Introduction and main results
Hyperuniformity means that the variance of the number of points in a large Euclidean ball grows more slowly than its volume. It occurs in periodic point configurations, certain quasicrystals, and some disordered systems [39 , 38 ] . We call a point process number rigid if its configuration outside any bounded region determines the number of points inside, and maximally rigid if it determines their positions as well. Examples arise from determinantal kernels, zeros of Gaussian analytic functions, and spectral gaps [17 , 15 , 16 , 13 , 32 ] .
Ghosh and Lebowitz [15 ] proved number rigidity for hyperuniform processes in dimensions one and two under additional decay assumptions on truncated pair correlations. They also raised the question whether hyperuniformity, possibly with correlation decay, is necessary for number rigidity. In [16 ] , they established maximal rigidity for processes whose structure function vanishes on an open set, and also under superpolynomial vanishing at the origin. Klatt and Last [26 ] constructed hyperfluctuating point processes for which the entire configuration in every bounded region can be recovered from a bounded exterior stopping set. Their strong rigidity implies maximal rigidity as defined here. Conversely, in dimensions at least three, sufficiently large independent Gaussian displacements of lattices give hyperuniform processes that are not number rigid [35 , 4 ] . Thus hyperuniformity does not imply number rigidity in dimensions at least three, while the hyperfluctuating examples of Klatt and Last show that number rigidity does not imply hyperuniformity.
These results concern particular families of point processes. In this paper, we study the Baire-category prevalence of hyperuniformity and rigidity among translation-invariant ergodic simple point processes of fixed intensity and finite local second moments. We prove, for the local L 2 L^{2} L 2 topology defined below, that hyperuniformity is dense but meagre, whereas linear maximal rigidity is residual.
We identify a locally finite point set with its counting measure. Let Conf ( R d ) \operatorname{Conf}(\mathbb{R}^{d}) Conf ( R d ) be the space of locally finite simple counting measures, equipped with the vague topology; see [8 Appendix A2.6] and [36 ] .
A point process is a probability measure η \eta η on Conf ( R d ) \operatorname{Conf}(\mathbb{R}^{d}) Conf ( R d ) . For t ∈ R d t \in\mathbb{R}^{d} t ∈ R d , set
( θ t ω ) ( A ) : = ω ( A + t ) , (\theta_t\omega)(A) := \omega(A+t), ( θ t ω ) ( A ) := ω ( A + t ) ,
and call η \eta η translation-invariant if ( θ t ) ∗ η = η (\theta_t)_*\eta=\eta ( θ t ) ∗ η = η for every t ∈ R d t \in\mathbb{R}^{d} t ∈ R d . For a bounded Borel set A ⊂ R d A \subset\mathbb{R}^{d} A ⊂ R d , let
N A ( ω ) : = ω ( A ) . N_A(\omega) := \omega(A). N A ( ω ) := ω ( A ) .
Such a process has intensity ρ > 0 \rho> 0 ρ > 0 if
∫ N A d η = ρ Vol d ( A ) \int N_A\,\mathrm{d}\eta= \rho\operatorname{Vol}_{d}(A) ∫ N A d η = ρ Vol d ( A )
for every bounded Borel set A ⊂ R d A \subset\mathbb{R}^{d} A ⊂ R d . Write P P ρ , 2 \mathbf{PP}_{\rho,2} PP ρ , 2 for the translation-invariant point processes of intensity ρ \rho ρ whose counts also satisfy
N A ∈ L 2 ( η ) N_A \in L^2(\eta) N A ∈ L 2 ( η )
for every bounded Borel A ⊂ R d A \subset\mathbb{R}^{d} A ⊂ R d . Finite intensity alone does not imply local square-integrability.
We equip P P ρ , 2 \mathbf{PP}_{\rho,2} PP ρ , 2 with the local L 2 L^2 L 2 topology . In this topology, convergence means weak convergence of point-process laws together with convergence of their count second moments in each cube Q m : = [ − m / 2 , m / 2 ) d Q_m := [-m/2,m/2)^d Q m := [ − m /2 , m /2 ) d , m ≥ 1 m \ge1 m ≥ 1 . We prove the precise convergence criterion in Proposition 2.7 and show in Proposition 2.6 that P P ρ , 2 \mathbf{PP}_{\rho,2} PP ρ , 2 is Polish and its subspace E P P ρ , 2 \mathbf{EPP}_{\rho,2} EPP ρ , 2 of ergodic processes is a G δ G_\delta G δ -subset, hence also Polish.
A set is nowhere dense if its closure has empty interior, and meagre if it is a countable union of nowhere-dense sets. A set is residual if its complement is meagre. In a Polish space this is equivalent to containing a dense G δ G_\delta G δ -subset.
Hyperuniformity
For R > 0 R > 0 R > 0 , let B R B_R B R be the closed Euclidean ball of radius R R R centred at the origin. We write
V η ( R ) : = Var η ( N B R ) . V_\eta(R) := \operatorname{Var}_\eta(N_{B_R}). V η ( R ) := Var η ( N B R ) .
We call η ∈ P P ρ , 2 \eta\in\mathbf{PP}_{\rho,2} η ∈ PP ρ , 2 hyperuniform if its ball-count variance grows sublinearly in volume, that is, if
V η ( R ) Vol d ( B R ) ⟶ 0 ( R → ∞ ) . \frac{V_\eta(R)}{\operatorname{Vol}_{d}(B_R)} \longrightarrow0 \qquad(R \to\infty). Vol d ( B R ) V η ( R ) ⟶ 0 ( R → ∞ ) .
Let H y p ρ ⊂ P P ρ , 2 \mathbf{Hyp}_{\rho} \subset\mathbf{PP}_{\rho,2} Hyp ρ ⊂ PP ρ , 2 denote the class of hyperuniform processes.
For d = 1 d=1 d = 1 , let Vol 0 \operatorname{Vol}_0 Vol 0 denote counting measure. Beck’s discrepancy theorem [2 Theorem 2A] and the spectral lower bound of [4 Theorem 5.1] imply that every η ∈ P P ρ , 2 \eta\in\mathbf{PP}_{\rho,2} η ∈ PP ρ , 2 , without an ergodicity assumption, satisfies
lim sup R → ∞ V η ( R ) Vol d − 1 ( ∂ B R ) > 0. (1) \limsup_{R\to\infty}\frac{V_\eta(R)}{\operatorname{Vol}_{d-1}(\partial B_R)} > 0. \tag*{(1)} R → ∞ lim sup Vol d − 1 ( ∂ B R ) V η ( R ) > 0. (1)
Equivalently, there are c = c ( η ) > 0 c=c(\eta)>0 c = c ( η ) > 0 and radii R k → ∞ R_k \to\infty R k → ∞ such that
V η ( R k ) ≥ c Vol d ( B R k ) ( d − 1 ) / d . V_\eta(R_k) \ge c\operatorname{Vol}_{d}(B_{R_k})^{(d-1)/d}. V η ( R k ) ≥ c Vol d ( B R k ) ( d − 1 ) / d .
At the other end of the scale, every ergodic η ∈ P P ρ , 2 \eta\in\mathbf{PP}_{\rho,2} η ∈ PP ρ , 2 satisfies
V η ( R ) = o ( Vol d ( B R ) 2 ) ( R → ∞ ) , V_\eta(R)=o\left(\operatorname{Vol}_{d}(B_R)^2\right) \qquad(R \to\infty), V η ( R ) = o ( Vol d ( B R ) 2 ) ( R → ∞ ) ,
by the mean ergodic theorem; see Proposition 6.2 .
We call η ∈ E P P ρ , 2 \eta\in\mathbf{EPP}_{\rho,2} η ∈ EPP ρ , 2 intermittently hyperuniform if, for every β > ( d − 1 ) / d \beta> (d-1)/d β > ( d − 1 ) / d ,
lim inf R → ∞ V η ( R ) Vol d ( B R ) β = 0 (2) \liminf_{R\to\infty}\frac{V_\eta(R)}{\operatorname{Vol}_{d}(B_R)^\beta}=0 \tag*{(2)} R → ∞ lim inf Vol d ( B R ) β V η ( R ) = 0 (2)
and, for every α < 2 \alpha< 2 α < 2 ,
lim sup R → ∞ V η ( R ) Vol d ( B R ) α = + ∞ . (3) \limsup_{R\to\infty}\frac{V_\eta(R)}{\operatorname{Vol}_{d}(B_R)^\alpha}=+\infty. \tag*{(3)} R → ∞ lim sup Vol d ( B R ) α V η ( R ) = + ∞. (3)
Write I H y p ρ \mathbf{IHyp}_{\rho} IHyp ρ for this class.
Hyperuniform examples include determinantal processes with projection kernels, such as the sine and Ginibre processes [14 ] , the two-dimensional one-component plasma [33 ] , certain cut-and-project quasicrystals [5 ] , and several transport constructions [27 ] . Stable matching with a lattice has also produced hyperuniform and number-rigid processes [28 ] .
Remark 1.1. The fair-partition construction of [27 Example 4.7] produces, from every ergodic finite-intensity point process, a hyperuniform process satisfying Cov η ( N A , N B ) ≤ 0 \operatorname{Cov}_{\eta}(N_A,N_B) \le0 Cov η ( N A , N B ) ≤ 0 for disjoint bounded Borel sets A A A , B B B . By Corollary 5.5 and Proposition 5.2 , this class is nowhere dense in E P P ρ , 2 \mathbf{EPP}_{\rho,2} EPP ρ , 2 ; the fair-partition construction therefore does not prove density of hyperuniform processes.
Maximal rigidity
For a Borel set B ⊂ R d B \subset\mathbb{R}^{d} B ⊂ R d , write ω ∣ B : = ω ( ⋅ ∩ B ) \omega|_{B} := \omega(\cdot\cap B) ω ∣ B := ω ( ⋅ ∩ B ) . For bounded Borel A A A , let
F A c : = σ ( N E : E ⊂ A c bounded Borel ) . \mathcal{F}_{A^{c}} := \sigma(N_E : E \subset A^{c}\text{ bounded Borel}). F A c := σ ( N E : E ⊂ A c bounded Borel ) .
The process η ∈ P P ρ , 2 \eta\in\mathbf{PP}_{\rho,2} η ∈ PP ρ , 2 is number rigid if, for every bounded Borel A A A , the count N A N_A N A is F A c \mathcal{F}_{A^{c}} F A c -measurable modulo η \eta η -null sets. It is maximally rigid if, for every such A A A , the restriction ω ∣ A \omega|_{A} ω ∣ A is F A c \mathcal{F}_{A^{c}} F A c -measurable modulo η \eta η -null sets. Equivalently, every N E N_E N E with E ⊂ A E \subset A E ⊂ A bounded Borel is F A c \mathcal{F}_{A^{c}} F A c -measurable. Write M R i g ρ ⊂ P P ρ , 2 \mathbf{MRig}_{\rho} \subset\mathbf{PP}_{\rho,2} MRig ρ ⊂ PP ρ , 2 for the class of maximally rigid processes.
We use real L 2 L^2 L 2 for rigidity and complex L 2 L^2 L 2 for spectral arguments.
We also consider approximation by exterior linear statistics. For a bounded, compactly supported Borel function f f f and ω ∈ Conf ( R d ) \omega\in\operatorname{Conf}(\mathbb{R}^{d}) ω ∈ Conf ( R d ) , set
S f ( ω ) : = ∫ R d f d ω , Z f η : = S f − ρ ∫ R d f . S_f(\omega) := \int_{\mathbb{R}^{d}} f\,d\omega,\qquad Z_f^\eta:= S_f-\rho\int_{\mathbb{R}^{d}}f. S f ( ω ) := ∫ R d f d ω , Z f η := S f − ρ ∫ R d f .
For each bounded Borel set A A A , let H A c η \mathcal{H}_{A^{c}}^\eta H A c η be the closed span in real L 2 ( η ) L^2(\eta) L 2 ( η ) of Z g η Z_g^\eta Z g η , over all bounded, compactly supported Borel functions g g g vanishing on A A A . We call η \eta η linearly maximally rigid if
Z f η ∈ H A c η Z_f^\eta\in\mathcal{H}_{A^{c}}^\eta Z f η ∈ H A c η
for every bounded Borel A A A and every bounded Borel function f f f vanishing outside A A A . Write L M R i g ρ ⊂ P P ρ , 2 \mathbf{LMRig}_{\rho} \subset\mathbf{PP}_{\rho,2} LMRig ρ ⊂ PP ρ , 2 for this class. Since H A c η ⊂ L 2 ( F A c , η ) \mathcal{H}_{A^{c}}^\eta\subset L^2(\mathcal{F}_{A^{c}},\eta) H A c η ⊂ L 2 ( F A c , η ) , linear maximal rigidity implies maximal rigidity. Indeed, for every bounded Borel E ⊂ A E \subset A E ⊂ A , the random variable N E − ρ Vol d ( E ) N_E-\rho\operatorname{Vol}_{d}(E) N E − ρ Vol d ( E ) is F A c \mathcal{F}_{A^{c}} F A c -measurable.
Theorem B (Rigidity). The set
L M R i g ρ ∩ E P P ρ , 2 \mathbf{LMRig}_{\rho} \cap\mathbf{EPP}_{\rho,2} LMRig ρ ∩ EPP ρ , 2 is residual in E P P ρ , 2 \mathbf{EPP}_{\rho,2} EPP ρ , 2 . In particular, M R i g ρ ∩ E P P ρ , 2 \mathbf{MRig}_{\rho} \cap\mathbf{EPP}_{\rho,2} MRig ρ ∩ EPP ρ , 2 is residual.
Corollary 1.2. The sets
I H y p ρ ∩ L M R i g ρ ∩ E P P ρ , 2 \mathbf{IHyp}_{\rho} \cap\mathbf{LMRig}_{\rho} \cap\mathbf{EPP}_{\rho,2} IHyp ρ ∩ LMRig ρ ∩ EPP ρ , 2
and
( MRig ρ ∖ Hyp ρ ) ∩ EPP ρ , 2 (\operatorname{MRig}_{\rho} \setminus\operatorname{Hyp}_{\rho}) \cap\operatorname{EPP}_{\rho,2} ( MRig ρ ∖ Hyp ρ ) ∩ EPP ρ , 2
are residual in EPP ρ , 2 \operatorname{EPP}_{\rho,2} EPP ρ , 2 .
In Appendix B, we construct a one-dimensional family of linearly maximally rigid point processes with uniformly discrete and relatively dense configurations. Their ball-count variance is bounded along a sequence of radii. The processes are hyperuniform for Lebesgue-almost every parameter, but intermittently hyperuniform for a dense G δ G_{\delta} G δ -set of Liouville parameters. Here the full-measure and residual statements refer to the parameter interval.
The distinction between number rigidity and maximal rigidity is visible in the Ginibre process and the zero process of the planar Gaussian analytic function: their exterior determines certain interior statistics, but not all interior point positions [17 ] . Through uniqueness-pair methods, maximal rigidity also holds for further examples including quasicrystals [32 ] . Spectral criteria for finite-order rigidity provide another approach [30 ] . Deletion and insertion tolerance provide related distinctions [22 ] .
Baire-category methods for measure-preserving transformations and actions originate in the work of Halmos and Rokhlin and are developed, among other places, in [20 , 37 , 18 GW19, 12 , 25 ] . We use related approximation methods for translation-invariant point processes, with intensity and local second moments fixed.
Fixed actions and the local-to-global principle
Let ( X , μ ) (X,\mu) ( X , μ ) be a nonatomic standard probability space and let
a : R d × X ⟶ X a : \mathbb{R}^{d} \times X \longrightarrow X a : R d × X ⟶ X
be an essentially free ergodic probability-measure-preserving (p.m.p.) action. Write t . x : = a ( t , x ) t.x := a(t,x) t . x := a ( t , x ) . A realization of a a a is a measurable map κ : X → Conf ( R d ) \kappa: X \to\operatorname{Conf}(\mathbb{R}^{d}) κ : X → Conf ( R d ) satisfying κ ( t . x ) = θ t κ ( x ) \kappa(t.x) = \theta_t\kappa(x) κ ( t . x ) = θ t κ ( x ) almost everywhere. It is generating if the σ \sigma σ -algebra generated by κ \kappa κ agrees with the full σ \sigma σ -algebra modulo null sets.
We denote by R ρ , 2 a ( X , μ ) \mathcal{R}^{a}_{\rho,2}(X,\mu) R ρ , 2 a ( X , μ ) the Polish space of realizations of intensity ρ \rho ρ with local second moments, in the rooted local L 2 L^{2} L 2 topology, and by R ρ , 2 a , i s o ( X , μ ) \mathcal{R}^{a,\mathrm{iso}}_{\rho,2}(X,\mu) R ρ , 2 a , iso ( X , μ ) its G δ G_{\delta} G δ -subspace of generating realizations. The rooted topology is defined in Subsections 2.1 and 2.2. The push-forward map
R ρ , 2 a ( X , μ ) ⟶ PP ρ , 2 , κ ⟼ κ ∗ μ , \mathcal{R}^{a}_{\rho,2}(X,\mu) \longrightarrow\operatorname{PP}_{\rho,2}, \qquad\kappa\longmapsto\kappa_{*}\mu, R ρ , 2 a ( X , μ ) ⟶ PP ρ , 2 , κ ⟼ κ ∗ μ ,
is continuous by Proposition 2.6 .
We first prove a fixed-action result from which Theorems A and B follow.
Proposition 1.3 (Fixed-action residuality). For every essentially free ergodic p.m.p. action a : R d × X → X a : \mathbb{R}^{d} \times X \to X a : R d × X → X , the set
{ κ ∈ R ρ , 2 a , i s o ( X , μ ) : κ ∗ μ ∈ IHyp ρ ∩ LMRig ρ } \left\{\kappa\in\mathcal{R}^{a,\mathrm{iso}}_{\rho,2}(X,\mu) : \kappa_{*}\mu\in\operatorname{IHyp}_{\rho} \cap\operatorname{LMRig}_{\rho}\right\} { κ ∈ R ρ , 2 a , iso ( X , μ ) : κ ∗ μ ∈ IHyp ρ ∩ LMRig ρ } is residual in R ρ , 2 a , i s o ( X , μ ) \mathcal{R}^{a,\mathrm{iso}}_{\rho,2}(X,\mu) R ρ , 2 a , iso ( X , μ ) .
The essentially free processes form a dense G δ G_{\delta} G δ -subset EPP ρ , 2 f r \operatorname{EPP}^{\mathrm{fr}}_{\rho,2} EPP ρ , 2 fr of EPP ρ , 2 \operatorname{EPP}_{\rho,2} EPP ρ , 2 , by Proposition 2.10. The local-to-global principle, Proposition 2.11 , says that a G δ G_{\delta} G δ -class A ⊂ EPP ρ , 2 \mathcal{A} \subset\operatorname{EPP}_{\rho,2} A ⊂ EPP ρ , 2 is residual whenever its pullback is dense in R ρ , 2 a , i s o ( X , μ ) \mathcal{R}^{a,\mathrm{iso}}_{\rho,2}(X,\mu) R ρ , 2 a , iso ( X , μ ) for every essentially free ergodic action a a a . With Proposition 1.3 , this yields Theorems A and B and Corollary 1.2.
The constructions in [3 ] give maximally rigid Delone (uniformly discrete and relatively dense) realizations of arbitrary essentially free ergodic actions at sufficiently large prescribed intensity.
Fluctuations along van Hove sequences
Let ( F n ) (F_n) ( F n ) be bounded Borel sets satisfying 0 < Vol d ( F n ) < ∞ 0 < \operatorname{Vol}_d(F_n) < \infty 0 < Vol d ( F n ) < ∞ . We call ( F n ) (F_n) ( F n ) a Følner sequence if, for every t ∈ R d t \in\mathbb{R}^d t ∈ R d ,
Vol d ( ( F n + t ) △ F n ) Vol d ( F n ) ⟶ 0 ( n → ∞ ) . \frac{\operatorname{Vol}_d((F_n+t)\mathbin{\triangle}F_n)}{\operatorname{Vol}_d(F_n)} \longrightarrow0 \qquad(n \to\infty). Vol d ( F n ) Vol d (( F n + t ) △ F n ) ⟶ 0 ( n → ∞ ) .
For r > 0 r > 0 r > 0 , define
∂ r F : = { x ∈ R d : dist ( x , ∂ F ) < r } . \partial_r F := \{x \in\mathbb{R}^d : \operatorname{dist}(x,\partial F) < r\}. ∂ r F := { x ∈ R d : dist ( x , ∂ F ) < r } .
The sequence is van Hove if
Vol d ( ∂ r F n ) Vol d ( F n ) ⟶ 0 ( n → ∞ , for every r > 0 ) . \frac{\operatorname{Vol}_d(\partial_r F_n)}{\operatorname{Vol}_d(F_n)} \longrightarrow0 \qquad(n \to\infty,\ \text{for every } r > 0). Vol d ( F n ) Vol d ( ∂ r F n ) ⟶ 0 ( n → ∞ , for every r > 0 ) .
If K ⊂ R d K \subset\mathbb{R}^d K ⊂ R d is a compact convex body with nonempty interior and R n → ∞ R_n \to\infty R n → ∞ , then ( R n K ) (R_nK) ( R n K ) is van Hove. This includes Euclidean balls and closed ℓ p \ell^p ℓ p -balls, 1 ≤ p ≤ ∞ 1 \le p \le\infty 1 ≤ p ≤ ∞ . Every van Hove sequence is Følner. For every η ∈ E P P ρ , 2 \eta\in\mathbf{EPP}_{\rho,2} η ∈ EPP ρ , 2 and every van Hove sequence ( F n ) (F_n) ( F n ) ,
Var η ( N F n ) = o ( Vol d ( F n ) 2 ) ; \operatorname{Var}_{\eta}(N_{F_n}) = o\left(\operatorname{Vol}_d(F_n)^2\right); Var η ( N F n ) = o ( Vol d ( F n ) 2 ) ;
see Proposition 6.2 . The bound can fail for Følner sequences that are not van Hove; see Theorem 6.3.
For η ∈ E P P ρ , 2 \eta\in\mathbf{EPP}_{\rho,2} η ∈ EPP ρ , 2 , we call a sequence ( t n ) ⊂ R d (t_n) \subset\mathbb{R}^d ( t n ) ⊂ R d a revival sequence if ∣ t n ∣ → ∞ \lvert t_n\rvert\to\infty ∣ t n ∣ → ∞ and
∥ N A + t n − N A ∥ L 2 ( η ) ⟶ 0 ( n → ∞ ) \left\|N_{A+t_n}-N_A\right\|_{L^2(\eta)} \longrightarrow0 \qquad(n \to\infty) ∥ N A + t n − N A ∥ L 2 ( η ) ⟶ 0 ( n → ∞ )
for every bounded Borel set A ⊂ R d A \subset\mathbb{R}^d A ⊂ R d with Vol d ( ∂ A ) = 0 \operatorname{Vol}_d(\partial A)=0 Vol d ( ∂ A ) = 0 . Let Rev ρ \operatorname{Rev}_{\rho} Rev ρ be the class of processes admitting a revival sequence. Every subsequence of a revival sequence is again a revival sequence.
For η ∈ E P P ρ , 2 \eta\in\mathbf{EPP}_{\rho,2} η ∈ EPP ρ , 2 , the property of near-quadratic van Hove fluctuations means that for every α n > 0 \alpha_n > 0 α n > 0 with α n → 0 \alpha_n \to0 α n → 0 and every M n > 0 M_n > 0 M n > 0 with M n → ∞ M_n \to\infty M n → ∞ , there is a van Hove sequence ( F n ) (F_n) ( F n ) with
Var η ( N F n ) ≥ M n Vol d ( F n ) 2 − α n ( n ≥ 1 ) . \operatorname{Var}_{\eta}(N_{F_n}) \ge M_n \operatorname{Vol}_d(F_n)^{2-\alpha_n} \qquad(n \ge1). Var η ( N F n ) ≥ M n Vol d ( F n ) 2 − α n ( n ≥ 1 ) .
Let MFVH ρ \operatorname{MFVH}_{\rho} MFVH ρ denote the class of processes with this property.
Theorem C (Revival and near-quadratic fluctuations). The set Rev ρ \operatorname{Rev}_{\rho} Rev ρ is residual in E P P ρ , 2 \mathbf{EPP}_{\rho,2} EPP ρ , 2 and Rev ρ ⊆ MFVH ρ \operatorname{Rev}_{\rho} \subseteq\operatorname{MFVH}_{\rho} Rev ρ ⊆ MFVH ρ . Consequently MFVH ρ \operatorname{MFVH}_{\rho} MFVH ρ is residual in E P P ρ , 2 \mathbf{EPP}_{\rho,2} EPP ρ , 2 .
By Propositions A.4 and A.5 , every process in Rev ρ \operatorname{Rev}_{\rho} Rev ρ is linearly maximally rigid.
Organization
§ 2 establishes the Polish spaces and the local-to-global principle. § 3 proves the residuality and meagreness assertions of Theorem A; its density assertion and the final proof appear in Subsection 5.1. The proofs of Proposition 1.3 , Theorem B, and Corollary 1.2 appear at the end of Subsection 4.2. Theorem C follows from Propositions 6.1 and 6.5 in § 7 . Appendix § 8 treats rigidity sequences for actions, and Appendix § 9 gives intermittently hyperuniform point processes.
Point-process spaces and fixed actions
Configuration maps and cross-sections
Let M + ( R d ) \mathcal{M}_{+}(\mathbb{R}^d) M + ( R d ) denote the space of locally finite positive Radon measures on R d \mathbb{R}^d R d . For φ ∈ C c ( R d ) \varphi\in C_c(\mathbb{R}^d) φ ∈ C c ( R d ) and ω ∈ M + ( R d ) \omega\in\mathcal{M}_{+}(\mathbb{R}^d) ω ∈ M + ( R d ) , set
S φ ( ω ) : = ∫ R d φ d ω . S_{\varphi}(\omega) := \int_{\mathbb{R}^d} \varphi\,\mathrm{d}\omega. S φ ( ω ) := ∫ R d φ d ω .
The vague topology on M + ( R d ) \mathcal{M}_{+}(\mathbb{R}^{d}) M + ( R d ) is the weakest topology making all maps S φ S_{\varphi} S φ , φ ∈ C c ( R d ) \varphi\in C_{c}(\mathbb{R}^{d}) φ ∈ C c ( R d ) , continuous.
Let N ( R d ) \mathcal{N}(\mathbb{R}^{d}) N ( R d ) be the subspace of locally finite counting measures, allowing multiplicities. For a locally finite set P ⊂ R d P \subset\mathbb{R}^{d} P ⊂ R d , write
δ P : = ∑ p ∈ P δ p . \delta_{P}:=\sum_{p\in P}\delta_{p}. δ P := p ∈ P ∑ δ p .
The space of locally finite simple counting measures is
Conf ( R d ) : = { δ P : P ⊂ R d is locally finite } . \operatorname{Conf}(\mathbb{R}^{d}):=\{\delta_{P}:P\subset\mathbb{R}^{d}\text{ is locally finite}\}. Conf ( R d ) := { δ P : P ⊂ R d is locally finite } .
With the vague topology, N ( R d ) \mathcal{N}(\mathbb{R}^{d}) N ( R d ) is Polish and its Borel σ \sigma σ -algebra is generated by the evaluation maps
N A ( ω ) : = ω ( A ) , A ⊂ R d bounded and Borel ; N_{A}(\omega):=\omega(A),\qquad A\subset\mathbb{R}^{d}\text{ bounded and Borel}; N A ( ω ) := ω ( A ) , A ⊂ R d bounded and Borel ;
see [9 Proposition 9.1.IV] . The subspace Conf ( R d ) \operatorname{Conf}(\mathbb{R}^{d}) Conf ( R d ) is a G δ G_{\delta} G δ subset of N ( R d ) \mathcal{N}(\mathbb{R}^{d}) N ( R d ) , hence is itself Polish; see [36 ] . Weak convergence of point-process laws will always refer to this vague topology unless stated otherwise.
We use ω \omega ω for the counting measure and supp ω \operatorname{supp}\omega supp ω for its underlying point set. Thus sums indexed by points of a configuration are written as ∑ u ∈ supp ω \sum_{u\in\operatorname{supp}\omega} ∑ u ∈ supp ω .
A point-process law η \eta η on Conf ( R d ) \operatorname{Conf}(\mathbb{R}^{d}) Conf ( R d ) has finite local second moments if
∫ N A ( ω ) 2 d η ( ω ) < ∞ for every bounded Borel A ⊂ R d . (4) \int N_{A}(\omega)^{2}\,d\eta(\omega)<\infty\qquad\text{for every bounded Borel }A\subset\mathbb{R}^{d}. \tag*{(4)} ∫ N A ( ω ) 2 d η ( ω ) < ∞ for every bounded Borel A ⊂ R d . (4)
Equivalently, it suffices to check the cubes Q m : = [ − m / 2 , m / 2 ) d Q_{m}:=[-m/2,m/2)^{d} Q m := [ − m /2 , m /2 ) d , m ≥ 1 m\geq1 m ≥ 1 .
Recall that, for t ∈ R d t\in\mathbb{R}^{d} t ∈ R d ,
( θ t ω ) ( A ) : = ω ( A + t ) . (\theta_{t}\omega)(A):=\omega(A+t). ( θ t ω ) ( A ) := ω ( A + t ) .
Then supp ( θ t ω ) = supp ( ω ) − t \operatorname{supp}(\theta_{t}\omega)=\operatorname{supp}(\omega)-t supp ( θ t ω ) = supp ( ω ) − t . The map ( t , ω ) ↦ θ t ω (t,\omega)\mapsto\theta_{t}\omega ( t , ω ) ↦ θ t ω is continuous. If t n → t t_{n}\to t t n → t and ω n → ω \omega_{n}\to\omega ω n → ω vaguely, then φ ( ⋅ − t n ) → φ ( ⋅ − t ) \varphi(\mathord\cdot-t_{n})\to\varphi(\mathord\cdot-t) φ ( ⋅ − t n ) → φ ( ⋅ − t ) uniformly for every φ ∈ C c ( R d ) \varphi\in C_{c}(\mathbb{R}^{d}) φ ∈ C c ( R d ) . These functions have support in a common compact set, whose ω n \omega_{n} ω n -masses are bounded. Consequently, S φ ( θ t n ω n ) → S φ ( θ t ω ) S_{\varphi}(\theta_{t_{n}}\omega_{n})\to S_{\varphi}(\theta_{t}\omega) S φ ( θ t n ω n ) → S φ ( θ t ω ) .
Throughout, ( X , μ ) (X,\mu) ( X , μ ) is a nonatomic standard probability space. Let Ad ( X , μ ) \operatorname{Ad}(X,\mu) Ad ( X , μ ) denote the essentially free ergodic p.m.p. actions
a : R d × X ⟶ X . a:\mathbb{R}^{d}\times X\longrightarrow X. a : R d × X ⟶ X .
For a ∈ Ad ( X , μ ) a\in\operatorname{Ad}(X,\mu) a ∈ Ad ( X , μ ) , set t . x : = a ( t , x ) t.x:=a(t,x) t . x := a ( t , x ) and denote by
L 0 ( X , Conf ( R d ) , μ ) L^{0}(X,\operatorname{Conf}(\mathbb{R}^{d}),\mu) L 0 ( X , Conf ( R d ) , μ )
the space of equivalence classes, modulo μ \mu μ -almost everywhere equality, of measurable maps from X X X to Conf ( R d ) \operatorname{Conf}(\mathbb{R}^{d}) Conf ( R d ) . A class κ \kappa κ is equivariant if, for every t ∈ R d t\in\mathbb{R}^{d} t ∈ R d ,
κ ( t . x ) = θ t κ ( x ) \kappa(t.x)=\theta_{t}\kappa(x) κ ( t . x ) = θ t κ ( x )
for μ \mu μ -almost every x ∈ X x\in X x ∈ X .
The exceptional null set may depend on t t t . For each fixed t t t , equivariance and invariance of μ \mu μ give
( θ t ) ∗ ( κ ∗ μ ) = ( κ ( t . ⋅ ) ) ∗ μ = κ ∗ μ . (\theta_{t})_{*}(\kappa_{*}\mu)=(\kappa(t.\mathord\cdot))_{*}\mu=\kappa_{*}\mu. ( θ t ) ∗ ( κ ∗ μ ) = ( κ ( t . ⋅ ) ) ∗ μ = κ ∗ μ .
Lemma 2.1. Let κ ∈ L 0 ( X , Conf ( R d ) , μ ) \kappa\in L^{0}(X,\operatorname{Conf}(\mathbb{R}^{d}),\mu) κ ∈ L 0 ( X , Conf ( R d ) , μ ) be equivariant. There exist an invariant conull Borel set X κ ⊂ X X_{\kappa}\subset X X κ ⊂ X and a Borel representative
κ ∘ : X κ ⟶ Conf ( R d ) \kappa^{\circ}:X_{\kappa}\longrightarrow\operatorname{Conf}(\mathbb{R}^{d}) κ ∘ : X κ ⟶ Conf ( R d ) of κ \kappa κ such that
κ ∘ ( t . x ) = θ t κ ∘ ( x ) ( t , x ) ∈ R d × X κ . \kappa^{\circ}(t.x)=\theta_{t}\kappa^{\circ}(x)\qquad(t,x)\in\mathbb{R}^{d}\times X_{\kappa}. κ ∘ ( t . x ) = θ t κ ∘ ( x ) ( t , x ) ∈ R d × X κ . If κ ( x ) \kappa(x) κ ( x ) is nonempty almost everywhere, then X κ X_{\kappa} X κ may be chosen so that κ ∘ ( x ) \kappa^{\circ}(x) κ ∘ ( x ) is nonempty for every x ∈ X κ x\in X_{\kappa} x ∈ X κ .
Proof. Apply [1 Proposition 5.7] to the measurable factor κ \kappa κ . □
Let X 0 ⊂ X X_0 \subset X X 0 ⊂ X be an invariant conull Borel set. A Borel set Y ⊂ X 0 Y \subset X_0 Y ⊂ X 0 is a cross-section if, for every x ∈ X 0 x \in X_0 x ∈ X 0 , the return-time set
Y x : = { t ∈ R d : t . x ∈ Y } Y_x := \{t \in\mathbb{R}^d : t.x \in Y\} Y x := { t ∈ R d : t . x ∈ Y }
is nonempty and locally finite.
Proposition 2.2. Let κ ∈ L 0 ( X , Conf ( R d ) , μ ) \kappa\in L^0(X,\operatorname{Conf}(\mathbb{R}^d),\mu) κ ∈ L 0 ( X , Conf ( R d ) , μ ) be equivariant and nonempty almost everywhere. Then there exist an invariant conull Borel set X κ ⊂ X X_\kappa\subset X X κ ⊂ X and a nonempty Borel representative κ ∘ : X κ → Conf ( R d ) \kappa^\circ: X_\kappa\to\operatorname{Conf}(\mathbb{R}^d) κ ∘ : X κ → Conf ( R d ) satisfying
κ ∘ ( t . x ) = θ t κ ∘ ( x ) ( t , x ) ∈ R d × X κ , \kappa^\circ(t.x) = \theta_t\kappa^\circ(x) \qquad(t,x) \in\mathbb{R}^d \times X_\kappa, κ ∘ ( t . x ) = θ t κ ∘ ( x ) ( t , x ) ∈ R d × X κ , such that
Y κ : = { x ∈ X κ : N { 0 } ( κ ∘ ( x ) ) = 1 } Y_\kappa:= \{x \in X_\kappa: N_{\{0\}}(\kappa^\circ(x)) = 1\} Y κ := { x ∈ X κ : N { 0 } ( κ ∘ ( x )) = 1 } is a cross-section and
( Y κ ) x = supp ( κ ∘ ( x ) ) , x ∈ X κ . (Y_\kappa)_x = \operatorname{supp}(\kappa^\circ(x)), \qquad x \in X_\kappa. ( Y κ ) x = supp ( κ ∘ ( x )) , x ∈ X κ .
Proof. Choose X κ X_\kappa X κ and κ ∘ \kappa^\circ κ ∘ from Lemma 2.1. The set Y κ Y_\kappa Y κ is Borel because N { 0 } N_{\{0\}} N { 0 } is Borel on Conf ( R d ) \operatorname{Conf}(\mathbb{R}^d) Conf ( R d ) . For t ∈ R d t \in\mathbb{R}^d t ∈ R d and x ∈ X κ x \in X_\kappa x ∈ X κ , pointwise equivariance gives
t ∈ ( Y κ ) x ⟺ 0 ∈ supp ( κ ∘ ( t . x ) ) ⟺ 0 ∈ supp ( κ ∘ ( x ) ) − t ⟺ t ∈ supp ( κ ∘ ( x ) ) . t \in(Y_\kappa)_x \Longleftrightarrow0 \in\operatorname{supp}(\kappa^\circ(t.x)) \Longleftrightarrow0 \in\operatorname{supp}(\kappa^\circ(x))-t \Longleftrightarrow t \in\operatorname{supp}(\kappa^\circ(x)). t ∈ ( Y κ ) x ⟺ 0 ∈ supp ( κ ∘ ( t . x )) ⟺ 0 ∈ supp ( κ ∘ ( x )) − t ⟺ t ∈ supp ( κ ∘ ( x )) .
Since supp ( κ ∘ ( x ) ) \operatorname{supp}(\kappa^\circ(x)) supp ( κ ∘ ( x )) is nonempty and locally finite, Y κ Y_\kappa Y κ is a cross-section. □
Conversely, a cross-section Y ⊂ X 0 Y \subset X_0 Y ⊂ X 0 gives the equivariant Borel map
κ Y ( x ) : = ∑ t ∈ Y x δ t . \kappa_Y(x) := \sum_{t \in Y_x} \delta_t. κ Y ( x ) := t ∈ Y x ∑ δ t .
The relation { ( x , t ) : x ∈ X 0 , t . x ∈ Y } \{(x,t) : x \in X_0,\ t.x \in Y\} {( x , t ) : x ∈ X 0 , t . x ∈ Y } is Borel, and its sections Y x Y_x Y x are countable by local finiteness. Lusin–Novikov [24 Theorem 18.10] expresses this relation as a union of graphs of partial Borel maps t j t_j t j . For each bounded Borel D D D , the function x ↦ N D ( κ Y ( x ) ) = ∑ j : x ∈ dom ( t j ) 1 D ( t j ( x ) ) x \mapsto N_D(\kappa_Y(x)) = \sum_{j:x\in\operatorname{dom}(t_j)} 1_D(t_j(x)) x ↦ N D ( κ Y ( x )) = ∑ j : x ∈ dom ( t j ) 1 D ( t j ( x )) is Borel. The identity Y t . x = Y x − t Y_{t.x} = Y_x-t Y t . x = Y x − t gives κ Y ( t . x ) = θ t κ Y ( x ) \kappa_Y(t.x) = \theta_t\kappa_Y(x) κ Y ( t . x ) = θ t κ Y ( x ) .
On an invariant conull Borel set X 0 X_0 X 0 where a a a is free, [7 Theorem A.3] , applied with the compact set B ‾ r / 2 \overline{B}_{r/2} B r /2 , gives a cocompact Borel cross-section C r C_r C r for every r > 0 r > 0 r > 0 such that each return set ( C r ) x (C_r)_x ( C r ) x is r r r -separated. Its configuration map has uniformly bounded local counts, but need not have intensity ρ \rho ρ .
Voronoi cells of a cross-section. Fix an r r r -separated cocompact Borel cross-section C ⊂ X 0 C \subset X_0 C ⊂ X 0 . Cocompactness gives R < ∞ R < \infty R < ∞ such that C x C_x C x is R R R -dense in R d \mathbb{R}^d R d for every x ∈ X 0 x \in X_0 x ∈ X 0 . Choose a Borel injection ℓ : C → [ 0 , 1 ] \ell: C \to[0,1] ℓ : C → [ 0 , 1 ] , which exists by [24 Theorem 15.6] . For c ∈ C c \in C c ∈ C , define
W c : = { t ∈ R d : ( ∥ t ∥ , ℓ ( c ) ) ≤ l e x ( ∥ t − v ∥ , ℓ ( v . c ) ) for every v ∈ C c } . (5) W_c := \{t \in\mathbb{R}^d : (\lVert t\rVert,\ell(c)) \le_{\mathrm{lex}} (\lVert t-v\rVert,\ell(v.c)) \text{ for every } v \in C_c\}. \tag*{(5)} W c := { t ∈ R d : (∥ t ∥ , ℓ ( c )) ≤ lex (∥ t − v ∥ , ℓ ( v . c )) for every v ∈ C c } . (5)
where ≤ l e x \le_{\mathrm{lex}} ≤ lex denotes the lexicographic order. Thus the cell W c W_c W c consists of the points for which 0 0 0 is a nearest return vector in C c C_c C c , with equal distances resolved by ℓ \ell ℓ . In particular, W c W_c W c depends on the chosen cross-section and its Borel injection; we retain the notation W c W_c W c when both are fixed.
The relation { ( c , v ) ∈ C × R d : v . c ∈ C } \{(c,v) \in C \times\mathbb{R}^d : v.c \in C\} {( c , v ) ∈ C × R d : v . c ∈ C } is Borel and has countable sections, since C c C_c C c is r r r -separated. By Lusin–Novikov [24 Theorem 18.10] , it is the union of graphs of partial Borel maps v j : D j → R d v_j : D_j \to\mathbb{R}^d v j : D j → R d . Applying the comparisons in (5) to these maps proves that { ( c , t ) : c ∈ C , t ∈ W c } \{(c,t) : c \in C,\ t \in W_c\} {( c , t ) : c ∈ C , t ∈ W c } is Borel. Each comparison defines an open or closed half-space, so W c W_c W c is convex. Separation and cocompactness give
B r / 2 ∘ ⊂ W c ⊂ B ‾ R . B^\circ_{r/2} \subset W_c \subset\overline{B}_R. B r /2 ∘ ⊂ W c ⊂ B R .
For every x ∈ X 0 x \in X_0 x ∈ X 0 , the sets u + W u . x u+W_{u.x} u + W u . x , u ∈ C x u \in C_x u ∈ C x , partition R d \mathbb{R}^d R d : each t t t has a nearest return vector u ∈ C x u \in C_x u ∈ C x , and ℓ \ell ℓ resolves ties between the finitely many nearest returns. The partition is equivariant under translation and locally finite, since all its cells lie within distance R R R of their return vectors.
Rooted realization spaces
Fix a complete compatible metric d C o n f d_{\mathrm{Conf}} d Conf on Conf ( R d ) \operatorname{Conf}(\mathbb{R}^d) Conf ( R d ) , bounded by 1. Choose a countable dense family ( ω j ) j ≥ 1 (\omega_j)_{j\geq1} ( ω j ) j ≥ 1 in this space, and let D D D be the countable unital Q \mathbb{Q} Q -algebra generated by the functions ω ↦ d C o n f ( ω , ω j ) \omega\mapsto d_{\mathrm{Conf}}(\omega,\omega_j) ω ↦ d Conf ( ω , ω j ) . Every element of D D D is bounded and continuous. For κ , λ ∈ L 0 ( X , Conf ( R d ) , μ ) \kappa,\lambda\in L^0(X,\operatorname{Conf}(\mathbb{R}^d),\mu) κ , λ ∈ L 0 ( X , Conf ( R d ) , μ ) , set
d 0 ( κ , λ ) : = ∫ X d C o n f ( κ ( x ) , λ ( x ) ) d μ ( x ) . d_0(\kappa,\lambda):=\int_X d_{\mathrm{Conf}}(\kappa(x),\lambda(x))\,d\mu(x). d 0 ( κ , λ ) := ∫ X d Conf ( κ ( x ) , λ ( x )) d μ ( x ) .
Convergence in d 0 d_0 d 0 is convergence in measure, since for every ε > 0 \varepsilon>0 ε > 0 ,
ε μ { d C o n f ( κ , λ ) > ε } ≤ d 0 ( κ , λ ) ≤ ε + μ { d C o n f ( κ , λ ) > ε } . \varepsilon\mu\{d_{\mathrm{Conf}}(\kappa,\lambda)>\varepsilon\}\leq d_0(\kappa,\lambda)\leq\varepsilon+\mu\{d_{\mathrm{Conf}}(\kappa,\lambda)>\varepsilon\}. ε μ { d Conf ( κ , λ ) > ε } ≤ d 0 ( κ , λ ) ≤ ε + μ { d Conf ( κ , λ ) > ε } .
Recall the cubes Q m Q_m Q m from (4) . An equivariant κ ∈ L 0 ( X , Conf ( R d ) , μ ) \kappa\in L^0(X,\operatorname{Conf}(\mathbb{R}^d),\mu) κ ∈ L 0 ( X , Conf ( R d ) , μ ) has finite local second moments if N Q m ∘ κ ∈ L 2 ( X , μ ) N_{Q_m}\circ\kappa\in L^2(X,\mu) N Q m ∘ κ ∈ L 2 ( X , μ ) for every $m\geq 1; equivalently, its law κ ∗ μ \kappa_*\mu κ ∗ μ satisfies (4) . It has intensity ρ \rho ρ if
∫ X N A ( κ ( x ) ) d μ ( x ) = ρ Vol d ( A ) \int_X N_A(\kappa(x))\,d\mu(x)=\rho\operatorname{Vol}_d(A) ∫ X N A ( κ ( x )) d μ ( x ) = ρ Vol d ( A )
for every bounded Borel set A ⊂ R d A\subset\mathbb{R}^d A ⊂ R d .
Let R ρ , 2 a ( X , μ ) \mathbb{R}_{\rho,2}^{a}(X,\mu) R ρ , 2 a ( X , μ ) be the set of a a a -equivariant classes of intensity ρ \rho ρ with finite local second moments, equipped with the metric
d r o o t , 2 ( κ , λ ) : = d 0 ( κ , λ ) + ∑ m = 1 ∞ 2 − m min { 1 , ∥ N Q m ∘ κ − N Q m ∘ λ ∥ 2 } . d_{\mathrm{root},2}(\kappa,\lambda):=d_0(\kappa,\lambda)+\sum_{m=1}^{\infty}2^{-m}\min\{1,\|N_{Q_m}\circ\kappa-N_{Q_m}\circ\lambda\|_2\}. d root , 2 ( κ , λ ) := d 0 ( κ , λ ) + m = 1 ∑ ∞ 2 − m min { 1 , ∥ N Q m ∘ κ − N Q m ∘ λ ∥ 2 } .
This metric defines the rooted local L 2 L^2 L 2 topology . A sequence ( κ n ) (\kappa_n) ( κ n ) converges to κ \kappa κ in R ρ , 2 a ( X , μ ) \mathbb{R}_{\rho,2}^{a}(X,\mu) R ρ , 2 a ( X , μ ) if and only if κ n → κ \kappa_n\to\kappa κ n → κ in measure and
N Q m ∘ κ n ⟶ N Q m ∘ κ in L 2 ( X , μ ) ( n → ∞ , m ≥ 1 ) . N_{Q_m}\circ\kappa_n\longrightarrow N_{Q_m}\circ\kappa\quad\text{in }L^2(X,\mu)\qquad(n\to\infty,\ m\geq1). N Q m ∘ κ n ⟶ N Q m ∘ κ in L 2 ( X , μ ) ( n → ∞ , m ≥ 1 ) .
For κ ∈ R ρ , 2 a ( X , μ ) \kappa\in\mathbb{R}_{\rho,2}^{a}(X,\mu) κ ∈ R ρ , 2 a ( X , μ ) , let F κ \mathcal{F}_\kappa F κ be the μ \mu μ -completion of the σ \sigma σ -algebra generated by κ \kappa κ . An element κ \kappa κ is generating if F κ = B X μ \mathcal{F}_\kappa=\mathcal{B}^{\mu}_X F κ = B X μ . Define
R ρ , 2 a , i s o ( X , μ ) : = { κ ∈ R ρ , 2 a ( X , μ ) : F κ = B X μ } . \mathbb{R}_{\rho,2}^{a,\mathrm{iso}}(X,\mu):=\{\kappa\in\mathbb{R}_{\rho,2}^{a}(X,\mu):\mathcal{F}_\kappa=\mathcal{B}^{\mu}_X\}. R ρ , 2 a , iso ( X , μ ) := { κ ∈ R ρ , 2 a ( X , μ ) : F κ = B X μ } .
For such κ \kappa κ , the factor map κ \kappa κ is a measurable isomorphism, modulo null sets, from R d ↷ a ( X , μ ) \mathbb{R}^d\mathrel{\overset{a}{\curvearrowright}}(X,\mu) R d ↷ a ( X , μ ) to the translation action on ( Conf ( R d ) , κ ∗ μ ) (\operatorname{Conf}(\mathbb{R}^d),\kappa_*\mu) ( Conf ( R d ) , κ ∗ μ ) : on standard Borel probability spaces, a generating factor admits a measurable inverse modulo null sets by [24 Theorem 15.1] .
Lemma 2.3. Let η \eta η be a translation-invariant probability measure on Conf ( R d ) \operatorname{Conf}(\mathbb{R}^d) Conf ( R d ) . If A ⊂ R d A\subset\mathbb{R}^d A ⊂ R d is bounded and Borel, with Vol d ( A ) = 0 \operatorname{Vol}_d(A)=0 Vol d ( A ) = 0 , then N A = 0 N_A=0 N A = 0 for η \eta η -almost every configuration.
Proof. Fix a bounded Borel set B ⊂ R d B\subset\mathbb{R}^d B ⊂ R d with positive volume. Tonelli’s theorem gives
∫ B ∫ N A + t ( ω ) d η ( ω ) d t = ∫ ∑ p ∈ supp ( ω ) Vol d ( B ∩ ( p − A ) ) d η ( ω ) = 0. \int_B\int N_{A+t}(\omega)\,d\eta(\omega)\,dt=\int\sum_{p\in\operatorname{supp}(\omega)}\operatorname{Vol}_d(B\cap(p-A))\,d\eta(\omega)=0. ∫ B ∫ N A + t ( ω ) d η ( ω ) d t = ∫ p ∈ supp ( ω ) ∑ Vol d ( B ∩ ( p − A )) d η ( ω ) = 0.
Translation invariance makes the inner expectation on the left independent of t t t , so ∫ N A d η = 0 \int N_A\,d\eta=0 ∫ N A d η = 0 . □ \square □
Lemma 2.4 (Vitali convergence). Let ( f n ) (f_n) ( f n ) be measurable functions on a probability space and let 1 ≤ p < ∞ 1\leq p<\infty 1 ≤ p < ∞ . If f n → f f_n\to f f n → f in measure and { ∣ f n ∣ p : n ≥ 1 } \{|f_n|^p:n\geq1\} { ∣ f n ∣ p : n ≥ 1 } is uniformly integrable, then f ∈ L p f\in L^p f ∈ L p and f n → f f_n\to f f n → f in L p L^p L p .
Proof. Every subsequence has a further subsequence converging almost everywhere to f f f . Fatou gives f ∈ L p f \in L^p f ∈ L p , while ∣ f n − f ∣ p ≤ 2 p − 1 ( ∣ f n ∣ p + ∣ f ∣ p ) |f_n-f|^p \le2^{p-1}(|f_n|^p+|f|^p) ∣ f n − f ∣ p ≤ 2 p − 1 ( ∣ f n ∣ p + ∣ f ∣ p ) makes the differences uniformly integrable. Vitali’s theorem gives L p L^p L p -convergence along the further subsequence, hence along the original sequence. □
Proposition 2.5 (Polish structure of realization spaces). The space R ρ , 2 a ( X , μ ) \mathbf{R}^{a}_{\rho,2}(X,\mu) R ρ , 2 a ( X , μ ) is Polish. Moreover:
if A ⊂ R d A \subset\mathbb{R}^d A ⊂ R d is bounded and Borel, with Vol d ( ∂ A ) = 0 \operatorname{Vol}_d(\partial A)=0 Vol d ( ∂ A ) = 0 , then κ ↦ N A ∘ κ \kappa\mapsto N_A \circ\kappa κ ↦ N A ∘ κ is continuous into L 2 ( X , μ ) L^2(X,\mu) L 2 ( X , μ ) ;
if φ ∈ C c ( R d ) \varphi\in C_c(\mathbb{R}^d) φ ∈ C c ( R d ) , then κ ↦ S φ ∘ κ \kappa\mapsto S_\varphi\circ\kappa κ ↦ S φ ∘ κ is continuous into L 2 ( X , μ ) L^2(X,\mu) L 2 ( X , μ ) ;
R ρ , 2 a , i s o ( X , μ ) \mathbf{R}^{a,\mathrm{iso}}_{\rho,2}(X,\mu) R ρ , 2 a , iso ( X , μ ) is a G δ G_\delta G δ subset of R ρ , 2 a ( X , μ ) \mathbf{R}^{a}_{\rho,2}(X,\mu) R ρ , 2 a ( X , μ ) , hence Polish.
Proof. The ambient space L 0 ( X , Conf ( R d ) , μ ) L^0(X,\operatorname{Conf}(\mathbb{R}^d),\mu) L 0 ( X , Conf ( R d ) , μ ) is Polish for convergence in measure. A d 0 d_0 d 0 -Cauchy sequence has a subsequence whose successive d 0 d_0 d 0 -distances are summable; Tonelli and completeness of d Conf d_{\operatorname{Conf}} d Conf give an almost-everywhere pointwise limit, and hence a d 0 d_0 d 0 -limit. Separability follows after identifying ( X , μ ) (X,\mu) ( X , μ ) with ( [ 0 , 1 ] , Leb ) ([0,1],\operatorname{Leb}) ([ 0 , 1 ] , Leb ) modulo null sets [24 Theorem 17.41] : step functions with rational intervals and values in a fixed countable dense subset of Conf ( R d ) \operatorname{Conf}(\mathbb{R}^d) Conf ( R d ) are dense.
Consider
J : R ρ , 2 a ( X , μ ) ⟶ L 0 ( X , Conf ( R d ) , μ ) × ∏ m ≥ 1 L 2 ( X , μ ) , J ( κ ) : = ( κ , ( N Q m ∘ κ ) m ≥ 1 ) . J:\mathbf{R}^{a}_{\rho,2}(X,\mu)\longrightarrow L^0(X,\operatorname{Conf}(\mathbb{R}^d),\mu)\times\prod_{m\ge1}L^2(X,\mu),\qquad J(\kappa):=\left(\kappa,(N_{Q_m}\circ\kappa)_{m\ge1}\right). J : R ρ , 2 a ( X , μ ) ⟶ L 0 ( X , Conf ( R d ) , μ ) × m ≥ 1 ∏ L 2 ( X , μ ) , J ( κ ) := ( κ , ( N Q m ∘ κ ) m ≥ 1 ) .
On the product we use d 0 d_0 d 0 on the first coordinate and ∑ m ≥ 1 2 − m min { 1 , ∥ g m − h m ∥ 2 } \sum_{m\ge1}2^{-m}\min\{1,\|g_m-h_m\|_2\} ∑ m ≥ 1 2 − m min { 1 , ∥ g m − h m ∥ 2 } on the L 2 L^2 L 2 coordinates. With these metrics, J J J is isometric. Let J ( κ n ) → ( κ , ( g m ) m ≥ 1 ) J(\kappa_n)\to(\kappa,(g_m)_{m\ge1}) J ( κ n ) → ( κ , ( g m ) m ≥ 1 ) in the product. For each fixed t ∈ R d t\in\mathbb{R}^d t ∈ R d , composition with x ↦ t . x x\mapsto t.x x ↦ t . x is an isometry of L 0 L^0 L 0 , while composition with θ t \theta_t θ t is continuous in measure because θ t \theta_t θ t is continuous on Conf ( R d ) \operatorname{Conf}(\mathbb{R}^d) Conf ( R d ) . Passing to the limit in κ n ( t . x ) = θ t κ n ( x ) \kappa_n(t.x)=\theta_t\kappa_n(x) κ n ( t . x ) = θ t κ n ( x ) shows that κ \kappa κ is equivariant. Hence κ ∗ μ \kappa_*\mu κ ∗ μ is translation-invariant.
After passing to a subsequence, still denoted ( κ n ) (\kappa_n) ( κ n ) , we may assume ∑ n d 0 ( κ n , κ ) < ∞ \sum_n d_0(\kappa_n,\kappa)<\infty ∑ n d 0 ( κ n , κ ) < ∞ . Tonelli’s theorem then gives κ n ( x ) → κ ( x ) \kappa_n(x)\to\kappa(x) κ n ( x ) → κ ( x ) vaguely for almost every x x x . By Lemma 2.3 , κ ( x ) ( ∂ Q m ) = 0 \kappa(x)(\partial Q_m)=0 κ ( x ) ( ∂ Q m ) = 0 for every m m m outside one null set. For such an x x x , the continuity-set criterion for vague convergence [8 Appendix A2.6] gives
N Q m ( κ n ( x ) ) ⟶ N Q m ( κ ( x ) ) ( n → ∞ ) . N_{Q_m}(\kappa_n(x))\longrightarrow N_{Q_m}(\kappa(x))\qquad(n\to\infty). N Q m ( κ n ( x )) ⟶ N Q m ( κ ( x )) ( n → ∞ ) .
Since N Q m ∘ κ n → g m N_{Q_m}\circ\kappa_n\to g_m N Q m ∘ κ n → g m in L 2 L^2 L 2 for every m m m , a diagonal subsequence gives g m = N Q m ∘ κ g_m=N_{Q_m}\circ\kappa g m = N Q m ∘ κ almost everywhere for all m m m . Hence N Q m ∘ κ ∈ L 2 ( X , μ ) N_{Q_m}\circ\kappa\in L^2(X,\mu) N Q m ∘ κ ∈ L 2 ( X , μ ) for every m m m .
Let A ⊂ R d A\subset\mathbb{R}^d A ⊂ R d be a bounded Borel set with Vol d ( ∂ A ) = 0 \operatorname{Vol}_d(\partial A)=0 Vol d ( ∂ A ) = 0 , and choose m m m with A ⊂ Q m A\subset Q_m A ⊂ Q m . By Lemma 2.3 and the continuity-set criterion for vague convergence, N A ∘ κ n → N A ∘ κ N_A\circ\kappa_n\to N_A\circ\kappa N A ∘ κ n → N A ∘ κ in measure. Since N Q m ∘ κ n → N Q m ∘ κ N_{Q_m}\circ\kappa_n\to N_{Q_m}\circ\kappa N Q m ∘ κ n → N Q m ∘ κ in L 2 L^2 L 2 , the squares of these local counts are uniformly integrable. Since N A ≤ N Q m N_A\le N_{Q_m} N A ≤ N Q m , the variables N A ∘ κ n N_A\circ\kappa_n N A ∘ κ n are uniformly integrable. Hence
∫ X N A ( κ ( x ) ) d μ ( x ) = lim n → ∞ ∫ X N A ( κ n ( x ) ) d μ ( x ) = ρ Vol d ( A ) . \int_X N_A(\kappa(x))\,d\mu(x)=\lim_{n\to\infty}\int_X N_A(\kappa_n(x))\,d\mu(x)=\rho\,\operatorname{Vol}_d(A). ∫ X N A ( κ ( x )) d μ ( x ) = n → ∞ lim ∫ X N A ( κ n ( x )) d μ ( x ) = ρ Vol d ( A ) .
Uniqueness of locally finite measures extends the identity from rational half-open rectangles to all bounded Borel sets. Hence κ \kappa κ has intensity ρ \rho ρ .
Thus J ( R ρ , 2 a ( X , μ ) ) J(\mathbf{R}^{a}_{\rho,2}(X,\mu)) J ( R ρ , 2 a ( X , μ )) is closed in a Polish space, and R ρ , 2 a ( X , μ ) \mathbf{R}^{a}_{\rho,2}(X,\mu) R ρ , 2 a ( X , μ ) is Polish.
Continuity of local statistics. Let κ n → κ \kappa_n\to\kappa κ n → κ in R ρ , 2 a ( X , μ ) \mathbf{R}^{a}_{\rho,2}(X,\mu) R ρ , 2 a ( X , μ ) , let A ⊂ R d A\subset\mathbb{R}^d A ⊂ R d be bounded and Borel, with Vol d ( ∂ A ) = 0 \operatorname{Vol}_d(\partial A)=0 Vol d ( ∂ A ) = 0 , and choose m m m with A ⊂ Q m A\subset Q_m A ⊂ Q m . By Lemma 2.3 , κ ( x ) ( ∂ A ) = 0 \kappa(x)(\partial A)=0 κ ( x ) ( ∂ A ) = 0 for almost every x x x . Vague convergence in measure therefore gives N A ∘ κ n → N A ∘ κ N_A\circ\kappa_n\to N_A\circ\kappa N A ∘ κ n → N A ∘ κ in measure, while
∣ N A ∘ κ n − N A ∘ κ ∣ 2 ≤ 2 ( N Q m ∘ κ n ) 2 + 2 ( N Q m ∘ κ ) 2 . |N_A\circ\kappa_n-N_A\circ\kappa|^2\le2(N_{Q_m}\circ\kappa_n)^2+2(N_{Q_m}\circ\kappa)^2. ∣ N A ∘ κ n − N A ∘ κ ∣ 2 ≤ 2 ( N Q m ∘ κ n ) 2 + 2 ( N Q m ∘ κ ) 2 .
Since N Q m ∘ κ n → N Q m ∘ κ N_{Q_m} \circ\kappa_n \to N_{Q_m} \circ\kappa N Q m ∘ κ n → N Q m ∘ κ in L 2 L^2 L 2 , the squares on the right are uniformly integrable. Hence Lemma 2.4 gives N A ∘ κ n → N A ∘ κ N_A \circ\kappa_n \to N_A \circ\kappa N A ∘ κ n → N A ∘ κ in L 2 L^2 L 2 .
Let φ ∈ C c ( R d ) \varphi\in C_c(\mathbb{R}^d) φ ∈ C c ( R d ) and choose m m m with supp φ ⊂ Q m \operatorname{supp}\varphi\subset Q_m supp φ ⊂ Q m . Vague convergence in measure gives S φ ∘ κ n → S φ ∘ κ S_\varphi\circ\kappa_n \to S_\varphi\circ\kappa S φ ∘ κ n → S φ ∘ κ in measure, while
∣ S φ ∘ κ n − S φ ∘ κ ∣ 2 ≤ 2 ∥ φ ∥ ∞ 2 ( ( N Q m ∘ κ n ) 2 + ( N Q m ∘ κ ) 2 ) . \left|S_\varphi\circ\kappa_n-S_\varphi\circ\kappa\right|^2 \leq2\|\varphi\|_\infty^2\left((N_{Q_m}\circ\kappa_n)^2+(N_{Q_m}\circ\kappa)^2\right). ∣ S φ ∘ κ n − S φ ∘ κ ∣ 2 ≤ 2∥ φ ∥ ∞ 2 ( ( N Q m ∘ κ n ) 2 + ( N Q m ∘ κ ) 2 ) .
By Lemma 2.4 , the linear statistics converge in L 2 L^2 L 2 .
Generating realizations. The algebra D \mathcal{D} D defined above separates points and generates the Borel σ \sigma σ -algebra. Hence the functional monotone-class theorem shows that D \mathcal{D} D is dense in real L 2 ( η ) L^2(\eta) L 2 ( η ) for every probability measure η \eta η on Conf ( R d ) \operatorname{Conf}(\mathbb{R}^d) Conf ( R d ) .
Let ( u j ) j ≥ 1 (u_j)_{j\geq1} ( u j ) j ≥ 1 be a dense family of bounded real-valued functions in real L 2 ( X , μ ) L^2(X,\mu) L 2 ( X , μ ) . The map κ \kappa κ is generating if and only if, for every j , n ≥ 1 j,n\geq1 j , n ≥ 1 , there exists h ∈ D h\in\mathcal{D} h ∈ D such that
∥ u j − h ∘ κ ∥ 2 < 1 n . \|u_j-h\circ\kappa\|_2<\frac{1}{n}. ∥ u j − h ∘ κ ∥ 2 < n 1 .
For a fixed bounded continuous h h h , convergence in d 0 d_0 d 0 gives h ∘ κ n → h ∘ κ h\circ\kappa_n\to h\circ\kappa h ∘ κ n → h ∘ κ in measure and, by boundedness, in L 2 L^2 L 2 . Thus each set
{ κ ∈ R ρ , 2 a ( X , μ ) : ∥ u j − h ∘ κ ∥ 2 < 1 n } \left\{\kappa\in\mathcal{R}^{a}_{\rho,2}(X,\mu):\|u_j-h\circ\kappa\|_2<\frac{1}{n}\right\} { κ ∈ R ρ , 2 a ( X , μ ) : ∥ u j − h ∘ κ ∥ 2 < n 1 }
is open. Therefore
R ρ , 2 a , i s o ( X , μ ) = ⋂ j , n ≥ 1 ⋃ h ∈ D { κ : ∥ u j − h ∘ κ ∥ 2 < 1 n } \mathcal{R}^{a,\mathrm{iso}}_{\rho,2}(X,\mu)
=
\bigcap_{j,n\geq1}\bigcup_{h\in\mathcal{D}}
\left\{\kappa:\|u_j-h\circ\kappa\|_2<\frac{1}{n}\right\} R ρ , 2 a , iso ( X , μ ) = j , n ≥ 1 ⋂ h ∈ D ⋃ { κ : ∥ u j − h ∘ κ ∥ 2 < n 1 }
is G δ G_\delta G δ . □
Global point-process spaces
For a Polish space Y Y Y and a continuous function Ψ : Y → [ 1 , ∞ ) \Psi:Y\to[1,\infty) Ψ : Y → [ 1 , ∞ ) , let P Ψ ( Y ) \mathcal{P}^{\Psi}(Y) P Ψ ( Y ) be the space of probability measures η \eta η on Y Y Y with ∫ Ψ d η < ∞ \int\Psi\,d\eta<\infty ∫ Ψ d η < ∞ . We equip it with the Ψ \Psi Ψ -weak topology, defined by convergence of the integrals of all continuous functions F : Y → R F:Y\to\mathbb{R} F : Y → R with ∣ F ∣ ≤ C Ψ |F|\leq C\Psi ∣ F ∣ ≤ C Ψ for some C < ∞ C<\infty C < ∞ . The map
η ⟼ Ψ η ∫ Ψ d η \eta\longmapsto\frac{\Psi\eta}{\int\Psi\,d\eta} η ⟼ ∫ Ψ d η Ψ η
is a homeomorphism onto the space of probability measures on Y Y Y with the weak topology; its inverse sends ν \nu ν to Ψ − 1 ν / ∫ Ψ − 1 d ν \Psi^{-1}\nu/\int\Psi^{-1}\,d\nu Ψ − 1 ν / ∫ Ψ − 1 d ν . Thus P Ψ ( Y ) \mathcal{P}^{\Psi}(Y) P Ψ ( Y ) is Polish. Equivalently, η n → η \eta_n\to\eta η n → η in the Ψ \Psi Ψ -weak topology if and only if η n ⇒ η \eta_n\Rightarrow\eta η n ⇒ η and ∫ Ψ d η n → ∫ Ψ d η \int\Psi\,d\eta_n\to\int\Psi\,d\eta ∫ Ψ d η n → ∫ Ψ d η as n → ∞ n\to\infty n → ∞ .
For each m ≥ 1 m\geq1 m ≥ 1 , choose φ m ∈ C c ( R d ) \varphi_m\in C_c(\mathbb{R}^d) φ m ∈ C c ( R d ) such that 0 ≤ φ m ≤ 1 0\leq\varphi_m\leq1 0 ≤ φ m ≤ 1 , φ m = 1 \varphi_m=1 φ m = 1 on Q m Q_m Q m , and supp φ m ⊂ Q m + 1 \operatorname{supp}\varphi_m\subset Q_{m+1} supp φ m ⊂ Q m + 1 . Define
Ψ m ( ω ) : = 1 + S φ m ( ω ) 2 \Psi_m(\omega):=1+S_{\varphi_m}(\omega)^2 Ψ m ( ω ) := 1 + S φ m ( ω ) 2
and abbreviate P Ψ m : = P Ψ m ( Conf ( R d ) ) \mathcal{P}_{\Psi_m}:=\mathcal{P}^{\Psi_m}(\operatorname{Conf}(\mathbb{R}^d)) P Ψ m := P Ψ m ( Conf ( R d )) . Each Ψ m \Psi_m Ψ m is continuous. Let
P l o c , 2 : = ⋂ m ≥ 1 P Ψ m , \mathcal{P}_{\mathrm{loc},2}:=\bigcap_{m\geq1}\mathcal{P}_{\Psi_m}, P loc , 2 := m ≥ 1 ⋂ P Ψ m ,
with the topology induced by the diagonal map into ∏ m ≥ 1 P Ψ m \prod_{m\geq1}\mathcal{P}_{\Psi_m} ∏ m ≥ 1 P Ψ m . Thus η n → η \eta_n\to\eta η n → η in P l o c , 2 \mathcal{P}_{\mathrm{loc},2} P loc , 2 if and only if η n ⇒ η \eta_n\Rightarrow\eta η n ⇒ η and
∫ S φ m 2 d η n ⟶ ∫ S φ m 2 d η ( n → ∞ , m ≥ 1 ) . \int S_{\varphi_m}^2\,d\eta_n\longrightarrow\int S_{\varphi_m}^2\,d\eta\qquad(n\to\infty,\ m\geq1). ∫ S φ m 2 d η n ⟶ ∫ S φ m 2 d η ( n → ∞ , m ≥ 1 ) .
Since N Q m ≤ S φ m ≤ N Q m + 1 N_{Q_m}\leq S_{\varphi_m}\leq N_{Q_{m+1}} N Q m ≤ S φ m ≤ N Q m + 1 , the condition η ∈ P l o c , 2 \eta\in\mathcal{P}_{\mathrm{loc},2} η ∈ P loc , 2 is equivalent to N Q m ∈ L 2 ( η ) N_{Q_m}\in L^2(\eta) N Q m ∈ L 2 ( η ) for every $m\geq 1.
Define P P ρ , 2 \mathrm{PP}_{\rho,2} PP ρ , 2 to be the set of translation-invariant Borel probability measures η \eta η on Conf ( R d ) \operatorname{Conf}(\mathbb{R}^{d}) Conf ( R d ) with intensity ρ \rho ρ and finite local second moments, in the sense of (4) . Equivalently, P P ρ , 2 \mathrm{PP}_{\rho,2} PP ρ , 2 is the translation-invariant intensity-ρ \rho ρ subspace of P l o c , 2 \mathcal{P}_{\mathrm{loc},2} P loc , 2 . We equip it with the subspace topology, called the local L 2 L^{2} L 2 topology. Let E P P ρ , 2 ⊂ P P ρ , 2 \mathrm{EPP}_{\rho,2} \subset\mathrm{PP}_{\rho,2} EPP ρ , 2 ⊂ PP ρ , 2 denote the ergodic elements.
Proposition 2.6. The space P P ρ , 2 \mathrm{PP}_{\rho,2} PP ρ , 2 is Polish for the local L 2 L^{2} L 2 topology, and E P P ρ , 2 \mathrm{EPP}_{\rho,2} EPP ρ , 2 is a G δ G_{\delta} G δ subset, hence Polish. Moreover:
if A ⊂ R d A \subset\mathbb{R}^{d} A ⊂ R d is bounded and Borel, with Vol d ( ∂ A ) = 0 \operatorname{Vol}_{d}(\partial A)=0 Vol d ( ∂ A ) = 0 , then η ↦ ∫ N A 2 d η \eta\mapsto\int N_{A}^{2}\,d\eta η ↦ ∫ N A 2 d η and η ↦ Var η ( N A ) \eta\mapsto\operatorname{Var}_{\eta}(N_{A}) η ↦ Var η ( N A ) are continuous on P P ρ , 2 \mathrm{PP}_{\rho,2} PP ρ , 2 ;
for every a ∈ A d ( X , μ ) a \in A_{d}(X,\mu) a ∈ A d ( X , μ ) , the map κ ↦ κ ∗ μ \kappa\mapsto\kappa_{*}\mu κ ↦ κ ∗ μ from R ρ , 2 a ( X , μ ) R^{a}_{\rho,2}(X,\mu) R ρ , 2 a ( X , μ ) to P P ρ , 2 \mathrm{PP}_{\rho,2} PP ρ , 2 is continuous.
Proof. The diagonal embedding of P l o c , 2 \mathcal{P}_{\mathrm{loc},2} P loc , 2 into ∏ m ≥ 1 P Ψ m \prod_{m\geq1}\mathcal{P}^{\Psi_{m}} ∏ m ≥ 1 P Ψ m has closed image: each Ψ m \Psi_{m} Ψ m -weak topology refines weak convergence, so all coordinates of a limit agree. Thus P l o c , 2 \mathcal{P}_{\mathrm{loc},2} P loc , 2 is Polish.
The set of translation-invariant probability measures is closed: if η n → η \eta_{n}\to\eta η n → η in P l o c , 2 \mathcal{P}_{\mathrm{loc},2} P loc , 2 and each η n \eta_{n} η n is translation-invariant, then for t ∈ R d t\in\mathbb{R}^{d} t ∈ R d and F ∈ C b ( Conf ( R d ) ) F\in C_{b}(\operatorname{Conf}(\mathbb{R}^{d})) F ∈ C b ( Conf ( R d )) , continuity of θ t \theta_{t} θ t gives
∫ F ∘ θ t d η = lim n → ∞ ∫ F ∘ θ t d η n = lim n → ∞ ∫ F d η n = ∫ F d η . \int F\circ\theta_{t}\,d\eta=\lim_{n\to\infty}\int F\circ\theta_{t}\,d\eta_{n}=\lim_{n\to\infty}\int F\,d\eta_{n}=\int F\,d\eta. ∫ F ∘ θ t d η = n → ∞ lim ∫ F ∘ θ t d η n = n → ∞ lim ∫ F d η n = ∫ F d η .
Suppose in addition that every η n \eta_{n} η n has intensity ρ \rho ρ . Fix a bounded Borel set A A A with Vol d ( ∂ A ) = 0 \operatorname{Vol}_{d}(\partial A)=0 Vol d ( ∂ A ) = 0 and choose m m m with A ⊂ Q m A\subset Q_{m} A ⊂ Q m . By translation invariance of η \eta η and Lemma 2.3 , η { ω : N ∂ A ( ω ) = 0 } = 1 \eta\{\omega:N_{\partial A}(\omega)=0\}=1 η { ω : N ∂ A ( ω ) = 0 } = 1 . The continuity-set criterion for vague convergence [8 Appendix A2.6] therefore shows that the distribution of N A N_{A} N A under η n \eta_{n} η n converges weakly on R \mathbb{R} R to its distribution under η \eta η . We also have sup n ∫ S φ m 2 d η n < ∞ \sup_{n}\int S_{\varphi_{m}}^{2}\,d\eta_{n}<\infty sup n ∫ S φ m 2 d η n < ∞ . By Cauchy–Schwarz and the bound sup n ≥ 1 ∫ S φ m 2 d η n < ∞ \sup_{n\geq1}\int S_{\varphi_{m}}^{2}\,d\eta_{n}<\infty sup n ≥ 1 ∫ S φ m 2 d η n < ∞ ,
lim R → ∞ sup n ≥ 1 ∫ S φ m 1 { S φ m > R } d η n = 0. \lim_{R\to\infty}\sup_{n\geq1}\int S_{\varphi_{m}}\mathbf{1}_{\{S_{\varphi_{m}}>R\}}\,d\eta_{n}=0. R → ∞ lim n ≥ 1 sup ∫ S φ m 1 { S φ m > R } d η n = 0.
Since N A ≤ S φ m N_{A}\leq S_{\varphi_{m}} N A ≤ S φ m , the same tail bound holds for N A N_{A} N A . Weak convergence of the distributions of N A N_{A} N A and this bound give
∫ N A d η = lim n → ∞ ∫ N A d η n = ρ Vol d ( A ) . \int N_{A}\,d\eta=\lim_{n\to\infty}\int N_{A}\,d\eta_{n}=\rho\operatorname{Vol}_{d}(A). ∫ N A d η = n → ∞ lim ∫ N A d η n = ρ Vol d ( A ) .
On each Q k Q_{k} Q k , the two first-moment measures are finite and agree on rational half-open rectangles. The monotone-class theorem extends the identity to all bounded Borel sets. Thus P P ρ , 2 \mathrm{PP}_{\rho,2} PP ρ , 2 is closed in P l o c , 2 \mathcal{P}_{\mathrm{loc},2} P loc , 2 and is Polish.
Enumerate the countable algebra D \mathcal{D} D defined in Subsection 2.2 as ( h j ) j ≥ 1 (h_{j})_{j\geq1} ( h j ) j ≥ 1 . For n ≥ 1 n\geq1 n ≥ 1 , put
A n h ( ω ) : = 1 Vol d ( Q n ) ∫ Q n h ( θ t ω ) d t , F j , n ( η ) : = ∫ ∣ A n h j − ∫ h j d η ∣ 2 d η . A_{n}h(\omega):=\frac{1}{\operatorname{Vol}_{d}(Q_{n})}\int_{Q_{n}}h(\theta_{t}\omega)\,dt,\qquad F_{j,n}(\eta):=\int\left|A_{n}h_{j}-\int h_{j}\,d\eta\right|^{2}\,d\eta. A n h ( ω ) := Vol d ( Q n ) 1 ∫ Q n h ( θ t ω ) d t , F j , n ( η ) := ∫ A n h j − ∫ h j d η 2 d η .
For fixed j , n j,n j , n , the function A n h j A_{n}h_{j} A n h j is bounded and continuous, so F j , n F_{j,n} F j , n is weakly continuous in η \eta η . For invariant η \eta η , let P η P_{\eta} P η be the orthogonal projection onto the invariant functions in L 2 ( η ) L^{2}(\eta) L 2 ( η ) . The mean ergodic theorem gives
lim n → ∞ F j , n ( η ) = ∥ P η h j − ∫ h j d η ∥ L 2 ( η ) 2 . \lim_{n\to\infty}F_{j,n}(\eta)=\left\|P_{\eta}h_{j}-\int h_{j}\,d\eta\right\|_{L^{2}(\eta)}^{2}. n → ∞ lim F j , n ( η ) = P η h j − ∫ h j d η L 2 ( η ) 2 .
Since D \mathcal{D} D is dense in L 2 ( η ) L^{2}(\eta) L 2 ( η ) for every η \eta η , this limit vanishes for all j j j exactly when η \eta η is ergodic. Consequently, among invariant probability measures,
{ η : η is ergodic } = ⋂ j , k , N ≥ 1 ⋃ n ≥ N { η : F j , n ( η ) < k − 1 } . \{\eta:\eta\text{ is ergodic}\}=\bigcap_{j,k,N\geq1}\bigcup_{n\geq N}\{\eta:F_{j,n}(\eta)<k^{-1}\}. { η : η is ergodic } = j , k , N ≥ 1 ⋂ n ≥ N ⋃ { η : F j , n ( η ) < k − 1 } .
The set of invariant laws is closed, and the displayed sets are open in the weak topology. Hence the ergodic laws form a G δ G_{\delta} G δ in the invariant-measure space and in the finer local L 2 L^{2} L 2 topology on P P ρ , 2 \mathrm{PP}_{\rho,2} PP ρ , 2 .
Let η n → η \eta_n \to\eta η n → η in P P ρ , 2 \mathrm{PP}_{\rho,2} PP ρ , 2 , let A ⊂ R d A \subset\mathbb{R}^d A ⊂ R d be bounded and Borel, with Vol d ( ∂ A ) = 0 \operatorname{Vol}_d(\partial A)=0 Vol d ( ∂ A ) = 0 , and choose m m m with A ⊂ Q m A \subset Q_m A ⊂ Q m . By Lemma 2.3 and the continuity-set criterion, the distribution of N A N_A N A under η n \eta_n η n converges weakly to its distribution under η \eta η . Moreover,
∫ S φ m 2 d η n ⟶ ∫ S φ m 2 d η . \int S_{\varphi_m}^2 \,d\eta_n \longrightarrow\int S_{\varphi_m}^2 \,d\eta. ∫ S φ m 2 d η n ⟶ ∫ S φ m 2 d η .
Since S φ m 2 S_{\varphi_m}^2 S φ m 2 is continuous and nonnegative, weak convergence and convergence of its expectations imply, by applying weak convergence to min { S φ m 2 , R } \min\{S_{\varphi_m}^2,R\} min { S φ m 2 , R } for each fixed R > 0 R>0 R > 0 , that
lim R → ∞ sup n ∫ ( S φ m 2 − R ) + d η n = 0. \lim_{R\to\infty}\sup_n\int(S_{\varphi_m}^2-R)_+\,d\eta_n=0. R → ∞ lim n sup ∫ ( S φ m 2 − R ) + d η n = 0.
Since u 1 { u > 2 R } ≤ 2 ( u − R ) + u1_{\{u>2R\}}\leq2(u-R)_+ u 1 { u > 2 R } ≤ 2 ( u − R ) + for u ≥ 0 u\geq0 u ≥ 0 , this implies
lim R → ∞ sup n ∫ S φ m 2 1 { S φ m 2 > R } d η n = 0. \lim_{R\to\infty}\sup_n\int S_{\varphi_m}^2 1_{\{S_{\varphi_m}^2>R\}}\,d\eta_n=0. R → ∞ lim n sup ∫ S φ m 2 1 { S φ m 2 > R } d η n = 0.
Since N A 2 ≤ S φ m 2 N_A^2\leq S_{\varphi_m}^2 N A 2 ≤ S φ m 2 , its corresponding tail integrals also tend to zero uniformly in n n n . Weak convergence of the distributions of N A N_A N A therefore yields
∫ N A 2 d η n ⟶ ∫ N A 2 d η . \int N_A^2\,d\eta_n\longrightarrow\int N_A^2\,d\eta. ∫ N A 2 d η n ⟶ ∫ N A 2 d η .
Since every process in P P ρ , 2 \mathrm{PP}_{\rho,2} PP ρ , 2 satisfies ∫ N A d η = ρ Vol d ( A ) \int N_A\,d\eta=\rho\operatorname{Vol}_d(A) ∫ N A d η = ρ Vol d ( A ) , we also have
Var η n ( N A ) = ∫ N A 2 d η n − ρ 2 Vol d ( A ) 2 ⟶ Var η ( N A ) . \operatorname{Var}_{\eta_n}(N_A)=\int N_A^2\,d\eta_n-\rho^2\operatorname{Vol}_d(A)^2\longrightarrow\operatorname{Var}_{\eta}(N_A). Var η n ( N A ) = ∫ N A 2 d η n − ρ 2 Vol d ( A ) 2 ⟶ Var η ( N A ) .
Let κ n → κ \kappa_n\to\kappa κ n → κ in R ρ , 2 a ( X , μ ) \mathbb{R}_{\rho,2}^a(X,\mu) R ρ , 2 a ( X , μ ) . For every F ∈ C b ( Conf ( R d ) ) F\in C_b(\operatorname{Conf}(\mathbb{R}^d)) F ∈ C b ( Conf ( R d )) , F ∘ κ n → F ∘ κ F\circ\kappa_n\to F\circ\kappa F ∘ κ n → F ∘ κ in measure and, because F F F is bounded, in L 2 ( X , μ ) L^2(X,\mu) L 2 ( X , μ ) , so ( κ n ) ∗ μ ⇒ κ ∗ μ (\kappa_n)_*\mu\Rightarrow\kappa_*\mu ( κ n ) ∗ μ ⇒ κ ∗ μ . By Proposition 2.5 , S φ m ∘ κ n → S φ m ∘ κ S_{\varphi_m}\circ\kappa_n\to S_{\varphi_m}\circ\kappa S φ m ∘ κ n → S φ m ∘ κ in L 2 L^2 L 2 for every m m m . Therefore
∫ S φ m 2 d ( ( κ n ) ∗ μ ) ⟶ ∫ S φ m 2 d ( κ ∗ μ ) . \int S_{\varphi_m}^2\,d((\kappa_n)_*\mu)\longrightarrow\int S_{\varphi_m}^2\,d(\kappa_*\mu). ∫ S φ m 2 d (( κ n ) ∗ μ ) ⟶ ∫ S φ m 2 d ( κ ∗ μ ) .
This is convergence of ( κ n ) ∗ μ (\kappa_n)_*\mu ( κ n ) ∗ μ to κ ∗ μ \kappa_*\mu κ ∗ μ in the local L 2 L^2 L 2 topology. □
Proposition 2.7 (Characterization of local L 2 L^2 L 2 convergence). Let η , η 1 , η 2 , … ∈ P P ρ , 2 \eta,\eta_1,\eta_2,\ldots\in\mathrm{PP}_{\rho,2} η , η 1 , η 2 , … ∈ PP ρ , 2 .
η n ⟶ η ⟺ η n ⇒ η and ∫ N Q m 2 d η n ⟶ ∫ N Q m 2 d η ( m ≥ 1 ) . \eta_n\longrightarrow\eta\quad\Longleftrightarrow\quad\eta_n\Rightarrow\eta\ \text{and}\ \int N_{Q_m}^2\,d\eta_n\longrightarrow\int N_{Q_m}^2\,d\eta\quad(m\geq1). η n ⟶ η ⟺ η n ⇒ η and ∫ N Q m 2 d η n ⟶ ∫ N Q m 2 d η ( m ≥ 1 ) . In particular, the topology does not depend on the cutoffs ( φ m ) (\varphi_m) ( φ m ) .
Proof. Assume first that η n → η \eta_n\to\eta η n → η in the local L 2 L^2 L 2 topology. Fix m m m . Weak convergence and translation invariance imply that the distribution of N Q m N_{Q_m} N Q m under η n \eta_n η n converges weakly on R \mathbb{R} R to its distribution under η \eta η , since Vol d ( ∂ Q m ) = 0 \operatorname{Vol}_d(\partial Q_m)=0 Vol d ( ∂ Q m ) = 0 . Choose j j j with Q m ⊂ Q j Q_m\subset Q_j Q m ⊂ Q j and φ j ≡ 1 \varphi_j\equiv1 φ j ≡ 1 on Q m Q_m Q m . Convergence of the S φ j S_{\varphi_j} S φ j -second moments gives
lim R → ∞ sup n ∫ S φ j 2 1 { S φ j 2 > R } d η n = 0. \lim_{R\to\infty}\sup_n\int S_{\varphi_j}^2 1_{\{S_{\varphi_j}^2>R\}}\,d\eta_n=0. R → ∞ lim n sup ∫ S φ j 2 1 { S φ j 2 > R } d η n = 0.
Since N Q m 2 ≤ S φ j 2 N_{Q_m}^2\leq S_{\varphi_j}^2 N Q m 2 ≤ S φ j 2 , the same tail bound holds for N Q m 2 N_{Q_m}^2 N Q m 2 . Hence
∫ N Q m 2 d η n ⟶ ∫ N Q m 2 d η . \int N_{Q_m}^2\,d\eta_n\longrightarrow\int N_{Q_m}^2\,d\eta. ∫ N Q m 2 d η n ⟶ ∫ N Q m 2 d η .
Conversely, suppose that η n ⇒ η \eta_n\Rightarrow\eta η n ⇒ η and that the displayed second-moment convergence holds for every m m m . Since supp ( φ m ) ⊂ Q m + 1 \operatorname{supp}(\varphi_m)\subset Q_{m+1} supp ( φ m ) ⊂ Q m + 1 ,
S φ m 2 ≤ N Q m + 1 2 . S_{\varphi_m}^2\leq N_{Q_{m+1}}^2. S φ m 2 ≤ N Q m + 1 2 .
Convergence of the Q m + 1 Q_{m+1} Q m + 1 -second moments and weak convergence give
lim R → ∞ sup n ∫ N Q m + 1 2 1 { N Q m + 1 2 > R } d η n = 0. \lim_{R\to\infty}\sup_n\int N_{Q_{m+1}}^2\mathbf{1}_{\{N_{Q_{m+1}}^2>R\}}\,d\eta_n=0. R → ∞ lim n sup ∫ N Q m + 1 2 1 { N Q m + 1 2 > R } d η n = 0.
Since S φ m 2 ≤ N Q m + 1 2 S_{\varphi_m}^2\le N_{Q_{m+1}}^2 S φ m 2 ≤ N Q m + 1 2 and S φ m S_{\varphi_m} S φ m is continuous, weak convergence and this tail bound give
∫ S φ m 2 d η n ⟶ ∫ S φ m 2 d η . \int S_{\varphi_m}^2\,d\eta_n\longrightarrow\int S_{\varphi_m}^2\,d\eta. ∫ S φ m 2 d η n ⟶ ∫ S φ m 2 d η .
This is convergence in the local L 2 L^2 L 2 topology. □
For φ ∈ L 1 ( R d ) \varphi\in L^1(\mathbb{R}^d) φ ∈ L 1 ( R d ) , set
φ ^ ( ξ ) : = ∫ R d φ ( x ) e − 2 π i ⟨ x , ξ ⟩ d x . \widehat{\varphi}(\xi):=\int_{\mathbb{R}^d}\varphi(x)e^{-2\pi i\langle x,\xi\rangle}\,dx. φ ( ξ ) := ∫ R d φ ( x ) e − 2 π i ⟨ x , ξ ⟩ d x .
If η ∈ P P ρ , 2 \eta\in\mathrm{PP}_{\rho,2} η ∈ PP ρ , 2 and φ ∈ S ( R d ) \varphi\in\mathcal{S}(\mathbb{R}^d) φ ∈ S ( R d ) , then S φ ∈ L 2 ( η ) S_\varphi\in L^2(\eta) S φ ∈ L 2 ( η ) . Indeed, the unit-cube decomposition gives, for every N > d N>d N > d ,
∥ S φ ∥ L 2 ( η ) ≤ ∥ N Q 1 ∥ L 2 ( η ) ∑ k ∈ Z d sup Q 1 + k ∣ φ ∣ ≤ C N sup x ( 1 + ∥ x ∥ ) N ∣ φ ( x ) ∣ . \|S_\varphi\|_{L^2(\eta)}\le\|N_{Q_1}\|_{L^2(\eta)}\sum_{k\in\mathbb{Z}^d}\sup_{Q_1+k}|\varphi|\le C_N\sup_x(1+\|x\|)^N|\varphi(x)|. ∥ S φ ∥ L 2 ( η ) ≤ ∥ N Q 1 ∥ L 2 ( η ) k ∈ Z d ∑ Q 1 + k sup ∣ φ ∣ ≤ C N x sup ( 1 + ∥ x ∥ ) N ∣ φ ( x ) ∣.
Define
Z φ η : = S φ − ρ ∫ R d φ ( x ) d x . Z_\varphi^\eta:=S_\varphi-\rho\int_{\mathbb{R}^d}\varphi(x)\,dx. Z φ η := S φ − ρ ∫ R d φ ( x ) d x .
Proposition 2.8 (Bartlett spectrum and continuity). For every η ∈ P P ρ , 2 \eta\in\mathrm{PP}_{\rho,2} η ∈ PP ρ , 2 , there is a unique positive symmetric translation-bounded Radon measure σ η \sigma_\eta σ η on R d \mathbb{R}^d R d such that
∫ Z φ η Z ψ η ‾ d η = ∫ R d φ ^ ( ξ ) ψ ^ ( ξ ) d σ η ( ξ ) , φ , ψ ∈ S ( R d ) . \int Z_\varphi^\eta\overline{Z_\psi^\eta}\,d\eta=\int_{\mathbb{R}^d}\widehat{\varphi}(\xi)\widehat{\psi}(\xi)\,d\sigma_\eta(\xi),\qquad\varphi,\psi\in\mathcal{S}(\mathbb{R}^d). ∫ Z φ η Z ψ η d η = ∫ R d φ ( ξ ) ψ ( ξ ) d σ η ( ξ ) , φ , ψ ∈ S ( R d ) . The measure σ η \sigma_\eta σ η is the Bartlett spectrum of η \eta η . If η n → η \eta_n\to\eta η n → η in P P ρ , 2 \mathrm{PP}_{\rho,2} PP ρ , 2 , then
σ η n ⟶ σ η ( n → ∞ ) \sigma_{\eta_n}\longrightarrow\sigma_\eta\qquad(n\to\infty) σ η n ⟶ σ η ( n → ∞ ) vaguely on R d \mathbb{R}^d R d .
Proof. The spectral representation follows from [8 Proposition 8.2.I and Definition 8.2.II] , with angular frequencies ω = 2 π ξ \omega=2\pi\xi ω = 2 π ξ in the notation of that reference. Uniqueness follows because every h ∈ C c ∞ ( R d ) h\in C^\infty_c(\mathbb{R}^d) h ∈ C c ∞ ( R d ) has the form φ ^ ψ ^ \widehat{\varphi}\widehat{\psi} φ ψ for some φ , ψ ∈ S ( R d ) \varphi,\psi\in\mathcal{S}(\mathbb{R}^d) φ , ψ ∈ S ( R d ) : take φ ^ = h \widehat{\varphi}=h φ = h and ψ ^ = 1 \widehat{\psi}=1 ψ = 1 on $\operatorname{supp}h. Since the covariance of real statistics is symmetric, so is σ η \sigma_\eta σ η . For translation boundedness, let χ ∈ C c ∞ ( R d ) \chi\in C^\infty_c(\mathbb{R}^d) χ ∈ C c ∞ ( R d ) satisfy ∣ χ ^ ∣ ≥ c > 0 |\widehat{\chi}|\ge c>0 ∣ χ ∣ ≥ c > 0 on a neighbourhood Q Q Q of the origin. Then
c 2 σ η ( a + Q ) ≤ ∥ Z e 2 π i ⟨ a , ⋅ ⟩ χ η ∥ L 2 2 ≤ ∫ S ∣ χ ∣ 2 d η , a ∈ R d . c^2\sigma_\eta(a+Q)\le\|Z_{e^{2\pi i\langle a,\cdot\rangle}\chi}^\eta\|_{L^2}^2\le\int S_{|\chi|}^2\,d\eta,\qquad a\in\mathbb{R}^d. c 2 σ η ( a + Q ) ≤ ∥ Z e 2 π i ⟨ a , ⋅ ⟩ χ η ∥ L 2 2 ≤ ∫ S ∣ χ ∣ 2 d η , a ∈ R d .
Fix a compact set K ⊂ R d K\subset\mathbb{R}^d K ⊂ R d . Choose a real χ ∈ C c ∞ ( R d ) \chi\in C^\infty_c(\mathbb{R}^d) χ ∈ C c ∞ ( R d ) such that χ ^ \widehat{\chi} χ has no zero on a neighbourhood of K K K , for instance a sufficiently concentrated smooth approximate identity. Set
ν n : = ∣ χ ^ ∣ 2 σ η n , ν : = ∣ χ ^ ∣ 2 σ η . \nu_n:=|\widehat{\chi}|^2\sigma_{\eta_n},\qquad\nu:=|\widehat{\chi}|^2\sigma_\eta. ν n := ∣ χ ∣ 2 σ η n , ν := ∣ χ ∣ 2 σ η .
For t ∈ R d t\in\mathbb{R}^d t ∈ R d , the Bartlett identity gives
∫ e − 2 π i ⟨ t , ξ ⟩ d ν n ( ξ ) = ∫ Z χ η n Z χ ( ⋅ + t ) η n ‾ d η n . \int e^{-2\pi i\langle t,\xi\rangle}\,d\nu_n(\xi)=\int Z_\chi^{\eta_n}\overline{Z_{\chi(\cdot+t)}^{\eta_n}}\,d\eta_n. ∫ e − 2 π i ⟨ t , ξ ⟩ d ν n ( ξ ) = ∫ Z χ η n Z χ ( ⋅+ t ) η n d η n .
Choose M M M so that supp χ ∪ ( supp χ − t ) ⊂ Q M \operatorname{supp}\chi\cup(\operatorname{supp}\chi-t)\subset Q_M supp χ ∪ ( supp χ − t ) ⊂ Q M . The integrand is continuous in the vague topology, and its absolute value is at most C ( 1 + N Q M 2 ) C(1+N_{Q_M}^2) C ( 1 + N Q M 2 ) . By Proposition 2.7 , the laws and second moments of N Q M N_{Q_M} N Q M converge. The resulting uniform tail estimate for N Q M 2 N_{Q_M}^2 N Q M 2 permits passage to the limit in the covariance integral. At t = 0 t=0 t = 0 this also gives ν n ( R d ) → ν ( R d ) \nu_n(\mathbb{R}^d)\to\nu(\mathbb{R}^d) ν n ( R d ) → ν ( R d ) . Lévy’s continuity theorem [Kal02, Theorem 5.3], after normalization when the limiting mass is nonzero, yields weak convergence of the finite measures ν n \nu_n ν n to ν \nu ν on R d \mathbb{R}^d R d .
If h ∈ C c ( R d ) h \in C_c(\mathbb{R}^d) h ∈ C c ( R d ) is supported in K K K , then
∫ h d σ η n = ∫ h ∣ χ ^ ∣ 2 d ν n ⟶ ∫ h ∣ χ ^ ∣ 2 d ν = ∫ h d σ η . \int h\,d\sigma_{\eta_n}=\int\frac{h}{|\widehat{\chi}|^2}\,d\nu_n\longrightarrow\int\frac{h}{|\widehat{\chi}|^2}\,d\nu=\int h\,d\sigma_\eta. ∫ h d σ η n = ∫ ∣ χ ∣ 2 h d ν n ⟶ ∫ ∣ χ ∣ 2 h d ν = ∫ h d σ η .
Hence σ η n → σ η \sigma_{\eta_n}\to\sigma_\eta σ η n → σ η vaguely. □
Lemma 2.9 (Bartlett identity for regular windows). Let η ∈ P P ρ , 2 \eta\in\mathrm{PP}_{\rho,2} η ∈ PP ρ , 2 , and let A ⊂ R d A \subset\mathbb{R}^d A ⊂ R d be bounded and Borel, with Vol d ( ∂ A ) = 0 \operatorname{Vol}_d(\partial A)=0 Vol d ( ∂ A ) = 0 . Then
Var η ( N A ) = ∫ R d ∣ 1 A ^ ( ξ ) ∣ 2 d σ η ( ξ ) . \operatorname{Var}_\eta(N_A)=\int_{\mathbb{R}^d}|\widehat{\mathbf{1}_A}(\xi)|^2\,d\sigma_\eta(\xi). Var η ( N A ) = ∫ R d ∣ 1 A ( ξ ) ∣ 2 d σ η ( ξ ) .
Proof. Let φ n ∈ C c ∞ ( R d ) \varphi_n \in C_c^\infty(\mathbb{R}^d) φ n ∈ C c ∞ ( R d ) be uniformly bounded and supported in a fixed compact set, with φ n → 1 A \varphi_n \to\mathbf{1}_A φ n → 1 A pointwise off ∂ A \partial A ∂ A and in L 1 L^1 L 1 . By Lemma 2.3 , S φ n → N A S_{\varphi_n}\to N_A S φ n → N A almost surely under η \eta η ; domination by a fixed local count gives convergence in L 2 ( η ) L^2(\eta) L 2 ( η ) . By the Bartlett identity, ( φ n ^ ) (\widehat{\varphi_n}) ( φ n ) is Cauchy in L 2 ( σ η ) L^2(\sigma_\eta) L 2 ( σ η ) . Since φ n ^ ( ξ ) → 1 A ^ ( ξ ) \widehat{\varphi_n}(\xi)\to\widehat{\mathbf{1}_A}(\xi) φ n ( ξ ) → 1 A ( ξ ) for every ξ \xi ξ , its L 2 ( σ η ) L^2(\sigma_\eta) L 2 ( σ η ) -limit is 1 A ^ \widehat{\mathbf{1}_A} 1 A . Passing to the limit in the variance identity gives the result. □
For every η ∈ P P ρ , 2 \eta\in\mathrm{PP}_{\rho,2} η ∈ PP ρ , 2 , the measure σ η \sigma_\eta σ η has positive mass on some compact annulus. Otherwise Z ∂ j φ η = 0 Z^\eta_{\partial_j\varphi}=0 Z ∂ j φ η = 0 almost surely for every φ ∈ C c ∞ ( R d ) \varphi\in C_c^\infty(\mathbb{R}^d) φ ∈ C c ∞ ( R d ) and 1 ≤ j ≤ d 1\le j\le d 1 ≤ j ≤ d . Taking a countable dense family gives ∂ j ω = 0 \partial_j\omega=0 ∂ j ω = 0 distributionally for almost every ω \omega ω . Thus ω \omega ω is a constant multiple of Lebesgue measure, contrary to simplicity and positive intensity. For 0 < a < b 0<a<b 0 < a < b , the Bessel asymptotic gives, uniformly for a ≤ ∥ ξ ∥ ≤ b a\le\|\xi\|\le b a ≤ ∥ ξ ∥ ≤ b ,
1 T ∫ T 2 T R 1 − d ∣ 1 B R ^ ( ξ ) ∣ 2 d R = 1 2 π 2 ∥ ξ ∥ d + 1 + O a , b , d ( T − 1 ) . \frac{1}{T}\int_T^{2T}R^{1-d}|\widehat{\mathbf{1}_{B_R}}(\xi)|^2\,dR=\frac{1}{2\pi^2\|\xi\|^{d+1}}+O_{a,b,d}(T^{-1}). T 1 ∫ T 2 T R 1 − d ∣ 1 B R ( ξ ) ∣ 2 d R = 2 π 2 ∥ ξ ∥ d + 1 1 + O a , b , d ( T − 1 ) .
Choose such an annulus of positive σ η \sigma_\eta σ η -mass. The Bartlett identity and Tonelli’s theorem give (1) ; cf. [4 Theorem 5.1] .
The local-to-global principle
Let E P P ρ , 2 f r \mathrm{EPP}^{\mathrm{fr}}_{\rho,2} EPP ρ , 2 fr consist of laws η ∈ E P P ρ , 2 \eta\in\mathrm{EPP}_{\rho,2} η ∈ EPP ρ , 2 for which the translation action R d ↷ ( Conf ( R d ) , η ) \mathbb{R}^d\curvearrowright(\operatorname{Conf}(\mathbb{R}^d),\eta) R d ↷ ( Conf ( R d ) , η ) is essentially free.
Proposition 2.10. The set E P P ρ , 2 f r \mathrm{EPP}^{\mathrm{fr}}_{\rho,2} EPP ρ , 2 fr is a dense G δ G_\delta G δ subset of E P P ρ , 2 \mathrm{EPP}_{\rho,2} EPP ρ , 2 .
Proof. For m ≥ 1 m\ge1 m ≥ 1 , let
K m : = { t ∈ R d : m − 1 ≤ ∥ t ∥ ≤ m } , h m ( ω ) : = min t ∈ K m d Conf ( θ t ω , ω ) . K_m:=\{t\in\mathbb{R}^d:m^{-1}\le\|t\|\le m\},\qquad h_m(\omega):=\min_{t\in K_m}d_{\operatorname{Conf}}(\theta_t\omega,\omega). K m := { t ∈ R d : m − 1 ≤ ∥ t ∥ ≤ m } , h m ( ω ) := t ∈ K m min d Conf ( θ t ω , ω ) .
Joint continuity of the translation action and compactness of K m K_m K m make h m h_m h m continuous, and ω \omega ω has trivial stabilizer exactly when h m ( ω ) > 0 h_m(\omega)>0 h m ( ω ) > 0 for every m m m . Thus, with C m : = { h m = 0 } C_m:=\{h_m=0\} C m := { h m = 0 } ,
E P P ρ , 2 f r = ⋂ m , k ≥ 1 { η ∈ E P P ρ , 2 : η ( C m ) < k − 1 } . \mathrm{EPP}^{\mathrm{fr}}_{\rho,2}=\bigcap_{m,k\ge1}\{\eta\in\mathrm{EPP}_{\rho,2}:\eta(C_m)<k^{-1}\}. EPP ρ , 2 fr = m , k ≥ 1 ⋂ { η ∈ EPP ρ , 2 : η ( C m ) < k − 1 } .
The sets C m C_m C m are closed, so η ↦ η ( C m ) \eta\mapsto\eta(C_m) η ↦ η ( C m ) is upper semicontinuous for weak convergence. Each strict sublevel set in the preceding intersection is therefore open.
Fix η ∈ E P P ρ , 2 \eta\in\mathrm{EPP}_{\rho,2} η ∈ EPP ρ , 2 and a probability measure υ ≪ Vol d \upsilon\ll\operatorname{Vol}_d υ ≪ Vol d supported on B 1 B_1 B 1 . The relation { ( ω , p ) : p ∈ supp ω } \{(\omega,p):p\in\operatorname{supp}\omega\} {( ω , p ) : p ∈ supp ω } is Borel: point incidence is equivalent to N B 1 / n ( θ p ω ) > 0 N_{B_{1/n}}(\theta_p\omega)>0 N B 1/ n ( θ p ω ) > 0 for all n ≥ 1 n\ge1 n ≥ 1 . These conditions are Borel by joint continuity of translation and Borel measurability of N B 1 / n N_{B_{1/n}} N B 1/ n . Its sections are countable by local finiteness. Lusin–Novikov [24 Theorem 18.10] gives partial Borel maps with disjoint graphs covering this relation. For each ω \omega ω , list their defined values in increasing index order as p 1 ( ω ) , p 2 ( ω ) , … p_1(\omega), p_2(\omega), \ldots p 1 ( ω ) , p 2 ( ω ) , … . The domains and values of these reindexed maps are Borel; the indices range from 1 1 1 to # supp ω \#\operatorname{supp}\omega # supp ω when this number is finite. On the product space Conf ( R d ) × B 1 N \operatorname{Conf}(\mathbb{R}^{d}) \times B_1^{\mathbb{N}} Conf ( R d ) × B 1 N with measure η ⊗ ν ⊗ N \eta\otimes\nu^{\otimes\mathbb{N}} η ⊗ ν ⊗ N , define
Φ ε ( ω , u ) : = ∑ j ≤ # supp ω δ p j ( ω ) + ε u j , η ε : = ( Φ ε ) ∗ ( η ⊗ ν ⊗ N ) . \Phi_\varepsilon(\omega,\mathbf{u}) := \sum_{j \le\#\operatorname{supp}\omega} \delta_{p_j(\omega)+\varepsilon u_j}, \qquad\eta_\varepsilon:= (\Phi_\varepsilon)_*(\eta\otimes\nu^{\otimes\mathbb{N}}). Φ ε ( ω , u ) := j ≤ # supp ω ∑ δ p j ( ω ) + ε u j , η ε := ( Φ ε ) ∗ ( η ⊗ ν ⊗ N ) .
This independent marking construction is Borel; see [29 Definition 5.3] . The resulting measure η ε \eta_\varepsilon η ε does not depend on the chosen enumeration. Translating the original configuration permutes its points and therefore their independent identically distributed displacements. Hence η ε \eta_\varepsilon η ε is translation-invariant. For φ ∈ C c ( R d ) \varphi\in C_c(\mathbb{R}^{d}) φ ∈ C c ( R d ) , the intensity identity gives
∫ S φ d η ε = ∫ ∑ p ∈ supp ω ∫ B 1 φ ( p + ε u ) d ν ( u ) d η ( ω ) = ρ ∫ φ d Vol d . \int S_\varphi\,d\eta_\varepsilon= \int\sum_{p\in\operatorname{supp}\omega} \int_{B_1} \varphi(p+\varepsilon u)\,d\nu(u)\,d\eta(\omega) = \rho\int\varphi\,d\operatorname{Vol}_d. ∫ S φ d η ε = ∫ p ∈ supp ω ∑ ∫ B 1 φ ( p + ε u ) d ν ( u ) d η ( ω ) = ρ ∫ φ d Vol d .
Independent marking preserves ergodicity. Let F , G F,G F , G be bounded observables of the displaced configuration on bounded regions. For H = F , G H=F,G H = F , G , set
H ‾ ( ω ) : = ∫ H ( Φ ε ( ω , u ) ) d ν ⊗ N ( u ) . \overline{H}(\omega) := \int H(\Phi_\varepsilon(\omega,\mathbf{u}))\,d\nu^{\otimes\mathbb{N}}(\mathbf{u}). H ( ω ) := ∫ H ( Φ ε ( ω , u )) d ν ⊗ N ( u ) .
For sufficiently large ∣ t ∣ |t| ∣ t ∣ , the marks affecting F F F and G ∘ θ t G \circ\theta_t G ∘ θ t belong to disjoint sets of original points. Conditional independence gives
∫ F ( G ∘ θ t ) d η ε = ∫ F ‾ ( G ‾ ∘ θ t ) d η . \int F(G\circ\theta_t)\,d\eta_\varepsilon= \int\overline{F}(\overline{G}\circ\theta_t)\,d\eta. ∫ F ( G ∘ θ t ) d η ε = ∫ F ( G ∘ θ t ) d η .
Averaging over t ∈ Q T t\in Q_T t ∈ Q T and applying the mean ergodic theorem to η \eta η shows that the averages of these correlations converge, as T → ∞ T\to\infty T → ∞ , to ( ∫ F d η ε ) ( ∫ G d η ε ) \left(\int F\,d\eta_\varepsilon\right)\left(\int G\,d\eta_\varepsilon\right) ( ∫ F d η ε ) ( ∫ G d η ε ) . Since bounded local functions generate the configuration σ \sigma σ -algebra, η ε \eta_\varepsilon η ε is ergodic.
For distinct p , q p,q p , q in any fixed configuration, absolute continuity of ν \nu ν gives
∫ B 1 2 1 { p + ε u = q + ε v } d ν ( u ) d ν ( v ) = 0. \int_{B_1^2} \mathbf{1}_{\{p+\varepsilon u=q+\varepsilon v\}}\,d\nu(u)\,d\nu(v)=0. ∫ B 1 2 1 { p + ε u = q + ε v } d ν ( u ) d ν ( v ) = 0.
For three distinct points p , q , r p,q,r p , q , r , likewise,
∫ B 1 3 1 { p + ε u + r + ε w = 2 q + 2 ε v } d ν ( u ) d ν ( v ) d ν ( w ) = 0. \int_{B_1^3} \mathbf{1}_{\{p+\varepsilon u+r+\varepsilon w=2q+2\varepsilon v\}}\,d\nu(u)\,d\nu(v)\,d\nu(w)=0. ∫ B 1 3 1 { p + ε u + r + εw = 2 q + 2 ε v } d ν ( u ) d ν ( v ) d ν ( w ) = 0.
There are only countably many such pairs and triples. Thus Φ ε ( ω , u ) \Phi_\varepsilon(\omega,\mathbf{u}) Φ ε ( ω , u ) is almost surely simple and has no three distinct points in arithmetic progression. The empty configuration is invariant under every translation. Since η ε \eta_\varepsilon η ε is ergodic and has positive intensity, it gives zero mass to the empty configuration. If a nonempty set P P P has period t ≠ 0 t\ne0 t = 0 , then x − t , x , x + t ∈ P x-t,x,x+t\in P x − t , x , x + t ∈ P for any x ∈ P x\in P x ∈ P , contradicting the absence of three-term progressions. Hence η ε \eta_\varepsilon η ε is essentially free.
For 0 < ε < 1 / 2 0<\varepsilon<1/2 0 < ε < 1/2 and every m ≥ 1 m\ge1 m ≥ 1 ,
S φ m ( Φ ε ( ω , u ) ) 2 ≤ N Q m + 2 ( ω ) 2 . S_{\varphi_m}(\Phi_\varepsilon(\omega,\mathbf{u}))^2 \le N_{Q_{m+2}}(\omega)^2. S φ m ( Φ ε ( ω , u ) ) 2 ≤ N Q m + 2 ( ω ) 2 .
The right-hand side is η \eta η -integrable, so η ε ∈ E P P ρ , 2 f r \eta_\varepsilon\in\mathbf{EPP}_{\rho,2}^{\mathrm{fr}} η ε ∈ EPP ρ , 2 fr . For each ω \omega ω and u \mathbf{u} u , the configurations Φ ε ( ω , u ) \Phi_\varepsilon(\omega,\mathbf{u}) Φ ε ( ω , u ) converge vaguely to ω \omega ω as ε → 0 \varepsilon\to0 ε → 0 . Dominated convergence gives η ε ⇒ η \eta_\varepsilon\Rightarrow\eta η ε ⇒ η and convergence of the S φ m S_{\varphi_m} S φ m -second moments for every m m m . Thus η ε → η \eta_\varepsilon\to\eta η ε → η in the local L 2 L^2 L 2 topology. □
Proposition 2.11 (Local-to-global principle). Let A ⊂ E P P ρ , 2 \mathcal{A}\subset\mathbf{EPP}_{\rho,2} A ⊂ EPP ρ , 2 be G δ G_\delta G δ . Suppose that for every a ∈ A d ( X , μ ) a\in A_d(X,\mu) a ∈ A d ( X , μ ) the set
{ κ ∈ R ρ , 2 a , i s o ( X , μ ) : κ ∗ μ ∈ A } \left\{\kappa\in\mathbb{R}_{\rho,2}^{a,\mathrm{iso}}(X,\mu):\kappa_*\mu\in\mathcal{A}\right\} { κ ∈ R ρ , 2 a , iso ( X , μ ) : κ ∗ μ ∈ A } is dense in R ρ , 2 a , i s o ( X , μ ) \mathbb{R}_{\rho,2}^{a,\mathrm{iso}}(X,\mu) R ρ , 2 a , iso ( X , μ ) . Then A \mathcal{A} A is residual in E P P ρ , 2 \mathbf{EPP}_{\rho,2} EPP ρ , 2 .
Proof. Let U ⊂ E P P ρ , 2 U \subset\mathrm{EPP}_{\rho,2} U ⊂ EPP ρ , 2 be nonempty and open. By Proposition 2.10, U U U contains some η ∈ E P P ρ , 2 f r \eta\in\mathrm{EPP}_{\rho,2}^{\mathrm{fr}} η ∈ EPP ρ , 2 fr . The probability space ( Conf ( R d ) , η ) (\operatorname{Conf}(\mathbb{R}^{d}),\eta) ( Conf ( R d ) , η ) is nonatomic. Indeed, since the action is essentially free, the translates of any atom would give uncountably many distinct atoms of the same positive mass. By [24 Theorem 17.41] , choose a measure-preserving Borel isomorphism modulo null sets
ψ : ( X , μ ) ⟶ ( Conf ( R d ) , η ) . \psi\colon(X,\mu) \longrightarrow(\operatorname{Conf}(\mathbb{R}^{d}),\eta). ψ : ( X , μ ) ⟶ ( Conf ( R d ) , η ) .
By shrinking the conull Borel sets, we may assume that their complements are uncountable Borel null sets. Extend ψ \psi ψ to a Borel bijection using a Borel isomorphism between these complements [24 Theorem 15.6] . Define
a ( t , x ) : = ψ − 1 ( θ t ψ ( x ) ) , ( t , x ) ∈ R d × X . a(t,x) := \psi^{-1}(\theta_t\psi(x)), \qquad(t,x) \in\mathbb{R}^{d} \times X. a ( t , x ) := ψ − 1 ( θ t ψ ( x )) , ( t , x ) ∈ R d × X .
Since ψ \psi ψ preserves measure and intertwines a a a with θ \theta θ , this action is p.m.p., ergodic, and essentially free. The map κ 0 : = ψ \kappa_0 := \psi κ 0 := ψ is equivariant and generating, with ( κ 0 ) ∗ μ = η (\kappa_0)_*\mu= \eta ( κ 0 ) ∗ μ = η .
Continuity of the push-forward map makes
{ κ ∈ R ρ , 2 a , i s o ( X , μ ) : κ ∗ μ ∈ U } \{\kappa\in\mathbb{R}_{\rho,2}^{a,\mathrm{iso}}(X,\mu) : \kappa_*\mu\in U\} { κ ∈ R ρ , 2 a , iso ( X , μ ) : κ ∗ μ ∈ U }
an open neighbourhood of κ 0 \kappa_0 κ 0 . By the density hypothesis it contains some κ \kappa κ with κ ∗ μ ∈ A \kappa_*\mu\in\mathcal{A} κ ∗ μ ∈ A . Hence U ∩ A ≠ ∅ U \cap\mathcal{A} \ne\varnothing U ∩ A = ∅ , so A \mathcal{A} A is dense. Since A \mathcal{A} A is G δ G_\delta G δ in the Polish space E P P ρ , 2 \mathrm{EPP}_{\rho,2} EPP ρ , 2 , it is residual. □
Intermittent hyperuniformity
In this section we prove that, for every essentially free ergodic action, realizations with small or large number variance at prescribed scales are dense in the local L 2 L^2 L 2 topology. For small variance, we replace points inside large cells so that their polynomial moments agree with Lebesgue measure. For large variance, we move points into smaller subcubes while preserving the number of points in each cell. We then obtain generating realizations by encoding the underlying action with sparse point triples.
Baire reduction
In this subsection we express intermittent hyperuniformity as a countable intersection of open variance conditions, reducing its residuality to two density assertions. Recall V η ( R ) V_\eta(R) V η ( R ) , H y p ρ \mathrm{Hyp}_\rho Hyp ρ , and I H y p ρ \mathrm{IHyp}_\rho IHyp ρ from Subsection 1.1. For a ∈ A d ( X , μ ) a \in A_d(X,\mu) a ∈ A d ( X , μ ) and κ ∈ R ρ , 2 a ( X , μ ) \kappa\in\mathbb{R}_{\rho,2}^{a}(X,\mu) κ ∈ R ρ , 2 a ( X , μ ) , set
V κ ( R ) : = V κ ∗ μ ( R ) . V_\kappa(R) := V_{\kappa_*\mu}(R). V κ ( R ) := V κ ∗ μ ( R ) .
For η ∈ E P P ρ , 2 \eta\in\mathrm{EPP}_{\rho,2} η ∈ EPP ρ , 2 and β > ( d − 1 ) / d \beta> (d-1)/d β > ( d − 1 ) / d , (2) holds if and only if, for every ε > 0 \varepsilon> 0 ε > 0 , there are arbitrarily large R R R with
V η ( R ) < ε Vol d ( B R ) β . V_\eta(R) < \varepsilon\operatorname{Vol}_d(B_R)^\beta. V η ( R ) < ε Vol d ( B R ) β .
For α < 2 \alpha< 2 α < 2 , (3) holds if and only if, for every M > 0 M > 0 M > 0 , there are arbitrarily large R R R with
V η ( R ) > M Vol d ( B R ) α . V_\eta(R) > M\operatorname{Vol}_d(B_R)^\alpha. V η ( R ) > M Vol d ( B R ) α .
Set
Q − : = Q ∩ ( − ∞ , 2 ) , Q + : = Q ∩ ( d − 1 d , ∞ ) . \mathbb{Q}_{-} := \mathbb{Q} \cap(-\infty,2), \qquad\mathbb{Q}_{+} := \mathbb{Q} \cap\left(\frac{d-1}{d},\infty\right). Q − := Q ∩ ( − ∞ , 2 ) , Q + := Q ∩ ( d d − 1 , ∞ ) .
Proposition 3.1. The set I H y p ρ \mathrm{IHyp}_\rho IHyp ρ is G δ G_\delta G δ in E P P ρ , 2 \mathrm{EPP}_{\rho,2} EPP ρ , 2 . For every a ∈ A d ( X , μ ) a \in A_d(X,\mu) a ∈ A d ( X , μ ) , the set
{ κ ∈ R ρ , 2 a ( X , μ ) : κ ∗ μ ∈ I H y p ρ } \{\kappa\in\mathbb{R}_{\rho,2}^{a}(X,\mu) : \kappa_*\mu\in\mathrm{IHyp}_\rho\} { κ ∈ R ρ , 2 a ( X , μ ) : κ ∗ μ ∈ IHyp ρ } is G δ G_\delta G δ in R ρ , 2 a ( X , μ ) \mathbb{R}_{\rho,2}^{a}(X,\mu) R ρ , 2 a ( X , μ ) .
Proof. Fix η ∈ P P ρ , 2 \eta\in\mathop{\mathrm{PP}}\nolimits_{\rho,2} η ∈ PP ρ , 2 . The map R ↦ V η ( R ) R \mapsto V_\eta(R) R ↦ V η ( R ) is continuous. Indeed, if R n → R R_n \to R R n → R , then N B R n → N B R N_{B_{R_n}} \to N_{B_R} N B R n → N B R η \eta η -almost surely by Lemma 2.3 . For all sufficiently large n n n , N B R n ≤ N B R + 1 N_{B_{R_n}} \le N_{B_{R+1}} N B R n ≤ N B R + 1 . Since η ∈ P P ρ , 2 \eta\in\mathop{\mathrm{PP}}\nolimits_{\rho,2} η ∈ PP ρ , 2 , we have N B R + 1 ∈ L 2 ( η ) N_{B_{R+1}} \in L^2(\eta) N B R + 1 ∈ L 2 ( η ) , so the convergence also holds in L 2 ( η ) L^2(\eta) L 2 ( η ) .
Hence rational radii suffice:
IHyp ρ = ⋂ β ∈ Q + j , N ≥ 1 ⋃ R ∈ Q R > N { η ∈ E P P ρ , 2 : V η ( R ) < j − 1 Vol d ( B R ) β } ∩ ⋂ α ∈ Q − j , N ≥ 1 ⋃ R ∈ Q R > N { η ∈ E P P ρ , 2 : V η ( R ) > j Vol d ( B R ) α } . \operatorname{IHyp}_{\rho}
=
\bigcap_{\substack{\beta\in\mathbb{Q}_{+}\\ j,N\ge1}}
\bigcup_{\substack{R\in\mathbb{Q}\\ R>N}}
\left\{\eta\in\mathop{\mathrm{EPP}}\nolimits_{\rho,2}:V_\eta(R)<j^{-1}\operatorname{Vol}_{d}(B_R)^\beta\right\}
\cap
\bigcap_{\substack{\alpha\in\mathbb{Q}_{-}\\ j,N\ge1}}
\bigcup_{\substack{R\in\mathbb{Q}\\ R>N}}
\left\{\eta\in\mathop{\mathrm{EPP}}\nolimits_{\rho,2}:V_\eta(R)>j\operatorname{Vol}_{d}(B_R)^\alpha\right\}. IHyp ρ = β ∈ Q + j , N ≥ 1 ⋂ R ∈ Q R > N ⋃ { η ∈ EPP ρ , 2 : V η ( R ) < j − 1 Vol d ( B R ) β } ∩ α ∈ Q − j , N ≥ 1 ⋂ R ∈ Q R > N ⋃ { η ∈ EPP ρ , 2 : V η ( R ) > j Vol d ( B R ) α } .
For fixed R > 0 R>0 R > 0 , the map η ↦ V η ( R ) \eta\mapsto V_\eta(R) η ↦ V η ( R ) is continuous by Proposition 2.6 . This proves the first assertion, and continuity of the push-forward map gives the second. □ \square □
For β > ( d − 1 ) / d \beta>(d-1)/d β > ( d − 1 ) / d , α < 2 \alpha<2 α < 2 , and j , N ≥ 1 j,N\ge1 j , N ≥ 1 , define
L β , j , N a : = ⋃ n ≥ N { κ ∈ R ρ , 2 a ( X , μ ) : V κ ( n ) < j − 1 Vol d ( B n ) β } . \mathcal{L}^{a}_{\beta,j,N}:=\bigcup_{n\ge N}\left\{\kappa\in\mathcal{R}^{a}_{\rho,2}(X,\mu):V_\kappa(n)<j^{-1}\operatorname{Vol}_{d}(B_n)^\beta\right\}. L β , j , N a := n ≥ N ⋃ { κ ∈ R ρ , 2 a ( X , μ ) : V κ ( n ) < j − 1 Vol d ( B n ) β } .
and
H α , j , N a : = ⋃ n ≥ N { κ ∈ R ρ , 2 a ( X , μ ) : V κ ( n ) > j Vol d ( B n ) α } . \mathcal{H}^{a}_{\alpha,j,N}:=\bigcup_{n\ge N}\left\{\kappa\in\mathcal{R}^{a}_{\rho,2}(X,\mu):V_\kappa(n)>j\operatorname{Vol}_{d}(B_n)^\alpha\right\}. H α , j , N a := n ≥ N ⋃ { κ ∈ R ρ , 2 a ( X , μ ) : V κ ( n ) > j Vol d ( B n ) α } .
These sets are open by Proposition 2.5 .
Proposition 3.2 (Baire reduction). Suppose that, for every a ∈ A d ( X , μ ) a\in A_d(X,\mu) a ∈ A d ( X , μ ) , β > ( d − 1 ) / d \beta>(d-1)/d β > ( d − 1 ) / d , α < 2 \alpha<2 α < 2 , and j , N ≥ 1 j,N\ge1 j , N ≥ 1 , the sets L β , j , N a \mathcal{L}^{a}_{\beta,j,N} L β , j , N a and H α , j , N a \mathcal{H}^{a}_{\alpha,j,N} H α , j , N a are dense in R ρ , 2 a ( X , μ ) \mathcal{R}^{a}_{\rho,2}(X,\mu) R ρ , 2 a ( X , μ ) , and that R ρ , 2 a , i s o ( X , μ ) \mathcal{R}^{a,\mathrm{iso}}_{\rho,2}(X,\mu) R ρ , 2 a , iso ( X , μ ) is dense there. Then
{ κ ∈ R ρ , 2 a , i s o ( X , μ ) : κ ∗ μ ∈ IHyp ρ } \left\{\kappa\in\mathcal{R}^{a,\mathrm{iso}}_{\rho,2}(X,\mu):\kappa_*\mu\in\operatorname{IHyp}_{\rho}\right\} { κ ∈ R ρ , 2 a , iso ( X , μ ) : κ ∗ μ ∈ IHyp ρ } is residual in R ρ , 2 a , i s o ( X , μ ) \mathcal{R}^{a,\mathrm{iso}}_{\rho,2}(X,\mu) R ρ , 2 a , iso ( X , μ ) for every a ∈ A d ( X , μ ) a\in A_d(X,\mu) a ∈ A d ( X , μ ) , and IHyp ρ \operatorname{IHyp}_{\rho} IHyp ρ is residual in E P P ρ , 2 \mathop{\mathrm{EPP}}\nolimits_{\rho,2} EPP ρ , 2 .
Proof. Let G a G_a G a be the intersection of R ρ , 2 a , i s o ( X , μ ) \mathcal{R}^{a,\mathrm{iso}}_{\rho,2}(X,\mu) R ρ , 2 a , iso ( X , μ ) with the sets L β , j , N a \mathcal{L}^{a}_{\beta,j,N} L β , j , N a and H α , j , N a \mathcal{H}^{a}_{\alpha,j,N} H α , j , N a over j , N ≥ 1 j,N\ge1 j , N ≥ 1 , β ∈ Q + \beta\in\mathbb{Q}_{+} β ∈ Q + , and α ∈ Q − \alpha\in\mathbb{Q}_{-} α ∈ Q − . The hypotheses and Proposition 2.5 show that G a G_a G a is a dense G δ G_\delta G δ subset of R ρ , 2 a ( X , μ ) \mathcal{R}^{a}_{\rho,2}(X,\mu) R ρ , 2 a ( X , μ ) .
For κ ∈ G a \kappa\in G_a κ ∈ G a and β > ( d − 1 ) / d \beta>(d-1)/d β > ( d − 1 ) / d , choose β ′ ∈ Q + \beta'\in\mathbb{Q}_{+} β ′ ∈ Q + with β ′ < β \beta'<\beta β ′ < β ; for α < 2 \alpha<2 α < 2 , choose α ′ ∈ Q − \alpha'\in\mathbb{Q}_{-} α ′ ∈ Q − with α < α ′ \alpha<\alpha' α < α ′ . Since Vol d ( B n ) → ∞ \operatorname{Vol}_{d}(B_n)\to\infty Vol d ( B n ) → ∞ as n → ∞ n\to\infty n → ∞ , the corresponding integer-radius inequalities imply κ ∗ μ ∈ IHyp ρ \kappa_*\mu\in\operatorname{IHyp}_{\rho} κ ∗ μ ∈ IHyp ρ . The pullback of IHyp ρ \operatorname{IHyp}_{\rho} IHyp ρ is therefore dense in every generating realization space and is G δ G_\delta G δ there by Proposition 3.1 . The local-to-global principle Proposition 2.11 yields the global conclusion. □ \square □
Low-variance approximation
In this subsection we prove that low-variance realizations are dense. The construction modifies a point configuration separately in the interior of each of many small cubes. In each cube, we retain the points outside a prescribed interior region and replace the points inside that region by a finite set. The union of the retained and replacement points has the prescribed count and polynomial moments on each small cube. These identities will allow us to apply Lemma 3.6.
Write I : = [ 0 , 1 ) d I:=[0,1)^d I := [ 0 , 1 ) d . For T > 0 T>0 T > 0 , a T T T -cell is a half-open cube v + T I v+TI v + T I , and a T T T -grid is the partition { v + T k + T I : k ∈ Z d } \{v+Tk+TI:k\in\mathbb{Z}^d\} { v + T k + T I : k ∈ Z d } for a fixed v ∈ R d v\in\mathbb{R}^d v ∈ R d . In particular, an L L L -cell has side length L L L ; each L L L -grid below has a specified translation.
Proposition 3.3 (Low-variance density). For every a ∈ A d ( X , μ ) a\in A_d(X,\mu) a ∈ A d ( X , μ ) , β > ( d − 1 ) / d \beta>(d-1)/d β > ( d − 1 ) / d , and j , N ≥ 1 j,N\ge1 j , N ≥ 1 , the set L β , j , N a \mathcal{L}^{a}_{\beta,j,N} L β , j , N a is dense in R ρ , 2 a ( X , μ ) \mathcal{R}^{a}_{\rho,2}(X,\mu) R ρ , 2 a ( X , μ ) .
Ergodic approximation of cell moments
We first estimate the number of points in large sets. Ergodicity gives L 2 L^2 L 2 convergence of scaled point counts, also when the points are weighted by a continuous function. We also need to control the expected number of points in a surrounding cube on the event that a prescribed weighted count deviates from its limit. This second estimate will be used to bound the points in cells where the replacement construction cannot be applied.
Lemma 3.4. Let a ∈ A d ( X , μ ) a \in A_d(X,\mu) a ∈ A d ( X , μ ) and κ ∈ R ρ , 2 a ( X , μ ) \kappa\in\mathbb{R}^{a}_{\rho,2}(X,\mu) κ ∈ R ρ , 2 a ( X , μ ) . Let A ⊂ R d A \subset\mathbb{R}^d A ⊂ R d be a bounded Borel set satisfying Vol d ( ∂ r A ) = O ( r ) \operatorname{Vol}_d(\partial_r A)=O(r) Vol d ( ∂ r A ) = O ( r ) as r ↓ 0 r \downarrow0 r ↓ 0 , and let g ∈ C ( A ‾ ) g \in C(\overline{A}) g ∈ C ( A ) . For s > 0 s>0 s > 0 , put
Σ s ( x ) : = s − d ∑ u ∈ supp κ ( x ) ∩ s A g ( u / s ) . (6) \Sigma_s(x):=s^{-d}\sum_{u\in\operatorname{supp}\kappa(x)\cap sA}g(u/s). \tag*{(6)} Σ s ( x ) := s − d u ∈ supp κ ( x ) ∩ s A ∑ g ( u / s ) . (6) Then
Σ s ⟶ ρ ∫ A g ( u ) d u in L 2 ( X , μ ) ( s → ∞ ) . (7) \Sigma_s \longrightarrow\rho\int_A g(u)\,du \quad\text{in }L^2(X,\mu)\quad(s\to\infty). \tag*{(7)} Σ s ⟶ ρ ∫ A g ( u ) d u in L 2 ( X , μ ) ( s → ∞ ) . (7) Moreover, for every bounded cube B ⊃ A B\supset A B ⊃ A and every ε > 0 \varepsilon>0 ε > 0 ,
s − d ∫ { x : ∣ Σ s ( x ) − ρ ∫ A g ∣ > ε } N s B ( κ ( x ) ) d μ ( x ) ⟶ 0 ( s → ∞ ) . (8) s^{-d}\int_{\{x:|\Sigma_s(x)-\rho\int_A g|>\varepsilon\}}N_{sB}(\kappa(x))\,d\mu(x)\longrightarrow0\quad(s\to\infty). \tag*{(8)} s − d ∫ { x : ∣ Σ s ( x ) − ρ ∫ A g ∣ > ε } N s B ( κ ( x )) d μ ( x ) ⟶ 0 ( s → ∞ ) . (8)
Proof. Fix a a a and κ \kappa κ as in the statement. We first consider a half-open cube Q ⊂ R d Q\subset\mathbb{R}^d Q ⊂ R d and the unweighted count N s Q ∘ κ N_{sQ}\circ\kappa N s Q ∘ κ . Choose a nonnegative φ ∈ C c ( R d ) \varphi\in C_c(\mathbb{R}^d) φ ∈ C c ( R d ) with integral one, and let r > 0 r>0 r > 0 satisfy supp φ ⊂ B r \operatorname{supp}\varphi\subset B_r supp φ ⊂ B r . Set H ( x ) : = ∑ u ∈ supp κ ( x ) φ ( u ) H(x):=\sum_{u\in\operatorname{supp}\kappa(x)}\varphi(u) H ( x ) := ∑ u ∈ supp κ ( x ) φ ( u ) . Equivariance gives
H ( t . x ) = ∑ u ∈ supp κ ( x ) φ ( u − t ) . (9) H(t.x)=\sum_{u\in\operatorname{supp}\kappa(x)}\varphi(u-t). \tag*{(9)} H ( t . x ) = u ∈ supp κ ( x ) ∑ φ ( u − t ) . (9)
The mean ergodic theorem and ∫ H d μ = ρ \int H\,d\mu=\rho ∫ H d μ = ρ give
1 Vol d ( s Q ) ∫ s Q H ( t . x ) d t ⟶ ρ in L 2 ( X , μ ) . (10) \frac{1}{\operatorname{Vol}_d(sQ)}\int_{sQ}H(t.x)\,dt\longrightarrow\rho\quad\text{in }L^2(X,\mu). \tag*{(10)} Vol d ( s Q ) 1 ∫ s Q H ( t . x ) d t ⟶ ρ in L 2 ( X , μ ) . (10)
On the other hand, Fubini and (9) give
∫ s Q H ( t . x ) d t = ∑ u ∈ supp κ ( x ) ( 1 s Q ∗ φ ) ( u ) . (11) \int_{sQ}H(t.x)\,dt=\sum_{u\in\operatorname{supp}\kappa(x)}(\mathbf{1}_{sQ}*\varphi)(u). \tag*{(11)} ∫ s Q H ( t . x ) d t = u ∈ supp κ ( x ) ∑ ( 1 s Q ∗ φ ) ( u ) . (11)
The functions 1 s Q ∗ φ \mathbf{1}_{sQ}*\varphi 1 s Q ∗ φ and 1 s Q \mathbf{1}_{sQ} 1 s Q coincide outside ∂ r ( s Q ) \partial_r(sQ) ∂ r ( s Q ) , and their difference has absolute value at most one. For s ≥ 1 s\geq1 s ≥ 1 , this boundary neighbourhood intersects at most C Q s d − 1 C_Qs^{d-1} C Q s d − 1 cubes of the unit grid. Since κ ∗ μ \kappa_*\mu κ ∗ μ is translation-invariant and has finite local second moments, the triangle inequality gives
∥ ∫ s Q H ( t . ⋅ ) d t − N s Q ∘ κ ∥ L 2 ( μ ) ≤ C Q s d − 1 ∥ N [ 0 , 1 ) d ∘ κ ∥ L 2 ( μ ) . (12) \left\|\int_{sQ}H(t.\mathord{\,\cdot\,})\,dt-N_{sQ}\circ\kappa\right\|_{L^2(\mu)}
\leq C_Qs^{d-1}\left\|N_{[0,1)^d}\circ\kappa\right\|_{L^2(\mu)}. \tag*{(12)} ∫ s Q H ( t . ⋅ ) d t − N s Q ∘ κ L 2 ( μ ) ≤ C Q s d − 1 N [ 0 , 1 ) d ∘ κ L 2 ( μ ) . (12)
Dividing (12) by s d s^d s d and using (10) yields
s − d N s Q ∘ κ ⟶ ρ Vol d ( Q ) in L 2 ( μ ) . (13) s^{-d}N_{sQ}\circ\kappa\longrightarrow\rho\operatorname{Vol}_d(Q)\quad\text{in }L^2(\mu). \tag*{(13)} s − d N s Q ∘ κ ⟶ ρ Vol d ( Q ) in L 2 ( μ ) . (13)
To pass from cube counts to the weighted statistic Σ s \Sigma_s Σ s , approximate g 1 A g\mathbf{1}_A g 1 A by step functions on a grid. We regard g 1 A g\mathbf{1}_A g 1 A as a function on R d \mathbb{R}^d R d , equal to zero outside A A A . For h > 0 h>0 h > 0 , let J h \mathcal{J}_h J h consist of the half-open cubes J = h ( k + I ) J=h(k+I) J = h ( k + I ) , k ∈ Z d k\in\mathbb{Z}^d k ∈ Z d , that are contained in A A A . This family is finite. Choose u J ∈ J u_J\in J u J ∈ J for each J ∈ J h J\in\mathcal{J}_h J ∈ J h , and put
ψ h : = ∑ J ∈ J h g ( u J ) 1 J , Σ s , h ( x ) : = s − d ∑ u ∈ supp κ ( x ) ψ h ( u / s ) . \psi_h:=\sum_{J\in\mathcal{J}_h}g(u_J)\mathbf{1}_J,\qquad
\Sigma_{s,h}(x):=s^{-d}\sum_{u\in\operatorname{supp}\kappa(x)}\psi_h(u/s). ψ h := J ∈ J h ∑ g ( u J ) 1 J , Σ s , h ( x ) := s − d u ∈ supp κ ( x ) ∑ ψ h ( u / s ) .
Our aim is to compare Σ s \Sigma_s Σ s in (6) with Σ s , h \Sigma_{s,h} Σ s , h , apply (13) with h h h fixed, and then let h ↓ 0 h\downarrow0 h ↓ 0 .
Define
ω g ( δ ) : = sup { ∣ g ( u ) − g ( v ) ∣ : u , v ∈ A ‾ , ∥ u − v ∥ ≤ δ } , U h : = A ∩ ∂ d h A . \omega_g(\delta) := \sup\{|g(u)-g(v)|:u,v\in\overline{A},\ \|u-v\|\le\delta\},\qquad U_h:=A\cap\partial_{\sqrt{d}h}A. ω g ( δ ) := sup { ∣ g ( u ) − g ( v ) ∣ : u , v ∈ A , ∥ u − v ∥ ≤ δ } , U h := A ∩ ∂ d h A .
Each cube of side length h h h has diameter d h \sqrt{d}\,h d h . Consequently, if u ∈ A ∖ U h u\in A\setminus U_h u ∈ A ∖ U h , its grid cube is contained in A A A : otherwise the segment from u u u to a point of the cube outside A A A would intersect ∂ A \partial A ∂ A at distance at most d h \sqrt{d}\,h d h from u u u . If J J J is this cube, then ∥ u − u J ∥ ≤ d h \|u-u_J\|\le\sqrt{d}\,h ∥ u − u J ∥ ≤ d h , so ∣ g ( u ) − ψ h ( u ) ∣ ≤ ω g ( d h ) |g(u)-\psi_h(u)|\le\omega_g(\sqrt{d}\,h) ∣ g ( u ) − ψ h ( u ) ∣ ≤ ω g ( d h ) . For u ∈ U h u\in U_h u ∈ U h we have ∣ g ( u ) − ψ h ( u ) ∣ ≤ 2 ∥ g ∥ ∞ |g(u)-\psi_h(u)|\le2\|g\|_\infty ∣ g ( u ) − ψ h ( u ) ∣ ≤ 2∥ g ∥ ∞ , and both functions vanish outside A A A .
Set M 2 : = ∥ N I ∘ κ ∥ L 2 ( μ ) M_2:=\|N_I\circ\kappa\|_{L^2(\mu)} M 2 := ∥ N I ∘ κ ∥ L 2 ( μ ) . If E ⊂ R d E\subset\mathbb{R}^d E ⊂ R d is bounded and Borel, the unit cubes intersecting E E E are disjoint and lie in E + [ − 1 , 1 ] d E+[-1,1]^d E + [ − 1 , 1 ] d . Translation invariance of κ ∗ μ \kappa_*\mu κ ∗ μ and the triangle inequality thus give
∥ N E ∘ κ ∥ L 2 ( μ ) ≤ M 2 Vol d ( E + [ − 1 , 1 ] d ) . (14) \|N_E\circ\kappa\|_{L^2(\mu)}\le M_2\operatorname{Vol}_d(E+[-1,1]^d). \tag*{(14)} ∥ N E ∘ κ ∥ L 2 ( μ ) ≤ M 2 Vol d ( E + [ − 1 , 1 ] d ) . (14)
For fixed h > 0 h>0 h > 0 and s → ∞ s\to\infty s → ∞ , this bound applied to s A sA s A and s U h sU_h s U h yields
lim sup s → ∞ s − d ∥ N s A ∘ κ ∥ 2 ≤ M 2 Vol d ( A ) , \limsup_{s\to\infty}s^{-d}\|N_{sA}\circ\kappa\|_2\le M_2\operatorname{Vol}_d(A), s → ∞ lim sup s − d ∥ N s A ∘ κ ∥ 2 ≤ M 2 Vol d ( A ) ,
lim sup s → ∞ s − d ∥ N s U h ∘ κ ∥ 2 ≤ M 2 Vol d ( ∂ 2 d h A ) . \limsup_{s\to\infty}s^{-d}\|N_{sU_h}\circ\kappa\|_2\le M_2\operatorname{Vol}_d(\partial_{2\sqrt{d}h}A). s → ∞ lim sup s − d ∥ N s U h ∘ κ ∥ 2 ≤ M 2 Vol d ( ∂ 2 d h A ) .
Indeed, for all sufficiently large s s s (depending on h h h ), U h + [ − 1 / s , 1 / s ] d ⊂ ∂ 2 d h A U_h+[-1/s,1/s]^d\subset\partial_{2\sqrt{d}h}A U h + [ − 1/ s , 1/ s ] d ⊂ ∂ 2 d h A . The pointwise bounds above therefore imply
lim sup s → ∞ ∥ Σ s − Σ s , h ∥ L 2 ( μ ) ≤ M 2 ( ω g ( d h ) Vol d ( A ) + 2 ∥ g ∥ ∞ Vol d ( ∂ 2 d h A ) ) . (15) \limsup_{s\to\infty}\|\Sigma_s-\Sigma_{s,h}\|_{L^2(\mu)}
\le M_2\left(\omega_g(\sqrt{d}h)\operatorname{Vol}_d(A)+2\|g\|_\infty\operatorname{Vol}_d(\partial_{2\sqrt{d}h}A)\right). \tag*{(15)} s → ∞ lim sup ∥ Σ s − Σ s , h ∥ L 2 ( μ ) ≤ M 2 ( ω g ( d h ) Vol d ( A ) + 2∥ g ∥ ∞ Vol d ( ∂ 2 d h A ) ) . (15)
For every fixed h > 0 h>0 h > 0 , ψ h \psi_h ψ h is a finite linear combination of indicators of half-open cubes. Hence (13) gives
Σ s , h ⟶ ρ ∫ R d ψ h ( u ) d u in L 2 ( μ ) ( s → ∞ ) . (16) \Sigma_{s,h}\longrightarrow\rho\int_{\mathbb{R}^d}\psi_h(u)\,du\quad\text{in }L^2(\mu)\quad(s\to\infty). \tag*{(16)} Σ s , h ⟶ ρ ∫ R d ψ h ( u ) d u in L 2 ( μ ) ( s → ∞ ) . (16)
Combining the preceding two equations with the triangle inequality produces, for every fixed h > 0 h>0 h > 0 ,
lim sup s → ∞ ∥ Σ s − ρ ∫ A g ∥ L 2 ( μ ) ≤ M 2 ( ω g ( d h ) Vol d ( A ) + 2 ∥ g ∥ ∞ Vol d ( ∂ 2 d h A ) ) + ρ ∣ ∫ R d ψ h − ∫ A g ∣ . (17) \limsup_{s\to\infty}\left\|\Sigma_s-\rho\int_A g\right\|_{L^2(\mu)}
\le M_2\left(\omega_g(\sqrt{d}h)\operatorname{Vol}_d(A)+2\|g\|_\infty\operatorname{Vol}_d(\partial_{2\sqrt{d}h}A)\right)
+\rho\left|\int_{\mathbb{R}^d}\psi_h-\int_A g\right|. \tag*{(17)} s → ∞ lim sup Σ s − ρ ∫ A g L 2 ( μ ) ≤ M 2 ( ω g ( d h ) Vol d ( A ) + 2∥ g ∥ ∞ Vol d ( ∂ 2 d h A ) ) + ρ ∫ R d ψ h − ∫ A g . (17)
The right-hand side tends to zero as h ↓ 0 h\downarrow0 h ↓ 0 : uniform continuity gives ω g ( d h ) → 0 \omega_g(\sqrt{d}h)\to0 ω g ( d h ) → 0 , the hypothesis on ∂ r A \partial_rA ∂ r A controls the boundary term, and the pointwise bounds above imply ∥ ψ h − g 1 A ∥ L 1 → 0 \|\psi_h-g1_A\|_{L^1}\to0 ∥ ψ h − g 1 A ∥ L 1 → 0 . Thus the left-hand side of (17) , which does not depend on h h h , is zero. This proves (7) .
Fix a cube B ⊃ A B\supset A B ⊃ A and ε > 0 \varepsilon>0 ε > 0 , and set
E s : = { x : ∣ Σ s ( x ) − ρ ∫ A g ∣ > ε } , Y s : = s − d N s B ∘ κ . E_s:=\left\{x:\left|\Sigma_s(x)-\rho\int_A g\right|>\varepsilon\right\},\qquad Y_s:=s^{-d}N_{sB}\circ\kappa. E s := { x : Σ s ( x ) − ρ ∫ A g > ε } , Y s := s − d N s B ∘ κ .
Convergence (7) implies μ ( E s ) → 0 \mu(E_s)\to0 μ ( E s ) → 0 . By (13) , Y s → ρ Vol d ( B ) Y_s\to\rho\operatorname{Vol}_d(B) Y s → ρ Vol d ( B ) in L 1 ( μ ) L^1(\mu) L 1 ( μ ) ; if B B B is not half-open, its boundary has zero expected count, so the choice of boundary convention does not affect Y s Y_s Y s almost surely. Since Y s ≥ 0 Y_s\ge0 Y s ≥ 0 ,
∫ E s Y s d μ ≤ ∥ Y s − ρ Vol d ( B ) ∥ L 1 ( μ ) + ρ Vol d ( B ) μ ( E s ) ⟶ 0 , \int_{E_s}Y_s\,d\mu\le\|Y_s-\rho\operatorname{Vol}_d(B)\|_{L^1(\mu)}+\rho\operatorname{Vol}_d(B)\mu(E_s)\longrightarrow0, ∫ E s Y s d μ ≤ ∥ Y s − ρ Vol d ( B ) ∥ L 1 ( μ ) + ρ Vol d ( B ) μ ( E s ) ⟶ 0 ,
which proves (8) . □
For a finite set of pairs ( A i , g i ) (A_i,g_i) ( A i , g i ) , 1 ≤ i ≤ n 1 \le i \le n 1 ≤ i ≤ n , each satisfying the hypotheses of Lemma 3.4 , fix ε i > 0 \varepsilon_i > 0 ε i > 0 and a bounded cube B B B containing all A i A_i A i . Define
E s , i : = { x : s − d ∑ u ∈ supp κ ( x ) ∩ s A i g i ( u / s ) − ρ ∫ A i g i > ε i } . E_{s,i}:=\left\{x:s^{-d}\sum_{u\in\operatorname{supp}\kappa(x)\cap sA_i}g_i(u/s)-\rho\int_{A_i}g_i>\varepsilon_i\right\}. E s , i := ⎩ ⎨ ⎧ x : s − d u ∈ supp κ ( x ) ∩ s A i ∑ g i ( u / s ) − ρ ∫ A i g i > ε i ⎭ ⎬ ⎫ .
Applying (7) and (8) to each of the n n n pairs gives μ ( ⋃ i = 1 n E s , i ) → 0 \mu\left(\bigcup_{i=1}^{n}E_{s,i}\right)\to0 μ ( ⋃ i = 1 n E s , i ) → 0 and
s − d ∫ ⋃ i = 1 n E s , i N s B ∘ κ d μ ≤ ∑ i = 1 n s − d ∫ E s , i N s B ∘ κ d μ ⟶ 0. (18) s^{-d}\int_{\bigcup_{i=1}^{n}E_{s,i}}N_{sB}\circ\kappa\,d\mu\leq\sum_{i=1}^{n}s^{-d}\int_{E_{s,i}}N_{sB}\circ\kappa\,d\mu\longrightarrow0. \tag*{(18)} s − d ∫ ⋃ i = 1 n E s , i N s B ∘ κ d μ ≤ i = 1 ∑ n s − d ∫ E s , i N s B ∘ κ d μ ⟶ 0. (18)
Completion of polynomial moments
Fix a closed cube D ⊂ ( 0 , 1 ) d D\subset(0,1)^d D ⊂ ( 0 , 1 ) d . Given a finite set of points F ⊂ I ∖ int D F\subset I\setminus\operatorname{int}D F ⊂ I ∖ int D , we seek a finite set A ⊂ int D A\subset\operatorname{int}D A ⊂ int D such that the union F ∪ A F\cup A F ∪ A has a prescribed number of points and its polynomial sums equal the corresponding integrals over I I I . Thus the points of F F F remain fixed; only the points of A A A are to be determined. The proposition gives a Borel choice of A A A when the count and polynomial sums of F F F are sufficiently close to their prescribed values. It also gives a lower bound on the distance between points of A A A and on their distance to ∂ D \partial D ∂ D .
For a positive integer p p p , let P p − 1 \mathcal{P}_{p-1} P p − 1 be the space of real polynomials on R d \mathbb{R}^d R d of degree less than p p p , and set
P p − 1 0 : = { P ∈ P p − 1 : P ( 0 ) = 0 } . \mathcal{P}^{0}_{p-1}:=\{P\in\mathcal{P}_{p-1}:P(0)=0\}. P p − 1 0 := { P ∈ P p − 1 : P ( 0 ) = 0 } .
For a finite set A ⊂ R d A\subset\mathbb{R}^d A ⊂ R d , write
sep ( A ) : = min { ∥ u − v ∥ : u , v ∈ A , u ≠ v } , \operatorname{sep}(A):=\min\{\lVert u-v\rVert:u,v\in A,\ u\ne v\}, sep ( A ) := min {∥ u − v ∥ : u , v ∈ A , u = v } ,
with sep ( A ) : = + ∞ \operatorname{sep}(A):=+\infty sep ( A ) := + ∞ when # A ≤ 1 \#A\leq1 # A ≤ 1 .
Proposition 3.5 (Completion of polynomial moments). Fix a closed cube D ⊂ ( 0 , 1 ) d D\subset(0,1)^d D ⊂ ( 0 , 1 ) d , a positive integer p p p , and a basis P 1 , … , P m P_1,\ldots,P_m P 1 , … , P m of P p − 1 0 \mathcal{P}^{0}_{p-1} P p − 1 0 . Put ϑ : = Vol d ( D ) \vartheta:=\operatorname{Vol}_d(D) ϑ := Vol d ( D ) and P : = ( P 1 , … , P m ) \mathbf{P}:=(P_1,\ldots,P_m) P := ( P 1 , … , P m ) ; for p = 1 p=1 p = 1 , the basis is empty. There exist ε 0 > 0 \varepsilon_0>0 ε 0 > 0 and q 0 ∈ N q_0\in\mathbb{N} q 0 ∈ N , depending on D D D , p p p , P \mathbf{P} P , and σ d > 0 \sigma_d>0 σ d > 0 , depending only on d d d , with the following property.
Fix an integer q ≥ q 0 q\geq q_0 q ≥ q 0 and 0 < ε ≤ ε 0 0<\varepsilon\leq\varepsilon_0 0 < ε ≤ ε 0 . Let F q , ε \mathcal{F}_{q,\varepsilon} F q , ε consist of the finite sets F ⊂ I ∖ int D F\subset I\setminus\operatorname{int}D F ⊂ I ∖ int D satisfying
∣ # F − ( 1 − ϑ ) q ∣ ≤ ε q and ∣ ∑ u ∈ F P i ( u ) − q ∫ I ∖ D P i ( u ) d u ∣ ≤ ε q ( 1 ≤ i ≤ m ) . (19) \left|\#F-(1-\vartheta)q\right|\leq\varepsilon q\quad\text{and}\quad\left|\sum_{u\in F}P_i(u)-q\int_{I\setminus D}P_i(u)\,du\right|\leq\varepsilon q\quad(1\leq i\leq m). \tag*{(19)} ∣ # F − ( 1 − ϑ ) q ∣ ≤ εq and u ∈ F ∑ P i ( u ) − q ∫ I ∖ D P i ( u ) d u ≤ εq ( 1 ≤ i ≤ m ) . (19) Then F q , ε \mathcal{F}_{q,\varepsilon} F q , ε is Borel in Conf ( R d ) \operatorname{Conf}(\mathbb{R}^d) Conf ( R d ) , and there is a Borel map
F ⟼ δ A ( q , F ) , F q , ε ⟶ Conf ( R d ) , F\longmapsto\delta_{A(q,F)},\qquad\mathcal{F}_{q,\varepsilon}\longrightarrow\operatorname{Conf}(\mathbb{R}^d), F ⟼ δ A ( q , F ) , F q , ε ⟶ Conf ( R d ) , with A ( q , F ) ⊂ int D A(q,F)\subset\operatorname{int}D A ( q , F ) ⊂ int D , such that
∑ u ∈ F ∪ A ( q , F ) P ( u ) = q ∫ I P ( u ) d u ( P ∈ P p − 1 ) , (20) \sum_{u\in F\cup A(q,F)}P(u)=q\int_I P(u)\,du\qquad(P\in\mathcal{P}_{p-1}), \tag*{(20)} u ∈ F ∪ A ( q , F ) ∑ P ( u ) = q ∫ I P ( u ) d u ( P ∈ P p − 1 ) , (20) and
sep ( A ( q , F ) ) ≥ σ d q − 1 / d , dist ( A ( q , F ) , ∂ D ) ≥ σ d q − 1 / d . (21) \operatorname{sep}(A(q,F))\geq\sigma_dq^{-1/d},\qquad\operatorname{dist}(A(q,F),\partial D)\geq\sigma_dq^{-1/d}. \tag*{(21)} sep ( A ( q , F )) ≥ σ d q − 1/ d , dist ( A ( q , F ) , ∂ D ) ≥ σ d q − 1/ d . (21)
Remark. For each fixed 0 < ε ≤ ε 0 0<\varepsilon\leq\varepsilon_0 0 < ε ≤ ε 0 , the family F q , ε \mathcal{F}_{q,\varepsilon} F q , ε is nonempty for all sufficiently large q q q . Indeed, let h = q − 1 / d h=q^{-1/d} h = q − 1/ d and take points of a translate of h Z d h\mathbb{Z}^d h Z d in I ∖ D I\setminus D I ∖ D whose distance from D D D is at least h / 4 h/4 h /4 . Riemann-sum estimates on the cube boundaries give an error O ( q 1 − 1 / d ) O(q^{1-1/d}) O ( q 1 − 1/ d ) in both the point count and each polynomial sum in (19) ; for fixed ε \varepsilon ε , this is at most ε q \varepsilon q εq when q q q is large. This nonemptiness assertion does not hold uniformly for all 0 < ε ≤ ε 0 0<\varepsilon\leq\varepsilon_0 0 < ε ≤ ε 0 at a fixed q q q .
Proof. Fix q ≥ q 0 q \ge q_0 q ≥ q 0 , 0 < ε ≤ ε 0 0 < \varepsilon\le\varepsilon_0 0 < ε ≤ ε 0 , and F ∈ F q , ε F \in\mathcal{F}_{q,\varepsilon} F ∈ F q , ε . Put k : = q − # F k := q - \#F k := q − # F . We seek A ⊂ int D A \subset\operatorname{int} D A ⊂ int D with exactly k k k points and, when m ≥ 1 m \ge1 m ≥ 1 , with
∑ u ∈ A P ( u ) = q ∫ I P ( u ) d u − ∑ u ∈ F P ( u ) . (22) \sum_{u\in A} \mathbf{P}(u) = q \int_I \mathbf{P}(u)\,du - \sum_{u\in F} \mathbf{P}(u). \tag*{(22)} u ∈ A ∑ P ( u ) = q ∫ I P ( u ) d u − u ∈ F ∑ P ( u ) . (22)
The count bound in (19) gives
∣ k − ϑ q ∣ ≤ ε q . |k-\vartheta q| \le\varepsilon q. ∣ k − ϑq ∣ ≤ εq .
We take ε 0 ≤ ϑ / 2 \varepsilon_0 \le\vartheta/2 ε 0 ≤ ϑ /2 , so that ϑ q / 2 ≤ k ≤ 3 ϑ q / 2 \vartheta q/2 \le k \le3\vartheta q/2 ϑq /2 ≤ k ≤ 3 ϑq /2 . In particular, k → ∞ k \to\infty k → ∞ uniformly over the sets F F F under consideration as q → ∞ q \to\infty q → ∞ .
There are constants c d , C d > 0 c_d, C_d > 0 c d , C d > 0 and K d ∈ N K_d \in\mathbb{N} K d ∈ N , depending only on d d d , such that for k ≥ K d k \ge K_d k ≥ K d the cube D D D can be partitioned into k k k axis-parallel rectangles of equal volume ϑ / k \vartheta/k ϑ / k , with every side length between 2 c d ( ϑ / k ) 1 / d 2c_d(\vartheta/k)^{1/d} 2 c d ( ϑ / k ) 1/ d and C d ( ϑ / k ) 1 / d C_d(\vartheta/k)^{1/d} C d ( ϑ / k ) 1/ d . Increase q 0 q_0 q 0 so that k ≥ K d k \ge K_d k ≥ K d . We construct these partitions inductively on d d d . For d = 1 d=1 d = 1 , use k k k equal intervals. For d ≥ 2 d \ge2 d ≥ 2 , divide the first coordinate into n : = ⌈ k 1 / d ⌉ n := \lceil k^{1/d}\rceil n := ⌈ k 1/ d ⌉ intervals. Choose integers k 1 , … , k n k_1,\ldots,k_n k 1 , … , k n in { ⌊ k / n ⌋ , ⌈ k / n ⌉ } \{\lfloor k/n\rfloor,\lceil k/n\rceil\} {⌊ k / n ⌋ , ⌈ k / n ⌉} with sum k k k , in a fixed order. Write ℓ : = ϑ 1 / d \ell:= \vartheta^{1/d} ℓ := ϑ 1/ d for the side length of D D D . Partition its first coordinate interval into consecutive intervals of lengths ℓ k i / k \ell k_i/k ℓ k i / k , 1 ≤ i ≤ n 1 \le i \le n 1 ≤ i ≤ n . Over the i i i th interval, partition the ( d − 1 ) (d-1) ( d − 1 ) -dimensional cube into k i k_i k i equal-volume rectangles by induction. The stated side-length bounds follow from k i ≍ k ( d − 1 ) / d k_i \asymp k^{(d-1)/d} k i ≍ k ( d − 1 ) / d , with constants depending only on d d d . Fix one such partition for each k k k , and let A k 0 A_k^0 A k 0 consist of its k k k rectangle centres. Then
sep ( A k 0 ) , dist ( A k 0 , ∂ D ) ≥ c d ( ϑ / k ) 1 / d ≥ c d ( 2 / 3 ) 1 / d q − 1 / d . \operatorname{sep}(A_k^0),\quad\operatorname{dist}(A_k^0,\partial D) \ge c_d(\vartheta/k)^{1/d} \ge c_d(2/3)^{1/d}q^{-1/d}. sep ( A k 0 ) , dist ( A k 0 , ∂ D ) ≥ c d ( ϑ / k ) 1/ d ≥ c d ( 2/3 ) 1/ d q − 1/ d .
For every f ∈ C 1 ( D ) f \in C^1(D) f ∈ C 1 ( D ) , integrating the difference between f f f at the centre and at each point of its rectangle yields
∣ 1 k ∑ u ∈ A k 0 f ( u ) − 1 ϑ ∫ D f ( u ) d u ∣ ≤ C d ( ϑ / k ) 1 / d ∥ ∇ f ∥ L ∞ ( D ) . (23) \left|\frac{1}{k}\sum_{u\in A_k^0} f(u)-\frac{1}{\vartheta}\int_D f(u)\,du\right|
\le C_d(\vartheta/k)^{1/d}\|\nabla f\|_{L^\infty(D)}. \tag*{(23)} k 1 u ∈ A k 0 ∑ f ( u ) − ϑ 1 ∫ D f ( u ) d u ≤ C d ( ϑ / k ) 1/ d ∥∇ f ∥ L ∞ ( D ) . (23)
If p = 1 p=1 p = 1 , the only polynomial condition concerns constants. Set A ( q , F ) : = A k 0 A(q,F):=A_k^0 A ( q , F ) := A k 0 . It has k k k points and satisfies the required separation bounds. The choice is Borel because the fixed set A k 0 A_k^0 A k 0 depends only on the integer k = q − # F k=q-\#F k = q − # F . For the rest of the proof, assume p ≥ 2 p\ge2 p ≥ 2 .
Dividing (22) by k k k , the average of P \mathbf{P} P over A A A must equal
y ( F ) : = 1 k ( q ∫ I P ( u ) d u − ∑ u ∈ F P ( u ) ) . y(F):=\frac{1}{k}\left(q\int_I \mathbf{P}(u)\,du-\sum_{u\in F}\mathbf{P}(u)\right). y ( F ) := k 1 ( q ∫ I P ( u ) d u − u ∈ F ∑ P ( u ) ) .
By splitting the integral over I I I into D D D and I ∖ D I\setminus D I ∖ D , we have
y ( F ) − 1 ϑ ∫ D P ( u ) d u = ( q k − 1 ϑ ) ∫ D P ( u ) d u + 1 k ( q ∫ I ∖ D P ( u ) d u − ∑ u ∈ F P ( u ) ) . y(F)-\frac{1}{\vartheta}\int_D \mathbf{P}(u)\,du
=
\left(\frac{q}{k}-\frac{1}{\vartheta}\right)\int_D \mathbf{P}(u)\,du
+\frac{1}{k}\left(q\int_{I\setminus D}\mathbf{P}(u)\,du-\sum_{u\in F}\mathbf{P}(u)\right). y ( F ) − ϑ 1 ∫ D P ( u ) d u = ( k q − ϑ 1 ) ∫ D P ( u ) d u + k 1 ( q ∫ I ∖ D P ( u ) d u − u ∈ F ∑ P ( u ) ) .
The bounds in (19) and k ≥ ϑ q / 2 k\ge\vartheta q/2 k ≥ ϑq /2 therefore imply
∥ y ( F ) − 1 ϑ ∫ D P ( u ) d u ∥ ≤ C ε . \left\|y(F)-\frac{1}{\vartheta}\int_D\mathbf{P}(u)\,du\right\|\le C\varepsilon. y ( F ) − ϑ 1 ∫ D P ( u ) d u ≤ C ε .
Here and below, constants may depend on D D D , p p p , and the fixed polynomial basis, but not on q q q , ε \varepsilon ε , or F F F .
To correct the polynomial sums of A k 0 A_k^0 A k 0 , we vary its points through an m m m -parameter family of maps and choose the directions so that the derivative of the moment map is invertible. Choose a nonnegative function χ ∈ C c ∞ ( int D ) \chi\in C^\infty_c(\operatorname{int} D) χ ∈ C c ∞ ( int D ) positive on a nonempty open set. For 1 ≤ j ≤ m 1 \le j \le m 1 ≤ j ≤ m , put X j ( u ) : = χ ( u ) ∇ P j ( u ) X_j(u) := \chi(u)\nabla P_j(u) X j ( u ) := χ ( u ) ∇ P j ( u ) , and define
Φ t ( u ) : = u + ∑ j = 1 m t j X j ( u ) ( t ∈ R m , u ∈ R d ) . \Phi_t(u) := u + \sum_{j=1}^{m} t_j X_j(u) \qquad(t \in\mathbb{R}^m,\ u \in\mathbb{R}^d). Φ t ( u ) := u + j = 1 ∑ m t j X j ( u ) ( t ∈ R m , u ∈ R d ) .
We shall apply Φ t \Phi_t Φ t to every point of A k 0 A_k^0 A k 0 . The derivative at t = 0 t=0 t = 0 of the vector ϑ − 1 ∫ D P ( Φ t ( u ) ) d u \vartheta^{-1}\int_D \mathbf{P}(\Phi_t(u))\,du ϑ − 1 ∫ D P ( Φ t ( u )) d u is the matrix
J i j : = 1 ϑ ∫ D χ ( u ) ∇ P i ( u ) ⋅ ∇ P j ( u ) d u . J_{ij} := \frac{1}{\vartheta}\int_D \chi(u)\nabla P_i(u)\cdot\nabla P_j(u)\,du. J ij := ϑ 1 ∫ D χ ( u ) ∇ P i ( u ) ⋅ ∇ P j ( u ) d u .
For ξ ∈ R m \xi\in\mathbb{R}^m ξ ∈ R m , let P ξ : = ∑ j ξ j P j P_\xi:= \sum_j \xi_jP_j P ξ := ∑ j ξ j P j . Then
ξ T J ξ = 1 ϑ ∫ D χ ( u ) ∥ ∇ P ξ ( u ) ∥ 2 d u . \xi^{\mathsf{T}}J\xi= \frac{1}{\vartheta}\int_D \chi(u)\lVert\nabla P_\xi(u)\rVert^2\,du. ξ T J ξ = ϑ 1 ∫ D χ ( u ) ∥ ∇ P ξ ( u ) ∥ 2 d u .
If this integral is zero, the polynomial ∇ P ξ \nabla P_\xi ∇ P ξ vanishes on a nonempty open set. Hence P ξ P_\xi P ξ is constant, and P ξ ( 0 ) = 0 P_\xi(0)=0 P ξ ( 0 ) = 0 makes it zero. Since the P j P_j P j form a basis, ξ = 0 \xi=0 ξ = 0 . Thus J J J is positive definite.
To obtain the prescribed polynomial sums, define
G k , F ( t ) : = 1 k ∑ u ∈ A k 0 P ( Φ t ( u ) ) − y ( F ) , J k : = D G k , F ( 0 ) . G_{k,F}(t) := \frac{1}{k}\sum_{u\in A_k^0}\mathbf{P}(\Phi_t(u))-y(F), \qquad J_k := DG_{k,F}(0). G k , F ( t ) := k 1 u ∈ A k 0 ∑ P ( Φ t ( u )) − y ( F ) , J k := D G k , F ( 0 ) .
The matrix J k J_k J k depends on k k k but not on F F F , since y ( F ) y(F) y ( F ) is independent of t t t . Applying (23) to the coordinates of P \mathbf{P} P and to the functions χ ∇ P i ⋅ ∇ P j \chi\nabla P_i\cdot\nabla P_j χ ∇ P i ⋅ ∇ P j gives
∥ G k , F ( 0 ) ∥ ≤ C ( ε + k − 1 / d ) , ∥ J k − J ∥ ≤ C k − 1 / d . \lVert G_{k,F}(0)\rVert\le C(\varepsilon+k^{-1/d}), \qquad\lVert J_k-J\rVert\le Ck^{-1/d}. ∥ G k , F ( 0 )∥ ≤ C ( ε + k − 1/ d ) , ∥ J k − J ∥ ≤ C k − 1/ d .
After increasing q 0 q_0 q 0 , all the matrices J k J_k J k that occur here are invertible, with uniformly bounded inverses. The second derivatives of G k , F G_{k,F} G k , F are uniformly bounded for ∥ t ∥ ≤ τ \lVert t\rVert\le\tau ∥ t ∥ ≤ τ , where τ > 0 \tau>0 τ > 0 is fixed; they too do not depend on F F F .
Set δ k : = C 1 ( ε + k − 1 / d ) \delta_k:=C_1(\varepsilon+k^{-1/d}) δ k := C 1 ( ε + k − 1/ d ) and
N k , F ( t ) : = t − J k − 1 G k , F ( t ) . N_{k,F}(t):=t-J_k^{-1}G_{k,F}(t). N k , F ( t ) := t − J k − 1 G k , F ( t ) .
Choose C 1 C_1 C 1 so that ∥ N k , F ( 0 ) ∥ ≤ δ k / 2 \lVert N_{k,F}(0)\rVert\le\delta_k/2 ∥ N k , F ( 0 )∥ ≤ δ k /2 . By decreasing ε 0 \varepsilon_0 ε 0 and increasing q 0 q_0 q 0 , we may assume that δ k ≤ τ \delta_k\le\tau δ k ≤ τ and, for all ∥ t ∥ ≤ δ k \lVert t\rVert\le\delta_k ∥ t ∥ ≤ δ k ,
∥ D N k , F ( t ) ∥ = ∥ J k − 1 ( J k − D G k , F ( t ) ) ∥ ≤ C δ k ≤ 1 2 . \lVert DN_{k,F}(t)\rVert=\lVert J_k^{-1}(J_k-DG_{k,F}(t))\rVert\le C\delta_k\le\frac{1}{2}. ∥ D N k , F ( t )∥ = ∥ J k − 1 ( J k − D G k , F ( t ))∥ ≤ C δ k ≤ 2 1 .
It follows that N k , F N_{k,F} N k , F maps the closed ball { ∥ t ∥ ≤ δ k } \{\lVert t\rVert\le\delta_k\} {∥ t ∥ ≤ δ k } into itself and is a contraction there. Let t ( F ) t(F) t ( F ) be its unique fixed point. Then G k , F ( t ( F ) ) = 0 G_{k,F}(t(F))=0 G k , F ( t ( F )) = 0 and ∥ t ( F ) ∥ ≤ δ k \lVert t(F)\rVert\le\delta_k ∥ t ( F )∥ ≤ δ k .
The vector fields X j X_j X j vanish outside int D \operatorname{int}D int D . By further decreasing ε 0 \varepsilon_0 ε 0 and increasing q 0 q_0 q 0 , we can ensure that ∥ D Φ t − I ∥ ≤ 1 / 2 \lVert D\Phi_t-I\rVert\le1/2 ∥ D Φ t − I ∥ ≤ 1/2 on R d \mathbb{R}^d R d whenever ∥ t ∥ ≤ δ k \lVert t\rVert\le\delta_k ∥ t ∥ ≤ δ k . Hence
1 2 ∥ u − v ∥ ≤ ∥ Φ t ( u ) − Φ t ( v ) ∥ ≤ 3 2 ∥ u − v ∥ ( u , v ∈ R d ) . \frac{1}{2}\lVert u-v\rVert\le\lVert\Phi_t(u)-\Phi_t(v)\rVert\le\frac{3}{2}\lVert u-v\rVert\qquad(u,v\in\mathbb{R}^d). 2 1 ∥ u − v ∥ ≤ ∥ Φ t ( u ) − Φ t ( v )∥ ≤ 2 3 ∥ u − v ∥ ( u , v ∈ R d ) .
Moreover, Φ t \Phi_t Φ t is a bijection of R d \mathbb{R}^d R d : for each y y y , the map u ↦ y − ( Φ t ( u ) − u ) u\mapsto y-(\Phi_t(u)-u) u ↦ y − ( Φ t ( u ) − u ) is a contraction. Since Φ t \Phi_t Φ t is the identity outside int D \operatorname{int}D int D , it maps D D D onto itself and fixes ∂ D \partial D ∂ D . Therefore, with
A ( q , F ) : = Φ t ( F ) ( A k 0 ) , A(q,F):=\Phi_{t(F)}(A_k^0), A ( q , F ) := Φ t ( F ) ( A k 0 ) ,
we have A ( q , F ) ⊂ int D A(q,F)\subset\operatorname{int}D A ( q , F ) ⊂ int D and
sep ( A ( q , F ) ) , dist ( A ( q , F ) , ∂ D ) ≥ 1 2 c d ( 2 / 3 ) 1 / d q − 1 / d . \operatorname{sep}(A(q,F)),\ \operatorname{dist}(A(q,F),\partial D)\ge\frac{1}{2}c_d(2/3)^{1/d}q^{-1/d}. sep ( A ( q , F )) , dist ( A ( q , F ) , ∂ D ) ≥ 2 1 c d ( 2/3 ) 1/ d q − 1/ d .
We may take σ d : = 1 2 c d ( 2 / 3 ) 1 / d \sigma_d:=\frac{1}{2}c_d(2/3)^{1/d} σ d := 2 1 c d ( 2/3 ) 1/ d . The identity G k , F ( t ( F ) ) = 0 G_{k,F}(t(F))=0 G k , F ( t ( F )) = 0 gives
∑ u ∈ A ( q , F ) P ( u ) = q ∫ I P ( u ) d u − ∑ u ∈ F P ( u ) . \sum_{u\in A(q,F)}\mathbf{P}(u)=q\int_I\mathbf{P}(u)\,du-\sum_{u\in F}\mathbf{P}(u). u ∈ A ( q , F ) ∑ P ( u ) = q ∫ I P ( u ) d u − u ∈ F ∑ P ( u ) .
The sets F F F and A ( q , F ) \mathcal{A}(q,F) A ( q , F ) are disjoint, since the former lies outside int D \operatorname{int} D int D and the latter lies inside. Thus their union has # F + k = q \#F+k=q # F + k = q points. The polynomial identities above and the identity for constants give
∑ u ∈ F ∪ A ( q , F ) P ( u ) = q ∫ I P ( u ) d u ( P ∈ P p − 1 ) . \sum_{u\in F\cup\mathcal{A}(q,F)} P(u)=q\int_I P(u)\,du \qquad(P\in\mathcal{P}_{p-1}). u ∈ F ∪ A ( q , F ) ∑ P ( u ) = q ∫ I P ( u ) d u ( P ∈ P p − 1 ) .
Finally, the set F q , ε \mathcal{F}_{q,\varepsilon} F q , ε is Borel: the condition F ⊂ I ∖ int D F\subset I\setminus\operatorname{int}D F ⊂ I ∖ int D , the finite-point condition, and the count and polynomial-sum inequalities are Borel conditions on configurations. On this domain, k = q − # F k=q-\#F k = q − # F and y ( F ) y(F) y ( F ) are Borel. The sets A k 0 A_k^0 A k 0 , matrices J k J_k J k , and maps Φ t \Phi_t Φ t fixed for each integer k k k make ( F , t ) ↦ N k , F ( t ) (F,t)\mapsto\mathcal{N}_{k,F}(t) ( F , t ) ↦ N k , F ( t ) Borel in F F F and continuous in t t t . Starting with t 0 ( F ) = 0 t_0(F)=0 t 0 ( F ) = 0 , define t n + 1 ( F ) : = N k , F ( t n ( F ) ) t_{n+1}(F):=\mathcal{N}_{k,F}(t_n(F)) t n + 1 ( F ) := N k , F ( t n ( F )) . Every t n t_n t n is Borel, and contraction gives t n ( F ) → t ( F ) t_n(F)\to t(F) t n ( F ) → t ( F ) . Hence F ↦ t ( F ) F\mapsto t(F) F ↦ t ( F ) and F ↦ δ A ( q , F ) F\mapsto\delta_{\mathcal{A}(q,F)} F ↦ δ A ( q , F ) are Borel. □ \square □
Fix q ≥ q 0 q\ge q_0 q ≥ q 0 and 0 < ε ≤ ε 0 0<\varepsilon\le\varepsilon_0 0 < ε ≤ ε 0 , and set L : = ( q / ρ ) 1 / d L:=(q/\rho)^{1/d} L := ( q / ρ ) 1/ d . For an L L L -cell P = v + L I P=v+LI P = v + L I , put D P : = v + L D D_P:=v+LD D P := v + L D . Given a locally finite point set Λ \Lambda Λ , define
F L , P ( Λ ) : = { u − v L : u ∈ Λ ∩ ( P ∖ int D P ) } . (24) F_{L,P}(\Lambda):=\left\{\frac{u-v}{L}:u\in\Lambda\cap(P\setminus\operatorname{int}D_P)\right\}. \tag*{(24)} F L , P ( Λ ) := { L u − v : u ∈ Λ ∩ ( P ∖ int D P ) } . (24)
The subscript L L L records the scale used to normalize the exterior points, and P P P specifies their location. Whenever F L , P ( Λ ) ∈ F q , ε F_{L,P}(\Lambda)\in\mathcal{F}_{q,\varepsilon} F L , P ( Λ ) ∈ F q , ε , define
R P ( δ Λ ) : = δ ( Λ ∖ int D P ) ∪ ( v + L A ( q , F L , P ( Λ ) ) ) . R_P(\delta_\Lambda):=\delta_{(\Lambda\setminus\operatorname{int}D_P)\cup(v+L\mathcal{A}(q,F_{L,P}(\Lambda)))}. R P ( δ Λ ) := δ ( Λ ∖ int D P ) ∪ ( v + L A ( q , F L , P ( Λ ))) .
The condition F L , P ( Λ ) ∈ F q , ε F_{L,P}(\Lambda)\in\mathcal{F}_{q,\varepsilon} F L , P ( Λ ) ∈ F q , ε is Borel in δ Λ \delta_\Lambda δ Λ . By Proposition 3.5 , the map R P R_P R P is Borel on this domain, agrees with δ Λ \delta_\Lambda δ Λ outside int D P \operatorname{int}D_P int D P , and satisfies
∫ P H ( u ) d R P ( δ Λ ) ( u ) = ρ ∫ P H ( u ) d u ( H ∈ P p − 1 ) . (25) \int_P H(u)\,dR_P(\delta_\Lambda)(u)=\rho\int_P H(u)\,du \qquad(H\in\mathcal{P}_{p-1}). \tag*{(25)} ∫ P H ( u ) d R P ( δ Λ ) ( u ) = ρ ∫ P H ( u ) d u ( H ∈ P p − 1 ) . (25)
In particular, R P ( δ Λ ) ( P ) = q R_P(\delta_\Lambda)(P)=q R P ( δ Λ ) ( P ) = q . Indeed, (25) is (20) after the change of variables u ↦ v + L u u\mapsto v+Lu u ↦ v + Lu .
Variance from exact cell moments
We estimate the number variance of periodic point sets whose polynomial moments on each L L L -cell agree with those of ρ d u \rho\,du ρ d u . The estimate is uniform in the period and in the point set; its constant depends on L L L . In the construction below, we choose an integer q q q sufficiently large and set L = ( q / ρ ) 1 / d L=(q/\rho)^{1/d} L = ( q / ρ ) 1/ d . Thus q = ρ L d q=\rho L^d q = ρ L d is the prescribed number of points in each modified cell.
Lemma 3.6 (Periodic variance from exact moments). Fix a positive integer p p p with 2 p > d + 1 2p>d+1 2 p > d + 1 , and an integer q ≥ 1 q\ge1 q ≥ 1 . Set L : = ( q / ρ ) 1 / d L:=(q/\rho)^{1/d} L := ( q / ρ ) 1/ d . There is K L < ∞ K_L<\infty K L < ∞ such that the following holds for every integer m ≥ 1 m\ge1 m ≥ 1 . Put S : = m L S:=mL S := m L and
C L , m : = { L k + [ 0 , L ) d : k ∈ { 0 , … , m − 1 } d } . \mathcal{C}_{L,m}:=\{Lk+[0,L)^d:k\in\{0,\ldots,m-1\}^d\}. C L , m := { L k + [ 0 , L ) d : k ∈ { 0 , … , m − 1 } d } . Suppose Λ ⊂ [ 0 , S ) d \Lambda\subset[0,S)^d Λ ⊂ [ 0 , S ) d is finite and, for every Q ∈ C L , m Q\in\mathcal{C}_{L,m} Q ∈ C L , m ,
# ( Λ ∩ Q ) = q , ∑ u ∈ Λ ∩ Q P ( u ) = ρ ∫ Q P ( u ) d u ( P ∈ P p − 1 ) . \#(\Lambda\cap Q)=q,\qquad\sum_{u\in\Lambda\cap Q}P(u)=\rho\int_QP(u)\,du \qquad(P\in\mathcal{P}_{p-1}). # ( Λ ∩ Q ) = q , u ∈ Λ ∩ Q ∑ P ( u ) = ρ ∫ Q P ( u ) d u ( P ∈ P p − 1 ) . Define the S Z d S\mathbb{Z}^d S Z d -periodic point set
Λ p e r : = ⋃ v ∈ S Z d ( Λ + v ) . \Lambda^{\mathrm{per}}:=\bigcup_{v\in S\mathbb{Z}^d}(\Lambda+v). Λ per := v ∈ S Z d ⋃ ( Λ + v ) . Then
1 S d ∫ [ 0 , S ) d ∣ N B R ( δ Λ p e r − t ) − ρ Vol d ( B R ) ∣ 2 d t ≤ K L R d − 1 ( R ≥ 1 ) . \frac{1}{S^d}\int_{[0,S)^d}\left|N_{B_R}(\delta_{\Lambda^{\mathrm{per}}-t})-\rho\operatorname{Vol}_d(B_R)\right|^2\,dt\le K_LR^{d-1}\qquad(R\ge1). S d 1 ∫ [ 0 , S ) d ∣ N B R ( δ Λ per − t ) − ρ Vol d ( B R ) ∣ 2 d t ≤ K L R d − 1 ( R ≥ 1 ) .
Proof. For each Q ∈ L L , m Q \in\mathcal{L}_{L,m} Q ∈ L L , m , define the signed measure
Δ Q : = δ Λ ∩ Q − ρ V o l d ∣ Q . \Delta_Q := \delta_{\Lambda\cap Q} - \rho\,\mathrm{Vol}_d\mathbin{|}_Q. Δ Q := δ Λ ∩ Q − ρ Vol d ∣ Q .
The hypotheses imply
∫ Q P ( u ) d Δ Q ( u ) = 0 ( P ∈ P p − 1 ) , ∥ Δ Q ∥ T V ≤ 2 q . (26) \int_Q P(u)\,d\Delta_Q(u)=0 \quad(P \in\mathcal{P}_{p-1}), \qquad\|\Delta_Q\|_{\mathrm{TV}} \le2q. \tag*{(26)} ∫ Q P ( u ) d Δ Q ( u ) = 0 ( P ∈ P p − 1 ) , ∥ Δ Q ∥ TV ≤ 2 q . (26)
We use (26) to express each Δ Q \Delta_Q Δ Q as distributional derivatives of order p p p .
Let c Q c_Q c Q be the centre of Q Q Q . For φ ∈ C c ∞ ( R d ) \varphi\in C^\infty_c(\mathbb{R}^d) φ ∈ C c ∞ ( R d ) , define its Taylor polynomial at c Q c_Q c Q of degree less than p p p by
P Q , φ ( u ) : = ∑ ∣ α ∣ < p ∂ α φ ( c Q ) α ! ( u − c Q ) α . P_{Q,\varphi}(u):=\sum_{|\alpha|<p}\frac{\partial^\alpha\varphi(c_Q)}{\alpha!}(u-c_Q)^\alpha. P Q , φ ( u ) := ∣ α ∣ < p ∑ α ! ∂ α φ ( c Q ) ( u − c Q ) α .
Here α ∈ N 0 d \alpha\in\mathbb{N}_0^d α ∈ N 0 d is a multi-index, α ! : = ∏ i α i ! \alpha!:=\prod_i\alpha_i! α ! := ∏ i α i ! , and ∂ α : = ∏ i ∂ i α i \partial^\alpha:=\prod_i\partial_i^{\alpha_i} ∂ α := ∏ i ∂ i α i . Taylor’s integral remainder formula gives
φ ( u ) − P Q , φ ( u ) = p ∑ ∣ α ∣ = p ( u − c Q ) α α ! ∫ 0 1 ( 1 − s ) p − 1 ∂ α φ ( c Q + s ( u − c Q ) ) d s . (27) \varphi(u)-P_{Q,\varphi}(u)=p\sum_{|\alpha|=p}\frac{(u-c_Q)^\alpha}{\alpha!}\int_0^1(1-s)^{p-1}\partial^\alpha\varphi(c_Q+s(u-c_Q))\,ds. \tag*{(27)} φ ( u ) − P Q , φ ( u ) = p ∣ α ∣ = p ∑ α ! ( u − c Q ) α ∫ 0 1 ( 1 − s ) p − 1 ∂ α φ ( c Q + s ( u − c Q )) d s . (27)
For each ∣ α ∣ = p |\alpha|=p ∣ α ∣ = p , define a signed measure Θ Q , α \Theta_{Q,\alpha} Θ Q , α by
∫ ψ d Θ Q , α : = ( − 1 ) p p α ! ∫ Q ∫ 0 1 ( 1 − s ) p − 1 ( u − c Q ) α ψ ( c Q + s ( u − c Q ) ) d s d Δ Q ( u ) \int\psi\,d\Theta_{Q,\alpha}:=\frac{(-1)^p p}{\alpha!}\int_Q\int_0^1(1-s)^{p-1}(u-c_Q)^\alpha\psi(c_Q+s(u-c_Q))\,ds\,d\Delta_Q(u) ∫ ψ d Θ Q , α := α ! ( − 1 ) p p ∫ Q ∫ 0 1 ( 1 − s ) p − 1 ( u − c Q ) α ψ ( c Q + s ( u − c Q )) d s d Δ Q ( u )
for ψ ∈ C c ( R d ) \psi\in C_c(\mathbb{R}^d) ψ ∈ C c ( R d ) . Since P Q , φ ∈ P p − 1 P_{Q,\varphi}\in\mathcal{P}_{p-1} P Q , φ ∈ P p − 1 , (26) and (27) give, as distributions,
Δ Q = ∑ ∣ α ∣ = p ∂ α Θ Q , α , supp Θ Q , α ⊂ Q ‾ , ∥ Θ Q , α ∥ T V ≤ C d , p q L p . (28) \Delta_Q=\sum_{|\alpha|=p}\partial^\alpha\Theta_{Q,\alpha}, \qquad\operatorname{supp}\Theta_{Q,\alpha}\subset\overline{Q}, \qquad\|\Theta_{Q,\alpha}\|_{\mathrm{TV}}\le C_{d,p}qL^p. \tag*{(28)} Δ Q = ∣ α ∣ = p ∑ ∂ α Θ Q , α , supp Θ Q , α ⊂ Q , ∥ Θ Q , α ∥ TV ≤ C d , p q L p . (28)
Extend ∑ Q ∈ L L , m Δ Q \sum_{Q\in\mathcal{L}_{L,m}}\Delta_Q ∑ Q ∈ L L , m Δ Q and ∑ Q ∈ L L , m Θ Q , α \sum_{Q\in\mathcal{L}_{L,m}}\Theta_{Q,\alpha} ∑ Q ∈ L L , m Θ Q , α periodically under S Z d S\mathbb{Z}^d S Z d , and denote the extensions by Δ \Delta Δ and Θ α \Theta_\alpha Θ α . For an S Z d S\mathbb{Z}^d S Z d -periodic signed measure ν \nu ν , use the Fourier coefficients
a ν ( k ) : = S − d ∫ [ 0 , S ) d e − 2 π i ⟨ k , u ⟩ / S d ν ( u ) ( k ∈ Z d ) . a_\nu(k):=S^{-d}\int_{[0,S)^d}e^{-2\pi i\langle k,u\rangle/S}\,d\nu(u)\qquad(k\in\mathbb{Z}^d). a ν ( k ) := S − d ∫ [ 0 , S ) d e − 2 π i ⟨ k , u ⟩ / S d ν ( u ) ( k ∈ Z d ) .
Taking Fourier coefficients in (28) yields
∣ a Δ ( k ) ∣ 2 ≤ C d , p ∥ k / S ∥ 2 p ∑ ∣ α ∣ = p ∣ a Θ α ( k ) ∣ 2 . (29) |a_\Delta(k)|^2\le C_{d,p}\|k/S\|^{2p}\sum_{|\alpha|=p}|a_{\Theta_\alpha}(k)|^2. \tag*{(29)} ∣ a Δ ( k ) ∣ 2 ≤ C d , p ∥ k / S ∥ 2 p ∣ α ∣ = p ∑ ∣ a Θ α ( k ) ∣ 2 . (29)
Also a Δ ( 0 ) = 0 a_\Delta(0)=0 a Δ ( 0 ) = 0 by (26) .
We next bound the squared Fourier coefficients of Δ \Delta Δ and Θ α \Theta_\alpha Θ α in balls of radius one, uniformly in S S S . Choose χ ∈ C c ∞ ( R d ) \chi\in C^\infty_c(\mathbb{R}^d) χ ∈ C c ∞ ( R d ) such that ∣ χ ^ ∣ ≥ 1 |\widehat{\chi}|\ge1 ∣ χ ∣ ≥ 1 on B 1 B_1 B 1 . By (26) and (28) , a translate of supp χ \operatorname{supp}\chi supp χ intersects at most C χ , L C_{\chi,L} C χ , L cells, and therefore
sup t ∈ R d ∫ ∣ χ ( u − t ) ∣ d ∣ ν ∣ ( u ) ≤ C L ( ν = Δ , Θ α ) . \sup_{t\in\mathbb{R}^d}\int|\chi(u-t)|\,d|\nu|(u)\le C_L \qquad(\nu=\Delta,\Theta_\alpha). t ∈ R d sup ∫ ∣ χ ( u − t ) ∣ d ∣ ν ∣ ( u ) ≤ C L ( ν = Δ , Θ α ) .
For a ∈ R d a\in\mathbb{R}^d a ∈ R d , Parseval on R d / S Z d \mathbb{R}^d/S\mathbb{Z}^d R d / S Z d , applied to the S Z d S\mathbb{Z}^d S Z d -periodic function
t ⟼ ∫ χ ( u − t ) e − 2 π i ⟨ a , u − t ⟩ d ν ( u ) , t\longmapsto\int\chi(u-t)e^{-2\pi i\langle a,u-t\rangle}\,d\nu(u), t ⟼ ∫ χ ( u − t ) e − 2 π i ⟨ a , u − t ⟩ d ν ( u ) ,
gives
∑ k ∈ Z d ∣ a ν ( k ) ∣ 2 ∣ χ ^ ( a − k / S ) ∣ 2 = S − d ∫ [ 0 , S ) d ∣ ∫ χ ( u − t ) e − 2 π i ⟨ a , u − t ⟩ d ν ( u ) ∣ 2 d t ≤ C L 2 . \sum_{k\in\mathbb{Z}^d}|a_\nu(k)|^2|\widehat{\chi}(a-k/S)|^2
=S^{-d}\int_{[0,S)^d}\left|\int\chi(u-t)e^{-2\pi i\langle a,u-t\rangle}\,d\nu(u)\right|^2dt
\le C_L^2. k ∈ Z d ∑ ∣ a ν ( k ) ∣ 2 ∣ χ ( a − k / S ) ∣ 2 = S − d ∫ [ 0 , S ) d ∫ χ ( u − t ) e − 2 π i ⟨ a , u − t ⟩ d ν ( u ) 2 d t ≤ C L 2 .
Hence
sup a ∈ R d ∑ k / S ∈ a + B 1 ∣ a ν ( k ) ∣ 2 ≤ C L 2 ( ν = Δ , Θ α ) . (30) \sup_{a\in\mathbb{R}^{d}}\sum_{k/S\in a+B_1}|a_\nu(k)|^2\le C_L^2\qquad(\nu=\Delta,\Theta_\alpha). \tag*{(30)} a ∈ R d sup k / S ∈ a + B 1 ∑ ∣ a ν ( k ) ∣ 2 ≤ C L 2 ( ν = Δ , Θ α ) . (30)
The constant C L C_L C L depends on L L L , d d d , p p p and the fixed χ \chi χ , but not on m m m , S S S , or Λ \Lambda Λ .
For 0 < ∥ k / S ∥ < 1 0<\lVert k/S\rVert<1 0 < ∥ k / S ∥ < 1 , the exponent 2 p > d + 1 2p>d+1 2 p > d + 1 and (29) give
∑ 0 < ∥ k / S ∥ < 1 ∣ a Δ ( k ) ∣ 2 ∥ k / S ∥ d + 1 ≤ C d , p ∑ ∣ α ∣ = p ∑ ∥ k / S ∥ < 1 ∣ a Θ α ( k ) ∣ 2 ≤ C L . \sum_{0<\lVert k/S\rVert<1}\frac{|a_\Delta(k)|^2}{\lVert k/S\rVert^{d+1}}\le C_{d,p}\sum_{|\alpha|=p}\sum_{\lVert k/S\rVert<1}|a_{\Theta_\alpha}(k)|^2\le C_L. 0 < ∥ k / S ∥ < 1 ∑ ∥ k / S ∥ d + 1 ∣ a Δ ( k ) ∣ 2 ≤ C d , p ∣ α ∣ = p ∑ ∥ k / S ∥ < 1 ∑ ∣ a Θ α ( k ) ∣ 2 ≤ C L .
For n ≥ 1 n\ge1 n ≥ 1 , the annulus n ≤ ∥ ξ ∥ < n + 1 n\le\lVert\xi\rVert<n+1 n ≤ ∥ ξ ∥ < n + 1 can be covered by O d ( n d − 1 ) O_d(n^{d-1}) O d ( n d − 1 ) balls of radius one. Applying (30) with ν = Δ \nu=\Delta ν = Δ therefore yields
∑ ∥ k / S ∥ ≥ 1 ∣ a Δ ( k ) ∣ 2 ∥ k / S ∥ d + 1 ≤ C L ∑ n ≥ 1 n − 2 < ∞ . \sum_{\lVert k/S\rVert\ge1}\frac{|a_\Delta(k)|^2}{\lVert k/S\rVert^{d+1}}\le C_L\sum_{n\ge1}n^{-2}<\infty. ∥ k / S ∥ ≥ 1 ∑ ∥ k / S ∥ d + 1 ∣ a Δ ( k ) ∣ 2 ≤ C L n ≥ 1 ∑ n − 2 < ∞.
Combining the estimates, we obtain
∑ k ≠ 0 ∣ a Δ ( k ) ∣ 2 ∥ k / S ∥ d + 1 ≤ K L , (31) \sum_{k\ne0}\frac{|a_\Delta(k)|^2}{\lVert k/S\rVert^{d+1}}\le K_L, \tag*{(31)} k = 0 ∑ ∥ k / S ∥ d + 1 ∣ a Δ ( k ) ∣ 2 ≤ K L , (31)
where K L K_L K L is independent of m m m and Λ \Lambda Λ .
By the definition of Δ \Delta Δ , its periodic extension is δ Λ p e r − ρ Vol d \delta_{\Lambda}^{\mathrm{per}}-\rho\operatorname{Vol}_d δ Λ per − ρ Vol d . Parseval now gives, with t t t as the translation variable,
S − d ∫ [ 0 , S ) d ∣ N B R ( δ Λ p e r − t ) − ρ Vol d ( B R ) ∣ 2 d t = ∑ k ≠ 0 ∣ a Δ ( k ) ∣ 2 ∣ 1 B R ^ ( k / S ) ∣ 2 . S^{-d}\int_{[0,S)^d}\left|N_{B_R}(\delta_{\Lambda}^{\mathrm{per}}-t)-\rho\operatorname{Vol}_d(B_R)\right|^2\,dt
=\sum_{k\ne0}|a_\Delta(k)|^2\left|\widehat{1_{B_R}}(k/S)\right|^2. S − d ∫ [ 0 , S ) d ∣ N B R ( δ Λ per − t ) − ρ Vol d ( B R ) ∣ 2 d t = k = 0 ∑ ∣ a Δ ( k ) ∣ 2 1 B R ( k / S ) 2 .
For R ≥ 1 R\ge1 R ≥ 1 and ξ ≠ 0 \xi\ne0 ξ = 0 , the Fourier transform of the Euclidean ball satisfies
∣ 1 B R ^ ( ξ ) ∣ 2 ≤ C d R d − 1 ∥ ξ ∥ − ( d + 1 ) ; \left|\widehat{1_{B_R}}(\xi)\right|^2\le C_dR^{d-1}\lVert\xi\rVert^{-(d+1)}; 1 B R ( ξ ) 2 ≤ C d R d − 1 ∥ ξ ∥ − ( d + 1 ) ;
see [39 Sec. IV and Appendix A] . The assertion follows from (31) . □
Construction of a low-variance realization
Fix a realization κ \kappa κ . We construct a nearby realization by changing its points inside disjoint S S S -cubes. In the core of each such cube, every L L L -cell will have exactly q q q points and satisfy (25) ; the points in the outer shell will be chosen so that the total point count in the S S S -cube is unchanged. We then apply Lemma 3.6 to the configurations on the cores and estimate the local L 2 L^2 L 2 change separately.
Let a ∈ A d ( X , μ ) a\in A_d(X,\mu) a ∈ A d ( X , μ ) and κ ∈ R p , 2 a ( X , μ ) \kappa\in\mathbb{R}_{p,2}^{a}(X,\mu) κ ∈ R p , 2 a ( X , μ ) . By Lemma 2.1 and essential freeness, take a pointwise equivariant Borel representative on an invariant conull Borel set X 0 X_0 X 0 on which a a a is free. Put Λ x : = supp κ ( x ) \Lambda_x:=\operatorname{supp}\kappa(x) Λ x := supp κ ( x ) for x ∈ X 0 x\in X_0 x ∈ X 0 .
Exact moment configurations in L L L -cells. Fix an integer p ≥ 1 p\ge1 p ≥ 1 with 2 p > d + 1 2p>d+1 2 p > d + 1 and a basis P 1 , … , P m P_1,\ldots,P_m P 1 , … , P m of P p − 1 0 \mathcal{P}_{p-1}^{0} P p − 1 0 . Choose a closed cube D ⊂ ( 0 , 1 ) d D\subset(0,1)^d D ⊂ ( 0 , 1 ) d , and set ϑ : = Vol d ( D ) \vartheta:=\operatorname{Vol}_d(D) ϑ := Vol d ( D ) . Let ε 0 \varepsilon_0 ε 0 , q 0 q_0 q 0 , σ d \sigma_d σ d be the constants of Proposition 3.5 for these choices, and put ε : = ε 0 / 2 \varepsilon:=\varepsilon_0/2 ε := ε 0 /2 .
For each integer q ≥ q 0 q\ge q_0 q ≥ q 0 , set L : = ( q / ρ ) 1 / d L:=(q/\rho)^{1/d} L := ( q / ρ ) 1/ d . For every v ∈ R d v\in\mathbb{R}^d v ∈ R d , put P : = v + L I P:=v+LI P := v + L I and D P : = v + L D D_P:=v+LD D P := v + L D . Recall that F L , P ( Λ ) F_{L,P}(\Lambda) F L , P ( Λ ) is the normalized exterior point set defined in (24) . We call P P P admissible for Λ \Lambda Λ if
F L , P ( Λ ) ∈ F q , ε , # ( Λ ∩ D P ) ≤ 2 ρ ϑ L d . (32) F_{L,P}(\Lambda)\in\mathcal{F}_{q,\varepsilon},\qquad\#(\Lambda\cap D_P)\le2\rho\vartheta L^d. \tag*{(32)} F L , P ( Λ ) ∈ F q , ε , # ( Λ ∩ D P ) ≤ 2 ρϑ L d . (32)
Define
E L : = { x ∈ X 0 : L I is not admissible for Λ x } . \mathcal{E}_L:=\{x\in X_0: LI\text{ is not admissible for }\Lambda_x\}. E L := { x ∈ X 0 : L I is not admissible for Λ x } .
Apply Lemma 3.4 with B = I B=I B = I to A = I ∖ D A=I\setminus D A = I ∖ D with weights 1 , P 1 , … , P m 1,P_1,\ldots,P_m 1 , P 1 , … , P m , and to A = D A=D A = D with weight 1 1 1 . Since the expected count on ∂ ( L D ) \partial(LD) ∂ ( L D ) is zero, we may disregard points there. With tolerance ρ min { ε , ϑ } \rho\min\{\varepsilon,\vartheta\} ρ min { ε , ϑ } , any x ∈ E L x\in\mathcal{E}_L x ∈ E L outside the boundary-null set violates at least one of these weighted-count inequalities, by the conditions in (19) and (32) . The finite-family estimate (18) therefore gives
μ ( E L ) ⟶ 0 , L − d ∫ E L N L I ( κ ( x ) ) d μ ( x ) ⟶ 0 ( q → ∞ ) . \mu(\mathcal{E}_L)\longrightarrow0,\qquad L^{-d}\int_{\mathcal{E}_L}N_{LI}(\kappa(x))\,d\mu(x)\longrightarrow0\qquad(q\to\infty). μ ( E L ) ⟶ 0 , L − d ∫ E L N L I ( κ ( x )) d μ ( x ) ⟶ 0 ( q → ∞ ) .
Put h q : = q − 1 / d h_q:=q^{-1/d} h q := q − 1/ d , and choose a translate Γ q \Gamma_q Γ q of h q Z d h_q\mathbb{Z}^d h q Z d whose points are disjoint from ∂ I ∪ ∂ D \partial I\cup\partial D ∂ I ∪ ∂ D .
Define
F q 0 : = { u ∈ Γ q ∩ ( I ∖ D ) : dist ( u , D ) ≥ h q / 4 } . F_q^0:=\{u\in\Gamma_q\cap(I\setminus D):\operatorname{dist}(u,D)\ge h_q/4\}. F q 0 := { u ∈ Γ q ∩ ( I ∖ D ) : dist ( u , D ) ≥ h q /4 } .
For each u ∈ Γ q u\in\Gamma_q u ∈ Γ q , put J u : = u + [ 0 , h q ) d J_u:=u+[0,h_q)^d J u := u + [ 0 , h q ) d . These cubes partition R d \mathbb{R}^d R d . If 1 F q 0 ( u ) ≠ 1 I ∖ D ( y ) \mathbf{1}_{F_q^0}(u)\ne\mathbf{1}_{I\setminus D}(y) 1 F q 0 ( u ) = 1 I ∖ D ( y ) for some y ∈ J u y\in J_u y ∈ J u , then J u J_u J u is contained in the ( d + 1 ) h q (\sqrt{d}+1)h_q ( d + 1 ) h q -neighbourhood of ∂ I ∪ ∂ D \partial I\cup\partial D ∂ I ∪ ∂ D . The union of these cubes has volume at most K d , D h q K_{d,D}h_q K d , D h q , since both boundaries consist of finitely many flat faces. For each u ∈ F q 0 u\in F_q^0 u ∈ F q 0 for which J u ⊂ I ∖ D J_u\subset I\setminus D J u ⊂ I ∖ D , the mean-value inequality and ∥ y − u ∥ ≤ d h q \lVert y-u\rVert\le\sqrt{d}\,h_q ∥ y − u ∥ ≤ d h q for y ∈ J u y\in J_u y ∈ J u give, for 1 ≤ i ≤ m 1\le i\le m 1 ≤ i ≤ m ,
∣ h q d P i ( u ) − ∫ J u P i ( y ) d y ∣ ≤ d h q d + 1 sup y ∈ [ − 1 , 2 ] d ∥ ∇ P i ( y ) ∥ . \left|h_q^dP_i(u)-\int_{J_u}P_i(y)\,dy\right|\le\sqrt{d}\,h_q^{d+1}\sup_{y\in[-1,2]^d}\lVert\nabla P_i(y)\rVert. h q d P i ( u ) − ∫ J u P i ( y ) d y ≤ d h q d + 1 y ∈ [ − 1 , 2 ] d sup ∥ ∇ P i ( y )∥ .
Every cube not covered by this estimate for which either its base point lies in F q 0 F_q^0 F q 0 or the cube intersects I ∖ D I\setminus D I ∖ D lies in the ( 2 d + 2 ) h q (2\sqrt{d}+2)h_q ( 2 d + 2 ) h q -neighbourhood of ∂ I ∪ ∂ D \partial I\cup\partial D ∂ I ∪ ∂ D . The total volume of these cubes is O d , D ( h q ) O_{d,D}(h_q) O d , D ( h q ) , and their contributions to both the Riemann sum and the integral are bounded by this volume times sup [ − 1 , 2 ] d ∣ P i ∣ \sup_{[-1,2]^d}|P_i| sup [ − 1 , 2 ] d ∣ P i ∣ . Combining these bounds, there is a constant K l a t K_{\mathrm{lat}} K lat , depending on d , D , P 1 , … , P m d,D,P_1,\ldots,P_m d , D , P 1 , … , P m , such that
∣ h q d # F q 0 − Vol d ( I ∖ D ) ∣ ≤ K l a t h q , \left|h_q^d\#F_q^0-\operatorname{Vol}_d(I\setminus D)\right|\le K_{\mathrm{lat}}h_q, h q d # F q 0 − Vol d ( I ∖ D ) ≤ K lat h q ,
and
∣ h q d ∑ u ∈ F q 0 P i ( u ) − ∫ I ∖ D P i ( u ) d u ∣ ≤ K l a t h q ( 1 ≤ i ≤ m ) . \left|h_q^d\sum_{u\in F_q^0}P_i(u)-\int_{I\setminus D}P_i(u)\,du\right|\le K_{\mathrm{lat}}h_q\qquad(1\le i\le m). h q d u ∈ F q 0 ∑ P i ( u ) − ∫ I ∖ D P i ( u ) d u ≤ K lat h q ( 1 ≤ i ≤ m ) .
Multiplying by q = h q − d q=h_q^{-d} q = h q − d gives an error at most K l a t q 1 − 1 / d K_{\mathrm{lat}}q^{1-1/d} K lat q 1 − 1/ d . For every q q q sufficiently large that K l a t q − 1 / d ≤ ε K_{\mathrm{lat}}q^{-1/d}\le\varepsilon K lat q − 1/ d ≤ ε , we have F q 0 ∈ F q , ε F_q^0\in\mathcal{F}_{q,\varepsilon} F q 0 ∈ F q , ε and
sep ( F q 0 ) ≥ h q , dist ( F q 0 , D ) ≥ h q / 4. \operatorname{sep}(F_q^0)\ge h_q,\qquad\operatorname{dist}(F_q^0,D)\ge h_q/4. sep ( F q 0 ) ≥ h q , dist ( F q 0 , D ) ≥ h q /4.
Set
Λ L 0 : = L ( F q 0 ∪ A ( q , F q 0 ) ) ⊂ L I . \Lambda_L^0:=L(F_q^0\cup A(q,F_q^0))\subset LI. Λ L 0 := L ( F q 0 ∪ A ( q , F q 0 )) ⊂ L I .
The configuration Λ L 0 \Lambda_L^0 Λ L 0 has q q q points and the exact polynomial moments required in Lemma 3.6. The estimates of Proposition 3.5 and the two preceding separation bounds imply
sep ( Λ L 0 ) ≥ σ ∗ ρ − 1 / d , σ ∗ : = min { σ d , 1 / 4 } . \operatorname{sep}(\Lambda_L^0)\ge\sigma_*\rho^{-1/d},\qquad\sigma_*:=\min\{\sigma_d,1/4\}. sep ( Λ L 0 ) ≥ σ ∗ ρ − 1/ d , σ ∗ := min { σ d , 1/4 } .
Let K L K_L K L be the constant in Lemma 3.6.
Large cubes and their cores. For 0 < ζ < 1 / 4 0<\zeta<1/4 0 < ζ < 1/4 and integers ℓ > ℓ ′ \ell>\ell' ℓ > ℓ ′ with ℓ − ℓ ′ \ell-\ell' ℓ − ℓ ′ even, put S : = ℓ L S:=\ell L S := ℓ L and S ′ : = ℓ ′ L S':=\ell'L S ′ := ℓ ′ L . Suppose that
ζ ≤ 1 − ( S ′ / S ) d ≤ 2 ζ . \zeta\le1-(S'/S)^d\le2\zeta. ζ ≤ 1 − ( S ′ / S ) d ≤ 2 ζ .
In particular, S ′ ≥ S / 2 S'\ge S/2 S ′ ≥ S /2 . Write 1 : = ( 1 , … , 1 ) \mathbf{1}:=(1,\ldots,1) 1 := ( 1 , … , 1 ) . For Q = u + S I Q=u+SI Q = u + S I , define
Q i n : = u + S − S ′ 2 1 + S ′ I , Q o u t : = Q ∖ Q i n , Q^{\mathrm{in}}:=u+\frac{S-S'}{2}\mathbf{1}+S'I,\qquad Q^{\mathrm{out}}:=Q\setminus Q^{\mathrm{in}}, Q in := u + 2 S − S ′ 1 + S ′ I , Q out := Q ∖ Q in ,
L ( Q ) : = { u + S − S ′ 2 1 + L k + L I : k ∈ { 0 , … , ℓ ′ − 1 } d } , V o u t : = S d − ( S ′ ) d . \mathcal{L}(Q):=\left\{u+\frac{S-S'}{2}\mathbf{1}+Lk+LI:k\in\{0,\ldots,\ell'-1\}^d\right\},\qquad V_{\mathrm{out}}:=S^d-(S')^d. L ( Q ) := { u + 2 S − S ′ 1 + L k + L I : k ∈ { 0 , … , ℓ ′ − 1 } d } , V out := S d − ( S ′ ) d .
We call Q Q Q an S S S -cube, Q i n Q^{\mathrm{in}} Q in its core, Q o u t Q^{\mathrm{out}} Q out its outer shell, and the members of L ( Q ) \mathcal{L}(Q) L ( Q ) the L L L -cells of its core. Since ℓ − ℓ ′ \ell-\ell' ℓ − ℓ ′ is even, the cells in L ( Q ) \mathcal{L}(Q) L ( Q ) belong to the tiling { v + L I : v ∈ u + L Z d } \{v+L I:v\in u+L\mathbb{Z}^{d}\} { v + L I : v ∈ u + L Z d } . They partition Q i n Q^{\mathrm{in}} Q in , and M : = # L ( Q ) = ( S ′ / L ) d M:=\#\mathcal{L}(Q)=(S'/L)^d M := # L ( Q ) = ( S ′ / L ) d .
For r ≥ S 2 r\ge S^2 r ≥ S 2 , choose an r r r -separated cocompact Borel cross-section C r ⊂ X 0 C_r\subset X_0 C r ⊂ X 0 and its Voronoi cells W c W_c W c as defined in (5) . The relation { ( x , u ) ∈ X 0 × R d : u . x ∈ C r } \{(x,u)\in X_0\times\mathbb{R}^d:u.x\in C_r\} {( x , u ) ∈ X 0 × R d : u . x ∈ C r } is Borel and has countable sections. Lusin–Novikov [24 Theorem 18.10] gives Borel sets D j ⊂ X 0 D_j\subset X_0 D j ⊂ X 0 and Borel maps u j : D j → R d u_j:D_j\to\mathbb{R}^d u j : D j → R d whose disjoint graphs cover the relation; these maps will be used below to establish measurability of the modified realization. Let ν r \nu_r ν r be the transverse measure normalized by the cross-section formula [6 Theorem 1.14] ,
∫ X H d μ = ∫ C r ∫ W c H ( t . c ) d t d ν r ( c ) . \int_X H\,\mathrm{d}\mu=\int_{C_r}\int_{W_c}H(t.c)\,\mathrm{d}t\,\mathrm{d}\nu_r(c). ∫ X H d μ = ∫ C r ∫ W c H ( t . c ) d t d ν r ( c ) .
The cross-section mass-transport formula [6 Theorem 1.14] gives
∫ C r N W c ( κ ( c ) ) d ν r ( c ) = ρ . \int_{C_r}N_{W_c}(\kappa(c))\,\mathrm{d}\nu_r(c)=\rho. ∫ C r N W c ( κ ( c )) d ν r ( c ) = ρ .
For s > 0 s>0 s > 0 , set
W c − s : = { u ∈ W c : dist ( u , ∂ W c ) > s } . W_c^{-s}:=\{u\in W_c:\operatorname{dist}(u,\partial W_c)>s\}. W c − s := { u ∈ W c : dist ( u , ∂ W c ) > s } .
The inclusion int B r / 2 ⊂ W c \operatorname{int}B_{r/2}\subset W_c int B r /2 ⊂ W c and convexity give, for 0 < s < r / 2 0<s<r/2 0 < s < r /2 ,
Vol d ( W c ∖ W c − s ) Vol d ( W c ) ≤ 2 d s r . \frac{\operatorname{Vol}_d(W_c\setminus W_c^{-s})}{\operatorname{Vol}_d(W_c)}\le\frac{2ds}{r}. Vol d ( W c ) Vol d ( W c ∖ W c − s ) ≤ r 2 d s .
For every z ∈ [ 0 , S ) d z\in[0,S)^d z ∈ [ 0 , S ) d , each point not covered by the cubes u + S I ⊂ W c u+SI\subset W_c u + S I ⊂ W c , u ∈ z + S Z d u\in z+S\mathbb{Z}^d u ∈ z + S Z d , lies within distance d S \sqrt{d}\,S d S of ∂ W c \partial W_c ∂ W c . Consequently,
Vol d ( W c ∖ ⋃ u ∈ z + S Z d u + S I ⊂ W c ( u + S I ) ) ≤ C d S r Vol d ( W c ) . (33) \operatorname{Vol}_d\left(W_c\setminus\bigcup_{\substack{u\in z+S\mathbb{Z}^d\\u+SI\subset W_c}}(u+SI)\right)\le C_d\frac{S}{r}\operatorname{Vol}_d(W_c). \tag*{(33)} Vol d W c ∖ u ∈ z + S Z d u + S I ⊂ W c ⋃ ( u + S I ) ≤ C d r S Vol d ( W c ) . (33)
For z ∈ [ 0 , S ) d z\in[0,S)^d z ∈ [ 0 , S ) d and c ∈ C r c\in C_r c ∈ C r , let
Q z ( c ) : = { u + S I : u ∈ z + S Z d , u + S I ⊂ W c } . Q_z(c):=\{u+SI:u\in z+S\mathbb{Z}^d,\ u+SI\subset W_c\}. Q z ( c ) := { u + S I : u ∈ z + S Z d , u + S I ⊂ W c } .
For Q = u + [ 0 , S ) d Q=u+[0,S)^d Q = u + [ 0 , S ) d and w ∈ R d w\in\mathbb{R}^d w ∈ R d ,
sup t ∈ Q ⟨ t , w ⟩ = ⟨ u , w ⟩ + S ∑ i = 1 d max { w i , 0 } . (34) \sup_{t\in Q}\langle t,w\rangle=\langle u,w\rangle+S\sum_{i=1}^{d}\max\{w_i,0\}. \tag*{(34)} t ∈ Q sup ⟨ t , w ⟩ = ⟨ u , w ⟩ + S i = 1 ∑ d max { w i , 0 } . (34)
This supremum is attained on Q Q Q exactly when every w i ≤ 0 w_i\le0 w i ≤ 0 . The inclusion of Q Q Q in a strict or non-strict affine half-space is therefore a Borel condition. For each j j j with c ∈ D j c\in D_j c ∈ D j , the half-space defined by u j ( c ) u_j(c) u j ( c ) depends Borelly on c c c ; intersecting these countably many conditions shows that ( c , z ) ↦ Q z ( c ) (c,z)\mapsto Q_z(c) ( c , z ) ↦ Q z ( c ) is Borel.
For Q ∈ Q z ( c ) Q\in Q_z(c) Q ∈ Q z ( c ) , put
n Q : = # ( Λ c ∩ Q ) , k Q : = n Q − q M . n_Q:=\#(\Lambda_c\cap Q),\qquad k_Q:=n_Q-qM. n Q := # ( Λ c ∩ Q ) , k Q := n Q − q M .
We call Q Q Q count-compatible if
∣ n Q − ρ S d ∣ ≤ ρ 4 V o u t . (35) \lvert n_Q-\rho S^d\rvert\le\frac{\rho}{4}V_{\mathrm{out}}. \tag*{(35)} ∣ n Q − ρ S d ∣ ≤ 4 ρ V out . (35)
Replacement in each large cube. For Q ∈ Q z ( c ) Q\in Q_z(c) Q ∈ Q z ( c ) , the definitions of n Q n_Q n Q , k Q k_Q k Q , and M M M give
k Q = ρ V o u t + ( n Q − ρ S d ) . k_Q=\rho V_{\mathrm{out}}+(n_Q-\rho S^d). k Q = ρ V out + ( n Q − ρ S d ) .
If Q Q Q is count-compatible, then (35) gives
3 4 ρ V o u t ≤ k Q ≤ 5 4 ρ V o u t . \frac{3}{4}\rho V_{\mathrm{out}}\le k_Q\le\frac{5}{4}\rho V_{\mathrm{out}}. 4 3 ρ V out ≤ k Q ≤ 4 5 ρ V out .
Write Q = u + S I Q=u+SI Q = u + S I , and define the finite set of lattice sites
G Q : = ( u + ( 2 ρ ) − 1 / d Z d ) ∩ Q o u t . G_Q := \left(u+(2\rho)^{-1/d}\mathbb{Z}^d\right)\cap Q^{\mathrm{out}}. G Q := ( u + ( 2 ρ ) − 1/ d Z d ) ∩ Q out .
It has 2 ρ V o u t + O ( S d − 1 ) 2\rho V_{\mathrm{out}}+O(S^{d-1}) 2 ρ V out + O ( S d − 1 ) points. Since ζ S d ≤ V o u t ≤ 2 ζ S d \zeta S^d\le V_{\mathrm{out}}\le2\zeta S^d ζ S d ≤ V out ≤ 2 ζ S d , for all sufficiently large S S S ,
# G Q ≥ 5 4 ρ V o u t ≥ k Q ( Q count-compatible ) . \#G_Q\ge\frac{5}{4}\rho V_{\mathrm{out}}\ge k_Q \qquad(Q\text{ count-compatible}). # G Q ≥ 4 5 ρ V out ≥ k Q ( Q count-compatible ) .
For a count-compatible Q Q Q , let G Q ( k Q ) G_Q(k_Q) G Q ( k Q ) be the first k Q k_Q k Q points of G Q G_Q G Q in lexicographic order.
Recall that an L L L -cell P P P is admissible for Λ c \Lambda_c Λ c precisely when F L , P ( Λ c ) ∈ F q , ε F_{L,P}(\Lambda_c)\in\mathcal{F}_{q,\varepsilon} F L , P ( Λ c ) ∈ F q , ε and # ( Λ c ∩ D P ) ≤ 2 ρ ϑ L d \#(\Lambda_c\cap D_P)\le2\rho\vartheta L^d # ( Λ c ∩ D P ) ≤ 2 ρϑ L d . For each P = v + L I ∈ L ( Q ) P=v+LI\in\mathcal{L}(Q) P = v + L I ∈ L ( Q ) , define
T P ( Λ c ) : = { ( Λ c ∩ ( P ∖ int D P ) ) ∪ ( v + L A ( q , F L , P ( Λ c ) ) ) , P admissible for Λ c , v + Λ L 0 , P not admissible for Λ c . \mathcal{T}_P(\Lambda_c):=
\begin{cases}
(\Lambda_c\cap(P\setminus\operatorname{int}D_P))\cup\left(v+L A(q,F_{L,P}(\Lambda_c))\right), & P\text{ admissible for }\Lambda_c,\\
v+\Lambda_L^0, & P\text{ not admissible for }\Lambda_c.
\end{cases} T P ( Λ c ) := { ( Λ c ∩ ( P ∖ int D P )) ∪ ( v + L A ( q , F L , P ( Λ c )) ) , v + Λ L 0 , P admissible for Λ c , P not admissible for Λ c .
By (25) , the first case has the required polynomial moments in P P P . The definition of Λ L 0 \Lambda_L^0 Λ L 0 gives the same identities in the second case. Thus, for each P ∈ L ( Q ) P\in\mathcal{L}(Q) P ∈ L ( Q ) ,
∑ u ∈ T P ( Λ c ) H ( u ) = ρ ∫ P H ( u ) d u ( H ∈ P p − 1 ) . (36) \sum_{u\in\mathcal{T}_P(\Lambda_c)}H(u)=\rho\int_P H(u)\,du\qquad(H\in\mathcal{P}_{p-1}). \tag*{(36)} u ∈ T P ( Λ c ) ∑ H ( u ) = ρ ∫ P H ( u ) d u ( H ∈ P p − 1 ) . (36)
In particular, # T P ( Λ c ) = q \#\mathcal{T}_P(\Lambda_c)=q # T P ( Λ c ) = q .
For each z ∈ [ 0 , S ) d z\in[0,S)^d z ∈ [ 0 , S ) d , define Λ c , z ′ \Lambda'_{c,z} Λ c , z ′ on every Q ∈ Q z ( c ) Q\in\mathcal{Q}_z(c) Q ∈ Q z ( c ) by
Λ c , z ′ ∩ Q : = { ⋃ P ∈ L ( Q ) T P ( Λ c ) ∪ G Q ( k Q ) , Q count-compatible , Λ c ∩ Q , Q not count-compatible . \Lambda'_{c,z}\cap Q:=
\begin{cases}
\bigcup_{P\in\mathcal{L}(Q)}\mathcal{T}_P(\Lambda_c)\cup G_Q(k_Q), & Q\text{ count-compatible},\\
\Lambda_c\cap Q, & Q\text{ not count-compatible}.
\end{cases} Λ c , z ′ ∩ Q := { ⋃ P ∈ L ( Q ) T P ( Λ c ) ∪ G Q ( k Q ) , Λ c ∩ Q , Q count-compatible , Q not count-compatible .
Outside these S S S -cubes, put Λ c , z ′ = Λ c \Lambda'_{c,z}=\Lambda_c Λ c , z ′ = Λ c . The union defining Λ c , z ′ ∩ Q \Lambda'_{c,z}\cap Q Λ c , z ′ ∩ Q is disjoint. For each count-compatible Q Q Q ,
# ( Λ c , z ′ ∩ Q ) = q M + k Q = n Q . \#(\Lambda'_{c,z}\cap Q)=qM+k_Q=n_Q. # ( Λ c , z ′ ∩ Q ) = q M + k Q = n Q .
Put Λ c , z W : = Λ c , z ′ ∩ W c \Lambda^W_{c,z}:=\Lambda'_{c,z}\cap W_c Λ c , z W := Λ c , z ′ ∩ W c . The global realization is the counting measure with support
supp κ z ( x ) : = ⋃ u ∈ ( C r ) x ( u + Λ u . x , z W ) . \operatorname{supp}\kappa_z(x):=\bigcup_{u\in(C_r)_x}(u+\Lambda^W_{u.x,z}). supp κ z ( x ) := u ∈ ( C r ) x ⋃ ( u + Λ u . x , z W ) .
The sets u + W u . x u+W_{u.x} u + W u . x partition R d \mathbb{R}^d R d and are locally finite, so the displayed union is locally finite. The maps c ↦ Λ c , z W c\mapsto\Lambda^W_{c,z} c ↦ Λ c , z W are Borel. The inclusion of cubes in Q z ( c ) \mathcal{Q}_z(c) Q z ( c ) is Borel by (34) and the calculation following it; count-compatibility is the inequality (35) , and admissibility is determined by the two conditions in its definition. The moment-completion map of Proposition 3.5 and the lexicographic choice from G Q G_Q G Q are Borel. Using the return maps u j u_j u j , for each bounded Borel set D D D we have
N D ( κ z ( x ) ) = ∑ j : x ∈ D j N D − u j ( x ) ( δ Λ u j ( x ) . x , z W ) . N_D(\kappa_z(x))=\sum_{j:x\in D_j}N_{D-u_j(x)}\left(\delta_{\Lambda^W_{u_j(x).x,z}}\right). N D ( κ z ( x )) = j : x ∈ D j ∑ N D − u j ( x ) ( δ Λ u j ( x ) . x , z W ) .
The summands are Borel, so x ↦ κ z ( x ) x\mapsto\kappa_z(x) x ↦ κ z ( x ) is Borel. For v ∈ R d v\in\mathbb{R}^d v ∈ R d ,
( C r ) v . x = ( C r ) x − v , ( u − v ) . ( v . x ) = u . x , (C_r)_{v.x}=(C_r)_x-v,\qquad(u-v).(v.x)=u.x, ( C r ) v . x = ( C r ) x − v , ( u − v ) . ( v . x ) = u . x ,
which gives κ z ( v . x ) = θ v κ z ( x ) \kappa_z(v.x)=\theta_v\kappa_z(x) κ z ( v . x ) = θ v κ z ( x ) . The equalities # ( Λ c , z ′ ∩ Q ) = # ( Λ c ∩ Q ) \#(\Lambda'_{c,z}\cap Q)=\#(\Lambda_c\cap Q) # ( Λ c , z ′ ∩ Q ) = # ( Λ c ∩ Q ) for modified cubes and Λ c , z ′ = Λ c \Lambda'_{c,z}=\Lambda_c Λ c , z ′ = Λ c elsewhere give
# Λ c , z W = # ( Λ c ∩ W c ) , ∫ C r # Λ c , z W d ν r ( c ) = ρ . \#\Lambda^W_{c,z}=\#(\Lambda_c\cap W_c),\qquad\int_{C_r}\#\Lambda^W_{c,z}\,d\nu_r(c)=\rho. # Λ c , z W = # ( Λ c ∩ W c ) , ∫ C r # Λ c , z W d ν r ( c ) = ρ .
Thus κ z \kappa_z κ z is a translation-equivariant simple counting-measure realization of intensity ρ \rho ρ .
For each z ∈ [ 0 , S ) d z\in[0,S)^d z ∈ [ 0 , S ) d , define
κ z + ( x ) : = δ supp κ z ( x ) ∖ supp κ ( x ) . \kappa_z^+(x):=\delta_{\operatorname{supp}\kappa_z(x)\setminus\operatorname{supp}\kappa(x)}. κ z + ( x ) := δ supp κ z ( x ) ∖ supp κ ( x ) .
Lemma 3.7 (Uniform inserted-point bound). For every n ≥ 1 n \ge1 n ≥ 1 there is a constant D n < ∞ D_n < \infty D n < ∞ , depending only on d d d , ρ \rho ρ , n n n , such that, for every choice of D D D , p p p , q q q , L L L , ϑ \vartheta ϑ , γ \gamma γ , ζ \zeta ζ , S S S , r r r satisfying the preceding hypotheses and every z ∈ [ 0 , S ) d z \in[0,S)^d z ∈ [ 0 , S ) d ,
N Q n ( κ z + ( x ) ) ≤ D n ( x ∈ X 0 ) . (37) N_{Q_n}\left(\kappa_z^+(x)\right) \le D_n \qquad(x \in X_0). \tag*{(37)} N Q n ( κ z + ( x ) ) ≤ D n ( x ∈ X 0 ) . (37) In particular, κ z ∈ R ρ , 2 a ( X , μ ) \kappa_z \in R^a_{\rho,2}(X,\mu) κ z ∈ R ρ , 2 a ( X , μ ) .
Proof. An L L L -cell P P P is admissible here if F L , P ( Λ c ) ∈ F q , ϵ F_{L,P}(\Lambda_c) \in\mathcal{F}_{q,\epsilon} F L , P ( Λ c ) ∈ F q , ϵ and # ( Λ c ∩ D P ) ≤ 2 ρ ϑ L d \#(\Lambda_c \cap D_P) \le2\rho\vartheta L^d # ( Λ c ∩ D P ) ≤ 2 ρϑ L d . The points inserted in an admissible L L L -cell, in an L L L -cell where either inequality fails, and in an outer shell have respective separation bounds σ d ρ − 1 / d \sigma_d\rho^{-1/d} σ d ρ − 1/ d , σ ∗ ρ − 1 / d \sigma_*\rho^{-1/d} σ ∗ ρ − 1/ d , and ( 2 ρ ) − 1 / d (2\rho)^{-1/d} ( 2 ρ ) − 1/ d . Set s 0 : = ρ − 1 / d min { σ ∗ , 2 − 1 / d } s_0 := \rho^{-1/d}\min\{\sigma_*,2^{-1/d}\} s 0 := ρ − 1/ d min { σ ∗ , 2 − 1/ d } . The core L L L -cells have disjoint interiors, as do the S S S -cubes containing the outer shells. Disjoint axis-parallel cubes of side T T T intersecting Q n Q_n Q n are contained in its coordinatewise T T T -enlargement, so their number is at most ( n / T + 2 ) d (n/T+2)^d ( n / T + 2 ) d . Since S ≥ L ≥ ρ − 1 / d S \ge L \ge\rho^{-1/d} S ≥ L ≥ ρ − 1/ d , at most M n : = ( ⌈ n ρ 1 / d ⌉ + 2 ) d M_n := (\lceil n\rho^{1/d}\rceil+2)^d M n := (⌈ n ρ 1/ d ⌉ + 2 ) d cubes of either type intersect Q n Q_n Q n . Let J n J_n J n bound the number of points in an s 0 s_0 s 0 -separated subset of Q n Q_n Q n . Then (3.32) holds with D n : = 2 M n J n D_n := 2M_nJ_n D n := 2 M n J n . The constants σ d , σ ∗ \sigma_d,\sigma_* σ d , σ ∗ depend only on d d d . Finally, N Q n ( κ z ) ≤ N Q n ( κ ) + D n N_{Q_n}(\kappa_z) \le N_{Q_n}(\kappa)+D_n N Q n ( κ z ) ≤ N Q n ( κ ) + D n . Since κ \kappa κ has local second moments and κ z \kappa_z κ z has intensity ρ \rho ρ , κ z ∈ R ρ , 2 a ( X , μ ) \kappa_z \in R^a_{\rho,2}(X,\mu) κ z ∈ R ρ , 2 a ( X , μ ) .
For each z ∈ [ 0 , S ) d z \in[0,S)^d z ∈ [ 0 , S ) d , define
e 1 ( z ) : = ∫ C r ∑ Q ∈ Q z ( c ) Q not count-compatible S d d ν r ( c ) , e_1(z):=\int_{C_r}\sum_{\substack{Q\in\mathcal{Q}_z(c)\\Q\text{ not count-compatible}}} S^d\,d\nu_r(c), e 1 ( z ) := ∫ C r Q ∈ Q z ( c ) Q not count-compatible ∑ S d d ν r ( c ) ,
e 2 ( z ) : = ∫ C r ∑ Q ∈ Q z ( c ) ∑ P ∈ L ( Q ) P not admissible for Λ c L d d ν r ( c ) , e_2(z):=\int_{C_r}\sum_{Q\in\mathcal{Q}_z(c)}\sum_{\substack{P\in\mathcal{L}(Q)\\P\text{ not admissible for }\Lambda_c}} L^d\,d\nu_r(c), e 2 ( z ) := ∫ C r Q ∈ Q z ( c ) ∑ P ∈ L ( Q ) P not admissible for Λ c ∑ L d d ν r ( c ) ,
e 3 ( z ) : = ∫ C r ∑ Q ∈ Q z ( c ) ∑ P ∈ L ( Q ) P not admissible for Λ c # ( Λ c ∩ P ) d ν r ( c ) , e_3(z):=\int_{C_r}\sum_{Q\in\mathcal{Q}_z(c)}\sum_{\substack{P\in\mathcal{L}(Q)\\P\text{ not admissible for }\Lambda_c}} \#(\Lambda_c\cap P)\,d\nu_r(c), e 3 ( z ) := ∫ C r Q ∈ Q z ( c ) ∑ P ∈ L ( Q ) P not admissible for Λ c ∑ # ( Λ c ∩ P ) d ν r ( c ) ,
e 4 ( z ) : = ∫ C r ∑ Q ∈ Q z ( c ) # ( Λ c ∩ Q o u t ) d ν r ( c ) . e_4(z):=\int_{C_r}\sum_{Q\in\mathcal{Q}_z(c)}\#(\Lambda_c\cap Q^{\mathrm{out}})\,d\nu_r(c). e 4 ( z ) := ∫ C r Q ∈ Q z ( c ) ∑ # ( Λ c ∩ Q out ) d ν r ( c ) .
Set
b S , ζ : = μ { x : ∣ N S I ( κ ( x ) ) − ρ S d ∣ > ρ ζ S d 4 } . b_{S,\zeta}:=\mu\left\{x:\left|N_{\mathrm{SI}}(\kappa(x))-\rho S^d\right|>\frac{\rho\zeta S^d}{4}\right\}. b S , ζ := μ { x : N SI ( κ ( x )) − ρ S d > 4 ρζ S d } .
By Lemma 3.4 , b S , ζ → 0 b_{S,\zeta}\to0 b S , ζ → 0 as S → ∞ S\to\infty S → ∞ for each fixed ζ > 0 \zeta>0 ζ > 0 .
Lemma 3.8 (Choice of grid shift). Let γ > 0 \gamma>0 γ > 0 and assume
μ ( E L ) < γ , L − d ∫ E L N L I ( κ ( x ) ) d μ ( x ) < γ . \mu(\mathcal{E}_L)<\gamma,\qquad L^{-d}\int_{\mathcal{E}_L}N_{\mathrm{LI}}(\kappa(x))\,d\mu(x)<\gamma. μ ( E L ) < γ , L − d ∫ E L N LI ( κ ( x )) d μ ( x ) < γ . Then there is z ∈ [ 0 , S ) d z\in[0,S)^d z ∈ [ 0 , S ) d such that
e 1 ( z ) ≤ 4 b S , ζ , e 2 ( z ) + e 3 ( z ) ≤ 8 γ , e 4 ( z ) ≤ 8 ρ ζ . e_1(z)\le4b_{S,\zeta},\qquad e_2(z)+e_3(z)\le8\gamma,\qquad e_4(z)\le8\rho\zeta. e 1 ( z ) ≤ 4 b S , ζ , e 2 ( z ) + e 3 ( z ) ≤ 8 γ , e 4 ( z ) ≤ 8 ρζ .
Proof. For T = S T=S T = S or L L L , let u T , z ( t ) ∈ [ 0 , T ) d u_{T,z}(t)\in[0,T)^d u T , z ( t ) ∈ [ 0 , T ) d be the position of t t t relative to the lower corner of its cell in the tiling { v + [ 0 , T ) d : v ∈ z + T Z d } \{v+[0,T)^d:v\in z+T\mathbb{Z}^d\} { v + [ 0 , T ) d : v ∈ z + T Z d } . For every nonnegative Borel H H H on X X X , the cross-section formula, translation invariance of μ \mu μ , and averaging z z z modulo T T T give
1 T d ∫ [ 0 , T ) d ∫ C r ∫ W c H ( ( t − u T , z ( t ) ) . c ) d t d ν r ( c ) d z = ∫ X H d μ . (38) \frac{1}{T^d}\int_{[0,T)^d}\int_{C_r}\int_{W_c}H((t-u_{T,z}(t)).c)\,dt\,d\nu_r(c)\,dz=\int_X H\,d\mu. \tag*{(38)} T d 1 ∫ [ 0 , T ) d ∫ C r ∫ W c H (( t − u T , z ( t )) . c ) d t d ν r ( c ) d z = ∫ X H d μ . (38)
Let P T , z \mathcal{P}_{T,z} P T , z be the grid of half-open T T T -cubes with lower corners in z + T Z d z+T\mathbb{Z}^{d} z + T Z d , and let a ( P ) a(P) a ( P ) denote the lower corner of P P P . For every nonnegative Borel H H H ,
∑ P ∈ P T , z P ⊂ W c T d H ( a ( P ) . c ) ≤ ∫ W c H ( t − u T , z ( t ) . c ) d t . (39) \sum_{\substack{P\in\mathcal{P}_{T,z}\\P\subset W_{c}}} T^{d}H(a(P).c)\le\int_{W_{c}}H(t-u_{T,z}(t).c)\,\mathrm{d}t. \tag*{(39)} P ∈ P T , z P ⊂ W c ∑ T d H ( a ( P ) . c ) ≤ ∫ W c H ( t − u T , z ( t ) . c ) d t . (39)
The integrand equals H ( a ( P ) . c ) H(a(P).c) H ( a ( P ) . c ) on P P P . The inequality also holds when the sum is restricted to fewer cells. For e 2 , e 3 e_{2},e_{3} e 2 , e 3 , the lower corner of a core L L L -cell differs from that of its containing S S S -cube by ℓ − ℓ ′ 2 L 1 + L k \frac{\ell-\ell'}{2}L_{1}+Lk 2 ℓ − ℓ ′ L 1 + L k , k ∈ Z d k\in\mathbb{Z}^{d} k ∈ Z d . Since ℓ − ℓ ′ \ell-\ell' ℓ − ℓ ′ is even, core L L L -cells belong to the z + L Z d z+L\mathbb{Z}^{d} z + L Z d grid. Applying (38) with T = S T=S T = S and
H ( x ) : = 1 { ∣ N S I ( κ ( x ) ) − ρ S d ∣ > ρ ζ S d / 4 } H(x):=\mathbf{1}_{\{\lvert N_{SI}(\kappa(x))-\rho S^{d}\rvert>\rho\zeta S^{d}/4\}} H ( x ) := 1 {∣ N S I ( κ ( x )) − ρ S d ∣ > ρζ S d /4 }
gives the bound for e 1 e_{1} e 1 . Since S = ℓ L S=\ell L S = ℓ L , averaging z z z over [ 0 , S ) d [0,S)^{d} [ 0 , S ) d also averages its residue modulo L L L uniformly. Taking T = L T=L T = L and then
H ( x ) : = 1 E L ( x ) , H ( x ) : = L − d N L I ( κ ( x ) ) 1 E L ( x ) H(x):=\mathbf{1}_{\mathcal{E}_{L}(x)},\qquad H(x):=L^{-d}N_{LI}(\kappa(x))\mathbf{1}_{\mathcal{E}_{L}(x)} H ( x ) := 1 E L ( x ) , H ( x ) := L − d N L I ( κ ( x )) 1 E L ( x )
gives the bounds for e 2 e_{2} e 2 and e 3 e_{3} e 3 . Thus
1 S d ∫ [ 0 , S ) d e 1 ( z ) d z ≤ b S , ζ , \frac{1}{S^{d}}\int_{[0,S)^{d}}e_{1}(z)\,\mathrm{d}z\le b_{S,\zeta}, S d 1 ∫ [ 0 , S ) d e 1 ( z ) d z ≤ b S , ζ ,
1 S d ∫ [ 0 , S ) d e 2 ( z ) d z ≤ μ ( E L ) < γ , \frac{1}{S^{d}}\int_{[0,S)^{d}}e_{2}(z)\,\mathrm{d}z\le\mu(\mathcal{E}_{L})<\gamma, S d 1 ∫ [ 0 , S ) d e 2 ( z ) d z ≤ μ ( E L ) < γ ,
1 S d ∫ [ 0 , S ) d e 3 ( z ) d z ≤ L − d ∫ E L N L I ( κ ( x ) ) d μ ( x ) < γ . \frac{1}{S^{d}}\int_{[0,S)^{d}}e_{3}(z)\,\mathrm{d}z\le L^{-d}\int_{\mathcal{E}_{L}}N_{LI}(\kappa(x))\,\mathrm{d}\mu(x)<\gamma. S d 1 ∫ [ 0 , S ) d e 3 ( z ) d z ≤ L − d ∫ E L N L I ( κ ( x )) d μ ( x ) < γ .
For fixed c c c and u ∈ Λ c ∩ W c u\in\Lambda_{c}\cap W_{c} u ∈ Λ c ∩ W c , the proportion of shifts z ∈ [ 0 , S ) d z\in[0,S)^{d} z ∈ [ 0 , S ) d for which u u u belongs to the complement of a core in the full S S S -grid is 1 − ( S ′ / S ) d 1-(S'/S)^{d} 1 − ( S ′ / S ) d . Integrating first in z z z and then in c c c , the mass-transport formula gives
1 S d ∫ [ 0 , S ) d e 4 ( z ) d z ≤ ρ ( 1 − ( S ′ / S ) d ) ≤ 2 ρ ζ . \frac{1}{S^{d}}\int_{[0,S)^{d}}e_{4}(z)\,\mathrm{d}z\le\rho\left(1-(S'/S)^{d}\right)\le2\rho\zeta. S d 1 ∫ [ 0 , S ) d e 4 ( z ) d z ≤ ρ ( 1 − ( S ′ / S ) d ) ≤ 2 ρζ .
Markov’s inequality gives a shift z ∈ [ 0 , S ) d z\in[0,S)^{d} z ∈ [ 0 , S ) d satisfying
e 1 ( z ) ≤ 4 b S , ζ , e 2 ( z ) + e 3 ( z ) ≤ 8 γ , e 4 ( z ) ≤ 8 ρ ζ . e_{1}(z)\le4b_{S,\zeta},\qquad e_{2}(z)+e_{3}(z)\le8\gamma,\qquad e_{4}(z)\le8\rho\zeta. e 1 ( z ) ≤ 4 b S , ζ , e 2 ( z ) + e 3 ( z ) ≤ 8 γ , e 4 ( z ) ≤ 8 ρζ .
Indeed, each of the three strict reverse inequalities holds on a set of shifts of measure at most S d / 4 S^{d}/4 S d /4 . □
Lemma 3.9 (Rooted approximation). *For every ε ∗ > 0 \varepsilon_{*}>0 ε ∗ > 0 there is δ ∗ = δ ∗ ( κ , ε ∗ ) > 0 \delta_{*}=\delta_{*}(\kappa,\varepsilon_{*})>0 δ ∗ = δ ∗ ( κ , ε ∗ ) > 0 with the following property. Assume that q ≥ q 0 q\ge q_{0} q ≥ q 0 , K l a t q − 1 / d ≤ ε K_{\mathrm{lat}}q^{-1/d}\le\varepsilon K lat q − 1/ d ≤ ε , 0 < ζ < 1 / 4 0<\zeta<1/4 0 < ζ < 1/4 , ℓ > ℓ ′ \ell>\ell' ℓ > ℓ ′ with ℓ − ℓ ′ \ell-\ell' ℓ − ℓ ′ even, ζ ≤ 1 − ( S ′ / S ) d ≤ 2 ζ \zeta\le1-(S'/S)^{d}\le2\zeta ζ ≤ 1 − ( S ′ / S ) d ≤ 2 ζ , r ≥ S 2 r\ge S^{2} r ≥ S 2 , and # G Q ≥ 5 4 ρ V o u t \#G_{Q}\ge\frac{5}{4}\rho V_{\mathrm{out}} # G Q ≥ 4 5 ρ V out for every S S S -cube Q = u + S I Q=u+SI Q = u + S I . Suppose that the hypotheses of Lemma 3.8 hold and z z z satisfies its three inequalities. If
ϑ + γ + ζ + S r < δ ∗ , (40) \vartheta+\gamma+\zeta+\frac{S}{r}<\delta_{*}, \tag*{(40)} ϑ + γ + ζ + r S < δ ∗ , (40) then d r o o t , 2 ( κ z , κ ) < ε ∗ d_{\mathrm{root},2}(\kappa_{z},\kappa)<\varepsilon_{*} d root , 2 ( κ z , κ ) < ε ∗ . The choice of δ ∗ \delta_{*} δ ∗ is independent of p p p , D D D , q q q , L L L , S S S , r r r , z z z , ϑ \vartheta ϑ , γ \gamma γ , ζ \zeta ζ , the cross-section C r C_{r} C r , and its Borel order.*
Proof. Put κ ′ : = κ z \kappa':=\kappa_{z} κ ′ := κ z , Λ c ′ : = Λ c , z ′ \Lambda'_{c}:=\Lambda'_{c,z} Λ c ′ := Λ c , z ′ , κ + : = κ z + \kappa^{+}:=\kappa^{+}_{z} κ + := κ z + , and
κ − ( x ) : = δ supp κ ( x ) ∖ supp κ ′ ( x ) . \kappa^{-}(x):=\delta_{\operatorname{supp}\kappa(x)\setminus\operatorname{supp}\kappa'(x)}. κ − ( x ) := δ supp κ ( x ) ∖ supp κ ′ ( x ) . Write N n ± : = N Q n ∘ κ ± N_{n}^{\pm}:=N_{Q_{n}}\circ\kappa^{\pm} N n ± := N Q n ∘ κ ± and let ρ ± \rho_{\pm} ρ ± be the intensities of κ ± \kappa^{\pm} κ ± . The numbers of points removed and inserted in an admissible core cell P P P satisfy
# ( ( Λ c ∖ Λ c ′ ) ∩ P ) ≤ 2 ρ ϑ L d , # ( ( Λ c ′ ∖ Λ c ) ∩ P ) ≤ 3 2 ρ ϑ L d . \#((\Lambda_{c}\setminus\Lambda'_{c})\cap P)\le2\rho\vartheta L^{d},\qquad
\#((\Lambda'_{c}\setminus\Lambda_{c})\cap P)\le\frac{3}{2}\rho\vartheta L^{d}. # (( Λ c ∖ Λ c ′ ) ∩ P ) ≤ 2 ρϑ L d , # (( Λ c ′ ∖ Λ c ) ∩ P ) ≤ 2 3 ρϑ L d .
For a core L L L -cell P P P that fails either of the admissibility requirements F L , P ( Λ c ) ∈ F q , ε F_{L,P}(\Lambda_c) \in\mathcal{F}_{q,\varepsilon} F L , P ( Λ c ) ∈ F q , ε or # ( Λ c ∩ D P ) ≤ 2 ρ ϑ L d \#(\Lambda_c \cap D_P) \le2\rho\vartheta L^d # ( Λ c ∩ D P ) ≤ 2 ρϑ L d , the total number of removed and inserted points is at most # ( Λ c ∩ P ) + ρ L d \#(\Lambda_c \cap P) + \rho L^d # ( Λ c ∩ P ) + ρ L d . Summing over such cells gives e 3 ( z ) + ρ e 2 ( z ) e_3(z) + \rho e_2(z) e 3 ( z ) + ρ e 2 ( z ) . A modified S S S -cube contributes at most # ( Λ c ∩ Q o u t ) + 5 ρ V o u t / 4 \#(\Lambda_c \cap Q^{\mathrm{out}}) + 5\rho V_{\mathrm{out}}/4 # ( Λ c ∩ Q out ) + 5 ρ V out /4 from its outer shell. Since V o u t / S d ≤ 2 ζ V_{\mathrm{out}}/S^d \le2\zeta V out / S d ≤ 2 ζ , Lemma 3.8 and the cross-section mass-transport identity give
ρ − + ρ + ≤ 7 2 ρ ϑ + e 3 ( z ) + ρ e 2 ( z ) + e 4 ( z ) + 5 2 ρ ζ ≤ 7 2 ρ ϑ + 8 max { 1 , ρ } γ + 21 2 ρ ζ . \begin{aligned}
\rho_- + \rho_+ \le\frac{7}{2}\rho\vartheta+ e_3(z) + \rho e_2(z) + e_4(z) + \frac{5}{2}\rho\zeta\\
&\le\frac{7}{2}\rho\vartheta+ 8\max\{1,\rho\}\gamma+ \frac{21}{2}\rho\zeta.
\end{aligned} ρ − + ρ + ≤ 2 7 ρϑ + e 3 ( z ) + ρ e 2 ( z ) + e 4 ( z ) + 2 5 ρζ ≤ 2 7 ρϑ + 8 max { 1 , ρ } γ + 2 21 ρζ .
In particular, for every fixed n n n ,
∫ X ( N n − + N n + ) d μ ≤ C n ( ϑ + γ + ζ ) . \int_X (N_n^- + N_n^+)\,d\mu\le C_n(\vartheta+\gamma+\zeta). ∫ X ( N n − + N n + ) d μ ≤ C n ( ϑ + γ + ζ ) .
By Lemma 3.7 , N Q n ( κ ′ ) ≤ N Q n ( κ ) + D n N_{Q_n}(\kappa') \le N_{Q_n}(\kappa) + D_n N Q n ( κ ′ ) ≤ N Q n ( κ ) + D n .
It follows that
( N n − ) 2 ≤ N Q n ( κ ) 2 1 { N n − > 0 } , ( N n + ) 2 ≤ D n N n + . (N_n^-)^2 \le N_{Q_n}(\kappa)^2\mathbf{1}_{\{N_n^->0\}}, \qquad(N_n^+)^2 \le D_nN_n^+. ( N n − ) 2 ≤ N Q n ( κ ) 2 1 { N n − > 0 } , ( N n + ) 2 ≤ D n N n + .
As ϑ + γ + ζ → 0 \vartheta+ \gamma+ \zeta\to0 ϑ + γ + ζ → 0 over the allowed parameter choices, the first squared L 2 L^2 L 2 -norm tends to zero by absolute continuity, since μ { N n − > 0 } ≤ ∫ X N n − d μ → 0 \mu\{N_n^- > 0\} \le\int_X N_n^-\,d\mu\to0 μ { N n − > 0 } ≤ ∫ X N n − d μ → 0 . The second squared norm tends to zero by the preceding L 1 L^1 L 1 bound. Thus ∥ N n − ∥ 2 + ∥ N n + ∥ 2 → 0 \|N_n^-\|_2 + \|N_n^+\|_2 \to0 ∥ N n − ∥ 2 + ∥ N n + ∥ 2 → 0 for each fixed n n n . For φ ∈ C c ( R d ) \varphi\in C_c(\mathbb{R}^d) φ ∈ C c ( R d ) vanishing outside Q n Q_n Q n ,
∣ S φ ( κ ′ ) − S φ ( κ ) ∣ ≤ ∥ φ ∥ ∞ ( N n − + N n + ) . |S_\varphi(\kappa') - S_\varphi(\kappa)| \le\|\varphi\|_\infty(N_n^- + N_n^+). ∣ S φ ( κ ′ ) − S φ ( κ ) ∣ ≤ ∥ φ ∥ ∞ ( N n − + N n + ) .
The same estimate with φ = 1 Q n \varphi= \mathbf{1}_{Q_n} φ = 1 Q n gives L 2 L^2 L 2 -convergence of each local count.
For any sequence of allowed parameter choices with ϑ + γ + ζ + S / r → 0 \vartheta+ \gamma+ \zeta+ S/r \to0 ϑ + γ + ζ + S / r → 0 , the probability that κ ′ \kappa' κ ′ and κ \kappa κ disagree on Q n Q_n Q n tends to zero for all n n n . Every subsequence has a further subsequence along which these restrictions eventually agree almost everywhere, simultaneously for every n n n . On that subsequence the configurations converge vaguely almost everywhere, hence in measure. Thus d 0 → 0 d_0 \to0 d 0 → 0 and d r o o t , 2 ( κ ′ , κ ) → 0 d_{\mathrm{root},2}(\kappa',\kappa) \to0 d root , 2 ( κ ′ , κ ) → 0 . If the asserted δ ∗ \delta_* δ ∗ did not exist, a sequence of allowed choices with the left side of (40) tending to zero and rooted distance at least ε ∗ \varepsilon_* ε ∗ would contradict this convergence. □
Lemma 3.10 (Variance on modified cores). Assume that # G Q ≥ 5 4 ρ V o u t \#G_Q \ge\frac{5}{4}\rho V_{\mathrm{out}} # G Q ≥ 4 5 ρ V out for every S S S -cube Q = u + S I Q = u+SI Q = u + S I . Under the hypotheses of Lemma 3.8 , let z z z satisfy its three inequalities, let κ ′ : = κ z \kappa' := \kappa_z κ ′ := κ z , and fix an integer R ≥ 1 R \ge1 R ≥ 1 with S > 4 R S > 4R S > 4 R . Define
E R : = { t . c : c ∈ C r , Q ∈ Q z ( c ) , Q count-compatible , t ∈ Q i n , dist ( t , ∂ Q i n ) > R } . E_R := \{t.c: c \in C_r,\ Q \in\mathcal{Q}_z(c),\ Q\ \text{count-compatible},\ t \in Q^{\mathrm{in}},\ \operatorname{dist}(t,\partial Q^{\mathrm{in}}) > R\}. E R := { t . c : c ∈ C r , Q ∈ Q z ( c ) , Q count-compatible , t ∈ Q in , dist ( t , ∂ Q in ) > R } . Then
∫ E R ∣ N B R ( κ ′ ) − ρ Vol d ( B R ) ∣ 2 d μ ≤ K L R d − 1 , \int_{E_R} \left|N_{B_R}(\kappa')-\rho\operatorname{Vol}_d(B_R)\right|^2\,d\mu\le K_LR^{d-1}, ∫ E R ∣ N B R ( κ ′ ) − ρ Vol d ( B R ) ∣ 2 d μ ≤ K L R d − 1 , and
μ ( E R c ) ≤ 4 b S , ζ + C d ζ + C d R / S + C d S / r . \mu(E_R^c) \le4b_{S,\zeta}+C_d\zeta+C_dR/S+C_dS/r. μ ( E R c ) ≤ 4 b S , ζ + C d ζ + C d R / S + C d S / r . There is an integrable function C R ( 1 + N B R ( κ ′ ) 2 ) C_R(1+N_{B_R}(\kappa')^2) C R ( 1 + N B R ( κ ′ ) 2 ) , independent of the later choices of ζ \zeta ζ , S S S , r r r , z z z , that bounds ∣ N B R ( κ ′ ) − ρ Vol d ( B R ) ∣ 2 \left|N_{B_R}(\kappa')-\rho\operatorname{Vol}_d(B_R)\right|^2 ∣ N B R ( κ ′ ) − ρ Vol d ( B R ) ∣ 2 on E R c E_R^c E R c .
Proof. Put Λ c ′ : = Λ c , z ′ \Lambda'_c := \Lambda'_{c,z} Λ c ′ := Λ c , z ′ . For each count-compatible Q ∈ Q z ( c ) Q \in\mathcal{Q}_z(c) Q ∈ Q z ( c ) , define the periodic point set
Λ Q p e r : = ⋃ v ∈ S ′ Z d ( ( Λ c ′ ∩ Q i n ) + v ) . \Lambda_Q^{\mathrm{per}} := \bigcup_{v\in S'\mathbb{Z}^d} \bigl((\Lambda'_c \cap Q^{\mathrm{in}})+v\bigr). Λ Q per := v ∈ S ′ Z d ⋃ ( ( Λ c ′ ∩ Q in ) + v ) .
Each L L L -cell in Q i n Q^{\mathrm{in}} Q in satisfies the exact polynomial moment identities, so Lemma 3.6 applies to Λ Q p e r \Lambda_Q^{\mathrm{per}} Λ Q per . For x = t . c ∈ E R x=t.c \in E_R x = t . c ∈ E R , the ball B R + t B_R+t B R + t lies in the core. Equivariance gives
N B R ( κ ′ ( x ) ) = # ( supp κ ′ ( c ) ∩ ( B R + t ) ) , N_{B_R}(\kappa'(x)) = \#\bigl(\operatorname{supp}\kappa'(c) \cap(B_R+t)\bigr), N B R ( κ ′ ( x )) = # ( supp κ ′ ( c ) ∩ ( B R + t ) ) ,
and Λ Q p e r \Lambda_Q^{\mathrm{per}} Λ Q per agrees with supp κ ′ ( c ) \operatorname{supp}\kappa'(c) supp κ ′ ( c ) on B R + t B_R+t B R + t . For fixed c , Q c,Q c , Q , enlarge the domain of integration from { t ∈ Q i n : dist ( t , ∂ Q i n ) > R } \{t\in Q^{\mathrm{in}}:\operatorname{dist}(t,\partial Q^{\mathrm{in}})>R\} { t ∈ Q in : dist ( t , ∂ Q in ) > R } to Q i n Q^{\mathrm{in}} Q in . The integrand is nonnegative, and the latter cube is a fundamental domain for Λ Q \Lambda_Q Λ Q . By Lemma 3.6, the integral is at most K L R d − 1 S d ′ K_L R^{d-1}S^{d'} K L R d − 1 S d ′ . The cores are disjoint. The cross-section formula gives
∫ C r ∑ Q ( S ′ ) d d ν r ( c ) ≤ ∫ C r Vol d ( W c ) d ν r ( c ) = 1 , \int_{C_r}\sum_Q (S')^d\,d\nu_r(c)\leq\int_{C_r}\operatorname{Vol}_d(W_c)\,d\nu_r(c)=1, ∫ C r Q ∑ ( S ′ ) d d ν r ( c ) ≤ ∫ C r Vol d ( W c ) d ν r ( c ) = 1 ,
where the sum is over modified S S S -cubes Q ⊂ W c Q\subset W_c Q ⊂ W c . Hence
∫ E R ∣ N B R ( κ ′ ( x ) ) − ρ Vol d ( B R ) ∣ 2 d μ ( x ) ≤ K L R d − 1 . \int_{E_R}\left|N_{B_R}(\kappa'(x))-\rho\operatorname{Vol}_d(B_R)\right|^2\,d\mu(x)\leq K_LR^{d-1}. ∫ E R ∣ N B R ( κ ′ ( x )) − ρ Vol d ( B R ) ∣ 2 d μ ( x ) ≤ K L R d − 1 .
If x = t . c ∉ E R x=t.c\notin E_R x = t . c ∈ / E R , then either t t t lies in an S S S -cube Q ∈ Q z ( c ) Q\in\mathcal{Q}_z(c) Q ∈ Q z ( c ) for which (35) fails, or t t t lies in an outer shell, within distance R R R of the boundary of a core, or in the part of W c W_c W c not covered by Q z ( c ) \mathcal{Q}_z(c) Q z ( c ) . By (33) and e 1 ( z ) ≤ 4 b S , ζ e_1(z)\leq4b_{S,\zeta} e 1 ( z ) ≤ 4 b S , ζ from Lemma 3.8 ,
μ ( E R c ) ≤ 4 b S , ζ + C d ζ + C d R S + C d S r . \mu(E_R^c)\leq4b_{S,\zeta}+C_d\zeta+C_d\frac{R}{S}+C_d\frac{S}{r}. μ ( E R c ) ≤ 4 b S , ζ + C d ζ + C d S R + C d r S .
Since B R ⊂ Q 2 R + 2 B_R\subset Q_{2R+2} B R ⊂ Q 2 R + 2 , (37) gives N B R ( κ t + ) ≤ D 2 R + 2 N_{B_R}(\kappa_t^+)\leq D_{2R+2} N B R ( κ t + ) ≤ D 2 R + 2 , independently of q , ϑ , γ , ζ , S , r q,\vartheta,\gamma,\zeta,S,r q , ϑ , γ , ζ , S , r . Therefore
N B R ( κ ′ ) ≤ N B R ( κ ) + D 2 R + 2 , N_{B_R}(\kappa')\leq N_{B_R}(\kappa)+D_{2R+2}, N B R ( κ ′ ) ≤ N B R ( κ ) + D 2 R + 2 ,
and, on E R c E_R^c E R c ,
∣ N B R ( κ ′ ) − ρ Vol d ( B R ) ∣ 2 ≤ C R ( 1 + N B R ( κ ) 2 ) , \left|N_{B_R}(\kappa')-\rho\operatorname{Vol}_d(B_R)\right|^2\leq C_R(1+N_{B_R}(\kappa)^2), ∣ N B R ( κ ′ ) − ρ Vol d ( B R ) ∣ 2 ≤ C R ( 1 + N B R ( κ ) 2 ) ,
where C R C_R C R depends only on R , d , ρ R,d,\rho R , d , ρ and the separation constants. The function 1 + N B R ( κ ) 2 1+N_{B_R}(\kappa)^2 1 + N B R ( κ ) 2 is integrable. □
Proof of Proposition 3.3 . Fix an open neighbourhood U U U of κ \kappa κ and ε U > 0 \varepsilon_U>0 ε U > 0 with B d r o o t , 2 ( κ , ε U ) ⊂ U B_{\mathrm{droot},2}(\kappa,\varepsilon_U)\subset U B droot , 2 ( κ , ε U ) ⊂ U . Let δ U : = δ ∗ ( κ , ε U ) \delta_U:=\delta_*(\kappa,\varepsilon_U) δ U := δ ∗ ( κ , ε U ) be given by Lemma 3.9 . Choose D D D and γ > 0 \gamma>0 γ > 0 so that
ϑ + γ < δ U 4 . \vartheta+\gamma<\frac{\delta_U}{4}. ϑ + γ < 4 δ U .
With this D D D , the constants ε 0 , q 0 , σ d \varepsilon_0,q_0,\sigma_d ε 0 , q 0 , σ d are fixed. Choose q ≥ q 0 q\geq q_0 q ≥ q 0 so that
μ ( E L ) < γ , L − d ∫ E L N L I ( κ ( x ) ) d μ ( x ) < γ , K l a t q − 1 / d ≤ ε . \mu(E_L)<\gamma,\qquad L^{-d}\int_{E_L}N_{LI}(\kappa(x))\,d\mu(x)<\gamma,\qquad K_{\mathrm{lat}}q^{-1/d}\leq\varepsilon. μ ( E L ) < γ , L − d ∫ E L N L I ( κ ( x )) d μ ( x ) < γ , K lat q − 1/ d ≤ ε .
This fixes L L L and K L K_L K L . Since β > ( d − 1 ) / d \beta>(d-1)/d β > ( d − 1 ) / d , choose an integer R > N R>N R > N such that
2 K L R d − 1 < j − 1 Vol d ( B R ) β . 2K_LR^{d-1}<j^{-1}\operatorname{Vol}_d(B_R)^\beta. 2 K L R d − 1 < j − 1 Vol d ( B R ) β .
For this R R R , the function C R ( 1 + N B R ( κ ) 2 ) C_R(1+N_{B_R}(\kappa)^2) C R ( 1 + N B R ( κ ) 2 ) in Lemma 3.10 is integrable. Choose δ R > 0 \delta_R>0 δ R > 0 such that
μ ( E ) < δ R ⟹ C R ∫ E ( 1 + N B R ( κ ) 2 ) d μ < K L R d − 1 . \mu(E)<\delta_R\quad\Longrightarrow\quad C_R\int_E(1+N_{B_R}(\kappa)^2)\,d\mu<K_LR^{d-1}. μ ( E ) < δ R ⟹ C R ∫ E ( 1 + N B R ( κ ) 2 ) d μ < K L R d − 1 .
Choose 0 < ζ < 1 / 4 0<\zeta<1/4 0 < ζ < 1/4 with
ζ < δ U 4 , C d ζ < δ R 3 . \zeta<\frac{\delta_U}{4},\qquad C_d\zeta<\frac{\delta_R}{3}. ζ < 4 δ U , C d ζ < 3 δ R .
For each sufficiently large integer ℓ \ell ℓ , choose ℓ ′ < ℓ \ell'<\ell ℓ ′ < ℓ of the same parity as ℓ \ell ℓ with
∣ ℓ ′ − ℓ ( 1 − 3 ζ / 2 ) 1 / d ∣ ≤ 2. \left|\ell'-\ell(1-3\zeta/2)^{1/d}\right|\leq2. ℓ ′ − ℓ ( 1 − 3 ζ /2 ) 1/ d ≤ 2.
Then ℓ ′ / ℓ → ( 1 − 3 ζ / 2 ) 1 / d \ell'/\ell\to(1-3\zeta/2)^{1/d} ℓ ′ / ℓ → ( 1 − 3 ζ /2 ) 1/ d , and
ζ ≤ 1 − ( ℓ ′ / ℓ ) d ≤ 2 ζ \zeta\leq1-(\ell'/\ell)^d\leq2\zeta ζ ≤ 1 − ( ℓ ′ / ℓ ) d ≤ 2 ζ
for all sufficiently large ℓ \ell ℓ . Since b S , ζ → 0 b_{S,\zeta}\to0 b S , ζ → 0 for fixed ζ \zeta ζ as S = ℓ L → ∞ S=\ell L\to\infty S = ℓ L → ∞ , we may also require
S > 4 R , 4 b S , ζ + C d R S < δ R 3 , # G Q ≥ 5 4 ρ V o u t ( Q = u + S I ) . S>4R,\qquad4b_{S,\zeta}+C_d\frac{R}{S}<\frac{\delta_R}{3},\qquad\#G_Q\geq\frac{5}{4}\rho V_{\mathrm{out}}\quad(Q=u+SI). S > 4 R , 4 b S , ζ + C d S R < 3 δ R , # G Q ≥ 4 5 ρ V out ( Q = u + S I ) .
Finally choose r ≥ S 2 r \ge S^2 r ≥ S 2 so that
C d S / r < δ R / 3 , S / r < δ U / 4. C_dS/r < \delta_R/3,\qquad S/r < \delta_U/4. C d S / r < δ R /3 , S / r < δ U /4.
Choose z z z by Lemma 3.8 and set κ ′ : = κ z \kappa' := \kappa_z κ ′ := κ z . The bounds give μ ( E R c ) < δ R \mu(E_R^c) < \delta_R μ ( E R c ) < δ R , while Lemma 3.9 gives d r o o t , 2 ( κ ′ , κ ) < ε U d_{\mathrm{root},2}(\kappa',\kappa) < \varepsilon_U d root , 2 ( κ ′ , κ ) < ε U . By the defining property of δ R \delta_R δ R ,
∫ E R c ∣ N B R ( κ ′ ) − ρ Vol d ( B R ) ∣ 2 d μ ≤ K L R d − 1 . \int_{E_R^c} \left|N_{B_R}(\kappa')-\rho\operatorname{Vol}_d(B_R)\right|^2\,d\mu\le K_LR^{d-1}. ∫ E R c ∣ N B R ( κ ′ ) − ρ Vol d ( B R ) ∣ 2 d μ ≤ K L R d − 1 .
Consequently κ ′ ∈ U \kappa' \in\mathcal{U} κ ′ ∈ U , and
V κ ′ ( R ) ≤ 2 K L R d − 1 < j − 1 Vol d ( B R ) β . \mathcal{V}_{\kappa'}(R) \le2K_LR^{d-1} < j^{-1}\operatorname{Vol}_d(B_R)^\beta. V κ ′ ( R ) ≤ 2 K L R d − 1 < j − 1 Vol d ( B R ) β .
High-variance approximation
In this subsection we prove density of realizations with arbitrarily large variance at prescribed scales. We preserve the number of points in each large cube but move them into smaller subcubes. A random translation of this arrangement produces a count fluctuation of order the squared expected number of points in an observation ball.
Proposition 3.11 (High-variance density). For every a ∈ A d ( X , μ ) a \in A_d(X,\mu) a ∈ A d ( X , μ ) , α < 2 \alpha< 2 α < 2 , and j , N ≥ 1 j,N \ge1 j , N ≥ 1 , the set H α , j , N a \mathcal{H}_{\alpha,j,N}^a H α , j , N a is dense in R ρ , 2 a ( X , μ ) \mathcal{R}_{\rho,2}^a(X,\mu) R ρ , 2 a ( X , μ ) .
Proof. Fix κ ∈ R ρ , 2 a ( X , μ ) \kappa\in\mathcal{R}_{\rho,2}^a(X,\mu) κ ∈ R ρ , 2 a ( X , μ ) and an open neighbourhood U \mathcal{U} U of κ \kappa κ . It suffices to find κ ′ ∈ U ∩ H α , j , N a \kappa' \in\mathcal{U} \cap\mathcal{H}_{\alpha,j,N}^a κ ′ ∈ U ∩ H α , j , N a .
Definition of κ δ , S , r \kappa_{\delta,S,r} κ δ , S , r . By Lemma 2.1 and essential freeness, use a pointwise equivariant Borel representative of κ \kappa κ on an invariant conull Borel set X 0 X_0 X 0 on which a a a is free. Write Λ x : = supp κ ( x ) \Lambda_x := \operatorname{supp}\kappa(x) Λ x := supp κ ( x ) . For 0 < δ < 6 − d 0 < \delta< 6^{-d} 0 < δ < 6 − d , S > 0 S > 0 S > 0 , and r ≥ S 2 r \ge S^2 r ≥ S 2 , choose an r r r -separated cocompact Borel cross-section C r ⊂ X 0 C_r \subset X_0 C r ⊂ X 0 , and equip it with the Voronoi cells W c W_c W c of (5) . Normalize the transverse measure ν r \nu_r ν r by
∫ X H d μ = ∫ C r ∫ W c H ( t . c ) d t d ν r ( c ) \int_X H\,d\mu= \int_{C_r}\int_{W_c} H(t.c)\,dt\,d\nu_r(c) ∫ X H d μ = ∫ C r ∫ W c H ( t . c ) d t d ν r ( c )
for nonnegative Borel H H H .
With h : = δ 1 / d S h := \delta^{1/d}S h := δ 1/ d S , for each c ∈ C r c \in C_r c ∈ C r put
Q S ( c ) : = { v + [ 0 , S ) d : v ∈ S Z d , v + [ 0 , S ) d ⊂ W c } . Q_S(c) := \{v+[0,S)^d : v \in S\mathbb{Z}^d,\ v+[0,S)^d \subset W_c\}. Q S ( c ) := { v + [ 0 , S ) d : v ∈ S Z d , v + [ 0 , S ) d ⊂ W c } .
By (33) ,
Vol d ( W c ∖ ⋃ Q ∈ Q S ( c ) Q ) ≤ C d S r Vol d ( W c ) . \operatorname{Vol}_d\left(W_c\setminus\bigcup_{Q\in Q_S(c)}Q\right)\le C_d\frac{S}{r}\operatorname{Vol}_d(W_c). Vol d W c ∖ Q ∈ Q S ( c ) ⋃ Q ≤ C d r S Vol d ( W c ) .
The cross-section formula gives
S − d ( 1 − C d S r ) ≤ ∫ C r # Q S ( c ) d ν r ( c ) ≤ S − d . S^{-d}\left(1-C_d\frac{S}{r}\right)\le\int_{C_r}\#Q_S(c)\,d\nu_r(c)\le S^{-d}. S − d ( 1 − C d r S ) ≤ ∫ C r # Q S ( c ) d ν r ( c ) ≤ S − d .
For Q = a Q + [ 0 , S ) d Q=a_Q+[0,S)^d Q = a Q + [ 0 , S ) d , set e : = ( 1 , … , 1 ) e:=(1,\ldots,1) e := ( 1 , … , 1 ) . Since h < S / 6 h<S/6 h < S /6 , the cubes
U Q : = a Q + S 4 e + [ 0 , 3 h ) d , U_Q:=a_Q+\frac{S}{4}e+[0,3h)^d, U Q := a Q + 4 S e + [ 0 , 3 h ) d ,
H Q : = a Q + S 4 e + [ 0 , h ) d , H_Q:=a_Q+\frac{S}{4}e+[0,h)^d, H Q := a Q + 4 S e + [ 0 , h ) d ,
D Q : = a Q + S 4 e + 2 h e + [ 0 , h ) d D_Q:=a_Q+\frac{S}{4}e+2he+[0,h)^d D Q := a Q + 4 S e + 2 h e + [ 0 , h ) d
satisfy H Q ⊂ U Q H_Q \subset U_Q H Q ⊂ U Q and U Q ‾ ⊂ int Q \overline{U_Q} \subset\operatorname{int} Q U Q ⊂ int Q , and D Q ⊂ U Q ∖ H Q D_Q \subset U_Q \setminus H_Q D Q ⊂ U Q ∖ H Q . Define
T Q ( y ) : = a Q + S 4 e + 2 h e + 1 3 ( y − a Q − S 4 e ) , y ∈ U Q . T_Q(y) := a_Q + \frac{S}{4}e + 2he + \frac{1}{3}\left(y-a_Q-\frac{S}{4}e\right), \qquad y \in U_Q. T Q ( y ) := a Q + 4 S e + 2 h e + 3 1 ( y − a Q − 4 S e ) , y ∈ U Q .
For each c ∈ C r c \in C_r c ∈ C r , set
U c : = ⋃ Q ∈ Q S ( c ) U Q , Λ ~ c : = ( Λ c ∩ ( W c ∖ U c ) ) ∪ ⋃ Q ∈ Q S ( c ) T Q ( Λ c ∩ U Q ) . U_c := \bigcup_{Q\in Q_S(c)} U_Q, \qquad\widetilde{\Lambda}_c := \left(\Lambda_c \cap(W_c \setminus U_c)\right) \cup\bigcup_{Q\in Q_S(c)} T_Q(\Lambda_c \cap U_Q). U c := Q ∈ Q S ( c ) ⋃ U Q , Λ c := ( Λ c ∩ ( W c ∖ U c ) ) ∪ Q ∈ Q S ( c ) ⋃ T Q ( Λ c ∩ U Q ) .
Define
κ δ , S , r ( x ) : = ∑ u ∈ ( C r ) x ∑ v ∈ Λ ~ u . x δ u + v , x ∈ X 0 . \kappa_{\delta,S,r}(x) := \sum_{u\in(C_r)_x}\sum_{v\in\widetilde{\Lambda}_{u.x}} \delta_{u+v}, \qquad x \in X_0. κ δ , S , r ( x ) := u ∈ ( C r ) x ∑ v ∈ Λ u . x ∑ δ u + v , x ∈ X 0 .
The map c ↦ Λ ~ c c \mapsto\widetilde{\Lambda}_c c ↦ Λ c is Borel. The condition Q ⊂ W c Q \subset W_c Q ⊂ W c is Borel by the half-space description of W c W_c W c in (5) and the cube supremum formula (34) . Point restrictions and affine images of locally finite configurations are also Borel. The cells u + W u . x u+W_{u.x} u + W u . x partition R d \mathbb{R}^d R d , and only finitely many intersect a compact set. Thus the sum defines a locally finite simple counting measure. The relation { ( x , u ) ∈ X 0 × R d : u . x ∈ C r } \{(x,u)\in X_0\times\mathbb{R}^d:u.x\in C_r\} {( x , u ) ∈ X 0 × R d : u . x ∈ C r } is Borel with countable sections, so Lusin–Novikov [24 Theorem 18.10] gives Borel partial maps u j : D j → R d u_j:D_j\to\mathbb{R}^d u j : D j → R d whose values enumerate ( C r ) x (C_r)_x ( C r ) x . For each bounded Borel set D D D ,
N D ( κ δ , S , r ( x ) ) = ∑ j : x ∈ D j # ( Λ ~ u j ( x ) . x ∩ ( D − u j ( x ) ) ) , N_D(\kappa_{\delta,S,r}(x))=\sum_{j:x\in D_j}\#\left(\widetilde{\Lambda}_{u_j(x).x}\cap(D-u_j(x))\right), N D ( κ δ , S , r ( x )) = j : x ∈ D j ∑ # ( Λ u j ( x ) . x ∩ ( D − u j ( x )) ) ,
which is Borel because c ↦ Λ ~ c c\mapsto\widetilde{\Lambda}_c c ↦ Λ c is a Borel locally finite point-set map. The return-time identity ( C r ) v . x = ( C r ) x − v (C_r)_{v.x}=(C_r)_x-v ( C r ) v . x = ( C r ) x − v gives equivariance. Since T Q : U Q → D Q ⊂ U Q T_Q:U_Q\to D_Q\subset U_Q T Q : U Q → D Q ⊂ U Q is injective and
# Λ ~ c = N W c ( κ ( c ) ) , \#\widetilde{\Lambda}_c=N_{W_c}(\kappa(c)), # Λ c = N W c ( κ ( c )) ,
the cross-section mass-transport formula gives intensity ρ \rho ρ . Moreover, ∥ T Q ( y ) − y ∥ ≤ C d S \lVert T_Q(y)-y\rVert\le C_dS ∥ T Q ( y ) − y ∥ ≤ C d S . For every m ≥ 1 m\ge1 m ≥ 1 , choose m ′ = m ′ ( m , S ) m'=m'(m,S) m ′ = m ′ ( m , S ) such that the C d S C_dS C d S -neighbourhood of Q m Q_m Q m is contained in Q m ′ Q_{m'} Q m ′ . Injectivity of the modifications gives
N Q m ( κ δ , S , r ( x ) ) ≤ N Q m ′ ( κ ( x ) ) ( x ∈ X 0 ) . N_{Q_m}(\kappa_{\delta,S,r}(x))\le N_{Q_{m'}}(\kappa(x)) \qquad(x\in X_0). N Q m ( κ δ , S , r ( x )) ≤ N Q m ′ ( κ ( x )) ( x ∈ X 0 ) .
Hence N Q m ∘ κ δ , S , r ∈ L 2 ( μ ) N_{Q_m}\circ\kappa_{\delta,S,r}\in L^2(\mu) N Q m ∘ κ δ , S , r ∈ L 2 ( μ ) .
Closeness in the rooted topology. Fix m ≥ 1 m\ge1 m ≥ 1 and restrict to S > max { 1 , 6 m , 2 d } S>\max\{1,6m,2\sqrt d\} S > max { 1 , 6 m , 2 d } . For Q = a Q + [ 0 , S ) d ∈ Q S ( c ) Q=a_Q+[0,S)^d\in Q_S(c) Q = a Q + [ 0 , S ) d ∈ Q S ( c ) put
a U : = a Q + S 4 e , a D : = a U + 2 h e , a_U:=a_Q+\frac{S}{4}e,\qquad a_D:=a_U+2he, a U := a Q + 4 S e , a D := a U + 2 h e ,
and define the closed cubes
A Q , m : = a D + [ − m / 2 , h + m / 2 ] d , B Q , m : = a U + [ − 3 m / 2 , 3 h + 3 m / 2 ] d . A_{Q,m}:=a_D+[-m/2,h+m/2]^d,\qquad B_{Q,m}:=a_U+[-3m/2,3h+3m/2]^d. A Q , m := a D + [ − m /2 , h + m /2 ] d , B Q , m := a U + [ − 3 m /2 , 3 h + 3 m /2 ] d .
Since h < S / 6 h<S/6 h < S /6 and S > 6 m S>6m S > 6 m , we have A Q , m , B Q , m ⊂ int Q ⊂ W c A_{Q,m},B_{Q,m}\subset\operatorname{int}Q\subset W_c A Q , m , B Q , m ⊂ int Q ⊂ W c . Moreover,
Q m ∩ ( U Q − t ) ≠ ∅ ⟹ t ∈ B Q , m , Q_m\cap(U_Q-t)\ne\varnothing\Longrightarrow t\in B_{Q,m}, Q m ∩ ( U Q − t ) = ∅ ⟹ t ∈ B Q , m ,
Q m ∩ ( D Q − t ) ≠ ∅ ⟹ t ∈ A Q , m . Q_m\cap(D_Q-t)\ne\varnothing\Longrightarrow t\in A_{Q,m}. Q m ∩ ( D Q − t ) = ∅ ⟹ t ∈ A Q , m .
The map ( c , t ) ↦ t . c (c,t)\mapsto t.c ( c , t ) ↦ t . c is a Borel bijection from { ( c , t ) : c ∈ C r , t ∈ W c } \{(c,t):c\in C_r,\ t\in W_c\} {( c , t ) : c ∈ C r , t ∈ W c } onto X 0 X_0 X 0 . Define
F m : = { t . c : c ∈ C r , Q ∈ Q S ( c ) , t ∈ B Q , m } . F_m:=\{t.c:c\in C_r,\ Q\in Q_S(c),\ t\in B_{Q,m}\}. F m := { t . c : c ∈ C r , Q ∈ Q S ( c ) , t ∈ B Q , m } .
If x ∉ F m x\notin F_m x ∈ / F m , the measures κ δ , S , r ( x ) \kappa_{\delta,S,r}(x) κ δ , S , r ( x ) and κ ( x ) \kappa(x) κ ( x ) agree on Q m Q_m Q m . The cross-section formula gives
μ ( F m ) = ( 3 h + 3 m ) d ∫ C r # Q S ( c ) d ν r ( c ) ≤ ( 3 δ 1 / d + 3 m S ) d ≤ C d δ + C d , m S . \mu(F_m)=(3h+3m)^d\int_{C_r}\#Q_S(c)\,d\nu_r(c)\le\left(3\delta^{1/d}+\frac{3m}{S}\right)^d\le C_d\delta+\frac{C_{d,m}}{S}. μ ( F m ) = ( 3 h + 3 m ) d ∫ C r # Q S ( c ) d ν r ( c ) ≤ ( 3 δ 1/ d + S 3 m ) d ≤ C d δ + S C d , m .
For x = t . c x=t.c x = t . c , with t ∈ W c t\in W_c t ∈ W c , define
I m ( x ) : = ∑ Q ∈ Q S ( c ) # ( ( T Q ( Λ c ∩ U Q ) ) ∩ ( t + Q m ) ) , I_m(x):=\sum_{Q\in\mathcal{Q}_S(c)}\#\bigl((T_Q(\Lambda_c\cap U_Q))\cap(t+Q_m)\bigr), I m ( x ) := Q ∈ Q S ( c ) ∑ # ( ( T Q ( Λ c ∩ U Q )) ∩ ( t + Q m ) ) ,
O m ( x ) : = ∑ Q ∈ Q S ( c ) # ( Λ c ∩ U Q ∩ ( t + Q m ) ) . O_m(x):=\sum_{Q\in\mathcal{Q}_S(c)}\#\bigl(\Lambda_c\cap U_Q\cap(t+Q_m)\bigr). O m ( x ) := Q ∈ Q S ( c ) ∑ # ( Λ c ∩ U Q ∩ ( t + Q m ) ) .
An inserted or removed point in u + U Q u+U_Q u + U Q , with u ∈ ( C r ) c ∖ { 0 } u\in(C_r)_c\setminus\{0\} u ∈ ( C r ) c ∖ { 0 } and Q ∈ Q S ( u . c ) Q\in\mathcal{Q}_S(u.c) Q ∈ Q S ( u . c ) , cannot lie in t + Q m t+Q_m t + Q m : otherwise t − u ∈ B Q , m ⊂ W u . c t-u\in B_{Q,m}\subset W_{u.c} t − u ∈ B Q , m ⊂ W u . c , contradicting the unique representation x = t . c x=t.c x = t . c . Unchanged points cancel in the count difference. Thus
N Q m ( κ δ , S , r ( x ) ) − N Q m ( κ ( x ) ) = I m ( x ) − O m ( x ) , O m ≤ ( N Q m ∘ κ ) 1 F m . N_{Q_m}(\kappa_{\delta,S,r}(x))-N_{Q_m}(\kappa(x))=I_m(x)-O_m(x),\qquad O_m\leq(N_{Q_m}\circ\kappa)1_{F_m}. N Q m ( κ δ , S , r ( x )) − N Q m ( κ ( x )) = I m ( x ) − O m ( x ) , O m ≤ ( N Q m ∘ κ ) 1 F m .
If x = t . c x=t.c x = t . c and moved points from U Q U_Q U Q enter t + Q m t+Q_m t + Q m , then t ∈ A Q , m t\in A_{Q,m} t ∈ A Q , m . Their preimages lie in s + Q 3 m s+Q_{3m} s + Q 3 m , where
s = 3 t + a U − 3 a D . s=3t+a_U-3a_D. s = 3 t + a U − 3 a D .
This affine map sends A Q , m A_{Q,m} A Q , m bijectively onto B Q , m B_{Q,m} B Q , m , with d s = 3 d d t ds=3^d\,dt d s = 3 d d t . For each c c c , the sets A Q , m A_{Q,m} A Q , m lie in pairwise disjoint cubes Q ⊂ W c Q\subset W_c Q ⊂ W c . Since ( c , t ) ↦ t . c (c,t)\mapsto t.c ( c , t ) ↦ t . c is injective on { ( c , t ) : c ∈ C r , t ∈ W c } \{(c,t):c\in C_r,\ t\in W_c\} {( c , t ) : c ∈ C r , t ∈ W c } , the sets { t . c : t ∈ A Q , m } \{t.c:t\in A_{Q,m}\} { t . c : t ∈ A Q , m } are disjoint as ( c , Q ) (c,Q) ( c , Q ) varies. The same argument applies to B Q , m B_{Q,m} B Q , m , so
∫ X I m 2 d μ ≤ ∫ C r ∑ Q ∈ Q S ( c ) ∫ A Q , m ( N Q 3 m ( κ ( ( 3 t + a U − 3 a D ) . c ) ) ) 2 d t d ν r ( c ) = 3 − d ∫ F m ( N Q 3 m ∘ κ ) 2 d μ . \begin{aligned}
\int_X I_m^2\,d\mu\leq\int_{C_r}\sum_{Q\in\mathcal{Q}_S(c)}\int_{A_{Q,m}}\bigl(N_{Q_{3m}}(\kappa((3t+a_U-3a_D).c))\bigr)^2\,dt\,d\nu_r(c) \\
&=3^{-d}\int_{F_m}(N_{Q_{3m}}\circ\kappa)^2\,d\mu.
\end{aligned} ∫ X I m 2 d μ ≤ ∫ C r Q ∈ Q S ( c ) ∑ ∫ A Q , m ( N Q 3 m ( κ (( 3 t + a U − 3 a D ) . c )) ) 2 d t d ν r ( c ) = 3 − d ∫ F m ( N Q 3 m ∘ κ ) 2 d μ .
It follows that
∥ N Q m ∘ κ δ , S , r − N Q m ∘ κ ∥ 2 2 ≤ 2 ( 1 + 3 − d ) ∫ F m ( N Q 3 m ∘ κ ) 2 d μ . \|N_{Q_m}\circ\kappa_{\delta,S,r}-N_{Q_m}\circ\kappa\|_2^2\leq2(1+3^{-d})\int_{F_m}(N_{Q_{3m}}\circ\kappa)^2\,d\mu. ∥ N Q m ∘ κ δ , S , r − N Q m ∘ κ ∥ 2 2 ≤ 2 ( 1 + 3 − d ) ∫ F m ( N Q 3 m ∘ κ ) 2 d μ .
The function ( N Q 3 m ∘ κ ) 2 (N_{Q_{3m}}\circ\kappa)^2 ( N Q 3 m ∘ κ ) 2 is integrable. By the bound on μ ( F m ) \mu(F_m) μ ( F m ) and absolute continuity of the integral,
lim δ ↓ 0 lim sup S → ∞ sup r ≥ S 2 ∥ N Q m ∘ κ δ , S , r − N Q m ∘ κ ∥ L 2 ( μ ) = 0. \lim_{\delta\downarrow0}\limsup_{S\to\infty}\sup_{r\geq S^2}\|N_{Q_m}\circ\kappa_{\delta,S,r}-N_{Q_m}\circ\kappa\|_{L^2(\mu)}=0. δ ↓ 0 lim S → ∞ lim sup r ≥ S 2 sup ∥ N Q m ∘ κ δ , S , r − N Q m ∘ κ ∥ L 2 ( μ ) = 0.
If φ ∈ C c ( R d ) \varphi\in C_c(\mathbb{R}^d) φ ∈ C c ( R d ) and supp φ ⊂ Q m \operatorname{supp}\varphi\subset Q_m supp φ ⊂ Q m , then S φ ( κ δ , S , r ) = S φ ( κ ) S_\varphi(\kappa_{\delta,S,r})=S_\varphi(\kappa) S φ ( κ δ , S , r ) = S φ ( κ ) on X ∖ F m X\setminus F_m X ∖ F m . The bound μ ( F m ) ≤ C d δ + C d , m / S \mu(F_m)\leq C_d\delta+C_{d,m}/S μ ( F m ) ≤ C d δ + C d , m / S and the subsequence argument in the proof of Lemma 3.9 give κ δ , S , r → κ \kappa_{\delta,S,r}\to\kappa κ δ , S , r → κ in measure for the vague topology. Together with the local-count L 2 L^2 L 2 -convergence, the definition of d r o o t , 2 d_{\mathrm{root},2} d root , 2 gives
lim δ ↓ 0 lim sup S → ∞ sup r ≥ S 2 d r o o t , 2 ( κ δ , S , r , κ ) = 0. \lim_{\delta\downarrow0}\limsup_{S\to\infty}\sup_{r\geq S^2}d_{\mathrm{root},2}(\kappa_{\delta,S,r},\kappa)=0. δ ↓ 0 lim S → ∞ lim sup r ≥ S 2 sup d root , 2 ( κ δ , S , r , κ ) = 0.
Let ε U > 0 \varepsilon_U>0 ε U > 0 satisfy B d r o o t , 2 ( κ , ε U ) ⊂ U B_{d_{\mathrm{root},2}}(\kappa,\varepsilon_U)\subset\mathcal{U} B d root , 2 ( κ , ε U ) ⊂ U , and take 0 < δ < 6 − d 0<\delta<6^{-d} 0 < δ < 6 − d so small that, for all sufficiently large S S S and every r ≥ S 2 r\geq S^2 r ≥ S 2 ,
d r o o t , 2 ( κ δ , S , r , κ ) < ε U . d_{\mathrm{root},2}(\kappa_{\delta,S,r},\kappa)<\varepsilon_U. d root , 2 ( κ δ , S , r , κ ) < ε U .
Variance from the empty region. Since α < 2 \alpha<2 α < 2 , choose an integer R > N R>N R > N so large that
δ 4 ρ 2 Vol d ( B R ) 2 > j Vol d ( B R ) α . \frac{\delta}{4}\rho^2\operatorname{Vol}_d(B_R)^2>j\operatorname{Vol}_d(B_R)^\alpha. 4 δ ρ 2 Vol d ( B R ) 2 > j Vol d ( B R ) α .
Choose S S S sufficiently large that d r o o t , 2 ( κ δ , S , r , κ ) < ε U d_{\mathrm{root},2}(\kappa_{\delta,S,r},\kappa)<\varepsilon_U d root , 2 ( κ δ , S , r , κ ) < ε U for every r ≥ S 2 r\geq S^2 r ≥ S 2 , and
h = δ 1 / d S > 2 R , ( 1 − 2 R h ) d ≥ 1 2 . h=\delta^{1/d}S>2R,\qquad\left(1-\frac{2R}{h}\right)^d\geq\frac12. h = δ 1/ d S > 2 R , ( 1 − h 2 R ) d ≥ 2 1 .
Then choose r ≥ S 2 r\geq S^2 r ≥ S 2 so that
1 − C d S r ≥ 1 2 . 1-\frac{C_dS}{r}\geq\frac12. 1 − r C d S ≥ 2 1 .
With κ ′ : = κ δ , S , r \kappa' := \kappa_{\delta,S,r} κ ′ := κ δ , S , r , we have κ ′ ∈ U \kappa' \in\mathcal{U} κ ′ ∈ U . Let E R 0 ⊂ X E_R^0 \subset X E R 0 ⊂ X consist of the points x = t . c x=t.c x = t . c for which t t t lies in some H Q H_Q H Q and dist ( t , ∂ H Q ) > R \operatorname{dist}(t,\partial H_Q)>R dist ( t , ∂ H Q ) > R . Since T Q ( U Q ) = D Q ⊂ U Q ∖ H Q T_Q(U_Q)=D_Q\subset U_Q\setminus H_Q T Q ( U Q ) = D Q ⊂ U Q ∖ H Q ,
N B R ( κ δ , S , r ( x ) ) = 0 ( x ∈ E R 0 ) . N_{B_R}(\kappa_{\delta,S,r}(x))=0 \qquad(x\in E_R^0). N B R ( κ δ , S , r ( x )) = 0 ( x ∈ E R 0 ) .
The lower bound for ∫ C r # Q S ( c ) d ν r ( c ) \int_{C_r}\# Q_S(c)\,d\nu_r(c) ∫ C r # Q S ( c ) d ν r ( c ) gives
μ ( E R 0 ) ≥ ( h − 2 R ) d S d ( 1 − C d S r ) = δ ( 1 − 2 R h ) d ( 1 − C d S r ) . \begin{aligned}
\mu(E_R^0)\ge\frac{(h-2R)^d}{S^d}\left(1-C_d\frac{S}{r}\right) \\
&=\delta\left(1-\frac{2R}{h}\right)^d\left(1-C_d\frac{S}{r}\right).
\end{aligned} μ ( E R 0 ) ≥ S d ( h − 2 R ) d ( 1 − C d r S ) = δ ( 1 − h 2 R ) d ( 1 − C d r S ) .
By the choices of S S S and r r r , the last expression is at least δ / 4 \delta/4 δ /4 .
Since κ ′ \kappa' κ ′ has intensity ρ \rho ρ ,
V κ ′ ( R ) = ∫ X ∣ N B R ( κ ′ ( x ) ) − ρ Vol d ( B R ) ∣ 2 d μ ( x ) ≥ μ ( E R 0 ) ρ 2 Vol d ( B R ) 2 ≥ δ 4 ρ 2 Vol d ( B R ) 2 > j Vol d ( B R ) α . \begin{aligned}
\mathcal{V}_{\kappa'}(R)=\int_X\left|N_{B_R}(\kappa'(x))-\rho\operatorname{Vol}_d(B_R)\right|^2\,d\mu(x) \\
&\ge\mu(E_R^0)\rho^2\operatorname{Vol}_d(B_R)^2 \\
&\ge\frac{\delta}{4}\rho^2\operatorname{Vol}_d(B_R)^2>j\operatorname{Vol}_d(B_R)^\alpha.
\end{aligned} V κ ′ ( R ) = ∫ X ∣ N B R ( κ ′ ( x )) − ρ Vol d ( B R ) ∣ 2 d μ ( x ) ≥ μ ( E R 0 ) ρ 2 Vol d ( B R ) 2 ≥ 4 δ ρ 2 Vol d ( B R ) 2 > j Vol d ( B R ) α .
Density of generating realizations
We prove that every realization can be approximated by one from which the original state can be recovered. For a separated cocompact cross-section, we retain all but three points in certain Voronoi cells and place three points near the corresponding return vector. Two of these points have a fixed nonzero difference, and the third records the cross-section element through a Borel injection into a Cantor set. Lemmas 3.13 and 3.14 give, respectively, the construction and the recovery map.
Proposition 3.12 (Density of generating realizations). Let a ∈ A d ( X , μ ) a\in A_d(X,\mu) a ∈ A d ( X , μ ) , let κ ∈ R ρ , 2 a ( X , μ ) \kappa\in R_{\rho,2}^a(X,\mu) κ ∈ R ρ , 2 a ( X , μ ) , and let U \mathcal{U} U be an open neighbourhood of κ \kappa κ . Then
U ∩ R ρ , 2 a , i s o ( X , μ ) ≠ ∅ . \mathcal{U}\cap R_{\rho,2}^{a,\mathrm{iso}}(X,\mu)\ne\varnothing. U ∩ R ρ , 2 a , iso ( X , μ ) = ∅ . Consequently, R ρ , 2 a , i s o ( X , μ ) R_{\rho,2}^{a,\mathrm{iso}}(X,\mu) R ρ , 2 a , iso ( X , μ ) is dense in R ρ , 2 a ( X , μ ) R_{\rho,2}^a(X,\mu) R ρ , 2 a ( X , μ ) .
Proof of Proposition 3.12
Lemma 3.13 (Sparse marker replacement). Let a ∈ A d ( X , μ ) a\in A_d(X,\mu) a ∈ A d ( X , μ ) , let κ ∈ R ρ , 2 a ( X , μ ) \kappa\in R_{\rho,2}^a(X,\mu) κ ∈ R ρ , 2 a ( X , μ ) , and fix ε > 0 \varepsilon>0 ε > 0 . For every sufficiently large r r r there exist an r r r -separated cocompact cross-section C r C_r C r with transverse measure ν r \nu_r ν r , a Borel equivariant realization κ r ∈ R ρ , 2 a ( X , μ ) \kappa_r\in R_{\rho,2}^a(X,\mu) κ r ∈ R ρ , 2 a ( X , μ ) , vectors 0 ≠ v ∗ ∈ B ε 0\ne v_*\in B_\varepsilon 0 = v ∗ ∈ B ε and a 0 ∈ B ε a_0\in B_\varepsilon a 0 ∈ B ε , a Borel set E r ′ ⊂ C r E'_r\subset C_r E r ′ ⊂ C r with ν r ( E r ′ ) > 0 \nu_r(E'_r)>0 ν r ( E r ′ ) > 0 , a compact Lebesgue-null Cantor set K ⊂ B ε K\subset B_\varepsilon K ⊂ B ε , and a Borel injection w : E r ′ → K w:E'_r\to K w : E r ′ → K satisfying the following.
(i) For μ \mu μ -almost every x x x , the ordered pairs of distinct points of supp κ r ( x ) \operatorname{supp}\kappa_r(x) supp κ r ( x ) having difference v ∗ v_* v ∗ are precisely
( u + a 0 , u + a 0 + v ∗ ) , u ∈ ( C r ) x , u . x ∈ E r ′ . (u+a_0,u+a_0+v_*),\qquad u\in(C_r)_x,\quad u.x\in E'_r. ( u + a 0 , u + a 0 + v ∗ ) , u ∈ ( C r ) x , u . x ∈ E r ′ . For each such u u u ,
supp κ r ( x ) ∩ ( u + a 0 + K ) = { u + a 0 + w ( u . x ) } . \operatorname{supp}\kappa_r(x)\cap(u+a_0+K)=\{u+a_0+w(u.x)\}. supp κ r ( x ) ∩ ( u + a 0 + K ) = { u + a 0 + w ( u . x )} . (ii) For the removed and inserted counting measures
κ r − ( x ) : = δ supp κ ( x ) ∖ supp κ r ( x ) , κ r + ( x ) : = δ supp κ r ( x ) ∖ supp κ ( x ) . \kappa_r^-(x):=\delta_{\operatorname{supp}\kappa(x)\setminus\operatorname{supp}\kappa_r(x)},\qquad
\kappa_r^+(x):=\delta_{\operatorname{supp}\kappa_r(x)\setminus\operatorname{supp}\kappa(x)}. κ r − ( x ) := δ supp κ ( x ) ∖ supp κ r ( x ) , κ r + ( x ) := δ supp κ r ( x ) ∖ supp κ ( x ) .
For every fixed m m m and all sufficiently large r r r ,
∫ X N Q m ( κ r + ( x ) ) d μ ( x ) ≤ 3 m d ν r ( C r ) , N Q m ( κ r + ( x ) ) ≤ 3 , ν r ( C r ) ≤ Vol d ( B r / 2 ) − 1 . \int_X N_{Q_m}(\kappa_r^+(x))\,d\mu(x) \le3m^d\nu_r(C_r),\qquad N_{Q_m}(\kappa_r^+(x))\le3,\qquad\nu_r(C_r)\le\operatorname{Vol}_d(B_{r/2})^{-1}. ∫ X N Q m ( κ r + ( x )) d μ ( x ) ≤ 3 m d ν r ( C r ) , N Q m ( κ r + ( x )) ≤ 3 , ν r ( C r ) ≤ Vol d ( B r /2 ) − 1 .
Lemma 3.14 (Recovery from sparse markers). The realization κ r \kappa_r κ r in Lemma 3.13 is generating.
Proof. Fix ε > 0 \varepsilon>0 ε > 0 and take κ r \kappa_r κ r from Lemmas 3.13 and 3.14. For fixed m m m , set N r , m ± : = N Q m ∘ κ r ± N_{r,m}^{\pm}:=N_{Q_m}\circ\kappa_r^{\pm} N r , m ± := N Q m ∘ κ r ± . Since ν r ( C r ) ≤ Vol d ( B r / 2 ) − 1 → 0 \nu_r(C_r)\le\operatorname{Vol}_d(B_{r/2})^{-1}\to0 ν r ( C r ) ≤ Vol d ( B r /2 ) − 1 → 0 ,
∫ X N r , m ± d μ ⟶ 0 ( r → ∞ ) . \int_X N_{r,m}^{\pm}\,d\mu\longrightarrow0\qquad(r\to\infty). ∫ X N r , m ± d μ ⟶ 0 ( r → ∞ ) .
The inequalities
( N r , m − ) 2 ≤ ( N Q m ∘ κ ) 2 1 { N r , m − > 0 } , ( N r , m + ) 2 ≤ 3 N r , m + (N_{r,m}^{-})^2\le(N_{Q_m}\circ\kappa)^2\mathbf{1}_{\{N_{r,m}^{-}>0\}},\qquad(N_{r,m}^{+})^2\le3N_{r,m}^{+} ( N r , m − ) 2 ≤ ( N Q m ∘ κ ) 2 1 { N r , m − > 0 } , ( N r , m + ) 2 ≤ 3 N r , m +
and μ { N r , m − > 0 } ≤ ∫ X N r , m − d μ \mu\{N_{r,m}^{-}>0\}\le\int_XN_{r,m}^{-}\,d\mu μ { N r , m − > 0 } ≤ ∫ X N r , m − d μ imply ∥ N r , m − ∥ 2 + ∥ N r , m + ∥ 2 → 0 \|N_{r,m}^{-}\|_2+\|N_{r,m}^{+}\|_2\to0 ∥ N r , m − ∥ 2 + ∥ N r , m + ∥ 2 → 0 . For φ ∈ C c ( R d ) \varphi\in C_c(\mathbb{R}^d) φ ∈ C c ( R d ) with support in Q m Q_m Q m ,
∣ S φ ( κ r ) − S φ ( κ ) ∣ ≤ ∥ φ ∥ ∞ ( N r , m − + N r , m + ) . \lvert S_\varphi(\kappa_r)-S_\varphi(\kappa)\rvert\le\|\varphi\|_\infty(N_{r,m}^{-}+N_{r,m}^{+}). ∣ S φ ( κ r ) − S φ ( κ )∣ ≤ ∥ φ ∥ ∞ ( N r , m − + N r , m + ) .
Moreover,
μ { κ r ∣ Q m ≠ κ ∣ Q m } ≤ ∫ X ( N r , m − + N r , m + ) d μ ⟶ 0. \mu\{\kappa_r|_{Q_m}\ne\kappa|_{Q_m}\}\le\int_X(N_{r,m}^{-}+N_{r,m}^{+})\,d\mu\longrightarrow0. μ { κ r ∣ Q m = κ ∣ Q m } ≤ ∫ X ( N r , m − + N r , m + ) d μ ⟶ 0.
Hence κ r → κ \kappa_r\to\kappa κ r → κ in measure, and d r o o t , 2 ( κ r , κ ) → 0 d_{\mathrm{root},2}(\kappa_r,\kappa)\to0 d root , 2 ( κ r , κ ) → 0 . Choose r r r large enough that κ r ∈ U \kappa_r\in\mathcal{U} κ r ∈ U . □
Proof of Lemma 3.13
We first choose a vector v ∗ v_* v ∗ that is not the difference of any two original configuration points, almost surely. The inserted pairs with difference v ∗ v_* v ∗ can then be distinguished from the retained points.
By Lemma 2.1 , we may restrict a a a and κ \kappa κ to an invariant conull Borel set X 0 X_0 X 0 on which a a a is free and κ \kappa κ is Borel and pointwise equivariant. Write Λ x : = supp κ ( x ) \Lambda_x:=\operatorname{supp}\kappa(x) Λ x := supp κ ( x ) for x ∈ X 0 x\in X_0 x ∈ X 0 . For a bounded Borel set D ⊂ R d D\subset\mathbb{R}^d D ⊂ R d , define
α ( D ) : = ∫ X ∑ p ∈ Λ x ∩ Q 1 ∑ q ∈ Λ x q ≠ p 1 D ( q − p ) d μ ( x ) . \alpha(D):=\int_{X}\sum_{p\in\Lambda_x\cap Q_1}\sum_{\substack{q\in\Lambda_x\\q\ne p}}\mathbf{1}_D(q-p)\,d\mu(x). α ( D ) := ∫ X p ∈ Λ x ∩ Q 1 ∑ q ∈ Λ x q = p ∑ 1 D ( q − p ) d μ ( x ) .
Its integrand is bounded by N Q 1 ( κ ( x ) ) N Q 1 + D ( κ ( x ) ) N_{Q_1}(\kappa(x))N_{Q_1+D}(\kappa(x)) N Q 1 ( κ ( x )) N Q 1 + D ( κ ( x )) , so α \alpha α is locally finite by the local second-moment assumption. Choose v ∗ ∈ B ε ∖ { 0 } v_*\in B_\varepsilon\setminus\{0\} v ∗ ∈ B ε ∖ { 0 } with α ( { v ∗ } ) = 0 \alpha(\{v_*\})=0 α ({ v ∗ }) = 0 . Almost surely, Λ x \Lambda_x Λ x has no ordered pair ( p , q ) (p,q) ( p , q ) with q − p = v ∗ q-p=v_* q − p = v ∗ and p ∈ Q 1 p\in Q_1 p ∈ Q 1 . Translation invariance and the integer translates of Q 1 Q_1 Q 1 show that, after restricting X 0 X_0 X 0 further, no two points of Λ x \Lambda_x Λ x differ by v ∗ v_* v ∗ for any x ∈ X 0 x\in X_0 x ∈ X 0 .
Cross-section and three-point sets. Fix r > 8 ε r>8\varepsilon r > 8 ε . Choose an r r r -separated cocompact Borel cross-section C r ⊂ X 0 C_r\subset X_0 C r ⊂ X 0 and its Voronoi cells W c W_c W c by (5) . The return relation
R r : = { ( x , u ) ∈ X 0 × R d : u . x ∈ C r } \mathcal{R}_r:=\{(x,u)\in X_0\times\mathbb{R}^d:u.x\in C_r\} R r := {( x , u ) ∈ X 0 × R d : u . x ∈ C r }
is Borel because the action and C r C_r C r are Borel. Each section ( C r ) x (C_r)_x ( C r ) x is r r r -separated and therefore countable. Lusin–Novikov [24 Theorem 18.10] gives Borel sets D j ⊂ X 0 D_j\subset X_0 D j ⊂ X 0 and Borel maps u j : D j → R d u_j:D_j\to\mathbb{R}^d u j : D j → R d whose disjoint graphs cover R r \mathcal{R}_r R r . In particular, ( C r ) x = { u j ( x ) : x ∈ D j } (C_r)_x=\{u_j(x):x\in D_j\} ( C r ) x = { u j ( x ) : x ∈ D j } . Let ν r \nu_r ν r be the transverse measure on C r C_r C r . The cross-section and mass-transport formulas [6 Theorem 1.14] give
1 = ∫ C r Vol d ( W c ) d ν r ( c ) , ρ = ∫ C r # ( Λ c ∩ W c ) d ν r ( c ) . 1=\int_{C_r}\operatorname{Vol}_d(W_c)\,d\nu_r(c),\qquad\rho=\int_{C_r}\#(\Lambda_c\cap W_c)\,d\nu_r(c). 1 = ∫ C r Vol d ( W c ) d ν r ( c ) , ρ = ∫ C r # ( Λ c ∩ W c ) d ν r ( c ) .
Consequently, ν r ( C r ) ≤ Vol d ( B r / 2 ) − 1 \nu_r(C_r)\le\operatorname{Vol}_d(B_{r/2})^{-1} ν r ( C r ) ≤ Vol d ( B r /2 ) − 1 . The relation
P : = { ( x , p ) ∈ X 0 × R d : p ∈ Λ x } \mathcal{P}:=\{(x,p)\in X_0\times\mathbb{R}^d:p\in\Lambda_x\} P := {( x , p ) ∈ X 0 × R d : p ∈ Λ x }
is Borel: since Λ x \Lambda_x Λ x is closed, p ∈ Λ x p \in\Lambda_x p ∈ Λ x if and only if N B 1 / n ( θ p κ ( x ) ) > 0 N_{B_{1/n}}(\theta_p\kappa(x)) > 0 N B 1/ n ( θ p κ ( x )) > 0 for every n ≥ 1 n \ge1 n ≥ 1 . The latter condition is Borel in ( x , p ) (x,p) ( x , p ) by the Borel map ( x , p ) ↦ θ p κ ( x ) (x,p) \mapsto\theta_p\kappa(x) ( x , p ) ↦ θ p κ ( x ) . Its sections are countable by local finiteness. Applying Lusin–Novikov to P \mathcal{P} P , choose Borel sets H j ⊂ X 0 H_j \subset X_0 H j ⊂ X 0 and Borel maps p j : H j → R d p_j : H_j \to\mathbb{R}^d p j : H j → R d whose disjoint graphs cover P \mathcal{P} P . Thus Λ x = { p j ( x ) : x ∈ H j } \Lambda_x = \{p_j(x) : x \in H_j\} Λ x = { p j ( x ) : x ∈ H j } .
We need a positive transverse-measure set of cells containing at least three original points. Take r r r sufficiently large that 2 ν r ( C r ) < ρ 2\nu_r(C_r) < \rho 2 ν r ( C r ) < ρ , and set
E r : = { c ∈ C r : # ( Λ c ∩ W c ) ≥ 3 } . E_r := \{c \in C_r : \#(\Lambda_c \cap W_c) \ge3\}. E r := { c ∈ C r : # ( Λ c ∩ W c ) ≥ 3 } .
The displayed description of Λ c \Lambda_c Λ c and the Borel relation defining W c W_c W c show that E r E_r E r is Borel. If ν r ( E r ) = 0 \nu_r(E_r) = 0 ν r ( E r ) = 0 , the second integral above is at most 2 ν r ( C r ) < ρ 2\nu_r(C_r) < \rho 2 ν r ( C r ) < ρ , a contradiction. Thus ν r ( E r ) > 0 \nu_r(E_r) > 0 ν r ( E r ) > 0 . For each c ∈ E r c \in E_r c ∈ E r , let F ( c ) F(c) F ( c ) consist of the three points p j ( c ) ∈ W c p_j(c) \in W_c p j ( c ) ∈ W c with the smallest indices j j j . The maps p j p_j p j make c ↦ F ( c ) c \mapsto F(c) c ↦ F ( c ) Borel. Choose a compact Lebesgue-null Cantor set
K ⊂ B ε ∖ { − v ∗ , 0 , v ∗ , 2 v ∗ } K \subset B_\varepsilon\setminus\{-v_*, 0, v_*, 2v_*\} K ⊂ B ε ∖ { − v ∗ , 0 , v ∗ , 2 v ∗ }
and a Borel injection w : E r → K w : E_r \to K w : E r → K , using [24 Theorem 15.6] when E r E_r E r is uncountable.
Choice of the inserted points. For c ∈ E r c \in E_r c ∈ E r and b ∈ B ε b \in B_\varepsilon b ∈ B ε , put A b ( c ) : = { b , b + v ∗ , b + w ( c ) } A_b(c) := \{b, b+v_*, b+w(c)\} A b ( c ) := { b , b + v ∗ , b + w ( c )} . We choose one vector b b b for all cells, excluding the values for which an inserted point would coincide with an original point, create an unintended pair of difference v ∗ v_* v ∗ , or place an original point in b + K b+K b + K . Let B c B_c B c be the set of b ∈ B ε b \in B_\varepsilon b ∈ B ε satisfying at least one of
A b ( c ) ∩ Λ c ≠ ∅ , A_b(c) \cap\Lambda_c \ne\varnothing, A b ( c ) ∩ Λ c = ∅ ,
∃ p ∈ A b ( c ) , q ∈ Λ c ∖ F ( c ) : q − p ∈ { − v ∗ , v ∗ } , \exists p \in A_b(c),\ q \in\Lambda_c \setminus F(c) : q-p \in\{-v_*,v_*\}, ∃ p ∈ A b ( c ) , q ∈ Λ c ∖ F ( c ) : q − p ∈ { − v ∗ , v ∗ } ,
( b + K ) ∩ Λ c ≠ ∅ . (b+K) \cap\Lambda_c \ne\varnothing. ( b + K ) ∩ Λ c = ∅ .
The first two conditions hold for at most countably many b b b . The set of b b b satisfying the third condition is contained in ⋃ q ∈ Λ c ( q − K ) \bigcup_{q \in\Lambda_c}(q-K) ⋃ q ∈ Λ c ( q − K ) and is Lebesgue null. The set { ( c , b ) : c ∈ E r , b ∈ B c } \{(c,b) : c \in E_r,\ b \in B_c\} {( c , b ) : c ∈ E r , b ∈ B c } is Borel. Indeed, in each condition defining B c B_c B c we can replace a quantifier over Λ c \Lambda_c Λ c by the countable disjunction over p j ( c ) p_j(c) p j ( c ) for which c ∈ H j c \in H_j c ∈ H j ; the condition p j ( c ) ∈ F ( c ) p_j(c) \in F(c) p j ( c ) ∈ F ( c ) is Borel by its construction, and the third condition becomes p j ( c ) − b ∈ K p_j(c)-b \in K p j ( c ) − b ∈ K for some such j j j . Fubini’s theorem therefore gives
∫ B ε ν r { c ∈ E r : b ∈ B c } d b = 0. \int_{B_\varepsilon} \nu_r\{c \in E_r : b \in B_c\}\,db = 0. ∫ B ε ν r { c ∈ E r : b ∈ B c } d b = 0.
Fix a 0 ∈ B ε a_0 \in B_\varepsilon a 0 ∈ B ε such that ν r { c ∈ E r : a 0 ∈ B c } = 0 \nu_r\{c \in E_r : a_0 \in B_c\} = 0 ν r { c ∈ E r : a 0 ∈ B c } = 0 , and define E r ′ : = { c ∈ E r : a 0 ∉ B c } E'_r := \{c \in E_r : a_0 \notin B_c\} E r ′ := { c ∈ E r : a 0 ∈ / B c } . Then
ν r ( E r ′ ) = ν r ( E r ) > 0. \nu_r(E'_r) = \nu_r(E_r) > 0. ν r ( E r ′ ) = ν r ( E r ) > 0.
Definition and measurability of the modified realization. We now replace F ( c ) F(c) F ( c ) by A a 0 ( c ) A_{a_0}(c) A a 0 ( c ) in each cell with c ∈ E r ′ c \in E'_r c ∈ E r ′ , leaving all other cells unchanged. For c ∈ C r c \in C_r c ∈ C r , define the finite set
Λ c W : = { ( ( Λ c ∩ W c ) ∖ F ( c ) ) ∪ A a 0 ( c ) , c ∈ E r ′ , Λ c ∩ W c , c ∉ E r ′ . \Lambda_c^W := \begin{cases}
((\Lambda_c \cap W_c) \setminus F(c)) \cup A_{a_0}(c), & c \in E'_r,\\
\Lambda_c \cap W_c, & c \notin E'_r.
\end{cases} Λ c W := { (( Λ c ∩ W c ) ∖ F ( c )) ∪ A a 0 ( c ) , Λ c ∩ W c , c ∈ E r ′ , c ∈ / E r ′ .
For x ∈ X 0 x \in X_0 x ∈ X 0 , set
supp κ r ( x ) : = ⋃ u ∈ ( C r ) x ( u + Λ u . x W ) . \operatorname{supp}\kappa_r(x) := \bigcup_{u \in(C_r)_x}(u+\Lambda^W_{u.x}). supp κ r ( x ) := u ∈ ( C r ) x ⋃ ( u + Λ u . x W ) .
Since a 0 , v ∗ , w ( c ) ∈ B ε a_0,v_*,w(c) \in B_\varepsilon a 0 , v ∗ , w ( c ) ∈ B ε , we have A a 0 ( c ) ⊂ B 2 ε ⊂ W c A_{a_0}(c) \subset B_{2\varepsilon} \subset W_c A a 0 ( c ) ⊂ B 2 ε ⊂ W c . The three points of A a 0 ( c ) A_{a_0}(c) A a 0 ( c ) are distinct because w ( c ) ∉ { 0 , v ∗ } w(c) \notin\{0,v_*\} w ( c ) ∈ / { 0 , v ∗ } . For c ∈ E r ′ c \in E'_r c ∈ E r ′ , the definition of B c B_c B c gives A a 0 ( c ) ∩ Λ c = ∅ A_{a_0}(c) \cap\Lambda_c = \varnothing A a 0 ( c ) ∩ Λ c = ∅ . Thus Λ c W \Lambda_c^W Λ c W contains exactly the points of Λ c ∩ W c \Lambda_c \cap W_c Λ c ∩ W c outside F ( c ) F(c) F ( c ) , together with the three points of A a 0 ( c ) A_{a_0}(c) A a 0 ( c ) . The partition by the sets u + W u . x u+W_{u.x} u + W u . x makes the union defining κ r ( x ) \kappa_r(x) κ r ( x ) disjoint. It is locally finite: the retained points form a subset of Λ x \Lambda_x Λ x , while any point of u + A a 0 ( u . x ) u+A_{a_0}(u.x) u + A a 0 ( u . x ) is within 2 ε 2\varepsilon 2 ε of an r r r -separated return vector u u u . The maps p j p_j p j , the Borel relation defining W c W_c W c , and the three points of A a 0 ( c ) A_{a_0}(c) A a 0 ( c ) show that c ↦ Λ c W c \mapsto\Lambda^W_c c ↦ Λ c W is a Borel finite-set map. For every bounded Borel set D D D ,
N D ( κ r ( x ) ) = ∑ j : x ∈ D j # ( ( u j ( x ) + Λ u j ( x ) . x W ) ∩ D ) . N_D(\kappa_r(x))=\sum_{j:x\in D_j}\#\left((u_j(x)+\Lambda^W_{u_j(x).x})\cap D\right). N D ( κ r ( x )) = j : x ∈ D j ∑ # ( ( u j ( x ) + Λ u j ( x ) . x W ) ∩ D ) .
Each summand is Borel, so κ r \kappa_r κ r is Borel. The identities ( C r ) t . x = ( C r ) x − t (C_r)_{t.x}=(C_r)_x-t ( C r ) t . x = ( C r ) x − t and ( u − t ) . ( t . x ) = u . x (u-t).(t.x)=u.x ( u − t ) . ( t . x ) = u . x prove equivariance.
Point counts and second moments. Each replacement removes and inserts exactly three points. Consequently, for every c ∈ C r c\in C_r c ∈ C r we have # Λ c W = # ( Λ c ∩ W c ) \#\Lambda^W_c=\#(\Lambda_c\cap W_c) # Λ c W = # ( Λ c ∩ W c ) . The cross-section mass-transport formula gives
∫ X N Q m ( κ r ( x ) ) d μ ( x ) = m d ∫ C r # Λ c W d ν r ( c ) = ρ m d . \int_X N_{Q_m}(\kappa_r(x))\,d\mu(x)=m^d\int_{C_r}\#\Lambda^W_c\,d\nu_r(c)=\rho m^d. ∫ X N Q m ( κ r ( x )) d μ ( x ) = m d ∫ C r # Λ c W d ν r ( c ) = ρ m d .
Translation invariance of the intensity measure then gives intensity ρ \rho ρ on every bounded Borel set. The points of Λ x \Lambda_x Λ x removed by the construction are precisely those in
⋃ u ∈ ( C r ) x u . x ∈ E r ′ ( u + F ( u . x ) ) . \bigcup_{\substack{u\in(C_r)_x\\u.x\in E'_r}}(u+F(u.x)). u ∈ ( C r ) x u . x ∈ E r ′ ⋃ ( u + F ( u . x )) .
and the points of supp κ r ( x ) \operatorname{supp}\kappa_r(x) supp κ r ( x ) not in Λ x \Lambda_x Λ x are precisely those in
⋃ u ∈ ( C r ) x u . x ∈ E r ′ ( u + A a 0 ( u . x ) ) . \bigcup_{\substack{u\in(C_r)_x\\u.x\in E'_r}}(u+A_{a_0}(u.x)). u ∈ ( C r ) x u . x ∈ E r ′ ⋃ ( u + A a 0 ( u . x )) .
Both unions are disjoint. Applying mass transport to their counting measures yields, for every m ≥ 1 m\geq1 m ≥ 1 ,
∫ X N Q m ( κ r − ( x ) ) d μ ( x ) = ∫ X N Q m ( κ r + ( x ) ) d μ ( x ) = 3 m d ν r ( E r ′ ) . \int_X N_{Q_m}(\kappa_r^-(x))\,d\mu(x)=\int_X N_{Q_m}(\kappa_r^+(x))\,d\mu(x)=3m^d\nu_r(E'_r). ∫ X N Q m ( κ r − ( x )) d μ ( x ) = ∫ X N Q m ( κ r + ( x )) d μ ( x ) = 3 m d ν r ( E r ′ ) .
If p ∈ ( u + A a 0 ( u . x ) ) ∩ Q m p\in(u+A_{a_0}(u.x))\cap Q_m p ∈ ( u + A a 0 ( u . x )) ∩ Q m and q ∈ ( v + A a 0 ( v . x ) ) ∩ Q m q\in(v+A_{a_0}(v.x))\cap Q_m q ∈ ( v + A a 0 ( v . x )) ∩ Q m , then ∥ u − p ∥ , ∥ v − q ∥ < 2 ε \lVert u-p\rVert,\lVert v-q\rVert<2\varepsilon ∥ u − p ∥ , ∥ v − q ∥ < 2 ε and ∥ u − v ∥ ≤ d m + 4 ε \lVert u-v\rVert\leq\sqrt{d}\,m+4\varepsilon ∥ u − v ∥ ≤ d m + 4 ε . For r > d m + 4 ε r>\sqrt{d}\,m+4\varepsilon r > d m + 4 ε , the r r r -separation forces u = v u=v u = v . Thus N Q m ( κ r + ( x ) ) ≤ 3 N_{Q_m}(\kappa_r^+(x))\leq3 N Q m ( κ r + ( x )) ≤ 3 . For each fixed r r r , m m m , the number of r r r -separated vectors in Q m + B 2 ε Q_m+B_{2\varepsilon} Q m + B 2 ε is bounded by a finite constant C r , m C_{r,m} C r , m , independently of x x x . Therefore
N Q m ( κ r ( x ) ) ≤ N Q m ( κ ( x ) ) + 3 C r , m , N_{Q_m}(\kappa_r(x))\leq N_{Q_m}(\kappa(x))+3C_{r,m}, N Q m ( κ r ( x )) ≤ N Q m ( κ ( x )) + 3 C r , m ,
which proves that κ r \kappa_r κ r has local second moments.
Identification of the pairs with difference v ∗ v_* v ∗ . We verify that the pairs with difference v ∗ v_* v ∗ are exactly those asserted in part (i). Fix x ∈ X 0 x\in X_0 x ∈ X 0 . No two points of Λ x \Lambda_x Λ x differ by v ∗ v_* v ∗ , so the points retained from Λ x \Lambda_x Λ x contain no such pair. For each u ∈ ( C r ) x u\in(C_r)_x u ∈ ( C r ) x with c : = u . x ∈ E r ′ c:=u.x\in E'_r c := u . x ∈ E r ′ , the only ordered pair of points of u + A a 0 ( c ) u+A_{a_0}(c) u + A a 0 ( c ) with difference v ∗ v_* v ∗ is ( u + a 0 , u + a 0 + v ∗ ) (u+a_0,u+a_0+v_*) ( u + a 0 , u + a 0 + v ∗ ) : all other possibilities would require w ( c ) ∈ { − v ∗ , 0 , v ∗ , 2 v ∗ } w(c)\in\{-v_*,0,v_*,2v_*\} w ( c ) ∈ { − v ∗ , 0 , v ∗ , 2 v ∗ } . The second condition defining B c B_c B c excludes a pair of difference v ∗ v_* v ∗ between u + A a 0 ( c ) u+A_{a_0}(c) u + A a 0 ( c ) and u + ( ( Λ c ∩ W c ) ∖ F ( c ) ) u+((\Lambda_c\cap W_c)\setminus F(c)) u + (( Λ c ∩ W c ) ∖ F ( c )) . If q ∈ Λ x ∖ ( u + W c ) q\in\Lambda_x\setminus(u+W_c) q ∈ Λ x ∖ ( u + W c ) , then ∥ q − u ∥ ≥ r / 2 \lVert q-u\rVert\geq r/2 ∥ q − u ∥ ≥ r /2 because B r / 2 ∘ ⊂ W c B^\circ_{r/2}\subset W_c B r /2 ∘ ⊂ W c . For p ∈ u + A a 0 ( c ) p\in u+A_{a_0}(c) p ∈ u + A a 0 ( c ) we have ∥ p − u ∥ < 2 ε \lVert p-u\rVert<2\varepsilon ∥ p − u ∥ < 2 ε , so ∥ p − q ∥ ≥ r / 2 − 2 ε > ∥ v ∗ ∥ \lVert p-q\rVert\geq r/2-2\varepsilon>\lVert v_*\rVert ∥ p − q ∥ ≥ r /2 − 2 ε > ∥ v ∗ ∥ . If u , v ∈ ( C r ) x u,v\in(C_r)_x u , v ∈ ( C r ) x are distinct and p ∈ u + A a 0 ( u . x ) p\in u+A_{a_0}(u.x) p ∈ u + A a 0 ( u . x ) , q ∈ v + A a 0 ( v . x ) q\in v+A_{a_0}(v.x) q ∈ v + A a 0 ( v . x ) , then ∥ p − q ∥ ≥ r − 4 ε > ∥ v ∗ ∥ \lVert p-q\rVert\geq r-4\varepsilon>\lVert v_*\rVert ∥ p − q ∥ ≥ r − 4 ε > ∥ v ∗ ∥ . Thus the ordered pairs of difference v ∗ v_* v ∗ in supp κ r ( x ) \operatorname{supp}\kappa_r(x) supp κ r ( x ) are exactly those in (i).
For u ∈ ( C r ) x u\in(C_r)_x u ∈ ( C r ) x with c : = u . x ∈ E r ′ c:=u.x\in E'_r c := u . x ∈ E r ′ , we have ( a 0 + K ) ∩ Λ c = ∅ (a_0+K)\cap\Lambda_c=\varnothing ( a 0 + K ) ∩ Λ c = ∅ by the definition of E r ′ E'_r E r ′ . The set a 0 + K a_0+K a 0 + K lies in B 2 ε ⊂ W c B_{2\varepsilon}\subset W_c B 2 ε ⊂ W c . Since 0 , v ∗ ∉ K 0,v_*\notin K 0 , v ∗ ∈ / K , its intersection with Λ c W \Lambda^W_c Λ c W is { a 0 + w ( c ) } \{a_0+w(c)\} { a 0 + w ( c )} . The partition by the sets u + W u . x u+W_{u.x} u + W u . x therefore gives
supp κ r ( x ) ∩ ( u + a 0 + K ) = { u + a 0 + w ( c ) } . \operatorname{supp}\kappa_r(x)\cap(u+a_0+K)=\{u+a_0+w(c)\}. supp κ r ( x ) ∩ ( u + a 0 + K ) = { u + a 0 + w ( c )} .
This proves (i) and (ii).
Maximal and linear rigidity
Proof of Lemma 3.14
First we show that almost every configuration contains an inserted triple. We then construct a Borel map recovering x x x from any such triple in κ r ( x ) \kappa_r(x) κ r ( x ) .
Let
S r : = { x ∈ X 0 : ∃ u ∈ ( C r ) x with u . x ∈ E r ′ } . S_r := \{x \in X_0 : \exists u \in(C_r)_x \text{ with } u.x \in E'_r\}. S r := { x ∈ X 0 : ∃ u ∈ ( C r ) x with u . x ∈ E r ′ } .
Retain the partial Borel maps u j : D j → R d u_j : D_j \to\mathbb{R}^d u j : D j → R d from the proof of Lemma 3.13 . They give
S r = ⋃ j ≥ 1 { x ∈ D j : u j ( x ) . x ∈ E r ′ } . S_r = \bigcup_{j \ge1} \{x \in D_j : u_j(x).x \in E'_r\}. S r = j ≥ 1 ⋃ { x ∈ D j : u j ( x ) . x ∈ E r ′ } .
Thus S r S_r S r is Borel. The return-time identity makes it invariant. Every point t . c t.c t . c with c ∈ E r ′ c \in E'_r c ∈ E r ′ and t ∈ W c t \in W_c t ∈ W c belongs to S r S_r S r . The cross-section formula therefore gives
μ ( S r ) ≥ ∫ E r ′ Vol d ( W c ) d ν r ( c ) ≥ Vol d ( B r / 2 ) ν r ( E r ′ ) > 0. \mu(S_r) \ge\int_{E'_r} \operatorname{Vol}_d(W_c)\,d\nu_r(c) \ge\operatorname{Vol}_d(B_{r/2})\,\nu_r(E'_r) > 0. μ ( S r ) ≥ ∫ E r ′ Vol d ( W c ) d ν r ( c ) ≥ Vol d ( B r /2 ) ν r ( E r ′ ) > 0.
Since a a a is ergodic, μ ( S r ) = 1 \mu(S_r) = 1 μ ( S r ) = 1 .
Borel reconstruction. Fix x ∈ S r x \in S_r x ∈ S r and write Λ x r : = supp κ r ( x ) \Lambda_x^r := \operatorname{supp}\kappa_r(x) Λ x r := supp κ r ( x ) . By Lemma 3.13 (i), there is an ordered pair ( z , z + v ∗ ) ∈ ( Λ x r ) 2 (z,z+v_*) \in(\Lambda_x^r)^2 ( z , z + v ∗ ) ∈ ( Λ x r ) 2 and exactly one point p ∈ Λ x r ∩ ( z + K ) p \in\Lambda_x^r \cap(z+K) p ∈ Λ x r ∩ ( z + K ) . For any such pair, the same lemma gives c ∈ E r ′ c \in E'_r c ∈ E r ′ and u ∈ ( C r ) x u \in(C_r)_x u ∈ ( C r ) x with
z = u + a 0 , p = z + w ( c ) , c = u . x . z = u + a_0, \qquad p = z + w(c), \qquad c = u.x. z = u + a 0 , p = z + w ( c ) , c = u . x .
The injection w w w determines c c c uniquely from p − z p-z p − z . It follows that
x = ( a 0 − z ) . w − 1 ( p − z ) . x = (a_0-z).w^{-1}(p-z). x = ( a 0 − z ) . w − 1 ( p − z ) .
Lusin–Souslin [24 Theorem 15.1] implies that w ( E r ′ ) w(E'_r) w ( E r ′ ) is Borel and w − 1 : w ( E r ′ ) → E r ′ w^{-1} : w(E'_r) \to E'_r w − 1 : w ( E r ′ ) → E r ′ is Borel. Let H \mathcal{H} H be the set of triples ( ω , z , p ) ∈ Conf ( R d ) × R d × R d (\omega,z,p) \in\operatorname{Conf}(\mathbb{R}^d) \times\mathbb{R}^d \times\mathbb{R}^d ( ω , z , p ) ∈ Conf ( R d ) × R d × R d satisfying
z , z + v ∗ , p ∈ supp ω , p − z ∈ w ( E r ′ ) , N z + K ( ω ) = 1. z, z+v_*,p \in\operatorname{supp}\omega, \qquad p-z \in w(E'_r), \qquad N_{z+K}(\omega) = 1. z , z + v ∗ , p ∈ supp ω , p − z ∈ w ( E r ′ ) , N z + K ( ω ) = 1.
The relation H \mathcal{H} H is Borel: point incidence in a locally finite configuration is Borel, and N z + K ( ω ) = N K ( θ z ω ) N_{z+K}(\omega) = N_K(\theta_z\omega) N z + K ( ω ) = N K ( θ z ω ) is Borel by joint continuity of translation and Borel measurability of N K N_K N K . For each ω \omega ω , the section H ω \mathcal{H}_\omega H ω is countable because both z z z and p p p belong to the countable set supp ω \operatorname{supp}\omega supp ω . Lusin–Novikov [24 Theorem 18.10] therefore shows that the projection Ω \Omega Ω of H \mathcal{H} H onto the configuration space is Borel and provides a Borel map ω ↦ ( z ( ω ) , p ( ω ) ) \omega\mapsto(z(\omega),p(\omega)) ω ↦ ( z ( ω ) , p ( ω )) on Ω \Omega Ω with ( ω , z ( ω ) , p ( ω ) ) ∈ H (\omega,z(\omega),p(\omega)) \in\mathcal{H} ( ω , z ( ω ) , p ( ω )) ∈ H . Consequently,
ω ⟼ ( a 0 − z ( ω ) ) . w − 1 ( p ( ω ) − z ( ω ) ) \omega\longmapsto(a_0-z(\omega)).w^{-1}(p(\omega)-z(\omega)) ω ⟼ ( a 0 − z ( ω )) . w − 1 ( p ( ω ) − z ( ω ))
is a Borel map from Ω \Omega Ω to X 0 X_0 X 0 . For every x ∈ S r x \in S_r x ∈ S r , Lemma 3.13 (i) implies that κ r ( x ) ∈ Ω \kappa_r(x) \in\Omega κ r ( x ) ∈ Ω and every pair ( z , p ) ∈ H κ r ( x ) (z,p) \in\mathcal{H}_{\kappa_r(x)} ( z , p ) ∈ H κ r ( x ) gives the value x x x . Thus this map sends κ r ( x ) \kappa_r(x) κ r ( x ) to x x x on the conull set S r S_r S r , and κ r \kappa_r κ r is generating. □
The density statements in Propositions 3.3 , 3.11 and 3.12 , together with Proposition 3.2 , give
{ κ ∈ R ρ , 2 a , i s o ( X , μ ) : κ ∗ μ ∈ I H y p ρ } \{\kappa\in\mathbb{R}^{a,\mathrm{iso}}_{\rho,2}(X,\mu) : \kappa_*\mu\in\mathrm{IHyp}_\rho\} { κ ∈ R ρ , 2 a , iso ( X , μ ) : κ ∗ μ ∈ IHyp ρ }
is residual in R ρ , 2 a , i s o ( X , μ ) \mathbb{R}^{a,\mathrm{iso}}_{\rho,2}(X,\mu) R ρ , 2 a , iso ( X , μ ) for every a ∈ A d ( X , μ ) a \in\mathrm{Ad}(X,\mu) a ∈ Ad ( X , μ ) , and I H y p ρ \mathrm{IHyp}_\rho IHyp ρ is residual in E P P ρ , 2 \mathrm{EPP}_{\rho,2} EPP ρ , 2 . Taking α = 1 \alpha= 1 α = 1 in the definition of intermittent hyperuniformity gives
lim sup R → ∞ V η ( R ) Vol d ( B R ) = + ∞ ( η ∈ I H y p ρ ) , \limsup_{R \to\infty} \frac{V_\eta(R)}{\operatorname{Vol}_d(B_R)} = +\infty\qquad(\eta\in\mathrm{IHyp}_\rho), R → ∞ lim sup Vol d ( B R ) V η ( R ) = + ∞ ( η ∈ IHyp ρ ) ,
so H y p ρ ∩ E P P ρ , 2 \mathrm{Hyp}_\rho\cap\mathrm{EPP}_{\rho,2} Hyp ρ ∩ EPP ρ , 2 is meagre in E P P ρ , 2 \mathrm{EPP}_{\rho,2} EPP ρ , 2 .
Maximal and linear rigidity
We prove that maximal rigidity is a G δ G_\delta G δ condition and obtain density by sparse marker coding. For linear maximal rigidity, polynomial moment identities allow local linear statistics to be approximated in L 2 L^2 L 2 by exterior statistics.
Maximal rigidity
Proposition 4.1. The set MRig ρ \operatorname{MRig}_{\rho} MRig ρ is G δ G_\delta G δ in PP ρ , 2 \operatorname{PP}_{\rho,2} PP ρ , 2 . Therefore, for every a ∈ A d ( X , μ ) a \in A_d(X,\mu) a ∈ A d ( X , μ ) ,
{ κ ∈ R ρ , 2 a ( X , μ ) : κ ∗ μ ∈ MRig ρ } \left\{\kappa\in R_{\rho,2}^{a}(X,\mu):\kappa_*\mu\in\operatorname{MRig}_{\rho}\right\} { κ ∈ R ρ , 2 a ( X , μ ) : κ ∗ μ ∈ MRig ρ }
is G δ G_\delta G δ in R ρ , 2 a ( X , μ ) R_{\rho,2}^{a}(X,\mu) R ρ , 2 a ( X , μ ) .
Proof. Fix an integer m ≥ 1 m \ge1 m ≥ 1 . Choose a countable family of nonnegative functions f m , ℓ ∈ C c ( int B m ) f_{m,\ell} \in C_c(\operatorname{int} B_m) f m , ℓ ∈ C c ( int B m ) such that every nonnegative f ∈ C c ( int B m ) f \in C_c(\operatorname{int} B_m) f ∈ C c ( int B m ) admits uniform approximants from the family whose supports lie in a common compact subset of int B m \operatorname{int} B_m int B m , and put
F m , ℓ ( ω ) : = e − S f m , ℓ ( ω ) . F_{m,\ell}(\omega) := e^{-S_{f_{m,\ell}}(\omega)}. F m , ℓ ( ω ) := e − S f m , ℓ ( ω ) .
For f ≥ 0 f \ge0 f ≥ 0 continuous and compactly supported, e − S f e^{-S_f} e − S f is bounded and continuous in the vague topology, and e − S f e − S g = e − S f + g e^{-S_f}e^{-S_g}=e^{-S_{f+g}} e − S f e − S g = e − S f + g . Since Laplace functionals separate locally finite counting measures, the functions F m , ℓ F_{m,\ell} F m , ℓ generate the σ \sigma σ -algebra of ω ∣ int B m \omega|_{\operatorname{int} B_m} ω ∣ int B m . Since N ∂ B m = 0 N_{\partial B_m}=0 N ∂ B m = 0 almost surely by Lemma 2.3 , it also generates the σ \sigma σ -algebra of ω ∣ B m \omega|_{B_m} ω ∣ B m modulo null sets.
Choose an analogous countable family of nonnegative functions in C c ( R d ∖ B m ) C_c(\mathbb{R}^d \setminus B_m) C c ( R d ∖ B m ) , and let A m o u t \mathcal{A}^{\mathrm{out}}_m A m out be the unital algebra of finite rational linear combinations of finite products of their Laplace functionals. This is a countable family of bounded continuous functions. The algebra generates F B m c \mathcal{F}_{B_m^c} F B m c : compactly supported continuous functions on B m c B_m^c B m c determine the exterior configuration. Since this algebra contains the constants and generates F B m c \mathcal{F}_{B_m^c} F B m c , the functional monotone-class theorem gives its density in L 2 ( F B m c , η ) L^2(\mathcal{F}_{B_m^c},\eta) L 2 ( F B m c , η ) . Define
D m , ℓ ( η ) : = inf G ∈ A m o u t ∫ ∣ F m , ℓ − G ∣ 2 d η . D_{m,\ell}(\eta):=\inf_{G\in\mathcal{A}^{\mathrm{out}}_m}\int|F_{m,\ell}-G|^2\,d\eta. D m , ℓ ( η ) := G ∈ A m out inf ∫ ∣ F m , ℓ − G ∣ 2 d η .
For fixed G ∈ A m o u t G \in\mathcal{A}^{\mathrm{out}}_m G ∈ A m out , the function ∣ F m , ℓ − G ∣ 2 |F_{m,\ell}-G|^2 ∣ F m , ℓ − G ∣ 2 is bounded and continuous on configuration space. Consequently the map η ↦ ∫ ∣ F m , ℓ − G ∣ 2 d η \eta\mapsto\int|F_{m,\ell}-G|^2\,d\eta η ↦ ∫ ∣ F m , ℓ − G ∣ 2 d η is continuous, even in the weak topology. Since
{ η : D m , ℓ ( η ) < ε } = ⋃ G ∈ A m o u t { η : ∫ ∣ F m , ℓ − G ∣ 2 d η < ε } , \{\eta:D_{m,\ell}(\eta)<\varepsilon\}
=
\bigcup_{G\in\mathcal{A}^{\mathrm{out}}_m}
\left\{\eta:\int|F_{m,\ell}-G|^2\,d\eta<\varepsilon\right\}, { η : D m , ℓ ( η ) < ε } = G ∈ A m out ⋃ { η : ∫ ∣ F m , ℓ − G ∣ 2 d η < ε } ,
this set is open.
Because the L 2 ( η ) L^2(\eta) L 2 ( η ) -closure of A m o u t \mathcal{A}^{\mathrm{out}}_m A m out is L 2 ( F B m c , η ) L^2(\mathcal{F}_{B_m^c},\eta) L 2 ( F B m c , η ) , the equality D m , ℓ ( η ) = 0 D_{m,\ell}(\eta)=0 D m , ℓ ( η ) = 0 is equivalent to F m , ℓ F_{m,\ell} F m , ℓ being F B m c \mathcal{F}_{B_m^c} F B m c -measurable modulo η \eta η -null sets. The inside functionals generate the restriction ω ∣ B m \omega|_{B_m} ω ∣ B m , so
MRig ρ = ⋂ m , ℓ , k ≥ 1 { η ∈ PP ρ , 2 : D m , ℓ ( η ) < k − 1 } . \operatorname{MRig}_{\rho}
=
\bigcap_{m,\ell,k\ge1}
\left\{\eta\in\operatorname{PP}_{\rho,2}:D_{m,\ell}(\eta)<k^{-1}\right\}. MRig ρ = m , ℓ , k ≥ 1 ⋂ { η ∈ PP ρ , 2 : D m , ℓ ( η ) < k − 1 } .
If D m , ℓ ( η ) = 0 D_{m,\ell}(\eta)=0 D m , ℓ ( η ) = 0 for all ℓ \ell ℓ , ω ∣ B m \omega|_{B_m} ω ∣ B m is a Borel function of ω ∣ B m c \omega|_{B_m^c} ω ∣ B m c almost surely, by the Doob–Dynkin lemma. For bounded Borel A ⊂ B m A\subset B_m A ⊂ B m we have F B m c ⊂ F A c \mathcal{F}_{B_m^c}\subset\mathcal{F}_{A^c} F B m c ⊂ F A c , so ω ∣ A \omega|_A ω ∣ A is then also determined by ω ∣ A c \omega|_{A^c} ω ∣ A c . Hence MRig ρ \operatorname{MRig}_{\rho} MRig ρ is G δ G_\delta G δ , and continuity of κ ↦ κ ∗ μ \kappa\mapsto\kappa_*\mu κ ↦ κ ∗ μ from Proposition 2.6 gives the rooted assertion. □
Proposition 4.2. Let a ∈ A d ( X , μ ) a\in A_d(X,\mu) a ∈ A d ( X , μ ) , let κ ∈ R ρ , 2 a ( X , μ ) \kappa\in R_{\rho,2}^{a}(X,\mu) κ ∈ R ρ , 2 a ( X , μ ) , and let U \mathcal{U} U be a neighbourhood of κ \kappa κ . Then
{ λ ∈ U ∩ R ρ , 2 a , i s o ( X , μ ) : λ ∗ μ ∈ MRig ρ } ≠ ∅ . \left\{\lambda\in\mathcal{U}\cap R_{\rho,2}^{a,\mathrm{iso}}(X,\mu):\lambda_*\mu\in\operatorname{MRig}_{\rho}\right\}\ne\varnothing. { λ ∈ U ∩ R ρ , 2 a , iso ( X , μ ) : λ ∗ μ ∈ MRig ρ } = ∅ .
Proof. Let κ r \kappa_r κ r be the realizations from the proof of Proposition 3.12 . Choose r r r sufficiently large that κ r ∈ U \kappa_r \in\mathcal{U} κ r ∈ U , using κ r → κ \kappa_r \to\kappa κ r → κ in the rooted local L 2 L^2 L 2 topology. Let E r ′ ⊂ C r E'_r \subset C_r E r ′ ⊂ C r be the Borel set on which the three-point replacement of Lemma 3.13 is performed. It has positive transverse measure. Define
T x : = { u ∈ ( C r ) x : u . x ∈ E r ′ } . T_x := \{u \in(C_r)_x : u.x \in E'_r\}. T x := { u ∈ ( C r ) x : u . x ∈ E r ′ } .
For u ∈ T x u \in T_x u ∈ T x , with c : = u . x c := u.x c := u . x , the three inserted points are
u + a 0 , u + a 0 + v ∗ , u + a 0 + w ( c ) . u+a_0,\qquad u+a_0+v_*,\qquad u+a_0+w(c). u + a 0 , u + a 0 + v ∗ , u + a 0 + w ( c ) .
By Lemma 3.13 (i), for almost every x x x , every ordered pair ( z , z + v ∗ ) (z,z+v_*) ( z , z + v ∗ ) of points in supp κ r ( x ) \operatorname{supp}\kappa_r(x) supp κ r ( x ) has the form ( u + a 0 , u + a 0 + v ∗ ) (u+a_0,u+a_0+v_*) ( u + a 0 , u + a 0 + v ∗ ) for some u ∈ T x u \in T_x u ∈ T x .
The translation-invariant process x ↦ δ T x x \mapsto\delta_{T_x} x ↦ δ T x has intensity ν r ( E r ′ ) > 0 \nu_r(E'_r)>0 ν r ( E r ′ ) > 0 . The cardinality # T x \#T_x # T x is invariant under x ↦ t . x x \mapsto t.x x ↦ t . x and hence, by ergodicity, equals a constant k ∈ N 0 ∪ { ∞ } k \in\mathbb{N}_0 \cup\{\infty\} k ∈ N 0 ∪ { ∞ } for almost every x x x . If k < ∞ k<\infty k < ∞ , then
ν r ( E r ′ ) Vol d ( Q n ) = ∫ X # ( T x ∩ Q n ) d μ ( x ) ≤ k ( n ≥ 1 ) , \nu_r(E'_r)\operatorname{Vol}_d(Q_n)=\int_X \#(T_x\cap Q_n)\,d\mu(x)\leq k\qquad(n\geq1), ν r ( E r ′ ) Vol d ( Q n ) = ∫ X # ( T x ∩ Q n ) d μ ( x ) ≤ k ( n ≥ 1 ) ,
a contradiction as n → ∞ n\to\infty n → ∞ . Thus T x T_x T x is infinite almost surely. Since T x T_x T x is r r r -separated, it is unbounded.
Fix m ≥ 1 m\geq1 m ≥ 1 and put A : = B m A:=B_m A := B m . Set
M : = { a 0 , a 0 + v ∗ } ∪ ( a 0 + K ) . M:=\{a_0,a_0+v_*\}\cup(a_0+K). M := { a 0 , a 0 + v ∗ } ∪ ( a 0 + K ) .
Since M M M is bounded and T x T_x T x is unbounded, almost surely some u ∈ T x u\in T_x u ∈ T x satisfies u + M ⊂ A c u+M\subset A^c u + M ⊂ A c . The first two points of the corresponding triple are ( z , z + v ∗ ) (z,z+v_*) ( z , z + v ∗ ) , where z = u + a 0 z=u+a_0 z = u + a 0 ; its third point is in z + K z+K z + K . Let ω o u t : = κ r ( x ) ∣ B m c \omega^{\mathrm{out}}:=\kappa_r(x)|_{B_m^c} ω out := κ r ( x ) ∣ B m c , and define
Z m ( ω o u t ) : = { z ∈ supp ( ω o u t ) : z + v ∗ ∈ supp ( ω o u t ) , ( z + K ) ∩ B m = ∅ , N z + K ( ω o u t ) = 1 } . Z_m(\omega^{\mathrm{out}}):=\{z\in\operatorname{supp}(\omega^{\mathrm{out}}):z+v_*\in\operatorname{supp}(\omega^{\mathrm{out}}),\ (z+K)\cap B_m=\varnothing,\ N_{z+K}(\omega^{\mathrm{out}})=1\}. Z m ( ω out ) := { z ∈ supp ( ω out ) : z + v ∗ ∈ supp ( ω out ) , ( z + K ) ∩ B m = ∅ , N z + K ( ω out ) = 1 } .
For x x x in the conull set on which Lemma 3.13 (i) holds, every z ∈ Z m ( ω o u t ) z\in Z_m(\omega^{\mathrm{out}}) z ∈ Z m ( ω out ) determines an ordered pair ( z , z + v ∗ ) (z,z+v_*) ( z , z + v ∗ ) in supp κ r ( x ) \operatorname{supp}\kappa_r(x) supp κ r ( x ) . By that lemma there is u ∈ ( C r ) x u\in(C_r)_x u ∈ ( C r ) x with c : = u . x ∈ E r ′ c:=u.x\in E'_r c := u . x ∈ E r ′ and z = u + a 0 z=u+a_0 z = u + a 0 . The same lemma gives supp κ r ( x ) ∩ ( z + K ) = { z + w ( c ) } \operatorname{supp}\kappa_r(x)\cap(z+K)=\{z+w(c)\} supp κ r ( x ) ∩ ( z + K ) = { z + w ( c )} . Since z + K ⊂ B m c z+K\subset B_m^c z + K ⊂ B m c , this is also the unique point of ω o u t \omega^{\mathrm{out}} ω out in z + K z+K z + K . Conversely, if u ∈ T x u\in T_x u ∈ T x satisfies u + M ⊂ B m c u+M\subset B_m^c u + M ⊂ B m c , then u + a 0 ∈ Z m ( ω o u t ) u+a_0\in Z_m(\omega^{\mathrm{out}}) u + a 0 ∈ Z m ( ω out ) . Thus Z m ( κ r ( x ) ∣ B m c ) ≠ ∅ Z_m(\kappa_r(x)|_{B_m^c})\neq\varnothing Z m ( κ r ( x ) ∣ B m c ) = ∅ for almost every x x x .
Consider the subset of the product of configuration space and R d \mathbb{R}^d R d
L m : = { ( ω , z ) : z ∈ Z m ( ω ) } . \mathcal{L}_m:=\{(\omega,z):z\in Z_m(\omega)\}. L m := {( ω , z ) : z ∈ Z m ( ω )} .
This subset is Borel. Indeed, the conditions z , z + v ∗ ∈ supp ω z,z+v_*\in\operatorname{supp}\omega z , z + v ∗ ∈ supp ω are Borel in ( ω , z ) (\omega,z) ( ω , z ) by joint Borel measurability of point incidence, N z + K ( ω ) = N K ( θ z ω ) N_{z+K}(\omega)=N_K(\theta_z\omega) N z + K ( ω ) = N K ( θ z ω ) is Borel, and { z : ( z + K ) ∩ B m = ∅ } \{z:(z+K)\cap B_m=\varnothing\} { z : ( z + K ) ∩ B m = ∅ } is open because K K K and B m B_m B m are compact. Each vertical section of L m \mathcal{L}_m L m is countable, since it is contained in supp ω \operatorname{supp}\omega supp ω . Lusin–Novikov implies that Ω m : = { ω : Z m ( ω ) ≠ ∅ } \Omega_m:=\{\omega:Z_m(\omega)\neq\varnothing\} Ω m := { ω : Z m ( ω ) = ∅ } is Borel and supplies a Borel map z : Ω m → R d z:\Omega_m\to\mathbb{R}^d z : Ω m → R d satisfying z ( ω ) ∈ Z m ( ω ) z(\omega)\in Z_m(\omega) z ( ω ) ∈ Z m ( ω ) . The subset
{ ( ω , p ) : ω ∈ Ω m , p ∈ supp ω ∩ ( z ( ω ) + K ) } \{(\omega,p):\omega\in\Omega_m,\ p\in\operatorname{supp}\omega\cap(z(\omega)+K)\} {( ω , p ) : ω ∈ Ω m , p ∈ supp ω ∩ ( z ( ω ) + K )}
of Ω m × R d \Omega_m\times\mathbb{R}^d Ω m × R d is Borel and has singleton vertical sections. Another application of Lusin–Novikov gives its unique point as a Borel map p : Ω m → R d p:\Omega_m\to\mathbb{R}^d p : Ω m → R d . For ω = κ r ( x ) ∣ B m c \omega=\kappa_r(x)|_{B_m^c} ω = κ r ( x ) ∣ B m c and almost every x x x , the preceding identities give p ( ω ) − z ( ω ) = w ( c ) p(\omega)-z(\omega)=w(c) p ( ω ) − z ( ω ) = w ( c ) and
x = ( a 0 − z ( ω ) ) ⋅ w − 1 ( p ( ω ) − z ( ω ) ) . x=(a_0-z(\omega))\cdot w^{-1}(p(\omega)-z(\omega)). x = ( a 0 − z ( ω )) ⋅ w − 1 ( p ( ω ) − z ( ω )) .
By Lusin–Souslin [24 Theorem 15.1] , w ( E r ′ ) w(E'_r) w ( E r ′ ) is Borel and w − 1 : w ( E r ′ ) → E r ′ w^{-1}:w(E'_r)\to E'_r w − 1 : w ( E r ′ ) → E r ′ is Borel. The set Ω m ′ : = { ω ∈ Ω m : p ( ω ) − z ( ω ) ∈ w ( E r ′ ) } \Omega'_m:=\{\omega\in\Omega_m:p(\omega)-z(\omega)\in w(E'_r)\} Ω m ′ := { ω ∈ Ω m : p ( ω ) − z ( ω ) ∈ w ( E r ′ )} is Borel and contains κ r ( x ) ∣ B m c \kappa_r(x)|_{B_m^c} κ r ( x ) ∣ B m c for almost every x x x . The displayed formula defines a Borel reconstruction on Ω m ′ \Omega'_m Ω m ′ . Thus κ r ( x ) \kappa_r(x) κ r ( x ) is determined by its restriction to B m c B_m^c B m c . For any bounded Borel A A A , choose m m m with A ⊂ B m A\subset B_m A ⊂ B m ; then F B m c ⊂ F A c \mathcal{F}_{B_m^c}\subset\mathcal{F}_{A^c} F B m c ⊂ F A c . Hence the restriction to A A A is determined by its exterior, and ( κ r ) ∗ μ (\kappa_r)_*\mu ( κ r ) ∗ μ is maximally rigid. □ \square □
Corollary 4.3. For every a ∈ Ad d ( X , μ ) a \in\operatorname{Ad}_{d}(X,\mu) a ∈ Ad d ( X , μ ) ,
{ κ ∈ R ρ , 2 a , i s o ( X , μ ) : κ ∗ μ ∈ M R i g ρ } \left\{\kappa\in\mathbb{R}_{\rho,2}^{a,\mathrm{iso}}(X,\mu) : \kappa_*\mu\in\mathrm{MRig}_{\rho}\right\} { κ ∈ R ρ , 2 a , iso ( X , μ ) : κ ∗ μ ∈ MRig ρ } is residual in R ρ , 2 a , i s o ( X , μ ) \mathbb{R}_{\rho,2}^{a,\mathrm{iso}}(X,\mu) R ρ , 2 a , iso ( X , μ ) . Moreover, M R i g ρ ∩ E P P ρ , 2 \mathrm{MRig}_{\rho} \cap\mathrm{EPP}_{\rho,2} MRig ρ ∩ EPP ρ , 2 is residual in E P P ρ , 2 \mathrm{EPP}_{\rho,2} EPP ρ , 2 .
Proof. The rooted statement follows from Propositions 4.1 and 4.2. The global statement follows from Proposition 2.11. □ \square □
Linear maximal rigidity
Proposition 4.4. The set L M R i g ρ \mathrm{LMRig}_{\rho} LMRig ρ is G δ G_{\delta} G δ in P P ρ , 2 \mathrm{PP}_{\rho,2} PP ρ , 2 . For every a ∈ Ad d ( X , μ ) a \in\operatorname{Ad}_{d}(X,\mu) a ∈ Ad d ( X , μ ) ,
{ κ ∈ R ρ , 2 a , i s o ( X , μ ) : κ ∗ μ ∈ L M R i g ρ } \left\{\kappa\in\mathbb{R}_{\rho,2}^{a,\mathrm{iso}}(X,\mu) : \kappa_*\mu\in\mathrm{LMRig}_{\rho}\right\} { κ ∈ R ρ , 2 a , iso ( X , μ ) : κ ∗ μ ∈ LMRig ρ } is residual in R ρ , 2 a , i s o ( X , μ ) \mathbb{R}_{\rho,2}^{a,\mathrm{iso}}(X,\mu) R ρ , 2 a , iso ( X , μ ) , while L M R i g ρ ∩ E P P ρ , 2 \mathrm{LMRig}_{\rho} \cap\mathrm{EPP}_{\rho,2} LMRig ρ ∩ EPP ρ , 2 is residual in E P P ρ , 2 \mathrm{EPP}_{\rho,2} EPP ρ , 2 .
For m ≥ 1 m \ge1 m ≥ 1 and j ∈ N 0 j \in\mathbb{N}_{0} j ∈ N 0 , set
γ m , j ( u ) : = u j 1 B m ( u ) . \gamma_{m,j}(u) := u^{j}\mathbf{1}_{B_m}(u). γ m , j ( u ) := u j 1 B m ( u ) .
Choose a countable family G m ⊂ C c ( R d ∖ B m ) \mathcal{G}_m \subset C_c(\mathbb{R}^{d}\setminus B_m) G m ⊂ C c ( R d ∖ B m ) whose rational span uniformly approximates every f ∈ C c ( R d ∖ B m ) f \in C_c(\mathbb{R}^{d}\setminus B_m) f ∈ C c ( R d ∖ B m ) with approximating supports in a compact set depending on f f f . Define
D m , j ( η ) : = inf g ∈ span Q G m ∥ Z γ m , j η − Z g η ∥ L 2 ( η ) . D_{m,j}(\eta) := \inf_{g \in\operatorname{span}_{\mathbb{Q}}\mathcal{G}_m} \left\|Z^{\eta}_{\gamma_{m,j}} - Z^{\eta}_{g}\right\|_{L^2(\eta)}. D m , j ( η ) := g ∈ span Q G m inf Z γ m , j η − Z g η L 2 ( η ) .
Lemma 4.5. The set L M R i g ρ \mathrm{LMRig}_{\rho} LMRig ρ is G δ G_{\delta} G δ in P P ρ , 2 \mathrm{PP}_{\rho,2} PP ρ , 2 .
Proof. For fixed g g g , put h : = γ m , j − g h := \gamma_{m,j} - g h := γ m , j − g . Then
∥ Z h η ∥ 2 2 = ∫ S h 2 d η − ( ρ ∫ h ) 2 . \left\|Z^{\eta}_{h}\right\|_{2}^{2} = \int S_h^2\,d\eta- \left(\rho\int h\right)^2. ∥ Z h η ∥ 2 2 = ∫ S h 2 d η − ( ρ ∫ h ) 2 .
The function h h h is bounded and compactly supported and is continuous outside ∂ B m \partial B_m ∂ B m . By Lemma 2.3, the configuration has no point on this sphere almost surely under any measure in P P ρ , 2 \mathrm{PP}_{\rho,2} PP ρ , 2 . The local L 2 L^2 L 2 characterization in Proposition 2.7 therefore gives continuity of ∫ S h 2 d η \int S_h^2\,d\eta ∫ S h 2 d η . Consequently { η : D m , j ( η ) < ε } \{\eta: D_{m,j}(\eta) < \varepsilon\} { η : D m , j ( η ) < ε } is open: it is a union, over g g g in the countable rational span, of strict sublevel sets of continuous functions.
For bounded Borel W W W , define the control measure on Borel sets C ⊂ R d C \subset\mathbb{R}^{d} C ⊂ R d by
ν W , η ( C ) : = ∫ N W ( ω ) ω ( C ∩ W ) d η ( ω ) . \nu_{W,\eta}(C) := \int N_W(\omega)\,\omega(C \cap W)\,d\eta(\omega). ν W , η ( C ) := ∫ N W ( ω ) ω ( C ∩ W ) d η ( ω ) .
This is a finite positive measure, because ν W , η ( W ) = ∫ N W 2 d η < ∞ \nu_{W,\eta}(W) = \int N_W^2\,d\eta< \infty ν W , η ( W ) = ∫ N W 2 d η < ∞ . For bounded Borel f , h f,h f , h vanishing outside W W W , Cauchy–Schwarz in each configuration gives
∥ Z f η − Z h η ∥ 2 2 ≤ ∫ W ∣ f − h ∣ 2 d ν W , η . (41) \left\|Z_f^\eta-Z_h^\eta\right\|_2^2 \le\int_W |f-h|^2\,d\nu_{W,\eta}. \tag*{(41)} Z f η − Z h η 2 2 ≤ ∫ W ∣ f − h ∣ 2 d ν W , η . (41)
Indeed, the left-hand side is at most the second moment of S f − h S_{f-h} S f − h , while ∣ S f − h ∣ 2 ≤ N W S ∣ f − h ∣ 2 |S_{f-h}|^2 \le N_W S_{|f-h|}^2 ∣ S f − h ∣ 2 ≤ N W S ∣ f − h ∣ 2 .
Let g g g be bounded, compactly supported, and zero on B m B_m B m , and choose a bounded Borel set W W W containing its support and a neighbourhood of that support. By Lemma 2.3, ν W , η ( ∂ B m ) = 0 \nu_{W,\eta}(\partial B_m)=0 ν W , η ( ∂ B m ) = 0 . For
U ε : = { u ∈ W : dist ( u , ∂ B m ) < ε } , U_\varepsilon:= \{u \in W : \operatorname{dist}(u,\partial B_m) < \varepsilon\}, U ε := { u ∈ W : dist ( u , ∂ B m ) < ε } ,
∥ Z g 1 U ε η ∥ 2 2 ≤ ∥ g ∥ ∞ 2 ν W , η ( U ε ) ⟶ 0 ( ε ↓ 0 ) , \left\|Z^\eta_{g\mathbf{1}_{U_\varepsilon}}\right\|_2^2 \le\|g\|_\infty^2 \nu_{W,\eta}(U_\varepsilon) \longrightarrow0 \qquad(\varepsilon\downarrow0), Z g 1 U ε η 2 2 ≤ ∥ g ∥ ∞ 2 ν W , η ( U ε ) ⟶ 0 ( ε ↓ 0 ) ,
by (41) . On the remaining compact region, bounded Borel functions can be approximated in L 2 ( ν W , η ) L^2(\nu_{W,\eta}) L 2 ( ν W , η ) by continuous functions supported away from B m B_m B m . The defining approximation property of $\mathcal{G}_m then approximates these functions uniformly, and hence in the same L 2 L^2 L 2 space, by its rational span. Another application of (41) yields
span ‾ { Z g n : g ∈ G m } = H B m c n . (42) \overline{\operatorname{span}}\{Z_g^n:g\in\mathcal{G}_m\}=\mathcal{H}_{B_m^c}^n. \tag*{(42)} span { Z g n : g ∈ G m } = H B m c n . (42)
By (42) , D m , j ( η ) = 0 D_{m,j}(\eta)=0 D m , j ( η ) = 0 if and only if Z γ m , j n ∈ H B m c n Z_{\gamma_{m,j}}^n\in\mathcal{H}_{B_m^c}^n Z γ m , j n ∈ H B m c n . Hence
LMRig ρ = ⋂ m ≥ 1 ⋂ j ∈ N 0 d ⋂ k ≥ 1 { η ∈ PP ρ , 2 : D m , j ( η ) < k − 1 } . \operatorname{LMRig}_{\rho}=\bigcap_{m\geq1}\bigcap_{j\in\mathbb{N}_0^d}\bigcap_{k\geq1}\{\eta\in\operatorname{PP}_{\rho,2}:D_{m,j}(\eta)<k^{-1}\}. LMRig ρ = m ≥ 1 ⋂ j ∈ N 0 d ⋂ k ≥ 1 ⋂ { η ∈ PP ρ , 2 : D m , j ( η ) < k − 1 } .
Let η \eta η belong to the right-hand side. Then Z γ m , j n ∈ H B m c n Z_{\gamma_{m,j}}^n\in\mathcal{H}_{B_m^c}^n Z γ m , j n ∈ H B m c n for every m , j m,j m , j , hence the same holds for Z p 1 B m n Z_{p1_{B_m}}^n Z p 1 B m n for every polynomial p p p . The control measure ν B m , η \nu_{B_m,\eta} ν B m , η is finite. Polynomials are uniformly dense in C ( B m ) C(B_m) C ( B m ) by Stone–Weierstrass, and continuous functions are dense in L 2 ( ν B m , η ) L^2(\nu_{B_m,\eta}) L 2 ( ν B m , η ) . Thus every bounded Borel function f f f vanishing outside B m B_m B m admits polynomials p n p_n p n with
∫ B m ∣ f − p n ∣ 2 d ν B m , η ⟶ 0 ( n → ∞ ) . \int_{B_m}|f-p_n|^2\,d\nu_{B_m,\eta}\longrightarrow0\qquad(n\to\infty). ∫ B m ∣ f − p n ∣ 2 d ν B m , η ⟶ 0 ( n → ∞ ) .
By (41) , Z p n 1 B m n → Z f n Z_{p_n1_{B_m}}^n\to Z_f^n Z p n 1 B m n → Z f n in L 2 ( η ) L^2(\eta) L 2 ( η ) , so Z f n ∈ H B m c n Z_f^n\in\mathcal{H}_{B_m^c}^n Z f n ∈ H B m c n . If A A A is bounded and Borel, choose m m m with A ⊂ B m A\subset B_m A ⊂ B m ; then H B m c n ⊂ H A c n \mathcal{H}_{B_m^c}^n\subset\mathcal{H}_{A^c}^n H B m c n ⊂ H A c n , and η \eta η is linearly maximally rigid.
Conversely, if η ∈ LMRig ρ \eta\in\operatorname{LMRig}_{\rho} η ∈ LMRig ρ , then Z γ m , j n ∈ H B m c n Z_{\gamma_{m,j}}^n\in\mathcal{H}_{B_m^c}^n Z γ m , j n ∈ H B m c n for every m , j m,j m , j . The description of H B m c n \mathcal{H}_{B_m^c}^n H B m c n just proved gives D m , j ( η ) = 0 D_{m,j}(\eta)=0 D m , j ( η ) = 0 . This proves the reverse inclusion and the G δ G_\delta G δ assertion. □
Lemma 4.6. Let a ∈ A d ( X , μ ) a\in A_d(X,\mu) a ∈ A d ( X , μ ) , m ≥ 1 m\geq1 m ≥ 1 , j ∈ N 0 d j\in\mathbb{N}_0^d j ∈ N 0 d , and ε > 0 \varepsilon>0 ε > 0 . Then
{ κ ∈ R ρ , 2 a ( X , μ ) : D m , j ( κ ∗ μ ) < ε } \{\kappa\in\mathbb{R}_{\rho,2}^a(X,\mu):D_{m,j}(\kappa_*\mu)<\varepsilon\} { κ ∈ R ρ , 2 a ( X , μ ) : D m , j ( κ ∗ μ ) < ε } is dense in R ρ , 2 a ( X , μ ) \mathbb{R}_{\rho,2}^a(X,\mu) R ρ , 2 a ( X , μ ) .
Proof. Fix a ∈ A d ( X , μ ) a\in A_d(X,\mu) a ∈ A d ( X , μ ) , m ≥ 1 m\geq1 m ≥ 1 , j ∈ N 0 d j\in\mathbb{N}_0^d j ∈ N 0 d , and ε > 0 \varepsilon>0 ε > 0 . Let κ ∈ R ρ , 2 a ( X , μ ) \kappa\in\mathbb{R}_{\rho,2}^a(X,\mu) κ ∈ R ρ , 2 a ( X , μ ) and let U U U be a neighbourhood of κ \kappa κ . Choose ε U > 0 \varepsilon_U>0 ε U > 0 with B d r o o t , 2 ( κ , ε U ) ⊂ U B_{\mathrm{droot},2}(\kappa,\varepsilon_U)\subset U B droot , 2 ( κ , ε U ) ⊂ U . Let s : = ∣ j ∣ s:=|j| s := ∣ j ∣ and fix an integer p > d + s p>d+s p > d + s . Since 2 p > d + 1 2p>d+1 2 p > d + 1 , the cell-replacement construction of Subsection 3.2 applies with this value of p p p . In every modified L L L -cell, the identity (36) holds for all polynomials of degree less than p p p . The construction preserves intensity and satisfies N Q m ( κ ′ ) ≤ N Q m ( κ ) + D m N_{Q_m}(\kappa')\leq N_{Q_m}(\kappa)+D_m N Q m ( κ ′ ) ≤ N Q m ( κ ) + D m for each fixed m m m ; see Lemma 3.7 .
Let φ ∈ C c ∞ ( R d ) \varphi\in C_c^\infty(\mathbb{R}^d) φ ∈ C c ∞ ( R d ) satisfy φ = 1 \varphi=1 φ = 1 on B 1 B_1 B 1 and supp φ ⊂ B 2 \operatorname{supp}\varphi\subset B_2 supp φ ⊂ B 2 , and set
f R ( u ) : = u j φ ( u / R ) . f_R(u):=u^j\varphi(u/R). f R ( u ) := u j φ ( u / R ) .
Fix δ U : = δ ∗ ( κ , ε U ) \delta_U:=\delta_*(\kappa,\varepsilon_U) δ U := δ ∗ ( κ , ε U ) from Lemma 3.9 . Choose the repair cube D D D and the tolerance γ > 0 \gamma>0 γ > 0 so that ϑ : = Vol d ( D ) \vartheta:=\operatorname{Vol}_{d}(D) ϑ := Vol d ( D ) satisfies ϑ + γ < δ U / 4 \vartheta+\gamma<\delta_U/4 ϑ + γ < δ U /4 . The choice of D D D fixes the moment-completion threshold ε 0 \varepsilon_0 ε 0 in Proposition 3.5 . Choose q q q large enough to satisfy that threshold and the estimates of Lemma 3.4 with tolerance γ \gamma γ , and put L = ( q / ρ ) 1 / d L=(q/\rho)^{1/d} L = ( q / ρ ) 1/ d . Each modified L L L -cell contains exactly q = ρ L d q=\rho L^d q = ρ L d points by (36) . Its point measure therefore differs from ρ Vol d ∣ Q \rho\operatorname{Vol}_{d}|_Q ρ Vol d ∣ Q by a signed measure of total variation at most 2 ρ L d 2\rho L^d 2 ρ L d .
With L L L fixed, choose R > max { m , L } R>\max\{m,L\} R > max { m , L } so large that
C L p R d + s − p < ε 4 , CL^pR^{d+s-p}<\frac{\varepsilon}{4}, C L p R d + s − p < 4 ε ,
where C C C depends on d , ρ , p , j , φ d,\rho,p,j,\varphi d , ρ , p , j , φ but is independent of R R R and the later choices of ζ , S , r , z \zeta,S,r,z ζ , S , r , z . Such an R R R exists because p > d + s p>d+s p > d + s . The local bound on inserted points in the construction gives an integrable majorant
Y R : = C R ( 1 + N B 2 R + 1 ( κ ) 2 ) . (43) Y_R:=C_R(1+N_{B_{2R+1}}(\kappa)^2). \tag*{(43)} Y R := C R ( 1 + N B 2 R + 1 ( κ ) 2 ) . (43)
The constant C R C_R C R does not depend on ζ \zeta ζ , S S S , r r r , z z z . Since κ \kappa κ has local second moments, Y R ∈ L 1 ( μ ) Y_R \in L^1(\mu) Y R ∈ L 1 ( μ ) . Absolute continuity of its integral provides δ R > 0 \delta_R > 0 δ R > 0 such that
μ ( E ) < δ R ⟹ ∫ E Y R d μ < ε 2 4 . \mu(E) < \delta_R \quad\Longrightarrow\quad\int_E Y_R\,\mathrm{d}\mu< \frac{\varepsilon^2}{4}. μ ( E ) < δ R ⟹ ∫ E Y R d μ < 4 ε 2 .
Choose 0 < ζ < 1 / 4 0 < \zeta< 1/4 0 < ζ < 1/4 so that
ζ < δ U / 4 , C d ζ < δ R / 3. \zeta< \delta_U/4,\qquad C_d\zeta< \delta_R/3. ζ < δ U /4 , C d ζ < δ R /3.
By Lemma 3.4 , b S , ζ → 0 b_{S,\zeta} \to0 b S , ζ → 0 as S → ∞ S \to\infty S → ∞ for this fixed ζ \zeta ζ , independently of r r r . Choose integers ℓ > ℓ ′ \ell> \ell' ℓ > ℓ ′ with ℓ − ℓ ′ \ell-\ell' ℓ − ℓ ′ even, and put S : = ℓ L S := \ell L S := ℓ L and S ′ : = ℓ ′ L S' := \ell' L S ′ := ℓ ′ L . Choose ℓ , ℓ ′ \ell,\ell' ℓ , ℓ ′ sufficiently large that the shell lattice satisfies the capacity bound # G Q ≥ 5 ρ V o u t / 4 \#G_Q \ge5\rho V_{\mathrm{out}}/4 # G Q ≥ 5 ρ V out /4 for every S S S -cube, and that
ζ ≤ 1 − ( S ′ / S ) d ≤ 2 ζ , S > 4 R + 2 d L , \zeta\le1-(S'/S)^d \le2\zeta,\qquad S > 4R+2\sqrt{d}\,L, ζ ≤ 1 − ( S ′ / S ) d ≤ 2 ζ , S > 4 R + 2 d L ,
and
4 b S , ζ + C d 2 R + d L S < δ R / 3. 4b_{S,\zeta}+C_d\frac{2R+\sqrt{d}\,L}{S}<\delta_R/3. 4 b S , ζ + C d S 2 R + d L < δ R /3.
Finally choose r ≥ S 2 r \ge S^2 r ≥ S 2 so that
S / r < δ U / 4 , C d S / r < δ R / 3. S/r < \delta_U/4,\qquad C_dS/r < \delta_R/3. S / r < δ U /4 , C d S / r < δ R /3.
Apply the replacement rule in Subsection 3.2 with the chosen parameters, obtaining a realization κ ′ \kappa' κ ′ and a shift z ∈ [ 0 , S ) d z \in[0,S)^d z ∈ [ 0 , S ) d . Set
E e x c : = X ∖ { t . c : c ∈ C r , Q ∈ Q z ( c ) , Q count-compatible , t ∈ Q i n , dist ( t , ∂ Q i n ) > 2 R + d L } . E_{\mathrm{exc}}:=X\setminus\left\{t.c:\begin{array}{l}
c\in C_r,\ Q\in\mathcal{Q}_z(c),\ Q\ \text{count-compatible},\\
t\in Q^{\mathrm{in}},\ \operatorname{dist}(t,\partial Q^{\mathrm{in}})>2R+\sqrt{d}\,L
\end{array}\right\}. E exc := X ∖ { t . c : c ∈ C r , Q ∈ Q z ( c ) , Q count-compatible , t ∈ Q in , dist ( t , ∂ Q in ) > 2 R + d L } .
The Voronoi covering estimate (33) and the bound e 1 ( z ) ≤ 4 b S , ζ e_1(z)\le4b_{S,\zeta} e 1 ( z ) ≤ 4 b S , ζ in Lemma 3.8 give
μ ( E e x c ) ≤ 4 b S , ζ + C d ζ + C d 2 R + d L S + C d S r < δ R . \mu(E_{\mathrm{exc}})\le4b_{S,\zeta}+C_d\zeta+C_d\frac{2R+\sqrt{d}\,L}{S}+C_d\frac{S}{r}<\delta_R. μ ( E exc ) ≤ 4 b S , ζ + C d ζ + C d S 2 R + d L + C d r S < δ R .
Also ϑ + γ + ζ + S / r < δ U \vartheta+\gamma+\zeta+S/r<\delta_U ϑ + γ + ζ + S / r < δ U , so d r o o t , 2 ( κ ′ , κ ) < ε U d_{\mathrm{root},2}(\kappa',\kappa)<\varepsilon_U d root , 2 ( κ ′ , κ ) < ε U . Thus κ ′ ∈ U \kappa'\in U κ ′ ∈ U , and the bound by Y R Y_R Y R gives
∫ E e x c ∣ Z f R ( κ ′ ) ( x ) μ ( κ ′ ( x ) ) ∣ 2 d μ ( x ) < ε 2 4 . \int_{E_{\mathrm{exc}}}\left|Z_{f_R}^{(\kappa')}(x)\mu(\kappa'(x))\right|^2\,\mathrm{d}\mu(x)<\frac{\varepsilon^2}{4}. ∫ E exc Z f R ( κ ′ ) ( x ) μ ( κ ′ ( x )) 2 d μ ( x ) < 4 ε 2 .
For x ∉ E e x c x\notin E_{\mathrm{exc}} x ∈ / E exc , the support of f R f_R f R lies in the union of modified L L L -cells in a single core: the extra margin d L \sqrt{d}\,L d L ensures that every cell intersecting B 2 R B_{2R} B 2 R belongs to that core. For each of these cells Q Q Q , define
Δ Q : = κ ′ ( x ) ∣ Q − ρ Vol d ∣ Q . \Delta_Q:=\kappa'(x)|_Q-\rho\,\operatorname{Vol}_d|_Q. Δ Q := κ ′ ( x ) ∣ Q − ρ Vol d ∣ Q .
By (36) ,
∫ P d Δ Q = 0 ( deg P < p ) . \int P\,\mathrm{d}\Delta_Q=0\qquad(\deg P<p). ∫ P d Δ Q = 0 ( deg P < p ) .
Also ∥ Δ Q ∥ T V ≤ 2 ρ L d \|\Delta_Q\|_{\mathrm{TV}}\le2\rho L^d ∥ Δ Q ∥ TV ≤ 2 ρ L d , since the cell contains exactly ρ L d \rho L^d ρ L d points. Let c Q c_Q c Q be the centre of Q Q Q . The Taylor polynomial T c Q p − 1 f R T_{c_Q}^{p-1}f_R T c Q p − 1 f R has degree less than p p p and integrates to zero against Δ Q \Delta_Q Δ Q . Since every derivative of f R f_R f R of order p p p is bounded by C R s − p CRs^{-p} C R s − p , Taylor’s theorem and the preceding total variation bound give
∣ ∫ f R d Δ Q ∣ ≤ ∥ Δ Q ∥ T V sup Q ∣ f R − T c Q p − 1 f R ∣ ≤ C L d + p R s − p . \left|\int f_R\,\mathrm{d}\Delta_Q\right|
\le\|\Delta_Q\|_{\mathrm{TV}}\sup_Q\left|f_R-T_{c_Q}^{p-1}f_R\right|
\le CL^{d+p}R s^{-p}. ∫ f R d Δ Q ≤ ∥ Δ Q ∥ TV Q sup f R − T c Q p − 1 f R ≤ C L d + p R s − p .
Since R > L R>L R > L , at most C d ( R / L ) d C_d(R/L)^d C d ( R / L ) d cells intersect B 2 R B_{2R} B 2 R . Each contributes at most C L d + p R s − p CL^{d+p}Rs^{-p} C L d + p R s − p ; the other cells contribute zero because f R f_R f R vanishes there. Therefore
∣ ∫ f R d ( κ ′ ( x ) − ρ Vol d ) ∣ ≤ C L p R d + s − p < ε 4 . \left|\int f_R\,\mathrm{d}\bigl(\kappa'(x)-\rho\operatorname{Vol}_d\bigr)\right|
\le CL^pR^{d+s-p}<\frac{\varepsilon}{4}. ∫ f R d ( κ ′ ( x ) − ρ Vol d ) ≤ C L p R d + s − p < 4 ε .
Thus
∫ X ∖ E e x c ∣ Z f R ( κ ′ ) ∗ μ ( κ ′ ( x ) ) ∣ 2 d μ ( x ) ≤ ε 2 16 . \int_{X\setminus E_{\mathrm{exc}}} \left|Z_{f_R}^{(\kappa')*\mu}(\kappa'(x))\right|^2\,d\mu(x) \le\frac{\varepsilon^2}{16}. ∫ X ∖ E exc Z f R ( κ ′ ) ∗ μ ( κ ′ ( x )) 2 d μ ( x ) ≤ 16 ε 2 .
Combining the two parts gives
∥ Z f R ( κ ′ ) ∗ μ ∥ L 2 ( ( κ ′ ) ∗ μ ) < ε . \left\|Z_{f_R}^{(\kappa')*\mu}\right\|_{L^2((\kappa')*\mu)} < \varepsilon. Z f R ( κ ′ ) ∗ μ L 2 (( κ ′ ) ∗ μ ) < ε .
Since R > m R>m R > m and φ = 1 \varphi=1 φ = 1 on B 1 B_1 B 1 , we have f R ( u ) = u j f_R(u)=u^j f R ( u ) = u j on B m B_m B m . The function
g R : = f R − γ m , j g_R:=f_R-\gamma_{m,j} g R := f R − γ m , j
is bounded, compactly supported, and vanishes on B m B_m B m . Hence Z − g R ( κ ′ ) ∗ μ ∈ H B m ( κ ′ ) ∗ μ Z_{-g_R}^{(\kappa')*\mu}\in H_{B_m}^{(\kappa')*\mu} Z − g R ( κ ′ ) ∗ μ ∈ H B m ( κ ′ ) ∗ μ . Since D m , j D_{m,j} D m , j is the distance to this closed subspace,
D m , j ( ( κ ′ ) ∗ μ ) ≤ ∥ Z γ m , j ( κ ′ ) ∗ μ − Z − g R ( κ ′ ) ∗ μ ∥ 2 = ∥ Z f R ( κ ′ ) ∗ μ ∥ 2 < ε . D_{m,j}((\kappa')*\mu)\le\left\|Z_{\gamma_{m,j}}^{(\kappa')*\mu}-Z_{-g_R}^{(\kappa')*\mu}\right\|_2=\left\|Z_{f_R}^{(\kappa')*\mu}\right\|_2<\varepsilon. D m , j (( κ ′ ) ∗ μ ) ≤ Z γ m , j ( κ ′ ) ∗ μ − Z − g R ( κ ′ ) ∗ μ 2 = Z f R ( κ ′ ) ∗ μ 2 < ε .
Since U \mathcal{U} U and ε \varepsilon ε were arbitrary, for every k ≥ 1 k\ge1 k ≥ 1 the preimage of { η : D m , j ( η ) < k − 1 } \{\eta:D_{m,j}(\eta)<k^{-1}\} { η : D m , j ( η ) < k − 1 } in R ρ , 2 a ( X , μ ) \mathcal{R}^{a}_{\rho,2}(X,\mu) R ρ , 2 a ( X , μ ) is dense. It is open by the G δ G_\delta G δ argument above and continuity of the push-forward map.
Proof of Proposition 4.4 . By Lemma 4.5 , the preimage of L M R i g ρ \mathrm{LMRig}_{\rho} LMRig ρ is a G δ G_\delta G δ in each rooted realization space. By Lemma 4.6 , its defining open sets are dense. Intersecting with the dense G δ G_\delta G δ of generating realizations from Proposition 3.12 proves the fixed-action assertion. The global conclusion follows from Proposition 2.11 .
Proof of Theorem B, Corollary 1.2, and Proposition 1.3 . For each a ∈ Ad ( X , μ ) a\in\operatorname{Ad}(X,\mu) a ∈ Ad ( X , μ ) , the intermittent-hyperuniform and linearly maximally rigid realizations are residual in R ρ , 2 a , i s o ( X , μ ) \mathcal{R}^{a,\mathrm{iso}}_{\rho,2}(X,\mu) R ρ , 2 a , iso ( X , μ ) by the end of § 3 and Proposition 4.4 ; their intersection gives Proposition 1.3 . The classes I H y p ρ \mathrm{IHyp}_{\rho} IHyp ρ and L M R i g ρ \mathrm{LMRig}_{\rho} LMRig ρ , intersected with E P P ρ , 2 \mathrm{EPP}_{\rho,2} EPP ρ , 2 , are residual in E P P ρ , 2 \mathrm{EPP}_{\rho,2} EPP ρ , 2 , which proves Theorem B and the first assertion of Corollary 1.2. Finally, H y p ρ ∩ E P P ρ , 2 \mathrm{Hyp}_{\rho}\cap\mathrm{EPP}_{\rho,2} Hyp ρ ∩ EPP ρ , 2 is meagre by the end of § 3 ; intersecting its residual complement with M R i g ρ ∩ E P P ρ , 2 \mathrm{MRig}_{\rho}\cap\mathrm{EPP}_{\rho,2} MRig ρ ∩ EPP ρ , 2 gives the last assertion.
Periodic approximation and pairwise negative correlation
The approximation of measure-preserving transformations by periodic ones, and its use in spaces of actions, are developed in the Rokhlin–Halmos tradition; see [25 Chaps. 1–2] . We adapt the block-approximation method to point-process measures, controlling both intensity and local second moments. Periodic approximation proves density of hyperuniform processes and, together with the covariance representation below, excludes pairwise negative correlation on a dense set.
Periodic approximation
For a full-rank lattice L < R d L<\mathbb{R}^d L < R d , write
L ∗ : = { ξ ∈ R d : ⟨ ξ , t ⟩ ∈ Z for every t ∈ L } , λ 1 ( L ) : = min t ∈ L ∖ { 0 } ∥ t ∥ . L^*:=\{\xi\in\mathbb{R}^d:\langle\xi,t\rangle\in\mathbb{Z}\text{ for every }t\in L\},\qquad\lambda_1(L):=\min_{t\in L\setminus\{0\}}\|t\|. L ∗ := { ξ ∈ R d : ⟨ ξ , t ⟩ ∈ Z for every t ∈ L } , λ 1 ( L ) := t ∈ L ∖ { 0 } min ∥ t ∥.
A nonempty locally finite set P ⊂ R d P\subset\mathbb{R}^d P ⊂ R d is periodic if its stabilizer
Stab ( P ) : = { t ∈ R d : P + t = P } \operatorname{Stab}(P):=\{t\in\mathbb{R}^d:P+t=P\} Stab ( P ) := { t ∈ R d : P + t = P }
contains a full-rank lattice. Its stabilizer is then itself a full-rank lattice, called the period lattice of P P P . If L ⊆ Stab ( P ) L\subseteq\operatorname{Stab}(P) L ⊆ Stab ( P ) is a full-rank lattice and D L D_L D L a half-open fundamental domain, define the density of P P P by
dens ( P ) : = # ( P ∩ D L ) covol ( L ) . \operatorname{dens}(P):=\frac{\#(P\cap D_L)}{\operatorname{covol}(L)}. dens ( P ) := covol ( L ) # ( P ∩ D L ) .
This is independent of the lattice and fundamental domain. Associate to P P P the probability measure
η P : = 1 covol ( L ) ∫ D L δ θ t δ P d t . \eta_P := \frac{1}{\operatorname{covol}(\mathbf{L})}\int_{D_{\mathbf{L}}}\delta_{\theta_t\delta_P}\,dt. η P := covol ( L ) 1 ∫ D L δ θ t δ P d t .
The measure η P \eta_P η P is the unique translation-invariant probability measure supported on the translation orbit of δ P \delta_P δ P , has intensity dens ( P ) \operatorname{dens}(P) dens ( P ) , and is independent of the chosen lattice L ⊆ Stab ( P ) \mathbf{L} \subseteq\operatorname{Stab}(P) L ⊆ Stab ( P ) .
A function h : Conf ( R d ) → R h:\operatorname{Conf}(\mathbb{R}^d)\to\mathbb{R} h : Conf ( R d ) → R is local if there is a bounded Borel set W ⊂ R d W\subset\mathbb{R}^d W ⊂ R d such that h ( ω ) = h ( ω ′ ) h(\omega)=h(\omega') h ( ω ) = h ( ω ′ ) whenever ω ∣ W = ω ′ ∣ W \omega|_W=\omega'|_W ω ∣ W = ω ′ ∣ W . Such a set W W W is a dependence window for h h h .
Lemma 5.1 (Periodic approximation). Let L 0 < R d \mathbf{L}_0<\mathbb{R}^d L 0 < R d be a full-rank lattice , let s n → ∞ s_n\to\infty s n → ∞ , and set L n : = s n L 0 \mathbf{L}_n:=s_n\mathbf{L}_0 L n := s n L 0 . Suppose N n ∈ N N_n\in\mathbb{N} N n ∈ N and
ρ n : = N n covol ( L n ) ≤ ρ , ρ n ⟶ ρ ( n → ∞ ) . \rho_n:=\frac{N_n}{\operatorname{covol}(\mathbf{L}_n)}\leq\rho,\qquad\rho_n\longrightarrow\rho\qquad(n\to\infty). ρ n := covol ( L n ) N n ≤ ρ , ρ n ⟶ ρ ( n → ∞ ) . Let η ∈ P P ρ , 2 \eta\in\mathsf{PP}_{\rho,2} η ∈ PP ρ , 2 , let h 1 , … , h k h_1,\ldots,h_k h 1 , … , h k be bounded continuous local functions, let Q 1 ′ , … , Q ℓ ′ Q'_1,\ldots,Q'_\ell Q 1 ′ , … , Q ℓ ′ be bounded half-open cubes, and let ε > 0 \varepsilon>0 ε > 0 . There is n 0 n_0 n 0 such that, for every n ≥ n 0 n\geq n_0 n ≥ n 0 , there is a simple L n \mathbf{L}_n L n -periodic set P n P_n P n with exactly N n N_n N n points in a fundamental domain and
∣ ∫ h i d η P n − ∫ h i d η ∣ < ε ( 1 ≤ i ≤ k ) , \left|\int h_i\,d\eta_{P_n}-\int h_i\,d\eta\right|<\varepsilon\qquad(1\leq i\leq k), ∫ h i d η P n − ∫ h i d η < ε ( 1 ≤ i ≤ k ) , ∣ ∫ N Q j ′ 2 d η P n − ∫ N Q j ′ 2 d η ∣ < ε ( 1 ≤ j ≤ ℓ ) . \left|\int N_{Q'_j}^{2}\,d\eta_{P_n}-\int N_{Q'_j}^{2}\,d\eta\right|<\varepsilon\qquad(1\leq j\leq\ell). ∫ N Q j ′ 2 d η P n − ∫ N Q j ′ 2 d η < ε ( 1 ≤ j ≤ ℓ ) . The measure η P n \eta_{P_n} η P n has intensity ρ n \rho_n ρ n .
Proof. Write V n : = covol ( L n ) V_n:=\operatorname{covol}(\mathbf{L}_n) V n := covol ( L n ) and choose a half-open fundamental parallelepiped D D D of L 0 \mathbf{L}_0 L 0 , so s n D s_nD s n D is one for L n \mathbf{L}_n L n . Choose r > 0 r>0 r > 0 such that all dependence windows of the functions h i h_i h i and all the cubes Q j ′ Q'_j Q j ′ are contained in B r B_r B r . Fix L > 10 r L>10r L > 10 r . For an L L L -cube Q = z + [ 0 , L ) d Q=z+[0,L)^d Q = z + [ 0 , L ) d , define
Q r e t : = z + [ 2 r , L − 2 r ) d , Q s a f e : = z + [ 3 r , L − 3 r ) d . Q^{\mathrm{ret}}:=z+[2r,L-2r)^d,\qquad Q^{\mathrm{safe}}:=z+[3r,L-3r)^d. Q ret := z + [ 2 r , L − 2 r ) d , Q safe := z + [ 3 r , L − 3 r ) d .
Their relative volumes are
b L : = ( 1 − 4 r L ) d , a L : = ( 1 − 6 r L ) d . b_L:=\left(1-\frac{4r}{L}\right)^d,\qquad a_L:=\left(1-\frac{6r}{L}\right)^d. b L := ( 1 − L 4 r ) d , a L := ( 1 − L 6 r ) d .
For t ∈ Q s a f e t\in Q^{\mathrm{safe}} t ∈ Q safe we have t + B r ⊂ Q r e t t+B_r\subset Q^{\mathrm{ret}} t + B r ⊂ Q ret . For large n n n , enumerate the cubes L k + [ 0 , L ) d Lk+[0,L)^d L k + [ 0 , L ) d , k ∈ Z d k\in\mathbb{Z}^d k ∈ Z d , contained in s n D s_nD s n D as Q 1 , … , Q J Q_1,\ldots,Q_J Q 1 , … , Q J , with J = J ( n , L ) J=J(n,L) J = J ( n , L ) . Their union leaves uncovered volume O ( L / s n ) V n O(L/s_n)V_n O ( L / s n ) V n , so
J L d V n = 1 + O ( L / s n ) . \frac{JL^d}{V_n}=1+O(L/s_n). V n J L d = 1 + O ( L / s n ) .
On Conf ( R d ) J \operatorname{Conf}(\mathbb{R}^d)^J Conf ( R d ) J , put m n , L : = η ⊗ J m_{n,L}:=\eta^{\otimes J} m n , L := η ⊗ J and denote the coordinate configurations by ω 1 , … , ω J \omega_1,\ldots,\omega_J ω 1 , … , ω J . Define P n 0 P_n^0 P n 0 on the fundamental domain by
P n 0 ∩ s n D : = ⋃ a = 1 J ( supp ω a ∩ Q a r e t ) P_n^0\cap s_nD:=\bigcup_{a=1}^{J}(\operatorname{supp}\omega_a\cap Q_a^{\mathrm{ret}}) P n 0 ∩ s n D := a = 1 ⋃ J ( supp ω a ∩ Q a ret )
and extend it L n \mathbf{L}_n L n -periodically. Put U n : = ⋃ a = 1 J Q a s a f e U_n:=\bigcup_{a=1}^{J}Q_a^{\mathrm{safe}} U n := ⋃ a = 1 J Q a safe . Distinct retained cubes, including their periodic translates, are separated by at least 4 r 4r 4 r . For t ∈ Q a s a f e t\in Q_a^{\mathrm{safe}} t ∈ Q a safe , each test depends only on the points in t + B r ⊂ Q a r e t t+B_r\subset Q_a^{\mathrm{ret}} t + B r ⊂ Q a ret . Thus
h i ( θ t δ P n 0 ) = h i ( θ t ω a ) , N Q j ′ ( θ t δ P n 0 ) 2 = N Q j ′ ( θ t ω a ) 2 . h_i(\theta_t\delta_{P_n^0})=h_i(\theta_t\omega_a),\qquad N_{Q'_j}(\theta_t\delta_{P_n^0})^2=N_{Q'_j}(\theta_t\omega_a)^2. h i ( θ t δ P n 0 ) = h i ( θ t ω a ) , N Q j ′ ( θ t δ P n 0 ) 2 = N Q j ′ ( θ t ω a ) 2 .
For 1 ≤ a ≤ J 1\leq a\leq J 1 ≤ a ≤ J , define
X a , i = L − d ∫ Q a s a f e h i ( θ t ω a ) d t , Y a , j = L − d ∫ Q a s a f e N Q j ′ ( θ t ω a ) 2 d t , Z a = N Q a r e t ( ω a ) . X_{a,i}=L^{-d}\int_{Q_a^{\mathrm{safe}}}h_i(\theta_t\omega_a)\,dt,\qquad Y_{a,j}=L^{-d}\int_{Q_a^{\mathrm{safe}}}N_{Q'_j}(\theta_t\omega_a)^2\,dt,\qquad Z_a=N_{Q_a^{\mathrm{ret}}}(\omega_a). X a , i = L − d ∫ Q a safe h i ( θ t ω a ) d t , Y a , j = L − d ∫ Q a safe N Q j ′ ( θ t ω a ) 2 d t , Z a = N Q a ret ( ω a ) .
The definitions and the local identities above give
1 V n ∫ U n h i ( θ t δ P n 0 ) d t = L d V n ∑ a = 1 J X a , i , \frac{1}{V_n}\int_{U_n} h_i(\theta_t\delta_{P_n^0})\,dt=\frac{L^d}{V_n}\sum_{a=1}^{J}X_{a,i}, V n 1 ∫ U n h i ( θ t δ P n 0 ) d t = V n L d a = 1 ∑ J X a , i ,
1 V n ∫ U n N Q j ′ ( θ t δ P n 0 ) 2 d t = L d V n ∑ a = 1 J Y a , j . \frac{1}{V_n}\int_{U_n}N_{Q'_j}(\theta_t\delta_{P_n^0})^2\,dt=\frac{L^d}{V_n}\sum_{a=1}^{J}Y_{a,j}. V n 1 ∫ U n N Q j ′ ( θ t δ P n 0 ) 2 d t = V n L d a = 1 ∑ J Y a , j .
For fixed L L L , the vectors ( X a , 1 , … , X a , k , Y a , 1 , … , Y a , ℓ , Z a ) (X_{a,1},\ldots,X_{a,k},Y_{a,1},\ldots,Y_{a,\ell},Z_a) ( X a , 1 , … , X a , k , Y a , 1 , … , Y a , ℓ , Z a ) are independent and identically distributed over a a a . Translation invariance gives
∫ X a , i d m n , L = a L ∫ h i d η , ∫ Y a , j d m n , L = a L ∫ N Q j ′ 2 d η , ∫ Z a d m n , L = ρ b L L d . \int X_{a,i}\,d\mathfrak{m}_{n,L}=a_L\int h_i\,d\eta,\qquad\int Y_{a,j}\,d\mathfrak{m}_{n,L}=a_L\int N_{Q'_j}^2\,d\eta,\qquad\int Z_a\,d\mathfrak{m}_{n,L}=\rho b_L L^d. ∫ X a , i d m n , L = a L ∫ h i d η , ∫ Y a , j d m n , L = a L ∫ N Q j ′ 2 d η , ∫ Z a d m n , L = ρ b L L d .
Define K n : = # ( P n 0 ∩ s n D ) = ∑ a = 1 J Z a K_n:=\#(P_n^0\cap s_nD)=\sum_{a=1}^{J}Z_a K n := # ( P n 0 ∩ s n D ) = ∑ a = 1 J Z a . The variables X a , i X_{a,i} X a , i are bounded, while Y a , j Y_{a,j} Y a , j and Z a Z_a Z a are integrable by the local second-moment assumption. For fixed L L L , we have J ( n , L ) → ∞ J(n,L)\to\infty J ( n , L ) → ∞ , so the weak law of large numbers applies to their sample averages. By the weak law, choose δ n ↓ 0 \delta_n\downarrow0 δ n ↓ 0 so that the event A n , L \mathcal{A}_{n,L} A n , L on which the sample means of X a , i X_{a,i} X a , i , Y a , j Y_{a,j} Y a , j and Z a Z_a Z a differ from their respective expectations by at most δ n \delta_n δ n satisfies m n , L ( A n , L ) → 1 \mathfrak{m}_{n,L}(\mathcal{A}_{n,L})\to1 m n , L ( A n , L ) → 1 . On A n , L \mathcal{A}_{n,L} A n , L ,
K n V n = ρ b L J L d V n + o ( 1 ) = ρ b L + O ( L / s n ) + o ( 1 ) , \frac{K_n}{V_n}=\rho b_L\frac{JL^d}{V_n}+o(1)=\rho b_L+O(L/s_n)+o(1), V n K n = ρ b L V n J L d + o ( 1 ) = ρ b L + O ( L / s n ) + o ( 1 ) ,
where the o ( 1 ) o(1) o ( 1 ) term is bounded by a constant multiple of δ n \delta_n δ n for fixed L L L .
The relative volume of s n D ∖ U n s_nD\setminus U_n s n D ∖ U n is O ( r / L ) + O ( L / s n ) O(r/L)+O(L/s_n) O ( r / L ) + O ( L / s n ) , so its contribution to each bounded h i h_i h i is bounded deterministically by this fraction times ∥ h i ∥ ∞ \|h_i\|_\infty ∥ h i ∥ ∞ . For the integrals of N Q j ′ 2 N_{Q'_j}^2 N Q j ′ 2 , the separation of the cubes Q a r e t Q_a^{\mathrm{ret}} Q a ret implies that t + Q j ′ t+Q'_j t + Q j ′ intersects at most one of them. For each a a a , the set of origins t t t outside the safe cubes for which t + Q j ′ t+Q'_j t + Q j ′ intersects Q a r e t Q_a^{\mathrm{ret}} Q a ret has volume O ( r L d − 1 ) O(rL^{d-1}) O ( r L d − 1 ) . On that set the square count of P n 0 P_n^0 P n 0 is bounded by N Q j ′ ( θ t ω a ) 2 N_{Q'_j}(\theta_t\omega_a)^2 N Q j ′ ( θ t ω a ) 2 . Translation invariance and Fubini give
∫ 1 V n ∫ s n D ∖ U n N Q j ′ ( θ t δ P n 0 ) 2 d t d m n , L ≤ C j r L ∫ N Q j ′ 2 d η . \int\frac{1}{V_n}\int_{s_nD\setminus U_n}N_{Q'_j}(\theta_t\delta_{P_n^0})^2\,dt\,d\mathfrak{m}_{n,L}\le C_j\frac{r}{L}\int N_{Q'_j}^2\,d\eta. ∫ V n 1 ∫ s n D ∖ U n N Q j ′ ( θ t δ P n 0 ) 2 d t d m n , L ≤ C j L r ∫ N Q j ′ 2 d η .
The integrals over U n U_n U n differ from the corresponding integrals against η \eta η by O ( 1 − a L ) + O ( L / s n ) + o ( 1 ) O(1-a_L)+O(L/s_n)+o(1) O ( 1 − a L ) + O ( L / s n ) + o ( 1 ) on A n , L \mathcal{A}_{n,L} A n , L . Choose L L L so that the deterministic losses and the sum of the Markov bounds for the complement integrals are small. For sufficiently large n n n , m n , L ( A n , L ) > 1 / 2 \mathfrak{m}_{n,L}(\mathcal{A}_{n,L})>1/2 m n , L ( A n , L ) > 1/2 and the union of the excluded Markov events has measure less than 1 / 2 1/2 1/2 . The intersection therefore contains a sample satisfying the test inequalities. For fixed L L L , the difference ρ − ρ b L = ρ ( 1 − b L ) \rho-\rho b_L=\rho(1-b_L) ρ − ρ b L = ρ ( 1 − b L ) is positive. Since ρ n → ρ \rho_n\to\rho ρ n → ρ , the bound on K n / V n K_n/V_n K n / V n gives K n / V n < ρ n = N n / V n K_n/V_n<\rho_n=N_n/V_n K n / V n < ρ n = N n / V n for sufficiently large n n n on A n , L \mathcal{A}_{n,L} A n , L . For this sample, put M n : = N n − K n M_n:=N_n-K_n M n := N n − K n . Then
0 ≤ M n V n = ρ ( 1 − b L ) + O ( L / s n ) + o ( 1 ) ≤ 2 ρ ( 1 − b L ) . 0\le\frac{M_n}{V_n}=\rho(1-b_L)+O(L/s_n)+o(1)\le2\rho(1-b_L). 0 ≤ V n M n = ρ ( 1 − b L ) + O ( L / s n ) + o ( 1 ) ≤ 2 ρ ( 1 − b L ) .
If M n > 0 M_n>0 M n > 0 , let k n k_n k n be the least integer with k n d ≥ M n k_n^d\ge M_n k n d ≥ M n . The grid k n − 1 L n / L n k_n^{-1}L_n/L_n k n − 1 L n / L n has k n d k_n^d k n d points. Since k n ≤ 2 M n 1 / d k_n\le2M_n^{1/d} k n ≤ 2 M n 1/ d , its separation is at least
c L 0 ( V n M n ) 1 / d , c L 0 : = λ 1 ( L 0 ) 2 covol ( L 0 ) 1 / d . c_{L_0}\left(\frac{V_n}{M_n}\right)^{1/d},\qquad c_{L_0}:=\frac{\lambda_1(L_0)}{2\operatorname{covol}(L_0)^{1/d}}. c L 0 ( M n V n ) 1/ d , c L 0 := 2 covol ( L 0 ) 1/ d λ 1 ( L 0 ) .
The estimates above permit L L L to be fixed sufficiently large, and then n n n sufficiently large, that M n / V n ≤ 2 ρ ( 1 − b L ) M_n/V_n\le2\rho(1-b_L) M n / V n ≤ 2 ρ ( 1 − b L ) and c L 0 ( 2 ρ ( 1 − b L ) ) − 1 / d > 2 r c_{L_0}(2\rho(1-b_L))^{-1/d}>2r c L 0 ( 2 ρ ( 1 − b L ) ) − 1/ d > 2 r . Choose a translate of the grid disjoint from P n 0 P_n^0 P n 0 modulo L n L_n L n , and let G n G_n G n consist of M n M_n M n of its points, extended L n L_n L n -periodically. Set P n : = P n 0 ∪ G n P_n:=P_n^0\cup G_n P n := P n 0 ∪ G n . This set is simple and L n L_n L n -periodic, and # ( P n ∩ s n D ) = K n + M n = N n \#(P_n\cap s_nD)=K_n+M_n=N_n # ( P n ∩ s n D ) = K n + M n = N n . For a bounded Borel window Q Q Q , put I Q ( t ) : = # ( G n ∩ ( t + Q ) ) I_Q(t) := \#(G_n \cap(t + Q)) I Q ( t ) := # ( G n ∩ ( t + Q )) . For each fixed Q Q Q , separation gives sup t I Q ( t ) ≤ C Q \sup_t I_Q(t) \le C_Q sup t I Q ( t ) ≤ C Q , with C Q C_Q C Q independent of sufficiently large L , n L,n L , n . Since I Q ( t ) 2 ≤ C Q I Q ( t ) I_Q(t)^2 \le C_Q I_Q(t) I Q ( t ) 2 ≤ C Q I Q ( t ) , integration over the period cell gives
1 V n ∫ s n D I Q ( t ) 2 d t ≤ C Q ∣ Q ∣ M n V n . (44) \frac{1}{V_n}\int_{s_nD} I_Q(t)^2\,dt \le C_Q|Q|\frac{M_n}{V_n}. \tag*{(44)} V n 1 ∫ s n D I Q ( t ) 2 d t ≤ C Q ∣ Q ∣ V n M n . (44)
Let W i W_i W i be a bounded dependence window for h i h_i h i . If ( t + W i ) (t+W_i) ( t + W i ) contains no inserted point, then h i ( θ t δ P n ) = h i ( θ t δ P n 0 ) h_i(\theta_t\delta_{P_n})=h_i(\theta_t\delta_{P_n^0}) h i ( θ t δ P n ) = h i ( θ t δ P n 0 ) . Consequently,
∣ ∫ h i d η P n − ∫ h i d η P n 0 ∣ ≤ 2 ∥ h i ∥ ∞ Vol d ( W i ) M n V n . \left|\int h_i\,d\eta_{P_n}-\int h_i\,d\eta_{P_n^0}\right|\le2\|h_i\|_\infty\operatorname{Vol}_d(W_i)\frac{M_n}{V_n}. ∫ h i d η P n − ∫ h i d η P n 0 ≤ 2∥ h i ∥ ∞ Vol d ( W i ) V n M n .
For the functions N Q j ′ 2 N_{Q'_j}^2 N Q j ′ 2 , for every t t t ,
N Q j ′ ( θ t δ P n ) = N Q j ′ ( θ t δ P n 0 ) + I Q j ′ ( t ) . N_{Q'_j}(\theta_t\delta_{P_n})=N_{Q'_j}(\theta_t\delta_{P_n^0})+I_{Q'_j}(t). N Q j ′ ( θ t δ P n ) = N Q j ′ ( θ t δ P n 0 ) + I Q j ′ ( t ) .
Put C ( t ) : = N Q j ′ ( θ t δ P n 0 ) C(t):=N_{Q'_j}(\theta_t\delta_{P_n^0}) C ( t ) := N Q j ′ ( θ t δ P n 0 ) and I ( t ) : = I Q j ′ ( t ) I(t):=I_{Q'_j}(t) I ( t ) := I Q j ′ ( t ) . Cauchy–Schwarz gives
1 V n ∫ s n D ( ( C + I ) 2 − C 2 ) d t ≤ 2 ( 1 V n ∫ s n D C 2 d t ) 1 / 2 ( 1 V n ∫ s n D I 2 d t ) 1 / 2 + 1 V n ∫ s n D I 2 d t . \frac{1}{V_n}\int_{s_nD}\bigl((C+I)^2-C^2\bigr)\,dt
\le2\left(\frac{1}{V_n}\int_{s_nD}C^2\,dt\right)^{1/2}
\left(\frac{1}{V_n}\int_{s_nD}I^2\,dt\right)^{1/2}
+\frac{1}{V_n}\int_{s_nD}I^2\,dt. V n 1 ∫ s n D ( ( C + I ) 2 − C 2 ) d t ≤ 2 ( V n 1 ∫ s n D C 2 d t ) 1/2 ( V n 1 ∫ s n D I 2 d t ) 1/2 + V n 1 ∫ s n D I 2 d t .
The choice of P n 0 P_n^0 P n 0 and (44) give
1 V n ∫ s n D C ( t ) 2 d t ≤ ∫ N Q j ′ 2 d η + ε , \frac{1}{V_n}\int_{s_nD}C(t)^2\,dt\le\int N_{Q'_j}^2\,d\eta+\varepsilon, V n 1 ∫ s n D C ( t ) 2 d t ≤ ∫ N Q j ′ 2 d η + ε ,
1 V n ∫ s n D I ( t ) 2 d t = O ( M n / V n ) . \frac{1}{V_n}\int_{s_nD}I(t)^2\,dt=O(M_n/V_n). V n 1 ∫ s n D I ( t ) 2 d t = O ( M n / V n ) .
Since M n / V n = O ( r / L ) + O ( L / s n ) + o ( 1 ) M_n/V_n=O(r/L)+O(L/s_n)+o(1) M n / V n = O ( r / L ) + O ( L / s n ) + o ( 1 ) , the difference of the N Q j ′ 2 N_{Q'_j}^2 N Q j ′ 2 -integrals caused by the M n M_n M n inserted points is O ( ( r / L ) 1 / 2 ) + o ( 1 ) O((r/L)^{1/2})+o(1) O (( r / L ) 1/2 ) + o ( 1 ) as n → ∞ n\to\infty n → ∞ with L L L fixed. Taking L L L large and then n n n large gives both required approximations. The point count per fundamental domain gives the asserted intensity ρ n \rho_n ρ n . □
Fix a countable family ( f k ) k ≥ 1 ⊂ C c + ( R d ) (f_k)_{k\ge1}\subset C^+_c(\mathbb{R}^d) ( f k ) k ≥ 1 ⊂ C c + ( R d ) such that, for every m ≥ 1 m\ge1 m ≥ 1 , every f ∈ C c + ( R d ) f\in C^+_c(\mathbb{R}^d) f ∈ C c + ( R d ) with supp f ⊂ B m \operatorname{supp} f\subset B_m supp f ⊂ B m is a uniform limit of functions from this family supported in B m + 1 B_{m+1} B m + 1 , and put
h k ( ω ) : = exp ( − S f k ( ω ) ) . h_k(\omega):=\exp(-S_{f_k}(\omega)). h k ( ω ) := exp ( − S f k ( ω )) .
Each h k h_k h k is a bounded continuous local function. If ∥ f − f k ∥ ∞ < ε \|f-f_k\|_\infty<\varepsilon ∥ f − f k ∥ ∞ < ε and both functions are supported in B m + 1 B_{m+1} B m + 1 , then, for every translation-invariant measure η ′ \eta' η ′ of intensity at most ρ \rho ρ ,
∣ ∫ e − S f d η ′ − ∫ e − S f k d η ′ ∣ ≤ ε ρ Vol d ( B m + 1 ) . \left|\int e^{-S_f}\,d\eta'-\int e^{-S_{f_k}}\,d\eta'\right|\le\varepsilon\rho\operatorname{Vol}_d(B_{m+1}). ∫ e − S f d η ′ − ∫ e − S f k d η ′ ≤ ερ Vol d ( B m + 1 ) .
The Laplace-functional criterion [23 ] shows that ( h k ) (h_k) ( h k ) determines weak convergence of point-process laws. Applying Lemma 5.1 successively to h 1 , … , h n h_1,\ldots,h_n h 1 , … , h n and Q 1 , … , Q n Q_1,\ldots,Q_n Q 1 , … , Q n gives weak convergence together with convergence of the required local second moments.
A process η ∈ E P P ρ , 2 \eta\in\mathbf{EPP}_{\rho,2} η ∈ EPP ρ , 2 is periodic if there is a full-rank lattice L < R d \mathbf{L}<\mathbb{R}^d L < R d such that θ t ω = ω \theta_t\omega=\omega θ t ω = ω for every t ∈ L t\in\mathbf{L} t ∈ L , for η \eta η -almost every ω \omega ω . Let Per ρ \operatorname{Per}_\rho Per ρ be the class of periodic processes in E P P ρ , 2 \mathbf{EPP}_{\rho,2} EPP ρ , 2 .
Proposition 5.2. The class Per ρ \operatorname{Per}_\rho Per ρ is contained in Hyp ρ \operatorname{Hyp}_\rho Hyp ρ and is dense in PP ρ , 2 \operatorname{PP}_{\rho,2} PP ρ , 2 . Consequently Hyp ρ ∩ E P P ρ , 2 \operatorname{Hyp}_\rho\cap\mathbf{EPP}_{\rho,2} Hyp ρ ∩ EPP ρ , 2 is dense in E P P ρ , 2 \mathbf{EPP}_{\rho,2} EPP ρ , 2 .
Proof. Fix η ∈ P P ρ , 2 \eta\in\mathrm{PP}_{\rho,2} η ∈ PP ρ , 2 . For each n n n , apply Lemma 5.1 to the full sequence
s j : = ( j / ρ ) 1 / d , L j : = s j Z d , N j : = j ( j ≥ 1 ) , s_j := (j/\rho)^{1/d}, \qquad\mathbf{L}_j := s_j\mathbb{Z}^d, \qquad N_j := j \quad(j \ge1), s j := ( j / ρ ) 1/ d , L j := s j Z d , N j := j ( j ≥ 1 ) ,
using the first n n n functions h 1 , … , h n h_1,\ldots,h_n h 1 , … , h n , the first n n n cubes Q 1 , … , Q n Q_1,\ldots,Q_n Q 1 , … , Q n , and tolerance 1 / n 1/n 1/ n . The lemma gives j n j_n j n such that every j ≥ j n j \ge j_n j ≥ j n admits the required periodic approximation. Choose increasing integers q n ≥ j n q_n \ge j_n q n ≥ j n , and put S n : = ( q n / ρ ) 1 / d S_n := (q_n/\rho)^{1/d} S n := ( q n / ρ ) 1/ d , L n : = S n Z d \mathbf{L}_n := S_n\mathbb{Z}^d L n := S n Z d . Let P n P_n P n be a L n \mathbf{L}_n L n -periodic set with exactly q n q_n q n points per fundamental domain. Thus dens ( P n ) = ρ \operatorname{dens}(P_n) = \rho dens ( P n ) = ρ and η P n ∈ Per ρ \eta_{P_n} \in\operatorname{Per}_\rho η P n ∈ Per ρ . The approximations of the Laplace functionals h k h_k h k give weak convergence. The second-moment estimates then give η P n → η \eta_{P_n} \to\eta η P n → η in P P ρ , 2 \mathrm{PP}_{\rho,2} PP ρ , 2 by Proposition 2.7 . Thus Per ρ \operatorname{Per}_\rho Per ρ is dense.
Fix η ∈ Per ρ \eta\in\operatorname{Per}_\rho η ∈ Per ρ and choose a full-rank lattice L \mathbf{L} L for which θ t ω = ω \theta_t\omega= \omega θ t ω = ω for every t ∈ L t \in\mathbf{L} t ∈ L , for η \eta η -almost every ω \omega ω . Let m L m_{\mathbf{L}} m L be Haar probability on R d / L \mathbb{R}^d/\mathbf{L} R d / L . For every bounded continuous function h h h on configuration space, the orbit average h # ( ω ) : = ∫ h ( θ t ω ) d m L ( t ) h^{\#}(\omega) := \int h(\theta_t\omega)\,dm_{\mathbf{L}}(t) h # ( ω ) := ∫ h ( θ t ω ) d m L ( t ) is invariant. Ergodicity makes this average equal to ∫ h d η \int h\,d\eta ∫ h d η for η \eta η -almost every ω \omega ω . Applying this to a countable determining family shows that η = η P \eta= \eta_P η = η P for a L \mathbf{L} L -periodic set P P P . Its density equals ρ \rho ρ . Choose a half-open fundamental domain D L D_{\mathbf{L}} D L for L \mathbf{L} L and put P 0 : = P ∩ D L P_0 := P \cap D_{\mathbf{L}} P 0 := P ∩ D L . For ξ ∈ L ∗ \xi\in\mathbf{L}^* ξ ∈ L ∗ , set
A ( ξ ) : = ∑ p ∈ P 0 e − 2 π i ⟨ ξ , p ⟩ . A(\xi) := \sum_{p\in P_0} e^{-2\pi i\langle\xi,p\rangle}. A ( ξ ) := p ∈ P 0 ∑ e − 2 π i ⟨ ξ , p ⟩ .
For f ∈ C c ( R d ) f \in C_c(\mathbb{R}^d) f ∈ C c ( R d ) , expand t ↦ S f ( δ P + t ) t \mapsto S_f(\delta_{P+t}) t ↦ S f ( δ P + t ) in Fourier series on R d / L \mathbb{R}^d/\mathbf{L} R d / L . After subtracting the zero-frequency term, the coefficient at ξ ∈ L ∗ ∖ { 0 } \xi\in\mathbf{L}^* \setminus\{0\} ξ ∈ L ∗ ∖ { 0 } has absolute value ∣ A ( ξ ) f ^ ( ξ ) ∣ / covol ( L ) |A(\xi)\widehat{f}(\xi)|/\operatorname{covol}(\mathbf{L}) ∣ A ( ξ ) f ( ξ ) ∣/ covol ( L ) . Orthogonality of the characters and Parseval’s identity give
σ η P = 1 covol ( L ) 2 ∑ ξ ∈ L ∗ ∖ { 0 } ∣ A ( ξ ) ∣ 2 δ ξ . \sigma_{\eta_P} = \frac{1}{\operatorname{covol}(\mathbf{L})^2}\sum_{\xi\in\mathbf{L}^*\setminus\{0\}} |A(\xi)|^2\,\delta_\xi. σ η P = covol ( L ) 2 1 ξ ∈ L ∗ ∖ { 0 } ∑ ∣ A ( ξ ) ∣ 2 δ ξ .
Parseval’s identity on R d / L \mathbb{R}^d/\mathbf{L} R d / L gives
V η P ( R ) = 1 covol ( L ) 2 ∑ ξ ∈ L ∗ ∖ { 0 } ∣ A ( ξ ) ∣ 2 ∣ 1 ^ B R ( ξ ) ∣ 2 . V_{\eta_P}(R) = \frac{1}{\operatorname{covol}(\mathbf{L})^2}\sum_{\xi\in\mathbf{L}^*\setminus\{0\}} |A(\xi)|^2\left|\widehat{\mathbf{1}}_{B_R}(\xi)\right|^2. V η P ( R ) = covol ( L ) 2 1 ξ ∈ L ∗ ∖ { 0 } ∑ ∣ A ( ξ ) ∣ 2 1 B R ( ξ ) 2 .
Since ∣ A ( ξ ) ∣ ≤ ρ covol ( L ) |A(\xi)| \le\rho\operatorname{covol}(\mathbf{L}) ∣ A ( ξ ) ∣ ≤ ρ covol ( L ) , for R ≥ λ 1 ( L ∗ ) − 1 R \ge\lambda_1(\mathbf{L}^*)^{-1} R ≥ λ 1 ( L ∗ ) − 1 the Bessel-function estimate
∣ 1 ^ B R ( ξ ) ∣ ≤ C d R ( d − 1 ) / 2 ∣ ξ ∣ − ( d + 1 ) / 2 ( R ∣ ξ ∣ ≥ 1 ) \left|\widehat{\mathbf{1}}_{B_R}(\xi)\right| \le C_d R^{(d-1)/2}|\xi|^{-(d+1)/2} \qquad(R|\xi| \ge1) 1 B R ( ξ ) ≤ C d R ( d − 1 ) /2 ∣ ξ ∣ − ( d + 1 ) /2 ( R ∣ ξ ∣ ≥ 1 )
(see [39 Sec. IV and Appendix A] ) yields
V η P ( R ) ≤ C P , d R d − 1 ∑ ξ ∈ L ∗ ∖ { 0 } ∣ ξ ∣ − ( d + 1 ) = O P ( R d − 1 ) . V_{\eta_P}(R) \le C_{P,d}R^{d-1}\sum_{\xi\in\mathbf{L}^*\setminus\{0\}}|\xi|^{-(d+1)} = O_P(R^{d-1}). V η P ( R ) ≤ C P , d R d − 1 ξ ∈ L ∗ ∖ { 0 } ∑ ∣ ξ ∣ − ( d + 1 ) = O P ( R d − 1 ) .
Thus every η ∈ Per ρ \eta\in\operatorname{Per}_\rho η ∈ Per ρ is hyperuniform. □
Proof of Theorem A. The residuality of I H y p ρ \mathrm{IHyp}_\rho IHyp ρ and the meagreness of H y p ρ ∩ E P P ρ , 2 \mathrm{Hyp}_\rho\cap\mathrm{EPP}_{\rho,2} Hyp ρ ∩ EPP ρ , 2 were proved at the end of § 3 ; Proposition 5.2 provides the remaining density assertion.
Pairwise negative correlation
The spectral representation of a reduced covariance measure belongs to the Bartlett-spectrum framework; see [8 Sec. 8.2] . For pairwise negative correlation we show that its off-diagonal part is a finite negative measure. This yields a bounded continuous spectral density and the variance limit used below.
A process η ∈ P P ρ , 2 \eta\in\mathrm{PP}_{\rho,2} η ∈ PP ρ , 2 is pairwise negatively correlated if
Cov η ( N A , N B ) ≤ 0 \operatorname{Cov}_{\eta}(N_A,N_B) \le0 Cov η ( N A , N B ) ≤ 0
for every pair of disjoint bounded Borel sets A , B ⊂ R d A, B \subset\mathbb{R}^{d} A , B ⊂ R d . Let P N C ρ \mathrm{PNC}_{\rho} PNC ρ denote this class. This condition is weaker than negative association [TPBG25].
Remark 5.3. The fair-partition construction of [27 Example 4.7] belongs to P N C ρ \mathrm{PNC}_{\rho} PNC ρ . Conditionally on the partition ( C i ) i (C_i)_i ( C i ) i , set
p i ( A ) : = ρ Vol d ( A ∩ C i ) . p_i(A) := \rho\operatorname{Vol}_{d}(A \cap C_i). p i ( A ) := ρ Vol d ( A ∩ C i ) .
For disjoint bounded Borel sets A , B A, B A , B , conditional independence between cells and the fact that each cell contributes at most one point give
Cov ( N A , N B ∣ ( C i ) i ) = − ∑ i p i ( A ) p i ( B ) . \operatorname{Cov}(N_A,N_B \mid(C_i)_i) = -\sum_i p_i(A)p_i(B). Cov ( N A , N B ∣ ( C i ) i ) = − i ∑ p i ( A ) p i ( B ) .
Since E [ N A ∣ ( C i ) i ] = ∑ i p i ( A ) = ρ Vol d ( A ) \mathbb{E}[N_A \mid(C_i)_i] = \sum_i p_i(A) = \rho\operatorname{Vol}_{d}(A) E [ N A ∣ ( C i ) i ] = ∑ i p i ( A ) = ρ Vol d ( A ) is deterministic,
Cov ( N A , N B ) = − E ∑ i p i ( A ) p i ( B ) ≤ 0. \operatorname{Cov}(N_A,N_B) = -\mathbb{E}\sum_i p_i(A)p_i(B) \le0. Cov ( N A , N B ) = − E i ∑ p i ( A ) p i ( B ) ≤ 0.
Recall the definition of a Følner sequence from Subsection 1.4. A process η ∈ P P ρ , 2 \eta\in\mathrm{PP}_{\rho,2} η ∈ PP ρ , 2 is Følner-hyperuniform if
Var η ( N F n ) Vol d ( F n ) ⟶ 0 ( n → ∞ ) . \frac{\operatorname{Var}_{\eta}(N_{F_n})}{\operatorname{Vol}_{d}(F_n)} \longrightarrow0 \qquad(n \to\infty). Vol d ( F n ) Var η ( N F n ) ⟶ 0 ( n → ∞ ) .
for every Følner sequence ( F n ) (F_n) ( F n ) . Denote this class by F H y p ρ \mathrm{FHyp}_{\rho} FHyp ρ .
Proposition 5.4. The class P N C ρ \mathrm{PNC}_{\rho} PNC ρ is closed in P P ρ , 2 \mathrm{PP}_{\rho,2} PP ρ , 2 . For every η ∈ P N C ρ \eta\in\mathrm{PNC}_{\rho} η ∈ PNC ρ there is a finite symmetric positive measure τ η \tau_{\eta} τ η on R d \mathbb{R}^{d} R d , with
τ η ( R d ) ≤ ρ , \tau_{\eta}(\mathbb{R}^{d}) \le\rho, τ η ( R d ) ≤ ρ , such that, for real bounded Borel functions f , g f,g f , g with compact support,
Cov η ( S f , S g ) = ρ ∫ f ( u ) g ( u ) d u − ∫ R d ∫ R d f ( u ) g ( u + t ) d u d τ η ( t ) . \operatorname{Cov}_{\eta}(S_f,S_g) = \rho\int f(u)g(u)\,\mathrm{d}u - \int_{\mathbb{R}^{d}}\int_{\mathbb{R}^{d}} f(u)g(u+t)\,\mathrm{d}u\,\mathrm{d}\tau_{\eta}(t). Cov η ( S f , S g ) = ρ ∫ f ( u ) g ( u ) d u − ∫ R d ∫ R d f ( u ) g ( u + t ) d u d τ η ( t ) . Equivalently,
Cov η ( S f , S g ) = ∫ R d f ^ ( ξ ) g ^ ( ξ ) s η ( ξ ) d ξ , s η ( ξ ) : = ρ − τ ^ η ( ξ ) , \operatorname{Cov}_{\eta}(S_f,S_g) = \int_{\mathbb{R}^{d}} \widehat{f}(\xi)\widehat{g}(\xi)s_{\eta}(\xi)\,\mathrm{d}\xi,\qquad s_{\eta}(\xi) := \rho- \widehat{\tau}_{\eta}(\xi), Cov η ( S f , S g ) = ∫ R d f ( ξ ) g ( ξ ) s η ( ξ ) d ξ , s η ( ξ ) := ρ − τ η ( ξ ) , so
σ η ( d ξ ) = s η ( ξ ) d ξ , 0 ≤ s η ≤ 2 ρ , \sigma_{\eta}(\mathrm{d}\xi) = s_{\eta}(\xi)\,\mathrm{d}\xi,\qquad0 \le s_{\eta} \le2\rho, σ η ( d ξ ) = s η ( ξ ) d ξ , 0 ≤ s η ≤ 2 ρ , with s η s_{\eta} s η bounded and continuous. Moreover, for every Følner sequence ( F n ) (F_n) ( F n ) ,
Var η ( N F n ) Vol d ( F n ) ⟶ ρ − τ η ( R d ) = s η ( 0 ) ( n → ∞ ) . \frac{\operatorname{Var}_{\eta}(N_{F_n})}{\operatorname{Vol}_{d}(F_n)} \longrightarrow\rho-\tau_{\eta}(\mathbb{R}^{d}) = s_{\eta}(0) \qquad(n \to\infty). Vol d ( F n ) Var η ( N F n ) ⟶ ρ − τ η ( R d ) = s η ( 0 ) ( n → ∞ ) . Consequently, for η ∈ P N C ρ \eta\in\mathrm{PNC}_{\rho} η ∈ PNC ρ , the following are equivalent:
η ∈ H y p ρ , η ∈ F H y p ρ , s η ( 0 ) = 0. \eta\in\mathrm{Hyp}_{\rho},\qquad\eta\in\mathrm{FHyp}_{\rho},\qquad s_{\eta}(0)=0. η ∈ Hyp ρ , η ∈ FHyp ρ , s η ( 0 ) = 0. Finally, for real f , g ∈ C c ( R d ) f,g \in C_c(\mathbb{R}^{d}) f , g ∈ C c ( R d ) ,
Cov η ( S f , S g ∘ θ t ) ⟶ 0 ( ∣ t ∣ → ∞ ) . \operatorname{Cov}_{\eta}(S_f,S_g \circ\theta_t) \longrightarrow0 \qquad(|t| \to\infty). Cov η ( S f , S g ∘ θ t ) ⟶ 0 ( ∣ t ∣ → ∞ ) .
Proof. Let R \mathcal{R} R be the countable ring generated by half-open rectangles with rational endpoints. Translation invariance and Lemma 2.3 imply that every A ∈ R A \in\mathcal{R} A ∈ R is a continuity set for every η ∈ P P ρ , 2 \eta\in\mathrm{PP}_{\rho,2} η ∈ PP ρ , 2 . Therefore, for disjoint A , B ∈ R A,B \in\mathcal{R} A , B ∈ R ,
η ⟼ Cov η ( N A , N B ) \eta\longmapsto\operatorname{Cov}_{\eta}(N_A,N_B) η ⟼ Cov η ( N A , N B )
is continuous in the local L 2 L^2 L 2 topology. Let α η ( 2 ) \alpha_\eta^{(2)} α η ( 2 ) be the second factorial moment measure,
∫ H ( x , y ) d α η ( 2 ) ( x , y ) : = ∫ ∑ p , q ∈ supp ω p ≠ q H ( p , q ) d η ( ω ) , H : R 2 d → [ 0 , ∞ ] Borel . \int H(x,y)\,d\alpha_\eta^{(2)}(x,y):=\int\sum_{\substack{p,q\in\operatorname{supp}\omega\\p\ne q}} H(p,q)\,d\eta(\omega),\qquad H:\mathbb{R}^{2d}\to[0,\infty] \text{ Borel}. ∫ H ( x , y ) d α η ( 2 ) ( x , y ) := ∫ p , q ∈ supp ω p = q ∑ H ( p , q ) d η ( ω ) , H : R 2 d → [ 0 , ∞ ] Borel .
For disjoint A , B A,B A , B ,
Cov η ( N A , N B ) ≤ 0 ⟺ α η ( 2 ) ( A × B ) ≤ ρ 2 Vol d ( A ) Vol d ( B ) . \operatorname{Cov}_\eta(N_A,N_B)\le0\qquad\Longleftrightarrow\qquad\alpha_\eta^{(2)}(A\times B)\le\rho^2\operatorname{Vol}_d(A)\operatorname{Vol}_d(B). Cov η ( N A , N B ) ≤ 0 ⟺ α η ( 2 ) ( A × B ) ≤ ρ 2 Vol d ( A ) Vol d ( B ) .
Thus the inequalities for disjoint A , B ∈ R A,B\in\mathcal{R} A , B ∈ R hold on a closed subset of P P ρ , 2 \mathrm{PP}_{\rho,2} PP ρ , 2 . To prove that they imply the inequalities for all disjoint bounded Borel sets, let H ∈ C c ( R d × R d ) H\in C_c(\mathbb{R}^d\times\mathbb{R}^d) H ∈ C c ( R d × R d ) be nonnegative with support disjoint from the diagonal. Partition a large box containing its support into sufficiently small sets from R \mathcal{R} R . Only products of distinct cells intersect the support of H H H , so step-function approximation and the rectangle inequalities give
∫ H d α η ( 2 ) ≤ ρ 2 ∫ H ( x , y ) d x d y . \int H\,d\alpha_\eta^{(2)}\le\rho^2\int H(x,y)\,dx\,dy. ∫ H d α η ( 2 ) ≤ ρ 2 ∫ H ( x , y ) d x d y .
The inequality holds for every such H H H . Regularity of the measures therefore gives
α η ( 2 ) ≤ ρ 2 d x d y off the diagonal . \alpha_\eta^{(2)}\le\rho^2\,dx\,dy\qquad\text{off the diagonal}. α η ( 2 ) ≤ ρ 2 d x d y off the diagonal .
In particular, the covariance inequality holds for all disjoint bounded Borel sets, so P N C ρ \mathrm{PNC}_\rho PNC ρ is closed.
Since η \eta η is simple, the factorial moment measure has no mass on the diagonal. Thus the off-diagonal inequality also gives α η ( 2 ) ≤ ρ 2 d x d y \alpha_\eta^{(2)}\le\rho^2dx\,dy α η ( 2 ) ≤ ρ 2 d x d y on all of R 2 d \mathbb{R}^{2d} R 2 d . Define the positive measure
Θ η : = ρ 2 d x d y − α η ( 2 ) . \Theta_\eta:=\rho^2\,dx\,dy-\alpha_\eta^{(2)}. Θ η := ρ 2 d x d y − α η ( 2 ) .
It is invariant under simultaneous translation of x x x and y y y . Let Θ η Δ \Theta_\eta^\Delta Θ η Δ be the push-forward of Θ η \Theta_\eta Θ η under ( x , y ) ↦ ( x , y − x ) (x,y)\mapsto(x,y-x) ( x , y ) ↦ ( x , y − x ) . For each bounded Borel C ⊂ R d ∖ { 0 } C\subset\mathbb{R}^d\setminus\{0\} C ⊂ R d ∖ { 0 } , the measure
A ⟼ Θ η Δ ( A × C ) A\longmapsto\Theta_\eta^\Delta(A\times C) A ⟼ Θ η Δ ( A × C )
is locally finite and translation-invariant. It is therefore a multiple of Lebesgue measure. Given a bounded Borel set U U U of positive volume, define
τ η ( C ) : = Θ η Δ ( U × C ) Vol d ( U ) . \tau_\eta(C):=\frac{\Theta_\eta^\Delta(U\times C)}{\operatorname{Vol}_d(U)}. τ η ( C ) := Vol d ( U ) Θ η Δ ( U × C ) .
Translation invariance makes this independent of U U U . It defines a locally finite positive measure on R d ∖ { 0 } \mathbb{R}^d\setminus\{0\} R d ∖ { 0 } . Set τ η ( { 0 } ) = 0 \tau_\eta(\{0\})=0 τ η ({ 0 }) = 0 . Then
Θ η Δ ( d x , d t ) = d x d τ η ( t ) , d Θ η ( x , y ) = d x d τ η ( y − x ) . \Theta_\eta^\Delta(dx,dt)=dx\,d\tau_\eta(t),\qquad d\Theta_\eta(x,y)=dx\,d\tau_\eta(y-x). Θ η Δ ( d x , d t ) = d x d τ η ( t ) , d Θ η ( x , y ) = d x d τ η ( y − x ) .
Symmetry of α η ( 2 ) \alpha_\eta^{(2)} α η ( 2 ) makes τ η \tau_\eta τ η symmetric.
For a bounded Borel set F F F with 0 < Vol d ( F ) < ∞ 0<\operatorname{Vol}_d(F)<\infty 0 < Vol d ( F ) < ∞ , the factorial-moment identity gives
Var η ( N F ) = ρ Vol d ( F ) + α η ( 2 ) ( F × F ) − ρ 2 Vol d ( F ) 2 = ρ Vol d ( F ) − Θ η ( F × F ) . \operatorname{Var}_\eta(N_F)=\rho\operatorname{Vol}_d(F)+\alpha_\eta^{(2)}(F\times F)-\rho^2\operatorname{Vol}_d(F)^2=\rho\operatorname{Vol}_d(F)-\Theta_\eta(F\times F). Var η ( N F ) = ρ Vol d ( F ) + α η ( 2 ) ( F × F ) − ρ 2 Vol d ( F ) 2 = ρ Vol d ( F ) − Θ η ( F × F ) .
Using d Θ η ( x , y ) = d x d τ η ( y − x ) d\Theta_\eta(x,y)=dx\,d\tau_\eta(y-x) d Θ η ( x , y ) = d x d τ η ( y − x ) , we obtain
Θ η ( F × F ) = ∫ R d Vol d ( F ∩ ( F − t ) ) d τ η ( t ) . \Theta_\eta(F\times F)=\int_{\mathbb{R}^d}\operatorname{Vol}_d(F\cap(F-t))\,d\tau_\eta(t). Θ η ( F × F ) = ∫ R d Vol d ( F ∩ ( F − t )) d τ η ( t ) .
Therefore
Var η ( N F ) Vol d ( F ) = ρ − ∫ R d c F ( t ) d τ η ( t ) , c F ( t ) : = Vol d ( F ∩ ( F − t ) ) Vol d ( F ) . (45) \frac{\operatorname{Var}_\eta(N_F)}{\operatorname{Vol}_d(F)}
=\rho-\int_{\mathbb{R}^d}c_F(t)\,d\tau_\eta(t),\qquad c_F(t):=\frac{\operatorname{Vol}_d(F\cap(F-t))}{\operatorname{Vol}_d(F)}. \tag*{(45)} Vol d ( F ) Var η ( N F ) = ρ − ∫ R d c F ( t ) d τ η ( t ) , c F ( t ) := Vol d ( F ) Vol d ( F ∩ ( F − t )) . (45)
For expanding cubes Q R Q_R Q R , the functions c Q R c_{Q_R} c Q R satisfy 0 ≤ c Q R ≤ 1 0\le c_{Q_R}\le1 0 ≤ c Q R ≤ 1 and c Q R ( t ) → 1 c_{Q_R}(t)\to1 c Q R ( t ) → 1 for every t t t . Since Var η ( N Q R ) ≥ 0 \operatorname{Var}_\eta(N_{Q_R})\ge0 Var η ( N Q R ) ≥ 0 , (45) gives ∫ c Q R d τ η ≤ ρ \int c_{Q_R}\,d\tau_\eta\le\rho ∫ c Q R d τ η ≤ ρ . Fatou’s lemma now yields
τ η ( R d ) ≤ ρ . \tau_\eta(\mathbb{R}^d)\le\rho. τ η ( R d ) ≤ ρ .
For bounded Borel functions f , g f,g f , g with compact support, the factorial moment identity is
∫ S f S g d η = ρ ∫ f ( u ) g ( u ) d u + ∬ f ( x ) g ( y ) d α η ( 2 ) ( x , y ) . \int S_fS_g\,d\eta=\rho\int f(u)g(u)\,du+\iint f(x)g(y)\,d\alpha_\eta^{(2)}(x,y). ∫ S f S g d η = ρ ∫ f ( u ) g ( u ) d u + ∬ f ( x ) g ( y ) d α η ( 2 ) ( x , y ) .
Substituting α η ( 2 ) = ρ 2 d x d y − Θ η \alpha_\eta^{(2)}=\rho^2dx\,dy-\Theta_\eta α η ( 2 ) = ρ 2 d x d y − Θ η and subtracting the product of the first moments gives
Cov η ( S f , S g ) = ρ ∫ f g − ∬ f ( u ) g ( u + t ) d u d τ η ( t ) . \operatorname{Cov}_\eta(S_f,S_g)=\rho\int fg-\iint f(u)g(u+t)\,du\,d\tau_\eta(t). Cov η ( S f , S g ) = ρ ∫ f g − ∬ f ( u ) g ( u + t ) d u d τ η ( t ) .
For real f , g ∈ C c ∞ f,g\in C_c^\infty f , g ∈ C c ∞ , Fourier inversion and Plancherel give the stated density s η = ρ − τ ^ η s_\eta=\rho-\widehat{\tau}_\eta s η = ρ − τ η . The corresponding identity for complex-valued smooth tests follows by sesquilinearity. For bounded Borel functions f , g f,g f , g with compact support, approximate them in L 2 ( d x ) L^2(dx) L 2 ( d x ) by smooth compactly supported functions with supports in one fixed compact set. As n → ∞ n\to\infty n → ∞ , Cauchy–Schwarz gives, uniformly in t t t ,
∣ ∫ f ( u ) g ( u + t ) d u − ∫ f n ( u ) g n ( u + t ) d u ∣ ≤ ∥ f − f n ∥ 2 ∥ g ∥ 2 + ∥ f n ∥ 2 ∥ g − g n ∥ 2 ⟶ 0. \left|\int f(u)g(u+t)\,du-\int f_n(u)g_n(u+t)\,du\right|\leq\|f-f_n\|_2\|g\|_2+\|f_n\|_2\|g-g_n\|_2\longrightarrow0. ∫ f ( u ) g ( u + t ) d u − ∫ f n ( u ) g n ( u + t ) d u ≤ ∥ f − f n ∥ 2 ∥ g ∥ 2 + ∥ f n ∥ 2 ∥ g − g n ∥ 2 ⟶ 0.
Since τ η \tau_\eta τ η is finite, the covariance formula passes to the limit and Plancherel gives the corresponding Fourier-side identity; since τ η \tau_\eta τ η is also symmetric, s η s_\eta s η is bounded, continuous, and real-valued. The covariance form is nonnegative when f = g f=g f = g , so s η ≥ 0 s_\eta\geq0 s η ≥ 0 almost everywhere; continuity gives s η ≥ 0 s_\eta\geq0 s η ≥ 0 everywhere. In addition,
s η ( ξ ) ≤ ρ + ∣ τ ^ η ( ξ ) ∣ ≤ 2 ρ . s_\eta(\xi)\leq\rho+\left|\widehat{\tau}_\eta(\xi)\right|\leq2\rho. s η ( ξ ) ≤ ρ + ∣ τ η ( ξ ) ∣ ≤ 2 ρ .
For f , g ∈ C c ∞ f,g\in C_c^\infty f , g ∈ C c ∞ , this is the Bartlett identity with spectral measure s η d ξ s_\eta\,d\xi s η d ξ . Approximation in the Schwartz topology, using translation boundedness, and uniqueness in Proposition 2.8 give σ η = s η d ξ \sigma_\eta=s_\eta\,d\xi σ η = s η d ξ .
If ( F n ) (F_n) ( F n ) is a Følner sequence, then
c F n ( t ) = 1 − Vol d ( F n △ ( F n − t ) ) 2 Vol d ( F n ) ⟶ 1 c_{F_n}(t)=1-\frac{\operatorname{Vol}_d(F_n\mathbin{\triangle}(F_n-t))}{2\operatorname{Vol}_d(F_n)}\longrightarrow1 c F n ( t ) = 1 − 2 Vol d ( F n ) Vol d ( F n △ ( F n − t )) ⟶ 1
for every t t t . Dominated convergence with respect to the finite measure τ η \tau_\eta τ η proves the Følner variance limit.
Since the balls form a Følner sequence, F H y p ρ ⊆ H y p ρ \mathrm{FHyp}_\rho\subseteq\mathrm{Hyp}_\rho FHyp ρ ⊆ Hyp ρ . If η \eta η is hyperuniform, the Følner variance limit along balls gives s η ( 0 ) = 0 s_\eta(0)=0 s η ( 0 ) = 0 ; conversely, s η ( 0 ) = 0 s_\eta(0)=0 s η ( 0 ) = 0 makes the displayed limit zero along every Følner sequence. This proves the three-way equivalence.
For real f , g ∈ C c ( R d ) f,g\in C_c(\mathbb{R}^d) f , g ∈ C c ( R d ) , the product s η f ^ g ^ s_\eta\widehat{f}\widehat{g} s η f g belongs to L 1 ( R d ) L^1(\mathbb{R}^d) L 1 ( R d ) . The Riemann–Lebesgue lemma therefore gives
Cov η ( S f , S g ∘ θ t ) ⟶ 0 ( ∣ t ∣ → ∞ ) . \operatorname{Cov}_\eta(S_f,S_g\circ\theta_t)\longrightarrow0\qquad(|t|\to\infty). Cov η ( S f , S g ∘ θ t ) ⟶ 0 ( ∣ t ∣ → ∞ ) .
Corollary 5.5. The class P N C ρ \mathrm{PNC}_\rho PNC ρ is nowhere dense in P P ρ , 2 \mathrm{PP}_{\rho,2} PP ρ , 2 .
Proof. By Proposition 5.4 , P N C ρ \mathrm{PNC}_\rho PNC ρ is closed. Let P P P be a periodic set with dens ( P ) = ρ \operatorname{dens}(P)=\rho dens ( P ) = ρ . Its orbit measure η P \eta_P η P is not a point mass, so Var η P ( S f ) > 0 \operatorname{Var}_{\eta_P}(S_f)>0 Var η P ( S f ) > 0 for some f ∈ C c ( R d ) f\in C_c(\mathbb{R}^d) f ∈ C c ( R d ) . If 0 ≠ λ ∈ Stab ( P ) 0\ne\lambda\in\operatorname{Stab}(P) 0 = λ ∈ Stab ( P ) , then
Cov η P ( S f , S f ∘ θ n λ ) = Var η P ( S f ) > 0 ( n ≥ 1 ) , \operatorname{Cov}_{\eta_P}(S_f,S_f\circ\theta_{n\lambda})=\operatorname{Var}_{\eta_P}(S_f)>0\qquad(n\geq1), Cov η P ( S f , S f ∘ θ nλ ) = Var η P ( S f ) > 0 ( n ≥ 1 ) ,
contradicting the final assertion of Proposition 5.4 . Hence P e r ρ ∩ P N C ρ = ∅ \mathrm{Per}_\rho\cap\mathrm{PNC}_\rho=\varnothing Per ρ ∩ PNC ρ = ∅ . Since P e r ρ \mathrm{Per}_\rho Per ρ is dense by Proposition 5.2 , the closed set P N C ρ \mathrm{PNC}_\rho PNC ρ has empty interior. □
Revival and near-quadratic van Hove fluctuations
Revival sequences for point processes
Proposition 6.1. The set Rev ρ \operatorname{Rev}_{\rho} Rev ρ is residual in EPP ρ , 2 \operatorname{EPP}_{\rho,2} EPP ρ , 2 .
Proof. Let ( A j ) j ≥ 1 (A_j)_{j\geq1} ( A j ) j ≥ 1 enumerate the countable ring of finite unions of bounded half-open rectangles with rational endpoints. For m , ℓ ≥ 1 m,\ell\geq1 m , ℓ ≥ 1 , define
V m , ℓ : = ⋃ ∣ t ∣ > ℓ ⋂ j = 1 m { η ∈ EPP ρ , 2 : ∥ N A j + t − N A j ∥ L 2 ( η ) < m − 1 } . \mathcal{V}_{m,\ell}:=\bigcup_{|t|>\ell}\bigcap_{j=1}^{m}\left\{\eta\in\operatorname{EPP}_{\rho,2}:\left\|N_{A_j+t}-N_{A_j}\right\|_{L^2(\eta)}<m^{-1}\right\}. V m , ℓ := ∣ t ∣ > ℓ ⋃ j = 1 ⋂ m { η ∈ EPP ρ , 2 : N A j + t − N A j L 2 ( η ) < m − 1 } .
Fix j j j and t t t , and suppose η k → η \eta_k\to\eta η k → η locally in L 2 L^2 L 2 . Choose m m m with A j ∪ ( A j + t ) ⊂ Q m A_j\cup(A_j+t)\subset Q_m A j ∪ ( A j + t ) ⊂ Q m . By Lemma 2.3, both count functions are almost surely continuous under η \eta η for the vague topology. The expectations of S φ m 2 S_{\varphi_m}^2 S φ m 2 converge by definition of the topology. Since S φ m 2 S_{\varphi_m}^2 S φ m 2 is continuous in the vague topology, weak convergence and convergence of its integrals give
lim M → ∞ sup k ≥ 1 ∫ S φ m 2 1 { S φ m 2 > M } d η k = 0. \lim_{M\to\infty}\sup_{k\geq1}\int S_{\varphi_m}^2\mathbf{1}_{\{S_{\varphi_m}^2>M\}}\,d\eta_k=0. M → ∞ lim k ≥ 1 sup ∫ S φ m 2 1 { S φ m 2 > M } d η k = 0.
The squared count difference is almost surely continuous under η \eta η and bounded by 4 S φ m 2 4S_{\varphi_m}^2 4 S φ m 2 . Thus the expectation of ( N A j + t − N A j ) 2 (N_{A_j+t}-N_{A_j})^2 ( N A j + t − N A j ) 2 is continuous in η \eta η . Each V m , ℓ \mathcal{V}_{m,\ell} V m , ℓ is therefore open.
If P P P is periodic, then for t ∈ Stab ( P ) t\in\operatorname{Stab}(P) t ∈ Stab ( P ) the configuration itself is unchanged by translation by t t t . Thus N A + t = N A N_{A+t}=N_A N A + t = N A η P \eta_P η P -almost surely. Since Stab ( P ) \operatorname{Stab}(P) Stab ( P ) contains vectors of arbitrarily large norm, every periodic invariant measure belongs to every V m , ℓ \mathcal{V}_{m,\ell} V m , ℓ . By Proposition 5.2, each V m , ℓ \mathcal{V}_{m,\ell} V m , ℓ is dense. Thus
V : = ⋂ m , ℓ ≥ 1 V m , ℓ \mathcal{V}:=\bigcap_{m,\ell\geq1}\mathcal{V}_{m,\ell} V := m , ℓ ≥ 1 ⋂ V m , ℓ
is a dense G δ G_\delta G δ subset of EPP ρ , 2 \operatorname{EPP}_{\rho,2} EPP ρ , 2 . Fix η ∈ V \eta\in\mathcal{V} η ∈ V . For each n n n , the condition η ∈ V n , n \eta\in\mathcal{V}_{n,n} η ∈ V n , n gives a vector t n t_n t n with ∣ t n ∣ > n |t_n|>n ∣ t n ∣ > n and
∥ N A j + t n − N A j ∥ L 2 ( η ) < n − 1 ( 1 ≤ j ≤ n ) . \left\|N_{A_j+t_n}-N_{A_j}\right\|_{L^2(\eta)}<n^{-1}\qquad(1\leq j\leq n). N A j + t n − N A j L 2 ( η ) < n − 1 ( 1 ≤ j ≤ n ) .
Let A A A be a bounded Borel set. Choose a bounded cube D ⊃ A D\supset A D ⊃ A . By approximation in Lebesgue measure, there are ring sets B k ⊂ D B_k\subset D B k ⊂ D such that
Vol d ( A △ B k ) ⟶ 0 ( k → ∞ ) . \operatorname{Vol}_{d}(A\mathbin{\triangle}B_k)\longrightarrow0\qquad(k\to\infty). Vol d ( A △ B k ) ⟶ 0 ( k → ∞ ) .
Translation invariance gives
∫ N A △ B k d η = ρ Vol d ( A △ B k ) ⟶ 0 ( k → ∞ ) . \int N_{A\mathbin{\triangle}B_k}\,d\eta=\rho\operatorname{Vol}_{d}(A\mathbin{\triangle}B_k)\longrightarrow0\qquad(k\to\infty). ∫ N A △ B k d η = ρ Vol d ( A △ B k ) ⟶ 0 ( k → ∞ ) .
Moreover 0 ≤ N A △ B k ≤ N D 0\leq N_{A\mathbin{\triangle}B_k}\leq N_D 0 ≤ N A △ B k ≤ N D and N D ∈ L 2 ( η ) N_D\in L^2(\eta) N D ∈ L 2 ( η ) . Markov’s inequality gives N A △ B k → 0 N_{A\mathbin{\triangle}B_k}\to0 N A △ B k → 0 in probability. Since N A △ B k 2 ≤ N D 2 ∈ L 1 ( η ) N_{A\mathbin{\triangle}B_k}^2\leq N_D^2\in L^1(\eta) N A △ B k 2 ≤ N D 2 ∈ L 1 ( η ) , uniform integrability upgrades convergence in probability to
∥ N A △ B k ∥ L 2 ( η ) ⟶ 0 ( k → ∞ ) . \left\|N_{A\mathbin{\triangle}B_k}\right\|_{L^2(\eta)}\longrightarrow0\qquad(k\to\infty). ∥ N A △ B k ∥ L 2 ( η ) ⟶ 0 ( k → ∞ ) .
For any such ring set B B B and every n n n , translation invariance and the triangle inequality give
∥ N A + t n − N A ∥ 2 ≤ ∥ N ( A △ B ) + t n ∥ 2 + ∥ N B + t n − N B ∥ 2 + ∥ N A △ B ∥ 2 ≤ 2 ∥ N A △ B ∥ 2 + ∥ N B + t n − N B ∥ 2 . \left\|N_{A+t_n}-N_A\right\|_2\leq\left\|N_{(A\mathbin{\triangle}B)+t_n}\right\|_2+\left\|N_{B+t_n}-N_B\right\|_2+\left\|N_{A\mathbin{\triangle}B}\right\|_2\leq2\left\|N_{A\mathbin{\triangle}B}\right\|_2+\left\|N_{B+t_n}-N_B\right\|_2. ∥ N A + t n − N A ∥ 2 ≤ N ( A △ B ) + t n 2 + ∥ N B + t n − N B ∥ 2 + ∥ N A △ B ∥ 2 ≤ 2 ∥ N A △ B ∥ 2 + ∥ N B + t n − N B ∥ 2 .
For each fixed B = B k B=B_k B = B k , the last term tends to zero as n → ∞ n\to\infty n → ∞ by the defining inequalities for the A j A_j A j . Then let k → ∞ k\to\infty k → ∞ , using ∥ N A △ B k ∥ 2 → 0 \left\|N_{A\mathbin{\triangle}B_k}\right\|_2\to0 ∥ N A △ B k ∥ 2 → 0 . It follows that ∥ N A + t n − N A ∥ 2 → 0 \left\|N_{A+t_n}-N_A\right\|_2\to0 ∥ N A + t n − N A ∥ 2 → 0 for every bounded Borel A A A . Thus ( t n ) (t_n) ( t n ) is a revival sequence and V ⊂ Rev ρ \mathcal{V}\subset\operatorname{Rev}_{\rho} V ⊂ Rev ρ . □ \square □
Near-quadratic van Hove fluctuations
Recall the van Hove condition and the notation ∂ r F \partial_r F ∂ r F from Subsection 1.4. Every van Hove sequence is Følner: for fixed t ∈ R d t \in\mathbb{R}^d t ∈ R d ,
( F + t ) △ F ⊆ ∂ ∣ t ∣ + 1 F , (F+t)\mathbin{\triangle} F \subseteq\partial_{\lvert t\rvert+1}F, ( F + t ) △ F ⊆ ∂ ∣ t ∣ + 1 F ,
so the Følner ratio is bounded by the corresponding van Hove boundary ratio.
Proposition 6.2. For every η ∈ E P P ρ , 2 \eta\in\mathrm{EPP}_{\rho,2} η ∈ EPP ρ , 2 and every van Hove sequence ( F n ) (F_n) ( F n ) ,
Var η ( N F n ) = o ( Vol d ( F n ) 2 ) . \operatorname{Var}_{\eta}(N_{F_n})=o\left(\operatorname{Vol}_d(F_n)^2\right). Var η ( N F n ) = o ( Vol d ( F n ) 2 ) .
Proof. Let φ ∈ C c ( R d ) \varphi\in C_{\mathrm{c}}(\mathbb{R}^d) φ ∈ C c ( R d ) be even and nonnegative, with ∫ φ = 1 \int\varphi=1 ∫ φ = 1 and supp φ ⊂ B r \operatorname{supp}\varphi\subset B_r supp φ ⊂ B r . For a unitary R d \mathbb{R}^d R d -representation, the normalized averages along a Følner sequence converge strongly to the invariant projection. To see this, note that the averages fix invariant vectors. On vectors H − U s H H-U_sH H − U s H they tend to zero by the Følner property; the closed span of these differences is the orthogonal complement of the invariant subspace. Since η \eta η is ergodic and ∫ ( S φ − ρ ) d η = 0 \int(S_\varphi-\rho)\,d\eta=0 ∫ ( S φ − ρ ) d η = 0 , the invariant projection of S φ − ρ S_\varphi-\rho S φ − ρ vanishes. Hence
1 Vol d ( F n ) ∫ F n ( S φ ∘ θ t − ρ ) d t ⟶ 0 in L 2 ( η ) as n → ∞ , \frac{1}{\operatorname{Vol}_d(F_n)}\int_{F_n}(S_\varphi\circ\theta_t-\rho)\,dt\longrightarrow0
\qquad\text{in }L^2(\eta)\text{ as }n\to\infty, Vol d ( F n ) 1 ∫ F n ( S φ ∘ θ t − ρ ) d t ⟶ 0 in L 2 ( η ) as n → ∞ ,
by the Følner property. By Fubini, the term on the left is Z 1 F n ∗ φ η / Vol d ( F n ) Z^\eta_{1_{F_n}*\varphi}/\operatorname{Vol}_d(F_n) Z 1 F n ∗ φ η / Vol d ( F n ) .
Define
g n : = 1 F n − 1 F n ∗ φ . g_n:=1_{F_n}-1_{F_n}*\varphi. g n := 1 F n − 1 F n ∗ φ .
We have ∣ g n ∣ ≤ 1 \lvert g_n\rvert\leq1 ∣ g n ∣ ≤ 1 , and g n g_n g n vanishes outside ∂ r F n \partial_rF_n ∂ r F n . Partition R d \mathbb{R}^d R d into half-open unit cubes C C C . If such a cube intersects ∂ r F n \partial_rF_n ∂ r F n , it lies inside ∂ r + d + 1 F n \partial_{r+\sqrt{d}+1}F_n ∂ r + d + 1 F n (up to its boundary), so the number of contributing cubes is bounded by the volume of that enlarged boundary neighbourhood. For each unit cube C C C , translation invariance and ∣ g n ∣ ≤ 1 \lvert g_n\rvert\leq1 ∣ g n ∣ ≤ 1 give
∥ Z g n 1 C η ∥ 2 ≤ ∥ N C ∥ 2 + ρ Vol d ( C ) = ∥ N Q 1 ∥ 2 + ρ . \left\|Z^\eta_{g_n1_C}\right\|_2\leq\left\|N_C\right\|_2+\rho\operatorname{Vol}_d(C)=\left\|N_{Q_1}\right\|_2+\rho. Z g n 1 C η 2 ≤ ∥ N C ∥ 2 + ρ Vol d ( C ) = ∥ N Q 1 ∥ 2 + ρ .
Apply Minkowski’s inequality to the contributing unit cubes and use the uniform L 2 L^2 L 2 bound above. There is a constant C η < ∞ C_\eta<\infty C η < ∞ such that
∥ Z g n η ∥ L 2 ( η ) ≤ C η Vol d ( ∂ r + d + 1 F n ) . \left\|Z^\eta_{g_n}\right\|_{L^2(\eta)}\leq C_\eta\operatorname{Vol}_d\left(\partial_{r+\sqrt{d}+1}F_n\right). Z g n η L 2 ( η ) ≤ C η Vol d ( ∂ r + d + 1 F n ) .
Since Vol d ( ∂ r + d + 1 F n ) / Vol d ( F n ) → 0 \operatorname{Vol}_d(\partial_{r+\sqrt{d}+1}F_n)/\operatorname{Vol}_d(F_n)\to0 Vol d ( ∂ r + d + 1 F n ) / Vol d ( F n ) → 0 ,
∥ Z g n η ∥ L 2 ( η ) Vol d ( F n ) ⟶ 0 ( n → ∞ ) . \frac{\left\|Z^\eta_{g_n}\right\|_{L^2(\eta)}}{\operatorname{Vol}_d(F_n)}\longrightarrow0
\qquad(n\to\infty). Vol d ( F n ) Z g n η L 2 ( η ) ⟶ 0 ( n → ∞ ) .
Since
N F n − ρ Vol d ( F n ) = Z 1 F n ∗ φ η + Z g n η , N_{F_n}-\rho\operatorname{Vol}_d(F_n)=Z^\eta_{1_{F_n}*\varphi}+Z^\eta_{g_n}, N F n − ρ Vol d ( F n ) = Z 1 F n ∗ φ η + Z g n η ,
the centered count divided by Vol d ( F n ) \operatorname{Vol}_d(F_n) Vol d ( F n ) converges to zero in L 2 ( η ) L^2(\eta) L 2 ( η ) . Taking squared norms proves the claim. □
Remark 6.3 (Følner versus van Hove sequences). The conclusion of Proposition 6.2 fails if “van Hove” is replaced by “Følner”. First take intensity 1 1 1 and let
∫ F d η : = ∫ [ 0 , 1 ) d F ( δ Z d + u ) d u ( F bounded Borel ) . \int F\,d\eta:=\int_{[0,1)^d}F(\delta_{\mathbb{Z}^d+u})\,du
\qquad(F\text{ bounded Borel}). ∫ F d η := ∫ [ 0 , 1 ) d F ( δ Z d + u ) d u ( F bounded Borel ) .
Let U U U denote the coordinate on [ 0 , 1 ) d [0,1)^d [ 0 , 1 ) d with Lebesgue measure. For ε n : = n − 3 / 2 \varepsilon_n:=n^{-3/2} ε n := n − 3/2 and K n : = n 2 d K_n:=n^{2d} K n := n 2 d , define
F n : = [ 0 , n ) d ∪ ⋃ i = 1 K n ( ( n + i ) e 1 + [ 0 , ε n ) d ) . F_n:=[0,n)^d\cup\bigcup_{i=1}^{K_n}\left((n+i)e_1+[0,\varepsilon_n)^d\right). F n := [ 0 , n ) d ∪ i = 1 ⋃ K n ( ( n + i ) e 1 + [ 0 , ε n ) d ) .
The microscopic cubes have total volume K n ε n d = n d / 2 K_n\varepsilon_n^d=n^{d/2} K n ε n d = n d /2 , so
Vol d ( F n ) = n d + n d / 2 ∼ n d . \operatorname{Vol}_d(F_n)=n^d+n^{d/2}\sim n^d. Vol d ( F n ) = n d + n d /2 ∼ n d .
For every fixed t ∈ R d t\in\mathbb{R}^d t ∈ R d ,
Vol d ( ( F n + t ) △ F n ) ≤ C d , t n d − 1 + 2 n d / 2 = o ( Vol d ( F n ) ) , \operatorname{Vol}_d((F_n+t)\mathbin{\triangle}F_n)\leq C_{d,t}n^{d-1}+2n^{d/2}=o(\operatorname{Vol}_d(F_n)), Vol d (( F n + t ) △ F n ) ≤ C d , t n d − 1 + 2 n d /2 = o ( Vol d ( F n )) ,
and hence ( F n ) (F_n) ( F n ) is Følner. Every microscopic cube contains a point of Z d + U \mathbb{Z}^d+U Z d + U if and only if U ∈ [ 0 , ε n ) d U\in[0,\varepsilon_n)^d U ∈ [ 0 , ε n ) d , simultaneously for all i i i , while [ 0 , n ) d [0,n)^d [ 0 , n ) d contains n d n^d n d points almost surely. Thus
N F n = n d + K n 1 { U ∈ [ 0 , ε n ) d } , N_{F_n}=n^d+K_n\mathbf{1}_{\{U\in[0,\varepsilon_n)^d\}}, N F n = n d + K n 1 { U ∈ [ 0 , ε n ) d } ,
and therefore
Var η ( N F n ) = K n 2 ε n d ( 1 − ε n d ) ∼ n 5 d / 2 . \operatorname{Var}_{\eta}(N_{F_n})=K_n^2\varepsilon_n^d(1-\varepsilon_n^d)\sim n^{5d/2}. Var η ( N F n ) = K n 2 ε n d ( 1 − ε n d ) ∼ n 5 d /2 .
Consequently
Var η ( N F n ) Vol d ( F n ) 2 ∼ n d / 2 ⟶ ∞ ( n → ∞ ) . \frac{\operatorname{Var}_{\eta}(N_{F_n})}{\operatorname{Vol}_d(F_n)^2}\sim n^{d/2}\longrightarrow\infty\qquad(n\to\infty). Vol d ( F n ) 2 Var η ( N F n ) ∼ n d /2 ⟶ ∞ ( n → ∞ ) .
The divergence comes from an event of measure ε n d = n − 3 d / 2 \varepsilon_n^d=n^{-3d/2} ε n d = n − 3 d /2 on which the microscopic cubes contribute K n = n 2 d K_n=n^{2d} K n = n 2 d points; their contribution to the variance is of order K n 2 ε n d = n 5 d / 2 K_n^2\varepsilon_n^d=n^{5d/2} K n 2 ε n d = n 5 d /2 . Dilating both the lattice and the sets by ρ − 1 / d \rho^{-1/d} ρ − 1/ d gives the same conclusion at any prescribed intensity ρ > 0 \rho>0 ρ > 0 .
For r = 1 / 4 r=1/4 r = 1/4 , the boundary neighbourhoods of the microscopic cubes have total volume of order n 2 d n^{2d} n 2 d , whereas Vol d ( F n ) ∼ n d \operatorname{Vol}_d(F_n)\sim n^d Vol d ( F n ) ∼ n d . Hence ( F n ) (F_n) ( F n ) is not van Hove.
Lemma 6.4. Let η ∈ E P P ρ , 2 \eta\in\mathrm{EPP}_{\rho,2} η ∈ EPP ρ , 2 and let A ⊂ R d A\subset\mathbb{R}^d A ⊂ R d be bounded Borel. Then, for s , t ∈ R d s,t\in\mathbb{R}^d s , t ∈ R d ,
Cov η ( N A + s , N A + t ) = Var η ( N A ) − 1 2 ∥ N A + s − N A + t ∥ L 2 ( η ) 2 . \operatorname{Cov}_{\eta}(N_{A+s},N_{A+t})=\operatorname{Var}_{\eta}(N_A)-\frac{1}{2}\lVert N_{A+s}-N_{A+t}\rVert_{L^2(\eta)}^2. Cov η ( N A + s , N A + t ) = Var η ( N A ) − 2 1 ∥ N A + s − N A + t ∥ L 2 ( η ) 2 . Consequently, if A + s 1 , … , A + s m A+s_1,\ldots,A+s_m A + s 1 , … , A + s m are pairwise disjoint and
∥ N A + s i − N A ∥ L 2 ( η ) ≤ δ ( 1 ≤ i ≤ m ) , \lVert N_{A+s_i}-N_A\rVert_{L^2(\eta)}\leq\delta\qquad(1\leq i\leq m), ∥ N A + s i − N A ∥ L 2 ( η ) ≤ δ ( 1 ≤ i ≤ m ) , then, with σ A : = Var η ( N A ) \sigma_A:=\sqrt{\operatorname{Var}_{\eta}(N_A)} σ A := Var η ( N A ) ,
Var η ( N ⋃ i = 1 m ( A + s i ) ) ≥ m ( σ A − δ ) . \sqrt{\operatorname{Var}_{\eta}\left(N_{\bigcup_{i=1}^{m}(A+s_i)}\right)}\geq m(\sigma_A-\delta). Var η ( N ⋃ i = 1 m ( A + s i ) ) ≥ m ( σ A − δ ) .
Proof. The random variables N A + s N_{A+s} N A + s and N A + t N_{A+t} N A + t have the same mean and variance as N A N_A N A . Expanding ∥ N A + s − N A + t ∥ 2 2 \lVert N_{A+s}-N_{A+t}\rVert_{2}^{2} ∥ N A + s − N A + t ∥ 2 2 gives the covariance identity. For the second assertion, write Y i : = N A + s i − ρ Vol d ( A ) Y_i:=N_{A+s_i}-\rho\operatorname{Vol}_d(A) Y i := N A + s i − ρ Vol d ( A ) and Y 0 : = N A − ρ Vol d ( A ) Y_0:=N_A-\rho\operatorname{Vol}_d(A) Y 0 := N A − ρ Vol d ( A ) . By the reverse triangle inequality,
∥ ∑ i = 1 m Y i ∥ 2 ≥ m ∥ Y 0 ∥ 2 − ∑ i = 1 m ∥ Y i − Y 0 ∥ 2 ≥ m ( σ A − δ ) . \left\lVert\sum_{i=1}^{m}Y_i\right\rVert_2\geq m\lVert Y_0\rVert_2-\sum_{i=1}^{m}\lVert Y_i-Y_0\rVert_2\geq m(\sigma_A-\delta). i = 1 ∑ m Y i 2 ≥ m ∥ Y 0 ∥ 2 − i = 1 ∑ m ∥ Y i − Y 0 ∥ 2 ≥ m ( σ A − δ ) .
Disjointness identifies ∑ i Y i \sum_iY_i ∑ i Y i with the centered count of the union. □
Proposition 6.5. Let η ∈ R e v ρ \eta\in\mathrm{Rev}_{\rho} η ∈ Rev ρ , let α n > 0 \alpha_n>0 α n > 0 with α n → 0 \alpha_n\to0 α n → 0 , and let M n > 0 M_n>0 M n > 0 with M n → ∞ M_n\to\infty M n → ∞ . Then there is a van Hove sequence ( F n ) (F_n) ( F n ) , depending on η \eta η , ( α n ) (\alpha_n) ( α n ) , and ( M n ) (M_n) ( M n ) , such that
Var η ( N F n ) ≥ M n Vol d ( F n ) 2 − α n ( n ≥ 1 ) . \operatorname{Var}_{\eta}(N_{F_n})\geq M_n\operatorname{Vol}_d(F_n)^{2-\alpha_n}\qquad(n\geq1). Var η ( N F n ) ≥ M n Vol d ( F n ) 2 − α n ( n ≥ 1 ) . Consequently R e v ρ ⊆ M F V H ρ \mathrm{Rev}_{\rho}\subseteq\mathrm{MFVH}_{\rho} Rev ρ ⊆ MFVH ρ .
Proof. Fix η ∈ Rev ρ \eta\in\operatorname{Rev}_{\rho} η ∈ Rev ρ and a revival sequence ( t k ) (t_k) ( t k ) . Choose A A A from the ring of finite unions of bounded rational rectangles such that
σ 2 : = Var η ( N A ) > 0. \sigma^2 := \operatorname{Var}_{\eta}(N_A)>0. σ 2 := Var η ( N A ) > 0.
Indeed, if the counts on this ring were all deterministic, η \eta η would be a Dirac mass. Translation invariance would then make its configuration translation-invariant, which is impossible for a locally finite counting measure of positive intensity.
Sets with variance bounded away from zero. Choose R n → ∞ R_n \to\infty R n → ∞ . For each n n n , ergodicity and the mean ergodic theorem give
∥ 1 Vol d ( Q T ) ∫ Q T + v ( N A + s − ρ Vol d ( A ) ) d s ∥ 2 ⟶ 0 ( T → ∞ ) , \left\Vert\frac{1}{\operatorname{Vol}_{d}(Q_T)}\int_{Q_T+v}(N_{A+s}-\rho\operatorname{Vol}_{d}(A))\,\mathrm{d}s \right\Vert_2 \longrightarrow0 \qquad(T\to\infty), Vol d ( Q T ) 1 ∫ Q T + v ( N A + s − ρ Vol d ( A )) d s 2 ⟶ 0 ( T → ∞ ) ,
uniformly in v v v : the averages over Q T + v Q_T+v Q T + v are unitary translates of the average over Q T Q_T Q T . Taking the inner product with N Q R n − ρ Vol d ( Q R n ) N_{Q_{R_n}}-\rho\operatorname{Vol}_{d}(Q_{R_n}) N Q R n − ρ Vol d ( Q R n ) shows that
1 Vol d ( Q T ) ∫ Q T + v Cov η ( N Q R n , N A + s ) d s ⟶ 0 ( T → ∞ ) \frac{1}{\operatorname{Vol}_{d}(Q_T)}\int_{Q_T+v}\operatorname{Cov}_{\eta}(N_{Q_{R_n}},N_{A+s})\,\mathrm{d}s \longrightarrow0 \qquad(T\to\infty) Vol d ( Q T ) 1 ∫ Q T + v Cov η ( N Q R n , N A + s ) d s ⟶ 0 ( T → ∞ )
uniformly in v v v . Choose T T T so that this average is greater than − σ 2 / 4 -\sigma^2/4 − σ 2 /4 . Then choose v v v so that every A + s A+s A + s , s ∈ Q T + v s\in Q_T+v s ∈ Q T + v , is disjoint from Q R n Q_{R_n} Q R n . Some s n ∈ Q T + v s_n\in Q_T+v s n ∈ Q T + v satisfies
( A + s n ) ∩ Q R n = ∅ , Cov η ( N Q R n , N A + s n ) ≥ − σ 2 4 . (A+s_n)\cap Q_{R_n}=\varnothing,\qquad\operatorname{Cov}_{\eta}(N_{Q_{R_n}},N_{A+s_n})\geq-\frac{\sigma^2}{4}. ( A + s n ) ∩ Q R n = ∅ , Cov η ( N Q R n , N A + s n ) ≥ − 4 σ 2 .
Set
E n : = Q R n ∪ ( A + s n ) . E_n:=Q_{R_n}\cup(A+s_n). E n := Q R n ∪ ( A + s n ) .
Then
Var η ( N E n ) ≥ σ 2 + 2 Cov η ( N Q R n , N A + s n ) ≥ σ 2 2 . \operatorname{Var}_{\eta}(N_{E_n})\geq\sigma^2+2\operatorname{Cov}_{\eta}(N_{Q_{R_n}},N_{A+s_n})\geq\frac{\sigma^2}{2}. Var η ( N E n ) ≥ σ 2 + 2 Cov η ( N Q R n , N A + s n ) ≥ 2 σ 2 .
For every fixed r > 0 r>0 r > 0 ,
Vol d ( ∂ r E n ) ≤ Vol d ( ∂ r Q R n ) + Vol d ( ∂ r A ) , \operatorname{Vol}_{d}(\partial_r E_n)\leq\operatorname{Vol}_{d}(\partial_r Q_{R_n})+\operatorname{Vol}_{d}(\partial_r A), Vol d ( ∂ r E n ) ≤ Vol d ( ∂ r Q R n ) + Vol d ( ∂ r A ) ,
where the second term is independent of the translation s n s_n s n and finite because A A A is bounded. The right-hand side is o ( Vol d ( Q R n ) ) o(\operatorname{Vol}_{d}(Q_{R_n})) o ( Vol d ( Q R n )) , while Vol d ( E n ) ≥ Vol d ( Q R n ) \operatorname{Vol}_{d}(E_n)\geq\operatorname{Vol}_{d}(Q_{R_n}) Vol d ( E n ) ≥ Vol d ( Q R n ) .
Variance amplification. Since Var η ( N E n ) > 0 \operatorname{Var}_{\eta}(N_{E_n})>0 Var η ( N E n ) > 0 and α n > 0 \alpha_n>0 α n > 0 , choose an integer m n m_n m n large enough that
m n α n ≥ 4 M n Vol d ( E n ) 2 − α n Var η ( N E n ) . m_n^{\alpha_n}\geq\frac{4M_n\operatorname{Vol}_{d}(E_n)^{2-\alpha_n}}{\operatorname{Var}_{\eta}(N_{E_n})}. m n α n ≥ Var η ( N E n ) 4 M n Vol d ( E n ) 2 − α n .
or equivalently,
m n 2 4 Var η ( N E n ) ≥ M n ( m n Vol d ( E n ) ) 2 − α n . \frac{m_n^2}{4}\operatorname{Var}_{\eta}(N_{E_n})\geq M_n\bigl(m_n\operatorname{Vol}_{d}(E_n)\bigr)^{2-\alpha_n}. 4 m n 2 Var η ( N E n ) ≥ M n ( m n Vol d ( E n ) ) 2 − α n .
Since ∂ E n ⊆ ∂ Q R n ∪ ( ∂ A + s n ) \partial E_n\subseteq\partial Q_{R_n}\cup(\partial A+s_n) ∂ E n ⊆ ∂ Q R n ∪ ( ∂ A + s n ) and both sets have zero Lebesgue measure, the revival property gives, for all sufficiently large k k k ,
∥ N E n + t k − N E n ∥ L 2 ( η ) ≤ 1 2 Var η ( N E n ) 1 / 2 . \left\Vert N_{E_n+t_k}-N_{E_n}\right\Vert_{L^2(\eta)}\leq\frac{1}{2}\operatorname{Var}_{\eta}(N_{E_n})^{1/2}. ∥ N E n + t k − N E n ∥ L 2 ( η ) ≤ 2 1 Var η ( N E n ) 1/2 .
Since ∣ t k ∣ → ∞ |t_k|\to\infty ∣ t k ∣ → ∞ , we may choose indices k 1 < ⋯ < k m n k_1<\cdots<k_{m_n} k 1 < ⋯ < k m n , each satisfying the preceding L 2 L^2 L 2 bound, with r n , i : = t k i r_{n,i}:=t_{k_i} r n , i := t k i and
∣ r n , i − r n , j ∣ > diam ( E n ) ( i ≠ j ) . |r_{n,i}-r_{n,j}|>\operatorname{diam}(E_n)\qquad(i\ne j). ∣ r n , i − r n , j ∣ > diam ( E n ) ( i = j ) .
Then the translates E n + r n , i E_n+r_{n,i} E n + r n , i are pairwise disjoint. Set
F n : = ⋃ i = 1 m n ( E n + r n , i ) . F_n:=\bigcup_{i=1}^{m_n}(E_n+r_{n,i}). F n := i = 1 ⋃ m n ( E n + r n , i ) .
By Lemma 6.4 ,
Var η ( N F n ) ≥ m n 2 4 Var η ( N E n ) ≥ M n Vol d ( F n ) 2 − α n . \operatorname{Var}_{\eta}(N_{F_n}) \ge\frac{m_n^2}{4}\operatorname{Var}_{\eta}(N_{E_n}) \ge M_n\operatorname{Vol}_{d}(F_n)2^{-\alpha_n}. Var η ( N F n ) ≥ 4 m n 2 Var η ( N E n ) ≥ M n Vol d ( F n ) 2 − α n .
The sequence ( F n ) (F_n) ( F n ) is van Hove: for every fixed r > 0 r > 0 r > 0 ,
∂ r F n ⊆ ⋃ i = 1 m n ( ∂ r E n + r n , i ) , \partial_r F_n \subseteq\bigcup_{i=1}^{m_n}(\partial_r E_n+r_{n,i}), ∂ r F n ⊆ i = 1 ⋃ m n ( ∂ r E n + r n , i ) ,
so
Vol d ( ∂ r F n ) Vol d ( F n ) ≤ Vol d ( ∂ r E n ) Vol d ( E n ) ⟶ 0 ( n → ∞ ) . \frac{\operatorname{Vol}_{d}(\partial_r F_n)}{\operatorname{Vol}_{d}(F_n)}
\le
\frac{\operatorname{Vol}_{d}(\partial_r E_n)}{\operatorname{Vol}_{d}(E_n)}
\longrightarrow0
\qquad(n \to\infty). Vol d ( F n ) Vol d ( ∂ r F n ) ≤ Vol d ( E n ) Vol d ( ∂ r E n ) ⟶ 0 ( n → ∞ ) .
Thus η ∈ M F V H ρ \eta\in\mathrm{MFVH}_{\rho} η ∈ MFVH ρ .
Corollary 6.6. The set M F V H ρ \mathrm{MFVH}_{\rho} MFVH ρ is residual in E P P ρ , 2 \mathrm{EPP}_{\rho,2} EPP ρ , 2 .
Proof. By Propositions 6.1 and 6.5 , the residual set R e v ρ \mathrm{Rev}_{\rho} Rev ρ is contained in M F V H ρ \mathrm{MFVH}_{\rho} MFVH ρ .
Rigidity sequences for actions
Let Aut ( X , μ ) \operatorname{Aut}(X,\mu) Aut ( X , μ ) be the group of measure-preserving automorphisms modulo null sets, equipped with its weak topology. This is a Polish group; see [25 Sec. 1] . For an action a : R d × X ⟶ X a:\mathbb{R}^{d}\times X\longrightarrow X a : R d × X ⟶ X , let a t ( x ) : = a ( t , x ) a_t(x):=a(t,x) a t ( x ) := a ( t , x ) . Identifying a a a with the continuous homomorphism t ↦ a t t\mapsto a_t t ↦ a t into Aut ( X , μ ) \operatorname{Aut}(X,\mu) Aut ( X , μ ) , equip A d ( X , μ ) \mathcal{A}_d(X,\mu) A d ( X , μ ) with the topology of uniform convergence on compact subsets of R d \mathbb{R}^{d} R d .
For a ∈ A d ( X , μ ) a\in\mathcal{A}_d(X,\mu) a ∈ A d ( X , μ ) , let U t a U_t^a U t a denote its Koopman operators on L 2 ( X , μ ) L^2(X,\mu) L 2 ( X , μ ) . The action a a a is weakly mixing if the diagonal action on ( X × X , μ × μ ) (X\times X,\mu\times\mu) ( X × X , μ × μ ) is ergodic. A sequence ( t n ) ⊂ R d (t_n)\subset\mathbb{R}^{d} ( t n ) ⊂ R d is a rigidity sequence for a a a if
∣ t n ∣ ⟶ ∞ and U t n a ⟶ I strongly on L 2 ( X , μ ) . \lvert t_n\rvert\longrightarrow\infty
\qquad\text{and}\qquad
U_{t_n}^a\longrightarrow I
\quad\text{strongly on }L^2(X,\mu). ∣ t n ∣ ⟶ ∞ and U t n a ⟶ I strongly on L 2 ( X , μ ) .
Proposition A.1. The set of weakly mixing actions in A d ( X , μ ) \mathcal{A}_d(X,\mu) A d ( X , μ ) which admit a rigidity sequence is a dense G δ G_\delta G δ -subset of A d ( X , μ ) \mathcal{A}_d(X,\mu) A d ( X , μ ) .
For a ∈ A d ( X , μ ) a\in\mathcal{A}_d(X,\mu) a ∈ A d ( X , μ ) and a bounded half-open rectangular box Q ⊂ R d Q\subset\mathbb{R}^{d} Q ⊂ R d , a Q Q Q -tower consists of a Borel set B ⊂ X B\subset X B ⊂ X and a finite measure ν B \nu_B ν B on B B B for which
Φ : Q × B ⟶ X , ( t , y ) ⟼ a t y \Phi:Q\times B\longrightarrow X,\qquad(t,y)\longmapsto a_t y Φ : Q × B ⟶ X , ( t , y ) ⟼ a t y
is injective and Borel, and μ ∣ Φ ( Q × B ) = Φ ∗ ( Vol d ∣ Q ⊗ ν B ) \mu|_{\Phi(Q\times B)}=\Phi_*(\operatorname{Vol}_d|_Q\otimes\nu_B) μ ∣ Φ ( Q × B ) = Φ ∗ ( Vol d ∣ Q ⊗ ν B ) . The measure ν B \nu_B ν B is the transverse measure on the base B B B ; it is not the restriction of μ \mu μ to B B B . The remainder is X ∖ Φ ( Q × B ) X\setminus\Phi(Q\times B) X ∖ Φ ( Q × B ) . The rectangular tower theorem of [34 Theorem 1 and Corollary 1, pp. 177–178] gives the following for essentially free measure-preserving actions: for every ϵ > 0 \epsilon>0 ϵ > 0 there exists L 0 > 0 L_0>0 L 0 > 0 such that, whenever Q = ∏ i = 1 d [ 0 , L i ) Q=\prod_{i=1}^{d}[0,L_i) Q = ∏ i = 1 d [ 0 , L i ) with L i ≥ L 0 L_i\ge L_0 L i ≥ L 0 for all i i i , there is a Q Q Q -tower satisfying
μ ( X ∖ Φ ( Q × B ) ) < ϵ . \mu(X\setminus\Phi(Q\times B))<\epsilon. μ ( X ∖ Φ ( Q × B )) < ϵ .
Its transverse base measure is nonatomic. Indeed, an atom of ν B \nu_B ν B would give positive μ \mu μ -measure to a bounded part of one orbit, contradicting finiteness and translation invariance of μ \mu μ .
Lemma A.2 (Density of conjugacy classes). Every essentially free action in A d ( X , μ ) \mathcal{A}_d(X,\mu) A d ( X , μ ) has a dense conjugacy class.
Proof. Fix two essentially free actions a , b a,b a , b . We show that a conjugate of b b b approximates a a a . Fix compact K ⊂ R d K\subset\mathbb{R}^{d} K ⊂ R d , measurable A 1 , … , A m ⊂ X A_1,\ldots,A_m\subset X A 1 , … , A m ⊂ X , and ϵ > 0 \epsilon>0 ϵ > 0 . Choose a half-open rectangular box Q Q Q with
Vol d { t ∈ Q : t + K ⊄ Q } < ϵ 8 Vol d ( Q ) . (46) \operatorname{Vol}_{d}\{t\in Q:t+K\not\subset Q\}<\frac{\epsilon}{8}\operatorname{Vol}_{d}(Q). \tag*{(46)} Vol d { t ∈ Q : t + K ⊂ Q } < 8 ϵ Vol d ( Q ) . (46)
Take Q Q Q -towers for a a a and b b b , with remainders of measure less than ε / 32 \varepsilon/32 ε /32 each, and bases B a , B b B_a,B_b B a , B b . Their transverse base measures are nonatomic. Restrict the larger base so that both have the same transverse measure, then choose a measure-preserving isomorphism J : B b → B a J:B_b\to B_a J : B b → B a . The two restricted towers have the same measure; their combined remainders have measure less than ε / 8 \varepsilon/8 ε /8 . Match these remainders to extend
T ( b t y ) : = a t ( J ( y ) ) , y ∈ B b , t ∈ Q , T(b_t y):=a_t(J(y)),\qquad y\in B_b,\quad t\in Q, T ( b t y ) := a t ( J ( y )) , y ∈ B b , t ∈ Q ,
to a measure-space automorphism of X X X . For c s : = T b s T − 1 c_s:=Tb_sT^{-1} c s := T b s T − 1 and t , t + s ∈ Q t,t+s\in Q t , t + s ∈ Q ,
c s ( a t x ) = a t + s x = a s ( a t x ) , x ∈ B a . c_s(a_t x)=a_{t+s}x=a_s(a_t x),\qquad x\in B_a. c s ( a t x ) = a t + s x = a s ( a t x ) , x ∈ B a .
Uniformly for s ∈ K s\in K s ∈ K , the set on which c s c_s c s and a s a_s a s differ has measure at most the remainder measure plus Vol d ( { t ∈ Q : t + K ⊄ Q } ) ν B a ( B a ) < ε / 4 \operatorname{Vol}_d(\{t\in Q:t+K\not\subset Q\})\nu_{B_a}(B_a)<\varepsilon/4 Vol d ({ t ∈ Q : t + K ⊂ Q }) ν B a ( B a ) < ε /4 . Hence
sup s ∈ K μ ( c s A i △ a s A i ) < ε ( 1 ≤ i ≤ m ) . \sup_{s\in K}\mu(c_sA_i\mathbin{\triangle}a_sA_i)<\varepsilon\qquad(1\le i\le m). s ∈ K sup μ ( c s A i △ a s A i ) < ε ( 1 ≤ i ≤ m ) .
Thus T b T − 1 TbT^{-1} T b T − 1 belongs to the prescribed neighbourhood of a a a .
Proof. Let ( f j ) j ≥ 1 ⊂ L 0 2 ( X , μ ) (f_j)_{j\ge1}\subset L^2_0(X,\mu) ( f j ) j ≥ 1 ⊂ L 0 2 ( X , μ ) be a countable dense family. Weak mixing is a G δ G_\delta G δ condition. Indeed, the mean ergodic theorem applied to the diagonal action (equivalently, to the tensor-square representation) shows that a a a is weakly mixing precisely when, for every i , j , k , ℓ ≥ 1 i,j,k,\ell\ge1 i , j , k , ℓ ≥ 1 , there is an integer R ≥ ℓ R\ge\ell R ≥ ℓ such that
1 Vol d ( Q R ) ∫ Q R ∣ ⟨ U t a f i , f j ⟩ ∣ 2 d t < 1 k . \frac{1}{\operatorname{Vol}_d(Q_R)}\int_{Q_R}\left|\langle U_t^a f_i,f_j\rangle\right|^2\,dt<\frac{1}{k}. Vol d ( Q R ) 1 ∫ Q R ∣ ⟨ U t a f i , f j ⟩ ∣ 2 d t < k 1 .
For fixed R R R , i i i , j j j , compact-open convergence of actions gives uniform convergence of the coefficients ⟨ U t a f i , f j ⟩ \langle U_t^a f_i,f_j\rangle ⟨ U t a f i , f j ⟩ for t ∈ Q R t\in Q_R t ∈ Q R . Their averaged squared moduli therefore depend continuously on a a a .
Choose a sequence ( h j ) j ≥ 1 (h_j)_{j\ge1} ( h j ) j ≥ 1 dense in the unit ball of L 2 ( X , μ ) L^2(X,\mu) L 2 ( X , μ ) . The existence of a rigidity sequence is equivalent to
∀ m , ℓ ≥ 1 ∃ t ∈ R d , ∣ t ∣ > ℓ : max 1 ≤ j ≤ m ∥ U t a h j − h j ∥ 2 < 1 m . \forall m,\ell\ge1\quad\exists t\in\mathbb{R}^d,\ |t|>\ell:\quad\max_{1\le j\le m}\|U_t^a h_j-h_j\|_2<\frac{1}{m}. ∀ m , ℓ ≥ 1 ∃ t ∈ R d , ∣ t ∣ > ℓ : 1 ≤ j ≤ m max ∥ U t a h j − h j ∥ 2 < m 1 .
For each fixed t t t , the finitely many inequalities define an open set in the action topology. Taking the union over ∣ t ∣ > ℓ |t|>\ell ∣ t ∣ > ℓ gives an open set. Conversely, applying it with m = ℓ = n m=\ell=n m = ℓ = n and selecting one t n t_n t n for each n n n gives ∣ t n ∣ → ∞ |t_n|\to\infty ∣ t n ∣ → ∞ and strong convergence to the identity on a dense family, hence on all of L 2 ( X , μ ) L^2(X,\mu) L 2 ( X , μ ) by unitarity.
Danilenko and Ryzhikov construct a rank-one weakly mixing rigid flow ( V t ) t ∈ R (V_t)_{t\in\mathbb{R}} ( V t ) t ∈ R with a rigidity sequence s n s_n s n [10 DR13] . Such a flow is essentially free: by ergodicity its stabilizer is almost surely a fixed closed subgroup of R \mathbb{R} R ; nonatomicity excludes R \mathbb{R} R , while a stabilizer c Z c\mathbb{Z} c Z , c > 0 c>0 c > 0 , would make the flow factor through R / c Z \mathbb{R}/c\mathbb{Z} R / c Z and hence give a nonconstant eigenfunction, contradicting weak mixing. The coordinatewise action
( t 1 , … , t d ) ⋅ ( x 1 , … , x d ) : = ( V t 1 x 1 , … , V t d x d ) (t_1,\ldots,t_d)\mathbin{\cdot}(x_1,\ldots,x_d):=(V_{t_1}x_1,\ldots,V_{t_d}x_d) ( t 1 , … , t d ) ⋅ ( x 1 , … , x d ) := ( V t 1 x 1 , … , V t d x d )
on the d d d -fold product is therefore essentially free and weakly mixing, and ( s n , … , s n ) (s_n,\ldots,s_n) ( s n , … , s n ) is a rigidity sequence because V s n → I V_{s_n}\to I V s n → I strongly. Transporting the product action to ( X , μ ) (X,\mu) ( X , μ ) gives an element of A d ( X , μ ) A_d(X,\mu) A d ( X , μ ) with both properties. Both properties are invariant under conjugacy. By Lemma A.2 , their intersection is dense.
Proposition A.3. Let a ∈ A d ( X , μ ) a\in A_d(X,\mu) a ∈ A d ( X , μ ) and κ ∈ R ρ , 2 a ( X , μ ) \kappa\in R^a_{\rho,2}(X,\mu) κ ∈ R ρ , 2 a ( X , μ ) . If a a a is weakly mixing, then the point process κ ∗ μ \kappa_{*}\mu κ ∗ μ is weakly mixing. If ( t n ) (t_n) ( t n ) is a rigidity sequence for a a a , then ( t n ) (t_n) ( t n ) is a revival sequence for κ ∗ μ \kappa_{*}\mu κ ∗ μ .
Proof. With η : = κ ∗ μ \eta:= \kappa_*\mu η := κ ∗ μ , weak mixing passes to factors. If ( t n ) (t_n) ( t n ) is a rigidity sequence for a a a , fix a bounded Borel continuity set A A A . The function N A ∘ κ N_A \circ\kappa N A ∘ κ belongs to L 2 ( X , μ ) L^2(X,\mu) L 2 ( X , μ ) , and equivariance identifies its Koopman translates with the translated window counts. Thus
∥ N A + t n − N A ∥ L 2 ( η ) = ∥ N A + t n ∘ κ − N A ∘ κ ∥ L 2 ( μ ) ⟶ 0 ( n → ∞ ) , \lVert N_{A+t_n}-N_A\rVert_{L^2(\eta)}=\lVert N_{A+t_n}\circ\kappa-N_A\circ\kappa\rVert_{L^2(\mu)}\longrightarrow0 \qquad(n\to\infty), ∥ N A + t n − N A ∥ L 2 ( η ) = ∥ N A + t n ∘ κ − N A ∘ κ ∥ L 2 ( μ ) ⟶ 0 ( n → ∞ ) ,
by strong convergence of U t n a U_{t_n}^a U t n a to the identity. □
Proposition A.4 (Revival and Koopman rigidity). Let η ∈ E P P ρ , 2 \eta\in\mathrm{EPP}_{\rho,2} η ∈ EPP ρ , 2 and let ( t n ) ⊂ R d (t_n)\subset\mathbb{R}^d ( t n ) ⊂ R d satisfy ∣ t n ∣ → ∞ |t_n|\to\infty ∣ t n ∣ → ∞ . Then ( t n ) (t_n) ( t n ) is a revival sequence for η \eta η if and only if
F ∘ θ t n ⟶ F in L 2 ( η ) ( F ∈ L 2 ( η ) ) . F\circ\theta_{t_n}\longrightarrow F \quad\text{in }L^2(\eta)\qquad(F\in L^2(\eta)). F ∘ θ t n ⟶ F in L 2 ( η ) ( F ∈ L 2 ( η )) .
Proof. Strong Koopman convergence implies revival by applying it to N A N_A N A for each bounded Borel set A A A with Vol d ( ∂ A ) = 0 \operatorname{Vol}_d(\partial A)=0 Vol d ( ∂ A ) = 0 . Conversely, suppose ( t n ) (t_n) ( t n ) is a revival sequence. Let Q \mathcal{Q} Q be the countable family of bounded half-open rectangles with rational endpoints. For each Q ∈ Q Q\in\mathcal{Q} Q ∈ Q ,
∥ N Q ∘ θ t n − N Q ∥ 2 = ∥ N Q + t n − N Q ∥ 2 ⟶ 0 ( n → ∞ ) . \lVert N_Q\circ\theta_{t_n}-N_Q\rVert_2=\lVert N_{Q+t_n}-N_Q\rVert_2\longrightarrow0 \qquad(n\to\infty). ∥ N Q ∘ θ t n − N Q ∥ 2 = ∥ N Q + t n − N Q ∥ 2 ⟶ 0 ( n → ∞ ) .
These counts are integer-valued. Hence, for every finite collection Q 1 , … , Q m ∈ Q Q_1,\ldots,Q_m\in\mathcal{Q} Q 1 , … , Q m ∈ Q , the probability that any of their counts changes under θ t n \theta_{t_n} θ t n tends to zero. Every bounded function of ( N Q 1 , … , N Q m ) (N_{Q_1},\ldots,N_{Q_m}) ( N Q 1 , … , N Q m ) therefore converges to its Koopman translate in L 2 ( η ) L^2(\eta) L 2 ( η ) . Such functions are dense in L 2 ( η ) L^2(\eta) L 2 ( η ) because the counts on Q \mathcal{Q} Q generate the Borel σ \sigma σ -algebra of Conf ( R d ) \operatorname{Conf}(\mathbb{R}^d) Conf ( R d ) . The Koopman operators are unitary, so convergence extends to every F ∈ L 2 ( η ) F\in L^2(\eta) F ∈ L 2 ( η ) . □
Proposition A.5 (Rigidity sequences and linear maximal rigidity). Let η ∈ P P ρ , 2 \eta\in\mathrm{PP}_{\rho,2} η ∈ PP ρ , 2 . Suppose that there are t n ∈ R d t_n\in\mathbb{R}^d t n ∈ R d with ∣ t n ∣ → ∞ |t_n|\to\infty ∣ t n ∣ → ∞ such that
F ∘ θ t n ⟶ F in L 2 ( η ) ( F ∈ L 2 ( η ) ) . F\circ\theta_{t_n}\longrightarrow F \quad\text{in }L^2(\eta)\qquad(F\in L^2(\eta)). F ∘ θ t n ⟶ F in L 2 ( η ) ( F ∈ L 2 ( η )) . Then η ∈ L M R i g ρ \eta\in\mathrm{LMRig}_\rho η ∈ LMRig ρ .
Proof. Fix a bounded Borel set A A A and a bounded Borel function h h h vanishing outside A A A . Put h n ( u ) : = h ( u − t n ) h_n(u):=h(u-t_n) h n ( u ) := h ( u − t n ) . For all sufficiently large n n n , h n h_n h n vanishes on A A A . Translation invariance and the definition of θ t \theta_t θ t give
Z h n η = Z h η ∘ θ t n ⟶ Z h η in L 2 ( η ) as n → ∞ . Z_{h_n}^{\eta}=Z_h^\eta\circ\theta_{t_n}\longrightarrow Z_h^\eta\quad\text{in }L^2(\eta)\text{ as }n\to\infty. Z h n η = Z h η ∘ θ t n ⟶ Z h η in L 2 ( η ) as n → ∞.
Thus Z h η Z_h^\eta Z h η belongs to the closed linear span of centered exterior statistics. □
In particular, Proposition A.4 gives R e v ρ ⊆ L M R i g ρ \mathrm{Rev}_\rho\subseteq\mathrm{LMRig}_\rho Rev ρ ⊆ LMRig ρ . If a ∈ A d ( X , μ ) a\in A_d(X,\mu) a ∈ A d ( X , μ ) admits a rigidity sequence and κ ∈ R ρ , 2 a ( X , μ ) \kappa\in R^a_{\rho,2}(X,\mu) κ ∈ R ρ , 2 a ( X , μ ) , then κ ∗ μ ∈ L M R i g ρ \kappa_*\mu\in\mathrm{LMRig}_\rho κ ∗ μ ∈ LMRig ρ : the Koopman representation of the factor is a subrepresentation of that of a a a .
Explicit intermittency and linear rigidity from irrational rotations
Let T : = R / Z T:=\mathbb{R}/\mathbb{Z} T := R / Z and I : = [ 0 , 1 / 2 ) ⊂ T I:=[0,1/2)\subset T I := [ 0 , 1/2 ) ⊂ T . For an irrational α ∈ ( 0 , 1 ) \alpha\in(0,1) α ∈ ( 0 , 1 ) and ( ϑ , u ) ∈ T × [ 0 , 1 ) (\vartheta,u)\in T\times[0,1) ( ϑ , u ) ∈ T × [ 0 , 1 ) , set
Λ α , ϑ , u : = { n + u : n ∈ Z , ϑ + n α ∈ I } . \Lambda_{\alpha,\vartheta,u}:=\{n+u:n\in\mathbb{Z},\ \vartheta+n\alpha\in I\}. Λ α , ϑ , u := { n + u : n ∈ Z , ϑ + n α ∈ I } .
Define the probability measure η α \eta_\alpha η α on Conf ( R ) \operatorname{Conf}(\mathbb{R}) Conf ( R ) by
∫ F d η α : = ∫ 0 1 ∫ T F ( δ Λ α , ϑ , u ) d ϑ d u ( F bounded Borel ) . \int F\,\mathrm{d}\eta_\alpha:=\int_0^1\int_T F(\delta_{\Lambda_{\alpha,\vartheta,u}})\,\mathrm{d}\vartheta\,\mathrm{d}u\qquad(F\text{ bounded Borel}). ∫ F d η α := ∫ 0 1 ∫ T F ( δ Λ α , ϑ , u ) d ϑ d u ( F bounded Borel ) .
This is the cut-and-project process of [5 ] for the lattice
Γ α : = { ( n , n α ) : n ∈ Z } < R × T \Gamma_\alpha:=\{(n,n\alpha):n\in\mathbb{Z}\}<\mathbb{R}\times T Γ α := {( n , n α ) : n ∈ Z } < R × T
and window I I I . Within the parameter interval ( 0 , 1 ) (0,1) ( 0 , 1 ) , the hyperuniform parameters have full Lebesgue measure but form a meagre set, whereas intermittent hyperuniformity holds for a dense G δ G_\delta G δ -set. The genericity results in the main part of the paper refer instead to Baire category on E P P ρ , 2 \mathrm{EPP}_{\rho,2} EPP ρ , 2 , which carries no specified probability measure.
Write the regular continued fraction of α \alpha α as α = [ 0 ; a 1 , a 2 , … ] \alpha=[0;a_1,a_2,\ldots] α = [ 0 ; a 1 , a 2 , … ] . For j ≥ 1 j\ge1 j ≥ 1 , let p j / q j p_j/q_j p j / q j be its j j j th convergent, in lowest terms, and put q 0 : = 1 q_0:=1 q 0 := 1 and q − 1 : = 0 q_{-1}:=0 q − 1 := 0 . Then q j + 1 = a j + 1 q j + q j − 1 q_{j+1}=a_{j+1}q_j+q_{j-1} q j + 1 = a j + 1 q j + q j − 1 and q j → ∞ q_j\to\infty q j → ∞ as j → ∞ j\to\infty j → ∞ .
Proposition B.1. For every irrational α ∈ ( 0 , 1 ) \alpha\in(0,1) α ∈ ( 0 , 1 ) , the measure η α \eta_\alpha η α is translation-invariant and ergodic, has intensity 1 / 2 1/2 1/2 , and belongs to L M R i g 1 / 2 \mathrm{LMRig}_{1/2} LMRig 1/2 . It is supported on uniformly discrete and relatively dense point configurations. Moreover,
V η α ( q j / 2 ) ≤ 4 ( j ≥ 1 ) . V_{\eta_\alpha}(q_j/2)\le4 \qquad(j\ge1). V η α ( q j /2 ) ≤ 4 ( j ≥ 1 ) . There is a dense G δ G_\delta G δ -set G ⊂ ( 0 , 1 ) \mathcal{G}\subset(0,1) G ⊂ ( 0 , 1 ) of Liouville numbers such that
η α ∈ I H y p 1 / 2 ∩ L M R i g 1 / 2 ( α ∈ G ) . \eta_\alpha\in\mathrm{IHyp}_{1/2}\cap\mathrm{LMRig}_{1/2}\qquad(\alpha\in\mathcal{G}). η α ∈ IHyp 1/2 ∩ LMRig 1/2 ( α ∈ G ) . One explicit parameter in G \mathcal{G} G is
α ∗ : = ∑ n = 1 ∞ 3 − n ! . \alpha_*:=\sum_{n=1}^{\infty}3^{-n!}. α ∗ := n = 1 ∑ ∞ 3 − n ! .
For β ∈ R \beta\in\mathbb{R} β ∈ R and an integer N ≥ 1 N\ge1 N ≥ 1 , set
D N β ( ϑ ) : = ∑ j = 0 N − 1 ( 1 I ( ϑ + j β ) − 1 / 2 ) . D_N^\beta(\vartheta):=\sum_{j=0}^{N-1}\left(\mathbf{1}_I(\vartheta+j\beta)-1/2\right). D N β ( ϑ ) := j = 0 ∑ N − 1 ( 1 I ( ϑ + j β ) − 1/2 ) .
Lemma B.2 (Rational shadowing). Let p , q p,q p , q be relatively prime integers with q q q positive and odd. If M ≥ 1 M\ge1 M ≥ 1 is an integer and
∣ β − p q ∣ < 1 100 M q 2 , \left|\beta-\frac{p}{q}\right|<\frac{1}{100Mq^2}, β − q p < 100 M q 2 1 , then
∫ T ∣ D M q β ( ϑ ) ∣ 2 d ϑ ≥ 6 25 M 2 . \int_{\mathbb{T}}\left|D_{Mq}^\beta(\vartheta)\right|^2\,\mathrm{d}\vartheta\ge\frac{6}{25}M^2. ∫ T D M q β ( ϑ ) 2 d ϑ ≥ 25 6 M 2 .
Proof. For 0 ≤ j < M q 0\le j<Mq 0 ≤ j < M q , the circle distance between ϑ + j β \vartheta+j\beta ϑ + j β and ϑ + j p / q \vartheta+jp/q ϑ + j p / q is less than 1 / ( 100 q ) 1/(100q) 1/ ( 100 q ) . Let B B B be the set of phases for which one of the q q q points ϑ + j p / q \vartheta+jp/q ϑ + j p / q , 0 ≤ j < q 0\le j<q 0 ≤ j < q , lies within 1 / ( 100 q ) 1/(100q) 1/ ( 100 q ) of either endpoint of I I I . The union bound gives
∣ B ∣ ≤ q ⋅ 2 ⋅ 2 100 q = 1 25 . |B|\le q\cdot2\cdot\frac{2}{100q}=\frac{1}{25}. ∣ B ∣ ≤ q ⋅ 2 ⋅ 100 q 2 = 25 1 .
Outside B B B , the two itineraries agree for all M q Mq M q steps. Since q q q is odd, a rational period contains either ( q − 1 ) / 2 (q-1)/2 ( q − 1 ) /2 or ( q + 1 ) / 2 (q+1)/2 ( q + 1 ) /2 visits to I I I . Its centered discrepancy is therefore 1 / 2 1/2 1/2 or − 1 / 2 -1/2 − 1/2 , with the same sign in every period. Thus ∣ D M q β ∣ = M / 2 \left|D_{Mq}^\beta\right|=M/2 D M q β = M /2 outside B B B , and
∫ T ∣ D M q β ∣ 2 ≥ 24 25 M 2 4 = 6 25 M 2 . \int_{\mathbb{T}}\left|D_{Mq}^\beta\right|^2\ge\frac{24}{25}\frac{M^2}{4}=\frac{6}{25}M^2. ∫ T D M q β 2 ≥ 25 24 4 M 2 = 25 6 M 2 .
Figure 1.
Proof of Proposition B.1 . Translation invariance and geometry. For t ∈ R t\in\mathbb{R} t ∈ R , put k = ⌊ u − t ⌋ k=\lfloor u-t\rfloor k = ⌊ u − t ⌋ . Translation of a configuration by − t -t − t corresponds to the transformation
( ϑ , u ) ⟼ ( ϑ − k α , u − t − k ) . (\vartheta,u)\longmapsto(\vartheta-k\alpha,u-t-k). ( ϑ , u ) ⟼ ( ϑ − k α , u − t − k ) .
This action preserves the product of Haar measure and Lebesgue measure. Its time-one map is ( ϑ , u ) ↦ ( ϑ + α , u ) (\vartheta,u)\mapsto(\vartheta+\alpha,u) ( ϑ , u ) ↦ ( ϑ + α , u ) . A function invariant under the whole action is therefore independent of ϑ \vartheta ϑ , by irrationality of α \alpha α , and invariance under the remaining times makes it constant in u u u . Hence the action, and its point-process factor, are ergodic.
For every bounded Borel set A ⊂ R A\subset\mathbb{R} A ⊂ R ,
∫ N A d η α = 1 2 ∫ 0 1 ∑ n ∈ Z 1 A ( n + u ) d u = 1 2 ∣ A ∣ . \int N_A\,d\eta_\alpha=\frac{1}{2}\int_0^1\sum_{n\in\mathbb{Z}}1_A(n+u)\,du=\frac{1}{2}|A|. ∫ N A d η α = 2 1 ∫ 0 1 n ∈ Z ∑ 1 A ( n + u ) d u = 2 1 ∣ A ∣.
All local counts are bounded by those of Z + u \mathbb{Z}+u Z + u , and distinct points have distance at least one. Minimality of the irrational rotation, applied to int I \operatorname{int} I int I , gives a finite bound on return gaps, uniform in the phase. Thus the configurations are relatively dense.
Linear maximal rigidity. Let A ⊂ R A\subset\mathbb{R} A ⊂ R be bounded and let h h h be a bounded Borel function vanishing outside A A A . Choose an integer K K K bounding the number of sites of every translate Z + u \mathbb{Z}+u Z + u in a bounded interval containing A A A . Put
h j ( v ) : = h ( v − q j ) , ε j : = dist ( q j α , Z ) . h_j(v):=h(v-q_j),\qquad\varepsilon_j:=\operatorname{dist}(q_j\alpha,\mathbb{Z}). h j ( v ) := h ( v − q j ) , ε j := dist ( q j α , Z ) .
The supports of h j h_j h j are disjoint from A A A for all sufficiently large j j j , and ε j → 0 \varepsilon_j\to0 ε j → 0 . Reindexing the sums gives
S h j − S h = ∑ n ∈ Z h ( n + u ) ( 1 I ( ϑ + n α + q j α ) − 1 I ( ϑ + n α ) ) . S h_j-S h=\sum_{n\in\mathbb{Z}}h(n+u)\left(1_I(\vartheta+n\alpha+q_j\alpha)-1_I(\vartheta+n\alpha)\right). S h j − S h = n ∈ Z ∑ h ( n + u ) ( 1 I ( ϑ + n α + q j α ) − 1 I ( ϑ + n α ) ) .
There are at most K K K nonzero summands. For each summand the two indicators disagree on a set of phases of measure 2 ε j 2\varepsilon_j 2 ε j . Cauchy–Schwarz and integration therefore give
∥ S h j − S h ∥ L 2 ( η α ) 2 ≤ 2 K 2 ∥ h ∥ ∞ 2 ε j ⟶ 0 ( j → ∞ ) . \lVert S h_j-S h\rVert_{L^2(\eta_\alpha)}^2\le2K^2\lVert h\rVert_\infty^2\varepsilon_j\longrightarrow0\qquad(j\to\infty). ∥ S h j − S h ∥ L 2 ( η α ) 2 ≤ 2 K 2 ∥ h ∥ ∞ 2 ε j ⟶ 0 ( j → ∞ ) .
Since ∫ h j = ∫ h \int h_j=\int h ∫ h j = ∫ h , the same convergence holds for the centered statistics. Hence Z h η α Z^{\eta_\alpha}_h Z h η α lies in the closed linear span of exterior centered statistics, proving η α ∈ L M R i g 1 / 2 \eta_\alpha\in\mathrm{LMRig}_{1/2} η α ∈ LMRig 1/2 .
Bounded variance. For an integer N ≥ 1 N\ge1 N ≥ 1 , the interval [ 0 , N ) [0,N) [ 0 , N ) contains exactly the sites u , u + 1 , … , u + N − 1 u,u+1,\ldots,u+N-1 u , u + 1 , … , u + N − 1 . Translation invariance therefore gives
V η α ( N / 2 ) = ∫ T ∣ D N α ( ϑ ) ∣ 2 d ϑ . (47) V_{\eta_\alpha}(N/2)=\int_{\mathbb{T}}\left|D_N^\alpha(\vartheta)\right|^2\,d\vartheta. \tag*{(47)} V η α ( N /2 ) = ∫ T ∣ D N α ( ϑ ) ∣ 2 d ϑ . (47)
The function 1 I − 1 / 2 1_I-1/2 1 I − 1/2 has mean zero and total variation two. The Denjoy–Koksma inequality [21 Chap. VI] yields ∣ D q j α ∣ ≤ 2 \lvert D_{q_j}^\alpha\rvert\le2 ∣ D q j α ∣ ≤ 2 , and hence V η α ( q j / 2 ) ≤ 4 V_{\eta_\alpha}(q_j/2)\le4 V η α ( q j /2 ) ≤ 4 . Since q j → ∞ q_j\to\infty q j → ∞ , for every β > 0 \beta>0 β > 0 we have
lim j → ∞ V η α ( q j / 2 ) ( q j / 2 ) β = 0. \lim_{j\to\infty}\frac{V_{\eta_\alpha}(q_j/2)}{(q_j/2)^\beta}=0. j → ∞ lim ( q j /2 ) β V η α ( q j /2 ) = 0.
Large variance. For k ≥ 6 k\ge6 k ≥ 6 , set
O k : = ⋃ q ≥ k o d d , p ∈ Z ( p , q ) = 1 ( p q − q − k , p q + q − k ) ∩ ( 0 , 1 ) , G : = ( ( 0 , 1 ) ∖ Q ) ∩ ⋂ k ≥ 6 O k . \mathcal{O}_k:=\bigcup_{\substack{q\ge k\ \mathrm{odd},\ p\in\mathbb{Z}\\(p,q)=1}}\left(\frac{p}{q}-q^{-k},\frac{p}{q}+q^{-k}\right)\cap(0,1),\qquad\mathcal{G}:=((0,1)\setminus\mathbb{Q})\cap\bigcap_{k\ge6}\mathcal{O}_k. O k := q ≥ k odd , p ∈ Z ( p , q ) = 1 ⋃ ( q p − q − k , q p + q − k ) ∩ ( 0 , 1 ) , G := (( 0 , 1 ) ∖ Q ) ∩ k ≥ 6 ⋂ O k .
Reduced rationals with arbitrarily large odd denominators are dense. Thus every O k \mathcal{O}_k O k is open and dense, and G \mathcal{G} G is a dense G δ G_\delta G δ -set consisting of Liouville numbers.
For α ∈ G \alpha\in\mathcal{G} α ∈ G and each k ≥ 6 k\ge6 k ≥ 6 , choose p , q p,q p , q as in the definition of O k \mathcal{O}_k O k and put
M : = ⌊ q k − 2 100 ⌋ , N : = M q . M:=\left\lfloor\frac{q^{k-2}}{100}\right\rfloor,\qquad N:=Mq. M := ⌊ 100 q k − 2 ⌋ , N := M q .
Then M ≥ q k − 2 / 200 M \ge q^{k-2}/200 M ≥ q k − 2 /200 and ∣ α − p / q ∣ < 1 / ( 100 M q 2 ) |\alpha-p/q|<1/(100Mq^2) ∣ α − p / q ∣ < 1/ ( 100 M q 2 ) . The rational-shadowing lemma gives
V η α ( N / 2 ) ≥ 6 25 M 2 . V_{\eta_\alpha}(N/2)\ge\frac{6}{25}M^2. V η α ( N /2 ) ≥ 25 6 M 2 .
Consequently, for every fixed γ < 2 \gamma<2 γ < 2 ,
V η α ( N / 2 ) N γ ≥ 6 25 200 − ( 2 − γ ) q ( k − 2 ) ( 2 − γ ) − γ ⟶ + ∞ . \frac{V_{\eta_\alpha}(N/2)}{N^\gamma}\ge\frac{6}{25}200^{-(2-\gamma)}q^{(k-2)(2-\gamma)-\gamma}\longrightarrow+\infty. N γ V η α ( N /2 ) ≥ 25 6 20 0 − ( 2 − γ ) q ( k − 2 ) ( 2 − γ ) − γ ⟶ + ∞.
Here q ≥ k q\ge k q ≥ k , so N → ∞ N\to\infty N → ∞ . Since R = N / 2 → ∞ R=N/2\to\infty R = N /2 → ∞ as k → ∞ k\to\infty k → ∞ , this gives lim sup R → ∞ V η α ( R ) / R γ = ∞ \limsup_{R\to\infty}V_{\eta_\alpha}(R)/R^\gamma=\infty lim sup R → ∞ V η α ( R ) / R γ = ∞ for every γ < 2 \gamma<2 γ < 2 . Together with the bounded-variance subsequence it gives η α ∈ I H y p 1 / 2 \eta_\alpha\in\mathrm{IHyp}_{1/2} η α ∈ IHyp 1/2 .
Explicit parameter. In the base-3 expansion of α ∗ = ∑ n ≥ 1 3 − n ! \alpha_*=\sum_{n\ge1}3^{-n!} α ∗ = ∑ n ≥ 1 3 − n ! , the digit at position n ! n! n ! is 1, and all other digits are 0. The increasing gaps between these positions show that the base-3 digit sequence is not eventually periodic, so α ∗ \alpha_* α ∗ is irrational. The n n n th partial sum of the displayed series is p n / q n p_n/q_n p n / q n , where q n = 3 n ! q_n=3^{n!} q n = 3 n ! and p n ≡ 1 ( m o d 3 ) p_n\equiv1\pmod3 p n ≡ 1 ( mod 3 ) . Thus the fraction is reduced and q n q_n q n is odd. Moreover,
0 < α ∗ − p n q n < 2 3 − ( n + 1 ) ! = 2 q n − ( n + 1 ) . 0<\alpha_*-\frac{p_n}{q_n}<2\,3^{-(n+1)!}=2q_n^{-(n+1)}. 0 < α ∗ − q n p n < 2 3 − ( n + 1 )! = 2 q n − ( n + 1 ) .
For every k ≥ 6 k\ge6 k ≥ 6 , choosing n n n sufficiently large gives q n ≥ k q_n\ge k q n ≥ k and ∣ α ∗ − p n / q n ∣ < q n − k |\alpha_*-p_n/q_n|<q_n^{-k} ∣ α ∗ − p n / q n ∣ < q n − k . Hence α ∗ ∈ G \alpha_*\in\mathcal{G} α ∗ ∈ G .
Corollary B.3 (Category and measure in the rotation family). The set of irrational α ∈ ( 0 , 1 ) \alpha\in(0,1) α ∈ ( 0 , 1 ) for which η α \eta_\alpha η α is hyperuniform has full Lebesgue measure but is meagre in ( 0 , 1 ) (0,1) ( 0 , 1 ) . The intermittently hyperuniform parameters contain the dense G δ G_\delta G δ -set G \mathcal{G} G from Proposition B.1 . Every irrational parameter gives a linearly maximally rigid process.
Proof. The last two assertions follow from Proposition B.1 . Since intermittent hyperuniformity excludes hyperuniformity, the set of hyperuniform parameters is contained in the meagre complement of G \mathcal{G} G .
Fix 0 < ε < 1 / 2 0<\varepsilon<1/2 0 < ε < 1/2 . For each integer q ≥ 1 q\ge1 q ≥ 1 , the set
{ α ∈ ( 0 , 1 ) : dist ( q α , Z ) < q − 1 − ε } \{\alpha\in(0,1):\operatorname{dist}(q\alpha,\mathbb{Z})<q^{-1-\varepsilon}\} { α ∈ ( 0 , 1 ) : dist ( q α , Z ) < q − 1 − ε }
has Lebesgue measure at most 2 q − 1 − ε 2q^{-1-\varepsilon} 2 q − 1 − ε . By Borel–Cantelli, for almost every α \alpha α there is q 0 q_0 q 0 such that dist ( q α , Z ) ≥ q − 1 − ε \operatorname{dist}(q\alpha,\mathbb{Z})\ge q^{-1-\varepsilon} dist ( q α , Z ) ≥ q − 1 − ε for q ≥ q 0 q\ge q_0 q ≥ q 0 . If p j / q j p_j/q_j p j / q j is a continued-fraction convergent of such an irrational α \alpha α , then
q j − 1 − ε ≤ ∣ q j α − p j ∣ < q j + 1 − 1 q_j^{-1-\varepsilon}\le|q_j\alpha-p_j|<q_{j+1}^{-1} q j − 1 − ε ≤ ∣ q j α − p j ∣ < q j + 1 − 1
for all sufficiently large j j j . Consequently a j + 1 ≤ C α q j ε a_{j+1}\le C_\alpha q_j^\varepsilon a j + 1 ≤ C α q j ε , where a j + 1 a_{j+1} a j + 1 is the next continued-fraction coefficient.
Fix an integer N ≥ 1 N\ge1 N ≥ 1 . Choose J J J maximal with q J ≤ N q_J\le N q J ≤ N and subtract successively the largest possible multiples of q J , q J − 1 , … , q 0 q_J,q_{J-1},\ldots,q_0 q J , q J − 1 , … , q 0 . The recurrence for q j q_j q j gives
N = ∑ j = 0 J b j q j , 0 ≤ b j ≤ a j + 1 . N=\sum_{j=0}^{J}b_jq_j,\qquad0\le b_j\le a_{j+1}. N = j = 0 ∑ J b j q j , 0 ≤ b j ≤ a j + 1 .
Since q j + 2 ≥ 2 q j q_{j+2}\ge2q_j q j + 2 ≥ 2 q j , we have J = O α ( log ( 2 N ) ) J=O_\alpha(\log(2N)) J = O α ( log ( 2 N )) . Partition the first N N N iterates into consecutive blocks of lengths q j q_j q j , with b j b_j b j blocks of length q j q_j q j . Applying Denjoy–Koksma at the initial phase of each block gives
sup ϑ ∈ T ∣ D N α ( ϑ ) ∣ ≤ 2 ∑ j = 0 J b j ≤ C α N ε log ( 2 N ) . \sup_{\vartheta\in\mathbb{T}}|D_N^{\alpha}(\vartheta)|\le2\sum_{j=0}^{J}b_j\le C_\alpha N^\varepsilon\log(2N). ϑ ∈ T sup ∣ D N α ( ϑ ) ∣ ≤ 2 j = 0 ∑ J b j ≤ C α N ε log ( 2 N ) .
By (B.1),
V η α ( N / 2 ) ≤ C α N 2 ε log 2 ( 2 N ) = o ( N ) . V_{\eta_\alpha}(N/2)\le C_\alpha N^{2\varepsilon}\log^2(2N)=o(N). V η α ( N /2 ) ≤ C α N 2 ε log 2 ( 2 N ) = o ( N ) .
If 2 R ∈ [ N , N + 1 ) 2R \in[N, N+1) 2 R ∈ [ N , N + 1 ) , the counts in [ 0 , 2 R ) [0,2R) [ 0 , 2 R ) and [ 0 , N ) [0,N) [ 0 , N ) differ by at most one. Their centered counts therefore differ in L 2 L^2 L 2 by at most two. Translation invariance identifies their variances with those of balls of radii R R R and N / 2 N/2 N /2 . Hence V η α ( R ) = o ( R ) V_{\eta_\alpha}(R)=o(R) V η α ( R ) = o ( R ) , which is hyperuniformity in dimension one. □ \square □
Funding
This work was supported by the Swedish Research Council under grant VR 11253322.
During the preparation of this manuscript, the author used ChatGPT (OpenAI) and Astra for mathematical exploration, proof analysis, and drafting; Aristotle (Harmonic) for substantial formalization and verification of selected arguments in Lean; and Claude Sonnet (Anthropic) for an independent adversarial mathematical review. The author reviewed and revised the AI-assisted contributions and takes full responsibility for the mathematical results and the final manuscript.
Department of Mathematics, Chalmers University of Technology and University of Gothenburg, Gothenburg, Sweden
Email address: micbjo@chalmers.se