hedgeResponseNat_isBestResponse
Online column response is a best response. Let $G$ be a finite zero-sum game with $M \ge 1$ row actions and $N \ge 1$ column actions, let $\eta \in \mathbb{R}$ be a learning rate, and let $T$ be a horizon. Write $j_s = hedgeResponseNat(G, \eta, s)$ for the online column response at round $s$, and let $\ell = ZeroSumGame.toLossSeq(G, (j_s)_{s < T})$ be the loss sequence it induces for the row player. Then for every round $t < T$ and every fixed column $j$, \[ \sum_{i} hedgeDist(\eta, \ell, t)_i \, A(i, j_t) \;\le\; \sum_{i} hedgeDist(\eta, \ell, t)_i \, A(i, j), \] i.e.\ the played column $j_t$ gives the row player, mixing according to the Hedge distribution at round $t$, no more payoff than any other column.
Statement
set_option relaxedAutoImplicit false in set_option autoImplicit false in open Real Finset BigOperators in lemma hedgeResponseNat_isBestResponse {M N T : ℕ} [NeZero M] [NeZero N] (G : ZeroSumGame M N) (η : ℝ) (t : Fin T) (j : Fin N) : ∑ i : Fin M, hedgeDist η (G.toLossSeq fun s : Fin T => hedgeResponseNat G η s.val) t.val i * G.payoff i (hedgeResponseNat G η t.val) ≤ ∑ i : Fin M, hedgeDist η (G.toLossSeq fun s : Fin T => hedgeResponseNat G η s.val) t.val i * G.payoff i j
- Source
- TCSlib, `hedgeResponseNat_isBestResponse`
- Lean proof
- By Karim Abdel Sadek, Mark Bedaywi, from github.com/Shilun-Allan-Li/tcslib at
c8a02591b1e5(Copyright (c) 2026 Karim Abdel Sadek and Mark Bedaywi. All rights reserved, Apache-2.0). Compiled here against leanprover/lean4:v4.33.1, Mathlib v4.33.1; written for leanprover/lean4:v4.25.0.hedgeResponseNat_isBestResponse← hedgeResponseNat_isBestResponse (TCSlib/LearningTheory/Minimax/HedgeInteraction.lean:344): verbatim
- Verified
- 07 Oct 2026
- Axioms
Classical.choiceQuot.soundpropext- Built on 9
- hedgeResponseNatThe sequence of column best responses generated online against Hedge: at time $t$, given the history $a_0, \dots, a_{t-1}$ of prior column…
- bestColumnGiven a game $G$ and a mixed row strategy $p$,
bestColumnis the pure column $j^* \in \mathrm{Fin}\,N$ that minimises… - ZeroSumGameA finite two-player zero-sum game with $M$ row actions and $N$ column actions is a record consisting of a payoff matrix…
- prefixHedgeMixedStrategy_weight_eq_hedgeDistPrefix Hedge strategy agrees with the abstract Hedge distribution. Let $G$ be a finite two-player zero-sum game with $M \ge 1$ row actions…
- bestColumn_specOptimality of the column best response. Let $G$ be a finite two-player zero-sum game with $M$ row actions and $N \ge 1$ column actions,…
- hedgeDistThe Hedge distribution at round $t$ is the normalization of the weight vector: \[ p_t(i) \;=\; \frac{w_t(i)}{W_t}. \]
- ZeroSumGame.toLossSeqGiven a game $G$ and a sequence of column responses $j_0, \dots, j_{T-1}$, the induced loss sequence for the row player is…
- prefixHedgeMixedStrategyThe normalised Hedge distribution after a prefix of $t$ column actions is the mixed row strategy with weights…
- payoffVsPureGiven a game $G$, a mixed row strategy $p$, and a pure column $j$, the expected payoff against column $j$ is…
- Used by 1
- hedge_constructionHedge yields an approximate minimax pair. Let $G$ be a finite two-player zero-sum game with $M \ge 2$ row actions and $N \ge 1$ column…
Lean source view module on GitHub
Choose a tactic — a marked line — to see the goals Lean had before and after it, as in Lean’s infoview. Step with ← → or j k.
set_option relaxedAutoImplicit false inset_option autoImplicit false inopen Real Finset BigOperators in/-- At each round, the online column response is a best response: against the Hedge distribution `hedgeDist η ℓ t` (where `ℓ` is the loss sequence induced by the responses themselves), the played column `hedgeResponseNat G η t` gives the row player no more payoff than any fixed column `j`. [FS99, §5, Eq. (10)]; [CBL06, proof of Thm 7.1]. **Proof sketch.** Step 1: the payoff of the prefix Hedge strategy against any column is the `hedgeDist`-weighted payoff sum at round `t`, by `prefixHedgeMixedStrategy_weight_eq_hedgeDist`. Step 2: by definition the played column is `bestColumn` of the prefix Hedge strategy. Step 3: `bestColumn_spec`, rewritten through Steps 1 and 2, is the claim. -/lemma hedgeResponseNat_isBestResponse {M N T : ℕ} [NeZero M] [NeZero N] (G : ZeroSumGame M N) (η : ℝ) (t : Fin T) (j : Fin N) : ∑ i : Fin M, hedgeDist η (G.toLossSeq fun s : Fin T => hedgeResponseNat G η s.val) t.val i * G.payoff i (hedgeResponseNat G η t.val) ≤ ∑ i : Fin M, hedgeDist η (G.toLossSeq fun s : Fin T => hedgeResponseNat G η s.val) t.val i * G.payoff i j := by let actions : Fin T → Fin N := fun s => hedgeResponseNat G η s.val -- Step 1: the prefix Hedge strategy's payoff against any column `j'` is the -- `hedgeDist`-weighted payoff sum at round `t`. have hpayoff : ∀ j' : Fin N, payoffVsPure G (prefixHedgeMixedStrategy G η (t := t.val) (fun s : Fin t.val => actions ⟨s.val, lt_trans s.isLt t.isLt⟩)) j' = ∑ i : Fin M, hedgeDist η (G.toLossSeq actions) t.val i * G.payoff i j' := by intro j' unfold payoffVsPure apply Finset.sum_congr rfl intro k _ rw [prefixHedgeMixedStrategy_weight_eq_hedgeDist G η actions t k] -- Step 2: the played column is `bestColumn` of the prefix Hedge strategy. have haction : actions t = bestColumn G (prefixHedgeMixedStrategy G η (t := t.val) (fun s : Fin t.val => actions ⟨s.val, lt_trans s.isLt t.isLt⟩)) := by dsimp [actions] rw [hedgeResponseNat.eq_1] -- Step 3: `bestColumn_spec`, read through Steps 1 and 2. have hbc := bestColumn_spec G (prefixHedgeMixedStrategy G η (t := t.val) (fun s : Fin t.val => actions ⟨s.val, lt_trans s.isLt t.isLt⟩)) j rw [hpayoff, hpayoff, ← haction] at hbc exact hbc