pith. sign in

arxiv: 2508.12850 · v2 · pith:PTUOVPMHnew · submitted 2025-08-18 · 🧮 math.OC

On Constraint Qualifications for MPECs with Applications to Bilevel Hyperparameter Optimization for Machine Learning

Jiani Li , Qingna Li , Alain Zemkoho This is my paper

Pith reviewed 2026-05-18 22:53 UTC · model grok-4.3

classification 🧮 math.OC
keywords MPECconstraint qualificationsMPEC-LICQbilevel optimizationhyperparameter optimizationsupport vector classificationL1-loss SVM
0
0 comments X

The pith

The MPEC-LICQ is completely characterized for the equilibrium-constrained program that arises when reformulating bilevel hyperparameter optimization of L1-loss support vector classification.

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

The paper studies constraint qualifications needed to analyze stationarity in mathematical programs that contain equilibrium constraints. It first reviews several standard qualifications and shows how they relate to one another. It then turns to a concrete MPEC obtained by converting a bilevel hyperparameter tuning problem for L1-loss support vector machines into a single-level equilibrium-constrained program. For this specific MPEC the authors derive exact conditions under which the MPEC linear independence constraint qualification holds or fails. A sympathetic reader would care because these qualifications determine whether first-order optimality conditions are reliable and whether numerical solvers are guaranteed to converge to meaningful solutions when tuning machine-learning models.

Core claim

In the MPEC that encodes bilevel hyperparameter optimization for L1-loss support vector classification, the MPEC-LICQ holds precisely when the gradients of the active upper-level constraints, the active lower-level constraints, and the complementarity multipliers satisfy a certain linear-independence relation that accounts for the complementarity structure; the paper supplies the complete list of index sets and algebraic conditions that make this relation true or false.

What carries the argument

The MPEC linear independence constraint qualification (MPEC-LICQ), which requires that the gradients associated with active upper-level inequalities, active lower-level inequalities, and the complementarity pairs remain linearly independent after the equilibrium constraints are incorporated.

If this is right

  • Whenever the derived conditions for MPEC-LICQ are satisfied, the standard KKT system correctly describes stationary points of the MPEC.
  • Convergence proofs for algorithms that solve the bilevel hyperparameter problem can invoke MPEC-LICQ on the instances that meet the characterization.
  • Cases in which the characterization shows MPEC-LICQ fails identify bilevel SVM tuning problems where first-order necessary conditions may not be sufficient.
  • The same index-set approach can be used to check MPEC-LICQ on other L1-regularized classification or regression models that admit an equivalent MPEC form.

Where Pith is reading between the lines

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

  • The characterization supplies a practical test that practitioners could run on a given data set before launching a hyperparameter search, flagging instances where standard stationarity-based solvers may behave unexpectedly.
  • The same style of index-set analysis could be applied to bilevel problems that arise in other machine-learning pipelines, such as neural-network architecture search or robust training, once they are cast as MPECs.
  • If the characterization extends to related qualifications such as MPEC-MFCQ, one could obtain a full hierarchy of constraint qualifications tailored to this class of bilevel SVM problems.

Load-bearing premise

The bilevel hyperparameter optimization problem for L1-loss support vector classification can be rewritten as the exact MPEC studied in the paper without any reformulation step that would change which gradients are active or how the complementarity conditions interact.

What would settle it

A concrete numerical instance of the L1-loss SVM bilevel problem in which the active gradients violate linear independence while the paper's index-set conditions predict that MPEC-LICQ should hold, or the converse.

Figures

Figures reproduced from arXiv: 2508.12850 by Alain Zemkoho, Jiani Li, Qingna Li.

Figure 1
Figure 1. Figure 1: : Summary of the fulfillment of MPEC-LICQ. [PITH_FULL_IMAGE:figures/full_fig_p017_1.png] view at source ↗
read the original abstract

Constraint qualifications for a Mathematical Program with Equilibrium Constraints (MPEC) are essential for analyzing stationarity properties and establishing convergence results. In this paper, we explore several classical MPEC constraint qualifications and clarify the relationships among them. We subsequently examine the behavior of these constraint qualifications in the context of a specific MPEC derived from bilevel hyperparameter optimization (BHO) for L1-loss support vector classification. In particular, for such an MPEC, we provide a complete characterization of the well-known MPEC linear independence constraint qualification (MPEC-LICQ), therefore, establishing conditions under which it holds or fails for our BHO for support vector machines.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper reviews classical constraint qualifications for Mathematical Programs with Equilibrium Constraints (MPECs) and their interrelationships. It then derives an MPEC from the bilevel hyperparameter optimization problem for L1-loss support vector classification and provides an explicit characterization of the conditions under which the MPEC linear independence constraint qualification (MPEC-LICQ) holds or fails for this instance.

Significance. If the MPEC reformulation is equivalent and the characterization is complete, the result supplies concrete, checkable conditions for MPEC-LICQ in a practically relevant bilevel machine-learning setting. This could support subsequent stationarity analysis and convergence proofs for algorithms applied to such hyperparameter optimization problems. The explicit (rather than abstract) treatment of the L1-SVM case is a positive feature.

major comments (2)
  1. [§3] §3 (MPEC reformulation): The derivation replaces the lower-level optimality condition with its KKT system to obtain the complementarity constraints. For the nonsmooth L1 (hinge) loss, the paper must confirm that the subdifferential is correctly incorporated and that any failure of lower-level constraint qualifications does not introduce additional degenerate multiplier cases outside the analyzed MPEC. This equivalence is load-bearing for transferring the MPEC-LICQ characterization back to the original bilevel problem.
  2. [§4] §4 (MPEC-LICQ characterization): The stated conditions for MPEC-LICQ to hold or fail rely on linear independence of certain gradients involving the complementarity constraints. Provide an explicit verification (e.g., via a low-dimensional example or counter-example) that these conditions are both necessary and sufficient for the specific structure arising from the L1-SVM lower level; otherwise the completeness claim is not fully substantiated.
minor comments (2)
  1. Notation for the complementarity constraints and the associated index sets (active, inactive, degenerate) should be introduced once and used consistently throughout the characterization.
  2. A brief comparison table of the classical MPEC CQs (MPEC-LICQ, MPEC-MFCQ, etc.) and their relationships would improve readability before the application section.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading of our manuscript and the constructive comments. We address each major comment in turn below, indicating the revisions we plan to incorporate.

read point-by-point responses

  1. Referee: [§3] §3 (MPEC reformulation): The derivation replaces the lower-level optimality condition with its KKT system to obtain the complementarity constraints. For the nonsmooth L1 (hinge) loss, the paper must confirm that the subdifferential is correctly incorporated and that any failure of lower-level constraint qualifications does not introduce additional degenerate multiplier cases outside the analyzed MPEC. This equivalence is load-bearing for transferring the MPEC-LICQ characterization back to the original bilevel problem.

    Authors: We agree that the equivalence between the original bilevel problem and the MPEC reformulation requires careful justification. In Section 3 the lower-level optimality condition for the L1-loss SVM is replaced by its KKT system, where the subdifferential of the nonsmooth hinge loss is incorporated via the standard convex subdifferential, yielding the complementarity constraints in the usual way. We work under the standing assumption that the lower-level problem satisfies a constraint qualification (e.g., LICQ) so that the KKT conditions are necessary and no extraneous degenerate multiplier cases arise. In the revised manuscript we will add an explicit remark after the derivation that states this assumption and briefly discusses its role in preserving equivalence. revision: yes

  2. Referee: [§4] §4 (MPEC-LICQ characterization): The stated conditions for MPEC-LICQ to hold or fail rely on linear independence of certain gradients involving the complementarity constraints. Provide an explicit verification (e.g., via a low-dimensional example or counter-example) that these conditions are both necessary and sufficient for the specific structure arising from the L1-SVM lower level; otherwise the completeness claim is not fully substantiated.

    Authors: The characterization in Section 4 is obtained by substituting the explicit gradients of the complementarity constraints that arise from the L1-SVM lower level into the definition of MPEC-LICQ and then enumerating the possible active-set combinations. Necessity and sufficiency therefore follow directly from the general MPEC-LICQ definition applied to this concrete instance. To make the argument more transparent we will insert a short low-dimensional numerical example (with explicit gradient matrices) in the revised version that verifies both a case in which the stated conditions hold and a case in which they fail. revision: yes

Circularity Check

0 steps flagged

No circularity: MPEC-LICQ characterization derived from classical definitions applied to reformulated BHO instance

full rationale

The paper first reviews standard MPEC constraint qualifications from the literature and then applies them to the specific MPEC obtained by replacing the lower-level optimality condition of the L1-loss SVM bilevel problem with its KKT system. The complete characterization of MPEC-LICQ is obtained by direct inspection of the resulting complementarity constraints and their gradients; no parameter is fitted inside the paper and then relabeled as a prediction, no self-citation supplies the uniqueness or validity of the central claim, and the reformulation step is presented as a standard equivalence rather than a definitional identity that would make the CQ result tautological. The derivation therefore remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Abstract-only review; no explicit free parameters, invented entities, or ad-hoc axioms are mentioned. The work relies on standard background results from MPEC theory.

axioms (1)
  • domain assumption Classical MPEC constraint qualifications and their relationships are well-defined and applicable to the bilevel reformulation.
    The paper begins by exploring several classical MPEC constraint qualifications before applying them to the BHO instance.

pith-pipeline@v0.9.0 · 5640 in / 1248 out tokens · 47335 ms · 2026-05-18T22:53:46.185744+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Bilevel learning

    math.OC 2026-05 unverdicted novelty 2.0

    Bilevel learning methods rely on implicit differentiation but are restricted by assumptions of unique lower-level solutions and struggle with constraints, and connections to broader bilevel optimization literature may...

Reference graph

Works this paper leans on

65 extracted references · 65 canonical work pages · cited by 1 Pith paper · 2 internal anchors

  1. [1]

    Practical bilevel optimization: algorithms and applications

    Bard JF. Practical bilevel optimization: algorithms and applications. Springer Science & Business Media, 2013

    work page 2013

  2. [2]

    Bilevel optimization: theory, algorithms and applications

    Dempe S. Bilevel optimization: theory, algorithms and applications. Freiberg, Germany: TU Bergakademie Freiberg, Fakult¨ at f¨ ur Mathematik und Informatik; 2018

    work page 2018

  3. [3]

    Bilevel optimization

    Dempe S, Zemkoho A. Bilevel optimization. In: Springer Optimization and Its Applications. Vol. 161. Springer; 2020. 17

    work page 2020

  4. [4]

    Bilevel programming problems

    Dempe S, Kalashnikov V, P´ erez-Vald´ es GA, et al. Bilevel programming problems. Energy Systems. Springer, Berlin, 2015; 10(978-3): 53-56

    work page 2015

  5. [5]

    A review on bilevel optimization: From classical to evolutionary approaches and applications

    Sinha A, Malo P, Deb K. A review on bilevel optimization: From classical to evolutionary approaches and applications. IEEE transactions on evolutionary computation. 2017; 22(2): 276-295

    work page 2017

  6. [6]

    Applications of bilevel optimization in energy and electricity markets

    Wogrin S, Pineda S, Tejada-Arango DA. Applications of bilevel optimization in energy and electricity markets. Bilevel Optimization: Advances and Next Challenges. 2020; 139-168

    work page 2020

  7. [7]

    Self-Tuning Networks: Bilevel Optimization of Hyperparameters using Structured Best-Response Functions

    MacKay M, Vicol P, Lorraine J, et al. Self-tuning networks: Bilevel optimization of hyper- parameters using structured best-response functions. 2019; arXiv preprint arXiv:1903.03088

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  8. [8]

    On ℓp-hyperparameter learning via bilevel nons- mooth optimization

    Okuno T, Takeda A, Kawana A, et al. On ℓp-hyperparameter learning via bilevel nons- mooth optimization. Journal of Machine Learning Research. 2021; 22(245): 1-47

    work page 2021

  9. [9]

    DARTS: Differentiable Architecture Search

    Liu H, Simonyan K, Yang Y. Darts: Differentiable architecture search. 2018; arXiv preprint arXiv:1806.09055

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  10. [10]

    Towards fast adaptation of neural architectures with meta learning

    Lian D, Zheng Y, Xu Y, et al. Towards fast adaptation of neural architectures with meta learning. International Conference on Learning Representations. 2020

    work page 2020

  11. [11]

    Bi-level actor-critic for multi-agent coordination

    Zhang H, Chen W, Huang Z, et al. Bi-level actor-critic for multi-agent coordination. Proceedings of the AAAI Conference on Artificial Intelligence. 2020; 34(05): 7325-7332

    work page 2020

  12. [12]

    Provably global convergence of actor-critic: A case for linear quadratic regulator with ergodic cost

    Yang Z, Chen Y, Hong M, et al. Provably global convergence of actor-critic: A case for linear quadratic regulator with ergodic cost. Advances in neural information processing systems. 2019; 32

    work page 2019

  13. [13]

    Bi-level probabilistic feature learning for deformable image registration

    Liu R, Li Z, Zhang Y, et al. Bi-level probabilistic feature learning for deformable image registration. Proceedings of the Twenty-Ninth International Conference on International Joint Conferences on Artificial Intelligence. 2021; 723-730

    work page 2021

  14. [14]

    Bilevel fast scene adaptation for low-light image enhancement

    Ma L, Jin D, An N, et al. Bilevel fast scene adaptation for low-light image enhancement. International Journal of Computer Vision. 2023; 1-19

    work page 2023

  15. [15]

    On the solution of convex bilevel optimization problems

    Dempe S, Franke S. On the solution of convex bilevel optimization problems. Computa- tional Optimization and Applications. 2016; 63: 685-703

    work page 2016

  16. [16]

    Mathematical programs with equilibrium constraints

    Luo ZQ, Pang JS, Ralph D. Mathematical programs with equilibrium constraints. Cam- bridge University Press. 1996

    work page 1996

  17. [17]

    The bilevel programming problem: reformulations, constraint qual- ifications and optimality conditions

    Dempe S, Zemkoho A. The bilevel programming problem: reformulations, constraint qual- ifications and optimality conditions. Mathematical Programming. 2013; 138: 447-473

    work page 2013

  18. [18]

    On the Karush–Kuhn–Tucker reformulation of the bilevel optimiza- tion problem

    Dempe S, Zemkoho A. On the Karush–Kuhn–Tucker reformulation of the bilevel optimiza- tion problem. Nonlinear Analysis: Theory, Methods & Applications. 2012; 75(3): 1202-1218

    work page 2012

  19. [19]

    Exact penalization and necessary optimality conditions for generalized bilevel programming problems

    Ye JJ, Zhu DL, Zhu QJ. Exact penalization and necessary optimality conditions for generalized bilevel programming problems. SIAM Journal on optimization. 1997; 7(2): 481- 507

    work page 1997

  20. [20]

    Abadie-type constraint qualification for mathematical programs with equilibrium constraints

    Flegel ML, Kanzow C. Abadie-type constraint qualification for mathematical programs with equilibrium constraints. Journal of Optimization Theory and Applications. 2005; 124(3): 595-614

    work page 2005

  21. [21]

    Mathematical programs with equilibrium constraints: a sequential optimality condition, new constraint qualifications and algorithmic consequences

    Ramos A. Mathematical programs with equilibrium constraints: a sequential optimality condition, new constraint qualifications and algorithmic consequences. Optimization Meth- ods and Software. 2021; 36(1): 45-81

    work page 2021

  22. [22]

    Complementarity constraint qualifications and simplified B-stationarity conditions for mathematical programs with equilibrium constraints

    Pang JS and Fukushima M. Complementarity constraint qualifications and simplified B-stationarity conditions for mathematical programs with equilibrium constraints. Compu- tational Optimization and Applications. 1999; 13: 111-136

    work page 1999

  23. [23]

    Scheel and S

    H. Scheel and S. Scholtes, Mathematical programs with complementarity constraints: Stationarity, optimality, and sensitivity. Mathematics of Operations Research, 2000, 25(1): 1-22

    work page 2000

  24. [24]

    Some properties of regularization and penalization schemes for MPECs

    Ralph* D, Wright SJ. Some properties of regularization and penalization schemes for MPECs. Optimization Methods and Software. 2004; 19(5): 527-556

    work page 2004

  25. [25]

    Necessary and sufficient optimality conditions for mathematical programs with equilibrium constraints

    Jane JY. Necessary and sufficient optimality conditions for mathematical programs with equilibrium constraints. Journal of Mathematical Analysis and Applications. 2005; 307(1): 350-369

    work page 2005

  26. [26]

    A direct proof for M-stationarity under MPEC-GCQ for mathe- 18 matical programs with equilibrium constraints

    Flegel ML, Kanzow C. A direct proof for M-stationarity under MPEC-GCQ for mathe- 18 matical programs with equilibrium constraints. Optimization with Multivalued Mappings: Theory, Applications, and Algorithms. Boston, MA: Springer US, 2006; 111-122

    work page 2006

  27. [27]

    Calmness of constraint systems with applications

    Henrion R, Outrata JV. Calmness of constraint systems with applications. Mathematical Programming. 2005; 104(2): 437-464

    work page 2005

  28. [28]

    On the Guignard constraint qualification for mathematical pro- grams with equilibrium constraints

    Flegel ML, Kanzow C. On the Guignard constraint qualification for mathematical pro- grams with equilibrium constraints. Optimization. 2005; 54(6): 517-534

    work page 2005

  29. [29]

    Optimality conditions for a class of mathematical programs with equilibrium constraints

    Outrata JV. Optimality conditions for a class of mathematical programs with equilibrium constraints. Mathematics of operations research. 1999; 24(3): 627-644

    work page 1999

  30. [30]

    A generalized mathematical program with equilibrium constraints

    Outrata JV. A generalized mathematical program with equilibrium constraints. SIAM Journal on Control and Optimization. 2000; 38(5): 1623-1638

    work page 2000

  31. [31]

    On M-stationary points for mathematical programs with equilib- rium constraints

    Flegel ML, Kanzow C. On M-stationary points for mathematical programs with equilib- rium constraints. Journal of Mathematical Analysis and Applications. 2005; 310(1): 286-302

    work page 2005

  32. [32]

    Optimality conditions for disjunctive programs with application to mathematical programs with equilibrium constraints

    Flegel ML, Kanzow C, Outrata JV. Optimality conditions for disjunctive programs with application to mathematical programs with equilibrium constraints. Set-Valued Analysis. 2007; 15(2): 139-162

    work page 2007

  33. [33]

    Convergence properties of a regularization scheme for mathematical programs with complementarity constraints

    Scholtes S. Convergence properties of a regularization scheme for mathematical programs with complementarity constraints. SIAM Journal on Optimization. 2001; 11(4): 918-936

    work page 2001

  34. [34]

    Workshop on Ill-Posed Variational Prob- lems and Regulation Techniques

    Fukushima M, Pang J S. Convergence of a smoothing continuation method for math- ematical programs with complementarity constraints. Ill-posed Variational Problems and Regularization Techniques: Proceedings of the “Workshop on Ill-Posed Variational Prob- lems and Regulation Techniques” held at the University of Trier, September 3–5, 1998. Springer Berlin H...

    work page 1998

  35. [35]

    Model selection via bilevel optimization

    Bennett KP, Hu J, Ji X, et al. Model selection via bilevel optimization. The 2006 IEEE International Joint Conference on Neural Network Proceedings. IEEE, 2006; 1922-1929

    work page 2006

  36. [36]

    Bilevel optimization and machine learning

    Bennett KP, Kunapuli G, Hu J, et al. Bilevel optimization and machine learning. IEEE world congress on computational intelligence. Berlin, Heidelberg: Springer Berlin Heidelberg, 2008; 25-47

    work page 2008

  37. [37]

    Classification model selection via bilevel program- ming

    Kunapuli G, Bennett KP, Hu J, et al. Classification model selection via bilevel program- ming. Optimization Methods & Software. 2008; 23(4): 475-489

    work page 2008

  38. [38]

    Bilevel model selection for support vector machines

    Kunapuli G, Bennett KP, Hu J, et al. Bilevel model selection for support vector machines. Data Mining and Mathematical Programming. 2008; 45: 129–158

    work page 2008

  39. [39]

    A unified framework and a case study for hyperparameter selection in machine learning via bilevel optimization

    Li Z, Qian Y, Li Q. A unified framework and a case study for hyperparameter selection in machine learning via bilevel optimization. 5th International Conference on Data Science and Information Technology (DSIT), 2022, pp. 1-8

    work page 2022

  40. [40]

    Bilevel hyperparameter optimization for support vector classi- fication: theoretical analysis and a solution method

    Li Q, Li Z, Zemkoho A. Bilevel hyperparameter optimization for support vector classi- fication: theoretical analysis and a solution method. Mathematical Methods of Operations Research. 2022; 96(3): 315-350

    work page 2022

  41. [41]

    Global relaxation-based LP-Newton method for multiple hy- perparameter selection in support vector classification with feature selection

    Qian Y, Li Q, Zemkoho A. Global relaxation-based LP-Newton method for multiple hy- perparameter selection in support vector classification with feature selection. arXiv preprint arXiv:2312.10848, 2023

    work page arXiv 2023

  42. [42]

    Bilevel hyperparameter optimization for nonlinear support vector machines

    Coniglio S, Dunn S, Li Q, et al. Bilevel hyperparameter optimization for nonlinear support vector machines. Optimization Online, 2023, pp. 1-78

    work page 2023

  43. [43]

    Mathematical programs with complementarity con- straints and application to hyperparameter tuning for nonlinear support vector machines, manuscript, 2025

    Ward S, Zemkoho A, Ahipasaoglu S. Mathematical programs with complementarity con- straints and application to hyperparameter tuning for nonlinear support vector machines, manuscript, 2025

    work page 2025

  44. [44]

    A fast smoothing Newton method for bilevel hyperparameter optimization for SVC with Logistic loss

    Wang Y, Li Q. A fast smoothing Newton method for bilevel hyperparameter optimization for SVC with Logistic loss. Optimization, 2024, DOI: 10.1080/02331934.2024.2394612

    work page doi:10.1080/02331934.2024.2394612 2024

  45. [45]

    Theoretical and numerical comparison of relaxation methods for mathematical programs with complementarity constraints

    Hoheisel T, Kanzow C, Schwartz A. Theoretical and numerical comparison of relaxation methods for mathematical programs with complementarity constraints. Mathematical Pro- gramming. 2013; 137: 257-288

    work page 2013

  46. [46]

    Multiplier rules under mixed assumptions of differentiability and Lipschitz conti- nuity

    Ye JJ. Multiplier rules under mixed assumptions of differentiability and Lipschitz conti- nuity. SIAM Journal on Control and Optimization, 2000, 39(5): 1441-1460

    work page 2000

  47. [47]

    Mathematical programs with equilibrium constraints: Enhanced Fritz John conditions, new constraint qualifications, and improved exact penalty results

    Kanzow C, Schwartz A. Mathematical programs with equilibrium constraints: Enhanced Fritz John conditions, new constraint qualifications, and improved exact penalty results. 19 SIAM Journal on Optimization. 2010; 20(5): 2730-2753

    work page 2010

  48. [48]

    Lipschitzian stability of constraint systems and generalized equations

    Mordukhovich B. Lipschitzian stability of constraint systems and generalized equations. Nonlinear Analysis: Theory, Methods & Applications. 1994; 22(2): 173-206

    work page 1994

  49. [49]

    Calmness as a constraint qualification for M-stationarity conditions in MPECs

    Henrion R. Calmness as a constraint qualification for M-stationarity conditions in MPECs. Generalized Nash Equilibrium Problems, Bilevel Programming and MPEC. Singapore: Springer Singapore, 2018: 21-41

    work page 2018

  50. [50]

    Misclassification minimization

    Mangasarian OL. Misclassification minimization. Journal of Global Optimization. 1994; 5(4): 309-323. Disclosure statement: There are no relevant financial or non-financial competing interests to report. Appendix A. Proof of Theorem 3.2 Recall IH := 4 ∪ k=1 IHk, IG := 4 ∪ k=1 IGk and IGH := 4 ∪ k=1 IGHk in [40], where IH1 := n i ∈ Qu | ζi = 0, (AB⊤α + z)i ...

    work page 1994

  51. [51]

    0(Λ+ 3 ,Λu) i , Γ5 e2 := h 0(Λc 3,Λ1) 0(Λc 3,Λ2) 0(Λc 3,Λ+ 3 ) I(Λc 3,Λc

    work page

  52. [52]

    0(Λc 3,Λu) i , Γ5 e3 := h 0(Λu,Λ1) 0(Λu,Λ2) 0(Λu,Λ+ 3 ) 0(Λu,Λc

    work page

  53. [53]

    I(Λu,Λu) i , Γ5 l1 := h I(Λ1,Λ1) 0(Λ1,Λ2) 0(Λ1,Λ+ 3 ) 0(Λ1,Λc

    work page

  54. [54]

    0(Λ1,Λu) i , Γ5 l2 := h 0(Λ2,Λ1) I(Λ2,Λ2) 0(Λ2,Λ+ 3 ) 0(Λ2,Λc

    work page

  55. [55]

    0(Λ2,Λu) i , Γ5 l3 := h 0(Λ+ 3 ,Λ1) 0(Λ+ 3 ,Λ2) I(Λ+ 3 ,Λ+ 3 ) 0(Λ+ 3 ,Λc

    work page

  56. [56]

    Proposition B.2

    0(Λ+ 3 ,Λu) i . Proposition B.2. By the definition of Lq, q = 1, ...,5, there are 2n − 1 columns in Γ. The number of rows is 2n − 2 + |Λ1| + |Λc 3| by the Γ in (B1). Related Lemmas and Proofs Define Γsub1 =   0(Λ+ 3 ,L1) (BB ⊤)(Λ+ 3 ,·) Γ5 e1 0(Λ2,L1) Γ4 h 0(Λ2,L5) 1(Λu,L1) Γ4 i 0(Λu,L5) 0(Λ+ 3 ,L1) 0(Λ+ 3 ,L4) Γ5 l3   . (B2) 23 Lemma B.3. Let v∗ ...

    work page

  57. [57]

    Meanwhile, ( BB ⊤)(Λ+ 3 ,Λu)1|Λu| is replaced by (BB ⊤)(Λ+ 3 ,Λc 3∪Λu)1|Λc 3∪Λu|

    replaced by zero. Meanwhile, ( BB ⊤)(Λ+ 3 ,Λu)1|Λu| is replaced by (BB ⊤)(Λ+ 3 ,Λc 3∪Λu)1|Λc 3∪Λu|. We carry on to conduct row transformation B-(2, 3) to obtain the second row block with ( BB ⊤)(Λc 3,Λ2) replaced by zero. Conduct B+(2, 4) to obtain the sec- ond row block with ( BB ⊤)(Λc 3,Λu) replaced by zero. Meanwhile, 0 (Λc 3,L1) is re- placed by ( BB ...

    work page

  58. [58]

    Meanwhile, ( BB ⊤)(Λc 3,Λu)1|Λu| is replaced by 28 (BB ⊤)(Λc 3,Λc 3∪Λu)1|Λc 3∪Λu|

    replaced by zero. Meanwhile, ( BB ⊤)(Λc 3,Λu)1|Λu| is replaced by 28 (BB ⊤)(Λc 3,Λc 3∪Λu)1|Λc 3∪Λu|. We reach the following matrix bΓ =   a1 4 Q1 4 0(Λ+ 3 ,L5) a2 4 Q2 4 0(Λc 3,L5) 0(Λ2,L1) Γ4 h 0(Λ2,L5) 1(Λu,L1) Γ4 i 0(Λu,L5) 1(Λc 3,L1) Γ4 j 0(Λc 3,L5) 0(Λc 3,L1) 0(Λc 3,L4) Γ5 k 0(Λ+ 3 ,L1) 0(Λ+ 3 ,L4) Γ5 l3   , where Q1 4 = [0(Λ+ 3 ,...

    work page

  59. [59]

    Assume there exists nonzero column vector ρ defined by ρ = ρ⊤ 1 , ρ⊤ 2 ,

    from left and add them to the first (forth) row block, we obtain the following matrix Γ =   a2 4 Q1 4 0(Λ+ 3 ,L5) a3 4 0(Λc 3,L4) 0(Λc 3,L5) 0(Λ2,L1) Γ4 h 0(Λ2,L5) 1(Λu,L1) Γ4 i 0(Λu,L5) 1(Λc 3,L1) Γ4 j 0(Λc 3,L5) 0(Λc 3,L1) 0(Λc 3,L4) Γ5 k 0(Λ+ 3 ,L1) 0(Λ+ 3 ,L4) Γ5 l3   , where a3 4 = a2 4 − A4a1 4. Assume there exists nonzero column...

    work page

  60. [60]

    0(Λ+ 3 ,Λu) i + ρ⊤ e2 h 0(Λc 3,Λ2) 0(Λc 3,Λ+ 3 ) I(Λc 3,Λc

    work page

  61. [61]

    0(Λc 3,Λu) i + ρ⊤ e3 h 0(Λu,Λ2) 0(Λu,Λ+ 3 ) 0(Λu,Λc

    work page

  62. [62]

    I(Λu,Λu) i + ρ⊤ k h 0(Λc 3,Λ2) 0(Λc 3,Λ+ 3 ) I(Λc 3,Λc

    work page

  63. [63]

    0(Λc 3,Λu) i + ρ⊤ l2 h I(Λ2,Λ2) 0(Λ2,Λ+ 3 ) 0(Λ2,Λc

    work page

  64. [64]

    0(Λ2,Λu) i + ρ⊤ l3 h 0(Λ+ 3 ,Λ2) I(Λ+ 3 ,Λ+ 3 ) 0(Λ+ 3 ,Λc

    work page

  65. [65]

    (B36) It implies that ρa = 0, ρf = 0, ρl2 = 0 and ρe3 = 0

    0(Λ+ 3 ,Λu) i = ρ⊤ l2 ρ⊤ e1 + ρ⊤ l3 ρ⊤ e2 + ρ⊤ k ρ⊤ e3 = 0. (B36) It implies that ρa = 0, ρf = 0, ρl2 = 0 and ρe3 = 0. Recall the definition of Γ sub3 in (B13). Then ρ⊤Γ4 reduces to ρ⊤ e1 ρ⊤ e2 ρ⊤ h ρ⊤ i ρ⊤ j ρ⊤ k ρ⊤ l3 Γsub3 = 0. (B37) By Lemma B.5, ifba4 ̸= 0, Γsub3 has full row rank. Therefore, (B37) implies that ρe1 = 0 , ρ e2 = 0 , ρ h = 0 , ρ i = 0 ...

    work page