Pith. sign in

REVIEW 4 major objections 4 minor 1 cited by

Stochastic Production Planning in Manufacturing Systems

T0 review · 4 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read The paper establishes that a stochastic production planning problem over any smooth convex capacity region is solved by a unique convex solution of a linear elliptic equation, from which the optimal production rate follows by…

desk verdict A promising but flawed note: the Cole-Hopf reduction is clean, but the convexity claim is tautological, the verification has a sign error, and the stopping rule actually confines the state to a ball. read the letter →

arxiv 2505.23149 v1 pith:H4OOHJDG submitted 2025-05-29 math.OC

Dragos-Patru Covei This is my paper
classification math.OC MSC 49K2049K3090C3190B30
keywords stochasticproductionplanningHamilton-Jacobi-Bellmanequationoptimalfeedbackcontrolconvexdomainellipticboundaryvalueproblemfunctionmartingaleverificationconcavity
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

This paper tries to show that a stochastic production planning problem—minimizing expected production and holding costs until inventory leaves a smooth convex capacity region—has a complete solution that is structurally simple. The central claim is that the value function can be recovered from a unique positive convex solution of the linear elliptic equation $\Delta u = b(y)u/\sigma^4$ with constant boundary data, and that the optimal production rate is then the gradient $p^*(y) = -\nabla z(y)/2$ of the concave transformed value $z(y) = -2\sigma^2 \ln u(y)$. If true, this turns a nonlinear Hamilton–Jacobi–Bellman problem into standard linear elliptic theory, giving existence, uniqueness, and a transparent feedback rule for manufacturing control. The paper also derives the limiting magnitude of the optimal control near the boundary and verifies the feedback law by a martingale argument. Numerical examples on an interval and on an elliptical domain illustrate the formulas.

What carries the argument

The load-bearing object is the exponential change of variable $u(y)=\exp(-z(y)/(2\sigma^2))$, which converts the nonlinear HJB equation $-\frac{\sigma^2}{2}\Delta z - b = -\frac{1}{4}|\nabla z|^2$ into the linear elliptic equation $\Delta u = \frac{1}{\sigma^4} b(y) u$. This is what makes existence and uniqueness accessible through sub- and supersolution iteration. Convexity of $u$ is then claimed to follow from an iterative scheme on exhausting balls (equation (10)), and concavity of $z$ follows from convexity of $\log u$ checked on one-dimensional slices. The verification argument uses Itô's lemma on the candidate value function to show that the cost process is a supermartingale for every control and a martingale under $p^* = -\nabla z/2$.

What would settle it

Solve equation (10) on the unit disk in $\mathbb{R}^2$ with $b\equiv 1$ and $z_d = 1-|y|^2$, a nonnegative convex source; the unique solution is $w(y) = (4|y|^2 - |y|^4 - 3)/16$, whose radial second derivative $1/2 - 3|y|^2/4$ is negative near $|y|=1$. This is a convex domain and a nonnegative source whose solution is not convex, so the preservation-of-convexity step used in Section 2.2 does not hold as stated.

Watch

Extended reading notes

Core claim

The paper's core claim is that the Hamilton–Jacobi–Bellman equation of stochastic production planning over a general smooth convex domain $\omega$ is exactly equivalent, through the logarithmic substitution $u = \exp(-z/(2\sigma^2))$, to the linear boundary value problem $\Delta u = b(y)u/\sigma^4$ in $\omega$ with $u = \exp(-Z_0/(2\sigma^2))$ on $\partial\omega$. It asserts that this problem has a unique positive solution $u \in C^2(\omega)\cap C^1(\omega)$, that $u$ is convex on $\omega$, and therefore that the value function $z(y) = -2\sigma^2 \ln u(y)$ is concave, so the optimal feedback production rate $p^*(y) = -\nabla z(y)/2$ is the negative half-gradient of a concave cost-to-go. The paper further claims that as the inventory approaches the boundary the magnitude of the optimal control tends to $\sigma^2\gamma \exp(Z_0/(2\sigma^2))$, where $\gamma$ is the outward normal derivative of $u$ at the boundary, and that the policy is verified by showing the associated cost process is a supermartingale for every admissible control and a true martingale under $p^*$.

Load-bearing premise

The load-bearing premise is the unproved step that solving the linear equation at each iteration—a nonnegative convex input on a convex region—always yields a convex output; this is not true for general linear equations of this type, and if it fails the claimed convexity and concavity of the value function are unsupported.

Editorial extensions

If this is right

  • On any smooth convex capacity domain, the nonlinear dynamic programming equation of this production model is exactly a linear elliptic problem, so classical linear elliptic solvers and maximum-principle tools apply directly.
  • The optimal production rate is completely determined by the gradient of the value function, $p^*(y) = -\nabla z(y)/2$, and can therefore be computed by differentiating the numerical solution of the linear PDE.
  • The value function is concave, so the optimal policy has the structural property that production adjustments are monotone along rays from any starting inventory.
  • Near the capacity boundary the optimal control has a known limiting magnitude, giving a checkable asymptotic rule for how production slows as the threshold is approached.
  • The martingale verification provides a certificate: using any other admissible production plan cannot yield lower expected cost than the candidate value function.

Reading between the lines

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

  • If the convexity step is supplied or replaced by a weaker structural condition, the same transformation would extend to other cost functionals, such as $a(p)=|p|^q$ for $q>1$, with different scaling constants and a different resulting PDE.
  • The paper does not discuss that the stopping time (1) halts production when the inventory leaves the largest ball centered at $x_0$ inside $\omega$, not when it reaches $\partial\omega$; for non-circular convex domains these are different boundaries, so the boundary condition on $\partial\omega$ only matches the stated exit rule when $\omega$ is a ball.
  • The concavity of the value function, if established, would justify convex-optimization-based numerical schemes and would guarantee that gradient-descent-like production policies do not get trapped in local minima.
  • A direct computational test of the convexity preservation step would be to solve (10) on a non-radial convex domain with a non-radial $b$ and inspect the Hessian of two successive iterates; the paper's numerical examples use an interval or a circle, where radial symmetry masks the issue.
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

4 major / 4 minor

Summary. The manuscript studies a stochastic production planning problem on a smooth convex domain omega, with inventory dynamics dy = p dt + sigma dw, running cost E[integral_0^tau (|p|^2 + b(y)) dt], and a stopping time defined by distance to the boundary. The author derives the associated HJB equation, applies the logarithmic transformation u = exp(-z/(2 sigma^2)) to obtain the linear PDE Delta u = b(y) u / sigma^4 with constant Dirichlet data, and claims existence, uniqueness, and convexity of u (Theorem 2.1), concavity of the value function z = -2 sigma^2 log u (Theorem 2.2), an asymptotic boundary estimate for the optimal control, and a martingale verification of the feedback law p* = -grad z / 2. Numerical finite-difference experiments in 1D and 2D are presented with Python code.

Significance. The transformation to a linear elliptic equation is standard and the existence/uniqueness of the linear PDE is plausible; the paper also supplies reproducible Python code for the numerical experiments. However, the claimed convexity of u is not proved, Theorem 2.2 merely restates its own hypothesis, the martingale verification leaves a nonzero drift, and the stopping-time definition appears inconsistent with the PDE domain except for balls. As a result, the concavity of the value function and the optimality of the feedback control are not established, and the advertised generalization to arbitrary smooth convex domains is not supported. If these gaps were repaired, the structural results would be potentially useful, but the current analysis leaves the central claims unproved.

major comments (4)
  1. [Section 2.2, Theorem 2.1 (equations (9)-(10))] The convexity proof is invalid. The induction step claims that if z_d is convex and the forcing term is nonnegative, then the solution z_{d+1} of Delta z_{d+1} = sigma^{-4} b(y) z_d on a ball with constant boundary is convex, citing 'standard arguments based on the convexity of the domain and maximum principle.' This is false in general: subharmonic functions need not be convex, and the maximum principle does not control the Hessian. No proof is supplied. Additionally, the patchwork argument omega = union of balls B_{R_d} cannot transfer local convexity of individual iterates on different balls to global convexity of the limit on omega. Thus the assertion in Theorem 2.1 that the solution of (7) is convex is unsupported.
  2. [Theorem 2.2] The theorem is tautological. The hypothesis phi(t) phi''(t) >= (phi'(t))^2 is precisely (log phi)''(t) >= 0, i.e., log-convexity along every one-dimensional slice, and the conclusion is the same property restated. The proof is a one-line computation and adds no information. More importantly, the paper never proves that the solution u of (7) satisfies this condition: convexity alone does not imply log-convexity (e.g., u(x) = x^2 on (0, infinity) is convex but log u is strictly concave). Consequently the concavity of z(y) = -2 sigma^2 log u(y) is not established.
  3. [Section 3.2 (equation (11) and subsequent drift computation)] The martingale verification fails. Using the stated HJB equation sigma^2/2 Delta U - (1/4)|grad U|^2 - b = 0, the drift of M^{p*} is computed in the manuscript as -1/2 |grad U|^2, not zero. The claim that 'any residual discrepancy is resolved provided the candidate U indeed is the solution to the HJB equation' is directly contradicted by the manuscript's own computation: substituting the identity (1/2) sigma^2 Delta U - b = (1/4)|grad U|^2 into the drift expression yields exactly the nonzero residual. Therefore M^{p*} is not a martingale, and the verification argument does not imply J(p*) = U(y(0)) or the optimality of p* = -grad z / 2.
  4. [Section 1.1, stopping time (1) versus PDE domain omega] The stopping time tau = inf{t > 0 : ||y(t) - x0|| > dist(x0, partial omega)} defines first exit from the ball centered at x0 with radius dist(x0, partial omega). This ball is contained in omega and equals omega only when omega is a ball. The PDE (7) is posed on omega with boundary conditions on partial omega, so the boundary condition and the stopping time correspond to different exit boundaries for non-ball domains. This undermines the claimed extension to general smooth convex domains; the model as stated is consistent only for balls.
minor comments (4)
  1. [Theorem 2.1, equation (8)] The displayed limit has a sign error: it writes lim_{y->partial omega} ||-grad z(y)/2|| = sigma^2 gamma exp(-Z0/(2 sigma^2)) = sigma^2 gamma exp(Z0/(2 sigma^2)), which is inconsistent unless Z0 = 0. The preceding derivation gives the second expression; the first exponential should be exp(Z0/(2 sigma^2)).
  2. [Section 2.2] The notation b(y) = max_{||t|| = y} b(t) and b(y) = min_{||t|| = y} b(t) is not well-defined because y is a vector; a radial variable r = ||y|| is presumably intended, and the text should be rewritten accordingly. The symbol BBR_d is also unusual and should be defined more clearly.
  3. [Section 4.1] The problem statement in the text says the inventory holding cost is b(x) = x^2, but the Python code in Section 7.1 sets b_func = lambda x: np.ones_like(x), i.e., a constant function equal to 1. The text and the code should be aligned.
  4. [Abstract] The abstract contains a grammatical error: 'a real-world examples' should be 'real-world examples.'

Circularity Check

2 steps flagged · score 6.0 of 10

Theorem 2.2 assumes log-convexity as its hypothesis, making the main concavity claim tautological, while the convexity of u is imported from self-citations [8,10] and an unproved induction step.

  1. self definitional [Section 2.2, Theorem 2.2 (statement and proof)]
    "Suppose that for every y ∈ ω and every ξ ∈ RN the one-dimensional slice φ(t) = u(y + t ξ), t ∈ I, with I = {t ∈ R : y + t ξ∈ ω}, satisfies the inequality φ(t) φ′′(t) ≥ (φ′(t))^2 for all t ∈ I. Then log u is convex on ω, and hence the value function defined by z(y) = −2σ^2 log u(y) is concave on ω."

    The hypothesis φφ'' ≥ (φ')^2 is exactly (log φ)'' ≥ 0 along every slice, which is the definition of log-convexity. The proof merely rewrites ψ'' = (φφ'' − (φ')^2)/φ^2 and reads off ψ'' ≥ 0. Thus the conclusion 'log u is convex' is the hypothesis restated, not a derived property of the PDE solution. The paper does not prove that the solution u of (7) satisfies this slice condition, and convexity of u alone (even if Theorem 2.1 were valid) does not imply log-convexity; for example u(x)=x^2 is convex while log u is concave. Hence the advertised concavity of z is assumed rather than derived.

  2. self citation load bearing [Section 2.2, proof of Theorem 2.1 (after Eq. (9), before Eq. (10))]
    "Results in the literature (see, e.g., [8, 10]) guarantee that the solutions z and z of the problems ... are convex in B_Rd. ... Since z0 is convex and the forcing term in the above problem is nonnegative, standard arguments based on the convexity of the domain and maximum principle imply that if zd is convex then so is zd+1."

    References [8] (Canepa, Covei, Pirvu) and [10] (Covei) are authored or co-authored by the present author. The convexity of the limiting solution u is a load-bearing premise for Theorem 2.1's claim of a convex value function, and it is imported from these self-references rather than proved here. The subsequent induction step 'if zd is convex then so is zd+1' is likewise asserted as 'standard arguments' without proof. Thus a central structural property of u rests on a self-citation chain rather than on a derivation contained in the paper.

full rationale

The transformation from the HJB equation to the linear PDE and the existence/uniqueness argument via sub- and supersolutions are not circular: they solve Δu = σ^{-4} b(y) u for a given b and constant boundary data. However, the principal qualitative result advertised in the introduction, concavity of the value function z(y) = -2σ² log u(y), is not actually derived from the PDE. Theorem 2.2 states as a hypothesis the very log-convexity condition that would yield concavity, and the paper supplies no argument showing that the PDE solution satisfies φφ'' ≥ (φ')^2. The convexity claim in Theorem 2.1 is also supported by self-citations [8,10] and an unproved 'standard' induction step. Because the linear-PDE existence part has independent content but the central concavity claim reduces to an assumption, the circularity is partial rather than total; I do not count the algebraic sign discrepancy in Section 3 as a circularity, since that is a correctness issue outside the present circularity pass.

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

The existence part relies only on standard elliptic theory. The convexity and concavity claims rest on an unproved and generally false monotonicity of convexity under the solution operator, plus a tautological log-convexity theorem. No free parameters are fitted; sigma, Z0, and b are model inputs.

assumptions (4)
  • standard math The auxiliary problem (9) has a unique positive solution w in C2 intersect C1, cited from Gilbarg and Trudinger.
    Standard linear elliptic theory for the Dirichlet problem with continuous nonnegative coefficient. Invoked in Section 2.2 to construct a subsolution.
  • standard math The strong maximum principle justifies uniqueness, and the Hopf lemma gives partial_n u < 0 at the boundary.
    Used in Section 2.2 for uniqueness and in the asymptotic estimate (8).
  • domain assumption The zero set of b is exactly the closure of an interior connected subdomain omega0, and b is continuous on omega.
    Stated in Section 2.1 as a hypothesis; used to construct sub- and supersolutions. It is a modeling restriction, not derived.
  • ad hoc to paper The exhaustion omega = union of B_{R_d} with boundary values matched on partial B_{R_d} preserves solvability and convexity of the iterates z_d.
    Equation (10) and the following paragraph assert 'standard arguments' that if z_d is convex then z_{d+1} is convex, but no such theorem is given and it is false in general. This is a load-bearing unproved premise.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Stochastic Production Planning in Manufacturing Systems." pith.science (2026). https://pith.science/paper/H4OOHJDG

@misc{pith2026250523149,
  author       = {Pith},
  title        = {Pith review of: Stochastic Production Planning in Manufacturing Systems},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/H4OOHJDG}},
  note         = {Machine review of arXiv:2505.23149}
}
abstract

We extend the stochastic production planning framework to manufacturing systems, where the set of admissible production configurations is described by a general smooth convex domain $\omega $. In our setting, production operations continue as long as the production inventory $y(t)$ remains inside the capacity limits of $\omega $ and are halted once the state exits this region, i.e.,% \begin{equation*} \tau =\inf \{t>0:\Vert y(t)-x_{0}\Vert >\text{dist}(x_{0},\partial \omega )\}. \end{equation*}% The running cost is partitioned into a quadratic production cost $% a(p)=\left\Vert p\right\Vert ^{2}$ and an inventory holding cost modeled by a positive continuous function $b(y)$. We derive the associated Hamilton--Jacobi--Bellman (HJB) equation, verify the supermartingale property of the value function, and characterize the optimal feedback control. Techniques inspired by Lasry, Lions and Alvarez enable us to prove existence and uniqueness within this generalized production planning framework. Numerical experiments and a real-world examples illustrate the practical relevance of our results.

Figures

Figures reproduced from arXiv: 2505.23149 by the authors.

Figure 1
Figure 1. Caption describing the two figures. 14 [PITH_FULL_IMAGE:figures/full_fig_p014_1.png] view at source ↗
Figure 2
Figure 2. Caption describing the two figures [PITH_FULL_IMAGE:figures/full_fig_p015_2.png] view at source ↗
Figure 3
Figure 3. Caption describing the two figures [PITH_FULL_IMAGE:figures/full_fig_p016_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Caption describing the two figures [PITH_FULL_IMAGE:figures/full_fig_p017_4.png]
Figure 5
Figure 5. Figure 5: Caption describing the two figures [PITH_FULL_IMAGE:figures/full_fig_p017_5.png]
Figure 1
Figure 1. Figure 1: fig1 . colorbar ( surf1 , ax = ax1 , shrink =0.5 , aspect =10) [PITH_FULL_IMAGE:figures/full_fig_p027_1.png]
Figure 2
Figure 2. Figure 2: fig2 . colorbar ( surf2 , ax = ax2 , shrink =0.5 , aspect =10) [PITH_FULL_IMAGE:figures/full_fig_p027_2.png]
Figure 3
Figure 3. Figure 3: fig3 . colorbar ( surf3 , ax = ax3 , shrink =0.5 , aspect =10) [PITH_FULL_IMAGE:figures/full_fig_p027_3.png]
Figure 4
Figure 4. Figure 4: fig4 . colorbar ( surf4 , ax = ax4 , shrink =0.5 , aspect =10) [PITH_FULL_IMAGE:figures/full_fig_p027_4.png]

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. An Iterative PDE Based Illumination Restoration Scheme for Image Enhancement

    math.AP 2025-06 reject novelty 4.0 of 10

    A logarithmic-potential PDE iteration is claimed to restore uneven illumination with PSNR 52.90 dB, but the printed implementation and parameters make the update collapse to a no-op.

Reference graph

Works this paper leans on

22 extracted references · 21 canonical work pages · cited by 1 Pith paper

  1. [1]

    O. Alvarez. A quasilinear elliptic equation in rn. Proc. Roy. Soc. Edinburgh Sect. A, 126:911–921, 1996

    work page 1996

  2. [2]

    H. Amann. Fixed point equations and nonlinear eigenvalue problems in ordered banach spaces. SIAM Review, 18(4):620–709, 1976

    work page 1976

  3. [3]

    C. Bayer. Discretization of sdes: Euler methods and beyond, 2006

    work page 2006

  4. [4]

    Bayram, T

    M. Bayram, T. Partal, and G. O. Buyukoz. Numerical methods for simulation of stochastic differential equations. Advances in Continuous and Discrete Models , 2018(17):1–10, 2018. 19

    work page 2018

  5. [5]

    H. J. Brascamp and E. H. Lieb. On extensions of the brunn–minkowski and pr´ ekopa– leindler theorems, including inequalities for log concave functions, and with an application to the diffusion equation. Journal of Functional Analysis , 22:366–389, 1976

    work page 1976

  6. [6]

    Cadenillas, P

    A. Cadenillas, P. Lakner, and M. Pinedo. Optimal control of a mean-reverting inventory. Operations Research, 58(6):1046–1062, 2010

    work page 2010

  7. [7]

    Cadenillas, P

    A. Cadenillas, P. Lakner, and M. Pinedo. Optimal production management when demand depends on the business cycle. Operations Research, 61(4):1046–1062, 2013

    work page 2013

  8. [8]

    Canepa, D.-P

    E.C. Canepa, D.-P. Covei, and T.A. Pirvu. A stochastic production planning prob- lem. Fixed Point Theory , 23(1):179–198, 2022

    work page 2022

Show all 22 references
  1. [9]

    D.-P. Covei. Image restoration via integration of optimal control techniques and the hamilton–jacobi–bellman equation. arXiv:2505.07699v1 [math.AP], may 2025. 12 May 2025

    2025 arXiv

  2. [10]

    D.-P. Covei. Stochastic production planning: Optimal control and analytical in- sights. arXiv:2505.12341v1 [math.OC], may 2025

    2025 arXiv

  3. [11]

    and Tubaro L

    Talay D. and Tubaro L. Expansion of the global error for numerical schemes solving stochastic differential equations. Stochastic Anal. Appl. , 8(4):483–509, 1990

    1990

  4. [12]

    Gharbi and J.P

    A. Gharbi and J.P. Kenne. Optimal production control problem in stochas- tic multiple-product multiple-machine manufacturing systems. IIE Transactions, 35:941–952, 2003

    2003

  5. [13]

    Gilbarg and N

    D. Gilbarg and N. S. Trudinger. Elliptic Partial Differential Equations of Second Order. Springer-Verlag, Berlin, Heidelberg, New York, 1977

    1977

  6. [14]

    Holmes-Cerfon

    M. Holmes-Cerfon. Applied Stochastic Analysis, volume 33. American Mathematical Society, Courant Institute of Mathematical Sciences at New York University, 2024

    2024

  7. [15]

    P. E. Kloeden and E. Platen. Numerical Solutions of Stochastic Differential Equa- tions, volume 23 of Applications of Mathematics . Springer, New York, 1992

    1992

  8. [16]

    Korevaar

    N. Korevaar. Convex solutions of nonlinear elliptic and parabolic boundary value problems. Indiana University Mathematics Journal , 36(3):687–704, 1987

    1987

  9. [17]

    Lasry and P.L

    J.M. Lasry and P.L. Lions. Nonlinear elliptic equations with singular boundary conditions and stochastic control with state constraints. Math. Ann., 283:583–630, 1989

    1989

  10. [18]

    G. M. Lieberman. Asymptotic behavior and uniqueness of blow-up solutions of quasilinear elliptic equations. Journal d’Analyse Math´ ematique, 115:213–249, 2011

    2011

  11. [19]

    G. M. Lieberman. Gradient estimates for singular fully nonlinear elliptic equations. Nonlinear Analysis, 119:382–397, 2015

    2015

  12. [20]

    G. M. Lieberman. Gradient estimates for elliptic oblique derivative problems via the maximum principle. Advances in Differential Equations , 25(11/12):709–754, November/December 2020. 20

    2020

  13. [21]

    Sethi and G.L

    S.P. Sethi and G.L. Thompson. Applied Optimal Control: Applications to Manage- ment Science. Nijho , Boston, 1981

    1981

  14. [22]

    S i m u l a t i o n ␣ halted ␣ at ␣ t ␣ = ␣ { t_vals [ i ]:.2 f } ␣ with ␣ x ␣ = ␣ { x_new :.4 f } ␣ ( out ␣ of ␣ bounds )

    G.L. Thompson, S.P. Sethi, and J. Teng. Strong planning and forecast horizons for a model with simultaneous price and production decisions. European Journal of Operational Research, 16:378–388, 1984. 7 Appendix These Python codes were created with the support of Microsoft Copi...

    1984

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.