Remark 1, bullets 1–2, p. 57 (corrected: ∂R/∂u continuous in u for continuous densities) — R_i, R, µ_i, r and ∂/∂u R are continuous in σ and u
ProvedDriverSurge.IC.remark_1_continuityLet the trip-length laws have bounded densities and finite means, and let the payments be continuous on and -integrable there. Measure the distance between policies by . Then, for each state :
- , and are continuous in ;
- is continuous in for fixed , and continuous in for fixed ;
- the marginal gain
(with respect to state 's policy) is continuous in for fixed , and, if moreover the densities are continuous on , continuous in for fixed .
These continuity properties are hypotheses 1–3 of Lemma 5, which turns the sign analysis of into the shape of the optimal policy.
Formalization Note Lean state 0 is the paper's non-surge state 1 and Lean state 1 the surge state 2; other i is the state . Item 3 is stated for every function that is the marginal gain of the reward at every policy and every .
Differences from the source:
- correction: continuous in under bounded densities → continuity in claimed when the densities are continuous on ; continuity in claimed unconditionally; reason: by Lemma 6 the derivative is , so it jumps in where jumps and : with all rates , uniform on , , (state 2 dominates: ) and accept-all, the marginal gain is at (no mass to the right) but tends to as (, )
- reading: "continuous in " → continuity with respect to ; reason: the metric used in the proof of Lemma 5 (p. 52: )
- encoding: "given assumptions on , " → bounded densities, finite means, continuous on and -integrable; reason: D.3.1's standing assumption and p. 5's continuity; finite means and integrability exclude the cases where , are undefined
- generalization: asymptotically sub-linear (p. 5) → not assumed; reason: the remark does not use it; integrability already makes the rates finite
- generalization: D.3.1's standing assumption (p. 56) that state 2 dominates state 1 ( with for all ) → not assumed; reason: the continuity properties do not involve it; they hold for every , so they specialize to the source under the assumption
import Mathlib import Definitions.Def_DriverSurge_IC_Pricing
namespace DriverSurge.IC
open MeasureTheory Set
/-- Remark 1, bullets 1–2 (p. 57; corrected: the continuity of `∂/∂u R(w, σ)` in `u` is claimed when
the densities are continuous on `(0, ∞)`). Let `F_i` have bounded densities and finite means, and let
the payments `w_i` be continuous on `(0, ∞)` and `F_i`-integrable there. Continuity in `σ` is with
respect to `F_1(σ_1 △ σ'_1) + F_2(σ_2 △ σ'_2)` over policies.
1. `R_i(σ)`, `R(σ)` and `µ_i(σ)` are continuous in `σ`;
2. `r(u, i, w, σ)` is continuous in `u > 0` for fixed `σ` and continuous in `σ` for fixed `u`;
3. every marginal gain `∂/∂u R(w, σ)` (with respect to state `i`'s policy) is continuous in `σ` for
fixed `u > 0`, and, if the densities are continuous on `(0, ∞)`, continuous in `u > 0` for fixed
`σ`. -/
theorem remark_1_continuity (P : Primitives) (f : Fin 2 → ℝ → ℝ) (hf : HasBoundedDensity P f)
(hmean : FiniteMean P) (w : Fin 2 → ℝ → ℝ) (hwc : ∀ k, ContinuousOn (w k) (Ioi 0))
(hwi : ∀ k, IntegrableOn (w k) (Ioi 0) (P.F k)) (i : Fin 2) :
ContinuousInPolicy P (fun σ => stateRate P i (w i) (σ i)) ∧
ContinuousInPolicy P (fun σ => rewardCF P w σ) ∧
ContinuousInPolicy P (fun σ => mu P σ i) ∧
(∀ σ, IsPolicy σ → ContinuousOn (rFun P w σ i) (Ioi 0)) ∧
(∀ u : ℝ, 0 < u → ContinuousInPolicy P (fun σ => rFun P w σ i u)) ∧
(∀ D : (Fin 2 → Set ℝ) → ℝ → ℝ,
(∀ σ, IsPolicy σ → ∀ u : ℝ, 0 < u → HasMarginalGain (rewardOn P w σ i) (σ i) u (D σ u)) →
(∀ u : ℝ, 0 < u → ContinuousInPolicy P (fun σ => D σ u)) ∧
((∀ k, ContinuousOn (f k) (Ioi 0)) → ∀ σ, IsPolicy σ → ContinuousOn (D σ) (Ioi 0))) := by sorry
end DriverSurge.IC
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.