A gradient model for the Bernstein polynomial basis

G. S. Nahum

REVIEW 2 major objections 4 minor 12 references

A gradient model for the Bernstein polynomial basis

T0 review · 2 major / 4 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read By introducing the Bernstein model, a symmetric exclusion process whose jump rates average to the Bernstein polynomial $B_{n,L}(\rho)$, this paper proves the gradient property for these rates and closes the gap of attainable diffusivities…

desk verdict The Bernstein model is a genuinely new construction, but the central gradient identity in Proposition 2.7 fails on a small example and Lemma 3.4 gives the wrong diffusivity, so the paper's main theorem is false as stated. read the letter →

arxiv 2411.15954 v2 pith:DDWDZKY7 submitted 2024-11-24 math.PR math-phmath.MPnlin.CG

G. S. Nahum This is my paper · ORCID

classification math.PRmath-phmath.MPnlin.CG MSC 60K3582C22

keywords gradientmodelBernsteinpolynomialbasisexclusionprocesskineticallyconstrainedlatticegasporousmediadiffusioncoefficientbinomialtransformhydrodynamiclimit

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 constructs a symmetric exclusion process, the Bernstein model, whose exchange rates are local box densities, and proves that it satisfies the gradient property: the microscopic current is the discrete gradient of an explicitly given potential. The consequence is a diffusion coefficient $D(\rho)=B_{n,L}(\rho)$, a Bernstein basis polynomial, which the paper argues cannot be obtained by superposing Porous Media Models. A companion reduced Porous Media Model attains $D(\rho)=\rho^\ell$ on the same interaction range and is tied to the Bernstein model by a binomial inversion formula. If the promised hydrodynamic limit goes through, these are the first kinetically constrained lattice gases with these diffusivities, and they give a microscopic underpinning for nonlinear PDEs with such polynomial diffusivities.

What carries the argument

The load-bearing object is the gradient identity for the Bernstein constraint. The proof converts the average over boxes $W_j^L$ into a sum over subsets $P$ of the enlarged box $\{0,\ldots,L+1\}$, rewrites the difference of occupation products as a discrete gradient using $1-\tau^{-j}=\sum_{i=1}^j \nabla\tau^{-i}$, and telescopes the boundary terms into an indicator that the density in $\{0,\ldots,L\}$ is at least $(n+1)/(L+1)$. The same combinatorial bookkeeping, together with binomial inversion, carries the gradient property to the reduced Porous Media Model and connects the two families of constraints.

What would settle it

Enumerate all particle configurations in a box of length $L+2$ for a small case such as $n=1,L=2$ and check the identity $b_{n,L}(\eta)(e_{0,1}(\eta)-e_{1,0}(\eta))=-\nabla H_{n,L}(\eta)$ with $H$ as defined; any violation refutes Proposition 2.7. Separately, for the reduced Porous Media Model, compute $\bar H'(\rho)$ from the stated $h_{\ell;L}$ and compare with $\rho^\ell$; the two must agree if the gradient potential integrates to the claimed diffusivity.

Watch

Extended reading notes

Core claim

The central claim is Proposition 2.7: for any $n\le L$, the Bernstein constraint $b_{n,L}(\eta)=\frac{1}{L+1}\sum_{j=0}^L \mathbf{1}\{\langle\eta\rangle_{W_j^L}=\frac{n}{L}\}$ satisfies $b_{n,L}(\eta)(e_{0,1}(\eta)-e_{1,0}(\eta))=-\nabla H_{n,L}(\eta)$, where $H_{n,L}=h_{n,L}+g_{n,L}$, $h_{n,L}(\eta)=\frac{1}{L+1}\mathbf{1}\{\langle\eta\rangle_L\ge\frac{n+1}{L+1}\}$, and $g_{n,L}$ is a sum of shifted antisymmetric terms. Since the constraint averages to $B_{n,L}(\rho)$ under the invariant Bernoulli measure, the model is a gradient model with diffusion coefficient $B_{n,L}(\rho)$, and it coincides with the Porous Media Model when $n=L$. The paper further identifies the reduced Porous Media Model with constraint $p_{\ell;L}$ and diffusivity $\rho^\ell$, and proves the binomial inversion formulas linking $b_{n,L}$ and $p_{\ell;L}$.

Load-bearing premise

The construction's physical conclusion depends on the standard gradient-model identification $D(\rho)=\int c\,d\nu_\rho=\bar H'(\rho)$, whose hydrodynamic-limit proof is deferred to a companion paper.

Editorial extensions

If this is right

If the companion hydrodynamic limit is completed, the Bernstein model yields the macroscopic equation $\partial_t\rho=\partial_u^2 \bar H(\rho)$ with diffusion coefficient $D(\rho)=B_{n,L}(\rho)$.
Superposition over $n$ gives a gradient model for every polynomial with nonnegative Bernstein coefficients, and since $\sum_n B_{n,L}=1$, the generators add to the SSEP generator (partition of unity).
The reduced Porous Media Model is gradient for each $0\le\ell\le L$ and has $D(\rho)=\rho^\ell$ while keeping interaction range $2L+2$, offering a fixed-range alternative to the PMM at each diffusivity.
The inversion formulas show the Bernstein constraints are a binomial transform of reduced-PMM constraints, so a target polynomial diffusivity can be translated into local rates by coefficient comparison.
The mobile-cluster structure (boxes of length $L+2$ with exactly $n+1$ particles) is preserved, so the dynamics remains non-cooperative and has the same type of blocked configurations.

Reading between the lines

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

One implicit direction: because Bernstein polynomials are dense in continuous functions on $[0,1]$, a construction of this kind could, in the large-$L$ limit, approximate arbitrary nonnegative diffusivities by superposing Bernstein models with gradient property.
The paper's root-multiplicity obstruction is specific to the monomial/PMN basis; in the Bernstein basis, nonnegativity of coefficients is automatic for nonnegative polynomials, so the Bernstein model may be the natural canonical gradient model for polynomial diffusivities.
A testable extension: replace boxes of fixed length $L$ with boxes of length growing with the system size; the gradient identity may still hold, but the hydrodynamic limit would then involve spatially dependent or nonlocal diffusivities.
The claimed obstruction to long-range jumps suggests the gradient property is sensitive to the locality of the constraint, so one could test whether convolutions of local Bernstein rates with a kernel of finite width preserve the gradient structure.

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.

Desk Editor's Note

read the letter

Quick take: the model construction is new and worth knowing about, but the main theorems as written do not hold. I checked Proposition 2.7 for L=1, n=0. With η(0)=1 and all other relevant sites empty, the left side b(e0,1-e1,0) is 1, while -∇H is -1/2. So the identity fails. This is not a sign typo in the proof; the proof's expansion carries the same problem. Lemma 3.4 also looks off: the stated h gives derivative ρ^ℓ(1-ρ)^{L-ℓ} under the Bernoulli measure, not ρ^ℓ. Since the g term has zero mean by symmetry, the full H̄' cannot equal the claimed diffusivity. So the reduced PMM gradient identity is also unproven.

The good part: the Bernstein model B(n,L) is a legitimately new dynamics, and the binomial inversion relating it to the reduced PMM is clean and correct (Proposition 2.6 and Lemma 3.1). The elementary properties in Propositions 2.8 and 2.9—partition of unity, symmetry, interpolation, mobile clusters—are clearly argued and appear sound. The paper also honestly states that the hydrodynamic limit is deferred to a companion paper, so the macroscopic equation is not at issue here.

The soft spots are load-bearing: the gradient property is the paper's main contribution, and it is contradicted by an explicit computation. The errors are specific and might be repairable—perhaps the correct H is different, or the intended range is n≥1—but as written the claims are false. A referee could help sort this out, because the construction itself is promising and fills a real gap in attainable diffusivities.

This paper is for people working on gradient exclusion processes and hydrodynamic limits. I would not cite it in its current form, but I would not dismiss the underlying idea. Recommendation: send it to a serious referee, with the expectation of major revision; the author needs to fix the gradient identity for all n and the reduced PMM potential.

Referee Report

2 major / 4 minor

Summary. The paper introduces two families of symmetric exclusion processes on the one-dimensional torus: the Bernstein model B(n,L), whose exchange rate at a node is the fraction of length-(L+2) windows containing exactly n+1 particles that contain the node, and the reduced porous media model PMM_L(ℓ), whose rates are built from hypergeometric probabilities. The central claims are (i) the gradient property for B(n,L) with an explicit potential H_{n,L}=h_{n,L}+g_{n,L} (Proposition 2.7), (ii) an analogous gradient representation for PMM_L(ℓ) (Lemma 3.4), yielding diffusivities B_{n,L}(ρ) and ρ^ℓ respectively, (iii) binomial inversion formulas relating the two families, and (iv) structural properties such as partition of unity, mobile clusters, and blocked configurations. The hydrodynamic limit is explicitly deferred to a companion paper.

Significance. The construction is well motivated and, if correct, would fill a genuine gap: the known superposition of PMMs cannot produce Bernstein-basis diffusivities with two high-multiplicity roots, and the paper proposes a natural combinatorial mechanism. The manuscript is self-contained and contains no fitted parameters; the inversion formulas (2.5)-(2.6) are clean and potentially useful. However, the main algebraic theorem (Proposition 2.7) is false as stated: the proposed H fails even for L=1,n=0. Since the gradient property is the paper's core contribution and the companion hydrodynamic paper would rely on it, the result cannot be accepted without a substantially corrected derivation and a correct potential.

major comments (2)

[Section 2.2, Proposition 2.7, Eq. (2.8); Section 3.2, Eq. (3.4)] Proposition 2.7 cannot be correct as stated. Take L=1, n=0, a torus with N≥5, and a configuration η with η(0)=1 and all other sites in {-1,0,1,2} empty. From (2.3), b_{0,1}(η)=1/2(1{η(-1)=0}+1{η(2)=0})=1. Since e_{0,1}(η)=1 and e_{1,0}(η)=0, the left-hand side b_{0,1}(η)(e_{0,1}(η)-e_{1,0}(η)) equals 1. On the other hand, Eq. (2.8) gives h_{0,1}(η)=1/2 and g_{0,1}(η)=-1/2, so H_{0,1}(η)=0; for τη one gets h_{0,1}(τη)=0 and g_{0,1}(τη)=1/2, so H_{0,1}(τη)=1/2. Hence -∇H = H(η)-H(τη) = -1/2, not 1. The same example shows that the expansion (3.4) carries the opposite sign: for b(e_{1,0}-e_{0,1}) the left-hand side is -1, while the first bracket equals 1/2 and (1/2)∇g equals 1/2, so the right-hand side of (3.4) is +1. This is a load-bearing error, not a typo.
[Section 3.2, Lemma 3.4] The stated potential in Lemma 3.4 is incompatible with the claimed diffusion coefficient. With h_{ℓ;L}(η)=(1/((L+1) binom(L,ℓ))) 1{⟨η⟩_L ≥ (ℓ+1)/(L+1)}, the canonical average under the Bernoulli measure ν_ρ is H̄(ρ)=P(Bin(L+1,ρ)≥ℓ+1)/((L+1) binom(L,ℓ)), whose derivative is ρ^ℓ(1-ρ)^{L-ℓ}, not ρ^ℓ for ℓ<L. This contradicts the identification D(ρ)=H̄'(ρ) from Section 2.1 together with Proposition 2.9(vi), which directly gives ∫ p_{ℓ;L} dν_ρ = ρ^ℓ. Thus the gradient potential for the reduced PMM is not established; a correct h obtained from the inversion formula (2.6) would be a sum of threshold terms over n, not the single threshold with a binomial denominator.

minor comments (4)

[Section 2.1, Eq. (2.1)] The displayed generator in Eq. (2.1) is written as a sum plus an identical sum; this is presumably a typo, and the two sums should involve the two different exclusion factors e_{0,1} and e_{1,0}.
[Section 1, Introduction] The sentence beginning "Throughout this work, for simplicity we fix Let N ∈ N+ be fixed" is garbled and needs rewriting.
[Section 3.3, proof of Proposition 2.8] In the proof of Proposition 2.8, the sentence "Regarding (iv), an example is..." should refer to property (v); moreover, Proposition 2.8(iv) first states b_{n,L}(η)∈[0,1) and then correctly observes that the maximum 1 is attained, so the interval should be [0,1].
[Definition 2.3] The notation ⟦a,b⟧ is used without definition; it should be defined as the integer interval {a,a+1,...,b} or replaced by standard notation.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the gradient identities are derived by self-contained combinatorics; prior-work citations are contextual.

full rationale

The central claim of the paper, Proposition 2.7, is obtained by a direct expansion of the constraint b_{n,L} into monomial indicators, rewriting the algebraic current b_{n,L}(e_{0,1}-e_{1,0}) as a gradient of an explicitly defined auxiliary function g_{n,L} plus a remainder, and then telescoping that remainder into h_{n,L} using the sets W_P and \tilde W_P and the recursion α_{i+1}=α_i-∇β_i. This is a self-contained combinatorial derivation: no parameter is fitted to the target diffusivity B_{n,L}(ρ), and no prediction is read back from the hydrodynamic equation. The inversion formula for p_{ℓ;L} is computed directly from the definitions of the constraints. The identification of the diffusion coefficient with ∫ c dν_ρ is standard gradient-model theory (Spohn, Sasada) used as background, and the hydrodynamic limit itself is explicitly deferred to a companion paper in the title footnote; that is an acknowledged limitation, not a circular reliance. The self-citations [4] and [7] appear only as motivation and as support for the claimed gap in attainable diffusivities; the proof of the gradient identity does not use them as premises. The statement in Section 1.1 that a long-range extension breaks the gradient property and is left as an open problem is likewise a limitation, not a circular step. A possible sign error in Proposition 2.7 or the h-term in Lemma 3.4 would be a correctness defect, not circularity, and does not change this assessment.

Assumptions & free parameters 0 free parameters · 3 assumptions · 2 invented entities

The paper introduces no fitted numerical parameters. The central derivation relies on standard gradient-model theory (the gradient property implies a hydrodynamic equation with D = H̄' = ∫c), which is an external domain assumption, and on standard combinatorial identities. The two new stochastic processes are invented entities without independent evidence beyond the paper's own (currently questionable) algebraic claims.

assumptions (3)

domain assumption The gradient property j = -∇H implies the hydrodynamic equation ∂_t ρ = ∂_u^2 H̄(ρ) with D(ρ)=H̄'(ρ)=∫c dν_ρ.
Invoked in Section 2.1 when identifying the diffusion coefficient; cites Spohn [11, Part II Subsection 2.4] and Sasada [9]. This is a load-bearing external result that connects the algebraic gradient identity to the macroscopic PDE.
standard math Inclusion-exclusion and binomial identities used in Lemmas 3.1 and 3.3.
Standard combinatorial identities; the proof of Lemma 3.3 uses the distributive rule and inclusion-exclusion, and Lemma 3.1 uses direct counting of subsets.
domain assumption The torus size N is large enough so that the boxes W_j^L are well-defined and do not wrap around.
Used throughout the definitions; Proposition 2.8(i) explicitly assumes 0 < L < N/2, and the model requires the boxes W_j^L (length L) to be proper subsets without periodic identification issues.

invented entities (2)

Bernstein model B(n,L)
purpose: A symmetric exclusion process whose diffusion coefficient is the Bernstein polynomial B_{n,L}(ρ).
The model is new; its claimed diffusivity is derived in the paper via the gradient identity, but no independent simulation, numerical check, or companion proof of the hydrodynamic limit is provided yet.
Reduced Porous Media Model PMM_L(ℓ)
purpose: An auxiliary gradient process with diffusion coefficient ρ^ℓ and interaction range 2L+2.
The paper claims the gradient property via Lemma 3.4, but the stated potential appears inconsistent with the claimed diffusivity (H̄'(ρ)=ρ^ℓ(1-ρ)^{L-ℓ} instead of ρ^ℓ), so this entity is not yet established.

how reviews work

0 comments

Cite this review

Pith. "Pith review of A gradient model for the Bernstein polynomial basis." pith.science (2026). https://pith.science/paper/DDWDZKY7

@misc{pith2026241115954,
  author       = {Pith},
  title        = {Pith review of: A gradient model for the Bernstein polynomial basis},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DDWDZKY7}},
  note         = {Machine review of arXiv:2411.15954}
}

read the original abstract

We introduce a symmetric, gradient exclusion process within the class of non-cooperative kinetically constrained lattice gases, modelling a non-linear diffusivity in which the exchange of occupation values between two neighbouring sites depends on the local density in specific boxes surrounding the pair. The existence of such a model satisfying the gradient property is the main novelty of this work, filling a gap in the literature regarding the types of diffusivities attainable within this class of models. The resulting dynamics exhibits similarities with the Bernstein polynomial basis and generalises the Porous Media Model. We also introduce an auxiliary collection of processes, which extend the Porous Media Model in a different direction and are related to the former process via an inversion formula.

Figures

Figures reproduced from arXiv: 2411.15954 by the authors.

**Figure 1.** PMM rate (n = 4) for an exchange in the node {x, x + 1}. The patterned rectangles represent the boxes where there are n aligned particles around {x, x + 1}. The total rate is 2/5. A convenient characteristic of the PMM, and one relevant to the present work, is that it enjoys the gradient property. This means precisely that the algebraic microscopic current can be expressed as the (discrete) gradient of some function… view at source ↗

**Figure 2.** Bernstein model (n = 2, L = 4) rate for an exchange in the node {x, x + 1}. The patterned rectangle represents the box containing {x, x + 1} with exactly n + 1 particles. The total rate is 1/5. Remarkably, this simple dynamics enjoys the gradient property and generalises the combinatorial mechanism of the PMM, while coinciding with a PMM for n = L. Moreover, it is associated with the diffusion coefficient D(ρ) = Bn… view at source ↗

**Figure 3.** Reduced PMM (ℓ = 2, L = 4) rates for a jump in {x, x + 1}. The patterned rectangles represent the boxes containing {x, x + 1} with at least two particles. The total rate is 3 5( 4 2) . The two previously introduced models are connected through an inversion formula. Alternatively, introducing the dynamics from question (2), one can see the Bernstein model’s constraints as a binomial transformation of a sequence of r… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

12 extracted references · 12 canonical work pages

[1]

Bonorino, R

L. Bonorino, R. De Paula, P. Gon¸ calves, and A. Neumann. Hydro dynamics for the porous medium model with slow reservoirs. Journal of Statistical Physics, 179(3):0022–4715, 2020

work page 2020
[2]

Cardoso, R

P. Cardoso, R. De Paula, and P. Gon¸ calves. Derivation of the fr actional porous medium equation from a microscopic dynamics. Non-linearity, 36(3):1840–1872, February 2023

work page 2023
[3]

Linear and Nonlinear Fractional PDEs from interacting particle systems

P. Cardoso and P. Gon¸ calves. Linear and nonlinear fractional p des from interacting particle systems. arXiv preprint arXiv:2407.02246, pages 1–27, 2024

work page Pith review arXiv 2024
[4]

Derivation of non-polynomial fractional diffusions from the generalized exclusion with a slow barrier

P. Cardoso, P. Gon¸ calves, and G. S. Nahum. Derivation of non- polynomial fractional diﬀusions from the generalized exclusion with a slow barrier. arXiv preprint arXiv:2412.12414, 2024. 81 pages, 4 ﬁgures

work page Pith review arXiv 2024
[5]

Hydrodynamic limit for an active exclusion pro cess

Cl´ ement Erignoux. Hydrodynamic limit for an active exclusion pro cess. M´ emoiresde la SMF, 169, 2021

work page 2021
[6]

Gon¸ calves, C

P. Gon¸ calves, C. Landim, and C. Toninelli. Hydrodynamic limit for a p article system with degenerate rates. Annales de l’I.H.P.: Probability and Statistics, 45(4):887–909, 2009

work page 2009
[7]

Gon¸ calves, G

P. Gon¸ calves, G. S. Nahum, and M. Simon. From exclusion to slow a nd fast diﬀusion. arXiv preprint (accepted in Annales de l’Institut Henri Poincar´ e(B) Probabilit´ eset Statistiques), January 2023

work page 2023
[8]

Kipnis and C

C. Kipnis and C. Landim. Scaling limits of interacting particle systems. Springer, Berlin, 1999

work page 1999

Show all 12 references

[9]

On the green–kubo formula and the gradient con dition on currents

Makiko Sasada. On the green–kubo formula and the gradient con dition on currents. The Annals of Applied Probability, 28(5):2727–2739, October 2018

2018
[10]

F. Spitzer. Interaction of markov processes. Advances in Mathematics, 5(2):246–290, 1970

1970
[11]

H. Spohn. Large Scale Dynamics of Interacting Particles. Springer-Verlag, 1991

1991
[12]

Herbert S. Wilf. Generatingfunctionology. Academic Press, 2nd edition, 1993. Har dback. eBook ISBN: 9780080571515. Gabriel S. Nahum, inria Lyon, DRACULA team, 56 Bd. Niels Bohr , 69199 Villeurbanne, France Email address : gabriel.nahum@protonmail.com

1993

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.

[1] [1]

Bonorino, R

L. Bonorino, R. De Paula, P. Gon¸ calves, and A. Neumann. Hydro dynamics for the porous medium model with slow reservoirs. Journal of Statistical Physics, 179(3):0022–4715, 2020

work page 2020

[2] [2]

Cardoso, R

P. Cardoso, R. De Paula, and P. Gon¸ calves. Derivation of the fr actional porous medium equation from a microscopic dynamics. Non-linearity, 36(3):1840–1872, February 2023

work page 2023

[3] [3]

Linear and Nonlinear Fractional PDEs from interacting particle systems

P. Cardoso and P. Gon¸ calves. Linear and nonlinear fractional p des from interacting particle systems. arXiv preprint arXiv:2407.02246, pages 1–27, 2024

work page Pith review arXiv 2024

[4] [4]

Derivation of non-polynomial fractional diffusions from the generalized exclusion with a slow barrier

P. Cardoso, P. Gon¸ calves, and G. S. Nahum. Derivation of non- polynomial fractional diﬀusions from the generalized exclusion with a slow barrier. arXiv preprint arXiv:2412.12414, 2024. 81 pages, 4 ﬁgures

work page Pith review arXiv 2024

[5] [5]

Hydrodynamic limit for an active exclusion pro cess

Cl´ ement Erignoux. Hydrodynamic limit for an active exclusion pro cess. M´ emoiresde la SMF, 169, 2021

work page 2021

[6] [6]

Gon¸ calves, C

P. Gon¸ calves, C. Landim, and C. Toninelli. Hydrodynamic limit for a p article system with degenerate rates. Annales de l’I.H.P.: Probability and Statistics, 45(4):887–909, 2009

work page 2009

[7] [7]

Gon¸ calves, G

P. Gon¸ calves, G. S. Nahum, and M. Simon. From exclusion to slow a nd fast diﬀusion. arXiv preprint (accepted in Annales de l’Institut Henri Poincar´ e(B) Probabilit´ eset Statistiques), January 2023

work page 2023

[8] [8]

Kipnis and C

C. Kipnis and C. Landim. Scaling limits of interacting particle systems. Springer, Berlin, 1999

work page 1999