Lemma 3.3: every edge of has a unique generating seed, and
ProvedHiddenPartition.lemma_3_3combinatoricscomplexity-theorygraph-theoryhardness-of-approximationp2o-gran-per-chapterp2o-plan-paperp2o-v1
Throughout, is a finite simple graph whose vertex set is partitioned into non-empty independent domains , and are finite non-empty sets with , and is a family of maps with copy maps , images and target graph on (Section 3.1).
Let and let be -locally coherent. Then
- every edge of is a copy of an edge of generated by a unique seed: for every edge of there is exactly one with for some edge of ;
- for every , the map is an isomorphism
Thus every copy of in the construction is an induced subgraph of , and edges are never shared between copies.
Differences from the source:
- encoding: standing assumptions of §3.1 (domain partition of G; A, B finite nonempty with |A| = |B| = M) → binders IsDomainPartition G dom, [Fintype A] [Nonempty A] [Fintype B], Fintype.card A = Fintype.card B; reason: the section's standing assumptions written as hypotheses; same instances
- encoding: domains 1, …, k and vertex set [k] × B of Z → domains Fin k = {0, …, k−1}, vertex type Fin k × B, family f : V → A → B; reason: Lean indexing convention; same instances
Preamble
import Mathlib import Definitions.Def_HiddenPartition_IsDomainPartition import Definitions.Def_HiddenPartition_IsLocallyCoherent set_option autoImplicit false
Formal statement
namespace HiddenPartition
/-- **Lemma 3.3** (Agrawal, Eua-anant, Haeupler, Laekhanukit, Mihov, Milenković, Santos, Zhang,
*From One Hidden Subgraph to Dense Graph Partitions with Applications to Hardness of Clustering*, STOC 2027
submission (manuscript sha256 17239d0b…), p.15 (pdf p.17)).
For `t ≥ 2` (and `F` a `t`-locally coherent family), every edge of `Z` is a copy of an edge in
`G` generated by a unique seed. Moreover, for every `a ∈ A`, the map `ϕ_a` is an isomorphism
from `G` to `Z[W_a]`.
**Formalization Note.** The standing assumptions of §3.1 (p.12) are binders: `G` is a
simple graph on a finite vertex type with a partition into `k` non-empty independent domains
(`IsDomainPartition`), `A` and `B` are finite non-empty sets with `|A| = |B|`, and
`f v = f_v`. `Z` is `target G dom f`, with vertex set `Fin k × B`.
"Generated by a unique seed" is `∃! a, Generates G dom f a s(x, y)`; "`ϕ_a` is an isomorphism
onto `Z[W_a]`" is a graph isomorphism `G ≃g Z[W_a]` whose underlying map is `ϕ_a`. -/
theorem lemma_3_3 {V A B : Type} [Fintype V] [DecidableEq V] [Fintype A] [Nonempty A]
[Fintype B] [DecidableEq B] {k : ℕ} (G : SimpleGraph V) (dom : V → Fin k)
(hdom : IsDomainPartition G dom) (hAB : Fintype.card A = Fintype.card B)
(f : V → A → B) (t : ℕ) (ht : 2 ≤ t)
(hf : IsLocallyCoherent t G dom f) :
(∀ x y, (target G dom f).Adj x y → ∃! a, Generates G dom f a s(x, y)) ∧
(∀ a, ∃ e : G ≃g (target G dom f).induce (W dom f a : Set (Fin k × B)),
∀ v, (e v : Fin k × B) = phi dom f a v) := by sorry
end HiddenPartition
Source
H. Agrawal, P. Eua-anant, B. Haeupler, B. Laekhanukit, P. Mihov, L. Milenković, R. Santos, Z. Zhang, From One Hidden Subgraph to Dense Graph Partitions with Applications to Hardness of Clustering, STOC 2027 submission (manuscript sha256 17239d0bf911912040cb82c45ae979b179c419fb0b9e69d6d1413bf70068b601), Lemma 3.3, p.15 (pdf p.17)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.