Theorem 6.5 — existence of fluid limits (milestone)
DisprovedProcessingNetworks.FluidStability.fluid_limit_existenceTheorem 6.5 (existence of fluid limits). Fix a sample point at which the SLLNs (2.14) (for external arrivals) and (2.15) (for processing variables) and the negligibility condition (6.38) all hold. For any unbounded set of initial states, there is a sequence with such that
for some and ; furthermore, satisfies the fluid equations (6.1)-(6.6) with .
This is what makes fluid limit paths (Definition 6.6) a nonempty, well-behaved family: not only do they exist along every unbounded sequence of initial states, but every one of them is automatically a fluid model solution, which is the bridge Theorem 6.2 needs.
Formalization note. is a normalization coming from the choice of
scaling sequence, not an added restriction on the SPN — it is stated here as part of the
conclusion, exactly as the book states it. Continuous Dh ∧ Continuous Fh ∧ Continuous Th ∧ Continuous Zh makes explicit the theorem's own " and " clause. The SLLNs
(2.14), (2.15) and the condition (6.38) are hypotheses on the fixed sample point , for
the core stochastic elements that the standard setup SPNProcessFamily is
built from; the fluid equations (6.1)–(6.6) for the limit follow from the system relations of
Section 2.5 satisfied by every version of the family.
import Mathlib import Definitions.Def_ProcessingNetworks_Stability_MarkovRepresentation import Definitions.Def_ProcessingNetworks_FluidStability_FluidEquationData import Definitions.Def_ProcessingNetworks_Stability_SPNModel import Definitions.Def_ProcessingNetworks_FluidStability_SPNProcessFamily
namespace ProcessingNetworks.FluidStability
open MeasureTheory ProbabilityTheory Filter ProcessingNetworks.Stability
/-- Theorem 6.5, Dai & Harrison p. 116 (PDF p. 132): fix `ω` such that the SLLNs (2.14), (2.15)
for the core stochastic elements and the negligibility condition (6.38) all hold. For any
unbounded set `C ⊂ 𝒳` of initial states, there is a sequence `{xₙ} ⊂ C` with `|xₙ| → ∞` such
that the fluid-scaled processes (6.37) converge u.o.c. to some continuous
`(D̂, F̂, T̂, Ẑ) ∈ C^I × C^J × C^J × C^I` (6.39), and furthermore `(D̂, F̂, T̂, Ẑ)` satisfies the
fluid equations (6.1)-(6.6) with `|Ẑ(0)| = 1`. -/
theorem fluid_limit_existence
{Xstate : Type*} [Countable Xstate] {Ω : Type*} [MeasureSpace Ω] {I J K : ℕ}
{N : ℝ → Ω → Fin J → ℕ} {Z : ℝ → Ω → Fin I → ℕ}
{Mrep : MarkovRepresentation Xstate I J N Z} {sd : SPNData I J K}
{dat : FluidEquationData I J K} {E : Fin I → ℝ → Ω → ℕ} {v : Fin J → ℕ → Ω → ℝ}
{φ : Fin J → ℕ → Ω → Fin I → ℕ} (fam : SPNProcessFamily Mrep sd dat E v φ)
(E : Fin I → ℝ → Ω → ℕ) (v : Fin J → ℕ → Ω → ℝ) (φ : Fin J → ℕ → Ω → Fin I → ℕ) (ω : Ω)
(C : Set Xstate) (hC : ¬ BddAbove ((spnSize Mrep) '' C))
(h214 : ∀ i, Tendsto (fun t : ℝ => (E i t ω : ℝ) / t) atTop (nhds (dat.lam i)))
(h215 :
∀ j, Tendsto (fun n : ℕ => (∑ ℓ ∈ Finset.range n, v j ℓ ω) / n) atTop (nhds (dat.m j)) ∧
∀ i, Tendsto (fun n : ℕ => (∑ ℓ ∈ Finset.range n, (φ j ℓ ω i : ℝ)) / n) atTop
(nhds (dat.Γ i j)))
(h638 :
∀ j, Tendsto (fun n : ℕ => (n : ℝ)⁻¹ * ⨆ ℓ ∈ Finset.range n, v j ℓ ω) atTop
(nhds 0)) :
∃ (x : ℕ → Xstate) (Dh : ℝ → Fin I → ℝ) (Fh Th : ℝ → Fin J → ℝ) (Zh : ℝ → Fin I → ℝ),
(∀ n, x n ∈ C) ∧ Tendsto (fun n => spnSize Mrep (x n)) atTop atTop ∧
Continuous Dh ∧ Continuous Fh ∧ Continuous Th ∧ Continuous Zh ∧
UOCConverges
(fun n t i => (spnSize Mrep (x n))⁻¹ * (fam.D (x n) (spnSize Mrep (x n) * t) ω i : ℝ)) Dh ∧
UOCConverges
(fun n t j => (spnSize Mrep (x n))⁻¹ * (fam.F (x n) (spnSize Mrep (x n) * t) ω j : ℝ)) Fh ∧
UOCConverges
(fun n t j => (spnSize Mrep (x n))⁻¹ * fam.T (x n) (spnSize Mrep (x n) * t) ω j) Th ∧
UOCConverges
(fun n t i => (spnSize Mrep (x n))⁻¹ * (fam.Zx (x n) (spnSize Mrep (x n) * t) ω i : ℝ)) Zh ∧
IsFluidModelSolution dat Dh Fh Th Zh ∧ (∑ i, Zh 0 i) = 1 := by sorry
end ProcessingNetworks.FluidStability
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.