Compactness of deterministic k-server cost bounds under finite testing
ProvedKServer.online_cost_compactnessFix a metric space , a natural number , an initial configuration of labeled servers, and any real-valued function on finite request sequences. Suppose every finite family of request sequences admits a deterministic online algorithm , starting at , with
Then there is a single deterministic online algorithm , starting at , such that
Every algorithm in the statement satisfies the service constraint at all histories. No compactness, boundedness, or finiteness of the metric space is assumed. The conclusion has exactly the same cost bound, with no approximation loss.
This permits the construction of one global policy from compatible finite cost tests, without requiring the separate test policies themselves to agree. In particular, it applies to competitive bounds with one prescribed additive constant. The assertion covers all natural , including degenerate cases; any existence required is supplied by the finite-family hypothesis.
import Definitions.Def_KServer_model open KServer
theorem KServer.online_cost_compactness (k : ℕ) (M : Type) [MetricSpace M]
(C₀ : Config k M) (b : List M → ℝ)
(h : ∀ F : Finset (List M), ∃ A : OnlineAlgorithm k M,
A.conf [] = C₀ ∧ ∀ σ ∈ F, A.cost σ ≤ b σ) :
∃ A : OnlineAlgorithm k M, A.conf [] = C₀ ∧
∀ σ : List M, A.cost σ ≤ b σ := by sorry