{"id":"d2584302-3880-44ba-8c64-2c0d3a85df8f","arxiv_id":"2411.08809","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For two-player quadratic games with social value orientation, the Nash equilibrium set is characterized by one-dimensional curves from eigenvalue problems, with ellipsoidal bounds when spectra are positive and explicit blow-up asymptotes when they are not.","lead":"This paper analyzes two-player games where each player partly cares about the other player's cost, an idea called social value orientation. It shows that the Nash equilibrium shifts with the cooperation level, and that even cooperative players can hit cooperation levels where the equilibrium becomes unbounded.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The spectral expansions and blow-up formulas in Props. 4, 5, and 7 rely on diagonalizability of MH^{-1}N^{-1}; for defective (Jordan) cases the asserted 'similar' behavior is unproven and the pole order can differ.","rationale":"The reader's conditional verdict identifies the diagonalizability assumption in Remark 2 as the weakest point, and my stress-test agrees: this is the single most load-bearing concern because the paper's main theorems—the spectral expansion (Prop. 2), the contraction and ellipsoid bounds (Lemma 1, Props. 4 and 5), and the blow-up asymptotics (Prop. 7)—are all expressed in terms of eigenvectors and eigenvalues of matrices that may be defective. The absence of a proof or even a precise statement for the Jordan case means the claims as written do not cover the full domain of two-player quadratic games.\n\nThe concern is concrete rather than philosophical: for a defective block with a negative eigenvalue, the resolvent has higher-order poles, so the asymptotic formula in Prop. 7, which assumes a simple factor t/(λ+t), cannot be correct as stated. A direct numerical or symbolic test on a small constructed example would settle the matter quickly. If the test shows that the generalized-eigenvector version reproduces the direct answer, then the paper only needs a careful rewrite; if it does not, the central characterization is incomplete.\n\nI credit the paper for the parts that are solid: the algebraic expansions in Props. 2 and 3 follow from the matrix identity in Lemma 2 and are valid without diagonalizability, and the qualitative bounded/unbounded distinction is likely robust. However, the quantitative claims—explicit ellipsoids and explicit asymptote directions—are exactly the parts that depend on the unproven Jordan extension. Therefore the conditional verdict is appropriate: accept only after the diagonalizable assumption is either stated in each theorem or the Jordan case is rigorously derived.","tokens_in":117300,"tokens_out":17036,"duration_ms":141075,"concrete_test":"Construct a concrete game with d1=d2=1 (or d=3) in which A = MH^{-1}N^{-1} has a Jordan block, e.g. choose cost matrices so that A equals [[λ,1],[0,λ]] with λ = -1, while A1,A2 > 0 so the game is well-posed for θ near the blow-up. Compute uθ for t approaching |λ| along the E1 curve in two independent ways: (i) directly from the KKT inversion in Eq. (5), and (ii) using Prop. 7's asymptote formula with the eigenvectors of A (or, if attempted, with generalized eigenvectors). If the direct solution has a leading pole (t-|λ|)^{-2} in any entry while Prop. 7 predicts only a (t-|λ|)^{-1} term, or if the limiting direction differs from uinf_φj, then the Jordan-case assertion in Remark 2 fails and the theorems require an explicit diagonalizability assumption or a separately derived Jordan generalization.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is that, for θ in (0,π/2)^2, the SVO-Nash equilibrium is characterized by spectral expansions (Props. 2 and 3), by intersection-of-ellipsoid bounds when the relevant spectra are positive (Props. 4 and 5), and by explicit blow-up directions when negative eigenvalues exist (Prop. 7). All of these characterizations are built on the eigendecomposition of A = MH^{-1}N^{-1} (and M1H^{-1}M2^{-1}): Eq. (12) writes G_φ(t) as Σ W_i [t/(λ_i+t)] V_i^T, Lemma 1 uses the eigenvector matrix V to define the P-norm and to prove contraction, and Prop. 7's uinf_φj is a sum over the corresponding right and left eigenvectors. The paper's Remark 2 acknowledges that diagonalizability is assumed and asserts that the general Jordan case would be 'similar,' but no proof or statement of the modified formulas is given.\n\nThis is load-bearing because the Jordan case is not a cosmetic extension. For a defective block with a negative real eigenvalue λ, the resolvent (I + A/t)^{-1} contains terms of order (t-|λ|)^{-k} with k equal to the size of the Jordan block, not just the simple (t-|λ|)^{-1} pole that appears in the diagonalizable formula. Consequently, Prop. 7's expression for the asymptotic direction and its leading-order coefficient, as written, is undefined or incorrect for such data. Likewise, when spec(Λ) > 0 but the matrix is defective, no full eigenvector matrix V exists, so the P-norm in Lemma 1 and the ellipsoid bounds in Props. 4 and 5 are not defined; the boundedness of the curve may still hold, but the stated intersection-of-ellipsoids conclusion is unsupported. Because the paper claims a characterization for general two-player quadratic games and explicitly relies on these spectral decompositions, this is the most serious gap in the argument.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies two-player quadratic games in which each player's cost is the SVO-weighted combination of their own and their opponent's cost. After introducing four coordinate reparametrizations of the SVO square, the authors derive exact expansions, notably u_theta = u_N + G_phi(t)(u_A - u_N) and u_theta = u_1 + G_psi(t)(u_2 - u_1), where G_phi and G_psi are expressed through the eigendecompositions of MH_phi^{-1}N^{-1} and M1H_psi^{-1}M2^{-1}. They use these expansions to prove ellipsoidal containment of the equilibrium set when the relevant spectra are positive and to give explicit asymptotic blow-up directions when negative eigenvalues exist. The results are applied to an open-loop linear-quadratic trajectory coordination problem.","tokens_in":117633,"tokens_out":9887,"duration_ms":103409,"significance":"If the spectral characterizations are correct, the paper offers a clean and surprisingly rich parametric description of SVO-Nash equilibria: a one-parameter family of curves, explicit ellipsoidal bounds, and the possibility of finite-time blow-up even in the cooperative quadrant. The derivation is self-contained linear algebra, has no fitted parameters, and makes explicit, falsifiable predictions about blow-up directions that are illustrated in the trajectory example. The significance is currently tempered, however, by the fact that the main theorems are stated without the diagonalizability hypothesis that the spectral expansions require, and by the absence of a proof for the central blow-up statement.","major_comments":[{"comment":"The spectral expansions and all results built on them assume that MH_phi^{-1}N^{-1} and M1H_psi^{-1}M2^{-1} are diagonalizable, but this hypothesis is not stated in the propositions. For a defective block with a negative real eigenvalue lambda, the resolvent (I + A/t)^{-1} contains terms of order (t - |lambda|)^{-k}, where k is the size of the Jordan block; therefore the simple-pole formula in Eq. (21a) and the direction uinf_phi_j in Prop. 7 are not the general answer. The bounded case is also affected because Lemma 1 defines the P_phi-norm via the full eigenvector matrix V_phi, which does not exist for defective matrices even when the spectrum is positive. The assertion in Remark 2 that 'the results would be similar in the general Jordan case' is unproven and is not a substitute for either a complete Jordan-case derivation or an explicit diagonalizability hypothesis in the theorem statements.","section":"Section 4.3, Remark 2; Props. 2, 4, 5, 7"},{"comment":"Prop. 7, which gives the asymptotic blow-up of Gamma_phi(t) as t -> |lambda_phi_j|, is stated without proof. Unlike Props. 2 and 3, whose proofs are deferred to Appendix 8.3, no derivation of Eq. (21a) is supplied anywhere in the visible manuscript. Since the explicit blow-up direction is one of the paper's main advertised contributions, this is not a cosmetic omission: the proposition must either be proved in the appendix or, if the intended proof is a direct spectral expansion, it should be written out, including the treatment of eigenvalue multiplicity and the conditions under which uinf_phi_j can vanish.","section":"Section 5.2.1, Prop. 7"},{"comment":"The well-posedness cutoff theta_i <= bar(theta)_i, defined by cos(theta_i) A_i + sin(theta_i) D_{-i} > 0 in Eq. (6), is not carried into the theorem statements. As written, Prop. 2 claims validity for all (theta_1, theta_2) in (0, pi/2)^2, but if D_i is indefinite and theta_i exceeds the cutoff, the player's SVO cost is indefinite and the first-order equation no longer characterizes a minimizer. The text says the assumption will be made without stating it in theorems; this overstates the range of validity of the main results. The authors should either add the cutoff condition as an explicit hypothesis or clarify, with proof, why all formulas continue to hold when the SVO costs are not well-posed.","section":"Eq. (6); Props. 2, 5, 7"}],"minor_comments":[{"comment":"Lemma 1 claims G_phi(t) is a contraction with respect to the P_phi-norm for all t in [0, infinity), but at t = 0 the spectral values are exactly 1 and the strict inequality in the proof fails. Consequently the open-ball inclusions in Prop. 4 are false at t = 0, where Gamma_phi(0) = u_N lies on the boundary rather than inside the open ball. The statements should either use closed balls or restrict the claim to t > 0.","section":"Lemma 1 and Prop. 4"},{"comment":"In Eq. (18b) the radius is written r' = ||u_1 - u_2||_{P_phi}, but the analogous bound for Gamma_psi(t) requires the P_psi-norm; this appears to be a typo for r' = ||u_1 - u_2||_{P_psi}.","section":"Prop. 4, Eq. (18b)"},{"comment":"The proof of 'Expansions 1 & 2' refers to 'apply Lemma 3 with w1 = t cos phi and w2 = t sin phi', but the lemma proved in that appendix is Lemma 2. The cross-reference should be corrected.","section":"Appendix 8.3"},{"comment":"Figure 9 lists theta = (3pi/8, 3pi/8) twice in the clockwise ordering, and Figures 15 and 16 describe the E3 and E4 curves with swapped references to theta_1 and theta_2. These captions should be cleaned up.","section":"Figure captions"}],"recommendation":"major_revision","confidential_remarks":"The paper is within the scope of math.OC and game theory, and I see no citation or novelty disclosure issue. The central framework is defensible, but the theorems as stated are not correct or complete outside the diagonalizable case, and one key proposition is unproved. These are fixable either by adding explicit diagonalizability and well-posedness hypotheses or by supplying the Jordan-case analysis and a proof of Prop. 7; I would not reject the paper on the basis of the current gaps."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper gives a clean spectral expansion for how social value orientation shifts Nash equilibria in two-player quadratic games. That is genuinely new and useful: the coordinate transformations in Prop. 1 and the resulting eigenvalue-parameterized curves are not in the prior SVO literature, and the bounded/unbounded classification with explicit blow-up directions goes beyond anything Schwarting et al. or Toghi et al. provide. The LQ trajectory example is a nice bonus, not a gimmick, and the derivations are mostly self-contained linear algebra.\n\nWhere it holds up: Props. 2 and 3 are correctly derived via the block matrix lemma, and the ellipsoidal bounds in Props. 4 and 5 follow from the spectral characterization. The paper is honest about the limits of Prop. 6. The central argument is sound under the stated diagonalizability assumption.\n\nThe soft spots are real but not fatal. Prop. 7, which is the main unboundedness result, has no proof in the main text, and the appendix does not seem to fill that gap. Remark 2's claim that the Jordan case is \"similar\" is exactly the kind of assertion that should be either proved or stated as a conjecture, because defective matrices can change pole orders in the resolvent. The stress-test note is right: for a Jordan block of size k, the blow-up is (t-t0)^(-k), not a simple pole, so the formulas as written simply do not apply. This does not invalidate the diagonalizable case, but it does mean the paper's sweep over \"general two-player quadratic games\" is too broad. The well-posedness cutoff theta_i <= theta-bar_i is also waved away in the body and never carried into the theorem statements; that is a minor consistency issue.\n\nThere are also some typos and a mislabeled lemma reference in the appendix. All of this is fixable, and the core contribution is worth preserving.\n\nWho is this for? People working on cooperative control, game-theoretic human-robot interaction, or any application where SVO parameters are tuned or estimated in LQ games. They will get practical formulas and a warning about blow-up that matters. The paper deserves a serious referee, but the referee should push on the Jordan case and demand a real proof for Prop. 7 before acceptance. I would engage with it and cite it, but I would not rely on the unboundedness formulas for defective data until the gap is closed.","headline":"A genuinely new spectral characterization of SVO-Nash equilibria, with a real but patchable gap around non-diagonalizable cases.","tokens_in":118217,"tokens_out":1292,"would_cite":true,"duration_ms":71796,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["91A10","91A05","49N70","93C05"],"pacs":[],"model":"deepseek-v4-flash","headline":"In two-player quadratic games, social value orientation traces Nash equilibria along curves that can blow up at discrete cooperation levels.","keywords":["social value orientation","Nash equilibrium","quadratic games","linear-quadratic games","two-player games","equilibrium parametrization","blow-up analysis","trajectory coordination"],"falsifier":"Take a concrete two-player quadratic game at a fixed $\\theta$ and compute $u_\\theta$ two ways: by solving the first-order conditions (5) directly, and by evaluating the spectral expansion (11) using the eigen-decomposition of $M H_\\phi^{-1} N^{-1}$. The formulas must agree for every diagonalizable choice; to test the Jordan claim, construct the game so that $M H_\\phi^{-1} N^{-1}$ is a single non-diagonalizable Jordan block (for instance, with $M=N=I$ and $H_\\phi$ chosen so that the product has a repeated eigenvalue with only one eigenvector) and check whether the expansion still matches the direct solve. A divergence between the two computations, or a failure to diverge at $t = |\\lambda_{\\phi j}|$ when Prop. 7 predicts a blow-up, would refute the claimed extension.","tokens_in":117043,"feed_emoji":"♟️","tokens_out":7680,"duration_ms":67458,"temperature":0.7,"pith_summary":"Two-player quadratic games are a standard model for competitive and cooperative interactions, and social value orientation turns each player's cost into a weighted mixture of their own cost and the opponent's cost. This paper shows that as the two cooperation angles sweep over the cooperative regime, the resulting Nash equilibrium is traced by one-dimensional curves, each obtained by solving an eigenvalue problem. The central result is an expansion of the SVO-Nash equilibrium around the standard Nash equilibrium and the altruistic Nash equilibrium, with a matrix that has eigenvalues $t/(\\lambda+t)$. If all the relevant eigenvalues are positive, the equilibrium is provably contained in the intersection of four ellipsoids centered at the four classic outcomes (Nash, altruistic Nash, and each player's individual optimum). If any eigenvalue is negative, the equilibrium blows up at finitely many cooperation levels, with an explicit asymptotic direction, meaning even purely cooperative players can produce unbounded actions. This matters for designing autonomous agents that have to predict or plan around human cooperation levels.","feed_headline":"Cooperation can make game equilibria unbounded","feed_subtitle":"A two-player quadratic game's Nash equilibrium moves along curves that may fly to infinity at discrete cooperation levels.","key_machinery":"The central object is a one-parameter family of matrices built from the game data: $G_\\phi(t) = ( (1/t) M H_\\phi^{-1} N^{-1} + I )^{-\\top}$, with $H_\\phi = \\mathrm{blkdg}(\\cos\\phi\\, I_{d_1},\\ \\sin\\phi\\, I_{d_2})$, plus its Player-opt counterpart $G_\\psi(t) = ( (1/t) M_1 H_\\psi^{-1} M_2^{-1} + I )^{-\\top}$. The spectral decomposition of the matrix products $M H_\\phi^{-1} N^{-1}$ and $M_1 H_\\psi^{-1} M_2^{-1}$ carries the argument: each eigenvalue $\\lambda$ acts as a scalar gain $t/(\\lambda + t)$ on a rank-one eigen-direction, so the sign of the real part of $\\lambda$ decides whether that mode is contractive or divergent. Contraction in every mode yields the ellipsoidal containment; a negative real eigenvalue yields a finite $t = |\\lambda|$ where the denominator vanishes, giving the blow-up and its explicit direction. The coordinate transformations $\\Theta_\\phi$ and $\\Theta_\\psi$ convert the two-dimensional SVO space into a fan of these one-dimensional curves, so the whole equilibrium geometry is understood through eigenvalue problems.","core_discovery":"The paper establishes that for every pair of social value orientations $\\theta \\in (0,\\pi/2)^2$, the SVO-Nash equilibrium $u_\\theta$ can be written as a point on a one-parameter curve. In the Nash expansion, $u_\\theta = \\Gamma_\\phi(t) = u_N + G_\\phi(t)(u_A - u_N)$, where the matrix $G_\\phi(t)$ has spectral decomposition with eigenvalues $t/(\\lambda_{\\phi i} + t)$; the Player-opt expansion analogously has $u_\\theta = u_1 + G_\\psi(t)(u_2 - u_1)$. When the spectra of the two governing matrices are positive, every SVO-Nash equilibrium lies in the intersection of four ellipsoids, $B_\\phi(u_N, r) \\cap B_\\phi(u_A, r) \\cap B_\\psi(u_1, r') \\cap B_\\psi(u_2, r')$. When a governing eigenvalue is negative real, the curve $\\Gamma_\\phi(t)$ blows up as $t$ approaches $|\\lambda_{\\phi j}|$, and the blow-up direction is the explicit vector $u_{\\mathrm{inf}}^{\\phi j} = \\sum_{j \\in J} W_{\\phi j} V_{\\phi j}^\\top (u_A - u_N)$. The paper also applies these formulas to an open-loop linear time-varying trajectory coordination problem and shows that the predicted blow-up directions appear in the sampled trajectories.","pith_inferences":["The results suggest that a pragmatic autonomous planner should treat cooperation levels near the negative eigenvalues as unresolvable regions: rather than solving for equilibria there, the planner can use the blow-up direction to detect when a human driver's inferred cooperation angle is drifting toward a pathological value.","Because the blow-up eigenvectors dominate the equilibrium near a singularity, estimating an opponent's cost or cooperation level from observed actions could be reduced to a low-dimensional problem: only the modes associated with the nearest negative eigenvalues need to be identified.","The ellipsoidal containment under positive spectra could serve as a certificate in human-autonomy interaction: if the inferred SVO interval lies in the positive-spectrum region, the autonomous agent can plan conservatively inside the intersection of ellipsoids without solving the game online.","The same expansion machinery may extend to Stackelberg equilibria; the paper notes that its $\\theta_1$- and $\\theta_2$-expansions were included partly for that reason."],"forward_implications":["In any game with positive spectra for $\\Lambda_\\phi$ and $\\Lambda_\\psi$, every cooperative SVO-Nash equilibrium is trapped inside the intersection of four ellipsoids, so the classic outcomes act as guaranteed bounds on cooperative play.","If even one negative real eigenvalue exists, the SVO-Nash equilibrium fails to exist as a finite action at finitely many cooperation values, and the explicit directions describe exactly how trajectories will be torn apart near those values.","The expansions turn the two-parameter equilibrium search into a sweep along one-dimensional curves, so predicting equilibria for many cooperation levels only requires solving eigenvalue problems once per curve.","In open-loop linear time-varying trajectory coordination, the blow-up directions in action space map directly to blow-up directions in state space, identifying the spatial patterns that become erratic near a bad cooperation level."],"supporting_citations":[{"why":"Introduces social value orientation to robotics and controls and motivates the SVO cost model in driving scenarios.","marker":"(Schwarting et al., 2019)"},{"why":"Supplies the psychological framework of social value orientation used to define the players' coupled costs.","marker":"(Liebrand and McClintock, 1988)"},{"why":"Further develops the SVO framework that the paper uses to weight own versus opponent cost.","marker":"(McClintock and Allison, 1989)"},{"why":"Defines the Nash equilibrium concept that the paper extends to the SVO-modified costs.","marker":"(Nash et al., 1950)"},{"why":"Provides the standard formulation of two-player linear-quadratic games and open-loop Nash equilibria that the paper builds on.","marker":"(Başar and Olsder, 1998)"},{"why":"Exemplifies SVO-style weighted rewards in autonomous-vehicle coordination, which motivates the analysis of cooperative equilibria.","marker":"(Toghi et al., 2021)"}],"fun_headline_variants":["Cooperation can send game equilibria to infinity","SVO-Nash equilibria can blow up, not just converge","Game theory: cooperative play may yield unbounded solutions","When players cooperate too much, equilibria escape to infinity","Unbounded Nash equilibria emerge from social value orientation"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing assumption is that the matrix products $M H_\\phi^{-1} N^{-1}$ and $M_1 H_\\psi^{-1} M_2^{-1}$ are diagonalizable; the paper states without proof that the results would be similar in the general Jordan case, so if that fails the explicit spectral formulas and blow-up directions may not hold. The standing well-posedness condition $\\theta_i \\le \\bar{\\theta}_i$ is also only assumed informally, not carried into the theorem statements.","fun_headline_variants_meta":{"raw":{"variants":["Cooperation can send game equilibria to infinity","SVO-Nash equilibria can blow up, not just converge","Game theory: cooperative play may yield unbounded solutions","When players cooperate too much, equilibria escape to infinity","Unbounded Nash equilibria emerge from social value orientation"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001188,"raw_usage":{"total_tokens":4979,"prompt_tokens":1095,"completion_tokens":3884,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":711,"completion_tokens_details":{"reasoning_tokens":3804}},"tokens_in":711,"tokens_out":3884,"duration_ms":25059,"temperature":1.0,"reasoning_tokens":3804,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T21:18:15.576503+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a concrete two-player quadratic game at a fixed $\\theta$ and compute $u_\\theta$ two ways: by solving the first-order conditions (5) directly, and by evaluating the spectral expansion (11) using the eigen-decomposition of $M H_\\phi^{-1} N^{-1}$. The formulas must agree for every diagonalizable choice; to test the Jordan claim, construct the game so that $M H_\\phi^{-1} N^{-1}$ is a single non-diagonalizable Jordan block (for instance, with $M=N=I$ and $H_\\phi$ chosen so that the product has a repeated eigenvalue with only one eigenvector) and check whether the expansion still matches the direct solve. A divergence between the two computations, or a failure to diverge at $t = |\\lambda_{\\phi j}|$ when Prop. 7 predicts a blow-up, would refute the claimed extension.","supporting_citations":[{"cited_title":"Social behavior for autonomous vehicles","cited_arxiv_id":null,"evidence_quote":"Introduces social value orientation to robotics and controls and motivates the SVO cost model in driving scenarios."},{"cited_title":"The ring measure of social values: A computerized procedure for assessing individual differences in information processing and social value orientation","cited_arxiv_id":null,"evidence_quote":"Supplies the psychological framework of social value orientation used to define the players' coupled costs."},{"cited_title":"Social value orientation and helping behavior 1","cited_arxiv_id":null,"evidence_quote":"Further develops the SVO framework that the paper uses to weight own versus opponent cost."},{"cited_title":"Non-cooperative games","cited_arxiv_id":null,"evidence_quote":"Defines the Nash equilibrium concept that the paper extends to the SVO-modified costs."},{"cited_title":"Dynamic noncooperative game theory","cited_arxiv_id":null,"evidence_quote":"Provides the standard formulation of two-player linear-quadratic games and open-loop Nash equilibria that the paper builds on."},{"cited_title":"Cooperative autonomous vehicles that sympathize with human drivers","cited_arxiv_id":null,"evidence_quote":"Exemplifies SVO-style weighted rewards in autonomous-vehicle coordination, which motivates the analysis of cooperative equilibria."}],"review_version":1}