arxiv: 2605.12972 · v1 · pith:QRKLIXN7new · submitted 2026-05-13 · ❄️ cond-mat.stat-mech

Matrix-noise Jacobians in stochastic-calculus inference and optimal paths

Surachate Limkumnerd This is my paper

Pith reviewed 2026-05-14 18:40 UTC · model grok-4.3

classification ❄️ cond-mat.stat-mech

keywords multiplicative noisestochastic calculusJacobianOnsager-Machluppath likelihooddiffusion processesmatrix-valued noisediscretized diffusions

0 comments

The pith

In multidimensional systems with matrix-valued multiplicative noise, a Jacobian term from the noise amplitude survives scalar cancellations and alters fitted stochastic prescriptions and optimal paths.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper establishes that multiplicative noise in stochastic dynamics depends on the interpretation of the white-noise limit, and in multidimensional cases with matrix-valued noise amplitudes this includes a local Jacobian contribution absent in scalar examples. It develops a finite-step path-likelihood framework for discretized diffusions whose short-time expansion isolates this Jacobian J_sigma. The quantity vanishes in one-dimensional, isotropic, and diagonal cases but persists when state-dependent noise mixes components, leading to measurable shifts in inferred dynamics and paths. Tests in two models confirm that including or removing this term changes the fitted prescription and the optimized transition path when the Jacobian is nonzero.

Core claim

For a specified noise-amplitude representation sigma, the Jacobian J_sigma equals partial_j sigma_ik partial_i sigma_jk minus (partial_i sigma_ik)(partial_l sigma_lk). This quantity vanishes in one-dimensional, scalar-isotropic, and strictly diagonal cases, but can survive when state-dependent noise directions mix different components. Paired comparisons holding the drift, diffusion matrix, interpolation point, and Gaussian increment fixed show that removing the off-diagonal determinant contribution shifts the fitted stochastic prescription, and removing the corresponding state-dependent action term changes a stable optimized transition path.

What carries the argument

The scalar Jacobian J_sigma extracted from the short-time expansion of the path likelihood in the finite-step framework for theta-discretized diffusions.

If this is right

Removing only the off-diagonal determinant contribution produces a shift in the fitted stochastic prescription that vanishes when J_sigma equals zero.
Removing the corresponding state-dependent action term changes a stable optimized transition path.
The Jacobian term survives the scalar cancellations familiar from one-dimensional or diagonal settings when noise directions mix components.
A genuinely matrix-noise part of the short-time path measure produces measurable changes in fitted stochastic prescriptions and Onsager-Machlup paths.

Where Pith is reading between the lines

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

Inference methods that omit the matrix Jacobian may introduce systematic bias in parameter estimates for systems whose noise couples different state components.
Recalculating optimal paths in applications such as reaction networks or population models to include this term could yield different trajectories than scalar approximations predict.
The framework could be tested by comparing continuous-limit limits against exact path integrals in low-dimensional matrix-noise examples.

Load-bearing premise

The finite-step path-likelihood framework for theta-discretized diffusions accurately isolates the continuous-limit Jacobian contribution without additional discretization artifacts that would cancel J_sigma.

What would settle it

Computing path probabilities or optimized paths in a multidimensional model with off-diagonal state-dependent noise both with and without the J_sigma term and finding identical results would falsify the claim that the Jacobian survives to produce measurable changes.

Figures

Figures reproduced from arXiv: 2605.12972 by Surachate Limkumnerd.

**Figure 2.** Figure 2: FIG. 2. Robustness of the paired inference shift in Model [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

**Figure 3.** Figure 3: FIG. 3. Most likely transition paths for Model B. The full [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗

read the original abstract

Multiplicative noise makes stochastic dynamics depend on how the white-noise limit is interpreted. In multidimensional systems with matrix-valued noise amplitudes $\sigma(x)$, this dependence includes a local Jacobian contribution that is absent from the scalar examples most often used to build intuition. We formulate a finite-step path-likelihood framework for $\theta$-discretized diffusions and show that its short-time expansion isolates the scalar $J_\sigma=\partial_j\sigma_{ik}\partial_i\sigma_{jk}-(\partial_i\sigma_{ik})(\partial_l\sigma_{lk})$. For a specified noise-amplitude representation $\sigma$, this quantity vanishes in one-dimensional, scalar-isotropic, and strictly diagonal cases, but can survive when state-dependent noise directions mix different components. We then test its consequences using paired comparisons that hold the drift, diffusion matrix, interpolation point, and Gaussian increment term fixed. In Model A, removing only the off-diagonal determinant contribution produces a shift in the fitted stochastic prescription that vanishes when $J_\sigma=0$. In Model B, removing the corresponding state-dependent action term changes a stable optimized transition path. These results show that a genuinely matrix-noise part of the short-time path measure can survive the scalar cancellations familiar from simpler settings and produce measurable changes in fitted stochastic prescriptions and Onsager--Machlup paths.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper isolates a matrix Jacobian J_σ in the short-time path measure for multidimensional multiplicative noise that survives scalar cancellations and shifts both fitted prescriptions and optimized Onsager-Machlup paths.

read the letter

The core contribution is the explicit isolation of the scalar J_σ = ∂_j σ_ik ∂_i σ_jk − (∂_i σ_ik)(∂_l σ_lk) from the short-time expansion of a θ-discretized path likelihood. It vanishes for 1D, isotropic, or diagonal noise but remains when off-diagonal state-dependent terms mix components, and the paired models show this produces measurable changes in fitted stochastic rules and transition paths while holding drift, diffusion matrix, and Gaussian increments fixed. That setup is the paper's strongest move: the comparisons are targeted and avoid the usual confounding variables. The derivation itself is direct and builds on standard finite-step likelihoods without obvious circularity. The numerical tests in Models A and B are presented cleanly enough to illustrate the point. The main soft spot is the stress-test concern about residual discretization artifacts. The finite-step expansion before the continuous limit could in principle leave θ-dependent or normalization terms that either cancel J_σ or mimic it, and the abstract does not spell out an explicit check that all O(Δt) pieces drop out cleanly. If the full manuscript only assumes the cancellation rather than verifying it term-by-term, that leaves a modest gap in the isolation claim. Otherwise the argument holds up on its own terms. This is a technical note for researchers already working on stochastic calculus interpretations or path inference in systems with matrix-valued multiplicative noise. It refines an existing corner of the literature rather than reopening a large field, but the demonstration is concrete enough that a serious referee could evaluate it quickly. I would send it to peer review.

Referee Report

2 major / 2 minor

Summary. The paper claims that multidimensional diffusions with matrix-valued multiplicative noise σ(x) possess a local Jacobian J_σ = ∂_j σ_ik ∂_i σ_jk − (∂_i σ_ik)(∂_l σ_lk) that survives in the short-time expansion of the finite-step path likelihood for θ-discretized processes. This term vanishes for one-dimensional, scalar-isotropic, and strictly diagonal noise but remains when off-diagonal or state-dependent noise directions mix components. Paired numerical tests (Models A and B) that hold drift, diffusion matrix, interpolation point, and Gaussian increment fixed show that removing the corresponding contribution shifts the fitted stochastic prescription and alters the optimized Onsager–Machlup transition path.

Significance. If the isolation of J_σ is rigorous, the work supplies a concrete, matrix-specific correction to the path measure that is invisible in the scalar examples used to build intuition. The paired-model construction offers a falsifiable test of its effect on inference and optimal paths, which could matter for systems with anisotropic multiplicative noise such as certain Langevin models in statistical mechanics.

major comments (2)

[§3] §3 (short-time expansion): The derivation asserts that the finite-step θ-discretized likelihood isolates exactly J_σ with no residual O(Δt) contributions from the θ-interpolation rule or the position-dependent Gaussian normalization. The manuscript must supply an explicit term-by-term cancellation check (or an auxiliary expansion) demonstrating that any θ-dependent drift or Jacobian-of-the-map terms vanish independently of the matrix-mixing pieces that define J_σ; without this, the numerical shifts in Models A and B could be contaminated by discretization artifacts.
[§4.1] §4.1 (Model A): The claim that removing only the off-diagonal determinant contribution produces a shift that vanishes precisely when J_σ = 0 requires a quantitative demonstration that the fitted stochastic prescription difference is proportional to J_σ and not to other held-fixed quantities (e.g., the diffusion-matrix eigenvalues). The current paired comparison does not report the magnitude of the shift relative to the statistical uncertainty of the fit.

minor comments (2)

[Abstract] Notation: the repeated index summation convention for J_σ should be stated once at first appearance; the current definition leaves the range of the repeated indices implicit.
[§4] Figure captions: the captions for the Model A and B results should explicitly state the value of θ and the step size Δt used in the likelihood evaluations.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading of our manuscript and the constructive comments. We address the major comments point by point below and will make the necessary revisions to strengthen the presentation.

read point-by-point responses

Referee: [§3] §3 (short-time expansion): The derivation asserts that the finite-step θ-discretized likelihood isolates exactly J_σ with no residual O(Δt) contributions from the θ-interpolation rule or the position-dependent Gaussian normalization. The manuscript must supply an explicit term-by-term cancellation check (or an auxiliary expansion) demonstrating that any θ-dependent drift or Jacobian-of-the-map terms vanish independently of the matrix-mixing pieces that define J_σ; without this, the numerical shifts in Models A and B could be contaminated by discretization artifacts.

Authors: We agree with the referee that an explicit term-by-term check is required to fully substantiate the isolation of J_σ. In the revised manuscript, we will add an auxiliary expansion in section 3 that verifies the cancellation of all θ-dependent and normalization terms independently of the matrix-mixing contributions defining J_σ. This will eliminate any concern about discretization artifacts affecting the numerical results in Models A and B. revision: yes
Referee: [§4.1] §4.1 (Model A): The claim that removing only the off-diagonal determinant contribution produces a shift that vanishes precisely when J_σ = 0 requires a quantitative demonstration that the fitted stochastic prescription difference is proportional to J_σ and not to other held-fixed quantities (e.g., the diffusion-matrix eigenvalues). The current paired comparison does not report the magnitude of the shift relative to the statistical uncertainty of the fit.

Authors: We will enhance the analysis in §4.1 by including quantitative measures of the shift in the fitted stochastic prescription, along with the associated statistical uncertainties from the fits. This will demonstrate that the observed difference scales with J_σ and is not attributable to other fixed quantities such as the diffusion-matrix eigenvalues. The revised manuscript will report these magnitudes explicitly. revision: yes

Circularity Check

0 steps flagged

Derivation self-contained; no load-bearing reductions to inputs or self-citations

full rationale

The paper derives J_σ via short-time expansion of a finite-step θ-discretized path likelihood, then tests consequences in paired Models A/B by holding drift, diffusion matrix, and Gaussian term fixed while selectively removing off-diagonal or state-dependent pieces. No quoted equation reduces the claimed Jacobian to a fitted parameter or prior self-citation by construction. The central isolation of the scalar J_σ expression is presented as an algebraic outcome of the expansion, not a renaming or ansatz smuggled from prior work. External benchmarks (vanishing in 1D/scalar/diagonal cases) are stated explicitly and used for validation rather than assumed.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the validity of the finite-step path-likelihood expansion for θ-discretized diffusions and the assumption that the paired models isolate only the Jacobian contribution.

axioms (1)

domain assumption The short-time expansion of the path likelihood for θ-discretized diffusions isolates the scalar J_σ without residual discretization terms.
Invoked in the formulation of the finite-step framework and the isolation of J_σ.

pith-pipeline@v0.9.0 · 5527 in / 1298 out tokens · 23053 ms · 2026-05-14T18:40:16.713159+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

21 extracted references · 21 canonical work pages · 3 internal anchors

[1]

J. M. Sancho, M. San Miguel, and D. D¨ urr, Journal of Statistical Physics28, 291 (1982)

work page 1982
[2]

Kupferman, G

R. Kupferman, G. A. Pavliotis, and A. M. Stuart, Phys- ical Review E70, 036120 (2004)

work page 2004
[3]

A. W. C. Lau and T. C. Lubensky, Physical Review E 76, 011123 (2007), arXiv:0707.2234

work page internal anchor Pith review Pith/arXiv arXiv 2007
[4]

Volpe and J

G. Volpe and J. Wehr, Reports on Progress in Physics 79, 053901 (2016)

work page 2016
[5]

Pesce, A

G. Pesce, A. McDaniel, S. Hottovy, J. Wehr, and G. Volpe, Nature Communications4, 2733 (2013)

work page 2013
[6]

Pacheco-Pozo, M

A. Pacheco-Pozo, M. Balcerek, A. Wy loma´ nska, K. Bur- necki, I. M. Sokolov, and D. Krapf, Physical Review Let- ters133, 067102 (2024), arXiv:2403.11928

work page arXiv 2024
[7]

Liu, R.-B

H.-C. Liu, R.-B. Zhong, J.-F. Zhang, L.-M. Fan, P.-C. Li, and M.-G. Li, Physical Review Research8, 013143 (2026)

work page 2026
[8]

A. S. Serov, F. Laurent, C. Floderer, K. Perronet, C. Favard, D. Muriaux, C. L. Vestergaard, and J.-B. Mas- son, Scientific Reports10, 3783 (2020)

work page 2020
[9]

Onsager and S

L. Onsager and S. Machlup, Physical Review91, 1505 (1953)

work page 1953
[10]

Graham, Zeitschrift f¨ ur Physik B26, 281 (1977)

R. Graham, Zeitschrift f¨ ur Physik B26, 281 (1977)

work page 1977
[11]

D¨ urr and A

D. D¨ urr and A. Bach, Communications in Mathematical Physics60, 153 (1978)

work page 1978
[12]

Arnold, Physical Review E61, 6099 (2000), arXiv:hep- ph/9912209

P. Arnold, Physical Review E61, 6099 (2000), arXiv:hep- ph/9912209

work page arXiv 2000
[13]

Z. G. Arenas and D. G. Barci, Physical Review E81, 051113 (2010)

work page 2010
[14]

Z. G. Arenas and D. G. Barci, Physical Review E85, 041122 (2012), arXiv:1111.6123

work page internal anchor Pith review Pith/arXiv arXiv 2012
[15]

M. V. Moreno, Z. G. Arenas, and D. G. Barci, Physical Review E91, 042103 (2015), arXiv:1412.7015

work page internal anchor Pith review Pith/arXiv arXiv 2015
[16]

L. F. Cugliandolo and V. Lecomte, Journal of Physics A: Mathematical and Theoretical50, 345001 (2017), arXiv:1704.03501

work page arXiv 2017
[17]

L. F. Cugliandolo, V. Lecomte, and F. van Wijland, Journal of Physics A: Mathematical and Theoretical52, 50LT01 (2019), arXiv:1806.09486

work page arXiv 2019
[18]

Arnoulx de Pirey, L

T. Arnoulx de Pirey, L. F. Cugliandolo, V. Lecomte, and F. van Wijland, Advances in Physics71, 1 (2022), arXiv:2211.09470

work page arXiv 2022
[19]

M. V. S. Moreno and D. G. Barci, Physical Review E99, 032125 (2019)

work page 2019
[20]

F. S. Abril-Berm´ udez, C. J. Quimbay, J. E. Trinidad- Segovia, and M. A. S´ anchez-Granero, Physical Review Research7, 023185 (2025), arXiv:2410.01387

work page arXiv 2025
[21]

Y. Li, J. Duan, and X. Liu, Physical Review E103, 012124 (2021), arXiv:2010.04114

work page arXiv 2021