Pith. sign in

REVIEW 3 major objections 5 minor 32 references

Batch Sample-wise Stochastic Optimal Control via Stochastic Maximum Principle

T0 review · 3 major / 5 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read This paper proves that a batch sample-wise projection method based on the stochastic maximum principle attains mean-squared error at most $C(1/K+1/N^2)$ for stochastic optimal control, improving the earlier $O(\sqrt{N/K+1/N})$ rate.

desk verdict A plausible batch-extension of sample-wise SMP whose headline rate relies on an unstated moment condition, with a suspicious timing table. read the letter →

arxiv 2505.02688 v2 pith:5U5VRI6O submitted 2025-05-05 math.OC

Hui Sun , Feng Bao This is my paper · ORCID
classification math.OC MSC 60H1060H3065C20
keywords stochasticoptimalcontrolmaximumprinciplebatchgradientdescentbackwarddifferentialequationsconvergenceratestrongconvexitydampedcontractionrandomizedneuralnetworks
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 tries to establish that the sample-wise backpropagation method for solving stochastic optimal control problems can be accelerated by replacing single-sample gradient estimates with batch averages and by simulating the forward SDE with a higher-order scheme. Under strong convexity of the cost functional, it proves that the resulting projection algorithm satisfies $E[\|u^*-u^{K+1}\|^2] \le C(1/K+1/N^2)$ after $K$ gradient iterations on an $N$-point time grid, improving the predecessor's $O(\sqrt{N/K+1/N})$ rate. The practical significance is that the bottleneck in such probabilistic control solvers is neither the optimization step nor the temporal resolution alone, but the variance and discretization error in estimating the adjoint processes $Y$ and $Z$ from the backward SDE; batching and a higher-order forward scheme attack exactly that bottleneck. The paper also proposes a damped contraction algorithm that uses the sufficient condition of the maximum principle, and proves convergence in a linear-quadratic special case at rate $O(\eta^{k+1}+N/M+1/N)$.

What carries the argument

The carrying mechanism is the sample-wise backward recursion $Y_n=Y_{n+1}+h(b_x'Y_{n+1}+f_x')$ with $Z_n=Y_{n+1}\Delta W_n/h$, paired with a batched gradient estimator $(j_u')_n=\frac1M\sum_i \partial_u H(t_n,X_n^i,Y_n^i,Z_n^i,u_n)$ and a higher-order forward integrator $\Psi$ of order $\Theta$. Proposition 4.6 identifies the sample-wise variables as unbiased conditional estimates of the classical BSDE discretization; Lemma 4.7 bounds the temporal error of $(Y,Z)$ by $O(h^2)$ under the condition $2\Theta\ge 3$ and the fourth-moment assumption on the forward scheme; Theorem 4.9 combines strong convexity with a variance estimate for the batched gradient to obtain the contraction recursion; Theorem 4.10 adds the $O(1/N^2)$ error from projecting the true optimum onto piecewise-constant controls.

What would settle it

Run a one-dimensional linear-quadratic control problem with known solution using Algorithm 1 with an Euler forward scheme but batched SGD; if the empirical error versus $N$ follows $1/N$ rather than $1/N^2$, the claimed $N^{-2}$ discretization term is not achievable without the higher-order scheme. Conversely, with the order-2 scheme of equation (7.30), $M=N$, and $K$ a constant multiple of $N^2$, the log-log error slope should be about $-1$; a plateau or a shallower slope would indicate that the variance bound or the fourth-moment assumption in Lemma 4.7 fails.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that the convergence rate of the sample-wise stochastic gradient projection method for deterministic controls can be improved from roughly $O(\sqrt{N/K+1/N})$ to roughly $O(\sqrt{1/K+1/N^2})$ by two coordinated changes: average the sample-wise Hamiltonian gradient over a batch of independent forward-backward paths, and discretize the forward SDE with a scheme of order at least $3/2$. Theorem 4.10 states that, under Lipschitz and linear-growth assumptions together with strong convexity of the loss, the iterates of Algorithm 1 satisfy $E[\|u^*-u^{K+1}\|^2] \le C(1/K+1/N^2)$, where the $K^{-1}$ term is the optimization error and the $N^{-2}$ term is the combined projection and BSDE discretization error. A supporting result, Proposition 4.6, shows that the sample-wise backward recursion produces unbiased conditional estimates of the classical numerical BSDE solution, so the batch average is an honest estimator of the numerical gradient rather than a heuristic smoothing device.

Load-bearing premise

The $N^{-2}$ part of the rate rests on Lemma 4.7's requirement that the forward SDE be simulated with a scheme of order at least $3/2$ and that the unusual condition $E[|X_N-X|^4] \le C E[|X_N-X|^2]$ holds; if only an Euler scheme is used or the moment condition fails, the discretization error reverts from $O(h^2)$ to $O(h)$, and the advertised rate degrades to $O(\sqrt{1/K+1/N})$.

Editorial extensions

If this is right

  • For a fixed error tolerance $\epsilon$, the number of gradient iterations needed drops from $O(\epsilon^{-4})$ to $O(\epsilon^{-2})$, and the temporal grid size drops from $O(\epsilon^{-2})$ to $O(\epsilon^{-1})$.
  • Batch size $M=N$ is sufficient to make the gradient estimator's variance independent of $N$, so the noise from the $Z$-term simulation no longer dominates the convergence rate.
  • The $1/N^2$ temporal term matches the optimal first-order rate for Euler-type approximations of the control, so further improvement requires a higher-order control representation, not just a smaller step size.
  • In the linear-quadratic special case, the damped contraction algorithm converges linearly in the iteration count, with a batch-penalty term $N/M$, giving a derivative-free alternative when the Hamiltonian minimizer is known explicitly.

Reading between the lines

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

  • The advertised $1/N^2$ is conditional on using a forward scheme of order at least $3/2$; if one retains the Euler scheme, the BSDE discretization error reverts to $O(h)$, so the rate degrades to $O(\sqrt{1/K+1/N})$. The paper states the requirement explicitly but leaves the practical cost trade-off implicit.
  • The variance analysis suggests an adaptive batch policy: choose $M$ proportional to $N$ when the diffusion is controlled, but smaller batch sizes may suffice when the diffusion is uncontrolled and the $Z$-term disappears from the gradient; the authors' numerics show faster-than-predicted decay in that case.
  • The randomized-neural-network application points to a derivative-free training loop: with random hidden weights fixed, the SMP gradient is linear in the output weights and can be computed by vector algebra, which would remove backpropagation from the parameter updates if extended to deeper networks.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper studies finite-horizon stochastic optimal control problems through the stochastic maximum principle, with deterministic controls restricted to piecewise-constant functions in time. It proposes two batch-sampling algorithms: a projected batch stochastic gradient descent update for the control, using a higher-order scheme for the forward SDE, and a damped contraction update for the control that minimizes the Hamiltonian. The main theoretical claim (Theorems 4.9--4.10) is that, under strong convexity of the cost functional, the projected batch SGD satisfies E||u* - u^{K+1}||^2 <= C(1/K + 1/N^2), improving the earlier rate O(sqrt(N/K + 1/N)) of Archibald--Bao--Cao--Sun [1]. A secondary result (Theorems 4.11--4.12) gives a contraction-type convergence bound for the damped algorithm under a linear-quadratic structure and an explicit contraction assumption. Numerical experiments for controlled and uncontrolled diffusions, a high-dimensional HJB equation, and a randomized neural network extension illustrate the behavior of the algorithms and their time efficiency relative to the original SGD method.

Significance. If the central rate bound were fully justified, the paper would provide a meaningful incremental advance over the sample-wise backpropagation analysis in [1]: it shows how batch sampling and a higher-order forward scheme turn the iteration complexity N/K into 1/K and the discretization complexity 1/N into 1/N^2 in the squared-error sense. The manuscript is transparent about the special assumptions needed for the damped contraction result, and it ships reproducible code and careful numerical comparisons against the prior SGD method. The high-dimensional HJB example with a randomized neural network is a useful practical demonstration, even though its optimizer study is heuristic rather than covered by the theorems. The main barrier to acceptance is the unsupported moment hypothesis in Lemma 4.7 on which the advertised N^{-2} rate rests.

major comments (3)
  1. [Section 4.1, Lemma 4.7 and Theorems 4.9-4.10] The advertised N^{-2} discretization rate is obtained only under the fourth-moment ratio condition sqrt(E|X^N_{t_n}-X_{t_n}|^4) <= C E|X^N_{t_n}-X_{t_n}|^2, which is stated as an assumption inside Lemma 4.7 and is not part of Assumption 1 or of the hypotheses of Theorems 4.9/4.10. In the proof of Lemma 4.7, this condition is used at the last inequality of (4.30) to control the cross term sqrt(E|\hat Y^N_{t_n}|^4 E|X^N_{t_n}-X_{t_n}|^4) by C h^{2\Theta}. Without it, (4.27) does not follow from the standard L2 strong-order estimate E|X^N-X|^2 <= C h^{2\Theta}, because the ratio (E|Y|^4)^{1/2}/E|Y|^2 can be unbounded when the discretization error has rare large excursions. Consequently, the step epsilon_N^2 <= C h^2 in Lemma 4.8, and hence the 1/N^2 term in Theorem 4.10, is unsupported as the theorems are stated. This is load-bearing: if the condition cannot be proved from Assumption 1, the main theorem degrades to O(1/K + 1/N), which removes the claimed improvement over [1].
  2. [Theorem 4.9, Eq. (4.52)] The batch variance estimate is written as * <= C(1 + N/B). As stated, the right-hand side does not tend to zero as B grows, which is inconsistent with the fact that the batch estimator converges to the conditional expectation; the proof then chooses B = N to obtain * <= C. However, Theorem 4.9 states no assumption linking the batch size M in Algorithm 1 to N, and the proof introduces B without relating it to M. The claimed passage from the N/K term in [1] to 1/K relies on this variance estimate. The theorem statement should be amended to include the needed batch-size assumption, for example M = N, and the estimate in (4.52) should be repaired so that it does not assert a non-vanishing variance for the sample mean.
  3. [Theorem 4.11, Eqs. (4.73)-(4.76)] The contraction recursion is not proved as written. In (4.76), the first line contains a term proportional to ||u^k - u^{N,*}|| on the right-hand side while the left-hand side is a squared norm, and the second line converts this into ||u^k - u^{N,*}||^2 without displaying the Young's inequality step that would justify the conversion. Such a step would introduce additional constants and would affect the condition on rho under which eta < 1 holds. The induction leading to (4.77) is therefore not a valid proof in the present form. This is fixable by rewriting the estimate, but the displayed argument needs to be corrected.
minor comments (5)
  1. [Section 4.1, Eq. (4.29)] In the display after (4.28), the first term on the right should be E[|Delta Y^N_{t_{n+1}}|^2] rather than E[|Y^N_{t_{n+1}}|^2]; otherwise the Gronwall argument does not control the quantity of interest.
  2. [Algorithm 1 and Section 4.1] The batch size is called M in Algorithm 1 and in Theorem 4.9, but Eq. (4.52) uses B for the same quantity. Please use one symbol consistently and state the relation between M and N in the theorem hypotheses.
  3. [Table 2 caption] The caption of Table 2 says 'Example 1 Controlled diffusion', but the table reports results for Example 2, which has uncontrolled diffusion. The caption should be corrected.
  4. [Section 5.3] The HJB numerical experiments compare SGD, Adam, and AdaGrad, but the convergence theorems in Section 4 cover only the SGD update. The text should state explicitly that the optimizer comparison is heuristic and outside the theoretical results, especially since the randomized neural network parametrization is not covered by the strong-convexity framework of Theorem 4.9.
  5. [Section 5.1.2] The text says 'we pick eta ~ O(1-rho) to be a large constant, in this case 0.995', but eta in Theorem 4.11 is defined as C(1-rho)^2 + rho^2. The authors should clarify how the numerical choice of eta corresponds to an admissible value of rho under that definition.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the claimed rate improvement is derived by a standard strong-convexity SGD argument, and the key N^{-2} term rests on an explicit (if unproved) moment condition rather than on a circular reduction.

full rationale

The central claim, Theorem 4.10, is obtained by combining the strong-convexity estimate (4.42), the iteration bound in Theorem 4.9, and the triangle inequality; no fitted constant or target quantity is reinserted as an input. The N^{-2} term is traced to Lemma 4.8's ε_N^2 ≤ C h^2, which is proved from Lemma 4.7. Lemma 4.7 is conditional on an explicit extra assumption, namely sqrt(E|X^N_tn - X_tn|^4) ≤ C E|X^N_tn - X_tn|^2 together with 2Θ ≥ 3; this hypothesis is not derived from Assumption 1, so the advertised 1/N^2 rate may degrade if the condition fails. This is an unsupported premise and a correctness risk, but it is not circular: the bound is not assumed as the conclusion, and the proof explicitly states the condition. The paper also cites prior work by the same authors for the sample-wise scheme and for bounds such as E|Z_n|^2 ≤ C N ([1], Lemma 3.3); those are published, externally checkable results rather than a redefinition of the target convergence, so the self-citations do not make the derivation circular. The damped contraction theorem is honestly conditional: Assumption 3 explicitly postulates the contraction-like inequality (4.5), and the abstract states that convergence is 'attained under some appropriate assumption.' The numerical experiments are consistency checks rather than fits used to produce the theorem. Overall, no step in the derivation reduces by construction to its own input.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The central claim rests on standard stochastic analysis plus two structural assumptions (smoothness, strong convexity) and an ad hoc moment condition in Lemma 4.7. No new physical entities are introduced. The free parameters are algorithm hyperparameters (batch size, damping, learning rate) that are chosen by hand rather than fitted to data.

free parameters (3)
  • Batch size M = M = N for projection, M = N^2 for contraction in experiments
    The advertised rate O(1/K + 1/N^2) requires the gradient variance to be O(1), which is achieved by taking the batch size proportional to N; with one sample the variance is O(N) and the iteration term becomes N/K.
  • Damping parameter rho = eta ~ 0.995 (so rho ~ 0.005) in experiments
    Theorem 4.11 only proves existence of rho such that the contraction factor eta < 1 without giving a constructive formula; experiments pick eta ~ 0.995 by hand.
  • Learning-rate schedule theta and offset M_offset in eta_k = theta/(k+M_offset) = Chosen to satisfy lambda*theta - 4CL*theta^2/(1+M_offset) > 2; HJB example uses constant learning rate 2e-3
    The convergence proof depends on this inequality involving unknown constants lambda and L; the numerical HJB experiment does not use the schedule.
assumptions (5)
  • domain assumption Assumption 1: b, sigma, f, g smooth with bounded derivatives and Lipschitz conditions
    Standard regularity for well-posedness of controlled SDEs and BSDEs, invoked throughout Section 4.
  • domain assumption Assumption 2: strong convexity of J with modulus lambda
    Needed for the gradient descent convergence proof and for converting functional gaps into control error bounds (Section 4.1).
  • domain assumption Assumption 3: linear-quadratic structure and contraction condition (4.5) with mu < 0
    Restricts the damped contraction proof to a special case; the paper notes it is implied by strong convexity when lambda > 1.
  • ad hoc to paper Lemma 4.7 moment condition E[|X_N - X|^4] <= C E[|X_N - X|^2] and scheme order Theta with 2*Theta >= 3
    Unusual condition not derived from Assumption 1; used to control the cross term in the BSDE discretization error.
  • standard math Standard Ito calculus, martingale representation, discrete Gronwall inequality, BSDE discretization errors from [8], [12], [32]
    Background results used without proof.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Batch Sample-wise Stochastic Optimal Control via Stochastic Maximum Principle." pith.science (2026). https://pith.science/paper/5U5VRI6O

@misc{pith2026250502688,
  author       = {Pith},
  title        = {Pith review of: Batch Sample-wise Stochastic Optimal Control via Stochastic Maximum Principle},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5U5VRI6O}},
  note         = {Machine review of arXiv:2505.02688}
}
abstract

In this work, we study the stochastic optimal control problem (SOC) mainly from the probabilistic view point, i.e. via the Stochastic Maximum principle (SMP) \cite{Peng4}. We adopt the sample-wise backpropagation scheme proposed in \cite{Hui1} to solve the SOC problem under the strong convexity assumption. Importantly, in the Stochastic Gradient Descent (SGD) procedure, we use batch samples with higher order scheme in the forward SDE to improve the convergence rate in \cite{Hui1} from $\sim \mathcal{O}(\sqrt{\frac{N}{K} + \frac{1}{N}})$ to $\sim \mathcal{O}(\sqrt{\frac{1}{K} + \frac{1}{N^2}})$ and note that the main source of uncertainty originates from the scheme for the simulation of $Z$ term in the BSDE. In the meantime, we note the SGD procedure uses only the necessary condition of the SMP, while the batch simulation of the approximating solution of BSDEs allows one to obtain a more accurate estimate of the control $u$ that minimizes the Hamiltonian. We then propose a damped contraction algorithm to solve the SOC problem whose proof of convergence for a special case is attained under some appropriate assumption. We then show numerical results to check the first order convergence rate of the projection algorithm and analyze the convergence behavior of the damped contraction algorithm. Lastly, we briefly discuss how to incorporate the proposed scheme in solving practical problems especially when the Randomized Neural Networks are used. We note that in this special case, the error backward propagation can be avoided and parameter update can be achieved via purely algebraic computation (vector algebra) which will potentially improve the efficiency of the whole training procedure. Such idea will require further exploration and we will leave it as our future work.

Figures

Figures reproduced from arXiv: 2505.02688 by the authors.

Figure 1
Figure 1. Top two figures: Demonstration of numerical solutions using the Batch projection algorithm. Bottom two [PITH_FULL_IMAGE:figures/full_fig_p019_1.png] view at source ↗
Figure 2
Figure 2. Top two figures: Demonstration of numerical solutions using the damped contraction algorithm. Bottom two [PITH_FULL_IMAGE:figures/full_fig_p020_2.png] view at source ↗
Figure 3
Figure 3. Top two figures: Demonstration of numerical solutions using the Batch projection algorithm. Bottom two [PITH_FULL_IMAGE:figures/full_fig_p022_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Top two figures: Demonstration of numerical solutions using the damped contraction algorithm. Bottom two [PITH_FULL_IMAGE:figures/full_fig_p023_4.png]
Figure 5
Figure 5. Figure 5: Top figure: Comparison between numerical solutions obtained via different optimizers. Bottom figure: Com [PITH_FULL_IMAGE:figures/full_fig_p025_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

32 extracted references · 27 canonical work pages

  1. [1]

    Archibald, F.Bao, Y.Cao, H.Sun

    R. Archibald, F.Bao, Y.Cao, H.Sun. Numerical analysis for convergence of a sample-wise backpropagation method for training stochastic neural networks. SIAM J. Numer. Anal. 2024, 62, 593–621

    work page 2024

  2. [2]

    Archibald, F

    R. Archibald, F. Bao, Y. Cao, and H. Zhang, A backward sde method for uncertainty quantification in deep learning, Discrete Contin. Dyn. Syst. Ser. S, 15 (2022), pp. 2807–2835

    work page 2022

  3. [3]

    Bensoussan, Lecture on stochastic control, in Nonlinear Filtering and Stochastic Control , Lecture Notes in Math

    A. Bensoussan, Lecture on stochastic control, in Nonlinear Filtering and Stochastic Control , Lecture Notes in Math. 972, Springer-Verlag, Berlin, New York, 1982, pp. 1–62

    work page 1982

  4. [4]

    Carmona and M

    R. Carmona and M. Lauri` ere. Convergence analysis of machine learning algorithms for the numerical solution of mean field control and games I: The ergodic case. SIAM Journal on Numerical Analysis, 59(3):1455–1485, 2021

    work page 2021

  5. [5]

    Carmona and M

    R. Carmona and M. Lauri` ere. Convergence analysis of machine learning algorithms for the numerical solution of mean field control and games: II—the finite horizon case. The Annals of Applied Probability, 32(6): 4065–4105, 2022

    work page 2022

  6. [6]

    Carmona and M

    R. Carmona and M. Lauri` ere. Deep Learning for Mean Field Games and Mean Field Control with Applica- tions to Finance, page 369–392. Cambridge University Press, 2023

    work page 2023

  7. [7]

    R. Carmona. Lectures on BSDEs, Stochastic Control, and Stochastic Differential Games with Financial Applications. SIAM. Feb 2016

    work page 2016

  8. [8]

    Chessari, R

    J. Chessari, R. Kawai, Y. Shinozaki, T. Yamada. Numerical methods for backward stochastic differential equations: A survey . Probability Surveys, Vol. 20 (2023) 486-567. ISSN: 1549-5787. https://doi.org/10. 1214/23-PS18. 29

    work page 2023

Show all 32 references
  1. [9]

    N. Du, J. T. Shi, and W. B. Liu, An effective gradient projection method for stochastic optimal control, Int. J. Numer. Anal. Model., 4 (2013), pp. 757–774

    2013

  2. [10]

    W. E., T. Li., E. Vanden-Eijinden. Applied Stochastic Analysis. Graduate Studies in Mathematics 199. American Mathematical Society, 2019

    2019

  3. [11]

    E., J.Han, and A

    W. E., J.Han, and A. Jentzen. Deep learning-based numerical methods for high dimensional parabolic partial differential equations and backward stochastic differential equations.communications in Mathematics and Statistics, 5(4): 349-3809, 2017

    2017

  4. [12]

    Gobet and C

    E. Gobet and C. Labart. Error expansion for the discretization of backward stochastic differential equations. Stochastic Processes and their Applications, 117(7):803–829, 2007

    2007

  5. [13]

    Gobet, J.-P

    E. Gobet, J.-P. Lemor, and X. Warin. A regression-based Monte Carlo method to solve backward stochastic differential equations. Annals of Applied Probability, 15(3):2172–2202, 2005

    2005

  6. [14]

    B. Gong, W. Liu, T. Tang, W. Zhao, T. Zhou An efficient gradient projection method for stochastic optimal control problem, SIAM J Numer Anal. Vol.55, No. 6 pp 2982-3005

  7. [15]

    Han and S

    Q. Han and S. Ji. A multi-step algorithm for BSDEs based on a predictor-corrector scheme and least-squares Monte Carlo. Methodology and Computing in Applied Probability, 24(4):2403–2426, 2022

    2022

  8. [16]

    F. B. Hanson. Applied Stochastic Processes and Control for Jump-Diffusions: Modeling, Analysis, and Computation, SIAM, Philadelphia, 2007

    2007

  9. [17]

    U. G. Haussmann. Some examples of optimal stochastic controls or: The stochastic maximum principle at work, SIAM Rev., 23 (1981), pp. 292–307, https://doi.org/10.1137/1023062

    1981 doi

  10. [18]

    Han and W

    J. Han and W. E. Deep learning approximation for stochastic control problems . NIPS, 11 2016

    2016

  11. [19]

    Hur´ e, H

    C. Hur´ e, H. Pham, and X. Warin. Deep backward schemes for high-dimensional nonlinear PDEs. Mathe- matics of Computation, 89:1547–1579, 2020

    2020

  12. [20]

    H. J. Kushner. Numerical Methods for Stochastic Control Problems in Continuous Time. SIAM Journal on Control and Optimization. Vol. 28, Iss. 5 (1990)10.1137/0328056

    1990 doi

  13. [21]

    Lemor, E

    J.-P. Lemor, E. Gobet, and X. Warin. Rate of convergence of an empirical regression method for solving generalized backward stochastic differential equations. Bernoulli, 12(5):889–916, 2006

    2006

  14. [22]

    M. Min, R. Hu. Signatured Deep Fictitious Play for Mean Field Games with Common Noise. PMLR. 2021

    2021

  15. [23]

    S. G. Peng, Backward stochastic differential equations and applications to optimal control , Appl. Math. Optim., 27 (1993), pp. 125–144

    1993

  16. [25]

    S. Peng. A General Stochastic Maximum Principle for Optimal Control Problems. SIAM Journal on Control and Optimization. Vol. 28, Iss. 4 (1990), 10.1137/0328054

    1990 doi

  17. [26]

    Pham, Continuous-time Stochastic Control and Optimization with Financial Applications

    H. Pham, Continuous-time Stochastic Control and Optimization with Financial Applications . Springer Nature 2009

    2009

  18. [27]

    Pham and X

    H. Pham and X. Warin. Mean-field neural networks-based algorithms for mckean-vlasov control problems. arXiv:2212.11518, 2022

    2022 arXiv

  19. [28]

    Pham and X

    H. Pham and X. Warin. Actor-critic learning algorithms for mean-field control with moment neural net- works. arXiv:2309.04317, 2023. 30

    2023 arXiv

  20. [29]

    Sun, Meshfree Approximation for Stochastic Optimal Control Problems

    H. Sun, Meshfree Approximation for Stochastic Optimal Control Problems. Communications in Mathemat- ical Research, Volume 37, Issue 3, pp. 387–420

  21. [30]

    H. M. Soner, J. Teichmann, Qinxin Yan. Learning algorithms for mean field optimal control arxiv: https: //arxiv.org/pdf/2503.17869

    arXiv

  22. [31]

    Yong and X

    J. Yong and X. Y. Zhou, Stochastic Controls: Hamiltonian Systems and HJB Equations, Springer, New York, 1999

    1999

  23. [32]

    J. Zhang. Backward Stochastic Differential Equations. From Linear to Fully Nonlinear Theory. Probability Theory and Stochastic Modelling, Volume 86. Springer. ISSN 2199-3130

  24. [33]

    W. Zhao, L. Chen, and S. Peng, A new kind of accurate numerical method for backward stochastic differ- ential equations, SIAM J. Sci. Comput., 28 (2006), pp. 1563–1581, https: //doi.org/10.1137/05063341X. 31

    2006 doi

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.