Catalog.Sparsity.Densification

Verification

Lemma

Let \(\preccurlyeq_d\) be as in Definition ShallowMinor.

For every \(\varepsilon \in (0, \frac{2}{3}]\) there exists \(M(\varepsilon) \in \mathbb{N}\), with \(M(\cdot)\) non-increasing, such that: if \(G\) is a graph on \(n \geq M(\varepsilon)\) vertices with minimum degree at least \(n^\varepsilon\), then either

  1. \(G\) contains \(K_t\) as a depth-\(1\) minor for some \(t \geq \log n\), or
  2. there is a depth-\(1\) minor \(G'\) of \(G\) with \(n' \geq n^{1-\varepsilon}\) vertices and at least \((n')^{1 + \varepsilon + \varepsilon^2}\) edges.

Review notes

The proof uses a probabilistic argument (Chernoff’s bound) to find a random subset \(A\) with good properties, then a greedy contraction process. The key output is that either a large clique minor is found, or edge density is amplified from \(n^\varepsilon\) (minimum degree) to \((n')^{\varepsilon + \varepsilon^2}\) in a depth-1 minor \(G'\). The exponent increase from \(\varepsilon\) to \(\varepsilon + \varepsilon^2\) is what drives the iteration in the proof of Theorem 3.1.

Lean encoding.

  • Minimum degree is encoded pointwise: \(\forall v,\; n^\varepsilon \leq \texttt{G.degree}\;v\), comparing \(\mathbb{R}\)-valued rpow against a cast of the natural-number degree.
  • \(\log n\) uses Real.log (natural logarithm). The source does not specify the base; the constant \(M\) absorbs any base choice.
  • The non-increasing property of \(M(\varepsilon)\) is not encoded; it is not needed downstream.
  • The existential over \(G'\) quantifies over a new vertex type \(W\) with all necessary instances (DecidableEq, Fintype, DecidableRel).

Catalog/Sparsity/Densification/Contract.lean (full)

1import Catalog.Sparsity.ShallowMinor.Contract
2import Mathlib.Combinatorics.SimpleGraph.Finite
3import Mathlib.Analysis.SpecialFunctions.Log.Basic
4import Mathlib.Analysis.SpecialFunctions.Pow.Real
5
6open Catalog.Sparsity.ShallowMinor
7
8namespace Catalog.Sparsity.Densification
9
10/-- Lemma 3.4/ch1: Densification lemma.
11
12 For every ε ∈ (0, 2/3], there exists a threshold M such that every graph G
13 on n ≥ M vertices with minimum degree ≥ n^ε either:
14 (1) contains K_t as a depth-1 minor for some t ≥ log n, or
15 (2) has a depth-1 minor G' with n' ≥ n^{1-ε} vertices and
16 ≥ (n')^{1+ε+ε²} edges. -/
17axiom densification (ε : ℝ) (hε_pos : 0 < ε) (hε_le : ε ≤ 2 / 3) :
18 ∃ M : ℕ, ∀ {V : Type} [DecidableEq V] [Fintype V]
19 (G : SimpleGraph V) [DecidableRel G.Adj],
20 M ≤ Fintype.card V →
21 (∀ v : V, (Fintype.card V : ℝ) ^ ε ≤ ↑(G.degree v)) →
22 (∃ t : ℕ, Real.log ↑(Fintype.card V) ≤ ↑t ∧
23 IsShallowMinor (SimpleGraph.completeGraph (Fin (t + 1))) G 1) ∨
24 (∃ (W : Type) (_ : DecidableEq W) (_ : Fintype W)
25 (G' : SimpleGraph W) (_ : DecidableRel G'.Adj),
26 IsShallowMinor G' G 1
27 (Fintype.card V : ℝ) ^ (1 - ε) ≤ ↑(Fintype.card W) ∧
28 (Fintype.card W : ℝ) ^ (1 + ε + ε ^ 2) ≤ ↑(G'.edgeFinset.card))
29
30end Catalog.Sparsity.Densification
31