Expected reward returned within steps transitions; unfinished paths contribute zero.
Equations
- One or more equations did not get rendered due to their size.
- model.rewardWithin 0 x✝ = match model.kind x✝ with | Determinize.Spec.FiniteModel.StateKind.returned reward => reward | x => 0
Instances For
theorem
Determinize.Proof.FiniteModel.outputWithin_mono
(model : Spec.FiniteModel.Model)
(state : Fin model.size)
:
Monotone fun (steps : ℕ) => model.outputWithin steps state
Increasing the exploration horizon only adds returned output mass.
theorem
Determinize.Proof.FiniteModel.returned_outputWithin
(model : Spec.FiniteModel.Model)
(state : Fin model.size)
(reward : ℚ)
(returned : model.kind state = Spec.FiniteModel.StateKind.returned reward)
(steps : ℕ)
:
theorem
Determinize.Proof.FiniteModel.rejected_outputWithin
(model : Spec.FiniteModel.Model)
(state : Fin model.size)
(rejected : model.kind state = Spec.FiniteModel.StateKind.rejected)
(steps : ℕ)
:
theorem
Determinize.Proof.FiniteModel.returned_output
(model : Spec.FiniteModel.Model)
(reward : ℚ)
(returned : model.kind model.initial = Spec.FiniteModel.StateKind.returned reward)
:
theorem
Determinize.Proof.FiniteModel.rejected_output
(model : Spec.FiniteModel.Model)
(rejected : model.kind model.initial = Spec.FiniteModel.StateKind.rejected)
:
theorem
Determinize.Proof.FiniteModel.integrable_outputWithin
(model : Spec.FiniteModel.Model)
(steps : ℕ)
(state : Fin model.size)
(f : ℝ → ℝ)
:
MeasureTheory.Integrable f (model.outputWithin steps state)
Every finite-horizon output law has finite support.
theorem
Determinize.Proof.FiniteModel.rewardWithin_integral
(model : Spec.FiniteModel.Model)
(steps : ℕ)
(state : Fin model.size)
:
Exact rational reward iteration agrees with the real-measure interpretation.