Counting induced three-vertex patterns in uniform triples of cells #
If three cells A 0, A 1, A 2 of a graph G are pairwise ε-uniform, then for every graph
J on Fin 3 the number of maps g with g i ∈ A i that induce J (edges of J go to edges of
G and nonedges to nonedges) is the count predicted by independence,
|A 0| |A 1| |A 2| ∏_{i < j} d_{ij}, up to an error 3 ε |A 0| |A 1| |A 2|. Here d_{ij} is the
edge density of (A i, A j) when i and j are adjacent in J and the nonedge density
1 - d(A i, A j) otherwise. Since nonedges are counted too, this predicts every induced
pattern, not only edge-preserving copies.
The argument has two steps.
- A
ε-uniform pair(s, t)has small rectangle discrepancy: the number of edges betweens' ⊆ sandt' ⊆ tis withinε |s| |t|ofd(s, t) |s'| |t'|. Large rectangles are controlled by uniformity itself, and small ones because both counts are then at most|s'| |t'| ≤ ε |s| |t|. Testing against functions with values in[0, 1]instead of indicator functions does not increase the discrepancy (SimpleGraph.IsUniform.abs_sum_sum_mul_sub_le). - Replacing the three pair indicators by their densities one at a time, every step is a test of
one pair against two functions with values in
[0, 1], summed over the third vertex.
The cells may overlap and need not be distinct; when they are pairwise disjoint, every counted map is injective. Summed over the triples of distinct parts of a uniform partition, this gives the contribution of transversal maps in counting lemmas for regular partitions, such as the induced counting lemma for strong graph regularity.
Main results #
SimpleGraph.abs_card_induced_fin_three_sub_le: the induced counting lemma for pairwise uniform triples of cells.
References #
- Y. Zhao, Graph Theory and Additive Combinatorics, Cambridge University Press (2023), Chapter 4 (the graph counting lemma).
Induced counting lemma for three cells. Let A 0, A 1, A 2 be pairwise ε-uniform
cells of G, and J a graph on Fin 3. Then the number of maps g with g i ∈ A i such that
g i and g j are adjacent in G exactly when i and j are adjacent in J differs from
|A 0| |A 1| |A 2| times the product, over the pairs i < j, of the edge density of
(A i, A j) (if i and j are adjacent in J) or of its nonedge density 1 - d(A i, A j)
(otherwise), that is SimpleGraph.patternDensity, by at most 3 ε |A 0| |A 1| |A 2|.