Pith. sign in

REVIEW 4 major objections 5 minor 27 references

Critical demand in a stochastic model of flows in supply networks

T0 review · 4 major / 5 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 Critical demand in supply networks depends on topology only when buffer stocks exist.

desk verdict A clearly specified supply-network flow model with an exact no-stock result and a plausible but condition-dependent message: stocks make topology matter only if capacities keep fluctuating. read the letter →

arxiv 2505.24813 v1 pith:DH2ZACPQ submitted 2025-05-30 physics.soc-ph cond-mat.dis-nncond-mat.stat-mech

Yannick Feld , Marc Barthelemy This is my paper
classification physics.soc-phcond-mat.dis-nncond-mat.stat-mech
keywords supplynetworksstochasticmodelcriticaldemandbufferstockstreetopologyproductioncapacitymemoryeffectsresilience
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 asks at what external demand rate $r$ a supply network can no longer satisfy customers, the critical demand $r^*$. It shows that when firms hold no buffer stocks, $r^*$ depends only on the number of firms: $r^* = 1/(N+1)$ for any tree-like network, so topology is irrelevant. When firms can store surplus in stocks, memory effects appear and topology becomes decisive: broad, short supply chains support higher $r^*$ than long, narrow ones. The practical claim is that restructuring supply chains toward local connectivity and short chains improves resilience.

What carries the argument

The engine is the per-time-step redrawing of each firm's maximal production capacity $m_i(t)$ from $U[0,1]$ together with the recursive production rule $I_i(t) = \min_{l \in C_i}[m_i(t), D_i(t), a_{il}(t)+k_{il}(t)]$, which makes the root's delivery the bottleneck minimum along the chain. Stock evolves as $k_{il}(t+1) = \min[s, a_{il}(t)+k_{il}(t)-I_i(t)]$, adding a memory term that couples consecutive time steps and lets topology alter the statistics of the bottleneck.

What would settle it

A simulation comparing two identical trees—one where capacities are redrawn each step (annealed) and one where capacities are drawn once (quenched)—with the same stock capacity $s>0$ would settle the claim: the paper predicts different $r^*$ values, while observing the same critical rate in both cases would falsify the memory and topology effect.

Watch

Extended reading notes

Core claim

The central discovery is that the critical demand rate $r^*$ in a stochastic supply network is exactly $1/(N+1)$ for all tree-like networks with no stocks, because the root's production is the minimum of $N$ iid uniform capacities. With positive stock capacity $s$, $r^*$ increases and follows a power-law fit in $N$ whose exponent depends on $s$; simulations for chains, open and closed chains, regular trees, and random trees show that topology matters only when stocks provide memory. The paper argues that nodes with multiple parents reduce $r^*$ because production must be split, while keeping nodes close to the root (broad and short trees) raises $r^*$.

Load-bearing premise

The claim that stocks make topology matter rests on the assumption that every firm's production capacity is drawn anew, independently, each time step; if capacities are fixed once, stocks become irrelevant and $r^*$ returns to $1/(N+1)$.

Editorial extensions

If this is right

  • Without buffer stocks, adding any firm to a tree lowers $r^*$ by exactly $1/(N(N+1))$, regardless of where the firm sits in the network.
  • With stocks, the exponent $\beta$ in the fit $r^* \sim (N+1)^{-\beta}$ changes with $s$, so buffer capacity changes how quickly long chains degrade.
  • For large stock capacity $s$, $r^*$ approaches $1/2$, the mean per-firm capacity, as surplus production is stored and waste vanishes.
  • Firms with multiple parents split their output and lower the collective critical rate, so keeping each firm on a single upstream path is beneficial.
  • Comparing random trees to regular trees, trees that avoid long linear segments perform better, which supports the 'broad and short' design guideline.

Reading between the lines

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

  • The paper's annealed-versus-quenched contrast implies a testable design rule: among networks with equal size, the one that minimizes height and maximizes branching near the root should maximize $r^*$; a direct simulation comparing such networks would isolate this effect.
  • The result suggests that real supply chains with fluctuating capacities (breakdowns, maintenance, demand spikes) benefit most from buffer stocks, while steady producers gain little from inventories — an implicit consequence not fully drawn out by the authors.
  • The exact stationary distribution derived for the second-to-last node in a chain could be extended to longer chains, potentially yielding $r^*$ predictions without numerical fitting.
  • The finding that $\chi$ peaks near $s \approx 0.315$ for trees suggests an optimal intermediate inventory level, which could be probed empirically in warehouse data.
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 / 5 minor

Summary. The paper proposes a stochastic model of material flow in directed acyclic supply networks, where each firm has an iid uniform production capacity redrawn at every time step, demands propagate according to unsatisfied orders, and buffer stocks of capacity s smooth fluctuations. The main quantity is the critical root demand r*, defined as the demand rate above which unsatisfied demand grows linearly. For s=0, the authors derive analytically that r*=1/(N+1) for any network in which each node has at most one parent, verifying the result numerically. For s>0, they derive an exact two-node result and the stock distribution near the leaf of a chain, but rely mostly on simulation to study chains, open chains, regular trees, and random trees. The central qualitative conclusion is that without stocks the topology is irrelevant, whereas with stocks memory effects make topology crucial, leading to the practical suggestion that firms should prefer broad, short supply chains over long, narrow ones. The quantitative support for the s>0 claims is based on three-parameter fits (r* = α/(N+1)^β + c) and data-collapse plots built from the same fitted parameters.

Significance. The paper has real strengths: the no-stock result is proved exactly and verified cleanly in Supp. Fig. S2, and the analytic treatment of the two-node chain with stocks (including the exact expression r* = 7/24 + tan(1/√2)/(4√2)) is a nontrivial and welcome contribution. The model is precisely specified, with pseudo-code that would allow reproducibility. If the central qualitative claim survives additional robustness checks, the paper would be a useful stylized-model contribution to the physics-inspired supply-chain literature. However, the main applied conclusion is contingent on a very specific annealed-capacity assumption, and the numerical exponents and asymptotic constants are not yet supported by error bars, out-of-sample validation, or a test of intermediate capacity persistence. The current evidence is suggestive rather than conclusive for the paper's policy-level takeaway.

major comments (4)
  1. [Main text, p.2; Supp. Mat., Sec. III.A and Algorithm 3] The central result that buffer stocks make topology relevant rests entirely on the annealed assumption that each production capacity m_i(t) is redrawn iid from U[0,1] every time step. The authors' own Supplementary Material (Sec. III.A) states that when the m_i are quenched (drawn once), 'the stocks kij become irrelevant' and the average critical demand returns to r*=1/(N+1) for any s. This is a strong sensitivity: the buffering effect that produces topology dependence disappears as soon as capacities are constant. Since real production capacities persist over timescales comparable to or longer than one time step, the practical design recommendation ('broad, short supply chains') needs a robustness test with intermediate persistence—for example, redrawing each m_i every T time steps, or using autocorrelated capacities—together with a statement of how the topology dependence and the exponent β decay as T increases. Without such a test, the external-validity claim is not supported.
  2. [Eq. (12), Fig. 3; Supp. Fig. S5] The 'data collapse' shown in Fig. 3 is a self-consistency check rather than an independent validation of the fitting function. The parameters α, β, c are obtained by fitting Eq. (12) to the very same r*(z,h) data that are then transformed into the collapse variable (α/(r*−c))^{1/β}−1. By construction, any fit with small residuals will produce points close to y=x, so the collapse does not provide new evidence that the functional form is correct. The same concern applies to the open-chain collapse in Supp. Fig. S5. I recommend reporting bootstrap or least-squares error bars on α, β, c, performing an out-of-sample test (e.g., fitting on half of the h range and testing on the other half), and comparing Eq. (12) with alternative forms, such as a pure power law without the constant or a stretched exponential, to assess whether the three parameters are actually identifiable.
  3. [Eq. (11), Fig. 2, Supp. Fig. S9] The quantitative conclusions about the s-dependence of β and c are based on three-parameter power-law fits (r* = α/(N+1)^β + c) performed on data without error bars. In Supp. Fig. S9, the power-law fit for c(s) yields different exponents depending on whether the fit is weighted logarithmically (c(s)=0.049 s^{0.71}, R²=0.974) or linearly (c(s)=0.057 s^{0.79}, R²>0.999). This sensitivity means that the statement that 'stocks not only modify the prefactor but also alter the exponent' and the recovery of the s→0 limit need uncertainty quantification. At minimum, the paper should provide error bars on r*, on the fitted parameters, and on the extrapolated c for N→∞, using a bootstrap or similar procedure.
  4. [Abstract; Supp. Mat., Sec. III.A] The abstract and the introductory summary claim that without stocks 'minimal production dictates behavior, making topology irrelevant.' This is too broad. The exact result Eq. (9) and the derivation preceding it are derived under the assumption that each node has at most one parent. The Supplementary Material explicitly notes that when any node has more than one parent, the production must be split and the topology affects r* even at s=0 (and even in the quenched case). Please qualify the claim as applying to tree-like networks (each node with a single parent), or extend the analysis to the multi-parent case in the no-stock regime.
minor comments (5)
  1. [Main text, p.4] The sentence 'In the rest of the paper, we will focus on ere and in the fo' appears to be garbled; please fix the missing words.
  2. [Eq. (7)] The symbol C_i is used both for the set of children in Eq. (4) and for the set of descendants in Eq. (7); this is confusing and should be disambiguated.
  3. [Fig. 4] The 'regular' tree used as the baseline is not explicitly defined for N=63; please state which regular tree is used (presumably the binary tree of height 6). In addition, comparing random trees with z∈{2,3} to a binary regular tree is a comparison of different mean degrees, so the conclusion about 'randomness' is confounded with connectivity; either control for the expected number of children or rephrase the conclusion as being about average branching.
  4. [Supp. Mat., Sec. IV] In Eqs. (S15)-(S22), the shifted variable is written with a hat, but several subsequent lines use an unhatted k for the shifted variable; please make the notation consistent.
  5. [Supp. Mat., Sec. VI and Fig. S7] The collapse exponent β=0.962 and the large-s power-law fit (χ∝s^{-1}) are based on fits to a single z0 curve (z0=2) with no uncertainty; please add error bars and state whether the exponent is the same for all z0.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the main analytical and numerical results are self-contained and do not reduce to their inputs by construction.

full rationale

The paper's derivation chain is self-contained. The no-stock result r* = 1/(N+1) follows directly from the iid uniform capacities and the minimum statistics in Eq. (7)-(9), with no fitted parameter or self-citation. The non-zero-stock analytics in the Supplementary Material derive the stock distribution and the N=2 critical demand from the model's master equation, and the numerics are checked against those independent expressions. The topology comparisons and the tree-versus-line production gains are direct simulation measurements, not predictions forced by a fitted ansatz. The data-collapse plots in Figs. 3 and S5 use parameters fitted to the same data and are therefore self-consistency checks rather than out-of-sample predictions, but the paper does not present them as independent predictions, and they are not load-bearing for the central conclusion. The quenched-capacity remark in Supp. III.A ('the stocks kij become irrelevant') is an explicit modeling limitation about external validity, not a circular step. There are no load-bearing self-citations, no imported uniqueness theorems, and no renaming of known results as new organization. Any concern about the annealed redrawing assumption is a correctness or robustness concern, not an internal circularity.

Assumptions & free parameters 4 free parameters · 8 assumptions · 0 invented entities

The model is built on several strong simplifications: acyclic topology, iid annealed capacities, uniform storage limits, and a proportional splitting rule. The central with-stock results require the annealed capacity assumption, since the quenched version makes stocks irrelevant. The scaling claims additionally depend on fitted power-law parameters rather than on a derivation from the model equations.

free parameters (4)
  • α, β, c in Eq. (11) for single chain, per s = s-dependent; e.g., β(s) shown in Fig. 2 inset, α and c in Fig. S9; for s→0: α≈1, β≈1, c≈0
    Three-parameter power-law fit to numerically measured r* vs N; the claim that stocks alter the exponent rests on these fitted parameters.
  • α, β, c in Eq. (12) for regular trees, per z = z-dependent; values in Fig. S10
    Fitted to r* vs height h; used for data collapse in Fig. 3.
  • β=0.962 in collapse Eq. (S35) for χ = 0.962
    Fitted to collapse χ/(z0-1)^β vs s in Supp. Mat. Fig. S7.
  • c(s) power-law fit = 0.049 s^0.71 (log weights) or 0.057 s^0.79 (linear weights)
    Fit to the asymptote c(s) from Eq. (11), used to claim c→0 as s→0 in Supp. Mat. Fig. S9.
assumptions (8)
  • domain assumption Network is a directed acyclic graph; no node is its own ancestor.
    Justified by empirical supply chain studies (Supp. Mat. Section I); excludes recycling and product recovery.
  • ad hoc to paper Production capacities m_i(t) are iid uniform on [0,1] and redrawn each time step.
    Central driver of fluctuations; the quenched version eliminates stock effects (Supp. Mat. Section III.A).
  • domain assumption All firms have the same maximum storage capacity s for each input product.
    Simplification; appears in Eq. (6).
  • domain assumption Demand of a non-root node is the sum of orders from its parents (Eq. 3).
    Defines demand propagation.
  • ad hoc to paper A firm splits its production among its parents in proportion to their share of its demand (Eq. 5).
    Allocation rule; motivates multi-parent inefficiency.
  • domain assumption At large times, unsatisfied demand grows as u(t)/t = max(0, r - r*).
    Measurement principle in Supp. Mat. Eq. (S1); no proof given.
  • domain assumption In the analytics for the 1D chain, demand is assumed large enough not to limit production (Eq. S9).
    Allows dropping the demand term from the minimum in Eq. (4).
  • ad hoc to paper Stationary stock distribution has two delta peaks at boundaries (Eq. S16).
    Ansatz in the master-equation solution, confirmed numerically in Fig. S3.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Critical demand in a stochastic model of flows in supply networks." pith.science (2026). https://pith.science/paper/DH2ZACPQ

@misc{pith2026250524813,
  author       = {Pith},
  title        = {Pith review of: Critical demand in a stochastic model of flows in supply networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DH2ZACPQ}},
  note         = {Machine review of arXiv:2505.24813}
}
read the original abstract

Supply networks are essential for modern production, yet their critical properties remain understudied. We present a stochastic model with random production capacities to analyze material flow to a root node, focusing on topology and buffer stocks. The critical demand, where unsatisfied demand diverges, is examined mostly through numerical simulations. Without stocks, minimal production dictates behavior, making topology irrelevant. With stocks, memory effects arise, making topology crucial. Increased local connectivity is beneficial: firms should favor broad, short supply chains over long, narrow ones.

Figures

Figures reproduced from arXiv: 2505.24813 by the authors.

Figure 1
Figure 1. FIG. 1. Schematic depiction of the dynamics. In a) we see a [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2. Numerical results for the critical root demand [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 4
Figure 4. FIG. 4. On the main plot, we show the numerically obtained [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

27 extracted references · 27 canonical work pages

  1. [1]

    Nagurney, M

    A. Nagurney, M. Yu, A. H. Masoumi, and L. S. Nagurney, Networks Against Time , 1st ed. (Springer New York, NY, 2013)

    work page 2013

  2. [2]

    Nagurney, Labor and Supply Chain Networks , 1st ed

    A. Nagurney, Labor and Supply Chain Networks , 1st ed. (Springer Cham, 2023)

    work page 2023

  3. [3]

    Nagurney, D

    A. Nagurney, D. Hassani, O. Nivievskyi, and P. Mar- tyshev, Multicommodity international agricultural trade network equilibrium: Competition for limited production and transportation capacity under disaster scenarios with implications for food security, European Journal of Op- erational Research 314, 1127 (2024)

    work page 2024

  4. [4]

    Bartezzaghi, R

    E. Bartezzaghi, R. Cagliano, F. Caniato, and S. Ronchi, eds., A Journey through Manufacturing and Supply Chain Strategy Research (Springer Cham, 2016)

    work page 2016

  5. [5]

    Katsaliaki, P

    K. Katsaliaki, P. Galetsi, and S. Kumar, Supply chain disruptions and resilience: a major review and future research agenda, Annals of Operations Research 319, 965 (2022)

    work page 2022

  6. [6]

    J. Dong, D. Zhang, and A. Nagurney, A supply chain network equilibrium model with random demands, Euro- pean Journal of Operational Research 156, 194 (2004), eURO Excellence in Practice Award 2001

    work page 2004

  7. [7]

    Critical demand in a stochastic model of flows in supply networks

    to major crises [8]. A key question is whether the sys- tem can absorb disruptions or if they trigger cascading failures [9, 10]. Complex, real-world models [11, 12] often obscure underlying resilience principles, and Moran et al. [13] introduced recently a stochastic model to study delay propagation in supply networks with dynamic topol- ogy, revealing c...

    work page Pith review arXiv 2025

  8. [8]

    Colon and M

    C. Colon and M. Ghil, Economic networks: Heterogeneity-induced vulnerability and loss of syn- chronization, Chaos: An Interdisciplinary Journal of Nonlinear Science 27, 126703 (2017)

    work page 2017

Show all 27 references
  1. [9]

    Caraiani, A

    P. Caraiani, A. M. Dima, C. P˘ aun, T. Stamule, and M. V. Vargas, Production networks and resilience: How dense production networks shield economies in financial crisis, PLOS ONE 19, 1 (2024)

    2024

  2. [10]

    Battiston, D

    S. Battiston, D. Delli Gatti, M. Gallegati, B. Greenwald, and J. E. Stiglitz, Credit chains and bankruptcy propa- gation in production networks, Journal of Economic Dy- namics and Control 31, 2061 (2007), tenth Workshop on Economic Heterogeneous Interacting Agents

    2007

  3. [11]

    E. J. Hearnshaw and M. M. Wilson, A complex network approach to supply chain network theory, International Journal of Operations & Production Management 33, 442 (2013)

    2013

  4. [12]

    Ramanathan and R

    U. Ramanathan and R. Ramanathan, eds., Sustainable Supply Chains: Strategies, Issues, and Models , 1st ed. (Springer Cham, 2020)

    2020

  5. [13]

    Sawik, Stochastic Programming in Supply Chain Risk Management (Springer Cham, 2024)

    T. Sawik, Stochastic Programming in Supply Chain Risk Management (Springer Cham, 2024)

    2024

  6. [14]

    Moran, M

    J. Moran, M. Romeijnders, P. L. Doussal, F. P. Pi- jpers, U. Weitzel, D. Panja, and J.-P. Bouchaud, Time- liness criticality in complex systems, Nature Physics 10.1038/s41567-024-02525-w (2024)

    2024 doi

  7. [15]

    M. J. Meixell and V. B. Gargeya, Global supply chain design: A literature review and critique, Transportation Research Part E: Logistics and Transportation Review 41, 531 (2005)

    2005

  8. [16]

    Wiedmer and S

    R. Wiedmer and S. E. Griffis, Structural characteristics of complex supply chain networks, Journal of Business Logistics 42, 264 (2021)

    2021

  9. [18]

    S. S. Perera, M. G. H. Bell, M. Piraveenan, D. Kasthuri- rathna, and M. Parhi, Topological structure of manufac- turing industry supply chain networks, Complexity 2018, 3924361 (2018)

    2018

  10. [19]

    Fleischmann, H

    M. Fleischmann, H. R. Krikke, R. Dekker, and S. D. P. Flapper, A characterisation of logistics networks for prod- uct recovery, Omega 28, 653 (2000)

    2000

  11. [20]

    Francas and S

    D. Francas and S. Minner, Manufacturing network con- figuration in supply chains with product recovery, Omega 37, 757 (2009), role of Flexibility in Supply Chain Design and Modeling

    2009

  12. [21]

    A. K. Hartmann, Big Practical Guide to Computer Sim- ulations, 2nd ed. (World Scientific, 2015). Supplementary Material for: Critical demand in a stochastic model of flows in supply networks Yannick Feld 1 and Marc Barthelemy 1,2 1Universit´ e Paris-Saclay, CNRS, CEA, Institut d...

    2015

  13. [22]

    8 0 0 . 2 0 . 4 0 . 6 0 . 8 1 ρk N − 2(k) k numeric f (k) analytic deltas analytical f (k) numeric deltas 0

  14. [23]

    8 0 0 . 2 0 . 4 0 . 6 0 . 8 1 FIG. S3. Here we show the probability density function ρk N −2(k) over k to compare our analytical result from Eq. (S29) to numeric measurements. The numeric probability density was ca lculated by creating a histogram of 1000 bins and then simulat...

  15. [24]

    1 1 10 100 ˜χ s z0 = 2 z0 = 3 z0 = 4 z0 = 5 fit FIG

    01 0 . 1 1 10 100 ˜χ s z0 = 2 z0 = 3 z0 = 4 z0 = 5 fit FIG. S7. We show ˜χ over s for different values of z0. The inset displays the same data in logarithmic scale. We use d β = 0.962 to create the collapse with the help of Eq. (S35). We also fitted a power law to the data for z0...

  16. [25]

    M. J. Meixell and V. B. Gargeya, Global supply chain design : A literature review and critique, Transportation Research Part E: Logistics and Transportation Review 41, 531 (2005)

    2005

  17. [26]

    Wiedmer and S

    R. Wiedmer and S. E. Griffis, Structural characteristics of comp lex supply chain networks, Journal of Business Logistics 42, 264 (2021)

    2021

  18. [27]

    T. Kito, A. Brintrup, S. New, and F. Reed-Tsochas, The struct ure of the toyota supply network: An empirical analysis, Sa ¨ ıd Business School WP 2014-3 10.2139/ssrn.2412512 (2014)

    2014 doi

  19. [28]

    S. S. Perera, M. G. H. Bell, M. Piraveenan, D. Kasthurirathna, an d M. Parhi, Topological structure of manufacturing industry supply chain networks, Complexity 2018, 3924361 (2018)

    2018

Pith tools

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