Documentation

LeanPool.RareSubtreePatterns.Applications.CompleteArySC

Complete r-ary marks: Assumptions (S) and (C) (Lemmas lem:S_rary and lem:C_rary) #

Section 6.2 of the paper, source lines 2623–2792 (inventory L6.2 (i),(ii), L6.3, P19, P21). Marks: G_k(T) = 1{H_{r,k} ≺ T} (Embeds (completeTree r k)) and its declumping Ghat_k(T) = 1{k_∅(T) ≠ r} 1{∃ u ∈ Γ_∅(T), G_k(T_u) = 1} (declumpedMark {d | d ≠ r} (Embeds (completeTree r k))). π_k = P(H_{r,k} ≺ 𝒯).

Main statements #

Proof of P19 (departure from the paper's bookkeeping) #

The paper splits the descendants u_{ℓ+1} of a root occurrence into the vertices of the displayed skeleton and those of the grafted forest, and bounds the forest part by a union bound over DFS positions of an infinite forest. Here both parts are handled by the cut-and-graft decomposition T = Cut_ℓ(T)[T_{u_{ℓ+1}}] at each fixed position ℓ: every position has a class j ≤ k (exists_replaceClass; j = 0 for forest vertices and skeleton leaves, j = k - a for internal skeleton vertices at depth a), and the cut-and-graft factorization gives exactly P(ℓ < |𝒯|, class j, G_k(𝒯_{u_{ℓ+1}})) = (π_k / π_j) P(ℓ < |𝒯|, G_k(𝒯), class j). The class 0 terms give M π_k² (the paper's forest bound M_n π_{k_n} · π_{k_n}; the "fresh increment" property at a fixed forest position is the factorization 𝐏(A[S]) = w(A) 𝐏(S)), and the classes j ≥ 1 give at most r^{k-j} positions each (classCount_le). With a = k - j this is the paper's skeleton sum π_k ∑_{a} r^a π_k / π_{k-a}; using π_k / π_{k-a} ≤ μ_r^{s_{r,a}} (superadditivity of s_{r,·}) it is bounded by K π_k with the k-independent convergent series K = ∑_a r^a μ_r^{s_{r,a}} (μ_r < 1). For r = 1 this is the paper's geometric continuation ∑_a μ_1^a; for r ≥ 2 the paper uses instead the bound k r^k μ_r^{r^{k-1}} + r^k π_k (its leaf term r^k π_k is part of our class 0). No forest DFS bookkeeping is needed.

Assumption (S) #

Under the decomposition embedsSkeletonLaw μ T' (U06), C = s(T') and L = ℓ(T') are deterministic, so E[C + L] ≤ s(T') + ℓ(T').

theorem PoissonDeclumping.Applications.structuralMicroscopic_embeds {μ : PMF ℕ} {α : ℝ} {a : ℕ → ℝ} (hT : StableToolkit μ α a) {η : ℝ} (hη : 0 < η ∧ η < 1 - 1 / α) (P : ℕ → PlaneTree) {k : ℕ → ℕ} (hk : ∀ (δ : ℝ), 0 < δ → ∀ᶠ (n : ℕ) in admissibleSizes μ, ↑((P (k n)).internalCount + (P (k n)).leafCount) ≤ ↑n ^ δ) :
StructuralMicroscopic μ a (fun (j : ℕ) => (P j).Embeds) k fun (n : ℕ) => max 1 ⌊↑n ^ (1 - η)⌋₊

Assumption (S) for non-fringe pattern marks G_j = 1{P_j ≺ ·} whose skeleton size s(P_{k_n}) + ℓ(P_{k_n}) (internal vertices plus leaves) is n^{o(1)}, with respect to the polynomial cutoff M_n = max 1 ⌊n^{1-η}⌋₊ (= ⌊n^{1-η}⌋ for n ≥ 1), η ∈ (0, 1 - 1/α): Lemma lem:critM applied to the decomposition of U06 (source lines 2655–2668).

theorem PoissonDeclumping.Applications.lem_S_rary_i {μ : PMF ℕ} {α : ℝ} {a : ℕ → ℝ} (hT : StableToolkit μ α a) {r : ℕ} {η : ℝ} (hη : 0 < η ∧ η < 1 - 1 / α) {k : ℕ → ℕ} (hk : ∀ (δ : ℝ), 0 < δ → ∀ᶠ (n : ℕ) in admissibleSizes μ, ↑((PlaneTree.completeTree r (k n)).internalCount + (PlaneTree.completeTree r (k n)).leafCount) ≤ ↑n ^ δ) :
StructuralMicroscopic μ a (fun (j : ℕ) => (PlaneTree.completeTree r j).Embeds) k fun (n : ℕ) => max 1 ⌊↑n ^ (1 - η)⌋₊

Lemma lem:S_rary(i) (source lines 2644–2668): if s_{r,k_n} + ℓ_{r,k_n} = n^{o(1)} (s_{r,k} = number of internal vertices and ℓ_{r,k} = number of leaves of H_{r,k}, identified with k, 1 resp. (r^k - 1)/(r - 1), r^k by U07), then G_k = 1{H_{r,k} ≺ ·} satisfies Assumption (S) along (k_n) with respect to M_n = max 1 ⌊n^{1-η}⌋₊, for every η ∈ (0, 1 - 1/α). The n^{o(1)} hypothesis is used only through its upper half ≤ n^δ (for every δ > 0, eventually along admissible sizes).

theorem PoissonDeclumping.Applications.lem_S_rary_i_one {μ : PMF ℕ} {α : ℝ} {a : ℕ → ℝ} (hT : StableToolkit μ α a) {η : ℝ} (hη : 0 < η ∧ η < 1 - 1 / α) {k : ℕ → ℕ} (hk : ∀ (δ : ℝ), 0 < δ → ∀ᶠ (n : ℕ) in admissibleSizes μ, ↑(k n) ≤ ↑n ^ δ) :
StructuralMicroscopic μ a (fun (j : ℕ) => (PlaneTree.completeTree 1 j).Embeds) k fun (n : ℕ) => max 1 ⌊↑n ^ (1 - η)⌋₊

Lemma lem:S_rary(i), r = 1: the hypothesis s_{1,k_n} + ℓ_{1,k_n} = k_n + 1 = n^{o(1)} is k_n = n^{o(1)}.

theorem PoissonDeclumping.Applications.lem_S_rary_i_of_two_le {μ : PMF ℕ} {α : ℝ} {a : ℕ → ℝ} (hT : StableToolkit μ α a) {r : ℕ} (hr : 2 ≤ r) {η : ℝ} (hη : 0 < η ∧ η < 1 - 1 / α) {k : ℕ → ℕ} (hk : ∀ (δ : ℝ), 0 < δ → ∀ᶠ (n : ℕ) in admissibleSizes μ, ↑r ^ k n ≤ ↑n ^ δ) :
StructuralMicroscopic μ a (fun (j : ℕ) => (PlaneTree.completeTree r j).Embeds) k fun (n : ℕ) => max 1 ⌊↑n ^ (1 - η)⌋₊

Lemma lem:S_rary(i), r ≥ 2: the hypothesis s_{r,k_n} + ℓ_{r,k_n} = (r^{k_n} - 1)/(r - 1) + r^{k_n} = n^{o(1)} is r^{k_n} = n^{o(1)}.

P19: the summed pair bound for the raw marks #

The cut trees A ∈ 𝒜_ℓ of class j: H_{r,k} ≺ A[S] ↔ H_{r,j} ≺ S for every S.

Equations
  • One or more equations did not get rendered due to their size.
Instances For

    The cut-and-graft factorization for one class (P19): P(ℓ < |𝒯|, class j, 𝒯_{u_{ℓ+1}} ∈ G_k) π_j = P(ℓ < |𝒯|, class j, 𝒯_{u_{ℓ+1}} ∈ G_j) π_k.

    The event {ℓ < |𝒯|, G_k(𝒯) = 1, position ℓ of class j}.

    Equations
    • One or more equations did not get rendered due to their size.
    Instances For

      π_j = P(H_{r,j} ≺ 𝒯) = μ_r^{s_{r,j}} as a real number (U07).

      P19, one position. A(ℓ) ≤ ∑_{j ≤ k} (π_k / π_j) P(ℓ < |𝒯|, G_k(𝒯) = 1, class j).

      theorem PoissonDeclumping.Applications.sum_bienayme_classEvent_le {μ : PMF ℕ} {r : ℕ} (hr : 1 ≤ r) (k M j : ℕ) (hj : 1 ≤ j) :

      P19, count of skeleton positions. For j ≥ 1, ∑_{ℓ=1}^{M} P(ℓ < |𝒯|, G_k(𝒯) = 1, class j) ≤ r^{k-j} π_k.

      s_{r,c} + s_{r,a} ≤ s_{r,a+c} (r ≥ 1), where s_{r,a} is the number of internal vertices of H_{r,a}.

      theorem PoissonDeclumping.Applications.summable_skeletonWeight {r : ℕ} (hr : 1 ≤ r) {x : ℝ} (hx0 : 0 ≤ x) (hx1 : x < 1) :
      Summable fun (a : ℕ) => ↑r ^ a * x ^ (PlaneTree.completeTree r a).internalCount

      The weights g_a = r^a μ_r^{s_{r,a}} of the skeleton contribution are summable (μ_r < 1): for r = 1 this is the geometric series ∑ μ_1^a, for r ≥ 2 the ratio r μ_r^{r^a} tends to 0.

      theorem PoissonDeclumping.Applications.pairSum_completeTree_le {μ : PMF ℕ} (hc : IsCritical μ) (hne : μ 1 ≠ 1) {r : ℕ} (hr : 1 ≤ r) (hμr : 0 < μ r) :

      P19 (source lines 2670–2750): for r ≥ 1 with μ_r > 0 there is one constant K such that, for every level k and every M, ∑_{ℓ=1}^{M} P(|𝒯| ≥ ℓ+1, G_k(𝒯) = 1, G_k(𝒯_{u_{ℓ+1}(𝒯)}) = 1) ≤ K π_k + M π_k².

      theorem PoissonDeclumping.Applications.lem_S_rary_ii {μ : PMF ℕ} (hc : IsCritical μ) (hne : μ 1 ≠ 1) {r : ℕ} (hr : 1 ≤ r) (hμr : 0 < μ r) {k M : ℕ → ℕ} (hM : (fun (n : ℕ) => ↑(M n)) =o[Filter.atTop] fun (n : ℕ) => ↑n) (hπ : Filter.Tendsto (fun (n : ℕ) => ↑n * markProb μ (fun (j : ℕ) => (PlaneTree.completeTree r j).Embeds) k n) (admissibleSizes μ) Filter.atTop) :
      NonClustering μ (fun (j : ℕ) => (PlaneTree.completeTree r j).Embeds) k M

      Lemma lem:S_rary(ii) (source lines 2670–2753), for an arbitrary cutoff with M_n = o(n) (the paper takes M_n = ⌊n^{1-η}⌋, see lem_S_rary): if μ_r > 0 (r ≥ 1) and n π_{k_n} → ∞ along admissible sizes, the raw marks G_k = 1{H_{r,k} ≺ ·} satisfy Assumption (C) along (k_n) with respect to (M_n). The hypothesis s_{r,k_n} + ℓ_{r,k_n} = n^{o(1)} of the lemma is not needed for this part.

      theorem PoissonDeclumping.Applications.lem_S_rary {μ : PMF ℕ} {α : ℝ} {a : ℕ → ℝ} (hT : StableToolkit μ α a) {r : ℕ} (hr : 1 ≤ r) (hμr : 0 < μ r) {η : ℝ} (hη : 0 < η ∧ η < 1 - 1 / α) {k : ℕ → ℕ} (hk : ∀ (δ : ℝ), 0 < δ → ∀ᶠ (n : ℕ) in admissibleSizes μ, ↑((PlaneTree.completeTree r (k n)).internalCount + (PlaneTree.completeTree r (k n)).leafCount) ≤ ↑n ^ δ) :
      (StructuralMicroscopic μ a (fun (j : ℕ) => (PlaneTree.completeTree r j).Embeds) k fun (n : ℕ) => max 1 ⌊↑n ^ (1 - η)⌋₊) ∧ (Filter.Tendsto (fun (n : ℕ) => ↑n * markProb μ (fun (j : ℕ) => (PlaneTree.completeTree r j).Embeds) k n) (admissibleSizes μ) Filter.atTop → NonClustering μ (fun (j : ℕ) => (PlaneTree.completeTree r j).Embeds) k fun (n : ℕ) => max 1 ⌊↑n ^ (1 - η)⌋₊)

      Lemma lem:S_rary(i) and (ii) with the paper's hypotheses (source lines 2644–2753): r ≥ 1, μ_r > 0, s_{r,k_n} + ℓ_{r,k_n} = n^{o(1)}, η ∈ (0, 1 - 1/α) and M_n = max 1 ⌊n^{1-η}⌋₊ (= ⌊n^{1-η}⌋ for n ≥ 1). Part (iii) (E[N_{k_n}(𝒯^n)] ∼ n π_{k_n}) is Corollary cor:SM(ii) and is not included.

      Lemma lem:C_rary: the declumped marks #

      P21 (source lines 2768–2786): for every r, k and ℓ ≥ 1, P(|𝒯| ≥ ℓ+1, Ghat_k(𝒯) = 1, Ghat_k(𝒯_{u_{ℓ+1}(𝒯)}) = 1) ≤ π̂_k², by Lemma lem:cut_graft with 𝓠_0 the whole tree space and 𝓡 the cut trees all of whose re-graftings are in {Ghat_k = 1}.

      theorem PoissonDeclumping.Applications.lem_C_rary {μ : PMF ℕ} (hc : IsCritical μ) (hne : μ 1 ≠ 1) (r : ℕ) (k M : ℕ → ℕ) (hM : (fun (n : ℕ) => ↑(M n)) =o[Filter.atTop] fun (n : ℕ) => ↑n) :

      Lemma lem:C_rary (source lines 2762–2791): for every r, every level sequence (k_n) and every cutoff sequence with M_n = o(n), the declumped complete r-ary marks Ghat_k = 1{k_∅ ≠ r} 1{∃ child c, H_{r,k} ≺ 𝒯_c} satisfy Assumption (C): ∑_{ℓ=1}^{M_n} Â_n(ℓ) ≤ M_n π̂_{k_n}² = o(n π̂_{k_n}²).

      theorem PoissonDeclumping.Applications.lem_C_rary_of_isCutoff {μ : PMF ℕ} (hc : IsCritical μ) (hne : μ 1 ≠ 1) (r : ℕ) (k : ℕ → ℕ) {M : ℕ → ℕ} (hM : IsCutoff M) :

      Lemma lem:C_rary for every cutoff sequence (IsCutoff M).