Paper
Erdős–Hajnal for the five-vertex path
19 pages · 22 marked passages · pdflatex · download PDF · lax-57
-
Erdős–Hajnal theorem for the five-vertex path
The five-vertex path has the Erdős–Hajnal property: there is a positive integer such that every finite graph with no induced copy of satisfies
1 import Lax57.GraphDefinitions 2 … module docstring, 13 lines 16 17 namespace Lax57.ErdosHajnalP5 18 19 open Lax57.GraphDefinitions 20 21 universe u 22 23 /-- The Erdős–Hajnal conjecture for the five-vertex path. -/ 24 axiom erdos_hajnal_P5 : 25 ∃ q : ℕ, 0 < q ∧ 26 ∀ {V : Type u} [Fintype V] (G : SimpleGraph V), 27 IsP5Free G → Fintype.card V ≤ homogeneousNumber G ^ q 28 29 end Lax57.ErdosHajnalP5 30 -
Rödl's theorem for induced-subgraph-free graphs
For every finite graph and every positive integer , there is a positive integer such that every finite induced--free graph contains a set satisfying and such that either or its complement has edge density at most . This is the cleared-denominator finite form of Rödl's theorem cited as Theorem 4.1 in the paper.
1 import Lax54.GraphDefinitions 2 … module docstring, 11 lines 14 15 open scoped SimpleGraph 16 17 namespace Lax54.RodlTheorem 18 19 open Lax54.GraphDefinitions 20 21 universe u v 22 23 /-- Rödl's theorem, with a reciprocal integer density parameter. -/ 24 axiom rodl_theorem : 25 ∀ {W : Type u} [Fintype W] (H : SimpleGraph W) (E : ℕ), 26 0 < E → ∃ D : ℕ, 0 < D ∧ 27 ∀ {V : Type v} [Fintype V] [DecidableEq V] (G : SimpleGraph V) 28 [DecidableRel G.Adj], 29 ¬ H ⊴ G → 30 ∃ X : Finset V, Fintype.card V ≤ D * X.card ∧ 31 HasSparseSide (G.induce (X : Set V)) E 32 33 end Lax54.RodlTheorem 34 -
Finite graph notions for the five-vertex path theorem
This module defines the five-vertex path , its complement (the house), and the notions of sparsity, restrictedness, and blockades used throughout the formalization. For a positive integer , means maximum degree at most , while is the corresponding edge-density condition with its denominator cleared.
1 import Lax54.GraphDefinitions 2 import Mathlib.Combinatorics.SimpleGraph.Density 3 import Mathlib.Combinatorics.SimpleGraph.Hasse 4 … module docstring, 11 lines 16 17 open Finset 18 open scoped SimpleGraph 19 20 namespace Lax57.GraphDefinitions 21 22 universe u 23 24 /-- The path on five vertices. -/ 25 abbrev P5 : SimpleGraph (Fin 5) := SimpleGraph.pathGraph 5 26 27 /-- The house graph, namely the complement of the five-vertex path. -/ 28 abbrev House : SimpleGraph (Fin 5) := P5ᶜ 29 30 /-- A graph has no induced copy of the five-vertex path. -/ 31 def IsP5Free {V : Type u} (G : SimpleGraph V) : Prop := 32 ¬ P5 ⊴ G 33 34 /-- A graph has no induced copy of the house. -/ 35 def IsHouseFree {V : Type u} (G : SimpleGraph V) : Prop := 36 ¬ House ⊴ G 37 38 /-- The largest cardinality of a clique or stable set. -/ 39 noncomputable abbrev homogeneousNumber {V : Type u} (G : SimpleGraph V) : ℕ := 40 Lax54.GraphDefinitions.homogeneousNumber G 41 42 /-- The product of the clique and independence numbers. -/ 43 noncomputable abbrev kappa {V : Type u} (G : SimpleGraph V) : ℕ := 44 Lax54.GraphDefinitions.kappa G 45 46 /-- `q`-criticality for the product `ω(G)α(G)`. -/ 47 abbrev IsQCritical {V : Type u} [Fintype V] (q : ℕ) (G : SimpleGraph V) : Prop := 48 Lax54.GraphDefinitions.IsQCritical q G 49 50 /-- The neighbors of `v` lying in a prescribed finite set. -/ 51 def neighborsIn {V : Type u} [DecidableEq V] (G : SimpleGraph V) 52 [DecidableRel G.Adj] (A : Finset V) (v : V) : Finset V := 53 A.filter fun x ↦ G.Adj v x 54 55 /-- `B` is `1/E`-sparse to `A`, with the denominator cleared. -/ 56 def ESparseTo {V : Type u} [DecidableEq V] (G : SimpleGraph V) 57 [DecidableRel G.Adj] (E : ℕ) (B A : Finset V) : Prop := 58 ∀ b ∈ B, E * (neighborsIn G A b).card ≤ A.card 59 60 /-- The pair `(A,B)` has edge density at most `1/E`. -/ 61 def WeaklyESparse {V : Type u} [DecidableEq V] (G : SimpleGraph V) 62 [DecidableRel G.Adj] (E : ℕ) (A B : Finset V) : Prop := 63 E * (G.interedges A B).card ≤ A.card * B.card 64 65 /-- The graph induced by `A` has maximum degree at most `|A|/E`. -/ 66 def ESparse {V : Type u} [DecidableEq V] (G : SimpleGraph V) 67 [DecidableRel G.Adj] (E : ℕ) (A : Finset V) : Prop := 68 ∀ v : {x : V // x ∈ A}, E * (G.induce (A : Set V)).degree v ≤ A.card 69 70 /-- One of the two complementary graphs induced by `A` is `1/E`-sparse. -/ 71 def ERestricted {V : Type u} [DecidableEq V] (G : SimpleGraph V) 72 [DecidableRel G.Adj] (E : ℕ) (A : Finset V) : Prop := 73 ESparse G E A ∨ ESparse Gᶜ E A 74 75 /-- A sequence of pairwise disjoint vertex blocks. -/ 76 structure Blockade {V : Type u} [DecidableEq V] (k : ℕ) where 77 block : Fin k → Finset V 78 disjoint : ∀ {i j : Fin k}, i ≠ j → Disjoint (block i) (block j) 79 80 /-- Every two different blocks are complete to one another. -/ 81 def Blockade.IsComplete {V : Type u} [DecidableEq V] {k : ℕ} 82 (G : SimpleGraph V) (B : Blockade (V := V) k) : Prop := 83 ∀ {i j : Fin k}, i ≠ j → 84 ∀ x ∈ B.block i, ∀ y ∈ B.block j, G.Adj x y 85 86 /-- Every two different blocks are anticomplete to one another. -/ 87 def Blockade.IsAnticomplete {V : Type u} [DecidableEq V] {k : ℕ} 88 (G : SimpleGraph V) (B : Blockade (V := V) k) : Prop := 89 ∀ {i j : Fin k}, i ≠ j → 90 ∀ x ∈ B.block i, ∀ y ∈ B.block j, ¬ G.Adj x y 91 92 /-- Each pair of blocks is either complete or anticomplete. -/ 93 def Blockade.IsPure {V : Type u} [DecidableEq V] {k : ℕ} 94 (G : SimpleGraph V) (B : Blockade (V := V) k) : Prop := 95 ∀ {i j : Fin k}, i ≠ j → 96 ((∀ x ∈ B.block i, ∀ y ∈ B.block j, G.Adj x y) ∨ 97 (∀ x ∈ B.block i, ∀ y ∈ B.block j, ¬ G.Adj x y)) 98 99 /-- A complete or anticomplete blockade. -/ 100 def Blockade.IsUniform {V : Type u} [DecidableEq V] {k : ℕ} 101 (G : SimpleGraph V) (B : Blockade (V := V) k) : Prop := 102 B.IsComplete G ∨ B.IsAnticomplete G 103 104 /-- The directed vertexwise sparsity condition of the paper. -/ 105 def Blockade.IsESparse {V : Type u} [DecidableEq V] {k : ℕ} 106 (G : SimpleGraph V) [DecidableRel G.Adj] (E : ℕ) 107 (B : Blockade (V := V) k) : Prop := 108 ∀ {i j : Fin k}, i < j → ESparseTo G E (B.block j) (B.block i) 109 110 /-- Each pair is complete or weakly `1/E`-sparse. -/ 111 def Blockade.IsSemisparse {V : Type u} [DecidableEq V] {k : ℕ} 112 (G : SimpleGraph V) [DecidableRel G.Adj] (E : ℕ) 113 (B : Blockade (V := V) k) : Prop := 114 ∀ {i j : Fin k}, i ≠ j → 115 ((∀ x ∈ B.block i, ∀ y ∈ B.block j, G.Adj x y) ∨ 116 WeaklyESparse G E (B.block i) (B.block j)) 117 118 /-- All blocks have size at least `|V(G)| / loss`. -/ 119 def Blockade.HasWidthLoss {V : Type u} [Fintype V] [DecidableEq V] 120 {k : ℕ} (B : Blockade (V := V) k) (loss : ℕ) : Prop := 121 ∀ i : Fin k, Fintype.card V ≤ loss * (B.block i).card 122 123 /-- Every block lies in a prescribed ambient vertex set. -/ 124 def Blockade.IsInside {V : Type u} [DecidableEq V] {k : ℕ} 125 (B : Blockade (V := V) k) (S : Finset V) : Prop := 126 ∀ i : Fin k, B.block i ⊆ S 127 128 /-- A uniform blockade with polynomial width and controlled length. -/ 129 def HasUniformBlockade {V : Type u} [Fintype V] [DecidableEq V] 130 (G : SimpleGraph V) (E a : ℕ) : Prop := 131 ∃ (k : ℕ) (B : Blockade (V := V) k), 132 2 ≤ k ∧ k ≤ E ∧ B.IsUniform G ∧ B.HasWidthLoss (k ^ a) 133 134 end Lax57.GraphDefinitions 135 -
Polynomial semisparse blockades for the house
This is the denominator-cleared form of Lemma 6.2 of Nguyen, Scott, and Seymour. A sufficiently large house-free graph contains disjoint polynomially large blocks, and every pair of blocks is either complete or has edge density at most .
1 import Lax57.GraphDefinitions 2 … module docstring, 10 lines 13 14 namespace Lax57.SemisparseBlockade 15 16 open Lax57.GraphDefinitions 17 18 universe u 19 20 /-- Polynomial semisparse blockades in house-free graphs. -/ 21 axiom semisparse_house_blockade : 22 ∃ d : ℕ, 40 ≤ d ∧ 23 ∀ E : ℕ, 2 ≤ E → 24 ∀ {V : Type u} [Fintype V] [DecidableEq V] 25 (G : SimpleGraph V) [DecidableRel G.Adj], 26 IsHouseFree G → E ^ (10 * d ^ 2) ≤ Fintype.card V → 27 ∃ B : Blockade (V := V) E, 28 B.IsSemisparse G (E ^ d) ∧ 29 B.HasWidthLoss (E ^ (10 * d ^ 2)) 30 31 end Lax57.SemisparseBlockade 32 -
Restricted set or uniform blockade in a house-free graph
There is an integer such that, for every , every finite house-free graph has one of two outcomes. Either an induced subgraph on is -restricted and , or has a complete or anticomplete blockade of length , where , whose blocks all have size at least .
This is the denominator-cleared form of Lemma 7.3 of Nguyen, Scott, and Seymour. The structural argument is stated for the house, the complement of the five-vertex path.
1 import Lax57.GraphDefinitions 2 … module docstring, 15 lines 18 19 namespace Lax57.HouseDichotomy 20 21 open Lax57.GraphDefinitions 22 23 universe u 24 25 /-- The structural dichotomy for finite house-free graphs. -/ 26 axiom house_dichotomy : 27 ∃ a : ℕ, 1 ≤ a ∧ 28 ∀ E : ℕ, 3 ≤ E → 29 ∀ {V : Type u} [Fintype V] [DecidableEq V] (G : SimpleGraph V) 30 [DecidableRel G.Adj], 31 IsHouseFree G → 32 (∃ X : Finset V, 33 Fintype.card V ≤ E ^ a * X.card ∧ ERestricted G E X) ∨ 34 HasUniformBlockade G E a 35 36 end Lax57.HouseDichotomy 37 -
Anticomponent or complete blockade
This denominator-cleared form of Lemma 4.1 groups the connected components of the complement. Either one anticonnected component has size at least , or there are pairwise complete groups, each of size at least .
1 import Lax57.GraphDefinitions 2 … module docstring, 10 lines 13 14 namespace Lax57.AnticomponentBlockade 15 16 open Lax57.GraphDefinitions 17 18 universe u 19 20 /-- A large anticonnected component or a long complete blockade. -/ 21 axiom anticomponent_or_complete_blockade : 22 ∀ {V : Type u} [Fintype V] [DecidableEq V] 23 (G : SimpleGraph V) [DecidableRel G.Adj] (T : Finset V) (Q : ℕ), 24 2 ≤ Q → 25 ( (∃ J : Finset V, J ⊆ T ∧ 26 (Gᶜ.induce (J : Set V)).Connected ∧ 27 T.card ≤ Q ^ 2 * J.card) ∨ 28 (∃ C : Blockade (V := V) Q, 29 C.IsInside T ∧ C.IsComplete G ∧ 30 ∀ i : Fin Q, 31 T.card ≤ 4 * Q ^ 3 * (C.block i).card) ) 32 33 end Lax57.AnticomponentBlockade 34 -
no assumptions
Consider the connected components of the complement induced on . A component of size at least gives the first outcome. Otherwise, greedily group whole components into disjoint families, each of size at least . Distinct families are anticomplete in the complement and therefore complete in .
-
Bipartite comb lemma
The case of Theorem 2.1, with denominators cleared. Let and be disjoint vertex sets. Suppose that every vertex of has a neighbor in and that every vertex of has at most neighbors in . For every , either there is a comb with teeth whose blocks satisfy , or .
The constant is an absolute constant obtained from an integral four-adic form of the peeling argument. Its precise value is not used later.
1 import Lax54.GraphDefinitions 2 … module docstring, 15 lines 18 19 namespace Lax54.BipartiteCombLemma 20 21 universe u 22 23 /-- 24 A comb with distinct teeth in `A` and pairwise disjoint blocks in `B`. Each 25 tooth is adjacent to its own block and nonadjacent to every other block. 26 -/ 27 structure CombBetween {V : Type u} [DecidableEq V] 28 (G : SimpleGraph V) (A B : Finset V) (t : ℕ) where 29 tooth : Fin t → V 30 block : Fin t → Finset V 31 tooth_mem : ∀ i, tooth i ∈ A 32 tooth_injective : Function.Injective tooth 33 block_subset : ∀ i, block i ⊆ B 34 blocks_disjoint : ∀ {i j}, i ≠ j → Disjoint (block i) (block j) 35 tooth_adj_block : ∀ i, ∀ x ∈ block i, G.Adj (tooth i) x 36 tooth_nonadj_other : ∀ {i j}, i ≠ j → 37 ∀ x ∈ block j, ¬ G.Adj (tooth i) x 38 39 /-- The sparse alternative in the `d = 1/2` case of Theorem 2.1. -/ 40 def SmallSideBound (C Gamma Delta b : ℕ) : Prop := 41 b ^ 2 ≤ C ^ 2 * Gamma * Delta 42 43 /-- The `d = 1/2` case of Theorem 2.1, with denominators cleared. -/ 44 axiom bipartite_comb_lemma : 45 ∀ {V : Type u} [Fintype V] [DecidableEq V] 46 (G : SimpleGraph V) [DecidableRel G.Adj] 47 (A B : Finset V) (Gamma Delta : ℕ), 48 Disjoint A B → 0 < Gamma → 49 (∀ b ∈ B, ∃ a ∈ A, G.Adj a b) → 50 (∀ a ∈ A, (B.filter fun b ↦ G.Adj a b).card ≤ Delta) → 51 (∃ (t : ℕ) (Cmb : CombBetween G A B t), 52 0 < t ∧ ∀ i : Fin t, Gamma ≤ t ^ 2 * (Cmb.block i).card) ∨ 53 SmallSideBound 128 Gamma Delta B.card 54 55 end Lax54.BipartiteCombLemma 56 -
Anticomplete pairs in sparse -free graphs
A -sparse -free graph on a set of at least two vertices has two anticomplete sets, each of size at least . This is Lemma 4.4 of Nguyen, Scott, and Seymour.
1 import Lax57.GraphDefinitions 2 … module docstring, 9 lines 12 13 namespace Lax57.SparseHouseTools 14 15 open Lax57.GraphDefinitions 16 17 universe u 18 19 /-- The linear anticomplete-pair lemma for sparse `P5`-free graphs. -/ 20 axiom sparse_P5_anticomplete_pair : 21 ∀ {V : Type u} [Fintype V] [DecidableEq V] 22 (G : SimpleGraph V) [DecidableRel G.Adj] (S : Finset V), 23 IsP5Free G → 2 ≤ S.card → ESparse G 32 S → 24 ∃ B : Blockade (V := V) 2, 25 B.IsInside S ∧ B.IsAnticomplete G ∧ 26 ∀ i : Fin 2, S.card ≤ 32 * (B.block i).card 27 28 end Lax57.SparseHouseTools 29 -
no assumptions
Repeatedly choose a large connected component, producing nested connected sets. Sparsity keeps the successive neighborhoods small. If no large anticomplete pair exists, two applications of connectivity across a cut produce five vertices that induce .
-
Polynomial semisparse blockades for the house
This is the denominator-cleared form of Lemma 6.2 of Nguyen, Scott, and Seymour. A sufficiently large house-free graph contains disjoint polynomially large blocks, and every pair of blocks is either complete or has edge density at most .
1 import Lax57.GraphDefinitions 2 … module docstring, 10 lines 13 14 namespace Lax57.SemisparseBlockade 15 16 open Lax57.GraphDefinitions 17 18 universe u 19 20 /-- Polynomial semisparse blockades in house-free graphs. -/ 21 axiom semisparse_house_blockade : 22 ∃ d : ℕ, 40 ≤ d ∧ 23 ∀ E : ℕ, 2 ≤ E → 24 ∀ {V : Type u} [Fintype V] [DecidableEq V] 25 (G : SimpleGraph V) [DecidableRel G.Adj], 26 IsHouseFree G → E ^ (10 * d ^ 2) ≤ Fintype.card V → 27 ∃ B : Blockade (V := V) E, 28 B.IsSemisparse G (E ^ d) ∧ 29 B.HasWidthLoss (E ^ (10 * d ^ 2)) 30 31 end Lax57.SemisparseBlockade 32 -
Start with the one-block layout and choose a maximal layout with fewer than blocks. Its conserved weight identifies a block containing a polynomial fraction of the graph. The local blockade theorem either gives at least blocks at once or refines the layout to a larger one, which must cross the -block threshold. The bound on exceptional ordered edges then gives the required semisparsity.
-
The sparse-house trichotomy
This is a denominator-cleared form of Lemma 7.1 of Nguyen, Scott, and Seymour. At scale , a sparse house-free graph either becomes polynomially sparser on a polynomial fraction of its vertices, contains a long complete blockade, or admits an anticomplete peel. In the last alternative, the set omits at most a fraction of the current vertex set.
1 import Lax57.GraphDefinitions 2 … module docstring, 11 lines 14 15 namespace Lax57.SparseHouseTrichotomy 16 17 open Lax57.GraphDefinitions 18 19 universe u 20 21 /-- Sparse refinement, a complete blockade, or an anticomplete peel. -/ 22 axiom sparse_house_trichotomy : 23 ∃ d : ℕ, 40 ≤ d ∧ 24 ∀ Q : ℕ, 8 ≤ Q → 25 ∀ {V : Type u} [Fintype V] [DecidableEq V] 26 (G : SimpleGraph V) [DecidableRel G.Adj] (S : Finset V), 27 IsHouseFree G → ESparse G Q S → 28 ( (∃ T : Finset V, T ⊆ S ∧ 29 S.card ≤ Q ^ (30 * d ^ 3) * T.card ∧ 30 ESparse G (Q ^ (2 * d)) T) ∨ 31 (∃ B : Blockade (V := V) Q, 32 B.IsInside S ∧ B.IsComplete G ∧ 33 ∀ i : Fin Q, 34 S.card ≤ Q ^ (33 * d ^ 3) * (B.block i).card) ∨ 35 (∃ X Y : Finset V, 36 X ⊆ S ∧ Y ⊆ S ∧ Disjoint X Y ∧ 37 (∀ x ∈ X, ∀ y ∈ Y, ¬ G.Adj x y) ∧ 38 S.card ≤ Q ^ (33 * d ^ 3) * X.card ∧ 39 Q * (S.card - Y.card) ≤ 3 * S.card) ) 40 41 end Lax57.SparseHouseTrichotomy 42 -
Use the prepared blockade from Claim 7.1.1. A vertex mixed on at least a fraction of the blocks cannot meet a complete pair among them, because anticonnectivity would supply four vertices that, together with it, induce a house. Their union gives the sparser outcome. Otherwise, double-counting mixed incidences finds a block with few mixed outside vertices. The original sparsity bound and the size of the prepared blockade leave a large set anticomplete to that block, giving the peel outcome.
-
Prepared sparse-or-complete house blockades
This is the denominator-cleared preparation carried out in Claim 7.1.1 of Nguyen, Scott, and Seymour. Starting from the semisparse blockade of Lemma 6.2, one samples equal-sized subblocks, removes high cross-degree vertices, and takes large anticonnected components. Failure to find such a component already gives the complete-blockade alternative.
1 import Lax57.GraphDefinitions 2 … module docstring, 11 lines 14 15 namespace Lax57.PreparedHouseBlockade 16 17 open Finset 18 open scoped SimpleGraph 19 open Lax57.GraphDefinitions 20 21 universe u 22 23 /-- A very long equal-width blockade whose noncomplete pairs are sparse in 24 both vertexwise directions, or the desired complete blockade already. -/ 25 axiom prepared_house_blockade : 26 ∃ d : ℕ, 40 ≤ d ∧ 27 ∀ Q : ℕ, 8 ≤ Q → 28 ∀ {V : Type u} [Fintype V] [DecidableEq V] 29 (G : SimpleGraph V) [DecidableRel G.Adj] (S : Finset V), 30 IsHouseFree G → Q ^ (30 * d ^ 3) ≤ S.card → 31 ( (∃ C : Blockade (V := V) Q, 32 C.IsInside S ∧ C.IsComplete G ∧ 33 ∀ i : Fin Q, 34 S.card ≤ Q ^ (33 * d ^ 3) * (C.block i).card) ∨ 35 (∃ (m : ℕ) (B : Blockade (V := V) (Q ^ (4 * d))), 36 0 < m ∧ B.IsInside S ∧ 37 (∀ i, m ≤ (B.block i).card ∧ 38 (B.block i).card ≤ 8 * Q ^ 2 * m) ∧ 39 (∀ i, (Gᶜ.induce (B.block i : Set V)).Connected) ∧ 40 (∀ {i j}, i ≠ j → 41 ((∀ x ∈ B.block i, ∀ y ∈ B.block j, G.Adj x y) ∨ 42 (ESparseTo G (Q ^ (4 * d)) (B.block i) (B.block j) ∧ 43 ESparseTo G (Q ^ (4 * d)) (B.block j) (B.block i)))) ∧ 44 Q * (∑ i, (B.block i).card) ≤ S.card ∧ 45 S.card ≤ Q ^ (30 * d ^ 3) * m) ) 46 47 end Lax57.PreparedHouseBlockade 48 -
Apply the semisparse-blockade theorem at scale and sample each block to a common size. Simultaneous cleaning retains at least half of every sample and makes each noncomplete pair sparse in both directions. In each cleaned block, either the complement has a large connected component, or its components yield the required complete -blockade. Choosing a large component in every block preserves the relations between blocks.
-
Simultaneous thinning of a semisparse blockade
This finite greedy lemma is the sampling-and-cleaning step in Claim 7.1.1. Subblocks are selected in order so that every weakly sparse pair has a controlled edge count. Removing vertices of excessive cross-degree then gives vertexwise sparsity in both directions while retaining at least half of each sample.
1 import Lax57.GraphDefinitions 2 … module docstring, 11 lines 14 15 namespace Lax57.BlockadeThinning 16 17 open Lax57.GraphDefinitions 18 19 universe u 20 21 /-- Thin every block simultaneously and convert weak pair density to 22 vertexwise sparsity. -/ 23 axiom semisparse_blockade_thinning : 24 ∀ {V : Type u} [Fintype V] [DecidableEq V] 25 (G : SimpleGraph V) [DecidableRel G.Adj] 26 (k P E t : ℕ) (B : Blockade (V := V) k), 27 2 ≤ k → 2 * k ≤ t → 128 * k ^ 3 * E ≤ P → 28 (∀ i, 2 * t ≤ (B.block i).card) → B.IsSemisparse G P → 29 ∃ C : Blockade (V := V) k, 30 (∀ i, C.block i ⊆ B.block i) ∧ 31 (∀ i, t ≤ 2 * (C.block i).card ∧ (C.block i).card ≤ t) ∧ 32 (∀ {i j}, i ≠ j → 33 ((∀ x ∈ C.block i, ∀ y ∈ C.block j, G.Adj x y) ∨ 34 (ESparseTo G E (C.block i) (C.block j) ∧ 35 ESparseTo G E (C.block j) (C.block i)))) 36 37 end Lax57.BlockadeThinning 38 -
no assumptions
Select vertices from each block in index order. Integer Markov bounds control all pairs involving earlier samples and later blocks. Delete from each sample the vertices whose degree to another sample exceeds . The hypothesis ensures that at least half of every sample remains, and the surviving noncomplete pairs are -sparse in both directions.
-
Sparse-house acceleration
For some integer , a house-free graph that is -sparse on a set , where , either becomes -sparse on a subset of size at least , or has a complete or anticomplete -blockade whose blocks have size at least . This is a reciprocal-square form of Lemma 7.2 of Nguyen, Scott, and Seymour.
1 import Lax57.GraphDefinitions 2 … module docstring, 11 lines 14 15 namespace Lax57.SparseHouseAcceleration 16 17 open Lax57.GraphDefinitions 18 19 universe u 20 21 /-- The sparse-house acceleration step, in reciprocal-square form. -/ 22 axiom sparse_house_acceleration : 23 ∃ d : ℕ, 2 ≤ d ∧ 24 ∀ R : ℕ, 64 ≤ R → 25 ∀ {V : Type u} [Fintype V] [DecidableEq V] 26 (G : SimpleGraph V) [DecidableRel G.Adj] (S : Finset V), 27 IsHouseFree G → ESparse G (R ^ 2) S → 28 (∃ T : Finset V, T ⊆ S ∧ 29 S.card ≤ R ^ (32 * d ^ 3) * T.card ∧ 30 ESparse G (R ^ (2 * d)) T) ∨ 31 (∃ B : Blockade (V := V) R, 32 B.IsInside S ∧ B.IsUniform G ∧ 33 ∀ i : Fin R, 34 S.card ≤ R ^ (36 * d ^ 3) * (B.block i).card) 35 36 end Lax57.SparseHouseAcceleration 37 -
Apply the sparse-house trichotomy at . Its sparse and complete outcomes give the required alternatives after rescaling. In the peel outcome, repeatedly remove the anticomplete piece. Each removal loses at most a fraction of the current set, so after steps at least half of the original set remains. The removed pieces form the required anticomplete -blockade.
-
Restricted set or uniform blockade in a house-free graph
There is an integer such that, for every , every finite house-free graph has one of two outcomes. Either an induced subgraph on is -restricted and , or has a complete or anticomplete blockade of length , where , whose blocks all have size at least .
This is the denominator-cleared form of Lemma 7.3 of Nguyen, Scott, and Seymour. The structural argument is stated for the house, the complement of the five-vertex path.
1 import Lax57.GraphDefinitions 2 … module docstring, 15 lines 18 19 namespace Lax57.HouseDichotomy 20 21 open Lax57.GraphDefinitions 22 23 universe u 24 25 /-- The structural dichotomy for finite house-free graphs. -/ 26 axiom house_dichotomy : 27 ∃ a : ℕ, 1 ≤ a ∧ 28 ∀ E : ℕ, 3 ≤ E → 29 ∀ {V : Type u} [Fintype V] [DecidableEq V] (G : SimpleGraph V) 30 [DecidableRel G.Adj], 31 IsHouseFree G → 32 (∃ X : Finset V, 33 Fintype.card V ≤ E ^ a * X.card ∧ ERestricted G E X) ∨ 34 HasUniformBlockade G E a 35 36 end Lax57.HouseDichotomy 37 -
Rödl's maximum-degree reduction gives a linearly large induced set on which one of the two complementary graphs is -sparse. In the direct orientation, iterate the sparse-house acceleration lemma until the requested parameter is reached. In the complementary orientation, the graph is -free, so the sparse anticomplete-pair lemma gives a complete two-block blockade in the original graph. A single exponent absorbs the fixed and polynomial losses.