§3.2, pp. 552–553 — the phase LP for snoopy caching has minimum
ProvedNonuniformCompetitive.Snoopy.lp_lower_boundcompetitive-analysislinear-programmingp2o-batch-p200bp2o-gran-per-chapterp2o-plan-paperp2o-v1
Let be an integer. Let be real numbers with , and let be a real number such that for every ,
Then
In the paper, is the probability that a phase-based algorithm has made the block private to the active processor before the -th write of a phase, the left-hand side is its expected cost on a phase with writes, and is the optimal cost of such a phase. The result is the LP lower bound of the proof of Theorem 4: no phase-based algorithm has an LP bound below .
Formalization Note As in the paper, the are free real variables: no constraint is imposed, which makes the statement stronger. The sequence is a function ; its values at and above are unused.
Preamble
import Mathlib import Definitions.Def_NonuniformCompetitive_Snoopy_ep
Formal statement
namespace NonuniformCompetitive.Snoopy
/-- The phase LP of Karlin–Manasse–McGeoch–Owicki (Algorithmica 11 (1994), §3.2, pp. 552–553)
has minimum `e_p / (e_p − 1)`. For a block-transfer cost `p ≥ 1`, let `π₁, …, π_{p+1}` be real
numbers (free variables: no sign or box constraints) with `π_{p+1} = 1`, and let `α` be a real
number such that for every `k = 0, …, p`,
`π_{k+1} · p + ∑_{i=1}^{k} (1 − π_i) ≤ α · k`.
Then `α ≥ e_p / (e_p − 1)`. The values `π 0` and `π k` for `k > p + 1` are unused. -/
theorem lp_lower_bound (p : ℕ) (hp : 1 ≤ p) (π : ℕ → ℝ) (α : ℝ)
(hπ : π (p + 1) = 1)
(hcon : ∀ k ≤ p, π (k + 1) * p + ∑ i ∈ Finset.Icc 1 k, (1 - π i) ≤ α * k) :
ep p / (ep p - 1) ≤ α := by sorry
end NonuniformCompetitive.Snoopy
Source
Karlin, Manasse, McGeoch, Owicki, Competitive Randomized Algorithms for Nonuniform Problems, Algorithmica 11 (1994), pp. 552–553, §3.2 (proof of Theorem 4), displays for E C_A(σ_k), E C_A(σ_k) ≤ α·k and α = e_p/(e_p − 1)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.