Theorem 3.2 — sufficient condition for an open-loop equilibrium (mean–variance module conventions)
OpenTimeInconsLQ.MeanVariance.theorem_3_2This is the sufficient condition for an open-loop equilibrium of the time-inconsistent stochastic LQ problem (Hu–Jin–Zhou, Theorem 3.2), stated in the conventions of the mean–variance modules (Model, Adjoint) so that it can be applied to the mean–variance instance of §5.
Let be a standard -dimensional Brownian motion with its natural filtration , horizon , and let the data satisfy the standing assumptions of §2. Let and let be a state process of , i.e. a solution of
Suppose that for every there is a solution on of the first-order adjoint equation
such that
satisfies condition (3.4): and almost surely. Then is an equilibrium control in the sense of Definition 2.1.
Combined with an explicit solution of the adjoint equation and a verification of (3.4), this theorem is how the explicit equilibria of §§4–5 (in particular Theorem 5.4) are established.
Formalization Note. The data, state, cost, spike variation and Definition 2.1 are those of Definitions.Def_TimeInconsLQ_MeanVariance_Model; the adjoint equation on (IsFirstAdjointAt), (Lam) and condition (3.4) at one (Cond34At) are those of Definitions.Def_TimeInconsLQ_MeanVariance_Adjoint. The adjoint pairs are given as families , constrained only for . The same theorem in the conventions of the Sufficient modules is TimeInconsLQ.Sufficient.theorem_3_2; the two versions differ in how (3.4)(b) is encoded (here the version of is only required to be jointly measurable).
import Mathlib import Definitions.Def_Peng1990_SMP_Stochastic import Definitions.Def_TimeInconsLQ_MeanVariance_Model import Definitions.Def_TimeInconsLQ_MeanVariance_Adjoint
namespace TimeInconsLQ.MeanVariance
open MeasureTheory ProbabilityTheory Filter Topology
open scoped ENNReal NNReal Matrix
/-- Theorem 3.2 (p. 7), in the conventions of the module `Model`/`Adjoint`: if `u ∈ L²_𝓕(0, T; ℝˡ)`
has a state process `X` and, for every `t ∈ [0, T)`, the first adjoint equation (3.1) along `X`
has a solution `(p(·; t), k(·; t))` on `[t, T]` such that
`Λ(·; t) = B p(·; t) + Σⱼ (Dʲ)′ kʲ(·; t) + R u` satisfies condition (3.4), then `u` is an
equilibrium control (Definition 2.1). -/
theorem theorem_3_2 {Ω : Type*} [MeasurableSpace Ω] {n l d : ℕ} (M : Data Ω n l d)
(hM : M.Standing) (u : ℝ≥0 → Ω → Fin l → ℝ) (X : ℝ≥0 → Ω → Fin n → ℝ)
(p : ℝ≥0 → ℝ≥0 → Ω → Fin n → ℝ) (k : ℝ≥0 → Fin d → ℝ≥0 → Ω → Fin n → ℝ)
(hu : M.Admissible u) (hX : M.IsState u X)
(hp : ∀ t : ℝ≥0, t < M.T → M.IsFirstAdjointAt X t (p t) (k t))
(h34 : ∀ t : ℝ≥0, t < M.T → M.Cond34At t (M.Lam u (p t) (k t))) :
M.IsEquilibrium u := by sorry
end TimeInconsLQ.MeanVariance