{"id":"9544d2ba-b3aa-46fe-9cca-c3e7cbf39cd7","arxiv_id":"2512.16141","paper_version":2,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":5.0,"correctness_risk":"high","formal_verification":"none","parameter_count":0,"one_line_summary":"For non-monotone variational inequalities, a solution exists if the normal mapping is norm-coercive and its generalized Jacobian has full rank off the zero set.","lead":"This paper gives new mathematical conditions that guarantee a solution exists for variational inequalities that are not monotone, using properties of the associated normal mapping. The authors apply these conditions to games, showing when a quasi-Nash or Nash equilibrium exists even when standard P-function assumptions fail.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The constrained-set verification chain relies on a false projection-Jacobian containment; for K=[0,∞)^2 it is violated, and the true projection calculus can make ∂(tF)_K^nor singular at nonzero points for every t>0.","rationale":"The paper's central theorem, Theorem 6, is a plausible extension of the unconstrained result and its proof via the Clarke inverse function theorem is coherent; I do not see a defect in that argument. The advertised value beyond the authors' prior work, however, is the constrained VI and game application, and that application is routed through Theorem 7 and the paragraph claiming that ∂Π_K[x]⊆Conv(G) is automatic for Cartesian products of one-dimensional convex sets. The reader's weakest-assumption identifies exactly this step, and my independent check confirms it is false: for K=[0,∞)^2, ∂Π_K(0) contains diag(0,0), which cannot lie in Conv(G). This alone invalidates the stated verification chain. I also checked whether the conclusion of Theorem 7 could survive on the true projection calculus: for F(x)=[[1,2],[3,1]]x on K=[0,∞)^2, the generalized Jacobian of (tF)_K^nor at x=(1,0) contains a singular matrix for every t>0 while (tF)_K^nor(x)≠0, despite ∇F being constant, nonsingular, and with nonzero principal minors. So the false containment is not a harmless over-simplification; the specific maximal-rank conclusion needed for constrained games fails for a simple Cartesian box. The unconstrained Example 1 and the proof of Theorem 6 are independent of this issue, so the paper is not without merit. But the constrained game-theoretic claims in Section 4 are conditional on repairing Theorem 7. Since the reader already assigned CONDITIONAL with high confidence and identified the same weak point, no verdict change is warranted.","tokens_in":21736,"tokens_out":20163,"duration_ms":204313,"concrete_test":"Analytic check that settles both the false containment and its consequence: (1) For K=[0,∞)^2, compute ∂Π_K(0) by sequences from the negative quadrant to confirm diag(0,0)∈∂Π_K(0); verify that every element of Conv(G) has diagonal entries summing to at least 1. (2) Let A=[[1,2],[3,1]], K=[0,∞)^2, and fix arbitrary t>0. At x=(1,0), take γ=1/(5t+1) and J=I+(tA−I)diag(1,γ). Confirm γ∈(0,1), diag(1,γ)∈∂Π_K(x), (tF)_K^nor(x)=(1+t,3t)≠0, and det J=0. This shows the actual projection calculus for a Cartesian product produces a singular generalized Jacobian at a nonzero normal-map value, for every t>0.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's advertised extension from unconstrained VIs to constrained/game VIs is carried by Theorem 7, whose key assumption is ∂Π_K[x] ⊆ Conv(G) for x∉K°, with G={I−e_i e_i^T : i=1,…,m}∪{I}. Immediately after Theorem 7 the paper asserts this holds when K=K_1×⋯×K_m is a Cartesian product of one-dimensional closed convex sets. That assertion is false. For K=[0,∞)^2 at x=(0,0), Π_K(y)=(max(y1,0),max(y2,0)), so ∂Π_K(0) contains diag(0,0) (approach from the negative quadrant). But every diagonal matrix in Conv(G) has the form diag(1−β α_1, 1−β α_2) with α_i≥0, α_1+α_2=1, β∈[0,1], hence trace ≥1; diag(0,0) has trace 0 and is not in Conv(G). The gap is not merely a missing proof. With the true projection Jacobian and F(x)=Ax, A=[[1,2],[3,1]], on K=[0,∞)^2, at x=(1,0) the matrix J=I+(tA−I)diag(1,γ) belongs to ∂(tF)_K^nor(x) for every γ∈[0,1]; det J = t(1−γ(5t+1)), which vanishes at γ=1/(5t+1)∈(0,1) for every t>0, while (tF)_K^nor(x)=(1+t,3t)≠0. Thus no t>0 makes the generalized Jacobian maximal-rank everywhere, even though ∇F is constant, full rank, with all principal minors nonzero. The constrained existence and quasi-Nash consequences therefore do not follow from the stated arguments; Theorem 6 itself is not implicated.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies existence of solutions to non-monotone variational inequalities VI(K,F) through the normal mapping F_K^nor(x)=x-Π_K[x]+F(Π_K[x]). Theorem 6 proves that if F_K^nor is norm coercive and its generalized Jacobian has maximal rank at every point where F_K^nor(x)≠0, then VI(K,F) has a solution. The paper then provides sufficient conditions for these hypotheses in terms of uniform P-function and uniform P-matrix conditions, and applies the results to games, claiming existence of quasi-Nash and Nash equilibria for constrained action sets. The central mechanism for the constrained case is Theorem 7, which needs the assumption ∂Π_K[x]⊆Conv(G) for x∉K°, where G={I−e_i e_i^⊤}∪{I}, and the paper asserts this assumption holds for Cartesian products of one-dimensional closed convex sets. The rest of the paper, including Section 4, builds on this assertion to obtain game-theoretic consequences.","tokens_in":22218,"tokens_out":6750,"duration_ms":72040,"significance":"If correct, the main existence result Theorem 6 would be a useful and clean extension of the unconstrained result in Theorem 1, and the paper contains instructive examples showing that existing uniform P-function and PΥ-matrix conditions fail for simple linear mappings. The proof of Theorem 6 itself is a clean application of Clarke's inverse mapping theorem and is not implicated in the main defect. However, the advertised extension to constrained VIs and games is carried by Theorem 7 and the verification chain based on ∂Π_K[x]⊆Conv(G), and that assumption is false. The specific counterexample K=[0,∞)^2 shows the claimed theorem-level conclusion also fails for a simple linear, full-rank, constant-Jacobian mapping. Consequently the central constrained-set and game-theoretic claims of the paper are not established.","major_comments":[{"comment":"The assertion that ∂Π_K[x]⊆Conv(G) holds when K is a Cartesian product of one-dimensional closed convex sets is false. For K=[0,∞)^2 at x=(0,0), the projection is Π_K(y)=(max(y1,0),max(y2,0)); approaching from the negative quadrant gives diag(0,0)∈∂Π_K(0). But every matrix in Conv(G) has the form I−β diag(α_1,...,α_m) with α_i≥0, Σα_i=1, β∈[0,1], hence trace ≥ m−1 = 1, while diag(0,0) has trace 0. This invalidates the constrained-set verification chain used to extend Theorem 6 to box/interval action sets and to the game results in Section 4.","section":"§3.3, Theorem 7 and the paragraph after it"},{"comment":"The intended conclusion of Theorem 7 is itself false for a natural case. Let K=[0,∞)^2 and F(x)=Ax with A=[[1,2],[3,1]]. Then ∇F is constant, full rank, and all principal minors are nonzero; however, at x=(1,0), (tF)_K^nor(x)=(1+t,3t)≠0. For every γ∈[0,1], the matrix J=I+(tA−I)diag(1,γ) lies in ∂(tF)_K^nor(x), and det J=t(1−γ(5t+1)). Choosing γ=1/(5t+1)∈(0,1) gives a singular matrix in the generalized Jacobian for every t>0. Thus no single t>0 makes ∂(tF)_K^nor(x) maximal rank everywhere, despite F satisfying all the Jacobian-side hypotheses in Theorem 7 except the false Conv(G) containment. This shows the gap is not a missing proof but a fundamental problem for a basic class of constrained sets.","section":"§3.3, Theorem 7"},{"comment":"The proof uses the step: for a vector w with w_i=v_i on M and w_i=0 off M, 'for some small enough ϵ>0, x+ϵw(x)∈K or x−ϵw(x)∈K'. This is false for Cartesian products of intervals when w has mixed signs. For K=[0,∞)^2, x=(0,0), and w=(1,−1), neither x+ϵw=(ϵ,−ϵ) nor x−ϵw=(−ϵ,ϵ) belongs to K. Hence the constructed y_ϵ may not lie in K, and the mean-value / uniform-P-function argument does not go through as written. Even if the final statement can be repaired by a more careful directional argument, the proof is currently invalid.","section":"§2.2, Theorem 3(a), proof"},{"comment":"The game-theoretic existence conclusions rely on the sentence that 'the conditions of Theorem 7 are satisfied' for Cartesian products of one-dimensional action sets. Since the ∂Π_K[x]⊆Conv(G) containment is false and Theorem 7's conclusion fails for the linear counterexample above, the chain leading to existence of quasi-Nash equilibria for constrained games is not established. The only worked example, Example 2, uses K=R^2, i.e., the unconstrained case, so it does not illustrate the constrained/game claim.","section":"§4, paragraph after Theorem 9"}],"minor_comments":[{"comment":"The reference to 'Clark Inverse' should be 'Clarke's inverse function theorem'.","section":"§3.1, proof of Theorem 6"},{"comment":"There is a typo in 'K=K_1×...×K_N n'; the intended dimension is ar n or a similar notation.","section":"§4.1, proof of Theorem 9"},{"comment":"The figures are referenced only loosely; it would help to state the corresponding implication formally in the surrounding text.","section":"Figures 1 and 2"},{"comment":"Both examples are unconstrained (K=R^2) and therefore do not test the paper's constrained-set theorem; this could be stated explicitly so readers are not misled.","section":"Examples 1 and 2"}],"recommendation":"reject","confidential_remarks":"The paper contains a sound and elegant Theorem 6, but the advertised extension to constrained VIs and games is invalid because Theorem 7 rests on a false projection-Jacobian containment. The explicit counterexample for K=[0,∞)^2 with a constant full-rank linear F also shows that the theorem's conclusion fails in a basic case, so a local proof patch is not feasible. If the authors later restrict the claims to unconstrained VIs or find a genuinely different way to verify maximal rank for polyhedral K, a revised submission could be worth considering; as it stands, the central claim is not established."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Punchline: Theorem 6, the main existence result, is a genuine and correctly proved extension of the authors' earlier unconstrained theorem to closed convex K via the normal mapping. The paper's bridge from that theorem to constrained and game settings — Theorem 7, carried by the claim that ∂Π_K[x] ⊆ Conv(G) when K is a Cartesian product of one-dimensional closed convex sets — is broken. The claim is asserted without proof right after Theorem 7, and it is false.\n\nFor K=[0,∞)^2 at x=(0,0), ∂Π_K[x] contains diag(0,0), but every matrix in Conv(G) has trace at least m−1 = 1. That is not a subtle miss: the corners of a box are exactly where multiple coordinates become active simultaneously, and the trace gap is structural. The follow-through fails too. Take F(x)=Ax with A=[[1,2],[3,1]] on K=[0,∞)^2. ∇F ≡ A is constant, full rank, all principal minors nonzero; all of Theorem 7's Jacobian hypotheses hold. At x=(1,0), a direct computation of the projection Jacobian gives ∂(tF)_K^nor(x) = Conv{[[t,0],[3t,1]], [[t,2t],[3t,t]]}. For every t>0, the convex combination with weight 1/(1+5t) is singular, while (tF)_K^nor(x) = (t,3t) ≠ 0. So no rescaling t>0 makes the generalized Jacobian maximal rank, and the constrained existence and quasi-Nash consequences do not follow from the stated arguments. The paper flags that the containment needs validation ('Later we will discuss when this assumption is valid') and then validates it with the false assertion. That is the load-bearing gap.\n\nWhat is good: Theorem 6's proof is clean — the inf-argument plus Clarke's inverse theorem is the right tool. Theorems 2 and 3, giving uniform lower bounds on singular values of principal submatrices from uniform P-matrix/P-function conditions, are new relative to the cited literature and look correctly proved. Examples 1 and 2 correctly demonstrate that the uniform P-function and PΥ-matrix conditions fail for a simple linear map, so the motivating examples are honest. Lemma 2's coercivity argument for Cartesian blocks is fine.\n\nThe gap is specific and possibly fixable: replace the false containment with correct projection-Jacobian calculus for boxes, or prove maximal rank directly for ∂(tF)_K^nor on the boundary. But as it stands, the constrained half of the paper should not be relied on.\n\nWho gets value: readers working on unconstrained/normal-mapping existence theory and on singular-value consequences of P-properties. Worth a serious referee — this is a paper with a sound core, a clear error, and a plausible repair path. I'd send it to review, expecting major revision of Section 3.3 and the dependent game consequences.","headline":"Theorem 6 is a clean, correct extension of the authors' earlier unconstrained result, but the constrained/game applications rest on a projection-Jacobian containment claim that is false for boxes.","tokens_in":22671,"tokens_out":12567,"would_cite":false,"duration_ms":107755,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["49J40","91A10","90C33"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper proves that a coercive normal map whose generalized Jacobian has full rank yields a solution to any non-monotone variational inequality, then applies this to games and Nash equilibria.","keywords":["variational inequalities","normal mapping","generalized Jacobian","norm coercivity","non-monotone VIs","Nash equilibrium","quasi-Nash equilibrium","P-matrix condition"],"falsifier":"Check K=[0,∞)^2 at x=(0,0). Sequences approaching the origin from the interior give projection Jacobian I, while sequences from the strictly negative quadrant give diag(0,0); so diag(0,0) ∈ ∂Π_K[0]. But every matrix in Conv({I − e_1e_1^T, I − e_2e_2^T, I}) has trace at least 1, whereas diag(0,0) has trace 0. Exhibiting diag(0,0) as a convex combination of matrices in G would refute this; otherwise Theorem 7 cannot be applied as stated to box action sets.","tokens_in":21614,"feed_emoji":"🎯","tokens_out":5787,"duration_ms":57483,"temperature":0.7,"pith_summary":"The paper is trying to establish a new existence theorem for variational inequalities whose defining map is not monotone. The central result says: if the normal map F^nor_K(x) = x − Π_K[x] + F(Π_K[x]) is norm coercive and its generalized Jacobian has maximal rank wherever F^nor_K is nonzero, then VI(K,F) has a solution. The proof reduces the constrained problem to finding a zero of the normal map, then uses a Clarke inverse function theorem to rule out a positive minimum of the norm. The paper then gives conditions on F and K — uniform P-function, uniform P-matrix, and PΥ-matrix conditions — under which these hypotheses hold, and shows in games that those conditions force strong convexity of each player's cost in her own variable, so solutions are Nash equilibria. A sympathetic reader will care because this offers an existence route for non-monotone problems where classical monotonicity-based results are silent.","feed_headline":"Normal-map rank condition proves solutions to non-monotone VIs","feed_subtitle":"A coercive normal map with full-rank generalized Jacobian forces a solution; game consequences follow.","key_machinery":"The normal mapping F^nor_K(x) = x − Π_K[x] + F(Π_K[x]) converts the constrained VI into an unconstrained root-finding problem: x* solves VI(K,F) iff F^nor_K(v)=0 for some v with x* = Π_K[v]. The generalized Jacobian ∂F^nor_K — the convex hull of limits of Jacobians at nearby differentiable points — together with Clarke's inverse function theorem supplies local invertibility off the zero set. A scaling trick (replace F by tF, t>0) is used to force the generalized Jacobian of the normal map to have full rank on the boundary while preserving the solution set.","core_discovery":"The paper's central claim is Theorem 6: for a nonempty closed convex set K and continuously differentiable F, if the normal mapping F^nor_K(x) = x − Π_K[x] + F(Π_K[x]) is norm coercive and its generalized Jacobian has maximal rank at every x where F^nor_K(x) ≠ 0, then VI(K,F) has a solution. The proof minimizes the norm of the normal map and uses the Clarke inverse function theorem to show the minimum cannot be positive. The paper also gives conditions on F and K under which these hypotheses are met, and in the game setting shows that the uniform P-matrix and PΥ-matrix conditions imply each player's cost is strongly convex in her own variable, so every solution to the VI formulation is actua","pith_inferences":["The paper's claimed extension to box/interval action sets relies on the assertion after Theorem 7 that ∂Π_K[x] ⊆ Conv(G) for Cartesian products of one-dimensional closed convex sets; that assertion has no proof in the paper and fails at the origin of [0,∞)^2, so the chain from Theorem 6 to box-constrained Nash existence is currently incomplete.","A concrete testable repair would be to prove or disprove the inclusion ∂Π_K[x] ⊆ Conv(G) for polyhedral sets; the trace obstruction suggests the inclusion may hold only for very special sets, or that a different rank argument is needed on the boundary.","The scaling-by-t argument suggests a computational avenue: instead of checking P-properties, one could certify coercivity and full-rank for large t, then use homotopy or continuation to locate the root of the normal map.","In nonconvex games, the PŁ-type condition of Theorem 11 upgrades quasi-Nash equilibria to Nash equilibria, so combining a quasi-Nash existence certificate with gradient dominance may give Nash existence without convexity."],"forward_implications":["Theorem 6 gives a new existence certificate: check norm coercivity of the normal map plus full rank of its generalized Jacobian, and a solution to the constrained VI exists without monotonicity.","Under the uniform P-function property on a Cartesian product set, all principal submatrices of ∇F have singular values uniformly bounded below, so the hypotheses of Theorem 7 are met after scaling by t.","In games, the uniform P-matrix condition on the game Jacobian implies each player's cost is strongly convex in her own variable; hence quasi-Nash equilibria coincide with Nash equilibria, and existence follows.","The PΥ-matrix condition yields a unique Nash equilibrium without separately requiring Lipschitz gradients or convexity of costs.","Examples show the conditions of Theorem 6 can be checked where uniform P-function and PΥ-matrix conditions fail, so the existence result covers VIs those criteria miss."],"fun_headline_variants":["Full-rank normal map solves non-monotone VIs and games","Coercive normal map with full-rank Jacobian forces VI solutions","Non-monotone VIs solved via full-rank normal mapping","New VI existence proof leads to Nash equilibrium conditions"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The whole path from Theorem 6 to box/interval-constrained games passes through the assertion, made after Theorem 7 without proof, that for x outside the interior of a Cartesian product of one-dimensional closed convex sets the generalized Jacobian of the projection satisfies ∂Π_K[x] ⊆ Conv({I − e_i e_i^T} ∪ {I}); this is false for K=[0,∞)^2 at x=(0,0), so that path collapses.","fun_headline_variants_meta":{"raw":{"variants":["Full-rank normal map solves non-monotone VIs and games","Coercive normal map with full-rank Jacobian forces VI solutions","Non-monotone VIs solved via full-rank normal mapping","New VI existence proof leads to Nash equilibrium conditions"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000132,"raw_usage":{"total_tokens":963,"prompt_tokens":729,"completion_tokens":234,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":473,"completion_tokens_details":{"reasoning_tokens":164}},"tokens_in":473,"tokens_out":234,"duration_ms":3485,"temperature":1.0,"reasoning_tokens":164,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-03T15:38:33.070490+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Check K=[0,∞)^2 at x=(0,0). Sequences approaching the origin from the interior give projection Jacobian I, while sequences from the strictly negative quadrant give diag(0,0); so diag(0,0) ∈ ∂Π_K[0]. But every matrix in Conv({I − e_1e_1^T, I − e_2e_2^T, I}) has trace at least 1, whereas diag(0,0) has trace 0. Exhibiting diag(0,0) as a convex combination of matrices in G would refute this; otherwise Theorem 7 cannot be applied as stated to box action sets.","supporting_citations":[],"review_version":1}