v ≥ 0 ,
Φ ‾ ( x ) = 1 2 π ∫ 0 ∞ e − ( x + v ) 2 / 2 d v ≤ e − x 2 / 2 2 π ∫ 0 ∞ e − v 2 / 2 d v = 1 2 e − x 2 / 2 . \overline{\Phi}(x)=\frac{1}{\sqrt{2\pi}}\int_0^\infty e^{-(x+v)^2/2}\,\mathrm{d}v \le\frac{e^{-x^2/2}}{\sqrt{2\pi}}\int_0^\infty e^{-v^2/2}\,\mathrm{d}v=\frac{1}{2}e^{-x^2/2}. Φ ( x ) = 2 π 1 ∫ 0 ∞ e − ( x + v ) 2 /2 d v ≤ 2 π e − x 2 /2 ∫ 0 ∞ e − v 2 /2 d v = 2 1 e − x 2 /2 .
(b) Let X X X be a standard linear Brownian motion and M ( s ) = max u ≤ s X ( u ) M(s)=\max_{u\le s}X(u) M ( s ) = max u ≤ s X ( u ) . For 0 < b ′ < b 0<b'<b 0 < b ′ < b , [25 Theorem 2.21] gives P ( M ( s ) ≥ b ) ≤ P ( M ( s ) > b ′ ) = 2 P ( X ( s ) > b ′ ) \mathbb{P}(M(s)\ge b)\le\mathbb{P}(M(s)>b')=2\mathbb{P}(X(s)>b') P ( M ( s ) ≥ b ) ≤ P ( M ( s ) > b ′ ) = 2 P ( X ( s ) > b ′ ) ; letting b ′ ↑ b b'\uparrow b b ′ ↑ b and using (a), P ( M ( s ) ≥ b ) ≤ 2 Φ ‾ ( b / s ) ≤ e − b 2 / ( 2 s ) \mathbb{P}(M(s)\ge b)\le2\overline{\Phi}(b/\sqrt{s})\le e^{-b^2/(2s)} P ( M ( s ) ≥ b ) ≤ 2 Φ ( b / s ) ≤ e − b 2 / ( 2 s ) . Applying this also to − X -X − X , P ( sup u ≤ s ∣ X ( u ) ∣ ≥ b ) ≤ 2 e − b 2 / ( 2 s ) \mathbb{P}(\sup_{u\le s}|X(u)|\ge b)\le2e^{-b^2/(2s)} P ( sup u ≤ s ∣ X ( u ) ∣ ≥ b ) ≤ 2 e − b 2 / ( 2 s ) . If max u ≤ s ∣ W k ( u ) ∣ < a / n \max_{u\le s}|W_k(u)|<a/\sqrt{n} max u ≤ s ∣ W k ( u ) ∣ < a / n for all coordinates k k k , then max u ≤ s ∣ W ( u ) ∣ 2 ≤ ∑ k max u ≤ s W k ( u ) 2 < a 2 \max_{u\le s}|W(u)|^2\le\sum_k\max_{u\le s}W_k(u)^2<a^2 max u ≤ s ∣ W ( u ) ∣ 2 ≤ ∑ k max u ≤ s W k ( u ) 2 < a 2 ; a union bound over the n n n coordinates gives the claim. ■
Lemma 3.3 (Exit from discs). Let W W W be a planar Brownian motion, R > 0 R>0 R > 0 and T R = inf { t ≥ 0 : ∣ W ( t ) ∣ ≥ R } T_R=\inf\{t\ge0:|W(t)|\ge R\} T R = inf { t ≥ 0 : ∣ W ( t ) ∣ ≥ R } . Then P w ( T R > a ) ≤ min { 1 , 9 R 2 / a } \mathbb{P}_w(T_R>a)\le\min\{1,9R^2/a\} P w ( T R > a ) ≤ min { 1 , 9 R 2 / a } for all w ∈ R 2 w\in\mathbb{R}^2 w ∈ R 2 and a > 0 a>0 a > 0 . In particular T R < ∞ T_R<\infty T R < ∞ almost surely.
Proof. Let h = 3 R 2 h=3R^2 h = 3 R 2 . For every v ∈ R 2 v\in\mathbb{R}^2 v ∈ R 2 the first coordinate of W ( h ) W(h) W ( h ) under P v \mathbb{P}_v P v has density bounded by ( 2 π h ) − 1 / 2 (2\pi h)^{-1/2} ( 2 π h ) − 1/2 , so P v ( ∣ W ( h ) ∣ < R ) ≤ 2 R ( 2 π h ) − 1 / 2 = 2 ( 6 π ) − 1 / 2 < 1 2 \mathbb{P}_v(|W(h)|<R)\le2R(2\pi h)^{-1/2}=2(6\pi)^{-1/2}<\frac12 P v ( ∣ W ( h ) ∣ < R ) ≤ 2 R ( 2 π h ) − 1/2 = 2 ( 6 π ) − 1/2 < 2 1 . Let E N = { ∣ W ( k h ) ∣ < R for 1 ≤ k ≤ N } E_N=\{|W(kh)|<R\text{ for }1\le k\le N\} E N = { ∣ W ( k h ) ∣ < R for 1 ≤ k ≤ N } and E 0 E_0 E 0 the sure event; E N − 1 E_{N-1} E N − 1 is determined by W ( h ) , … , W ( ( N − 1 ) h ) W(h),\ldots,W((N-1)h) W ( h ) , … , W (( N − 1 ) h ) and hence lies in F + ( ( N − 1 ) h ) \mathcal{F}^+((N-1)h) F + (( N − 1 ) h ) . By (3.1) at the time ( N − 1 ) h (N-1)h ( N − 1 ) h with E = E N − 1 E=E_{N-1} E = E N − 1 , P w ( E N ) = E w [ 1 E N − 1 P W ( ( N − 1 ) h ) ( ∣ W ( h ) ∣ < R ) ] ≤ 1 2 P w ( E N − 1 ) \mathbb{P}_w(E_N)=\mathbb{E}_w[1_{E_{N-1}}\mathbb{P}_{W((N-1)h)}(|W(h)|<R)]\le\frac12\mathbb{P}_w(E_{N-1}) P w ( E N ) = E w [ 1 E N − 1 P W (( N − 1 ) h ) ( ∣ W ( h ) ∣ < R )] ≤ 2 1 P w ( E N − 1 ) , so P w ( T R > N h ) ≤ P w ( E N ) ≤ 2 − N \mathbb{P}_w(T_R>Nh)\le\mathbb{P}_w(E_N)\le2^{-N} P w ( T R > N h ) ≤ P w ( E N ) ≤ 2 − N . With N = ⌊ a / h ⌋ N=\lfloor a/h\rfloor N = ⌊ a / h ⌋ and e − y ≤ 1 / y e^{-y}\le1/y e − y ≤ 1/ y ,
P w ( T R > a ) ≤ 2 − N ≤ 2 exp ( − a log 2 3 R 2 ) ≤ 6 R 2 a log 2 ≤ 9 R 2 a . \mathbb{P}_w(T_R>a)\le2^{-N}\le2\exp\left(-\frac{a\log2}{3R^2}\right)\le\frac{6R^2}{a\log2}\le\frac{9R^2}{a}. P w ( T R > a ) ≤ 2 − N ≤ 2 exp ( − 3 R 2 a log 2 ) ≤ a log 2 6 R 2 ≤ a 9 R 2 .
Lemma 3.4 (Planar excursions). Let W W W be a planar Brownian motion started in w w w with ∣ w ∣ = 2 |w|=2 ∣ w ∣ = 2 , and κ = inf { t ≥ 0 : ∣ W ( t ) ∣ ≤ 1 } \kappa=\inf\{t\ge0:|W(t)|\le1\} κ = inf { t ≥ 0 : ∣ W ( t ) ∣ ≤ 1 } . Then P w ( κ > a ) ≤ 12 / log ( 2 + a ) \mathbb{P}_w(\kappa>a)\le12/\log(2+a) P w ( κ > a ) ≤ 12/ log ( 2 + a ) for all a > 0 a>0 a > 0 ; in particular κ < ∞ \kappa<\infty κ < ∞ almost surely.
Proof. Let m ≥ 2 m\ge2 m ≥ 2 be an integer and T = T 2 m T=T_{2^m} T = T 2 m as in Theorem 3.3. Since 1 < ∣ w ∣ < 2 m 1<|w|<2^m 1 < ∣ w ∣ < 2 m , by continuity κ \kappa κ and T T T are the hitting times of the circles of radii 1 1 1 and 2 m 2^m 2 m , and T < ∞ T<\infty T < ∞ almost surely. By [25 Theorem 3.18] (planar case), P w ( T < κ ) = 1 − log 2 m − log 2 log 2 m = 1 m \mathbb{P}_w(T<\kappa)=1-\frac{\log2^m-\log2}{\log2^m}=\frac{1}{m} P w ( T < κ ) = 1 − l o g 2 m l o g 2 m − l o g 2 = m 1 . Since ∣ W ( T ) ∣ = 2 m ≠ 1 = ∣ W ( κ ) ∣ |W(T)|=2^m\ne1=|W(\kappa)| ∣ W ( T ) ∣ = 2 m = 1 = ∣ W ( κ ) ∣ whenever both are finite, { T = κ } \{T=\kappa\} { T = κ } is a null set, and { κ > a } ⊂ { T < κ } ∪ { T > a } \{\kappa>a\}\subset\{T<\kappa\}\cup\{T>a\} { κ > a } ⊂ { T < κ } ∪ { T > a } up to this null set. By Theorem 3.3,
P w ( κ > a ) ≤ 1 m + 9 ⋅ 4 m a . \mathbb{P}_w(\kappa>a)\le\frac{1}{m}+\frac{9\cdot4^m}{a}. P w ( κ > a ) ≤ m 1 + a 9 ⋅ 4 m .
For a ≥ 2 16 a\ge2^{16} a ≥ 2 16 take m = ⌊ log 2 ( a ) / 4 ⌋ ≥ 4 m=\lfloor\log_2(a)/4\rfloor\ge4 m = ⌊ log 2 ( a ) /4 ⌋ ≥ 4 ; then 4 m ≤ a 1 / 2 4^m\le a^{1/2} 4 m ≤ a 1/2 and m ≥ log 2 ( a ) / 4 − 1 ≥ log 2 ( a ) / 8 m\ge\log_2(a)/4-1\ge\log_2(a)/8 m ≥ log 2 ( a ) /4 − 1 ≥ log 2 ( a ) /8 , so P w ( κ > a ) ≤ 8 log 2 / log a + 9 a − 1 / 2 \mathbb{P}_w(\kappa>a)\le8\log2/\log a+9a^{-1/2} P w ( κ > a ) ≤ 8 log 2/ log a + 9 a − 1/2 . Since v ↦ log ( v ) / v v\mapsto\log(v)/\sqrt{v} v ↦ log ( v ) / v decreases on [ e 2 , ∞ ) [e^2,\infty) [ e 2 , ∞ ) , log ( a ) / a ≤ 16 log 2 / 2 8 < 0.044 \log(a)/\sqrt{a}\le16\log2/2^8<0.044 log ( a ) / a ≤ 16 log 2/ 2 8 < 0.044 , hence 9 a − 1 / 2 ≤ 0.4 / log a 9a^{-1/2}\le0.4/\log a 9 a − 1/2 ≤ 0.4/ log a and P w ( κ > a ) ≤ 6 / log a ≤ 12 / log ( 2 + a ) \mathbb{P}_w(\kappa>a)\le6/\log a\le12/\log(2+a) P w ( κ > a ) ≤ 6/ log a ≤ 12/ log ( 2 + a ) , using 2 + a ≤ a 2 2+a\le a^2 2 + a ≤ a 2 . For 0 < a < 2 16 0<a<2^{16} 0 < a < 2 16 the bound is trivial, as 12 / log ( 2 + a ) > 12 / log ( 2 17 ) > 1 12/\log(2+a)>12/\log(2^{17})>1 12/ log ( 2 + a ) > 12/ log ( 2 17 ) > 1 . ■
Exterior harmonic extension
In this subsection, we will prove an adaption of [15 Theorem 1.23] to exterior domains for functions with suitable decay.
Theorem 3.5 (Exterior harmonic extension). Let n ≥ 3 n\ge3 n ≥ 3 and let { D i } i ∈ I \{D_i\}_{i\in I} { D i } i ∈ I be a locally finite, countable family of bounded Lipschitz domains in R n \mathbb{R}^n R n with pairwise disjoint closures. Set D = ⋃ i ∈ I D i D=\bigcup_{i\in I}D_i D = ⋃ i ∈ I D i and Ω = R n ∖ D ‾ \Omega=\mathbb{R}^n\setminus\overline{D} Ω = R n ∖ D . Write τ = τ ( ∂ D ) \tau=\tau(\partial D) τ = τ ( ∂ D ) . For any continuous function φ : ∂ D → R \varphi:\partial D\to\mathbb{R} φ : ∂ D → R satisfying
lim R → ∞ sup x ∈ ∂ D ∩ B R ( 0 ) c ∣ φ ( x ) ∣ = 0 (17) \lim_{R\to\infty}\sup_{x\in\partial D\cap B_R(0)^c}|\varphi(x)|=0 \tag*{(17)} R → ∞ lim x ∈ ∂ D ∩ B R ( 0 ) c sup ∣ φ ( x ) ∣ = 0 (17) the function defined by
u ( x ) = E x [ φ ( B ( τ ( ∂ D ) ) ) 1 { τ < ∞ } ] (18) u(x)=\mathbb{E}_x\left[\varphi(B(\tau(\partial D)))1_{\{\tau<\infty\}}\right] \tag*{(18)} u ( x ) = E x [ φ ( B ( τ ( ∂ D ))) 1 { τ < ∞ } ] (18) for x ∈ Ω x\in\Omega x ∈ Ω , extended by φ \varphi φ on ∂ Ω \partial\Omega ∂ Ω , is the unique bounded function in C ( Ω ‾ ) C(\overline{\Omega}) C ( Ω ) that is harmonic in Ω \Omega Ω , has boundary values φ \varphi φ , and vanishes at infinity.
Proof. The continuity of φ \varphi φ and its decay at infinity imply that it is bounded. Existence and continuity hence follows from [15 Theorem 1.23] since the Lipschitz condition ensures regularity of the boundary by the exterior cone criterion [15 Proposition 1.22] . Next, we show that u u u vanishes at infinity. For any J ⊂ I \mathcal{J} \subset\mathcal{I} J ⊂ I finite define
φ J ( x ) = { φ ( x ) , x ∈ ∂ D ∩ ⋃ i ∈ J ∂ D i , 0 , else . (19) \varphi^{\mathcal{J}}(x) =
\begin{cases}
\varphi(x), & x \in\partial D \cap\bigcup_{i\in\mathcal{J}} \partial D_i,\\
0, & \text{else}.
\end{cases} \tag*{(19)} φ J ( x ) = { φ ( x ) , 0 , x ∈ ∂ D ∩ ⋃ i ∈ J ∂ D i , else . (19)
Fix ε > 0 \varepsilon> 0 ε > 0 . By local finiteness and the decay of φ \varphi φ at infinity, choose a finite subset J ⊂ I \mathcal{J} \subset\mathcal{I} J ⊂ I such that ∣ φ − φ J ∣ < ε \lvert\varphi-\varphi^{\mathcal{J}}\rvert< \varepsilon ∣ φ − φ J ∣ < ε . Define
u J ( x ) = E x [ φ J ( B ( τ ( ∂ D ) ) ) 1 { τ < ∞ } ] . (20) u^{\mathcal{J}}(x) = \mathbb{E}_x\left[\varphi^{\mathcal{J}}(B(\tau(\partial D)))1_{\{\tau<\infty\}}\right]. \tag*{(20)} u J ( x ) = E x [ φ J ( B ( τ ( ∂ D ))) 1 { τ < ∞ } ] . (20)
Choose R ε R_\varepsilon R ε so that the support of φ J \varphi^{\mathcal{J}} φ J is contained in B R ε ( 0 ) B_{R_\varepsilon}(0) B R ε ( 0 ) . A path contributing to u J ( x ) u^{\mathcal{J}}(x) u J ( x ) must hit this ball. The Brownian hitting probability of a ball in dimension n ≥ 3 n \ge3 n ≥ 3 therefore gives, for ∣ x ∣ > R ε \lvert x\rvert> R_\varepsilon ∣ x ∣ > R ε ,
∣ u J ( x ) ∣ ≤ ∥ φ ∥ ∞ ( R ε ∣ x ∣ ) n − 2 . \lvert u^{\mathcal{J}}(x)\rvert\le\lVert\varphi\rVert_\infty\left(\frac{R_\varepsilon}{\lvert x\rvert}\right)^{n-2}. ∣ u J ( x )∣ ≤ ∥ φ ∥ ∞ ( ∣ x ∣ R ε ) n − 2 .
Since ∥ u − u J ∥ ∞ ≤ ε \lVert u-u^{\mathcal{J}}\rVert_\infty\le\varepsilon ∥ u − u J ∥ ∞ ≤ ε , it follows that
∣ u ( x ) ∣ ≤ ε + ∥ φ ∥ ∞ ( R ε ∣ x ∣ ) n − 2 . (21) \lvert u(x)\rvert\le\varepsilon+\lVert\varphi\rVert_\infty\left(\frac{R_\varepsilon}{\lvert x\rvert}\right)^{n-2}. \tag*{(21)} ∣ u ( x )∣ ≤ ε + ∥ φ ∥ ∞ ( ∣ x ∣ R ε ) n − 2 . (21)
Letting ∣ x ∣ → ∞ \lvert x\rvert\to\infty ∣ x ∣ → ∞ and then ε ↓ 0 \varepsilon\downarrow0 ε ↓ 0 proves the decay of u u u .
Finally, we show uniqueness. Let v v v be another solution with the same boundary values and decay at infinity. Then h = u − v h=u-v h = u − v is harmonic in the exterior domain, vanishes on ∂ D \partial D ∂ D , and tends to zero at infinity. Apply the maximum principle to h h h and − h -h − h on each component of ( R n ∖ D ‾ ) ∩ B R ( 0 ) (\mathbb{R}^n\setminus\overline{D})\cap B_R(0) ( R n ∖ D ) ∩ B R ( 0 ) , and let R → ∞ R\to\infty R → ∞ . ■
Lemma 3.6. Suppose the family { D i } i ∈ I \{D_i\}_{i\in\mathcal{I}} { D i } i ∈ I satisfies (2.7) and (2.8). Then { D i } i ∈ I \{D_i\}_{i\in\mathcal{I}} { D i } i ∈ I is locally finite and Ω \Omega Ω is connected. In particular, the assumptions of Theorem 3.5 are satisfied.
Proof. Because r i ≤ r max r_i \le r_{\max} r i ≤ r m a x , any D i D_i D i intersecting B R ( 0 ) B_R(0) B R ( 0 ) for some R > 0 R>0 R > 0 must be contained in B R + 2 r max ( 0 ) B_{R+2r_{\max}}(0) B R + 2 r m a x ( 0 ) . However, ∣ D i ∣ ≥ 4 π r min 3 / 3 \lvert D_i\rvert\ge4\pi r_{\min}^3/3 ∣ D i ∣ ≥ 4 π r m i n 3 /3 and hence any bounded set can contain only finitely many inclusions. To see that Ω \Omega Ω is connected, cover each D i D_i D i with B r i + d min / 3 ( x i ) B_{r_i+d_{\min}/3}(x_i) B r i + d m i n /3 ( x i ) . Let now x , y ∈ Ω x,y\in\Omega x , y ∈ Ω be two arbitrary points and γ \gamma γ the straight segment connecting them, which meets only finitely many of these balls. Whenever γ \gamma γ enters some B r i + d min / 3 ( x i ) B_{r_i+d_{\min}/3}(x_i) B r i + d m i n /3 ( x i ) , we may connect the entry and exit point by an arc on ∂ B r i + d min / 3 ( x i ) ⊂ Ω \partial B_{r_i+d_{\min}/3}(x_i)\subset\Omega ∂ B r i + d m i n /3 ( x i ) ⊂ Ω ; if x x x or y y y lies inside such a ball, we first move it radially to the sphere, which is possible as x , y ∉ D ‾ x,y\notin\overline{D} x , y ∈ / D . This yields a path γ ′ ⊂ Ω \gamma'\subset\Omega γ ′ ⊂ Ω . ■
We thus obtain the following Corollary
Corollary 3.7. Under the assumptions (2.7) and (2.8), the unique solution to the problem (2.3) is given by
V i ( x ) = E x [ 1 { τ < ∞ } 1 ∂ D i ( B ( τ ) ) ] = P x ( τ < ∞ , B ( τ ) ∈ ∂ D i ) . (22) V_i(x)=\mathbb{E}_x\left[1_{\{\tau<\infty\}}1_{\partial D_i}(B(\tau))\right]=\mathbb{P}_x(\tau<\infty,B(\tau)\in\partial D_i). \tag*{(22)} V i ( x ) = E x [ 1 { τ < ∞ } 1 ∂ D i ( B ( τ )) ] = P x ( τ < ∞ , B ( τ ) ∈ ∂ D i ) . (22)
As an immediate consequence, one obtains the following decay estimate.
Proposition 3.8. For every z ∈ Ω z\in\Omega z ∈ Ω ,
0 ≤ V i ( z ) ≤ r i ∣ z − x i ∣ . (23) 0\le V_i(z)\le\frac{r_i}{\lvert z-x_i\rvert}. \tag*{(23)} 0 ≤ V i ( z ) ≤ ∣ z − x i ∣ r i . (23) Proof. Let T = inf { t > 0 : ∣ B ( t ) − x i ∣ = r i } T=\inf\{t>0:\lvert B(t)-x_i\rvert=r_i\} T = inf { t > 0 : ∣ B ( t ) − x i ∣ = r i } . On { τ < ∞ , B ( τ ) ∈ ∂ D i } \{\tau<\infty,B(\tau)\in\partial D_i\} { τ < ∞ , B ( τ ) ∈ ∂ D i } we have T ≤ τ < ∞ T\le\tau<\infty T ≤ τ < ∞ . Since ∣ z − x i ∣ > r i \lvert z-x_i\rvert>r_i ∣ z − x i ∣ > r i , [25 Corollary 3.19] (three-dimensional case), applied to B − x i B-x_i B − x i , gives V i ( z ) ≤ P z ( T < ∞ ) = r i / ∣ z − x i ∣ V_i(z)\le\mathbb{P}_z(T<\infty)=r_i/\lvert z-x_i\rvert V i ( z ) ≤ P z ( T < ∞ ) = r i / ∣ z − x i ∣ . ■
From flux to hitting probabilities
The next lemma is the final step in going from estimating a flux to upper bounding hitting probabilities. First however, we need to define the open shells
A j δ = { z ∈ R 3 : r j < ∣ x j − z ∣ < δ + r j } . (24) A_j^\delta=\{z\in\mathbb{R}^3:r_j<|x_j-z|<\delta+r_j\}. \tag*{(24)} A j δ = { z ∈ R 3 : r j < ∣ x j − z ∣ < δ + r j } . (24)
Note for δ ∈ ( 0 , d min / 2 ] \delta\in(0,d_{\min}/2] δ ∈ ( 0 , d m i n /2 ] we have
A i δ ‾ ∩ D k ‾ = ∅ ( k ≠ i ) , A i δ ‾ ∩ D i ‾ = ∂ D i , \overline{A_i^\delta}\cap\overline{D_k}=\varnothing\ (k\ne i),\qquad\overline{A_i^\delta}\cap\overline{D_i}=\partial D_i, A i δ ∩ D k = ∅ ( k = i ) , A i δ ∩ D i = ∂ D i ,
in particular A i δ ⊂ Ω , ∂ B r i + δ ( x i ) ⊂ Ω . (25) \text{in particular }A_i^\delta\subset\Omega,\quad\partial B_{r_i+\delta}(x_i)\subset\Omega. \tag*{(25)} in particular A i δ ⊂ Ω , ∂ B r i + δ ( x i ) ⊂ Ω. (25)
Lemma 3.9 (Flux bound). Let j ∈ I j\in\mathcal{I} j ∈ I and δ ∈ ( 0 , d min / 2 ] \delta\in(0,d_{\min}/2] δ ∈ ( 0 , d m i n /2 ] .
(a) Let u u u be harmonic on A j δ A_j^\delta A j δ and M ≥ 0 M\geq0 M ≥ 0 with
0 ≤ u ( z ) ≤ M r j − 1 − ∣ z − x j ∣ − 1 r j − 1 − ( r j + δ ) − 1 , z ∈ A j δ . (26) 0\leq u(z)\leq M\frac{r_j^{-1}-|z-x_j|^{-1}}{r_j^{-1}-(r_j+\delta)^{-1}},\qquad z\in A_j^\delta. \tag*{(26)} 0 ≤ u ( z ) ≤ M r j − 1 − ( r j + δ ) − 1 r j − 1 − ∣ z − x j ∣ − 1 , z ∈ A j δ . (26) Then Ψ u ( ϱ ) = ∫ ∂ B ϱ ( x j ) ∂ r u d σ \Psi_u(\varrho)=\int_{\partial B_\varrho(x_j)}\partial_r u\,\mathrm{d}\sigma Ψ u ( ϱ ) = ∫ ∂ B ϱ ( x j ) ∂ r u d σ does not depend on ϱ ∈ ( r j , r j + δ ) \varrho\in(r_j,r_j+\delta) ϱ ∈ ( r j , r j + δ ) , and its value Ψ u \Psi_u Ψ u satisfies 0 ≤ Ψ u ≤ 4 π r j ( r j + δ ) M / δ 0\leq\Psi_u\leq4\pi r_j(r_j+\delta)M/\delta 0 ≤ Ψ u ≤ 4 π r j ( r j + δ ) M / δ .
(b) For i ≠ j i\ne j i = j , u = V i u=V_i u = V i satisfies (3.11) with M = M i j δ = sup { V i ( z ) : ∣ z − x j ∣ = r j + δ } M=M_{ij}^\delta=\sup\{V_i(z):|z-x_j|=r_j+\delta\} M = M ij δ = sup { V i ( z ) : ∣ z − x j ∣ = r j + δ } , and u = 1 − V j u=1-V_j u = 1 − V j satisfies (3.11) with M = 1 M=1 M = 1 .
Proof. (a) For r j < ϱ 1 < ϱ 2 < r j + δ r_j<\varrho_1<\varrho_2<r_j+\delta r j < ϱ 1 < ϱ 2 < r j + δ the divergence theorem on the closed shell { ϱ 1 ≤ ∣ z − x j ∣ ≤ ϱ 2 } ⊂ A j δ \{\varrho_1\leq|z-x_j|\leq\varrho_2\}\subset A_j^\delta { ϱ 1 ≤ ∣ z − x j ∣ ≤ ϱ 2 } ⊂ A j δ gives Ψ u ( ϱ 2 ) − Ψ u ( ϱ 1 ) = ∫ Δ u = 0 \Psi_u(\varrho_2)-\Psi_u(\varrho_1)=\int\Delta u=0 Ψ u ( ϱ 2 ) − Ψ u ( ϱ 1 ) = ∫ Δ u = 0 . This establishes (a) for ϱ ∈ ( r j , r j + δ ) \varrho\in(r_j,r_j+\delta) ϱ ∈ ( r j , r j + δ ) .
Let μ ( ϱ ) = 1 4 π ∫ ∂ B 1 ( 0 ) u ( x j + ϱ ω ) d σ ( ω ) \mu(\varrho)=\frac{1}{4\pi}\int_{\partial B_1(0)}u(x_j+\varrho\omega)\,\mathrm{d}\sigma(\omega) μ ( ϱ ) = 4 π 1 ∫ ∂ B 1 ( 0 ) u ( x j + ϱ ω ) d σ ( ω ) . Since ∇ u \nabla u ∇ u is bounded on closed shells contained in A j δ A_j^\delta A j δ , we may differentiate under the integral and get
μ ′ ( ϱ ) = 1 4 π ∫ ∂ B 1 ( 0 ) ∇ u ( x j + ϱ ω ) ⋅ ω d σ ( ω ) = Ψ u 4 π ϱ 2 , \mu'(\varrho)=\frac{1}{4\pi}\int_{\partial B_1(0)}\nabla u(x_j+\varrho\omega)\cdot\omega\,\mathrm{d}\sigma(\omega)=\frac{\Psi_u}{4\pi\varrho^2}, μ ′ ( ϱ ) = 4 π 1 ∫ ∂ B 1 ( 0 ) ∇ u ( x j + ϱ ω ) ⋅ ω d σ ( ω ) = 4 π ϱ 2 Ψ u ,
so μ ( ϱ ) = μ ( ϱ 0 ) + Ψ u 4 π ( ϱ 0 − 1 − ϱ − 1 ) \mu(\varrho)=\mu(\varrho_0)+\frac{\Psi_u}{4\pi}(\varrho_0^{-1}-\varrho^{-1}) μ ( ϱ ) = μ ( ϱ 0 ) + 4 π Ψ u ( ϱ 0 − 1 − ϱ − 1 ) for r j < ϱ 0 < ϱ < r j + δ r_j<\varrho_0<\varrho<r_j+\delta r j < ϱ 0 < ϱ < r j + δ . By (3.11), 0 ≤ μ ( ϱ 0 ) ≤ M ( r j − 1 − ϱ 0 − 1 ) / ( r j − 1 − ( r j + δ ) − 1 ) → 0 0\leq\mu(\varrho_0)\leq M(r_j^{-1}-\varrho_0^{-1})/(r_j^{-1}-(r_j+\delta)^{-1})\to0 0 ≤ μ ( ϱ 0 ) ≤ M ( r j − 1 − ϱ 0 − 1 ) / ( r j − 1 − ( r j + δ ) − 1 ) → 0 as ϱ 0 ↓ r j \varrho_0\downarrow r_j ϱ 0 ↓ r j , hence
μ ( ϱ ) = Ψ u 4 π ( 1 r j − 1 ϱ ) , ϱ ∈ ( r j , r j + δ ) . (27) \mu(\varrho)=\frac{\Psi_u}{4\pi}\left(\frac{1}{r_j}-\frac{1}{\varrho}\right),\qquad\varrho\in(r_j,r_j+\delta). \tag*{(27)} μ ( ϱ ) = 4 π Ψ u ( r j 1 − ϱ 1 ) , ϱ ∈ ( r j , r j + δ ) . (27)
As μ ≥ 0 \mu\geq0 μ ≥ 0 , Ψ u ≥ 0 \Psi_u\geq0 Ψ u ≥ 0 ; comparing (3.12) with the bound on μ ( ϱ ) \mu(\varrho) μ ( ϱ ) from (3.11) gives Ψ u ≤ 4 π M / ( r j − 1 − ( r j + δ ) − 1 ) = 4 π r j ( r j + δ ) M / δ \Psi_u\leq4\pi M/(r_j^{-1}-(r_j+\delta)^{-1})=4\pi r_j(r_j+\delta)M/\delta Ψ u ≤ 4 π M / ( r j − 1 − ( r j + δ ) − 1 ) = 4 π r j ( r j + δ ) M / δ .
To extend the statement to ϱ = r j \varrho=r_j ϱ = r j , fix δ ′ ∈ ( 0 , δ ) \delta'\in(0,\delta) δ ′ ∈ ( 0 , δ ) and set A = { z : r j < ∣ z − x j ∣ < r j + δ ′ } A=\{z:r_j<|z-x_j|<r_j+\delta'\} A = { z : r j < ∣ z − x j ∣ < r j + δ ′ } . By (3.11), u ∈ C 0 ( A ‾ ) ∩ C ∞ ( A ) u\in C^0(\overline{A})\cap C^\infty(A) u ∈ C 0 ( A ) ∩ C ∞ ( A ) with boundary values φ ∈ C ∞ ( A ‾ ) \varphi\in C^\infty(\overline{A}) φ ∈ C ∞ ( A ) , where φ \varphi φ vanishes near ∂ D j \partial D_j ∂ D j and coincides with u u u near the outer sphere, so that u ∈ C 2 , α ( A ‾ ) u\in C^{2,\alpha}(\overline{A}) u ∈ C 2 , α ( A ) by classical regularity theory [20 Theorem 6.19] . Hence ∇ u \nabla u ∇ u is continuous to the boundary, and Ψ u ( r j ) = lim ϱ ↓ r j Ψ u ( ϱ ) = Ψ u \Psi_u(r_j)=\lim_{\varrho\downarrow r_j}\Psi_u(\varrho)=\Psi_u Ψ u ( r j ) = lim ϱ ↓ r j Ψ u ( ϱ ) = Ψ u .
(b) Define
H j δ ( z ) = r j − 1 − ∣ z − x j ∣ − 1 r j − 1 − ( r j + δ ) − 1 . H_j^\delta(z)=\frac{r_j^{-1}-|z-x_j|^{-1}}{r_j^{-1}-(r_j+\delta)^{-1}}. H j δ ( z ) = r j − 1 − ( r j + δ ) − 1 r j − 1 − ∣ z − x j ∣ − 1 .
This function is harmonic on A j δ A_j^\delta A j δ , equals zero on the inner sphere, and equals one on the outer sphere. By Theorem 3.5, both V i V_i V i for i ≠ j i\ne j i = j and 1 − V j 1-V_j 1 − V j are harmonic on the shell, continuous on its closure, and zero on the inner sphere. On the outer sphere they are bounded above by M i j δ M_{ij}^\delta M ij δ and 1 1 1 , respectively. The maximum principle applied to V i − M i j δ H j δ V_i-M_{ij}^\delta H_j^\delta V i − M ij δ H j δ and to 1 − V j − H j δ 1-V_j-H_j^\delta 1 − V j − H j δ proves (3.11) in both cases. ■
Consequently, applying Theorem 3.9(a) with the constants from Theorem 3.9(b) to (4) , we thus obtain
∣ C i j ∣ ≤ 4 π r j ( r j + δ ) δ M i j δ , i ≠ j , (28) |C_{ij}|\leq\frac{4\pi r_j(r_j+\delta)}{\delta}M_{ij}^\delta,\qquad i\ne j, \tag*{(28)} ∣ C ij ∣ ≤ δ 4 π r j ( r j + δ ) M ij δ , i = j , (28)
thereby successfully going from having to estimate the flux to having to estimate hitting probabilities on a shell around the inclusions. As an immediate application we recover the classical 1 / d i j 1/d_{ij} 1/ d ij decay and can control the effect of removing distant resonators.
Corollary 3.10. For i ≠ j i \ne j i = j and δ ∈ ( 0 , d min / 2 ] \delta\in(0,d_{\min}/2] δ ∈ ( 0 , d m i n /2 ] ,
∣ C i j ∣ ≤ 4 π r i r j ( r j + δ ) δ 1 ∣ x i − x j ∣ − ( r j + δ ) . (29) |C_{ij}| \le\frac{4\pi r_i r_j(r_j+\delta)}{\delta}\frac{1}{|x_i-x_j|-(r_j+\delta)}. \tag*{(29)} ∣ C ij ∣ ≤ δ 4 π r i r j ( r j + δ ) ∣ x i − x j ∣ − ( r j + δ ) 1 . (29)
Proof. Since dist ( D i , D j ) = ∣ x i − x j ∣ − r i − r j \operatorname{dist}(D_i,D_j)=|x_i-x_j|-r_i-r_j dist ( D i , D j ) = ∣ x i − x j ∣ − r i − r j , (2.8) gives ∣ x i − x j ∣ ≥ r i + r j + d min |x_i-x_j|\ge r_i+r_j+d_{\min} ∣ x i − x j ∣ ≥ r i + r j + d m i n , so every z z z with ∣ z − x j ∣ = r j + δ |z-x_j|=r_j+\delta ∣ z − x j ∣ = r j + δ satisfies ∣ z − x i ∣ ≥ ∣ x i − x j ∣ − r j − δ ≥ r i + d min − δ > 0 |z-x_i|\ge|x_i-x_j|-r_j-\delta\ge r_i+d_{\min}-\delta>0 ∣ z − x i ∣ ≥ ∣ x i − x j ∣ − r j − δ ≥ r i + d m i n − δ > 0 . By Theorem 3.8, M i j δ ≤ r i / ( ∣ x i − x j ∣ − r j − δ ) M_{ij}^{\delta}\le r_i/(|x_i-x_j|-r_j-\delta) M ij δ ≤ r i / ( ∣ x i − x j ∣ − r j − δ ) , and (3.13) gives the claimed bound. ■
Proposition 3.11. Let ∅ ≠ J ⊂ I \varnothing\ne\mathcal{J}\subset\mathcal{I} ∅ = J ⊂ I , and denote by C J C^{\mathcal{J}} C J and C I = C C^{\mathcal{I}}=C C I = C the capacitance matrices (2.4) of D J = ⋃ k ∈ J D k D^{\mathcal{J}}=\bigcup_{k\in\mathcal{J}}D_k D J = ⋃ k ∈ J D k and D I = D D^{\mathcal{I}}=D D I = D . Let i , j ∈ J i,j\in\mathcal{J} i , j ∈ J and R = inf m ∈ I ∖ J dist ( x i , D m ) R=\inf_{m\in\mathcal{I}\setminus\mathcal{J}}\operatorname{dist}(x_i,D_m) R = inf m ∈ I ∖ J dist ( x i , D m ) (with R = ∞ R=\infty R = ∞ if J = I \mathcal{J}=\mathcal{I} J = I ). Then for every δ ∈ ( 0 , d min / 2 ] \delta\in(0,d_{\min}/2] δ ∈ ( 0 , d m i n /2 ]
0 ≤ C i j I − C i j J ≤ 4 π r i r j ( r j + δ ) δ R . (30) 0\le C_{ij}^{\mathcal{I}}-C_{ij}^{\mathcal{J}}\le\frac{4\pi r_i r_j(r_j+\delta)}{\delta R}. \tag*{(30)} 0 ≤ C ij I − C ij J ≤ δ R 4 π r i r j ( r j + δ ) . (30)
Proof. The case J = I \mathcal{J}=\mathcal{I} J = I is immediate. Otherwise, set h = V i J − V i I h=V_i^{\mathcal{J}}-V_i^{\mathcal{I}} h = V i J − V i I on Ω I \Omega^{\mathcal{I}} Ω I . By Theorem 3.5, h h h is harmonic in Ω I \Omega^{\mathcal{I}} Ω I , continuous on its closure, and vanishes at infinity.
On a retained sphere ∂ D k \partial D_k ∂ D k , k ∈ J k\in\mathcal{J} k ∈ J , both potentials have the same boundary value, so h = 0 h=0 h = 0 . On an omitted sphere ∂ D m \partial D_m ∂ D m , m ∉ J m\notin\mathcal{J} m ∈ / J , we have V i I = 0 V_i^{\mathcal{I}}=0 V i I = 0 . The one-ball bound of Theorem 3.8, applied to the family J \mathcal{J} J , gives
0 ≤ h ( w ) = V i J ( w ) ≤ r i ∣ w − x i ∣ ≤ r i R , w ∈ ∂ D m . 0\le h(w)=V_i^{\mathcal{J}}(w)\le\frac{r_i}{|w-x_i|}\le\frac{r_i}{R},\qquad w\in\partial D_m. 0 ≤ h ( w ) = V i J ( w ) ≤ ∣ w − x i ∣ r i ≤ R r i , w ∈ ∂ D m .
Apply the maximum principle to h h h and − h -h − h on the components of Ω I ∩ B ϱ ( 0 ) \Omega^{\mathcal{I}}\cap B_\varrho(0) Ω I ∩ B ϱ ( 0 ) , and let ϱ → ∞ \varrho\to\infty ϱ → ∞ . Since h h h vanishes at infinity, this yields 0 ≤ h ≤ r i / R 0\le h\le r_i/R 0 ≤ h ≤ r i / R throughout Ω I \Omega^{\mathcal{I}} Ω I .
On A j δ A_j^\delta A j δ , the function h h h is harmonic and continuous up to both spheres, with inner boundary value zero and outer boundary values at most r i / R r_i/R r i / R . Comparing it with the radial barrier H j δ H_j^\delta H j δ from the proof of Theorem 3.9(b) gives
0 ≤ h ( z ) ≤ r i R H j δ ( z ) , z ∈ A j δ . 0\le h(z)\le\frac{r_i}{R}H_j^\delta(z),\qquad z\in A_j^\delta. 0 ≤ h ( z ) ≤ R r i H j δ ( z ) , z ∈ A j δ .
Thus Theorem 3.9(a) applies with M = r i / R M=r_i/R M = r i / R . By linearity of the flux and (2.4),
0 ≤ C i j I − C i j J = Ψ h ≤ 4 π r i r j ( r j + δ ) δ R . 0\le C_{ij}^{\mathcal{I}}-C_{ij}^{\mathcal{J}}=\Psi_h\le\frac{4\pi r_i r_j(r_j+\delta)}{\delta R}. 0 ≤ C ij I − C ij J = Ψ h ≤ δ R 4 π r i r j ( r j + δ ) .
■
Killed Brownian Motion
We begin this section by introducing the probability density of killed Brownian motion. This then allows us to define the Green’s function for the exterior domain Ω \Omega Ω , which will then be used to obtain a quantitative estimate on M i j M_{ij} M ij from Theorem 3.9.
The killed transition density and Green’s function
The killed transition density describes Brownian motion before its first contact with a resonator. We follow [15 Chapter 2] .
Let p t ( z ) = ( 2 π t ) − 3 / 2 e − ∣ z ∣ 2 / ( 2 t ) p_t(z)=(2\pi t)^{-3/2}e^{-|z|^2/(2t)} p t ( z ) = ( 2 π t ) − 3/2 e − ∣ z ∣ 2 / ( 2 t ) denote the Gaussian kernel in R 3 \mathbb{R}^3 R 3 and define
r t Ω ( x , y ) = E x [ 1 { τ < t } p t − τ ( B ( τ ) − y ) ] . (31) r_t^\Omega(x,y)=\mathbb{E}_x[\mathbf{1}_{\{\tau<t\}}p_{t-\tau}(B(\tau)-y)]. \tag*{(31)} r t Ω ( x , y ) = E x [ 1 { τ < t } p t − τ ( B ( τ ) − y )] . (31)
The killed probability density is then defined as
p t Ω ( x , y ) = p t ( x − y ) − r t Ω ( x , y ) . (32) p_t^\Omega(x,y)=p_t(x-y)-r_t^\Omega(x,y). \tag*{(32)} p t Ω ( x , y ) = p t ( x − y ) − r t Ω ( x , y ) . (32)
The following result can be found in [22 ] , [15 Theorem 2.4] .
Theorem 4.1. Let Ω \Omega Ω be a domain in R d \mathbb{R}^d R d . Then for any t > 0 t>0 t > 0 , x ∈ Ω x\in\Omega x ∈ Ω and A A A in the Borel σ \sigma σ -algebra of R d \mathbb{R}^d R d
P x ( τ > t , B ( t ) ∈ A ) = ∫ A p t Ω ( x , y ) d y . (33) \mathbb{P}_x(\tau>t,B(t)\in A)=\int_A p_t^\Omega(x,y)\,dy. \tag*{(33)} P x ( τ > t , B ( t ) ∈ A ) = ∫ A p t Ω ( x , y ) d y . (33)
The function p t Ω p_t^{\Omega} p t Ω is symmetric on R d × R d \mathbb{R}^d \times\mathbb{R}^d R d × R d , continuous on ( R d ∖ ∂ Ω ) × ( R d ∖ ∂ Ω ) (\mathbb{R}^d \setminus\partial\Omega) \times(\mathbb{R}^d \setminus\partial\Omega) ( R d ∖ ∂ Ω ) × ( R d ∖ ∂ Ω ) , and strictly positive on Ω × Ω \Omega\times\Omega Ω × Ω . For any t > s > 0 t > s > 0 t > s > 0 , x , y ∈ R d x,y \in\mathbb{R}^d x , y ∈ R d , we have
p t Ω ( x , y ) = ∫ R d p s Ω ( x , z ) p t − s Ω ( z , y ) d z . (34) p_t^{\Omega}(x,y)=\int_{\mathbb{R}^d}p_s^{\Omega}(x,z)p_{t-s}^{\Omega}(z,y)\,dz. \tag*{(34)} p t Ω ( x , y ) = ∫ R d p s Ω ( x , z ) p t − s Ω ( z , y ) d z . (34)
For any t > 0 t>0 t > 0 , y ∈ Ω y\in\Omega y ∈ Ω and regular point z ∈ ∂ Ω z\in\partial\Omega z ∈ ∂ Ω , we have
lim x → z x ∈ Ω ‾ p t Ω ( x , y ) = 0. (35) \lim_{\substack{x\to z\\x\in\overline{\Omega}}}p_t^{\Omega}(x,y)=0. \tag*{(35)} x → z x ∈ Ω lim p t Ω ( x , y ) = 0. (35)
With the killed transition density in hand, we can now define the killed Green’s function and operator
G Ω ( x , y ) = ∫ 0 ∞ p t Ω ( x , y ) d t (36) G^{\Omega}(x,y)=\int_0^\infty p_t^{\Omega}(x,y)\,dt \tag*{(36)} G Ω ( x , y ) = ∫ 0 ∞ p t Ω ( x , y ) d t (36)
G Ω f ( x ) = ∫ R d G Ω ( x , y ) f ( y ) d y = E x [ ∫ 0 τ f ( B ( t ) ) d t ] . (37) G^{\Omega}f(x)=\int_{\mathbb{R}^d}G^{\Omega}(x,y)f(y)\,dy=\mathbb{E}_x\left[\int_0^\tau f(B(t))\,dt\right]. \tag*{(37)} G Ω f ( x ) = ∫ R d G Ω ( x , y ) f ( y ) d y = E x [ ∫ 0 τ f ( B ( t )) d t ] . (37)
Finally, writing G 0 ( x , y ) = 1 2 π ∣ x − y ∣ G^0(x,y)=\frac{1}{2\pi|x-y|} G 0 ( x , y ) = 2 π ∣ x − y ∣ 1 for the free Green’s function, the following identity holds [15, Theorem 2.5}:
G Ω ( x , y ) = G 0 ( x , y ) − E x [ 1 { τ < ∞ } G 0 ( B ( τ ) , y ) ] . (38) G^{\Omega}(x,y)=G^0(x,y)-\mathbb{E}_x\left[1_{\{\tau<\infty\}}G^0(B(\tau),y)\right]. \tag*{(38)} G Ω ( x , y ) = G 0 ( x , y ) − E x [ 1 { τ < ∞ } G 0 ( B ( τ ) , y ) ] . (38)
By an abuse of notation, we define
G Ω ( x , Λ ) = G Ω 1 Λ ( x ) = E x [ ∫ 0 τ 1 Λ ( B ( t ) ) d t ] = ∫ 0 ∞ P x ( τ > t , B ( t ) ∈ Λ ) d t . (39) G^{\Omega}(x,\Lambda)=G^{\Omega}1_{\Lambda}(x)=\mathbb{E}_x\left[\int_0^\tau1_{\Lambda}(B(t))\,dt\right]=\int_0^\infty\mathbb{P}_x(\tau>t,B(t)\in\Lambda)\,dt. \tag*{(39)} G Ω ( x , Λ ) = G Ω 1 Λ ( x ) = E x [ ∫ 0 τ 1 Λ ( B ( t )) d t ] = ∫ 0 ∞ P x ( τ > t , B ( t ) ∈ Λ ) d t . (39)
where the last equality follows from Theorem 4.1 . It is not difficult to see that G Ω ( x , Λ ) G^{\Omega}(x,\Lambda) G Ω ( x , Λ ) is thus the expected occupation time in Λ \Lambda Λ of a Brownian motion started in x x x until it is killed. In Theorem 4.3, we will use the occupation time G Ω ( ⋅ , A i δ ) G^{\Omega}(\cdot,A_i^\delta) G Ω ( ⋅ , A i δ ) to upper bound the potential V i V_i V i . Consequently, we require good control on G Ω ( ⋅ , A i δ ) G^{\Omega}(\cdot,A_i^\delta) G Ω ( ⋅ , A i δ ) , which will be done through estimating the killed kernel density. This is the first time where geometry will enter the picture. To avoid essentially proving the same result twice, we state a general version, which will later be adapted to the respective setup.
Lemma 4.2. Let Π \Pi Π be an orthogonal projection and x , y ∈ Ω x,y\in\Omega x , y ∈ Ω with R = ∣ Π x − Π y ∣ > 0 R=|\Pi x-\Pi y|>0 R = ∣Π x − Π y ∣ > 0 . Then it holds that
p t Ω ( x , y ) ≤ ( 2 / π t ) 3 / 2 [ P y ( τ > t / 4 ) P x { τ > t / 2 , ∣ Π ( B ( t / 2 ) − x ) ∣ ≥ R / 2 } + P x ( τ > t / 4 ) P y { τ > t / 2 , ∣ Π ( B ( t / 2 ) − y ) ∣ ≥ R / 2 } ] . (40) \begin{aligned}
p_t^{\Omega}(x,y)\leq(2/\pi t)^{3/2}\big[&
\mathbb{P}_y(\tau>t/4)\mathbb{P}_x\{\tau>t/2,\ |\Pi(B(t/2)-x)|\geq R/2\}\\
&+\mathbb{P}_x(\tau>t/4)\mathbb{P}_y\{\tau>t/2,\ |\Pi(B(t/2)-y)|\geq R/2\}\big]. \tag*{(40)}
\end{aligned} p t Ω ( x , y ) ≤ ( 2/ π t ) 3/2 [ P y ( τ > t /4 ) P x { τ > t /2 , ∣Π ( B ( t /2 ) − x ) ∣ ≥ R /2 } + P x ( τ > t /4 ) P y { τ > t /2 , ∣Π ( B ( t /2 ) − y ) ∣ ≥ R /2 } ] . (40) For any p > 0 p>0 p > 0 and η > 0 \eta>0 η > 0 ,
P z ( τ > s , ∣ Π ( B ( s ) − z ) ∣ ≥ η ) ≤ min { P z ( τ > s ) , η − p E z ∣ Π ( B ( τ ∧ s ) − z ) ∣ p } . (41) \mathbb{P}_z(\tau>s,\ |\Pi(B(s)-z)|\geq\eta)\leq\min\{\mathbb{P}_z(\tau>s),\eta^{-p}\mathbb{E}_z|\Pi(B(\tau\wedge s)-z)|^p\}. \tag*{(41)} P z ( τ > s , ∣Π ( B ( s ) − z ) ∣ ≥ η ) ≤ min { P z ( τ > s ) , η − p E z ∣Π ( B ( τ ∧ s ) − z ) ∣ p } . (41)
Proof. Let E x = { u : ∣ Π ( u − x ) ∣ ≥ R / 2 } E_x=\{u:|\Pi(u-x)|\geq R/2\} E x = { u : ∣Π ( u − x ) ∣ ≥ R /2 } and define E y E_y E y similarly. Since ∣ Π x − Π y ∣ = R |\Pi x-\Pi y|=R ∣Π x − Π y ∣ = R , the triangle inequality implies E x ∪ E y = R 3 E_x\cup E_y=\mathbb{R}^3 E x ∪ E y = R 3 . Theorem 4.1 now implies
p t Ω ( x , y ) ≤ ∫ E x p t / 2 Ω ( x , z ) p t / 2 Ω ( z , y ) d z + ∫ E y p t / 2 Ω ( x , z ) p t / 2 Ω ( z , y ) d z . (42) p_t^{\Omega}(x,y)\leq\int_{E_x}p_{t/2}^{\Omega}(x,z)p_{t/2}^{\Omega}(z,y)\,dz+\int_{E_y}p_{t/2}^{\Omega}(x,z)p_{t/2}^{\Omega}(z,y)\,dz. \tag*{(42)} p t Ω ( x , y ) ≤ ∫ E x p t /2 Ω ( x , z ) p t /2 Ω ( z , y ) d z + ∫ E y p t /2 Ω ( x , z ) p t /2 Ω ( z , y ) d z . (42)
Applying the semigroup property once more, we have
p t / 2 Ω ( x , y ) = ∫ R 3 p t / 4 Ω ( x , z ) p t / 4 Ω ( z , y ) d z ≤ ( 2 / π t ) 3 / 2 P y ( τ > t / 4 ) . (43) \begin{aligned}
p_{t/2}^{\Omega}(x,y)&=\int_{\mathbb{R}^3}p_{t/4}^{\Omega}(x,z)p_{t/4}^{\Omega}(z,y)\,dz\\
&\leq(2/\pi t)^{3/2}\mathbb{P}_y(\tau>t/4). \tag*{(43)}
\end{aligned} p t /2 Ω ( x , y ) = ∫ R 3 p t /4 Ω ( x , z ) p t /4 Ω ( z , y ) d z ≤ ( 2/ π t ) 3/2 P y ( τ > t /4 ) . (43)
Combining the above equation with Equation (42) , we obtain the Equation (40) :
p t Ω ( x , y ) ≤ ( 2 / π t ) 3 / 2 [ P y ( τ > t / 4 ) ∫ E x p t / 2 Ω ( x , z ) d z + P x ( τ > t / 4 ) ∫ E y p t / 2 Ω ( y , z ) d z ] = ( 2 / π t ) 3 / 2 [ P y ( τ > t / 4 ) P x { τ > t / 2 , ∣ Π ( B ( t / 2 ) − x ) ∣ ≥ R / 2 } + P x ( τ > t / 4 ) P y { τ > t / 2 , ∣ Π ( B ( t / 2 ) − y ) ∣ ≥ R / 2 } ] . (44) \begin{aligned}
p_t^{\Omega}(x,y)\leq(2/\pi t)^{3/2}\big[&
\mathbb{P}_y(\tau>t/4)\int_{E_x}p_{t/2}^{\Omega}(x,z)\,dz+
\mathbb{P}_x(\tau>t/4)\int_{E_y}p_{t/2}^{\Omega}(y,z)\,dz\big]\\
=(2/\pi t)^{3/2}\big[&
\mathbb{P}_y(\tau>t/4)\mathbb{P}_x\{\tau>t/2,\ |\Pi(B(t/2)-x)|\geq R/2\}\\
&+\mathbb{P}_x(\tau>t/4)\mathbb{P}_y\{\tau>t/2,\ |\Pi(B(t/2)-y)|\geq R/2\}\big]. \tag*{(44)}
\end{aligned} p t Ω ( x , y ) ≤ ( 2/ π t ) 3/2 [ = ( 2/ π t ) 3/2 [ P y ( τ > t /4 ) ∫ E x p t /2 Ω ( x , z ) d z + P x ( τ > t /4 ) ∫ E y p t /2 Ω ( y , z ) d z ] P y ( τ > t /4 ) P x { τ > t /2 , ∣Π ( B ( t /2 ) − x ) ∣ ≥ R /2 } + P x ( τ > t /4 ) P y { τ > t /2 , ∣Π ( B ( t /2 ) − y ) ∣ ≥ R /2 } ] . (44)
It thus remains to prove Equation (41) . The estimate
P z ( τ > s , ∣ Π ( B ( s ) − z ) ∣ ≥ η ) ≤ P z ( τ > s ) , (45) \mathbb{P}_z\left(\tau>s,\left|\Pi(B(s)-z)\right|\ge\eta\right)\le\mathbb{P}_z(\tau>s), \tag*{(45)} P z ( τ > s , ∣ Π ( B ( s ) − z ) ∣ ≥ η ) ≤ P z ( τ > s ) , (45)
is trivial. We thus concern ourselves with the second bound. To this end, note that
P z ( τ > s , ∣ Π ( B ( s ) − z ) ∣ ≥ η ) = P z ( τ > s , ∣ Π ( B ( τ ∧ s ) − z ) ∣ ≥ η ) ≤ P z ( ∣ Π ( B ( τ ∧ s ) − z ) ∣ ≥ η ) ≤ E z [ ∣ Π ( B ( τ ∧ s ) − z ) ∣ p ] η − p , (46) \begin{aligned}
\mathbb{P}_z\left(\tau>s,\left|\Pi(B(s)-z)\right|\ge\eta\right)
&=\mathbb{P}_z\left(\tau>s,\left|\Pi(B(\tau\wedge s)-z)\right|\ge\eta\right)\\
&\le\mathbb{P}_z\left(\left|\Pi(B(\tau\wedge s)-z)\right|\ge\eta\right)\\
&\le\mathbb{E}_z\left[\left|\Pi(B(\tau\wedge s)-z)\right|^p\right]\eta^{-p}, \tag*{(46)}
\end{aligned} P z ( τ > s , ∣ Π ( B ( s ) − z ) ∣ ≥ η ) = P z ( τ > s , ∣ Π ( B ( τ ∧ s ) − z ) ∣ ≥ η ) ≤ P z ( ∣ Π ( B ( τ ∧ s ) − z ) ∣ ≥ η ) ≤ E z [ ∣ Π ( B ( τ ∧ s ) − z ) ∣ p ] η − p , (46)
where we used a Chebyshev inequality in the last line.
From hitting probability to occupation time
Lemma 4.3. For A i δ A_i^\delta A i δ with δ ∈ ( 0 , d min / 2 ] \delta\in(0,d_{\min}/2] δ ∈ ( 0 , d m i n /2 ] and z ∈ Ω z\in\Omega z ∈ Ω satisfying ∣ z − x i ∣ ≥ r i + δ / 2 \lvert z-x_i\rvert\ge r_i+\delta/2 ∣ z − x i ∣ ≥ r i + δ /2 it holds that
V i ( z ) ≤ 12 δ 2 G Ω ( z , A i δ ) , ∣ z − x i ∣ ≥ r i + δ / 2. (47) V_i(z)\le\frac{12}{\delta^2}\mathcal{G}^{\Omega}(z,A_i^\delta),\qquad\lvert z-x_i\rvert\ge r_i+\delta/2. \tag*{(47)} V i ( z ) ≤ δ 2 12 G Ω ( z , A i δ ) , ∣ z − x i ∣ ≥ r i + δ /2. (47) Proof. Let T T T be the first hit of the sphere of radius r i + δ / 2 r_i+\delta/2 r i + δ /2 about x i x_i x i . Continuity implies that every path first absorbed at D i D_i D i has T < τ T<\tau T < τ , including T = 0 T=0 T = 0 if the starting point is on this sphere.
If w w w is on the sphere of radius r i + δ / 2 r_i+\delta/2 r i + δ /2 , then B δ / 2 ( w ) B_{\delta/2}(w) B δ /2 ( w ) lies in A i δ A_i^\delta A i δ . Write σ \sigma σ for the exit time of B B B from B δ / 2 ( w ) B_{\delta/2}(w) B δ /2 ( w ) . Optional stopping of ∣ B ( t ) − w ∣ 2 − 3 t \lvert B(t)-w\rvert^2-3t ∣ B ( t ) − w ∣ 2 − 3 t at σ \sigma σ gives 3 E w ( σ ) = E w ∣ B ( σ ) − w ∣ 2 = δ 2 / 4 3\mathbb{E}_w(\sigma)=\mathbb{E}_w\lvert B(\sigma)-w\rvert^2=\delta^2/4 3 E w ( σ ) = E w ∣ B ( σ ) − w ∣ 2 = δ 2 /4 . Hence E w σ = δ 2 / 12 \mathbb{E}_w\sigma=\delta^2/12 E w σ = δ 2 /12 . Before σ \sigma σ no absorption occurs, and the whole time is spent in A i δ A_i^\delta A i δ . Strong Markov at T T T , on the event { T < τ } ∈ F + ( T ) \{T<\tau\}\in\mathcal{F}^+(T) { T < τ } ∈ F + ( T ) , therefore gives
G Ω ( z , A i δ ) = E z [ ∫ 0 τ 1 A i δ ( B ( t ) ) d t ] ≥ E z [ 1 T < τ ∫ T τ 1 A i δ ( B ( t ) ) d t ] = E z [ 1 T < τ E B ( T ) [ ∫ 0 τ 1 A i δ ( B ( t ) ) d t ] ] ≥ δ 2 12 P z ( T < τ ) , (48) \begin{aligned}
\mathcal{G}^{\Omega}(z,A_i^\delta)
&=\mathbb{E}_z\left[\int_0^\tau\mathbf{1}_{A_i^\delta}(B(t))\,dt\right]\\
&\ge\mathbb{E}_z\left[\mathbf{1}_{T<\tau}\int_T^\tau\mathbf{1}_{A_i^\delta}(B(t))\,dt\right]\\
&=\mathbb{E}_z\left[\mathbf{1}_{T<\tau}\mathbb{E}_{B(T)}\left[\int_0^\tau\mathbf{1}_{A_i^\delta}(B(t))\,dt\right]\right]\\
&\ge\frac{\delta^2}{12}\mathbb{P}_z(T<\tau), \tag*{(48)}
\end{aligned} G Ω ( z , A i δ ) = E z [ ∫ 0 τ 1 A i δ ( B ( t )) d t ] ≥ E z [ 1 T < τ ∫ T τ 1 A i δ ( B ( t )) d t ] = E z [ 1 T < τ E B ( T ) [ ∫ 0 τ 1 A i δ ( B ( t )) d t ] ] ≥ 12 δ 2 P z ( T < τ ) , (48) which yields the claim.
Upper bounds
In this section, we will derive the upper bounds of Theorems 2.5 to 2.7 . While the arguments are morally the same, the estimates have to be adapted to the respective geometry, hence the split into subsections. By Theorem 2.8, we may assume ρ ≤ 1 / 2 \rho\le1/2 ρ ≤ 1/2 throughout this and the following section. We denote by Π ∥ z = z ∥ \Pi_{\parallel}z=z_{\parallel} Π ∥ z = z ∥ and Π ⊥ z = z ⊥ \Pi_{\perp}z=z_{\perp} Π ⊥ z = z ⊥ the orthogonal projections onto the chain axis and onto the screen plane, respectively. In each subsection, q − q_- q − denotes the constant of the corresponding killing lemma (Lemmas 5.1 , 5.4 and 5.7 ). Roughly, the argument is as follows
Derive a killing probability that holds uniformly for points on the boundary of a cylinder/slab.
Use this uniform killing to prove that the number of visits to a fixed shell before absorption has a geometric tail.
Turn the geometric tail of the visits into a tail estimate for P z ( τ > t ) \mathbb{P}_z(\tau>t) P z ( τ > t ) .
Use the tail estimate for P z ( τ > t ) \mathbb{P}_z(\tau>t) P z ( τ > t ) to obtain an estimate for the killed density p t Ω p_t^\Omega p t Ω .
Integrate in time to obtain an upper bound for the Green’s function G Ω ( x , y ) \mathcal{G}^{\Omega}(x,y) G Ω ( x , y ) .
Chains
As outlined above, we first derive a uniform lower bound on the probability of absorption during a visit to ∂ Z 0 \partial Z_0 ∂ Z 0 .
Lemma 5.1. Let α = inf { t ≥ 0 : ∣ B ⊥ ( t ) ∣ ≥ 2 } \alpha= \inf\{t \ge0 : |B_{\perp}(t)| \ge2\} α = inf { t ≥ 0 : ∣ B ⊥ ( t ) ∣ ≥ 2 } be the exit time from Z 1 Z_1 Z 1 . For every z ∈ ∂ Z 0 z \in\partial Z_0 z ∈ ∂ Z 0 ,
P z ( τ < α ) ≥ q − ∈ ( 0 , 1 ) . (49) \mathbb{P}_z(\tau< \alpha) \ge q_- \in(0,1). \tag*{(49)} P z ( τ < α ) ≥ q − ∈ ( 0 , 1 ) . (49) The constant q − q_- q − depends only on r min r_{\min} r m i n and L L L .
Proof. Fix T = 1 T=1 T = 1 and ε = 1 2 min { r min , 1 } \varepsilon= \frac{1}{2}\min\{r_{\min},1\} ε = 2 1 min { r m i n , 1 } . The assumptions of Theorem 2.5 imply that for any z ∈ ∂ Z 0 z \in\partial Z_0 z ∈ ∂ Z 0 , there exists some x i x_i x i such that d = ∣ x i − z ∣ ≤ L 2 + 9 / 4 d = |x_i-z| \le\sqrt{L^2+9/4} d = ∣ x i − z ∣ ≤ L 2 + 9/4 . Set F ( t ) = z + t ( x i − z ) T F(t)=z+\frac{t(x_i-z)}{T} F ( t ) = z + T t ( x i − z ) and consider the event E = { max t ≤ T ∣ B ( t ) − F ( t ) ∣ < ε } E=\{\max_{t\le T}|B(t)-F(t)|<\varepsilon\} E = { max t ≤ T ∣ B ( t ) − F ( t ) ∣ < ε } . That is, B B B stays within a tube of radius ε \varepsilon ε of the path F F F until time T T T . It is not difficult to see that
{ τ < α } ⊃ E , (50) \{\tau< \alpha\} \supset E, \tag*{(50)} { τ < α } ⊃ E , (50) since on E E E we have ∣ B ( T ) − x i ∣ < ε < r i |B(T)-x_i|<\varepsilon<r_i ∣ B ( T ) − x i ∣ < ε < r i , i.e. B ( T ) ∈ D i B(T)\in D_i B ( T ) ∈ D i , while ∣ B ⊥ ( t ) ∣ ≤ max { ∣ z ⊥ ∣ , ∣ a i ∣ } + ε < 2 |B_{\perp}(t)|\le\max\{|z_{\perp}|,|a_i|\}+\varepsilon<2 ∣ B ⊥ ( t ) ∣ ≤ max { ∣ z ⊥ ∣ , ∣ a i ∣ } + ε < 2 for all t ≤ T t\le T t ≤ T . Hence P z ( τ < α ) ≥ P z ( E ) \mathbb{P}_z(\tau<\alpha)\ge\mathbb{P}_z(E) P z ( τ < α ) ≥ P z ( E ) . The Cameron–Martin Theorem [25 Theorem 1.38, Remark 1.43] yields
P z ( E ) = E 0 [ 1 A exp ( ∫ 0 T F ′ d B − 1 2 ∫ 0 T ∣ F ′ ∣ 2 ( s ) d s ) ] , (51) \mathbb{P}_z(E)=\mathbb{E}_0\left[\mathbf{1}_A\exp\left(\int_0^T F'\,\mathrm{d}B-\frac{1}{2}\int_0^T|F'|^2(s)\,\mathrm{d}s\right)\right], \tag*{(51)} P z ( E ) = E 0 [ 1 A exp ( ∫ 0 T F ′ d B − 2 1 ∫ 0 T ∣ F ′ ∣ 2 ( s ) d s ) ] , (51) where A = { max t ≤ T ∣ B ( t ) ∣ < ε } A=\{\max_{t\le T}|B(t)|<\varepsilon\} A = { max t ≤ T ∣ B ( t ) ∣ < ε } . Noting that A A A is symmetric with respect to negation, i.e. B ∈ A ⟺ − B ∈ A B\in A\Longleftrightarrow-B\in A B ∈ A ⟺ − B ∈ A , we obtain
P z ( E ) = E 0 [ 1 A cosh ( ⟨ x i − z , B ( T ) ⟩ T ) ] exp ( − ∣ x i − z ∣ 2 2 T ) \mathbb{P}_z(E)=\mathbb{E}_0\left[\mathbf{1}_A\cosh\left(\frac{\langle x_i-z,B(T)\rangle}{T}\right)\right]\exp\left(-\frac{|x_i-z|^2}{2T}\right) P z ( E ) = E 0 [ 1 A cosh ( T ⟨ x i − z , B ( T )⟩ ) ] exp ( − 2 T ∣ x i − z ∣ 2 ) ≥ exp ( − ∣ x i − z ∣ 2 2 T ) P 0 ( A ) . (52) \ge\exp\left(-\frac{|x_i-z|^2}{2T}\right)\mathbb{P}_0(A). \tag*{(52)} ≥ exp ( − 2 T ∣ x i − z ∣ 2 ) P 0 ( A ) . (52) Let σ = inf { t ≥ 0 : ∣ B ( t ) ∣ ≥ 1 } \sigma=\inf\{t\ge0:|B(t)|\ge1\} σ = inf { t ≥ 0 : ∣ B ( t ) ∣ ≥ 1 } . Using the scaling property, it holds that
P 0 ( A ) = P 0 ( σ > T / ε 2 ) > 0. (53) \mathbb{P}_0(A)=\mathbb{P}_0(\sigma>T/\varepsilon^2)>0. \tag*{(53)} P 0 ( A ) = P 0 ( σ > T / ε 2 ) > 0. (53) Hence we may choose q − = e − ( L 2 + 9 / 4 ) / 2 P 0 ( A ) q_-=e^{-(L^2+9/4)/2}\mathbb{P}_0(A) q − = e − ( L 2 + 9/4 ) /2 P 0 ( A ) from which the claim follows. ■
The next Lemma achieves the second and third step of the outline: Observe that the Brownian motion alternates between visits to Z 1 Z_1 Z 1 and excursions from ∂ Z 1 \partial Z_1 ∂ Z 1 back to ∂ Z 0 \partial Z_0 ∂ Z 0 . A resonator can only be hit during the time it spends inside Z 1 Z_1 Z 1 , that is during a visit. This allows us to derive tail estimates for the durations of visits and excursions.
Lemma 5.2. Let z ∈ Ω z\in\Omega z ∈ Ω with ∣ z ⊥ ∣ ≤ 2 |z_{\perp}|\le2 ∣ z ⊥ ∣ ≤ 2 . Let β 0 = 0 \beta_0=0 β 0 = 0 , α 1 = inf { t ≥ 0 : ∣ B ⊥ ( t ) ∣ ≥ 2 } \alpha_1=\inf\{t\ge0:|B_{\perp}(t)|\ge2\} α 1 = inf { t ≥ 0 : ∣ B ⊥ ( t ) ∣ ≥ 2 } (the time α \alpha α of Theorem 5.1) and, for k ≥ 1 k\ge1 k ≥ 1 ,
β k = inf { t ≥ α k : ∣ B ⊥ ( t ) ∣ ≤ 1 } , α k + 1 = inf { t ≥ β k : ∣ B ⊥ ( t ) ∣ ≥ 2 } , \beta_k=\inf\{t\ge\alpha_k:|B_{\perp}(t)|\le1\},\qquad\alpha_{k+1}=\inf\{t\ge\beta_k:|B_{\perp}(t)|\ge2\}, β k = inf { t ≥ α k : ∣ B ⊥ ( t ) ∣ ≤ 1 } , α k + 1 = inf { t ≥ β k : ∣ B ⊥ ( t ) ∣ ≥ 2 } , and write ( a ) + = max { a , 0 } (a)_+=\max\{a,0\} ( a ) + = max { a , 0 } . Let U k = α k − β k − 1 U_k=\alpha_k-\beta_{k-1} U k = α k − β k − 1 (the k k k -th visit to Z 1 Z_1 Z 1 ) and Y k = β k − α k Y_k=\beta_k-\alpha_k Y k = β k − α k (the k k k -th excursion out of Z 0 Z_0 Z 0 ). Then:
(a) The α k , β k \alpha_k,\beta_k α k , β k are stopping times; P z \mathbb{P}_z P z -almost surely they are finite, 0 = β 0 ≤ α 1 < β 1 < α 2 < ⋯ 0=\beta_0\le\alpha_1<\beta_1<\alpha_2<\cdots 0 = β 0 ≤ α 1 < β 1 < α 2 < ⋯ , and α k → ∞ \alpha_k\to\infty α k → ∞ .
(b) τ ∉ [ α k , β k ] \tau\notin[\alpha_k,\beta_k] τ ∈ / [ α k , β k ] for all k ≥ 1 k\ge1 k ≥ 1 .
(c) P z ( τ > α k ) ≤ ( 1 − q − ) k − 1 \mathbb{P}_z(\tau>\alpha_k)\le(1-q_-)^{k-1} P z ( τ > α k ) ≤ ( 1 − q − ) k − 1 for all k ≥ 1 k\ge1 k ≥ 1 ; in particular τ < ∞ \tau<\infty τ < ∞ P z \mathbb{P}_z P z -almost surely.
(d) For all k ≥ 1 k\ge1 k ≥ 1 and s > 0 s>0 s > 0 ,
P z ( τ > β k − 1 , U k > s ) ≤ ( 1 − q − ) ( k − 2 ) + min { 1 , 36 / s } , (54) \mathbb{P}_z(\tau>\beta_{k-1},U_k>s)\le(1-q_-)^{(k-2)_+}\min\{1,36/s\}, \tag*{(54)} P z ( τ > β k − 1 , U k > s ) ≤ ( 1 − q − ) ( k − 2 ) + min { 1 , 36/ s } , (54) P z ( τ > α k , Y k > s ) ≤ ( 1 − q − ) k − 1 12 log ( 2 + s ) . (55) \mathbb{P}_z(\tau>\alpha_k,Y_k>s)\le(1-q_-)^{k-1}\frac{12}{\log(2+s)}. \tag*{(55)} P z ( τ > α k , Y k > s ) ≤ ( 1 − q − ) k − 1 log ( 2 + s ) 12 . (55)
Proof. (a) The α k \alpha_k α k , β k \beta_k β k are successive hitting times of the closed sets R 3 ∖ Z 1 \mathbb{R}^3 \setminus Z_1 R 3 ∖ Z 1 and Z ‾ 0 \overline{Z}_0 Z 0 , hence stopping times [15 Proposition 1.15] . Moreover they are functions of B ⊥ B_\perp B ⊥ only. Whenever finite, ∣ B ⊥ ( α k ) ∣ = 2 \lvert B_\perp(\alpha_k)\rvert=2 ∣ B ⊥ ( α k )∣ = 2 and ∣ B ⊥ ( β k ) ∣ = 1 \lvert B_\perp(\beta_k)\rvert=1 ∣ B ⊥ ( β k )∣ = 1 by continuity, which gives the strict inequalities. Finiteness follows inductively from Theorems 3.3 and 3.4 and (16) at α k \alpha_k α k and β k \beta_k β k . If α k \alpha_k α k increased to a finite limit, B ⊥ B_\perp B ⊥ would not be uniformly continuous on a compact interval, since ∣ B ⊥ ( α k ) ∣ − ∣ B ⊥ ( β k ) ∣ = 1 \lvert B_\perp(\alpha_k)\rvert-\lvert B_\perp(\beta_k)\rvert=1 ∣ B ⊥ ( α k )∣ − ∣ B ⊥ ( β k )∣ = 1 and α k < β k < α k + 1 \alpha_k<\beta_k<\alpha_{k+1} α k < β k < α k + 1 .
(b) On [ α k , β k ] [\alpha_k,\beta_k] [ α k , β k ] we have ∣ B ⊥ ∣ ≥ 1 > ρ ≥ ∣ a i ∣ + r i \lvert B_\perp\rvert\geq1>\rho\geq\lvert a_i\rvert+r_i ∣ B ⊥ ∣ ≥ 1 > ρ ≥ ∣ a i ∣ + r i for all i i i by (2.9) and ρ ≤ 1 / 2 \rho\leq1/2 ρ ≤ 1/2 , so B ∈ Ω B\in\Omega B ∈ Ω .
(c) For k = 1 k=1 k = 1 the claim is trivial. By (b), { τ > α k } = { β k < τ } \{\tau>\alpha_k\}=\{\beta_k<\tau\} { τ > α k } = { β k < τ } up to a null set; on this event, τ = β k + τ ∘ θ β k \tau=\beta_k+\tau\circ\theta_{\beta_k} τ = β k + τ ∘ θ β k and α k + 1 = β k + α 1 ∘ θ β k \alpha_{k+1}=\beta_k+\alpha_1\circ\theta_{\beta_k} α k + 1 = β k + α 1 ∘ θ β k . By (16) , P z ( τ > α k + 1 ) = E z [ 1 { τ > β k } P B ( β k ) ( τ > α 1 ) ] \mathbb{P}_z(\tau>\alpha_{k+1})=\mathbb{E}_z[\mathbf{1}_{\{\tau>\beta_k\}}\mathbb{P}_{B(\beta_k)}(\tau>\alpha_1)] P z ( τ > α k + 1 ) = E z [ 1 { τ > β k } P B ( β k ) ( τ > α 1 )] . For w ∈ ∂ Z 0 w\in\partial Z_0 w ∈ ∂ Z 0 we have P w ( τ = α 1 ) = 0 \mathbb{P}_w(\tau=\alpha_1)=0 P w ( τ = α 1 ) = 0 (as B ( α 1 ) ∈ ∂ Z 1 ⊂ Ω B(\alpha_1)\in\partial Z_1\subset\Omega B ( α 1 ) ∈ ∂ Z 1 ⊂ Ω ), so P w ( τ > α 1 ) ≤ 1 − q − \mathbb{P}_w(\tau>\alpha_1)\leq1-q_- P w ( τ > α 1 ) ≤ 1 − q − by Theorem 5.1. As B ( β k ) ∈ ∂ Z 0 B(\beta_k)\in\partial Z_0 B ( β k ) ∈ ∂ Z 0 , P z ( τ > α k + 1 ) ≤ ( 1 − q − ) P z ( τ > α k ) \mathbb{P}_z(\tau>\alpha_{k+1})\leq(1-q_-)\mathbb{P}_z(\tau>\alpha_k) P z ( τ > α k + 1 ) ≤ ( 1 − q − ) P z ( τ > α k ) , and (c) follows by induction; finally P z ( τ = ∞ ) ≤ P z ( τ > α k ) → 0 \mathbb{P}_z(\tau=\infty)\leq\mathbb{P}_z(\tau>\alpha_k)\to0 P z ( τ = ∞ ) ≤ P z ( τ > α k ) → 0 .
(d) U k U_k U k is the exit time from the disc of radius 2 2 2 of the planar motion restarted at B ⊥ ( β k − 1 ) B_\perp(\beta_{k-1}) B ⊥ ( β k − 1 ) , ∣ B ⊥ ( β k − 1 ) ∣ ≤ 2 \lvert B_\perp(\beta_{k-1})\rvert\leq2 ∣ B ⊥ ( β k − 1 )∣ ≤ 2 . By (16) and Theorem 3.3 (R = 2 R=2 R = 2 ), P z ( τ > β k − 1 , U k > s ) ≤ P z ( τ > β k − 1 ) min { 1 , 36 / s } \mathbb{P}_z(\tau>\beta_{k-1},U_k>s)\leq\mathbb{P}_z(\tau>\beta_{k-1})\min\{1,36/s\} P z ( τ > β k − 1 , U k > s ) ≤ P z ( τ > β k − 1 ) min { 1 , 36/ s } ; here P z ( τ > β 0 ) ≤ 1 \mathbb{P}_z(\tau>\beta_0)\leq1 P z ( τ > β 0 ) ≤ 1 , and P z ( τ > β k − 1 ) = P z ( τ > α k − 1 ) ≤ ( 1 − q − ) k − 2 \mathbb{P}_z(\tau>\beta_{k-1})=\mathbb{P}_z(\tau>\alpha_{k-1})\leq(1-q_-)^{k-2} P z ( τ > β k − 1 ) = P z ( τ > α k − 1 ) ≤ ( 1 − q − ) k − 2 for k ≥ 2 k\geq2 k ≥ 2 by (b), (c). Likewise, Y k Y_k Y k is the hitting time of the closed unit disc by the planar motion restarted at B ⊥ ( α k ) B_\perp(\alpha_k) B ⊥ ( α k ) , ∣ B ⊥ ( α k ) ∣ = 2 \lvert B_\perp(\alpha_k)\rvert=2 ∣ B ⊥ ( α k )∣ = 2 , and (16) , Theorem 3.4 and (c) give (5.7). ■
Using the geometric tail of the number of visits together with the duration estimates, we obtain tail estimates on the survival duration. This is step three of the outline.
Proposition 5.3 (Logarithmic survival bound). Let K 0 = 635 q − − 3 K_0=635q_-^{-3} K 0 = 635 q − − 3 , which depends only on r min r_{\min} r m i n and L L L . For all z ∈ Ω z\in\Omega z ∈ Ω with ∣ z ⊥ ∣ ≤ 2 \lvert z_\perp\rvert\leq2 ∣ z ⊥ ∣ ≤ 2 and t ≥ 0 t\geq0 t ≥ 0 ,
P z ( τ > t ) ≤ K 0 log ( 2 + t ) and E z [ τ ∧ t ] ≤ 3 K 0 t log ( 2 + t ) . (56) \mathbb{P}_z(\tau>t)\leq\frac{K_0}{\log(2+t)}
\qquad\text{and}\qquad
\mathbb{E}_z[\tau\wedge t]\leq\frac{3K_0t}{\log(2+t)}. \tag*{(56)} P z ( τ > t ) ≤ log ( 2 + t ) K 0 and E z [ τ ∧ t ] ≤ log ( 2 + t ) 3 K 0 t . (56)
Proof. We begin with the tail estimate on P z ( τ > t ) \mathbb{P}_z(\tau>t) P z ( τ > t ) . Let w k = 1 k ( k + 1 ) w_k=\frac{1}{k(k+1)} w k = k ( k + 1 ) 1 , so that ∑ k ≥ 1 w k = 1 \sum_{k\geq1}w_k=1 ∑ k ≥ 1 w k = 1 . Almost surely τ < ∞ \tau<\infty τ < ∞ , and since α k → ∞ \alpha_k\to\infty α k → ∞ we may define G = min { k ≥ 1 : τ < α k } G=\min\{k\geq1:\tau<\alpha_k\} G = min { k ≥ 1 : τ < α k } . As
τ < α G = ∑ k = 1 G U k + ∑ k = 1 G − 1 Y k , \tau<\alpha_G=\sum_{k=1}^{G}U_k+\sum_{k=1}^{G-1}Y_k, τ < α G = k = 1 ∑ G U k + k = 1 ∑ G − 1 Y k ,
on { τ > t } \{\tau>t\} { τ > t } it is impossible that U k ≤ w k t / 2 U_k\leq w_kt/2 U k ≤ w k t /2 for all k ≤ G k\leq G k ≤ G and Y k ≤ w k t / 2 Y_k\leq w_kt/2 Y k ≤ w k t /2 for all k ≤ G − 1 k\leq G-1 k ≤ G − 1 . Hence, as { k ≤ G } ⊂ { τ > β k − 1 } \{k\leq G\}\subset\{\tau>\beta_{k-1}\} { k ≤ G } ⊂ { τ > β k − 1 } and { k ≤ G − 1 } ⊂ { τ > α k } \{k\leq G-1\}\subset\{\tau>\alpha_k\} { k ≤ G − 1 } ⊂ { τ > α k } by Theorem 5.2(b), by Theorem 5.2(d),
P z ( τ > t ) ≤ P z ( α G > t ) ≤ P z ( ⋃ k ≥ 1 { τ > α k , Y k > w k t / 2 } ∪ { τ > β k − 1 , U k > w k t / 2 } ) ≤ ∑ k ≥ 1 ( 1 − q − ) ( k − 2 ) + min { 1 , 72 k ( k + 1 ) t } \begin{aligned}
\mathbb{P}_z(\tau>t)\leq\mathbb{P}_z(\alpha_G>t) \\
&\leq\mathbb{P}_z\left(\bigcup_{k\geq1}\{\tau>\alpha_k,Y_k>w_kt/2\}\cup\{\tau>\beta_{k-1},U_k>w_kt/2\}\right) \\
&\leq\sum_{k\geq1}(1-q_-)^{(k-2)+}\min\left\{1,\frac{72k(k+1)}{t}\right\}
\end{aligned} P z ( τ > t ) ≤ P z ( α G > t ) ≤ P z ( k ≥ 1 ⋃ { τ > α k , Y k > w k t /2 } ∪ { τ > β k − 1 , U k > w k t /2 } ) ≤ k ≥ 1 ∑ ( 1 − q − ) ( k − 2 ) + min { 1 , t 72 k ( k + 1 ) }
+ ∑ k ≥ 1 ( 1 − q − ) k − 1 min { 1 , 12 log ( 2 + t 2 k ( k + 1 ) ) } . (57) \qquad+\sum_{k\geq1}(1-q_-)^{k-1}\min\left\{1,\frac{12}{\log\left(2+\frac{t}{2k(k+1)}\right)}\right\}. \tag*{(57)} + k ≥ 1 ∑ ( 1 − q − ) k − 1 min ⎩ ⎨ ⎧ 1 , log ( 2 + 2 k ( k + 1 ) t ) 12 ⎭ ⎬ ⎫ . (57)
We thus have to upper bound these two sums. Let t ≥ 512 t\geq512 t ≥ 512 . Using
∑ j ≥ 0 ( j + 1 ) ( j + 2 ) x j = 2 ( 1 − x ) − 3 and ( j + 2 ) ( j + 3 ) ≤ 3 ( j + 1 ) ( j + 2 ) , (58) \sum_{j\geq0}(j+1)(j+2)x^j=2(1-x)^{-3}
\qquad\text{and}\qquad
(j+2)(j+3)\leq3(j+1)(j+2), \tag*{(58)} j ≥ 0 ∑ ( j + 1 ) ( j + 2 ) x j = 2 ( 1 − x ) − 3 and ( j + 2 ) ( j + 3 ) ≤ 3 ( j + 1 ) ( j + 2 ) , (58)
we get
∑ k ≥ 1 k ( k + 1 ) ( 1 − q − ) ( k − 2 ) + ≤ 2 + 6 q − − 3 ≤ 8 q − − 3 . (59) \sum_{k\geq1}k(k+1)(1-q_-)^{(k-2)+}\leq2+6q_-^{-3}\leq8q_-^{-3}. \tag*{(59)} k ≥ 1 ∑ k ( k + 1 ) ( 1 − q − ) ( k − 2 ) + ≤ 2 + 6 q − − 3 ≤ 8 q − − 3 . (59)
So the first sum in (5.9) is at most 576 / ( q − 3 t ) ≤ 576 / ( q − 3 log ( 2 + t ) ) 576/(q_-^3t)\leq576/(q_-^3\log(2+t)) 576/ ( q − 3 t ) ≤ 576/ ( q − 3 log ( 2 + t )) .
In the second sum let k t = ⌊ t 1 / 4 ⌋ k_t=\lfloor t^{1/4}\rfloor k t = ⌊ t 1/4 ⌋ . For k ≤ k t k\leq k_t k ≤ k t we have 2 k ( k + 1 ) ≤ 4 t 2k(k+1)\leq4\sqrt{t} 2 k ( k + 1 ) ≤ 4 t and hence log ( 2 + t 2 k ( k + 1 ) ) ≥ 1 2 log t − log 4 ≥ 1 4 log ( 2 t ) ≥ 1 4 log ( 2 + t ) \log\left(2+\frac{t}{2k(k+1)}\right)\geq\frac12\log t-\log4\geq\frac14\log(2t)\geq\frac14\log(2+t) log ( 2 + 2 k ( k + 1 ) t ) ≥ 2 1 log t − log 4 ≥ 4 1 log ( 2 t ) ≥ 4 1 log ( 2 + t ) , these terms contribute at most 48/( q − log ( 2 + t ) ) (q_- \log(2+t)) ( q − log ( 2 + t )) . The terms with k > k t k>k_t k > k t contribute at most
∑ k > k t ( 1 − q − ) k − 1 ≤ e − q − ( t 1 / 4 − 1 ) q − ≤ 4 e q − 2 log ( 2 + t ) . \sum_{k>k_t}(1-q_-)^{k-1}\le\frac{e^{-q_-(t^{1/4}-1)}}{q_-}\le\frac{4e}{q_-^2\log(2+t)}. k > k t ∑ ( 1 − q − ) k − 1 ≤ q − e − q − ( t 1/4 − 1 ) ≤ q − 2 log ( 2 + t ) 4 e .
where we used log ( 2 + t ) ≤ 4 log ( 2 + t 1 / 4 ) ≤ 4 ( 1 + t 1 / 4 ) \log(2+t)\le4\log(2+t^{1/4})\le4(1+t^{1/4}) log ( 2 + t ) ≤ 4 log ( 2 + t 1/4 ) ≤ 4 ( 1 + t 1/4 ) and ( 1 + y ) e − q − y ≤ q − − 1 (1+y)e^{-q_-y}\le q_-^{-1} ( 1 + y ) e − q − y ≤ q − − 1 for y ≥ 0 y\ge0 y ≥ 0 . Altogether we therefore obtain
P z ( τ > t ) ≤ 576 q − 3 log ( 2 + t ) + 48 q − log ( 2 + t ) + 4 e q − 2 log ( 2 + t ) ≤ 635 q − − 3 log ( 2 + t ) , (60) \begin{aligned}
\mathbb{P}_z(\tau>t)\le\frac{576}{q_-^3\log(2+t)}+\frac{48}{q_-\log(2+t)}+\frac{4e}{q_-^2\log(2+t)} \\
&\le\frac{635q_-^{-3}}{\log(2+t)}, \tag*{(60)}
\end{aligned} P z ( τ > t ) ≤ q − 3 log ( 2 + t ) 576 + q − log ( 2 + t ) 48 + q − 2 log ( 2 + t ) 4 e ≤ log ( 2 + t ) 635 q − − 3 , (60)
which proves the first claim for t ≥ 512 t\ge512 t ≥ 512 . For t < 512 t<512 t < 512 it holds trivially, as 635 / log ( 514 ) > 1 635/\log(514)>1 635/ log ( 514 ) > 1 .
The second bound now follows by integrating over the tail estimate on P z ( τ > t ) \mathbb{P}_z(\tau>t) P z ( τ > t ) . That is, for t > 4 t>4 t > 4 we have
E z ( τ ∧ t ) = ∫ 0 t P z ( τ > s ) d s ≤ t + ∫ t t K 0 log ( 2 + s ) d s ≤ t + K 0 t log ( 2 + t ) ≤ 3 K 0 t log ( 2 + t ) , (61) \begin{aligned}
\mathbb{E}_z(\tau\wedge t)=\int_0^t\mathbb{P}_z(\tau>s)\,ds \\
&\le\sqrt{t}+\int_{\sqrt{t}}^t\frac{K_0}{\log(2+s)}\,ds \\
&\le\sqrt{t}+\frac{K_0t}{\log(2+\sqrt{t})} \\
&\le\frac{3K_0t}{\log(2+t)}, \tag*{(61)}
\end{aligned} E z ( τ ∧ t ) = ∫ 0 t P z ( τ > s ) d s ≤ t + ∫ t t log ( 2 + s ) K 0 d s ≤ t + log ( 2 + t ) K 0 t ≤ log ( 2 + t ) 3 K 0 t , (61)
where in the last step we used ( 2 + t ) 2 ≥ 2 + t (2+\sqrt{t})^2\ge2+t ( 2 + t ) 2 ≥ 2 + t and t ≤ t / log ( 2 + t ) ≤ K 0 t / log ( 2 + t ) \sqrt{t}\le t/\log(2+t)\le K_0t/\log(2+t) t ≤ t / log ( 2 + t ) ≤ K 0 t / log ( 2 + t ) , as log ( 2 + t ) ≤ t \log(2+t)\le\sqrt{t} log ( 2 + t ) ≤ t for t > 4 t>4 t > 4 . For t ≤ 4 t\le4 t ≤ 4 the bound follows from E z ( τ ∧ t ) ≤ t \mathbb{E}_z(\tau\wedge t)\le t E z ( τ ∧ t ) ≤ t and log ( 2 + t ) ≤ log 6 < 3 K 0 \log(2+t)\le\log6<3K_0 log ( 2 + t ) ≤ log 6 < 3 K 0 . ■
5.1.1. Proof of the upper bound in Theorem 2.5. By Theorem 3.9 it is sufficient to upper bound V i V_i V i on a shell around D j D_j D j . Theorem 4.3 turns this into an estimate of the occupation time G Ω ( z , A i δ ) \mathcal{G}^{\Omega}(z,A_i^\delta) G Ω ( z , A i δ ) . To this end, fix δ = min { d min / 2 , 1 } \delta=\min\{d_{\min}/2,1\} δ = min { d m i n /2 , 1 } , let z ∈ ∂ B r j + δ ( x j ) z\in\partial B_{r_j+\delta}(x_j) z ∈ ∂ B r j + δ ( x j ) and y ∈ A i δ y\in A_i^\delta y ∈ A i δ . Since ρ ≤ 1 / 2 \rho\le1/2 ρ ≤ 1/2 , we have ∣ z ⊥ ∣ , ∣ y ⊥ ∣ ≤ ρ + δ ≤ 2 |z_\perp|,|y_\perp|\le\rho+\delta\le2 ∣ z ⊥ ∣ , ∣ y ⊥ ∣ ≤ ρ + δ ≤ 2 , and (2.8) yields ∣ z − x i ∣ ≥ d i j − r j − δ ≥ r i + δ |z-x_i|\ge d_{ij}-r_j-\delta\ge r_i+\delta ∣ z − x i ∣ ≥ d ij − r j − δ ≥ r i + δ , so that Theorem 4.3 and Theorem 5.3 apply. Rewriting as in (4.9) allows us to make use of Theorem 4.2. Concretely, we apply it with Π = Π ∥ \Pi=\Pi_\parallel Π = Π ∥ : let x , y ∈ Ω x,y\in\Omega x , y ∈ Ω with ∣ x ⊥ ∣ , ∣ y ⊥ ∣ ≤ 2 |x_\perp|,|y_\perp|\le2 ∣ x ⊥ ∣ , ∣ y ⊥ ∣ ≤ 2 and R = ∣ Π ∥ ( x − y ) ∣ > 0 R=|\Pi_\parallel(x-y)|>0 R = ∣ Π ∥ ( x − y ) ∣ > 0 . We then have
p t Ω ( x , y ) ≤ ( 2 / π t ) 3 / 2 [ P y ( τ > t / 4 ) P x ( τ > t / 2 , ∣ Π ∥ ( B ( t / 2 ) − x ) ∣ ≥ R / 2 ) + P x ( τ > t / 4 ) P y ( τ > t / 2 , ∣ Π ∥ ( B ( t / 2 ) − y ) ∣ ≥ R / 2 ) ] . (62) p_t^\Omega(x,y)\le(2/\pi t)^{3/2}\bigl[\mathbb{P}_y(\tau>t/4)\mathbb{P}_x(\tau>t/2,|\Pi_\parallel(B(t/2)-x)|\ge R/2)
+\mathbb{P}_x(\tau>t/4)\mathbb{P}_y(\tau>t/2,|\Pi_\parallel(B(t/2)-y)|\ge R/2)\bigr]. \tag*{(62)} p t Ω ( x , y ) ≤ ( 2/ π t ) 3/2 [ P y ( τ > t /4 ) P x ( τ > t /2 , ∣ Π ∥ ( B ( t /2 ) − x ) ∣ ≥ R /2 ) + P x ( τ > t /4 ) P y ( τ > t /2 , ∣ Π ∥ ( B ( t /2 ) − y ) ∣ ≥ R /2 ) ] . (62)
To bound P x ( τ > t / 2 , ∣ Π ∥ ( B ( t / 2 ) − x ) ∣ ≥ R / 2 ) \mathbb{P}_x(\tau>t/2,|\Pi_\parallel(B(t/2)-x)|\ge R/2) P x ( τ > t /2 , ∣ Π ∥ ( B ( t /2 ) − x ) ∣ ≥ R /2 ) , let T = τ ∧ t / 2 T=\tau\wedge t/2 T = τ ∧ t /2 and set M ( t ) = ∣ Π ∥ ( B ( t ) − x ) ∣ 2 − t \mathcal{M}(t)=|\Pi_\parallel(B(t)-x)|^2-t M ( t ) = ∣ Π ∥ ( B ( t ) − x ) ∣ 2 − t . This is a martingale; applying the optional stopping theorem [25 Proposition 2.42] to T ≤ t / 2 T\le t/2 T ≤ t /2 (admissible, as sup u ≤ t / 2 ∣ M ( u ) ∣ \sup_{u\le t/2}|\mathcal{M}(u)| sup u ≤ t /2 ∣ M ( u ) ∣ is integrable by Doob’s inequality), we therefore have
E x [ M ( T ) ] = E x [ M ( 0 ) ] = 0. (63) \mathbb{E}_x[\mathcal{M}(T)]=\mathbb{E}_x[\mathcal{M}(0)]=0. \tag*{(63)} E x [ M ( T )] = E x [ M ( 0 )] = 0. (63)
A Markov inequality with Theorem 5.3 thus yields
P x ( τ > t / 2 , ∣ Π ∥ ( B ( t / 2 ) − x ) ∣ ≥ R / 2 ) ≤ 6 K 0 t R 2 log ( 2 + t / 2 ) . (64) \mathbb{P}_x(\tau>t/2,|\Pi_\parallel(B(t/2)-x)|\ge R/2)\le\frac{6K_0t}{R^2\log(2+t/2)}. \tag*{(64)} P x ( τ > t /2 , ∣ Π ∥ ( B ( t /2 ) − x ) ∣ ≥ R /2 ) ≤ R 2 log ( 2 + t /2 ) 6 K 0 t . (64)
The above bound together with the last part of Theorem 4.2 thus yields
p t Ω ( x , y ) ≤ 12 ( 2 / π t ) 3 / 2 K 0 log ( 2 + t / 4 ) K 0 t R 2 log ( 2 + t / 2 ) p_t^\Omega(x,y)\le12(2/\pi t)^{3/2}\frac{K_0}{\log(2+t/4)}\frac{K_0t}{R^2\log(2+t/2)} p t Ω ( x , y ) ≤ 12 ( 2/ π t ) 3/2 log ( 2 + t /4 ) K 0 R 2 log ( 2 + t /2 ) K 0 t
≤ 12 ( 2 / π t ) 3 / 2 K 0 2 t R 2 log 2 ( 2 + t / 4 ) . (65) \le12(2/\pi t)^{3/2}\frac{K_0^2t}{R^2\log^2(2+t/4)}. \tag*{(65)} ≤ 12 ( 2/ π t ) 3/2 R 2 log 2 ( 2 + t /4 ) K 0 2 t . (65)
Alternatively, we may apply Theorem 5.3 together with the semi-group property from Theorem 4.1 to obtain
p t Ω ( x , y ) ≤ ( 2 / π ) 3 / 2 K 0 2 t 3 / 2 log 2 ( 2 + t / 4 ) min ( 1 , 12 t R 2 ) . (66) p_t^{\Omega}(x,y) \le(2/\pi)^{3/2}\frac{K_0^2}{t^{3/2}\log^2(2+t/4)}\min\left(1,\frac{12t}{R^2}\right). \tag*{(66)} p t Ω ( x , y ) ≤ ( 2/ π ) 3/2 t 3/2 log 2 ( 2 + t /4 ) K 0 2 min ( 1 , R 2 12 t ) . (66)
Let now t ∗ = R 2 / log 4 R t^*=R^2/\log^4 R t ∗ = R 2 / log 4 R and assume R ≥ e 6 R\ge e^6 R ≥ e 6 , we then have
G Ω ( x , y ) = ∫ 0 ∞ p t Ω ( x , y ) d t = ∫ 0 t ∗ p t Ω ( x , y ) d t + ∫ t ∗ R 2 p t Ω ( x , y ) d t + ∫ R 2 ∞ p t Ω ( x , y ) d t . (67) \begin{aligned}
\mathcal{G}^{\Omega}(x,y)=\int_0^\infty p_t^{\Omega}(x,y)\,dt \\
&=\int_0^{t^*}p_t^{\Omega}(x,y)\,dt+\int_{t^*}^{R^2}p_t^{\Omega}(x,y)\,dt+\int_{R^2}^{\infty}p_t^{\Omega}(x,y)\,dt. \tag*{(67)}
\end{aligned} G Ω ( x , y ) = ∫ 0 ∞ p t Ω ( x , y ) d t = ∫ 0 t ∗ p t Ω ( x , y ) d t + ∫ t ∗ R 2 p t Ω ( x , y ) d t + ∫ R 2 ∞ p t Ω ( x , y ) d t . (67)
For the first integral, we have
∫ 0 t ∗ p t Ω ( x , y ) d t ≤ ∫ 0 t ∗ ( 2 / π ) 3 / 2 K 0 2 t log 2 ( 2 + t / 4 ) 12 R 2 d t ≤ ( 2 / π ) 3 / 2 24 t ∗ K 0 2 R 2 log 2 2 = ( 2 / π ) 3 / 2 24 K 0 2 R ( log R log 2 ) 2 . (68) \begin{aligned}
\int_0^{t^*}p_t^{\Omega}(x,y)\,dt\le\int_0^{t^*}(2/\pi)^{3/2}\frac{K_0^2}{\sqrt{t}\log^2(2+t/4)}\frac{12}{R^2}\,dt \\
&\le(2/\pi)^{3/2}\frac{24\sqrt{t^*}K_0^2}{R^2\log^2 2} \\
&=(2/\pi)^{3/2}\frac{24K_0^2}{R(\log R\log2)^2}. \tag*{(68)}
\end{aligned} ∫ 0 t ∗ p t Ω ( x , y ) d t ≤ ∫ 0 t ∗ ( 2/ π ) 3/2 t log 2 ( 2 + t /4 ) K 0 2 R 2 12 d t ≤ ( 2/ π ) 3/2 R 2 log 2 2 24 t ∗ K 0 2 = ( 2/ π ) 3/2 R ( log R log 2 ) 2 24 K 0 2 . (68)
The second integral yields
∫ t ∗ R 2 p t Ω ( x , y ) d t ≤ ( 2 / π ) 3 / 2 12 K 0 2 R 2 log 2 ( 2 + t ∗ / 4 ) ∫ t ∗ R 2 d t t ≤ ( 2 / π ) 3 / 2 24 K 0 2 R log 2 ( 2 + t ∗ / 4 ) ≤ ( 2 / π ) 3 / 2 96 K 0 2 R log 2 R , (69) \begin{aligned}
\int_{t^*}^{R^2}p_t^{\Omega}(x,y)\,dt\le(2/\pi)^{3/2}\frac{12K_0^2}{R^2\log^2(2+t^*/4)}\int_{t^*}^{R^2}\frac{dt}{\sqrt{t}}
\le(2/\pi)^{3/2}\frac{24K_0^2}{R\log^2(2+t^*/4)} \\
&\le(2/\pi)^{3/2}\frac{96K_0^2}{R\log^2 R}, \tag*{(69)}
\end{aligned} ∫ t ∗ R 2 p t Ω ( x , y ) d t ≤ ( 2/ π ) 3/2 R 2 log 2 ( 2 + t ∗ /4 ) 12 K 0 2 ∫ t ∗ R 2 t d t ≤ ( 2/ π ) 3/2 R log 2 ( 2 + t ∗ /4 ) 24 K 0 2 ≤ ( 2/ π ) 3/2 R log 2 R 96 K 0 2 , (69)
where we used that log ( 2 + t ∗ / 4 ) ≥ 2 log R − 4 log log R − log 4 ≥ 1 2 log R \log(2+t^*/4)\ge2\log R-4\log\log R-\log4\ge\frac{1}{2}\log R log ( 2 + t ∗ /4 ) ≥ 2 log R − 4 log log R − log 4 ≥ 2 1 log R for R ≥ e 6 R\ge e^6 R ≥ e 6 . For the third one, we obtain
∫ R 2 ∞ p t Ω ( x , y ) d t ≤ ( 2 / π ) 3 / 2 2 K 0 2 R log 2 ( 2 + R 2 / 4 ) . (70) \int_{R^2}^{\infty}p_t^{\Omega}(x,y)\,dt\le(2/\pi)^{3/2}\frac{2K_0^2}{R\log^2(2+R^2/4)}. \tag*{(70)} ∫ R 2 ∞ p t Ω ( x , y ) d t ≤ ( 2/ π ) 3/2 R log 2 ( 2 + R 2 /4 ) 2 K 0 2 . (70)
Putting everything together, we have
G Ω ( x , y ) ≤ ( 2 / π ) 3 / 2 ( 24 K 0 2 R ( log R log 2 ) 2 + 96 K 0 2 R log 2 R + 2 K 0 2 R log 2 ( 2 + R 2 / 4 ) ) ≤ C K 0 2 R log 2 R , (71) \begin{aligned}
\mathcal{G}^{\Omega}(x,y)\le(2/\pi)^{3/2}\left(\frac{24K_0^2}{R(\log R\log2)^2}+\frac{96K_0^2}{R\log^2 R}+\frac{2K_0^2}{R\log^2(2+R^2/4)}\right) \\
&\le\frac{CK_0^2}{R\log^2 R}, \tag*{(71)}
\end{aligned} G Ω ( x , y ) ≤ ( 2/ π ) 3/2 ( R ( log R log 2 ) 2 24 K 0 2 + R log 2 R 96 K 0 2 + R log 2 ( 2 + R 2 /4 ) 2 K 0 2 ) ≤ R log 2 R C K 0 2 , (71)
with an absolute constant C C C . Integrating over A i δ A_i^\delta A i δ , we therefore have
V i ( z ) ≤ 12 δ 2 G Ω ( z , A i δ ) ≤ C ′ K 0 2 δ 2 d i j log 2 ( 2 + d i j ) , d i j ≥ 2 e 6 , (72) V_i(z)\le\frac{12}{\delta^2}\mathcal{G}^{\Omega}(z,A_i^\delta)\le\frac{C'K_0^2}{\delta^2d_{ij}\log^2(2+d_{ij})},\qquad d_{ij}\ge2e^6, \tag*{(72)} V i ( z ) ≤ δ 2 12 G Ω ( z , A i δ ) ≤ δ 2 d ij log 2 ( 2 + d ij ) C ′ K 0 2 , d ij ≥ 2 e 6 , (72)
where we used that R = ∣ Π ∥ ( z − y ) ∣ ≥ ∣ s i − s j ∣ − ( r i + δ ) − ( r j + δ ) ≥ d i j − 4 ≥ d i j / 2 ≥ e 6 R=|\Pi_{\parallel}(z-y)|\ge|s_i-s_j|-(r_i+\delta)-(r_j+\delta)\ge d_{ij}-4\ge d_{ij}/2\ge e^6 R = ∣ Π ∥ ( z − y ) ∣ ≥ ∣ s i − s j ∣ − ( r i + δ ) − ( r j + δ ) ≥ d ij − 4 ≥ d ij /2 ≥ e 6 for every y ∈ A i δ y\in A_i^\delta y ∈ A i δ , as ∣ a i − a j ∣ ≤ 2 ρ ≤ 1 |a_i-a_j|\le2\rho\le1 ∣ a i − a j ∣ ≤ 2 ρ ≤ 1 and r k + δ ≤ 3 / 2 r_k+\delta\le3/2 r k + δ ≤ 3/2 , that R ↦ R log 2 R R\mapsto R\log^2R R ↦ R log 2 R is increasing, and that log ( d i j / 2 ) ≥ 1 2 log ( 2 + d i j ) \log(d_{ij}/2)\ge\frac{1}{2}\log(2+d_{ij}) log ( d ij /2 ) ≥ 2 1 log ( 2 + d ij ) for d i j ≥ 8 d_{ij}\ge8 d ij ≥ 8 . Taking the supremum over z ∈ ∂ B r j + δ ( x j ) z\in\partial B_{r_j+\delta}(x_j) z ∈ ∂ B r j + δ ( x j ) in (3.13) yields the upper bound for d i j ≥ 2 e 6 d_{ij}\ge2e^6 d ij ≥ 2 e 6 . For d i j < 2 e 6 d_{ij}<2e^6 d ij < 2 e 6 it follows from V i ≤ 1 V_i\le1 V i ≤ 1 and (3.13), as d ↦ d log 2 ( 2 + d ) d\mapsto d\log^2(2+d) d ↦ d log 2 ( 2 + d ) is bounded on bounded sets.
Screens
The upper bounds for screens follows the same outline of the decay estimate for chains. We begin by defining the slabs
S k : = { z ∈ R 3 : ∣ z ∥ ∣ < 2 k } . (73) S_k:=\{z\in\mathbb{R}^3:|z_{\parallel}|<2^k\}. \tag*{(73)} S k := { z ∈ R 3 : ∣ z ∥ ∣ < 2 k } . (73)
As before, a uniform killing bound holds:
Lemma 5.4. Let α = inf { t ≥ 0 : ∣ B ∥ ( t ) ∣ ≥ 2 } \alpha=\inf\{t\geq0:|B_{\parallel}(t)|\geq2\} α = inf { t ≥ 0 : ∣ B ∥ ( t ) ∣ ≥ 2 } be the exit time from S 1 S_1 S 1 . For every z ∈ ∂ S 0 z\in\partial S_0 z ∈ ∂ S 0 ,
P z ( τ < α ) ≥ q − ∈ ( 0 , 1 ) . (74) \mathbb{P}_z(\tau<\alpha)\geq q_{-}\in(0,1). \tag*{(74)} P z ( τ < α ) ≥ q − ∈ ( 0 , 1 ) . (74) The constant q − q_{-} q − depends only on r min r_{\min} r m i n and L L L .
Proof. The proof follows mutatis mutandis from Theorem 5.1.
As the reader may expect, the next step is to adapt Theorem 5.2.
Lemma 5.5. Let z ∈ Ω z\in\Omega z ∈ Ω with ∣ z ∥ ∣ ≤ 2 \lvert z_{\parallel}\rvert\leq2 ∣ z ∥ ∣ ≤ 2 . Let β 0 = 0 \beta_0=0 β 0 = 0 , α 1 = inf { t ≥ 0 : ∣ B ∥ ( t ) ∣ ≥ 2 } \alpha_1=\inf\{t\geq0:\lvert B_{\parallel}(t)\rvert\geq2\} α 1 = inf { t ≥ 0 : ∣ B ∥ ( t )∣ ≥ 2 } (the time α \alpha α of Theorem 5.4) and, for k ≥ 1 k\geq1 k ≥ 1 ,
β k = inf { t ≥ α k : ∣ B ∥ ( t ) ∣ ≤ 1 } , α k + 1 = inf { t ≥ β k : ∣ B ∥ ( t ) ∣ ≥ 2 } , \beta_k=\inf\{t\geq\alpha_k:\lvert B_{\parallel}(t)\rvert\leq1\},\qquad\alpha_{k+1}=\inf\{t\geq\beta_k:\lvert B_{\parallel}(t)\rvert\geq2\}, β k = inf { t ≥ α k : ∣ B ∥ ( t )∣ ≤ 1 } , α k + 1 = inf { t ≥ β k : ∣ B ∥ ( t )∣ ≥ 2 } , and write ( a ) + = max { a , 0 } (a)_+=\max\{a,0\} ( a ) + = max { a , 0 } . Let U k = α k − β k − 1 U_k=\alpha_k-\beta_{k-1} U k = α k − β k − 1 (the k k k -th visit to S 1 S_1 S 1 ) and Y k = β k − α k Y_k=\beta_k-\alpha_k Y k = β k − α k (the k k k -th excursion out of S 0 S_0 S 0 ). Then:
(a) The α k , β k \alpha_k,\beta_k α k , β k are stopping times; P z \mathbb{P}_z P z -almost surely they are finite, 0 = β 0 ≤ α 1 < β 1 < α 2 < ⋯ 0=\beta_0\leq\alpha_1<\beta_1<\alpha_2<\cdots 0 = β 0 ≤ α 1 < β 1 < α 2 < ⋯ , and α k → ∞ \alpha_k\to\infty α k → ∞ .
(b) τ ∉ [ α k , β k ] \tau\notin[\alpha_k,\beta_k] τ ∈ / [ α k , β k ] for all k ≥ 1 k\geq1 k ≥ 1 .
(c) P z ( τ > α k ) ≤ ( 1 − q − ) k − 1 \mathbb{P}_z(\tau>\alpha_k)\leq(1-q_-)^{k-1} P z ( τ > α k ) ≤ ( 1 − q − ) k − 1 for all k ≥ 1 k\geq1 k ≥ 1 ; in particular τ < ∞ \tau<\infty τ < ∞ P z \mathbb{P}_z P z -almost surely.
(d) For all k ≥ 1 k\geq1 k ≥ 1 and s > 0 s>0 s > 0 ,
P z ( τ > β k − 1 , U k > s ) ≤ ( 1 − q − ) ( k − 2 ) + min { 1 , 4 / s } , (75) \mathbb{P}_z(\tau>\beta_{k-1},U_k>s)\leq(1-q_-)^{(k-2)_+}\min\{1,4/s\}, \tag*{(75)} P z ( τ > β k − 1 , U k > s ) ≤ ( 1 − q − ) ( k − 2 ) + min { 1 , 4/ s } , (75) P z ( τ > α k , Y k > s ) ≤ ( 1 − q − ) k − 1 min { 1 , s − 1 / 2 } . (76) \mathbb{P}_z(\tau>\alpha_k,Y_k>s)\leq(1-q_-)^{k-1}\min\{1,s^{-1/2}\}. \tag*{(76)} P z ( τ > α k , Y k > s ) ≤ ( 1 − q − ) k − 1 min { 1 , s − 1/2 } . (76)
Proof. Items (a)-(c) follow from the same arguments as in Theorem 5.2, where in (b) we now use ∣ B ∥ ∣ ≥ 1 > ρ ≥ ∣ s i ∣ + r i \lvert B_{\parallel}\rvert\geq1>\rho\geq\lvert s_i\rvert+r_i ∣ B ∥ ∣ ≥ 1 > ρ ≥ ∣ s i ∣ + r i on [ α k , β k ] [\alpha_k,\beta_k] [ α k , β k ] , by (2.11) and ρ ≤ 1 / 2 \rho\leq1/2 ρ ≤ 1/2 . It thus remains to prove (d). We now note that U k U_k U k is the exit time from ( − 2 , 2 ) (-2,2) ( − 2 , 2 ) of the one dimensional Brownian motion B ∥ B_{\parallel} B ∥ started at B ∥ ( β k − 1 ) ∈ [ − 2 , 2 ] B_{\parallel}(\beta_{k-1})\in[-2,2] B ∥ ( β k − 1 ) ∈ [ − 2 , 2 ] . Let u < 0 < v u<0<v u < 0 < v , if we denote by T T T the exit time of standard scalar Brownian motion from the interval [ u , v ] [u,v] [ u , v ] , it holds that E 0 [ T ] = ∣ u ∣ v \mathbb{E}_0[T]=\lvert u\rvert v E 0 [ T ] = ∣ u ∣ v [25 Theorem 2.49] , so that the expected exit time from ( − 2 , 2 ) (-2,2) ( − 2 , 2 ) of a Brownian motion started at c ∈ [ − 2 , 2 ] c\in[-2,2] c ∈ [ − 2 , 2 ] equals ( 2 − c ) ( 2 + c ) ≤ 4 (2-c)(2+c)\leq4 ( 2 − c ) ( 2 + c ) ≤ 4 . The strong Markov property together with an application of the Markov inequality thus yields
P z ( τ > β k − 1 , U k > s ) ≤ ( 1 − q − ) ( k − 2 ) + min { 1 , 4 / s } . (77) \mathbb{P}_z(\tau>\beta_{k-1},U_k>s)\leq(1-q_-)^{(k-2)_+}\min\{1,4/s\}. \tag*{(77)} P z ( τ > β k − 1 , U k > s ) ≤ ( 1 − q − ) ( k − 2 ) + min { 1 , 4/ s } . (77)
Similarly, Y k Y_k Y k can be understood as the hitting time of a point at distance 1 1 1 by a linear Brownian motion W W W started at 0 0 0 . By the reflection principle [25 Theorem 2.21] , P ( max [ 0 , s ] W < 1 ) = P ( ∣ W ( s ) ∣ < 1 ) ≤ 2 / 2 π s ≤ s − 1 / 2 \mathbb{P}(\max_{[0,s]}W<1)=\mathbb{P}(\lvert W(s)\rvert<1)\leq2/\sqrt{2\pi s}\leq s^{-1/2} P ( max [ 0 , s ] W < 1 ) = P (∣ W ( s )∣ < 1 ) ≤ 2/ 2 π s ≤ s − 1/2 , hence
P z ( τ > α k , Y k > s ) ≤ ( 1 − q − ) k − 1 min { 1 , s − 1 / 2 } . (78) \mathbb{P}_z(\tau>\alpha_k,Y_k>s)\leq(1-q_-)^{k-1}\min\{1,s^{-1/2}\}. \tag*{(78)} P z ( τ > α k , Y k > s ) ≤ ( 1 − q − ) k − 1 min { 1 , s − 1/2 } . (78)
Proposition 5.6 (Logarithmic survival bound). For all z ∈ Ω z\in\Omega z ∈ Ω with ∣ z ∥ ∣ ≤ 2 \lvert z_{\parallel}\rvert\leq2 ∣ z ∥ ∣ ≤ 2 and t ≥ 0 t\geq0 t ≥ 0 ,
P z ( τ > t ) ≤ 68 q − − 3 min { 1 , t − 1 / 2 } and E z [ ( τ ∧ t ) 2 ] ≤ 272 t 3 / 2 3 q − 3 . (79) \mathbb{P}_z(\tau>t)\leq68q_-^{-3}\min\{1,t^{-1/2}\}
\qquad\text{and}\qquad
\mathbb{E}_z[(\tau\wedge t)^2]\leq\frac{272t^{3/2}}{3q_-^3}. \tag*{(79)} P z ( τ > t ) ≤ 68 q − − 3 min { 1 , t − 1/2 } and E z [( τ ∧ t ) 2 ] ≤ 3 q − 3 272 t 3/2 . (79)
Proof. The proof is similar to that of Theorem 5.3. Letting again w k = 1 k ( k + 1 ) w_k=\frac{1}{k(k+1)} w k = k ( k + 1 ) 1 and using Theorem 5.5, we obtain
P z ( τ > t ) ≤ P z ( α G > t ) ≤ P z ( ⋃ k ≥ 1 { τ > α k , Y k > w k t / 2 } ∪ { τ > β k − 1 , U k > w k t / 2 } ) ≤ ∑ k ≥ 1 ( 1 − q − ) ( k − 2 ) + min { 1 , 8 k ( k + 1 ) t } + ∑ k ≥ 1 ( 1 − q − ) k − 1 min { 1 , ( 2 k ( k + 1 ) t ) 1 / 2 } . (80) \begin{aligned}
\mathbb{P}_z(\tau>t)\leq\mathbb{P}_z(\alpha_G>t) \\
&\leq\mathbb{P}_z\left(\bigcup_{k\geq1}\{\tau>\alpha_k,Y_k>w_kt/2\}\cup\{\tau>\beta_{k-1},U_k>w_kt/2\}\right) \\
&\leq\sum_{k\geq1}(1-q_-)^{(k-2)_+}\min\left\{1,\frac{8k(k+1)}{t}\right\} \\
&+\sum_{k\geq1}(1-q_-)^{k-1}\min\left\{1,\left(\frac{2k(k+1)}{t}\right)^{1/2}\right\}. \tag*{(80)}
\end{aligned} P z ( τ > t ) ≤ P z ( α G > t ) ≤ P z ( k ≥ 1 ⋃ { τ > α k , Y k > w k t /2 } ∪ { τ > β k − 1 , U k > w k t /2 } ) ≤ k ≥ 1 ∑ ( 1 − q − ) ( k − 2 ) + min { 1 , t 8 k ( k + 1 ) } + k ≥ 1 ∑ ( 1 − q − ) k − 1 min { 1 , ( t 2 k ( k + 1 ) ) 1/2 } . (80)
An argument as in Theorem 5.3 yields
∑ k ≥ 1 ( 1 − q − ) ( k − 2 ) + min { 1 , 8 k ( k + 1 ) t } ≤ 64 q − − 3 min ( 1 , t − 1 ) , (81) \sum_{k\ge1}(1-q_-)^{(k-2)+}\min\left\{1,\frac{8k(k+1)}{t}\right\}\le64q_-^{-3}\min\left(1,t^{-1}\right), \tag*{(81)} k ≥ 1 ∑ ( 1 − q − ) ( k − 2 ) + min { 1 , t 8 k ( k + 1 ) } ≤ 64 q − − 3 min ( 1 , t − 1 ) , (81)
while the second sum can be bounded by
∑ k ≥ 1 ( 1 − q − ) k − 1 min { 1 , ( 2 k ( k + 1 ) t ) 1 / 2 } ≤ 4 q − 2 min ( 1 , t − 1 / 2 ) . (82) \sum_{k\ge1}(1-q_-)^{k-1}\min\left\{1,\left(\frac{2k(k+1)}{t}\right)^{1/2}\right\}\le\frac{4}{q_-^2}\min\left(1,t^{-1/2}\right). \tag*{(82)} k ≥ 1 ∑ ( 1 − q − ) k − 1 min { 1 , ( t 2 k ( k + 1 ) ) 1/2 } ≤ q − 2 4 min ( 1 , t − 1/2 ) . (82)
Putting these two together, we have
P z ( τ > t ) ≤ 68 q − − 3 min ( 1 , t − 1 / 2 ) . (83) \mathbb{P}_z(\tau>t)\le68q_-^{-3}\min\left(1,t^{-1/2}\right). \tag*{(83)} P z ( τ > t ) ≤ 68 q − − 3 min ( 1 , t − 1/2 ) . (83)
Integrating this tail bound, we obtain
E z [ ( τ ∧ t ) 2 ] = 2 ∫ 0 t s P z ( τ > s ) d s ≤ 136 q − − 3 ∫ 0 t s 1 / 2 d s ≤ 272 t 3 / 2 3 q − 3 . (84) \begin{aligned}
\mathbb{E}_z[(\tau\wedge t)^2]&=2\int_0^t s\mathbb{P}_z(\tau>s)\,ds\\
&\le136q_-^{-3}\int_0^t s^{1/2}\,ds\\
&\le\frac{272t^{3/2}}{3q_-^3}. \tag*{(84)}
\end{aligned} E z [( τ ∧ t ) 2 ] = 2 ∫ 0 t s P z ( τ > s ) d s ≤ 136 q − − 3 ∫ 0 t s 1/2 d s ≤ 3 q − 3 272 t 3/2 . (84)
5.2.1. Proof of the upper bound in Theorem 2.6. The strategy remains the same as in the upper bound of Theorem 2.5 . Let again δ = min { d min / 2 , 1 } \delta=\min\{d_{\min}/2,1\} δ = min { d m i n /2 , 1 } , and let x , y ∈ Ω x,y\in\Omega x , y ∈ Ω with ∣ x ∥ ∣ , ∣ y ∥ ∣ ≤ 2 |x_\parallel|,|y_\parallel|\le2 ∣ x ∥ ∣ , ∣ y ∥ ∣ ≤ 2 and R = ∣ Π ⊥ ( x − y ) ∣ > 0 R=|\Pi_\perp(x-y)|>0 R = ∣ Π ⊥ ( x − y ) ∣ > 0 . By Theorem 4.2 (applied with Π = Π ⊥ \Pi=\Pi_\perp Π = Π ⊥ and p = 4 p=4 p = 4 ) we thus have
p t Ω ( x , y ) ≤ ( 2 / π t ) 3 / 2 16 R 4 [ P y ( τ > t / 4 ) E x [ ∣ Π ⊥ ( B ( τ ∧ t / 2 ) − x ) ∣ 4 ] + P x ( τ > t / 4 ) E y [ ∣ Π ⊥ ( B ( τ ∧ t / 2 ) − y ) ∣ 4 ] ] . (85) p_t^\Omega(x,y)\le(2/\pi t)^{3/2}\frac{16}{R^4}\left[\mathbb{P}_y(\tau>t/4)\mathbb{E}_x\left[|\Pi_\perp(B(\tau\wedge t/2)-x)|^4\right]+\mathbb{P}_x(\tau>t/4)\mathbb{E}_y\left[|\Pi_\perp(B(\tau\wedge t/2)-y)|^4\right]\right]. \tag*{(85)} p t Ω ( x , y ) ≤ ( 2/ π t ) 3/2 R 4 16 [ P y ( τ > t /4 ) E x [ ∣ Π ⊥ ( B ( τ ∧ t /2 ) − x ) ∣ 4 ] + P x ( τ > t /4 ) E y [ ∣ Π ⊥ ( B ( τ ∧ t /2 ) − y ) ∣ 4 ] ] . (85)
Once again, noting that W ( t ) = Π ⊥ ( B ( t ) − x ) W(t)=\Pi_\perp(B(t)-x) W ( t ) = Π ⊥ ( B ( t ) − x ) is a planar Brownian motion under P x \mathbb{P}_x P x and that
M ( t ) = ∣ W ( t ) ∣ 4 − 8 t ∣ W ( t ) ∣ 2 + 8 t 2 (86) \mathcal{M}(t)=|W(t)|^4-8t|W(t)|^2+8t^2 \tag*{(86)} M ( t ) = ∣ W ( t ) ∣ 4 − 8 t ∣ W ( t ) ∣ 2 + 8 t 2 (86)
is a martingale, applying the optional stopping Theorem [25 Proposition 2.42] to the bounded stopping time T = τ ∧ t / 2 T=\tau\wedge t/2 T = τ ∧ t /2 (admissible, as sup u ≤ t / 2 ∣ M ( u ) ∣ \sup_{u\le t/2}|\mathcal{M}(u)| sup u ≤ t /2 ∣ M ( u ) ∣ is integrable by Doob’s inequality) yields E x [ M ( T ) ] = 0 \mathbb{E}_x[\mathcal{M}(T)]=0 E x [ M ( T )] = 0 , hence
E x [ ∣ W ( T ) ∣ 4 ] = 8 E x [ T ∣ W ( T ) ∣ 2 ] − 8 E x [ T 2 ] ≤ 8 E x [ T 2 ] 1 / 2 E x [ ∣ W ( T ) ∣ 4 ] 1 / 2 . (87) \mathbb{E}_x\left[|W(T)|^4\right]=8\mathbb{E}_x\left[T|W(T)|^2\right]-8\mathbb{E}_x\left[T^2\right]\le8\mathbb{E}_x\left[T^2\right]^{1/2}\mathbb{E}_x\left[|W(T)|^4\right]^{1/2}. \tag*{(87)} E x [ ∣ W ( T ) ∣ 4 ] = 8 E x [ T ∣ W ( T ) ∣ 2 ] − 8 E x [ T 2 ] ≤ 8 E x [ T 2 ] 1/2 E x [ ∣ W ( T ) ∣ 4 ] 1/2 . (87)
that is, E x [ ∣ Π ⊥ ( B ( τ ∧ t / 2 ) − x ) ∣ 4 ] ≤ 64 E x [ ( τ ∧ t / 2 ) 2 ] \mathbb{E}_x[|\Pi_\perp(B(\tau\wedge t/2)-x)|^4]\le64\mathbb{E}_x[(\tau\wedge t/2)^2] E x [ ∣ Π ⊥ ( B ( τ ∧ t /2 ) − x ) ∣ 4 ] ≤ 64 E x [( τ ∧ t /2 ) 2 ] . Combining this, Theorem 5.6 and Equation (85) , we obtain
p t Ω ( x , y ) ≤ ( 2 / π t ) 3 / 2 16 R 4 ⋅ 2 ⋅ 68 q − − 3 min ( 1 , 2 t − 1 / 2 ) ⋅ 64 ⋅ 272 3 q − 3 ( t 2 ) 3 / 2 ≤ K 1 R 4 min ( 1 , t − 1 / 2 ) , (88) p_t^\Omega(x,y)\le(2/\pi t)^{3/2}\frac{16}{R^4}\cdot2\cdot68q_-^{-3}\min\left(1,2t^{-1/2}\right)\cdot64\cdot\frac{272}{3q_-^3}\left(\frac{t}{2}\right)^{3/2}\le\frac{K_1}{R^4}\min\left(1,t^{-1/2}\right), \tag*{(88)} p t Ω ( x , y ) ≤ ( 2/ π t ) 3/2 R 4 16 ⋅ 2 ⋅ 68 q − − 3 min ( 1 , 2 t − 1/2 ) ⋅ 64 ⋅ 3 q − 3 272 ( 2 t ) 3/2 ≤ R 4 K 1 min ( 1 , t − 1/2 ) , (88)
where K 1 = 4.6 ⋅ 10 6 q − − 6 K_1=4.6\cdot10^6q_-^{-6} K 1 = 4.6 ⋅ 1 0 6 q − − 6 . On the other hand, the semigroup property from Theorem 4.1 and Theorem 5.6 give, as in the proof of the upper bound of Theorem 2.5 ,
p t Ω ( x , y ) ≤ ( 2 / π t ) 3 / 2 P y ( τ > t / 4 ) P x ( τ > t / 2 ) ≤ K 2 t 5 / 2 , t ≥ 4 , (89) p_t^\Omega(x,y)\le(2/\pi t)^{3/2}\mathbb{P}_y(\tau>t/4)\mathbb{P}_x(\tau>t/2)\le\frac{K_2}{t^{5/2}},\qquad t\ge4, \tag*{(89)} p t Ω ( x , y ) ≤ ( 2/ π t ) 3/2 P y ( τ > t /4 ) P x ( τ > t /2 ) ≤ t 5/2 K 2 , t ≥ 4 , (89)
with K 2 = 6644 q − − 6 K_2=6644q_-^{-6} K 2 = 6644 q − − 6 . Let now R ≥ 2 R\ge2 R ≥ 2 . Using (88) on [ 0 , R 2 ] [0,R^2] [ 0 , R 2 ] and (89) on [ R 2 , ∞ ) [R^2,\infty) [ R 2 , ∞ ) , we obtain
G Ω ( x , y ) = ∫ 0 ∞ p t Ω ( x , y ) d t ≤ K 1 R 4 ( 1 + ∫ 1 R 2 d t t ) + K 2 ∫ R 2 ∞ d t t 5 / 2 ≤ 2 K 1 R 3 + 2 K 2 3 R 3 ≤ 3 K 1 R 3 . (90) \mathcal{G}^\Omega(x,y)=\int_0^\infty p_t^\Omega(x,y)\,dt\le\frac{K_1}{R^4}\left(1+\int_1^{R^2}\frac{dt}{\sqrt{t}}\right)+K_2\int_{R^2}^{\infty}\frac{dt}{t^{5/2}}\le\frac{2K_1}{R^3}+\frac{2K_2}{3R^3}\le\frac{3K_1}{R^3}. \tag*{(90)} G Ω ( x , y ) = ∫ 0 ∞ p t Ω ( x , y ) d t ≤ R 4 K 1 ( 1 + ∫ 1 R 2 t d t ) + K 2 ∫ R 2 ∞ t 5/2 d t ≤ R 3 2 K 1 + 3 R 3 2 K 2 ≤ R 3 3 K 1 . (90)
Let finally z ∈ ∂ B r j + δ ( x j ) z \in\partial B_{r_j+\delta}(x_j) z ∈ ∂ B r j + δ ( x j ) and y ∈ A i δ y \in A_i^\delta y ∈ A i δ . Then ∥ z ∥ , ∥ y ∥ ≤ ρ + δ ≤ 2 \lVert z\rVert,\lVert y\rVert\le\rho+\delta\le2 ∥ z ∥ , ∥ y ∥ ≤ ρ + δ ≤ 2 , Theorem 4.3 applies as in the proof of Theorem 2.5 , and R = ∣ Π ⊥ ( z − y ) ∣ ≥ ∣ a i − a j ∣ − ( r i + δ ) − ( r j + δ ) ≥ d i j − 4 ≥ d i j / 2 R=\lvert\Pi^\perp(z-y)\rvert\ge\lvert a_i-a_j\rvert-(r_i+\delta)-(r_j+\delta) \ge d_{ij}-4 \ge d_{ij}/2 R = ∣ Π ⊥ ( z − y )∣ ≥ ∣ a i − a j ∣ − ( r i + δ ) − ( r j + δ ) ≥ d ij − 4 ≥ d ij /2 for d i j ≥ 8 d_{ij}\ge8 d ij ≥ 8 , as ∣ s i − s j ∣ ≤ 2 ρ ≤ 1 \lvert s_i-s_j\rvert\le2\rho\le1 ∣ s i − s j ∣ ≤ 2 ρ ≤ 1 and r k + δ ≤ 3 / 2 r_k+\delta\le3/2 r k + δ ≤ 3/2 . Finally, integrating over A i δ A_i^\delta A i δ , we obtain
V i ( z ) ≤ 12 ∣ A i δ ∣ 3 K 1 δ 2 R 3 ≤ 288 ∣ A i δ ∣ K 1 δ 2 d i j 3 , d i j ≥ 8 , (91) V_i(z) \le\frac{12\lvert A_i^\delta\rvert3K_1}{\delta^2R^3} \le\frac{288\lvert A_i^\delta\rvert K_1}{\delta^2d_{ij}^3}, \qquad d_{ij}\ge8, \tag*{(91)} V i ( z ) ≤ δ 2 R 3 12 ∣ A i δ ∣ 3 K 1 ≤ δ 2 d ij 3 288 ∣ A i δ ∣ K 1 , d ij ≥ 8 , (91)
which proves the claim for d i j ≥ 8 d_{ij}\ge8 d ij ≥ 8 after taking the supremum over z z z in (28) ; for d i j < 8 d_{ij}<8 d ij < 8 it follows from V i ≤ 1 V_i\le1 V i ≤ 1 and (28) .
Crystals
Proving the upper bound for crystals follows a slightly different approach. The assumptions of Theorem 2.7 imply that at any point z ∈ Ω z\in\Omega z ∈ Ω , there exists some q − q_- q − such that
P z ( τ < 1 ) ≥ q − ∈ ( 0 , 1 ) . (92) \mathbb{P}_z(\tau<1)\ge q_-\in(0,1). \tag*{(92)} P z ( τ < 1 ) ≥ q − ∈ ( 0 , 1 ) . (92)
This differs from Theorems 5.1 and 5.4 in that we do not require z z z to lie inside some slab or cylinder. Consequently, this estimate will be enough to force exponential decay.
Lemma 5.7. Under the assumptions of Theorem 2.7 , there exists some q − > 0 q_->0 q − > 0 such that for every z ∈ Ω z\in\Omega z ∈ Ω
P z ( τ < 1 ) ≥ q − . (93) \mathbb{P}_z(\tau<1)\ge q_-. \tag*{(93)} P z ( τ < 1 ) ≥ q − . (93) The constant q − q_- q − depends only on r min r_{\min} r m i n and L L L .
Proof. The proof follows mutatis mutandis from Theorem 5.1. ■
Proposition 5.8. There exist C , μ > 0 C,\mu>0 C , μ > 0 such that for all z ∈ Ω z\in\Omega z ∈ Ω
P z ( τ > t ) ≤ C e − μ t . (94) \mathbb{P}_z(\tau>t)\le Ce^{-\mu t}. \tag*{(94)} P z ( τ > t ) ≤ C e − μ t . (94) Proof. Let E n = { τ > n } E_n=\{\tau>n\} E n = { τ > n } , the strong Markov property then yields
P z ( E n ) = E z [ 1 E n − 1 P B ( n − 1 ) ( τ > 1 ) ] ≤ ( 1 − q − ) P z ( E n − 1 ) . (95) \mathbb{P}_z(E_n)=\mathbb{E}_z\left[1_{E_{n-1}}\mathbb{P}_{B(n-1)}(\tau>1)\right]\le(1-q_-)\mathbb{P}_z(E_{n-1}). \tag*{(95)} P z ( E n ) = E z [ 1 E n − 1 P B ( n − 1 ) ( τ > 1 ) ] ≤ ( 1 − q − ) P z ( E n − 1 ) . (95) By induction, we therefore have
P z ( τ > n ) ≤ ( 1 − q − ) n , (96) \mathbb{P}_z(\tau>n)\le(1-q_-)^n, \tag*{(96)} P z ( τ > n ) ≤ ( 1 − q − ) n , (96) or more generally,
P z ( τ > t ) ≤ ( 1 − q − ) t − 1 . (97) \mathbb{P}_z(\tau>t)\le(1-q_-)^{t-1}. \tag*{(97)} P z ( τ > t ) ≤ ( 1 − q − ) t − 1 . (97) Setting μ = − log ( 1 − q − ) \mu=-\log(1-q_-) μ = − log ( 1 − q − ) and C = ( 1 − q − ) − 1 C=(1-q_-)^{-1} C = ( 1 − q − ) − 1 , the claim follows. ■
Proof of the upper bound in Theorem 2.7
Once again we want to obtain an upper bound on the killed transition density p t Ω p_t^\Omega p t Ω . Thanks to the exponential decay bound of Theorem 5.8, the estimate is much less involved: The semigroup property of Theorem 4.1 together with p t / 2 Ω ≤ p t / 2 ≤ ( π t ) − 3 / 2 p_{t/2}^\Omega\le p_{t/2}\le(\pi t)^{-3/2} p t /2 Ω ≤ p t /2 ≤ ( π t ) − 3/2 and Theorem 5.8 yield
p t Ω ( x , y ) ≤ C π − 3 / 2 t − 3 / 2 e − μ t / 2 . (98) p_t^\Omega(x,y)\le C\pi^{-3/2}t^{-3/2}e^{-\mu t/2}. \tag*{(98)} p t Ω ( x , y ) ≤ C π − 3/2 t − 3/2 e − μ t /2 . (98)
On the other hand, we always have p t Ω ≤ p t p_t^\Omega\le p_t p t Ω ≤ p t . Taking the geometric mean of these two upper bounds, we thus have
p t Ω ( x , y ) ≤ C t − 3 / 2 exp ( − ∣ x − y ∣ 2 4 t − μ t 4 ) . (99) p_t^\Omega(x,y)\le Ct^{-3/2}\exp\left(-\frac{\lvert x-y\rvert^2}{4t}-\frac{\mu t}{4}\right). \tag*{(99)} p t Ω ( x , y ) ≤ C t − 3/2 exp ( − 4 t ∣ x − y ∣ 2 − 4 μ t ) . (99)
Define
J ( c ) = ∫ 0 ∞ exp ( − u 2 − c 2 u 2 ) d u , (100) J(c)=\int_0^\infty\exp\left(-u^2-\frac{c^2}{u^2}\right)\,du, \tag*{(100)} J ( c ) = ∫ 0 ∞ exp ( − u 2 − u 2 c 2 ) d u , (100)
by substituting u = ∣ x − y ∣ / ( 2 t ) u=\lvert x-y\rvert/(2\sqrt{t}) u = ∣ x − y ∣ / ( 2 t ) one can show that
G Ω ( x , y ) = ∫ 0 ∞ p t Ω ( x , y ) d t ≤ 4 C ∣ x − y ∣ J ( ∣ x − y ∣ μ 4 ) . (101) G^\Omega(x,y)=\int_0^\infty p_t^\Omega(x,y)\,dt\le\frac{4C}{\lvert x-y\rvert}J\left(\frac{\lvert x-y\rvert\sqrt{\mu}}{4}\right). \tag*{(101)} G Ω ( x , y ) = ∫ 0 ∞ p t Ω ( x , y ) d t ≤ ∣ x − y ∣ 4 C J ( 4 ∣ x − y ∣ μ ) . (101)
Differentiating under the integral sign and substituting v = c / u v=c/u v = c / u gives
J ′ ( c ) = − 2 J ( c ) . (102) J'(c)=-2J(c). \tag*{(102)} J ′ ( c ) = − 2 J ( c ) . (102)
Simultaneously, we have J ( 0 ) = π / 2 J(0)=\sqrt{\pi}/2 J ( 0 ) = π /2 and hence J ( c ) = π 2 e − 2 c J(c)=\frac{\sqrt{\pi}}{2}e^{-2c} J ( c ) = 2 π e − 2 c . We therefore have
G Ω ( x , y ) ≤ C ∣ x − y ∣ exp ( − ∣ x − y ∣ μ / 2 ) . (103) \mathcal{G}^{\Omega}(x,y)\leq\frac{C}{\lvert x-y\rvert}\exp\left(-\lvert x-y\rvert\sqrt{\mu/2}\right). \tag*{(103)} G Ω ( x , y ) ≤ ∣ x − y ∣ C exp ( − ∣ x − y ∣ μ /2 ) . (103)
Integrating over A i δ A_i^\delta A i δ with δ = min { d min / 2 , 1 } \delta=\min\{d_{\min}/2,1\} δ = min { d m i n /2 , 1 } , where ∣ z − y ∣ ≥ d i j − 2 r max − 2 \lvert z-y\rvert\geq d_{ij}-2r_{\max}-2 ∣ z − y ∣ ≥ d ij − 2 r m a x − 2 for z ∈ ∂ B r j + δ ( x j ) z\in\partial B_{r_j+\delta}(x_j) z ∈ ∂ B r j + δ ( x j ) and y ∈ A i δ y\in A_i^\delta y ∈ A i δ , together with Theorem 4.3 and (28) yields the claim.
Lower bounds
Throughout this section, we may assume that d min ≤ 1 d_{\min}\leq1 d m i n ≤ 1 , replacing the lower separation bound in (2.8) by min { d min , 1 } \min\{d_{\min},1\} min { d m i n , 1 } if necessary. This does not restrict the admissible configurations, and the constants obtained below still depend only on the geometric parameters in the statements of the main theorems. For chains and screens, we also retain the normalisation ρ ≤ 1 / 2 \rho\leq1/2 ρ ≤ 1/2 provided by Theorem 2.8.
Finally, we will derive the lower bounds of Theorems 2.5 and 2.6 and Theorem 2.7 . Once again, the arguments in the chain and screen setting follow the same strategy:
We show that there exists a small ball Q Q Q in Z 1 c Z^c_1 Z 1 c , respectively S 1 c S^c_1 S 1 c , on which V i V_i V i is uniformly lower bounded.
Using this lower bound on Q Q Q , we derive a lower bound for all x x x in some subset U ⊂ Ω U\subset\Omega U ⊂ Ω in terms of P x ( τ Q < τ U ) \mathbb{P}_x(\tau_Q<\tau_U) P x ( τ Q < τ U ) .
Using Theorem 6.2, we lower bound P x ( τ Q < τ U ) \mathbb{P}_x(\tau_Q<\tau_U) P x ( τ Q < τ U ) in terms of G U 1 Q ( x ) \mathcal{G}^U1_Q(x) G U 1 Q ( x ) .
The respective geometries then yield a lower bound on G U \mathcal{G}^U G U .
Finally, we turn the lower bound on V i V_i V i into a lower bound for ∣ C i j ∣ \lvert C_{ij}\rvert ∣ C ij ∣ .
We will develop this machinery in detail in Subsection 6.1, afterwards, in Subsection 6.2 we will be a bit laissez-faire with the arguments as they are so similar. For crystals, a shorter argument can be made using the Cameron-Martin Theorem.
Chains
As a first step, we show that to each resonator D i D_i D i there exists some ball Q i ⊂ Ω Q_i\subset\Omega Q i ⊂ Ω , on which V i V_i V i is uniformly lower bounded.
Lemma 6.1. Let n ∈ R 2 × { 0 } n\in\mathbb{R}^2\times\{0\} n ∈ R 2 × { 0 } have unit norm. Set z i = x i + 3 n z_i=x_i+3n z i = x i + 3 n and Q i = B 1 / 2 ( z i ) Q_i=B_{1/2}(z_i) Q i = B 1/2 ( z i ) . Then we have Q i ⊂ Z 1 c ⊂ Ω Q_i\subset Z^c_1\subset\Omega Q i ⊂ Z 1 c ⊂ Ω . Moreover, there exists a constant c 0 > 0 c_0>0 c 0 > 0 , depending only on r min r_{\min} r m i n and d min d_{\min} d m i n in the normalised configuration, and independent of i i i , n n n , and the particular configuration such that
V i ∣ Q i ≥ c 0 . (104) V_i\big|_{Q_i}\geq c_0. \tag*{(104)} V i Q i ≥ c 0 . (104)
Proof. Fix q ∈ Q i q\in Q_i q ∈ Q i . Since ∣ a i ∣ + r i ≤ 1 / 2 \lvert a_i\rvert+r_i\leq1/2 ∣ a i ∣ + r i ≤ 1/2 ,
∣ q ⊥ ∣ ≥ 3 − ∣ a i ∣ − 1 / 2 ≥ 2 + r i > 2. \lvert q_\perp\rvert\geq3-\lvert a_i\rvert-1/2\geq2+r_i>2. ∣ q ⊥ ∣ ≥ 3 − ∣ a i ∣ − 1/2 ≥ 2 + r i > 2.
Thus Q i ⊂ Z 1 c ⊂ Ω Q_i\subset Z^c_1\subset\Omega Q i ⊂ Z 1 c ⊂ Ω , and every point of Q i Q_i Q i has distance greater than 3 / 2 3/2 3/2 from D \mathcal{D} D .
For each k ≠ i k\ne i k = i , consider the enlarged ball B r k + d min / 3 ( x k ) B_{r_k+d_{\min}/3}(x_k) B r k + d m i n /3 ( x k ) . Their closures are pairwise disjoint. Moreover, both q q q and x i x_i x i lie outside all these balls. The segment joining q q q to x i x_i x i meets only finitely many of them, by local finiteness. Replace each portion of the segment inside an enlarged ball by a shortest great-circle arc on its boundary.
The resulting rectifiable path joins q q q to x i x_i x i and remains at distance at least d min / 3 d_{\min}/3 d m i n /3 from every D k D_k D k with k ≠ i k\ne i k = i . A shortest spherical arc has length at most π / 2 \pi/2 π /2 times the length of its chord. Since the replaced portions of the segment are disjoint, the length ℓ \ell ℓ of the resulting path satisfies
ℓ ≤ π 2 ∣ q − x i ∣ ≤ 7 π 4 = : L ∗ . \ell\leq\frac{\pi}{2}\lvert q-x_i\rvert\leq\frac{7\pi}{4}=:L_*. ℓ ≤ 2 π ∣ q − x i ∣ ≤ 4 7 π =: L ∗ .
Parametrise this path at constant speed on [ 0 , 1 ] [0,1] [ 0 , 1 ] , and denote the parametrisation by γ ~ \widetilde{\gamma} γ . Then
γ ~ ( 0 ) = q , γ ~ ( 1 ) = x i , ∫ 0 1 ∣ γ ~ ′ ( t ) ∣ 2 d t = ℓ 2 ≤ L ∗ 2 . \widetilde{\gamma}(0)=q,\qquad\widetilde{\gamma}(1)=x_i,\qquad\int_0^1\lvert\widetilde{\gamma}'(t)\rvert^2\,dt=\ell^2\leq L_*^2. γ ( 0 ) = q , γ ( 1 ) = x i , ∫ 0 1 ∣ γ ′ ( t ) ∣ 2 d t = ℓ 2 ≤ L ∗ 2 .
Set
ε ∗ = min { r min / 2 , d min / 6 } , E q = { sup 0 ≤ t ≤ 1 ∣ B ( t ) − γ ~ ( t ) ∣ < ε ∗ } . \varepsilon_*=\min\{r_{\min}/2,d_{\min}/6\},\qquad E_q=\left\{\sup_{0\le t\le1}|B(t)-\widetilde{\gamma}(t)|<\varepsilon_*\right\}. ε ∗ = min { r m i n /2 , d m i n /6 } , E q = { 0 ≤ t ≤ 1 sup ∣ B ( t ) − γ ( t ) ∣ < ε ∗ } .
On E q E_q E q , the Brownian path avoids every resonator other than D i D_i D i until time 1 1 1 , and B ( 1 ) ∈ D i B(1)\in D_i B ( 1 ) ∈ D i . By continuity,
E q ⊂ { τ < ∞ , B ( τ ) ∈ ∂ D i } . E_q\subset\{\tau<\infty,\ B(\tau)\in\partial D_i\}. E q ⊂ { τ < ∞ , B ( τ ) ∈ ∂ D i } .
The Cameron–Martin theorem, together with the symmetry under B ↦ − B B\mapsto-B B ↦ − B of the event { sup 0 ≤ t ≤ 1 ∣ B ( t ) ∣ < ε ∗ } \{\sup_{0\le t\le1}|B(t)|<\varepsilon_*\} { sup 0 ≤ t ≤ 1 ∣ B ( t ) ∣ < ε ∗ } , therefore gives, as in the proof of Theorem 5.1,
V i ( q ) ≥ P q ( E q ) ≥ exp ( − 1 2 ∫ 0 1 ∣ γ ~ ′ ( t ) ∣ 2 d t ) P 0 ( sup 0 ≤ t ≤ 1 ∣ B ( t ) ∣ < ε ∗ ) ≥ e − L ∗ 2 / 2 P 0 ( sup 0 ≤ t ≤ 1 ∣ B ( t ) ∣ < ε ∗ ) = : c 0 > 0. \begin{aligned}
V_i(q)\ge\mathbb{P}_q(E_q) \\
&\ge\exp\left(-\frac{1}{2}\int_0^1|\widetilde{\gamma}'(t)|^2\,dt\right)\mathbb{P}_0\left(\sup_{0\le t\le1}|B(t)|<\varepsilon_*\right) \\
&\ge e^{-L_*^2/2}\mathbb{P}_0\left(\sup_{0\le t\le1}|B(t)|<\varepsilon_*\right)=:c_0>0.
\end{aligned} V i ( q ) ≥ P q ( E q ) ≥ exp ( − 2 1 ∫ 0 1 ∣ γ ′ ( t ) ∣ 2 d t ) P 0 ( 0 ≤ t ≤ 1 sup ∣ B ( t ) ∣ < ε ∗ ) ≥ e − L ∗ 2 /2 P 0 ( 0 ≤ t ≤ 1 sup ∣ B ( t ) ∣ < ε ∗ ) =: c 0 > 0.
The last expression depends only on r min r_{\min} r m i n and d min d_{\min} d m i n and is independent of q q q , i i i , n n n , and the particular configuration. This proves the claim.
Lemma 6.2
Let Q = B ϱ ( u ) ‾ ⊂ U Q=\overline{B_\varrho(u)}\subset U Q = B ϱ ( u ) ⊂ U and x ∈ U ∖ Q x\in U\setminus Q x ∈ U ∖ Q . For τ Q = inf { t ≥ 0 : B ( t ) ∈ Q } \tau_Q=\inf\{t\ge0:B(t)\in Q\} τ Q = inf { t ≥ 0 : B ( t ) ∈ Q } and τ U = inf { t ≥ 0 : B ( t ) ∉ U } \tau_U=\inf\{t\ge0:B(t)\notin U\} τ U = inf { t ≥ 0 : B ( t ) ∈ / U } ,
P x ( τ Q < τ U ) ≥ ϱ − 2 ∫ Q G U ( x , v ) d v . (105) \mathbb{P}_x(\tau_Q<\tau_U)\ge\varrho^{-2}\int_Q G^U(x,v)\,dv. \tag*{(105)} P x ( τ Q < τ U ) ≥ ϱ − 2 ∫ Q G U ( x , v ) d v . (105)
If also U ⊂ Ω U\subset\Omega U ⊂ Ω and V i ≥ m > 0 V_i\ge m>0 V i ≥ m > 0 on Q Q Q , then V i ( x ) ≥ m P x ( τ Q < τ U ) V_i(x)\ge m\mathbb{P}_x(\tau_Q<\tau_U) V i ( x ) ≥ m P x ( τ Q < τ U ) .
Proof. There is no occupation of Q Q Q before τ Q \tau_Q τ Q . Applying the strong Markov property (3.1) at τ Q \tau_Q τ Q with E = { τ Q < τ U } ∈ F + ( τ Q ) E=\{\tau_Q<\tau_U\}\in\mathcal{F}^+(\tau_Q) E = { τ Q < τ U } ∈ F + ( τ Q ) together with domain monotonicity of the Green’s function therefore yields
∫ Q G U ( x , y ) d y = G U 1 Q ( x ) = E x [ ∫ 0 τ U 1 Q ( B ( t ) ) d t ] = E x [ 1 τ Q < τ U ∫ τ Q τ U 1 Q ( B ( t ) ) d t ] = E x [ 1 τ Q < τ U E B ( τ Q ) [ ∫ 0 τ U 1 Q ( B ( t ) ) d t ] ] ≤ P x ( τ Q < τ U ) sup w ∈ Q ∫ Q G Ω ( w , y ) d y . (106) \begin{aligned}
\int_QG^U(x,y)\,dy=G^U\mathbf{1}_Q(x) \\
&=\mathbb{E}_x\left[\int_0^{\tau_U}\mathbf{1}_Q(B(t))\,dt\right] \\
&=\mathbb{E}_x\left[\mathbf{1}_{\tau_Q<\tau_U}\int_{\tau_Q}^{\tau_U}\mathbf{1}_Q(B(t))\,dt\right] \\
&=\mathbb{E}_x\left[\mathbf{1}_{\tau_Q<\tau_U}\mathbb{E}_{B(\tau_Q)}\left[\int_0^{\tau_U}\mathbf{1}_Q(B(t))\,dt\right]\right] \\
&\le\mathbb{P}_x(\tau_Q<\tau_U)\sup_{w\in Q}\int_QG^\Omega(w,y)\,dy. \tag*{(106)}
\end{aligned} ∫ Q G U ( x , y ) d y = G U 1 Q ( x ) = E x [ ∫ 0 τ U 1 Q ( B ( t )) d t ] = E x [ 1 τ Q < τ U ∫ τ Q τ U 1 Q ( B ( t )) d t ] = E x [ 1 τ Q < τ U E B ( τ Q ) [ ∫ 0 τ U 1 Q ( B ( t )) d t ] ] ≤ P x ( τ Q < τ U ) w ∈ Q sup ∫ Q G Ω ( w , y ) d y . (106)
For every w w w , the layer-cake formula gives
∫ Q d v ∣ w − v ∣ = ∫ 0 ∞ ∣ Q ∩ B 1 / s ( w ) ∣ d s ≤ ∫ 0 1 / ϱ 4 π ϱ 3 3 d s + ∫ 1 / ϱ ∞ 4 π 3 s 3 d s = 2 π ϱ 2 . \begin{aligned}
\int_Q\frac{dv}{|w-v|}
=\int_0^\infty|Q\cap B_{1/s}(w)|\,ds \\
&\le\int_0^{1/\varrho}\frac{4\pi\varrho^3}{3}\,ds+\int_{1/\varrho}^{\infty}\frac{4\pi}{3s^3}\,ds=2\pi\varrho^2.
\end{aligned} ∫ Q ∣ w − v ∣ d v = ∫ 0 ∞ ∣ Q ∩ B 1/ s ( w ) ∣ d s ≤ ∫ 0 1/ ϱ 3 4 π ϱ 3 d s + ∫ 1/ ϱ ∞ 3 s 3 4 π d s = 2 π ϱ 2 .
This proves (6.2). To obtain the last claim, note that
V i ( x ) = E x [ 1 τ < ∞ 1 ∂ D i ( B ( τ ) ) ] ≥ E x [ 1 τ Q < τ U 1 τ < ∞ 1 ∂ D i ( B ( τ ) ) ] = E x [ 1 τ Q < τ U V i ( B ( τ Q ) ) ] ≥ m P x ( τ Q < τ U ) . (107) \begin{aligned}
V_i(x)=\mathbb{E}_x[\mathbf{1}_{\tau<\infty}\mathbf{1}_{\partial D_i}(B(\tau))] \\
&\ge\mathbb{E}_x[\mathbf{1}_{\tau_Q<\tau_U}\mathbf{1}_{\tau<\infty}\mathbf{1}_{\partial D_i}(B(\tau))] \\
&=\mathbb{E}_x[\mathbf{1}_{\tau_Q<\tau_U}V_i(B(\tau_Q))] \\
&\ge m\mathbb{P}_x(\tau_Q<\tau_U). \tag*{(107)}
\end{aligned} V i ( x ) = E x [ 1 τ < ∞ 1 ∂ D i ( B ( τ ))] ≥ E x [ 1 τ Q < τ U 1 τ < ∞ 1 ∂ D i ( B ( τ ))] = E x [ 1 τ Q < τ U V i ( B ( τ Q ))] ≥ m P x ( τ Q < τ U ) . (107)
where we used that { τ Q < τ U } ⊂ { τ Q < τ } \{\tau_Q < \tau_U\} \subset\{\tau_Q < \tau\} { τ Q < τ U } ⊂ { τ Q < τ } when U ⊂ Ω U \subset\Omega U ⊂ Ω , so that τ = τ Q + τ ∘ θ τ Q \tau= \tau_Q + \tau\circ\theta_{\tau_Q} τ = τ Q + τ ∘ θ τ Q on this event and (3.1) together with (3.7) give the middle equality, and that V i ≥ m V_i \ge m V i ≥ m on Q ∋ B ( τ Q ) Q \ni B(\tau_Q) Q ∋ B ( τ Q ) . ■
Setting U = Z 1 c U = Z_1^c U = Z 1 c , an application of Theorem 6.2 immediately yields
V i ( x ) ≥ c 0 P x ( τ Q i < τ Z c ) ≥ 4 c 0 G 1 Z 1 c 1 Q i ( x ) . (108) V_i(x) \ge c_0 P_x(\tau_{Q_i} < \tau_{Z^c}) \ge4c_0 G_1^{Z_1^c} 1_{Q_i}(x). \tag*{(108)} V i ( x ) ≥ c 0 P x ( τ Q i < τ Z c ) ≥ 4 c 0 G 1 Z 1 c 1 Q i ( x ) . (108)
The next step is thus to derive an expression for G 1 Z 1 c G_1^{Z_1^c} G 1 Z 1 c . To this end, we require the following result on the Green’s function of planar exterior domains [16 Theorem 1.2] .
Theorem 6.3. Let Ω = R 2 ∖ B r ( 0 ) ‾ \Omega= \mathbb{R}^2 \setminus\overline{B_r(0)} Ω = R 2 ∖ B r ( 0 ) for some r > 0 r > 0 r > 0 . For any x , y ∈ Ω x,y \in\Omega x , y ∈ Ω it holds that
lim t → ∞ t ( log t ) 2 p t Ω ( x , y ) = 2 log ( ∣ x ∣ / r ) log ( ∣ y ∣ / r ) π . (109) \lim_{t \to\infty} t(\log t)^2 p_t^\Omega(x,y) = \frac{2 \log(|x|/r)\log(|y|/r)}{\pi}. \tag*{(109)} t → ∞ lim t ( log t ) 2 p t Ω ( x , y ) = π 2 log ( ∣ x ∣/ r ) log ( ∣ y ∣/ r ) . (109) The convergence is uniform on any compact subset of Ω × Ω \Omega\times\Omega Ω × Ω .
The case r = 1 r = 1 r = 1 is [16 Theorem 1.2] ; the general case follows from the scaling identity p t Ω ( x , y ) = r − 2 p t / r 2 Ω ( x / r , y / r ) p_t^\Omega(x,y) = r^{-2}p_{t/r^2}^{\Omega}(x/r,y/r) p t Ω ( x , y ) = r − 2 p t / r 2 Ω ( x / r , y / r ) together with ( log t ) 2 / ( log ( t / r 2 ) ) 2 → 1 (\log t)^2/(\log(t/r^2))^2 \to1 ( log t ) 2 / ( log ( t / r 2 ) ) 2 → 1 .
Upper and lower bounds for Dirichlet heat kernels in the exterior of compact sets were established by Grigor’yan and Saloff-Coste [21 ] . Uchiyama [26 ] obtained uniform large-time asymptotics for the transition density of Brownian motion killed on a bounded set, allowing the spatial variables to vary with time. For our purposes, the convergence on compact subsets stated in Theorem 6.3 is sufficient.
With this in hand, we now can prove the following Theorem
Theorem 6.4. Fix 2 < ζ − < ζ + < ∞ 2 < \zeta_- < \zeta_+ < \infty 2 < ζ − < ζ + < ∞ and ε > 0 \varepsilon> 0 ε > 0 . Then there exists some R ε = R ε ( ζ − , ζ + , ε ) R_\varepsilon= R_\varepsilon(\zeta_-, \zeta_+, \varepsilon) R ε = R ε ( ζ − , ζ + , ε ) such that for any x , y ∈ Z 1 c x,y \in Z_1^c x , y ∈ Z 1 c with ζ − < ∣ x ⊥ ∣ , ∣ y ⊥ ∣ < ζ + \zeta_- < |x_\perp|, |y_\perp| < \zeta_+ ζ − < ∣ x ⊥ ∣ , ∣ y ⊥ ∣ < ζ + and R = ∣ Π ∥ ( x − y ) ∣ > R ε R = |\Pi_\parallel(x-y)| > R_\varepsilon R = ∣ Π ∥ ( x − y ) ∣ > R ε we have
G 1 Z 1 c ( x , y ) ≥ ( 1 − ε ) log ( ∣ x ⊥ ∣ / 2 ) log ( ∣ y ⊥ ∣ / 2 ) 2 π R ( log R ) 2 . (110) G_1^{Z_1^c}(x,y) \ge(1-\varepsilon)\frac{\log(|x_\perp|/2)\log(|y_\perp|/2)}{2\pi R(\log R)^2}. \tag*{(110)} G 1 Z 1 c ( x , y ) ≥ ( 1 − ε ) 2 π R ( log R ) 2 log ( ∣ x ⊥ ∣/2 ) log ( ∣ y ⊥ ∣/2 ) . (110) Proof. Let q t ( x , y ) q_t(x,y) q t ( x , y ) denote the killed density of B 2 ( 0 ) ‾ c \overline{B_2(0)}^c B 2 ( 0 ) c , so that Theorem 6.3 applies with r = 2 r = 2 r = 2 . Independence of axial and radial Brownian motion then implies that
p t Z 1 c ( x , y ) = q t ( x ⊥ , y ⊥ ) p t ( 1 ) ( Π ∥ ( x − y ) ) , (111) p_t^{Z_1^c}(x,y) = q_t(x_\perp,y_\perp)p_t^{(1)}(\Pi_\parallel(x-y)), \tag*{(111)} p t Z 1 c ( x , y ) = q t ( x ⊥ , y ⊥ ) p t ( 1 ) ( Π ∥ ( x − y )) , (111) where p t ( 1 ) ( u ) = ( 2 π t ) − 1 / 2 e − u 2 / ( 2 t ) p_t^{(1)}(u) = (2\pi t)^{-1/2}e^{-u^2/(2t)} p t ( 1 ) ( u ) = ( 2 π t ) − 1/2 e − u 2 / ( 2 t ) denotes the one-dimensional Gaussian kernel. Suppose the claim fails. Then there is a sequence x n , y n ∈ Z 1 c x_n,y_n \in Z_1^c x n , y n ∈ Z 1 c with R n = ∣ Π ∥ ( x n − y n ) ∣ → ∞ R_n = |\Pi_\parallel(x_n-y_n)| \to\infty R n = ∣ Π ∥ ( x n − y n ) ∣ → ∞ and ζ − < ∣ ( x n ) ⊥ ∣ , ∣ ( y n ) ⊥ ∣ < ζ + \zeta_- < |(x_n)_\perp|, |(y_n)_\perp| < \zeta_+ ζ − < ∣ ( x n ) ⊥ ∣ , ∣ ( y n ) ⊥ ∣ < ζ + for which the asserted inequality fails. Writing ξ n = ( x n ) ⊥ \xi_n = (x_n)_\perp ξ n = ( x n ) ⊥ , ξ n ′ = ( y n ) ⊥ \xi'_n = (y_n)_\perp ξ n ′ = ( y n ) ⊥ and ℓ n = log ( ∣ ξ n ∣ / 2 ) log ( ∣ ξ n ′ ∣ / 2 ) \ell_n = \log(|\xi_n|/2)\log(|\xi'_n|/2) ℓ n = log ( ∣ ξ n ∣/2 ) log ( ∣ ξ n ′ ∣/2 ) , Theorem 6.3 then yields
lim n → ∞ q R n 2 s ( ξ n , ξ n ′ ) 2 π R n 2 ( log R n ) 2 ℓ n = s − 1 (112) \lim_{n\to\infty} q_{R_n^2s}(\xi_n,\xi'_n)\frac{2\pi R_n^2(\log R_n)^2}{\ell_n} = s^{-1} \tag*{(112)} n → ∞ lim q R n 2 s ( ξ n , ξ n ′ ) ℓ n 2 π R n 2 ( log R n ) 2 = s − 1 (112) for every $s > 0.
The scaling property of Brownian motion and Fatou’s Lemma then yield
lim inf n → ∞ 2 π R n ( log R n ) 2 ℓ n G 1 Z 1 c ( x n , y n ) = lim inf n → ∞ ∫ 0 ∞ 2 π R n ( log R n ) 2 ℓ n q t ( ξ n , ξ n ′ ) e − R n 2 / ( 2 t ) 2 π t d t ≥ ∫ 0 ∞ lim inf n → ∞ 2 π R n 2 ( log R n ) 2 ℓ n q R n 2 s ( ξ n , ξ n ′ ) e − 1 / ( 2 s ) 2 π s d s ≥ ∫ 0 ∞ s − 1 e − 1 / ( 2 s ) 2 π s d s = 1. (113) \begin{aligned}
\liminf_{n\to\infty}\frac{2\pi R_n(\log R_n)^2}{\ell_n}G_1^{Z_1^c}(x_n,y_n)
&= \liminf_{n\to\infty}\int_0^\infty\frac{2\pi R_n(\log R_n)^2}{\ell_n}q_t(\xi_n,\xi'_n)\frac{e^{-R_n^2/(2t)}}{\sqrt{2\pi t}}\,dt \\
&\ge\int_0^\infty\liminf_{n\to\infty}\frac{2\pi R_n^2(\log R_n)^2}{\ell_n}q_{R_n^2s}(\xi_n,\xi'_n)\frac{e^{-1/(2s)}}{\sqrt{2\pi s}}\,ds \\
&\ge\int_0^\infty s^{-1}\frac{e^{-1/(2s)}}{\sqrt{2\pi s}}\,ds = 1. \tag*{(113)}
\end{aligned} n → ∞ lim inf ℓ n 2 π R n ( log R n ) 2 G 1 Z 1 c ( x n , y n ) = n → ∞ lim inf ∫ 0 ∞ ℓ n 2 π R n ( log R n ) 2 q t ( ξ n , ξ n ′ ) 2 π t e − R n 2 / ( 2 t ) d t ≥ ∫ 0 ∞ n → ∞ lim inf ℓ n 2 π R n 2 ( log R n ) 2 q R n 2 s ( ξ n , ξ n ′ ) 2 π s e − 1/ ( 2 s ) d s ≥ ∫ 0 ∞ s − 1 2 π s e − 1/ ( 2 s ) d s = 1. (113) This contradicts the choice of the sequence, which proves the claim. ■
Since 5 / 2 ≤ ∣ ( z j ) ⊥ ∣ ≤ 7 / 2 5/2 \leq|(z_j)_{\perp}| \leq7/2 5/2 ≤ ∣ ( z j ) ⊥ ∣ ≤ 7/2 and 9 / 4 < ∣ y ⊥ ∣ < 15 / 4 9/4 < |y_{\perp}| < 15/4 9/4 < ∣ y ⊥ ∣ < 15/4 for y ∈ B 1 / 4 ( z i ) ⊂ Q i y \in B_{1/4}(z_i) \subset Q_i y ∈ B 1/4 ( z i ) ⊂ Q i , while d i j − 5 / 4 ≤ R ≤ d i j + 2 d_{ij} - 5/4 \leq R \leq d_{ij} + 2 d ij − 5/4 ≤ R ≤ d ij + 2 for such y y y , for d i j ≥ R ε + 2 d_{ij} \geq R_{\varepsilon} + 2 d ij ≥ R ε + 2 we thus obtain from Theorem 6.2 and the previous theorem (with ζ − = 9 / 4 \zeta_{-} = 9/4 ζ − = 9/4 , ζ + = 4 \zeta_{+} = 4 ζ + = 4 )
V i ( z j ) ≥ c 0 c 1 ( d i j + 2 ) ( log ( d i j + 2 ) ) 2 , (114) V_i(z_j) \geq\frac{c_0c_1}{(d_{ij}+2)(\log(d_{ij}+2))^2}, \tag*{(114)} V i ( z j ) ≥ ( d ij + 2 ) ( log ( d ij + 2 ) ) 2 c 0 c 1 , (114)
where c 1 c_1 c 1 is a constant independent of the geometric parameters.
Proof of the lower bound in Theorem 2.5 . Fix ε = 1 / 2 \varepsilon= 1/2 ε = 1/2 in (114) and set R 0 = R ε + 2 R_0 = R_{\varepsilon} + 2 R 0 = R ε + 2 , which is an absolute constant. For d i j < R 0 d_{ij} < R_0 d ij < R 0 the segment from z j z_j z j to x i x_i x i has length at most R 0 + 2 r max + 3 R_0 + 2r_{\max} + 3 R 0 + 2 r m a x + 3 , and the construction in the proof of Theorem 6.1, applied to this segment, yields V i ( z j ) ≥ c 0 ′ V_i(z_j) \geq c'_0 V i ( z j ) ≥ c 0 ′ for some c 0 ′ > 0 c'_0 > 0 c 0 ′ > 0 depending only on r min , r max , d min r_{\min}, r_{\max}, d_{\min} r m i n , r m a x , d m i n . As d i j ≥ d min d_{ij} \geq d_{\min} d ij ≥ d m i n , and d i j + 2 ≤ 2 d i j d_{ij} + 2 \leq2d_{ij} d ij + 2 ≤ 2 d ij for d i j ≥ R 0 d_{ij} \geq R_0 d ij ≥ R 0 , we thus have in all cases
V i ( z j ) ≥ c d i j log 2 ( 2 + d i j ) , (115) V_i(z_j) \geq\frac{c}{d_{ij}\log^2(2+d_{ij})}, \tag*{(115)} V i ( z j ) ≥ d ij log 2 ( 2 + d ij ) c , (115)
with c > 0 c > 0 c > 0 depending only on r min , r max , d min r_{\min}, r_{\max}, d_{\min} r m i n , r m a x , d m i n .
Consequently, B 4 r h ( y ) ⊂ Ω B_{4r_h}(y) \subset\Omega B 4 r h ( y ) ⊂ Ω for every point y y y on the path. A chain of overlapping balls of radius r h r_h r h , followed by repeated application of Harnack’s inequality, therefore gives
V i ( z ′ ) ≥ 27 − N V i ( z j ) , (116) V_i(z') \geq27^{-N}V_i(z_j), \tag*{(116)} V i ( z ′ ) ≥ 2 7 − N V i ( z j ) , (116)
where N N N depends only on the prescribed geometric bounds.
Every point of ∂ B r j + d min / 3 ( x j ) \partial B_{r_j+d_{\min}/3}(x_j) ∂ B r j + d m i n /3 ( x j ) has distance at least d min / 3 d_{\min}/3 d m i n /3 from D D D . Joining z ′ z' z ′ to any other point of this sphere by a shortest great-circle arc and applying the same argument gives
min z ∈ ∂ B r j + d min / 3 ( x j ) V i ( z ) ≥ 27 − ( N + N ′ ) V i ( z j ) , (117) \min_{z\in\partial B_{r_j+d_{\min}/3}(x_j)} V_i(z) \geq27^{-(N+N')}V_i(z_j), \tag*{(117)} z ∈ ∂ B r j + d m i n /3 ( x j ) min V i ( z ) ≥ 2 7 − ( N + N ′ ) V i ( z j ) , (117)
where N ′ N' N ′ depends only on r max r_{\max} r m a x and d min d_{\min} d m i n .
Finally, apply (27) with δ = d min / 2 \delta= d_{\min}/2 δ = d m i n /2 , u = V i u = V_i u = V i , and ϱ = r j + d min / 3 \varrho= r_j + d_{\min}/3 ϱ = r j + d m i n /3 . Writing μ ( ϱ ) \mu(\varrho) μ ( ϱ ) for the spherical mean of V i V_i V i , we obtain
∣ C i j ∣ = 4 π r j ϱ ϱ − r j μ ( ϱ ) ≥ 12 π r min 2 d min 27 − ( N + N ′ ) V i ( z j ) . (118) |C_{ij}| = \frac{4\pi r_j\varrho}{\varrho-r_j}\mu(\varrho) \geq\frac{12\pi r_{\min}^2}{d_{\min}}27^{-(N+N')}V_i(z_j). \tag*{(118)} ∣ C ij ∣ = ϱ − r j 4 π r j ϱ μ ( ϱ ) ≥ d m i n 12 π r m i n 2 2 7 − ( N + N ′ ) V i ( z j ) . (118)
Together with (115) , this proves the lower bound.
Screens
The proof of the lower bound for screen geometries follows the arguments from the previous subsection. The proof of Theorem 6.1 applies to screens with n = ( 0 , 0 , 1 ) n=(0,0,1) n = ( 0 , 0 , 1 ) : the balls Q i Q_i Q i then lie above the plane { z ∥ = 2 } \{z_{\parallel}=2\} { z ∥ = 2 } and have distance greater than 3 / 2 3/2 3/2 from D D D . The path construction and the Cameron–Martin estimate are unchanged. Lemma 6.2 also applies. It thus remains to choose an appropriate subdomain U ⊂ Ω U \subset\Omega U ⊂ Ω and obtain lower bounds on its Green’s function.
To this end, set H = S 1 c ∩ R 2 × R + = { z ∈ R 3 : z ∥ > 2 } \mathcal{H}=S^c_1\cap\mathbb{R}^2\times\mathbb{R}_+=\{z\in\mathbb{R}^3:z_{\parallel}>2\} H = S 1 c ∩ R 2 × R + = { z ∈ R 3 : z ∥ > 2 } and U = H U=\mathcal{H} U = H ; in Theorem 6.1 we take n = ( 0 , 0 , 1 ) n=(0,0,1) n = ( 0 , 0 , 1 ) , so that Q i = B 1 / 2 ( x i + 3 n ) ⊂ H Q_i=B_{1/2}(x_i+3n)\subset\mathcal{H} Q i = B 1/2 ( x i + 3 n ) ⊂ H by ∣ s i ∣ < 1 / 2 |s_i|<1/2 ∣ s i ∣ < 1/2 . It is well known, see e.g. [19 ] that the Green’s function for a halfspace is given by
G H ( x , y ) = G 0 ( x , y ) − G 0 ( x ∗ , y ) , (119) G^{\mathcal{H}}(x,y)=G^0(x,y)-G^0(x^*,y), \tag*{(119)} G H ( x , y ) = G 0 ( x , y ) − G 0 ( x ∗ , y ) , (119)
where x ∗ x^* x ∗ is the reflection of x x x across the plane ∂ H = { z ∥ = 2 } \partial\mathcal{H}=\{z_{\parallel}=2\} ∂ H = { z ∥ = 2 } . Writing h x = x ∥ − 2 h_x=x_{\parallel}-2 h x = x ∥ − 2 and h y = y ∥ − 2 h_y=y_{\parallel}-2 h y = y ∥ − 2 for the distances of x , y ∈ H x,y\in\mathcal{H} x , y ∈ H to ∂ H \partial\mathcal{H} ∂ H , we have ∣ x ∗ − y ∣ 2 − ∣ x − y ∣ 2 = 4 h x h y |x^*-y|^2-|x-y|^2=4h_xh_y ∣ x ∗ − y ∣ 2 − ∣ x − y ∣ 2 = 4 h x h y and hence
G H ( x , y ) = 2 h x h y π ∣ x − y ∣ ∣ x ∗ − y ∣ ( ∣ x − y ∣ + ∣ x ∗ − y ∣ ) . (120) G^{\mathcal{H}}(x,y)=\frac{2h_xh_y}{\pi|x-y||x^*-y|(|x-y|+|x^*-y|)}. \tag*{(120)} G H ( x , y ) = π ∣ x − y ∣∣ x ∗ − y ∣ ( ∣ x − y ∣ + ∣ x ∗ − y ∣ ) 2 h x h y . (120)
The following Lemma follows from a simple calculation.
Lemma 6.5. Fix 2 < ζ − < ζ + < ∞ 2 < \zeta_- < \zeta_+ < \infty 2 < ζ − < ζ + < ∞ . Then for any x , y ∈ H x,y \in\mathcal{H} x , y ∈ H with ζ − < ∥ x ∥ , ∥ y ∥ < ζ + \zeta_- < \lVert x\rVert,\lVert y\rVert< \zeta_+ ζ − < ∥ x ∥ , ∥ y ∥ < ζ + and ∣ x − y ∣ ≥ 1 |x-y| \ge1 ∣ x − y ∣ ≥ 1 it holds that
G H ( x , y ) ≥ C ∣ x − y ∣ 3 , (121) \mathcal{G}^{\mathcal{H}}(x,y) \ge\frac{C}{|x-y|^3}, \tag*{(121)} G H ( x , y ) ≥ ∣ x − y ∣ 3 C , (121)
where C > 0 C>0 C > 0 depends only on ζ − , ζ + \zeta_-,\zeta_+ ζ − , ζ + .
Proof. We have ζ − − 2 < h x , h y < ζ + − 2 \zeta_- - 2 < h_x,h_y < \zeta_+ - 2 ζ − − 2 < h x , h y < ζ + − 2 , hence ∣ x ∗ − y ∣ ≤ ∣ x − y ∣ + 2 h x ≤ ( 2 ζ + − 3 ) ∣ x − y ∣ |x^*-y| \le|x-y|+2h_x \le(2\zeta_+-3)|x-y| ∣ x ∗ − y ∣ ≤ ∣ x − y ∣ + 2 h x ≤ ( 2 ζ + − 3 ) ∣ x − y ∣ for ∣ x − y ∣ ≥ 1 |x-y| \ge1 ∣ x − y ∣ ≥ 1 . Inserting this into (120) gives the claim with C = 2 ( ζ − − 2 ) 2 / ( π ( 2 ζ + − 3 ) ( 2 ζ + − 2 ) ) C=2(\zeta_- -2)^2/(\pi(2\zeta_+-3)(2\zeta_+-2)) C = 2 ( ζ − − 2 ) 2 / ( π ( 2 ζ + − 3 ) ( 2 ζ + − 2 )) .
The lower bound of Theorem 2.6 now follows mutatis mutandis from the proof of the lower bound of Theorem 2.5 : for y ∈ B 1 / 4 ( z i ) ⊂ Q i y \in B_{1/4}(z_i) \subset Q_i y ∈ B 1/4 ( z i ) ⊂ Q i we have 9 / 4 < ∥ y ∥ < 15 / 4 9/4 < \lVert y\rVert< 15/4 9/4 < ∥ y ∥ < 15/4 and 5 / 2 ≤ ( z j ) ∥ ≤ 7 / 2 5/2 \le(z_j)\| \le7/2 5/2 ≤ ( z j ) ∥ ≤ 7/2 , while d i j − 1 / 4 ≤ ∣ z j − y ∣ ≤ d i j + 5 / 4 d_{ij}-1/4 \le|z_j-y| \le d_{ij}+5/4 d ij − 1/4 ≤ ∣ z j − y ∣ ≤ d ij + 5/4 , so that for d i j ≥ 5 / 4 d_{ij} \ge5/4 d ij ≥ 5/4 Theorem 6.2 and Theorem 6.5 (with ζ − = 9 / 4 \zeta_-=9/4 ζ − = 9/4 , ζ + = 4 \zeta_+=4 ζ + = 4 ) yield V i ( z j ) ≥ c d i j − 3 V_i(z_j) \ge c d_{ij}^{-3} V i ( z j ) ≥ c d ij − 3 . For d i j < 5 / 4 d_{ij}<5/4 d ij < 5/4 the bound V i ( z j ) ≥ c 0 ′ V_i(z_j) \ge c'_0 V i ( z j ) ≥ c 0 ′ holds as in Section 6.1, and the Harnack argument given there together with (118) transfers these bounds to ∣ C i j ∣ |C_{ij}| ∣ C ij ∣ .
Crystals
It remains to prove the lower bound for crystals. In this regime, a simpler Cameron-Martin based argument is sufficient.
Once again, we proceed as in Theorem 3.6, constructing a path γ ~ \tilde{\gamma} γ ~ from a point z ∗ ∈ ∂ B r j + d min / 3 ( x j ) z^* \in\partial B_{r_j+d_{\min}/3}(x_j) z ∗ ∈ ∂ B r j + d m i n /3 ( x j ) to a point of ∂ B r i + d min / 3 ( x i ) \partial B_{r_i+d_{\min}/3}(x_i) ∂ B r i + d m i n /3 ( x i ) , i ≠ j i \ne j i = j , such that dist ( γ ~ , D ) ≥ d min / 3 \operatorname{dist}(\tilde{\gamma},D) \ge d_{\min}/3 dist ( γ ~ , D ) ≥ d m i n /3 , whose length is upper bounded by d i j d_{ij} d ij times a constant C C C depending on the geometric parameters, but not i , j i,j i , j , and append the radial segment to x i x_i x i . We parametrise γ ~ \tilde{\gamma} γ ~ by arc length, so that γ ~ : [ 0 , ℓ ] → R 3 \tilde{\gamma}:[0,\ell]\to\mathbb{R}^3 γ ~ : [ 0 , ℓ ] → R 3 has unit speed and ℓ ≤ C d i j + r max + d min / 3 \ell\le Cd_{ij}+r_{\max}+d_{\min}/3 ℓ ≤ C d ij + r m a x + d m i n /3 , and set ε = min { d min / 6 , r min } \varepsilon=\min\{d_{\min}/6,r_{\min}\} ε = min { d m i n /6 , r m i n } . Since under P z ∗ \mathbb{P}_{z^*} P z ∗ { τ < ∞ , B ( τ ) ∈ ∂ D i } ⊃ { max 0 ≤ t ≤ ℓ ∣ B ( t ) − γ ~ ( t ) ∣ < ε } \{\tau<\infty,B(\tau)\in\partial D_i\}\supset\{\max_{0\le t\le\ell}|B(t)-\tilde{\gamma}(t)|<\varepsilon\} { τ < ∞ , B ( τ ) ∈ ∂ D i } ⊃ { max 0 ≤ t ≤ ℓ ∣ B ( t ) − γ ~ ( t ) ∣ < ε } (on this event B B B keeps a distance greater than d min / 6 d_{\min}/6 d m i n /6 from D ∖ D i D\setminus D_i D ∖ D i and satisfies ∣ B ( ℓ ) − x i ∣ < ε ≤ r i |B(\ell)-x_i|<\varepsilon\le r_i ∣ B ( ℓ ) − x i ∣ < ε ≤ r i ), Cameron Martin as in the proof of Theorem 5.1, now with ∫ 0 ℓ ∣ γ ~ ′ ∣ 2 d t = ℓ \int_0^\ell|\tilde{\gamma}'|^2\,dt=\ell ∫ 0 ℓ ∣ γ ~ ′ ∣ 2 d t = ℓ , yields
V i ( z ∗ ) ≥ exp ( − ℓ / 2 ) P 0 ( max 0 ≤ t ≤ ℓ ∣ B ( t ) ∣ < ε ) . (122) V_i(z^*) \ge\exp(-\ell/2)\mathbb{P}_0\left(\max_{0\le t\le\ell}|B(t)|<\varepsilon\right). \tag*{(122)} V i ( z ∗ ) ≥ exp ( − ℓ /2 ) P 0 ( 0 ≤ t ≤ ℓ max ∣ B ( t ) ∣ < ε ) . (122)
The last probability decays at most exponentially in ℓ \ell ℓ . Indeed, for ∣ w ∣ ≤ ε / 2 |w|\le\varepsilon/2 ∣ w ∣ ≤ ε /2 the event { max 0 ≤ t ≤ 1 ∣ B ( t ) − ( 1 − t ) w ∣ < ε / 2 } \{\max_{0\le t\le1}|B(t)-(1-t)w|<\varepsilon/2\} { max 0 ≤ t ≤ 1 ∣ B ( t ) − ( 1 − t ) w ∣ < ε /2 } is contained in { max 0 ≤ t ≤ 1 ∣ B ( t ) ∣ < ε , ∣ B ( 1 ) ∣ < ε / 2 } \{\max_{0\le t\le1}|B(t)|<\varepsilon,\ |B(1)|<\varepsilon/2\} { max 0 ≤ t ≤ 1 ∣ B ( t ) ∣ < ε , ∣ B ( 1 ) ∣ < ε /2 } under P w \mathbb{P}_w P w , so that Cameron Martin gives
p 0 = inf ∣ w ∣ ≤ ε / 2 P w ( max 0 ≤ t ≤ 1 ∣ B ( t ) ∣ < ε , ∣ B ( 1 ) ∣ < ε / 2 ) ≥ e − ε 2 / 8 P 0 ( max 0 ≤ t ≤ 1 ∣ B ( t ) ∣ < ε / 2 ) > 0 , (123) p_0=\inf_{|w|\le\varepsilon/2}\mathbb{P}_w\left(\max_{0\le t\le1}|B(t)|<\varepsilon,\ |B(1)|<\varepsilon/2\right)
\ge e^{-\varepsilon^2/8}\mathbb{P}_0\left(\max_{0\le t\le1}|B(t)|<\varepsilon/2\right)>0, \tag*{(123)} p 0 = ∣ w ∣ ≤ ε /2 inf P w ( 0 ≤ t ≤ 1 max ∣ B ( t ) ∣ < ε , ∣ B ( 1 ) ∣ < ε /2 ) ≥ e − ε 2 /8 P 0 ( 0 ≤ t ≤ 1 max ∣ B ( t ) ∣ < ε /2 ) > 0 , (123)
and the Markov property at the times 1 , … , ⌈ ℓ ⌉ − 1 1,\ldots,\lceil\ell\rceil-1 1 , … , ⌈ ℓ ⌉ − 1 yields P 0 ( max 0 ≤ t ≤ ℓ ∣ B ( t ) ∣ < ε ) ≥ p 0 ⌈ ℓ ⌉ \mathbb{P}_0(\max_{0\le t\le\ell}|B(t)|<\varepsilon)\ge p_0^{\lceil\ell\rceil} P 0 ( max 0 ≤ t ≤ ℓ ∣ B ( t ) ∣ < ε ) ≥ p 0 ⌈ ℓ ⌉ . Consequently, V i ( z ∗ ) ≥ p 0 exp ( − ℓ ( 1 / 2 + log ( 1 / p 0 ) ) ) ≥ c exp ( − c ′ d i j ) V_i(z^*)\ge p_0\exp(-\ell(1/2+\log(1/p_0)))\ge c\exp(-c'd_{ij}) V i ( z ∗ ) ≥ p 0 exp ( − ℓ ( 1/2 + log ( 1/ p 0 ))) ≥ c exp ( − c ′ d ij ) with c , c ′ > 0 c,c'>0 c , c ′ > 0 depending only on the geometric parameters. Applying Harnack’s inequality to obtain a uniform lower bound on ∂ B r j + d min / 3 ( x j ) \partial B_{r_j+d_{\min}/3}(x_j) ∂ B r j + d m i n /3 ( x j ) , as in the proof of the lower bound of Theorem 2.5 , together with Theorem 3.9 yields the claim.
Conclusion
In this paper, we have derived upper and lower bounds off-diagonal decay estimates for the capacitance matrix of resonator systems in chain, screen and crystal geometries, using the classical connection between Brownian motion and the Dirichlet problem. Our main results, Theorems 2.5 to 2.7, show that the rate of decay is determined by the dimensionality of the structure.
Many interesting questions remain. For one, it is natural to expect that similar estimates hold for systems of high-index dielectric resonators [9 ] . Another promising direction would be to derive sharper truncation estimates [10 ] , which can be used to design fast computational methods [8 ] . More ambitious, but equally important, would be decay estimates in the non-subwavelength regime, where the capacitance matrix becomes frequency dependent [7 ] .
Statements and declarations
Funding
C. Thalhammer was supported in part by the Swiss National Science Foundation grant number 200021-236472.
Competing interests
The authors have no competing interests to declare.
Declaration of AI use
The authors interacted with and guided ChatGPT 6 and Claude Fable 5.1, Opus 5.5 whilst developing the proofs and then wrote the paper. The authors take full responsibility for the accuracy and integrity of all content in this paper. Habib Ammari
ETH Zürich, Department of Mathematics, Rämistrasse 101, 8092 Zürich, Switzerland. ORCID.org/0000-0001-7278-4877 Email address: habib.ammari@math.ethz.ch
Clemens Thalhammer
ETH Zürich, Department of Mathematics, Rämistrasse 101, 8092 Zürich, Switzerland. ORCID.org/0009-0006-7218-260X Email address: clemens.thalhammer@sam.math.ethz.ch