Lemma 10.14 — the Dini-derivative bound at regular points (milestone)
ProvedProcessingNetworks.ProportionalFairness.within_group_entropy_dini_bound_regularlyapunov-methodsreal-analysis
Lemma 10.14. At each regular point ,
Specializing Lemma 10.13's bound to a regular point of (where ), the last two terms of (10.56) telescope to (Eq. 10.68), leaving exactly this bound — feeding directly into Lemma 10.9's bound on (mission IX).
Formalization note. Specializing Lemma 10.13's bound to a regular point of (where ), the last two terms of (10.56) telescope to (Eq. 10.68), leaving exactly this bound — feeding directly into Lemma 10.9's bound on (mission IX). is Lipschitz on , as in Lemma 10.13.
Preamble
import Mathlib import Definitions.Def_ProcessingNetworks_ProportionalFairness_WithinGroupEntropy
Formal statement
namespace ProcessingNetworks.ProportionalFairness
/-- Lemma 10.14, Dai & Harrison p. 203 (PDF p. 219): at each regular point `t > 0` of `Z`, the
upper-right Dini derivative of `withinGroupEntropy` is bounded by
`∑_i Ż_i(t) log(Z_i(t)/Y_{grp i}(t))` over classes `i` with `Z_i(t) > 0` (Eq. 10.66). -/
theorem within_group_entropy_dini_bound_regular
{I L : ℕ} (grp : Fin I → Fin L) (Zh : ℝ → Fin I → ℝ)
(hZnn : ∀ t, 0 ≤ t → ∀ i, 0 ≤ Zh t i)
(hZlip : ∃ Kc : ℝ, ∀ i (s t : ℝ), 0 ≤ s → s ≤ t → |Zh t i - Zh s i| ≤ Kc * (t - s))
(t : ℝ) (ht : 0 < t)
(hreg : ∀ i, DifferentiableAt ℝ (fun s => Zh s i) t) :
diniUpperRight (withinGroupEntropy grp Zh) t ≤
((∑ i, if Zh t i = 0 then 0 else
deriv (fun s => Zh s i) t * Real.log (Zh t i / groupAggregate grp (Zh t) (grp i)) : ℝ) :
EReal) := by sorry
end ProcessingNetworks.ProportionalFairness
Source
Dai & Harrison, Processing Networks: Fluid Models and Stability, pre-publication draft 2020-4-2, p. 203, Lemma 10.14, Eq. (10.66)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.