{"id":"ed5c0ec5-3b20-4540-9b00-bcb7a363c3ae","arxiv_id":"2509.03205","paper_version":1,"verdict":"REJECT","confidence":"MODERATE","novelty_score":3.0,"correctness_risk":"high","formal_verification":"none","parameter_count":0,"one_line_summary":"For nonsmooth MPECs with tangentially convex data, the paper defines Abadie-type constraint qualifications and GA/GS stationary points via tangential subdifferentials and proves implication and optimality theorems.","lead":"This paper derives optimality conditions for nonsmooth optimization problems with equilibrium constraints, using tangential subdifferentials for tangentially convex functions. It defines generalized Abadie and Zangwill constraint qualifications and two stationary-point concepts, then proves necessary and sufficient conditions and relationships between them.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Bounded tangential subdifferentials do not imply closed multiplier cones, so Theorem 4 and Corollary 3 are unsupported.","rationale":"The reader's weakest assumption identifies exactly the closedness of the cone used in the separation argument, and I agree. The paper's main construction is a standard translation of convexificator MPEC theory to tangential subdifferentials, and Theorem 2 follows the usual template: separate a compact subdifferential set from a closed cone, use ACQ to produce a feasible descent direction, and contradict local optimality. The breakdown is the automatic closedness claim. The counterexample with C = {(x,y): 0≤x≤1, x^2≤y≤1} shows that in finite dimensions a compact convex set containing the origin can generate a nonclosed conic hull, and this C is indeed a tangential subdifferential of a convex function. Thus Corollary 3's inference from boundedness to closedness is invalid, and Theorem 4's unproved assertion that Λ is closed cannot be justified from the stated assumptions. Since the separation step in Theorem 4 depends on closedness of Λ, the GA-stationary necessary condition is unsupported as stated. Theorem 2 itself is not refuted because it explicitly assumes Δ is closed, but the paper claims closedness follows automatically in Theorem 4 and Corollary 3. Additional errors in the examples reinforce the correctness risk: Example 1's directional derivative in direction (1,0) is 0, not d1, because the function is 0 when k2=0; Example 2 lists a nonconvex three-point set as a tangential subdifferential, whereas ∂T J(0) for J=|k1|+k2^2 is the segment [-1,1]×{0}. These are secondary but consistent with the rejection. The framework is likely repairable by adding explicit closedness hypotheses or proving them under stronger assumptions, so the appropriate disposition remains the reader's REJECT.","tokens_in":11481,"tokens_out":13428,"duration_ms":137078,"concrete_test":"Compute cone(C) for C={(x,y): 0≤x≤1, x^2≤y≤1}. For each n, c_n=(1/n,1/n^2)∈C and n c_n=(1,1/n)∈cone(C), so (1,1/n)→(1,0). Show (1,0)∉cone(C): any finite nonnegative combination with y=0 and x=1 would require some contributing c with y>0, because every c in C with x>0 has y≥x^2>0, while c with y=0 have x=0. Then set f(x,y)=h_C(x,y) and verify f is convex, hence tangentially convex, with ∂T f(0)=C. This settles that the implication used in Corollary 3 and the unproved closure assertion in Theorem 4 are false.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The load-bearing weakness is the closedness of the multiplier cone in the separation argument. In the proof of Theorem 4, the cone Λ is asserted to be \"closed and convex\" without proof; Corollary 3 similarly infers closedness of Δ from boundedness of tangential subdifferentials. This inference is false. A compact convex set containing the origin need not have a closed conic hull: for C = {(x,y) : 0≤x≤1, x^2≤y≤1}, each (1,1/n)=n(1/n,1/n^2) lies in cone(C), yet (1,0) does not—any representation with y-coordinate 0 and positive x-coordinate is impossible. C is realizable as the tangential subdifferential of the convex, tangentially convex function f(x,y)=h_C(x,y) at 0. Hence bounded/compact subdifferentials do not force Δ or Λ to be closed. Because the contradiction argument in Theorem 4 strictly separates compact ∂TJ(k*) from −Λ, the asserted GA-stationary conclusion is unsupported; Corollary 3's proof fails for the same reason. Theorem 2's statement is not itself refuted, since it assumes Δ closed, but the paper's claims that closedness is automatic are invalid.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops constraint qualifications, stationary concepts, and optimality conditions for nonsmooth mathematical programs with equilibrium constraints (MPEC) using tangential subdifferentials. It defines generalized standard Abadie (GS-ACQ), MPEC Abadie (MPEC-ACQ), MPEC Zangwill, and weak reverse convex constraint qualifications, introduces generalized strong (GS) and generalized alternative (GA) stationarity, and proves implications among the constraint qualifications. The main necessary optimality result is Theorem 2, which states that a local minimum satisfying GS-ACQ and a closed cone Δ is GS-stationary, and Theorem 4, which states that a local minimum satisfying MPEC-ACQ is GA-stationary. Theorem 5 gives a sufficient optimality condition under generalized convexity assumptions. The paper also provides examples illustrating the results.","tokens_in":11706,"tokens_out":13189,"duration_ms":125664,"significance":"If the results were fully valid, the paper would offer a natural extension of Abadie-type constraint qualifications and stationarity conditions for MPEC to the class of tangentially convex functions, whose tangential subdifferentials are compact and convex and have useful calculus rules. The paper is not circular: the proofs are direct separation arguments from explicit assumptions, and the dependence on cited results for tangential subdifferentials is normal foundational support. However, the main necessary conditions, especially Theorem 4 and Corollary 3, rest on an unjustified closedness assertion for the multiplier cone. This is a load-bearing gap because the separating hyperplane argument cannot be applied unless the cone is closed. Theorem 2 is conditionally sound because it explicitly assumes closedness of Δ, but the paper's claim that closedness is automatic is false. The paper therefore needs a substantive revision before its central claims can be accepted.","major_comments":[{"comment":"The proof states that Λ is closed and convex without any argument. This matters because the separation theorem is the only mechanism producing the multiplier system and the subsequent stationarity conclusion. A compact convex set containing the origin need not have a closed conic hull: for C = {(x,y) : 0≤x≤1, x²≤y≤1}, the points (1,1/n) = n(1/n,1/n²) lie in con C, but (1,0) does not. Such a C is realizable as the tangential subdifferential of the convex, hence tangentially convex, support function h(x)=sup_{c∈C}⟨c,x⟩ at the origin. Therefore boundedness or compactness of tangential subdifferentials does not imply closedness of Λ, and Theorem 4 is unsupported as stated; an explicit closedness assumption for Λ is required.","section":"§4, Theorem 4, proof of Eq. (13)"},{"comment":"The proof of Corollary 3 asserts that Δ is closed because the tangential subdifferentials are bounded. The same counterexample as above shows that a nonempty compact convex set containing the origin can have a non-closed conic hull. Hence the corollary is not established, and the claimed automatic closedness is false. Theorem 2 itself is conditionally sound because its statement explicitly assumes Δ is closed; the defect is the claim that closedness follows from the standing assumptions.","section":"§4, Corollary 3"},{"comment":"After equation (14), the proof sets λ^G_Θ = λ^H_Υ = μ^G_Υ = μ^H_Θ = 0, but Definition 10 requires λ^G_Υ = λ^H_Θ = 0 for GA-stationarity. The preceding sentence contains the correct conditions, so this appears to be a typo; nevertheless, as written the proof's conclusion does not match the definition being proved.","section":"§4, Theorem 4, final displayed conditions"}],"minor_comments":[{"comment":"For J(k1,k2)=|k1|+k2² at (0,0), the tangential subdifferential is the segment {(t,0) : -1≤t≤1}, not just the three listed points {(1,0),(-1,0),(0,0)}.","section":"§2, Example 2"},{"comment":"The symbol β appears in the definition of GS-stationarity but is never defined; from the proof of Theorem 2 it is evidently meant to be the degenerate set Ω.","section":"§4, Definition 11"},{"comment":"The proof says that Δ is 'closed and bounded'; a nonzero cone is never bounded. The separation argument needs only closedness, so this should be corrected to avoid confusion.","section":"§4, Theorem 2 proof"},{"comment":"The statement's condition 'ΩG_μ ∪ μH_i ∪ Θ+_μ ∪ Υ+_μ = φ' is garbled; it should presumably be ΩG_μ ∪ ΩH_μ ∪ Θ+_μ ∪ Υ+_μ = ∅. There is also a typo 'pseudoconvax' for 'pseudoconvex'.","section":"§4, Theorem 5"},{"comment":"The remark says that for a smooth function 'any set that contains the gradient' is the tangential subdifferential; by Definition 3 the tangential subdifferential of a smooth function is exactly {∇J(k)}, so the wording should be corrected.","section":"§3, Remark 2"}],"recommendation":"major_revision","confidential_remarks":"The paper is not ready in its current form because Theorem 4 and Corollary 3 depend on a closedness assertion that is both unproved and, for Corollary 3, false. I do not recommend outright rejection, because Theorem 2 is conditional on an explicit closedness assumption and the proofs are otherwise direct; a revision that adds the same closedness assumption to Theorem 4 and removes or corrects Corollary 3 could make the central necessary-condition result sound. However, if the authors wish to keep the MPEC-ACQ-only formulation, a substantially different argument would be needed."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear colleague,\n\nQuick take: this is a workmanlike translation of the convexificator-based MPEC optimality theory in Ansari-Ardali et al. into the language of tangential subdifferentials. The main definitions and proof template are not new, but the specific statements—GS-ACQ/MPEC-ACQ in terms of ∂T, GA/GS stationarity, the relationships between CQs, and a sufficient optimality result—are stated and proved for this subdifferential, which is legitimate incremental progress. If correct, they would be a modest but useful contribution for people working on nonsmooth MPECs.\n\nWhat is good: Theorem 2 is conditionally sound. It assumes the cone Δ is closed, and given that assumption the separation argument is standard and the GS-stationarity conclusion follows. The paper is not circular; it uses tangential subdifferential calculus as foundational support, and it explicitly notes that the CQs reduce to the smooth versions when gradients replace subdifferentials.\n\nThe soft spots are real and load-bearing.\n\n- Theorem 4 asserts Λ = cone(ℓ∪ℏ∪GΘ∪HΥ∪(GH)Ω∪GΩ) is \"closed and convex\" without proof. That is not automatic. A cone over a compact convex set containing the origin need not be closed: for C = {(x,y): 0≤x≤1, x²≤y≤1}, each (1,1/n) lies in cone(C) but (1,0) does not, and such a set can be realized as a tangential subdifferential of a convex tangentially convex function at a point. So the separation step in Theorem 4 is unsupported.\n- Corollary 3 says Δ is closed because tangential subdifferentials are bounded. That inference is false for the same reason. The corollary collapses.\n- Examples 1 and 2 are inconsistent with Definition 3. In Example 2, ∂TJ(0) is listed as three points, but the set { (a,0): |a|≤1 } is required to be convex; in Example 1, the claimed directional derivative d1 is not correct in directions with d2=0, since the function vanishes on the k2-axis.\n- Minor issue: the proof of Theorem 2 calls Δ \"closed and bounded,\" but a nonzero cone is never bounded. The assumption is only closedness, so the proof should say that.\n\nNet: the central theorem can be repaired by adding explicit closedness assumptions, but as written Theorem 4 and Corollary 3 are false as stated, and the examples undermine credibility.\n\nWho is this for: researchers working on constraint qualifications for nonsmooth MPECs who want a tangential-subdifferential version of known results. Not a broad audience.\n\nI would send it to a referee rather than desk reject, because the conditional Theorem 2 and the translation framework are substantive enough to warrant careful checking. But I would not accept as-is; the closedness issue and examples need to be fixed before the paper is publishable.\n\nBest.","headline":"A workmanlike translation of convexificator-based MPEC optimality conditions to tangential subdifferentials; the conditional main theorem is plausible, but Theorem 4 and Corollary 3 rely on false closedness claims and the examples contain basic errors.","tokens_in":12200,"tokens_out":3892,"would_cite":false,"duration_ms":34635,"reading_group":"no","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["90C26","90C46","49K99"],"pacs":[],"model":"deepseek-v4-flash","headline":"For tangentially convex nonsmooth MPECs, Abadie-type constraint qualifications imply generalized stationary conditions; with generalized convexity, stationarity becomes global optimality.","keywords":["mathematical programs with equilibrium constraints","tangential subdifferentials","constraint qualifications","stationary point","nonsmooth analysis","generalized convexity","Abadie constraint qualification","optimality conditions"],"falsifier":"Take the compact convex set $K=\\{(x,y):(x-1)^2+y^2\\le 1\\}$, which contains the origin; its conic hull $\\operatorname{cone}K=\\{(x,y):x>0\\}\\cup\\{(0,0)\\}$ is not closed. A tangentially convex MPEC whose active constraint subdifferentials generate such a cone therefore escapes the closedness hypothesis of Theorem 2; checking whether a local minimizer at that point still satisfies GS-ACQ and admits GS-multipliers would determine whether the closedness condition can be dropped or whether the claimed necessity fails.","tokens_in":11285,"feed_emoji":"📐","tokens_out":12100,"duration_ms":108910,"temperature":0.7,"pith_summary":"The paper studies optimization problems whose constraints include a complementarity condition—mathematical programs with equilibrium constraints—when the objective and constraint functions are tangentially convex but not necessarily differentiable. It claims that the usual Abadie-type constraint qualifications can be rephrased using tangential subdifferentials, and that these conditions certify stationarity of local minimizers: a generalized strong stationarity under the GS-Abadie condition, and a generalized alternative stationarity under the MPEC-Abadie condition. It further claims that, under a generalized quasiconvexity assumption, the alternative stationarity condition is sufficient for global optimality. The interest is that the tangentially convex class covers convex, differentiable, and Clarke-regular functions, so a single framework yields multiplier certificates across a broad family of nonsmooth equilibrium-constrained problems.","feed_headline":"Abadie-type conditions yield stationarity in nonsmooth MPECs","feed_subtitle":"Tangential subdifferentials extend the usual MPEC stationarity theory to nondifferentiable programs.","key_machinery":"The central object is the tangential subdifferential of a tangentially convex function: $\\partial_T J(k)=\\{\\xi:\\langle\\xi,d\\rangle\\le J'(k,d)\\ \\forall d\\}$, a nonempty compact convex set whose support function recovers the directional derivative. Around this object the paper builds dual constraint cones—$\\Pi(k^*)$ and $\\Psi(k^*)$ from unions of tangential subdifferentials of the active inequality, equality, and complementarity constraints—and defines GS-ACQ and MPEC-ACQ by requiring $\\Pi(k^*)\\subseteq T(K,k^*)$ or $\\Psi(k^*)\\subseteq T(K,k^*)$. The argument then runs on convex separation: if $0\\notin \\partial_T J(k^*)+\\Delta$, separation produces a direction $\\nu$ with negative directional derivative for $J$; GS-ACQ places $\\nu$ in the contingent cone of the feasible set, and local Lipschitzness turns the contingent direction into feasible points with strictly lower objective values, contradicting local optimality. The sufficient-optimality part uses $\\partial_T$-pseudoconvexity and $\\partial_T$-quasiconvexity to convert the multiplier inequality into a global comparison $J(k)\\ge J(k^*)$.","core_discovery":"On the paper's own terms, the central discovery is that the tangential subdifferential $\\partial_T J(k)=\\{\\xi\\in\\mathbb{R}^n:\\langle\\xi,d\\rangle\\le J'(k,d)\\ \\text{for all } d\\}$—a nonempty compact convex set whose support function is the directional derivative—can carry the entire MPEC stationarity theory. Theorem 2 proves that if $k^*$ is a local solution, $J$ is tangentially convex and locally Lipschitz at $k^*$, GS-ACQ holds at $k^*$, and the cone $\\Delta$ generated by the active constraint subdifferentials is closed, then $k^*$ is a GS-stationary point. Theorem 4 proves the analogous implication from MPEC-ACQ to GA-stationarity, and Theorem 5 shows that GA-stationarity plus $\\partial_T$-pseudoconvexity of the objective and $\\partial_T$-quasiconvexity of the active constraints makes $k^*$ globally optimal. The paper also establishes a chain among constraint qualifications: MPEC weak reverse convex CQ implies MPEC Zangwill CQ, which implies MPEC Abadie CQ, and GS-ACQ implies MPEC-ACQ.","pith_inferences":["The closedness gap suggests a natural repair: replace $\\Delta$ or $\\Lambda$ by its closure and aim for approximately stationary multipliers, or impose a separate closed-cone constraint qualification; the theorems would then survive in a limiting form.","The same machinery might extend to semidefinite or infinite-dimensional equilibrium constraints, since tangential subdifferentials remain compact and convex there, but separation arguments would need a closed cone in the ambient topology.","Because the sufficiency theorem only needs the generalized convexity on the active index sets, it may apply to complementarity reformulations of bilevel problems where the upper-level objective is mildly nonconvex.","A sharper testable extension is to construct an MPEC whose active constraint subdifferentials generate a nonclosed cone and check whether MPEC-ACQ plus a local minimum can coexist with the absence of GA-multipliers; if it can, Theorem 4 as stated is false."],"forward_implications":["Local minimizers of nonsmooth tangentially convex MPECs satisfying GS-ACQ with closed $\\Delta$ are certified by GS-stationarity multipliers, giving a KKT-type certificate without differentiability.","When only MPEC-ACQ is available, GA-stationarity still holds, so weaker regularity still forces a multiplier condition at local minima.","Because GS-stationary points are GA-stationary, the two necessary conditions form a hierarchy in which stronger geometric regularity yields stronger multipliers.","The implication chain among constraint qualifications lets practitioners verify the weakest geometric assumption, for instance checking MPEC weak reverse convex CQ, and then inherit MPEC-ACQ stationarity conclusions.","Under $\\partial_T$-pseudoconvex and $\\partial_T$-quasiconvex hypotheses, GA-stationarity is both necessary and sufficient for global optimality, so the multiplier test completely characterizes solutions in that class."],"supporting_citations":[{"why":"Supplies the definition and convex calculus of tangential subdifferentials, the objects that replace derivatives throughout.","marker":"[22]"},{"why":"Provides the convexificator-based MPEC constraint qualifications and stationary concepts that the paper adapts to tangential subdifferentials.","marker":"[2]"},{"why":"Introduces the stationarity hierarchy (S-, M-, A-, C-stationarity) that GS and GA stationarity generalize.","marker":"[32]"},{"why":"Introduced MPEC Abadie constraint qualifications for nonsmooth MPEC, the model for Definition 7.","marker":"[26]"},{"why":"Origin of the smooth MPEC Abadie-type approach that the paper extends.","marker":"[6]"},{"why":"Established the class of tangentially convex functions and its role in nonsmooth optimization.","marker":"[17]"},{"why":"Source of the weak reverse convex constraint qualification used in Theorem 1.","marker":"[35]"},{"why":"The Zangwill constraint qualification whose MPEC analogue is compared to Abadie-type conditions.","marker":"[37]"}],"fun_headline_variants":["Tangential subdifferentials unlock optimality for nonsmooth MPECs","Nonsmooth MPEC theory sharpened by tangential subdifferentials","Tangential subdifferentials yield global optimality in nonsmooth MPECs","Constraint qualifications for MPECs via tangential subdifferentials","Extending MPEC stationarity to nondifferentiable programs"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the constraint cone $\\Delta$ (and, in Theorem 4, $\\Lambda$) built from tangential subdifferentials is closed, because the convex separation step needs closedness to produce stationarity multipliers; boundedness of the subdifferentials alone does not guarantee the cone is closed.","fun_headline_variants_meta":{"raw":{"variants":["Tangential subdifferentials unlock optimality for nonsmooth MPECs","Nonsmooth MPEC theory sharpened by tangential subdifferentials","Tangential subdifferentials yield global optimality in nonsmooth MPECs","Constraint qualifications for MPECs via tangential subdifferentials","Extending MPEC stationarity to nondifferentiable programs"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001126,"raw_usage":{"total_tokens":4654,"prompt_tokens":891,"completion_tokens":3763,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":507,"completion_tokens_details":{"reasoning_tokens":3672}},"tokens_in":507,"tokens_out":3763,"duration_ms":22604,"temperature":1.0,"reasoning_tokens":3672,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T16:34:39.929467+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take the compact convex set $K=\\{(x,y):(x-1)^2+y^2\\le 1\\}$, which contains the origin; its conic hull $\\operatorname{cone}K=\\{(x,y):x>0\\}\\cup\\{(0,0)\\}$ is not closed. A tangentially convex MPEC whose active constraint subdifferentials generate such a cone therefore escapes the closedness hypothesis of Theorem 2; checking whether a local minimizer at that point still satisfies GS-ACQ and admits GS-multipliers would determine whether the closedness condition can be dropped or whether the claimed necessity fails.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the definition and convex calculus of tangential subdifferentials, the objects that replace derivatives throughout."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the convexificator-based MPEC constraint qualifications and stationary concepts that the paper adapts to tangential subdifferentials."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Introduces the stationarity hierarchy (S-, M-, A-, C-stationarity) that GS and GA stationarity generalize."},{"cited_title":"Set Valued Anal","cited_arxiv_id":null,"evidence_quote":"Introduced MPEC Abadie constraint qualifications for nonsmooth MPEC, the model for Definition 7."},{"cited_title":"Institute of Applied Mathematics and Statistics, University of Wurzburg","cited_arxiv_id":null,"evidence_quote":"Origin of the smooth MPEC Abadie-type approach that the paper extends."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Established the class of tangentially convex functions and its role in nonsmooth optimization."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Source of the weak reverse convex constraint qualification used in Theorem 1."},{"cited_title":"(1969) 17","cited_arxiv_id":null,"evidence_quote":"The Zangwill constraint qualification whose MPEC analogue is compared to Abadie-type conditions."}],"review_version":1}