{"id":"bbe53f7b-98bd-4bbe-b504-bc4b782b7932","arxiv_id":"1908.03643","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A zero-sum differential game is uniquely ergodic iff the players have no disjoint 'dominions', approximately invariant sets each can defend, under an equicontinuity or uniform-time-estimate condition.","lead":"This paper proves that a deterministic two-player zero-sum differential game has a unique long-term average value precisely when the two players have no disjoint regions, called dominions, that they can each defend. It gives mathematicians a single symmetric test for a class of games that previously required separate controllability conditions for each player.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Proposition 4.6 is false as stated: an irrational linear flow on the torus is uniquely ergodic yet its invariant parallel lines are closed dominions of both players whose torus images are disjoint; the proof's claim that π^{-1}(π(D)) is a dominion fails because this set need not be closed.","rationale":"The reader's CONDITIONAL verdict focused on the sufficiency direction's regularity assumptions. I found a more fundamental problem in the necessary direction. The counterexample uses only the paper's own definitions and a textbook uniquely ergodic flow: a no-control irrational linear flow on T^2. Since Proposition 4.6 is false as stated, the central 'if and only if' characterization cannot be accepted in its current form. The paper contains substantial usable material, and replacing 'dominion in the torus' by its closure appears to restore the intended theorems; however, this changes Definition 4.5 and the statements of Proposition 4.6 and Theorems 4.13 and 5.3, so a correction to the central claim is required. Because the flaw is a false assertion in the main equivalence rather than a missing proof of a regularity condition, I would reject the current version while recognizing that the underlying method may be salvageable.","tokens_in":25998,"tokens_out":45083,"duration_ms":520337,"concrete_test":"Run the following formal check. Set γ irrational, f(x,a,b) = (1,γ), A = B = {0}, and ℓ ≡ 0. (i) Verify that D1 = {(s, γs)} and D2 = {(s, γs + 1/2)} are closed and exactly invariant, hence dominions of both players under Definition 4.2. (ii) Verify that π(D1) ∩ π(D2) = ∅ by showing 1/2 ∉ γZ + Z. (iii) Verify unique ergodicity: for every continuous 1-periodic g, the discounted values δvδ(x) = δ∫_0^∞ e^{-δs} g(x + s(1,γ)) ds converge uniformly to the torus average of g, by unique ergodicity of irrational flows. These three checks make the counterexample to Proposition 4.6 rigorous. Additionally, directly compute D1 + Z^2 for D1; it is dense and not closed, disproving the proof's invariance step.","verdict_should_be":"REJECT","load_bearing_attack":"The necessity half of the central characterization breaks before any regularity assumption is needed. Proposition 4.6 asserts that unique ergodicity implies π(D1) ∩ π(D2) ≠ ∅ for every pair of dominions. Its proof replaces D by π^{-1}(π(D)) to make it Z^n-invariant, claiming this is again a dominion. This is false: a dominion must be closed (Definition 4.2), but π^{-1}(π(D)) = D + Z^n need not be closed. Take γ irrational and the no-control game f(x,a,b) = (1,γ) with A = B = {0}. D1 = {(s, γs)} and D2 = {(s, γs + 1/2)} are closed invariant lines, hence dominions of both players. Their images in the torus are disjoint: an intersection would force 1/2 ∈ γZ + Z, impossible for irrational γ. Yet this is the irrational linear flow on T^2, which is uniquely ergodic; for every continuous periodic g, δ∫_0^∞ e^{-δs} g(x + s(1,γ)) ds converges uniformly to the torus average of g. Thus a uniquely ergodic game has disjoint dominions in the torus, contradicting Proposition 4.6 and the only-if direction of Theorems 4.13 and 5.3. The fix is to work with closures of π(D); with the paper's K = π(D) definition, the main characterization is false as stated.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies unique ergodicity of deterministic zero-sum differential games on the torus, defined as the uniform convergence of the discounted value δv_δ (or the finite-horizon average v(t,·)/t) to a constant for every state-dependent perturbation of the running payoff. The central objects are 'dominions', closed sets that a player can make approximately invariant. The main claim is that unique ergodicity is equivalent to the absence of disjoint dominions of the two players in the torus, under either structural equicontinuity (Theorem 4.13) or a uniform reachability time estimate (Theorem 5.3). The paper proves a necessary condition in Proposition 4.6 and provides operator-theoretic characterizations of dominions in Section 6.","tokens_in":26244,"tokens_out":12469,"duration_ms":122091,"significance":"If the characterization were valid, it would give a clean, symmetric, dynamics-only criterion for unique ergodicity in two-player games, extending Arisawa's results for optimal control and connecting to viability theory and repeated games. The paper contains several substantial and apparently correct auxiliary results, including Theorem 3.2 (unique ergodicity, structural equicontinuity, and the strong maximum principle), Lemma 4.11 (argmin and argmax of cell-problem solutions are dominions), and Theorem 6.2 (proximal-normal characterization of dominions). However, the central necessity result, Proposition 4.6, is false as stated, and the same counterexample invalidates Proposition 5.2. The manuscript therefore does not establish its advertised characterization.","major_comments":[{"comment":"The proof of Proposition 4.6 uses the claim, stated in the note after Definition 4.5, that π^{-1}(π(D)) is again a dominion whenever D is a dominion. This is false as stated because Definition 4.2 requires dominions to be closed, while π^{-1}(π(D)) = D + Z^n need not be closed. A concrete obstruction is the control-free game on R^2 with f(x,a,b) = (1,γ), A = B = {0}, and γ irrational. The sets D1 = {(s,γs) : s ∈ R} and D2 = {(s,γs + 1/2) : s ∈ R} are closed invariant lines, hence dominions of both players, and π(D1) ∩ π(D2) = ∅ because an intersection would force 1/2 ∈ γZ + Z, impossible for irrational γ. At the same time the game is uniquely ergodic, since the underlying dynamics is the uniquely ergodic irrational linear flow on T^2; for every continuous periodic perturbation g, δv_δ^g converges uniformly to the torus average of ℓ + g. Thus Proposition 4.6 is false, and the only-if direction of Theorems 4.13 and 5.3 is not established. The fix should involve closed images in the torus (e.g., require K closed with π^{-1}(K) a dominion, or replace π(D) by its closure) and a separation argument adapted to such sets.","section":"§4.1–4.2, Definition 4.5 and Proposition 4.6"},{"comment":"Independently of the closedness issue, the construction in the proof of Proposition 4.6 requires the ε-neighborhoods D1^ε and D2^ε in R^n to be disjoint for some ε > 0, which is possible only if the torus images π(D1) and π(D2) are positively separated. In the irrational-flow example the images are dense, so no such ε exists and the payoff perturbation g satisfying (10) cannot be defined. Any corrected version of the necessity theorem must either add a separation hypothesis or use an argument that works for non-closed images.","section":"§4.2, proof of Proposition 4.6 (separation step)"},{"comment":"Proposition 5.2 is stated as a stronger version of Proposition 4.6 for Lipschitz perturbations and is used in Theorem 5.3 to prove (ii) ⇒ (iii). The control-free irrational-flow example also disproves Proposition 5.2: the game satisfies the proposition's hypothesis (it is uniquely ergodic, hence ergodic for every Lipschitz perturbation) but the players have disjoint dominions in the torus. Consequently the proof of Theorem 5.3 is invalid as written. The uniform time estimate Assumption A2 does not repair Proposition 5.2, which is stated independently of A2; a corrected proof of the necessity direction under A2 would still require a different argument.","section":"§5, Proposition 5.2 and Theorem 5.3"}],"minor_comments":[{"comment":"There are several typographical errors, including 'W e', 's um', and 'identify' for 'identified'; the manuscript would benefit from a careful proofreading pass.","section":"Abstract and Introduction"},{"comment":"The notation in the note after Definition 4.5 is garbled in the rendering, and the claim that π^{-1}(π(D)) is a dominion is precisely the point that fails; this should be corrected in conjunction with the major revision.","section":"Definition 4.5 and following note"},{"comment":"The proof correctly observes that for argmin and argmax sets of a continuous periodic solution w of the cell problem, the images π(D1) and π(D2) are closed; this contrasts with the general definition of a dominion in the torus, where π(D) need not be closed, and this distinction should be made explicit.","section":"Theorem 4.13"},{"comment":"The argument for irrational γ is heuristic, particularly the statement that 'any deviation of a trajectory from one of these half-lines eventually leads to the intersection'; a rigorous proof or a reference to a standard density argument would strengthen the example.","section":"Example 4.14"}],"recommendation":"major_revision","confidential_remarks":"The false claim in Proposition 4.6 is not a minor technicality: the counterexample is a control-free uniquely ergodic game, so the submitted version's main theorem is false as stated. I believe the program is salvageable by redefining 'dominion in the torus' via closed sets or closures and proving the separation step for such sets, which is why I recommend major revision rather than rejection. The paper's auxiliary results and the general framework are valuable, but the central equivalence must be corrected before the manuscript can be considered for publication."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The main theorem is false as stated. Proposition 4.6 claims that unique ergodicity forces every pair of dominions to intersect in the torus. Here is a counterexample: take A = B = {0}, f(x, a, b) = (1, γ) with γ irrational, and state space R^2/Z^2. The lines D1 = {(s, γs)} and D2 = {(s, γs + 1/2)} are closed, invariant, hence dominions of both players. Their images in the torus are disjoint (an intersection would force 1/2 ∈ γZ + Z, impossible for irrational γ). The game is just the irrational linear flow on T^2, which is uniquely ergodic: for every continuous periodic g, the discounted averages converge uniformly to the torus average. So the necessary condition in Proposition 4.6 fails, and with it the only-if directions of Theorems 4.13 and 5.3.\n\nThe proof's invalid step is the claim that π^{-1}(π(D)) is again a dominion. That set is D + Z^n, which need not be closed even when D is closed; Definition 4.2 requires dominions to be closed. This is not a subtle hidden-assumption issue: the author himself notes earlier that images of closed sets in the torus need not be closed. The fix is to work with closures of π(D), or to require dominions to be Z^n-invariant and closed in R^n, but that changes the paper's stated definition and theorem.\n\nWhat is still good: the operator-theoretic characterization of dominions (Theorems 6.1 and 6.2) is a solid link to viability theory and likely correct. The sufficiency arguments under structural equicontinuity or Assumption A2 may also survive once the necessary condition is repaired. The paper is not a waste; it introduces a useful notion and proves several nontrivial partial results.\n\nFor a quick verdict: the central equivalence as written is wrong, so the paper cannot be accepted. But it deserves a serious referee, because the framework is valuable and the flaw is identifiable and fixable. If the author redefines dominions appropriately and re-proves the necessity, the result may hold. I would not cite the paper in its current form, but I would read a revised version.\n\nRecommendation: send to peer review, but flag the counterexample and the non-closedness issue right away; the paper needs major revision before it is publishable.","headline":"The central characterization is false: an irrational linear flow on the torus is uniquely ergodic yet has disjoint dominions, so Proposition 4.6 and the only-if directions of the main theorems fail as stated.","tokens_in":26852,"tokens_out":3713,"would_cite":false,"duration_ms":41247,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["91A23","49N70","37A99","49L25","35F21","35B40"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper characterizes unique ergodicity of a deterministic zero-sum differential game by the absence of disjoint dominions, the closed sets each player can keep approximately invariant.","keywords":["differential games","zero-sum games","Hamilton-Jacobi equations","viscosity solutions","ergodicity","unique ergodicity","dominions","limit value"],"falsifier":"A concrete check: for the two-dimensional game $f(x,a,b)=(a,\\gamma b)$, the theory predicts unique ergodicity exactly when $\\gamma$ is irrational; computing $\\delta v_\\delta$ for rational and irrational $\\gamma$ and a family of perturbations $g$, and testing uniform convergence to a constant, would settle the characterization in this model.","tokens_in":25710,"feed_emoji":"🎮","tokens_out":12635,"duration_ms":121742,"temperature":0.7,"pith_summary":"The paper studies deterministic two-player zero-sum differential games on the n-torus, asking when the game has a unique long-run value: the discounted value $\\delta v_\\delta(x)$ converges uniformly to a constant as the discount factor $\\delta$ tends to $0$, not just for one payoff but for every continuous state-dependent perturbation of the running payoff. The central claim is that this property, unique ergodicity, is equivalent to a symmetric condition on the controlled dynamics: the two players must have no disjoint \"dominions,\" meaning no nonempty closed sets, one defensible by each player, that can be kept approximately invariant for arbitrarily long horizons. The necessity of the dominion condition is proved in full generality (Proposition 4.6), and sufficiency is proved under either structural equicontinuity of the discounted value family (Theorem 4.13) or a uniform approximate-reachability time estimate (Assumption A2, Theorem 5.3). A sympathetic reader should care because the result converts a delicate question about limits of Hamilton-Jacobi equations into a geometric check on which regions each player can defend, and it treats the two players symmetrically, unlike earlier controllability conditions.","feed_headline":"No disjoint dominions, one long-run value for the game","feed_subtitle":"A zero-sum differential game is uniquely ergodic exactly when the players have no disjoint dominions.","key_machinery":"The central object is the dominion: a closed set that one player can render approximately invariant for arbitrarily long times. The argument runs through the Hamilton-Jacobi-Isaacs equation of the lower game, whose data are encoded in the Hamiltonian $H(x,p)=\\min_{b\\in B}\\max_{a\\in A}\\{-\\langle f(x,a,b),p\\rangle-\\ell(x,a,b)\\}$, and its recession function $H_\\infty(x,p)=\\min_{b\\in B}\\max_{a\\in A}\\{-\\langle f(x,a,b),p\\rangle\\}$; the cell problem $c+H(x,Dw)=0$ links discounted values to the ergodic constant. Theorem 3.2 decomposes unique ergodicity into two ingredients: structural equicontinuity of the family $\\{\\delta u_\\delta\\}$ under every perturbation $g\\in C^0_{\\rm per}$, and the strong maximum principle for $H_\\infty$. Dominions do the work of the second ingredient: by Lemma 4.11, the argmin and argmax of any periodic solution of $H_\\infty(x,Dw)=0$ are respectively dominions of player 1 and player 2, so the no-disjoint-dominions condition forces every such solution to be constant. Section 6 gives the operator-theoretic handle, characterizing dominions by inequalities involving proximal normals, which is exactly the leadership/discriminating domain criterion of viability theory.","core_discovery":"On the paper's own terms, the discovery is that unique ergodicity of a deterministic zero-sum differential game is governed by dominions. A dominion of a player is a nonempty closed set $D$ such that, from any $x\\in D$, that player can keep the trajectory inside the $\\varepsilon$-neighborhood of $D$ for any horizon $T$, no matter what the opponent does; the relevant objects live on the torus $\\mathbb{R}^n/\\mathbb{Z}^n$ because a dominion in $\\mathbb{R}^n$ may project to a dense set. Proposition 4.6 shows uniquely ergodic games cannot admit two disjoint dominions held by the two players. Theorem 4.13 adds structural equicontinuity and proves the converse: if every dominion of player 1 meets every dominion of player 2 in the torus, then the Hamiltonian satisfies the strong maximum principle---the only periodic viscosity solutions of $H_\\infty(x,Dw)=0$ are constants---and the game is uniquely ergodic. Theorem 5.3 proves the same equivalence under Assumption A2, when the game is not necessarily Lipschitz-controllable but reachable points can be approached in time at most $C(-\\log\\varepsilon)^\\gamma$. Section 6 completes the picture by characterizing dominions through proximal-normal inequalities, identifying them with leadership and discriminating domains from viability theory.","pith_inferences":["The proximal-normal characterization of dominions suggests a computational route the paper does not develop: unique ergodicity could be certified by a search over closed sets satisfying the normal-cone inequalities of Theorems 6.1 and 6.2, stopping when a disjoint pair is found or none exists.","The regularity gap between necessity (unconditional) and sufficiency (conditional) leaves room for a genuinely non-equicontinuous game with no disjoint dominions but no unique ergodic limit; constructing or ruling out such an example would locate the exact boundary of the theorem.","Because the companion discrete-time results use the same 'dominion' terminology, the dichotomy may carry over to zero-sum repeated or stochastic games, giving a unified explanation of when long-run values exist; the paper only gestures at this connection."],"forward_implications":["If the equivalence holds, unique ergodicity can be certified from the bare controlled dynamics $f$, independent of the running payoff $\\ell$: check that no closed set defensible by player 1 is disjoint from one defensible by player 2 on the torus.","The theorem recovers and unifies earlier controllability results: if one player can uniformly control the system, the other player's only dominion is the whole torus, so the condition holds automatically and the game is uniquely ergodic.","For one-player optimal control viewed as a two-player game, the dominion condition reduces to the existence and uniqueness of the ergodic attractor of Arisawa: a unique minimal positively invariant set that intersects every dominion.","In the explicit two-dimensional example $f(x,a,b)=(a,\\gamma b)$, the characterization yields a sharp dichotomy: for irrational $\\gamma$ the game is uniquely ergodic, while for rational $\\gamma$ the players hold disjoint dominion lines and the game fails to be uniquely ergodic."],"supporting_citations":[{"why":"Establishes the three-way equivalence between discounted, finite-horizon, and cell-problem ergodicity used in Theorem 2.3 as the definition of ergodicity.","marker":"[AB03]"},{"why":"Provides the characterization of unique ergodicity by equicontinuity plus a strong maximum principle that Theorem 3.2 adapts and generalizes.","marker":"[AB10]"},{"why":"Gives the ergodic attractor for one-player control problems that the paper recovers as a unique minimal dominion in Remark 4.10.","marker":"[Ari97]"},{"why":"Supplies the uniform reachability time estimate, Assumption A2, that drives the sufficiency proof of Theorem 5.3.","marker":"[Ari98]"},{"why":"Defines leadership and discriminating domains of viability theory, which Section 6 identifies with dominions.","marker":"[Car96]"},{"why":"Supplies the two-dimensional example and the criterion that the cell problem has nonconstant solutions exactly when $\\gamma\\in\\mathbb{Q}$.","marker":"[Car10]"},{"why":"Provides the viscosity-solution characterization of value functions and the H\\\"older regularity estimate used in the PDE proofs.","marker":"[BCD97]"},{"why":"Provides the measurable selection theorem used to construct strategies in the sufficiency direction of Theorem 6.2.","marker":"[AF09]"},{"why":"Introduces the cell problem and corrector method for Hamilton-Jacobi equations that the paper's ergodic framework extends.","marker":"[LPV87]"}],"fun_headline_variants":["Unique ergodicity tied to no disjoint dominions","No disjoint dominions means one limit value","Dominion condition decides unique ergodicity","Zero-sum games: when all dominions intersect","Ergodic iff player dominions always overlap"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the discounted value functions stay uniformly continuous as the discount rate vanishes (or satisfy the uniform approximate-reachability time estimate A2); if that regularity fails, the no-disjoint-dominions condition is only necessary, not sufficient, and the equivalence collapses.","fun_headline_variants_meta":{"raw":{"variants":["Unique ergodicity tied to no disjoint dominions","No disjoint dominions means one limit value","Dominion condition decides unique ergodicity","Zero-sum games: when all dominions intersect","Ergodic iff player dominions always overlap"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.0002,"raw_usage":{"total_tokens":1364,"prompt_tokens":924,"completion_tokens":440,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":540,"completion_tokens_details":{"reasoning_tokens":370}},"tokens_in":540,"tokens_out":440,"duration_ms":3966,"temperature":1.0,"reasoning_tokens":370,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T14:07:24.642007+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A concrete check: for the two-dimensional game $f(x,a,b)=(a,\\gamma b)$, the theory predicts unique ergodicity exactly when $\\gamma$ is irrational; computing $\\delta v_\\delta$ for rational and irrational $\\gamma$ and a family of perturbations $g$, and testing uniform convergence to a constant, would settle the characterization in this model.","supporting_citations":[],"review_version":1}