Pith. sign in

REVIEW 2 major objections 6 minor 51 references

Approximating Rockafellians Mitigate Distributional Perturbations: Discontinuous Integrands and Chance-Constrained Applications

T0 review · 2 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read For stochastic programs with discontinuous integrands, the paper proves that carefully designed approximating Rockafellians epi-converge to the true objective, so that near-minimizers converge to the true solution set even when the…

desk verdict Solid qualitative extension of Rockafellian stability to discontinuous integrands and general Borel measures; the S2 rate proof in Theorem 4.7 has a real but likely repairable gap. read the letter →

arxiv 2507.15801 v1 pith:HSQUTTBI submitted 2025-07-21 math.OC

Lai Tian , Johannes O. Royset This is my paper
classification math.OC MSC 90C1590C3149J5360B10
keywords distributionalperturbationsRockafellianrelaxationepi-convergencechance-constrainedprogrammingdiscontinuousintegrandsprobabilitymetricsmetricsubregularityouterMinkowskicontent
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper claims that the disproportionate instability of stochastic programs under small distributional changes can be corrected by solving an approximating Rockafellian rather than the naïve plug-in problem. A Rockafellian embeds the original minimization problem into a parametric family indexed by a perturbation variable $u$, and the approximating version penalizes $\|u\|^\alpha$ while optionally replacing the integrand by a smoothed mapping $G^\nu$. For essentially arbitrary Borel probability distributions and lower semicontinuous, possibly discontinuous integrands, the authors show that the approximating Rockafellians epi-converge to the true Rockafellian and that their near-minimizers converge to the true solutions. The payoff is a stability theory for chance-constrained programs, where the plug-in problem can fail dramatically under weak convergence; the paper shows that the penalized formulations (4.5) and (4.7) are stable under mild assumptions, and quantifies rates using metric subregularity and finite upper outer-Minkowski content.

What carries the argument

The load-bearing object is the approximating Rockafellian $f^\nu$ in (2.3): the true Rockafellian $f(u,x)=g_0(x)+h(u+\mathbb{E}_\mu[G(\xi,x)])+\iota_{\{0\}}(u)$ is modified by using the perturbed distribution $\mu^\nu$, a possibly regularized integrand $G^\nu$, and the penalty $\frac{1}{\alpha\lambda_\nu}\|u\|^\alpha$ in place of the indicator $\iota_{\{0\}}(u)$. The proof is carried by an epi-approximation theorem (Theorem 2.2) showing that a liminf inequality holds for all convergent sequences and a one-point limsup inequality holds at a true minimizer; Sections 3 and 4 then give principled constructions of $G^\nu$ via epigraphical regularization, $g^\nu_i(\xi,x)=\inf_{\zeta\in\Xi} g_i(\zeta,x)+\frac{1}{\beta\theta_\nu}\|\zeta-\xi\|^\beta$, which makes the $\xi$-variation Lipschitz and thereby lets weak convergence of measures be exploited. For chance-constrained programs, the explicit penalized objectives $\phi^\nu_f$ in (4.5) and (4.7) inherit stability, and Theorem 4.7 relates their near-minimizers to near-minimizers of $\phi$ through metric subregularity of the inverse feasible-set map and upper outer-Minkowski bounds on $H_i(x)$.

What would settle it

Inspect the proof of Theorem 4.7(S2): it takes a $\gamma$-net of $B^{m+n}(0,\rho)\cap(\operatorname{lev}_{\le\rho} f^\nu_\tau)$ lying in $D_\rho$, but $D_\rho$ is only dense in $B^n(0,\rho)$. A concrete test: choose $\mu^\nu\to\mu$ weakly with nonzero bounded-Lipschitz error, sets $H_i(x)$ satisfying (4.8) only on a dense set of $x$, and near-minimizers $(u^\nu,x^\nu)$ whose $u^\nu$-components are not within any $\gamma$ of a point in $D_\rho$; if for every $\gamma>0$ such a net fails to exist and the claimed bound $\eta_\nu$ does not vanish, the S2 rate estimate collapses.

Watch

Extended reading notes

Core claim

The central discovery is that a stochastic program with lower semicontinuous, even discontinuous integrands can be made stable under distributional perturbations by solving an approximating Rockafellian rather than the plug-in problem. For $f(u,x)=g_0(x)+h(u+\mathbb{E}_\mu[G(\xi,x)])+\iota_{\{0\}}(u)$ and its approximations $f^\nu(u,x)=g_0(x)+h(u+\mathbb{E}_{\mu^\nu}[G^\nu(\xi,x)])+\frac{1}{\alpha\lambda_\nu}\|u\|^\alpha$, the authors show that conditions (i)--(iii) of Theorem 2.2 imply epi-convergence $f^\nu\to_e f$ and, when the true solution set is nonempty, convergence of $\varepsilon^\nu$-minimizers of $f^\nu$ to $\{0\}\times\operatorname*{argmin}\phi$ (Theorem 2.3). In the chance-constrained setting, the penalized formulations $\phi^\nu_f$ in (4.5) and (4.7) inherit this stability: with the Pasch--Hausdorff envelope of the constraint indicator as $G^\nu$, weak convergence of $\mu^\nu$ suffices, and under metric subregularity of the feasible-set inverse plus finite upper outer-Minkowski content, Theorem 4.7 gives the stated rates with a vanishing $\gamma$-term in the weak-convergence case.

Load-bearing premise

The rate result for the weak-convergence setting needs the finite-outer-Minkowski-content condition to hold on a set that is dense in the full space containing the near-minimizers' $u$-coordinates, whereas Assumption 4.5 only states it is dense in the $x$-coordinate ball; that missing density is the load-bearing unstated premise.

Editorial extensions

If this is right

  • For chance-constrained programs, minimizing $\phi^\nu_f$ in (4.5) or (4.7) instead of the plug-in objective recovers the true solution set as $\nu\to\infty$ (Corollaries 4.1--4.3).
  • The stability holds without metric regularity or inner semicontinuity of the feasible-set map; the quantitative result uses only metric subregularity (Assumption 4.4) and finite outer-Minkowski content (Assumption 4.5).
  • Weak convergence of distributions is enough when $G^\nu$ is the Pasch--Hausdorff envelope, including the bounded-Lipschitz rate $d_{\mathrm{BL}}(\mu^\nu,\mu)$ (Proposition 3.8).
  • In setting (S1), the solution-set distance decays like $d_{\mathrm{mi}}(\mu^\nu,\mu)^{\alpha/(\alpha+1)}$; in (S2), it reaches any $\gamma$-neighborhood of the true solutions at rate $d_{\mathrm{BL}}(\mu^\nu,\mu)^{\alpha/(2\alpha+2)}$ (Theorem 4.7).
  • Empirical (sample-average) approximations are almost-surely stable even though empirical measures may fail to converge in minimal-information or total-variation metrics (Corollary 4.2).

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A natural testable extension is to choose $\lambda_\nu$ and $\theta_\nu$ as in Proposition 3.8 (e.g., $\lambda_\nu=\theta_\nu=d_{\mathrm{BL}}(\mu^\nu,\mu)^{1/2-\varepsilon}$) and solve $\phi^\nu_f$ on a finite-dimensional chance-constrained instance, checking whether the observed solution error tracks the predicted order.
  • Because the machinery only uses lower semicontinuity and boundedness, the same approximating-Rockafellian construction should apply to mixed-integer stochastic programs, whose integrands are discontinuous for the same reason as chance constraints; the paper notes this as an open practical direction.
  • The $\gamma$-term in the (S2) rate is controlled only by the non-uniformity in Assumption 4.5; verifying the uniform bound (4.10) for a concrete family of sets $H_i(x)$ would decide whether the cleaner $\gamma=0$ rate of Remark 4.8 is available.
  • A stationary-point analogue would likely require extra assumptions, as the paper states; one could probe whether epi-convergence still holds when the near-minimizers are replaced by stationary points of $f^\nu$ in simple one-dimensional chance constraints.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 6 minor

Summary. The paper studies parametric stability of stochastic programs of composite form (1.1), allowing discontinuous random lsc integrands and general Borel probability distributions. It replaces the exact Rockafellian (2.1) with approximating Rockafellians (2.3) and shows, under three verifiable conditions, that these approximations epi-converge to the exact Rockafellian and that their near-minimizers converge to the true solution set (Theorems 2.2 and 2.3). The paper then gives principled constructions of the approximating integrands G^nu, including epigraphical regularization, and parameter selections for weak convergence, setwise convergence, bounded Lipschitz/Fortet-Mourier/Wasserstein/minimal-information/total-variation discrepancies, and empirical measures. The final section specializes to chance-constrained programs, deriving penalized formulations (4.5) and (4.7), qualitative convergence results (Corollaries 4.1-4.3), and a quantitative rate result (Theorem 4.7) under metric subregularity and finite upper outer-Minkowski content. The central claim is that solving these approximating Rockafellians stabilizes the underlying stochastic program as the distributional perturbation vanishes, even for discontinuous integrands and without metric regularity assumptions.

Significance. If the quantitative results are repaired, this is a genuinely useful contribution. The paper removes metric regularity assumptions that are standard in the qualitative stability literature, handles discontinuous integrands that arise in chance constraints, and provides explicit finite-dimensional correctors, notably the construction u^nu = E_mu[G^nu(xi,x0)] - E_mu^nu[G^nu(xi,x0)] in Theorem 2.2(b), which is correct and non-circular. The systematic treatment of probability metrics and the explicit rate comparison with Henrion-Roemisch are valuable. The qualitative convergence arguments are carefully presented and mostly standard. However, the proof of the rate result in setting (S2) of Theorem 4.7 contains a genuine gap involving the use of the dense set D_rho, and this is a load-bearing point because the (S2) rate is one of the paper's headline claims. The gap appears repairable by a fixed finite net argument, but the current manuscript does not contain that argument.

major comments (2)
  1. [Section 4.2, proof of Theorem 4.7, setting (S2)] The proof chooses a gamma-net N_gamma subset of B^{m+n}(0,rho) cap (lev_{<=rho} f^nu_tau) cap D_rho, but D_rho from Assumption 4.5 is only a dense subset of B^n(0,rho), not of R^{m+n}. Even if D_rho is embedded as {0} x D_rho, density in the x-coordinate does not give a gamma-net for points with nonzero u-coordinate. This net is then used to obtain the uniform bound mu((H_i(hat x)+B^d(0,theta_nu))\H_i(hat x)) <= C_mc theta_nu, which drives the eta_2 bound and hence the claimed rate in (S2). As written, the rate result in setting (S2) is not established. The gap is repairable: since lambda_nu -> 0, for large nu every point in the level set satisfies ||u||_2 <= (2 alpha lambda_nu rho)^{1/alpha} < gamma/2, and a fixed finite gamma/2-net of S_tau cap B^n(0,rho) drawn from D_rho covers the x-coordinates; the points {(0,y)} then form a gamma-net of the product level set. This argument should be stated explicitly.
  2. [Section 4.2, proof of Theorem 4.7, setting (S2)] Even if the gamma-net is allowed to depend on nu, the constant C_mc = 0.001 + max_{y in N_gamma} max_i limsup_eps (1/eps) mu((H_i(y)+B^d(0,eps))\H_i(y)) would depend on nu through N_gamma. Assumption 4.5 only gives finiteness pointwise for each x in D_rho, so the maximum over a moving finite net does not yield a uniform bound independent of nu. The subsequent absorption of C_mc into a single constant C_2 is therefore not justified as written. The fixed finite net repair from the previous comment also resolves this issue, because it makes C_mc depend only on gamma; without that repair, the proof needs an additional uniformity condition beyond Assumption 4.5.
minor comments (6)
  1. [Section 1, introduction] The sentence 'motivating our development has has the indicator function' contains a duplicated word 'has'; one occurrence should be deleted.
  2. [Section 3.1, Proposition 3.1] In the proof, 'an extended Fautou's lemma' should read 'an extended Fatou's lemma'.
  3. [Section 1, Terminology] The definition of iota_C and 1_C is written as a single run-on sentence: 'we define iota_C(x) := 0, 1_C(x) := 1 if x in C; otherwise iota_C(x) := infinity, 1_C(x) := 0.' The two cases should be separated more clearly, for example with a semicolon and an explicit 'if x in C' and 'if x not in C'.
  4. [Proof of Proposition 4.6, case (ii-c)] The phrase 'which is proportion to the upper (d-1)-dimensional Minkowski content' should read 'which is proportional to the upper (d-1)-dimensional Minkowski content'.
  5. [Remark 4.8] Remark 4.8 states that a 'slight (omitted) modification of the proof' yields gamma = 0 under the uniform condition (4.10). Since this is presented as a result rather than as a conjecture, the omitted argument should be supplied or the statement should be explicitly labeled as an unproved extension. This is especially relevant because the proof of the gamma > 0 case already requires the fixed-net repair described in the major comments.
  6. [Proof of Theorem 4.7, setting (S2)] In the displayed definition of C_mc, the ball should be written as B^d(0,varepsilon) to match the notation in Assumption 4.5; the current text uses B(0,varepsilon) without specifying the dimension.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the convergence mechanism is a constructive corrector, and the flagged (S2) issue is a proof gap, not a circular step.

full rationale

The central derivation chain is not circular. The approximating Rockafellians f^ν in (2.3) are defined with an explicit penalty on the corrector u, and Theorem 2.2 establishes epi-convergence from hypotheses (i)-(iii) by constructing u^ν = E_μ[G^ν(ξ,x0)] − E_{μ^ν}[G^ν(ξ,x0)]; this corrector is not assumed to equal the target limit but is shown to vanish under condition (iii). The transfer from epi-convergence to near-minimizer convergence in Theorem 2.3 uses standard epi-convergence arguments, citing [43, Theorem 5.5] as a proof template rather than as an assumption of the paper's conclusion. The G^ν constructions in Section 3 are parameter-free regularizations (Pasch-Hausdorff and Moreau partial envelopes) with explicit Lipschitz and boundedness estimates, so the resulting rates in Theorem 4.7 are derived bounds, not fitted quantities relabeled as predictions. The authors' prior work [42, 2, 8] is used for context and comparison, not to import the main theorem, and the 'resembles the classic penalty method' remark in Section 4.1 explicitly distinguishes the stabilization motivation from the computational penalty-method literature. The only load-bearing concern I found is a proof gap, not circularity: in the (S2) part of Theorem 4.7, the proof selects a γ-net N_γ ⊆ B^{m+n}(0,ρ) ∩ (lev_{≤ρ} f^ν_τ) ∩ D_ρ, although D_ρ from Assumption 4.5 is defined only as a dense subset of B^n(0,ρ), so the net is not guaranteed to exist for points with nonzero u-coordinate. This is a completeness gap in the quantitative claim, and it is consistent with the paper's own Remark 4.8 tying the γ term to nonuniformity in Assumption 4.5, but it does not make the derivation circular.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

No new physical or computational entities are introduced. The paper's qualitative results depend only on standard variational analysis background and the stated structural hypotheses; the quantitative rate theorem additionally assumes metric subregularity and outer Minkowski content, which are explicit domain assumptions. The sequences lambda_nu, theta_nu and exponents alpha, beta are user-chosen design parameters, not fitted constants.

free parameters (4)
  • lambda_nu = to 0, rates specified per corollary
    Penalty parameter in the approximating Rockafellian (2.3); its rate relative to distribution discrepancies dmi, dBL, dTV and to nu (empirical) drives convergence. It is a design parameter, not fitted to data.
  • theta_nu = to 0, rates specified per corollary
    Regularization width in the Pasch-Hausdorff/Moreau partial envelopes (3.4); its rate relative to dBL or dFM controls the error in E_munu[Gnu] - E_mu[Gnu].
  • alpha = >= 1, user-chosen
    Exponent in the penalty (1/(alpha lambda_nu)) ||u||^alpha; affects rate exponents in Theorem 4.7.
  • beta = >= 1, user-chosen
    Order of the epigraphical regularization (3.4) and of the Fortet-Mourier metric; user-chosen, not fitted.
assumptions (5)
  • standard math Epigraphical convergence and variational analysis machinery (Rockafellar-Wets [34]; Royset-Wets [43, Theorems 5.5, 6.56, Prop. 6.58])
    Invoked throughout Sections 2-4 to justify lsc properties, epi-convergence, and truncated Hausdorff distance estimates.
  • standard math Extended Fatou lemmas for weakly and setwise convergent measures ([12, Theorem 3.4], [13, Theorem 4.1])
    Used in Proposition 3.1(a) and Proposition 3.6(a) to pass liminf through expectations under changing measures.
  • domain assumption Metric subregularity of M^{-1} at feasible points (Assumption 4.4)
    Central to the rate bounds in Theorem 4.7; it bounds the distance to the true feasible set by the constraint violation.
  • domain assumption Finite upper outer-Minkowski content of H_i(x) with respect to mu on a dense set D_rho (Assumption 4.5)
    Used in the (S2) argument of Theorem 4.7 to control the error between Gnu and G; as discussed in the weakest_assumption field, its application in the proof has a gap.
  • domain assumption Random lsc integrands, local uniform boundedness, and the structural hypotheses on g0, h, Hi used throughout (Sections 2 and 4)
    The main modeling assumptions that make expectations lsc and the constructions well-defined.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Approximating Rockafellians Mitigate Distributional Perturbations: Discontinuous Integrands and Chance-Constrained Applications." pith.science (2026). https://pith.science/paper/HSQUTTBI

@misc{pith2026250715801,
  author       = {Pith},
  title        = {Pith review of: Approximating Rockafellians Mitigate Distributional Perturbations: Discontinuous Integrands and Chance-Constrained Applications},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HSQUTTBI}},
  note         = {Machine review of arXiv:2507.15801}
}
read the original abstract

In this paper, we show how approximating Rockafellians serve as a principled and effective alternative for improving the stability of stochastic programs under distributional changes. Unlike previous efforts that focus on special distributions and continuous integrands, our results accommodate general probability distributions and discontinuous integrands. Thus, our results apply to chance-constrained programs, for which we obtain improved qualitative and quantitative stability results under weaker assumptions pertaining to metric subregularity and upper outer-Minkowski content.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

51 extracted references · 44 canonical work pages

  1. [19]

    Henrion and W

    R. Henrion and W. R¨ omisch. Metric regularity and quantitative stability in stochastic programs with probabilistic constraints.Mathematical Programming, 84:55–88, 1999

    work page 1999

  2. [1]

    Ambrosio, A

    L. Ambrosio, A. Colesanti, and E. Villa. Outer Minkowski content for some classes of closed sets. Mathematische Annalen, 342(4):727–748, 2008

    work page 2008

  3. [2]

    Antil, S

    H. Antil, S. P. Carney, H. D ´ ıaz, and J. O. Royset. Rockafellian relaxation for PDE-constrained optimization with distributional uncertainty.Preprint arXiv:2405.00176, 2024

    arXiv 2024

  4. [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

    work page 1995

  5. [4]

    Attouch and R

    H. Attouch and R. J-B Wets. Quantitative stability of variational systems: I. The epigraphical distance.Transactions of the American Mathematical Society, 328(2):695–729, 1991. 30

    work page 1991

  6. [5]

    Blanchet, P

    J. Blanchet, P. Cui, J. Li, and J. Liu. Stability evaluation through distributional perturbation analysis. InProceedings of the 41st International Conference on Machine Learning, pages 4140– 4159, 2024

    work page 2024

  7. [6]

    Blanchet, J

    J. Blanchet, J. Li, S. Lin, and X. Zhang. Distributionally robust optimization and robust statistics. Preprint arXiv:2401.14655, 2024

    arXiv 2024

  8. [7]

    W. Chen, M. Sim, J. Sun, and C.-P. Teo. From CVaR to uncertainty set: Implications in joint chance-constrained optimization.Operations Research, 58(2):470–485, 2010

    work page 2010

Show all 51 references
  1. [8]

    Deride and J

    J. Deride and J. O. Royset. Approximations of Rockafellians, Lagrangians, and dual functions. SIAM J. Optimization, to appear, 2025

    2025

  2. [9]

    R. M. Dudley.Real Analysis and Probability. Cambridge Studies in Advanced Mathematics. Cambridge University Press, 2 edition, 2002

    2002

  3. [10]

    Durrett.Probability: Theory and Examples, volume 49

    R. Durrett.Probability: Theory and Examples, volume 49. Cambridge University Press, 2019

    2019

  4. [11]

    Federer.Geometric Measure Theory

    H. Federer.Geometric Measure Theory. Springer, 1996

    1996

  5. [12]

    E. A. Feinberg, P. O. Kasyanov, and J. O. Royset. Epi-convergence of expectation functions under varying measures and integrands.Journal of Convex Analysis, 30(3):917–936, 2023

    2023

  6. [13]

    E. A. Feinberg, P. O. Kasyanov, and N. V. Zadoianchuk. Fatou’s lemma for weakly converging probabilities.Theory of Probability and Its Applications, 58(4):683–689, 2014

    2014

  7. [14]

    Fournier and A

    N. Fournier and A. Guillin. On the rate of convergence in Wasserstein distance of the empirical measure.Probability Theory and Related Fields, 162(3):707–738, 2015

    2015

  8. [15]

    A. L. Gibbs and F. E. Su. On choosing and bounding probability metrics.International Statistical Review, 70(3):419–435, 2002

    2002

  9. [16]

    S. Guo, H. Xu, and L. Zhang. Convergence analysis for mathematical programs with distribution- ally robust chance constraint.SIAM Journal on Optimization, 27(2):784–816, 2017

    2017

  10. [17]

    G. A. Hanasusanto, V. Roitch, D. Kuhn, and W. Wiesemann. A distributionally robust perspective on uncertainty quantification and chance constrained programming.Mathematical Programming, 151(1):35–62, 2015

    2015

  11. [18]

    G. A. Hanasusanto, V. Roitch, D. Kuhn, and W. Wiesemann. Ambiguous joint chance constraints under mean and dispersion information.Operations Research, 65(3):751–767, 2017

    2017

  12. [20]

    Henrion and W

    R. Henrion and W. R¨ omisch. H¨ older and Lipschitz stability of solution sets in programs with probabilistic constraints.Mathematical Programming, 100:589–611, 2004. 31

    2004

  13. [21]

    Jiang and W

    N. Jiang and W. Xie. Distributionally favorable optimization: A framework for data-driven decision-making with endogenous outliers.SIAM Journal on Optimization, 34(1):419–458, 2024

    2024

  14. [22]

    P. Kall. On approximations and stability in stochastic programmin. InParametric Optimization and Related Topics, pages 387–407, 1987

    1987

  15. [23]

    D. Klatte. A note on quantitative stability results in nonlinear optimization. InProceedings of the

  16. [24]

    Jahrestagung Mathematische Optimierung, Seminarbericht, volume 90, pages 77–86, 1987

    1987

  17. [25]

    Li and Y

    H. Li and Y. Cui. A decomposition algorithm for two-stage stochastic programs with nonconvex recourse functions.SIAM Journal on Optimization, 34(1):306–335, 2024

    2024

  18. [26]

    Mattila.Geometry of Sets and Measures in Euclidean Spaces: Fractals and Rectifiability, vol- ume 44

    P. Mattila.Geometry of Sets and Measures in Euclidean Spaces: Fractals and Rectifiability, vol- ume 44. Cambridge University Press, 1999

    1999

  19. [27]

    V. A. Nguyen, S. Shafieezadeh Abadeh, M.-C. Yue, D. Kuhn, and W. Wiesemann. Optimistic distributionally robust optimization for nonparametric likelihood approximation.Advances in Neural Information Processing Systems, 32, 2019

    2019

  20. [28]

    S. T. Rachev.Probability Metrics and the Stability of Stochastic Models. John Wiley & Sons, 1991

    1991

  21. [29]

    S. T. Rachev and W. R¨ omisch. Quantitative stability in stochastic programming: The method of probability metrics.Mathematics of Operations Research, 27(4):792–818, 2002

    2002

  22. [30]

    S. M. Robinson. Local epi-continuity and local optimization.Mathematical Programming, 37(2):208–222, 1987

    1987

  23. [31]

    S. M. Robinson and R. J-B Wets. Stability in two-stage stochastic programming.SIAM Journal on Control and Optimization, 25(6):1409–1416, 1987

    1987

  24. [32]

    R. T. Rockafellar.Convex Functions and Dual Extremum Problems. PhD thesis, Harvard Univer- sity, 1963

    1963

  25. [33]

    R. T. Rockafellar.Convex Analysis. Princeton University Press, 1970

    1970

  26. [34]

    R. T. Rockafellar.Conjugate Duality and Optimization. SIAM, 1974

    1974

  27. [35]

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

    2009

  28. [36]

    R¨ omisch

    W. R¨ omisch. Stability of stochastic programming problems.Handbooks in Operations Research and Management Science, 10:483–554, 2003

    2003

  29. [37]

    R¨ omisch and R

    W. R¨ omisch and R. Schultz. Distribution sensitivity in stochastic programming.Mathematical Programming, 50:197–226, 1991. 32

    1991

  30. [38]

    R¨ omisch and R

    W. R¨ omisch and R. Schultz. Stability analysis for stochastic programs.Annals of Operations Research, 30:241–266, 1991

    1991

  31. [39]

    H. L. Royden.Real Analysis 2ed. Macmillan, 1968

    1968

  32. [40]

    J. O. Royset. Approximations and solution estimates in optimization.Mathematical Programming, 170(2):479–506, 2018

    2018

  33. [41]

    J. O. Royset. Good and bad optimization models: Insights from Rockafellians. In J. G. Carlsson, editor,Tutorials in Operations Research: Emerging Optimization Methods and Modeling Tech- niques with Applications, pages 131–160. INFORMS, Cantonsville, MD, 2021

    2021

  34. [42]

    J. O. Royset. Risk-adaptive approaches to stochastic optimization: A survey.SIAM Review, 67(1):3–70, 2025

    2025

  35. [43]

    J. O. Royset, L. L. Chen, and E. Eckstrand. Rockafellian relaxation and stochastic optimization under perturbations.Mathematics of Operations Research, to appear, 2024

    2024

  36. [44]

    J. O. Royset and R. J-B Wets.An Optimization Primer. Springer, 2021

    2021

  37. [45]

    Schneider.Convex Bodies: The Brunn–Minkowski Theory, volume 151

    R. Schneider.Convex Bodies: The Brunn–Minkowski Theory, volume 151. Cambridge University Press, 2013

    2013

  38. [46]

    R. Schultz. Some aspects of stability in stochastic programming.Annals of Operations Research, 100:55–84, 2000

    2000

  39. [47]

    Shapiro, D

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

    2021

  40. [48]

    A. W. van der Vaart and J. A. Wellner.Weak Convergence and Empirical Processes. Springer, 2nd printing 2000 edition, 1996

    2000

  41. [49]

    van Ackooij

    W. van Ackooij. A discussion of probability functions and constraints from a variational perspec- tive.Set-Valued and Variational Analysis, 28(4):585–609, 2020

    2020

  42. [50]

    van Ackooij, P

    W. van Ackooij, P. P´ erez-Aros, C. Soto, and E. Vilches. Inner Moreau envelope of nonsmooth conic chance-constrained optimization problems.Mathematics of Operations Research, 49(3):1419–1451, 2024

    2024

  43. [51]

    Villani.Optimal Transport: Old and New, volume 338

    C. Villani.Optimal Transport: Old and New, volume 338. Springer, 2009. 33

    2009

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.