{"id":"58910486-8c11-400f-9b26-1382f7358d63","arxiv_id":"2507.16755","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"A new Macaulay2 package computes Nash, correlated, dependency, and conditional independence equilibria for finite normal-form games using algebraic geometry and polyhedral methods.","lead":"This paper presents GameTheory 1.0, a free software package for the Macaulay2 system that computes four kinds of game theory equilibria from payoff data. Researchers can use it to calculate Nash, correlated, dependency, and conditional independence equilibria, along with their underlying algebraic geometry.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Paper's Proposition 5.1 misstates the conditional independence equations; this is a definite error in the documented mathematics, though the package's implementation uses the correct minors.","rationale":"The reader's weakest_assumption concerns the order of saturations in intersectWithCImodel. That order is not actually a correctness risk: after computing I_sat = spohnIdeal:W^∞ and J_sat = ciIdeal:W^∞, the method computes (I_sat + J_sat):W^∞. Since I ⊆ I_sat and J ⊆ J_sat, we have I+J ⊆ I_sat+J_sat, so (I+J):W^∞ ⊆ (I_sat+J_sat):W^∞. Conversely, I_sat ⊆ (I+J):W^∞ and J_sat ⊆ (I+J):W^∞, so I_sat+J_sat ⊆ (I+J):W^∞, and saturating both sides gives (I_sat+J_sat):W^∞ ⊆ (I+J):W^∞. Hence the two are equal; the intermediate order cannot change the final ideal. The paper's statement that the order is 'experimental' refers only to performance, not correctness. The real issue is the wrong equation (4). This is the most load-bearing because it is a false mathematical statement in the definition of one of the four equilibrium notions the paper claims to explain. Any reader implementing or checking the CI model from the paper would be led astray. The package itself uses the correct minors (Example 5.2), so the computational claim survives, but the documentation does not. Since this is a definite, easily fixable error, CONDITIONAL acceptance is appropriate. I also note the absence of an independent check of the spohnCI output in the paper; a simple dimension/degree check against known cases from [13, 11] would strengthen the evidence, but this is secondary to the equation error. The package's other methods are supported by examples, by checks against codimension/degree theorems for spohnIdeal, and by agreement between numberTMNE and blockDerangements. No evidence of fabrication or circularity.","tokens_in":12141,"tokens_out":15262,"duration_ms":148044,"concrete_test":"Use the (2×2×2) joint distribution p_{000}=p_{001}=p_{110}=p_{111}=1/4, all other entries zero. This distribution satisfies X_1⊥⊥X_3|X_2 because each slice for fixed X_2 has rank one. Substituting into equation (4) with iA=0, jA=1, iB=0, jB=1, iC=0 gives 1/4+1/4 = 1/2 ≠ 0, so the displayed equation fails; the correct minor condition gives 0, matching the package output in Example 5.2.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Proposition 5.1 and equation (4) state that X_A⊥⊥X_B|X_C holds iff p_{iA iB iC} + p_{jA jB iC} − p_{iA jB iC} + p_{jA iB iC} = 0. This is not the conditional-independence condition; the correct statement is the vanishing of the 2×2 minor p_{iA iB iC} p_{jA jB iC} − p_{iA jB iC} p_{jA iB iC} = 0. The package's own Example 5.2 computes ideals generated by minors, so the implementation is right and the paper is wrong. Because the abstract promises to 'explain the four equilibrium notions' and the introduction claims the package lets users 'construct and analyze the defining ideals,' a false defining equation for one of the four notions undermines the paper's central documentation claim. The saturation-order issue highlighted by the reader is not load-bearing, since the final saturation of the sum of the individually saturated ideals is equal to the saturation of the original sum, as shown in the rationale.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper describes GameTheory, a Macaulay2 package (version 1.0, available since Macaulay2 1.25.05) that implements algebraic and polyhedral models for four equilibrium notions in finite normal-form games: Nash equilibria (via nashEquilibriumIdeal), correlated equilibria (via correlatedEquilibria), dependency equilibria (via spohnIdeal and spohnMatrices), and conditional independence equilibria (via ciIdeal and spohnCI). The paper explains the underlying mathematics, presents Macaulay2 session excerpts as examples (Bach-Stravinsky, a 2x2x2 game with a double totally mixed Nash equilibrium, a random 2x2x2 game, and a perturbed 2x2 game), and discusses auxiliary tools such as numberTMNE for mixed-volume computations and intersectWithCImodel for computing Spohn CI ideals. The central claim is that the package lets users construct and analyze the defining ideals and polyhedral models of these equilibria.","tokens_in":12277,"tokens_out":9146,"duration_ms":85938,"significance":"If correct, the package provides a unified, exact symbolic-algebraic alternative to numerical algorithms for equilibrium computation, with reproducible Macaulay2 code and concrete examples that match known results, including the reducible Spohn variety for Bach-Stravinsky, the prime elliptic curve after a generic perturbation, and the degree-2 equilibrium ideal of the 2x2x2 double-point example. The mixed-volume computation via Chern classes is a fast and elegant addition. The paper is essentially a software documentation piece, so its value depends on the accuracy of its mathematical statements; two statements need correction before the documentation can be considered reliable.","major_comments":[{"comment":"The conditional independence condition is stated incorrectly. The displayed equation p_{iA iB iC} + p_{jA jB iC} − p_{iA jB iC} + p_{jA iB iC} = 0 is not the defining condition for X_A ⊥⊥ X_B | X_C. The correct condition is the vanishing of all 2×2 minors p_{iA iB iC} p_{jA jB iC} − p_{iA jB iC} p_{jA iB iC} = 0, for i_A, j_A ∈ R_A, i_B, j_B ∈ R_B, and i_C ∈ R_C. In addition, the quantifier in Proposition 5.1 includes a variable j_C that does not appear in the displayed equation. The package's own Example 5.2 uses the correct minor equations, so this is a documentation error rather than an implementation error, but it is load-bearing because the paper's abstract and introduction promise to explain the defining ideals of the four equilibrium notions, and Section 5's CI model is introduced through this false equation. Please correct Proposition 5.1, Eq. (4), and the quantifier.","section":"§5.1, Proposition 5.1 and Eq. (4)"},{"comment":"The statement \"If n = d0 = d1 = 2, then the Spohn variety V_X is an elliptic curve\" is false as written. The Bach-Stravinsky game, which has n = d0 = d1 = 2, is shown in Example 4.4 to have a reducible Spohn variety with three primary components. The theorem presumably assumes X is generic, as in the surrounding discussion of \"generic payoff tensors,\" but the theorem's hypothesis omits this genericity assumption, creating an internal inconsistency between the stated theorem and the paper's own computation. Please add the genericity hypothesis or otherwise align the statement with the cited source [13, Theorem 8].","section":"§4, Theorem 4.3"}],"minor_comments":[{"comment":"The remark that saturating each ideal separately before intersecting, and then saturating the sum, is justified only by experimental performance can be strengthened by observing that saturate(saturate(I,W) + saturate(J,W), W) = saturate(I+J, W), so the strategy is mathematically safe. Stating this equality would remove any concern about the lack of a correctness proof.","section":"§5.3"},{"comment":"The quantifier \"for all i_A, j_A ∈ R_A, i_B, j_B ∈ R_B, i_C, j_C ∈ R_C\" in Proposition 5.1 is inconsistent with Equ. (4), which does not depend on j_C; this should be fixed along with the equation itself.","section":"§5.1"},{"comment":"The sentence \"This discrepancy does not affect the indexing of variables\" is somewhat cryptic; a brief example linking one-based vertex labels to zero-based probability variables would improve readability.","section":"§1"}],"recommendation":"major_revision","confidential_remarks":"The package is partly authored by the same group that developed much of the theory it implements (e.g., [13], [11], [12]), so independent validation is limited; this is common for software papers and does not by itself undermine correctness, but an external verification of the Spohn CI outputs would increase confidence. The two mathematical errors identified in the major comments are readily fixable within the manuscript's scope, so I recommend major revision rather than rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear [Colleague],\n\nYou should know two things about this paper. First, the GameTheory package is a real, working contribution: it gives a single Macaulay2 interface for Nash, correlated, Spohn/dependency, and Spohn-CI equilibria, with examples that actually run and match known results. Second, the paper as written has a clear mathematical error in Proposition 5.1: equation (4) is not the conditional independence condition. The correct condition is the vanishing of the 2x2 minors, which is exactly what the package's own ciIdeal computes (see Example 5.2). So the implementation is right and the text is wrong.\n\nWhat's genuinely new here is the integration: bringing the four equilibrium notions into one package with compatible data structures, plus the bridge to GraphicalModels.m2 for CI ideals. That is a convenience tool, not new theory, but it is a solid one. The examples are well chosen, and the computations are reproducible from the shipped code. The authors also get credit for showing the perturbed Bach-Stravinsky game and the reducible Spohn variety, which makes the behavior of the package concrete.\n\nThe soft spots are real but manageable. The equation (4) error is the most serious; it is in the central documentation claim. Anyone who reads the paper to learn the definition of CI equilibrium would be misled. That needs to be fixed, not just as a typo: the displayed equation should be the minor equation. The introduction's phrase 'offering new theoretical insights' overstates things; the theory is in Portakal-Sturmfels and related papers. The numberTMNE method leans on an unreviewed co-authored preprint (Abo-Portakal-Sodomaco), and the authors don't provide an independent cross-check for that number. That's a minor concern for a software paper, as long as that method is flagged as depending on a preprint.\n\nThe saturation-order issue the reader worried about is not load-bearing. The paper's experimental 'saturate before and after' strategy does not invalidate the output, since the final saturation of the sum is what it is; the order is a performance choice, not a correctness risk.\n\nWho is this for? People who actually compute equilibria and want to do it in Macaulay2, or those studying Spohn varieties computationally. It belongs in a serious referee process, but the referee should request a corrected Proposition 5.1 and a toned-down introduction. I would accept it after minor revision.\n\nRecommendation: send it to review, but make sure the error in equation (4) is fixed before publication.","headline":"A genuinely useful Macaulay2 package, but the paper misstates the conditional independence equations in Proposition 5.1 and needs a revision before it should be cited as documentation.","tokens_in":12886,"tokens_out":2321,"would_cite":true,"duration_ms":22151,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["14-04","91A10","13P10"],"pacs":[],"model":"deepseek-v4-flash","headline":"A new Macaulay2 package computes Nash, correlated, dependency, and conditional independence equilibria as polynomial ideals and polytopes.","keywords":["game theory","Nash equilibrium","correlated equilibrium","dependency equilibrium","Spohn variety","conditional independence","Macaulay2","polyhedral geometry"],"falsifier":"Run spohnCI on a random (2x2x2) game over a finite field with the CI statement that player 1 is independent of player 3 given player 2; then compute the same variety by directly saturating the sum of spohnIdeal and ciIdeal with respect to the product of all coordinate and sum hyperplanes using a different term order. If the two resulting ideals differ, the experimental saturation strategy is incorrect.","tokens_in":11891,"feed_emoji":"🎲","tokens_out":8306,"duration_ms":81013,"temperature":0.7,"pith_summary":"This paper introduces GameTheory, a package for the Macaulay2 computer algebra system, and presents it as a unified algebraic setting for four equilibrium concepts in finite normal-form games. The package computes totally mixed Nash equilibria from their defining multilinear equations, correlated equilibria as a convex polytope, dependency equilibria as the Spohn variety, and conditional independence equilibria as a saturated intersection of the Spohn variety with a conditional independence model. If the package is correct, researchers can investigate questions that were previously hard to access: the dimension and degree of equilibrium sets, primary decompositions, and fast counts of equilibrium bounds via mixed volume. The paper supports the claim with worked examples including the Bach or Stravinsky game and a three-player game with a unique totally mixed Nash equilibrium.","feed_headline":"New package computes four equilibrium types with algebra","feed_subtitle":"Nash, correlated, dependency, and CI equilibria become ideals and polytopes in GameTheory.","key_machinery":"The machinery is an algebraic translation dictionary. Each equilibrium concept is rewritten as a variety or polytope inside the probability ring: Nash equilibria as the solutions of multilinear payoff comparisons, correlated equilibria as a convex polytope defined by linear incentive constraints, dependency equilibria as the Spohn variety generated by the 2x2 minors of the Spohn matrices, and conditional independence equilibria as the saturation of the sum of the Spohn ideal and the conditional independence ideal away from the hyperplanes {p=0} and {sum p=0}. Groebner basis and polyhedral computations on these objects produce the package's outputs, including dimension, degree, irreducibility, vertices, and facets.","core_discovery":"The paper's central claim is that four equilibrium notions for finite normal-form games can be treated uniformly as algebraic or polyhedral objects, and that the GameTheory package version 1.0 implements this treatment correctly and usefully. Totally mixed Nash equilibria are the interior solutions of the multilinear equations defining nashEquilibriumIdeal; correlated equilibria are the points of the correlated equilibrium polytope; dependency equilibria are cut out by the 2x2 minors of the Spohn matrices, defining the Spohn variety; and conditional independence equilibria are obtained by intersecting the Spohn variety with a discrete conditional independence model and saturating away the boundary. The package also computes invariants such as dimension, degree, vertices, and facets, and it provides fast mixed-volume counts for the maximal number of totally mixed Nash equilibria.","pith_inferences":["A natural next step the authors do not take is to scan random games of a fixed format and measure how often the Spohn CI variety is nonempty; the package makes such a scan straightforward.","The fast mixed-volume routines could be reused as a subroutine for counting or bounding Nash equilibria in larger games where direct polytope mixed-volume computations stall.","Because the package links game-theoretic varieties to conditional independence models, known classification results for graphical models could be translated into statements about which game graphs admit nontrivial dependency equilibria."],"forward_implications":["Totally mixed Nash equilibria can be computed as the solutions of an explicit multilinear system, so users get not just numerical solutions but the ideal's dimension, degree, and primary decomposition.","The correlated equilibrium polytope is available as a polyhedron, giving direct access to its vertices, facets, and f-vector.","For generic games, spohnIdeal reports the Spohn variety's codimension and degree, and users can test irreducibility.","spohnCI removes boundary hyperplanes from the intersection of the Spohn variety and a conditional independence model, so graphical-model restrictions can be imposed on equilibrium computations."],"supporting_citations":[{"why":"Develops the geometry of dependency equilibria and defines the Spohn varieties that spohnIdeal and spohnMatrices implement.","marker":"[13]"},{"why":"Gives the multilinear system for totally mixed Nash equilibria that nashEquilibriumIdeal encodes.","marker":"[15]"},{"why":"Introduces dependency equilibria, the notion behind the Spohn variety construction.","marker":"[14]"},{"why":"Bernstein's theorem supplies the mixed-volume bound on isolated solutions that deltaList and numberTMNE use.","marker":"[4]"},{"why":"Establishes that the mixed-volume bound is attained generically through block derangements, which blockDerangements implements.","marker":"[8]"},{"why":"Gives the vector bundle formula that numberTMNE uses to compute the mixed volume.","marker":"[1]"},{"why":"Provides the conditional independence equations that ciIdeal encodes.","marker":"[16]"},{"why":"Defines the parametrized graphical model used in the Spohn CI construction and the saturation approach.","marker":"[12]"},{"why":"Gives the incentive-constraint characterization of correlated equilibria that defines the correlated equilibrium polytope.","marker":"[3]"}],"fun_headline_variants":["Four equilibria become ideals and polytopes","GameTheory: equilibria as algebraic objects","Macaulay2 package unifies four equilibrium notions","Algebra meets game theory in Macaulay2","Equilibria via algebra: new GameTheory package"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the experimentally chosen saturation order in the Spohn CI routine always computes the correct saturated ideal and terminates; the paper gives performance evidence but no correctness proof for that step.","fun_headline_variants_meta":{"raw":{"variants":["Four equilibria become ideals and polytopes","GameTheory: equilibria as algebraic objects","Macaulay2 package unifies four equilibrium notions","Algebra meets game theory in Macaulay2","Equilibria via algebra: new GameTheory package"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000265,"raw_usage":{"total_tokens":1501,"prompt_tokens":732,"completion_tokens":769,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":348,"completion_tokens_details":{"reasoning_tokens":696}},"tokens_in":348,"tokens_out":769,"duration_ms":8001,"temperature":1.0,"reasoning_tokens":696,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T15:02:39.544355+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run spohnCI on a random (2x2x2) game over a finite field with the CI statement that player 1 is independent of player 3 given player 2; then compute the same variety by directly saturating the sum of spohnIdeal and ciIdeal with respect to the product of all coordinate and sum hyperplanes using a different term order. If the two resulting ideals differ, the experimental saturation strategy is incorrect.","supporting_citations":[{"cited_title":"Geometry of dependency equilibria.Rendiconti dell’Istituto di Matematica dell’Universit`a di Trieste: an International Journal of Mathematics, 54, 2022","cited_arxiv_id":null,"evidence_quote":"Develops the geometry of dependency equilibria and defines the Spohn varieties that spohnIdeal and spohnMatrices implement."},{"cited_title":"Solving Systems of Polynomial Equations, volume 97 of CBMS Regional Conference Series in Mathematics","cited_arxiv_id":null,"evidence_quote":"Gives the multilinear system for totally mixed Nash equilibria that nashEquilibriumIdeal encodes."},{"cited_title":"Dependency equilibria","cited_arxiv_id":null,"evidence_quote":"Introduces dependency equilibria, the notion behind the Spohn variety construction."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Bernstein's theorem supplies the mixed-volume bound on isolated solutions that deltaList and numberTMNE use."},{"cited_title":"The maximal number of regular totally mixed nash equilibria","cited_arxiv_id":null,"evidence_quote":"Establishes that the mixed-volume bound is attained generically through block derangements, which blockDerangements implements."},{"cited_title":"Sullivant","cited_arxiv_id":null,"evidence_quote":"Provides the conditional independence equations that ciIdeal encodes."},{"cited_title":"Game theory of undirected graphical models","cited_arxiv_id":null,"evidence_quote":"Defines the parametrized graphical model used in the Spohn CI construction and the saturation approach."},{"cited_title":"Correlated equilibrium as an expression of bayesian rationality","cited_arxiv_id":null,"evidence_quote":"Gives the incentive-constraint characterization of correlated equilibria that defines the correlated equilibrium polytope."}],"review_version":1}