Source lean/Determinize/Proof/RewardModel/Control.lean
import Determinize.Spec.RewardModel.Modelnamespace Determinize.Proof.RewardModelopen Spec.RewardModeltheorem controlWeight_nonnegative (model : Model) (i j : Fin model.size) :    0 ≤ controlWeight model i j := by  apply List.sum_nonneg  intro x hx  obtain ⟨e, he, rfl⟩ := List.mem_map.mp hx  split_ifs  · exact model.nonnegative i e he  · rflprivate theorem sum_weights {n : Nat} (edges : List (Edge n)) :    (∑ j : Fin n, (edges.map fun e ↦ if e.target = j then e.probability else 0).sum) =      (edges.map Edge.probability).sum := by  induction edges with  | nil => simp  | cons e es ih =>    simp only [List.map_cons, List.sum_cons, Finset.sum_add_distrib, ih]    simptheorem controlWeight_normalized (model : Model) (i : Fin model.size) :    ∑ j, controlWeight model i j = 1 := by  unfold controlWeight  rw [sum_weights, model.normalized]theorem edge_le_controlWeight (model : Model) (i : Fin model.size) (e : Edge model.size)    (member : e ∈ model.edges i) : e.probability ≤ controlWeight model i e.target := by  apply List.single_le_sum  · intro x hx    obtain ⟨a, ha, rfl⟩ := List.mem_map.mp hx    split_ifs    · exact model.nonnegative i a ha    · rfl  · exact List.mem_map.mpr ⟨e, member, by simp⟩theorem sum_controlWeight_mul (model : Model) (i : Fin model.size) (v : Fin model.size → Rat) :    (∑ j, controlWeight model i j * v j) =      ((model.edges i).map fun e ↦ e.probability * v e.target).sum := by  unfold controlWeight  generalize model.edges i = edges  induction edges with  | nil => simp  | cons e es ih =>    simp only [List.map_cons, List.sum_cons, add_mul, Finset.sum_add_distrib, ih]    congr 1    simptheorem sum_controlWeight_mul_real (model : Model) (i : Fin model.size) (v : Fin model.size → ℝ) :    (∑ j, (controlWeight model i j : ℝ) * v j) =      ((model.edges i).map fun e ↦ (e.probability : ℝ) * v e.target).sum := by  unfold controlWeight  generalize model.edges i = edges  induction edges with  | nil => simp  | cons e es ih =>    simp only [List.map_cons, List.sum_cons, Rat.cast_add, add_mul, Finset.sum_add_distrib, ih]    congr 1    have cast (j : Fin model.size) : ((if e.target = j then e.probability else 0 : Rat) : ℝ) =        if e.target = j then (e.probability : ℝ) else 0 := by split_ifs <;> simp    simp_rw [cast]    simpend Determinize.Proof.RewardModelnamespace Determinize.Spec.RewardModelabbrev Model.control (model : Model) : FiniteModel.Model where  size := model.size  initial := model.initial  kind := model.kind  transition := controlWeight model  nonnegative := Proof.RewardModel.controlWeight_nonnegative model  normalized := Proof.RewardModel.controlWeight_normalized modelend Determinize.Spec.RewardModel