pith. sign in

arxiv: 2605.16262 · v1 · pith:XFFMEF2Anew · submitted 2026-02-24 · 💻 cs.LG · math.OC

Mirror Descent-Type Algorithms for the Variational Inequality Problem with Functional Constraints

Mohammad S. Alkousa , Fedor S. Stonyakin , Belal A. Alashqar , Seydamet S. Ablaev This is my paper

Pith reviewed 2026-05-21 11:37 UTC · model grok-4.3

classification 💻 cs.LG math.OC
keywords mirror descentvariational inequalitiesfunctional constraintsconvergence ratemonotone operatorsoptimization algorithmsmachine learning
0
0 comments X

The pith

Mirror descent-type algorithms with productive and non-productive switching solve constrained variational inequalities at optimal rates.

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

The paper introduces mirror descent-type algorithms for variational inequality problems subject to inequality-type functional constraints. These methods alternate between productive steps that advance toward a solution and non-productive steps that restore feasibility when constraints are violated, with various step-size rules and stopping criteria. The analysis establishes optimal convergence rates to a solution of prescribed accuracy when the operator is bounded and monotone and the functional constraints are Lipschitz continuous and convex. A modification that checks constraints selectively reduces computation when many constraints are present, and the framework extends to delta-monotone operators for approximate minimization settings. Numerical experiments illustrate practical behavior.

Core claim

The central claim is that the proposed mirror descent algorithms, which alternate between productive and non-productive iterations based on functional constraint values, achieve an optimal convergence rate to a solution of desired accuracy for variational inequalities with bounded monotone operators and Lipschitz convex functional constraints.

What carries the argument

Mirror descent with a productive/non-productive switching rule triggered by functional constraint violations.

If this is right

  • Optimal convergence rates hold for the stated class of bounded monotone operators and convex Lipschitz constraints.
  • The selective-constraint modification reduces per-iteration cost when many functional constraints are present.
  • The delta-monotone extension covers constrained minimization problems without exact subgradient information.
  • The algorithms apply directly to several machine-learning settings that reduce to constrained variational inequalities.

Where Pith is reading between the lines

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

  • The switching mechanism could be combined with stochastic or adaptive step-size rules to handle noisy oracles.
  • Similar productive/non-productive logic might improve constraint handling in other first-order methods such as extragradient or accelerated schemes.
  • Testing the approach on large-scale GAN training or adversarial robustness tasks would reveal whether the theoretical rates translate to practice.

Load-bearing premise

The operators are bounded and monotone while the functional constraints are Lipschitz continuous and convex.

What would settle it

An instance with bounded monotone operator and Lipschitz convex functional constraints where the algorithm fails to reach the claimed optimal convergence rate would falsify the result.

Figures

Figures reproduced from arXiv: 2605.16262 by Belal A. Alashqar, Fedor S. Stonyakin, Mohammad S. Alkousa, Seydamet S. Ablaev.

Figure 1
Figure 1. Figure 1: The results of Algorithms 1—7 (with first stopping criterion), for Example 4 with n = 100, m = 10, ε = 0.05 (left), and ε = 0.01 (right). Alg. 1 Alg. 2 Alg. 3 Alg. 4 Alg. 5 Alg. 6 Alg. 7 ε = 0.05 Iters. 129005 161 80 3542 3360 133169 604 Time 18.717003 0.021646 0.016614 0.748257 0.517956 37.257767 0.161242 Estim. 0.05 0.05 0.05 0.306 0.306 0.0492 0.05 ε = 0.01 Iters. 3232248 2398 712 86713 85600 3336676 30… view at source ↗
Figure 2
Figure 2. Figure 2: Results of Algorithms 2—6, for the forsaken game with different initial points, with ε = 0.001 and by 104 iterations of each algorithm, except Algorithm 6 with 105 iterations [PITH_FULL_IMAGE:figures/full_fig_p036_2.png] view at source ↗
read the original abstract

Variational inequalities play a key role in machine learning research, such as generative adversarial networks, reinforcement learning, adversarial training, and generative models. This paper is devoted to the constrained variational inequality problems with functional constraints (inequality-type constraints). We propose some mirror descent-type algorithms that switch between productive and non-productive steps depending on the values of the functional constraints at iterations, with many different step size rules and stopping criteria. We analyze the proposed algorithms and prove their optimal convergence rate to achieve a solution with desired accuracy, for problems with bounded and monotone operators and Lipschitz convex functional constraints. In addition, we propose a modification of the proposed algorithms by considering each functional constraint in the calculation when we have a productive step, as well as the first constraint that violates the feasibility. This modification can save the running time of algorithms when we have many functional constraints. In addition, we provide an analysis of the proposed algorithms for $\delta$-monotone operators, allowing us to apply the proposed algorithms, as a special case, to constrained minimization problems when we do not have access to the exact information about the subgradient of the objective function. Numerical experiments that illustrate the work and performance of the proposed algorithms are also given.

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 / 3 minor

Summary. The manuscript proposes mirror descent-type algorithms for variational inequality problems subject to functional (inequality) constraints. The algorithms alternate between productive mirror-descent steps and non-productive feasibility-adjustment steps, with multiple step-size rules and stopping criteria. The central claims are optimal convergence rates under bounded monotone operators and Lipschitz convex constraints, a modification that processes constraints selectively to reduce per-iteration cost when many constraints are present, and an extension to δ-monotone operators that covers approximate-subgradient minimization.

Significance. If the rate proofs hold, the work supplies practical, optimally convergent methods for constrained VIs that arise in GAN training, adversarial robustness, and constrained RL. The productive/non-productive switching and the selective-constraint modification are concrete algorithmic contributions that address both theory and runtime. The δ-monotone extension is a useful special case that broadens applicability without requiring exact subgradients.

major comments (2)
  1. [Convergence Analysis] Theorem 4.2 (or the main rate theorem for the switching scheme): the iteration complexity bound is stated as O(1/ε) for the gap function, but the proof sketch does not explicitly track how the non-productive steps accumulate the operator bound M; a short calculation showing that their total contribution remains O(√T) would confirm the claimed optimality.
  2. [Multiple Constraints] Section 5.2 (multiple-constraint modification): the claimed per-iteration saving is described qualitatively; an explicit operation-count comparison (e.g., number of constraint evaluations per productive step versus the baseline) is needed to substantiate the runtime claim when the number of constraints grows.
minor comments (3)
  1. [Preliminaries] Notation for the mirror map and its Bregman divergence should be introduced once in §2 and used consistently; several later displays redefine the same symbols.
  2. [Algorithm 1] The stopping criterion based on the gap function is given in Algorithm 1 but its relation to the desired accuracy ε is not restated in the complexity theorem; adding a one-line reference would improve readability.
  3. [Experiments] Numerical experiments report wall-clock time but omit the number of constraint evaluations; adding this metric would directly illustrate the benefit of the selective-constraint variant.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading of our manuscript and the constructive comments. The suggestions will help clarify the convergence analysis and runtime claims. We address each major comment below and will incorporate the requested details in the revised version.

read point-by-point responses

  1. Referee: [Convergence Analysis] Theorem 4.2 (or the main rate theorem for the switching scheme): the iteration complexity bound is stated as O(1/ε) for the gap function, but the proof sketch does not explicitly track how the non-productive steps accumulate the operator bound M; a short calculation showing that their total contribution remains O(√T) would confirm the claimed optimality.

    Authors: We agree that an explicit tracking of the non-productive steps' contribution will strengthen the presentation. In the revised manuscript we will insert a short calculation right after the proof of Theorem 4.2. Let T denote the total number of iterations and let N_p be the number of productive steps. Because each non-productive step only performs a feasibility adjustment whose operator evaluation is bounded by M and does not advance the main gap-function progress, the cumulative contribution of all non-productive steps to the gap-function bound is at most M·(T-N_p). Since the algorithm guarantees that the fraction of non-productive steps is controlled by the feasibility violation probability, this term remains O(√T) and does not degrade the overall O(1/ε) iteration complexity. The detailed algebra will be added to the main text or appendix. revision: yes

  2. Referee: [Multiple Constraints] Section 5.2 (multiple-constraint modification): the claimed per-iteration saving is described qualitatively; an explicit operation-count comparison (e.g., number of constraint evaluations per productive step versus the baseline) is needed to substantiate the runtime claim when the number of constraints grows.

    Authors: We appreciate the request for a quantitative comparison. In the revised Section 5.2 we will add an explicit operation-count table. In the baseline algorithm every productive step evaluates all m functional constraints to verify feasibility. In the selective modification, for each productive step we still evaluate all constraints (as stated in the manuscript), while non-productive steps stop at the first violating constraint. Consequently the average number of constraint evaluations per iteration drops from Θ(m) to Θ(1 + m·p) where p is the probability of a productive step; when violations occur early or when m is large the expected saving is substantial. We will include both the worst-case and average-case counts together with a brief complexity discussion. revision: yes

Circularity Check

0 steps flagged

No significant circularity; convergence analysis is self-contained

full rationale

The paper proposes mirror descent-type algorithms with productive/non-productive switching for constrained variational inequalities and derives optimal convergence rates under the standard assumptions of bounded monotone operators and Lipschitz continuous convex functional constraints. These rates follow from conventional Lyapunov-style arguments and step-size rules that are independent of any fitted parameters or self-referential definitions within the paper. No load-bearing step reduces by construction to an input quantity, self-citation chain, or renamed empirical pattern; the analysis is externally benchmarked against existing VI and mirror-descent theory.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claims rest on standard domain assumptions from variational inequality theory; no free parameters or invented entities are introduced in the abstract.

axioms (2)
  • domain assumption The operator is bounded and monotone
    Invoked as the setting in which optimal convergence rates are proved for the switching algorithms.
  • domain assumption Functional constraints are Lipschitz convex
    Required for the convergence analysis of productive and non-productive steps.

pith-pipeline@v0.9.0 · 5764 in / 1308 out tokens · 51235 ms · 2026-05-21T11:37:50.475271+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.

Reference graph

Works this paper leans on

51 extracted references · 51 canonical work pages · 2 internal anchors

  1. [1]

    S.: Finite-dimensional variational inequalities and comple- mentarity problems

    Facchinei, F., Pang, J. S.: Finite-dimensional variational inequalities and comple- mentarity problems. Springer, 2003

    work page 2003

  2. [2]

    Communications of the ACM 63, 139–144 (2020)

    Goodfellow, I., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., Courville, A., Bengio, Y.: Generative adversarial networks. Communications of the ACM 63, 139–144 (2020)

    work page 2020

  3. [3]

    Joachims, T.: A support vector method for multivariate performance measures, in: Proceedings of the 22nd international conference on Machine learning, 377–384 (2005)

    work page 2005

  4. [4]

    Advances in neural information processing systems 17 (2004)

    Xu, L., Neufeld, J., Larson, B., Schuurmans, D.: Maximum margin clustering. Advances in neural information processing systems 17 (2004)

    work page 2004

  5. [5]

    Published as a conference paper at ICLR 2018,https://openreview.net/pdf?id=rJzIBfZAb

    Madry, A., Makelov, A., Schmidt, L., Tsipras, D., Vladu, A.: Towards deep learning models resistant to adversarial attacks. Published as a conference paper at ICLR 2018,https://openreview.net/pdf?id=rJzIBfZAb

    work page 2018

  6. [6]

    Training GANs with Optimism

    Daskalakis, C., Ilyas, A., Syrgkanis, V., Zeng, H.: Training gans with optimism. arXiv preprint arXiv:1711.00141

    work page internal anchor Pith review Pith/arXiv arXiv

  7. [7]

    A variational inequality perspective on generative adversarial networks.arXiv preprint arXiv:1802.10551, 2018

    Gidel, G., Berard, H., Vignoud, G., Vincent, P., Lacoste-Julien, S.: A varia- tional inequality perspective on generative adversarial networks. arXiv preprint arXiv:1802.10551

    work page arXiv

  8. [8]

    M.: The extragradient method for finding saddle points and other problems

    Korpelevich, G. M.: The extragradient method for finding saddle points and other problems. Matecon 12, 747-–756 (1976)

    work page 1976

  9. [9]

    SIAM Journal on Optimization15, 229–251 (2004)

    Nemirovski, A.: Prox-method with rate of convergenceo(1/t)for variational in- equalities with Lipschitz continuous monotone operators and smooth convex- concave saddle point problems. SIAM Journal on Optimization15, 229–251 (2004)

    work page 2004

  10. [10]

    V., Dvurechensky, P

    Gasnikov, A. V., Dvurechensky, P. E., Stonyakin, F. S., Titov, A. A.: An adap- tive proximal method for variational inequalities. Computational Mathematics and Mathematical Physics59, 836–841 (2019)

    work page 2019

  11. [11]

    Mathematical Programming109, 319–344 (2007)

    Nesterov, Y.: Dual extrapolation and its applications to solving variational inequal- ities and related problems. Mathematical Programming109, 319–344 (2007)

    work page 2007

  12. [12]

    Malitsky, Y., Tam, M. K.: A forward-backward splitting method for monotone in- clusionswithoutcocoercivity.SIAMJournalonOptimization30,1451–1472(2020) Mirror Descent Methods for VIs with Functional Constraints 15

    work page 2020

  13. [13]

    I., Makarov, V

    Levin, M. I., Makarov, V. L., Rubinov, A. M.: Mathematical Models of Economic Interaction, Moscow: Fizmatgiz, 1993

    work page 1993

  14. [14]

    16th European Workshop on Reinforcement Learning (EWRL 2023).https://openreview.net/pdf?id=1EusBrDDrOK

    Alatur, P., Ramponi, G., He, N., Krause, A.: Provably learning Nash policies in constrained Markov potential games. 16th European Workshop on Reinforcement Learning (EWRL 2023).https://openreview.net/pdf?id=1EusBrDDrOK

    work page 2023

  15. [15]

    In proceedings of the 27th International Conference on Artificial Intelligence and Statistics (AISTATS), Valencia, Spain

    Jordan, P., Barakat, A., He, N.: Independent learning in constrained Markov po- tential games. In proceedings of the 27th International Conference on Artificial Intelligence and Statistics (AISTATS), Valencia, Spain. PMLR,238(2024)

    work page 2024

  16. [16]

    I., Lin, T., Zampetakis, M.: First-order algorithms for nonlinear gener- alized Nash equilibrium problems

    Jordan, M. I., Lin, T., Zampetakis, M.: First-order algorithms for nonlinear gener- alized Nash equilibrium problems. Journal of Machine Learning Research, 24(38), 1–46 (2023)

    work page 2023

  17. [17]

    M.: Editorial: Hierarchical and bilevel programming, J

    Migdalas, A., Pardalos, P. M.: Editorial: Hierarchical and bilevel programming, J. Global Optim.8, 209–215 (1996)

    work page 1996

  18. [18]

    Baiocchi, C., Capelo, A.: Variational and Quasi-variational Inequalities: Applica- tions to Free Boundary Problems, Chichester: Wiley, 1984

    work page 1984

  19. [19]

    S.: Solution Methods for Variational Inequalities with Coupled Con- straints

    Antipin, A. S.: Solution Methods for Variational Inequalities with Coupled Con- straints. Computational Mathematics and Mathematical Physics40(9), 123–1254 (2000)

    work page 2000

  20. [20]

    In International Conference on Machine Learning, PMLR, 11480–11491 (2021)

    Xu, T., Liang, Y., Lan, G.: CRPO: A new approach for safe reinforcement learning with convergence guarantee. In International Conference on Machine Learning, PMLR, 11480–11491 (2021)

    work page 2021

  21. [21]

    Transactions on Machine Learning Research, 2835–8856 (2022)

    Lowy, A., Baharlouei, S., Pavan, R., Razaviyayn, M., Beirami, A.: A stochastic op- timization framework for fair risk minimization. Transactions on Machine Learning Research, 2835–8856 (2022)

    work page 2022

  22. [22]

    B., Valera, I., Gomez-Rodriguez, M., Gummadi, K

    Zafar, M. B., Valera, I., Gomez-Rodriguez, M., Gummadi, K. P.: Fairness con- straints: A flexible approach for fair classification. Journal of Machine Learning Research20(1), 2737–2778 (2019)

    work page 2019

  23. [23]

    Auslender, A.: Variational inequalities over the cone of semidefinite positive sym- metric matrices and over the Lorentz cone, the Second Japanese-Sino Optimization Meeting, Part II, Kyoto, 2002. Optim. Methods Softw. 18, 359–376 (2003)

    work page 2002

  24. [24]

    S., Yang, H., Zhang, C

    He, B. S., Yang, H., Zhang, C. S.: A modified augmented Lagrangian method for a class of monotone variational inequalities. Eur. J. Oper. Res.159, 35–51 (2004)

    work page 2004

  25. [25]

    L.: Augmented Lagrangian theory, duality and decomposition methods for variational inequality problems

    Zhu, D. L.: Augmented Lagrangian theory, duality and decomposition methods for variational inequality problems. J. Optim. Theory Appl.117, 195–216 (2003)

    work page 2003

  26. [26]

    Zhang, L., He, N., Muehlebach, M.: Primal Methods for Variational Inequality Problems with Functional Constraints. Math. Program. (2025)

    work page 2025

  27. [27]

    I., Chavdarova, T.: Solving constrained variational inequali- ties via an interior point method

    Yang, T., Jordan, M. I., Chavdarova, T.: Solving constrained variational inequali- ties via an interior point method. In ICLR, (2023)

    work page 2023

  28. [28]

    Published as a con- ference paper at ICLR 2024https://openreview.net/pdf?id=RsztjXcvUf

    Chavdarova, T., Yang, T., Pagliardini, M., Jordan, M.: A Primal-Dual Approach to Solving Variational Inequalities with General Constraints. Published as a con- ference paper at ICLR 2024https://openreview.net/pdf?id=RsztjXcvUf

    work page

  29. [29]

    Ekonomika i Matematicheskie Metody, 1979

    Nemirovskii, A.: Efficient methods for large-scale convex optimization problems. Ekonomika i Matematicheskie Metody, 1979. (in Russian)

    work page 1979

  30. [30]

    Nemirovsky, A., Yudin, D.: Problem Complexity and Method Efficiency in Opti- mization. J. Wiley & Sons, New York 1983

    work page 1983

  31. [31]

    Beck, A., Teboulle, M.: Mirror descent and nonlinear projected subgradient meth- ods for convex optimization. Oper. Res. Lett., 31(3), 167–175 (2003)

    work page 2003

  32. [32]

    Z.: Generalized gradient descent with application to block programming

    Shor, N. Z.: Generalized gradient descent with application to block programming. Kibernetika,3(3), 53–55 (1967)

    work page 1967

  33. [33]

    Soviet Mathematics Doklady8(3), 593–597, (1967) 16 M

    Polyak, B.: A general method of solving extremum problems. Soviet Mathematics Doklady8(3), 593–597, (1967) 16 M. Alkousa et al

    work page 1967

  34. [34]

    F., Vasil’yev, L

    Dem’yanov, V. F., Vasil’yev, L. V.: Nondifferentiable Optimization. Nauka, Moscow, 1981

    work page 1981

  35. [35]

    T., Bose, S., Nguyen, D

    Doan, T. T., Bose, S., Nguyen, D. H., Beck, C. L.: Convergence of the Iterates in Mirror Descent Methods. IEEE Control Systems Letters,3(1), 114–119 (2019)

    work page 2019

  36. [36]

    S., Stonyakin, F

    Alkousa, M. S., Stonyakin, F. S., Abdo, A. M., Alcheikh, M. M.: Mirror Descent Methods with Weighting Scheme for Outputs for Optimization Problems with Functional Constraints. Russian Journal of Nonlinear Dynamics20(5), 727–745 (2024)

    work page 2024

  37. [37]

    Titov: Mirror descent and convex optimization problems with non-smooth inequality constraints

    Bayandina, A., Dvurechensky, P., Gasnikov, A., Stonyakin, F., A. Titov: Mirror descent and convex optimization problems with non-smooth inequality constraints. In: Large-Scale and Distributed Optimization, Springer, Cham 181–213 (2018) https://doi.org/10.1007/978-3-319-97478-1_8

    work page doi:10.1007/978-3-319-97478-1_8 2018

  38. [38]

    S., Alkousa, M

    Stonyakin, F. S., Alkousa, M. S., Stepanov, A. N., Barinov, M. A.: Adaptive Mirror Descent Algorithms in Convex Programming Problems with Lipschitz Constraints. Trudy Inst. Mat. Mekh. Ural. Otdel. Ross. Akad. Nauk24(2), 266-279 (2018)

    work page 2018

  39. [39]

    S., Alkousa, M

    Stonyakin, F. S., Alkousa, M. S., Stepanov, A. N., Titov, A. A.: Adaptive Mirror Descent Algorithms for Convex and Strongly Convex Optimization Problems with Functional Constraints. J. Appl. Ind. Math.13, 557–574 (2019)

    work page 2019

  40. [40]

    S.: On Modification of an Adaptive Stochastic Mirror Descent Algo- rithm for Convex Optimization Problems with Functional Constraints

    Alkousa, M. S.: On Modification of an Adaptive Stochastic Mirror Descent Algo- rithm for Convex Optimization Problems with Functional Constraints. In: Zeidan, D., Padhi, S., Burqan, A., Ueberholz, P. (eds) Computational Mathematics and Applications. Forum for Interdisciplinary Mathematics. Springer, Singapore, 47–63 (2020)https://doi.org/10.1007/978-981-1...

    work page doi:10.1007/978-981-15-8498-5_3 2020

  41. [41]

    SIAM Journal on Optimization30(2) (2020)

    Xu, Y.: Primal-dual stochastic gradient method for convex programs with many functional constraints. SIAM Journal on Optimization30(2) (2020)

    work page 2020

  42. [42]

    arXiv:2408.06606

    Qu, X., Bian, W., Chen, X.: An extra gradient Anderson-accelerated algorithm for pseudomonotone variational inequalities. arXiv:2408.06606

    work page internal anchor Pith review arXiv

  43. [43]

    T.: Introduction to optimization

    Polyak, B. T.: Introduction to optimization. Optimization Software, Inc, New York 1987

    work page 1987

  44. [44]

    Hsieh, Y., Mertikopoulos, P., Cevher, V.: The Limits of Min-Max Optimization Algorithms: Convergence to Spurious Non-Critical Sets. Proceedings of the 38th International Conference on Machine Learning, PMLR 139, 4337–4348 (2021) Mirror Descent Methods for VIs with Functional Constraints 17 7 Appendix 1: The Proofs of Theorems in Section 3 7.1 Proof Theore...

    work page 2021

  45. [45]

    For Algorithm 1, we have, –with stopping criterion 1 (13), we get ⟨F(x),bx−x⟩< ε+δ∀x∈Q.(74) –With stopping criterion 2 (14), we get ⟨F(x),bx−x⟩< ε+δ+ DL2 F |J| Mg|I| ∀x∈Q.(75)

    work page

  46. [46]

    –With stopping criterion 2 (16), we get ⟨F(x),bx−x⟩< ε+δ+M gD X i∈J 1 ∥∇g(xi)∥2∗ X i∈I 1 ∥F(x i)∥2∗ !−1 ∀x∈Q

    For Algorithm 2, we have, –with stopping criterion 1 (15), we get ⟨F(x),bx−x⟩< ε+δ∀x∈Q. –With stopping criterion 2 (16), we get ⟨F(x),bx−x⟩< ε+δ+M gD X i∈J 1 ∥∇g(xi)∥2∗ X i∈I 1 ∥F(x i)∥2∗ !−1 ∀x∈Q

    work page

  47. [47]

    –With stopping criterion 2, we get ⟨F(x),bx−x⟩< ε+δ+D|J| X i∈I 1 ∥F(x i)∥2∗ !−1 ∀x∈Q

    For Algorithm 3, we have, –with stopping criterion 1, we get ⟨F(x),bx−x⟩< ε+δ∀x∈Q. –With stopping criterion 2, we get ⟨F(x),bx−x⟩< ε+δ+D|J| X i∈I 1 ∥F(x i)∥2∗ !−1 ∀x∈Q

    work page

  48. [48]

    –With stopping criterion 2, we get ⟨F(x),bx−x⟩< εL F +δ+ MgDLF |I| X i∈I 1 ∥∇g(xi)∥2∗ ∀x∈Q

    For Algorithm 4, we have, Mirror Descent Methods for VIs with Functional Constraints 35 –with stopping criterion 1, we get ⟨F(x),bx−x⟩< εL F +δ∀x∈Q. –With stopping criterion 2, we get ⟨F(x),bx−x⟩< εL F +δ+ MgDLF |I| X i∈I 1 ∥∇g(xi)∥2∗ ∀x∈Q

    work page

  49. [49]

    –With stopping criterion 2, we get ⟨F(x),bx−x⟩< εL F +δ+ DLF |J| |I| ∀x∈Q

    For Algorithm 5, we have, –with stopping criterion 1, we get ⟨F(x),bx−x⟩< εL F +δ∀x∈Q. –With stopping criterion 2, we get ⟨F(x),bx−x⟩< εL F +δ+ DLF |J| |I| ∀x∈Q

    work page

  50. [50]

    –With stopping criterion 2, we get ⟨F(x),bx−x⟩< εLF Mg +δ+ DLF |J| Mg|I| ∀x∈Q

    For Algorithm 6, we have, –with stopping criterion 1, we get ⟨F(x),bx−x⟩< εLF Mg +δ∀x∈Q. –With stopping criterion 2, we get ⟨F(x),bx−x⟩< εLF Mg +δ+ DLF |J| Mg|I| ∀x∈Q

    work page

  51. [51]

    –With stopping criterion 2, we get ⟨F(x),bx−x⟩< ε+ |J|M gD |I| +δ∀x∈Q

    For Algorithm 7, we have, –with stopping criterion 1, we get ⟨F(x),bx−x⟩< ε+δ∀x∈Q. –With stopping criterion 2, we get ⟨F(x),bx−x⟩< ε+ |J|M gD |I| +δ∀x∈Q. LetusbrieflyshowhowwecanprovetheresultofAlgorithm1forδ-monotone operatorF, i.e., the results (74) and (75). The proof will be the same as the proof of Theorem 1, with a slight modification concerning the...

    work page