{"id":"84785072-348c-42cd-b86c-90dd1bd4127d","arxiv_id":"2412.06037","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"Discrete-time revision-protocol dynamics in 2x2 anti-coordination games can be made Li-Yorke chaotic for any interior Nash equilibrium using suitable imitative or innovative protocols.","lead":"This paper proves that even the simplest two-strategy anti-coordination games can produce chaotic, unpredictable population dynamics in discrete time, when agents compare payoffs via revision protocols. The authors construct exact protocols, including a rescaled 'imitate the better' rule, under which Li-Yorke chaos and periodic orbits of every period appear for large time steps.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 3's repelling-Nash conclusion rests on Proposition 7, whose appendix proof is an outline that omits the key inequality checks; an error in those thresholds would invalidate the central observable-chaos claim.","rationale":"The reader's weakest assumption points to the first-order Euler discretization and to the omitted calculations in Proposition 7. I agree with the latter but not the former as the load-bearing issue. The map (map) is the paper's explicit definition of the discrete-time dynamics; whether a different revision-timing protocol would give different dynamics is a modeling question, not a correctness flaw in Theorems 2 and 3 as stated. By contrast, Proposition 7 is the keystone of Theorem 3, the paper's main new contribution: without repelling p, the Li-Yorke chaotic set could have measure zero and be unobservable, undercutting the economic message. The appendix explicitly says it skips the calculations that verify the Proposition 1 hypotheses and the derivative bounds. Those calculations are not routine one-line checks; they involve the piecewise map (21), the Lipschitz constant L, and the threshold δ*_3 obtained from a quadratic inequality. The paper gives the thresholds but not the derivations, so a reader cannot certify Theorem 3 from the text. I do not claim the proposition is false — the formulas appear plausible and the outline is checkable — but this is exactly the kind of missing support that should gate acceptance. Hence the verdict stays CONDITIONAL: the concern does not change the reader's assessment, and it would be resolved by the concrete verification described.","tokens_in":35492,"tokens_out":31910,"duration_ms":265636,"concrete_test":"Independently verify Proposition 7 by direct computation: for p = 0.2 and p = 0.4, with γ = p + p²/2 and the stated η, ξ, evaluate F* from (21) and check numerically that F*(γ) < p/2, F*(cl) > γ, and (F*)²(cl) > γ for δ just above δ*_p = max{δ*_1, δ*_3}, and that the one-sided derivatives of F* at p are < −1 for δ > max{δ*_4, δ*_5}. Then repeat symbolically (or with exact rational arithmetic for rational p) to confirm the thresholds δ*_i are correct and δ*_p < 1; if any inequality fails at the claimed threshold, Theorem 3 needs revision.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim that the truncated PPI protocol (Section 6) makes the Nash equilibrium p repelling while preserving Li-Yorke chaos is Theorem 3. For p ∈ (0, 1/2) this is exactly Proposition 7 (Appendix D), and the p > 1/2 case is delegated to it by topological conjugacy (Proposition 8). The proof of Proposition 7, however, is not a proof: after defining γ = p + p²/2, η = 4/[p(2−2p−p²)(b−d)], ξ = 4/[p(b−d)] and thresholds δ*_1, ..., δ*_5, the text says 'we will skip all calculations and present only its outline.' The chaos part requires verifying that the points cl = p/2 and γ satisfy the hypotheses of Proposition 1: F*(γ) < cl, F*(cl) > γ, and (F*)²(cl) > γ for every δ ∈ (δ*_p, 1], where δ*_p = max{δ*_1, δ*_3}. These are nontrivial inequalities involving the explicit piecewise map (21) and the stated Lipschitz bound L = 1 + 2δ/(2−2p−p²). The repelling part requires the one-sided derivative bounds (F*)'_−(p) < −1 and (F*)'_+(p) < −1, which are only asserted to follow from δ > δ*_4, δ*_5. If any of these threshold computations contains a sign or algebra error, the conclusion that chaos is observable (p repelling) fails. Since Theorem 3 is the paper's main added value beyond [23], this omitted verification is the load-bearing point.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies discrete-time revision-protocol dynamics for 2×2 anti-coordination games with payoff matrix (game) and a unique mixed Nash equilibrium p. The state map is the first-order Euler map F(x)=x+δ[(1−x)ρ_BA(x)−xρ_AB(x)]. The main results are: Theorem 1, which constructs innovative and imitative revision protocols for every such game so that, for large δ, the dynamics is Li-Yorke chaotic, has periodic orbits of all periods, and has repelling Nash equilibrium; Theorem 2, which shows that a two-parameter perturbation of pairwise proportional imitation (rates (14)) yields Li-Yorke chaos for every game; and Theorem 3, which uses truncated rates (18)-(19) to obtain Li-Yorke chaos together with a repelling Nash equilibrium. The proofs are constructive, based on the period-three criterion of Proposition 1 and explicit piecewise-linear bimodal maps. Most technical estimates are collected in the appendices; the exception is Proposition 7, whose proof is explicitly only an outline.","tokens_in":35926,"tokens_out":20029,"duration_ms":173650,"significance":"If the missing verification is supplied, this is a valuable contribution to the discrete-time theory of population games. The paper gives explicit, game-dependent constructions rather than abstract existence results, and it shows that a natural perturbation of pairwise proportional imitation—the standard microfoundation of the replicator dynamics—can produce Li-Yorke chaos with a repelling Nash equilibrium, in sharp contrast to the convergent continuous-time picture. The explicit computations in Propositions 2 and 5, Lemma 7, and Lemma 3 are detailed and check out, and the topological-conjugacy reductions are elegant. The main new claim concerning observable chaos, however, currently rests on an incomplete proof, so the result is not yet fully established as written.","major_comments":[{"comment":"The proof of Proposition 7 states that it will 'skip all calculations and present only its outline' and then asserts the inequalities F*(γ)<cl, F*(cl)>γ, (F*)²(cl)>γ for δ>δ*_p, as well as the one-sided derivative bounds (F*)'_−(p)<−1 and (F*)'_+(p)<−1 for δ>max{δ*_4,δ*_5}. This proposition is the sole proof of Theorem 3 for p∈(0,1/2) and, through Proposition 8, for p>1/2. Since Theorem 3 is the paper's main added value beyond [23], these omitted calculations are load-bearing. Please supply the complete algebra for the thresholds δ*_1,...,δ*_5, the Lipschitz estimate with L=1+2δ/(2−2p−p²), and the derivative bounds, or provide a machine-checked verification.","section":"Appendix D, proof of Proposition 7"},{"comment":"The abstract and the concluding section state that unpredictability 'is encoded into any imitative revision protocol.' The supporting statement, Proposition 4, requires the special case d−b=a−c (i.e., p=1/2) and the extra symmetry condition (13). For general p and general imitative protocols the paper proves existence of some protocol with chaotic dynamics, not inevitability across all imitative protocols. Please qualify the abstract and conclusions accordingly.","section":"Abstract and Section 7"}],"minor_comments":[{"comment":"The assertion that (η,ξ,δ)∈∆p is equivalent to (ξ,η,δ)∈∆ep is not true as stated; for example, with p=0.4 and ep=0.6, (η,ξ)=(5,5) satisfies the bounds of ∆ep but not those of ∆p. The subsequent conjugacy argument can be formulated directly without this equivalence and should be restated.","section":"Section 5.1, Step 3"},{"comment":"The analogous equivalence between ∆*_p and Γ*_ep also appears to be one-directional at best. Please state precisely which inclusion is needed for the proof of Proposition 8 and verify it for the specific maximal parameters used there.","section":"Section 6.1, Step 3"},{"comment":"The notation c^δ_l appears to be a typo for z^δ_l, and the displayed formula for δ* does not solve F(z)=1. From F(z)=z[1+δ(1−z)h(z)], the correct value is δ*=1/[z h(z)] (with z=z^δ_l), so the formula should be corrected.","section":"Appendix C, proof of Proposition 4"},{"comment":"There are typos in displayed formulas: in the expression for ρAB(x) on [2p,1], 'b(1−2b)' should be 'p(1−2p)', and in Proposition 9 the interval '[b + p²/2, 1]' should read '[p + p²/2, 1]'.","section":"Appendix A, proof of Proposition 2 and Proposition 9"},{"comment":"The sentence 'application of Lemma 4, Lemma 5 and Proposition 7 completes the proof' should explicitly verify that the truncated rates defined by (22) satisfy the conjugacy condition (29). The verification is short but is currently omitted.","section":"Appendix D, proof of Proposition 8"}],"recommendation":"major_revision","confidential_remarks":"I believe the paper has the right core idea and that the missing calculations in Proposition 7 are likely routine, given the quality of the explicit estimates elsewhere. I recommend major revision rather than rejection. The authors should also tighten the abstract and fix the set-equivalence statements in Sections 5 and 6, which are false as written even though the surrounding conjugacy arguments can be repaired locally."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a solid existence-theorem paper in discrete-time evolutionary game theory. The genuinely new part is that for every interior Nash equilibrium of a 2x2 anti-coordination game one can rescale pairwise proportional imitation rates and get Li-Yorke chaos for sufficiently large time step, and that a truncated version makes the equilibrium repelling, so the chaos is observable rather than confined to a measure-zero scrambled set. That is worth taking seriously.\n\nWhat the paper does well: the constructions are explicit. Theorems 1 and 2 come with concrete switch rates or parameter choices, and the inequalities in Proposition 5 are written out and check out. Proposition 4's observation about symmetric imitative protocols is a nice unifying remark, though it is not new in substance—it follows from [23]—and the abstract oversells it by saying chaos is 'encoded into any imitative revision protocol' when the proposition requires d-b=a-c and condition (13).\n\nThe soft spot is exactly where the stress-test note lands. Theorem 3, the main added value, rests on Proposition 7, and the proof in Appendix D says 'we will skip all calculations and present only its outline.' The threshold inequalities for F*(gamma)<cl, F*(cl)>gamma, (F*)²(cl)>gamma, and the two one-sided derivative bounds are asserted but not derived. These are exactly the load-bearing checks: if one of them has a sign or algebra error, the observable-chaos claim fails. This is not a fatal flaw—the formulas are plausible and the pattern matches the fully verified Proposition 5—but it is an incomplete proof as written. A referee should ask the authors to supply the algebra.\n\nThe modeling assumption that the discrete law of motion is the first-order Euler map, with no compounding within a period, is a real scope condition, but it is stated plainly and is standard in this line of work. I would not hold it against the paper.\n\nBottom line: send it to a serious referee. The paper is for people working on discrete-time population dynamics, learning in games, and interval maps; they will get value from the explicit constructions. It deserves referee time, not a desk reject, but the referee should insist on a complete proof of Proposition 7 before acceptance.","headline":"Theorems 2 and 3 are the real contribution, and they look right; but Theorem 3's key lemma is a proof sketch, so the main new claim needs referee verification.","tokens_in":36459,"tokens_out":5039,"would_cite":true,"duration_ms":46531,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["37E05","37D45","91A22"],"pacs":[],"model":"deepseek-v4-flash","headline":"For every 2x2 anti-coordination game, there exist revision protocols and a time-step size such that the discrete-time dynamics is Li-Yorke chaotic, with periodic orbits of every period, even though the continuous-time dynamics converges…","keywords":["anti-coordination games","revision protocols","discrete-time game dynamics","Li-Yorke chaos","pairwise proportional imitation","period-3 orbits","interval maps","Nash equilibrium instability"],"falsifier":"For a specific test, take payoffs with $p=1/4$, $b-d=1$, and the maximum perturbations $\\eta=4p/((1-p)^2(b-d))$, $\\xi=4/(p(b-d))$ from Proposition 5 with $\\delta=1$; the proof asserts that the map $F$ in (15) satisfies the crossing conditions of Proposition 1 at $c_l=p/2$ and $c_r=(p+1)/2$. One can plot the third iterate $F^3$ and check whether there is an $x\\in(c_l,c_r)$ with $F(x)<x<F^3(x)$; if no such crossing exists, the period-3 route is absent. A second falsification target is the Euler-map assumption itself: simulate the asynchronous Poisson process with the same rates and check whether the long-run distribution matches the deterministic chaotic map.","tokens_in":35288,"feed_emoji":"🎲","tokens_out":13790,"duration_ms":124846,"temperature":0.7,"pith_summary":"The paper is trying to establish that discrete-time population game dynamics driven by revision protocols can be chaotic even in the simplest nontrivial case: a symmetric 2x2 anti-coordination game with payoff inequalities $a<c$ and $d<b$, whose continuous-time process converges to the unique mixed Nash equilibrium $p=(b-d)/(c-a+b-d)$. For every such game one can choose a revision protocol—either innovative, based on direct payoff comparison, or imitative, based on copying a randomly observed agent—and a time-step length $\\delta$ such that the update map on the unit interval is Li-Yorke chaotic and has periodic orbits of every period, with the Nash equilibrium repelling. This is not an artifact of contrived switch rates: a perturbed version of the standard pairwise proportional imitation protocol produces the same chaos for sufficiently large $\\delta$, and in the symmetric case $p=1/2$ every imitative protocol satisfying a natural symmetry condition becomes chaotic. A final truncation of the imitation rates removes the possibility that the chaos is hidden on a measure-zero set by making the Nash equilibrium repelling for large $\\delta$.","feed_headline":"Chaos can be forced in every 2x2 anti-coordination game","feed_subtitle":"Discrete-time imitation becomes unpredictable for large revision steps, even when continuous time converges.","key_machinery":"The carrying object is the one-dimensional interval map $F:[0,1]\\to[0,1]$ obtained from the inflow-outflow equation $F(x)=x+\\delta[(1-x)\\rho_{BA}(x)-x\\rho_{AB}(x)]$. The load-bearing dynamical criterion is Proposition 1: if two points $0\\le z_l<z_r\\le1$ satisfy $F^2(z_l)>z_r$ and one of two geometric crossing conditions ($F(z_r)<z_l<F(z_l)>z_r$, or $F(z_r)>z_l$ and $F(z_l)<z_l$), then some $x\\in(z_l,z_r)$ has $F(x)<x<F^3(x)$; a form of the odd-chaos theorem then yields a period-3 orbit, and the Li-Yorke and Sharkovsky theorems yield Li-Yorke chaos with all periods. For imitative protocols the same map reads $F(x)=x(1+\\delta(1-x)(r_{BA}(x)-r_{AB}(x)))$, with fixed points $0$, $1$ and $p$, so chaos is controlled by the sign of the difference of the two conditional imitation rates. The explicit constructions place the two critical points at $c_l=p/2$ and $c_r=(p+1)/2$ (or their mirrors under $x\\mapsto1-x$) and verify Proposition 1 with formulas; the truncation in Section 6 replaces one rate by a constant past $\\gamma$, enlarging the admissible $\\eta$ and making $p$ repelling while preserving the self-map property $F([0,1])=[0,1]$.","core_discovery":"The paper states the result as three theorems. Theorem 1: for any $2\\times2$ anti-coordination game with $a<c$ and $d<b$ there is a revision protocol—either innovative or imitative—and a choice of $\\delta$ for which the map $F(x)=x+\\delta[(1-x)\\rho_{BA}(x)-x\\rho_{AB}(x)]$ is Li-Yorke chaotic, has periodic orbits of any period, and has the unique Nash equilibrium $p$ repelling. Theorem 2 specializes to a perturbation of pairwise proportional imitation: with conditional imitation rates $r_{AB}(x)=\\eta[(b-d)/p(x-p)]_+$ and $r_{BA}(x)=\\xi[(b-d)/p(p-x)]_+$, explicit multipliers $\\eta,\\xi$ are given such that for every $\\delta$ above a threshold $\\delta_p$ the dynamics is Li-Yorke chaotic with all periods; for $p\\in(0,1/2)$ the values are $\\eta=4p/((1-p)^2(b-d))$ and $\\xi=4/(p(b-d))$, and the case $p>1/2$ follows by the topological conjugacy $x\\mapsto 1-x$. Theorem 3 modifies the rates by truncating one of them to a constant beyond a level $\\gamma\\in(0,1)\\setminus\\{p\\}$; with $\\gamma=p+p^2/2$ for $p<1/2$, the same chaotic conclusions hold and, for $\\delta>\\delta_p$, $p$ is repelling, so the chaotic behavior cannot be masked by almost-sure convergence to equilibrium.","pith_inferences":["Beyond the paper, the same period-3 crossing criterion should be testable in larger anti-coordination games by projecting onto a one-dimensional factor of the state space; a numerical search for period-3 points would show whether the chaotic phenomenon extends beyond $2\\times2$ games.","Beyond the paper, because the deterministic map is the exact Euler step, simulating the underlying asynchronous Poisson process would reveal whether allowing multiple revision opportunities within one period shifts or suppresses the chaos thresholds; this is a direct robustness check on the model.","Beyond the paper, the truncation parameter $\\gamma$ has a behavioral interpretation as a satisficing threshold beyond which agents ignore further payoff differences, which suggests an experimental prediction: capping payoff information should make volatile, chaotic swings more likely rather than less."],"forward_implications":["For any $2\\times2$ anti-coordination game, a sufficiently large revision step $\\delta$ can destroy equilibrium prediction: the discrete-time dynamics can have periodic orbits of every period and an uncountable scrambled set of initial conditions.","The perturbed pairwise proportional imitation protocol of Theorem 2 is close to the microeconomic protocol that yields replicator dynamics in continuous time, so the instability is not confined to exotic or hard-to-interpret switch rates.","In the symmetric case $p=1/2$, Proposition 4 shows that any imitative protocol satisfying the symmetry condition $r_{AB}(x)+r_{AB}(1-x)=r_{BA}(x)+r_{BA}(1-x)$ is chaotic for large $\\delta$; chaos is thus an inherent property of imitation, not a special construction.","The truncated rates of Theorem 3 make the Nash equilibrium repelling for sufficiently large $\\delta$ for every equilibrium position $p$, so the chaotic set cannot be dismissed as a measure-zero phenomenon that almost all trajectories avoid.","All conclusions for equilibria above $1/2$ are obtained by topological conjugacy, so the result covers the whole range $p\\in(0,1)$."],"supporting_citations":[{"why":"Establishes Li-Yorke chaos for pairwise proportional imitation in the symmetric congestion case, which the paper extends to all anti-coordination games.","marker":"[23]"},{"why":"Introduces the pairwise proportional imitation protocol whose perturbed rates are used in Theorems 2 and 3.","marker":"[31]"},{"why":"Defines Li-Yorke chaos and supplies the 'period three implies chaos' criterion used throughout.","marker":"[37]"},{"why":"Gives the odd-chaos theorem that converts the inequality $F(x)<x<F^3(x)$ from Proposition 1 into a period-3 orbit.","marker":"[38]"},{"why":"Provides the revision-protocol framework and inflow-outflow equation on which the interval map $F$ is based.","marker":"[63]"},{"why":"Sharkovsky's theorem, used to conclude from a period-3 orbit that periodic orbits of every period exist.","marker":"[69]"}],"fun_headline_variants":["Discrete time flips stable Nash to chaos in anti-coordination games","Imitation protocols can force chaos in any 2x2 anti-coordination game","Large discrete steps induce chaos in anti-coordination games","From stability to chaos: discrete revision in 2x2 games"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the discrete-time law of motion is exactly the first-order Euler map $F(x)=x+\\delta[(1-x)\\rho_{BA}(x)-x\\rho_{AB}(x)]$, meaning each agent has at most one revision opportunity in a period and there is no within-period compounding; if agents can revise several times inside a period, the map and the chaos thresholds change. A second load-bearing point is that the proof of the repelling-fixed-point part of Theorem 3 uses Proposition 7, whose Appendix D proof explicitly skips calculations and gives only an outline.","fun_headline_variants_meta":{"raw":{"variants":["Discrete time flips stable Nash to chaos in anti-coordination games","Imitation protocols can force chaos in any 2x2 anti-coordination game","Large discrete steps induce chaos in anti-coordination games","From stability to chaos: discrete revision in 2x2 games"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000458,"raw_usage":{"total_tokens":2369,"prompt_tokens":1088,"completion_tokens":1281,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":704,"completion_tokens_details":{"reasoning_tokens":1205}},"tokens_in":704,"tokens_out":1281,"duration_ms":12674,"temperature":1.0,"reasoning_tokens":1205,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T20:06:18.850772+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"For a specific test, take payoffs with $p=1/4$, $b-d=1$, and the maximum perturbations $\\eta=4p/((1-p)^2(b-d))$, $\\xi=4/(p(b-d))$ from Proposition 5 with $\\delta=1$; the proof asserts that the map $F$ in (15) satisfies the crossing conditions of Proposition 1 at $c_l=p/2$ and $c_r=(p+1)/2$. One can plot the third iterate $F^3$ and check whether there is an $x\\in(c_l,c_r)$ with $F(x)<x<F^3(x)$; if no such crossing exists, the period-3 route is absent. A second falsification target is the Euler-map assumption itself: simulate the asynchronous Poisson process with the same rates and check whether the long-run distribution matches the deterministic chaotic map.","supporting_citations":[{"cited_title":"Falniowski and P","cited_arxiv_id":null,"evidence_quote":"Establishes Li-Yorke chaos for pairwise proportional imitation in the symmetric congestion case, which the paper extends to all anti-coordination games."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Introduces the pairwise proportional imitation protocol whose perturbed rates are used in Theorems 2 and 3."},{"cited_title":"Li and J","cited_arxiv_id":null,"evidence_quote":"Defines Li-Yorke chaos and supplies the 'period three implies chaos' criterion used throughout."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the odd-chaos theorem that converts the inequality $F(x)<x<F^3(x)$ from Proposition 1 into a period-3 orbit."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the revision-protocol framework and inflow-outflow equation on which the interval map $F$ is based."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Sharkovsky's theorem, used to conclude from a period-3 orbit that periodic orbits of every period exist."}],"review_version":1}