LeanPool.AsymptoticTrianglePacking.Internal — deterministic pruning counts #
Each nibble round produces a small set B of vertices whose safe degree left the target band; the
tight-band invariant is restored by DELETING them (and the edges through them). Deleting B costs
every other vertex the edges it shares with B, and this file bounds that cost:
card_edges_meeting_le— at most|B|·Δedges meetB;card_heavyLoss_le— at mostr·|B|·Δ/tvertices lose≥ tedges to the deletion;degree_prune_ge— the degree in the pruned hypergraph is the old degree minus the lost edges.
Pure Finset combinatorics, no probability. placeholder-free and axiom-clean
[propext, Classical.choice, Quot.sound].
def
LeanPool.AsymptoticTrianglePacking.Internal.lostDegree
{V : Type u_1}
[DecidableEq V]
(H : Finset (Finset V))
(B : Finset V)
(v : V)
:
The edges lost at v when the vertex set B is deleted.
Equations
Instances For
def
LeanPool.AsymptoticTrianglePacking.Internal.prune
{V : Type u_1}
[DecidableEq V]
(H : Finset (Finset V))
(B : Finset V)
:
The hypergraph with all edges meeting B removed.
Equations
- LeanPool.AsymptoticTrianglePacking.Internal.prune H B = {e ∈ H | Disjoint e B}
Instances For
theorem
LeanPool.AsymptoticTrianglePacking.Internal.degree_prune_ge
{V : Type u_1}
[DecidableEq V]
{H : Finset (Finset V)}
(B : Finset V)
(v : V)
:
Pruning splits the degree.
theorem
LeanPool.AsymptoticTrianglePacking.Internal.card_heavyLoss_le
{V : Type u_1}
[DecidableEq V]
[Fintype V]
{H : Finset (Finset V)}
{r Δ : ℕ}
(hr : Hypergraph.IsUniform H r)
(hΔ : ∀ (x : V), Hypergraph.degree H x ≤ Δ)
(B : Finset V)
(t : ℕ)
:
Few vertices lose many edges. At most r·|B|·Δ/t vertices lose ≥ t edges when B is
deleted.