Pith. sign in

REVIEW 3 major objections 5 minor 29 references

This paper establishes a distributionally robust chance-constraint bound: for any true distribution within a relative-entropy distance R of a Gaussian reference, the true violation probability is bounded by a closed-form expression, and it

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 ↗

A relative-entropy ambiguity set with a Donsker–Varadhan bound, plus a quadratic-truncation-based estimate of its radius, yields a distributionally robust risk upper bound for nonlinear covariance steering.

T0 review reviewed 2026-08-01 challenge →

load-bearing objection A genuinely useful ambiguity-radius construction that is honestly labeled as heuristic; the paper's own caveat about g_t means the risk guarantee is not yet a certificate. the 3 major comments →

arxiv 2607.16977 v1 pith:W4PBVAXH submitted 2026-07-18 math.OC cs.ITcs.ROmath.IT

Relative Entropy-Bounded Ambiguous Chance Constraints for Robust Planning in Nonlinear Systems

Trevor N. Wolf , Jay W. McMahon This is my paper
classification math.OC cs.ITcs.ROmath.IT MSC 90C1593E20
keywords distributionally robust chance constraintsrelative entropyKullback-Leibler divergencecovariance steeringnonlinear stochastic controlrisk upper boundspacecraft guidanceDonsker-Varadhan variational formula
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 reading

The paper aims to make chance constraints in stochastic control valid even when the true state distribution is unknown and non-Gaussian. It shows that if the true distribution lies within a relative-entropy distance R of a Gaussian reference, then the true risk of violating a constraint is bounded by a computable expression, and it provides a way to estimate a valid R over time using only the reference covariance and second-order truncation errors of the nonlinear dynamics. If correct, this converts structural approximation error—initial Gaussianity and linearization of the drift—into a distributionally robust risk guarantee, with the usual Gaussian risk recovered as R goes to zero. A spacecraft guidance example demonstrates that the nominal Gaussian risk underestimates the true collision risk, while the proposed bound covers it.

Core claim

The central claim is that for any true distribution πt whose relative entropy to the Gaussian reference π̂t is at most R, the true violation probability satisfies the bound in Eq. (11); combined with Theorem 1, this yields a tight, λ-optimizable upper bound that reduces to the exact Gaussian risk in the zero-divergence limit. The paper then derives a time-varying upper bound on R (Eq. 35) that depends only on the reference covariance Pt and the Hessians of the approximate drift, integrated forward in time. The claim is that this converts the structural approximation errors of Gaussianity and linearization into a distributionally robust risk guarantee, without ever knowing the true distributi

What carries the argument

The Donsker–Varadhan variational formula (Theorem 1), a variational expression for exponential integrals, turns an expectation under an unknown distribution into a supremum over distributions penalized by relative entropy, producing the risk upper bound in Eq. (11). Lemma 1 bounds the relative-entropy growth rate by a quadratic drift mismatch under the true expectation. That expectation is made reference-only via Hoeffding's Lemma and Lemma 2, a closed-form Gaussian expectation of quadratic forms, tr(SP)^2 + 2tr((SP)^2). The resulting rate expression (35) is the engine: it integrates to give a time-varying R using only the reference covariance and Hessians of the drift.

Load-bearing premise

The guarantee rests on treating (1+γ)Eπ̂t[gt] as a genuine upper bound on the truncation error term gt; the paper adopts this as a heuristic and does not prove it.

What would settle it

Run a Monte Carlo truth model for a strongly nonlinear system with large cubic terms, compute the Gaussian-reduced KLD between the truth and the linear-Gaussian reference, and check whether it ever exceeds the Eq. (35) bound for the chosen γ; if it does, the claimed bound on R is false.

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

If this is right

  • Any Gaussian-reference chance constraint can be made robust to distribution ambiguity by adding a relative-entropy term, requiring no extra distributional knowledge.
  • The ambiguity radius becomes a design variable in covariance steering: since the bound in Eq. (35) depends on the reference covariance, optimizing the covariance can actively shrink the ambiguity set.
  • The framework extends standard covariance steering risk assessment to nonlinear systems where Monte Carlo dispersions diverge from the Gaussian reference.
  • In the presented spacecraft scenario, the nominal Gaussian risk underestimates the true risk, while the proposed bound covers both, making it usable as a conservative risk metric in safety-critical guidance.

Where Pith is reading between the lines

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

  • Beyond the paper: the γ heuristic could be replaced by a data-driven upper bound on the truncation term gt from a short Monte Carlo run, turning the method into a formal guarantee without changing the variational core.
  • Beyond the paper: the same variational machinery likely extends to other divergence measures, but the growth inequality in Lemma 1 is specific to relative entropy, so new growth inequalities would be needed.
  • Beyond the paper: co-optimizing λ with the steering policy could reduce conservatism, since the paper optimizes λ post-hoc rather than as part of the optimization.
  • Beyond the paper: the framework only handles structural error; if parameter uncertainty in the drift is included, as the paper plans, the same bound could also cover epistemic model-parameter ambiguity.
Share X Bluesky LinkedIn Reddit HN

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

3 major / 5 minor

Summary. The paper proposes a distributionally robust chance-constraint framework for nonlinear covariance steering. It uses the Donsker–Varadhan variational formula (Theorem 1) to upper-bound the true violation probability by a term depending on the Gaussian reference risk and the relative entropy between the true and reference distributions. The main claimed contribution is a computable time-varying bound on this relative entropy, derived from the reference covariance evolution and second-order truncation errors of the dynamics (Section IV, Eq. (35)). The bound is validated against Monte Carlo simulations in a circular restricted three-body problem spacecraft guidance scenario.

Significance. If the central relative-entropy radius (Eq. (35)) were a valid upper bound, the paper would provide a valuable risk certificate for chance-constrained covariance steering under model-structure ambiguity, without needing the true distribution. The use of the variational upper bound in Theorem 1 is standard and correctly applied, and the zero-divergence recovery of the nominal risk is a desirable consistency property. The authors are also transparent about the heuristic nature of their truncation-error bound. However, the main novelty — the computable ambiguity radius — rests on an acknowledged non-formal step, and the numerical validation uses a Gaussian-reduced surrogate for the true relative entropy. As a result, the paper's central claim is not established as stated.

major comments (3)
  1. [§IV.A, Eqs. (25)–(29) and (35)] The derivation of the relative-entropy growth bound applies Hoeffding's Lemma (Eq. (26)) to the random variable g_t = ||D^{-1/2}(φ_t − f̂_t)||², which is unbounded above for a non-degenerate Gaussian reference and nonlinear drift. The paper itself notes that no finite upper bound exists and that g_t is not sub-Gaussian or sub-exponential. Setting ḡ_t = (1+γ)E_{π̂_t}[g_t] does not repair this: for any non-constant g_t with finite mean, P(g_t > (1+γ)E_{π̂_t}[g_t]) > 0, so ḡ_t is not an almost-sure upper bound and Hoeffding's inequality is not applicable. Consequently Eq. (29) and the integrated bound in Eq. (51) are not certified upper bounds on R(π_t||π̂_t), and the risk bound in Eq. (14) can understate the true violation probability. This is explicitly acknowledged in the text as a heuristic for future work, but it is the load-bearing step of the paper's main contribution.
  2. [§VI.B, Eq. (54)] The numerical validation compares the proposed upper bound to the KLD between a Gaussian reduction of the Monte Carlo ensemble and the Gaussian reference. This is not the true KLD R(π_t||π̂_t); by information processing / Gaussian reduction, the surrogate can be substantially smaller than the true relative entropy. Thus Figures 3 and 4 cannot confirm that Eq. (35) bounds the true relative entropy for all distributions in the ambiguity set, even for the two scenarios shown. The statement that the upper bound is 'valid for all values of γ' is therefore only a statement about the Gaussian-reduced surrogate, not about the actual risk guarantee.
  3. [§IV.A, Eq. (35) and Figs. 3–4] The parameter γ > 1 is free and no procedure is given to choose it so that ḡ_t is a valid upper bound. The plots show γ ∈ {0,1,2} all above the Monte Carlo surrogate, but this does not establish worst-case validity: the probability that g_t exceeds (1+γ)E_{π̂_t}[g_t] remains positive for every finite γ. Without a calibration rule tied to a certified probability, the ambiguity radius is not a distributionally robust guarantee but a heuristic estimate. The paper's stated goal of providing a 'principled approach to determine an appropriate time-varying size of the ambiguity set' is therefore not achieved.
minor comments (5)
  1. [Lemma 1, Eq. (22)] In the second integral, the expression should involve ∇_x ln π̂_t, not ∇_x π̂_t. Similarly, Eq. (23) should read ||D^{1/2}(∇_x ln π_t − ∇_x ln π̂_t)||². This appears to be a typographical error in an otherwise standard contraction estimate.
  2. [Eq. (26)] Hoeffding's Lemma is stated with a typo: the exponential argument should be E_q[e^{λ(X−E_q[X])}], not E_q[e^{λ(X−E_q[X]}], and the bounding interval should be stated explicitly.
  3. [Eq. (27) and surrounding notation] The notation g_t, g_t, and ḡ_t is visually confusing; using e.g. g_*, g^*, or ̲{g}_t and ̄{g}_t would improve readability.
  4. [§VI.B and Figs. 3–4] The text refers to the Gaussian-reduced KLD as the 'true' relative entropy; this is misleading. The figures and text should use a term such as 'Gaussian surrogate KLD' throughout.
  5. [References] Reference [26] appears to be an unpublished preprint without complete bibliographic information; please provide the archive or conference/journal details.

Circularity Check

0 steps flagged

No significant circularity: the risk bound is anchored in external variational results, and the ambiguity-radius computation is self-contained apart from a non-circular heuristic.

full rationale

The derivation is not circular. Theorem 1 (Eq. 10) is the Donsker–Varadhan variational formula, cited to the external text [22] (Dupuis–Ellis), and Eqs. (11)–(14) are a direct application, with the zero-divergence limit (15) providing a consistency check rather than a fitted input. Lemma 1's KLD growth bound (16) is derived from the Fokker–Planck equation and Young's inequality, following the external reference [23], not a self-citation. Equations (25)–(29) and (35) combine Theorem 1 with Hoeffding's Lemma [24] and Lemma 2 [25]; no fitted parameter is renamed as a prediction, and the Monte Carlo KLD surrogate (54) is not used to fit the bound. The one soft spot is the explicit heuristic in Section IV.A: the paper states 'there does not exist a finite upper bound g_t' and that 'the distribution of (δx_t^T H δx_t)^2 is not sub-Gaussian or sub-exponential,' then sets g_t = (1+γ) E_{π̂_t}[g_t]. This makes Eq. (35) not a formally guaranteed upper bound, but it is a correctness/rigor gap, not a circularity: the bound is not defined in terms of the quantity it predicts, and no load-bearing claim reduces to a self-citation. The reference risk is recovered in the zero-divergence limit independently, and the overall framework stands on external mathematical results.

Axiom & Free-Parameter Ledger

3 free parameters · 5 axioms · 0 invented entities

The ledger is light on invented entities: no new particles/forces/dimensions. The main load-bearing choices are (1) γ heuristics, (2) Hoeffding boundedness assumption, (3) same-diffusion Fokker–Planck assumption. All are acknowledged explicitly in the text.

free parameters (3)
  • γ = tested at 0.0, 1.0, 2.0
    Multiplier in the heuristic upper bound gt = (1+γ)E[gt]. No formal calibration; the authors call it 'a simple assumption' and admit future work is needed. The actual bound validity depends on it.
  • λ (per time step)
    Tunable parameter in the variational bound; the paper minimizes over it (λ*) rather than fitting it, so it is not exactly a free parameter in the pejorative sense, but the bound's tightness depends on it.
  • γ sweep in numerics = {0,1,2}
    Used to demonstrate the bound holds across γ values; the choice of the sweep is manual.
axioms (5)
  • standard math Donsker–Varadhan variational formula (Theorem 1, cited to Dupuis–Ellis Proposition 4.5.1).
    Unproved in this paper, taken from cited literature. Standard result, low risk.
  • standard math Hoeffding's Lemma applied to gt
    Used to bound Eπ[gt] − Eπ̂[gt]. Requires gt to be bounded, which is exactly what the heuristic γ is doing. The paper admits no finite bound exists in general.
  • standard math Lemma 2 (expectation of products of quadratic forms in a Gaussian)
    Cited to Magnus & Neudecker; used in Eq. (34).
  • domain assumption True and reference densities satisfy Fokker–Planck with the same diffusion D.
    Needed for Lemma 1. The paper treats the process noise as identical in both, which is consistent with its setup, but it rules out misspecified noise magnitude as a source of ambiguity.
  • domain assumption Truncation error is dominated by the quadratic Taylor remainder and the Gaussian reference covariance Pt is an adequate description of the spread δx.
    Around Eq. (30), the paper replaces ft − f̂t by the Hessian term. This is an approximation assumption, standard in covariance steering but not a theorem.

reviewed 2026-08-01 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Relative Entropy-Bounded Ambiguous Chance Constraints for Robust Planning in Nonlinear Systems." pith.science (2026). https://pith.science/paper/W4PBVAXH

@misc{pith2026260716977,
  author       = {Pith},
  title        = {Pith review of: Relative Entropy-Bounded Ambiguous Chance Constraints for Robust Planning in Nonlinear Systems},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/W4PBVAXH}},
  note         = {Machine review of arXiv:2607.16977}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

We consider defining risk probability in stochastic control problems under distribution ambiguity. Current approaches for chance-constrained control typically assume that the true state distribution is known and Gaussian distributed. These assumptions are not amenable to many real-world engineering applications where system dynamics are nonlinear and only approximately modeled. In this work, we define a distribution ambiguity set and, with a variational expression for exponential integrals, bound the expected risk value under an unknown distribution that resides within a relative entropy distance of a nominal Gaussian reference distribution. Our bound recovers the reference risk value in the zero-divergence limit. A method is presented to determine the relative entropy distance defining the ambiguity set that is a function of the reference covariance evolution and second-order dynamical truncation errors. The resulting contributions provide a framework for handling distributional ambiguity in nonlinear covariance steering problems. A stochastic spacecraft guidance example is presented to demonstrate our contributions.

Figures

Figures reproduced from arXiv: 2607.16977 by Jay W. McMahon, Trevor N. Wolf.

Figure 1
Figure 1. Figure 1: Evaluation of the risk upper bound under distri [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The top panel shows the steered nominal covari [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: The true R(πMC t ∥πˆt) and its upper bound, derived in the present study, as a function of time for the scenario provided. The left- and right-hand sides correspond to σ = 5×10−6 km/s 3/2 and σ = 1×10−6 km/s 3/2 , respectively. σ = 1 × 10−6 km/s 3/2 cases, respectively. At the start of the trial, the Gaussian reference coincides closely with the true distribution, so the true relative entropy value is quit… view at source ↗
Figure 4
Figure 4. Figure 4: Risk probability evaluated under the nominal reference and true distributions, and the risk upper bound [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: λ ∗ as a function of time for the scenario using a diffusion magnitude σ = 5 × 10−6 km/s 3/2 . [9] E. Erdogan and G. Iyengar, “Ambiguous chance constrained problems and robust optimization,” Mathematical Programming, vol. 107, no. 1, pp. 37–61, 2006. [Online]. Available: https://doi.org/10.1007/s10107-005-0678-0 [10] D. Kuhn, S. Shafiee, and W. Wiesemann, “Distribu￾tionally robust optimization,” 2025. [Onl… view at source ↗

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

29 extracted references · 4 canonical work pages

  1. [1]

    Chance constrained programming approach to process optimization under uncer- tainty,

    P. Li, H. Arellano-Garcia, and G. Wozny, “Chance constrained programming approach to process optimization under uncer- tainty,” Computers & Chemical Engineering, vol. 32, no. 1, pp. 25–45, 2008, process Systems Engineering: Contributions on the State-of-the-Art

    2008

  2. [2]

    Stochastic spacecraft trajectory optimization with the consid- eration of chance constraints,

    R. Chai, A. Savvaris, A. Tsourdos, S. Chai, and Y. Xia, “Stochastic spacecraft trajectory optimization with the consid- eration of chance constraints,” IEEE Transactions on Control Systems Technology, vol. 28, no. 4, pp. 1550–1559, 2020

    2020

  3. [3]

    Robust spacecraft guidance around small bodies under uncertainty: Stochastic optimal control approach,

    K. Oguri and J. W. McMahon, “Robust spacecraft guidance around small bodies under uncertainty: Stochastic optimal control approach,” Journal of Guidance, Control, and Dynam- ics, vol. 44, no. 7, pp. 1295–1313, 2021

    2021

  4. [4]

    Covariance control theory,

    A. Hotz and R. E. Skelton, “Covariance control theory,” International Journal of Control, vol. 46, no. 1, pp. 13–32, 1987

    1987

  5. [5]

    Optimal covari- ance control for stochastic systems under chance constraints,

    K. Okamoto, M. Goldshtein, and P. Tsiotras, “Optimal covari- ance control for stochastic systems under chance constraints,” IEEE Control Systems Letters, vol. 2, no. 2, pp. 266–271, 2018

    2018

  6. [6]

    Power laws of topography and gravity spectra of the solar system bodies,

    A. I. Ermakov, R. S. Park, and B. G. Bills, “Power laws of topography and gravity spectra of the solar system bodies,” Journal of Geophysical Research: Planets, vol. 123, no. 8, pp. 2038–2064, 2018

    2038

  7. [7]

    Non-gaussian chance-constrained trajectory control using gaussian mixtures and risk alloca- tion,

    S. Boone and J. McMahon, “Non-gaussian chance-constrained trajectory control using gaussian mixtures and risk alloca- tion,” in 2022 IEEE 61st Conference on Decision and Control (CDC), 2022, pp. 3592–3597

    2022

  8. [8]

    Chance-constrained gaussian mixture steering to a terminal gaussian distribution,

    N. Kumagai and K. Oguri, “Chance-constrained gaussian mixture steering to a terminal gaussian distribution,” 2024. [Online]. A vailable: https://arxiv.org/abs/2403.16302 0 1 2 30.00 0.25 0.50 0.75 1.00 Time [TU] Probability Reference Gaussian MC γ = 0.0 γ = 1.0 γ = 2.0 (a) 0 1 2 30.00 0.25 0.50 0.75 1.00 Time [TU] Probability Reference Gaussian MC γ = 0.0...

    Pith/arXiv arXiv 2024

  9. [9]

    Ambiguous chance constrained problems and robust optimization,

    E. Erdogan and G. Iyengar, “Ambiguous chance constrained problems and robust optimization,” Mathematical Programming, vol. 107, no. 1, pp. 37–61, 2006. [Online]. A vailable: https://doi.org/10.1007/s10107-005-0678-0

    work page doi:10.1007/s10107-005-0678-0 2006

  10. [10]

    Distribu- tionally robust optimization,

    D. Kuhn, S. Shafiee, and W. Wiesemann, “Distribu- tionally robust optimization,” 2025. [Online]. A vailable: https://arxiv.org/abs/2411.02549

    Pith/arXiv arXiv 2025

  11. [11]

    Wasserstein tube mpc with exact uncertainty propagation,

    L. Aolaritei, M. Fochesato, J. Lygeros, and F. Dörfler, “Wasserstein tube mpc with exact uncertainty propagation,”

  12. [12]

    Distributionally robust density control with wasserstein ambiguity sets,

    J. Pilipovsky and P. Tsiotras, “Distributionally robust density control with wasserstein ambiguity sets,” 2024. [Online]. A vailable: https://arxiv.org/abs/2403.12378

    Pith/arXiv arXiv 2024

  13. [13]

    Dembo and O

    A. Dembo and O. Zeitouni, Large Deviations Techniques and Applications, 2nd ed., ser. Stochastic Modelling and Applied Probability. New York: Springer-Verlag, 1998, vol. 38

    1998

  14. [14]

    Kullback-leibler divergence constrained distributionally robust optimization,

    Z. Hu and L. J. Hong, “Kullback-leibler divergence constrained distributionally robust optimization,” A vailable at Optimiza- tion Online, vol. 1, no. 2, p. 9, 2013

    2013

  15. [15]

    Convex approximations of chance constrained programs,

    A. Nemirovski and A. Shapiro, “Convex approximations of chance constrained programs,” SIAM Journal on Optimization, vol. 17, no. 4, pp. 969–996, 2007. [Online]. A vailable: https://doi.org/10.1137/050622328

    doi:10.1137/050622328 2007

  16. [16]

    Satellite orbits: Models, methods, and applications,

    O. Montenbruck, E. Gill, and F. Lutze, “Satellite orbits: Models, methods, and applications,” Applied Mechanics Reviews, vol. 55, no. 2, pp. B27–B28, 04 2002. [Online]. A vailable: https://doi.org/10.1115/1.1451162

    doi:10.1115/1.1451162 2002

  17. [17]

    Chance-constrained covariance steering in a gaussian random field via successive convex programming,

    J. Ridderhof and P. Tsiotras, “Chance-constrained covariance steering in a gaussian random field via successive convex programming,” Journal of Guidance, Control, and Dynamics, vol. 45, no. 4, pp. 599–610, 2022. [Online]. A vailable: https://doi.org/10.2514/1.G005941

    work page doi:10.2514/1.g005941 2022

  18. [18]

    Robust cislunar low- thrust trajectory optimization under uncertainties via sequential covariance steering,

    N. Kumagai and K. Oguri, “Robust cislunar low- thrust trajectory optimization under uncertainties via sequential covariance steering,” Journal of Guidance, Control, and Dynamics, pp. 1–19, 2025. [Online]. A vailable: https://doi.org/10.2514/1.G009092

    work page doi:10.2514/1.g009092 2025

  19. [19]

    Distinguishing and integrating aleatoric and epistemic variation in uncertainty quantification,

    Chowdhary, Kamaljit and Dupuis, Paul, “Distinguishing and integrating aleatoric and epistemic variation in uncertainty quantification,” ESAIM: M2AN, vol. 47, no. 3, pp. 635–662, 2013. [Online]. A vailable: https://doi.org/10.1051/m2an/2012038

    arXiv 2013

  20. [20]

    Robust bounds on risk-sensitive functionals via rényi divergence,

    R. Atar, K. Chowdhary, and P. Dupuis, “Robust bounds on risk-sensitive functionals via rényi divergence,” SIAM/ASA Journal on Uncertainty Quantification, vol. 3, no. 1, pp. 18–33,

  21. [21]

    Quantifying aleatoric and epistemic uncertainty with proper scoring rules,

    P. Hofman, Y. Sale, and E. Hüllermeier, “Quantifying aleatoric and epistemic uncertainty with proper scoring rules,” arXiv preprint arXiv:2404.12215, 2024

    Pith/arXiv arXiv 2024

  22. [22]

    A weak convergence approach to the theory of large deviations,

    P. Dupuis and R. S. Ellis, “A weak convergence approach to the theory of large deviations,” New York, 1997

    1997

  23. [23]

    Improved bounds for discretization of langevin diffusions: Near-optimal rates without convexity,

    W. Mou, N. Flammarion, M. J. Wainwright, and P. L. Bartlett, “Improved bounds for discretization of langevin diffusions: Near-optimal rates without convexity,” Bernoulli, vol. 28, no. 3, pp. 1577–1601, 2022

    2022

  24. [24]

    Probability inequalities for sums of bounded random variables,

    W. Hoeffding, “Probability inequalities for sums of bounded random variables,” Journal of the American Statistical Association, vol. 58, no. 301, pp. 13–30, 1963. [Online]. A vailable: http://www.jstor.org/stable/2282952

    arXiv 1963

  25. [25]

    The Commutation Matrix: Some Properties and Applications,

    J. R. Magnus and H. Neudecker, “The Commutation Matrix: Some Properties and Applications,” The Annals of Statistics, vol. 7, no. 2, pp. 381 – 394, 1979. [Online]. A vailable: https://doi.org/10.1214/aos/1176344621

    arXiv 1979

  26. [26]

    J. A. Rataczak, I. D. Boyd, and J. W. McMahon, Stochastic Aerocapture Trajectory Optimization Under Aerothermody- namic Uncertainty

  27. [27]

    On the properties of kullback-leibler divergence between multivariate gaussian distributions,

    Y. Zhang, W. Liu, Z. Chen, J. Wang, and K. Li, “On the properties of kullback-leibler divergence between multivariate gaussian distributions,” 2023. [Online]. A vailable: https://arxiv.org/abs/2102.05485

    Pith/arXiv arXiv 2023

  28. [2015]

    A vailable: https://doi.org/10.1137/130939730

    [Online]. A vailable: https://doi.org/10.1137/130939730

    work page doi:10.1137/130939730

  29. [2023]

    A vailable: https://arxiv.org/abs/2304.12093

    [Online]. A vailable: https://arxiv.org/abs/2304.12093

    Pith/arXiv arXiv

This paper was first reviewed by deepseek-v4-flash on August 1, 2026.