BanditRLlib
Lean gate passed before this site build; local proof declarations are shown as compiled.Lean-verified build · exact declarations linked.

Lean module · Foundations

BanditRLProof.Algorithms.HOOPathComparison

Actual HOO search-path comparison with a supremum-preserving branch.

Module map

Declarations
3
Placeholders
0

Imports

BanditRLProof.HOOOptimalBranch, BanditRLProof.Algorithms.HOOPrefix

Imported by

BanditRLProof.Algorithms.HOOSelectionTail

Declarations

Open an item to read its exact compact statement and source link. Detailed teaching notes are linked when registered.

theorem BanditRLProof.HOO.Covering.branch_root_optimistic Compiled

No declaration docstring is present; use the chapter context and exact statement below.

Used in these reading views: Bandit Book

1. Finite bandits, traces, and regret

Canonical node identitydeclaration:BanditRLProof.HOO.Covering.branch_root_optimistic

Reading membership is not a proof dependency. Exact assumptions remain in the Lean statement.

theorem Covering.branch_root_optimistic {X : Type*} (C : Covering X) (f : X → ℝ) (S : Finset Node) (U : Node → WithTop ℝ) (best : ℝ) (k : ℕ) (hk : C.optimalPath f k ∉ S) (hpre : ∀ j < k, C.optimalPath f j ∈ S) (hU : ∀ j < k, (best : WithTop ℝ) ≤ U (C.optimalPath f j)) : (best : WithTop ℝ) ≤ bValue S U []
theorem BanditRLProof.HOO.Covering.selected_underestimate_implies_branch_underestimate Compiled

If an expanded region lies on the actual selected path but has U below `best`, some node of the comparison branch has U below `best`. The branch need only be inspected through the finite tree's maximum depth.

Used in these reading views: Bandit Book

1. Finite bandits, traces, and regret

Canonical node identitydeclaration:BanditRLProof.HOO.Covering.selected_underestimate_implies_branch_underestimate

Reading membership is not a proof dependency. Exact assumptions remain in the Lean statement.

theorem Covering.selected_underestimate_implies_branch_underestimate {X : Type*} (C : Covering X) (f : X → ℝ) (S : Finset Node) (U : Node → WithTop ℝ) (best : ℝ) (v : Node) (hv : v ∈ S) (hpath : v <+: select S U) (hU : U v < (best : WithTop ℝ)) : ∃ j ≤ depthBound S, U (C.optimalPath f j) < (best : WithTop ℝ)
theorem BanditRLProof.HOO.Covering.history_selected_underestimate Compiled

The same comparison stated directly on the generated chronological state.

Used in these reading views: Bandit Book

1. Finite bandits, traces, and regret

Canonical node identitydeclaration:BanditRLProof.HOO.Covering.history_selected_underestimate

Reading membership is not a proof dependency. Exact assumptions remain in the Lean statement.

theorem Covering.history_selected_underestimate {X : Type*} (C : Covering X) (f : X → ℝ) (ν ρ best : ℝ) (Y : ℕ → ℝ) (n : ℕ) (v : Node) (hv : v ∈ expanded (history ν ρ Y n)) (hpath : v <+: action ν ρ Y n) (hu : upper ν ρ (history ν ρ Y n) v < (best : WithTop ℝ)) : ∃ j ≤ n, upper ν ρ (history ν ρ Y n) (C.optimalPath f j) < (best : WithTop ℝ)