Depth and approximation loss of balanced recursion
Lax263307.RecursionDepth · concepts/Lax263307/RecursionDepth.lean · lax-263307
No public endorsements yet.
Loading review…
Sign in with ORCIDNatural Language Statement
Theorem
A recursive call on a balanced component has at most vertices. The number of nontrivial calls on a branch is therefore logarithmic in the original graph order. A factor lost at each of at most levels retains at least of a certified upper bound.
Concept map
Evidence
Lean source view on GitHub
| 1 | import Lax263307.WeightedIndependentSet |
| 2 | import Mathlib.Data.Nat.Log |
| 3 | |
| 4 | /-! |
| 5 | --- |
| 6 | title: Depth and approximation loss of balanced recursion |
| 7 | type: theorem |
| 8 | --- |
| 9 | A recursive call on a balanced component has at most |
| 10 | vertices. The number of nontrivial calls on a branch is therefore logarithmic |
| 11 | in the original graph order. A factor lost at each of at most |
| 12 | levels retains at least of a certified upper bound. |
| 13 | |
| 14 | # Formalization notes |
| 15 | |
| 16 | The integer depth function avoids numerical evaluation of logarithms with |
| 17 | base . Empty graphs and singletons have depth zero. The explicit upper |
| 18 | bound uses twice the binary logarithm; this constant change does not affect |
| 19 | the QPTAS bound. |
| 20 | |
| 21 | # Attribution |
| 22 | |
| 23 | Bonnet–Chang, the recursion-depth argument in Lemma 14 and Bernoulli's |
| 24 | inequality in Theorem 15. |
| 25 | -/ |
| 26 | |
| 27 | namespace Lax263307.RecursionDepth |
| 28 | |
| 29 | /-- Number of balanced reductions needed to reach order at most one. -/ |
| 30 | def depth (n : ℕ) : ℕ := |
| 31 | if n ≤ 1 then 0 else 1 + depth (2 * n / 3) |
| 32 | termination_by n |
| 33 | decreasing_by omega |
| 34 | |
| 35 | /-- A balanced recursive call consumes one level. -/ |
| 36 | axiom balanced_decrease (n m : ℕ) (nontrivial : 1 < n) (balanced : 3 * m ≤ 2 * n) : |
| 37 | depth m + 1 ≤ depth n |
| 38 | |
| 39 | /-- At most two reductions are needed per binary-logarithm level. -/ |
| 40 | axiom logarithmic_bound (n : ℕ) : depth n ≤ 2 * (Nat.log2 n + 1) |
| 41 | |
| 42 | /-- A certified independence upper bound turns the accumulated loss into approximation. -/ |
| 43 | axiom approximation_transfer {n : ℕ} (G : SimpleGraph (Fin n)) |
| 44 | (I : Finset (Fin n)) (u levels : ℕ) (δ ε : ℚ) |
| 45 | (independent : G.IsIndepSet (I : Set (Fin n))) |
| 46 | (upper_bound : ∀ J : Finset (Fin n), G.IsIndepSet (J : Set (Fin n)) → J.card ≤ u) |
| 47 | (δ_nonnegative : 0 ≤ δ) (δ_small : δ ≤ 1) |
| 48 | (ε_small : ε ≤ 1) (loss : (levels : ℚ) * δ ≤ ε) |
| 49 | (size : (1 - δ) ^ levels * (u : ℚ) ≤ I.card) : |
| 50 | WeightedIndependentSet.IsApproximation G (fun _ => 1) ε I |
| 51 | |
| 52 | end Lax263307.RecursionDepth |
| 53 |
Formalization notes
The integer depth function avoids numerical evaluation of logarithms with base . Empty graphs and singletons have depth zero. The explicit upper bound uses twice the binary logarithm; this constant change does not affect the QPTAS bound.
Attribution
Bonnet–Chang, the recursion-depth argument in Lemma 14 and Bernoulli's inequality in Theorem 15.
Builds on
From Mathlib
Discussion
Ask a question or add context. Endorsements and structured flags are kept in the review panel above.
0 comments