Pith. sign in

REVIEW 2 major objections 4 minor 39 references

The paper's central claim is that radial epiderivatives — an infimum-based replacement for derivatives — make Fritz John and KKT conditions able to certify global, not just local, minima for nonsmooth nonconvex inequality-constrained proble

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-05 12:46 UTC pith:KGHH74YY

load-bearing objection The main necessity theorem is unproved because the proof applies Gordan's lemma to conic combinations without any additivity property; the sufficiency half is worth a look. the 2 major comments →

arxiv 2509.01272 v1 pith:KGHH74YY submitted 2025-09-01 math.OC

Radial Epiderivative Based Fritz John and KKT Conditions in Nonsmooth Nonconvex Optimization

Refail Kasimbeyli , Jian-Wen Peng , Jen-Chih Yao This is my paper
classification math.OC MSC 49J5290C2690C30
keywords radial epiderivativenonsmooth optimizationnonconvex optimizationglobal optimality conditionsFritz John conditionsKKT conditionsconstraint qualificationdiscrete feasible sets
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper aims to establish global optimality conditions for nonsmooth and nonconvex constrained optimization using the radial epiderivative, a derivative-like construction based on infima rather than limits. It defines a radial gradient vector and a feasible-direction cone built from the radial cone, then proves Fritz John and KKT-type theorems that certify global minima without requiring convexity, classical differentiability, or even a continuous feasible set. The main claim is that Theorem 12 gives necessary FJ conditions and Theorem 13 gives sufficient KKT conditions for global optimality, with versions that avoid using active constraints altogether. A sympathetic reader would care because classical KKT theory is local and needs differentiability, whereas this framework promises certificates for genuinely nonlocal, discrete, and nonsmooth problems.

Core claim

For the problem of minimizing f(x) subject to gi(x) ≤ 0 with all functions radially epidifferentiable, the paper introduces the cone of feasible directions D(x), the set of radial descent directions, and the radial gradient vector whose entries are radial epiderivatives evaluated along a basis of D(x). Theorem 12 states that if x is a global minimum and Assumption 1 holds, there exist multipliers v0, v1, ..., vm ≥ 0, not all zero, satisfying v0∇^r_d f(x) + Σ vi∇^r_d gi(x) ≥ 0; if the radial gradients of the constraints are linearly independent and Assumption 2 (D(x) ⊆ ilde G1(x)) holds, then v0 > 0. Theorem 13 states the converse direction: if such nonnegative multipliers exist for every ba

What carries the argument

The radial epiderivative f^r(x; d) = inf_{t>0} liminf_{u→d} [f(x+tu) − f(x)]/t is the load-bearing object: it is positively homogeneous, it makes a direction globally descending exactly when f^r(x; d) < 0, and it turns a global minimum into the absence of feasible descending directions. The paper builds a radial gradient vector by evaluating f^r along basis feasible directions of the radial cone, and uses a Gordan-type theorem to convert the absence of a common descent direction into multiplier inequalities.

Load-bearing premise

The necessity proof assumes the radial epiderivative behaves linearly enough that checking basis directions also rules out nonnegative combinations of them; only positive homogeneity, not additivity, is established.

What would settle it

Construct a radially epidifferentiable problem satisfying Assumptions 1 and 2 on a two-direction feasible cone where f^r(x; d1) ≥ 0 and f^r(x; d2) ≥ 0, but f^r(x; d1 + d2) < 0 with x + d1 + d2 feasible. If such a point is a true global minimizer yet no multipliers satisfy (21)–(23), the Gordan step in Theorem 12 fails. A concrete search over piecewise-linear functions on small discrete feasible sets would settle whether such a counterexample exists.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

Share X Bluesky LinkedIn Reddit HN

If this is right

  • KKT certificates can be written and checked for global minima without knowing which constraints are active, covering interior points and isolated feasible points.
  • The conditions apply on discrete domains where classical or Clarke derivatives are undefined, because the radial epiderivative uses an infimum rather than a limit.
  • The sufficient condition in Theorem 13 relies on D(x) ⊆ ilde G1(x), a weaker-looking analogue of Abadie's constraint qualification formulated with radial epiderivatives.
  • The multiplier inequalities are expressed directly through basis feasible directions, so they can be verified numerically by computing finitely many radial epiderivative values.
  • The active-constraint versions of the theorems provide global optimality conditions using only the constraints tight at the candidate point.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the necessity proof is strengthened by an additivity or subadditivity condition on the radial epiderivative over the feasible cone, the same multiplier form would likely yield separating-hyperplane-style duality and zero-duality-gap results for nonconvex problems.
  • The radial gradient could be used as a global descent oracle inside iterative methods: generate feasible basis directions, compute radial epiderivatives, and move along any direction with negative value, thereby escaping local minima.
  • A testable extension is to compare these certificates against global solvers on small nonconvex and discrete problems; the subtle case is when no single basis direction is descending but a nonnegative combination of basis directions is descending, since the current Gordan step does not explicitly handle that possibility.
  • The active-constraint-free formulation is a promising route toward infinite-dimensional problems where the active set is hard to identify, a direction the paper mentions only implicitly.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

2 major / 4 minor

Summary. The paper develops Fritz John (FJ) and Karush-Kuhn-Tucker (KKT) optimality conditions for finite-dimensional nonsmooth, nonconvex inequality-constrained programs by replacing classical derivatives with radial epiderivatives. It introduces a radial gradient vector taken along a basis of the cone of feasible directions, separates the cases with and without active constraints, and states necessary and sufficient conditions for global optimality (Theorems 12, 13, 16, 17). The paper also derives calculus rules for radial epiderivatives and provides illustrative examples. The central assertion is that global FJ/KKT conditions can be formulated without active constraints under Assumptions 1 and 2, and that these are, possibly for the first time, global optimality conditions for this class of problems.

Significance. The motivation is genuine: radial epiderivatives apply to lower Lipschitz functions, including functions on discrete domains, and global optimality conditions of this type would be a valuable contribution. The examples are helpful, and the calculus rules in Section 4 are a useful supplement. However, the main necessity theorem contains a load-bearing proof gap: the passage from global optimality to the inconsistency of a linear system Ax<0 uses additivity of the radial epiderivative for nonnegative combinations of directions, a property that is neither proved nor implied by positive homogeneity. Since this step is the bridge from optimality to the multiplier inequality, the central claim of the paper is not established as written. The same flaw propagates to the active-constraint version, Theorem 16.

major comments (2)
  1. [Section 3.1, Theorem 12] The step 'Then, the system Ax < 0 for x in R^k_+ is inconsistent' is not justified. Global optimality, via Theorem 11, gives only F1(x) cap G1(x) = empty, i.e., no single feasible direction d satisfies simultaneously f^r(x;d)<0 and g_i^r(x;d)<0 for all i. This rules out each basis direction d_j individually. However, the matrix A has rows evaluated at the basis directions, and a nonnegative combination x with sum_j x_j f^r(d_j)<0 and sum_j x_j g_i^r(d_j)<0 for all i does not imply that the direction d = sum_j x_j d_j belongs to F1(x) cap G1(x). Lemma 1 gives only positive homogeneity, not additivity, so f^r(x;d) is not equal to sum_j x_j f^r(x;d_j), and similarly for the constraints. Consequently, Gordan's theorem (Lemma 2) cannot be applied to A in the way the proof does. This is the central bridge from global optimality to the FJ multiplier inequality; without it, the necessity half of
  2. [Section 3, Theorem 11] The converse implication of Theorem 11 is asserted rather than proved. From F1(x) cap G1(x) = empty, the proof states that 'there cannot exist a descent direction which is feasble' and concludes global optimality. But feasibility is defined by D(x), not by G1(x). A feasible direction d may satisfy f^r(x;d)<0 while, for an active constraint, g_i^r(x;d)=0 (or even g_i^r(x;d)>0 for some positive step while x+td remains feasible), so d is not in G1(x). Such a direction is compatible with F1(x) cap G1(x) = empty but, if it exists, it contradicts global optimality. No argument is supplied to connect D(x) with G1(x) in this implication. Since Theorem 12 invokes Theorem 11, the gap affects the main result independently of the additivity issue discussed above.
minor comments (4)
  1. [Throughout] There are numerous typos: 'pozitive' for 'positive', 'inverstigate' for 'investigate', 'leed' for 'lead', 'minimums' for 'minima', 'Defnition' for 'Definition', 'Propsition' for 'Proposition'.
  2. [Example 2, Section 3.2] The computation for f^r(x;d1) is presented twice with the identical expression; the value for f^r(x;d2) is not shown. This makes verification of the example harder.
  3. [Remark 11] The claim that Assumption 2 is 'weaker than Abadie's CQ' is not substantiated. Abadie's condition is local, tangent-cone based, and uses active constraints, while Assumption 2 is a global inclusion involving the radial feasible-direction cone and radial epiderivatives. The two conditions are of different type; a genuine comparison requires a proof or an explicit counterexample rather than a heuristic discussion.
  4. [References] Reference [39] reads 'A Seies of Comprehensive Studies in Mathematics'; this should be corrected.

Circularity Check

0 steps flagged

No significant circularity: the FJ/KKT derivation is a new application of a prior derivative concept, not a restatement of its inputs. (A proof gap in Theorem 12 is a correctness issue, not circularity.)

full rationale

Circularity pass: I found no step where a claimed derivation reduces to its own input by construction. The radial-epiderivative framework (Definition 2, Lemma 1, Theorems 4–5) is imported from the first author's earlier papers [15,27]; those are general, parameter-free mathematical theorems about descent directions and global minima, not the FJ/KKT conclusions of this paper. Per hard rule 4, such citations are real evidence and do not by themselves raise the circularity score. The new FJ/KKT statements (Theorems 12, 13, 16, 17) are derived through explicit multiplier conditions and the newly introduced radial-gradient and feasible-direction sets; they are not restatements of Definition 4 or Assumption 2. The examples involve no fitted parameters called predictions and no data fitting. One caveat that is a correctness issue rather than circularity: the necessity proof of Theorem 12 passes from F1(x)∩G1(x)=∅ to inconsistency of Ax<0 for x∈R^k_+. This would require the radial epiderivative to be additive over nonnegative combinations, whereas Lemma 1 only gives positive homogeneity. That is an invalid inference, but it is not an equivalence-by-construction; the theorem's content is not identical to its assumptions. Thus it does not affect the circularity score.

Axiom & Free-Parameter Ledger

0 free parameters · 7 axioms · 0 invented entities

The central claim rests on several domain assumptions about radial epidifferentiability and lower semicontinuity, plus ad hoc assumptions (Assumptions 1, 2, 3) that function as constraint qualifications. The foundational representation formula and descent-direction criteria are taken from the authors' own prior work without independent verification. The main proof additionally assumes an additivity property of the radial epiderivative that is not established.

axioms (7)
  • domain assumption All functions f and g_i are radially epidifferentiable at x, equivalently lower Lipschitz at x (Theorem 3).
    Assumed throughout; enables use of the radial epiderivative and prior results [15,27].
  • domain assumption The representation f^r(x;d)=inf_{t>0} liminf_{u->d} (f(x+tu)-f(x))/t (Proposition 2) is taken from [15] without independent verification.
    This formula is the computational basis for all subsequent proofs, including the descent-direction criterion.
  • ad hoc to paper Assumption 1: G1(x) ⊆ G0(x) via nonempty intersection of the sets Λ_i; used to prove Theorem 11.
    Introduced specifically so that the derivative-defined descent set equals the literal feasible-descent set; no argument beyond 'natural requirement' is given.
  • ad hoc to paper Assumption 2: D(x) ⊆ ~G1(x); used to prove the KKT necessity (v0>0) and sufficiency.
    A constraint qualification in radial terms; the paper claims it is weak, but it is a substantial restriction on the feasible set.
  • ad hoc to paper Assumption 3: active-constraint analogue of Assumption 1; used in Theorems 16-17.
    Same status as Assumption 1, limited to active constraints.
  • domain assumption D(x), the cone of feasible directions, is assumed to be a k-dimensional subset of Rn admitting a basis of feasible directions.
    Needed to build the matrix A in Theorem 12; not justified for general feasible sets S.
  • standard math Gordan's theorem / separation of convex sets (Lemma 2) is used.
    Classical result; the point of failure is its application, not the theorem itself.

pith-pipeline@v1.4.0-alltime-deepseek-medium · 18464 in / 38027 out tokens · 436220 ms · 2026-08-05T12:46:12.095926+00:00 · methodology

0 comments
Cite this review

Pith. "Pith review of Radial Epiderivative Based Fritz John and KKT Conditions in Nonsmooth Nonconvex Optimization." pith.science (2026). https://pith.science/paper/KGHH74YY

@misc{pith2026250901272,
  author       = {Pith},
  title        = {Pith review of: Radial Epiderivative Based Fritz John and KKT Conditions in Nonsmooth Nonconvex Optimization},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KGHH74YY}},
  note         = {Machine review of arXiv:2509.01272}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

In this study, we examine Fritz John (FJ) and Karush-Kuhn-Tucker (KKT) type optimality conditions for a class of nonsmooth and nonconvex optimization problems with inequality constraints, where the objective and constraint functions all are assumed to be radially epidifferentiable. The concept of the radial epiderivative constitutes a distinct generalization of classical derivative notions, as it replaces the conventional limit operation with an infimum-based construction. This formulation permits the analysis of directional behavior without invoking standard neighborhood-based assumptions and constructions commonly required in generalized differentiation. This leads to one of the key advantages of the radial epiderivative which lies in its applicability even to discrete domains, where classical and generalized derivatives are often inapplicable or undefined. The other advantage of this concept is that it provides a possibility to inverstigate KKT conditions for global minimums. Consequently, this approach necessitates a reformulation of fundamental analytical tools such as gradient vectors, feasible direction sets, constraint qualifications and other concepts which are central to the derivation of optimality conditions in smooth optimization theory. The primary contribution of this paper is the development of a comprehensive theoretical framework for KKT conditions tailored to the radially epidifferentiable setting. We introduce novel definitions of the radial gradient vector, of the set of feasible directions, and examine how classical constraint qualifications may be interpreted and extended within this new framework.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

39 extracted references · 39 canonical work pages · 2 internal anchors

  1. [1]

    In: Nonlinear Programming, Abadie, J., ed

    Abadie, J.: On the Kuhn-Tucker theorem. In: Nonlinear Programming, Abadie, J., ed. John Wiley & Sons, New York, 21–36 (1967) 24

    work page 1967

  2. [2]

    Achtziger, W., Kanzow, C.: Mathematical programs with vanishing constraints: Optimality conditions and constraint qualifications. Math. Program., Ser. A 114, 69–99 (2008)

    work page 2008

  3. [3]

    Andreani, R., Echague, C.E., Schuverdt, M.L.: Constant-rank condition and second-order constraint qualification. J. Optim. Theory Appl. 146, 255–266 (2010)

    work page 2010

  4. [4]

    Andreani, R., Haeser, G., Schuverdt, M.L., Silva, P.J.S.: Two new constraint qualifications and applications. SIAM J. Optim. 22, 1109–1135 (2012)

    work page 2012

  5. [5]

    Andreani, R., Haeser, G., Schuverdt, M.L., Silva, P.J.S.: A relaxed constant posi- tive linear dependence constraint qualification and applications. Math. Program., Ser. A 135, 255–273 (2012)

    work page 2012

  6. [6]

    Andreani, R., Martinez, J.M., Svaiter, B.F.: A new sequential optimality condi- tion for constrained optimization and algorithmic consequences. SIAM J. Optim. 20, 3533–3554 (2010)

    work page 2010

  7. [7]

    Andreani, R., Martinez, J.M., Ramos, A., Silva, P.J.S.: A cone-continuity con- straint qualification and algorithmic consequences. SIAM J. Optim. 26, 96–110 (2016)

    work page 2016

  8. [8]

    Naval Res

    Arrow, S., Hurwicz, L., Uzawa, H.: Constraint qualifications in maximization problems. Naval Res. Logist. Quart. 8, 175–191 (1961)

    work page 1961

  9. [9]

    John Wiley & Sons, New Jersey (2006)

    Bazaraa, M.S., Sherali, H.D., Shetty, C.M.: Nonlinear Programming: Theory and Algorithms, 3rd edn. John Wiley & Sons, New Jersey (2006)

    work page 2006

  10. [10]

    Athena Scientific, Belmont, MA (2003)

    Bertsekas, D.P.: Convex Analysis and Optimization. Athena Scientific, Belmont, MA (2003)

    work page 2003

  11. [11]

    Springer, New York (2006)

    Borwein, J., Lewis, A.S.: Convex Analysis and Nonlinear Optimization: Theory and Examples. Springer, New York (2006)

    work page 2006

  12. [12]

    Second-Order Necessary Conditions, Constraint Qualifications and Exact Penalty for Mathematical Programs with Switching Constraints

    Chen, J., Liu, L., L., Y., Zhao, K.: Second-Order Necessary Conditions, Con- straint Qualifications and Exact Penalty for Mathematical Programs with Switching Constraints. arXiv:2407.17285v2 [math.OC] (2024)

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  13. [13]

    CMS of Monographs and Advanced Texts

    Clarke, F.H.: Optimization and Nonsmooth Analysis. CMS of Monographs and Advanced Texts. Wiley, New York (1983)

    work page 1983

  14. [14]

    Optimization 70(7), 1513–1553 (2021)

    Dinc Yalcin, G., Kasimbeyli, R.: Weak subgradient method for solving nonsmooth nonconvex optimization problems. Optimization 70(7), 1513–1553 (2021)

    work page 2021

  15. [15]

    Dinc Yalcin, G., Kasimbeyli, R.: Generalized derivatives and optimality condi- tions in nonconvex optimization. Bull. Malays. Math. Sci. Soc. 47(3) (2024) 25

    work page 2024

  16. [16]

    Flores-Bazan, F.: Radial epiderivatives and asymptotic functions in nonconvex vector optimization. SIAM J. Optim. 14, 284–305 (2003)

    work page 2003

  17. [17]

    Flores-Bazan, F., Mastroeni, G.: Characterizing FJ and KKT conditions in non- convex mathematical programming with applications. SIAM J. Optim. 25(1), 647–676 (2015)

    work page 2015

  18. [18]

    Flores-Bazan, F., Flores-Baz´ an, F., Vera, C.: Maximizing and minimizing qua- siconvex functions: Related properties, existence and optimality conditions via radial epiderivates. J. Glob. Optim. 63, 99–123 (2015)

    work page 2015

  19. [19]

    University of Pavia, Department of Economics and Management Working Paper Series, No 160 (2018)

    Giorgi, G.: A guided tour in constraint qualifications for nonlinear program- ming under differentiability assumptions. University of Pavia, Department of Economics and Management Working Paper Series, No 160 (2018)

    work page 2018

  20. [20]

    Giorgi, G.: Notes on constraint qualifications for second-order optimality condi- tions. J. Math. Res. 11(5), 16–32 (2019)

    work page 2019

  21. [21]

    Gould, F.J., Tolle, J.W.: A necessary and sufficient qualification for constrained optimization. SIAM J. Appl. Math. 20, 164–172 (1971)

    work page 1971

  22. [22]

    Gould, F.J., Tolle, J.W.: Geometry of optimality conditions and constraint qualifications. Math. Program. 2, 1–18 (1972)

    work page 1972

  23. [23]

    Guignard, M.: Generalized Kuhn-Tucker conditions for mathematical program- ming problems in a Banach space. SIAM J. Control 7, 232–241 (1969)

    work page 1969

  24. [24]

    In: Math

    Janin, R.: Directional derivative of the marginal function in nonlinear program- ming. In: Math. Program. Stud. vol. 21, pp. 110–126. North-Holland, Amsterdam (1984)

    work page 1984

  25. [25]

    Studies and Essays, Courant Anniversary Volume, Interscience, New York (1948)

    John, F.: Extremum Problems with Inequalities as Subsidiary Conditions. Studies and Essays, Courant Anniversary Volume, Interscience, New York (1948)

    work page 1948

  26. [26]

    Karush, W.: Minima of functions of several variables with inequalities and side conditions. M.S. thesis, Department of Mathematics, University of Chicago. (1939)

    work page 1939

  27. [27]

    Optimization 58, 521–534 (2009)

    Kasimbeyli, R.: Radial epiderivatives and set-valued optimization. Optimization 58, 521–534 (2009)

    work page 2009

  28. [28]

    Kasimbeyli, R.: A nonlinear cone separation theorem and scalarization in nonconvex vector optimization. SIAM J. Optim. 20, 1591–1619 (2010)

    work page 2010

  29. [29]

    Radial Epiderivative Based Line Search Methods in Nonconvex and Nonsmooth Box-Constrained Optimization

    Kasimbeyli, R., Dinc Yalcin, G., Yildiz, G.B., Ozcetin, E.: Radial epideriva- tive based line search methods in nonconvex and nonsmooth box-constrained optimization (2025). http://arxiv.org/abs/2504.05090 26

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  30. [30]

    Kasimbeyli, R., Mammadov, M.: On weak subdifferentials, directional derivatives and radial epiderivatives for nonconvex functions. SIAM J. Optim.20(2), 841–855 (2009)

    work page 2009

  31. [31]

    Nonlinear Anal., Theory Methods Appl

    Kasimbeyli, R., Mammadov, M.: Optimality conditions in nonconvex optimiza- tion via weak subdifferentials. Nonlinear Anal., Theory Methods Appl. 74(7), 2534–2547 (2011)

    work page 2011

  32. [32]

    Koushki, J., Soleimani-damaneh, M.: Characterization of generalized FJ and KKT conditions in nonsmooth nonconvex optimization. J. Glob. Optim. 76, 407–431 (2020)

    work page 2020

  33. [33]

    W., Tucker, A

    Kuhn, H. W., Tucker, A. W.: Nonlinear programming. In: Proceedings of the Sec- ond Berkeley Symposium on Mathematical Statistics and Probability, Neyman, J., ed. University of California Press, Berkeley, 481–492 (1950)

    work page 1950

  34. [34]

    Lara, F.: Optimality conditions for nonconvex nonsmooth optimization via global derivatives. J. Optim. Theory Appl. 185, 134–150 (2020)

    work page 2020

  35. [35]

    Mangasarian, O.L., Fromovitz, S.: The Fritz John necessary optimality conditions in the presence of equality and inequality constraints. J. Math. Anal. Appl. 17, 37–47 (1967)

    work page 1967

  36. [36]

    Optimization 73, 1993–2005 (2024)

    Ribeiro, A., Sachine, M.: On strong second-order necessary optimality conditions under relaxed constant rank constraint qualification. Optimization 73, 1993–2005 (2024)

    work page 1993

  37. [37]

    Rockafellar, R.T.: First and second-order epi-differentiability in nonlinear pro- gramming. Trans. Am. Math. Soc. 307, 75–108 (1988)

    work page 1988

  38. [38]

    SIAM Rev

    Rockafellar, R.T.: Lagrange multipliers and optimality. SIAM Rev. 35(2), 183– 238 (1993)

    work page 1993

  39. [39]

    A Seies of Comprehensive Studies in Mathematics Springer, Berlin (2009) 27

    Rockafellar, R.T., Wets, R.J.-B.: Variational Analysis, 3rd edn. A Seies of Comprehensive Studies in Mathematics Springer, Berlin (2009) 27

    work page 2009