Theorem 1 — the fraction of abandoning customers tends to
ProvedErlangA.Abandonment.theorem_1Consider a sequence of Erlang-A queues indexed by the number of agents : the -th system has arrival rate , service rate (the same for all ) and patience rate , and its traffic intensity is
Assume, as in §4 of the paper, that and with . Let be the probability that a customer arriving to the -th system in steady state abandons. If for some , then
with .
Thus, to leading order, the fraction of abandoning callers does not depend on the patience distribution's rate: a critically loaded center () already has vanishing abandonment, while an overloaded one sheds exactly the excess load.
Formalization Note The limit is an extended nonnegative real, and is convergence of in ; for it means . The value is computed in and converted to a real number, which gives at . The hypotheses and are the standing assumptions of §4 and are kept as the paper states them; positivity of , is imposed for .
import Mathlib import Definitions.Def_ErlangA_Abandonment_Model open Filter Topology
namespace ErlangA.Abandonment
/-- **Theorem 1**, p. 215. Standing assumptions of §4 (p. 215): `μ_N ≡ μ > 0`, `λ_N > 0` with
`λ_N → ∞`, `θ_N > 0` with `θ_N → θ ∈ (0, ∞)`. If `ρ_N = λ_N/(Nμ) → ρ_∞ ∈ [0, ∞]` (convergence
in `ℝ≥0∞`), then `P_N{Ab} → 0` when `ρ_∞ ≤ 1` and `P_N{Ab} → 1 − 1/ρ_∞` when `ρ_∞ > 1`
(with `1 − 1/∞ = 1`). -/
theorem theorem_1 (μ θ : ℝ) (lam θN : ℕ → ℝ) (hμ : 0 < μ)
(hlam : ∀ N : ℕ, 1 ≤ N → 0 < lam N) (hlam_infty : Tendsto lam atTop atTop)
(hθN : ∀ N : ℕ, 1 ≤ N → 0 < θN N) (hθ : 0 < θ) (hθlim : Tendsto θN atTop (𝓝 θ))
(ρinf : ENNReal)
(hρ : Tendsto (fun N : ℕ => ENNReal.ofReal (lam N / ((N : ℝ) * μ))) atTop (𝓝 ρinf)) :
Tendsto (fun N : ℕ => probAbandon N (lam N) μ (θN N)) atTop
(𝓝 (if ρinf ≤ 1 then 0 else (1 - ρinf⁻¹).toReal)) := by sorry
end ErlangA.Abandonment
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.