Theorem 4.7.1 — explicit solution of the auxiliary problem P(lambda,b)
ProvedMDPFinance.MeanVariance.plambdab_solutionWith as in Eq. defining MRcdSeq: a) if , the value functions are
and is the value of ; if ,
the value of is . b) If , the optimal policy
is stationary,
This is 's own Bellman solution via the Structure Assumption (SAN), with value class and feedback class ; it is the building block Proposition 4.7.2 specializes to solve .
Formalization Note. The book's Remark 2.3.14/Theorem 2.5.4 aside — that the strong Integrability Assumption (AN) fails here but the weaker Theorem 2.5.4 still applies — is proof machinery, not part of the numbered statement, and is not restated as a hypothesis.
Formalization Note (moderation). (Assumption (FM)(ii)) is carried: the proof's equivalence for all rests on it.
import Mathlib import Definitions.Def_MDPFinance_MeanVariance_MeanRiskMarket import Definitions.Def_MDPFinance_MeanVariance_PLambdaBValueFunction import Definitions.Def_MDPFinance_MeanVariance_MRcdSeq open MeasureTheory ProbabilityTheory
namespace MDPFinance.MeanVariance
/-- Theorem 4.7.1 (Bäuerle–Rieder, p. 126, PDF 140), under the section's standing assumptions
(`d < 1 < u`, `0 < x0 < μ` in the model; (FM)(ii) `p > q`, `hpq`). With `c_n, d_n` as in
`MRcdSeq`: a) if `c_0 ≥ d_0`, the value functions are `V_n(x) = c_n(x+b)^- - d_n(x+b)^+`, and
`V_0(x_0)` is the value of `P(λ,b)`; if `c_0 < d_0`, the value of `P(λ,b)` is `-∞`. b) if
`c_0 ≥ d_0`, the optimal policy `π^* = (f_0^*,…,f_{N-1}^*)` is stationary,
`f_n^*(x) = f^*(x) := max((x+b)/(1-u), (x+b)/(1-d))`. -/
theorem plambdab_solution {Ω : Type*} [MeasurableSpace Ω] (M : MeanRiskMarket Ω)
(hpq : M.q < M.p) (lam b : ℝ) (hlam : 0 ≤ lam) :
(M.cSeq lam 0 ≥ M.dSeq lam 0 →
(∀ n ≤ M.N, ∀ x : ℝ,
M.VPlambdabFrom lam b n x = ((M.cSeq lam n * max (-(x + b)) 0) -
M.dSeq lam n * max (x + b) 0 : EReal)) ∧
(∃ πstar : ℕ → ℝ → ℝ, M.IsAdmissible 0 πstar ∧
(∀ n < M.N, ∀ x : ℝ, πstar n x = max ((x + b) / (1 - M.u)) ((x + b) / (1 - M.d))) ∧
M.IsOptimalPlambdab lam b πstar)) ∧
(M.cSeq lam 0 < M.dSeq lam 0 → M.VPlambdab lam b = ⊥) := by sorry
end MDPFinance.MeanVariance
Confirmed by the mission captain (proposal self-audit).
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.