Pith. sign in

REVIEW 1 major objections 7 minor 60 references

Sample averages converge to the true objective in a Lipschitz norm under separability or NIP-definability of the random function family.

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

2026-08-01 09:54 UTC pith:VKRJBQC3

load-bearing objection A strong, credible paper that proves Lipschitzian SLLNs under separability or NIP-style definability and derives uniform subdifferential convergence; worth serious refereeing. the 1 major comments →

arxiv 2607.20411 v1 pith:VKRJBQC3 submitted 2026-07-22 math.OC cs.LGmath.STstat.TH

Lipschitzian SLLNs for random functions

Lai Tian , Johannes O. Royset This is my paper
classification math.OC cs.LGmath.STstat.TH MSC 60F1590C1549J5203C64
keywords strong law of large numbersLipschitz pseudometricsample average approximationnonsmooth stochastic programminglimiting subdifferentialClarke subdifferentialNIP structuresGlivenko–Cantelli
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 establishes strong laws of large numbers for possibly nonconvex, nonsmooth random functions, comparing sample-average and population objectives in a Lipschitz pseudometric rather than pointwise or epigraphically. Under either separability of the function family or a model-theoretic definability condition called piecewise uniform definability in NIP structures, the distance d_X(E_ν f, E f) tends to zero almost surely, and at the canonical root-n rate when only finitely many definable pieces are present. Because this metric controls uniform function values on bounded sets and Hausdorff distances between limiting and Clarke subdifferentials, the theorem yields uniform SLLNs for subdifferentials and consistency of stationary points without interchanging expectation and subdifferentiation. It also gives finite-sample identification of global and stationary solutions under sharpness conditions, and it identifies broad classes, including many deep-learning losses, where earlier failure phenomena do not occur.

Core claim

The paper's central claim is that the sample-average function E_ν f converges to the population objective E f in the Lipschitz pseudometric d_X, defined by d_X(h,g) = max{|h(0)-g(0)|, g-lip_X(h-g)}, under two structural conditions: an explicit separability assumption on the family of slices, or piecewise uniform definability in NIP expansions of the real ordered field. The definability condition holds automatically for jointly definable functions in o-minimal structures, such as semialgebraic and exponential-field-definable losses, and extends beyond them to functions with uniformly or slicewise definable restrictions. The authors prove the almost-sure limit and, under finitely many pieces a

What carries the argument

The central object is the Lipschitz extended pseudometric d_X(h,g) = max{|h(0)-g(0)|, g-lip_X(h-g)}, which makes the space of Lipschitz functions on X a Banach space under the associated Lipschitz norm. The proof reduces convergence in this metric to a Glivenko–Cantelli property of the class of difference quotients g_{x,y}(ξ) = (f(ξ,x)-f(ξ,y))/||x-y|| indexed by a countable dense subset of X. Under NIP definability, the subgraphs of these quotients form a VC-class, so uniform entropy bounds and empirical process theory apply; under separability, the slices form a Bochner-integrable random variable in a separable Banach space, reducing the claim to a classical Banach-valued strong law.

Load-bearing premise

The results hinge on Assumption 3.7, which requires that on each piece of a countable cover of the sample space, the values of every slice on a fixed countable dense set be simultaneously definable by a single formula in some NIP structure; without this uniformity the VC-class reduction and the Glivenko–Cantelli argument collapse, and the alternative separability condition excludes even simple functions like |x-ξ|.

What would settle it

Simulate the nonseparable but semialgebraic case f(ξ,x)=|x-ξ| with ξ uniform on [0,1] and X=[0,1]; the theorem predicts d_X(E_ν f, E f)→0 at rate O(ν^{-1/2}). If a direct computation or simulation shows the distance remaining bounded away from zero, the definability-based theorem is false. Conversely, construct slices that are pointwise definable but not uniformly definable on any countable dense set, violating Assumption 3.7's uniformity, and check whether convergence fails, as the authors conjecture it does.

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

If this is right

  • Empirical (sample-average) objectives converge to the population objective in a metric that controls both function values on bounded sets and global Lipschitz constants, for broad classes of nonconvex nonsmooth stochastic programs.
  • Uniform Hausdorff convergence of limiting and Clarke subdifferentials holds without interchanging expectation and subdifferentiation, yielding consistency of limiting stationary points for irregular locally Lipschitz functions.
  • Under sharpness, the set of minimizers (and of stationary points) of the sample-average problem is eventually contained in the true solution set, after finitely many samples, without requiring convexity, polyhedrality, or finite support of the distribution.
  • The convergence rate is O_P(ν^{-1/2}) when the definable cover has finitely many pieces and second moments exist; otherwise convergence can be arbitrarily slow, as shown by a missing-mass construction.
  • Many deterministic tame classes of functions arising in modern learning, such as losses definable in the real exponential field, satisfy the Lipschitzian SLLN, ruling out previously known failure phenomena.

Where Pith is reading between the lines

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

  • A natural extension is that the same Lipschitz pseudometric controls other first-order objects, such as metric slopes and generalized directional derivatives, so analogous SLLNs should follow with essentially the same proofs.
  • The NIP-based machinery suggests a testable criterion: a family of slices satisfies the Lipschitzian SLLN as soon as its difference-quotient subgraphs form a VC-class, which one could verify computationally for parametric losses without explicit model-theoretic definability.
  • The missing-mass construction implies that moment conditions alone cannot guarantee the root-n rate; rate bounds must incorporate the combinatorial complexity of the function class, paralleling the Glivenko-Cantelli versus Donsker distinction in empirical process theory.
  • The paper conjectures that the countability assumptions in the slicewise definability theorem are essential; a concrete test is to seek a counterexample with an uncountable language or uncountable index set, which would sharpen the boundary of the theory.

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

1 major / 7 minor

Summary. The paper establishes strong laws of large numbers for sample-average approximations of random locally Lipschitz functions in the Lipschitz pseudometric d_X(h,g)=max{|h(0)-g(0)|, g-lip_X(h-g)}. Under a blanket integrability/Lipschitz assumption (Assumption 3.1), Theorem 3.3 proves d_X(E_ν f, E f) -> 0 a.s. under a separability condition on the slice family in Lip(X); Theorem 3.8 proves the same under a piecewise uniform NIP-definability condition; Theorem 3.9 adds the rate O_P(ν^{-1/2}) when there are finitely many definability pieces and second moments; Proposition 3.10 shows the rate can be arbitrarily slow with countably many semialgebraic pieces. Theorem 4.10 upgrades slicewise definability in countable-language totally Borel NIP structures, in particular o-minimal slices, to Assumption 3.7. Applications are given to uniform Hausdorff convergence of limiting and Clarke subdifferentials, stationarity consistency, finite-sample identification of minimizers and stationary points, and permanence properties.

Significance. The main positive results are significant. They supply a uniform convergence mode that controls subdifferentials of SAA objectives without interchanging expectation and subdifferentiation, and they cover broad classes of nonconvex nonsmooth objectives, including o-minimal-definable losses and many deep-learning training losses. The proof architecture is sound: Theorem 3.3 is a clean reduction to the Bochner SLLN in separable Banach spaces; Theorem 3.8 reduces the global Lipschitz modulus to Glivenko-Cantelli for VC-subgraph classes of difference quotients; Theorem 3.9 uses a Donsker argument. The delicate uniform-definability requirement in Assumption 3.7 is handled carefully, and Theorem 4.10 gives a valid mechanism for obtaining it from pointwise slicewise definability via a Borel projection argument. The negative baseline Proposition 3.6 and the slow-rate example Proposition 3.10 calibrate the assumptions. One display in the subdifferential application needs correction, but the intended conclusion is recoverable from the proof as written.

major comments (1)
  1. [Section 5.2, Proposition 5.2; also Eq. (4) in Section 1] The displayed chain is false as stated: d_l(∂h,∂g) for limiting subdifferentials is not generally ≤ d_l(∂̄h,∂̄g) for Clarke subdifferentials; the reverse inequality holds because taking convex hulls cannot increase Hausdorff distance. For example, h(x)=|x| and g(x)=-|x| at 0 give d_l(∂h(0),∂g(0))=1 but d_l(∂̄h(0),∂̄g(0))=0. The proof of Proposition 5.2 actually proves sup_x d_l(∂̄h,∂̄g) ≤ sup_x d_l(∂h,∂g) ≤ sup_x lip(h-g)(x) ≤ d_X(h,g). Please reverse the first inequality in Eq. (4) and in Proposition 5.2. The applications in Corollary 5.3 and the stationarity-consistency results remain valid after this correction, since both subdifferential quantities are still bounded by d_X(h,g).
minor comments (7)
  1. [Section 2.1] The definitions of g-lip_X h and lip h(x) use the signed quotient (h(x)-h(y))/||x-y|| without an absolute value. Since the supremum is over ordered pairs this is equivalent to the standard absolute-value modulus, but the text should say so explicitly to avoid ambiguity.
  2. [Section 3.3.1, proof of Theorem 3.3] The sentence 'By restricting the defining supremum to the countable set D' is terse. It is valid because each slice is L(ξ)-Lipschitz on X, so the modulus over X equals the modulus over the dense set D; adding one sentence explaining this would help the reader.
  3. [Section 3.3.4, Proposition 3.10] In the definition of g_k, the formula appears as ⌊2kt⌋; this should presumably be ⌊2^k t⌋. Please correct the typographical loss of the exponent.
  4. [Section 4.2.2, Example 4.7] The notation 'Let Z(ξ)=1_Z(ξ)' overloads the letter Z, which is also used for the set of integers. Use something like 'Let z(ξ)=𝟙_ℤ(ξ)'.
  5. [Section 4.3.3] The subsection title 'Missing details in Remark 4.12' is odd; it is actually a supplement to the remark. Rename it, e.g., 'Supplement to Remark 4.12'.
  6. [Section 5.2] In the first display of the subdifferential definitions, the Clarke subdifferential is typeset as ∂h(x)=con∂h(x); the overline or other distinguishing notation for the Clarke object should be restored so that the subsequent proposition and proof are unambiguous.
  7. [Proposition 3.6 and references] The failure baseline relies on the companion preprint [53], and the reader must consult that paper for the construction of f. If a published version of [53] exists, it would be helpful to cite it; otherwise consider stating the needed conclusion more explicitly.

Circularity Check

0 steps flagged

No significant circularity: central theorems derived from stated assumptions plus external results; self-citation used only as failure baseline.

full rationale

Walking the derivation chain, the positive laws (Theorems 3.3, 3.8, 3.9, and 4.10) follow from the stated assumptions (3.1, 3.2, 3.7, and 4.9) via external results: separable Banach-valued SLLNs, VC/Glivenko–Cantelli/Donsker theory, and NIP/o-minimal definability facts. Assumption 3.7 is a sufficient structural condition, not an encoding of the target convergence; Theorem 4.10 proves Assumption 3.7 from slicewise definability by constructing universally measurable pieces and value-sequence identifiability, and this argument does not presuppose the conclusion. The only self-citation, to the authors' prior negative result [53], is used to establish a failure baseline and to motivate additional assumptions (e.g., Proposition 3.6); it is not an input to the positive law, so it is not load-bearing. Minor notation issues, such as the missing absolute value in the displayed definition of g-lip, do not affect the logical derivation. Therefore, no circular step is present.

Axiom & Free-Parameter Ledger

0 free parameters · 10 axioms · 0 invented entities

No fitted free parameters or invented entities appear. The central results rest on standard empirical process/model-theoretic theorems plus the three structural domain assumptions (separability, piecewise uniform definability, slicewise definability). The self-cited negative result [53] is used only to establish that Assumption 3.1 alone is insufficient and does not feed the positive derivations.

axioms (10)
  • standard math Bochner SLLN for iid Bochner-integrable random variables in separable Banach spaces (Ledoux–Talagrand).
    Used in proof of Theorem 3.3 to obtain convergence of sample averages of F(ξ)=f(ξ,·)|X in Lip(X).
  • standard math Glivenko–Cantelli theorem for VC-subgraph classes with integrable envelope, and P-Donsker theorem for VC-subgraph classes with square-integrable envelope (van der Vaart–Wellner).
    Basis for Theorem 3.8 and Theorem 3.9.
  • standard math NIP formulas define VC classes, and o-minimal structures are NIP (Laskowski; Simon).
    Bridges model-theoretic definability to VC theory, used in Lemma 3.11 and Section 4.2.
  • standard math Clarke subdifferential calculus: ∂h(x)⊂∂g(x)+∂(h-g)(x) and ∂(h-g)(x)⊂lip(h-g)(x)B, plus local Lipschitz modulus bounds (Rockafellar–Wets).
    Transfer Lipschitz pseudometric convergence to Hausdorff convergence of subdifferentials in Proposition 5.2.
  • standard math Every o-minimal structure is totally Borel; R_exp is o-minimal; countable-language reducts of R_an are o-minimal (Kaiser; Wilkie).
    Used in Theorem 4.10 and Corollary 4.11 to convert slicewise definability into Assumption 3.7.
  • domain assumption The negative result in [53, Theorem 3]: under Assumption 3.1 alone, d_X(E_ν f,E_f) may fail to converge, even for convex f.
    Used in Proposition 3.6 to show Assumption 3.1 alone is insufficient; motivates Assumptions 3.2 and 3.7.
  • domain assumption Assumption 3.1: A-measurability in ξ, E|f(ξ,0)|<∞, and L(ξ)-Lipschitz slices with EL<∞.
    Blanket well-definedness and integrability condition for all theorems.
  • domain assumption Assumption 3.2: {f(ξ,·)|X} lies in a separable subspace of Lip(X).
    Sufficient condition for Theorem 3.3.
  • domain assumption Assumption 3.7: piecewise uniform definability in NIP structures over a countable dense set.
    Main model-theoretic condition for Theorem 3.8.
  • domain assumption Assumption 4.9: slicewise definability in countable-language totally Borel NIP structures.
    Derived sufficient condition for Assumption 3.7 in Theorem 4.10.

pith-pipeline@v1.3.0-alltime-deepseek · 28360 in / 22871 out tokens · 170787 ms · 2026-08-01T09:54:26.975918+00:00 · methodology

0 comments
read the original abstract

We prove strong laws of large numbers for locally Lipschitz functions in the Lipschitz pseudometric. Our results hold under either a topological or a model-theoretic condition, with the latter encompassing functions jointly definable in o-minimal structures but extending substantially beyond this class. Applications include uniform convergence of limiting and Clarke subdifferentials and finite-sample identification of solutions. Consequently, we identify broad classes of functions for which the failure phenomena revealed by our previous negative results [Tian and Royset, arXiv:2511.16568, 2025] do not occur.

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

60 extracted references · 4 linked inside Pith

  1. [1]

    C. D. Aliprantis and K. C. Border.Infinite Dimensional Analysis: A Hitchhiker’s Guide. Springer, 3rd edition, 2006

    2006

  2. [2]

    Areces, J

    F. Areces, J. Duchi, and M. Sommers. Finding a stationary point of a stochastic convex problem.arXiv preprint arXiv:2607.06883, 2026

    Pith/arXiv arXiv 2026

  3. [3]

    Artstein and R

    Z. Artstein and R. J-B Wets. Consistency of minimizers and the SLLN for stochastic programs. Journal of Convex Analysis, 2(1-2):1–17, 1995

    1995

  4. [4]

    R. J. Aumann. Integrals of set-valued functions.Journal of Mathematical Analysis and Ap- plications, 12(1):1–12, 1965. 26

    1965

  5. [5]

    Bareilles, A

    G. Bareilles, A. Gehret, J. Aspman, J. Lepˇ sov´ a, and J. Mareˇ cek. Deep learning as the disci- plined construction of tame objects.arXiv preprint arXiv:2509.18025, 2025

    Pith/arXiv arXiv 2025

  6. [6]

    Beer.Bornologies and Lipschitz Analysis

    G. Beer.Bornologies and Lipschitz Analysis. CRC Press, 2023

    2023

  7. [7]

    Beer and M

    G. Beer and M. J. Hoffman. The Lipschitz metric for real-valued continuous functions.Journal of Mathematical Analysis and Applications, 406(1):229–236, 2013

    2013

  8. [8]

    Berend and A

    D. Berend and A. Kontorovich. The missing mass problem.Statistics & Probability Letters, 82(6):1102–1110, 2012

    2012

  9. [9]

    Bolte and E

    J. Bolte and E. Pauwels. Conservative set valued fields, automatic differentiation, stochastic gradient methods and deep learning.Mathematical Programming, 188(1):19–51, 2021

    2021

  10. [10]

    Bolte, A

    J. Bolte, A. Daniilidis, A. Lewis, and M. Shiota. Clarke subgradients of stratifiable functions. SIAM Journal on Optimization, 18(2):556–572, 2007

    2007

  11. [11]

    Bolte, T

    J. Bolte, T. Le, and E. Pauwels. Subgradient sampling for nonsmooth nonconvex minimization. SIAM Journal on Optimization, 33(4):2542–2569, 2023

    2023

  12. [12]

    J. V. Burke and M. C. Ferris. Weak sharp minima in mathematical programming.SIAM Journal on Control and Optimization, 31(5):1340–1359, 1993

    1993

  13. [13]

    J. V. Burke, X. Chen, and H. Sun. The subdifferential of measurable composite max integrands and smoothing approximation.Mathematical Programming, 181(2):229–264, 2020

    2020

  14. [14]

    C. C. Chang and H. J. Keisler.Model Theory, volume 73. Elsevier, 3rd edition, 1990

    1990

  15. [15]

    Chase and J

    H. Chase and J. Freitag. Model theory and machine learning.Bulletin of Symbolic Logic, 25 (3):319–332, 2019

    2019

  16. [16]

    F. H. Clarke.Optimization and Nonsmooth Analysis. SIAM, 1990

    1990

  17. [17]

    D. L. Cohn.Measure Theory. Birkh¨ auser Advanced Texts Basler Lehrb¨ ucher. Springer, 2nd edition, 2013

    2013

  18. [18]

    Coste.An Introduction to O-minimal Geometry

    M. Coste.An Introduction to O-minimal Geometry. RAAG Notes. Institut de Recherche Math´ ematiques de Rennes, 1999

    1999

  19. [19]

    Coste.An Introduction to Semialgebraic Geometry

    M. Coste.An Introduction to Semialgebraic Geometry. RAAG Notes. Institut de Recherche Math´ ematiques de Rennes, 2002

    2002

  20. [20]

    Davis and D

    D. Davis and D. Drusvyatskiy. Graphical convergence of subgradients in nonconvex optimiza- tion and learning.Mathematics of Operations Research, 47(1):209–231, 2022

    2022

  21. [21]

    Davis, D

    D. Davis, D. Drusvyatskiy, S. Kakade, and J. D. Lee. Stochastic subgradient method converges on tame functions.Foundations of Computational Mathematics, 20(1):119–154, 2020

    2020

  22. [22]

    Devale, P

    T. Devale, P. Devulapalli, and S. Hanneke. Uniform convergence beyond Glivenko-Cantelli. In Matus Telgarsky and Jonathan Ullman, editors,International Conference on Algorithmic Learning Theory, volume 313 ofProceedings of Machine Learning Research, pages 1–21. PMLR, 23–26 Feb 2026. 27

    2026

  23. [23]

    G. B. Folland.Real Analysis: Modern Techniques and Their Applications. John Wiley & Sons, 2nd edition, 1999

    1999

  24. [24]

    D. J. Foster, A. Sekhari, and K. Sridharan. Uniform convergence of gradients for non-convex learning and optimization.Advances in Neural Information Processing Systems, 31:8759–8770, 2018

    2018

  25. [25]

    R. M. Gray and D. L. Neuhoff. Quantization.IEEE Transactions on Information Theory, 44 (6):2325–2383, 2002

    2002

  26. [26]

    G¨ unaydin and P

    A. G¨ unaydin and P. Hieronymi. Dependent pairs.The Journal of Symbolic Logic, 76(2): 377–390, 2011

    2011

  27. [27]

    Hieronymi

    P. Hieronymi. Defining the set of integers in expansions of the real field by a closed discrete set.Proceedings of the American Mathematical Society, 138(6):2163–2168, 2010

    2010

  28. [28]

    A. D. Ioffe. An invitation to tame optimization.SIAM Journal on Optimization, 19(4):1894– 1917, 2009

    1917

  29. [29]

    Iusem and A

    A. Iusem and A. Seeger. Distances between closed convex cones: Old and new results.Journal of Convex Analysis, 17(3-4):1033–1055, 2010

    2010

  30. [30]

    T. Kaiser. First order tameness of measures.Annals of Pure and Applied Logic, 163(12): 1903–1927, 2012

    1903

  31. [31]

    Karpinski and A

    M. Karpinski and A. Macintyre. Polynomial bounds for VC dimension of sigmoidal and general Pfaffian neural networks.Journal of Computer and System Sciences, 54(1):169–176, 1997

    1997

  32. [32]

    A. J. Kleywegt, A. Shapiro, and T. Homem-de-Mello. The sample average approximation method for stochastic discrete optimization.SIAM Journal on Optimization, 12(2):479–502, 2002

    2002

  33. [33]

    L. S. Krapp, M. Vermeil, and L. Wirth. On tameness, measurability and the independence property.arXiv preprint arXiv:2506.08733, 2025

    Pith/arXiv arXiv 2025

  34. [34]

    M. C. Laskowski. Vapnik-Chervonenkis classes of definable sets.Journal of the London Math- ematical Society, 2(2):377–384, 1992

    1992

  35. [35]

    Ledoux and M

    M. Ledoux and M. Talagrand.Probability in Banach Spaces: Isoperimetry and Processes, volume 23. Springer, 1991

    1991

  36. [36]

    Livni and P

    R. Livni and P. Simon. Honest compressions and their application to compression schemes. In Conference on Learning Theory, pages 77–92. PMLR, 2013

    2013

  37. [37]

    Macintyre and E

    A. Macintyre and E. D. Sontag. Finiteness results for sigmoidal “neural” networks. InACM Symposium on Theory of Computing, pages 325–334, 1993

    1993

  38. [38]

    S. Mei, Y. Bai, and A. Montanari. The landscape of empirical risk for nonconvex losses.Annals of Statistics, 46(6A):2747–2774, 2018

    2018

  39. [39]

    H. D. Nguyen, J. Westerhout, and X. Guo. On rates of convergence for sample average approximations without smoothness.arXiv preprint arXiv:2604.25153, 2026. 28

    Pith/arXiv arXiv 2026

  40. [40]

    V. I. Norkin and R. J-B Wets. On strong graphical law of large numbers for random semicon- tinuous mappings.Vestnik Sankt-Peterburgskogo Universiteta, Seriya 10(3):102–111, 2013

    2013

  41. [41]

    R. T. Rockafellar. Favorable classes of Lipschitz-continuous functions in subgradient optimiza- tion. In E. Nurminski, editor,Progress in Nondifferentiable Optimization, IIASA Collaborative Proceedings Series, pages 125–144. International Institute for Applied Systems Analysis, Lax- enburg, Austria, 1982

    1982

  42. [42]

    R. T. Rockafellar and R. J-B Wets.Variational Analysis. Springer, 3rd printing-2009 edition, 1998

    2009

  43. [43]

    F. Ruan. On the uniform convergence of subdifferentials in stochastic optimization and learn- ing.Mathematics of Operations Research, to appear, 2025

    2025

  44. [44]

    Schechtman

    S. Schechtman. The gradient’s limit of a definable family of functions admits a variational stratification.SIAM Journal on Optimization, 36(2):1075–1099, 2026

    2026

  45. [45]

    Shalev-Shwartz and S

    S. Shalev-Shwartz and S. Ben-David.Understanding Machine Learning: From Theory to Algorithms. Cambridge University Press, 2014

    2014

  46. [46]

    Shapiro and T

    A. Shapiro and T. Homem-de-Mello. On the rate of convergence of optimal solutions of Monte Carlo approximations of stochastic programs.SIAM Journal on Optimization, 11(1):70–86, 2000

    2000

  47. [47]

    Shapiro and H

    A. Shapiro and H. Xu. Uniform laws of large numbers for set-valued mappings and subdif- ferentials of random functions.Journal of Mathematical Analysis and Applications, 325(2): 1390–1399, 2007

    2007

  48. [48]

    Shapiro, T

    A. Shapiro, T. Homem-de Mello, and J. Kim. Conditioning of convex piecewise linear stochastic programs.Mathematical Programming, 94(1):1–19, 2002

    2002

  49. [49]

    Shapiro, D

    A. Shapiro, D. Dentcheva, and A. Ruszczy´ nski.Lectures on Stochastic Programming: Modeling and Theory. SIAM, 3rd edition, 2021

    2021

  50. [50]

    Simon.A Guide to NIP Theories

    P. Simon.A Guide to NIP Theories. Lecture Notes in Logic. Cambridge University Press, 2015

    2015

  51. [51]

    C. I. Steinhorn. Chapter XVI: Borel structures and measure and category logics. InModel- Theoretic Logics, volume 8, pages 579–597. Association for Symbolic Logic, 1985

    1985

  52. [52]

    P. Ter´ an. On a uniform law of large numbers for random sets and subdifferentials of random functions.Statistics & Probability Letters, 78(1):42–49, 2008

    2008

  53. [53]

    Tian and J

    L. Tian and J. O. Royset. Failure of uniform laws of large numbers for subdifferentials and beyond.arXiv preprint arXiv:2511.16568, 2025

    arXiv 2025

  54. [54]

    A. W. van der Vaart and J. A. Wellner.Weak Convergence and Empirical Processes. Springer, 2nd edition, 2023

    2023

  55. [55]

    van den Dries.Tame Topology and O-minimal Structures, volume 248

    L. van den Dries.Tame Topology and O-minimal Structures, volume 248. Cambridge University Press, 1998. 29

    1998

  56. [56]

    van den Dries and C

    L. van den Dries and C. Miller. Geometric categories and o-minimal structures.Duke Mathe- matical Journal, 84(2), 1996

    1996

  57. [57]

    A. W. van der Vaart. New Donsker classes.Annals of Probability, 24(4):2128–2140, 1996

    1996

  58. [58]

    A. W. van der Vaart and J. A. Wellner. Preservation theorems for Glivenko-Cantelli and uniform Glivenko-Cantelli classes. InHigh Dimensional Probability II, pages 115–133. Springer, 2000

    2000

  59. [59]

    A. J. Wilkie. Model completeness results for expansions of the ordered field of real num- bers by restricted Pfaffian functions and the exponential function.Journal of the American Mathematical Society, 9(4):1051–1094, 1996

    1996

  60. [60]

    H. Xu. Uniform exponential convergence of sample average random functions under general sampling with applications in stochastic programming.Journal of Mathematical Analysis and Applications, 368(2):692–710, 2010. 30

    2010